On Mon, Jan 13, 2025 at 07:18:05AM -0700, Jeff Law via Gcc wrote: > > > On 1/13/25 2:56 AM, Stefan Schulze Frielinghaus via Gcc wrote: > > Hi everyone, > > > > In order to better test our s390 builtins, I have been coming up with a > > small tool in order to automatically generate tests. The tool consumes > > s390-builtins.def and generates currently > > > > $ ls -1 | wc -l > > 4703 > > > > many files which consume > > > > $ du -s -h . > > 19M . > > > > space in total. Since these are so many generated files which consume > > in total quite some space, I tend to not check in the files itself but > > would rather like to generate those automatically whenever the s390 > > testsuite is run. Generation time is negligible: > > > > $ time ./foo.py --workdir=/gcc/src --test > > ./foo.py --workdir=/gcc/src --test 0.16s user 0.18s system 96% cpu 0.346 > > total > > > > This has also the benefit that one cannot forget to regenerate the tests > > whenever s390-builtins.def is changed. > > > > The tool is written in python. This in turn would mean that whoever > > wants to run the s390 testsuite also needs python (and a proper version > > of it). I guess this is acceptable as long as I require a python > > version which matches with gcc_cv_prog_python3_modern. > > > > Do we have a similar setup I could try to follow in the sense that I > > need to integrate the generator tool into the testsuite and have it > > generate testcases automatically into some (build) directory which > > dejagnu then sources? Any pointers are highly appreciated. > We've got the ABI compatibility tests that are generated. > > Peek at gcc.dg/compat/struct-layout-1.exp and see if gives you any > insights. The generator in that case is C, but it might give you a > structure to work within.
That looks very promising. > > I'd also suggest wandering the aarch64 testsuite directory to see if it's > got anything generated (particular for builtin testing). I will have a look at those exp files, too. Thanks for the pointers! Cheers, Stefan > > Jeff > > > > Cheers, > > Stefan >
