Line data    Source code

       1             : /*
       2             :  *  Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
       3             :  *
       4             :  *  Use of this source code is governed by a BSD-style license
       5             :  *  that can be found in the LICENSE file in the root of the source
       6             :  *  tree. An additional intellectual property rights grant can be found
       7             :  *  in the file PATENTS.  All contributing project authors may
       8             :  *  be found in the AUTHORS file in the root of the source tree.
       9             :  */
      10             : 
      11             : #include "webrtc/common_audio/fir_filter_sse.h"
      12             : 
      13             : #include <stdint.h>
      14             : #include <string.h>
      15             : #include <xmmintrin.h>
      16             : 
      17             : #include "webrtc/base/checks.h"
      18             : #include "webrtc/system_wrappers/include/aligned_malloc.h"
      19             : 
      20             : namespace webrtc {
      21             : 
      22           0 : FIRFilterSSE2::FIRFilterSSE2(const float* coefficients,
      23             :                              size_t coefficients_length,
      24           0 :                              size_t max_input_length)
      25             :     :  // Closest higher multiple of four.
      26           0 :       coefficients_length_((coefficients_length + 3) & ~0x03),
      27           0 :       state_length_(coefficients_length_ - 1),
      28             :       coefficients_(static_cast<float*>(
      29           0 :           AlignedMalloc(sizeof(float) * coefficients_length_, 16))),
      30             :       state_(static_cast<float*>(
      31           0 :           AlignedMalloc(sizeof(float) * (max_input_length + state_length_),
      32           0 :                         16))) {
      33             :   // Add zeros at the end of the coefficients.
      34           0 :   size_t padding = coefficients_length_ - coefficients_length;
      35           0 :   memset(coefficients_.get(), 0, padding * sizeof(coefficients_[0]));
      36             :   // The coefficients are reversed to compensate for the order in which the
      37             :   // input samples are acquired (most recent last).
      38           0 :   for (size_t i = 0; i < coefficients_length; ++i) {
      39           0 :     coefficients_[i + padding] = coefficients[coefficients_length - i - 1];
      40             :   }
      41           0 :   memset(state_.get(),
      42             :          0,
      43           0 :          (max_input_length + state_length_) * sizeof(state_[0]));
      44           0 : }
      45             : 
      46           0 : void FIRFilterSSE2::Filter(const float* in, size_t length, float* out) {
      47           0 :   RTC_DCHECK_GT(length, 0);
      48             : 
      49           0 :   memcpy(&state_[state_length_], in, length * sizeof(*in));
      50             : 
      51             :   // Convolves the input signal |in| with the filter kernel |coefficients_|
      52             :   // taking into account the previous state.
      53           0 :   for (size_t i = 0; i < length; ++i) {
      54           0 :     float* in_ptr = &state_[i];
      55           0 :     float* coef_ptr = coefficients_.get();
      56             : 
      57           0 :     __m128 m_sum = _mm_setzero_ps();
      58             :     __m128 m_in;
      59             : 
      60             :     // Depending on if the pointer is aligned with 16 bytes or not it is loaded
      61             :     // differently.
      62           0 :     if (reinterpret_cast<uintptr_t>(in_ptr) & 0x0F) {
      63           0 :       for (size_t j = 0; j < coefficients_length_; j += 4) {
      64           0 :         m_in = _mm_loadu_ps(in_ptr + j);
      65           0 :         m_sum = _mm_add_ps(m_sum, _mm_mul_ps(m_in, _mm_load_ps(coef_ptr + j)));
      66             :       }
      67             :     } else {
      68           0 :       for (size_t j = 0; j < coefficients_length_; j += 4) {
      69           0 :         m_in = _mm_load_ps(in_ptr + j);
      70           0 :         m_sum = _mm_add_ps(m_sum, _mm_mul_ps(m_in, _mm_load_ps(coef_ptr + j)));
      71             :       }
      72             :     }
      73           0 :     m_sum = _mm_add_ps(_mm_movehl_ps(m_sum, m_sum), m_sum);
      74           0 :     _mm_store_ss(out + i, _mm_add_ss(m_sum, _mm_shuffle_ps(m_sum, m_sum, 1)));
      75             :   }
      76             : 
      77             :   // Update current state.
      78           0 :   memmove(state_.get(), &state_[length], state_length_ * sizeof(state_[0]));
      79           0 : }
      80             : 
      81             : }  // namespace webrtc