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

ARGBToUV 64 bit use ymm8 for shuffler

Browse Source 
Bug: 381138208
Change-Id: I5e69bc1610bd6269bf9a4113e729cf307dd36f60
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/6536833
Reviewed-by: richard winterton <rrwinterton@gmail.com>
This commit is contained in:
Frank Barchard 2025-05-12 15:01:02 -07:00
parent 61bdaee13a
commit 0853c9353f
6 changed files with 575 additions and 568 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.chromium
View File

@ -1,6 +1,6 @@
Name: libyuv Name: libyuv
URL: https://chromium.googlesource.com/libyuv/libyuv/ URL: https://chromium.googlesource.com/libyuv/libyuv/
Version: 1909 Version: 1910
License: BSD-3-Clause License: BSD-3-Clause
License File: LICENSE License File: LICENSE
Shipped: yes Shipped: yes

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ #ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1909 #define LIBYUV_VERSION 1910
#endif // INCLUDE_LIBYUV_VERSION_H_ #endif // INCLUDE_LIBYUV_VERSION_H_

3
source/compare_neon64.cc
View File

@ -116,7 +116,8 @@ uint32_t HashDjb2_NEON(const uint8_t* src, int count, uint32_t seed) {
uint32_t hash = seed; uint32_t hash = seed;
const uint32_t c16 = 0x92d9e201; // 33^16 const uint32_t c16 = 0x92d9e201; // 33^16
uint32_t tmp, tmp2; uint32_t tmp, tmp2;
asm("ld1 {v16.4s, v17.4s, v18.4s, v19.4s}, [%[kIdx]] \n" asm(
"ld1 {v16.4s, v17.4s, v18.4s, v19.4s}, [%[kIdx]] \n"
"ld1 {v4.4s, v5.4s, v6.4s, v7.4s}, [%[kMuls]] \n" "ld1 {v4.4s, v5.4s, v6.4s, v7.4s}, [%[kMuls]] \n"
// count is always a multiple of 16. // count is always a multiple of 16.

1018
source/row_gcc.cc
View File

File diff suppressed because it is too large Load Diff

56
source/row_neon64.cc
View File

@ -291,12 +291,12 @@ void I210ToAR30Row_NEON(const uint16_t* src_y,
uint16_t limit = 0x3ff0; uint16_t limit = 0x3ff0;
uint16_t alpha = 0xc000; uint16_t alpha = 0xc000;
asm volatile(YUVTORGB_SETUP asm volatile(YUVTORGB_SETUP
"dup v22.8h, %w[limit] \n" "dup v22.8h, %w[limit] \n"
"dup v23.8h, %w[alpha] \n" "dup v23.8h, %w[alpha] \n"
"1: \n" // "1: \n" //
READYUV210 READYUV210
"subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30 "subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30
"b.gt 1b \n" "b.gt 1b \n"
: [src_y] "+r"(src_y), // %[src_y] : [src_y] "+r"(src_y), // %[src_y]
[src_u] "+r"(src_u), // %[src_u] [src_u] "+r"(src_u), // %[src_u]
[src_v] "+r"(src_v), // %[src_v] [src_v] "+r"(src_v), // %[src_v]
@ -320,12 +320,12 @@ void I410ToAR30Row_NEON(const uint16_t* src_y,
uint16_t limit = 0x3ff0; uint16_t limit = 0x3ff0;
uint16_t alpha = 0xc000; uint16_t alpha = 0xc000;
asm volatile(YUVTORGB_SETUP asm volatile(YUVTORGB_SETUP
"dup v22.8h, %w[limit] \n" "dup v22.8h, %w[limit] \n"
"dup v23.8h, %w[alpha] \n" "dup v23.8h, %w[alpha] \n"
"1: \n" // "1: \n" //
READYUV410 READYUV410
"subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30 "subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30
"b.gt 1b \n" "b.gt 1b \n"
: [src_y] "+r"(src_y), // %[src_y] : [src_y] "+r"(src_y), // %[src_y]
[src_u] "+r"(src_u), // %[src_u] [src_u] "+r"(src_u), // %[src_u]
[src_v] "+r"(src_v), // %[src_v] [src_v] "+r"(src_v), // %[src_v]
@ -348,12 +348,12 @@ void I212ToAR30Row_NEON(const uint16_t* src_y,
const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias; const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias;
const uint16_t limit = 0x3ff0; const uint16_t limit = 0x3ff0;
asm volatile(YUVTORGB_SETUP asm volatile(YUVTORGB_SETUP
"dup v22.8h, %w[limit] \n" "dup v22.8h, %w[limit] \n"
"movi v23.8h, #0xc0, lsl #8 \n" // A "movi v23.8h, #0xc0, lsl #8 \n" // A
"1: \n" // "1: \n" //
READYUV212 READYUV212
"subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30 "subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30
"b.gt 1b \n" "b.gt 1b \n"
: [src_y] "+r"(src_y), // %[src_y] : [src_y] "+r"(src_y), // %[src_y]
[src_u] "+r"(src_u), // %[src_u] [src_u] "+r"(src_u), // %[src_u]
[src_v] "+r"(src_v), // %[src_v] [src_v] "+r"(src_v), // %[src_v]
@ -530,13 +530,13 @@ void P210ToAR30Row_NEON(const uint16_t* src_y,
const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias; const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias;
const uint16_t limit = 0x3ff0; const uint16_t limit = 0x3ff0;
asm volatile(YUVTORGB_SETUP asm volatile(YUVTORGB_SETUP
"dup v22.8h, %w[limit] \n" "dup v22.8h, %w[limit] \n"
"movi v23.8h, #0xc0, lsl #8 \n" // A "movi v23.8h, #0xc0, lsl #8 \n" // A
"ldr q2, [%[kIndices]] \n" "ldr q2, [%[kIndices]] \n"
"1: \n" // "1: \n" //
READYUVP210 READYUVP210
"subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30 "subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30
"b.gt 1b \n" "b.gt 1b \n"
: [src_y] "+r"(src_y), // %[src_y] : [src_y] "+r"(src_y), // %[src_y]
[src_uv] "+r"(src_uv), // %[src_uv] [src_uv] "+r"(src_uv), // %[src_uv]
[dst_ar30] "+r"(dst_ar30), // %[dst_ar30] [dst_ar30] "+r"(dst_ar30), // %[dst_ar30]
@ -557,13 +557,13 @@ void P410ToAR30Row_NEON(const uint16_t* src_y,
const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias; const vec16* rgb_coeff = &yuvconstants->kRGBCoeffBias;
uint16_t limit = 0x3ff0; uint16_t limit = 0x3ff0;
asm volatile(YUVTORGB_SETUP asm volatile(YUVTORGB_SETUP
"dup v22.8h, %w[limit] \n" "dup v22.8h, %w[limit] \n"
"movi v23.8h, #0xc0, lsl #8 \n" // A "movi v23.8h, #0xc0, lsl #8 \n" // A
"ldr q2, [%[kIndices]] \n" "ldr q2, [%[kIndices]] \n"
"1: \n" // "1: \n" //
READYUVP410 READYUVP410
"subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30 "subs %w[width], %w[width], #8 \n" NVTORGB STOREAR30
"b.gt 1b \n" "b.gt 1b \n"
: [src_y] "+r"(src_y), // %[src_y] : [src_y] "+r"(src_y), // %[src_y]
[src_uv] "+r"(src_uv), // %[src_uv] [src_uv] "+r"(src_uv), // %[src_uv]
[dst_ar30] "+r"(dst_ar30), // %[dst_ar30] [dst_ar30] "+r"(dst_ar30), // %[dst_ar30]
@ -3461,7 +3461,7 @@ static void ABCDToUVMatrixRow_NEON_I8MM(const uint8_t* src,
// 16-bit) // 16-bit)
"ld2r {v24.4s, v25.4s}, [%[uvconstants]] \n" "ld2r {v24.4s, v25.4s}, [%[uvconstants]] \n"
"1: \n" "1: \n"
"ld2 {v0.4s, v1.4s}, [%[src]], #32 \n" // load 8 pixels "ld2 {v0.4s, v1.4s}, [%[src]], #32 \n" // load 8 pixels
"ld2 {v2.4s, v3.4s}, [%[src]], #32 \n" // load 8 pixels "ld2 {v2.4s, v3.4s}, [%[src]], #32 \n" // load 8 pixels
"subs %w[width], %w[width], #16 \n" // 16 processed per loop "subs %w[width], %w[width], #16 \n" // 16 processed per loop

62
source/scale_gcc.cc
View File

@ -1759,25 +1759,25 @@ void ScaleRowUp2_Bilinear_16_AVX2(const uint16_t* src_ptr,
void ScaleAddRow_SSE2(const uint8_t* src_ptr, void ScaleAddRow_SSE2(const uint8_t* src_ptr,
uint16_t* dst_ptr, uint16_t* dst_ptr,
int src_width) { int src_width) {
asm volatile("pxor %%xmm5,%%xmm5 \n" asm volatile("pxor %%xmm5,%%xmm5 \n"
// 16 pixel loop. // 16 pixel loop.
LABELALIGN LABELALIGN
"1: \n" "1: \n"
"movdqu (%0),%%xmm3 \n" "movdqu (%0),%%xmm3 \n"
"lea 0x10(%0),%0 \n" // src_ptr += 16 "lea 0x10(%0),%0 \n" // src_ptr += 16
"movdqu (%1),%%xmm0 \n" "movdqu (%1),%%xmm0 \n"
"movdqu 0x10(%1),%%xmm1 \n" "movdqu 0x10(%1),%%xmm1 \n"
"movdqa %%xmm3,%%xmm2 \n" "movdqa %%xmm3,%%xmm2 \n"
"punpcklbw %%xmm5,%%xmm2 \n" "punpcklbw %%xmm5,%%xmm2 \n"
"punpckhbw %%xmm5,%%xmm3 \n" "punpckhbw %%xmm5,%%xmm3 \n"
"paddusw %%xmm2,%%xmm0 \n" "paddusw %%xmm2,%%xmm0 \n"
"paddusw %%xmm3,%%xmm1 \n" "paddusw %%xmm3,%%xmm1 \n"
"movdqu %%xmm0,(%1) \n" "movdqu %%xmm0,(%1) \n"
"movdqu %%xmm1,0x10(%1) \n" "movdqu %%xmm1,0x10(%1) \n"
"lea 0x20(%1),%1 \n" "lea 0x20(%1),%1 \n"
"sub $0x10,%2 \n" "sub $0x10,%2 \n"
"jg 1b \n" "jg 1b \n"
: "+r"(src_ptr), // %0 : "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1 "+r"(dst_ptr), // %1
"+r"(src_width) // %2 "+r"(src_width) // %2
@ -1790,23 +1790,23 @@ void ScaleAddRow_SSE2(const uint8_t* src_ptr,
void ScaleAddRow_AVX2(const uint8_t* src_ptr, void ScaleAddRow_AVX2(const uint8_t* src_ptr,
uint16_t* dst_ptr, uint16_t* dst_ptr,
int src_width) { int src_width) {
asm volatile("vpxor %%ymm5,%%ymm5,%%ymm5 \n" asm volatile("vpxor %%ymm5,%%ymm5,%%ymm5 \n"
LABELALIGN LABELALIGN
"1: \n" "1: \n"
"vmovdqu (%0),%%ymm3 \n" "vmovdqu (%0),%%ymm3 \n"
"lea 0x20(%0),%0 \n" // src_ptr += 32 "lea 0x20(%0),%0 \n" // src_ptr += 32
"vpermq $0xd8,%%ymm3,%%ymm3 \n" "vpermq $0xd8,%%ymm3,%%ymm3 \n"
"vpunpcklbw %%ymm5,%%ymm3,%%ymm2 \n" "vpunpcklbw %%ymm5,%%ymm3,%%ymm2 \n"
"vpunpckhbw %%ymm5,%%ymm3,%%ymm3 \n" "vpunpckhbw %%ymm5,%%ymm3,%%ymm3 \n"
"vpaddusw (%1),%%ymm2,%%ymm0 \n" "vpaddusw (%1),%%ymm2,%%ymm0 \n"
"vpaddusw 0x20(%1),%%ymm3,%%ymm1 \n" "vpaddusw 0x20(%1),%%ymm3,%%ymm1 \n"
"vmovdqu %%ymm0,(%1) \n" "vmovdqu %%ymm0,(%1) \n"
"vmovdqu %%ymm1,0x20(%1) \n" "vmovdqu %%ymm1,0x20(%1) \n"
"lea 0x40(%1),%1 \n" "lea 0x40(%1),%1 \n"
"sub $0x20,%2 \n" "sub $0x20,%2 \n"
"jg 1b \n" "jg 1b \n"
"vzeroupper \n" "vzeroupper \n"
: "+r"(src_ptr), // %0 : "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1 "+r"(dst_ptr), // %1
"+r"(src_width) // %2 "+r"(src_width) // %2