dsputil_mmx_rnd.h
资源名称:tcpmp.rar [点击查看]
上传用户:wstnjxml
上传日期:2014-04-03
资源大小:7248k
文件大小:16k
源码类别:
Windows CE
开发平台:
C/C++
- /*
- * DSP utils mmx functions are compiled twice for rnd/no_rnd
- * Copyright (c) 2000, 2001 Fabrice Bellard.
- * Copyright (c) 2003-2004 Michael Niedermayer <michaelni@gmx.at>
- *
- * 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>
- */
- // put_pixels
- static void DEF(put, pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- __asm __volatile(
- "lea (%3, %3), %%"REG_a" nt"
- ".balign 8 nt"
- "1: nt"
- "movq (%1), %%mm0 nt"
- "movq 1(%1), %%mm1 nt"
- "movq (%1, %3), %%mm2 nt"
- "movq 1(%1, %3), %%mm3 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%2) nt"
- "movq %%mm5, (%2, %3) nt"
- "add %%"REG_a", %1 nt"
- "add %%"REG_a", %2 nt"
- "movq (%1), %%mm0 nt"
- "movq 1(%1), %%mm1 nt"
- "movq (%1, %3), %%mm2 nt"
- "movq 1(%1, %3), %%mm3 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%2) nt"
- "movq %%mm5, (%2, %3) nt"
- "add %%"REG_a", %1 nt"
- "add %%"REG_a", %2 nt"
- "subl $4, %0 nt"
- "jnz 1b nt"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r"((long)line_size)
- :REG_a, "memory");
- }
- static void attribute_unused DEF(put, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- MOVQ_BFE(mm6);
- __asm __volatile(
- "testl $1, %0 nt"
- " jz 1f nt"
- "movq (%1), %%mm0 nt"
- "movq (%2), %%mm1 nt"
- "add %4, %1 nt"
- "add $8, %2 nt"
- PAVGB(%%mm0, %%mm1, %%mm4, %%mm6)
- "movq %%mm4, (%3) nt"
- "add %5, %3 nt"
- "decl %0 nt"
- ".balign 8 nt"
- "1: nt"
- "movq (%1), %%mm0 nt"
- "movq (%2), %%mm1 nt"
- "add %4, %1 nt"
- "movq (%1), %%mm2 nt"
- "movq 8(%2), %%mm3 nt"
- "add %4, %1 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%3) nt"
- "add %5, %3 nt"
- "movq %%mm5, (%3) nt"
- "add %5, %3 nt"
- "movq (%1), %%mm0 nt"
- "movq 16(%2), %%mm1 nt"
- "add %4, %1 nt"
- "movq (%1), %%mm2 nt"
- "movq 24(%2), %%mm3 nt"
- "add %4, %1 nt"
- "add $32, %2 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%3) nt"
- "add %5, %3 nt"
- "movq %%mm5, (%3) nt"
- "add %5, %3 nt"
- "subl $4, %0 nt"
- "jnz 1b nt"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((long)src1Stride), "D"((long)dstStride)
- :"memory");
- }
- static void DEF(put, pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- __asm __volatile(
- "lea (%3, %3), %%"REG_a" nt"
- ".balign 8 nt"
- "1: nt"
- "movq (%1), %%mm0 nt"
- "movq 1(%1), %%mm1 nt"
- "movq (%1, %3), %%mm2 nt"
- "movq 1(%1, %3), %%mm3 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%2) nt"
- "movq %%mm5, (%2, %3) nt"
- "movq 8(%1), %%mm0 nt"
- "movq 9(%1), %%mm1 nt"
- "movq 8(%1, %3), %%mm2 nt"
- "movq 9(%1, %3), %%mm3 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, 8(%2) nt"
- "movq %%mm5, 8(%2, %3) nt"
- "add %%"REG_a", %1 nt"
- "add %%"REG_a", %2 nt"
- "movq (%1), %%mm0 nt"
- "movq 1(%1), %%mm1 nt"
- "movq (%1, %3), %%mm2 nt"
- "movq 1(%1, %3), %%mm3 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%2) nt"
- "movq %%mm5, (%2, %3) nt"
- "movq 8(%1), %%mm0 nt"
- "movq 9(%1), %%mm1 nt"
- "movq 8(%1, %3), %%mm2 nt"
- "movq 9(%1, %3), %%mm3 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, 8(%2) nt"
- "movq %%mm5, 8(%2, %3) nt"
- "add %%"REG_a", %1 nt"
- "add %%"REG_a", %2 nt"
- "subl $4, %0 nt"
- "jnz 1b nt"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r"((long)line_size)
- :REG_a, "memory");
- }
- static void attribute_unused DEF(put, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- MOVQ_BFE(mm6);
- __asm __volatile(
- "testl $1, %0 nt"
- " jz 1f nt"
- "movq (%1), %%mm0 nt"
- "movq (%2), %%mm1 nt"
- "movq 8(%1), %%mm2 nt"
- "movq 8(%2), %%mm3 nt"
- "add %4, %1 nt"
- "add $16, %2 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%3) nt"
- "movq %%mm5, 8(%3) nt"
- "add %5, %3 nt"
- "decl %0 nt"
- ".balign 8 nt"
- "1: nt"
- "movq (%1), %%mm0 nt"
- "movq (%2), %%mm1 nt"
- "movq 8(%1), %%mm2 nt"
- "movq 8(%2), %%mm3 nt"
- "add %4, %1 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%3) nt"
- "movq %%mm5, 8(%3) nt"
- "add %5, %3 nt"
- "movq (%1), %%mm0 nt"
- "movq 16(%2), %%mm1 nt"
- "movq 8(%1), %%mm2 nt"
- "movq 24(%2), %%mm3 nt"
- "add %4, %1 nt"
- PAVGBP(%%mm0, %%mm1, %%mm4, %%mm2, %%mm3, %%mm5)
- "movq %%mm4, (%3) nt"
- "movq %%mm5, 8(%3) nt"
- "add %5, %3 nt"
- "add $32, %2 nt"
- "subl $2, %0 nt"
- "jnz 1b nt"
- #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used
- :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #else
- :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
- #endif
- :"S"((long)src1Stride), "D"((long)dstStride)
- :"memory");
- }
- static void DEF(put, pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- __asm __volatile(
- "lea (%3, %3), %%"REG_a" nt"
- "movq (%1), %%mm0 nt"
- ".balign 8 nt"
- "1: nt"
- "movq (%1, %3), %%mm1 nt"
- "movq (%1, %%"REG_a"),%%mm2 nt"
- PAVGBP(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5)
- "movq %%mm4, (%2) nt"
- "movq %%mm5, (%2, %3) nt"
- "add %%"REG_a", %1 nt"
- "add %%"REG_a", %2 nt"
- "movq (%1, %3), %%mm1 nt"
- "movq (%1, %%"REG_a"),%%mm0 nt"
- PAVGBP(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5)
- "movq %%mm4, (%2) nt"
- "movq %%mm5, (%2, %3) nt"
- "add %%"REG_a", %1 nt"
- "add %%"REG_a", %2 nt"
- "subl $4, %0 nt"
- "jnz 1b nt"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r"((long)line_size)
- :REG_a, "memory");
- }
- static void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_ZERO(mm7);
- SET_RND(mm6); // =2 for rnd and =1 for no_rnd version
- __asm __volatile(
- "movq (%1), %%mm0 nt"
- "movq 1(%1), %%mm4 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm4, %%mm5 nt"
- "punpcklbw %%mm7, %%mm0 nt"
- "punpcklbw %%mm7, %%mm4 nt"
- "punpckhbw %%mm7, %%mm1 nt"
- "punpckhbw %%mm7, %%mm5 nt"
- "paddusw %%mm0, %%mm4 nt"
- "paddusw %%mm1, %%mm5 nt"
- "xor %%"REG_a", %%"REG_a" nt"
- "add %3, %1 nt"
- ".balign 8 nt"
- "1: nt"
- "movq (%1, %%"REG_a"), %%mm0 nt"
- "movq 1(%1, %%"REG_a"), %%mm2 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm2, %%mm3 nt"
- "punpcklbw %%mm7, %%mm0 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpckhbw %%mm7, %%mm1 nt"
- "punpckhbw %%mm7, %%mm3 nt"
- "paddusw %%mm2, %%mm0 nt"
- "paddusw %%mm3, %%mm1 nt"
- "paddusw %%mm6, %%mm4 nt"
- "paddusw %%mm6, %%mm5 nt"
- "paddusw %%mm0, %%mm4 nt"
- "paddusw %%mm1, %%mm5 nt"
- "psrlw $2, %%mm4 nt"
- "psrlw $2, %%mm5 nt"
- "packuswb %%mm5, %%mm4 nt"
- "movq %%mm4, (%2, %%"REG_a") nt"
- "add %3, %%"REG_a" nt"
- "movq (%1, %%"REG_a"), %%mm2 nt" // 0 <-> 2 1 <-> 3
- "movq 1(%1, %%"REG_a"), %%mm4 nt"
- "movq %%mm2, %%mm3 nt"
- "movq %%mm4, %%mm5 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpcklbw %%mm7, %%mm4 nt"
- "punpckhbw %%mm7, %%mm3 nt"
- "punpckhbw %%mm7, %%mm5 nt"
- "paddusw %%mm2, %%mm4 nt"
- "paddusw %%mm3, %%mm5 nt"
- "paddusw %%mm6, %%mm0 nt"
- "paddusw %%mm6, %%mm1 nt"
- "paddusw %%mm4, %%mm0 nt"
- "paddusw %%mm5, %%mm1 nt"
- "psrlw $2, %%mm0 nt"
- "psrlw $2, %%mm1 nt"
- "packuswb %%mm1, %%mm0 nt"
- "movq %%mm0, (%2, %%"REG_a") nt"
- "add %3, %%"REG_a" nt"
- "subl $2, %0 nt"
- "jnz 1b nt"
- :"+g"(h), "+S"(pixels)
- :"D"(block), "r"((long)line_size)
- :REG_a, "memory");
- }
- // avg_pixels
- static void attribute_unused DEF(avg, pixels4)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- JUMPALIGN();
- do {
- __asm __volatile(
- "movd %0, %%mm0 nt"
- "movd %1, %%mm1 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- "movd %%mm2, %0 nt"
- :"+m"(*block)
- :"m"(*pixels)
- :"memory");
- pixels += line_size;
- block += line_size;
- }
- while (--h);
- }
- // in case more speed is needed - unroling would certainly help
- static void DEF(avg, pixels8)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- JUMPALIGN();
- do {
- __asm __volatile(
- "movq %0, %%mm0 nt"
- "movq %1, %%mm1 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- "movq %%mm2, %0 nt"
- :"+m"(*block)
- :"m"(*pixels)
- :"memory");
- pixels += line_size;
- block += line_size;
- }
- while (--h);
- }
- static void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- JUMPALIGN();
- do {
- __asm __volatile(
- "movq %0, %%mm0 nt"
- "movq %1, %%mm1 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- "movq %%mm2, %0 nt"
- "movq 8%0, %%mm0 nt"
- "movq 8%1, %%mm1 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- "movq %%mm2, 8%0 nt"
- :"+m"(*block)
- :"m"(*pixels)
- :"memory");
- pixels += line_size;
- block += line_size;
- }
- while (--h);
- }
- static void DEF(avg, pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- JUMPALIGN();
- do {
- __asm __volatile(
- "movq %1, %%mm0 nt"
- "movq 1%1, %%mm1 nt"
- "movq %0, %%mm3 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
- "movq %%mm0, %0 nt"
- :"+m"(*block)
- :"m"(*pixels)
- :"memory");
- pixels += line_size;
- block += line_size;
- } while (--h);
- }
- static __attribute__((unused)) void DEF(avg, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- MOVQ_BFE(mm6);
- JUMPALIGN();
- do {
- __asm __volatile(
- "movq %1, %%mm0 nt"
- "movq %2, %%mm1 nt"
- "movq %0, %%mm3 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
- "movq %%mm0, %0 nt"
- :"+m"(*dst)
- :"m"(*src1), "m"(*src2)
- :"memory");
- dst += dstStride;
- src1 += src1Stride;
- src2 += 8;
- } while (--h);
- }
- static void DEF(avg, pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- JUMPALIGN();
- do {
- __asm __volatile(
- "movq %1, %%mm0 nt"
- "movq 1%1, %%mm1 nt"
- "movq %0, %%mm3 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
- "movq %%mm0, %0 nt"
- "movq 8%1, %%mm0 nt"
- "movq 9%1, %%mm1 nt"
- "movq 8%0, %%mm3 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
- "movq %%mm0, 8%0 nt"
- :"+m"(*block)
- :"m"(*pixels)
- :"memory");
- pixels += line_size;
- block += line_size;
- } while (--h);
- }
- static __attribute__((unused)) void DEF(avg, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
- {
- MOVQ_BFE(mm6);
- JUMPALIGN();
- do {
- __asm __volatile(
- "movq %1, %%mm0 nt"
- "movq %2, %%mm1 nt"
- "movq %0, %%mm3 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
- "movq %%mm0, %0 nt"
- "movq 8%1, %%mm0 nt"
- "movq 8%2, %%mm1 nt"
- "movq 8%0, %%mm3 nt"
- PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
- PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
- "movq %%mm0, 8%0 nt"
- :"+m"(*dst)
- :"m"(*src1), "m"(*src2)
- :"memory");
- dst += dstStride;
- src1 += src1Stride;
- src2 += 16;
- } while (--h);
- }
- static void DEF(avg, pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_BFE(mm6);
- __asm __volatile(
- "lea (%3, %3), %%"REG_a" nt"
- "movq (%1), %%mm0 nt"
- ".balign 8 nt"
- "1: nt"
- "movq (%1, %3), %%mm1 nt"
- "movq (%1, %%"REG_a"), %%mm2 nt"
- PAVGBP(%%mm1, %%mm0, %%mm4, %%mm2, %%mm1, %%mm5)
- "movq (%2), %%mm3 nt"
- PAVGB(%%mm3, %%mm4, %%mm0, %%mm6)
- "movq (%2, %3), %%mm3 nt"
- PAVGB(%%mm3, %%mm5, %%mm1, %%mm6)
- "movq %%mm0, (%2) nt"
- "movq %%mm1, (%2, %3) nt"
- "add %%"REG_a", %1 nt"
- "add %%"REG_a", %2 nt"
- "movq (%1, %3), %%mm1 nt"
- "movq (%1, %%"REG_a"), %%mm0 nt"
- PAVGBP(%%mm1, %%mm2, %%mm4, %%mm0, %%mm1, %%mm5)
- "movq (%2), %%mm3 nt"
- PAVGB(%%mm3, %%mm4, %%mm2, %%mm6)
- "movq (%2, %3), %%mm3 nt"
- PAVGB(%%mm3, %%mm5, %%mm1, %%mm6)
- "movq %%mm2, (%2) nt"
- "movq %%mm1, (%2, %3) nt"
- "add %%"REG_a", %1 nt"
- "add %%"REG_a", %2 nt"
- "subl $4, %0 nt"
- "jnz 1b nt"
- :"+g"(h), "+S"(pixels), "+D"(block)
- :"r"((long)line_size)
- :REG_a, "memory");
- }
- // this routine is 'slightly' suboptimal but mostly unused
- static void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)
- {
- MOVQ_ZERO(mm7);
- SET_RND(mm6); // =2 for rnd and =1 for no_rnd version
- __asm __volatile(
- "movq (%1), %%mm0 nt"
- "movq 1(%1), %%mm4 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm4, %%mm5 nt"
- "punpcklbw %%mm7, %%mm0 nt"
- "punpcklbw %%mm7, %%mm4 nt"
- "punpckhbw %%mm7, %%mm1 nt"
- "punpckhbw %%mm7, %%mm5 nt"
- "paddusw %%mm0, %%mm4 nt"
- "paddusw %%mm1, %%mm5 nt"
- "xor %%"REG_a", %%"REG_a" nt"
- "add %3, %1 nt"
- ".balign 8 nt"
- "1: nt"
- "movq (%1, %%"REG_a"), %%mm0 nt"
- "movq 1(%1, %%"REG_a"), %%mm2 nt"
- "movq %%mm0, %%mm1 nt"
- "movq %%mm2, %%mm3 nt"
- "punpcklbw %%mm7, %%mm0 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpckhbw %%mm7, %%mm1 nt"
- "punpckhbw %%mm7, %%mm3 nt"
- "paddusw %%mm2, %%mm0 nt"
- "paddusw %%mm3, %%mm1 nt"
- "paddusw %%mm6, %%mm4 nt"
- "paddusw %%mm6, %%mm5 nt"
- "paddusw %%mm0, %%mm4 nt"
- "paddusw %%mm1, %%mm5 nt"
- "psrlw $2, %%mm4 nt"
- "psrlw $2, %%mm5 nt"
- "movq (%2, %%"REG_a"), %%mm3 nt"
- "packuswb %%mm5, %%mm4 nt"
- "pcmpeqd %%mm2, %%mm2 nt"
- "paddb %%mm2, %%mm2 nt"
- PAVGB(%%mm3, %%mm4, %%mm5, %%mm2)
- "movq %%mm5, (%2, %%"REG_a") nt"
- "add %3, %%"REG_a" nt"
- "movq (%1, %%"REG_a"), %%mm2 nt" // 0 <-> 2 1 <-> 3
- "movq 1(%1, %%"REG_a"), %%mm4 nt"
- "movq %%mm2, %%mm3 nt"
- "movq %%mm4, %%mm5 nt"
- "punpcklbw %%mm7, %%mm2 nt"
- "punpcklbw %%mm7, %%mm4 nt"
- "punpckhbw %%mm7, %%mm3 nt"
- "punpckhbw %%mm7, %%mm5 nt"
- "paddusw %%mm2, %%mm4 nt"
- "paddusw %%mm3, %%mm5 nt"
- "paddusw %%mm6, %%mm0 nt"
- "paddusw %%mm6, %%mm1 nt"
- "paddusw %%mm4, %%mm0 nt"
- "paddusw %%mm5, %%mm1 nt"
- "psrlw $2, %%mm0 nt"
- "psrlw $2, %%mm1 nt"
- "movq (%2, %%"REG_a"), %%mm3 nt"
- "packuswb %%mm1, %%mm0 nt"
- "pcmpeqd %%mm2, %%mm2 nt"
- "paddb %%mm2, %%mm2 nt"
- PAVGB(%%mm3, %%mm0, %%mm1, %%mm2)
- "movq %%mm1, (%2, %%"REG_a") nt"
- "add %3, %%"REG_a" nt"
- "subl $2, %0 nt"
- "jnz 1b nt"
- :"+g"(h), "+S"(pixels)
- :"D"(block), "r"((long)line_size)
- :REG_a, "memory");
- }
- //FIXME optimize
- static void DEF(put, pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(put, pixels8_y2)(block , pixels , line_size, h);
- DEF(put, pixels8_y2)(block+8, pixels+8, line_size, h);
- }
- static void DEF(put, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(put, pixels8_xy2)(block , pixels , line_size, h);
- DEF(put, pixels8_xy2)(block+8, pixels+8, line_size, h);
- }
- static void DEF(avg, pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(avg, pixels8_y2)(block , pixels , line_size, h);
- DEF(avg, pixels8_y2)(block+8, pixels+8, line_size, h);
- }
- static void DEF(avg, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){
- DEF(avg, pixels8_xy2)(block , pixels , line_size, h);
- DEF(avg, pixels8_xy2)(block+8, pixels+8, line_size, h);
- }