INSANEMODE/yuzu-android
1
0
Fork 0
mirror of https://github.com/yuzu-emu/yuzu-android.git synced 2025-06-11 09:37:57 -05:00
Code Issues Packages Projects Releases Wiki Activity

MaxwellDMA: Optimize micro copies.

Browse Source
This commit is contained in:
Fernando Sahmkow
2020-04-27 23:57:24 -04:00
parent b87422a86f
commit 37c690576f
3 changed files with 57 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
src/video_core/engines/maxwell_dma.cpp
View File

@ -90,7 +90,47 @@ void MaxwellDMA::HandleCopy() {
ASSERT(regs.exec.enable_2d == 1);
if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
ASSERT(regs.src_params.BlockDepth() == 0);
// Optimized path for micro copies.
if (regs.dst_pitch * regs.y_count < Texture::GetGOBSize() && regs.dst_pitch <= 64) {
const u32 bytes_per_pixel = regs.dst_pitch / regs.x_count;
const std::size_t src_size = Texture::GetGOBSize();
const std::size_t dst_size = regs.dst_pitch * regs.y_count;
u32 pos_x = regs.src_params.pos_x;
u32 pos_y = regs.src_params.pos_y;
const u64 offset =
Texture::GetGOBOffset(regs.src_params.size_x, regs.src_params.size_y, pos_x, pos_y,
regs.src_params.BlockDepth(), bytes_per_pixel);
const u32 x_in_gob = 64 / bytes_per_pixel;
pos_x = pos_x % x_in_gob;
pos_y = pos_y % 8;
if (read_buffer.size() < src_size) {
read_buffer.resize(src_size);
}
if (write_buffer.size() < dst_size) {
write_buffer.resize(dst_size);
}
if (Settings::IsGPULevelExtreme()) {
memory_manager.ReadBlock(source + offset, read_buffer.data(), src_size);
memory_manager.ReadBlock(dest, write_buffer.data(), dst_size);
} else {
memory_manager.ReadBlockUnsafe(source + offset, read_buffer.data(), src_size);
memory_manager.ReadBlockUnsafe(dest, write_buffer.data(), dst_size);
}
Texture::UnswizzleSubrect(regs.x_count, regs.y_count, regs.dst_pitch,
regs.src_params.size_x, bytes_per_pixel, read_buffer.data(),
write_buffer.data(), regs.src_params.BlockHeight(), pos_x,
pos_y);
memory_manager.WriteBlock(dest, write_buffer.data(), dst_size);
return;
}
// If the input is tiled and the output is linear, deswizzle the input and copy it over.
const u32 bytes_per_pixel = regs.dst_pitch / regs.x_count;
const std::size_t src_size = Texture::CalculateSize(