From 1329cb8d8992c34365fcc2ac0447356708157dfb Mon Sep 17 00:00:00 2001
From: Chris Lattner <sabre@nondot.org>
Date: Sat, 26 Jan 2008 19:45:50 +0000
Subject: [PATCH] Infer alignment of loads and increase their alignment when we
 can tell they are from the stack.  This allows us to compile stack-align.ll
 to:

_test:
	movsd	LCPI1_0, %xmm0
	movapd	%xmm0, %xmm1
***	andpd	4(%esp), %xmm1
	andpd	_G, %xmm0
	addsd	%xmm1, %xmm0
	movl	20(%esp), %eax
	movsd	%xmm0, (%eax)
	ret

instead of:

_test:
	movsd	LCPI1_0, %xmm0
**	movsd	4(%esp), %xmm1
**	andpd	%xmm0, %xmm1
	andpd	_G, %xmm0
	addsd	%xmm1, %xmm0
	movl	20(%esp), %eax
	movsd	%xmm0, (%eax)
	ret



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@46401 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 37 +++++++++++++++++++++---
 test/CodeGen/X86/stack-align.ll          | 22 ++++++++++++++
 2 files changed, 55 insertions(+), 4 deletions(-)
 create mode 100644 test/CodeGen/X86/stack-align.ll

diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index ee7ef198aeb..473f9080ef3 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -18,6 +18,7 @@
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetFrameInfo.h"
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
@@ -4073,12 +4074,40 @@ bool DAGCombiner::CombineToPostIndexedLoadStore(SDNode *N) {
 static unsigned InferAlignment(SDOperand Ptr, SelectionDAG &DAG) {
   // If this is a direct reference to a stack slot, use information about the
   // stack slot's alignment.
+  int FrameIdx = 1 << 31;
+  int64_t FrameOffset = 0;
   if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(Ptr)) {
-    return DAG.getMachineFunction().getFrameInfo()->
-         getObjectAlignment(FI->getIndex());
+    FrameIdx = FI->getIndex();
+  } else if (Ptr.getOpcode() == ISD::ADD && 
+             isa<ConstantSDNode>(Ptr.getOperand(1)) &&
+             isa<FrameIndexSDNode>(Ptr.getOperand(0))) {
+    FrameIdx = cast<FrameIndexSDNode>(Ptr.getOperand(0))->getIndex();
+    FrameOffset = Ptr.getConstantOperandVal(1);
+  }
+             
+  if (FrameIdx != (1 << 31)) {
+    // FIXME: Handle FI+CST.
+    const MachineFrameInfo &MFI = *DAG.getMachineFunction().getFrameInfo();
+    if (MFI.isFixedObjectIndex(FrameIdx)) {
+      int64_t ObjectOffset = MFI.getObjectOffset(FrameIdx);
+
+      // The alignment of the frame index can be determined from its offset from
+      // the incoming frame position.  If the frame object is at offset 32 and
+      // the stack is guaranteed to be 16-byte aligned, then we know that the
+      // object is 16-byte aligned.
+      unsigned StackAlign = DAG.getTarget().getFrameInfo()->getStackAlignment();
+      unsigned Align = MinAlign(ObjectOffset, StackAlign);
+      
+      // Finally, the frame object itself may have a known alignment.  Factor
+      // the alignment + offset into a new alignment.  For example, if we know
+      // the  FI is 8 byte aligned, but the pointer is 4 off, we really have a
+      // 4-byte alignment of the resultant pointer.  Likewise align 4 + 4-byte
+      // offset = 4-byte alignment, align 4 + 1-byte offset = align 1, etc.
+      unsigned FIInfoAlign = MinAlign(MFI.getObjectAlignment(FrameIdx), 
+                                      FrameOffset);
+      return std::max(Align, FIInfoAlign);
+    }
   }
-  
-  // FIXME: Handle FI+CST.
   
   return 0;
 }
diff --git a/test/CodeGen/X86/stack-align.ll b/test/CodeGen/X86/stack-align.ll
new file mode 100644
index 00000000000..1e6c2b2af06
--- /dev/null
+++ b/test/CodeGen/X86/stack-align.ll
@@ -0,0 +1,22 @@
+; RUN: llvm-as < %s | llc -relocation-model=static -mcpu=yonah | grep {andpd.*4(%esp), %xmm}
+
+; The double argument is at 4(esp) which is 16-byte aligned, allowing us to
+; fold the load into the andpd.
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target triple = "i686-apple-darwin8"
+@G = external global double
+
+define void @test({ double, double }* byval  %z, double* %P) {
+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]
+	%tmp3 = load double* @G, align 16		; <double> [#uses=1]
+	%tmp4 = tail call double @fabs( double %tmp3 )		; <double> [#uses=1]
+	%tmp6 = add double %tmp4, %tmp2		; <double> [#uses=1]
+	store double %tmp6, double* %P, align 8
+	ret void
+}
+
+declare double @fabs(double)