| OLD | NEW |
| 1 /* | 1 /* |
| 2 * Copyright (c) 2013 The WebRTC project authors. All Rights Reserved. | 2 * Copyright (c) 2013 The WebRTC project authors. All Rights Reserved. |
| 3 * | 3 * |
| 4 * Use of this source code is governed by a BSD-style license | 4 * Use of this source code is governed by a BSD-style license |
| 5 * that can be found in the LICENSE file in the root of the source | 5 * that can be found in the LICENSE file in the root of the source |
| 6 * tree. An additional intellectual property rights grant can be found | 6 * tree. An additional intellectual property rights grant can be found |
| 7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
| 8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
| 9 */ | 9 */ |
| 10 | 10 |
| (...skipping 69 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 80 // 8) Else, if we're not on the second load, goto (4). | 80 // 8) Else, if we're not on the second load, goto (4). |
| 81 // | 81 // |
| 82 // Note: we're glossing over how the sub-sample handling works with | 82 // Note: we're glossing over how the sub-sample handling works with |
| 83 // |virtual_source_idx_|, etc. | 83 // |virtual_source_idx_|, etc. |
| 84 | 84 |
| 85 // MSVC++ requires this to be set before any other includes to get M_PI. | 85 // MSVC++ requires this to be set before any other includes to get M_PI. |
| 86 #define _USE_MATH_DEFINES | 86 #define _USE_MATH_DEFINES |
| 87 | 87 |
| 88 #include "webrtc/common_audio/resampler/sinc_resampler.h" | 88 #include "webrtc/common_audio/resampler/sinc_resampler.h" |
| 89 | 89 |
| 90 #include <assert.h> | |
| 91 #include <math.h> | 90 #include <math.h> |
| 92 #include <string.h> | 91 #include <string.h> |
| 93 | 92 |
| 94 #include <limits> | 93 #include <limits> |
| 95 | 94 |
| 95 #include "webrtc/base/checks.h" |
| 96 #include "webrtc/system_wrappers/include/cpu_features_wrapper.h" | 96 #include "webrtc/system_wrappers/include/cpu_features_wrapper.h" |
| 97 #include "webrtc/typedefs.h" | 97 #include "webrtc/typedefs.h" |
| 98 | 98 |
| 99 namespace webrtc { | 99 namespace webrtc { |
| 100 | 100 |
| 101 namespace { | 101 namespace { |
| 102 | 102 |
| 103 double SincScaleFactor(double io_ratio) { | 103 double SincScaleFactor(double io_ratio) { |
| 104 // |sinc_scale_factor| is basically the normalized cutoff frequency of the | 104 // |sinc_scale_factor| is basically the normalized cutoff frequency of the |
| 105 // low-pass filter. | 105 // low-pass filter. |
| (...skipping 52 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 158 AlignedMalloc(sizeof(float) * kKernelStorageSize, 16))), | 158 AlignedMalloc(sizeof(float) * kKernelStorageSize, 16))), |
| 159 input_buffer_(static_cast<float*>( | 159 input_buffer_(static_cast<float*>( |
| 160 AlignedMalloc(sizeof(float) * input_buffer_size_, 16))), | 160 AlignedMalloc(sizeof(float) * input_buffer_size_, 16))), |
| 161 #if defined(WEBRTC_CPU_DETECTION) | 161 #if defined(WEBRTC_CPU_DETECTION) |
| 162 convolve_proc_(NULL), | 162 convolve_proc_(NULL), |
| 163 #endif | 163 #endif |
| 164 r1_(input_buffer_.get()), | 164 r1_(input_buffer_.get()), |
| 165 r2_(input_buffer_.get() + kKernelSize / 2) { | 165 r2_(input_buffer_.get() + kKernelSize / 2) { |
| 166 #if defined(WEBRTC_CPU_DETECTION) | 166 #if defined(WEBRTC_CPU_DETECTION) |
| 167 InitializeCPUSpecificFeatures(); | 167 InitializeCPUSpecificFeatures(); |
| 168 assert(convolve_proc_); | 168 RTC_DCHECK(convolve_proc_); |
| 169 #endif | 169 #endif |
| 170 assert(request_frames_ > 0); | 170 RTC_DCHECK_GT(request_frames_, 0); |
| 171 Flush(); | 171 Flush(); |
| 172 assert(block_size_ > kKernelSize); | 172 RTC_DCHECK_GT(block_size_, kKernelSize); |
| 173 | 173 |
| 174 memset(kernel_storage_.get(), 0, | 174 memset(kernel_storage_.get(), 0, |
| 175 sizeof(*kernel_storage_.get()) * kKernelStorageSize); | 175 sizeof(*kernel_storage_.get()) * kKernelStorageSize); |
| 176 memset(kernel_pre_sinc_storage_.get(), 0, | 176 memset(kernel_pre_sinc_storage_.get(), 0, |
| 177 sizeof(*kernel_pre_sinc_storage_.get()) * kKernelStorageSize); | 177 sizeof(*kernel_pre_sinc_storage_.get()) * kKernelStorageSize); |
| 178 memset(kernel_window_storage_.get(), 0, | 178 memset(kernel_window_storage_.get(), 0, |
| 179 sizeof(*kernel_window_storage_.get()) * kKernelStorageSize); | 179 sizeof(*kernel_window_storage_.get()) * kKernelStorageSize); |
| 180 | 180 |
| 181 InitializeKernel(); | 181 InitializeKernel(); |
| 182 } | 182 } |
| 183 | 183 |
| 184 SincResampler::~SincResampler() {} | 184 SincResampler::~SincResampler() {} |
| 185 | 185 |
| 186 void SincResampler::UpdateRegions(bool second_load) { | 186 void SincResampler::UpdateRegions(bool second_load) { |
| 187 // Setup various region pointers in the buffer (see diagram above). If we're | 187 // Setup various region pointers in the buffer (see diagram above). If we're |
| 188 // on the second load we need to slide r0_ to the right by kKernelSize / 2. | 188 // on the second load we need to slide r0_ to the right by kKernelSize / 2. |
| 189 r0_ = input_buffer_.get() + (second_load ? kKernelSize : kKernelSize / 2); | 189 r0_ = input_buffer_.get() + (second_load ? kKernelSize : kKernelSize / 2); |
| 190 r3_ = r0_ + request_frames_ - kKernelSize; | 190 r3_ = r0_ + request_frames_ - kKernelSize; |
| 191 r4_ = r0_ + request_frames_ - kKernelSize / 2; | 191 r4_ = r0_ + request_frames_ - kKernelSize / 2; |
| 192 block_size_ = r4_ - r2_; | 192 block_size_ = r4_ - r2_; |
| 193 | 193 |
| 194 // r1_ at the beginning of the buffer. | 194 // r1_ at the beginning of the buffer. |
| 195 assert(r1_ == input_buffer_.get()); | 195 RTC_DCHECK_EQ(r1_, input_buffer_.get()); |
| 196 // r1_ left of r2_, r4_ left of r3_ and size correct. | 196 // r1_ left of r2_, r4_ left of r3_ and size correct. |
| 197 assert(r2_ - r1_ == r4_ - r3_); | 197 RTC_DCHECK_EQ(r2_ - r1_, r4_ - r3_); |
| 198 // r2_ left of r3. | 198 // r2_ left of r3. |
| 199 assert(r2_ < r3_); | 199 RTC_DCHECK_LT(r2_, r3_); |
| 200 } | 200 } |
| 201 | 201 |
| 202 void SincResampler::InitializeKernel() { | 202 void SincResampler::InitializeKernel() { |
| 203 // Blackman window parameters. | 203 // Blackman window parameters. |
| 204 static const double kAlpha = 0.16; | 204 static const double kAlpha = 0.16; |
| 205 static const double kA0 = 0.5 * (1.0 - kAlpha); | 205 static const double kA0 = 0.5 * (1.0 - kAlpha); |
| 206 static const double kA1 = 0.5; | 206 static const double kA1 = 0.5; |
| 207 static const double kA2 = 0.5 * kAlpha; | 207 static const double kA2 = 0.5 * kAlpha; |
| 208 | 208 |
| 209 // Generates a set of windowed sinc() kernels. | 209 // Generates a set of windowed sinc() kernels. |
| (...skipping 66 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 276 const float* const kernel_ptr = kernel_storage_.get(); | 276 const float* const kernel_ptr = kernel_storage_.get(); |
| 277 while (remaining_frames) { | 277 while (remaining_frames) { |
| 278 // |i| may be negative if the last Resample() call ended on an iteration | 278 // |i| may be negative if the last Resample() call ended on an iteration |
| 279 // that put |virtual_source_idx_| over the limit. | 279 // that put |virtual_source_idx_| over the limit. |
| 280 // | 280 // |
| 281 // Note: The loop construct here can severely impact performance on ARM | 281 // Note: The loop construct here can severely impact performance on ARM |
| 282 // or when built with clang. See https://codereview.chromium.org/18566009/ | 282 // or when built with clang. See https://codereview.chromium.org/18566009/ |
| 283 for (int i = static_cast<int>( | 283 for (int i = static_cast<int>( |
| 284 ceil((block_size_ - virtual_source_idx_) / current_io_ratio)); | 284 ceil((block_size_ - virtual_source_idx_) / current_io_ratio)); |
| 285 i > 0; --i) { | 285 i > 0; --i) { |
| 286 assert(virtual_source_idx_ < block_size_); | 286 RTC_DCHECK_LT(virtual_source_idx_, block_size_); |
| 287 | 287 |
| 288 // |virtual_source_idx_| lies in between two kernel offsets so figure out | 288 // |virtual_source_idx_| lies in between two kernel offsets so figure out |
| 289 // what they are. | 289 // what they are. |
| 290 const int source_idx = static_cast<int>(virtual_source_idx_); | 290 const int source_idx = static_cast<int>(virtual_source_idx_); |
| 291 const double subsample_remainder = virtual_source_idx_ - source_idx; | 291 const double subsample_remainder = virtual_source_idx_ - source_idx; |
| 292 | 292 |
| 293 const double virtual_offset_idx = | 293 const double virtual_offset_idx = |
| 294 subsample_remainder * kKernelOffsetCount; | 294 subsample_remainder * kKernelOffsetCount; |
| 295 const int offset_idx = static_cast<int>(virtual_offset_idx); | 295 const int offset_idx = static_cast<int>(virtual_offset_idx); |
| 296 | 296 |
| 297 // We'll compute "convolutions" for the two kernels which straddle | 297 // We'll compute "convolutions" for the two kernels which straddle |
| 298 // |virtual_source_idx_|. | 298 // |virtual_source_idx_|. |
| 299 const float* const k1 = kernel_ptr + offset_idx * kKernelSize; | 299 const float* const k1 = kernel_ptr + offset_idx * kKernelSize; |
| 300 const float* const k2 = k1 + kKernelSize; | 300 const float* const k2 = k1 + kKernelSize; |
| 301 | 301 |
| 302 // Ensure |k1|, |k2| are 16-byte aligned for SIMD usage. Should always be | 302 // Ensure |k1|, |k2| are 16-byte aligned for SIMD usage. Should always be |
| 303 // true so long as kKernelSize is a multiple of 16. | 303 // true so long as kKernelSize is a multiple of 16. |
| 304 assert(0u == (reinterpret_cast<uintptr_t>(k1) & 0x0F)); | 304 RTC_DCHECK_EQ(0, reinterpret_cast<uintptr_t>(k1) % 16); |
| 305 assert(0u == (reinterpret_cast<uintptr_t>(k2) & 0x0F)); | 305 RTC_DCHECK_EQ(0, reinterpret_cast<uintptr_t>(k2) % 16); |
| 306 | 306 |
| 307 // Initialize input pointer based on quantized |virtual_source_idx_|. | 307 // Initialize input pointer based on quantized |virtual_source_idx_|. |
| 308 const float* const input_ptr = r1_ + source_idx; | 308 const float* const input_ptr = r1_ + source_idx; |
| 309 | 309 |
| 310 // Figure out how much to weight each kernel's "convolution". | 310 // Figure out how much to weight each kernel's "convolution". |
| 311 const double kernel_interpolation_factor = | 311 const double kernel_interpolation_factor = |
| 312 virtual_offset_idx - offset_idx; | 312 virtual_offset_idx - offset_idx; |
| 313 *destination++ = CONVOLVE_FUNC( | 313 *destination++ = CONVOLVE_FUNC( |
| 314 input_ptr, k1, k2, kernel_interpolation_factor); | 314 input_ptr, k1, k2, kernel_interpolation_factor); |
| 315 | 315 |
| (...skipping 47 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 363 sum1 += *input_ptr * *k1++; | 363 sum1 += *input_ptr * *k1++; |
| 364 sum2 += *input_ptr++ * *k2++; | 364 sum2 += *input_ptr++ * *k2++; |
| 365 } | 365 } |
| 366 | 366 |
| 367 // Linearly interpolate the two "convolutions". | 367 // Linearly interpolate the two "convolutions". |
| 368 return static_cast<float>((1.0 - kernel_interpolation_factor) * sum1 + | 368 return static_cast<float>((1.0 - kernel_interpolation_factor) * sum1 + |
| 369 kernel_interpolation_factor * sum2); | 369 kernel_interpolation_factor * sum2); |
| 370 } | 370 } |
| 371 | 371 |
| 372 } // namespace webrtc | 372 } // namespace webrtc |
| OLD | NEW |