rocmPackages: 6.3.3 -> 6.4.3 #427944

LunNova · 2025-07-24T02:19:36Z

Fixes #412640
Fixes #418825
Fixes #434280

TODO

Fix torch segfault in pythonImportsCheckPhase
~~Test gfx11-generic handles strix halo if anyone is available to test with one of the new ryzen AI max boxes~~
~~Test gfx9-generic handles early vega~~
Fix aqlprofile load error in rocprofiler (ok to leave to followup if difficult, it's even more broken in 6.3.3)
Turn aotriton preinstalled triton patch into an upstream PR - CMakeLists.txt: AOTRITON_INHERIT_SYSTEM_SITE_TRITON flag ROCm/aotriton#116
- This was separated into rocmPackages.aotriton: 0.9.2b -> 0.10b #439056 as there was no need to apply the aotriton bump at the same time as the ROCm version bump
Turn rccl HW ID and -generic build fix into an upstream issue - [Issue]: RCCL HWID accessor defines break for -generic targets ROCm/rccl#1862
Retest after rebasing on top of recent aotriton, rocm-core and rocminfo changes and applying 6.4.3 bump

Things done

Add a 👍 reaction to pull requests you find important.

pkgs/development/rocm-modules/6/aqlprofile/default.nix

pkgs/development/rocm-modules/6/llvm/default.nix

pkgs/development/rocm-modules/6/amdsmi/default.nix

pkgs/development/rocm-modules/6/aqlprofile/default.nix

pkgs/development/rocm-modules/6/composable_kernel/base.nix

emilazy · 2025-09-14T22:31:11Z

pkgs/development/rocm-modules/6/default.nix

+      rocm-core = self.callPackage ./rocm-core {
+        inherit stdenv;
      };


Shouldn’t need to inherit it if it’s in the scope. (Applies to a bunch of stuff in this file.)

so this actually changes the stdenv because this is inheriting the original normal stdenv from pkgs from outside the package set. I need to do some renaming.

renamed but it's still not very clear.

pkgs/development/rocm-modules/6/hipblas/default.nix

pkgs/development/rocm-modules/6/llvm/default.nix

LunNova · 2025-09-15T00:14:42Z

pkgs/development/rocm-modules/6/clr/default.nix

+      # "950" #  MI350X TODO: Expected in ROCm 7.x
+      # "10-1-generic" # fine for all RDNA1 cards
      "1010"
-      "1012"


It was an error for us to build 1012 by default in previous versions, our ISA compat patches allow 1010 to load for 1012 so this just wasted resources.

gfx1151 support in torch was fixed by the ROCm 6.4 bump

… supported targets are set Allows building torchWithRocm with clr's localGpuTargets set to gfx906 and other arches which composable_kernel does not support.

…lly bad

Will be required for CMake 4 bump which is hopefully happening in nixpkgs soon

LunNova · 2025-09-15T00:52:08Z

pkgs/development/rocm-modules/6/migraphx/msgpack-6-compat.patch

@emilazy sadly migraphx still needs a patch. going to try to upstream this.

LunNova · 2025-09-15T05:44:22Z

llama-cpp failure to investigate:

In file included from /build/source/ggml/src/ggml-cuda/fattn-tile.cu:1:
/build/source/ggml/src/ggml-cuda/common.cuh:559:18: error: instruction not supported on this GPU
  559 |     asm volatile("v_dot2_f32_f16 %0, %1, %2, %0" : "+v"(acc) : "v"(v), "v"(u));
      |                  ^
<inline asm>:1:2: note: instantiated into assembly here
    1 |         v_dot2_f32_f16 v8, v103, v99, v8

unsupported insn in a very recently added llama-cpp feature ggml-org/llama.cpp#15884

LunNova · 2025-09-15T06:17:07Z

`nixpkgs-review` result

Generated using nixpkgs-review.

Command: nixpkgs-review pr 427944 --checkout commit --additional-package rocmPackages.gfx906.rocm-tests.passthru.tests.torch --additional-package rocmPackages.gfx1100.rocm-tests.passthru.tests.torch
Commit: d844a2f11124be2a736e9ac3eb6fcefefd8012bd

`x86_64-linux`

❌ 1 package failed to build:

llama-cpp-rocm

✅ 137 packages built:

adaptivecpp
adaptivecppWithCuda
adaptivecppWithRocm
blender-hip
btop-rocm
magma (magma-hip)
magma-cuda-static
magma-cuda-static.test
magma.test (magma-hip.test)
ollama-rocm
opensplatWithCuda
python312Packages.torchWithCuda
python312Packages.torchWithCuda.cxxdev
python312Packages.torchWithCuda.dev
python312Packages.torchWithCuda.dist
python312Packages.torchWithCuda.lib
python312Packages.torchWithRocm
python312Packages.torchWithRocm.cxxdev
python312Packages.torchWithRocm.dev
python312Packages.torchWithRocm.dist
python312Packages.torchWithRocm.lib
python313Packages.torchWithCuda
python313Packages.torchWithCuda.cxxdev
python313Packages.torchWithCuda.dev
python313Packages.torchWithCuda.dist
python313Packages.torchWithCuda.lib
python313Packages.torchWithRocm
python313Packages.torchWithRocm.cxxdev
python313Packages.torchWithRocm.dev
python313Packages.torchWithRocm.dist
python313Packages.torchWithRocm.lib
rocmPackages.amdsmi (rocmPackages_6.amdsmi)
rocmPackages.aotriton (rocmPackages_6.aotriton)
rocmPackages.aotriton.debug (rocmPackages_6.aotriton.debug)
rocmPackages.aqlprofile (rocmPackages_6.aqlprofile)
rocmPackages.ck4inductor (rocmPackages_6.ck4inductor)
rocmPackages.ck4inductor.dist (rocmPackages_6.ck4inductor.dist)
rocmPackages.clang (rocmPackages.llvm.clang, rocmPackages.llvm.rocmcxx, rocmPackages_6.clang, rocmPackages_6.llvm.clang, rocmPackages_6.llvm.rocmcxx)
rocmPackages.clr (rocmPackages_6.clr)
rocmPackages.clr.debug (rocmPackages_6.clr.debug)
rocmPackages.clr.icd (rocmPackages_6.clr.icd)
rocmPackages.composable_kernel (rocmPackages_6.composable_kernel)
rocmPackages.gfx1100.rocm-tests.passthru.tests.torch
rocmPackages.gfx1100.rocm-tests.passthru.tests.torch.cxxdev
rocmPackages.gfx1100.rocm-tests.passthru.tests.torch.dev
rocmPackages.gfx1100.rocm-tests.passthru.tests.torch.dist
rocmPackages.gfx1100.rocm-tests.passthru.tests.torch.lib
rocmPackages.gfx906.rocm-tests.passthru.tests.torch
rocmPackages.gfx906.rocm-tests.passthru.tests.torch.cxxdev
rocmPackages.gfx906.rocm-tests.passthru.tests.torch.dev
rocmPackages.gfx906.rocm-tests.passthru.tests.torch.dist
rocmPackages.gfx906.rocm-tests.passthru.tests.torch.lib
rocmPackages.half (rocmPackages_6.half)
rocmPackages.hip-common (rocmPackages_6.hip-common)
rocmPackages.hipblas (rocmPackages_6.hipblas)
rocmPackages.hipblas-common (rocmPackages_6.hipblas-common)
rocmPackages.hipblaslt (rocmPackages_6.hipblaslt)
rocmPackages.hipblaslt.benchmark (rocmPackages_6.hipblaslt.benchmark)
rocmPackages.hipcc (rocmPackages_6.hipcc)
rocmPackages.hipcub (rocmPackages_6.hipcub)
rocmPackages.hipfft (rocmPackages_6.hipfft)
rocmPackages.hipfort (rocmPackages_6.hipfort)
rocmPackages.hipify (rocmPackages_6.hipify)
rocmPackages.hiprand (rocmPackages_6.hiprand)
rocmPackages.hiprt (rocmPackages_6.hiprt)
rocmPackages.hipsolver (rocmPackages_6.hipsolver)
rocmPackages.hipsparse (rocmPackages_6.hipsparse)
rocmPackages.hsakmt (rocmPackages.rocm-runtime, rocmPackages_6.hsakmt, rocmPackages_6.rocm-runtime)
rocmPackages.hsakmt.debug (rocmPackages.rocm-runtime.debug, rocmPackages_6.hsakmt.debug, rocmPackages_6.rocm-runtime.debug)
rocmPackages.llvm.bintools (rocmPackages_6.llvm.bintools)
rocmPackages.llvm.clang-tools (rocmPackages_6.llvm.clang-tools)
rocmPackages.llvm.clang-unwrapped (rocmPackages_6.llvm.clang-unwrapped)
rocmPackages.llvm.clang-unwrapped.debug (rocmPackages_6.llvm.clang-unwrapped.debug)
rocmPackages.llvm.clang-unwrapped.dev (rocmPackages_6.llvm.clang-unwrapped.dev)
rocmPackages.llvm.clang-unwrapped.lib (rocmPackages_6.llvm.clang-unwrapped.lib)
rocmPackages.llvm.clang-unwrapped.python (rocmPackages_6.llvm.clang-unwrapped.python)
rocmPackages.llvm.compiler-rt (rocmPackages.llvm.compiler-rt-libc, rocmPackages_6.llvm.compiler-rt, rocmPackages_6.llvm.compiler-rt-libc)
rocmPackages.llvm.compiler-rt.dev (rocmPackages.llvm.compiler-rt-libc.dev, rocmPackages_6.llvm.compiler-rt-libc.dev, rocmPackages_6.llvm.compiler-rt.dev)
rocmPackages.llvm.libcxx (rocmPackages_6.llvm.libcxx)
rocmPackages.llvm.libcxx.dev (rocmPackages_6.llvm.libcxx.dev)
rocmPackages.llvm.libunwind (rocmPackages_6.llvm.libunwind)
rocmPackages.llvm.libunwind.dev (rocmPackages_6.llvm.libunwind.dev)
rocmPackages.llvm.lld (rocmPackages_6.llvm.lld)
rocmPackages.llvm.lld.dev (rocmPackages_6.llvm.lld.dev)
rocmPackages.llvm.lld.lib (rocmPackages_6.llvm.lld.lib)
rocmPackages.llvm.llvm (rocmPackages_6.llvm.llvm)
rocmPackages.llvm.llvm.dev (rocmPackages_6.llvm.llvm.dev)
rocmPackages.llvm.llvm.lib (rocmPackages_6.llvm.llvm.lib)
rocmPackages.llvm.llvm.python (rocmPackages_6.llvm.llvm.python)
rocmPackages.openmp (rocmPackages.llvm.openmp, rocmPackages_6.llvm.openmp, rocmPackages_6.openmp)
rocmPackages.openmp.dev (rocmPackages.llvm.openmp.dev, rocmPackages_6.llvm.openmp.dev, rocmPackages_6.openmp.dev)
rocmPackages.rocm-merged-llvm (rocmPackages.llvm.rocm-merged-llvm, rocmPackages_6.llvm.rocm-merged-llvm, rocmPackages_6.rocm-merged-llvm)
rocmPackages.stdenv (rocmPackages.rocmClangStdenv, rocmPackages_6.llvm.rocmClangStdenv, rocmPackages_6.rocmClangStdenv, rocmPackages_6.stdenv)
rocmPackages.migraphx (rocmPackages_6.migraphx)
rocmPackages.miopen (rocmPackages.miopen-hip, rocmPackages_6.miopen, rocmPackages_6.miopen-hip)
rocmPackages.mivisionx (rocmPackages.mivisionx-hip, rocmPackages_6.mivisionx, rocmPackages_6.mivisionx-hip)
rocmPackages.mivisionx-cpu (rocmPackages_6.mivisionx-cpu)
rocmPackages.mpi (rocmPackages.openmpi, rocmPackages_6.mpi, rocmPackages_6.openmpi)
rocmPackages.mpi.dev (rocmPackages.openmpi.dev, rocmPackages_6.mpi.dev, rocmPackages_6.openmpi.dev)
rocmPackages.mpi.man (rocmPackages.openmpi.man, rocmPackages_6.mpi.man, rocmPackages_6.openmpi.man)
rocmPackages.mscclpp (rocmPackages_6.mscclpp)
rocmPackages.rccl (rocmPackages_6.rccl)
rocmPackages.rdc (rocmPackages_6.rdc)
rocmPackages.rdc.doc (rocmPackages_6.rdc.doc)
rocmPackages.rocalution (rocmPackages_6.rocalution)
rocmPackages.rocblas (rocmPackages_6.rocblas)
rocmPackages.rocdbgapi (rocmPackages_6.rocdbgapi)
rocmPackages.rocdbgapi.doc (rocmPackages_6.rocdbgapi.doc)
rocmPackages.rocfft (rocmPackages_6.rocfft)
rocmPackages.rocgdb (rocmPackages_6.rocgdb)
rocmPackages.rocm-bandwidth-test (rocmPackages_6.rocm-bandwidth-test)
rocmPackages.rocm-cmake (rocmPackages_6.rocm-cmake)
rocmPackages.rocm-comgr (rocmPackages_6.rocm-comgr)
rocmPackages.rocm-core (rocmPackages_6.rocm-core)
rocmPackages.rocm-device-libs (rocmPackages_6.rocm-device-libs)
rocmPackages.rocm-docs-core (rocmPackages_6.rocm-docs-core)
rocmPackages.rocm-docs-core.dist (rocmPackages_6.rocm-docs-core.dist)
rocmPackages.rocm-smi (rocmPackages_6.rocm-smi)
rocmPackages.rocm-tests (rocmPackages_6.rocm-tests)
rocmPackages.rocmPath (rocmPackages_6.rocmPath)
rocmPackages.rocminfo (rocmPackages_6.rocminfo)
rocmPackages.rocmlir (rocmPackages.rocmlir-rock, rocmPackages_6.rocmlir, rocmPackages_6.rocmlir-rock)
rocmPackages.rocprim (rocmPackages_6.rocprim)
rocmPackages.rocprofiler (rocmPackages_6.rocprofiler)
rocmPackages.rocprofiler-register (rocmPackages_6.rocprofiler-register)
rocmPackages.rocr-debug-agent (rocmPackages_6.rocr-debug-agent)
rocmPackages.rocrand (rocmPackages_6.rocrand)
rocmPackages.rocsolver (rocmPackages_6.rocsolver)
rocmPackages.rocsparse (rocmPackages_6.rocsparse)
rocmPackages.rocthrust (rocmPackages_6.rocthrust)
rocmPackages.roctracer (rocmPackages_6.roctracer)
rocmPackages.rocwmma (rocmPackages_6.rocwmma)
rocmPackages.rpp (rocmPackages.rpp-hip, rocmPackages_6.rpp, rocmPackages_6.rpp-hip)
rocmPackages.rpp-cpu (rocmPackages_6.rpp-cpu)
rocmPackages.tensile (rocmPackages_6.tensile)
rocmPackages.tensile.dist (rocmPackages_6.tensile.dist)
zluda

Error logs: `x86_64-linux`

llama-cpp-rocm

Job completed: /build/mmq-instance-q5_0-gfx908-4eed84.o
Job completed: /build/mmq-instance-q5_0-gfx90a-4aae84.o
Job completed: /build/mmq-instance-q5_0-gfx942-67d1da.o
Job completed: ggml/src/ggml-hip/CMakeFiles/ggml-hip.dir/__/ggml-cuda/template-instances/mmq-instance-q5_0.cu.o
[191/213] Building HIP object ggml/src/ggml-hip/CMakeFiles/ggml-hip.dir/__/ggml-cuda/template-instances/mmq-instance-q2_k.cu.o
Job completed: /build/mmq-instance-q2_k-gfx1010-671ae6.o
Job completed: /build/mmq-instance-q2_k-gfx1030-3d38cd.o
Job completed: /build/mmq-instance-q2_k-gfx1100-d8e9fa.o
Job completed: /build/mmq-instance-q2_k-gfx1101-e3873e.o
Job completed: /build/mmq-instance-q2_k-gfx1102-f8663d.o
Job completed: /build/mmq-instance-q2_k-gfx1151-92a513.o
Job completed: /build/mmq-instance-q2_k-gfx1200-6c9408.o
Job completed: /build/mmq-instance-q2_k-gfx1201-019565.o
Job completed: /build/mmq-instance-q2_k-gfx900-09a182.o
Job completed: /build/mmq-instance-q2_k-gfx906-4a4bf7.o
Job completed: /build/mmq-instance-q2_k-gfx908-05e6ce.o
Job completed: /build/mmq-instance-q2_k-gfx90a-c5abc9.o
Job completed: /build/mmq-instance-q2_k-gfx942-c35912.o
Job completed: ggml/src/ggml-hip/CMakeFiles/ggml-hip.dir/__/ggml-cuda/template-instances/mmq-instance-q2_k.cu.o
ninja: build stopped: subcommand failed.

LunNova mentioned this pull request Jul 24, 2025

Update Request: rocm 6.3.3 → 6.4.1 #412640

Open

3 tasks

LunNova changed the title ~~WIP 6.4.1 bump~~ rocmPackages: 6.3.3 -> 6.4.x Jul 24, 2025

LunNova mentioned this pull request Jul 24, 2025

rocmPackages: 6.3.3 -> 6.4.x #427942

Closed

LunNova force-pushed the lunnova/rocm-6.4.x branch from e8872b0 to f034276 Compare July 24, 2025 02:30

LunNova commented Jul 24, 2025

View reviewed changes

pkgs/development/rocm-modules/6/aqlprofile/default.nix Outdated Show resolved Hide resolved

This comment was marked as outdated.

Sign in to view

nixpkgs-ci bot added the 2.status: merge conflict This PR has merge conflicts with the target branch label Jul 24, 2025

noebm reviewed Jul 27, 2025

View reviewed changes

pkgs/development/rocm-modules/6/llvm/default.nix Outdated Show resolved Hide resolved

LunNova force-pushed the lunnova/rocm-6.4.x branch from ed91505 to 8d8c252 Compare August 6, 2025 00:16

nixpkgs-ci bot removed the 2.status: merge conflict This PR has merge conflicts with the target branch label Aug 6, 2025

LunNova force-pushed the lunnova/rocm-6.4.x branch from 8d8c252 to 11a7af6 Compare August 6, 2025 01:22

LunNova changed the title ~~rocmPackages: 6.3.3 -> 6.4.x~~ rocmPackages: 6.3.3 -> 6.4.2 Aug 6, 2025

LunNova force-pushed the lunnova/rocm-6.4.x branch 4 times, most recently from 29e7d0c to 2758d52 Compare August 7, 2025 05:15

This comment was marked as outdated.

Sign in to view

LunNova force-pushed the lunnova/rocm-6.4.x branch 3 times, most recently from 8089e29 to 87e341b Compare August 7, 2025 16:17

nixpkgs-ci bot added the 2.status: merge conflict This PR has merge conflicts with the target branch label Aug 7, 2025

LunNova force-pushed the lunnova/rocm-6.4.x branch 3 times, most recently from 6715162 to 8b7c3c1 Compare August 7, 2025 17:00

nixpkgs-ci bot added the 6.topic: python Python is a high-level, general-purpose programming language. label Aug 7, 2025

LunNova force-pushed the lunnova/rocm-6.4.x branch from 8b7c3c1 to 251034d Compare August 8, 2025 00:26

This comment was marked as outdated.

Sign in to view

LunNova force-pushed the lunnova/rocm-6.4.x branch from 286e46c to 29c2885 Compare September 13, 2025 19:13

emilazy reviewed Sep 14, 2025

View reviewed changes

LunNova force-pushed the lunnova/rocm-6.4.x branch 2 times, most recently from 9127b6c to 9c6d9ff Compare September 15, 2025 00:02

LunNova commented Sep 15, 2025

View reviewed changes

LunNova added 10 commits September 14, 2025 17:45

rocmPackages: 6.3.3 -> 6.4.2

b592c8b

magma: set -fopenmp for rocm, use modern rocm c/xx

83d1827

ucx: drop no longer needed rocm stdenv workaround

5e31e13

python3Packages.torchWithRocm: allow gfx1151 (Strix Halo)

3c2cbd0

gfx1151 support in torch was fixed by the ROCm 6.4 bump

rocmPackages: 6.4.2 -> 6.4.3

a9df0e8

rocmPackages.aotriton: correctness and build perf improvements

43911cb

python3Packages.torchWithRocm: use vendored composable_kernel when no…

05ddfcf

… supported targets are set Allows building torchWithRocm with clr's localGpuTargets set to gfx906 and other arches which composable_kernel does not support.

rocmPackages.miopen: note that the broken lfsconfig workaround is rea…

561e8e5

…lly bad

rocmPackages.clr: backport cmake_minimum_required patch

3ac14a6

Will be required for CMake 4 bump which is hopefully happening in nixpkgs soon

llama-cpp: use local target override if set

ec1179b

LunNova force-pushed the lunnova/rocm-6.4.x branch 2 times, most recently from f7734fc to 62a423d Compare September 15, 2025 00:49

LunNova commented Sep 15, 2025

View reviewed changes

LunNova force-pushed the lunnova/rocm-6.4.x branch 2 times, most recently from f5eacbb to ec546ee Compare September 15, 2025 02:07

rocmPackages: msgpack → msgpack-cxx

d844a2f

LunNova force-pushed the lunnova/rocm-6.4.x branch from ec546ee to d844a2f Compare September 15, 2025 02:17

LunNova changed the base branch from master to staging September 15, 2025 02:42

nixpkgs-ci bot closed this Sep 15, 2025

nixpkgs-ci bot reopened this Sep 15, 2025

emilazy approved these changes Sep 15, 2025

View reviewed changes

emilazy merged commit c535ff1 into NixOS:staging Sep 15, 2025
54 of 57 checks passed

LunNova mentioned this pull request Sep 18, 2025

rocmPackages.llvm: use same LLVM version for bootstrap and override #443943

Merged

13 tasks

Uh oh!

rocmPackages: 6.3.3 -> 6.4.3 #427944

rocmPackages: 6.3.3 -> 6.4.3 #427944

Uh oh!

Conversation

LunNova commented Jul 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

TODO

Things done

Uh oh!

Uh oh!

This comment was marked as outdated.

Uh oh!

This comment was marked as outdated.

This comment was marked as outdated.

Uh oh!

Uh oh!

Uh oh!

emilazy Sep 14, 2025

Choose a reason for hiding this comment

Uh oh!

LunNova Sep 14, 2025

Choose a reason for hiding this comment

Uh oh!

LunNova Sep 15, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

LunNova Sep 15, 2025

Choose a reason for hiding this comment

Uh oh!

LunNova Sep 15, 2025

Choose a reason for hiding this comment

Uh oh!

LunNova commented Sep 15, 2025

Uh oh!

LunNova commented Sep 15, 2025

nixpkgs-review result

x86_64-linux

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

12 participants

LunNova commented Jul 24, 2025 •

edited

Loading

`nixpkgs-review` result

`x86_64-linux`