mirror of
https://github.com/oxen-io/session-android.git
synced 2024-12-19 14:37:32 +00:00
165 lines
6.1 KiB
C++
165 lines
6.1 KiB
C++
|
/*
|
||
|
* Copyright (c) 2012 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.
|
||
|
*/
|
||
|
|
||
|
#include "webrtc/modules/audio_coding/neteq/sync_buffer.h"
|
||
|
|
||
|
#include "gtest/gtest.h"
|
||
|
|
||
|
namespace webrtc {
|
||
|
|
||
|
TEST(SyncBuffer, CreateAndDestroy) {
|
||
|
// Create a SyncBuffer with two channels and 10 samples each.
|
||
|
static const size_t kLen = 10;
|
||
|
static const size_t kChannels = 2;
|
||
|
SyncBuffer sync_buffer(kChannels, kLen);
|
||
|
EXPECT_EQ(kChannels, sync_buffer.Channels());
|
||
|
EXPECT_EQ(kLen, sync_buffer.Size());
|
||
|
// When the buffer is empty, the next index to play out is at the end.
|
||
|
EXPECT_EQ(kLen, sync_buffer.next_index());
|
||
|
// Verify that all elements are zero.
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
for (size_t i = 0; i < kLen; ++i) {
|
||
|
EXPECT_EQ(0, sync_buffer[channel][i]);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
TEST(SyncBuffer, SetNextIndex) {
|
||
|
// Create a SyncBuffer with two channels and 100 samples each.
|
||
|
static const size_t kLen = 100;
|
||
|
static const size_t kChannels = 2;
|
||
|
SyncBuffer sync_buffer(kChannels, kLen);
|
||
|
sync_buffer.set_next_index(0);
|
||
|
EXPECT_EQ(0u, sync_buffer.next_index());
|
||
|
sync_buffer.set_next_index(kLen / 2);
|
||
|
EXPECT_EQ(kLen / 2, sync_buffer.next_index());
|
||
|
sync_buffer.set_next_index(kLen);
|
||
|
EXPECT_EQ(kLen, sync_buffer.next_index());
|
||
|
// Try to set larger than the buffer size; should cap at buffer size.
|
||
|
sync_buffer.set_next_index(kLen + 1);
|
||
|
EXPECT_EQ(kLen, sync_buffer.next_index());
|
||
|
}
|
||
|
|
||
|
TEST(SyncBuffer, PushBackAndFlush) {
|
||
|
// Create a SyncBuffer with two channels and 100 samples each.
|
||
|
static const size_t kLen = 100;
|
||
|
static const size_t kChannels = 2;
|
||
|
SyncBuffer sync_buffer(kChannels, kLen);
|
||
|
static const size_t kNewLen = 10;
|
||
|
AudioMultiVector new_data(kChannels, kNewLen);
|
||
|
// Populate |new_data|.
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
for (size_t i = 0; i < kNewLen; ++i) {
|
||
|
new_data[channel][i] = i;
|
||
|
}
|
||
|
}
|
||
|
// Push back |new_data| into |sync_buffer|. This operation should pop out
|
||
|
// data from the front of |sync_buffer|, so that the size of the buffer
|
||
|
// remains the same. The |next_index_| should also move with the same length.
|
||
|
sync_buffer.PushBack(new_data);
|
||
|
ASSERT_EQ(kLen, sync_buffer.Size());
|
||
|
// Verify that |next_index_| moved accordingly.
|
||
|
EXPECT_EQ(kLen - kNewLen, sync_buffer.next_index());
|
||
|
// Verify the new contents.
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
for (size_t i = 0; i < kNewLen; ++i) {
|
||
|
EXPECT_EQ(new_data[channel][i],
|
||
|
sync_buffer[channel][sync_buffer.next_index() + i]);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
// Now flush the buffer, and verify that it is all zeros, and that next_index
|
||
|
// points to the end.
|
||
|
sync_buffer.Flush();
|
||
|
ASSERT_EQ(kLen, sync_buffer.Size());
|
||
|
EXPECT_EQ(kLen, sync_buffer.next_index());
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
for (size_t i = 0; i < kLen; ++i) {
|
||
|
EXPECT_EQ(0, sync_buffer[channel][i]);
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
TEST(SyncBuffer, PushFrontZeros) {
|
||
|
// Create a SyncBuffer with two channels and 100 samples each.
|
||
|
static const size_t kLen = 100;
|
||
|
static const size_t kChannels = 2;
|
||
|
SyncBuffer sync_buffer(kChannels, kLen);
|
||
|
static const size_t kNewLen = 10;
|
||
|
AudioMultiVector new_data(kChannels, kNewLen);
|
||
|
// Populate |new_data|.
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
for (size_t i = 0; i < kNewLen; ++i) {
|
||
|
new_data[channel][i] = 1000 + i;
|
||
|
}
|
||
|
}
|
||
|
sync_buffer.PushBack(new_data);
|
||
|
EXPECT_EQ(kLen, sync_buffer.Size());
|
||
|
|
||
|
// Push |kNewLen| - 1 zeros into each channel in the front of the SyncBuffer.
|
||
|
sync_buffer.PushFrontZeros(kNewLen - 1);
|
||
|
EXPECT_EQ(kLen, sync_buffer.Size()); // Size should remain the same.
|
||
|
// Verify that |next_index_| moved accordingly. Should be at the end - 1.
|
||
|
EXPECT_EQ(kLen - 1, sync_buffer.next_index());
|
||
|
// Verify the zeros.
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
for (size_t i = 0; i < kNewLen - 1; ++i) {
|
||
|
EXPECT_EQ(0, sync_buffer[channel][i]);
|
||
|
}
|
||
|
}
|
||
|
// Verify that the correct data is at the end of the SyncBuffer.
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
EXPECT_EQ(1000, sync_buffer[channel][sync_buffer.next_index()]);
|
||
|
}
|
||
|
}
|
||
|
|
||
|
TEST(SyncBuffer, GetNextAudioInterleaved) {
|
||
|
// Create a SyncBuffer with two channels and 100 samples each.
|
||
|
static const size_t kLen = 100;
|
||
|
static const size_t kChannels = 2;
|
||
|
SyncBuffer sync_buffer(kChannels, kLen);
|
||
|
static const size_t kNewLen = 10;
|
||
|
AudioMultiVector new_data(kChannels, kNewLen);
|
||
|
// Populate |new_data|.
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
for (size_t i = 0; i < kNewLen; ++i) {
|
||
|
new_data[channel][i] = i;
|
||
|
}
|
||
|
}
|
||
|
// Push back |new_data| into |sync_buffer|. This operation should pop out
|
||
|
// data from the front of |sync_buffer|, so that the size of the buffer
|
||
|
// remains the same. The |next_index_| should also move with the same length.
|
||
|
sync_buffer.PushBack(new_data);
|
||
|
|
||
|
// Read to interleaved output. Read in two batches, where each read operation
|
||
|
// should automatically update the |net_index_| in the SyncBuffer.
|
||
|
int16_t output[kChannels * kNewLen];
|
||
|
// Note that |samples_read| is the number of samples read from each channel.
|
||
|
// That is, the number of samples written to |output| is
|
||
|
// |samples_read| * |kChannels|.
|
||
|
size_t samples_read = sync_buffer.GetNextAudioInterleaved(kNewLen / 2,
|
||
|
output);
|
||
|
samples_read +=
|
||
|
sync_buffer.GetNextAudioInterleaved(kNewLen / 2,
|
||
|
&output[samples_read * kChannels]);
|
||
|
EXPECT_EQ(kNewLen, samples_read);
|
||
|
|
||
|
// Verify the data.
|
||
|
int16_t* output_ptr = output;
|
||
|
for (size_t i = 0; i < kNewLen; ++i) {
|
||
|
for (size_t channel = 0; channel < kChannels; ++channel) {
|
||
|
EXPECT_EQ(new_data[channel][i], *output_ptr);
|
||
|
++output_ptr;
|
||
|
}
|
||
|
}
|
||
|
}
|
||
|
|
||
|
} // namespace webrtc
|