coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2026-10-28 18:07:12 +08:00
Code Issues Packages Projects Releases Wiki Activity
libyuv/unit_test/scale_plane_test.cc
Frank Barchard a84bf45a21 [libyuv] Fix MSan uninitialized errors in AVX512 alpha and scale rows 
Clang MemorySanitizer builds define LIBYUV_ENABLE_ROWWIN and compile
row_win.cc and scale_win.cc instead of inline assembly. Two AVX-512
intrinsic patterns in those files are not modeled by LLVM MSan:

1. _mm512_mask_cvtepi32_storeu_epi8 and _mm512_mask_cvtepi64_storeu_epi8
   lower to @llvm.x86.avx512.mask.pmov.db.mem.512 and
   @llvm.x86.avx512.mask.pmov.qb.mem.512, which LLVM MSan handles as
   unknown intrinsics without unpoisoning destination shadow memory.
   Replace them in ARGBExtractAlphaRow_AVX512BW, ScaleAddCols1_AVX512BW,
   and ScaleAddCols2_AVX512BW with _mm512_cvtepi32_epi8 /
   _mm512_cvtepi64_epi8 followed by _mm_mask_storeu_epi8, which lowers
   to @llvm.masked.store and still compiles to a single vpmovdb/vpmovqb
   instruction.

2. ARGBCopyYToAlphaRow_AVX512BW loaded dst and blended the alpha channel
   using _mm512_ternarylogic_epi32, which LLVM MSan does not model
   bitwise and flags when any destination byte (such as the old alpha
   channel) is uninitialized. Replace the load + vpternlogd + store in
   both row_win.cc and row_gcc.cc with byte-masked stores
   (_mm512_mask_storeu_epi8 / vmovdqu8 with mask 0x8888888888888888)
   that write only the alpha bytes without reading dst.

Expand ARGBCopyAlpha, ARGBExtractAlpha, and ARGBCopyYToAlpha unit tests
into _Any, _Unaligned, _Invert, and _Opt variants and poison destination
buffers under MSan to catch unmodeled masked stores. Document MSan
AVX-512 intrinsic guidelines in GEMINI.md.

Test: out/msan/libyuv_unittest --gtest_filter=*Alpha*:*ScalePlaneBox_Any*
Bug: 578615529
Change-Id: I54ac4a5844aca6cfe3aa3445fa04b8221c610b35
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/8672001
Commit-Queue: Frank Barchard <fbarchard@google.com>
Reviewed-by: Dale Curtis <dalecurtis@chromium.org>
2026-10-27 16:27:29 -07:00

913 lines
31 KiB
C++
Raw Blame History

/*
* Copyright 2023 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 <limits.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <time.h>
#include <new>
#include "../unit_test/unit_test.h"
#include "libyuv/cpu_id.h"
#include "libyuv/scale.h"
#include "libyuv/scale_uv.h"
#ifdef ENABLE_ROW_TESTS
#include "libyuv/scale_row.h" // For ScaleRowDown2Box_Odd_C
#endif
#if defined(__has_feature)
#if __has_feature(memory_sanitizer)
#include <sanitizer/msan_interface.h>
#define MSAN_POISON(ptr, size) __msan_allocated_memory(ptr, size)
#endif
#endif
#ifndef MSAN_POISON
#define MSAN_POISON(ptr, size)
#endif
#define STRINGIZE(line) #line
#define FILELINESTR(file, line) file ":" STRINGIZE(line)
#if (defined(__riscv) && !defined(__clang__)) || defined(__hexagon__)
#define DISABLE_SLOW_TESTS
#undef ENABLE_FULL_TESTS
#undef ENABLE_ROW_TESTS
#define LEAN_TESTS
#endif
#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
// SLOW TESTS are those that are unoptimized C code.
// FULL TESTS are optimized but test many variations of the same code.
#define ENABLE_FULL_TESTS
#endif
namespace libyuv {
// POC: int row_stride = src_stride * 2 overflows to a small negative value
// when src_stride is close to INT_MAX, causing src_ptr to walk backward
// past the start of the source allocation on the second loop iteration.
// With src_stride = 0x7FFFFFFE, row_stride = (int)0xFFFFFFFC = -4, so on
// y=1 ScaleRowDown2Box reads 4 bytes before the heap allocation.
TEST_F(LibYUVScaleTest, ScalePlaneDown2_RowStrideOverflow) {
constexpr int kSrcStride = 0x7FFFFFFE; // INT_MAX - 1
constexpr int kSrcW = 64;
constexpr int kSrcH = 4;
constexpr int kDstW = 32;
constexpr int kDstH = 2;
// src_size = (kSrcH - 1) * stride + width.
size_t src_size = kSrcH - 1;
if (src_size > SIZE_MAX / kSrcStride) {
GTEST_SKIP() << "could not represent allocation size in size_t";
}
src_size *= kSrcStride;
if (src_size > SIZE_MAX - kSrcW) {
GTEST_SKIP() << "could not represent allocation size in size_t";
}
src_size += kSrcW;
#if defined(__aarch64__)
// Infer malloc can accept a large size for cpu with dot product (a76/a55)
int has_large_malloc = TestCpuFlag(kCpuHasNeonDotProd);
#else
int has_large_malloc = 1;
#endif
if (!has_large_malloc) {
GTEST_SKIP() << "large allocation may assert for " << src_size << " bytes";
}
uint8_t* src = new (std::nothrow) uint8_t[src_size];
if (!src) {
GTEST_SKIP() << "could not allocate " << src_size << " bytes";
}
uint8_t dst[kDstW * kDstH];
uint8_t* src_row = src;
for (int i = 0; i < kSrcH; i++) {
memset(src_row, 0x41, kSrcW);
src_row += kSrcStride;
}
// Force the C row kernel: the SIMD kernels are inline asm that ASAN does not
// instrument, so they silently read OOB without a report.
MaskCpuFlags(1);
// 2*dst == src on both axes -> ScalePlane dispatches to ScalePlaneDown2.
// int row_stride = kSrcStride * 2 wraps to -4; on y=1 src_ptr underflows.
ScalePlane(src, kSrcStride, kSrcW, kSrcH, dst, kDstW, kDstW, kDstH,
kFilterBox);
MaskCpuFlags(0);
delete[] src;
}
// POC: same defect in the 1/4 fast path. src_stride = 0x3FFFFFFF gives
// int row_stride = src_stride * 4 = (int)0xFFFFFFFC = -4.
TEST_F(LibYUVScaleTest, ScalePlaneDown4_RowStrideOverflow) {
constexpr int kSrcStride = 0x3FFFFFFF; // INT_MAX / 4 (rounded down)
constexpr int kSrcW = 64;
constexpr int kSrcH = 8;
constexpr int kDstW = 16;
constexpr int kDstH = 2;
// src_size = (kSrcH - 1) * stride + width.
size_t src_size = kSrcH - 1;
if (src_size > SIZE_MAX / kSrcStride) {
GTEST_SKIP() << "could not represent allocation size in size_t";
}
src_size *= kSrcStride;
if (src_size > SIZE_MAX - kSrcW) {
GTEST_SKIP() << "could not represent allocation size in size_t";
}
src_size += kSrcW;
#if defined(__aarch64__)
// Infer malloc can accept a large size for cpu with dot product (a76/a55)
int has_large_malloc = TestCpuFlag(kCpuHasNeonDotProd);
#else
int has_large_malloc = 1;
#endif
if (!has_large_malloc) {
GTEST_SKIP() << "large allocation may assert for " << src_size << " bytes";
}
uint8_t* src = new (std::nothrow) uint8_t[src_size];
if (!src) {
GTEST_SKIP() << "could not allocate " << src_size << " bytes";
}
uint8_t dst[kDstW * kDstH];
uint8_t* src_row = src;
for (int i = 0; i < kSrcH; i++) {
memset(src_row, 0x41, kSrcW);
src_row += kSrcStride;
}
// Force the C row kernel: the SIMD kernels are inline asm that ASAN does not
// instrument, so they silently read OOB without a report.
MaskCpuFlags(1);
// 4*dst == src on both axes with kFilterBox -> ScalePlaneDown4.
ScalePlane(src, kSrcStride, kSrcW, kSrcH, dst, kDstW, kDstW, kDstH,
kFilterBox);
MaskCpuFlags(0);
delete[] src;
}
#ifdef ENABLE_ROW_TESTS
#ifdef HAS_SCALEROWDOWN2_SSSE3
TEST_F(LibYUVScaleTest, TestScaleRowDown2Box_Odd_SSSE3) {
SIMD_ALIGNED(uint8_t orig_pixels[128 * 2]);
SIMD_ALIGNED(uint8_t dst_pixels_opt[64]);
SIMD_ALIGNED(uint8_t dst_pixels_c[64]);
memset(orig_pixels, 0, sizeof(orig_pixels));
memset(dst_pixels_opt, 0, sizeof(dst_pixels_opt));
memset(dst_pixels_c, 0, sizeof(dst_pixels_c));
int has_ssse3 = TestCpuFlag(kCpuHasSSSE3);
if (!has_ssse3) {
printf("Warning SSSE3 not detected; Skipping test.\n");
} else {
// TL.
orig_pixels[0] = 255u;
orig_pixels[1] = 0u;
orig_pixels[128 + 0] = 0u;
orig_pixels[128 + 1] = 0u;
// TR.
orig_pixels[2] = 0u;
orig_pixels[3] = 100u;
orig_pixels[128 + 2] = 0u;
orig_pixels[128 + 3] = 0u;
// BL.
orig_pixels[4] = 0u;
orig_pixels[5] = 0u;
orig_pixels[128 + 4] = 50u;
orig_pixels[128 + 5] = 0u;
// BR.
orig_pixels[6] = 0u;
orig_pixels[7] = 0u;
orig_pixels[128 + 6] = 0u;
orig_pixels[128 + 7] = 20u;
// Odd.
orig_pixels[126] = 4u;
orig_pixels[127] = 255u;
orig_pixels[128 + 126] = 16u;
orig_pixels[128 + 127] = 255u;
// Test regular half size.
ScaleRowDown2Box_C(orig_pixels, 128, dst_pixels_c, 64);
ASSERT_EQ(64u, dst_pixels_c[0]);
ASSERT_EQ(25u, dst_pixels_c[1]);
ASSERT_EQ(13u, dst_pixels_c[2]);
ASSERT_EQ(5u, dst_pixels_c[3]);
ASSERT_EQ(0u, dst_pixels_c[4]);
ASSERT_EQ(133u, dst_pixels_c[63]);
// Test Odd width version - Last pixel is just 1 horizontal pixel.
ScaleRowDown2Box_Odd_C(orig_pixels, 128, dst_pixels_c, 64);
ASSERT_EQ(64u, dst_pixels_c[0]);
ASSERT_EQ(25u, dst_pixels_c[1]);
ASSERT_EQ(13u, dst_pixels_c[2]);
ASSERT_EQ(5u, dst_pixels_c[3]);
ASSERT_EQ(0u, dst_pixels_c[4]);
ASSERT_EQ(10u, dst_pixels_c[63]);
// Test one pixel less, should skip the last pixel.
memset(dst_pixels_c, 0, sizeof(dst_pixels_c));
ScaleRowDown2Box_Odd_C(orig_pixels, 128, dst_pixels_c, 63);
ASSERT_EQ(64u, dst_pixels_c[0]);
ASSERT_EQ(25u, dst_pixels_c[1]);
ASSERT_EQ(13u, dst_pixels_c[2]);
ASSERT_EQ(5u, dst_pixels_c[3]);
ASSERT_EQ(0u, dst_pixels_c[4]);
ASSERT_EQ(0u, dst_pixels_c[63]);
// Test regular half size SSSE3.
ScaleRowDown2Box_SSSE3(orig_pixels, 128, dst_pixels_opt, 64);
ASSERT_EQ(64u, dst_pixels_opt[0]);
ASSERT_EQ(25u, dst_pixels_opt[1]);
ASSERT_EQ(13u, dst_pixels_opt[2]);
ASSERT_EQ(5u, dst_pixels_opt[3]);
ASSERT_EQ(0u, dst_pixels_opt[4]);
ASSERT_EQ(133u, dst_pixels_opt[63]);
// Compare C and SSSE3 match.
ScaleRowDown2Box_Odd_C(orig_pixels, 128, dst_pixels_c, 64);
ScaleRowDown2Box_Odd_SSSE3(orig_pixels, 128, dst_pixels_opt, 64);
for (int i = 0; i < 64; ++i) {
ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
}
}
}
#endif // HAS_SCALEROWDOWN2_SSSE3
TEST_F(LibYUVScaleTest, TestScaleRowDown2Box_16) {
SIMD_ALIGNED(uint16_t orig_pixels[2560 * 2]);
SIMD_ALIGNED(uint16_t dst_pixels_c[1280]);
SIMD_ALIGNED(uint16_t dst_pixels_opt[1280]);
memset(orig_pixels, 0, sizeof(orig_pixels));
memset(dst_pixels_c, 1, sizeof(dst_pixels_c));
memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt));
for (int i = 0; i < 2560 * 2; ++i) {
orig_pixels[i] = i;
}
ScaleRowDown2Box_16_C(&orig_pixels[0], 2560, &dst_pixels_c[0], 1280);
for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
int has_neon = TestCpuFlag(kCpuHasNEON);
if (has_neon) {
ScaleRowDown2Box_16_NEON(&orig_pixels[0], 2560, &dst_pixels_opt[0], 1280);
} else {
ScaleRowDown2Box_16_C(&orig_pixels[0], 2560, &dst_pixels_opt[0], 1280);
}
#else
ScaleRowDown2Box_16_C(&orig_pixels[0], 2560, &dst_pixels_opt[0], 1280);
#endif
}
for (int i = 0; i < 1280; ++i) {
ASSERT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
}
ASSERT_EQ(dst_pixels_c[0], (0 + 1 + 2560 + 2561 + 2) / 4);
ASSERT_EQ(dst_pixels_c[1279], 3839);
}
#endif // ENABLE_ROW_TESTS
// Test scaling plane with 8 bit C vs 12 bit C and return maximum pixel
// difference.
// 0 = exact.
static int TestPlaneFilter_16(int src_width,
int src_height,
int dst_width,
int dst_height,
FilterMode f,
int benchmark_iterations,
int disable_cpu_flags,
int benchmark_cpu_info) {
if (!SizeValid(src_width, src_height, dst_width, dst_height)) {
return 0;
}
int i;
int64_t src_y_plane_size = (Abs(src_width)) * (Abs(src_height));
int src_stride_y = Abs(src_width);
int dst_y_plane_size = dst_width * dst_height;
int dst_stride_y = dst_width;
align_buffer_page_end(src_y, src_y_plane_size);
align_buffer_page_end(src_y_16, src_y_plane_size * 2);
align_buffer_page_end(dst_y_8, dst_y_plane_size);
align_buffer_page_end(dst_y_16, dst_y_plane_size * 2);
uint16_t* p_src_y_16 = reinterpret_cast<uint16_t*>(src_y_16);
uint16_t* p_dst_y_16 = reinterpret_cast<uint16_t*>(dst_y_16);
MemRandomize(src_y, src_y_plane_size);
memset(dst_y_8, 0, dst_y_plane_size);
memset(dst_y_16, 1, dst_y_plane_size * 2);
for (i = 0; i < src_y_plane_size; ++i) {
p_src_y_16[i] = src_y[i] & 255;
}
MaskCpuFlags(disable_cpu_flags); // Disable all CPU optimization.
ScalePlane(src_y, src_stride_y, src_width, src_height, dst_y_8, dst_stride_y,
dst_width, dst_height, f);
MaskCpuFlags(benchmark_cpu_info); // Enable all CPU optimization.
for (i = 0; i < benchmark_iterations; ++i) {
ScalePlane_16(p_src_y_16, src_stride_y, src_width, src_height, p_dst_y_16,
dst_stride_y, dst_width, dst_height, f);
}
// Expect an exact match.
int max_diff = 0;
for (i = 0; i < dst_y_plane_size; ++i) {
int abs_diff = Abs(dst_y_8[i] - p_dst_y_16[i]);
if (abs_diff > max_diff) {
max_diff = abs_diff;
}
}
free_aligned_buffer_page_end(dst_y_8);
free_aligned_buffer_page_end(dst_y_16);
free_aligned_buffer_page_end(src_y);
free_aligned_buffer_page_end(src_y_16);
return max_diff;
}
// The following adjustments in dimensions ensure the scale factor will be
// exactly achieved.
// 2 is chroma subsample.
#define DX(x, nom, denom) static_cast<int>(((Abs(x) / nom + 1) / 2) * nom * 2)
#define SX(x, nom, denom) static_cast<int>(((x / nom + 1) / 2) * denom * 2)
#define TEST_FACTOR1(name, filter, nom, denom, max_diff) \
TEST_F(LibYUVScaleTest, DISABLED_##ScalePlaneDownBy##name##_##filter##_16) { \
int diff = TestPlaneFilter_16( \
SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
kFilter##filter, benchmark_iterations_, disable_cpu_flags_, \
benchmark_cpu_info_); \
ASSERT_LE(diff, max_diff); \
}
// Test a scale factor with all 4 filters. Expect unfiltered to be exact, but
// filtering is different fixed point implementations for SSSE3, Neon and C.
#define TEST_FACTOR(name, nom, denom, boxdiff) \
TEST_FACTOR1(name, None, nom, denom, 0) \
TEST_FACTOR1(name, Linear, nom, denom, boxdiff) \
TEST_FACTOR1(name, Bilinear, nom, denom, boxdiff) \
TEST_FACTOR1(name, Box, nom, denom, boxdiff)
TEST_FACTOR(2, 1, 2, 0)
TEST_FACTOR(4, 1, 4, 0)
// TEST_FACTOR(8, 1, 8, 0) Disable for benchmark performance. Takes 90 seconds.
TEST_FACTOR(3by4, 3, 4, 1)
TEST_FACTOR(3by8, 3, 8, 1)
TEST_FACTOR(3, 1, 3, 0)
#undef TEST_FACTOR1
#undef TEST_FACTOR
#undef SX
#undef DX
TEST_F(LibYUVScaleTest, PlaneTest3x) {
const int kSrcStride = 480;
const int kDstStride = 160;
const int kSize = kSrcStride * 3;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < 480 * 3; ++i) {
orig_pixels[i] = i;
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
benchmark_iterations_;
for (int i = 0; i < iterations160; ++i) {
ScalePlane(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
kFilterBilinear);
}
ASSERT_EQ(225, dest_pixels[0]);
ScalePlane(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
kFilterNone);
ASSERT_EQ(225, dest_pixels[0]);
free_aligned_buffer_page_end(dest_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
TEST_F(LibYUVScaleTest, PlaneTest4x) {
const int kSrcStride = 640;
const int kDstStride = 160;
const int kSize = kSrcStride * 4;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < 640 * 4; ++i) {
orig_pixels[i] = i;
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
benchmark_iterations_;
for (int i = 0; i < iterations160; ++i) {
ScalePlane(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
kFilterBilinear);
}
ASSERT_EQ(66, dest_pixels[0]);
ScalePlane(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
kFilterNone);
ASSERT_EQ(2, dest_pixels[0]); // expect the 3rd pixel of the 3rd row
free_aligned_buffer_page_end(dest_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
// Intent is to test 200x50 to 50x200 but width and height can be parameters.
TEST_F(LibYUVScaleTest, PlaneTestRotate_None) {
const int kSize = benchmark_width_ * benchmark_height_;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < kSize; ++i) {
orig_pixels[i] = i;
}
align_buffer_page_end(dest_opt_pixels, kSize);
align_buffer_page_end(dest_c_pixels, kSize);
MaskCpuFlags(disable_cpu_flags_); // Disable all CPU optimization.
ScalePlane(orig_pixels, benchmark_width_, benchmark_width_, benchmark_height_,
dest_c_pixels, benchmark_height_, benchmark_height_,
benchmark_width_, kFilterNone);
MaskCpuFlags(benchmark_cpu_info_); // Enable all CPU optimization.
for (int i = 0; i < benchmark_iterations_; ++i) {
ScalePlane(orig_pixels, benchmark_width_, benchmark_width_,
benchmark_height_, dest_opt_pixels, benchmark_height_,
benchmark_height_, benchmark_width_, kFilterNone);
}
for (int i = 0; i < kSize; ++i) {
ASSERT_EQ(dest_c_pixels[i], dest_opt_pixels[i]);
}
free_aligned_buffer_page_end(dest_c_pixels);
free_aligned_buffer_page_end(dest_opt_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
TEST_F(LibYUVScaleTest, PlaneTestRotate_Bilinear) {
const int kSize = benchmark_width_ * benchmark_height_;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < kSize; ++i) {
orig_pixels[i] = i;
}
align_buffer_page_end(dest_opt_pixels, kSize);
align_buffer_page_end(dest_c_pixels, kSize);
MaskCpuFlags(disable_cpu_flags_); // Disable all CPU optimization.
ScalePlane(orig_pixels, benchmark_width_, benchmark_width_, benchmark_height_,
dest_c_pixels, benchmark_height_, benchmark_height_,
benchmark_width_, kFilterBilinear);
MaskCpuFlags(benchmark_cpu_info_); // Enable all CPU optimization.
for (int i = 0; i < benchmark_iterations_; ++i) {
ScalePlane(orig_pixels, benchmark_width_, benchmark_width_,
benchmark_height_, dest_opt_pixels, benchmark_height_,
benchmark_height_, benchmark_width_, kFilterBilinear);
}
for (int i = 0; i < kSize; ++i) {
ASSERT_EQ(dest_c_pixels[i], dest_opt_pixels[i]);
}
free_aligned_buffer_page_end(dest_c_pixels);
free_aligned_buffer_page_end(dest_opt_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
// Intent is to test 200x50 to 50x200 but width and height can be parameters.
TEST_F(LibYUVScaleTest, PlaneTestRotate_Box) {
const int kSize = benchmark_width_ * benchmark_height_;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < kSize; ++i) {
orig_pixels[i] = i;
}
align_buffer_page_end(dest_opt_pixels, kSize);
align_buffer_page_end(dest_c_pixels, kSize);
MaskCpuFlags(disable_cpu_flags_); // Disable all CPU optimization.
ScalePlane(orig_pixels, benchmark_width_, benchmark_width_, benchmark_height_,
dest_c_pixels, benchmark_height_, benchmark_height_,
benchmark_width_, kFilterBox);
MaskCpuFlags(benchmark_cpu_info_); // Enable all CPU optimization.
for (int i = 0; i < benchmark_iterations_; ++i) {
ScalePlane(orig_pixels, benchmark_width_, benchmark_width_,
benchmark_height_, dest_opt_pixels, benchmark_height_,
benchmark_height_, benchmark_width_, kFilterBox);
}
for (int i = 0; i < kSize; ++i) {
ASSERT_EQ(dest_c_pixels[i], dest_opt_pixels[i]);
}
free_aligned_buffer_page_end(dest_c_pixels);
free_aligned_buffer_page_end(dest_opt_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
TEST_F(LibYUVScaleTest, ScalePlaneBox_Any) {
const int kDstW = 29;
const int kDstH = 3;
const int kSrcH = 9;
for (int src_w : {87, 174, 73, 155}) {
const int kSrcSize = src_w * kSrcH;
const int kDstSize = kDstW * kDstH;
align_buffer_page_end(src_pixels, kSrcSize);
align_buffer_page_end(dst_c_pixels, kDstSize);
align_buffer_page_end(dst_opt_pixels, kDstSize);
MemRandomize(src_pixels, kSrcSize);
MSAN_POISON(dst_opt_pixels, kDstSize);
MaskCpuFlags(disable_cpu_flags_);
ScalePlane(src_pixels, src_w, src_w, kSrcH, dst_c_pixels, kDstW, kDstW,
kDstH, kFilterBox);
MaskCpuFlags(benchmark_cpu_info_);
ScalePlane(src_pixels, src_w, src_w, kSrcH, dst_opt_pixels, kDstW, kDstW,
kDstH, kFilterBox);
for (int i = 0; i < kDstSize; ++i) {
ASSERT_EQ(dst_c_pixels[i], dst_opt_pixels[i]);
}
free_aligned_buffer_page_end(dst_opt_pixels);
free_aligned_buffer_page_end(dst_c_pixels);
free_aligned_buffer_page_end(src_pixels);
}
}
TEST_F(LibYUVScaleTest, PlaneTest1_Box) {
align_buffer_page_end(orig_pixels, 3);
align_buffer_page_end(dst_pixels, 3);
// Pad the 1x1 byte image with invalid values before and after in case libyuv
// reads outside the memory boundaries.
orig_pixels[0] = 0;
orig_pixels[1] = 1; // scale this pixel
orig_pixels[2] = 2;
dst_pixels[0] = 3;
dst_pixels[1] = 3;
dst_pixels[2] = 3;
libyuv::ScalePlane(orig_pixels + 1, /* src_stride= */ 1, /* src_width= */ 1,
/* src_height= */ 1, dst_pixels, /* dst_stride= */ 1,
/* dst_width= */ 1, /* dst_height= */ 2,
libyuv::kFilterBox);
ASSERT_EQ(dst_pixels[0], 1);
ASSERT_EQ(dst_pixels[1], 1);
ASSERT_EQ(dst_pixels[2], 3);
free_aligned_buffer_page_end(dst_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
TEST_F(LibYUVScaleTest, PlaneTest1_16_Box) {
align_buffer_page_end(orig_pixels_alloc, 3 * 2);
align_buffer_page_end(dst_pixels_alloc, 3 * 2);
uint16_t* orig_pixels = (uint16_t*)orig_pixels_alloc;
uint16_t* dst_pixels = (uint16_t*)dst_pixels_alloc;
// Pad the 1x1 byte image with invalid values before and after in case libyuv
// reads outside the memory boundaries.
orig_pixels[0] = 0;
orig_pixels[1] = 1; // scale this pixel
orig_pixels[2] = 2;
dst_pixels[0] = 3;
dst_pixels[1] = 3;
dst_pixels[2] = 3;
libyuv::ScalePlane_16(
orig_pixels + 1, /* src_stride= */ 1, /* src_width= */ 1,
/* src_height= */ 1, dst_pixels, /* dst_stride= */ 1,
/* dst_width= */ 1, /* dst_height= */ 2, libyuv::kFilterNone);
ASSERT_EQ(dst_pixels[0], 1);
ASSERT_EQ(dst_pixels[1], 1);
ASSERT_EQ(dst_pixels[2], 3);
free_aligned_buffer_page_end(dst_pixels_alloc);
free_aligned_buffer_page_end(orig_pixels_alloc);
}
TEST_F(LibYUVScaleTest, ScalePlaneVerticalPointUp) {
const int kWidth = 32;
const uint8_t kSentinel = 0xFF;
const uint8_t expected[] = {10, 10, 20, 30, 30};
// An extra sentinel row exposes accidental interpolation past the last valid
// row. 3->5 is the uneven ratio that used to drop the last source row.
align_buffer_page_end(src, kWidth * 4);
align_buffer_page_end(dst, kWidth * 5);
memset(src, kSentinel, kWidth * 4);
memset(dst, kSentinel, kWidth * 5);
memset(src, 10, kWidth);
memset(src + kWidth, 20, kWidth);
memset(src + 2 * kWidth, 30, kWidth);
ASSERT_EQ(0, ScalePlane(src, kWidth, kWidth, 3, dst, kWidth, kWidth, 5,
kFilterNone));
for (int y = 0; y < 5; ++y) {
for (int x = 0; x < kWidth; ++x) {
EXPECT_EQ(expected[y], dst[y * kWidth + x]);
}
}
free_aligned_buffer_page_end(dst);
free_aligned_buffer_page_end(src);
}
TEST_F(LibYUVScaleTest, ScalePlaneVerticalPointUp_16) {
const int kWidth = 32;
const uint16_t kSentinel = 0xFFFF;
const uint16_t expected[] = {10, 10, 20, 30, 30};
align_buffer_page_end(src_alloc, kWidth * 4 * 2);
align_buffer_page_end(dst_alloc, kWidth * 5 * 2);
uint16_t* src = reinterpret_cast<uint16_t*>(src_alloc);
uint16_t* dst = reinterpret_cast<uint16_t*>(dst_alloc);
for (int i = 0; i < kWidth * 4; ++i) {
src[i] = kSentinel;
}
for (int i = 0; i < kWidth * 5; ++i) {
dst[i] = kSentinel;
}
for (int x = 0; x < kWidth; ++x) {
src[x] = 10;
src[kWidth + x] = 20;
src[2 * kWidth + x] = 30;
}
ASSERT_EQ(0, ScalePlane_16(src, kWidth, kWidth, 3, dst, kWidth, kWidth, 5,
kFilterNone));
for (int y = 0; y < 5; ++y) {
for (int x = 0; x < kWidth; ++x) {
EXPECT_EQ(expected[y], dst[y * kWidth + x]);
}
}
free_aligned_buffer_page_end(dst_alloc);
free_aligned_buffer_page_end(src_alloc);
}
TEST_F(LibYUVScaleTest, ScalePlaneVerticalBilinearUp) {
const int kWidth = 32;
const uint8_t kSentinel = 0xFF;
// Same-width 2x upsampling still uses FixedDiv1. With y = 0 the 8.8
// fractions are 0, 85, 170, 255. InterpolateRow_C:
// (a * (256 - yf) + b * yf + 128) >> 8.
// Nearest 2x would duplicate rows as {10, 10, 200, 200}.
const uint8_t expected[] = {10, 73, 136, 199};
align_buffer_page_end(src, kWidth * 3);
align_buffer_page_end(dst, kWidth * 4);
memset(src, kSentinel, kWidth * 3);
memset(dst, kSentinel, kWidth * 4);
memset(src, 10, kWidth);
memset(src + kWidth, 200, kWidth);
ASSERT_EQ(0, ScalePlane(src, kWidth, kWidth, 2, dst, kWidth, kWidth, 4,
kFilterBilinear));
for (int y = 0; y < 4; ++y) {
for (int x = 0; x < kWidth; ++x) {
EXPECT_EQ(expected[y], dst[y * kWidth + x]);
}
}
free_aligned_buffer_page_end(dst);
free_aligned_buffer_page_end(src);
}
TEST_F(LibYUVScaleTest, ScalePlaneVerticalBilinearUp_16) {
const int kWidth = 32;
const uint16_t kSentinel = 0xFFFF;
const uint16_t expected[] = {10, 73, 136, 199};
align_buffer_page_end(src_alloc, kWidth * 3 * 2);
align_buffer_page_end(dst_alloc, kWidth * 4 * 2);
uint16_t* src = reinterpret_cast<uint16_t*>(src_alloc);
uint16_t* dst = reinterpret_cast<uint16_t*>(dst_alloc);
for (int i = 0; i < kWidth * 3; ++i) {
src[i] = kSentinel;
}
for (int i = 0; i < kWidth * 4; ++i) {
dst[i] = kSentinel;
}
for (int x = 0; x < kWidth; ++x) {
src[x] = 10;
src[kWidth + x] = 200;
}
ASSERT_EQ(0, ScalePlane_16(src, kWidth, kWidth, 2, dst, kWidth, kWidth, 4,
kFilterBilinear));
for (int y = 0; y < 4; ++y) {
for (int x = 0; x < kWidth; ++x) {
EXPECT_EQ(expected[y], dst[y * kWidth + x]);
}
}
free_aligned_buffer_page_end(dst_alloc);
free_aligned_buffer_page_end(src_alloc);
}
// POC: int * int overflow in ScalePlaneVertical (scale_common.cc).
//
// `yi * src_stride` is evaluated as int * int. When the product exceeds
// INT_MAX it wraps negative and InterpolateRow reads from BEFORE the
// source allocation.
//
// Parameters:
// - dst_width == src_width
// -> ScalePlane dispatches to ScalePlaneVertical
// - src_height == 5, dst_height == 1
// -> single iteration with yi == 2
// - src_stride == 0x7FFFFFF8
// -> 2 * 0x7FFFFFF8 == 0xFFFFFFF0 == -16 (int)
//
// The source buffer is sized so that the *correct* 64-bit offset
// (2 * 0x7FFFFFF8 == 4294967280) plus kWidth bytes is in-bounds. With the
// bug, the 32-bit product is -16 and ASAN reports a heap-buffer-overflow
// READ "16 bytes before" the allocation.
TEST_F(LibYUVScaleTest, ScalePlaneVertical_IntStrideOverflow) {
const int kWidth = 16;
const int kSrcHeight = 5;
const int kDstHeight = 1;
const int kStride = 0x7FFFFFF8; // 2147483640
// src_size is big enough for the only row this call legitimately touches
// (yi == 2) when computed in 64-bit: 2 * stride + width = 4 GiB.
size_t src_size = kStride;
if (src_size > SIZE_MAX / 2) {
GTEST_SKIP() << "could not represent allocation size in size_t";
}
src_size *= 2;
if (src_size > SIZE_MAX - kWidth) {
GTEST_SKIP() << "could not represent allocation size in size_t";
}
src_size += kWidth;
#if defined(__aarch64__)
// Infer malloc can accept a large size for cpu with dot product (a76/a55)
int has_large_malloc = TestCpuFlag(kCpuHasNeonDotProd);
#else
int has_large_malloc = 1;
#endif
if (!has_large_malloc) {
GTEST_SKIP() << "large allocation may assert for " << src_size << " bytes";
}
uint8_t* src = new (std::nothrow) uint8_t[src_size];
if (!src) {
GTEST_SKIP() << "could not allocate " << src_size << " bytes";
}
uint8_t* dst = new uint8_t[kWidth];
memset(dst, 0, kWidth);
// Force the scalar path so the crash site is deterministic
// (InterpolateRow_C -> memcpy when yf == 0).
MaskCpuFlags(disable_cpu_flags_);
int r = ScalePlane(src, kStride, kWidth, kSrcHeight, dst, kWidth, kWidth,
kDstHeight, kFilterNone);
// Not reached under ASAN.
ASSERT_EQ(0, r);
delete[] src;
delete[] dst;
}
TEST_F(LibYUVScaleTest, ScalePlane_InvalidInputs) {
uint8_t src[16] = {0};
uint8_t dst[16] = {0};
// NULL src/dst
EXPECT_EQ(-1, ScalePlane(nullptr, 4, 4, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, 4, 4, nullptr, 4, 4, 4, kFilterNone));
// Width/height <= 0 (except src_height which can be negative but not 0)
EXPECT_EQ(-1, ScalePlane(src, 4, 0, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, -1, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, 4, 0, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, 4, 4, dst, 4, 0, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, 4, 4, dst, 4, -1, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, 4, 4, dst, 4, 4, 0, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, 4, 4, dst, 4, 4, -1, kFilterNone));
// Width/height too large (> 32768)
EXPECT_EQ(-1, ScalePlane(src, 4, 32769, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, 4, 32769, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane(src, 4, 4, -32769, dst, 4, 4, 4, kFilterNone));
// Valid edge cases
EXPECT_EQ(0, ScalePlane(src, 4, 1, 1, dst, 4, 1, 1, kFilterNone));
EXPECT_EQ(0, ScalePlane(src, 4, 1, -1, dst, 4, 1, 1, kFilterNone));
}
TEST_F(LibYUVScaleTest, ScalePlane_16_InvalidInputs) {
uint16_t src[16] = {0};
uint16_t dst[16] = {0};
EXPECT_EQ(-1, ScalePlane_16(nullptr, 4, 4, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane_16(src, 4, 4, 4, nullptr, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane_16(src, 4, 0, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane_16(src, 4, 32769, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane_16(src, 4, 4, -32769, dst, 4, 4, 4, kFilterNone));
}
TEST_F(LibYUVScaleTest, ScalePlane_12_InvalidInputs) {
uint16_t src[16] = {0};
uint16_t dst[16] = {0};
EXPECT_EQ(-1, ScalePlane_12(nullptr, 4, 4, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane_12(src, 4, 4, 4, nullptr, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane_12(src, 4, 0, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane_12(src, 4, 32769, 4, dst, 4, 4, 4, kFilterNone));
EXPECT_EQ(-1, ScalePlane_12(src, 4, 4, -32769, dst, 4, 4, 4, kFilterNone));
}
TEST_F(LibYUVScaleTest, ScalePlaneOversizeHeight) {
const int kSrcWidth = 4;
const int kSrcHeight = 32768;
uint8_t* src = new uint8_t[kSrcWidth * kSrcHeight];
memset(src, 0, kSrcWidth * kSrcHeight);
uint8_t dst[4 * 2] = {};
for (int dst_w : {4, 2}) {
for (int dst_h : {1, 2}) {
for (FilterMode f :
{kFilterNone, kFilterLinear, kFilterBilinear, kFilterBox}) {
EXPECT_EQ(0, ScalePlane(src, kSrcWidth, kSrcWidth, kSrcHeight, dst,
dst_w, dst_w, dst_h, f));
EXPECT_EQ(0, ScalePlane(src, kSrcWidth, kSrcWidth, -kSrcHeight, dst,
dst_w, dst_w, dst_h, f));
}
}
}
delete[] src;
}
TEST_F(LibYUVScaleTest, ScalePlaneBoxMaxSrcHeight) {
const int kSrcWidth = 4;
const int kSrcHeight = 32768;
uint8_t* src = new uint8_t[kSrcWidth * kSrcHeight];
memset(src, 0, kSrcWidth * kSrcHeight);
uint8_t dst[2 * 4] = {};
EXPECT_EQ(0, ScalePlane(src, kSrcWidth, kSrcWidth, kSrcHeight, dst, 2, 2, 4,
kFilterBox));
delete[] src;
}
#if defined(__LP64__) || defined(_WIN64)
TEST_F(LibYUVScaleTest, ScaleLargeImageOver2GB) {
const int kSrcWidth = 4;
const int kSrcHeight = 32768;
const int kSrcStride = 131072; // 32768 * 131072 = 4 GB (> INT_MAX)
const size_t kSrcSize = (size_t)kSrcStride * kSrcHeight;
uint8_t* src = static_cast<uint8_t*>(calloc(1, kSrcSize));
if (!src) {
GTEST_SKIP() << "could not allocate " << kSrcSize << " bytes";
}
uint8_t dst[8 * 4 * 4] = {};
for (FilterMode f :
{kFilterNone, kFilterLinear, kFilterBilinear, kFilterBox}) {
EXPECT_EQ(
0, ScalePlane(src, kSrcStride, kSrcWidth, kSrcHeight, dst, 4, 4, 2, f));
EXPECT_EQ(0, ScalePlane(src, kSrcStride, kSrcWidth, -kSrcHeight, dst, 4, 4,
2, f));
EXPECT_EQ(
0, ScalePlane(src, kSrcStride, kSrcWidth, kSrcHeight, dst, 8, 8, 2, f));
EXPECT_EQ(0, UVScale(src, kSrcStride, kSrcWidth, kSrcHeight, dst, 4 * 2, 4,
2, f));
EXPECT_EQ(0, UVScale(src, kSrcStride, kSrcWidth, -kSrcHeight, dst, 4 * 2, 4,
2, f));
EXPECT_EQ(0, UVScale(src, kSrcStride, kSrcWidth, kSrcHeight, dst, 8 * 2, 8,
2, f));
}
free(src);
}
#endif
} // namespace libyuv
Reference in New Issue View Git Blame Copy Permalink