yuv422 to rgb565 doesn't work well
uint32_t mCoefTbl32[516];
uint8_t *mCoefTbl = NULL;
static void init_coff()
{
uint8_t *clip;
int i;
mCoefTbl = (uint8_t *)mCoefTbl32;
*((uint32_t*)mCoefTbl) = (int)(65536*0.4681); //0.714);
*((uint32_t*)(mCoefTbl+4)) = (int)(65536*1.5748);//1.402);
*((uint32_t*)(mCoefTbl+8)) = (int)(65536*0.1873);//0.344);
*((uint32_t*)(mCoefTbl+12)) = (int)(65536*1.8556);//1.772);
clip = mCoefTbl + 400;
/* do 5 bit conversion */
memset( &clip[-384], 0, 385*sizeof( *clip));
memset( &clip[ 640], 0, 385*sizeof( *clip));
for (i=1; i<255; i++){ // range of (x>>3) between -24 and 56
clip[i] = i>>3;
clip[i+1024] = i>>2;
}
memset( &clip[255], 31, 385*sizeof( *clip));
memset( &clip[1279], 63, 385*sizeof( *clip));
}
static int32_t cc16(uint16_t *pY,uint8_t *pCb, uint8_t *pCr, uint8_t
*dst, int32_t *disp, uint8_t *coff_tbl)
{
#define OFFSET_5_0 2
#define OFFSET_6_0 (1+1024)
#define OFFSET_5_1 6
#define OFFSET_6_1 (3+1024)
uint16_t *pDst;
int32_t src_pitch, dst_pitch, src_width;
int32_t Y, Cb, Cr, Cg;
int32_t deltaY, deltaDst, deltaCbCr;
int32_t row, col;
int32_t tmp0, tmp1, tmp2;
uint32_t rgb;
uint8_t *clip = coff_tbl+400;
int32_t cc1 = (*((int32_t*)(clip - 400)));
int32_t cc3 = (*((int32_t*)(clip - 396)));
int32_t cc2 = (*((int32_t*)(clip - 392)));
int32_t cc4 = (*((int32_t*)(clip - 388)));
src_pitch = disp[0];
dst_pitch = disp[1];
src_width = disp[2];
if(disp[6]) /* rotate 180 and flip */
{ /* move the starting point to the bottom-left corner of the
picture */
deltaY = src_pitch*(disp[3]-1);
deltaY = (src_pitch>>1)*((disp[3]>>1)-1);
deltaY = -src_width-(src_pitch<<1);
deltaCbCr = -((src_width+src_pitch)>>1);
src_pitch = -(src_pitch>>1);
}
else
{
deltaY = (src_pitch<<1)-src_width;
deltaCbCr = (src_pitch-src_width)>>1;
src_pitch >>= 1;
}
deltaDst = (dst_pitch<<1)-src_width;
pDst = (uint16_t *)dst;
for(row = disp[3]; row >0; row-=2){
for(col = src_width-1; col >=0; col-=2){
Cb = *pCb++; Cr = *pCr++;
Y = pY[src_pitch];
Cb -= 128; Cr -= 128;
Cg = Cr*cc1;
Cr *= cc3;
Cg += Cb*cc2;
Cb *= cc4;
tmp0 = (Y & 0xFF); //Low endian left pixel
tmp0 += OFFSET_5_0;
tmp1 = tmp0 - (Cg>>16);
tmp2 = tmp0 + (Cb>>16);
tmp0 = tmp0 + (Cr>>16);
tmp0 = clip[tmp0];
tmp1 = clip[tmp1 + OFFSET_6_0 - OFFSET_5_0];
tmp2 = clip[tmp2];
//RGB_565
rgb = tmp1|(tmp0<<6);
rgb = tmp2|(rgb<<5);
Y = (Y>>8) & 0xFF;
Y += OFFSET_5_1;
tmp1 = (Y) - (Cg>>16);
tmp2 = (Y) + (Cb>>16);
tmp0 = (Y) + (Cr>>16);
tmp0 = clip[tmp0];
tmp1 = clip[tmp1 + OFFSET_6_1 - OFFSET_5_1];
tmp2 = clip[tmp2];
//RGB_565
tmp0 = tmp1|(tmp0<<6);
tmp0 = tmp2|(tmp0<<5);
rgb |= (tmp0<<16);
*( (uint32_t*)(pDst+dst_pitch) ) = rgb;
//load the top two pixels
Y = *pY++;
tmp0 = (Y & 0xFF); //Low endian left pixel
tmp0 += OFFSET_5_1;
tmp1 = tmp0 - (Cg>>16);
tmp2 = tmp0 + (Cb>>16);
tmp0 = tmp0 + (Cr>>16);
tmp0 = clip[tmp0];
tmp1 = clip[tmp1 + OFFSET_6_1 - OFFSET_5_1];
tmp2 = clip[tmp2];
//RGB_565
rgb = tmp1|(tmp0<<6);
rgb = tmp2|(rgb<<5);
Y = (Y>>8) & 0xFF;
Y += OFFSET_5_0;
tmp1 = (Y) - (Cg>>16);
tmp2 = (Y) + (Cb>>16);
tmp0 = (Y) + (Cr>>16);
tmp0 = clip[tmp0];
tmp1 = clip[tmp1 + OFFSET_6_0 - OFFSET_5_0];
tmp2 = clip[tmp2];
//RGB_565
tmp0 = tmp1|(tmp0<<6);
tmp0 = tmp2|(tmp0<<5);
rgb |= (tmp0<<16);
*( (uint32_t *)pDst) = rgb; pDst+=2;
}//end of COL
pY += (deltaY>>1);
pCb += deltaCbCr;
pCr += deltaCbCr;
pDst+= (deltaDst); //coz pDst defined as UINT *
}
return 1;
}
On 11月19日, 下午5时02分, "
forest...@gmail.com" <
forest...@gmail.com> wrote:
> thanks very much,cpu is pxa300,camera only provides yuv422 & RawRGB,
>
> "On the G1 we use another code path to draw video frames" which
> path,thanks~
>
> On 11月19日, 下午4时07分, Mathias Agopian <
pixelflin...@google.com> wrote:
>
>
>
> > Hi,
>
> > On Tue, Nov 18, 2008 at 11:15 PM,
forest...@gmail.com
>
> > <
forest...@gmail.com> wrote:
>
> > > the camera provide yuv422 data,butopenGL"just show the Y plane of
> > > YUV buffers"(frameworks\base\libs\surfaceflinger\LayerBase.cpp
> > > 624),must convert yuv422 to rgb565,camera preview is ok,a little
> > > slow,why,thanks
>
> > No,OpenGLdoesn't support YUV textures. That's just that.
>
> > On the G1 we use another code path to draw video frames, it doesn't
> > useOpenGLES, instead it uses the 2D engine. The code that you are
> > referring to exists only so that the YUV video node will display
> > "something" when used on the emulator or on a device that doesn't do
> > YUV (in which case the video node should not try to produce YUV
> > output). It's mostly there for debugging.
>
> > Of course, *some* YUV extensions forOpenGLES exist, and the code
> > in SurfaceFlinger doesn't make use of them at this time. Exercise left
> > to the reader... or until real h/w with these capabilities falls into
> > my hands.
>
> > What h/w are you using?
>
> > Mathias- 隐藏被引用文字 -
>
> - 显示引用的文字 -