Subzero: workaround bad codegen for Windows x86 at Om1 optimization level

See the referenced bug for the gory details. For now, this makes it so that Subzero's codegen doesn't crash even in the simple ReactorUnitTest cases. Bug: b/171222930 Change-Id: Ie22a91d7e3e4f3d940b7b9120067469ac4264020 Reviewed-on: https://swiftshader-review.googlesource.com/c/SwiftShader/+/49388 Kokoro-Result: kokoro <noreply+kokoro@google.com> Tested-by: Antonio Maiorano <amaiorano@google.com> Reviewed-by: Nicolas Capens <nicolascapens@google.com>

Subzero: workaround bad codegen for Windows x86 at Om1 optimization level
85d9d5d4 · Antonio Maiorano · ee18f390 · 85d9d5d4
Commit 85d9d5d4 authored Oct 19, 2020 by Antonio Maiorano
Hide whitespace changes
Inline Side-by-side

Showing with 2 additions and 1 deletion

IceTargetLoweringX86BaseImpl.h third_party/subzero/src/IceTargetLoweringX86BaseImpl.h +2 -1

No files found.
--- a/third_party/subzero/src/IceTargetLoweringX86BaseImpl.h
+++ b/third_party/subzero/src/IceTargetLoweringX86BaseImpl.h
@@ -614,7 +614,8 @@ template <typename TraitsType> void TargetX86Base<TraitsType>::translateOm1() {
  genTargetHelperCalls();

  // Do not merge Alloca instructions, and lay out the stack.
-  static constexpr bool SortAndCombineAllocas = false;
+  // static constexpr bool SortAndCombineAllocas = false;
+  static constexpr bool SortAndCombineAllocas = true; // TODO(b/171222930): Fix Win32 bug when this is false
  Func->processAllocas(SortAndCombineAllocas);
  Func->dump("After Alloca processing");