Add av1_get_br_level_counts_sse2() Change-Id: I6ce7aea19e3bdeef24d3fe66ac6eba7b8d585f9a
diff --git a/av1/av1.cmake b/av1/av1.cmake index 6ea0cfd..804941a 100644 --- a/av1/av1.cmake +++ b/av1/av1.cmake
@@ -254,7 +254,8 @@ set(AOM_AV1_COMMON_INTRIN_SSE2 ${AOM_AV1_COMMON_INTRIN_SSE2} - "${AOM_ROOT}/av1/common/cdef_block_sse2.c") + "${AOM_ROOT}/av1/common/cdef_block_sse2.c" + "${AOM_ROOT}/av1/common/x86/mem_sse2.h") set(AOM_AV1_COMMON_INTRIN_SSSE3 ${AOM_AV1_COMMON_INTRIN_SSSE3} @@ -384,6 +385,10 @@ "${AOM_ROOT}/av1/common/txb_common.c" "${AOM_ROOT}/av1/common/txb_common.h") + set(AOM_AV1_COMMON_INTRIN_SSE2 + ${AOM_AV1_COMMON_INTRIN_SSE2} + "${AOM_ROOT}/av1/common/x86/txb_sse2.c") + set(AOM_AV1_DECODER_SOURCES ${AOM_AV1_DECODER_SOURCES} "${AOM_ROOT}/av1/decoder/decodetxb.c"
diff --git a/av1/av1_common.mk b/av1/av1_common.mk index f0c65c2..ed5a7ec 100644 --- a/av1/av1_common.mk +++ b/av1/av1_common.mk
@@ -76,6 +76,7 @@ AV1_COMMON_SRCS-yes += common/av1_inv_txfm2d.c AV1_COMMON_SRCS-yes += common/av1_inv_txfm1d_cfg.h AV1_COMMON_SRCS-$(HAVE_AVX2) += common/x86/convolve_avx2.c +AV1_COMMON_SRCS-$(HAVE_SSE2) += common/x86/mem_sse2.h AV1_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/av1_convolve_ssse3.c ifeq ($(CONFIG_CONVOLVE_ROUND)x$(CONFIG_COMPOUND_ROUND),yesx) AV1_COMMON_SRCS-$(HAVE_SSE4_1) += common/x86/av1_convolve_scale_sse4.c @@ -167,6 +168,10 @@ endif endif +ifeq ($(CONFIG_LV_MAP),yes) +AV1_COMMON_SRCS-$(HAVE_SSE2) += common/x86/txb_sse2.c +endif + ifeq ($(CONFIG_Q_ADAPT_PROBS),yes) AV1_COMMON_SRCS-yes += common/token_cdfs.h endif
diff --git a/av1/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl index ab0fa48..6b2f1ed 100755 --- a/av1/common/av1_rtcd_defs.pl +++ b/av1/common/av1_rtcd_defs.pl
@@ -67,6 +67,14 @@ } # +# txb +# +if (aom_config("CONFIG_LV_MAP") eq "yes") { + add_proto qw/void av1_get_br_level_counts/, "const uint8_t *const levels, const int width, const int height, uint8_t *const level_counts"; + specialize qw/av1_get_br_level_counts sse2/; +} + +# # Inverse dct # add_proto qw/void av1_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, const struct txfm_param *param";
diff --git a/av1/common/txb_common.c b/av1/common/txb_common.c index 38ac984..4607428 100644 --- a/av1/common/txb_common.c +++ b/av1/common/txb_common.c
@@ -333,8 +333,8 @@ } } -void av1_get_br_level_counts(const uint8_t *const levels, const int width, - const int height, uint8_t *const level_counts) { +void av1_get_br_level_counts_c(const uint8_t *const levels, const int width, + const int height, uint8_t *const level_counts) { const int stride = width + TX_PAD_HOR; const int level_minus_1 = NUM_BASE_LEVELS;
diff --git a/av1/common/txb_common.h b/av1/common/txb_common.h index c609f29..edae7f7 100644 --- a/av1/common/txb_common.h +++ b/av1/common/txb_common.h
@@ -391,7 +391,7 @@ const int width = 1 << bwl; int ctx = 0; - int tx_class = get_tx_class(tx_type); + const TX_CLASS tx_class = get_tx_class(tx_type); int offset; if (tx_class == TX_CLASS_2D) offset = 0; @@ -455,7 +455,7 @@ const int row = coeff_idx >> bwl; const int col = coeff_idx - (row << bwl); - int tx_class = get_tx_class(tx_type); + const TX_CLASS tx_class = get_tx_class(tx_type); #if USE_CAUSAL_BASE_CTX int mag = 0; int count = get_nz_count_mag(levels, bwl, row, col, tx_class, &mag); @@ -578,7 +578,4 @@ const int level_minus_1, const int width, const int height, uint8_t *const level_counts); -void av1_get_br_level_counts(const uint8_t *const levels, const int width, - const int height, uint8_t *const level_counts); - #endif // AV1_COMMON_TXB_COMMON_H_
diff --git a/av1/common/x86/mem_sse2.h b/av1/common/x86/mem_sse2.h new file mode 100644 index 0000000..432fd99 --- /dev/null +++ b/av1/common/x86/mem_sse2.h
@@ -0,0 +1,25 @@ +/* + * Copyright (c) 2017, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#ifndef AV1_COMMON_X86_MEM_SSE2_H_ +#define AV1_COMMON_X86_MEM_SSE2_H_ + +#include <emmintrin.h> // SSE2 + +#include "./aom_config.h" +#include "aom/aom_integer.h" + +static INLINE __m128i loadh_epi64(const void *const src, const __m128i s) { + return _mm_castps_si128( + _mm_loadh_pi(_mm_castsi128_ps(s), (const __m64 *)src)); +} + +#endif // AV1_COMMON_X86_MEM_SSE2_H_
diff --git a/av1/common/x86/txb_sse2.c b/av1/common/x86/txb_sse2.c new file mode 100644 index 0000000..ebe4e8c --- /dev/null +++ b/av1/common/x86/txb_sse2.c
@@ -0,0 +1,248 @@ +/* + * Copyright (c) 2017, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include <assert.h> +#include <emmintrin.h> // SSE2 + +#include "aom/aom_integer.h" +#include "av1/common/onyxc_int.h" +#include "av1/common/txb_common.h" +#include "av1/common/x86/mem_sse2.h" + +static INLINE __m128i load_8bit_4x4_sse2(const void *const s0, + const void *const s1, + const void *const s2, + const void *const s3) { + return _mm_setr_epi32(*(const uint32_t *)s0, *(const uint32_t *)s1, + *(const uint32_t *)s2, *(const uint32_t *)s3); +} + +static INLINE void load_levels_4x4x3_sse2(const uint8_t *const s0, + const uint8_t *const s1, + const uint8_t *const s2, + const uint8_t *const s3, + __m128i *const level) { + level[0] = load_8bit_4x4_sse2(s0 - 1, s1 - 1, s2 - 1, s3 - 1); + level[1] = load_8bit_4x4_sse2(s0 + 0, s1 + 0, s2 + 0, s3 + 0); + level[2] = load_8bit_4x4_sse2(s0 + 1, s1 + 1, s2 + 1, s3 + 1); +} + +static INLINE __m128i get_level_counts_kernel_sse2(__m128i *const level) { + const __m128i level_minus_1 = _mm_set1_epi8(NUM_BASE_LEVELS); + __m128i count; + + level[6] = _mm_cmpgt_epi8(level[6], level_minus_1); + level[7] = _mm_cmpgt_epi8(level[7], level_minus_1); + level[8] = _mm_cmpgt_epi8(level[8], level_minus_1); + count = _mm_setzero_si128(); + count = _mm_sub_epi8(count, level[0]); + count = _mm_sub_epi8(count, level[1]); + count = _mm_sub_epi8(count, level[2]); + count = _mm_sub_epi8(count, level[3]); + count = _mm_sub_epi8(count, level[5]); + count = _mm_sub_epi8(count, level[6]); + count = _mm_sub_epi8(count, level[7]); + count = _mm_sub_epi8(count, level[8]); + level[0] = level[3]; + level[1] = level[4]; + level[2] = level[5]; + level[3] = level[6]; + level[4] = level[7]; + level[5] = level[8]; + + return count; +} + +static INLINE void get_4_level_counts_sse2(const uint8_t *const levels, + const int height, + uint8_t *const level_counts) { + const int stride = 4 + TX_PAD_HOR; + const __m128i level_minus_1 = _mm_set1_epi8(NUM_BASE_LEVELS); + int row = height; + __m128i count; + __m128i level[9]; + + /* level_counts must be 16 byte aligned. */ + assert(!((intptr_t)level_counts & 0xf)); + assert(!(height % 4)); + + if (height == 4) { + load_levels_4x4x3_sse2(levels - 1 * stride, levels + 0 * stride, + levels + 1 * stride, levels + 2 * stride, &level[0]); + load_levels_4x4x3_sse2(levels + 0 * stride, levels + 1 * stride, + levels + 2 * stride, levels + 3 * stride, &level[3]); + load_levels_4x4x3_sse2(levels + 1 * stride, levels + 2 * stride, + levels + 3 * stride, levels + 4 * stride, &level[6]); + level[0] = _mm_cmpgt_epi8(level[0], level_minus_1); + level[1] = _mm_cmpgt_epi8(level[1], level_minus_1); + level[2] = _mm_cmpgt_epi8(level[2], level_minus_1); + level[3] = _mm_cmpgt_epi8(level[3], level_minus_1); + level[4] = _mm_cmpgt_epi8(level[4], level_minus_1); + level[5] = _mm_cmpgt_epi8(level[5], level_minus_1); + count = get_level_counts_kernel_sse2(level); + _mm_store_si128((__m128i *)level_counts, count); + } else { + const uint8_t *ls[4]; + uint8_t *lcs[4]; + + ls[0] = levels + 0 * stride * height / 4; + ls[1] = levels + 1 * stride * height / 4; + ls[2] = levels + 2 * stride * height / 4; + ls[3] = levels + 3 * stride * height / 4; + lcs[0] = level_counts + 0 * 4 * height / 4; + lcs[1] = level_counts + 1 * 4 * height / 4; + lcs[2] = level_counts + 2 * 4 * height / 4; + lcs[3] = level_counts + 3 * 4 * height / 4; + + load_levels_4x4x3_sse2(ls[0] - 1 * stride, ls[1] - 1 * stride, + ls[2] - 1 * stride, ls[3] - 1 * stride, &level[0]); + load_levels_4x4x3_sse2(ls[0] + 0 * stride, ls[1] + 0 * stride, + ls[2] + 0 * stride, ls[3] + 0 * stride, &level[3]); + level[0] = _mm_cmpgt_epi8(level[0], level_minus_1); + level[1] = _mm_cmpgt_epi8(level[1], level_minus_1); + level[2] = _mm_cmpgt_epi8(level[2], level_minus_1); + level[3] = _mm_cmpgt_epi8(level[3], level_minus_1); + level[4] = _mm_cmpgt_epi8(level[4], level_minus_1); + level[5] = _mm_cmpgt_epi8(level[5], level_minus_1); + + do { + load_levels_4x4x3_sse2(ls[0] + 1 * stride, ls[1] + 1 * stride, + ls[2] + 1 * stride, ls[3] + 1 * stride, &level[6]); + + count = get_level_counts_kernel_sse2(level); + *(int *)(lcs[0]) = _mm_cvtsi128_si32(count); + *(int *)(lcs[1]) = _mm_cvtsi128_si32(_mm_srli_si128(count, 4)); + *(int *)(lcs[2]) = _mm_cvtsi128_si32(_mm_srli_si128(count, 8)); + *(int *)(lcs[3]) = _mm_cvtsi128_si32(_mm_srli_si128(count, 12)); + ls[0] += stride; + ls[1] += stride; + ls[2] += stride; + ls[3] += stride; + lcs[0] += 4; + lcs[1] += 4; + lcs[2] += 4; + lcs[3] += 4; + row -= 4; + } while (row); + } +} + +static INLINE void get_8_level_counts_sse2(const uint8_t *const levels, + const int height, + uint8_t *const level_counts) { + const int stride = 8 + TX_PAD_HOR; + const __m128i level_minus_1 = _mm_set1_epi8(NUM_BASE_LEVELS); + int row = height; + __m128i count; + __m128i level[9]; + const uint8_t *ls[2]; + uint8_t *lcs[2]; + + assert(!(height % 2)); + + ls[0] = levels; + ls[1] = levels + stride * height / 2; + lcs[0] = level_counts; + lcs[1] = level_counts + 8 * height / 2; + + level[0] = _mm_loadl_epi64((__m128i *)(ls[0] - 1 * stride - 1)); + level[1] = _mm_loadl_epi64((__m128i *)(ls[0] - 1 * stride + 0)); + level[2] = _mm_loadl_epi64((__m128i *)(ls[0] - 1 * stride + 1)); + level[3] = _mm_loadl_epi64((__m128i *)(ls[0] + 0 * stride - 1)); + level[4] = _mm_loadl_epi64((__m128i *)(ls[0] + 0 * stride + 0)); + level[5] = _mm_loadl_epi64((__m128i *)(ls[0] + 0 * stride + 1)); + level[0] = loadh_epi64(ls[1] - 1 * stride - 1, level[0]); + level[1] = loadh_epi64(ls[1] - 1 * stride + 0, level[1]); + level[2] = loadh_epi64(ls[1] - 1 * stride + 1, level[2]); + level[3] = loadh_epi64(ls[1] + 0 * stride - 1, level[3]); + level[4] = loadh_epi64(ls[1] + 0 * stride + 0, level[4]); + level[5] = loadh_epi64(ls[1] + 0 * stride + 1, level[5]); + level[0] = _mm_cmpgt_epi8(level[0], level_minus_1); + level[1] = _mm_cmpgt_epi8(level[1], level_minus_1); + level[2] = _mm_cmpgt_epi8(level[2], level_minus_1); + level[3] = _mm_cmpgt_epi8(level[3], level_minus_1); + level[4] = _mm_cmpgt_epi8(level[4], level_minus_1); + level[5] = _mm_cmpgt_epi8(level[5], level_minus_1); + + do { + level[6] = _mm_loadl_epi64((__m128i *)(ls[0] + 1 * stride - 1)); + level[7] = _mm_loadl_epi64((__m128i *)(ls[0] + 1 * stride + 0)); + level[8] = _mm_loadl_epi64((__m128i *)(ls[0] + 1 * stride + 1)); + level[6] = loadh_epi64(ls[1] + 1 * stride - 1, level[6]); + level[7] = loadh_epi64(ls[1] + 1 * stride + 0, level[7]); + level[8] = loadh_epi64(ls[1] + 1 * stride + 1, level[8]); + + count = get_level_counts_kernel_sse2(level); + _mm_storel_epi64((__m128i *)(lcs[0]), count); + _mm_storeh_pi((__m64 *)(lcs[1]), _mm_castsi128_ps(count)); + ls[0] += stride; + ls[1] += stride; + lcs[0] += 8; + lcs[1] += 8; + row -= 2; + } while (row); +} + +static INLINE void get_16x_level_counts_sse2(const uint8_t *levels, + const int width, const int height, + uint8_t *level_counts) { + const int stride = width + TX_PAD_HOR; + const __m128i level_minus_1 = _mm_set1_epi8(NUM_BASE_LEVELS); + __m128i count; + __m128i level[9]; + + /* level_counts must be 16 byte aligned. */ + assert(!((intptr_t)level_counts & 0xf)); + assert(!(width % 16)); + + for (int i = 0; i < width; i += 16) { + int row = height; + + level[0] = _mm_loadu_si128((__m128i *)(levels + i - 1 * stride - 1)); + level[1] = _mm_loadu_si128((__m128i *)(levels + i - 1 * stride + 0)); + level[2] = _mm_loadu_si128((__m128i *)(levels + i - 1 * stride + 1)); + level[3] = _mm_loadu_si128((__m128i *)(levels + i + 0 * stride - 1)); + level[4] = _mm_loadu_si128((__m128i *)(levels + i + 0 * stride + 0)); + level[5] = _mm_loadu_si128((__m128i *)(levels + i + 0 * stride + 1)); + level[0] = _mm_cmpgt_epi8(level[0], level_minus_1); + level[1] = _mm_cmpgt_epi8(level[1], level_minus_1); + level[2] = _mm_cmpgt_epi8(level[2], level_minus_1); + level[3] = _mm_cmpgt_epi8(level[3], level_minus_1); + level[4] = _mm_cmpgt_epi8(level[4], level_minus_1); + level[5] = _mm_cmpgt_epi8(level[5], level_minus_1); + + do { + level[6] = _mm_loadu_si128((__m128i *)(levels + i + 1 * stride - 1)); + level[7] = _mm_loadu_si128((__m128i *)(levels + i + 1 * stride + 0)); + level[8] = _mm_loadu_si128((__m128i *)(levels + i + 1 * stride + 1)); + + count = get_level_counts_kernel_sse2(level); + _mm_store_si128((__m128i *)(level_counts + i), count); + levels += stride; + level_counts += width; + } while (--row); + + levels -= stride * height; + level_counts -= width * height; + } +} + +void av1_get_br_level_counts_sse2(const uint8_t *const levels, const int width, + const int height, + uint8_t *const level_counts) { + if (width == 4) { + get_4_level_counts_sse2(levels, height, level_counts); + } else if (width == 8) { + get_8_level_counts_sse2(levels, height, level_counts); + } else { + get_16x_level_counts_sse2(levels, width, height, level_counts); + } +}
diff --git a/av1/decoder/decodetxb.c b/av1/decoder/decodetxb.c index edcb0aa..ff19776 100644 --- a/av1/decoder/decodetxb.c +++ b/av1/decoder/decodetxb.c
@@ -9,6 +9,7 @@ * PATENTS file, you can obtain it at www.aomedia.org/license/patent. */ +#include "aom_ports/mem.h" #include "av1/common/scan.h" #include "av1/common/idct.h" #include "av1/common/txb_common.h" @@ -74,7 +75,7 @@ int cul_level = 0; uint8_t levels_buf[TX_PAD_2D]; uint8_t *const levels = set_levels(levels_buf, width); - uint8_t level_counts[MAX_TX_SQUARE]; + DECLARE_ALIGNED(16, uint8_t, level_counts[MAX_TX_SQUARE]); int8_t signs[MAX_TX_SQUARE]; memset(tcoeffs, 0, sizeof(*tcoeffs) * seg_eob);
diff --git a/av1/encoder/encodetxb.c b/av1/encoder/encodetxb.c index d459f89..21d4a1d 100644 --- a/av1/encoder/encodetxb.c +++ b/av1/encoder/encodetxb.c
@@ -9,6 +9,7 @@ * PATENTS file, you can obtain it at www.aomedia.org/license/patent. */ +#include "aom_ports/mem.h" #include "av1/common/scan.h" #include "av1/common/blockd.h" #include "av1/common/idct.h" @@ -321,7 +322,7 @@ FRAME_CONTEXT *ec_ctx = xd->tile_ctx; uint8_t levels_buf[TX_PAD_2D]; uint8_t *const levels = set_levels(levels_buf, width); - uint8_t level_counts[MAX_TX_SQUARE]; + DECLARE_ALIGNED(16, uint8_t, level_counts[MAX_TX_SQUARE]); (void)blk_row; (void)blk_col; @@ -635,7 +636,7 @@ const int16_t *scan = scan_order->scan; uint8_t levels_buf[TX_PAD_2D]; uint8_t *const levels = set_levels(levels_buf, width); - uint8_t level_counts[MAX_TX_SQUARE]; + DECLARE_ALIGNED(16, uint8_t, level_counts[MAX_TX_SQUARE]); LV_MAP_COEFF_COST *coeff_costs = &x->coeff_costs[txs_ctx][plane_type]; @@ -2144,7 +2145,7 @@ const int height = tx_size_high[tx_size]; uint8_t levels_buf[TX_PAD_2D]; uint8_t *const levels = set_levels(levels_buf, width); - uint8_t level_counts[MAX_TX_SQUARE]; + DECLARE_ALIGNED(16, uint8_t, level_counts[MAX_TX_SQUARE]); const uint8_t allow_update_cdf = args->allow_update_cdf; TX_SIZE txsize_ctx = get_txsize_context(tx_size);
diff --git a/test/test.cmake b/test/test.cmake index df6ed85..7a0d09a 100644 --- a/test/test.cmake +++ b/test/test.cmake
@@ -169,6 +169,12 @@ endif () endif () + if (CONFIG_LV_MAP) + set(AOM_UNIT_TEST_COMMON_SOURCES + ${AOM_UNIT_TEST_COMMON_SOURCES} + "${AOM_ROOT}/test/txb_test.cc") + endif () + set(AOM_UNIT_TEST_COMMON_INTRIN_NEON ${AOM_UNIT_TEST_COMMON_INTRIN_NEON} "${AOM_ROOT}/test/simd_cmp_neon.cc")
diff --git a/test/test.mk b/test/test.mk index 17bdb81..c2191b3 100644 --- a/test/test.mk +++ b/test/test.mk
@@ -154,6 +154,7 @@ LIBAOM_TEST_SRCS-$(HAVE_AVX2) += simd_avx2_test.cc LIBAOM_TEST_SRCS-$(HAVE_NEON) += simd_neon_test.cc LIBAOM_TEST_SRCS-yes += intrapred_test.cc +LIBAOM_TEST_SRCS-$(CONFIG_LV_MAP) += txb_test.cc LIBAOM_TEST_SRCS-$(CONFIG_INTRABC) += intrabc_test.cc #LIBAOM_TEST_SRCS-$(CONFIG_AV1_DECODER) += av1_thread_test.cc LIBAOM_TEST_SRCS-$(CONFIG_AV1_ENCODER) += dct16x16_test.cc
diff --git a/test/txb_test.cc b/test/txb_test.cc new file mode 100644 index 0000000..83bba51 --- /dev/null +++ b/test/txb_test.cc
@@ -0,0 +1,149 @@ +/* + * Copyright (c) 2017, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include "third_party/googletest/src/googletest/include/gtest/gtest.h" + +#include "./aom_config.h" +#include "./av1_rtcd.h" +#include "aom_ports/aom_timer.h" +#include "aom_ports/mem.h" +#include "av1/common/onyxc_int.h" +#include "av1/common/txb_common.h" +#include "test/acm_random.h" +#include "test/clear_system_state.h" +#include "test/register_state_check.h" +#include "test/util.h" + +namespace { +using libaom_test::ACMRandom; + +typedef void (*GetLevelCountsFunc)(const uint8_t *const levels, const int width, + const int height, + uint8_t *const level_counts); + +class TxbTest : public ::testing::TestWithParam<GetLevelCountsFunc> { + public: + TxbTest() : get_br_level_counts_func_(GetParam()) {} + + virtual ~TxbTest() {} + + virtual void SetUp() { + level_counts_ref_ = reinterpret_cast<uint8_t *>( + aom_memalign(16, sizeof(*level_counts_ref_) * MAX_TX_SQUARE)); + ASSERT_TRUE(level_counts_ref_ != NULL); + level_counts_ = reinterpret_cast<uint8_t *>( + aom_memalign(16, sizeof(*level_counts_) * MAX_TX_SQUARE)); + ASSERT_TRUE(level_counts_ != NULL); + } + + virtual void TearDown() { + aom_free(level_counts_ref_); + aom_free(level_counts_); + libaom_test::ClearSystemState(); + } + + void GetLevelCountsRun() { + const int kNumTests = 10000; + int result = 0; + + for (int tx_size = TX_4X4; tx_size < TX_SIZES_ALL; ++tx_size) { + const int width = tx_size_wide[tx_size]; + const int height = tx_size_high[tx_size]; + levels_ = set_levels(levels_buf_, width); + memset(levels_buf_, 0, sizeof(*levels_buf_) * MAX_TX_SQUARE); + + for (int i = 0; i < kNumTests && !result; ++i) { + InitLevels(width, height); + + av1_get_br_level_counts_c(levels_, width, height, level_counts_ref_); + get_br_level_counts_func_(levels_, width, height, level_counts_); + + PrintDiff(width, height); + + result = memcmp(level_counts_, level_counts_ref_, + sizeof(*level_counts_ref_) * MAX_TX_SQUARE); + + EXPECT_EQ(result, 0) << " width " << width << " height " << height; + } + } + } + + void SpeedTestGetLevelCountsRun() { + const int kNumTests = 10000000; + aom_usec_timer timer; + + for (int tx_size = TX_4X4; tx_size < TX_SIZES_ALL; ++tx_size) { + const int width = tx_size_wide[tx_size]; + const int height = tx_size_high[tx_size]; + levels_ = set_levels(levels_buf_, width); + memset(levels_buf_, 0, sizeof(*levels_buf_) * MAX_TX_SQUARE); + InitLevels(width, height); + + aom_usec_timer_start(&timer); + for (int i = 0; i < kNumTests; ++i) { + get_br_level_counts_func_(levels_, width, height, level_counts_); + } + aom_usec_timer_mark(&timer); + + const int elapsed_time = static_cast<int>(aom_usec_timer_elapsed(&timer)); + printf("get_br_level_counts_%2dx%2d: %7.1f ms\n", width, height, + elapsed_time / 1000.0); + } + } + + void InitLevels(const int width, const int height) { + const int stride = width + TX_PAD_HOR; + + for (int i = 0; i < height; ++i) { + for (int j = 0; j < width; ++j) { + levels_[i * stride + j] = rnd_.Rand8() % (NUM_BASE_LEVELS + 2); + } + } + for (int i = 0; i < MAX_TX_SQUARE; ++i) { + level_counts_ref_[i] = level_counts_[i] = 255; + } + } + + void PrintDiff(const int width, const int height) const { + if (memcmp(level_counts_, level_counts_ref_, + sizeof(*level_counts_ref_) * MAX_TX_SQUARE)) { + for (int y = 0; y < height; y++) { + for (int x = 0; x < width; x++) { + if (level_counts_ref_[y * width + x] != + level_counts_[y * width + x]) { + printf("count[%d][%d] diff:%6d (ref),%6d (opt)\n", y, x, + level_counts_ref_[y * width + x], + level_counts_[y * width + x]); + break; + } + } + } + } + } + + GetLevelCountsFunc get_br_level_counts_func_; + ACMRandom rnd_; + tran_low_t *coeff_; + uint8_t levels_buf_[TX_PAD_2D]; + uint8_t *levels_; + uint8_t *level_counts_ref_; + uint8_t *level_counts_; +}; + +TEST_P(TxbTest, BitExact) { GetLevelCountsRun(); } + +TEST_P(TxbTest, DISABLED_Speed) { SpeedTestGetLevelCountsRun(); } + +#if HAVE_SSE2 +INSTANTIATE_TEST_CASE_P(SSE2, TxbTest, + ::testing::Values(av1_get_br_level_counts_sse2)); +#endif +} // namespace