Hi,
In order to add the support for the 'set' directive of OpenACC linked
to this patch in progress:
https://gcc.gnu.org/pipermail/gcc-patches/2026-June/722082.html
I've found the following code in
libgomop/oacc-async.c:validate_async_val
if (async >= 0)
/* TODO: we reserve 0 for acc_async_noval before
we can clarify the semantics of "default_async". */
return 1 + async;
Since the OpenACC specification states the default_async() clause, I'd
like to clarify this in order to integrate this clause in the patch
that I mentioned before.
Since currently we have:
typedef enum acc_async_t {
/* Keep in sync with include/gomp-constants.h. */
acc_async_noval = -1,
acc_async_sync = -2
} acc_async_t;
But in libgomop/oacc-async.c:validate_async_val we have:
if (async == acc_async_sync)
return -1;
if (async == acc_async_noval)
return 0;
if (async >= 0)
/* TODO: we reserve 0 for acc_async_noval before we can clarify
the semantics of "default_async". */
return 1 + async;
- Should we set the default_async to acc_async_noval initially?
And also,
- Should we integrate the default_async value for all the threads in
libgomp/env.c (as an env variable)?
- Or should we include the default_async value for each thread in
goacc_thread instead?
The second one might imply asking for an active device when the 'set'
directive is called (or init a new one? I don't see a good reason to
init a hole thread just for a set call honestly, but I don't know), in
order to set the default value to the specific thread.
---
In my opinion, leaving the default_async to acc_async_noval is
just fine, and we should also set the enum equal to the actual
validate_async_val function work, is just shifting by 1 the values
in the enum. But I would say that I don't know the exact details so
probably I get wrong something or I'm not seeing something critical for
the implementation.
Best,
Sebastian
