https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #21 from CVS Commits ---
The master branch has been updated by Jakub Jelinek :
https://gcc.gnu.org/g:95d67762171f83277a5700b270c0d1e2756f83f4
commit r12-1066-g95d67762171f83277a5700b270c0d1e2756f83f4
Author: Jakub Jelinek
Date:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #20 from Jakub Jelinek ---
Yeah, that is likely what happens, I'll debug tomorrow.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #19 from Alexander Monakov ---
Ah, does the issue arise because foo._omp_fn.0 is (before the patch) callable
in two contexts, in one it's called from host and should be 'omp target
entrypoint', and in the other it's called from
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #18 from Tobias Burnus ---
I think the problem is:
create_omp_child_function(omp_context*, bool)
...
1916 DECL_ATTRIBUTES (decl) = DECL_ATTRIBUTES (current_function_decl);
The code removes then 'omp declare simd' but not 'omp
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #17 from Alexander Monakov ---
Yes, I'd agree normally it's present in the offload table, but ideally if
you're trying to stub out the call, it should not be present in the offload
table.
I think Tobias is saying that on GIMPLE
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #16 from Jakub Jelinek ---
(In reply to Alexander Monakov from comment #14)
> I would break in gdb on cuModuleGetFunction and
>
> x/s $rdx
>
> to print the failing symbol (it's the third argument to the function).
>
> It seems
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #15 from Jakub Jelinek ---
(In reply to Tobias Burnus from comment #13)
> (In reply to Tobias Burnus from comment #9)
> > not found name: 'test_d_normal._omp_fn.0.kd'
>
> I think the problem is the following:
>
> (a) working:
>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #14 from Alexander Monakov ---
I would break in gdb on cuModuleGetFunction and
x/s $rdx
to print the failing symbol (it's the third argument to the function).
It seems the "inner" entrypoint (which your patch attempted to
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #13 from Tobias Burnus ---
(In reply to Tobias Burnus from comment #9)
> not found name: 'test_d_normal._omp_fn.0.kd'
I think the problem is the following:
(a) working:
foo()
#pragma target
bar()
Here, 'foo._omp_fn.0' as as
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #12 from Jakub Jelinek ---
With incremental
--- gcc/omp-offload.c.jj2021-05-25 13:43:01.341137265 +0200
+++ gcc/omp-offload.c 2021-05-25 20:07:01.934506823 +0200
@@ -2696,8 +2696,16 @@ pass_omp_target_link::execute
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
Jakub Jelinek changed:
What|Removed |Added
CC||amonakov at gcc dot gnu.org
---
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #10 from Jakub Jelinek ---
I didn't have the nvidia binary module loaded and cuda installed when doing the
light testing, now I've installed that and see
FAIL: libgomp.c/../libgomp.c-c++-common/for-3.c execution test
FAIL:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #9 from Tobias Burnus ---
(In reply to Jakub Jelinek from comment #8)
> Lightly tested patch.
Just quick manually testing "for-3.c" (I tried -O0 and -O3):
* With nvptx offloading, it compiles + links – but at run time, I get on
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #8 from Jakub Jelinek ---
Created attachment 50867
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50867=edit
gcc12-pr100573.patch
Lightly tested patch.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #7 from Jakub Jelinek ---
Guess that is because the functions that have #pragma omp target teams
directive in it are marked declare target to.
So, either we'd need to play with macros etc. to make sure that those functions
aren't
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #6 from Tobias Burnus ---
(In reply to Jakub Jelinek from comment #5)
> I think we want to fix both for-3.c and for-9.c similarly to
> r11-2571-g916c7a201a9a1dc94f2c056a773826a26d1daca9 i.e.
> #define DO_PRAGMA(x) _Pragma (#x)
>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #5 from Jakub Jelinek ---
The intent of this particular test is to test how orphaned distribute works and
was done before the host teams construct support has been added, which means we
do not want the target teams visible in the
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #4 from Jakub Jelinek ---
(In reply to Tom de Vries from comment #3)
> Hmm, I reproduced the problem on the original test-case:
> libgomp.c-c++-common/for-3.c, and minimized from there:
> ...
> $ cat
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #3 from Tom de Vries ---
Hmm, I reproduced the problem on the original test-case:
libgomp.c-c++-common/for-3.c, and minimized from there:
...
$ cat libgomp/testsuite/libgomp.c-c++-common/for-3.c
/* { dg-additional-options
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #2 from Tom de Vries ---
(In reply to Tobias Burnus from comment #1)
> Created attachment 50803 [details]
> Reduced testcase - works with hostfall back but fails with GCN and nvptx
Is this not an invalid test-case?
The semantics
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100573
--- Comment #1 from Tobias Burnus ---
Created attachment 50803
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50803=edit
Reduced testcase - works with hostfall back but fails with GCN and nvptx
21 matches
Mail list logo
