opal/asm: silence xlc warning about typeof usage

[hjelmn]

Signed-off-by: Nathan Hjelm [email protected]

[ibm-ompi]

The IBM CI (PGI) build failed! Please review the log, linked below.

Gist: https://gist.github.com/3f5da063ed91dc374a76f61c90cea2b6

[jjhursey]

IBM Ci had an issue over the weekend. Retrying. bot:ibm:retest

[hjelmn]

@jjhursey Thanks. Looks good now.

[jjhursey]

Looking at the build log the warning is still present, and this section of code wasn't activated for some reason.

The -qkeyword is defined for the AIX version of XL, but not the Power version. We'll have to look at what we need ot do diferently.

*** C compiler and preprocessor
checking for gcc... (cached) xlc_r
checking whether we are using the GNU C compiler... (cached) yes
checking whether xlc_r accepts -g... (cached) yes
checking for xlc_r option to accept ISO C89... (cached) none needed
checking whether xlc_r understands -c and -o together... (cached) yes
checking if xlc_r requires a flag for C11... yes
configure: checking if xlc_r supports C11 with a flag
checking if xlc_r -std=gnu11 supports C11 _Thread_local... yes
checking if xlc_r -std=gnu11 supports C11 atomic variables... no
checking if xlc_r -std=gnu11 supports C11 _Atomic keyword... no
checking if xlc_r -std=gnu11 supports C11 _Generic keyword... yes
checking if xlc_r -std=gnu11 supports C11 _Static_assert... yes
checking if xlc_r -std=gnu11 supports C11 atomic_fetch_xor_explicit... no
checking if xlc_r -std=c11 supports C11 _Thread_local... yes
checking if xlc_r -std=c11 supports C11 atomic variables... no
checking if xlc_r -std=c11 supports C11 _Atomic keyword... no
checking if xlc_r -std=c11 supports C11 _Generic keyword... yes
checking if xlc_r -std=c11 supports C11 _Static_assert... yes
checking if xlc_r -std=c11 supports C11 atomic_fetch_xor_explicit... no
checking if xlc_r -c11 supports C11 _Thread_local... no
checking if xlc_r -c11 supports C11 atomic variables... no
checking if xlc_r -c11 supports C11 _Atomic keyword... no
checking if xlc_r -c11 supports C11 _Generic keyword... no
checking if xlc_r -c11 supports C11 _Static_assert... no
checking if xlc_r -c11 supports C11 atomic_fetch_xor_explicit... no
checking for xlc_r option to add a directory only to the search path for the quote form of include... -iquote
checking for xlc_r option to accept ISO C99... none needed
checking if xlc_r  supports C11 _Thread_local... yes
checking if xlc_r  supports C11 atomic variables... no
checking if xlc_r  supports C11 _Atomic keyword... no
checking if xlc_r  supports C11 _Generic keyword... yes
checking if xlc_r  supports C11 _Static_assert... yes
checking if xlc_r  supports C11 atomic_fetch_xor_explicit... no
checking if xlc_r  supports __thread... yes
checking if xlc_r  supports C11 _Thread_local... yes
checking for the C compiler vendor... gnu
checking for ANSI C header files... (cached) yes
checking if xlc_r supports -Wno-long-double... no
configure: WARNING:  -Wall -Wundef -Wno-long-long -Wsign-compare -Wmissing-prototypes -Wstrict-prototypes -Wcomment -pedantic -Werror-implicit-function-declaration  has been added to CFLAGS (--enable-picky)
checking if xlc_r supports -finline-functions... yes
checking if xlc_r supports -fno-strict-aliasing... yes
configure: WARNING:  -fno-strict-aliasing has been added to CFLAGS
checking if xlc_r supports __builtin_expect... yes
checking if xlc_r supports __builtin_prefetch... yes
checking if xlc_r supports __builtin_clz... yes
checking for C optimization flags... -O3 -DNDEBUG -Wall -Wundef -Wno-long-long -Wsign-compare -Wmissing-prototypes -Wstrict-prototypes -Wcomment -pedantic -Werror-implicit-function-declaration -finline-functions -fno-strict-aliasing
checking for Interix environment... no

[awlauria]

@jjhursey @hjelmn it looks like it may be because the compiler got labeled as 'gnu', and not 'ibm'. I fixed that in this PR:
062a72b and 6d7a4d3

#8444

[hjelmn]

@awlauria Can you confirm the warning goes away with the PR?

[awlauria]

I verified that this fix worked for the opal/powerpc atomics:
09dfb12

Since @jjhursey mentioned -qkeyword isn't available on Power machines, maybe we should go for that approach if xl is used. I can add that to #8444 or you can here, either works for me.

[hjelmn]

@awlauria Unfortunate that they removed that option yet kept the error. Maybe we should updating the --std flag selection to select c99 (or c11) with extensions? That should also fix it since this is complaining about extensions. I believe for gcc --std=gnu11 is default now so there is a precedent to use the extensions. I will leave it up to you on how we proceed.

[awlauria]

@hjelmn this should do it:

2487ed9

[awlauria]

If adding --std=c99 fixes it, I'd be for that as well. I can test it on our power nodes.

[hjelmn]

@awlauria Actually this: -qlanglvl=extc1x. And for non-C11 versions: -qlanglvl=extc99 should do it. Can you check if they do?

[awlauria]

@hjelmn I can't seem to reproduce this warning on master anymore, so I can't verify either way.

Not sure what commit did it.

[gpaulsen]

@hjelmn, once #8528 goes in, can we just close this?

[hjelmn]

@gpaulsen Ideally the LL/SC atomics will still be used even when using C11. I think @awlauria has already fixed this issue by using __typeof__ instead of typeof with xlc so this can be closed.

[gpaulsen]

retest

[awlauria]

I believe this has long been fixed, but if not this definitely did it:
7590b31