[llvm-commits] [llvm] r42484 - in /llvm/trunk/lib: CodeGen/SelectionDAG/LegalizeDAG.cpp CodeGen/SelectionDAG/SelectionDAG.cpp ExecutionEngine/ExecutionEngine.cpp VMCore/ConstantFold.cpp

[Chris Lattner]

On Oct 1, 2007, at 6:07 AM, Neil Booth wrote:
>> Author: johannes
>> Date: Sun Sep 30 13:19:03 2007
>> New Revision: 42484
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=42484&view=rev
>> Log:
>> Constant fold int-to-long-double conversions;
>> use APFloat for int-to-float/double; use
>> round-to-nearest for these (implementation-defined,
>> seems to match gcc).
>
> Seems sensible.  Ultimately FENV_ACCESS pragma should be taken into
> account, for C at least.

Unfortunately, LLVM doesn't propagate this down into the optimizers  
(yet).

-Chris