00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022
00023
00024
00025 #include "libavutil/cpu.h"
00026 #include "libavutil/x86_cpu.h"
00027 #include "libavcodec/dsputil.h"
00028 #include "libavcodec/h264dsp.h"
00029 #include "libavcodec/mpegvideo.h"
00030 #include "libavcodec/simple_idct.h"
00031 #include "libavcodec/ac3dec.h"
00032 #include "dsputil_mmx.h"
00033 #include "idct_xvid.h"
00034 #include "diracdsp_mmx.h"
00035
00036
00037
00038
00039
00040 DECLARE_ALIGNED(8, const uint64_t, ff_bone) = 0x0101010101010101ULL;
00041 DECLARE_ALIGNED(8, const uint64_t, ff_wtwo) = 0x0002000200020002ULL;
00042
00043 DECLARE_ALIGNED(16, const uint64_t, ff_pdw_80000000)[2] =
00044 { 0x8000000080000000ULL, 0x8000000080000000ULL };
00045
00046 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_1) = { 0x0001000100010001ULL, 0x0001000100010001ULL };
00047 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_2) = { 0x0002000200020002ULL, 0x0002000200020002ULL };
00048 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_3) = { 0x0003000300030003ULL, 0x0003000300030003ULL };
00049 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_4) = { 0x0004000400040004ULL, 0x0004000400040004ULL };
00050 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_5) = { 0x0005000500050005ULL, 0x0005000500050005ULL };
00051 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_8) = { 0x0008000800080008ULL, 0x0008000800080008ULL };
00052 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_9) = { 0x0009000900090009ULL, 0x0009000900090009ULL };
00053 DECLARE_ALIGNED(8, const uint64_t, ff_pw_15) = 0x000F000F000F000FULL;
00054 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_16) = { 0x0010001000100010ULL, 0x0010001000100010ULL };
00055 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_17) = { 0x0011001100110011ULL, 0x0011001100110011ULL };
00056 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_18) = { 0x0012001200120012ULL, 0x0012001200120012ULL };
00057 DECLARE_ALIGNED(8, const uint64_t, ff_pw_20) = 0x0014001400140014ULL;
00058 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_27) = { 0x001B001B001B001BULL, 0x001B001B001B001BULL };
00059 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_28) = { 0x001C001C001C001CULL, 0x001C001C001C001CULL };
00060 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_32) = { 0x0020002000200020ULL, 0x0020002000200020ULL };
00061 DECLARE_ALIGNED(8, const uint64_t, ff_pw_42) = 0x002A002A002A002AULL;
00062 DECLARE_ALIGNED(8, const uint64_t, ff_pw_53) = 0x0035003500350035ULL;
00063 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_63) = { 0x003F003F003F003FULL, 0x003F003F003F003FULL };
00064 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_64) = { 0x0040004000400040ULL, 0x0040004000400040ULL };
00065 DECLARE_ALIGNED(8, const uint64_t, ff_pw_96) = 0x0060006000600060ULL;
00066 DECLARE_ALIGNED(8, const uint64_t, ff_pw_128) = 0x0080008000800080ULL;
00067 DECLARE_ALIGNED(8, const uint64_t, ff_pw_255) = 0x00ff00ff00ff00ffULL;
00068 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_512) = { 0x0200020002000200ULL, 0x0200020002000200ULL };
00069 DECLARE_ALIGNED(16, const xmm_reg, ff_pw_1019) = { 0x03FB03FB03FB03FBULL, 0x03FB03FB03FB03FBULL };
00070
00071 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_0) = { 0x0000000000000000ULL, 0x0000000000000000ULL };
00072 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_1) = { 0x0101010101010101ULL, 0x0101010101010101ULL };
00073 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_3) = { 0x0303030303030303ULL, 0x0303030303030303ULL };
00074 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_4) = { 0x0404040404040404ULL, 0x0404040404040404ULL };
00075 DECLARE_ALIGNED(8, const uint64_t, ff_pb_7) = 0x0707070707070707ULL;
00076 DECLARE_ALIGNED(8, const uint64_t, ff_pb_1F) = 0x1F1F1F1F1F1F1F1FULL;
00077 DECLARE_ALIGNED(8, const uint64_t, ff_pb_3F) = 0x3F3F3F3F3F3F3F3FULL;
00078 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_80) = { 0x8080808080808080ULL, 0x8080808080808080ULL };
00079 DECLARE_ALIGNED(8, const uint64_t, ff_pb_81) = 0x8181818181818181ULL;
00080 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_A1) = { 0xA1A1A1A1A1A1A1A1ULL, 0xA1A1A1A1A1A1A1A1ULL };
00081 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_F8) = { 0xF8F8F8F8F8F8F8F8ULL, 0xF8F8F8F8F8F8F8F8ULL };
00082 DECLARE_ALIGNED(8, const uint64_t, ff_pb_FC) = 0xFCFCFCFCFCFCFCFCULL;
00083 DECLARE_ALIGNED(16, const xmm_reg, ff_pb_FE) = { 0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL };
00084
00085 DECLARE_ALIGNED(16, const double, ff_pd_1)[2] = { 1.0, 1.0 };
00086 DECLARE_ALIGNED(16, const double, ff_pd_2)[2] = { 2.0, 2.0 };
00087
00088 #define JUMPALIGN() __asm__ volatile (".p2align 3"::)
00089 #define MOVQ_ZERO(regd) __asm__ volatile ("pxor %%"#regd", %%"#regd ::)
00090
00091 #define MOVQ_BFE(regd) \
00092 __asm__ volatile ( \
00093 "pcmpeqd %%"#regd", %%"#regd" \n\t" \
00094 "paddb %%"#regd", %%"#regd" \n\t" ::)
00095
00096 #ifndef PIC
00097 #define MOVQ_BONE(regd) __asm__ volatile ("movq %0, %%"#regd" \n\t" :: "m"(ff_bone))
00098 #define MOVQ_WTWO(regd) __asm__ volatile ("movq %0, %%"#regd" \n\t" :: "m"(ff_wtwo))
00099 #else
00100
00101
00102 #define MOVQ_BONE(regd) \
00103 __asm__ volatile ( \
00104 "pcmpeqd %%"#regd", %%"#regd" \n\t" \
00105 "psrlw $15, %%"#regd" \n\t" \
00106 "packuswb %%"#regd", %%"#regd" \n\t" ::)
00107
00108 #define MOVQ_WTWO(regd) \
00109 __asm__ volatile ( \
00110 "pcmpeqd %%"#regd", %%"#regd" \n\t" \
00111 "psrlw $15, %%"#regd" \n\t" \
00112 "psllw $1, %%"#regd" \n\t"::)
00113
00114 #endif
00115
00116
00117
00118
00119 #define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \
00120 "movq "#rega", "#regr" \n\t" \
00121 "pand "#regb", "#regr" \n\t" \
00122 "pxor "#rega", "#regb" \n\t" \
00123 "pand "#regfe", "#regb" \n\t" \
00124 "psrlq $1, "#regb" \n\t" \
00125 "paddb "#regb", "#regr" \n\t"
00126
00127 #define PAVGB_MMX(rega, regb, regr, regfe) \
00128 "movq "#rega", "#regr" \n\t" \
00129 "por "#regb", "#regr" \n\t" \
00130 "pxor "#rega", "#regb" \n\t" \
00131 "pand "#regfe", "#regb" \n\t" \
00132 "psrlq $1, "#regb" \n\t" \
00133 "psubb "#regb", "#regr" \n\t"
00134
00135
00136 #define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \
00137 "movq "#rega", "#regr" \n\t" \
00138 "movq "#regc", "#regp" \n\t" \
00139 "pand "#regb", "#regr" \n\t" \
00140 "pand "#regd", "#regp" \n\t" \
00141 "pxor "#rega", "#regb" \n\t" \
00142 "pxor "#regc", "#regd" \n\t" \
00143 "pand %%mm6, "#regb" \n\t" \
00144 "pand %%mm6, "#regd" \n\t" \
00145 "psrlq $1, "#regb" \n\t" \
00146 "psrlq $1, "#regd" \n\t" \
00147 "paddb "#regb", "#regr" \n\t" \
00148 "paddb "#regd", "#regp" \n\t"
00149
00150 #define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \
00151 "movq "#rega", "#regr" \n\t" \
00152 "movq "#regc", "#regp" \n\t" \
00153 "por "#regb", "#regr" \n\t" \
00154 "por "#regd", "#regp" \n\t" \
00155 "pxor "#rega", "#regb" \n\t" \
00156 "pxor "#regc", "#regd" \n\t" \
00157 "pand %%mm6, "#regb" \n\t" \
00158 "pand %%mm6, "#regd" \n\t" \
00159 "psrlq $1, "#regd" \n\t" \
00160 "psrlq $1, "#regb" \n\t" \
00161 "psubb "#regb", "#regr" \n\t" \
00162 "psubb "#regd", "#regp" \n\t"
00163
00164
00165
00166 #define DEF(x, y) x ## _no_rnd_ ## y ## _mmx
00167 #define SET_RND MOVQ_WONE
00168 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f)
00169 #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e)
00170 #define OP_AVG(a, b, c, e) PAVGB_MMX(a, b, c, e)
00171
00172 #include "dsputil_mmx_rnd_template.c"
00173
00174 #undef DEF
00175 #undef SET_RND
00176 #undef PAVGBP
00177 #undef PAVGB
00178
00179
00180
00181 #define DEF(x, y) x ## _ ## y ## _mmx
00182 #define SET_RND MOVQ_WTWO
00183 #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f)
00184 #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e)
00185
00186 #include "dsputil_mmx_rnd_template.c"
00187
00188 #undef DEF
00189 #undef SET_RND
00190 #undef PAVGBP
00191 #undef PAVGB
00192 #undef OP_AVG
00193
00194
00195
00196
00197 #define DEF(x) x ## _3dnow
00198 #define PAVGB "pavgusb"
00199 #define OP_AVG PAVGB
00200
00201 #include "dsputil_mmx_avg_template.c"
00202
00203 #undef DEF
00204 #undef PAVGB
00205 #undef OP_AVG
00206
00207
00208
00209
00210 #define DEF(x) x ## _mmx2
00211
00212
00213 #define PAVGB "pavgb"
00214 #define OP_AVG PAVGB
00215
00216 #include "dsputil_mmx_avg_template.c"
00217
00218 #undef DEF
00219 #undef PAVGB
00220 #undef OP_AVG
00221
00222 #define put_no_rnd_pixels16_mmx put_pixels16_mmx
00223 #define put_no_rnd_pixels8_mmx put_pixels8_mmx
00224 #define put_pixels16_mmx2 put_pixels16_mmx
00225 #define put_pixels8_mmx2 put_pixels8_mmx
00226 #define put_pixels4_mmx2 put_pixels4_mmx
00227 #define put_no_rnd_pixels16_mmx2 put_no_rnd_pixels16_mmx
00228 #define put_no_rnd_pixels8_mmx2 put_no_rnd_pixels8_mmx
00229 #define put_pixels16_3dnow put_pixels16_mmx
00230 #define put_pixels8_3dnow put_pixels8_mmx
00231 #define put_pixels4_3dnow put_pixels4_mmx
00232 #define put_no_rnd_pixels16_3dnow put_no_rnd_pixels16_mmx
00233 #define put_no_rnd_pixels8_3dnow put_no_rnd_pixels8_mmx
00234
00235
00236
00237
00238 void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
00239 int line_size)
00240 {
00241 const DCTELEM *p;
00242 uint8_t *pix;
00243
00244
00245 p = block;
00246 pix = pixels;
00247
00248 __asm__ volatile (
00249 "movq %3, %%mm0 \n\t"
00250 "movq 8%3, %%mm1 \n\t"
00251 "movq 16%3, %%mm2 \n\t"
00252 "movq 24%3, %%mm3 \n\t"
00253 "movq 32%3, %%mm4 \n\t"
00254 "movq 40%3, %%mm5 \n\t"
00255 "movq 48%3, %%mm6 \n\t"
00256 "movq 56%3, %%mm7 \n\t"
00257 "packuswb %%mm1, %%mm0 \n\t"
00258 "packuswb %%mm3, %%mm2 \n\t"
00259 "packuswb %%mm5, %%mm4 \n\t"
00260 "packuswb %%mm7, %%mm6 \n\t"
00261 "movq %%mm0, (%0) \n\t"
00262 "movq %%mm2, (%0, %1) \n\t"
00263 "movq %%mm4, (%0, %1, 2) \n\t"
00264 "movq %%mm6, (%0, %2) \n\t"
00265 :: "r"(pix), "r"((x86_reg)line_size), "r"((x86_reg)line_size * 3),
00266 "m"(*p)
00267 : "memory");
00268 pix += line_size * 4;
00269 p += 32;
00270
00271
00272
00273
00274 __asm__ volatile (
00275 "movq (%3), %%mm0 \n\t"
00276 "movq 8(%3), %%mm1 \n\t"
00277 "movq 16(%3), %%mm2 \n\t"
00278 "movq 24(%3), %%mm3 \n\t"
00279 "movq 32(%3), %%mm4 \n\t"
00280 "movq 40(%3), %%mm5 \n\t"
00281 "movq 48(%3), %%mm6 \n\t"
00282 "movq 56(%3), %%mm7 \n\t"
00283 "packuswb %%mm1, %%mm0 \n\t"
00284 "packuswb %%mm3, %%mm2 \n\t"
00285 "packuswb %%mm5, %%mm4 \n\t"
00286 "packuswb %%mm7, %%mm6 \n\t"
00287 "movq %%mm0, (%0) \n\t"
00288 "movq %%mm2, (%0, %1) \n\t"
00289 "movq %%mm4, (%0, %1, 2) \n\t"
00290 "movq %%mm6, (%0, %2) \n\t"
00291 :: "r"(pix), "r"((x86_reg)line_size), "r"((x86_reg)line_size * 3), "r"(p)
00292 : "memory");
00293 }
00294
00295 #define put_signed_pixels_clamped_mmx_half(off) \
00296 "movq "#off"(%2), %%mm1 \n\t" \
00297 "movq 16 + "#off"(%2), %%mm2 \n\t" \
00298 "movq 32 + "#off"(%2), %%mm3 \n\t" \
00299 "movq 48 + "#off"(%2), %%mm4 \n\t" \
00300 "packsswb 8 + "#off"(%2), %%mm1 \n\t" \
00301 "packsswb 24 + "#off"(%2), %%mm2 \n\t" \
00302 "packsswb 40 + "#off"(%2), %%mm3 \n\t" \
00303 "packsswb 56 + "#off"(%2), %%mm4 \n\t" \
00304 "paddb %%mm0, %%mm1 \n\t" \
00305 "paddb %%mm0, %%mm2 \n\t" \
00306 "paddb %%mm0, %%mm3 \n\t" \
00307 "paddb %%mm0, %%mm4 \n\t" \
00308 "movq %%mm1, (%0) \n\t" \
00309 "movq %%mm2, (%0, %3) \n\t" \
00310 "movq %%mm3, (%0, %3, 2) \n\t" \
00311 "movq %%mm4, (%0, %1) \n\t"
00312
00313 void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
00314 int line_size)
00315 {
00316 x86_reg line_skip = line_size;
00317 x86_reg line_skip3;
00318
00319 __asm__ volatile (
00320 "movq "MANGLE(ff_pb_80)", %%mm0 \n\t"
00321 "lea (%3, %3, 2), %1 \n\t"
00322 put_signed_pixels_clamped_mmx_half(0)
00323 "lea (%0, %3, 4), %0 \n\t"
00324 put_signed_pixels_clamped_mmx_half(64)
00325 : "+&r"(pixels), "=&r"(line_skip3)
00326 : "r"(block), "r"(line_skip)
00327 : "memory");
00328 }
00329
00330 void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels,
00331 int line_size)
00332 {
00333 const DCTELEM *p;
00334 uint8_t *pix;
00335 int i;
00336
00337
00338 p = block;
00339 pix = pixels;
00340 MOVQ_ZERO(mm7);
00341 i = 4;
00342 do {
00343 __asm__ volatile (
00344 "movq (%2), %%mm0 \n\t"
00345 "movq 8(%2), %%mm1 \n\t"
00346 "movq 16(%2), %%mm2 \n\t"
00347 "movq 24(%2), %%mm3 \n\t"
00348 "movq %0, %%mm4 \n\t"
00349 "movq %1, %%mm6 \n\t"
00350 "movq %%mm4, %%mm5 \n\t"
00351 "punpcklbw %%mm7, %%mm4 \n\t"
00352 "punpckhbw %%mm7, %%mm5 \n\t"
00353 "paddsw %%mm4, %%mm0 \n\t"
00354 "paddsw %%mm5, %%mm1 \n\t"
00355 "movq %%mm6, %%mm5 \n\t"
00356 "punpcklbw %%mm7, %%mm6 \n\t"
00357 "punpckhbw %%mm7, %%mm5 \n\t"
00358 "paddsw %%mm6, %%mm2 \n\t"
00359 "paddsw %%mm5, %%mm3 \n\t"
00360 "packuswb %%mm1, %%mm0 \n\t"
00361 "packuswb %%mm3, %%mm2 \n\t"
00362 "movq %%mm0, %0 \n\t"
00363 "movq %%mm2, %1 \n\t"
00364 : "+m"(*pix), "+m"(*(pix + line_size))
00365 : "r"(p)
00366 : "memory");
00367 pix += line_size * 2;
00368 p += 16;
00369 } while (--i);
00370 }
00371
00372 static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels,
00373 int line_size, int h)
00374 {
00375 __asm__ volatile (
00376 "lea (%3, %3), %%"REG_a" \n\t"
00377 ".p2align 3 \n\t"
00378 "1: \n\t"
00379 "movd (%1 ), %%mm0 \n\t"
00380 "movd (%1, %3), %%mm1 \n\t"
00381 "movd %%mm0, (%2) \n\t"
00382 "movd %%mm1, (%2, %3) \n\t"
00383 "add %%"REG_a", %1 \n\t"
00384 "add %%"REG_a", %2 \n\t"
00385 "movd (%1 ), %%mm0 \n\t"
00386 "movd (%1, %3), %%mm1 \n\t"
00387 "movd %%mm0, (%2) \n\t"
00388 "movd %%mm1, (%2, %3) \n\t"
00389 "add %%"REG_a", %1 \n\t"
00390 "add %%"REG_a", %2 \n\t"
00391 "subl $4, %0 \n\t"
00392 "jnz 1b \n\t"
00393 : "+g"(h), "+r"(pixels), "+r"(block)
00394 : "r"((x86_reg)line_size)
00395 : "%"REG_a, "memory"
00396 );
00397 }
00398
00399 static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels,
00400 int line_size, int h)
00401 {
00402 __asm__ volatile (
00403 "lea (%3, %3), %%"REG_a" \n\t"
00404 ".p2align 3 \n\t"
00405 "1: \n\t"
00406 "movq (%1 ), %%mm0 \n\t"
00407 "movq (%1, %3), %%mm1 \n\t"
00408 "movq %%mm0, (%2) \n\t"
00409 "movq %%mm1, (%2, %3) \n\t"
00410 "add %%"REG_a", %1 \n\t"
00411 "add %%"REG_a", %2 \n\t"
00412 "movq (%1 ), %%mm0 \n\t"
00413 "movq (%1, %3), %%mm1 \n\t"
00414 "movq %%mm0, (%2) \n\t"
00415 "movq %%mm1, (%2, %3) \n\t"
00416 "add %%"REG_a", %1 \n\t"
00417 "add %%"REG_a", %2 \n\t"
00418 "subl $4, %0 \n\t"
00419 "jnz 1b \n\t"
00420 : "+g"(h), "+r"(pixels), "+r"(block)
00421 : "r"((x86_reg)line_size)
00422 : "%"REG_a, "memory"
00423 );
00424 }
00425
00426 static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels,
00427 int line_size, int h)
00428 {
00429 __asm__ volatile (
00430 "lea (%3, %3), %%"REG_a" \n\t"
00431 ".p2align 3 \n\t"
00432 "1: \n\t"
00433 "movq (%1 ), %%mm0 \n\t"
00434 "movq 8(%1 ), %%mm4 \n\t"
00435 "movq (%1, %3), %%mm1 \n\t"
00436 "movq 8(%1, %3), %%mm5 \n\t"
00437 "movq %%mm0, (%2) \n\t"
00438 "movq %%mm4, 8(%2) \n\t"
00439 "movq %%mm1, (%2, %3) \n\t"
00440 "movq %%mm5, 8(%2, %3) \n\t"
00441 "add %%"REG_a", %1 \n\t"
00442 "add %%"REG_a", %2 \n\t"
00443 "movq (%1 ), %%mm0 \n\t"
00444 "movq 8(%1 ), %%mm4 \n\t"
00445 "movq (%1, %3), %%mm1 \n\t"
00446 "movq 8(%1, %3), %%mm5 \n\t"
00447 "movq %%mm0, (%2) \n\t"
00448 "movq %%mm4, 8(%2) \n\t"
00449 "movq %%mm1, (%2, %3) \n\t"
00450 "movq %%mm5, 8(%2, %3) \n\t"
00451 "add %%"REG_a", %1 \n\t"
00452 "add %%"REG_a", %2 \n\t"
00453 "subl $4, %0 \n\t"
00454 "jnz 1b \n\t"
00455 : "+g"(h), "+r"(pixels), "+r"(block)
00456 : "r"((x86_reg)line_size)
00457 : "%"REG_a, "memory"
00458 );
00459 }
00460
00461 static void put_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
00462 int line_size, int h)
00463 {
00464 __asm__ volatile (
00465 "1: \n\t"
00466 "movdqu (%1 ), %%xmm0 \n\t"
00467 "movdqu (%1, %3 ), %%xmm1 \n\t"
00468 "movdqu (%1, %3, 2), %%xmm2 \n\t"
00469 "movdqu (%1, %4 ), %%xmm3 \n\t"
00470 "lea (%1, %3, 4), %1 \n\t"
00471 "movdqa %%xmm0, (%2) \n\t"
00472 "movdqa %%xmm1, (%2, %3) \n\t"
00473 "movdqa %%xmm2, (%2, %3, 2) \n\t"
00474 "movdqa %%xmm3, (%2, %4) \n\t"
00475 "subl $4, %0 \n\t"
00476 "lea (%2, %3, 4), %2 \n\t"
00477 "jnz 1b \n\t"
00478 : "+g"(h), "+r"(pixels), "+r"(block)
00479 : "r"((x86_reg)line_size), "r"((x86_reg)3L * line_size)
00480 : "memory"
00481 );
00482 }
00483
00484 static void avg_pixels16_sse2(uint8_t *block, const uint8_t *pixels,
00485 int line_size, int h)
00486 {
00487 __asm__ volatile (
00488 "1: \n\t"
00489 "movdqu (%1 ), %%xmm0 \n\t"
00490 "movdqu (%1, %3 ), %%xmm1 \n\t"
00491 "movdqu (%1, %3, 2), %%xmm2 \n\t"
00492 "movdqu (%1, %4 ), %%xmm3 \n\t"
00493 "lea (%1, %3, 4), %1 \n\t"
00494 "pavgb (%2 ), %%xmm0 \n\t"
00495 "pavgb (%2, %3 ), %%xmm1 \n\t"
00496 "pavgb (%2, %3, 2), %%xmm2 \n\t"
00497 "pavgb (%2, %4), %%xmm3 \n\t"
00498 "movdqa %%xmm0, (%2) \n\t"
00499 "movdqa %%xmm1, (%2, %3) \n\t"
00500 "movdqa %%xmm2, (%2, %3, 2) \n\t"
00501 "movdqa %%xmm3, (%2, %4) \n\t"
00502 "subl $4, %0 \n\t"
00503 "lea (%2, %3, 4), %2 \n\t"
00504 "jnz 1b \n\t"
00505 : "+g"(h), "+r"(pixels), "+r"(block)
00506 : "r"((x86_reg)line_size), "r"((x86_reg)3L * line_size)
00507 : "memory"
00508 );
00509 }
00510
00511 #define CLEAR_BLOCKS(name, n) \
00512 static void name(DCTELEM *blocks) \
00513 { \
00514 __asm__ volatile ( \
00515 "pxor %%mm7, %%mm7 \n\t" \
00516 "mov %1, %%"REG_a" \n\t" \
00517 "1: \n\t" \
00518 "movq %%mm7, (%0, %%"REG_a") \n\t" \
00519 "movq %%mm7, 8(%0, %%"REG_a") \n\t" \
00520 "movq %%mm7, 16(%0, %%"REG_a") \n\t" \
00521 "movq %%mm7, 24(%0, %%"REG_a") \n\t" \
00522 "add $32, %%"REG_a" \n\t" \
00523 "js 1b \n\t" \
00524 :: "r"(((uint8_t *)blocks) + 128 * n), \
00525 "i"(-128 * n) \
00526 : "%"REG_a \
00527 ); \
00528 }
00529 CLEAR_BLOCKS(clear_blocks_mmx, 6)
00530 CLEAR_BLOCKS(clear_block_mmx, 1)
00531
00532 static void clear_block_sse(DCTELEM *block)
00533 {
00534 __asm__ volatile (
00535 "xorps %%xmm0, %%xmm0 \n"
00536 "movaps %%xmm0, (%0) \n"
00537 "movaps %%xmm0, 16(%0) \n"
00538 "movaps %%xmm0, 32(%0) \n"
00539 "movaps %%xmm0, 48(%0) \n"
00540 "movaps %%xmm0, 64(%0) \n"
00541 "movaps %%xmm0, 80(%0) \n"
00542 "movaps %%xmm0, 96(%0) \n"
00543 "movaps %%xmm0, 112(%0) \n"
00544 :: "r"(block)
00545 : "memory"
00546 );
00547 }
00548
00549 static void clear_blocks_sse(DCTELEM *blocks)
00550 {
00551 __asm__ volatile (
00552 "xorps %%xmm0, %%xmm0 \n"
00553 "mov %1, %%"REG_a" \n"
00554 "1: \n"
00555 "movaps %%xmm0, (%0, %%"REG_a") \n"
00556 "movaps %%xmm0, 16(%0, %%"REG_a") \n"
00557 "movaps %%xmm0, 32(%0, %%"REG_a") \n"
00558 "movaps %%xmm0, 48(%0, %%"REG_a") \n"
00559 "movaps %%xmm0, 64(%0, %%"REG_a") \n"
00560 "movaps %%xmm0, 80(%0, %%"REG_a") \n"
00561 "movaps %%xmm0, 96(%0, %%"REG_a") \n"
00562 "movaps %%xmm0, 112(%0, %%"REG_a") \n"
00563 "add $128, %%"REG_a" \n"
00564 "js 1b \n"
00565 :: "r"(((uint8_t *)blocks) + 128 * 6),
00566 "i"(-128 * 6)
00567 : "%"REG_a
00568 );
00569 }
00570
00571 static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w)
00572 {
00573 x86_reg i = 0;
00574 __asm__ volatile (
00575 "jmp 2f \n\t"
00576 "1: \n\t"
00577 "movq (%1, %0), %%mm0 \n\t"
00578 "movq (%2, %0), %%mm1 \n\t"
00579 "paddb %%mm0, %%mm1 \n\t"
00580 "movq %%mm1, (%2, %0) \n\t"
00581 "movq 8(%1, %0), %%mm0 \n\t"
00582 "movq 8(%2, %0), %%mm1 \n\t"
00583 "paddb %%mm0, %%mm1 \n\t"
00584 "movq %%mm1, 8(%2, %0) \n\t"
00585 "add $16, %0 \n\t"
00586 "2: \n\t"
00587 "cmp %3, %0 \n\t"
00588 "js 1b \n\t"
00589 : "+r"(i)
00590 : "r"(src), "r"(dst), "r"((x86_reg)w - 15)
00591 );
00592 for ( ; i < w; i++)
00593 dst[i + 0] += src[i + 0];
00594 }
00595
00596 #if HAVE_7REGS
00597 static void add_hfyu_median_prediction_cmov(uint8_t *dst, const uint8_t *top,
00598 const uint8_t *diff, int w,
00599 int *left, int *left_top)
00600 {
00601 x86_reg w2 = -w;
00602 x86_reg x;
00603 int l = *left & 0xff;
00604 int tl = *left_top & 0xff;
00605 int t;
00606 __asm__ volatile (
00607 "mov %7, %3 \n"
00608 "1: \n"
00609 "movzbl (%3, %4), %2 \n"
00610 "mov %2, %k3 \n"
00611 "sub %b1, %b3 \n"
00612 "add %b0, %b3 \n"
00613 "mov %2, %1 \n"
00614 "cmp %0, %2 \n"
00615 "cmovg %0, %2 \n"
00616 "cmovg %1, %0 \n"
00617 "cmp %k3, %0 \n"
00618 "cmovg %k3, %0 \n"
00619 "mov %7, %3 \n"
00620 "cmp %2, %0 \n"
00621 "cmovl %2, %0 \n"
00622 "add (%6, %4), %b0 \n"
00623 "mov %b0, (%5, %4) \n"
00624 "inc %4 \n"
00625 "jl 1b \n"
00626 : "+&q"(l), "+&q"(tl), "=&r"(t), "=&q"(x), "+&r"(w2)
00627 : "r"(dst + w), "r"(diff + w), "rm"(top + w)
00628 );
00629 *left = l;
00630 *left_top = tl;
00631 }
00632 #endif
00633
00634 #define H263_LOOP_FILTER \
00635 "pxor %%mm7, %%mm7 \n\t" \
00636 "movq %0, %%mm0 \n\t" \
00637 "movq %0, %%mm1 \n\t" \
00638 "movq %3, %%mm2 \n\t" \
00639 "movq %3, %%mm3 \n\t" \
00640 "punpcklbw %%mm7, %%mm0 \n\t" \
00641 "punpckhbw %%mm7, %%mm1 \n\t" \
00642 "punpcklbw %%mm7, %%mm2 \n\t" \
00643 "punpckhbw %%mm7, %%mm3 \n\t" \
00644 "psubw %%mm2, %%mm0 \n\t" \
00645 "psubw %%mm3, %%mm1 \n\t" \
00646 "movq %1, %%mm2 \n\t" \
00647 "movq %1, %%mm3 \n\t" \
00648 "movq %2, %%mm4 \n\t" \
00649 "movq %2, %%mm5 \n\t" \
00650 "punpcklbw %%mm7, %%mm2 \n\t" \
00651 "punpckhbw %%mm7, %%mm3 \n\t" \
00652 "punpcklbw %%mm7, %%mm4 \n\t" \
00653 "punpckhbw %%mm7, %%mm5 \n\t" \
00654 "psubw %%mm2, %%mm4 \n\t" \
00655 "psubw %%mm3, %%mm5 \n\t" \
00656 "psllw $2, %%mm4 \n\t" \
00657 "psllw $2, %%mm5 \n\t" \
00658 "paddw %%mm0, %%mm4 \n\t" \
00659 "paddw %%mm1, %%mm5 \n\t" \
00660 "pxor %%mm6, %%mm6 \n\t" \
00661 "pcmpgtw %%mm4, %%mm6 \n\t" \
00662 "pcmpgtw %%mm5, %%mm7 \n\t" \
00663 "pxor %%mm6, %%mm4 \n\t" \
00664 "pxor %%mm7, %%mm5 \n\t" \
00665 "psubw %%mm6, %%mm4 \n\t" \
00666 "psubw %%mm7, %%mm5 \n\t" \
00667 "psrlw $3, %%mm4 \n\t" \
00668 "psrlw $3, %%mm5 \n\t" \
00669 "packuswb %%mm5, %%mm4 \n\t" \
00670 "packsswb %%mm7, %%mm6 \n\t" \
00671 "pxor %%mm7, %%mm7 \n\t" \
00672 "movd %4, %%mm2 \n\t" \
00673 "punpcklbw %%mm2, %%mm2 \n\t" \
00674 "punpcklbw %%mm2, %%mm2 \n\t" \
00675 "punpcklbw %%mm2, %%mm2 \n\t" \
00676 "psubusb %%mm4, %%mm2 \n\t" \
00677 "movq %%mm2, %%mm3 \n\t" \
00678 "psubusb %%mm4, %%mm3 \n\t" \
00679 "psubb %%mm3, %%mm2 \n\t" \
00680 "movq %1, %%mm3 \n\t" \
00681 "movq %2, %%mm4 \n\t" \
00682 "pxor %%mm6, %%mm3 \n\t" \
00683 "pxor %%mm6, %%mm4 \n\t" \
00684 "paddusb %%mm2, %%mm3 \n\t" \
00685 "psubusb %%mm2, %%mm4 \n\t" \
00686 "pxor %%mm6, %%mm3 \n\t" \
00687 "pxor %%mm6, %%mm4 \n\t" \
00688 "paddusb %%mm2, %%mm2 \n\t" \
00689 "packsswb %%mm1, %%mm0 \n\t" \
00690 "pcmpgtb %%mm0, %%mm7 \n\t" \
00691 "pxor %%mm7, %%mm0 \n\t" \
00692 "psubb %%mm7, %%mm0 \n\t" \
00693 "movq %%mm0, %%mm1 \n\t" \
00694 "psubusb %%mm2, %%mm0 \n\t" \
00695 "psubb %%mm0, %%mm1 \n\t" \
00696 "pand %5, %%mm1 \n\t" \
00697 "psrlw $2, %%mm1 \n\t" \
00698 "pxor %%mm7, %%mm1 \n\t" \
00699 "psubb %%mm7, %%mm1 \n\t" \
00700 "movq %0, %%mm5 \n\t" \
00701 "movq %3, %%mm6 \n\t" \
00702 "psubb %%mm1, %%mm5 \n\t" \
00703 "paddb %%mm1, %%mm6 \n\t"
00704
00705 static void h263_v_loop_filter_mmx(uint8_t *src, int stride, int qscale)
00706 {
00707 if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
00708 const int strength = ff_h263_loop_filter_strength[qscale];
00709
00710 __asm__ volatile (
00711 H263_LOOP_FILTER
00712
00713 "movq %%mm3, %1 \n\t"
00714 "movq %%mm4, %2 \n\t"
00715 "movq %%mm5, %0 \n\t"
00716 "movq %%mm6, %3 \n\t"
00717 : "+m"(*(uint64_t*)(src - 2 * stride)),
00718 "+m"(*(uint64_t*)(src - 1 * stride)),
00719 "+m"(*(uint64_t*)(src + 0 * stride)),
00720 "+m"(*(uint64_t*)(src + 1 * stride))
00721 : "g"(2 * strength), "m"(ff_pb_FC)
00722 );
00723 }
00724 }
00725
00726 static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale)
00727 {
00728 if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
00729 const int strength = ff_h263_loop_filter_strength[qscale];
00730 DECLARE_ALIGNED(8, uint64_t, temp)[4];
00731 uint8_t *btemp = (uint8_t*)temp;
00732
00733 src -= 2;
00734
00735 transpose4x4(btemp, src, 8, stride);
00736 transpose4x4(btemp + 4, src + 4 * stride, 8, stride);
00737 __asm__ volatile (
00738 H263_LOOP_FILTER
00739
00740 : "+m"(temp[0]),
00741 "+m"(temp[1]),
00742 "+m"(temp[2]),
00743 "+m"(temp[3])
00744 : "g"(2 * strength), "m"(ff_pb_FC)
00745 );
00746
00747 __asm__ volatile (
00748 "movq %%mm5, %%mm1 \n\t"
00749 "movq %%mm4, %%mm0 \n\t"
00750 "punpcklbw %%mm3, %%mm5 \n\t"
00751 "punpcklbw %%mm6, %%mm4 \n\t"
00752 "punpckhbw %%mm3, %%mm1 \n\t"
00753 "punpckhbw %%mm6, %%mm0 \n\t"
00754 "movq %%mm5, %%mm3 \n\t"
00755 "movq %%mm1, %%mm6 \n\t"
00756 "punpcklwd %%mm4, %%mm5 \n\t"
00757 "punpcklwd %%mm0, %%mm1 \n\t"
00758 "punpckhwd %%mm4, %%mm3 \n\t"
00759 "punpckhwd %%mm0, %%mm6 \n\t"
00760 "movd %%mm5, (%0) \n\t"
00761 "punpckhdq %%mm5, %%mm5 \n\t"
00762 "movd %%mm5, (%0, %2) \n\t"
00763 "movd %%mm3, (%0, %2, 2) \n\t"
00764 "punpckhdq %%mm3, %%mm3 \n\t"
00765 "movd %%mm3, (%0, %3) \n\t"
00766 "movd %%mm1, (%1) \n\t"
00767 "punpckhdq %%mm1, %%mm1 \n\t"
00768 "movd %%mm1, (%1, %2) \n\t"
00769 "movd %%mm6, (%1, %2, 2) \n\t"
00770 "punpckhdq %%mm6, %%mm6 \n\t"
00771 "movd %%mm6, (%1, %3) \n\t"
00772 :: "r"(src),
00773 "r"(src + 4 * stride),
00774 "r"((x86_reg)stride),
00775 "r"((x86_reg)(3 * stride))
00776 );
00777 }
00778 }
00779
00780
00781
00782 static void draw_edges_mmx(uint8_t *buf, int wrap, int width, int height,
00783 int w, int h, int sides)
00784 {
00785 uint8_t *ptr, *last_line;
00786 int i;
00787
00788 last_line = buf + (height - 1) * wrap;
00789
00790 ptr = buf;
00791 if (w == 8) {
00792 __asm__ volatile (
00793 "1: \n\t"
00794 "movd (%0), %%mm0 \n\t"
00795 "punpcklbw %%mm0, %%mm0 \n\t"
00796 "punpcklwd %%mm0, %%mm0 \n\t"
00797 "punpckldq %%mm0, %%mm0 \n\t"
00798 "movq %%mm0, -8(%0) \n\t"
00799 "movq -8(%0, %2), %%mm1 \n\t"
00800 "punpckhbw %%mm1, %%mm1 \n\t"
00801 "punpckhwd %%mm1, %%mm1 \n\t"
00802 "punpckhdq %%mm1, %%mm1 \n\t"
00803 "movq %%mm1, (%0, %2) \n\t"
00804 "add %1, %0 \n\t"
00805 "cmp %3, %0 \n\t"
00806 "jb 1b \n\t"
00807 : "+r"(ptr)
00808 : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
00809 );
00810 } else {
00811 __asm__ volatile (
00812 "1: \n\t"
00813 "movd (%0), %%mm0 \n\t"
00814 "punpcklbw %%mm0, %%mm0 \n\t"
00815 "punpcklwd %%mm0, %%mm0 \n\t"
00816 "punpckldq %%mm0, %%mm0 \n\t"
00817 "movq %%mm0, -8(%0) \n\t"
00818 "movq %%mm0, -16(%0) \n\t"
00819 "movq -8(%0, %2), %%mm1 \n\t"
00820 "punpckhbw %%mm1, %%mm1 \n\t"
00821 "punpckhwd %%mm1, %%mm1 \n\t"
00822 "punpckhdq %%mm1, %%mm1 \n\t"
00823 "movq %%mm1, (%0, %2) \n\t"
00824 "movq %%mm1, 8(%0, %2) \n\t"
00825 "add %1, %0 \n\t"
00826 "cmp %3, %0 \n\t"
00827 "jb 1b \n\t"
00828 : "+r"(ptr)
00829 : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height)
00830 );
00831 }
00832
00833
00834 if (sides & EDGE_TOP) {
00835 for (i = 0; i < h; i += 4) {
00836 ptr = buf - (i + 1) * wrap - w;
00837 __asm__ volatile (
00838 "1: \n\t"
00839 "movq (%1, %0), %%mm0 \n\t"
00840 "movq %%mm0, (%0) \n\t"
00841 "movq %%mm0, (%0, %2) \n\t"
00842 "movq %%mm0, (%0, %2, 2) \n\t"
00843 "movq %%mm0, (%0, %3) \n\t"
00844 "add $8, %0 \n\t"
00845 "cmp %4, %0 \n\t"
00846 "jb 1b \n\t"
00847 : "+r"(ptr)
00848 : "r"((x86_reg)buf - (x86_reg)ptr - w), "r"((x86_reg) -wrap),
00849 "r"((x86_reg) -wrap * 3), "r"(ptr + width + 2 * w)
00850 );
00851 }
00852 }
00853
00854 if (sides & EDGE_BOTTOM) {
00855 for (i = 0; i < h; i += 4) {
00856 ptr = last_line + (i + 1) * wrap - w;
00857 __asm__ volatile (
00858 "1: \n\t"
00859 "movq (%1, %0), %%mm0 \n\t"
00860 "movq %%mm0, (%0) \n\t"
00861 "movq %%mm0, (%0, %2) \n\t"
00862 "movq %%mm0, (%0, %2, 2) \n\t"
00863 "movq %%mm0, (%0, %3) \n\t"
00864 "add $8, %0 \n\t"
00865 "cmp %4, %0 \n\t"
00866 "jb 1b \n\t"
00867 : "+r"(ptr)
00868 : "r"((x86_reg)last_line - (x86_reg)ptr - w),
00869 "r"((x86_reg)wrap), "r"((x86_reg)wrap * 3),
00870 "r"(ptr + width + 2 * w)
00871 );
00872 }
00873 }
00874 }
00875
00876 #define QPEL_V_LOW(m3, m4, m5, m6, pw_20, pw_3, rnd, \
00877 in0, in1, in2, in7, out, OP) \
00878 "paddw "#m4", "#m3" \n\t" \
00879 "movq "MANGLE(ff_pw_20)", %%mm4 \n\t" \
00880 "pmullw "#m3", %%mm4 \n\t" \
00881 "movq "#in7", "#m3" \n\t" \
00882 "movq "#in0", %%mm5 \n\t" \
00883 "paddw "#m3", %%mm5 \n\t" \
00884 "psubw %%mm5, %%mm4 \n\t" \
00885 "movq "#in1", %%mm5 \n\t" \
00886 "movq "#in2", %%mm6 \n\t" \
00887 "paddw "#m6", %%mm5 \n\t" \
00888 "paddw "#m5", %%mm6 \n\t" \
00889 "paddw %%mm6, %%mm6 \n\t" \
00890 "psubw %%mm6, %%mm5 \n\t" \
00891 "pmullw "MANGLE(ff_pw_3)", %%mm5 \n\t" \
00892 "paddw "#rnd", %%mm4 \n\t" \
00893 "paddw %%mm4, %%mm5 \n\t" \
00894 "psraw $5, %%mm5 \n\t" \
00895 "packuswb %%mm5, %%mm5 \n\t" \
00896 OP(%%mm5, out, %%mm7, d)
00897
00898 #define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW) \
00899 static void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, \
00900 uint8_t *src, \
00901 int dstStride, \
00902 int srcStride, \
00903 int h) \
00904 { \
00905 uint64_t temp; \
00906 \
00907 __asm__ volatile ( \
00908 "pxor %%mm7, %%mm7 \n\t" \
00909 "1: \n\t" \
00910 "movq (%0), %%mm0 \n\t" \
00911 "movq %%mm0, %%mm1 \n\t" \
00912 "movq %%mm0, %%mm2 \n\t" \
00913 "punpcklbw %%mm7, %%mm0 \n\t" \
00914 "punpckhbw %%mm7, %%mm1 \n\t" \
00915 "pshufw $0x90, %%mm0, %%mm5 \n\t" \
00916 "pshufw $0x41, %%mm0, %%mm6 \n\t" \
00917 "movq %%mm2, %%mm3 \n\t" \
00918 "movq %%mm2, %%mm4 \n\t" \
00919 "psllq $8, %%mm2 \n\t" \
00920 "psllq $16, %%mm3 \n\t" \
00921 "psllq $24, %%mm4 \n\t" \
00922 "punpckhbw %%mm7, %%mm2 \n\t" \
00923 "punpckhbw %%mm7, %%mm3 \n\t" \
00924 "punpckhbw %%mm7, %%mm4 \n\t" \
00925 "paddw %%mm3, %%mm5 \n\t" \
00926 "paddw %%mm2, %%mm6 \n\t" \
00927 "paddw %%mm5, %%mm5 \n\t" \
00928 "psubw %%mm5, %%mm6 \n\t" \
00929 "pshufw $0x06, %%mm0, %%mm5 \n\t" \
00930 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" \
00931 "paddw %%mm4, %%mm0 \n\t" \
00932 "paddw %%mm1, %%mm5 \n\t" \
00933 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" \
00934 "psubw %%mm5, %%mm0 \n\t" \
00935 "paddw %6, %%mm6 \n\t" \
00936 "paddw %%mm6, %%mm0 \n\t" \
00937 "psraw $5, %%mm0 \n\t" \
00938 "movq %%mm0, %5 \n\t" \
00939 \
00940 \
00941 "movq 5(%0), %%mm0 \n\t" \
00942 "movq %%mm0, %%mm5 \n\t" \
00943 "movq %%mm0, %%mm6 \n\t" \
00944 "psrlq $8, %%mm0 \n\t" \
00945 "psrlq $16, %%mm5 \n\t" \
00946 "punpcklbw %%mm7, %%mm0 \n\t" \
00947 "punpcklbw %%mm7, %%mm5 \n\t" \
00948 "paddw %%mm0, %%mm2 \n\t" \
00949 "paddw %%mm5, %%mm3 \n\t" \
00950 "paddw %%mm2, %%mm2 \n\t" \
00951 "psubw %%mm2, %%mm3 \n\t" \
00952 "movq %%mm6, %%mm2 \n\t" \
00953 "psrlq $24, %%mm6 \n\t" \
00954 "punpcklbw %%mm7, %%mm2 \n\t" \
00955 "punpcklbw %%mm7, %%mm6 \n\t" \
00956 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" \
00957 "paddw %%mm2, %%mm1 \n\t" \
00958 "paddw %%mm6, %%mm4 \n\t" \
00959 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" \
00960 "psubw %%mm4, %%mm3 \n\t" \
00961 "paddw %6, %%mm1 \n\t" \
00962 "paddw %%mm1, %%mm3 \n\t" \
00963 "psraw $5, %%mm3 \n\t" \
00964 "movq %5, %%mm1 \n\t" \
00965 "packuswb %%mm3, %%mm1 \n\t" \
00966 OP_MMX2(%%mm1, (%1), %%mm4, q) \
00967 \
00968 \
00969 "movq 9(%0), %%mm1 \n\t" \
00970 "movq %%mm1, %%mm4 \n\t" \
00971 "movq %%mm1, %%mm3 \n\t" \
00972 "psrlq $8, %%mm1 \n\t" \
00973 "psrlq $16, %%mm4 \n\t" \
00974 "punpcklbw %%mm7, %%mm1 \n\t" \
00975 "punpcklbw %%mm7, %%mm4 \n\t" \
00976 "paddw %%mm1, %%mm5 \n\t" \
00977 "paddw %%mm4, %%mm0 \n\t" \
00978 "paddw %%mm5, %%mm5 \n\t" \
00979 "psubw %%mm5, %%mm0 \n\t" \
00980 "movq %%mm3, %%mm5 \n\t" \
00981 "psrlq $24, %%mm3 \n\t" \
00982 "pmullw "MANGLE(ff_pw_3)", %%mm0 \n\t" \
00983 "punpcklbw %%mm7, %%mm3 \n\t" \
00984 "paddw %%mm3, %%mm2 \n\t" \
00985 "psubw %%mm2, %%mm0 \n\t" \
00986 "movq %%mm5, %%mm2 \n\t" \
00987 "punpcklbw %%mm7, %%mm2 \n\t" \
00988 "punpckhbw %%mm7, %%mm5 \n\t" \
00989 "paddw %%mm2, %%mm6 \n\t" \
00990 "pmullw "MANGLE(ff_pw_20)", %%mm6 \n\t" \
00991 "paddw %6, %%mm0 \n\t" \
00992 "paddw %%mm6, %%mm0 \n\t" \
00993 "psraw $5, %%mm0 \n\t" \
00994 \
00995 \
00996 \
00997 "paddw %%mm5, %%mm3 \n\t" \
00998 "pshufw $0xF9, %%mm5, %%mm6 \n\t" \
00999 "paddw %%mm4, %%mm6 \n\t" \
01000 "pshufw $0xBE, %%mm5, %%mm4 \n\t" \
01001 "pshufw $0x6F, %%mm5, %%mm5 \n\t" \
01002 "paddw %%mm1, %%mm4 \n\t" \
01003 "paddw %%mm2, %%mm5 \n\t" \
01004 "paddw %%mm6, %%mm6 \n\t" \
01005 "psubw %%mm6, %%mm4 \n\t" \
01006 "pmullw "MANGLE(ff_pw_20)", %%mm3 \n\t" \
01007 "pmullw "MANGLE(ff_pw_3)", %%mm4 \n\t" \
01008 "psubw %%mm5, %%mm3 \n\t" \
01009 "paddw %6, %%mm4 \n\t" \
01010 "paddw %%mm3, %%mm4 \n\t" \
01011 "psraw $5, %%mm4 \n\t" \
01012 "packuswb %%mm4, %%mm0 \n\t" \
01013 OP_MMX2(%%mm0, 8(%1), %%mm4, q) \
01014 \
01015 "add %3, %0 \n\t" \
01016 "add %4, %1 \n\t" \
01017 "decl %2 \n\t" \
01018 "jnz 1b \n\t" \
01019 : "+a"(src), "+c"(dst), "+D"(h) \
01020 : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), \
01021 "m"(temp), "m"(ROUNDER) \
01022 : "memory" \
01023 ); \
01024 } \
01025 \
01026 static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, \
01027 uint8_t *src, \
01028 int dstStride, \
01029 int srcStride, \
01030 int h) \
01031 { \
01032 int i; \
01033 int16_t temp[16]; \
01034 \
01035 for (i = 0; i < h; i++) { \
01036 temp[ 0] = (src[ 0] + src[ 1]) * 20 - (src[ 0] + src[ 2]) * 6 + \
01037 (src[ 1] + src[ 3]) * 3 - (src[ 2] + src[ 4]); \
01038 temp[ 1] = (src[ 1] + src[ 2]) * 20 - (src[ 0] + src[ 3]) * 6 + \
01039 (src[ 0] + src[ 4]) * 3 - (src[ 1] + src[ 5]); \
01040 temp[ 2] = (src[ 2] + src[ 3]) * 20 - (src[ 1] + src[ 4]) * 6 + \
01041 (src[ 0] + src[ 5]) * 3 - (src[ 0] + src[ 6]); \
01042 temp[ 3] = (src[ 3] + src[ 4]) * 20 - (src[ 2] + src[ 5]) * 6 + \
01043 (src[ 1] + src[ 6]) * 3 - (src[ 0] + src[ 7]); \
01044 temp[ 4] = (src[ 4] + src[ 5]) * 20 - (src[ 3] + src[ 6]) * 6 + \
01045 (src[ 2] + src[ 7]) * 3 - (src[ 1] + src[ 8]); \
01046 temp[ 5] = (src[ 5] + src[ 6]) * 20 - (src[ 4] + src[ 7]) * 6 + \
01047 (src[ 3] + src[ 8]) * 3 - (src[ 2] + src[ 9]); \
01048 temp[ 6] = (src[ 6] + src[ 7]) * 20 - (src[ 5] + src[ 8]) * 6 + \
01049 (src[ 4] + src[ 9]) * 3 - (src[ 3] + src[10]); \
01050 temp[ 7] = (src[ 7] + src[ 8]) * 20 - (src[ 6] + src[ 9]) * 6 + \
01051 (src[ 5] + src[10]) * 3 - (src[ 4] + src[11]); \
01052 temp[ 8] = (src[ 8] + src[ 9]) * 20 - (src[ 7] + src[10]) * 6 + \
01053 (src[ 6] + src[11]) * 3 - (src[ 5] + src[12]); \
01054 temp[ 9] = (src[ 9] + src[10]) * 20 - (src[ 8] + src[11]) * 6 + \
01055 (src[ 7] + src[12]) * 3 - (src[ 6] + src[13]); \
01056 temp[10] = (src[10] + src[11]) * 20 - (src[ 9] + src[12]) * 6 + \
01057 (src[ 8] + src[13]) * 3 - (src[ 7] + src[14]); \
01058 temp[11] = (src[11] + src[12]) * 20 - (src[10] + src[13]) * 6 + \
01059 (src[ 9] + src[14]) * 3 - (src[ 8] + src[15]); \
01060 temp[12] = (src[12] + src[13]) * 20 - (src[11] + src[14]) * 6 + \
01061 (src[10] + src[15]) * 3 - (src[ 9] + src[16]); \
01062 temp[13] = (src[13] + src[14]) * 20 - (src[12] + src[15]) * 6 + \
01063 (src[11] + src[16]) * 3 - (src[10] + src[16]); \
01064 temp[14] = (src[14] + src[15]) * 20 - (src[13] + src[16]) * 6 + \
01065 (src[12] + src[16]) * 3 - (src[11] + src[15]); \
01066 temp[15] = (src[15] + src[16]) * 20 - (src[14] + src[16]) * 6 + \
01067 (src[13] + src[15]) * 3 - (src[12] + src[14]); \
01068 __asm__ volatile ( \
01069 "movq (%0), %%mm0 \n\t" \
01070 "movq 8(%0), %%mm1 \n\t" \
01071 "paddw %2, %%mm0 \n\t" \
01072 "paddw %2, %%mm1 \n\t" \
01073 "psraw $5, %%mm0 \n\t" \
01074 "psraw $5, %%mm1 \n\t" \
01075 "packuswb %%mm1, %%mm0 \n\t" \
01076 OP_3DNOW(%%mm0, (%1), %%mm1, q) \
01077 "movq 16(%0), %%mm0 \n\t" \
01078 "movq 24(%0), %%mm1 \n\t" \
01079 "paddw %2, %%mm0 \n\t" \
01080 "paddw %2, %%mm1 \n\t" \
01081 "psraw $5, %%mm0 \n\t" \
01082 "psraw $5, %%mm1 \n\t" \
01083 "packuswb %%mm1, %%mm0 \n\t" \
01084 OP_3DNOW(%%mm0, 8(%1), %%mm1, q) \
01085 :: "r"(temp), "r"(dst), "m"(ROUNDER) \
01086 : "memory" \
01087 ); \
01088 dst += dstStride; \
01089 src += srcStride; \
01090 } \
01091 } \
01092 \
01093 static void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, \
01094 uint8_t *src, \
01095 int dstStride, \
01096 int srcStride, \
01097 int h) \
01098 { \
01099 __asm__ volatile ( \
01100 "pxor %%mm7, %%mm7 \n\t" \
01101 "1: \n\t" \
01102 "movq (%0), %%mm0 \n\t" \
01103 "movq %%mm0, %%mm1 \n\t" \
01104 "movq %%mm0, %%mm2 \n\t" \
01105 "punpcklbw %%mm7, %%mm0 \n\t" \
01106 "punpckhbw %%mm7, %%mm1 \n\t" \
01107 "pshufw $0x90, %%mm0, %%mm5 \n\t" \
01108 "pshufw $0x41, %%mm0, %%mm6 \n\t" \
01109 "movq %%mm2, %%mm3 \n\t" \
01110 "movq %%mm2, %%mm4 \n\t" \
01111 "psllq $8, %%mm2 \n\t" \
01112 "psllq $16, %%mm3 \n\t" \
01113 "psllq $24, %%mm4 \n\t" \
01114 "punpckhbw %%mm7, %%mm2 \n\t" \
01115 "punpckhbw %%mm7, %%mm3 \n\t" \
01116 "punpckhbw %%mm7, %%mm4 \n\t" \
01117 "paddw %%mm3, %%mm5 \n\t" \
01118 "paddw %%mm2, %%mm6 \n\t" \
01119 "paddw %%mm5, %%mm5 \n\t" \
01120 "psubw %%mm5, %%mm6 \n\t" \
01121 "pshufw $0x06, %%mm0, %%mm5 \n\t" \
01122 "pmullw "MANGLE(ff_pw_3)", %%mm6 \n\t" \
01123 "paddw %%mm4, %%mm0 \n\t" \
01124 "paddw %%mm1, %%mm5 \n\t" \
01125 "pmullw "MANGLE(ff_pw_20)", %%mm0 \n\t" \
01126 "psubw %%mm5, %%mm0 \n\t" \
01127 "paddw %5, %%mm6 \n\t" \
01128 "paddw %%mm6, %%mm0 \n\t" \
01129 "psraw $5, %%mm0 \n\t" \
01130 \
01131 \
01132 "movd 5(%0), %%mm5 \n\t" \
01133 "punpcklbw %%mm7, %%mm5 \n\t" \
01134 "pshufw $0xF9, %%mm5, %%mm6 \n\t" \
01135 "paddw %%mm5, %%mm1 \n\t" \
01136 "paddw %%mm6, %%mm2 \n\t" \
01137 "pshufw $0xBE, %%mm5, %%mm6 \n\t" \
01138 "pshufw $0x6F, %%mm5, %%mm5 \n\t" \
01139 "paddw %%mm6, %%mm3 \n\t" \
01140 "paddw %%mm5, %%mm4 \n\t" \
01141 "paddw %%mm2, %%mm2 \n\t" \
01142 "psubw %%mm2, %%mm3 \n\t" \
01143 "pmullw "MANGLE(ff_pw_20)", %%mm1 \n\t" \
01144 "pmullw "MANGLE(ff_pw_3)", %%mm3 \n\t" \
01145 "psubw %%mm4, %%mm3 \n\t" \
01146 "paddw %5, %%mm1 \n\t" \
01147 "paddw %%mm1, %%mm3 \n\t" \
01148 "psraw $5, %%mm3 \n\t" \
01149 "packuswb %%mm3, %%mm0 \n\t" \
01150 OP_MMX2(%%mm0, (%1), %%mm4, q) \
01151 \
01152 "add %3, %0 \n\t" \
01153 "add %4, %1 \n\t" \
01154 "decl %2 \n\t" \
01155 "jnz 1b \n\t" \
01156 : "+a"(src), "+c"(dst), "+d"(h) \
01157 : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), \
01158 "m"(ROUNDER) \
01159 : "memory" \
01160 ); \
01161 } \
01162 \
01163 static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, \
01164 uint8_t *src, \
01165 int dstStride, \
01166 int srcStride, \
01167 int h) \
01168 { \
01169 int i; \
01170 int16_t temp[8]; \
01171 \
01172 for (i = 0; i < h; i++) { \
01173 temp[0] = (src[0] + src[1]) * 20 - (src[0] + src[2]) * 6 + \
01174 (src[1] + src[3]) * 3 - (src[2] + src[4]); \
01175 temp[1] = (src[1] + src[2]) * 20 - (src[0] + src[3]) * 6 + \
01176 (src[0] + src[4]) * 3 - (src[1] + src[5]); \
01177 temp[2] = (src[2] + src[3]) * 20 - (src[1] + src[4]) * 6 + \
01178 (src[0] + src[5]) * 3 - (src[0] + src[6]); \
01179 temp[3] = (src[3] + src[4]) * 20 - (src[2] + src[5]) * 6 + \
01180 (src[1] + src[6]) * 3 - (src[0] + src[7]); \
01181 temp[4] = (src[4] + src[5]) * 20 - (src[3] + src[6]) * 6 + \
01182 (src[2] + src[7]) * 3 - (src[1] + src[8]); \
01183 temp[5] = (src[5] + src[6]) * 20 - (src[4] + src[7]) * 6 + \
01184 (src[3] + src[8]) * 3 - (src[2] + src[8]); \
01185 temp[6] = (src[6] + src[7]) * 20 - (src[5] + src[8]) * 6 + \
01186 (src[4] + src[8]) * 3 - (src[3] + src[7]); \
01187 temp[7] = (src[7] + src[8]) * 20 - (src[6] + src[8]) * 6 + \
01188 (src[5] + src[7]) * 3 - (src[4] + src[6]); \
01189 __asm__ volatile ( \
01190 "movq (%0), %%mm0 \n\t" \
01191 "movq 8(%0), %%mm1 \n\t" \
01192 "paddw %2, %%mm0 \n\t" \
01193 "paddw %2, %%mm1 \n\t" \
01194 "psraw $5, %%mm0 \n\t" \
01195 "psraw $5, %%mm1 \n\t" \
01196 "packuswb %%mm1, %%mm0 \n\t" \
01197 OP_3DNOW(%%mm0, (%1), %%mm1, q) \
01198 :: "r"(temp), "r"(dst), "m"(ROUNDER) \
01199 : "memory" \
01200 ); \
01201 dst += dstStride; \
01202 src += srcStride; \
01203 } \
01204 }
01205
01206 #define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX) \
01207 static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, \
01208 uint8_t *src, \
01209 int dstStride, \
01210 int srcStride) \
01211 { \
01212 uint64_t temp[17 * 4]; \
01213 uint64_t *temp_ptr = temp; \
01214 int count = 17; \
01215 \
01216 \
01217 __asm__ volatile ( \
01218 "pxor %%mm7, %%mm7 \n\t" \
01219 "1: \n\t" \
01220 "movq (%0), %%mm0 \n\t" \
01221 "movq (%0), %%mm1 \n\t" \
01222 "movq 8(%0), %%mm2 \n\t" \
01223 "movq 8(%0), %%mm3 \n\t" \
01224 "punpcklbw %%mm7, %%mm0 \n\t" \
01225 "punpckhbw %%mm7, %%mm1 \n\t" \
01226 "punpcklbw %%mm7, %%mm2 \n\t" \
01227 "punpckhbw %%mm7, %%mm3 \n\t" \
01228 "movq %%mm0, (%1) \n\t" \
01229 "movq %%mm1, 17 * 8(%1) \n\t" \
01230 "movq %%mm2, 2 * 17 * 8(%1) \n\t" \
01231 "movq %%mm3, 3 * 17 * 8(%1) \n\t" \
01232 "add $8, %1 \n\t" \
01233 "add %3, %0 \n\t" \
01234 "decl %2 \n\t" \
01235 "jnz 1b \n\t" \
01236 : "+r"(src), "+r"(temp_ptr), "+r"(count) \
01237 : "r"((x86_reg)srcStride) \
01238 : "memory" \
01239 ); \
01240 \
01241 temp_ptr = temp; \
01242 count = 4; \
01243 \
01244 \
01245 __asm__ volatile ( \
01246 \
01247 "1: \n\t" \
01248 "movq (%0), %%mm0 \n\t" \
01249 "movq 8(%0), %%mm1 \n\t" \
01250 "movq 16(%0), %%mm2 \n\t" \
01251 "movq 24(%0), %%mm3 \n\t" \
01252 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP) \
01253 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP) \
01254 "add %4, %1 \n\t" \
01255 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP) \
01256 \
01257 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP) \
01258 "add %4, %1 \n\t" \
01259 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP) \
01260 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP) \
01261 "add %4, %1 \n\t" \
01262 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP) \
01263 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP) \
01264 "add %4, %1 \n\t" \
01265 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP) \
01266 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0), 104(%0), (%1, %3), OP) \
01267 "add %4, %1 \n\t" \
01268 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0), 112(%0), (%1), OP) \
01269 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0), 120(%0), (%1, %3), OP) \
01270 "add %4, %1 \n\t" \
01271 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0), 128(%0), (%1), OP) \
01272 \
01273 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0), 128(%0), (%1, %3), OP) \
01274 "add %4, %1 \n\t" \
01275 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0), 104(%0), 120(%0), (%1), OP) \
01276 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0), 104(%0), 112(%0), 112(%0), (%1, %3), OP) \
01277 \
01278 "add $136, %0 \n\t" \
01279 "add %6, %1 \n\t" \
01280 "decl %2 \n\t" \
01281 "jnz 1b \n\t" \
01282 \
01283 : "+r"(temp_ptr), "+r"(dst), "+g"(count) \
01284 : "r"((x86_reg)dstStride), "r"(2 * (x86_reg)dstStride), \
01285 "m"(ROUNDER), \
01286 "g"(4 - 14 * (x86_reg)dstStride) \
01287 : "memory" \
01288 ); \
01289 } \
01290 \
01291 static void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, \
01292 uint8_t *src, \
01293 int dstStride, \
01294 int srcStride) \
01295 { \
01296 uint64_t temp[9 * 2]; \
01297 uint64_t *temp_ptr = temp; \
01298 int count = 9; \
01299 \
01300 \
01301 __asm__ volatile ( \
01302 "pxor %%mm7, %%mm7 \n\t" \
01303 "1: \n\t" \
01304 "movq (%0), %%mm0 \n\t" \
01305 "movq (%0), %%mm1 \n\t" \
01306 "punpcklbw %%mm7, %%mm0 \n\t" \
01307 "punpckhbw %%mm7, %%mm1 \n\t" \
01308 "movq %%mm0, (%1) \n\t" \
01309 "movq %%mm1, 9*8(%1) \n\t" \
01310 "add $8, %1 \n\t" \
01311 "add %3, %0 \n\t" \
01312 "decl %2 \n\t" \
01313 "jnz 1b \n\t" \
01314 : "+r"(src), "+r"(temp_ptr), "+r"(count) \
01315 : "r"((x86_reg)srcStride) \
01316 : "memory" \
01317 ); \
01318 \
01319 temp_ptr = temp; \
01320 count = 2; \
01321 \
01322 \
01323 __asm__ volatile ( \
01324 \
01325 "1: \n\t" \
01326 "movq (%0), %%mm0 \n\t" \
01327 "movq 8(%0), %%mm1 \n\t" \
01328 "movq 16(%0), %%mm2 \n\t" \
01329 "movq 24(%0), %%mm3 \n\t" \
01330 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0), 8(%0), (%0), 32(%0), (%1), OP) \
01331 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 8(%0), (%0), (%0), 40(%0), (%1, %3), OP) \
01332 "add %4, %1 \n\t" \
01333 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, (%0), (%0), 8(%0), 48(%0), (%1), OP) \
01334 \
01335 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, (%0), 8(%0), 16(%0), 56(%0), (%1, %3), OP) \
01336 "add %4, %1 \n\t" \
01337 QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 8(%0), 16(%0), 24(%0), 64(%0), (%1), OP) \
01338 \
01339 QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP) \
01340 "add %4, %1 \n\t" \
01341 QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP) \
01342 QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP) \
01343 \
01344 "add $72, %0 \n\t" \
01345 "add %6, %1 \n\t" \
01346 "decl %2 \n\t" \
01347 "jnz 1b \n\t" \
01348 \
01349 : "+r"(temp_ptr), "+r"(dst), "+g"(count) \
01350 : "r"((x86_reg)dstStride), "r"(2 * (x86_reg)dstStride), \
01351 "m"(ROUNDER), \
01352 "g"(4 - 6 * (x86_reg)dstStride) \
01353 : "memory" \
01354 ); \
01355 } \
01356 \
01357 static void OPNAME ## qpel8_mc00_ ## MMX (uint8_t *dst, uint8_t *src, \
01358 int stride) \
01359 { \
01360 OPNAME ## pixels8_ ## MMX(dst, src, stride, 8); \
01361 } \
01362 \
01363 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src, \
01364 int stride) \
01365 { \
01366 uint64_t temp[8]; \
01367 uint8_t * const half = (uint8_t*)temp; \
01368 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, \
01369 stride, 8); \
01370 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8); \
01371 } \
01372 \
01373 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src, \
01374 int stride) \
01375 { \
01376 OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, \
01377 stride, 8); \
01378 } \
01379 \
01380 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src, \
01381 int stride) \
01382 { \
01383 uint64_t temp[8]; \
01384 uint8_t * const half = (uint8_t*)temp; \
01385 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, \
01386 stride, 8); \
01387 OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride, \
01388 stride, 8); \
01389 } \
01390 \
01391 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src, \
01392 int stride) \
01393 { \
01394 uint64_t temp[8]; \
01395 uint8_t * const half = (uint8_t*)temp; \
01396 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride); \
01397 OPNAME ## pixels8_l2_ ## MMX(dst, src, half, stride, stride, 8); \
01398 } \
01399 \
01400 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src, \
01401 int stride) \
01402 { \
01403 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride); \
01404 } \
01405 \
01406 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src, \
01407 int stride) \
01408 { \
01409 uint64_t temp[8]; \
01410 uint8_t * const half = (uint8_t*)temp; \
01411 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride); \
01412 OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride, \
01413 stride, 8); \
01414 } \
01415 \
01416 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src, \
01417 int stride) \
01418 { \
01419 uint64_t half[8 + 9]; \
01420 uint8_t * const halfH = ((uint8_t*)half) + 64; \
01421 uint8_t * const halfHV = ((uint8_t*)half); \
01422 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01423 stride, 9); \
01424 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9); \
01425 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8); \
01426 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8); \
01427 } \
01428 \
01429 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src, \
01430 int stride) \
01431 { \
01432 uint64_t half[8 + 9]; \
01433 uint8_t * const halfH = ((uint8_t*)half) + 64; \
01434 uint8_t * const halfHV = ((uint8_t*)half); \
01435 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01436 stride, 9); \
01437 put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \
01438 stride, 9); \
01439 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8); \
01440 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8); \
01441 } \
01442 \
01443 static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src, \
01444 int stride) \
01445 { \
01446 uint64_t half[8 + 9]; \
01447 uint8_t * const halfH = ((uint8_t*)half) + 64; \
01448 uint8_t * const halfHV = ((uint8_t*)half); \
01449 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01450 stride, 9); \
01451 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9); \
01452 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8); \
01453 OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, stride, 8, 8); \
01454 } \
01455 \
01456 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src, \
01457 int stride) \
01458 { \
01459 uint64_t half[8 + 9]; \
01460 uint8_t * const halfH = ((uint8_t*)half) + 64; \
01461 uint8_t * const halfHV = ((uint8_t*)half); \
01462 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01463 stride, 9); \
01464 put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \
01465 stride, 9); \
01466 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8); \
01467 OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, stride, 8, 8); \
01468 } \
01469 \
01470 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src, \
01471 int stride) \
01472 { \
01473 uint64_t half[8 + 9]; \
01474 uint8_t * const halfH = ((uint8_t*)half) + 64; \
01475 uint8_t * const halfHV = ((uint8_t*)half); \
01476 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01477 stride, 9); \
01478 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8); \
01479 OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV, stride, 8, 8); \
01480 } \
01481 \
01482 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src, \
01483 int stride) \
01484 { \
01485 uint64_t half[8 + 9]; \
01486 uint8_t * const halfH = ((uint8_t*)half) + 64; \
01487 uint8_t * const halfHV = ((uint8_t*)half); \
01488 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01489 stride, 9); \
01490 put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8); \
01491 OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV, stride, 8, 8); \
01492 } \
01493 \
01494 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src, \
01495 int stride) \
01496 { \
01497 uint64_t half[8 + 9]; \
01498 uint8_t * const halfH = ((uint8_t*)half); \
01499 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01500 stride, 9); \
01501 put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8, stride, 9); \
01502 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8); \
01503 } \
01504 \
01505 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src, \
01506 int stride) \
01507 { \
01508 uint64_t half[8 + 9]; \
01509 uint8_t * const halfH = ((uint8_t*)half); \
01510 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01511 stride, 9); \
01512 put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8, \
01513 stride, 9); \
01514 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8); \
01515 } \
01516 \
01517 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src, \
01518 int stride) \
01519 { \
01520 uint64_t half[9]; \
01521 uint8_t * const halfH = ((uint8_t*)half); \
01522 put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, \
01523 stride, 9); \
01524 OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8); \
01525 } \
01526 \
01527 static void OPNAME ## qpel16_mc00_ ## MMX (uint8_t *dst, uint8_t *src, \
01528 int stride) \
01529 { \
01530 OPNAME ## pixels16_ ## MMX(dst, src, stride, 16); \
01531 } \
01532 \
01533 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src, \
01534 int stride) \
01535 { \
01536 uint64_t temp[32]; \
01537 uint8_t * const half = (uint8_t*)temp; \
01538 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, \
01539 stride, 16); \
01540 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16); \
01541 } \
01542 \
01543 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src, \
01544 int stride) \
01545 { \
01546 OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, \
01547 stride, stride, 16); \
01548 } \
01549 \
01550 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src, \
01551 int stride) \
01552 { \
01553 uint64_t temp[32]; \
01554 uint8_t * const half = (uint8_t*)temp; \
01555 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, \
01556 stride, 16); \
01557 OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half, \
01558 stride, stride, 16); \
01559 } \
01560 \
01561 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src, \
01562 int stride) \
01563 { \
01564 uint64_t temp[32]; \
01565 uint8_t * const half = (uint8_t*)temp; \
01566 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, \
01567 stride); \
01568 OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride, stride, 16); \
01569 } \
01570 \
01571 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src, \
01572 int stride) \
01573 { \
01574 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride); \
01575 } \
01576 \
01577 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src, \
01578 int stride) \
01579 { \
01580 uint64_t temp[32]; \
01581 uint8_t * const half = (uint8_t*)temp; \
01582 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, \
01583 stride); \
01584 OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half, \
01585 stride, stride, 16); \
01586 } \
01587 \
01588 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src, \
01589 int stride) \
01590 { \
01591 uint64_t half[16 * 2 + 17 * 2]; \
01592 uint8_t * const halfH = ((uint8_t*)half) + 256; \
01593 uint8_t * const halfHV = ((uint8_t*)half); \
01594 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01595 stride, 17); \
01596 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \
01597 stride, 17); \
01598 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
01599 16, 16); \
01600 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16); \
01601 } \
01602 \
01603 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src, \
01604 int stride) \
01605 { \
01606 uint64_t half[16 * 2 + 17 * 2]; \
01607 uint8_t * const halfH = ((uint8_t*)half) + 256; \
01608 uint8_t * const halfHV = ((uint8_t*)half); \
01609 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01610 stride, 17); \
01611 put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \
01612 stride, 17); \
01613 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
01614 16, 16); \
01615 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16); \
01616 } \
01617 \
01618 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src, \
01619 int stride) \
01620 { \
01621 uint64_t half[16 * 2 + 17 * 2]; \
01622 uint8_t * const halfH = ((uint8_t*)half) + 256; \
01623 uint8_t * const halfHV = ((uint8_t*)half); \
01624 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01625 stride, 17); \
01626 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \
01627 stride, 17); \
01628 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
01629 16, 16); \
01630 OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, stride, \
01631 16, 16); \
01632 } \
01633 \
01634 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src, \
01635 int stride) \
01636 { \
01637 uint64_t half[16 * 2 + 17 * 2]; \
01638 uint8_t * const halfH = ((uint8_t*)half) + 256; \
01639 uint8_t * const halfHV = ((uint8_t*)half); \
01640 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01641 stride, 17); \
01642 put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \
01643 stride, 17); \
01644 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
01645 16, 16); \
01646 OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, stride, \
01647 16, 16); \
01648 } \
01649 \
01650 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src, \
01651 int stride) \
01652 { \
01653 uint64_t half[16 * 2 + 17 * 2]; \
01654 uint8_t * const halfH = ((uint8_t*)half) + 256; \
01655 uint8_t * const halfHV = ((uint8_t*)half); \
01656 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01657 stride, 17); \
01658 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
01659 16, 16); \
01660 OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV, stride, 16, 16); \
01661 } \
01662 \
01663 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src, \
01664 int stride) \
01665 { \
01666 uint64_t half[16 * 2 + 17 * 2]; \
01667 uint8_t * const halfH = ((uint8_t*)half) + 256; \
01668 uint8_t * const halfHV = ((uint8_t*)half); \
01669 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01670 stride, 17); \
01671 put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, \
01672 16, 16); \
01673 OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV, stride, \
01674 16, 16); \
01675 } \
01676 \
01677 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src, \
01678 int stride) \
01679 { \
01680 uint64_t half[17 * 2]; \
01681 uint8_t * const halfH = ((uint8_t*)half); \
01682 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01683 stride, 17); \
01684 put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16, \
01685 stride, 17); \
01686 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16); \
01687 } \
01688 \
01689 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src, \
01690 int stride) \
01691 { \
01692 uint64_t half[17 * 2]; \
01693 uint8_t * const halfH = ((uint8_t*)half); \
01694 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01695 stride, 17); \
01696 put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16, \
01697 stride, 17); \
01698 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16); \
01699 } \
01700 \
01701 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src, \
01702 int stride) \
01703 { \
01704 uint64_t half[17 * 2]; \
01705 uint8_t * const halfH = ((uint8_t*)half); \
01706 put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, \
01707 stride, 17); \
01708 OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16); \
01709 }
01710
01711 #define PUT_OP(a, b, temp, size) \
01712 "mov"#size" "#a", "#b" \n\t"
01713
01714 #define AVG_3DNOW_OP(a, b, temp, size) \
01715 "mov"#size" "#b", "#temp" \n\t" \
01716 "pavgusb "#temp", "#a" \n\t" \
01717 "mov"#size" "#a", "#b" \n\t"
01718
01719 #define AVG_MMX2_OP(a, b, temp, size) \
01720 "mov"#size" "#b", "#temp" \n\t" \
01721 "pavgb "#temp", "#a" \n\t" \
01722 "mov"#size" "#a", "#b" \n\t"
01723
01724 QPEL_BASE(put_, ff_pw_16, _, PUT_OP, PUT_OP)
01725 QPEL_BASE(avg_, ff_pw_16, _, AVG_MMX2_OP, AVG_3DNOW_OP)
01726 QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
01727 QPEL_OP(put_, ff_pw_16, _, PUT_OP, 3dnow)
01728 QPEL_OP(avg_, ff_pw_16, _, AVG_3DNOW_OP, 3dnow)
01729 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
01730 QPEL_OP(put_, ff_pw_16, _, PUT_OP, mmx2)
01731 QPEL_OP(avg_, ff_pw_16, _, AVG_MMX2_OP, mmx2)
01732 QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
01733
01734
01735
01736
01737 #define QPEL_2TAP_XY(OPNAME, SIZE, MMX, XY, HPEL) \
01738 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, \
01739 uint8_t *src, \
01740 int stride) \
01741 { \
01742 OPNAME ## pixels ## SIZE ## HPEL(dst, src, stride, SIZE); \
01743 }
01744
01745 #define QPEL_2TAP_L3(OPNAME, SIZE, MMX, XY, S0, S1, S2) \
01746 static void OPNAME ## 2tap_qpel ## SIZE ## _mc ## XY ## _ ## MMX(uint8_t *dst, \
01747 uint8_t *src, \
01748 int stride) \
01749 { \
01750 OPNAME ## 2tap_qpel ## SIZE ## _l3_ ## MMX(dst, src + S0, stride, SIZE, \
01751 S1, S2); \
01752 }
01753
01754 #define QPEL_2TAP(OPNAME, SIZE, MMX) \
01755 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 20, _x2_ ## MMX) \
01756 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 02, _y2_ ## MMX) \
01757 QPEL_2TAP_XY(OPNAME, SIZE, MMX, 22, _xy2_mmx) \
01758 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc00_ ## MMX = \
01759 OPNAME ## qpel ## SIZE ## _mc00_ ## MMX; \
01760 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc21_ ## MMX = \
01761 OPNAME ## 2tap_qpel ## SIZE ## _mc20_ ## MMX; \
01762 static const qpel_mc_func OPNAME ## 2tap_qpel ## SIZE ## _mc12_ ## MMX = \
01763 OPNAME ## 2tap_qpel ## SIZE ## _mc02_ ## MMX; \
01764 static void OPNAME ## 2tap_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, \
01765 uint8_t *src, \
01766 int stride) \
01767 { \
01768 OPNAME ## pixels ## SIZE ## _y2_ ## MMX(dst, src + 1, stride, SIZE); \
01769 } \
01770 static void OPNAME ## 2tap_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, \
01771 uint8_t *src, \
01772 int stride) \
01773 { \
01774 OPNAME ## pixels ## SIZE ## _x2_ ## MMX(dst, src + stride, \
01775 stride, SIZE); \
01776 } \
01777 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 10, 0, 1, 0) \
01778 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 30, 1, -1, 0) \
01779 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 01, 0, stride, 0) \
01780 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 03, stride, -stride, 0) \
01781 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 11, 0, stride, 1) \
01782 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 31, 1, stride, -1) \
01783 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 13, stride, -stride, 1) \
01784 QPEL_2TAP_L3(OPNAME, SIZE, MMX, 33, stride + 1, -stride, -1) \
01785
01786 QPEL_2TAP(put_, 16, mmx2)
01787 QPEL_2TAP(avg_, 16, mmx2)
01788 QPEL_2TAP(put_, 8, mmx2)
01789 QPEL_2TAP(avg_, 8, mmx2)
01790 QPEL_2TAP(put_, 16, 3dnow)
01791 QPEL_2TAP(avg_, 16, 3dnow)
01792 QPEL_2TAP(put_, 8, 3dnow)
01793 QPEL_2TAP(avg_, 8, 3dnow)
01794
01795 void ff_put_rv40_qpel8_mc33_mmx(uint8_t *dst, uint8_t *src, int stride)
01796 {
01797 put_pixels8_xy2_mmx(dst, src, stride, 8);
01798 }
01799 void ff_put_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src, int stride)
01800 {
01801 put_pixels16_xy2_mmx(dst, src, stride, 16);
01802 }
01803 void ff_avg_rv40_qpel8_mc33_mmx(uint8_t *dst, uint8_t *src, int stride)
01804 {
01805 avg_pixels8_xy2_mmx(dst, src, stride, 8);
01806 }
01807 void ff_avg_rv40_qpel16_mc33_mmx(uint8_t *dst, uint8_t *src, int stride)
01808 {
01809 avg_pixels16_xy2_mmx(dst, src, stride, 16);
01810 }
01811
01812 #if HAVE_YASM
01813 typedef void emu_edge_core_func(uint8_t *buf, const uint8_t *src,
01814 x86_reg linesize, x86_reg start_y,
01815 x86_reg end_y, x86_reg block_h,
01816 x86_reg start_x, x86_reg end_x,
01817 x86_reg block_w);
01818 extern emu_edge_core_func ff_emu_edge_core_mmx;
01819 extern emu_edge_core_func ff_emu_edge_core_sse;
01820
01821 static av_always_inline void emulated_edge_mc(uint8_t *buf, const uint8_t *src,
01822 int linesize,
01823 int block_w, int block_h,
01824 int src_x, int src_y,
01825 int w, int h,
01826 emu_edge_core_func *core_fn)
01827 {
01828 int start_y, start_x, end_y, end_x, src_y_add = 0;
01829
01830 if (src_y >= h) {
01831 src_y_add = h - 1 - src_y;
01832 src_y = h - 1;
01833 } else if (src_y <= -block_h) {
01834 src_y_add = 1 - block_h - src_y;
01835 src_y = 1 - block_h;
01836 }
01837 if (src_x >= w) {
01838 src += w - 1 - src_x;
01839 src_x = w - 1;
01840 } else if (src_x <= -block_w) {
01841 src += 1 - block_w - src_x;
01842 src_x = 1 - block_w;
01843 }
01844
01845 start_y = FFMAX(0, -src_y);
01846 start_x = FFMAX(0, -src_x);
01847 end_y = FFMIN(block_h, h-src_y);
01848 end_x = FFMIN(block_w, w-src_x);
01849 assert(start_x < end_x && block_w > 0);
01850 assert(start_y < end_y && block_h > 0);
01851
01852
01853 src += (src_y_add + start_y) * linesize + start_x;
01854 buf += start_x;
01855 core_fn(buf, src, linesize, start_y, end_y,
01856 block_h, start_x, end_x, block_w);
01857 }
01858
01859 #if ARCH_X86_32
01860 static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src,
01861 int linesize,
01862 int block_w, int block_h,
01863 int src_x, int src_y, int w, int h)
01864 {
01865 emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
01866 w, h, &ff_emu_edge_core_mmx);
01867 }
01868 #endif
01869
01870 static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src,
01871 int linesize,
01872 int block_w, int block_h,
01873 int src_x, int src_y, int w, int h)
01874 {
01875 emulated_edge_mc(buf, src, linesize, block_w, block_h, src_x, src_y,
01876 w, h, &ff_emu_edge_core_sse);
01877 }
01878 #endif
01879
01880 typedef void emulated_edge_mc_func(uint8_t *dst, const uint8_t *src,
01881 int linesize, int block_w, int block_h,
01882 int src_x, int src_y, int w, int h);
01883
01884 static av_always_inline void gmc(uint8_t *dst, uint8_t *src,
01885 int stride, int h, int ox, int oy,
01886 int dxx, int dxy, int dyx, int dyy,
01887 int shift, int r, int width, int height,
01888 emulated_edge_mc_func *emu_edge_fn)
01889 {
01890 const int w = 8;
01891 const int ix = ox >> (16 + shift);
01892 const int iy = oy >> (16 + shift);
01893 const int oxs = ox >> 4;
01894 const int oys = oy >> 4;
01895 const int dxxs = dxx >> 4;
01896 const int dxys = dxy >> 4;
01897 const int dyxs = dyx >> 4;
01898 const int dyys = dyy >> 4;
01899 const uint16_t r4[4] = { r, r, r, r };
01900 const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys };
01901 const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys };
01902 const uint64_t shift2 = 2 * shift;
01903 uint8_t edge_buf[(h + 1) * stride];
01904 int x, y;
01905
01906 const int dxw = (dxx - (1 << (16 + shift))) * (w - 1);
01907 const int dyh = (dyy - (1 << (16 + shift))) * (h - 1);
01908 const int dxh = dxy * (h - 1);
01909 const int dyw = dyx * (w - 1);
01910 if (
01911 ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) |
01912 (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift)
01913
01914 || (dxx | dxy | dyx | dyy) & 15) {
01915
01916 ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy,
01917 shift, r, width, height);
01918 return;
01919 }
01920
01921 src += ix + iy * stride;
01922 if ((unsigned)ix >= width - w ||
01923 (unsigned)iy >= height - h) {
01924 emu_edge_fn(edge_buf, src, stride, w + 1, h + 1, ix, iy, width, height);
01925 src = edge_buf;
01926 }
01927
01928 __asm__ volatile (
01929 "movd %0, %%mm6 \n\t"
01930 "pxor %%mm7, %%mm7 \n\t"
01931 "punpcklwd %%mm6, %%mm6 \n\t"
01932 "punpcklwd %%mm6, %%mm6 \n\t"
01933 :: "r"(1<<shift)
01934 );
01935
01936 for (x = 0; x < w; x += 4) {
01937 uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0),
01938 oxs - dxys + dxxs * (x + 1),
01939 oxs - dxys + dxxs * (x + 2),
01940 oxs - dxys + dxxs * (x + 3) };
01941 uint16_t dy4[4] = { oys - dyys + dyxs * (x + 0),
01942 oys - dyys + dyxs * (x + 1),
01943 oys - dyys + dyxs * (x + 2),
01944 oys - dyys + dyxs * (x + 3) };
01945
01946 for (y = 0; y < h; y++) {
01947 __asm__ volatile (
01948 "movq %0, %%mm4 \n\t"
01949 "movq %1, %%mm5 \n\t"
01950 "paddw %2, %%mm4 \n\t"
01951 "paddw %3, %%mm5 \n\t"
01952 "movq %%mm4, %0 \n\t"
01953 "movq %%mm5, %1 \n\t"
01954 "psrlw $12, %%mm4 \n\t"
01955 "psrlw $12, %%mm5 \n\t"
01956 : "+m"(*dx4), "+m"(*dy4)
01957 : "m"(*dxy4), "m"(*dyy4)
01958 );
01959
01960 __asm__ volatile (
01961 "movq %%mm6, %%mm2 \n\t"
01962 "movq %%mm6, %%mm1 \n\t"
01963 "psubw %%mm4, %%mm2 \n\t"
01964 "psubw %%mm5, %%mm1 \n\t"
01965 "movq %%mm2, %%mm0 \n\t"
01966 "movq %%mm4, %%mm3 \n\t"
01967 "pmullw %%mm1, %%mm0 \n\t"
01968 "pmullw %%mm5, %%mm3 \n\t"
01969 "pmullw %%mm5, %%mm2 \n\t"
01970 "pmullw %%mm4, %%mm1 \n\t"
01971
01972 "movd %4, %%mm5 \n\t"
01973 "movd %3, %%mm4 \n\t"
01974 "punpcklbw %%mm7, %%mm5 \n\t"
01975 "punpcklbw %%mm7, %%mm4 \n\t"
01976 "pmullw %%mm5, %%mm3 \n\t"
01977 "pmullw %%mm4, %%mm2 \n\t"
01978
01979 "movd %2, %%mm5 \n\t"
01980 "movd %1, %%mm4 \n\t"
01981 "punpcklbw %%mm7, %%mm5 \n\t"
01982 "punpcklbw %%mm7, %%mm4 \n\t"
01983 "pmullw %%mm5, %%mm1 \n\t"
01984 "pmullw %%mm4, %%mm0 \n\t"
01985 "paddw %5, %%mm1 \n\t"
01986 "paddw %%mm3, %%mm2 \n\t"
01987 "paddw %%mm1, %%mm0 \n\t"
01988 "paddw %%mm2, %%mm0 \n\t"
01989
01990 "psrlw %6, %%mm0 \n\t"
01991 "packuswb %%mm0, %%mm0 \n\t"
01992 "movd %%mm0, %0 \n\t"
01993
01994 : "=m"(dst[x + y * stride])
01995 : "m"(src[0]), "m"(src[1]),
01996 "m"(src[stride]), "m"(src[stride + 1]),
01997 "m"(*r4), "m"(shift2)
01998 );
01999 src += stride;
02000 }
02001 src += 4 - h * stride;
02002 }
02003 }
02004
02005 #if HAVE_YASM
02006 #if ARCH_X86_32
02007 static void gmc_mmx(uint8_t *dst, uint8_t *src,
02008 int stride, int h, int ox, int oy,
02009 int dxx, int dxy, int dyx, int dyy,
02010 int shift, int r, int width, int height)
02011 {
02012 gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
02013 width, height, &emulated_edge_mc_mmx);
02014 }
02015 #endif
02016 static void gmc_sse(uint8_t *dst, uint8_t *src,
02017 int stride, int h, int ox, int oy,
02018 int dxx, int dxy, int dyx, int dyy,
02019 int shift, int r, int width, int height)
02020 {
02021 gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
02022 width, height, &emulated_edge_mc_sse);
02023 }
02024 #else
02025 static void gmc_mmx(uint8_t *dst, uint8_t *src,
02026 int stride, int h, int ox, int oy,
02027 int dxx, int dxy, int dyx, int dyy,
02028 int shift, int r, int width, int height)
02029 {
02030 gmc(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r,
02031 width, height, &ff_emulated_edge_mc_8);
02032 }
02033 #endif
02034
02035 #define PREFETCH(name, op) \
02036 static void name(void *mem, int stride, int h) \
02037 { \
02038 const uint8_t *p = mem; \
02039 do { \
02040 __asm__ volatile (#op" %0" :: "m"(*p)); \
02041 p += stride; \
02042 } while (--h); \
02043 }
02044
02045 PREFETCH(prefetch_mmx2, prefetcht0)
02046 PREFETCH(prefetch_3dnow, prefetch)
02047 #undef PREFETCH
02048
02049 #include "h264_qpel_mmx.c"
02050
02051 void ff_put_h264_chroma_mc8_mmx_rnd (uint8_t *dst, uint8_t *src,
02052 int stride, int h, int x, int y);
02053 void ff_avg_h264_chroma_mc8_mmx2_rnd (uint8_t *dst, uint8_t *src,
02054 int stride, int h, int x, int y);
02055 void ff_avg_h264_chroma_mc8_3dnow_rnd(uint8_t *dst, uint8_t *src,
02056 int stride, int h, int x, int y);
02057
02058 void ff_put_h264_chroma_mc4_mmx (uint8_t *dst, uint8_t *src,
02059 int stride, int h, int x, int y);
02060 void ff_avg_h264_chroma_mc4_mmx2 (uint8_t *dst, uint8_t *src,
02061 int stride, int h, int x, int y);
02062 void ff_avg_h264_chroma_mc4_3dnow (uint8_t *dst, uint8_t *src,
02063 int stride, int h, int x, int y);
02064
02065 void ff_put_h264_chroma_mc2_mmx2 (uint8_t *dst, uint8_t *src,
02066 int stride, int h, int x, int y);
02067 void ff_avg_h264_chroma_mc2_mmx2 (uint8_t *dst, uint8_t *src,
02068 int stride, int h, int x, int y);
02069
02070 void ff_put_h264_chroma_mc8_ssse3_rnd(uint8_t *dst, uint8_t *src,
02071 int stride, int h, int x, int y);
02072 void ff_put_h264_chroma_mc4_ssse3 (uint8_t *dst, uint8_t *src,
02073 int stride, int h, int x, int y);
02074
02075 void ff_avg_h264_chroma_mc8_ssse3_rnd(uint8_t *dst, uint8_t *src,
02076 int stride, int h, int x, int y);
02077 void ff_avg_h264_chroma_mc4_ssse3 (uint8_t *dst, uint8_t *src,
02078 int stride, int h, int x, int y);
02079
02080 #define CHROMA_MC(OP, NUM, DEPTH, OPT) \
02081 void ff_ ## OP ## _h264_chroma_mc ## NUM ## _ ## DEPTH ## _ ## OPT \
02082 (uint8_t *dst, uint8_t *src, \
02083 int stride, int h, int x, int y);
02084
02085 CHROMA_MC(put, 2, 10, mmxext)
02086 CHROMA_MC(avg, 2, 10, mmxext)
02087 CHROMA_MC(put, 4, 10, mmxext)
02088 CHROMA_MC(avg, 4, 10, mmxext)
02089 CHROMA_MC(put, 8, 10, sse2)
02090 CHROMA_MC(avg, 8, 10, sse2)
02091 CHROMA_MC(put, 8, 10, avx)
02092 CHROMA_MC(avg, 8, 10, avx)
02093
02094
02095 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride)
02096 {
02097 put_pixels8_mmx(dst, src, stride, 8);
02098 }
02099
02100 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride)
02101 {
02102 avg_pixels8_mmx(dst, src, stride, 8);
02103 }
02104
02105 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride)
02106 {
02107 put_pixels16_mmx(dst, src, stride, 16);
02108 }
02109
02110 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride)
02111 {
02112 avg_pixels16_mmx(dst, src, stride, 16);
02113 }
02114
02115
02116 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src,
02117 int stride, int rnd)
02118 {
02119 put_pixels8_mmx(dst, src, stride, 8);
02120 }
02121
02122 void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src,
02123 int stride, int rnd)
02124 {
02125 avg_pixels8_mmx2(dst, src, stride, 8);
02126 }
02127
02128
02129 static void put_vp_no_rnd_pixels8_l2_mmx(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h)
02130 {
02131
02132 MOVQ_BFE(mm6);
02133 __asm__ volatile(
02134 "1: \n\t"
02135 "movq (%1), %%mm0 \n\t"
02136 "movq (%2), %%mm1 \n\t"
02137 "movq (%1,%4), %%mm2 \n\t"
02138 "movq (%2,%4), %%mm3 \n\t"
02139 PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
02140 "movq %%mm4, (%3) \n\t"
02141 "movq %%mm5, (%3,%4) \n\t"
02142
02143 "movq (%1,%4,2), %%mm0 \n\t"
02144 "movq (%2,%4,2), %%mm1 \n\t"
02145 "movq (%1,%5), %%mm2 \n\t"
02146 "movq (%2,%5), %%mm3 \n\t"
02147 "lea (%1,%4,4), %1 \n\t"
02148 "lea (%2,%4,4), %2 \n\t"
02149 PAVGBP_MMX_NO_RND(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
02150 "movq %%mm4, (%3,%4,2) \n\t"
02151 "movq %%mm5, (%3,%5) \n\t"
02152 "lea (%3,%4,4), %3 \n\t"
02153 "subl $4, %0 \n\t"
02154 "jnz 1b \n\t"
02155 :"+r"(h), "+r"(a), "+r"(b), "+r"(dst)
02156 :"r"((x86_reg)stride), "r"((x86_reg)3L*stride)
02157 :"memory");
02158
02159 }
02160 static void put_vp_no_rnd_pixels16_l2_mmx(uint8_t *dst, const uint8_t *a, const uint8_t *b, int stride, int h)
02161 {
02162 put_vp_no_rnd_pixels8_l2_mmx(dst, a, b, stride, h);
02163 put_vp_no_rnd_pixels8_l2_mmx(dst+8, a+8, b+8, stride, h);
02164 }
02165
02166 #if CONFIG_DIRAC_DECODER
02167 #define DIRAC_PIXOP(OPNAME, EXT)\
02168 void ff_ ## OPNAME ## _dirac_pixels8_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
02169 {\
02170 OPNAME ## _pixels8_ ## EXT(dst, src[0], stride, h);\
02171 }\
02172 void ff_ ## OPNAME ## _dirac_pixels16_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
02173 {\
02174 OPNAME ## _pixels16_ ## EXT(dst, src[0], stride, h);\
02175 }\
02176 void ff_ ## OPNAME ## _dirac_pixels32_ ## EXT(uint8_t *dst, const uint8_t *src[5], int stride, int h)\
02177 {\
02178 OPNAME ## _pixels16_ ## EXT(dst , src[0] , stride, h);\
02179 OPNAME ## _pixels16_ ## EXT(dst+16, src[0]+16, stride, h);\
02180 }
02181
02182 DIRAC_PIXOP(put, mmx)
02183 DIRAC_PIXOP(avg, mmx)
02184 DIRAC_PIXOP(avg, mmx2)
02185
02186 void ff_put_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
02187 {
02188 put_pixels16_sse2(dst, src[0], stride, h);
02189 }
02190 void ff_avg_dirac_pixels16_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
02191 {
02192 avg_pixels16_sse2(dst, src[0], stride, h);
02193 }
02194 void ff_put_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
02195 {
02196 put_pixels16_sse2(dst , src[0] , stride, h);
02197 put_pixels16_sse2(dst+16, src[0]+16, stride, h);
02198 }
02199 void ff_avg_dirac_pixels32_sse2(uint8_t *dst, const uint8_t *src[5], int stride, int h)
02200 {
02201 avg_pixels16_sse2(dst , src[0] , stride, h);
02202 avg_pixels16_sse2(dst+16, src[0]+16, stride, h);
02203 }
02204 #endif
02205
02206
02207
02208 #if CONFIG_GPL
02209 static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size,
02210 DCTELEM *block)
02211 {
02212 ff_mmx_idct(block);
02213 ff_put_pixels_clamped_mmx(block, dest, line_size);
02214 }
02215
02216 static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size,
02217 DCTELEM *block)
02218 {
02219 ff_mmx_idct(block);
02220 ff_add_pixels_clamped_mmx(block, dest, line_size);
02221 }
02222
02223 static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size,
02224 DCTELEM *block)
02225 {
02226 ff_mmxext_idct(block);
02227 ff_put_pixels_clamped_mmx(block, dest, line_size);
02228 }
02229
02230 static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size,
02231 DCTELEM *block)
02232 {
02233 ff_mmxext_idct(block);
02234 ff_add_pixels_clamped_mmx(block, dest, line_size);
02235 }
02236 #endif
02237
02238 static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block)
02239 {
02240 ff_idct_xvid_mmx(block);
02241 ff_put_pixels_clamped_mmx(block, dest, line_size);
02242 }
02243
02244 static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block)
02245 {
02246 ff_idct_xvid_mmx(block);
02247 ff_add_pixels_clamped_mmx(block, dest, line_size);
02248 }
02249
02250 static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block)
02251 {
02252 ff_idct_xvid_mmx2(block);
02253 ff_put_pixels_clamped_mmx(block, dest, line_size);
02254 }
02255
02256 static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block)
02257 {
02258 ff_idct_xvid_mmx2(block);
02259 ff_add_pixels_clamped_mmx(block, dest, line_size);
02260 }
02261
02262 static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize)
02263 {
02264 int i;
02265 __asm__ volatile ("pxor %%mm7, %%mm7":);
02266 for (i = 0; i < blocksize; i += 2) {
02267 __asm__ volatile (
02268 "movq %0, %%mm0 \n\t"
02269 "movq %1, %%mm1 \n\t"
02270 "movq %%mm0, %%mm2 \n\t"
02271 "movq %%mm1, %%mm3 \n\t"
02272 "pfcmpge %%mm7, %%mm2 \n\t"
02273 "pfcmpge %%mm7, %%mm3 \n\t"
02274 "pslld $31, %%mm2 \n\t"
02275 "pxor %%mm2, %%mm1 \n\t"
02276 "movq %%mm3, %%mm4 \n\t"
02277 "pand %%mm1, %%mm3 \n\t"
02278 "pandn %%mm1, %%mm4 \n\t"
02279 "pfadd %%mm0, %%mm3 \n\t"
02280 "pfsub %%mm4, %%mm0 \n\t"
02281 "movq %%mm3, %1 \n\t"
02282 "movq %%mm0, %0 \n\t"
02283 : "+m"(mag[i]), "+m"(ang[i])
02284 :: "memory"
02285 );
02286 }
02287 __asm__ volatile ("femms");
02288 }
02289
02290 static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize)
02291 {
02292 int i;
02293
02294 __asm__ volatile (
02295 "movaps %0, %%xmm5 \n\t"
02296 :: "m"(ff_pdw_80000000[0])
02297 );
02298 for (i = 0; i < blocksize; i += 4) {
02299 __asm__ volatile (
02300 "movaps %0, %%xmm0 \n\t"
02301 "movaps %1, %%xmm1 \n\t"
02302 "xorps %%xmm2, %%xmm2 \n\t"
02303 "xorps %%xmm3, %%xmm3 \n\t"
02304 "cmpleps %%xmm0, %%xmm2 \n\t"
02305 "cmpleps %%xmm1, %%xmm3 \n\t"
02306 "andps %%xmm5, %%xmm2 \n\t"
02307 "xorps %%xmm2, %%xmm1 \n\t"
02308 "movaps %%xmm3, %%xmm4 \n\t"
02309 "andps %%xmm1, %%xmm3 \n\t"
02310 "andnps %%xmm1, %%xmm4 \n\t"
02311 "addps %%xmm0, %%xmm3 \n\t"
02312 "subps %%xmm4, %%xmm0 \n\t"
02313 "movaps %%xmm3, %1 \n\t"
02314 "movaps %%xmm0, %0 \n\t"
02315 : "+m"(mag[i]), "+m"(ang[i])
02316 :: "memory"
02317 );
02318 }
02319 }
02320
02321 #define IF1(x) x
02322 #define IF0(x)
02323
02324 #define MIX5(mono, stereo) \
02325 __asm__ volatile ( \
02326 "movss 0(%2), %%xmm5 \n" \
02327 "movss 8(%2), %%xmm6 \n" \
02328 "movss 24(%2), %%xmm7 \n" \
02329 "shufps $0, %%xmm5, %%xmm5 \n" \
02330 "shufps $0, %%xmm6, %%xmm6 \n" \
02331 "shufps $0, %%xmm7, %%xmm7 \n" \
02332 "1: \n" \
02333 "movaps (%0, %1), %%xmm0 \n" \
02334 "movaps 0x400(%0, %1), %%xmm1 \n" \
02335 "movaps 0x800(%0, %1), %%xmm2 \n" \
02336 "movaps 0xc00(%0, %1), %%xmm3 \n" \
02337 "movaps 0x1000(%0, %1), %%xmm4 \n" \
02338 "mulps %%xmm5, %%xmm0 \n" \
02339 "mulps %%xmm6, %%xmm1 \n" \
02340 "mulps %%xmm5, %%xmm2 \n" \
02341 "mulps %%xmm7, %%xmm3 \n" \
02342 "mulps %%xmm7, %%xmm4 \n" \
02343 stereo("addps %%xmm1, %%xmm0 \n") \
02344 "addps %%xmm1, %%xmm2 \n" \
02345 "addps %%xmm3, %%xmm0 \n" \
02346 "addps %%xmm4, %%xmm2 \n" \
02347 mono("addps %%xmm2, %%xmm0 \n") \
02348 "movaps %%xmm0, (%0, %1) \n" \
02349 stereo("movaps %%xmm2, 0x400(%0, %1) \n") \
02350 "add $16, %0 \n" \
02351 "jl 1b \n" \
02352 : "+&r"(i) \
02353 : "r"(samples[0] + len), "r"(matrix) \
02354 : XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \
02355 "%xmm4", "%xmm5", "%xmm6", "%xmm7",) \
02356 "memory" \
02357 );
02358
02359 #define MIX_MISC(stereo) \
02360 __asm__ volatile ( \
02361 "1: \n" \
02362 "movaps (%3, %0), %%xmm0 \n" \
02363 stereo("movaps %%xmm0, %%xmm1 \n") \
02364 "mulps %%xmm4, %%xmm0 \n" \
02365 stereo("mulps %%xmm5, %%xmm1 \n") \
02366 "lea 1024(%3, %0), %1 \n" \
02367 "mov %5, %2 \n" \
02368 "2: \n" \
02369 "movaps (%1), %%xmm2 \n" \
02370 stereo("movaps %%xmm2, %%xmm3 \n") \
02371 "mulps (%4, %2), %%xmm2 \n" \
02372 stereo("mulps 16(%4, %2), %%xmm3 \n") \
02373 "addps %%xmm2, %%xmm0 \n" \
02374 stereo("addps %%xmm3, %%xmm1 \n") \
02375 "add $1024, %1 \n" \
02376 "add $32, %2 \n" \
02377 "jl 2b \n" \
02378 "movaps %%xmm0, (%3, %0) \n" \
02379 stereo("movaps %%xmm1, 1024(%3, %0) \n") \
02380 "add $16, %0 \n" \
02381 "jl 1b \n" \
02382 : "+&r"(i), "=&r"(j), "=&r"(k) \
02383 : "r"(samples[0] + len), "r"(matrix_simd + in_ch), \
02384 "g"((intptr_t) - 32 * (in_ch - 1)) \
02385 : "memory" \
02386 );
02387
02388 static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2],
02389 int out_ch, int in_ch, int len)
02390 {
02391 int (*matrix_cmp)[2] = (int(*)[2])matrix;
02392 intptr_t i, j, k;
02393
02394 i = -len * sizeof(float);
02395 if (in_ch == 5 && out_ch == 2 &&
02396 !(matrix_cmp[0][1] | matrix_cmp[2][0] |
02397 matrix_cmp[3][1] | matrix_cmp[4][0] |
02398 (matrix_cmp[1][0] ^ matrix_cmp[1][1]) |
02399 (matrix_cmp[0][0] ^ matrix_cmp[2][1]))) {
02400 MIX5(IF0, IF1);
02401 } else if (in_ch == 5 && out_ch == 1 &&
02402 matrix_cmp[0][0] == matrix_cmp[2][0] &&
02403 matrix_cmp[3][0] == matrix_cmp[4][0]) {
02404 MIX5(IF1, IF0);
02405 } else {
02406 DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
02407 j = 2 * in_ch * sizeof(float);
02408 __asm__ volatile (
02409 "1: \n"
02410 "sub $8, %0 \n"
02411 "movss (%2, %0), %%xmm4 \n"
02412 "movss 4(%2, %0), %%xmm5 \n"
02413 "shufps $0, %%xmm4, %%xmm4 \n"
02414 "shufps $0, %%xmm5, %%xmm5 \n"
02415 "movaps %%xmm4, (%1, %0, 4) \n"
02416 "movaps %%xmm5, 16(%1, %0, 4) \n"
02417 "jg 1b \n"
02418 : "+&r"(j)
02419 : "r"(matrix_simd), "r"(matrix)
02420 : "memory"
02421 );
02422 if (out_ch == 2) {
02423 MIX_MISC(IF1);
02424 } else {
02425 MIX_MISC(IF0);
02426 }
02427 }
02428 }
02429
02430 #if HAVE_6REGS
02431 static void vector_fmul_window_3dnow2(float *dst, const float *src0,
02432 const float *src1, const float *win,
02433 int len)
02434 {
02435 x86_reg i = -len * 4;
02436 x86_reg j = len * 4 - 8;
02437 __asm__ volatile (
02438 "1: \n"
02439 "pswapd (%5, %1), %%mm1 \n"
02440 "movq (%5, %0), %%mm0 \n"
02441 "pswapd (%4, %1), %%mm5 \n"
02442 "movq (%3, %0), %%mm4 \n"
02443 "movq %%mm0, %%mm2 \n"
02444 "movq %%mm1, %%mm3 \n"
02445 "pfmul %%mm4, %%mm2 \n"
02446 "pfmul %%mm5, %%mm3 \n"
02447 "pfmul %%mm4, %%mm1 \n"
02448 "pfmul %%mm5, %%mm0 \n"
02449 "pfadd %%mm3, %%mm2 \n"
02450 "pfsub %%mm0, %%mm1 \n"
02451 "pswapd %%mm2, %%mm2 \n"
02452 "movq %%mm1, (%2, %0) \n"
02453 "movq %%mm2, (%2, %1) \n"
02454 "sub $8, %1 \n"
02455 "add $8, %0 \n"
02456 "jl 1b \n"
02457 "femms \n"
02458 : "+r"(i), "+r"(j)
02459 : "r"(dst + len), "r"(src0 + len), "r"(src1), "r"(win + len)
02460 );
02461 }
02462
02463 static void vector_fmul_window_sse(float *dst, const float *src0,
02464 const float *src1, const float *win, int len)
02465 {
02466 x86_reg i = -len * 4;
02467 x86_reg j = len * 4 - 16;
02468 __asm__ volatile (
02469 "1: \n"
02470 "movaps (%5, %1), %%xmm1 \n"
02471 "movaps (%5, %0), %%xmm0 \n"
02472 "movaps (%4, %1), %%xmm5 \n"
02473 "movaps (%3, %0), %%xmm4 \n"
02474 "shufps $0x1b, %%xmm1, %%xmm1 \n"
02475 "shufps $0x1b, %%xmm5, %%xmm5 \n"
02476 "movaps %%xmm0, %%xmm2 \n"
02477 "movaps %%xmm1, %%xmm3 \n"
02478 "mulps %%xmm4, %%xmm2 \n"
02479 "mulps %%xmm5, %%xmm3 \n"
02480 "mulps %%xmm4, %%xmm1 \n"
02481 "mulps %%xmm5, %%xmm0 \n"
02482 "addps %%xmm3, %%xmm2 \n"
02483 "subps %%xmm0, %%xmm1 \n"
02484 "shufps $0x1b, %%xmm2, %%xmm2 \n"
02485 "movaps %%xmm1, (%2, %0) \n"
02486 "movaps %%xmm2, (%2, %1) \n"
02487 "sub $16, %1 \n"
02488 "add $16, %0 \n"
02489 "jl 1b \n"
02490 : "+r"(i), "+r"(j)
02491 : "r"(dst + len), "r"(src0 + len), "r"(src1), "r"(win + len)
02492 );
02493 }
02494 #endif
02495
02496 static void vector_clipf_sse(float *dst, const float *src,
02497 float min, float max, int len)
02498 {
02499 x86_reg i = (len - 16) * 4;
02500 __asm__ volatile (
02501 "movss %3, %%xmm4 \n\t"
02502 "movss %4, %%xmm5 \n\t"
02503 "shufps $0, %%xmm4, %%xmm4 \n\t"
02504 "shufps $0, %%xmm5, %%xmm5 \n\t"
02505 "1: \n\t"
02506 "movaps (%2, %0), %%xmm0 \n\t"
02507 "movaps 16(%2, %0), %%xmm1 \n\t"
02508 "movaps 32(%2, %0), %%xmm2 \n\t"
02509 "movaps 48(%2, %0), %%xmm3 \n\t"
02510 "maxps %%xmm4, %%xmm0 \n\t"
02511 "maxps %%xmm4, %%xmm1 \n\t"
02512 "maxps %%xmm4, %%xmm2 \n\t"
02513 "maxps %%xmm4, %%xmm3 \n\t"
02514 "minps %%xmm5, %%xmm0 \n\t"
02515 "minps %%xmm5, %%xmm1 \n\t"
02516 "minps %%xmm5, %%xmm2 \n\t"
02517 "minps %%xmm5, %%xmm3 \n\t"
02518 "movaps %%xmm0, (%1, %0) \n\t"
02519 "movaps %%xmm1, 16(%1, %0) \n\t"
02520 "movaps %%xmm2, 32(%1, %0) \n\t"
02521 "movaps %%xmm3, 48(%1, %0) \n\t"
02522 "sub $64, %0 \n\t"
02523 "jge 1b \n\t"
02524 : "+&r"(i)
02525 : "r"(dst), "r"(src), "m"(min), "m"(max)
02526 : "memory"
02527 );
02528 }
02529
02530 void ff_vp3_idct_mmx(int16_t *input_data);
02531 void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block);
02532 void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block);
02533
02534 void ff_vp3_idct_dc_add_mmx2(uint8_t *dest, int line_size,
02535 const DCTELEM *block);
02536
02537 void ff_vp3_v_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
02538 void ff_vp3_h_loop_filter_mmx2(uint8_t *src, int stride, int *bounding_values);
02539
02540 void ff_vp3_idct_sse2(int16_t *input_data);
02541 void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block);
02542 void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block);
02543
02544 int32_t ff_scalarproduct_int16_mmx2(const int16_t *v1, const int16_t *v2,
02545 int order);
02546 int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2,
02547 int order);
02548 int32_t ff_scalarproduct_and_madd_int16_mmx2(int16_t *v1, const int16_t *v2,
02549 const int16_t *v3,
02550 int order, int mul);
02551 int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2,
02552 const int16_t *v3,
02553 int order, int mul);
02554 int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2,
02555 const int16_t *v3,
02556 int order, int mul);
02557
02558 void ff_apply_window_int16_mmxext (int16_t *output, const int16_t *input,
02559 const int16_t *window, unsigned int len);
02560 void ff_apply_window_int16_mmxext_ba (int16_t *output, const int16_t *input,
02561 const int16_t *window, unsigned int len);
02562 void ff_apply_window_int16_sse2 (int16_t *output, const int16_t *input,
02563 const int16_t *window, unsigned int len);
02564 void ff_apply_window_int16_sse2_ba (int16_t *output, const int16_t *input,
02565 const int16_t *window, unsigned int len);
02566 void ff_apply_window_int16_ssse3 (int16_t *output, const int16_t *input,
02567 const int16_t *window, unsigned int len);
02568 void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
02569 const int16_t *window, unsigned int len);
02570
02571 void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w);
02572 void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w);
02573
02574 void ff_add_hfyu_median_prediction_mmx2(uint8_t *dst, const uint8_t *top,
02575 const uint8_t *diff, int w,
02576 int *left, int *left_top);
02577 int ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src,
02578 int w, int left);
02579 int ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src,
02580 int w, int left);
02581
02582 float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
02583
02584 void ff_vector_fmul_sse(float *dst, const float *src0, const float *src1,
02585 int len);
02586 void ff_vector_fmul_avx(float *dst, const float *src0, const float *src1,
02587 int len);
02588
02589 void ff_vector_fmul_reverse_sse(float *dst, const float *src0,
02590 const float *src1, int len);
02591 void ff_vector_fmul_reverse_avx(float *dst, const float *src0,
02592 const float *src1, int len);
02593
02594 void ff_vector_fmul_add_sse(float *dst, const float *src0, const float *src1,
02595 const float *src2, int len);
02596 void ff_vector_fmul_add_avx(float *dst, const float *src0, const float *src1,
02597 const float *src2, int len);
02598
02599 void ff_vector_clip_int32_mmx (int32_t *dst, const int32_t *src,
02600 int32_t min, int32_t max, unsigned int len);
02601 void ff_vector_clip_int32_sse2 (int32_t *dst, const int32_t *src,
02602 int32_t min, int32_t max, unsigned int len);
02603 void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src,
02604 int32_t min, int32_t max, unsigned int len);
02605 void ff_vector_clip_int32_sse4 (int32_t *dst, const int32_t *src,
02606 int32_t min, int32_t max, unsigned int len);
02607
02608 extern void ff_butterflies_float_interleave_sse(float *dst, const float *src0,
02609 const float *src1, int len);
02610 extern void ff_butterflies_float_interleave_avx(float *dst, const float *src0,
02611 const float *src1, int len);
02612
02613 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX) \
02614 do { \
02615 c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
02616 c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
02617 c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
02618 c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
02619 c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
02620 c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
02621 c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
02622 c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
02623 c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
02624 c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
02625 c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
02626 c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
02627 c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
02628 c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
02629 c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
02630 c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
02631 } while (0)
02632
02633 #define SET_HPEL_FUNCS(PFX, IDX, SIZE, CPU) \
02634 do { \
02635 c->PFX ## _pixels_tab[IDX][0] = PFX ## _pixels ## SIZE ## _ ## CPU; \
02636 c->PFX ## _pixels_tab[IDX][1] = PFX ## _pixels ## SIZE ## _x2_ ## CPU; \
02637 c->PFX ## _pixels_tab[IDX][2] = PFX ## _pixels ## SIZE ## _y2_ ## CPU; \
02638 c->PFX ## _pixels_tab[IDX][3] = PFX ## _pixels ## SIZE ## _xy2_ ## CPU; \
02639 } while (0)
02640
02641 #define H264_QPEL_FUNCS(x, y, CPU) \
02642 do { \
02643 c->put_h264_qpel_pixels_tab[0][x + y * 4] = put_h264_qpel16_mc ## x ## y ## _ ## CPU; \
02644 c->put_h264_qpel_pixels_tab[1][x + y * 4] = put_h264_qpel8_mc ## x ## y ## _ ## CPU; \
02645 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = avg_h264_qpel16_mc ## x ## y ## _ ## CPU; \
02646 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = avg_h264_qpel8_mc ## x ## y ## _ ## CPU; \
02647 } while (0)
02648
02649 #define H264_QPEL_FUNCS_10(x, y, CPU) \
02650 do { \
02651 c->put_h264_qpel_pixels_tab[0][x + y * 4] = ff_put_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
02652 c->put_h264_qpel_pixels_tab[1][x + y * 4] = ff_put_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
02653 c->avg_h264_qpel_pixels_tab[0][x + y * 4] = ff_avg_h264_qpel16_mc ## x ## y ## _10_ ## CPU; \
02654 c->avg_h264_qpel_pixels_tab[1][x + y * 4] = ff_avg_h264_qpel8_mc ## x ## y ## _10_ ## CPU; \
02655 } while (0)
02656
02657 static void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx, int mm_flags)
02658 {
02659 const int high_bit_depth = avctx->bits_per_raw_sample > 8;
02660
02661 c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
02662 c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
02663 c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
02664
02665 if (!high_bit_depth) {
02666 c->clear_block = clear_block_mmx;
02667 c->clear_blocks = clear_blocks_mmx;
02668 c->draw_edges = draw_edges_mmx;
02669
02670 SET_HPEL_FUNCS(put, 0, 16, mmx);
02671 SET_HPEL_FUNCS(put_no_rnd, 0, 16, mmx);
02672 SET_HPEL_FUNCS(avg, 0, 16, mmx);
02673 SET_HPEL_FUNCS(avg_no_rnd, 0, 16, mmx);
02674 SET_HPEL_FUNCS(put, 1, 8, mmx);
02675 SET_HPEL_FUNCS(put_no_rnd, 1, 8, mmx);
02676 SET_HPEL_FUNCS(avg, 1, 8, mmx);
02677 SET_HPEL_FUNCS(avg_no_rnd, 1, 8, mmx);
02678 }
02679
02680 #if ARCH_X86_32 || !HAVE_YASM
02681 c->gmc = gmc_mmx;
02682 #endif
02683 #if ARCH_X86_32 && HAVE_YASM
02684 if (!high_bit_depth)
02685 c->emulated_edge_mc = emulated_edge_mc_mmx;
02686 #endif
02687
02688 c->add_bytes = add_bytes_mmx;
02689
02690 c->put_no_rnd_pixels_l2[0]= put_vp_no_rnd_pixels16_l2_mmx;
02691 c->put_no_rnd_pixels_l2[1]= put_vp_no_rnd_pixels8_l2_mmx;
02692
02693 if (CONFIG_H263_DECODER || CONFIG_H263_ENCODER) {
02694 c->h263_v_loop_filter = h263_v_loop_filter_mmx;
02695 c->h263_h_loop_filter = h263_h_loop_filter_mmx;
02696 }
02697
02698 #if HAVE_YASM
02699 if (!high_bit_depth && CONFIG_H264CHROMA) {
02700 c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_mmx_rnd;
02701 c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_mmx;
02702 }
02703
02704 c->vector_clip_int32 = ff_vector_clip_int32_mmx;
02705 #endif
02706
02707 }
02708
02709 static void dsputil_init_mmx2(DSPContext *c, AVCodecContext *avctx,
02710 int mm_flags)
02711 {
02712 const int bit_depth = avctx->bits_per_raw_sample;
02713 const int high_bit_depth = bit_depth > 8;
02714
02715 c->prefetch = prefetch_mmx2;
02716
02717 if (!high_bit_depth) {
02718 c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
02719 c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
02720
02721 c->avg_pixels_tab[0][0] = avg_pixels16_mmx2;
02722 c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2;
02723 c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2;
02724
02725 c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2;
02726 c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2;
02727
02728 c->avg_pixels_tab[1][0] = avg_pixels8_mmx2;
02729 c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
02730 c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
02731 }
02732
02733 if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
02734 if (!high_bit_depth) {
02735 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
02736 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2;
02737 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2;
02738 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2;
02739
02740 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2;
02741 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
02742 }
02743
02744 if (CONFIG_VP3_DECODER && HAVE_YASM) {
02745 c->vp3_v_loop_filter = ff_vp3_v_loop_filter_mmx2;
02746 c->vp3_h_loop_filter = ff_vp3_h_loop_filter_mmx2;
02747 }
02748 }
02749 if (CONFIG_VP3_DECODER && HAVE_YASM)
02750 c->vp3_idct_dc_add = ff_vp3_idct_dc_add_mmx2;
02751
02752 if (CONFIG_VP3_DECODER && (avctx->codec_id == CODEC_ID_VP3 ||
02753 avctx->codec_id == CODEC_ID_THEORA)) {
02754 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_mmx2;
02755 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_mmx2;
02756 }
02757
02758 if (CONFIG_H264QPEL) {
02759 SET_QPEL_FUNCS(put_qpel, 0, 16, mmx2, );
02760 SET_QPEL_FUNCS(put_qpel, 1, 8, mmx2, );
02761 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmx2, );
02762 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, mmx2, );
02763 SET_QPEL_FUNCS(avg_qpel, 0, 16, mmx2, );
02764 SET_QPEL_FUNCS(avg_qpel, 1, 8, mmx2, );
02765
02766 if (!high_bit_depth) {
02767 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, mmx2, );
02768 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, mmx2, );
02769 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, mmx2, );
02770 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, mmx2, );
02771 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, mmx2, );
02772 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, mmx2, );
02773 } else if (bit_depth == 10) {
02774 #if HAVE_YASM
02775 #if !ARCH_X86_64
02776 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_mmxext, ff_);
02777 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_mmxext, ff_);
02778 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_mmxext, ff_);
02779 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_mmxext, ff_);
02780 #endif
02781 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 10_mmxext, ff_);
02782 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 10_mmxext, ff_);
02783 #endif
02784 }
02785
02786 SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, mmx2, );
02787 SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, mmx2, );
02788 SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, mmx2, );
02789 SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, mmx2, );
02790 }
02791
02792 #if HAVE_YASM
02793 if (!high_bit_depth && CONFIG_H264CHROMA) {
02794 c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_mmx2_rnd;
02795 c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_mmx2;
02796 c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_mmx2;
02797 c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_mmx2;
02798 }
02799 if (bit_depth == 10 && CONFIG_H264CHROMA) {
02800 c->put_h264_chroma_pixels_tab[2] = ff_put_h264_chroma_mc2_10_mmxext;
02801 c->avg_h264_chroma_pixels_tab[2] = ff_avg_h264_chroma_mc2_10_mmxext;
02802 c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_10_mmxext;
02803 c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_10_mmxext;
02804 }
02805
02806 c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmx2;
02807
02808 c->scalarproduct_int16 = ff_scalarproduct_int16_mmx2;
02809 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmx2;
02810
02811 if (avctx->flags & CODEC_FLAG_BITEXACT) {
02812 c->apply_window_int16 = ff_apply_window_int16_mmxext_ba;
02813 } else {
02814 c->apply_window_int16 = ff_apply_window_int16_mmxext;
02815 }
02816 #endif
02817 }
02818
02819 static void dsputil_init_3dnow(DSPContext *c, AVCodecContext *avctx,
02820 int mm_flags)
02821 {
02822 const int high_bit_depth = avctx->bits_per_raw_sample > 8;
02823
02824 c->prefetch = prefetch_3dnow;
02825
02826 if (!high_bit_depth) {
02827 c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
02828 c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
02829
02830 c->avg_pixels_tab[0][0] = avg_pixels16_3dnow;
02831 c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow;
02832 c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow;
02833
02834 c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow;
02835 c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow;
02836
02837 c->avg_pixels_tab[1][0] = avg_pixels8_3dnow;
02838 c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
02839 c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
02840
02841 if (!(avctx->flags & CODEC_FLAG_BITEXACT)){
02842 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow;
02843 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow;
02844 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow;
02845 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow;
02846
02847 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow;
02848 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
02849 }
02850 }
02851
02852 if (CONFIG_VP3_DECODER && (avctx->codec_id == CODEC_ID_VP3 ||
02853 avctx->codec_id == CODEC_ID_THEORA)) {
02854 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_exact_3dnow;
02855 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_exact_3dnow;
02856 }
02857
02858 if (CONFIG_H264QPEL) {
02859 SET_QPEL_FUNCS(put_qpel, 0, 16, 3dnow, );
02860 SET_QPEL_FUNCS(put_qpel, 1, 8, 3dnow, );
02861 SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, 3dnow, );
02862 SET_QPEL_FUNCS(put_no_rnd_qpel, 1, 8, 3dnow, );
02863 SET_QPEL_FUNCS(avg_qpel, 0, 16, 3dnow, );
02864 SET_QPEL_FUNCS(avg_qpel, 1, 8, 3dnow, );
02865
02866 if (!high_bit_depth) {
02867 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 3dnow, );
02868 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 3dnow, );
02869 SET_QPEL_FUNCS(put_h264_qpel, 2, 4, 3dnow, );
02870 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 3dnow, );
02871 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 3dnow, );
02872 SET_QPEL_FUNCS(avg_h264_qpel, 2, 4, 3dnow, );
02873 }
02874
02875 SET_QPEL_FUNCS(put_2tap_qpel, 0, 16, 3dnow, );
02876 SET_QPEL_FUNCS(put_2tap_qpel, 1, 8, 3dnow, );
02877 SET_QPEL_FUNCS(avg_2tap_qpel, 0, 16, 3dnow, );
02878 SET_QPEL_FUNCS(avg_2tap_qpel, 1, 8, 3dnow, );
02879 }
02880
02881 #if HAVE_YASM
02882 if (!high_bit_depth && CONFIG_H264CHROMA) {
02883 c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_3dnow_rnd;
02884 c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_3dnow;
02885 }
02886 #endif
02887
02888 c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
02889
02890 #if HAVE_7REGS
02891 c->add_hfyu_median_prediction = add_hfyu_median_prediction_cmov;
02892 #endif
02893 }
02894
02895 static void dsputil_init_3dnow2(DSPContext *c, AVCodecContext *avctx,
02896 int mm_flags)
02897 {
02898 #if HAVE_6REGS
02899 c->vector_fmul_window = vector_fmul_window_3dnow2;
02900 #endif
02901 }
02902
02903 static void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, int mm_flags)
02904 {
02905 const int high_bit_depth = avctx->bits_per_raw_sample > 8;
02906
02907 if (!high_bit_depth) {
02908 if (!(CONFIG_MPEG_XVMC_DECODER && avctx->xvmc_acceleration > 1)) {
02909
02910 c->clear_block = clear_block_sse;
02911 c->clear_blocks = clear_blocks_sse;
02912 }
02913 }
02914
02915 c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
02916 c->ac3_downmix = ac3_downmix_sse;
02917 #if HAVE_YASM
02918 c->vector_fmul = ff_vector_fmul_sse;
02919 c->vector_fmul_reverse = ff_vector_fmul_reverse_sse;
02920 c->vector_fmul_add = ff_vector_fmul_add_sse;
02921 #endif
02922
02923 #if HAVE_6REGS
02924 c->vector_fmul_window = vector_fmul_window_sse;
02925 #endif
02926
02927 c->vector_clipf = vector_clipf_sse;
02928
02929 #if HAVE_YASM
02930 c->scalarproduct_float = ff_scalarproduct_float_sse;
02931 c->butterflies_float_interleave = ff_butterflies_float_interleave_sse;
02932
02933 if (!high_bit_depth)
02934 c->emulated_edge_mc = emulated_edge_mc_sse;
02935 c->gmc = gmc_sse;
02936 #endif
02937 }
02938
02939 static void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
02940 int mm_flags)
02941 {
02942 const int bit_depth = avctx->bits_per_raw_sample;
02943 const int high_bit_depth = bit_depth > 8;
02944
02945 if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
02946
02947 if (!high_bit_depth) {
02948 c->put_pixels_tab[0][0] = put_pixels16_sse2;
02949 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_sse2;
02950 c->avg_pixels_tab[0][0] = avg_pixels16_sse2;
02951 if (CONFIG_H264QPEL)
02952 H264_QPEL_FUNCS(0, 0, sse2);
02953 }
02954 }
02955
02956 if (!high_bit_depth && CONFIG_H264QPEL) {
02957 H264_QPEL_FUNCS(0, 1, sse2);
02958 H264_QPEL_FUNCS(0, 2, sse2);
02959 H264_QPEL_FUNCS(0, 3, sse2);
02960 H264_QPEL_FUNCS(1, 1, sse2);
02961 H264_QPEL_FUNCS(1, 2, sse2);
02962 H264_QPEL_FUNCS(1, 3, sse2);
02963 H264_QPEL_FUNCS(2, 1, sse2);
02964 H264_QPEL_FUNCS(2, 2, sse2);
02965 H264_QPEL_FUNCS(2, 3, sse2);
02966 H264_QPEL_FUNCS(3, 1, sse2);
02967 H264_QPEL_FUNCS(3, 2, sse2);
02968 H264_QPEL_FUNCS(3, 3, sse2);
02969 }
02970
02971 #if HAVE_YASM
02972 if (bit_depth == 10) {
02973 if (CONFIG_H264QPEL) {
02974 SET_QPEL_FUNCS(put_h264_qpel, 0, 16, 10_sse2, ff_);
02975 SET_QPEL_FUNCS(put_h264_qpel, 1, 8, 10_sse2, ff_);
02976 SET_QPEL_FUNCS(avg_h264_qpel, 0, 16, 10_sse2, ff_);
02977 SET_QPEL_FUNCS(avg_h264_qpel, 1, 8, 10_sse2, ff_);
02978 H264_QPEL_FUNCS_10(1, 0, sse2_cache64);
02979 H264_QPEL_FUNCS_10(2, 0, sse2_cache64);
02980 H264_QPEL_FUNCS_10(3, 0, sse2_cache64);
02981 }
02982 if (CONFIG_H264CHROMA) {
02983 c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_10_sse2;
02984 c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_10_sse2;
02985 }
02986 }
02987
02988 c->scalarproduct_int16 = ff_scalarproduct_int16_sse2;
02989 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
02990 if (mm_flags & AV_CPU_FLAG_ATOM) {
02991 c->vector_clip_int32 = ff_vector_clip_int32_int_sse2;
02992 } else {
02993 c->vector_clip_int32 = ff_vector_clip_int32_sse2;
02994 }
02995 if (avctx->flags & CODEC_FLAG_BITEXACT) {
02996 c->apply_window_int16 = ff_apply_window_int16_sse2_ba;
02997 } else if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
02998 c->apply_window_int16 = ff_apply_window_int16_sse2;
02999 }
03000 c->bswap_buf = ff_bswap32_buf_sse2;
03001 #endif
03002 }
03003
03004 static void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx,
03005 int mm_flags)
03006 {
03007 #if HAVE_SSSE3
03008 const int high_bit_depth = avctx->bits_per_raw_sample > 8;
03009 const int bit_depth = avctx->bits_per_raw_sample;
03010
03011 if (!high_bit_depth && CONFIG_H264QPEL) {
03012 H264_QPEL_FUNCS(1, 0, ssse3);
03013 H264_QPEL_FUNCS(1, 1, ssse3);
03014 H264_QPEL_FUNCS(1, 2, ssse3);
03015 H264_QPEL_FUNCS(1, 3, ssse3);
03016 H264_QPEL_FUNCS(2, 0, ssse3);
03017 H264_QPEL_FUNCS(2, 1, ssse3);
03018 H264_QPEL_FUNCS(2, 2, ssse3);
03019 H264_QPEL_FUNCS(2, 3, ssse3);
03020 H264_QPEL_FUNCS(3, 0, ssse3);
03021 H264_QPEL_FUNCS(3, 1, ssse3);
03022 H264_QPEL_FUNCS(3, 2, ssse3);
03023 H264_QPEL_FUNCS(3, 3, ssse3);
03024 }
03025 #if HAVE_YASM
03026 else if (bit_depth == 10 && CONFIG_H264QPEL) {
03027 H264_QPEL_FUNCS_10(1, 0, ssse3_cache64);
03028 H264_QPEL_FUNCS_10(2, 0, ssse3_cache64);
03029 H264_QPEL_FUNCS_10(3, 0, ssse3_cache64);
03030 }
03031 if (!high_bit_depth && CONFIG_H264CHROMA) {
03032 c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_ssse3_rnd;
03033 c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_ssse3_rnd;
03034 c->put_h264_chroma_pixels_tab[1] = ff_put_h264_chroma_mc4_ssse3;
03035 c->avg_h264_chroma_pixels_tab[1] = ff_avg_h264_chroma_mc4_ssse3;
03036 }
03037 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
03038 if (mm_flags & AV_CPU_FLAG_SSE4)
03039 c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
03040
03041 if (mm_flags & AV_CPU_FLAG_ATOM)
03042 c->apply_window_int16 = ff_apply_window_int16_ssse3_atom;
03043 else
03044 c->apply_window_int16 = ff_apply_window_int16_ssse3;
03045 if (!(mm_flags & (AV_CPU_FLAG_SSE42|AV_CPU_FLAG_3DNOW)))
03046 c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
03047 c->bswap_buf = ff_bswap32_buf_ssse3;
03048 #endif
03049 #endif
03050 }
03051
03052 static void dsputil_init_sse4(DSPContext *c, AVCodecContext *avctx,
03053 int mm_flags)
03054 {
03055 #if HAVE_YASM
03056 c->vector_clip_int32 = ff_vector_clip_int32_sse4;
03057 #endif
03058 }
03059
03060 static void dsputil_init_avx(DSPContext *c, AVCodecContext *avctx, int mm_flags)
03061 {
03062 #if HAVE_AVX && HAVE_YASM
03063 const int bit_depth = avctx->bits_per_raw_sample;
03064
03065 if (bit_depth == 10) {
03066
03067
03068 if (CONFIG_H264QPEL) {
03069 H264_QPEL_FUNCS_10(1, 0, sse2);
03070 H264_QPEL_FUNCS_10(2, 0, sse2);
03071 H264_QPEL_FUNCS_10(3, 0, sse2);
03072 }
03073
03074 if (CONFIG_H264CHROMA) {
03075 c->put_h264_chroma_pixels_tab[0] = ff_put_h264_chroma_mc8_10_avx;
03076 c->avg_h264_chroma_pixels_tab[0] = ff_avg_h264_chroma_mc8_10_avx;
03077 }
03078 }
03079 c->butterflies_float_interleave = ff_butterflies_float_interleave_avx;
03080 c->vector_fmul = ff_vector_fmul_avx;
03081 c->vector_fmul_reverse = ff_vector_fmul_reverse_avx;
03082 c->vector_fmul_add = ff_vector_fmul_add_avx;
03083 #endif
03084 }
03085
03086 void ff_dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx)
03087 {
03088 int mm_flags = av_get_cpu_flags();
03089
03090 #if 0
03091 av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:");
03092 if (mm_flags & AV_CPU_FLAG_MMX)
03093 av_log(avctx, AV_LOG_INFO, " mmx");
03094 if (mm_flags & AV_CPU_FLAG_MMX2)
03095 av_log(avctx, AV_LOG_INFO, " mmx2");
03096 if (mm_flags & AV_CPU_FLAG_3DNOW)
03097 av_log(avctx, AV_LOG_INFO, " 3dnow");
03098 if (mm_flags & AV_CPU_FLAG_SSE)
03099 av_log(avctx, AV_LOG_INFO, " sse");
03100 if (mm_flags & AV_CPU_FLAG_SSE2)
03101 av_log(avctx, AV_LOG_INFO, " sse2");
03102 av_log(avctx, AV_LOG_INFO, "\n");
03103 #endif
03104
03105 if (mm_flags & AV_CPU_FLAG_MMX) {
03106 const int idct_algo = avctx->idct_algo;
03107
03108 if (avctx->lowres == 0 && avctx->bits_per_raw_sample <= 8) {
03109 if (idct_algo == FF_IDCT_AUTO || idct_algo == FF_IDCT_SIMPLEMMX) {
03110 c->idct_put = ff_simple_idct_put_mmx;
03111 c->idct_add = ff_simple_idct_add_mmx;
03112 c->idct = ff_simple_idct_mmx;
03113 c->idct_permutation_type = FF_SIMPLE_IDCT_PERM;
03114 #if CONFIG_GPL
03115 } else if (idct_algo == FF_IDCT_LIBMPEG2MMX) {
03116 if (mm_flags & AV_CPU_FLAG_MMX2) {
03117 c->idct_put = ff_libmpeg2mmx2_idct_put;
03118 c->idct_add = ff_libmpeg2mmx2_idct_add;
03119 c->idct = ff_mmxext_idct;
03120 } else {
03121 c->idct_put = ff_libmpeg2mmx_idct_put;
03122 c->idct_add = ff_libmpeg2mmx_idct_add;
03123 c->idct = ff_mmx_idct;
03124 }
03125 c->idct_permutation_type = FF_LIBMPEG2_IDCT_PERM;
03126 #endif
03127 } else if ((CONFIG_VP3_DECODER || CONFIG_VP5_DECODER ||
03128 CONFIG_VP6_DECODER) &&
03129 idct_algo == FF_IDCT_VP3 && HAVE_YASM) {
03130 if (mm_flags & AV_CPU_FLAG_SSE2) {
03131 c->idct_put = ff_vp3_idct_put_sse2;
03132 c->idct_add = ff_vp3_idct_add_sse2;
03133 c->idct = ff_vp3_idct_sse2;
03134 c->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
03135 } else {
03136 c->idct_put = ff_vp3_idct_put_mmx;
03137 c->idct_add = ff_vp3_idct_add_mmx;
03138 c->idct = ff_vp3_idct_mmx;
03139 c->idct_permutation_type = FF_PARTTRANS_IDCT_PERM;
03140 }
03141 } else if (idct_algo == FF_IDCT_CAVS) {
03142 c->idct_permutation_type = FF_TRANSPOSE_IDCT_PERM;
03143 } else if (idct_algo == FF_IDCT_XVIDMMX) {
03144 if (mm_flags & AV_CPU_FLAG_SSE2) {
03145 c->idct_put = ff_idct_xvid_sse2_put;
03146 c->idct_add = ff_idct_xvid_sse2_add;
03147 c->idct = ff_idct_xvid_sse2;
03148 c->idct_permutation_type = FF_SSE2_IDCT_PERM;
03149 } else if (mm_flags & AV_CPU_FLAG_MMX2) {
03150 c->idct_put = ff_idct_xvid_mmx2_put;
03151 c->idct_add = ff_idct_xvid_mmx2_add;
03152 c->idct = ff_idct_xvid_mmx2;
03153 } else {
03154 c->idct_put = ff_idct_xvid_mmx_put;
03155 c->idct_add = ff_idct_xvid_mmx_add;
03156 c->idct = ff_idct_xvid_mmx;
03157 }
03158 }
03159 }
03160
03161 dsputil_init_mmx(c, avctx, mm_flags);
03162 }
03163
03164 if (mm_flags & AV_CPU_FLAG_MMX2)
03165 dsputil_init_mmx2(c, avctx, mm_flags);
03166
03167 if (mm_flags & AV_CPU_FLAG_3DNOW && HAVE_AMD3DNOW)
03168 dsputil_init_3dnow(c, avctx, mm_flags);
03169
03170 if (mm_flags & AV_CPU_FLAG_3DNOWEXT && HAVE_AMD3DNOWEXT)
03171 dsputil_init_3dnow2(c, avctx, mm_flags);
03172
03173 if (mm_flags & AV_CPU_FLAG_SSE && HAVE_SSE)
03174 dsputil_init_sse(c, avctx, mm_flags);
03175
03176 if (mm_flags & AV_CPU_FLAG_SSE2)
03177 dsputil_init_sse2(c, avctx, mm_flags);
03178
03179 if (mm_flags & AV_CPU_FLAG_SSSE3)
03180 dsputil_init_ssse3(c, avctx, mm_flags);
03181
03182 if (mm_flags & AV_CPU_FLAG_SSE4 && HAVE_SSE)
03183 dsputil_init_sse4(c, avctx, mm_flags);
03184
03185 if (mm_flags & AV_CPU_FLAG_AVX)
03186 dsputil_init_avx(c, avctx, mm_flags);
03187
03188 if (CONFIG_ENCODERS)
03189 ff_dsputilenc_init_mmx(c, avctx);
03190 }
