Mercurial > libavcodec.hg

comparison x86/vp3dsp_sse2.c @ 8430:7768bdfd4f7b libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Rename libavcodec/i386/ --> libavcodec/x86/. It contains optimizations that are not specific to i386 and libavutil uses this naming scheme already.
author diego
date Mon, 22 Dec 2008 09:12:42 +0000
parents
children e9d9d946f213
comparison
equal deleted inserted replaced
8429:b3ecaba81501 8430:7768bdfd4f7b
1 /*
2 * Copyright (C) 2004 the ffmpeg project
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 /**
22 * @file vp3dsp_sse2.c
23 * SSE2-optimized functions cribbed from the original VP3 source code.
24 */
25
26 #include "libavcodec/dsputil.h"
27 #include "dsputil_mmx.h"
28
29 DECLARE_ALIGNED_16(const uint16_t, ff_vp3_idct_data[7 * 8]) =
30 {
31 64277,64277,64277,64277,64277,64277,64277,64277,
32 60547,60547,60547,60547,60547,60547,60547,60547,
33 54491,54491,54491,54491,54491,54491,54491,54491,
34 46341,46341,46341,46341,46341,46341,46341,46341,
35 36410,36410,36410,36410,36410,36410,36410,36410,
36 25080,25080,25080,25080,25080,25080,25080,25080,
37 12785,12785,12785,12785,12785,12785,12785,12785
38 };
39
40
41 #define VP3_1D_IDCT_SSE2(ADD, SHIFT) \
42 "movdqa "I(3)", %%xmm2 \n\t" /* xmm2 = i3 */ \
43 "movdqa "C(3)", %%xmm6 \n\t" /* xmm6 = c3 */ \
44 "movdqa %%xmm2, %%xmm4 \n\t" /* xmm4 = i3 */ \
45 "movdqa "I(5)", %%xmm7 \n\t" /* xmm7 = i5 */ \
46 "pmulhw %%xmm6, %%xmm4 \n\t" /* xmm4 = c3 * i3 - i3 */ \
47 "movdqa "C(5)", %%xmm1 \n\t" /* xmm1 = c5 */ \
48 "pmulhw %%xmm7, %%xmm6 \n\t" /* xmm6 = c3 * i5 - i5 */ \
49 "movdqa %%xmm1, %%xmm5 \n\t" /* xmm5 = c5 */ \
50 "pmulhw %%xmm2, %%xmm1 \n\t" /* xmm1 = c5 * i3 - i3 */ \
51 "movdqa "I(1)", %%xmm3 \n\t" /* xmm3 = i1 */ \
52 "pmulhw %%xmm7, %%xmm5 \n\t" /* xmm5 = c5 * i5 - i5 */ \
53 "movdqa "C(1)", %%xmm0 \n\t" /* xmm0 = c1 */ \
54 "paddw %%xmm2, %%xmm4 \n\t" /* xmm4 = c3 * i3 */ \
55 "paddw %%xmm7, %%xmm6 \n\t" /* xmm6 = c3 * i5 */ \
56 "paddw %%xmm1, %%xmm2 \n\t" /* xmm2 = c5 * i3 */ \
57 "movdqa "I(7)", %%xmm1 \n\t" /* xmm1 = i7 */ \
58 "paddw %%xmm5, %%xmm7 \n\t" /* xmm7 = c5 * i5 */ \
59 "movdqa %%xmm0, %%xmm5 \n\t" /* xmm5 = c1 */ \
60 "pmulhw %%xmm3, %%xmm0 \n\t" /* xmm0 = c1 * i1 - i1 */ \
61 "paddsw %%xmm7, %%xmm4 \n\t" /* xmm4 = c3 * i3 + c5 * i5 = C */ \
62 "pmulhw %%xmm1, %%xmm5 \n\t" /* xmm5 = c1 * i7 - i7 */ \
63 "movdqa "C(7)", %%xmm7 \n\t" /* xmm7 = c7 */ \
64 "psubsw %%xmm2, %%xmm6 \n\t" /* xmm6 = c3 * i5 - c5 * i3 = D */ \
65 "paddw %%xmm3, %%xmm0 \n\t" /* xmm0 = c1 * i1 */ \
66 "pmulhw %%xmm7, %%xmm3 \n\t" /* xmm3 = c7 * i1 */ \
67 "movdqa "I(2)", %%xmm2 \n\t" /* xmm2 = i2 */ \
68 "pmulhw %%xmm1, %%xmm7 \n\t" /* xmm7 = c7 * i7 */ \
69 "paddw %%xmm1, %%xmm5 \n\t" /* xmm5 = c1 * i7 */ \
70 "movdqa %%xmm2, %%xmm1 \n\t" /* xmm1 = i2 */ \
71 "pmulhw "C(2)", %%xmm2 \n\t" /* xmm2 = i2 * c2 -i2 */ \
72 "psubsw %%xmm5, %%xmm3 \n\t" /* xmm3 = c7 * i1 - c1 * i7 = B */ \
73 "movdqa "I(6)", %%xmm5 \n\t" /* xmm5 = i6 */ \
74 "paddsw %%xmm7, %%xmm0 \n\t" /* xmm0 = c1 * i1 + c7 * i7 = A */ \
75 "movdqa %%xmm5, %%xmm7 \n\t" /* xmm7 = i6 */ \
76 "psubsw %%xmm4, %%xmm0 \n\t" /* xmm0 = A - C */ \
77 "pmulhw "C(2)", %%xmm5 \n\t" /* xmm5 = c2 * i6 - i6 */ \
78 "paddw %%xmm1, %%xmm2 \n\t" /* xmm2 = i2 * c2 */ \
79 "pmulhw "C(6)", %%xmm1 \n\t" /* xmm1 = c6 * i2 */ \
80 "paddsw %%xmm4, %%xmm4 \n\t" /* xmm4 = C + C */ \
81 "paddsw %%xmm0, %%xmm4 \n\t" /* xmm4 = A + C = C. */ \
82 "psubsw %%xmm6, %%xmm3 \n\t" /* xmm3 = B - D */ \
83 "paddw %%xmm7, %%xmm5 \n\t" /* xmm5 = c2 * i6 */ \
84 "paddsw %%xmm6, %%xmm6 \n\t" /* xmm6 = D + D */ \
85 "pmulhw "C(6)", %%xmm7 \n\t" /* xmm7 = c6 * i6 */ \
86 "paddsw %%xmm3, %%xmm6 \n\t" /* xmm6 = B + D = D. */ \
87 "movdqa %%xmm4, "I(1)" \n\t" /* Save C. at I(1) */ \
88 "psubsw %%xmm5, %%xmm1 \n\t" /* xmm1 = c6 * i2 - c2 * i6 = H */ \
89 "movdqa "C(4)", %%xmm4 \n\t" /* xmm4 = c4 */ \
90 "movdqa %%xmm3, %%xmm5 \n\t" /* xmm5 = B - D */ \
91 "pmulhw %%xmm4, %%xmm3 \n\t" /* xmm3 = ( c4 -1 ) * ( B - D ) */ \
92 "paddsw %%xmm2, %%xmm7 \n\t" /* xmm7 = c2 * i2 + c6 * i6 = G */ \
93 "movdqa %%xmm6, "I(2)" \n\t" /* Save D. at I(2) */ \
94 "movdqa %%xmm0, %%xmm2 \n\t" /* xmm2 = A - C */ \
95 "movdqa "I(0)", %%xmm6 \n\t" /* xmm6 = i0 */ \
96 "pmulhw %%xmm4, %%xmm0 \n\t" /* xmm0 = ( c4 - 1 ) * ( A - C ) = A. */ \
97 "paddw %%xmm3, %%xmm5 \n\t" /* xmm5 = c4 * ( B - D ) = B. */ \
98 "movdqa "I(4)", %%xmm3 \n\t" /* xmm3 = i4 */ \
99 "psubsw %%xmm1, %%xmm5 \n\t" /* xmm5 = B. - H = B.. */ \
100 "paddw %%xmm0, %%xmm2 \n\t" /* xmm2 = c4 * ( A - C) = A. */ \
101 "psubsw %%xmm3, %%xmm6 \n\t" /* xmm6 = i0 - i4 */ \
102 "movdqa %%xmm6, %%xmm0 \n\t" /* xmm0 = i0 - i4 */ \
103 "pmulhw %%xmm4, %%xmm6 \n\t" /* xmm6 = (c4 - 1) * (i0 - i4) = F */ \
104 "paddsw %%xmm3, %%xmm3 \n\t" /* xmm3 = i4 + i4 */ \
105 "paddsw %%xmm1, %%xmm1 \n\t" /* xmm1 = H + H */ \
106 "paddsw %%xmm0, %%xmm3 \n\t" /* xmm3 = i0 + i4 */ \
107 "paddsw %%xmm5, %%xmm1 \n\t" /* xmm1 = B. + H = H. */ \
108 "pmulhw %%xmm3, %%xmm4 \n\t" /* xmm4 = ( c4 - 1 ) * ( i0 + i4 ) */ \
109 "paddw %%xmm0, %%xmm6 \n\t" /* xmm6 = c4 * ( i0 - i4 ) */ \
110 "psubsw %%xmm2, %%xmm6 \n\t" /* xmm6 = F - A. = F. */ \
111 "paddsw %%xmm2, %%xmm2 \n\t" /* xmm2 = A. + A. */ \
112 "movdqa "I(1)", %%xmm0 \n\t" /* Load C. from I(1) */ \
113 "paddsw %%xmm6, %%xmm2 \n\t" /* xmm2 = F + A. = A.. */ \
114 "paddw %%xmm3, %%xmm4 \n\t" /* xmm4 = c4 * ( i0 + i4 ) = 3 */ \
115 "psubsw %%xmm1, %%xmm2 \n\t" /* xmm2 = A.. - H. = R2 */ \
116 ADD(%%xmm2) /* Adjust R2 and R1 before shifting */ \
117 "paddsw %%xmm1, %%xmm1 \n\t" /* xmm1 = H. + H. */ \
118 "paddsw %%xmm2, %%xmm1 \n\t" /* xmm1 = A.. + H. = R1 */ \
119 SHIFT(%%xmm2) /* xmm2 = op2 */ \
120 "psubsw %%xmm7, %%xmm4 \n\t" /* xmm4 = E - G = E. */ \
121 SHIFT(%%xmm1) /* xmm1 = op1 */ \
122 "movdqa "I(2)", %%xmm3 \n\t" /* Load D. from I(2) */ \
123 "paddsw %%xmm7, %%xmm7 \n\t" /* xmm7 = G + G */ \
124 "paddsw %%xmm4, %%xmm7 \n\t" /* xmm7 = E + G = G. */ \
125 "psubsw %%xmm3, %%xmm4 \n\t" /* xmm4 = E. - D. = R4 */ \
126 ADD(%%xmm4) /* Adjust R4 and R3 before shifting */ \
127 "paddsw %%xmm3, %%xmm3 \n\t" /* xmm3 = D. + D. */ \
128 "paddsw %%xmm4, %%xmm3 \n\t" /* xmm3 = E. + D. = R3 */ \
129 SHIFT(%%xmm4) /* xmm4 = op4 */ \
130 "psubsw %%xmm5, %%xmm6 \n\t" /* xmm6 = F. - B..= R6 */ \
131 SHIFT(%%xmm3) /* xmm3 = op3 */ \
132 ADD(%%xmm6) /* Adjust R6 and R5 before shifting */ \
133 "paddsw %%xmm5, %%xmm5 \n\t" /* xmm5 = B.. + B.. */ \
134 "paddsw %%xmm6, %%xmm5 \n\t" /* xmm5 = F. + B.. = R5 */ \
135 SHIFT(%%xmm6) /* xmm6 = op6 */ \
136 SHIFT(%%xmm5) /* xmm5 = op5 */ \
137 "psubsw %%xmm0, %%xmm7 \n\t" /* xmm7 = G. - C. = R7 */ \
138 ADD(%%xmm7) /* Adjust R7 and R0 before shifting */ \
139 "paddsw %%xmm0, %%xmm0 \n\t" /* xmm0 = C. + C. */ \
140 "paddsw %%xmm7, %%xmm0 \n\t" /* xmm0 = G. + C. */ \
141 SHIFT(%%xmm7) /* xmm7 = op7 */ \
142 SHIFT(%%xmm0) /* xmm0 = op0 */
143
144 #define PUT_BLOCK(r0, r1, r2, r3, r4, r5, r6, r7) \
145 "movdqa " #r0 ", " O(0) "\n\t" \
146 "movdqa " #r1 ", " O(1) "\n\t" \
147 "movdqa " #r2 ", " O(2) "\n\t" \
148 "movdqa " #r3 ", " O(3) "\n\t" \
149 "movdqa " #r4 ", " O(4) "\n\t" \
150 "movdqa " #r5 ", " O(5) "\n\t" \
151 "movdqa " #r6 ", " O(6) "\n\t" \
152 "movdqa " #r7 ", " O(7) "\n\t"
153
154 #define NOP(xmm)
155 #define SHIFT4(xmm) "psraw $4, "#xmm"\n\t"
156 #define ADD8(xmm) "paddsw %2, "#xmm"\n\t"
157
158 void ff_vp3_idct_sse2(int16_t *input_data)
159 {
160 #define I(x) AV_STRINGIFY(16*x)"(%0)"
161 #define O(x) I(x)
162 #define C(x) AV_STRINGIFY(16*(x-1))"(%1)"
163
164 __asm__ volatile (
165 VP3_1D_IDCT_SSE2(NOP, NOP)
166
167 TRANSPOSE8(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7, (%0))
168 PUT_BLOCK(%%xmm0, %%xmm5, %%xmm7, %%xmm3, %%xmm6, %%xmm4, %%xmm2, %%xmm1)
169
170 VP3_1D_IDCT_SSE2(ADD8, SHIFT4)
171 PUT_BLOCK(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7)
172 :: "r"(input_data), "r"(ff_vp3_idct_data), "m"(ff_pw_8)
173 );
174 }
175
176 void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block)
177 {
178 ff_vp3_idct_sse2(block);
179 put_signed_pixels_clamped_mmx(block, dest, line_size);
180 }
181
182 void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block)
183 {
184 ff_vp3_idct_sse2(block);
185 add_pixels_clamped_mmx(block, dest, line_size);
186 }