coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 16:56:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

[AArch64] Replace UQXTN{,2} with UZP2 in Convert16To8Row_NEON

Browse Source 
The existing code makes use of a pair of shifts to put the bits we want
in the low part of each vector lane and then a pair of UQXTN and UQXTN2
instructions to perform a saturating cast down from 16-bit elements to
8-bit elements. We can instead achieve the same thing by adding eight to
the first shift amount so that the bits we want appear in the high half
of the lane, doing the saturation at the same time, and then simply use
UZP2 to pull out the high halves of each lane in a single instruction.

Reduction in runtime for Convert16To8Row_NEON:

  Cortex-A55: -19.7%
 Cortex-A510: -23.5%
  Cortex-A76: -35.4%
   Cortex-X2: -34.1%

Bug: libyuv:976
Change-Id: I9a80c0f4f2c6b5203f23e422c0970d3167052f91
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/5463950
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
George Steed 2024-03-14 05:12:21 +00:00 committed by Frank Barchard
parent 4f52235a67
commit 356232b687
1 changed files with 7 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
source/row_neon64.cc
View File

@ -4701,16 +4701,18 @@ void Convert16To8Row_NEON(const uint16_t* src_y,
uint8_t* dst_y, uint8_t* dst_y,
int scale, int scale,
int width) { int width) {
int shift = 15 - __builtin_clz((int32_t)scale); // Negative shl is shr // 15 - clz(scale), + 8 to shift result into the high half of the lane to
// saturate, then we can just use UZP2 to narrow rather than a pair of
// saturating narrow instructions.
int shift = 23 - __builtin_clz((int32_t)scale);
asm volatile( asm volatile(
"dup v2.8h, %w3 \n" "dup v2.8h, %w3 \n"
"1: \n" "1: \n"
"ldp q0, q1, [%0], #32 \n" "ldp q0, q1, [%0], #32 \n"
"ushl v0.8h, v0.8h, v2.8h \n" // shr = v2 is negative "uqshl v0.8h, v0.8h, v2.8h \n"
"ushl v1.8h, v1.8h, v2.8h \n" "uqshl v1.8h, v1.8h, v2.8h \n"
"prfm pldl1keep, [%0, 448] \n" "prfm pldl1keep, [%0, 448] \n"
"uqxtn v0.8b, v0.8h \n" "uzp2 v0.16b, v0.16b, v1.16b \n"
"uqxtn2 v0.16b, v1.8h \n"
"subs %w2, %w2, #16 \n" // 16 src pixels per loop "subs %w2, %w2, #16 \n" // 16 src pixels per loop
"str q0, [%1], #16 \n" // store 16 pixels "str q0, [%1], #16 \n" // store 16 pixels
"b.gt 1b \n" "b.gt 1b \n"