annotate ppc/dsputil_altivec.h @ 4863:fa4951ffc434 libavcodec

Simplify, patch by Ramiro Polla. ramiro ta lisha tod ufsc tod br
author banan
date Wed, 18 Apr 2007 19:47:22 +0000
parents fec25fd9febf
children ad0c45e0008c
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
828
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
1 /*
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
2 * Copyright (c) 2002 Brian Foley
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
3 * Copyright (c) 2002 Dieter Shirley
1949
66215baae7b9 hadamard8_diff8x8 in AltiVec, the 16bits edition by (Romain Dolbeau <dolbeau at irisa dot fr>)
michael
parents: 1708
diff changeset
4 * Copyright (c) 2003-2004 Romain Dolbeau <romain@dolbeau.org>
828
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
5 *
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3945
diff changeset
6 * This file is part of FFmpeg.
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3945
diff changeset
7 *
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3945
diff changeset
8 * FFmpeg is free software; you can redistribute it and/or
828
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
9 * modify it under the terms of the GNU Lesser General Public
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
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: 3945
diff changeset
11 * version 2.1 of the License, or (at your option) any later version.
828
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
12 *
3947
c8c591fe26f8 Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents: 3945
diff changeset
13 * FFmpeg is distributed in the hope that it will be useful,
828
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
16 * Lesser General Public License for more details.
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
17 *
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
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: 3945
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: 2057
diff changeset
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
828
ace3ccd18dd2 Altivec Patch (Mark III) by (Dieter Shirley <dieters at schemasoft dot com>)
michaelni
parents: 638
diff changeset
21 */
878
6ea69518e5f7 altivec optimizations patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents: 828
diff changeset
22
1015
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
23 #ifndef _DSPUTIL_ALTIVEC_
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
24 #define _DSPUTIL_ALTIVEC_
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
25
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
26 #include "dsputil_ppc.h"
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
27
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
28 #ifdef HAVE_ALTIVEC
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
29
638
0012f75c92bb altivec build tidyup patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
diff changeset
30 extern int has_altivec(void);
981
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
31
3945
9544ad38f02a fix a warning
lu_zero
parents: 3546
diff changeset
32 void put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h);
9544ad38f02a fix a warning
lu_zero
parents: 3546
diff changeset
33
9544ad38f02a fix a warning
lu_zero
parents: 3546
diff changeset
34 void avg_pixels16_altivec(uint8_t *block, const uint8_t *pixels, int line_size, int h);
9544ad38f02a fix a warning
lu_zero
parents: 3546
diff changeset
35
981
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
36 // used to build registers permutation vectors (vcprm)
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
37 // the 's' are for words in the _s_econd vector
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
38 #define WORD_0 0x00,0x01,0x02,0x03
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
39 #define WORD_1 0x04,0x05,0x06,0x07
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
40 #define WORD_2 0x08,0x09,0x0a,0x0b
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
41 #define WORD_3 0x0c,0x0d,0x0e,0x0f
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
42 #define WORD_s0 0x10,0x11,0x12,0x13
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
43 #define WORD_s1 0x14,0x15,0x16,0x17
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
44 #define WORD_s2 0x18,0x19,0x1a,0x1b
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
45 #define WORD_s3 0x1c,0x1d,0x1e,0x1f
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
46
1033
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
47 #ifdef CONFIG_DARWIN
981
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
48 #define vcprm(a,b,c,d) (const vector unsigned char)(WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d)
1033
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
49 #else
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
50 #define vcprm(a,b,c,d) (const vector unsigned char){WORD_ ## a, WORD_ ## b, WORD_ ## c, WORD_ ## d}
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
51 #endif
981
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
52
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
53 // vcprmle is used to keep the same index as in the SSE version.
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
54 // it's the same as vcprm, with the index inversed
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
55 // ('le' is Little Endian)
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
56 #define vcprmle(a,b,c,d) vcprm(d,c,b,a)
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
57
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
58 // used to build inverse/identity vectors (vcii)
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
59 // n is _n_egative, p is _p_ositive
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
60 #define FLOAT_n -1.
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
61 #define FLOAT_p 1.
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
62
1033
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
63
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
64 #ifdef CONFIG_DARWIN
981
8bec850dc9c7 altivec patches by Romain Dolbeau
bellard
parents: 978
diff changeset
65 #define vcii(a,b,c,d) (const vector float)(FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d)
1033
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
66 #else
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
67 #define vcii(a,b,c,d) (const vector float){FLOAT_ ## a, FLOAT_ ## b, FLOAT_ ## c, FLOAT_ ## d}
b4172ff70d27 Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents: 1024
diff changeset
68 #endif
1009
3b7cc8e4b83f AltiVec perf (take 2), plus a couple AltiVec functions by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 995
diff changeset
69
3983
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
70 // Transpose 8x8 matrix of 16-bit elements (in-place)
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
71 #define TRANSPOSE8(a,b,c,d,e,f,g,h) \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
72 do { \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
73 vector signed short A1, B1, C1, D1, E1, F1, G1, H1; \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
74 vector signed short A2, B2, C2, D2, E2, F2, G2, H2; \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
75 \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
76 A1 = vec_mergeh (a, e); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
77 B1 = vec_mergel (a, e); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
78 C1 = vec_mergeh (b, f); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
79 D1 = vec_mergel (b, f); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
80 E1 = vec_mergeh (c, g); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
81 F1 = vec_mergel (c, g); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
82 G1 = vec_mergeh (d, h); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
83 H1 = vec_mergel (d, h); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
84 \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
85 A2 = vec_mergeh (A1, E1); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
86 B2 = vec_mergel (A1, E1); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
87 C2 = vec_mergeh (B1, F1); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
88 D2 = vec_mergel (B1, F1); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
89 E2 = vec_mergeh (C1, G1); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
90 F2 = vec_mergel (C1, G1); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
91 G2 = vec_mergeh (D1, H1); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
92 H2 = vec_mergel (D1, H1); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
93 \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
94 a = vec_mergeh (A2, E2); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
95 b = vec_mergel (A2, E2); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
96 c = vec_mergeh (B2, F2); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
97 d = vec_mergel (B2, F2); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
98 e = vec_mergeh (C2, G2); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
99 f = vec_mergel (C2, G2); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
100 g = vec_mergeh (D2, H2); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
101 h = vec_mergel (D2, H2); \
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
102 } while (0)
fec25fd9febf Move TRANSPOSE8 macro to dsputil_altivec.h.
gpoirier
parents: 3973
diff changeset
103
1009
3b7cc8e4b83f AltiVec perf (take 2), plus a couple AltiVec functions by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 995
diff changeset
104 #endif /* HAVE_ALTIVEC */
1015
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
105
35cf2f4a0f8c PPC perf, PPC clear_block, AltiVec put_pixels8_xy2 patch by (Romain Dolbeau <dolbeau at irisa dot fr>)
michaelni
parents: 1009
diff changeset
106 #endif /* _DSPUTIL_ALTIVEC_ */