raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 13:56:59 +01:00
Code Issues Packages Projects Releases Wiki Activity

shader: own control flow on IR blocks with direct targets

Browse Source
This commit is contained in:
nmzik
2026-10-09 01:59:41 +02:00
parent 8abf99c56e
commit 3df456675d
13 changed files with 477 additions and 472 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp
+18 -22
View File
@@ -680,30 +680,29 @@ bool BuildSamplerPlan(const ShaderInfo& base, SamplerPlan& plan) {
template <typename Predicate>
static std::vector<ResourceBlock> ResourceControlFlow(const Program& program, const Predicate& predicate) {
// The renderer checks captured scalar reads against final buffer and image write ranges.
if (program.blocks.size() != program.block_info.size() || program.has_address_writes) {
if (program.has_address_writes) {
return {};
}
std::unordered_map<uint32_t, uint32_t> indices;
for (uint32_t i = 0; i < program.block_info.size(); i++) {
if (!indices.emplace(program.block_info[i].id, i).second) {
std::unordered_map<const Block*, uint32_t> indices;
for (uint32_t i = 0; i < program.blocks.size(); i++) {
if (!indices.emplace(program.blocks[i], i).second) {
return {};
}
}
std::vector<ResourceBlock> blocks(program.blocks.size());
for (uint32_t i = 0; i < blocks.size(); i++) {
auto& block = blocks[i];
const auto& info = program.block_info[i];
const auto& info = *program.blocks[i];
const auto& terminator = info.terminator;
std::vector<uint32_t> successors;
const std::array targets {terminator.true_block, terminator.false_block};
std::span<Block* const> successors;
switch (terminator.kind) {
case CFG::TerminatorKind::Branch: successors.push_back(terminator.true_block); break;
case CFG::TerminatorKind::Branch: successors = std::span(targets).first(1); break;
case CFG::TerminatorKind::ConditionalBranch:
successors = {terminator.true_block, terminator.false_block};
successors = targets;
block.condition = info.condition;
break;
case CFG::TerminatorKind::IndirectBranch:
successors = terminator.indirect_targets;
break;
case CFG::TerminatorKind::IndirectBranch: successors = info.ImmSuccessors(); break;
case CFG::TerminatorKind::Return: break;
default: return {};
}
@@ -824,17 +823,16 @@ static std::optional<std::array<uint64_t, 3>> FillIndex(Value value, uint32_t ax
}
static UniformFillPlan AnalyzeUniformFill(const Program& program) {
if (program.stage != ShaderType::Compute || program.blocks.empty() ||
program.blocks.size() != program.block_info.size() || program.info.uses_dma ||
if (program.stage != ShaderType::Compute || program.blocks.empty() || program.info.uses_dma ||
!program.info.samplers.empty()) {
return {};
}
std::unordered_set<uint32_t> visited;
uint32_t index = 0;
const Inst* store = nullptr;
std::unordered_set<const Block*> visited;
const auto* block = program.blocks.front();
const Inst* store = nullptr;
for (;;) {
if (!visited.insert(index).second) return {};
for (const auto& inst: *program.blocks[index]) {
if (block == nullptr || !visited.insert(block).second) return {};
for (const auto& inst: *block) {
if (AddressOpcodeInfoOf(inst.GetOpcode()).access != AddressAccess::None) return {};
if (!inst.MayHaveSideEffects()) continue;
if (store != nullptr || (BufferAccessOf(inst.GetOpcode()) != BufferAccess::Write &&
@@ -842,12 +840,10 @@ static UniformFillPlan AnalyzeUniformFill(const Program& program) {
return {};
store = &inst;
}
const auto& term = program.block_info[index].terminator;
const auto& term = block->terminator;
if (term.kind == CFG::TerminatorKind::Return) break;
if (term.kind != CFG::TerminatorKind::Branch) return {};
const auto next = std::ranges::find(program.block_info, term.true_block, &BlockInfo::id);
if (next == program.block_info.end()) return {};
index = static_cast<uint32_t>(next - program.block_info.begin());
block = term.true_block;
}
if (store == nullptr || visited.size() != program.blocks.size()) return {};
for (const auto& buffer: program.info.buffers) {
src/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp
+23 -49
View File
@@ -613,8 +613,7 @@ private:
value = value.Resolve();
const auto* phi = value.TryInstruction();
if (m_shader_writes || phi == nullptr || phi->GetOpcode() != ValueOpcode::Phi ||
phi->NumArgs() != 2u || phi->NumPhiBlocks() != 2u || phi->GetType() != Type::U32 ||
m_program.blocks.size() != m_program.block_info.size()) {
phi->NumArgs() != 2u || phi->NumPhiBlocks() != 2u || phi->GetType() != Type::U32) {
return value;
}
const auto* merge = phi->Parent();
@@ -637,7 +636,7 @@ private:
if (branch == merge || branch->ImmSuccessors().size() != 2u) {
return value;
}
std::array<uint32_t, 2> target_ids;
std::array<const Block*, 2> targets;
for (uint32_t arm = 0; arm < 2; arm++) {
const auto* incoming = phi->PhiBlock(arm);
if (incoming == merge ||
@@ -648,29 +647,20 @@ private:
incoming->ImmSuccessors()[0] != merge))) {
return value;
}
const auto* target = incoming == branch ? merge : incoming;
const auto it = std::ranges::find(m_program.blocks, target);
if (it == m_program.blocks.end()) {
return value;
}
target_ids[arm] = m_program.block_info[it - m_program.blocks.begin()].id;
targets[arm] = incoming == branch ? merge : incoming;
}
const auto branch_it = std::ranges::find(m_program.blocks, branch);
if (branch_it == m_program.blocks.end()) {
return value;
}
const auto& info = m_program.block_info[branch_it - m_program.blocks.begin()];
const auto& info = *branch;
const auto& term = info.terminator;
if (term.kind != CFG::TerminatorKind::ConditionalBranch ||
!((term.true_block == target_ids[0] && term.false_block == target_ids[1]) ||
(term.false_block == target_ids[0] && term.true_block == target_ids[1])) ||
!((term.true_block == targets[0] && term.false_block == targets[1]) ||
(term.false_block == targets[0] && term.true_block == targets[1])) ||
!ValidateRuntimeValue(m_program, info.condition, RuntimeValueType::Integer) ||
!ValidateRuntimeValue(m_program, phi->Arg(0)) ||
!ValidateRuntimeValue(m_program, phi->Arg(1))) {
return value;
}
// Retain a host expression; replacing the GPU Phi would break SSA dominance.
const auto true_arg = term.true_block == target_ids[0] ? 0u : 1u;
const auto true_arg = term.true_block == targets[0] ? 0u : 1u;
auto& selected = m_program.value_storage.emplace_back(ValueOpcode::SelectU32);
selected.SetArg(0, info.condition);
selected.SetArg(1, phi->Arg(true_arg));
@@ -852,9 +842,9 @@ private:
if (entry.source.dwords[word].Resolve() == Value(read))
entry.source.dwords[word] = flat;
}
for (auto& info: m_program.block_info) {
if (info.condition.Resolve() == Value(read)) info.condition = flat;
if (info.indirect_target.Resolve() == Value(read)) info.indirect_target = flat;
for (auto* block: m_program.blocks) {
if (block->condition.Resolve() == Value(read)) block->condition = flat;
if (block->indirect_target.Resolve() == Value(read)) block->indirect_target = flat;
}
if (keep) {
if (resolved->planning_handle.IsEmpty()) {
@@ -1022,9 +1012,6 @@ private:
};
bool NonzeroOnEntry(Value value, const Block* block) const {
if (m_program.blocks.size() != m_program.block_info.size()) {
return false;
}
// Each unique predecessor must execute before this use. Stop at joins: an
// unrelated comparison is not a bound on FindILsb's zero-input sentinel.
for (size_t depth = 0; block != nullptr && depth < m_program.blocks.size(); ++depth) {
@@ -1098,15 +1085,11 @@ private:
}
std::optional<EdgePredicate> ConditionalEdge(const Block* from, const Block* to) const {
const auto position = std::ranges::find(m_program.blocks, from);
const auto target = std::ranges::find(m_program.blocks, to);
if (position == m_program.blocks.end() || target == m_program.blocks.end()) return {};
const auto& info = m_program.block_info[position - m_program.blocks.begin()];
const auto& term = info.terminator;
const auto id = m_program.block_info[target - m_program.blocks.begin()].id;
const auto& term = from->terminator;
if (term.kind != CFG::TerminatorKind::ConditionalBranch ||
(term.true_block == id) == (term.false_block == id)) return {};
EdgePredicate edge {info.condition, term.true_block == id};
(term.true_block == to) == (term.false_block == to))
return {};
EdgePredicate edge {from->condition, term.true_block == to};
while (const auto* inst = edge.condition.Resolve().TryInstruction()) {
if (inst->GetOpcode() == ValueOpcode::LogicalNot) {
edge.positive = !edge.positive;
@@ -1470,9 +1453,9 @@ private:
const Inst* BoundedLoop(Value key, const Block* use, const auto& accepts_bound) const {
const auto* phi = key.Resolve().TryInstruction();
if (phi == nullptr || phi->GetOpcode() != ValueOpcode::Phi ||
phi->GetType() != Type::U32 || phi->NumArgs() != 2u || phi->NumPhiBlocks() != 2u ||
m_program.blocks.size() != m_program.block_info.size()) return {};
if (phi == nullptr || phi->GetOpcode() != ValueOpcode::Phi || phi->GetType() != Type::U32 ||
phi->NumArgs() != 2u || phi->NumPhiBlocks() != 2u)
return {};
const Block* increment_block = nullptr;
uint32_t initial_arm = 0;
for (uint32_t initial = 0; initial < 2u; ++initial) {
@@ -1904,22 +1887,13 @@ private:
}
bool ImpossibleSwitchEdge(const Block* from, const Block* to) const {
const auto from_it = std::ranges::find(m_program.blocks, from);
const auto to_it = std::ranges::find(m_program.blocks, to);
if (from_it == m_program.blocks.end() || to_it == m_program.blocks.end() ||
m_program.block_info.size() != m_program.blocks.size()) return false;
const auto& info = m_program.block_info[from_it - m_program.blocks.begin()];
const auto& term = info.terminator;
if (term.kind != CFG::TerminatorKind::IndirectBranch ||
term.indirect_selector_code == UINT32_MAX ||
term.indirect_selector_values.empty() ||
term.indirect_selector_values.size() != term.indirect_selector_targets.size()) return false;
const auto target = m_program.block_info[to_it - m_program.blocks.begin()].id;
const auto maximum = SelectorMaximum(info.indirect_target);
const auto& term = from->terminator;
if (term.kind != CFG::TerminatorKind::IndirectBranch || !term.indexed) return false;
const auto maximum = SelectorMaximum(from->indirect_target);
bool found = false;
for (size_t i = 0; i < term.indirect_selector_values.size(); ++i) {
if (term.indirect_selector_targets[i] != target) continue;
if (term.indirect_selector_values[i] <= maximum) return false;
for (const auto& branch: term.cases) {
if (branch.target != to) continue;
if (branch.value <= maximum) return false;
found = true;
}
return found;