emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
- // NOTE: the +1 is for the MMX scaler which reads over the end
- FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail);
+ // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
+ FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
if (FFABS(xInc - 0x10000) <10) { // unscaled
int i;
// Note the +1 is for the MMX scaler which reads over the end
/* align at 16 for AltiVec (needed by hScale_altivec_real) */
- FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail);
+ FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
/* normalize & store in outFilter */
for (i=0; i<dstW; i++) {
}
}
- (*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end
+ (*filterPos)[dstW+0] =
+ (*filterPos)[dstW+1] =
+ (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
for (i=0; i<*outFilterSize; i++) {
- int j= dstW*(*outFilterSize);
- (*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)];
+ int k= (dstW - 1) * (*outFilterSize) + i;
+ (*outFilter)[k + 1 * (*outFilterSize)] =
+ (*outFilter)[k + 2 * (*outFilterSize)] =
+ (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
}
ret=0;
}
}
- c->scalingBpp = FFMAX(av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1,
- av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1) >= 8 ? 16 : 8;
- if (c->scalingBpp == 16)
+ c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
+ if (c->srcBpc < 8)
+ c->srcBpc = 8;
+ c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
+ if (c->dstBpc < 8)
+ c->dstBpc = 8;
+ if (c->dstBpc == 16)
dst_stride <<= 1;
- FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW, 16) * 2 * c->scalingBpp >> 3, fail);
- if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->scalingBpp == 8) {
+ FF_ALLOC_OR_GOTO(c, c->formatConvBuffer,
+ FFALIGN(srcW, 16) * 2 * FFALIGN(c->srcBpc, 8) >> 3,
+ fail);
+ if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
if (flags&SWS_PRINT_INFO)
FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
}
- // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
- c->uv_off_px = dst_stride_px + 64 / c->scalingBpp;
+ // 64 / (c->dstBpc & ~7) is the same as 16 / sizeof(scaling_intermediate)
+ c->uv_off_px = dst_stride_px + 64 / (c->dstBpc &~ 7);
c->uv_off_byte = dst_stride + 16;
for (i=0; i<c->vChrBufSize; i++) {
FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);