[Llvm-bgq-discuss] Interpreting CLANG assembly
Hal Finkel
hfinkel at anl.gov
Thu Apr 18 13:34:22 CDT 2013
----- Original Message -----
> From: "Ian Karlin" <karlin1 at llnl.gov>
> To: "Hal Finkel" <hfinkel at anl.gov>
> Cc: llvm-bgq-discuss at lists.alcf.anl.gov
> Sent: Thursday, April 18, 2013 1:26:49 PM
> Subject: Re: [Llvm-bgq-discuss] Interpreting CLANG assembly
>
> Hi Hal,
>
> I'm not using intrinsics I was more trying to figure out what was
> being
> made SIMD by the compiler and what instructions. The hint to the
> assym
> language was most of what I needed. I'll try out your flags and if
> that
> gets me the rest I'll be good. If not I'll send you the code.
>
> On a side note is there a way to turn off SIMD generation only? That
> is
> another experiment I"m planning on trying.
Okay, sounds good. If you'd like to turn on SIMD generation completely, while still otherwise remaining at -O3, then: pass the flag: -fno-vectorize and also set the environmental variable: BGCLANG_VECTORIZE=no (this latter part will be improved in the future).
-Hal
>
> Thanks!
>
> Ian
>
> On 4/17/13 9:06 PM, "Hal Finkel" <hfinkel at anl.gov> wrote:
>
> >----- Original Message -----
> >> From: "Ian Karlin" <karlin1 at llnl.gov>
> >> To: llvm-bgq-discuss at lists.alcf.anl.gov
> >> Sent: Wednesday, April 17, 2013 8:12:23 PM
> >> Subject: [Llvm-bgq-discuss] Interpreting CLANG assembly
> >>
> >>
> >>
> >> Hi,
> >>
> >>
> >> I'm trying to figure out if CLANG is capable of issuing SIMD
> >> instructions when xlc is not. I'm having two issues. The first is
> >> all the debug information is not put inline with the assembly. At
> >> a
> >> minimum it would be nice to know what function or loop(s) I was
> >> looking at. Second I see you're using a different assembly
> >> language
> >> than xlc. Is there a guide for interpreting it?
> >
> >Can you be more specific about at what you're looking? The
> >underlying
> >assembly language is indeed the same (and, as far as I know, both
> >compilers use the same underlying system assembler).
> >
> >The lack of good feedback on loop vectorization is indeed an issue,
> >if
> >you can provide me with access to your source files, I can help you
> >out.
> >Also, if you run clang with the flags: -mllvm
> >-debug-only=loop-vectorize
> >then you'll see output on the loop vectorizer (which is not the only
> >one
> >in there, but this will give you a good start).
> >
> >To Jeff's comment, there is little reason to use inline assembly
> >with
> >bgclang for SIMD, bgclang supports the same QPX intrinsics as xlc
> >(vec_add, etc.). Use those instead.
> >
> > -Hal
> >
> >>
> >>
> >> Thanks!
> >>
> >>
> >> Ian
> >>
> >>
> >> _______________________________________________
> >> llvm-bgq-discuss mailing list
> >> llvm-bgq-discuss at lists.alcf.anl.gov
> >> https://lists.alcf.anl.gov/mailman/listinfo/llvm-bgq-discuss
> >>
>
>
More information about the llvm-bgq-discuss
mailing list