DO NOT COMMIT. Diff of downstream webrtc resampler commit for review purposes.

media/base/sinc_resampler.cc

Index: media/base/sinc_resampler.cc
diff --git a/media/base/sinc_resampler.cc b/media/base/sinc_resampler.cc
index a2918c3f0d2e590c2eedc48ed29aa92a3e015343..b66ed345a3f13d148148ea29075216b9acca771c 100644
--- a/media/base/sinc_resampler.cc
+++ b/media/base/sinc_resampler.cc
@@ -1,7 +1,16 @@
-// Copyright (c) 2012 The Chromium Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-//
+/*
+ *  Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+// Modified from the Chromium original:
+// src/media/base/sinc_resampler.cc
+
 // Initial input buffer layout, dividing into regions r0_ to r4_ (note: r0_, r3_
 // and r4_ will move after the first load):
 //
@@ -76,19 +85,16 @@
 // MSVC++ requires this to be set before any other includes to get M_PI.
 #define _USE_MATH_DEFINES
 
-#include "media/base/sinc_resampler.h"
+#include "webrtc/common_audio/resampler/sinc_resampler.h"
+#include "webrtc/system_wrappers/interface/compile_assert.h"
+#include "webrtc/system_wrappers/interface/cpu_features_wrapper.h"
+#include "webrtc/typedefs.h"
 
 #include <cmath>
+#include <cstring>
 #include <limits>
 
-#include "base/cpu.h"
-#include "base/logging.h"
-
-#if defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
-#include <arm_neon.h>
-#endif
-
-namespace media {
+namespace webrtc {
 
 static double SincScaleFactor(double io_ratio) {
   // |sinc_scale_factor| is basically the normalized cutoff frequency of the
@@ -107,31 +113,36 @@ static double SincScaleFactor(double io_ratio) {
 }
 
 // If we know the minimum architecture at compile time, avoid CPU detection.
-// Force NaCl code to use C routines since (at present) nothing there uses these
-// methods and plumbing the -msse built library is non-trivial.  iOS lies
-// about its architecture, so we also need to exclude it here.
-#if defined(ARCH_CPU_X86_FAMILY) && !defined(OS_NACL) && !defined(OS_IOS)
+// iOS lies about its architecture, so we also need to exclude it here.
+#if defined(WEBRTC_ARCH_X86_FAMILY) && !defined(WEBRTC_IOS)
 #if defined(__SSE__)
 #define CONVOLVE_FUNC Convolve_SSE
 void SincResampler::InitializeCPUSpecificFeatures() {}
 #else
-// X86 CPU detection required.  Functions will be set by
+// X86 CPU detection required.  Function will be set by
 // InitializeCPUSpecificFeatures().
 // TODO(dalecurtis): Once Chrome moves to an SSE baseline this can be removed.
-#define CONVOLVE_FUNC g_convolve_proc_
-
-typedef float (*ConvolveProc)(const float*, const float*, const float*, double);
-static ConvolveProc g_convolve_proc_ = NULL;
+#define CONVOLVE_FUNC convolve_proc_
 
 void SincResampler::InitializeCPUSpecificFeatures() {
-  CHECK(!g_convolve_proc_);
-  g_convolve_proc_ = base::CPU().has_sse() ? Convolve_SSE : Convolve_C;
+  convolve_proc_ = WebRtc_GetCPUInfo(kSSE2) ? Convolve_SSE : Convolve_C;
 }
 #endif
-#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
+#elif defined(WEBRTC_ARCH_ARM_V7)
+#if defined(WEBRTC_ARCH_ARM_NEON)
 #define CONVOLVE_FUNC Convolve_NEON
 void SincResampler::InitializeCPUSpecificFeatures() {}
 #else
+// NEON CPU detection required.  Function will be set by
+// InitializeCPUSpecificFeatures().
+#define CONVOLVE_FUNC convolve_proc_
+
+void SincResampler::InitializeCPUSpecificFeatures() {
+  convolve_proc_ = WebRtc_GetCPUFeaturesARM() & kCPUFeatureNEON ?
+      Convolve_NEON : Convolve_C;
+}
+#endif
+#else
 // Unknown architecture.
 #define CONVOLVE_FUNC Convolve_C
 void SincResampler::InitializeCPUSpecificFeatures() {}
@@ -139,26 +150,32 @@ void SincResampler::InitializeCPUSpecificFeatures() {}
 
 SincResampler::SincResampler(double io_sample_rate_ratio,
                              int request_frames,
-                             const ReadCB& read_cb)
+                             SincResamplerCallback* read_cb)
     : io_sample_rate_ratio_(io_sample_rate_ratio),
       read_cb_(read_cb),
       request_frames_(request_frames),
       input_buffer_size_(request_frames_ + kKernelSize),
       // Create input buffers with a 16-byte alignment for SSE optimizations.
       kernel_storage_(static_cast<float*>(
-          base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))),
+          AlignedMalloc(sizeof(float) * kKernelStorageSize, 16))),
       kernel_pre_sinc_storage_(static_cast<float*>(
-          base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))),
+          AlignedMalloc(sizeof(float) * kKernelStorageSize, 16))),
       kernel_window_storage_(static_cast<float*>(
-          base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))),
+          AlignedMalloc(sizeof(float) * kKernelStorageSize, 16))),
       input_buffer_(static_cast<float*>(
-          base::AlignedAlloc(sizeof(float) * input_buffer_size_, 16))),
+          AlignedMalloc(sizeof(float) * input_buffer_size_, 16))),
+#if defined(WEBRTC_RESAMPLER_CPU_DETECTION)
+      convolve_proc_(NULL),
+#endif
       r1_(input_buffer_.get()),
       r2_(input_buffer_.get() + kKernelSize / 2) {
-  CHECK_GT(request_frames_, 0);
+#if defined(WEBRTC_RESAMPLER_CPU_DETECTION)
+  InitializeCPUSpecificFeatures();
+  assert(convolve_proc_);
+#endif
+  assert(request_frames_ > 0);
   Flush();
-  CHECK_GT(block_size_, kKernelSize)
-      << "block_size must be greater than kKernelSize!";
+  assert(block_size_ > kKernelSize);
 
   memset(kernel_storage_.get(), 0,
          sizeof(*kernel_storage_.get()) * kKernelStorageSize);
@@ -181,11 +198,11 @@ void SincResampler::UpdateRegions(bool second_load) {
   block_size_ = r4_ - r2_;
 
   // r1_ at the beginning of the buffer.
-  CHECK_EQ(r1_, input_buffer_.get());
+  assert(r1_ == input_buffer_.get());
   // r1_ left of r2_, r4_ left of r3_ and size correct.
-  CHECK_EQ(r2_ - r1_, r4_ - r3_);
+  assert(r2_ - r1_ == r4_ - r3_);
   // r2_ left of r3.
-  CHECK_LT(r2_, r3_);
+  assert(r2_ < r3_);
 }
 
 void SincResampler::InitializeKernel() {
@@ -205,20 +222,20 @@ void SincResampler::InitializeKernel() {
     for (int i = 0; i < kKernelSize; ++i) {
       const int idx = i + offset_idx * kKernelSize;
       const float pre_sinc = M_PI * (i - kKernelSize / 2 - subsample_offset);
-      kernel_pre_sinc_storage_[idx] = pre_sinc;
+      kernel_pre_sinc_storage_.get()[idx] = pre_sinc;
 
       // Compute Blackman window, matching the offset of the sinc().
       const float x = (i - subsample_offset) / kKernelSize;
       const float window = kA0 - kA1 * cos(2.0 * M_PI * x) + kA2
           * cos(4.0 * M_PI * x);
-      kernel_window_storage_[idx] = window;
+      kernel_window_storage_.get()[idx] = window;
 
       // Compute the sinc with offset, then window the sinc() function and store
       // at the correct offset.
       if (pre_sinc == 0) {
-        kernel_storage_[idx] = sinc_scale_factor * window;
+        kernel_storage_.get()[idx] = sinc_scale_factor * window;
       } else {
-        kernel_storage_[idx] =
+        kernel_storage_.get()[idx] =
             window * sin(sinc_scale_factor * pre_sinc) / pre_sinc;
       }
     }
@@ -239,13 +256,13 @@ void SincResampler::SetRatio(double io_sample_rate_ratio) {
   for (int offset_idx = 0; offset_idx <= kKernelOffsetCount; ++offset_idx) {
     for (int i = 0; i < kKernelSize; ++i) {
       const int idx = i + offset_idx * kKernelSize;
-      const float window = kernel_window_storage_[idx];
-      const float pre_sinc = kernel_pre_sinc_storage_[idx];
+      const float window = kernel_window_storage_.get()[idx];
+      const float pre_sinc = kernel_pre_sinc_storage_.get()[idx];
 
       if (pre_sinc == 0) {
-        kernel_storage_[idx] = sinc_scale_factor * window;
+        kernel_storage_.get()[idx] = sinc_scale_factor * window;
       } else {
-        kernel_storage_[idx] =
+        kernel_storage_.get()[idx] =
             window * sin(sinc_scale_factor * pre_sinc) / pre_sinc;
       }
     }
@@ -257,7 +274,7 @@ void SincResampler::Resample(int frames, float* destination) {
 
   // Step (1) -- Prime the input buffer at the start of the input stream.
   if (!buffer_primed_ && remaining_frames) {
-    read_cb_.Run(request_frames_, r0_);
+    read_cb_->Run(request_frames_, r0_);
     buffer_primed_ = true;
   }
 
@@ -273,7 +290,7 @@ void SincResampler::Resample(int frames, float* destination) {
     // or when built with clang.  See https://codereview.chromium.org/18566009/
     for (int i = ceil((block_size_ - virtual_source_idx_) / current_io_ratio);
          i > 0; --i) {
-      DCHECK_LT(virtual_source_idx_, block_size_);
+      assert(virtual_source_idx_ < block_size_);
 
       // |virtual_source_idx_| lies in between two kernel offsets so figure out
       // what they are.
@@ -291,8 +308,8 @@ void SincResampler::Resample(int frames, float* destination) {
 
       // Ensure |k1|, |k2| are 16-byte aligned for SIMD usage.  Should always be
       // true so long as kKernelSize is a multiple of 16.
-      DCHECK_EQ(0u, reinterpret_cast<uintptr_t>(k1) & 0x0F);
-      DCHECK_EQ(0u, reinterpret_cast<uintptr_t>(k2) & 0x0F);
+      assert(0u == (reinterpret_cast<uintptr_t>(k1) & 0x0F));
+      assert(0u == (reinterpret_cast<uintptr_t>(k2) & 0x0F));
 
       // Initialize input pointer based on quantized |virtual_source_idx_|.
       const float* const input_ptr = r1_ + source_idx;
@@ -322,7 +339,7 @@ void SincResampler::Resample(int frames, float* destination) {
       UpdateRegions(true);
 
     // Step (5) -- Refresh the buffer with more input.
-    read_cb_.Run(request_frames_, r0_);
+    read_cb_->Run(request_frames_, r0_);
   }
 }
 
@@ -359,33 +376,4 @@ float SincResampler::Convolve_C(const float* input_ptr, const float* k1,
       + kernel_interpolation_factor * sum2;
 }
 
-#if defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
-float SincResampler::Convolve_NEON(const float* input_ptr, const float* k1,
-                                   const float* k2,
-                                   double kernel_interpolation_factor) {
-  float32x4_t m_input;
-  float32x4_t m_sums1 = vmovq_n_f32(0);
-  float32x4_t m_sums2 = vmovq_n_f32(0);
-
-  const float* upper = input_ptr + kKernelSize;
-  for (; input_ptr < upper; ) {
-    m_input = vld1q_f32(input_ptr);
-    input_ptr += 4;
-    m_sums1 = vmlaq_f32(m_sums1, m_input, vld1q_f32(k1));
-    k1 += 4;
-    m_sums2 = vmlaq_f32(m_sums2, m_input, vld1q_f32(k2));
-    k2 += 4;
-  }
-
-  // Linearly interpolate the two "convolutions".
-  m_sums1 = vmlaq_f32(
-      vmulq_f32(m_sums1, vmovq_n_f32(1.0 - kernel_interpolation_factor)),
-      m_sums2, vmovq_n_f32(kernel_interpolation_factor));
-
-  // Sum components together.
-  float32x2_t m_half = vadd_f32(vget_high_f32(m_sums1), vget_low_f32(m_sums1));
-  return vget_lane_f32(vpadd_f32(m_half, m_half), 0);
-}
-#endif
-
-}  // namespace media
+}  // namespace webrtc