RPCS3/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2024-11-22 18:53:28 +01:00
Code Issues Releases Wiki Activity

gl: Optimize image operations

Browse Source 
- Avoid double transfers where a transfer to a temp image is done
without scaling and then a secondary transfer follows. Combines the two
steps into one whenever possible which can significantly alleviate
bandwidth problems at higher resolutions. Significant speedup, upto 90%
in some cases (PDF, PDF2)
This commit is contained in:
kd-11 2020-01-16 17:24:04 +03:00 committed by kd-11
parent 47b196e9d0
commit 82af17beb1
2 changed files with 40 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
rpcs3/Emu/RSX/GL/GLHelpers.cpp
View File

@ -382,6 +382,35 @@ namespace gl
const gl::texture* real_src = src;
const gl::texture* real_dst = dst;
// Optimization pass; check for pass-through data transfer
if (!xfer_info.flip_horizontal && !xfer_info.flip_vertical && src_rect.height() == dst_rect.height())
{
auto src_w = src_rect.width();
auto dst_w = dst_rect.width();
if (xfer_info.src_is_typeless) src_w *= xfer_info.src_scaling_hint;
if (xfer_info.dst_is_typeless) dst_w *= xfer_info.dst_scaling_hint;
if (src_w == dst_w)
{
// Final dimensions are a match
if (xfer_info.src_is_typeless || xfer_info.dst_is_typeless)
{
const coord3i src_region = { { src_rect.x1, src_rect.y1, 0 }, { src_w, src_rect.height(), 1 } };
const coord3i dst_region = { { dst_rect.x1, dst_rect.y1, 0 }, { src_w, src_rect.height(), 1 } };
gl::copy_typeless(dst, src, dst_region, src_region);
}
else
{
glCopyImageSubData(src->id(), dst->id(), 0, src_rect.x1, src_rect.y1, 0,
dst->id(), GL_TEXTURE_2D, 0, dst_rect.x1, dst_rect.y1, 0,
src_w, src_rect.height(), 1);
}
return;
}
}
if (xfer_info.src_is_typeless)
{
const auto internal_fmt = xfer_info.src_native_format_override ?

11
rpcs3/Emu/RSX/GL/GLTextureCache.h
View File

@ -609,6 +609,17 @@ namespace gl
const u16 src_w2 = u16(src_w * src_bpp) / dst_bpp;
const u16 src_x2 = u16(src_x * src_bpp) / dst_bpp;
if (src_w2 == slice.dst_w && src_h == slice.dst_h && slice.level == 0)
{
// Optimization, avoid typeless copy to tmp followed by data copy to dst
// Combine the two transfers into one
const coord3u src_region = { { src_x, src_y, 0 }, { src_w, src_h, 1 } };
const coord3u dst_region = { { slice.dst_x, slice.dst_y, slice.dst_z }, { slice.dst_w, slice.dst_h, 1 } };
gl::copy_typeless(dst_image, slice.src, dst_region, src_region);
continue;
}
const coord3u src_region = { { src_x, src_y, 0 }, { src_w, src_h, 1 } };
const coord3u dst_region = { { src_x2, src_y, 0 }, { src_w2, src_h, 1 } };
gl::copy_typeless(src_image, slice.src, dst_region, src_region);