[LLVMdev] ConstantFoldBinaryFP and cross compilation

[Krzysztof Parzyszek]

On 4/27/2013 6:17 PM, Chandler Carruth wrote:
>
> I think that long-term LLVM should, out of the box, support strictly
> correct (according to IEEE) floating point folding outside of fast-math,
> and should provide aggressive as hell folding inside of fast-math.

I agree with the idea of avoiding host's FP to emulate target's FP, 
especially via direct calculations on float/double data types.  Having a 
floating point "folder" (or in general, a model of target's FP) would be 
very useful not only for ensuring the same behavior for all targets 
using the same model, but also would allow support for targets that do 
not conform to IEEE, or implement their own floating point model in 
addition to IEEE (such as 390).

-Krzysztof


-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, 
hosted by The Linux Foundation