Add support for the 'alignstack' attribute to the x86 backend. Fixes PR5254.

Also, FileCheck'ize a test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@96686 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/MachineFunction.cpp b/lib/CodeGen/MachineFunction.cpp
index f141c567708798cc63da5d19da469b94b2a1c383..4377d5bd71acfc8dc3f182a919519e5eaf55d2f3 100644 (file)
--- a/lib/CodeGen/MachineFunction.cpp
+++ b/lib/CodeGen/MachineFunction.cpp
@@ -95,6 +95,9 @@ MachineFunction::MachineFunction(Function *F, const TargetMachine &TM,
   MFInfo = 0;
   FrameInfo = new (Allocator.Allocate<MachineFrameInfo>())
                   MachineFrameInfo(*TM.getFrameInfo());
+  if (Fn->hasFnAttr(Attribute::StackAlignment))
+    FrameInfo->setMaxAlignment(Attribute::getStackAlignmentFromAttrs(
+        Fn->getAttributes().getFnAttributes()));
   ConstantPool = new (Allocator.Allocate<MachineConstantPool>())
                      MachineConstantPool(TM.getTargetData());
   Alignment = TM.getTargetLowering()->getFunctionAlignment(F);

diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp
index 85242368c3cf9f8df14088f1e1206f6e2adf10ad..45ff29750868f9229d52a58699b40ea88130e087 100644 (file)
--- a/lib/Target/X86/X86RegisterInfo.cpp
+++ b/lib/Target/X86/X86RegisterInfo.cpp
@@ -446,8 +446,10 @@ bool X86RegisterInfo::canRealignStack(const MachineFunction &MF) const {
 
 bool X86RegisterInfo::needsStackRealignment(const MachineFunction &MF) const {
   const MachineFrameInfo *MFI = MF.getFrameInfo();
+  const Function *F = MF.getFunction();
   bool requiresRealignment =
-    RealignStack && (MFI->getMaxAlignment() > StackAlign);
+    RealignStack && ((MFI->getMaxAlignment() > StackAlign) ||
+                     F->hasFnAttr(Attribute::StackAlignment));
 
   // FIXME: Currently we don't support stack realignment for functions with
   //        variable-sized allocas.

diff --git a/test/CodeGen/X86/stack-align.ll b/test/CodeGen/X86/stack-align.ll
index cb65e9b50fe2fbf4ab021bc923574839a469678f..e971ef70dbd4a65d06804b9f22afd8c93102a975 100644 (file)
--- a/test/CodeGen/X86/stack-align.ll
+++ b/test/CodeGen/X86/stack-align.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -relocation-model=static -mcpu=yonah | grep {andpd.*4(%esp), %xmm}
+; RUN: llc < %s -relocation-model=static -realign-stack=1 -mcpu=yonah | FileCheck %s
 
 ; The double argument is at 4(esp) which is 16-byte aligned, allowing us to
 ; fold the load into the andpd.
@@ -12,6 +12,7 @@ entry:
        %tmp = getelementptr { double, double }* %z, i32 0, i32 0               ; <double*> [#uses=1]
        %tmp1 = load double* %tmp, align 8              ; <double> [#uses=1]
        %tmp2 = tail call double @fabs( double %tmp1 )          ; <double> [#uses=1]
+    ; CHECK: andpd{{.*}}4(%esp), %xmm
        %tmp3 = load double* @G, align 16               ; <double> [#uses=1]
        %tmp4 = tail call double @fabs( double %tmp3 )          ; <double> [#uses=1]
        %tmp6 = fadd double %tmp4, %tmp2                ; <double> [#uses=1]
@@ -19,4 +20,20 @@ entry:
        ret void
 }
 
+define void @test2() alignstack(16) {
+entry:
+    ; CHECK: andl{{.*}}$-16, %esp
+    ret void
+}
+
+; Use a call to force a spill.
+define <2 x double> @test3(<2 x double> %x, <2 x double> %y) alignstack(32) {
+entry:
+    ; CHECK: andl{{.*}}$-32, %esp
+    call void @test2()
+    %A = mul <2 x double> %x, %y
+    ret <2 x double> %A
+}
+
 declare double @fabs(double)
+