From 7369c7d7a5eed32963d8af37658286617919f91c Mon Sep 17 00:00:00 2001 From: trav90 Date: Thu, 18 Oct 2018 06:04:57 -0500 Subject: Update aom to commit id f5bdeac22930ff4c6b219be49c843db35970b918 --- third_party/aom/test/quantize_func_test.cc | 288 ++++++++++++++++++++++------- 1 file changed, 220 insertions(+), 68 deletions(-) (limited to 'third_party/aom/test/quantize_func_test.cc') diff --git a/third_party/aom/test/quantize_func_test.cc b/third_party/aom/test/quantize_func_test.cc index 32b1d5139..94dd056b4 100644 --- a/third_party/aom/test/quantize_func_test.cc +++ b/third_party/aom/test/quantize_func_test.cc @@ -12,8 +12,10 @@ #include "third_party/googletest/src/googletest/include/gtest/gtest.h" #include "./aom_config.h" +#include "./aom_dsp_rtcd.h" #include "./av1_rtcd.h" #include "aom/aom_codec.h" +#include "aom_ports/aom_timer.h" #include "av1/encoder/encoder.h" #include "av1/encoder/av1_quantize.h" #include "test/acm_random.h" @@ -24,26 +26,44 @@ namespace { using libaom_test::ACMRandom; -#if !CONFIG_AOM_QM -typedef void (*QuantizeFunc)(const tran_low_t *coeff_ptr, intptr_t n_coeffs, - int skip_block, const int16_t *zbin_ptr, - const int16_t *round_ptr, const int16_t *quant_ptr, - const int16_t *quant_shift_ptr, - tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, - const int16_t *dequant_ptr, uint16_t *eob_ptr, - const int16_t *scan, const int16_t *iscan); -#else -typedef void (*QuantizeFunc)(const tran_low_t *coeff_ptr, intptr_t n_coeffs, - int skip_block, const int16_t *zbin_ptr, - const int16_t *round_ptr, const int16_t *quant_ptr, - const int16_t *quant_shift_ptr, - tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, - const int16_t *dequant_ptr, uint16_t *eob_ptr, - const int16_t *scan, const int16_t *iscan, - const qm_val_t *qm_ptr, const qm_val_t *iqm_ptr); -#endif +#define QUAN_PARAM_LIST \ + const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, \ + const int16_t *zbin_ptr, const int16_t *round_ptr, \ + const int16_t *quant_ptr, const int16_t *quant_shift_ptr, \ + tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, \ + const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, \ + const int16_t *iscan + +typedef void (*QuantizeFunc)(QUAN_PARAM_LIST); +typedef void (*QuantizeFuncHbd)(QUAN_PARAM_LIST, int log_scale); + +#define HBD_QUAN_FUNC \ + fn(coeff_ptr, n_coeffs, skip_block, zbin_ptr, round_ptr, quant_ptr, \ + quant_shift_ptr, qcoeff_ptr, dqcoeff_ptr, dequant_ptr, eob_ptr, scan, \ + iscan, log_scale) + +template +void highbd_quan16x16_wrapper(QUAN_PARAM_LIST) { + const int log_scale = 0; + HBD_QUAN_FUNC; +} -typedef std::tr1::tuple +template +void highbd_quan32x32_wrapper(QUAN_PARAM_LIST) { + const int log_scale = 1; + HBD_QUAN_FUNC; +} + +template +void highbd_quan64x64_wrapper(QUAN_PARAM_LIST) { + const int log_scale = 2; + HBD_QUAN_FUNC; +} + +typedef enum { TYPE_B, TYPE_DC, TYPE_FP } QuantType; + +typedef std::tr1::tuple QuantizeParam; typedef struct { @@ -57,15 +77,15 @@ class QuantizeTest : public ::testing::TestWithParam { protected: QuantizeTest() : quant_ref_(GET_PARAM(0)), quant_(GET_PARAM(1)), tx_size_(GET_PARAM(2)), - bd_(GET_PARAM(3)) {} + type_(GET_PARAM(3)), bd_(GET_PARAM(4)) {} virtual ~QuantizeTest() {} virtual void SetUp() { - qtab_ = reinterpret_cast(aom_memalign(16, sizeof(*qtab_))); - const int n_coeffs = getCoeffNum(); + qtab_ = reinterpret_cast(aom_memalign(32, sizeof(*qtab_))); + const int n_coeffs = coeff_num(); coeff_ = reinterpret_cast( - aom_memalign(16, 6 * n_coeffs * sizeof(tran_low_t))); + aom_memalign(32, 6 * n_coeffs * sizeof(tran_low_t))); InitQuantizer(); } @@ -81,9 +101,9 @@ class QuantizeTest : public ::testing::TestWithParam { av1_build_quantizer(bd_, 0, 0, 0, &qtab_->quant, &qtab_->dequant); } - void QuantizeRun(bool isLoop, int q = 0, int testNum = 1) { + void QuantizeRun(bool is_loop, int q = 0, int test_num = 1) { tran_low_t *coeff_ptr = coeff_; - const intptr_t n_coeffs = getCoeffNum(); + const intptr_t n_coeffs = coeff_num(); const int skip_block = 0; tran_low_t *qcoeff_ref = coeff_ptr + n_coeffs; @@ -98,31 +118,47 @@ class QuantizeTest : public ::testing::TestWithParam { // Testing uses luminance quantization table const int16_t *zbin = qtab_->quant.y_zbin[q]; - const int16_t *round_fp = qtab_->quant.y_round_fp[q]; - const int16_t *quant_fp = qtab_->quant.y_quant_fp[q]; + + const int16_t *round = 0; + const int16_t *quant = 0; + if (type_ == TYPE_B) { + round = qtab_->quant.y_round[q]; + quant = qtab_->quant.y_quant[q]; + } else if (type_ == TYPE_FP) { + round = qtab_->quant.y_round_fp[q]; + quant = qtab_->quant.y_quant_fp[q]; + } + const int16_t *quant_shift = qtab_->quant.y_quant_shift[q]; const int16_t *dequant = qtab_->dequant.y_dequant[q]; - const size_t bufferSize = n_coeffs; - int i = 0; - while (i < testNum) { - if (isLoop) FillCoeffRandom(); + for (int i = 0; i < test_num; ++i) { + if (is_loop) FillCoeffRandom(); memset(qcoeff_ref, 0, 5 * n_coeffs * sizeof(*qcoeff_ref)); - quant_ref_(coeff_ptr, n_coeffs, skip_block, zbin, round_fp, quant_fp, + quant_ref_(coeff_ptr, n_coeffs, skip_block, zbin, round, quant, quant_shift, qcoeff_ref, dqcoeff_ref, dequant, &eob[0], sc->scan, sc->iscan); ASM_REGISTER_STATE_CHECK(quant_( - coeff_ptr, n_coeffs, skip_block, zbin, round_fp, quant_fp, - quant_shift, qcoeff, dqcoeff, dequant, &eob[1], sc->scan, sc->iscan)); + coeff_ptr, n_coeffs, skip_block, zbin, round, quant, quant_shift, + qcoeff, dqcoeff, dequant, &eob[1], sc->scan, sc->iscan)); - CompareResults(qcoeff_ref, qcoeff, bufferSize, "Qcoeff", q, i); - CompareResults(dqcoeff_ref, dqcoeff, bufferSize, "Dqcoeff", q, i); - ASSERT_EQ(eob[0], eob[1]) << "eobs mismatch on test: " << i; + for (int j = 0; j < n_coeffs; ++j) { + ASSERT_EQ(qcoeff_ref[j], qcoeff[j]) + << "Q mismatch on test: " << i << " at position: " << j + << " Q: " << q << " coeff: " << coeff_ptr[j]; + } - i++; + for (int j = 0; j < n_coeffs; ++j) { + ASSERT_EQ(dqcoeff_ref[j], dqcoeff[j]) + << "Dq mismatch on test: " << i << " at position: " << j + << " Q: " << q << " coeff: " << coeff_ptr[j]; + } + + ASSERT_EQ(eob[0], eob[1]) << "eobs mismatch on test: " << i + << " Q: " << q; } } @@ -135,29 +171,29 @@ class QuantizeTest : public ::testing::TestWithParam { } } - int getCoeffNum() { return tx_size_2d[tx_size_]; } + int coeff_num() const { return tx_size_2d[tx_size_]; } - void FillCoeffGeneric(bool isConstant, tran_low_t c = 0) { - const int n_coeffs = getCoeffNum(); - int i; - if (isConstant) { - for (i = 0; i < n_coeffs; ++i) { - coeff_[i] = c; - } - } else { - FillCoeffZero(); - int num = rnd_.Rand16() % n_coeffs; - for (i = 0; i < num; ++i) { - coeff_[i] = GetRandomCoeff(); - } + void FillCoeff(tran_low_t c) { + const int n_coeffs = coeff_num(); + for (int i = 0; i < n_coeffs; ++i) { + coeff_[i] = c; } } - void FillCoeffZero() { FillCoeffGeneric(true); } + void FillCoeffRandom() { + const int n_coeffs = coeff_num(); + FillCoeffZero(); + int num = rnd_.Rand16() % n_coeffs; + for (int i = 0; i < num; ++i) { + coeff_[i] = GetRandomCoeff(); + } + } + + void FillCoeffZero() { FillCoeff(0); } void FillCoeffConstant() { tran_low_t c = GetRandomCoeff(); - FillCoeffGeneric(true, c); + FillCoeff(c); } void FillDcOnly() { @@ -173,10 +209,17 @@ class QuantizeTest : public ::testing::TestWithParam { coeff_[0] = -8191; } - void FillCoeffRandom() { FillCoeffGeneric(false); } - tran_low_t GetRandomCoeff() { - return clamp((int16_t)rnd_.Rand16(), INT16_MIN + 1, INT16_MAX); + tran_low_t coeff; + if (bd_ == AOM_BITS_8) { + coeff = + clamp(static_cast(rnd_.Rand16()), INT16_MIN + 1, INT16_MAX); + } else { + tran_low_t min = -(1 << (7 + bd_)); + tran_low_t max = -min - 1; + coeff = clamp(static_cast(rnd_.Rand31()), min, max); + } + return coeff; } ACMRandom rnd_; @@ -185,6 +228,7 @@ class QuantizeTest : public ::testing::TestWithParam { QuantizeFunc quant_ref_; QuantizeFunc quant_; TX_SIZE tx_size_; + QuantType type_; aom_bit_depth_t bd_; }; @@ -195,12 +239,12 @@ TEST_P(QuantizeTest, ZeroInput) { TEST_P(QuantizeTest, LargeNegativeInput) { FillDcLargeNegative(); - QuantizeRun(false); + QuantizeRun(false, 0, 1); } TEST_P(QuantizeTest, DcOnlyInput) { FillDcOnly(); - QuantizeRun(false); + QuantizeRun(false, 0, 1); } TEST_P(QuantizeTest, RandomInput) { QuantizeRun(true, 0, kTestNum); } @@ -211,26 +255,134 @@ TEST_P(QuantizeTest, MultipleQ) { } } +TEST_P(QuantizeTest, DISABLED_Speed) { + tran_low_t *coeff_ptr = coeff_; + const intptr_t n_coeffs = coeff_num(); + const int skip_block = 0; + + tran_low_t *qcoeff_ref = coeff_ptr + n_coeffs; + tran_low_t *dqcoeff_ref = qcoeff_ref + n_coeffs; + + tran_low_t *qcoeff = dqcoeff_ref + n_coeffs; + tran_low_t *dqcoeff = qcoeff + n_coeffs; + uint16_t *eob = (uint16_t *)(dqcoeff + n_coeffs); + + // Testing uses 2-D DCT scan order table + const SCAN_ORDER *const sc = get_default_scan(tx_size_, DCT_DCT, 0); + + // Testing uses luminance quantization table + const int q = 22; + const int16_t *zbin = qtab_->quant.y_zbin[q]; + const int16_t *round_fp = qtab_->quant.y_round_fp[q]; + const int16_t *quant_fp = qtab_->quant.y_quant_fp[q]; + const int16_t *quant_shift = qtab_->quant.y_quant_shift[q]; + const int16_t *dequant = qtab_->dequant.y_dequant[q]; + const int kNumTests = 5000000; + aom_usec_timer timer; + + FillCoeffRandom(); + + aom_usec_timer_start(&timer); + for (int n = 0; n < kNumTests; ++n) { + quant_(coeff_ptr, n_coeffs, skip_block, zbin, round_fp, quant_fp, + quant_shift, qcoeff, dqcoeff, dequant, eob, sc->scan, sc->iscan); + } + aom_usec_timer_mark(&timer); + + const int elapsed_time = static_cast(aom_usec_timer_elapsed(&timer)); + printf("Elapsed time: %d us\n", elapsed_time); +} + using std::tr1::make_tuple; +#if HAVE_AVX2 +const QuantizeParam kQParamArrayAvx2[] = { + make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_avx2, TX_16X16, TYPE_FP, + AOM_BITS_8), + make_tuple(&av1_quantize_fp_32x32_c, &av1_quantize_fp_32x32_avx2, TX_32X32, + TYPE_FP, AOM_BITS_8), +#if CONFIG_HIGHBITDEPTH + make_tuple(&highbd_quan16x16_wrapper, + &highbd_quan16x16_wrapper, TX_16X16, + TYPE_FP, AOM_BITS_8), + make_tuple(&highbd_quan16x16_wrapper, + &highbd_quan16x16_wrapper, TX_16X16, + TYPE_FP, AOM_BITS_10), + make_tuple(&highbd_quan16x16_wrapper, + &highbd_quan16x16_wrapper, TX_16X16, + TYPE_FP, AOM_BITS_12), + make_tuple(&highbd_quan32x32_wrapper, + &highbd_quan32x32_wrapper, TX_32X32, + TYPE_FP, AOM_BITS_8), + make_tuple(&highbd_quan32x32_wrapper, + &highbd_quan32x32_wrapper, TX_32X32, + TYPE_FP, AOM_BITS_10), + make_tuple(&highbd_quan32x32_wrapper, + &highbd_quan32x32_wrapper, TX_32X32, + TYPE_FP, AOM_BITS_12), +#if CONFIG_TX64X64 + make_tuple(&highbd_quan64x64_wrapper, + &highbd_quan64x64_wrapper, TX_64X64, + TYPE_FP, AOM_BITS_8), + make_tuple(&highbd_quan64x64_wrapper, + &highbd_quan64x64_wrapper, TX_64X64, + TYPE_FP, AOM_BITS_10), + make_tuple(&highbd_quan64x64_wrapper, + &highbd_quan64x64_wrapper, TX_64X64, + TYPE_FP, AOM_BITS_12), +#endif // CONFIG_TX64X64 + make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_avx2, TX_16X16, + TYPE_B, AOM_BITS_8), + make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_avx2, TX_16X16, + TYPE_B, AOM_BITS_10), + make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_avx2, TX_16X16, + TYPE_B, AOM_BITS_12), +#endif // CONFIG_HIGHBITDEPTH +}; + +INSTANTIATE_TEST_CASE_P(AVX2, QuantizeTest, + ::testing::ValuesIn(kQParamArrayAvx2)); +#endif // HAVE_AVX2 + #if HAVE_SSE2 -const QuantizeParam kQParamArraySSE2[] = { make_tuple( - &av1_quantize_fp_c, &av1_quantize_fp_sse2, TX_16X16, AOM_BITS_8) }; +const QuantizeParam kQParamArraySSE2[] = { + make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_sse2, TX_16X16, TYPE_FP, + AOM_BITS_8), +#if CONFIG_HIGHBITDEPTH + make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_sse2, TX_16X16, + TYPE_B, AOM_BITS_8), + make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_sse2, TX_16X16, + TYPE_B, AOM_BITS_10), + make_tuple(&aom_highbd_quantize_b_c, &aom_highbd_quantize_b_sse2, TX_16X16, + TYPE_B, AOM_BITS_12), + make_tuple(&aom_highbd_quantize_b_32x32_c, &aom_highbd_quantize_b_32x32_sse2, + TX_32X32, TYPE_B, AOM_BITS_8), + make_tuple(&aom_highbd_quantize_b_32x32_c, &aom_highbd_quantize_b_32x32_sse2, + TX_32X32, TYPE_B, AOM_BITS_10), + make_tuple(&aom_highbd_quantize_b_32x32_c, &aom_highbd_quantize_b_32x32_sse2, + TX_32X32, TYPE_B, AOM_BITS_12), +#endif +}; INSTANTIATE_TEST_CASE_P(SSE2, QuantizeTest, ::testing::ValuesIn(kQParamArraySSE2)); #endif #if !CONFIG_HIGHBITDEPTH && HAVE_SSSE3 && ARCH_X86_64 -const QuantizeParam kQParamArraySSSE3[] = { - make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_ssse3, TX_16X16, AOM_BITS_8), - // TODO(any): - // The following test couldn't pass yet - // make_tuple(av1_quantize_fp_c, av1_quantize_fp_32x32_ssse3, TX_32X32, - // AOM_BITS_8) +const QuantizeParam kQ16x16ParamArraySSSE3[] = { + make_tuple(&av1_quantize_fp_c, &av1_quantize_fp_ssse3, TX_16X16, TYPE_FP, + AOM_BITS_8), }; INSTANTIATE_TEST_CASE_P(SSSE3, QuantizeTest, - ::testing::ValuesIn(kQParamArraySSSE3)); + ::testing::ValuesIn(kQ16x16ParamArraySSSE3)); + +// TODO(any): +// The following test does not pass yet +const QuantizeParam kQ32x32ParamArraySSSE3[] = { make_tuple( + &av1_quantize_fp_32x32_c, &av1_quantize_fp_32x32_ssse3, TX_32X32, TYPE_FP, + AOM_BITS_8) }; +INSTANTIATE_TEST_CASE_P(DISABLED_SSSE3, QuantizeTest, + ::testing::ValuesIn(kQ32x32ParamArraySSSE3)); #endif } // namespace -- cgit v1.2.3