gremling/KytyPS5
1
0 Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 08:13:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: preserve affine image heap candidate domains

Browse Source 
Co-authored-by: fxpw <fxpw@outlook.com>
This commit is contained in:
nmzikandfxpw committed 2026-10-15 22:28:01 +02:00
1 parent fcab4e5d0d
commit 5916a184cc
4 files changed
+116 -34
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp
+23 -9
View File
@@ -972,6 +972,7 @@ private:
uint32_t InternSource(const DescriptorSource& descriptor) {
for (uint32_t candidate = 0; candidate < m_sources.size(); candidate++) {
auto& current = m_sources[candidate];
bool shared_image_heap = false;
if (current.dword_count != descriptor.dword_count ||
current.indirect_descriptor.has_value() != descriptor.indirect_descriptor.has_value()) {
continue;
@@ -979,11 +980,16 @@ private:
if (current.indirect_descriptor.has_value()) {
const auto& a = *current.indirect_descriptor;
const auto& b = *descriptor.indirect_descriptor;
// An indexed image heap already covers every DWORD start, so its domain
// subsumes an affine scalar use of the same table and shares its snapshot.
shared_image_heap = descriptor.dword_count == 8u &&
(a.table_record_bytes != 0u || b.table_record_bytes != 0u);
if (a.selector != b.selector || a.table_maximum != b.table_maximum ||
a.pointer_offset != b.pointer_offset ||
a.table_source != b.table_source ||
a.table_offset != b.table_offset || a.table_immediate != b.table_immediate ||
a.table_stride != b.table_stride || a.indexed != b.indexed ||
(!shared_image_heap && (a.table_offset != b.table_offset ||
a.table_immediate != b.table_immediate ||
a.table_stride != b.table_stride)) || a.indexed != b.indexed ||
a.workgroup_axis != b.workgroup_axis || a.sources != b.sources ||
a.signed_count != b.signed_count ||
!EquivalentValue(m_program, a.key_count, b.key_count) ||
@@ -1000,6 +1006,11 @@ private:
}
if (same) {
if (current.indirect_descriptor) {
if (shared_image_heap) {
current.indirect_descriptor->table_stride = 0u;
current.indirect_descriptor->table_offset = 0u;
current.indirect_descriptor->table_immediate = 0u;
}
current.indirect_descriptor->table_scalar |= descriptor.indirect_descriptor->table_scalar;
current.indirect_descriptor->table_record_bytes = std::max(
current.indirect_descriptor->table_record_bytes,
@@ -2920,16 +2931,19 @@ private:
}
if ((table_offset & 3u) != 0u) return false;
} else {
// A bounded V# supplies the complete image table. Leave every GPU selector
// and descriptor read in the shader; the host only translates table bytes.
// Keep the GPU selector and descriptor reads, and preserve the proven scalar
// offset domain within the bounded image table.
if (sampler || plan.split_offsets) return false;
indirect.table_stride = 0u;
indirect.table_offset = 0u;
indirect.table_immediate = 0u;
indirect.workgroup_axis = UINT32_MAX;
plan.table_read = plan.reads[0];
if (plan.table_indexed) indirect.table_record_bytes = table_immediate + plan.read_dwords * 4u;
else indirect.table_scalar = true;
if (plan.table_indexed) {
indirect.table_stride = 0u;
indirect.table_offset = 0u;
indirect.table_immediate = 0u;
indirect.table_record_bytes = table_immediate + plan.read_dwords * 4u;
} else {
indirect.table_scalar = true;
}
plan.retain_reads = true;
}