[X86] Preserve BasePtr for LEA64_32r

When replacing FrameIndex with BasePtr, we must preserve BasePtr for
LEA64_32r since BasePtr is used later for stack adjustment if it is
the same as StackPtr.

Patch by H.J Lu <hjl.tools@gmail.com>

Differential Revision: https://reviews.llvm.org/D23575

llvm-svn: 283486

2 files changed, 49 insertions, 3 deletions

diff --git a/llvm/lib/Target/X86/X86RegisterInfo.cpp b/llvm/lib/Target/X86/X86RegisterInfo.cpp
index 5a33b5defde..4401d1769ba 100644
--- a/llvm/lib/Target/X86/X86RegisterInfo.cpp
+++ b/llvm/lib/Target/X86/X86RegisterInfo.cpp
@@ -669,12 +669,14 @@ X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   // For LEA64_32r when BasePtr is 32-bits (X32) we can use full-size 64-bit
   // register as source operand, semantic is the same and destination is
   // 32-bits. It saves one byte per lea in code since 0x67 prefix is avoided.
+  // Don't change BasePtr since it is used later for stack adjustment.
+  unsigned MachineBasePtr = BasePtr;
   if (Opc == X86::LEA64_32r && X86::GR32RegClass.contains(BasePtr))
-    BasePtr = getX86SubSuperRegister(BasePtr, 64);
+    MachineBasePtr = getX86SubSuperRegister(BasePtr, 64);
 
   // This must be part of a four operand memory reference.  Replace the
-  // FrameIndex with base register with EBP.  Add an offset to the offset.
-  MI.getOperand(FIOperandNum).ChangeToRegister(BasePtr, false);
+  // FrameIndex with base register.  Add an offset to the offset.
+  MI.getOperand(FIOperandNum).ChangeToRegister(MachineBasePtr, false);
 
   // Now add the frame object offset to the offset from EBP.
   int FIOffset;
diff --git a/llvm/test/CodeGen/X86/x32-movtopush64.ll b/llvm/test/CodeGen/X86/x32-movtopush64.ll
new file mode 100644
index 00000000000..d31628d7ba5
--- /dev/null
+++ b/llvm/test/CodeGen/X86/x32-movtopush64.ll
@@ -0,0 +1,44 @@
+; RUN: llc < %s -mtriple=x86_64-linux-gnux32 | FileCheck %s -check-prefix=CHECK
+
+declare void @bar(i32*, i32*, i32*, i32*, i32*, i64*, i32, i32, i32)
+
+; Function Attrs: nounwind uwtable
+define void @foo() {
+entry:
+  %i1 = alloca i32, align 4
+  %i2 = alloca i32, align 4
+  %i3 = alloca i32, align 4
+  %i4 = alloca i32, align 4
+  %i5 = alloca i32, align 4
+  %i6 = alloca i64, align 8
+  %0 = bitcast i32* %i1 to i8*
+  store i32 1, i32* %i1, align 4
+; CHECK: movl $1, 28(%esp)
+  %1 = bitcast i32* %i2 to i8*
+  store i32 2, i32* %i2, align 4
+; CHECK-NEXT: movl $2, 24(%esp)
+  %2 = bitcast i32* %i3 to i8*
+  store i32 3, i32* %i3, align 4
+; CHECK-NEXT: movl $3, 20(%esp)
+  %3 = bitcast i32* %i4 to i8*
+  store i32 4, i32* %i4, align 4
+; CHECK-NEXT: movl $4, 16(%esp)
+  %4 = bitcast i32* %i5 to i8*
+  store i32 5, i32* %i5, align 4
+; CHECK-NEXT: movl $5, 12(%esp)
+  %5 = bitcast i64* %i6 to i8*
+  store i64 6, i64* %i6, align 8
+; CHECK-NEXT: movq $6, 32(%esp)
+; CHECK-NEXT: subl $8, %esp
+; CHECK: leal 36(%rsp), %edi
+; CHECK-NEXT: leal 32(%rsp), %esi
+; CHECK-NEXT: leal 28(%rsp), %edx
+; CHECK-NEXT: leal 24(%rsp), %ecx
+; CHECK-NEXT: leal 20(%rsp), %r8d
+; CHECK-NEXT: leal 40(%rsp), %r9d
+; CHECK: pushq $0
+; CHECK: pushq $0
+; CHECK: pushq $0
+  call void @bar(i32* nonnull %i1, i32* nonnull %i2, i32* nonnull %i3, i32* nonnull %i4, i32* nonnull %i5, i64* nonnull %i6, i32 0, i32 0, i32 0)
+  ret void
+}