[llvm-commits] CVS: llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
Chris Lattner
lattner at cs.uiuc.edu
Sun Apr 10 16:37:32 PDT 2005
Changes in directory llvm/lib/CodeGen/SelectionDAG:
SelectionDAG.cpp updated: 1.70 -> 1.71
---
Log message:
Don't bother sign/zext_inreg'ing the result of an and operation if we know
the result does change as a result of the extend.
This improves codegen for Alpha on this testcase:
int %a(ushort* %i) {
%tmp.1 = load ushort* %i
%tmp.2 = cast ushort %tmp.1 to int
%tmp.4 = and int %tmp.2, 1
ret int %tmp.4
}
Generating:
a:
ldgp $29, 0($27)
ldwu $0,0($16)
and $0,1,$0
ret $31,($26),1
instead of:
a:
ldgp $29, 0($27)
ldwu $0,0($16)
and $0,1,$0
addl $0,0,$0
ret $31,($26),1
btw, alpha really should switch to livein/outs for args :)
---
Diffs of the changes: (+19 -0)
SelectionDAG.cpp | 19 +++++++++++++++++++
1 files changed, 19 insertions(+)
Index: llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
diff -u llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:1.70 llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:1.71
--- llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:1.70 Sat Apr 9 23:33:08 2005
+++ llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp Sun Apr 10 18:37:16 2005
@@ -1119,6 +1119,25 @@
if (Opcode == ISD::SIGN_EXTEND_INREG) return N1;
break;
}
+
+ // If we are sign extending the result of an (and X, C) operation, and we
+ // know the extended bits are zeros already, don't do the extend.
+ if (N1.getOpcode() == ISD::AND)
+ if (ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getOperand(1))) {
+ uint64_t Mask = N1C->getValue();
+ unsigned NumBits = MVT::getSizeInBits(EVT);
+ if (Opcode == ISD::ZERO_EXTEND_INREG) {
+ if ((Mask & (~0ULL << NumBits)) == 0)
+ return N1;
+ else
+ return getNode(ISD::AND, VT, N1.getOperand(0),
+ getConstant(Mask & (~0ULL >> (64-NumBits)), VT));
+ } else {
+ assert(Opcode == ISD::SIGN_EXTEND_INREG);
+ if ((Mask & (~0ULL << (NumBits-1))) == 0)
+ return N1;
+ }
+ }
break;
}
More information about the llvm-commits
mailing list