Merge 187732 "Don't use magic statics in SincResampler for threa..."

media/base/sinc_resampler.cc

 // 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.
 //
 // Input buffer layout, dividing the total buffer into regions (r0_ - r5_):
 //
 // |----------------|-----------------------------------------|----------------|
 //
 //                                   kBlockSize + kKernelSize / 2
 //                   <--------------------------------------------------------->
@@ skipping 39 matching lines @@
 SincResampler::SincResampler(double io_sample_rate_ratio, const ReadCB& read_cb)
     : io_sample_rate_ratio_(io_sample_rate_ratio),
       virtual_source_idx_(0),
       buffer_primed_(false),
       read_cb_(read_cb),
       // Create input buffers with a 16-byte alignment for SSE optimizations.
       kernel_storage_(static_cast<float*>(
           base::AlignedAlloc(sizeof(float) * kKernelStorageSize, 16))),
       input_buffer_(static_cast<float*>(
           base::AlignedAlloc(sizeof(float) * kBufferSize, 16))),
+#if defined(ARCH_CPU_X86_FAMILY) && !defined(__SSE__)
+      convolve_proc_(base::CPU().has_sse() ? Convolve_SSE : Convolve_C),
+#endif
       // Setup various region pointers in the buffer (see diagram above).
       r0_(input_buffer_.get() + kKernelSize / 2),
       r1_(input_buffer_.get()),
       r2_(r0_),
       r3_(r0_ + kBlockSize - kKernelSize / 2),
       r4_(r0_ + kBlockSize),
       r5_(r0_ + kKernelSize / 2) {
   // Ensure kKernelSize is a multiple of 32 for easy SSE optimizations; causes
   // r0_ and r5_ (used for input) to always be 16-byte aligned by virtue of
   // input_buffer_ being 16-byte aligned.
@@ skipping 59 matching lines @@
       double x = (i - subsample_offset) / kKernelSize;
       double window = kA0 - kA1 * cos(2.0 * M_PI * x) + kA2
           * cos(4.0 * M_PI * x);
 
       // Window the sinc() function and store at the correct offset.
       kernel_storage_.get()[i + offset_idx * kKernelSize] = sinc * window;
     }
   }
 }
 
+// If we know the minimum architecture avoid function hopping for CPU detection.
+#if defined(ARCH_CPU_X86_FAMILY)
+#if defined(__SSE__)
+#define CONVOLVE_FUNC Convolve_SSE
+#else
+// X86 CPU detection required.  |convolve_proc_| will be set upon construction.
+// TODO(dalecurtis): Once Chrome moves to a SSE baseline this can be removed.
+#define CONVOLVE_FUNC convolve_proc_
+#endif
+#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
+#define CONVOLVE_FUNC Convolve_NEON
+#else
+// Unknown architecture.
+#define CONVOLVE_FUNC Convolve_C
+#endif
+
 void SincResampler::Resample(float* destination, int frames) {
   int remaining_frames = frames;
 
   // Step (1) -- Prime the input buffer at the start of the input stream.
   if (!buffer_primed_) {
     read_cb_.Run(r0_, kBlockSize + kKernelSize / 2);
     buffer_primed_ = true;
   }
 
   // Step (2) -- Resample!
   while (remaining_frames) {
     while (virtual_source_idx_ < kBlockSize) {
       // |virtual_source_idx_| lies in between two kernel offsets so figure out
       // what they are.
       int source_idx = static_cast<int>(virtual_source_idx_);
       double subsample_remainder = virtual_source_idx_ - source_idx;
 
       double virtual_offset_idx = subsample_remainder * kKernelOffsetCount;
       int offset_idx = static_cast<int>(virtual_offset_idx);
 
       // We'll compute "convolutions" for the two kernels which straddle
       // |virtual_source_idx_|.
       float* k1 = kernel_storage_.get() + offset_idx * kKernelSize;
       float* k2 = k1 + kKernelSize;
 
+      // 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);
+
       // Initialize input pointer based on quantized |virtual_source_idx_|.
       float* input_ptr = r1_ + source_idx;
 
       // Figure out how much to weight each kernel's "convolution".
       double kernel_interpolation_factor = virtual_offset_idx - offset_idx;
-      *destination++ = Convolve(
+      *destination++ = CONVOLVE_FUNC(
           input_ptr, k1, k2, kernel_interpolation_factor);
 
       // Advance the virtual index.
       virtual_source_idx_ += io_sample_rate_ratio_;
 
       if (!--remaining_frames)
         return;
     }
 
     // Wrap back around to the start.
     virtual_source_idx_ -= kBlockSize;
 
     // Step (3) Copy r3_ to r1_ and r4_ to r2_.
     // This wraps the last input frames back to the start of the buffer.
     memcpy(r1_, r3_, sizeof(*input_buffer_.get()) * (kKernelSize / 2));
     memcpy(r2_, r4_, sizeof(*input_buffer_.get()) * (kKernelSize / 2));
 
     // Step (4)
     // Refresh the buffer with more input.
     read_cb_.Run(r5_, kBlockSize);
   }
 }
 
+#undef CONVOLVE_FUNC
+
 int SincResampler::ChunkSize() const {
   return kBlockSize / io_sample_rate_ratio_;
 }
 
 void SincResampler::Flush() {
   virtual_source_idx_ = 0;
   buffer_primed_ = false;
   memset(input_buffer_.get(), 0, sizeof(*input_buffer_.get()) * kBufferSize);
 }
 
-float SincResampler::Convolve(const float* input_ptr, const float* k1,
-                              const float* k2,
-                              double kernel_interpolation_factor) {
-  // Ensure |k1|, |k2| are 16-byte aligned for SSE 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);
-
-  // Rely on function level static initialization to keep ConvolveProc selection
-  // thread safe.
-  typedef float (*ConvolveProc)(const float* src, const float* k1,
-                                const float* k2,
-                                double kernel_interpolation_factor);
-#if defined(ARCH_CPU_X86_FAMILY)
-#if defined(__SSE__)
-  static const ConvolveProc kConvolveProc = Convolve_SSE;
-#else
-  static const ConvolveProc kConvolveProc =
-      base::CPU().has_sse() ? Convolve_SSE : Convolve_C;
-#endif
-#elif defined(ARCH_CPU_ARM_FAMILY) && defined(USE_NEON)
-  static const ConvolveProc kConvolveProc = Convolve_NEON;
-#else
-  static const ConvolveProc kConvolveProc = Convolve_C;
-#endif
-
-  return kConvolveProc(input_ptr, k1, k2, kernel_interpolation_factor);
-}
-
 float SincResampler::Convolve_C(const float* input_ptr, const float* k1,
                                 const float* k2,
                                 double kernel_interpolation_factor) {
   float sum1 = 0;
   float sum2 = 0;
 
   // Generate a single output sample.  Unrolling this loop hurt performance in
   // local testing.
   int n = kKernelSize;
   while (n--) {
@@ skipping 29 matching lines @@
       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