NixOS · prusnak · Mar 24, 2025 · Jan 4, 2025 · Jan 4, 2025 · Feb 26, 2025
diff --git a/nixos/doc/manual/release-notes/rl-2505.section.md b/nixos/doc/manual/release-notes/rl-2505.section.md
@@ -380,6 +380,10 @@
 
 - `python3Packages.jaeger-client` was removed because it was deprecated upstream. [OpenTelemetry](https://opentelemetry.io) is the recommended replacement.
 
+- `rocmPackages.rocm-thunk` has been removed and its functionality has been integrated with the ROCm CLR. Use `rocmPackages.clr` instead.
+
+- `rocmPackages.clang-ocl` has been removed. [It was deprecated by AMD in 2023.](https://github.com/ROCm/clang-ocl)
+
 - `nodePackages.meshcommander` has been removed, as the package was deprecated by Intel.
 
 - The default version of `z3` has been updated from 4.8 to 4.13. There are still a few packages that need specific older versions; those will continue to be maintained as long as other packages depend on them but may be removed in the future.

diff --git a/pkgs/by-name/ll/llama-cpp/package.nix b/pkgs/by-name/ll/llama-cpp/package.nix
@@ -152,16 +152,11 @@ effectiveStdenv.mkDerivation (finalAttrs: {
     ++ optionals cudaSupport [
       (cmakeFeature "CMAKE_CUDA_ARCHITECTURES" cudaPackages.flags.cmakeCudaArchitecturesString)
     ]
-    ++ optionals rocmSupport [
-      (cmakeFeature "CMAKE_C_COMPILER" "hipcc")
-      (cmakeFeature "CMAKE_CXX_COMPILER" "hipcc")
-
-      # Build all targets supported by rocBLAS. When updating search for TARGET_LIST_ROCM
-      # in https://github.com/ROCmSoftwarePlatform/rocBLAS/blob/develop/CMakeLists.txt
-      # and select the line that matches the current nixpkgs version of rocBLAS.
-      # Should likely use `rocmPackages.clr.gpuTargets`.
-      "-DAMDGPU_TARGETS=gfx803;gfx900;gfx906:xnack-;gfx908:xnack-;gfx90a:xnack+;gfx90a:xnack-;gfx940;gfx941;gfx942;gfx1010;gfx1012;gfx1030;gfx1100;gfx1101;gfx1102"
-    ]
+    ++ optionals rocmSupport ([
+      (cmakeFeature "CMAKE_HIP_COMPILER" "${rocmPackages.clr.hipClangPath}/clang++")
+      # TODO: this should become `clr.gpuTargets` in the future.
+      (cmakeFeature "CMAKE_HIP_ARCHITECTURES" rocmPackages.rocblas.amdgpu_targets)
+    ])
     ++ optionals metalSupport [
       (cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1")
       (cmakeBool "LLAMA_METAL_EMBED_LIBRARY" true)

diff --git a/pkgs/by-name/ol/ollama/package.nix b/pkgs/by-name/ol/ollama/package.nix
@@ -72,17 +72,17 @@ let
 
   rocmLibs = [
     rocmPackages.clr
+    rocmPackages.hipblas-common
     rocmPackages.hipblas
     rocmPackages.rocblas
     rocmPackages.rocsolver
     rocmPackages.rocsparse
     rocmPackages.rocm-device-libs
     rocmPackages.rocm-smi
   ];
-  rocmClang = linkFarm "rocm-clang" { llvm = rocmPackages.llvm.clang; };
   rocmPath = buildEnv {
     name = "rocm-path";
-    paths = rocmLibs ++ [ rocmClang ];
+    paths = rocmLibs;
   };
 
   cudaLibs = [
@@ -133,6 +133,8 @@ let
   goBuild =
     if enableCuda then
       buildGoModule.override { stdenv = cudaPackages.backendStdenv; }
+    else if enableRocm then
+      buildGoModule.override { stdenv = rocmPackages.stdenv; }
     else
       buildGoModule;
   inherit (lib) licenses platforms maintainers;
@@ -150,6 +152,14 @@ goBuild {
       ROCM_PATH = rocmPath;
       CLBlast_DIR = "${clblast}/lib/cmake/CLBlast";
       HIP_PATH = rocmPath;
+      CFLAGS = "-Wno-c++17-extensions -I${rocmPath}/include";
+      CXXFLAGS = "-Wno-c++17-extensions -I${rocmPath}/include";
+    }
+    // lib.optionalAttrs (enableRocm && (rocmPackages.clr.localGpuTargets or false)) {
+
+      # If rocm CLR is set to build for an exact set of targets reuse that target list,
+      # otherwise let ollama use its builtin defaults
+      HIP_ARCHS = lib.concatStringsSep ";" rocmPackages.clr.localGpuTargets;
     }
     // lib.optionalAttrs enableCuda { CUDA_PATH = cudaPath; };
 

diff --git a/pkgs/by-name/uc/ucx/package.nix b/pkgs/by-name/uc/ucx/package.nix
@@ -33,8 +33,10 @@ let
     paths = rocmList;
   };
 
+  # rocm build fails with gcc stdenv due to unrecognised arg parallel-jobs
+  stdenv' = if enableRocm then rocmPackages.stdenv else stdenv;
 in
-stdenv.mkDerivation rec {
+stdenv'.mkDerivation rec {
   pname = "ucx";
   version = "1.18.0";
 

diff --git a/pkgs/development/libraries/science/math/magma/generic.nix b/pkgs/development/libraries/science/math/magma/generic.nix
@@ -117,18 +117,11 @@ stdenv.mkDerivation {
     "test"
   ];
 
-  patches = lib.optionals (version == "2.9.0") [
-    # get ROCm version directly
-    # https://github.com/icl-utk-edu/magma/pull/27
-    (fetchpatch {
-      url = "https://github.com/icl-utk-edu/magma/commit/10fe816b763c41099fa1c978a79d6869246671cf.patch";
-      hash = "sha256-qSY5ACMHyHofJdQKyPqx8sI8GbPD6IZezmCd8qOS5OM=";
-    })
-  ];
-
-  # Fixup for the python test runners
   postPatch =
     ''
+      # For rocm version script invoked by cmake
+      patchShebangs tools/
+      # Fixup for the python test runners
       patchShebangs ./testing/run_{tests,summarize}.py
     ''
     + lib.optionalString (strings.versionOlder version "2.9.0") ''
@@ -201,6 +194,10 @@ stdenv.mkDerivation {
       (strings.cmakeFeature "MIN_ARCH" minArch) # Disarms magma's asserts
     ]
     ++ lists.optionals rocmSupport [
+      # Can be removed once https://github.com/icl-utk-edu/magma/pull/27 is merged
+      # Can't easily apply the PR as a patch because we rely on the tarball with pregenerated
+      # hipified files ∴ fetchpatch of the PR will apply cleanly but fail to build
+      (strings.cmakeFeature "ROCM_CORE" "${effectiveRocmPackages.clr}")
       (strings.cmakeFeature "CMAKE_C_COMPILER" "${effectiveRocmPackages.clr}/bin/hipcc")
       (strings.cmakeFeature "CMAKE_CXX_COMPILER" "${effectiveRocmPackages.clr}/bin/hipcc")
     ];

diff --git a/pkgs/development/python-modules/accelerate/default.nix b/pkgs/development/python-modules/accelerate/default.nix
@@ -92,8 +92,9 @@ buildPythonPackage rec {
       # set the environment variable, CC, which conflicts with standard environment
       "test_patch_environment_key_exists"
     ]
-    ++ lib.optionals (pythonAtLeast "3.13") [
+    ++ lib.optionals ((pythonAtLeast "3.13") || (torch.rocmSupport or false)) [
       # RuntimeError: Dynamo is not supported on Python 3.13+
+      # OR torch.compile tests broken on torch 2.5 + rocm
       "test_can_unwrap_distributed_compiled_model_keep_torch_compile"
       "test_can_unwrap_distributed_compiled_model_remove_torch_compile"
       "test_convert_to_fp32"

diff --git a/pkgs/development/python-modules/torch/default.nix b/pkgs/development/python-modules/torch/default.nix
@@ -36,6 +36,7 @@
   symlinkJoin,
   which,
   pybind11,
+  pkg-config,
   removeReferencesTo,
 
   # Build inputs
@@ -66,7 +67,13 @@
   #          (dependencies without cuda support).
   #          Instead we should rely on overlays and nixpkgsFun.
   # (@SomeoneSerge)
-  _tritonEffective ? if cudaSupport then triton-cuda else triton,
+  _tritonEffective ?
+    if cudaSupport then
+      triton-cuda
+    else if rocmSupport then
+      rocmPackages.triton
+    else
+      triton,
   triton-cuda,
 
   # Unit tests
@@ -92,7 +99,7 @@
 
   # ROCm dependencies
   rocmSupport ? config.rocmSupport,
-  rocmPackages_5,
+  rocmPackages,
   gpuTargets ? [ ],
 
   vulkanSupport ? false,
@@ -112,8 +119,6 @@ let
 
   triton = throw "python3Packages.torch: use _tritonEffective instead of triton to avoid divergence";
 
-  rocmPackages = rocmPackages_5;
-
   setBool = v: if v then "1" else "0";
 
   # https://github.com/pytorch/pytorch/blob/v2.4.0/torch/utils/cpp_extension.py#L1953
@@ -181,7 +186,7 @@ let
       clr
       rccl
       miopen
-      miopengemm
+      aotriton
       rocrand
       rocblas
       rocsparse
@@ -193,10 +198,12 @@ let
       rocfft
       rocsolver
       hipfft
+      hiprand
       hipsolver
+      hipblas-common
       hipblas
+      hipblaslt
       rocminfo
-      rocm-thunk
       rocm-comgr
       rocm-device-libs
       rocm-runtime
@@ -226,8 +233,6 @@ let
     # In particular, this triggered warnings from cuda's `aliases.nix`
     "Magma cudaPackages does not match cudaPackages" =
       cudaSupport && (effectiveMagma.cudaPackages.cudaVersion != cudaPackages.cudaVersion);
-    "Rocm support is currently broken because `rocmPackages.hipblaslt` is unpackaged. (2024-06-09)" =
-      rocmSupport;
   };
 
   unroll-src = writeShellScript "unroll-src" ''
@@ -294,6 +299,11 @@ buildPythonPackage rec {
 
       # annotations (3.7), print_function (3.0), with_statement (2.6) are all supported
       sed -i -e "/from __future__ import/d" **.py
+      substituteInPlace third_party/NNPACK/CMakeLists.txt \
+        --replace-fail "PYTHONPATH=" 'PYTHONPATH=$ENV{PYTHONPATH}:'
+      # flag from cmakeFlags doesn't work, not clear why
+      # setting it at the top of NNPACK's own CMakeLists does
+      sed -i '2s;^;set(PYTHON_SIX_SOURCE_DIR ${six.src})\n;' third_party/NNPACK/CMakeLists.txt
     ''
     + lib.optionalString rocmSupport ''
       # https://github.com/facebookincubator/gloo/pull/297
@@ -366,6 +376,10 @@ buildPythonPackage rec {
   # We only do an imports check, so do not build tests either.
   BUILD_TEST = setBool false;
 
+  # ninja hook doesn't automatically turn on ninja
+  # because pytorch setup.py is responsible for this
+  CMAKE_GENERATOR = "Ninja";
+
   # Unlike MKL, oneDNN (née MKLDNN) is FOSS, so we enable support for
   # it by default. PyTorch currently uses its own vendored version
   # of oneDNN through Intel iDeep.
@@ -376,14 +390,15 @@ buildPythonPackage rec {
   # Also avoids pytorch exporting the headers of pybind11
   USE_SYSTEM_PYBIND11 = true;
 
-  # NB technical debt: building without NNPACK as workaround for missing `six`
-  USE_NNPACK = 0;
+  # Multicore CPU convnet support
+  USE_NNPACK = 1;
 
   # Explicitly enable MPS for Darwin
   USE_MPS = setBool stdenv.hostPlatform.isDarwin;
 
   cmakeFlags =
     [
+      (lib.cmakeFeature "PYTHON_SIX_SOURCE_DIR" "${six.src}")
       # (lib.cmakeBool "CMAKE_FIND_DEBUG_MODE" true)
       (lib.cmakeFeature "CUDAToolkit_VERSION" cudaPackages.cudaVersion)
     ]
@@ -441,6 +456,9 @@ buildPythonPackage rec {
     }
     // lib.optionalAttrs vulkanSupport {
       VULKAN_SDK = shaderc.bin;
+    }
+    // lib.optionalAttrs rocmSupport {
+      AOTRITON_INSTALLED_PREFIX = "${rocmPackages.aotriton}";
     };
 
   nativeBuildInputs =
@@ -449,6 +467,7 @@ buildPythonPackage rec {
       which
       ninja
       pybind11
+      pkg-config
       removeReferencesTo
     ]
     ++ lib.optionals cudaSupport (
@@ -502,7 +521,10 @@ buildPythonPackage rec {
     ]
     ++ lib.optionals tritonSupport [ _tritonEffective ]
     ++ lib.optionals MPISupport [ mpi ]
-    ++ lib.optionals rocmSupport [ rocmtoolkit_joined ];
+    ++ lib.optionals rocmSupport [
+      rocmtoolkit_joined
+      rocmPackages.clr # Added separately so setup hook applies
+    ];
 
   pythonRelaxDeps = [
     "sympy"

diff --git a/pkgs/development/python-modules/torchaudio/default.nix b/pkgs/development/python-modules/torchaudio/default.nix
@@ -36,7 +36,6 @@ let
       clr
       rccl
       miopen
-      miopengemm
       rocrand
       rocblas
       rocsparse
@@ -49,9 +48,9 @@ let
       rocsolver
       hipfft
       hipsolver
+      hipblas-common
       hipblas
       rocminfo
-      rocm-thunk
       rocm-comgr
       rocm-device-libs
       rocm-runtime

diff --git a/pkgs/development/rocm-modules/6/amdsmi/default.nix b/pkgs/development/rocm-modules/6/amdsmi/default.nix
@@ -0,0 +1,89 @@
+{
+  lib,
+  stdenv,
+  fetchFromGitHub,
+  fetchpatch,
+  rocmUpdateScript,
+  cmake,
+  pkg-config,
+  libdrm,
+  wrapPython,
+  autoPatchelfHook,
+}:
+
+let
+  esmi_ib_src = fetchFromGitHub {
+    owner = "amd";
+    repo = "esmi_ib_library";
+    rev = "esmi_pkg_ver-3.0.3";
+    hash = "sha256-q0w5c5c+CpXkklmSyfzc+sbkt4cHNxscGJA3AXwvHxQ=";
+  };
+in
+stdenv.mkDerivation (finalAttrs: {
+  pname = "amdsmi";
+  version = "6.3.3";
+  src = fetchFromGitHub {
+    owner = "rocm";
+    repo = "amdsmi";
+    rev = "rocm-${finalAttrs.version}";
+    hash = "sha256-hrPqd4ZWqzTG7JRfVwc1SZx6TNS0Q/LFg8yDxrM3mPo=";
+  };
+
+  postPatch = ''
+    substituteInPlace goamdsmi_shim/CMakeLists.txt \
+      --replace-fail "amd_smi)" ${"'"}''${AMD_SMI_TARGET})' \
+      --replace-fail 'target_link_libraries(''${GOAMDSMI_SHIM_TARGET} -L' '#'
+
+    # Manually unpack esmi_ib_src and add amd_hsmp.h so execute-process git clone doesn't run
+    cp -rf --no-preserve=mode ${esmi_ib_src} ./esmi_ib_library
+    mkdir -p ./esmi_ib_library/include/asm
+    cp ./include/amd_smi/impl/amd_hsmp.h ./esmi_ib_library/include/asm/amd_hsmp.h
+  '';
+
+  patches = [
+    # Fix ld.lld undefined reference: drmGetVersion
+    (fetchpatch {
+      url = "https://github.com/ROCm/amdsmi/commit/c3864bf6171970d86dc50fd23f06377736823997.patch";
+      hash = "sha256-zRG1tBD8sIQCWdKfCbXC/Z/6d6NTrRYvRpddPWdM4j8=";
+    })
+  ];
+
+  nativeBuildInputs = [
+    cmake
+    pkg-config
+    wrapPython
+    autoPatchelfHook
+  ];
+
+  buildInputs = [
+    libdrm
+  ];
+
+  cmakeFlags = [
+    # Manually define CMAKE_INSTALL_<DIR>
+    # See: https://github.com/NixOS/nixpkgs/pull/197838
+    "-DCMAKE_INSTALL_BINDIR=bin"
+    "-DCMAKE_INSTALL_LIBDIR=lib"
+    "-DCMAKE_INSTALL_INCLUDEDIR=include"
+  ];
+
+  postInstall = ''
+    wrapPythonProgramsIn $out
+    rm $out/bin/amd-smi
+    ln -sf $out/libexec/amdsmi_cli/amdsmi_cli.py $out/bin/amd-smi
+  '';
+
+  passthru.updateScript = rocmUpdateScript {
+    name = finalAttrs.pname;
+    inherit (finalAttrs.src) owner;
+    inherit (finalAttrs.src) repo;
+  };
+
+  meta = with lib; {
+    description = "System management interface for AMD GPUs supported by ROCm";
+    homepage = "https://github.com/ROCm/rocm_smi_lib";
+    license = with licenses; [ mit ];
+    maintainers = with maintainers; [ lovesegfault ] ++ teams.rocm.members;
+    platforms = [ "x86_64-linux" ];
+  };
+})