[llvm] [AArch64] Disable loop alignment for Windows targets (PR #67894)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-aarch64

<details>
<summary>Changes</summary>

This should fix #<!-- -->66912. When emitting SEH unwind info, we need to be able to calculate the exact length of functions before alignments are fixed. Until that limitation is overcome, just disable all loop alignment on Windows targets.

---
Full diff: https://github.com/llvm/llvm-project/pull/67894.diff


3 Files Affected:

- (modified) llvm/lib/MC/MCWin64EH.cpp (+3) 
- (modified) llvm/lib/Target/AArch64/AArch64ISelLowering.cpp (+6-1) 
- (added) llvm/test/CodeGen/AArch64/sched-loop-align.ll (+21) 


``````````diff
diff --git a/llvm/lib/MC/MCWin64EH.cpp b/llvm/lib/MC/MCWin64EH.cpp
index a2d61da722af870..bb3492bec8aad8a 100644
--- a/llvm/lib/MC/MCWin64EH.cpp
+++ b/llvm/lib/MC/MCWin64EH.cpp
@@ -1402,6 +1402,9 @@ static void ARM64EmitUnwindInfo(MCStreamer &streamer, WinEH::FrameInfo *info,
     // here, but we'd have to emit the pdata, the xdata header, and the
     // epilogue scopes later, since they depend on whether the we need to
     // split the unwind data.
+    //
+    // If this is fixed, remove code in AArch64ISelLowering.cpp that
+    // disables loop alignment on Windows.
     RawFuncLength = GetAbsDifference(streamer, info->FuncletOrFuncEnd,
                                      info->Begin);
   }
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 932b36587f0694e..46064bc7e46783f 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -1051,7 +1051,12 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
   // Set required alignment.
   setMinFunctionAlignment(Align(4));
   // Set preferred alignments.
-  setPrefLoopAlignment(STI.getPrefLoopAlignment());
+
+  // Don't align loops on Windows. The SEH unwind info generation needs to
+  // know the exact length of functions before the alignments have been
+  // expanded.
+  if (!Subtarget->isTargetWindows())
+    setPrefLoopAlignment(STI.getPrefLoopAlignment());
   setMaxBytesForAlignment(STI.getMaxBytesForLoopAlignment());
   setPrefFunctionAlignment(STI.getPrefFunctionAlignment());
 
diff --git a/llvm/test/CodeGen/AArch64/sched-loop-align.ll b/llvm/test/CodeGen/AArch64/sched-loop-align.ll
new file mode 100644
index 000000000000000..5b8e42c2790a439
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sched-loop-align.ll
@@ -0,0 +1,21 @@
+; RUN: llc < %s -mtriple=aarch64-windows | FileCheck %s --check-prefix=WINDOWS
+; RUN: llc < %s -mtriple=aarch64-linux | FileCheck %s --check-prefix=LINUX
+
+define dso_local void @b() #0 {
+entry:
+  br label %for.cond
+
+for.cond:
+  tail call void @a()
+  br label %for.cond
+}
+
+declare dso_local void @a(...)
+
+attributes #0 = { noreturn nounwind uwtable "tune-cpu"="cortex-a53" }
+
+; LINUX-LABEL: b:
+; LINUX: .p2align 4
+
+; WINDOWS-LABEL: b:
+; WINDOWS-NOT: .p2align

``````````

</details>


https://github.com/llvm/llvm-project/pull/67894