fineg74 wrote: > This is way too much for a single PR. The first PR should just set up the > minimum infrastructure required to build the `libc` library for SPIR-V. The > header changes can be easily changed and those are for `offload` so it's a > different category. > > Please use `spirv` instead of `spirv64` everywhere, theoretically this stuff > will all work with 32-bit SPIR-V if the backend and host machine handles it. > Same reason I don't specify `nvptx64` (usually). > > The biggest requirement is a loader that can actually run the tests. Tests > require `quick_exit` and `write_to_stderr` as the minimum set, that should be > the first step. Running the tests requires a loader utility. I have #162739 > which needs to be rebased and merged, but should probably make this easier > once SPIR-V is usable from here. The only reason I've delayed this is because > I don't like how it loads every plugin. I was going to make a change to let > the user specify which backends they want at initialization time but never > got around to it.
I removed whatever I could without making big changes for the building infrastructure. Libc is still a long way before getting ready for testing. There is a bunch of active PRs and I expect several more before saying it is ready for testing. While not optimal, some of the stuff can be tested as part of offloading testing, so hopefully no loader is required for these tests, as I saw some tests failing because they use some libc API https://github.com/llvm/llvm-project/pull/181049 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
