| 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 // Initial input buffer layout, dividing into regions r0_ to r4_ (note: r0_, r3_ | 5 // Initial input buffer layout, dividing into regions r0_ to r4_ (note: r0_, r3_ |
| 6 // and r4_ will move after the first load): | 6 // and r4_ will move after the first load): |
| 7 // | 7 // |
| 8 // |----------------|-----------------------------------------|----------------| | 8 // |----------------|-----------------------------------------|----------------| |
| 9 // | 9 // |
| 10 // request_frames_ | 10 // request_frames_ |
| (...skipping 135 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 146 // Create input buffers with a 16-byte alignment for SSE optimizations. | 146 // Create input buffers with a 16-byte alignment for SSE optimizations. |
| 147 kernel_storage_(static_cast<float*>( | 147 kernel_storage_(static_cast<float*>( |
| 148 base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))), | 148 base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))), |
| 149 kernel_pre_sinc_storage_(static_cast<float*>( | 149 kernel_pre_sinc_storage_(static_cast<float*>( |
| 150 base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))), | 150 base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))), |
| 151 kernel_window_storage_(static_cast<float*>( | 151 kernel_window_storage_(static_cast<float*>( |
| 152 base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))), | 152 base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))), |
| 153 input_buffer_(static_cast<float*>( | 153 input_buffer_(static_cast<float*>( |
| 154 base::AlignedAlloc(sizeof(float) * input_buffer_size_, 16))), | 154 base::AlignedAlloc(sizeof(float) * input_buffer_size_, 16))), |
| 155 r1_(input_buffer_.get()), | 155 r1_(input_buffer_.get()), |
| 156 r2_(input_buffer_.get() + kKernelSize / 2), | 156 r2_(input_buffer_.get() + kKernelSize / 2) { |
| 157 not_currently_resampling_(1) { | |
| 158 CHECK_GT(request_frames_, 0); | 157 CHECK_GT(request_frames_, 0); |
| 159 Flush(); | 158 Flush(); |
| 160 CHECK_GT(block_size_, kKernelSize) | 159 CHECK_GT(block_size_, kKernelSize) |
| 161 << "block_size must be greater than kKernelSize!"; | 160 << "block_size must be greater than kKernelSize!"; |
| 162 | 161 |
| 163 memset(kernel_storage_.get(), 0, | 162 memset(kernel_storage_.get(), 0, |
| 164 sizeof(*kernel_storage_.get()) * kKernelStorageSize); | 163 sizeof(*kernel_storage_.get()) * kKernelStorageSize); |
| 165 memset(kernel_pre_sinc_storage_.get(), 0, | 164 memset(kernel_pre_sinc_storage_.get(), 0, |
| 166 sizeof(*kernel_pre_sinc_storage_.get()) * kKernelStorageSize); | 165 sizeof(*kernel_pre_sinc_storage_.get()) * kKernelStorageSize); |
| 167 memset(kernel_window_storage_.get(), 0, | 166 memset(kernel_window_storage_.get(), 0, |
| 168 sizeof(*kernel_window_storage_.get()) * kKernelStorageSize); | 167 sizeof(*kernel_window_storage_.get()) * kKernelStorageSize); |
| 169 | 168 |
| 170 InitializeKernel(); | 169 InitializeKernel(); |
| 171 } | 170 } |
| 172 | 171 |
| 173 SincResampler::~SincResampler() { | 172 SincResampler::~SincResampler() {} |
| 174 // TODO(dalecurtis): Remove debugging for http://crbug.com/295278 | |
| 175 CHECK(!base::AtomicRefCountDec(¬_currently_resampling_)); | |
| 176 } | |
| 177 | 173 |
| 178 void SincResampler::UpdateRegions(bool second_load) { | 174 void SincResampler::UpdateRegions(bool second_load) { |
| 179 // Setup various region pointers in the buffer (see diagram above). If we're | 175 // Setup various region pointers in the buffer (see diagram above). If we're |
| 180 // on the second load we need to slide r0_ to the right by kKernelSize / 2. | 176 // on the second load we need to slide r0_ to the right by kKernelSize / 2. |
| 181 r0_ = input_buffer_.get() + (second_load ? kKernelSize : kKernelSize / 2); | 177 r0_ = input_buffer_.get() + (second_load ? kKernelSize : kKernelSize / 2); |
| 182 r3_ = r0_ + request_frames_ - kKernelSize; | 178 r3_ = r0_ + request_frames_ - kKernelSize; |
| 183 r4_ = r0_ + request_frames_ - kKernelSize / 2; | 179 r4_ = r0_ + request_frames_ - kKernelSize / 2; |
| 184 block_size_ = r4_ - r2_; | 180 block_size_ = r4_ - r2_; |
| 185 | 181 |
| 186 // r1_ at the beginning of the buffer. | 182 // r1_ at the beginning of the buffer. |
| (...skipping 62 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 249 kernel_storage_[idx] = sinc_scale_factor * window; | 245 kernel_storage_[idx] = sinc_scale_factor * window; |
| 250 } else { | 246 } else { |
| 251 kernel_storage_[idx] = | 247 kernel_storage_[idx] = |
| 252 window * sin(sinc_scale_factor * pre_sinc) / pre_sinc; | 248 window * sin(sinc_scale_factor * pre_sinc) / pre_sinc; |
| 253 } | 249 } |
| 254 } | 250 } |
| 255 } | 251 } |
| 256 } | 252 } |
| 257 | 253 |
| 258 void SincResampler::Resample(int frames, float* destination) { | 254 void SincResampler::Resample(int frames, float* destination) { |
| 259 CHECK(!base::AtomicRefCountDec(¬_currently_resampling_)); | |
| 260 | |
| 261 int remaining_frames = frames; | 255 int remaining_frames = frames; |
| 262 | 256 |
| 263 // Step (1) -- Prime the input buffer at the start of the input stream. | 257 // Step (1) -- Prime the input buffer at the start of the input stream. |
| 264 if (!buffer_primed_ && remaining_frames) { | 258 if (!buffer_primed_ && remaining_frames) { |
| 265 read_cb_.Run(request_frames_, r0_); | 259 read_cb_.Run(request_frames_, r0_); |
| 266 buffer_primed_ = true; | 260 buffer_primed_ = true; |
| 267 } | 261 } |
| 268 | 262 |
| 269 // Step (2) -- Resample! const what we can outside of the loop for speed. It | 263 // Step (2) -- Resample! const what we can outside of the loop for speed. It |
| 270 // actually has an impact on ARM performance. See inner loop comment below. | 264 // actually has an impact on ARM performance. See inner loop comment below. |
| (...skipping 28 matching lines...) Expand all Loading... |
| 299 // Figure out how much to weight each kernel's "convolution". | 293 // Figure out how much to weight each kernel's "convolution". |
| 300 const double kernel_interpolation_factor = | 294 const double kernel_interpolation_factor = |
| 301 virtual_offset_idx - offset_idx; | 295 virtual_offset_idx - offset_idx; |
| 302 *destination++ = CONVOLVE_FUNC( | 296 *destination++ = CONVOLVE_FUNC( |
| 303 input_ptr, k1, k2, kernel_interpolation_factor); | 297 input_ptr, k1, k2, kernel_interpolation_factor); |
| 304 | 298 |
| 305 // Advance the virtual index. | 299 // Advance the virtual index. |
| 306 virtual_source_idx_ += current_io_ratio; | 300 virtual_source_idx_ += current_io_ratio; |
| 307 source_idx = virtual_source_idx_; | 301 source_idx = virtual_source_idx_; |
| 308 | 302 |
| 309 if (!--remaining_frames) { | 303 if (!--remaining_frames) |
| 310 base::AtomicRefCountInc(¬_currently_resampling_); | |
| 311 return; | 304 return; |
| 312 } | |
| 313 } | 305 } |
| 314 | 306 |
| 315 // Wrap back around to the start. | 307 // Wrap back around to the start. |
| 316 DCHECK_GE(virtual_source_idx_, block_size_); | 308 DCHECK_GE(virtual_source_idx_, block_size_); |
| 317 virtual_source_idx_ -= block_size_; | 309 virtual_source_idx_ -= block_size_; |
| 318 | 310 |
| 319 // Step (3) -- Copy r3_, r4_ to r1_, r2_. | 311 // Step (3) -- Copy r3_, r4_ to r1_, r2_. |
| 320 // This wraps the last input frames back to the start of the buffer. | 312 // This wraps the last input frames back to the start of the buffer. |
| 321 memcpy(r1_, r3_, sizeof(*input_buffer_.get()) * kKernelSize); | 313 memcpy(r1_, r3_, sizeof(*input_buffer_.get()) * kKernelSize); |
| 322 | 314 |
| 323 // Step (4) -- Reinitialize regions if necessary. | 315 // Step (4) -- Reinitialize regions if necessary. |
| 324 if (r0_ == r2_) | 316 if (r0_ == r2_) |
| 325 UpdateRegions(true); | 317 UpdateRegions(true); |
| 326 | 318 |
| 327 // Step (5) -- Refresh the buffer with more input. | 319 // Step (5) -- Refresh the buffer with more input. |
| 328 read_cb_.Run(request_frames_, r0_); | 320 read_cb_.Run(request_frames_, r0_); |
| 329 } | 321 } |
| 330 | |
| 331 base::AtomicRefCountInc(¬_currently_resampling_); | |
| 332 } | 322 } |
| 333 | 323 |
| 334 #undef CONVOLVE_FUNC | 324 #undef CONVOLVE_FUNC |
| 335 | 325 |
| 336 int SincResampler::ChunkSize() const { | 326 int SincResampler::ChunkSize() const { |
| 337 return block_size_ / io_sample_rate_ratio_; | 327 return block_size_ / io_sample_rate_ratio_; |
| 338 } | 328 } |
| 339 | 329 |
| 340 void SincResampler::Flush() { | 330 void SincResampler::Flush() { |
| 341 CHECK(base::AtomicRefCountIsOne(¬_currently_resampling_)); | |
| 342 virtual_source_idx_ = 0; | 331 virtual_source_idx_ = 0; |
| 343 buffer_primed_ = false; | 332 buffer_primed_ = false; |
| 344 memset(input_buffer_.get(), 0, | 333 memset(input_buffer_.get(), 0, |
| 345 sizeof(*input_buffer_.get()) * input_buffer_size_); | 334 sizeof(*input_buffer_.get()) * input_buffer_size_); |
| 346 UpdateRegions(false); | 335 UpdateRegions(false); |
| 347 } | 336 } |
| 348 | 337 |
| 349 float SincResampler::Convolve_C(const float* input_ptr, const float* k1, | 338 float SincResampler::Convolve_C(const float* input_ptr, const float* k1, |
| 350 const float* k2, | 339 const float* k2, |
| 351 double kernel_interpolation_factor) { | 340 double kernel_interpolation_factor) { |
| (...skipping 36 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 388 vmulq_f32(m_sums1, vmovq_n_f32(1.0 - kernel_interpolation_factor)), | 377 vmulq_f32(m_sums1, vmovq_n_f32(1.0 - kernel_interpolation_factor)), |
| 389 m_sums2, vmovq_n_f32(kernel_interpolation_factor)); | 378 m_sums2, vmovq_n_f32(kernel_interpolation_factor)); |
| 390 | 379 |
| 391 // Sum components together. | 380 // Sum components together. |
| 392 float32x2_t m_half = vadd_f32(vget_high_f32(m_sums1), vget_low_f32(m_sums1)); | 381 float32x2_t m_half = vadd_f32(vget_high_f32(m_sums1), vget_low_f32(m_sums1)); |
| 393 return vget_lane_f32(vpadd_f32(m_half, m_half), 0); | 382 return vget_lane_f32(vpadd_f32(m_half, m_half), 0); |
| 394 } | 383 } |
| 395 #endif | 384 #endif |
| 396 | 385 |
| 397 } // namespace media | 386 } // namespace media |
| OLD | NEW |