miguelr/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

video_core: implement 1D copies based on VMM 'kind'

Browse source
This commit is contained in:
FengChen 2022-10-14 22:20:45 +08:00
parent 99507d0188
commit 23b6569fc2
2 changed files with 73 additions and 56 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

83
src/video_core/engines/maxwell_dma.cpp
View file

@ -56,6 +56,7 @@ void MaxwellDMA::Launch() {
ASSERT(launch.interrupt_type == LaunchDMA::InterruptType::NONE);
ASSERT(launch.data_transfer_type == LaunchDMA::DataTransferType::NON_PIPELINED);
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;
@ -66,56 +67,74 @@ void MaxwellDMA::Launch() {
}
if (is_src_pitch && is_dst_pitch) {
CopyPitchToPitch();
for (u32 line = 0; line < regs.line_count; ++line) {
const GPUVAddr source_line =
regs.offset_in + static_cast<size_t>(line) * regs.pitch_in;
const GPUVAddr dest_line =
regs.offset_out + static_cast<size_t>(line) * regs.pitch_out;
memory_manager.CopyBlock(dest_line, source_line, regs.line_length_in);
}
} else {
ASSERT(launch.multi_line_enable == 1);
if (!is_src_pitch && is_dst_pitch) {
CopyBlockLinearToPitch();
} else {
CopyPitchToBlockLinear();
}
}
ReleaseSemaphore();
}
void MaxwellDMA::CopyPitchToPitch() {
// When `multi_line_enable` bit is enabled we copy a 2D image of dimensions
// (line_length_in, line_count).
// Otherwise the copy is performed as if we were copying a 1D buffer of length line_length_in.
const bool remap_enabled = regs.launch_dma.remap_enable != 0;
if (regs.launch_dma.multi_line_enable) {
UNIMPLEMENTED_IF(remap_enabled);
// Perform a line-by-line copy.
// We're going to take a subrect of size (line_length_in, line_count) from the source
// rectangle. There is no need to manually flush/invalidate the regions because CopyBlock
// does that for us.
for (u32 line = 0; line < regs.line_count; ++line) {
const GPUVAddr source_line = regs.offset_in + static_cast<size_t>(line) * regs.pitch_in;
const GPUVAddr dest_line = regs.offset_out + static_cast<size_t>(line) * regs.pitch_out;
memory_manager.CopyBlock(dest_line, source_line, regs.line_length_in);
}
return;
}
} else {
// TODO: allow multisized components.
auto& accelerate = rasterizer->AccessAccelerateDMA();
const bool is_const_a_dst = regs.remap_const.dst_x == RemapConst::Swizzle::CONST_A;
const bool is_buffer_clear = remap_enabled && is_const_a_dst;
if (is_buffer_clear) {
if (regs.launch_dma.remap_enable != 0 && is_const_a_dst) {
ASSERT(regs.remap_const.component_size_minus_one == 3);
accelerate.BufferClear(regs.offset_out, regs.line_length_in, regs.remap_consta_value);
std::vector<u32> tmp_buffer(regs.line_length_in, regs.remap_consta_value);
memory_manager.WriteBlockUnsafe(regs.offset_out, reinterpret_cast<u8*>(tmp_buffer.data()),
memory_manager.WriteBlockUnsafe(regs.offset_out,
reinterpret_cast<u8*>(tmp_buffer.data()),
regs.line_length_in * sizeof(u32));
return;
} else {
auto convert_linear_2_blocklinear_addr = [](u64 address) {
return (address & ~0x1f0ULL) | ((address & 0x40) >> 2) | ((address & 0x10) << 1) |
((address & 0x180) >> 1) | ((address & 0x20) << 3);
};
auto src_kind = memory_manager.GetPageKind(regs.offset_in);
auto dst_kind = memory_manager.GetPageKind(regs.offset_out);
const bool is_src_pitch = IsPitchKind(static_cast<PTEKind>(src_kind));
const bool is_dst_pitch = IsPitchKind(static_cast<PTEKind>(dst_kind));
if (!is_src_pitch && is_dst_pitch) {
std::vector<u8> tmp_buffer(regs.line_length_in);
std::vector<u8> dst_buffer(regs.line_length_in);
memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(),
regs.line_length_in);
for (u32 offset = 0; offset < regs.line_length_in; ++offset) {
dst_buffer[offset] =
tmp_buffer[convert_linear_2_blocklinear_addr(regs.offset_in + offset) -
regs.offset_in];
}
UNIMPLEMENTED_IF(remap_enabled);
memory_manager.WriteBlock(regs.offset_out, dst_buffer.data(), regs.line_length_in);
} else if (is_src_pitch && !is_dst_pitch) {
std::vector<u8> tmp_buffer(regs.line_length_in);
std::vector<u8> dst_buffer(regs.line_length_in);
memory_manager.ReadBlockUnsafe(regs.offset_in, tmp_buffer.data(),
regs.line_length_in);
for (u32 offset = 0; offset < regs.line_length_in; ++offset) {
dst_buffer[convert_linear_2_blocklinear_addr(regs.offset_out + offset) -
regs.offset_out] = tmp_buffer[offset];
}
memory_manager.WriteBlock(regs.offset_out, dst_buffer.data(), regs.line_length_in);
} else {
if (!accelerate.BufferCopy(regs.offset_in, regs.offset_out, regs.line_length_in)) {
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(), 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(),
regs.line_length_in);
}
}
}
}
ReleaseSemaphore();
}
void MaxwellDMA::CopyBlockLinearToPitch() {

2
src/video_core/engines/maxwell_dma.h
View file

@ -219,8 +219,6 @@ private:
/// registers.
void Launch();
void CopyPitchToPitch();
void CopyBlockLinearToPitch();
void CopyPitchToBlockLinear();