inference-engine
inference-engine copied to clipboard
Work around ifx automatic-offloading bugs
This PR works around issues that arise when the following build command is used to tell the ifx compiler to automatically offload do concurrent constructs to a GPU:
fpm test --compiler ifx --flag "-coarray -coarray-num-images=1 -fopenmp-target-do-concurrent -qopenmp -fopenmp-targets=spir64"
The workarounds are
- further reduce iterations in
trainable_engine_test_mso training converges withifxwithout crashing - convert
do concurrentwith array statements todoloops