yading@10: /*
yading@10:  * MMX optimized DSP utils
yading@10:  * Copyright (c) 2000, 2001 Fabrice Bellard
yading@10:  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
yading@10:  *
yading@10:  * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #include "libavutil/attributes.h"
yading@10: #include "libavutil/cpu.h"
yading@10: #include "libavutil/x86/asm.h"
yading@10: #include "libavutil/x86/cpu.h"
yading@10: #include "libavcodec/dct.h"
yading@10: #include "libavcodec/dsputil.h"
yading@10: #include "libavcodec/mpegvideo.h"
yading@10: #include "libavcodec/mathops.h"
yading@10: #include "dsputil_mmx.h"
yading@10: 
yading@10: void ff_get_pixels_mmx(int16_t *block, const uint8_t *pixels, int line_size);
yading@10: void ff_get_pixels_sse2(int16_t *block, const uint8_t *pixels, int line_size);
yading@10: void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2, int stride);
yading@10: int ff_pix_sum16_mmx(uint8_t * pix, int line_size);
yading@10: int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
yading@10: 
yading@10: #if HAVE_INLINE_ASM
yading@10: 
yading@10: static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
yading@10:     int tmp;
yading@10:   __asm__ volatile (
yading@10:       "movl %4,%%ecx\n"
yading@10:       "shr $1,%%ecx\n"
yading@10:       "pxor %%mm0,%%mm0\n"      /* mm0 = 0 */
yading@10:       "pxor %%mm7,%%mm7\n"      /* mm7 holds the sum */
yading@10:       "1:\n"
yading@10:       "movq (%0),%%mm1\n"       /* mm1 = pix1[0][0-7] */
yading@10:       "movq (%1),%%mm2\n"       /* mm2 = pix2[0][0-7] */
yading@10:       "movq (%0,%3),%%mm3\n"    /* mm3 = pix1[1][0-7] */
yading@10:       "movq (%1,%3),%%mm4\n"    /* mm4 = pix2[1][0-7] */
yading@10: 
yading@10:       /* todo: mm1-mm2, mm3-mm4 */
yading@10:       /* algo: subtract mm1 from mm2 with saturation and vice versa */
yading@10:       /*       OR the results to get absolute difference */
yading@10:       "movq %%mm1,%%mm5\n"
yading@10:       "movq %%mm3,%%mm6\n"
yading@10:       "psubusb %%mm2,%%mm1\n"
yading@10:       "psubusb %%mm4,%%mm3\n"
yading@10:       "psubusb %%mm5,%%mm2\n"
yading@10:       "psubusb %%mm6,%%mm4\n"
yading@10: 
yading@10:       "por %%mm1,%%mm2\n"
yading@10:       "por %%mm3,%%mm4\n"
yading@10: 
yading@10:       /* now convert to 16-bit vectors so we can square them */
yading@10:       "movq %%mm2,%%mm1\n"
yading@10:       "movq %%mm4,%%mm3\n"
yading@10: 
yading@10:       "punpckhbw %%mm0,%%mm2\n"
yading@10:       "punpckhbw %%mm0,%%mm4\n"
yading@10:       "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
yading@10:       "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
yading@10: 
yading@10:       "pmaddwd %%mm2,%%mm2\n"
yading@10:       "pmaddwd %%mm4,%%mm4\n"
yading@10:       "pmaddwd %%mm1,%%mm1\n"
yading@10:       "pmaddwd %%mm3,%%mm3\n"
yading@10: 
yading@10:       "lea (%0,%3,2), %0\n"     /* pix1 += 2*line_size */
yading@10:       "lea (%1,%3,2), %1\n"     /* pix2 += 2*line_size */
yading@10: 
yading@10:       "paddd %%mm2,%%mm1\n"
yading@10:       "paddd %%mm4,%%mm3\n"
yading@10:       "paddd %%mm1,%%mm7\n"
yading@10:       "paddd %%mm3,%%mm7\n"
yading@10: 
yading@10:       "decl %%ecx\n"
yading@10:       "jnz 1b\n"
yading@10: 
yading@10:       "movq %%mm7,%%mm1\n"
yading@10:       "psrlq $32, %%mm7\n"      /* shift hi dword to lo */
yading@10:       "paddd %%mm7,%%mm1\n"
yading@10:       "movd %%mm1,%2\n"
yading@10:       : "+r" (pix1), "+r" (pix2), "=r"(tmp)
yading@10:       : "r" ((x86_reg)line_size) , "m" (h)
yading@10:       : "%ecx");
yading@10:     return tmp;
yading@10: }
yading@10: 
yading@10: static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
yading@10:     int tmp;
yading@10:   __asm__ volatile (
yading@10:       "movl %4,%%ecx\n"
yading@10:       "pxor %%mm0,%%mm0\n"      /* mm0 = 0 */
yading@10:       "pxor %%mm7,%%mm7\n"      /* mm7 holds the sum */
yading@10:       "1:\n"
yading@10:       "movq (%0),%%mm1\n"       /* mm1 = pix1[0-7] */
yading@10:       "movq (%1),%%mm2\n"       /* mm2 = pix2[0-7] */
yading@10:       "movq 8(%0),%%mm3\n"      /* mm3 = pix1[8-15] */
yading@10:       "movq 8(%1),%%mm4\n"      /* mm4 = pix2[8-15] */
yading@10: 
yading@10:       /* todo: mm1-mm2, mm3-mm4 */
yading@10:       /* algo: subtract mm1 from mm2 with saturation and vice versa */
yading@10:       /*       OR the results to get absolute difference */
yading@10:       "movq %%mm1,%%mm5\n"
yading@10:       "movq %%mm3,%%mm6\n"
yading@10:       "psubusb %%mm2,%%mm1\n"
yading@10:       "psubusb %%mm4,%%mm3\n"
yading@10:       "psubusb %%mm5,%%mm2\n"
yading@10:       "psubusb %%mm6,%%mm4\n"
yading@10: 
yading@10:       "por %%mm1,%%mm2\n"
yading@10:       "por %%mm3,%%mm4\n"
yading@10: 
yading@10:       /* now convert to 16-bit vectors so we can square them */
yading@10:       "movq %%mm2,%%mm1\n"
yading@10:       "movq %%mm4,%%mm3\n"
yading@10: 
yading@10:       "punpckhbw %%mm0,%%mm2\n"
yading@10:       "punpckhbw %%mm0,%%mm4\n"
yading@10:       "punpcklbw %%mm0,%%mm1\n" /* mm1 now spread over (mm1,mm2) */
yading@10:       "punpcklbw %%mm0,%%mm3\n" /* mm4 now spread over (mm3,mm4) */
yading@10: 
yading@10:       "pmaddwd %%mm2,%%mm2\n"
yading@10:       "pmaddwd %%mm4,%%mm4\n"
yading@10:       "pmaddwd %%mm1,%%mm1\n"
yading@10:       "pmaddwd %%mm3,%%mm3\n"
yading@10: 
yading@10:       "add %3,%0\n"
yading@10:       "add %3,%1\n"
yading@10: 
yading@10:       "paddd %%mm2,%%mm1\n"
yading@10:       "paddd %%mm4,%%mm3\n"
yading@10:       "paddd %%mm1,%%mm7\n"
yading@10:       "paddd %%mm3,%%mm7\n"
yading@10: 
yading@10:       "decl %%ecx\n"
yading@10:       "jnz 1b\n"
yading@10: 
yading@10:       "movq %%mm7,%%mm1\n"
yading@10:       "psrlq $32, %%mm7\n"      /* shift hi dword to lo */
yading@10:       "paddd %%mm7,%%mm1\n"
yading@10:       "movd %%mm1,%2\n"
yading@10:       : "+r" (pix1), "+r" (pix2), "=r"(tmp)
yading@10:       : "r" ((x86_reg)line_size) , "m" (h)
yading@10:       : "%ecx");
yading@10:     return tmp;
yading@10: }
yading@10: 
yading@10: static int hf_noise8_mmx(uint8_t * pix1, int line_size, int h) {
yading@10:     int tmp;
yading@10:   __asm__ volatile (
yading@10:       "movl %3,%%ecx\n"
yading@10:       "pxor %%mm7,%%mm7\n"
yading@10:       "pxor %%mm6,%%mm6\n"
yading@10: 
yading@10:       "movq (%0),%%mm0\n"
yading@10:       "movq %%mm0, %%mm1\n"
yading@10:       "psllq $8, %%mm0\n"
yading@10:       "psrlq $8, %%mm1\n"
yading@10:       "psrlq $8, %%mm0\n"
yading@10:       "movq %%mm0, %%mm2\n"
yading@10:       "movq %%mm1, %%mm3\n"
yading@10:       "punpcklbw %%mm7,%%mm0\n"
yading@10:       "punpcklbw %%mm7,%%mm1\n"
yading@10:       "punpckhbw %%mm7,%%mm2\n"
yading@10:       "punpckhbw %%mm7,%%mm3\n"
yading@10:       "psubw %%mm1, %%mm0\n"
yading@10:       "psubw %%mm3, %%mm2\n"
yading@10: 
yading@10:       "add %2,%0\n"
yading@10: 
yading@10:       "movq (%0),%%mm4\n"
yading@10:       "movq %%mm4, %%mm1\n"
yading@10:       "psllq $8, %%mm4\n"
yading@10:       "psrlq $8, %%mm1\n"
yading@10:       "psrlq $8, %%mm4\n"
yading@10:       "movq %%mm4, %%mm5\n"
yading@10:       "movq %%mm1, %%mm3\n"
yading@10:       "punpcklbw %%mm7,%%mm4\n"
yading@10:       "punpcklbw %%mm7,%%mm1\n"
yading@10:       "punpckhbw %%mm7,%%mm5\n"
yading@10:       "punpckhbw %%mm7,%%mm3\n"
yading@10:       "psubw %%mm1, %%mm4\n"
yading@10:       "psubw %%mm3, %%mm5\n"
yading@10:       "psubw %%mm4, %%mm0\n"
yading@10:       "psubw %%mm5, %%mm2\n"
yading@10:       "pxor %%mm3, %%mm3\n"
yading@10:       "pxor %%mm1, %%mm1\n"
yading@10:       "pcmpgtw %%mm0, %%mm3\n\t"
yading@10:       "pcmpgtw %%mm2, %%mm1\n\t"
yading@10:       "pxor %%mm3, %%mm0\n"
yading@10:       "pxor %%mm1, %%mm2\n"
yading@10:       "psubw %%mm3, %%mm0\n"
yading@10:       "psubw %%mm1, %%mm2\n"
yading@10:       "paddw %%mm0, %%mm2\n"
yading@10:       "paddw %%mm2, %%mm6\n"
yading@10: 
yading@10:       "add %2,%0\n"
yading@10:       "1:\n"
yading@10: 
yading@10:       "movq (%0),%%mm0\n"
yading@10:       "movq %%mm0, %%mm1\n"
yading@10:       "psllq $8, %%mm0\n"
yading@10:       "psrlq $8, %%mm1\n"
yading@10:       "psrlq $8, %%mm0\n"
yading@10:       "movq %%mm0, %%mm2\n"
yading@10:       "movq %%mm1, %%mm3\n"
yading@10:       "punpcklbw %%mm7,%%mm0\n"
yading@10:       "punpcklbw %%mm7,%%mm1\n"
yading@10:       "punpckhbw %%mm7,%%mm2\n"
yading@10:       "punpckhbw %%mm7,%%mm3\n"
yading@10:       "psubw %%mm1, %%mm0\n"
yading@10:       "psubw %%mm3, %%mm2\n"
yading@10:       "psubw %%mm0, %%mm4\n"
yading@10:       "psubw %%mm2, %%mm5\n"
yading@10:       "pxor %%mm3, %%mm3\n"
yading@10:       "pxor %%mm1, %%mm1\n"
yading@10:       "pcmpgtw %%mm4, %%mm3\n\t"
yading@10:       "pcmpgtw %%mm5, %%mm1\n\t"
yading@10:       "pxor %%mm3, %%mm4\n"
yading@10:       "pxor %%mm1, %%mm5\n"
yading@10:       "psubw %%mm3, %%mm4\n"
yading@10:       "psubw %%mm1, %%mm5\n"
yading@10:       "paddw %%mm4, %%mm5\n"
yading@10:       "paddw %%mm5, %%mm6\n"
yading@10: 
yading@10:       "add %2,%0\n"
yading@10: 
yading@10:       "movq (%0),%%mm4\n"
yading@10:       "movq %%mm4, %%mm1\n"
yading@10:       "psllq $8, %%mm4\n"
yading@10:       "psrlq $8, %%mm1\n"
yading@10:       "psrlq $8, %%mm4\n"
yading@10:       "movq %%mm4, %%mm5\n"
yading@10:       "movq %%mm1, %%mm3\n"
yading@10:       "punpcklbw %%mm7,%%mm4\n"
yading@10:       "punpcklbw %%mm7,%%mm1\n"
yading@10:       "punpckhbw %%mm7,%%mm5\n"
yading@10:       "punpckhbw %%mm7,%%mm3\n"
yading@10:       "psubw %%mm1, %%mm4\n"
yading@10:       "psubw %%mm3, %%mm5\n"
yading@10:       "psubw %%mm4, %%mm0\n"
yading@10:       "psubw %%mm5, %%mm2\n"
yading@10:       "pxor %%mm3, %%mm3\n"
yading@10:       "pxor %%mm1, %%mm1\n"
yading@10:       "pcmpgtw %%mm0, %%mm3\n\t"
yading@10:       "pcmpgtw %%mm2, %%mm1\n\t"
yading@10:       "pxor %%mm3, %%mm0\n"
yading@10:       "pxor %%mm1, %%mm2\n"
yading@10:       "psubw %%mm3, %%mm0\n"
yading@10:       "psubw %%mm1, %%mm2\n"
yading@10:       "paddw %%mm0, %%mm2\n"
yading@10:       "paddw %%mm2, %%mm6\n"
yading@10: 
yading@10:       "add %2,%0\n"
yading@10:       "subl $2, %%ecx\n"
yading@10:       " jnz 1b\n"
yading@10: 
yading@10:       "movq %%mm6, %%mm0\n"
yading@10:       "punpcklwd %%mm7,%%mm0\n"
yading@10:       "punpckhwd %%mm7,%%mm6\n"
yading@10:       "paddd %%mm0, %%mm6\n"
yading@10: 
yading@10:       "movq %%mm6,%%mm0\n"
yading@10:       "psrlq $32, %%mm6\n"
yading@10:       "paddd %%mm6,%%mm0\n"
yading@10:       "movd %%mm0,%1\n"
yading@10:       : "+r" (pix1), "=r"(tmp)
yading@10:       : "r" ((x86_reg)line_size) , "g" (h-2)
yading@10:       : "%ecx");
yading@10:       return tmp;
yading@10: }
yading@10: 
yading@10: static int hf_noise16_mmx(uint8_t * pix1, int line_size, int h) {
yading@10:     int tmp;
yading@10:     uint8_t * pix= pix1;
yading@10:   __asm__ volatile (
yading@10:       "movl %3,%%ecx\n"
yading@10:       "pxor %%mm7,%%mm7\n"
yading@10:       "pxor %%mm6,%%mm6\n"
yading@10: 
yading@10:       "movq (%0),%%mm0\n"
yading@10:       "movq 1(%0),%%mm1\n"
yading@10:       "movq %%mm0, %%mm2\n"
yading@10:       "movq %%mm1, %%mm3\n"
yading@10:       "punpcklbw %%mm7,%%mm0\n"
yading@10:       "punpcklbw %%mm7,%%mm1\n"
yading@10:       "punpckhbw %%mm7,%%mm2\n"
yading@10:       "punpckhbw %%mm7,%%mm3\n"
yading@10:       "psubw %%mm1, %%mm0\n"
yading@10:       "psubw %%mm3, %%mm2\n"
yading@10: 
yading@10:       "add %2,%0\n"
yading@10: 
yading@10:       "movq (%0),%%mm4\n"
yading@10:       "movq 1(%0),%%mm1\n"
yading@10:       "movq %%mm4, %%mm5\n"
yading@10:       "movq %%mm1, %%mm3\n"
yading@10:       "punpcklbw %%mm7,%%mm4\n"
yading@10:       "punpcklbw %%mm7,%%mm1\n"
yading@10:       "punpckhbw %%mm7,%%mm5\n"
yading@10:       "punpckhbw %%mm7,%%mm3\n"
yading@10:       "psubw %%mm1, %%mm4\n"
yading@10:       "psubw %%mm3, %%mm5\n"
yading@10:       "psubw %%mm4, %%mm0\n"
yading@10:       "psubw %%mm5, %%mm2\n"
yading@10:       "pxor %%mm3, %%mm3\n"
yading@10:       "pxor %%mm1, %%mm1\n"
yading@10:       "pcmpgtw %%mm0, %%mm3\n\t"
yading@10:       "pcmpgtw %%mm2, %%mm1\n\t"
yading@10:       "pxor %%mm3, %%mm0\n"
yading@10:       "pxor %%mm1, %%mm2\n"
yading@10:       "psubw %%mm3, %%mm0\n"
yading@10:       "psubw %%mm1, %%mm2\n"
yading@10:       "paddw %%mm0, %%mm2\n"
yading@10:       "paddw %%mm2, %%mm6\n"
yading@10: 
yading@10:       "add %2,%0\n"
yading@10:       "1:\n"
yading@10: 
yading@10:       "movq (%0),%%mm0\n"
yading@10:       "movq 1(%0),%%mm1\n"
yading@10:       "movq %%mm0, %%mm2\n"
yading@10:       "movq %%mm1, %%mm3\n"
yading@10:       "punpcklbw %%mm7,%%mm0\n"
yading@10:       "punpcklbw %%mm7,%%mm1\n"
yading@10:       "punpckhbw %%mm7,%%mm2\n"
yading@10:       "punpckhbw %%mm7,%%mm3\n"
yading@10:       "psubw %%mm1, %%mm0\n"
yading@10:       "psubw %%mm3, %%mm2\n"
yading@10:       "psubw %%mm0, %%mm4\n"
yading@10:       "psubw %%mm2, %%mm5\n"
yading@10:       "pxor %%mm3, %%mm3\n"
yading@10:       "pxor %%mm1, %%mm1\n"
yading@10:       "pcmpgtw %%mm4, %%mm3\n\t"
yading@10:       "pcmpgtw %%mm5, %%mm1\n\t"
yading@10:       "pxor %%mm3, %%mm4\n"
yading@10:       "pxor %%mm1, %%mm5\n"
yading@10:       "psubw %%mm3, %%mm4\n"
yading@10:       "psubw %%mm1, %%mm5\n"
yading@10:       "paddw %%mm4, %%mm5\n"
yading@10:       "paddw %%mm5, %%mm6\n"
yading@10: 
yading@10:       "add %2,%0\n"
yading@10: 
yading@10:       "movq (%0),%%mm4\n"
yading@10:       "movq 1(%0),%%mm1\n"
yading@10:       "movq %%mm4, %%mm5\n"
yading@10:       "movq %%mm1, %%mm3\n"
yading@10:       "punpcklbw %%mm7,%%mm4\n"
yading@10:       "punpcklbw %%mm7,%%mm1\n"
yading@10:       "punpckhbw %%mm7,%%mm5\n"
yading@10:       "punpckhbw %%mm7,%%mm3\n"
yading@10:       "psubw %%mm1, %%mm4\n"
yading@10:       "psubw %%mm3, %%mm5\n"
yading@10:       "psubw %%mm4, %%mm0\n"
yading@10:       "psubw %%mm5, %%mm2\n"
yading@10:       "pxor %%mm3, %%mm3\n"
yading@10:       "pxor %%mm1, %%mm1\n"
yading@10:       "pcmpgtw %%mm0, %%mm3\n\t"
yading@10:       "pcmpgtw %%mm2, %%mm1\n\t"
yading@10:       "pxor %%mm3, %%mm0\n"
yading@10:       "pxor %%mm1, %%mm2\n"
yading@10:       "psubw %%mm3, %%mm0\n"
yading@10:       "psubw %%mm1, %%mm2\n"
yading@10:       "paddw %%mm0, %%mm2\n"
yading@10:       "paddw %%mm2, %%mm6\n"
yading@10: 
yading@10:       "add %2,%0\n"
yading@10:       "subl $2, %%ecx\n"
yading@10:       " jnz 1b\n"
yading@10: 
yading@10:       "movq %%mm6, %%mm0\n"
yading@10:       "punpcklwd %%mm7,%%mm0\n"
yading@10:       "punpckhwd %%mm7,%%mm6\n"
yading@10:       "paddd %%mm0, %%mm6\n"
yading@10: 
yading@10:       "movq %%mm6,%%mm0\n"
yading@10:       "psrlq $32, %%mm6\n"
yading@10:       "paddd %%mm6,%%mm0\n"
yading@10:       "movd %%mm0,%1\n"
yading@10:       : "+r" (pix1), "=r"(tmp)
yading@10:       : "r" ((x86_reg)line_size) , "g" (h-2)
yading@10:       : "%ecx");
yading@10:       return tmp + hf_noise8_mmx(pix+8, line_size, h);
yading@10: }
yading@10: 
yading@10: static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
yading@10:     MpegEncContext *c = p;
yading@10:     int score1, score2;
yading@10: 
yading@10:     if(c) score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);
yading@10:     else  score1 = sse16_mmx(c, pix1, pix2, line_size, h);
yading@10:     score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);
yading@10: 
yading@10:     if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
yading@10:     else  return score1 + FFABS(score2)*8;
yading@10: }
yading@10: 
yading@10: static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
yading@10:     MpegEncContext *c = p;
yading@10:     int score1= sse8_mmx(c, pix1, pix2, line_size, h);
yading@10:     int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);
yading@10: 
yading@10:     if(c) return score1 + FFABS(score2)*c->avctx->nsse_weight;
yading@10:     else  return score1 + FFABS(score2)*8;
yading@10: }
yading@10: 
yading@10: static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {
yading@10:     int tmp;
yading@10: 
yading@10:     av_assert2( (((int)pix) & 7) == 0);
yading@10:     av_assert2((line_size &7) ==0);
yading@10: 
yading@10: #define SUM(in0, in1, out0, out1) \
yading@10:       "movq (%0), %%mm2\n"\
yading@10:       "movq 8(%0), %%mm3\n"\
yading@10:       "add %2,%0\n"\
yading@10:       "movq %%mm2, " #out0 "\n"\
yading@10:       "movq %%mm3, " #out1 "\n"\
yading@10:       "psubusb " #in0 ", %%mm2\n"\
yading@10:       "psubusb " #in1 ", %%mm3\n"\
yading@10:       "psubusb " #out0 ", " #in0 "\n"\
yading@10:       "psubusb " #out1 ", " #in1 "\n"\
yading@10:       "por %%mm2, " #in0 "\n"\
yading@10:       "por %%mm3, " #in1 "\n"\
yading@10:       "movq " #in0 ", %%mm2\n"\
yading@10:       "movq " #in1 ", %%mm3\n"\
yading@10:       "punpcklbw %%mm7, " #in0 "\n"\
yading@10:       "punpcklbw %%mm7, " #in1 "\n"\
yading@10:       "punpckhbw %%mm7, %%mm2\n"\
yading@10:       "punpckhbw %%mm7, %%mm3\n"\
yading@10:       "paddw " #in1 ", " #in0 "\n"\
yading@10:       "paddw %%mm3, %%mm2\n"\
yading@10:       "paddw %%mm2, " #in0 "\n"\
yading@10:       "paddw " #in0 ", %%mm6\n"
yading@10: 
yading@10: 
yading@10:   __asm__ volatile (
yading@10:       "movl %3,%%ecx\n"
yading@10:       "pxor %%mm6,%%mm6\n"
yading@10:       "pxor %%mm7,%%mm7\n"
yading@10:       "movq (%0),%%mm0\n"
yading@10:       "movq 8(%0),%%mm1\n"
yading@10:       "add %2,%0\n"
yading@10:       "jmp 2f\n"
yading@10:       "1:\n"
yading@10: 
yading@10:       SUM(%%mm4, %%mm5, %%mm0, %%mm1)
yading@10:       "2:\n"
yading@10:       SUM(%%mm0, %%mm1, %%mm4, %%mm5)
yading@10: 
yading@10:       "subl $2, %%ecx\n"
yading@10:       "jnz 1b\n"
yading@10: 
yading@10:       "movq %%mm6,%%mm0\n"
yading@10:       "psrlq $32, %%mm6\n"
yading@10:       "paddw %%mm6,%%mm0\n"
yading@10:       "movq %%mm0,%%mm6\n"
yading@10:       "psrlq $16, %%mm0\n"
yading@10:       "paddw %%mm6,%%mm0\n"
yading@10:       "movd %%mm0,%1\n"
yading@10:       : "+r" (pix), "=r"(tmp)
yading@10:       : "r" ((x86_reg)line_size) , "m" (h)
yading@10:       : "%ecx");
yading@10:     return tmp & 0xFFFF;
yading@10: }
yading@10: #undef SUM
yading@10: 
yading@10: static int vsad_intra16_mmxext(void *v, uint8_t *pix, uint8_t *dummy,
yading@10:                                int line_size, int h)
yading@10: {
yading@10:     int tmp;
yading@10: 
yading@10:     av_assert2( (((int)pix) & 7) == 0);
yading@10:     av_assert2((line_size &7) ==0);
yading@10: 
yading@10: #define SUM(in0, in1, out0, out1) \
yading@10:       "movq (%0), " #out0 "\n"\
yading@10:       "movq 8(%0), " #out1 "\n"\
yading@10:       "add %2,%0\n"\
yading@10:       "psadbw " #out0 ", " #in0 "\n"\
yading@10:       "psadbw " #out1 ", " #in1 "\n"\
yading@10:       "paddw " #in1 ", " #in0 "\n"\
yading@10:       "paddw " #in0 ", %%mm6\n"
yading@10: 
yading@10:   __asm__ volatile (
yading@10:       "movl %3,%%ecx\n"
yading@10:       "pxor %%mm6,%%mm6\n"
yading@10:       "pxor %%mm7,%%mm7\n"
yading@10:       "movq (%0),%%mm0\n"
yading@10:       "movq 8(%0),%%mm1\n"
yading@10:       "add %2,%0\n"
yading@10:       "jmp 2f\n"
yading@10:       "1:\n"
yading@10: 
yading@10:       SUM(%%mm4, %%mm5, %%mm0, %%mm1)
yading@10:       "2:\n"
yading@10:       SUM(%%mm0, %%mm1, %%mm4, %%mm5)
yading@10: 
yading@10:       "subl $2, %%ecx\n"
yading@10:       "jnz 1b\n"
yading@10: 
yading@10:       "movd %%mm6,%1\n"
yading@10:       : "+r" (pix), "=r"(tmp)
yading@10:       : "r" ((x86_reg)line_size) , "m" (h)
yading@10:       : "%ecx");
yading@10:     return tmp;
yading@10: }
yading@10: #undef SUM
yading@10: 
yading@10: static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {
yading@10:     int tmp;
yading@10: 
yading@10:     av_assert2( (((int)pix1) & 7) == 0);
yading@10:     av_assert2( (((int)pix2) & 7) == 0);
yading@10:     av_assert2((line_size &7) ==0);
yading@10: 
yading@10: #define SUM(in0, in1, out0, out1) \
yading@10:       "movq (%0),%%mm2\n"\
yading@10:       "movq (%1)," #out0 "\n"\
yading@10:       "movq 8(%0),%%mm3\n"\
yading@10:       "movq 8(%1)," #out1 "\n"\
yading@10:       "add %3,%0\n"\
yading@10:       "add %3,%1\n"\
yading@10:       "psubb " #out0 ", %%mm2\n"\
yading@10:       "psubb " #out1 ", %%mm3\n"\
yading@10:       "pxor %%mm7, %%mm2\n"\
yading@10:       "pxor %%mm7, %%mm3\n"\
yading@10:       "movq %%mm2, " #out0 "\n"\
yading@10:       "movq %%mm3, " #out1 "\n"\
yading@10:       "psubusb " #in0 ", %%mm2\n"\
yading@10:       "psubusb " #in1 ", %%mm3\n"\
yading@10:       "psubusb " #out0 ", " #in0 "\n"\
yading@10:       "psubusb " #out1 ", " #in1 "\n"\
yading@10:       "por %%mm2, " #in0 "\n"\
yading@10:       "por %%mm3, " #in1 "\n"\
yading@10:       "movq " #in0 ", %%mm2\n"\
yading@10:       "movq " #in1 ", %%mm3\n"\
yading@10:       "punpcklbw %%mm7, " #in0 "\n"\
yading@10:       "punpcklbw %%mm7, " #in1 "\n"\
yading@10:       "punpckhbw %%mm7, %%mm2\n"\
yading@10:       "punpckhbw %%mm7, %%mm3\n"\
yading@10:       "paddw " #in1 ", " #in0 "\n"\
yading@10:       "paddw %%mm3, %%mm2\n"\
yading@10:       "paddw %%mm2, " #in0 "\n"\
yading@10:       "paddw " #in0 ", %%mm6\n"
yading@10: 
yading@10: 
yading@10:   __asm__ volatile (
yading@10:       "movl %4,%%ecx\n"
yading@10:       "pxor %%mm6,%%mm6\n"
yading@10:       "pcmpeqw %%mm7,%%mm7\n"
yading@10:       "psllw $15, %%mm7\n"
yading@10:       "packsswb %%mm7, %%mm7\n"
yading@10:       "movq (%0),%%mm0\n"
yading@10:       "movq (%1),%%mm2\n"
yading@10:       "movq 8(%0),%%mm1\n"
yading@10:       "movq 8(%1),%%mm3\n"
yading@10:       "add %3,%0\n"
yading@10:       "add %3,%1\n"
yading@10:       "psubb %%mm2, %%mm0\n"
yading@10:       "psubb %%mm3, %%mm1\n"
yading@10:       "pxor %%mm7, %%mm0\n"
yading@10:       "pxor %%mm7, %%mm1\n"
yading@10:       "jmp 2f\n"
yading@10:       "1:\n"
yading@10: 
yading@10:       SUM(%%mm4, %%mm5, %%mm0, %%mm1)
yading@10:       "2:\n"
yading@10:       SUM(%%mm0, %%mm1, %%mm4, %%mm5)
yading@10: 
yading@10:       "subl $2, %%ecx\n"
yading@10:       "jnz 1b\n"
yading@10: 
yading@10:       "movq %%mm6,%%mm0\n"
yading@10:       "psrlq $32, %%mm6\n"
yading@10:       "paddw %%mm6,%%mm0\n"
yading@10:       "movq %%mm0,%%mm6\n"
yading@10:       "psrlq $16, %%mm0\n"
yading@10:       "paddw %%mm6,%%mm0\n"
yading@10:       "movd %%mm0,%2\n"
yading@10:       : "+r" (pix1), "+r" (pix2), "=r"(tmp)
yading@10:       : "r" ((x86_reg)line_size) , "m" (h)
yading@10:       : "%ecx");
yading@10:     return tmp & 0x7FFF;
yading@10: }
yading@10: #undef SUM
yading@10: 
yading@10: static int vsad16_mmxext(void *v, uint8_t *pix1, uint8_t *pix2,
yading@10:                          int line_size, int h)
yading@10: {
yading@10:     int tmp;
yading@10: 
yading@10:     av_assert2( (((int)pix1) & 7) == 0);
yading@10:     av_assert2( (((int)pix2) & 7) == 0);
yading@10:     av_assert2((line_size &7) ==0);
yading@10: 
yading@10: #define SUM(in0, in1, out0, out1) \
yading@10:       "movq (%0)," #out0 "\n"\
yading@10:       "movq (%1),%%mm2\n"\
yading@10:       "movq 8(%0)," #out1 "\n"\
yading@10:       "movq 8(%1),%%mm3\n"\
yading@10:       "add %3,%0\n"\
yading@10:       "add %3,%1\n"\
yading@10:       "psubb %%mm2, " #out0 "\n"\
yading@10:       "psubb %%mm3, " #out1 "\n"\
yading@10:       "pxor %%mm7, " #out0 "\n"\
yading@10:       "pxor %%mm7, " #out1 "\n"\
yading@10:       "psadbw " #out0 ", " #in0 "\n"\
yading@10:       "psadbw " #out1 ", " #in1 "\n"\
yading@10:       "paddw " #in1 ", " #in0 "\n"\
yading@10:       "paddw " #in0 ", %%mm6\n"
yading@10: 
yading@10:   __asm__ volatile (
yading@10:       "movl %4,%%ecx\n"
yading@10:       "pxor %%mm6,%%mm6\n"
yading@10:       "pcmpeqw %%mm7,%%mm7\n"
yading@10:       "psllw $15, %%mm7\n"
yading@10:       "packsswb %%mm7, %%mm7\n"
yading@10:       "movq (%0),%%mm0\n"
yading@10:       "movq (%1),%%mm2\n"
yading@10:       "movq 8(%0),%%mm1\n"
yading@10:       "movq 8(%1),%%mm3\n"
yading@10:       "add %3,%0\n"
yading@10:       "add %3,%1\n"
yading@10:       "psubb %%mm2, %%mm0\n"
yading@10:       "psubb %%mm3, %%mm1\n"
yading@10:       "pxor %%mm7, %%mm0\n"
yading@10:       "pxor %%mm7, %%mm1\n"
yading@10:       "jmp 2f\n"
yading@10:       "1:\n"
yading@10: 
yading@10:       SUM(%%mm4, %%mm5, %%mm0, %%mm1)
yading@10:       "2:\n"
yading@10:       SUM(%%mm0, %%mm1, %%mm4, %%mm5)
yading@10: 
yading@10:       "subl $2, %%ecx\n"
yading@10:       "jnz 1b\n"
yading@10: 
yading@10:       "movd %%mm6,%2\n"
yading@10:       : "+r" (pix1), "+r" (pix2), "=r"(tmp)
yading@10:       : "r" ((x86_reg)line_size) , "m" (h)
yading@10:       : "%ecx");
yading@10:     return tmp;
yading@10: }
yading@10: #undef SUM
yading@10: 
yading@10: static void diff_bytes_mmx(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w){
yading@10:     x86_reg i=0;
yading@10:     if(w>=16)
yading@10:     __asm__ volatile(
yading@10:         "1:                             \n\t"
yading@10:         "movq  (%2, %0), %%mm0          \n\t"
yading@10:         "movq  (%1, %0), %%mm1          \n\t"
yading@10:         "psubb %%mm0, %%mm1             \n\t"
yading@10:         "movq %%mm1, (%3, %0)           \n\t"
yading@10:         "movq 8(%2, %0), %%mm0          \n\t"
yading@10:         "movq 8(%1, %0), %%mm1          \n\t"
yading@10:         "psubb %%mm0, %%mm1             \n\t"
yading@10:         "movq %%mm1, 8(%3, %0)          \n\t"
yading@10:         "add $16, %0                    \n\t"
yading@10:         "cmp %4, %0                     \n\t"
yading@10:         " jb 1b                         \n\t"
yading@10:         : "+r" (i)
yading@10:         : "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w-15)
yading@10:     );
yading@10:     for(; i<w; i++)
yading@10:         dst[i+0] = src1[i+0]-src2[i+0];
yading@10: }
yading@10: 
yading@10: static void sub_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *src1,
yading@10:                                               const uint8_t *src2, int w,
yading@10:                                               int *left, int *left_top)
yading@10: {
yading@10:     x86_reg i=0;
yading@10:     uint8_t l, lt;
yading@10: 
yading@10:     __asm__ volatile(
yading@10:         "movq  (%1, %0), %%mm0          \n\t" // LT
yading@10:         "psllq $8, %%mm0                \n\t"
yading@10:         "1:                             \n\t"
yading@10:         "movq  (%1, %0), %%mm1          \n\t" // T
yading@10:         "movq  -1(%2, %0), %%mm2        \n\t" // L
yading@10:         "movq  (%2, %0), %%mm3          \n\t" // X
yading@10:         "movq %%mm2, %%mm4              \n\t" // L
yading@10:         "psubb %%mm0, %%mm2             \n\t"
yading@10:         "paddb %%mm1, %%mm2             \n\t" // L + T - LT
yading@10:         "movq %%mm4, %%mm5              \n\t" // L
yading@10:         "pmaxub %%mm1, %%mm4            \n\t" // max(T, L)
yading@10:         "pminub %%mm5, %%mm1            \n\t" // min(T, L)
yading@10:         "pminub %%mm2, %%mm4            \n\t"
yading@10:         "pmaxub %%mm1, %%mm4            \n\t"
yading@10:         "psubb %%mm4, %%mm3             \n\t" // dst - pred
yading@10:         "movq %%mm3, (%3, %0)           \n\t"
yading@10:         "add $8, %0                     \n\t"
yading@10:         "movq -1(%1, %0), %%mm0         \n\t" // LT
yading@10:         "cmp %4, %0                     \n\t"
yading@10:         " jb 1b                         \n\t"
yading@10:         : "+r" (i)
yading@10:         : "r"(src1), "r"(src2), "r"(dst), "r"((x86_reg)w)
yading@10:     );
yading@10: 
yading@10:     l= *left;
yading@10:     lt= *left_top;
yading@10: 
yading@10:     dst[0]= src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt)&0xFF);
yading@10: 
yading@10:     *left_top= src1[w-1];
yading@10:     *left    = src2[w-1];
yading@10: }
yading@10: 
yading@10: #define MMABS_MMX(a,z)\
yading@10:     "pxor " #z ", " #z "              \n\t"\
yading@10:     "pcmpgtw " #a ", " #z "           \n\t"\
yading@10:     "pxor " #z ", " #a "              \n\t"\
yading@10:     "psubw " #z ", " #a "             \n\t"
yading@10: 
yading@10: #define MMABS_MMXEXT(a, z)                 \
yading@10:     "pxor " #z ", " #z "              \n\t"\
yading@10:     "psubw " #a ", " #z "             \n\t"\
yading@10:     "pmaxsw " #z ", " #a "            \n\t"
yading@10: 
yading@10: #define MMABS_SSSE3(a,z)\
yading@10:     "pabsw " #a ", " #a "             \n\t"
yading@10: 
yading@10: #define MMABS_SUM(a,z, sum)\
yading@10:     MMABS(a,z)\
yading@10:     "paddusw " #a ", " #sum "         \n\t"
yading@10: 
yading@10: /* FIXME: HSUM_* saturates at 64k, while an 8x8 hadamard or dct block can get up to
yading@10:  * about 100k on extreme inputs. But that's very unlikely to occur in natural video,
yading@10:  * and it's even more unlikely to not have any alternative mvs/modes with lower cost. */
yading@10: #define HSUM_MMX(a, t, dst)\
yading@10:     "movq "#a", "#t"                  \n\t"\
yading@10:     "psrlq $32, "#a"                  \n\t"\
yading@10:     "paddusw "#t", "#a"               \n\t"\
yading@10:     "movq "#a", "#t"                  \n\t"\
yading@10:     "psrlq $16, "#a"                  \n\t"\
yading@10:     "paddusw "#t", "#a"               \n\t"\
yading@10:     "movd "#a", "#dst"                \n\t"\
yading@10: 
yading@10: #define HSUM_MMXEXT(a, t, dst)             \
yading@10:     "pshufw $0x0E, "#a", "#t"         \n\t"\
yading@10:     "paddusw "#t", "#a"               \n\t"\
yading@10:     "pshufw $0x01, "#a", "#t"         \n\t"\
yading@10:     "paddusw "#t", "#a"               \n\t"\
yading@10:     "movd "#a", "#dst"                \n\t"\
yading@10: 
yading@10: #define HSUM_SSE2(a, t, dst)\
yading@10:     "movhlps "#a", "#t"               \n\t"\
yading@10:     "paddusw "#t", "#a"               \n\t"\
yading@10:     "pshuflw $0x0E, "#a", "#t"        \n\t"\
yading@10:     "paddusw "#t", "#a"               \n\t"\
yading@10:     "pshuflw $0x01, "#a", "#t"        \n\t"\
yading@10:     "paddusw "#t", "#a"               \n\t"\
yading@10:     "movd "#a", "#dst"                \n\t"\
yading@10: 
yading@10: #define DCT_SAD4(m,mm,o)\
yading@10:     "mov"#m" "#o"+ 0(%1), "#mm"2      \n\t"\
yading@10:     "mov"#m" "#o"+16(%1), "#mm"3      \n\t"\
yading@10:     "mov"#m" "#o"+32(%1), "#mm"4      \n\t"\
yading@10:     "mov"#m" "#o"+48(%1), "#mm"5      \n\t"\
yading@10:     MMABS_SUM(mm##2, mm##6, mm##0)\
yading@10:     MMABS_SUM(mm##3, mm##7, mm##1)\
yading@10:     MMABS_SUM(mm##4, mm##6, mm##0)\
yading@10:     MMABS_SUM(mm##5, mm##7, mm##1)\
yading@10: 
yading@10: #define DCT_SAD_MMX\
yading@10:     "pxor %%mm0, %%mm0                \n\t"\
yading@10:     "pxor %%mm1, %%mm1                \n\t"\
yading@10:     DCT_SAD4(q, %%mm, 0)\
yading@10:     DCT_SAD4(q, %%mm, 8)\
yading@10:     DCT_SAD4(q, %%mm, 64)\
yading@10:     DCT_SAD4(q, %%mm, 72)\
yading@10:     "paddusw %%mm1, %%mm0             \n\t"\
yading@10:     HSUM(%%mm0, %%mm1, %0)
yading@10: 
yading@10: #define DCT_SAD_SSE2\
yading@10:     "pxor %%xmm0, %%xmm0              \n\t"\
yading@10:     "pxor %%xmm1, %%xmm1              \n\t"\
yading@10:     DCT_SAD4(dqa, %%xmm, 0)\
yading@10:     DCT_SAD4(dqa, %%xmm, 64)\
yading@10:     "paddusw %%xmm1, %%xmm0           \n\t"\
yading@10:     HSUM(%%xmm0, %%xmm1, %0)
yading@10: 
yading@10: #define DCT_SAD_FUNC(cpu) \
yading@10: static int sum_abs_dctelem_##cpu(int16_t *block){\
yading@10:     int sum;\
yading@10:     __asm__ volatile(\
yading@10:         DCT_SAD\
yading@10:         :"=r"(sum)\
yading@10:         :"r"(block)\
yading@10:     );\
yading@10:     return sum&0xFFFF;\
yading@10: }
yading@10: 
yading@10: #define DCT_SAD       DCT_SAD_MMX
yading@10: #define HSUM(a,t,dst) HSUM_MMX(a,t,dst)
yading@10: #define MMABS(a,z)    MMABS_MMX(a,z)
yading@10: DCT_SAD_FUNC(mmx)
yading@10: #undef MMABS
yading@10: #undef HSUM
yading@10: 
yading@10: #define HSUM(a,t,dst) HSUM_MMXEXT(a,t,dst)
yading@10: #define MMABS(a,z)    MMABS_MMXEXT(a,z)
yading@10: DCT_SAD_FUNC(mmxext)
yading@10: #undef HSUM
yading@10: #undef DCT_SAD
yading@10: 
yading@10: #define DCT_SAD       DCT_SAD_SSE2
yading@10: #define HSUM(a,t,dst) HSUM_SSE2(a,t,dst)
yading@10: DCT_SAD_FUNC(sse2)
yading@10: #undef MMABS
yading@10: 
yading@10: #if HAVE_SSSE3_INLINE
yading@10: #define MMABS(a,z)    MMABS_SSSE3(a,z)
yading@10: DCT_SAD_FUNC(ssse3)
yading@10: #undef MMABS
yading@10: #endif
yading@10: #undef HSUM
yading@10: #undef DCT_SAD
yading@10: 
yading@10: static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int size){
yading@10:     int sum;
yading@10:     x86_reg i=size;
yading@10:     __asm__ volatile(
yading@10:         "pxor %%mm4, %%mm4 \n"
yading@10:         "1: \n"
yading@10:         "sub $8, %0 \n"
yading@10:         "movq (%2,%0), %%mm2 \n"
yading@10:         "movq (%3,%0,2), %%mm0 \n"
yading@10:         "movq 8(%3,%0,2), %%mm1 \n"
yading@10:         "punpckhbw %%mm2, %%mm3 \n"
yading@10:         "punpcklbw %%mm2, %%mm2 \n"
yading@10:         "psraw $8, %%mm3 \n"
yading@10:         "psraw $8, %%mm2 \n"
yading@10:         "psubw %%mm3, %%mm1 \n"
yading@10:         "psubw %%mm2, %%mm0 \n"
yading@10:         "pmaddwd %%mm1, %%mm1 \n"
yading@10:         "pmaddwd %%mm0, %%mm0 \n"
yading@10:         "paddd %%mm1, %%mm4 \n"
yading@10:         "paddd %%mm0, %%mm4 \n"
yading@10:         "jg 1b \n"
yading@10:         "movq %%mm4, %%mm3 \n"
yading@10:         "psrlq $32, %%mm3 \n"
yading@10:         "paddd %%mm3, %%mm4 \n"
yading@10:         "movd %%mm4, %1 \n"
yading@10:         :"+r"(i), "=r"(sum)
yading@10:         :"r"(pix1), "r"(pix2)
yading@10:     );
yading@10:     return sum;
yading@10: }
yading@10: 
yading@10: #define PHADDD(a, t)\
yading@10:     "movq "#a", "#t"                  \n\t"\
yading@10:     "psrlq $32, "#a"                  \n\t"\
yading@10:     "paddd "#t", "#a"                 \n\t"
yading@10: /*
yading@10:    pmulhw: dst[0-15]=(src[0-15]*dst[0-15])[16-31]
yading@10:    pmulhrw: dst[0-15]=(src[0-15]*dst[0-15] + 0x8000)[16-31]
yading@10:    pmulhrsw: dst[0-15]=(src[0-15]*dst[0-15] + 0x4000)[15-30]
yading@10:  */
yading@10: #define PMULHRW(x, y, s, o)\
yading@10:     "pmulhw " #s ", "#x "            \n\t"\
yading@10:     "pmulhw " #s ", "#y "            \n\t"\
yading@10:     "paddw " #o ", "#x "             \n\t"\
yading@10:     "paddw " #o ", "#y "             \n\t"\
yading@10:     "psraw $1, "#x "                 \n\t"\
yading@10:     "psraw $1, "#y "                 \n\t"
yading@10: #define DEF(x) x ## _mmx
yading@10: #define SET_RND MOVQ_WONE
yading@10: #define SCALE_OFFSET 1
yading@10: 
yading@10: #include "dsputil_qns_template.c"
yading@10: 
yading@10: #undef DEF
yading@10: #undef SET_RND
yading@10: #undef SCALE_OFFSET
yading@10: #undef PMULHRW
yading@10: 
yading@10: #define DEF(x) x ## _3dnow
yading@10: #define SET_RND(x)
yading@10: #define SCALE_OFFSET 0
yading@10: #define PMULHRW(x, y, s, o)\
yading@10:     "pmulhrw " #s ", "#x "           \n\t"\
yading@10:     "pmulhrw " #s ", "#y "           \n\t"
yading@10: 
yading@10: #include "dsputil_qns_template.c"
yading@10: 
yading@10: #undef DEF
yading@10: #undef SET_RND
yading@10: #undef SCALE_OFFSET
yading@10: #undef PMULHRW
yading@10: 
yading@10: #if HAVE_SSSE3_INLINE
yading@10: #undef PHADDD
yading@10: #define DEF(x) x ## _ssse3
yading@10: #define SET_RND(x)
yading@10: #define SCALE_OFFSET -1
yading@10: #define PHADDD(a, t)\
yading@10:     "pshufw $0x0E, "#a", "#t"         \n\t"\
yading@10:     "paddd "#t", "#a"                 \n\t" /* faster than phaddd on core2 */
yading@10: #define PMULHRW(x, y, s, o)\
yading@10:     "pmulhrsw " #s ", "#x "          \n\t"\
yading@10:     "pmulhrsw " #s ", "#y "          \n\t"
yading@10: 
yading@10: #include "dsputil_qns_template.c"
yading@10: 
yading@10: #undef DEF
yading@10: #undef SET_RND
yading@10: #undef SCALE_OFFSET
yading@10: #undef PMULHRW
yading@10: #undef PHADDD
yading@10: #endif /* HAVE_SSSE3_INLINE */
yading@10: 
yading@10: #endif /* HAVE_INLINE_ASM */
yading@10: 
yading@10: int ff_sse16_sse2(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h);
yading@10: 
yading@10: #define hadamard_func(cpu) \
yading@10: int ff_hadamard8_diff_##cpu  (void *s, uint8_t *src1, uint8_t *src2, \
yading@10:                               int stride, int h); \
yading@10: int ff_hadamard8_diff16_##cpu(void *s, uint8_t *src1, uint8_t *src2, \
yading@10:                               int stride, int h);
yading@10: 
yading@10: hadamard_func(mmx)
yading@10: hadamard_func(mmxext)
yading@10: hadamard_func(sse2)
yading@10: hadamard_func(ssse3)
yading@10: 
yading@10: av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx)
yading@10: {
yading@10:     int mm_flags = av_get_cpu_flags();
yading@10:     int bit_depth = avctx->bits_per_raw_sample;
yading@10: 
yading@10: #if HAVE_YASM
yading@10:     if (EXTERNAL_MMX(mm_flags)) {
yading@10:         if (bit_depth <= 8)
yading@10:             c->get_pixels = ff_get_pixels_mmx;
yading@10:         c->diff_pixels = ff_diff_pixels_mmx;
yading@10:         c->pix_sum = ff_pix_sum16_mmx;
yading@10: 
yading@10:         c->pix_norm1 = ff_pix_norm1_mmx;
yading@10:     }
yading@10:     if (EXTERNAL_SSE2(mm_flags))
yading@10:         if (bit_depth <= 8)
yading@10:             c->get_pixels = ff_get_pixels_sse2;
yading@10: #endif /* HAVE_YASM */
yading@10: 
yading@10: #if HAVE_INLINE_ASM
yading@10:     if (mm_flags & AV_CPU_FLAG_MMX) {
yading@10:         const int dct_algo = avctx->dct_algo;
yading@10:         if (avctx->bits_per_raw_sample <= 8 &&
yading@10:             (dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX)) {
yading@10:             if(mm_flags & AV_CPU_FLAG_SSE2){
yading@10:                 c->fdct = ff_fdct_sse2;
yading@10:             } else if (mm_flags & AV_CPU_FLAG_MMXEXT) {
yading@10:                 c->fdct = ff_fdct_mmxext;
yading@10:             }else{
yading@10:                 c->fdct = ff_fdct_mmx;
yading@10:             }
yading@10:         }
yading@10: 
yading@10: 
yading@10:         c->diff_bytes= diff_bytes_mmx;
yading@10:         c->sum_abs_dctelem= sum_abs_dctelem_mmx;
yading@10: 
yading@10:         c->sse[0] = sse16_mmx;
yading@10:         c->sse[1] = sse8_mmx;
yading@10:         c->vsad[4]= vsad_intra16_mmx;
yading@10: 
yading@10:         c->nsse[0] = nsse16_mmx;
yading@10:         c->nsse[1] = nsse8_mmx;
yading@10:         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
yading@10:             c->vsad[0] = vsad16_mmx;
yading@10:         }
yading@10: 
yading@10:         if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
yading@10:             c->try_8x8basis= try_8x8basis_mmx;
yading@10:         }
yading@10:         c->add_8x8basis= add_8x8basis_mmx;
yading@10: 
yading@10:         c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;
yading@10: 
yading@10:         if (mm_flags & AV_CPU_FLAG_MMXEXT) {
yading@10:             c->sum_abs_dctelem = sum_abs_dctelem_mmxext;
yading@10:             c->vsad[4]         = vsad_intra16_mmxext;
yading@10: 
yading@10:             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
yading@10:                 c->vsad[0] = vsad16_mmxext;
yading@10:             }
yading@10: 
yading@10:             c->sub_hfyu_median_prediction = sub_hfyu_median_prediction_mmxext;
yading@10:         }
yading@10: 
yading@10:         if(mm_flags & AV_CPU_FLAG_SSE2){
yading@10:             c->sum_abs_dctelem= sum_abs_dctelem_sse2;
yading@10:         }
yading@10: 
yading@10: #if HAVE_SSSE3_INLINE
yading@10:         if(mm_flags & AV_CPU_FLAG_SSSE3){
yading@10:             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
yading@10:                 c->try_8x8basis= try_8x8basis_ssse3;
yading@10:             }
yading@10:             c->add_8x8basis= add_8x8basis_ssse3;
yading@10:             c->sum_abs_dctelem= sum_abs_dctelem_ssse3;
yading@10:         }
yading@10: #endif
yading@10: 
yading@10:         if(mm_flags & AV_CPU_FLAG_3DNOW){
yading@10:             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
yading@10:                 c->try_8x8basis= try_8x8basis_3dnow;
yading@10:             }
yading@10:             c->add_8x8basis= add_8x8basis_3dnow;
yading@10:         }
yading@10:     }
yading@10: #endif /* HAVE_INLINE_ASM */
yading@10: 
yading@10:     if (EXTERNAL_MMX(mm_flags)) {
yading@10:         c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx;
yading@10:         c->hadamard8_diff[1] = ff_hadamard8_diff_mmx;
yading@10: 
yading@10:         if (EXTERNAL_MMXEXT(mm_flags)) {
yading@10:             c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext;
yading@10:             c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext;
yading@10:         }
yading@10: 
yading@10:         if (EXTERNAL_SSE2(mm_flags)) {
yading@10:             c->sse[0] = ff_sse16_sse2;
yading@10: 
yading@10: #if HAVE_ALIGNED_STACK
yading@10:             c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;
yading@10:             c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;
yading@10: #endif
yading@10:         }
yading@10: 
yading@10:         if (EXTERNAL_SSSE3(mm_flags) && HAVE_ALIGNED_STACK) {
yading@10:             c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;
yading@10:             c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;
yading@10:         }
yading@10:     }
yading@10: 
yading@10:     ff_dsputil_init_pix_mmx(c, avctx);
yading@10: }