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] Add SVE2 implementation of I422ToRGB24Row

Browse Source 
Observed reduction in runtime compared to the existing Neon code:

Cortex-A510: -57.8%
Cortex-A520: -41.7%
Cortex-A715: -28.0%
Cortex-A720: -28.1%
  Cortex-X2: -29.7%
  Cortex-X3: -28.7%
  Cortex-X4: -30.5%
Cortex-X925: -30.3%

Bug: b/42280942
Change-Id: I328bd16babda75fb089c8da8f2714465f658187e
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/5802965
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: Justin Green <greenjustin@google.com>
This commit is contained in:
George Steed 2024-05-03 15:22:23 +01:00 committed by Frank Barchard
parent 6ac7c8f251
commit 0dce974ca0
3 changed files with 62 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
include/libyuv/row.h
View File

@ -551,6 +551,7 @@ extern "C" {
#define HAS_I400TOARGBROW_SVE2
#define HAS_I422ALPHATOARGBROW_SVE2
#define HAS_I422TOARGBROW_SVE2
#define HAS_I422TORGB24ROW_SVE2
#define HAS_I422TORGBAROW_SVE2
#define HAS_I444ALPHATOARGBROW_SVE2
#define HAS_I444TOARGBROW_SVE2
@ -1168,6 +1169,12 @@ void I422ToRGB24Row_NEON(const uint8_t* src_y,
uint8_t* dst_rgb24,
const struct YuvConstants* yuvconstants,
int width);
void I422ToRGB24Row_SVE2(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
uint8_t* dst_rgb24,
const struct YuvConstants* yuvconstants,
int width);
void I422ToRGB565Row_NEON(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,

10
source/convert_argb.cc
View File

@ -5362,6 +5362,11 @@ int I420ToRGB24Matrix(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422TORGB24ROW_SVE2)
if (TestCpuFlag(kCpuHasSVE2)) {
I422ToRGB24Row = I422ToRGB24Row_SVE2;
}
#endif
#if defined(HAS_I422TORGB24ROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
I422ToRGB24Row = I422ToRGB24Row_Any_MSA;
@ -5564,6 +5569,11 @@ int I422ToRGB24Matrix(const uint8_t* src_y,
}
}
#endif
#if defined(HAS_I422TORGB24ROW_SVE2)
if (TestCpuFlag(kCpuHasSVE2)) {
I422ToRGB24Row = I422ToRGB24Row_SVE2;
}
#endif
#if defined(HAS_I422TORGB24ROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
I422ToRGB24Row = I422ToRGB24Row_Any_MSA;

45
source/row_sve.cc
View File

@ -332,6 +332,51 @@ void I422ToARGBRow_SVE2(const uint8_t* src_y,
: "cc", "memory", YUVTORGB_SVE_REGS);
}
void I422ToRGB24Row_SVE2(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,
uint8_t* dst_argb,
const struct YuvConstants* yuvconstants,
int width) {
uint64_t vl;
asm volatile(
"cntb %[vl] \n"
"ptrue p0.b \n" YUVTORGB_SVE_SETUP
"subs %w[width], %w[width], %w[vl] \n"
"b.lt 2f \n"
// Run bulk of computation with an all-true predicate to avoid predicate
// generation overhead.
"ptrue p1.b \n"
"1: \n" READYUV422_SVE_2X
I422TORGB_SVE_2X RGBTOARGB8_SVE_2X
"subs %w[width], %w[width], %w[vl] \n"
"st3b {z16.b, z17.b, z18.b}, p1, [%[dst_argb]] \n"
"incb %[dst_argb], all, mul #3 \n"
"b.ge 1b \n"
"2: \n"
"adds %w[width], %w[width], %w[vl] \n"
"b.eq 99f \n"
// Calculate a predicate for the final iteration to deal with the tail.
"cnth %[vl] \n"
"whilelt p1.b, wzr, %w[width] \n" READYUV422_SVE_2X
I422TORGB_SVE_2X RGBTOARGB8_SVE_2X
"st3b {z16.b, z17.b, z18.b}, p1, [%[dst_argb]] \n"
"99: \n"
: [src_y] "+r"(src_y), // %[src_y]
[src_u] "+r"(src_u), // %[src_u]
[src_v] "+r"(src_v), // %[src_v]
[dst_argb] "+r"(dst_argb), // %[dst_argb]
[width] "+r"(width), // %[width]
[vl] "=&r"(vl) // %[vl]
: [kUVCoeff] "r"(&yuvconstants->kUVCoeff), // %[kUVCoeff]
[kRGBCoeffBias] "r"(&yuvconstants->kRGBCoeffBias) // %[kRGBCoeffBias]
: "cc", "memory", YUVTORGB_SVE_REGS);
}
void I422ToRGBARow_SVE2(const uint8_t* src_y,
const uint8_t* src_u,
const uint8_t* src_v,