Yu-Hsiang M. Tsai comments

Results 177 comments of


                                            Yu-Hsiang M. Tsai

Building branch glu_experimental fails with default options

need `experimental::` before `factorization`, I think? `gko::experimental::factorization::Glu`

Easy to mismatch residual norm criteria and actual norm used by solver

Any solver can use the ResidualNorm as the stop criterion. If the solver does not have it during solving, ResidualNorm will introduce an additional spmv and norm as requests. Thus,...

generate the test pipeline if the build job require it with specific runner

Actually, a unified build is not too hard because gitlab provides CI_RUNNER_TAGS. Thus, we can select the slurm script when CI_RUNNER_TAGS contains horeka or slurm or normal scrip otherwise. However,...

generate the test pipeline if the build job require it with specific runner

> Would it not be better to have the variables from the job itself ? I dont see a reason why the additional complexity of first filling a .yml file...

generate the test pipeline if the build job require it with specific runner

Generating the job is based on which runner picks the job up (runtime). Thus, it can be some jobs on horeka, some on amdci, and some on other machines. generate_test_pipeline...

generate the test pipeline if the build job require it with specific runner

okay, so you would like to have the pre-generated job yml. we still need some runtime variable replacement due to the `USE_NAME`, which is related to the build_job pipeline id....

generate the test pipeline if the build job require it with specific runner

@tcojean Yes, it is easy for merging them. However, the bill or limitation of computing resources is the issue to avoid the merging way when I ask for this usage....

generate the test pipeline if the build job require it with specific runner

If we do not have limit on CI resources from the slurm system, unified build is easy and good as the second point I mentioned. Unfortunately, we have limited resources...

generate the test pipeline if the build job require it with specific runner

It is considered from the build time and test time. From the current develop pipeline https://gitlab.com/ginkgo-project/ginkgo-public-ci/-/pipelines/986527225 with mpi, build: 27min, test: 58 min -> require 46% more if we combine...

generate the test pipeline if the build job require it with specific runner

I know it's not the real case for combining build/test on computing nodes, but it is the only real number we have for guessing the situation. As I said, it...