blob: e908ef766de189a47666474dda7be643c43a4d8d [file] [log] [blame]
// Copyright 2012 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifdef UNSAFE_BUFFERS_BUILD
// TODO(crbug.com/40285824): Remove this and convert code to safer constructs.
#pragma allow_unsafe_buffers
#endif
#include "media/base/audio_bus.h"
#include <stddef.h>
#include <stdint.h>
#include <limits>
#include <memory>
#include "base/memory/aligned_memory.h"
#include "base/memory/raw_ptr.h"
#include "base/strings/stringprintf.h"
#include "base/test/bind.h"
#include "base/time/time.h"
#include "build/build_config.h"
#include "media/base/audio_parameters.h"
#include "media/base/audio_sample_types.h"
#include "media/base/channel_layout.h"
#include "media/base/fake_audio_render_callback.h"
#include "testing/gtest/include/gtest/gtest.h"
namespace media {
using AlignedFloatArray = base::AlignedHeapArray<float>;
static const int kChannels = 6;
static constexpr ChannelLayout kChannelLayout = CHANNEL_LAYOUT_5_1;
// Use a buffer size which is intentionally not a multiple of kChannelAlignment.
static const size_t kFrameCount = media::AudioBus::kChannelAlignment * 32 - 1;
static const int kSampleRate = 48000;
class AudioBusTest : public testing::Test {
public:
AudioBusTest() = default;
AudioBusTest(const AudioBusTest&) = delete;
AudioBusTest& operator=(const AudioBusTest&) = delete;
~AudioBusTest() override = default;
void VerifyChannelAndFrameCount(AudioBus* bus) {
EXPECT_EQ(kChannels, bus->channels());
EXPECT_EQ(static_cast<int>(kFrameCount), bus->frames());
}
void VerifySpanIsFilledWithValue(AudioBus::ConstChannel data, float value) {
for (size_t i = 0; i < data.size(); ++i) {
ASSERT_FLOAT_EQ(value, data[i]) << "i=" << i;
}
}
// Verify values for each channel in |result| are within |epsilon| of
// |expected|. If |epsilon| exactly equals 0, uses FLOAT_EQ macro.
void VerifyAreEqualWithEpsilon(const AudioBus* result,
const AudioBus* expected,
float epsilon) {
ASSERT_EQ(expected->channels(), result->channels());
ASSERT_EQ(expected->frames(), result->frames());
ASSERT_EQ(expected->is_bitstream_format(), result->is_bitstream_format());
if (expected->is_bitstream_format()) {
ASSERT_EQ(expected->GetBitstreamFrames(), result->GetBitstreamFrames());
ASSERT_EQ(expected->bitstream_data(), result->bitstream_data());
return;
}
for (int ch = 0; ch < result->channels(); ++ch) {
for (int i = 0; i < result->frames(); ++i) {
SCOPED_TRACE(base::StringPrintf("ch=%d, i=%d", ch, i));
if (epsilon == 0) {
ASSERT_FLOAT_EQ(expected->channel_span(ch)[i],
result->channel_span(ch)[i]);
} else {
ASSERT_NEAR(expected->channel_span(ch)[i],
result->channel_span(ch)[i], epsilon);
}
}
}
}
// Verify values for each channel in |result| against |expected|.
void VerifyAreEqual(const AudioBus* result, const AudioBus* expected) {
VerifyAreEqualWithEpsilon(result, expected, 0);
}
// Read and write to the full extent of the allocated channel data. Also test
// the Zero() method and verify it does as advertised. Also test data if data
// is 16-byte aligned as advertised (see kChannelAlignment in audio_bus.h).
void VerifyReadWriteAndAlignment(AudioBus* bus) {
int channel_count = 0;
for (auto channel : bus->AllChannels()) {
// Verify that the address returned by channel(i) is a multiple of
// AudioBus::kChannelAlignment.
ASSERT_TRUE(AudioBus::IsAligned(channel));
// Write into the channel buffer.
std::ranges::fill(channel, channel_count++);
}
for (int i = 0; i < bus->channels(); ++i) {
VerifySpanIsFilledWithValue(bus->channel_span(i), i);
}
bus->Zero();
for (int i = 0; i < bus->channels(); ++i) {
VerifySpanIsFilledWithValue(bus->channel_span(i), 0);
}
}
// Verify copying to and from |bus1| and |bus2|.
void CopyTest(AudioBus* bus1, AudioBus* bus2) {
// Fill |bus1| with dummy data.
if (bus1->is_bitstream_format()) {
uint8_t filler = 1;
std::ranges::generate(bus1->bitstream_data(),
[filler]() mutable { return filler++; });
} else {
int filler = 1;
for (auto channel : bus1->AllChannels()) {
std::ranges::fill(channel, filler);
}
}
// Verify copy from |bus1| to |bus2|.
bus2->Zero();
bus1->CopyTo(bus2);
VerifyAreEqual(bus1, bus2);
// Verify copy from |bus2| to |bus1|.
bus1->Zero();
bus2->CopyTo(bus1);
VerifyAreEqual(bus2, bus1);
}
protected:
void AllocateDataPerChannel() {
data_.reserve(kChannels);
for (int i = 0; i < kChannels; ++i) {
data_.push_back(
base::AlignedUninit<float>(kFrameCount, AudioBus::kChannelAlignment));
}
}
// TODO(crbug.com/373960632): remove this when all ctors take spans.
std::vector<float*> GetRawPointers(std::vector<AlignedFloatArray>& data) {
std::vector<float*> result;
result.reserve(data_.size());
for (AlignedFloatArray& array : data) {
result.push_back(array.as_span().data());
}
return result;
}
std::vector<AlignedFloatArray> data_;
};
// Verify basic Create(...) method works as advertised.
TEST_F(AudioBusTest, Create) {
std::unique_ptr<AudioBus> bus = AudioBus::Create(kChannels, kFrameCount);
VerifyChannelAndFrameCount(bus.get());
VerifyReadWriteAndAlignment(bus.get());
}
// Verify Create(...) using AudioParameters works as advertised.
TEST_F(AudioBusTest, CreateUsingAudioParameters) {
std::unique_ptr<AudioBus> bus = AudioBus::Create(
AudioParameters(AudioParameters::AUDIO_PCM_LINEAR,
ChannelLayoutConfig::FromLayout<kChannelLayout>(),
kSampleRate, kFrameCount));
VerifyChannelAndFrameCount(bus.get());
VerifyReadWriteAndAlignment(bus.get());
}
// Verify an AudioBus created via CreateWrapper(...) works as advertised.
TEST_F(AudioBusTest, CreateWrapper) {
AllocateDataPerChannel();
std::unique_ptr<AudioBus> bus = AudioBus::CreateWrapper(kChannels);
bus->set_frames(kFrameCount);
for (int i = 0; i < bus->channels(); ++i) {
bus->SetChannelData(i, data_[i].as_span());
}
bool deleted = false;
bus->SetWrappedDataDeleter(
base::BindLambdaForTesting([&]() { deleted = true; }));
VerifyChannelAndFrameCount(bus.get());
VerifyReadWriteAndAlignment(bus.get());
EXPECT_FALSE(deleted);
bus.reset();
EXPECT_TRUE(deleted);
}
TEST_F(AudioBusTest, AllChannels) {
AllocateDataPerChannel();
std::unique_ptr<AudioBus> bus = AudioBus::CreateWrapper(kChannels);
bus->set_frames(kFrameCount);
AudioBus::ChannelVector channels;
int value = 1;
for (AlignedFloatArray& data : data_) {
AudioBus::Channel channel(data);
// Fill each channel with a different value.
std::ranges::fill(channel, value++);
channels.push_back(channel);
}
bus->SetAllChannels(channels);
VerifyChannelAndFrameCount(bus.get());
// Verify looping through `AllChannels()` is equivalent to getting each
// channel individually.
int current_channel = 0;
for (auto channel : bus->AllChannels()) {
EXPECT_EQ(channel, bus->channel_span(current_channel++));
}
EXPECT_EQ(current_channel, kChannels);
}
TEST_F(AudioBusTest, AllChannelsSubspan) {
AllocateDataPerChannel();
std::unique_ptr<AudioBus> bus = AudioBus::CreateWrapper(kChannels);
bus->set_frames(kFrameCount);
AudioBus::ChannelVector channels;
int value = 1;
for (AlignedFloatArray& data : data_) {
AudioBus::Channel channel(data);
// Fill each sample with a different value.
for (float& sample : channel) {
sample = value++;
}
channels.push_back(channel);
}
bus->SetAllChannels(channels);
// Verify looping through `AllChannelsSubspan()` is equivalent to getting each
// channel individually and applying `subspan()` to them.
int current_channel = 0;
constexpr size_t kOffset = 3;
constexpr size_t kCount = 25;
for (auto channel : bus->AllChannelsSubspan(kOffset, kCount)) {
EXPECT_EQ(channel,
bus->channel_span(current_channel++).subspan(kOffset, kCount));
}
EXPECT_EQ(current_channel, kChannels);
}
// Verify an AudioBus created via wrapping a vector works as advertised.
TEST_F(AudioBusTest, WrapVector) {
AllocateDataPerChannel();
std::vector<float*> data_pointers;
data_pointers.reserve(kChannels);
for (AlignedFloatArray& data : data_) {
data_pointers.push_back(data.as_span().data());
}
std::unique_ptr<AudioBus> bus =
AudioBus::WrapVector(kFrameCount, GetRawPointers(data_));
VerifyChannelAndFrameCount(bus.get());
VerifyReadWriteAndAlignment(bus.get());
}
// Verify an AudioBus created via wrapping a memory block works as advertised.
TEST_F(AudioBusTest, WrapMemory) {
AudioParameters params(AudioParameters::AUDIO_PCM_LINEAR,
ChannelLayoutConfig::FromLayout<kChannelLayout>(),
kSampleRate, kFrameCount);
size_t total_frame_count =
AudioBus::CalculateMemorySize(params) / sizeof(float);
AlignedFloatArray data = base::AlignedUninit<float>(
total_frame_count, AudioBus::kChannelAlignment);
// Fill the memory with a test value we can check for after wrapping.
static const float kTestValue = 3;
std::ranges::fill(data, kTestValue);
std::unique_ptr<AudioBus> bus =
AudioBus::WrapMemory(params, data.as_span().data());
// Verify the test value we filled prior to wrapping.
for (auto channel : bus->AllChannels()) {
VerifySpanIsFilledWithValue(channel, kTestValue);
}
VerifyChannelAndFrameCount(bus.get());
VerifyReadWriteAndAlignment(bus.get());
// Verify the channel vectors lie within the provided memory block.
const float* backing_memory_start = data.as_span().data();
const float* backing_memory_end = backing_memory_start + total_frame_count;
EXPECT_GE(bus->channel_span(0).data(), backing_memory_start);
EXPECT_LT(bus->channel_span(bus->channels() - 1).data() + bus->frames(),
backing_memory_end);
}
// Simulate a shared memory transfer and verify results.
TEST_F(AudioBusTest, CopyTo) {
// Create one bus with AudioParameters and the other through direct values to
// test for parity between the Create() functions.
AudioParameters params(AudioParameters::AUDIO_PCM_LINEAR,
ChannelLayoutConfig::FromLayout<kChannelLayout>(),
kSampleRate, kFrameCount);
std::unique_ptr<AudioBus> bus1 = AudioBus::Create(kChannels, kFrameCount);
std::unique_ptr<AudioBus> bus2 = AudioBus::Create(params);
{
SCOPED_TRACE("Created");
CopyTest(bus1.get(), bus2.get());
}
{
SCOPED_TRACE("Wrapped Vector");
// Try a copy to an AudioBus wrapping a vector.
AllocateDataPerChannel();
bus2 = AudioBus::WrapVector(kFrameCount, GetRawPointers(data_));
CopyTest(bus1.get(), bus2.get());
}
{
SCOPED_TRACE("Wrapped Memory");
// Try a copy to an AudioBus wrapping a memory block.
std::unique_ptr<float, base::AlignedFreeDeleter> data(static_cast<float*>(
base::AlignedAlloc(AudioBus::CalculateMemorySize(params),
AudioBus::kChannelAlignment)));
bus2 = AudioBus::WrapMemory(params, data.get());
CopyTest(bus1.get(), bus2.get());
}
}
// Verify Zero() and ZeroFrames(...) utility methods work as advertised.
TEST_F(AudioBusTest, Zero) {
std::unique_ptr<AudioBus> bus = AudioBus::Create(kChannels, kFrameCount);
// Fill the bus with dummy data.
int value = 1;
for (auto channel : bus->AllChannels()) {
std::ranges::fill(channel, value++);
}
EXPECT_FALSE(bus->AreFramesZero());
// Zero first half the frames of each channel.
bus->ZeroFrames(kFrameCount / 2);
value = 1;
for (auto channel : bus->AllChannels()) {
SCOPED_TRACE("First Half Zero");
auto [first_half, second_half] = channel.split_at(kFrameCount / 2);
VerifySpanIsFilledWithValue(first_half, 0);
VerifySpanIsFilledWithValue(second_half, value++);
}
EXPECT_FALSE(bus->AreFramesZero());
// Fill the bus with dummy data.
value = 1;
for (auto channel : bus->AllChannels()) {
std::ranges::fill(channel, value++);
}
// Zero the last half of the frames.
static constexpr size_t kRemainingFrames = kFrameCount - kFrameCount / 2;
bus->ZeroFramesPartial(kFrameCount / 2, kRemainingFrames);
value = 1;
for (auto channel : bus->AllChannels()) {
SCOPED_TRACE("Last Half Zero");
auto [first_half, second_half] = channel.split_at(kFrameCount / 2);
VerifySpanIsFilledWithValue(first_half, value++);
VerifySpanIsFilledWithValue(second_half, 0);
}
EXPECT_FALSE(bus->AreFramesZero());
// Fill the bus with dummy data.
value = 1;
for (auto channel : bus->AllChannels()) {
std::ranges::fill(channel, value++);
}
// Zero all the frames of each channel.
bus->Zero();
for (auto channel : bus->AllChannels()) {
SCOPED_TRACE("All Zero");
VerifySpanIsFilledWithValue(channel, 0);
}
EXPECT_TRUE(bus->AreFramesZero());
}
// Each test vector represents two channels of data in the following arbitrary
// layout: <min, zero, max, min, max / 2, min / 2, zero, max, zero, zero>.
static constexpr int kTestVectorSize = 10;
static const uint8_t kTestVectorUint8[kTestVectorSize] = {
0, -INT8_MIN, UINT8_MAX,
0, INT8_MAX / 2 + 128, INT8_MIN / 2 + 128,
-INT8_MIN, UINT8_MAX, -INT8_MIN,
-INT8_MIN};
static const int16_t kTestVectorInt16[kTestVectorSize] = {
INT16_MIN, 0, INT16_MAX, INT16_MIN, INT16_MAX / 2,
INT16_MIN / 2, 0, INT16_MAX, 0, 0};
static const int32_t kTestVectorInt32[kTestVectorSize] = {
INT32_MIN, 0, INT32_MAX, INT32_MIN, INT32_MAX / 2,
INT32_MIN / 2, 0, INT32_MAX, 0, 0};
static const float kTestVectorFloat32[kTestVectorSize] = {
-1.0f, 0.0f, 1.0f, -1.0f, 0.5f, -0.5f, 0.0f, 1.0f, 0.0f, 0.0f};
// This is based on kTestVectorFloat32, but has some of the values outside of
// sanity.
static const float kTestVectorFloat32Invalid[kTestVectorSize] = {
-5.0f,
0.0f,
5.0f,
-1.0f,
0.5f,
-0.5f,
0.0f,
std::numeric_limits<float>::infinity(),
std::numeric_limits<float>::signaling_NaN(),
std::numeric_limits<float>::quiet_NaN()};
static const float kTestVectorFloat32Sanitized[kTestVectorSize] = {
-1.0f, 0.0f, 1.0f, -1.0f, 0.5f, -0.5f, 0.0f, 1.0f, 0.0f, 0.0f};
// Expected results.
static constexpr size_t kTestVectorFrameCount = kTestVectorSize / 2;
static const float kTestVectorResult[][kTestVectorFrameCount] = {
{-1.0f, 1.0f, 0.5f, 0.0f, 0.0f},
{0.0f, -1.0f, -0.5f, 1.0f, 0.0f}};
static const int kTestVectorChannelCount = std::size(kTestVectorResult);
// Verify FromInterleaved() deinterleaves audio in supported formats correctly.
TEST_F(AudioBusTest, FromInterleaved) {
std::unique_ptr<AudioBus> bus =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
std::unique_ptr<AudioBus> expected =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
for (int ch = 0; ch < kTestVectorChannelCount; ++ch) {
expected->channel_span(ch).copy_from(base::span(kTestVectorResult[ch]));
}
{
SCOPED_TRACE("UnsignedInt8SampleTypeTraits");
bus->Zero();
bus->FromInterleaved<UnsignedInt8SampleTypeTraits>(kTestVectorUint8,
kTestVectorFrameCount);
// Biased uint8_t calculations have poor precision, so the epsilon here is
// slightly more permissive than int16_t and int32_t calculations.
VerifyAreEqualWithEpsilon(bus.get(), expected.get(),
1.0f / (std::numeric_limits<uint8_t>::max() - 1));
}
{
SCOPED_TRACE("SignedInt16SampleTypeTraits");
bus->Zero();
bus->FromInterleaved<SignedInt16SampleTypeTraits>(kTestVectorInt16,
kTestVectorFrameCount);
VerifyAreEqualWithEpsilon(
bus.get(), expected.get(),
1.0f /
(static_cast<float>(std::numeric_limits<uint16_t>::max()) + 1.0f));
}
{
SCOPED_TRACE("SignedInt32SampleTypeTraits");
bus->Zero();
bus->FromInterleaved<SignedInt32SampleTypeTraits>(kTestVectorInt32,
kTestVectorFrameCount);
VerifyAreEqualWithEpsilon(
bus.get(), expected.get(),
1.0f / static_cast<float>(std::numeric_limits<uint32_t>::max()));
}
{
SCOPED_TRACE("Float32SampleTypeTraits");
bus->Zero();
bus->FromInterleaved<Float32SampleTypeTraits>(kTestVectorFloat32,
kTestVectorFrameCount);
VerifyAreEqual(bus.get(), expected.get());
}
}
// Verify FromInterleavedPartial() deinterleaves audio correctly.
TEST_F(AudioBusTest, FromInterleavedPartial) {
// Only deinterleave the middle two frames in each channel.
static constexpr size_t kPartialStart = 1;
static constexpr size_t kPartialFrames = 2;
ASSERT_LE(kPartialStart + kPartialFrames, kTestVectorFrameCount);
std::unique_ptr<AudioBus> bus =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
std::unique_ptr<AudioBus> expected =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
expected->Zero();
int current_channel = 0;
for (auto partial_channel :
expected->AllChannelsSubspan(kPartialStart, kPartialFrames)) {
partial_channel.copy_from(base::span(kTestVectorResult[current_channel++])
.subspan(kPartialStart, kPartialFrames));
}
{
SCOPED_TRACE("SignedInt32SampleTypeTraits");
bus->Zero();
bus->FromInterleavedPartial<SignedInt32SampleTypeTraits>(
kTestVectorInt32 + kPartialStart * bus->channels(), kPartialStart,
kPartialFrames);
VerifyAreEqual(bus.get(), expected.get());
}
}
// Verify ToInterleaved() interleaves audio in supported formats correctly.
TEST_F(AudioBusTest, ToInterleaved) {
std::unique_ptr<AudioBus> bus =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
// Fill the bus with our test vector.
for (int ch = 0; ch < bus->channels(); ++ch) {
bus->channel_span(ch).copy_from(base::span(kTestVectorResult[ch]));
}
{
SCOPED_TRACE("UnsignedInt8SampleTypeTraits");
uint8_t test_array[std::size(kTestVectorUint8)];
bus->ToInterleaved<UnsignedInt8SampleTypeTraits>(bus->frames(), test_array);
ASSERT_EQ(0,
memcmp(test_array, kTestVectorUint8, sizeof(kTestVectorUint8)));
}
{
SCOPED_TRACE("SignedInt16SampleTypeTraits");
int16_t test_array[std::size(kTestVectorInt16)];
bus->ToInterleaved<SignedInt16SampleTypeTraits>(bus->frames(), test_array);
ASSERT_EQ(0,
memcmp(test_array, kTestVectorInt16, sizeof(kTestVectorInt16)));
}
{
SCOPED_TRACE("SignedInt32SampleTypeTraits");
int32_t test_array[std::size(kTestVectorInt32)];
bus->ToInterleaved<SignedInt32SampleTypeTraits>(bus->frames(), test_array);
// Some compilers get better precision than others on the half-max test, so
// let the test pass with an off by one check on the half-max.
int32_t alternative_acceptable_result[std::size(kTestVectorInt32)];
memcpy(alternative_acceptable_result, kTestVectorInt32,
sizeof(kTestVectorInt32));
ASSERT_EQ(alternative_acceptable_result[4],
std::numeric_limits<int32_t>::max() / 2);
alternative_acceptable_result[4]++;
ASSERT_TRUE(
memcmp(test_array, kTestVectorInt32, sizeof(kTestVectorInt32)) == 0 ||
memcmp(test_array, alternative_acceptable_result,
sizeof(alternative_acceptable_result)) == 0);
}
{
SCOPED_TRACE("Float32SampleTypeTraits");
float test_array[std::size(kTestVectorFloat32)];
bus->ToInterleaved<Float32SampleTypeTraits>(bus->frames(), test_array);
ASSERT_EQ(
0, memcmp(test_array, kTestVectorFloat32, sizeof(kTestVectorFloat32)));
}
}
TEST_F(AudioBusTest, ToInterleavedSanitized) {
std::unique_ptr<AudioBus> bus =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
bus->FromInterleaved<Float32SampleTypeTraits>(kTestVectorFloat32Invalid,
bus->frames());
// Verify FromInterleaved applied no sanity.
ASSERT_EQ(bus->channel_span(0)[0], kTestVectorFloat32Invalid[0]);
float test_array[std::size(kTestVectorFloat32Sanitized)];
bus->ToInterleaved<Float32SampleTypeTraits>(bus->frames(), test_array);
for (size_t i = 0; i < std::size(kTestVectorFloat32Sanitized); ++i)
ASSERT_EQ(kTestVectorFloat32Sanitized[i], test_array[i]);
// Verify that Float32SampleTypeTraitsNoClip applied no sanity. Note: We don't
// use memcmp() here since the NaN type may change on x86 platforms in certain
// circumstances, see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57484
bus->ToInterleaved<Float32SampleTypeTraitsNoClip>(bus->frames(), test_array);
for (int i = 0; i < kTestVectorSize; ++i) {
if (std::isnan(test_array[i]))
EXPECT_TRUE(std::isnan(kTestVectorFloat32Invalid[i]));
else
EXPECT_FLOAT_EQ(test_array[i], kTestVectorFloat32Invalid[i]);
}
}
TEST_F(AudioBusTest, CopyAndClipTo) {
auto bus = AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
bus->FromInterleaved<Float32SampleTypeTraits>(kTestVectorFloat32Invalid,
bus->frames());
auto expected =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
expected->FromInterleaved<Float32SampleTypeTraits>(
kTestVectorFloat32Sanitized, bus->frames());
// Verify FromInterleaved applied no sanity.
ASSERT_EQ(bus->channel_span(0)[0], kTestVectorFloat32Invalid[0]);
std::unique_ptr<AudioBus> copy_to_bus =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
bus->CopyAndClipTo(copy_to_bus.get());
for (int ch = 0; ch < expected->channels(); ++ch) {
for (int i = 0; i < expected->frames(); ++i)
ASSERT_EQ(copy_to_bus->channel_span(ch)[i],
expected->channel_span(ch)[i]);
}
ASSERT_EQ(expected->channels(), copy_to_bus->channels());
ASSERT_EQ(expected->frames(), copy_to_bus->frames());
ASSERT_EQ(expected->is_bitstream_format(),
copy_to_bus->is_bitstream_format());
}
// Verify ToInterleavedPartial() interleaves audio correctly.
TEST_F(AudioBusTest, ToInterleavedPartial) {
// Only interleave the middle two frames in each channel.
static constexpr size_t kPartialStart = 1;
static constexpr size_t kPartialFrames = 2;
ASSERT_LE(kPartialStart + kPartialFrames, kTestVectorFrameCount);
std::unique_ptr<AudioBus> expected =
AudioBus::Create(kTestVectorChannelCount, kTestVectorFrameCount);
for (int ch = 0; ch < kTestVectorChannelCount; ++ch) {
expected->channel_span(ch).copy_from(base::span(kTestVectorResult[ch]));
}
{
SCOPED_TRACE("Float32SampleTypeTraits");
float test_array[std::size(kTestVectorFloat32)];
expected->ToInterleavedPartial<Float32SampleTypeTraits>(
kPartialStart, kPartialFrames, test_array);
ASSERT_EQ(0, memcmp(test_array, kTestVectorFloat32 +
kPartialStart * kTestVectorChannelCount,
kPartialFrames * sizeof(*kTestVectorFloat32) *
kTestVectorChannelCount));
}
}
struct ZeroingOutTestData {
static constexpr int kChannelCount = 2;
static constexpr int kFrameCount = 10;
static constexpr int kInterleavedFrameCount = 3;
std::unique_ptr<AudioBus> bus_under_test;
std::vector<float> interleaved_dummy_frames;
ZeroingOutTestData() {
// Create a bus and fill each channel with a test pattern of form
// [1.0, 2.0, 3.0, ...]
bus_under_test = AudioBus::Create(kChannelCount, kFrameCount);
for (auto channel : bus_under_test->AllChannels()) {
int value = 1;
for (float& sample : channel) {
sample = value++;
}
}
// Create a vector containing dummy interleaved samples.
static const float kDummySampleValue = 0.123f;
interleaved_dummy_frames.resize(kChannelCount * kInterleavedFrameCount);
std::fill(interleaved_dummy_frames.begin(), interleaved_dummy_frames.end(),
kDummySampleValue);
}
};
TEST_F(AudioBusTest, FromInterleavedZerosOutUntouchedFrames) {
ZeroingOutTestData test_data;
// Exercise
test_data.bus_under_test->FromInterleaved<Float32SampleTypeTraits>(
&test_data.interleaved_dummy_frames[0], test_data.kInterleavedFrameCount);
const size_t untouched_frame_count =
test_data.kFrameCount - test_data.kInterleavedFrameCount;
// Verification
for (auto partial_channel : test_data.bus_under_test->AllChannelsSubspan(
test_data.kInterleavedFrameCount, untouched_frame_count)) {
for (auto sample : partial_channel) {
ASSERT_EQ(0.0f, sample);
}
}
}
TEST_F(AudioBusTest, FromInterleavedPartialDoesNotZeroOutUntouchedFrames) {
{
SCOPED_TRACE("Zero write offset");
ZeroingOutTestData test_data;
static const int kWriteOffsetInFrames = 0;
// Exercise
test_data.bus_under_test->FromInterleavedPartial<Float32SampleTypeTraits>(
&test_data.interleaved_dummy_frames[0], kWriteOffsetInFrames,
test_data.kInterleavedFrameCount);
// Verification
for (int ch = 0; ch < test_data.kChannelCount; ++ch) {
auto sample_array_for_current_channel =
test_data.bus_under_test->channel_span(ch);
for (int frame_index =
test_data.kInterleavedFrameCount + kWriteOffsetInFrames;
frame_index < test_data.kFrameCount; frame_index++) {
ASSERT_EQ(frame_index + 1,
sample_array_for_current_channel[frame_index]);
}
}
}
{
SCOPED_TRACE("Positive write offset");
ZeroingOutTestData test_data;
static const int kWriteOffsetInFrames = 2;
// Exercise
test_data.bus_under_test->FromInterleavedPartial<Float32SampleTypeTraits>(
&test_data.interleaved_dummy_frames[0], kWriteOffsetInFrames,
test_data.kInterleavedFrameCount);
// Verification
for (int ch = 0; ch < test_data.kChannelCount; ++ch) {
auto sample_array_for_current_channel =
test_data.bus_under_test->channel_span(ch);
// Check untouched frames before write offset
for (int frame_index = 0; frame_index < kWriteOffsetInFrames;
frame_index++) {
ASSERT_EQ(frame_index + 1,
sample_array_for_current_channel[frame_index]);
}
// Check untouched frames after write
for (int frame_index =
test_data.kInterleavedFrameCount + kWriteOffsetInFrames;
frame_index < test_data.kFrameCount; frame_index++) {
ASSERT_EQ(frame_index + 1,
sample_array_for_current_channel[frame_index]);
}
}
}
}
TEST_F(AudioBusTest, Scale) {
std::unique_ptr<AudioBus> bus = AudioBus::Create(kChannels, kFrameCount);
// Fill the bus with dummy data.
static constexpr float kFillValue = 1;
for (auto channel : bus->AllChannels()) {
std::ranges::fill(channel, kFillValue);
}
// Adjust by an invalid volume and ensure volume is unchanged.
bus->Scale(-1);
for (auto channel : bus->AllChannels()) {
SCOPED_TRACE("Invalid Scale");
VerifySpanIsFilledWithValue(channel, kFillValue);
}
// Verify correct volume adjustment.
static constexpr float kVolume = 0.5;
bus->Scale(kVolume);
for (auto channel : bus->AllChannels()) {
SCOPED_TRACE("Half Scale");
VerifySpanIsFilledWithValue(channel, kFillValue * kVolume);
}
// Verify zero volume case.
bus->Scale(0);
for (auto channel : bus->AllChannels()) {
SCOPED_TRACE("Zero Scale");
VerifySpanIsFilledWithValue(channel, 0);
}
}
TEST_F(AudioBusTest, Bitstream) {
static const size_t kDataSize = kFrameCount / 2;
std::unique_ptr<AudioBus> bus = AudioBus::Create(1, kFrameCount);
EXPECT_FALSE(bus->is_bitstream_format());
bus->set_is_bitstream_format(true);
EXPECT_TRUE(bus->is_bitstream_format());
EXPECT_EQ(0u, bus->bitstream_data().size());
bus->SetBitstreamSize(kDataSize);
EXPECT_EQ(kDataSize, bus->bitstream_data().size());
EXPECT_EQ(0, bus->GetBitstreamFrames());
bus->SetBitstreamFrames(kFrameCount);
EXPECT_EQ(static_cast<int>(kFrameCount), bus->GetBitstreamFrames());
std::unique_ptr<AudioBus> bus2 = AudioBus::Create(1, kFrameCount);
CopyTest(bus.get(), bus2.get());
bus->Zero();
EXPECT_EQ(0u, bus->bitstream_data().size());
EXPECT_EQ(0, bus->GetBitstreamFrames());
}
} // namespace media