summaryrefslogtreecommitdiffstats
path: root/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
diff options
context:
space:
mode:
authorSimon Pilgrim <llvm-dev@redking.me.uk>2015-04-03 10:02:21 +0000
committerSimon Pilgrim <llvm-dev@redking.me.uk>2015-04-03 10:02:21 +0000
commited2ba33ba0b3f123f22ffd3b09947487f4ac0844 (patch)
tree7e55c262860fc93d7a1513f8766a8e5c044b476a /llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
parent488c89ed77255695ddc1ffc57ee7996caa308778 (diff)
downloadbcm5719-llvm-ed2ba33ba0b3f123f22ffd3b09947487f4ac0844.tar.gz
bcm5719-llvm-ed2ba33ba0b3f123f22ffd3b09947487f4ac0844.zip
[DAGCombiner] Combine shuffles of BUILD_VECTOR and SCALAR_TO_VECTOR
This patch attempts to fold the shuffling of 'scalar source' inputs - BUILD_VECTOR and SCALAR_TO_VECTOR nodes - if the shuffle node is the only user. This folds away a lot of unnecessary shuffle nodes, and allows quite a bit of constant folding that was being missed. Differential Revision: http://reviews.llvm.org/D8516 llvm-svn: 234004
Diffstat (limited to 'llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp')
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp37
1 files changed, 37 insertions, 0 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 19950f5dad1..70f6185964c 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -11980,6 +11980,43 @@ SDValue DAGCombiner::visitVECTOR_SHUFFLE(SDNode *N) {
return V;
}
+ // Attempt to combine a shuffle of 2 inputs of 'scalar sources' -
+ // BUILD_VECTOR or SCALAR_TO_VECTOR into a single BUILD_VECTOR.
+ if (Level < AfterLegalizeVectorOps && TLI.isTypeLegal(VT)) {
+ SmallVector<SDValue, 8> Ops;
+ for (int M : SVN->getMask()) {
+ SDValue Op = DAG.getUNDEF(VT.getScalarType());
+ if (M >= 0) {
+ int Idx = M % NumElts;
+ SDValue &S = (M < (int)NumElts ? N0 : N1);
+ if (S.getOpcode() == ISD::BUILD_VECTOR && S.hasOneUse()) {
+ Op = S.getOperand(Idx);
+ } else if (S.getOpcode() == ISD::SCALAR_TO_VECTOR && S.hasOneUse()) {
+ if (Idx == 0)
+ Op = S.getOperand(0);
+ } else {
+ // Operand can't be combined - bail out.
+ break;
+ }
+ }
+ Ops.push_back(Op);
+ }
+ if (Ops.size() == VT.getVectorNumElements()) {
+ // BUILD_VECTOR requires all inputs to be of the same type, find the
+ // maximum type and extend them all.
+ EVT SVT = VT.getScalarType();
+ if (SVT.isInteger())
+ for (SDValue &Op : Ops)
+ SVT = (SVT.bitsLT(Op.getValueType()) ? Op.getValueType() : SVT);
+ if (SVT != VT.getScalarType())
+ for (SDValue &Op : Ops)
+ Op = TLI.isZExtFree(Op.getValueType(), SVT)
+ ? DAG.getZExtOrTrunc(Op, SDLoc(N), SVT)
+ : DAG.getSExtOrTrunc(Op, SDLoc(N), SVT);
+ return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(N), VT, Ops);
+ }
+ }
+
// If this shuffle only has a single input that is a bitcasted shuffle,
// attempt to merge the 2 shuffles and suitably bitcast the inputs/output
// back to their original types.
OpenPOWER on IntegriCloud