Signal when SPIR-V legalization needs loop unroll

[AnastaZIuk]

Summary

• add dedicated SpirvEmitter bits for legalization-time loop unroll and legalization-time SSA rewrite
• set those bits only for the known lowering paths that still need those expensive transforms for correctness
• pass the bits into the narrowed SPIR-V legalization overload
• update a few CodeGenSPIRV checks so they validate the semantically relevant SPIR-V shape instead of the older inflated IR shape
• update the nested SPIR-V submodule pointers so this branch materializes the validated companion optimizer state
• use the DXC trunk Godbolt reproducer, which is the preprocessed output of our path tracer at about 58k LoC

Root cause

The expensive blanket policy lives in the SPIR-V optimizer recipe, not in HLSL parsing or the DXC CLI front-end.

DXC is still the producer that knows when a specific lowering path genuinely needs those expensive legalization transforms for correctness, so the optimizer should not have to guess that globally.

Two confirmed correctness-sensitive cases are:

• variable image sample offsets, which still need materialized loop unroll to legalize into forms such as ConstOffset
• combined image sampler conversion, which does not need loop unroll but still needs legalize-time SSA rewrite cleanup to avoid invalid SPIR-V such as storing sampled image types

This patch provides that producer-side signal. The companion optimizer change in KhronosGroup/SPIRV-Tools#6612 then uses it to avoid the broader blanket defaults.

Validation

• reproducer: godbolt.org/z/o5xf1hq36 (note: Compiler Explorer cache can make repeated runs look much faster than a cold compile)
• shader payload: preprocessed output of our path tracer at about 58k LoC
• local machine: AMD Ryzen 5 5600G with Radeon Graphics, 6 physical cores, 12 logical processors, Windows-reported max clock 3901 MHz
• on the same payload and the same machine, SPIRV-Tools@487ff843bd8a + DXC@bd9a8b1c5365 reduced the workload from 19.161 s to 6.042 s
• with SPIRV-Tools@57007cf46bb4 + DXC@b02b772e0b50, the same payload measured 4.702 s
• with the current branch pair SPIRV-Tools@7134be5024ff + DXC@55112e338fd2, the same payload now measures 2.464 s
• full local CodeGenSPIRV lit/FileCheck passes: 1403 expected passes, 2 expected failures, 0 unexpected

Companion SPIR-V optimizer PR:
KhronosGroup/SPIRV-Tools#6612

[github-actions]

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff 54afd2c6a5f1a5bcbef66967515fa69f5a3650fd 9ac57982725c4264348b20ad3fb271acc0993754 -- tools/clang/lib/Frontend/Rewrite/RewriteObjC.cpp tools/clang/lib/SPIRV/SpirvEmitter.cpp tools/clang/lib/SPIRV/SpirvEmitter.h

View the diff from clang-format here.

diff --git a/tools/clang/lib/SPIRV/SpirvEmitter.cpp b/tools/clang/lib/SPIRV/SpirvEmitter.cpp
index 4a616a00..0920fc1c 100644
--- a/tools/clang/lib/SPIRV/SpirvEmitter.cpp
+++ b/tools/clang/lib/SPIRV/SpirvEmitter.cpp
@@ -593,8 +593,7 @@ SpirvEmitter::SpirvEmitter(CompilerInstance &ci)
       constEvaluator(astContext, spvBuilder), entryFunction(nullptr),
       curFunction(nullptr), curThis(nullptr), seenPushConstantAt(),
       isSpecConstantMode(false), needsLegalization(false),
-      needsLegalizationLoopUnroll(false),
-      needsLegalizationSsaRewrite(false),
+      needsLegalizationLoopUnroll(false), needsLegalizationSsaRewrite(false),
       beforeHlslLegalization(false), mainSourceFile(nullptr) {
 
   // Get ShaderModel from command line hlsl profile option.
@@ -957,8 +956,7 @@ void SpirvEmitter::HandleTranslationUnit(ASTContext &context) {
       !dsetbindingsToCombineImageSampler.empty() ||
       spirvOptions.signaturePacking;
   needsLegalizationSsaRewrite =
-      needsLegalizationSsaRewrite ||
-      !dsetbindingsToCombineImageSampler.empty();
+      needsLegalizationSsaRewrite || !dsetbindingsToCombineImageSampler.empty();
 
   // Run legalization passes
   if (spirvOptions.codeGenHighLevel) {
@@ -16679,9 +16677,9 @@ bool SpirvEmitter::spirvToolsLegalize(std::vector<uint32_t> *mod,
   } else if (needsLegalizationSsaRewrite) {
     legalizationSsaRewriteMode = spvtools::SSARewriteMode::OpaqueOnly;
   }
-  optimizer.RegisterLegalizationPasses(
-      spirvOptions.preserveInterface, needsLegalizationLoopUnroll,
-      legalizationSsaRewriteMode);
+  optimizer.RegisterLegalizationPasses(spirvOptions.preserveInterface,
+                                       needsLegalizationLoopUnroll,
+                                       legalizationSsaRewriteMode);
   // Add flattening of resources if needed.
   if (spirvOptions.flattenResourceArrays) {
     optimizer.RegisterPass(

• Check this box to apply formatting changes to this branch.