ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

[APU] Added new XMA decoder as config option.

This commit is contained in:
Gliniak 2024-03-25 21:59:49 +01:00 committed by Radosław Gliński
parent 26ea81624a
commit 122f58c9dd
9 changed files with 1976 additions and 996 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

912
src/xenia/apu/xma_context.cc
View File

@ -13,7 +13,6 @@
#include <cstring>
#include "xenia/apu/xma_decoder.h"
#include "xenia/apu/xma_helpers.h"
#include "xenia/base/bit_stream.h"
#include "xenia/base/logging.h"
#include "xenia/base/platform.h"
@ -39,259 +38,11 @@ namespace apu {
XmaContext::XmaContext() = default;
XmaContext::~XmaContext() {
if (av_context_) {
if (avcodec_is_open(av_context_)) {
avcodec_close(av_context_);
}
av_free(av_context_);
}
if (av_frame_) {
av_frame_free(&av_frame_);
}
// if (current_frame_) {
// delete[] current_frame_;
// }
}
XmaContext::~XmaContext() {}
int XmaContext::Setup(uint32_t id, Memory* memory, uint32_t guest_ptr) {
id_ = id;
memory_ = memory;
guest_ptr_ = guest_ptr;
// Allocate ffmpeg stuff:
av_packet_ = av_packet_alloc();
assert_not_null(av_packet_);
// chrispy: preallocate this buffer so that ffmpeg isn't reallocating it for
// every packet, these allocations were causing RtlSubsegmentInitialize
av_packet_->buf = av_buffer_alloc(128 * 1024);
// find the XMA2 audio decoder
av_codec_ = avcodec_find_decoder(AV_CODEC_ID_XMAFRAMES);
if (!av_codec_) {
XELOGE("XmaContext {}: Codec not found", id);
return 1;
}
av_context_ = avcodec_alloc_context3(av_codec_);
if (!av_context_) {
XELOGE("XmaContext {}: Couldn't allocate context", id);
return 1;
}
// Initialize these to 0. They'll actually be set later.
av_context_->channels = 0;
av_context_->sample_rate = 0;
av_frame_ = av_frame_alloc();
if (!av_frame_) {
XELOGE("XmaContext {}: Couldn't allocate frame", id);
return 1;
}
// FYI: We're purposely not opening the codec here. That is done later.
return 0;
}
bool XmaContext::Work() {
if (!is_enabled() || !is_allocated()) {
return false;
}
{
std::lock_guard<xe_mutex> lock(lock_);
set_is_enabled(false);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
Decode(&data);
data.Store(context_ptr);
return true;
}
}
void XmaContext::Enable() {
std::lock_guard<xe_mutex> lock(lock_);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
XELOGAPU("XmaContext: kicking context {} (buffer {} {}/{} bits)", id(),
data.current_buffer, data.input_buffer_read_offset,
(data.current_buffer == 0 ? data.input_buffer_0_packet_count
: data.input_buffer_1_packet_count) *
kBitsPerPacket);
data.Store(context_ptr);
set_is_enabled(true);
}
bool XmaContext::Block(bool poll) {
if (!lock_.try_lock()) {
if (poll) {
return false;
}
lock_.lock();
}
lock_.unlock();
return true;
}
void XmaContext::Clear() {
std::lock_guard<xe_mutex> lock(lock_);
XELOGAPU("XmaContext: reset context {}", id());
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
data.input_buffer_0_valid = 0;
data.input_buffer_1_valid = 0;
data.output_buffer_valid = 0;
data.input_buffer_read_offset = 0;
data.output_buffer_read_offset = 0;
data.output_buffer_write_offset = 0;
xma_frame_.fill(0);
split_frame_len_ = 0;
split_frame_len_partial_ = 0;
split_frame_padding_start_ = 0;
data.Store(context_ptr);
}
void XmaContext::Disable() {
std::lock_guard<xe_mutex> lock(lock_);
XELOGAPU("XmaContext: disabling context {}", id());
set_is_enabled(false);
}
void XmaContext::Release() {
// Lock it in case the decoder thread is working on it now.
std::lock_guard<xe_mutex> lock(lock_);
assert_true(is_allocated_ == true);
set_is_allocated(false);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
std::memset(context_ptr, 0, sizeof(XMA_CONTEXT_DATA)); // Zero it.
}
void XmaContext::SwapInputBuffer(XMA_CONTEXT_DATA* data) {
// No more frames.
if (data->current_buffer == 0) {
data->input_buffer_0_valid = 0;
} else {
data->input_buffer_1_valid = 0;
}
data->current_buffer ^= 1;
data->input_buffer_read_offset = kBitsPerHeader;
}
bool XmaContext::TrySetupNextLoop(XMA_CONTEXT_DATA* data,
bool ignore_input_buffer_offset) {
// Setup the input buffer offset if next loop exists.
// TODO(Pseudo-Kernel): Need to handle loop in the following cases.
// 1. loop_start == loop_end == 0
// 2. loop_start > loop_end && loop_count > 0
if (data->loop_count > 0 && data->loop_start < data->loop_end &&
(ignore_input_buffer_offset ||
data->input_buffer_read_offset >= data->loop_end)) {
// Loop back to the beginning.
data->input_buffer_read_offset = data->loop_start;
if (data->loop_count < 255) {
data->loop_count--;
}
return true;
}
return false;
}
/*
void XmaContext::NextPacket(
uint8_t* input_buffer,
uint32_t input_size,
uint32_t input_buffer_read_offset) {
*/
void XmaContext::NextPacket(XMA_CONTEXT_DATA* data) {
// auto packet_idx = GetFramePacketNumber(input_buffer, input_size,
// input_buffer_read_offset);
// packet_idx++;
// if (packet_idx++ >= input_size)
}
int XmaContext::GetSampleRate(int id) {
switch (id) {
case 0:
return 24000;
case 1:
return 32000;
case 2:
return 44100;
case 3:
return 48000;
}
assert_always();
return 0;
}
bool XmaContext::ValidFrameOffset(uint8_t* block, size_t size_bytes,
size_t frame_offset_bits) {
uint32_t packet_num =
GetFramePacketNumber(block, size_bytes, frame_offset_bits);
if (packet_num == -1) {
// Invalid packet number
XELOGAPU("ValidFrameOffset: Invalid packet number");
return false;
}
uint8_t* packet = block + (packet_num * kBytesPerPacket);
size_t relative_offset_bits = frame_offset_bits % kBitsPerPacket;
uint32_t first_frame_offset = xma::GetPacketFrameOffset(packet);
if (first_frame_offset == -1 || first_frame_offset > kBitsPerPacket) {
XELOGAPU("ValidFrameOffset: Invalid frame offset {}", first_frame_offset);
// Packet only contains a partial frame, so no frames can start here.
return false;
}
BitStream stream(packet, kBitsPerPacket);
stream.SetOffset(first_frame_offset);
while (true) {
if (stream.offset_bits() == relative_offset_bits) {
return true;
}
if (stream.BitsRemaining() < 15) {
XELOGAPU("ValidFrameOffset: No room for next frame header {}",
first_frame_offset);
// Not enough room for another frame header.
return false;
}
uint64_t size = stream.Read(15);
if ((size - 15) > stream.BitsRemaining()) {
XELOGAPU("ValidFrameOffset: Last frame {} - {}", first_frame_offset,
size);
// Last frame.
return false;
} else if (size == 0x7FFF) {
// Invalid frame (and last of this packet)
return false;
}
stream.Advance(size - 16);
// Read the trailing bit to see if frames follow
if (stream.Read(1) == 0) {
break;
}
}
return false;
}
static void dump_raw(AVFrame* frame, int id) {
FILE* outfile = fopen(fmt::format("out{}.raw", id).c_str(), "ab");
void XmaContext::DumpRaw(AVFrame* frame, int id) {
FILE* outfile =
xe::filesystem::OpenFile(fmt::format("out{}.raw", id).c_str(), "ab");
if (!outfile) {
return;
}
@ -304,661 +55,6 @@ static void dump_raw(AVFrame* frame, int id) {
fclose(outfile);
}
void XmaContext::Decode(XMA_CONTEXT_DATA* data) {
SCOPE_profile_cpu_f("apu");
// What I see:
// XMA outputs 2 bytes per sample
// 512 samples per frame (128 per subframe)
// Max output size is data.output_buffer_block_count * 256
// This decoder is fed packets (max 4095 per buffer)
// Packets contain "some" frames
// 32bit header (big endian)
// Frames are the smallest thing the SPUs can decode.
// They can and usually will span packets.
// Sample rates (data.sample_rate):
// 0 - 24 kHz
// 1 - 32 kHz
// 2 - 44.1 kHz
// 3 - 48 kHz
// SPUs also support stereo decoding. (data.is_stereo)
// Check the output buffer - we cannot decode anything else if it's
// unavailable.
if (!data->output_buffer_valid) {
return;
}
// No available data.
if (!data->input_buffer_0_valid && !data->input_buffer_1_valid) {
return;
}
// XAudio Loops
// loop_count:
// - XAUDIO2_MAX_LOOP_COUNT = 254
// - XAUDIO2_LOOP_INFINITE = 255
// loop_start/loop_end are bit offsets to a specific frame
// Translate pointers for future use.
// Sometimes the game will use rolling input buffers. If they do, we cannot
// assume they form a complete block! In addition, the buffers DO NOT have
// to be contiguous!
uint8_t* in0 = data->input_buffer_0_valid
? memory()->TranslatePhysical(data->input_buffer_0_ptr)
: nullptr;
uint8_t* in1 = data->input_buffer_1_valid
? memory()->TranslatePhysical(data->input_buffer_1_ptr)
: nullptr;
uint8_t* current_input_buffer = data->current_buffer ? in1 : in0;
if (!current_input_buffer) {
XELOGE("XmaContext {}: Error - input buffer pointer is invalid!", id());
return;
}
if (!data->output_buffer_block_count) {
XELOGE("XmaContext {}: Error - Received 0 for output_buffer_block_count!",
id());
return;
}
XELOGAPU(
"Processing context {} (offset {}, buffer {}, ptr {:p}, output buffer "
"{:08X}, output buffer count {})",
id(), data->input_buffer_read_offset, data->current_buffer,
current_input_buffer, data->output_buffer_ptr,
data->output_buffer_block_count);
if (is_stream_done_) {
is_stream_done_ = false;
packets_skip_ = 0;
SwapInputBuffer(data);
return;
}
size_t input_buffer_0_size =
data->input_buffer_0_packet_count * kBytesPerPacket;
size_t input_buffer_1_size =
data->input_buffer_1_packet_count * kBytesPerPacket;
size_t current_input_size =
data->current_buffer ? input_buffer_1_size : input_buffer_0_size;
size_t current_input_packet_count = current_input_size / kBytesPerPacket;
bool is_streaming = data->input_buffer_0_packet_count == 1 &&
data->input_buffer_1_packet_count == 1;
// Output buffers are in raw PCM samples, 256 bytes per block.
// Output buffer is a ring buffer. We need to write from the write offset
// to the read offset.
uint8_t* output_buffer = memory()->TranslatePhysical(data->output_buffer_ptr);
uint32_t output_capacity =
data->output_buffer_block_count * kBytesPerSubframeChannel;
uint32_t output_read_offset =
data->output_buffer_read_offset * kBytesPerSubframeChannel;
uint32_t output_write_offset =
data->output_buffer_write_offset * kBytesPerSubframeChannel;
RingBuffer output_rb(output_buffer, output_capacity);
output_rb.set_read_offset(output_read_offset);
output_rb.set_write_offset(output_write_offset);
// We can only decode an entire frame and write it out at a time, so
// don't save any samples.
// TODO(JoelLinn): subframes when looping
size_t output_remaining_bytes = output_rb.write_count();
output_remaining_bytes -=
output_remaining_bytes % (kBytesPerFrameChannel << data->is_stereo);
// is_dirty_ = true; // TODO
// is_dirty_ = false; // TODO
assert_false(data->stop_when_done);
assert_false(data->interrupt_when_done);
static int total_samples = 0;
// Decode until we can't write any more data.
while (output_remaining_bytes > 0) {
if (!data->input_buffer_0_valid && !data->input_buffer_1_valid) {
// Out of data.
break;
}
// Setup the input buffer if we are at loop_end.
// The input buffer must not be swapped out until all loops are processed.
bool reuse_input_buffer = TrySetupNextLoop(data, false);
// assert_true(packets_skip_ == 0);
// assert_true(split_frame_len_ == 0);
// assert_true(split_frame_len_partial_ == 0);
// Where are we in the buffer (in XMA jargon)
int packet_idx, frame_idx, frame_count;
uint8_t* packet;
bool frame_last_split;
BitStream stream(current_input_buffer, current_input_size * 8);
stream.SetOffset(data->input_buffer_read_offset);
if (data->input_buffer_read_offset > current_input_size * 8) {
XELOGE(
"XmaContext {}: Error - Provided input offset exceed input buffer "
"size! ({} > {})",
id(), data->input_buffer_read_offset, current_input_size * 8);
SwapInputBuffer(data);
return;
}
// if we had a buffer swap try to skip packets first
if (packets_skip_ > 0) {
packet_idx =
GetFramePacketNumber(current_input_buffer, current_input_size,
data->input_buffer_read_offset);
while (packets_skip_ > 0) {
packets_skip_--;
packet_idx++;
if (packet_idx > current_input_packet_count) {
if (!reuse_input_buffer) {
// Last packet. Try setup once more.
reuse_input_buffer = TrySetupNextLoop(data, true);
}
if (!reuse_input_buffer) {
if (is_streaming) {
SwapInputBuffer(data);
} else {
is_stream_done_ = true;
}
}
return;
}
}
// invalid frame pointer but needed for us
data->input_buffer_read_offset = packet_idx * kBitsPerPacket;
// continue;
}
if (split_frame_len_) {
// handle a frame that was split over two packages
packet_idx =
GetFramePacketNumber(current_input_buffer, current_input_size,
data->input_buffer_read_offset);
packet = current_input_buffer + packet_idx * kBytesPerPacket;
std::tie(frame_count, frame_last_split) = GetPacketFrameCount(packet);
frame_idx = -1;
stream =
BitStream(current_input_buffer, (packet_idx + 1) * kBitsPerPacket);
stream.SetOffset(packet_idx * kBitsPerPacket + kBitsPerHeader);
if (split_frame_len_ > xma::kMaxFrameLength) {
// TODO write CopyPeekMethod
auto offset = stream.offset_bits();
stream.Copy(
xma_frame_.data() + 1 +
((split_frame_len_partial_ + split_frame_padding_start_) / 8),
15 - split_frame_len_partial_);
stream.SetOffset(offset);
BitStream slen(xma_frame_.data() + 1, 15 + split_frame_padding_start_);
slen.Advance(split_frame_padding_start_);
split_frame_len_ = static_cast<int>(slen.Read(15));
}
if (frame_count > 0) {
// assert_true(xma::GetPacketFrameOffset(packet) - 32 ==
// split_frame_len_ - split_frame_len_partial_);
}
auto offset = stream.Copy(
xma_frame_.data() + 1 +
((split_frame_len_partial_ + split_frame_padding_start_) / 8),
split_frame_len_ - split_frame_len_partial_);
assert_true(offset ==
(split_frame_padding_start_ + split_frame_len_partial_) % 8);
} else {
if (data->input_buffer_read_offset % kBitsPerPacket == 0) {
// Invalid offset. Go ahead and set it.
int packet_number =
GetFramePacketNumber(current_input_buffer, current_input_size,
data->input_buffer_read_offset);
if (packet_number == -1) {
return;
}
auto offset =
xma::GetPacketFrameOffset(current_input_buffer +
kBytesPerPacket * packet_number) +
data->input_buffer_read_offset;
if (offset == -1) {
// No more frames.
SwapInputBuffer(data);
// TODO partial frames? end?
XELOGE("XmaContext {}: TODO partial frames? end?", id());
assert_always("TODO");
return;
} else {
data->input_buffer_read_offset = offset;
}
}
if (!ValidFrameOffset(current_input_buffer, current_input_size,
data->input_buffer_read_offset)) {
XELOGAPU("XmaContext {}: Error - Invalid read offset {}!", id(),
data->input_buffer_read_offset);
SwapInputBuffer(data);
return;
}
// Where are we in the buffer (in XMA jargon)
std::tie(packet_idx, frame_idx) =
GetFrameNumber(current_input_buffer, current_input_size,
data->input_buffer_read_offset);
// TODO handle
assert_true(packet_idx >= 0);
assert_true(frame_idx >= 0);
packet = current_input_buffer + packet_idx * kBytesPerPacket;
// frames that belong to this packet
std::tie(frame_count, frame_last_split) = GetPacketFrameCount(packet);
assert_true(frame_count >= 0); // TODO end
PrepareDecoder(packet, data->sample_rate, bool(data->is_stereo));
// Current frame is split to next packet:
bool frame_is_split = frame_last_split && (frame_idx >= frame_count - 1);
stream =
BitStream(current_input_buffer, (packet_idx + 1) * kBitsPerPacket);
stream.SetOffset(data->input_buffer_read_offset);
// int frame_len;
// int frame_len_partial
split_frame_len_partial_ = static_cast<int>(stream.BitsRemaining());
if (split_frame_len_partial_ >= 15) {
split_frame_len_ = static_cast<int>(stream.Peek(15));
} else {
// assert_always();
split_frame_len_ = xma::kMaxFrameLength + 1;
}
assert_true(frame_is_split ==
(split_frame_len_ > split_frame_len_partial_));
// TODO fix bitstream copy
std::memset(xma_frame_.data(), 0, xma_frame_.size());
{
int32_t bits_to_copy =
std::min(split_frame_len_, split_frame_len_partial_);
if (!stream.IsOffsetValid(bits_to_copy)) {
XELOGAPU(
"XmaContext {}: Error - Invalid amount of bits to copy! "
"split_frame_len: {}, split_partial: {}, offset_bits: {}",
id(), split_frame_len_, split_frame_len_partial_,
stream.offset_bits());
SwapInputBuffer(data);
return;
}
auto offset = stream.Copy(xma_frame_.data() + 1, bits_to_copy);
assert_true(offset < 8);
split_frame_padding_start_ = static_cast<uint8_t>(offset);
}
if (frame_is_split) {
// go to next xma packet of this stream
packets_skip_ = xma::GetPacketSkipCount(packet) + 1;
while (packets_skip_ > 0) {
packets_skip_--;
packet += kBytesPerPacket;
packet_idx++;
if (packet_idx >= current_input_packet_count) {
if (!reuse_input_buffer) {
// Last packet. Try setup once more.
reuse_input_buffer = TrySetupNextLoop(data, true);
}
if (!reuse_input_buffer) {
if (is_streaming) {
SwapInputBuffer(data);
} else {
is_stream_done_ = true;
}
}
return;
}
}
// TODO guest might read this:
data->input_buffer_read_offset = packet_idx * kBitsPerPacket;
continue;
}
}
av_packet_->data = xma_frame_.data();
av_packet_->size = static_cast<int>(
1 + ((split_frame_padding_start_ + split_frame_len_) / 8) +
(((split_frame_padding_start_ + split_frame_len_) % 8) ? 1 : 0));
auto padding_end = av_packet_->size * 8 -
(8 + split_frame_padding_start_ + split_frame_len_);
assert_true(padding_end < 8);
xma_frame_[0] =
((split_frame_padding_start_ & 7) << 5) | ((padding_end & 7) << 2);
split_frame_len_ = 0;
split_frame_len_partial_ = 0;
split_frame_padding_start_ = 0;
auto ret = avcodec_send_packet(av_context_, av_packet_);
if (ret < 0) {
XELOGE("XmaContext {}: Error - Sending packet for decoding failed", id());
// TODO bail out
assert_always();
}
ret = avcodec_receive_frame(av_context_, av_frame_);
/*
if (ret == AVERROR(EAGAIN) || ret == AVERROR_EOF)
// TODO AVERROR_EOF???
break;
else
*/
if (ret < 0) {
XELOGE("XmaContext {}: Error - Decoding failed", id());
data->parser_error_status = 4; // TODO(Gliniak): Find all parsing errors
// and create enumerator from them
SwapInputBuffer(data);
assert_always();
return; // TODO bail out
}
assert_true(ret == 0);
{
// copy over 1 frame
// update input buffer read offset
// assert(decoded_consumed_samples_ + kSamplesPerFrame <=
// current_frame_.size());
assert_true(av_context_->sample_fmt == AV_SAMPLE_FMT_FLTP);
// assert_true(frame_is_split == (frame_idx == -1));
// dump_raw(av_frame_, id());
ConvertFrame((const uint8_t**)av_frame_->data,
bool(av_frame_->channels > 1), raw_frame_.data());
// decoded_consumed_samples_ += kSamplesPerFrame;
auto byte_count = kBytesPerFrameChannel << data->is_stereo;
assert_true(output_remaining_bytes >= byte_count);
output_rb.Write(raw_frame_.data(), byte_count);
output_remaining_bytes -= byte_count;
data->output_buffer_write_offset = output_rb.write_offset() / 256;
total_samples += id_ == 0 ? kSamplesPerFrame : 0;
uint32_t offset =
std::max(kBitsPerHeader, data->input_buffer_read_offset);
offset = static_cast<uint32_t>(
GetNextFrame(current_input_buffer, current_input_size, offset));
XELOGAPU(
"XmaContext {}: Next Offset: {} (Frame: {}/{} Packet: {}/{} Packet "
"Skip: {} - {})",
id(), offset, frame_idx, frame_count - 1, packet_idx,
current_input_packet_count, xma::GetPacketSkipCount(packet),
data->input_buffer_read_offset);
if (frame_idx + 1 >= frame_count) {
// Skip to next packet (no split frame)
packets_skip_ = xma::GetPacketSkipCount(packet) + 1;
while (packets_skip_ > 0) {
packets_skip_--;
packet_idx++;
if (packet_idx >= current_input_packet_count) {
if (!reuse_input_buffer) {
// Last packet. Try setup once more.
reuse_input_buffer = TrySetupNextLoop(data, true);
}
if (!reuse_input_buffer) {
if (is_streaming) {
SwapInputBuffer(data);
data->input_buffer_read_offset =
GetPacketFirstFrameOffset(data);
} else {
is_stream_done_ = true;
}
if (output_rb.write_offset() == output_rb.read_offset()) {
data->output_buffer_valid = 0;
}
}
return;
}
}
packet = current_input_buffer + packet_idx * kBytesPerPacket;
// TODO(Gliniak): There might be an edge-case when we're in packet 26/27
// and GetPacketFrameOffset returns that there is no data in this packet
// aka. FrameOffset is set to more than 0x7FFF-0x20
offset =
xma::GetPacketFrameOffset(packet) + packet_idx * kBitsPerPacket;
}
if (offset == 0 || frame_idx == -1) {
// Next packet but we already skipped to it
if (packet_idx >= current_input_packet_count) {
// Buffer is fully used
if (!reuse_input_buffer) {
// Last packet. Try setup once more.
reuse_input_buffer = TrySetupNextLoop(data, true);
}
if (!reuse_input_buffer) {
if (is_streaming) {
SwapInputBuffer(data);
} else {
is_stream_done_ = true;
}
}
break;
}
offset =
xma::GetPacketFrameOffset(packet) + packet_idx * kBitsPerPacket;
}
// TODO buffer bounds check
assert_true(data->input_buffer_read_offset < offset);
data->input_buffer_read_offset = offset;
}
}
// assert_true((split_frame_len_ != 0) == (data->input_buffer_read_offset ==
// 0));
// The game will kick us again with a new output buffer later.
// It's important that we only invalidate this if we actually wrote to it!!
if (output_rb.write_offset() == output_rb.read_offset()) {
data->output_buffer_valid = 0;
}
}
uint32_t XmaContext::GetPacketFirstFrameOffset(const XMA_CONTEXT_DATA* data) {
uint32_t first_frame_offset = kBitsPerHeader;
uint8_t* in0 = data->input_buffer_0_valid
? memory()->TranslatePhysical(data->input_buffer_0_ptr)
: nullptr;
uint8_t* in1 = data->input_buffer_1_valid
? memory()->TranslatePhysical(data->input_buffer_1_ptr)
: nullptr;
uint8_t* current_input_buffer = data->current_buffer ? in1 : in0;
if (current_input_buffer) {
first_frame_offset = xma::GetPacketFrameOffset(current_input_buffer);
}
return first_frame_offset;
}
size_t XmaContext::GetNextFrame(uint8_t* block, size_t size,
size_t bit_offset) {
// offset = xma::GetPacketFrameOffset(packet);
// TODO meh
// auto next_packet = bit_offset - bit_offset % kBitsPerPacket +
// kBitsPerPacket;
auto packet_idx = GetFramePacketNumber(block, size, bit_offset);
BitStream stream(block, size * 8);
stream.SetOffset(bit_offset);
if (stream.BitsRemaining() < 15) {
return 0;
}
uint64_t len = stream.Read(15);
if ((len - 15) > stream.BitsRemaining()) {
// assert_always("TODO");
// *bit_offset = next_packet;
// return false;
// return next_packet;
return 0;
} else if (len >= xma::kMaxFrameLength) {
assert_always("TODO");
// *bit_offset = next_packet;
// return false;
return 0;
// return next_packet;
}
stream.Advance(len - (15 + 1));
// Read the trailing bit to see if frames follow
if (stream.Read(1) == 0) {
return 0;
}
bit_offset += len;
if (packet_idx < GetFramePacketNumber(block, size, bit_offset)) {
return 0;
}
return bit_offset;
}
int XmaContext::GetFramePacketNumber(uint8_t* block, size_t size,
size_t bit_offset) {
size *= 8;
if (bit_offset >= size) {
// Not good :(
assert_always();
return -1;
}
size_t byte_offset = bit_offset >> 3;
size_t packet_number = byte_offset / kBytesPerPacket;
return (uint32_t)packet_number;
}
std::tuple<int, int> XmaContext::GetFrameNumber(uint8_t* block, size_t size,
size_t bit_offset) {
auto packet_idx = GetFramePacketNumber(block, size, bit_offset);
if (packet_idx < 0 || (packet_idx + 1) * kBytesPerPacket > size) {
assert_always();
return {packet_idx, -2};
}
if (bit_offset == 0) {
return {packet_idx, -1};
}
uint8_t* packet = block + (packet_idx * kBytesPerPacket);
auto first_frame_offset = xma::GetPacketFrameOffset(packet);
BitStream stream(block, size * 8);
stream.SetOffset(packet_idx * kBitsPerPacket + first_frame_offset);
int frame_idx = 0;
while (true) {
if (stream.BitsRemaining() < 15) {
break;
}
if (stream.offset_bits() == bit_offset) {
break;
}
uint64_t size = stream.Read(15);
if ((size - 15) > stream.BitsRemaining()) {
// Last frame.
break;
} else if (size == 0x7FFF) {
// Invalid frame (and last of this packet)
break;
}
stream.Advance(size - (15 + 1));
// Read the trailing bit to see if frames follow
if (stream.Read(1) == 0) {
break;
}
frame_idx++;
}
return {packet_idx, frame_idx};
}
std::tuple<int, bool> XmaContext::GetPacketFrameCount(uint8_t* packet) {
auto first_frame_offset = xma::GetPacketFrameOffset(packet);
if (first_frame_offset > kBitsPerPacket - kBitsPerHeader) {
// frame offset is beyond packet end
return {0, false};
}
BitStream stream(packet, kBitsPerPacket);
stream.SetOffset(first_frame_offset);
int frame_count = 0;
while (true) {
if (stream.BitsRemaining() < 15) {
return {frame_count, false};
}
frame_count++;
uint64_t size = stream.Read(15);
if ((size - 15) > stream.BitsRemaining()) {
return {frame_count, true};
} else if (size == 0x7FFF) {
assert_always();
return {frame_count, true};
}
stream.Advance(size - (15 + 1));
if (stream.Read(1) == 0) {
return {frame_count, false};
}
// There is a case when frame ends EXACTLY at the end of packet.
// In such case we shouldn't increase frame count by additional not existing
// frame and don't mark it as splitted, but as a normal frame
if (!stream.BitsRemaining()) {
return {frame_count, false};
}
}
}
int XmaContext::PrepareDecoder(uint8_t* packet, int sample_rate,
bool is_two_channel) {
// Sanity check: Packet metadata is always 1 for XMA2/0 for XMA
assert_true((packet[2] & 0x7) == 1 || (packet[2] & 0x7) == 0);
sample_rate = GetSampleRate(sample_rate);
// Re-initialize the context with new sample rate and channels.
uint32_t channels = is_two_channel ? 2 : 1;
if (av_context_->sample_rate != sample_rate ||
av_context_->channels != channels) {
// We have to reopen the codec so it'll realloc whatever data it needs.
// TODO(DrChat): Find a better way.
avcodec_close(av_context_);
av_context_->sample_rate = sample_rate;
av_context_->channels = channels;
if (avcodec_open2(av_context_, av_codec_, NULL) < 0) {
XELOGE("XmaContext: Failed to reopen FFmpeg context");
return -1;
}
return 1;
}
return 0;
}
void XmaContext::ConvertFrame(const uint8_t** samples, bool is_two_channel,
uint8_t* output_buffer) {
// Loop through every sample, convert and drop it into the output array.

99
src/xenia/apu/xma_context.h
View File

@ -76,7 +76,7 @@ struct XMA_CONTEXT_DATA {
// DWORD 2
uint32_t input_buffer_read_offset : 26; // XMAGetInputBufferReadOffset
uint32_t unk_dword_2 : 6; // ErrorStatus/ErrorSet (?)
uint32_t error_status : 6; // ErrorStatus/ErrorSet (?)
// DWORD 3
uint32_t loop_start : 26; // XMASetLoopData LoopStartOffset
@ -119,6 +119,34 @@ struct XMA_CONTEXT_DATA {
reinterpret_cast<const uint32_t*>(this),
sizeof(XMA_CONTEXT_DATA) / 4);
}
bool IsInputBufferValid(uint8_t buffer_index) const {
return buffer_index == 0 ? input_buffer_0_valid : input_buffer_1_valid;
}
bool IsCurrentInputBufferValid() const {
return IsInputBufferValid(current_buffer);
}
bool IsAnyInputBufferValid() const {
return input_buffer_0_valid || input_buffer_1_valid;
}
const uint32_t GetInputBufferAddress(uint8_t buffer_index) const {
return buffer_index == 0 ? input_buffer_0_ptr : input_buffer_1_ptr;
}
const uint32_t GetCurrentInputBufferAddress() const {
return GetInputBufferAddress(current_buffer);
}
const uint32_t GetInputBufferPacketCount(uint8_t buffer_index) const {
return buffer_index == 0 ? input_buffer_0_packet_count
: input_buffer_1_packet_count;
}
const uint32_t GetCurrentInputBufferPacketCount() const {
return GetInputBufferPacketCount(current_buffer);
}
};
static_assert_size(XMA_CONTEXT_DATA, 64);
@ -150,14 +178,16 @@ class XmaContext {
explicit XmaContext();
~XmaContext();
int Setup(uint32_t id, Memory* memory, uint32_t guest_ptr);
bool Work();
virtual int Setup(uint32_t id, Memory* memory, uint32_t guest_ptr) {
return 0;
};
virtual bool Work() { return false; };
void Enable();
bool Block(bool poll);
void Clear();
void Disable();
void Release();
virtual void Enable(){};
virtual bool Block(bool poll) { return 0; };
virtual void Clear(){};
virtual void Disable(){};
virtual void Release(){};
Memory* memory() const { return memory_; }
@ -169,37 +199,12 @@ class XmaContext {
void set_is_allocated(bool is_allocated) { is_allocated_ = is_allocated; }
void set_is_enabled(bool is_enabled) { is_enabled_ = is_enabled; }
private:
static void SwapInputBuffer(XMA_CONTEXT_DATA* data);
static bool TrySetupNextLoop(XMA_CONTEXT_DATA* data,
bool ignore_input_buffer_offset);
static void NextPacket(XMA_CONTEXT_DATA* data);
static int GetSampleRate(int id);
// Get the offset of the next frame. Does not traverse packets.
static size_t GetNextFrame(uint8_t* block, size_t size, size_t bit_offset);
// Get the containing packet number of the frame pointed to by the offset.
static int GetFramePacketNumber(uint8_t* block, size_t size,
size_t bit_offset);
// Get the packet number and the index of the frame inside that packet
static std::tuple<int, int> GetFrameNumber(uint8_t* block, size_t size,
size_t bit_offset);
// Get the number of frames contained in the packet (including truncated) and
// if the last frame is split.
static std::tuple<int, bool> GetPacketFrameCount(uint8_t* packet);
protected:
static void DumpRaw(AVFrame* frame, int id);
// Convert sample format and swap bytes
static void ConvertFrame(const uint8_t** samples, bool is_two_channel,
uint8_t* output_buffer);
bool ValidFrameOffset(uint8_t* block, size_t size_bytes,
size_t frame_offset_bits);
void Decode(XMA_CONTEXT_DATA* data);
int PrepareDecoder(uint8_t* packet, int sample_rate, bool is_two_channel);
// This method should be used ONLY when we're at the last packet of the stream
// and we want to find offset in next buffer
uint32_t GetPacketFirstFrameOffset(const XMA_CONTEXT_DATA* data);
Memory* memory_ = nullptr;
uint32_t id_ = 0;
@ -207,36 +212,12 @@ class XmaContext {
xe_mutex lock_;
volatile bool is_allocated_ = false;
volatile bool is_enabled_ = false;
// bool is_dirty_ = true;
// ffmpeg structures
AVPacket* av_packet_ = nullptr;
AVCodec* av_codec_ = nullptr;
AVCodecContext* av_context_ = nullptr;
AVFrame* av_frame_ = nullptr;
// uint32_t decoded_consumed_samples_ = 0; // TODO do this dynamically
// int decoded_idx_ = -1;
// bool partial_frame_saved_ = false;
// bool partial_frame_size_known_ = false;
// size_t partial_frame_total_size_bits_ = 0;
// size_t partial_frame_start_offset_bits_ = 0;
// size_t partial_frame_offset_bits_ = 0; // blah internal don't use this
// std::vector<uint8_t> partial_frame_buffer_;
uint32_t packets_skip_ = 0;
bool is_stream_done_ = false;
// bool split_frame_pending_ = false;
uint32_t split_frame_len_ = 0;
uint32_t split_frame_len_partial_ = 0;
uint8_t split_frame_padding_start_ = 0;
// first byte contains bit offset information
std::array<uint8_t, 1 + 4096> xma_frame_;
// uint8_t* current_frame_ = nullptr;
// conversion buffer for 2 channel frame
std::array<uint8_t, kBytesPerFrameChannel * 2> raw_frame_;
// std::vector<uint8_t> current_frame_ = std::vector<uint8_t>(0);
};
} // namespace apu

694
src/xenia/apu/xma_context_new.cc Normal file
View File

@ -0,0 +1,694 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2024 Xenia Canary. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/apu/xma_context_new.h"
#include "xenia/apu/xma_helpers.h"
#include <algorithm>
#include "xenia/base/logging.h"
#include "xenia/base/platform.h"
#include "xenia/base/profiling.h"
extern "C" {
#if XE_COMPILER_MSVC
#pragma warning(push)
#pragma warning(disable : 4101 4244 5033)
#endif
#include "third_party/FFmpeg/libavcodec/avcodec.h"
#if XE_COMPILER_MSVC
#pragma warning(pop)
#endif
} // extern "C"
// Credits for most of this code goes to:
// https://github.com/koolkdev/libertyv/blob/master/libav_wrapper/xma2dec.c
namespace xe {
namespace apu {
XmaContextNew::XmaContextNew() = default;
XmaContextNew::~XmaContextNew() {
if (av_context_) {
if (avcodec_is_open(av_context_)) {
avcodec_close(av_context_);
}
av_free(av_context_);
}
if (av_frame_) {
av_frame_free(&av_frame_);
}
}
int XmaContextNew::Setup(uint32_t id, Memory* memory, uint32_t guest_ptr) {
id_ = id;
memory_ = memory;
guest_ptr_ = guest_ptr;
// Allocate ffmpeg stuff:
av_packet_ = av_packet_alloc();
assert_not_null(av_packet_);
av_packet_->buf = av_buffer_alloc(128 * 1024);
// find the XMA2 audio decoder
av_codec_ = avcodec_find_decoder(AV_CODEC_ID_XMAFRAMES);
if (!av_codec_) {
XELOGE("XmaContext {}: Codec not found", id);
return 1;
}
av_context_ = avcodec_alloc_context3(av_codec_);
if (!av_context_) {
XELOGE("XmaContext {}: Couldn't allocate context", id);
return 1;
}
// Initialize these to 0. They'll actually be set later.
av_context_->channels = 0;
av_context_->sample_rate = 0;
av_frame_ = av_frame_alloc();
if (!av_frame_) {
XELOGE("XmaContext {}: Couldn't allocate frame", id);
return 1;
}
// FYI: We're purposely not opening the codec here. That is done later.
return 0;
}
RingBuffer XmaContextNew::PrepareOutputRingBuffer(XMA_CONTEXT_DATA* data) {
const uint32_t output_capacity =
data->output_buffer_block_count * kOutputBytesPerBlock;
const uint32_t output_read_offset =
data->output_buffer_read_offset * kOutputBytesPerBlock;
const uint32_t output_write_offset =
data->output_buffer_write_offset * kOutputBytesPerBlock;
if (output_capacity > kOutputMaxSizeBytes) {
XELOGW(
"XmaContext {}: Output buffer uses more space than expected! "
"(Actual: {} Max: {})",
id(), output_capacity, kOutputMaxSizeBytes);
}
uint8_t* output_buffer = memory()->TranslatePhysical(data->output_buffer_ptr);
// Output buffers are in raw PCM samples, 256 bytes per block.
// Output buffer is a ring buffer. We need to write from the write offset
// to the read offset.
RingBuffer output_rb(output_buffer, output_capacity);
output_rb.set_read_offset(output_read_offset);
output_rb.set_write_offset(output_write_offset);
remaining_subframe_blocks_in_output_buffer_ =
(int32_t)output_rb.write_count() / kOutputBytesPerBlock;
return output_rb;
}
bool XmaContextNew::Work() {
if (!is_enabled() || !is_allocated()) {
return false;
}
std::lock_guard<xe_mutex> lock(lock_);
set_is_enabled(false);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
if (!data.output_buffer_valid) {
return true;
}
RingBuffer output_rb = PrepareOutputRingBuffer(&data);
const int32_t minimum_subframe_decode_count =
(data.subframe_decode_count * 2) - 1;
// We don't have enough space to even make one pass
// Waiting for decoder to return more space.
if (minimum_subframe_decode_count >
remaining_subframe_blocks_in_output_buffer_) {
XELOGD("XmaContext {}: No space for subframe decoding {}/{}!", id(),
minimum_subframe_decode_count,
remaining_subframe_blocks_in_output_buffer_);
data.Store(context_ptr);
return true;
}
while (remaining_subframe_blocks_in_output_buffer_ >=
minimum_subframe_decode_count) {
XELOGAPU(
"XmaContext {}: Write Count: {}, Capacity: {} - {} {} Subframes: {} "
"Skip: {}",
id(), (uint32_t)output_rb.write_count(),
remaining_subframe_blocks_in_output_buffer_,
data.input_buffer_0_valid + (data.input_buffer_1_valid << 1),
data.output_buffer_valid, data.subframe_decode_count,
data.subframe_skip_count);
Decode(&data);
Consume(&output_rb, &data);
if (!data.IsAnyInputBufferValid() || data.error_status == 4) {
break;
}
}
data.output_buffer_write_offset =
output_rb.write_offset() / kOutputBytesPerBlock;
XELOGAPU("XmaContext {}: Read Output: {} Write Output: {}", id(),
data.output_buffer_read_offset, data.output_buffer_write_offset);
// That's a bit misleading due to nature of ringbuffer
// when write and read offset matches it might mean that we wrote nothing
// or we fully saturated allocated space.
if (output_rb.empty()) {
data.output_buffer_valid = 0;
}
// TODO: Rewrite!
// There is a case when game can modify certain parts of context mid-play
// and decoder should be aware of it
data.Store(context_ptr);
return true;
}
void XmaContextNew::Enable() {
std::lock_guard<xe_mutex> lock(lock_);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
XELOGAPU("XmaContext: kicking context {} (buffer {} {}/{} bits)", id(),
data.current_buffer, data.input_buffer_read_offset,
data.GetCurrentInputBufferPacketCount() * kBitsPerPacket);
data.Store(context_ptr);
set_is_enabled(true);
}
bool XmaContextNew::Block(bool poll) {
if (!lock_.try_lock()) {
if (poll) {
return false;
}
lock_.lock();
}
lock_.unlock();
return true;
}
void XmaContextNew::Clear() {
std::lock_guard<xe_mutex> lock(lock_);
XELOGAPU("XmaContext: reset context {}", id());
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
data.input_buffer_0_valid = 0;
data.input_buffer_1_valid = 0;
data.output_buffer_valid = 0;
data.input_buffer_read_offset = 0;
data.output_buffer_read_offset = 0;
data.output_buffer_write_offset = 0;
data.input_buffer_read_offset = kBitsPerPacketHeader;
current_frame_remaining_subframes_ = 0;
data.Store(context_ptr);
}
void XmaContextNew::Disable() {
std::lock_guard<xe_mutex> lock(lock_);
XELOGAPU("XmaContext: disabling context {}", id());
set_is_enabled(false);
}
void XmaContextNew::Release() {
// Lock it in case the decoder thread is working on it now.
std::lock_guard<xe_mutex> lock(lock_);
assert_true(is_allocated_ == true);
set_is_allocated(false);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
std::memset(context_ptr, 0, sizeof(XMA_CONTEXT_DATA)); // Zero it.
}
int XmaContextNew::GetSampleRate(int id) {
return kIdToSampleRate[std::min(id, 3)];
}
void XmaContextNew::SwapInputBuffer(XMA_CONTEXT_DATA* data) {
// No more frames.
if (data->current_buffer == 0) {
data->input_buffer_0_valid = 0;
} else {
data->input_buffer_1_valid = 0;
}
data->current_buffer ^= 1;
data->input_buffer_read_offset = kBitsPerPacketHeader;
}
void XmaContextNew::Consume(RingBuffer* output_rb, XMA_CONTEXT_DATA* data) {
if (!current_frame_remaining_subframes_) {
return;
}
const int8_t subframes_to_write =
std::min((int8_t)current_frame_remaining_subframes_,
(int8_t)data->subframe_decode_count);
const int8_t raw_frame_read_offset =
((kBytesPerFrameChannel / kOutputBytesPerBlock) << data->is_stereo) -
current_frame_remaining_subframes_;
// + data->subframe_skip_count;
output_rb->Write(
raw_frame_.data() + (kOutputBytesPerBlock * raw_frame_read_offset),
subframes_to_write * kOutputBytesPerBlock);
remaining_subframe_blocks_in_output_buffer_ -= subframes_to_write;
current_frame_remaining_subframes_ -= subframes_to_write;
XELOGAPU("XmaContext {}: Consume: {} - {} - {} - {} - {}", id(),
remaining_subframe_blocks_in_output_buffer_,
data->output_buffer_write_offset, data->output_buffer_read_offset,
output_rb->write_offset(), current_frame_remaining_subframes_);
}
void XmaContextNew::Decode(XMA_CONTEXT_DATA* data) {
SCOPE_profile_cpu_f("apu");
// No available data.
if (!data->IsAnyInputBufferValid()) {
// data->error_status = 4;
return;
}
if (current_frame_remaining_subframes_ > 0) {
return;
}
uint8_t* current_input_buffer = GetCurrentInputBuffer(data);
if (!data->IsCurrentInputBufferValid()) {
XELOGE(
"XmaContext {}: Invalid current buffer! Selected Buffer: {} Valid: {} "
"Pointer: {:08X}",
id(), data->current_buffer, data->IsCurrentInputBufferValid(),
data->GetCurrentInputBufferAddress());
return;
}
input_buffer_.fill(0);
UpdateLoopStatus(data);
if (!data->output_buffer_block_count) {
XELOGE("XmaContext {}: Error - Received 0 for output_buffer_block_count!",
id());
return;
}
XELOGAPU(
"Processing context {} (offset {}, buffer {}, ptr {:p}, output buffer "
"{:08X}, output buffer count {})",
id(), data->input_buffer_read_offset, data->current_buffer,
current_input_buffer, data->output_buffer_ptr,
data->output_buffer_block_count);
const uint32_t current_input_size = GetCurrentInputBufferSize(data);
const uint32_t current_input_packet_count =
current_input_size / kBytesPerPacket;
const int16_t packet_index =
GetPacketNumber(current_input_size, data->input_buffer_read_offset);
if (packet_index == -1) {
XELOGE("XmaContext {}: Invalid packet index. Input read offset: {}", id(),
data->input_buffer_read_offset);
return;
}
uint8_t* packet = current_input_buffer + (packet_index * kBytesPerPacket);
// Because game can reset read offset. We must assure that new offset is
// valid. Splitted frames aren't handled here, so it's not a big deal.
const uint32_t frame_offset = xma::GetPacketFrameOffset(packet);
if (data->input_buffer_read_offset < frame_offset) {
data->input_buffer_read_offset = frame_offset;
}
const uint32_t relative_offset =
data->input_buffer_read_offset % kBitsPerPacket;
const kPacketInfo packet_info = GetPacketInfo(packet, relative_offset);
const uint32_t packet_to_skip = xma::GetPacketSkipCount(packet) + 1;
const uint32_t next_packet_index = packet_index + packet_to_skip;
BitStream stream =
BitStream(current_input_buffer, (packet_index + 1) * kBitsPerPacket);
stream.SetOffset(data->input_buffer_read_offset);
const uint64_t bits_to_copy = GetAmountOfBitsToRead(
(uint32_t)stream.BitsRemaining(), packet_info.current_frame_size_);
if (bits_to_copy == 0) {
XELOGE("XmaContext {}: There is no bits to copy!", id());
SwapInputBuffer(data);
return;
}
if (packet_info.isLastFrameInPacket()) {
// Frame is a splitted frame
if (stream.BitsRemaining() < packet_info.current_frame_size_) {
const uint8_t* next_packet =
GetNextPacket(data, next_packet_index, current_input_packet_count);
if (!next_packet) {
// Error path
// Decoder probably should return error here
// Not sure what error code should be returned
data->error_status = 4;
return;
}
// Copy next packet to buffer
std::memcpy(input_buffer_.data() + kBytesPerPacketData,
next_packet + kBytesPerPacketHeader, kBytesPerPacketData);
}
}
// Copy current packet to buffer
std::memcpy(input_buffer_.data(), packet + kBytesPerPacketHeader,
kBytesPerPacketData);
stream = BitStream(input_buffer_.data(),
(kBitsPerPacket - kBitsPerPacketHeader) * 2);
stream.SetOffset(relative_offset - kBitsPerPacketHeader);
xma_frame_.fill(0);
XELOGAPU(
"XmaContext {}: Reading Frame {}/{} (size: {}) From Packet "
"{}/{}",
id(), (int32_t)packet_info.current_frame_, packet_info.frame_count_,
packet_info.current_frame_size_, packet_index,
current_input_packet_count);
const uint32_t padding_start = static_cast<uint8_t>(
stream.Copy(xma_frame_.data() + 1, packet_info.current_frame_size_));
raw_frame_.fill(0);
PrepareDecoder(data->sample_rate, bool(data->is_stereo));
PreparePacket(packet_info.current_frame_size_, padding_start);
if (DecodePacket(av_context_, av_packet_, av_frame_)) {
// dump_raw(av_frame_, id());
ConvertFrame(reinterpret_cast<const uint8_t**>(&av_frame_->data),
bool(data->is_stereo), raw_frame_.data());
}
// TODO: Write function to regenerate decoder
// TODO: Be aware of subframe_skips & loops subframes skips
current_frame_remaining_subframes_ = 4 << data->is_stereo;
// Compute where to go next.
if (!packet_info.isLastFrameInPacket()) {
const uint32_t next_frame_offset =
(data->input_buffer_read_offset + bits_to_copy) % kBitsPerPacket;
XELOGAPU("XmaContext {}: Index: {}/{} - Next frame offset: {}", id(),
(int32_t)packet_info.current_frame_, packet_info.frame_count_,
next_frame_offset);
data->input_buffer_read_offset =
(packet_index * kBitsPerPacket) + next_frame_offset;
return;
}
const uint8_t* next_packet =
GetNextPacket(data, next_packet_index, current_input_packet_count);
if (!next_packet) {
// Error path
// Decoder probably should return error here
// Not sure what error code should be returned
// data->error_status = 4;
// data->output_buffer_valid = 0;
// return;
}
uint32_t next_input_offset = GetNextPacketReadOffset(
current_input_buffer, next_packet_index, current_input_packet_count);
if (next_input_offset == kBitsPerPacketHeader) {
SwapInputBuffer(data);
// We're at start of next buffer
// If it have any frame in this packet decoder should go to first frame in
// packet If it doesn't have any frame then it should immediatelly go to
// next packet
if (data->IsAnyInputBufferValid()) {
next_input_offset = xma::GetPacketFrameOffset(
memory()->TranslatePhysical(data->GetCurrentInputBufferAddress()));
if (next_input_offset > kMaxFrameSizeinBits) {
XELOGAPU(
"XmaContext {}: Next buffer contains no frames in packet! Frame "
"offset: {}",
id(), next_input_offset);
SwapInputBuffer(data);
return;
}
XELOGAPU("XmaContext {}: Next buffer first frame starts at: {}", id(),
next_input_offset);
}
}
data->input_buffer_read_offset = next_input_offset;
return;
}
// Frame & Packet searching methods
void XmaContextNew::UpdateLoopStatus(XMA_CONTEXT_DATA* data) {
if (data->loop_count == 0) {
return;
}
const uint32_t loop_start = std::max(kBitsPerPacketHeader, data->loop_start);
const uint32_t loop_end = std::max(kBitsPerPacketHeader, data->loop_end);
XELOGAPU("XmaContext {}: Looped Data: {} < {} (Start: {}) Remaining: {}",
id(), data->input_buffer_read_offset, data->loop_end,
data->loop_start, data->loop_count);
if (data->input_buffer_read_offset != loop_end) {
return;
}
data->input_buffer_read_offset = loop_start;
if (data->loop_count != 255) {
data->loop_count--;
}
}
const uint8_t* XmaContextNew::GetNextPacket(
XMA_CONTEXT_DATA* data, uint32_t next_packet_index,
uint32_t current_input_packet_count) {
if (next_packet_index < current_input_packet_count) {
return memory()->TranslatePhysical(data->GetCurrentInputBufferAddress()) +
next_packet_index * kBytesPerPacket;
}
const uint8_t next_buffer_index = data->current_buffer ^ 1;
if (!data->IsInputBufferValid(next_buffer_index)) {
return nullptr;
}
const uint32_t next_buffer_address =
data->GetInputBufferAddress(next_buffer_index);
if (!next_buffer_address) {
// This should never occur but there is always a chance
XELOGE(
"XmaContext {}: Buffer is marked as valid, but doesn't have valid "
"pointer!",
id());
return nullptr;
}
return memory()->TranslatePhysical(next_buffer_address);
}
const uint32_t XmaContextNew::GetNextPacketReadOffset(
uint8_t* buffer, uint32_t next_packet_index,
uint32_t current_input_packet_count) {
if (next_packet_index >= current_input_packet_count) {
return kBitsPerPacketHeader;
}
uint8_t* next_packet = buffer + (next_packet_index * kBytesPerPacket);
const uint32_t packet_frame_offset = xma::GetPacketFrameOffset(next_packet);
if (packet_frame_offset > kMaxFrameSizeinBits) {
const uint32_t offset = GetNextPacketReadOffset(
buffer, next_packet_index + 1, current_input_packet_count);
return offset;
}
const uint32_t new_input_buffer_offset =
(next_packet_index * kBitsPerPacket) + packet_frame_offset;
XELOGAPU("XmaContext {}: new offset: {} packet_offset: {} packet: {}/{}",
id(), new_input_buffer_offset, packet_frame_offset,
next_packet_index, current_input_packet_count);
return new_input_buffer_offset;
}
const uint32_t XmaContextNew::GetAmountOfBitsToRead(
const uint32_t remaining_stream_bits, const uint32_t frame_size) {
return std::min(remaining_stream_bits, frame_size);
}
uint32_t XmaContextNew::GetCurrentInputBufferSize(XMA_CONTEXT_DATA* data) {
return data->GetCurrentInputBufferPacketCount() * kBytesPerPacket;
}
uint8_t* XmaContextNew::GetCurrentInputBuffer(XMA_CONTEXT_DATA* data) {
return memory()->TranslatePhysical(data->GetCurrentInputBufferAddress());
}
const kPacketInfo XmaContextNew::GetPacketInfo(uint8_t* packet,
uint32_t frame_offset) {
kPacketInfo packet_info = {};
const uint32_t first_frame_offset = xma::GetPacketFrameOffset(packet);
BitStream stream(packet, kBitsPerPacket);
stream.SetOffset(first_frame_offset);
// Handling of splitted frame
if (frame_offset < first_frame_offset) {
packet_info.current_frame_ = 0;
packet_info.current_frame_size_ = first_frame_offset - frame_offset;
}
while (true) {
if (stream.BitsRemaining() < kBitsPerFrameHeader) {
break;
}
const uint64_t frame_size = stream.Peek(kBitsPerFrameHeader);
if (frame_size == xma::kMaxFrameLength) {
break;
}
if (stream.offset_bits() == frame_offset) {
packet_info.current_frame_ = packet_info.frame_count_;
packet_info.current_frame_size_ = (uint32_t)frame_size;
}
packet_info.frame_count_++;
if (frame_size > stream.BitsRemaining()) {
// Last frame.
break;
}
stream.Advance(frame_size - 1);
// Read the trailing bit to see if frames follow
if (stream.Read(1) == 0) {
break;
}
}
if (xma::IsPacketXma2Type(packet)) {
const uint8_t xma2_frame_count = xma::GetPacketFrameCount(packet);
if (xma2_frame_count != packet_info.frame_count_) {
XELOGE(
"XmaContext {}: XMA2 packet header defines different amount of "
"frames than internally found! (Header: {} Found: {})",
id(), xma2_frame_count, packet_info.frame_count_);
}
}
return packet_info;
}
int16_t XmaContextNew::GetPacketNumber(size_t size, size_t bit_offset) {
if (bit_offset < kBitsPerPacketHeader) {
assert_always();
return -1;
}
if (bit_offset >= (size << 3)) {
assert_always();
return -1;
}
size_t byte_offset = bit_offset >> 3;
size_t packet_number = byte_offset / kBytesPerPacket;
return (int16_t)packet_number;
}
int XmaContextNew::PrepareDecoder(int sample_rate, bool is_two_channel) {
sample_rate = GetSampleRate(sample_rate);
// Re-initialize the context with new sample rate and channels.
uint32_t channels = is_two_channel ? 2 : 1;
if (av_context_->sample_rate != sample_rate ||
av_context_->channels != channels) {
// We have to reopen the codec so it'll realloc whatever data it needs.
// TODO(DrChat): Find a better way.
avcodec_close(av_context_);
av_context_->sample_rate = sample_rate;
av_context_->channels = channels;
if (avcodec_open2(av_context_, av_codec_, NULL) < 0) {
XELOGE("XmaContext: Failed to reopen FFmpeg context");
return -1;
}
return 1;
}
return 0;
}
void XmaContextNew::PreparePacket(const uint32_t frame_size,
const uint32_t frame_padding) {
av_packet_->data = xma_frame_.data();
av_packet_->size =
static_cast<int>(1 + ((frame_padding + frame_size) / 8) +
(((frame_padding + frame_size) % 8) ? 1 : 0));
auto padding_end = av_packet_->size * 8 - (8 + frame_padding + frame_size);
assert_true(padding_end < 8);
xma_frame_[0] = ((frame_padding & 7) << 5) | ((padding_end & 7) << 2);
}
bool XmaContextNew::DecodePacket(AVCodecContext* av_context,
const AVPacket* av_packet, AVFrame* av_frame) {
auto ret = avcodec_send_packet(av_context, av_packet);
if (ret < 0) {
XELOGE("XmaContext {}: Error sending packet for decoding", id());
return false;
}
ret = avcodec_receive_frame(av_context, av_frame);
if (ret < 0) {
XELOGE("XmaContext {}: Error during decoding", id());
return false;
}
return true;
}
} // namespace apu
} // namespace xe

136
src/xenia/apu/xma_context_new.h Normal file
View File

@ -0,0 +1,136 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2024 Xenia Canary. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_APU_XMA_CONTEXT_NEW_H_
#define XENIA_APU_XMA_CONTEXT_NEW_H_
#include <array>
#include <atomic>
#include <mutex>
#include <queue>
#include "xenia/apu/xma_context.h"
#include "xenia/base/bit_stream.h"
#include "xenia/base/ring_buffer.h"
#include "xenia/memory.h"
#include "xenia/xbox.h"
// Forward declarations
struct AVCodec;
struct AVCodecParserContext;
struct AVCodecContext;
struct AVFrame;
struct AVPacket;
namespace xe {
namespace apu {
struct kPacketInfo {
uint8_t frame_count_;
uint8_t current_frame_;
uint32_t current_frame_size_;
const bool isLastFrameInPacket() const {
return current_frame_ == frame_count_ - 1;
}
};
static constexpr int kIdToSampleRate[4] = {24000, 32000, 44100, 48000};
class XmaContextNew : public XmaContext {
public:
static const uint32_t kBytesPerPacket = 2048;
static const uint32_t kBytesPerPacketHeader = 4;
static const uint32_t kBytesPerPacketData =
kBytesPerPacket - kBytesPerPacketHeader;
static const uint32_t kBitsPerPacket = kBytesPerPacket * 8;
static const uint32_t kBitsPerPacketHeader = 32;
static const uint32_t kBitsPerFrameHeader = 15;
static const uint32_t kBytesPerSample = 2;
static const uint32_t kSamplesPerFrame = 512;
static const uint32_t kSamplesPerSubframe = 128;
static const uint32_t kBytesPerFrameChannel =
kSamplesPerFrame * kBytesPerSample;
static const uint32_t kBytesPerSubframeChannel =
kSamplesPerSubframe * kBytesPerSample;
static const uint32_t kOutputBytesPerBlock = 256;
static const uint32_t kOutputMaxSizeBytes = 31 * kOutputBytesPerBlock;
static const uint32_t kLastFrameMarker = 0x7FFF;
static const uint32_t kMaxFrameSizeinBits = 0x4000 - kBitsPerPacketHeader;
explicit XmaContextNew();
~XmaContextNew();
int Setup(uint32_t id, Memory* memory, uint32_t guest_ptr);
bool Work();
void Enable();
bool Block(bool poll);
void Clear();
void Disable();
void Release();
private:
static void SwapInputBuffer(XMA_CONTEXT_DATA* data);
// Convert sampling rate from ID to frequency.
static int GetSampleRate(int id);
// Get the containing packet number of the frame pointed to by the offset.
static int16_t GetPacketNumber(size_t size, size_t bit_offset);
const kPacketInfo GetPacketInfo(uint8_t* packet, uint32_t frame_offset);
const uint32_t GetAmountOfBitsToRead(const uint32_t remaining_stream_bits,
const uint32_t frame_size);
const uint8_t* GetNextPacket(XMA_CONTEXT_DATA* data,
uint32_t next_packet_index,
uint32_t current_input_packet_count);
const uint32_t GetNextPacketReadOffset(uint8_t* buffer,
uint32_t next_packet_index,
uint32_t current_input_packet_count);
// Returns currently used buffer
uint8_t* GetCurrentInputBuffer(XMA_CONTEXT_DATA* data);
static uint32_t GetCurrentInputBufferSize(XMA_CONTEXT_DATA* data);
void Decode(XMA_CONTEXT_DATA* data);
void Consume(RingBuffer* output_rb, XMA_CONTEXT_DATA* data);
void UpdateLoopStatus(XMA_CONTEXT_DATA* data);
int PrepareDecoder(int sample_rate, bool is_two_channel);
void PreparePacket(const uint32_t frame_size, const uint32_t frame_padding);
RingBuffer PrepareOutputRingBuffer(XMA_CONTEXT_DATA* data);
bool DecodePacket(AVCodecContext* av_context, const AVPacket* av_packet,
AVFrame* av_frame);
// This method should be used ONLY when we're at the last packet of the stream
// and we want to find offset in next buffer
uint32_t GetPacketFirstFrameOffset(const XMA_CONTEXT_DATA* data);
std::array<uint8_t, kBytesPerPacketData * 2> input_buffer_;
// first byte contains bit offset information
std::array<uint8_t, 1 + 4096> xma_frame_;
std::array<uint8_t, kBytesPerFrameChannel * 2> raw_frame_;
int32_t remaining_subframe_blocks_in_output_buffer_ = 0;
uint8_t current_frame_remaining_subframes_ = 0;
};
} // namespace apu
} // namespace xe
#endif // XENIA_APU_XMA_CONTEXT_H_

950
src/xenia/apu/xma_context_old.cc Normal file
View File

@ -0,0 +1,950 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2024 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/apu/xma_context_old.h"
#include <algorithm>
#include <cstring>
#include "xenia/apu/xma_decoder.h"
#include "xenia/apu/xma_helpers.h"
#include "xenia/base/bit_stream.h"
#include "xenia/base/logging.h"
#include "xenia/base/platform.h"
#include "xenia/base/profiling.h"
#include "xenia/base/ring_buffer.h"
extern "C" {
#if XE_COMPILER_MSVC
#pragma warning(push)
#pragma warning(disable : 4101 4244 5033)
#endif
#include "third_party/FFmpeg/libavcodec/avcodec.h"
#if XE_COMPILER_MSVC
#pragma warning(pop)
#endif
} // extern "C"
// Credits for most of this code goes to:
// https://github.com/koolkdev/libertyv/blob/master/libav_wrapper/xma2dec.c
namespace xe {
namespace apu {
XmaContextOld::XmaContextOld() = default;
XmaContextOld::~XmaContextOld() {
if (av_context_) {
if (avcodec_is_open(av_context_)) {
avcodec_close(av_context_);
}
av_free(av_context_);
}
if (av_frame_) {
av_frame_free(&av_frame_);
}
// if (current_frame_) {
// delete[] current_frame_;
// }
}
int XmaContextOld::Setup(uint32_t id, Memory* memory, uint32_t guest_ptr) {
id_ = id;
memory_ = memory;
guest_ptr_ = guest_ptr;
// Allocate ffmpeg stuff:
av_packet_ = av_packet_alloc();
assert_not_null(av_packet_);
// chrispy: preallocate this buffer so that ffmpeg isn't reallocating it for
// every packet, these allocations were causing RtlSubsegmentInitialize
av_packet_->buf = av_buffer_alloc(128 * 1024);
// find the XMA2 audio decoder
av_codec_ = avcodec_find_decoder(AV_CODEC_ID_XMAFRAMES);
if (!av_codec_) {
XELOGE("XmaContext {}: Codec not found", id);
return 1;
}
av_context_ = avcodec_alloc_context3(av_codec_);
if (!av_context_) {
XELOGE("XmaContext {}: Couldn't allocate context", id);
return 1;
}
// Initialize these to 0. They'll actually be set later.
av_context_->channels = 0;
av_context_->sample_rate = 0;
av_frame_ = av_frame_alloc();
if (!av_frame_) {
XELOGE("XmaContext {}: Couldn't allocate frame", id);
return 1;
}
// FYI: We're purposely not opening the codec here. That is done later.
return 0;
}
bool XmaContextOld::Work() {
if (!is_enabled() || !is_allocated()) {
return false;
}
{
std::lock_guard<xe_mutex> lock(lock_);
set_is_enabled(false);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
Decode(&data);
data.Store(context_ptr);
return true;
}
}
void XmaContextOld::Enable() {
std::lock_guard<xe_mutex> lock(lock_);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
XELOGAPU("XmaContext: kicking context {} (buffer {} {}/{} bits)", id(),
data.current_buffer, data.input_buffer_read_offset,
(data.current_buffer == 0 ? data.input_buffer_0_packet_count
: data.input_buffer_1_packet_count) *
kBitsPerPacket);
data.Store(context_ptr);
set_is_enabled(true);
}
bool XmaContextOld::Block(bool poll) {
if (!lock_.try_lock()) {
if (poll) {
return false;
}
lock_.lock();
}
lock_.unlock();
return true;
}
void XmaContextOld::Clear() {
std::lock_guard<xe_mutex> lock(lock_);
XELOGAPU("XmaContext: reset context {}", id());
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
XMA_CONTEXT_DATA data(context_ptr);
data.input_buffer_0_valid = 0;
data.input_buffer_1_valid = 0;
data.output_buffer_valid = 0;
data.input_buffer_read_offset = 0;
data.output_buffer_read_offset = 0;
data.output_buffer_write_offset = 0;
xma_frame_.fill(0);
split_frame_len_ = 0;
split_frame_len_partial_ = 0;
split_frame_padding_start_ = 0;
data.Store(context_ptr);
}
void XmaContextOld::Disable() {
std::lock_guard<xe_mutex> lock(lock_);
XELOGAPU("XmaContext: disabling context {}", id());
set_is_enabled(false);
}
void XmaContextOld::Release() {
// Lock it in case the decoder thread is working on it now.
std::lock_guard<xe_mutex> lock(lock_);
assert_true(is_allocated_ == true);
set_is_allocated(false);
auto context_ptr = memory()->TranslateVirtual(guest_ptr());
std::memset(context_ptr, 0, sizeof(XMA_CONTEXT_DATA)); // Zero it.
}
void XmaContextOld::SwapInputBuffer(XMA_CONTEXT_DATA* data) {
// No more frames.
if (data->current_buffer == 0) {
data->input_buffer_0_valid = 0;
} else {
data->input_buffer_1_valid = 0;
}
data->current_buffer ^= 1;
data->input_buffer_read_offset = kBitsPerHeader;
}
bool XmaContextOld::TrySetupNextLoop(XMA_CONTEXT_DATA* data,
bool ignore_input_buffer_offset) {
// Setup the input buffer offset if next loop exists.
// TODO(Pseudo-Kernel): Need to handle loop in the following cases.
// 1. loop_start == loop_end == 0
// 2. loop_start > loop_end && loop_count > 0
if (data->loop_count > 0 && data->loop_start < data->loop_end &&
(ignore_input_buffer_offset ||
data->input_buffer_read_offset >= data->loop_end)) {
// Loop back to the beginning.
data->input_buffer_read_offset = data->loop_start;
if (data->loop_count < 255) {
data->loop_count--;
}
return true;
}
return false;
}
/*
void XmaContext::NextPacket(
uint8_t* input_buffer,
uint32_t input_size,
uint32_t input_buffer_read_offset) {
*/
void XmaContextOld::NextPacket(XMA_CONTEXT_DATA* data) {
// auto packet_idx = GetFramePacketNumber(input_buffer, input_size,
// input_buffer_read_offset);
// packet_idx++;
// if (packet_idx++ >= input_size)
}
int XmaContextOld::GetSampleRate(int id) {
switch (id) {
case 0:
return 24000;
case 1:
return 32000;
case 2:
return 44100;
case 3:
return 48000;
}
assert_always();
return 0;
}
bool XmaContextOld::ValidFrameOffset(uint8_t* block, size_t size_bytes,
size_t frame_offset_bits) {
uint32_t packet_num =
GetFramePacketNumber(block, size_bytes, frame_offset_bits);
if (packet_num == -1) {
// Invalid packet number
XELOGAPU("ValidFrameOffset: Invalid packet number");
return false;
}
uint8_t* packet = block + (packet_num * kBytesPerPacket);
size_t relative_offset_bits = frame_offset_bits % kBitsPerPacket;
uint32_t first_frame_offset = xma::GetPacketFrameOffset(packet);
if (first_frame_offset == -1 || first_frame_offset > kBitsPerPacket) {
XELOGAPU("ValidFrameOffset: Invalid frame offset {}", first_frame_offset);
// Packet only contains a partial frame, so no frames can start here.
return false;
}
BitStream stream(packet, kBitsPerPacket);
stream.SetOffset(first_frame_offset);
while (true) {
if (stream.offset_bits() == relative_offset_bits) {
return true;
}
if (stream.BitsRemaining() < 15) {
XELOGAPU("ValidFrameOffset: No room for next frame header {}",
first_frame_offset);
// Not enough room for another frame header.
return false;
}
uint64_t size = stream.Read(15);
if ((size - 15) > stream.BitsRemaining()) {
XELOGAPU("ValidFrameOffset: Last frame {} - {}", first_frame_offset,
size);
// Last frame.
return false;
} else if (size == 0x7FFF) {
// Invalid frame (and last of this packet)
return false;
}
stream.Advance(size - 16);
// Read the trailing bit to see if frames follow
if (stream.Read(1) == 0) {
break;
}
}
return false;
}
void XmaContextOld::Decode(XMA_CONTEXT_DATA* data) {
SCOPE_profile_cpu_f("apu");
// What I see:
// XMA outputs 2 bytes per sample
// 512 samples per frame (128 per subframe)
// Max output size is data.output_buffer_block_count * 256
// This decoder is fed packets (max 4095 per buffer)
// Packets contain "some" frames
// 32bit header (big endian)
// Frames are the smallest thing the SPUs can decode.
// They can and usually will span packets.
// Sample rates (data.sample_rate):
// 0 - 24 kHz
// 1 - 32 kHz
// 2 - 44.1 kHz
// 3 - 48 kHz
// SPUs also support stereo decoding. (data.is_stereo)
// Check the output buffer - we cannot decode anything else if it's
// unavailable.
if (!data->output_buffer_valid) {
return;
}
// No available data.
if (!data->input_buffer_0_valid && !data->input_buffer_1_valid) {
return;
}
// XAudio Loops
// loop_count:
// - XAUDIO2_MAX_LOOP_COUNT = 254
// - XAUDIO2_LOOP_INFINITE = 255
// loop_start/loop_end are bit offsets to a specific frame
// Translate pointers for future use.
// Sometimes the game will use rolling input buffers. If they do, we cannot
// assume they form a complete block! In addition, the buffers DO NOT have
// to be contiguous!
uint8_t* in0 = data->input_buffer_0_valid
? memory()->TranslatePhysical(data->input_buffer_0_ptr)
: nullptr;
uint8_t* in1 = data->input_buffer_1_valid
? memory()->TranslatePhysical(data->input_buffer_1_ptr)
: nullptr;
uint8_t* current_input_buffer = data->current_buffer ? in1 : in0;
if (!current_input_buffer) {
XELOGE("XmaContext {}: Error - input buffer pointer is invalid!", id());
return;
}
if (!data->output_buffer_block_count) {
XELOGE("XmaContext {}: Error - Received 0 for output_buffer_block_count!",
id());
return;
}
XELOGAPU(
"Processing context {} (offset {}, buffer {}, ptr {:p}, output buffer "
"{:08X}, output buffer count {})",
id(), data->input_buffer_read_offset, data->current_buffer,
current_input_buffer, data->output_buffer_ptr,
data->output_buffer_block_count);
if (is_stream_done_) {
is_stream_done_ = false;
packets_skip_ = 0;
SwapInputBuffer(data);
return;
}
size_t input_buffer_0_size =
data->input_buffer_0_packet_count * kBytesPerPacket;
size_t input_buffer_1_size =
data->input_buffer_1_packet_count * kBytesPerPacket;
size_t current_input_size =
data->current_buffer ? input_buffer_1_size : input_buffer_0_size;
size_t current_input_packet_count = current_input_size / kBytesPerPacket;
bool is_streaming = data->input_buffer_0_packet_count == 1 &&
data->input_buffer_1_packet_count == 1;
// Output buffers are in raw PCM samples, 256 bytes per block.
// Output buffer is a ring buffer. We need to write from the write offset
// to the read offset.
uint8_t* output_buffer = memory()->TranslatePhysical(data->output_buffer_ptr);
uint32_t output_capacity =
data->output_buffer_block_count * kBytesPerSubframeChannel;
uint32_t output_read_offset =
data->output_buffer_read_offset * kBytesPerSubframeChannel;
uint32_t output_write_offset =
data->output_buffer_write_offset * kBytesPerSubframeChannel;
RingBuffer output_rb(output_buffer, output_capacity);
output_rb.set_read_offset(output_read_offset);
output_rb.set_write_offset(output_write_offset);
// We can only decode an entire frame and write it out at a time, so
// don't save any samples.
// TODO(JoelLinn): subframes when looping
size_t output_remaining_bytes = output_rb.write_count();
output_remaining_bytes -=
output_remaining_bytes % (kBytesPerFrameChannel << data->is_stereo);
// is_dirty_ = true; // TODO
// is_dirty_ = false; // TODO
assert_false(data->stop_when_done);
assert_false(data->interrupt_when_done);
static int total_samples = 0;
// Decode until we can't write any more data.
while (output_remaining_bytes > 0) {
if (!data->input_buffer_0_valid && !data->input_buffer_1_valid) {
// Out of data.
break;
}
// Setup the input buffer if we are at loop_end.
// The input buffer must not be swapped out until all loops are processed.
bool reuse_input_buffer = TrySetupNextLoop(data, false);
// assert_true(packets_skip_ == 0);
// assert_true(split_frame_len_ == 0);
// assert_true(split_frame_len_partial_ == 0);
// Where are we in the buffer (in XMA jargon)
int packet_idx, frame_idx, frame_count;
uint8_t* packet;
bool frame_last_split;
BitStream stream(current_input_buffer, current_input_size * 8);
stream.SetOffset(data->input_buffer_read_offset);
if (data->input_buffer_read_offset > current_input_size * 8) {
XELOGE(
"XmaContext {}: Error - Provided input offset exceed input buffer "
"size! ({} > {})",
id(), data->input_buffer_read_offset, current_input_size * 8);
SwapInputBuffer(data);
return;
}
// if we had a buffer swap try to skip packets first
if (packets_skip_ > 0) {
packet_idx =
GetFramePacketNumber(current_input_buffer, current_input_size,
data->input_buffer_read_offset);
while (packets_skip_ > 0) {
packets_skip_--;
packet_idx++;
if (packet_idx > current_input_packet_count) {
if (!reuse_input_buffer) {
// Last packet. Try setup once more.
reuse_input_buffer = TrySetupNextLoop(data, true);
}
if (!reuse_input_buffer) {
if (is_streaming) {
SwapInputBuffer(data);
} else {
is_stream_done_ = true;
}
}
return;
}
}
// invalid frame pointer but needed for us
data->input_buffer_read_offset = packet_idx * kBitsPerPacket;
// continue;
}
if (split_frame_len_) {
// handle a frame that was split over two packages
packet_idx =
GetFramePacketNumber(current_input_buffer, current_input_size,
data->input_buffer_read_offset);
packet = current_input_buffer + packet_idx * kBytesPerPacket;
std::tie(frame_count, frame_last_split) = GetPacketFrameCount(packet);
frame_idx = -1;
stream =
BitStream(current_input_buffer, (packet_idx + 1) * kBitsPerPacket);
stream.SetOffset(packet_idx * kBitsPerPacket + kBitsPerHeader);
if (split_frame_len_ > xma::kMaxFrameLength) {
// TODO write CopyPeekMethod
auto offset = stream.offset_bits();
stream.Copy(
xma_frame_.data() + 1 +
((split_frame_len_partial_ + split_frame_padding_start_) / 8),
15 - split_frame_len_partial_);
stream.SetOffset(offset);
BitStream slen(xma_frame_.data() + 1, 15 + split_frame_padding_start_);
slen.Advance(split_frame_padding_start_);
split_frame_len_ = static_cast<int>(slen.Read(15));
}
if (frame_count > 0) {
// assert_true(xma::GetPacketFrameOffset(packet) - 32 ==
// split_frame_len_ - split_frame_len_partial_);
}
auto offset = stream.Copy(
xma_frame_.data() + 1 +
((split_frame_len_partial_ + split_frame_padding_start_) / 8),
split_frame_len_ - split_frame_len_partial_);
assert_true(offset ==
(split_frame_padding_start_ + split_frame_len_partial_) % 8);
} else {
if (data->input_buffer_read_offset % kBitsPerPacket == 0) {
// Invalid offset. Go ahead and set it.
int packet_number =
GetFramePacketNumber(current_input_buffer, current_input_size,
data->input_buffer_read_offset);
if (packet_number == -1) {
return;
}
auto offset =
xma::GetPacketFrameOffset(current_input_buffer +
kBytesPerPacket * packet_number) +
data->input_buffer_read_offset;
if (offset == -1) {
// No more frames.
SwapInputBuffer(data);
// TODO partial frames? end?
XELOGE("XmaContext {}: TODO partial frames? end?", id());
assert_always("TODO");
return;
} else {
data->input_buffer_read_offset = offset;
}
}
if (!ValidFrameOffset(current_input_buffer, current_input_size,
data->input_buffer_read_offset)) {
XELOGAPU("XmaContext {}: Error - Invalid read offset {}!", id(),
data->input_buffer_read_offset);
SwapInputBuffer(data);
return;
}
// Where are we in the buffer (in XMA jargon)
std::tie(packet_idx, frame_idx) =
GetFrameNumber(current_input_buffer, current_input_size,
data->input_buffer_read_offset);
// TODO handle
assert_true(packet_idx >= 0);
assert_true(frame_idx >= 0);
packet = current_input_buffer + packet_idx * kBytesPerPacket;
// frames that belong to this packet
std::tie(frame_count, frame_last_split) = GetPacketFrameCount(packet);
assert_true(frame_count >= 0); // TODO end
PrepareDecoder(packet, data->sample_rate, bool(data->is_stereo));
// Current frame is split to next packet:
bool frame_is_split = frame_last_split && (frame_idx >= frame_count - 1);
stream =
BitStream(current_input_buffer, (packet_idx + 1) * kBitsPerPacket);
stream.SetOffset(data->input_buffer_read_offset);
// int frame_len;
// int frame_len_partial
split_frame_len_partial_ = static_cast<int>(stream.BitsRemaining());
if (split_frame_len_partial_ >= 15) {
split_frame_len_ = static_cast<int>(stream.Peek(15));
} else {
// assert_always();
split_frame_len_ = xma::kMaxFrameLength + 1;
}
assert_true(frame_is_split ==
(split_frame_len_ > split_frame_len_partial_));
// TODO fix bitstream copy
std::memset(xma_frame_.data(), 0, xma_frame_.size());
{
int32_t bits_to_copy =
std::min(split_frame_len_, split_frame_len_partial_);
if (!stream.IsOffsetValid(bits_to_copy)) {
XELOGAPU(
"XmaContext {}: Error - Invalid amount of bits to copy! "
"split_frame_len: {}, split_partial: {}, offset_bits: {}",
id(), split_frame_len_, split_frame_len_partial_,
stream.offset_bits());
SwapInputBuffer(data);
return;
}
auto offset = stream.Copy(xma_frame_.data() + 1, bits_to_copy);
assert_true(offset < 8);
split_frame_padding_start_ = static_cast<uint8_t>(offset);
}
if (frame_is_split) {
// go to next xma packet of this stream
packets_skip_ = xma::GetPacketSkipCount(packet) + 1;
while (packets_skip_ > 0) {
packets_skip_--;
packet += kBytesPerPacket;
packet_idx++;
if (packet_idx >= current_input_packet_count) {
if (!reuse_input_buffer) {
// Last packet. Try setup once more.
reuse_input_buffer = TrySetupNextLoop(data, true);
}
if (!reuse_input_buffer) {
if (is_streaming) {
SwapInputBuffer(data);
} else {
is_stream_done_ = true;
}
}
return;
}
}
// TODO guest might read this:
data->input_buffer_read_offset = packet_idx * kBitsPerPacket;
continue;
}
}
av_packet_->data = xma_frame_.data();
av_packet_->size = static_cast<int>(
1 + ((split_frame_padding_start_ + split_frame_len_) / 8) +
(((split_frame_padding_start_ + split_frame_len_) % 8) ? 1 : 0));
auto padding_end = av_packet_->size * 8 -
(8 + split_frame_padding_start_ + split_frame_len_);
assert_true(padding_end < 8);
xma_frame_[0] =
((split_frame_padding_start_ & 7) << 5) | ((padding_end & 7) << 2);
split_frame_len_ = 0;
split_frame_len_partial_ = 0;
split_frame_padding_start_ = 0;
auto ret = avcodec_send_packet(av_context_, av_packet_);
if (ret < 0) {
XELOGE("XmaContext {}: Error - Sending packet for decoding failed", id());
// TODO bail out
assert_always();
}
ret = avcodec_receive_frame(av_context_, av_frame_);
/*
if (ret == AVERROR(EAGAIN) || ret == AVERROR_EOF)
// TODO AVERROR_EOF???
break;
else
*/
if (ret < 0) {
XELOGE("XmaContext {}: Error - Decoding failed", id());
data->parser_error_status = 4; // TODO(Gliniak): Find all parsing errors
// and create enumerator from them
SwapInputBuffer(data);
assert_always();
return; // TODO bail out
}
assert_true(ret == 0);
{
// copy over 1 frame
// update input buffer read offset
// assert(decoded_consumed_samples_ + kSamplesPerFrame <=
// current_frame_.size());
assert_true(av_context_->sample_fmt == AV_SAMPLE_FMT_FLTP);
// assert_true(frame_is_split == (frame_idx == -1));
// dump_raw(av_frame_, id());
ConvertFrame(reinterpret_cast<const uint8_t**>(&av_frame_->data),
bool(av_frame_->channels > 1), raw_frame_.data());
// decoded_consumed_samples_ += kSamplesPerFrame;
auto byte_count = kBytesPerFrameChannel << data->is_stereo;
assert_true(output_remaining_bytes >= byte_count);
output_rb.Write(raw_frame_.data(), byte_count);
output_remaining_bytes -= byte_count;
data->output_buffer_write_offset = output_rb.write_offset() / 256;
total_samples += id_ == 0 ? kSamplesPerFrame : 0;
uint32_t offset =
std::max(kBitsPerHeader, data->input_buffer_read_offset);
offset = static_cast<uint32_t>(
GetNextFrame(current_input_buffer, current_input_size, offset));
XELOGAPU(
"XmaContext {}: Next Offset: {} (Frame: {}/{} Packet: {}/{} Packet "
"Skip: {} - {})",
id(), offset, frame_idx, frame_count - 1, packet_idx,
current_input_packet_count, xma::GetPacketSkipCount(packet),
data->input_buffer_read_offset);
if (frame_idx + 1 >= frame_count) {
// Skip to next packet (no split frame)
packets_skip_ = xma::GetPacketSkipCount(packet) + 1;
while (packets_skip_ > 0) {
packets_skip_--;
packet_idx++;
if (packet_idx >= current_input_packet_count) {
if (!reuse_input_buffer) {
// Last packet. Try setup once more.
reuse_input_buffer = TrySetupNextLoop(data, true);
}
if (!reuse_input_buffer) {
if (is_streaming) {
SwapInputBuffer(data);
data->input_buffer_read_offset =
GetPacketFirstFrameOffset(data);
} else {
is_stream_done_ = true;
}
if (output_rb.write_offset() == output_rb.read_offset()) {
data->output_buffer_valid = 0;
}
}
return;
}
}
packet = current_input_buffer + packet_idx * kBytesPerPacket;
// TODO(Gliniak): There might be an edge-case when we're in packet 26/27
// and GetPacketFrameOffset returns that there is no data in this packet
// aka. FrameOffset is set to more than 0x7FFF-0x20
offset =
xma::GetPacketFrameOffset(packet) + packet_idx * kBitsPerPacket;
}
if (offset == 0 || frame_idx == -1) {
// Next packet but we already skipped to it
if (packet_idx >= current_input_packet_count) {
// Buffer is fully used
if (!reuse_input_buffer) {
// Last packet. Try setup once more.
reuse_input_buffer = TrySetupNextLoop(data, true);
}
if (!reuse_input_buffer) {
if (is_streaming) {
SwapInputBuffer(data);
} else {
is_stream_done_ = true;
}
}
break;
}
offset =
xma::GetPacketFrameOffset(packet) + packet_idx * kBitsPerPacket;
}
// TODO buffer bounds check
assert_true(data->input_buffer_read_offset < offset);
data->input_buffer_read_offset = offset;
}
}
// assert_true((split_frame_len_ != 0) == (data->input_buffer_read_offset ==
// 0));
// The game will kick us again with a new output buffer later.
// It's important that we only invalidate this if we actually wrote to it!!
if (output_rb.write_offset() == output_rb.read_offset()) {
data->output_buffer_valid = 0;
}
}
uint32_t XmaContextOld::GetPacketFirstFrameOffset(
const XMA_CONTEXT_DATA* data) {
uint32_t first_frame_offset = kBitsPerHeader;
uint8_t* in0 = data->input_buffer_0_valid
? memory()->TranslatePhysical(data->input_buffer_0_ptr)
: nullptr;
uint8_t* in1 = data->input_buffer_1_valid
? memory()->TranslatePhysical(data->input_buffer_1_ptr)
: nullptr;
uint8_t* current_input_buffer = data->current_buffer ? in1 : in0;
if (current_input_buffer) {
first_frame_offset = xma::GetPacketFrameOffset(current_input_buffer);
}
return first_frame_offset;
}
size_t XmaContextOld::GetNextFrame(uint8_t* block, size_t size,
size_t bit_offset) {
// offset = xma::GetPacketFrameOffset(packet);
// TODO meh
// auto next_packet = bit_offset - bit_offset % kBitsPerPacket +
// kBitsPerPacket;
auto packet_idx = GetFramePacketNumber(block, size, bit_offset);
BitStream stream(block, size * 8);
stream.SetOffset(bit_offset);
if (stream.BitsRemaining() < 15) {
return 0;
}
uint64_t len = stream.Read(15);
if ((len - 15) > stream.BitsRemaining()) {
// assert_always("TODO");
// *bit_offset = next_packet;
// return false;
// return next_packet;
return 0;
} else if (len >= xma::kMaxFrameLength) {
assert_always("TODO");
// *bit_offset = next_packet;
// return false;
return 0;
// return next_packet;
}
stream.Advance(len - (15 + 1));
// Read the trailing bit to see if frames follow
if (stream.Read(1) == 0) {
return 0;
}
bit_offset += len;
if (packet_idx < GetFramePacketNumber(block, size, bit_offset)) {
return 0;
}
return bit_offset;
}
int XmaContextOld::GetFramePacketNumber(uint8_t* block, size_t size,
size_t bit_offset) {
size *= 8;
if (bit_offset >= size) {
// Not good :(
assert_always();
return -1;
}
size_t byte_offset = bit_offset >> 3;
size_t packet_number = byte_offset / kBytesPerPacket;
return (uint32_t)packet_number;
}
std::tuple<int, int> XmaContextOld::GetFrameNumber(uint8_t* block, size_t size,
size_t bit_offset) {
auto packet_idx = GetFramePacketNumber(block, size, bit_offset);
if (packet_idx < 0 || (packet_idx + 1) * kBytesPerPacket > size) {
assert_always();
return {packet_idx, -2};
}
if (bit_offset == 0) {
return {packet_idx, -1};
}
uint8_t* packet = block + (packet_idx * kBytesPerPacket);
auto first_frame_offset = xma::GetPacketFrameOffset(packet);
BitStream stream(block, size * 8);
stream.SetOffset(packet_idx * kBitsPerPacket + first_frame_offset);
int frame_idx = 0;
while (true) {
if (stream.BitsRemaining() < 15) {
break;
}
if (stream.offset_bits() == bit_offset) {
break;
}
uint64_t size = stream.Read(15);
if ((size - 15) > stream.BitsRemaining()) {
// Last frame.
break;
} else if (size == 0x7FFF) {
// Invalid frame (and last of this packet)
break;
}
stream.Advance(size - (15 + 1));
// Read the trailing bit to see if frames follow
if (stream.Read(1) == 0) {
break;
}
frame_idx++;
}
return {packet_idx, frame_idx};
}
std::tuple<int, bool> XmaContextOld::GetPacketFrameCount(uint8_t* packet) {
auto first_frame_offset = xma::GetPacketFrameOffset(packet);
if (first_frame_offset > kBitsPerPacket - kBitsPerHeader) {
// frame offset is beyond packet end
return {0, false};
}
BitStream stream(packet, kBitsPerPacket);
stream.SetOffset(first_frame_offset);
int frame_count = 0;
while (true) {
if (stream.BitsRemaining() < 15) {
return {frame_count, false};
}
frame_count++;
uint64_t size = stream.Read(15);
if ((size - 15) > stream.BitsRemaining()) {
return {frame_count, true};
} else if (size == 0x7FFF) {
assert_always();
return {frame_count, true};
}
stream.Advance(size - (15 + 1));
if (stream.Read(1) == 0) {
return {frame_count, false};
}
// There is a case when frame ends EXACTLY at the end of packet.
// In such case we shouldn't increase frame count by additional not existing
// frame and don't mark it as splitted, but as a normal frame
if (!stream.BitsRemaining()) {
return {frame_count, false};
}
}
}
int XmaContextOld::PrepareDecoder(uint8_t* packet, int sample_rate,
bool is_two_channel) {
// Sanity check: Packet metadata is always 1 for XMA2/0 for XMA
assert_true((packet[2] & 0x7) == 1 || (packet[2] & 0x7) == 0);
sample_rate = GetSampleRate(sample_rate);
// Re-initialize the context with new sample rate and channels.
uint32_t channels = is_two_channel ? 2 : 1;
if (av_context_->sample_rate != sample_rate ||
av_context_->channels != channels) {
// We have to reopen the codec so it'll realloc whatever data it needs.
// TODO(DrChat): Find a better way.
avcodec_close(av_context_);
av_context_->sample_rate = sample_rate;
av_context_->channels = channels;
if (avcodec_open2(av_context_, av_codec_, NULL) < 0) {
XELOGE("XmaContext: Failed to reopen FFmpeg context");
return -1;
}
return 1;
}
return 0;
}
} // namespace apu
} // namespace xe

101
src/xenia/apu/xma_context_old.h Normal file
View File

@ -0,0 +1,101 @@
/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2024 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#ifndef XENIA_APU_XMA_CONTEXT_OLD_H_
#define XENIA_APU_XMA_CONTEXT_OLD_H_
#include <array>
#include <atomic>
#include <mutex>
#include <queue>
#include "xenia/apu/xma_context.h"
#include "xenia/memory.h"
#include "xenia/xbox.h"
// Forward declarations
struct AVCodec;
struct AVCodecParserContext;
struct AVCodecContext;
struct AVFrame;
struct AVPacket;
namespace xe {
namespace apu {
class XmaContextOld : public XmaContext {
public:
explicit XmaContextOld();
~XmaContextOld();
int Setup(uint32_t id, Memory* memory, uint32_t guest_ptr);
bool Work();
void Enable();
bool Block(bool poll);
void Clear();
void Disable();
void Release();
private:
static void SwapInputBuffer(XMA_CONTEXT_DATA* data);
static bool TrySetupNextLoop(XMA_CONTEXT_DATA* data,
bool ignore_input_buffer_offset);
static void NextPacket(XMA_CONTEXT_DATA* data);
static int GetSampleRate(int id);
// Get the offset of the next frame. Does not traverse packets.
static size_t GetNextFrame(uint8_t* block, size_t size, size_t bit_offset);
// Get the containing packet number of the frame pointed to by the offset.
static int GetFramePacketNumber(uint8_t* block, size_t size,
size_t bit_offset);
// Get the packet number and the index of the frame inside that packet
static std::tuple<int, int> GetFrameNumber(uint8_t* block, size_t size,
size_t bit_offset);
// Get the number of frames contained in the packet (including truncated) and
// if the last frame is split.
static std::tuple<int, bool> GetPacketFrameCount(uint8_t* packet);
bool ValidFrameOffset(uint8_t* block, size_t size_bytes,
size_t frame_offset_bits);
void Decode(XMA_CONTEXT_DATA* data);
int PrepareDecoder(uint8_t* packet, int sample_rate, bool is_two_channel);
// This method should be used ONLY when we're at the last packet of the stream
// and we want to find offset in next buffer
uint32_t GetPacketFirstFrameOffset(const XMA_CONTEXT_DATA* data);
// uint32_t decoded_consumed_samples_ = 0; // TODO do this dynamically
// int decoded_idx_ = -1;
// bool partial_frame_saved_ = false;
// bool partial_frame_size_known_ = false;
// size_t partial_frame_total_size_bits_ = 0;
// size_t partial_frame_start_offset_bits_ = 0;
// size_t partial_frame_offset_bits_ = 0; // blah internal don't use this
// std::vector<uint8_t> partial_frame_buffer_;
uint32_t packets_skip_ = 0;
bool is_stream_done_ = false;
// bool split_frame_pending_ = false;
uint32_t split_frame_len_ = 0;
uint32_t split_frame_len_partial_ = 0;
uint8_t split_frame_padding_start_ = 0;
// first byte contains bit offset information
std::array<uint8_t, 1 + 4096> xma_frame_;
// uint8_t* current_frame_ = nullptr;
// conversion buffer for 2 channel frame
std::array<uint8_t, kBytesPerFrameChannel * 2> raw_frame_;
// std::vector<uint8_t> current_frame_ = std::vector<uint8_t>(0);
};
} // namespace apu
} // namespace xe
#endif // XENIA_APU_XMA_CONTEXT_H_

42
src/xenia/apu/xma_decoder.cc
View File

@ -10,6 +10,9 @@
#include "xenia/apu/xma_decoder.h"
#include "xenia/apu/xma_context.h"
#include "xenia/apu/xma_context_new.h"
#include "xenia/apu/xma_context_old.h"
#include "xenia/base/cvar.h"
#include "xenia/base/logging.h"
#include "xenia/base/math.h"
@ -51,6 +54,14 @@ extern "C" {
DEFINE_bool(ffmpeg_verbose, false, "Verbose FFmpeg output (debug and above)",
"APU");
DEFINE_bool(use_new_decoder, false,
"Enables usage of new experimental XMA audio decoder.", "APU");
DEFINE_bool(use_dedicated_xma_thread, true,
"Enables XMA decoding on separate thread. Disabled should produce "
"better results, but decrease performance a bit.",
"APU");
namespace xe {
namespace apu {
@ -128,9 +139,14 @@ X_STATUS XmaDecoder::Setup(kernel::KernelState* kernel_state) {
// Setup XMA contexts.
for (int i = 0; i < kContextCount; ++i) {
if (cvars::use_new_decoder) {
contexts_[i] = new XmaContextNew();
} else {
contexts_[i] = new XmaContextOld();
}
uint32_t guest_ptr = context_data_first_ptr_ + i * sizeof(XMA_CONTEXT_DATA);
XmaContext& context = contexts_[i];
if (context.Setup(i, memory(), guest_ptr)) {
if (contexts_[i]->Setup(i, memory(), guest_ptr)) {
assert_always();
}
}
@ -144,7 +160,9 @@ X_STATUS XmaDecoder::Setup(kernel::KernelState* kernel_state) {
kernel::object_ref<kernel::XHostThread>(new kernel::XHostThread(
kernel_state, 128 * 1024, 0,
[this]() {
WorkerThreadMain();
if (cvars::use_dedicated_xma_thread) {
WorkerThreadMain();
}
return 0;
},
kernel_state
@ -163,8 +181,7 @@ void XmaDecoder::WorkerThreadMain() {
// Okay, let's loop through XMA contexts to find ones we need to decode!
bool did_work = false;
for (uint32_t n = 0; n < kContextCount; n++) {
XmaContext& context = contexts_[n];
did_work = context.Work() || did_work;
did_work = contexts_[n]->Work() || did_work;
// TODO: Need thread safety to do this.
// Probably not too important though.
@ -228,7 +245,7 @@ uint32_t XmaDecoder::AllocateContext() {
return 0;
}
XmaContext& context = contexts_[index];
XmaContext& context = *contexts_[index];
assert_false(context.is_allocated());
context.set_is_allocated(true);
return context.guest_ptr();
@ -238,7 +255,7 @@ void XmaDecoder::ReleaseContext(uint32_t guest_ptr) {
auto context_id = GetContextId(guest_ptr);
assert_true(context_id >= 0);
XmaContext& context = contexts_[context_id];
XmaContext& context = *contexts_[context_id];
assert_true(context.is_allocated());
context.Release();
context_bitmap_.Release(context_id);
@ -248,7 +265,7 @@ bool XmaDecoder::BlockOnContext(uint32_t guest_ptr, bool poll) {
auto context_id = GetContextId(guest_ptr);
assert_true(context_id >= 0);
XmaContext& context = contexts_[context_id];
XmaContext& context = *contexts_[context_id];
return context.Block(poll);
}
@ -309,8 +326,11 @@ void XmaDecoder::WriteRegister(uint32_t addr, uint32_t value) {
for (int i = 0; value && i < 32; ++i, value >>= 1) {
if (value & 1) {
uint32_t context_id = base_context_id + i;
auto& context = contexts_[context_id];
auto& context = *contexts_[context_id];
context.Enable();
if (!cvars::use_dedicated_xma_thread) {
context.Work();
}
}
}
// Signal the decoder thread to start processing.
@ -323,7 +343,7 @@ void XmaDecoder::WriteRegister(uint32_t addr, uint32_t value) {
for (int i = 0; value && i < 32; ++i, value >>= 1) {
if (value & 1) {
uint32_t context_id = base_context_id + i;
auto& context = contexts_[context_id];
auto& context = *contexts_[context_id];
context.Disable();
}
}
@ -337,7 +357,7 @@ void XmaDecoder::WriteRegister(uint32_t addr, uint32_t value) {
for (int i = 0; value && i < 32; ++i, value >>= 1) {
if (value & 1) {
uint32_t context_id = base_context_id + i;
XmaContext& context = contexts_[context_id];
XmaContext& context = *contexts_[context_id];
context.Clear();
}
}

2
src/xenia/apu/xma_decoder.h
View File

@ -81,7 +81,7 @@ class XmaDecoder {
XmaRegisterFile register_file_;
static const uint32_t kContextCount = 320;
XmaContext contexts_[kContextCount];
XmaContext* contexts_[kContextCount];
BitMap context_bitmap_;
uint32_t context_data_first_ptr_ = 0;

36
src/xenia/apu/xma_helpers.h
View File

@ -2,7 +2,7 @@
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2021 Ben Vanik. All rights reserved. *
* Copyright 2023 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
@ -20,31 +20,33 @@ namespace xma {
static const uint32_t kMaxFrameLength = 0x7FFF;
// Get number of frames that /begin/ in this packet.
uint32_t GetPacketFrameCount(uint8_t* packet) {
return (uint8_t)(packet[0] >> 2);
// Get number of frames that /begin/ in this packet. This is valid only for XMA2
// packets
static const uint8_t GetPacketFrameCount(const uint8_t* packet) {
return packet[0] >> 2;
}
static const uint8_t GetPacketMetadata(const uint8_t* packet) {
return packet[2] & 0x7;
}
static const bool IsPacketXma2Type(const uint8_t* packet) {
return GetPacketMetadata(packet) == 1;
}
static const uint8_t GetPacketSkipCount(const uint8_t* packet) {
return packet[3];
}
// Get the first frame offset in bits
uint32_t GetPacketFrameOffset(uint8_t* packet) {
static uint32_t GetPacketFrameOffset(const uint8_t* packet) {
uint32_t val = (uint16_t)(((packet[0] & 0x3) << 13) | (packet[1] << 5) |
(packet[2] >> 3));
// if (val > kBitsPerPacket - kBitsPerHeader) {
// // There is no data in this packet
// return -1;
// } else {
return val + 32;
// }
}
uint32_t GetPacketMetadata(uint8_t* packet) {
return (uint8_t)(packet[2] & 0x7);
}
uint32_t GetPacketSkipCount(uint8_t* packet) { return (uint8_t)(packet[3]); }
} // namespace xma
} // namespace apu
} // namespace xe
#endif // XENIA_APU_XMA_HELPERS_H_
#endif // XENIA_APU_XMA_HELPERS_H_