[SDAG] allow partial undef vector constants with select->logic folds

This is an enhancement suggested in the original review/commit:
D97730 / 7fce3322a283

diff --git a/llvm/include/llvm/CodeGen/SelectionDAGNodes.h b/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
index 7adbc5d..4a8dd95 100644 (file)
--- a/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
+++ b/llvm/include/llvm/CodeGen/SelectionDAGNodes.h
@@ -1677,12 +1677,12 @@ bool isNullOrNullSplat(SDValue V, bool AllowUndefs = false);
 /// Return true if the value is a constant 1 integer or a splatted vector of a
 /// constant 1 integer (with no undefs).
 /// Does not permit build vector implicit truncation.
-bool isOneOrOneSplat(SDValue V);
+bool isOneOrOneSplat(SDValue V, bool AllowUndefs = false);
 
 /// Return true if the value is a constant -1 integer or a splatted vector of a
 /// constant -1 integer (with no undefs).
 /// Does not permit build vector implicit truncation.
-bool isAllOnesOrAllOnesSplat(SDValue V);
+bool isAllOnesOrAllOnesSplat(SDValue V, bool AllowUndefs = false);
 
 class GlobalAddressSDNode : public SDNode {
   friend class SelectionDAG;

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 39003d6..caee795 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -9312,22 +9312,22 @@ static SDValue foldBoolSelectToLogic(SDNode *N, SelectionDAG &DAG) {
 
   // select Cond, Cond, F --> or Cond, F
   // select Cond, 1, F    --> or Cond, F
-  if (Cond == T || isOneOrOneSplat(T))
+  if (Cond == T || isOneOrOneSplat(T, /* AllowUndefs */ true))
     return DAG.getNode(ISD::OR, SDLoc(N), VT, Cond, F);
 
   // select Cond, T, Cond --> and Cond, T
   // select Cond, T, 0    --> and Cond, T
-  if (Cond == F || isNullOrNullSplat(F))
+  if (Cond == F || isNullOrNullSplat(F, /* AllowUndefs */ true))
     return DAG.getNode(ISD::AND, SDLoc(N), VT, Cond, T);
 
   // select Cond, T, 1 --> or (not Cond), T
-  if (isOneOrOneSplat(F)) {
+  if (isOneOrOneSplat(F, /* AllowUndefs */ true)) {
     SDValue NotCond = DAG.getNOT(SDLoc(N), Cond, VT);
     return DAG.getNode(ISD::OR, SDLoc(N), VT, NotCond, T);
   }
 
   // select Cond, 0, F --> and (not Cond), F
-  if (isNullOrNullSplat(T)) {
+  if (isNullOrNullSplat(T, /* AllowUndefs */ true)) {
     SDValue NotCond = DAG.getNOT(SDLoc(N), Cond, VT);
     return DAG.getNode(ISD::AND, SDLoc(N), VT, NotCond, F);
   }

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 0c592b2..2b74aae 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -9253,17 +9253,17 @@ bool llvm::isNullOrNullSplat(SDValue N, bool AllowUndefs) {
   return C && C->isNullValue();
 }
 
-bool llvm::isOneOrOneSplat(SDValue N) {
+bool llvm::isOneOrOneSplat(SDValue N, bool AllowUndefs) {
   // TODO: may want to use peekThroughBitcast() here.
   unsigned BitWidth = N.getScalarValueSizeInBits();
-  ConstantSDNode *C = isConstOrConstSplat(N);
+  ConstantSDNode *C = isConstOrConstSplat(N, AllowUndefs);
   return C && C->isOne() && C->getValueSizeInBits(0) == BitWidth;
 }
 
-bool llvm::isAllOnesOrAllOnesSplat(SDValue N) {
+bool llvm::isAllOnesOrAllOnesSplat(SDValue N, bool AllowUndefs) {
   N = peekThroughBitcasts(N);
   unsigned BitWidth = N.getScalarValueSizeInBits();
-  ConstantSDNode *C = isConstOrConstSplat(N);
+  ConstantSDNode *C = isConstOrConstSplat(N, AllowUndefs);
   return C && C->isAllOnesValue() && C->getValueSizeInBits(0) == BitWidth;
 }
 

diff --git a/llvm/test/CodeGen/AArch64/select-with-and-or.ll b/llvm/test/CodeGen/AArch64/select-with-and-or.ll
index a6e1825..94e6860 100644 (file)
--- a/llvm/test/CodeGen/AArch64/select-with-and-or.ll
+++ b/llvm/test/CodeGen/AArch64/select-with-and-or.ll
@@ -122,7 +122,7 @@ define <4 x i1> @and_vec_undef(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i3
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    cmgt v1.4s, v2.4s, v3.4s
-; CHECK-NEXT:    and v0.16b, v1.16b, v0.16b
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    xtn v0.4h, v0.4s
 ; CHECK-NEXT:    ret
   %a = icmp eq <4 x i32> %x, %y
@@ -136,10 +136,8 @@ define <4 x i1> @or_vec_undef(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x i32
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    cmgt v1.4s, v2.4s, v3.4s
+; CHECK-NEXT:    orr v0.16b, v0.16b, v1.16b
 ; CHECK-NEXT:    xtn v0.4h, v0.4s
-; CHECK-NEXT:    xtn v1.4h, v1.4s
-; CHECK-NEXT:    movi v2.4h, #1
-; CHECK-NEXT:    bsl v0.8b, v2.8b, v1.8b
 ; CHECK-NEXT:    ret
   %a = icmp eq <4 x i32> %x, %y
   %b = icmp sgt <4 x i32> %z, %w
@@ -152,9 +150,8 @@ define <4 x i1> @and_not_vec_undef(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    cmgt v1.4s, v2.4s, v3.4s
+; CHECK-NEXT:    bic v0.16b, v1.16b, v0.16b
 ; CHECK-NEXT:    xtn v0.4h, v0.4s
-; CHECK-NEXT:    xtn v1.4h, v1.4s
-; CHECK-NEXT:    bic v0.8b, v1.8b, v0.8b
 ; CHECK-NEXT:    ret
   %a = icmp eq <4 x i32> %x, %y
   %b = icmp sgt <4 x i32> %z, %w
@@ -167,12 +164,8 @@ define <4 x i1> @or_not_vec_undef(<4 x i32> %x, <4 x i32> %y, <4 x i32> %z, <4 x
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    cmeq v0.4s, v0.4s, v1.4s
 ; CHECK-NEXT:    cmgt v1.4s, v2.4s, v3.4s
-; CHECK-NEXT:    movi v2.4h, #1
-; CHECK-NEXT:    xtn v3.4h, v0.4s
-; CHECK-NEXT:    and v0.16b, v1.16b, v0.16b
+; CHECK-NEXT:    orn v0.16b, v1.16b, v0.16b
 ; CHECK-NEXT:    xtn v0.4h, v0.4s
-; CHECK-NEXT:    bic v1.8b, v2.8b, v3.8b
-; CHECK-NEXT:    orr v0.8b, v0.8b, v1.8b
 ; CHECK-NEXT:    ret
   %a = icmp eq <4 x i32> %x, %y
   %b = icmp sgt <4 x i32> %z, %w