With explicit unrolling, we can skip half of the sign bit flips, and
the compiler is then better able to optimise the scalar loop:
predictor_c: 31376.0 (before)
predictor_c: 23703.0 (after)
---
libavcodec/exrdsp.c | 16 +++++++++++++---
1 file changed, 13 insertions(+), 3 deletions(-)
diff --git a/libavcodec/exrdsp.c b/libavcodec/exrdsp.c
index 752e1eb553..248cb93c5a 100644
--- a/libavcodec/exrdsp.c
+++ b/libavcodec/exrdsp.c
@@ -40,10 +40,20 @@ static void reorder_pixels_scalar(uint8_t *dst, const
uint8_t *src, ptrdiff_t si
static void predictor_scalar(uint8_t *src, ptrdiff_t size)
{
- ptrdiff_t i;
+ /* Unrolled: `src[i + 1] += src[i] - 128;` */
+ if ((size & 1) == 0) {
+ src[1] += src[0] ^ 0x80;
+ src++;
+ size--;
+ }
+
+ for (ptrdiff_t i = 1; i < size; i += 2) {
+ uint8_t a = src[i] + src[i - 1];
- for (i = 1; i < size; i++)
- src[i] += src[i-1] - 128;
+ src[i] = a;
+ src[i + 1] += a;
+ src[i] ^= 0x80;
+ }
}
av_cold void ff_exrdsp_init(ExrDSPContext *c)
--
2.42.0
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".
