[llvm-commits] [llvm] r125393 - in /llvm/trunk: lib/VMCore/ConstantFold.cpp test/Transforms/InstCombine/bitcast-vec-uniform.ll test/Transforms/InstCombine/fold-vector-select.ll

[Duncan Sands]

Hi Nadav,

> I attached a revised patch which supports getAllOnesValue on FP values. I added minor changes to APFloat as well as new tests.


> --- include/llvm/ADT/APFloat.h	(revision 125681)
> +++ include/llvm/ADT/APFloat.h	(working copy)
> @@ -246,6 +246,13 @@
>      static APFloat getSmallestNormalized(const fltSemantics &Sem,
>                                           bool Negative = false);
>
> +    /// getAllOnesValue - Returns a float which is bitcasted from
> +    /// an all one value int

I don't think you should mention bitcasting here.  How about:
   Returns a float with every bit set to one.

> --- lib/VMCore/Constants.cpp	(revision 125681)
> +++ lib/VMCore/Constants.cpp	(working copy)
> @@ -93,7 +93,26 @@
>    if (const IntegerType *ITy = dyn_cast<IntegerType>(Ty))
>      return ConstantInt::get(Ty->getContext(),
>                              APInt::getAllOnesValue(ITy->getBitWidth()));
> -
> +
> +  if (Ty->isFloatingPointTy()) {
> +    APFloat FL(0.0f);
> +    switch (Ty->getTypeID()) {

You can get the bitwidth using Ty->getPrimitiveSizeInBits().  Thus you only
have to worry about setting the IEEE flag right.  So the switch etc can be
turned into something like this:
   APFloat FL = APFloat::getAllOnesValue(Ty->getPrimitiveSizeInBits(), 
Ty->isPPC_FP128Ty());

Ciao, Duncan.