Reapply 122353-122355 with fixes. 122354 was wrong;
the shift type was needed one place, the shift count
type another. The transform in 123555 had the same
problem.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122366 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 75d6013..38c7c05 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -2967,11 +2967,37 @@
N0.getOperand(1).getOpcode() == ISD::Constant) {
uint64_t c1 = cast<ConstantSDNode>(N0.getOperand(1))->getZExtValue();
uint64_t c2 = N1C->getZExtValue();
- if (c1 + c2 > OpSizeInBits)
+ if (c1 + c2 >= OpSizeInBits)
return DAG.getConstant(0, VT);
return DAG.getNode(ISD::SHL, N->getDebugLoc(), VT, N0.getOperand(0),
DAG.getConstant(c1 + c2, N1.getValueType()));
}
+
+ // fold (shl (ext (shl x, c1)), c2) -> (ext (shl x, (add c1, c2)))
+ // For this to be valid, the second form must not preserve any of the bits
+ // that are shifted out by the inner shift in the first form. This means
+ // the outer shift size must be >= the number of bits added by the ext.
+ // As a corollary, we don't care what kind of ext it is.
+ if (N1C && (N0.getOpcode() == ISD::ZERO_EXTEND ||
+ N0.getOpcode() == ISD::ANY_EXTEND ||
+ N0.getOpcode() == ISD::SIGN_EXTEND) &&
+ N0.getOperand(0).getOpcode() == ISD::SHL &&
+ isa<ConstantSDNode>(N0.getOperand(0)->getOperand(1))) {
+ uint64_t c1 =
+ cast<ConstantSDNode>(N0.getOperand(0)->getOperand(1))->getZExtValue();
+ uint64_t c2 = N1C->getZExtValue();
+ EVT InnerShiftVT = N0.getOperand(0).getValueType();
+ uint64_t InnerShiftSize = InnerShiftVT.getScalarType().getSizeInBits();
+ if (c2 >= OpSizeInBits - InnerShiftSize) {
+ if (c1 + c2 >= OpSizeInBits)
+ return DAG.getConstant(0, VT);
+ return DAG.getNode(ISD::SHL, N0->getDebugLoc(), VT,
+ DAG.getNode(N0.getOpcode(), N0->getDebugLoc(), VT,
+ N0.getOperand(0)->getOperand(0)),
+ DAG.getConstant(c1 + c2, N1.getValueType()));
+ }
+ }
+
// fold (shl (srl x, c1), c2) -> (shl (and x, (shl -1, c1)), (sub c2, c1)) or
// (srl (and x, (shl -1, c1)), (sub c1, c2))
if (N1C && N0.getOpcode() == ISD::SRL &&
@@ -3165,7 +3191,7 @@
N0.getOperand(1).getOpcode() == ISD::Constant) {
uint64_t c1 = cast<ConstantSDNode>(N0.getOperand(1))->getZExtValue();
uint64_t c2 = N1C->getZExtValue();
- if (c1 + c2 > OpSizeInBits)
+ if (c1 + c2 >= OpSizeInBits)
return DAG.getConstant(0, VT);
return DAG.getNode(ISD::SRL, N->getDebugLoc(), VT, N0.getOperand(0),
DAG.getConstant(c1 + c2, N1.getValueType()));
@@ -3178,7 +3204,8 @@
uint64_t c1 =
cast<ConstantSDNode>(N0.getOperand(0)->getOperand(1))->getZExtValue();
uint64_t c2 = N1C->getZExtValue();
- EVT InnerShiftVT = N0.getOperand(0)->getOperand(1).getValueType();
+ EVT InnerShiftVT = N0.getOperand(0).getValueType();
+ EVT ShiftCountVT = N0.getOperand(0)->getOperand(1).getValueType();
uint64_t InnerShiftSize = InnerShiftVT.getScalarType().getSizeInBits();
// This is only valid if the OpSizeInBits + c1 = size of inner shift.
if (c1 + OpSizeInBits == InnerShiftSize) {
@@ -3187,7 +3214,7 @@
return DAG.getNode(ISD::TRUNCATE, N0->getDebugLoc(), VT,
DAG.getNode(ISD::SRL, N0->getDebugLoc(), InnerShiftVT,
N0.getOperand(0)->getOperand(0),
- DAG.getConstant(c1 + c2, InnerShiftVT)));
+ DAG.getConstant(c1 + c2, ShiftCountVT)));
}
}