mirror/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-03-29 22:20:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

rsx: Implement atlas trivialization pass

Browse Source 
- Decompose sequences of copies into the minimally required set.
  This does 2 things:
  a. Reduces GPU workload by doing less transfers
  b. Allows the clipping optimization to kick in and skip a ton of GPU work
This commit is contained in:
kd-11 2023-06-20 19:21:24 +03:00 committed by kd-11
parent b788e05396
commit 69b02f7e46
1 changed files with 39 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
rpcs3/Emu/RSX/Common/texture_cache.h
View File

@ -218,24 +218,51 @@ namespace rsx
texcoord_xform.clamp = false; texcoord_xform.clamp = false;
} }
bool section_fills_target(const copy_region_descriptor& cpy) const
{
return cpy.dst_x == 0 && cpy.dst_y == 0 &&
cpy.dst_w == external_subresource_desc.width && cpy.dst_h == external_subresource_desc.height &&
cpy.src_w == cpy.dst_w && cpy.src_h == cpy.dst_h;
}
void simplify() void simplify()
{ {
// Optimizations in the straightforward methods copy_image_static and copy_image_dynamic make them preferred over the atlas method if (external_subresource_desc.op != deferred_request_command::atlas_gather)
if (external_subresource_desc.op == deferred_request_command::atlas_gather &&
external_subresource_desc.sections_to_copy.size() == 1)
{ {
// Check if the subresource fills the target, if so, change the command to copy_image_static // Only atlas simplification supported for now
const auto &cpy = external_subresource_desc.sections_to_copy.front(); return;
if (cpy.dst_x == 0 && cpy.dst_y == 0 && }
cpy.dst_w == external_subresource_desc.width && cpy.dst_h == external_subresource_desc.height &&
cpy.src_w == cpy.dst_w && cpy.src_h == cpy.dst_h) auto& sections = external_subresource_desc.sections_to_copy;
if (sections.size() > 1)
{
// GPU image copies are expensive, cull unnecessary transfers if possible
for (auto idx = sections.size() - 1; idx >= 1; idx--)
{ {
external_subresource_desc.external_handle = cpy.src; if (section_fills_target(sections[idx]))
external_subresource_desc.x = cpy.src_x; {
external_subresource_desc.y = cpy.src_y; const auto remaining = sections.size() - idx;
external_subresource_desc.op = deferred_request_command::copy_image_static; std::memcpy(
sections.data(),
&sections[idx],
remaining * sizeof(sections[0])
);
sections.resize(remaining);
break;
}
} }
} }
// Optimizations in the straightforward methods copy_image_static and copy_image_dynamic make them preferred over the atlas method
if (sections.size() == 1 && section_fills_target(sections[0]))
{
// Change the command to copy_image_static
const auto cpy = sections[0];
external_subresource_desc.external_handle = cpy.src;
external_subresource_desc.x = cpy.src_x;
external_subresource_desc.y = cpy.src_y;
external_subresource_desc.op = deferred_request_command::copy_image_static;
}
} }
// Returns true if at least threshold% is covered in pixels // Returns true if at least threshold% is covered in pixels