coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 08:46:47 +08:00
Code Issues Packages Projects Releases Wiki Activity

[AArch64] Add Neon implementation of ScaleRowDown2Linear_16

Browse Source 
Reduction in runtime observed relative to the auto-vectorized C
implementation compiled with LLVM 19:

  Cortex-A55: -13.7%
 Cortex-A510: -49.0%
 Cortex-A520: -32.0%
  Cortex-A76: -34.3%
 Cortex-A710: -56.7%
 Cortex-A715: -45.4%
 Cortex-A720: -44.7%
   Cortex-X1: -70.6%
   Cortex-X2: -67.9%
   Cortex-X3: -72.2%
   Cortex-X4: -40.0%
 Cortex-X925: -24.1%

Bug: b/42280942
Change-Id: I977899a2239e752400c9901f4d8482a76841269a
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/6040154
Reviewed-by: Justin Green <greenjustin@google.com>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
George Steed 2024-11-15 15:32:56 +00:00 committed by Frank Barchard
parent 11c57f4f12
commit 9a9752134e
3 changed files with 29 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
include/libyuv/scale_row.h
View File

@ -1440,6 +1440,10 @@ void ScaleRowDown2Linear_NEON(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width);
void ScaleRowDown2Linear_16_NEON(const uint16_t* src_ptr,
ptrdiff_t src_stride,
uint16_t* dst,
int dst_width);
void ScaleRowDown2Linear_SME(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,

5
source/scale.cc
View File

@ -188,8 +188,9 @@ static void ScalePlaneDown2_16(int src_width,
#if defined(HAS_SCALEROWDOWN2_16_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(dst_width, 16)) {
ScaleRowDown2 =
filtering ? ScaleRowDown2Box_16_NEON : ScaleRowDown2_16_NEON;
ScaleRowDown2 = filtering == kFilterNone ? ScaleRowDown2_16_NEON
: filtering == kFilterLinear ? ScaleRowDown2Linear_16_NEON
: ScaleRowDown2Box_16_NEON;
}
#endif
#if defined(HAS_SCALEROWDOWN2_16_SSE2)

22
source/scale_neon64.cc
View File

@ -1399,6 +1399,28 @@ void ScaleRowDown2_16_NEON(const uint16_t* src_ptr,
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7");
}
void ScaleRowDown2Linear_16_NEON(const uint16_t* src_ptr,
ptrdiff_t src_stride,
uint16_t* dst,
int dst_width) {
(void)src_stride;
asm volatile(
"1: \n"
"ld2 {v0.8h, v1.8h}, [%[src_ptr]], #32 \n"
"ld2 {v2.8h, v3.8h}, [%[src_ptr]], #32 \n"
"subs %w[dst_width], %w[dst_width], #16 \n"
"urhadd v0.8h, v0.8h, v1.8h \n"
"urhadd v1.8h, v2.8h, v3.8h \n"
"prfm pldl1keep, [%[src_ptr], 448] \n"
"stp q0, q1, [%[dst_ptr]], #32 \n"
"b.gt 1b \n"
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
[dst_ptr] "+r"(dst), // %[dst_ptr]
[dst_width] "+r"(dst_width) // %[dst_width]
:
: "memory", "cc", "v0", "v1", "v2", "v3");
}
// Read 16x2 average down and write 8x1.
void ScaleRowDown2Box_16_NEON(const uint16_t* src_ptr,
ptrdiff_t src_stride,