A comparison of the "Build OCaml" task, sequential versus parallel. According to the docs, runners have 4 CPUs. "cached" means that the configure script results have been cached. Considering these ad-hoc benchmarks of GitHub Actions on my fork, I'm going with the -j option of Make, which will not limit the number of jobs that can run simultaneously.
This is similar to what we already do in AppVeyor for MinGW-w64 builds, although we bound the number of simultaneous jobs with -j$NUMBER_OF_PROCESSORS (equivalent to -j$(nproc) on Linux).
sequential
parallel, -j4
parallel, -j4, cached
parallel, -j, cached
MSVC 64 bits
13m 26s
6m 21s
6m 19s
6m 5s
clang-cl 64 bits
14m 58s
8m 17s
7m 21s
7m 4s
MSVC 32 bits
8m 28s
4m 31s
4m 35s
3m 56s
This PR currently sits on my previous PR enabling the Autoconf cache. I'll rebase once it's merged.
A comparison of the "Build OCaml" task, sequential versus parallel. According to the docs, runners have 4 CPUs. "cached" means that the configure script results have been cached. Considering these ad-hoc benchmarks of GitHub Actions on my fork, I'm going with the
-joption of Make, which will not limit the number of jobs that can run simultaneously.This is similar to what we already do in AppVeyor for MinGW-w64 builds, although we bound the number of simultaneous jobs with
-j$NUMBER_OF_PROCESSORS(equivalent to-j$(nproc)on Linux).-j4-j4, cached-j, cachedThis PR currently sits on my previous PR enabling the Autoconf cache. I'll rebase once it's merged.