On Thu, 14 Jun 2018, Jakub Jelinek wrote:
> Hi!
>
> #pragma omp atomic is now allowed inside of simd regions.
> Tested on x86_64-linux, committed to gomp-5_0-branch.
>
> We will actually not vectorize it then though, so some further work will be
> needed in the vectorizer to handle it. Either, if we have hw atomics for both
> the size of the scalar accesses and size of the whole vector type, the
> accesses are adjacent and known to be aligned, we could replace it with
> atomic on the whole vector, or emit as a small loop or unrolled loop doing
> the extraction, scalar atomics and if needed insert result back into
> vectors. Richard, thoughts on that?
What's the semantic of this? Generally for non-vectorizable stmts
we could implement partial vectorization by doing VF number of copies
of the scalar stmt:
tem10 = vinput1[0];
tem20 = vinput2[0];
...
soutput0 = OP (tem10, tem20,...);
...
voutput = { soutput0, ... };
or alternatively specific for the OMP atomic handling, outline the
atomic stmts to a OMP simd function?
Anyway, still need to have an idea of the semantics of an atomic
stmt inside a SIMD loop.
Richard.
> 2018-06-14 Jakub Jelinek <ja...@redhat.com>
>
> * omp-low.c (check_omp_nesting_restrictions): Allow OpenMP atomics
> inside of simd regions.
>
> * c-c++-common/gomp/simd7.c: New test.
>
> --- gcc/omp-low.c.jj 2018-06-08 12:12:34.377111135 +0200
> +++ gcc/omp-low.c 2018-06-14 11:09:55.439824132 +0200
> @@ -2463,9 +2463,13 @@ check_omp_nesting_restrictions (gimple *
> return true;
> }
> }
> + else if (gimple_code (stmt) == GIMPLE_OMP_ATOMIC_LOAD
> + || gimple_code (stmt) == GIMPLE_OMP_ATOMIC_STORE)
> + return true;
> error_at (gimple_location (stmt),
> "OpenMP constructs other than %<#pragma omp ordered simd%>"
> - " may not be nested inside %<simd%> region");
> + " or %<#pragma omp atomic%> may not be nested inside"
> + " %<simd%> region");
> return false;
> }
> else if (gimple_code (ctx->stmt) == GIMPLE_OMP_TEAMS)
> --- gcc/testsuite/c-c++-common/gomp/simd7.c.jj 2018-06-14
> 11:19:30.466274188 +0200
> +++ gcc/testsuite/c-c++-common/gomp/simd7.c 2018-06-14 10:55:52.520186067
> +0200
> @@ -0,0 +1,21 @@
> +int a[64];
> +
> +#pragma omp declare simd linear(x)
> +int
> +bar (int x, int y)
> +{
> + int v;
> + #pragma omp atomic capture
> + v = a[x] += y;
> + return v;
> +}
> +
> +void
> +foo (void)
> +{
> + int i;
> + #pragma omp simd
> + for (i = 0; i < 64; i++)
> + #pragma omp atomic
> + a[i] += 1;
> +}
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Graham Norton, HRB
21284 (AG Nuernberg)
