Re: Some observations (results) on BPF acquire and release

On Wed, Nov 13, 2024 at 8:25 PM Andrea Parri <[email protected]> wrote:
>
> [...]
>
> > I guess the next question (once clarified the intentions for the R
> > and Z6.3 tests seen earlier) is "Does BPF really care about 2+2W
> > and B-cumulativity for store-release?"; I mentioned some tradeoff,
> > but in the end this is a call for the BPF community.
>
> Interpreting the radio silence as an unanimous "No, it doesn't", please find
> tentative fixes/patch (on top of the bpf_acquire_release branch cited in an
> earlier post) at the bottom of this email.
>
> While testing the changes in question, I noticed an (unrelated) omission in
> the current PPO relation; the second patch below addresses that.
>
> Both patches were tested using the "BPF catalogue" available in the tree at
> stake: as expected, the only differences in outcomes were for the new/added
> five tests.
>
> Please use and integrate according to your preference, any feedback welcome.

Hi Andrea

Sorry for the silence and a huge thanks for working on this and fixing
the cat file.

I have applied your patches and modified them to add the new tests to
kinds.txt and shelf.py
now these tests will run with all other tests using 'make cata-bpf-test'

All 175 tests, including new tests added by you, pass :D

make cata-bpf-test

_build/default/internal/herd_catalogue_regression_test.exe \
        -j 32 \
        -herd-timeout 16.0 \
        -herd-path _build/install/default/bin/herd7 \
        -libdir-path ./herd/libdir \
        -kinds-path catalogue/bpf/tests/kinds.txt \
        -shelf-path catalogue/bpf/shelf.py \
        test
herd7 catalogue bpf tests: OK


I have pushed it and sent a PR so we can get all this merged to master.
https://github.com/herd/herdtools7/pull/1050

P.S. - I am writing this using gmail so the formatting might be
broken, sorry for that.

Thanks,
Puranjay Mohan