pggpp2-llvm TERMINATED by signal 11

I have some example code that compares different ways to run BLAS at https://github.com/mnicely/computeWorks_examples/tree/master/computeWorks_mm. Everything compiled and executed correctly with CUDA 10.1 + PGI 19.4.

But as soon as I try to compile with PGI 19.10, I get the following error. Any ideas where to begin in solving this issue?

pgc++-Fatal-/opt/pgi/linux86-64-llvm/19.10/bin/pggpp2-llvm TERMINATED by signal 11
Arguments to /opt/pgi/linux86-64-llvm/19.10/bin/pggpp2-llvm
/opt/pgi/linux86-64-llvm/19.10/bin/pggpp2-llvm /tmp/tmpxft_00005126_00000000-5_computeWorks_mm.cudafe1.cpp -opt 2 -terse 1 -inform warn -x 119 0xa10000 -x 122 0x40 -x 123 0x1000 -x 127 4 -x 127 17 -x 19 0x400000 -x 28 0x40000 -x 120 0x10000000 -x 70 0x8000 -x 122 1 -x 125 0x20000 -quad -vect 56 -y 34 16 -x 34 0x8 -x 32 52428800 -y 19 8 -y 35 0 -x 42 0x30 -x 39 0x40 -x 39 0x80 -x 59 4 -x 129 2 -tp haswell -x 120 0x1000 -astype 0 -x 121 1 -fn /tmp/tmpxft_00005126_00000000-5_computeWorks_mm.cudafe1.cpp -il /tmp/pgc++XCfft0JzkBfr.il -x 117 0x200 -x 123 0x80000000 -x 123 4 -x 119 0x20 -def __pgnu_vsn=70400 -x 70 0x40000000 -x 183 4 -x 121 0x800 -x 6 0x20000 -autoinl 10 -x 168 400 -x 174 128000 -x 14 0x200000 -x 14 0x400000 -x 249 70 -x 176 0x100 -cudacap 70 -cudaver 10010 -x 120 0x200000 -x 70 0x40000000 -x 8 0x40000000 -x 164 0x800000 -x 39 4 -x 68 0x1 -accel tesla -x 186 0x80000 -x 180 0x4000400 -x 121 0xc00 -x 194 0x40000 -x 186 0x80 -x 180 0x4000400 -x 121 0xc00 -x 186 0x80 -x 163 0x1 -x 186 0x80000 -cudaver 10010 -x 194 0x40000 -cudaroot /opt/pgi/linux86-64-llvm/2019/cuda/10.1 -x 176 0x100 -cudacap 70 -x 180 0x4000400 -x 121 0xc00 -x 186 0x80 -x 176 0x100 -cudacap 70 -y 163 0xc0000000 -x 189 0x10 -y 189 0x4000000 -cudaroot /opt/pgi/linux86-64-llvm/2019/cuda/10.1 -mp -x 69 0x200 -x 69 0x400 -x 69 2 -x 137 1 -x 0 0x1000000 -x 2 0x100000 -x 0 0x2000000 -x 161 16384 -x 162 16384 -x 9 1 -x 72 0x1 -x 136 0x11 -x 37 0x481000 -gnuvsn 70400 -x 69 0x200 -x 123 0x400 -x 118 0x2 -cmdline '+pgc++ /tmp/pgc++XCfft0JzkBfr.il --nvcchost -cppsuffix=cpp -V19.10 -m64 -D__CUDA_ARCH__=700 -c -DCUDA_DOUBLE_MATH_FUNCTIONS -Bstatic_pgi -acc -mp -Mcuda -Minfo=accel -ta=tesla:nordc -ta=time -O2 -Mvect=simd -I/usr/local/cuda-10.1/bin/../targets/x86_64-linux/include --no_preincludes -o /tmp/tmpxft_00005126_00000000-10_computeWorks_mm.o' -asm /tmp/pgc++bCffdmvx7CXg.ll
makefile:20: recipe for target 'computeWorks_mm' failed
make: *** [computeWorks_mm] Error 127

Hi nmicely.

This looks related to an issue that I found a few weeks ago, TPR #27934, where the combination of using “-ta=tesla:nordc” and “-Mcuda” causes the LLVM compiler to error. Though in case it’s a separate issue, I added a new problem report, TPR #27983, and sent it to engineering for investigation.

The error does not occur with our noLLVM compilers, so one work around is to use “/opt/pgi/linux86-64-nollvm/19.10/bin/pgc++”.

Alternatively, if you split out the OpenACC call to a separate cpp file, compile this file with pgc++ (without nordc), then link with pgc++ (again without nordc and add -lrt), the code will compile and run correctly.

Thanks for the report,
Mat

That worked perfect! Thanks