Fix PR13899

- Update maximal stack alignment when stack arguments are prepared before a
  call.
- Test cases are enhanced to show it's not a Win32 specific issue but a generic
  one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@164946 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/CodeGen/CallingConvLower.h b/include/llvm/CodeGen/CallingConvLower.h
index 3afe309..436918b 100644 (file)
--- a/include/llvm/CodeGen/CallingConvLower.h
+++ b/include/llvm/CodeGen/CallingConvLower.h
@@ -17,6 +17,7 @@
 
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/ValueTypes.h"
 #include "llvm/Target/TargetCallingConv.h"
 #include "llvm/CallingConv.h"
@@ -288,6 +289,7 @@ public:
     StackOffset = ((StackOffset + Align-1) & ~(Align-1));
     unsigned Result = StackOffset;
     StackOffset += Size;
+    MF.getFrameInfo()->ensureMaxAlignment(Align);
     return Result;
   }
 

diff --git a/test/CodeGen/X86/pr13899.ll b/test/CodeGen/X86/pr13899.ll
new file mode 100644 (file)
index 0000000..bc81e34
--- /dev/null
+++ b/test/CodeGen/X86/pr13899.ll
@@ -0,0 +1,58 @@
+; RUN: llc < %s -mtriple=i386-pc-win32 -mcpu=corei7 | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 | FileCheck %s --check-prefix=X64
+
+; ModuleID = 'a.bc'
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f80:128:128-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32-S32"
+target triple = "i386-pc-win32"
+
+%v4_varying_big_struct = type { [4 x <4 x i32>] }
+
+declare <4 x i32> @"foo"(%v4_varying_big_struct, <4 x i32>) nounwind
+
+define <4 x i32> @"bar"(%v4_varying_big_struct %s, <4 x i32> %__mask) nounwind {
+allocas:
+  %calltmp = call <4 x i32> @"foo"(%v4_varying_big_struct %s, <4 x i32> %__mask)
+  ret <4 x i32> %calltmp
+; CHECK: bar
+; CHECK: andl
+; CHECK: call
+; CHECK: ret
+}
+
+declare <8 x float> @bar64(<8 x float> %i0, <8 x float> %i1,
+                         <8 x float> %i2, <8 x float> %i3,
+                         <8 x float> %i4, <8 x float> %i5,
+                         <8 x float> %i6, <8 x float> %i7,
+                         <8 x float> %i8, <8 x float> %i9)
+
+define <8 x float> @foo64(<8 x float>* %p) {
+  %1 = load <8 x float>* %p
+  %idx1 = getelementptr inbounds <8 x float>* %p, i64 1
+  %2 = load <8 x float>* %idx1
+  %idx2 = getelementptr inbounds <8 x float>* %p, i64 2
+  %3 = load <8 x float>* %idx2
+  %idx3 = getelementptr inbounds <8 x float>* %p, i64 3
+  %4 = load <8 x float>* %idx3
+  %idx4 = getelementptr inbounds <8 x float>* %p, i64 4
+  %5 = load <8 x float>* %idx4
+  %idx5 = getelementptr inbounds <8 x float>* %p, i64 5
+  %6 = load <8 x float>* %idx5
+  %idx6 = getelementptr inbounds <8 x float>* %p, i64 6
+  %7 = load <8 x float>* %idx6
+  %idx7 = getelementptr inbounds <8 x float>* %p, i64 7
+  %8 = load <8 x float>* %idx7
+  %idx8 = getelementptr inbounds <8 x float>* %p, i64 8
+  %9 = load <8 x float>* %idx8
+  %idx9 = getelementptr inbounds <8 x float>* %p, i64 9
+  %10 = load <8 x float>* %idx9
+  %r = tail call <8 x float> @bar64(<8 x float> %1, <8 x float> %2,
+                                    <8 x float> %3, <8 x float> %4,
+                                    <8 x float> %5, <8 x float> %6,
+                                    <8 x float> %7, <8 x float> %8,
+                                    <8 x float> %9, <8 x float> %10)
+  ret <8 x float> %r
+; X64: foo
+; X64: and
+; X64: call
+; X64: ret
+}