INSANEMODE/yuzu-android
1
0
Fork 0
mirror of https://github.com/yuzu-emu/yuzu-android.git synced 2025-06-11 17:57:58 -05:00
Code Issues Packages Projects Releases Wiki Activity

video_core: Cache GPU internal writes.

Browse Source
This commit is contained in:
Fernando Sahmkow
2023-01-04 22:05:20 -05:00
parent b78328f19a
commit 6c7eb81f7d
10 changed files with 185 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/video_core/engines/engine_upload.cpp
View File

@ -76,7 +76,7 @@ void State::ProcessData(std::span<const u8> read_buffer) {
regs.dest.height, regs.dest.depth, x_offset, regs.dest.y,
x_elements, regs.line_count, regs.dest.BlockHeight(),
regs.dest.BlockDepth(), regs.line_length_in);
memory_manager.WriteBlock(address, tmp_buffer.data(), dst_size);
memory_manager.WriteBlockCached(address, tmp_buffer.data(), dst_size);
}
}

7
src/video_core/engines/maxwell_3d.cpp
View File

@ -485,11 +485,6 @@ void Maxwell3D::StampQueryResult(u64 payload, bool long_query) {
}
void Maxwell3D::ProcessQueryGet() {
// TODO(Subv): Support the other query units.
if (regs.report_semaphore.query.location != Regs::ReportSemaphore::Location::All) {
LOG_DEBUG(HW_GPU, "Locations other than ALL are unimplemented");
}
switch (regs.report_semaphore.query.operation) {
case Regs::ReportSemaphore::Operation::Release:
if (regs.report_semaphore.query.short_query != 0) {
@ -649,7 +644,7 @@ void Maxwell3D::ProcessCBMultiData(const u32* start_base, u32 amount) {
const GPUVAddr address{buffer_address + regs.const_buffer.offset};
const size_t copy_size = amount * sizeof(u32);
memory_manager.WriteBlock(address, start_base, copy_size);
memory_manager.WriteBlockCached(address, start_base, copy_size);
// Increment the current buffer position.
regs.const_buffer.offset += static_cast<u32>(copy_size);

17
src/video_core/engines/maxwell_dma.cpp
View File

@ -69,7 +69,7 @@ void MaxwellDMA::Launch() {
if (launch.multi_line_enable) {
const bool is_src_pitch = launch.src_memory_layout == LaunchDMA::MemoryLayout::PITCH;
const bool is_dst_pitch = launch.dst_memory_layout == LaunchDMA::MemoryLayout::PITCH;
memory_manager.FlushCaching();
if (!is_src_pitch && !is_dst_pitch) {
// If both the source and the destination are in block layout, assert.
CopyBlockLinearToBlockLinear();
@ -104,6 +104,7 @@ void MaxwellDMA::Launch() {
reinterpret_cast<u8*>(tmp_buffer.data()),
regs.line_length_in * sizeof(u32));
} else {
memory_manager.FlushCaching();
const auto convert_linear_2_blocklinear_addr = [](u64 address) {
return (address & ~0x1f0ULL) | ((address & 0x40) >> 2) | ((address & 0x10) << 1) |
((address & 0x180) >> 1) | ((address & 0x20) << 3);
@ -121,7 +122,7 @@ void MaxwellDMA::Launch() {
memory_manager.ReadBlockUnsafe(
convert_linear_2_blocklinear_addr(regs.offset_in + offset),
tmp_buffer.data(), tmp_buffer.size());
memory_manager.WriteBlock(regs.offset_out + offset, tmp_buffer.data(),
memory_manager.WriteBlockCached(regs.offset_out + offset, tmp_buffer.data(),
tmp_buffer.size());
}
} else if (is_src_pitch && !is_dst_pitch) {
@ -132,7 +133,7 @@ void MaxwellDMA::Launch() {
for (u32 offset = 0; offset < regs.line_length_in; offset += 16) {
memory_manager.ReadBlockUnsafe(regs.offset_in + offset, tmp_buffer.data(),
tmp_buffer.size());
memory_manager.WriteBlock(
memory_manager.WriteBlockCached(
convert_linear_2_blocklinear_addr(regs.offset_out + offset),
tmp_buffer.data(), tmp_buffer.size());
}
@ -141,7 +142,7 @@ void MaxwellDMA::Launch() {
std::vector<u8> tmp_buffer(regs.line_length_in);
memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(),
regs.line_length_in);
memory_manager.WriteBlock(regs.offset_out, tmp_buffer.data(),
memory_manager.WriteBlockCached(regs.offset_out, tmp_buffer.data(),
regs.line_length_in);
}
}
@ -204,7 +205,7 @@ void MaxwellDMA::CopyBlockLinearToPitch() {
src_params.origin.y, x_elements, regs.line_count, block_height, block_depth,
regs.pitch_out);
memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
memory_manager.WriteBlockCached(regs.offset_out, write_buffer.data(), dst_size);
}
void MaxwellDMA::CopyPitchToBlockLinear() {
@ -256,7 +257,7 @@ void MaxwellDMA::CopyPitchToBlockLinear() {
dst_params.origin.y, x_elements, regs.line_count, block_height, block_depth,
regs.pitch_in);
memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
memory_manager.WriteBlockCached(regs.offset_out, write_buffer.data(), dst_size);
}
void MaxwellDMA::FastCopyBlockLinearToPitch() {
@ -287,7 +288,7 @@ void MaxwellDMA::FastCopyBlockLinearToPitch() {
regs.src_params.block_size.height, regs.src_params.block_size.depth,
regs.pitch_out);
memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
memory_manager.WriteBlockCached(regs.offset_out, write_buffer.data(), dst_size);
}
void MaxwellDMA::CopyBlockLinearToBlockLinear() {
@ -347,7 +348,7 @@ void MaxwellDMA::CopyBlockLinearToBlockLinear() {
dst.depth, dst_x_offset, dst.origin.y, x_elements, regs.line_count,
dst.block_size.height, dst.block_size.depth, pitch);
memory_manager.WriteBlock(regs.offset_out, write_buffer.data(), dst_size);
memory_manager.WriteBlockCached(regs.offset_out, write_buffer.data(), dst_size);
}
void MaxwellDMA::ReleaseSemaphore() {