summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorCraig Topper <craig.topper@intel.com>2017-12-24 02:05:18 +0000
committerCraig Topper <craig.topper@intel.com>2017-12-24 02:05:18 +0000
commitf65a6e4ed827fbbec486ab3d6d4a3e56a2311016 (patch)
tree2d558f4c8dedd23e30bcfb34f3aba63e3cc55dc0
parent2e308a9b2f871b48b393179d68742f1b51ab9ff1 (diff)
downloadbcm5719-llvm-f65a6e4ed827fbbec486ab3d6d4a3e56a2311016.tar.gz
bcm5719-llvm-f65a6e4ed827fbbec486ab3d6d4a3e56a2311016.zip
[DAGCombiners] Don't turn ANDs to shuffles with zero so early. Give some other combines a chance to run.
This moves the combine for turning ANDs into shuffle with zero out of SimplifyVBinOps and places it only in visitAND below the reassociate handling. This fixes the specific case I noticed where we failed to combine two ands with constants. llvm-svn: 321417
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp15
-rw-r--r--llvm/test/CodeGen/X86/v8i1-masks.ll2
2 files changed, 8 insertions, 9 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 17f907eb07e..388663eb1db 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -3987,6 +3987,12 @@ SDValue DAGCombiner::visitAND(SDNode *N) {
// reassociate and
if (SDValue RAND = ReassociateOps(ISD::AND, SDLoc(N), N0, N1))
return RAND;
+
+ // Try to convert a constant mask AND into a shuffle clear mask.
+ if (VT.isVector())
+ if (SDValue Shuffle = XformToShuffleWithZero(N))
+ return Shuffle;
+
// fold (and (or x, C), D) -> D if (C & D) == D
auto MatchSubset = [](ConstantSDNode *LHS, ConstantSDNode *RHS) {
return RHS->getAPIntValue().isSubsetOf(LHS->getAPIntValue());
@@ -16480,6 +16486,8 @@ SDValue DAGCombiner::visitFP16_TO_FP(SDNode *N) {
/// e.g. AND V, <0xffffffff, 0, 0xffffffff, 0>. ==>
/// vector_shuffle V, Zero, <0, 4, 2, 4>
SDValue DAGCombiner::XformToShuffleWithZero(SDNode *N) {
+ assert(N->getOpcode() == ISD::AND && "Unexpected opcode!");
+
EVT VT = N->getValueType(0);
SDValue LHS = N->getOperand(0);
SDValue RHS = peekThroughBitcast(N->getOperand(1));
@@ -16490,9 +16498,6 @@ SDValue DAGCombiner::XformToShuffleWithZero(SDNode *N) {
if (LegalOperations)
return SDValue();
- if (N->getOpcode() != ISD::AND)
- return SDValue();
-
if (RHS.getOpcode() != ISD::BUILD_VECTOR)
return SDValue();
@@ -16581,10 +16586,6 @@ SDValue DAGCombiner::SimplifyVBinOp(SDNode *N) {
N->getOpcode(), SDLoc(LHS), LHS.getValueType(), Ops, N->getFlags()))
return Fold;
- // Try to convert a constant mask AND into a shuffle clear mask.
- if (SDValue Shuffle = XformToShuffleWithZero(N))
- return Shuffle;
-
// Type legalization might introduce new shuffles in the DAG.
// Fold (VBinOp (shuffle (A, Undef, Mask)), (shuffle (B, Undef, Mask)))
// -> (shuffle (VBinOp (A, B)), Undef, Mask).
diff --git a/llvm/test/CodeGen/X86/v8i1-masks.ll b/llvm/test/CodeGen/X86/v8i1-masks.ll
index e3755de3693..a799b0e6f12 100644
--- a/llvm/test/CodeGen/X86/v8i1-masks.ll
+++ b/llvm/test/CodeGen/X86/v8i1-masks.ll
@@ -160,7 +160,6 @@ define <8 x i32> @and_mask_constant(<8 x i32> %v0, <8 x i32> %v1) {
; X32-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X32-AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm0
; X32-AVX2-NEXT: vpand LCPI2_0, %ymm0, %ymm0
-; X32-AVX2-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7],ymm0[8],ymm1[9],ymm0[10],ymm1[11],ymm0[12],ymm1[13],ymm0[14],ymm1[15]
; X32-AVX2-NEXT: retl
;
; X64-AVX2-LABEL: and_mask_constant:
@@ -168,7 +167,6 @@ define <8 x i32> @and_mask_constant(<8 x i32> %v0, <8 x i32> %v1) {
; X64-AVX2-NEXT: vpxor %xmm1, %xmm1, %xmm1
; X64-AVX2-NEXT: vpcmpeqd %ymm1, %ymm0, %ymm0
; X64-AVX2-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0
-; X64-AVX2-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7],ymm0[8],ymm1[9],ymm0[10],ymm1[11],ymm0[12],ymm1[13],ymm0[14],ymm1[15]
; X64-AVX2-NEXT: retq
%m = icmp eq <8 x i32> %v0, zeroinitializer
%mand = and <8 x i1> %m, <i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 true, i1 false>
OpenPOWER on IntegriCloud