summaryrefslogtreecommitdiffstats
path: root/llvm
diff options
context:
space:
mode:
authorMon P Wang <wangmp@apple.com>2008-12-15 21:44:00 +0000
committerMon P Wang <wangmp@apple.com>2008-12-15 21:44:00 +0000
commit580f2c7b61ec345d50f5d503baad5087a2d0ce1d (patch)
tree73f9606e70a55612740a41444d071271d9f80749 /llvm
parent1349b457eee7240bb8022483db59d7edc87d6b47 (diff)
downloadbcm5719-llvm-580f2c7b61ec345d50f5d503baad5087a2d0ce1d.tar.gz
bcm5719-llvm-580f2c7b61ec345d50f5d503baad5087a2d0ce1d.zip
Added support for splitting and scalarizing vector shifts.
llvm-svn: 61050
Diffstat (limited to 'llvm')
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h1
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp18
-rw-r--r--llvm/test/CodeGen/X86/vshift_scalar.ll11
-rw-r--r--llvm/test/CodeGen/X86/vshift_split2.ll11
4 files changed, 41 insertions, 0 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
index 36c89adebc6..9a42c5656ee 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeTypes.h
@@ -496,6 +496,7 @@ private:
// Vector Result Scalarization: <1 x ty> -> ty.
void ScalarizeVectorResult(SDNode *N, unsigned OpNo);
SDValue ScalarizeVecRes_BinOp(SDNode *N);
+ SDValue ScalarizeVecRes_ShiftOp(SDNode *N);
SDValue ScalarizeVecRes_UnaryOp(SDNode *N);
SDValue ScalarizeVecRes_BIT_CONVERT(SDNode *N);
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index 35eb7cd7db5..5f15fa506d3 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -91,6 +91,10 @@ void DAGTypeLegalizer::ScalarizeVectorResult(SDNode *N, unsigned ResNo) {
case ISD::UDIV:
case ISD::UREM:
case ISD::XOR: R = ScalarizeVecRes_BinOp(N); break;
+
+ case ISD::SHL:
+ case ISD::SRA:
+ case ISD::SRL: R = ScalarizeVecRes_ShiftOp(N); break;
}
// If R is null, the sub-method took care of registering the result.
@@ -104,6 +108,17 @@ SDValue DAGTypeLegalizer::ScalarizeVecRes_BinOp(SDNode *N) {
return DAG.getNode(N->getOpcode(), LHS.getValueType(), LHS, RHS);
}
+SDValue DAGTypeLegalizer::ScalarizeVecRes_ShiftOp(SDNode *N) {
+ SDValue LHS = GetScalarizedVector(N->getOperand(0));
+ SDValue ShiftAmt = GetScalarizedVector(N->getOperand(1));
+ if (TLI.getShiftAmountTy().bitsLT(ShiftAmt.getValueType()))
+ ShiftAmt = DAG.getNode(ISD::TRUNCATE, TLI.getShiftAmountTy(), ShiftAmt);
+ else if (TLI.getShiftAmountTy().bitsGT(ShiftAmt.getValueType()))
+ ShiftAmt = DAG.getNode(ISD::ANY_EXTEND, TLI.getShiftAmountTy(), ShiftAmt);
+
+ return DAG.getNode(N->getOpcode(), LHS.getValueType(), LHS, ShiftAmt);
+}
+
SDValue DAGTypeLegalizer::ScalarizeVecRes_BIT_CONVERT(SDNode *N) {
MVT NewVT = N->getValueType(0).getVectorElementType();
return DAG.getNode(ISD::BIT_CONVERT, NewVT, N->getOperand(0));
@@ -392,6 +407,9 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
case ISD::AND:
case ISD::OR:
case ISD::XOR:
+ case ISD::SHL:
+ case ISD::SRA:
+ case ISD::SRL:
case ISD::UREM:
case ISD::SREM:
case ISD::FREM: SplitVecRes_BinOp(N, Lo, Hi); break;
diff --git a/llvm/test/CodeGen/X86/vshift_scalar.ll b/llvm/test/CodeGen/X86/vshift_scalar.ll
new file mode 100644
index 00000000000..8895cdf8aff
--- /dev/null
+++ b/llvm/test/CodeGen/X86/vshift_scalar.ll
@@ -0,0 +1,11 @@
+; RUN: llvm-as < %s | llc
+
+; Legalization test that requires scalarizing a vector.
+
+define void @update(<1 x i32> %val, <1 x i32>* %dst) nounwind {
+entry:
+ %shl = shl <1 x i32> %val, < i32 2>
+ %shr = ashr <1 x i32> %val, < i32 4>
+ store <1 x i32> %shr, <1 x i32>* %dst
+ ret void
+}
diff --git a/llvm/test/CodeGen/X86/vshift_split2.ll b/llvm/test/CodeGen/X86/vshift_split2.ll
new file mode 100644
index 00000000000..356e0fd1a64
--- /dev/null
+++ b/llvm/test/CodeGen/X86/vshift_split2.ll
@@ -0,0 +1,11 @@
+; RUN: llvm-as < %s | llc
+
+; Legalization example that requires splitting a large vector into smaller pieces.
+
+define void @update(<8 x i32> %val, <8 x i32>* %dst) nounwind {
+entry:
+ %shl = shl <8 x i32> %val, < i32 2, i32 2, i32 2, i32 2, i32 4, i32 4, i32 4, i32 4 >
+ %shr = ashr <8 x i32> %val, < i32 2, i32 2, i32 2, i32 2, i32 4, i32 4, i32 4, i32 4 >
+ store <8 x i32> %shr, <8 x i32>* %dst
+ ret void
+}
OpenPOWER on IntegriCloud