Teach InferPtrAlignment to infer GV+cst alignment and use it to simplify x86 isl...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@90925 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index a4b328553b53e20f98a9ac7059194ad6cd3b7c58..3f44b60b303d35a62c5e016f660e0588d5654f6c 100644 (file)
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5914,6 +5914,12 @@ bool SelectionDAG::isConsecutiveLoad(LoadSDNode *LD, LoadSDNode *Base,
 /// InferPtrAlignment - Infer alignment of a load / store address. Return 0 if
 /// it cannot be inferred.
 unsigned SelectionDAG::InferPtrAlignment(SDValue Ptr) const {
+  // If this is a GlobalAddress + cst, return the alignment.
+  GlobalValue *GV;
+  int64_t GVOffset = 0;
+  if (TLI.isGAPlusOffset(Ptr.getNode(), GV, GVOffset))
+    return MinAlign(GV->getAlignment(), GVOffset);
+
   // If this is a direct reference to a stack slot, use information about the
   // stack slot's alignment.
   int FrameIdx = 1 << 31;

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index d66f75436b10f801084bd0902552e83077980f1b..38e1ffe89244ac9bc5d926bbb803e2d3e456fe99 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8327,16 +8327,6 @@ bool X86TargetLowering::isGAPlusOffset(SDNode *N,
   return TargetLowering::isGAPlusOffset(N, GA, Offset);
 }
 
-static bool isBaseAlignmentOfN(unsigned N, SDNode *Base,
-                               const TargetLowering &TLI) {
-  GlobalValue *GV;
-  int64_t Offset = 0;
-  if (TLI.isGAPlusOffset(Base, GV, Offset))
-    return (GV->getAlignment() >= N && (Offset % N) == 0);
-  // DAG combine handles the stack object case.
-  return false;
-}
-
 static bool EltsFromConsecutiveLoads(ShuffleVectorSDNode *N, unsigned NumElems,
                                      EVT EltVT, LoadSDNode *&LDBase,
                                      unsigned &LastLoadedElt,
@@ -8399,7 +8389,7 @@ static SDValue PerformShuffleCombine(SDNode *N, SelectionDAG &DAG,
     return SDValue();
 
   if (LastLoadedElt == NumElems - 1) {
-    if (isBaseAlignmentOfN(16, LD->getBasePtr().getNode(), TLI))
+    if (DAG.InferPtrAlignment(LD->getBasePtr()) >= 16)
       return DAG.getLoad(VT, dl, LD->getChain(), LD->getBasePtr(),
                          LD->getSrcValue(), LD->getSrcValueOffset(),
                          LD->isVolatile());

diff --git a/test/CodeGen/CellSPU/call_indirect.ll b/test/CodeGen/CellSPU/call_indirect.ll
index 639c794424f38ae882fc906980207a81d378bb92..f25d6b5810f5f64ecd903c7d19a76c3936411da8 100644 (file)
--- a/test/CodeGen/CellSPU/call_indirect.ll
+++ b/test/CodeGen/CellSPU/call_indirect.ll
@@ -2,17 +2,17 @@
 ; RUN: llc < %s -march=cellspu -mattr=large_mem > %t2.s
 ; RUN: grep bisl    %t1.s | count 7
 ; RUN: grep ila     %t1.s | count 1
-; RUN: grep rotqby  %t1.s | count 6
+; RUN: grep rotqby  %t1.s | count 5
 ; RUN: grep lqa     %t1.s | count 1
 ; RUN: grep lqd     %t1.s | count 12
 ; RUN: grep dispatch_tab %t1.s | count 5
 ; RUN: grep bisl    %t2.s | count 7
 ; RUN: grep ilhu    %t2.s | count 2
 ; RUN: grep iohl    %t2.s | count 2
-; RUN: grep rotqby  %t2.s | count 6
+; RUN: grep rotqby  %t2.s | count 5
 ; RUN: grep lqd     %t2.s | count 13
 ; RUN: grep ilhu    %t2.s | count 2
-; RUN: grep ai      %t2.s | count 9
+; RUN: grep ai      %t2.s | count 8
 ; RUN: grep dispatch_tab %t2.s | count 6
 
 ; ModuleID = 'call_indirect.bc'