Increased Compile time with function inlining


We have a relatively large fortran/OpenACC codebase that compiling regularly takes about 30mins. (Average of: 31m19.825s)

To increase performance of the code we used the PGI/NVIDIA INLINE LIBRARY. With this we inline ca. 90 functions and subroutines across 24 source files. In the source code they are called from within OpenACC kernels and inlined at 1840 places. This increases the kernel performance quite a bit but increases compile time by an hour to 90 mins. (Average of 90m53.525s) Obviously inlining increases compile time but this amount seems excessive.

We are already in contact with Nvidia Devtecs that help with our project and they have a copy of our code. I will forward this issue directly to them.

pgi/nvidia compiler version 21.3.0
GPU target: Tesla P100