RPCS3/llvm
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm.git synced 2025-02-24 21:00:36 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Patch by Shuxin Yang <shuxin.llvm@gmail.com>.

Browse Source 
Original message:

The attached is the fix to radar://11663049. The optimization can be outlined by following rules:

   (select (x != c), e, c) -> select (x != c), e, x),
   (select (x == c), c, e) -> select (x == c), x, e)
where the <c> is an integer constant.

 The reason for this change is that : on x86, conditional-move-from-constant needs two instructions;
however, conditional-move-from-register need only one instruction.

  While the LowerSELECT() sounds to be the most convenient place for this optimization, it turns out to be a bad place. The reason is that by replacing the constant <c> with a symbolic value, it obscure some instruction-combining opportunities which would otherwise be very easy to spot. For that reason, I have to postpone the change to last instruction-combining phase.

  The change passes the test of "make check-all -C <build-root/test" and "make -C project/test-suite/SingleSource".



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@165661 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Nadav Rotem 2012-10-10 21:31:55 +00:00
parent 26160f4e64
commit 87255a431b
2 changed files with 56 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
lib/Target/X86/X86ISelLowering.cpp
View File

@ -14418,6 +14418,7 @@ static SDValue PerformCMOVCombine(SDNode *N, SelectionDAG &DAG,
if (TrueC->getAPIntValue().ult(FalseC->getAPIntValue())) {
CC = X86::GetOppositeBranchCondition(CC);
std::swap(TrueC, FalseC);
std::swap(TrueOp, FalseOp);
}
// Optimize C ? 8 : 0 -> zext(setcc(C)) << 3. Likewise for any pow2/0.
@ -14500,6 +14501,45 @@ static SDValue PerformCMOVCombine(SDNode *N, SelectionDAG &DAG,
}
}
}
// Handle these cases:
// (select (x != c), e, c) -> select (x != c), e, x),
// (select (x == c), c, e) -> select (x == c), x, e)
// where the c is an integer constant, and the "select" is the combination
// of CMOV and CMP.
//
// The rationale for this change is that the conditional-move from a constant
// needs two instructions, however, conditional-move from a register needs
// only one instruction.
//
// CAVEAT: By replacing a constant with a symbolic value, it may obscure
// some instruction-combining opportunities. This opt needs to be
// postponed as late as possible.
//
if (!DCI.isBeforeLegalize() && !DCI.isBeforeLegalizeOps()) {
// the DCI.xxxx conditions are provided to postpone the optimization as
// late as possible.
ConstantSDNode *CmpAgainst = 0;
if ((Cond.getOpcode() == X86ISD::CMP || Cond.getOpcode() == X86ISD::SUB) &&
(CmpAgainst = dyn_cast<ConstantSDNode>(Cond.getOperand(1))) &&
dyn_cast<ConstantSDNode>(Cond.getOperand(0)) == 0) {
if (CC == X86::COND_NE &&
CmpAgainst == dyn_cast<ConstantSDNode>(FalseOp)) {
CC = X86::GetOppositeBranchCondition(CC);
std::swap(TrueOp, FalseOp);
}
if (CC == X86::COND_E &&
CmpAgainst == dyn_cast<ConstantSDNode>(TrueOp)) {
SDValue Ops[] = { FalseOp, Cond.getOperand(0), N->getOperand(2), Cond };
return DAG.getNode(X86ISD::CMOV, DL, N->getVTList (), Ops,
array_lengthof(Ops));
}
}
}
return SDValue();
}

16
test/CodeGen/X86/select_const.ll Normal file
View File

@ -0,0 +1,16 @@
; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -mcpu=corei7 | FileCheck %s
define i64 @test1(i64 %x) nounwind {
entry:
%cmp = icmp eq i64 %x, 2
%add = add i64 %x, 1
%retval.0 = select i1 %cmp, i64 2, i64 %add
ret i64 %retval.0
; CHECK: test1:
; CHECK: leaq 1(%rdi), %rax
; CHECK: cmpq $2, %rdi
; CHECK: cmoveq %rdi, %rax
; CHECK: ret
}