This is a part of YUV422 image resize function,
#pragma MUST_ITERATE(30,60,2)
for(k = 0; k < height/4; k++)
{
memcpy(g_acBuffA,(src->imageData+k*iSrcBufsz),iSrcBufsz); //(1)
psrc = g_acBuffA; //g_acBuffA is located in L1D SRAM
pdst = g_acBuffB; //g_acBuffB is located in L1D SRAM
//(2)
#pragma MUST_ITERATE(4,4,4)
for (j = 0; j < 4; j++) {
#pragma MUST_ITERATE(358,718,2)
for (i = 0; i < width - 2; i++)
{
_mem4(pdst) = _amem4_const(psrc);
pdst += 3;
psrc += 4;
}
//last two iteration
{
_mem4(pdst) = _amem4_const(psrc);
pdst += 3;
psrc += 4;
*pdst++ = *psrc++;
*pdst++ = *psrc++;
*pdst++ = *psrc++;
psrc += 1;
}
psrc += s_step;
}
//(3)
memcpy((dst->imageData+k*iDesBufsz),g_acBuffB,iDesBufsz); //(4)
}
(1) Copy source image data from DDR to L1D SRAM buffer A
(2)~(3) Resize image and store it to L1D SRAM buffer B
(4) Copy L1D SRAM buffer B to DDR
I used the CCSV3.3 profile tool and found (2) ~(3) has Large L1D Miss.
How can I get a high L1D hits?
Thanks
