comparison postproc/rgb2rgb_template.c @ 11072:3254b413ef1c

yv12->uyvy MMX
author michael
date Fri, 10 Oct 2003 00:19:16 +0000
parents f33f908ae085
children b5c106b694e4
comparison
equal deleted inserted replaced
11071:9db8700a5a16 11072:3254b413ef1c
1574 { 1574 {
1575 unsigned y; 1575 unsigned y;
1576 const unsigned chromWidth= width>>1; 1576 const unsigned chromWidth= width>>1;
1577 for(y=0; y<height; y++) 1577 for(y=0; y<height; y++)
1578 { 1578 {
1579 #ifdef HAVE_MMX
1580 //FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)
1581 asm volatile(
1582 "xorl %%eax, %%eax \n\t"
1583 ".balign 16 \n\t"
1584 "1: \n\t"
1585 PREFETCH" 32(%1, %%eax, 2) \n\t"
1586 PREFETCH" 32(%2, %%eax) \n\t"
1587 PREFETCH" 32(%3, %%eax) \n\t"
1588 "movq (%2, %%eax), %%mm0 \n\t" // U(0)
1589 "movq %%mm0, %%mm2 \n\t" // U(0)
1590 "movq (%3, %%eax), %%mm1 \n\t" // V(0)
1591 "punpcklbw %%mm1, %%mm0 \n\t" // UVUV UVUV(0)
1592 "punpckhbw %%mm1, %%mm2 \n\t" // UVUV UVUV(8)
1593
1594 "movq (%1, %%eax,2), %%mm3 \n\t" // Y(0)
1595 "movq 8(%1, %%eax,2), %%mm5 \n\t" // Y(8)
1596 "movq %%mm0, %%mm4 \n\t" // Y(0)
1597 "movq %%mm2, %%mm6 \n\t" // Y(8)
1598 "punpcklbw %%mm3, %%mm0 \n\t" // YUYV YUYV(0)
1599 "punpckhbw %%mm3, %%mm4 \n\t" // YUYV YUYV(4)
1600 "punpcklbw %%mm5, %%mm2 \n\t" // YUYV YUYV(8)
1601 "punpckhbw %%mm5, %%mm6 \n\t" // YUYV YUYV(12)
1602
1603 MOVNTQ" %%mm0, (%0, %%eax, 4) \n\t"
1604 MOVNTQ" %%mm4, 8(%0, %%eax, 4) \n\t"
1605 MOVNTQ" %%mm2, 16(%0, %%eax, 4) \n\t"
1606 MOVNTQ" %%mm6, 24(%0, %%eax, 4) \n\t"
1607
1608 "addl $8, %%eax \n\t"
1609 "cmpl %4, %%eax \n\t"
1610 " jb 1b \n\t"
1611 ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)
1612 : "%eax"
1613 );
1614 #else
1615 //FIXME adapt the alpha asm code from yv12->yuy2
1616
1579 #if __WORDSIZE >= 64 1617 #if __WORDSIZE >= 64
1580 int i; 1618 int i;
1581 uint64_t *ldst = (uint64_t *) dst; 1619 uint64_t *ldst = (uint64_t *) dst;
1582 const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc; 1620 const uint8_t *yc = ysrc, *uc = usrc, *vc = vsrc;
1583 for(i = 0; i < chromWidth; i += 2){ 1621 for(i = 0; i < chromWidth; i += 2){
1601 yc += 2; 1639 yc += 2;
1602 uc++; 1640 uc++;
1603 vc++; 1641 vc++;
1604 } 1642 }
1605 #endif 1643 #endif
1644 #endif
1606 if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) ) 1645 if((y&(vertLumPerChroma-1))==(vertLumPerChroma-1) )
1607 { 1646 {
1608 usrc += chromStride; 1647 usrc += chromStride;
1609 vsrc += chromStride; 1648 vsrc += chromStride;
1610 } 1649 }
1611 ysrc += lumStride; 1650 ysrc += lumStride;
1612 dst += dstStride; 1651 dst += dstStride;
1613 } 1652 }
1653 #ifdef HAVE_MMX
1654 asm( EMMS" \n\t"
1655 SFENCE" \n\t"
1656 :::"memory");
1657 #endif
1614 } 1658 }
1615 1659
1616 /** 1660 /**
1617 * 1661 *
1618 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a 1662 * height should be a multiple of 2 and width should be a multiple of 16 (if this is a