2 * SIMD-optimized halfpel functions
3 * Copyright (c) 2000, 2001 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
25 #include "libavutil/attributes.h"
26 #include "libavutil/cpu.h"
27 #include "libavutil/x86/cpu.h"
28 #include "libavcodec/avcodec.h"
29 #include "libavcodec/hpeldsp.h"
30 #include "libavcodec/pixels.h"
34 void ff_put_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
35 ptrdiff_t line_size, int h);
36 void ff_put_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
37 ptrdiff_t line_size, int h);
38 void ff_put_pixels16_x2_mmxext(uint8_t *block, const uint8_t *pixels,
39 ptrdiff_t line_size, int h);
40 void ff_put_pixels16_x2_3dnow(uint8_t *block, const uint8_t *pixels,
41 ptrdiff_t line_size, int h);
42 void ff_put_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
43 ptrdiff_t line_size, int h);
44 void ff_avg_pixels16_x2_sse2(uint8_t *block, const uint8_t *pixels,
45 ptrdiff_t line_size, int h);
46 void ff_put_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
47 ptrdiff_t line_size, int h);
48 void ff_avg_pixels16_y2_sse2(uint8_t *block, const uint8_t *pixels,
49 ptrdiff_t line_size, int h);
50 void ff_put_no_rnd_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
51 ptrdiff_t line_size, int h);
52 void ff_put_no_rnd_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
53 ptrdiff_t line_size, int h);
54 void ff_put_no_rnd_pixels8_x2_exact_mmxext(uint8_t *block,
55 const uint8_t *pixels,
56 ptrdiff_t line_size, int h);
57 void ff_put_no_rnd_pixels8_x2_exact_3dnow(uint8_t *block,
58 const uint8_t *pixels,
59 ptrdiff_t line_size, int h);
60 void ff_put_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
61 ptrdiff_t line_size, int h);
62 void ff_put_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
63 ptrdiff_t line_size, int h);
64 void ff_put_no_rnd_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
65 ptrdiff_t line_size, int h);
66 void ff_put_no_rnd_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
67 ptrdiff_t line_size, int h);
68 void ff_put_no_rnd_pixels8_y2_exact_mmxext(uint8_t *block,
69 const uint8_t *pixels,
70 ptrdiff_t line_size, int h);
71 void ff_put_no_rnd_pixels8_y2_exact_3dnow(uint8_t *block,
72 const uint8_t *pixels,
73 ptrdiff_t line_size, int h);
74 void ff_avg_pixels8_3dnow(uint8_t *block, const uint8_t *pixels,
75 ptrdiff_t line_size, int h);
76 void ff_avg_pixels8_x2_mmxext(uint8_t *block, const uint8_t *pixels,
77 ptrdiff_t line_size, int h);
78 void ff_avg_pixels8_x2_3dnow(uint8_t *block, const uint8_t *pixels,
79 ptrdiff_t line_size, int h);
80 void ff_avg_pixels8_y2_mmxext(uint8_t *block, const uint8_t *pixels,
81 ptrdiff_t line_size, int h);
82 void ff_avg_pixels8_y2_3dnow(uint8_t *block, const uint8_t *pixels,
83 ptrdiff_t line_size, int h);
84 void ff_avg_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
85 ptrdiff_t line_size, int h);
86 void ff_avg_approx_pixels8_xy2_mmxext(uint8_t *block, const uint8_t *pixels,
87 ptrdiff_t line_size, int h);
88 void ff_avg_approx_pixels8_xy2_3dnow(uint8_t *block, const uint8_t *pixels,
89 ptrdiff_t line_size, int h);
91 #define avg_pixels8_mmx ff_avg_pixels8_mmx
92 #define avg_pixels8_x2_mmx ff_avg_pixels8_x2_mmx
93 #define avg_pixels16_mmx ff_avg_pixels16_mmx
94 #define avg_pixels8_xy2_mmx ff_avg_pixels8_xy2_mmx
95 #define avg_pixels16_xy2_mmx ff_avg_pixels16_xy2_mmx
96 #define put_pixels8_mmx ff_put_pixels8_mmx
97 #define put_pixels16_mmx ff_put_pixels16_mmx
98 #define put_pixels8_xy2_mmx ff_put_pixels8_xy2_mmx
99 #define put_pixels16_xy2_mmx ff_put_pixels16_xy2_mmx
100 #define avg_no_rnd_pixels16_mmx ff_avg_pixels16_mmx
101 #define put_no_rnd_pixels8_mmx ff_put_pixels8_mmx
102 #define put_no_rnd_pixels16_mmx ff_put_pixels16_mmx
106 /***********************************/
107 /* MMX no rounding */
108 #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
109 #define SET_RND MOVQ_WONE
110 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
111 #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e)
112 #define STATIC static
114 #include "rnd_template.c"
115 #include "hpeldsp_rnd_template.c"
124 CALL_2X_PIXELS(avg_no_rnd_pixels16_y2_mmx, avg_no_rnd_pixels8_y2_mmx, 8)
125 CALL_2X_PIXELS(put_no_rnd_pixels16_y2_mmx, put_no_rnd_pixels8_y2_mmx, 8)
127 CALL_2X_PIXELS(avg_no_rnd_pixels16_xy2_mmx, avg_no_rnd_pixels8_xy2_mmx, 8)
128 CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_mmx, put_no_rnd_pixels8_xy2_mmx, 8)
131 /***********************************/
134 #define DEF(x, y) x ## _ ## y ## _mmx
135 #define SET_RND MOVQ_WTWO
136 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f)
137 #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e)
139 #include "hpeldsp_rnd_template.c"
142 #define DEF(x, y) ff_ ## x ## _ ## y ## _mmx
145 #include "rnd_template.c"
153 CALL_2X_PIXELS(avg_pixels16_y2_mmx, avg_pixels8_y2_mmx, 8)
154 CALL_2X_PIXELS(put_pixels16_y2_mmx, put_pixels8_y2_mmx, 8)
156 CALL_2X_PIXELS_EXPORT(ff_avg_pixels16_xy2_mmx, ff_avg_pixels8_xy2_mmx, 8)
157 CALL_2X_PIXELS_EXPORT(ff_put_pixels16_xy2_mmx, ff_put_pixels8_xy2_mmx, 8)
160 #endif /* HAVE_INLINE_ASM */
165 #define HPELDSP_AVG_PIXELS16(CPUEXT) \
166 CALL_2X_PIXELS(put_no_rnd_pixels16_x2 ## CPUEXT, ff_put_no_rnd_pixels8_x2 ## CPUEXT, 8) \
167 CALL_2X_PIXELS(put_pixels16_y2 ## CPUEXT, ff_put_pixels8_y2 ## CPUEXT, 8) \
168 CALL_2X_PIXELS(put_no_rnd_pixels16_y2 ## CPUEXT, ff_put_no_rnd_pixels8_y2 ## CPUEXT, 8) \
169 CALL_2X_PIXELS(avg_pixels16 ## CPUEXT, ff_avg_pixels8 ## CPUEXT, 8) \
170 CALL_2X_PIXELS(avg_pixels16_x2 ## CPUEXT, ff_avg_pixels8_x2 ## CPUEXT, 8) \
171 CALL_2X_PIXELS(avg_pixels16_y2 ## CPUEXT, ff_avg_pixels8_y2 ## CPUEXT, 8) \
172 CALL_2X_PIXELS(avg_pixels16_xy2 ## CPUEXT, ff_avg_pixels8_xy2 ## CPUEXT, 8) \
173 CALL_2X_PIXELS(avg_approx_pixels16_xy2## CPUEXT, ff_avg_approx_pixels8_xy2## CPUEXT, 8)
175 HPELDSP_AVG_PIXELS16(_3dnow)
176 HPELDSP_AVG_PIXELS16(_mmxext)
178 #endif /* HAVE_YASM */
180 #define SET_HPEL_FUNCS_EXT(PFX, IDX, SIZE, CPU) \
181 if (HAVE_MMX_EXTERNAL) \
182 c->PFX ## _pixels_tab IDX [0] = PFX ## _pixels ## SIZE ## _ ## CPU;
185 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
187 SET_HPEL_FUNCS_EXT(PFX, IDX, SIZE, CPU) \
188 c->PFX ## _pixels_tab IDX [1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
189 c->PFX ## _pixels_tab IDX [2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
190 c->PFX ## _pixels_tab IDX [3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \
193 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
195 SET_HPEL_FUNCS_EXT(PFX, IDX, SIZE, CPU) \
199 static void hpeldsp_init_mmx(HpelDSPContext *c, int flags, int cpu_flags)
201 SET_HPEL_FUNCS(put, [0], 16, mmx);
202 SET_HPEL_FUNCS(put_no_rnd, [0], 16, mmx);
203 SET_HPEL_FUNCS(avg, [0], 16, mmx);
204 SET_HPEL_FUNCS(avg_no_rnd, , 16, mmx);
205 SET_HPEL_FUNCS(put, [1], 8, mmx);
206 SET_HPEL_FUNCS(put_no_rnd, [1], 8, mmx);
207 if (HAVE_MMX_EXTERNAL) {
208 c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmx;
209 c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmx;
212 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx;
213 c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmx;
217 static void hpeldsp_init_mmxext(HpelDSPContext *c, int flags, int cpu_flags)
219 #if HAVE_MMXEXT_EXTERNAL
220 c->put_pixels_tab[0][1] = ff_put_pixels16_x2_mmxext;
221 c->put_pixels_tab[0][2] = put_pixels16_y2_mmxext;
223 c->avg_pixels_tab[0][0] = avg_pixels16_mmxext;
224 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmxext;
225 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmxext;
226 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmxext;
228 c->put_pixels_tab[1][1] = ff_put_pixels8_x2_mmxext;
229 c->put_pixels_tab[1][2] = ff_put_pixels8_y2_mmxext;
231 c->avg_pixels_tab[1][0] = ff_avg_pixels8_mmxext;
232 c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_mmxext;
233 c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_mmxext;
234 c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_mmxext;
236 if (!(flags & AV_CODEC_FLAG_BITEXACT)) {
237 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmxext;
238 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmxext;
239 c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_mmxext;
240 c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_mmxext;
242 c->avg_pixels_tab[0][3] = avg_approx_pixels16_xy2_mmxext;
243 c->avg_pixels_tab[1][3] = ff_avg_approx_pixels8_xy2_mmxext;
246 if (CONFIG_VP3_DECODER && flags & AV_CODEC_FLAG_BITEXACT) {
247 c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_mmxext;
248 c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_mmxext;
250 #endif /* HAVE_MMXEXT_EXTERNAL */
253 static void hpeldsp_init_3dnow(HpelDSPContext *c, int flags, int cpu_flags)
255 #if HAVE_AMD3DNOW_EXTERNAL
256 c->put_pixels_tab[0][1] = ff_put_pixels16_x2_3dnow;
257 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
259 c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
260 c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
261 c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
262 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
264 c->put_pixels_tab[1][1] = ff_put_pixels8_x2_3dnow;
265 c->put_pixels_tab[1][2] = ff_put_pixels8_y2_3dnow;
267 c->avg_pixels_tab[1][0] = ff_avg_pixels8_3dnow;
268 c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_3dnow;
269 c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_3dnow;
270 c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_3dnow;
272 if (!(flags & AV_CODEC_FLAG_BITEXACT)){
273 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
274 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
275 c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_3dnow;
276 c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_3dnow;
278 c->avg_pixels_tab[0][3] = avg_approx_pixels16_xy2_3dnow;
279 c->avg_pixels_tab[1][3] = ff_avg_approx_pixels8_xy2_3dnow;
282 if (CONFIG_VP3_DECODER && flags & AV_CODEC_FLAG_BITEXACT) {
283 c->put_no_rnd_pixels_tab[1][1] = ff_put_no_rnd_pixels8_x2_exact_3dnow;
284 c->put_no_rnd_pixels_tab[1][2] = ff_put_no_rnd_pixels8_y2_exact_3dnow;
286 #endif /* HAVE_AMD3DNOW_EXTERNAL */
289 static void hpeldsp_init_sse2(HpelDSPContext *c, int flags, int cpu_flags)
291 #if HAVE_SSE2_EXTERNAL
292 if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW)) {
293 // these functions are slower than mmx on AMD, but faster on Intel
294 c->put_pixels_tab[0][0] = ff_put_pixels16_sse2;
295 c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_sse2;
296 c->put_pixels_tab[0][1] = ff_put_pixels16_x2_sse2;
297 c->put_pixels_tab[0][2] = ff_put_pixels16_y2_sse2;
298 c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_sse2;
299 c->avg_pixels_tab[0][0] = ff_avg_pixels16_sse2;
300 c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_sse2;
301 c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_sse2;
302 c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_sse2;
304 #endif /* HAVE_SSE2_EXTERNAL */
307 static void hpeldsp_init_ssse3(HpelDSPContext *c, int flags, int cpu_flags)
309 #if HAVE_SSSE3_EXTERNAL
310 c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_ssse3;
311 c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_ssse3;
312 c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_ssse3;
313 c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_ssse3;
317 av_cold void ff_hpeldsp_init_x86(HpelDSPContext *c, int flags)
319 int cpu_flags = av_get_cpu_flags();
321 if (INLINE_MMX(cpu_flags))
322 hpeldsp_init_mmx(c, flags, cpu_flags);
324 if (EXTERNAL_AMD3DNOW(cpu_flags))
325 hpeldsp_init_3dnow(c, flags, cpu_flags);
327 if (EXTERNAL_MMXEXT(cpu_flags))
328 hpeldsp_init_mmxext(c, flags, cpu_flags);
330 if (EXTERNAL_SSE2(cpu_flags))
331 hpeldsp_init_sse2(c, flags, cpu_flags);
333 if (EXTERNAL_SSSE3(cpu_flags))
334 hpeldsp_init_ssse3(c, flags, cpu_flags);