diff options
Diffstat (limited to 'media/libaom/src/test/wiener_test.cc')
| -rw-r--r-- | media/libaom/src/test/wiener_test.cc | 425 |
1 files changed, 366 insertions, 59 deletions
diff --git a/media/libaom/src/test/wiener_test.cc b/media/libaom/src/test/wiener_test.cc index dfec09119d..81839fd56f 100644 --- a/media/libaom/src/test/wiener_test.cc +++ b/media/libaom/src/test/wiener_test.cc @@ -9,36 +9,38 @@ * PATENTS file, you can obtain it at www.aomedia.org/license/patent. */ +#include <tuple> #include <vector> #include "third_party/googletest/src/googletest/include/gtest/gtest.h" -#include "test/function_equivalence_test.h" #include "test/register_state_check.h" +#include "test/acm_random.h" +#include "test/util.h" #include "config/aom_config.h" #include "config/aom_dsp_rtcd.h" #include "aom/aom_integer.h" +#include "aom_ports/aom_timer.h" #include "av1/encoder/pickrst.h" #define MAX_WIENER_BLOCK 384 #define MAX_DATA_BLOCK (MAX_WIENER_BLOCK + WIENER_WIN) -using libaom_test::FunctionEquivalenceTest; -namespace { +// 8-bit-depth tests +namespace wiener_lowbd { static void compute_stats_win_opt_c(int wiener_win, const uint8_t *dgd, const uint8_t *src, int h_start, int h_end, int v_start, int v_end, int dgd_stride, - int src_stride, double *M, double *H) { + int src_stride, int64_t *M, int64_t *H) { ASSERT_TRUE(wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA); int i, j, k, l, m, n; const int pixel_count = (h_end - h_start) * (v_end - v_start); const int wiener_win2 = wiener_win * wiener_win; const int wiener_halfwin = (wiener_win >> 1); - const double avg = - find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride); + uint8_t avg = find_average(dgd, h_start, h_end, v_start, v_end, dgd_stride); std::vector<std::vector<int64_t> > M_int(wiener_win, std::vector<int64_t>(wiener_win, 0)); @@ -75,16 +77,16 @@ static void compute_stats_win_opt_c(int wiener_win, const uint8_t *dgd, } } - const double avg_square_sum = avg * avg * pixel_count; + const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count; for (k = 0; k < wiener_win; k++) { for (l = 0; l < wiener_win; l++) { M[l * wiener_win + k] = - M_int[l][k] + avg_square_sum - avg * (sumX + sumY[k][l]); + M_int[l][k] + avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]); for (m = 0; m < wiener_win; m++) { for (n = 0; n < wiener_win; n++) { H[(l * wiener_win + k) * wiener_win2 + m * wiener_win + n] = H_int[(l * wiener_win + k)][n * 8 + m] + avg_square_sum - - avg * (sumY[k][l] + sumY[n][m]); + (int64_t)avg * (sumY[k][l] + sumY[n][m]); } } } @@ -93,7 +95,8 @@ static void compute_stats_win_opt_c(int wiener_win, const uint8_t *dgd, void compute_stats_opt_c(int wiener_win, const uint8_t *dgd, const uint8_t *src, int h_start, int h_end, int v_start, int v_end, - int dgd_stride, int src_stride, double *M, double *H) { + int dgd_stride, int src_stride, int64_t *M, + int64_t *H) { if (wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA) { compute_stats_win_opt_c(wiener_win, dgd, src, h_start, h_end, v_start, v_end, dgd_stride, src_stride, M, H); @@ -104,40 +107,47 @@ void compute_stats_opt_c(int wiener_win, const uint8_t *dgd, const uint8_t *src, } static const int kIterations = 100; -static const double min_error = (double)(0.01); typedef void (*compute_stats_Func)(int wiener_win, const uint8_t *dgd, const uint8_t *src, int h_start, int h_end, int v_start, int v_end, int dgd_stride, - int src_stride, double *M, double *H); - -typedef libaom_test::FuncParam<compute_stats_Func> TestFuncs; + int src_stride, int64_t *M, int64_t *H); //////////////////////////////////////////////////////////////////////////////// // 8 bit //////////////////////////////////////////////////////////////////////////////// -typedef ::testing::tuple<const compute_stats_Func> WienerTestParam; +typedef std::tuple<const compute_stats_Func> WienerTestParam; class WienerTest : public ::testing::TestWithParam<WienerTestParam> { public: - virtual void SetUp() { target_func_ = GET_PARAM(0); } - void runWienerTest(const int32_t wiener_win, int32_t run_times); - void runWienerTest_ExtremeValues(const int32_t wiener_win); + virtual void SetUp() { + src_buf = (uint8_t *)aom_memalign( + 32, MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*src_buf)); + dgd_buf = (uint8_t *)aom_memalign( + 32, MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*dgd_buf)); + target_func_ = GET_PARAM(0); + } + virtual void TearDown() { + aom_free(src_buf); + aom_free(dgd_buf); + } + void RunWienerTest(const int32_t wiener_win, int32_t run_times); + void RunWienerTest_ExtremeValues(const int32_t wiener_win); private: compute_stats_Func target_func_; - ACMRandom rng_; + libaom_test::ACMRandom rng_; + uint8_t *src_buf; + uint8_t *dgd_buf; }; -void WienerTest::runWienerTest(const int32_t wiener_win, int32_t run_times) { +void WienerTest::RunWienerTest(const int32_t wiener_win, int32_t run_times) { const int32_t wiener_halfwin = wiener_win >> 1; const int32_t wiener_win2 = wiener_win * wiener_win; - DECLARE_ALIGNED(32, uint8_t, dgd_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]); - DECLARE_ALIGNED(32, uint8_t, src_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]); - DECLARE_ALIGNED(32, double, M_ref[WIENER_WIN2]); - DECLARE_ALIGNED(32, double, H_ref[WIENER_WIN2 * WIENER_WIN2]); - DECLARE_ALIGNED(32, double, M_test[WIENER_WIN2]); - DECLARE_ALIGNED(32, double, H_test[WIENER_WIN2 * WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, M_ref[WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, H_ref[WIENER_WIN2 * WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, M_test[WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, H_test[WIENER_WIN2 * WIENER_WIN2]); const int h_start = ((rng_.Rand16() % (MAX_WIENER_BLOCK / 2)) & (~7)); int h_end = run_times != 1 ? 256 : ((rng_.Rand16() % MAX_WIENER_BLOCK) & (~7)) + 8; @@ -177,19 +187,18 @@ void WienerTest::runWienerTest(const int32_t wiener_win, int32_t run_times) { } int failed = 0; for (int i = 0; i < wiener_win2; ++i) { - if (fabs(M_ref[i] - M_test[i]) > min_error) { + if (M_ref[i] != M_test[i]) { failed = 1; - printf("win %d M iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win, - iter, i, M_ref[i], M_test[i]); + printf("win %d M iter %d [%4d] ref %6" PRId64 " test %6" PRId64 " \n", + wiener_win, iter, i, M_ref[i], M_test[i]); break; } } - // ASSERT_EQ(failed, 0); for (int i = 0; i < wiener_win2 * wiener_win2; ++i) { - if (fabs(H_ref[i] - H_test[i]) > min_error) { + if (H_ref[i] != H_test[i]) { failed = 1; - printf("win %d H iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win, - iter, i, H_ref[i], H_test[i]); + printf("win %d H iter %d [%4d] ref %6" PRId64 " test %6" PRId64 " \n", + wiener_win, iter, i, H_ref[i], H_test[i]); break; } } @@ -197,15 +206,13 @@ void WienerTest::runWienerTest(const int32_t wiener_win, int32_t run_times) { } } -void WienerTest::runWienerTest_ExtremeValues(const int32_t wiener_win) { +void WienerTest::RunWienerTest_ExtremeValues(const int32_t wiener_win) { const int32_t wiener_halfwin = wiener_win >> 1; const int32_t wiener_win2 = wiener_win * wiener_win; - DECLARE_ALIGNED(32, uint8_t, dgd_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]); - DECLARE_ALIGNED(32, uint8_t, src_buf[MAX_DATA_BLOCK * MAX_DATA_BLOCK]); - DECLARE_ALIGNED(32, double, M_ref[WIENER_WIN2]); - DECLARE_ALIGNED(32, double, H_ref[WIENER_WIN2 * WIENER_WIN2]); - DECLARE_ALIGNED(32, double, M_test[WIENER_WIN2]); - DECLARE_ALIGNED(32, double, H_test[WIENER_WIN2 * WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, M_ref[WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, H_ref[WIENER_WIN2 * WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, M_test[WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, H_test[WIENER_WIN2 * WIENER_WIN2]); const int h_start = 16; const int h_end = MAX_WIENER_BLOCK; const int v_start = 16; @@ -229,19 +236,18 @@ void WienerTest::runWienerTest_ExtremeValues(const int32_t wiener_win) { int failed = 0; for (int i = 0; i < wiener_win2; ++i) { - if (fabs(M_ref[i] - M_test[i]) > min_error) { + if (M_ref[i] != M_test[i]) { failed = 1; - printf("win %d M iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win, - iter, i, M_ref[i], M_test[i]); + printf("win %d M iter %d [%4d] ref %6" PRId64 " test %6" PRId64 " \n", + wiener_win, iter, i, M_ref[i], M_test[i]); break; } } - // ASSERT_EQ(failed, 0); for (int i = 0; i < wiener_win2 * wiener_win2; ++i) { - if (fabs(H_ref[i] - H_test[i]) > min_error) { + if (H_ref[i] != H_test[i]) { failed = 1; - printf("win %d H iter %d [%4d] ref %6.0f test %6.0f \n", wiener_win, - iter, i, H_ref[i], H_test[i]); + printf("win %d H iter %d [%4d] ref %6" PRId64 " test %6" PRId64 " \n", + wiener_win, iter, i, H_ref[i], H_test[i]); break; } } @@ -250,31 +256,332 @@ void WienerTest::runWienerTest_ExtremeValues(const int32_t wiener_win) { } TEST_P(WienerTest, RandomValues) { - runWienerTest(WIENER_WIN, 1); - runWienerTest(WIENER_WIN_CHROMA, 1); + RunWienerTest(WIENER_WIN, 1); + RunWienerTest(WIENER_WIN_CHROMA, 1); } TEST_P(WienerTest, ExtremeValues) { - runWienerTest_ExtremeValues(WIENER_WIN); - runWienerTest_ExtremeValues(WIENER_WIN_CHROMA); + RunWienerTest_ExtremeValues(WIENER_WIN); + RunWienerTest_ExtremeValues(WIENER_WIN_CHROMA); } TEST_P(WienerTest, DISABLED_Speed) { - runWienerTest(WIENER_WIN, 200); - runWienerTest(WIENER_WIN_CHROMA, 200); + RunWienerTest(WIENER_WIN, 200); + RunWienerTest(WIENER_WIN_CHROMA, 200); } -INSTANTIATE_TEST_CASE_P(C, WienerTest, ::testing::Values(compute_stats_opt_c)); +INSTANTIATE_TEST_SUITE_P(C, WienerTest, ::testing::Values(compute_stats_opt_c)); #if HAVE_SSE4_1 -INSTANTIATE_TEST_CASE_P(SSE4_1, WienerTest, - ::testing::Values(av1_compute_stats_sse4_1)); +INSTANTIATE_TEST_SUITE_P(SSE4_1, WienerTest, + ::testing::Values(av1_compute_stats_sse4_1)); #endif // HAVE_SSE4_1 #if HAVE_AVX2 -INSTANTIATE_TEST_CASE_P(AVX2, WienerTest, - ::testing::Values(av1_compute_stats_avx2)); +INSTANTIATE_TEST_SUITE_P(AVX2, WienerTest, + ::testing::Values(av1_compute_stats_avx2)); +#endif // HAVE_AVX2 + +} // namespace wiener_lowbd + +#if CONFIG_AV1_HIGHBITDEPTH +// High bit-depth tests: +namespace wiener_highbd { + +static void compute_stats_highbd_win_opt_c(int wiener_win, const uint8_t *dgd8, + const uint8_t *src8, int h_start, + int h_end, int v_start, int v_end, + int dgd_stride, int src_stride, + int64_t *M, int64_t *H, + aom_bit_depth_t bit_depth) { + ASSERT_TRUE(wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA); + int i, j, k, l, m, n; + const int pixel_count = (h_end - h_start) * (v_end - v_start); + const int wiener_win2 = wiener_win * wiener_win; + const int wiener_halfwin = (wiener_win >> 1); + const uint16_t *src = CONVERT_TO_SHORTPTR(src8); + const uint16_t *dgd = CONVERT_TO_SHORTPTR(dgd8); + const uint16_t avg = + find_average_highbd(dgd, h_start, h_end, v_start, v_end, dgd_stride); + + std::vector<std::vector<int64_t> > M_int(wiener_win, + std::vector<int64_t>(wiener_win, 0)); + std::vector<std::vector<int64_t> > H_int( + wiener_win * wiener_win, std::vector<int64_t>(wiener_win * 8, 0)); + std::vector<std::vector<int32_t> > sumY(wiener_win, + std::vector<int32_t>(wiener_win, 0)); + + memset(M, 0, sizeof(*M) * wiener_win2); + memset(H, 0, sizeof(*H) * wiener_win2 * wiener_win2); + + int64_t sumX = 0; + const uint16_t *dgd_win = dgd - wiener_halfwin * dgd_stride - wiener_halfwin; + + for (i = v_start; i < v_end; i++) { + for (j = h_start; j < h_end; j += 2) { + const uint16_t X1 = src[i * src_stride + j]; + const uint16_t X2 = src[i * src_stride + j + 1]; + sumX += X1 + X2; + + const uint16_t *dgd_ij = dgd_win + i * dgd_stride + j; + for (k = 0; k < wiener_win; k++) { + for (l = 0; l < wiener_win; l++) { + const uint16_t *dgd_ijkl = dgd_ij + k * dgd_stride + l; + int64_t *H_int_temp = &H_int[(l * wiener_win + k)][0]; + const uint16_t D1 = dgd_ijkl[0]; + const uint16_t D2 = dgd_ijkl[1]; + sumY[k][l] += D1 + D2; + M_int[l][k] += D1 * X1 + D2 * X2; + for (m = 0; m < wiener_win; m++) { + for (n = 0; n < wiener_win; n++) { + H_int_temp[m * 8 + n] += D1 * dgd_ij[n + dgd_stride * m] + + D2 * dgd_ij[n + dgd_stride * m + 1]; + } + } + } + } + } + } + + uint8_t bit_depth_divider = 1; + if (bit_depth == AOM_BITS_12) + bit_depth_divider = 16; + else if (bit_depth == AOM_BITS_10) + bit_depth_divider = 4; + + const int64_t avg_square_sum = (int64_t)avg * (int64_t)avg * pixel_count; + for (k = 0; k < wiener_win; k++) { + for (l = 0; l < wiener_win; l++) { + M[l * wiener_win + k] = + (M_int[l][k] + + (avg_square_sum - (int64_t)avg * (sumX + sumY[k][l]))) / + bit_depth_divider; + for (m = 0; m < wiener_win; m++) { + for (n = 0; n < wiener_win; n++) { + H[(l * wiener_win + k) * wiener_win2 + m * wiener_win + n] = + (H_int[(l * wiener_win + k)][n * 8 + m] + + (avg_square_sum - (int64_t)avg * (sumY[k][l] + sumY[n][m]))) / + bit_depth_divider; + } + } + } + } +} + +void compute_stats_highbd_opt_c(int wiener_win, const uint8_t *dgd, + const uint8_t *src, int h_start, int h_end, + int v_start, int v_end, int dgd_stride, + int src_stride, int64_t *M, int64_t *H, + aom_bit_depth_t bit_depth) { + if (wiener_win == WIENER_WIN || wiener_win == WIENER_WIN_CHROMA) { + compute_stats_highbd_win_opt_c(wiener_win, dgd, src, h_start, h_end, + v_start, v_end, dgd_stride, src_stride, M, H, + bit_depth); + } else { + av1_compute_stats_highbd_c(wiener_win, dgd, src, h_start, h_end, v_start, + v_end, dgd_stride, src_stride, M, H, bit_depth); + } +} + +static const int kIterations = 100; +typedef void (*compute_stats_Func)(int wiener_win, const uint8_t *dgd, + const uint8_t *src, int h_start, int h_end, + int v_start, int v_end, int dgd_stride, + int src_stride, int64_t *M, int64_t *H, + aom_bit_depth_t bit_depth); + +typedef std::tuple<const compute_stats_Func> WienerTestParam; + +class WienerTestHighbd : public ::testing::TestWithParam<WienerTestParam> { + public: + virtual void SetUp() { + src_buf = (uint16_t *)aom_memalign( + 32, MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*src_buf)); + dgd_buf = (uint16_t *)aom_memalign( + 32, MAX_DATA_BLOCK * MAX_DATA_BLOCK * sizeof(*dgd_buf)); + target_func_ = GET_PARAM(0); + } + virtual void TearDown() { + aom_free(src_buf); + aom_free(dgd_buf); + } + void RunWienerTest(const int32_t wiener_win, int32_t run_times, + aom_bit_depth_t bit_depth); + void RunWienerTest_ExtremeValues(const int32_t wiener_win, + aom_bit_depth_t bit_depth); + + private: + compute_stats_Func target_func_; + libaom_test::ACMRandom rng_; + uint16_t *src_buf; + uint16_t *dgd_buf; +}; + +void WienerTestHighbd::RunWienerTest(const int32_t wiener_win, + int32_t run_times, + aom_bit_depth_t bit_depth) { + const int32_t wiener_halfwin = wiener_win >> 1; + const int32_t wiener_win2 = wiener_win * wiener_win; + DECLARE_ALIGNED(32, int64_t, M_ref[WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, H_ref[WIENER_WIN2 * WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, M_test[WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, H_test[WIENER_WIN2 * WIENER_WIN2]); + const int h_start = ((rng_.Rand16() % (MAX_WIENER_BLOCK / 2)) & (~7)); + const int h_end = + run_times != 1 ? 256 : ((rng_.Rand16() % MAX_WIENER_BLOCK) & (~7)) + 8; + const int v_start = ((rng_.Rand16() % (MAX_WIENER_BLOCK / 2)) & (~7)); + const int v_end = + run_times != 1 ? 256 : ((rng_.Rand16() % MAX_WIENER_BLOCK) & (~7)) + 8; + const int dgd_stride = h_end; + const int src_stride = MAX_DATA_BLOCK; + const int iters = run_times == 1 ? kIterations : 2; + for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) { + for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) { + dgd_buf[i] = rng_.Rand16() % (1 << bit_depth); + src_buf[i] = rng_.Rand16() % (1 << bit_depth); + } + const uint8_t *dgd8 = CONVERT_TO_BYTEPTR( + dgd_buf + wiener_halfwin * MAX_DATA_BLOCK + wiener_halfwin); + const uint8_t *src8 = CONVERT_TO_BYTEPTR(src_buf); + + aom_usec_timer timer; + aom_usec_timer_start(&timer); + for (int i = 0; i < run_times; ++i) { + av1_compute_stats_highbd_c(wiener_win, dgd8, src8, h_start, h_end, + v_start, v_end, dgd_stride, src_stride, M_ref, + H_ref, bit_depth); + } + aom_usec_timer_mark(&timer); + const double time1 = static_cast<double>(aom_usec_timer_elapsed(&timer)); + aom_usec_timer_start(&timer); + for (int i = 0; i < run_times; ++i) { + target_func_(wiener_win, dgd8, src8, h_start, h_end, v_start, v_end, + dgd_stride, src_stride, M_test, H_test, bit_depth); + } + aom_usec_timer_mark(&timer); + const double time2 = static_cast<double>(aom_usec_timer_elapsed(&timer)); + if (run_times > 10) { + printf("win %d bd %d %3dx%-3d:%7.2f/%7.2fns", wiener_win, bit_depth, + h_end, v_end, time1, time2); + printf("(%3.2f)\n", time1 / time2); + } + int failed = 0; + for (int i = 0; i < wiener_win2; ++i) { + if (M_ref[i] != M_test[i]) { + failed = 1; + printf("win %d bd %d M iter %d [%4d] ref %6" PRId64 " test %6" PRId64 + " \n", + wiener_win, bit_depth, iter, i, M_ref[i], M_test[i]); + break; + } + } + for (int i = 0; i < wiener_win2 * wiener_win2; ++i) { + if (H_ref[i] != H_test[i]) { + failed = 1; + printf("win %d bd %d H iter %d [%4d] ref %6" PRId64 " test %6" PRId64 + " \n", + wiener_win, bit_depth, iter, i, H_ref[i], H_test[i]); + break; + } + } + ASSERT_EQ(failed, 0); + } +} + +void WienerTestHighbd::RunWienerTest_ExtremeValues(const int32_t wiener_win, + aom_bit_depth_t bit_depth) { + const int32_t wiener_halfwin = wiener_win >> 1; + const int32_t wiener_win2 = wiener_win * wiener_win; + DECLARE_ALIGNED(32, int64_t, M_ref[WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, H_ref[WIENER_WIN2 * WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, M_test[WIENER_WIN2]); + DECLARE_ALIGNED(32, int64_t, H_test[WIENER_WIN2 * WIENER_WIN2]); + const int h_start = 16; + const int h_end = MAX_WIENER_BLOCK; + const int v_start = 16; + const int v_end = MAX_WIENER_BLOCK; + const int dgd_stride = h_end; + const int src_stride = MAX_DATA_BLOCK; + const int iters = 1; + for (int iter = 0; iter < iters && !HasFatalFailure(); ++iter) { + for (int i = 0; i < MAX_DATA_BLOCK * MAX_DATA_BLOCK; ++i) { + dgd_buf[i] = ((uint16_t)1 << bit_depth) - 1; + src_buf[i] = ((uint16_t)1 << bit_depth) - 1; + } + const uint8_t *dgd8 = CONVERT_TO_BYTEPTR( + dgd_buf + wiener_halfwin * MAX_DATA_BLOCK + wiener_halfwin); + const uint8_t *src8 = CONVERT_TO_BYTEPTR(src_buf); + + av1_compute_stats_highbd_c(wiener_win, dgd8, src8, h_start, h_end, v_start, + v_end, dgd_stride, src_stride, M_ref, H_ref, + bit_depth); + + target_func_(wiener_win, dgd8, src8, h_start, h_end, v_start, v_end, + dgd_stride, src_stride, M_test, H_test, bit_depth); + + int failed = 0; + for (int i = 0; i < wiener_win2; ++i) { + if (M_ref[i] != M_test[i]) { + failed = 1; + printf("win %d bd %d M iter %d [%4d] ref %6" PRId64 " test %6" PRId64 + " \n", + wiener_win, bit_depth, iter, i, M_ref[i], M_test[i]); + break; + } + } + for (int i = 0; i < wiener_win2 * wiener_win2; ++i) { + if (H_ref[i] != H_test[i]) { + failed = 1; + printf("win %d bd %d H iter %d [%4d] ref %6" PRId64 " test %6" PRId64 + " \n", + wiener_win, bit_depth, iter, i, H_ref[i], H_test[i]); + break; + } + } + ASSERT_EQ(failed, 0); + } +} + +TEST_P(WienerTestHighbd, RandomValues) { + RunWienerTest(WIENER_WIN, 1, AOM_BITS_8); + RunWienerTest(WIENER_WIN_CHROMA, 1, AOM_BITS_8); + RunWienerTest(WIENER_WIN, 1, AOM_BITS_10); + RunWienerTest(WIENER_WIN_CHROMA, 1, AOM_BITS_10); + RunWienerTest(WIENER_WIN, 1, AOM_BITS_12); + RunWienerTest(WIENER_WIN_CHROMA, 1, AOM_BITS_12); +} + +TEST_P(WienerTestHighbd, ExtremeValues) { + RunWienerTest_ExtremeValues(WIENER_WIN, AOM_BITS_8); + RunWienerTest_ExtremeValues(WIENER_WIN_CHROMA, AOM_BITS_8); + RunWienerTest_ExtremeValues(WIENER_WIN, AOM_BITS_10); + RunWienerTest_ExtremeValues(WIENER_WIN_CHROMA, AOM_BITS_10); + RunWienerTest_ExtremeValues(WIENER_WIN, AOM_BITS_12); + RunWienerTest_ExtremeValues(WIENER_WIN_CHROMA, AOM_BITS_12); +} + +TEST_P(WienerTestHighbd, DISABLED_Speed) { + RunWienerTest(WIENER_WIN, 200, AOM_BITS_8); + RunWienerTest(WIENER_WIN_CHROMA, 200, AOM_BITS_8); + RunWienerTest(WIENER_WIN, 200, AOM_BITS_10); + RunWienerTest(WIENER_WIN_CHROMA, 200, AOM_BITS_10); + RunWienerTest(WIENER_WIN, 200, AOM_BITS_12); + RunWienerTest(WIENER_WIN_CHROMA, 200, AOM_BITS_12); +} + +INSTANTIATE_TEST_SUITE_P(C, WienerTestHighbd, + ::testing::Values(compute_stats_highbd_opt_c)); + +#if HAVE_SSE4_1 +INSTANTIATE_TEST_SUITE_P(SSE4_1, WienerTestHighbd, + ::testing::Values(av1_compute_stats_highbd_sse4_1)); +#endif // HAVE_SSE4_1 + +#if HAVE_AVX2 +INSTANTIATE_TEST_SUITE_P(AVX2, WienerTestHighbd, + ::testing::Values(av1_compute_stats_highbd_avx2)); #endif // HAVE_AVX2 -} // namespace +} // namespace wiener_highbd +#endif // CONFIG_AV1_HIGHBITDEPTH |