On Thu, 5 Nov 2015, Martin Jambor wrote:
> Hi,
>
> in the previous email I wrote we need to "change behavior" of a few
> optimization passes. One was the flattening of GPU functions and the
> other two are in the patch below. It all comes to that, at the
> moment, we need to switch off the vectorizer (only for the GPU
> functions, of course).
>
> We are actually quite close to being able to handle gimple vector
> input in HSA back-end but not all the way yet, and before allowing the
> vectorizer again, we will have to make sure it never produces vectors
> bigger than 128bits (in GPU functions).
Hmm. I'd rather have this modify
DECL_FUNCTION_SPECIFIC_OPTIMIZATION of the hsa function to get this
effect. I think I mentioned this to the OACC guys as well for a
similar needs of them.
Richard.
> Thanks,
>
> Martin
>
>
> 2015-11-05 Martin Jambor <mjam...@suse.cz>
>
> * tree-ssa-loop.c: Include cgraph.c, symbol-summary.c and hsa.h.
> (pass_vectorize::gate): Do not run on HSA functions.
> * tree-vectorizer.c: Include symbol-summary.c and hsa.h.
> (pass_slp_vectorize::gate): Do not run on HSA functions.
>
> diff --git a/gcc/tree-ssa-loop.c b/gcc/tree-ssa-loop.c
> index 8ecd140..0d119e2 100644
> --- a/gcc/tree-ssa-loop.c
> +++ b/gcc/tree-ssa-loop.c
> @@ -35,6 +35,9 @@ along with GCC; see the file COPYING3. If not see
> #include "tree-inline.h"
> #include "tree-scalar-evolution.h"
> #include "tree-vectorizer.h"
> +#include "cgraph.h"
> +#include "symbol-summary.h"
> +#include "hsa.h"
>
>
> /* A pass making sure loops are fixed up. */
> @@ -257,7 +260,8 @@ public:
> /* opt_pass methods: */
> virtual bool gate (function *fun)
> {
> - return flag_tree_loop_vectorize || fun->has_force_vectorize_loops;
> + return (flag_tree_loop_vectorize || fun->has_force_vectorize_loops)
> + && !hsa_gpu_implementation_p (fun->decl);
> }
>
> virtual unsigned int execute (function *);
> diff --git a/gcc/tree-vectorizer.c b/gcc/tree-vectorizer.c
> index b80a8dd..366138c 100644
> --- a/gcc/tree-vectorizer.c
> +++ b/gcc/tree-vectorizer.c
> @@ -75,6 +75,8 @@ along with GCC; see the file COPYING3. If not see
> #include "tree-ssa-propagate.h"
> #include "dbgcnt.h"
> #include "tree-scalar-evolution.h"
> +#include "symbol-summary.h"
> +#include "hsa.h"
>
>
> /* Loop or bb location. */
> @@ -675,7 +677,10 @@ public:
>
> /* opt_pass methods: */
> opt_pass * clone () { return new pass_slp_vectorize (m_ctxt); }
> - virtual bool gate (function *) { return flag_tree_slp_vectorize != 0; }
> + virtual bool gate (function *fun)
> + {
> + return flag_tree_slp_vectorize && !hsa_gpu_implementation_p (fun->decl);
> + }
> virtual unsigned int execute (function *);
>
> }; // class pass_slp_vectorize
>
>
--
Richard Biener <rguent...@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Graham Norton, HRB
21284 (AG Nuernberg)
