[llvm-commits] [llvm] r72872 - in /llvm/trunk/lib: CodeGen/SelectionDAG/DAGCombiner.cpp CodeGen/SelectionDAG/SelectionDAG.cpp Transforms/Scalar/InstructionCombining.cpp

[Bill Wendling]

Please look at this test which is now failing because of this change:

Running /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/test/Transforms/InstCombine/dg.exp
...
FAIL: /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/test/Transforms/InstCombine/mul.ll
for PR2642
Failed with exit(1) at line 1
while running:  llvm-as <
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore/test/Transforms/InstCombine/mul.ll
|  opt -instcombine |  llvm-dis | not /usr/bin/grep mul
	%3 = mul <4 x float> %2, zeroinitializer		; <<4 x float>> [#uses=1]
child process exited abnormally

-bw

On Thu, Jun 4, 2009 at 10:12 AM, Dan Gohman <gohman at apple.com> wrote:
> Author: djg
> Date: Thu Jun  4 12:12:12 2009
> New Revision: 72872
>
> URL: http://llvm.org/viewvc/llvm-project?rev=72872&view=rev
> Log:
> Don't do the X * 0.0 -> 0.0 transformation in instcombine, because
> instcombine doesn't know when it's safe. To partially compensate
> for this, introduce new code to do this transformation in
> dagcombine, which can use UnsafeFPMath.
>
> Modified:
>    llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
>    llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
>    llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
>
> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=72872&r1=72871&r2=72872&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
> +++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Thu Jun  4 12:12:12 2009
> @@ -4019,6 +4019,9 @@
>   // fold (fmul A, 0) -> 0
>   if (UnsafeFPMath && N1CFP && N1CFP->getValueAPF().isZero())
>     return N1;
> +  // fold (fmul A, 0) -> 0, vector edition.
> +  if (UnsafeFPMath && ISD::isBuildVectorAllZeros(N1.getNode()))
> +    return N1;
>   // fold (fmul X, 2.0) -> (fadd X, X)
>   if (N1CFP && N1CFP->isExactlyValue(+2.0))
>     return DAG.getNode(ISD::FADD, N->getDebugLoc(), VT, N0, N0);
>
> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp?rev=72872&r1=72871&r2=72872&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp (original)
> +++ llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp Thu Jun  4 12:12:12 2009
> @@ -2807,16 +2807,19 @@
>     case ISD::ADDC:
>     case ISD::ADDE:
>     case ISD::SUB:
> -    case ISD::FADD:
> -    case ISD::FSUB:
> -    case ISD::FMUL:
> -    case ISD::FDIV:
> -    case ISD::FREM:
>     case ISD::UDIV:
>     case ISD::SDIV:
>     case ISD::UREM:
>     case ISD::SREM:
>       return N2;       // fold op(arg1, undef) -> undef
> +    case ISD::FADD:
> +    case ISD::FSUB:
> +    case ISD::FMUL:
> +    case ISD::FDIV:
> +    case ISD::FREM:
> +      if (UnsafeFPMath)
> +        return N2;
> +      break;
>     case ISD::MUL:
>     case ISD::AND:
>     case ISD::SRL:
>
> Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=72872&r1=72871&r2=72872&view=diff
>
> ==============================================================================
> --- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
> +++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Thu Jun  4 12:12:12 2009
> @@ -2585,7 +2585,9 @@
>   bool Changed = SimplifyCommutative(I);
>   Value *Op0 = I.getOperand(0);
>
> -  if (isa<UndefValue>(I.getOperand(1)))              // undef * X -> 0
> +  // TODO: If Op1 is undef and Op0 is finite, return zero.
> +  if (!I.getType()->isFPOrFPVector() &&
> +      isa<UndefValue>(I.getOperand(1)))              // undef * X -> 0
>     return ReplaceInstUsesWith(I, Constant::getNullValue(I.getType()));
>
>   // Simplify mul instructions with a constant RHS...
> @@ -2612,16 +2614,14 @@
>                  ConstantInt::get(Op0->getType(), Val.logBase2()));
>       }
>     } else if (ConstantFP *Op1F = dyn_cast<ConstantFP>(Op1)) {
> -      if (Op1F->isNullValue())
> -        return ReplaceInstUsesWith(I, Op1);
> +      // TODO: If Op1 is zero and Op0 is finite, return zero.
>
>       // "In IEEE floating point, x*1 is not equivalent to x for nans.  However,
>       // ANSI says we can drop signals, so we can do this anyway." (from GCC)
>       if (Op1F->isExactlyValue(1.0))
>         return ReplaceInstUsesWith(I, Op0);  // Eliminate 'mul double %X, 1.0'
>     } else if (isa<VectorType>(Op1->getType())) {
> -      if (isa<ConstantAggregateZero>(Op1))
> -        return ReplaceInstUsesWith(I, Op1);
> +      // TODO: If Op1 is all zeros and Op0 is all finite, return all zeros.
>
>       if (ConstantVector *Op1V = dyn_cast<ConstantVector>(Op1)) {
>         if (Op1V->isAllOnesValue())              // X * -1 == 0 - X
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>