On 08/07/15 01:39, Cesar Philippidis wrote:
The attached patch fixes it by assigning worker_red_sym to a scratch
register. Is this OK gomp-4_0-branch?
It'd be simpler to not create the PLUS rtx if offset == 0
On 08/04/2015 04:50 AM, Nathan Sidwell wrote:
> +/* Worker reduction address expander. */
> +static rtx
> +nvptx_expand_work_red_addr (tree exp, rtx target,
> + machine_mode ARG_UNUSED (mode),
> + int ignore)
> {
> - return nvptx_expand_lock_unloc
I've committed this to gomp4 branch. It creates a new builtin to be used for
worker-level reductions that Cesar is working on. When the builtin is expanded
it allocates a slot in a new .shared array to hold the reduction variable. This
array is reused for reductions on different loops.
I a