/* * MMX optimized DSP utils * Copyright (c) 2000, 2001 Fabrice Bellard. * Copyright (c) 2002-2004 Michael Niedermayer * * This library is free software; you can redistribute it and/or * modify it under the terms of the GNU Lesser General Public * License as published by the Free Software Foundation; either * version 2 of the License, or (at your option) any later version. * * This library is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * Lesser General Public License for more details. * * You should have received a copy of the GNU Lesser General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA * * MMX optimization by Nick Kurshev */ #include "../dsputil.h" #include "../simple_idct.h" #include "../mpegvideo.h" #include "x86_cpu.h" #include "mmx.h" //#undef NDEBUG //#include extern const uint8_t ff_h263_loop_filter_strength[32]; extern void ff_idct_xvid_mmx(short *block); extern void ff_idct_xvid_mmx2(short *block); int mm_flags; /* multimedia extension flags */ /* pixel operations */ static const uint64_t mm_bone attribute_used __attribute__ ((aligned(8))) = 0x0101010101010101ULL; static const uint64_t mm_wone attribute_used __attribute__ ((aligned(8))) = 0x0001000100010001ULL; static const uint64_t mm_wtwo attribute_used __attribute__ ((aligned(8))) = 0x0002000200020002ULL; static const uint64_t ff_pdw_80000000[2] attribute_used __attribute__ ((aligned(16))) = {0x8000000080000000ULL, 0x8000000080000000ULL}; static const uint64_t ff_pw_20 attribute_used __attribute__ ((aligned(8))) = 0x0014001400140014ULL; static const uint64_t ff_pw_3 attribute_used __attribute__ ((aligned(8))) = 0x0003000300030003ULL; static const uint64_t ff_pw_4 attribute_used __attribute__ ((aligned(8))) = 0x0004000400040004ULL; static const uint64_t ff_pw_5 attribute_used __attribute__ ((aligned(8))) = 0x0005000500050005ULL; static const uint64_t ff_pw_8 attribute_used __attribute__ ((aligned(8))) = 0x0008000800080008ULL; static const uint64_t ff_pw_16 attribute_used __attribute__ ((aligned(8))) = 0x0010001000100010ULL; static const uint64_t ff_pw_32 attribute_used __attribute__ ((aligned(8))) = 0x0020002000200020ULL; static const uint64_t ff_pw_64 attribute_used __attribute__ ((aligned(8))) = 0x0040004000400040ULL; static const uint64_t ff_pw_15 attribute_used __attribute__ ((aligned(8))) = 0x000F000F000F000FULL; static const uint64_t ff_pb_3F attribute_used __attribute__ ((aligned(8))) = 0x3F3F3F3F3F3F3F3FULL; static const uint64_t ff_pb_FC attribute_used __attribute__ ((aligned(8))) = 0xFCFCFCFCFCFCFCFCULL; #define JUMPALIGN() __asm __volatile (".balign 8"::) #define MOVQ_ZERO(regd) __asm __volatile ("pxor %%" #regd ", %%" #regd ::) #define MOVQ_WONE(regd) \ __asm __volatile ( \ "pcmpeqd %%" #regd ", %%" #regd " \n\t" \ "psrlw $15, %%" #regd ::) #define MOVQ_BFE(regd) \ __asm __volatile ( \ "pcmpeqd %%" #regd ", %%" #regd " \n\t"\ "paddb %%" #regd ", %%" #regd " \n\t" ::) #ifndef PIC #define MOVQ_BONE(regd) __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(mm_bone)) #define MOVQ_WTWO(regd) __asm __volatile ("movq %0, %%" #regd " \n\t" ::"m"(mm_wtwo)) #else // for shared library it's better to use this way for accessing constants // pcmpeqd -> -1 #define MOVQ_BONE(regd) \ __asm __volatile ( \ "pcmpeqd %%" #regd ", %%" #regd " \n\t" \ "psrlw $15, %%" #regd " \n\t" \ "packuswb %%" #regd ", %%" #regd " \n\t" ::) #define MOVQ_WTWO(regd) \ __asm __volatile ( \ "pcmpeqd %%" #regd ", %%" #regd " \n\t" \ "psrlw $15, %%" #regd " \n\t" \ "psllw $1, %%" #regd " \n\t"::) #endif // using regr as temporary and for the output result // first argument is unmodifed and second is trashed // regfe is supposed to contain 0xfefefefefefefefe #define PAVGB_MMX_NO_RND(rega, regb, regr, regfe) \ "movq " #rega ", " #regr " \n\t"\ "pand " #regb ", " #regr " \n\t"\ "pxor " #rega ", " #regb " \n\t"\ "pand " #regfe "," #regb " \n\t"\ "psrlq $1, " #regb " \n\t"\ "paddb " #regb ", " #regr " \n\t" #define PAVGB_MMX(rega, regb, regr, regfe) \ "movq " #rega ", " #regr " \n\t"\ "por " #regb ", " #regr " \n\t"\ "pxor " #rega ", " #regb " \n\t"\ "pand " #regfe "," #regb " \n\t"\ "psrlq $1, " #regb " \n\t"\ "psubb " #regb ", " #regr " \n\t" // mm6 is supposed to contain 0xfefefefefefefefe #define PAVGBP_MMX_NO_RND(rega, regb, regr, regc, regd, regp) \ "movq " #rega ", " #regr " \n\t"\ "movq " #regc ", " #regp " \n\t"\ "pand " #regb ", " #regr " \n\t"\ "pand " #regd ", " #regp " \n\t"\ "pxor " #rega ", " #regb " \n\t"\ "pxor " #regc ", " #regd " \n\t"\ "pand %%mm6, " #regb " \n\t"\ "pand %%mm6, " #regd " \n\t"\ "psrlq $1, " #regb " \n\t"\ "psrlq $1, " #regd " \n\t"\ "paddb " #regb ", " #regr " \n\t"\ "paddb " #regd ", " #regp " \n\t" #define PAVGBP_MMX(rega, regb, regr, regc, regd, regp) \ "movq " #rega ", " #regr " \n\t"\ "movq " #regc ", " #regp " \n\t"\ "por " #regb ", " #regr " \n\t"\ "por " #regd ", " #regp " \n\t"\ "pxor " #rega ", " #regb " \n\t"\ "pxor " #regc ", " #regd " \n\t"\ "pand %%mm6, " #regb " \n\t"\ "pand %%mm6, " #regd " \n\t"\ "psrlq $1, " #regd " \n\t"\ "psrlq $1, " #regb " \n\t"\ "psubb " #regb ", " #regr " \n\t"\ "psubb " #regd ", " #regp " \n\t" /***********************************/ /* MMX no rounding */ #define DEF(x, y) x ## _no_rnd_ ## y ##_mmx #define SET_RND MOVQ_WONE #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX_NO_RND(a, b, c, d, e, f) #define PAVGB(a, b, c, e) PAVGB_MMX_NO_RND(a, b, c, e) #include "dsputil_mmx_rnd.h" #undef DEF #undef SET_RND #undef PAVGBP #undef PAVGB /***********************************/ /* MMX rounding */ #define DEF(x, y) x ## _ ## y ##_mmx #define SET_RND MOVQ_WTWO #define PAVGBP(a, b, c, d, e, f) PAVGBP_MMX(a, b, c, d, e, f) #define PAVGB(a, b, c, e) PAVGB_MMX(a, b, c, e) #include "dsputil_mmx_rnd.h" #undef DEF #undef SET_RND #undef PAVGBP #undef PAVGB /***********************************/ /* 3Dnow specific */ #define DEF(x) x ## _3dnow /* for Athlons PAVGUSB is prefered */ #define PAVGB "pavgusb" #include "dsputil_mmx_avg.h" #undef DEF #undef PAVGB /***********************************/ /* MMX2 specific */ #define DEF(x) x ## _mmx2 /* Introduced only in MMX2 set */ #define PAVGB "pavgb" #include "dsputil_mmx_avg.h" #undef DEF #undef PAVGB #define SBUTTERFLY(a,b,t,n)\ "movq " #a ", " #t " \n\t" /* abcd */\ "punpckl" #n " " #b ", " #a " \n\t" /* aebf */\ "punpckh" #n " " #b ", " #t " \n\t" /* cgdh */\ /***********************************/ /* standard MMX */ #ifdef CONFIG_ENCODERS static void get_pixels_mmx(DCTELEM *block, const uint8_t *pixels, int line_size) { asm volatile( "mov $-128, %%"REG_a" \n\t" "pxor %%mm7, %%mm7 \n\t" ".balign 16 \n\t" "1: \n\t" "movq (%0), %%mm0 \n\t" "movq (%0, %2), %%mm2 \n\t" "movq %%mm0, %%mm1 \n\t" "movq %%mm2, %%mm3 \n\t" "punpcklbw %%mm7, %%mm0 \n\t" "punpckhbw %%mm7, %%mm1 \n\t" "punpcklbw %%mm7, %%mm2 \n\t" "punpckhbw %%mm7, %%mm3 \n\t" "movq %%mm0, (%1, %%"REG_a") \n\t" "movq %%mm1, 8(%1, %%"REG_a") \n\t" "movq %%mm2, 16(%1, %%"REG_a") \n\t" "movq %%mm3, 24(%1, %%"REG_a") \n\t" "add %3, %0 \n\t" "add $32, %%"REG_a" \n\t" "js 1b \n\t" : "+r" (pixels) : "r" (block+64), "r" ((long)line_size), "r" ((long)line_size*2) : "%"REG_a ); } static inline void diff_pixels_mmx(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride) { asm volatile( "pxor %%mm7, %%mm7 \n\t" "mov $-128, %%"REG_a" \n\t" ".balign 16 \n\t" "1: \n\t" "movq (%0), %%mm0 \n\t" "movq (%1), %%mm2 \n\t" "movq %%mm0, %%mm1 \n\t" "movq %%mm2, %%mm3 \n\t" "punpcklbw %%mm7, %%mm0 \n\t" "punpckhbw %%mm7, %%mm1 \n\t" "punpcklbw %%mm7, %%mm2 \n\t" "punpckhbw %%mm7, %%mm3 \n\t" "psubw %%mm2, %%mm0 \n\t" "psubw %%mm3, %%mm1 \n\t" "movq %%mm0, (%2, %%"REG_a") \n\t" "movq %%mm1, 8(%2, %%"REG_a") \n\t" "add %3, %0 \n\t" "add %3, %1 \n\t" "add $16, %%"REG_a" \n\t" "jnz 1b \n\t" : "+r" (s1), "+r" (s2) : "r" (block+64), "r" ((long)stride) : "%"REG_a ); } #endif //CONFIG_ENCODERS void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) { const DCTELEM *p; uint8_t *pix; /* read the pixels */ p = block; pix = pixels; /* unrolled loop */ __asm __volatile( "movq %3, %%mm0 \n\t" "movq 8%3, %%mm1 \n\t" "movq 16%3, %%mm2 \n\t" "movq 24%3, %%mm3 \n\t" "movq 32%3, %%mm4 \n\t" "movq 40%3, %%mm5 \n\t" "movq 48%3, %%mm6 \n\t" "movq 56%3, %%mm7 \n\t" "packuswb %%mm1, %%mm0 \n\t" "packuswb %%mm3, %%mm2 \n\t" "packuswb %%mm5, %%mm4 \n\t" "packuswb %%mm7, %%mm6 \n\t" "movq %%mm0, (%0) \n\t" "movq %%mm2, (%0, %1) \n\t" "movq %%mm4, (%0, %1, 2) \n\t" "movq %%mm6, (%0, %2) \n\t" ::"r" (pix), "r" ((long)line_size), "r" ((long)line_size*3), "m"(*p) :"memory"); pix += line_size*4; p += 32; // if here would be an exact copy of the code above // compiler would generate some very strange code // thus using "r" __asm __volatile( "movq (%3), %%mm0 \n\t" "movq 8(%3), %%mm1 \n\t" "movq 16(%3), %%mm2 \n\t" "movq 24(%3), %%mm3 \n\t" "movq 32(%3), %%mm4 \n\t" "movq 40(%3), %%mm5 \n\t" "movq 48(%3), %%mm6 \n\t" "movq 56(%3), %%mm7 \n\t" "packuswb %%mm1, %%mm0 \n\t" "packuswb %%mm3, %%mm2 \n\t" "packuswb %%mm5, %%mm4 \n\t" "packuswb %%mm7, %%mm6 \n\t" "movq %%mm0, (%0) \n\t" "movq %%mm2, (%0, %1) \n\t" "movq %%mm4, (%0, %1, 2) \n\t" "movq %%mm6, (%0, %2) \n\t" ::"r" (pix), "r" ((long)line_size), "r" ((long)line_size*3), "r"(p) :"memory"); } static DECLARE_ALIGNED_8(const unsigned char, vector128[8]) = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 }; void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) { int i; movq_m2r(*vector128, mm1); for (i = 0; i < 8; i++) { movq_m2r(*(block), mm0); packsswb_m2r(*(block + 4), mm0); block += 8; paddb_r2r(mm1, mm0); movq_r2m(mm0, *pixels); pixels += line_size; } } void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size) { const DCTELEM *p; uint8_t *pix; int i; /* read the pixels */ p = block; pix = pixels; MOVQ_ZERO(mm7); i = 4; do { __asm __volatile( "movq (%2), %%mm0 \n\t" "movq 8(%2), %%mm1 \n\t" "movq 16(%2), %%mm2 \n\t" "movq 24(%2), %%mm3 \n\t" "movq %0, %%mm4 \n\t" "movq %1, %%mm6 \n\t" "movq %%mm4, %%mm5 \n\t" "punpcklbw %%mm7, %%mm4 \n\t" "punpckhbw %%mm7, %%mm5 \n\t" "paddsw %%mm4, %%mm0 \n\t" "paddsw %%mm5, %%mm1 \n\t" "movq %%mm6, %%mm5 \n\t" "punpcklbw %%mm7, %%mm6 \n\t" "punpckhbw %%mm7, %%mm5 \n\t" "paddsw %%mm6, %%mm2 \n\t" "paddsw %%mm5, %%mm3 \n\t" "packuswb %%mm1, %%mm0 \n\t" "packuswb %%mm3, %%mm2 \n\t" "movq %%mm0, %0 \n\t" "movq %%mm2, %1 \n\t" :"+m"(*pix), "+m"(*(pix+line_size)) :"r"(p) :"memory"); pix += line_size*2; p += 16; } while (--i); } static void put_pixels4_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h) { __asm __volatile( "lea (%3, %3), %%"REG_a" \n\t" ".balign 8 \n\t" "1: \n\t" "movd (%1), %%mm0 \n\t" "movd (%1, %3), %%mm1 \n\t" "movd %%mm0, (%2) \n\t" "movd %%mm1, (%2, %3) \n\t" "add %%"REG_a", %1 \n\t" "add %%"REG_a", %2 \n\t" "movd (%1), %%mm0 \n\t" "movd (%1, %3), %%mm1 \n\t" "movd %%mm0, (%2) \n\t" "movd %%mm1, (%2, %3) \n\t" "add %%"REG_a", %1 \n\t" "add %%"REG_a", %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" : "+g"(h), "+r" (pixels), "+r" (block) : "r"((long)line_size) : "%"REG_a, "memory" ); } static void put_pixels8_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h) { __asm __volatile( "lea (%3, %3), %%"REG_a" \n\t" ".balign 8 \n\t" "1: \n\t" "movq (%1), %%mm0 \n\t" "movq (%1, %3), %%mm1 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "add %%"REG_a", %1 \n\t" "add %%"REG_a", %2 \n\t" "movq (%1), %%mm0 \n\t" "movq (%1, %3), %%mm1 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "add %%"REG_a", %1 \n\t" "add %%"REG_a", %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" : "+g"(h), "+r" (pixels), "+r" (block) : "r"((long)line_size) : "%"REG_a, "memory" ); } static void put_pixels16_mmx(uint8_t *block, const uint8_t *pixels, int line_size, int h) { __asm __volatile( "lea (%3, %3), %%"REG_a" \n\t" ".balign 8 \n\t" "1: \n\t" "movq (%1), %%mm0 \n\t" "movq 8(%1), %%mm4 \n\t" "movq (%1, %3), %%mm1 \n\t" "movq 8(%1, %3), %%mm5 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm4, 8(%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "movq %%mm5, 8(%2, %3) \n\t" "add %%"REG_a", %1 \n\t" "add %%"REG_a", %2 \n\t" "movq (%1), %%mm0 \n\t" "movq 8(%1), %%mm4 \n\t" "movq (%1, %3), %%mm1 \n\t" "movq 8(%1, %3), %%mm5 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm4, 8(%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "movq %%mm5, 8(%2, %3) \n\t" "add %%"REG_a", %1 \n\t" "add %%"REG_a", %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" : "+g"(h), "+r" (pixels), "+r" (block) : "r"((long)line_size) : "%"REG_a, "memory" ); } static void clear_blocks_mmx(DCTELEM *blocks) { __asm __volatile( "pxor %%mm7, %%mm7 \n\t" "mov $-128*6, %%"REG_a" \n\t" "1: \n\t" "movq %%mm7, (%0, %%"REG_a") \n\t" "movq %%mm7, 8(%0, %%"REG_a") \n\t" "movq %%mm7, 16(%0, %%"REG_a") \n\t" "movq %%mm7, 24(%0, %%"REG_a") \n\t" "add $32, %%"REG_a" \n\t" " js 1b \n\t" : : "r" (((uint8_t *)blocks)+128*6) : "%"REG_a ); } #ifdef CONFIG_ENCODERS static int pix_sum16_mmx(uint8_t * pix, int line_size){ const int h=16; int sum; long index= -line_size*h; __asm __volatile( "pxor %%mm7, %%mm7 \n\t" "pxor %%mm6, %%mm6 \n\t" "1: \n\t" "movq (%2, %1), %%mm0 \n\t" "movq (%2, %1), %%mm1 \n\t" "movq 8(%2, %1), %%mm2 \n\t" "movq 8(%2, %1), %%mm3 \n\t" "punpcklbw %%mm7, %%mm0 \n\t" "punpckhbw %%mm7, %%mm1 \n\t" "punpcklbw %%mm7, %%mm2 \n\t" "punpckhbw %%mm7, %%mm3 \n\t" "paddw %%mm0, %%mm1 \n\t" "paddw %%mm2, %%mm3 \n\t" "paddw %%mm1, %%mm3 \n\t" "paddw %%mm3, %%mm6 \n\t" "add %3, %1 \n\t" " js 1b \n\t" "movq %%mm6, %%mm5 \n\t" "psrlq $32, %%mm6 \n\t" "paddw %%mm5, %%mm6 \n\t" "movq %%mm6, %%mm5 \n\t" "psrlq $16, %%mm6 \n\t" "paddw %%mm5, %%mm6 \n\t" "movd %%mm6, %0 \n\t" "andl $0xFFFF, %0 \n\t" : "=&r" (sum), "+r" (index) : "r" (pix - index), "r" ((long)line_size) ); return sum; } #endif //CONFIG_ENCODERS static void add_bytes_mmx(uint8_t *dst, uint8_t *src, int w){ long i=0; asm volatile( "1: \n\t" "movq (%1, %0), %%mm0 \n\t" "movq (%2, %0), %%mm1 \n\t" "paddb %%mm0, %%mm1 \n\t" "movq %%mm1, (%2, %0) \n\t" "movq 8(%1, %0), %%mm0 \n\t" "movq 8(%2, %0), %%mm1 \n\t" "paddb %%mm0, %%mm1 \n\t" "movq %%mm1, 8(%2, %0) \n\t" "add $16, %0 \n\t" "cmp %3, %0 \n\t" " jb 1b \n\t" : "+r" (i) : "r"(src), "r"(dst), "r"((long)w-15) ); for(; idsp.sse[0](c, pix1, pix2, line_size, h); else score1 = sse16_mmx(c, pix1, pix2, line_size, h); score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h); if(c) return score1 + ABS(score2)*c->avctx->nsse_weight; else return score1 + ABS(score2)*8; } static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) { MpegEncContext *c = p; int score1= sse8_mmx(c, pix1, pix2, line_size, h); int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h); if(c) return score1 + ABS(score2)*c->avctx->nsse_weight; else return score1 + ABS(score2)*8; } static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) { int tmp; assert( (((int)pix) & 7) == 0); assert((line_size &7) ==0); #define SUM(in0, in1, out0, out1) \ "movq (%0), %%mm2\n"\ "movq 8(%0), %%mm3\n"\ "add %2,%0\n"\ "movq %%mm2, " #out0 "\n"\ "movq %%mm3, " #out1 "\n"\ "psubusb " #in0 ", %%mm2\n"\ "psubusb " #in1 ", %%mm3\n"\ "psubusb " #out0 ", " #in0 "\n"\ "psubusb " #out1 ", " #in1 "\n"\ "por %%mm2, " #in0 "\n"\ "por %%mm3, " #in1 "\n"\ "movq " #in0 ", %%mm2\n"\ "movq " #in1 ", %%mm3\n"\ "punpcklbw %%mm7, " #in0 "\n"\ "punpcklbw %%mm7, " #in1 "\n"\ "punpckhbw %%mm7, %%mm2\n"\ "punpckhbw %%mm7, %%mm3\n"\ "paddw " #in1 ", " #in0 "\n"\ "paddw %%mm3, %%mm2\n"\ "paddw %%mm2, " #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" "subl $2, %%ecx\n" SUM(%%mm0, %%mm1, %%mm4, %%mm5) "1:\n" SUM(%%mm4, %%mm5, %%mm0, %%mm1) SUM(%%mm0, %%mm1, %%mm4, %%mm5) "subl $2, %%ecx\n" "jnz 1b\n" "movq %%mm6,%%mm0\n" "psrlq $32, %%mm6\n" "paddw %%mm6,%%mm0\n" "movq %%mm0,%%mm6\n" "psrlq $16, %%mm0\n" "paddw %%mm6,%%mm0\n" "movd %%mm0,%1\n" : "+r" (pix), "=r"(tmp) : "r" ((long)line_size) , "m" (h) : "%ecx"); return tmp & 0xFFFF; } #undef SUM static int vsad_intra16_mmx2(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) { int tmp; assert( (((int)pix) & 7) == 0); assert((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" "subl $2, %%ecx\n" SUM(%%mm0, %%mm1, %%mm4, %%mm5) "1:\n" SUM(%%mm4, %%mm5, %%mm0, %%mm1) SUM(%%mm0, %%mm1, %%mm4, %%mm5) "subl $2, %%ecx\n" "jnz 1b\n" "movd %%mm6,%1\n" : "+r" (pix), "=r"(tmp) : "r" ((long)line_size) , "m" (h) : "%ecx"); return tmp; } #undef SUM static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) { int tmp; assert( (((int)pix1) & 7) == 0); assert( (((int)pix2) & 7) == 0); assert((line_size &7) ==0); #define SUM(in0, in1, out0, out1) \ "movq (%0),%%mm2\n"\ "movq (%1)," #out0 "\n"\ "movq 8(%0),%%mm3\n"\ "movq 8(%1)," #out1 "\n"\ "add %3,%0\n"\ "add %3,%1\n"\ "psubb " #out0 ", %%mm2\n"\ "psubb " #out1 ", %%mm3\n"\ "pxor %%mm7, %%mm2\n"\ "pxor %%mm7, %%mm3\n"\ "movq %%mm2, " #out0 "\n"\ "movq %%mm3, " #out1 "\n"\ "psubusb " #in0 ", %%mm2\n"\ "psubusb " #in1 ", %%mm3\n"\ "psubusb " #out0 ", " #in0 "\n"\ "psubusb " #out1 ", " #in1 "\n"\ "por %%mm2, " #in0 "\n"\ "por %%mm3, " #in1 "\n"\ "movq " #in0 ", %%mm2\n"\ "movq " #in1 ", %%mm3\n"\ "punpcklbw %%mm7, " #in0 "\n"\ "punpcklbw %%mm7, " #in1 "\n"\ "punpckhbw %%mm7, %%mm2\n"\ "punpckhbw %%mm7, %%mm3\n"\ "paddw " #in1 ", " #in0 "\n"\ "paddw %%mm3, %%mm2\n"\ "paddw %%mm2, " #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" "subl $2, %%ecx\n" "psubb %%mm2, %%mm0\n" "psubb %%mm3, %%mm1\n" "pxor %%mm7, %%mm0\n" "pxor %%mm7, %%mm1\n" SUM(%%mm0, %%mm1, %%mm4, %%mm5) "1:\n" SUM(%%mm4, %%mm5, %%mm0, %%mm1) SUM(%%mm0, %%mm1, %%mm4, %%mm5) "subl $2, %%ecx\n" "jnz 1b\n" "movq %%mm6,%%mm0\n" "psrlq $32, %%mm6\n" "paddw %%mm6,%%mm0\n" "movq %%mm0,%%mm6\n" "psrlq $16, %%mm0\n" "paddw %%mm6,%%mm0\n" "movd %%mm0,%2\n" : "+r" (pix1), "+r" (pix2), "=r"(tmp) : "r" ((long)line_size) , "m" (h) : "%ecx"); return tmp & 0x7FFF; } #undef SUM static int vsad16_mmx2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) { int tmp; assert( (((int)pix1) & 7) == 0); assert( (((int)pix2) & 7) == 0); assert((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" "subl $2, %%ecx\n" "psubb %%mm2, %%mm0\n" "psubb %%mm3, %%mm1\n" "pxor %%mm7, %%mm0\n" "pxor %%mm7, %%mm1\n" SUM(%%mm0, %%mm1, %%mm4, %%mm5) "1:\n" SUM(%%mm4, %%mm5, %%mm0, %%mm1) SUM(%%mm0, %%mm1, %%mm4, %%mm5) "subl $2, %%ecx\n" "jnz 1b\n" "movd %%mm6,%2\n" : "+r" (pix1), "+r" (pix2), "=r"(tmp) : "r" ((long)line_size) , "m" (h) : "%ecx"); return tmp; } #undef SUM static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){ long i=0; asm volatile( "1: \n\t" "movq (%2, %0), %%mm0 \n\t" "movq (%1, %0), %%mm1 \n\t" "psubb %%mm0, %%mm1 \n\t" "movq %%mm1, (%3, %0) \n\t" "movq 8(%2, %0), %%mm0 \n\t" "movq 8(%1, %0), %%mm1 \n\t" "psubb %%mm0, %%mm1 \n\t" "movq %%mm1, 8(%3, %0) \n\t" "add $16, %0 \n\t" "cmp %4, %0 \n\t" " jb 1b \n\t" : "+r" (i) : "r"(src1), "r"(src2), "r"(dst), "r"((long)w-15) ); for(; iput_ ## postfix1 = put_ ## postfix2;\ c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\ c->avg_ ## postfix1 = avg_ ## postfix2; static void gmc_mmx(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height){ const int w = 8; const int ix = ox>>(16+shift); const int iy = oy>>(16+shift); const int oxs = ox>>4; const int oys = oy>>4; const int dxxs = dxx>>4; const int dxys = dxy>>4; const int dyxs = dyx>>4; const int dyys = dyy>>4; const uint16_t r4[4] = {r,r,r,r}; const uint16_t dxy4[4] = {dxys,dxys,dxys,dxys}; const uint16_t dyy4[4] = {dyys,dyys,dyys,dyys}; const uint64_t shift2 = 2*shift; uint8_t edge_buf[(h+1)*stride]; int x, y; const int dxw = (dxx-(1<<(16+shift)))*(w-1); const int dyh = (dyy-(1<<(16+shift)))*(h-1); const int dxh = dxy*(h-1); const int dyw = dyx*(w-1); if( // non-constant fullpel offset (3% of blocks) (ox^(ox+dxw) | ox^(ox+dxh) | ox^(ox+dxw+dxh) | oy^(oy+dyw) | oy^(oy+dyh) | oy^(oy+dyw+dyh)) >> (16+shift) // uses more than 16 bits of subpel mv (only at huge resolution) || (dxx|dxy|dyx|dyy)&15 ) { //FIXME could still use mmx for some of the rows ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, shift, r, width, height); return; } src += ix + iy*stride; if( (unsigned)ix >= width-w || (unsigned)iy >= height-h ) { ff_emulated_edge_mc(edge_buf, src, stride, w+1, h+1, ix, iy, width, height); src = edge_buf; } asm volatile( "movd %0, %%mm6 \n\t" "pxor %%mm7, %%mm7 \n\t" "punpcklwd %%mm6, %%mm6 \n\t" "punpcklwd %%mm6, %%mm6 \n\t" :: "r"(1<>(BASIS_SHIFT - RECON_SHIFT); } } } #define PREFETCH(name, op) \ void name(void *mem, int stride, int h){\ const uint8_t *p= mem;\ do{\ asm volatile(#op" %0" :: "m"(*p));\ p+= stride;\ }while(--h);\ } PREFETCH(prefetch_mmx2, prefetcht0) PREFETCH(prefetch_3dnow, prefetch) #undef PREFETCH #include "h264dsp_mmx.c" /* AVS specific */ void ff_cavsdsp_init_mmx2(DSPContext* c, AVCodecContext *avctx); void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) { put_pixels8_mmx(dst, src, stride, 8); } void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) { avg_pixels8_mmx(dst, src, stride, 8); } void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) { put_pixels16_mmx(dst, src, stride, 16); } void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride) { avg_pixels16_mmx(dst, src, stride, 16); } /* external functions, from idct_mmx.c */ void ff_mmx_idct(DCTELEM *block); void ff_mmxext_idct(DCTELEM *block); void ff_vp3_idct_sse2(int16_t *input_data); void ff_vp3_idct_mmx(int16_t *data); void ff_vp3_dsp_init_mmx(void); /* XXX: those functions should be suppressed ASAP when all IDCTs are converted */ static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block) { ff_mmx_idct (block); put_pixels_clamped_mmx(block, dest, line_size); } static void ff_libmpeg2mmx_idct_add(uint8_t *dest, int line_size, DCTELEM *block) { ff_mmx_idct (block); add_pixels_clamped_mmx(block, dest, line_size); } static void ff_libmpeg2mmx2_idct_put(uint8_t *dest, int line_size, DCTELEM *block) { ff_mmxext_idct (block); put_pixels_clamped_mmx(block, dest, line_size); } static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *block) { ff_mmxext_idct (block); add_pixels_clamped_mmx(block, dest, line_size); } static void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block) { ff_vp3_idct_sse2(block); put_signed_pixels_clamped_mmx(block, dest, line_size); } static void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block) { ff_vp3_idct_sse2(block); add_pixels_clamped_mmx(block, dest, line_size); } static void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block) { ff_vp3_idct_mmx(block); put_signed_pixels_clamped_mmx(block, dest, line_size); } static void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block) { ff_vp3_idct_mmx(block); add_pixels_clamped_mmx(block, dest, line_size); } #ifdef CONFIG_GPL static void ff_idct_xvid_mmx_put(uint8_t *dest, int line_size, DCTELEM *block) { ff_idct_xvid_mmx (block); put_pixels_clamped_mmx(block, dest, line_size); } static void ff_idct_xvid_mmx_add(uint8_t *dest, int line_size, DCTELEM *block) { ff_idct_xvid_mmx (block); add_pixels_clamped_mmx(block, dest, line_size); } static void ff_idct_xvid_mmx2_put(uint8_t *dest, int line_size, DCTELEM *block) { ff_idct_xvid_mmx2 (block); put_pixels_clamped_mmx(block, dest, line_size); } static void ff_idct_xvid_mmx2_add(uint8_t *dest, int line_size, DCTELEM *block) { ff_idct_xvid_mmx2 (block); add_pixels_clamped_mmx(block, dest, line_size); } #endif static void vorbis_inverse_coupling_3dnow(float *mag, float *ang, int blocksize) { int i; asm volatile("pxor %%mm7, %%mm7":); for(i=0; i0) & (a ^ sign(m))) "movq %%mm3, %1 \n\t" "movq %%mm0, %0 \n\t" :"+m"(mag[i]), "+m"(ang[i]) ::"memory" ); } asm volatile("femms"); } static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize) { int i; asm volatile( "movaps %0, %%xmm5 \n\t" ::"m"(ff_pdw_80000000[0]) ); for(i=0; i0) & (a ^ sign(m))) "movaps %%xmm3, %1 \n\t" "movaps %%xmm0, %0 \n\t" :"+m"(mag[i]), "+m"(ang[i]) ::"memory" ); } } static void vector_fmul_3dnow(float *dst, const float *src, int len){ long i; len >>= 1; for(i=0; i>= 2; for(i=0; idsp_mask) { if (avctx->dsp_mask & FF_MM_FORCE) mm_flags |= (avctx->dsp_mask & 0xffff); else mm_flags &= ~(avctx->dsp_mask & 0xffff); } #if 0 av_log(avctx, AV_LOG_INFO, "libavcodec: CPU flags:"); if (mm_flags & MM_MMX) av_log(avctx, AV_LOG_INFO, " mmx"); if (mm_flags & MM_MMXEXT) av_log(avctx, AV_LOG_INFO, " mmxext"); if (mm_flags & MM_3DNOW) av_log(avctx, AV_LOG_INFO, " 3dnow"); if (mm_flags & MM_SSE) av_log(avctx, AV_LOG_INFO, " sse"); if (mm_flags & MM_SSE2) av_log(avctx, AV_LOG_INFO, " sse2"); av_log(avctx, AV_LOG_INFO, "\n"); #endif if (mm_flags & MM_MMX) { const int idct_algo= avctx->idct_algo; #ifdef CONFIG_ENCODERS const int dct_algo = avctx->dct_algo; if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){ if(mm_flags & MM_SSE2){ c->fdct = ff_fdct_sse2; }else if(mm_flags & MM_MMXEXT){ c->fdct = ff_fdct_mmx2; }else{ c->fdct = ff_fdct_mmx; } } #endif //CONFIG_ENCODERS if(avctx->lowres==0){ if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_SIMPLEMMX){ c->idct_put= ff_simple_idct_put_mmx; c->idct_add= ff_simple_idct_add_mmx; c->idct = ff_simple_idct_mmx; c->idct_permutation_type= FF_SIMPLE_IDCT_PERM; }else if(idct_algo==FF_IDCT_LIBMPEG2MMX){ if(mm_flags & MM_MMXEXT){ c->idct_put= ff_libmpeg2mmx2_idct_put; c->idct_add= ff_libmpeg2mmx2_idct_add; c->idct = ff_mmxext_idct; }else{ c->idct_put= ff_libmpeg2mmx_idct_put; c->idct_add= ff_libmpeg2mmx_idct_add; c->idct = ff_mmx_idct; } c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM; #if 0 }else if(idct_algo==FF_IDCT_VP3){ if(mm_flags & MM_SSE2){ c->idct_put= ff_vp3_idct_put_sse2; c->idct_add= ff_vp3_idct_add_sse2; c->idct = ff_vp3_idct_sse2; c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM; }else{ ff_vp3_dsp_init_mmx(); c->idct_put= ff_vp3_idct_put_mmx; c->idct_add= ff_vp3_idct_add_mmx; c->idct = ff_vp3_idct_mmx; c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM; } #endif }else if(idct_algo==FF_IDCT_CAVS){ c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM; #ifdef CONFIG_GPL }else if(idct_algo==FF_IDCT_XVIDMMX){ if(mm_flags & MM_MMXEXT){ c->idct_put= ff_idct_xvid_mmx2_put; c->idct_add= ff_idct_xvid_mmx2_add; c->idct = ff_idct_xvid_mmx2; }else{ c->idct_put= ff_idct_xvid_mmx_put; c->idct_add= ff_idct_xvid_mmx_add; c->idct = ff_idct_xvid_mmx; } #endif } } #ifdef CONFIG_ENCODERS c->get_pixels = get_pixels_mmx; c->diff_pixels = diff_pixels_mmx; #endif //CONFIG_ENCODERS c->put_pixels_clamped = put_pixels_clamped_mmx; c->put_signed_pixels_clamped = put_signed_pixels_clamped_mmx; c->add_pixels_clamped = add_pixels_clamped_mmx; c->clear_blocks = clear_blocks_mmx; #ifdef CONFIG_ENCODERS c->pix_sum = pix_sum16_mmx; #endif //CONFIG_ENCODERS c->put_pixels_tab[0][0] = put_pixels16_mmx; c->put_pixels_tab[0][1] = put_pixels16_x2_mmx; c->put_pixels_tab[0][2] = put_pixels16_y2_mmx; c->put_pixels_tab[0][3] = put_pixels16_xy2_mmx; c->put_no_rnd_pixels_tab[0][0] = put_pixels16_mmx; c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx; c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx; c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_mmx; c->avg_pixels_tab[0][0] = avg_pixels16_mmx; c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx; c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx; c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx; c->avg_no_rnd_pixels_tab[0][0] = avg_no_rnd_pixels16_mmx; c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_mmx; c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_mmx; c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_mmx; c->put_pixels_tab[1][0] = put_pixels8_mmx; c->put_pixels_tab[1][1] = put_pixels8_x2_mmx; c->put_pixels_tab[1][2] = put_pixels8_y2_mmx; c->put_pixels_tab[1][3] = put_pixels8_xy2_mmx; c->put_no_rnd_pixels_tab[1][0] = put_pixels8_mmx; c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx; c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx; c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_mmx; c->avg_pixels_tab[1][0] = avg_pixels8_mmx; c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx; c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx; c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx; c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_mmx; c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx; c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx; c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx; c->gmc= gmc_mmx; c->add_bytes= add_bytes_mmx; #ifdef CONFIG_ENCODERS c->diff_bytes= diff_bytes_mmx; c->hadamard8_diff[0]= hadamard8_diff16_mmx; c->hadamard8_diff[1]= hadamard8_diff_mmx; c->pix_norm1 = pix_norm1_mmx; c->sse[0] = (mm_flags & MM_SSE2) ? sse16_sse2 : sse16_mmx; c->sse[1] = sse8_mmx; c->vsad[4]= vsad_intra16_mmx; c->nsse[0] = nsse16_mmx; c->nsse[1] = nsse8_mmx; if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ c->vsad[0] = vsad16_mmx; } if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ c->try_8x8basis= try_8x8basis_mmx; } c->add_8x8basis= add_8x8basis_mmx; #endif //CONFIG_ENCODERS c->h263_v_loop_filter= h263_v_loop_filter_mmx; c->h263_h_loop_filter= h263_h_loop_filter_mmx; c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx; c->put_h264_chroma_pixels_tab[1]= put_h264_chroma_mc4_mmx; c->h264_idct_dc_add= c->h264_idct_add= ff_h264_idct_add_mmx; c->h264_idct8_dc_add= c->h264_idct8_add= ff_h264_idct8_add_mmx; if (mm_flags & MM_MMXEXT) { c->prefetch = prefetch_mmx2; c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2; c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2; c->avg_pixels_tab[0][0] = avg_pixels16_mmx2; c->avg_pixels_tab[0][1] = avg_pixels16_x2_mmx2; c->avg_pixels_tab[0][2] = avg_pixels16_y2_mmx2; c->put_pixels_tab[1][1] = put_pixels8_x2_mmx2; c->put_pixels_tab[1][2] = put_pixels8_y2_mmx2; c->avg_pixels_tab[1][0] = avg_pixels8_mmx2; c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2; c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2; #ifdef CONFIG_ENCODERS c->hadamard8_diff[0]= hadamard8_diff16_mmx2; c->hadamard8_diff[1]= hadamard8_diff_mmx2; c->vsad[4]= vsad_intra16_mmx2; #endif //CONFIG_ENCODERS c->h264_idct_dc_add= ff_h264_idct_dc_add_mmx2; c->h264_idct8_dc_add= ff_h264_idct8_dc_add_mmx2; if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2; c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_mmx2; c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_mmx2; c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_mmx2; c->avg_pixels_tab[0][3] = avg_pixels16_xy2_mmx2; c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2; #ifdef CONFIG_ENCODERS c->vsad[0] = vsad16_mmx2; #endif //CONFIG_ENCODERS } #if 1 SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2) SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2) #endif //FIXME 3dnow too #define dspfunc(PFX, IDX, NUM) \ c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_mmx2; \ c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_mmx2; \ c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_mmx2; \ c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_mmx2; \ c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_mmx2; \ c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_mmx2; \ c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_mmx2; \ c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_mmx2; \ c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_mmx2; \ c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_mmx2; \ c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_mmx2; \ c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_mmx2; \ c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_mmx2; \ c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_mmx2; \ c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_mmx2; \ c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_mmx2 dspfunc(put_h264_qpel, 0, 16); dspfunc(put_h264_qpel, 1, 8); dspfunc(put_h264_qpel, 2, 4); dspfunc(avg_h264_qpel, 0, 16); dspfunc(avg_h264_qpel, 1, 8); dspfunc(avg_h264_qpel, 2, 4); #undef dspfunc c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2; c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_mmx2; c->avg_h264_chroma_pixels_tab[2]= avg_h264_chroma_mc2_mmx2; c->put_h264_chroma_pixels_tab[2]= put_h264_chroma_mc2_mmx2; c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2; c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2; c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2; c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2; c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2; c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2; c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2; c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2; c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2; c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2; c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2; c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2; c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2; c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2; c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2; c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2; c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2; c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2; c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2; c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2; c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2; c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2; #ifdef CONFIG_CAVS_DECODER ff_cavsdsp_init_mmx2(c, avctx); #endif #ifdef CONFIG_ENCODERS c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2; #endif //CONFIG_ENCODERS } else if (mm_flags & MM_3DNOW) { c->prefetch = prefetch_3dnow; c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow; c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow; c->avg_pixels_tab[0][0] = avg_pixels16_3dnow; c->avg_pixels_tab[0][1] = avg_pixels16_x2_3dnow; c->avg_pixels_tab[0][2] = avg_pixels16_y2_3dnow; c->put_pixels_tab[1][1] = put_pixels8_x2_3dnow; c->put_pixels_tab[1][2] = put_pixels8_y2_3dnow; c->avg_pixels_tab[1][0] = avg_pixels8_3dnow; c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow; c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow; if(!(avctx->flags & CODEC_FLAG_BITEXACT)){ c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_3dnow; c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_3dnow; c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_3dnow; c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_3dnow; c->avg_pixels_tab[0][3] = avg_pixels16_xy2_3dnow; c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow; } SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow) SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow) #define dspfunc(PFX, IDX, NUM) \ c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_3dnow; \ c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_3dnow; \ c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_3dnow; \ c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_3dnow; \ c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_3dnow; \ c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_3dnow; \ c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_3dnow; \ c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_3dnow; \ c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_3dnow; \ c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_3dnow; \ c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_3dnow; \ c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_3dnow; \ c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_3dnow; \ c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_3dnow; \ c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_3dnow; \ c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_3dnow dspfunc(put_h264_qpel, 0, 16); dspfunc(put_h264_qpel, 1, 8); dspfunc(put_h264_qpel, 2, 4); dspfunc(avg_h264_qpel, 0, 16); dspfunc(avg_h264_qpel, 1, 8); dspfunc(avg_h264_qpel, 2, 4); c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow; c->avg_h264_chroma_pixels_tab[1]= avg_h264_chroma_mc4_3dnow; } #ifdef CONFIG_SNOW_ENCODER if(mm_flags & MM_SSE2){ c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2; c->vertical_compose97i = ff_snow_vertical_compose97i_sse2; c->inner_add_yblock = ff_snow_inner_add_yblock_sse2; } else{ c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx; c->vertical_compose97i = ff_snow_vertical_compose97i_mmx; c->inner_add_yblock = ff_snow_inner_add_yblock_mmx; } #endif if(mm_flags & MM_3DNOW){ c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow; c->vector_fmul = vector_fmul_3dnow; if(!(avctx->flags & CODEC_FLAG_BITEXACT)) c->float_to_int16 = float_to_int16_3dnow; } if(mm_flags & MM_3DNOWEXT) c->vector_fmul_reverse = vector_fmul_reverse_3dnow2; if(mm_flags & MM_SSE){ c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse; c->vector_fmul = vector_fmul_sse; c->float_to_int16 = float_to_int16_sse; c->vector_fmul_reverse = vector_fmul_reverse_sse; c->vector_fmul_add_add = vector_fmul_add_add_sse; } if(mm_flags & MM_3DNOW) c->vector_fmul_add_add = vector_fmul_add_add_3dnow; // faster than sse2 } #ifdef CONFIG_ENCODERS dsputil_init_pix_mmx(c, avctx); #endif //CONFIG_ENCODERS #if 0 // for speed testing get_pixels = just_return; put_pixels_clamped = just_return; add_pixels_clamped = just_return; pix_abs16x16 = just_return; pix_abs16x16_x2 = just_return; pix_abs16x16_y2 = just_return; pix_abs16x16_xy2 = just_return; put_pixels_tab[0] = just_return; put_pixels_tab[1] = just_return; put_pixels_tab[2] = just_return; put_pixels_tab[3] = just_return; put_no_rnd_pixels_tab[0] = just_return; put_no_rnd_pixels_tab[1] = just_return; put_no_rnd_pixels_tab[2] = just_return; put_no_rnd_pixels_tab[3] = just_return; avg_pixels_tab[0] = just_return; avg_pixels_tab[1] = just_return; avg_pixels_tab[2] = just_return; avg_pixels_tab[3] = just_return; avg_no_rnd_pixels_tab[0] = just_return; avg_no_rnd_pixels_tab[1] = just_return; avg_no_rnd_pixels_tab[2] = just_return; avg_no_rnd_pixels_tab[3] = just_return; //av_fdct = just_return; //ff_idct = just_return; #endif }