[AArch64] Enable vscale_range with +sme (#124466)

If we have +sme but not +sve, we would not set vscale_range on
functions. It should be valid to apply it with the same range with just
+sme, which can help mitigate some performance regressions in cases such
as scalable vector bitcasts (https://godbolt.org/z/exhe4jd8d).

1 files changed, 9 insertions, 8 deletions

diff --git a/clang/lib/CodeGen/CodeGenFunction.cpp b/clang/lib/CodeGen/CodeGenFunction.cpp
index bbef277..08165e0 100644
--- a/clang/lib/CodeGen/CodeGenFunction.cpp
+++ b/clang/lib/CodeGen/CodeGenFunction.cpp
@@ -551,14 +551,6 @@ void CodeGenFunction::FinishFunction(SourceLocation EndLoc) {
     CurFn->addFnAttr("min-legal-vector-width",
                      llvm::utostr(LargestVectorWidth));
 
-  // Add vscale_range attribute if appropriate.
-  std::optional<std::pair<unsigned, unsigned>> VScaleRange =
-      getContext().getTargetInfo().getVScaleRange(getLangOpts());
-  if (VScaleRange) {
-    CurFn->addFnAttr(llvm::Attribute::getWithVScaleRangeArgs(
-        getLLVMContext(), VScaleRange->first, VScaleRange->second));
-  }
-
   // If we generated an unreachable return block, delete it now.
   if (ReturnBlock.isValid() && ReturnBlock.getBlock()->use_empty()) {
     Builder.ClearInsertionPoint();
@@ -1110,6 +1102,15 @@ void CodeGenFunction::StartFunction(GlobalDecl GD, QualType RetTy,
   if (FD && FD->isMain())
     Fn->removeFnAttr("zero-call-used-regs");
 
+  // Add vscale_range attribute if appropriate.
+  std::optional<std::pair<unsigned, unsigned>> VScaleRange =
+      getContext().getTargetInfo().getVScaleRange(
+          getLangOpts(), FD ? IsArmStreamingFunction(FD, true) : false);
+  if (VScaleRange) {
+    CurFn->addFnAttr(llvm::Attribute::getWithVScaleRangeArgs(
+        getLLVMContext(), VScaleRange->first, VScaleRange->second));
+  }
+
   llvm::BasicBlock *EntryBB = createBasicBlock("entry", CurFn);
 
   // Create a marker to make it easy to insert allocas into the entryblock