raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 02:47:02 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: specialize guarded packed material selectors

Browse Source
This commit is contained in:
nmzik
2026-10-13 06:07:49 +02:00
parent 907c65f54c
commit d08a6cef64
4 changed files with 165 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp
+20 -11
View File
@@ -205,16 +205,14 @@ void MarkCleanReads(const std::unordered_set<const Inst*>& planned_reads,
}
}
bool ReadScalarTable(uint64_t base, uint64_t size, uint64_t dynamic_offset,
const SrtRuntime& runtime, std::span<uint32_t> words) {
const auto offset = dynamic_offset & ~uint64_t {3};
bool ReadTable(uint64_t base, uint64_t size, uint64_t offset,
const SrtRuntime& runtime, std::span<uint32_t> words) {
const auto count = std::min<uint64_t>(words.size(), offset < size ? (size - offset) / 4u : 0u);
std::ranges::fill(words.subspan(count), 0u);
if (count == 0u) {
return true;
}
base &= AddressMask & ~uint64_t {3};
if (offset > AddressMask - base) {
if (base > AddressMask || offset > AddressMask - base) {
return false;
}
const auto address = base + offset;
@@ -224,6 +222,11 @@ bool ReadScalarTable(uint64_t base, uint64_t size, uint64_t dynamic_offset,
runtime.read_specialization_memory(runtime.userdata, address, prefix);
}
bool ReadScalarTable(uint64_t base, uint64_t size, uint64_t offset,
const SrtRuntime& runtime, std::span<uint32_t> words) {
return ReadTable(base & AddressMask & ~uint64_t {3}, size, offset & ~uint64_t {3}, runtime, words);
}
bool NormalizeIndirectStoreBuffer(DescriptorValue& value, bool indexed, bool formatted) {
ShaderBufferResource descriptor;
if (!DecodeBufferDescriptor(value, descriptor)) return false;
@@ -282,14 +285,19 @@ bool MaterializeIndirectDescriptor(const ResourcePlan& p
}
return true;
};
const auto read_selector = [&](uint64_t base, uint64_t size, uint64_t offset,
std::span<uint32_t> words) {
return selector->global ? ReadTable(base, size, offset, runtime, words)
: ReadScalarTable(base, size, offset, runtime, words);
};
const auto read_keys = [&](uint64_t base, uint64_t size, uint64_t first,
uint64_t step, uint64_t count) {
keys.resize(count);
if (step == 4u) {
return ReadScalarTable(base, size, first, runtime, keys);
return read_selector(base, size, first, keys);
}
for (auto& key: keys) {
if (!ReadScalarTable(base, size, first, runtime, {&key, 1}))
if (!read_selector(base, size, first, {&key, 1}))
return false;
first += step;
}
@@ -516,8 +524,8 @@ bool MaterializeIndirectDescriptor(const ResourcePlan& p
static_cast<uint64_t>(index) * selector->stride;
if (offset > UINT32_MAX) return false;
uint32_t key = 0;
if (!ReadScalarTable(material_base, UINT64_MAX, static_cast<uint32_t>(offset),
runtime, {&key, 1})) return false;
if (!read_selector(material_base, UINT64_MAX, static_cast<uint32_t>(offset),
{&key, 1})) return false;
keys.push_back(key);
mask &= mask - 1u;
}
@@ -538,8 +546,9 @@ bool MaterializeIndirectDescriptor(const ResourcePlan& p
if (count < domain_count) keys.push_back(0u); // An out-of-bounds scalar read returns zero.
}
if (selector != nullptr && selector->byte_ranges.empty()) {
if (selector->maximum != UINT32_MAX)
for (auto& key: keys) key = std::min(key, selector->maximum);
if (selector->bit_offset != 0u || selector->bit_mask != UINT32_MAX || selector->maximum != UINT32_MAX)
for (auto& key: keys)
key = std::min((key >> selector->bit_offset) & selector->bit_mask, selector->maximum);
std::ranges::sort(keys);
keys.erase(std::unique(keys.begin(), keys.end()), keys.end());
}