[llvm] 56ed3dd - Pre-commit reproducer for argument copy elison related bug

[Bjorn Pettersson via llvm-commits]

Author: Bjorn Pettersson
Date: 2024-04-23T13:43:44+02:00
New Revision: 56ed3dd77f4ef07a9b35cbb3d0ca868cc0999d01

URL: https://github.com/llvm/llvm-project/commit/56ed3dd77f4ef07a9b35cbb3d0ca868cc0999d01
DIFF: https://github.com/llvm/llvm-project/commit/56ed3dd77f4ef07a9b35cbb3d0ca868cc0999d01.diff

LOG: Pre-commit reproducer for argument copy elison related bug

Adding test case related to
  https://github.com/llvm/llvm-project/issues/89060

It shows that after argument copy elison the scheduler may reorder
a load of the input argument and a store to the same fixed stack
entry (the fixed stack entry that is reused for the local variable).

Added: 
    llvm/test/CodeGen/Hexagon/arg-copy-elison.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/CodeGen/Hexagon/arg-copy-elison.ll b/llvm/test/CodeGen/Hexagon/arg-copy-elison.ll
new file mode 100644
index 00000000000000..14f61f7de351a7
--- /dev/null
+++ b/llvm/test/CodeGen/Hexagon/arg-copy-elison.ll
@@ -0,0 +1,41 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -mtriple hexagon-- -o - %s | FileCheck %s
+
+; Reproducer for https://github.com/llvm/llvm-project/issues/89060
+;
+; Problem was a bug in argument copy elison. Given that the %alloca is
+; eliminated, the same frame index will be used for accessing %alloca and %a
+; on the fixed stack. Care must be taken when setting up
+; MachinePointerInfo/MemOperands for those accesses to either make sure that
+; we always refer to the fixed stack slot the same way (not using the
+; ir.alloca name), or make sure that we still detect that they alias each
+; other if using 
diff erent kinds of MemOperands to identify the same fixed
+; stack entry.
+;
+define i32 @f(i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32 %q1, i32 %a, i32 %q2) {
+; CHECK-LABEL: f:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  // %bb.0:
+; CHECK-NEXT:    {
+; CHECK-NEXT:     r0 = memw(r29+#36)
+; CHECK-NEXT:     memw(r29+#32) = ##666
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     r1 = memw(r29+#28)
+; CHECK-NEXT:     r2 = memw(r29+#32)
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     r0 = sub(r1,r0)
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     r0 = xor(r0,r2)
+; CHECK-NEXT:     jumpr r31
+; CHECK-NEXT:    }
+  %alloca = alloca i32
+  store i32 %a, ptr %alloca     ; Should be elided.
+  store i32 666, ptr %alloca
+  %x = sub i32 %q1, %q2
+  %y = xor i32 %x, %a           ; Results in a load of %a from fixed stack.
+                                ; Using same frame index as elided %alloca.
+  ret i32 %y
+}