Finalization of the first version of EchoCanceller 3

webrtc/modules/audio_processing/aec3/matched_filter.cc

Index: webrtc/modules/audio_processing/aec3/matched_filter.cc
diff --git a/webrtc/modules/audio_processing/aec3/matched_filter.cc b/webrtc/modules/audio_processing/aec3/matched_filter.cc
index f187159911295aaf18771401d3f3cc71fed8c44b..20ceb916a97c85f350fc80f104498548be0c8532 100644
--- a/webrtc/modules/audio_processing/aec3/matched_filter.cc
+++ b/webrtc/modules/audio_processing/aec3/matched_filter.cc
@@ -9,6 +9,10 @@
  */
 #include "webrtc/modules/audio_processing/aec3/matched_filter.h"
 
+#include "webrtc/typedefs.h"
+#if defined(WEBRTC_ARCH_X86_FAMILY)
+#include <emmintrin.h>
+#endif
 #include <algorithm>
 #include <numeric>
 
@@ -16,6 +20,131 @@
 #include "webrtc/modules/audio_processing/logging/apm_data_dumper.h"
 
 namespace webrtc {
+namespace aec3 {
+
+#if defined(WEBRTC_ARCH_X86_FAMILY)
+
+void MatchedFilterCore_SSE2(size_t x_start_index,
+                            float x2_sum_threshold,
+                            rtc::ArrayView<const float> x,
+                            rtc::ArrayView<const float> y,
+                            rtc::ArrayView<float> h,
+                            bool* filters_updated,
+                            float* error_sum) {
+  // Process for all samples in the sub-block.
+  for (size_t i = 0; i < kSubBlockSize; ++i) {
+    // Apply the matched filter as filter * x.  and compute x * x.
+    float x2_sum = 0.f;
+    float s = 0;
+    size_t x_index = x_start_index;
+    RTC_DCHECK_EQ(h.size(), (h.size() / 4) * 4);

[ivoc, 2017/02/22 17:08:13]

I would prefer a check for h.size() % 4 == 0.

[peah-webrtc, 2017/02/22 23:51:39]

Done.

+
+    __m128 s_128 = _mm_set1_ps(0);
+    __m128 x2_sum_128 = _mm_set1_ps(0);
+
+    size_t k = 0;
+    if (h.size() > (x.size() - x_index)) {
+      const size_t limit = x.size() - x_index;
+      for (; (k + 3) < limit; k += 4, x_index += 4) {
+        const __m128 x_k = _mm_loadu_ps(&x[x_index]);
+        const __m128 h_k = _mm_loadu_ps(&h[k]);
+        const __m128 xx = _mm_mul_ps(x_k, x_k);
+        x2_sum_128 = _mm_add_ps(x2_sum_128, xx);
+        const __m128 hx = _mm_mul_ps(h_k, x_k);
+        s_128 = _mm_add_ps(s_128, hx);
+      }
+
+      for (; k < limit; ++k, ++x_index) {
+        x2_sum += x[x_index] * x[x_index];
+        s += h[k] * x[x_index];
+      }
+      x_index = 0;
+    }
+
+    for (; k + 3 < h.size(); k += 4, x_index += 4) {
+      const __m128 x_k = _mm_loadu_ps(&x[x_index]);
+      const __m128 h_k = _mm_loadu_ps(&h[k]);
+      const __m128 xx = _mm_mul_ps(x_k, x_k);
+      x2_sum_128 = _mm_add_ps(x2_sum_128, xx);
+      const __m128 hx = _mm_mul_ps(h_k, x_k);
+      s_128 = _mm_add_ps(s_128, hx);
+    }
+
+    for (; k < h.size(); ++k, ++x_index) {
+      x2_sum += x[x_index] * x[x_index];
+      s += h[k] * x[x_index];
+    }
+
+    float* v = reinterpret_cast<float*>(&x2_sum_128);
+    x2_sum += v[0] + v[1] + v[2] + v[3];
+    v = reinterpret_cast<float*>(&s_128);
+    s += v[0] + v[1] + v[2] + v[3];
+
+    // Compute the matched filter error.
+    const float e = std::min(32767.f, std::max(-32768.f, y[i] - s));
+    (*error_sum) += e * e;
+
+    // Update the matched filter estimate in an NLMS manner.
+    if (x2_sum > x2_sum_threshold) {
+      RTC_DCHECK_LT(0.f, x2_sum);
+      const float alpha = 0.7f * e / x2_sum;
+
+      // filter = filter + 0.7 * (y - filter * x) / x * x.
+      size_t x_index = x_start_index;
+      for (size_t k = 0; k < h.size(); ++k) {
+        h[k] += alpha * x[x_index];
+        x_index = x_index < (x.size() - 1) ? x_index + 1 : 0;
+      }
+      *filters_updated = true;
+    }
+
+    x_start_index = x_start_index > 0 ? x_start_index - 1 : x.size() - 1;
+  }
+}
+#endif
+
+void MatchedFilterCore(size_t x_start_index,
+                       float x2_sum_threshold,
+                       rtc::ArrayView<const float> x,
+                       rtc::ArrayView<const float> y,
+                       rtc::ArrayView<float> h,
+                       bool* filters_updated,
+                       float* error_sum) {
+  // Process for all samples in the sub-block.
+  for (size_t i = 0; i < kSubBlockSize; ++i) {
+    // Apply the matched filter as filter * x.  and compute x * x.
+    float x2_sum = 0.f;
+    float s = 0;
+    size_t x_index = x_start_index;
+    for (size_t k = 0; k < h.size(); ++k) {
+      x2_sum += x[x_index] * x[x_index];
+      s += h[k] * x[x_index];
+      x_index = x_index < (x.size() - 1) ? x_index + 1 : 0;
+    }
+
+    // Compute the matched filter error.
+    const float e = std::min(32767.f, std::max(-32768.f, y[i] - s));
+    (*error_sum) += e * e;
+
+    // Update the matched filter estimate in an NLMS manner.
+    if (x2_sum > x2_sum_threshold) {
+      RTC_DCHECK_LT(0.f, x2_sum);
+      const float alpha = 0.7f * e / x2_sum;
+
+      // filter = filter + 0.7 * (y - filter * x) / x * x.
+      size_t x_index = x_start_index;
+      for (size_t k = 0; k < h.size(); ++k) {
+        h[k] += alpha * x[x_index];
+        x_index = x_index < (x.size() - 1) ? x_index + 1 : 0;
+      }
+      *filters_updated = true;
+    }
+
+    x_start_index = x_start_index > 0 ? x_start_index - 1 : x.size() - 1;
+  }
+}
+
+}  // namespace aec3
 
 MatchedFilter::IndexedBuffer::IndexedBuffer(size_t size) : data(size, 0.f) {
   RTC_DCHECK_EQ(0, size % kSubBlockSize);
@@ -24,10 +153,12 @@ MatchedFilter::IndexedBuffer::IndexedBuffer(size_t size) : data(size, 0.f) {
 MatchedFilter::IndexedBuffer::~IndexedBuffer() = default;
 
 MatchedFilter::MatchedFilter(ApmDataDumper* data_dumper,
+                             Aec3Optimization optimization,
                              size_t window_size_sub_blocks,
                              int num_matched_filters,
                              size_t alignment_shift_sub_blocks)
     : data_dumper_(data_dumper),
+      optimization_(optimization),
       filter_intra_lag_shift_(alignment_shift_sub_blocks * kSubBlockSize),
       filters_(num_matched_filters,
                std::vector<float>(window_size_sub_blocks * kSubBlockSize, 0.f)),
@@ -65,65 +196,17 @@ void MatchedFilter::Update(const std::array<float, kSubBlockSize>& render,
         (x_buffer_.index + alignment_shift + kSubBlockSize - 1) %
         x_buffer_.data.size();
 
-    // Process for all samples in the sub-block.
-    for (size_t i = 0; i < kSubBlockSize; ++i) {
-      // As x_buffer is a circular buffer, all of the processing is split into
-      // two loops around the wrapping of the buffer.
-      const size_t loop_size_1 =
-          std::min(filters_[n].size(), x_buffer_.data.size() - x_start_index);
-      const size_t loop_size_2 = filters_[n].size() - loop_size_1;
-      RTC_DCHECK_EQ(filters_[n].size(), loop_size_1 + loop_size_2);
-
-      // x * x.
-      float x2_sum = std::inner_product(
-          x_buffer_.data.begin() + x_start_index,
-          x_buffer_.data.begin() + x_start_index + loop_size_1,
-          x_buffer_.data.begin() + x_start_index, 0.f);
-      // Apply the matched filter as filter * x.
-      float s = std::inner_product(filters_[n].begin(),
-                                   filters_[n].begin() + loop_size_1,
-                                   x_buffer_.data.begin() + x_start_index, 0.f);
-
-      if (loop_size_2 > 0) {
-        // Update the cumulative sum of x * x.
-        x2_sum = std::inner_product(x_buffer_.data.begin(),
-                                    x_buffer_.data.begin() + loop_size_2,
-                                    x_buffer_.data.begin(), x2_sum);
-
-        // Compute the matched filter output filter * x in a cumulative manner.
-        s = std::inner_product(x_buffer_.data.begin(),
-                               x_buffer_.data.begin() + loop_size_2,
-                               filters_[n].begin() + loop_size_1, s);
-      }
-
-      // Compute the matched filter error.
-      const float e = std::min(32767.f, std::max(-32768.f, y[i] - s));
-      error_sum += e * e;
-
-      // Update the matched filter estimate in an NLMS manner.
-      if (x2_sum > x2_sum_threshold) {
-        filters_updated = true;
-        RTC_DCHECK_LT(0.f, x2_sum);
-        const float alpha = 0.7f * e / x2_sum;
-
-        // filter = filter + 0.7 * (y - filter * x) / x * x.
-        std::transform(filters_[n].begin(), filters_[n].begin() + loop_size_1,
-                       x_buffer_.data.begin() + x_start_index,
-                       filters_[n].begin(),
-                       [&](float a, float b) { return a + alpha * b; });
-
-        if (loop_size_2 > 0) {
-          // filter = filter + 0.7 * (y - filter * x) / x * x.
-          std::transform(x_buffer_.data.begin(),
-                         x_buffer_.data.begin() + loop_size_2,
-                         filters_[n].begin() + loop_size_1,
-                         filters_[n].begin() + loop_size_1,
-                         [&](float a, float b) { return b + alpha * a; });
-        }
-      }
-
-      x_start_index =
-          x_start_index > 0 ? x_start_index - 1 : x_buffer_.data.size() - 1;
+    switch (optimization_) {
+#if defined(WEBRTC_ARCH_X86_FAMILY)
+      case Aec3Optimization::kSse2:
+        aec3::MatchedFilterCore_SSE2(x_start_index, x2_sum_threshold,
+                                     x_buffer_.data, y, filters_[n],
+                                     &filters_updated, &error_sum);
+        break;
+#endif
+      default:
+        aec3::MatchedFilterCore(x_start_index, x2_sum_threshold, x_buffer_.data,
+                                y, filters_[n], &filters_updated, &error_sum);
     }
 
     // Compute anchor for the matched filter error.
@@ -140,11 +223,12 @@ void MatchedFilter::Update(const std::array<float, kSubBlockSize>& render,
             [](float a, float b) -> bool { return a * a < b * b; }));
 
     // Update the lag estimates for the matched filter.
-    const float kMatchingFilterThreshold = 0.3f;
-    lag_estimates_[n] =
-        LagEstimate(error_sum_anchor - error_sum,
-                    error_sum < kMatchingFilterThreshold * error_sum_anchor,
-                    lag_estimate + alignment_shift, filters_updated);
+    const float kMatchingFilterThreshold = 0.1f;
+    lag_estimates_[n] = LagEstimate(
+        error_sum_anchor - error_sum,
+        (lag_estimate > 2 && lag_estimate < (filters_[n].size() - 10) &&
+         error_sum < kMatchingFilterThreshold * error_sum_anchor),
+        lag_estimate + alignment_shift, filters_updated);
 
     // TODO(peah): Remove once development of EchoCanceller3 is fully done.
     RTC_DCHECK_EQ(4, filters_.size());