Finalization of the first version of EchoCanceller 3

webrtc/modules/audio_processing/aec3/fft_data.h

Index: webrtc/modules/audio_processing/aec3/fft_data.h
diff --git a/webrtc/modules/audio_processing/aec3/fft_data.h b/webrtc/modules/audio_processing/aec3/fft_data.h
new file mode 100644
index 0000000000000000000000000000000000000000..5a92d91d62b91c23d0647a595771e88199bee790
--- /dev/null
+++ b/webrtc/modules/audio_processing/aec3/fft_data.h
@@ -0,0 +1,98 @@
+/*
+ *  Copyright (c) 2017 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.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC3_FFT_DATA_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC3_FFT_DATA_H_
+
+#include "webrtc/typedefs.h"
+#if defined(WEBRTC_ARCH_X86_FAMILY)
+#include <emmintrin.h>
+#endif
+#include <algorithm>
+#include <array>
+
+#include "webrtc/base/array_view.h"
+#include "webrtc/modules/audio_processing/aec3/aec3_common.h"
+
+namespace webrtc {
+
+// Struct that holds imaginary data produced from 128 point real-valued FFTs.
+struct FftData {
+  // Copies the data in src.
+  void Assign(const FftData& src) {
+    std::copy(src.re.begin(), src.re.end(), re.begin());
+    std::copy(src.im.begin(), src.im.end(), im.begin());
+    im[0] = im[kFftLengthBy2] = 0;
+  }
+
+  // Clears all the imaginary.
+  void Clear() {
+    re.fill(0.f);
+    im.fill(0.f);
+  }
+
+  // Computes the power spectrum of the data.
+  void Spectrum(Aec3Optimization optimization,
+                std::array<float, kFftLengthBy2Plus1>* power_spectrum) const {
+    RTC_DCHECK(power_spectrum);
+    switch (optimization) {
+#if defined(WEBRTC_ARCH_X86_FAMILY)
+      case Aec3Optimization::kSse2: {
+        constexpr int kNumFourBinBands = kFftLengthBy2 / 4;
+        constexpr int kLimit = kNumFourBinBands * 4;
+        for (size_t k = 0; k < kLimit; k += 4) {
+          const __m128 r = _mm_loadu_ps(&re[k]);
+          const __m128 i = _mm_loadu_ps(&im[k]);
+          const __m128 ii = _mm_mul_ps(i, i);
+          const __m128 rr = _mm_mul_ps(r, r);
+          const __m128 rrii = _mm_add_ps(rr, ii);
+          _mm_storeu_ps(&(*power_spectrum)[k], rrii);
+        }
+        (*power_spectrum)[kFftLengthBy2] =

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

Is it necessary to compute this coefficient separate from the SSE loop? It looks
like the same calculation, right?

[peah-webrtc, 2017/02/22 23:51:38]

There is no gain in doing this in the SSE loop as the SSE2 operations operate on
blocks of 4 floats. Since the array is 65 long, that leaves the last operation
to operate on a 1 float long block. Therefore, I partitioned it such that the
SSE loop takes care of the first 64 values (divides evenly by 4) and value 65 is
computed separately.

+            re[kFftLengthBy2] * re[kFftLengthBy2] +
+            im[kFftLengthBy2] * im[kFftLengthBy2];
+      } break;
+#endif
+      default:
+        std::transform(re.begin(), re.end(), im.begin(),
+                       power_spectrum->begin(),
+                       [](float a, float b) { return a * a + b * b; });
+    }
+  }
+
+  // Copy the data from an interleaved array.
+  void CopyFromPackedArray(const std::array<float, kFftLength>& v) {
+    re[0] = v[0];
+    re[kFftLengthBy2] = v[1];
+    im[0] = im[kFftLengthBy2] = 0;
+    for (size_t k = 1, j = 2; k < kFftLengthBy2; ++k) {
+      re[k] = v[j++];
+      im[k] = v[j++];
+    }
+  }
+
+  // Copies the data into an interleaved array.
+  void CopyToPackedArray(std::array<float, kFftLength>* v) const {
+    RTC_DCHECK(v);
+    (*v)[0] = re[0];
+    (*v)[1] = re[kFftLengthBy2];
+    for (size_t k = 1, j = 2; k < kFftLengthBy2; ++k) {
+      (*v)[j++] = re[k];
+      (*v)[j++] = im[k];
+    }
+  }
+
+  std::array<float, kFftLengthBy2Plus1> re;
+  std::array<float, kFftLengthBy2Plus1> im;
+};
+
+}  // namespace webrtc
+
+#endif  // WEBRTC_MODULES_AUDIO_PROCESSING_AEC3_FFT_DATA_H_