histo_src/alternates/histo_exstack2_goto__c.c, line 55: error: "goto *expr" is nonstandard

[jeffhammond]

I do not understand abused C well enough to figure out what this code is doing but the NVIDIA (nee PGI) C compiler rejects it (GCC accepts it).

I have already filed a bug against NVC to request support for this non-standard usage but perhaps Bale could also avoid the apparently non-standard usage.

jehammond@dgx-a100-math:~/SHMEM/bale/src/bale_classic/build_unknown/apps$ /opt/nvidia/hpc_sdk/Linux_x86_64/21.5/comm_libs/openmpi4/openmpi-4.0.5/bin/oshcc -DHAVE_SHMEM_FREE=1 -I. -I/local/jehammond/SHMEM/bale/src/bale_classic/apps   -I/local/jehammond/SHMEM/bale/src/bale_classic/build_unknown/include -I/local/jehammond/SHMEM/bale/src/bale_classic/apps/histo_src -DUSE_SHMEM   -g -O2 -std=c11 -c histo_src/alternates/histo_exstack2_goto__c.c  -o histo_src/alternates/histo-histo_exstack2_goto__c.o 
"histo_src/alternates/histo_exstack2_goto__c.c", line 55: error: "goto *expr" is nonstandard
       goto *loop_ptr;
       ^

"histo_src/alternates/histo_exstack2_goto__c.c", line 31: warning: variable "ret" was declared but never referenced
    int ret;
        ^

"histo_src/alternates/histo_exstack2_goto__c.c", line 33: warning: variable "idxp" was declared but never referenced
    int64_t pe, col, idx, *idxp;
                           ^

1 error detected in the compilation of "histo_src/alternates/histo_exstack2_goto__c.c".

[PRMerkey]

Hi Jeff,

This is on me.
The story for this version was an attempt to look at different
loop designs to help with resetting state and hiding the "end game".

The idea was that being able to jump back into the middle
of the for loop would allow you to hold any state you were
using in the push process. Then we you got done pushing,
hit the bottom of the for loop, you would stay in the end game
until you skip the bottom goto.

I thought that if this looping design had any traction,
then we could think about some macros or something else
to hide the gotos. It got no traction.
I left it in the alternates directory as a record
of things we tried. I obviously didn't spend enough time
cleaning it up. I am embarrassed that there were
unused variables. I don't understand the pointers
to labels so my use of them was probably a cut and paste
job from stackoverflow.

I will look into this some more just to see if I can find
a spec compliant way to do it.
Your take-away might be that this was basically a failed
attempt at saving state and hiding the end game.
Fixing this syntax is less important than finding
a good way to do those things.

Thank you for interest and all your work on Bale.

Phil

[jeffhammond]

Thanks Phil. If this is in the alternatives directory because it's not essential, that's good enough for me. As. you know, this repo is recommended for investigation by HPC vendors but doesn't always come with detailed instructions on what to study 😄

I filed the bug asking for the GCC extension you are using to be supported in NVC and they were not enthusiastic about fixing it, but if it's reasonable to skip this implementation altogether, that's a win for everybody.