Merge pull request #20 from liushuyu/dsp_aac
Media Foundation Improvement
This commit is contained in:
commit
c93df434c5
5 changed files with 93 additions and 92 deletions
|
@ -20,4 +20,4 @@ ADTSData ParseADTS(const char* buffer);
|
|||
|
||||
// last two bytes of MF AAC decoder user data
|
||||
// see https://docs.microsoft.com/en-us/windows/desktop/medfound/aac-decoder#example-media-types
|
||||
u16 MFGetAACTag(const ADTSData input);
|
||||
u16 MFGetAACTag(const ADTSData& input);
|
||||
|
|
|
@ -50,7 +50,7 @@ ADTSData ParseADTS(const char* buffer) {
|
|||
// Frame length flag (1 bit)
|
||||
// Depends on core coder (1 bit)
|
||||
// Extension flag (1 bit)
|
||||
u16 MFGetAACTag(const ADTSData input) {
|
||||
u16 MFGetAACTag(const ADTSData& input) {
|
||||
u16 tag = 0;
|
||||
|
||||
tag |= input.profile << 11;
|
||||
|
|
|
@ -16,14 +16,12 @@ public:
|
|||
private:
|
||||
std::optional<BinaryResponse> Initalize(const BinaryRequest& request);
|
||||
|
||||
void Clear();
|
||||
|
||||
std::optional<BinaryResponse> Decode(const BinaryRequest& request);
|
||||
|
||||
MFOutputState DecodingLoop(ADTSData adts_header, std::array<std::vector<u8>, 2>& out_streams);
|
||||
|
||||
bool initalized = false;
|
||||
bool selected = false;
|
||||
bool transform_initialized = false;
|
||||
bool format_selected = false;
|
||||
|
||||
Memory::MemorySystem& memory;
|
||||
|
||||
|
@ -33,10 +31,51 @@ private:
|
|||
};
|
||||
|
||||
WMFDecoder::Impl::Impl(Memory::MemorySystem& memory) : memory(memory) {
|
||||
MFCoInit();
|
||||
HRESULT hr = S_OK;
|
||||
hr = CoInitialize(NULL);
|
||||
// S_FALSE will be returned when COM has already been initialized
|
||||
if (hr != S_OK && hr != S_FALSE) {
|
||||
ReportError("Failed to start COM components", hr);
|
||||
}
|
||||
|
||||
WMFDecoder::Impl::~Impl() = default;
|
||||
// lite startup is faster and all what we need is included
|
||||
hr = MFStartup(MF_VERSION, MFSTARTUP_LITE);
|
||||
if (hr != S_OK) {
|
||||
// Do you know you can't initialize MF in test mode or safe mode?
|
||||
ReportError("Failed to initialize Media Foundation", hr);
|
||||
}
|
||||
|
||||
LOG_INFO(Audio_DSP, "Media Foundation activated");
|
||||
|
||||
// initialize transform
|
||||
transform = MFDecoderInit();
|
||||
if (transform == nullptr) {
|
||||
LOG_CRITICAL(Audio_DSP, "Can't initialize decoder");
|
||||
return;
|
||||
}
|
||||
|
||||
hr = transform->GetStreamIDs(1, &in_stream_id, 1, &out_stream_id);
|
||||
if (hr == E_NOTIMPL) {
|
||||
// if not implemented, it means this MFT does not assign stream ID for you
|
||||
in_stream_id = 0;
|
||||
out_stream_id = 0;
|
||||
} else if (FAILED(hr)) {
|
||||
ReportError("Decoder failed to initialize the stream ID", hr);
|
||||
return;
|
||||
}
|
||||
transform_initialized = true;
|
||||
}
|
||||
|
||||
WMFDecoder::Impl::~Impl() {
|
||||
if (transform_initialized) {
|
||||
MFFlush(transform.get());
|
||||
// delete the transform object before shutting down MF
|
||||
// otherwise access violation will occur
|
||||
transform.reset();
|
||||
}
|
||||
MFShutdown();
|
||||
CoUninitialize();
|
||||
}
|
||||
|
||||
std::optional<BinaryResponse> WMFDecoder::Impl::ProcessRequest(const BinaryRequest& request) {
|
||||
if (request.codec != DecoderCodec::AAC) {
|
||||
|
@ -65,43 +104,14 @@ std::optional<BinaryResponse> WMFDecoder::Impl::ProcessRequest(const BinaryReque
|
|||
}
|
||||
|
||||
std::optional<BinaryResponse> WMFDecoder::Impl::Initalize(const BinaryRequest& request) {
|
||||
if (initalized) {
|
||||
Clear();
|
||||
}
|
||||
|
||||
BinaryResponse response;
|
||||
std::memcpy(&response, &request, sizeof(response));
|
||||
response.unknown1 = 0x0;
|
||||
transform = MFDecoderInit();
|
||||
|
||||
if (transform == nullptr) {
|
||||
LOG_CRITICAL(Audio_DSP, "Can't init decoder");
|
||||
format_selected = false; // select format again if application request initialize the DSP
|
||||
return response;
|
||||
}
|
||||
|
||||
HRESULT hr = transform->GetStreamIDs(1, &in_stream_id, 1, &out_stream_id);
|
||||
if (hr == E_NOTIMPL) {
|
||||
// if not implemented, it means this MFT does not assign stream ID for you
|
||||
in_stream_id = 0;
|
||||
out_stream_id = 0;
|
||||
} else if (FAILED(hr)) {
|
||||
ReportError("Decoder failed to initialize the stream ID", hr);
|
||||
return response;
|
||||
}
|
||||
|
||||
initalized = true;
|
||||
return response;
|
||||
}
|
||||
|
||||
void WMFDecoder::Impl::Clear() {
|
||||
if (initalized) {
|
||||
MFFlush(transform.get());
|
||||
MFDeInit(transform.get());
|
||||
}
|
||||
initalized = false;
|
||||
selected = false;
|
||||
}
|
||||
|
||||
MFOutputState WMFDecoder::Impl::DecodingLoop(ADTSData adts_header,
|
||||
std::array<std::vector<u8>, 2>& out_streams) {
|
||||
MFOutputState output_status = MFOutputState::OK;
|
||||
|
@ -117,7 +127,7 @@ MFOutputState WMFDecoder::Impl::DecodingLoop(ADTSData adts_header,
|
|||
|
||||
// the following was taken from ffmpeg version of the decoder
|
||||
f32 val_f32;
|
||||
for (size_t i = 0; i < output_buffer->size();) {
|
||||
for (std::size_t i = 0; i < output_buffer->size();) {
|
||||
for (std::size_t channel = 0; channel < adts_header.channels; channel++) {
|
||||
val_f32 = output_buffer->at(i);
|
||||
s16 val = static_cast<s16>(0x7FFF * val_f32);
|
||||
|
@ -135,8 +145,8 @@ MFOutputState WMFDecoder::Impl::DecodingLoop(ADTSData adts_header,
|
|||
|
||||
// for status = 2, reset MF
|
||||
if (output_status == MFOutputState::NeedReconfig) {
|
||||
Clear();
|
||||
return MFOutputState::FatalError;
|
||||
format_selected = false;
|
||||
return MFOutputState::NeedReconfig;
|
||||
}
|
||||
|
||||
// for status = 3, try again with new buffer
|
||||
|
@ -161,8 +171,8 @@ std::optional<BinaryResponse> WMFDecoder::Impl::Decode(const BinaryRequest& requ
|
|||
response.num_channels = 2;
|
||||
response.num_samples = 1024;
|
||||
|
||||
if (!initalized) {
|
||||
LOG_DEBUG(Audio_DSP, "Decoder not initalized");
|
||||
if (!transform_initialized) {
|
||||
LOG_DEBUG(Audio_DSP, "Decoder not initialized");
|
||||
// This is a hack to continue games when decoder failed to initialize
|
||||
return response;
|
||||
}
|
||||
|
@ -177,6 +187,7 @@ std::optional<BinaryResponse> WMFDecoder::Impl::Decode(const BinaryRequest& requ
|
|||
std::array<std::vector<u8>, 2> out_streams;
|
||||
unique_mfptr<IMFSample> sample;
|
||||
MFInputState input_status = MFInputState::OK;
|
||||
MFOutputState output_status = MFOutputState::OK;
|
||||
std::optional<ADTSMeta> adts_meta = DetectMediaType((char*)data, request.size);
|
||||
|
||||
if (!adts_meta) {
|
||||
|
@ -186,7 +197,7 @@ std::optional<BinaryResponse> WMFDecoder::Impl::Decode(const BinaryRequest& requ
|
|||
|
||||
response.num_channels = adts_meta->ADTSHeader.channels;
|
||||
|
||||
if (!selected) {
|
||||
if (!format_selected) {
|
||||
LOG_DEBUG(Audio_DSP, "New ADTS stream: channels = {}, sample rate = {}",
|
||||
adts_meta->ADTSHeader.channels, adts_meta->ADTSHeader.samplerate);
|
||||
SelectInputMediaType(transform.get(), in_stream_id, adts_meta->ADTSHeader,
|
||||
|
@ -196,7 +207,7 @@ std::optional<BinaryResponse> WMFDecoder::Impl::Decode(const BinaryRequest& requ
|
|||
// cache the result from detect_mediatype and call select_*_mediatype only once
|
||||
// This could increase performance very slightly
|
||||
transform->ProcessMessage(MFT_MESSAGE_NOTIFY_BEGIN_STREAMING, 0);
|
||||
selected = true;
|
||||
format_selected = true;
|
||||
}
|
||||
|
||||
sample = CreateSample((void*)data, request.size, 1, 0);
|
||||
|
@ -204,8 +215,9 @@ std::optional<BinaryResponse> WMFDecoder::Impl::Decode(const BinaryRequest& requ
|
|||
|
||||
while (true) {
|
||||
input_status = SendSample(transform.get(), in_stream_id, sample.get());
|
||||
output_status = DecodingLoop(adts_meta->ADTSHeader, out_streams);
|
||||
|
||||
if (DecodingLoop(adts_meta->ADTSHeader, out_streams) == MFOutputState::FatalError) {
|
||||
if (output_status == MFOutputState::FatalError) {
|
||||
// if the decode issues are caused by MFT not accepting new samples, try again
|
||||
// NOTICE: you are required to check the output even if you already knew/guessed
|
||||
// MFT didn't accept the input sample
|
||||
|
@ -216,6 +228,11 @@ std::optional<BinaryResponse> WMFDecoder::Impl::Decode(const BinaryRequest& requ
|
|||
|
||||
LOG_ERROR(Audio_DSP, "Errors occurred when receiving output");
|
||||
return response;
|
||||
} else if (output_status == MFOutputState::NeedReconfig) {
|
||||
// flush the transform
|
||||
MFFlush(transform.get());
|
||||
// decode again
|
||||
return this->Decode(request);
|
||||
}
|
||||
|
||||
break; // jump out of the loop if at least we don't have obvious issues
|
||||
|
|
|
@ -2,6 +2,7 @@
|
|||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
#include "common/logging/log.h"
|
||||
#include "common/string_util.h"
|
||||
#include "wmf_decoder_utils.h"
|
||||
|
||||
// utility functions
|
||||
|
@ -9,42 +10,21 @@ void ReportError(std::string msg, HRESULT hr) {
|
|||
if (SUCCEEDED(hr)) {
|
||||
return;
|
||||
}
|
||||
LPSTR err;
|
||||
FormatMessage(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_ALLOCATE_BUFFER |
|
||||
LPWSTR err;
|
||||
FormatMessageW(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_ALLOCATE_BUFFER |
|
||||
FORMAT_MESSAGE_IGNORE_INSERTS,
|
||||
nullptr, hr,
|
||||
// hardcode to use en_US because if any user had problems with this
|
||||
// we can help them w/o translating anything
|
||||
// default is to use the language currently active on the operating system
|
||||
MAKELANGID(LANG_ENGLISH, SUBLANG_ENGLISH_US), (LPSTR)&err, 0, nullptr);
|
||||
MAKELANGID(LANG_ENGLISH, SUBLANG_ENGLISH_US), (LPWSTR)&err, 0, nullptr);
|
||||
if (err != nullptr) {
|
||||
LOG_CRITICAL(Audio_DSP, "{}: {}", msg, err);
|
||||
LOG_CRITICAL(Audio_DSP, "{}: {}", msg, Common::UTF16ToUTF8(err));
|
||||
LocalFree(err);
|
||||
}
|
||||
LOG_CRITICAL(Audio_DSP, "{}: {:08x}", msg, hr);
|
||||
}
|
||||
|
||||
bool MFCoInit() {
|
||||
HRESULT hr = S_OK;
|
||||
hr = CoInitialize(NULL);
|
||||
// S_FALSE will be returned when COM has already been initialized
|
||||
if (hr != S_OK && hr != S_FALSE) {
|
||||
ReportError("Failed to start COM components", hr);
|
||||
return false;
|
||||
}
|
||||
|
||||
// lite startup is faster and all what we need is included
|
||||
hr = MFStartup(MF_VERSION, MFSTARTUP_LITE);
|
||||
if (hr != S_OK) {
|
||||
// Do you know you can't initialize MF in test mode or safe mode?
|
||||
ReportError("Failed to initialize Media Foundation", hr);
|
||||
return false;
|
||||
}
|
||||
|
||||
LOG_INFO(Audio_DSP, "Media Foundation activated");
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
unique_mfptr<IMFTransform> MFDecoderInit(GUID audio_format) {
|
||||
HRESULT hr = S_OK;
|
||||
MFT_REGISTER_TYPE_INFO reg = {0};
|
||||
|
@ -72,6 +52,8 @@ unique_mfptr<IMFTransform> MFDecoderInit(GUID audio_format) {
|
|||
if (FAILED(hr))
|
||||
transform = nullptr;
|
||||
activate[n]->Release();
|
||||
if (SUCCEEDED(hr))
|
||||
break;
|
||||
}
|
||||
if (transform == nullptr) {
|
||||
ReportError("Failed to initialize MFT", hr);
|
||||
|
@ -79,15 +61,11 @@ unique_mfptr<IMFTransform> MFDecoderInit(GUID audio_format) {
|
|||
return nullptr;
|
||||
}
|
||||
CoTaskMemFree(activate);
|
||||
return std::move(transform);
|
||||
return transform;
|
||||
}
|
||||
|
||||
void MFDeInit(IMFTransform* transform) {
|
||||
MFShutdownObject(transform);
|
||||
CoUninitialize();
|
||||
}
|
||||
|
||||
unique_mfptr<IMFSample> CreateSample(void* data, DWORD len, DWORD alignment, LONGLONG duration) {
|
||||
unique_mfptr<IMFSample> CreateSample(const void* data, DWORD len, DWORD alignment,
|
||||
LONGLONG duration) {
|
||||
HRESULT hr = S_OK;
|
||||
unique_mfptr<IMFMediaBuffer> buf;
|
||||
unique_mfptr<IMFSample> sample;
|
||||
|
@ -126,11 +104,11 @@ unique_mfptr<IMFSample> CreateSample(void* data, DWORD len, DWORD alignment, LON
|
|||
ReportError("Unable to set sample duration, but continuing anyway", hr);
|
||||
}
|
||||
|
||||
return std::move(sample);
|
||||
return sample;
|
||||
}
|
||||
|
||||
bool SelectInputMediaType(IMFTransform* transform, int in_stream_id, const ADTSData& adts,
|
||||
UINT8* user_data, UINT32 user_data_len, GUID audio_format) {
|
||||
const UINT8* user_data, UINT32 user_data_len, GUID audio_format) {
|
||||
HRESULT hr = S_OK;
|
||||
unique_mfptr<IMFMediaType> t;
|
||||
|
||||
|
@ -209,7 +187,7 @@ bool SelectOutputMediaType(IMFTransform* transform, int out_stream_id, GUID audi
|
|||
return false;
|
||||
}
|
||||
|
||||
std::optional<ADTSMeta> DetectMediaType(char* buffer, size_t len) {
|
||||
std::optional<ADTSMeta> DetectMediaType(char* buffer, std::size_t len) {
|
||||
if (len < 7) {
|
||||
return std::nullopt;
|
||||
}
|
||||
|
|
|
@ -29,6 +29,14 @@ struct MFRelease {
|
|||
};
|
||||
};
|
||||
|
||||
template <>
|
||||
struct MFRelease<IMFTransform> {
|
||||
void operator()(IMFTransform* pointer) const {
|
||||
MFShutdownObject(pointer);
|
||||
pointer->Release();
|
||||
};
|
||||
};
|
||||
|
||||
// wrapper facilities for dealing with pointers
|
||||
template <typename T>
|
||||
using unique_mfptr = std::unique_ptr<T, MFRelease<T>>;
|
||||
|
@ -65,15 +73,13 @@ struct ADTSMeta {
|
|||
};
|
||||
|
||||
// exported functions
|
||||
bool MFCoInit();
|
||||
unique_mfptr<IMFTransform> MFDecoderInit(GUID audio_format = MFAudioFormat_AAC);
|
||||
void MFDeInit(IMFTransform* transform);
|
||||
unique_mfptr<IMFSample> CreateSample(void* data, DWORD len, DWORD alignment = 1,
|
||||
unique_mfptr<IMFSample> CreateSample(const void* data, DWORD len, DWORD alignment = 1,
|
||||
LONGLONG duration = 0);
|
||||
bool SelectInputMediaType(IMFTransform* transform, int in_stream_id, const ADTSData& adts,
|
||||
UINT8* user_data, UINT32 user_data_len,
|
||||
const UINT8* user_data, UINT32 user_data_len,
|
||||
GUID audio_format = MFAudioFormat_AAC);
|
||||
std::optional<ADTSMeta> DetectMediaType(char* buffer, size_t len);
|
||||
std::optional<ADTSMeta> DetectMediaType(char* buffer, std::size_t len);
|
||||
bool SelectOutputMediaType(IMFTransform* transform, int out_stream_id,
|
||||
GUID audio_format = MFAudioFormat_PCM);
|
||||
void MFFlush(IMFTransform* transform);
|
||||
|
|
Loading…
Reference in a new issue