Remove dead function: av1_get_br_level_counts Change-Id: I83bb5daf20774fcc5191f9137ea824a4935df301
diff --git a/av1/av1.cmake b/av1/av1.cmake index 8e163f3..583c9dc 100644 --- a/av1/av1.cmake +++ b/av1/av1.cmake
@@ -372,10 +372,6 @@ "${AOM_ROOT}/av1/common/txb_common.c" "${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/common/av1_rtcd_defs.pl b/av1/common/av1_rtcd_defs.pl index 4dcda55..a8ada97 100755 --- a/av1/common/av1_rtcd_defs.pl +++ b/av1/common/av1_rtcd_defs.pl
@@ -80,12 +80,6 @@ specialize qw/av1_highbd_wiener_convolve_add_src avx2/; # -# txb -# -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 1d70be8..c0f61fa 100644 --- a/av1/common/txb_common.c +++ b/av1/common/txb_common.c
@@ -235,35 +235,3 @@ const int16_t k_eob_group_start[12] = { 0, 1, 2, 3, 5, 9, 17, 33, 65, 129, 257, 513 }; const int16_t k_eob_offset_bits[12] = { 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 }; - -// Note: because of the SSE2 optimization, levels[] must be in the range [0, -// 127], inclusive. -void av1_get_base_level_counts(const uint8_t *const levels, - const int level_minus_1, const int width, - const int height, uint8_t *const level_counts) { - const int stride = width + TX_PAD_HOR; - - for (int row = 0; row < height; ++row) { - for (int col = 0; col < width; ++col) { - level_counts[row * width + col] = - get_level_count(levels, stride, row, col, level_minus_1, - base_ref_offset, BASE_CONTEXT_POSITION_NUM); - } - } -} - -// Note: because of the SSE2 optimization, levels[] must be in the range [0, -// 127], inclusive. -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; - - for (int row = 0; row < height; ++row) { - for (int col = 0; col < width; ++col) { - level_counts[row * width + col] = - get_level_count(levels, stride, row, col, level_minus_1, - br_ref_offset, BR_CONTEXT_POSITION_NUM); - } - } -}
diff --git a/av1/common/txb_common.h b/av1/common/txb_common.h index 29ad05b..336e1cc 100644 --- a/av1/common/txb_common.h +++ b/av1/common/txb_common.h
@@ -671,8 +671,4 @@ void av1_init_lv_map(AV1_COMMON *cm); -void av1_get_base_level_counts(const uint8_t *const levels, - const int level_minus_1, const int width, - const int height, uint8_t *const level_counts); - #endif // AV1_COMMON_TXB_COMMON_H_
diff --git a/av1/common/x86/txb_sse2.c b/av1/common/x86/txb_sse2.c deleted file mode 100644 index 775d083..0000000 --- a/av1/common/x86/txb_sse2.c +++ /dev/null
@@ -1,256 +0,0 @@ -/* - * 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 "aom_dsp/x86/mem_sse2.h" -#include "av1/common/onyxc_int.h" -#include "av1/common/txb_common.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); -} - -// This function calculates 16 pixels' level counts, and updates levels history. -// These 16 pixels may be from different parts of input levels[] if the width is -// less than 16. -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); - __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]; - int row = height; - - // levels[] are divided into 4 even parts. In each loop, totally 4 rows from - // different parts are processed together. - 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)); - - // levels[] are divided into 2 even parts. In each loop, totally 2 rows from - // different parts are processed together. - 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; - } -} - -// Note: levels[] must be in the range [0, 127], inclusive. -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/test/test.cmake b/test/test.cmake index 092b31c..80be8d9 100644 --- a/test/test.cmake +++ b/test/test.cmake
@@ -162,10 +162,6 @@ "${AOM_ROOT}/test/hiprec_convolve_test_util.h") endif () - set(AOM_UNIT_TEST_COMMON_SOURCES - ${AOM_UNIT_TEST_COMMON_SOURCES} - "${AOM_ROOT}/test/txb_test.cc") - set(AOM_UNIT_TEST_ENCODER_SOURCES ${AOM_UNIT_TEST_ENCODER_SOURCES} "${AOM_ROOT}/test/encodetxb_test.cc")
diff --git a/test/txb_test.cc b/test/txb_test.cc deleted file mode 100644 index d91e4fb..0000000 --- a/test/txb_test.cc +++ /dev/null
@@ -1,158 +0,0 @@ -/* - * 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 <stdint.h> -#include <stdio.h> -#include <string.h> - -#include "third_party/googletest/src/googletest/include/gtest/gtest.h" - -#include "./aom_config.h" -#include "./av1_rtcd.h" -#include "aom/aom_integer.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_)); - - for (int i = 0; i < kNumTests && !result; ++i) { - InitLevels(width, height, i); - - 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_)); - InitLevels(width, height, 0); - - 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); - } - } - - private: - void InitLevels(const int width, const int height, const int idx) { - const int stride = width + TX_PAD_HOR; - // levels_[] is initialized to either the whole possible range, or small - // values around base level. - const int max_value = (idx & 1) ? INT8_MAX : (NUM_BASE_LEVELS + 2); - - for (int i = 0; i < height; ++i) { - for (int j = 0; j < width; ++j) { - levels_[i * stride + j] = - static_cast<uint8_t>(clamp(rnd_.Rand8(), 0, max_value)); - } - } - for (int i = 0; i < MAX_TX_SQUARE; ++i) { - level_counts_ref_[i] = level_counts_[i] = rnd_.Rand8(); - } - } - - 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_; - 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