Reapply r174343, with a fix for a scary DAG combine bug where it failed to differenti...

base point of a load, and the overall alignment of the load.  This caused infinite loops in DAG combine with the
original application of this patch.

ORIGINAL COMMIT LOG:
When the target-independent DAGCombiner inferred a higher alignment for a load,
it would replace the load with one with the higher alignment.  However, it did
not place the new load in the worklist, which prevented later DAG combines in
the same phase (for example, target-specific combines) from ever seeing it.

This patch corrects that oversight, and updates some tests whose output changed
due to slightly different DAGCombine outputs.

llvm-svn: 174431

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 79ec227..d694bc7 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -7199,12 +7199,15 @@ SDValue DAGCombiner::visitLOAD(SDNode *N) {
   // Try to infer better alignment information than the load already has.
   if (OptLevel != CodeGenOpt::None && LD->isUnindexed()) {
     if (unsigned Align = DAG.InferPtrAlignment(Ptr)) {
-      if (Align > LD->getAlignment())
-        return DAG.getExtLoad(LD->getExtensionType(), N->getDebugLoc(),
+      if (Align > LD->getMemOperand()->getBaseAlignment()) {
+        SDValue NewLoad =
+               DAG.getExtLoad(LD->getExtensionType(), N->getDebugLoc(),
                               LD->getValueType(0),
                               Chain, Ptr, LD->getPointerInfo(),
                               LD->getMemoryVT(),
                               LD->isVolatile(), LD->isNonTemporal(), Align);
+        return CombineTo(N, NewLoad, SDValue(NewLoad.getNode(), 1), true);
+      }
     }
   }
 

diff --git a/llvm/test/CodeGen/AArch64/func-argpassing.ll b/llvm/test/CodeGen/AArch64/func-argpassing.ll
index db58798..760c8d4 100644 (file)
--- a/llvm/test/CodeGen/AArch64/func-argpassing.ll
+++ b/llvm/test/CodeGen/AArch64/func-argpassing.ll
@@ -178,7 +178,8 @@ define void @check_i128_stackalign(i32 %val0, i32 %val1, i32 %val2, i32 %val3,
     ; Nothing local on stack in current codegen, so first stack is 16 away
 ; CHECK: ldr {{x[0-9]+}}, [sp, #16]
     ; Important point is that we address sp+24 for second dword
-; CHECK: ldr {{x[0-9]+}}, [sp, #24]
+; CHECK: add     [[REG:x[0-9]+]], sp, #16
+; CHECK: ldr     {{x[0-9]+}}, {{\[}}[[REG]], #8]
     ret void
 }
 

diff --git a/llvm/test/CodeGen/X86/divide-by-constant.ll b/llvm/test/CodeGen/X86/divide-by-constant.ll
index 8e7c13d..9669d97 100644 (file)
--- a/llvm/test/CodeGen/X86/divide-by-constant.ll
+++ b/llvm/test/CodeGen/X86/divide-by-constant.ll
@@ -56,9 +56,9 @@ entry:
   %div = sdiv i16 %x, 10
   ret i16 %div
 ; CHECK: test6:
-; CHECK: imull $26215, %eax, %eax
-; CHECK: shrl  $31, %ecx
-; CHECK: sarl  $18, %eax
+; CHECK: imull $26215, %eax, %ecx
+; CHECK: sarl $18, %ecx
+; CHECK: shrl $15, %eax
 }
 
 define i32 @test7(i32 %x) nounwind {