rocmPackages: use CMake 3.9.2 when building clr

[GZGavinZhao]

Fixes #305641.

Description of changes

Ensure CMake 3.9.2 is used to build clr, both for ROCm 5 and 6. The changes can be reverted once CMake 3.9.2 lands in unstable, if desired.

I have tested that rocsparse (5 and 6), rocfft (6), and composable_kernel (6) all build.

See #305641 for more details.

cc @mschwaig

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandboxing enabled in nix.conf? (See Nix manual)
  • sandbox = relaxed
  • sandbox = true
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 24.05 Release Notes (or backporting 23.05 and 23.11 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
• Fits CONTRIBUTING.md.

---

Add a 👍 reaction to pull requests you find important.

[mschwaig]

I have reviewed the code and it looks good to me.
The version hashes match the ones from the update to 3.9.2 at #304111, and I have tested that the downloads for the updated source work.

I'm a bit surprised it was really only necessary to change the CMake version used by clr and not anywhere else, but your testing seems to confirm that.

That's enough for me to approve this change.

I am currently still running nixpkgs-review, to check if this change fixes all of the broken packages, and confirm what you saw in your testing.

Since the build is broken on master anyways, if anyone decides they want to just merge this and not wait for nixpkgs-review to finish, go ahead.

[mschwaig]

Result of nixpkgs-review pr 306375 run on x86_64-linux 1

12 packages marked as broken and skipped:

• rocmPackages.llvm.flang
• rocmPackages.llvm.flang.doc
• rocmPackages.llvm.flang.info
• rocmPackages.llvm.flang.man
• rocmPackages.migraphx
• rocmPackages.mivisionx
• rocmPackages.mivisionx-cpu
• rocmPackages.mivisionx-hip
• rocmPackages_5.llvm.flang
• rocmPackages_5.llvm.flang.doc
• rocmPackages_5.llvm.flang.info
• rocmPackages_5.llvm.flang.man

1 package failed to build:

• rocmPackages_5.rocprofiler

82 packages built:

• blender-hip
• magma (magma-hip ,magma_2_7_2)
• magma.test (magma-hip.test ,magma_2_7_2.test)
• magma_2_6_2
• magma_2_6_2.test
• opensyclWithRocm
• python311Packages.torchWithRocm
• python311Packages.torchWithRocm.cxxdev
• python311Packages.torchWithRocm.dev
• python311Packages.torchWithRocm.dist
• python311Packages.torchWithRocm.lib
• python312Packages.torchWithRocm
• python312Packages.torchWithRocm.cxxdev
• python312Packages.torchWithRocm.dev
• python312Packages.torchWithRocm.dist
• python312Packages.torchWithRocm.lib
• rocmPackages.clr
• rocmPackages.clr.icd
• rocmPackages.composable_kernel
• rocmPackages.hipblas
• rocmPackages.hipcub
• rocmPackages.hipfft
• rocmPackages.hiprand
• rocmPackages.hipsolver
• rocmPackages.hipsparse
• rocmPackages.llvm.mlir
• rocmPackages.miopen
• rocmPackages.rccl
• rocmPackages.rocalution
• rocmPackages.rocblas
• rocmPackages.rocfft
• rocmPackages.rocmlir
• rocmPackages.rocmlir-rock
• rocmPackages.rocmlir.external
• rocmPackages.rocprim
• rocmPackages.rocprofiler
• rocmPackages.rocr-debug-agent
• rocmPackages.rocrand
• rocmPackages.rocsolver
• rocmPackages.rocsparse
• rocmPackages.rocthrust
• rocmPackages.roctracer
• rocmPackages.rocwmma
• rocmPackages.rpp (rocmPackages.rpp-hip)
• rocmPackages.rpp-cpu
• rocmPackages.rpp-opencl
• rocmPackages_5.clr
• rocmPackages_5.clr.icd
• rocmPackages_5.composable_kernel
• rocmPackages_5.hipblas
• rocmPackages_5.hipcub
• rocmPackages_5.hipfft
• rocmPackages_5.hiprand
• rocmPackages_5.hipsolver
• rocmPackages_5.hipsparse
• rocmPackages_5.llvm.mlir
• rocmPackages_5.migraphx
• rocmPackages_5.miopen (rocmPackages_5.miopen-hip)
• rocmPackages_5.miopen-opencl
• rocmPackages_5.miopengemm
• rocmPackages_5.miopengemm.doc
• rocmPackages_5.mivisionx (rocmPackages_5.mivisionx-hip)
• rocmPackages_5.mivisionx-cpu
• rocmPackages_5.mivisionx-opencl
• rocmPackages_5.rccl
• rocmPackages_5.rocalution
• rocmPackages_5.rocblas
• rocmPackages_5.rocfft
• rocmPackages_5.rocmlir
• rocmPackages_5.rocmlir-rock
• rocmPackages_5.rocmlir.external
• rocmPackages_5.rocprim
• rocmPackages_5.rocr-debug-agent
• rocmPackages_5.rocsolver
• rocmPackages_5.rocsparse
• rocmPackages_5.rocthrust
• rocmPackages_5.roctracer
• rocmPackages_5.rocwmma
• rocmPackages_5.rpp (rocmPackages_5.rpp-hip)
• rocmPackages_5.rpp-cpu
• rocmPackages_5.rpp-opencl
• zluda

I have some unknown issue with the rocprofiler being flaky locally, so this is as good as it gets. LGTM.

[nixos-discourse]

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/prs-ready-for-review/3032/3826

[mschwaig]

@ulrikstrid do you have time to take a look at this?

I just mentioned this PR on Discourse as well, to spread the review load a bit/in case you are busy.
Thanks for all the reviews.

[ulrikstrid]

I'm okay with this since it unblocks rocm until the cmake update has landed in master

[mschwaig]

I'm okay with this since it unblocks rocm until the cmake update has landed in master

Great, thanks.

[jonas-w]

Seems like this can be reverted, as 3.29.2 is in unstable:

https://nixpk.gs/pr-tracker.html?pr=304111