[X86] Check if machine loop is passed while getting loop alignment

[boomanaiden154]

After d6bb96e, calling getPrefLoopAlignment without passing in a pointer to a MachineLoop causes a segmentation fault. This conflicts with the API in TargetLoweringBase where the default MachineLoop pointer passed is nullptr. This patch fixes this by checking if the pointer points to something before enabling the optional functionality.

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Aiden Grossman (boomanaiden154)

Changes

After d6bb96e, calling getPrefLoopAlignment without passing in a pointer to a MachineLoop causes a segmentation fault. This conflicts with the API in TargetLoweringBase where the default MachineLoop pointer passed is nullptr. This patch fixes this by checking if the pointer points to something before enabling the optional functionality.

---

Full diff: https://github.com/llvm/llvm-project/pull/77283.diff

1 Files Affected:

• (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+1-1)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 8c4f091c793dcb..c14e03197aaef1 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -57672,7 +57672,7 @@ X86TargetLowering::getStackProbeSize(const MachineFunction &MF) const {
 }
 
 Align X86TargetLowering::getPrefLoopAlignment(MachineLoop *ML) const {
-  if (ML->isInnermost() &&
+  if (ML && ML->isInnermost() &&
       ExperimentalPrefInnermostLoopAlignment.getNumOccurrences())
     return Align(1ULL << ExperimentalPrefInnermostLoopAlignment);
   return TargetLowering::getPrefLoopAlignment();

[boomanaiden154]

This is needed for #77264. Opened this as a separate PR as I wanted to confirm this is appropriate use of the API with X86 maintainers. There is at least one call to getPrefLoopAlignment inside the ARM backend that doesn't pass a MachineLoop, but nothing in X86. Given that, I think this makes sense, but wanted to confirm.

[RKSimon]

LGTM