raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: canonicalize CFG loop headers

Browse Source
This commit is contained in:
nmzik
2026-08-19 16:02:17 +02:00
parent 85cf965d54
commit 4710522e87
4 changed files with 91 additions and 184 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/ShaderRecompiler.cpp
-142
View File
@@ -78,85 +78,6 @@ void LogDispatcherFallback(const CompileOptions& options, const CFG::Graph& cfg,
static_cast<uint64_t>(cfg.back_edges.size()), reason.c_str());
}
bool InstructionMaySplitSpirvBlock(const IR::Instruction& inst) {
switch (inst.op) {
case IR::Opcode::SLoadDword:
case IR::Opcode::SBufferLoadDword:
case IR::Opcode::BufferLoadUbyte:
case IR::Opcode::BufferLoadSbyte:
case IR::Opcode::BufferLoadUshort:
case IR::Opcode::BufferLoadSshort:
case IR::Opcode::BufferLoadDword:
case IR::Opcode::BufferStoreByte:
case IR::Opcode::BufferStoreShort:
case IR::Opcode::BufferStoreDword:
case IR::Opcode::AtomicSwapU32:
case IR::Opcode::AtomicAddU32:
case IR::Opcode::AtomicSubU32:
case IR::Opcode::AtomicSMinI32:
case IR::Opcode::AtomicUMinU32:
case IR::Opcode::AtomicSMaxI32:
case IR::Opcode::AtomicUMaxU32:
case IR::Opcode::AtomicAndU32:
case IR::Opcode::AtomicOrU32:
case IR::Opcode::AtomicXorU32:
case IR::Opcode::AtomicFMinF32:
case IR::Opcode::AtomicFMaxF32:
case IR::Opcode::FlatLoadUbyte:
case IR::Opcode::FlatLoadSbyte:
case IR::Opcode::FlatLoadUshort:
case IR::Opcode::FlatLoadSshort:
case IR::Opcode::FlatLoadDword:
case IR::Opcode::FlatStoreByte:
case IR::Opcode::FlatStoreShort:
case IR::Opcode::FlatStoreDword:
case IR::Opcode::DsReadUbyte:
case IR::Opcode::DsReadSbyte:
case IR::Opcode::DsReadUshort:
case IR::Opcode::DsReadSshort:
case IR::Opcode::DsReadB32:
case IR::Opcode::DsMinF32:
case IR::Opcode::DsMaxF32:
case IR::Opcode::DsWriteByte:
case IR::Opcode::DsWriteShort:
case IR::Opcode::DsWriteB32:
case IR::Opcode::DsWriteAddtidB32:
case IR::Opcode::DsReadAddtidB32:
case IR::Opcode::DsAppend:
case IR::Opcode::DsConsume:
case IR::Opcode::ImageLoad:
case IR::Opcode::ImageStore:
case IR::Opcode::ImageSample:
case IR::Opcode::ImageGather4:
case IR::Opcode::Export: return true;
default: return false;
}
}
bool NeedsDispatcherForStructuredLoopHeader(const IR::Program& ir, uint32_t* failure_block,
std::string* reason) {
for (const auto& block: ir.blocks) {
if (!block.terminator.loop_header) {
continue;
}
for (const auto& inst: block.instructions) {
if (!InstructionMaySplitSpirvBlock(inst)) {
continue;
}
if (reason != nullptr) {
*reason = fmt::format("loop header block {} contains an instruction whose SPIR-V "
"lowering emits internal control flow: {}",
block.id, IR::InstructionToString(inst).c_str());
}
if (failure_block != nullptr) {
*failure_block = block.id;
}
return true;
}
}
return false;
}
enum class EmbeddedFetchValueType {
Unknown,
Constant,
@@ -771,69 +692,6 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
return false;
}
if (!dispatcher_fallback) {
const auto select_dispatcher = [&](uint32_t failure_block, const std::string& reason,
const CFG::Graph& diagnostic_cfg) {
dispatcher_fallback = true;
dispatcher_reason = reason;
cfg.failure_kind = CFG::FailureKind::StructuredControlFlow;
cfg.failure_block = failure_block;
cfg.unsupported_reason = reason;
ir.cfg_failure_kind = CFG::FailureKind::StructuredControlFlow;
ir.fallback_reason = reason;
LogDispatcherFallback(options, diagnostic_cfg, "emit", reason);
};
const auto retry_budget = cfg.natural_loops.size();
for (size_t retry = 0;; retry++) {
std::string emitter_reason;
uint32_t failure_block = UINT32_MAX;
if (!NeedsDispatcherForStructuredLoopHeader(ir, &failure_block, &emitter_reason)) {
break;
}
if (retry >= retry_budget) {
select_dispatcher(failure_block,
emitter_reason + "; no-clone loop-header isolation exhausted",
cfg);
break;
}
const auto* source_block = cfg.FindBlock(failure_block);
const auto source_pc = source_block != nullptr ? source_block->start_pc : UINT32_MAX;
const auto previous_cfg = cfg;
const auto blocks_before = cfg.blocks.size();
std::string isolation_error;
if (!CFG::IsolateLoopHeader(cfg, failure_block, &isolation_error)) {
const auto failed_cfg = cfg;
cfg = previous_cfg;
select_dispatcher(failure_block,
fmt::format("{}; no-clone loop-header isolation failed: {}",
emitter_reason, isolation_error),
failed_cfg);
break;
}
IR::Program isolated_ir;
if (!lower_program(isolated_ir)) {
const auto lower_error = error != nullptr ? *error : "IR lowering failed";
if (error != nullptr) {
error->clear();
}
cfg = previous_cfg;
select_dispatcher(failure_block,
fmt::format("{}; isolated loop-header lowering failed: {}",
emitter_reason, lower_error),
cfg);
break;
}
ir = std::move(isolated_ir);
LOGF("%s CFG isolated semantic loop header: stage=%s hash=0x%016" PRIx64
" source_block=%" PRIu32 " pc=0x%08" PRIx32 " blocks=%" PRIu64 "->%" PRIu64
" reason=%s\n",
GetDumpLabel(options), StageName(options.stage), options.shader_hash,
failure_block, source_pc, static_cast<uint64_t>(blocks_before),
static_cast<uint64_t>(cfg.blocks.size()), emitter_reason.c_str());
}
}
LOGF("%s phase end: stage=%s hash=0x%016" PRIx64 " IR LowerProgram blocks=%" PRIu64
" elapsed_ms=%" PRIu64 "\n",
GetDumpLabel(options), StageName(options.stage), options.shader_hash,