Commit c0d6dee4 authored by fbarchard@google.com's avatar fbarchard@google.com

Fix for ARGBColorTable x86 assembly. Sav

BUG=51
TEST=TestARGBColorTable
Review URL: https://webrtc-codereview.appspot.com/667011

git-svn-id: http://libyuv.googlecode.com/svn/trunk@307 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 452472df
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 306
Version: 307
License: BSD
License File: LICENSE
......
......@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 306
#define LIBYUV_VERSION 307
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
......@@ -59,10 +59,11 @@ enum FourCC {
FOURCC_RGBP = FOURCC('R', 'G', 'B', 'P'), // bgr565.
FOURCC_RGBO = FOURCC('R', 'G', 'B', 'O'), // abgr1555.
FOURCC_R444 = FOURCC('R', '4', '4', '4'), // argb4444.
FOURCC_MJPG = FOURCC('M', 'J', 'P', 'G'),
FOURCC_RAW = FOURCC('r', 'a', 'w', ' '),
FOURCC_NV21 = FOURCC('N', 'V', '2', '1'),
FOURCC_NV12 = FOURCC('N', 'V', '1', '2'),
FOURCC_MJPG = FOURCC('M', 'J', 'P', 'G'),
FOURCC_H264 = FOURCC('H', '2', '6', '4'),
// Next four are Bayer RGB formats. The four characters define the order of
// the colours in each 2x2 pixel grid, going left-to-right and top-to-bottom.
FOURCC_RGGB = FOURCC('R', 'G', 'G', 'B'),
......@@ -112,10 +113,11 @@ enum FourCCBpp {
FOURCC_BPP_RGBP = 16,
FOURCC_BPP_RGBO = 16,
FOURCC_BPP_R444 = 16,
FOURCC_BPP_MJPG = 0, // 0 means unknown.
FOURCC_BPP_RAW = 24,
FOURCC_BPP_NV21 = 12,
FOURCC_BPP_NV12 = 12,
FOURCC_BPP_MJPG = 0, // 0 means unknown.
FOURCC_BPP_H264 = 0,
// Next four are Bayer RGB formats. The four characters define the order of
// the colours in each 2x2 pixel grid, going left-to-right and top-to-bottom.
FOURCC_BPP_RGGB = 8,
......
......@@ -86,7 +86,7 @@ extern "C" {
// The following are Windows only:
#if !defined(YUV_DISABLE_ASM) && defined(_M_IX86)
// TODO(fbarchard): Investigate possible issue in this function and reenable.
#define HAS_ARGBCOLORTABLEROW_X86_DISABLED
#define HAS_ARGBCOLORTABLEROW_X86
#endif
// The following are disabled when SSSE3 is available:
......
......@@ -3031,11 +3031,12 @@ void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb,
int width) {
__asm {
push ebx
push esi
push edi
push ebp
mov eax, [esp + 12 + 4] /* dst_argb */
mov edi, [esp + 12 + 8] /* table_argb */
mov ecx, [esp + 12 + 12] /* width */
mov eax, [esp + 16 + 4] /* dst_argb */
mov edi, [esp + 16 + 8] /* table_argb */
mov ecx, [esp + 16 + 12] /* width */
xor ebx, ebx
xor edx, edx
......@@ -3064,6 +3065,7 @@ void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb,
jg convertloop
pop ebp
pop edi
pop esi
pop ebx
ret
}
......
......@@ -46,18 +46,18 @@ TEST_F(libyuvTest, FMT_PLANAR##To##FMT_B##_OptVsC) { \
} \
MaskCpuFlags(kCpuInitialized); \
FMT_PLANAR##To##FMT_B(src_y, kWidth, \
src_u, kWidth / SUBSAMP_X, \
src_v, kWidth / SUBSAMP_X, \
dst_argb_c, kWidth * BPP_B, \
kWidth, kHeight); \
src_u, kWidth / SUBSAMP_X, \
src_v, kWidth / SUBSAMP_X, \
dst_argb_c, kWidth * BPP_B, \
kWidth, kHeight); \
MaskCpuFlags(-1); \
const int runs = 1000; \
for (int i = 0; i < runs; ++i) { \
FMT_PLANAR##To##FMT_B(src_y, kWidth, \
src_u, kWidth / SUBSAMP_X, \
src_v, kWidth / SUBSAMP_X, \
dst_argb_opt, kWidth * BPP_B, \
kWidth, kHeight); \
src_u, kWidth / SUBSAMP_X, \
src_v, kWidth / SUBSAMP_X, \
dst_argb_opt, kWidth * BPP_B, \
kWidth, kHeight); \
} \
int err = 0; \
for (int i = 0; i < kHeight; ++i) { \
......@@ -108,16 +108,16 @@ TEST_F(libyuvTest, FMT_PLANAR##To##FMT_B##_OptVsC) { \
} \
MaskCpuFlags(kCpuInitialized); \
FMT_PLANAR##To##FMT_B(src_y, kWidth, \
src_uv, kWidth / SUBSAMP_X * 2, \
dst_argb_c, kWidth * BPP_B, \
kWidth, kHeight); \
src_uv, kWidth / SUBSAMP_X * 2, \
dst_argb_c, kWidth * BPP_B, \
kWidth, kHeight); \
MaskCpuFlags(-1); \
const int runs = 1000; \
for (int i = 0; i < runs; ++i) { \
FMT_PLANAR##To##FMT_B(src_y, kWidth, \
src_uv, kWidth / SUBSAMP_X * 2, \
dst_argb_opt, kWidth * BPP_B, \
kWidth, kHeight); \
src_uv, kWidth / SUBSAMP_X * 2, \
dst_argb_opt, kWidth * BPP_B, \
kWidth, kHeight); \
} \
int err = 0; \
for (int i = 0; i < kHeight; ++i) { \
......@@ -158,18 +158,18 @@ TEST_F(libyuvTest, FMT_A##To##FMT_PLANAR##_OptVsC) { \
src_argb[(i * kWidth * BPP_A) + j] = (random() & 0xff); \
MaskCpuFlags(kCpuInitialized); \
FMT_A##To##FMT_PLANAR(src_argb, kWidth * BPP_A, \
dst_y_c, kWidth, \
dst_u_c, kWidth / SUBSAMP_X, \
dst_v_c, kWidth / SUBSAMP_X, \
kWidth, kHeight); \
dst_y_c, kWidth, \
dst_u_c, kWidth / SUBSAMP_X, \
dst_v_c, kWidth / SUBSAMP_X, \
kWidth, kHeight); \
MaskCpuFlags(-1); \
const int runs = 1000; \
for (int i = 0; i < runs; ++i) { \
FMT_A##To##FMT_PLANAR(src_argb, kWidth * BPP_A, \
dst_y_opt, kWidth, \
dst_u_opt, kWidth / SUBSAMP_X, \
dst_v_opt, kWidth / SUBSAMP_X, \
kWidth, kHeight); \
dst_y_opt, kWidth, \
dst_u_opt, kWidth / SUBSAMP_X, \
dst_v_opt, kWidth / SUBSAMP_X, \
kWidth, kHeight); \
} \
int err = 0; \
for (int i = 0; i < kHeight; ++i) { \
......@@ -237,14 +237,14 @@ TEST_F(libyuvTest, FMT_A##To##FMT_B##_OptVsC) { \
} \
MaskCpuFlags(kCpuInitialized); \
FMT_A##To##FMT_B(src_argb, kWidth * STRIDE_A, \
dst_argb_c, kWidth * BPP_B, \
kWidth, kHeight); \
dst_argb_c, kWidth * BPP_B, \
kWidth, kHeight); \
MaskCpuFlags(-1); \
const int runs = 1000; \
for (int i = 0; i < runs; ++i) { \
FMT_A##To##FMT_B(src_argb, kWidth * STRIDE_A, \
dst_argb_opt, kWidth * BPP_B, \
kWidth, kHeight); \
dst_argb_opt, kWidth * BPP_B, \
kWidth, kHeight); \
} \
int err = 0; \
for (int i = 0; i < kHeight * kWidth * BPP_B; ++i) { \
......@@ -294,12 +294,12 @@ TEST_F(libyuvTest, FMT_A##To##FMT_B##_Random) { \
} \
MaskCpuFlags(kCpuInitialized); \
FMT_A##To##FMT_B(src_argb, kWidth * STRIDE_A, \
dst_argb_c, kWidth * BPP_B, \
kWidth, kHeight); \
dst_argb_c, kWidth * BPP_B, \
kWidth, kHeight); \
MaskCpuFlags(-1); \
FMT_A##To##FMT_B(src_argb, kWidth * STRIDE_A, \
dst_argb_opt, kWidth * BPP_B, \
kWidth, kHeight); \
dst_argb_opt, kWidth * BPP_B, \
kWidth, kHeight); \
int err = 0; \
for (int i = 0; i < kHeight * kWidth * BPP_B; ++i) { \
int diff = static_cast<int>(dst_argb_c[i]) - \
......@@ -601,6 +601,65 @@ TEST_F(libyuvTest, TestARGBColorMatrix) {
}
}
TEST_F(libyuvTest, TestARGBColorTable) {
SIMD_ALIGNED(uint8 orig_pixels[256][4]);
// Matrix for Sepia.
static const uint8 kARGBTable[256 * 4] = {
1u, 2u, 3u, 4u,
5u, 6u, 7u, 8u,
9u, 10u, 11u, 12u,
13u, 14u, 15u, 16u,
};
orig_pixels[0][0] = 0u;
orig_pixels[0][1] = 0u;
orig_pixels[0][2] = 0u;
orig_pixels[0][3] = 0u;
orig_pixels[1][0] = 1u;
orig_pixels[1][1] = 1u;
orig_pixels[1][2] = 1u;
orig_pixels[1][3] = 1u;
orig_pixels[2][0] = 2u;
orig_pixels[2][1] = 2u;
orig_pixels[2][2] = 2u;
orig_pixels[2][3] = 2u;
orig_pixels[3][0] = 0u;
orig_pixels[3][1] = 1u;
orig_pixels[3][2] = 2u;
orig_pixels[3][3] = 3u;
// Do 16 to test asm version.
ARGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 16, 1);
EXPECT_EQ(1u, orig_pixels[0][0]);
EXPECT_EQ(2u, orig_pixels[0][1]);
EXPECT_EQ(3u, orig_pixels[0][2]);
EXPECT_EQ(4u, orig_pixels[0][3]);
EXPECT_EQ(5u, orig_pixels[1][0]);
EXPECT_EQ(6u, orig_pixels[1][1]);
EXPECT_EQ(7u, orig_pixels[1][2]);
EXPECT_EQ(8u, orig_pixels[1][3]);
EXPECT_EQ(9u, orig_pixels[2][0]);
EXPECT_EQ(10u, orig_pixels[2][1]);
EXPECT_EQ(11u, orig_pixels[2][2]);
EXPECT_EQ(12u, orig_pixels[2][3]);
EXPECT_EQ(1u, orig_pixels[3][0]);
EXPECT_EQ(6u, orig_pixels[3][1]);
EXPECT_EQ(11u, orig_pixels[3][2]);
EXPECT_EQ(16u, orig_pixels[3][3]);
for (int i = 0; i < 256; ++i) {
orig_pixels[i][0] = i;
orig_pixels[i][1] = i / 2;
orig_pixels[i][2] = i / 3;
orig_pixels[i][3] = i;
}
for (int i = 0; i < 1000 * 1280 * 720 / 256; ++i) {
ARGBColorTable(&orig_pixels[0][0], 0, &kARGBTable[0], 0, 0, 256, 1);
}
}
TEST_F(libyuvTest, TestARGBQuantize) {
SIMD_ALIGNED(uint8 orig_pixels[256][4]);
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment