Re: RFC: Centralized ROCm target list + utility
Hi all,
On 2025-07-03 10:27, Cordell Bloor wrote:
> The term "arch" is used to refer to gfx targets in many places. It is the terminology used by cmake [1], clang [2][3], and gcc [4]
I also looked around, and believe no unambiguous solution exists.
Actually, "arch" might even be the most ambiguous one (eg: I've seen it
used for RDNA3, gfx1100, and gfx1100 all with different meanings).
However, as various other tools already seem to be using, I think we
might as well, too, just for consistency, so I'll upload again with
rocm-target-archs.
We can always change this in future, with not too much pain. I think
it's likely that we'll come up with a better name altogether eventually,
but to get there, we'll need some real-world usage and data first. As
we've done with all our other tooling.
Also, all of this is currently happening in experimental anyway. If we
do things right, we'll manage to iron out all the kinks until unstable
is open for regular development again.
> On 2025-06-23 21:55, Christian Kastner wrote:
>> X-ROCm-GPU-ISA
> I would just suggest that you consider whether it's better to be "ROCm" or "AMD". If a package were built with GCC's support for AMD GPUs [4] (which is not part of ROCm), would you want it to have this tag? If so, I would replace "ROCm" with "AMD".
In that case, I'd stick with "ROCm". We don't have experience with the
GCC use case yet, so I don't think we should be signalling for it at
this point.
As the utility is going by "archs" now, I guess ISA should depart here
as well. I'd use X-ROCm-GPU-Architecture instead of some abbreviation,
as the utility does, because the field is functionally quite similar to
the already established "Architecture" field for CPUs.
Again, this is going to experimental, so we're just aiming for "good"
for now. Let's figure out "perfect" with more data.
Best,
Christian
> [1]: https://cmake.org/cmake/help/v4.1/variable/CMAKE_HIP_ARCHITECTURES.html
> [2]: https://rocm.docs.amd.com/projects/llvm-project/en/docs-6.4.1/conceptual/code-portability.html
> [3]: e.g., --offload-arch=<target-id>, and the amdgpu-arch utility
> [4]: https://gcc.gnu.org/onlinedocs/gcc/AMD-GCN-Options.html
Reply to: