setting to draft to check if upgrading embedded nvptx-tools does the trick already

after checking with an example
gcc -save-temps -fopenacc -foffload=nvptx-none=-save-temps pi.c
for

#include <stdio.h>

#define N 2000000000

#define vl 1024

int main(void) {

  double pi = 0.0f;
  long long i;

  #pragma acc parallel vector_length(vl) 
  #pragma acc loop reduction(+:pi)
  for (i=0; i<N; i++) {
    double t= (double)((i+0.5)/N);
    pi +=4.0/(1.0+t*t);
  }

  printf("pi=%11.10f\n",pi/N);

  return 0;

}

I found the generated ptxas asm file (pi.s) has .target sm_30 so will cause newer nvptx-tools to pass that to ptxas thereby not solving the issue. So I'll keep this change as is, as upgrading nvptx-tools is neither sufficient nor necessary to fix the reported issue.

@boegelbot please test @ generoso

@bartoldeman: Request for testing this PR well received on generoso

PR test command 'EB_PR=13290 EB_ARGS= /apps/slurm/default/bin/sbatch --job-name test_PR_13290 --ntasks=4 ~/boegelbot/eb_from_pr_upload_generoso.sh' executed!

• exit code: 0
• output:

Submitted batch job 17653

Test results coming soon (I hope)...

@bartoldeman my attempted fix for the problem based on your other comment (using a newer version of nvptx-tools) failed in a build on a node with CUDA enabled. I will try this one instead.

@verdurin thanks for confirming, as this is expected "as upgrading nvptx-tools is neither sufficient nor necessary to fix the reported issue.", see above.

@boegelbot please test @ generoso
CORE_CNT=16

@bartoldeman Building 3 GCCcore easyconfigs using only 4 cores is going to take forever, so I've cancelled the original test and requested a new one using 16 cores.

@boegel: Request for testing this PR well received on generoso

PR test command 'EB_PR=13290 EB_ARGS= /apps/slurm/default/bin/sbatch --job-name test_PR_13290 --ntasks="16" ~/boegelbot/eb_from_pr_upload_generoso.sh' executed!

• exit code: 0
• output:

Submitted batch job 17655

Test results coming soon (I hope)...

Test build of GCCcore-9.3.0on CUDA-enabled node worked for me (it's not setup for test report upload).

Test report by @boegel
SUCCESS
Build succeeded for 3 out of 3 (3 easyconfigs in total)
node3302.joltik.os - Linux centos linux 7.9.2009, x86_64, Intel(R) Xeon(R) Gold 6242 CPU @ 2.80GHz (cascadelake), Python 3.6.8
See https://gist.github.com/0830592c41013fdf51332ae9438850a9 for a full test report.

worked for me with GCC 10.2.
to reproduce at runtime instead of build-time you just need to load a CUDA 11 module and then
gcc -save-temps -fopenacc -foffload=nvptx-none pi.c
for the above example.

$ gcc -fopenacc -foffload=nvptx-none pi.c
<no output -- correct>

# with unmodified GCCcore:
$ gcc -fopenacc -foffload=nvptx-none pi.c
ptxas fatal   : Value 'sm_30' is not defined for option 'gpu-name'
nvptx-as: ptxas returned 255 exit status
mkoffload: fatal error: x86_64-pc-linux-gnu-accel-nvptx-none-gcc returned 1 exit status
compilation terminated.
lto-wrapper: fatal error: /cvmfs/soft.computecanada.ca/easybuild/software/2020/Core/gcccore/9.3.0/libexec/gcc/x86_64-pc-linux-gnu/9.3.0//accel/nvptx-none/mkoffload returned 1 exit status
compilation terminated.
/cvmfs/soft.computecanada.ca/gentoo/2020/usr/x86_64-pc-linux-gnu/binutils-bin/2.33.1/ld.gold: fatal error: lto-wrapper failed
collect2: error: ld returned 1 exit status

Test report by @boegelbot
SUCCESS
Build succeeded for 3 out of 3 (3 easyconfigs in total)
generoso-x-2 - Linux centos linux 8.2.2004, x86_64, Intel(R) Xeon(R) CPU E5-2667 v3 @ 3.20GHz (haswell), Python 3.6.8
See https://gist.github.com/cbbbe1af8ba3977af6e333ce448aec8d for a full test report.

Test report by @bartoldeman
SUCCESS
Build succeeded for 3 out of 3 (3 easyconfigs in total)
build-node.computecanada.ca - Linux centos linux 7.9.2009, x86_64, Intel Xeon Processor (Skylake, IBRS), Python 3.7.7
See https://gist.github.com/4a10ad15f00374ccfd5891190a7a0869 for a full test report.