[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [patches] PowerPC fesetexceptflag fixes
- To: Kumar Gala <galak@xxxxxxxxxxxxxxxxxxx>
- Subject: Re: [patches] PowerPC fesetexceptflag fixes
- From: Steven Munroe <munroesj@xxxxxxxxxx>
- Date: Wed, 29 Aug 2007 15:02:49 -0500
Kumar Gala wrote:
>
> On Aug 27, 2007, at 12:59 PM, Joseph S. Myers wrote:
>
>> On Fri, 24 Aug 2007, Daniel Jacobowitz wrote:
>>
>>> The e500 and -msoft-float versions of fesetexceptflag set the state
>>> of all exceptions. They're supposed to leave the state of other
>>> exceptions alone, based on the text in C99. While I was there I
>>> updated some of the functions which should raise (or should not raise)
>>> exceptions to match my understanding of the language standard. I
>>> don't know of anything that tests whether these functions raise
>>> signals in particular conditions, though.
>>>
>>> Does this look OK to commit? If so I'd like to apply it everywhere
>>> (that's trunk, 2.6, 2.5, and FSF libc-ports).
>>
>> It looks reasonable to me, but I'd like Steven to review it.
>
> I'd like to think about this for a day or so. I had come up with some
> usage in the kernel with prctl (PR_GET_FPEXC/PR_SET_FPEXC) for SPE,
> but maybe that's not relevant.
Does the FP performance of E500 change with the MSR FE0/FE1 setting
(interrupts disabled vs Precise)?
I think the new kernel defaults to 0/0 Interrupts disabled so you may
have to the do the prctl call just to get SIGFP to work.