Cory Bloor

> If using OpenMP requires additional flags with some (broken?) toolchains, these additional flags shouldn't be hard-coded into the build rules imho. Instead a user (with a broken toolchain?) should...

@kiritigowda, thanks for the information. We're working around the lack of libffts for the moment by building with audio support disabled. I'm not sure if Debian will package FFTS separately...

> Ooura doesn't seem to be separately packaged by any distros, but seems to be vendored quite frequently (e.g., in Chromium). Two hours after I said this, it was no...

When I added the rocsolver dependency on fmtlib for `fmt::format` and `fmt::print` in #236, the `std::format` function was recently standardized but not yet implemented in the standard library and `std::print`...

> This shouldn't auto merge, you aren't building on Linux or Windows the targets you expected, use "gfxall" label Thanks for the tip on `gfxall`. You're right that the CI...

There is a bug in `rocm_check_target_ids`, so making the target optional would require https://github.com/ROCm/rocm-cmake/pull/265.

The gfx11-generic target requires `-mcode-object-version=6` for amdclang++ older than ROCm 6.4. Using COV6 with versions of clang from before [this December commit](https://github.com/ROCm/llvm-project/commit/d3eebcd1eab4b6694772a901d6d01c032874e0ca) will result in a warning that "code object...

The [User Guide for AMDGPU Backend](https://llvm.org/docs/AMDGPUUsage.html) has some excellent information about the generic targets. For gfx12-generic, there should be no drawbacks. For gfx11-generic, there are some instruction restrictions that may...

Honestly, that whole section should probably be deleted.

> I know. But it's still wierd that a gpu wich worked perfectly fine with pytorch compiled for an older romc version + HSA_OVERRIDE_GFX_VERSION=10.3.0 (even if not officially supported) suddently...