| OLD | NEW |
| 1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. | 1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. |
| 2 // Use of this source code is governed by a BSD-style license that can be | 2 // Use of this source code is governed by a BSD-style license that can be |
| 3 // found in the LICENSE file. | 3 // found in the LICENSE file. |
| 4 | 4 |
| 5 // MSVC++ requires this to be set before any other includes to get M_PI. | 5 // MSVC++ requires this to be set before any other includes to get M_PI. |
| 6 #define _USE_MATH_DEFINES | 6 #define _USE_MATH_DEFINES |
| 7 #include <cmath> | 7 #include <cmath> |
| 8 | 8 |
| 9 #include "base/command_line.h" | 9 #include "base/command_line.h" |
| 10 #include "base/cpu.h" |
| 10 #include "base/memory/aligned_memory.h" | 11 #include "base/memory/aligned_memory.h" |
| 11 #include "base/memory/scoped_ptr.h" | 12 #include "base/memory/scoped_ptr.h" |
| 12 #include "base/string_number_conversions.h" | 13 #include "base/string_number_conversions.h" |
| 13 #include "base/time.h" | 14 #include "base/time.h" |
| 14 #include "media/base/vector_math.h" | 15 #include "media/base/vector_math.h" |
| 15 #include "media/base/vector_math_testing.h" | 16 #include "media/base/vector_math_testing.h" |
| 16 #include "testing/gtest/include/gtest/gtest.h" | 17 #include "testing/gtest/include/gtest/gtest.h" |
| 17 | 18 |
| 18 using base::TimeTicks; | 19 using base::TimeTicks; |
| 19 using std::fill; | 20 using std::fill; |
| (...skipping 63 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 83 } | 84 } |
| 84 | 85 |
| 85 { | 86 { |
| 86 SCOPED_TRACE("FMAC_C"); | 87 SCOPED_TRACE("FMAC_C"); |
| 87 FillTestVectors(kInputFillValue, kOutputFillValue); | 88 FillTestVectors(kInputFillValue, kOutputFillValue); |
| 88 vector_math::FMAC_C( | 89 vector_math::FMAC_C( |
| 89 input_vector.get(), kScale, kVectorSize, output_vector.get()); | 90 input_vector.get(), kScale, kVectorSize, output_vector.get()); |
| 90 VerifyOutput(kResult); | 91 VerifyOutput(kResult); |
| 91 } | 92 } |
| 92 | 93 |
| 93 #if defined(ARCH_CPU_X86_FAMILY) && defined(__SSE__) | 94 #if defined(ARCH_CPU_X86_FAMILY) |
| 94 { | 95 { |
| 96 ASSERT_TRUE(base::CPU().has_sse()); |
| 95 SCOPED_TRACE("FMAC_SSE"); | 97 SCOPED_TRACE("FMAC_SSE"); |
| 96 FillTestVectors(kInputFillValue, kOutputFillValue); | 98 FillTestVectors(kInputFillValue, kOutputFillValue); |
| 97 vector_math::FMAC_SSE( | 99 vector_math::FMAC_SSE( |
| 98 input_vector.get(), kScale, kVectorSize, output_vector.get()); | 100 input_vector.get(), kScale, kVectorSize, output_vector.get()); |
| 99 VerifyOutput(kResult); | 101 VerifyOutput(kResult); |
| 100 } | 102 } |
| 101 #endif | 103 #endif |
| 102 } | 104 } |
| 103 | 105 |
| 104 // Benchmark for each optimized vector_math::FMAC() method. Original benchmarks | 106 // Benchmark for each optimized vector_math::FMAC() method. Original benchmarks |
| 105 // were run with --vector-fmac-iterations=200000. | 107 // were run with --vector-fmac-iterations=200000. |
| 106 TEST_F(VectorMathTest, FMACBenchmark) { | 108 TEST_F(VectorMathTest, FMACBenchmark) { |
| 107 static const int kBenchmarkIterations = BenchmarkIterations(); | 109 static const int kBenchmarkIterations = BenchmarkIterations(); |
| 108 | 110 |
| 109 printf("Benchmarking %d iterations:\n", kBenchmarkIterations); | 111 printf("Benchmarking %d iterations:\n", kBenchmarkIterations); |
| 110 | 112 |
| 111 // Benchmark FMAC_C(). | 113 // Benchmark FMAC_C(). |
| 112 FillTestVectors(kInputFillValue, kOutputFillValue); | 114 FillTestVectors(kInputFillValue, kOutputFillValue); |
| 113 TimeTicks start = TimeTicks::HighResNow(); | 115 TimeTicks start = TimeTicks::HighResNow(); |
| 114 for (int i = 0; i < kBenchmarkIterations; ++i) { | 116 for (int i = 0; i < kBenchmarkIterations; ++i) { |
| 115 vector_math::FMAC_C( | 117 vector_math::FMAC_C( |
| 116 input_vector.get(), kScale, kVectorSize, output_vector.get()); | 118 input_vector.get(), kScale, kVectorSize, output_vector.get()); |
| 117 } | 119 } |
| 118 double total_time_c_ms = (TimeTicks::HighResNow() - start).InMillisecondsF(); | 120 double total_time_c_ms = (TimeTicks::HighResNow() - start).InMillisecondsF(); |
| 119 printf("FMAC_C took %.2fms.\n", total_time_c_ms); | 121 printf("FMAC_C took %.2fms.\n", total_time_c_ms); |
| 120 | 122 |
| 121 #if defined(ARCH_CPU_X86_FAMILY) && defined(__SSE__) | 123 #if defined(ARCH_CPU_X86_FAMILY) |
| 124 ASSERT_TRUE(base::CPU().has_sse()); |
| 125 |
| 122 // Benchmark FMAC_SSE() with unaligned size. | 126 // Benchmark FMAC_SSE() with unaligned size. |
| 123 ASSERT_NE((kVectorSize - 1) % (vector_math::kRequiredAlignment / | 127 ASSERT_NE((kVectorSize - 1) % (vector_math::kRequiredAlignment / |
| 124 sizeof(float)), 0U); | 128 sizeof(float)), 0U); |
| 125 FillTestVectors(kInputFillValue, kOutputFillValue); | 129 FillTestVectors(kInputFillValue, kOutputFillValue); |
| 126 start = TimeTicks::HighResNow(); | 130 start = TimeTicks::HighResNow(); |
| 127 for (int j = 0; j < kBenchmarkIterations; ++j) { | 131 for (int j = 0; j < kBenchmarkIterations; ++j) { |
| 128 vector_math::FMAC_SSE( | 132 vector_math::FMAC_SSE( |
| 129 input_vector.get(), kScale, kVectorSize - 1, output_vector.get()); | 133 input_vector.get(), kScale, kVectorSize - 1, output_vector.get()); |
| 130 } | 134 } |
| 131 double total_time_sse_unaligned_ms = | 135 double total_time_sse_unaligned_ms = |
| (...skipping 14 matching lines...) Expand all Loading... |
| 146 double total_time_sse_aligned_ms = | 150 double total_time_sse_aligned_ms = |
| 147 (TimeTicks::HighResNow() - start).InMillisecondsF(); | 151 (TimeTicks::HighResNow() - start).InMillisecondsF(); |
| 148 printf("FMAC_SSE (aligned size) took %.2fms; which is %.2fx faster than" | 152 printf("FMAC_SSE (aligned size) took %.2fms; which is %.2fx faster than" |
| 149 " FMAC_C and %.2fx faster than FMAC_SSE (unaligned size).\n", | 153 " FMAC_C and %.2fx faster than FMAC_SSE (unaligned size).\n", |
| 150 total_time_sse_aligned_ms, total_time_c_ms / total_time_sse_aligned_ms, | 154 total_time_sse_aligned_ms, total_time_c_ms / total_time_sse_aligned_ms, |
| 151 total_time_sse_unaligned_ms / total_time_sse_aligned_ms); | 155 total_time_sse_unaligned_ms / total_time_sse_aligned_ms); |
| 152 #endif | 156 #endif |
| 153 } | 157 } |
| 154 | 158 |
| 155 } // namespace media | 159 } // namespace media |
| OLD | NEW |