openh264/test/encoder/EncUT_DecodeMbAux.cpp
Sindre Aamås c8c74903f8 [Encoder] Add single-block AVX2 4x4 DCT/IDCT routines
We do four blocks at a time when possible, but need to handle
single blocks at a time for intra prediction.

~3.15x speedup over MMX for the DCT on Haswell.
~2.94x speedup over MMX for the IDCT on Haswell.

Returns diminish with increasing vector length because a larger
proportion of the time is spent on load/store/shuffling.
2016-02-02 17:22:49 +01:00

379 lines
13 KiB
C++

#include <gtest/gtest.h>
#include "decode_mb_aux.h"
#include "wels_common_basis.h"
#include "macros.h"
#include "cpu.h"
using namespace WelsEnc;
TEST (DecodeMbAuxTest, TestIhdm_4x4_dc) {
short W[16], T[16], Y[16];
for (int i = 0; i < 16; i++)
W[i] = rand() % 256 + 1;
T[0] = W[0] + W[4] + W[8] + W[12];
T[1] = W[1] + W[5] + W[9] + W[13];
T[2] = W[2] + W[6] + W[10] + W[14];
T[3] = W[3] + W[7] + W[11] + W[15];
T[4] = W[0] + W[4] - W[8] - W[12];
T[5] = W[1] + W[5] - W[9] - W[13];
T[6] = W[2] + W[6] - W[10] - W[14];
T[7] = W[3] + W[7] - W[11] - W[15];
T[8] = W[0] - W[4] - W[8] + W[12];
T[9] = W[1] - W[5] - W[9] + W[13];
T[10] = W[2] - W[6] - W[10] + W[14];
T[11] = W[3] - W[7] - W[11] + W[15];
T[12] = W[0] - W[4] + W[8] - W[12];
T[13] = W[1] - W[5] + W[9] - W[13];
T[14] = W[2] - W[6] + W[10] - W[14];
T[15] = W[3] - W[7] + W[11] - W[15];
Y[0] = T[0] + T[1] + T[2] + T[3];
Y[1] = T[0] + T[1] - T[2] - T[3];
Y[2] = T[0] - T[1] - T[2] + T[3];
Y[3] = T[0] - T[1] + T[2] - T[3];
Y[4] = T[4] + T[5] + T[6] + T[7];
Y[5] = T[4] + T[5] - T[6] - T[7];
Y[6] = T[4] - T[5] - T[6] + T[7];
Y[7] = T[4] - T[5] + T[6] - T[7];
Y[8] = T[8] + T[9] + T[10] + T[11];
Y[9] = T[8] + T[9] - T[10] - T[11];
Y[10] = T[8] - T[9] - T[10] + T[11];
Y[11] = T[8] - T[9] + T[10] - T[11];
Y[12] = T[12] + T[13] + T[14] + T[15];
Y[13] = T[12] + T[13] - T[14] - T[15];
Y[14] = T[12] - T[13] - T[14] + T[15];
Y[15] = T[12] - T[13] + T[14] - T[15];
WelsIHadamard4x4Dc (W);
for (int i = 0; i < 16; i++)
EXPECT_EQ (Y[i], W[i]);
}
TEST (DecodeMbAuxTest, TestDequant_4x4_luma_dc) {
short T[16], W[16];
for (int qp = 0; qp < 12; qp++) {
for (int i = 0; i < 16; i++) {
T[i] = rand() % 256 + 1;
W[i] = T[i];
}
WelsDequantLumaDc4x4 (W, qp);
for (int i = 0; i < 16; i++) {
T[i] = (((T[i] * g_kuiDequantCoeff[qp % 6][0] + (1 << (1 - qp / 6)))) >> (2 - qp / 6));
EXPECT_EQ (T[i], W[i]);
}
}
}
TEST (DecodeMbAuxTest, TestDequant_ihdm_4x4_c) {
short W[16], T[16], Y[16];
const unsigned short mf = rand() % 16 + 1;
for (int i = 0; i < 16; i++)
W[i] = rand() % 256 + 1;
T[0] = W[0] + W[4] + W[8] + W[12];
T[1] = W[1] + W[5] + W[9] + W[13];
T[2] = W[2] + W[6] + W[10] + W[14];
T[3] = W[3] + W[7] + W[11] + W[15];
T[4] = W[0] + W[4] - W[8] - W[12];
T[5] = W[1] + W[5] - W[9] - W[13];
T[6] = W[2] + W[6] - W[10] - W[14];
T[7] = W[3] + W[7] - W[11] - W[15];
T[8] = W[0] - W[4] - W[8] + W[12];
T[9] = W[1] - W[5] - W[9] + W[13];
T[10] = W[2] - W[6] - W[10] + W[14];
T[11] = W[3] - W[7] - W[11] + W[15];
T[12] = W[0] - W[4] + W[8] - W[12];
T[13] = W[1] - W[5] + W[9] - W[13];
T[14] = W[2] - W[6] + W[10] - W[14];
T[15] = W[3] - W[7] + W[11] - W[15];
Y[0] = (T[0] + T[1] + T[2] + T[3]) * mf;
Y[1] = (T[0] + T[1] - T[2] - T[3]) * mf;
Y[2] = (T[0] - T[1] - T[2] + T[3]) * mf;
Y[3] = (T[0] - T[1] + T[2] - T[3]) * mf;
Y[4] = (T[4] + T[5] + T[6] + T[7]) * mf;
Y[5] = (T[4] + T[5] - T[6] - T[7]) * mf;
Y[6] = (T[4] - T[5] - T[6] + T[7]) * mf;
Y[7] = (T[4] - T[5] + T[6] - T[7]) * mf;
Y[8] = (T[8] + T[9] + T[10] + T[11]) * mf;
Y[9] = (T[8] + T[9] - T[10] - T[11]) * mf;
Y[10] = (T[8] - T[9] - T[10] + T[11]) * mf;
Y[11] = (T[8] - T[9] + T[10] - T[11]) * mf;
Y[12] = (T[12] + T[13] + T[14] + T[15]) * mf;
Y[13] = (T[12] + T[13] - T[14] - T[15]) * mf;
Y[14] = (T[12] - T[13] - T[14] + T[15]) * mf;
Y[15] = (T[12] - T[13] + T[14] - T[15]) * mf;
WelsDequantIHadamard4x4_c (W, mf);
for (int i = 0; i < 16; i++)
EXPECT_EQ (Y[i], W[i]);
}
TEST (DecodeMbAuxTest, TestDequant_4x4_c) {
short W[16], T[16];
unsigned short mf[16];
for (int i = 0; i < 16; i++) {
W[i] = rand() % 256 + 1;
T[i] = W[i];
}
for (int i = 0; i < 8; i++)
mf[i] = rand() % 16 + 1;
WelsDequant4x4_c (W, mf);
for (int i = 0; i < 16; i++)
EXPECT_EQ (T[i]*mf[i % 8], W[i]);
}
TEST (DecodeMbAuxTest, TestDequant_4_4x4_c) {
short W[64], T[64];
unsigned short mf[16];
for (int i = 0; i < 64; i++) {
W[i] = rand() % 256 + 1;
T[i] = W[i];
}
for (int i = 0; i < 8; i++)
mf[i] = rand() % 16 + 1;
WelsDequantFour4x4_c (W, mf);
for (int i = 0; i < 64; i++)
EXPECT_EQ (T[i]*mf[i % 8], W[i]);
}
void WelsDequantHadamard2x2DcAnchor (int16_t* pDct, int16_t iMF) {
const int16_t iSumU = pDct[0] + pDct[2];
const int16_t iDelU = pDct[0] - pDct[2];
const int16_t iSumD = pDct[1] + pDct[3];
const int16_t iDelD = pDct[1] - pDct[3];
pDct[0] = ((iSumU + iSumD) * iMF) >> 1;
pDct[1] = ((iSumU - iSumD) * iMF) >> 1;
pDct[2] = ((iDelU + iDelD) * iMF) >> 1;
pDct[3] = ((iDelU - iDelD) * iMF) >> 1;
}
TEST (DecodeMbAuxTest, WelsDequantIHadamard2x2Dc) {
int16_t iDct[4], iRefDct[4];
int16_t iMF;
iMF = rand() & 127;
for (int i = 0; i < 4; i++)
iDct[i] = iRefDct[i] = (rand() & 65535) - 32768;
WelsDequantHadamard2x2DcAnchor (iRefDct, iMF);
WelsDequantIHadamard2x2Dc (iDct, iMF);
bool ok = true;
for (int i = 0; i < 4; i++) {
if (iDct[i] != iRefDct[i]) {
ok = false;
break;
}
}
EXPECT_TRUE (ok);
}
#define FDEC_STRIDE 32
template<typename clip_t>
void WelsIDctT4Anchor (uint8_t* p_dst, int16_t dct[16]) {
int16_t tmp[16];
int32_t iStridex2 = (FDEC_STRIDE << 1);
int32_t iStridex3 = iStridex2 + FDEC_STRIDE;
uint8_t uiDst = 0;
int i;
for (i = 0; i < 4; i++) {
tmp[i << 2] = dct[i << 2] + dct[ (i << 2) + 1] + dct[ (i << 2) + 2] + (dct[ (i << 2) + 3] >> 1);
tmp[ (i << 2) + 1] = dct[i << 2] + (dct[ (i << 2) + 1] >> 1) - dct[ (i << 2) + 2] - dct[ (i << 2) + 3];
tmp[ (i << 2) + 2] = dct[i << 2] - (dct[ (i << 2) + 1] >> 1) - dct[ (i << 2) + 2] + dct[ (i << 2) + 3];
tmp[ (i << 2) + 3] = dct[i << 2] - dct[ (i << 2) + 1] + dct[ (i << 2) + 2] - (dct[ (i << 2) + 3] >> 1);
}
for (i = 0; i < 4; i++) {
uiDst = p_dst[i];
p_dst[i] = WelsClip1 (uiDst + (clip_t (tmp[i] + tmp[4 + i] + tmp[8 + i] + (tmp[12 + i] >> 1) + 32) >> 6));
uiDst = p_dst[i + FDEC_STRIDE];
p_dst[i + FDEC_STRIDE] = WelsClip1 (uiDst + (clip_t (tmp[i] + (tmp[4 + i] >> 1) - tmp[8 + i] - tmp[12 + i] + 32) >> 6));
uiDst = p_dst[i + iStridex2];
p_dst[i + iStridex2] = WelsClip1 (uiDst + (clip_t (tmp[i] - (tmp[4 + i] >> 1) - tmp[8 + i] + tmp[12 + i] + 32) >> 6));
uiDst = p_dst[i + iStridex3];
p_dst[i + iStridex3] = WelsClip1 (uiDst + (clip_t (tmp[i] - tmp[4 + i] + tmp[8 + i] - (tmp[12 + i] >> 1) + 32) >> 6));
}
}
template<typename clip_t>
void TestIDctT4Rec (PIDctFunc func) {
int16_t iRefDct[16];
uint8_t iRefDst[16 * FDEC_STRIDE];
ENFORCE_STACK_ALIGN_1D (int16_t, iDct, 16, 16);
ENFORCE_STACK_ALIGN_1D (uint8_t, iPred, 16 * FDEC_STRIDE, 16);
ENFORCE_STACK_ALIGN_1D (uint8_t, iRec, 16 * FDEC_STRIDE, 16);
for (int i = 0; i < 4; i++) {
for (int j = 0; j < 4; j++) {
iRefDct[i * 4 + j] = iDct[i * 4 + j] = (rand() & 65535) - 32768;
iPred[i * FDEC_STRIDE + j] = iRefDst[i * FDEC_STRIDE + j] = rand() & 255;
}
}
WelsIDctT4Anchor<clip_t> (iRefDst, iRefDct);
func (iRec, FDEC_STRIDE, iPred, FDEC_STRIDE, iDct);
int ok = -1;
for (int i = 0; i < 4; i++) {
for (int j = 0; j < 4; j++) {
if (iRec[i * FDEC_STRIDE + j] != iRefDst[i * FDEC_STRIDE + j]) {
ok = i * 4 + j;
break;
}
}
}
EXPECT_EQ (ok, -1);
}
TEST (DecodeMbAuxTest, WelsIDctT4Rec_c) {
TestIDctT4Rec<int32_t> (WelsIDctT4Rec_c);
}
#if defined(X86_ASM)
TEST (DecodeMbAuxTest, WelsIDctT4Rec_mmx) {
TestIDctT4Rec<int16_t> (WelsIDctT4Rec_mmx);
}
TEST (DecodeMbAuxTest, WelsIDctT4Rec_sse2) {
TestIDctT4Rec<int16_t> (WelsIDctT4Rec_sse2);
}
TEST (DecodeMbAuxTest, WelsIDctT4Rec_avx2) {
if (WelsCPUFeatureDetect (0) & WELS_CPU_AVX2)
TestIDctT4Rec<int16_t> (WelsIDctT4Rec_avx2);
}
#endif
template<typename clip_t>
void WelsIDctT8Anchor (uint8_t* p_dst, int16_t dct[4][16]) {
WelsIDctT4Anchor<clip_t> (&p_dst[0], dct[0]);
WelsIDctT4Anchor<clip_t> (&p_dst[4], dct[1]);
WelsIDctT4Anchor<clip_t> (&p_dst[4 * FDEC_STRIDE + 0], dct[2]);
WelsIDctT4Anchor<clip_t> (&p_dst[4 * FDEC_STRIDE + 4], dct[3]);
}
template<typename clip_t>
void TestIDctFourT4Rec (PIDctFunc func) {
int16_t iRefDct[4][16];
uint8_t iRefDst[16 * FDEC_STRIDE];
ENFORCE_STACK_ALIGN_1D (int16_t, iDct, 64, 16);
ENFORCE_STACK_ALIGN_1D (uint8_t, iPred, 16 * FDEC_STRIDE, 16);
ENFORCE_STACK_ALIGN_1D (uint8_t, iRec, 16 * FDEC_STRIDE, 16);
for (int k = 0; k < 4; k++)
for (int i = 0; i < 16; i++)
iRefDct[k][i] = iDct[k * 16 + i] = (rand() & 65535) - 32768;
for (int i = 0; i < 8; i++)
for (int j = 0; j < 8; j++)
iPred[i * FDEC_STRIDE + j] = iRefDst[i * FDEC_STRIDE + j] = rand() & 255;
WelsIDctT8Anchor<clip_t> (iRefDst, iRefDct);
func (iRec, FDEC_STRIDE, iPred, FDEC_STRIDE, iDct);
int ok = -1;
for (int i = 0; i < 8; i++) {
for (int j = 0; j < 8; j++) {
if (iRec[i * FDEC_STRIDE + j] != iRefDst[i * FDEC_STRIDE + j]) {
ok = i * 8 + j;
break;
}
}
}
EXPECT_EQ (ok, -1);
}
TEST (DecodeMbAuxTest, WelsIDctFourT4Rec_c) {
TestIDctFourT4Rec<int32_t> (WelsIDctFourT4Rec_c);
}
void WelsIDctRecI16x4DcAnchor (uint8_t* p_dst, int16_t dct[4]) {
for (int i = 0; i < 4; i++, p_dst += FDEC_STRIDE) {
p_dst[0] = WelsClip1 (p_dst[0] + ((dct[0] + 32) >> 6));
p_dst[1] = WelsClip1 (p_dst[1] + ((dct[0] + 32) >> 6));
p_dst[2] = WelsClip1 (p_dst[2] + ((dct[0] + 32) >> 6));
p_dst[3] = WelsClip1 (p_dst[3] + ((dct[0] + 32) >> 6));
p_dst[4] = WelsClip1 (p_dst[4] + ((dct[1] + 32) >> 6));
p_dst[5] = WelsClip1 (p_dst[5] + ((dct[1] + 32) >> 6));
p_dst[6] = WelsClip1 (p_dst[6] + ((dct[1] + 32) >> 6));
p_dst[7] = WelsClip1 (p_dst[7] + ((dct[1] + 32) >> 6));
p_dst[8] = WelsClip1 (p_dst[8] + ((dct[2] + 32) >> 6));
p_dst[9] = WelsClip1 (p_dst[9] + ((dct[2] + 32) >> 6));
p_dst[10] = WelsClip1 (p_dst[10] + ((dct[2] + 32) >> 6));
p_dst[11] = WelsClip1 (p_dst[11] + ((dct[2] + 32) >> 6));
p_dst[12] = WelsClip1 (p_dst[12] + ((dct[3] + 32) >> 6));
p_dst[13] = WelsClip1 (p_dst[13] + ((dct[3] + 32) >> 6));
p_dst[14] = WelsClip1 (p_dst[14] + ((dct[3] + 32) >> 6));
p_dst[15] = WelsClip1 (p_dst[15] + ((dct[3] + 32) >> 6));
}
}
void WelsIDctRecI16x16DcAnchor (uint8_t* p_dst, int16_t dct[4][4]) {
for (int i = 0; i < 4; i++, p_dst += 4 * FDEC_STRIDE)
WelsIDctRecI16x4DcAnchor (&p_dst[0], dct[i]);
}
TEST (DecodeMbAuxTest, WelsIDctRecI16x16Dc_c) {
uint8_t iRefDst[16 * FDEC_STRIDE];
int16_t iRefDct[4][4];
ENFORCE_STACK_ALIGN_1D (int16_t, iDct, 16, 16);
ENFORCE_STACK_ALIGN_1D (uint8_t, iPred, 16 * FDEC_STRIDE, 16);
ENFORCE_STACK_ALIGN_1D (uint8_t, iRec, 16 * FDEC_STRIDE, 16);
for (int i = 0; i < 16; i++)
for (int j = 0; j < 16; j++)
iRefDst[i * FDEC_STRIDE + j] = iPred[i * FDEC_STRIDE + j] = rand() & 255;
for (int i = 0; i < 4; i++)
for (int j = 0; j < 4; j++)
iRefDct[i][j] = iDct[i * 4 + j] = (rand() & 65535) - 32768;
WelsIDctRecI16x16DcAnchor (iRefDst, iRefDct);
WelsIDctRecI16x16Dc_c (iRec, FDEC_STRIDE, iPred, FDEC_STRIDE, iDct);
int ok = -1;
for (int i = 0; i < 16; i++) {
for (int j = 0; j < 16; j++) {
if (iRec[i * FDEC_STRIDE + j] != iRefDst[i * FDEC_STRIDE + j]) {
ok = i * 16 + j;
break;
}
}
}
EXPECT_EQ (ok, -1);
}
#if defined(X86_ASM)
TEST (DecodeMbAuxTest, WelsIDctFourT4Rec_sse2) {
TestIDctFourT4Rec<int16_t> (WelsIDctFourT4Rec_sse2);
}
TEST (DecodeMbAuxTest, WelsIDctFourT4Rec_avx2) {
if (WelsCPUFeatureDetect (0) & WELS_CPU_AVX2)
TestIDctFourT4Rec<int16_t> (WelsIDctFourT4Rec_avx2);
}
TEST (DecodeMbAuxTest, WelsIDctRecI16x16Dc_sse2) {
int32_t iCpuCores = 0;
uint32_t uiCpuFeatureFlag = WelsCPUFeatureDetect (&iCpuCores);
if (uiCpuFeatureFlag & WELS_CPU_SSE2) {
uint8_t iRefDst[16 * FDEC_STRIDE];
int16_t iRefDct[4][4];
ENFORCE_STACK_ALIGN_1D (int16_t, iDct, 16, 16);
ENFORCE_STACK_ALIGN_1D (uint8_t, iPred, 16 * FDEC_STRIDE, 16);
ENFORCE_STACK_ALIGN_1D (uint8_t, iRec, 16 * FDEC_STRIDE, 16);
for (int i = 0; i < 16; i++)
for (int j = 0; j < 16; j++)
iRefDst[i * FDEC_STRIDE + j] = iPred[i * FDEC_STRIDE + j] = rand() & 255;
for (int i = 0; i < 4; i++)
for (int j = 0; j < 4; j++)
iRefDct[i][j] = iDct[i * 4 + j] = (rand() & ((1 << 15) - 1)) - (1 <<
14); //2^14 limit, (2^15+32) will cause overflow for SSE2.
WelsIDctRecI16x16DcAnchor (iRefDst, iRefDct);
WelsIDctRecI16x16Dc_sse2 (iRec, FDEC_STRIDE, iPred, FDEC_STRIDE, iDct);
int ok = -1;
for (int i = 0; i < 16; i++) {
for (int j = 0; j < 16; j++) {
if (iRec[i * FDEC_STRIDE + j] != iRefDst[i * FDEC_STRIDE + j]) {
ok = i * 16 + j;
break;
}
}
}
EXPECT_EQ (ok, -1);
}
}
#endif