summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--lib/CodeGen/SelectionDAG/DAGCombiner.cpp53
-rw-r--r--test/CodeGen/X86/rotate3.ll76
2 files changed, 4 insertions, 125 deletions
diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 805cc9e0f2..7dab11a7e9 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -3342,7 +3342,6 @@ SDNode *DAGCombiner::MatchRotate(SDValue LHS, SDValue RHS, SDLoc DL) {
unsigned OpSizeInBits = VT.getSizeInBits();
SDValue LHSShiftArg = LHSShift.getOperand(0);
SDValue LHSShiftAmt = LHSShift.getOperand(1);
- SDValue RHSShiftArg = RHSShift.getOperand(0);
SDValue RHSShiftAmt = RHSShift.getOperand(1);
// fold (or (shl x, C1), (srl x, C2)) -> (rotl x, C1)
@@ -3402,32 +3401,10 @@ SDNode *DAGCombiner::MatchRotate(SDValue LHS, SDValue RHS, SDLoc DL) {
// fold (or (shl x, (*ext y)), (srl x, (*ext (sub 32, y)))) ->
// (rotr x, (sub 32, y))
if (ConstantSDNode *SUBC =
- dyn_cast<ConstantSDNode>(RExtOp0.getOperand(0))) {
- if (SUBC->getAPIntValue() == OpSizeInBits) {
+ dyn_cast<ConstantSDNode>(RExtOp0.getOperand(0)))
+ if (SUBC->getAPIntValue() == OpSizeInBits)
return DAG.getNode(HasROTL ? ISD::ROTL : ISD::ROTR, DL, VT, LHSShiftArg,
HasROTL ? LHSShiftAmt : RHSShiftAmt).getNode();
- } else if (LHSShiftArg.getOpcode() == ISD::ZERO_EXTEND ||
- LHSShiftArg.getOpcode() == ISD::ANY_EXTEND) {
- // fold (or (shl (*ext x), (*ext y)),
- // (srl (*ext x), (*ext (sub 32, y)))) ->
- // (*ext (rotl x, y))
- // fold (or (shl (*ext x), (*ext y)),
- // (srl (*ext x), (*ext (sub 32, y)))) ->
- // (*ext (rotr x, (sub 32, y)))
- SDValue LArgExtOp0 = LHSShiftArg.getOperand(0);
- EVT LArgVT = LArgExtOp0.getValueType();
- bool HasROTRWithLArg = TLI.isOperationLegalOrCustom(ISD::ROTR, LArgVT);
- bool HasROTLWithLArg = TLI.isOperationLegalOrCustom(ISD::ROTL, LArgVT);
- if (HasROTRWithLArg || HasROTLWithLArg) {
- if (LArgVT.getSizeInBits() == SUBC->getAPIntValue()) {
- SDValue V =
- DAG.getNode(HasROTLWithLArg ? ISD::ROTL : ISD::ROTR, DL, LArgVT,
- LArgExtOp0, HasROTL ? LHSShiftAmt : RHSShiftAmt);
- return DAG.getNode(LHSShiftArg.getOpcode(), DL, VT, V).getNode();
- }
- }
- }
- }
} else if (LExtOp0.getOpcode() == ISD::SUB &&
RExtOp0 == LExtOp0.getOperand(1)) {
// fold (or (shl x, (*ext (sub 32, y))), (srl x, (*ext y))) ->
@@ -3435,32 +3412,10 @@ SDNode *DAGCombiner::MatchRotate(SDValue LHS, SDValue RHS, SDLoc DL) {
// fold (or (shl x, (*ext (sub 32, y))), (srl x, (*ext y))) ->
// (rotl x, (sub 32, y))
if (ConstantSDNode *SUBC =
- dyn_cast<ConstantSDNode>(LExtOp0.getOperand(0))) {
- if (SUBC->getAPIntValue() == OpSizeInBits) {
+ dyn_cast<ConstantSDNode>(LExtOp0.getOperand(0)))
+ if (SUBC->getAPIntValue() == OpSizeInBits)
return DAG.getNode(HasROTR ? ISD::ROTR : ISD::ROTL, DL, VT, LHSShiftArg,
HasROTR ? RHSShiftAmt : LHSShiftAmt).getNode();
- } else if (RHSShiftArg.getOpcode() == ISD::ZERO_EXTEND ||
- RHSShiftArg.getOpcode() == ISD::ANY_EXTEND) {
- // fold (or (shl (*ext x), (*ext (sub 32, y))),
- // (srl (*ext x), (*ext y))) ->
- // (*ext (rotl x, y))
- // fold (or (shl (*ext x), (*ext (sub 32, y))),
- // (srl (*ext x), (*ext y))) ->
- // (*ext (rotr x, (sub 32, y)))
- SDValue RArgExtOp0 = RHSShiftArg.getOperand(0);
- EVT RArgVT = RArgExtOp0.getValueType();
- bool HasROTRWithRArg = TLI.isOperationLegalOrCustom(ISD::ROTR, RArgVT);
- bool HasROTLWithRArg = TLI.isOperationLegalOrCustom(ISD::ROTL, RArgVT);
- if (HasROTRWithRArg || HasROTLWithRArg) {
- if (RArgVT.getSizeInBits() == SUBC->getAPIntValue()) {
- SDValue V =
- DAG.getNode(HasROTRWithRArg ? ISD::ROTR : ISD::ROTL, DL, RArgVT,
- RArgExtOp0, HasROTR ? RHSShiftAmt : LHSShiftAmt);
- return DAG.getNode(RHSShiftArg.getOpcode(), DL, VT, V).getNode();
- }
- }
- }
- }
}
return 0;
diff --git a/test/CodeGen/X86/rotate3.ll b/test/CodeGen/X86/rotate3.ll
deleted file mode 100644
index b92f7c2a60..0000000000
--- a/test/CodeGen/X86/rotate3.ll
+++ /dev/null
@@ -1,76 +0,0 @@
-; Check that (or (shl x, y), (srl x, (sub 32, y))) is folded into (rotl x, y)
-; and (or (shl x, (sub 32, y)), (srl x, r)) into (rotr x, y) even if the
-; argument is zero extended. Fix for PR16726.
-
-; RUN: llc < %s -march=x86-64 -mcpu=corei7 | FileCheck %s
-
-define zeroext i8 @rolbyte(i32 %nBits_arg, i8 %x_arg) nounwind readnone {
-entry:
- %tmp1 = zext i8 %x_arg to i32
- %tmp3 = shl i32 %tmp1, %nBits_arg
- %tmp8 = sub i32 8, %nBits_arg
- %tmp10 = lshr i32 %tmp1, %tmp8
- %tmp11 = or i32 %tmp3, %tmp10
- %tmp12 = trunc i32 %tmp11 to i8
- ret i8 %tmp12
-}
-; CHECK: rolb %cl, %{{[a-z0-9]+}}
-
-
-define zeroext i8 @rorbyte(i32 %nBits_arg, i8 %x_arg) nounwind readnone {
-entry:
- %tmp1 = zext i8 %x_arg to i32
- %tmp3 = lshr i32 %tmp1, %nBits_arg
- %tmp8 = sub i32 8, %nBits_arg
- %tmp10 = shl i32 %tmp1, %tmp8
- %tmp11 = or i32 %tmp3, %tmp10
- %tmp12 = trunc i32 %tmp11 to i8
- ret i8 %tmp12
-}
-; CHECK: rorb %cl, %{{[a-z0-9]+}}
-
-define zeroext i16 @rolword(i32 %nBits_arg, i16 %x_arg) nounwind readnone {
-entry:
- %tmp1 = zext i16 %x_arg to i32
- %tmp3 = shl i32 %tmp1, %nBits_arg
- %tmp8 = sub i32 16, %nBits_arg
- %tmp10 = lshr i32 %tmp1, %tmp8
- %tmp11 = or i32 %tmp3, %tmp10
- %tmp12 = trunc i32 %tmp11 to i16
- ret i16 %tmp12
-}
-; CHECK: rolw %cl, %{{[a-z0-9]+}}
-
-define zeroext i16 @rorword(i32 %nBits_arg, i16 %x_arg) nounwind readnone {
-entry:
- %tmp1 = zext i16 %x_arg to i32
- %tmp3 = lshr i32 %tmp1, %nBits_arg
- %tmp8 = sub i32 16, %nBits_arg
- %tmp10 = shl i32 %tmp1, %tmp8
- %tmp11 = or i32 %tmp3, %tmp10
- %tmp12 = trunc i32 %tmp11 to i16
- ret i16 %tmp12
-}
-; CHECK: rorw %cl, %{{[a-z0-9]+}}
-
-define i64 @roldword(i64 %nBits_arg, i32 %x_arg) nounwind readnone {
-entry:
- %tmp1 = zext i32 %x_arg to i64
- %tmp3 = shl i64 %tmp1, %nBits_arg
- %tmp8 = sub i64 32, %nBits_arg
- %tmp10 = lshr i64 %tmp1, %tmp8
- %tmp11 = or i64 %tmp3, %tmp10
- ret i64 %tmp11
-}
-; CHECK: roll %cl, %{{[a-z0-9]+}}
-
-define zeroext i64 @rordword(i64 %nBits_arg, i32 %x_arg) nounwind readnone {
-entry:
- %tmp1 = zext i32 %x_arg to i64
- %tmp3 = lshr i64 %tmp1, %nBits_arg
- %tmp8 = sub i64 32, %nBits_arg
- %tmp10 = shl i64 %tmp1, %tmp8
- %tmp11 = or i64 %tmp3, %tmp10
- ret i64 %tmp11
-}
-; CHECK: rorl %cl, %{{[a-z0-9]+}}
generated by cgit v1.2.3 (git 2.25.1) at 2024-05-22 17:00:16 +0000