[libcxx-commits] [libcxx] [libcxx] Set `_LIBCPP_HAS_CLOCK_GETTIME` for GPU targets (PR #99243)

[Joseph Huber via libcxx-commits]

jhuber6 wrote:

> In general we like to have a CI to validate the changes. These changes will fail for several tests so they need to be excluded from testing. Do you have plans to provide a CI. (This can be discussed further on Discord.)

I don't know much about that myself, but it's difficult to set up a CI before I have a working build configuration, so I figured a new (mostly) NFC patches that get a working build would be a good start.

If https://github.com/llvm/llvm-project/pull/99259, https://github.com/llvm/llvm-project/pull/99243, https://github.com/llvm/llvm-project/pull/99287, and https://github.com/llvm/llvm-project/pull/99333 land I will be able to set up a working build which can then be added on top of the existing NVPTX / AMDGPU buildbots for `libc`.

Getting testing working would then be a separate issue. The problem is that it seems `llvm-lit` doesn't provide an easy way to "emulate" the binaries, but I plan on just doing what I do with the GPU `libc` target. That is described in https://libc.llvm.org/gpu/testing.html, TL;DR I have a program called `amdhsa-loader` and `crt1.o` that will allow us to treat the GPU like a regular target and execute a `main` function.

https://github.com/llvm/llvm-project/pull/99243