Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

MMX & MMX2 optimizations (MMX2 is buggy and commented out)

Browse Source 
32, 24, 16, 15 bit support in C (only 32 & 16 tested)
32, 16 bit support in mmx (16 with dithering)
ranges of some variables changed so mmx likes them more

Originally committed as revision 2233 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc
pull/126/head
Michael Niedermayer 24 years ago
parent d4cb9e70a5
commit d3f41512a8
2 changed files with 1126 additions and 72 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 599
      postproc/swscale.c
  2. 599
      postproc/swscale_template.c

599
postproc/swscale.c
Unescape View File

@ -1,9 +1,18 @@
// Software scaling and colorspace conversion routines for MPlayer
#include <inttypes.h>
#include "../config.h"
#undef HAVE_MMX2 //code is buggy
//#undef HAVE_MMX
#define RET 0xC3 //near return opcode
// temporary storage for 4 yuv lines:
static unsigned int pix_buf_y[4][2048];
static unsigned int pix_buf_uv[2][2048*2];
// 16bit for now (mmx likes it more compact)
static uint16_t pix_buf_y[4][2048];
static uint16_t pix_buf_uv[2][2048*2];
// clipping helper table for C implementations:
static unsigned char clip_table[768];
@ -15,6 +24,31 @@ static int yuvtab_0c92[256];
static int yuvtab_1a1e[256];
static int yuvtab_40cf[256];
static uint64_t yCoeff= 0x2568256825682568LL;
static uint64_t ubCoeff= 0x3343334333433343LL;
static uint64_t vrCoeff= 0x40cf40cf40cf40cfLL;
static uint64_t ugCoeff= 0xE5E2E5E2E5E2E5E2LL;
static uint64_t vgCoeff= 0xF36EF36EF36EF36ELL;
static uint64_t w80= 0x0080008000800080LL;
static uint64_t w10= 0x0010001000100010LL;
static uint64_t b16Dither= 0x0004000400040004LL;
static uint64_t b16Dither1=0x0004000400040004LL;
static uint64_t b16Dither2=0x0602060206020602LL;
static uint64_t g16Dither= 0x0002000200020002LL;
static uint64_t g16Dither1=0x0002000200020002LL;
static uint64_t g16Dither2=0x0301030103010301LL;
static uint64_t b16Mask= 0x001F001F001F001FLL;
static uint64_t g16Mask= 0x07E007E007E007E0LL;
static uint64_t r16Mask= 0xF800F800F800F800LL;
static uint64_t temp0;
static uint8_t funnyYCode[10000];
static uint8_t funnyUVCode[10000];
// *** bilinear scaling and yuv->rgb conversion of yv12 slices:
// *** Note: it's called multiple times while decoding a frame, first time y==0
// *** Designed to upscale, but may work for downscale too.
@ -33,25 +67,149 @@ unsigned int s_xinc2=s_xinc>>1;
static int s_srcypos;
static int s_ypos;
static int s_last_ypos;
static int static_dstw;
#ifdef HAVE_MMX2
static int old_dstw= -1;
static int old_s_xinc= -1;
#endif
s_xinc&= -2; //clear last bit or uv and y might be shifted relative to each other
if(y==0){
s_srcypos=-2*s_yinc;
s_ypos=-2;
s_last_ypos=-2;
#ifdef HAVE_MMX2
// cant downscale !!!
if(old_s_xinc != s_xinc || old_dstw!=dstw)
{
uint8_t *fragment;
int imm8OfPShufW1;
int imm8OfPShufW2;
int fragmentLength;
int xpos, xx, xalpha, i;
old_s_xinc= s_xinc;
old_dstw= dstw;
static_dstw= dstw;
// create an optimized horizontal scaling routine
//code fragment
// fragmentLength=0;
// printf("%d, %d\n", fragmentLength,imm8OfPShufW1);
asm volatile(
"jmp 9f \n\t"
// Begin
"0: \n\t"
"movq (%%esi, %%ebx), %%mm0 \n\t" //FIXME Alignment
"movq %%mm0, %%mm1 \n\t"
"psrlq $8, %%mm0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"pshufw $0xFF, %%mm1, %%mm1 \n\t"
"1: \n\t"
"pshufw $0xFF, %%mm0, %%mm0 \n\t"
"2: \n\t"
"psubw %%mm1, %%mm0 \n\t"
"psraw $1, %%mm0 \n\t"
"pmullw %%mm2, %%mm0 \n\t"
"psllw $7, %%mm1 \n\t"
"paddw %%mm1, %%mm0 \n\t"
"movq %%mm0, (%%edi, %%eax) \n\t"
"paddb %%mm6, %%mm2 \n\t" // 2*alpha += xpos&0xFF
"addb %%ch, %%cl \n\t" //2*xalpha += (4*s_xinc)&0xFF
"adcl %%edx, %%ebx \n\t" //xx+= (4*s_xinc)>>8 + carry
"addl $8, %%eax \n\t"
// End
"9: \n\t"
// "int $3\n\t"
"leal 0b, %0 \n\t"
"leal 1b, %1 \n\t"
"leal 2b, %2 \n\t"
"decl %1 \n\t"
"decl %2 \n\t"
"subl %0, %1 \n\t"
"subl %0, %2 \n\t"
"leal 9b, %3 \n\t"
"subl %0, %3 \n\t"
:"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
"=r" (fragmentLength)
);
xpos= xx=xalpha= 0;
//FIXME choose size and or xinc so that they fit exactly
for(i=0; i<dstw/8; i++)
{
int xx=xpos>>8;
if((i&3) == 0)
{
int a=0;
int b=((xpos+s_xinc)>>8) - xx;
int c=((xpos+s_xinc*2)>>8) - xx;
int d=((xpos+s_xinc*3)>>8) - xx;
memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
a | (b<<2) | (c<<4) | (d<<6);
funnyYCode[fragmentLength*(i+4)/4]= RET;
}
xpos+=s_xinc;
}
xpos= xx=xalpha= 0;
//FIXME choose size and or xinc so that they fit exactly
for(i=0; i<dstw/8; i++)
{
int xx=xpos>>8;
if((i&3) == 0)
{
int a=0;
int b=((xpos+s_xinc2)>>8) - xx;
int c=((xpos+s_xinc2*2)>>8) - xx;
int d=((xpos+s_xinc2*3)>>8) - xx;
memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
a | (b<<2) | (c<<4) | (d<<6);
funnyUVCode[fragmentLength*(i+4)/4]= RET;
}
xpos+=s_xinc2;
}
// funnyCode[0]= RET;
}
#endif
} // reset counters
while(1){
unsigned char *dest=dstptr+dststride*s_ypos;
int y0=2+(s_srcypos>>16);
int y1=1+(s_srcypos>>17);
int yalpha=(s_srcypos&0xFFFF)>>8;
int yalpha1=yalpha^255;
int uvalpha=((s_srcypos>>1)&0xFFFF)>>8;
int uvalpha1=uvalpha^255;
unsigned int *buf0=pix_buf_y[y0&3];
unsigned int *buf1=pix_buf_y[((y0+1)&3)];
unsigned int *uvbuf0=pix_buf_uv[y1&1];
unsigned int *uvbuf1=pix_buf_uv[(y1&1)^1];
int yalpha=(s_srcypos&0xFFFF)>>7;
int yalpha1=yalpha^511;
int uvalpha=((s_srcypos>>1)&0xFFFF)>>7;
int uvalpha1=uvalpha^511;
uint16_t *buf0=pix_buf_y[y0&3];
uint16_t *buf1=pix_buf_y[((y0+1)&3)];
uint16_t *uvbuf0=pix_buf_uv[y1&1];
uint16_t *uvbuf1=pix_buf_uv[(y1&1)^1];
int i;
if(y0>=y+h) break;
@ -64,54 +222,423 @@ static int s_last_ypos;
s_last_ypos=y0;
// *** horizontal scale Y line to temp buffer
// this loop should be rewritten in MMX assembly!!!!
#ifdef HAVE_MMX2
asm volatile(
"pxor %%mm7, %%mm7 \n\t"
"pxor %%mm2, %%mm2 \n\t" // 2*xalpha
"movd %5, %%mm6 \n\t" // s_xinc&0xFF
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"movq %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t"
"paddb %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t"
"paddb %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t" //0,t,2t,3t t=s_xinc&0xFF
"movq %%mm2, temp0 \n\t"
"movd %4, %%mm6 \n\t" //(s_xinc*4)&0xFF
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"movl %0, %%esi \n\t" // src
"movl %1, %%edi \n\t" // buf1
"movl %3, %%edx \n\t" // (s_xinc*4)>>8
"xorl %%ecx, %%ecx \n\t"
"movb %4, %%ch \n\t" // (s_xinc*4)&0xFF
// "int $3\n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
:: "m" (src), "m" (buf1), "m" (dstw), "m" ((s_xinc*4)>>8),
"m" ((s_xinc*4)&0xFF), "m" (s_xinc&0xFF)
: "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
);
#elif defined (ARCH_X86)
//NO MMX just normal asm ... FIXME try/write funny MMX2 variant
//FIXME add prefetch
asm volatile(
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"xorl %%ecx, %%ecx \n\t" // 2*xalpha
"1: \n\t"
"movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
"movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
"subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
"imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
"shll $8, %%edi \n\t"
"addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
"shrl $1, %%esi \n\t"
"movw %%si, (%1, %%eax, 2) \n\t"
"addb %4, %%cl \n\t" //2*xalpha += s_xinc&0xFF
"adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
"movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
"movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
"subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
"imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
"shll $8, %%edi \n\t"
"addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
"shrl $1, %%esi \n\t"
"movw %%si, 2(%1, %%eax, 2) \n\t"
"addb %4, %%cl \n\t" //2*xalpha += s_xinc&0xFF
"adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
"addl $2, %%eax \n\t"
"cmpl %2, %%eax \n\t"
" jb 1b \n\t"
:: "r" (src), "r" (buf1), "m" (dstw), "m" (s_xinc>>8), "m" (s_xinc&0xFF)
: "%eax", "%ebx", "%ecx", "%edi", "%esi"
);
#else
for(i=0;i<dstw;i++){
register unsigned int xx=xpos>>8;
register unsigned int xalpha=xpos&0xFF;
buf1[i]=(src[xx]*(xalpha^255)+src[xx+1]*xalpha);
register unsigned int xalpha=(xpos&0xFF)>>1;
buf1[i]=(src[xx]*(xalpha^127)+src[xx+1]*xalpha);
xpos+=s_xinc;
}
#endif
// *** horizontal scale U and V lines to temp buffer
if(!(y0&1)){
unsigned char *src1=srcptr[1]+(y1-y/2)*stride[1];
unsigned char *src2=srcptr[2]+(y1-y/2)*stride[2];
xpos=0;
// this loop should be rewritten in MMX assembly!!!!
#ifdef HAVE_MMX2
asm volatile(
"pxor %%mm7, %%mm7 \n\t"
"pxor %%mm2, %%mm2 \n\t" // 2*xalpha
"movd %5, %%mm6 \n\t" // s_xinc&0xFF
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"movq %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t"
"paddb %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t"
"paddb %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t" //0,t,2t,3t t=s_xinc&0xFF
"movq %%mm2, temp0 \n\t"
"movd %4, %%mm6 \n\t" //(s_xinc*4)&0xFF
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"movl %0, %%esi \n\t" // src
"movl %1, %%edi \n\t" // buf1
"movl %3, %%edx \n\t" // (s_xinc*4)>>8
"xorl %%ecx, %%ecx \n\t"
"movb %4, %%ch \n\t" // (s_xinc*4)&0xFF
// "int $3\n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"movl %6, %%esi \n\t" // src
"movl %1, %%edi \n\t" // buf1
"addl $4096, %%edi \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
:: "m" (src1), "m" (uvbuf1), "m" (dstw), "m" ((s_xinc2*4)>>8),
"m" ((s_xinc2*4)&0xFF), "m" (s_xinc2&0xFF), "m" (src2)
: "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
);
#elif defined (ARCH_X86)
//NO MMX just normal asm ... FIXME try/write funny MMX2 variant
asm volatile(
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"xorl %%ecx, %%ecx \n\t" // 2*xalpha
"1: \n\t"
"movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
"movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
"subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
"imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
"shll $8, %%edi \n\t"
"addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
"movl %1, %%edi \n\t"
"shrl $1, %%esi \n\t"
"movw %%si, (%%edi, %%eax, 2) \n\t"
"movzbl (%5, %%ebx), %%edi \n\t" //src[xx]
"movzbl 1(%5, %%ebx), %%esi \n\t" //src[xx+1]
"subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
"imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
"shll $8, %%edi \n\t"
"addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
"movl %1, %%edi \n\t"
"shrl $1, %%esi \n\t"
"movw %%si, 4096(%%edi, %%eax, 2)\n\t"
"addb %4, %%cl \n\t" //2*xalpha += s_xinc&0xFF
"adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
"addl $1, %%eax \n\t"
"cmpl %2, %%eax \n\t"
" jb 1b \n\t"
:: "r" (src1), "m" (uvbuf1), "m" (dstw), "m" (s_xinc2>>8), "m" (s_xinc2&0xFF),
"r" (src2)
: "%eax", "%ebx", "%ecx", "%edi", "%esi"
);
#else
for(i=0;i<dstw;i++){
register unsigned int xx=xpos>>8;
register unsigned int xalpha=xpos&0xFF;
uvbuf1[i]=(src1[xx]*(xalpha^255)+src1[xx+1]*xalpha);
uvbuf1[i+2048]=(src2[xx]*(xalpha^255)+src2[xx+1]*xalpha);
register unsigned int xalpha=(xpos&0xFF)>>1;
uvbuf1[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
uvbuf1[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
xpos+=s_xinc2;
}
#endif
}
if(!y0) continue;
}
// this loop should be rewritten in MMX assembly!!!!
// Note1: this code can be resticted to n*8 (or n*16) width lines to simplify optimization...
// Re: Note1: ok n*4 for now
// Note2: instead of using lookup tabs, mmx version could do the multiply...
// Re: Note2: yep
// Note3: maybe we should make separated 15/16, 24 and 32bpp version of this:
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
#if 1
// 24/32 bpp
dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
// Re: done (32 & 16) and 16 has dithering :) but 16 is untested
#ifdef HAVE_MMX
//FIXME write lq version with less uv ...
//FIXME reorder / optimize
if(dstbpp == 4)
{
asm volatile(
#define YSCALEYUV2RGB \
"pxor %%mm7, %%mm7 \n\t"\
"movd %6, %%mm6 \n\t" /*yalpha1*/\
"punpcklwd %%mm6, %%mm6 \n\t"\
"punpcklwd %%mm6, %%mm6 \n\t"\
"movd %7, %%mm5 \n\t" /*uvalpha1*/\
"punpcklwd %%mm5, %%mm5 \n\t"\
"punpcklwd %%mm5, %%mm5 \n\t"\
"xorl %%eax, %%eax \n\t"\
"1: \n\t"\
"movq (%0, %%eax, 2), %%mm0 \n\t" /*buf0[eax]*/\
"movq (%1, %%eax, 2), %%mm1 \n\t" /*buf1[eax]*/\
"psubw %%mm1, %%mm0 \n\t" /* buf0[eax] - buf1[eax]*/\
"pmulhw %%mm6, %%mm0 \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
"psraw $7, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>7*/\
"paddw %%mm0, %%mm1 \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
"psubw w10, %%mm1 \n\t" /* Y-16*/\
"psllw $3, %%mm1 \n\t" /* (y-16)*8*/\
"pmulhw yCoeff, %%mm1 \n\t"\
\
"movq (%2, %%eax,2), %%mm2 \n\t" /* uvbuf0[eax]*/\
"movq (%3, %%eax,2), %%mm3 \n\t" /* uvbuf1[eax]*/\
"psubw %%mm3, %%mm2 \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
"pmulhw %%mm5, %%mm2 \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
"psraw $7, %%mm3 \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>7*/\
"paddw %%mm2, %%mm3 \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
"psubw w80, %%mm3 \n\t" /* (U-128)*/\
"psllw $3, %%mm3 \n\t" /*(U-128)8*/\
\
"movq 4096(%2, %%eax,2), %%mm4 \n\t" /* uvbuf0[eax+2048]*/\
"movq 4096(%3, %%eax,2), %%mm0 \n\t" /* uvbuf1[eax+2048]*/\
"psubw %%mm0, %%mm4 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
"pmulhw %%mm5, %%mm4 \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
"psraw $7, %%mm0 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>7*/\
"paddw %%mm4, %%mm0 \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
"psubw w80, %%mm0 \n\t" /* (V-128)*/\
"psllw $3, %%mm0 \n\t" /* (V-128)8*/\
\
"movq %%mm3, %%mm2 \n\t" /* (U-128)8*/\
"pmulhw ubCoeff, %%mm3 \n\t"\
"paddw %%mm1, %%mm3 \n\t" /* B*/\
\
"movq %%mm0, %%mm4 \n\t" /* (V-128)8*/\
"pmulhw vrCoeff, %%mm0 \n\t"\
"paddw %%mm1, %%mm0 \n\t" /* R*/\
\
"pmulhw ugCoeff, %%mm2 \n\t"\
"pmulhw vgCoeff, %%mm4 \n\t"\
"paddw %%mm4, %%mm2 \n\t"\
"paddw %%mm2, %%mm1 \n\t" /* G*/\
\
"packuswb %%mm3, %%mm3 \n\t"\
"packuswb %%mm0, %%mm0 \n\t"\
"packuswb %%mm1, %%mm1 \n\t"
YSCALEYUV2RGB
"punpcklbw %%mm1, %%mm3 \n\t" // BGBGBGBG
"punpcklbw %%mm7, %%mm0 \n\t" // R0R0R0R0
"movq %%mm3, %%mm1 \n\t"
"punpcklwd %%mm0, %%mm3 \n\t" // BGR0BGR0
"punpckhwd %%mm0, %%mm1 \n\t" // BGR0BGR0
#ifdef HAVE_MMX2
"movntq %%mm3, (%4, %%eax, 4) \n\t"
"movntq %%mm1, 8(%4, %%eax, 4) \n\t"
#else
unsigned short *d=dest;
unsigned int b=clip_table[((Y + yuvtab_3343[U]) >>13)];
unsigned int g=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
unsigned int r=clip_table[((Y + yuvtab_40cf[V]) >>13)];
// d[0]=((r>>3)<<10)|((g>>3)<<5)|((b>>3)); // 15bpp
d[0]=((r>>3)<<11)|((g>>2)<<5)|((b>>3)); // 16bpp
"movq %%mm3, (%4, %%eax, 4) \n\t"
"movq %%mm1, 8(%4, %%eax, 4) \n\t"
#endif
dest+=dstbpp;
}
"addl $4, %%eax \n\t"
"cmpl %5, %%eax \n\t"
" jb 1b \n\t"
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax"
);
}
else if(dstbpp==2)
{
asm volatile(
YSCALEYUV2RGB
"paddusb g16Dither, %%mm1 \n\t"
"paddusb b16Dither, %%mm0 \n\t"
"paddusb b16Dither, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t" // 0G0G0G0G
"punpcklbw %%mm7, %%mm3 \n\t" // 0B0B0B0B
"punpcklbw %%mm7, %%mm0 \n\t" // 0R0R0R0R
"psrlw $3, %%mm3 \n\t"
"psllw $3, %%mm1 \n\t"
"psllw $8, %%mm0 \n\t"
"pand g16Mask, %%mm1 \n\t"
"pand r16Mask, %%mm0 \n\t"
"por %%mm3, %%mm1 \n\t"
"por %%mm1, %%mm0 \n\t"
#ifdef HAVE_MMX2
"movntq %%mm0, (%4, %%eax, 2) \n\t"
#else
"movq %%mm0, (%4, %%eax, 2) \n\t"
#endif
"addl $4, %%eax \n\t"
"cmpl %5, %%eax \n\t"
" jb 1b \n\t"
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax"
);
}
#else
if(dstbpp==4 || dstbpp==3)
{
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
dest+=dstbpp;
}
}
else if(dstbpp==2) //16bit
{
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
((uint16_t*)dest)[0] =
(clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
(clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<3)&0x07E0 |
(clip_table[((Y + yuvtab_40cf[V]) >>13)]<<8)&0xF800;
dest+=dstbpp;
}
}
else if(dstbpp==2) //15bit FIXME how do i figure out if its 15 or 16?
{
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
((uint16_t*)dest)[0] =
(clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
(clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<2)&0x03E0 |
(clip_table[((Y + yuvtab_40cf[V]) >>13)]<<7)&0x7C00;
dest+=dstbpp;
}
}
#endif
b16Dither= b16Dither1;
b16Dither1= b16Dither2;
b16Dither2= b16Dither;
g16Dither= g16Dither1;
g16Dither1= g16Dither2;
g16Dither2= g16Dither;
}
}

599
postproc/swscale_template.c
Unescape View File

@ -1,9 +1,18 @@
// Software scaling and colorspace conversion routines for MPlayer
#include <inttypes.h>
#include "../config.h"
#undef HAVE_MMX2 //code is buggy
//#undef HAVE_MMX
#define RET 0xC3 //near return opcode
// temporary storage for 4 yuv lines:
static unsigned int pix_buf_y[4][2048];
static unsigned int pix_buf_uv[2][2048*2];
// 16bit for now (mmx likes it more compact)
static uint16_t pix_buf_y[4][2048];
static uint16_t pix_buf_uv[2][2048*2];
// clipping helper table for C implementations:
static unsigned char clip_table[768];
@ -15,6 +24,31 @@ static int yuvtab_0c92[256];
static int yuvtab_1a1e[256];
static int yuvtab_40cf[256];
static uint64_t yCoeff= 0x2568256825682568LL;
static uint64_t ubCoeff= 0x3343334333433343LL;
static uint64_t vrCoeff= 0x40cf40cf40cf40cfLL;
static uint64_t ugCoeff= 0xE5E2E5E2E5E2E5E2LL;
static uint64_t vgCoeff= 0xF36EF36EF36EF36ELL;
static uint64_t w80= 0x0080008000800080LL;
static uint64_t w10= 0x0010001000100010LL;
static uint64_t b16Dither= 0x0004000400040004LL;
static uint64_t b16Dither1=0x0004000400040004LL;
static uint64_t b16Dither2=0x0602060206020602LL;
static uint64_t g16Dither= 0x0002000200020002LL;
static uint64_t g16Dither1=0x0002000200020002LL;
static uint64_t g16Dither2=0x0301030103010301LL;
static uint64_t b16Mask= 0x001F001F001F001FLL;
static uint64_t g16Mask= 0x07E007E007E007E0LL;
static uint64_t r16Mask= 0xF800F800F800F800LL;
static uint64_t temp0;
static uint8_t funnyYCode[10000];
static uint8_t funnyUVCode[10000];
// *** bilinear scaling and yuv->rgb conversion of yv12 slices:
// *** Note: it's called multiple times while decoding a frame, first time y==0
// *** Designed to upscale, but may work for downscale too.
@ -33,25 +67,149 @@ unsigned int s_xinc2=s_xinc>>1;
static int s_srcypos;
static int s_ypos;
static int s_last_ypos;
static int static_dstw;
#ifdef HAVE_MMX2
static int old_dstw= -1;
static int old_s_xinc= -1;
#endif
s_xinc&= -2; //clear last bit or uv and y might be shifted relative to each other
if(y==0){
s_srcypos=-2*s_yinc;
s_ypos=-2;
s_last_ypos=-2;
#ifdef HAVE_MMX2
// cant downscale !!!
if(old_s_xinc != s_xinc || old_dstw!=dstw)
{
uint8_t *fragment;
int imm8OfPShufW1;
int imm8OfPShufW2;
int fragmentLength;
int xpos, xx, xalpha, i;
old_s_xinc= s_xinc;
old_dstw= dstw;
static_dstw= dstw;
// create an optimized horizontal scaling routine
//code fragment
// fragmentLength=0;
// printf("%d, %d\n", fragmentLength,imm8OfPShufW1);
asm volatile(
"jmp 9f \n\t"
// Begin
"0: \n\t"
"movq (%%esi, %%ebx), %%mm0 \n\t" //FIXME Alignment
"movq %%mm0, %%mm1 \n\t"
"psrlq $8, %%mm0 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t"
"punpcklbw %%mm7, %%mm0 \n\t"
"pshufw $0xFF, %%mm1, %%mm1 \n\t"
"1: \n\t"
"pshufw $0xFF, %%mm0, %%mm0 \n\t"
"2: \n\t"
"psubw %%mm1, %%mm0 \n\t"
"psraw $1, %%mm0 \n\t"
"pmullw %%mm2, %%mm0 \n\t"
"psllw $7, %%mm1 \n\t"
"paddw %%mm1, %%mm0 \n\t"
"movq %%mm0, (%%edi, %%eax) \n\t"
"paddb %%mm6, %%mm2 \n\t" // 2*alpha += xpos&0xFF
"addb %%ch, %%cl \n\t" //2*xalpha += (4*s_xinc)&0xFF
"adcl %%edx, %%ebx \n\t" //xx+= (4*s_xinc)>>8 + carry
"addl $8, %%eax \n\t"
// End
"9: \n\t"
// "int $3\n\t"
"leal 0b, %0 \n\t"
"leal 1b, %1 \n\t"
"leal 2b, %2 \n\t"
"decl %1 \n\t"
"decl %2 \n\t"
"subl %0, %1 \n\t"
"subl %0, %2 \n\t"
"leal 9b, %3 \n\t"
"subl %0, %3 \n\t"
:"=r" (fragment), "=r" (imm8OfPShufW1), "=r" (imm8OfPShufW2),
"=r" (fragmentLength)
);
xpos= xx=xalpha= 0;
//FIXME choose size and or xinc so that they fit exactly
for(i=0; i<dstw/8; i++)
{
int xx=xpos>>8;
if((i&3) == 0)
{
int a=0;
int b=((xpos+s_xinc)>>8) - xx;
int c=((xpos+s_xinc*2)>>8) - xx;
int d=((xpos+s_xinc*3)>>8) - xx;
memcpy(funnyYCode + fragmentLength*i/4, fragment, fragmentLength);
funnyYCode[fragmentLength*i/4 + imm8OfPShufW1]=
funnyYCode[fragmentLength*i/4 + imm8OfPShufW2]=
a | (b<<2) | (c<<4) | (d<<6);
funnyYCode[fragmentLength*(i+4)/4]= RET;
}
xpos+=s_xinc;
}
xpos= xx=xalpha= 0;
//FIXME choose size and or xinc so that they fit exactly
for(i=0; i<dstw/8; i++)
{
int xx=xpos>>8;
if((i&3) == 0)
{
int a=0;
int b=((xpos+s_xinc2)>>8) - xx;
int c=((xpos+s_xinc2*2)>>8) - xx;
int d=((xpos+s_xinc2*3)>>8) - xx;
memcpy(funnyUVCode + fragmentLength*i/4, fragment, fragmentLength);
funnyUVCode[fragmentLength*i/4 + imm8OfPShufW1]=
funnyUVCode[fragmentLength*i/4 + imm8OfPShufW2]=
a | (b<<2) | (c<<4) | (d<<6);
funnyUVCode[fragmentLength*(i+4)/4]= RET;
}
xpos+=s_xinc2;
}
// funnyCode[0]= RET;
}
#endif
} // reset counters
while(1){
unsigned char *dest=dstptr+dststride*s_ypos;
int y0=2+(s_srcypos>>16);
int y1=1+(s_srcypos>>17);
int yalpha=(s_srcypos&0xFFFF)>>8;
int yalpha1=yalpha^255;
int uvalpha=((s_srcypos>>1)&0xFFFF)>>8;
int uvalpha1=uvalpha^255;
unsigned int *buf0=pix_buf_y[y0&3];
unsigned int *buf1=pix_buf_y[((y0+1)&3)];
unsigned int *uvbuf0=pix_buf_uv[y1&1];
unsigned int *uvbuf1=pix_buf_uv[(y1&1)^1];
int yalpha=(s_srcypos&0xFFFF)>>7;
int yalpha1=yalpha^511;
int uvalpha=((s_srcypos>>1)&0xFFFF)>>7;
int uvalpha1=uvalpha^511;
uint16_t *buf0=pix_buf_y[y0&3];
uint16_t *buf1=pix_buf_y[((y0+1)&3)];
uint16_t *uvbuf0=pix_buf_uv[y1&1];
uint16_t *uvbuf1=pix_buf_uv[(y1&1)^1];
int i;
if(y0>=y+h) break;
@ -64,54 +222,423 @@ static int s_last_ypos;
s_last_ypos=y0;
// *** horizontal scale Y line to temp buffer
// this loop should be rewritten in MMX assembly!!!!
#ifdef HAVE_MMX2
asm volatile(
"pxor %%mm7, %%mm7 \n\t"
"pxor %%mm2, %%mm2 \n\t" // 2*xalpha
"movd %5, %%mm6 \n\t" // s_xinc&0xFF
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"movq %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t"
"paddb %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t"
"paddb %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t" //0,t,2t,3t t=s_xinc&0xFF
"movq %%mm2, temp0 \n\t"
"movd %4, %%mm6 \n\t" //(s_xinc*4)&0xFF
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"movl %0, %%esi \n\t" // src
"movl %1, %%edi \n\t" // buf1
"movl %3, %%edx \n\t" // (s_xinc*4)>>8
"xorl %%ecx, %%ecx \n\t"
"movb %4, %%ch \n\t" // (s_xinc*4)&0xFF
// "int $3\n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyYCode \n\t"
:: "m" (src), "m" (buf1), "m" (dstw), "m" ((s_xinc*4)>>8),
"m" ((s_xinc*4)&0xFF), "m" (s_xinc&0xFF)
: "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
);
#elif defined (ARCH_X86)
//NO MMX just normal asm ... FIXME try/write funny MMX2 variant
//FIXME add prefetch
asm volatile(
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"xorl %%ecx, %%ecx \n\t" // 2*xalpha
"1: \n\t"
"movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
"movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
"subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
"imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
"shll $8, %%edi \n\t"
"addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
"shrl $1, %%esi \n\t"
"movw %%si, (%1, %%eax, 2) \n\t"
"addb %4, %%cl \n\t" //2*xalpha += s_xinc&0xFF
"adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
"movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
"movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
"subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
"imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
"shll $8, %%edi \n\t"
"addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
"shrl $1, %%esi \n\t"
"movw %%si, 2(%1, %%eax, 2) \n\t"
"addb %4, %%cl \n\t" //2*xalpha += s_xinc&0xFF
"adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
"addl $2, %%eax \n\t"
"cmpl %2, %%eax \n\t"
" jb 1b \n\t"
:: "r" (src), "r" (buf1), "m" (dstw), "m" (s_xinc>>8), "m" (s_xinc&0xFF)
: "%eax", "%ebx", "%ecx", "%edi", "%esi"
);
#else
for(i=0;i<dstw;i++){
register unsigned int xx=xpos>>8;
register unsigned int xalpha=xpos&0xFF;
buf1[i]=(src[xx]*(xalpha^255)+src[xx+1]*xalpha);
register unsigned int xalpha=(xpos&0xFF)>>1;
buf1[i]=(src[xx]*(xalpha^127)+src[xx+1]*xalpha);
xpos+=s_xinc;
}
#endif
// *** horizontal scale U and V lines to temp buffer
if(!(y0&1)){
unsigned char *src1=srcptr[1]+(y1-y/2)*stride[1];
unsigned char *src2=srcptr[2]+(y1-y/2)*stride[2];
xpos=0;
// this loop should be rewritten in MMX assembly!!!!
#ifdef HAVE_MMX2
asm volatile(
"pxor %%mm7, %%mm7 \n\t"
"pxor %%mm2, %%mm2 \n\t" // 2*xalpha
"movd %5, %%mm6 \n\t" // s_xinc&0xFF
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"movq %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t"
"paddb %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t"
"paddb %%mm6, %%mm2 \n\t"
"psllq $16, %%mm2 \n\t" //0,t,2t,3t t=s_xinc&0xFF
"movq %%mm2, temp0 \n\t"
"movd %4, %%mm6 \n\t" //(s_xinc*4)&0xFF
"punpcklwd %%mm6, %%mm6 \n\t"
"punpcklwd %%mm6, %%mm6 \n\t"
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"movl %0, %%esi \n\t" // src
"movl %1, %%edi \n\t" // buf1
"movl %3, %%edx \n\t" // (s_xinc*4)>>8
"xorl %%ecx, %%ecx \n\t"
"movb %4, %%ch \n\t" // (s_xinc*4)&0xFF
// "int $3\n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"movl %6, %%esi \n\t" // src
"movl %1, %%edi \n\t" // buf1
"addl $4096, %%edi \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
"movq temp0, %%mm2 \n\t"
"xorb %%cl, %%cl \n\t"
"call funnyUVCode \n\t"
:: "m" (src1), "m" (uvbuf1), "m" (dstw), "m" ((s_xinc2*4)>>8),
"m" ((s_xinc2*4)&0xFF), "m" (s_xinc2&0xFF), "m" (src2)
: "%eax", "%ebx", "%ecx", "%edx", "%esi", "%edi"
);
#elif defined (ARCH_X86)
//NO MMX just normal asm ... FIXME try/write funny MMX2 variant
asm volatile(
"xorl %%eax, %%eax \n\t" // i
"xorl %%ebx, %%ebx \n\t" // xx
"xorl %%ecx, %%ecx \n\t" // 2*xalpha
"1: \n\t"
"movzbl (%0, %%ebx), %%edi \n\t" //src[xx]
"movzbl 1(%0, %%ebx), %%esi \n\t" //src[xx+1]
"subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
"imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
"shll $8, %%edi \n\t"
"addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
"movl %1, %%edi \n\t"
"shrl $1, %%esi \n\t"
"movw %%si, (%%edi, %%eax, 2) \n\t"
"movzbl (%5, %%ebx), %%edi \n\t" //src[xx]
"movzbl 1(%5, %%ebx), %%esi \n\t" //src[xx+1]
"subl %%edi, %%esi \n\t" //src[xx+1] - src[xx]
"imull %%ecx, %%esi \n\t" //(src[xx+1] - src[xx])*2*xalpha
"shll $8, %%edi \n\t"
"addl %%edi, %%esi \n\t" //src[xx+1]*2*xalpha + src[xx]*(1-2*xalpha)
"movl %1, %%edi \n\t"
"shrl $1, %%esi \n\t"
"movw %%si, 4096(%%edi, %%eax, 2)\n\t"
"addb %4, %%cl \n\t" //2*xalpha += s_xinc&0xFF
"adcl %3, %%ebx \n\t" //xx+= s_xinc>>8 + carry
"addl $1, %%eax \n\t"
"cmpl %2, %%eax \n\t"
" jb 1b \n\t"
:: "r" (src1), "m" (uvbuf1), "m" (dstw), "m" (s_xinc2>>8), "m" (s_xinc2&0xFF),
"r" (src2)
: "%eax", "%ebx", "%ecx", "%edi", "%esi"
);
#else
for(i=0;i<dstw;i++){
register unsigned int xx=xpos>>8;
register unsigned int xalpha=xpos&0xFF;
uvbuf1[i]=(src1[xx]*(xalpha^255)+src1[xx+1]*xalpha);
uvbuf1[i+2048]=(src2[xx]*(xalpha^255)+src2[xx+1]*xalpha);
register unsigned int xalpha=(xpos&0xFF)>>1;
uvbuf1[i]=(src1[xx]*(xalpha^127)+src1[xx+1]*xalpha);
uvbuf1[i+2048]=(src2[xx]*(xalpha^127)+src2[xx+1]*xalpha);
xpos+=s_xinc2;
}
#endif
}
if(!y0) continue;
}
// this loop should be rewritten in MMX assembly!!!!
// Note1: this code can be resticted to n*8 (or n*16) width lines to simplify optimization...
// Re: Note1: ok n*4 for now
// Note2: instead of using lookup tabs, mmx version could do the multiply...
// Re: Note2: yep
// Note3: maybe we should make separated 15/16, 24 and 32bpp version of this:
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
#if 1
// 24/32 bpp
dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
// Re: done (32 & 16) and 16 has dithering :) but 16 is untested
#ifdef HAVE_MMX
//FIXME write lq version with less uv ...
//FIXME reorder / optimize
if(dstbpp == 4)
{
asm volatile(
#define YSCALEYUV2RGB \
"pxor %%mm7, %%mm7 \n\t"\
"movd %6, %%mm6 \n\t" /*yalpha1*/\
"punpcklwd %%mm6, %%mm6 \n\t"\
"punpcklwd %%mm6, %%mm6 \n\t"\
"movd %7, %%mm5 \n\t" /*uvalpha1*/\
"punpcklwd %%mm5, %%mm5 \n\t"\
"punpcklwd %%mm5, %%mm5 \n\t"\
"xorl %%eax, %%eax \n\t"\
"1: \n\t"\
"movq (%0, %%eax, 2), %%mm0 \n\t" /*buf0[eax]*/\
"movq (%1, %%eax, 2), %%mm1 \n\t" /*buf1[eax]*/\
"psubw %%mm1, %%mm0 \n\t" /* buf0[eax] - buf1[eax]*/\
"pmulhw %%mm6, %%mm0 \n\t" /* (buf0[eax] - buf1[eax])yalpha1>>16*/\
"psraw $7, %%mm1 \n\t" /* buf0[eax] - buf1[eax] >>7*/\
"paddw %%mm0, %%mm1 \n\t" /* buf0[eax]yalpha1 + buf1[eax](1-yalpha1) >>16*/\
"psubw w10, %%mm1 \n\t" /* Y-16*/\
"psllw $3, %%mm1 \n\t" /* (y-16)*8*/\
"pmulhw yCoeff, %%mm1 \n\t"\
\
"movq (%2, %%eax,2), %%mm2 \n\t" /* uvbuf0[eax]*/\
"movq (%3, %%eax,2), %%mm3 \n\t" /* uvbuf1[eax]*/\
"psubw %%mm3, %%mm2 \n\t" /* uvbuf0[eax] - uvbuf1[eax]*/\
"pmulhw %%mm5, %%mm2 \n\t" /* (uvbuf0[eax] - uvbuf1[eax])uvalpha1>>16*/\
"psraw $7, %%mm3 \n\t" /* uvbuf0[eax] - uvbuf1[eax] >>7*/\
"paddw %%mm2, %%mm3 \n\t" /* uvbuf0[eax]uvalpha1 - uvbuf1[eax](1-uvalpha1)*/\
"psubw w80, %%mm3 \n\t" /* (U-128)*/\
"psllw $3, %%mm3 \n\t" /*(U-128)8*/\
\
"movq 4096(%2, %%eax,2), %%mm4 \n\t" /* uvbuf0[eax+2048]*/\
"movq 4096(%3, %%eax,2), %%mm0 \n\t" /* uvbuf1[eax+2048]*/\
"psubw %%mm0, %%mm4 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048]*/\
"pmulhw %%mm5, %%mm4 \n\t" /* (uvbuf0[eax+2048] - uvbuf1[eax+2048])uvalpha1>>16*/\
"psraw $7, %%mm0 \n\t" /* uvbuf0[eax+2048] - uvbuf1[eax+2048] >>7*/\
"paddw %%mm4, %%mm0 \n\t" /* uvbuf0[eax+2048]uvalpha1 - uvbuf1[eax+2048](1-uvalpha1)*/\
"psubw w80, %%mm0 \n\t" /* (V-128)*/\
"psllw $3, %%mm0 \n\t" /* (V-128)8*/\
\
"movq %%mm3, %%mm2 \n\t" /* (U-128)8*/\
"pmulhw ubCoeff, %%mm3 \n\t"\
"paddw %%mm1, %%mm3 \n\t" /* B*/\
\
"movq %%mm0, %%mm4 \n\t" /* (V-128)8*/\
"pmulhw vrCoeff, %%mm0 \n\t"\
"paddw %%mm1, %%mm0 \n\t" /* R*/\
\
"pmulhw ugCoeff, %%mm2 \n\t"\
"pmulhw vgCoeff, %%mm4 \n\t"\
"paddw %%mm4, %%mm2 \n\t"\
"paddw %%mm2, %%mm1 \n\t" /* G*/\
\
"packuswb %%mm3, %%mm3 \n\t"\
"packuswb %%mm0, %%mm0 \n\t"\
"packuswb %%mm1, %%mm1 \n\t"
YSCALEYUV2RGB
"punpcklbw %%mm1, %%mm3 \n\t" // BGBGBGBG
"punpcklbw %%mm7, %%mm0 \n\t" // R0R0R0R0
"movq %%mm3, %%mm1 \n\t"
"punpcklwd %%mm0, %%mm3 \n\t" // BGR0BGR0
"punpckhwd %%mm0, %%mm1 \n\t" // BGR0BGR0
#ifdef HAVE_MMX2
"movntq %%mm3, (%4, %%eax, 4) \n\t"
"movntq %%mm1, 8(%4, %%eax, 4) \n\t"
#else
unsigned short *d=dest;
unsigned int b=clip_table[((Y + yuvtab_3343[U]) >>13)];
unsigned int g=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
unsigned int r=clip_table[((Y + yuvtab_40cf[V]) >>13)];
// d[0]=((r>>3)<<10)|((g>>3)<<5)|((b>>3)); // 15bpp
d[0]=((r>>3)<<11)|((g>>2)<<5)|((b>>3)); // 16bpp
"movq %%mm3, (%4, %%eax, 4) \n\t"
"movq %%mm1, 8(%4, %%eax, 4) \n\t"
#endif
dest+=dstbpp;
}
"addl $4, %%eax \n\t"
"cmpl %5, %%eax \n\t"
" jb 1b \n\t"
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax"
);
}
else if(dstbpp==2)
{
asm volatile(
YSCALEYUV2RGB
"paddusb g16Dither, %%mm1 \n\t"
"paddusb b16Dither, %%mm0 \n\t"
"paddusb b16Dither, %%mm3 \n\t"
"punpcklbw %%mm7, %%mm1 \n\t" // 0G0G0G0G
"punpcklbw %%mm7, %%mm3 \n\t" // 0B0B0B0B
"punpcklbw %%mm7, %%mm0 \n\t" // 0R0R0R0R
"psrlw $3, %%mm3 \n\t"
"psllw $3, %%mm1 \n\t"
"psllw $8, %%mm0 \n\t"
"pand g16Mask, %%mm1 \n\t"
"pand r16Mask, %%mm0 \n\t"
"por %%mm3, %%mm1 \n\t"
"por %%mm1, %%mm0 \n\t"
#ifdef HAVE_MMX2
"movntq %%mm0, (%4, %%eax, 2) \n\t"
#else
"movq %%mm0, (%4, %%eax, 2) \n\t"
#endif
"addl $4, %%eax \n\t"
"cmpl %5, %%eax \n\t"
" jb 1b \n\t"
:: "r" (buf0), "r" (buf1), "r" (uvbuf0), "r" (uvbuf1), "r" (dest), "m" (dstw),
"m" (yalpha1), "m" (uvalpha1)
: "%eax"
);
}
#else
if(dstbpp==4 || dstbpp==3)
{
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
dest[0]=clip_table[((Y + yuvtab_3343[U]) >>13)];
dest[1]=clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)];
dest[2]=clip_table[((Y + yuvtab_40cf[V]) >>13)];
dest+=dstbpp;
}
}
else if(dstbpp==2) //16bit
{
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
((uint16_t*)dest)[0] =
(clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
(clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<3)&0x07E0 |
(clip_table[((Y + yuvtab_40cf[V]) >>13)]<<8)&0xF800;
dest+=dstbpp;
}
}
else if(dstbpp==2) //15bit FIXME how do i figure out if its 15 or 16?
{
for(i=0;i<dstw;i++){
// vertical linear interpolation && yuv2rgb in a single step:
int Y=yuvtab_2568[((buf0[i]*yalpha1+buf1[i]*yalpha)>>16)];
int U=((uvbuf0[i]*uvalpha1+uvbuf1[i]*uvalpha)>>16);
int V=((uvbuf0[i+2048]*uvalpha1+uvbuf1[i+2048]*uvalpha)>>16);
((uint16_t*)dest)[0] =
(clip_table[((Y + yuvtab_3343[U]) >>13)]>>3) |
(clip_table[((Y + yuvtab_0c92[V] + yuvtab_1a1e[U]) >>13)]<<2)&0x03E0 |
(clip_table[((Y + yuvtab_40cf[V]) >>13)]<<7)&0x7C00;
dest+=dstbpp;
}
}
#endif
b16Dither= b16Dither1;
b16Dither1= b16Dither2;
b16Dither2= b16Dither;
g16Dither= g16Dither1;
g16Dither1= g16Dither2;
g16Dither2= g16Dither;
}
}

Write Preview
Loading…
Cancel
Save