20635, "DanilaFe", "Internal error in 'foreach' loops over empty domains on GPUs", "2022-09-06T19:00:52Z"
When using the prototype GPU support, a loop over an empty domain in
an on here.gpus[0] block leads to a runtime error. The minimal snippet
to reproduce is as follows:
on here.gpus[0] {
foreach i in {1..<1, 1..<1} {}
}
The error looks like this:
internal error: gpu-cuda.c:221: Error calling CUDA function: invalid argument (Code: 1)
Currently, this issue affects the gpu/native/nestedForeach test, which tests
iteration over a multi-dimensional, but empty, domain. When this issue
is resolved, that test should be updated to run the (currently commented
out, and failing) multi-dimensional-but-empty iteration.
Compile command:
chpl nested.chpl
Execution command:
./nested
Configuration Information
- Output of
chpl --version:chpl version 1.28.0 pre-release (76bbdcc00f) built with LLVM version 14.0.0 - Output of
$CHPL_HOME/util/printchplenv --anonymize:CHPL_TARGET_PLATFORM: cray-cs * CHPL_TARGET_COMPILER: llvm CHPL_TARGET_ARCH: x86_64 CHPL_TARGET_CPU: native * CHPL_LOCALE_MODEL: gpu * CHPL_COMM: none * CHPL_TASKS: qthreads CHPL_LAUNCHER: slurm-srun CHPL_TIMERS: generic CHPL_UNWIND: none CHPL_MEM: jemalloc CHPL_ATOMICS: cstdlib CHPL_GMP: bundled CHPL_HWLOC: bundled CHPL_RE2: bundled CHPL_LLVM: bundled CHPL_AUX_FILESYS: none