annotate i386/dsputil_mmx.c @ 6111:ed3937d4f2aa libavcodec

do not set channels before checking request_channels
author jbr
date Sat, 05 Jan 2008 15:18:15 +0000
parents 03febc8f506f
children 1c67967d893b
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
1 /*
986e461dc072 Initial revision
glantau
parents:
diff changeset
2 * MMX optimized DSP utils
429
718a22dc121f license/copyright change
glantau
parents: 422
diff changeset
3 * Copyright (c) 2000, 2001 Fabrice Bellard.
1739
07a484280a82 copyright year update of the files i touched and remembered, things look annoyingly unmaintained otherwise
michael
parents: 1729
diff changeset
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
5 *
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3932
diff changeset
6 * This file is part of FFmpeg.
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3932
diff changeset
7 *
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3932
diff changeset
8 * FFmpeg is free software; you can redistribute it and/or
429
718a22dc121f license/copyright change
glantau
parents: 422
diff changeset
9 * modify it under the terms of the GNU Lesser General Public
718a22dc121f license/copyright change
glantau
parents: 422
diff changeset
10 * License as published by the Free Software Foundation; either
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3932
diff changeset
11 * version 2.1 of the License, or (at your option) any later version.
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
12 *
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3932
diff changeset
13 * FFmpeg is distributed in the hope that it will be useful,
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
429
718a22dc121f license/copyright change
glantau
parents: 422
diff changeset
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
718a22dc121f license/copyright change
glantau
parents: 422
diff changeset
16 * Lesser General Public License for more details.
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
17 *
429
718a22dc121f license/copyright change
glantau
parents: 422
diff changeset
18 * You should have received a copy of the GNU Lesser General Public
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3932
diff changeset
19 * License along with FFmpeg; if not, write to the Free Software
3036
0b546eab515d Update licensing information: The FSF changed postal address.
diego
parents: 2979
diff changeset
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
21 *
986e461dc072 Initial revision
glantau
parents:
diff changeset
22 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
986e461dc072 Initial revision
glantau
parents:
diff changeset
23 */
986e461dc072 Initial revision
glantau
parents:
diff changeset
24
5010
d5ba514e3f4a Add libavcodec to compiler include flags in order to simplify header
diego
parents: 5007
diff changeset
25 #include "dsputil.h"
5946
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
26 #include "dsputil_mmx.h"
5010
d5ba514e3f4a Add libavcodec to compiler include flags in order to simplify header
diego
parents: 5007
diff changeset
27 #include "simple_idct.h"
d5ba514e3f4a Add libavcodec to compiler include flags in order to simplify header
diego
parents: 5007
diff changeset
28 #include "mpegvideo.h"
3398
e0927bc44a10 Move REG_* macros from libavcodec/i386/mmx.h to libavutil/x86_cpu.h
lucabe
parents: 3250
diff changeset
29 #include "x86_cpu.h"
1984
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
30 #include "mmx.h"
5014
42b99a3aadde better separation of vp3dsp functions from dsputil_mmx.c
aurel
parents: 5010
diff changeset
31 #include "vp3dsp_mmx.h"
42b99a3aadde better separation of vp3dsp functions from dsputil_mmx.c
aurel
parents: 5010
diff changeset
32 #include "vp3dsp_sse2.h"
5277
7b3fcb7c61ce Avoid linking with h263.c functions when the relevant codecs
aurel
parents: 5255
diff changeset
33 #include "h263.h"
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
34
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
35 //#undef NDEBUG
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
36 //#include <assert.h>
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
37
2868
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
38 extern void ff_idct_xvid_mmx(short *block);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
39 extern void ff_idct_xvid_mmx2(short *block);
1647
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
40
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
41 int mm_flags; /* multimedia extension flags */
936
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
42
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
43 /* pixel operations */
5947
37a03989871b use ff_ prefix for extern vars
aurel
parents: 5946
diff changeset
44 DECLARE_ALIGNED_8 (const uint64_t, ff_bone) = 0x0101010101010101ULL;
37a03989871b use ff_ prefix for extern vars
aurel
parents: 5946
diff changeset
45 DECLARE_ALIGNED_8 (const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
5946
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
46
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
47 DECLARE_ALIGNED_16(const uint64_t, ff_pdw_80000000[2]) =
3557
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
48 {0x8000000080000000ULL, 0x8000000080000000ULL};
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
49
5946
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
50 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_3 ) = 0x0003000300030003ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
51 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_4 ) = 0x0004000400040004ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
52 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_5 ) = 0x0005000500050005ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
53 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_8 ) = 0x0008000800080008ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
54 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_15 ) = 0x000F000F000F000FULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
55 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_16 ) = 0x0010001000100010ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
56 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_20 ) = 0x0014001400140014ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
57 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_32 ) = 0x0020002000200020ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
58 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_42 ) = 0x002A002A002A002AULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
59 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_64 ) = 0x0040004000400040ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
60 DECLARE_ALIGNED_8 (const uint64_t, ff_pw_96 ) = 0x0060006000600060ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
61 DECLARE_ALIGNED_16(const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
62
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
63 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_1 ) = 0x0101010101010101ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
64 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3 ) = 0x0303030303030303ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
65 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_7 ) = 0x0707070707070707ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
66 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_3F ) = 0x3F3F3F3F3F3F3F3FULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
67 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_A1 ) = 0xA1A1A1A1A1A1A1A1ULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
68 DECLARE_ALIGNED_8 (const uint64_t, ff_pb_FC ) = 0xFCFCFCFCFCFCFCFCULL;
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
69
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
70 DECLARE_ALIGNED_16(const double, ff_pd_1[2]) = { 1.0, 1.0 };
55251379b5b1 make ff_p* vars extern so that they can be used in various *_mmx.c files
aurel
parents: 5933
diff changeset
71 DECLARE_ALIGNED_16(const double, ff_pd_2[2]) = { 2.0, 2.0 };
5737
efa3c1f9259a sse2 version of compute_autocorr().
lorenm
parents: 5602
diff changeset
72
3576
f7125bf10892 Support for MacIntel, last part: balign directives
gpoirier
parents: 3574
diff changeset
73 #define JUMPALIGN() __asm __volatile (ASMALIGN(3)::)
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
74 #define MOVQ_ZERO(regd) __asm __volatile ("pxor %%" #regd ", %%" #regd ::)
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
75
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
76 #define MOVQ_WONE(regd) \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
77 __asm __volatile ( \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
78 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
79 "psrlw $15, %%" #regd ::)
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
80
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
81 #define MOVQ_BFE(regd) \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
82 __asm __volatile ( \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
83 "pcmpeqd %%" #regd ", %%" #regd " \n\t"\
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
84 "paddb %%" #regd ", %%" #regd " \n\t" ::)
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
85
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
86 #ifndef PIC
5947
37a03989871b use ff_ prefix for extern vars
aurel
parents: 5946
diff changeset
87 #define MOVQ_BONE(regd) __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_bone))
37a03989871b use ff_ prefix for extern vars
aurel
parents: 5946
diff changeset
88 #define MOVQ_WTWO(regd) __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(ff_wtwo))
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
89 #else
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
90 // for shared library it's better to use this way for accessing constants
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
91 // pcmpeqd -> -1
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
92 #define MOVQ_BONE(regd) \
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
93 __asm __volatile ( \
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
94 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
95 "psrlw $15, %%" #regd " \n\t" \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
96 "packuswb %%" #regd ", %%" #regd " \n\t" ::)
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
97
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
98 #define MOVQ_WTWO(regd) \
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
99 __asm __volatile ( \
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
100 "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
101 "psrlw $15, %%" #regd " \n\t" \
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
102 "psllw $1, %%" #regd " \n\t"::)
387
b8f3affeb8e1 shared lib support (req by kabi) ...
michaelni
parents: 386
diff changeset
103
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
104 #endif
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
105
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
106 // using regr as temporary and for the output result
444
a5edef76dac6 * new mmx code - based upon http://aggregate.org/MAGIC
kabi
parents: 438
diff changeset
107 // first argument is unmodifed and second is trashed
471
d7f65ea52aaa * reimplemented remaing avg_ pixel functions
kabi
parents: 448
diff changeset
108 // regfe is supposed to contain 0xfefefefefefefefe
d7f65ea52aaa * reimplemented remaing avg_ pixel functions
kabi
parents: 448
diff changeset
109 #define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
110 "movq " #rega ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
111 "pand " #regb ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
112 "pxor " #rega ", " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
113 "pand " #regfe "," #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
114 "psrlq $1, " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
115 "paddb " #regb ", " #regr " \n\t"
444
a5edef76dac6 * new mmx code - based upon http://aggregate.org/MAGIC
kabi
parents: 438
diff changeset
116
471
d7f65ea52aaa * reimplemented remaing avg_ pixel functions
kabi
parents: 448
diff changeset
117 #define PAVGB_MMX(rega, regb, regr, regfe) \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
118 "movq " #rega ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
119 "por " #regb ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
120 "pxor " #rega ", " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
121 "pand " #regfe "," #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
122 "psrlq $1, " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
123 "psubb " #regb ", " #regr " \n\t"
445
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
124
471
d7f65ea52aaa * reimplemented remaing avg_ pixel functions
kabi
parents: 448
diff changeset
125 // mm6 is supposed to contain 0xfefefefefefefefe
446
efe0c0d40577 * reenabled original xy2 put routine - rounding error is really bad with
kabi
parents: 445
diff changeset
126 #define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
127 "movq " #rega ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
128 "movq " #regc ", " #regp " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
129 "pand " #regb ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
130 "pand " #regd ", " #regp " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
131 "pxor " #rega ", " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
132 "pxor " #regc ", " #regd " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
133 "pand %%mm6, " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
134 "pand %%mm6, " #regd " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
135 "psrlq $1, " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
136 "psrlq $1, " #regd " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
137 "paddb " #regb ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
138 "paddb " #regd ", " #regp " \n\t"
446
efe0c0d40577 * reenabled original xy2 put routine - rounding error is really bad with
kabi
parents: 445
diff changeset
139
efe0c0d40577 * reenabled original xy2 put routine - rounding error is really bad with
kabi
parents: 445
diff changeset
140 #define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
141 "movq " #rega ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
142 "movq " #regc ", " #regp " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
143 "por " #regb ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
144 "por " #regd ", " #regp " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
145 "pxor " #rega ", " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
146 "pxor " #regc ", " #regd " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
147 "pand %%mm6, " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
148 "pand %%mm6, " #regd " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
149 "psrlq $1, " #regd " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
150 "psrlq $1, " #regb " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
151 "psubb " #regb ", " #regr " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
152 "psubb " #regd ", " #regp " \n\t"
446
efe0c0d40577 * reenabled original xy2 put routine - rounding error is really bad with
kabi
parents: 445
diff changeset
153
445
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
154 /***********************************/
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
155 /* MMX no rounding */
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
156 #define DEF(x, y) x ## _no_rnd_ ## y ##_mmx
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
157 #define SET_RND MOVQ_WONE
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
158 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
159 #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e)
445
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
160
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
161 #include "dsputil_mmx_rnd.h"
444
a5edef76dac6 * new mmx code - based upon http://aggregate.org/MAGIC
kabi
parents: 438
diff changeset
162
445
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
163 #undef DEF
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
164 #undef SET_RND
446
efe0c0d40577 * reenabled original xy2 put routine - rounding error is really bad with
kabi
parents: 445
diff changeset
165 #undef PAVGBP
471
d7f65ea52aaa * reimplemented remaing avg_ pixel functions
kabi
parents: 448
diff changeset
166 #undef PAVGB
445
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
167 /***********************************/
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
168 /* MMX rounding */
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
169
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
170 #define DEF(x, y) x ## _ ## y ##_mmx
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
171 #define SET_RND MOVQ_WTWO
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
172 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
173 #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e)
445
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
174
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
175 #include "dsputil_mmx_rnd.h"
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
176
62c01dbdc1e0 * code with new PAVGB for MMX only CPU splited into separate file
kabi
parents: 444
diff changeset
177 #undef DEF
448
e8c8ca9106aa * removed MANGLE from macros for setting constants
kabi
parents: 446
diff changeset
178 #undef SET_RND
446
efe0c0d40577 * reenabled original xy2 put routine - rounding error is really bad with
kabi
parents: 445
diff changeset
179 #undef PAVGBP
471
d7f65ea52aaa * reimplemented remaing avg_ pixel functions
kabi
parents: 448
diff changeset
180 #undef PAVGB
387
b8f3affeb8e1 shared lib support (req by kabi) ...
michaelni
parents: 386
diff changeset
181
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
182 /***********************************/
986e461dc072 Initial revision
glantau
parents:
diff changeset
183 /* 3Dnow specific */
986e461dc072 Initial revision
glantau
parents:
diff changeset
184
986e461dc072 Initial revision
glantau
parents:
diff changeset
185 #define DEF(x) x ## _3dnow
986e461dc072 Initial revision
glantau
parents:
diff changeset
186 #define PAVGB "pavgusb"
986e461dc072 Initial revision
glantau
parents:
diff changeset
187
986e461dc072 Initial revision
glantau
parents:
diff changeset
188 #include "dsputil_mmx_avg.h"
986e461dc072 Initial revision
glantau
parents:
diff changeset
189
986e461dc072 Initial revision
glantau
parents:
diff changeset
190 #undef DEF
986e461dc072 Initial revision
glantau
parents:
diff changeset
191 #undef PAVGB
986e461dc072 Initial revision
glantau
parents:
diff changeset
192
986e461dc072 Initial revision
glantau
parents:
diff changeset
193 /***********************************/
986e461dc072 Initial revision
glantau
parents:
diff changeset
194 /* MMX2 specific */
986e461dc072 Initial revision
glantau
parents:
diff changeset
195
386
f49629bab18d hopefully faster mmx2&3dnow MC
michaelni
parents: 342
diff changeset
196 #define DEF(x) x ## _mmx2
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
197
986e461dc072 Initial revision
glantau
parents:
diff changeset
198 /* Introduced only in MMX2 set */
986e461dc072 Initial revision
glantau
parents:
diff changeset
199 #define PAVGB "pavgb"
986e461dc072 Initial revision
glantau
parents:
diff changeset
200
986e461dc072 Initial revision
glantau
parents:
diff changeset
201 #include "dsputil_mmx_avg.h"
986e461dc072 Initial revision
glantau
parents:
diff changeset
202
986e461dc072 Initial revision
glantau
parents:
diff changeset
203 #undef DEF
986e461dc072 Initial revision
glantau
parents:
diff changeset
204 #undef PAVGB
986e461dc072 Initial revision
glantau
parents:
diff changeset
205
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
206 #define SBUTTERFLY(a,b,t,n,m)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
207 "mov" #m " " #a ", " #t " \n\t" /* abcd */\
3416
fb9d94637fb2 #define SBUTTERFLY outside CONFIG_ENCODERS
mru
parents: 3398
diff changeset
208 "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\
fb9d94637fb2 #define SBUTTERFLY outside CONFIG_ENCODERS
mru
parents: 3398
diff changeset
209 "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */\
fb9d94637fb2 #define SBUTTERFLY outside CONFIG_ENCODERS
mru
parents: 3398
diff changeset
210
4939
3409dc0e7797 cosmetics: remove duplicate transpose macro
lorenm
parents: 4796
diff changeset
211 #define TRANSPOSE4(a,b,c,d,t)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
212 SBUTTERFLY(a,b,t,wd,q) /* a=aebf t=cgdh */\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
213 SBUTTERFLY(c,d,b,wd,q) /* c=imjn b=kolp */\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
214 SBUTTERFLY(a,c,d,dq,q) /* a=aeim d=bfjn */\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
215 SBUTTERFLY(t,b,c,dq,q) /* t=cgko c=dhlp */
4939
3409dc0e7797 cosmetics: remove duplicate transpose macro
lorenm
parents: 4796
diff changeset
216
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
217 /***********************************/
986e461dc072 Initial revision
glantau
parents:
diff changeset
218 /* standard MMX */
986e461dc072 Initial revision
glantau
parents:
diff changeset
219
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
220 #ifdef CONFIG_ENCODERS
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
221 static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size)
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
222 {
386
f49629bab18d hopefully faster mmx2&3dnow MC
michaelni
parents: 342
diff changeset
223 asm volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
224 "mov $-128, %%"REG_a" \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
225 "pxor %%mm7, %%mm7 \n\t"
3576
f7125bf10892 Support for MacIntel, last part: balign directives
gpoirier
parents: 3574
diff changeset
226 ASMALIGN(4)
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
227 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
228 "movq (%0), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
229 "movq (%0, %2), %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
230 "movq %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
231 "movq %%mm2, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
232 "punpcklbw %%mm7, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
233 "punpckhbw %%mm7, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
234 "punpcklbw %%mm7, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
235 "punpckhbw %%mm7, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
236 "movq %%mm0, (%1, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
237 "movq %%mm1, 8(%1, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
238 "movq %%mm2, 16(%1, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
239 "movq %%mm3, 24(%1, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
240 "add %3, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
241 "add $32, %%"REG_a" \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
242 "js 1b \n\t"
386
f49629bab18d hopefully faster mmx2&3dnow MC
michaelni
parents: 342
diff changeset
243 : "+r" (pixels)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
244 : "r" (block+64), "r" ((long)line_size), "r" ((long)line_size*2)
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
245 : "%"REG_a
386
f49629bab18d hopefully faster mmx2&3dnow MC
michaelni
parents: 342
diff changeset
246 );
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
247 }
986e461dc072 Initial revision
glantau
parents:
diff changeset
248
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
249 static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride)
324
9c6f056f0e41 fixed mpeg4 time stuff on encoding
michaelni
parents: 296
diff changeset
250 {
9c6f056f0e41 fixed mpeg4 time stuff on encoding
michaelni
parents: 296
diff changeset
251 asm volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
252 "pxor %%mm7, %%mm7 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
253 "mov $-128, %%"REG_a" \n\t"
3576
f7125bf10892 Support for MacIntel, last part: balign directives
gpoirier
parents: 3574
diff changeset
254 ASMALIGN(4)
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
255 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
256 "movq (%0), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
257 "movq (%1), %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
258 "movq %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
259 "movq %%mm2, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
260 "punpcklbw %%mm7, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
261 "punpckhbw %%mm7, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
262 "punpcklbw %%mm7, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
263 "punpckhbw %%mm7, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
264 "psubw %%mm2, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
265 "psubw %%mm3, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
266 "movq %%mm0, (%2, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
267 "movq %%mm1, 8(%2, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
268 "add %3, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
269 "add %3, %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
270 "add $16, %%"REG_a" \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
271 "jnz 1b \n\t"
324
9c6f056f0e41 fixed mpeg4 time stuff on encoding
michaelni
parents: 296
diff changeset
272 : "+r" (s1), "+r" (s2)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
273 : "r" (block+64), "r" ((long)stride)
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
274 : "%"REG_a
324
9c6f056f0e41 fixed mpeg4 time stuff on encoding
michaelni
parents: 296
diff changeset
275 );
9c6f056f0e41 fixed mpeg4 time stuff on encoding
michaelni
parents: 296
diff changeset
276 }
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
277 #endif //CONFIG_ENCODERS
324
9c6f056f0e41 fixed mpeg4 time stuff on encoding
michaelni
parents: 296
diff changeset
278
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
279 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
280 {
986e461dc072 Initial revision
glantau
parents:
diff changeset
281 const DCTELEM *p;
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
282 uint8_t *pix;
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
283
986e461dc072 Initial revision
glantau
parents:
diff changeset
284 /* read the pixels */
986e461dc072 Initial revision
glantau
parents:
diff changeset
285 p = block;
986e461dc072 Initial revision
glantau
parents:
diff changeset
286 pix = pixels;
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
287 /* unrolled loop */
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
288 __asm __volatile(
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
289 "movq %3, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
290 "movq 8%3, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
291 "movq 16%3, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
292 "movq 24%3, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
293 "movq 32%3, %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
294 "movq 40%3, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
295 "movq 48%3, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
296 "movq 56%3, %%mm7 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
297 "packuswb %%mm1, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
298 "packuswb %%mm3, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
299 "packuswb %%mm5, %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
300 "packuswb %%mm7, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
301 "movq %%mm0, (%0) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
302 "movq %%mm2, (%0, %1) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
303 "movq %%mm4, (%0, %1, 2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
304 "movq %%mm6, (%0, %2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
305 ::"r" (pix), "r" ((long)line_size), "r" ((long)line_size*3), "m"(*p)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
306 :"memory");
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
307 pix += line_size*4;
986e461dc072 Initial revision
glantau
parents:
diff changeset
308 p += 32;
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
309
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
310 // if here would be an exact copy of the code above
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
311 // compiler would generate some very strange code
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
312 // thus using "r"
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
313 __asm __volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
314 "movq (%3), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
315 "movq 8(%3), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
316 "movq 16(%3), %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
317 "movq 24(%3), %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
318 "movq 32(%3), %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
319 "movq 40(%3), %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
320 "movq 48(%3), %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
321 "movq 56(%3), %%mm7 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
322 "packuswb %%mm1, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
323 "packuswb %%mm3, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
324 "packuswb %%mm5, %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
325 "packuswb %%mm7, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
326 "movq %%mm0, (%0) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
327 "movq %%mm2, (%0, %1) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
328 "movq %%mm4, (%0, %1, 2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
329 "movq %%mm6, (%0, %2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
330 ::"r" (pix), "r" ((long)line_size), "r" ((long)line_size*3), "r"(p)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
331 :"memory");
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
332 }
986e461dc072 Initial revision
glantau
parents:
diff changeset
333
3089
072dbc669253 MSVC-compatible __align8/__align16 declaration
diego
parents: 3036
diff changeset
334 static DECLARE_ALIGNED_8(const unsigned char, vector128[8]) =
1985
b2bc62fdecc0 move the 0x80 vector outside of the function, thus saving the compiler
melanson
parents: 1984
diff changeset
335 { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
b2bc62fdecc0 move the 0x80 vector outside of the function, thus saving the compiler
melanson
parents: 1984
diff changeset
336
1984
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
337 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
338 {
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
339 int i;
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
340
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
341 movq_m2r(*vector128, mm1);
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
342 for (i = 0; i < 8; i++) {
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
343 movq_m2r(*(block), mm0);
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
344 packsswb_m2r(*(block + 4), mm0);
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
345 block += 8;
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
346 paddb_r2r(mm1, mm0);
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
347 movq_r2m(mm0, *pixels);
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
348 pixels += line_size;
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
349 }
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
350 }
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
351
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
352 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size)
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
353 {
986e461dc072 Initial revision
glantau
parents:
diff changeset
354 const DCTELEM *p;
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
355 uint8_t *pix;
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
356 int i;
986e461dc072 Initial revision
glantau
parents:
diff changeset
357
986e461dc072 Initial revision
glantau
parents:
diff changeset
358 /* read the pixels */
986e461dc072 Initial revision
glantau
parents:
diff changeset
359 p = block;
986e461dc072 Initial revision
glantau
parents:
diff changeset
360 pix = pixels;
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
361 MOVQ_ZERO(mm7);
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
362 i = 4;
342
8635a7036395 * fixes problem with -funroll-loops and buggy gcc compiler
kabi
parents: 324
diff changeset
363 do {
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
364 __asm __volatile(
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
365 "movq (%2), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
366 "movq 8(%2), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
367 "movq 16(%2), %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
368 "movq 24(%2), %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
369 "movq %0, %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
370 "movq %1, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
371 "movq %%mm4, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
372 "punpcklbw %%mm7, %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
373 "punpckhbw %%mm7, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
374 "paddsw %%mm4, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
375 "paddsw %%mm5, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
376 "movq %%mm6, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
377 "punpcklbw %%mm7, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
378 "punpckhbw %%mm7, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
379 "paddsw %%mm6, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
380 "paddsw %%mm5, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
381 "packuswb %%mm1, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
382 "packuswb %%mm3, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
383 "movq %%mm0, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
384 "movq %%mm2, %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
385 :"+m"(*pix), "+m"(*(pix+line_size))
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
386 :"r"(p)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
387 :"memory");
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
388 pix += line_size*2;
986e461dc072 Initial revision
glantau
parents:
diff changeset
389 p += 16;
342
8635a7036395 * fixes problem with -funroll-loops and buggy gcc compiler
kabi
parents: 324
diff changeset
390 } while (--i);
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
391 }
986e461dc072 Initial revision
glantau
parents:
diff changeset
392
2209
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
393 static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
394 {
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
395 __asm __volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
396 "lea (%3, %3), %%"REG_a" \n\t"
3576
f7125bf10892 Support for MacIntel, last part: balign directives
gpoirier
parents: 3574
diff changeset
397 ASMALIGN(3)
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
398 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
399 "movd (%1), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
400 "movd (%1, %3), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
401 "movd %%mm0, (%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
402 "movd %%mm1, (%2, %3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
403 "add %%"REG_a", %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
404 "add %%"REG_a", %2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
405 "movd (%1), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
406 "movd (%1, %3), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
407 "movd %%mm0, (%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
408 "movd %%mm1, (%2, %3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
409 "add %%"REG_a", %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
410 "add %%"REG_a", %2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
411 "subl $4, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
412 "jnz 1b \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
413 : "+g"(h), "+r" (pixels), "+r" (block)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
414 : "r"((long)line_size)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
415 : "%"REG_a, "memory"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
416 );
2209
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
417 }
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
418
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
419 static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
420 {
471
d7f65ea52aaa * reimplemented remaing avg_ pixel functions
kabi
parents: 448
diff changeset
421 __asm __volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
422 "lea (%3, %3), %%"REG_a" \n\t"
3576
f7125bf10892 Support for MacIntel, last part: balign directives
gpoirier
parents: 3574
diff changeset
423 ASMALIGN(3)
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
424 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
425 "movq (%1), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
426 "movq (%1, %3), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
427 "movq %%mm0, (%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
428 "movq %%mm1, (%2, %3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
429 "add %%"REG_a", %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
430 "add %%"REG_a", %2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
431 "movq (%1), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
432 "movq (%1, %3), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
433 "movq %%mm0, (%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
434 "movq %%mm1, (%2, %3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
435 "add %%"REG_a", %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
436 "add %%"REG_a", %2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
437 "subl $4, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
438 "jnz 1b \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
439 : "+g"(h), "+r" (pixels), "+r" (block)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
440 : "r"((long)line_size)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
441 : "%"REG_a, "memory"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
442 );
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
443 }
986e461dc072 Initial revision
glantau
parents:
diff changeset
444
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
445 static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h)
651
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
446 {
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
447 __asm __volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
448 "lea (%3, %3), %%"REG_a" \n\t"
3576
f7125bf10892 Support for MacIntel, last part: balign directives
gpoirier
parents: 3574
diff changeset
449 ASMALIGN(3)
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
450 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
451 "movq (%1), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
452 "movq 8(%1), %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
453 "movq (%1, %3), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
454 "movq 8(%1, %3), %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
455 "movq %%mm0, (%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
456 "movq %%mm4, 8(%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
457 "movq %%mm1, (%2, %3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
458 "movq %%mm5, 8(%2, %3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
459 "add %%"REG_a", %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
460 "add %%"REG_a", %2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
461 "movq (%1), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
462 "movq 8(%1), %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
463 "movq (%1, %3), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
464 "movq 8(%1, %3), %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
465 "movq %%mm0, (%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
466 "movq %%mm4, 8(%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
467 "movq %%mm1, (%2, %3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
468 "movq %%mm5, 8(%2, %3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
469 "add %%"REG_a", %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
470 "add %%"REG_a", %2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
471 "subl $4, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
472 "jnz 1b \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
473 : "+g"(h), "+r" (pixels), "+r" (block)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
474 : "r"((long)line_size)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
475 : "%"REG_a, "memory"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
476 );
651
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
477 }
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
478
296
c1a8a1b4a24b sizeof(s->block) isnt 64*6*2 anymore bugfix
michaelni
parents: 294
diff changeset
479 static void clear_blocks_mmx(DCTELEM *blocks)
c1a8a1b4a24b sizeof(s->block) isnt 64*6*2 anymore bugfix
michaelni
parents: 294
diff changeset
480 {
471
d7f65ea52aaa * reimplemented remaing avg_ pixel functions
kabi
parents: 448
diff changeset
481 __asm __volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
482 "pxor %%mm7, %%mm7 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
483 "mov $-128*6, %%"REG_a" \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
484 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
485 "movq %%mm7, (%0, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
486 "movq %%mm7, 8(%0, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
487 "movq %%mm7, 16(%0, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
488 "movq %%mm7, 24(%0, %%"REG_a") \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
489 "add $32, %%"REG_a" \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
490 " js 1b \n\t"
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
491 : : "r" (((uint8_t *)blocks)+128*6)
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
492 : "%"REG_a
296
c1a8a1b4a24b sizeof(s->block) isnt 64*6*2 anymore bugfix
michaelni
parents: 294
diff changeset
493 );
c1a8a1b4a24b sizeof(s->block) isnt 64*6*2 anymore bugfix
michaelni
parents: 294
diff changeset
494 }
c1a8a1b4a24b sizeof(s->block) isnt 64*6*2 anymore bugfix
michaelni
parents: 294
diff changeset
495
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
496 #ifdef CONFIG_ENCODERS
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
497 static int pix_sum16_mmx(uint8_t * pix, int line_size){
688
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
498 const int h=16;
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
499 int sum;
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
500 long index= -line_size*h;
688
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
501
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
502 __asm __volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
503 "pxor %%mm7, %%mm7 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
504 "pxor %%mm6, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
505 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
506 "movq (%2, %1), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
507 "movq (%2, %1), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
508 "movq 8(%2, %1), %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
509 "movq 8(%2, %1), %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
510 "punpcklbw %%mm7, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
511 "punpckhbw %%mm7, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
512 "punpcklbw %%mm7, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
513 "punpckhbw %%mm7, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
514 "paddw %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
515 "paddw %%mm2, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
516 "paddw %%mm1, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
517 "paddw %%mm3, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
518 "add %3, %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
519 " js 1b \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
520 "movq %%mm6, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
521 "psrlq $32, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
522 "paddw %%mm5, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
523 "movq %%mm6, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
524 "psrlq $16, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
525 "paddw %%mm5, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
526 "movd %%mm6, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
527 "andl $0xFFFF, %0 \n\t"
688
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
528 : "=&r" (sum), "+r" (index)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
529 : "r" (pix - index), "r" ((long)line_size)
688
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
530 );
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
531
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
532 return sum;
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
533 }
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
534 #endif //CONFIG_ENCODERS
688
894b61908734 pix_sum16_mmx()
michaelni
parents: 651
diff changeset
535
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
536 static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
537 long i=0;
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
538 asm volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
539 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
540 "movq (%1, %0), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
541 "movq (%2, %0), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
542 "paddb %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
543 "movq %%mm1, (%2, %0) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
544 "movq 8(%1, %0), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
545 "movq 8(%2, %0), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
546 "paddb %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
547 "movq %%mm1, 8(%2, %0) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
548 "add $16, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
549 "cmp %3, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
550 " jb 1b \n\t"
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
551 : "+r" (i)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
552 : "r"(src), "r"(dst), "r"((long)w-15)
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
553 );
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
554 for(; i<w; i++)
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
555 dst[i+0] += src[i+0];
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
556 }
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
557
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
558 #define H263_LOOP_FILTER \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
559 "pxor %%mm7, %%mm7 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
560 "movq %0, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
561 "movq %0, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
562 "movq %3, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
563 "movq %3, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
564 "punpcklbw %%mm7, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
565 "punpckhbw %%mm7, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
566 "punpcklbw %%mm7, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
567 "punpckhbw %%mm7, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
568 "psubw %%mm2, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
569 "psubw %%mm3, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
570 "movq %1, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
571 "movq %1, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
572 "movq %2, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
573 "movq %2, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
574 "punpcklbw %%mm7, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
575 "punpckhbw %%mm7, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
576 "punpcklbw %%mm7, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
577 "punpckhbw %%mm7, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
578 "psubw %%mm2, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
579 "psubw %%mm3, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
580 "psllw $2, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
581 "psllw $2, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
582 "paddw %%mm0, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
583 "paddw %%mm1, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
584 "pxor %%mm6, %%mm6 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
585 "pcmpgtw %%mm4, %%mm6 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
586 "pcmpgtw %%mm5, %%mm7 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
587 "pxor %%mm6, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
588 "pxor %%mm7, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
589 "psubw %%mm6, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
590 "psubw %%mm7, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
591 "psrlw $3, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
592 "psrlw $3, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
593 "packuswb %%mm5, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
594 "packsswb %%mm7, %%mm6 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
595 "pxor %%mm7, %%mm7 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
596 "movd %4, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
597 "punpcklbw %%mm2, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
598 "punpcklbw %%mm2, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
599 "punpcklbw %%mm2, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
600 "psubusb %%mm4, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
601 "movq %%mm2, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
602 "psubusb %%mm4, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
603 "psubb %%mm3, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
604 "movq %1, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
605 "movq %2, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
606 "pxor %%mm6, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
607 "pxor %%mm6, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
608 "paddusb %%mm2, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
609 "psubusb %%mm2, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
610 "pxor %%mm6, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
611 "pxor %%mm6, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
612 "paddusb %%mm2, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
613 "packsswb %%mm1, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
614 "pcmpgtb %%mm0, %%mm7 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
615 "pxor %%mm7, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
616 "psubb %%mm7, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
617 "movq %%mm0, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
618 "psubusb %%mm2, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
619 "psubb %%mm0, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
620 "pand %5, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
621 "psrlw $2, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
622 "pxor %%mm7, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
623 "psubb %%mm7, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
624 "movq %0, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
625 "movq %3, %%mm6 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
626 "psubb %%mm1, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
627 "paddb %%mm1, %%mm6 \n\t"
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
628
1647
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
629 static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale){
5394
e9a6215f4e3a help some gcc version to optimize out those functions
aurel
parents: 5278
diff changeset
630 if(ENABLE_ANY_H263) {
1647
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
631 const int strength= ff_h263_loop_filter_strength[qscale];
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
632
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
633 asm volatile(
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
634
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
635 H263_LOOP_FILTER
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
636
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
637 "movq %%mm3, %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
638 "movq %%mm4, %2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
639 "movq %%mm5, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
640 "movq %%mm6, %3 \n\t"
1647
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
641 : "+m" (*(uint64_t*)(src - 2*stride)),
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
642 "+m" (*(uint64_t*)(src - 1*stride)),
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
643 "+m" (*(uint64_t*)(src + 0*stride)),
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
644 "+m" (*(uint64_t*)(src + 1*stride))
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
645 : "g" (2*strength), "m"(ff_pb_FC)
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
646 );
5394
e9a6215f4e3a help some gcc version to optimize out those functions
aurel
parents: 5278
diff changeset
647 }
1647
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
648 }
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
649
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
650 static inline void transpose4x4(uint8_t *dst, uint8_t *src, int dst_stride, int src_stride){
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
651 asm volatile( //FIXME could save 1 instruction if done as 8x4 ...
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
652 "movd %4, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
653 "movd %5, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
654 "movd %6, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
655 "movd %7, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
656 "punpcklbw %%mm1, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
657 "punpcklbw %%mm3, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
658 "movq %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
659 "punpcklwd %%mm2, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
660 "punpckhwd %%mm2, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
661 "movd %%mm0, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
662 "punpckhdq %%mm0, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
663 "movd %%mm0, %1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
664 "movd %%mm1, %2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
665 "punpckhdq %%mm1, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
666 "movd %%mm1, %3 \n\t"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
667
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
668 : "=m" (*(uint32_t*)(dst + 0*dst_stride)),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
669 "=m" (*(uint32_t*)(dst + 1*dst_stride)),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
670 "=m" (*(uint32_t*)(dst + 2*dst_stride)),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
671 "=m" (*(uint32_t*)(dst + 3*dst_stride))
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
672 : "m" (*(uint32_t*)(src + 0*src_stride)),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
673 "m" (*(uint32_t*)(src + 1*src_stride)),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
674 "m" (*(uint32_t*)(src + 2*src_stride)),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
675 "m" (*(uint32_t*)(src + 3*src_stride))
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
676 );
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
677 }
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
678
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
679 static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
5394
e9a6215f4e3a help some gcc version to optimize out those functions
aurel
parents: 5278
diff changeset
680 if(ENABLE_ANY_H263) {
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
681 const int strength= ff_h263_loop_filter_strength[qscale];
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
682 uint64_t temp[4] __attribute__ ((aligned(8)));
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
683 uint8_t *btemp= (uint8_t*)temp;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
684
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
685 src -= 2;
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
686
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
687 transpose4x4(btemp , src , 8, stride);
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
688 transpose4x4(btemp+4, src + 4*stride, 8, stride);
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
689 asm volatile(
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
690 H263_LOOP_FILTER // 5 3 4 6
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
691
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
692 : "+m" (temp[0]),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
693 "+m" (temp[1]),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
694 "+m" (temp[2]),
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
695 "+m" (temp[3])
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
696 : "g" (2*strength), "m"(ff_pb_FC)
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
697 );
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
698
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
699 asm volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
700 "movq %%mm5, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
701 "movq %%mm4, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
702 "punpcklbw %%mm3, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
703 "punpcklbw %%mm6, %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
704 "punpckhbw %%mm3, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
705 "punpckhbw %%mm6, %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
706 "movq %%mm5, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
707 "movq %%mm1, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
708 "punpcklwd %%mm4, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
709 "punpcklwd %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
710 "punpckhwd %%mm4, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
711 "punpckhwd %%mm0, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
712 "movd %%mm5, (%0) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
713 "punpckhdq %%mm5, %%mm5 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
714 "movd %%mm5, (%0,%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
715 "movd %%mm3, (%0,%2,2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
716 "punpckhdq %%mm3, %%mm3 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
717 "movd %%mm3, (%0,%3) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
718 "movd %%mm1, (%1) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
719 "punpckhdq %%mm1, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
720 "movd %%mm1, (%1,%2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
721 "movd %%mm6, (%1,%2,2) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
722 "punpckhdq %%mm6, %%mm6 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
723 "movd %%mm6, (%1,%3) \n\t"
2505
86e2b1424801 optimization and gcc 4.0 bug workaround patch by (Martin Drab >drab kepler.fjfi.cvut cz<)
michael
parents: 2293
diff changeset
724 :: "r" (src),
86e2b1424801 optimization and gcc 4.0 bug workaround patch by (Martin Drab >drab kepler.fjfi.cvut cz<)
michael
parents: 2293
diff changeset
725 "r" (src + 4*stride),
86e2b1424801 optimization and gcc 4.0 bug workaround patch by (Martin Drab >drab kepler.fjfi.cvut cz<)
michael
parents: 2293
diff changeset
726 "r" ((long) stride ),
86e2b1424801 optimization and gcc 4.0 bug workaround patch by (Martin Drab >drab kepler.fjfi.cvut cz<)
michael
parents: 2293
diff changeset
727 "r" ((long)(3*stride))
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
728 );
5394
e9a6215f4e3a help some gcc version to optimize out those functions
aurel
parents: 5278
diff changeset
729 }
1648
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
730 }
de28264c3dc3 h263_h_loop_filter_mmx
michael
parents: 1647
diff changeset
731
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
732 #ifdef CONFIG_ENCODERS
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
733 static int pix_norm1_mmx(uint8_t *pix, int line_size) {
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
734 int tmp;
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
735 asm volatile (
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
736 "movl $16,%%ecx\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
737 "pxor %%mm0,%%mm0\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
738 "pxor %%mm7,%%mm7\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
739 "1:\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
740 "movq (%0),%%mm2\n" /* mm2 = pix[0-7] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
741 "movq 8(%0),%%mm3\n" /* mm3 = pix[8-15] */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
742
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
743 "movq %%mm2,%%mm1\n" /* mm1 = mm2 = pix[0-7] */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
744
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
745 "punpckhbw %%mm0,%%mm1\n" /* mm1 = [pix4-7] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
746 "punpcklbw %%mm0,%%mm2\n" /* mm2 = [pix0-3] */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
747
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
748 "movq %%mm3,%%mm4\n" /* mm4 = mm3 = pix[8-15] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
749 "punpckhbw %%mm0,%%mm3\n" /* mm3 = [pix12-15] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
750 "punpcklbw %%mm0,%%mm4\n" /* mm4 = [pix8-11] */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
751
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
752 "pmaddwd %%mm1,%%mm1\n" /* mm1 = (pix0^2+pix1^2,pix2^2+pix3^2) */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
753 "pmaddwd %%mm2,%%mm2\n" /* mm2 = (pix4^2+pix5^2,pix6^2+pix7^2) */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
754
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
755 "pmaddwd %%mm3,%%mm3\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
756 "pmaddwd %%mm4,%%mm4\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
757
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
758 "paddd %%mm1,%%mm2\n" /* mm2 = (pix0^2+pix1^2+pix4^2+pix5^2,
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
759 pix2^2+pix3^2+pix6^2+pix7^2) */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
760 "paddd %%mm3,%%mm4\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
761 "paddd %%mm2,%%mm7\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
762
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
763 "add %2, %0\n"
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
764 "paddd %%mm4,%%mm7\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
765 "dec %%ecx\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
766 "jnz 1b\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
767
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
768 "movq %%mm7,%%mm1\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
769 "psrlq $32, %%mm7\n" /* shift hi dword to lo */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
770 "paddd %%mm7,%%mm1\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
771 "movd %%mm1,%1\n"
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
772 : "+r" (pix), "=r"(tmp) : "r" ((long)line_size) : "%ecx" );
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
773 return tmp;
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
774 }
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
775
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
776 static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
777 int tmp;
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
778 asm volatile (
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
779 "movl %4,%%ecx\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
780 "shr $1,%%ecx\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
781 "pxor %%mm0,%%mm0\n" /* mm0 = 0 */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
782 "pxor %%mm7,%%mm7\n" /* mm7 holds the sum */
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
783 "1:\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
784 "movq (%0),%%mm1\n" /* mm1 = pix1[0][0-7] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
785 "movq (%1),%%mm2\n" /* mm2 = pix2[0][0-7] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
786 "movq (%0,%3),%%mm3\n" /* mm3 = pix1[1][0-7] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
787 "movq (%1,%3),%%mm4\n" /* mm4 = pix2[1][0-7] */
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
788
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
789 /* todo: mm1-mm2, mm3-mm4 */
5963
80103098c797 spelling
vitor
parents: 5952
diff changeset
790 /* algo: subtract mm1 from mm2 with saturation and vice versa */
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
791 /* OR the results to get absolute difference */
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
792 "movq %%mm1,%%mm5\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
793 "movq %%mm3,%%mm6\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
794 "psubusb %%mm2,%%mm1\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
795 "psubusb %%mm4,%%mm3\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
796 "psubusb %%mm5,%%mm2\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
797 "psubusb %%mm6,%%mm4\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
798
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
799 "por %%mm1,%%mm2\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
800 "por %%mm3,%%mm4\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
801
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
802 /* now convert to 16-bit vectors so we can square them */
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
803 "movq %%mm2,%%mm1\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
804 "movq %%mm4,%%mm3\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
805
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
806 "punpckhbw %%mm0,%%mm2\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
807 "punpckhbw %%mm0,%%mm4\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
808 "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
809 "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
810
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
811 "pmaddwd %%mm2,%%mm2\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
812 "pmaddwd %%mm4,%%mm4\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
813 "pmaddwd %%mm1,%%mm1\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
814 "pmaddwd %%mm3,%%mm3\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
815
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
816 "lea (%0,%3,2), %0\n" /* pix1 += 2*line_size */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
817 "lea (%1,%3,2), %1\n" /* pix2 += 2*line_size */
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
818
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
819 "paddd %%mm2,%%mm1\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
820 "paddd %%mm4,%%mm3\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
821 "paddd %%mm1,%%mm7\n"
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
822 "paddd %%mm3,%%mm7\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
823
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
824 "decl %%ecx\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
825 "jnz 1b\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
826
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
827 "movq %%mm7,%%mm1\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
828 "psrlq $32, %%mm7\n" /* shift hi dword to lo */
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
829 "paddd %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
830 "movd %%mm1,%2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
831 : "+r" (pix1), "+r" (pix2), "=r"(tmp)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
832 : "r" ((long)line_size) , "m" (h)
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
833 : "%ecx");
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
834 return tmp;
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
835 }
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
836
1708
dea5b2946999 interlaced motion estimation
michael
parents: 1686
diff changeset
837 static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
838 int tmp;
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
839 asm volatile (
1708
dea5b2946999 interlaced motion estimation
michael
parents: 1686
diff changeset
840 "movl %4,%%ecx\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
841 "pxor %%mm0,%%mm0\n" /* mm0 = 0 */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
842 "pxor %%mm7,%%mm7\n" /* mm7 holds the sum */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
843 "1:\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
844 "movq (%0),%%mm1\n" /* mm1 = pix1[0-7] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
845 "movq (%1),%%mm2\n" /* mm2 = pix2[0-7] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
846 "movq 8(%0),%%mm3\n" /* mm3 = pix1[8-15] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
847 "movq 8(%1),%%mm4\n" /* mm4 = pix2[8-15] */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
848
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
849 /* todo: mm1-mm2, mm3-mm4 */
5963
80103098c797 spelling
vitor
parents: 5952
diff changeset
850 /* algo: subtract mm1 from mm2 with saturation and vice versa */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
851 /* OR the results to get absolute difference */
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
852 "movq %%mm1,%%mm5\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
853 "movq %%mm3,%%mm6\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
854 "psubusb %%mm2,%%mm1\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
855 "psubusb %%mm4,%%mm3\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
856 "psubusb %%mm5,%%mm2\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
857 "psubusb %%mm6,%%mm4\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
858
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
859 "por %%mm1,%%mm2\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
860 "por %%mm3,%%mm4\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
861
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
862 /* now convert to 16-bit vectors so we can square them */
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
863 "movq %%mm2,%%mm1\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
864 "movq %%mm4,%%mm3\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
865
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
866 "punpckhbw %%mm0,%%mm2\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
867 "punpckhbw %%mm0,%%mm4\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
868 "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
869 "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
870
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
871 "pmaddwd %%mm2,%%mm2\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
872 "pmaddwd %%mm4,%%mm4\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
873 "pmaddwd %%mm1,%%mm1\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
874 "pmaddwd %%mm3,%%mm3\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
875
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
876 "add %3,%0\n"
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
877 "add %3,%1\n"
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
878
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
879 "paddd %%mm2,%%mm1\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
880 "paddd %%mm4,%%mm3\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
881 "paddd %%mm1,%%mm7\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
882 "paddd %%mm3,%%mm7\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
883
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
884 "decl %%ecx\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
885 "jnz 1b\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
886
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
887 "movq %%mm7,%%mm1\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
888 "psrlq $32, %%mm7\n" /* shift hi dword to lo */
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
889 "paddd %%mm7,%%mm1\n"
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
890 "movd %%mm1,%2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
891 : "+r" (pix1), "+r" (pix2), "=r"(tmp)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
892 : "r" ((long)line_size) , "m" (h)
1708
dea5b2946999 interlaced motion estimation
michael
parents: 1686
diff changeset
893 : "%ecx");
997
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
894 return tmp;
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
895 }
4dfe15ae0078 sse16 & pix_norm1 optimization patch by (Felix von Leitner <felix-ffmpeg at fefe dot de>) (with some modifications)
michaelni
parents: 984
diff changeset
896
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
897 static int sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
898 int tmp;
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
899 asm volatile (
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
900 "shr $1,%2\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
901 "pxor %%xmm0,%%xmm0\n" /* mm0 = 0 */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
902 "pxor %%xmm7,%%xmm7\n" /* mm7 holds the sum */
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
903 "1:\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
904 "movdqu (%0),%%xmm1\n" /* mm1 = pix1[0][0-15] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
905 "movdqu (%1),%%xmm2\n" /* mm2 = pix2[0][0-15] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
906 "movdqu (%0,%4),%%xmm3\n" /* mm3 = pix1[1][0-15] */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
907 "movdqu (%1,%4),%%xmm4\n" /* mm4 = pix2[1][0-15] */
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
908
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
909 /* todo: mm1-mm2, mm3-mm4 */
5963
80103098c797 spelling
vitor
parents: 5952
diff changeset
910 /* algo: subtract mm1 from mm2 with saturation and vice versa */
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
911 /* OR the results to get absolute difference */
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
912 "movdqa %%xmm1,%%xmm5\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
913 "movdqa %%xmm3,%%xmm6\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
914 "psubusb %%xmm2,%%xmm1\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
915 "psubusb %%xmm4,%%xmm3\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
916 "psubusb %%xmm5,%%xmm2\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
917 "psubusb %%xmm6,%%xmm4\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
918
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
919 "por %%xmm1,%%xmm2\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
920 "por %%xmm3,%%xmm4\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
921
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
922 /* now convert to 16-bit vectors so we can square them */
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
923 "movdqa %%xmm2,%%xmm1\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
924 "movdqa %%xmm4,%%xmm3\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
925
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
926 "punpckhbw %%xmm0,%%xmm2\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
927 "punpckhbw %%xmm0,%%xmm4\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
928 "punpcklbw %%xmm0,%%xmm1\n" /* mm1 now spread over (mm1,mm2) */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
929 "punpcklbw %%xmm0,%%xmm3\n" /* mm4 now spread over (mm3,mm4) */
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
930
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
931 "pmaddwd %%xmm2,%%xmm2\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
932 "pmaddwd %%xmm4,%%xmm4\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
933 "pmaddwd %%xmm1,%%xmm1\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
934 "pmaddwd %%xmm3,%%xmm3\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
935
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
936 "lea (%0,%4,2), %0\n" /* pix1 += 2*line_size */
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
937 "lea (%1,%4,2), %1\n" /* pix2 += 2*line_size */
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
938
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
939 "paddd %%xmm2,%%xmm1\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
940 "paddd %%xmm4,%%xmm3\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
941 "paddd %%xmm1,%%xmm7\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
942 "paddd %%xmm3,%%xmm7\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
943
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
944 "decl %2\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
945 "jnz 1b\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
946
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
947 "movdqa %%xmm7,%%xmm1\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
948 "psrldq $8, %%xmm7\n" /* shift hi qword to lo */
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
949 "paddd %%xmm1,%%xmm7\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
950 "movdqa %%xmm7,%%xmm1\n"
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
951 "psrldq $4, %%xmm7\n" /* shift hi dword to lo */
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
952 "paddd %%xmm1,%%xmm7\n"
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
953 "movd %%xmm7,%3\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
954 : "+r" (pix1), "+r" (pix2), "+r"(h), "=r"(tmp)
2899
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
955 : "r" ((long)line_size));
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
956 return tmp;
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
957 }
d3a726717baf sse2 16x16 sum squared diff (306=>268 cycles on a K8)
lorenm
parents: 2892
diff changeset
958
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
959 static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
960 int tmp;
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
961 asm volatile (
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
962 "movl %3,%%ecx\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
963 "pxor %%mm7,%%mm7\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
964 "pxor %%mm6,%%mm6\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
965
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
966 "movq (%0),%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
967 "movq %%mm0, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
968 "psllq $8, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
969 "psrlq $8, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
970 "psrlq $8, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
971 "movq %%mm0, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
972 "movq %%mm1, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
973 "punpcklbw %%mm7,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
974 "punpcklbw %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
975 "punpckhbw %%mm7,%%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
976 "punpckhbw %%mm7,%%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
977 "psubw %%mm1, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
978 "psubw %%mm3, %%mm2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
979
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
980 "add %2,%0\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
981
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
982 "movq (%0),%%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
983 "movq %%mm4, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
984 "psllq $8, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
985 "psrlq $8, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
986 "psrlq $8, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
987 "movq %%mm4, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
988 "movq %%mm1, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
989 "punpcklbw %%mm7,%%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
990 "punpcklbw %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
991 "punpckhbw %%mm7,%%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
992 "punpckhbw %%mm7,%%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
993 "psubw %%mm1, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
994 "psubw %%mm3, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
995 "psubw %%mm4, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
996 "psubw %%mm5, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
997 "pxor %%mm3, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
998 "pxor %%mm1, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
999 "pcmpgtw %%mm0, %%mm3\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1000 "pcmpgtw %%mm2, %%mm1\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1001 "pxor %%mm3, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1002 "pxor %%mm1, %%mm2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1003 "psubw %%mm3, %%mm0\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1004 "psubw %%mm1, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1005 "paddw %%mm0, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1006 "paddw %%mm2, %%mm6\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1007
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1008 "add %2,%0\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1009 "1:\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1010
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1011 "movq (%0),%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1012 "movq %%mm0, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1013 "psllq $8, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1014 "psrlq $8, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1015 "psrlq $8, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1016 "movq %%mm0, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1017 "movq %%mm1, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1018 "punpcklbw %%mm7,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1019 "punpcklbw %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1020 "punpckhbw %%mm7,%%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1021 "punpckhbw %%mm7,%%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1022 "psubw %%mm1, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1023 "psubw %%mm3, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1024 "psubw %%mm0, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1025 "psubw %%mm2, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1026 "pxor %%mm3, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1027 "pxor %%mm1, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1028 "pcmpgtw %%mm4, %%mm3\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1029 "pcmpgtw %%mm5, %%mm1\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1030 "pxor %%mm3, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1031 "pxor %%mm1, %%mm5\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1032 "psubw %%mm3, %%mm4\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1033 "psubw %%mm1, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1034 "paddw %%mm4, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1035 "paddw %%mm5, %%mm6\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1036
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1037 "add %2,%0\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1038
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1039 "movq (%0),%%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1040 "movq %%mm4, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1041 "psllq $8, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1042 "psrlq $8, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1043 "psrlq $8, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1044 "movq %%mm4, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1045 "movq %%mm1, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1046 "punpcklbw %%mm7,%%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1047 "punpcklbw %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1048 "punpckhbw %%mm7,%%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1049 "punpckhbw %%mm7,%%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1050 "psubw %%mm1, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1051 "psubw %%mm3, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1052 "psubw %%mm4, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1053 "psubw %%mm5, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1054 "pxor %%mm3, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1055 "pxor %%mm1, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1056 "pcmpgtw %%mm0, %%mm3\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1057 "pcmpgtw %%mm2, %%mm1\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1058 "pxor %%mm3, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1059 "pxor %%mm1, %%mm2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1060 "psubw %%mm3, %%mm0\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1061 "psubw %%mm1, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1062 "paddw %%mm0, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1063 "paddw %%mm2, %%mm6\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1064
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1065 "add %2,%0\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1066 "subl $2, %%ecx\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1067 " jnz 1b\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1068
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1069 "movq %%mm6, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1070 "punpcklwd %%mm7,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1071 "punpckhwd %%mm7,%%mm6\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1072 "paddd %%mm0, %%mm6\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1073
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1074 "movq %%mm6,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1075 "psrlq $32, %%mm6\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1076 "paddd %%mm6,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1077 "movd %%mm0,%1\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1078 : "+r" (pix1), "=r"(tmp)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1079 : "r" ((long)line_size) , "g" (h-2)
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1080 : "%ecx");
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1081 return tmp;
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1082 }
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1083
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1084 static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1085 int tmp;
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1086 uint8_t * pix= pix1;
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1087 asm volatile (
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1088 "movl %3,%%ecx\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1089 "pxor %%mm7,%%mm7\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1090 "pxor %%mm6,%%mm6\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1091
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1092 "movq (%0),%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1093 "movq 1(%0),%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1094 "movq %%mm0, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1095 "movq %%mm1, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1096 "punpcklbw %%mm7,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1097 "punpcklbw %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1098 "punpckhbw %%mm7,%%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1099 "punpckhbw %%mm7,%%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1100 "psubw %%mm1, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1101 "psubw %%mm3, %%mm2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1102
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1103 "add %2,%0\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1104
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1105 "movq (%0),%%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1106 "movq 1(%0),%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1107 "movq %%mm4, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1108 "movq %%mm1, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1109 "punpcklbw %%mm7,%%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1110 "punpcklbw %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1111 "punpckhbw %%mm7,%%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1112 "punpckhbw %%mm7,%%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1113 "psubw %%mm1, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1114 "psubw %%mm3, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1115 "psubw %%mm4, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1116 "psubw %%mm5, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1117 "pxor %%mm3, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1118 "pxor %%mm1, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1119 "pcmpgtw %%mm0, %%mm3\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1120 "pcmpgtw %%mm2, %%mm1\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1121 "pxor %%mm3, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1122 "pxor %%mm1, %%mm2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1123 "psubw %%mm3, %%mm0\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1124 "psubw %%mm1, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1125 "paddw %%mm0, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1126 "paddw %%mm2, %%mm6\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1127
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1128 "add %2,%0\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1129 "1:\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1130
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1131 "movq (%0),%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1132 "movq 1(%0),%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1133 "movq %%mm0, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1134 "movq %%mm1, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1135 "punpcklbw %%mm7,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1136 "punpcklbw %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1137 "punpckhbw %%mm7,%%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1138 "punpckhbw %%mm7,%%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1139 "psubw %%mm1, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1140 "psubw %%mm3, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1141 "psubw %%mm0, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1142 "psubw %%mm2, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1143 "pxor %%mm3, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1144 "pxor %%mm1, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1145 "pcmpgtw %%mm4, %%mm3\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1146 "pcmpgtw %%mm5, %%mm1\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1147 "pxor %%mm3, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1148 "pxor %%mm1, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1149 "psubw %%mm3, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1150 "psubw %%mm1, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1151 "paddw %%mm4, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1152 "paddw %%mm5, %%mm6\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1153
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1154 "add %2,%0\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1155
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1156 "movq (%0),%%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1157 "movq 1(%0),%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1158 "movq %%mm4, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1159 "movq %%mm1, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1160 "punpcklbw %%mm7,%%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1161 "punpcklbw %%mm7,%%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1162 "punpckhbw %%mm7,%%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1163 "punpckhbw %%mm7,%%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1164 "psubw %%mm1, %%mm4\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1165 "psubw %%mm3, %%mm5\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1166 "psubw %%mm4, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1167 "psubw %%mm5, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1168 "pxor %%mm3, %%mm3\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1169 "pxor %%mm1, %%mm1\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1170 "pcmpgtw %%mm0, %%mm3\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1171 "pcmpgtw %%mm2, %%mm1\n\t"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1172 "pxor %%mm3, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1173 "pxor %%mm1, %%mm2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1174 "psubw %%mm3, %%mm0\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1175 "psubw %%mm1, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1176 "paddw %%mm0, %%mm2\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1177 "paddw %%mm2, %%mm6\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1178
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1179 "add %2,%0\n"
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1180 "subl $2, %%ecx\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1181 " jnz 1b\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1182
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1183 "movq %%mm6, %%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1184 "punpcklwd %%mm7,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1185 "punpckhwd %%mm7,%%mm6\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1186 "paddd %%mm0, %%mm6\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1187
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1188 "movq %%mm6,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1189 "psrlq $32, %%mm6\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1190 "paddd %%mm6,%%mm0\n"
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1191 "movd %%mm0,%1\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1192 : "+r" (pix1), "=r"(tmp)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1193 : "r" ((long)line_size) , "g" (h-2)
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1194 : "%ecx");
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1195 return tmp + hf_noise8_mmx(pix+8, line_size, h);
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1196 }
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1197
2864
95bac7109ff0 Kill some compiler warnings. Compiled code verified identical after changes.
mru
parents: 2754
diff changeset
1198 static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
95bac7109ff0 Kill some compiler warnings. Compiled code verified identical after changes.
mru
parents: 2754
diff changeset
1199 MpegEncContext *c = p;
2940
8aa244d7c274 use sse16_sse2() in nsse
lorenm
parents: 2922
diff changeset
1200 int score1, score2;
8aa244d7c274 use sse16_sse2() in nsse
lorenm
parents: 2922
diff changeset
1201
8aa244d7c274 use sse16_sse2() in nsse
lorenm
parents: 2922
diff changeset
1202 if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
8aa244d7c274 use sse16_sse2() in nsse
lorenm
parents: 2922
diff changeset
1203 else score1 = sse16_mmx(c, pix1, pix2, line_size, h);
8aa244d7c274 use sse16_sse2() in nsse
lorenm
parents: 2922
diff changeset
1204 score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1205
4001
34fdffe98bd0 Rename ABS macro to FFABS.
diego
parents: 3947
diff changeset
1206 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
34fdffe98bd0 Rename ABS macro to FFABS.
diego
parents: 3947
diff changeset
1207 else return score1 + FFABS(score2)*8;
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1208 }
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1209
2864
95bac7109ff0 Kill some compiler warnings. Compiled code verified identical after changes.
mru
parents: 2754
diff changeset
1210 static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
95bac7109ff0 Kill some compiler warnings. Compiled code verified identical after changes.
mru
parents: 2754
diff changeset
1211 MpegEncContext *c = p;
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1212 int score1= sse8_mmx(c, pix1, pix2, line_size, h);
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1213 int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1214
4001
34fdffe98bd0 Rename ABS macro to FFABS.
diego
parents: 3947
diff changeset
1215 if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
34fdffe98bd0 Rename ABS macro to FFABS.
diego
parents: 3947
diff changeset
1216 else return score1 + FFABS(score2)*8;
2067
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1217 }
f37b6ffc81ed sse8 and nsse in mmx
michael
parents: 2024
diff changeset
1218
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1219 static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1220 int tmp;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1221
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1222 assert( (((int)pix) & 7) == 0);
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1223 assert((line_size &7) ==0);
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1224
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1225 #define SUM(in0, in1, out0, out1) \
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1226 "movq (%0), %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1227 "movq 8(%0), %%mm3\n"\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1228 "add %2,%0\n"\
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1229 "movq %%mm2, " #out0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1230 "movq %%mm3, " #out1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1231 "psubusb " #in0 ", %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1232 "psubusb " #in1 ", %%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1233 "psubusb " #out0 ", " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1234 "psubusb " #out1 ", " #in1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1235 "por %%mm2, " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1236 "por %%mm3, " #in1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1237 "movq " #in0 ", %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1238 "movq " #in1 ", %%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1239 "punpcklbw %%mm7, " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1240 "punpcklbw %%mm7, " #in1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1241 "punpckhbw %%mm7, %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1242 "punpckhbw %%mm7, %%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1243 "paddw " #in1 ", " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1244 "paddw %%mm3, %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1245 "paddw %%mm2, " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1246 "paddw " #in0 ", %%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1247
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1248
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1249 asm volatile (
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1250 "movl %3,%%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1251 "pxor %%mm6,%%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1252 "pxor %%mm7,%%mm7\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1253 "movq (%0),%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1254 "movq 8(%0),%%mm1\n"
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1255 "add %2,%0\n"
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1256 "subl $2, %%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1257 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1258 "1:\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1259
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1260 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1261
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1262 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1263
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1264 "subl $2, %%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1265 "jnz 1b\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1266
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1267 "movq %%mm6,%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1268 "psrlq $32, %%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1269 "paddw %%mm6,%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1270 "movq %%mm0,%%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1271 "psrlq $16, %%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1272 "paddw %%mm6,%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1273 "movd %%mm0,%1\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1274 : "+r" (pix), "=r"(tmp)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1275 : "r" ((long)line_size) , "m" (h)
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1276 : "%ecx");
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1277 return tmp & 0xFFFF;
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1278 }
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1279 #undef SUM
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1280
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1281 static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1282 int tmp;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1283
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1284 assert( (((int)pix) & 7) == 0);
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1285 assert((line_size &7) ==0);
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1286
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1287 #define SUM(in0, in1, out0, out1) \
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1288 "movq (%0), " #out0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1289 "movq 8(%0), " #out1 "\n"\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1290 "add %2,%0\n"\
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1291 "psadbw " #out0 ", " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1292 "psadbw " #out1 ", " #in1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1293 "paddw " #in1 ", " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1294 "paddw " #in0 ", %%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1295
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1296 asm volatile (
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1297 "movl %3,%%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1298 "pxor %%mm6,%%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1299 "pxor %%mm7,%%mm7\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1300 "movq (%0),%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1301 "movq 8(%0),%%mm1\n"
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1302 "add %2,%0\n"
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1303 "subl $2, %%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1304 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1305 "1:\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1306
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1307 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1308
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1309 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1310
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1311 "subl $2, %%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1312 "jnz 1b\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1313
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1314 "movd %%mm6,%1\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1315 : "+r" (pix), "=r"(tmp)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1316 : "r" ((long)line_size) , "m" (h)
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1317 : "%ecx");
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1318 return tmp;
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1319 }
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1320 #undef SUM
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1321
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1322 static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1323 int tmp;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1324
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1325 assert( (((int)pix1) & 7) == 0);
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1326 assert( (((int)pix2) & 7) == 0);
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1327 assert((line_size &7) ==0);
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1328
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1329 #define SUM(in0, in1, out0, out1) \
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1330 "movq (%0),%%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1331 "movq (%1)," #out0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1332 "movq 8(%0),%%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1333 "movq 8(%1)," #out1 "\n"\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1334 "add %3,%0\n"\
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1335 "add %3,%1\n"\
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1336 "psubb " #out0 ", %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1337 "psubb " #out1 ", %%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1338 "pxor %%mm7, %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1339 "pxor %%mm7, %%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1340 "movq %%mm2, " #out0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1341 "movq %%mm3, " #out1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1342 "psubusb " #in0 ", %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1343 "psubusb " #in1 ", %%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1344 "psubusb " #out0 ", " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1345 "psubusb " #out1 ", " #in1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1346 "por %%mm2, " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1347 "por %%mm3, " #in1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1348 "movq " #in0 ", %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1349 "movq " #in1 ", %%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1350 "punpcklbw %%mm7, " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1351 "punpcklbw %%mm7, " #in1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1352 "punpckhbw %%mm7, %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1353 "punpckhbw %%mm7, %%mm3\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1354 "paddw " #in1 ", " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1355 "paddw %%mm3, %%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1356 "paddw %%mm2, " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1357 "paddw " #in0 ", %%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1358
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1359
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1360 asm volatile (
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1361 "movl %4,%%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1362 "pxor %%mm6,%%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1363 "pcmpeqw %%mm7,%%mm7\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1364 "psllw $15, %%mm7\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1365 "packsswb %%mm7, %%mm7\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1366 "movq (%0),%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1367 "movq (%1),%%mm2\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1368 "movq 8(%0),%%mm1\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1369 "movq 8(%1),%%mm3\n"
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1370 "add %3,%0\n"
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1371 "add %3,%1\n"
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1372 "subl $2, %%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1373 "psubb %%mm2, %%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1374 "psubb %%mm3, %%mm1\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1375 "pxor %%mm7, %%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1376 "pxor %%mm7, %%mm1\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1377 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1378 "1:\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1379
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1380 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1381
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1382 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1383
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1384 "subl $2, %%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1385 "jnz 1b\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1386
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1387 "movq %%mm6,%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1388 "psrlq $32, %%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1389 "paddw %%mm6,%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1390 "movq %%mm0,%%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1391 "psrlq $16, %%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1392 "paddw %%mm6,%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1393 "movd %%mm0,%2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1394 : "+r" (pix1), "+r" (pix2), "=r"(tmp)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1395 : "r" ((long)line_size) , "m" (h)
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1396 : "%ecx");
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1397 return tmp & 0x7FFF;
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1398 }
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1399 #undef SUM
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1400
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1401 static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1402 int tmp;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1403
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1404 assert( (((int)pix1) & 7) == 0);
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1405 assert( (((int)pix2) & 7) == 0);
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1406 assert((line_size &7) ==0);
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1407
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1408 #define SUM(in0, in1, out0, out1) \
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1409 "movq (%0)," #out0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1410 "movq (%1),%%mm2\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1411 "movq 8(%0)," #out1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1412 "movq 8(%1),%%mm3\n"\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1413 "add %3,%0\n"\
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1414 "add %3,%1\n"\
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1415 "psubb %%mm2, " #out0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1416 "psubb %%mm3, " #out1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1417 "pxor %%mm7, " #out0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1418 "pxor %%mm7, " #out1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1419 "psadbw " #out0 ", " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1420 "psadbw " #out1 ", " #in1 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1421 "paddw " #in1 ", " #in0 "\n"\
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1422 "paddw " #in0 ", %%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1423
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1424 asm volatile (
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1425 "movl %4,%%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1426 "pxor %%mm6,%%mm6\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1427 "pcmpeqw %%mm7,%%mm7\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1428 "psllw $15, %%mm7\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1429 "packsswb %%mm7, %%mm7\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1430 "movq (%0),%%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1431 "movq (%1),%%mm2\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1432 "movq 8(%0),%%mm1\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1433 "movq 8(%1),%%mm3\n"
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1434 "add %3,%0\n"
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1435 "add %3,%1\n"
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1436 "subl $2, %%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1437 "psubb %%mm2, %%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1438 "psubb %%mm3, %%mm1\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1439 "pxor %%mm7, %%mm0\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1440 "pxor %%mm7, %%mm1\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1441 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1442 "1:\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1443
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1444 SUM(%%mm4, %%mm5, %%mm0, %%mm1)
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1445
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1446 SUM(%%mm0, %%mm1, %%mm4, %%mm5)
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1447
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1448 "subl $2, %%ecx\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1449 "jnz 1b\n"
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1450
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1451 "movd %%mm6,%2\n"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1452 : "+r" (pix1), "+r" (pix2), "=r"(tmp)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1453 : "r" ((long)line_size) , "m" (h)
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1454 : "%ecx");
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1455 return tmp;
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1456 }
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1457 #undef SUM
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
1458
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1459 static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1460 long i=0;
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1461 asm volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1462 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1463 "movq (%2, %0), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1464 "movq (%1, %0), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1465 "psubb %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1466 "movq %%mm1, (%3, %0) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1467 "movq 8(%2, %0), %%mm0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1468 "movq 8(%1, %0), %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1469 "psubb %%mm0, %%mm1 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1470 "movq %%mm1, 8(%3, %0) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1471 "add $16, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1472 "cmp %4, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1473 " jb 1b \n\t"
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1474 : "+r" (i)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1475 : "r"(src1), "r"(src2), "r"(dst), "r"((long)w-15)
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1476 );
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1477 for(; i<w; i++)
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1478 dst[i+0] = src1[i+0]-src2[i+0];
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1479 }
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1480
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1481 static void sub_hfyu_median_prediction_mmx2(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top){
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1482 long i=0;
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1483 uint8_t l, lt;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1484
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1485 asm volatile(
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1486 "1: \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1487 "movq -1(%1, %0), %%mm0 \n\t" // LT
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1488 "movq (%1, %0), %%mm1 \n\t" // T
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1489 "movq -1(%2, %0), %%mm2 \n\t" // L
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1490 "movq (%2, %0), %%mm3 \n\t" // X
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1491 "movq %%mm2, %%mm4 \n\t" // L
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1492 "psubb %%mm0, %%mm2 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1493 "paddb %%mm1, %%mm2 \n\t" // L + T - LT
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1494 "movq %%mm4, %%mm5 \n\t" // L
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1495 "pmaxub %%mm1, %%mm4 \n\t" // max(T, L)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1496 "pminub %%mm5, %%mm1 \n\t" // min(T, L)
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1497 "pminub %%mm2, %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1498 "pmaxub %%mm1, %%mm4 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1499 "psubb %%mm4, %%mm3 \n\t" // dst - pred
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1500 "movq %%mm3, (%3, %0) \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1501 "add $8, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1502 "cmp %4, %0 \n\t"
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1503 " jb 1b \n\t"
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1504 : "+r" (i)
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
1505 : "r"(src1), "r"(src2), "r"(dst), "r"((long)w)
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1506 );
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1507
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1508 l= *left;
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1509 lt= *left_top;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1510
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1511 dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1512
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1513 *left_top= src1[w-1];
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1514 *left = src2[w-1];
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1515 }
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
1516
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1517 #define DIFF_PIXELS_1(m,a,t,p1,p2)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1518 "mov"#m" "#p1", "#a" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1519 "mov"#m" "#p2", "#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1520 "punpcklbw "#a", "#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1521 "punpcklbw "#a", "#a" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1522 "psubw "#t", "#a" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1523
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1524 #define DIFF_PIXELS_8(m0,m1,mm,p1,p2,stride,temp) {\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1525 uint8_t *p1b=p1, *p2b=p2;\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1526 asm volatile(\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1527 DIFF_PIXELS_1(m0, mm##0, mm##7, (%1), (%2))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1528 DIFF_PIXELS_1(m0, mm##1, mm##7, (%1,%3), (%2,%3))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1529 DIFF_PIXELS_1(m0, mm##2, mm##7, (%1,%3,2), (%2,%3,2))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1530 "add %4, %1 \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1531 "add %4, %2 \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1532 DIFF_PIXELS_1(m0, mm##3, mm##7, (%1), (%2))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1533 DIFF_PIXELS_1(m0, mm##4, mm##7, (%1,%3), (%2,%3))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1534 DIFF_PIXELS_1(m0, mm##5, mm##7, (%1,%3,2), (%2,%3,2))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1535 DIFF_PIXELS_1(m0, mm##6, mm##7, (%1,%4), (%2,%4))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1536 "mov"#m1" "#mm"0, %0 \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1537 DIFF_PIXELS_1(m0, mm##7, mm##0, (%1,%3,4), (%2,%3,4))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1538 "mov"#m1" %0, "#mm"0 \n\t"\
5912
f75ee7ea171b tring to workaround gcc 2.95 bug which causes random failures
michael
parents: 5737
diff changeset
1539 : "+m"(temp), "+r"(p1b), "+r"(p2b)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1540 : "r"((long)stride), "r"((long)stride*3)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1541 );\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1542 }
5912
f75ee7ea171b tring to workaround gcc 2.95 bug which causes random failures
michael
parents: 5737
diff changeset
1543 //the "+m"(temp) is needed as gcc 2.95 sometimes fails to compile "=m"(temp)
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1544
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1545 #define DIFF_PIXELS_4x8(p1,p2,stride,temp) DIFF_PIXELS_8(d, q, %%mm, p1, p2, stride, temp)
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1546 #define DIFF_PIXELS_8x8(p1,p2,stride,temp) DIFF_PIXELS_8(q, dqa, %%xmm, p1, p2, stride, temp)
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1547
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1548 #ifdef ARCH_X86_64
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1549 // permutes 01234567 -> 05736421
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1550 #define TRANSPOSE8(a,b,c,d,e,f,g,h,t)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1551 SBUTTERFLY(a,b,%%xmm8,wd,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1552 SBUTTERFLY(c,d,b,wd,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1553 SBUTTERFLY(e,f,d,wd,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1554 SBUTTERFLY(g,h,f,wd,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1555 SBUTTERFLY(a,c,h,dq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1556 SBUTTERFLY(%%xmm8,b,c,dq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1557 SBUTTERFLY(e,g,b,dq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1558 SBUTTERFLY(d,f,g,dq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1559 SBUTTERFLY(a,e,f,qdq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1560 SBUTTERFLY(%%xmm8,d,e,qdq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1561 SBUTTERFLY(h,b,d,qdq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1562 SBUTTERFLY(c,g,b,qdq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1563 "movdqa %%xmm8, "#g" \n\t"
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1564 #else
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1565 #define TRANSPOSE8(a,b,c,d,e,f,g,h,t)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1566 "movdqa "#h", "#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1567 SBUTTERFLY(a,b,h,wd,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1568 "movdqa "#h", 16"#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1569 "movdqa "#t", "#h" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1570 SBUTTERFLY(c,d,b,wd,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1571 SBUTTERFLY(e,f,d,wd,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1572 SBUTTERFLY(g,h,f,wd,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1573 SBUTTERFLY(a,c,h,dq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1574 "movdqa "#h", "#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1575 "movdqa 16"#t", "#h" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1576 SBUTTERFLY(h,b,c,dq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1577 SBUTTERFLY(e,g,b,dq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1578 SBUTTERFLY(d,f,g,dq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1579 SBUTTERFLY(a,e,f,qdq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1580 SBUTTERFLY(h,d,e,qdq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1581 "movdqa "#h", 16"#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1582 "movdqa "#t", "#h" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1583 SBUTTERFLY(h,b,d,qdq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1584 SBUTTERFLY(c,g,b,qdq,dqa)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1585 "movdqa 16"#t", "#g" \n\t"
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1586 #endif
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1587
1153
2725c8eb3c81 faster hadamard transform
michaelni
parents: 1122
diff changeset
1588 #define LBUTTERFLY2(a1,b1,a2,b2)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1589 "paddw " #b1 ", " #a1 " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1590 "paddw " #b2 ", " #a2 " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1591 "paddw " #b1 ", " #b1 " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1592 "paddw " #b2 ", " #b2 " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1593 "psubw " #a1 ", " #b1 " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1594 "psubw " #a2 ", " #b2 " \n\t"
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1595
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1596 #define HADAMARD8(m0, m1, m2, m3, m4, m5, m6, m7)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1597 LBUTTERFLY2(m0, m1, m2, m3)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1598 LBUTTERFLY2(m4, m5, m6, m7)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1599 LBUTTERFLY2(m0, m2, m1, m3)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1600 LBUTTERFLY2(m4, m6, m5, m7)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1601 LBUTTERFLY2(m0, m4, m1, m5)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1602 LBUTTERFLY2(m2, m6, m3, m7)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1603
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1604 #define HADAMARD48 HADAMARD8(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm6, %%mm7)
936
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
1605
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1606 #define MMABS_MMX(a,z)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1607 "pxor " #z ", " #z " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1608 "pcmpgtw " #a ", " #z " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1609 "pxor " #z ", " #a " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1610 "psubw " #z ", " #a " \n\t"
936
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
1611
1153
2725c8eb3c81 faster hadamard transform
michaelni
parents: 1122
diff changeset
1612 #define MMABS_MMX2(a,z)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1613 "pxor " #z ", " #z " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1614 "psubw " #a ", " #z " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1615 "pmaxsw " #z ", " #a " \n\t"
1153
2725c8eb3c81 faster hadamard transform
michaelni
parents: 1122
diff changeset
1616
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1617 #define MMABS_SSSE3(a,z)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1618 "pabsw " #a ", " #a " \n\t"
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1619
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1620 #define MMABS_SUM(a,z, sum)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1621 MMABS(a,z)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1622 "paddusw " #a ", " #sum " \n\t"
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1623
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1624 #define MMABS_SUM_8x8_NOSPILL\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1625 MMABS(%%xmm0, %%xmm8)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1626 MMABS(%%xmm1, %%xmm9)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1627 MMABS_SUM(%%xmm2, %%xmm8, %%xmm0)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1628 MMABS_SUM(%%xmm3, %%xmm9, %%xmm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1629 MMABS_SUM(%%xmm4, %%xmm8, %%xmm0)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1630 MMABS_SUM(%%xmm5, %%xmm9, %%xmm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1631 MMABS_SUM(%%xmm6, %%xmm8, %%xmm0)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1632 MMABS_SUM(%%xmm7, %%xmm9, %%xmm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1633 "paddusw %%xmm1, %%xmm0 \n\t"
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1634
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1635 #ifdef ARCH_X86_64
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1636 #define MMABS_SUM_8x8_SSE2 MMABS_SUM_8x8_NOSPILL
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1637 #else
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1638 #define MMABS_SUM_8x8_SSE2\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1639 "movdqa %%xmm7, (%1) \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1640 MMABS(%%xmm0, %%xmm7)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1641 MMABS(%%xmm1, %%xmm7)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1642 MMABS_SUM(%%xmm2, %%xmm7, %%xmm0)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1643 MMABS_SUM(%%xmm3, %%xmm7, %%xmm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1644 MMABS_SUM(%%xmm4, %%xmm7, %%xmm0)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1645 MMABS_SUM(%%xmm5, %%xmm7, %%xmm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1646 MMABS_SUM(%%xmm6, %%xmm7, %%xmm0)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1647 "movdqa (%1), %%xmm2 \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1648 MMABS_SUM(%%xmm2, %%xmm7, %%xmm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1649 "paddusw %%xmm1, %%xmm0 \n\t"
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1650 #endif
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
1651
936
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
1652 #define LOAD4(o, a, b, c, d)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1653 "movq "#o"(%1), "#a" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1654 "movq "#o"+8(%1), "#b" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1655 "movq "#o"+16(%1), "#c" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1656 "movq "#o"+24(%1), "#d" \n\t"\
936
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
1657
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
1658 #define STORE4(o, a, b, c, d)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1659 "movq "#a", "#o"(%1) \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1660 "movq "#b", "#o"+8(%1) \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1661 "movq "#c", "#o"+16(%1) \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1662 "movq "#d", "#o"+24(%1) \n\t"\
936
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
1663
4988
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1664 /* FIXME: HSUM_* saturates at 64k, while an 8x8 hadamard or dct block can get up to
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1665 * about 100k on extreme inputs. But that's very unlikely to occur in natural video,
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1666 * and it's even more unlikely to not have any alternative mvs/modes with lower cost. */
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1667 #define HSUM_MMX(a, t, dst)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1668 "movq "#a", "#t" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1669 "psrlq $32, "#a" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1670 "paddusw "#t", "#a" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1671 "movq "#a", "#t" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1672 "psrlq $16, "#a" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1673 "paddusw "#t", "#a" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1674 "movd "#a", "#dst" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1675
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1676 #define HSUM_MMX2(a, t, dst)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1677 "pshufw $0x0E, "#a", "#t" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1678 "paddusw "#t", "#a" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1679 "pshufw $0x01, "#a", "#t" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1680 "paddusw "#t", "#a" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1681 "movd "#a", "#dst" \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1682
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1683 #define HSUM_SSE2(a, t, dst)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1684 "movhlps "#a", "#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1685 "paddusw "#t", "#a" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1686 "pshuflw $0x0E, "#a", "#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1687 "paddusw "#t", "#a" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1688 "pshuflw $0x01, "#a", "#t" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1689 "paddusw "#t", "#a" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1690 "movd "#a", "#dst" \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1691
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1692 #define HADAMARD8_DIFF_MMX(cpu) \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1693 static int hadamard8_diff_##cpu(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1694 DECLARE_ALIGNED_8(uint64_t, temp[13]);\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1695 int sum;\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1696 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1697 assert(h==8);\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1698 \
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1699 DIFF_PIXELS_4x8(src1, src2, stride, temp[0]);\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1700 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1701 asm volatile(\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1702 HADAMARD48\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1703 \
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1704 "movq %%mm7, 96(%1) \n\t"\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1705 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1706 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1707 STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1708 \
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1709 "movq 96(%1), %%mm7 \n\t"\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1710 TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1711 STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1712 \
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1713 : "=r" (sum)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1714 : "r"(temp)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1715 );\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1716 \
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1717 DIFF_PIXELS_4x8(src1+4, src2+4, stride, temp[4]);\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1718 \
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1719 asm volatile(\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1720 HADAMARD48\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1721 \
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1722 "movq %%mm7, 96(%1) \n\t"\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1723 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1724 TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1725 STORE4(32, %%mm0, %%mm3, %%mm7, %%mm2)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1726 \
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1727 "movq 96(%1), %%mm7 \n\t"\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1728 TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1729 "movq %%mm7, %%mm5 \n\t"/*FIXME remove*/\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1730 "movq %%mm6, %%mm7 \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1731 "movq %%mm0, %%mm6 \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1732 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1733 LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1734 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1735 HADAMARD48\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1736 "movq %%mm7, 64(%1) \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1737 MMABS(%%mm0, %%mm7)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1738 MMABS(%%mm1, %%mm7)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1739 MMABS_SUM(%%mm2, %%mm7, %%mm0)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1740 MMABS_SUM(%%mm3, %%mm7, %%mm1)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1741 MMABS_SUM(%%mm4, %%mm7, %%mm0)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1742 MMABS_SUM(%%mm5, %%mm7, %%mm1)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1743 MMABS_SUM(%%mm6, %%mm7, %%mm0)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1744 "movq 64(%1), %%mm2 \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1745 MMABS_SUM(%%mm2, %%mm7, %%mm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1746 "paddusw %%mm1, %%mm0 \n\t"\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1747 "movq %%mm0, 64(%1) \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1748 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1749 LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1750 LOAD4(32, %%mm4, %%mm5, %%mm6, %%mm7)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1751 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1752 HADAMARD48\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1753 "movq %%mm7, (%1) \n\t"\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1754 MMABS(%%mm0, %%mm7)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1755 MMABS(%%mm1, %%mm7)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1756 MMABS_SUM(%%mm2, %%mm7, %%mm0)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1757 MMABS_SUM(%%mm3, %%mm7, %%mm1)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1758 MMABS_SUM(%%mm4, %%mm7, %%mm0)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1759 MMABS_SUM(%%mm5, %%mm7, %%mm1)\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1760 MMABS_SUM(%%mm6, %%mm7, %%mm0)\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1761 "movq (%1), %%mm2 \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1762 MMABS_SUM(%%mm2, %%mm7, %%mm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1763 "paddusw 64(%1), %%mm0 \n\t"\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1764 "paddusw %%mm1, %%mm0 \n\t"\
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1765 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1766 HSUM(%%mm0, %%mm1, %0)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1767 \
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1768 : "=r" (sum)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1769 : "r"(temp)\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1770 );\
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1771 return sum&0xFFFF;\
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1772 }\
6056
558c1fd0ee72 Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents: 6030
diff changeset
1773 WRAPPER8_16_SQ(hadamard8_diff_##cpu, hadamard8_diff16_##cpu)
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1774
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1775 #define HADAMARD8_DIFF_SSE2(cpu) \
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1776 static int hadamard8_diff_##cpu(void *s, uint8_t *src1, uint8_t *src2, int stride, int h){\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1777 DECLARE_ALIGNED_16(uint64_t, temp[4]);\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1778 int sum;\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1779 \
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1780 assert(h==8);\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1781 \
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1782 DIFF_PIXELS_8x8(src1, src2, stride, temp[0]);\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1783 \
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1784 asm volatile(\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1785 HADAMARD8(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1786 TRANSPOSE8(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7, (%1))\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1787 HADAMARD8(%%xmm0, %%xmm5, %%xmm7, %%xmm3, %%xmm6, %%xmm4, %%xmm2, %%xmm1)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1788 MMABS_SUM_8x8\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1789 HSUM_SSE2(%%xmm0, %%xmm1, %0)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1790 : "=r" (sum)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1791 : "r"(temp)\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1792 );\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1793 return sum&0xFFFF;\
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1794 }\
6056
558c1fd0ee72 Fix typo in macro name: WARPER8_16_SQ --> WRAPPER8_16_SQ.
diego
parents: 6030
diff changeset
1795 WRAPPER8_16_SQ(hadamard8_diff_##cpu, hadamard8_diff16_##cpu)
936
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
1796
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1797 #define MMABS(a,z) MMABS_MMX(a,z)
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1798 #define HSUM(a,t,dst) HSUM_MMX(a,t,dst)
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1799 HADAMARD8_DIFF_MMX(mmx)
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1800 #undef MMABS
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1801 #undef HSUM
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1802
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1803 #define MMABS(a,z) MMABS_MMX2(a,z)
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1804 #define MMABS_SUM_8x8 MMABS_SUM_8x8_SSE2
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1805 #define HSUM(a,t,dst) HSUM_MMX2(a,t,dst)
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1806 HADAMARD8_DIFF_MMX(mmx2)
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1807 HADAMARD8_DIFF_SSE2(sse2)
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1808 #undef MMABS
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1809 #undef MMABS_SUM_8x8
4946
c1fb4544bd59 cosmetics: remove code duplication in hadamard8_diff_mmx
lorenm
parents: 4939
diff changeset
1810 #undef HSUM
1153
2725c8eb3c81 faster hadamard transform
michaelni
parents: 1122
diff changeset
1811
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1812 #ifdef HAVE_SSSE3
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1813 #define MMABS(a,z) MMABS_SSSE3(a,z)
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1814 #define MMABS_SUM_8x8 MMABS_SUM_8x8_NOSPILL
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1815 HADAMARD8_DIFF_SSE2(ssse3)
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1816 #undef MMABS
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1817 #undef MMABS_SUM_8x8
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
1818 #endif
4749
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1819
4988
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1820 #define DCT_SAD4(m,mm,o)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1821 "mov"#m" "#o"+ 0(%1), "#mm"2 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1822 "mov"#m" "#o"+16(%1), "#mm"3 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1823 "mov"#m" "#o"+32(%1), "#mm"4 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1824 "mov"#m" "#o"+48(%1), "#mm"5 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1825 MMABS_SUM(mm##2, mm##6, mm##0)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1826 MMABS_SUM(mm##3, mm##7, mm##1)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1827 MMABS_SUM(mm##4, mm##6, mm##0)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1828 MMABS_SUM(mm##5, mm##7, mm##1)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1829
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1830 #define DCT_SAD_MMX\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1831 "pxor %%mm0, %%mm0 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1832 "pxor %%mm1, %%mm1 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1833 DCT_SAD4(q, %%mm, 0)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1834 DCT_SAD4(q, %%mm, 8)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1835 DCT_SAD4(q, %%mm, 64)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1836 DCT_SAD4(q, %%mm, 72)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1837 "paddusw %%mm1, %%mm0 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1838 HSUM(%%mm0, %%mm1, %0)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1839
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1840 #define DCT_SAD_SSE2\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1841 "pxor %%xmm0, %%xmm0 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1842 "pxor %%xmm1, %%xmm1 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1843 DCT_SAD4(dqa, %%xmm, 0)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1844 DCT_SAD4(dqa, %%xmm, 64)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1845 "paddusw %%xmm1, %%xmm0 \n\t"\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1846 HSUM(%%xmm0, %%xmm1, %0)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1847
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1848 #define DCT_SAD_FUNC(cpu) \
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1849 static int sum_abs_dctelem_##cpu(DCTELEM *block){\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1850 int sum;\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1851 asm volatile(\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1852 DCT_SAD\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1853 :"=r"(sum)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1854 :"r"(block)\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1855 );\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1856 return sum&0xFFFF;\
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1857 }
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1858
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1859 #define DCT_SAD DCT_SAD_MMX
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1860 #define HSUM(a,t,dst) HSUM_MMX(a,t,dst)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1861 #define MMABS(a,z) MMABS_MMX(a,z)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1862 DCT_SAD_FUNC(mmx)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1863 #undef MMABS
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1864 #undef HSUM
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1865
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1866 #define HSUM(a,t,dst) HSUM_MMX2(a,t,dst)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1867 #define MMABS(a,z) MMABS_MMX2(a,z)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1868 DCT_SAD_FUNC(mmx2)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1869 #undef HSUM
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1870 #undef DCT_SAD
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1871
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1872 #define DCT_SAD DCT_SAD_SSE2
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1873 #define HSUM(a,t,dst) HSUM_SSE2(a,t,dst)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1874 DCT_SAD_FUNC(sse2)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1875 #undef MMABS
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1876
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1877 #ifdef HAVE_SSSE3
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1878 #define MMABS(a,z) MMABS_SSSE3(a,z)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1879 DCT_SAD_FUNC(ssse3)
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1880 #undef MMABS
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1881 #endif
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1882 #undef HSUM
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1883 #undef DCT_SAD
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
1884
5255
669a97223dc7 make arguments to ssd_int8_vs_int16() const
mru
parents: 5049
diff changeset
1885 static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int size){
4749
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1886 int sum;
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1887 long i=size;
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1888 asm volatile(
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1889 "pxor %%mm4, %%mm4 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1890 "1: \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1891 "sub $8, %0 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1892 "movq (%2,%0), %%mm2 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1893 "movq (%3,%0,2), %%mm0 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1894 "movq 8(%3,%0,2), %%mm1 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1895 "punpckhbw %%mm2, %%mm3 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1896 "punpcklbw %%mm2, %%mm2 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1897 "psraw $8, %%mm3 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1898 "psraw $8, %%mm2 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1899 "psubw %%mm3, %%mm1 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1900 "psubw %%mm2, %%mm0 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1901 "pmaddwd %%mm1, %%mm1 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1902 "pmaddwd %%mm0, %%mm0 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1903 "paddd %%mm1, %%mm4 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1904 "paddd %%mm0, %%mm4 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1905 "jg 1b \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1906 "movq %%mm4, %%mm3 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1907 "psrlq $32, %%mm3 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1908 "paddd %%mm3, %%mm4 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1909 "movd %%mm4, %1 \n"
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1910 :"+r"(i), "=r"(sum)
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1911 :"r"(pix1), "r"(pix2)
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1912 );
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1913 return sum;
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1914 }
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
1915
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
1916 #endif //CONFIG_ENCODERS
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
1917
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
1918 #define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
1919 #define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
1920
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
1921 #define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1922 "paddw " #m4 ", " #m3 " \n\t" /* x1 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1923 "movq "MANGLE(ff_pw_20)", %%mm4 \n\t" /* 20 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1924 "pmullw " #m3 ", %%mm4 \n\t" /* 20x1 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1925 "movq "#in7", " #m3 " \n\t" /* d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1926 "movq "#in0", %%mm5 \n\t" /* D */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1927 "paddw " #m3 ", %%mm5 \n\t" /* x4 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1928 "psubw %%mm5, %%mm4 \n\t" /* 20x1 - x4 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1929 "movq "#in1", %%mm5 \n\t" /* C */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1930 "movq "#in2", %%mm6 \n\t" /* B */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1931 "paddw " #m6 ", %%mm5 \n\t" /* x3 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1932 "paddw " #m5 ", %%mm6 \n\t" /* x2 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1933 "paddw %%mm6, %%mm6 \n\t" /* 2x2 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1934 "psubw %%mm6, %%mm5 \n\t" /* -2x2 + x3 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1935 "pmullw "MANGLE(ff_pw_3)", %%mm5 \n\t" /* -6x2 + 3x3 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1936 "paddw " #rnd ", %%mm4 \n\t" /* x2 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1937 "paddw %%mm4, %%mm5 \n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1938 "psraw $5, %%mm5 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1939 "packuswb %%mm5, %%mm5 \n\t"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
1940 OP(%%mm5, out, %%mm7, d)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
1941
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
1942 #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
1057
bb5de8a59da8 * static,const,compiler warning cleanup
kabi
parents: 997
diff changeset
1943 static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
1944 uint64_t temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
1945 \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
1946 asm volatile(\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1947 "pxor %%mm7, %%mm7 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1948 "1: \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1949 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1950 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1951 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1952 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1953 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1954 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1955 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1956 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1957 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1958 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1959 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1960 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1961 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1962 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1963 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1964 "paddw %%mm3, %%mm5 \n\t" /* b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1965 "paddw %%mm2, %%mm6 \n\t" /* c */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1966 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1967 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1968 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1969 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1970 "paddw %%mm4, %%mm0 \n\t" /* a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1971 "paddw %%mm1, %%mm5 \n\t" /* d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1972 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1973 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1974 "paddw %6, %%mm6 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1975 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1976 "psraw $5, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1977 "movq %%mm0, %5 \n\t"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
1978 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
1979 \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1980 "movq 5(%0), %%mm0 \n\t" /* FGHIJKLM */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1981 "movq %%mm0, %%mm5 \n\t" /* FGHIJKLM */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1982 "movq %%mm0, %%mm6 \n\t" /* FGHIJKLM */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1983 "psrlq $8, %%mm0 \n\t" /* GHIJKLM0 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1984 "psrlq $16, %%mm5 \n\t" /* HIJKLM00 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1985 "punpcklbw %%mm7, %%mm0 \n\t" /* 0G0H0I0J */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1986 "punpcklbw %%mm7, %%mm5 \n\t" /* 0H0I0J0K */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1987 "paddw %%mm0, %%mm2 \n\t" /* b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1988 "paddw %%mm5, %%mm3 \n\t" /* c */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1989 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1990 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1991 "movq %%mm6, %%mm2 \n\t" /* FGHIJKLM */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1992 "psrlq $24, %%mm6 \n\t" /* IJKLM000 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1993 "punpcklbw %%mm7, %%mm2 \n\t" /* 0F0G0H0I */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1994 "punpcklbw %%mm7, %%mm6 \n\t" /* 0I0J0K0L */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1995 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1996 "paddw %%mm2, %%mm1 \n\t" /* a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1997 "paddw %%mm6, %%mm4 \n\t" /* d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1998 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
1999 "psubw %%mm4, %%mm3 \n\t" /* - 6b +3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2000 "paddw %6, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2001 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b +3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2002 "psraw $5, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2003 "movq %5, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2004 "packuswb %%mm3, %%mm1 \n\t"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2005 OP_MMX2(%%mm1, (%1),%%mm4, q)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2006 /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2007 \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2008 "movq 9(%0), %%mm1 \n\t" /* JKLMNOPQ */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2009 "movq %%mm1, %%mm4 \n\t" /* JKLMNOPQ */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2010 "movq %%mm1, %%mm3 \n\t" /* JKLMNOPQ */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2011 "psrlq $8, %%mm1 \n\t" /* KLMNOPQ0 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2012 "psrlq $16, %%mm4 \n\t" /* LMNOPQ00 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2013 "punpcklbw %%mm7, %%mm1 \n\t" /* 0K0L0M0N */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2014 "punpcklbw %%mm7, %%mm4 \n\t" /* 0L0M0N0O */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2015 "paddw %%mm1, %%mm5 \n\t" /* b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2016 "paddw %%mm4, %%mm0 \n\t" /* c */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2017 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2018 "psubw %%mm5, %%mm0 \n\t" /* c - 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2019 "movq %%mm3, %%mm5 \n\t" /* JKLMNOPQ */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2020 "psrlq $24, %%mm3 \n\t" /* MNOPQ000 */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2021 "pmullw "MANGLE(ff_pw_3)", %%mm0 \n\t" /* 3c - 6b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2022 "punpcklbw %%mm7, %%mm3 \n\t" /* 0M0N0O0P */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2023 "paddw %%mm3, %%mm2 \n\t" /* d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2024 "psubw %%mm2, %%mm0 \n\t" /* -6b + 3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2025 "movq %%mm5, %%mm2 \n\t" /* JKLMNOPQ */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2026 "punpcklbw %%mm7, %%mm2 \n\t" /* 0J0K0L0M */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2027 "punpckhbw %%mm7, %%mm5 \n\t" /* 0N0O0P0Q */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2028 "paddw %%mm2, %%mm6 \n\t" /* a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2029 "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" /* 20a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2030 "paddw %6, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2031 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2032 "psraw $5, %%mm0 \n\t"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2033 /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2034 \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2035 "paddw %%mm5, %%mm3 \n\t" /* a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2036 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0O0P0Q0Q */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2037 "paddw %%mm4, %%mm6 \n\t" /* b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2038 "pshufw $0xBE, %%mm5, %%mm4 \n\t" /* 0P0Q0Q0P */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2039 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0Q0Q0P0O */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2040 "paddw %%mm1, %%mm4 \n\t" /* c */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2041 "paddw %%mm2, %%mm5 \n\t" /* d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2042 "paddw %%mm6, %%mm6 \n\t" /* 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2043 "psubw %%mm6, %%mm4 \n\t" /* c - 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2044 "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" /* 20a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2045 "pmullw "MANGLE(ff_pw_3)", %%mm4 \n\t" /* 3c - 6b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2046 "psubw %%mm5, %%mm3 \n\t" /* -6b + 3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2047 "paddw %6, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2048 "paddw %%mm3, %%mm4 \n\t" /* 20a - 6b + 3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2049 "psraw $5, %%mm4 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2050 "packuswb %%mm4, %%mm0 \n\t"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2051 OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2052 \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2053 "add %3, %0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2054 "add %4, %1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2055 "decl %2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2056 " jnz 1b \n\t"\
967
274b518c4ecb PIC / ebx fix
michaelni
parents: 966
diff changeset
2057 : "+a"(src), "+c"(dst), "+m"(h)\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
2058 : "d"((long)srcStride), "S"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
966
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2059 : "memory"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2060 );\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2061 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2062 \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2063 static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2064 int i;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2065 int16_t temp[16];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2066 /* quick HACK, XXX FIXME MUST be optimized */\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2067 for(i=0; i<h; i++)\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2068 {\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2069 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2070 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2071 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2072 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2073 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2074 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2075 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2076 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2077 temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2078 temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2079 temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2080 temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2081 temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2082 temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2083 temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2084 temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2085 asm volatile(\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2086 "movq (%0), %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2087 "movq 8(%0), %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2088 "paddw %2, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2089 "paddw %2, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2090 "psraw $5, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2091 "psraw $5, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2092 "packuswb %%mm1, %%mm0 \n\t"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2093 OP_3DNOW(%%mm0, (%1), %%mm1, q)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2094 "movq 16(%0), %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2095 "movq 24(%0), %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2096 "paddw %2, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2097 "paddw %2, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2098 "psraw $5, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2099 "psraw $5, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2100 "packuswb %%mm1, %%mm0 \n\t"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2101 OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2102 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
966
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2103 : "memory"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2104 );\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2105 dst+=dstStride;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2106 src+=srcStride;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2107 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2108 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2109 \
1057
bb5de8a59da8 * static,const,compiler warning cleanup
kabi
parents: 997
diff changeset
2110 static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2111 uint64_t temp;\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2112 \
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2113 asm volatile(\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2114 "pxor %%mm7, %%mm7 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2115 "1: \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2116 "movq (%0), %%mm0 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2117 "movq %%mm0, %%mm1 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2118 "movq %%mm0, %%mm2 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2119 "punpcklbw %%mm7, %%mm0 \n\t" /* 0A0B0C0D */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2120 "punpckhbw %%mm7, %%mm1 \n\t" /* 0E0F0G0H */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2121 "pshufw $0x90, %%mm0, %%mm5 \n\t" /* 0A0A0B0C */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2122 "pshufw $0x41, %%mm0, %%mm6 \n\t" /* 0B0A0A0B */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2123 "movq %%mm2, %%mm3 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2124 "movq %%mm2, %%mm4 \n\t" /* ABCDEFGH */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2125 "psllq $8, %%mm2 \n\t" /* 0ABCDEFG */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2126 "psllq $16, %%mm3 \n\t" /* 00ABCDEF */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2127 "psllq $24, %%mm4 \n\t" /* 000ABCDE */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2128 "punpckhbw %%mm7, %%mm2 \n\t" /* 0D0E0F0G */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2129 "punpckhbw %%mm7, %%mm3 \n\t" /* 0C0D0E0F */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2130 "punpckhbw %%mm7, %%mm4 \n\t" /* 0B0C0D0E */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2131 "paddw %%mm3, %%mm5 \n\t" /* b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2132 "paddw %%mm2, %%mm6 \n\t" /* c */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2133 "paddw %%mm5, %%mm5 \n\t" /* 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2134 "psubw %%mm5, %%mm6 \n\t" /* c - 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2135 "pshufw $0x06, %%mm0, %%mm5 \n\t" /* 0C0B0A0A */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2136 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" /* 3c - 6b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2137 "paddw %%mm4, %%mm0 \n\t" /* a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2138 "paddw %%mm1, %%mm5 \n\t" /* d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2139 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" /* 20a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2140 "psubw %%mm5, %%mm0 \n\t" /* 20a - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2141 "paddw %6, %%mm6 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2142 "paddw %%mm6, %%mm0 \n\t" /* 20a - 6b + 3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2143 "psraw $5, %%mm0 \n\t"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2144 /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2145 \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2146 "movd 5(%0), %%mm5 \n\t" /* FGHI */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2147 "punpcklbw %%mm7, %%mm5 \n\t" /* 0F0G0H0I */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2148 "pshufw $0xF9, %%mm5, %%mm6 \n\t" /* 0G0H0I0I */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2149 "paddw %%mm5, %%mm1 \n\t" /* a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2150 "paddw %%mm6, %%mm2 \n\t" /* b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2151 "pshufw $0xBE, %%mm5, %%mm6 \n\t" /* 0H0I0I0H */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2152 "pshufw $0x6F, %%mm5, %%mm5 \n\t" /* 0I0I0H0G */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2153 "paddw %%mm6, %%mm3 \n\t" /* c */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2154 "paddw %%mm5, %%mm4 \n\t" /* d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2155 "paddw %%mm2, %%mm2 \n\t" /* 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2156 "psubw %%mm2, %%mm3 \n\t" /* c - 2b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2157 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" /* 20a */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2158 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" /* 3c - 6b */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2159 "psubw %%mm4, %%mm3 \n\t" /* -6b + 3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2160 "paddw %6, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2161 "paddw %%mm1, %%mm3 \n\t" /* 20a - 6b + 3c - d */\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2162 "psraw $5, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2163 "packuswb %%mm3, %%mm0 \n\t"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2164 OP_MMX2(%%mm0, (%1), %%mm4, q)\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2165 \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2166 "add %3, %0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2167 "add %4, %1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2168 "decl %2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2169 " jnz 1b \n\t"\
967
274b518c4ecb PIC / ebx fix
michaelni
parents: 966
diff changeset
2170 : "+a"(src), "+c"(dst), "+m"(h)\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
2171 : "S"((long)srcStride), "D"((long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
966
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2172 : "memory"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2173 );\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2174 }\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2175 \
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2176 static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2177 int i;\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2178 int16_t temp[8];\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2179 /* quick HACK, XXX FIXME MUST be optimized */\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2180 for(i=0; i<h; i++)\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2181 {\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2182 temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2183 temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2184 temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2185 temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2186 temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2187 temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2188 temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2189 temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2190 asm volatile(\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2191 "movq (%0), %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2192 "movq 8(%0), %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2193 "paddw %2, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2194 "paddw %2, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2195 "psraw $5, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2196 "psraw $5, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2197 "packuswb %%mm1, %%mm0 \n\t"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2198 OP_3DNOW(%%mm0, (%1), %%mm1, q)\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2199 :: "r"(temp), "r"(dst), "m"(ROUNDER)\
966
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2200 :"memory"\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2201 );\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2202 dst+=dstStride;\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2203 src+=srcStride;\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2204 }\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2205 }
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2206
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2207 #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2208 \
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2209 static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2210 uint64_t temp[17*4];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2211 uint64_t *temp_ptr= temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2212 int count= 17;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2213 \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2214 /*FIXME unroll */\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2215 asm volatile(\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2216 "pxor %%mm7, %%mm7 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2217 "1: \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2218 "movq (%0), %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2219 "movq (%0), %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2220 "movq 8(%0), %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2221 "movq 8(%0), %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2222 "punpcklbw %%mm7, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2223 "punpckhbw %%mm7, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2224 "punpcklbw %%mm7, %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2225 "punpckhbw %%mm7, %%mm3 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2226 "movq %%mm0, (%1) \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2227 "movq %%mm1, 17*8(%1) \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2228 "movq %%mm2, 2*17*8(%1) \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2229 "movq %%mm3, 3*17*8(%1) \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2230 "add $8, %1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2231 "add %3, %0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2232 "decl %2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2233 " jnz 1b \n\t"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2234 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
2235 : "r" ((long)srcStride)\
966
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2236 : "memory"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2237 );\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2238 \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2239 temp_ptr= temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2240 count=4;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2241 \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2242 /*FIXME reorder for speed */\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2243 asm volatile(\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2244 /*"pxor %%mm7, %%mm7 \n\t"*/\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2245 "1: \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2246 "movq (%0), %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2247 "movq 8(%0), %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2248 "movq 16(%0), %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2249 "movq 24(%0), %%mm3 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2250 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2251 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2252 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2253 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2254 \
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2255 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2256 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2257 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2258 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2259 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2260 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2261 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2262 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2263 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2264 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2265 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2266 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2267 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2268 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2269 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2270 \
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2271 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2272 "add %4, %1 \n\t" \
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2273 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2274 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2275 \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2276 "add $136, %0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2277 "add %6, %1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2278 "decl %2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2279 " jnz 1b \n\t"\
958
9bb668034ecf slowdown / gcc 2.95.* bug workaround (this should be reversed as soon as gcc 2.95.* support is droped)
michaelni
parents: 954
diff changeset
2280 \
967
274b518c4ecb PIC / ebx fix
michaelni
parents: 966
diff changeset
2281 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
2282 : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*(long)dstStride)\
966
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2283 :"memory"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2284 );\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2285 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2286 \
1057
bb5de8a59da8 * static,const,compiler warning cleanup
kabi
parents: 997
diff changeset
2287 static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
2209
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
2288 uint64_t temp[9*2];\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2289 uint64_t *temp_ptr= temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2290 int count= 9;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2291 \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2292 /*FIXME unroll */\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2293 asm volatile(\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2294 "pxor %%mm7, %%mm7 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2295 "1: \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2296 "movq (%0), %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2297 "movq (%0), %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2298 "punpcklbw %%mm7, %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2299 "punpckhbw %%mm7, %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2300 "movq %%mm0, (%1) \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2301 "movq %%mm1, 9*8(%1) \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2302 "add $8, %1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2303 "add %3, %0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2304 "decl %2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2305 " jnz 1b \n\t"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2306 : "+r" (src), "+r" (temp_ptr), "+r"(count)\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
2307 : "r" ((long)srcStride)\
966
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2308 : "memory"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2309 );\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2310 \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2311 temp_ptr= temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2312 count=2;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2313 \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2314 /*FIXME reorder for speed */\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2315 asm volatile(\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2316 /*"pxor %%mm7, %%mm7 \n\t"*/\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2317 "1: \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2318 "movq (%0), %%mm0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2319 "movq 8(%0), %%mm1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2320 "movq 16(%0), %%mm2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2321 "movq 24(%0), %%mm3 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2322 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP)\
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2323 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2324 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2325 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2326 \
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2327 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2328 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2329 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2330 \
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2331 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2332 "add %4, %1 \n\t"\
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2333 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2334 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2335 \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2336 "add $72, %0 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2337 "add %6, %1 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2338 "decl %2 \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2339 " jnz 1b \n\t"\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2340 \
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
2341 : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
2293
15cfba1b97b5 adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)
michael
parents: 2256
diff changeset
2342 : "r"((long)dstStride), "r"(2*(long)dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*(long)dstStride)\
966
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2343 : "memory"\
7ef9226f430e more gcc bug workarounds
michaelni
parents: 961
diff changeset
2344 );\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2345 }\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2346 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2347 static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2348 OPNAME ## pixels8_mmx(dst, src, stride, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2349 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2350 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2351 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2352 uint64_t temp[8];\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2353 uint8_t * const half= (uint8_t*)temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2354 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2355 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2356 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2357 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2358 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2359 OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2360 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2361 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2362 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2363 uint64_t temp[8];\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2364 uint8_t * const half= (uint8_t*)temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2365 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2366 OPNAME ## pixels8_l2_ ## MMX(dst, src+1, half, stride, stride, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2367 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2368 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2369 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2370 uint64_t temp[8];\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2371 uint8_t * const half= (uint8_t*)temp;\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2372 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2373 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2374 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2375 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2376 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2377 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2378 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2379 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2380 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2381 uint64_t temp[8];\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2382 uint8_t * const half= (uint8_t*)temp;\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2383 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2384 OPNAME ## pixels8_l2_ ## MMX(dst, src+stride, half, stride, stride, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2385 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2386 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2387 uint64_t half[8 + 9];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2388 uint8_t * const halfH= ((uint8_t*)half) + 64;\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2389 uint8_t * const halfHV= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2390 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2391 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2392 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2393 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2394 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2395 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2396 uint64_t half[8 + 9];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2397 uint8_t * const halfH= ((uint8_t*)half) + 64;\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2398 uint8_t * const halfHV= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2399 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2400 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2401 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2402 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2403 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2404 static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2405 uint64_t half[8 + 9];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2406 uint8_t * const halfH= ((uint8_t*)half) + 64;\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2407 uint8_t * const halfHV= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2408 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2409 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2410 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2411 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2412 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2413 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2414 uint64_t half[8 + 9];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2415 uint8_t * const halfH= ((uint8_t*)half) + 64;\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2416 uint8_t * const halfHV= ((uint8_t*)half);\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2417 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2418 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2419 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2420 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2421 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2422 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2423 uint64_t half[8 + 9];\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2424 uint8_t * const halfH= ((uint8_t*)half) + 64;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2425 uint8_t * const halfHV= ((uint8_t*)half);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2426 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2427 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2428 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2429 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2430 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2431 uint64_t half[8 + 9];\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2432 uint8_t * const halfH= ((uint8_t*)half) + 64;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2433 uint8_t * const halfHV= ((uint8_t*)half);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2434 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2435 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2436 OPNAME ## pixels8_l2_ ## MMX(dst, halfH+8, halfHV, stride, 8, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2437 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2438 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2439 uint64_t half[8 + 9];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2440 uint8_t * const halfH= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2441 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2442 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9);\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2443 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2444 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2445 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2446 uint64_t half[8 + 9];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2447 uint8_t * const halfH= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2448 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2449 put ## RND ## pixels8_l2_ ## MMX(halfH, src+1, halfH, 8, stride, 9);\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2450 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2451 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2452 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2453 uint64_t half[9];\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2454 uint8_t * const halfH= ((uint8_t*)half);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2455 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2456 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2457 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2458 static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2459 OPNAME ## pixels16_mmx(dst, src, stride, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2460 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2461 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2462 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2463 uint64_t temp[32];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2464 uint8_t * const half= (uint8_t*)temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2465 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2466 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2467 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2468 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2469 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2470 OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2471 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2472 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2473 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2474 uint64_t temp[32];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2475 uint8_t * const half= (uint8_t*)temp;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2476 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2477 OPNAME ## pixels16_l2_ ## MMX(dst, src+1, half, stride, stride, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2478 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2479 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2480 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2481 uint64_t temp[32];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2482 uint8_t * const half= (uint8_t*)temp;\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2483 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2484 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2485 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2486 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2487 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2488 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2489 }\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2490 \
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2491 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2492 uint64_t temp[32];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2493 uint8_t * const half= (uint8_t*)temp;\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2494 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2495 OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, stride, stride, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2496 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2497 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2498 uint64_t half[16*2 + 17*2];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2499 uint8_t * const halfH= ((uint8_t*)half) + 256;\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2500 uint8_t * const halfHV= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2501 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2502 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2503 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2504 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2505 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2506 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2507 uint64_t half[16*2 + 17*2];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2508 uint8_t * const halfH= ((uint8_t*)half) + 256;\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2509 uint8_t * const halfHV= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2510 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2511 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2512 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2513 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2514 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2515 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2516 uint64_t half[16*2 + 17*2];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2517 uint8_t * const halfH= ((uint8_t*)half) + 256;\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2518 uint8_t * const halfHV= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2519 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2520 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2521 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2522 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2523 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2524 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2525 uint64_t half[16*2 + 17*2];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2526 uint8_t * const halfH= ((uint8_t*)half) + 256;\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2527 uint8_t * const halfHV= ((uint8_t*)half);\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2528 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2529 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2530 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2531 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2532 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2533 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2534 uint64_t half[16*2 + 17*2];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2535 uint8_t * const halfH= ((uint8_t*)half) + 256;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2536 uint8_t * const halfHV= ((uint8_t*)half);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2537 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2538 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2539 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2540 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2541 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2542 uint64_t half[16*2 + 17*2];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2543 uint8_t * const halfH= ((uint8_t*)half) + 256;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2544 uint8_t * const halfHV= ((uint8_t*)half);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2545 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2546 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2547 OPNAME ## pixels16_l2_ ## MMX(dst, halfH+16, halfHV, stride, 16, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2548 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2549 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2550 uint64_t half[17*2];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2551 uint8_t * const halfH= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2552 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2553 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, stride, 17);\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2554 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2555 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2556 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2557 uint64_t half[17*2];\
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2558 uint8_t * const halfH= ((uint8_t*)half);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2559 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
2207
22b768f1261a 10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
michael
parents: 2067
diff changeset
2560 put ## RND ## pixels16_l2_ ## MMX(halfH, src+1, halfH, 16, stride, 17);\
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
2561 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2562 }\
1064
b32afefe7d33 * UINTX -> uintx_t INTX -> intx_t
kabi
parents: 1057
diff changeset
2563 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2564 uint64_t half[17*2];\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2565 uint8_t * const halfH= ((uint8_t*)half);\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2566 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2567 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2568 }
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2569
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2570 #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b " \n\t"
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2571 #define AVG_3DNOW_OP(a,b,temp, size) \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2572 "mov" #size " " #b ", " #temp " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2573 "pavgusb " #temp ", " #a " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2574 "mov" #size " " #a ", " #b " \n\t"
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2575 #define AVG_MMX2_OP(a,b,temp, size) \
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2576 "mov" #size " " #b ", " #temp " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2577 "pavgb " #temp ", " #a " \n\t"\
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
2578 "mov" #size " " #a ", " #b " \n\t"
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2579
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2580 QPEL_BASE(put_ , ff_pw_16, _ , PUT_OP, PUT_OP)
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2581 QPEL_BASE(avg_ , ff_pw_16, _ , AVG_MMX2_OP, AVG_3DNOW_OP)
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2582 QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2583 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, 3dnow)
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2584 QPEL_OP(avg_ , ff_pw_16, _ , AVG_3DNOW_OP, 3dnow)
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2585 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2586 QPEL_OP(put_ , ff_pw_16, _ , PUT_OP, mmx2)
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
2587 QPEL_OP(avg_ , ff_pw_16, _ , AVG_MMX2_OP, mmx2)
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2588 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2589
3807
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2590 /***********************************/
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2591 /* bilinear qpel: not compliant to any spec, only for -lavdopts fast */
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2592
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2593 #define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2594 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2595 OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE);\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2596 }
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2597 #define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2598 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2599 OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src+S0, stride, SIZE, S1, S2);\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2600 }
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2601
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2602 #define QPEL_2TAP(OPNAME, SIZE, MMX)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2603 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2604 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2605 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2606 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX =\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2607 OPNAME ## qpel ## SIZE ## _mc00_ ## MMX;\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2608 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX =\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2609 OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX;\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2610 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX =\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2611 OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX;\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2612 static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2613 OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src+1, stride, SIZE);\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2614 }\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2615 static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2616 OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src+stride, stride, SIZE);\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2617 }\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2618 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0, 1, 0)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2619 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1, -1, 0)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2620 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0, stride, 0)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2621 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride, -stride, 0)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2622 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0, stride, 1)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2623 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1, stride, -1)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2624 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride, -stride, 1)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2625 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride+1, -stride, -1)\
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2626
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2627 QPEL_2TAP(put_, 16, mmx2)
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2628 QPEL_2TAP(avg_, 16, mmx2)
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2629 QPEL_2TAP(put_, 8, mmx2)
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2630 QPEL_2TAP(avg_, 8, mmx2)
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2631 QPEL_2TAP(put_, 16, 3dnow)
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2632 QPEL_2TAP(avg_, 16, 3dnow)
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2633 QPEL_2TAP(put_, 8, 3dnow)
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2634 QPEL_2TAP(avg_, 8, 3dnow)
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2635
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
2636
393
bf164fce2c14 removed debug function
glantau
parents: 387
diff changeset
2637 #if 0
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
2638 static void just_return() { return; }
393
bf164fce2c14 removed debug function
glantau
parents: 387
diff changeset
2639 #endif
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
2640
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2641 #define SET_QPEL_FUNC(postfix1, postfix2) \
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2642 c->put_ ## postfix1 = put_ ## postfix2;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2643 c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
2644 c->avg_ ## postfix1 = avg_ ## postfix2;
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2645
3248
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2646 static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2647 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2648 const int w = 8;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2649 const int ix = ox>>(16+shift);
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2650 const int iy = oy>>(16+shift);
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2651 const int oxs = ox>>4;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2652 const int oys = oy>>4;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2653 const int dxxs = dxx>>4;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2654 const int dxys = dxy>>4;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2655 const int dyxs = dyx>>4;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2656 const int dyys = dyy>>4;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2657 const uint16_t r4[4] = {r,r,r,r};
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2658 const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys};
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2659 const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys};
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2660 const uint64_t shift2 = 2*shift;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2661 uint8_t edge_buf[(h+1)*stride];
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2662 int x, y;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2663
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2664 const int dxw = (dxx-(1<<(16+shift)))*(w-1);
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2665 const int dyh = (dyy-(1<<(16+shift)))*(h-1);
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2666 const int dxh = dxy*(h-1);
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2667 const int dyw = dyx*(w-1);
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2668 if( // non-constant fullpel offset (3% of blocks)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2669 (ox^(ox+dxw) | ox^(ox+dxh) | ox^(ox+dxw+dxh) |
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2670 oy^(oy+dyw) | oy^(oy+dyh) | oy^(oy+dyw+dyh)) >> (16+shift)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2671 // uses more than 16 bits of subpel mv (only at huge resolution)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2672 || (dxx|dxy|dyx|dyy)&15 )
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2673 {
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2674 //FIXME could still use mmx for some of the rows
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2675 ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height);
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2676 return;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2677 }
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2678
3250
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2679 src += ix + iy*stride;
3248
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2680 if( (unsigned)ix >= width-w ||
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2681 (unsigned)iy >= height-h )
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2682 {
3250
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2683 ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height);
3248
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2684 src = edge_buf;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2685 }
3250
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2686
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2687 asm volatile(
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2688 "movd %0, %%mm6 \n\t"
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2689 "pxor %%mm7, %%mm7 \n\t"
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2690 "punpcklwd %%mm6, %%mm6 \n\t"
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2691 "punpcklwd %%mm6, %%mm6 \n\t"
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2692 :: "r"(1<<shift)
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2693 );
3248
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2694
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2695 for(x=0; x<w; x+=4){
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2696 uint16_t dx4[4] = { oxs - dxys + dxxs*(x+0),
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2697 oxs - dxys + dxxs*(x+1),
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2698 oxs - dxys + dxxs*(x+2),
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2699 oxs - dxys + dxxs*(x+3) };
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2700 uint16_t dy4[4] = { oys - dyys + dyxs*(x+0),
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2701 oys - dyys + dyxs*(x+1),
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2702 oys - dyys + dyxs*(x+2),
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2703 oys - dyys + dyxs*(x+3) };
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2704
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2705 for(y=0; y<h; y++){
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2706 asm volatile(
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2707 "movq %0, %%mm4 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2708 "movq %1, %%mm5 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2709 "paddw %2, %%mm4 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2710 "paddw %3, %%mm5 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2711 "movq %%mm4, %0 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2712 "movq %%mm5, %1 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2713 "psrlw $12, %%mm4 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2714 "psrlw $12, %%mm5 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2715 : "+m"(*dx4), "+m"(*dy4)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2716 : "m"(*dxy4), "m"(*dyy4)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2717 );
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2718
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2719 asm volatile(
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2720 "movq %%mm6, %%mm2 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2721 "movq %%mm6, %%mm1 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2722 "psubw %%mm4, %%mm2 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2723 "psubw %%mm5, %%mm1 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2724 "movq %%mm2, %%mm0 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2725 "movq %%mm4, %%mm3 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2726 "pmullw %%mm1, %%mm0 \n\t" // (s-dx)*(s-dy)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2727 "pmullw %%mm5, %%mm3 \n\t" // dx*dy
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2728 "pmullw %%mm5, %%mm2 \n\t" // (s-dx)*dy
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2729 "pmullw %%mm4, %%mm1 \n\t" // dx*(s-dy)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2730
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2731 "movd %4, %%mm5 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2732 "movd %3, %%mm4 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2733 "punpcklbw %%mm7, %%mm5 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2734 "punpcklbw %%mm7, %%mm4 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2735 "pmullw %%mm5, %%mm3 \n\t" // src[1,1] * dx*dy
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2736 "pmullw %%mm4, %%mm2 \n\t" // src[0,1] * (s-dx)*dy
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2737
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2738 "movd %2, %%mm5 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2739 "movd %1, %%mm4 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2740 "punpcklbw %%mm7, %%mm5 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2741 "punpcklbw %%mm7, %%mm4 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2742 "pmullw %%mm5, %%mm1 \n\t" // src[1,0] * dx*(s-dy)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2743 "pmullw %%mm4, %%mm0 \n\t" // src[0,0] * (s-dx)*(s-dy)
3250
fec9bc8d63fc gmc_mmx tweaks
lorenm
parents: 3248
diff changeset
2744 "paddw %5, %%mm1 \n\t"
3248
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2745 "paddw %%mm3, %%mm2 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2746 "paddw %%mm1, %%mm0 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2747 "paddw %%mm2, %%mm0 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2748
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2749 "psrlw %6, %%mm0 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2750 "packuswb %%mm0, %%mm0 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2751 "movd %%mm0, %0 \n\t"
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2752
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2753 : "=m"(dst[x+y*stride])
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2754 : "m"(src[0]), "m"(src[1]),
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2755 "m"(src[stride]), "m"(src[stride+1]),
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2756 "m"(*r4), "m"(shift2)
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2757 );
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2758 src += stride;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2759 }
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2760 src += 4-h*stride;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2761 }
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2762 }
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
2763
3777
20545fbb6f7c add some #ifdef CONFIG_ENCODERS/DECODERS
mru
parents: 3721
diff changeset
2764 #ifdef CONFIG_ENCODERS
5024
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2765
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2766 #define PHADDD(a, t)\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2767 "movq "#a", "#t" \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2768 "psrlq $32, "#a" \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2769 "paddd "#t", "#a" \n\t"
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2770 /*
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2771 pmulhw: dst[0-15]=(src[0-15]*dst[0-15])[16-31]
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2772 pmulhrw: dst[0-15]=(src[0-15]*dst[0-15] + 0x8000)[16-31]
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2773 pmulhrsw: dst[0-15]=(src[0-15]*dst[0-15] + 0x4000)[15-30]
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2774 */
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2775 #define PMULHRW(x, y, s, o)\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2776 "pmulhw " #s ", "#x " \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2777 "pmulhw " #s ", "#y " \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2778 "paddw " #o ", "#x " \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2779 "paddw " #o ", "#y " \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2780 "psraw $1, "#x " \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2781 "psraw $1, "#y " \n\t"
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2782 #define DEF(x) x ## _mmx
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2783 #define SET_RND MOVQ_WONE
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2784 #define SCALE_OFFSET 1
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2785
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2786 #include "dsputil_mmx_qns.h"
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2787
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2788 #undef DEF
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2789 #undef SET_RND
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2790 #undef SCALE_OFFSET
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2791 #undef PMULHRW
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2792
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2793 #define DEF(x) x ## _3dnow
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2794 #define SET_RND(x)
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2795 #define SCALE_OFFSET 0
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2796 #define PMULHRW(x, y, s, o)\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2797 "pmulhrw " #s ", "#x " \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2798 "pmulhrw " #s ", "#y " \n\t"
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2799
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2800 #include "dsputil_mmx_qns.h"
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2801
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2802 #undef DEF
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2803 #undef SET_RND
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2804 #undef SCALE_OFFSET
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2805 #undef PMULHRW
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2806
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2807 #ifdef HAVE_SSSE3
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2808 #undef PHADDD
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2809 #define DEF(x) x ## _ssse3
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2810 #define SET_RND(x)
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2811 #define SCALE_OFFSET -1
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2812 #define PHADDD(a, t)\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2813 "pshufw $0x0E, "#a", "#t" \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2814 "paddd "#t", "#a" \n\t" /* faster than phaddd on core2 */
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2815 #define PMULHRW(x, y, s, o)\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2816 "pmulhrsw " #s ", "#x " \n\t"\
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2817 "pmulhrsw " #s ", "#y " \n\t"
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2818
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2819 #include "dsputil_mmx_qns.h"
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2820
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2821 #undef DEF
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2822 #undef SET_RND
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2823 #undef SCALE_OFFSET
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2824 #undef PMULHRW
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2825 #undef PHADDD
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2826 #endif //HAVE_SSSE3
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
2827
3777
20545fbb6f7c add some #ifdef CONFIG_ENCODERS/DECODERS
mru
parents: 3721
diff changeset
2828 #endif /* CONFIG_ENCODERS */
2754
a49f140179e9 sort H.264 mmx dsp functions into their own file
lorenm
parents: 2753
diff changeset
2829
3215
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2830 #define PREFETCH(name, op) \
4172
608e2dfcb86e adding more static keywords
mru
parents: 4127
diff changeset
2831 static void name(void *mem, int stride, int h){\
3215
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2832 const uint8_t *p= mem;\
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2833 do{\
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2834 asm volatile(#op" %0" :: "m"(*p));\
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2835 p+= stride;\
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2836 }while(--h);\
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2837 }
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2838 PREFETCH(prefetch_mmx2, prefetcht0)
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2839 PREFETCH(prefetch_3dnow, prefetch)
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2840 #undef PREFETCH
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
2841
2754
a49f140179e9 sort H.264 mmx dsp functions into their own file
lorenm
parents: 2753
diff changeset
2842 #include "h264dsp_mmx.c"
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
2843
6009
ecfdc0bfb233 typo/clarification
diego
parents: 5963
diff changeset
2844 /* CAVS specific */
3524
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2845 void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx);
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2846
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2847 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2848 put_pixels8_mmx(dst, src, stride, 8);
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2849 }
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2850 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2851 avg_pixels8_mmx(dst, src, stride, 8);
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2852 }
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2853 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2854 put_pixels16_mmx(dst, src, stride, 16);
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2855 }
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2856 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) {
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2857 avg_pixels16_mmx(dst, src, stride, 16);
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2858 }
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
2859
6030
fb99890ee609 move FLAC mmx dsp to its own file
aurel
parents: 6009
diff changeset
2860 /* FLAC specific */
fb99890ee609 move FLAC mmx dsp to its own file
aurel
parents: 6009
diff changeset
2861 void ff_flac_compute_autocorr_sse2(const int32_t *data, int len, int lag,
fb99890ee609 move FLAC mmx dsp to its own file
aurel
parents: 6009
diff changeset
2862 double *autoc);
fb99890ee609 move FLAC mmx dsp to its own file
aurel
parents: 6009
diff changeset
2863
5948
db875a610973 build vc1dsp_mmx.c in its own compilation unit
aurel
parents: 5947
diff changeset
2864 /* VC1 specific */
db875a610973 build vc1dsp_mmx.c in its own compilation unit
aurel
parents: 5947
diff changeset
2865 void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx);
db875a610973 build vc1dsp_mmx.c in its own compilation unit
aurel
parents: 5947
diff changeset
2866
db875a610973 build vc1dsp_mmx.c in its own compilation unit
aurel
parents: 5947
diff changeset
2867 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd) {
db875a610973 build vc1dsp_mmx.c in its own compilation unit
aurel
parents: 5947
diff changeset
2868 put_pixels8_mmx(dst, src, stride, 8);
db875a610973 build vc1dsp_mmx.c in its own compilation unit
aurel
parents: 5947
diff changeset
2869 }
db875a610973 build vc1dsp_mmx.c in its own compilation unit
aurel
parents: 5947
diff changeset
2870
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2871 /* external functions, from idct_mmx.c */
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2872 void ff_mmx_idct(DCTELEM *block);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2873 void ff_mmxext_idct(DCTELEM *block);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2874
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2875 /* XXX: those functions should be suppressed ASAP when all IDCTs are
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2876 converted */
4020
723818b5de0f Put libmpeg2 IDCT functions under CONFIG_GPL, fixes link failure
diego
parents: 4001
diff changeset
2877 #ifdef CONFIG_GPL
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2878 static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2879 {
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2880 ff_mmx_idct (block);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2881 put_pixels_clamped_mmx(block, dest, line_size);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2882 }
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2883 static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2884 {
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2885 ff_mmx_idct (block);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2886 add_pixels_clamped_mmx(block, dest, line_size);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2887 }
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2888 static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2889 {
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2890 ff_mmxext_idct (block);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2891 put_pixels_clamped_mmx(block, dest, line_size);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2892 }
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2893 static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2894 {
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2895 ff_mmxext_idct (block);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2896 add_pixels_clamped_mmx(block, dest, line_size);
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
2897 }
4020
723818b5de0f Put libmpeg2 IDCT functions under CONFIG_GPL, fixes link failure
diego
parents: 4001
diff changeset
2898 #endif
2868
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2899 static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2900 {
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2901 ff_idct_xvid_mmx (block);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2902 put_pixels_clamped_mmx(block, dest, line_size);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2903 }
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2904 static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2905 {
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2906 ff_idct_xvid_mmx (block);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2907 add_pixels_clamped_mmx(block, dest, line_size);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2908 }
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2909 static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2910 {
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2911 ff_idct_xvid_mmx2 (block);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2912 put_pixels_clamped_mmx(block, dest, line_size);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2913 }
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2914 static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2915 {
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2916 ff_idct_xvid_mmx2 (block);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2917 add_pixels_clamped_mmx(block, dest, line_size);
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
2918 }
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
2919
3541
3fbddeb13686 10l, vorbis_inverse_coupling_sse() was really 3dnow
lorenm
parents: 3536
diff changeset
2920 static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
3536
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2921 {
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2922 int i;
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2923 asm volatile("pxor %%mm7, %%mm7":);
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2924 for(i=0; i<blocksize; i+=2) {
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2925 asm volatile(
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2926 "movq %0, %%mm0 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2927 "movq %1, %%mm1 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2928 "movq %%mm0, %%mm2 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2929 "movq %%mm1, %%mm3 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2930 "pfcmpge %%mm7, %%mm2 \n\t" // m <= 0.0
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2931 "pfcmpge %%mm7, %%mm3 \n\t" // a <= 0.0
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2932 "pslld $31, %%mm2 \n\t" // keep only the sign bit
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2933 "pxor %%mm2, %%mm1 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2934 "movq %%mm3, %%mm4 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2935 "pand %%mm1, %%mm3 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2936 "pandn %%mm1, %%mm4 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2937 "pfadd %%mm0, %%mm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2938 "pfsub %%mm4, %%mm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2939 "movq %%mm3, %1 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2940 "movq %%mm0, %0 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2941 :"+m"(mag[i]), "+m"(ang[i])
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2942 ::"memory"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2943 );
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2944 }
3561
97325fecd35a emms -> femms
lorenm
parents: 3557
diff changeset
2945 asm volatile("femms");
3536
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2946 }
3557
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2947 static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
3536
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2948 {
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2949 int i;
3557
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2950
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2951 asm volatile(
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2952 "movaps %0, %%xmm5 \n\t"
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2953 ::"m"(ff_pdw_80000000[0])
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2954 );
3536
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2955 for(i=0; i<blocksize; i+=4) {
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2956 asm volatile(
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2957 "movaps %0, %%xmm0 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2958 "movaps %1, %%xmm1 \n\t"
3557
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2959 "xorps %%xmm2, %%xmm2 \n\t"
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2960 "xorps %%xmm3, %%xmm3 \n\t"
3536
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2961 "cmpleps %%xmm0, %%xmm2 \n\t" // m <= 0.0
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2962 "cmpleps %%xmm1, %%xmm3 \n\t" // a <= 0.0
3557
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2963 "andps %%xmm5, %%xmm2 \n\t" // keep only the sign bit
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2964 "xorps %%xmm2, %%xmm1 \n\t"
3536
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2965 "movaps %%xmm3, %%xmm4 \n\t"
3557
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2966 "andps %%xmm1, %%xmm3 \n\t"
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
2967 "andnps %%xmm1, %%xmm4 \n\t"
3536
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2968 "addps %%xmm0, %%xmm3 \n\t" // a = m + ((a<0) & (a ^ sign(m)))
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2969 "subps %%xmm4, %%xmm0 \n\t" // m = m + ((a>0) & (a ^ sign(m)))
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2970 "movaps %%xmm3, %1 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2971 "movaps %%xmm0, %0 \n\t"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2972 :"+m"(mag[i]), "+m"(ang[i])
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2973 ::"memory"
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2974 );
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2975 }
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2976 }
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
2977
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
2978 static void vector_fmul_3dnow(float *dst, const float *src, int len){
3574
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2979 long i = (len-4)*4;
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2980 asm volatile(
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2981 "1: \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2982 "movq (%1,%0), %%mm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2983 "movq 8(%1,%0), %%mm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2984 "pfmul (%2,%0), %%mm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2985 "pfmul 8(%2,%0), %%mm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2986 "movq %%mm0, (%1,%0) \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2987 "movq %%mm1, 8(%1,%0) \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2988 "sub $16, %0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2989 "jge 1b \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2990 "femms \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2991 :"+r"(i)
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2992 :"r"(dst), "r"(src)
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2993 :"memory"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2994 );
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
2995 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
2996 static void vector_fmul_sse(float *dst, const float *src, int len){
3574
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2997 long i = (len-8)*4;
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2998 asm volatile(
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
2999 "1: \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3000 "movaps (%1,%0), %%xmm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3001 "movaps 16(%1,%0), %%xmm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3002 "mulps (%2,%0), %%xmm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3003 "mulps 16(%2,%0), %%xmm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3004 "movaps %%xmm0, (%1,%0) \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3005 "movaps %%xmm1, 16(%1,%0) \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3006 "sub $32, %0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3007 "jge 1b \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3008 :"+r"(i)
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3009 :"r"(dst), "r"(src)
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3010 :"memory"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3011 );
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3012 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3013
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3014 static void vector_fmul_reverse_3dnow2(float *dst, const float *src0, const float *src1, int len){
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3015 long i = len*4-16;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3016 asm volatile(
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3017 "1: \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3018 "pswapd 8(%1), %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3019 "pswapd (%1), %%mm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3020 "pfmul (%3,%0), %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3021 "pfmul 8(%3,%0), %%mm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3022 "movq %%mm0, (%2,%0) \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3023 "movq %%mm1, 8(%2,%0) \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3024 "add $16, %1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3025 "sub $16, %0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3026 "jge 1b \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3027 :"+r"(i), "+r"(src1)
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3028 :"r"(dst), "r"(src0)
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3029 );
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3030 asm volatile("femms");
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3031 }
3569
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3032 static void vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len){
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3033 long i = len*4-32;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3034 asm volatile(
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3035 "1: \n\t"
3569
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3036 "movaps 16(%1), %%xmm0 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3037 "movaps (%1), %%xmm1 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3038 "shufps $0x1b, %%xmm0, %%xmm0 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3039 "shufps $0x1b, %%xmm1, %%xmm1 \n\t"
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3040 "mulps (%3,%0), %%xmm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3041 "mulps 16(%3,%0), %%xmm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3042 "movaps %%xmm0, (%2,%0) \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3043 "movaps %%xmm1, 16(%2,%0) \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3044 "add $32, %1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3045 "sub $32, %0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3046 "jge 1b \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3047 :"+r"(i), "+r"(src1)
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3048 :"r"(dst), "r"(src0)
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3049 );
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3050 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3051
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3052 static void vector_fmul_add_add_3dnow(float *dst, const float *src0, const float *src1,
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3053 const float *src2, int src3, int len, int step){
3574
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3054 long i = (len-4)*4;
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3055 if(step == 2 && src3 == 0){
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3056 dst += (len-4)*2;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3057 asm volatile(
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3058 "1: \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3059 "movq (%2,%0), %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3060 "movq 8(%2,%0), %%mm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3061 "pfmul (%3,%0), %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3062 "pfmul 8(%3,%0), %%mm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3063 "pfadd (%4,%0), %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3064 "pfadd 8(%4,%0), %%mm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3065 "movd %%mm0, (%1) \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3066 "movd %%mm1, 16(%1) \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3067 "psrlq $32, %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3068 "psrlq $32, %%mm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3069 "movd %%mm0, 8(%1) \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3070 "movd %%mm1, 24(%1) \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3071 "sub $32, %1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3072 "sub $16, %0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3073 "jge 1b \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3074 :"+r"(i), "+r"(dst)
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3075 :"r"(src0), "r"(src1), "r"(src2)
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3076 :"memory"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3077 );
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3078 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3079 else if(step == 1 && src3 == 0){
3574
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3080 asm volatile(
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3081 "1: \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3082 "movq (%2,%0), %%mm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3083 "movq 8(%2,%0), %%mm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3084 "pfmul (%3,%0), %%mm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3085 "pfmul 8(%3,%0), %%mm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3086 "pfadd (%4,%0), %%mm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3087 "pfadd 8(%4,%0), %%mm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3088 "movq %%mm0, (%1,%0) \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3089 "movq %%mm1, 8(%1,%0) \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3090 "sub $16, %0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3091 "jge 1b \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3092 :"+r"(i)
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3093 :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3094 :"memory"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3095 );
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3096 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3097 else
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3098 ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3099 asm volatile("femms");
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3100 }
3569
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3101 static void vector_fmul_add_add_sse(float *dst, const float *src0, const float *src1,
3574
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3102 const float *src2, int src3, int len, int step){
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3103 long i = (len-8)*4;
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3104 if(step == 2 && src3 == 0){
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3105 dst += (len-8)*2;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3106 asm volatile(
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3107 "1: \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3108 "movaps (%2,%0), %%xmm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3109 "movaps 16(%2,%0), %%xmm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3110 "mulps (%3,%0), %%xmm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3111 "mulps 16(%3,%0), %%xmm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3112 "addps (%4,%0), %%xmm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3113 "addps 16(%4,%0), %%xmm1 \n\t"
3569
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3114 "movss %%xmm0, (%1) \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3115 "movss %%xmm1, 32(%1) \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3116 "movhlps %%xmm0, %%xmm2 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3117 "movhlps %%xmm1, %%xmm3 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3118 "movss %%xmm2, 16(%1) \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3119 "movss %%xmm3, 48(%1) \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3120 "shufps $0xb1, %%xmm0, %%xmm0 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3121 "shufps $0xb1, %%xmm1, %%xmm1 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3122 "movss %%xmm0, 8(%1) \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3123 "movss %%xmm1, 40(%1) \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3124 "movhlps %%xmm0, %%xmm2 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3125 "movhlps %%xmm1, %%xmm3 \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3126 "movss %%xmm2, 24(%1) \n\t"
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3127 "movss %%xmm3, 56(%1) \n\t"
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3128 "sub $64, %1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3129 "sub $32, %0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3130 "jge 1b \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3131 :"+r"(i), "+r"(dst)
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3132 :"r"(src0), "r"(src1), "r"(src2)
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3133 :"memory"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3134 );
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3135 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3136 else if(step == 1 && src3 == 0){
3574
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3137 asm volatile(
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3138 "1: \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3139 "movaps (%2,%0), %%xmm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3140 "movaps 16(%2,%0), %%xmm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3141 "mulps (%3,%0), %%xmm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3142 "mulps 16(%3,%0), %%xmm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3143 "addps (%4,%0), %%xmm0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3144 "addps 16(%4,%0), %%xmm1 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3145 "movaps %%xmm0, (%1,%0) \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3146 "movaps %%xmm1, 16(%1,%0) \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3147 "sub $32, %0 \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3148 "jge 1b \n\t"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3149 :"+r"(i)
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3150 :"r"(dst), "r"(src0), "r"(src1), "r"(src2)
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3151 :"memory"
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3152 );
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3153 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3154 else
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3155 ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3156 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3157
4172
608e2dfcb86e adding more static keywords
mru
parents: 4127
diff changeset
3158 static void float_to_int16_3dnow(int16_t *dst, const float *src, int len){
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3159 // not bit-exact: pf2id uses different rounding than C and SSE
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3160 int i;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3161 for(i=0; i<len; i+=4) {
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3162 asm volatile(
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3163 "pf2id %1, %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3164 "pf2id %2, %%mm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3165 "packssdw %%mm1, %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3166 "movq %%mm0, %0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3167 :"=m"(dst[i])
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3168 :"m"(src[i]), "m"(src[i+2])
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3169 );
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3170 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3171 asm volatile("femms");
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3172 }
4172
608e2dfcb86e adding more static keywords
mru
parents: 4127
diff changeset
3173 static void float_to_int16_sse(int16_t *dst, const float *src, int len){
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3174 int i;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3175 for(i=0; i<len; i+=4) {
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3176 asm volatile(
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3177 "cvtps2pi %1, %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3178 "cvtps2pi %2, %%mm1 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3179 "packssdw %%mm1, %%mm0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3180 "movq %%mm0, %0 \n\t"
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3181 :"=m"(dst[i])
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3182 :"m"(src[i]), "m"(src[i+2])
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3183 );
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3184 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3185 asm volatile("emms");
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3186 }
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3187
3210
81cafbc23b8d snow mmx+sse2 optimizations, part 4
corey
parents: 3207
diff changeset
3188 extern void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width);
81cafbc23b8d snow mmx+sse2 optimizations, part 4
corey
parents: 3207
diff changeset
3189 extern void ff_snow_horizontal_compose97i_mmx(DWTELEM *b, int width);
3207
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3190 extern void ff_snow_vertical_compose97i_sse2(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3191 extern void ff_snow_vertical_compose97i_mmx(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
4436
d3e389536b0a Add the const specifier as needed to reduce the number of warnings.
takis
parents: 4197
diff changeset
3192 extern void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
3211
b77b5e7072d6 add MMX and SSE versions of ff_snow_inner_add_yblock
gpoirier
parents: 3210
diff changeset
3193 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
4436
d3e389536b0a Add the const specifier as needed to reduce the number of warnings.
takis
parents: 4197
diff changeset
3194 extern void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
3211
b77b5e7072d6 add MMX and SSE versions of ff_snow_inner_add_yblock
gpoirier
parents: 3210
diff changeset
3195 int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
3207
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3196
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3197 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
3198 {
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3199 mm_flags = mm_support();
1115
74a46d77e061 * support FF_MM_FORCE
kabi
parents: 1092
diff changeset
3200
1122
ddc3b0140b8f * oooooops - sorry for this one - wrong logic
kabi
parents: 1115
diff changeset
3201 if (avctx->dsp_mask) {
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
3202 if (avctx->dsp_mask & FF_MM_FORCE)
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3203 mm_flags |= (avctx->dsp_mask & 0xffff);
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
3204 else
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3205 mm_flags &= ~(avctx->dsp_mask & 0xffff);
1122
ddc3b0140b8f * oooooops - sorry for this one - wrong logic
kabi
parents: 1115
diff changeset
3206 }
1115
74a46d77e061 * support FF_MM_FORCE
kabi
parents: 1092
diff changeset
3207
631
47a8964ba5cd be less verbose patch by (Lennert Buytenhek <buytenh at math dot leidenuniv dot nl>)
michaelni
parents: 629
diff changeset
3208 #if 0
1868
771dcc2d4a0c use optimized VP3 functions where appropriate
melanson
parents: 1845
diff changeset
3209 av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3210 if (mm_flags & MM_MMX)
1868
771dcc2d4a0c use optimized VP3 functions where appropriate
melanson
parents: 1845
diff changeset
3211 av_log(avctx, AV_LOG_INFO, " mmx");
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3212 if (mm_flags & MM_MMXEXT)
1868
771dcc2d4a0c use optimized VP3 functions where appropriate
melanson
parents: 1845
diff changeset
3213 av_log(avctx, AV_LOG_INFO, " mmxext");
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3214 if (mm_flags & MM_3DNOW)
1868
771dcc2d4a0c use optimized VP3 functions where appropriate
melanson
parents: 1845
diff changeset
3215 av_log(avctx, AV_LOG_INFO, " 3dnow");
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3216 if (mm_flags & MM_SSE)
1868
771dcc2d4a0c use optimized VP3 functions where appropriate
melanson
parents: 1845
diff changeset
3217 av_log(avctx, AV_LOG_INFO, " sse");
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3218 if (mm_flags & MM_SSE2)
1868
771dcc2d4a0c use optimized VP3 functions where appropriate
melanson
parents: 1845
diff changeset
3219 av_log(avctx, AV_LOG_INFO, " sse2");
771dcc2d4a0c use optimized VP3 functions where appropriate
melanson
parents: 1845
diff changeset
3220 av_log(avctx, AV_LOG_INFO, "\n");
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
3221 #endif
986e461dc072 Initial revision
glantau
parents:
diff changeset
3222
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3223 if (mm_flags & MM_MMX) {
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3224 const int idct_algo= avctx->idct_algo;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3225
1232
e88d3b1fb2a1 more #ifdef CONFIG_ENCODERS by (Wolfgang Hesseler <qv at multimediaware dot com>)
michaelni
parents: 1186
diff changeset
3226 #ifdef CONFIG_ENCODERS
2024
f65d87bfdd5a some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)
michael
parents: 1985
diff changeset
3227 const int dct_algo = avctx->dct_algo;
1565
1a9a63f59849 minor mmx2 optimization if the dct
michael
parents: 1530
diff changeset
3228 if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3229 if(mm_flags & MM_SSE2){
1765
e31754bc5b65 SSE2 fdct by (Balatoni Denes <pnis at coder dot hu>)
michael
parents: 1739
diff changeset
3230 c->fdct = ff_fdct_sse2;
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3231 }else if(mm_flags & MM_MMXEXT){
1565
1a9a63f59849 minor mmx2 optimization if the dct
michael
parents: 1530
diff changeset
3232 c->fdct = ff_fdct_mmx2;
1a9a63f59849 minor mmx2 optimization if the dct
michael
parents: 1530
diff changeset
3233 }else{
1a9a63f59849 minor mmx2 optimization if the dct
michael
parents: 1530
diff changeset
3234 c->fdct = ff_fdct_mmx;
1a9a63f59849 minor mmx2 optimization if the dct
michael
parents: 1530
diff changeset
3235 }
1a9a63f59849 minor mmx2 optimization if the dct
michael
parents: 1530
diff changeset
3236 }
1232
e88d3b1fb2a1 more #ifdef CONFIG_ENCODERS by (Wolfgang Hesseler <qv at multimediaware dot com>)
michaelni
parents: 1186
diff changeset
3237 #endif //CONFIG_ENCODERS
2256
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3238 if(avctx->lowres==0){
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3239 if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3240 c->idct_put= ff_simple_idct_put_mmx;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3241 c->idct_add= ff_simple_idct_add_mmx;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3242 c->idct = ff_simple_idct_mmx;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3243 c->idct_permutation_type= FF_SIMPLE_IDCT_PERM;
3717
ea9fe1c9d126 Remove the LGPL exception clause as discussed on ffmpeg-devel
diego
parents: 3712
diff changeset
3244 #ifdef CONFIG_GPL
2256
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3245 }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3246 if(mm_flags & MM_MMXEXT){
2256
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3247 c->idct_put= ff_libmpeg2mmx2_idct_put;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3248 c->idct_add= ff_libmpeg2mmx2_idct_add;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3249 c->idct = ff_mmxext_idct;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3250 }else{
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3251 c->idct_put= ff_libmpeg2mmx_idct_put;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3252 c->idct_add= ff_libmpeg2mmx_idct_add;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3253 c->idct = ff_mmx_idct;
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3254 }
7e0b2e86afa9 1/2 resolution decoding
michael
parents: 2217
diff changeset
3255 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
3717
ea9fe1c9d126 Remove the LGPL exception clause as discussed on ffmpeg-devel
diego
parents: 3712
diff changeset
3256 #endif
5007
f7edc4fe94db Make vp3dsp*.c compilation optional.
takis
parents: 4988
diff changeset
3257 }else if((ENABLE_VP3_DECODER || ENABLE_VP5_DECODER || ENABLE_VP6_DECODER) &&
f7edc4fe94db Make vp3dsp*.c compilation optional.
takis
parents: 4988
diff changeset
3258 idct_algo==FF_IDCT_VP3 &&
3721
2000e401593d disable vp3 mmx idct for theora files to avoid artifacts
aurel
parents: 3717
diff changeset
3259 avctx->codec->id!=CODEC_ID_THEORA &&
3712
f7f75f718efb Enables back the mmx/sse optimized version of the vp3 idct.
aurel
parents: 3666
diff changeset
3260 !(avctx->flags & CODEC_FLAG_BITEXACT)){
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3261 if(mm_flags & MM_SSE2){
2696
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3262 c->idct_put= ff_vp3_idct_put_sse2;
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3263 c->idct_add= ff_vp3_idct_add_sse2;
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3264 c->idct = ff_vp3_idct_sse2;
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3265 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3266 }else{
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3267 ff_vp3_dsp_init_mmx();
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3268 c->idct_put= ff_vp3_idct_put_mmx;
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3269 c->idct_add= ff_vp3_idct_add_mmx;
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3270 c->idct = ff_vp3_idct_mmx;
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3271 c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
9699d325049d porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API
michael
parents: 2691
diff changeset
3272 }
3524
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
3273 }else if(idct_algo==FF_IDCT_CAVS){
419409926166 some MMX optimizations for the CAVS decoder
stefang
parents: 3496
diff changeset
3274 c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
2868
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3275 }else if(idct_algo==FF_IDCT_XVIDMMX){
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3276 if(mm_flags & MM_MMXEXT){
2868
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3277 c->idct_put= ff_idct_xvid_mmx2_put;
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3278 c->idct_add= ff_idct_xvid_mmx2_add;
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3279 c->idct = ff_idct_xvid_mmx2;
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3280 }else{
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3281 c->idct_put= ff_idct_xvid_mmx_put;
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3282 c->idct_add= ff_idct_xvid_mmx_add;
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3283 c->idct = ff_idct_xvid_mmx;
666064f710d4 xvids mmx&mmx2 idcts
michael
parents: 2864
diff changeset
3284 }
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3285 }
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3286 }
1868
771dcc2d4a0c use optimized VP3 functions where appropriate
melanson
parents: 1845
diff changeset
3287
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3288 #ifdef CONFIG_ENCODERS
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3289 c->get_pixels = get_pixels_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3290 c->diff_pixels = diff_pixels_mmx;
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3291 #endif //CONFIG_ENCODERS
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3292 c->put_pixels_clamped = put_pixels_clamped_mmx;
1984
ef919e9ef73e separate out put_signed_pixels_clamped() into its own function and
melanson
parents: 1977
diff changeset
3293 c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx;
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3294 c->add_pixels_clamped = add_pixels_clamped_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3295 c->clear_blocks = clear_blocks_mmx;
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3296 #ifdef CONFIG_ENCODERS
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3297 c->pix_sum = pix_sum16_mmx;
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3298 #endif //CONFIG_ENCODERS
415
1c3f42442fba * added simple test main - see comments about how to
kabi
parents: 402
diff changeset
3299
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3300 c->put_pixels_tab[0][0] = put_pixels16_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3301 c->put_pixels_tab[0][1] = put_pixels16_x2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3302 c->put_pixels_tab[0][2] = put_pixels16_y2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3303 c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx;
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
3304
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3305 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3306 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3307 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3308 c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx;
651
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
3309
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3310 c->avg_pixels_tab[0][0] = avg_pixels16_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3311 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3312 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3313 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx;
415
1c3f42442fba * added simple test main - see comments about how to
kabi
parents: 402
diff changeset
3314
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3315 c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3316 c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3317 c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3318 c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3319
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3320 c->put_pixels_tab[1][0] = put_pixels8_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3321 c->put_pixels_tab[1][1] = put_pixels8_x2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3322 c->put_pixels_tab[1][2] = put_pixels8_y2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3323 c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx;
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
3324
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3325 c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3326 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3327 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3328 c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx;
651
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
3329
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3330 c->avg_pixels_tab[1][0] = avg_pixels8_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3331 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3332 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3333 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx;
651
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
3334
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3335 c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3336 c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3337 c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3338 c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
3339
3248
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
3340 c->gmc= gmc_mmx;
7aa9f80e7954 mmx implementation of 3-point GMC. (5x faster than C)
lorenm
parents: 3215
diff changeset
3341
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
3342 c->add_bytes= add_bytes_mmx;
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3343 #ifdef CONFIG_ENCODERS
866
725ef4ea3ecc huffyuv
michaelni
parents: 853
diff changeset
3344 c->diff_bytes= diff_bytes_mmx;
4988
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
3345 c->sum_abs_dctelem= sum_abs_dctelem_mmx;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
3346
936
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
3347 c->hadamard8_diff[0]= hadamard8_diff16_mmx;
caa77cd960c0 qpel encoding
michaelni
parents: 866
diff changeset
3348 c->hadamard8_diff[1]= hadamard8_diff_mmx;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
3349
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
3350 c->pix_norm1 = pix_norm1_mmx;
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3351 c->sse[0] = (mm_flags & MM_SSE2) ? sse16_sse2 : sse16_mmx;
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
3352 c->sse[1] = sse8_mmx;
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
3353 c->vsad[4]= vsad_intra16_mmx;
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
3354
2979
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
3355 c->nsse[0] = nsse16_mmx;
bfabfdf9ce55 COSMETICS: tabs --> spaces, some prettyprinting
diego
parents: 2967
diff changeset
3356 c->nsse[1] = nsse8_mmx;
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
3357 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
3358 c->vsad[0] = vsad16_mmx;
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
3359 }
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
3360
1784
65f7bd09f37b quantizer noise shaping optimization
michael
parents: 1772
diff changeset
3361 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
65f7bd09f37b quantizer noise shaping optimization
michael
parents: 1772
diff changeset
3362 c->try_8x8basis= try_8x8basis_mmx;
65f7bd09f37b quantizer noise shaping optimization
michael
parents: 1772
diff changeset
3363 }
65f7bd09f37b quantizer noise shaping optimization
michael
parents: 1772
diff changeset
3364 c->add_8x8basis= add_8x8basis_mmx;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
3365
4749
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
3366 c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;
7011f597e473 mmx 16-bit ssd. 2.3x faster svq1 encoding.
lorenm
parents: 4589
diff changeset
3367
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3368 #endif //CONFIG_ENCODERS
1647
c943c1d2d099 h263_v_loop_filter_mmx
michael
parents: 1566
diff changeset
3369
5277
7b3fcb7c61ce Avoid linking with h263.c functions when the relevant codecs
aurel
parents: 5255
diff changeset
3370 if (ENABLE_ANY_H263) {
5278
ef85411bb7e8 cosmetics: indentation
aurel
parents: 5277
diff changeset
3371 c->h263_v_loop_filter= h263_v_loop_filter_mmx;
ef85411bb7e8 cosmetics: indentation
aurel
parents: 5277
diff changeset
3372 c->h263_h_loop_filter= h263_h_loop_filter_mmx;
5277
7b3fcb7c61ce Avoid linking with h263.c functions when the relevant codecs
aurel
parents: 5255
diff changeset
3373 }
6057
03febc8f506f add MMX version for put_no_rnd_h264_chroma_mc8_c, used in VC-1 decoding.
gpoirier
parents: 6056
diff changeset
3374 c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx_rnd;
2922
d772011258ec faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents: 2902
diff changeset
3375 c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx;
6057
03febc8f506f add MMX version for put_no_rnd_h264_chroma_mc8_c, used in VC-1 decoding.
gpoirier
parents: 6056
diff changeset
3376 c->put_no_rnd_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx_nornd;
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
3377
3173
9a2cc7b0fbdb h264_idct_add only needs mmx1
lorenm
parents: 3105
diff changeset
3378 c->h264_idct_dc_add=
9a2cc7b0fbdb h264_idct_add only needs mmx1
lorenm
parents: 3105
diff changeset
3379 c->h264_idct_add= ff_h264_idct_add_mmx;
3174
b65cbae9d940 h264_idct8_add_mmx
lorenm
parents: 3173
diff changeset
3380 c->h264_idct8_dc_add=
b65cbae9d940 h264_idct8_add_mmx
lorenm
parents: 3173
diff changeset
3381 c->h264_idct8_add= ff_h264_idct8_add_mmx;
3173
9a2cc7b0fbdb h264_idct_add only needs mmx1
lorenm
parents: 3105
diff changeset
3382
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3383 if (mm_flags & MM_MMXEXT) {
3215
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
3384 c->prefetch = prefetch_mmx2;
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
3385
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3386 c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3387 c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
651
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
3388
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3389 c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3390 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3391 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
415
1c3f42442fba * added simple test main - see comments about how to
kabi
parents: 402
diff changeset
3392
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3393 c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3394 c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
651
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
3395
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3396 c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3397 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3398 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3399
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3400 #ifdef CONFIG_ENCODERS
4988
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
3401 c->sum_abs_dctelem= sum_abs_dctelem_mmx2;
1153
2725c8eb3c81 faster hadamard transform
michaelni
parents: 1122
diff changeset
3402 c->hadamard8_diff[0]= hadamard8_diff16_mmx2;
2725c8eb3c81 faster hadamard transform
michaelni
parents: 1122
diff changeset
3403 c->hadamard8_diff[1]= hadamard8_diff_mmx2;
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
3404 c->vsad[4]= vsad_intra16_mmx2;
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3405 #endif //CONFIG_ENCODERS
1153
2725c8eb3c81 faster hadamard transform
michaelni
parents: 1122
diff changeset
3406
3105
2d35fb3cb940 h264: special case dc-only idct. ~1% faster overall
lorenm
parents: 3089
diff changeset
3407 c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2;
2d35fb3cb940 h264: special case dc-only idct. ~1% faster overall
lorenm
parents: 3089
diff changeset
3408 c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2;
2745
42d3e9068e32 MMX for H.264 iDCT (adapted from x264)
lorenm
parents: 2732
diff changeset
3409
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3410 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3411 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3412 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3413 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3414 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3415 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3416 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
1772
8cd5257195c9 vsad16_mmx2 only applies if encoders are turned on
melanson
parents: 1765
diff changeset
3417 #ifdef CONFIG_ENCODERS
1729
a4a5e7521339 interlaced dct decision cleanup
michael
parents: 1708
diff changeset
3418 c->vsad[0] = vsad16_mmx2;
1772
8cd5257195c9 vsad16_mmx2 only applies if encoders are turned on
melanson
parents: 1765
diff changeset
3419 #endif //CONFIG_ENCODERS
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3420 }
959
3ec070eef24a qpel in b frames bugfixes
michaelni
parents: 958
diff changeset
3421
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
3422 #if 1
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3423 SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3424 SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3425 SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3426 SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3427 SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3428 SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3429 SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3430 SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3431 SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3432 SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3433 SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3434 SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3435 SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3436 SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3437 SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3438 SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3439 SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3440 SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3441 SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3442 SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3443 SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3444 SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3445 SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3446 SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3447 SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3448 SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3449 SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3450 SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3451 SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3452 SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3453 SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3454 SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)
961
f8c5babc7b4e 1000l (push & esp) using mangle now ...
michaelni
parents: 959
diff changeset
3455 #endif
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
3456
2209
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3457 //FIXME 3dnow too
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3458 #define dspfunc(PFX, IDX, NUM) \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3459 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3460 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3461 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3462 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3463 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3464 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3465 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3466 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3467 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3468 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3469 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3470 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3471 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3472 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3473 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_mmx2; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3474 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_mmx2
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3475
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3476 dspfunc(put_h264_qpel, 0, 16);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3477 dspfunc(put_h264_qpel, 1, 8);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3478 dspfunc(put_h264_qpel, 2, 4);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3479 dspfunc(avg_h264_qpel, 0, 16);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3480 dspfunc(avg_h264_qpel, 1, 8);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3481 dspfunc(avg_h264_qpel, 2, 4);
3807
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3482
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3483 dspfunc(put_2tap_qpel, 0, 16);
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3484 dspfunc(put_2tap_qpel, 1, 8);
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3485 dspfunc(avg_2tap_qpel, 0, 16);
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3486 dspfunc(avg_2tap_qpel, 1, 8);
2209
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3487 #undef dspfunc
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3488
6057
03febc8f506f add MMX version for put_no_rnd_h264_chroma_mc8_c, used in VC-1 decoding.
gpoirier
parents: 6056
diff changeset
3489 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2_rnd;
2922
d772011258ec faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents: 2902
diff changeset
3490 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2;
3213
57d31bdbebe8 added mmx implementation of h264_chroma_mc2
lorenm
parents: 3211
diff changeset
3491 c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2;
57d31bdbebe8 added mmx implementation of h264_chroma_mc2
lorenm
parents: 3211
diff changeset
3492 c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2;
2633
72e6ffa1f3a5 MMX for H.264 deblocking filter
lorenm
parents: 2505
diff changeset
3493 c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;
72e6ffa1f3a5 MMX for H.264 deblocking filter
lorenm
parents: 2505
diff changeset
3494 c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;
72e6ffa1f3a5 MMX for H.264 deblocking filter
lorenm
parents: 2505
diff changeset
3495 c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;
72e6ffa1f3a5 MMX for H.264 deblocking filter
lorenm
parents: 2505
diff changeset
3496 c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;
2707
360024d31dab H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents: 2696
diff changeset
3497 c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;
360024d31dab H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)
lorenm
parents: 2696
diff changeset
3498 c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;
3645
47821be55b6c mmx implementation of deblocking strength decision.
lorenm
parents: 3576
diff changeset
3499 c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;
2633
72e6ffa1f3a5 MMX for H.264 deblocking filter
lorenm
parents: 2505
diff changeset
3500
2902
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3501 c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3502 c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3503 c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3504 c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3505 c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3506 c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3507 c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3508 c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3509
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3510 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3511 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3512 c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3513 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3514 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3515 c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3516 c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3517 c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;
3c79bc9f3aa9 h264 mmx weighted prediction. up to 3% overall speedup.
lorenm
parents: 2899
diff changeset
3518
5949
d7ad613197d2 convert some #ifdef CONFIG_ to if(ENABLE_
aurel
parents: 5948
diff changeset
3519 if (ENABLE_CAVS_DECODER)
5950
e419e6d4e7eb cosmetics: indentation
aurel
parents: 5949
diff changeset
3520 ff_cavsdsp_init_mmx2(c, avctx);
5949
d7ad613197d2 convert some #ifdef CONFIG_ to if(ENABLE_
aurel
parents: 5948
diff changeset
3521
d7ad613197d2 convert some #ifdef CONFIG_ to if(ENABLE_
aurel
parents: 5948
diff changeset
3522 if (ENABLE_VC1_DECODER || ENABLE_WMV3_DECODER)
5950
e419e6d4e7eb cosmetics: indentation
aurel
parents: 5949
diff changeset
3523 ff_vc1dsp_init_mmx(c, avctx);
5933
6ce8f15fc02b add VC-1 MMX DSP functions, under MIT license.
gpoirier
parents: 5912
diff changeset
3524
1686
68abbec33289 Here are juste two added #ifdef CONFIG_ENCODERS to allow
michael
parents: 1648
diff changeset
3525 #ifdef CONFIG_ENCODERS
1527
8ffd0c00e6df mmx2 optimization of huffyuv median encoding
michael
parents: 1324
diff changeset
3526 c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;
1686
68abbec33289 Here are juste two added #ifdef CONFIG_ENCODERS to allow
michael
parents: 1648
diff changeset
3527 #endif //CONFIG_ENCODERS
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3528 } else if (mm_flags & MM_3DNOW) {
3215
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
3529 c->prefetch = prefetch_3dnow;
06f98047ff26 prefetch pixels for future motion compensation. 2-5% faster h264.
lorenm
parents: 3213
diff changeset
3530
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3531 c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3532 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
393
bf164fce2c14 removed debug function
glantau
parents: 387
diff changeset
3533
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3534 c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3535 c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3536 c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
651
45e8f39fda50 put/avg_pixels16
michaelni
parents: 631
diff changeset
3537
853
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3538 c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3539 c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3540
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3541 c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3542 c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
eacc2dd8fd9d * using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents: 706
diff changeset
3543 c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3544
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3545 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3546 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3547 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3548 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3549 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3550 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3551 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3552 }
984
e162c09efbe7 qpel fix
michaelni
parents: 967
diff changeset
3553
954
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3554 SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3555 SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3556 SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3557 SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3558 SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3559 SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3560 SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3561 SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3562 SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3563 SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3564 SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3565 SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3566 SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3567 SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3568 SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3569 SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3570 SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3571 SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3572 SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3573 SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3574 SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3575 SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3576 SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3577 SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3578 SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3579 SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3580 SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3581 SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3582 SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3583 SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3584 SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)
13aec7e50c52 qpel in mmx2/3dnow
michaelni
parents: 936
diff changeset
3585 SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
2209
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3586
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3587 #define dspfunc(PFX, IDX, NUM) \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3588 c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3589 c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3590 c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3591 c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3592 c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3593 c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3594 c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3595 c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3596 c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3597 c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3598 c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3599 c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3600 c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3601 c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3602 c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_3dnow; \
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3603 c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_3dnow
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3604
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3605 dspfunc(put_h264_qpel, 0, 16);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3606 dspfunc(put_h264_qpel, 1, 8);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3607 dspfunc(put_h264_qpel, 2, 4);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3608 dspfunc(avg_h264_qpel, 0, 16);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3609 dspfunc(avg_h264_qpel, 1, 8);
c4a476971abc h264 luma motion compensation in mmx2/3dnow
michael
parents: 2207
diff changeset
3610 dspfunc(avg_h264_qpel, 2, 4);
2732
473ee06ec3a1 MMX code for (put|avg)_h264_chroma_mc8
hzoli
parents: 2707
diff changeset
3611
3807
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3612 dspfunc(put_2tap_qpel, 0, 16);
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3613 dspfunc(put_2tap_qpel, 1, 8);
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3614 dspfunc(avg_2tap_qpel, 0, 16);
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3615 dspfunc(avg_2tap_qpel, 1, 8);
6a40092eb9e6 approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.
lorenm
parents: 3777
diff changeset
3616
6057
03febc8f506f add MMX version for put_no_rnd_h264_chroma_mc8_c, used in VC-1 decoding.
gpoirier
parents: 6056
diff changeset
3617 c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow_rnd;
2922
d772011258ec faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
lorenm
parents: 2902
diff changeset
3618 c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow;
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
3619 }
3207
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3620
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3621 #ifdef CONFIG_ENCODERS
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3622 if(mm_flags & MM_SSE2){
4988
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
3623 c->sum_abs_dctelem= sum_abs_dctelem_sse2;
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3624 c->hadamard8_diff[0]= hadamard8_diff16_sse2;
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3625 c->hadamard8_diff[1]= hadamard8_diff_sse2;
6030
fb99890ee609 move FLAC mmx dsp to its own file
aurel
parents: 6009
diff changeset
3626 if (ENABLE_FLAC_ENCODER)
fb99890ee609 move FLAC mmx dsp to its own file
aurel
parents: 6009
diff changeset
3627 c->flac_compute_autocorr = ff_flac_compute_autocorr_sse2;
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3628 }
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3629
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3630 #ifdef HAVE_SSSE3
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3631 if(mm_flags & MM_SSSE3){
5024
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3632 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3633 c->try_8x8basis= try_8x8basis_ssse3;
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3634 }
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3635 c->add_8x8basis= add_8x8basis_ssse3;
4988
689490842cf5 factor sum_abs_dctelem out of dct_sad, and simd it.
lorenm
parents: 4987
diff changeset
3636 c->sum_abs_dctelem= sum_abs_dctelem_ssse3;
4987
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3637 c->hadamard8_diff[0]= hadamard8_diff16_ssse3;
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3638 c->hadamard8_diff[1]= hadamard8_diff_ssse3;
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3639 }
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3640 #endif
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3641 #endif
02199b094850 sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
lorenm
parents: 4946
diff changeset
3642
4589
30261f4ed12d Fix wrong conditional, Snow decoding, not encoding, was SIMD-accelerated.
diego
parents: 4436
diff changeset
3643 #ifdef CONFIG_SNOW_DECODER
5591
642588a60570 update mmx code to latest snow changes
michael
parents: 5587
diff changeset
3644 if(mm_flags & MM_SSE2 & 0){
3210
81cafbc23b8d snow mmx+sse2 optimizations, part 4
corey
parents: 3207
diff changeset
3645 c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
5602
3b21f3268707 CONFIG_7REGS has been renamed to HAVE_7REGS
ramiro
parents: 5601
diff changeset
3646 #ifdef HAVE_7REGS
3207
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3647 c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
5601
b26025b9586d workaround gcc bug, untested as my gcc is not complaining
michael
parents: 5594
diff changeset
3648 #endif
3211
b77b5e7072d6 add MMX and SSE versions of ff_snow_inner_add_yblock
gpoirier
parents: 3210
diff changeset
3649 c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
3207
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3650 }
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3651 else{
5594
384629ebcb93 avoid overflow in the 3rd lifting step, this now needs mmx2 at minimum
michael
parents: 5591
diff changeset
3652 if(mm_flags & MM_MMXEXT){
3210
81cafbc23b8d snow mmx+sse2 optimizations, part 4
corey
parents: 3207
diff changeset
3653 c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
5602
3b21f3268707 CONFIG_7REGS has been renamed to HAVE_7REGS
ramiro
parents: 5601
diff changeset
3654 #ifdef HAVE_7REGS
3207
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3655 c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
5601
b26025b9586d workaround gcc bug, untested as my gcc is not complaining
michael
parents: 5594
diff changeset
3656 #endif
5594
384629ebcb93 avoid overflow in the 3rd lifting step, this now needs mmx2 at minimum
michael
parents: 5591
diff changeset
3657 }
3211
b77b5e7072d6 add MMX and SSE versions of ff_snow_inner_add_yblock
gpoirier
parents: 3210
diff changeset
3658 c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
3207
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3659 }
33110c1008a4 Add the mmx and sse2 implementations of ff_snow_vertical_compose().
gpoirier
parents: 3174
diff changeset
3660 #endif
3536
545a15c19c91 sse & sse2 implementations of vorbis channel coupling.
lorenm
parents: 3524
diff changeset
3661
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3662 if(mm_flags & MM_3DNOW){
5024
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3663 #ifdef CONFIG_ENCODERS
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3664 if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3665 c->try_8x8basis= try_8x8basis_3dnow;
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3666 }
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3667 c->add_8x8basis= add_8x8basis_3dnow;
8a3bc96c366f 3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw
zuxy
parents: 5014
diff changeset
3668 #endif //CONFIG_ENCODERS
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3669 c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3670 c->vector_fmul = vector_fmul_3dnow;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3671 if(!(avctx->flags & CODEC_FLAG_BITEXACT))
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3672 c->float_to_int16 = float_to_int16_3dnow;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3673 }
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3674 if(mm_flags & MM_3DNOWEXT)
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3675 c->vector_fmul_reverse = vector_fmul_reverse_3dnow2;
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3676 if(mm_flags & MM_SSE){
3557
8e13ec0f8aa3 change vorbis_inverse_coupling_sse2() so it works on sse1 cpus
michael
parents: 3541
diff changeset
3677 c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3678 c->vector_fmul = vector_fmul_sse;
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3679 c->float_to_int16 = float_to_int16_sse;
3569
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3680 c->vector_fmul_reverse = vector_fmul_reverse_sse;
c42c03f3b402 convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
michael
parents: 3568
diff changeset
3681 c->vector_fmul_add_add = vector_fmul_add_add_sse;
3568
945caa35ee9a sse and 3dnow implementations of float->int conversion and mdct windowing.
lorenm
parents: 3561
diff changeset
3682 }
4197
bbe0bc387a19 revert bad checkin
mru
parents: 4196
diff changeset
3683 if(mm_flags & MM_3DNOW)
3574
f549d1e685f7 vorbis simd tweaks
lorenm
parents: 3569
diff changeset
3684 c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
3685 }
2967
ef2149182f1c COSMETICS: Remove all trailing whitespace.
diego
parents: 2940
diff changeset
3686
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3687 #ifdef CONFIG_ENCODERS
1092
f59c3f66363b MpegEncContext.(i)dct_* -> DspContext.(i)dct_*
michaelni
parents: 1065
diff changeset
3688 dsputil_init_pix_mmx(c, avctx);
1530
3b31998fe22f disable encoders where appropriate (patch courtesy of BERO
melanson
parents: 1527
diff changeset
3689 #endif //CONFIG_ENCODERS
247
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3690 #if 0
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3691 // for speed testing
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3692 get_pixels = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3693 put_pixels_clamped = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3694 add_pixels_clamped = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3695
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3696 pix_abs16x16 = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3697 pix_abs16x16_x2 = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3698 pix_abs16x16_y2 = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3699 pix_abs16x16_xy2 = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3700
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3701 put_pixels_tab[0] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3702 put_pixels_tab[1] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3703 put_pixels_tab[2] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3704 put_pixels_tab[3] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3705
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3706 put_no_rnd_pixels_tab[0] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3707 put_no_rnd_pixels_tab[1] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3708 put_no_rnd_pixels_tab[2] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3709 put_no_rnd_pixels_tab[3] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3710
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3711 avg_pixels_tab[0] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3712 avg_pixels_tab[1] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3713 avg_pixels_tab[2] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3714 avg_pixels_tab[3] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3715
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3716 avg_no_rnd_pixels_tab[0] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3717 avg_no_rnd_pixels_tab[1] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3718 avg_no_rnd_pixels_tab[2] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3719 avg_no_rnd_pixels_tab[3] = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3720
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3721 //av_fdct = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3722 //ff_idct = just_return;
6f48cacd9ed9 * some modifications to allow gcc to compile same code for -fPIC
kabi
parents: 188
diff changeset
3723 #endif
0
986e461dc072 Initial revision
glantau
parents:
diff changeset
3724 }