OLD | NEW |
1 /* | 1 /* |
2 * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved. | 2 * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license | 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 | 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 | 6 * tree. An additional intellectual property rights grant can be found |
7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
9 */ | 9 */ |
10 | 10 |
11 #include "webrtc/common_audio/fir_filter_neon.h" | 11 #include "webrtc/common_audio/fir_filter_neon.h" |
12 | 12 |
13 #include <arm_neon.h> | 13 #include <arm_neon.h> |
14 #include <assert.h> | |
15 #include <string.h> | 14 #include <string.h> |
16 | 15 |
| 16 #include "webrtc/base/checks.h" |
17 #include "webrtc/system_wrappers/include/aligned_malloc.h" | 17 #include "webrtc/system_wrappers/include/aligned_malloc.h" |
18 | 18 |
19 namespace webrtc { | 19 namespace webrtc { |
20 | 20 |
21 FIRFilterNEON::FIRFilterNEON(const float* coefficients, | 21 FIRFilterNEON::FIRFilterNEON(const float* coefficients, |
22 size_t coefficients_length, | 22 size_t coefficients_length, |
23 size_t max_input_length) | 23 size_t max_input_length) |
24 : // Closest higher multiple of four. | 24 : // Closest higher multiple of four. |
25 coefficients_length_((coefficients_length + 3) & ~0x03), | 25 coefficients_length_((coefficients_length + 3) & ~0x03), |
26 state_length_(coefficients_length_ - 1), | 26 state_length_(coefficients_length_ - 1), |
27 coefficients_(static_cast<float*>( | 27 coefficients_(static_cast<float*>( |
28 AlignedMalloc(sizeof(float) * coefficients_length_, 16))), | 28 AlignedMalloc(sizeof(float) * coefficients_length_, 16))), |
29 state_(static_cast<float*>( | 29 state_(static_cast<float*>( |
30 AlignedMalloc(sizeof(float) * (max_input_length + state_length_), | 30 AlignedMalloc(sizeof(float) * (max_input_length + state_length_), |
31 16))) { | 31 16))) { |
32 // Add zeros at the end of the coefficients. | 32 // Add zeros at the end of the coefficients. |
33 size_t padding = coefficients_length_ - coefficients_length; | 33 size_t padding = coefficients_length_ - coefficients_length; |
34 memset(coefficients_.get(), 0.f, padding * sizeof(coefficients_[0])); | 34 memset(coefficients_.get(), 0.f, padding * sizeof(coefficients_[0])); |
35 // The coefficients are reversed to compensate for the order in which the | 35 // The coefficients are reversed to compensate for the order in which the |
36 // input samples are acquired (most recent last). | 36 // input samples are acquired (most recent last). |
37 for (size_t i = 0; i < coefficients_length; ++i) { | 37 for (size_t i = 0; i < coefficients_length; ++i) { |
38 coefficients_[i + padding] = coefficients[coefficients_length - i - 1]; | 38 coefficients_[i + padding] = coefficients[coefficients_length - i - 1]; |
39 } | 39 } |
40 memset(state_.get(), | 40 memset(state_.get(), |
41 0.f, | 41 0.f, |
42 (max_input_length + state_length_) * sizeof(state_[0])); | 42 (max_input_length + state_length_) * sizeof(state_[0])); |
43 } | 43 } |
44 | 44 |
45 void FIRFilterNEON::Filter(const float* in, size_t length, float* out) { | 45 void FIRFilterNEON::Filter(const float* in, size_t length, float* out) { |
46 assert(length > 0); | 46 RTC_DCHECK_GT(length, 0); |
47 | 47 |
48 memcpy(&state_[state_length_], in, length * sizeof(*in)); | 48 memcpy(&state_[state_length_], in, length * sizeof(*in)); |
49 | 49 |
50 // Convolves the input signal |in| with the filter kernel |coefficients_| | 50 // Convolves the input signal |in| with the filter kernel |coefficients_| |
51 // taking into account the previous state. | 51 // taking into account the previous state. |
52 for (size_t i = 0; i < length; ++i) { | 52 for (size_t i = 0; i < length; ++i) { |
53 float* in_ptr = &state_[i]; | 53 float* in_ptr = &state_[i]; |
54 float* coef_ptr = coefficients_.get(); | 54 float* coef_ptr = coefficients_.get(); |
55 | 55 |
56 float32x4_t m_sum = vmovq_n_f32(0); | 56 float32x4_t m_sum = vmovq_n_f32(0); |
57 float32x4_t m_in; | 57 float32x4_t m_in; |
58 | 58 |
59 for (size_t j = 0; j < coefficients_length_; j += 4) { | 59 for (size_t j = 0; j < coefficients_length_; j += 4) { |
60 m_in = vld1q_f32(in_ptr + j); | 60 m_in = vld1q_f32(in_ptr + j); |
61 m_sum = vmlaq_f32(m_sum, m_in, vld1q_f32(coef_ptr + j)); | 61 m_sum = vmlaq_f32(m_sum, m_in, vld1q_f32(coef_ptr + j)); |
62 } | 62 } |
63 | 63 |
64 float32x2_t m_half = vadd_f32(vget_high_f32(m_sum), vget_low_f32(m_sum)); | 64 float32x2_t m_half = vadd_f32(vget_high_f32(m_sum), vget_low_f32(m_sum)); |
65 out[i] = vget_lane_f32(vpadd_f32(m_half, m_half), 0); | 65 out[i] = vget_lane_f32(vpadd_f32(m_half, m_half), 0); |
66 } | 66 } |
67 | 67 |
68 // Update current state. | 68 // Update current state. |
69 memmove(state_.get(), &state_[length], state_length_ * sizeof(state_[0])); | 69 memmove(state_.get(), &state_[length], state_length_ * sizeof(state_[0])); |
70 } | 70 } |
71 | 71 |
72 } // namespace webrtc | 72 } // namespace webrtc |
OLD | NEW |