https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
Andrew Pinski changed:
What|Removed |Added
CC||thomas at monjalon dot net
--- Comment #
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #33 from janus at gcc dot gnu.org ---
(In reply to janus from comment #30)
> I have verified that the problem goes away when using the following as
> version:
> GNU assembler (GNU Binutils) 2.31.51.20181124
Same ǵoes for the 2.30 bran
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #32 from H.J. Lu ---
*** Bug 86831 has been marked as a duplicate of this bug. ***
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #31 from H.J. Lu ---
*** Bug 86828 has been marked as a duplicate of this bug. ***
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #30 from janus at gcc dot gnu.org ---
I have verified that the problem goes away when using the following as version:
GNU assembler (GNU Binutils) 2.31.51.20181124
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
H.J. Lu changed:
What|Removed |Added
Status|NEW |RESOLVED
See Also|
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #28 from Alexander Monakov ---
It seems like object file is not correctly assembled, note wrong offset to the
last instruction (movdqu):
.s:
vpgatherqd (%rax,%ymm3,4), %xmm4{%k4}
vpgatherqd (%rax,%ymm1,4),
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #27 from janus at gcc dot gnu.org ---
Created attachment 45086
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=45086&action=edit
object file
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #26 from janus at gcc dot gnu.org ---
(In reply to janus from comment #25)
> Created attachment 45085 [details]
> assembler output
In fact this seems to be identical to Jakub's assembly (except for filenames
and versions strings), but
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #25 from janus at gcc dot gnu.org ---
Created attachment 45085
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=45085&action=edit
assembler output
assembler output obtained from comment 0 via "gfortran-9 c0.f90
-march=skylake-avx5
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #24 from Alexander Monakov ---
Janus, can you attach your .s and .o files please?
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #23 from Jakub Jelinek ---
Created attachment 45083
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=45083&action=edit
pr86735.s
Assembly I get which works for me.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #22 from Jakub Jelinek ---
Fedora 27, kernel 4.16.4-200
Intel(R) Core(TM) i9-7960X CPU
glibc 2.26-30, binutils, 2.29-13
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #21 from janus at gcc dot gnu.org ---
(In reply to Jakub Jelinek from comment #20)
> Tried again, this time on real hw (i9-7960X) and still can't reproduce,
Thanks for the data point, Jakub. Could you please share your kernel and glib
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #20 from Jakub Jelinek ---
Tried again, this time on real hw (i9-7960X) and still can't reproduce, tried
current trunk, current 8 branch and r263070 (i.e. trunk from the time this PR
has been filed), none of this reproduces it.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #19 from janus at gcc dot gnu.org ---
(In reply to janus from comment #17)
> Huh, sounds possible. However, I see the buggy behavior also with sde64 (on
> the same machine where I also see it in real HW, see comment 16).
Another machi
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #18 from janus at gcc dot gnu.org ---
(In reply to janus from comment #17)
> However, I see the buggy behavior also with sde64 (on
> the same machine where I also see it in real HW, see comment 16).
But then again, I don't see it with
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #17 from janus at gcc dot gnu.org ---
(In reply to Jakub Jelinek from comment #14)
> /usr/src/sde/sde64 -- ./pr86735
>1 2 3 4 5 6
> 7 8 9 10
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #16 from janus at gcc dot gnu.org ---
(In reply to H.J. Lu from comment #15)
> I can no longer reproduce it under kernel 4.17.11.
But five days ago you could, right? What changed in the meantime?
I still see the problem with 8.2.0 an
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #15 from H.J. Lu ---
I can no longer reproduce it under kernel 4.17.11.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
Jakub Jelinek changed:
What|Removed |Added
CC||jakub at gcc dot gnu.org
--- Comment #14
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #13 from janus at gcc dot gnu.org ---
gfortran.fortran-torture/execute/in-pack.f90 in the testsuite exhibits the same
problem as comment 0, I think. It also contains Fortran array operations and
fails with "-march=skylake-avx512 -Ofast
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #12 from janus at gcc dot gnu.org ---
(In reply to Dominique d'Humieres from comment #11)
> Thanks for the explanation. Note that
>
> % gfcp pr86735.f90 -Ofast -march=skylake -Wall -Wextra -fcheck=all
That probably just means that -f
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #11 from Dominique d'Humieres ---
> If you build for haswell or skylake architecture, that executable will
> not run on ivybridge, of course.
Thanks for the explanation. Note that
% gfcp pr86735.f90 -Ofast -march=skylake -Wall -Wext
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #10 from janus at gcc dot gnu.org ---
(In reply to Dominique d'Humieres from comment #8)
> On x86_64-apple-darwin17 without access to avx512, I see
>
>[..]
>
> Is this the same bug or should I file a new one?
That is not a bug at al
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #9 from H.J. Lu ---
(In reply to Dominique d'Humieres from comment #8)
> On x86_64-apple-darwin17 without access to avx512, I see
>
> Is this the same bug or should I file a new one?
Please file a new one.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #8 from Dominique d'Humieres ---
On x86_64-apple-darwin17 without access to avx512, I see
% gfcp pr86735.f90 -Ofast -march=skylake
% ./a.out
1 2 3 4 5 6
7
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #7 from H.J. Lu ---
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index fc3d6f0aebc..a8709972e9c 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -40402,6 +40402,10 @@ ix86_multiplication_cost (const
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
H.J. Lu changed:
What|Removed |Added
Status|UNCONFIRMED |NEW
Last reconfirmed|
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
--- Comment #5 from janus at gcc dot gnu.org ---
(In reply to Alexander Monakov from comment #3)
> An observation in case it might help to narrow down the problem: gcc-8
> changed to prefer 256-bit-wide vectors on skylake-avx512, so adding
> -mpre
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86735
Richard Biener changed:
What|Removed |Added
Target||x86_64-*-*
CC|
31 matches
Mail list logo
