Files
palemoon27/dom/media/platforms/wmf/WMFAudioMFTManager.cpp
T
roytam1 93f846cd1f import changes from `dev' branch of rmottola/Arctic-Fox:
- Bug 1275016 - Rename Endian.h to EndianUtils.h to avoid #include confusion with Android's endian.h stdlib header. r=froydnj (b54a25f572)
- add crashreporter stuff (aa7ef15337)
- Bug 1261168 - Add AlignedAutoTArray type in Web Audio; r=padenot (285d2cb88b)
- Bug 1273390. Part 1 - move some functions to private. r=jya. (07a3037e59)
- Bug 1273390. Part 2 - add assertions. r=jya. (2cae7c596a)
- Bug 1273390. Part 3 - rename some functions to be consistent with other sub-classes of MediaDataDecoder. r=jya. (c48c7060ce)
- Bug 1273390. Part 4 - remove use of FlushableTaskQueue::Flush(). r=jya. (00565a65f4)
- Bug 1273390. Part 5 - remove use of FlushableTaskQueue. r=jya. (30600b204e)
- Bug 1273774. Part 1 - remove unused members and thread assertions. r=jya (f5177ed641)
- Bug 1273774. Part 2 - do decoding jobs synchronously without dispatching. r=jya. (62d840d27c)
- Bug 1273774. Part 3 - remove members no longer used. r=jya. (e957ca512a)
- Bug 1244410: [ffmpeg] Ensure the last drained frame has the proper duration set. r=gerald (d5521bfdd4)
- Bug 1271508. Part 1 - refactor FFmpegAudioDecoder code to be similar to FFmpegVideoDecoder::Input() so it would be easier to extract common code to the parent class. r=jya. (613e6c624c)
- Bug 1271508. Part 2 - rename functions so they are the same as those of FFmpegAudioDecoder so it would be easier to extract common code to the parent class. r=jya. (cb281cba26)
- Bug 1270350 - per comment 0, use SyncRunnable to repalce the boilerplate code. r=jya. (b99460e571)
- Bug 1271508. Part 3 - extract code to the parent class and remove use of mTaskQueue from sub-classes. r=jya. (2a7ff4dd1e)
- Bug 1274216 - remove use of FlushableTaskQueue from PlatformDecoderModule. r=jya. (eb160c5fa2)
- Bug 1271517. Part 1 - remove use of FlushableTaskQueue::Flush() from FFmpegDataDecoder::Flush(). r=jya. (fdf10da4ab)
- Bug 1271517. Part 2 - remove use of FlushableTaskQueue. r=jya. (a7016d8506)
- Bug 1273397. Part 1 - rename some functions to be consistent with other MediaDataDecoder sub-classes. r=jya. (7eecb164be)
- Bug 1273397. Part 2 - constify some members. r=jya. (e4482f9a23)
- Bug 1273397. Part 3 - remove use of FlushableTaskQueue::Flush(). r=jya. (0b7ee073fe)
- Bug 1273397. Part 4 - remove use of FlushableTaskQueue. r=jya. (6a612161d5)
- Bug 1273397. Part 5 - add assertions. r=jya. (ff3a62a6fb)
- Bug 1274199 - remove use of FlushableTaskQueue. r=cpearce. (adc4c84ede)
- Bug 1273405. Part 1 - rename some functions to be consistent with other MediaDataDecoder sub-classes. r=jya. (af123d6c21)
- Bug 1273405. Part 2 - remove use of FlushableTaskQueue::Flush(). r=jya. (2d144bfbcd)
- Bug 1273405. Part 3 - remove use of FlushableTaskQueue. r=jya. (1e9ea3c2c7)
- Bug 1273405. Part 4 - add assertions. r=jya. (b400647323)
- Bug 1271491: [WMF] P1. Don't use main thread only preferences methods. r=cpearce (7177454dfb)
- Bug 1262427. Don't try D3D11 harder. r=dvander (404147d6fa)
- Use gfxConfig for D3D9 preferences. (bug 1270650, r=jrmuizel) (40d89c154c)
- Bug 1271491: P2. Allow initialization of WMFPlatformDecoderModule from any threads. r=mattwoodrow (c8fe0bf009)
- Bug 1271491: P3. Remove refcounting the number of time apple's linkers are called. r=cpearce (0324ffe876)
- Bug 1271491: [ffmpeg] P4. Remove requirements to call Init on the main thread. r=cpearce (b511d7dfd5)
- Bug 1271491: [GMP] P5. Allow GMPDecoderModule::Init() to be called off the main thread. r=cpearce (2131eb0b2e)
- Bug 1266102 - Don't run VP9 benchmark on Android r=jya (57d7b573fe)
- Bug 1271491: P6. Remove the need to call PDMFactory::Init(). r=cpearce (5726cfe49c)
- Bug 1271491: P7. Remove unused members. r=alfredo (0f8a9dde73)
- Bug 1268905 - Disable D3D11 with some Toshiba DLLs - r=cpearce (b5bf77442e)
- Bug 1269204 - Disable D3D11 with idg10umd32 9.17.10.2857 - r=cpearce (7eb6a3d96b)
- Bug 1273406 - Disable D3D11 with some iSonyVideoProcessor DLLs - r=cpearce (d9b6f0cefe)
- Bug 1273406 - Ugly macros transform into beautiful constexpr goodness - r=cpearce (0671483695)
- Bug 1273691 - Implement 'media.wmf.disable-d3d11-for-dlls' pref - r=cpearce (193ae53070)
- Bug 1272225. Part 1 - add assertions to make thread constraints clear. r=jya. (83c620470e)
- Bug 1272225. Part 2 - remove use of FlushableTaskQueue::Flush(). r=jya. (9473e092d1)
- Bug 1272553. Part 1 - move code around to be able to extract common code in P2. r=jya. (d727f97ee8)
- Bug 1272553. Part 2 - extract common code to the parent class. r=jya. (2fb3cd4bd9)
- Bug 1272553. Part 3 - make mTaskQueue private. r=jya. (93fea98cb6)
- Bug 1272232. Part 1 - move code around so we can extract common code in P2. r=jya. (8cdaab9078)
- Bug 1272232. Part 2 - extract common code to the parent class. r=jya. (27156668b3)
- Bug 1272232. Part 3 - constify some members and make them private when possible. r=jya. (550b963d97)
- Bug 1272232. Part 4 - remove use of FlushableTaskQueue::Flush(). r=jya. (bdbfdeb6bc)
- Bug 1272232. Part 5 - remove use of FlushableTaskQueue. r=jya. (640f889a9d)
- Bug 1274913 - Move PDM log definition to header. r=njn (823b07f42b)
- Bug 1275538: P1. Abort early if a skip request is in progress. r=gerald,kamidphish (d67b8a2236)
- Bug 1272422 - Part 1: Expose control of suspending background video. r=cpearce (ec7193773f)
- Bug 1272422 - Part 2: Vidoe -> Video. r=cpearce (97390aee69)
- Bug 1272422 - Part 3: Don't reset audio queue. r=jya (e183db1062)
- Bug 1272964: P1. Only activate skip to next keyframe logic when next keyframe time is known. r=gerald (1be74df027)
- Bug 1272964: P2. Don't activate skip to next keyframe until we passed the internal seek target. r=gerald (c55b6ae003)
- Bug 1258922: [MSE] P1. Initialise variable. r=gerald (56a5acb345)
- Bug 1258922: [MSE] P2. Do not go over gap when attempting to find the next key frame. r=gerald (db1319f080)
- Bug 1258922: [MSE] P3. Check that the data we are attempting to skip to is buffered. r=gerald (621d71d5d6)
- Bug 1258922: [MSE] P4. Set draining flag to true when skip to next keyframe failed. r=gerald (6c75613faf)
- Bug 1272916: [MSE] P1. Don't rely only on dts gap to establish if we have a gap in our source buffer. r=gerald (8770113b83)
- Bug 1272964: [MSE] P3. Do not skip over gaps when searching for the next keyframe. r=gerald (76916c5ac6)
- Bug 1272964: P4. Only flush decoder if skip to next keyframe actually succeeds. r=cpearce (5394708eef)
- Bug 1270323: P1. Don't reset flag indicating that new data was received. r=cpearce (d32f06ef34)
- Bug 1270323: P2. Don't process new incoming data while a skip to next keyframe is pending. r=cpearce (bca7909de9)
- Bug 1270323: [ffmpeg] P3. Use the dts of the last sample input, not the dts of the last decoded sample (0d768c33ef)
- Bug 1270323: P4. Don't drain decoder if we're already waiting for new data. r=cpearce (679302cb6e)
- Bug 1270323: P5. Prevent potential null deref. r=cpearce (cc63270e06)
- Bug 1275538: P2. Drop decoded frames that we know are already too late. r=kamidphish (4e7af9398c)
- Bug 1273018: P1. Rename some members. r=gerald (3a92fbd994)
- Bug 1273018: P2. Don't reject audio waiting promise when performing a video only seek. r=gerald (34e4988db1)
- Bug 1273018: P3. Adjust range of audio assertions. r=gerald (feb2afd0ae)
- Bug 1249706 - Backout a085ea2d24bb for blowing telemetry server's mind. r=backout (d61fb51f52)
- Bug 1249706 - Fix 8fe22dd4fc8a (backout of a085ea2d24bb). r=bustage (ba65251db7)
- Bug 1272964: [MSE] P5. Default to skipping to the next keyframe if no keyframe was found past currentTime. (29086fcf56)
- Bug 1272964: P6. Exclude frames dropped due to internal seeking from calculations. r=cpearce (bf6faa7612)
- Bug 1068151 - keep decoding a corrupted video. r=jya (3b5462e5b6)
- Bug 1273947 - Update ResetDecode() to ResetDecode(TargetQueue) r=jya (6c28d46974)
- Bug 1277508: P1. Don't attempt to demux new samples while we're currently draining. r=kamidphish (64f200b921)
- Bug 1274933: Reject data promise when EOS is encountered following waiting for data. r=gerald (5bba4a7853)
- Bug 1277508: P2. Add HasPendingDrain convenience method. r=kamidphish (3d89a90a97)
2024-10-08 23:23:56 +08:00

345 lines
11 KiB
C++

/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim:set ts=2 sw=2 sts=2 et cindent: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "WMFAudioMFTManager.h"
#include "MediaInfo.h"
#include "VideoUtils.h"
#include "WMFUtils.h"
#include "nsTArray.h"
#include "TimeUnits.h"
#include "mozilla/Telemetry.h"
#include "mozilla/Logging.h"
#define LOG(...) MOZ_LOG(sPDMLog, mozilla::LogLevel::Debug, (__VA_ARGS__))
namespace mozilla {
static void
AACAudioSpecificConfigToUserData(uint8_t aAACProfileLevelIndication,
const uint8_t* aAudioSpecConfig,
uint32_t aConfigLength,
nsTArray<BYTE>& aOutUserData)
{
MOZ_ASSERT(aOutUserData.IsEmpty());
// The MF_MT_USER_DATA for AAC is defined here:
// http://msdn.microsoft.com/en-us/library/windows/desktop/dd742784%28v=vs.85%29.aspx
//
// For MFAudioFormat_AAC, MF_MT_USER_DATA contains the portion of
// the HEAACWAVEINFO structure that appears after the WAVEFORMATEX
// structure (that is, after the wfx member). This is followed by
// the AudioSpecificConfig() data, as defined by ISO/IEC 14496-3.
// [...]
// The length of the AudioSpecificConfig() data is 2 bytes for AAC-LC
// or HE-AAC with implicit signaling of SBR/PS. It is more than 2 bytes
// for HE-AAC with explicit signaling of SBR/PS.
//
// The value of audioObjectType as defined in AudioSpecificConfig()
// must be 2, indicating AAC-LC. The value of extensionAudioObjectType
// must be 5 for SBR or 29 for PS.
//
// HEAACWAVEINFO structure:
// typedef struct heaacwaveinfo_tag {
// WAVEFORMATEX wfx;
// WORD wPayloadType;
// WORD wAudioProfileLevelIndication;
// WORD wStructType;
// WORD wReserved1;
// DWORD dwReserved2;
// }
const UINT32 heeInfoLen = 4 * sizeof(WORD) + sizeof(DWORD);
// The HEAACWAVEINFO must have payload and profile set,
// the rest can be all 0x00.
BYTE heeInfo[heeInfoLen] = {0};
WORD* w = (WORD*)heeInfo;
w[0] = 0x0; // Payload type raw AAC packet
w[1] = aAACProfileLevelIndication;
aOutUserData.AppendElements(heeInfo, heeInfoLen);
aOutUserData.AppendElements(aAudioSpecConfig, aConfigLength);
}
WMFAudioMFTManager::WMFAudioMFTManager(
const AudioInfo& aConfig)
: mAudioChannels(aConfig.mChannels)
, mAudioRate(aConfig.mRate)
, mAudioFrameSum(0)
, mMustRecaptureAudioPosition(true)
{
MOZ_COUNT_CTOR(WMFAudioMFTManager);
if (aConfig.mMimeType.EqualsLiteral("audio/mpeg")) {
mStreamType = MP3;
} else if (aConfig.mMimeType.EqualsLiteral("audio/mp4a-latm")) {
mStreamType = AAC;
AACAudioSpecificConfigToUserData(aConfig.mProfile,
aConfig.mCodecSpecificConfig->Elements(),
aConfig.mCodecSpecificConfig->Length(),
mUserData);
} else {
mStreamType = Unknown;
}
}
WMFAudioMFTManager::~WMFAudioMFTManager()
{
MOZ_COUNT_DTOR(WMFAudioMFTManager);
}
const GUID&
WMFAudioMFTManager::GetMFTGUID()
{
MOZ_ASSERT(mStreamType != Unknown);
switch (mStreamType) {
case AAC: return CLSID_CMSAACDecMFT;
case MP3: return CLSID_CMP3DecMediaObject;
default: return GUID_NULL;
};
}
const GUID&
WMFAudioMFTManager::GetMediaSubtypeGUID()
{
MOZ_ASSERT(mStreamType != Unknown);
switch (mStreamType) {
case AAC: return MFAudioFormat_AAC;
case MP3: return MFAudioFormat_MP3;
default: return GUID_NULL;
};
}
bool
WMFAudioMFTManager::Init()
{
NS_ENSURE_TRUE(mStreamType != Unknown, false);
RefPtr<MFTDecoder> decoder(new MFTDecoder());
HRESULT hr = decoder->Create(GetMFTGUID());
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
// Setup input/output media types
RefPtr<IMFMediaType> inputType;
hr = wmf::MFCreateMediaType(getter_AddRefs(inputType));
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = inputType->SetGUID(MF_MT_MAJOR_TYPE, MFMediaType_Audio);
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = inputType->SetGUID(MF_MT_SUBTYPE, GetMediaSubtypeGUID());
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = inputType->SetUINT32(MF_MT_AUDIO_SAMPLES_PER_SECOND, mAudioRate);
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = inputType->SetUINT32(MF_MT_AUDIO_NUM_CHANNELS, mAudioChannels);
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
if (mStreamType == AAC) {
hr = inputType->SetUINT32(MF_MT_AAC_PAYLOAD_TYPE, 0x0); // Raw AAC packet
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = inputType->SetBlob(MF_MT_USER_DATA,
mUserData.Elements(),
mUserData.Length());
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
}
RefPtr<IMFMediaType> outputType;
hr = wmf::MFCreateMediaType(getter_AddRefs(outputType));
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = outputType->SetGUID(MF_MT_MAJOR_TYPE, MFMediaType_Audio);
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = outputType->SetGUID(MF_MT_SUBTYPE, MFAudioFormat_PCM);
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = outputType->SetUINT32(MF_MT_AUDIO_BITS_PER_SAMPLE, 16);
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
hr = decoder->SetMediaTypes(inputType, outputType);
NS_ENSURE_TRUE(SUCCEEDED(hr), false);
mDecoder = decoder;
return true;
}
HRESULT
WMFAudioMFTManager::Input(MediaRawData* aSample)
{
return mDecoder->Input(aSample->Data(),
uint32_t(aSample->Size()),
aSample->mTime);
}
HRESULT
WMFAudioMFTManager::UpdateOutputType()
{
HRESULT hr;
RefPtr<IMFMediaType> type;
hr = mDecoder->GetOutputMediaType(type);
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
hr = type->GetUINT32(MF_MT_AUDIO_SAMPLES_PER_SECOND, &mAudioRate);
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
hr = type->GetUINT32(MF_MT_AUDIO_NUM_CHANNELS, &mAudioChannels);
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
AudioConfig::ChannelLayout layout(mAudioChannels);
if (!layout.IsValid()) {
return E_FAIL;
}
return S_OK;
}
HRESULT
WMFAudioMFTManager::Output(int64_t aStreamOffset,
RefPtr<MediaData>& aOutData)
{
aOutData = nullptr;
RefPtr<IMFSample> sample;
HRESULT hr;
int typeChangeCount = 0;
while (true) {
hr = mDecoder->Output(&sample);
if (hr == MF_E_TRANSFORM_NEED_MORE_INPUT) {
return hr;
}
if (hr == MF_E_TRANSFORM_STREAM_CHANGE) {
hr = UpdateOutputType();
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
// Catch infinite loops, but some decoders perform at least 2 stream
// changes on consecutive calls, so be permissive.
// 100 is arbitrarily > 2.
NS_ENSURE_TRUE(typeChangeCount < 100, MF_E_TRANSFORM_STREAM_CHANGE);
++typeChangeCount;
continue;
}
break;
}
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
if (!sample) {
LOG("Audio MFTDecoder returned success but null output.");
nsCOMPtr<nsIRunnable> task = NS_NewRunnableFunction([]() -> void {
LOG("Reporting telemetry AUDIO_MFT_OUTPUT_NULL_SAMPLES");
Telemetry::Accumulate(Telemetry::ID::AUDIO_MFT_OUTPUT_NULL_SAMPLES, 1);
});
AbstractThread::MainThread()->Dispatch(task.forget());
return E_FAIL;
}
RefPtr<IMFMediaBuffer> buffer;
hr = sample->ConvertToContiguousBuffer(getter_AddRefs(buffer));
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
BYTE* data = nullptr; // Note: *data will be owned by the IMFMediaBuffer, we don't need to free it.
DWORD maxLength = 0, currentLength = 0;
hr = buffer->Lock(&data, &maxLength, &currentLength);
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
// Sometimes when starting decoding, the AAC decoder gives us samples
// with a negative timestamp. AAC does usually have preroll (or encoder
// delay) encoded into its bitstream, but the amount encoded to the stream
// is variable, and it not signalled in-bitstream. There is sometimes
// signalling in the MP4 container what the preroll amount, but it's
// inconsistent. It looks like WMF's AAC encoder may take this into
// account, so strip off samples with a negative timestamp to get us
// to a 0-timestamp start. This seems to maintain A/V sync, so we can run
// with this until someone complains...
// We calculate the timestamp and the duration based on the number of audio
// frames we've already played. We don't trust the timestamp stored on the
// IMFSample, as sometimes it's wrong, possibly due to buggy encoders?
// If this sample block comes after a discontinuity (i.e. a gap or seek)
// reset the frame counters, and capture the timestamp. Future timestamps
// will be offset from this block's timestamp.
UINT32 discontinuity = false;
sample->GetUINT32(MFSampleExtension_Discontinuity, &discontinuity);
if (mMustRecaptureAudioPosition || discontinuity) {
// Update the output type, in case this segment has a different
// rate. This also triggers on the first sample, which can have a
// different rate than is advertised in the container, and sometimes we
// don't get a MF_E_TRANSFORM_STREAM_CHANGE when the rate changes.
hr = UpdateOutputType();
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
mAudioFrameSum = 0;
LONGLONG timestampHns = 0;
hr = sample->GetSampleTime(&timestampHns);
NS_ENSURE_TRUE(SUCCEEDED(hr), hr);
mAudioTimeOffset = media::TimeUnit::FromMicroseconds(timestampHns / 10);
mMustRecaptureAudioPosition = false;
}
// We can assume PCM 16 output.
int32_t numSamples = currentLength / 2;
int32_t numFrames = numSamples / mAudioChannels;
MOZ_ASSERT(numFrames >= 0);
MOZ_ASSERT(numSamples >= 0);
if (numFrames == 0) {
// All data from this chunk stripped, loop back and try to output the next
// frame, if possible.
return S_OK;
}
AlignedAudioBuffer audioData(numSamples);
if (!audioData) {
return E_OUTOFMEMORY;
}
int16_t* pcm = (int16_t*)data;
for (int32_t i = 0; i < numSamples; ++i) {
#ifdef MOZ_SAMPLE_TYPE_FLOAT32
audioData[i] = AudioSampleToFloat(pcm[i]);
#else
audioData[i] = pcm[i];
#endif
}
buffer->Unlock();
media::TimeUnit timestamp =
mAudioTimeOffset + FramesToTimeUnit(mAudioFrameSum, mAudioRate);
NS_ENSURE_TRUE(timestamp.IsValid(), E_FAIL);
mAudioFrameSum += numFrames;
media::TimeUnit duration = FramesToTimeUnit(numFrames, mAudioRate);
NS_ENSURE_TRUE(duration.IsValid(), E_FAIL);
aOutData = new AudioData(aStreamOffset,
timestamp.ToMicroseconds(),
duration.ToMicroseconds(),
numFrames,
Move(audioData),
mAudioChannels,
mAudioRate);
#ifdef LOG_SAMPLE_DECODE
LOG("Decoded audio sample! timestamp=%lld duration=%lld currentLength=%u",
timestamp.ToMicroseconds(), duration.ToMicroseconds(), currentLength);
#endif
return S_OK;
}
void
WMFAudioMFTManager::Shutdown()
{
mDecoder = nullptr;
}
} // namespace mozilla