ANSCENTER/ANSCORE
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update

Browse Source
This commit is contained in:
Tuan Nghia Nguyen
2026-04-22 16:52:40 +10:00
parent fc47a42c6a
commit a0d5c81814
217 changed files with 139100 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

435
3rdparty/libyuv/source/compare.cc vendored Normal file
View File

@@ -0,0 +1,435 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/compare.h"
#include <float.h>
#include <math.h>
#ifdef _OPENMP
#include <omp.h>
#endif
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#include "libyuv/cpu_id.h"
#include "libyuv/row.h"
#include "libyuv/video_common.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// hash seed of 5381 recommended.
LIBYUV_API
uint32_t HashDjb2(const uint8_t* src, uint64_t count, uint32_t seed) {
const int kBlockSize = 1 << 15; // 32768;
int remainder;
uint32_t (*HashDjb2_SSE)(const uint8_t* src, int count, uint32_t seed) =
HashDjb2_C;
#if defined(HAS_HASHDJB2_SSE41)
if (TestCpuFlag(kCpuHasSSE41)) {
HashDjb2_SSE = HashDjb2_SSE41;
}
#endif
#if defined(HAS_HASHDJB2_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
HashDjb2_SSE = HashDjb2_AVX2;
}
#endif
#if defined(HAS_HASHDJB2_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
HashDjb2_SSE = HashDjb2_NEON;
}
#endif
while (count >= (uint64_t)kBlockSize) {
seed = HashDjb2_SSE(src, kBlockSize, seed);
src += kBlockSize;
count -= kBlockSize;
}
remainder = (int)count & ~15;
if (remainder) {
seed = HashDjb2_SSE(src, remainder, seed);
src += remainder;
count -= remainder;
}
remainder = (int)count & 15;
if (remainder) {
seed = HashDjb2_C(src, remainder, seed);
}
return seed;
}
static uint32_t ARGBDetectRow_C(const uint8_t* argb, int width) {
int x;
for (x = 0; x < width - 1; x += 2) {
if (argb[0] != 255) { // First byte is not Alpha of 255, so not ARGB.
return FOURCC_BGRA;
}
if (argb[3] != 255) { // Fourth byte is not Alpha of 255, so not BGRA.
return FOURCC_ARGB;
}
if (argb[4] != 255) { // Second pixel first byte is not Alpha of 255.
return FOURCC_BGRA;
}
if (argb[7] != 255) { // Second pixel fourth byte is not Alpha of 255.
return FOURCC_ARGB;
}
argb += 8;
}
if (width & 1) {
if (argb[0] != 255) { // First byte is not Alpha of 255, so not ARGB.
return FOURCC_BGRA;
}
if (argb[3] != 255) { // 4th byte is not Alpha of 255, so not BGRA.
return FOURCC_ARGB;
}
}
return 0;
}
// Scan an opaque argb image and return fourcc based on alpha offset.
// Returns FOURCC_ARGB, FOURCC_BGRA, or 0 if unknown.
LIBYUV_API
uint32_t ARGBDetect(const uint8_t* argb,
int stride_argb,
int width,
int height) {
uint32_t fourcc = 0;
int h;
// Coalesce rows.
if (stride_argb == width * 4) {
width *= height;
height = 1;
stride_argb = 0;
}
for (h = 0; h < height && fourcc == 0; ++h) {
fourcc = ARGBDetectRow_C(argb, width);
argb += stride_argb;
}
return fourcc;
}
// NEON version accumulates in 16 bit shorts which overflow at 65536 bytes.
// So actual maximum is 1 less loop, which is 64436 - 32 bytes.
LIBYUV_API
uint64_t ComputeHammingDistance(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
const int kBlockSize = 1 << 15; // 32768;
const int kSimdSize = 64;
// SIMD for multiple of 64, and C for remainder
int remainder = count & (kBlockSize - 1) & ~(kSimdSize - 1);
uint64_t diff = 0;
int i;
uint32_t (*HammingDistance)(const uint8_t* src_a, const uint8_t* src_b,
int count) = HammingDistance_C;
#if defined(HAS_HAMMINGDISTANCE_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
HammingDistance = HammingDistance_NEON;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_NEON_DOTPROD)
if (TestCpuFlag(kCpuHasNeonDotProd)) {
HammingDistance = HammingDistance_NEON_DotProd;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
HammingDistance = HammingDistance_SSSE3;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_SSE42)
if (TestCpuFlag(kCpuHasSSE42)) {
HammingDistance = HammingDistance_SSE42;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
HammingDistance = HammingDistance_AVX2;
}
#endif
#ifdef _OPENMP
#pragma omp parallel for reduction(+ : diff)
#endif
for (i = 0; i < (count - (kBlockSize - 1)); i += kBlockSize) {
diff += HammingDistance(src_a + i, src_b + i, kBlockSize);
}
src_a += count & ~(kBlockSize - 1);
src_b += count & ~(kBlockSize - 1);
if (remainder) {
diff += HammingDistance(src_a, src_b, remainder);
src_a += remainder;
src_b += remainder;
}
remainder = count & (kSimdSize - 1);
if (remainder) {
diff += HammingDistance_C(src_a, src_b, remainder);
}
return diff;
}
// TODO(fbarchard): Refactor into row function.
LIBYUV_API
uint64_t ComputeSumSquareError(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
// SumSquareError returns values 0 to 65535 for each squared difference.
// Up to 65536 of those can be summed and remain within a uint32_t.
// After each block of 65536 pixels, accumulate into a uint64_t.
const int kBlockSize = 65536;
int remainder = count & (kBlockSize - 1) & ~31;
uint64_t sse = 0;
int i;
uint32_t (*SumSquareError)(const uint8_t* src_a, const uint8_t* src_b,
int count) = SumSquareError_C;
#if defined(HAS_SUMSQUAREERROR_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
SumSquareError = SumSquareError_NEON;
}
#endif
#if defined(HAS_SUMSQUAREERROR_NEON_DOTPROD)
if (TestCpuFlag(kCpuHasNeonDotProd)) {
SumSquareError = SumSquareError_NEON_DotProd;
}
#endif
#if defined(HAS_SUMSQUAREERROR_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
// Note only used for multiples of 16 so count is not checked.
SumSquareError = SumSquareError_SSE2;
}
#endif
#if defined(HAS_SUMSQUAREERROR_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
// Note only used for multiples of 32 so count is not checked.
SumSquareError = SumSquareError_AVX2;
}
#endif
#ifdef _OPENMP
#pragma omp parallel for reduction(+ : sse)
#endif
for (i = 0; i < (count - (kBlockSize - 1)); i += kBlockSize) {
sse += SumSquareError(src_a + i, src_b + i, kBlockSize);
}
src_a += count & ~(kBlockSize - 1);
src_b += count & ~(kBlockSize - 1);
if (remainder) {
sse += SumSquareError(src_a, src_b, remainder);
src_a += remainder;
src_b += remainder;
}
remainder = count & 31;
if (remainder) {
sse += SumSquareError_C(src_a, src_b, remainder);
}
return sse;
}
LIBYUV_API
uint64_t ComputeSumSquareErrorPlane(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height) {
uint64_t sse = 0;
int h;
// Coalesce rows.
if (stride_a == width && stride_b == width) {
width *= height;
height = 1;
stride_a = stride_b = 0;
}
for (h = 0; h < height; ++h) {
sse += ComputeSumSquareError(src_a, src_b, width);
src_a += stride_a;
src_b += stride_b;
}
return sse;
}
LIBYUV_API
double SumSquareErrorToPsnr(uint64_t sse, uint64_t count) {
double psnr;
if (sse > 0) {
double mse = (double)count / (double)sse;
psnr = 10.0 * log10(255.0 * 255.0 * mse);
} else {
psnr = kMaxPsnr; // Limit to prevent divide by 0
}
if (psnr > kMaxPsnr) {
psnr = kMaxPsnr;
}
return psnr;
}
LIBYUV_API
double CalcFramePsnr(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height) {
const uint64_t samples = (uint64_t)width * (uint64_t)height;
const uint64_t sse = ComputeSumSquareErrorPlane(src_a, stride_a, src_b,
stride_b, width, height);
return SumSquareErrorToPsnr(sse, samples);
}
LIBYUV_API
double I420Psnr(const uint8_t* src_y_a,
int stride_y_a,
const uint8_t* src_u_a,
int stride_u_a,
const uint8_t* src_v_a,
int stride_v_a,
const uint8_t* src_y_b,
int stride_y_b,
const uint8_t* src_u_b,
int stride_u_b,
const uint8_t* src_v_b,
int stride_v_b,
int width,
int height) {
const uint64_t sse_y = ComputeSumSquareErrorPlane(
src_y_a, stride_y_a, src_y_b, stride_y_b, width, height);
const int width_uv = (width + 1) >> 1;
const int height_uv = (height + 1) >> 1;
const uint64_t sse_u = ComputeSumSquareErrorPlane(
src_u_a, stride_u_a, src_u_b, stride_u_b, width_uv, height_uv);
const uint64_t sse_v = ComputeSumSquareErrorPlane(
src_v_a, stride_v_a, src_v_b, stride_v_b, width_uv, height_uv);
const uint64_t samples = (uint64_t)width * (uint64_t)height +
2 * ((uint64_t)width_uv * (uint64_t)height_uv);
const uint64_t sse = sse_y + sse_u + sse_v;
return SumSquareErrorToPsnr(sse, samples);
}
static const int64_t cc1 = 26634; // (64^2*(.01*255)^2
static const int64_t cc2 = 239708; // (64^2*(.03*255)^2
static double Ssim8x8_C(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b) {
int64_t sum_a = 0;
int64_t sum_b = 0;
int64_t sum_sq_a = 0;
int64_t sum_sq_b = 0;
int64_t sum_axb = 0;
int i;
for (i = 0; i < 8; ++i) {
int j;
for (j = 0; j < 8; ++j) {
sum_a += src_a[j];
sum_b += src_b[j];
sum_sq_a += src_a[j] * src_a[j];
sum_sq_b += src_b[j] * src_b[j];
sum_axb += src_a[j] * src_b[j];
}
src_a += stride_a;
src_b += stride_b;
}
{
const int64_t count = 64;
// scale the constants by number of pixels
const int64_t c1 = (cc1 * count * count) >> 12;
const int64_t c2 = (cc2 * count * count) >> 12;
const int64_t sum_a_x_sum_b = sum_a * sum_b;
const int64_t ssim_n = (2 * sum_a_x_sum_b + c1) *
(2 * count * sum_axb - 2 * sum_a_x_sum_b + c2);
const int64_t sum_a_sq = sum_a * sum_a;
const int64_t sum_b_sq = sum_b * sum_b;
const int64_t ssim_d =
(sum_a_sq + sum_b_sq + c1) *
(count * sum_sq_a - sum_a_sq + count * sum_sq_b - sum_b_sq + c2);
if (ssim_d == 0) {
return DBL_MAX;
}
return (double)ssim_n / (double)ssim_d;
}
}
// We are using a 8x8 moving window with starting location of each 8x8 window
// on the 4x4 pixel grid. Such arrangement allows the windows to overlap
// block boundaries to penalize blocking artifacts.
LIBYUV_API
double CalcFrameSsim(const uint8_t* src_a,
int stride_a,
const uint8_t* src_b,
int stride_b,
int width,
int height) {
int samples = 0;
double ssim_total = 0;
double (*Ssim8x8)(const uint8_t* src_a, int stride_a, const uint8_t* src_b,
int stride_b) = Ssim8x8_C;
// sample point start with each 4x4 location
int i;
for (i = 0; i < height - 8; i += 4) {
int j;
for (j = 0; j < width - 8; j += 4) {
ssim_total += Ssim8x8(src_a + j, stride_a, src_b + j, stride_b);
samples++;
}
src_a += stride_a * 4;
src_b += stride_b * 4;
}
ssim_total /= samples;
return ssim_total;
}
LIBYUV_API
double I420Ssim(const uint8_t* src_y_a,
int stride_y_a,
const uint8_t* src_u_a,
int stride_u_a,
const uint8_t* src_v_a,
int stride_v_a,
const uint8_t* src_y_b,
int stride_y_b,
const uint8_t* src_u_b,
int stride_u_b,
const uint8_t* src_v_b,
int stride_v_b,
int width,
int height) {
const double ssim_y =
CalcFrameSsim(src_y_a, stride_y_a, src_y_b, stride_y_b, width, height);
const int width_uv = (width + 1) >> 1;
const int height_uv = (height + 1) >> 1;
const double ssim_u = CalcFrameSsim(src_u_a, stride_u_a, src_u_b, stride_u_b,
width_uv, height_uv);
const double ssim_v = CalcFrameSsim(src_v_a, stride_v_a, src_v_b, stride_v_b,
width_uv, height_uv);
return ssim_y * 0.8 + 0.1 * (ssim_u + ssim_v);
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

74
3rdparty/libyuv/source/compare_common.cc vendored Normal file
View File

@@ -0,0 +1,74 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Hakmem method for hamming distance.
uint32_t HammingDistance_C(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
int i;
for (i = 0; i < count - 3; i += 4) {
uint32_t x = *((const uint32_t*)src_a) ^ *((const uint32_t*)src_b);
uint32_t u = x - ((x >> 1) & 0x55555555);
u = ((u >> 2) & 0x33333333) + (u & 0x33333333);
diff += ((((u + (u >> 4)) & 0x0f0f0f0f) * 0x01010101) >> 24);
src_a += 4;
src_b += 4;
}
for (; i < count; ++i) {
uint32_t x = *src_a ^ *src_b;
uint32_t u = x - ((x >> 1) & 0x55);
u = ((u >> 2) & 0x33) + (u & 0x33);
diff += (u + (u >> 4)) & 0x0f;
src_a += 1;
src_b += 1;
}
return diff;
}
uint32_t SumSquareError_C(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse = 0u;
int i;
for (i = 0; i < count; ++i) {
int diff = src_a[i] - src_b[i];
sse += (uint32_t)(diff * diff);
}
return sse;
}
// hash seed of 5381 recommended.
// Internal C version of HashDjb2 with int sized count for efficiency.
uint32_t HashDjb2_C(const uint8_t* src, int count, uint32_t seed) {
uint32_t hash = seed;
int i;
for (i = 0; i < count; ++i) {
hash += (hash << 5) + src[i];
}
return hash;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

362
3rdparty/libyuv/source/compare_gcc.cc vendored Normal file
View File

@@ -0,0 +1,362 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for GCC x86 and x64.
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || defined(__i386__)) && \
!defined(LIBYUV_ENABLE_ROWWIN)
// "memory" clobber prevents the reads from being removed
#if defined(__x86_64__)
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint64_t diff;
asm volatile(
"xor %3,%3 \n"
"xor %%r8,%%r8 \n"
"xor %%r9,%%r9 \n"
"xor %%r10,%%r10 \n"
// Process 32 bytes per loop.
LABELALIGN
"1: \n"
"mov (%0),%%rcx \n"
"mov 0x8(%0),%%rdx \n"
"xor (%1),%%rcx \n"
"xor 0x8(%1),%%rdx \n"
"popcnt %%rcx,%%rcx \n"
"popcnt %%rdx,%%rdx \n"
"mov 0x10(%0),%%rsi \n"
"mov 0x18(%0),%%rdi \n"
"xor 0x10(%1),%%rsi \n"
"xor 0x18(%1),%%rdi \n"
"popcnt %%rsi,%%rsi \n"
"popcnt %%rdi,%%rdi \n"
"add $0x20,%0 \n"
"add $0x20,%1 \n"
"add %%rcx,%3 \n"
"add %%rdx,%%r8 \n"
"add %%rsi,%%r9 \n"
"add %%rdi,%%r10 \n"
"sub $0x20,%2 \n"
"jg 1b \n"
"add %%r8, %3 \n"
"add %%r9, %3 \n"
"add %%r10, %3 \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=&r"(diff) // %3
:
: "cc", "memory", "rcx", "rdx", "rsi", "rdi", "r8", "r9", "r10");
return (uint32_t)(diff);
}
#else
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
asm volatile(
// Process 16 bytes per loop.
LABELALIGN
"1: \n"
"mov (%0),%%ecx \n"
"mov 0x4(%0),%%edx \n"
"xor (%1),%%ecx \n"
"xor 0x4(%1),%%edx \n"
"popcnt %%ecx,%%ecx \n"
"add %%ecx,%3 \n"
"popcnt %%edx,%%edx \n"
"add %%edx,%3 \n"
"mov 0x8(%0),%%ecx \n"
"mov 0xc(%0),%%edx \n"
"xor 0x8(%1),%%ecx \n"
"xor 0xc(%1),%%edx \n"
"popcnt %%ecx,%%ecx \n"
"add %%ecx,%3 \n"
"popcnt %%edx,%%edx \n"
"add %%edx,%3 \n"
"add $0x10,%0 \n"
"add $0x10,%1 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"+r"(diff) // %3
:
: "cc", "memory", "ecx", "edx");
return diff;
}
#endif
static const vec8 kNibbleMask = {15, 15, 15, 15, 15, 15, 15, 15,
15, 15, 15, 15, 15, 15, 15, 15};
static const vec8 kBitCount = {0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4};
uint32_t HammingDistance_SSSE3(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff;
asm volatile(
"movdqa %4,%%xmm2 \n"
"movdqa %5,%%xmm3 \n"
"pxor %%xmm0,%%xmm0 \n"
"pxor %%xmm1,%%xmm1 \n"
"sub %0,%1 \n"
LABELALIGN
"1: \n"
"movdqa (%0),%%xmm4 \n"
"movdqa 0x10(%0), %%xmm5 \n"
"pxor (%0,%1), %%xmm4 \n"
"movdqa %%xmm4,%%xmm6 \n"
"pand %%xmm2,%%xmm6 \n"
"psrlw $0x4,%%xmm4 \n"
"movdqa %%xmm3,%%xmm7 \n"
"pshufb %%xmm6,%%xmm7 \n"
"pand %%xmm2,%%xmm4 \n"
"movdqa %%xmm3,%%xmm6 \n"
"pshufb %%xmm4,%%xmm6 \n"
"paddb %%xmm7,%%xmm6 \n"
"pxor 0x10(%0,%1),%%xmm5 \n"
"add $0x20,%0 \n"
"movdqa %%xmm5,%%xmm4 \n"
"pand %%xmm2,%%xmm5 \n"
"psrlw $0x4,%%xmm4 \n"
"movdqa %%xmm3,%%xmm7 \n"
"pshufb %%xmm5,%%xmm7 \n"
"pand %%xmm2,%%xmm4 \n"
"movdqa %%xmm3,%%xmm5 \n"
"pshufb %%xmm4,%%xmm5 \n"
"paddb %%xmm7,%%xmm5 \n"
"paddb %%xmm5,%%xmm6 \n"
"psadbw %%xmm1,%%xmm6 \n"
"paddd %%xmm6,%%xmm0 \n"
"sub $0x20,%2 \n"
"jg 1b \n"
"pshufd $0xaa,%%xmm0,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"movd %%xmm0, %3 \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=r"(diff) // %3
: "m"(kNibbleMask), // %4
"m"(kBitCount) // %5
: "cc", "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
return diff;
}
#ifdef HAS_HAMMINGDISTANCE_AVX2
uint32_t HammingDistance_AVX2(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff;
asm volatile(
"vbroadcastf128 %4,%%ymm2 \n"
"vbroadcastf128 %5,%%ymm3 \n"
"vpxor %%ymm0,%%ymm0,%%ymm0 \n"
"vpxor %%ymm1,%%ymm1,%%ymm1 \n"
"sub %0,%1 \n"
LABELALIGN
"1: \n"
"vmovdqa (%0),%%ymm4 \n"
"vmovdqa 0x20(%0), %%ymm5 \n"
"vpxor (%0,%1), %%ymm4, %%ymm4 \n"
"vpand %%ymm2,%%ymm4,%%ymm6 \n"
"vpsrlw $0x4,%%ymm4,%%ymm4 \n"
"vpshufb %%ymm6,%%ymm3,%%ymm6 \n"
"vpand %%ymm2,%%ymm4,%%ymm4 \n"
"vpshufb %%ymm4,%%ymm3,%%ymm4 \n"
"vpaddb %%ymm4,%%ymm6,%%ymm6 \n"
"vpxor 0x20(%0,%1),%%ymm5,%%ymm4 \n"
"add $0x40,%0 \n"
"vpand %%ymm2,%%ymm4,%%ymm5 \n"
"vpsrlw $0x4,%%ymm4,%%ymm4 \n"
"vpshufb %%ymm5,%%ymm3,%%ymm5 \n"
"vpand %%ymm2,%%ymm4,%%ymm4 \n"
"vpshufb %%ymm4,%%ymm3,%%ymm4 \n"
"vpaddb %%ymm5,%%ymm4,%%ymm4 \n"
"vpaddb %%ymm6,%%ymm4,%%ymm4 \n"
"vpsadbw %%ymm1,%%ymm4,%%ymm4 \n"
"vpaddd %%ymm0,%%ymm4,%%ymm0 \n"
"sub $0x40,%2 \n"
"jg 1b \n"
"vpermq $0xb1,%%ymm0,%%ymm1 \n"
"vpaddd %%ymm1,%%ymm0,%%ymm0 \n"
"vpermq $0xaa,%%ymm0,%%ymm1 \n"
"vpaddd %%ymm1,%%ymm0,%%ymm0 \n"
"vmovd %%xmm0,%3 \n"
"vzeroupper \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=r"(diff) // %3
: "m"(kNibbleMask), // %4
"m"(kBitCount) // %5
: "cc", "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6");
return diff;
}
#endif // HAS_HAMMINGDISTANCE_AVX2
uint32_t SumSquareError_SSE2(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse;
asm volatile(
"pxor %%xmm0,%%xmm0 \n"
"pxor %%xmm5,%%xmm5 \n"
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm1 \n"
"lea 0x10(%0),%0 \n"
"movdqu (%1),%%xmm2 \n"
"lea 0x10(%1),%1 \n"
"movdqa %%xmm1,%%xmm3 \n"
"psubusb %%xmm2,%%xmm1 \n"
"psubusb %%xmm3,%%xmm2 \n"
"por %%xmm2,%%xmm1 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklbw %%xmm5,%%xmm1 \n"
"punpckhbw %%xmm5,%%xmm2 \n"
"pmaddwd %%xmm1,%%xmm1 \n"
"pmaddwd %%xmm2,%%xmm2 \n"
"paddd %%xmm1,%%xmm0 \n"
"paddd %%xmm2,%%xmm0 \n"
"sub $0x10,%2 \n"
"jg 1b \n"
"pshufd $0xee,%%xmm0,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"pshufd $0x1,%%xmm0,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"movd %%xmm0,%3 \n"
: "+r"(src_a), // %0
"+r"(src_b), // %1
"+r"(count), // %2
"=r"(sse) // %3
:
: "cc", "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5");
return sse;
}
static const uvec32 kHash16x33 = {0x92d9e201, 0, 0, 0}; // 33 ^ 16
static const uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
};
static const uvec32 kHashMul1 = {
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
};
static const uvec32 kHashMul2 = {
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
};
static const uvec32 kHashMul3 = {
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
};
uint32_t HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed) {
uint32_t hash;
asm volatile(
"movd %2,%%xmm0 \n"
"pxor %%xmm7,%%xmm7 \n"
"movdqa %4,%%xmm6 \n"
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm1 \n"
"lea 0x10(%0),%0 \n"
"pmulld %%xmm6,%%xmm0 \n"
"movdqa %5,%%xmm5 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklbw %%xmm7,%%xmm2 \n"
"movdqa %%xmm2,%%xmm3 \n"
"punpcklwd %%xmm7,%%xmm3 \n"
"pmulld %%xmm5,%%xmm3 \n"
"movdqa %6,%%xmm5 \n"
"movdqa %%xmm2,%%xmm4 \n"
"punpckhwd %%xmm7,%%xmm4 \n"
"pmulld %%xmm5,%%xmm4 \n"
"movdqa %7,%%xmm5 \n"
"punpckhbw %%xmm7,%%xmm1 \n"
"movdqa %%xmm1,%%xmm2 \n"
"punpcklwd %%xmm7,%%xmm2 \n"
"pmulld %%xmm5,%%xmm2 \n"
"movdqa %8,%%xmm5 \n"
"punpckhwd %%xmm7,%%xmm1 \n"
"pmulld %%xmm5,%%xmm1 \n"
"paddd %%xmm4,%%xmm3 \n"
"paddd %%xmm2,%%xmm1 \n"
"paddd %%xmm3,%%xmm1 \n"
"pshufd $0xe,%%xmm1,%%xmm2 \n"
"paddd %%xmm2,%%xmm1 \n"
"pshufd $0x1,%%xmm1,%%xmm2 \n"
"paddd %%xmm2,%%xmm1 \n"
"paddd %%xmm1,%%xmm0 \n"
"sub $0x10,%1 \n"
"jg 1b \n"
"movd %%xmm0,%3 \n"
: "+r"(src), // %0
"+r"(count), // %1
"+rm"(seed), // %2
"=r"(hash) // %3
: "m"(kHash16x33), // %4
"m"(kHashMul0), // %5
"m"(kHashMul1), // %6
"m"(kHashMul2), // %7
"m"(kHashMul3) // %8
: "cc", "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
return hash;
}
#endif // defined(__x86_64__) || (defined(__i386__) && !defined(__pic__)))
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

96
3rdparty/libyuv/source/compare_neon.cc vendored Normal file
View File

@@ -0,0 +1,96 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__) && \
!defined(__aarch64__)
// 256 bits at a time
// uses short accumulator which restricts count to 131 KB
uint32_t HammingDistance_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff;
asm volatile(
"vmov.u16 q4, #0 \n" // accumulator
"1: \n"
"vld1.8 {q0, q1}, [%0]! \n"
"vld1.8 {q2, q3}, [%1]! \n"
"veor.32 q0, q0, q2 \n"
"veor.32 q1, q1, q3 \n"
"vcnt.i8 q0, q0 \n"
"vcnt.i8 q1, q1 \n"
"subs %2, %2, #32 \n"
"vadd.u8 q0, q0, q1 \n" // 16 byte counts
"vpadal.u8 q4, q0 \n" // 8 shorts
"bgt 1b \n"
"vpaddl.u16 q0, q4 \n" // 4 ints
"vpadd.u32 d0, d0, d1 \n"
"vpadd.u32 d0, d0, d0 \n"
"vmov.32 %3, d0[0] \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(diff)
:
: "cc", "q0", "q1", "q2", "q3", "q4");
return diff;
}
uint32_t SumSquareError_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse;
asm volatile(
"vmov.u8 q8, #0 \n"
"vmov.u8 q10, #0 \n"
"vmov.u8 q9, #0 \n"
"vmov.u8 q11, #0 \n"
"1: \n"
"vld1.8 {q0}, [%0]! \n"
"vld1.8 {q1}, [%1]! \n"
"subs %2, %2, #16 \n"
"vsubl.u8 q2, d0, d2 \n"
"vsubl.u8 q3, d1, d3 \n"
"vmlal.s16 q8, d4, d4 \n"
"vmlal.s16 q9, d6, d6 \n"
"vmlal.s16 q10, d5, d5 \n"
"vmlal.s16 q11, d7, d7 \n"
"bgt 1b \n"
"vadd.u32 q8, q8, q9 \n"
"vadd.u32 q10, q10, q11 \n"
"vadd.u32 q11, q8, q10 \n"
"vpaddl.u32 q1, q11 \n"
"vadd.u64 d0, d2, d3 \n"
"vmov.32 %3, d0[0] \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(sse)
:
: "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11");
return sse;
}
#endif // defined(__ARM_NEON__) && !defined(__aarch64__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

223
3rdparty/libyuv/source/compare_neon64.cc vendored Normal file
View File

@@ -0,0 +1,223 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
// 256 bits at a time
// uses short accumulator which restricts count to 131 KB
uint32_t HammingDistance_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff;
asm volatile(
"movi v4.8h, #0 \n"
"1: \n"
"ld1 {v0.16b, v1.16b}, [%0], #32 \n"
"ld1 {v2.16b, v3.16b}, [%1], #32 \n"
"eor v0.16b, v0.16b, v2.16b \n"
"prfm pldl1keep, [%0, 448] \n" // prefetch 7 lines ahead
"eor v1.16b, v1.16b, v3.16b \n"
"cnt v0.16b, v0.16b \n"
"prfm pldl1keep, [%1, 448] \n"
"cnt v1.16b, v1.16b \n"
"subs %w2, %w2, #32 \n"
"add v0.16b, v0.16b, v1.16b \n"
"uadalp v4.8h, v0.16b \n"
"b.gt 1b \n"
"uaddlv s4, v4.8h \n"
"fmov %w3, s4 \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(diff)
:
: "memory", "cc", "v0", "v1", "v2", "v3", "v4");
return diff;
}
uint32_t SumSquareError_NEON(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t sse;
asm volatile(
"movi v16.16b, #0 \n"
"movi v17.16b, #0 \n"
"movi v18.16b, #0 \n"
"movi v19.16b, #0 \n"
"1: \n"
"ld1 {v0.16b}, [%0], #16 \n"
"ld1 {v1.16b}, [%1], #16 \n"
"subs %w2, %w2, #16 \n"
"usubl v2.8h, v0.8b, v1.8b \n"
"usubl2 v3.8h, v0.16b, v1.16b \n"
"prfm pldl1keep, [%0, 448] \n" // prefetch 7 lines ahead
"smlal v16.4s, v2.4h, v2.4h \n"
"smlal v17.4s, v3.4h, v3.4h \n"
"prfm pldl1keep, [%1, 448] \n"
"smlal2 v18.4s, v2.8h, v2.8h \n"
"smlal2 v19.4s, v3.8h, v3.8h \n"
"b.gt 1b \n"
"add v16.4s, v16.4s, v17.4s \n"
"add v18.4s, v18.4s, v19.4s \n"
"add v19.4s, v16.4s, v18.4s \n"
"addv s0, v19.4s \n"
"fmov %w3, s0 \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(sse)
:
: "memory", "cc", "v0", "v1", "v2", "v3", "v16", "v17", "v18", "v19");
return sse;
}
static const uvec32 kDjb2Multiplicands[] = {
{0x0c3525e1, // 33^15
0xa3476dc1, // 33^14
0x3b4039a1, // 33^13
0x4f5f0981}, // 33^12
{0x30f35d61, // 33^11
0x855cb541, // 33^10
0x040a9121, // 33^9
0x747c7101}, // 33^8
{0xec41d4e1, // 33^7
0x4cfa3cc1, // 33^6
0x025528a1, // 33^5
0x00121881}, // 33^4
{0x00008c61, // 33^3
0x00000441, // 33^2
0x00000021, // 33^1
0x00000001}, // 33^0
};
static const uvec32 kDjb2WidenIndices[] = {
{0xffffff00U, 0xffffff01U, 0xffffff02U, 0xffffff03U},
{0xffffff04U, 0xffffff05U, 0xffffff06U, 0xffffff07U},
{0xffffff08U, 0xffffff09U, 0xffffff0aU, 0xffffff0bU},
{0xffffff0cU, 0xffffff0dU, 0xffffff0eU, 0xffffff0fU},
};
uint32_t HashDjb2_NEON(const uint8_t* src, int count, uint32_t seed) {
uint32_t hash = seed;
const uint32_t c16 = 0x92d9e201; // 33^16
uint32_t tmp, tmp2;
asm("ld1 {v16.4s, v17.4s, v18.4s, v19.4s}, [%[kIdx]] \n"
"ld1 {v4.4s, v5.4s, v6.4s, v7.4s}, [%[kMuls]] \n"
// count is always a multiple of 16.
// maintain two accumulators, reduce and then final sum in scalar since
// this has better performance on little cores.
"1: \n"
"ldr q0, [%[src]], #16 \n"
"subs %w[count], %w[count], #16 \n"
"tbl v3.16b, {v0.16b}, v19.16b \n"
"tbl v2.16b, {v0.16b}, v18.16b \n"
"tbl v1.16b, {v0.16b}, v17.16b \n"
"tbl v0.16b, {v0.16b}, v16.16b \n"
"mul v3.4s, v3.4s, v7.4s \n"
"mul v2.4s, v2.4s, v6.4s \n"
"mla v3.4s, v1.4s, v5.4s \n"
"mla v2.4s, v0.4s, v4.4s \n"
"addv s1, v3.4s \n"
"addv s0, v2.4s \n"
"fmov %w[tmp2], s1 \n"
"fmov %w[tmp], s0 \n"
"add %w[tmp], %w[tmp], %w[tmp2] \n"
"madd %w[hash], %w[hash], %w[c16], %w[tmp] \n"
"b.gt 1b \n"
: [hash] "+r"(hash), // %[hash]
[count] "+r"(count), // %[count]
[tmp] "=&r"(tmp), // %[tmp]
[tmp2] "=&r"(tmp2) // %[tmp2]
: [src] "r"(src), // %[src]
[kMuls] "r"(kDjb2Multiplicands), // %[kMuls]
[kIdx] "r"(kDjb2WidenIndices), // %[kIdx]
[c16] "r"(c16) // %[c16]
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
"v17", "v18", "v19");
return hash;
}
uint32_t HammingDistance_NEON_DotProd(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff;
asm volatile(
"movi v4.4s, #0 \n"
"movi v5.4s, #0 \n"
"movi v6.16b, #1 \n"
"1: \n"
"ldp q0, q1, [%0], #32 \n"
"ldp q2, q3, [%1], #32 \n"
"eor v0.16b, v0.16b, v2.16b \n"
"prfm pldl1keep, [%0, 448] \n" // prefetch 7 lines ahead
"eor v1.16b, v1.16b, v3.16b \n"
"cnt v0.16b, v0.16b \n"
"prfm pldl1keep, [%1, 448] \n"
"cnt v1.16b, v1.16b \n"
"subs %w2, %w2, #32 \n"
"udot v4.4s, v0.16b, v6.16b \n"
"udot v5.4s, v1.16b, v6.16b \n"
"b.gt 1b \n"
"add v0.4s, v4.4s, v5.4s \n"
"addv s0, v0.4s \n"
"fmov %w3, s0 \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(diff)
:
: "memory", "cc", "v0", "v1", "v2", "v3", "v4", "v5", "v6");
return diff;
}
uint32_t SumSquareError_NEON_DotProd(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
// count is guaranteed to be a multiple of 32.
uint32_t sse;
asm volatile(
"movi v4.4s, #0 \n"
"movi v5.4s, #0 \n"
"1: \n"
"ldp q0, q2, [%0], #32 \n"
"ldp q1, q3, [%1], #32 \n"
"subs %w2, %w2, #32 \n"
"uabd v0.16b, v0.16b, v1.16b \n"
"uabd v1.16b, v2.16b, v3.16b \n"
"prfm pldl1keep, [%0, 448] \n" // prefetch 7 lines ahead
"udot v4.4s, v0.16b, v0.16b \n"
"udot v5.4s, v1.16b, v1.16b \n"
"prfm pldl1keep, [%1, 448] \n"
"b.gt 1b \n"
"add v0.4s, v4.4s, v5.4s \n"
"addv s0, v0.4s \n"
"fmov %w3, s0 \n"
: "+r"(src_a), "+r"(src_b), "+r"(count), "=r"(sse)
:
: "memory", "cc", "v0", "v1", "v2", "v3", "v4", "v5");
return sse;
}
#endif // !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

241
3rdparty/libyuv/source/compare_win.cc vendored Normal file
View File

@@ -0,0 +1,241 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#include "libyuv/row.h"
#if defined(_MSC_VER)
#include <intrin.h> // For __popcnt
#endif
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for 32 bit Visual C x86
#if !defined(LIBYUV_DISABLE_X86) && defined(_MSC_VER) && defined(_M_IX86) && \
(!defined(__clang__) || defined(LIBYUV_ENABLE_ROWWIN))
uint32_t HammingDistance_SSE42(const uint8_t* src_a,
const uint8_t* src_b,
int count) {
uint32_t diff = 0u;
int i;
for (i = 0; i < count - 3; i += 4) {
uint32_t x = *((uint32_t*)src_a) ^ *((uint32_t*)src_b); // NOLINT
src_a += 4;
src_b += 4;
diff += __popcnt(x);
}
return diff;
}
__declspec(naked) uint32_t
SumSquareError_SSE2(const uint8_t* src_a, const uint8_t* src_b, int count) {
__asm {
mov eax, [esp + 4] // src_a
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
pxor xmm0, xmm0
pxor xmm5, xmm5
wloop:
movdqu xmm1, [eax]
lea eax, [eax + 16]
movdqu xmm2, [edx]
lea edx, [edx + 16]
movdqa xmm3, xmm1 // abs trick
psubusb xmm1, xmm2
psubusb xmm2, xmm3
por xmm1, xmm2
movdqa xmm2, xmm1
punpcklbw xmm1, xmm5
punpckhbw xmm2, xmm5
pmaddwd xmm1, xmm1
pmaddwd xmm2, xmm2
paddd xmm0, xmm1
paddd xmm0, xmm2
sub ecx, 16
jg wloop
pshufd xmm1, xmm0, 0xee
paddd xmm0, xmm1
pshufd xmm1, xmm0, 0x01
paddd xmm0, xmm1
movd eax, xmm0
ret
}
}
#ifdef HAS_SUMSQUAREERROR_AVX2
// C4752: found Intel(R) Advanced Vector Extensions; consider using /arch:AVX.
#pragma warning(disable : 4752)
__declspec(naked) uint32_t
SumSquareError_AVX2(const uint8_t* src_a, const uint8_t* src_b, int count) {
__asm {
mov eax, [esp + 4] // src_a
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
vpxor ymm0, ymm0, ymm0 // sum
vpxor ymm5, ymm5, ymm5 // constant 0 for unpck
sub edx, eax
wloop:
vmovdqu ymm1, [eax]
vmovdqu ymm2, [eax + edx]
lea eax, [eax + 32]
vpsubusb ymm3, ymm1, ymm2 // abs difference trick
vpsubusb ymm2, ymm2, ymm1
vpor ymm1, ymm2, ymm3
vpunpcklbw ymm2, ymm1, ymm5 // u16. mutates order.
vpunpckhbw ymm1, ymm1, ymm5
vpmaddwd ymm2, ymm2, ymm2 // square + hadd to u32.
vpmaddwd ymm1, ymm1, ymm1
vpaddd ymm0, ymm0, ymm1
vpaddd ymm0, ymm0, ymm2
sub ecx, 32
jg wloop
vpshufd ymm1, ymm0, 0xee // 3, 2 + 1, 0 both lanes.
vpaddd ymm0, ymm0, ymm1
vpshufd ymm1, ymm0, 0x01 // 1 + 0 both lanes.
vpaddd ymm0, ymm0, ymm1
vpermq ymm1, ymm0, 0x02 // high + low lane.
vpaddd ymm0, ymm0, ymm1
vmovd eax, xmm0
vzeroupper
ret
}
}
#endif // HAS_SUMSQUAREERROR_AVX2
uvec32 kHash16x33 = {0x92d9e201, 0, 0, 0}; // 33 ^ 16
uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
};
uvec32 kHashMul1 = {
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
};
uvec32 kHashMul2 = {
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
};
uvec32 kHashMul3 = {
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
};
__declspec(naked) uint32_t
HashDjb2_SSE41(const uint8_t* src, int count, uint32_t seed) {
__asm {
mov eax, [esp + 4] // src
mov ecx, [esp + 8] // count
movd xmm0, [esp + 12] // seed
pxor xmm7, xmm7 // constant 0 for unpck
movdqa xmm6, xmmword ptr kHash16x33
wloop:
movdqu xmm1, [eax] // src[0-15]
lea eax, [eax + 16]
pmulld xmm0, xmm6 // hash *= 33 ^ 16
movdqa xmm5, xmmword ptr kHashMul0
movdqa xmm2, xmm1
punpcklbw xmm2, xmm7 // src[0-7]
movdqa xmm3, xmm2
punpcklwd xmm3, xmm7 // src[0-3]
pmulld xmm3, xmm5
movdqa xmm5, xmmword ptr kHashMul1
movdqa xmm4, xmm2
punpckhwd xmm4, xmm7 // src[4-7]
pmulld xmm4, xmm5
movdqa xmm5, xmmword ptr kHashMul2
punpckhbw xmm1, xmm7 // src[8-15]
movdqa xmm2, xmm1
punpcklwd xmm2, xmm7 // src[8-11]
pmulld xmm2, xmm5
movdqa xmm5, xmmword ptr kHashMul3
punpckhwd xmm1, xmm7 // src[12-15]
pmulld xmm1, xmm5
paddd xmm3, xmm4 // add 16 results
paddd xmm1, xmm2
paddd xmm1, xmm3
pshufd xmm2, xmm1, 0x0e // upper 2 dwords
paddd xmm1, xmm2
pshufd xmm2, xmm1, 0x01
paddd xmm1, xmm2
paddd xmm0, xmm1
sub ecx, 16
jg wloop
movd eax, xmm0 // return hash
ret
}
}
// Visual C 2012 required for AVX2.
#ifdef HAS_HASHDJB2_AVX2
__declspec(naked) uint32_t
HashDjb2_AVX2(const uint8_t* src, int count, uint32_t seed) {
__asm {
mov eax, [esp + 4] // src
mov ecx, [esp + 8] // count
vmovd xmm0, [esp + 12] // seed
wloop:
vpmovzxbd xmm3, [eax] // src[0-3]
vpmulld xmm0, xmm0, xmmword ptr kHash16x33 // hash *= 33 ^ 16
vpmovzxbd xmm4, [eax + 4] // src[4-7]
vpmulld xmm3, xmm3, xmmword ptr kHashMul0
vpmovzxbd xmm2, [eax + 8] // src[8-11]
vpmulld xmm4, xmm4, xmmword ptr kHashMul1
vpmovzxbd xmm1, [eax + 12] // src[12-15]
vpmulld xmm2, xmm2, xmmword ptr kHashMul2
lea eax, [eax + 16]
vpmulld xmm1, xmm1, xmmword ptr kHashMul3
vpaddd xmm3, xmm3, xmm4 // add 16 results
vpaddd xmm1, xmm1, xmm2
vpaddd xmm1, xmm1, xmm3
vpshufd xmm2, xmm1, 0x0e // upper 2 dwords
vpaddd xmm1, xmm1,xmm2
vpshufd xmm2, xmm1, 0x01
vpaddd xmm1, xmm1, xmm2
vpaddd xmm0, xmm0, xmm1
sub ecx, 16
jg wloop
vmovd eax, xmm0 // return hash
vzeroupper
ret
}
}
#endif // HAS_HASHDJB2_AVX2
#endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

5005
3rdparty/libyuv/source/convert.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

9179
3rdparty/libyuv/source/convert_argb.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

866
3rdparty/libyuv/source/convert_from.cc vendored Normal file
View File

@@ -0,0 +1,866 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/convert_from.h"
#include "libyuv/basic_types.h"
#include "libyuv/convert.h" // For I420Copy
#include "libyuv/cpu_id.h"
#include "libyuv/planar_functions.h"
#include "libyuv/rotate.h"
#include "libyuv/row.h"
#include "libyuv/scale.h" // For ScalePlane()
#include "libyuv/video_common.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#define SUBSAMPLE(v, a, s) (v < 0) ? (-((-v + a) >> s)) : ((v + a) >> s)
static __inline int Abs(int v) {
return v >= 0 ? v : -v;
}
// I420 To any I4xx YUV format with mirroring.
static int I420ToI4xx(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_u,
int dst_stride_u,
uint8_t* dst_v,
int dst_stride_v,
int src_y_width,
int src_y_height,
int dst_uv_width,
int dst_uv_height) {
const int src_uv_width = SUBSAMPLE(src_y_width, 1, 1);
const int src_uv_height = SUBSAMPLE(src_y_height, 1, 1);
int r;
if ((!src_y && dst_y) || !src_u || !src_v || !dst_u || !dst_v ||
src_y_width <= 0 || src_y_height == 0 || dst_uv_width <= 0 ||
dst_uv_height <= 0) {
return -1;
}
if (dst_y) {
CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, src_y_width,
src_y_height);
}
r = ScalePlane(src_u, src_stride_u, src_uv_width, src_uv_height, dst_u,
dst_stride_u, dst_uv_width, dst_uv_height, kFilterBilinear);
if (r != 0) {
return r;
}
r = ScalePlane(src_v, src_stride_v, src_uv_width, src_uv_height, dst_v,
dst_stride_v, dst_uv_width, dst_uv_height, kFilterBilinear);
return r;
}
// Convert 8 bit YUV to 10 bit.
LIBYUV_API
int I420ToI010(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint16_t* dst_y,
int dst_stride_y,
uint16_t* dst_u,
int dst_stride_u,
uint16_t* dst_v,
int dst_stride_v,
int width,
int height) {
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
if ((!src_y && dst_y) || !src_u || !src_v || !dst_u || !dst_v || width <= 0 ||
height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
halfheight = (height + 1) >> 1;
src_y = src_y + (height - 1) * src_stride_y;
src_u = src_u + (halfheight - 1) * src_stride_u;
src_v = src_v + (halfheight - 1) * src_stride_v;
src_stride_y = -src_stride_y;
src_stride_u = -src_stride_u;
src_stride_v = -src_stride_v;
}
// Convert Y plane.
Convert8To16Plane(src_y, src_stride_y, dst_y, dst_stride_y, 1024, width,
height);
// Convert UV planes.
Convert8To16Plane(src_u, src_stride_u, dst_u, dst_stride_u, 1024, halfwidth,
halfheight);
Convert8To16Plane(src_v, src_stride_v, dst_v, dst_stride_v, 1024, halfwidth,
halfheight);
return 0;
}
// Convert 8 bit YUV to 12 bit.
LIBYUV_API
int I420ToI012(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint16_t* dst_y,
int dst_stride_y,
uint16_t* dst_u,
int dst_stride_u,
uint16_t* dst_v,
int dst_stride_v,
int width,
int height) {
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
if ((!src_y && dst_y) || !src_u || !src_v || !dst_u || !dst_v || width <= 0 ||
height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
halfheight = (height + 1) >> 1;
src_y = src_y + (height - 1) * src_stride_y;
src_u = src_u + (halfheight - 1) * src_stride_u;
src_v = src_v + (halfheight - 1) * src_stride_v;
src_stride_y = -src_stride_y;
src_stride_u = -src_stride_u;
src_stride_v = -src_stride_v;
}
// Convert Y plane.
Convert8To16Plane(src_y, src_stride_y, dst_y, dst_stride_y, 4096, width,
height);
// Convert UV planes.
Convert8To16Plane(src_u, src_stride_u, dst_u, dst_stride_u, 4096, halfwidth,
halfheight);
Convert8To16Plane(src_v, src_stride_v, dst_v, dst_stride_v, 4096, halfwidth,
halfheight);
return 0;
}
// 420 chroma is 1/2 width, 1/2 height
// 422 chroma is 1/2 width, 1x height
LIBYUV_API
int I420ToI422(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_u,
int dst_stride_u,
uint8_t* dst_v,
int dst_stride_v,
int width,
int height) {
const int dst_uv_width = (Abs(width) + 1) >> 1;
const int dst_uv_height = Abs(height);
return I420ToI4xx(src_y, src_stride_y, src_u, src_stride_u, src_v,
src_stride_v, dst_y, dst_stride_y, dst_u, dst_stride_u,
dst_v, dst_stride_v, width, height, dst_uv_width,
dst_uv_height);
}
// 420 chroma is 1/2 width, 1/2 height
// 444 chroma is 1x width, 1x height
LIBYUV_API
int I420ToI444(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_u,
int dst_stride_u,
uint8_t* dst_v,
int dst_stride_v,
int width,
int height) {
const int dst_uv_width = Abs(width);
const int dst_uv_height = Abs(height);
return I420ToI4xx(src_y, src_stride_y, src_u, src_stride_u, src_v,
src_stride_v, dst_y, dst_stride_y, dst_u, dst_stride_u,
dst_v, dst_stride_v, width, height, dst_uv_width,
dst_uv_height);
}
// 420 chroma to 444 chroma, 10/12 bit version
LIBYUV_API
int I010ToI410(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint16_t* dst_y,
int dst_stride_y,
uint16_t* dst_u,
int dst_stride_u,
uint16_t* dst_v,
int dst_stride_v,
int width,
int height) {
int r;
if ((!src_y && dst_y) || !src_u || !src_v || !dst_u || !dst_v || width <= 0 ||
height == 0) {
return -1;
}
if (dst_y) {
CopyPlane_16(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
}
r = ScalePlane_12(src_u, src_stride_u, SUBSAMPLE(width, 1, 1),
SUBSAMPLE(height, 1, 1), dst_u, dst_stride_u, width,
Abs(height), kFilterBilinear);
if (r != 0) {
return r;
}
r = ScalePlane_12(src_v, src_stride_v, SUBSAMPLE(width, 1, 1),
SUBSAMPLE(height, 1, 1), dst_v, dst_stride_v, width,
Abs(height), kFilterBilinear);
return r;
}
// 422 chroma to 444 chroma, 10/12 bit version
LIBYUV_API
int I210ToI410(const uint16_t* src_y,
int src_stride_y,
const uint16_t* src_u,
int src_stride_u,
const uint16_t* src_v,
int src_stride_v,
uint16_t* dst_y,
int dst_stride_y,
uint16_t* dst_u,
int dst_stride_u,
uint16_t* dst_v,
int dst_stride_v,
int width,
int height) {
int r;
if ((!src_y && dst_y) || !src_u || !src_v || !dst_u || !dst_v || width <= 0 ||
height == 0) {
return -1;
}
if (dst_y) {
CopyPlane_16(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
}
r = ScalePlane_12(src_u, src_stride_u, SUBSAMPLE(width, 1, 1), height, dst_u,
dst_stride_u, width, Abs(height), kFilterBilinear);
if (r != 0) {
return r;
}
r = ScalePlane_12(src_v, src_stride_v, SUBSAMPLE(width, 1, 1), height, dst_v,
dst_stride_v, width, Abs(height), kFilterBilinear);
return r;
}
// 422 chroma is 1/2 width, 1x height
// 444 chroma is 1x width, 1x height
LIBYUV_API
int I422ToI444(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_u,
int dst_stride_u,
uint8_t* dst_v,
int dst_stride_v,
int width,
int height) {
int r;
if ((!src_y && dst_y) || !src_u || !src_v || !dst_u || !dst_v || width <= 0 ||
height == 0) {
return -1;
}
if (dst_y) {
CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
}
r = ScalePlane(src_u, src_stride_u, SUBSAMPLE(width, 1, 1), height, dst_u,
dst_stride_u, width, Abs(height), kFilterBilinear);
if (r != 0) {
return r;
}
r = ScalePlane(src_v, src_stride_v, SUBSAMPLE(width, 1, 1), height, dst_v,
dst_stride_v, width, Abs(height), kFilterBilinear);
return r;
}
// Copy to I400. Source can be I420,422,444,400,NV12,NV21
LIBYUV_API
int I400Copy(const uint8_t* src_y,
int src_stride_y,
uint8_t* dst_y,
int dst_stride_y,
int width,
int height) {
if (!src_y || !dst_y || width <= 0 || height == 0) {
return -1;
}
CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
return 0;
}
LIBYUV_API
int I422ToYUY2(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_yuy2,
int dst_stride_yuy2,
int width,
int height) {
int y;
void (*I422ToYUY2Row)(const uint8_t* src_y, const uint8_t* src_u,
const uint8_t* src_v, uint8_t* dst_yuy2, int width) =
I422ToYUY2Row_C;
if (!src_y || !src_u || !src_v || !dst_yuy2 || width <= 0 || height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
dst_yuy2 = dst_yuy2 + (height - 1) * dst_stride_yuy2;
dst_stride_yuy2 = -dst_stride_yuy2;
}
// Coalesce rows.
if (src_stride_y == width && src_stride_u * 2 == width &&
src_stride_v * 2 == width && dst_stride_yuy2 == width * 2) {
width *= height;
height = 1;
src_stride_y = src_stride_u = src_stride_v = dst_stride_yuy2 = 0;
}
#if defined(HAS_I422TOYUY2ROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
I422ToYUY2Row = I422ToYUY2Row_Any_SSE2;
if (IS_ALIGNED(width, 16)) {
I422ToYUY2Row = I422ToYUY2Row_SSE2;
}
}
#endif
#if defined(HAS_I422TOYUY2ROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
I422ToYUY2Row = I422ToYUY2Row_Any_AVX2;
if (IS_ALIGNED(width, 32)) {
I422ToYUY2Row = I422ToYUY2Row_AVX2;
}
}
#endif
#if defined(HAS_I422TOYUY2ROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
I422ToYUY2Row = I422ToYUY2Row_Any_NEON;
if (IS_ALIGNED(width, 16)) {
I422ToYUY2Row = I422ToYUY2Row_NEON;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToYUY2Row(src_y, src_u, src_v, dst_yuy2, width);
src_y += src_stride_y;
src_u += src_stride_u;
src_v += src_stride_v;
dst_yuy2 += dst_stride_yuy2;
}
return 0;
}
LIBYUV_API
int I420ToYUY2(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_yuy2,
int dst_stride_yuy2,
int width,
int height) {
int y;
void (*I422ToYUY2Row)(const uint8_t* src_y, const uint8_t* src_u,
const uint8_t* src_v, uint8_t* dst_yuy2, int width) =
I422ToYUY2Row_C;
if (!src_y || !src_u || !src_v || !dst_yuy2 || width <= 0 || height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
dst_yuy2 = dst_yuy2 + (height - 1) * dst_stride_yuy2;
dst_stride_yuy2 = -dst_stride_yuy2;
}
#if defined(HAS_I422TOYUY2ROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
I422ToYUY2Row = I422ToYUY2Row_Any_SSE2;
if (IS_ALIGNED(width, 16)) {
I422ToYUY2Row = I422ToYUY2Row_SSE2;
}
}
#endif
#if defined(HAS_I422TOYUY2ROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
I422ToYUY2Row = I422ToYUY2Row_Any_AVX2;
if (IS_ALIGNED(width, 32)) {
I422ToYUY2Row = I422ToYUY2Row_AVX2;
}
}
#endif
#if defined(HAS_I422TOYUY2ROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
I422ToYUY2Row = I422ToYUY2Row_Any_NEON;
if (IS_ALIGNED(width, 16)) {
I422ToYUY2Row = I422ToYUY2Row_NEON;
}
}
#endif
#if defined(HAS_I422TOYUY2ROW_LSX)
if (TestCpuFlag(kCpuHasLSX)) {
I422ToYUY2Row = I422ToYUY2Row_Any_LSX;
if (IS_ALIGNED(width, 16)) {
I422ToYUY2Row = I422ToYUY2Row_LSX;
}
}
#endif
#if defined(HAS_I422TOYUY2ROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToYUY2Row = I422ToYUY2Row_Any_LASX;
if (IS_ALIGNED(width, 32)) {
I422ToYUY2Row = I422ToYUY2Row_LASX;
}
}
#endif
for (y = 0; y < height - 1; y += 2) {
I422ToYUY2Row(src_y, src_u, src_v, dst_yuy2, width);
I422ToYUY2Row(src_y + src_stride_y, src_u, src_v,
dst_yuy2 + dst_stride_yuy2, width);
src_y += src_stride_y * 2;
src_u += src_stride_u;
src_v += src_stride_v;
dst_yuy2 += dst_stride_yuy2 * 2;
}
if (height & 1) {
I422ToYUY2Row(src_y, src_u, src_v, dst_yuy2, width);
}
return 0;
}
LIBYUV_API
int I422ToUYVY(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_uyvy,
int dst_stride_uyvy,
int width,
int height) {
int y;
void (*I422ToUYVYRow)(const uint8_t* src_y, const uint8_t* src_u,
const uint8_t* src_v, uint8_t* dst_uyvy, int width) =
I422ToUYVYRow_C;
if (!src_y || !src_u || !src_v || !dst_uyvy || width <= 0 || height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
dst_uyvy = dst_uyvy + (height - 1) * dst_stride_uyvy;
dst_stride_uyvy = -dst_stride_uyvy;
}
// Coalesce rows.
if (src_stride_y == width && src_stride_u * 2 == width &&
src_stride_v * 2 == width && dst_stride_uyvy == width * 2) {
width *= height;
height = 1;
src_stride_y = src_stride_u = src_stride_v = dst_stride_uyvy = 0;
}
#if defined(HAS_I422TOUYVYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
I422ToUYVYRow = I422ToUYVYRow_Any_SSE2;
if (IS_ALIGNED(width, 16)) {
I422ToUYVYRow = I422ToUYVYRow_SSE2;
}
}
#endif
#if defined(HAS_I422TOUYVYROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
I422ToUYVYRow = I422ToUYVYRow_Any_AVX2;
if (IS_ALIGNED(width, 32)) {
I422ToUYVYRow = I422ToUYVYRow_AVX2;
}
}
#endif
#if defined(HAS_I422TOUYVYROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
I422ToUYVYRow = I422ToUYVYRow_Any_NEON;
if (IS_ALIGNED(width, 16)) {
I422ToUYVYRow = I422ToUYVYRow_NEON;
}
}
#endif
#if defined(HAS_I422TOUYVYROW_LSX)
if (TestCpuFlag(kCpuHasLSX)) {
I422ToUYVYRow = I422ToUYVYRow_Any_LSX;
if (IS_ALIGNED(width, 16)) {
I422ToUYVYRow = I422ToUYVYRow_LSX;
}
}
#endif
#if defined(HAS_I422TOUYVYROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToUYVYRow = I422ToUYVYRow_Any_LASX;
if (IS_ALIGNED(width, 32)) {
I422ToUYVYRow = I422ToUYVYRow_LASX;
}
}
#endif
for (y = 0; y < height; ++y) {
I422ToUYVYRow(src_y, src_u, src_v, dst_uyvy, width);
src_y += src_stride_y;
src_u += src_stride_u;
src_v += src_stride_v;
dst_uyvy += dst_stride_uyvy;
}
return 0;
}
LIBYUV_API
int I420ToUYVY(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_uyvy,
int dst_stride_uyvy,
int width,
int height) {
int y;
void (*I422ToUYVYRow)(const uint8_t* src_y, const uint8_t* src_u,
const uint8_t* src_v, uint8_t* dst_uyvy, int width) =
I422ToUYVYRow_C;
if (!src_y || !src_u || !src_v || !dst_uyvy || width <= 0 || height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
dst_uyvy = dst_uyvy + (height - 1) * dst_stride_uyvy;
dst_stride_uyvy = -dst_stride_uyvy;
}
#if defined(HAS_I422TOUYVYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
I422ToUYVYRow = I422ToUYVYRow_Any_SSE2;
if (IS_ALIGNED(width, 16)) {
I422ToUYVYRow = I422ToUYVYRow_SSE2;
}
}
#endif
#if defined(HAS_I422TOUYVYROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
I422ToUYVYRow = I422ToUYVYRow_Any_AVX2;
if (IS_ALIGNED(width, 32)) {
I422ToUYVYRow = I422ToUYVYRow_AVX2;
}
}
#endif
#if defined(HAS_I422TOUYVYROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
I422ToUYVYRow = I422ToUYVYRow_Any_NEON;
if (IS_ALIGNED(width, 16)) {
I422ToUYVYRow = I422ToUYVYRow_NEON;
}
}
#endif
#if defined(HAS_I422TOUYVYROW_LSX)
if (TestCpuFlag(kCpuHasLSX)) {
I422ToUYVYRow = I422ToUYVYRow_Any_LSX;
if (IS_ALIGNED(width, 16)) {
I422ToUYVYRow = I422ToUYVYRow_LSX;
}
}
#endif
#if defined(HAS_I422TOUYVYROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
I422ToUYVYRow = I422ToUYVYRow_Any_LASX;
if (IS_ALIGNED(width, 32)) {
I422ToUYVYRow = I422ToUYVYRow_LASX;
}
}
#endif
for (y = 0; y < height - 1; y += 2) {
I422ToUYVYRow(src_y, src_u, src_v, dst_uyvy, width);
I422ToUYVYRow(src_y + src_stride_y, src_u, src_v,
dst_uyvy + dst_stride_uyvy, width);
src_y += src_stride_y * 2;
src_u += src_stride_u;
src_v += src_stride_v;
dst_uyvy += dst_stride_uyvy * 2;
}
if (height & 1) {
I422ToUYVYRow(src_y, src_u, src_v, dst_uyvy, width);
}
return 0;
}
LIBYUV_API
int I420ToNV12(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int width,
int height) {
int halfwidth = (width + 1) / 2;
int halfheight = (height + 1) / 2;
if ((!src_y && dst_y) || !src_u || !src_v || !dst_uv || width <= 0 ||
height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
halfheight = (height + 1) >> 1;
src_y = src_y + (height - 1) * src_stride_y;
src_u = src_u + (halfheight - 1) * src_stride_u;
src_v = src_v + (halfheight - 1) * src_stride_v;
src_stride_y = -src_stride_y;
src_stride_u = -src_stride_u;
src_stride_v = -src_stride_v;
}
if (dst_y) {
CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
}
MergeUVPlane(src_u, src_stride_u, src_v, src_stride_v, dst_uv, dst_stride_uv,
halfwidth, halfheight);
return 0;
}
LIBYUV_API
int I420ToNV21(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_u,
int src_stride_u,
const uint8_t* src_v,
int src_stride_v,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int width,
int height) {
return I420ToNV12(src_y, src_stride_y, src_v, src_stride_v, src_u,
src_stride_u, dst_y, dst_stride_y, dst_vu, dst_stride_vu,
width, height);
}
// Convert I420 to specified format
LIBYUV_API
int ConvertFromI420(const uint8_t* y,
int y_stride,
const uint8_t* u,
int u_stride,
const uint8_t* v,
int v_stride,
uint8_t* dst_sample,
int dst_sample_stride,
int width,
int height,
uint32_t fourcc) {
uint32_t format = CanonicalFourCC(fourcc);
int r = 0;
if (!y || !u || !v || !dst_sample || width <= 0 || height == 0) {
return -1;
}
switch (format) {
// Single plane formats
case FOURCC_YUY2:
r = I420ToYUY2(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 2, width,
height);
break;
case FOURCC_UYVY:
r = I420ToUYVY(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 2, width,
height);
break;
case FOURCC_RGBP:
r = I420ToRGB565(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 2, width,
height);
break;
case FOURCC_RGBO:
r = I420ToARGB1555(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 2,
width, height);
break;
case FOURCC_R444:
r = I420ToARGB4444(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 2,
width, height);
break;
case FOURCC_24BG:
r = I420ToRGB24(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 3, width,
height);
break;
case FOURCC_RAW:
r = I420ToRAW(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 3, width,
height);
break;
case FOURCC_ARGB:
r = I420ToARGB(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 4, width,
height);
break;
case FOURCC_BGRA:
r = I420ToBGRA(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 4, width,
height);
break;
case FOURCC_ABGR:
r = I420ToABGR(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 4, width,
height);
break;
case FOURCC_RGBA:
r = I420ToRGBA(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 4, width,
height);
break;
case FOURCC_AR30:
r = I420ToAR30(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width * 4, width,
height);
break;
case FOURCC_I400:
r = I400Copy(y, y_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width, width,
height);
break;
case FOURCC_NV12: {
int dst_y_stride = dst_sample_stride ? dst_sample_stride : width;
uint8_t* dst_uv = dst_sample + dst_y_stride * height;
r = I420ToNV12(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width, dst_uv,
dst_sample_stride ? dst_sample_stride : width, width,
height);
break;
}
case FOURCC_NV21: {
int dst_y_stride = dst_sample_stride ? dst_sample_stride : width;
uint8_t* dst_vu = dst_sample + dst_y_stride * height;
r = I420ToNV21(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride ? dst_sample_stride : width, dst_vu,
dst_sample_stride ? dst_sample_stride : width, width,
height);
break;
}
// Triplanar formats
case FOURCC_I420:
case FOURCC_YV12: {
dst_sample_stride = dst_sample_stride ? dst_sample_stride : width;
int halfstride = (dst_sample_stride + 1) / 2;
int halfheight = (height + 1) / 2;
uint8_t* dst_u;
uint8_t* dst_v;
if (format == FOURCC_YV12) {
dst_v = dst_sample + dst_sample_stride * height;
dst_u = dst_v + halfstride * halfheight;
} else {
dst_u = dst_sample + dst_sample_stride * height;
dst_v = dst_u + halfstride * halfheight;
}
r = I420Copy(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride, dst_u, halfstride, dst_v, halfstride,
width, height);
break;
}
case FOURCC_I422:
case FOURCC_YV16: {
dst_sample_stride = dst_sample_stride ? dst_sample_stride : width;
int halfstride = (dst_sample_stride + 1) / 2;
uint8_t* dst_u;
uint8_t* dst_v;
if (format == FOURCC_YV16) {
dst_v = dst_sample + dst_sample_stride * height;
dst_u = dst_v + halfstride * height;
} else {
dst_u = dst_sample + dst_sample_stride * height;
dst_v = dst_u + halfstride * height;
}
r = I420ToI422(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride, dst_u, halfstride, dst_v, halfstride,
width, height);
break;
}
case FOURCC_I444:
case FOURCC_YV24: {
dst_sample_stride = dst_sample_stride ? dst_sample_stride : width;
uint8_t* dst_u;
uint8_t* dst_v;
if (format == FOURCC_YV24) {
dst_v = dst_sample + dst_sample_stride * height;
dst_u = dst_v + dst_sample_stride * height;
} else {
dst_u = dst_sample + dst_sample_stride * height;
dst_v = dst_u + dst_sample_stride * height;
}
r = I420ToI444(y, y_stride, u, u_stride, v, v_stride, dst_sample,
dst_sample_stride, dst_u, dst_sample_stride, dst_v,
dst_sample_stride, width, height);
break;
}
// Formats not supported - MJPG, biplanar, some rgb formats.
default:
return -1; // unknown fourcc - return failure code.
}
return r;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

4226
3rdparty/libyuv/source/convert_from_argb.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

602
3rdparty/libyuv/source/convert_jpeg.cc vendored Normal file
View File

@@ -0,0 +1,602 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/convert.h"
#include "libyuv/convert_argb.h"
#ifdef HAVE_JPEG
#include "libyuv/mjpeg_decoder.h"
#endif
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#ifdef HAVE_JPEG
struct I420Buffers {
uint8_t* y;
int y_stride;
uint8_t* u;
int u_stride;
uint8_t* v;
int v_stride;
int w;
int h;
};
static void JpegCopyI420(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I420Copy(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
dest->h -= rows;
}
static void JpegI422ToI420(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I422ToI420(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
dest->h -= rows;
}
static void JpegI444ToI420(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I444ToI420(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
dest->h -= rows;
}
static void JpegI400ToI420(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I400ToI420(data[0], strides[0], dest->y, dest->y_stride, dest->u,
dest->u_stride, dest->v, dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
dest->h -= rows;
}
// Query size of MJPG in pixels.
LIBYUV_API
int MJPGSize(const uint8_t* src_mjpg,
size_t src_size_mjpg,
int* width,
int* height) {
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(src_mjpg, src_size_mjpg);
if (ret) {
*width = mjpeg_decoder.GetWidth();
*height = mjpeg_decoder.GetHeight();
}
mjpeg_decoder.UnloadFrame();
return ret ? 0 : -1; // -1 for runtime failure.
}
// MJPG (Motion JPeg) to I420
// TODO(fbarchard): review src_width and src_height requirement. dst_width and
// dst_height may be enough.
LIBYUV_API
int MJPGToI420(const uint8_t* src_mjpg,
size_t src_size_mjpg,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_u,
int dst_stride_u,
uint8_t* dst_v,
int dst_stride_v,
int src_width,
int src_height,
int dst_width,
int dst_height) {
if (src_size_mjpg == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
}
// TODO(fbarchard): Port MJpeg to C.
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(src_mjpg, src_size_mjpg);
if (ret && (mjpeg_decoder.GetWidth() != src_width ||
mjpeg_decoder.GetHeight() != src_height)) {
// ERROR: MJPEG frame has unexpected dimensions
mjpeg_decoder.UnloadFrame();
return 1; // runtime failure
}
if (ret) {
I420Buffers bufs = {dst_y, dst_stride_y, dst_u, dst_stride_u,
dst_v, dst_stride_v, dst_width, dst_height};
// YUV420
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegCopyI420, &bufs, dst_width,
dst_height);
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToI420, &bufs, dst_width,
dst_height);
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToI420, &bufs, dst_width,
dst_height);
// YUV400
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceGrayscale &&
mjpeg_decoder.GetNumComponents() == 1 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToI420, &bufs, dst_width,
dst_height);
} else {
// TODO(fbarchard): Implement conversion for any other
// colorspace/subsample factors that occur in practice. ERROR: Unable to
// convert MJPEG frame because format is not supported
mjpeg_decoder.UnloadFrame();
return 1;
}
}
return ret ? 0 : 1;
}
struct NV21Buffers {
uint8_t* y;
int y_stride;
uint8_t* vu;
int vu_stride;
int w;
int h;
};
static void JpegI420ToNV21(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
I420ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI422ToNV21(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
I422ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI444ToNV21(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
I444ToNV21(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI400ToNV21(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
I400ToNV21(data[0], strides[0], dest->y, dest->y_stride, dest->vu,
dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
// MJPG (Motion JPeg) to NV21
LIBYUV_API
int MJPGToNV21(const uint8_t* src_mjpg,
size_t src_size_mjpg,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_vu,
int dst_stride_vu,
int src_width,
int src_height,
int dst_width,
int dst_height) {
if (src_size_mjpg == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
}
// TODO(fbarchard): Port MJpeg to C.
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(src_mjpg, src_size_mjpg);
if (ret && (mjpeg_decoder.GetWidth() != src_width ||
mjpeg_decoder.GetHeight() != src_height)) {
// ERROR: MJPEG frame has unexpected dimensions
mjpeg_decoder.UnloadFrame();
return 1; // runtime failure
}
if (ret) {
NV21Buffers bufs = {dst_y, dst_stride_y, dst_vu,
dst_stride_vu, dst_width, dst_height};
// YUV420
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToNV21, &bufs, dst_width,
dst_height);
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToNV21, &bufs, dst_width,
dst_height);
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToNV21, &bufs, dst_width,
dst_height);
// YUV400
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceGrayscale &&
mjpeg_decoder.GetNumComponents() == 1 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToNV21, &bufs, dst_width,
dst_height);
} else {
// Unknown colorspace.
mjpeg_decoder.UnloadFrame();
return 1;
}
}
return ret ? 0 : 1;
}
static void JpegI420ToNV12(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
// Use NV21 with VU swapped.
I420ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI422ToNV12(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
// Use NV21 with VU swapped.
I422ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI444ToNV12(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
// Use NV21 with VU swapped.
I444ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1],
dest->y, dest->y_stride, dest->vu, dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
static void JpegI400ToNV12(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque);
// Use NV21 since there is no UV plane.
I400ToNV21(data[0], strides[0], dest->y, dest->y_stride, dest->vu,
dest->vu_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->vu += ((rows + 1) >> 1) * dest->vu_stride;
dest->h -= rows;
}
// MJPG (Motion JPEG) to NV12.
LIBYUV_API
int MJPGToNV12(const uint8_t* sample,
size_t sample_size,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int src_width,
int src_height,
int dst_width,
int dst_height) {
if (sample_size == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
}
// TODO(fbarchard): Port MJpeg to C.
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size);
if (ret && (mjpeg_decoder.GetWidth() != src_width ||
mjpeg_decoder.GetHeight() != src_height)) {
// ERROR: MJPEG frame has unexpected dimensions
mjpeg_decoder.UnloadFrame();
return 1; // runtime failure
}
if (ret) {
// Use NV21Buffers but with UV instead of VU.
NV21Buffers bufs = {dst_y, dst_stride_y, dst_uv,
dst_stride_uv, dst_width, dst_height};
// YUV420
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToNV12, &bufs, dst_width,
dst_height);
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToNV12, &bufs, dst_width,
dst_height);
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToNV12, &bufs, dst_width,
dst_height);
// YUV400
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceGrayscale &&
mjpeg_decoder.GetNumComponents() == 1 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToNV12, &bufs, dst_width,
dst_height);
} else {
// Unknown colorspace.
mjpeg_decoder.UnloadFrame();
return 1;
}
}
return ret ? 0 : 1;
}
struct ARGBBuffers {
uint8_t* argb;
int argb_stride;
int w;
int h;
};
static void JpegI420ToARGB(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I420ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
static void JpegI422ToARGB(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I422ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
static void JpegI444ToARGB(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I444ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
static void JpegI400ToARGB(void* opaque,
const uint8_t* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I400ToARGB(data[0], strides[0], dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
// MJPG (Motion JPeg) to ARGB
// TODO(fbarchard): review src_width and src_height requirement. dst_width and
// dst_height may be enough.
LIBYUV_API
int MJPGToARGB(const uint8_t* src_mjpg,
size_t src_size_mjpg,
uint8_t* dst_argb,
int dst_stride_argb,
int src_width,
int src_height,
int dst_width,
int dst_height) {
if (src_size_mjpg == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
}
// TODO(fbarchard): Port MJpeg to C.
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(src_mjpg, src_size_mjpg);
if (ret && (mjpeg_decoder.GetWidth() != src_width ||
mjpeg_decoder.GetHeight() != src_height)) {
// ERROR: MJPEG frame has unexpected dimensions
mjpeg_decoder.UnloadFrame();
return 1; // runtime failure
}
if (ret) {
ARGBBuffers bufs = {dst_argb, dst_stride_argb, dst_width, dst_height};
// YUV420
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToARGB, &bufs, dst_width,
dst_height);
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToARGB, &bufs, dst_width,
dst_height);
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToARGB, &bufs, dst_width,
dst_height);
// YUV400
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceGrayscale &&
mjpeg_decoder.GetNumComponents() == 1 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToARGB, &bufs, dst_width,
dst_height);
} else {
// TODO(fbarchard): Implement conversion for any other
// colorspace/subsample factors that occur in practice. ERROR: Unable to
// convert MJPEG frame because format is not supported
mjpeg_decoder.UnloadFrame();
return 1;
}
}
return ret ? 0 : 1;
}
#endif // HAVE_JPEG
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

391
3rdparty/libyuv/source/convert_to_argb.cc vendored Normal file
View File

@@ -0,0 +1,391 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/convert_argb.h"
#include <limits.h>
#include <stdint.h>
#include <stdlib.h>
#include "libyuv/cpu_id.h"
#ifdef HAVE_JPEG
#include "libyuv/mjpeg_decoder.h"
#endif
#include "libyuv/rotate_argb.h"
#include "libyuv/row.h"
#include "libyuv/video_common.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Convert camera sample to ARGB with cropping, rotation and vertical flip.
// src_width is used for source stride computation
// src_height is used to compute location of planes, and indicate inversion
// sample_size is measured in bytes and is the size of the frame.
// With MJPEG it is the compressed size of the frame.
// TODO(fbarchard): Add the following:
// H010ToARGB
// I010ToARGB
LIBYUV_API
int ConvertToARGB(const uint8_t* sample,
size_t sample_size,
uint8_t* dst_argb,
int dst_stride_argb,
int crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32_t fourcc) {
uint32_t format = CanonicalFourCC(fourcc);
int aligned_src_width = (src_width + 1) & ~1;
const uint8_t* src;
const uint8_t* src_uv;
int abs_src_height = (src_height < 0) ? -src_height : src_height;
int inv_crop_height = (crop_height < 0) ? -crop_height : crop_height;
int r = 0;
// One pass rotation is available for some formats. For the rest, convert
// to ARGB (with optional vertical flipping) into a temporary ARGB buffer,
// and then rotate the ARGB to the final destination buffer.
// For in-place conversion, if destination dst_argb is same as source sample,
// also enable temporary buffer.
LIBYUV_BOOL need_buf =
(rotation && format != FOURCC_ARGB) || dst_argb == sample;
uint8_t* dest_argb = dst_argb;
int dest_dst_stride_argb = dst_stride_argb;
uint8_t* rotate_buffer = NULL;
int abs_crop_height = (crop_height < 0) ? -crop_height : crop_height;
if (dst_argb == NULL || sample == NULL || src_width <= 0 ||
src_width > INT_MAX / 4 || crop_width <= 0 || crop_width > INT_MAX / 4 ||
src_height == 0 || crop_height == 0) {
return -1;
}
if (src_height < 0) {
inv_crop_height = -inv_crop_height;
}
if (need_buf) {
const uint64_t rotate_buffer_size =
(uint64_t)crop_width * 4 * abs_crop_height;
if (rotate_buffer_size > SIZE_MAX) {
return -1; // Invalid size.
}
rotate_buffer = (uint8_t*)malloc((size_t)rotate_buffer_size);
if (!rotate_buffer) {
return 1; // Out of memory runtime error.
}
dst_argb = rotate_buffer;
dst_stride_argb = crop_width * 4;
}
switch (format) {
// Single plane formats
case FOURCC_YUY2:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = YUY2ToARGB(src, aligned_src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_UYVY:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = UYVYToARGB(src, aligned_src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_24BG:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RGB24ToARGB(src, src_width * 3, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_RAW:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RAWToARGB(src, src_width * 3, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_ARGB:
if (!need_buf && !rotation) {
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBToARGB(src, src_width * 4, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
}
break;
case FOURCC_BGRA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = BGRAToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_ABGR:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ABGRToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_RGBA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = RGBAToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_AR30:
src = sample + (src_width * crop_y + crop_x) * 4;
r = AR30ToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_AB30:
src = sample + (src_width * crop_y + crop_x) * 4;
r = AB30ToARGB(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_RGBP:
src = sample + (src_width * crop_y + crop_x) * 2;
r = RGB565ToARGB(src, src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_RGBO:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB1555ToARGB(src, src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_R444:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB4444ToARGB(src, src_width * 2, dst_argb, dst_stride_argb,
crop_width, inv_crop_height);
break;
case FOURCC_I400:
src = sample + src_width * crop_y + crop_x;
r = I400ToARGB(src, src_width, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
case FOURCC_J400:
src = sample + src_width * crop_y + crop_x;
r = J400ToARGB(src, src_width, dst_argb, dst_stride_argb, crop_width,
inv_crop_height);
break;
// Biplanar formats
case FOURCC_NV12:
src = sample + (src_width * crop_y + crop_x);
src_uv =
sample + aligned_src_width * (abs_src_height + crop_y / 2) + crop_x;
r = NV12ToARGB(src, src_width, src_uv, aligned_src_width, dst_argb,
dst_stride_argb, crop_width, inv_crop_height);
break;
case FOURCC_NV21:
src = sample + (src_width * crop_y + crop_x);
src_uv =
sample + aligned_src_width * (abs_src_height + crop_y / 2) + crop_x;
// Call NV12 but with u and v parameters swapped.
r = NV21ToARGB(src, src_width, src_uv, aligned_src_width, dst_argb,
dst_stride_argb, crop_width, inv_crop_height);
break;
// Triplanar formats
case FOURCC_I420:
case FOURCC_YV12: {
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u;
const uint8_t* src_v;
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
if (format == FOURCC_YV12) {
src_v = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
src_u = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
} else {
src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
}
r = I420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_J420: {
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
r = J420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_H420: {
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
r = H420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_U420: {
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
r = U420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_I422:
case FOURCC_YV16: {
int halfwidth = (src_width + 1) / 2;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
if (format == FOURCC_YV16) {
src_v = sample + src_width * abs_src_height + halfwidth * crop_y +
crop_x / 2;
src_u = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
} else {
src_u = sample + src_width * abs_src_height + halfwidth * crop_y +
crop_x / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
}
r = I422ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_J422: {
int halfwidth = (src_width + 1) / 2;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u =
sample + src_width * abs_src_height + halfwidth * crop_y + crop_x / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
r = J422ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_H422: {
int halfwidth = (src_width + 1) / 2;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u =
sample + src_width * abs_src_height + halfwidth * crop_y + crop_x / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
r = H422ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_U422: {
int halfwidth = (src_width + 1) / 2;
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u =
sample + src_width * abs_src_height + halfwidth * crop_y + crop_x / 2;
const uint8_t* src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
r = H422ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_I444:
case FOURCC_YV24: {
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
if (format == FOURCC_YV24) {
src_v = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_u = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
} else {
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
}
r = I444ToARGB(src_y, src_width, src_u, src_width, src_v, src_width,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_J444: {
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
r = J444ToARGB(src_y, src_width, src_u, src_width, src_v, src_width,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_H444: {
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
r = H444ToARGB(src_y, src_width, src_u, src_width, src_v, src_width,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
case FOURCC_U444: {
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
r = U444ToARGB(src_y, src_width, src_u, src_width, src_v, src_width,
dst_argb, dst_stride_argb, crop_width, inv_crop_height);
break;
}
#ifdef HAVE_JPEG
case FOURCC_MJPG:
r = MJPGToARGB(sample, sample_size, dst_argb, dst_stride_argb, src_width,
abs_src_height, crop_width, inv_crop_height);
break;
#endif
default:
r = -1; // unknown fourcc - return failure code.
}
if (need_buf) {
if (!r) {
r = ARGBRotate(dst_argb, dst_stride_argb, dest_argb, dest_dst_stride_argb,
crop_width, abs_crop_height, rotation);
}
free(rotate_buffer);
} else if (rotation) {
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBRotate(src, src_width * 4, dst_argb, dst_stride_argb, crop_width,
inv_crop_height, rotation);
}
return r;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

288
3rdparty/libyuv/source/convert_to_i420.cc vendored Normal file
View File

@@ -0,0 +1,288 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/convert.h"
#include <limits.h>
#include <stddef.h>
#include <stdint.h>
#include <stdlib.h>
#include "libyuv/video_common.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Convert camera sample to I420 with cropping, rotation and vertical flip.
// src_width is used for source stride computation
// src_height is used to compute location of planes, and indicate inversion
// sample_size is measured in bytes and is the size of the frame.
// With MJPEG it is the compressed size of the frame.
LIBYUV_API
int ConvertToI420(const uint8_t* sample,
size_t sample_size,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_u,
int dst_stride_u,
uint8_t* dst_v,
int dst_stride_v,
int crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32_t fourcc) {
uint32_t format = CanonicalFourCC(fourcc);
int aligned_src_width = (src_width + 1) & ~1;
const uint8_t* src;
const uint8_t* src_uv;
const int abs_src_height = (src_height < 0) ? -src_height : src_height;
const int abs_crop_height = (crop_height < 0) ? -crop_height : crop_height;
int r = 0;
LIBYUV_BOOL need_buf =
(rotation && format != FOURCC_I420 && format != FOURCC_NV12 &&
format != FOURCC_NV21 && format != FOURCC_YV12) ||
dst_y == sample;
uint8_t* tmp_y = dst_y;
uint8_t* tmp_u = dst_u;
uint8_t* tmp_v = dst_v;
int tmp_y_stride = dst_stride_y;
int tmp_u_stride = dst_stride_u;
int tmp_v_stride = dst_stride_v;
uint8_t* rotate_buffer = NULL;
const int inv_crop_height =
(src_height < 0) ? -abs_crop_height : abs_crop_height;
if (!dst_y || !dst_u || !dst_v || !sample || src_width <= 0 ||
src_width > INT_MAX / 4 || crop_width <= 0 || src_height == 0 ||
crop_height == 0) {
return -1;
}
// One pass rotation is available for some formats. For the rest, convert
// to I420 (with optional vertical flipping) into a temporary I420 buffer,
// and then rotate the I420 to the final destination buffer.
// For in-place conversion, if destination dst_y is same as source sample,
// also enable temporary buffer.
if (need_buf) {
int y_size = crop_width * abs_crop_height;
int uv_size = ((crop_width + 1) / 2) * ((abs_crop_height + 1) / 2);
const uint64_t rotate_buffer_size =
(uint64_t)y_size + (uint64_t)uv_size * 2;
if (rotate_buffer_size > SIZE_MAX) {
return -1; // Invalid size.
}
rotate_buffer = (uint8_t*)malloc((size_t)rotate_buffer_size);
if (!rotate_buffer) {
return 1; // Out of memory runtime error.
}
dst_y = rotate_buffer;
dst_u = dst_y + y_size;
dst_v = dst_u + uv_size;
dst_stride_y = crop_width;
dst_stride_u = dst_stride_v = ((crop_width + 1) / 2);
}
switch (format) {
// Single plane formats
case FOURCC_YUY2: { // TODO(fbarchard): Find better odd crop fix.
uint8_t* u = (crop_x & 1) ? dst_v : dst_u;
uint8_t* v = (crop_x & 1) ? dst_u : dst_v;
int stride_u = (crop_x & 1) ? dst_stride_v : dst_stride_u;
int stride_v = (crop_x & 1) ? dst_stride_u : dst_stride_v;
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = YUY2ToI420(src, aligned_src_width * 2, dst_y, dst_stride_y, u,
stride_u, v, stride_v, crop_width, inv_crop_height);
break;
}
case FOURCC_UYVY: {
uint8_t* u = (crop_x & 1) ? dst_v : dst_u;
uint8_t* v = (crop_x & 1) ? dst_u : dst_v;
int stride_u = (crop_x & 1) ? dst_stride_v : dst_stride_u;
int stride_v = (crop_x & 1) ? dst_stride_u : dst_stride_v;
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = UYVYToI420(src, aligned_src_width * 2, dst_y, dst_stride_y, u,
stride_u, v, stride_v, crop_width, inv_crop_height);
break;
}
case FOURCC_RGBP:
src = sample + (src_width * crop_y + crop_x) * 2;
r = RGB565ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_RGBO:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB1555ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_R444:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB4444ToI420(src, src_width * 2, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_24BG:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RGB24ToI420(src, src_width * 3, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_RAW:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RAWToI420(src, src_width * 3, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_ARGB:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_BGRA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = BGRAToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_ABGR:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ABGRToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
case FOURCC_RGBA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = RGBAToI420(src, src_width * 4, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, crop_width,
inv_crop_height);
break;
// TODO(fbarchard): Add AR30 and AB30
case FOURCC_I400:
src = sample + src_width * crop_y + crop_x;
r = I400ToI420(src, src_width, dst_y, dst_stride_y, dst_u, dst_stride_u,
dst_v, dst_stride_v, crop_width, inv_crop_height);
break;
// Biplanar formats
case FOURCC_NV12:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + (src_width * abs_src_height) +
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, dst_y,
dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, crop_width, inv_crop_height, rotation);
break;
case FOURCC_NV21:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + (src_width * abs_src_height) +
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
// Call NV12 but with dst_u and dst_v parameters swapped.
r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, dst_y,
dst_stride_y, dst_v, dst_stride_v, dst_u,
dst_stride_u, crop_width, inv_crop_height, rotation);
break;
// Triplanar formats
case FOURCC_I420:
case FOURCC_YV12: {
const uint8_t* src_y = sample + (src_width * crop_y + crop_x);
const uint8_t* src_u;
const uint8_t* src_v;
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
if (format == FOURCC_YV12) {
src_v = sample + src_width * abs_src_height + halfwidth * (crop_y / 2) +
(crop_x / 2);
src_u = sample + src_width * abs_src_height +
halfwidth * (halfheight + (crop_y / 2)) + (crop_x / 2);
} else {
src_u = sample + src_width * abs_src_height + halfwidth * (crop_y / 2) +
(crop_x / 2);
src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + (crop_y / 2)) + (crop_x / 2);
}
r = I420Rotate(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, crop_width, inv_crop_height, rotation);
break;
}
case FOURCC_I422:
case FOURCC_YV16: {
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
int halfwidth = (src_width + 1) / 2;
if (format == FOURCC_YV16) {
src_v = sample + src_width * abs_src_height + halfwidth * crop_y +
(crop_x / 2);
src_u = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + (crop_x / 2);
} else {
src_u = sample + src_width * abs_src_height + halfwidth * crop_y +
(crop_x / 2);
src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + (crop_x / 2);
}
r = I422ToI420(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, crop_width, inv_crop_height);
break;
}
case FOURCC_I444:
case FOURCC_YV24: {
const uint8_t* src_y = sample + src_width * crop_y + crop_x;
const uint8_t* src_u;
const uint8_t* src_v;
if (format == FOURCC_YV24) {
src_v = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_u = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
} else {
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
}
r = I444ToI420(src_y, src_width, src_u, src_width, src_v, src_width,
dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, crop_width, inv_crop_height);
break;
}
#ifdef HAVE_JPEG
case FOURCC_MJPG:
r = MJPGToI420(sample, sample_size, dst_y, dst_stride_y, dst_u,
dst_stride_u, dst_v, dst_stride_v, src_width,
abs_src_height, crop_width, inv_crop_height);
break;
#endif
default:
r = -1; // unknown fourcc - return failure code.
}
if (need_buf) {
if (!r) {
r = I420Rotate(dst_y, dst_stride_y, dst_u, dst_stride_u, dst_v,
dst_stride_v, tmp_y, tmp_y_stride, tmp_u, tmp_u_stride,
tmp_v, tmp_v_stride, crop_width, abs_crop_height,
rotation);
}
free(rotate_buffer);
}
return r;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

501
3rdparty/libyuv/source/cpu_id.cc vendored Normal file
View File

@@ -0,0 +1,501 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/cpu_id.h"
#if defined(_MSC_VER)
#include <intrin.h> // For __cpuidex()
#endif
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
!defined(__native_client__) && (defined(_M_IX86) || defined(_M_X64)) && \
defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
#include <immintrin.h> // For _xgetbv()
#endif
// For ArmCpuCaps() but unittested on all platforms
#include <stdio.h> // For fopen()
#include <string.h>
#if defined(__linux__) && (defined(__aarch64__) || defined(__loongarch__))
#include <sys/auxv.h> // For getauxval()
#endif
#if defined(_WIN32) && defined(__aarch64__)
#undef WIN32_LEAN_AND_MEAN
#define WIN32_LEAN_AND_MEAN
#undef WIN32_EXTRA_LEAN
#define WIN32_EXTRA_LEAN
#include <windows.h> // For IsProcessorFeaturePresent()
#endif
#if defined(__APPLE__) && defined(__aarch64__)
#include <sys/sysctl.h> // For sysctlbyname()
#endif
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// For functions that use the stack and have runtime checks for overflow,
// use SAFEBUFFERS to avoid additional check.
#if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219) && \
!defined(__clang__)
#define SAFEBUFFERS __declspec(safebuffers)
#else
#define SAFEBUFFERS
#endif
// cpu_info_ variable for SIMD instruction sets detected.
LIBYUV_API int cpu_info_ = 0;
// Low level cpuid for X86.
#if (defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || \
defined(__x86_64__)) && \
!defined(__pnacl__) && !defined(__CLR_VER)
LIBYUV_API
void CpuId(int info_eax, int info_ecx, int* cpu_info) {
#if defined(_MSC_VER)
// Visual C version uses intrinsic or inline x86 assembly.
#if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
__cpuidex(cpu_info, info_eax, info_ecx);
#elif defined(_M_IX86)
__asm {
mov eax, info_eax
mov ecx, info_ecx
mov edi, cpu_info
cpuid
mov [edi], eax
mov [edi + 4], ebx
mov [edi + 8], ecx
mov [edi + 12], edx
}
#else // Visual C but not x86
if (info_ecx == 0) {
__cpuid(cpu_info, info_eax);
} else {
cpu_info[3] = cpu_info[2] = cpu_info[1] = cpu_info[0] = 0u;
}
#endif
// GCC version uses inline x86 assembly.
#else // defined(_MSC_VER)
int info_ebx, info_edx;
asm volatile(
#if defined(__i386__) && defined(__PIC__)
// Preserve ebx for fpic 32 bit.
"mov %%ebx, %%edi \n"
"cpuid \n"
"xchg %%edi, %%ebx \n"
: "=D"(info_ebx),
#else
"cpuid \n"
: "=b"(info_ebx),
#endif // defined( __i386__) && defined(__PIC__)
"+a"(info_eax), "+c"(info_ecx), "=d"(info_edx));
cpu_info[0] = info_eax;
cpu_info[1] = info_ebx;
cpu_info[2] = info_ecx;
cpu_info[3] = info_edx;
#endif // defined(_MSC_VER)
}
#else // (defined(_M_IX86) || defined(_M_X64) ...
LIBYUV_API
void CpuId(int eax, int ecx, int* cpu_info) {
(void)eax;
(void)ecx;
cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0;
}
#endif
// For VS2010 and earlier emit can be used:
// _asm _emit 0x0f _asm _emit 0x01 _asm _emit 0xd0 // For VS2010 and earlier.
// __asm {
// xor ecx, ecx // xcr 0
// xgetbv
// mov xcr0, eax
// }
// For VS2013 and earlier 32 bit, the _xgetbv(0) optimizer produces bad code.
// https://code.google.com/p/libyuv/issues/detail?id=529
#if defined(_M_IX86) && defined(_MSC_VER) && (_MSC_VER < 1900)
#pragma optimize("g", off)
#endif
#if (defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || \
defined(__x86_64__)) && \
!defined(__pnacl__) && !defined(__CLR_VER) && !defined(__native_client__)
// X86 CPUs have xgetbv to detect OS saves high parts of ymm registers.
static int GetXCR0() {
int xcr0 = 0;
#if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
xcr0 = (int)_xgetbv(0); // VS2010 SP1 required. NOLINT
#elif defined(__i386__) || defined(__x86_64__)
asm(".byte 0x0f, 0x01, 0xd0" : "=a"(xcr0) : "c"(0) : "%edx");
#endif // defined(__i386__) || defined(__x86_64__)
return xcr0;
}
#else
// xgetbv unavailable to query for OSSave support. Return 0.
#define GetXCR0() 0
#endif // defined(_M_IX86) || defined(_M_X64) ..
// Return optimization to previous setting.
#if defined(_M_IX86) && defined(_MSC_VER) && (_MSC_VER < 1900)
#pragma optimize("g", on)
#endif
static int cpuinfo_search(const char* cpuinfo_line,
const char* needle,
int needle_len) {
const char* p = strstr(cpuinfo_line, needle);
return p && (p[needle_len] == ' ' || p[needle_len] == '\n');
}
// Based on libvpx arm_cpudetect.c
// For Arm, but public to allow testing on any CPU
LIBYUV_API SAFEBUFFERS int ArmCpuCaps(const char* cpuinfo_name) {
char cpuinfo_line[512];
FILE* f = fopen(cpuinfo_name, "re");
if (!f) {
// Assume Neon if /proc/cpuinfo is unavailable.
// This will occur for Chrome sandbox for Pepper or Render process.
return kCpuHasNEON;
}
memset(cpuinfo_line, 0, sizeof(cpuinfo_line));
int features = 0;
while (fgets(cpuinfo_line, sizeof(cpuinfo_line), f)) {
if (memcmp(cpuinfo_line, "Features", 8) == 0) {
if (cpuinfo_search(cpuinfo_line, " neon", 5)) {
features |= kCpuHasNEON;
}
}
}
fclose(f);
return features;
}
#ifdef __aarch64__
#ifdef __linux__
// Define hwcap values ourselves: building with an old auxv header where these
// hwcap values are not defined should not prevent features from being enabled.
#define YUV_AARCH64_HWCAP_ASIMDDP (1UL << 20)
#define YUV_AARCH64_HWCAP_SVE (1UL << 22)
#define YUV_AARCH64_HWCAP2_SVE2 (1UL << 1)
#define YUV_AARCH64_HWCAP2_SVEF32MM (1UL << 10)
#define YUV_AARCH64_HWCAP2_I8MM (1UL << 13)
#define YUV_AARCH64_HWCAP2_SME (1UL << 23)
#define YUV_AARCH64_HWCAP2_SME2 (1UL << 37)
// For AArch64, but public to allow testing on any CPU.
LIBYUV_API SAFEBUFFERS int AArch64CpuCaps(unsigned long hwcap,
unsigned long hwcap2) {
// Neon is mandatory on AArch64, so enable regardless of hwcaps.
int features = kCpuHasNEON;
// Don't try to enable later extensions unless earlier extensions are also
// reported available. Some of these constraints aren't strictly required by
// the architecture, but are satisfied by all micro-architectures of
// interest. This also avoids an issue on some emulators where true
// architectural constraints are not satisfied, e.g. SVE2 may be reported as
// available while SVE is not.
if (hwcap & YUV_AARCH64_HWCAP_ASIMDDP) {
features |= kCpuHasNeonDotProd;
if (hwcap2 & YUV_AARCH64_HWCAP2_I8MM) {
features |= kCpuHasNeonI8MM;
if (hwcap & YUV_AARCH64_HWCAP_SVE) {
features |= kCpuHasSVE;
if (hwcap2 & YUV_AARCH64_HWCAP2_SVEF32MM) {
features |= kCpuHasSVEF32MM;
}
if (hwcap2 & YUV_AARCH64_HWCAP2_SVE2) {
features |= kCpuHasSVE2;
}
}
// SME may be present without SVE
if (hwcap2 & YUV_AARCH64_HWCAP2_SME) {
features |= kCpuHasSME;
if (hwcap2 & YUV_AARCH64_HWCAP2_SME2) {
features |= kCpuHasSME2;
}
}
}
}
return features;
}
#elif defined(_WIN32)
// For AArch64, but public to allow testing on any CPU.
LIBYUV_API SAFEBUFFERS int AArch64CpuCaps() {
// Neon is mandatory on AArch64, so enable unconditionally.
int features = kCpuHasNEON;
// For more information on IsProcessorFeaturePresent(), see:
// https://learn.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-isprocessorfeaturepresent#parameters
#ifdef PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE
if (IsProcessorFeaturePresent(PF_ARM_V82_DP_INSTRUCTIONS_AVAILABLE)) {
features |= kCpuHasNeonDotProd;
}
#endif
// No Neon I8MM or SVE feature detection available here at time of writing.
return features;
}
#elif defined(__APPLE__)
static bool have_feature(const char* feature) {
// For more information on sysctlbyname(), see:
// https://developer.apple.com/documentation/kernel/1387446-sysctlbyname/determining_instruction_set_characteristics
int64_t feature_present = 0;
size_t size = sizeof(feature_present);
if (sysctlbyname(feature, &feature_present, &size, NULL, 0) != 0) {
return false;
}
return feature_present;
}
// For AArch64, but public to allow testing on any CPU.
LIBYUV_API SAFEBUFFERS int AArch64CpuCaps() {
// Neon is mandatory on AArch64, so enable unconditionally.
int features = kCpuHasNEON;
if (have_feature("hw.optional.arm.FEAT_DotProd")) {
features |= kCpuHasNeonDotProd;
if (have_feature("hw.optional.arm.FEAT_I8MM")) {
features |= kCpuHasNeonI8MM;
if (have_feature("hw.optional.arm.FEAT_SME")) {
features |= kCpuHasSME;
if (have_feature("hw.optional.arm.FEAT_SME2")) {
features |= kCpuHasSME2;
}
}
}
}
// No SVE feature detection available here at time of writing.
return features;
}
#else // !defined(__linux__) && !defined(_WIN32) && !defined(__APPLE__)
// For AArch64, but public to allow testing on any CPU.
LIBYUV_API SAFEBUFFERS int AArch64CpuCaps() {
// Neon is mandatory on AArch64, so enable unconditionally.
int features = kCpuHasNEON;
// TODO(libyuv:980) support feature detection on other platforms.
return features;
}
#endif
#endif // defined(__aarch64__)
LIBYUV_API SAFEBUFFERS int RiscvCpuCaps(const char* cpuinfo_name) {
char cpuinfo_line[512];
int flag = 0;
FILE* f = fopen(cpuinfo_name, "re");
if (!f) {
#if defined(__riscv_vector)
// Assume RVV if /proc/cpuinfo is unavailable.
// This will occur for Chrome sandbox for Pepper or Render process.
return kCpuHasRVV;
#else
return 0;
#endif
}
memset(cpuinfo_line, 0, sizeof(cpuinfo_line));
while (fgets(cpuinfo_line, sizeof(cpuinfo_line), f)) {
if (memcmp(cpuinfo_line, "isa", 3) == 0) {
// ISA string must begin with rv64{i,e,g} for a 64-bit processor.
char* isa = strstr(cpuinfo_line, "rv64");
if (isa) {
size_t isa_len = strlen(isa);
char* extensions;
size_t extensions_len = 0;
size_t std_isa_len;
// Remove the new-line character at the end of string
if (isa[isa_len - 1] == '\n') {
isa[--isa_len] = '\0';
}
// 5 ISA characters
if (isa_len < 5) {
fclose(f);
return 0;
}
// Skip {i,e,g} canonical checking.
// Skip rvxxx
isa += 5;
// Find the very first occurrence of 's', 'x' or 'z'.
// To detect multi-letter standard, non-standard, and
// supervisor-level extensions.
extensions = strpbrk(isa, "zxs");
if (extensions) {
extensions_len = strlen(extensions);
// Multi-letter extensions are seperated by a single underscore
// as described in RISC-V User-Level ISA V2.2.
char* ext = extensions;
while (ext) {
char* next = strchr(ext, '_');
if (next) {
*next = '\0';
next++;
}
// Search for the ZVFH (Vector FP16) extension.
if (!strcmp(ext, "zvfh")) {
flag |= kCpuHasRVVZVFH;
}
ext = next;
}
}
std_isa_len = isa_len - extensions_len - 5;
// Detect the v in the standard single-letter extensions.
if (memchr(isa, 'v', std_isa_len)) {
// The RVV implied the F extension.
flag |= kCpuHasRVV;
}
}
}
#if defined(__riscv_vector)
// Assume RVV if /proc/cpuinfo is from x86 host running QEMU.
else if ((memcmp(cpuinfo_line, "vendor_id\t: GenuineIntel", 24) == 0) ||
(memcmp(cpuinfo_line, "vendor_id\t: AuthenticAMD", 24) == 0)) {
fclose(f);
return kCpuHasRVV;
}
#endif
}
fclose(f);
return flag;
}
#if defined(__loongarch__) && defined(__linux__)
// Define hwcap values ourselves: building with an old auxv header where these
// hwcap values are not defined should not prevent features from being enabled.
#define YUV_LOONGARCH_HWCAP_LSX (1 << 4)
#define YUV_LOONGARCH_HWCAP_LASX (1 << 5)
LIBYUV_API SAFEBUFFERS int LoongArchCpuCaps(void) {
int flag = 0;
unsigned long hwcap = getauxval(AT_HWCAP);
if (hwcap & YUV_LOONGARCH_HWCAP_LSX)
flag |= kCpuHasLSX;
if (hwcap & YUV_LOONGARCH_HWCAP_LASX)
flag |= kCpuHasLASX;
return flag;
}
#endif
static SAFEBUFFERS int GetCpuFlags(void) {
int cpu_info = 0;
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
(defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || \
defined(_M_IX86))
int cpu_info0[4] = {0, 0, 0, 0};
int cpu_info1[4] = {0, 0, 0, 0};
int cpu_info7[4] = {0, 0, 0, 0};
int cpu_einfo7[4] = {0, 0, 0, 0};
int cpu_info24[4] = {0, 0, 0, 0};
int cpu_amdinfo21[4] = {0, 0, 0, 0};
CpuId(0, 0, cpu_info0);
CpuId(1, 0, cpu_info1);
if (cpu_info0[0] >= 7) {
CpuId(7, 0, cpu_info7);
CpuId(7, 1, cpu_einfo7);
CpuId(0x80000021, 0, cpu_amdinfo21);
}
if (cpu_info0[0] >= 0x24) {
CpuId(0x24, 0, cpu_info24);
}
cpu_info = kCpuHasX86 | ((cpu_info1[3] & 0x04000000) ? kCpuHasSSE2 : 0) |
((cpu_info1[2] & 0x00000200) ? kCpuHasSSSE3 : 0) |
((cpu_info1[2] & 0x00080000) ? kCpuHasSSE41 : 0) |
((cpu_info1[2] & 0x00100000) ? kCpuHasSSE42 : 0) |
((cpu_info7[1] & 0x00000200) ? kCpuHasERMS : 0) |
((cpu_info7[3] & 0x00000010) ? kCpuHasFSMR : 0);
// AVX requires OS saves YMM registers.
if (((cpu_info1[2] & 0x1c000000) == 0x1c000000) && // AVX and OSXSave
((GetXCR0() & 6) == 6)) { // Test OS saves YMM registers
cpu_info |= kCpuHasAVX | ((cpu_info7[1] & 0x00000020) ? kCpuHasAVX2 : 0) |
((cpu_info1[2] & 0x00001000) ? kCpuHasFMA3 : 0) |
((cpu_info1[2] & 0x20000000) ? kCpuHasF16C : 0) |
((cpu_einfo7[0] & 0x00000010) ? kCpuHasAVXVNNI : 0) |
((cpu_einfo7[3] & 0x00000010) ? kCpuHasAVXVNNIINT8 : 0);
cpu_info |= ((cpu_amdinfo21[0] & 0x00008000) ? kCpuHasERMS : 0);
// Detect AVX512bw
if ((GetXCR0() & 0xe0) == 0xe0 && (cpu_info7[1] & 0x00010000)) {
cpu_info |= ((cpu_info7[1] & 0x40000000) ? kCpuHasAVX512BW : 0) |
((cpu_info7[1] & 0x80000000) ? kCpuHasAVX512VL : 0) |
((cpu_info7[2] & 0x00000002) ? kCpuHasAVX512VBMI : 0) |
((cpu_info7[2] & 0x00000040) ? kCpuHasAVX512VBMI2 : 0) |
((cpu_info7[2] & 0x00000800) ? kCpuHasAVX512VNNI : 0) |
((cpu_info7[2] & 0x00001000) ? kCpuHasAVX512VBITALG : 0) |
((cpu_einfo7[3] & 0x00080000) ? kCpuHasAVX10 : 0) |
((cpu_info7[3] & 0x02000000) ? kCpuHasAMXINT8 : 0);
if (cpu_info0[0] >= 0x24 && (cpu_einfo7[3] & 0x00080000)) {
cpu_info |= ((cpu_info24[1] & 0xFF) >= 2) ? kCpuHasAVX10_2 : 0;
}
}
}
#endif
#if defined(__loongarch__) && defined(__linux__)
cpu_info = LoongArchCpuCaps();
cpu_info |= kCpuHasLOONGARCH;
#endif
#if defined(__aarch64__)
#if defined(__linux__)
// getauxval is supported since Android SDK version 18, minimum at time of
// writing is 21, so should be safe to always use this. If getauxval is
// somehow disabled then getauxval returns 0, which will leave Neon enabled
// since Neon is mandatory on AArch64.
unsigned long hwcap = getauxval(AT_HWCAP);
unsigned long hwcap2 = getauxval(AT_HWCAP2);
cpu_info = AArch64CpuCaps(hwcap, hwcap2);
#else
cpu_info = AArch64CpuCaps();
#endif
cpu_info |= kCpuHasARM;
#endif // __aarch64__
#if defined(__arm__)
// gcc -mfpu=neon defines __ARM_NEON__
// __ARM_NEON__ generates code that requires Neon. NaCL also requires Neon.
// For Linux, /proc/cpuinfo can be tested but without that assume Neon.
// Linux arm parse text file for neon detect.
#if defined(__linux__)
cpu_info = ArmCpuCaps("/proc/cpuinfo");
#elif defined(__ARM_NEON__)
cpu_info = kCpuHasNEON;
#else
cpu_info = 0;
#endif
cpu_info |= kCpuHasARM;
#endif // __arm__
#if defined(__riscv) && defined(__linux__)
cpu_info = RiscvCpuCaps("/proc/cpuinfo");
cpu_info |= kCpuHasRISCV;
#endif // __riscv
cpu_info |= kCpuInitialized;
return cpu_info;
}
// Note that use of this function is not thread safe.
LIBYUV_API
int MaskCpuFlags(int enable_flags) {
int cpu_info = GetCpuFlags() & enable_flags;
SetCpuFlags(cpu_info);
return cpu_info;
}
LIBYUV_API
int InitCpuFlags(void) {
return MaskCpuFlags(-1);
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

580
3rdparty/libyuv/source/mjpeg_decoder.cc vendored Normal file
View File

@@ -0,0 +1,580 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/mjpeg_decoder.h"
#ifdef HAVE_JPEG
#include <assert.h>
#if !defined(__pnacl__) && !defined(__CLR_VER) && !defined(COVERAGE_ENABLED)
// Must be included before jpeglib.
#include <setjmp.h>
#define HAVE_SETJMP
#if defined(_MSC_VER)
// disable warning 4324: structure was padded due to __declspec(align())
#pragma warning(disable : 4324)
#endif
#endif
#include <stdio.h> // For jpeglib.h.
// C++ build requires extern C for jpeg internals.
#ifdef __cplusplus
extern "C" {
#endif
#include <jpeglib.h>
#ifdef __cplusplus
} // extern "C"
#endif
#include "libyuv/planar_functions.h" // For CopyPlane().
namespace libyuv {
#ifdef HAVE_SETJMP
struct SetJmpErrorMgr {
jpeg_error_mgr base; // Must be at the top
jmp_buf setjmp_buffer;
};
#endif
const int MJpegDecoder::kColorSpaceUnknown = JCS_UNKNOWN;
const int MJpegDecoder::kColorSpaceGrayscale = JCS_GRAYSCALE;
const int MJpegDecoder::kColorSpaceRgb = JCS_RGB;
const int MJpegDecoder::kColorSpaceYCbCr = JCS_YCbCr;
const int MJpegDecoder::kColorSpaceCMYK = JCS_CMYK;
const int MJpegDecoder::kColorSpaceYCCK = JCS_YCCK;
// Methods that are passed to jpeglib.
boolean fill_input_buffer(jpeg_decompress_struct* cinfo);
void init_source(jpeg_decompress_struct* cinfo);
void skip_input_data(jpeg_decompress_struct* cinfo, long num_bytes); // NOLINT
void term_source(jpeg_decompress_struct* cinfo);
void ErrorHandler(jpeg_common_struct* cinfo);
void OutputHandler(jpeg_common_struct* cinfo);
MJpegDecoder::MJpegDecoder()
: has_scanline_padding_(LIBYUV_FALSE),
num_outbufs_(0),
scanlines_(NULL),
scanlines_sizes_(NULL),
databuf_(NULL),
databuf_strides_(NULL) {
decompress_struct_ = new jpeg_decompress_struct;
source_mgr_ = new jpeg_source_mgr;
#ifdef HAVE_SETJMP
error_mgr_ = new SetJmpErrorMgr;
decompress_struct_->err = jpeg_std_error(&error_mgr_->base);
// Override standard exit()-based error handler.
error_mgr_->base.error_exit = &ErrorHandler;
error_mgr_->base.output_message = &OutputHandler;
#endif
decompress_struct_->client_data = NULL;
source_mgr_->init_source = &init_source;
source_mgr_->fill_input_buffer = &fill_input_buffer;
source_mgr_->skip_input_data = &skip_input_data;
source_mgr_->resync_to_restart = &jpeg_resync_to_restart;
source_mgr_->term_source = &term_source;
jpeg_create_decompress(decompress_struct_);
decompress_struct_->src = source_mgr_;
buf_vec_.buffers = &buf_;
buf_vec_.len = 1;
}
MJpegDecoder::~MJpegDecoder() {
jpeg_destroy_decompress(decompress_struct_);
delete decompress_struct_;
delete source_mgr_;
#ifdef HAVE_SETJMP
delete error_mgr_;
#endif
DestroyOutputBuffers();
}
LIBYUV_BOOL MJpegDecoder::LoadFrame(const uint8_t* src, size_t src_len) {
if (!ValidateJpeg(src, src_len)) {
return LIBYUV_FALSE;
}
buf_.data = src;
buf_.len = (int)src_len;
buf_vec_.pos = 0;
decompress_struct_->client_data = &buf_vec_;
#ifdef HAVE_SETJMP
if (setjmp(error_mgr_->setjmp_buffer)) {
// We called jpeg_read_header, it experienced an error, and we called
// longjmp() and rewound the stack to here. Return error.
return LIBYUV_FALSE;
}
#endif
if (jpeg_read_header(decompress_struct_, TRUE) != JPEG_HEADER_OK) {
// ERROR: Bad MJPEG header
return LIBYUV_FALSE;
}
AllocOutputBuffers(GetNumComponents());
for (int i = 0; i < num_outbufs_; ++i) {
int scanlines_size = GetComponentScanlinesPerImcuRow(i);
if (scanlines_sizes_[i] != scanlines_size) {
if (scanlines_[i]) {
delete scanlines_[i];
}
scanlines_[i] = new uint8_t*[scanlines_size];
scanlines_sizes_[i] = scanlines_size;
}
// We allocate padding for the final scanline to pad it up to DCTSIZE bytes
// to avoid memory errors, since jpeglib only reads full MCUs blocks. For
// the preceding scanlines, the padding is not needed/wanted because the
// following addresses will already be valid (they are the initial bytes of
// the next scanline) and will be overwritten when jpeglib writes out that
// next scanline.
int databuf_stride = GetComponentStride(i);
int databuf_size = scanlines_size * databuf_stride;
if (databuf_strides_[i] != databuf_stride) {
if (databuf_[i]) {
delete databuf_[i];
}
databuf_[i] = new uint8_t[databuf_size];
databuf_strides_[i] = databuf_stride;
}
if (GetComponentStride(i) != GetComponentWidth(i)) {
has_scanline_padding_ = LIBYUV_TRUE;
}
}
return LIBYUV_TRUE;
}
static int DivideAndRoundUp(int numerator, int denominator) {
return (numerator + denominator - 1) / denominator;
}
static int DivideAndRoundDown(int numerator, int denominator) {
return numerator / denominator;
}
// Returns width of the last loaded frame.
int MJpegDecoder::GetWidth() {
return decompress_struct_->image_width;
}
// Returns height of the last loaded frame.
int MJpegDecoder::GetHeight() {
return decompress_struct_->image_height;
}
// Returns format of the last loaded frame. The return value is one of the
// kColorSpace* constants.
int MJpegDecoder::GetColorSpace() {
return decompress_struct_->jpeg_color_space;
}
// Number of color components in the color space.
int MJpegDecoder::GetNumComponents() {
return decompress_struct_->num_components;
}
// Sample factors of the n-th component.
int MJpegDecoder::GetHorizSampFactor(int component) {
return decompress_struct_->comp_info[component].h_samp_factor;
}
int MJpegDecoder::GetVertSampFactor(int component) {
return decompress_struct_->comp_info[component].v_samp_factor;
}
int MJpegDecoder::GetHorizSubSampFactor(int component) {
return decompress_struct_->max_h_samp_factor / GetHorizSampFactor(component);
}
int MJpegDecoder::GetVertSubSampFactor(int component) {
return decompress_struct_->max_v_samp_factor / GetVertSampFactor(component);
}
int MJpegDecoder::GetImageScanlinesPerImcuRow() {
return decompress_struct_->max_v_samp_factor * DCTSIZE;
}
int MJpegDecoder::GetComponentScanlinesPerImcuRow(int component) {
int vs = GetVertSubSampFactor(component);
return DivideAndRoundUp(GetImageScanlinesPerImcuRow(), vs);
}
int MJpegDecoder::GetComponentWidth(int component) {
int hs = GetHorizSubSampFactor(component);
return DivideAndRoundUp(GetWidth(), hs);
}
int MJpegDecoder::GetComponentHeight(int component) {
int vs = GetVertSubSampFactor(component);
return DivideAndRoundUp(GetHeight(), vs);
}
// Get width in bytes padded out to a multiple of DCTSIZE
int MJpegDecoder::GetComponentStride(int component) {
return (GetComponentWidth(component) + DCTSIZE - 1) & ~(DCTSIZE - 1);
}
int MJpegDecoder::GetComponentSize(int component) {
return GetComponentWidth(component) * GetComponentHeight(component);
}
LIBYUV_BOOL MJpegDecoder::UnloadFrame() {
#ifdef HAVE_SETJMP
if (setjmp(error_mgr_->setjmp_buffer)) {
// We called jpeg_abort_decompress, it experienced an error, and we called
// longjmp() and rewound the stack to here. Return error.
return LIBYUV_FALSE;
}
#endif
jpeg_abort_decompress(decompress_struct_);
return LIBYUV_TRUE;
}
// TODO(fbarchard): Allow rectangle to be specified: x, y, width, height.
LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(uint8_t** planes,
int dst_width,
int dst_height) {
if (dst_width != GetWidth() || dst_height > GetHeight()) {
// ERROR: Bad dimensions
return LIBYUV_FALSE;
}
#ifdef HAVE_SETJMP
if (setjmp(error_mgr_->setjmp_buffer)) {
// We called into jpeglib, it experienced an error sometime during this
// function call, and we called longjmp() and rewound the stack to here.
// Return error.
return LIBYUV_FALSE;
}
#endif
if (!StartDecode()) {
return LIBYUV_FALSE;
}
SetScanlinePointers(databuf_);
int lines_left = dst_height;
// Compute amount of lines to skip to implement vertical crop.
// TODO(fbarchard): Ensure skip is a multiple of maximum component
// subsample. ie 2
int skip = (GetHeight() - dst_height) / 2;
if (skip > 0) {
// There is no API to skip lines in the output data, so we read them
// into the temp buffer.
while (skip >= GetImageScanlinesPerImcuRow()) {
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
skip -= GetImageScanlinesPerImcuRow();
}
if (skip > 0) {
// Have a partial iMCU row left over to skip. Must read it and then
// copy the parts we want into the destination.
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
for (int i = 0; i < num_outbufs_; ++i) {
// TODO(fbarchard): Compute skip to avoid this
assert(skip % GetVertSubSampFactor(i) == 0);
int rows_to_skip = DivideAndRoundDown(skip, GetVertSubSampFactor(i));
int scanlines_to_copy =
GetComponentScanlinesPerImcuRow(i) - rows_to_skip;
int data_to_skip = rows_to_skip * GetComponentStride(i);
CopyPlane(databuf_[i] + data_to_skip, GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i),
scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
lines_left -= (GetImageScanlinesPerImcuRow() - skip);
}
}
// Read full MCUs but cropped horizontally
for (; lines_left > GetImageScanlinesPerImcuRow();
lines_left -= GetImageScanlinesPerImcuRow()) {
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
for (int i = 0; i < num_outbufs_; ++i) {
int scanlines_to_copy = GetComponentScanlinesPerImcuRow(i);
CopyPlane(databuf_[i], GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i), scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
}
if (lines_left > 0) {
// Have a partial iMCU row left over to decode.
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
for (int i = 0; i < num_outbufs_; ++i) {
int scanlines_to_copy =
DivideAndRoundUp(lines_left, GetVertSubSampFactor(i));
CopyPlane(databuf_[i], GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i), scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
}
return FinishDecode();
}
LIBYUV_BOOL MJpegDecoder::DecodeToCallback(CallbackFunction fn,
void* opaque,
int dst_width,
int dst_height) {
if (dst_width != GetWidth() || dst_height > GetHeight()) {
// ERROR: Bad dimensions
return LIBYUV_FALSE;
}
#ifdef HAVE_SETJMP
if (setjmp(error_mgr_->setjmp_buffer)) {
// We called into jpeglib, it experienced an error sometime during this
// function call, and we called longjmp() and rewound the stack to here.
// Return error.
return LIBYUV_FALSE;
}
#endif
if (!StartDecode()) {
return LIBYUV_FALSE;
}
SetScanlinePointers(databuf_);
int lines_left = dst_height;
// TODO(fbarchard): Compute amount of lines to skip to implement vertical crop
int skip = (GetHeight() - dst_height) / 2;
if (skip > 0) {
while (skip >= GetImageScanlinesPerImcuRow()) {
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
skip -= GetImageScanlinesPerImcuRow();
}
if (skip > 0) {
// Have a partial iMCU row left over to skip.
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
for (int i = 0; i < num_outbufs_; ++i) {
// TODO(fbarchard): Compute skip to avoid this
assert(skip % GetVertSubSampFactor(i) == 0);
int rows_to_skip = DivideAndRoundDown(skip, GetVertSubSampFactor(i));
int data_to_skip = rows_to_skip * GetComponentStride(i);
// Change our own data buffer pointers so we can pass them to the
// callback.
databuf_[i] += data_to_skip;
}
int scanlines_to_copy = GetImageScanlinesPerImcuRow() - skip;
(*fn)(opaque, databuf_, databuf_strides_, scanlines_to_copy);
// Now change them back.
for (int i = 0; i < num_outbufs_; ++i) {
int rows_to_skip = DivideAndRoundDown(skip, GetVertSubSampFactor(i));
int data_to_skip = rows_to_skip * GetComponentStride(i);
databuf_[i] -= data_to_skip;
}
lines_left -= scanlines_to_copy;
}
}
// Read full MCUs until we get to the crop point.
for (; lines_left >= GetImageScanlinesPerImcuRow();
lines_left -= GetImageScanlinesPerImcuRow()) {
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
(*fn)(opaque, databuf_, databuf_strides_, GetImageScanlinesPerImcuRow());
}
if (lines_left > 0) {
// Have a partial iMCU row left over to decode.
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
(*fn)(opaque, databuf_, databuf_strides_, lines_left);
}
return FinishDecode();
}
void init_source(j_decompress_ptr cinfo) {
fill_input_buffer(cinfo);
}
boolean fill_input_buffer(j_decompress_ptr cinfo) {
BufferVector* buf_vec = reinterpret_cast<BufferVector*>(cinfo->client_data);
if (buf_vec->pos >= buf_vec->len) {
// ERROR: No more data
return FALSE;
}
cinfo->src->next_input_byte = buf_vec->buffers[buf_vec->pos].data;
cinfo->src->bytes_in_buffer = buf_vec->buffers[buf_vec->pos].len;
++buf_vec->pos;
return TRUE;
}
void skip_input_data(j_decompress_ptr cinfo, long num_bytes) { // NOLINT
jpeg_source_mgr* src = cinfo->src;
size_t bytes = (size_t)num_bytes;
if (bytes > src->bytes_in_buffer) {
src->next_input_byte = nullptr;
src->bytes_in_buffer = 0;
} else {
src->next_input_byte += bytes;
src->bytes_in_buffer -= bytes;
}
}
void term_source(j_decompress_ptr cinfo) {
(void)cinfo; // Nothing to do.
}
#ifdef HAVE_SETJMP
void ErrorHandler(j_common_ptr cinfo) {
// This is called when a jpeglib command experiences an error. Unfortunately
// jpeglib's error handling model is not very flexible, because it expects the
// error handler to not return--i.e., it wants the program to terminate. To
// recover from errors we use setjmp() as shown in their example. setjmp() is
// C's implementation for the "call with current continuation" functionality
// seen in some functional programming languages.
// A formatted message can be output, but is unsafe for release.
#ifdef DEBUG
char buf[JMSG_LENGTH_MAX];
(*cinfo->err->format_message)(cinfo, buf);
// ERROR: Error in jpeglib: buf
#endif
SetJmpErrorMgr* mgr = reinterpret_cast<SetJmpErrorMgr*>(cinfo->err);
// This rewinds the call stack to the point of the corresponding setjmp()
// and causes it to return (for a second time) with value 1.
longjmp(mgr->setjmp_buffer, 1);
}
// Suppress fprintf warnings.
void OutputHandler(j_common_ptr cinfo) {
(void)cinfo;
}
#endif // HAVE_SETJMP
void MJpegDecoder::AllocOutputBuffers(int num_outbufs) {
if (num_outbufs != num_outbufs_) {
// We could perhaps optimize this case to resize the output buffers without
// necessarily having to delete and recreate each one, but it's not worth
// it.
DestroyOutputBuffers();
scanlines_ = new uint8_t**[num_outbufs];
scanlines_sizes_ = new int[num_outbufs];
databuf_ = new uint8_t*[num_outbufs];
databuf_strides_ = new int[num_outbufs];
for (int i = 0; i < num_outbufs; ++i) {
scanlines_[i] = NULL;
scanlines_sizes_[i] = 0;
databuf_[i] = NULL;
databuf_strides_[i] = 0;
}
num_outbufs_ = num_outbufs;
}
}
void MJpegDecoder::DestroyOutputBuffers() {
for (int i = 0; i < num_outbufs_; ++i) {
delete[] scanlines_[i];
delete[] databuf_[i];
}
delete[] scanlines_;
delete[] databuf_;
delete[] scanlines_sizes_;
delete[] databuf_strides_;
scanlines_ = NULL;
databuf_ = NULL;
scanlines_sizes_ = NULL;
databuf_strides_ = NULL;
num_outbufs_ = 0;
}
// JDCT_IFAST and do_block_smoothing improve performance substantially.
LIBYUV_BOOL MJpegDecoder::StartDecode() {
decompress_struct_->raw_data_out = TRUE;
decompress_struct_->dct_method = JDCT_IFAST; // JDCT_ISLOW is default
decompress_struct_->dither_mode = JDITHER_NONE;
// Not applicable to 'raw':
decompress_struct_->do_fancy_upsampling = (boolean)(LIBYUV_FALSE);
// Only for buffered mode:
decompress_struct_->enable_2pass_quant = (boolean)(LIBYUV_FALSE);
// Blocky but fast:
decompress_struct_->do_block_smoothing = (boolean)(LIBYUV_FALSE);
if (!jpeg_start_decompress(decompress_struct_)) {
// ERROR: Couldn't start JPEG decompressor";
return LIBYUV_FALSE;
}
return LIBYUV_TRUE;
}
LIBYUV_BOOL MJpegDecoder::FinishDecode() {
// jpeglib considers it an error if we finish without decoding the whole
// image, so we call "abort" rather than "finish".
jpeg_abort_decompress(decompress_struct_);
return LIBYUV_TRUE;
}
void MJpegDecoder::SetScanlinePointers(uint8_t** data) {
for (int i = 0; i < num_outbufs_; ++i) {
uint8_t* data_i = data[i];
for (int j = 0; j < scanlines_sizes_[i]; ++j) {
scanlines_[i][j] = data_i;
data_i += GetComponentStride(i);
}
}
}
inline LIBYUV_BOOL MJpegDecoder::DecodeImcuRow() {
return (unsigned int)(GetImageScanlinesPerImcuRow()) ==
jpeg_read_raw_data(decompress_struct_, scanlines_,
GetImageScanlinesPerImcuRow());
}
// The helper function which recognizes the jpeg sub-sampling type.
JpegSubsamplingType MJpegDecoder::JpegSubsamplingTypeHelper(
int* subsample_x,
int* subsample_y,
int number_of_components) {
if (number_of_components == 3) { // Color images.
if (subsample_x[0] == 1 && subsample_y[0] == 1 && subsample_x[1] == 2 &&
subsample_y[1] == 2 && subsample_x[2] == 2 && subsample_y[2] == 2) {
return kJpegYuv420;
}
if (subsample_x[0] == 1 && subsample_y[0] == 1 && subsample_x[1] == 2 &&
subsample_y[1] == 1 && subsample_x[2] == 2 && subsample_y[2] == 1) {
return kJpegYuv422;
}
if (subsample_x[0] == 1 && subsample_y[0] == 1 && subsample_x[1] == 1 &&
subsample_y[1] == 1 && subsample_x[2] == 1 && subsample_y[2] == 1) {
return kJpegYuv444;
}
} else if (number_of_components == 1) { // Grey-scale images.
if (subsample_x[0] == 1 && subsample_y[0] == 1) {
return kJpegYuv400;
}
}
return kJpegUnknown;
}
} // namespace libyuv
#endif // HAVE_JPEG

71
3rdparty/libyuv/source/mjpeg_validate.cc vendored Normal file
View File

@@ -0,0 +1,71 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/mjpeg_decoder.h"
#include <string.h> // For memchr.
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Helper function to scan for EOI marker (0xff 0xd9).
static LIBYUV_BOOL ScanEOI(const uint8_t* src_mjpg, size_t src_size_mjpg) {
if (src_size_mjpg >= 2) {
const uint8_t* end = src_mjpg + src_size_mjpg - 1;
const uint8_t* it = src_mjpg;
while (it < end) {
// TODO(fbarchard): scan for 0xd9 instead.
it = (const uint8_t*)(memchr(it, 0xff, end - it));
if (it == NULL) {
break;
}
if (it[1] == 0xd9) {
return LIBYUV_TRUE; // Success: Valid jpeg.
}
++it; // Skip over current 0xff.
}
}
// ERROR: Invalid jpeg end code not found. Size src_size_mjpg
return LIBYUV_FALSE;
}
// Helper function to validate the jpeg appears intact.
LIBYUV_BOOL ValidateJpeg(const uint8_t* src_mjpg, size_t src_size_mjpg) {
// Maximum size that ValidateJpeg will consider valid.
const size_t kMaxJpegSize = 0x7fffffffull;
const size_t kBackSearchSize = 1024;
if (src_size_mjpg < 64 || src_size_mjpg > kMaxJpegSize || !src_mjpg) {
// ERROR: Invalid jpeg size: src_size_mjpg
return LIBYUV_FALSE;
}
// SOI marker
if (src_mjpg[0] != 0xff || src_mjpg[1] != 0xd8 || src_mjpg[2] != 0xff) {
// ERROR: Invalid jpeg initial start code
return LIBYUV_FALSE;
}
// Look for the End Of Image (EOI) marker near the end of the buffer.
if (src_size_mjpg > kBackSearchSize) {
if (ScanEOI(src_mjpg + src_size_mjpg - kBackSearchSize, kBackSearchSize)) {
return LIBYUV_TRUE; // Success: Valid jpeg.
}
// Reduce search size for forward search.
src_size_mjpg = src_size_mjpg - kBackSearchSize + 1;
}
// Step over SOI marker and scan for EOI.
return ScanEOI(src_mjpg + 2, src_size_mjpg - 2);
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

5671
3rdparty/libyuv/source/planar_functions.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1241
3rdparty/libyuv/source/rotate.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

76
3rdparty/libyuv/source/rotate_any.cc vendored Normal file
View File

@@ -0,0 +1,76 @@
/*
* Copyright 2015 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate.h"
#include "libyuv/rotate_row.h"
#include "libyuv/basic_types.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#define TANY(NAMEANY, TPOS_SIMD, TPOS_C, MASK) \
void NAMEANY(const uint8_t* src, int src_stride, uint8_t* dst, \
int dst_stride, int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst, dst_stride, n); \
} \
TPOS_C(src + n, src_stride, dst + n * dst_stride, dst_stride, r); \
}
#ifdef HAS_TRANSPOSEWX8_NEON
TANY(TransposeWx8_Any_NEON, TransposeWx8_NEON, TransposeWx8_C, 7)
#endif
#ifdef HAS_TRANSPOSEWX16_NEON
TANY(TransposeWx16_Any_NEON, TransposeWx16_NEON, TransposeWx16_C, 15)
#endif
#ifdef HAS_TRANSPOSEWX8_SSSE3
TANY(TransposeWx8_Any_SSSE3, TransposeWx8_SSSE3, TransposeWx8_C, 7)
#endif
#ifdef HAS_TRANSPOSEWX8_FAST_SSSE3
TANY(TransposeWx8_Fast_Any_SSSE3, TransposeWx8_Fast_SSSE3, TransposeWx8_C, 15)
#endif
#ifdef HAS_TRANSPOSEWX16_LSX
TANY(TransposeWx16_Any_LSX, TransposeWx16_LSX, TransposeWx16_C, 15)
#endif
#undef TANY
#define TUVANY(NAMEANY, TPOS_SIMD, MASK) \
void NAMEANY(const uint8_t* src, int src_stride, uint8_t* dst_a, \
int dst_stride_a, uint8_t* dst_b, int dst_stride_b, \
int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, n); \
} \
TransposeUVWx8_C(src + n * 2, src_stride, dst_a + n * dst_stride_a, \
dst_stride_a, dst_b + n * dst_stride_b, dst_stride_b, r); \
}
#ifdef HAS_TRANSPOSEUVWX8_NEON
TUVANY(TransposeUVWx8_Any_NEON, TransposeUVWx8_NEON, 7)
#endif
#ifdef HAS_TRANSPOSEUVWX8_SSE2
TUVANY(TransposeUVWx8_Any_SSE2, TransposeUVWx8_SSE2, 7)
#endif
#ifdef HAS_TRANSPOSEUVWX16_LSX
TUVANY(TransposeUVWx16_Any_LSX, TransposeUVWx16_LSX, 7)
#endif
#undef TUVANY
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

259
3rdparty/libyuv/source/rotate_argb.cc vendored Normal file
View File

@@ -0,0 +1,259 @@
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_argb.h"
#include "libyuv/convert.h"
#include "libyuv/cpu_id.h"
#include "libyuv/planar_functions.h"
#include "libyuv/rotate.h"
#include "libyuv/row.h"
#include "libyuv/scale_row.h" /* for ScaleARGBRowDownEven_ */
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
static int ARGBTranspose(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height) {
int i;
int src_pixel_step = src_stride_argb >> 2;
void (*ScaleARGBRowDownEven)(
const uint8_t* src_argb, ptrdiff_t src_stride_argb, int src_step,
uint8_t* dst_argb, int dst_width) = ScaleARGBRowDownEven_C;
// Check stride is a multiple of 4.
if (src_stride_argb & 3) {
return -1;
}
#if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
ScaleARGBRowDownEven = ScaleARGBRowDownEven_Any_SSE2;
if (IS_ALIGNED(height, 4)) { // Width of dest.
ScaleARGBRowDownEven = ScaleARGBRowDownEven_SSE2;
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWNEVEN_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
ScaleARGBRowDownEven = ScaleARGBRowDownEven_Any_NEON;
if (IS_ALIGNED(height, 4)) { // Width of dest.
ScaleARGBRowDownEven = ScaleARGBRowDownEven_NEON;
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWNEVEN_LSX)
if (TestCpuFlag(kCpuHasLSX)) {
ScaleARGBRowDownEven = ScaleARGBRowDownEven_Any_LSX;
if (IS_ALIGNED(height, 4)) { // Width of dest.
ScaleARGBRowDownEven = ScaleARGBRowDownEven_LSX;
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWNEVEN_RVV)
if (TestCpuFlag(kCpuHasRVV)) {
ScaleARGBRowDownEven = ScaleARGBRowDownEven_RVV;
}
#endif
for (i = 0; i < width; ++i) { // column of source to row of dest.
ScaleARGBRowDownEven(src_argb, 0, src_pixel_step, dst_argb, height);
dst_argb += dst_stride_argb;
src_argb += 4;
}
return 0;
}
static int ARGBRotate90(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height) {
// Rotate by 90 is a ARGBTranspose with the source read
// from bottom to top. So set the source pointer to the end
// of the buffer and flip the sign of the source stride.
src_argb += src_stride_argb * (height - 1);
src_stride_argb = -src_stride_argb;
return ARGBTranspose(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height);
}
static int ARGBRotate270(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height) {
// Rotate by 270 is a ARGBTranspose with the destination written
// from bottom to top. So set the destination pointer to the end
// of the buffer and flip the sign of the destination stride.
dst_argb += dst_stride_argb * (width - 1);
dst_stride_argb = -dst_stride_argb;
return ARGBTranspose(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height);
}
static int ARGBRotate180(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height) {
// Swap first and last row and mirror the content. Uses a temporary row.
const uint8_t* src_bot = src_argb + src_stride_argb * (height - 1);
uint8_t* dst_bot = dst_argb + dst_stride_argb * (height - 1);
int half_height = (height + 1) >> 1;
int y;
void (*ARGBMirrorRow)(const uint8_t* src_argb, uint8_t* dst_argb, int width) =
ARGBMirrorRow_C;
void (*CopyRow)(const uint8_t* src_argb, uint8_t* dst_argb, int width) =
CopyRow_C;
align_buffer_64(row, width * 4);
if (!row)
return 1;
#if defined(HAS_ARGBMIRRORROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
ARGBMirrorRow = ARGBMirrorRow_Any_NEON;
if (IS_ALIGNED(width, 8)) {
ARGBMirrorRow = ARGBMirrorRow_NEON;
}
}
#endif
#if defined(HAS_ARGBMIRRORROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
ARGBMirrorRow = ARGBMirrorRow_Any_SSE2;
if (IS_ALIGNED(width, 4)) {
ARGBMirrorRow = ARGBMirrorRow_SSE2;
}
}
#endif
#if defined(HAS_ARGBMIRRORROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
ARGBMirrorRow = ARGBMirrorRow_Any_AVX2;
if (IS_ALIGNED(width, 8)) {
ARGBMirrorRow = ARGBMirrorRow_AVX2;
}
}
#endif
#if defined(HAS_ARGBMIRRORROW_LSX)
if (TestCpuFlag(kCpuHasLSX)) {
ARGBMirrorRow = ARGBMirrorRow_Any_LSX;
if (IS_ALIGNED(width, 8)) {
ARGBMirrorRow = ARGBMirrorRow_LSX;
}
}
#endif
#if defined(HAS_ARGBMIRRORROW_LASX)
if (TestCpuFlag(kCpuHasLASX)) {
ARGBMirrorRow = ARGBMirrorRow_Any_LASX;
if (IS_ALIGNED(width, 16)) {
ARGBMirrorRow = ARGBMirrorRow_LASX;
}
}
#endif
#if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
CopyRow = IS_ALIGNED(width * 4, 32) ? CopyRow_SSE2 : CopyRow_Any_SSE2;
}
#endif
#if defined(HAS_COPYROW_AVX)
if (TestCpuFlag(kCpuHasAVX)) {
CopyRow = IS_ALIGNED(width * 4, 64) ? CopyRow_AVX : CopyRow_Any_AVX;
}
#endif
#if defined(HAS_COPYROW_AVX512BW)
if (TestCpuFlag(kCpuHasAVX512BW)) {
CopyRow =
IS_ALIGNED(width * 4, 128) ? CopyRow_AVX512BW : CopyRow_Any_AVX512BW;
}
#endif
#if defined(HAS_COPYROW_ERMS)
if (TestCpuFlag(kCpuHasERMS)) {
CopyRow = CopyRow_ERMS;
}
#endif
#if defined(HAS_COPYROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
CopyRow = IS_ALIGNED(width * 4, 32) ? CopyRow_NEON : CopyRow_Any_NEON;
}
#endif
#if defined(HAS_COPYROW_SME)
if (TestCpuFlag(kCpuHasSME)) {
CopyRow = CopyRow_SME;
}
#endif
#if defined(HAS_COPYROW_RVV)
if (TestCpuFlag(kCpuHasRVV)) {
CopyRow = CopyRow_RVV;
}
#endif
// Odd height will harmlessly mirror the middle row twice.
for (y = 0; y < half_height; ++y) {
ARGBMirrorRow(src_argb, row, width); // Mirror first row into a buffer
ARGBMirrorRow(src_bot, dst_argb, width); // Mirror last row into first row
CopyRow(row, dst_bot, width * 4); // Copy first mirrored row into last
src_argb += src_stride_argb;
dst_argb += dst_stride_argb;
src_bot -= src_stride_argb;
dst_bot -= dst_stride_argb;
}
free_aligned_buffer_64(row);
return 0;
}
LIBYUV_API
int ARGBRotate(const uint8_t* src_argb,
int src_stride_argb,
uint8_t* dst_argb,
int dst_stride_argb,
int width,
int height,
enum RotationMode mode) {
if (!src_argb || width <= 0 || height == 0 || !dst_argb) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
src_argb = src_argb + (height - 1) * src_stride_argb;
src_stride_argb = -src_stride_argb;
}
switch (mode) {
case kRotate0:
// copy frame
return ARGBCopy(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height);
case kRotate90:
return ARGBRotate90(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height);
case kRotate270:
return ARGBRotate270(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height);
case kRotate180:
return ARGBRotate180(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height);
default:
break;
}
return -1;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

208
3rdparty/libyuv/source/rotate_common.cc vendored Normal file
View File

@@ -0,0 +1,208 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
void TransposeWx8_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
int i;
for (i = 0; i < width; ++i) {
dst[0] = src[0 * src_stride];
dst[1] = src[1 * src_stride];
dst[2] = src[2 * src_stride];
dst[3] = src[3 * src_stride];
dst[4] = src[4 * src_stride];
dst[5] = src[5 * src_stride];
dst[6] = src[6 * src_stride];
dst[7] = src[7 * src_stride];
++src;
dst += dst_stride;
}
}
void TransposeWx16_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
TransposeWx8_C(src, src_stride, dst, dst_stride, width);
TransposeWx8_C((src + 8 * src_stride), src_stride, (dst + 8), dst_stride,
width);
}
void TransposeUVWx8_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
int i;
for (i = 0; i < width; ++i) {
dst_a[0] = src[0 * src_stride + 0];
dst_b[0] = src[0 * src_stride + 1];
dst_a[1] = src[1 * src_stride + 0];
dst_b[1] = src[1 * src_stride + 1];
dst_a[2] = src[2 * src_stride + 0];
dst_b[2] = src[2 * src_stride + 1];
dst_a[3] = src[3 * src_stride + 0];
dst_b[3] = src[3 * src_stride + 1];
dst_a[4] = src[4 * src_stride + 0];
dst_b[4] = src[4 * src_stride + 1];
dst_a[5] = src[5 * src_stride + 0];
dst_b[5] = src[5 * src_stride + 1];
dst_a[6] = src[6 * src_stride + 0];
dst_b[6] = src[6 * src_stride + 1];
dst_a[7] = src[7 * src_stride + 0];
dst_b[7] = src[7 * src_stride + 1];
src += 2;
dst_a += dst_stride_a;
dst_b += dst_stride_b;
}
}
void TransposeWxH_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height) {
int i;
for (i = 0; i < width; ++i) {
int j;
for (j = 0; j < height; ++j) {
dst[i * dst_stride + j] = src[j * src_stride + i];
}
}
}
void TransposeUVWxH_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height) {
int i;
for (i = 0; i < width * 2; i += 2) {
int j;
for (j = 0; j < height; ++j) {
dst_a[((i >> 1) * dst_stride_a) + j] = src[i + (j * src_stride)];
dst_b[((i >> 1) * dst_stride_b) + j] = src[i + (j * src_stride) + 1];
}
}
}
void TransposeWx8_16_C(const uint16_t* src,
int src_stride,
uint16_t* dst,
int dst_stride,
int width) {
int i;
for (i = 0; i < width; ++i) {
dst[0] = src[0 * src_stride];
dst[1] = src[1 * src_stride];
dst[2] = src[2 * src_stride];
dst[3] = src[3 * src_stride];
dst[4] = src[4 * src_stride];
dst[5] = src[5 * src_stride];
dst[6] = src[6 * src_stride];
dst[7] = src[7 * src_stride];
++src;
dst += dst_stride;
}
}
void TransposeWxH_16_C(const uint16_t* src,
int src_stride,
uint16_t* dst,
int dst_stride,
int width,
int height) {
int i;
for (i = 0; i < width; ++i) {
int j;
for (j = 0; j < height; ++j) {
dst[i * dst_stride + j] = src[j * src_stride + i];
}
}
}
// Transpose 32 bit values (ARGB)
void Transpose4x4_32_C(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
const uint8_t* src1 = src + src_stride;
const uint8_t* src2 = src1 + src_stride;
const uint8_t* src3 = src2 + src_stride;
uint8_t* dst1 = dst + dst_stride;
uint8_t* dst2 = dst1 + dst_stride;
uint8_t* dst3 = dst2 + dst_stride;
int i;
for (i = 0; i < width; i += 4) {
uint32_t p00 = ((uint32_t*)(src))[0];
uint32_t p10 = ((uint32_t*)(src))[1];
uint32_t p20 = ((uint32_t*)(src))[2];
uint32_t p30 = ((uint32_t*)(src))[3];
uint32_t p01 = ((uint32_t*)(src1))[0];
uint32_t p11 = ((uint32_t*)(src1))[1];
uint32_t p21 = ((uint32_t*)(src1))[2];
uint32_t p31 = ((uint32_t*)(src1))[3];
uint32_t p02 = ((uint32_t*)(src2))[0];
uint32_t p12 = ((uint32_t*)(src2))[1];
uint32_t p22 = ((uint32_t*)(src2))[2];
uint32_t p32 = ((uint32_t*)(src2))[3];
uint32_t p03 = ((uint32_t*)(src3))[0];
uint32_t p13 = ((uint32_t*)(src3))[1];
uint32_t p23 = ((uint32_t*)(src3))[2];
uint32_t p33 = ((uint32_t*)(src3))[3];
((uint32_t*)(dst))[0] = p00;
((uint32_t*)(dst))[1] = p01;
((uint32_t*)(dst))[2] = p02;
((uint32_t*)(dst))[3] = p03;
((uint32_t*)(dst1))[0] = p10;
((uint32_t*)(dst1))[1] = p11;
((uint32_t*)(dst1))[2] = p12;
((uint32_t*)(dst1))[3] = p13;
((uint32_t*)(dst2))[0] = p20;
((uint32_t*)(dst2))[1] = p21;
((uint32_t*)(dst2))[2] = p22;
((uint32_t*)(dst2))[3] = p23;
((uint32_t*)(dst3))[0] = p30;
((uint32_t*)(dst3))[1] = p31;
((uint32_t*)(dst3))[2] = p32;
((uint32_t*)(dst3))[3] = p33;
src += src_stride * 4; // advance 4 rows
src1 += src_stride * 4;
src2 += src_stride * 4;
src3 += src_stride * 4;
dst += 4 * 4; // advance 4 columns
dst1 += 4 * 4;
dst2 += 4 * 4;
dst3 += 4 * 4;
}
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

505
3rdparty/libyuv/source/rotate_gcc.cc vendored Normal file
View File

@@ -0,0 +1,505 @@
/*
* Copyright 2015 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for GCC x86 and x64.
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || defined(__i386__)) && \
!defined(LIBYUV_ENABLE_ROWWIN)
// Transpose 8x8. 32 or 64 bit, but not NaCL for 64 bit.
#if defined(HAS_TRANSPOSEWX8_SSSE3)
void TransposeWx8_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movq (%0),%%xmm0 \n"
"movq (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"movq (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"movq (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movq (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"movq (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movq (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"lea 0x8(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"sub $0x8,%2 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
}
#endif // defined(HAS_TRANSPOSEWX8_SSSE3)
// Transpose 16x8. 64 bit
#if defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
void TransposeWx8_Fast_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqu (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"movdqa %%xmm8,%%xmm9 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"palignr $0x8,%%xmm9,%%xmm9 \n"
"movdqu (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm2,%%xmm10 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm10 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movdqa %%xmm10,%%xmm11 \n"
"movdqu (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"movdqu (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm4,%%xmm12 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm12 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movdqa %%xmm12,%%xmm13 \n"
"movdqu (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movdqu (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm6,%%xmm14 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"punpckhbw %%xmm7,%%xmm14 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"movdqa %%xmm14,%%xmm15 \n"
"lea 0x10(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"punpcklwd %%xmm10,%%xmm8 \n"
"punpcklwd %%xmm11,%%xmm9 \n"
"movdqa %%xmm8,%%xmm10 \n"
"movdqa %%xmm9,%%xmm11 \n"
"palignr $0x8,%%xmm10,%%xmm10 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"punpcklwd %%xmm14,%%xmm12 \n"
"punpcklwd %%xmm15,%%xmm13 \n"
"movdqa %%xmm12,%%xmm14 \n"
"movdqa %%xmm13,%%xmm15 \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm12,%%xmm8 \n"
"movq %%xmm8,(%1) \n"
"movdqa %%xmm8,%%xmm12 \n"
"palignr $0x8,%%xmm12,%%xmm12 \n"
"movq %%xmm12,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm14,%%xmm10 \n"
"movdqa %%xmm10,%%xmm14 \n"
"movq %%xmm10,(%1) \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"punpckldq %%xmm13,%%xmm9 \n"
"movq %%xmm14,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm9,%%xmm13 \n"
"movq %%xmm9,(%1) \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movq %%xmm13,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm15,%%xmm11 \n"
"movq %%xmm11,(%1) \n"
"movdqa %%xmm11,%%xmm15 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"sub $0x10,%2 \n"
"movq %%xmm15,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7", "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14",
"xmm15");
}
#endif // defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
// Transpose UV 8x8. 64 bit.
#if defined(HAS_TRANSPOSEUVWX8_SSE2)
void TransposeUVWx8_SSE2(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%4),%%xmm1 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqa %%xmm8,%%xmm1 \n"
"movdqu (%0),%%xmm2 \n"
"movdqu (%0,%4),%%xmm3 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm8 \n"
"movdqa %%xmm8,%%xmm3 \n"
"movdqu (%0),%%xmm4 \n"
"movdqu (%0,%4),%%xmm5 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm4,%%xmm8 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm8 \n"
"movdqa %%xmm8,%%xmm5 \n"
"movdqu (%0),%%xmm6 \n"
"movdqu (%0,%4),%%xmm7 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm6,%%xmm8 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %4 \n"
"lea 0x10(%0,%4,8),%0 \n"
"punpckhbw %%xmm7,%%xmm8 \n"
"movdqa %%xmm8,%%xmm7 \n"
"neg %4 \n"
// Second round of bit swap.
"movdqa %%xmm0,%%xmm8 \n"
"movdqa %%xmm1,%%xmm9 \n"
"punpckhwd %%xmm2,%%xmm8 \n"
"punpckhwd %%xmm3,%%xmm9 \n"
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm8,%%xmm2 \n"
"movdqa %%xmm9,%%xmm3 \n"
"movdqa %%xmm4,%%xmm8 \n"
"movdqa %%xmm5,%%xmm9 \n"
"punpckhwd %%xmm6,%%xmm8 \n"
"punpckhwd %%xmm7,%%xmm9 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm8,%%xmm6 \n"
"movdqa %%xmm9,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"movdqa %%xmm0,%%xmm8 \n"
"punpckldq %%xmm4,%%xmm0 \n"
"movlpd %%xmm0,(%1) \n" // Write back U channel
"movhpd %%xmm0,(%2) \n" // Write back V channel
"punpckhdq %%xmm4,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movlpd %%xmm2,(%1) \n"
"movhpd %%xmm2,(%2) \n"
"punpckhdq %%xmm6,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm1,%%xmm8 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movlpd %%xmm1,(%1) \n"
"movhpd %%xmm1,(%2) \n"
"punpckhdq %%xmm5,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm3,%%xmm8 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movlpd %%xmm3,(%1) \n"
"movhpd %%xmm3,(%2) \n"
"punpckhdq %%xmm7,%%xmm8 \n"
"sub $0x8,%3 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst_a), // %1
"+r"(dst_b), // %2
"+r"(width) // %3
: "r"((intptr_t)(src_stride)), // %4
"r"((intptr_t)(dst_stride_a)), // %5
"r"((intptr_t)(dst_stride_b)) // %6
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7", "xmm8", "xmm9");
}
#endif // defined(HAS_TRANSPOSEUVWX8_SSE2)
#if defined(HAS_TRANSPOSE4X4_32_SSE2)
// 4 values, little endian view
// a b c d
// e f g h
// i j k l
// m n o p
// transpose 2x2
// a e b f from row 0, 1
// i m j n from row 2, 3
// c g d h from row 0, 1
// k o l p from row 2, 3
// transpose 4x4
// a e i m from row 0, 1
// b f j n from row 0, 1
// c g k o from row 2, 3
// d h l p from row 2, 3
// Transpose 32 bit values (ARGB)
void Transpose4x4_32_SSE2(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
asm volatile(
// Main loop transpose 4x4. Read a column, write a row.
"1: \n"
"movdqu (%0),%%xmm0 \n" // a b c d
"movdqu (%0,%3),%%xmm1 \n" // e f g h
"lea (%0,%3,2),%0 \n" // src += stride * 2
"movdqu (%0),%%xmm2 \n" // i j k l
"movdqu (%0,%3),%%xmm3 \n" // m n o p
"lea (%0,%3,2),%0 \n" // src += stride * 2
// Transpose 2x2
"movdqa %%xmm0,%%xmm4 \n"
"movdqa %%xmm2,%%xmm5 \n"
"movdqa %%xmm0,%%xmm6 \n"
"movdqa %%xmm2,%%xmm7 \n"
"punpckldq %%xmm1,%%xmm4 \n" // a e b f from row 0, 1
"punpckldq %%xmm3,%%xmm5 \n" // i m j n from row 2, 3
"punpckhdq %%xmm1,%%xmm6 \n" // c g d h from row 0, 1
"punpckhdq %%xmm3,%%xmm7 \n" // k o l p from row 2, 3
// Transpose 4x4
"movdqa %%xmm4,%%xmm0 \n"
"movdqa %%xmm4,%%xmm1 \n"
"movdqa %%xmm6,%%xmm2 \n"
"movdqa %%xmm6,%%xmm3 \n"
"punpcklqdq %%xmm5,%%xmm0 \n" // a e i m from row 0, 1
"punpckhqdq %%xmm5,%%xmm1 \n" // b f j n from row 0, 1
"punpcklqdq %%xmm7,%%xmm2 \n" // c g k o from row 2, 3
"punpckhqdq %%xmm7,%%xmm3 \n" // d h l p from row 2, 3
"movdqu %%xmm0,(%1) \n"
"lea 16(%1,%4),%1 \n" // dst += stride + 16
"movdqu %%xmm1,-16(%1) \n"
"movdqu %%xmm2,-16(%1,%4) \n"
"movdqu %%xmm3,-16(%1,%4,2) \n"
"sub %4,%1 \n"
"sub $0x4,%2 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+rm"(width) // %2
: "r"((ptrdiff_t)(src_stride)), // %3
"r"((ptrdiff_t)(dst_stride)) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
}
#endif // defined(HAS_TRANSPOSE4X4_32_SSE2)
#if defined(HAS_TRANSPOSE4X4_32_AVX2)
// Transpose 32 bit values (ARGB)
void Transpose4x4_32_AVX2(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
asm volatile(
// Main loop transpose 2 blocks of 4x4. Read a column, write a row.
"1: \n"
"vmovdqu (%0),%%xmm0 \n" // a b c d
"vmovdqu (%0,%3),%%xmm1 \n" // e f g h
"lea (%0,%3,2),%0 \n" // src += stride * 2
"vmovdqu (%0),%%xmm2 \n" // i j k l
"vmovdqu (%0,%3),%%xmm3 \n" // m n o p
"lea (%0,%3,2),%0 \n" // src += stride * 2
"vinserti128 $1,(%0),%%ymm0,%%ymm0 \n" // a b c d
"vinserti128 $1,(%0,%3),%%ymm1,%%ymm1 \n" // e f g h
"lea (%0,%3,2),%0 \n" // src += stride * 2
"vinserti128 $1,(%0),%%ymm2,%%ymm2 \n" // i j k l
"vinserti128 $1,(%0,%3),%%ymm3,%%ymm3 \n" // m n o p
"lea (%0,%3,2),%0 \n" // src += stride * 2
// Transpose 2x2
"vpunpckldq %%ymm1,%%ymm0,%%ymm4 \n" // a e b f from row 0, 1
"vpunpckldq %%ymm3,%%ymm2,%%ymm5 \n" // i m j n from row 2, 3
"vpunpckhdq %%ymm1,%%ymm0,%%ymm6 \n" // c g d h from row 0, 1
"vpunpckhdq %%ymm3,%%ymm2,%%ymm7 \n" // k o l p from row 2, 3
// Transpose 4x4
"vpunpcklqdq %%ymm5,%%ymm4,%%ymm0 \n" // a e i m from row 0, 1
"vpunpckhqdq %%ymm5,%%ymm4,%%ymm1 \n" // b f j n from row 0, 1
"vpunpcklqdq %%ymm7,%%ymm6,%%ymm2 \n" // c g k o from row 2, 3
"vpunpckhqdq %%ymm7,%%ymm6,%%ymm3 \n" // d h l p from row 2, 3
"vmovdqu %%ymm0,(%1) \n"
"lea 32(%1,%4),%1 \n" // dst += stride + 32
"vmovdqu %%ymm1,-32(%1) \n"
"vmovdqu %%ymm2,-32(%1,%4) \n"
"vmovdqu %%ymm3,-32(%1,%4,2) \n"
"sub %4,%1 \n"
"sub $0x8,%2 \n"
"jg 1b \n"
"vzeroupper \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+rm"(width) // %2
: "r"((ptrdiff_t)(src_stride)), // %3
"r"((ptrdiff_t)(dst_stride)) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
}
#endif // defined(HAS_TRANSPOSE4X4_32_AVX2)
#endif // defined(__x86_64__) || defined(__i386__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

233
3rdparty/libyuv/source/rotate_lsx.cc vendored Normal file
View File

@@ -0,0 +1,233 @@
/*
* Copyright 2022 The LibYuv Project Authors. All rights reserved.
*
* Copyright (c) 2022 Loongson Technology Corporation Limited
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
#if !defined(LIBYUV_DISABLE_LSX) && defined(__loongarch_sx)
#include "libyuv/loongson_intrinsics.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#define ILVLH_B(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
DUP2_ARG2(__lsx_vilvl_b, in1, in0, in3, in2, out0, out2); \
DUP2_ARG2(__lsx_vilvh_b, in1, in0, in3, in2, out1, out3); \
}
#define ILVLH_H(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
DUP2_ARG2(__lsx_vilvl_h, in1, in0, in3, in2, out0, out2); \
DUP2_ARG2(__lsx_vilvh_h, in1, in0, in3, in2, out1, out3); \
}
#define ILVLH_W(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
DUP2_ARG2(__lsx_vilvl_w, in1, in0, in3, in2, out0, out2); \
DUP2_ARG2(__lsx_vilvh_w, in1, in0, in3, in2, out1, out3); \
}
#define ILVLH_D(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
DUP2_ARG2(__lsx_vilvl_d, in1, in0, in3, in2, out0, out2); \
DUP2_ARG2(__lsx_vilvh_d, in1, in0, in3, in2, out1, out3); \
}
#define LSX_ST_4(_dst0, _dst1, _dst2, _dst3, _dst, _stride, _stride2, \
_stride3, _stride4) \
{ \
__lsx_vst(_dst0, _dst, 0); \
__lsx_vstx(_dst1, _dst, _stride); \
__lsx_vstx(_dst2, _dst, _stride2); \
__lsx_vstx(_dst3, _dst, _stride3); \
_dst += _stride4; \
}
#define LSX_ST_2(_dst0, _dst1, _dst, _stride, _stride2) \
{ \
__lsx_vst(_dst0, _dst, 0); \
__lsx_vstx(_dst1, _dst, _stride); \
_dst += _stride2; \
}
void TransposeUVWx16_C(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
TransposeUVWx8_C(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width);
TransposeUVWx8_C((src + 8 * src_stride), src_stride, (dst_a + 8),
dst_stride_a, (dst_b + 8), dst_stride_b, width);
}
void TransposeWx16_LSX(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
int x;
int len = width / 16;
uint8_t* s;
int src_stride2 = src_stride << 1;
int src_stride3 = src_stride + src_stride2;
int src_stride4 = src_stride2 << 1;
int dst_stride2 = dst_stride << 1;
int dst_stride3 = dst_stride + dst_stride2;
int dst_stride4 = dst_stride2 << 1;
__m128i src0, src1, src2, src3, dst0, dst1, dst2, dst3;
__m128i tmp0, tmp1, tmp2, tmp3;
__m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
__m128i res0, res1, res2, res3, res4, res5, res6, res7, res8, res9;
for (x = 0; x < len; x++) {
s = (uint8_t*)src;
src0 = __lsx_vld(s, 0);
src1 = __lsx_vldx(s, src_stride);
src2 = __lsx_vldx(s, src_stride2);
src3 = __lsx_vldx(s, src_stride3);
s += src_stride4;
ILVLH_B(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
ILVLH_H(tmp0, tmp2, tmp1, tmp3, reg0, reg1, reg2, reg3);
src0 = __lsx_vld(s, 0);
src1 = __lsx_vldx(s, src_stride);
src2 = __lsx_vldx(s, src_stride2);
src3 = __lsx_vldx(s, src_stride3);
s += src_stride4;
ILVLH_B(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
ILVLH_H(tmp0, tmp2, tmp1, tmp3, reg4, reg5, reg6, reg7);
ILVLH_W(reg0, reg4, reg1, reg5, res0, res1, res2, res3);
ILVLH_W(reg2, reg6, reg3, reg7, res4, res5, res6, res7);
src0 = __lsx_vld(s, 0);
src1 = __lsx_vldx(s, src_stride);
src2 = __lsx_vldx(s, src_stride2);
src3 = __lsx_vldx(s, src_stride3);
s += src_stride4;
ILVLH_B(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
ILVLH_H(tmp0, tmp2, tmp1, tmp3, reg0, reg1, reg2, reg3);
src0 = __lsx_vld(s, 0);
src1 = __lsx_vldx(s, src_stride);
src2 = __lsx_vldx(s, src_stride2);
src3 = __lsx_vldx(s, src_stride3);
s += src_stride4;
ILVLH_B(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
ILVLH_H(tmp0, tmp2, tmp1, tmp3, reg4, reg5, reg6, reg7);
res8 = __lsx_vilvl_w(reg4, reg0);
res9 = __lsx_vilvh_w(reg4, reg0);
ILVLH_D(res0, res8, res1, res9, dst0, dst1, dst2, dst3);
LSX_ST_4(dst0, dst1, dst2, dst3, dst, dst_stride, dst_stride2, dst_stride3,
dst_stride4);
res8 = __lsx_vilvl_w(reg5, reg1);
res9 = __lsx_vilvh_w(reg5, reg1);
ILVLH_D(res2, res8, res3, res9, dst0, dst1, dst2, dst3);
LSX_ST_4(dst0, dst1, dst2, dst3, dst, dst_stride, dst_stride2, dst_stride3,
dst_stride4);
res8 = __lsx_vilvl_w(reg6, reg2);
res9 = __lsx_vilvh_w(reg6, reg2);
ILVLH_D(res4, res8, res5, res9, dst0, dst1, dst2, dst3);
LSX_ST_4(dst0, dst1, dst2, dst3, dst, dst_stride, dst_stride2, dst_stride3,
dst_stride4);
res8 = __lsx_vilvl_w(reg7, reg3);
res9 = __lsx_vilvh_w(reg7, reg3);
ILVLH_D(res6, res8, res7, res9, dst0, dst1, dst2, dst3);
LSX_ST_4(dst0, dst1, dst2, dst3, dst, dst_stride, dst_stride2, dst_stride3,
dst_stride4);
src += 16;
}
}
void TransposeUVWx16_LSX(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
int x;
int len = width / 8;
uint8_t* s;
int src_stride2 = src_stride << 1;
int src_stride3 = src_stride + src_stride2;
int src_stride4 = src_stride2 << 1;
int dst_stride_a2 = dst_stride_a << 1;
int dst_stride_b2 = dst_stride_b << 1;
__m128i src0, src1, src2, src3, dst0, dst1, dst2, dst3;
__m128i tmp0, tmp1, tmp2, tmp3;
__m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
__m128i res0, res1, res2, res3, res4, res5, res6, res7, res8, res9;
for (x = 0; x < len; x++) {
s = (uint8_t*)src;
src0 = __lsx_vld(s, 0);
src1 = __lsx_vldx(s, src_stride);
src2 = __lsx_vldx(s, src_stride2);
src3 = __lsx_vldx(s, src_stride3);
s += src_stride4;
ILVLH_B(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
ILVLH_H(tmp0, tmp2, tmp1, tmp3, reg0, reg1, reg2, reg3);
src0 = __lsx_vld(s, 0);
src1 = __lsx_vldx(s, src_stride);
src2 = __lsx_vldx(s, src_stride2);
src3 = __lsx_vldx(s, src_stride3);
s += src_stride4;
ILVLH_B(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
ILVLH_H(tmp0, tmp2, tmp1, tmp3, reg4, reg5, reg6, reg7);
ILVLH_W(reg0, reg4, reg1, reg5, res0, res1, res2, res3);
ILVLH_W(reg2, reg6, reg3, reg7, res4, res5, res6, res7);
src0 = __lsx_vld(s, 0);
src1 = __lsx_vldx(s, src_stride);
src2 = __lsx_vldx(s, src_stride2);
src3 = __lsx_vldx(s, src_stride3);
s += src_stride4;
ILVLH_B(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
ILVLH_H(tmp0, tmp2, tmp1, tmp3, reg0, reg1, reg2, reg3);
src0 = __lsx_vld(s, 0);
src1 = __lsx_vldx(s, src_stride);
src2 = __lsx_vldx(s, src_stride2);
src3 = __lsx_vldx(s, src_stride3);
s += src_stride4;
ILVLH_B(src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3);
ILVLH_H(tmp0, tmp2, tmp1, tmp3, reg4, reg5, reg6, reg7);
res8 = __lsx_vilvl_w(reg4, reg0);
res9 = __lsx_vilvh_w(reg4, reg0);
ILVLH_D(res0, res8, res1, res9, dst0, dst1, dst2, dst3);
LSX_ST_2(dst0, dst2, dst_a, dst_stride_a, dst_stride_a2);
LSX_ST_2(dst1, dst3, dst_b, dst_stride_b, dst_stride_b2);
res8 = __lsx_vilvl_w(reg5, reg1);
res9 = __lsx_vilvh_w(reg5, reg1);
ILVLH_D(res2, res8, res3, res9, dst0, dst1, dst2, dst3);
LSX_ST_2(dst0, dst2, dst_a, dst_stride_a, dst_stride_a2);
LSX_ST_2(dst1, dst3, dst_b, dst_stride_b, dst_stride_b2);
res8 = __lsx_vilvl_w(reg6, reg2);
res9 = __lsx_vilvh_w(reg6, reg2);
ILVLH_D(res4, res8, res5, res9, dst0, dst1, dst2, dst3);
LSX_ST_2(dst0, dst2, dst_a, dst_stride_a, dst_stride_a2);
LSX_ST_2(dst1, dst3, dst_b, dst_stride_b, dst_stride_b2);
res8 = __lsx_vilvl_w(reg7, reg3);
res9 = __lsx_vilvh_w(reg7, reg3);
ILVLH_D(res6, res8, res7, res9, dst0, dst1, dst2, dst3);
LSX_ST_2(dst0, dst2, dst_a, dst_stride_a, dst_stride_a2);
LSX_ST_2(dst1, dst3, dst_b, dst_stride_b, dst_stride_b2);
src += 16;
}
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // !defined(LIBYUV_DISABLE_LSX) && defined(__loongarch_sx)

219
3rdparty/libyuv/source/rotate_neon.cc vendored Normal file
View File

@@ -0,0 +1,219 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#include "libyuv/basic_types.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__) && \
!defined(__aarch64__)
void TransposeWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
const uint8_t* temp;
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %[width], #8 \n"
"1: \n"
"mov %[temp], %[src] \n"
"vld1.8 {d0}, [%[temp]], %[src_stride] \n"
"vld1.8 {d1}, [%[temp]], %[src_stride] \n"
"vld1.8 {d2}, [%[temp]], %[src_stride] \n"
"vld1.8 {d3}, [%[temp]], %[src_stride] \n"
"vld1.8 {d4}, [%[temp]], %[src_stride] \n"
"vld1.8 {d5}, [%[temp]], %[src_stride] \n"
"vld1.8 {d6}, [%[temp]], %[src_stride] \n"
"vld1.8 {d7}, [%[temp]] \n"
"add %[src], #8 \n"
"vtrn.8 d1, d0 \n"
"vtrn.8 d3, d2 \n"
"vtrn.8 d5, d4 \n"
"vtrn.8 d7, d6 \n"
"subs %[width], #8 \n"
"vtrn.16 d1, d3 \n"
"vtrn.16 d0, d2 \n"
"vtrn.16 d5, d7 \n"
"vtrn.16 d4, d6 \n"
"vtrn.32 d1, d5 \n"
"vtrn.32 d0, d4 \n"
"vtrn.32 d3, d7 \n"
"vtrn.32 d2, d6 \n"
"vrev16.8 q0, q0 \n"
"vrev16.8 q1, q1 \n"
"vrev16.8 q2, q2 \n"
"vrev16.8 q3, q3 \n"
"mov %[temp], %[dst] \n"
"vst1.8 {d1}, [%[temp]], %[dst_stride] \n"
"vst1.8 {d0}, [%[temp]], %[dst_stride] \n"
"vst1.8 {d3}, [%[temp]], %[dst_stride] \n"
"vst1.8 {d2}, [%[temp]], %[dst_stride] \n"
"vst1.8 {d5}, [%[temp]], %[dst_stride] \n"
"vst1.8 {d4}, [%[temp]], %[dst_stride] \n"
"vst1.8 {d7}, [%[temp]], %[dst_stride] \n"
"vst1.8 {d6}, [%[temp]] \n"
"add %[dst], %[dst], %[dst_stride], lsl #3 \n"
"bge 1b \n"
: [temp] "=&r"(temp), // %[temp]
[src] "+r"(src), // %[src]
[dst] "+r"(dst), // %[dst]
[width] "+r"(width) // %[width]
: [src_stride] "r"(src_stride), // %[src_stride]
[dst_stride] "r"(dst_stride) // %[dst_stride]
: "memory", "cc", "q0", "q1", "q2", "q3");
}
void TransposeUVWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
const uint8_t* temp;
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %[width], #8 \n"
"1: \n"
"mov %[temp], %[src] \n"
"vld2.8 {d0, d1}, [%[temp]], %[src_stride] \n"
"vld2.8 {d2, d3}, [%[temp]], %[src_stride] \n"
"vld2.8 {d4, d5}, [%[temp]], %[src_stride] \n"
"vld2.8 {d6, d7}, [%[temp]], %[src_stride] \n"
"vld2.8 {d16, d17}, [%[temp]], %[src_stride] \n"
"vld2.8 {d18, d19}, [%[temp]], %[src_stride] \n"
"vld2.8 {d20, d21}, [%[temp]], %[src_stride] \n"
"vld2.8 {d22, d23}, [%[temp]] \n"
"add %[src], #8*2 \n"
"vtrn.8 q1, q0 \n"
"vtrn.8 q3, q2 \n"
"vtrn.8 q9, q8 \n"
"vtrn.8 q11, q10 \n"
"subs %[width], #8 \n"
"vtrn.16 q1, q3 \n"
"vtrn.16 q0, q2 \n"
"vtrn.16 q9, q11 \n"
"vtrn.16 q8, q10 \n"
"vtrn.32 q1, q9 \n"
"vtrn.32 q0, q8 \n"
"vtrn.32 q3, q11 \n"
"vtrn.32 q2, q10 \n"
"vrev16.8 q0, q0 \n"
"vrev16.8 q1, q1 \n"
"vrev16.8 q2, q2 \n"
"vrev16.8 q3, q3 \n"
"vrev16.8 q8, q8 \n"
"vrev16.8 q9, q9 \n"
"vrev16.8 q10, q10 \n"
"vrev16.8 q11, q11 \n"
"mov %[temp], %[dst_a] \n"
"vst1.8 {d2}, [%[temp]], %[dst_stride_a] \n"
"vst1.8 {d0}, [%[temp]], %[dst_stride_a] \n"
"vst1.8 {d6}, [%[temp]], %[dst_stride_a] \n"
"vst1.8 {d4}, [%[temp]], %[dst_stride_a] \n"
"vst1.8 {d18}, [%[temp]], %[dst_stride_a] \n"
"vst1.8 {d16}, [%[temp]], %[dst_stride_a] \n"
"vst1.8 {d22}, [%[temp]], %[dst_stride_a] \n"
"vst1.8 {d20}, [%[temp]] \n"
"add %[dst_a], %[dst_a], %[dst_stride_a], lsl #3 \n"
"mov %[temp], %[dst_b] \n"
"vst1.8 {d3}, [%[temp]], %[dst_stride_b] \n"
"vst1.8 {d1}, [%[temp]], %[dst_stride_b] \n"
"vst1.8 {d7}, [%[temp]], %[dst_stride_b] \n"
"vst1.8 {d5}, [%[temp]], %[dst_stride_b] \n"
"vst1.8 {d19}, [%[temp]], %[dst_stride_b] \n"
"vst1.8 {d17}, [%[temp]], %[dst_stride_b] \n"
"vst1.8 {d23}, [%[temp]], %[dst_stride_b] \n"
"vst1.8 {d21}, [%[temp]] \n"
"add %[dst_b], %[dst_b], %[dst_stride_b], lsl #3 \n"
"bge 1b \n"
: [temp] "=&r"(temp), // %[temp]
[src] "+r"(src), // %[src]
[dst_a] "+r"(dst_a), // %[dst_a]
[dst_b] "+r"(dst_b), // %[dst_b]
[width] "+r"(width) // %[width]
: [src_stride] "r"(src_stride), // %[src_stride]
[dst_stride_a] "r"(dst_stride_a), // %[dst_stride_a]
[dst_stride_b] "r"(dst_stride_b) // %[dst_stride_b]
: "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11");
}
// Transpose 32 bit values (ARGB)
void Transpose4x4_32_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
const uint8_t* src1 = src + src_stride;
const uint8_t* src2 = src1 + src_stride;
const uint8_t* src3 = src2 + src_stride;
uint8_t* dst1 = dst + dst_stride;
uint8_t* dst2 = dst1 + dst_stride;
uint8_t* dst3 = dst2 + dst_stride;
asm volatile(
// Main loop transpose 4x4. Read a column, write a row.
"1: \n"
"vld4.32 {d0[0], d2[0], d4[0], d6[0]}, [%0], %9 \n"
"vld4.32 {d0[1], d2[1], d4[1], d6[1]}, [%1], %9 \n"
"vld4.32 {d1[0], d3[0], d5[0], d7[0]}, [%2], %9 \n"
"vld4.32 {d1[1], d3[1], d5[1], d7[1]}, [%3], %9 \n"
"subs %8, %8, #4 \n" // w -= 4
"vst1.8 {q0}, [%4]! \n"
"vst1.8 {q1}, [%5]! \n"
"vst1.8 {q2}, [%6]! \n"
"vst1.8 {q3}, [%7]! \n"
"bgt 1b \n"
: "+r"(src), // %0
"+r"(src1), // %1
"+r"(src2), // %2
"+r"(src3), // %3
"+r"(dst), // %4
"+r"(dst1), // %5
"+r"(dst2), // %6
"+r"(dst3), // %7
"+r"(width) // %8
: "r"((ptrdiff_t)(src_stride * 4)) // %9
: "memory", "cc", "q0", "q1", "q2", "q3");
}
#endif // defined(__ARM_NEON__) && !defined(__aarch64__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

273
3rdparty/libyuv/source/rotate_neon64.cc vendored Normal file
View File

@@ -0,0 +1,273 @@
/*
* Copyright 2014 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#include "libyuv/basic_types.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for GCC Neon armv8 64 bit.
#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
void TransposeWx16_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
const uint8_t* src_temp;
asm volatile(
"1: \n"
"mov %[src_temp], %[src] \n"
"ld1 {v16.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v17.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v18.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v19.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v20.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v21.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v22.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v23.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v24.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v25.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v26.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v27.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v28.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v29.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v30.16b}, [%[src_temp]], %[src_stride] \n"
"ld1 {v31.16b}, [%[src_temp]], %[src_stride] \n"
"add %[src], %[src], #16 \n"
// Transpose bytes within each 2x2 block.
"trn1 v0.16b, v16.16b, v17.16b \n"
"trn2 v1.16b, v16.16b, v17.16b \n"
"trn1 v2.16b, v18.16b, v19.16b \n"
"trn2 v3.16b, v18.16b, v19.16b \n"
"trn1 v4.16b, v20.16b, v21.16b \n"
"trn2 v5.16b, v20.16b, v21.16b \n"
"trn1 v6.16b, v22.16b, v23.16b \n"
"trn2 v7.16b, v22.16b, v23.16b \n"
"trn1 v8.16b, v24.16b, v25.16b \n"
"trn2 v9.16b, v24.16b, v25.16b \n"
"trn1 v10.16b, v26.16b, v27.16b \n"
"trn2 v11.16b, v26.16b, v27.16b \n"
"trn1 v12.16b, v28.16b, v29.16b \n"
"trn2 v13.16b, v28.16b, v29.16b \n"
"trn1 v14.16b, v30.16b, v31.16b \n"
"trn2 v15.16b, v30.16b, v31.16b \n"
// Transpose 2x2-byte blocks within each 4x4 block.
"trn1 v16.8h, v0.8h, v2.8h \n"
"trn1 v17.8h, v1.8h, v3.8h \n"
"trn2 v18.8h, v0.8h, v2.8h \n"
"trn2 v19.8h, v1.8h, v3.8h \n"
"trn1 v20.8h, v4.8h, v6.8h \n"
"trn1 v21.8h, v5.8h, v7.8h \n"
"trn2 v22.8h, v4.8h, v6.8h \n"
"trn2 v23.8h, v5.8h, v7.8h \n"
"trn1 v24.8h, v8.8h, v10.8h \n"
"trn1 v25.8h, v9.8h, v11.8h \n"
"trn2 v26.8h, v8.8h, v10.8h \n"
"trn2 v27.8h, v9.8h, v11.8h \n"
"trn1 v28.8h, v12.8h, v14.8h \n"
"trn1 v29.8h, v13.8h, v15.8h \n"
"trn2 v30.8h, v12.8h, v14.8h \n"
"trn2 v31.8h, v13.8h, v15.8h \n"
"subs %w[width], %w[width], #16 \n"
// Transpose 4x4-byte blocks within each 8x8 block.
"trn1 v0.4s, v16.4s, v20.4s \n"
"trn1 v2.4s, v17.4s, v21.4s \n"
"trn1 v4.4s, v18.4s, v22.4s \n"
"trn1 v6.4s, v19.4s, v23.4s \n"
"trn2 v8.4s, v16.4s, v20.4s \n"
"trn2 v10.4s, v17.4s, v21.4s \n"
"trn2 v12.4s, v18.4s, v22.4s \n"
"trn2 v14.4s, v19.4s, v23.4s \n"
"trn1 v1.4s, v24.4s, v28.4s \n"
"trn1 v3.4s, v25.4s, v29.4s \n"
"trn1 v5.4s, v26.4s, v30.4s \n"
"trn1 v7.4s, v27.4s, v31.4s \n"
"trn2 v9.4s, v24.4s, v28.4s \n"
"trn2 v11.4s, v25.4s, v29.4s \n"
"trn2 v13.4s, v26.4s, v30.4s \n"
"trn2 v15.4s, v27.4s, v31.4s \n"
// Transpose 8x8-byte blocks and store.
"st2 {v0.d, v1.d}[0], [%[dst]], %[dst_stride] \n"
"st2 {v2.d, v3.d}[0], [%[dst]], %[dst_stride] \n"
"st2 {v4.d, v5.d}[0], [%[dst]], %[dst_stride] \n"
"st2 {v6.d, v7.d}[0], [%[dst]], %[dst_stride] \n"
"st2 {v8.d, v9.d}[0], [%[dst]], %[dst_stride] \n"
"st2 {v10.d, v11.d}[0], [%[dst]], %[dst_stride] \n"
"st2 {v12.d, v13.d}[0], [%[dst]], %[dst_stride] \n"
"st2 {v14.d, v15.d}[0], [%[dst]], %[dst_stride] \n"
"st2 {v0.d, v1.d}[1], [%[dst]], %[dst_stride] \n"
"st2 {v2.d, v3.d}[1], [%[dst]], %[dst_stride] \n"
"st2 {v4.d, v5.d}[1], [%[dst]], %[dst_stride] \n"
"st2 {v6.d, v7.d}[1], [%[dst]], %[dst_stride] \n"
"st2 {v8.d, v9.d}[1], [%[dst]], %[dst_stride] \n"
"st2 {v10.d, v11.d}[1], [%[dst]], %[dst_stride] \n"
"st2 {v12.d, v13.d}[1], [%[dst]], %[dst_stride] \n"
"st2 {v14.d, v15.d}[1], [%[dst]], %[dst_stride] \n"
"b.gt 1b \n"
: [src] "+r"(src), // %[src]
[src_temp] "=&r"(src_temp), // %[src_temp]
[dst] "+r"(dst), // %[dst]
[width] "+r"(width) // %[width]
: [src_stride] "r"((ptrdiff_t)src_stride), // %[src_stride]
[dst_stride] "r"((ptrdiff_t)dst_stride) // %[dst_stride]
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8",
"v9", "v10", "v11", "v12", "v13", "v14", "v15", "v16", "v17", "v18",
"v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28",
"v29", "v30", "v31");
}
void TransposeUVWx8_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width) {
const uint8_t* temp;
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %w[width], %w[width], #8 \n"
"1: \n"
"mov %[temp], %[src] \n"
"ld1 {v0.16b}, [%[temp]], %[src_stride] \n"
"ld1 {v1.16b}, [%[temp]], %[src_stride] \n"
"ld1 {v2.16b}, [%[temp]], %[src_stride] \n"
"ld1 {v3.16b}, [%[temp]], %[src_stride] \n"
"ld1 {v4.16b}, [%[temp]], %[src_stride] \n"
"ld1 {v5.16b}, [%[temp]], %[src_stride] \n"
"ld1 {v6.16b}, [%[temp]], %[src_stride] \n"
"ld1 {v7.16b}, [%[temp]] \n"
"add %[src], %[src], #16 \n"
"trn1 v16.16b, v0.16b, v1.16b \n"
"trn2 v17.16b, v0.16b, v1.16b \n"
"trn1 v18.16b, v2.16b, v3.16b \n"
"trn2 v19.16b, v2.16b, v3.16b \n"
"trn1 v20.16b, v4.16b, v5.16b \n"
"trn2 v21.16b, v4.16b, v5.16b \n"
"trn1 v22.16b, v6.16b, v7.16b \n"
"trn2 v23.16b, v6.16b, v7.16b \n"
"subs %w[width], %w[width], #8 \n"
"trn1 v0.8h, v16.8h, v18.8h \n"
"trn2 v1.8h, v16.8h, v18.8h \n"
"trn1 v2.8h, v20.8h, v22.8h \n"
"trn2 v3.8h, v20.8h, v22.8h \n"
"trn1 v4.8h, v17.8h, v19.8h \n"
"trn2 v5.8h, v17.8h, v19.8h \n"
"trn1 v6.8h, v21.8h, v23.8h \n"
"trn2 v7.8h, v21.8h, v23.8h \n"
"trn1 v16.4s, v0.4s, v2.4s \n"
"trn2 v17.4s, v0.4s, v2.4s \n"
"trn1 v18.4s, v1.4s, v3.4s \n"
"trn2 v19.4s, v1.4s, v3.4s \n"
"trn1 v20.4s, v4.4s, v6.4s \n"
"trn2 v21.4s, v4.4s, v6.4s \n"
"trn1 v22.4s, v5.4s, v7.4s \n"
"trn2 v23.4s, v5.4s, v7.4s \n"
"mov %[temp], %[dst_a] \n"
"st1 {v16.d}[0], [%[temp]], %[dst_stride_a] \n"
"st1 {v18.d}[0], [%[temp]], %[dst_stride_a] \n"
"st1 {v17.d}[0], [%[temp]], %[dst_stride_a] \n"
"st1 {v19.d}[0], [%[temp]], %[dst_stride_a] \n"
"st1 {v16.d}[1], [%[temp]], %[dst_stride_a] \n"
"st1 {v18.d}[1], [%[temp]], %[dst_stride_a] \n"
"st1 {v17.d}[1], [%[temp]], %[dst_stride_a] \n"
"st1 {v19.d}[1], [%[temp]] \n"
"add %[dst_a], %[dst_a], %[dst_stride_a], lsl #3 \n"
"mov %[temp], %[dst_b] \n"
"st1 {v20.d}[0], [%[temp]], %[dst_stride_b] \n"
"st1 {v22.d}[0], [%[temp]], %[dst_stride_b] \n"
"st1 {v21.d}[0], [%[temp]], %[dst_stride_b] \n"
"st1 {v23.d}[0], [%[temp]], %[dst_stride_b] \n"
"st1 {v20.d}[1], [%[temp]], %[dst_stride_b] \n"
"st1 {v22.d}[1], [%[temp]], %[dst_stride_b] \n"
"st1 {v21.d}[1], [%[temp]], %[dst_stride_b] \n"
"st1 {v23.d}[1], [%[temp]] \n"
"add %[dst_b], %[dst_b], %[dst_stride_b], lsl #3 \n"
"b.ge 1b \n"
: [temp] "=&r"(temp), // %[temp]
[src] "+r"(src), // %[src]
[dst_a] "+r"(dst_a), // %[dst_a]
[dst_b] "+r"(dst_b), // %[dst_b]
[width] "+r"(width) // %[width]
: [src_stride] "r"((ptrdiff_t)src_stride), // %[src_stride]
[dst_stride_a] "r"((ptrdiff_t)dst_stride_a), // %[dst_stride_a]
[dst_stride_b] "r"((ptrdiff_t)dst_stride_b) // %[dst_stride_b]
: "memory", "cc", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v16",
"v17", "v18", "v19", "v20", "v21", "v22", "v23", "v30", "v31");
}
// Transpose 32 bit values (ARGB)
void Transpose4x4_32_NEON(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
const uint8_t* src1 = src + src_stride;
const uint8_t* src2 = src1 + src_stride;
const uint8_t* src3 = src2 + src_stride;
uint8_t* dst1 = dst + dst_stride;
uint8_t* dst2 = dst1 + dst_stride;
uint8_t* dst3 = dst2 + dst_stride;
asm volatile(
// Main loop transpose 4x4. Read a column, write a row.
"1: \n"
"ld4 {v0.s, v1.s, v2.s, v3.s}[0], [%0], %9 \n"
"ld4 {v0.s, v1.s, v2.s, v3.s}[1], [%1], %9 \n"
"ld4 {v0.s, v1.s, v2.s, v3.s}[2], [%2], %9 \n"
"ld4 {v0.s, v1.s, v2.s, v3.s}[3], [%3], %9 \n"
"subs %w8, %w8, #4 \n" // w -= 4
"st1 {v0.4s}, [%4], 16 \n"
"st1 {v1.4s}, [%5], 16 \n"
"st1 {v2.4s}, [%6], 16 \n"
"st1 {v3.4s}, [%7], 16 \n"
"b.gt 1b \n"
: "+r"(src), // %0
"+r"(src1), // %1
"+r"(src2), // %2
"+r"(src3), // %3
"+r"(dst), // %4
"+r"(dst1), // %5
"+r"(dst2), // %6
"+r"(dst3), // %7
"+r"(width) // %8
: "r"((ptrdiff_t)(src_stride * 4)) // %9
: "memory", "cc", "v0", "v1", "v2", "v3");
}
#endif // !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

174
3rdparty/libyuv/source/rotate_sme.cc vendored Normal file
View File

@@ -0,0 +1,174 @@
/*
* Copyright 2024 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#include "libyuv/basic_types.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#if !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) && \
defined(__aarch64__)
__arm_locally_streaming __arm_new("za") void TransposeWxH_SME(
const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width,
int height) {
int vl;
asm("cntb %x0" : "=r"(vl));
do {
const uint8_t* src2 = src;
uint8_t* dst2 = dst;
// Process up to VL elements per iteration of the inner loop.
int block_height = height > vl ? vl : height;
int width2 = width;
do {
const uint8_t* src3 = src2;
// Process up to VL elements per iteration of the inner loop.
int block_width = width2 > vl ? vl : width2;
asm volatile(
"mov w12, #0 \n"
// Create a predicate to handle loading partial rows.
"whilelt p0.b, wzr, %w[block_width] \n"
// Load H <= VL rows into ZA0.
"1: \n"
"ld1b {za0h.b[w12, 0]}, p0/z, [%[src3]] \n"
"add %[src3], %[src3], %[src_stride] \n"
"add w12, w12, #1 \n"
"cmp w12, %w[block_height] \n"
"b.ne 1b \n"
// Create a predicate to handle storing partial columns.
"whilelt p0.b, wzr, %w[block_height] \n"
"mov w12, #0 \n"
// Store W <= VL columns from ZA0.
"2: \n"
"st1b {za0v.b[w12, 0]}, p0, [%[dst2]] \n"
"add %[dst2], %[dst2], %[dst_stride] \n"
"add w12, w12, #1 \n"
"cmp w12, %w[block_width] \n"
"b.ne 2b \n"
: [src3] "+r"(src3), // %[src3]
[dst2] "+r"(dst2) // %[dst2]
: [src_stride] "r"((ptrdiff_t)src_stride), // %[src_stride]
[dst_stride] "r"((ptrdiff_t)dst_stride), // %[dst_stride]
[block_width] "r"(block_width), // %[block_width]
[block_height] "r"(block_height) // %[block_height]
: "cc", "memory", "p0", "w12", "za");
src2 += vl;
width2 -= vl;
} while (width2 > 0);
src += vl * src_stride;
dst += vl;
height -= vl;
} while (height > 0);
}
__arm_locally_streaming __arm_new("za") void TransposeUVWxH_SME(
const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int width,
int height) {
int vl;
asm("cnth %x0" : "=r"(vl));
do {
const uint8_t* src2 = src;
uint8_t* dst2_a = dst_a;
uint8_t* dst2_b = dst_b;
// Process up to VL bytes per iteration of the inner loop.
int block_height = height > vl * 2 ? vl * 2 : height;
int width2 = width;
do {
const uint8_t* src3 = src2;
// Process up to VL 16-bit elements per iteration of the inner loop.
int block_width = width2 > vl ? vl : width2;
asm volatile(
"mov w12, #0 \n"
// Create a predicate to handle loading partial rows,
// %[block_width] is always a multiple of two here.
"whilelt p0.b, wzr, %w[block_width] \n"
// Load H <= VL rows into ZA0, such that U/V components exist in
// alternating columns.
"1: \n"
"ld1b {za0h.b[w12, 0]}, p0/z, [%[src]] \n"
"add %[src], %[src], %[src_stride] \n"
"add w12, w12, #1 \n"
"cmp w12, %w[block_height] \n"
"b.ne 1b \n"
// Create a predicate to handle storing partial columns.
"whilelt p0.b, wzr, %w[block_height] \n"
"mov w12, #0 \n"
// Store alternating UV data from pairs of ZA0 columns.
"2: \n"
"st1b {za0v.b[w12, 0]}, p0, [%[dst_a]] \n"
"st1b {za0v.b[w12, 1]}, p0, [%[dst_b]] \n"
"add %[dst_a], %[dst_a], %[dst_stride_a] \n"
"add %[dst_b], %[dst_b], %[dst_stride_b] \n"
"add w12, w12, #2 \n"
"cmp w12, %w[block_width] \n"
"b.ne 2b \n"
: [src] "+r"(src3), // %[src]
[dst_a] "+r"(dst2_a), // %[dst_a]
[dst_b] "+r"(dst2_b) // %[dst_b]
: [src_stride] "r"((ptrdiff_t)src_stride), // %[src_stride]
[dst_stride_a] "r"((ptrdiff_t)dst_stride_a), // %[dst_stride_a]
[dst_stride_b] "r"((ptrdiff_t)dst_stride_b), // %[dst_stride_b]
[block_width] "r"(block_width * 2), // %[block_width]
[block_height] "r"(block_height) // %[block_height]
: "cc", "memory", "p0", "w12", "za");
src2 += 2 * vl;
width2 -= vl;
} while (width2 > 0);
src += 2 * vl * src_stride;
dst_a += 2 * vl;
dst_b += 2 * vl;
height -= 2 * vl;
} while (height > 0);
}
#endif // !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) &&
// defined(__aarch64__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

253
3rdparty/libyuv/source/rotate_win.cc vendored Normal file
View File

@@ -0,0 +1,253 @@
/*
* Copyright 2013 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for 32 bit Visual C x86
#if !defined(LIBYUV_DISABLE_X86) && defined(_MSC_VER) && defined(_M_IX86) && \
(!defined(__clang__) || defined(LIBYUV_ENABLE_ROWWIN))
__declspec(naked) void TransposeWx8_SSSE3(const uint8_t* src,
int src_stride,
uint8_t* dst,
int dst_stride,
int width) {
__asm {
push edi
push esi
push ebp
mov eax, [esp + 12 + 4] // src
mov edi, [esp + 12 + 8] // src_stride
mov edx, [esp + 12 + 12] // dst
mov esi, [esp + 12 + 16] // dst_stride
mov ecx, [esp + 12 + 20] // width
// Read in the data from the source pointer.
// First round of bit swap.
align 4
convertloop:
movq xmm0, qword ptr [eax]
lea ebp, [eax + 8]
movq xmm1, qword ptr [eax + edi]
lea eax, [eax + 2 * edi]
punpcklbw xmm0, xmm1
movq xmm2, qword ptr [eax]
movdqa xmm1, xmm0
palignr xmm1, xmm1, 8
movq xmm3, qword ptr [eax + edi]
lea eax, [eax + 2 * edi]
punpcklbw xmm2, xmm3
movdqa xmm3, xmm2
movq xmm4, qword ptr [eax]
palignr xmm3, xmm3, 8
movq xmm5, qword ptr [eax + edi]
punpcklbw xmm4, xmm5
lea eax, [eax + 2 * edi]
movdqa xmm5, xmm4
movq xmm6, qword ptr [eax]
palignr xmm5, xmm5, 8
movq xmm7, qword ptr [eax + edi]
punpcklbw xmm6, xmm7
mov eax, ebp
movdqa xmm7, xmm6
palignr xmm7, xmm7, 8
// Second round of bit swap.
punpcklwd xmm0, xmm2
punpcklwd xmm1, xmm3
movdqa xmm2, xmm0
movdqa xmm3, xmm1
palignr xmm2, xmm2, 8
palignr xmm3, xmm3, 8
punpcklwd xmm4, xmm6
punpcklwd xmm5, xmm7
movdqa xmm6, xmm4
movdqa xmm7, xmm5
palignr xmm6, xmm6, 8
palignr xmm7, xmm7, 8
// Third round of bit swap.
// Write to the destination pointer.
punpckldq xmm0, xmm4
movq qword ptr [edx], xmm0
movdqa xmm4, xmm0
palignr xmm4, xmm4, 8
movq qword ptr [edx + esi], xmm4
lea edx, [edx + 2 * esi]
punpckldq xmm2, xmm6
movdqa xmm6, xmm2
palignr xmm6, xmm6, 8
movq qword ptr [edx], xmm2
punpckldq xmm1, xmm5
movq qword ptr [edx + esi], xmm6
lea edx, [edx + 2 * esi]
movdqa xmm5, xmm1
movq qword ptr [edx], xmm1
palignr xmm5, xmm5, 8
punpckldq xmm3, xmm7
movq qword ptr [edx + esi], xmm5
lea edx, [edx + 2 * esi]
movq qword ptr [edx], xmm3
movdqa xmm7, xmm3
palignr xmm7, xmm7, 8
sub ecx, 8
movq qword ptr [edx + esi], xmm7
lea edx, [edx + 2 * esi]
jg convertloop
pop ebp
pop esi
pop edi
ret
}
}
__declspec(naked) void TransposeUVWx8_SSE2(const uint8_t* src,
int src_stride,
uint8_t* dst_a,
int dst_stride_a,
uint8_t* dst_b,
int dst_stride_b,
int w) {
__asm {
push ebx
push esi
push edi
push ebp
mov eax, [esp + 16 + 4] // src
mov edi, [esp + 16 + 8] // src_stride
mov edx, [esp + 16 + 12] // dst_a
mov esi, [esp + 16 + 16] // dst_stride_a
mov ebx, [esp + 16 + 20] // dst_b
mov ebp, [esp + 16 + 24] // dst_stride_b
mov ecx, esp
sub esp, 4 + 16
and esp, ~15
mov [esp + 16], ecx
mov ecx, [ecx + 16 + 28] // w
align 4
// Read in the data from the source pointer.
// First round of bit swap.
convertloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + edi]
lea eax, [eax + 2 * edi]
movdqa xmm7, xmm0 // use xmm7 as temp register.
punpcklbw xmm0, xmm1
punpckhbw xmm7, xmm1
movdqa xmm1, xmm7
movdqu xmm2, [eax]
movdqu xmm3, [eax + edi]
lea eax, [eax + 2 * edi]
movdqa xmm7, xmm2
punpcklbw xmm2, xmm3
punpckhbw xmm7, xmm3
movdqa xmm3, xmm7
movdqu xmm4, [eax]
movdqu xmm5, [eax + edi]
lea eax, [eax + 2 * edi]
movdqa xmm7, xmm4
punpcklbw xmm4, xmm5
punpckhbw xmm7, xmm5
movdqa xmm5, xmm7
movdqu xmm6, [eax]
movdqu xmm7, [eax + edi]
lea eax, [eax + 2 * edi]
movdqu [esp], xmm5 // backup xmm5
neg edi
movdqa xmm5, xmm6 // use xmm5 as temp register.
punpcklbw xmm6, xmm7
punpckhbw xmm5, xmm7
movdqa xmm7, xmm5
lea eax, [eax + 8 * edi + 16]
neg edi
// Second round of bit swap.
movdqa xmm5, xmm0
punpcklwd xmm0, xmm2
punpckhwd xmm5, xmm2
movdqa xmm2, xmm5
movdqa xmm5, xmm1
punpcklwd xmm1, xmm3
punpckhwd xmm5, xmm3
movdqa xmm3, xmm5
movdqa xmm5, xmm4
punpcklwd xmm4, xmm6
punpckhwd xmm5, xmm6
movdqa xmm6, xmm5
movdqu xmm5, [esp] // restore xmm5
movdqu [esp], xmm6 // backup xmm6
movdqa xmm6, xmm5 // use xmm6 as temp register.
punpcklwd xmm5, xmm7
punpckhwd xmm6, xmm7
movdqa xmm7, xmm6
// Third round of bit swap.
// Write to the destination pointer.
movdqa xmm6, xmm0
punpckldq xmm0, xmm4
punpckhdq xmm6, xmm4
movdqa xmm4, xmm6
movdqu xmm6, [esp] // restore xmm6
movlpd qword ptr [edx], xmm0
movhpd qword ptr [ebx], xmm0
movlpd qword ptr [edx + esi], xmm4
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm4
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm2 // use xmm0 as the temp register.
punpckldq xmm2, xmm6
movlpd qword ptr [edx], xmm2
movhpd qword ptr [ebx], xmm2
punpckhdq xmm0, xmm6
movlpd qword ptr [edx + esi], xmm0
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm0
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm1 // use xmm0 as the temp register.
punpckldq xmm1, xmm5
movlpd qword ptr [edx], xmm1
movhpd qword ptr [ebx], xmm1
punpckhdq xmm0, xmm5
movlpd qword ptr [edx + esi], xmm0
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm0
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm3 // use xmm0 as the temp register.
punpckldq xmm3, xmm7
movlpd qword ptr [edx], xmm3
movhpd qword ptr [ebx], xmm3
punpckhdq xmm0, xmm7
sub ecx, 8
movlpd qword ptr [edx + esi], xmm0
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm0
lea ebx, [ebx + 2 * ebp]
jg convertloop
mov esp, [esp + 16]
pop ebp
pop edi
pop esi
pop ebx
ret
}
}
#endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

2659
3rdparty/libyuv/source/row_any.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

4668
3rdparty/libyuv/source/row_common.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

9742
3rdparty/libyuv/source/row_gcc.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

2343
3rdparty/libyuv/source/row_lasx.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

3030
3rdparty/libyuv/source/row_lsx.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

3983
3rdparty/libyuv/source/row_neon.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

5617
3rdparty/libyuv/source/row_neon64.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1670
3rdparty/libyuv/source/row_rvv.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1183
3rdparty/libyuv/source/row_sme.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1088
3rdparty/libyuv/source/row_sve.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

221
3rdparty/libyuv/source/row_win.cc vendored Normal file
View File

@@ -0,0 +1,221 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/convert_from_argb.h" // For ArgbConstants
// This module is for Visual C 32/64 bit
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || defined(__i386__) || \
defined(_M_X64) || defined(_M_X86)) && \
((defined(_MSC_VER) && !defined(__clang__)) || \
defined(LIBYUV_ENABLE_ROWWIN))
#include <emmintrin.h>
#include <tmmintrin.h> // For _mm_maddubs_epi16
#include <immintrin.h> // For AVX2 intrinsics
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Read 8 UV from 444
#define READYUV444 \
xmm3 = _mm_loadl_epi64((__m128i*)u_buf); \
xmm1 = _mm_loadl_epi64((__m128i*)(u_buf + offset)); \
xmm3 = _mm_unpacklo_epi8(xmm3, xmm1); \
u_buf += 8; \
xmm4 = _mm_loadl_epi64((__m128i*)y_buf); \
xmm4 = _mm_unpacklo_epi8(xmm4, xmm4); \
y_buf += 8;
// Read 8 UV from 444, With 8 Alpha.
#define READYUVA444 \
xmm3 = _mm_loadl_epi64((__m128i*)u_buf); \
xmm1 = _mm_loadl_epi64((__m128i*)(u_buf + offset)); \
xmm3 = _mm_unpacklo_epi8(xmm3, xmm1); \
u_buf += 8; \
xmm4 = _mm_loadl_epi64((__m128i*)y_buf); \
xmm4 = _mm_unpacklo_epi8(xmm4, xmm4); \
y_buf += 8; \
xmm5 = _mm_loadl_epi64((__m128i*)a_buf); \
a_buf += 8;
// Read 4 UV from 422, upsample to 8 UV.
#define READYUV422 \
xmm3 = _mm_cvtsi32_si128(*(uint32_t*)u_buf); \
xmm1 = _mm_cvtsi32_si128(*(uint32_t*)(u_buf + offset)); \
xmm3 = _mm_unpacklo_epi8(xmm3, xmm1); \
xmm3 = _mm_unpacklo_epi16(xmm3, xmm3); \
u_buf += 4; \
xmm4 = _mm_loadl_epi64((__m128i*)y_buf); \
xmm4 = _mm_unpacklo_epi8(xmm4, xmm4); \
y_buf += 8;
// Read 4 UV from 422, upsample to 8 UV. With 8 Alpha.
#define READYUVA422 \
xmm3 = _mm_cvtsi32_si128(*(uint32_t*)u_buf); \
xmm1 = _mm_cvtsi32_si128(*(uint32_t*)(u_buf + offset)); \
xmm3 = _mm_unpacklo_epi8(xmm3, xmm1); \
xmm3 = _mm_unpacklo_epi16(xmm3, xmm3); \
u_buf += 4; \
xmm4 = _mm_loadl_epi64((__m128i*)y_buf); \
xmm4 = _mm_unpacklo_epi8(xmm4, xmm4); \
y_buf += 8; \
xmm5 = _mm_loadl_epi64((__m128i*)a_buf); \
a_buf += 8;
// Convert 8 pixels: 8 UV and 8 Y.
#define YUVTORGB(yuvconstants) \
xmm3 = _mm_sub_epi8(xmm3, _mm_set1_epi8((char)0x80)); \
xmm4 = _mm_mulhi_epu16(xmm4, *(__m128i*)yuvconstants->kYToRgb); \
xmm4 = _mm_add_epi16(xmm4, *(__m128i*)yuvconstants->kYBiasToRgb); \
xmm0 = _mm_maddubs_epi16(*(__m128i*)yuvconstants->kUVToB, xmm3); \
xmm1 = _mm_maddubs_epi16(*(__m128i*)yuvconstants->kUVToG, xmm3); \
xmm2 = _mm_maddubs_epi16(*(__m128i*)yuvconstants->kUVToR, xmm3); \
xmm0 = _mm_adds_epi16(xmm4, xmm0); \
xmm1 = _mm_subs_epi16(xmm4, xmm1); \
xmm2 = _mm_adds_epi16(xmm4, xmm2); \
xmm0 = _mm_srai_epi16(xmm0, 6); \
xmm1 = _mm_srai_epi16(xmm1, 6); \
xmm2 = _mm_srai_epi16(xmm2, 6); \
xmm0 = _mm_packus_epi16(xmm0, xmm0); \
xmm1 = _mm_packus_epi16(xmm1, xmm1); \
xmm2 = _mm_packus_epi16(xmm2, xmm2);
// Store 8 ARGB values.
#define STOREARGB \
xmm0 = _mm_unpacklo_epi8(xmm0, xmm1); \
xmm2 = _mm_unpacklo_epi8(xmm2, xmm5); \
xmm1 = _mm_loadu_si128(&xmm0); \
xmm0 = _mm_unpacklo_epi16(xmm0, xmm2); \
xmm1 = _mm_unpackhi_epi16(xmm1, xmm2); \
_mm_storeu_si128((__m128i*)dst_argb, xmm0); \
_mm_storeu_si128((__m128i*)(dst_argb + 16), xmm1); \
dst_argb += 32;
#if defined(HAS_I422TOARGBROW_SSSE3)
#endif
#if defined(HAS_I422ALPHATOARGBROW_SSSE3)
#endif
#if defined(HAS_I444TOARGBROW_SSSE3)
#endif
#if defined(HAS_I444ALPHATOARGBROW_SSSE3)
#endif
#if defined(HAS_ARGBTOYROW_AVX2)
#if defined(__clang__) || defined(__GNUC__)
#define LIBYUV_TARGET_AVX2 __attribute__((target("avx2")))
#else
#define LIBYUV_TARGET_AVX2
#endif
LIBYUV_TARGET_AVX2
void ARGBToYMatrixRow_AVX2(const uint8_t* src_argb,
uint8_t* dst_y,
int width,
const struct ArgbConstants* c) {
__m256i ymm5 = _mm256_set1_epi8((char)0x80);
__m128i kRGBToY = _mm_loadu_si128((const __m128i*)c->kRGBToY);
__m256i ymm4 = _mm256_broadcastsi128_si256(kRGBToY);
__m128i kAddY = _mm_loadu_si128((const __m128i*)c->kAddY);
__m256i ymm7 = _mm256_broadcastsi128_si256(kAddY);
__m256i ymm6 = _mm256_maddubs_epi16(ymm4, ymm5);
ymm6 = _mm256_hadd_epi16(ymm6, ymm6);
ymm7 = _mm256_sub_epi16(ymm7, ymm6);
__m256i perm_mask = _mm256_setr_epi32(0, 4, 1, 5, 2, 6, 3, 7);
while (width > 0) {
__m256i ymm0 = _mm256_loadu_si256((const __m256i*)src_argb);
__m256i ymm1 = _mm256_loadu_si256((const __m256i*)(src_argb + 32));
__m256i ymm2 = _mm256_loadu_si256((const __m256i*)(src_argb + 64));
__m256i ymm3 = _mm256_loadu_si256((const __m256i*)(src_argb + 96));
src_argb += 128;
ymm0 = _mm256_sub_epi8(ymm0, ymm5);
ymm1 = _mm256_sub_epi8(ymm1, ymm5);
ymm2 = _mm256_sub_epi8(ymm2, ymm5);
ymm3 = _mm256_sub_epi8(ymm3, ymm5);
ymm0 = _mm256_maddubs_epi16(ymm4, ymm0);
ymm1 = _mm256_maddubs_epi16(ymm4, ymm1);
ymm2 = _mm256_maddubs_epi16(ymm4, ymm2);
ymm3 = _mm256_maddubs_epi16(ymm4, ymm3);
ymm0 = _mm256_hadd_epi16(ymm0, ymm1);
ymm2 = _mm256_hadd_epi16(ymm2, ymm3);
ymm0 = _mm256_add_epi16(ymm0, ymm7);
ymm2 = _mm256_add_epi16(ymm2, ymm7);
ymm0 = _mm256_srli_epi16(ymm0, 8);
ymm2 = _mm256_srli_epi16(ymm2, 8);
ymm0 = _mm256_packus_epi16(ymm0, ymm2);
ymm0 = _mm256_permutevar8x32_epi32(ymm0, perm_mask);
_mm256_storeu_si256((__m256i*)dst_y, ymm0);
dst_y += 32;
width -= 32;
}
}
LIBYUV_TARGET_AVX2
void ARGBToYRow_AVX2(const uint8_t* src_argb, uint8_t* dst_y, int width) {
ARGBToYMatrixRow_AVX2(src_argb, dst_y, width, &kArgbI601Constants);
}
LIBYUV_TARGET_AVX2
void ABGRToYRow_AVX2(const uint8_t* src_abgr, uint8_t* dst_y, int width) {
ARGBToYMatrixRow_AVX2(src_abgr, dst_y, width, &kAbgrI601Constants);
}
LIBYUV_TARGET_AVX2
void ARGBToYJRow_AVX2(const uint8_t* src_argb, uint8_t* dst_y, int width) {
ARGBToYMatrixRow_AVX2(src_argb, dst_y, width, &kArgbJPEGConstants);
}
LIBYUV_TARGET_AVX2
void ABGRToYJRow_AVX2(const uint8_t* src_abgr, uint8_t* dst_y, int width) {
ARGBToYMatrixRow_AVX2(src_abgr, dst_y, width, &kAbgrJPEGConstants);
}
LIBYUV_TARGET_AVX2
void RGBAToYJRow_AVX2(const uint8_t* src_rgba, uint8_t* dst_y, int width) {
ARGBToYMatrixRow_AVX2(src_rgba, dst_y, width, &kRgbaJPEGConstants);
}
LIBYUV_TARGET_AVX2
void RGBAToYRow_AVX2(const uint8_t* src_rgba, uint8_t* dst_y, int width) {
ARGBToYMatrixRow_AVX2(src_rgba, dst_y, width, &kRgbaI601Constants);
}
LIBYUV_TARGET_AVX2
void BGRAToYRow_AVX2(const uint8_t* src_bgra, uint8_t* dst_y, int width) {
ARGBToYMatrixRow_AVX2(src_bgra, dst_y, width, &kBgraI601Constants);
}
#endif
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // !defined(LIBYUV_DISABLE_X86) && (defined(__x86_64__) || defined(__i386__) || defined(_M_X64) || defined(_M_X86)) && ((defined(_MSC_VER) && !defined(__clang__)) || defined(LIBYUV_ENABLE_ROWWIN))

2710
3rdparty/libyuv/source/scale.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

991
3rdparty/libyuv/source/scale_any.cc vendored Normal file
View File

@@ -0,0 +1,991 @@
/*
* Copyright 2015 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <string.h> // For memset/memcpy
#include "libyuv/scale.h"
#include "libyuv/scale_row.h"
#include "libyuv/basic_types.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Fixed scale down.
// Mask may be non-power of 2, so use MOD
#define SDANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK) \
void NAMEANY(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, \
int dst_width) { \
int r = (int)((unsigned int)dst_width % (MASK + 1)); /* NOLINT */ \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r); \
}
// Fixed scale down for odd source width. Used by I420Blend subsampling.
// Since dst_width is (width + 1) / 2, this function scales one less pixel
// and copies the last pixel.
#define SDODD(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK) \
void NAMEANY(const uint8_t* src_ptr, ptrdiff_t src_stride, uint8_t* dst_ptr, \
int dst_width) { \
int r = (int)((unsigned int)(dst_width - 1) % (MASK + 1)); /* NOLINT */ \
int n = (dst_width - 1) - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r + 1); \
}
#ifdef HAS_SCALEROWDOWN2_SSSE3
SDANY(ScaleRowDown2_Any_SSSE3, ScaleRowDown2_SSSE3, ScaleRowDown2_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_SSSE3,
ScaleRowDown2Linear_SSSE3,
ScaleRowDown2Linear_C,
2,
1,
15)
SDANY(ScaleRowDown2Box_Any_SSSE3,
ScaleRowDown2Box_SSSE3,
ScaleRowDown2Box_C,
2,
1,
15)
SDODD(ScaleRowDown2Box_Odd_SSSE3,
ScaleRowDown2Box_SSSE3,
ScaleRowDown2Box_Odd_C,
2,
1,
15)
#endif
#ifdef HAS_SCALEUVROWDOWN2BOX_SSSE3
SDANY(ScaleUVRowDown2Box_Any_SSSE3,
ScaleUVRowDown2Box_SSSE3,
ScaleUVRowDown2Box_C,
2,
2,
3)
#endif
#ifdef HAS_SCALEUVROWDOWN2BOX_AVX2
SDANY(ScaleUVRowDown2Box_Any_AVX2,
ScaleUVRowDown2Box_AVX2,
ScaleUVRowDown2Box_C,
2,
2,
7)
#endif
#ifdef HAS_SCALEROWDOWN2_AVX2
SDANY(ScaleRowDown2_Any_AVX2, ScaleRowDown2_AVX2, ScaleRowDown2_C, 2, 1, 31)
SDANY(ScaleRowDown2Linear_Any_AVX2,
ScaleRowDown2Linear_AVX2,
ScaleRowDown2Linear_C,
2,
1,
31)
SDANY(ScaleRowDown2Box_Any_AVX2,
ScaleRowDown2Box_AVX2,
ScaleRowDown2Box_C,
2,
1,
31)
SDODD(ScaleRowDown2Box_Odd_AVX2,
ScaleRowDown2Box_AVX2,
ScaleRowDown2Box_Odd_C,
2,
1,
31)
#endif
#ifdef HAS_SCALEROWDOWN2_NEON
SDANY(ScaleRowDown2_Any_NEON, ScaleRowDown2_NEON, ScaleRowDown2_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_NEON,
ScaleRowDown2Linear_NEON,
ScaleRowDown2Linear_C,
2,
1,
15)
SDANY(ScaleRowDown2Box_Any_NEON,
ScaleRowDown2Box_NEON,
ScaleRowDown2Box_C,
2,
1,
15)
SDODD(ScaleRowDown2Box_Odd_NEON,
ScaleRowDown2Box_NEON,
ScaleRowDown2Box_Odd_C,
2,
1,
15)
#endif
#ifdef HAS_SCALEUVROWDOWN2_NEON
SDANY(ScaleUVRowDown2_Any_NEON,
ScaleUVRowDown2_NEON,
ScaleUVRowDown2_C,
2,
2,
7)
#endif
#ifdef HAS_SCALEUVROWDOWN2LINEAR_NEON
SDANY(ScaleUVRowDown2Linear_Any_NEON,
ScaleUVRowDown2Linear_NEON,
ScaleUVRowDown2Linear_C,
2,
2,
7)
#endif
#ifdef HAS_SCALEUVROWDOWN2BOX_NEON
SDANY(ScaleUVRowDown2Box_Any_NEON,
ScaleUVRowDown2Box_NEON,
ScaleUVRowDown2Box_C,
2,
2,
7)
#endif
#ifdef HAS_SCALEROWDOWN2_LSX
SDANY(ScaleRowDown2_Any_LSX, ScaleRowDown2_LSX, ScaleRowDown2_C, 2, 1, 31)
SDANY(ScaleRowDown2Linear_Any_LSX,
ScaleRowDown2Linear_LSX,
ScaleRowDown2Linear_C,
2,
1,
31)
SDANY(ScaleRowDown2Box_Any_LSX,
ScaleRowDown2Box_LSX,
ScaleRowDown2Box_C,
2,
1,
31)
#endif
#ifdef HAS_SCALEROWDOWN4_SSSE3
SDANY(ScaleRowDown4_Any_SSSE3, ScaleRowDown4_SSSE3, ScaleRowDown4_C, 4, 1, 7)
SDANY(ScaleRowDown4Box_Any_SSSE3,
ScaleRowDown4Box_SSSE3,
ScaleRowDown4Box_C,
4,
1,
7)
#endif
#ifdef HAS_SCALEROWDOWN4_AVX2
SDANY(ScaleRowDown4_Any_AVX2, ScaleRowDown4_AVX2, ScaleRowDown4_C, 4, 1, 15)
SDANY(ScaleRowDown4Box_Any_AVX2,
ScaleRowDown4Box_AVX2,
ScaleRowDown4Box_C,
4,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN4_NEON
SDANY(ScaleRowDown4_Any_NEON, ScaleRowDown4_NEON, ScaleRowDown4_C, 4, 1, 15)
SDANY(ScaleRowDown4Box_Any_NEON,
ScaleRowDown4Box_NEON,
ScaleRowDown4Box_C,
4,
1,
7)
#endif
#ifdef HAS_SCALEROWDOWN4_LSX
SDANY(ScaleRowDown4_Any_LSX, ScaleRowDown4_LSX, ScaleRowDown4_C, 4, 1, 15)
SDANY(ScaleRowDown4Box_Any_LSX,
ScaleRowDown4Box_LSX,
ScaleRowDown4Box_C,
4,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN34_SSSE3
SDANY(ScaleRowDown34_Any_SSSE3,
ScaleRowDown34_SSSE3,
ScaleRowDown34_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_0_Box_Any_SSSE3,
ScaleRowDown34_0_Box_SSSE3,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_1_Box_Any_SSSE3,
ScaleRowDown34_1_Box_SSSE3,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
23)
#endif
#ifdef HAS_SCALEROWDOWN34_NEON
#ifdef __aarch64__
SDANY(ScaleRowDown34_Any_NEON,
ScaleRowDown34_NEON,
ScaleRowDown34_C,
4 / 3,
1,
47)
SDANY(ScaleRowDown34_0_Box_Any_NEON,
ScaleRowDown34_0_Box_NEON,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
47)
SDANY(ScaleRowDown34_1_Box_Any_NEON,
ScaleRowDown34_1_Box_NEON,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
47)
#else
SDANY(ScaleRowDown34_Any_NEON,
ScaleRowDown34_NEON,
ScaleRowDown34_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_0_Box_Any_NEON,
ScaleRowDown34_0_Box_NEON,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_1_Box_Any_NEON,
ScaleRowDown34_1_Box_NEON,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
23)
#endif
#endif
#ifdef HAS_SCALEROWDOWN34_LSX
SDANY(ScaleRowDown34_Any_LSX,
ScaleRowDown34_LSX,
ScaleRowDown34_C,
4 / 3,
1,
47)
SDANY(ScaleRowDown34_0_Box_Any_LSX,
ScaleRowDown34_0_Box_LSX,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
47)
SDANY(ScaleRowDown34_1_Box_Any_LSX,
ScaleRowDown34_1_Box_LSX,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
47)
#endif
#ifdef HAS_SCALEROWDOWN38_SSSE3
SDANY(ScaleRowDown38_Any_SSSE3,
ScaleRowDown38_SSSE3,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_SSSE3,
ScaleRowDown38_3_Box_SSSE3,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
5)
SDANY(ScaleRowDown38_2_Box_Any_SSSE3,
ScaleRowDown38_2_Box_SSSE3,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
5)
#endif
#ifdef HAS_SCALEROWDOWN38_NEON
SDANY(ScaleRowDown38_Any_NEON,
ScaleRowDown38_NEON,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_NEON,
ScaleRowDown38_3_Box_NEON,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_2_Box_Any_NEON,
ScaleRowDown38_2_Box_NEON,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
11)
#endif
#ifdef HAS_SCALEROWDOWN38_LSX
SDANY(ScaleRowDown38_Any_LSX,
ScaleRowDown38_LSX,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_LSX,
ScaleRowDown38_3_Box_LSX,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_2_Box_Any_LSX,
ScaleRowDown38_2_Box_LSX,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
11)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_SSE2
SDANY(ScaleARGBRowDown2_Any_SSE2,
ScaleARGBRowDown2_SSE2,
ScaleARGBRowDown2_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Linear_Any_SSE2,
ScaleARGBRowDown2Linear_SSE2,
ScaleARGBRowDown2Linear_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Box_Any_SSE2,
ScaleARGBRowDown2Box_SSE2,
ScaleARGBRowDown2Box_C,
2,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_NEON
SDANY(ScaleARGBRowDown2_Any_NEON,
ScaleARGBRowDown2_NEON,
ScaleARGBRowDown2_C,
2,
4,
7)
SDANY(ScaleARGBRowDown2Linear_Any_NEON,
ScaleARGBRowDown2Linear_NEON,
ScaleARGBRowDown2Linear_C,
2,
4,
7)
SDANY(ScaleARGBRowDown2Box_Any_NEON,
ScaleARGBRowDown2Box_NEON,
ScaleARGBRowDown2Box_C,
2,
4,
7)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_LSX
SDANY(ScaleARGBRowDown2_Any_LSX,
ScaleARGBRowDown2_LSX,
ScaleARGBRowDown2_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Linear_Any_LSX,
ScaleARGBRowDown2Linear_LSX,
ScaleARGBRowDown2Linear_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Box_Any_LSX,
ScaleARGBRowDown2Box_LSX,
ScaleARGBRowDown2Box_C,
2,
4,
3)
#endif
#undef SDANY
// Scale down by even scale factor.
#define SDAANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, BPP, MASK) \
void NAMEANY(const uint8_t* src_ptr, ptrdiff_t src_stride, int src_stepx, \
uint8_t* dst_ptr, int dst_width) { \
int r = dst_width & MASK; \
int n = dst_width & ~MASK; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, src_stepx, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * src_stepx) * BPP, src_stride, src_stepx, \
dst_ptr + n * BPP, r); \
}
#ifdef HAS_SCALEARGBROWDOWNEVEN_SSE2
SDAANY(ScaleARGBRowDownEven_Any_SSE2,
ScaleARGBRowDownEven_SSE2,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_SSE2,
ScaleARGBRowDownEvenBox_SSE2,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWNEVEN_NEON
SDAANY(ScaleARGBRowDownEven_Any_NEON,
ScaleARGBRowDownEven_NEON,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_NEON,
ScaleARGBRowDownEvenBox_NEON,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWNEVEN_LSX
SDAANY(ScaleARGBRowDownEven_Any_LSX,
ScaleARGBRowDownEven_LSX,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_LSX,
ScaleARGBRowDownEvenBox_LSX,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
#ifdef HAS_SCALEUVROWDOWNEVEN_NEON
SDAANY(ScaleUVRowDownEven_Any_NEON,
ScaleUVRowDownEven_NEON,
ScaleUVRowDownEven_C,
2,
3)
#endif
#ifdef SASIMDONLY
// This also works and uses memcpy and SIMD instead of C, but is slower on ARM
// Add rows box filter scale down. Using macro from row_any
#define SAROW(NAMEANY, ANY_SIMD, SBPP, BPP, MASK) \
void NAMEANY(const uint8_t* src_ptr, uint16_t* dst_ptr, int width) { \
SIMD_ALIGNED(uint16_t dst_temp[32]); \
SIMD_ALIGNED(uint8_t src_temp[32]); \
memset(dst_temp, 0, 32 * 2); /* for msan */ \
int r = width & MASK; \
int n = width & ~MASK; \
if (n > 0) { \
ANY_SIMD(src_ptr, dst_ptr, n); \
} \
memcpy(src_temp, src_ptr + n * SBPP, r * SBPP); \
memcpy(dst_temp, dst_ptr + n * BPP, r * BPP); \
ANY_SIMD(src_temp, dst_temp, MASK + 1); \
memcpy(dst_ptr + n * BPP, dst_temp, r * BPP); \
}
#ifdef HAS_SCALEADDROW_SSE2
SAROW(ScaleAddRow_Any_SSE2, ScaleAddRow_SSE2, 1, 2, 15)
#endif
#ifdef HAS_SCALEADDROW_AVX2
SAROW(ScaleAddRow_Any_AVX2, ScaleAddRow_AVX2, 1, 2, 31)
#endif
#ifdef HAS_SCALEADDROW_NEON
SAROW(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, 1, 2, 15)
#endif
#ifdef HAS_SCALEADDROW_LSX
SAROW(ScaleAddRow_Any_LSX, ScaleAddRow_LSX, 1, 2, 15)
#endif
#undef SAANY
#else
// Add rows box filter scale down.
#define SAANY(NAMEANY, SCALEADDROW_SIMD, SCALEADDROW_C, MASK) \
void NAMEANY(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width) { \
int n = src_width & ~MASK; \
if (n > 0) { \
SCALEADDROW_SIMD(src_ptr, dst_ptr, n); \
} \
SCALEADDROW_C(src_ptr + n, dst_ptr + n, src_width & MASK); \
}
#ifdef HAS_SCALEADDROW_SSE2
SAANY(ScaleAddRow_Any_SSE2, ScaleAddRow_SSE2, ScaleAddRow_C, 15)
#endif
#ifdef HAS_SCALEADDROW_AVX2
SAANY(ScaleAddRow_Any_AVX2, ScaleAddRow_AVX2, ScaleAddRow_C, 31)
#endif
#ifdef HAS_SCALEADDROW_NEON
SAANY(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, ScaleAddRow_C, 15)
#endif
#ifdef HAS_SCALEADDROW_LSX
SAANY(ScaleAddRow_Any_LSX, ScaleAddRow_LSX, ScaleAddRow_C, 15)
#endif
#undef SAANY
#endif // SASIMDONLY
// Definition for ScaleFilterCols, ScaleARGBCols and ScaleARGBFilterCols
#define CANY(NAMEANY, TERP_SIMD, TERP_C, BPP, MASK) \
void NAMEANY(uint8_t* dst_ptr, const uint8_t* src_ptr, int dst_width, int x, \
int dx) { \
int r = dst_width & MASK; \
int n = dst_width & ~MASK; \
if (n > 0) { \
TERP_SIMD(dst_ptr, src_ptr, n, x, dx); \
} \
TERP_C(dst_ptr + n * BPP, src_ptr, r, x + n * dx, dx); \
}
#ifdef HAS_SCALEFILTERCOLS_NEON
CANY(ScaleFilterCols_Any_NEON, ScaleFilterCols_NEON, ScaleFilterCols_C, 1, 7)
#endif
#ifdef HAS_SCALEFILTERCOLS_LSX
CANY(ScaleFilterCols_Any_LSX, ScaleFilterCols_LSX, ScaleFilterCols_C, 1, 15)
#endif
#ifdef HAS_SCALEARGBCOLS_NEON
CANY(ScaleARGBCols_Any_NEON, ScaleARGBCols_NEON, ScaleARGBCols_C, 4, 7)
#endif
#ifdef HAS_SCALEARGBCOLS_LSX
CANY(ScaleARGBCols_Any_LSX, ScaleARGBCols_LSX, ScaleARGBCols_C, 4, 3)
#endif
#ifdef HAS_SCALEARGBFILTERCOLS_NEON
CANY(ScaleARGBFilterCols_Any_NEON,
ScaleARGBFilterCols_NEON,
ScaleARGBFilterCols_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBFILTERCOLS_LSX
CANY(ScaleARGBFilterCols_Any_LSX,
ScaleARGBFilterCols_LSX,
ScaleARGBFilterCols_C,
4,
7)
#endif
#undef CANY
// Scale up horizontally 2 times using linear filter.
#define SUH2LANY(NAME, SIMD, C, MASK, PTYPE) \
void NAME(const PTYPE* src_ptr, PTYPE* dst_ptr, int dst_width) { \
int work_width = (dst_width - 1) & ~1; \
int r = work_width & MASK; \
int n = work_width & ~MASK; \
dst_ptr[0] = src_ptr[0]; \
if (work_width > 0) { \
if (n != 0) { \
SIMD(src_ptr, dst_ptr + 1, n); \
} \
C(src_ptr + (n / 2), dst_ptr + n + 1, r); \
} \
dst_ptr[dst_width - 1] = src_ptr[(dst_width - 1) / 2]; \
}
// Even the C versions need to be wrapped, because boundary pixels have to
// be handled differently
SUH2LANY(ScaleRowUp2_Linear_Any_C,
ScaleRowUp2_Linear_C,
ScaleRowUp2_Linear_C,
0,
uint8_t)
SUH2LANY(ScaleRowUp2_Linear_16_Any_C,
ScaleRowUp2_Linear_16_C,
ScaleRowUp2_Linear_16_C,
0,
uint16_t)
#ifdef HAS_SCALEROWUP2_LINEAR_SSE2
SUH2LANY(ScaleRowUp2_Linear_Any_SSE2,
ScaleRowUp2_Linear_SSE2,
ScaleRowUp2_Linear_C,
15,
uint8_t)
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_SSSE3
SUH2LANY(ScaleRowUp2_Linear_Any_SSSE3,
ScaleRowUp2_Linear_SSSE3,
ScaleRowUp2_Linear_C,
15,
uint8_t)
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_12_SSSE3
SUH2LANY(ScaleRowUp2_Linear_12_Any_SSSE3,
ScaleRowUp2_Linear_12_SSSE3,
ScaleRowUp2_Linear_16_C,
15,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_16_SSE2
SUH2LANY(ScaleRowUp2_Linear_16_Any_SSE2,
ScaleRowUp2_Linear_16_SSE2,
ScaleRowUp2_Linear_16_C,
7,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_AVX2
SUH2LANY(ScaleRowUp2_Linear_Any_AVX2,
ScaleRowUp2_Linear_AVX2,
ScaleRowUp2_Linear_C,
31,
uint8_t)
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_12_AVX2
SUH2LANY(ScaleRowUp2_Linear_12_Any_AVX2,
ScaleRowUp2_Linear_12_AVX2,
ScaleRowUp2_Linear_16_C,
31,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_16_AVX2
SUH2LANY(ScaleRowUp2_Linear_16_Any_AVX2,
ScaleRowUp2_Linear_16_AVX2,
ScaleRowUp2_Linear_16_C,
15,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_NEON
#ifdef __aarch64__
SUH2LANY(ScaleRowUp2_Linear_Any_NEON,
ScaleRowUp2_Linear_NEON,
ScaleRowUp2_Linear_C,
31,
uint8_t)
#else
SUH2LANY(ScaleRowUp2_Linear_Any_NEON,
ScaleRowUp2_Linear_NEON,
ScaleRowUp2_Linear_C,
15,
uint8_t)
#endif
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_12_NEON
SUH2LANY(ScaleRowUp2_Linear_12_Any_NEON,
ScaleRowUp2_Linear_12_NEON,
ScaleRowUp2_Linear_16_C,
15,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_LINEAR_16_NEON
SUH2LANY(ScaleRowUp2_Linear_16_Any_NEON,
ScaleRowUp2_Linear_16_NEON,
ScaleRowUp2_Linear_16_C,
15,
uint16_t)
#endif
#undef SUH2LANY
// Scale up 2 times using bilinear filter.
// This function produces 2 rows at a time.
#define SU2BLANY(NAME, SIMD, C, MASK, PTYPE) \
void NAME(const PTYPE* src_ptr, ptrdiff_t src_stride, PTYPE* dst_ptr, \
ptrdiff_t dst_stride, int dst_width) { \
int work_width = (dst_width - 1) & ~1; \
int r = work_width & MASK; \
int n = work_width & ~MASK; \
const PTYPE* sa = src_ptr; \
const PTYPE* sb = src_ptr + src_stride; \
PTYPE* da = dst_ptr; \
PTYPE* db = dst_ptr + dst_stride; \
da[0] = (3 * sa[0] + sb[0] + 2) >> 2; \
db[0] = (sa[0] + 3 * sb[0] + 2) >> 2; \
if (work_width > 0) { \
if (n != 0) { \
SIMD(sa, sb - sa, da + 1, db - da, n); \
} \
C(sa + (n / 2), sb - sa, da + n + 1, db - da, r); \
} \
da[dst_width - 1] = \
(3 * sa[(dst_width - 1) / 2] + sb[(dst_width - 1) / 2] + 2) >> 2; \
db[dst_width - 1] = \
(sa[(dst_width - 1) / 2] + 3 * sb[(dst_width - 1) / 2] + 2) >> 2; \
}
SU2BLANY(ScaleRowUp2_Bilinear_Any_C,
ScaleRowUp2_Bilinear_C,
ScaleRowUp2_Bilinear_C,
0,
uint8_t)
SU2BLANY(ScaleRowUp2_Bilinear_16_Any_C,
ScaleRowUp2_Bilinear_16_C,
ScaleRowUp2_Bilinear_16_C,
0,
uint16_t)
#ifdef HAS_SCALEROWUP2_BILINEAR_SSE2
SU2BLANY(ScaleRowUp2_Bilinear_Any_SSE2,
ScaleRowUp2_Bilinear_SSE2,
ScaleRowUp2_Bilinear_C,
15,
uint8_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_12_SSSE3
SU2BLANY(ScaleRowUp2_Bilinear_12_Any_SSSE3,
ScaleRowUp2_Bilinear_12_SSSE3,
ScaleRowUp2_Bilinear_16_C,
15,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_16_SSE2
SU2BLANY(ScaleRowUp2_Bilinear_16_Any_SSE2,
ScaleRowUp2_Bilinear_16_SSE2,
ScaleRowUp2_Bilinear_16_C,
7,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_SSSE3
SU2BLANY(ScaleRowUp2_Bilinear_Any_SSSE3,
ScaleRowUp2_Bilinear_SSSE3,
ScaleRowUp2_Bilinear_C,
15,
uint8_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_AVX2
SU2BLANY(ScaleRowUp2_Bilinear_Any_AVX2,
ScaleRowUp2_Bilinear_AVX2,
ScaleRowUp2_Bilinear_C,
31,
uint8_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_12_AVX2
SU2BLANY(ScaleRowUp2_Bilinear_12_Any_AVX2,
ScaleRowUp2_Bilinear_12_AVX2,
ScaleRowUp2_Bilinear_16_C,
15,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_16_AVX2
SU2BLANY(ScaleRowUp2_Bilinear_16_Any_AVX2,
ScaleRowUp2_Bilinear_16_AVX2,
ScaleRowUp2_Bilinear_16_C,
15,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_NEON
SU2BLANY(ScaleRowUp2_Bilinear_Any_NEON,
ScaleRowUp2_Bilinear_NEON,
ScaleRowUp2_Bilinear_C,
15,
uint8_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_12_NEON
SU2BLANY(ScaleRowUp2_Bilinear_12_Any_NEON,
ScaleRowUp2_Bilinear_12_NEON,
ScaleRowUp2_Bilinear_16_C,
15,
uint16_t)
#endif
#ifdef HAS_SCALEROWUP2_BILINEAR_16_NEON
SU2BLANY(ScaleRowUp2_Bilinear_16_Any_NEON,
ScaleRowUp2_Bilinear_16_NEON,
ScaleRowUp2_Bilinear_16_C,
7,
uint16_t)
#endif
#undef SU2BLANY
// Scale bi-planar plane up horizontally 2 times using linear filter.
#define SBUH2LANY(NAME, SIMD, C, MASK, PTYPE) \
void NAME(const PTYPE* src_ptr, PTYPE* dst_ptr, int dst_width) { \
int work_width = (dst_width - 1) & ~1; \
int r = work_width & MASK; \
int n = work_width & ~MASK; \
dst_ptr[0] = src_ptr[0]; \
dst_ptr[1] = src_ptr[1]; \
if (work_width > 0) { \
if (n != 0) { \
SIMD(src_ptr, dst_ptr + 2, n); \
} \
C(src_ptr + n, dst_ptr + 2 * n + 2, r); \
} \
dst_ptr[2 * dst_width - 2] = src_ptr[((dst_width + 1) & ~1) - 2]; \
dst_ptr[2 * dst_width - 1] = src_ptr[((dst_width + 1) & ~1) - 1]; \
}
SBUH2LANY(ScaleUVRowUp2_Linear_Any_C,
ScaleUVRowUp2_Linear_C,
ScaleUVRowUp2_Linear_C,
0,
uint8_t)
SBUH2LANY(ScaleUVRowUp2_Linear_16_Any_C,
ScaleUVRowUp2_Linear_16_C,
ScaleUVRowUp2_Linear_16_C,
0,
uint16_t)
#ifdef HAS_SCALEUVROWUP2_LINEAR_SSSE3
SBUH2LANY(ScaleUVRowUp2_Linear_Any_SSSE3,
ScaleUVRowUp2_Linear_SSSE3,
ScaleUVRowUp2_Linear_C,
7,
uint8_t)
#endif
#ifdef HAS_SCALEUVROWUP2_LINEAR_AVX2
SBUH2LANY(ScaleUVRowUp2_Linear_Any_AVX2,
ScaleUVRowUp2_Linear_AVX2,
ScaleUVRowUp2_Linear_C,
15,
uint8_t)
#endif
#ifdef HAS_SCALEUVROWUP2_LINEAR_16_SSE41
SBUH2LANY(ScaleUVRowUp2_Linear_16_Any_SSE41,
ScaleUVRowUp2_Linear_16_SSE41,
ScaleUVRowUp2_Linear_16_C,
3,
uint16_t)
#endif
#ifdef HAS_SCALEUVROWUP2_LINEAR_16_AVX2
SBUH2LANY(ScaleUVRowUp2_Linear_16_Any_AVX2,
ScaleUVRowUp2_Linear_16_AVX2,
ScaleUVRowUp2_Linear_16_C,
7,
uint16_t)
#endif
#ifdef HAS_SCALEUVROWUP2_LINEAR_NEON
SBUH2LANY(ScaleUVRowUp2_Linear_Any_NEON,
ScaleUVRowUp2_Linear_NEON,
ScaleUVRowUp2_Linear_C,
15,
uint8_t)
#endif
#ifdef HAS_SCALEUVROWUP2_LINEAR_16_NEON
SBUH2LANY(ScaleUVRowUp2_Linear_16_Any_NEON,
ScaleUVRowUp2_Linear_16_NEON,
ScaleUVRowUp2_Linear_16_C,
15,
uint16_t)
#endif
#undef SBUH2LANY
// Scale bi-planar plane up 2 times using bilinear filter.
// This function produces 2 rows at a time.
#define SBU2BLANY(NAME, SIMD, C, MASK, PTYPE) \
void NAME(const PTYPE* src_ptr, ptrdiff_t src_stride, PTYPE* dst_ptr, \
ptrdiff_t dst_stride, int dst_width) { \
int work_width = (dst_width - 1) & ~1; \
int r = work_width & MASK; \
int n = work_width & ~MASK; \
const PTYPE* sa = src_ptr; \
const PTYPE* sb = src_ptr + src_stride; \
PTYPE* da = dst_ptr; \
PTYPE* db = dst_ptr + dst_stride; \
da[0] = (3 * sa[0] + sb[0] + 2) >> 2; \
db[0] = (sa[0] + 3 * sb[0] + 2) >> 2; \
da[1] = (3 * sa[1] + sb[1] + 2) >> 2; \
db[1] = (sa[1] + 3 * sb[1] + 2) >> 2; \
if (work_width > 0) { \
if (n != 0) { \
SIMD(sa, sb - sa, da + 2, db - da, n); \
} \
C(sa + n, sb - sa, da + 2 * n + 2, db - da, r); \
} \
da[2 * dst_width - 2] = (3 * sa[((dst_width + 1) & ~1) - 2] + \
sb[((dst_width + 1) & ~1) - 2] + 2) >> \
2; \
db[2 * dst_width - 2] = (sa[((dst_width + 1) & ~1) - 2] + \
3 * sb[((dst_width + 1) & ~1) - 2] + 2) >> \
2; \
da[2 * dst_width - 1] = (3 * sa[((dst_width + 1) & ~1) - 1] + \
sb[((dst_width + 1) & ~1) - 1] + 2) >> \
2; \
db[2 * dst_width - 1] = (sa[((dst_width + 1) & ~1) - 1] + \
3 * sb[((dst_width + 1) & ~1) - 1] + 2) >> \
2; \
}
SBU2BLANY(ScaleUVRowUp2_Bilinear_Any_C,
ScaleUVRowUp2_Bilinear_C,
ScaleUVRowUp2_Bilinear_C,
0,
uint8_t)
SBU2BLANY(ScaleUVRowUp2_Bilinear_16_Any_C,
ScaleUVRowUp2_Bilinear_16_C,
ScaleUVRowUp2_Bilinear_16_C,
0,
uint16_t)
#ifdef HAS_SCALEUVROWUP2_BILINEAR_SSSE3
SBU2BLANY(ScaleUVRowUp2_Bilinear_Any_SSSE3,
ScaleUVRowUp2_Bilinear_SSSE3,
ScaleUVRowUp2_Bilinear_C,
7,
uint8_t)
#endif
#ifdef HAS_SCALEUVROWUP2_BILINEAR_AVX2
SBU2BLANY(ScaleUVRowUp2_Bilinear_Any_AVX2,
ScaleUVRowUp2_Bilinear_AVX2,
ScaleUVRowUp2_Bilinear_C,
15,
uint8_t)
#endif
#ifdef HAS_SCALEUVROWUP2_BILINEAR_16_SSE41
SBU2BLANY(ScaleUVRowUp2_Bilinear_16_Any_SSE41,
ScaleUVRowUp2_Bilinear_16_SSE41,
ScaleUVRowUp2_Bilinear_16_C,
7,
uint16_t)
#endif
#ifdef HAS_SCALEUVROWUP2_BILINEAR_16_AVX2
SBU2BLANY(ScaleUVRowUp2_Bilinear_16_Any_AVX2,
ScaleUVRowUp2_Bilinear_16_AVX2,
ScaleUVRowUp2_Bilinear_16_C,
7,
uint16_t)
#endif
#ifdef HAS_SCALEUVROWUP2_BILINEAR_NEON
SBU2BLANY(ScaleUVRowUp2_Bilinear_Any_NEON,
ScaleUVRowUp2_Bilinear_NEON,
ScaleUVRowUp2_Bilinear_C,
7,
uint8_t)
#endif
#ifdef HAS_SCALEUVROWUP2_BILINEAR_16_NEON
SBU2BLANY(ScaleUVRowUp2_Bilinear_16_Any_NEON,
ScaleUVRowUp2_Bilinear_16_NEON,
ScaleUVRowUp2_Bilinear_16_C,
7,
uint16_t)
#endif
#undef SBU2BLANY
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

1158
3rdparty/libyuv/source/scale_argb.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1977
3rdparty/libyuv/source/scale_common.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

2947
3rdparty/libyuv/source/scale_gcc.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

739
3rdparty/libyuv/source/scale_lsx.cc vendored Normal file
View File

@@ -0,0 +1,739 @@
/*
* Copyright 2022 The LibYuv Project Authors. All rights reserved.
*
* Copyright (c) 2022 Loongson Technology Corporation Limited
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <assert.h>
#include "libyuv/scale_row.h"
#if !defined(LIBYUV_DISABLE_LSX) && defined(__loongarch_sx)
#include "libyuv/loongson_intrinsics.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#define LOAD_DATA(_src, _in, _out) \
{ \
int _tmp1, _tmp2, _tmp3, _tmp4; \
DUP4_ARG2(__lsx_vpickve2gr_w, _in, 0, _in, 1, _in, 2, _in, 3, _tmp1, \
_tmp2, _tmp3, _tmp4); \
_out = __lsx_vinsgr2vr_w(_out, _src[_tmp1], 0); \
_out = __lsx_vinsgr2vr_w(_out, _src[_tmp2], 1); \
_out = __lsx_vinsgr2vr_w(_out, _src[_tmp3], 2); \
_out = __lsx_vinsgr2vr_w(_out, _src[_tmp4], 3); \
}
void ScaleARGBRowDown2_LSX(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
int x;
int len = dst_width / 4;
(void)src_stride;
__m128i src0, src1, dst0;
for (x = 0; x < len; x++) {
DUP2_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src0, src1);
dst0 = __lsx_vpickod_w(src1, src0);
__lsx_vst(dst0, dst_argb, 0);
src_argb += 32;
dst_argb += 16;
}
}
void ScaleARGBRowDown2Linear_LSX(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
int x;
int len = dst_width / 4;
(void)src_stride;
__m128i src0, src1, tmp0, tmp1, dst0;
for (x = 0; x < len; x++) {
DUP2_ARG2(__lsx_vld, src_argb, 0, src_argb, 16, src0, src1);
tmp0 = __lsx_vpickev_w(src1, src0);
tmp1 = __lsx_vpickod_w(src1, src0);
dst0 = __lsx_vavgr_bu(tmp1, tmp0);
__lsx_vst(dst0, dst_argb, 0);
src_argb += 32;
dst_argb += 16;
}
}
void ScaleARGBRowDown2Box_LSX(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
int x;
int len = dst_width / 4;
const uint8_t* s = src_argb;
const uint8_t* t = src_argb + src_stride;
__m128i src0, src1, src2, src3, tmp0, tmp1, tmp2, tmp3, dst0;
__m128i reg0, reg1, reg2, reg3;
__m128i shuff = {0x0703060205010400, 0x0F0B0E0A0D090C08};
for (x = 0; x < len; x++) {
DUP2_ARG2(__lsx_vld, s, 0, s, 16, src0, src1);
DUP2_ARG2(__lsx_vld, t, 0, t, 16, src2, src3);
DUP4_ARG3(__lsx_vshuf_b, src0, src0, shuff, src1, src1, shuff, src2, src2,
shuff, src3, src3, shuff, tmp0, tmp1, tmp2, tmp3);
DUP4_ARG2(__lsx_vhaddw_hu_bu, tmp0, tmp0, tmp1, tmp1, tmp2, tmp2, tmp3,
tmp3, reg0, reg1, reg2, reg3);
DUP2_ARG2(__lsx_vsadd_hu, reg0, reg2, reg1, reg3, reg0, reg1);
dst0 = __lsx_vsrarni_b_h(reg1, reg0, 2);
__lsx_vst(dst0, dst_argb, 0);
s += 32;
t += 32;
dst_argb += 16;
}
}
void ScaleARGBRowDownEven_LSX(const uint8_t* src_argb,
ptrdiff_t src_stride,
int32_t src_stepx,
uint8_t* dst_argb,
int dst_width) {
int x;
int len = dst_width / 4;
int32_t stepx = src_stepx << 2;
(void)src_stride;
__m128i dst0, dst1, dst2, dst3;
for (x = 0; x < len; x++) {
dst0 = __lsx_vldrepl_w(src_argb, 0);
src_argb += stepx;
dst1 = __lsx_vldrepl_w(src_argb, 0);
src_argb += stepx;
dst2 = __lsx_vldrepl_w(src_argb, 0);
src_argb += stepx;
dst3 = __lsx_vldrepl_w(src_argb, 0);
src_argb += stepx;
__lsx_vstelm_w(dst0, dst_argb, 0, 0);
__lsx_vstelm_w(dst1, dst_argb, 4, 0);
__lsx_vstelm_w(dst2, dst_argb, 8, 0);
__lsx_vstelm_w(dst3, dst_argb, 12, 0);
dst_argb += 16;
}
}
void ScaleARGBRowDownEvenBox_LSX(const uint8_t* src_argb,
ptrdiff_t src_stride,
int src_stepx,
uint8_t* dst_argb,
int dst_width) {
int x;
int len = dst_width / 4;
int32_t stepx = src_stepx * 4;
const uint8_t* next_argb = src_argb + src_stride;
__m128i src0, src1, src2, src3;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
__m128i reg0, reg1, dst0;
for (x = 0; x < len; x++) {
tmp0 = __lsx_vldrepl_d(src_argb, 0);
src_argb += stepx;
tmp1 = __lsx_vldrepl_d(src_argb, 0);
src_argb += stepx;
tmp2 = __lsx_vldrepl_d(src_argb, 0);
src_argb += stepx;
tmp3 = __lsx_vldrepl_d(src_argb, 0);
src_argb += stepx;
tmp4 = __lsx_vldrepl_d(next_argb, 0);
next_argb += stepx;
tmp5 = __lsx_vldrepl_d(next_argb, 0);
next_argb += stepx;
tmp6 = __lsx_vldrepl_d(next_argb, 0);
next_argb += stepx;
tmp7 = __lsx_vldrepl_d(next_argb, 0);
next_argb += stepx;
DUP4_ARG2(__lsx_vilvl_d, tmp1, tmp0, tmp3, tmp2, tmp5, tmp4, tmp7, tmp6,
src0, src1, src2, src3);
DUP2_ARG2(__lsx_vaddwev_h_bu, src0, src2, src1, src3, tmp0, tmp2);
DUP2_ARG2(__lsx_vaddwod_h_bu, src0, src2, src1, src3, tmp1, tmp3);
DUP2_ARG2(__lsx_vpackev_w, tmp1, tmp0, tmp3, tmp2, reg0, reg1);
DUP2_ARG2(__lsx_vpackod_w, tmp1, tmp0, tmp3, tmp2, tmp4, tmp5);
DUP2_ARG2(__lsx_vadd_h, reg0, tmp4, reg1, tmp5, reg0, reg1);
dst0 = __lsx_vsrarni_b_h(reg1, reg0, 2);
dst0 = __lsx_vshuf4i_b(dst0, 0xD8);
__lsx_vst(dst0, dst_argb, 0);
dst_argb += 16;
}
}
void ScaleRowDown2_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
int len = dst_width / 32;
__m128i src0, src1, src2, src3, dst0, dst1;
(void)src_stride;
for (x = 0; x < len; x++) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_ptr, 32, src_ptr, 48,
src0, src1, src2, src3);
DUP2_ARG2(__lsx_vpickod_b, src1, src0, src3, src2, dst0, dst1);
__lsx_vst(dst0, dst, 0);
__lsx_vst(dst1, dst, 16);
src_ptr += 64;
dst += 32;
}
}
void ScaleRowDown2Linear_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
int len = dst_width / 32;
__m128i src0, src1, src2, src3;
__m128i tmp0, tmp1, tmp2, tmp3, dst0, dst1;
(void)src_stride;
for (x = 0; x < len; x++) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_ptr, 32, src_ptr, 48,
src0, src1, src2, src3);
DUP2_ARG2(__lsx_vpickev_b, src1, src0, src3, src2, tmp0, tmp2);
DUP2_ARG2(__lsx_vpickod_b, src1, src0, src3, src2, tmp1, tmp3);
DUP2_ARG2(__lsx_vavgr_bu, tmp0, tmp1, tmp2, tmp3, dst0, dst1);
__lsx_vst(dst0, dst, 0);
__lsx_vst(dst1, dst, 16);
src_ptr += 64;
dst += 32;
}
}
void ScaleRowDown2Box_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
int len = dst_width / 32;
const uint8_t* src_nex = src_ptr + src_stride;
__m128i src0, src1, src2, src3, src4, src5, src6, src7;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
__m128i dst0, dst1;
for (x = 0; x < len; x++) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_ptr, 32, src_ptr, 48,
src0, src1, src2, src3);
DUP4_ARG2(__lsx_vld, src_nex, 0, src_nex, 16, src_nex, 32, src_nex, 48,
src4, src5, src6, src7);
DUP4_ARG2(__lsx_vaddwev_h_bu, src0, src4, src1, src5, src2, src6, src3,
src7, tmp0, tmp2, tmp4, tmp6);
DUP4_ARG2(__lsx_vaddwod_h_bu, src0, src4, src1, src5, src2, src6, src3,
src7, tmp1, tmp3, tmp5, tmp7);
DUP4_ARG2(__lsx_vadd_h, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
tmp0, tmp1, tmp2, tmp3);
DUP2_ARG3(__lsx_vsrarni_b_h, tmp1, tmp0, 2, tmp3, tmp2, 2, dst0, dst1);
__lsx_vst(dst0, dst, 0);
__lsx_vst(dst1, dst, 16);
src_ptr += 64;
src_nex += 64;
dst += 32;
}
}
void ScaleRowDown4_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
int len = dst_width / 16;
__m128i src0, src1, src2, src3, tmp0, tmp1, dst0;
(void)src_stride;
for (x = 0; x < len; x++) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_ptr, 32, src_ptr, 48,
src0, src1, src2, src3);
DUP2_ARG2(__lsx_vpickev_b, src1, src0, src3, src2, tmp0, tmp1);
dst0 = __lsx_vpickod_b(tmp1, tmp0);
__lsx_vst(dst0, dst, 0);
src_ptr += 64;
dst += 16;
}
}
void ScaleRowDown4Box_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
int len = dst_width / 16;
const uint8_t* ptr1 = src_ptr + src_stride;
const uint8_t* ptr2 = ptr1 + src_stride;
const uint8_t* ptr3 = ptr2 + src_stride;
__m128i src0, src1, src2, src3, src4, src5, src6, src7;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
__m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7, dst0;
for (x = 0; x < len; x++) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_ptr, 32, src_ptr, 48,
src0, src1, src2, src3);
DUP4_ARG2(__lsx_vld, ptr1, 0, ptr1, 16, ptr1, 32, ptr1, 48, src4, src5,
src6, src7);
DUP4_ARG2(__lsx_vaddwev_h_bu, src0, src4, src1, src5, src2, src6, src3,
src7, tmp0, tmp2, tmp4, tmp6);
DUP4_ARG2(__lsx_vaddwod_h_bu, src0, src4, src1, src5, src2, src6, src3,
src7, tmp1, tmp3, tmp5, tmp7);
DUP4_ARG2(__lsx_vadd_h, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
reg0, reg1, reg2, reg3);
DUP4_ARG2(__lsx_vld, ptr2, 0, ptr2, 16, ptr2, 32, ptr2, 48, src0, src1,
src2, src3);
DUP4_ARG2(__lsx_vld, ptr3, 0, ptr3, 16, ptr3, 32, ptr3, 48, src4, src5,
src6, src7);
DUP4_ARG2(__lsx_vaddwev_h_bu, src0, src4, src1, src5, src2, src6, src3,
src7, tmp0, tmp2, tmp4, tmp6);
DUP4_ARG2(__lsx_vaddwod_h_bu, src0, src4, src1, src5, src2, src6, src3,
src7, tmp1, tmp3, tmp5, tmp7);
DUP4_ARG2(__lsx_vadd_h, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7,
reg4, reg5, reg6, reg7);
DUP4_ARG2(__lsx_vadd_h, reg0, reg4, reg1, reg5, reg2, reg6, reg3, reg7,
reg0, reg1, reg2, reg3);
DUP4_ARG2(__lsx_vhaddw_wu_hu, reg0, reg0, reg1, reg1, reg2, reg2, reg3,
reg3, reg0, reg1, reg2, reg3);
DUP2_ARG3(__lsx_vsrarni_h_w, reg1, reg0, 4, reg3, reg2, 4, tmp0, tmp1);
dst0 = __lsx_vpickev_b(tmp1, tmp0);
__lsx_vst(dst0, dst, 0);
src_ptr += 64;
ptr1 += 64;
ptr2 += 64;
ptr3 += 64;
dst += 16;
}
}
void ScaleRowDown38_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x, len;
__m128i src0, src1, tmp0;
__m128i shuff = {0x13100E0B08060300, 0x000000001E1B1816};
assert(dst_width % 3 == 0);
len = dst_width / 12;
(void)src_stride;
for (x = 0; x < len; x++) {
DUP2_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src0, src1);
tmp0 = __lsx_vshuf_b(src1, src0, shuff);
__lsx_vstelm_d(tmp0, dst, 0, 0);
__lsx_vstelm_w(tmp0, dst, 8, 2);
src_ptr += 32;
dst += 12;
}
}
void ScaleRowDown38_2_Box_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst_ptr,
int dst_width) {
int x, len;
const uint8_t* src_nex = src_ptr + src_stride;
__m128i src0, src1, src2, src3, dst0;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
__m128i reg0, reg1, reg2, reg3;
__m128i shuff = {0x0A08160604120200, 0x000000001E0E0C1A};
__m128i const_0x2AAA = __lsx_vreplgr2vr_h(0x2AAA);
__m128i const_0x4000 = __lsx_vreplgr2vr_w(0x4000);
assert((dst_width % 3 == 0) && (dst_width > 0));
len = dst_width / 12;
for (x = 0; x < len; x++) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_nex, 0, src_nex, 16, src0,
src1, src2, src3);
DUP2_ARG2(__lsx_vaddwev_h_bu, src0, src2, src1, src3, tmp0, tmp2);
DUP2_ARG2(__lsx_vaddwod_h_bu, src0, src2, src1, src3, tmp1, tmp3);
DUP2_ARG2(__lsx_vpickev_h, tmp2, tmp0, tmp3, tmp1, reg0, reg1);
DUP2_ARG2(__lsx_vpackod_h, tmp1, tmp0, tmp3, tmp2, reg2, reg3);
tmp4 = __lsx_vpickev_w(reg3, reg2);
tmp5 = __lsx_vadd_h(reg0, reg1);
tmp6 = __lsx_vadd_h(tmp5, tmp4);
tmp7 = __lsx_vmuh_h(tmp6, const_0x2AAA);
tmp0 = __lsx_vpickod_w(reg3, reg2);
tmp1 = __lsx_vhaddw_wu_hu(tmp0, tmp0);
tmp2 = __lsx_vmul_w(tmp1, const_0x4000);
dst0 = __lsx_vshuf_b(tmp2, tmp7, shuff);
__lsx_vstelm_d(dst0, dst_ptr, 0, 0);
__lsx_vstelm_w(dst0, dst_ptr, 8, 2);
src_ptr += 32;
src_nex += 32;
dst_ptr += 12;
}
}
void ScaleRowDown38_3_Box_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst_ptr,
int dst_width) {
int x, len;
const uint8_t* ptr1 = src_ptr + src_stride;
const uint8_t* ptr2 = ptr1 + src_stride;
__m128i src0, src1, src2, src3, src4, src5;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
__m128i reg0, reg1, reg2, reg3, dst0;
__m128i zero = __lsx_vldi(0);
__m128i shuff = {0x0A08160604120200, 0x000000001E0E0C1A};
__m128i const_0x1C71 = __lsx_vreplgr2vr_h(0x1C71);
__m128i const_0x2AAA = __lsx_vreplgr2vr_w(0x2AAA);
assert((dst_width % 3 == 0) && (dst_width > 0));
len = dst_width / 12;
for (x = 0; x < len; x++) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, ptr1, 0, ptr1, 16, src0, src1,
src2, src3);
DUP2_ARG2(__lsx_vld, ptr2, 0, ptr2, 16, src4, src5);
DUP2_ARG2(__lsx_vaddwev_h_bu, src0, src2, src1, src3, tmp0, tmp2);
DUP2_ARG2(__lsx_vaddwod_h_bu, src0, src2, src1, src3, tmp1, tmp3);
DUP2_ARG2(__lsx_vpackev_b, zero, src4, zero, src5, tmp4, tmp6);
DUP2_ARG2(__lsx_vpackod_b, zero, src4, zero, src5, tmp5, tmp7);
DUP4_ARG2(__lsx_vadd_h, tmp0, tmp4, tmp1, tmp5, tmp2, tmp6, tmp3, tmp7,
tmp0, tmp1, tmp2, tmp3);
DUP2_ARG2(__lsx_vpickev_h, tmp2, tmp0, tmp3, tmp1, reg0, reg1);
DUP2_ARG2(__lsx_vpackod_h, tmp1, tmp0, tmp3, tmp2, reg2, reg3);
tmp4 = __lsx_vpickev_w(reg3, reg2);
tmp5 = __lsx_vadd_h(reg0, reg1);
tmp6 = __lsx_vadd_h(tmp5, tmp4);
tmp7 = __lsx_vmuh_h(tmp6, const_0x1C71);
tmp0 = __lsx_vpickod_w(reg3, reg2);
tmp1 = __lsx_vhaddw_wu_hu(tmp0, tmp0);
tmp2 = __lsx_vmul_w(tmp1, const_0x2AAA);
dst0 = __lsx_vshuf_b(tmp2, tmp7, shuff);
__lsx_vstelm_d(dst0, dst_ptr, 0, 0);
__lsx_vstelm_w(dst0, dst_ptr, 8, 2);
src_ptr += 32;
ptr1 += 32;
ptr2 += 32;
dst_ptr += 12;
}
}
void ScaleAddRow_LSX(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width) {
int x;
int len = src_width / 16;
__m128i src0, tmp0, tmp1, dst0, dst1;
__m128i zero = __lsx_vldi(0);
assert(src_width > 0);
for (x = 0; x < len; x++) {
src0 = __lsx_vld(src_ptr, 0);
DUP2_ARG2(__lsx_vld, dst_ptr, 0, dst_ptr, 16, dst0, dst1);
tmp0 = __lsx_vilvl_b(zero, src0);
tmp1 = __lsx_vilvh_b(zero, src0);
DUP2_ARG2(__lsx_vadd_h, dst0, tmp0, dst1, tmp1, dst0, dst1);
__lsx_vst(dst0, dst_ptr, 0);
__lsx_vst(dst1, dst_ptr, 16);
src_ptr += 16;
dst_ptr += 16;
}
}
void ScaleFilterCols_LSX(uint8_t* dst_ptr,
const uint8_t* src_ptr,
int dst_width,
int x,
int dx) {
int j;
int len = dst_width / 16;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
__m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
__m128i vec0, vec1, dst0;
__m128i vec_x = __lsx_vreplgr2vr_w(x);
__m128i vec_dx = __lsx_vreplgr2vr_w(dx);
__m128i const1 = __lsx_vreplgr2vr_w(0xFFFF);
__m128i const2 = __lsx_vreplgr2vr_w(0x40);
__m128i const_tmp = {0x0000000100000000, 0x0000000300000002};
vec0 = __lsx_vmul_w(vec_dx, const_tmp);
vec1 = __lsx_vslli_w(vec_dx, 2);
vec_x = __lsx_vadd_w(vec_x, vec0);
for (j = 0; j < len; j++) {
tmp0 = __lsx_vsrai_w(vec_x, 16);
tmp4 = __lsx_vand_v(vec_x, const1);
vec_x = __lsx_vadd_w(vec_x, vec1);
tmp1 = __lsx_vsrai_w(vec_x, 16);
tmp5 = __lsx_vand_v(vec_x, const1);
vec_x = __lsx_vadd_w(vec_x, vec1);
tmp2 = __lsx_vsrai_w(vec_x, 16);
tmp6 = __lsx_vand_v(vec_x, const1);
vec_x = __lsx_vadd_w(vec_x, vec1);
tmp3 = __lsx_vsrai_w(vec_x, 16);
tmp7 = __lsx_vand_v(vec_x, const1);
vec_x = __lsx_vadd_w(vec_x, vec1);
DUP4_ARG2(__lsx_vsrai_w, tmp4, 9, tmp5, 9, tmp6, 9, tmp7, 9, tmp4, tmp5,
tmp6, tmp7);
LOAD_DATA(src_ptr, tmp0, reg0);
LOAD_DATA(src_ptr, tmp1, reg1);
LOAD_DATA(src_ptr, tmp2, reg2);
LOAD_DATA(src_ptr, tmp3, reg3);
DUP4_ARG2(__lsx_vaddi_wu, tmp0, 1, tmp1, 1, tmp2, 1, tmp3, 1, tmp0, tmp1,
tmp2, tmp3);
LOAD_DATA(src_ptr, tmp0, reg4);
LOAD_DATA(src_ptr, tmp1, reg5);
LOAD_DATA(src_ptr, tmp2, reg6);
LOAD_DATA(src_ptr, tmp3, reg7);
DUP4_ARG2(__lsx_vsub_w, reg4, reg0, reg5, reg1, reg6, reg2, reg7, reg3,
reg4, reg5, reg6, reg7);
DUP4_ARG2(__lsx_vmul_w, reg4, tmp4, reg5, tmp5, reg6, tmp6, reg7, tmp7,
reg4, reg5, reg6, reg7);
DUP4_ARG2(__lsx_vadd_w, reg4, const2, reg5, const2, reg6, const2, reg7,
const2, reg4, reg5, reg6, reg7);
DUP4_ARG2(__lsx_vsrai_w, reg4, 7, reg5, 7, reg6, 7, reg7, 7, reg4, reg5,
reg6, reg7);
DUP4_ARG2(__lsx_vadd_w, reg0, reg4, reg1, reg5, reg2, reg6, reg3, reg7,
reg0, reg1, reg2, reg3);
DUP2_ARG2(__lsx_vpickev_h, reg1, reg0, reg3, reg2, tmp0, tmp1);
dst0 = __lsx_vpickev_b(tmp1, tmp0);
__lsx_vst(dst0, dst_ptr, 0);
dst_ptr += 16;
}
}
void ScaleARGBCols_LSX(uint8_t* dst_argb,
const uint8_t* src_argb,
int dst_width,
int x,
int dx) {
const uint32_t* src = (const uint32_t*)src_argb;
uint32_t* dst = (uint32_t*)dst_argb;
int j;
int len = dst_width / 4;
__m128i tmp0, tmp1, tmp2, dst0;
__m128i vec_x = __lsx_vreplgr2vr_w(x);
__m128i vec_dx = __lsx_vreplgr2vr_w(dx);
__m128i const_tmp = {0x0000000100000000, 0x0000000300000002};
tmp0 = __lsx_vmul_w(vec_dx, const_tmp);
tmp1 = __lsx_vslli_w(vec_dx, 2);
vec_x = __lsx_vadd_w(vec_x, tmp0);
for (j = 0; j < len; j++) {
tmp2 = __lsx_vsrai_w(vec_x, 16);
vec_x = __lsx_vadd_w(vec_x, tmp1);
LOAD_DATA(src, tmp2, dst0);
__lsx_vst(dst0, dst, 0);
dst += 4;
}
}
void ScaleARGBFilterCols_LSX(uint8_t* dst_argb,
const uint8_t* src_argb,
int dst_width,
int x,
int dx) {
const uint32_t* src = (const uint32_t*)src_argb;
int j;
int len = dst_width / 8;
__m128i src0, src1, src2, src3;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
__m128i reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
__m128i vec0, vec1, dst0, dst1;
__m128i vec_x = __lsx_vreplgr2vr_w(x);
__m128i vec_dx = __lsx_vreplgr2vr_w(dx);
__m128i const_tmp = {0x0000000100000000, 0x0000000300000002};
__m128i const_7f = __lsx_vldi(0x7F);
vec0 = __lsx_vmul_w(vec_dx, const_tmp);
vec1 = __lsx_vslli_w(vec_dx, 2);
vec_x = __lsx_vadd_w(vec_x, vec0);
for (j = 0; j < len; j++) {
tmp0 = __lsx_vsrai_w(vec_x, 16);
reg0 = __lsx_vsrai_w(vec_x, 9);
vec_x = __lsx_vadd_w(vec_x, vec1);
tmp1 = __lsx_vsrai_w(vec_x, 16);
reg1 = __lsx_vsrai_w(vec_x, 9);
vec_x = __lsx_vadd_w(vec_x, vec1);
DUP2_ARG2(__lsx_vand_v, reg0, const_7f, reg1, const_7f, reg0, reg1);
DUP2_ARG2(__lsx_vshuf4i_b, reg0, 0, reg1, 0, reg0, reg1);
DUP2_ARG2(__lsx_vxor_v, reg0, const_7f, reg1, const_7f, reg2, reg3);
DUP2_ARG2(__lsx_vilvl_b, reg0, reg2, reg1, reg3, reg4, reg6);
DUP2_ARG2(__lsx_vilvh_b, reg0, reg2, reg1, reg3, reg5, reg7);
LOAD_DATA(src, tmp0, src0);
LOAD_DATA(src, tmp1, src1);
DUP2_ARG2(__lsx_vaddi_wu, tmp0, 1, tmp1, 1, tmp0, tmp1);
LOAD_DATA(src, tmp0, src2);
LOAD_DATA(src, tmp1, src3);
DUP2_ARG2(__lsx_vilvl_b, src2, src0, src3, src1, tmp4, tmp6);
DUP2_ARG2(__lsx_vilvh_b, src2, src0, src3, src1, tmp5, tmp7);
DUP4_ARG2(__lsx_vdp2_h_bu, tmp4, reg4, tmp5, reg5, tmp6, reg6, tmp7, reg7,
tmp0, tmp1, tmp2, tmp3);
DUP2_ARG3(__lsx_vsrani_b_h, tmp1, tmp0, 7, tmp3, tmp2, 7, dst0, dst1);
__lsx_vst(dst0, dst_argb, 0);
__lsx_vst(dst1, dst_argb, 16);
dst_argb += 32;
}
}
void ScaleRowDown34_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
(void)src_stride;
__m128i src0, src1, src2, src3;
__m128i dst0, dst1, dst2;
__m128i shuff0 = {0x0908070504030100, 0x141311100F0D0C0B};
__m128i shuff1 = {0x0F0D0C0B09080705, 0x1918171514131110};
__m128i shuff2 = {0x141311100F0D0C0B, 0x1F1D1C1B19181715};
assert((dst_width % 3 == 0) && (dst_width > 0));
for (x = 0; x < dst_width; x += 48) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_ptr, 32, src_ptr, 48,
src0, src1, src2, src3);
DUP2_ARG3(__lsx_vshuf_b, src1, src0, shuff0, src2, src1, shuff1, dst0,
dst1);
dst2 = __lsx_vshuf_b(src3, src2, shuff2);
__lsx_vst(dst0, dst, 0);
__lsx_vst(dst1, dst, 16);
__lsx_vst(dst2, dst, 32);
src_ptr += 64;
dst += 48;
}
}
void ScaleRowDown34_0_Box_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* d,
int dst_width) {
const uint8_t* src_nex = src_ptr + src_stride;
int x;
__m128i src0, src1, src2, src3, src4, src5, src6, src7;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8, tmp9;
__m128i tmp10, tmp11, dst0, dst1, dst2;
__m128i const0 = {0x0103030101010103, 0x0101010303010101};
__m128i const1 = {0x0301010101030301, 0x0103030101010103};
__m128i const2 = {0x0101010303010101, 0x0301010101030301};
__m128i shuff0 = {0x0504030202010100, 0x0A09090807060605};
__m128i shuff1 = {0x0F0E0E0D0D0C0B0A, 0x1514131212111110};
__m128i shuff2 = {0x0A09090807060605, 0x0F0E0E0D0D0C0B0A};
__m128i shift0 = {0x0002000200010002, 0x0001000200020001};
__m128i shift1 = {0x0002000100020002, 0x0002000200010002};
__m128i shift2 = {0x0001000200020001, 0x0002000100020002};
assert((dst_width % 3 == 0) && (dst_width > 0));
for (x = 0; x < dst_width; x += 48) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_ptr, 32, src_ptr, 48,
src0, src1, src2, src3);
DUP4_ARG2(__lsx_vld, src_nex, 0, src_nex, 16, src_nex, 32, src_nex, 48,
src4, src5, src6, src7);
DUP4_ARG3(__lsx_vshuf_b, src0, src0, shuff0, src1, src0, shuff1, src1, src1,
shuff2, src2, src2, shuff0, tmp0, tmp1, tmp2, tmp3);
DUP4_ARG3(__lsx_vshuf_b, src3, src2, shuff1, src3, src3, shuff2, src4, src4,
shuff0, src5, src4, shuff1, tmp4, tmp5, tmp6, tmp7);
DUP4_ARG3(__lsx_vshuf_b, src5, src5, shuff2, src6, src6, shuff0, src7, src6,
shuff1, src7, src7, shuff2, tmp8, tmp9, tmp10, tmp11);
DUP4_ARG2(__lsx_vdp2_h_bu, tmp0, const0, tmp1, const1, tmp2, const2, tmp3,
const0, src0, src1, src2, src3);
DUP4_ARG2(__lsx_vdp2_h_bu, tmp4, const1, tmp5, const2, tmp6, const0, tmp7,
const1, src4, src5, src6, src7);
DUP4_ARG2(__lsx_vdp2_h_bu, tmp8, const2, tmp9, const0, tmp10, const1, tmp11,
const2, tmp0, tmp1, tmp2, tmp3);
DUP4_ARG2(__lsx_vsrar_h, src0, shift0, src1, shift1, src2, shift2, src3,
shift0, src0, src1, src2, src3);
DUP4_ARG2(__lsx_vsrar_h, src4, shift1, src5, shift2, src6, shift0, src7,
shift1, src4, src5, src6, src7);
DUP4_ARG2(__lsx_vsrar_h, tmp0, shift2, tmp1, shift0, tmp2, shift1, tmp3,
shift2, tmp0, tmp1, tmp2, tmp3);
DUP4_ARG2(__lsx_vslli_h, src0, 1, src1, 1, src2, 1, src3, 1, tmp5, tmp6,
tmp7, tmp8);
DUP2_ARG2(__lsx_vslli_h, src4, 1, src5, 1, tmp9, tmp10);
DUP4_ARG2(__lsx_vadd_h, src0, tmp5, src1, tmp6, src2, tmp7, src3, tmp8,
src0, src1, src2, src3);
DUP2_ARG2(__lsx_vadd_h, src4, tmp9, src5, tmp10, src4, src5);
DUP4_ARG2(__lsx_vadd_h, src0, src6, src1, src7, src2, tmp0, src3, tmp1,
src0, src1, src2, src3);
DUP2_ARG2(__lsx_vadd_h, src4, tmp2, src5, tmp3, src4, src5);
DUP2_ARG3(__lsx_vsrarni_b_h, src1, src0, 2, src3, src2, 2, dst0, dst1);
dst2 = __lsx_vsrarni_b_h(src5, src4, 2);
__lsx_vst(dst0, d, 0);
__lsx_vst(dst1, d, 16);
__lsx_vst(dst2, d, 32);
src_ptr += 64;
src_nex += 64;
d += 48;
}
}
void ScaleRowDown34_1_Box_LSX(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* d,
int dst_width) {
const uint8_t* src_nex = src_ptr + src_stride;
int x;
__m128i src0, src1, src2, src3, src4, src5, src6, src7;
__m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8, tmp9;
__m128i tmp10, tmp11, dst0, dst1, dst2;
__m128i const0 = {0x0103030101010103, 0x0101010303010101};
__m128i const1 = {0x0301010101030301, 0x0103030101010103};
__m128i const2 = {0x0101010303010101, 0x0301010101030301};
__m128i shuff0 = {0x0504030202010100, 0x0A09090807060605};
__m128i shuff1 = {0x0F0E0E0D0D0C0B0A, 0x1514131212111110};
__m128i shuff2 = {0x0A09090807060605, 0x0F0E0E0D0D0C0B0A};
__m128i shift0 = {0x0002000200010002, 0x0001000200020001};
__m128i shift1 = {0x0002000100020002, 0x0002000200010002};
__m128i shift2 = {0x0001000200020001, 0x0002000100020002};
assert((dst_width % 3 == 0) && (dst_width > 0));
for (x = 0; x < dst_width; x += 48) {
DUP4_ARG2(__lsx_vld, src_ptr, 0, src_ptr, 16, src_ptr, 32, src_ptr, 48,
src0, src1, src2, src3);
DUP4_ARG2(__lsx_vld, src_nex, 0, src_nex, 16, src_nex, 32, src_nex, 48,
src4, src5, src6, src7);
DUP4_ARG3(__lsx_vshuf_b, src0, src0, shuff0, src1, src0, shuff1, src1, src1,
shuff2, src2, src2, shuff0, tmp0, tmp1, tmp2, tmp3);
DUP4_ARG3(__lsx_vshuf_b, src3, src2, shuff1, src3, src3, shuff2, src4, src4,
shuff0, src5, src4, shuff1, tmp4, tmp5, tmp6, tmp7);
DUP4_ARG3(__lsx_vshuf_b, src5, src5, shuff2, src6, src6, shuff0, src7, src6,
shuff1, src7, src7, shuff2, tmp8, tmp9, tmp10, tmp11);
DUP4_ARG2(__lsx_vdp2_h_bu, tmp0, const0, tmp1, const1, tmp2, const2, tmp3,
const0, src0, src1, src2, src3);
DUP4_ARG2(__lsx_vdp2_h_bu, tmp4, const1, tmp5, const2, tmp6, const0, tmp7,
const1, src4, src5, src6, src7);
DUP4_ARG2(__lsx_vdp2_h_bu, tmp8, const2, tmp9, const0, tmp10, const1, tmp11,
const2, tmp0, tmp1, tmp2, tmp3);
DUP4_ARG2(__lsx_vsrar_h, src0, shift0, src1, shift1, src2, shift2, src3,
shift0, src0, src1, src2, src3);
DUP4_ARG2(__lsx_vsrar_h, src4, shift1, src5, shift2, src6, shift0, src7,
shift1, src4, src5, src6, src7);
DUP4_ARG2(__lsx_vsrar_h, tmp0, shift2, tmp1, shift0, tmp2, shift1, tmp3,
shift2, tmp0, tmp1, tmp2, tmp3);
DUP4_ARG2(__lsx_vadd_h, src0, src6, src1, src7, src2, tmp0, src3, tmp1,
src0, src1, src2, src3);
DUP2_ARG2(__lsx_vadd_h, src4, tmp2, src5, tmp3, src4, src5);
DUP2_ARG3(__lsx_vsrarni_b_h, src1, src0, 1, src3, src2, 1, dst0, dst1);
dst2 = __lsx_vsrarni_b_h(src5, src4, 1);
__lsx_vst(dst0, d, 0);
__lsx_vst(dst1, d, 16);
__lsx_vst(dst2, d, 32);
src_ptr += 64;
src_nex += 64;
d += 48;
}
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // !defined(LIBYUV_DISABLE_LSX) && defined(__loongarch_sx)

1449
3rdparty/libyuv/source/scale_neon.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1552
3rdparty/libyuv/source/scale_neon64.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

82
3rdparty/libyuv/source/scale_rgb.cc vendored Normal file
View File

@@ -0,0 +1,82 @@
/*
* Copyright 2022 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/scale.h" /* For FilterMode */
#include <assert.h>
#include <limits.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include "libyuv/convert_argb.h"
#include "libyuv/convert_from_argb.h"
#include "libyuv/row.h"
#include "libyuv/scale_argb.h"
#include "libyuv/scale_rgb.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// Scale a 24 bit image.
// Converts to ARGB as intermediate step
LIBYUV_API
int RGBScale(const uint8_t* src_rgb,
int src_stride_rgb,
int src_width,
int src_height,
uint8_t* dst_rgb,
int dst_stride_rgb,
int dst_width,
int dst_height,
enum FilterMode filtering) {
int r;
if (!src_rgb || !dst_rgb || src_width <= 0 || src_width > INT_MAX / 4 ||
src_height == 0 || dst_width <= 0 || dst_width > INT_MAX / 4 ||
dst_height <= 0) {
return -1;
}
const int abs_src_height = (src_height < 0) ? -src_height : src_height;
const uint64_t src_argb_size = (uint64_t)src_width * abs_src_height * 4;
const uint64_t dst_argb_size = (uint64_t)dst_width * dst_height * 4;
if (src_argb_size > (UINT64_MAX - dst_argb_size)) {
return -1; // Invalid size.
}
const uint64_t argb_size = src_argb_size + dst_argb_size;
if (argb_size > SIZE_MAX) {
return -1; // Invalid size.
}
uint8_t* src_argb = (uint8_t*)malloc((size_t)argb_size);
if (!src_argb) {
return 1; // Out of memory runtime error.
}
uint8_t* dst_argb = src_argb + (size_t)src_argb_size;
r = RGB24ToARGB(src_rgb, src_stride_rgb, src_argb, src_width * 4, src_width,
src_height);
if (!r) {
r = ARGBScale(src_argb, src_width * 4, src_width, abs_src_height, dst_argb,
dst_width * 4, dst_width, dst_height, filtering);
if (!r) {
r = ARGBToRGB24(dst_argb, dst_width * 4, dst_rgb, dst_stride_rgb,
dst_width, dst_height);
}
}
free(src_argb);
return r;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

1124
3rdparty/libyuv/source/scale_rvv.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

555
3rdparty/libyuv/source/scale_sme.cc vendored Normal file
View File

@@ -0,0 +1,555 @@
/*
* Copyright 2024 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/scale_row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#if !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) && \
defined(__aarch64__)
__arm_locally_streaming void ScaleRowDown2_SME(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cntb %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"1: \n"
"ptrue p0.b \n"
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n"
"incb %[src_ptr], all, mul #2 \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"st1b {z1.b}, p0, [%[dst_ptr]] \n"
"incb %[dst_ptr] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.b, wzr, %w[dst_width] \n"
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n"
"st1b {z1.b}, p0, [%[dst_ptr]] \n"
"99: \n"
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
[dst_ptr] "+r"(dst), // %[dst_ptr]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0");
}
__arm_locally_streaming void ScaleRowDown2_16_SME(const uint16_t* src_ptr,
ptrdiff_t src_stride,
uint16_t* dst,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cnth %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"1: \n"
"ptrue p0.h \n"
"ld2h {z0.h, z1.h}, p0/z, [%[src_ptr]] \n"
"incb %[src_ptr], all, mul #2 \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"st1h {z1.h}, p0, [%[dst_ptr]] \n"
"incb %[dst_ptr] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.h, wzr, %w[dst_width] \n"
"ld2h {z0.h, z1.h}, p0/z, [%[src_ptr]] \n"
"st1h {z1.h}, p0, [%[dst_ptr]] \n"
"99: \n"
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
[dst_ptr] "+r"(dst), // %[dst_ptr]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0");
}
__arm_locally_streaming void ScaleRowDown2Linear_SME(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cntb %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"1: \n"
"ptrue p0.b \n"
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n"
"incb %[src_ptr], all, mul #2 \n"
"urhadd z0.b, p0/m, z0.b, z1.b \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"st1b {z0.b}, p0, [%[dst_ptr]] \n"
"incb %[dst_ptr] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.b, wzr, %w[dst_width] \n"
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n"
"urhadd z0.b, p0/m, z0.b, z1.b \n"
"st1b {z0.b}, p0, [%[dst_ptr]] \n"
"99: \n"
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
[dst_ptr] "+r"(dst), // %[dst_ptr]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0");
}
__arm_locally_streaming void ScaleRowDown2Linear_16_SME(const uint16_t* src_ptr,
ptrdiff_t src_stride,
uint16_t* dst,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cnth %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"1: \n"
"ptrue p0.h \n"
"ld2h {z0.h, z1.h}, p0/z, [%[src_ptr]] \n"
"incb %[src_ptr], all, mul #2 \n"
"urhadd z0.h, p0/m, z0.h, z1.h \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"st1h {z0.h}, p0, [%[dst_ptr]] \n"
"incb %[dst_ptr] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.h, wzr, %w[dst_width] \n"
"ld2h {z0.h, z1.h}, p0/z, [%[src_ptr]] \n"
"urhadd z0.h, p0/m, z0.h, z1.h \n"
"st1h {z0.h}, p0, [%[dst_ptr]] \n"
"99: \n"
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
[dst_ptr] "+r"(dst), // %[dst_ptr]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0");
}
#define SCALEROWDOWN2BOX_SVE \
"ld2b {z0.b, z1.b}, p0/z, [%[src_ptr]] \n" \
"ld2b {z2.b, z3.b}, p0/z, [%[src2_ptr]] \n" \
"incb %[src_ptr], all, mul #2 \n" \
"incb %[src2_ptr], all, mul #2 \n" \
"uaddlb z4.h, z0.b, z1.b \n" \
"uaddlt z5.h, z0.b, z1.b \n" \
"uaddlb z6.h, z2.b, z3.b \n" \
"uaddlt z7.h, z2.b, z3.b \n" \
"add z4.h, z4.h, z6.h \n" \
"add z5.h, z5.h, z7.h \n" \
"rshrnb z0.b, z4.h, #2 \n" \
"rshrnt z0.b, z5.h, #2 \n" \
"subs %w[dst_width], %w[dst_width], %w[vl] \n" \
"st1b {z0.b}, p0, [%[dst_ptr]] \n" \
"incb %[dst_ptr] \n"
__arm_locally_streaming void ScaleRowDown2Box_SME(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
const uint8_t* src2_ptr = src_ptr + src_stride;
int vl;
asm volatile(
"cntb %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"ptrue p0.b \n"
"1: \n" //
SCALEROWDOWN2BOX_SVE
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.b, wzr, %w[dst_width] \n" //
SCALEROWDOWN2BOX_SVE
"99: \n"
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
[src2_ptr] "+r"(src2_ptr), // %[src2_ptr]
[dst_ptr] "+r"(dst), // %[dst_ptr]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "p0");
}
#undef SCALEROWDOWN2BOX_SVE
#define SCALEROWDOWN2BOX_16_SVE \
"ld2h {z0.h, z1.h}, p0/z, [%[src_ptr]] \n" \
"ld2h {z2.h, z3.h}, p0/z, [%[src2_ptr]] \n" \
"incb %[src_ptr], all, mul #2 \n" \
"incb %[src2_ptr], all, mul #2 \n" \
"uaddlb z4.s, z0.h, z1.h \n" \
"uaddlt z5.s, z0.h, z1.h \n" \
"uaddlb z6.s, z2.h, z3.h \n" \
"uaddlt z7.s, z2.h, z3.h \n" \
"add z4.s, z4.s, z6.s \n" \
"add z5.s, z5.s, z7.s \n" \
"rshrnb z0.h, z4.s, #2 \n" \
"rshrnt z0.h, z5.s, #2 \n" \
"subs %w[dst_width], %w[dst_width], %w[vl] \n" \
"st1h {z0.h}, p0, [%[dst_ptr]] \n" \
"incb %[dst_ptr] \n"
__arm_locally_streaming void ScaleRowDown2Box_16_SME(const uint16_t* src_ptr,
ptrdiff_t src_stride,
uint16_t* dst,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
const uint16_t* src2_ptr = src_ptr + src_stride;
int vl;
asm volatile(
"cnth %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"ptrue p0.h \n"
"1: \n" //
SCALEROWDOWN2BOX_16_SVE
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.h, wzr, %w[dst_width] \n" //
SCALEROWDOWN2BOX_16_SVE
"99: \n"
: [src_ptr] "+r"(src_ptr), // %[src_ptr]
[src2_ptr] "+r"(src2_ptr), // %[src2_ptr]
[dst_ptr] "+r"(dst), // %[dst_ptr]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "p0");
}
#undef SCALEROWDOWN2BOX_16_SVE
__arm_locally_streaming void ScaleUVRowDown2_SME(const uint8_t* src_uv,
ptrdiff_t src_stride,
uint8_t* dst_uv,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cnth %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"1: \n"
"ptrue p0.b \n"
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n"
"incb %[src_uv], all, mul #2 \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"st1h {z1.h}, p0, [%[dst_uv]] \n"
"incb %[dst_uv] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.h, wzr, %w[dst_width] \n"
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n"
"st1h {z1.h}, p0, [%[dst_uv]] \n"
"99: \n"
: [src_uv] "+r"(src_uv), // %[src_uv]
[dst_uv] "+r"(dst_uv), // %[dst_uv]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0");
}
__arm_locally_streaming void ScaleUVRowDown2Linear_SME(const uint8_t* src_uv,
ptrdiff_t src_stride,
uint8_t* dst_uv,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cnth %x[vl] \n"
"ptrue p1.b \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"ptrue p0.h \n"
"1: \n"
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n"
"incb %[src_uv], all, mul #2 \n"
"urhadd z0.b, p1/m, z0.b, z1.b \n"
"st1h {z0.h}, p0, [%[dst_uv]] \n"
"incb %[dst_uv], all, mul #1 \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.h, wzr, %w[dst_width] \n"
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n"
"urhadd z0.b, p1/m, z0.b, z1.b \n"
"st1h {z0.h}, p0, [%[dst_uv]] \n"
"99: \n"
: [src_uv] "+r"(src_uv), // %[src_uv]
[dst_uv] "+r"(dst_uv), // %[dst_uv]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "z0", "z1", "p0", "p1");
}
#define SCALEUVROWDOWN2BOX_SVE \
"ld2h {z0.h, z1.h}, p0/z, [%[src_uv]] \n" \
"ld2h {z2.h, z3.h}, p0/z, [%[src2_uv]] \n" \
"incb %[src_uv], all, mul #2 \n" \
"incb %[src2_uv], all, mul #2 \n" \
"uaddlb z4.h, z0.b, z1.b \n" \
"uaddlt z5.h, z0.b, z1.b \n" \
"uaddlb z6.h, z2.b, z3.b \n" \
"uaddlt z7.h, z2.b, z3.b \n" \
"add z4.h, z4.h, z6.h \n" \
"add z5.h, z5.h, z7.h \n" \
"rshrnb z0.b, z4.h, #2 \n" \
"rshrnt z0.b, z5.h, #2 \n" \
"st1h {z0.h}, p0, [%[dst_uv]] \n" \
"incb %[dst_uv], all, mul #1 \n"
__arm_locally_streaming void ScaleUVRowDown2Box_SME(const uint8_t* src_uv,
ptrdiff_t src_stride,
uint8_t* dst_uv,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
const uint8_t* src2_uv = src_uv + src_stride;
int vl;
asm volatile(
"cnth %x[vl] \n"
"ptrue p1.b \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"ptrue p0.h \n"
"1: \n" //
SCALEUVROWDOWN2BOX_SVE
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.h, wzr, %w[dst_width] \n" //
SCALEUVROWDOWN2BOX_SVE
"99: \n"
: [src_uv] "+r"(src_uv), // %[src_uv]
[src2_uv] "+r"(src2_uv), // %[src2_uv]
[dst_uv] "+r"(dst_uv), // %[dst_uv]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "p0", "p1");
}
#undef SCALEUVROWDOWN2BOX_SVE
__arm_locally_streaming void ScaleARGBRowDown2_SME(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cntw %x[vl] \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"1: \n"
"ptrue p0.b \n"
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n"
"incb %[src_argb], all, mul #2 \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"st1w {z1.s}, p0, [%[dst_argb]] \n"
"incb %[dst_argb] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.s, wzr, %w[dst_width] \n"
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n"
"st1w {z1.s}, p0, [%[dst_argb]] \n"
"99: \n"
: [src_argb] "+r"(src_argb), // %[src_argb]
[dst_argb] "+r"(dst_argb), // %[dst_argb]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0");
}
__arm_locally_streaming void ScaleARGBRowDown2Linear_SME(
const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
(void)src_stride;
int vl;
asm volatile(
"cntw %x[vl] \n"
"ptrue p1.b \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"ptrue p0.s \n"
"1: \n"
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n"
"incb %[src_argb], all, mul #2 \n"
"urhadd z0.b, p1/m, z0.b, z1.b \n"
"st1w {z0.s}, p0, [%[dst_argb]] \n"
"incb %[dst_argb], all, mul #1 \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.s, wzr, %w[dst_width] \n"
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n"
"urhadd z0.b, p1/m, z0.b, z1.b \n"
"st1w {z0.s}, p0, [%[dst_argb]] \n"
"99: \n"
: [src_argb] "+r"(src_argb), // %[src_argb]
[dst_argb] "+r"(dst_argb), // %[dst_argb]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "p0", "p1");
}
#define SCALEARGBROWDOWN2BOX_SVE \
"ld2w {z0.s, z1.s}, p0/z, [%[src_argb]] \n" \
"ld2w {z2.s, z3.s}, p0/z, [%[src2_argb]] \n" \
"incb %[src_argb], all, mul #2 \n" \
"incb %[src2_argb], all, mul #2 \n" \
"uaddlb z4.h, z0.b, z1.b \n" \
"uaddlt z5.h, z0.b, z1.b \n" \
"uaddlb z6.h, z2.b, z3.b \n" \
"uaddlt z7.h, z2.b, z3.b \n" \
"add z4.h, z4.h, z6.h \n" \
"add z5.h, z5.h, z7.h \n" \
"rshrnb z0.b, z4.h, #2 \n" \
"rshrnt z0.b, z5.h, #2 \n" \
"st1w {z0.s}, p0, [%[dst_argb]] \n" \
"incb %[dst_argb], all, mul #1 \n"
__arm_locally_streaming void ScaleARGBRowDown2Box_SME(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
// Streaming-SVE only, no use of ZA tile.
const uint8_t* src2_argb = src_argb + src_stride;
int vl;
asm volatile(
"cntw %x[vl] \n"
"ptrue p1.b \n"
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.lt 2f \n"
"ptrue p0.s \n"
"1: \n" //
SCALEARGBROWDOWN2BOX_SVE
"subs %w[dst_width], %w[dst_width], %w[vl] \n"
"b.ge 1b \n"
"2: \n"
"adds %w[dst_width], %w[dst_width], %w[vl] \n"
"b.eq 99f \n"
"whilelt p0.s, wzr, %w[dst_width] \n" //
SCALEARGBROWDOWN2BOX_SVE
"99: \n"
: [src_argb] "+r"(src_argb), // %[src_argb]
[src2_argb] "+r"(src2_argb), // %[src2_argb]
[dst_argb] "+r"(dst_argb), // %[dst_argb]
[dst_width] "+r"(dst_width), // %[dst_width]
[vl] "=r"(vl) // %[vl]
:
: "memory", "cc", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "p0",
"p1");
}
#endif // !defined(LIBYUV_DISABLE_SME) && defined(CLANG_HAS_SME) &&
// defined(__aarch64__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

1159
3rdparty/libyuv/source/scale_uv.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1392
3rdparty/libyuv/source/scale_win.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

35
3rdparty/libyuv/source/test.sh vendored Normal file
View File

@@ -0,0 +1,35 @@
#!/bin/bash
set -x
function runbenchmark1 {
perf record /google/src/cloud/fbarchard/clean/google3/blaze-bin/third_party/libyuv/libyuv_test --gunit_filter=*$1 --libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=1000 --libyuv_flags=-1 --libyuv_cpu_info=-1
perf report | grep AVX
}
runbenchmark1 ABGRToI420
runbenchmark1 Android420ToI420
runbenchmark1 ARGBToI420
runbenchmark1 Convert16To8Plane
runbenchmark1 ConvertToARGB
runbenchmark1 ConvertToI420
runbenchmark1 CopyPlane
runbenchmark1 H010ToAB30
runbenchmark1 H010ToAR30
runbenchmark1 HalfFloatPlane
runbenchmark1 I010ToAB30
runbenchmark1 I010ToAR30
runbenchmark1 I420Copy
runbenchmark1 I420Psnr
runbenchmark1 I420Scale
runbenchmark1 I420Ssim
runbenchmark1 I420ToARGB
runbenchmark1 I420ToNV12
runbenchmark1 I420ToUYVY
runbenchmark1 I422ToI420
runbenchmark1 InitCpuFlags
runbenchmark1 J420ToARGB
runbenchmark1 NV12ToARGB
runbenchmark1 NV12ToI420
runbenchmark1 NV12ToI420Rotate
runbenchmark1 SetCpuFlags
runbenchmark1 YUY2ToI420

62
3rdparty/libyuv/source/video_common.cc vendored Normal file
View File

@@ -0,0 +1,62 @@
/*
* Copyright 2011 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/video_common.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
struct FourCCAliasEntry {
uint32_t alias;
uint32_t canonical;
};
#define NUM_ALIASES 18
static const struct FourCCAliasEntry kFourCCAliases[NUM_ALIASES] = {
{FOURCC_IYUV, FOURCC_I420},
{FOURCC_YU12, FOURCC_I420},
{FOURCC_YU16, FOURCC_I422},
{FOURCC_YU24, FOURCC_I444},
{FOURCC_YUYV, FOURCC_YUY2},
{FOURCC_YUVS, FOURCC_YUY2}, // kCMPixelFormat_422YpCbCr8_yuvs
{FOURCC_HDYC, FOURCC_UYVY},
{FOURCC_2VUY, FOURCC_UYVY}, // kCMPixelFormat_422YpCbCr8
{FOURCC_JPEG, FOURCC_MJPG}, // Note: JPEG has DHT while MJPG does not.
{FOURCC_DMB1, FOURCC_MJPG},
{FOURCC_BA81, FOURCC_BGGR}, // deprecated.
{FOURCC_RGB3, FOURCC_RAW},
{FOURCC_BGR3, FOURCC_24BG},
{FOURCC_CM32, FOURCC_BGRA}, // kCMPixelFormat_32ARGB
{FOURCC_CM24, FOURCC_RAW}, // kCMPixelFormat_24RGB
{FOURCC_L555, FOURCC_RGBO}, // kCMPixelFormat_16LE555
{FOURCC_L565, FOURCC_RGBP}, // kCMPixelFormat_16LE565
{FOURCC_5551, FOURCC_RGBO}, // kCMPixelFormat_16LE5551
};
// TODO(fbarchard): Consider mapping kCMPixelFormat_32BGRA to FOURCC_ARGB.
// {FOURCC_BGRA, FOURCC_ARGB}, // kCMPixelFormat_32BGRA
LIBYUV_API
uint32_t CanonicalFourCC(uint32_t fourcc) {
int i;
for (i = 0; i < NUM_ALIASES; ++i) {
if (kFourCCAliases[i].alias == fourcc) {
return kFourCCAliases[i].canonical;
}
}
// Not an alias, so return it as-is.
return fourcc;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif