X-Git-Url: https://git.piment-noir.org/?p=deb_ffmpeg.git;a=blobdiff_plain;f=ffmpeg%2Flibavcodec%2Fx86%2Fme_cmp_init.c;fp=ffmpeg%2Flibavcodec%2Fx86%2Fme_cmp_init.c;h=255df5065d6dcebfb60c7de11c3c2e8136b91c40;hp=21db2214c6fe0151d922f78aa753990912077872;hb=f6fa7814ccfe3e76514b36cf04f5cd3cb657c8cf;hpb=2ba45a602cbfa7b771effba9b11bb4245c21bc00 diff --git a/ffmpeg/libavcodec/x86/me_cmp_init.c b/ffmpeg/libavcodec/x86/me_cmp_init.c index 21db221..255df50 100644 --- a/ffmpeg/libavcodec/x86/me_cmp_init.c +++ b/ffmpeg/libavcodec/x86/me_cmp_init.c @@ -34,19 +34,55 @@ int ff_sum_abs_dctelem_mmxext(int16_t *block); int ff_sum_abs_dctelem_sse2(int16_t *block); int ff_sum_abs_dctelem_ssse3(int16_t *block); int ff_sse8_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, - int line_size, int h); + ptrdiff_t stride, int h); int ff_sse16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, - int line_size, int h); + ptrdiff_t stride, int h); int ff_sse16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, - int line_size, int h); -int ff_hf_noise8_mmx(uint8_t *pix1, int lsize, int h); -int ff_hf_noise16_mmx(uint8_t *pix1, int lsize, int h); - -#define hadamard_func(cpu) \ - int ff_hadamard8_diff_ ## cpu(MpegEncContext *s, uint8_t *src1, \ - uint8_t *src2, int stride, int h); \ - int ff_hadamard8_diff16_ ## cpu(MpegEncContext *s, uint8_t *src1, \ - uint8_t *src2, int stride, int h); + ptrdiff_t stride, int h); +int ff_hf_noise8_mmx(uint8_t *pix1, ptrdiff_t stride, int h); +int ff_hf_noise16_mmx(uint8_t *pix1, ptrdiff_t stride, int h); +int ff_sad8_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad16_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad8_x2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad16_x2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad16_x2_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad8_y2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad16_y2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad16_y2_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad8_approx_xy2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad16_approx_xy2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_sad16_approx_xy2_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_vsad_intra8_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_vsad_intra16_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_vsad_intra16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_vsad8_approx_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_vsad16_approx_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); +int ff_vsad16_approx_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, + ptrdiff_t stride, int h); + +#define hadamard_func(cpu) \ + int ff_hadamard8_diff_ ## cpu(MpegEncContext *s, uint8_t *src1, \ + uint8_t *src2, ptrdiff_t stride, int h); \ + int ff_hadamard8_diff16_ ## cpu(MpegEncContext *s, uint8_t *src1, \ + uint8_t *src2, ptrdiff_t stride, int h); hadamard_func(mmx) hadamard_func(mmxext) @@ -55,16 +91,16 @@ hadamard_func(ssse3) #if HAVE_YASM static int nsse16_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2, - int line_size, int h) + ptrdiff_t stride, int h) { int score1, score2; if (c) - score1 = c->mecc.sse[0](c, pix1, pix2, line_size, h); + score1 = c->mecc.sse[0](c, pix1, pix2, stride, h); else - score1 = ff_sse16_mmx(c, pix1, pix2, line_size, h); - score2 = ff_hf_noise16_mmx(pix1, line_size, h) + ff_hf_noise8_mmx(pix1+8, line_size, h) - - ff_hf_noise16_mmx(pix2, line_size, h) - ff_hf_noise8_mmx(pix2+8, line_size, h); + score1 = ff_sse16_mmx(c, pix1, pix2, stride, h); + score2 = ff_hf_noise16_mmx(pix1, stride, h) + ff_hf_noise8_mmx(pix1+8, stride, h) + - ff_hf_noise16_mmx(pix2, stride, h) - ff_hf_noise8_mmx(pix2+8, stride, h); if (c) return score1 + FFABS(score2) * c->avctx->nsse_weight; @@ -73,11 +109,11 @@ static int nsse16_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2, } static int nsse8_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2, - int line_size, int h) + ptrdiff_t stride, int h) { - int score1 = ff_sse8_mmx(c, pix1, pix2, line_size, h); - int score2 = ff_hf_noise8_mmx(pix1, line_size, h) - - ff_hf_noise8_mmx(pix2, line_size, h); + int score1 = ff_sse8_mmx(c, pix1, pix2, stride, h); + int score2 = ff_hf_noise8_mmx(pix1, stride, h) - + ff_hf_noise8_mmx(pix2, stride, h); if (c) return score1 + FFABS(score2) * c->avctx->nsse_weight; @@ -90,12 +126,12 @@ static int nsse8_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2, #if HAVE_INLINE_ASM static int vsad_intra16_mmx(MpegEncContext *v, uint8_t *pix, uint8_t *dummy, - int line_size, int h) + ptrdiff_t stride, int h) { int tmp; av_assert2((((int) pix) & 7) == 0); - av_assert2((line_size & 7) == 0); + av_assert2((stride & 7) == 0); #define SUM(in0, in1, out0, out1) \ "movq (%0), %%mm2\n" \ @@ -146,64 +182,21 @@ static int vsad_intra16_mmx(MpegEncContext *v, uint8_t *pix, uint8_t *dummy, "paddw %%mm6, %%mm0\n" "movd %%mm0, %1\n" : "+r" (pix), "=r" (tmp) - : "r" ((x86_reg) line_size), "m" (h) + : "r" (stride), "m" (h) : "%ecx"); return tmp & 0xFFFF; } #undef SUM -static int vsad_intra16_mmxext(MpegEncContext *v, uint8_t *pix, uint8_t *dummy, - int line_size, int h) -{ - int tmp; - - av_assert2((((int) pix) & 7) == 0); - av_assert2((line_size & 7) == 0); - -#define SUM(in0, in1, out0, out1) \ - "movq (%0), " #out0 "\n" \ - "movq 8(%0), " #out1 "\n" \ - "add %2, %0\n" \ - "psadbw " #out0 ", " #in0 "\n" \ - "psadbw " #out1 ", " #in1 "\n" \ - "paddw " #in1 ", " #in0 "\n" \ - "paddw " #in0 ", %%mm6\n" - - __asm__ volatile ( - "movl %3, %%ecx\n" - "pxor %%mm6, %%mm6\n" - "pxor %%mm7, %%mm7\n" - "movq (%0), %%mm0\n" - "movq 8(%0), %%mm1\n" - "add %2, %0\n" - "jmp 2f\n" - "1:\n" - - SUM(%%mm4, %%mm5, %%mm0, %%mm1) - "2:\n" - SUM(%%mm0, %%mm1, %%mm4, %%mm5) - - "subl $2, %%ecx\n" - "jnz 1b\n" - - "movd %%mm6, %1\n" - : "+r" (pix), "=r" (tmp) - : "r" ((x86_reg) line_size), "m" (h) - : "%ecx"); - - return tmp; -} -#undef SUM - static int vsad16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, - int line_size, int h) + ptrdiff_t stride, int h) { int tmp; av_assert2((((int) pix1) & 7) == 0); av_assert2((((int) pix2) & 7) == 0); - av_assert2((line_size & 7) == 0); + av_assert2((stride & 7) == 0); #define SUM(in0, in1, out0, out1) \ "movq (%0), %%mm2\n" \ @@ -270,86 +263,23 @@ static int vsad16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, "paddw %%mm6, %%mm0\n" "movd %%mm0, %2\n" : "+r" (pix1), "+r" (pix2), "=r" (tmp) - : "r" ((x86_reg) line_size), "m" (h) + : "r" (stride), "m" (h) : "%ecx"); return tmp & 0x7FFF; } #undef SUM -static int vsad16_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2, - int line_size, int h) -{ - int tmp; - - av_assert2((((int) pix1) & 7) == 0); - av_assert2((((int) pix2) & 7) == 0); - av_assert2((line_size & 7) == 0); - -#define SUM(in0, in1, out0, out1) \ - "movq (%0), " #out0 "\n" \ - "movq (%1), %%mm2\n" \ - "movq 8(%0), " #out1 "\n" \ - "movq 8(%1), %%mm3\n" \ - "add %3, %0\n" \ - "add %3, %1\n" \ - "psubb %%mm2, " #out0 "\n" \ - "psubb %%mm3, " #out1 "\n" \ - "pxor %%mm7, " #out0 "\n" \ - "pxor %%mm7, " #out1 "\n" \ - "psadbw " #out0 ", " #in0 "\n" \ - "psadbw " #out1 ", " #in1 "\n" \ - "paddw " #in1 ", " #in0 "\n" \ - "paddw " #in0 ", %%mm6\n " - - __asm__ volatile ( - "movl %4, %%ecx\n" - "pxor %%mm6, %%mm6\n" - "pcmpeqw %%mm7, %%mm7\n" - "psllw $15, %%mm7\n" - "packsswb %%mm7, %%mm7\n" - "movq (%0), %%mm0\n" - "movq (%1), %%mm2\n" - "movq 8(%0), %%mm1\n" - "movq 8(%1), %%mm3\n" - "add %3, %0\n" - "add %3, %1\n" - "psubb %%mm2, %%mm0\n" - "psubb %%mm3, %%mm1\n" - "pxor %%mm7, %%mm0\n" - "pxor %%mm7, %%mm1\n" - "jmp 2f\n" - "1:\n" - - SUM(%%mm4, %%mm5, %%mm0, %%mm1) - "2:\n" - SUM(%%mm0, %%mm1, %%mm4, %%mm5) - - "subl $2, %%ecx\n" - "jnz 1b\n" - - "movd %%mm6, %2\n" - : "+r" (pix1), "+r" (pix2), "=r" (tmp) - : "r" ((x86_reg) line_size), "m" (h) - : "%ecx"); - - return tmp; -} -#undef SUM - - - DECLARE_ASM_CONST(8, uint64_t, round_tab)[3] = { 0x0000000000000000ULL, 0x0001000100010001ULL, 0x0002000200020002ULL, }; -DECLARE_ASM_CONST(8, uint64_t, bone) = 0x0101010101010101LL; - -static inline void sad8_1_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h) +static inline void sad8_1_mmx(uint8_t *blk1, uint8_t *blk2, + ptrdiff_t stride, int h) { - x86_reg len = -(x86_reg)stride * h; + x86_reg len = -stride * h; __asm__ volatile ( ".p2align 4 \n\t" "1: \n\t" @@ -379,137 +309,13 @@ static inline void sad8_1_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h) "add %3, %%"REG_a" \n\t" " js 1b \n\t" : "+a" (len) - : "r" (blk1 - len), "r" (blk2 - len), "r" ((x86_reg) stride)); -} - -static inline void sad8_1_mmxext(uint8_t *blk1, uint8_t *blk2, - int stride, int h) -{ - __asm__ volatile ( - ".p2align 4 \n\t" - "1: \n\t" - "movq (%1), %%mm0 \n\t" - "movq (%1, %3), %%mm1 \n\t" - "psadbw (%2), %%mm0 \n\t" - "psadbw (%2, %3), %%mm1 \n\t" - "paddw %%mm0, %%mm6 \n\t" - "paddw %%mm1, %%mm6 \n\t" - "lea (%1,%3,2), %1 \n\t" - "lea (%2,%3,2), %2 \n\t" - "sub $2, %0 \n\t" - " jg 1b \n\t" - : "+r" (h), "+r" (blk1), "+r" (blk2) - : "r" ((x86_reg) stride)); -} - -static int sad16_sse2(MpegEncContext *v, uint8_t *blk2, uint8_t *blk1, - int stride, int h) -{ - int ret; - __asm__ volatile ( - "pxor %%xmm2, %%xmm2 \n\t" - ".p2align 4 \n\t" - "1: \n\t" - "movdqu (%1), %%xmm0 \n\t" - "movdqu (%1, %4), %%xmm1 \n\t" - "psadbw (%2), %%xmm0 \n\t" - "psadbw (%2, %4), %%xmm1 \n\t" - "paddw %%xmm0, %%xmm2 \n\t" - "paddw %%xmm1, %%xmm2 \n\t" - "lea (%1,%4,2), %1 \n\t" - "lea (%2,%4,2), %2 \n\t" - "sub $2, %0 \n\t" - " jg 1b \n\t" - "movhlps %%xmm2, %%xmm0 \n\t" - "paddw %%xmm0, %%xmm2 \n\t" - "movd %%xmm2, %3 \n\t" - : "+r" (h), "+r" (blk1), "+r" (blk2), "=r" (ret) - : "r" ((x86_reg) stride)); - return ret; -} - -static inline void sad8_x2a_mmxext(uint8_t *blk1, uint8_t *blk2, - int stride, int h) -{ - __asm__ volatile ( - ".p2align 4 \n\t" - "1: \n\t" - "movq (%1), %%mm0 \n\t" - "movq (%1, %3), %%mm1 \n\t" - "pavgb 1(%1), %%mm0 \n\t" - "pavgb 1(%1, %3), %%mm1 \n\t" - "psadbw (%2), %%mm0 \n\t" - "psadbw (%2, %3), %%mm1 \n\t" - "paddw %%mm0, %%mm6 \n\t" - "paddw %%mm1, %%mm6 \n\t" - "lea (%1,%3,2), %1 \n\t" - "lea (%2,%3,2), %2 \n\t" - "sub $2, %0 \n\t" - " jg 1b \n\t" - : "+r" (h), "+r" (blk1), "+r" (blk2) - : "r" ((x86_reg) stride)); -} - -static inline void sad8_y2a_mmxext(uint8_t *blk1, uint8_t *blk2, - int stride, int h) -{ - __asm__ volatile ( - "movq (%1), %%mm0 \n\t" - "add %3, %1 \n\t" - ".p2align 4 \n\t" - "1: \n\t" - "movq (%1), %%mm1 \n\t" - "movq (%1, %3), %%mm2 \n\t" - "pavgb %%mm1, %%mm0 \n\t" - "pavgb %%mm2, %%mm1 \n\t" - "psadbw (%2), %%mm0 \n\t" - "psadbw (%2, %3), %%mm1 \n\t" - "paddw %%mm0, %%mm6 \n\t" - "paddw %%mm1, %%mm6 \n\t" - "movq %%mm2, %%mm0 \n\t" - "lea (%1,%3,2), %1 \n\t" - "lea (%2,%3,2), %2 \n\t" - "sub $2, %0 \n\t" - " jg 1b \n\t" - : "+r" (h), "+r" (blk1), "+r" (blk2) - : "r" ((x86_reg) stride)); -} - -static inline void sad8_4_mmxext(uint8_t *blk1, uint8_t *blk2, - int stride, int h) -{ - __asm__ volatile ( - "movq "MANGLE(bone)", %%mm5 \n\t" - "movq (%1), %%mm0 \n\t" - "pavgb 1(%1), %%mm0 \n\t" - "add %3, %1 \n\t" - ".p2align 4 \n\t" - "1: \n\t" - "movq (%1), %%mm1 \n\t" - "movq (%1,%3), %%mm2 \n\t" - "pavgb 1(%1), %%mm1 \n\t" - "pavgb 1(%1,%3), %%mm2 \n\t" - "psubusb %%mm5, %%mm1 \n\t" - "pavgb %%mm1, %%mm0 \n\t" - "pavgb %%mm2, %%mm1 \n\t" - "psadbw (%2), %%mm0 \n\t" - "psadbw (%2,%3), %%mm1 \n\t" - "paddw %%mm0, %%mm6 \n\t" - "paddw %%mm1, %%mm6 \n\t" - "movq %%mm2, %%mm0 \n\t" - "lea (%1,%3,2), %1 \n\t" - "lea (%2,%3,2), %2 \n\t" - "sub $2, %0 \n\t" - " jg 1b \n\t" - : "+r" (h), "+r" (blk1), "+r" (blk2) - : "r" ((x86_reg) stride) - NAMED_CONSTRAINTS_ADD(bone)); + : "r" (blk1 - len), "r" (blk2 - len), "r" (stride)); } static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, - int stride, int h) + ptrdiff_t stride, int h) { - x86_reg len = -(x86_reg)stride * h; + x86_reg len = -stride * h; __asm__ volatile ( ".p2align 4 \n\t" "1: \n\t" @@ -542,12 +348,13 @@ static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2, " js 1b \n\t" : "+a" (len) : "r" (blk1a - len), "r" (blk1b - len), "r" (blk2 - len), - "r" ((x86_reg) stride)); + "r" (stride)); } -static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h) +static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, + ptrdiff_t stride, int h) { - x86_reg len = -(x86_reg)stride * h; + x86_reg len = -stride * h; __asm__ volatile ( "movq (%1, %%"REG_a"), %%mm0 \n\t" "movq 1(%1, %%"REG_a"), %%mm2 \n\t" @@ -595,7 +402,7 @@ static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h) " js 1b \n\t" : "+a" (len) : "r" (blk1 - len), "r" (blk1 - len + stride), "r" (blk2 - len), - "r" ((x86_reg) stride), "m" (round_tab[2])); + "r" (stride), "m" (round_tab[2])); } static inline int sum_mmx(void) @@ -613,28 +420,21 @@ static inline int sum_mmx(void) return ret & 0xFFFF; } -static inline int sum_mmxext(void) -{ - int ret; - __asm__ volatile ( - "movd %%mm6, %0 \n\t" - : "=r" (ret)); - return ret; -} - -static inline void sad8_x2a_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h) +static inline void sad8_x2a_mmx(uint8_t *blk1, uint8_t *blk2, + ptrdiff_t stride, int h) { sad8_2_mmx(blk1, blk1 + 1, blk2, stride, h); } -static inline void sad8_y2a_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h) +static inline void sad8_y2a_mmx(uint8_t *blk1, uint8_t *blk2, + ptrdiff_t stride, int h) { sad8_2_mmx(blk1, blk1 + stride, blk2, stride, h); } #define PIX_SAD(suf) \ static int sad8_ ## suf(MpegEncContext *v, uint8_t *blk2, \ - uint8_t *blk1, int stride, int h) \ + uint8_t *blk1, ptrdiff_t stride, int h) \ { \ av_assert2(h == 8); \ __asm__ volatile ( \ @@ -648,7 +448,7 @@ static int sad8_ ## suf(MpegEncContext *v, uint8_t *blk2, \ } \ \ static int sad8_x2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ - uint8_t *blk1, int stride, int h) \ + uint8_t *blk1, ptrdiff_t stride, int h) \ { \ av_assert2(h == 8); \ __asm__ volatile ( \ @@ -663,7 +463,7 @@ static int sad8_x2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ } \ \ static int sad8_y2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ - uint8_t *blk1, int stride, int h) \ + uint8_t *blk1, ptrdiff_t stride, int h) \ { \ av_assert2(h == 8); \ __asm__ volatile ( \ @@ -678,7 +478,7 @@ static int sad8_y2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ } \ \ static int sad8_xy2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ - uint8_t *blk1, int stride, int h) \ + uint8_t *blk1, ptrdiff_t stride, int h) \ { \ av_assert2(h == 8); \ __asm__ volatile ( \ @@ -692,7 +492,7 @@ static int sad8_xy2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ } \ \ static int sad16_ ## suf(MpegEncContext *v, uint8_t *blk2, \ - uint8_t *blk1, int stride, int h) \ + uint8_t *blk1, ptrdiff_t stride, int h) \ { \ __asm__ volatile ( \ "pxor %%mm7, %%mm7 \n\t" \ @@ -706,7 +506,7 @@ static int sad16_ ## suf(MpegEncContext *v, uint8_t *blk2, \ } \ \ static int sad16_x2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ - uint8_t *blk1, int stride, int h) \ + uint8_t *blk1, ptrdiff_t stride, int h) \ { \ __asm__ volatile ( \ "pxor %%mm7, %%mm7 \n\t" \ @@ -721,7 +521,7 @@ static int sad16_x2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ } \ \ static int sad16_y2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ - uint8_t *blk1, int stride, int h) \ + uint8_t *blk1, ptrdiff_t stride, int h) \ { \ __asm__ volatile ( \ "pxor %%mm7, %%mm7 \n\t" \ @@ -736,7 +536,7 @@ static int sad16_y2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ } \ \ static int sad16_xy2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ - uint8_t *blk1, int stride, int h) \ + uint8_t *blk1, ptrdiff_t stride, int h) \ { \ __asm__ volatile ( \ "pxor %%mm7, %%mm7 \n\t" \ @@ -750,7 +550,6 @@ static int sad16_xy2_ ## suf(MpegEncContext *v, uint8_t *blk2, \ } \ PIX_SAD(mmx) -PIX_SAD(mmxext) #endif /* HAVE_INLINE_ASM */ @@ -779,32 +578,6 @@ av_cold void ff_me_cmp_init_x86(MECmpContext *c, AVCodecContext *avctx) } } - if (INLINE_MMXEXT(cpu_flags)) { - c->vsad[4] = vsad_intra16_mmxext; - - c->pix_abs[0][0] = sad16_mmxext; - c->pix_abs[1][0] = sad8_mmxext; - - c->sad[0] = sad16_mmxext; - c->sad[1] = sad8_mmxext; - - c->pix_abs[0][1] = sad16_x2_mmxext; - c->pix_abs[0][2] = sad16_y2_mmxext; - c->pix_abs[1][1] = sad8_x2_mmxext; - c->pix_abs[1][2] = sad8_y2_mmxext; - - if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { - c->pix_abs[0][3] = sad16_xy2_mmxext; - c->pix_abs[1][3] = sad8_xy2_mmxext; - - c->vsad[0] = vsad16_mmxext; - } - } - - if (INLINE_SSE2(cpu_flags) && !(cpu_flags & AV_CPU_FLAG_SSE2SLOW) && avctx->codec_id != AV_CODEC_ID_SNOW) { - c->sad[0] = sad16_sse2; - } - #endif /* HAVE_INLINE_ASM */ if (EXTERNAL_MMX(cpu_flags)) { @@ -823,6 +596,27 @@ av_cold void ff_me_cmp_init_x86(MECmpContext *c, AVCodecContext *avctx) c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext; c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext; c->sum_abs_dctelem = ff_sum_abs_dctelem_mmxext; + + c->sad[0] = ff_sad16_mmxext; + c->sad[1] = ff_sad8_mmxext; + + c->pix_abs[0][0] = ff_sad16_mmxext; + c->pix_abs[0][1] = ff_sad16_x2_mmxext; + c->pix_abs[0][2] = ff_sad16_y2_mmxext; + c->pix_abs[1][0] = ff_sad8_mmxext; + c->pix_abs[1][1] = ff_sad8_x2_mmxext; + c->pix_abs[1][2] = ff_sad8_y2_mmxext; + + c->vsad[4] = ff_vsad_intra16_mmxext; + c->vsad[5] = ff_vsad_intra8_mmxext; + + if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { + c->pix_abs[0][3] = ff_sad16_approx_xy2_mmxext; + c->pix_abs[1][3] = ff_sad8_approx_xy2_mmxext; + + c->vsad[0] = ff_vsad16_approx_mmxext; + c->vsad[1] = ff_vsad8_approx_mmxext; + } } if (EXTERNAL_SSE2(cpu_flags)) { @@ -833,6 +627,18 @@ av_cold void ff_me_cmp_init_x86(MECmpContext *c, AVCodecContext *avctx) c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2; c->hadamard8_diff[1] = ff_hadamard8_diff_sse2; #endif + if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW) && avctx->codec_id != AV_CODEC_ID_SNOW) { + c->sad[0] = ff_sad16_sse2; + c->pix_abs[0][0] = ff_sad16_sse2; + c->pix_abs[0][1] = ff_sad16_x2_sse2; + c->pix_abs[0][2] = ff_sad16_y2_sse2; + + c->vsad[4] = ff_vsad_intra16_sse2; + if (!(avctx->flags & CODEC_FLAG_BITEXACT)) { + c->pix_abs[0][3] = ff_sad16_approx_xy2_sse2; + c->vsad[0] = ff_vsad16_approx_sse2; + } + } } if (EXTERNAL_SSSE3(cpu_flags)) {