Mercurial > libavcodec.hg
view i386/dsputil_mmx_avg.h @ 480:a7cbee351b55 libavcodec
added ff_idct_put/add
author | bellard |
---|---|
date | Wed, 05 Jun 2002 18:48:07 +0000 |
parents | e8c8ca9106aa |
children | 45e8f39fda50 |
line wrap: on
line source
/* * DSP utils : average functions are compiled twice for 3dnow/mmx2 * Copyright (c) 2000, 2001 Fabrice Bellard. * Copyright (c) 2002 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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * * MMX optimization by Nick Kurshev <nickols_k@mail.ru> * mostly rewritten by Michael Niedermayer <michaelni@gmx.at> * and improved by Zdenek Kabelac <kabi@users.sf.net> */ /* XXX: we use explicit registers to avoid a gcc 2.95.2 register asm clobber bug - now it will work with 2.95.2 and also with -fPIC */ static void DEF(put_pixels_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { __asm __volatile( "lea (%3, %3), %%eax \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" "movq %%mm0, (%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "addl %%eax, %1 \n\t" "addl %%eax, %2 \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" "addl %%eax, %1 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" :"+g"(h), "+S"(pixels), "+D"(block) :"r" (line_size) :"%eax", "memory"); } /* GL: this function does incorrect rounding if overflow */ static void DEF(put_no_rnd_pixels_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { MOVQ_BONE(mm6); __asm __volatile( "lea (%3, %3), %%eax \n\t" "1: \n\t" "movq (%1), %%mm0 \n\t" "movq (%1, %3), %%mm2 \n\t" "movq 1(%1), %%mm1 \n\t" "movq 1(%1, %3), %%mm3 \n\t" "addl %%eax, %1 \n\t" "psubusb %%mm6, %%mm0 \n\t" "psubusb %%mm6, %%mm2 \n\t" PAVGB" %%mm1, %%mm0 \n\t" PAVGB" %%mm3, %%mm2 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm2, (%2, %3) \n\t" "movq (%1), %%mm0 \n\t" "movq 1(%1), %%mm1 \n\t" "movq (%1, %3), %%mm2 \n\t" "movq 1(%1, %3), %%mm3 \n\t" "addl %%eax, %2 \n\t" "addl %%eax, %1 \n\t" "psubusb %%mm6, %%mm0 \n\t" "psubusb %%mm6, %%mm2 \n\t" PAVGB" %%mm1, %%mm0 \n\t" PAVGB" %%mm3, %%mm2 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm2, (%2, %3) \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" :"+g"(h), "+S"(pixels), "+D"(block) :"r" (line_size) :"%eax", "memory"); } static void DEF(put_pixels_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { __asm __volatile( "lea (%3, %3), %%eax \n\t" "movq (%1), %%mm0 \n\t" "subl %3, %2 \n\t" "1: \n\t" "movq (%1, %3), %%mm1 \n\t" "movq (%1, %%eax), %%mm2 \n\t" "addl %%eax, %1 \n\t" PAVGB" %%mm1, %%mm0 \n\t" PAVGB" %%mm2, %%mm1 \n\t" "movq %%mm0, (%2, %3) \n\t" "movq %%mm1, (%2, %%eax) \n\t" "movq (%1, %3), %%mm1 \n\t" "movq (%1, %%eax), %%mm0 \n\t" "addl %%eax, %2 \n\t" "addl %%eax, %1 \n\t" PAVGB" %%mm1, %%mm2 \n\t" PAVGB" %%mm0, %%mm1 \n\t" "movq %%mm2, (%2, %3) \n\t" "movq %%mm1, (%2, %%eax) \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" :"+g"(h), "+S"(pixels), "+D" (block) :"r" (line_size) :"%eax", "memory"); } /* GL: this function does incorrect rounding if overflow */ static void DEF(put_no_rnd_pixels_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { MOVQ_BONE(mm6); __asm __volatile( "lea (%3, %3), %%eax \n\t" "movq (%1), %%mm0 \n\t" "subl %3, %2 \n\t" "1: \n\t" "movq (%1, %3), %%mm1 \n\t" "movq (%1, %%eax), %%mm2 \n\t" "addl %%eax, %1 \n\t" "psubusb %%mm6, %%mm1 \n\t" PAVGB" %%mm1, %%mm0 \n\t" PAVGB" %%mm2, %%mm1 \n\t" "movq %%mm0, (%2, %3) \n\t" "movq %%mm1, (%2, %%eax) \n\t" "movq (%1, %3), %%mm1 \n\t" "movq (%1, %%eax), %%mm0 \n\t" "addl %%eax, %2 \n\t" "addl %%eax, %1 \n\t" "psubusb %%mm6, %%mm1 \n\t" PAVGB" %%mm1, %%mm2 \n\t" PAVGB" %%mm0, %%mm1 \n\t" "movq %%mm2, (%2, %3) \n\t" "movq %%mm1, (%2, %%eax) \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" :"+g"(h), "+S"(pixels), "+D" (block) :"r" (line_size) :"%eax", "memory"); } static void DEF(avg_pixels)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { __asm __volatile( "lea (%3, %3), %%eax \n\t" "1: \n\t" "movq (%2), %%mm0 \n\t" "movq (%2, %3), %%mm1 \n\t" PAVGB" (%1), %%mm0 \n\t" PAVGB" (%1, %3), %%mm1 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "addl %%eax, %1 \n\t" "addl %%eax, %2 \n\t" "movq (%2), %%mm0 \n\t" "movq (%2, %3), %%mm1 \n\t" PAVGB" (%1), %%mm0 \n\t" PAVGB" (%1, %3), %%mm1 \n\t" "addl %%eax, %1 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" :"+g"(h), "+S"(pixels), "+D"(block) :"r" (line_size) :"%eax", "memory"); } static void DEF(avg_pixels_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { __asm __volatile( "lea (%3, %3), %%eax \n\t" "1: \n\t" "movq (%1), %%mm0 \n\t" "movq (%1, %3), %%mm2 \n\t" PAVGB" 1(%1), %%mm0 \n\t" PAVGB" 1(%1, %3), %%mm2 \n\t" PAVGB" (%2), %%mm0 \n\t" PAVGB" (%2, %3), %%mm2 \n\t" "addl %%eax, %1 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm2, (%2, %3) \n\t" "movq (%1), %%mm0 \n\t" "movq (%1, %3), %%mm2 \n\t" PAVGB" 1(%1), %%mm0 \n\t" PAVGB" 1(%1, %3), %%mm2 \n\t" "addl %%eax, %2 \n\t" "addl %%eax, %1 \n\t" PAVGB" (%2), %%mm0 \n\t" PAVGB" (%2, %3), %%mm2 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm2, (%2, %3) \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" :"+g"(h), "+S"(pixels), "+D"(block) :"r" (line_size) :"%eax", "memory"); } static void DEF(avg_pixels_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { __asm __volatile( "lea (%3, %3), %%eax \n\t" "movq (%1), %%mm0 \n\t" "subl %3, %2 \n\t" "1: \n\t" "movq (%1, %3), %%mm1 \n\t" "movq (%1, %%eax), %%mm2 \n\t" "addl %%eax, %1 \n\t" PAVGB" %%mm1, %%mm0 \n\t" PAVGB" %%mm2, %%mm1 \n\t" "movq (%2, %3), %%mm3 \n\t" "movq (%2, %%eax), %%mm4 \n\t" PAVGB" %%mm3, %%mm0 \n\t" PAVGB" %%mm4, %%mm1 \n\t" "movq %%mm0, (%2, %3) \n\t" "movq %%mm1, (%2, %%eax) \n\t" "movq (%1, %3), %%mm1 \n\t" "movq (%1, %%eax), %%mm0 \n\t" PAVGB" %%mm1, %%mm2 \n\t" PAVGB" %%mm0, %%mm1 \n\t" "addl %%eax, %2 \n\t" "addl %%eax, %1 \n\t" "movq (%2, %3), %%mm3 \n\t" "movq (%2, %%eax), %%mm4 \n\t" PAVGB" %%mm3, %%mm2 \n\t" PAVGB" %%mm4, %%mm1 \n\t" "movq %%mm2, (%2, %3) \n\t" "movq %%mm1, (%2, %%eax) \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" :"+g"(h), "+S"(pixels), "+D"(block) :"r" (line_size) :"%eax", "memory"); } // Note this is not correctly rounded, but this function is only used for b frames so it doesnt matter static void DEF(avg_pixels_xy2)(UINT8 *block, const UINT8 *pixels, int line_size, int h) { MOVQ_BONE(mm6); __asm __volatile( "lea (%3, %3), %%eax \n\t" "movq (%1), %%mm0 \n\t" PAVGB" 1(%1), %%mm0 \n\t" ".balign 8 \n\t" "1: \n\t" "movq (%1, %%eax), %%mm2 \n\t" "movq (%1, %3), %%mm1 \n\t" "psubusb %%mm6, %%mm2 \n\t" PAVGB" 1(%1, %3), %%mm1 \n\t" PAVGB" 1(%1, %%eax), %%mm2 \n\t" "addl %%eax, %1 \n\t" PAVGB" %%mm1, %%mm0 \n\t" PAVGB" %%mm2, %%mm1 \n\t" PAVGB" (%2), %%mm0 \n\t" PAVGB" (%2, %3), %%mm1 \n\t" "movq %%mm0, (%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "movq (%1, %3), %%mm1 \n\t" "movq (%1, %%eax), %%mm0 \n\t" PAVGB" 1(%1, %3), %%mm1 \n\t" PAVGB" 1(%1, %%eax), %%mm0 \n\t" "addl %%eax, %2 \n\t" "addl %%eax, %1 \n\t" PAVGB" %%mm1, %%mm2 \n\t" PAVGB" %%mm0, %%mm1 \n\t" PAVGB" (%2), %%mm2 \n\t" PAVGB" (%2, %3), %%mm1 \n\t" "movq %%mm2, (%2) \n\t" "movq %%mm1, (%2, %3) \n\t" "addl %%eax, %2 \n\t" "subl $4, %0 \n\t" "jnz 1b \n\t" :"+g"(h), "+S"(pixels), "+D"(block) :"r" (line_size) :"%eax", "memory"); }