coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2026-01-01 03:12:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

[AArch64] Use full Neon vectors in ARGB4444ToARGBRow_NEON

Browse Source 
The existing Neon code narrows the input 16-bit packed data to 8-bit
elements and separates the color channels, causing us to only process
half a Neon vector per instruction for the channel widening from 4-bit
color data to 8-bits.

We can note that the processing being done is identical for all color
channels and therefore we can keep them partially interleaved during the
widening step. This allows us to use full Neon vectors for the whole
loop body.

Reductions in runtimes observed for ARGB4444ToARGBRow_NEON:

 Cortex-A55: -30.7%
Cortex-A510: -44.3%
 Cortex-A76: -51.6%
  Cortex-X2: -54.2%

Bug: libyuv:976
Change-Id: I9d9cda7e16eb07619c6d7f1de2e6b8c0fb6d64cf
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/5594389
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
George Steed 2024-03-23 20:02:41 +00:00 committed by Frank Barchard
parent 7633c818ec
commit dff7bad43d
1 changed files with 7 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
source/row_neon64.cc
View File

@ -1911,16 +1911,11 @@ void ARGB1555ToARGBRow_NEON(const uint8_t* src_argb1555,
);
}
#define ARGB4444TOARGB \
/* Input: v0.8h = AAAARRRRGGGGBBBB */ \
"xtn v1.8b, v0.8h \n" /* GGGGBBBB */ \
"shrn v2.8b, v0.8h, #4 \n" /* RRRRxxxx */ \
"shrn v3.8b, v0.8h, #8 \n" /* AAAAxxxx */ \
"shl v0.8b, v1.8b, #4 \n" /* BBBB0000 */ \
"sri v1.8b, v1.8b, #4 \n" /* GGGGGGGG */ \
"sri v2.8b, v2.8b, #4 \n" /* RRRRRRRR */ \
"sri v3.8b, v3.8b, #4 \n" /* AAAAAAAA */ \
"sri v0.8b, v0.8b, #4 \n" /* BBBBBBBB */
#define ARGB4444TOARGB \
/* Input: v1.8h = AAAARRRR_GGGGBBBB */ \
"shl v0.16b, v1.16b, #4 \n" /* RRRR0000_BBBB0000 */ \
"sri v1.16b, v1.16b, #4 \n" /* AAAAAAAA_GGGGGGGG */ \
"sri v0.16b, v0.16b, #4 \n" /* RRRRRRRR_BBBBBBBB */
#define ARGB4444TORGB \
/* Input: v0.8h = xxxxRRRRGGGGBBBB */ \
@ -1936,10 +1931,10 @@ void ARGB4444ToARGBRow_NEON(const uint8_t* src_argb4444,
int width) {
asm volatile(
"1: \n"
"ld1 {v0.16b}, [%0], #16 \n" // load 8 ARGB4444 pixels.
"ld1 {v1.16b}, [%0], #16 \n" // load 8 ARGB4444 pixels.
"subs %w2, %w2, #8 \n" // 8 processed per loop.
"prfm pldl1keep, [%0, 448] \n" ARGB4444TOARGB
"st4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%1], #32 \n" // store 8 ARGB
"st2 {v0.16b, v1.16b}, [%1], #32 \n" // store 8 ARGB.
"b.gt 1b \n"
: "+r"(src_argb4444), // %0
"+r"(dst_argb), // %1