[FFmpeg-devel] [PATCH] avcodec/scpr: optimize shift loop.

[Brian Matherly]

On 9/9/2017 1:27 PM, Michael Niedermayer wrote:
> +            // If the image is sufficiently aligned, compute 8 samples at once
> +            if (!(((uintptr_t)dst) & 7)) {
> +                uint64_t *dst64 = (uint64_t *)dst;
> +                int w = avctx->width>>1;
> +                for (x = 0; x < w; x++) {
> +                    dst64[x] = (dst64[x] << 3) & 0xFCFCFCFCFCFCFCFCULL;
> +                }
> +                x *= 8;
> +            } else
> +                x = 0;
> +            for (; x < avctx->width * 4; x++) {
>                   dst[x] = dst[x] << 3;
>               }

Forgive me if I'm not understanding the code correctly, but couldn't you 
always apply the optimization if you align the first (up to) 7 samples?

Pseudocode:

uint64_t *dst64 = (uint64_t *)dst;
int w = avctx->width>>1;
x=0
// compute un-aligned beginning samples
for (; x < (avctx->width * 4) && (((uintptr_t)dst) & 7); x++) {
     dst[x] = dst[x] << 3;
}
// compute aligned samples
for (; x < w; x+=8) {
     dst64[x] = (dst64[x] << 3) & 0xFCFCFCFCFCFCFCFCULL;
}
x -= 8;
// compute un-aligned ending samples
for (; x < avctx->width * 4; x++) {
     dst[x] = dst[x] << 3;
}