2 * This file is part of FFmpeg.
4 * FFmpeg is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2.1 of the License, or (at your option) any later version.
9 * FFmpeg is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with FFmpeg; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "libavutil/attributes.h"
20 #include "libavutil/avassert.h"
21 #include "libavutil/cpu.h"
22 #include "libavutil/x86/cpu.h"
23 #include "libavcodec/avcodec.h"
24 #include "libavcodec/mpegvideoencdsp.h"
26 int ff_pix_sum16_mmx(uint8_t *pix, int line_size);
27 int ff_pix_sum16_sse2(uint8_t *pix, int line_size);
28 int ff_pix_sum16_xop(uint8_t *pix, int line_size);
29 int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
30 int ff_pix_norm1_sse2(uint8_t *pix, int line_size);
34 #define PHADDD(a, t) \
35 "movq " #a ", " #t " \n\t" \
36 "psrlq $32, " #a " \n\t" \
37 "paddd " #t ", " #a " \n\t"
40 * pmulhw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15])[16 - 31]
41 * pmulhrw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x8000)[16 - 31]
42 * pmulhrsw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x4000)[15 - 30]
44 #define PMULHRW(x, y, s, o) \
45 "pmulhw " #s ", " #x " \n\t" \
46 "pmulhw " #s ", " #y " \n\t" \
47 "paddw " #o ", " #x " \n\t" \
48 "paddw " #o ", " #y " \n\t" \
49 "psraw $1, " #x " \n\t" \
50 "psraw $1, " #y " \n\t"
51 #define DEF(x) x ## _mmx
52 #define SET_RND MOVQ_WONE
53 #define SCALE_OFFSET 1
55 #include "mpegvideoenc_qns_template.c"
62 #define DEF(x) x ## _3dnow
64 #define SCALE_OFFSET 0
65 #define PMULHRW(x, y, s, o) \
66 "pmulhrw " #s ", " #x " \n\t" \
67 "pmulhrw " #s ", " #y " \n\t"
69 #include "mpegvideoenc_qns_template.c"
78 #define DEF(x) x ## _ssse3
80 #define SCALE_OFFSET -1
82 #define PHADDD(a, t) \
83 "pshufw $0x0E, " #a ", " #t " \n\t" \
84 /* faster than phaddd on core2 */ \
85 "paddd " #t ", " #a " \n\t"
87 #define PMULHRW(x, y, s, o) \
88 "pmulhrsw " #s ", " #x " \n\t" \
89 "pmulhrsw " #s ", " #y " \n\t"
91 #include "mpegvideoenc_qns_template.c"
98 #endif /* HAVE_SSSE3_INLINE */
100 /* Draw the edges of width 'w' of an image of size width, height
101 * this MMX version can only handle w == 8 || w == 16. */
102 static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
103 int w, int h, int sides)
105 uint8_t *ptr, *last_line;
108 last_line = buf + (height - 1) * wrap;
114 "movd (%0), %%mm0 \n\t"
115 "punpcklbw %%mm0, %%mm0 \n\t"
116 "punpcklwd %%mm0, %%mm0 \n\t"
117 "punpckldq %%mm0, %%mm0 \n\t"
118 "movq %%mm0, -8(%0) \n\t"
119 "movq -8(%0, %2), %%mm1 \n\t"
120 "punpckhbw %%mm1, %%mm1 \n\t"
121 "punpckhwd %%mm1, %%mm1 \n\t"
122 "punpckhdq %%mm1, %%mm1 \n\t"
123 "movq %%mm1, (%0, %2) \n\t"
128 : "r" ((x86_reg) wrap), "r" ((x86_reg) width),
129 "r" (ptr + wrap * height));
130 } else if (w == 16) {
133 "movd (%0), %%mm0 \n\t"
134 "punpcklbw %%mm0, %%mm0 \n\t"
135 "punpcklwd %%mm0, %%mm0 \n\t"
136 "punpckldq %%mm0, %%mm0 \n\t"
137 "movq %%mm0, -8(%0) \n\t"
138 "movq %%mm0, -16(%0) \n\t"
139 "movq -8(%0, %2), %%mm1 \n\t"
140 "punpckhbw %%mm1, %%mm1 \n\t"
141 "punpckhwd %%mm1, %%mm1 \n\t"
142 "punpckhdq %%mm1, %%mm1 \n\t"
143 "movq %%mm1, (%0, %2) \n\t"
144 "movq %%mm1, 8(%0, %2) \n\t"
149 : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
155 "movd (%0), %%mm0 \n\t"
156 "punpcklbw %%mm0, %%mm0 \n\t"
157 "punpcklwd %%mm0, %%mm0 \n\t"
158 "movd %%mm0, -4(%0) \n\t"
159 "movd -4(%0, %2), %%mm1 \n\t"
160 "punpcklbw %%mm1, %%mm1 \n\t"
161 "punpckhwd %%mm1, %%mm1 \n\t"
162 "punpckhdq %%mm1, %%mm1 \n\t"
163 "movd %%mm1, (%0, %2) \n\t"
168 : "r" ((x86_reg) wrap), "r" ((x86_reg) width),
169 "r" (ptr + wrap * height));
172 /* top and bottom (and hopefully also the corners) */
173 if (sides & EDGE_TOP) {
174 for (i = 0; i < h; i += 4) {
175 ptr = buf - (i + 1) * wrap - w;
178 "movq (%1, %0), %%mm0 \n\t"
179 "movq %%mm0, (%0) \n\t"
180 "movq %%mm0, (%0, %2) \n\t"
181 "movq %%mm0, (%0, %2, 2) \n\t"
182 "movq %%mm0, (%0, %3) \n\t"
187 : "r" ((x86_reg) buf - (x86_reg) ptr - w),
188 "r" ((x86_reg) - wrap), "r" ((x86_reg) - wrap * 3),
189 "r" (ptr + width + 2 * w));
193 if (sides & EDGE_BOTTOM) {
194 for (i = 0; i < h; i += 4) {
195 ptr = last_line + (i + 1) * wrap - w;
198 "movq (%1, %0), %%mm0 \n\t"
199 "movq %%mm0, (%0) \n\t"
200 "movq %%mm0, (%0, %2) \n\t"
201 "movq %%mm0, (%0, %2, 2) \n\t"
202 "movq %%mm0, (%0, %3) \n\t"
207 : "r" ((x86_reg) last_line - (x86_reg) ptr - w),
208 "r" ((x86_reg) wrap), "r" ((x86_reg) wrap * 3),
209 "r" (ptr + width + 2 * w));
214 #endif /* HAVE_INLINE_ASM */
216 av_cold void ff_mpegvideoencdsp_init_x86(MpegvideoEncDSPContext *c,
217 AVCodecContext *avctx)
219 int cpu_flags = av_get_cpu_flags();
221 if (EXTERNAL_MMX(cpu_flags)) {
222 c->pix_sum = ff_pix_sum16_mmx;
223 c->pix_norm1 = ff_pix_norm1_mmx;
226 if (EXTERNAL_SSE2(cpu_flags)) {
227 c->pix_sum = ff_pix_sum16_sse2;
228 c->pix_norm1 = ff_pix_norm1_sse2;
231 if (EXTERNAL_XOP(cpu_flags)) {
232 c->pix_sum = ff_pix_sum16_xop;
237 if (INLINE_MMX(cpu_flags)) {
238 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
239 c->try_8x8basis = try_8x8basis_mmx;
241 c->add_8x8basis = add_8x8basis_mmx;
243 if (avctx->bits_per_raw_sample <= 8) {
244 c->draw_edges = draw_edges_mmx;
248 if (INLINE_AMD3DNOW(cpu_flags)) {
249 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
250 c->try_8x8basis = try_8x8basis_3dnow;
252 c->add_8x8basis = add_8x8basis_3dnow;
255 #if HAVE_SSSE3_INLINE
256 if (INLINE_SSSE3(cpu_flags)) {
257 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
258 c->try_8x8basis = try_8x8basis_ssse3;
260 c->add_8x8basis = add_8x8basis_ssse3;
262 #endif /* HAVE_SSSE3_INLINE */
264 #endif /* HAVE_INLINE_ASM */