This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Ping Re: [gomp4] Dumping gimple for offload.

On Fri, Nov 29, 2013 at 1:50 PM, Bernd Schmidt <> wrote:
> On 11/29/2013 01:36 PM, Richard Biener wrote:
>> Note that we (SUSE/AMD) sofar think we can go an easier route, not
>> adding a real backend that targets HSAIL/BRIG but instead use a
>> custom GIMPLE SSA -> HSAIL/BRIG translator (including a SSA
>> based register allocator).  Which if course simplifies driving this a bit
>> as we don't need to write/read any GIMPLE.
>> The idea is of course that the "highlevel" target languages, being it
>> HSAIL/BRIG or PTX run through another compiler + optimizer anyway,
>> so machine specific optimization is not necessary (fingers crossing...).
>> Not sure if anybody announced it yet (but gcc-cvs readers may have
>> noticed), there is a 'hsa' branch in svn covering work done sofar
>> (see gcc/README.hsa for how to use it).
> That's also an interesting idea. Did you resurrect the gimple-backend
> branch that I think existed a while ago?

No - I didn't even know there was one ;)  I know of the gimple-frontend

> I'm not sure ptx is really high-level enough for that approach to work
> well though. And gimple looks different for x86 and ptx due to the use
> of address spaces, so I have doubts whether such an approach would be
> suitable.

We'll see - it's good to have both variants tried.  I doubt HSAIL is
more high-level than PTX though, but the unified address space
you have with the HSA framework certainly simplifies things
(not that there isn't talks about optional different memory spaces
in the HSA specs ...).


> Bernd

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]