bool DAGCombiner::MergeConsecutiveStores(StoreSDNode* St) {
EVT MemVT = St->getMemoryVT();
int64_t ElementSizeBytes = MemVT.getSizeInBits()/8;
+ bool NoVectors = DAG.getMachineFunction().getFunction()->getAttributes().
+ hasAttribute(AttributeSet::FunctionIndex, Attribute::NoImplicitFloat);
// Don't merge vectors into wider inputs.
if (MemVT.isVector() || !MemVT.isSimple())
// We only use vectors if the constant is known to be zero and the
// function is not marked with the noimplicitfloat attribute.
- if (NonZero || (DAG.getMachineFunction().getFunction()->getAttributes().
- hasAttribute(AttributeSet::FunctionIndex,
- Attribute::NoImplicitFloat)))
+ if (NonZero || NoVectors)
LastLegalVectorType = 0;
// Check if we found a legal integer type to store.
if (LastLegalType == 0 && LastLegalVectorType == 0)
return false;
- bool UseVector = LastLegalVectorType > LastLegalType;
+ bool UseVector = (LastLegalVectorType > LastLegalType) && !NoVectors;
unsigned NumElem = UseVector ? LastLegalVectorType : LastLegalType;
// Make sure we have something to merge.
// All loads much share the same chain.
if (LoadNodes[i].MemNode->getChain() != FirstChain)
break;
-
+
int64_t CurrAddress = LoadNodes[i].OffsetFromBase;
if (CurrAddress - StartAddress != (ElementSizeBytes * i))
break;
// Only use vector types if the vector type is larger than the integer type.
// If they are the same, use integers.
- bool UseVectorTy = LastLegalVectorType > LastLegalIntegerType;
+ bool UseVectorTy = LastLegalVectorType > LastLegalIntegerType && !NoVectors;
unsigned LastLegalType = std::max(LastLegalVectorType, LastLegalIntegerType);
// We add +1 here because the LastXXX variables refer to location while
ret void
}
+; No vectors because we use noimplicitfloat
+; CHECK: merge_const_store_no_vec
+; CHECK-NOT: vmovups
+; CHECK: ret
+define void @merge_const_store_no_vec(i32 %count, %struct.B* nocapture %p) noimplicitfloat{
+ %1 = icmp sgt i32 %count, 0
+ br i1 %1, label %.lr.ph, label %._crit_edge
+.lr.ph:
+ %i.02 = phi i32 [ %10, %.lr.ph ], [ 0, %0 ]
+ %.01 = phi %struct.B* [ %11, %.lr.ph ], [ %p, %0 ]
+ %2 = getelementptr inbounds %struct.B* %.01, i64 0, i32 0
+ store i32 0, i32* %2, align 4
+ %3 = getelementptr inbounds %struct.B* %.01, i64 0, i32 1
+ store i32 0, i32* %3, align 4
+ %4 = getelementptr inbounds %struct.B* %.01, i64 0, i32 2
+ store i32 0, i32* %4, align 4
+ %5 = getelementptr inbounds %struct.B* %.01, i64 0, i32 3
+ store i32 0, i32* %5, align 4
+ %6 = getelementptr inbounds %struct.B* %.01, i64 0, i32 4
+ store i32 0, i32* %6, align 4
+ %7 = getelementptr inbounds %struct.B* %.01, i64 0, i32 5
+ store i32 0, i32* %7, align 4
+ %8 = getelementptr inbounds %struct.B* %.01, i64 0, i32 6
+ store i32 0, i32* %8, align 4
+ %9 = getelementptr inbounds %struct.B* %.01, i64 0, i32 7
+ store i32 0, i32* %9, align 4
+ %10 = add nsw i32 %i.02, 1
+ %11 = getelementptr inbounds %struct.B* %.01, i64 1
+ %exitcond = icmp eq i32 %10, %count
+ br i1 %exitcond, label %._crit_edge, label %.lr.ph
+._crit_edge:
+ ret void
+}
+
; Move the constants using a single vector store.
; CHECK: merge_const_store_vec
; CHECK: vmovups