From 8e800cedaf24f5ad9717463b809b0beef7677000 Mon Sep 17 00:00:00 2001 From: Connor Baker Date: Tue, 7 Nov 2023 14:35:11 +0000 Subject: cudaPackages: move derivations to cuda-modules & support aarch64 cudaPackages.cuda_compat: ignore missing libs provided at runtime cudaPackages.gpus: Jetson should never build by default cudaPackages.flags: don't build Jetson capabilities by default cudaPackages: re-introduce filter for pre-existing CUDA redist packages in overrides cudaPackages: only recurseIntoAttrs for the latest of each major version cudaPackages.nvccCompatabilities: use GCC 10 through CUDA 11.5 to avoid a GLIBC incompatability cudaPackages.cutensor: acquire libcublas through cudatoolkit prior to 11.4 cudaPackages.cuda_compat: mark as broken on aarch64-linux if not targeting Jetson cudaPackages.cutensor_1_4: fix build cudaPackages: adjust use of autoPatchelfIgnoreMissingDeps cudaPackages.cuda_nvprof: remove unecessary override to add addOpenGLRunpath cudaPackages: use getExe' to avoid patchelf warning about missing meta.mainProgram cudaPackages: fix evaluation with Nix 2.3 cudaPackages: fix platform detection for Jetson/non-Jetson aarch64-linux python3Packages.tensorrt: mark as broken if required packages are missing Note: evaluating the name of the derivation will fail if tensorrt is not present, which is why we wrap the value in `lib.optionalString`. cudaPackages.flags.getNixSystem: add guard based on jetsonTargets cudaPackages.cudnn: use explicit path to patchelf cudaPackages.tensorrt: use explicit path to patchelf --- .../cudatoolkit/auto-add-opengl-runpath-hook.sh | 28 -- pkgs/development/compilers/cudatoolkit/common.nix | 367 -------------------- .../hooks/mark-for-cudatoolkit-root-hook.sh | 14 - .../compilers/cudatoolkit/hooks/setup-cuda-hook.sh | 139 -------- .../redist/build-cuda-redist-package.nix | 178 ---------- .../compilers/cudatoolkit/redist/extension.nix | 139 -------- .../compilers/cudatoolkit/redist/overrides.nix | 119 ------- .../compilers/cudatoolkit/saxpy/CMakeLists.txt | 12 - .../compilers/cudatoolkit/saxpy/default.nix | 52 --- .../compilers/cudatoolkit/saxpy/saxpy.cu | 68 ---- pkgs/development/compilers/cudatoolkit/stdenv.nix | 33 -- pkgs/development/cuda-modules/backend-stdenv.nix | 39 +++ pkgs/development/cuda-modules/cuda/extension.nix | 101 ++++++ pkgs/development/cuda-modules/cuda/overrides.nix | 129 +++++++ .../cuda-modules/cudatoolkit/default.nix | 374 +++++++++++++++++++++ pkgs/development/cuda-modules/cudnn/fixup.nix | 69 ++++ pkgs/development/cuda-modules/cudnn/shims.nix | 13 + .../cuda-modules/cutensor/extension.nix | 164 +++++++++ pkgs/development/cuda-modules/flags.nix | 49 +-- .../cuda-modules/generic-builders/manifest.nix | 249 ++++++++++++++ .../cuda-modules/generic-builders/multiplex.nix | 131 ++++++++ pkgs/development/cuda-modules/gpus.nix | 2 +- .../cuda-modules/modules/cuda/default.nix | 1 + .../cuda-modules/modules/cudnn/default.nix | 12 + .../cuda-modules/modules/cutensor/default.nix | 1 + pkgs/development/cuda-modules/modules/default.nix | 10 + .../cuda-modules/modules/generic/default.nix | 7 + .../modules/generic/manifests/default.nix | 7 + .../modules/generic/manifests/feature/manifest.nix | 10 + .../modules/generic/manifests/feature/outputs.nix | 60 ++++ .../modules/generic/manifests/feature/package.nix | 10 + .../modules/generic/manifests/feature/release.nix | 10 + .../generic/manifests/redistrib/manifest.nix | 33 ++ .../generic/manifests/redistrib/package.nix | 32 ++ .../generic/manifests/redistrib/release.nix | 36 ++ .../modules/generic/releases/default.nix | 45 +++ .../cuda-modules/modules/generic/types/default.nix | 39 +++ .../cuda-modules/modules/tensorrt/default.nix | 16 + .../cuda-modules/nccl-tests/default.nix | 84 +++++ pkgs/development/cuda-modules/nccl/default.nix | 112 ++++++ .../cuda-modules/nvcc-compatibilities.nix | 12 +- pkgs/development/cuda-modules/saxpy/CMakeLists.txt | 12 + pkgs/development/cuda-modules/saxpy/default.nix | 56 +++ pkgs/development/cuda-modules/saxpy/saxpy.cu | 68 ++++ .../setup-hooks/auto-add-opengl-runpath-hook.sh | 28 ++ .../cuda-modules/setup-hooks/extension.nix | 47 +++ .../setup-hooks/mark-for-cudatoolkit-root-hook.sh | 14 + .../cuda-modules/setup-hooks/setup-cuda-hook.sh | 139 ++++++++ pkgs/development/cuda-modules/tensorrt/fixup.nix | 113 +++++++ .../development/cuda-modules/tensorrt/releases.nix | 5 +- pkgs/development/cuda-modules/tensorrt/shims.nix | 16 + .../libraries/science/math/cudnn/extension.nix | 66 ---- .../libraries/science/math/cudnn/generic.nix | 170 ---------- .../libraries/science/math/cutensor/generic.nix | 88 ----- .../libraries/science/math/nccl/default.nix | 113 ------- .../libraries/science/math/nccl/tests.nix | 84 ----- .../libraries/science/math/tensorrt/generic.nix | 95 ------ .../python-modules/tensorrt/default.nix | 5 +- pkgs/test/cuda/cuda-library-samples/extension.nix | 17 +- pkgs/test/cuda/cuda-library-samples/generic.nix | 74 ++-- pkgs/test/cuda/cuda-samples/extension.nix | 49 ++- pkgs/test/cuda/cuda-samples/generic.nix | 125 +++---- pkgs/test/cuda/default.nix | 12 +- pkgs/top-level/all-packages.nix | 9 +- pkgs/top-level/cuda-packages.nix | 202 ++++++----- 65 files changed, 2616 insertions(+), 1997 deletions(-) delete mode 100644 pkgs/development/compilers/cudatoolkit/auto-add-opengl-runpath-hook.sh delete mode 100644 pkgs/development/compilers/cudatoolkit/common.nix delete mode 100644 pkgs/development/compilers/cudatoolkit/hooks/mark-for-cudatoolkit-root-hook.sh delete mode 100644 pkgs/development/compilers/cudatoolkit/hooks/setup-cuda-hook.sh delete mode 100644 pkgs/development/compilers/cudatoolkit/redist/build-cuda-redist-package.nix delete mode 100644 pkgs/development/compilers/cudatoolkit/redist/extension.nix delete mode 100644 pkgs/development/compilers/cudatoolkit/redist/overrides.nix delete mode 100644 pkgs/development/compilers/cudatoolkit/saxpy/CMakeLists.txt delete mode 100644 pkgs/development/compilers/cudatoolkit/saxpy/default.nix delete mode 100644 pkgs/development/compilers/cudatoolkit/saxpy/saxpy.cu delete mode 100644 pkgs/development/compilers/cudatoolkit/stdenv.nix create mode 100644 pkgs/development/cuda-modules/backend-stdenv.nix create mode 100644 pkgs/development/cuda-modules/cuda/extension.nix create mode 100644 pkgs/development/cuda-modules/cuda/overrides.nix create mode 100644 pkgs/development/cuda-modules/cudatoolkit/default.nix create mode 100644 pkgs/development/cuda-modules/cudnn/fixup.nix create mode 100644 pkgs/development/cuda-modules/cudnn/shims.nix create mode 100644 pkgs/development/cuda-modules/cutensor/extension.nix create mode 100644 pkgs/development/cuda-modules/generic-builders/manifest.nix create mode 100644 pkgs/development/cuda-modules/generic-builders/multiplex.nix create mode 100644 pkgs/development/cuda-modules/modules/cuda/default.nix create mode 100644 pkgs/development/cuda-modules/modules/cudnn/default.nix create mode 100644 pkgs/development/cuda-modules/modules/cutensor/default.nix create mode 100644 pkgs/development/cuda-modules/modules/default.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/default.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/manifests/default.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/manifests/feature/manifest.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/manifests/feature/outputs.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/manifests/feature/package.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/manifests/feature/release.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/manifests/redistrib/manifest.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/manifests/redistrib/package.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/manifests/redistrib/release.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/releases/default.nix create mode 100644 pkgs/development/cuda-modules/modules/generic/types/default.nix create mode 100644 pkgs/development/cuda-modules/modules/tensorrt/default.nix create mode 100644 pkgs/development/cuda-modules/nccl-tests/default.nix create mode 100644 pkgs/development/cuda-modules/nccl/default.nix create mode 100644 pkgs/development/cuda-modules/saxpy/CMakeLists.txt create mode 100644 pkgs/development/cuda-modules/saxpy/default.nix create mode 100644 pkgs/development/cuda-modules/saxpy/saxpy.cu create mode 100644 pkgs/development/cuda-modules/setup-hooks/auto-add-opengl-runpath-hook.sh create mode 100644 pkgs/development/cuda-modules/setup-hooks/extension.nix create mode 100644 pkgs/development/cuda-modules/setup-hooks/mark-for-cudatoolkit-root-hook.sh create mode 100644 pkgs/development/cuda-modules/setup-hooks/setup-cuda-hook.sh create mode 100644 pkgs/development/cuda-modules/tensorrt/fixup.nix create mode 100644 pkgs/development/cuda-modules/tensorrt/shims.nix delete mode 100644 pkgs/development/libraries/science/math/cudnn/extension.nix delete mode 100644 pkgs/development/libraries/science/math/cudnn/generic.nix delete mode 100644 pkgs/development/libraries/science/math/cutensor/generic.nix delete mode 100644 pkgs/development/libraries/science/math/nccl/default.nix delete mode 100644 pkgs/development/libraries/science/math/nccl/tests.nix delete mode 100644 pkgs/development/libraries/science/math/tensorrt/generic.nix diff --git a/pkgs/development/compilers/cudatoolkit/auto-add-opengl-runpath-hook.sh b/pkgs/development/compilers/cudatoolkit/auto-add-opengl-runpath-hook.sh deleted file mode 100644 index f50a5f6c25c66..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/auto-add-opengl-runpath-hook.sh +++ /dev/null @@ -1,28 +0,0 @@ -# shellcheck shell=bash -# Run addOpenGLRunpath on all dynamically linked, ELF files -echo "Sourcing auto-add-opengl-runpath-hook" - -elfHasDynamicSection() { - patchelf --print-rpath "$1" >& /dev/null -} - -autoAddOpenGLRunpathPhase() ( - local outputPaths - mapfile -t outputPaths < <(for o in $(getAllOutputNames); do echo "${!o}"; done) - find "${outputPaths[@]}" -type f -executable -print0 | while IFS= read -rd "" f; do - if isELF "$f"; then - # patchelf returns an error on statically linked ELF files - if elfHasDynamicSection "$f" ; then - echo "autoAddOpenGLRunpathHook: patching $f" - addOpenGLRunpath "$f" - elif (( "${NIX_DEBUG:-0}" >= 1 )) ; then - echo "autoAddOpenGLRunpathHook: skipping a statically-linked ELF file $f" - fi - fi - done -) - -if [ -z "${dontUseAutoAddOpenGLRunpath-}" ]; then - echo "Using autoAddOpenGLRunpathPhase" - postFixupHooks+=(autoAddOpenGLRunpathPhase) -fi diff --git a/pkgs/development/compilers/cudatoolkit/common.nix b/pkgs/development/compilers/cudatoolkit/common.nix deleted file mode 100644 index 681549fa62dbe..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/common.nix +++ /dev/null @@ -1,367 +0,0 @@ -args@ -{ version -, sha256 -, url ? "" -, name ? "" -, developerProgram ? false -, runPatches ? [] -, autoPatchelfHook -, autoAddOpenGLRunpathHook -, addOpenGLRunpath -, alsa-lib -, curlMinimal -, expat -, fetchurl -, fontconfig -, freetype -, gdk-pixbuf -, glib -, glibc -, gst_all_1 -, gtk2 -, lib -, libxkbcommon -, libkrb5 -, krb5 -, makeWrapper -, markForCudatoolkitRootHook -, ncurses5 -, numactl -, nss -, perl -, python3 # FIXME: CUDAToolkit 10 may still need python27 -, pulseaudio -, requireFile -, setupCudaHook -, stdenv -, backendStdenv # E.g. gcc11Stdenv, set in extension.nix -, unixODBC -, wayland -, xorg -, zlib -, freeglut -, libGLU -, libsForQt5 -, libtiff -, qt6Packages -, qt6 -, rdma-core -, ucx -, rsync -}: - -backendStdenv.mkDerivation rec { - pname = "cudatoolkit"; - inherit version runPatches; - - dontPatchELF = true; - dontStrip = true; - - src = - if developerProgram then - requireFile { - message = '' - This nix expression requires that ${args.name} is already part of the store. - Register yourself to NVIDIA Accelerated Computing Developer Program, retrieve the CUDA toolkit - at https://developer.nvidia.com/cuda-toolkit, and run the following command in the download directory: - nix-prefetch-url file://\$PWD/${args.name} - ''; - inherit (args) name sha256; - } - else - fetchurl { - inherit (args) url sha256; - }; - - outputs = [ "out" "lib" "doc" ]; - - nativeBuildInputs = [ - perl - makeWrapper - rsync - addOpenGLRunpath - autoPatchelfHook - autoAddOpenGLRunpathHook - markForCudatoolkitRootHook - ] ++ lib.optionals (lib.versionOlder version "11") [ - libsForQt5.wrapQtAppsHook - ] ++ lib.optionals (lib.versionAtLeast version "11.8") [ - qt6Packages.wrapQtAppsHook - ]; - depsTargetTargetPropagated = [ - setupCudaHook - ]; - buildInputs = lib.optionals (lib.versionOlder version "11") [ - libsForQt5.qt5.qtwebengine - freeglut - libGLU - ] ++ [ - # To get $GDK_PIXBUF_MODULE_FILE via setup-hook - gdk-pixbuf - - # For autoPatchelf - ncurses5 - expat - python3 - zlib - glibc - xorg.libX11 - xorg.libXext - xorg.libXrender - xorg.libXt - xorg.libXtst - xorg.libXi - xorg.libXext - xorg.libXdamage - xorg.libxcb - xorg.xcbutilimage - xorg.xcbutilrenderutil - xorg.xcbutilwm - xorg.xcbutilkeysyms - pulseaudio - libxkbcommon - libkrb5 - krb5 - gtk2 - glib - fontconfig - freetype - numactl - nss - unixODBC - alsa-lib - wayland - ] ++ lib.optionals (lib.versionAtLeast version "11.8") [ - (lib.getLib libtiff) - qt6Packages.qtwayland - rdma-core - (ucx.override { enableCuda = false; }) # Avoid infinite recursion - xorg.libxshmfence - xorg.libxkbfile - ] ++ (lib.optionals (lib.versionAtLeast version "12") (map lib.getLib ([ - # Used by `/target-linux-x64/CollectX/clx` and `/target-linux-x64/CollectX/libclx_api.so` for: - # - `libcurl.so.4` - curlMinimal - - # Used by `/host-linux-x64/Scripts/WebRTCContainer/setup/neko/server/bin/neko` - gst_all_1.gstreamer - gst_all_1.gst-plugins-base - ]) ++ (with qt6; [ - qtmultimedia - qttools - qtpositioning - qtscxml - qtsvg - qtwebchannel - qtwebengine - ]))); - - # Prepended to runpaths by autoPatchelf. - # The order inherited from older rpath preFixup code - runtimeDependencies = [ - (placeholder "lib") - (placeholder "out") - "${placeholder "out"}/nvvm" - # NOTE: use the same libstdc++ as the rest of nixpkgs, not from backendStdenv - "${lib.getLib stdenv.cc.cc}/lib64" - "${placeholder "out"}/jre/lib/amd64/jli" - "${placeholder "out"}/lib64" - "${placeholder "out"}/nvvm/lib64" - ]; - - autoPatchelfIgnoreMissingDeps = [ - # This is the hardware-dependent userspace driver that comes from - # nvidia_x11 package. It must be deployed at runtime in - # /run/opengl-driver/lib or pointed at by LD_LIBRARY_PATH variable, rather - # than pinned in runpath - "libcuda.so.1" - - # The krb5 expression ships libcom_err.so.3 but cudatoolkit asks for the - # older - # This dependency is asked for by target-linux-x64/CollectX/RedHat/x86_64/libssl.so.10 - # - do we even want to use nvidia-shipped libssl? - "libcom_err.so.2" - ]; - - preFixup = if lib.versionOlder version "11" then '' - patchelf $out/targets/*/lib/libnvrtc.so --add-needed libnvrtc-builtins.so - '' else '' - patchelf $out/lib64/libnvrtc.so --add-needed libnvrtc-builtins.so - ''; - - unpackPhase = '' - sh $src --keep --noexec - - ${lib.optionalString (lib.versionOlder version "10.1") '' - cd pkg/run_files - sh cuda-linux*.run --keep --noexec - sh cuda-samples*.run --keep --noexec - mv pkg ../../$(basename $src) - cd ../.. - rm -rf pkg - - for patch in $runPatches; do - sh $patch --keep --noexec - mv pkg $(basename $patch) - done - ''} - ''; - - installPhase = '' - runHook preInstall - mkdir $out - ${lib.optionalString (lib.versionOlder version "10.1") '' - cd $(basename $src) - export PERL5LIB=. - perl ./install-linux.pl --prefix="$out" - cd .. - for patch in $runPatches; do - cd $(basename $patch) - perl ./install_patch.pl --silent --accept-eula --installdir="$out" - cd .. - done - ''} - ${lib.optionalString (lib.versionAtLeast version "10.1" && lib.versionOlder version "11") '' - cd pkg/builds/cuda-toolkit - mv * $out/ - ''} - ${lib.optionalString (lib.versionAtLeast version "11") '' - mkdir -p $out/bin $out/lib64 $out/include $doc - for dir in pkg/builds/* pkg/builds/cuda_nvcc/nvvm pkg/builds/cuda_cupti/extras/CUPTI; do - if [ -d $dir/bin ]; then - mv $dir/bin/* $out/bin - fi - if [ -d $dir/doc ]; then - (cd $dir/doc && find . -type d -exec mkdir -p $doc/\{} \;) - (cd $dir/doc && find . \( -type f -o -type l \) -exec mv \{} $doc/\{} \;) - fi - if [ -L $dir/include ] || [ -d $dir/include ]; then - (cd $dir/include && find . -type d -exec mkdir -p $out/include/\{} \;) - (cd $dir/include && find . \( -type f -o -type l \) -exec mv \{} $out/include/\{} \;) - fi - if [ -L $dir/lib64 ] || [ -d $dir/lib64 ]; then - (cd $dir/lib64 && find . -type d -exec mkdir -p $out/lib64/\{} \;) - (cd $dir/lib64 && find . \( -type f -o -type l \) -exec mv \{} $out/lib64/\{} \;) - fi - done - mv pkg/builds/cuda_nvcc/nvvm $out/nvvm - - mv pkg/builds/cuda_sanitizer_api $out/cuda_sanitizer_api - ln -s $out/cuda_sanitizer_api/compute-sanitizer/compute-sanitizer $out/bin/compute-sanitizer - - mv pkg/builds/nsight_systems/target-linux-x64 $out/target-linux-x64 - mv pkg/builds/nsight_systems/host-linux-x64 $out/host-linux-x64 - rm $out/host-linux-x64/libstdc++.so* - ''} - ${lib.optionalString (lib.versionAtLeast version "11.8") - # error: auto-patchelf could not satisfy dependency libtiff.so.5 wanted by /nix/store/.......-cudatoolkit-12.0.1/host-linux-x64/Plugins/imageformats/libqtiff.so - # we only ship libtiff.so.6, so let's use qt plugins built by Nix. - # TODO: don't copy, come up with a symlink-based "merge" - '' - rsync ${lib.getLib qt6Packages.qtimageformats}/lib/qt-6/plugins/ $out/host-linux-x64/Plugins/ -aP - ''} - - rm -f $out/tools/CUDA_Occupancy_Calculator.xls # FIXME: why? - - ${lib.optionalString (lib.versionOlder version "10.1") '' - # let's remove the 32-bit libraries, they confuse the lib64->lib mover - rm -rf $out/lib - ''} - - ${lib.optionalString (lib.versionAtLeast version "12.0") '' - rm $out/host-linux-x64/libQt6* - ''} - - # Remove some cruft. - ${lib.optionalString ((lib.versionAtLeast version "7.0") && (lib.versionOlder version "10.1")) - "rm $out/bin/uninstall*"} - - # Fixup path to samples (needed for cuda 6.5 or else nsight will not find them) - if [ -d "$out"/cuda-samples ]; then - mv "$out"/cuda-samples "$out"/samples - fi - - # Change the #error on GCC > 4.9 to a #warning. - sed -i $out/include/host_config.h -e 's/#error\(.*unsupported GNU version\)/#warning\1/' - - # Fix builds with newer glibc version - sed -i "1 i#define _BITS_FLOATN_H" "$out/include/host_defines.h" - '' + - # Point NVCC at a compatible compiler - # CUDA_TOOLKIT_ROOT_DIR is legacy, - # Cf. https://cmake.org/cmake/help/latest/module/FindCUDA.html#input-variables - '' - mkdir -p $out/nix-support - cat <> $out/nix-support/setup-hook - cmakeFlags+=' -DCUDA_TOOLKIT_ROOT_DIR=$out' - EOF - - # Move some libraries to the lib output so that programs that - # depend on them don't pull in this entire monstrosity. - mkdir -p $lib/lib - mv -v $out/lib64/libcudart* $lib/lib/ - - # Remove OpenCL libraries as they are provided by ocl-icd and driver. - rm -f $out/lib64/libOpenCL* - ${lib.optionalString (lib.versionAtLeast version "10.1" && (lib.versionOlder version "11")) '' - mv $out/lib64 $out/lib - mv $out/extras/CUPTI/lib64/libcupti* $out/lib - ''} - - # nvprof do not find any program to profile if LD_LIBRARY_PATH is not set - wrapProgram $out/bin/nvprof \ - --prefix LD_LIBRARY_PATH : $out/lib - '' + lib.optionalString (lib.versionOlder version "8.0") '' - # Hack to fix building against recent Glibc/GCC. - echo "NIX_CFLAGS_COMPILE+=' -D_FORCE_INLINES'" >> $out/nix-support/setup-hook - '' - # 11.8 includes a broken symlink, include/include, pointing to targets/x86_64-linux/include - + lib.optionalString (lib.versions.majorMinor version == "11.8") '' - rm $out/include/include - '' + '' - runHook postInstall - ''; - - postInstall = '' - for b in nvvp ${lib.optionalString (lib.versionOlder version "11") "nsight"}; do - wrapProgram "$out/bin/$b" \ - --set GDK_PIXBUF_MODULE_FILE "$GDK_PIXBUF_MODULE_FILE" - done - ''; - - - # cuda-gdb doesn't run correctly when not using sandboxing, so - # temporarily disabling the install check. This should be set to true - # when we figure out how to get `cuda-gdb --version` to run correctly - # when not using sandboxing. - doInstallCheck = false; - postInstallCheck = let - in '' - # Smoke test binaries - pushd $out/bin - for f in *; do - case $f in - crt) continue;; - nvcc.profile) continue;; - nsight_ee_plugins_manage.sh) continue;; - uninstall_cuda_toolkit_6.5.pl) continue;; - computeprof|nvvp|nsight) continue;; # GUIs don't feature "--version" - *) echo "Executing '$f --version':"; ./$f --version;; - esac - done - popd - ''; - passthru = { - inherit (backendStdenv) cc; - majorMinorVersion = lib.versions.majorMinor version; - majorVersion = lib.versions.majorMinor version; - }; - - meta = with lib; { - description = "A compiler for NVIDIA GPUs, math libraries, and tools"; - homepage = "https://developer.nvidia.com/cuda-toolkit"; - platforms = [ "x86_64-linux" ]; - license = licenses.nvidiaCuda; - maintainers = teams.cuda.members; - }; -} diff --git a/pkgs/development/compilers/cudatoolkit/hooks/mark-for-cudatoolkit-root-hook.sh b/pkgs/development/compilers/cudatoolkit/hooks/mark-for-cudatoolkit-root-hook.sh deleted file mode 100644 index ba04c2e0806af..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/hooks/mark-for-cudatoolkit-root-hook.sh +++ /dev/null @@ -1,14 +0,0 @@ -# shellcheck shell=bash - -# Should we mimick cc-wrapper's "hygiene"? -[[ -z ${strictDeps-} ]] || (( "$hostOffset" < 0 )) || return 0 - -echo "Sourcing mark-for-cudatoolkit-root-hook" >&2 - -markForCUDAToolkit_ROOT() { - mkdir -p "${prefix}/nix-support" - [[ -f "${prefix}/nix-support/include-in-cudatoolkit-root" ]] && return - echo "$pname-$output" > "${prefix}/nix-support/include-in-cudatoolkit-root" -} - -fixupOutputHooks+=(markForCUDAToolkit_ROOT) diff --git a/pkgs/development/compilers/cudatoolkit/hooks/setup-cuda-hook.sh b/pkgs/development/compilers/cudatoolkit/hooks/setup-cuda-hook.sh deleted file mode 100644 index 7b7b3bdde80e3..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/hooks/setup-cuda-hook.sh +++ /dev/null @@ -1,139 +0,0 @@ -# shellcheck shell=bash - -# Only run the hook from nativeBuildInputs -(( "$hostOffset" == -1 && "$targetOffset" == 0)) || return 0 - -guard=Sourcing -reason= - -[[ -n ${cudaSetupHookOnce-} ]] && guard=Skipping && reason=" because the hook has been propagated more than once" - -if (( "${NIX_DEBUG:-0}" >= 1 )) ; then - echo "$guard hostOffset=$hostOffset targetOffset=$targetOffset setupCudaHook$reason" >&2 -else - echo "$guard setup-cuda-hook$reason" >&2 -fi - -[[ "$guard" = Sourcing ]] || return 0 - -declare -g cudaSetupHookOnce=1 -declare -Ag cudaHostPathsSeen=() -declare -Ag cudaOutputToPath=() - -extendcudaHostPathsSeen() { - (( "${NIX_DEBUG:-0}" >= 1 )) && echo "extendcudaHostPathsSeen $1" >&2 - - local markerPath="$1/nix-support/include-in-cudatoolkit-root" - [[ ! -f "${markerPath}" ]] && return - [[ -v cudaHostPathsSeen[$1] ]] && return - - cudaHostPathsSeen["$1"]=1 - - # E.g. cuda_cudart-lib - local cudaOutputName - read -r cudaOutputName < "$markerPath" - - [[ -z "$cudaOutputName" ]] && return - - local oldPath="${cudaOutputToPath[$cudaOutputName]-}" - [[ -n "$oldPath" ]] && echo "extendcudaHostPathsSeen: warning: overwriting $cudaOutputName from $oldPath to $1" >&2 - cudaOutputToPath["$cudaOutputName"]="$1" -} -addEnvHooks "$targetOffset" extendcudaHostPathsSeen - -setupCUDAToolkit_ROOT() { - (( "${NIX_DEBUG:-0}" >= 1 )) && echo "setupCUDAToolkit_ROOT: cudaHostPathsSeen=${!cudaHostPathsSeen[*]}" >&2 - - for path in "${!cudaHostPathsSeen[@]}" ; do - addToSearchPathWithCustomDelimiter ";" CUDAToolkit_ROOT "$path" - if [[ -d "$path/include" ]] ; then - addToSearchPathWithCustomDelimiter ";" CUDAToolkit_INCLUDE_DIR "$path/include" - fi - done - - export cmakeFlags+=" -DCUDAToolkit_INCLUDE_DIR=$CUDAToolkit_INCLUDE_DIR -DCUDAToolkit_ROOT=$CUDAToolkit_ROOT" -} -preConfigureHooks+=(setupCUDAToolkit_ROOT) - -setupCUDAToolkitCompilers() { - echo Executing setupCUDAToolkitCompilers >&2 - - if [[ -n "${dontSetupCUDAToolkitCompilers-}" ]] ; then - return - fi - - # Point NVCC at a compatible compiler - - # For CMake-based projects: - # https://cmake.org/cmake/help/latest/module/FindCUDA.html#input-variables - # https://cmake.org/cmake/help/latest/envvar/CUDAHOSTCXX.html - # https://cmake.org/cmake/help/latest/variable/CMAKE_CUDA_HOST_COMPILER.html - - export cmakeFlags+=" -DCUDA_HOST_COMPILER=@ccFullPath@" - export cmakeFlags+=" -DCMAKE_CUDA_HOST_COMPILER=@ccFullPath@" - - # For non-CMake projects: - # We prepend --compiler-bindir to nvcc flags. - # Downstream packages can override these, because NVCC - # uses the last --compiler-bindir it gets on the command line. - # FIXME: this results in "incompatible redefinition" warnings. - # https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#compiler-bindir-directory-ccbin - if [ -z "${CUDAHOSTCXX-}" ]; then - export CUDAHOSTCXX="@ccFullPath@"; - fi - - export NVCC_PREPEND_FLAGS+=" --compiler-bindir=@ccRoot@/bin" - - # NOTE: We set -Xfatbin=-compress-all, which reduces the size of the compiled - # binaries. If binaries grow over 2GB, they will fail to link. This is a problem for us, as - # the default set of CUDA capabilities we build can regularly cause this to occur (for - # example, with Magma). - # - # @SomeoneSerge: original comment was made by @ConnorBaker in .../cudatoolkit/common.nix - if [[ -z "${dontCompressFatbin-}" ]]; then - export NVCC_PREPEND_FLAGS+=" -Xfatbin=-compress-all" - fi - - # CMake's enable_language(CUDA) runs a compiler test and it doesn't account for - # CUDAToolkit_ROOT. We have to help it locate libcudart - if [[ -z "${nvccDontPrependCudartFlags-}" ]] ; then - if [[ ! -v cudaOutputToPath["cuda_cudart-out"] ]] ; then - echo "setupCUDAToolkitCompilers: missing cudaPackages.cuda_cudart. This may become an an error in the future" >&2 - # exit 1 - fi - for pkg in "${!cudaOutputToPath[@]}" ; do - [[ ! "$pkg" = cuda_cudart* ]] && continue - - local path="${cudaOutputToPath[$pkg]}" - if [[ -d "$path/include" ]] ; then - export NVCC_PREPEND_FLAGS+=" -I$path/include" - fi - if [[ -d "$path/lib" ]] ; then - export NVCC_PREPEND_FLAGS+=" -L$path/lib" - fi - done - fi -} -preConfigureHooks+=(setupCUDAToolkitCompilers) - -propagateCudaLibraries() { - (( "${NIX_DEBUG:-0}" >= 1 )) && echo "propagateCudaLibraries: cudaPropagateToOutput=$cudaPropagateToOutput cudaHostPathsSeen=${!cudaHostPathsSeen[*]}" >&2 - - [[ -z "${cudaPropagateToOutput-}" ]] && return - - mkdir -p "${!cudaPropagateToOutput}/nix-support" - # One'd expect this should be propagated-bulid-build-deps, but that doesn't seem to work - echo "@setupCudaHook@" >> "${!cudaPropagateToOutput}/nix-support/propagated-native-build-inputs" - - local propagatedBuildInputs=( "${!cudaHostPathsSeen[@]}" ) - for output in $(getAllOutputNames) ; do - if [[ ! "$output" = "$cudaPropagateToOutput" ]] ; then - propagatedBuildInputs+=( "${!output}" ) - fi - break - done - - # One'd expect this should be propagated-host-host-deps, but that doesn't seem to work - printWords "${propagatedBuildInputs[@]}" >> "${!cudaPropagateToOutput}/nix-support/propagated-build-inputs" -} -postFixupHooks+=(propagateCudaLibraries) diff --git a/pkgs/development/compilers/cudatoolkit/redist/build-cuda-redist-package.nix b/pkgs/development/compilers/cudatoolkit/redist/build-cuda-redist-package.nix deleted file mode 100644 index 1e4a3ab178559..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/redist/build-cuda-redist-package.nix +++ /dev/null @@ -1,178 +0,0 @@ -# Type Aliases -# -# See ./extension.nix: -# - ReleaseAttrs -# - ReleaseFeaturesAttrs -# -# General callPackage-supplied arguments -{ lib -, stdenv -, backendStdenv -, fetchurl -, autoPatchelfHook -, autoAddOpenGLRunpathHook -, markForCudatoolkitRootHook -, lndir -, symlinkJoin -}: -# Function arguments -{ - # Short package name (e.g., "cuda_cccl") - # pname : String - pname -, # Long package name (e.g., "CXX Core Compute Libraries") - # description : String - description -, # platforms : List System - platforms -, # version : Version - version -, # releaseAttrs : ReleaseAttrs - releaseAttrs -, # releaseFeaturesAttrs : ReleaseFeaturesAttrs - releaseFeaturesAttrs -, -}: -let - # Useful imports - inherit (lib.lists) optionals; - inherit (lib.meta) getExe; - inherit (lib.strings) optionalString; -in -backendStdenv.mkDerivation (finalAttrs: { - # NOTE: Even though there's no actual buildPhase going on here, the derivations of the - # redistributables are sensitive to the compiler flags provided to stdenv. The patchelf package - # is sensitive to the compiler flags provided to stdenv, and we depend on it. As such, we are - # also sensitive to the compiler flags provided to stdenv. - inherit pname version; - strictDeps = true; - - outputs = with releaseFeaturesAttrs; - [ "out" ] - ++ optionals hasBin [ "bin" ] - ++ optionals hasLib [ "lib" ] - ++ optionals hasStatic [ "static" ] - ++ optionals hasDev [ "dev" ] - ++ optionals hasDoc [ "doc" ] - ++ optionals hasSample [ "sample" ]; - - src = fetchurl { - url = "https://developer.download.nvidia.com/compute/cuda/redist/${releaseAttrs.relative_path}"; - inherit (releaseAttrs) sha256; - }; - - # We do need some other phases, like configurePhase, so the multiple-output setup hook works. - dontBuild = true; - - nativeBuildInputs = [ - autoPatchelfHook - # This hook will make sure libcuda can be found - # in typically /lib/opengl-driver by adding that - # directory to the rpath of all ELF binaries. - # Check e.g. with `patchelf --print-rpath path/to/my/binary - autoAddOpenGLRunpathHook - markForCudatoolkitRootHook - ]; - - buildInputs = [ - # autoPatchelfHook will search for a libstdc++ and we're giving it - # one that is compatible with the rest of nixpkgs, even when - # nvcc forces us to use an older gcc - # NB: We don't actually know if this is the right thing to do - stdenv.cc.cc.lib - ]; - - # Picked up by autoPatchelf - # Needed e.g. for libnvrtc to locate (dlopen) libnvrtc-builtins - appendRunpaths = [ - "$ORIGIN" - ]; - - installPhase = with releaseFeaturesAttrs; - # Pre-install hook - '' - runHook preInstall - '' - # doc and dev have special output handling. Other outputs need to be moved to their own - # output. - # Note that moveToOutput operates on all outputs: - # https://github.com/NixOS/nixpkgs/blob/2920b6fc16a9ed5d51429e94238b28306ceda79e/pkgs/build-support/setup-hooks/multiple-outputs.sh#L105-L107 - + '' - mkdir -p "$out" - rm LICENSE - mv * "$out" - '' - # Handle bin, which defaults to out - + optionalString hasBin '' - moveToOutput "bin" "$bin" - '' - # Handle lib, which defaults to out - + optionalString hasLib '' - moveToOutput "lib" "$lib" - '' - # Handle static libs, which isn't handled by the setup hook - + optionalString hasStatic '' - moveToOutput "**/*.a" "$static" - '' - # Handle samples, which isn't handled by the setup hook - + optionalString hasSample '' - moveToOutput "samples" "$sample" - '' - # Post-install hook - + '' - runHook postInstall - ''; - - # The out output leverages the same functionality which backs the `symlinkJoin` function in - # Nixpkgs: - # https://github.com/NixOS/nixpkgs/blob/d8b2a92df48f9b08d68b0132ce7adfbdbc1fbfac/pkgs/build-support/trivial-builders/default.nix#L510 - # - # That should allow us to emulate "fat" default outputs without having to actually create them. - # - # It is important that this run after the autoPatchelfHook, otherwise the symlinks in out will reference libraries in lib, creating a circular dependency. - postPhases = [ "postPatchelf" ]; - # For each output, create a symlink to it in the out output. - # NOTE: We must recreate the out output here, because the setup hook will have deleted it - # if it was empty. - # NOTE: Do not use optionalString based on whether `outputs` contains only `out` -- phases - # which are empty strings are skipped/unset and result in errors of the form "command not - # found: ". - postPatchelf = '' - mkdir -p "$out" - for output in $outputs; do - if [ "$output" = "out" ]; then - continue - fi - ${getExe lndir} "''${!output}" "$out" - done - ''; - - # Make the CUDA-patched stdenv available - passthru.stdenv = backendStdenv; - - # Setting propagatedBuildInputs to false will prevent outputs known to the multiple-outputs - # from depending on `out` by default. - # https://github.com/NixOS/nixpkgs/blob/2920b6fc16a9ed5d51429e94238b28306ceda79e/pkgs/build-support/setup-hooks/multiple-outputs.sh#L196 - # Indeed, we want to do the opposite -- fat "out" outputs that contain all the other outputs. - propagatedBuildOutputs = false; - - # By default, if the dev output exists it just uses that. - # However, because we disabled propagatedBuildOutputs, dev doesn't contain libraries or - # anything of the sort. To remedy this, we set outputSpecified to true, and use - # outputsToInstall, which tells Nix which outputs to use when the package name is used - # unqualified (that is, without an explicit output). - outputSpecified = true; - - meta = { - inherit platforms; - description = "${description}. By downloading and using the packages you accept the terms and conditions of the ${finalAttrs.meta.license.shortName}"; - license = lib.licenses.nvidiaCudaRedist // { - url = "https://developer.download.nvidia.com/compute/cuda/redist/${releaseAttrs.license_path or "${pname}/LICENSE.txt"}"; - }; - sourceProvenance = [ lib.sourceTypes.binaryNativeCode ]; - maintainers = lib.teams.cuda.members; - # Force the use of the default, fat output by default (even though `dev` exists, which - # causes Nix to prefer that output over the others if outputSpecified isn't set). - outputsToInstall = [ "out" ]; - }; -}) diff --git a/pkgs/development/compilers/cudatoolkit/redist/extension.nix b/pkgs/development/compilers/cudatoolkit/redist/extension.nix deleted file mode 100644 index ea6fd581cde4f..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/redist/extension.nix +++ /dev/null @@ -1,139 +0,0 @@ -# Type Aliases -# -# ReleaseAttrs : { -# "relative_path" : String, -# "sha256" : String, -# "md5" : String, -# "size" : String, -# } -# -# NOTE: PackageAttrs must have at least one of the arches. -# PackageAttrs : { -# "name" : String, -# "license" : String, -# "version" : String, -# "license_path" : None | String, -# "linux-aarch64" : None | ReleaseAttrs, -# "linux-ppc64le" : None | ReleaseAttrs, -# "linux-sbsa" : None | ReleaseAttrs, -# "linux-x86_64" : None | ReleaseAttrs, -# "windows-x86_64" : None | ReleaseAttrs, -# } -# -# ReleaseFeaturesAttrs : { -# "hasBin" : Boolean, -# "hasDev" : Boolean, -# "hasDoc" : Boolean, -# "hasLib" : Boolean, -# "hasOut" : Boolean, -# "hasSample" : Boolean, -# "hasStatic" : Boolean, -# "rootDirs" : List String, -# } -# -# NOTE: PackageFeatureAttrs must have at least one of the arches. -# PackageFeatureAttrs : { -# "linux-aarch64" : None | ReleaseFeaturesAttrs, -# "linux-ppc64le" : None | ReleaseFeaturesAttrs, -# "linux-sbsa" : None | ReleaseFeaturesAttrs, -# "linux-x86_64" : None | ReleaseFeaturesAttrs, -# "windows-x86_64" : None | ReleaseFeaturesAttrs, -# } -# -final: prev: -let - # NOTE: We use hasAttr throughout instead of the (?) operator because hasAttr does not require - # us to interpolate our variables into strings (like ${attrName}). - inherit (builtins) attrNames concatMap hasAttr listToAttrs removeAttrs; - inherit (final) callPackage; - inherit (prev) cudaVersion; - inherit (prev.lib.attrsets) nameValuePair optionalAttrs; - inherit (prev.lib.lists) optionals; - inherit (prev.lib.trivial) flip importJSON pipe; - - # Manifest files for CUDA redistributables (aka redist). These can be found at - # https://developer.download.nvidia.com/compute/cuda/redist/ - # Maps a cuda version to the specific version of the manifest. - cudaVersionMap = { - "11.4" = "11.4.4"; - "11.5" = "11.5.2"; - "11.6" = "11.6.2"; - "11.7" = "11.7.0"; - "11.8" = "11.8.0"; - "12.0" = "12.0.1"; - "12.1" = "12.1.1"; - "12.2" = "12.2.0"; - }; - - # Check if the current CUDA version is supported. - cudaVersionMappingExists = hasAttr cudaVersion cudaVersionMap; - - # Maps a cuda version to its manifest files. - # The manifest itself is from NVIDIA, but the features manifest is generated - # by us ahead of time and allows us to split pacakges into multiple outputs. - # Package names (e.g., "cuda_cccl") are mapped to their attributes or features. - # Since we map each attribute to a package name, we need to make sure to get rid of meta - # attributes included in the manifest. Currently, these are any of the following: - # - release_date - # - release_label - # - release_product - redistManifests = - let - # Remove meta attributes from the manifest - # removeAttrs : AttrSet String b -> Attr String b - removeMetaAttrs = flip removeAttrs [ "release_date" "release_label" "release_product" ]; - # processManifest : Path -> Attr Set (String PackageAttrs) - processManifest = flip pipe [ importJSON removeMetaAttrs ]; - # fullCudaVersion : String - fullCudaVersion = cudaVersionMap.${cudaVersion}; - in - { - # features : Attr Set (String PackageFeatureAttrs) - features = processManifest (./manifests + "/redistrib_features_${fullCudaVersion}.json"); - # manifest : Attr Set (String PackageAttrs) - manifest = processManifest (./manifests + "/redistrib_${fullCudaVersion}.json"); - }; - - # Function to build a single redist package - buildRedistPackage = callPackage ./build-cuda-redist-package.nix { }; - - # Function that builds all redist packages given manifests - buildRedistPackages = { features, manifest }: - let - wrapper = pname: - let - # Get the redist architectures the package provides distributables for - packageAttrs = manifest.${pname}; - - # Check if supported - # TODO(@connorbaker): Currently hardcoding x86_64-linux as the only supported platform. - isSupported = packageAttrs ? linux-x86_64; - - # Build the derivation - drv = buildRedistPackage { - inherit pname; - # TODO(@connorbaker): We currently discard the license attribute. - inherit (manifest.${pname}) version; - description = manifest.${pname}.name; - platforms = [ "x86_64-linux" ]; - releaseAttrs = manifest.${pname}.linux-x86_64; - releaseFeaturesAttrs = features.${pname}.linux-x86_64; - }; - - # Wrap in an optional so we can filter out the empty lists created by unsupported - # packages with concatMap. - wrapped = optionals isSupported [ (nameValuePair pname drv) ]; - in - wrapped; - - # concatMap provides us an easy way to filter out packages for unsupported platforms. - # We wrap the buildRedistPackage call in a list to prevent errors when the package is not - # supported (by returning an empty list). - redistPackages = listToAttrs (concatMap wrapper (attrNames manifest)); - in - redistPackages; - - # All redistributable packages for the current CUDA version - redistPackages = optionalAttrs cudaVersionMappingExists (buildRedistPackages redistManifests); -in -redistPackages diff --git a/pkgs/development/compilers/cudatoolkit/redist/overrides.nix b/pkgs/development/compilers/cudatoolkit/redist/overrides.nix deleted file mode 100644 index 71e70e8d7b704..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/redist/overrides.nix +++ /dev/null @@ -1,119 +0,0 @@ -final: prev: -let - inherit (prev) lib pkgs; - cudaVersionOlder = lib.versionOlder final.cudaVersion; - cudaVersionAtLeast = lib.versionAtLeast final.cudaVersion; -in -(lib.filterAttrs (attr: _: (prev ? "${attr}")) { - ### Overrides to fix the components of cudatoolkit-redist - - # Attributes that don't exist in the previous set are removed. - # That means only overrides can go here, and not new expressions! - - libcufile = prev.libcufile.overrideAttrs (oldAttrs: { - buildInputs = oldAttrs.buildInputs ++ [ - final.libcublas.lib - pkgs.numactl - pkgs.rdma-core - ]; - # libcuda needs to be resolved during runtime - autoPatchelfIgnoreMissingDeps = - ["libcuda.so.1"] - # Before 12.0 libcufile depends on itself for some reason. - ++ lib.optionals (cudaVersionOlder "12.0") [ - "libcufile.so.0" - ]; - }); - - libcusolver = final.addBuildInputs prev.libcusolver ( - # Always depends on this - [final.libcublas.lib] - # Dependency from 12.0 and on - ++ lib.optionals (cudaVersionAtLeast "12.0") [ - final.libnvjitlink.lib - ] - # Dependency from 12.1 and on - ++ lib.optionals (cudaVersionAtLeast "12.1") [ - final.libcusparse.lib - ] - ); - - libcusparse = final.addBuildInputs prev.libcusparse ( - lib.optionals (cudaVersionAtLeast "12.0") [ - final.libnvjitlink.lib - ] - ); - - cuda_gdb = final.addBuildInputs prev.cuda_gdb ( - # x86_64 only needs gmp from 12.0 and on - lib.optionals (cudaVersionAtLeast "12.0") [ - pkgs.gmp - ] - ); - - cuda_nvcc = prev.cuda_nvcc.overrideAttrs (oldAttrs: { - propagatedBuildInputs = [ - final.setupCudaHook - ]; - - meta = (oldAttrs.meta or { }) // { - mainProgram = "nvcc"; - }; - }); - - cuda_nvprof = prev.cuda_nvprof.overrideAttrs (oldAttrs: { - nativeBuildInputs = oldAttrs.nativeBuildInputs ++ [ pkgs.addOpenGLRunpath ]; - buildInputs = oldAttrs.buildInputs ++ [ final.cuda_cupti.lib ]; - # libcuda needs to be resolved during runtime - autoPatchelfIgnoreMissingDeps = ["libcuda.so.1"]; - }); - - cuda_demo_suite = final.addBuildInputs prev.cuda_demo_suite [ - pkgs.freeglut - pkgs.libGLU - pkgs.libglvnd - pkgs.mesa - final.libcufft.lib - final.libcurand.lib - ]; - - nsight_compute = prev.nsight_compute.overrideAttrs (oldAttrs: { - nativeBuildInputs = oldAttrs.nativeBuildInputs - ++ (if (lib.versionOlder prev.nsight_compute.version "2022.2.0") - then [ pkgs.qt5.wrapQtAppsHook ] - else [ pkgs.qt6.wrapQtAppsHook ]); - buildInputs = oldAttrs.buildInputs - ++ (if (lib.versionOlder prev.nsight_compute.version "2022.2.0") - then [ pkgs.qt5.qtwebview ] - else [ pkgs.qt6.qtwebview ]); - }); - - nsight_systems = prev.nsight_systems.overrideAttrs (oldAttrs: { - nativeBuildInputs = oldAttrs.nativeBuildInputs ++ [ - pkgs.addOpenGLRunpath - pkgs.qt5.wrapQtAppsHook - ]; - buildInputs = oldAttrs.buildInputs ++ [ - pkgs.alsa-lib - pkgs.e2fsprogs - pkgs.nss - pkgs.numactl - pkgs.pulseaudio - pkgs.wayland - pkgs.xorg.libXcursor - pkgs.xorg.libXdamage - pkgs.xorg.libXrandr - pkgs.xorg.libXtst - ]; - # libcuda needs to be resolved during runtime - autoPatchelfIgnoreMissingDeps = true; - }); - - nvidia_driver = prev.nvidia_driver.overrideAttrs (oldAttrs: { - # libcuda needs to be resolved during runtime - autoPatchelfIgnoreMissingDeps = ["libcuda.so.1"]; - # No need to support this package as we have drivers already - # in linuxPackages. - meta.broken = true; - }); -}) diff --git a/pkgs/development/compilers/cudatoolkit/saxpy/CMakeLists.txt b/pkgs/development/compilers/cudatoolkit/saxpy/CMakeLists.txt deleted file mode 100644 index a6954e6e8bee2..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/saxpy/CMakeLists.txt +++ /dev/null @@ -1,12 +0,0 @@ -cmake_minimum_required(VERSION 3.25) -project(saxpy LANGUAGES CXX CUDA) - -find_package(CUDAToolkit REQUIRED COMPONENTS cudart cublas) - -add_executable(saxpy saxpy.cu) -target_link_libraries(saxpy PUBLIC CUDA::cublas CUDA::cudart m) -target_compile_features(saxpy PRIVATE cxx_std_14) -target_compile_options(saxpy PRIVATE $<$: - --expt-relaxed-constexpr>) - -install(TARGETS saxpy) diff --git a/pkgs/development/compilers/cudatoolkit/saxpy/default.nix b/pkgs/development/compilers/cudatoolkit/saxpy/default.nix deleted file mode 100644 index 2da6da29004dc..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/saxpy/default.nix +++ /dev/null @@ -1,52 +0,0 @@ -{ autoAddOpenGLRunpathHook -, backendStdenv -, cmake -, cuda_cccl ? null -, cuda_cudart ? null -, cudaFlags -, cuda_nvcc ? null -, cudatoolkit ? null -, lib -, libcublas ? null -, setupCudaHook -, stdenv -}: - -backendStdenv.mkDerivation { - pname = "saxpy"; - version = "unstable-2023-07-11"; - - src = ./.; - - buildInputs = lib.optionals (cuda_cudart != null) [ - libcublas - cuda_cudart - cuda_cccl - ] ++ lib.optionals (cuda_cudart == null) [ - cudatoolkit - ]; - nativeBuildInputs = [ - cmake - - # Alternatively, we could remove the propagated hook from cuda_nvcc and add - # directly: - # setupCudaHook - autoAddOpenGLRunpathHook - ] ++ lib.optionals (cuda_nvcc != null) [ - cuda_nvcc - ] ++ lib.optionals (cuda_nvcc == null) [ - cudatoolkit - ]; - - cmakeFlags = [ - "-DCMAKE_VERBOSE_MAKEFILE=ON" - "-DCMAKE_CUDA_ARCHITECTURES=${with cudaFlags; builtins.concatStringsSep ";" (map dropDot cudaCapabilities)}" - ]; - - meta = { - description = "A simple (Single-precision AX Plus Y) FindCUDAToolkit.cmake example for testing cross-compilation"; - license = lib.licenses.mit; - maintainers = lib.teams.cuda.members; - platforms = lib.platforms.unix; - }; -} diff --git a/pkgs/development/compilers/cudatoolkit/saxpy/saxpy.cu b/pkgs/development/compilers/cudatoolkit/saxpy/saxpy.cu deleted file mode 100644 index 912a6d1647b14..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/saxpy/saxpy.cu +++ /dev/null @@ -1,68 +0,0 @@ -#include -#include -#include - -#include - -static inline void check(cudaError_t err, const char *context) { - if (err != cudaSuccess) { - fprintf(stderr, "CUDA error at %s: %s\n", context, cudaGetErrorString(err)); - std::exit(EXIT_FAILURE); - } -} - -#define CHECK(x) check(x, #x) - -__global__ void saxpy(int n, float a, float *x, float *y) { - int i = blockIdx.x * blockDim.x + threadIdx.x; - if (i < n) - y[i] = a * x[i] + y[i]; -} - -int main(void) { - setbuf(stderr, NULL); - fprintf(stderr, "Start\n"); - - int rtVersion, driverVersion; - CHECK(cudaRuntimeGetVersion(&rtVersion)); - CHECK(cudaDriverGetVersion(&driverVersion)); - - fprintf(stderr, "Runtime version: %d\n", rtVersion); - fprintf(stderr, "Driver version: %d\n", driverVersion); - - constexpr int N = 1 << 10; - - std::vector xHost(N), yHost(N); - for (int i = 0; i < N; i++) { - xHost[i] = 1.0f; - yHost[i] = 2.0f; - } - - fprintf(stderr, "Host memory initialized, copying to the device\n"); - fflush(stderr); - - float *xDevice, *yDevice; - CHECK(cudaMalloc(&xDevice, N * sizeof(float))); - CHECK(cudaMalloc(&yDevice, N * sizeof(float))); - - CHECK(cudaMemcpy(xDevice, xHost.data(), N * sizeof(float), - cudaMemcpyHostToDevice)); - CHECK(cudaMemcpy(yDevice, yHost.data(), N * sizeof(float), - cudaMemcpyHostToDevice)); - fprintf(stderr, "Scheduled a cudaMemcpy, calling the kernel\n"); - - saxpy<<<(N + 255) / 256, 256>>>(N, 2.0f, xDevice, yDevice); - fprintf(stderr, "Scheduled a kernel call\n"); - CHECK(cudaGetLastError()); - - CHECK(cudaMemcpy(yHost.data(), yDevice, N * sizeof(float), - cudaMemcpyDeviceToHost)); - - float maxError = 0.0f; - for (int i = 0; i < N; i++) - maxError = max(maxError, abs(yHost[i] - 4.0f)); - fprintf(stderr, "Max error: %f\n", maxError); - - CHECK(cudaFree(xDevice)); - CHECK(cudaFree(yDevice)); -} diff --git a/pkgs/development/compilers/cudatoolkit/stdenv.nix b/pkgs/development/compilers/cudatoolkit/stdenv.nix deleted file mode 100644 index 95e783a682bff..0000000000000 --- a/pkgs/development/compilers/cudatoolkit/stdenv.nix +++ /dev/null @@ -1,33 +0,0 @@ -{ lib -, nixpkgsCompatibleLibstdcxx -, nvccCompatibleCC -, overrideCC -, stdenv -, wrapCCWith -}: - -let - cc = wrapCCWith - { - cc = nvccCompatibleCC; - - # This option is for clang's libcxx, but we (ab)use it for gcc's libstdc++. - # Note that libstdc++ maintains forward-compatibility: if we load a newer - # libstdc++ into the process, we can still use libraries built against an - # older libstdc++. This, in practice, means that we should use libstdc++ from - # the same stdenv that the rest of nixpkgs uses. - # We currently do not try to support anything other than gcc and linux. - libcxx = nixpkgsCompatibleLibstdcxx; - }; - cudaStdenv = overrideCC stdenv cc; - passthruExtra = { - inherit nixpkgsCompatibleLibstdcxx; - # cc already exposed - }; - assertCondition = true; -in -lib.extendDerivation - assertCondition - passthruExtra - cudaStdenv - diff --git a/pkgs/development/cuda-modules/backend-stdenv.nix b/pkgs/development/cuda-modules/backend-stdenv.nix new file mode 100644 index 0000000000000..10fedd1e6f271 --- /dev/null +++ b/pkgs/development/cuda-modules/backend-stdenv.nix @@ -0,0 +1,39 @@ +{ + lib, + nvccCompatibilities, + cudaVersion, + buildPackages, + overrideCC, + stdenv, + wrapCCWith, +}: +let + gccMajorVersion = nvccCompatibilities.${cudaVersion}.gccMaxMajorVersion; + # We use buildPackages (= pkgsBuildHost) because we look for a gcc that + # runs on our build platform, and that produces executables for the host + # platform (= platform on which we deploy and run the downstream packages). + # The target platform of buildPackages.gcc is our host platform, so its + # .lib output should be the libstdc++ we want to be writing in the runpaths + # Cf. https://github.com/NixOS/nixpkgs/pull/225661#discussion_r1164564576 + nixpkgsCompatibleLibstdcxx = buildPackages.gcc.cc.lib; + nvccCompatibleCC = buildPackages."gcc${gccMajorVersion}".cc; + + cc = wrapCCWith { + cc = nvccCompatibleCC; + + # This option is for clang's libcxx, but we (ab)use it for gcc's libstdc++. + # Note that libstdc++ maintains forward-compatibility: if we load a newer + # libstdc++ into the process, we can still use libraries built against an + # older libstdc++. This, in practice, means that we should use libstdc++ from + # the same stdenv that the rest of nixpkgs uses. + # We currently do not try to support anything other than gcc and linux. + libcxx = nixpkgsCompatibleLibstdcxx; + }; + cudaStdenv = overrideCC stdenv cc; + passthruExtra = { + inherit nixpkgsCompatibleLibstdcxx; + # cc already exposed + }; + assertCondition = true; +in +lib.extendDerivation assertCondition passthruExtra cudaStdenv diff --git a/pkgs/development/cuda-modules/cuda/extension.nix b/pkgs/development/cuda-modules/cuda/extension.nix new file mode 100644 index 0000000000000..20ec90d058468 --- /dev/null +++ b/pkgs/development/cuda-modules/cuda/extension.nix @@ -0,0 +1,101 @@ +{cudaVersion, lib}: +let + inherit (lib) attrsets modules trivial; + redistName = "cuda"; + + # Manifest files for CUDA redistributables (aka redist). These can be found at + # https://developer.download.nvidia.com/compute/cuda/redist/ + # Maps a cuda version to the specific version of the manifest. + cudaVersionMap = { + "11.4" = "11.4.4"; + "11.5" = "11.5.2"; + "11.6" = "11.6.2"; + "11.7" = "11.7.1"; + "11.8" = "11.8.0"; + "12.0" = "12.0.1"; + "12.1" = "12.1.1"; + "12.2" = "12.2.2"; + }; + + # Check if the current CUDA version is supported. + cudaVersionMappingExists = builtins.hasAttr cudaVersion cudaVersionMap; + + # fullCudaVersion : String + fullCudaVersion = cudaVersionMap.${cudaVersion}; + + evaluatedModules = modules.evalModules { + modules = [ + ../modules + # We need to nest the manifests in a config.cuda.manifests attribute so the + # module system can evaluate them. + { + cuda.manifests = { + redistrib = trivial.importJSON (./manifests + "/redistrib_${fullCudaVersion}.json"); + feature = trivial.importJSON (./manifests + "/feature_${fullCudaVersion}.json"); + }; + } + ]; + }; + + # Generally we prefer to do things involving getting attribute names with feature_manifest instead + # of redistrib_manifest because the feature manifest will have *only* the redist architecture + # names as the keys, whereas the redistrib manifest will also have things like version, name, license, + # and license_path. + featureManifest = evaluatedModules.config.cuda.manifests.feature; + redistribManifest = evaluatedModules.config.cuda.manifests.redistrib; + + # Builder function which builds a single redist package for a given platform. + # buildRedistPackage : callPackage -> PackageName -> Derivation + buildRedistPackage = + callPackage: pname: + let + redistribRelease = redistribManifest.${pname}; + featureRelease = featureManifest.${pname}; + drv = + (callPackage ../generic-builders/manifest.nix { + # We pass the whole release to the builder because it has logic to handle + # the case we're trying to build on an unsupported platform. + inherit + pname + redistName + redistribRelease + featureRelease + ; + }).overrideAttrs + ( + prevAttrs: { + # Add the package-specific license. + meta = prevAttrs.meta // { + license = + let + licensePath = + if redistribRelease.license_path != null then + redistribRelease.license_path + else + "${pname}/LICENSE.txt"; + url = "https://developer.download.nvidia.com/compute/cuda/redist/${licensePath}"; + in + lib.licenses.nvidiaCudaRedist // {inherit url;}; + }; + } + ); + in + drv; + + # Build all the redist packages given final and prev. + redistPackages = + final: _prev: + # Wrap the whole thing in an optionalAttrs so we can return an empty set if the CUDA version + # is not supported. + # NOTE: We cannot include the call to optionalAttrs *in* the pipe as we would strictly evaluate the + # attrNames before we check if the CUDA version is supported. + attrsets.optionalAttrs cudaVersionMappingExists ( + trivial.pipe featureManifest [ + # Get all the package names + builtins.attrNames + # Build the redist packages + (trivial.flip attrsets.genAttrs (buildRedistPackage final.callPackage)) + ] + ); +in +redistPackages diff --git a/pkgs/development/cuda-modules/cuda/overrides.nix b/pkgs/development/cuda-modules/cuda/overrides.nix new file mode 100644 index 0000000000000..061d5da16bb59 --- /dev/null +++ b/pkgs/development/cuda-modules/cuda/overrides.nix @@ -0,0 +1,129 @@ +{cudaVersion, lib}: +let + inherit (lib) attrsets lists strings; + # cudaVersionOlder : Version -> Boolean + cudaVersionOlder = strings.versionOlder cudaVersion; + # cudaVersionAtLeast : Version -> Boolean + cudaVersionAtLeast = strings.versionAtLeast cudaVersion; + + addBuildInputs = + drv: buildInputs: + drv.overrideAttrs (prevAttrs: {buildInputs = prevAttrs.buildInputs ++ buildInputs;}); +in +# NOTE: Filter out attributes that are not present in the previous version of +# the package set. This is necessary to prevent the appearance of attributes +# like `cuda_nvcc` in `cudaPackages_10_0, which predates redistributables. +final: prev: +attrsets.filterAttrs (attr: _: (builtins.hasAttr attr prev)) { + libcufile = prev.libcufile.overrideAttrs ( + prevAttrs: { + buildInputs = prevAttrs.buildInputs ++ [ + final.libcublas.lib + final.pkgs.numactl + final.pkgs.rdma-core + ]; + # Before 11.7 libcufile depends on itself for some reason. + env.autoPatchelfIgnoreMissingDeps = + prevAttrs.env.autoPatchelfIgnoreMissingDeps + + strings.optionalString (cudaVersionOlder "11.7") " libcufile.so.0"; + } + ); + + libcusolver = addBuildInputs prev.libcusolver ( + # Always depends on this + [final.libcublas.lib] + # Dependency from 12.0 and on + ++ lists.optionals (cudaVersionAtLeast "12.0") [final.libnvjitlink.lib] + # Dependency from 12.1 and on + ++ lists.optionals (cudaVersionAtLeast "12.1") [final.libcusparse.lib] + ); + + libcusparse = addBuildInputs prev.libcusparse ( + lists.optionals (cudaVersionAtLeast "12.0") [final.libnvjitlink.lib] + ); + + cuda_compat = prev.cuda_compat.overrideAttrs ( + prevAttrs: { + env.autoPatchelfIgnoreMissingDeps = + prevAttrs.env.autoPatchelfIgnoreMissingDeps + " libnvrm_gpu.so libnvrm_mem.so"; + # `cuda_compat` only works on aarch64-linux, and only when building for Jetson devices. + brokenConditions = prevAttrs.brokenConditions // { + "Trying to use cuda_compat on aarch64-linux targeting non-Jetson devices" = + !final.flags.isJetsonBuild; + }; + } + ); + + cuda_gdb = addBuildInputs prev.cuda_gdb ( + # x86_64 only needs gmp from 12.0 and on + lists.optionals (cudaVersionAtLeast "12.0") [final.pkgs.gmp] + ); + + cuda_nvcc = prev.cuda_nvcc.overrideAttrs ( + oldAttrs: { + propagatedBuildInputs = [final.setupCudaHook]; + + meta = (oldAttrs.meta or {}) // { + mainProgram = "nvcc"; + }; + } + ); + + cuda_nvprof = prev.cuda_nvprof.overrideAttrs ( + prevAttrs: {buildInputs = prevAttrs.buildInputs ++ [final.cuda_cupti.lib];} + ); + + cuda_demo_suite = addBuildInputs prev.cuda_demo_suite [ + final.pkgs.freeglut + final.pkgs.libGLU + final.pkgs.libglvnd + final.pkgs.mesa + final.libcufft.lib + final.libcurand.lib + ]; + + nsight_compute = prev.nsight_compute.overrideAttrs ( + prevAttrs: { + nativeBuildInputs = + prevAttrs.nativeBuildInputs + ++ ( + if (strings.versionOlder prev.nsight_compute.version "2022.2.0") then + [final.pkgs.qt5.wrapQtAppsHook] + else + [final.pkgs.qt6.wrapQtAppsHook] + ); + buildInputs = + prevAttrs.buildInputs + ++ ( + if (strings.versionOlder prev.nsight_compute.version "2022.2.0") then + [final.pkgs.qt5.qtwebview] + else + [final.pkgs.qt6.qtwebview] + ); + } + ); + + nsight_systems = prev.nsight_systems.overrideAttrs ( + prevAttrs: { + nativeBuildInputs = prevAttrs.nativeBuildInputs ++ [final.pkgs.qt5.wrapQtAppsHook]; + buildInputs = prevAttrs.buildInputs ++ [ + final.pkgs.alsa-lib + final.pkgs.e2fsprogs + final.pkgs.nss + final.pkgs.numactl + final.pkgs.pulseaudio + final.pkgs.wayland + final.pkgs.xorg.libXcursor + final.pkgs.xorg.libXdamage + final.pkgs.xorg.libXrandr + final.pkgs.xorg.libXtst + ]; + } + ); + + nvidia_driver = prev.nvidia_driver.overrideAttrs { + # No need to support this package as we have drivers already + # in linuxPackages. + meta.broken = true; + }; +} diff --git a/pkgs/development/cuda-modules/cudatoolkit/default.nix b/pkgs/development/cuda-modules/cudatoolkit/default.nix new file mode 100644 index 0000000000000..e0868e2b1c2cb --- /dev/null +++ b/pkgs/development/cuda-modules/cudatoolkit/default.nix @@ -0,0 +1,374 @@ +{ + cudaVersion, + runPatches ? [], + autoPatchelfHook, + autoAddOpenGLRunpathHook, + addOpenGLRunpath, + alsa-lib, + curlMinimal, + expat, + fetchurl, + fontconfig, + freetype, + gdk-pixbuf, + glib, + glibc, + gst_all_1, + gtk2, + lib, + libxkbcommon, + libkrb5, + krb5, + makeWrapper, + markForCudatoolkitRootHook, + ncurses5, + numactl, + nss, + patchelf, + perl, + python3, # FIXME: CUDAToolkit 10 may still need python27 + pulseaudio, + setupCudaHook, + stdenv, + backendStdenv, # E.g. gcc11Stdenv, set in extension.nix + unixODBC, + wayland, + xorg, + zlib, + freeglut, + libGLU, + libsForQt5, + libtiff, + qt6Packages, + qt6, + rdma-core, + ucx, + rsync, +}: + +let + # Version info for the classic cudatoolkit packages that contain everything that is in redist. + releases = builtins.import ./releases.nix; + release = releases.${cudaVersion}; +in + +backendStdenv.mkDerivation rec { + pname = "cudatoolkit"; + inherit (release) version; + inherit runPatches; + + dontPatchELF = true; + dontStrip = true; + + src = fetchurl {inherit (release) url sha256;}; + + outputs = [ + "out" + "lib" + "doc" + ]; + + nativeBuildInputs = + [ + perl + makeWrapper + rsync + addOpenGLRunpath + autoPatchelfHook + autoAddOpenGLRunpathHook + markForCudatoolkitRootHook + ] + ++ lib.optionals (lib.versionOlder version "11") [libsForQt5.wrapQtAppsHook] + ++ lib.optionals (lib.versionAtLeast version "11.8") [qt6Packages.wrapQtAppsHook]; + depsTargetTargetPropagated = [setupCudaHook]; + buildInputs = + lib.optionals (lib.versionOlder version "11") [ + libsForQt5.qt5.qtwebengine + freeglut + libGLU + ] + ++ [ + # To get $GDK_PIXBUF_MODULE_FILE via setup-hook + gdk-pixbuf + + # For autoPatchelf + ncurses5 + expat + python3 + zlib + glibc + xorg.libX11 + xorg.libXext + xorg.libXrender + xorg.libXt + xorg.libXtst + xorg.libXi + xorg.libXext + xorg.libXdamage + xorg.libxcb + xorg.xcbutilimage + xorg.xcbutilrenderutil + xorg.xcbutilwm + xorg.xcbutilkeysyms + pulseaudio + libxkbcommon + libkrb5 + krb5 + gtk2 + glib + fontconfig + freetype + numactl + nss + unixODBC + alsa-lib + wayland + ] + ++ lib.optionals (lib.versionAtLeast version "11.8") [ + (lib.getLib libtiff) + qt6Packages.qtwayland + rdma-core + (ucx.override {enableCuda = false;}) # Avoid infinite recursion + xorg.libxshmfence + xorg.libxkbfile + ] + ++ (lib.optionals (lib.versionAtLeast version "12") ( + map lib.getLib ([ + # Used by `/target-linux-x64/CollectX/clx` and `/target-linux-x64/CollectX/libclx_api.so` for: + # - `libcurl.so.4` + curlMinimal + + # Used by `/host-linux-x64/Scripts/WebRTCContainer/setup/neko/server/bin/neko` + gst_all_1.gstreamer + gst_all_1.gst-plugins-base + ]) + ++ ( + with qt6; [ + qtmultimedia + qttools + qtpositioning + qtscxml + qtsvg + qtwebchannel + qtwebengine + ] + ) + )); + + # Prepended to runpaths by autoPatchelf. + # The order inherited from older rpath preFixup code + runtimeDependencies = [ + (placeholder "lib") + (placeholder "out") + "${placeholder "out"}/nvvm" + # NOTE: use the same libstdc++ as the rest of nixpkgs, not from backendStdenv + "${lib.getLib stdenv.cc.cc}/lib64" + "${placeholder "out"}/jre/lib/amd64/jli" + "${placeholder "out"}/lib64" + "${placeholder "out"}/nvvm/lib64" + ]; + + autoPatchelfIgnoreMissingDeps = [ + # This is the hardware-dependent userspace driver that comes from + # nvidia_x11 package. It must be deployed at runtime in + # /run/opengl-driver/lib or pointed at by LD_LIBRARY_PATH variable, rather + # than pinned in runpath + "libcuda.so.1" + + # The krb5 expression ships libcom_err.so.3 but cudatoolkit asks for the + # older + # This dependency is asked for by target-linux-x64/CollectX/RedHat/x86_64/libssl.so.10 + # - do we even want to use nvidia-shipped libssl? + "libcom_err.so.2" + ]; + + preFixup = + if lib.versionOlder version "11" then + '' + ${lib.getExe' patchelf "patchelf"} $out/targets/*/lib/libnvrtc.so --add-needed libnvrtc-builtins.so + '' + else + '' + ${lib.getExe' patchelf "patchelf"} $out/lib64/libnvrtc.so --add-needed libnvrtc-builtins.so + ''; + + unpackPhase = '' + sh $src --keep --noexec + + ${lib.optionalString (lib.versionOlder version "10.1") '' + cd pkg/run_files + sh cuda-linux*.run --keep --noexec + sh cuda-samples*.run --keep --noexec + mv pkg ../../$(basename $src) + cd ../.. + rm -rf pkg + + for patch in $runPatches; do + sh $patch --keep --noexec + mv pkg $(basename $patch) + done + ''} + ''; + + installPhase = + '' + runHook preInstall + mkdir $out + ${lib.optionalString (lib.versionOlder version "10.1") '' + cd $(basename $src) + export PERL5LIB=. + perl ./install-linux.pl --prefix="$out" + cd .. + for patch in $runPatches; do + cd $(basename $patch) + perl ./install_patch.pl --silent --accept-eula --installdir="$out" + cd .. + done + ''} + ${lib.optionalString (lib.versionAtLeast version "10.1" && lib.versionOlder version "11") '' + cd pkg/builds/cuda-toolkit + mv * $out/ + ''} + ${lib.optionalString (lib.versionAtLeast version "11") '' + mkdir -p $out/bin $out/lib64 $out/include $doc + for dir in pkg/builds/* pkg/builds/cuda_nvcc/nvvm pkg/builds/cuda_cupti/extras/CUPTI; do + if [ -d $dir/bin ]; then + mv $dir/bin/* $out/bin + fi + if [ -d $dir/doc ]; then + (cd $dir/doc && find . -type d -exec mkdir -p $doc/\{} \;) + (cd $dir/doc && find . \( -type f -o -type l \) -exec mv \{} $doc/\{} \;) + fi + if [ -L $dir/include ] || [ -d $dir/include ]; then + (cd $dir/include && find . -type d -exec mkdir -p $out/include/\{} \;) + (cd $dir/include && find . \( -type f -o -type l \) -exec mv \{} $out/include/\{} \;) + fi + if [ -L $dir/lib64 ] || [ -d $dir/lib64 ]; then + (cd $dir/lib64 && find . -type d -exec mkdir -p $out/lib64/\{} \;) + (cd $dir/lib64 && find . \( -type f -o -type l \) -exec mv \{} $out/lib64/\{} \;) + fi + done + mv pkg/builds/cuda_nvcc/nvvm $out/nvvm + + mv pkg/builds/cuda_sanitizer_api $out/cuda_sanitizer_api + ln -s $out/cuda_sanitizer_api/compute-sanitizer/compute-sanitizer $out/bin/compute-sanitizer + + mv pkg/builds/nsight_systems/target-linux-x64 $out/target-linux-x64 + mv pkg/builds/nsight_systems/host-linux-x64 $out/host-linux-x64 + rm $out/host-linux-x64/libstdc++.so* + ''} + ${ + lib.optionalString (lib.versionAtLeast version "11.8") + # error: auto-patchelf could not satisfy dependency libtiff.so.5 wanted by /nix/store/.......-cudatoolkit-12.0.1/host-linux-x64/Plugins/imageformats/libqtiff.so + # we only ship libtiff.so.6, so let's use qt plugins built by Nix. + # TODO: don't copy, come up with a symlink-based "merge" + '' + rsync ${lib.getLib qt6Packages.qtimageformats}/lib/qt-6/plugins/ $out/host-linux-x64/Plugins/ -aP + '' + } + + rm -f $out/tools/CUDA_Occupancy_Calculator.xls # FIXME: why? + + ${lib.optionalString (lib.versionOlder version "10.1") '' + # let's remove the 32-bit libraries, they confuse the lib64->lib mover + rm -rf $out/lib + ''} + + ${lib.optionalString (lib.versionAtLeast version "12.0") '' + rm $out/host-linux-x64/libQt6* + ''} + + # Remove some cruft. + ${lib.optionalString ((lib.versionAtLeast version "7.0") && (lib.versionOlder version "10.1")) + "rm $out/bin/uninstall*"} + + # Fixup path to samples (needed for cuda 6.5 or else nsight will not find them) + if [ -d "$out"/cuda-samples ]; then + mv "$out"/cuda-samples "$out"/samples + fi + + # Change the #error on GCC > 4.9 to a #warning. + sed -i $out/include/host_config.h -e 's/#error\(.*unsupported GNU version\)/#warning\1/' + + # Fix builds with newer glibc version + sed -i "1 i#define _BITS_FLOATN_H" "$out/include/host_defines.h" + '' + + + # Point NVCC at a compatible compiler + # CUDA_TOOLKIT_ROOT_DIR is legacy, + # Cf. https://cmake.org/cmake/help/latest/module/FindCUDA.html#input-variables + '' + mkdir -p $out/nix-support + cat <> $out/nix-support/setup-hook + cmakeFlags+=' -DCUDA_TOOLKIT_ROOT_DIR=$out' + EOF + + # Move some libraries to the lib output so that programs that + # depend on them don't pull in this entire monstrosity. + mkdir -p $lib/lib + mv -v $out/lib64/libcudart* $lib/lib/ + + # Remove OpenCL libraries as they are provided by ocl-icd and driver. + rm -f $out/lib64/libOpenCL* + ${lib.optionalString (lib.versionAtLeast version "10.1" && (lib.versionOlder version "11")) '' + mv $out/lib64 $out/lib + mv $out/extras/CUPTI/lib64/libcupti* $out/lib + ''} + + # nvprof do not find any program to profile if LD_LIBRARY_PATH is not set + wrapProgram $out/bin/nvprof \ + --prefix LD_LIBRARY_PATH : $out/lib + '' + + lib.optionalString (lib.versionOlder version "8.0") '' + # Hack to fix building against recent Glibc/GCC. + echo "NIX_CFLAGS_COMPILE+=' -D_FORCE_INLINES'" >> $out/nix-support/setup-hook + '' + # 11.8 includes a broken symlink, include/include, pointing to targets/x86_64-linux/include + + lib.optionalString (lib.versions.majorMinor version == "11.8") '' + rm $out/include/include + '' + + '' + runHook postInstall + ''; + + postInstall = '' + for b in nvvp ${lib.optionalString (lib.versionOlder version "11") "nsight"}; do + wrapProgram "$out/bin/$b" \ + --set GDK_PIXBUF_MODULE_FILE "$GDK_PIXBUF_MODULE_FILE" + done + ''; + + # cuda-gdb doesn't run correctly when not using sandboxing, so + # temporarily disabling the install check. This should be set to true + # when we figure out how to get `cuda-gdb --version` to run correctly + # when not using sandboxing. + doInstallCheck = false; + postInstallCheck = '' + # Smoke test binaries + pushd $out/bin + for f in *; do + case $f in + crt) continue;; + nvcc.profile) continue;; + nsight_ee_plugins_manage.sh) continue;; + uninstall_cuda_toolkit_6.5.pl) continue;; + computeprof|nvvp|nsight) continue;; # GUIs don't feature "--version" + *) echo "Executing '$f --version':"; ./$f --version;; + esac + done + popd + ''; + passthru = { + inherit (backendStdenv) cc; + majorMinorVersion = lib.versions.majorMinor version; + majorVersion = lib.versions.majorMinor version; + }; + + meta = with lib; { + description = "A compiler for NVIDIA GPUs, math libraries, and tools"; + homepage = "https://developer.nvidia.com/cuda-toolkit"; + platforms = ["x86_64-linux"]; + license = licenses.nvidiaCuda; + maintainers = teams.cuda.members; + }; +} diff --git a/pkgs/development/cuda-modules/cudnn/fixup.nix b/pkgs/development/cuda-modules/cudnn/fixup.nix new file mode 100644 index 0000000000000..1fb5a6ad015e4 --- /dev/null +++ b/pkgs/development/cuda-modules/cudnn/fixup.nix @@ -0,0 +1,69 @@ +{ + cudaVersion, + fetchurl, + final, + lib, + package, + patchelf, + zlib, + ... +}: +let + inherit (lib) + lists + maintainers + meta + strings + ; +in +finalAttrs: prevAttrs: { + src = fetchurl {inherit (package) url hash;}; + + # Useful for inspecting why something went wrong. + brokenConditions = + let + cudaTooOld = strings.versionOlder cudaVersion package.minCudaVersion; + cudaTooNew = + (package.maxCudaVersion != null) && strings.versionOlder package.maxCudaVersion cudaVersion; + in + prevAttrs.brokenConditions + // { + "CUDA version is too old" = cudaTooOld; + "CUDA version is too new" = cudaTooNew; + }; + + buildInputs = + prevAttrs.buildInputs + ++ [zlib] + ++ lists.optionals finalAttrs.passthru.useCudatoolkitRunfile [final.cudatoolkit] + ++ lists.optionals (!finalAttrs.passthru.useCudatoolkitRunfile) [final.libcublas.lib]; + + # Tell autoPatchelf about runtime dependencies. + # NOTE: Versions from CUDNN releases have four components. + postFixup = strings.optionalString (strings.versionAtLeast finalAttrs.version "8.0.5.0") '' + ${meta.getExe' patchelf "patchelf"} $lib/lib/libcudnn.so --add-needed libcudnn_cnn_infer.so + ${meta.getExe' patchelf "patchelf"} $lib/lib/libcudnn_ops_infer.so --add-needed libcublas.so --add-needed libcublasLt.so + ''; + + passthru.useCudatoolkitRunfile = strings.versionOlder cudaVersion "11.3.999"; + + meta = prevAttrs.meta // { + homepage = "https://developer.nvidia.com/cudnn"; + maintainers = + prevAttrs.meta.maintainers + ++ ( + with maintainers; [ + mdaiter + samuela + connorbaker + ] + ); + license = { + shortName = "cuDNN EULA"; + fullName = "NVIDIA cuDNN Software License Agreement (EULA)"; + url = "https://docs.nvidia.com/deeplearning/sdk/cudnn-sla/index.html#supplement"; + free = false; + redistributable = !finalAttrs.passthru.useCudatoolkitRunfile; + }; + }; +} diff --git a/pkgs/development/cuda-modules/cudnn/shims.nix b/pkgs/development/cuda-modules/cudnn/shims.nix new file mode 100644 index 0000000000000..e9eca8ef7c8b9 --- /dev/null +++ b/pkgs/development/cuda-modules/cudnn/shims.nix @@ -0,0 +1,13 @@ +# Shims to mimic the shape of ../modules/generic/manifests/{feature,redistrib}/release.nix +{package, redistArch}: +{ + featureRelease.${redistArch}.outputs = { + lib = true; + static = true; + dev = true; + }; + redistribRelease = { + name = "NVIDIA CUDA Deep Neural Network library (cuDNN)"; + inherit (package) version; + }; +} diff --git a/pkgs/development/cuda-modules/cutensor/extension.nix b/pkgs/development/cuda-modules/cutensor/extension.nix new file mode 100644 index 0000000000000..b762fd22ede88 --- /dev/null +++ b/pkgs/development/cuda-modules/cutensor/extension.nix @@ -0,0 +1,164 @@ +# Support matrix can be found at +# https://docs.nvidia.com/deeplearning/cudnn/archives/cudnn-880/support-matrix/index.html +# +# TODO(@connorbaker): +# This is a very similar strategy to CUDA/CUDNN: +# +# - Get all versions supported by the current release of CUDA +# - Build all of them +# - Make the newest the default +# +# Unique twists: +# +# - Instead of providing different releases for each version of CUDA, CuTensor has multiple subdirectories in `lib` +# -- one for each version of CUDA. +{ + cudaVersion, + flags, + hostPlatform, + lib, + mkVersionedPackageName, +}: +let + inherit (lib) + attrsets + lists + modules + versions + strings + trivial + ; + + redistName = "cutensor"; + pname = "libcutensor"; + + cutensorVersions = [ + "1.3.3" + "1.4.0" + "1.5.0" + "1.6.2" + "1.7.0" + ]; + + # Manifests :: { redistrib, feature } + + # Each release of cutensor gets mapped to an evaluated module for that release. + # From there, we can get the min/max CUDA versions supported by that release. + # listOfManifests :: List Manifests + listOfManifests = + let + configEvaluator = + fullCutensorVersion: + modules.evalModules { + modules = [ + ../modules + # We need to nest the manifests in a config.cutensor.manifests attribute so the + # module system can evaluate them. + { + cutensor.manifests = { + redistrib = trivial.importJSON (./manifests + "/redistrib_${fullCutensorVersion}.json"); + feature = trivial.importJSON (./manifests + "/feature_${fullCutensorVersion}.json"); + }; + } + ]; + }; + # Un-nest the manifests attribute set. + releaseGrabber = evaluatedModules: evaluatedModules.config.cutensor.manifests; + in + lists.map + (trivial.flip trivial.pipe [ + configEvaluator + releaseGrabber + ]) + cutensorVersions; + + # Our cudaVersion tells us which version of CUDA we're building against. + # The subdirectories in lib/ tell us which versions of CUDA are supported. + # Typically the names will look like this: + # + # - 10.2 + # - 11 + # - 11.0 + # - 12 + + # libPath :: String + libPath = + let + cudaMajorMinor = versions.majorMinor cudaVersion; + cudaMajor = versions.major cudaVersion; + in + if cudaMajorMinor == "10.2" then cudaMajorMinor else cudaMajor; + + # A release is supported if it has a libPath that matches our CUDA version for our platform. + # LibPath are not constant across the same release -- one platform may support fewer + # CUDA versions than another. + redistArch = flags.getRedistArch hostPlatform.system; + # platformIsSupported :: Manifests -> Boolean + platformIsSupported = + {feature, ...}: + (attrsets.attrByPath + [ + pname + redistArch + ] + null + feature + ) != null; + + # TODO(@connorbaker): With an auxilliary file keeping track of the CUDA versions each release supports, + # we could filter out releases that don't support our CUDA version. + # However, we don't have that currently, so we make a best-effort to try to build TensorRT with whatever + # libPath corresponds to our CUDA version. + # supportedManifests :: List Manifests + supportedManifests = builtins.filter platformIsSupported listOfManifests; + + # Compute versioned attribute name to be used in this package set + # Patch version changes should not break the build, so we only use major and minor + # computeName :: RedistribRelease -> String + computeName = {version, ...}: mkVersionedPackageName redistName version; +in +final: _: +let + # buildCutensorPackage :: Manifests -> AttrSet Derivation + buildCutensorPackage = + {redistrib, feature}: + let + drv = final.callPackage ../generic-builders/manifest.nix { + inherit pname redistName libPath; + redistribRelease = redistrib.${pname}; + featureRelease = feature.${pname}; + }; + fixedDrv = drv.overrideAttrs ( + prevAttrs: { + buildInputs = + prevAttrs.buildInputs + ++ lists.optionals (strings.versionOlder cudaVersion "11.4") [final.cudatoolkit] + ++ lists.optionals (strings.versionAtLeast cudaVersion "11.4") ( + [final.libcublas.lib] + # For some reason, the 1.4.x release of cuTENSOR requires the cudart library. + ++ lists.optionals (strings.hasPrefix "1.4" redistrib.${pname}.version) [final.cuda_cudart.lib] + ); + meta = prevAttrs.meta // { + description = "cuTENSOR: A High-Performance CUDA Library For Tensor Primitives"; + homepage = "https://developer.nvidia.com/cutensor"; + maintainers = prevAttrs.meta.maintainers ++ [lib.maintainers.obsidian-systems-maintenance]; + license = lib.licenses.unfreeRedistributable // { + shortName = "cuTENSOR EULA"; + name = "cuTENSOR SUPPLEMENT TO SOFTWARE LICENSE AGREEMENT FOR NVIDIA SOFTWARE DEVELOPMENT KITS"; + url = "https://docs.nvidia.com/cuda/cutensor/license.html"; + }; + }; + } + ); + in + attrsets.nameValuePair (computeName redistrib.${pname}) fixedDrv; + + extension = + let + nameOfNewest = computeName (lists.last supportedManifests).redistrib.${pname}; + drvs = builtins.listToAttrs (lists.map buildCutensorPackage supportedManifests); + containsDefault = attrsets.optionalAttrs (drvs != {}) {cutensor = drvs.${nameOfNewest};}; + in + drvs // containsDefault; +in +extension diff --git a/pkgs/development/cuda-modules/flags.nix b/pkgs/development/cuda-modules/flags.nix index 7b7922085ff66..139952bc9dfd9 100644 --- a/pkgs/development/cuda-modules/flags.nix +++ b/pkgs/development/cuda-modules/flags.nix @@ -53,11 +53,11 @@ let isDefault = gpu: let - inherit (gpu) dontDefaultAfter; + inherit (gpu) dontDefaultAfter isJetson; newGpu = dontDefaultAfter == null; recentGpu = newGpu || strings.versionAtLeast dontDefaultAfter cudaVersion; in - recentGpu; + recentGpu && !isJetson; # supportedGpus :: List Gpu # GPUs which are supported by the provided CUDA version. @@ -100,11 +100,11 @@ let ]; # Find the intersection with the user-specified list of cudaCapabilities. - # NOTE: Jetson devices are never built by default because they cannot be targeted along + # NOTE: Jetson devices are never built by default because they cannot be targeted along with # non-Jetson devices and require an aarch64 host platform. As such, if they're present anywhere, # they must be in the user-specified cudaCapabilities. # NOTE: We don't need to worry about mixes of Jetson and non-Jetson devices here -- there's - # sanity-checking for all that in cudaFlags. + # sanity-checking for all that in below. jetsonTargets = lists.intersectLists jetsonComputeCapabilities cudaCapabilities; # dropDot :: String -> String @@ -146,14 +146,15 @@ let builtins.throw "Unsupported Nix system: ${nixSystem}"; # Maps NVIDIA redist arch to Nix system. + # It is imperative that we include the boolean condition based on jetsonTargets to ensure + # we don't advertise availability of packages only available on server-grade ARM + # as being available for the Jetson, since both `linux-sbsa` and `linux-aarch64` are + # mapped to the Nix system `aarch64-linux`. getNixSystem = redistArch: - if - lists.elem redistArch [ - "linux-aarch64" - "linux-sbsa" - ] - then + if redistArch == "linux-sbsa" && jetsonTargets == [] then + "aarch64-linux" + else if redistArch == "linux-aarch64" && jetsonTargets != [] then "aarch64-linux" else if redistArch == "linux-x86_64" then "x86_64-linux" @@ -217,26 +218,28 @@ let # isJetsonBuild :: Boolean isJetsonBuild = let - # List of booleans representing whether any of the currently targeted capabilities are - # Jetson devices. - # isJetsons :: List Boolean - isJetsons = - lists.map (trivial.flip builtins.getAttr cudaComputeCapabilityToIsJetson) + requestedJetsonDevices = + lists.filter (cap: cudaComputeCapabilityToIsJetson.${cap}) + cudaCapabilities; + requestedNonJetsonDevices = + lists.filter (cap: !(builtins.elem cap requestedJetsonDevices)) cudaCapabilities; - anyJetsons = lists.any (trivial.id) isJetsons; - allJetsons = lists.all (trivial.id) isJetsons; - hostIsAarch64 = hostPlatform.isAarch64; + jetsonBuildSufficientCondition = requestedJetsonDevices != []; + jetsonBuildNecessaryCondition = requestedNonJetsonDevices == [] && hostPlatform.isAarch64; in - trivial.throwIfNot (anyJetsons -> (allJetsons && hostIsAarch64)) + trivial.throwIf (jetsonBuildSufficientCondition && !jetsonBuildNecessaryCondition) '' Jetson devices cannot be targeted with non-Jetson devices. Additionally, they require hostPlatform to be aarch64. You requested ${builtins.toJSON cudaCapabilities} for host platform ${hostPlatform.system}. + Requested Jetson devices: ${builtins.toJSON requestedJetsonDevices}. + Requested non-Jetson devices: ${builtins.toJSON requestedNonJetsonDevices}. Exactly one of the following must be true: - - All CUDA capabilities belong to Jetson devices (${trivial.boolToString allJetsons}) and the hostPlatform is aarch64 (${trivial.boolToString hostIsAarch64}). - - No CUDA capabilities belong to Jetson devices (${trivial.boolToString (!anyJetsons)}). + - All CUDA capabilities belong to Jetson devices and hostPlatform is aarch64. + - No CUDA capabilities belong to Jetson devices. See ${./gpus.nix} for a list of architectures supported by this version of Nixpkgs. '' - allJetsons; + jetsonBuildSufficientCondition + && jetsonBuildNecessaryCondition; }; in # When changing names or formats: pause, validate, and update the assert @@ -283,7 +286,7 @@ assert let }; actualWrapped = (builtins.tryEval (builtins.deepSeq actual actual)).value; in -asserts.assertMsg (expected == actualWrapped) '' +asserts.assertMsg ((strings.versionAtLeast cudaVersion "11.2") -> (expected == actualWrapped)) '' This test should only fail when using a version of CUDA older than 11.2, the first to support 8.6. Expected: ${builtins.toJSON expected} diff --git a/pkgs/development/cuda-modules/generic-builders/manifest.nix b/pkgs/development/cuda-modules/generic-builders/manifest.nix new file mode 100644 index 0000000000000..01398d1165116 --- /dev/null +++ b/pkgs/development/cuda-modules/generic-builders/manifest.nix @@ -0,0 +1,249 @@ +{ + # General callPackage-supplied arguments + autoAddOpenGLRunpathHook, + autoPatchelfHook, + backendStdenv, + fetchurl, + lib, + lndir, + markForCudatoolkitRootHook, + flags, + stdenv, + hostPlatform, + # Builder-specific arguments + # Short package name (e.g., "cuda_cccl") + # pname : String + pname, + # Common name (e.g., "cutensor" or "cudnn") -- used in the URL. + # Also known as the Redistributable Name. + # redistName : String, + redistName, + # If libPath is non-null, it must be a subdirectory of `lib`. + # The contents of `libPath` will be moved to the root of `lib`. + libPath ? null, + # See ./modules/generic/manifests/redistrib/release.nix + redistribRelease, + # See ./modules/generic/manifests/feature/release.nix + featureRelease, +}: +let + inherit (lib) + attrsets + lists + meta + strings + trivial + licenses + teams + sourceTypes + ; + + # Get the redist architectures for which package provides distributables. + # These are used by meta.platforms. + supportedRedistArchs = builtins.attrNames featureRelease; + redistArch = flags.getRedistArch hostPlatform.system; +in +backendStdenv.mkDerivation ( + finalAttrs: { + # NOTE: Even though there's no actual buildPhase going on here, the derivations of the + # redistributables are sensitive to the compiler flags provided to stdenv. The patchelf package + # is sensitive to the compiler flags provided to stdenv, and we depend on it. As such, we are + # also sensitive to the compiler flags provided to stdenv. + inherit pname; + inherit (redistribRelease) version; + + # Don't force serialization to string for structured attributes, like outputToPatterns + # and brokenConditions. + # Avoids "set cannot be coerced to string" errors. + __structuredAttrs = true; + + # Keep better track of dependencies. + strictDeps = true; + + # TODO(@connorbaker): Update `cuda-redist-find-features` to produce an attrset of boolean values for the + # outputs instead of `has*` attributes. + # NOTE: Outputs are evaluated jointly with meta, so in the case that this is an unsupported platform, + # we still need to provide a list of outputs. + outputs = + let + # Checks whether the redistributable provides an output. + hasOutput = + output: + attrsets.attrByPath + [ + redistArch + "outputs" + output + ] + false + featureRelease; + # Order is important here so we use a list. + additionalOutputs = builtins.filter hasOutput [ + "bin" + "lib" + "static" + "dev" + "doc" + "sample" + "python" + ]; + # The out output is special -- it's the default output and we always include it. + outputs = ["out"] ++ additionalOutputs; + in + outputs; + + # Traversed in the order of the outputs speficied in outputs; + # entries are skipped if they don't exist in outputs. + outputToPatterns = { + bin = ["bin"]; + lib = [ + "lib" + "lib64" + ]; + static = ["**/*.a"]; + sample = ["samples"]; + python = ["**/*.whl"]; + }; + + # Useful for introspecting why something went wrong. + # Maps descriptions of why the derivation would be marked broken to + # booleans indicating whether that description is true. + brokenConditions = {}; + + src = fetchurl { + url = "https://developer.download.nvidia.com/compute/${redistName}/redist/${ + redistribRelease.${redistArch}.relative_path + }"; + inherit (redistribRelease.${redistArch}) sha256; + }; + + # We do need some other phases, like configurePhase, so the multiple-output setup hook works. + dontBuild = true; + + nativeBuildInputs = [ + autoPatchelfHook + # This hook will make sure libcuda can be found + # in typically /lib/opengl-driver by adding that + # directory to the rpath of all ELF binaries. + # Check e.g. with `patchelf --print-rpath path/to/my/binary + autoAddOpenGLRunpathHook + markForCudatoolkitRootHook + ]; + + buildInputs = + [ + # autoPatchelfHook will search for a libstdc++ and we're giving it + # one that is compatible with the rest of nixpkgs, even when + # nvcc forces us to use an older gcc + # NB: We don't actually know if this is the right thing to do + stdenv.cc.cc.lib + ]; + + # Picked up by autoPatchelf + # Needed e.g. for libnvrtc to locate (dlopen) libnvrtc-builtins + appendRunpaths = ["$ORIGIN"]; + + # NOTE: We don't need to check for dev or doc, because those outputs are handled by + # the multiple-outputs setup hook. + # NOTE: moveToOutput operates on all outputs: + # https://github.com/NixOS/nixpkgs/blob/2920b6fc16a9ed5d51429e94238b28306ceda79e/pkgs/build-support/setup-hooks/multiple-outputs.sh#L105-L107 + installPhase = + let + mkMoveToOutputCommand = + output: + let + template = pattern: ''moveToOutput "${pattern}" "${"$" + output}"''; + patterns = finalAttrs.outputToPatterns.${output} or []; + in + strings.concatMapStringsSep "\n" template patterns; + in + # Pre-install hook + '' + runHook preInstall + '' + # Handle the existence of libPath, which requires us to re-arrange the lib directory + + strings.optionalString (libPath != null) '' + if [[ ! -d "${libPath}" ]] ; then + echo "${finalAttrs.pname}: ${libPath} does not exist, only found:" >&2 + find "$(dirname ${libPath})"/ -maxdepth 1 >&2 + echo "This release might not support your CUDA version" >&2 + exit 1 + fi + mv "lib/${libPath}" lib_new + rm -r lib + mv lib_new lib + '' + + '' + mkdir -p "$out" + mv * "$out" + ${strings.concatMapStringsSep "\n" mkMoveToOutputCommand (builtins.tail finalAttrs.outputs)} + runHook postInstall + ''; + + # libcuda needs to be resolved during runtime + # NOTE: Due to the use of __structuredAttrs, we can't use a list for autoPatchelfIgnoreMissingDeps, since it + # will take only the first value. Instead, we produce a string with the values separated by spaces. + # Using the `env` attribute ensures that the value is representable as one of the primitives allowed by + # bash's environment variables. + env.autoPatchelfIgnoreMissingDeps = "libcuda.so libcuda.so.*"; + + # The out output leverages the same functionality which backs the `symlinkJoin` function in + # Nixpkgs: + # https://github.com/NixOS/nixpkgs/blob/d8b2a92df48f9b08d68b0132ce7adfbdbc1fbfac/pkgs/build-support/trivial-builders/default.nix#L510 + # + # That should allow us to emulate "fat" default outputs without having to actually create them. + # + # It is important that this run after the autoPatchelfHook, otherwise the symlinks in out will reference libraries in lib, creating a circular dependency. + postPhases = ["postPatchelf"]; + + # For each output, create a symlink to it in the out output. + # NOTE: We must recreate the out output here, because the setup hook will have deleted it + # if it was empty. + postPatchelf = + let + # Note the double dollar sign -- we want to interpolate the variable in bash, not the string. + mkJoinWithOutOutputCommand = output: ''${meta.getExe lndir} "${"$" + output}" "$out"''; + in + '' + mkdir -p "$out" + ${strings.concatMapStringsSep "\n" mkJoinWithOutOutputCommand (builtins.tail finalAttrs.outputs)} + ''; + + # Make the CUDA-patched stdenv available + passthru.stdenv = backendStdenv; + + # Setting propagatedBuildInputs to false will prevent outputs known to the multiple-outputs + # from depending on `out` by default. + # https://github.com/NixOS/nixpkgs/blob/2920b6fc16a9ed5d51429e94238b28306ceda79e/pkgs/build-support/setup-hooks/multiple-outputs.sh#L196 + # Indeed, we want to do the opposite -- fat "out" outputs that contain all the other outputs. + propagatedBuildOutputs = false; + + # By default, if the dev output exists it just uses that. + # However, because we disabled propagatedBuildOutputs, dev doesn't contain libraries or + # anything of the sort. To remedy this, we set outputSpecified to true, and use + # outputsToInstall, which tells Nix which outputs to use when the package name is used + # unqualified (that is, without an explicit output). + outputSpecified = true; + + meta = { + description = "${redistribRelease.name}. By downloading and using the packages you accept the terms and conditions of the ${finalAttrs.meta.license.shortName}"; + sourceProvenance = [sourceTypes.binaryNativeCode]; + platforms = + lists.concatMap + ( + redistArch: + let + nixSystem = builtins.tryEval (flags.getNixSystem redistArch); + in + if nixSystem.success then [nixSystem.value] else [] + ) + supportedRedistArchs; + broken = lists.any trivial.id (attrsets.attrValues finalAttrs.brokenConditions); + license = licenses.unfree; + maintainers = teams.cuda.members; + # Force the use of the default, fat output by default (even though `dev` exists, which + # causes Nix to prefer that output over the others if outputSpecified isn't set). + outputsToInstall = ["out"]; + }; + } +) diff --git a/pkgs/development/cuda-modules/generic-builders/multiplex.nix b/pkgs/development/cuda-modules/generic-builders/multiplex.nix new file mode 100644 index 0000000000000..b8053094bcc82 --- /dev/null +++ b/pkgs/development/cuda-modules/generic-builders/multiplex.nix @@ -0,0 +1,131 @@ +{ + # callPackage-provided arguments + lib, + cudaVersion, + flags, + hostPlatform, + # Expected to be passed by the caller + mkVersionedPackageName, + # pname :: String + pname, + # releasesModule :: Path + # A path to a module which provides a `releases` attribute + releasesModule, + # shims :: Path + # A path to a module which provides a `shims` attribute + # The redistribRelease is only used in ./manifest.nix for the package version + # and the package description (which NVIDIA's manifest calls the "name"). + # It's also used for fetching the source, but we override that since we can't + # re-use that portion of the functionality (different URLs, etc.). + # The featureRelease is used to populate meta.platforms (by way of looking at the attribute names) + # and to determine the outputs of the package. + # shimFn :: {package, redistArch} -> AttrSet + shimsFn ? ({package, redistArch}: throw "shimsFn must be provided"), + # fixupFn :: Path + # A path (or nix expression) to be evaluated with callPackage and then + # provided to the package's overrideAttrs function. + # It must accept at least the following arguments: + # - final + # - cudaVersion + # - mkVersionedPackageName + # - package + fixupFn ? ( + { + final, + cudaVersion, + mkVersionedPackageName, + package, + ... + }: + throw "fixupFn must be provided" + ), +}: +let + inherit (lib) + attrsets + lists + modules + strings + ; + + evaluatedModules = modules.evalModules { + modules = [ + ../modules + releasesModule + ]; + }; + + # NOTE: Important types: + # - Releases: ../modules/${pname}/releases/releases.nix + # - Package: ../modules/${pname}/releases/package.nix + + # All releases across all platforms + # See ../modules/${pname}/releases/releases.nix + allReleases = evaluatedModules.config.${pname}.releases; + + # Compute versioned attribute name to be used in this package set + # Patch version changes should not break the build, so we only use major and minor + # computeName :: Package -> String + computeName = {version, ...}: mkVersionedPackageName pname version; + + # Check whether a package supports our CUDA version + # isSupported :: Package -> Bool + isSupported = + package: + strings.versionAtLeast cudaVersion package.minCudaVersion + && strings.versionAtLeast package.maxCudaVersion cudaVersion; + + # Get all of the packages for our given platform. + redistArch = flags.getRedistArch hostPlatform.system; + + # All the supported packages we can build for our platform. + # supportedPackages :: List (AttrSet Packages) + supportedPackages = builtins.filter isSupported (allReleases.${redistArch} or []); + + # newestToOldestSupportedPackage :: List (AttrSet Packages) + newestToOldestSupportedPackage = lists.reverseList supportedPackages; + + nameOfNewest = computeName (builtins.head newestToOldestSupportedPackage); + + # A function which takes the `final` overlay and the `package` being built and returns + # a function to be consumed via `overrideAttrs`. + overrideAttrsFixupFn = + final: package: + final.callPackage fixupFn { + inherit + final + cudaVersion + mkVersionedPackageName + package + ; + }; + + extension = + final: _: + let + # Builds our package into derivation and wraps it in a nameValuePair, where the name is the versioned name + # of the package. + buildPackage = + package: + let + shims = final.callPackage shimsFn {inherit package redistArch;}; + name = computeName package; + drv = final.callPackage ./manifest.nix { + inherit pname; + redistName = pname; + inherit (shims) redistribRelease featureRelease; + }; + fixedDrv = drv.overrideAttrs (overrideAttrsFixupFn final package); + in + attrsets.nameValuePair name fixedDrv; + + # versionedDerivations :: AttrSet Derivation + versionedDerivations = builtins.listToAttrs (lists.map buildPackage newestToOldestSupportedPackage); + + defaultDerivation = attrsets.optionalAttrs (versionedDerivations != {}) { + ${pname} = versionedDerivations.${nameOfNewest}; + }; + in + versionedDerivations // defaultDerivation; +in +extension diff --git a/pkgs/development/cuda-modules/gpus.nix b/pkgs/development/cuda-modules/gpus.nix index d170739361383..cf6e0a1eaf360 100644 --- a/pkgs/development/cuda-modules/gpus.nix +++ b/pkgs/development/cuda-modules/gpus.nix @@ -86,7 +86,7 @@ computeCapability = "5.3"; isJetson = true; minCudaVersion = "10.0"; - dontDefaultAfter = "11.0"; + dontDefaultAfter = null; maxCudaVersion = null; } { diff --git a/pkgs/development/cuda-modules/modules/cuda/default.nix b/pkgs/development/cuda-modules/modules/cuda/default.nix new file mode 100644 index 0000000000000..4ea35d0482265 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/cuda/default.nix @@ -0,0 +1 @@ +{options, ...}: {options.cuda.manifests = options.generic.manifests;} diff --git a/pkgs/development/cuda-modules/modules/cudnn/default.nix b/pkgs/development/cuda-modules/modules/cudnn/default.nix new file mode 100644 index 0000000000000..dd52cbaa24b4d --- /dev/null +++ b/pkgs/development/cuda-modules/modules/cudnn/default.nix @@ -0,0 +1,12 @@ +{options, ...}: +{ + options.cudnn.releases = options.generic.releases; + # TODO(@connorbaker): Figure out how to add additional options to the + # to the generic release. + # { + # url = options.mkOption { + # description = "The URL to download the tarball from"; + # type = types.str; + # }; + # } +} diff --git a/pkgs/development/cuda-modules/modules/cutensor/default.nix b/pkgs/development/cuda-modules/modules/cutensor/default.nix new file mode 100644 index 0000000000000..8ec2189fee4cc --- /dev/null +++ b/pkgs/development/cuda-modules/modules/cutensor/default.nix @@ -0,0 +1 @@ +{options, ...}: {options.cutensor.manifests = options.generic.manifests;} diff --git a/pkgs/development/cuda-modules/modules/default.nix b/pkgs/development/cuda-modules/modules/default.nix new file mode 100644 index 0000000000000..ccccd871479e1 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/default.nix @@ -0,0 +1,10 @@ +{ + imports = [ + ./generic + # Always after generic + ./cuda + ./cudnn + ./cutensor + ./tensorrt + ]; +} diff --git a/pkgs/development/cuda-modules/modules/generic/default.nix b/pkgs/development/cuda-modules/modules/generic/default.nix new file mode 100644 index 0000000000000..b68aa614f2400 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/default.nix @@ -0,0 +1,7 @@ +{ + imports = [ + ./types + ./manifests + ./releases + ]; +} diff --git a/pkgs/development/cuda-modules/modules/generic/manifests/default.nix b/pkgs/development/cuda-modules/modules/generic/manifests/default.nix new file mode 100644 index 0000000000000..6c12919ff4000 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/manifests/default.nix @@ -0,0 +1,7 @@ +{lib, config, ...}: +{ + options.generic.manifests = { + feature = import ./feature/manifest.nix {inherit lib config;}; + redistrib = import ./redistrib/manifest.nix {inherit lib;}; + }; +} diff --git a/pkgs/development/cuda-modules/modules/generic/manifests/feature/manifest.nix b/pkgs/development/cuda-modules/modules/generic/manifests/feature/manifest.nix new file mode 100644 index 0000000000000..29ca678e0e5a5 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/manifests/feature/manifest.nix @@ -0,0 +1,10 @@ +{lib, config, ...}: +let + inherit (lib) options trivial types; + Release = import ./release.nix {inherit lib config;}; +in +options.mkOption { + description = "A feature manifest is an attribute set which includes a mapping from package name to release"; + example = trivial.importJSON ../../../../cuda/manifests/feature_11.5.2.json; + type = types.attrsOf Release.type; +} diff --git a/pkgs/development/cuda-modules/modules/generic/manifests/feature/outputs.nix b/pkgs/development/cuda-modules/modules/generic/manifests/feature/outputs.nix new file mode 100644 index 0000000000000..db6dff769e145 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/manifests/feature/outputs.nix @@ -0,0 +1,60 @@ +{lib, ...}: +let + inherit (lib) options types; +in +# https://github.com/ConnorBaker/cuda-redist-find-features/blob/603407bea2fab47f2dfcd88431122a505af95b42/cuda_redist_find_features/manifest/feature/package/package.py +options.mkOption { + description = "A set of outputs that a package can provide."; + example = { + bin = true; + dev = true; + doc = false; + lib = false; + sample = false; + static = false; + }; + type = types.submodule { + options = { + bin = options.mkOption { + description = "A `bin` output requires that we have a non-empty `bin` directory containing at least one file with the executable bit set."; + type = types.bool; + }; + dev = options.mkOption { + description = '' + A `dev` output requires that we have at least one of the following non-empty directories: + + - `include` + - `lib/pkgconfig` + - `share/pkgconfig` + - `lib/cmake` + - `share/aclocal` + ''; + type = types.bool; + }; + doc = options.mkOption { + description = '' + A `doc` output requires that we have at least one of the following non-empty directories: + + - `share/info` + - `share/doc` + - `share/gtk-doc` + - `share/devhelp` + - `share/man` + ''; + type = types.bool; + }; + lib = options.mkOption { + description = "A `lib` output requires that we have a non-empty lib directory containing at least one shared library."; + type = types.bool; + }; + sample = options.mkOption { + description = "A `sample` output requires that we have a non-empty `samples` directory."; + type = types.bool; + }; + static = options.mkOption { + description = "A `static` output requires that we have a non-empty lib directory containing at least one static library."; + type = types.bool; + }; + }; + }; +} diff --git a/pkgs/development/cuda-modules/modules/generic/manifests/feature/package.nix b/pkgs/development/cuda-modules/modules/generic/manifests/feature/package.nix new file mode 100644 index 0000000000000..2c36a3e0cb270 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/manifests/feature/package.nix @@ -0,0 +1,10 @@ +{lib, ...}: +let + inherit (lib) options types; + Outputs = import ./outputs.nix {inherit lib;}; +in +options.mkOption { + description = "A package in the manifest"; + example = (import ./release.nix {inherit lib;}).linux-x86_64; + type = types.submodule {options.outputs = Outputs;}; +} diff --git a/pkgs/development/cuda-modules/modules/generic/manifests/feature/release.nix b/pkgs/development/cuda-modules/modules/generic/manifests/feature/release.nix new file mode 100644 index 0000000000000..be3a30ffdc59d --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/manifests/feature/release.nix @@ -0,0 +1,10 @@ +{lib, config, ...}: +let + inherit (lib) options types; + Package = import ./package.nix {inherit lib config;}; +in +options.mkOption { + description = "A release is an attribute set which includes a mapping from platform to package"; + example = (import ./manifest.nix {inherit lib;}).cuda_cccl; + type = types.attrsOf Package.type; +} diff --git a/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/manifest.nix b/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/manifest.nix new file mode 100644 index 0000000000000..0cfa40241fdc0 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/manifest.nix @@ -0,0 +1,33 @@ +{lib, ...}: +let + inherit (lib) options trivial types; + Release = import ./release.nix {inherit lib;}; +in +options.mkOption { + description = "A redistributable manifest is an attribute set which includes a mapping from package name to release"; + example = trivial.importJSON ../../../../cuda/manifests/redistrib_11.5.2.json; + type = types.submodule { + # Allow any attribute name as these will be the package names + freeformType = types.attrsOf Release.type; + options = { + release_date = options.mkOption { + description = "The release date of the manifest"; + type = types.nullOr types.str; + default = null; + example = "2023-08-29"; + }; + release_label = options.mkOption { + description = "The release label of the manifest"; + type = types.nullOr types.str; + default = null; + example = "12.2.2"; + }; + release_product = options.mkOption { + example = "cuda"; + description = "The release product of the manifest"; + type = types.nullOr types.str; + default = null; + }; + }; + }; +} diff --git a/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/package.nix b/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/package.nix new file mode 100644 index 0000000000000..8d18c06b893f4 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/package.nix @@ -0,0 +1,32 @@ +{lib, ...}: +let + inherit (lib) options types; +in +options.mkOption { + description = "A package in the manifest"; + example = (import ./release.nix {inherit lib;}).linux-x86_64; + type = types.submodule { + options = { + relative_path = options.mkOption { + description = "The relative path to the package"; + example = "cuda_cccl/linux-x86_64/cuda_cccl-linux-x86_64-11.5.62-archive.tar.xz"; + type = types.str; + }; + sha256 = options.mkOption { + description = "The sha256 hash of the package"; + example = "bbe633d6603d5a96a214dcb9f3f6f6fd2fa04d62e53694af97ae0c7afe0121b0"; + type = types.str; + }; + md5 = options.mkOption { + description = "The md5 hash of the package"; + example = "e5deef4f6cb71f14aac5be5d5745dafe"; + type = types.str; + }; + size = options.mkOption { + description = "The size of the package as a string"; + type = types.str; + example = "960968"; + }; + }; + }; +} diff --git a/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/release.nix b/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/release.nix new file mode 100644 index 0000000000000..dd2b206fede41 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/manifests/redistrib/release.nix @@ -0,0 +1,36 @@ +{lib, ...}: +let + inherit (lib) options types; + Package = import ./package.nix {inherit lib;}; +in +options.mkOption { + description = "A release is an attribute set which includes a mapping from platform to package"; + example = (import ./manifest.nix {inherit lib;}).cuda_cccl; + type = types.submodule { + # Allow any attribute name as these will be the platform names + freeformType = types.attrsOf Package.type; + options = { + name = options.mkOption { + description = "The full name of the package"; + example = "CXX Core Compute Libraries"; + type = types.str; + }; + license = options.mkOption { + description = "The license of the package"; + example = "CUDA Toolkit"; + type = types.str; + }; + license_path = options.mkOption { + description = "The path to the license of the package"; + example = "cuda_cccl/LICENSE.txt"; + default = null; + type = types.nullOr types.str; + }; + version = options.mkOption { + description = "The version of the package"; + example = "11.5.62"; + type = types.str; + }; + }; + }; +} diff --git a/pkgs/development/cuda-modules/modules/generic/releases/default.nix b/pkgs/development/cuda-modules/modules/generic/releases/default.nix new file mode 100644 index 0000000000000..8da6f0d5cc79c --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/releases/default.nix @@ -0,0 +1,45 @@ +{lib, config, ...}: +let + inherit (config.generic.types) majorMinorVersion majorMinorPatchBuildVersion; + inherit (lib) options types; +in +{ + options.generic.releases = options.mkOption { + description = "A collection of packages targeting different platforms"; + type = + let + Package = options.mkOption { + description = "A package for a specific platform"; + example = { + version = "8.0.3.4"; + minCudaVersion = "10.2"; + maxCudaVersion = "10.2"; + hash = "sha256-LxcXgwe1OCRfwDsEsNLIkeNsOcx3KuF5Sj+g2dY6WD0="; + }; + type = types.submodule { + # TODO(@connorbaker): Figure out how to extend option sets. + freeformType = types.attrsOf types.anything; + options = { + version = options.mkOption { + description = "The version of the package"; + type = majorMinorPatchBuildVersion; + }; + minCudaVersion = options.mkOption { + description = "The minimum CUDA version supported"; + type = majorMinorVersion; + }; + maxCudaVersion = options.mkOption { + description = "The maximum CUDA version supported"; + type = majorMinorVersion; + }; + hash = options.mkOption { + description = "The hash of the tarball"; + type = types.str; + }; + }; + }; + }; + in + types.attrsOf (types.listOf Package.type); + }; +} diff --git a/pkgs/development/cuda-modules/modules/generic/types/default.nix b/pkgs/development/cuda-modules/modules/generic/types/default.nix new file mode 100644 index 0000000000000..61d13b3cc8d2b --- /dev/null +++ b/pkgs/development/cuda-modules/modules/generic/types/default.nix @@ -0,0 +1,39 @@ +{lib, ...}: +let + inherit (lib) options types; +in +{ + options.generic.types = options.mkOption { + type = types.attrsOf types.optionType; + default = {}; + description = "A set of generic types."; + }; + config.generic.types = { + cudaArch = types.strMatching "^sm_[[:digit:]]+[a-z]?$" // { + name = "cudaArch"; + description = "A CUDA architecture name."; + }; + # https://github.com/ConnorBaker/cuda-redist-find-features/blob/c841980e146f8664bbcd0ba1399e486b7910617b/cuda_redist_find_features/types/_lib_so_name.py + libSoName = types.strMatching ".*\\.so(\\.[[:digit:]]+)*$" // { + name = "libSoName"; + description = "The name of a shared object file."; + }; + + majorMinorVersion = types.strMatching "^([[:digit:]]+)\\.([[:digit:]]+)$" // { + name = "majorMinorVersion"; + description = "A version number with a major and minor component."; + }; + + majorMinorPatchVersion = types.strMatching "^([[:digit:]]+)\\.([[:digit:]]+)\\.([[:digit:]]+)$" // { + name = "majorMinorPatchVersion"; + description = "A version number with a major, minor, and patch component."; + }; + + majorMinorPatchBuildVersion = + types.strMatching "^([[:digit:]]+)\\.([[:digit:]]+)\\.([[:digit:]]+)\\.([[:digit:]]+)$" + // { + name = "majorMinorPatchBuildVersion"; + description = "A version number with a major, minor, patch, and build component."; + }; + }; +} diff --git a/pkgs/development/cuda-modules/modules/tensorrt/default.nix b/pkgs/development/cuda-modules/modules/tensorrt/default.nix new file mode 100644 index 0000000000000..e62942c679aa0 --- /dev/null +++ b/pkgs/development/cuda-modules/modules/tensorrt/default.nix @@ -0,0 +1,16 @@ +{options, ...}: +{ + options.tensorrt.releases = options.generic.releases; + # TODO(@connorbaker): Figure out how to add additional options to the + # to the generic release. + # { + # cudnnVersion = lib.options.mkOption { + # description = "The CUDNN version supported"; + # type = types.nullOr majorMinorVersion; + # }; + # filename = lib.options.mkOption { + # description = "The tarball name"; + # type = types.str; + # }; + # } +} diff --git a/pkgs/development/cuda-modules/nccl-tests/default.nix b/pkgs/development/cuda-modules/nccl-tests/default.nix new file mode 100644 index 0000000000000..5c2f29b7ed564 --- /dev/null +++ b/pkgs/development/cuda-modules/nccl-tests/default.nix @@ -0,0 +1,84 @@ +# NOTE: Though NCCL tests is called within the cudaPackages package set, we avoid passing in +# the names of dependencies from that package set directly to avoid evaluation errors +# in the case redistributable packages are not available. +{ + config, + cudaPackages, + fetchFromGitHub, + gitUpdater, + lib, + mpi, + mpiSupport ? false, + which, +}: +let + inherit (cudaPackages) + backendStdenv + cuda_cccl + cuda_cudart + cuda_nvcc + cudatoolkit + cudaVersion + nccl + ; +in +backendStdenv.mkDerivation ( + finalAttrs: { + + pname = "nccl-tests"; + version = "2.13.8"; + + src = fetchFromGitHub { + owner = "NVIDIA"; + repo = finalAttrs.pname; + rev = "v${finalAttrs.version}"; + hash = "sha256-dxLoflsTHDBnZRTzoXdm30OyKpLlRa73b784YWALBHg="; + }; + + strictDeps = true; + + nativeBuildInputs = + [which] + ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] + ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [cuda_nvcc]; + + buildInputs = + [nccl] + ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] + ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [ + cuda_nvcc.dev # crt/host_config.h + cuda_cudart + ] + ++ lib.optionals (lib.versionAtLeast cudaVersion "12.0") [ + cuda_cccl.dev # + ] + ++ lib.optionals mpiSupport [mpi]; + + makeFlags = + ["NCCL_HOME=${nccl}"] + ++ lib.optionals (lib.versionOlder cudaVersion "11.4") ["CUDA_HOME=${cudatoolkit}"] + ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") ["CUDA_HOME=${cuda_nvcc}"] + ++ lib.optionals mpiSupport ["MPI=1"]; + + enableParallelBuilding = true; + + installPhase = '' + mkdir -p $out/bin + cp -r build/* $out/bin/ + ''; + + passthru.updateScript = gitUpdater { + inherit (finalAttrs) pname version; + rev-prefix = "v"; + }; + + meta = with lib; { + description = "Tests to check both the performance and the correctness of NVIDIA NCCL operations"; + homepage = "https://github.com/NVIDIA/nccl-tests"; + platforms = platforms.linux; + license = licenses.bsd3; + broken = !config.cudaSupport || (mpiSupport && mpi == null); + maintainers = with maintainers; [jmillerpdt] ++ teams.cuda.members; + }; + } +) diff --git a/pkgs/development/cuda-modules/nccl/default.nix b/pkgs/development/cuda-modules/nccl/default.nix new file mode 100644 index 0000000000000..c56d59cb42068 --- /dev/null +++ b/pkgs/development/cuda-modules/nccl/default.nix @@ -0,0 +1,112 @@ +# NOTE: Though NCCL is called within the cudaPackages package set, we avoid passing in +# the names of dependencies from that package set directly to avoid evaluation errors +# in the case redistributable packages are not available. +{ + lib, + fetchFromGitHub, + python3, + which, + cudaPackages, + # passthru.updateScript + gitUpdater, +}: +let + inherit (cudaPackages) + autoAddOpenGLRunpathHook + backendStdenv + cuda_cccl + cuda_cudart + cuda_nvcc + cudaFlags + cudatoolkit + cudaVersion + ; +in +backendStdenv.mkDerivation ( + finalAttrs: { + pname = "nccl"; + version = "2.19.3-1"; + + src = fetchFromGitHub { + owner = "NVIDIA"; + repo = finalAttrs.pname; + rev = "v${finalAttrs.version}"; + hash = "sha256-59FlOKM5EB5Vkm4dZBRCkn+IgIcdQehE+FyZAdTCT/A="; + }; + + strictDeps = true; + + outputs = [ + "out" + "dev" + ]; + + nativeBuildInputs = + [ + which + autoAddOpenGLRunpathHook + python3 + ] + ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] + ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [cuda_nvcc]; + + buildInputs = + lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] + ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [ + cuda_nvcc.dev # crt/host_config.h + cuda_cudart + ] + # NOTE: CUDA versions in Nixpkgs only use a major and minor version. When we do comparisons + # against other version, like below, it's important that we use the same format. Otherwise, + # we'll get incorrect results. + # For example, lib.versionAtLeast "12.0" "12.0.0" == false. + ++ lib.optionals (lib.versionAtLeast cudaVersion "12.0") [cuda_cccl]; + + env.NIX_CFLAGS_COMPILE = toString ["-Wno-unused-function"]; + + preConfigure = '' + patchShebangs ./src/device/generate.py + makeFlagsArray+=( + "NVCC_GENCODE=${lib.concatStringsSep " " cudaFlags.gencode}" + ) + ''; + + makeFlags = + ["PREFIX=$(out)"] + ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [ + "CUDA_HOME=${cudatoolkit}" + "CUDA_LIB=${lib.getLib cudatoolkit}/lib" + "CUDA_INC=${lib.getDev cudatoolkit}/include" + ] + ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [ + "CUDA_HOME=${cuda_nvcc}" + "CUDA_LIB=${lib.getLib cuda_cudart}/lib" + "CUDA_INC=${lib.getDev cuda_cudart}/include" + ]; + + enableParallelBuilding = true; + + postFixup = '' + moveToOutput lib/libnccl_static.a $dev + ''; + + passthru.updateScript = gitUpdater { + inherit (finalAttrs) pname version; + rev-prefix = "v"; + }; + + meta = with lib; { + description = "Multi-GPU and multi-node collective communication primitives for NVIDIA GPUs"; + homepage = "https://developer.nvidia.com/nccl"; + license = licenses.bsd3; + platforms = platforms.linux; + maintainers = + with maintainers; + [ + mdaiter + orivej + ] + ++ teams.cuda.members; + }; + } +) diff --git a/pkgs/development/cuda-modules/nvcc-compatibilities.nix b/pkgs/development/cuda-modules/nvcc-compatibilities.nix index 50f05f8af8ebf..3981869e4b315 100644 --- a/pkgs/development/cuda-modules/nvcc-compatibilities.nix +++ b/pkgs/development/cuda-modules/nvcc-compatibilities.nix @@ -70,16 +70,22 @@ let # Added support for Clang 12 and GCC 11 # https://docs.nvidia.com/cuda/archive/11.4.4/cuda-toolkit-release-notes/index.html#cuda-general-new-features - "11.4" = { + "11.4" = attrs."11.3" // { clangMaxMajorVersion = "12"; - gccMaxMajorVersion = "11"; + # NOTE: There is a bug in the version of GLIBC that GCC 11 uses which causes it to fail to compile some CUDA + # code. As such, we skip it for this release, and do the bump in 11.6 (skipping 11.5). + # https://forums.developer.nvidia.com/t/cuda-11-5-samples-throw-multiple-error-attribute-malloc-does-not-take-arguments/192750/15 + # gccMaxMajorVersion = "11"; }; # No changes from 11.4 to 11.5 "11.5" = attrs."11.4"; # No changes from 11.5 to 11.6 - "11.6" = attrs."11.5"; + # However, as mentioned above, we add GCC 11 this release. + "11.6" = attrs."11.5" // { + gccMaxMajorVersion = "11"; + }; # Added support for Clang 13 # https://docs.nvidia.com/cuda/archive/11.7.1/cuda-toolkit-release-notes/index.html#cuda-compiler-new-features diff --git a/pkgs/development/cuda-modules/saxpy/CMakeLists.txt b/pkgs/development/cuda-modules/saxpy/CMakeLists.txt new file mode 100644 index 0000000000000..a6954e6e8bee2 --- /dev/null +++ b/pkgs/development/cuda-modules/saxpy/CMakeLists.txt @@ -0,0 +1,12 @@ +cmake_minimum_required(VERSION 3.25) +project(saxpy LANGUAGES CXX CUDA) + +find_package(CUDAToolkit REQUIRED COMPONENTS cudart cublas) + +add_executable(saxpy saxpy.cu) +target_link_libraries(saxpy PUBLIC CUDA::cublas CUDA::cudart m) +target_compile_features(saxpy PRIVATE cxx_std_14) +target_compile_options(saxpy PRIVATE $<$: + --expt-relaxed-constexpr>) + +install(TARGETS saxpy) diff --git a/pkgs/development/cuda-modules/saxpy/default.nix b/pkgs/development/cuda-modules/saxpy/default.nix new file mode 100644 index 0000000000000..fff52801e3cca --- /dev/null +++ b/pkgs/development/cuda-modules/saxpy/default.nix @@ -0,0 +1,56 @@ +{ + cmake, + cudaPackages, + lib, +}: +let + inherit (cudaPackages) + autoAddOpenGLRunpathHook + backendStdenv + cuda_cccl + cuda_cudart + cuda_nvcc + cudatoolkit + cudaVersion + flags + libcublas + ; +in +backendStdenv.mkDerivation { + pname = "saxpy"; + version = "unstable-2023-07-11"; + + src = ./.; + + strictDeps = true; + + nativeBuildInputs = + [ + cmake + autoAddOpenGLRunpathHook + ] + ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] + ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [cuda_nvcc]; + + buildInputs = + lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] + ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [ + libcublas + cuda_cudart + ] + ++ lib.optionals (lib.versionAtLeast cudaVersion "12.0") [cuda_cccl]; + + cmakeFlags = [ + (lib.cmakeBool "CMAKE_VERBOSE_MAKEFILE" true) + (lib.cmakeFeature "CMAKE_CUDA_ARCHITECTURES" ( + with flags; lib.concatStringsSep ";" (lib.lists.map dropDot cudaCapabilities) + )) + ]; + + meta = { + description = "A simple (Single-precision AX Plus Y) FindCUDAToolkit.cmake example for testing cross-compilation"; + license = lib.licenses.mit; + maintainers = lib.teams.cuda.members; + platforms = lib.platforms.unix; + }; +} diff --git a/pkgs/development/cuda-modules/saxpy/saxpy.cu b/pkgs/development/cuda-modules/saxpy/saxpy.cu new file mode 100644 index 0000000000000..912a6d1647b14 --- /dev/null +++ b/pkgs/development/cuda-modules/saxpy/saxpy.cu @@ -0,0 +1,68 @@ +#include +#include +#include + +#include + +static inline void check(cudaError_t err, const char *context) { + if (err != cudaSuccess) { + fprintf(stderr, "CUDA error at %s: %s\n", context, cudaGetErrorString(err)); + std::exit(EXIT_FAILURE); + } +} + +#define CHECK(x) check(x, #x) + +__global__ void saxpy(int n, float a, float *x, float *y) { + int i = blockIdx.x * blockDim.x + threadIdx.x; + if (i < n) + y[i] = a * x[i] + y[i]; +} + +int main(void) { + setbuf(stderr, NULL); + fprintf(stderr, "Start\n"); + + int rtVersion, driverVersion; + CHECK(cudaRuntimeGetVersion(&rtVersion)); + CHECK(cudaDriverGetVersion(&driverVersion)); + + fprintf(stderr, "Runtime version: %d\n", rtVersion); + fprintf(stderr, "Driver version: %d\n", driverVersion); + + constexpr int N = 1 << 10; + + std::vector xHost(N), yHost(N); + for (int i = 0; i < N; i++) { + xHost[i] = 1.0f; + yHost[i] = 2.0f; + } + + fprintf(stderr, "Host memory initialized, copying to the device\n"); + fflush(stderr); + + float *xDevice, *yDevice; + CHECK(cudaMalloc(&xDevice, N * sizeof(float))); + CHECK(cudaMalloc(&yDevice, N * sizeof(float))); + + CHECK(cudaMemcpy(xDevice, xHost.data(), N * sizeof(float), + cudaMemcpyHostToDevice)); + CHECK(cudaMemcpy(yDevice, yHost.data(), N * sizeof(float), + cudaMemcpyHostToDevice)); + fprintf(stderr, "Scheduled a cudaMemcpy, calling the kernel\n"); + + saxpy<<<(N + 255) / 256, 256>>>(N, 2.0f, xDevice, yDevice); + fprintf(stderr, "Scheduled a kernel call\n"); + CHECK(cudaGetLastError()); + + CHECK(cudaMemcpy(yHost.data(), yDevice, N * sizeof(float), + cudaMemcpyDeviceToHost)); + + float maxError = 0.0f; + for (int i = 0; i < N; i++) + maxError = max(maxError, abs(yHost[i] - 4.0f)); + fprintf(stderr, "Max error: %f\n", maxError); + + CHECK(cudaFree(xDevice)); + CHECK(cudaFree(yDevice)); +} diff --git a/pkgs/development/cuda-modules/setup-hooks/auto-add-opengl-runpath-hook.sh b/pkgs/development/cuda-modules/setup-hooks/auto-add-opengl-runpath-hook.sh new file mode 100644 index 0000000000000..f50a5f6c25c66 --- /dev/null +++ b/pkgs/development/cuda-modules/setup-hooks/auto-add-opengl-runpath-hook.sh @@ -0,0 +1,28 @@ +# shellcheck shell=bash +# Run addOpenGLRunpath on all dynamically linked, ELF files +echo "Sourcing auto-add-opengl-runpath-hook" + +elfHasDynamicSection() { + patchelf --print-rpath "$1" >& /dev/null +} + +autoAddOpenGLRunpathPhase() ( + local outputPaths + mapfile -t outputPaths < <(for o in $(getAllOutputNames); do echo "${!o}"; done) + find "${outputPaths[@]}" -type f -executable -print0 | while IFS= read -rd "" f; do + if isELF "$f"; then + # patchelf returns an error on statically linked ELF files + if elfHasDynamicSection "$f" ; then + echo "autoAddOpenGLRunpathHook: patching $f" + addOpenGLRunpath "$f" + elif (( "${NIX_DEBUG:-0}" >= 1 )) ; then + echo "autoAddOpenGLRunpathHook: skipping a statically-linked ELF file $f" + fi + fi + done +) + +if [ -z "${dontUseAutoAddOpenGLRunpath-}" ]; then + echo "Using autoAddOpenGLRunpathPhase" + postFixupHooks+=(autoAddOpenGLRunpathPhase) +fi diff --git a/pkgs/development/cuda-modules/setup-hooks/extension.nix b/pkgs/development/cuda-modules/setup-hooks/extension.nix new file mode 100644 index 0000000000000..762dad9ea8765 --- /dev/null +++ b/pkgs/development/cuda-modules/setup-hooks/extension.nix @@ -0,0 +1,47 @@ +final: _: { + # Internal hook, used by cudatoolkit and cuda redist packages + # to accommodate automatic CUDAToolkit_ROOT construction + markForCudatoolkitRootHook = + final.callPackage + ( + {makeSetupHook}: + makeSetupHook {name = "mark-for-cudatoolkit-root-hook";} ./mark-for-cudatoolkit-root-hook.sh + ) + {}; + + # Currently propagated by cuda_nvcc or cudatoolkit, rather than used directly + setupCudaHook = + (final.callPackage + ( + {makeSetupHook, backendStdenv}: + makeSetupHook + { + name = "setup-cuda-hook"; + + substitutions.setupCudaHook = placeholder "out"; + + # Point NVCC at a compatible compiler + substitutions.ccRoot = "${backendStdenv.cc}"; + + # Required in addition to ccRoot as otherwise bin/gcc is looked up + # when building CMakeCUDACompilerId.cu + substitutions.ccFullPath = "${backendStdenv.cc}/bin/${backendStdenv.cc.targetPrefix}c++"; + } + ./setup-cuda-hook.sh + ) + {} + ); + + autoAddOpenGLRunpathHook = + final.callPackage + ( + {addOpenGLRunpath, makeSetupHook}: + makeSetupHook + { + name = "auto-add-opengl-runpath-hook"; + propagatedBuildInputs = [addOpenGLRunpath]; + } + ./auto-add-opengl-runpath-hook.sh + ) + {}; +} diff --git a/pkgs/development/cuda-modules/setup-hooks/mark-for-cudatoolkit-root-hook.sh b/pkgs/development/cuda-modules/setup-hooks/mark-for-cudatoolkit-root-hook.sh new file mode 100644 index 0000000000000..ba04c2e0806af --- /dev/null +++ b/pkgs/development/cuda-modules/setup-hooks/mark-for-cudatoolkit-root-hook.sh @@ -0,0 +1,14 @@ +# shellcheck shell=bash + +# Should we mimick cc-wrapper's "hygiene"? +[[ -z ${strictDeps-} ]] || (( "$hostOffset" < 0 )) || return 0 + +echo "Sourcing mark-for-cudatoolkit-root-hook" >&2 + +markForCUDAToolkit_ROOT() { + mkdir -p "${prefix}/nix-support" + [[ -f "${prefix}/nix-support/include-in-cudatoolkit-root" ]] && return + echo "$pname-$output" > "${prefix}/nix-support/include-in-cudatoolkit-root" +} + +fixupOutputHooks+=(markForCUDAToolkit_ROOT) diff --git a/pkgs/development/cuda-modules/setup-hooks/setup-cuda-hook.sh b/pkgs/development/cuda-modules/setup-hooks/setup-cuda-hook.sh new file mode 100644 index 0000000000000..7b7b3bdde80e3 --- /dev/null +++ b/pkgs/development/cuda-modules/setup-hooks/setup-cuda-hook.sh @@ -0,0 +1,139 @@ +# shellcheck shell=bash + +# Only run the hook from nativeBuildInputs +(( "$hostOffset" == -1 && "$targetOffset" == 0)) || return 0 + +guard=Sourcing +reason= + +[[ -n ${cudaSetupHookOnce-} ]] && guard=Skipping && reason=" because the hook has been propagated more than once" + +if (( "${NIX_DEBUG:-0}" >= 1 )) ; then + echo "$guard hostOffset=$hostOffset targetOffset=$targetOffset setupCudaHook$reason" >&2 +else + echo "$guard setup-cuda-hook$reason" >&2 +fi + +[[ "$guard" = Sourcing ]] || return 0 + +declare -g cudaSetupHookOnce=1 +declare -Ag cudaHostPathsSeen=() +declare -Ag cudaOutputToPath=() + +extendcudaHostPathsSeen() { + (( "${NIX_DEBUG:-0}" >= 1 )) && echo "extendcudaHostPathsSeen $1" >&2 + + local markerPath="$1/nix-support/include-in-cudatoolkit-root" + [[ ! -f "${markerPath}" ]] && return + [[ -v cudaHostPathsSeen[$1] ]] && return + + cudaHostPathsSeen["$1"]=1 + + # E.g. cuda_cudart-lib + local cudaOutputName + read -r cudaOutputName < "$markerPath" + + [[ -z "$cudaOutputName" ]] && return + + local oldPath="${cudaOutputToPath[$cudaOutputName]-}" + [[ -n "$oldPath" ]] && echo "extendcudaHostPathsSeen: warning: overwriting $cudaOutputName from $oldPath to $1" >&2 + cudaOutputToPath["$cudaOutputName"]="$1" +} +addEnvHooks "$targetOffset" extendcudaHostPathsSeen + +setupCUDAToolkit_ROOT() { + (( "${NIX_DEBUG:-0}" >= 1 )) && echo "setupCUDAToolkit_ROOT: cudaHostPathsSeen=${!cudaHostPathsSeen[*]}" >&2 + + for path in "${!cudaHostPathsSeen[@]}" ; do + addToSearchPathWithCustomDelimiter ";" CUDAToolkit_ROOT "$path" + if [[ -d "$path/include" ]] ; then + addToSearchPathWithCustomDelimiter ";" CUDAToolkit_INCLUDE_DIR "$path/include" + fi + done + + export cmakeFlags+=" -DCUDAToolkit_INCLUDE_DIR=$CUDAToolkit_INCLUDE_DIR -DCUDAToolkit_ROOT=$CUDAToolkit_ROOT" +} +preConfigureHooks+=(setupCUDAToolkit_ROOT) + +setupCUDAToolkitCompilers() { + echo Executing setupCUDAToolkitCompilers >&2 + + if [[ -n "${dontSetupCUDAToolkitCompilers-}" ]] ; then + return + fi + + # Point NVCC at a compatible compiler + + # For CMake-based projects: + # https://cmake.org/cmake/help/latest/module/FindCUDA.html#input-variables + # https://cmake.org/cmake/help/latest/envvar/CUDAHOSTCXX.html + # https://cmake.org/cmake/help/latest/variable/CMAKE_CUDA_HOST_COMPILER.html + + export cmakeFlags+=" -DCUDA_HOST_COMPILER=@ccFullPath@" + export cmakeFlags+=" -DCMAKE_CUDA_HOST_COMPILER=@ccFullPath@" + + # For non-CMake projects: + # We prepend --compiler-bindir to nvcc flags. + # Downstream packages can override these, because NVCC + # uses the last --compiler-bindir it gets on the command line. + # FIXME: this results in "incompatible redefinition" warnings. + # https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html#compiler-bindir-directory-ccbin + if [ -z "${CUDAHOSTCXX-}" ]; then + export CUDAHOSTCXX="@ccFullPath@"; + fi + + export NVCC_PREPEND_FLAGS+=" --compiler-bindir=@ccRoot@/bin" + + # NOTE: We set -Xfatbin=-compress-all, which reduces the size of the compiled + # binaries. If binaries grow over 2GB, they will fail to link. This is a problem for us, as + # the default set of CUDA capabilities we build can regularly cause this to occur (for + # example, with Magma). + # + # @SomeoneSerge: original comment was made by @ConnorBaker in .../cudatoolkit/common.nix + if [[ -z "${dontCompressFatbin-}" ]]; then + export NVCC_PREPEND_FLAGS+=" -Xfatbin=-compress-all" + fi + + # CMake's enable_language(CUDA) runs a compiler test and it doesn't account for + # CUDAToolkit_ROOT. We have to help it locate libcudart + if [[ -z "${nvccDontPrependCudartFlags-}" ]] ; then + if [[ ! -v cudaOutputToPath["cuda_cudart-out"] ]] ; then + echo "setupCUDAToolkitCompilers: missing cudaPackages.cuda_cudart. This may become an an error in the future" >&2 + # exit 1 + fi + for pkg in "${!cudaOutputToPath[@]}" ; do + [[ ! "$pkg" = cuda_cudart* ]] && continue + + local path="${cudaOutputToPath[$pkg]}" + if [[ -d "$path/include" ]] ; then + export NVCC_PREPEND_FLAGS+=" -I$path/include" + fi + if [[ -d "$path/lib" ]] ; then + export NVCC_PREPEND_FLAGS+=" -L$path/lib" + fi + done + fi +} +preConfigureHooks+=(setupCUDAToolkitCompilers) + +propagateCudaLibraries() { + (( "${NIX_DEBUG:-0}" >= 1 )) && echo "propagateCudaLibraries: cudaPropagateToOutput=$cudaPropagateToOutput cudaHostPathsSeen=${!cudaHostPathsSeen[*]}" >&2 + + [[ -z "${cudaPropagateToOutput-}" ]] && return + + mkdir -p "${!cudaPropagateToOutput}/nix-support" + # One'd expect this should be propagated-bulid-build-deps, but that doesn't seem to work + echo "@setupCudaHook@" >> "${!cudaPropagateToOutput}/nix-support/propagated-native-build-inputs" + + local propagatedBuildInputs=( "${!cudaHostPathsSeen[@]}" ) + for output in $(getAllOutputNames) ; do + if [[ ! "$output" = "$cudaPropagateToOutput" ]] ; then + propagatedBuildInputs+=( "${!output}" ) + fi + break + done + + # One'd expect this should be propagated-host-host-deps, but that doesn't seem to work + printWords "${propagatedBuildInputs[@]}" >> "${!cudaPropagateToOutput}/nix-support/propagated-build-inputs" +} +postFixupHooks+=(propagateCudaLibraries) diff --git a/pkgs/development/cuda-modules/tensorrt/fixup.nix b/pkgs/development/cuda-modules/tensorrt/fixup.nix new file mode 100644 index 0000000000000..d713189328ed7 --- /dev/null +++ b/pkgs/development/cuda-modules/tensorrt/fixup.nix @@ -0,0 +1,113 @@ +{ + cudaVersion, + final, + hostPlatform, + lib, + mkVersionedPackageName, + package, + patchelf, + requireFile, + ... +}: +let + inherit (lib) + maintainers + meta + strings + versions + ; +in +finalAttrs: prevAttrs: { + # Useful for inspecting why something went wrong. + brokenConditions = + let + cudaTooOld = strings.versionOlder cudaVersion package.minCudaVersion; + cudaTooNew = + (package.maxCudaVersion != null) && strings.versionOlder package.maxCudaVersion cudaVersion; + cudnnVersionIsSpecified = package.cudnnVersion != null; + cudnnVersionSpecified = versions.majorMinor package.cudnnVersion; + cudnnVersionProvided = versions.majorMinor finalAttrs.passthru.cudnn.version; + cudnnTooOld = + cudnnVersionIsSpecified && (strings.versionOlder cudnnVersionProvided cudnnVersionSpecified); + cudnnTooNew = + cudnnVersionIsSpecified && (strings.versionOlder cudnnVersionSpecified cudnnVersionProvided); + in + prevAttrs.brokenConditions + // { + "CUDA version is too old" = cudaTooOld; + "CUDA version is too new" = cudaTooNew; + "CUDNN version is too old" = cudnnTooOld; + "CUDNN version is too new" = cudnnTooNew; + }; + + src = requireFile { + name = package.filename; + inherit (package) hash; + message = '' + To use the TensorRT derivation, you must join the NVIDIA Developer Program and + download the ${package.version} TAR package for CUDA ${cudaVersion} from + ${finalAttrs.meta.homepage}. + + Once you have downloaded the file, add it to the store with the following + command, and try building this derivation again. + + $ nix-store --add-fixed sha256 ${package.filename} + ''; + }; + + # We need to look inside the extracted output to get the files we need. + sourceRoot = "TensorRT-${finalAttrs.version}"; + + buildInputs = prevAttrs.buildInputs ++ [finalAttrs.passthru.cudnn.lib]; + + preInstall = + let + targetArch = + if hostPlatform.isx86_64 then + "x86_64-linux-gnu" + else if hostPlatform.isAarch64 then + "aarch64-linux-gnu" + else + throw "Unsupported architecture"; + in + (prevAttrs.preInstall or "") + + '' + # Replace symlinks to bin and lib with the actual directories from targets. + for dir in bin lib; do + rm "$dir" + mv "targets/${targetArch}/$dir" "$dir" + done + ''; + + # Tell autoPatchelf about runtime dependencies. + postFixup = + let + versionTriple = "${versions.majorMinor finalAttrs.version}.${versions.patch finalAttrs.version}"; + in + (prevAttrs.postFixup or "") + + '' + ${meta.getExe' patchelf "patchelf"} --add-needed libnvinfer.so \ + "$lib/lib/libnvinfer.so.${versionTriple}" \ + "$lib/lib/libnvinfer_plugin.so.${versionTriple}" \ + "$lib/lib/libnvinfer_builder_resource.so.${versionTriple}" + ''; + + passthru = { + useCudatoolkitRunfile = strings.versionOlder cudaVersion "11.3.999"; + # The CUDNN used with TensorRT. + # If null, the default cudnn derivation will be used. + # If a version is specified, the cudnn derivation with that version will be used, + # unless it is not available, in which case the default cudnn derivation will be used. + cudnn = + let + desiredName = mkVersionedPackageName "cudnn" package.cudnnVersion; + desiredIsAvailable = final ? desiredName; + in + if package.cudnnVersion == null || !desiredIsAvailable then final.cudnn else final.${desiredName}; + }; + + meta = prevAttrs.meta // { + homepage = "https://developer.nvidia.com/tensorrt"; + maintainers = prevAttrs.meta.maintainers ++ [maintainers.aidalgol]; + }; +} diff --git a/pkgs/development/cuda-modules/tensorrt/releases.nix b/pkgs/development/cuda-modules/tensorrt/releases.nix index b5cae4aca4e1c..d6a1f0487dd43 100644 --- a/pkgs/development/cuda-modules/tensorrt/releases.nix +++ b/pkgs/development/cuda-modules/tensorrt/releases.nix @@ -1,4 +1,5 @@ # NOTE: Check https://developer.nvidia.com/nvidia-tensorrt-8x-download. +# Version policy is to keep the latest minor release for each major release. { tensorrt.releases = { # jetson @@ -112,7 +113,7 @@ version = "8.6.1.6"; minCudaVersion = "11.0"; maxCudaVersion = "11.8"; - cudnnVersion = null; + cudnnVersion = "8.9"; filename = "TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz"; hash = "sha256-Fb/mBT1F/uxF7McSOpEGB2sLQ/oENfJC2J3KB3gzd1k="; } @@ -120,7 +121,7 @@ version = "8.6.1.6"; minCudaVersion = "12.0"; maxCudaVersion = "12.1"; - cudnnVersion = null; + cudnnVersion = "8.9"; filename = "TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-12.0.tar.gz"; hash = "sha256-D4FXpfxTKZQ7M4uJNZE3M1CvqQyoEjnNrddYDNHrolQ="; } diff --git a/pkgs/development/cuda-modules/tensorrt/shims.nix b/pkgs/development/cuda-modules/tensorrt/shims.nix new file mode 100644 index 0000000000000..8be3e7988bb34 --- /dev/null +++ b/pkgs/development/cuda-modules/tensorrt/shims.nix @@ -0,0 +1,16 @@ +# Shims to mimic the shape of ../modules/generic/manifests/{feature,redistrib}/release.nix +{package, redistArch}: +{ + featureRelease.${redistArch}.outputs = { + bin = true; + lib = true; + static = true; + dev = true; + sample = true; + python = true; + }; + redistribRelease = { + name = "TensorRT: a high-performance deep learning interface"; + inherit (package) version; + }; +} diff --git a/pkgs/development/libraries/science/math/cudnn/extension.nix b/pkgs/development/libraries/science/math/cudnn/extension.nix deleted file mode 100644 index d4c83428980d3..0000000000000 --- a/pkgs/development/libraries/science/math/cudnn/extension.nix +++ /dev/null @@ -1,66 +0,0 @@ -# Support matrix can be found at -# https://docs.nvidia.com/deeplearning/cudnn/archives/cudnn-880/support-matrix/index.html -# Type aliases -# Release = { -# version: String, -# minCudaVersion: String, -# maxCudaVersion: String, -# url: String, -# hash: String, -# } -final: prev: let - inherit (final) callPackage; - inherit (prev) cudaVersion; - inherit (prev.lib) attrsets lists versions; - inherit (prev.lib.strings) replaceStrings versionAtLeast versionOlder; - - # Compute versioned attribute name to be used in this package set - # Patch version changes should not break the build, so we only use major and minor - # computeName :: String -> String - computeName = version: "cudnn_${replaceStrings ["."] ["_"] (versions.majorMinor version)}"; - - # Check whether a CUDNN release supports our CUDA version - # Thankfully we're able to do lexicographic comparison on the version strings - # isSupported :: Release -> Bool - isSupported = release: - versionAtLeast cudaVersion release.minCudaVersion - && versionAtLeast release.maxCudaVersion cudaVersion; - - # useCudatoolkitRunfile :: Bool - useCudatoolkitRunfile = versionOlder cudaVersion "11.3.999"; - - # buildCuDnnPackage :: Release -> Derivation - buildCuDnnPackage = callPackage ./generic.nix {inherit useCudatoolkitRunfile;}; - - # Reverse the list to have the latest release first - # cudnnReleases :: List Release - cudnnReleases = lists.reverseList (builtins.import ./releases.nix); - - # Check whether a CUDNN release supports our CUDA version - # supportedReleases :: List Release - supportedReleases = builtins.filter isSupported cudnnReleases; - - # Function to transform our releases into build attributes - # toBuildAttrs :: Release -> { name: String, value: Derivation } - toBuildAttrs = release: { - name = computeName release.version; - value = buildCuDnnPackage release; - }; - - # Add all supported builds as attributes - # allBuilds :: AttrSet String Derivation - allBuilds = builtins.listToAttrs (builtins.map toBuildAttrs supportedReleases); - - defaultBuild = attrsets.optionalAttrs (supportedReleases != []) { - cudnn = let - # The latest release is the first element of the list and will be our default choice - # latestReleaseName :: String - latestReleaseName = computeName (builtins.head supportedReleases).version; - in - allBuilds.${latestReleaseName}; - }; - - # builds :: AttrSet String Derivation - builds = allBuilds // defaultBuild; -in - builds diff --git a/pkgs/development/libraries/science/math/cudnn/generic.nix b/pkgs/development/libraries/science/math/cudnn/generic.nix deleted file mode 100644 index b9f101d80fa30..0000000000000 --- a/pkgs/development/libraries/science/math/cudnn/generic.nix +++ /dev/null @@ -1,170 +0,0 @@ -{ stdenv, - backendStdenv, - lib, - lndir, - zlib, - useCudatoolkitRunfile ? false, - cudaVersion, - cudaMajorVersion, - cudatoolkit, # For cuda < 11 - libcublas ? null, # cuda <11 doesn't ship redist packages - autoPatchelfHook, - autoAddOpenGLRunpathHook, - fetchurl, -}: { - version, - url, - hash, - minCudaVersion, - maxCudaVersion, -}: -assert useCudatoolkitRunfile || (libcublas != null); let - inherit (lib) lists strings trivial versions; - - # majorMinorPatch :: String -> String - majorMinorPatch = (trivial.flip trivial.pipe) [ - (versions.splitVersion) - (lists.take 3) - (strings.concatStringsSep ".") - ]; - - # versionTriple :: String - # Version with three components: major.minor.patch - versionTriple = majorMinorPatch version; -in - backendStdenv.mkDerivation { - pname = "cudatoolkit-${cudaMajorVersion}-cudnn"; - version = versionTriple; - strictDeps = true; - outputs = ["out" "lib" "static" "dev"]; - - src = fetchurl { - inherit url hash; - }; - - # We do need some other phases, like configurePhase, so the multiple-output setup hook works. - dontBuild = true; - - # Check and normalize Runpath against DT_NEEDED using autoPatchelf. - # Prepend /run/opengl-driver/lib using addOpenGLRunpath for dlopen("libcudacuda.so") - nativeBuildInputs = [ - autoPatchelfHook - autoAddOpenGLRunpathHook - ]; - - # Used by autoPatchelfHook - buildInputs = [ - # Note this libstdc++ isn't from the (possibly older) nvcc-compatible - # stdenv, but from the (newer) stdenv that the rest of nixpkgs uses - stdenv.cc.cc.lib - - zlib - ] ++ lists.optionals useCudatoolkitRunfile [ - cudatoolkit - ] ++ lists.optionals (!useCudatoolkitRunfile) [ - libcublas.lib - ]; - - # We used to patch Runpath here, but now we use autoPatchelfHook - # - # Note also that version <=8.3.0 contained a subdirectory "lib64/" but in - # version 8.3.2 it seems to have been renamed to simply "lib/". - # - # doc and dev have special output handling. Other outputs need to be moved to their own - # output. - # Note that moveToOutput operates on all outputs: - # https://github.com/NixOS/nixpkgs/blob/2920b6fc16a9ed5d51429e94238b28306ceda79e/pkgs/build-support/setup-hooks/multiple-outputs.sh#L105-L107 - installPhase = - '' - runHook preInstall - - mkdir -p "$out" - mv * "$out" - moveToOutput "lib64" "$lib" - moveToOutput "lib" "$lib" - moveToOutput "**/*.a" "$static" - - runHook postInstall - ''; - - # Without --add-needed autoPatchelf forgets $ORIGIN on cuda>=8.0.5. - postFixup = strings.optionalString (strings.versionAtLeast versionTriple "8.0.5") '' - patchelf $lib/lib/libcudnn.so --add-needed libcudnn_cnn_infer.so - patchelf $lib/lib/libcudnn_ops_infer.so --add-needed libcublas.so --add-needed libcublasLt.so - ''; - - # The out output leverages the same functionality which backs the `symlinkJoin` function in - # Nixpkgs: - # https://github.com/NixOS/nixpkgs/blob/d8b2a92df48f9b08d68b0132ce7adfbdbc1fbfac/pkgs/build-support/trivial-builders/default.nix#L510 - # - # That should allow us to emulate "fat" default outputs without having to actually create them. - # - # It is important that this run after the autoPatchelfHook, otherwise the symlinks in out will reference libraries in lib, creating a circular dependency. - postPhases = ["postPatchelf"]; - # For each output, create a symlink to it in the out output. - # NOTE: We must recreate the out output here, because the setup hook will have deleted it - # if it was empty. - # NOTE: Do not use optionalString based on whether `outputs` contains only `out` -- phases - # which are empty strings are skipped/unset and result in errors of the form "command not - # found: ". - postPatchelf = '' - mkdir -p "$out" - ${lib.meta.getExe lndir} "$lib" "$out" - ${lib.meta.getExe lndir} "$static" "$out" - ${lib.meta.getExe lndir} "$dev" "$out" - ''; - - passthru = { - inherit useCudatoolkitRunfile; - - cudatoolkit = - trivial.warn - '' - cudnn.cudatoolkit passthru attribute is deprecated; - if your derivation uses cudnn directly, it should probably consume cudaPackages instead - '' - cudatoolkit; - - majorVersion = versions.major versionTriple; - }; - - # Setting propagatedBuildInputs to false will prevent outputs known to the multiple-outputs - # from depending on `out` by default. - # https://github.com/NixOS/nixpkgs/blob/2920b6fc16a9ed5d51429e94238b28306ceda79e/pkgs/build-support/setup-hooks/multiple-outputs.sh#L196 - # Indeed, we want to do the opposite -- fat "out" outputs that contain all the other outputs. - propagatedBuildOutputs = false; - - # By default, if the dev output exists it just uses that. - # However, because we disabled propagatedBuildOutputs, dev doesn't contain libraries or - # anything of the sort. To remedy this, we set outputSpecified to true, and use - # outputsToInstall, which tells Nix which outputs to use when the package name is used - # unqualified (that is, without an explicit output). - outputSpecified = true; - - meta = with lib; { - # Check that the cudatoolkit version satisfies our min/max constraints (both - # inclusive). We mark the package as broken if it fails to satisfies the - # official version constraints (as recorded in default.nix). In some cases - # you _may_ be able to smudge version constraints, just know that you're - # embarking into unknown and unsupported territory when doing so. - broken = - strings.versionOlder cudaVersion minCudaVersion - || strings.versionOlder maxCudaVersion cudaVersion; - description = "NVIDIA CUDA Deep Neural Network library (cuDNN)"; - homepage = "https://developer.nvidia.com/cudnn"; - sourceProvenance = with sourceTypes; [binaryNativeCode]; - license = { - shortName = "cuDNN EULA"; - fullName = "NVIDIA cuDNN Software License Agreement (EULA)"; - url = "https://docs.nvidia.com/deeplearning/sdk/cudnn-sla/index.html#supplement"; - free = false; - } // lib.optionalAttrs (!useCudatoolkitRunfile) { - redistributable = true; - }; - platforms = ["x86_64-linux"]; - maintainers = with maintainers; [mdaiter samuela]; - # Force the use of the default, fat output by default (even though `dev` exists, which - # causes Nix to prefer that output over the others if outputSpecified isn't set). - outputsToInstall = ["out"]; - }; - } diff --git a/pkgs/development/libraries/science/math/cutensor/generic.nix b/pkgs/development/libraries/science/math/cutensor/generic.nix deleted file mode 100644 index 02fe13851620b..0000000000000 --- a/pkgs/development/libraries/science/math/cutensor/generic.nix +++ /dev/null @@ -1,88 +0,0 @@ -{ stdenv -, lib -, libPath -, cuda_cudart -, cudaMajorVersion -, cuda_nvcc -, cudatoolkit -, libcublas -, fetchurl -, autoPatchelfHook -, addOpenGLRunpath - -, version -, hash -}: - -let - mostOfVersion = builtins.concatStringsSep "." - (lib.take 3 (lib.versions.splitVersion version)); - platform = "${stdenv.hostPlatform.parsed.kernel.name}-${stdenv.hostPlatform.parsed.cpu.name}"; -in - -stdenv.mkDerivation { - pname = "cutensor-cu${cudaMajorVersion}"; - inherit version; - - src = fetchurl { - url = if lib.versionOlder mostOfVersion "1.3.3" - then "https://developer.download.nvidia.com/compute/cutensor/${mostOfVersion}/local_installers/libcutensor-${platform}-${version}.tar.gz" - else "https://developer.download.nvidia.com/compute/cutensor/redist/libcutensor/${platform}/libcutensor-${platform}-${version}-archive.tar.xz"; - inherit hash; - }; - - outputs = [ "out" "dev" ]; - - nativeBuildInputs = [ - autoPatchelfHook - addOpenGLRunpath - cuda_nvcc - ]; - - buildInputs = [ - stdenv.cc.cc.lib - cuda_cudart - libcublas - ]; - - # Set RUNPATH so that libcuda in /run/opengl-driver(-32)/lib can be found. - # See the explanation in addOpenGLRunpath. - installPhase = '' - mkdir -p "$out" "$dev" - - if [[ ! -d "${libPath}" ]] ; then - echo "Cutensor: ${libPath} does not exist, only found:" >&2 - find "$(dirname ${libPath})"/ -maxdepth 1 >&2 - echo "This cutensor release might not support your cudatoolkit version" >&2 - exit 1 - fi - - mv include "$dev" - mv ${libPath} "$out/lib" - - function finalRPathFixups { - for lib in $out/lib/lib*.so; do - addOpenGLRunpath $lib - done - } - postFixupHooks+=(finalRPathFixups) - ''; - - passthru = { - cudatoolkit = lib.warn "cutensor.passthru: cudaPackages.cudatoolkit is deprecated" cudatoolkit; - majorVersion = lib.versions.major version; - }; - - meta = with lib; { - description = "cuTENSOR: A High-Performance CUDA Library For Tensor Primitives"; - homepage = "https://developer.nvidia.com/cutensor"; - sourceProvenance = with sourceTypes; [ binaryNativeCode ]; - license = licenses.unfreeRedistributable // { - shortName = "cuTENSOR EULA"; - name = "cuTENSOR SUPPLEMENT TO SOFTWARE LICENSE AGREEMENT FOR NVIDIA SOFTWARE DEVELOPMENT KITS"; - url = "https://docs.nvidia.com/cuda/cutensor/license.html"; - }; - platforms = [ "x86_64-linux" ]; - maintainers = with maintainers; [ obsidian-systems-maintenance ]; - }; -} diff --git a/pkgs/development/libraries/science/math/nccl/default.nix b/pkgs/development/libraries/science/math/nccl/default.nix deleted file mode 100644 index d877e19a6dd68..0000000000000 --- a/pkgs/development/libraries/science/math/nccl/default.nix +++ /dev/null @@ -1,113 +0,0 @@ -# NOTE: Though NCCL is called within the cudaPackages package set, we avoid passing in -# the names of dependencies from that package set directly to avoid evaluation errors -# in the case redistributable packages are not available. -{ - lib, - fetchFromGitHub, - python3, - which, - cudaPackages, - # passthru.updateScript - gitUpdater, -}: -let - - inherit (cudaPackages) - autoAddOpenGLRunpathHook - backendStdenv - cuda_cccl - cuda_cudart - cuda_nvcc - cudaFlags - cudatoolkit - cudaVersion - ; -in -backendStdenv.mkDerivation ( - finalAttrs: { - pname = "nccl"; - version = "2.19.3-1"; - - src = fetchFromGitHub { - owner = "NVIDIA"; - repo = finalAttrs.pname; - rev = "v${finalAttrs.version}"; - hash = "sha256-59FlOKM5EB5Vkm4dZBRCkn+IgIcdQehE+FyZAdTCT/A="; - }; - - strictDeps = true; - - outputs = [ - "out" - "dev" - ]; - - nativeBuildInputs = - [ - which - autoAddOpenGLRunpathHook - python3 - ] - ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] - ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [cuda_nvcc]; - - buildInputs = - lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] - ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [ - cuda_nvcc.dev # crt/host_config.h - cuda_cudart - ] - # NOTE: CUDA versions in Nixpkgs only use a major and minor version. When we do comparisons - # against other version, like below, it's important that we use the same format. Otherwise, - # we'll get incorrect results. - # For example, lib.versionAtLeast "12.0" "12.0.0" == false. - ++ lib.optionals (lib.versionAtLeast cudaVersion "12.0") [cuda_cccl]; - - env.NIX_CFLAGS_COMPILE = toString ["-Wno-unused-function"]; - - preConfigure = '' - patchShebangs ./src/device/generate.py - makeFlagsArray+=( - "NVCC_GENCODE=${lib.concatStringsSep " " cudaFlags.gencode}" - ) - ''; - - makeFlags = - ["PREFIX=$(out)"] - ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [ - "CUDA_HOME=${cudatoolkit}" - "CUDA_LIB=${lib.getLib cudatoolkit}/lib" - "CUDA_INC=${lib.getDev cudatoolkit}/include" - ] - ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [ - "CUDA_HOME=${cuda_nvcc}" - "CUDA_LIB=${lib.getLib cuda_cudart}/lib" - "CUDA_INC=${lib.getDev cuda_cudart}/include" - ]; - - enableParallelBuilding = true; - - postFixup = '' - moveToOutput lib/libnccl_static.a $dev - ''; - - passthru.updateScript = gitUpdater { - inherit (finalAttrs) pname version; - rev-prefix = "v"; - }; - - meta = with lib; { - description = "Multi-GPU and multi-node collective communication primitives for NVIDIA GPUs"; - homepage = "https://developer.nvidia.com/nccl"; - license = licenses.bsd3; - platforms = platforms.linux; - maintainers = - with maintainers; - [ - mdaiter - orivej - ] - ++ teams.cuda.members; - }; - } -) diff --git a/pkgs/development/libraries/science/math/nccl/tests.nix b/pkgs/development/libraries/science/math/nccl/tests.nix deleted file mode 100644 index 5c2f29b7ed564..0000000000000 --- a/pkgs/development/libraries/science/math/nccl/tests.nix +++ /dev/null @@ -1,84 +0,0 @@ -# NOTE: Though NCCL tests is called within the cudaPackages package set, we avoid passing in -# the names of dependencies from that package set directly to avoid evaluation errors -# in the case redistributable packages are not available. -{ - config, - cudaPackages, - fetchFromGitHub, - gitUpdater, - lib, - mpi, - mpiSupport ? false, - which, -}: -let - inherit (cudaPackages) - backendStdenv - cuda_cccl - cuda_cudart - cuda_nvcc - cudatoolkit - cudaVersion - nccl - ; -in -backendStdenv.mkDerivation ( - finalAttrs: { - - pname = "nccl-tests"; - version = "2.13.8"; - - src = fetchFromGitHub { - owner = "NVIDIA"; - repo = finalAttrs.pname; - rev = "v${finalAttrs.version}"; - hash = "sha256-dxLoflsTHDBnZRTzoXdm30OyKpLlRa73b784YWALBHg="; - }; - - strictDeps = true; - - nativeBuildInputs = - [which] - ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] - ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [cuda_nvcc]; - - buildInputs = - [nccl] - ++ lib.optionals (lib.versionOlder cudaVersion "11.4") [cudatoolkit] - ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") [ - cuda_nvcc.dev # crt/host_config.h - cuda_cudart - ] - ++ lib.optionals (lib.versionAtLeast cudaVersion "12.0") [ - cuda_cccl.dev # - ] - ++ lib.optionals mpiSupport [mpi]; - - makeFlags = - ["NCCL_HOME=${nccl}"] - ++ lib.optionals (lib.versionOlder cudaVersion "11.4") ["CUDA_HOME=${cudatoolkit}"] - ++ lib.optionals (lib.versionAtLeast cudaVersion "11.4") ["CUDA_HOME=${cuda_nvcc}"] - ++ lib.optionals mpiSupport ["MPI=1"]; - - enableParallelBuilding = true; - - installPhase = '' - mkdir -p $out/bin - cp -r build/* $out/bin/ - ''; - - passthru.updateScript = gitUpdater { - inherit (finalAttrs) pname version; - rev-prefix = "v"; - }; - - meta = with lib; { - description = "Tests to check both the performance and the correctness of NVIDIA NCCL operations"; - homepage = "https://github.com/NVIDIA/nccl-tests"; - platforms = platforms.linux; - license = licenses.bsd3; - broken = !config.cudaSupport || (mpiSupport && mpi == null); - maintainers = with maintainers; [jmillerpdt] ++ teams.cuda.members; - }; - } -) diff --git a/pkgs/development/libraries/science/math/tensorrt/generic.nix b/pkgs/development/libraries/science/math/tensorrt/generic.nix deleted file mode 100644 index 2bcdd8e588cf0..0000000000000 --- a/pkgs/development/libraries/science/math/tensorrt/generic.nix +++ /dev/null @@ -1,95 +0,0 @@ -{ lib -, backendStdenv -, requireFile -, autoPatchelfHook -, autoAddOpenGLRunpathHook -, cudaVersion -, cudatoolkit -, cudnn -}: - -{ enable ? true -, fullVersion -, fileVersionCudnn ? null -, tarball -, sha256 -, supportedCudaVersions ? [ ] -}: - -assert !enable || fileVersionCudnn == null || lib.assertMsg (lib.strings.versionAtLeast cudnn.version fileVersionCudnn) - "This version of TensorRT requires at least cuDNN ${fileVersionCudnn} (current version is ${cudnn.version})"; - -backendStdenv.mkDerivation rec { - pname = "cudatoolkit-${cudatoolkit.majorVersion}-tensorrt"; - version = fullVersion; - src = if !enable then null else - requireFile rec { - name = tarball; - inherit sha256; - message = '' - To use the TensorRT derivation, you must join the NVIDIA Developer Program and - download the ${version} Linux x86_64 TAR package for CUDA ${cudaVersion} from - ${meta.homepage}. - - Once you have downloaded the file, add it to the store with the following - command, and try building this derivation again. - - $ nix-store --add-fixed sha256 ${name} - ''; - }; - - outputs = [ "out" "dev" ]; - - nativeBuildInputs = lib.optionals enable [ - autoPatchelfHook - autoAddOpenGLRunpathHook - ]; - - # Used by autoPatchelfHook - buildInputs = lib.optionals enable [ - backendStdenv.cc.cc.lib # libstdc++ - cudatoolkit - cudnn - ]; - - sourceRoot = "TensorRT-${version}"; - - installPhase = '' - install --directory "$dev" "$out" - mv include "$dev" - mv targets/x86_64-linux-gnu/lib "$out" - install -D --target-directory="$out/bin" targets/x86_64-linux-gnu/bin/trtexec - ''; - - # Tell autoPatchelf about runtime dependencies. - # (postFixup phase is run before autoPatchelfHook.) - postFixup = - let - mostOfVersion = builtins.concatStringsSep "." - (lib.take 3 (lib.versions.splitVersion version)); - in - '' - echo 'Patching RPATH of libnvinfer libs' - patchelf --debug --add-needed libnvinfer.so \ - "$out/lib/libnvinfer.so.${mostOfVersion}" \ - "$out/lib/libnvinfer_plugin.so.${mostOfVersion}" \ - "$out/lib/libnvinfer_builder_resource.so.${mostOfVersion}" - ''; - - passthru.stdenv = backendStdenv; - passthru.enable = enable; - - meta = with lib; { - # Check that the cudatoolkit version satisfies our min/max constraints (both - # inclusive). We mark the package as broken if it fails to satisfies the - # official version constraints (as recorded in default.nix). In some cases - # you _may_ be able to smudge version constraints, just know that you're - # embarking into unknown and unsupported territory when doing so. - broken = !enable || !(elem cudaVersion supportedCudaVersions); - description = "TensorRT: a high-performance deep learning interface"; - homepage = "https://developer.nvidia.com/tensorrt"; - license = licenses.unfree; - platforms = [ "x86_64-linux" ]; - maintainers = with maintainers; [ aidalgol ]; - }; -} diff --git a/pkgs/development/python-modules/tensorrt/default.nix b/pkgs/development/python-modules/tensorrt/default.nix index 475e7627e4baa..e108b1a773cba 100644 --- a/pkgs/development/python-modules/tensorrt/default.nix +++ b/pkgs/development/python-modules/tensorrt/default.nix @@ -11,7 +11,7 @@ let in buildPythonPackage rec { pname = "tensorrt"; - version = cudaPackages.tensorrt.version; + version = lib.optionalString (cudaPackages ? tensorrt) cudaPackages.tensorrt.version; src = cudaPackages.tensorrt.src; @@ -48,5 +48,8 @@ buildPythonPackage rec { license = licenses.unfree; platforms = [ "x86_64-linux" ]; maintainers = with maintainers; [ aidalgol ]; + broken = + !(cudaPackages ? tensorrt) + || !(cudaPackages ? cudnn); }; } diff --git a/pkgs/test/cuda/cuda-library-samples/extension.nix b/pkgs/test/cuda/cuda-library-samples/extension.nix index 4c721a9e9e1bc..62de715fd0b4f 100644 --- a/pkgs/test/cuda/cuda-library-samples/extension.nix +++ b/pkgs/test/cuda/cuda-library-samples/extension.nix @@ -1,3 +1,14 @@ -final: prev: { - cuda-library-samples = final.callPackage ./generic.nix { }; -} +{hostPlatform, lib}: +let + # Samples are built around the CUDA Toolkit, which is not available for + # aarch64. Check for both CUDA version and platform. + platformIsSupported = hostPlatform.isx86_64; + + # Build our extension + extension = + final: _: + lib.attrsets.optionalAttrs platformIsSupported { + cuda-library-samples = final.callPackage ./generic.nix {}; + }; +in +extension diff --git a/pkgs/test/cuda/cuda-library-samples/generic.nix b/pkgs/test/cuda/cuda-library-samples/generic.nix index e9a481c94a7a4..d4182536654e1 100644 --- a/pkgs/test/cuda/cuda-library-samples/generic.nix +++ b/pkgs/test/cuda/cuda-library-samples/generic.nix @@ -1,7 +1,11 @@ -{ lib, backendStdenv, fetchFromGitHub -, cmake, addOpenGLRunpath -, cudatoolkit -, cutensor +{ + lib, + backendStdenv, + fetchFromGitHub, + cmake, + addOpenGLRunpath, + cudatoolkit, + cutensor, }: let @@ -14,8 +18,11 @@ let }; commonAttrs = { version = lib.strings.substring 0 7 rev + "-" + lib.versions.majorMinor cudatoolkit.version; - nativeBuildInputs = [ cmake addOpenGLRunpath ]; - buildInputs = [ cudatoolkit ]; + nativeBuildInputs = [ + cmake + addOpenGLRunpath + ]; + buildInputs = [cudatoolkit]; postFixup = '' for exe in $out/bin/*; do addOpenGLRunpath $exe @@ -29,43 +36,50 @@ let cuSPARSE, cuSOLVER, cuFFT, cuRAND, NPP and nvJPEG. ''; license = lib.licenses.bsd3; - maintainers = with lib.maintainers; [ obsidian-systems-maintenance ]; + maintainers = with lib.maintainers; [obsidian-systems-maintenance] ++ lib.teams.cuda.members; }; }; in { - cublas = backendStdenv.mkDerivation (commonAttrs // { - pname = "cuda-library-samples-cublas"; + cublas = backendStdenv.mkDerivation ( + commonAttrs + // { + pname = "cuda-library-samples-cublas"; - src = "${src}/cuBLASLt"; - }); + src = "${src}/cuBLASLt"; + } + ); - cusolver = backendStdenv.mkDerivation (commonAttrs // { - pname = "cuda-library-samples-cusolver"; + cusolver = backendStdenv.mkDerivation ( + commonAttrs + // { + pname = "cuda-library-samples-cusolver"; - src = "${src}/cuSOLVER"; + src = "${src}/cuSOLVER"; - sourceRoot = "cuSOLVER/gesv"; - }); + sourceRoot = "cuSOLVER/gesv"; + } + ); - cutensor = backendStdenv.mkDerivation (commonAttrs // { - pname = "cuda-library-samples-cutensor"; + cutensor = backendStdenv.mkDerivation ( + commonAttrs + // { + pname = "cuda-library-samples-cutensor"; - src = "${src}/cuTENSOR"; + src = "${src}/cuTENSOR"; - buildInputs = [ cutensor ]; + buildInputs = [cutensor]; - cmakeFlags = [ - "-DCUTENSOR_EXAMPLE_BINARY_INSTALL_DIR=${builtins.placeholder "out"}/bin" - ]; + cmakeFlags = ["-DCUTENSOR_EXAMPLE_BINARY_INSTALL_DIR=${builtins.placeholder "out"}/bin"]; - # CUTENSOR_ROOT is double escaped - postPatch = '' - substituteInPlace CMakeLists.txt \ - --replace "\''${CUTENSOR_ROOT}/include" "${cutensor.dev}/include" - ''; + # CUTENSOR_ROOT is double escaped + postPatch = '' + substituteInPlace CMakeLists.txt \ + --replace "\''${CUTENSOR_ROOT}/include" "${cutensor.dev}/include" + ''; - CUTENSOR_ROOT = cutensor; - }); + CUTENSOR_ROOT = cutensor; + } + ); } diff --git a/pkgs/test/cuda/cuda-samples/extension.nix b/pkgs/test/cuda/cuda-samples/extension.nix index 664349416b713..d41da90cd5d0e 100644 --- a/pkgs/test/cuda/cuda-samples/extension.nix +++ b/pkgs/test/cuda/cuda-samples/extension.nix @@ -1,14 +1,18 @@ -final: prev: let - - sha256 = { - "10.0" = "1zvh4xsdyc59m87brpcmssxsjlp9dkynh4asnkcmc3g94f53l0jw"; - "10.1" = "1s8ka0hznrni36ajhzf2gqpdrl8kd8fi047qijxks5l2abc093qd"; - "10.2" = "01p1innzgh9siacpld6nsqimj8jkg93rk4gj8q4crn62pa5vhd94"; - "11.0" = "1n3vjc8c7zdig2xgl5fppavrphqzhdiv9m9nk6smh4f99fwi0705"; - "11.1" = "1kjixk50i8y1bkiwbdn5lkv342crvkmbvy1xl5j3lsa1ica21kwh"; - "11.2" = "1p1qjvfbm28l933mmnln02rqrf0cy9kbpsyb488d1haiqzvrazl1"; - "11.3" = "0kbibb6pgz8j5iq6284axcnmycaha9bw8qlmdp6yfwmnahq1v0yz"; - "11.4" = "082dkk5y34wyvjgj2p5j1d00rk8xaxb9z0mhvz16bd469r1bw2qk"; +{ + cudaVersion, + hostPlatform, + lib, +}: +let + cudaVersionToHash = { + "10.0" = "sha256-XAI6iiPpDVbZtFoRaP1s6VKpu9aV3bwOqqkw33QncP8="; + "10.1" = "sha256-DY8E2FKCFj27jPgQEB1qE9HcLn7CfSiVGdFm+yFQE+k="; + "10.2" = "sha256-JDW4i7rC2MwIRvKRmUd6UyJZI9bWNHqZijrB962N4QY="; + "11.0" = "sha256-BRwQuUvJEVi1mTbVtGODH8Obt7rXFfq6eLH9wxCTe9g="; + "11.1" = "sha256-kM8gFItBaTpkoT34vercmQky9qTFtsXjXMGjCMrsUc4="; + "11.2" = "sha256-gX6V98dRwdAQIsvru2byDLiMswCW2lrHSBSJutyWONw="; + "11.3" = "sha256-34MdMFS2cufNbZVixFdSUDFfLeuKIGFwLBL9d81acU0="; + "11.4" = "sha256-Ewu+Qk6GtGXC37CCn1ZXHc0MQAuyXCGf3J6T4cucTSA="; "11.5" = "sha256-AKRZbke0K59lakhTi8dX2cR2aBuWPZkiQxyKaZTvHrI="; "11.6" = "sha256-AsLNmAplfuQbXg9zt09tXAuFJ524EtTYsQuUlV1tPkE="; # The tag 11.7 of cuda-samples does not exist @@ -16,10 +20,23 @@ final: prev: let "12.0" = "sha256-Lj2kbdVFrJo5xPYPMiE4BS7Z8gpU5JLKXVJhZABUe/g="; "12.1" = "sha256-xE0luOMq46zVsIEWwK4xjLs7NorcTIi9gbfZPVjIlqo="; "12.2" = "sha256-pOy0qfDjA/Nr0T9PNKKefK/63gQnJV2MQsN2g3S2yng="; + "12.3" = "sha256-fjVp0G6uRCWxsfe+gOwWTN+esZfk0O5uxS623u0REAk="; }; -in prev.lib.attrsets.optionalAttrs (builtins.hasAttr prev.cudaVersion sha256) { - cuda-samples = final.callPackage ./generic.nix { - sha256 = sha256.${prev.cudaVersion}; - }; -} + # Samples are built around the CUDA Toolkit, which is not available for + # aarch64. Check for both CUDA version and platform. + cudaVersionIsSupported = cudaVersionToHash ? ${cudaVersion}; + platformIsSupported = hostPlatform.isx86_64; + isSupported = cudaVersionIsSupported && platformIsSupported; + + # Build our extension + extension = + final: _: + lib.attrsets.optionalAttrs isSupported { + cuda-samples = final.callPackage ./generic.nix { + inherit cudaVersion; + hash = cudaVersionToHash.${cudaVersion}; + }; + }; +in +extension diff --git a/pkgs/test/cuda/cuda-samples/generic.nix b/pkgs/test/cuda/cuda-samples/generic.nix index e690f32959f2a..fb3d7cc99da95 100644 --- a/pkgs/test/cuda/cuda-samples/generic.nix +++ b/pkgs/test/cuda/cuda-samples/generic.nix @@ -1,70 +1,79 @@ -{ autoAddOpenGLRunpathHook -, backendStdenv -, cmake -, cudatoolkit -, cudaVersion -, fetchFromGitHub -, fetchpatch -, freeimage -, glfw3 -, lib -, pkg-config -, sha256 +{ + autoAddOpenGLRunpathHook, + backendStdenv, + cmake, + cudatoolkit, + cudaVersion, + fetchFromGitHub, + fetchpatch, + freeimage, + glfw3, + hash, + lib, + pkg-config, }: -backendStdenv.mkDerivation (finalAttrs: { - pname = "cuda-samples"; - version = cudaVersion; +let + inherit (lib) lists strings; +in +backendStdenv.mkDerivation ( + finalAttrs: { + strictDeps = true; - src = fetchFromGitHub { - owner = "NVIDIA"; - repo = finalAttrs.pname; - rev = "v${finalAttrs.version}"; - inherit sha256; - }; + pname = "cuda-samples"; + version = cudaVersion; - nativeBuildInputs = [ - pkg-config - autoAddOpenGLRunpathHook - glfw3 - freeimage - ] - # CMake has to run as a native, build-time dependency for libNVVM samples. - ++ lib.lists.optionals (lib.strings.versionAtLeast finalAttrs.version "12.2") [ - cmake - ]; + src = fetchFromGitHub { + owner = "NVIDIA"; + repo = finalAttrs.pname; + rev = "v${finalAttrs.version}"; + inherit hash; + }; - # CMake is not the primary build tool -- that's still make. - # As such, we disable CMake's build system. - dontUseCmakeConfigure = true; + nativeBuildInputs = + [ + autoAddOpenGLRunpathHook + pkg-config + ] + # CMake has to run as a native, build-time dependency for libNVVM samples. + # However, it's not the primary build tool -- that's still make. + # As such, we disable CMake's build system. + ++ lists.optionals (strings.versionAtLeast finalAttrs.version "12.2") [cmake]; - buildInputs = [ cudatoolkit ]; + dontUseCmakeConfigure = true; - # See https://github.com/NVIDIA/cuda-samples/issues/75. - patches = lib.optionals (finalAttrs.version == "11.3") [ - (fetchpatch { - url = "https://github.com/NVIDIA/cuda-samples/commit/5c3ec60faeb7a3c4ad9372c99114d7bb922fda8d.patch"; - sha256 = "sha256-0XxdmNK9MPpHwv8+qECJTvXGlFxc+fIbta4ynYprfpU="; - }) - ]; + buildInputs = [ + cudatoolkit + freeimage + glfw3 + ]; - enableParallelBuilding = true; + # See https://github.com/NVIDIA/cuda-samples/issues/75. + patches = lib.optionals (finalAttrs.version == "11.3") [ + (fetchpatch { + url = "https://github.com/NVIDIA/cuda-samples/commit/5c3ec60faeb7a3c4ad9372c99114d7bb922fda8d.patch"; + hash = "sha256-0XxdmNK9MPpHwv8+qECJTvXGlFxc+fIbta4ynYprfpU="; + }) + ]; - preConfigure = '' - export CUDA_PATH=${cudatoolkit} - ''; + enableParallelBuilding = true; - installPhase = '' - runHook preInstall + preConfigure = '' + export CUDA_PATH=${cudatoolkit} + ''; - install -Dm755 -t $out/bin bin/${backendStdenv.hostPlatform.parsed.cpu.name}/${backendStdenv.hostPlatform.parsed.kernel.name}/release/* + installPhase = '' + runHook preInstall - runHook postInstall - ''; + install -Dm755 -t $out/bin bin/${backendStdenv.hostPlatform.parsed.cpu.name}/${backendStdenv.hostPlatform.parsed.kernel.name}/release/* - meta = { - description = "Samples for CUDA Developers which demonstrates features in CUDA Toolkit"; - # CUDA itself is proprietary, but these sample apps are not. - license = lib.licenses.bsd3; - maintainers = with lib.maintainers; [ obsidian-systems-maintenance ] ++ lib.teams.cuda.members; - }; -}) + runHook postInstall + ''; + + meta = { + description = "Samples for CUDA Developers which demonstrates features in CUDA Toolkit"; + # CUDA itself is proprietary, but these sample apps are not. + license = lib.licenses.bsd3; + maintainers = with lib.maintainers; [obsidian-systems-maintenance] ++ lib.teams.cuda.members; + }; + } +) diff --git a/pkgs/test/cuda/default.nix b/pkgs/test/cuda/default.nix index c7b790e35e259..be88bd3820a90 100644 --- a/pkgs/test/cuda/default.nix +++ b/pkgs/test/cuda/default.nix @@ -1,7 +1,7 @@ -{ callPackage }: +{callPackage}: rec { - cuda-samplesPackages = callPackage ./cuda-samples/generic.nix { }; + cuda-samplesPackages = callPackage ./cuda-samples/generic.nix {}; inherit (cuda-samplesPackages) cuda-samples_cudatoolkit_10 cuda-samples_cudatoolkit_10_0 @@ -12,9 +12,10 @@ rec { cuda-samples_cudatoolkit_11_1 cuda-samples_cudatoolkit_11_2 cuda-samples_cudatoolkit_11_3 - cuda-samples_cudatoolkit_11_4; + cuda-samples_cudatoolkit_11_4 + ; - cuda-library-samplesPackages = callPackage ./cuda-library-samples/generic.nix { }; + cuda-library-samplesPackages = callPackage ./cuda-library-samples/generic.nix {}; inherit (cuda-library-samplesPackages) cuda-library-samples_cudatoolkit_10 cuda-library-samples_cudatoolkit_10_1 @@ -24,5 +25,6 @@ rec { cuda-library-samples_cudatoolkit_11_1 cuda-library-samples_cudatoolkit_11_2 cuda-library-samples_cudatoolkit_11_3 - cuda-library-samples_cudatoolkit_11_4; + cuda-library-samples_cudatoolkit_11_4 + ; } diff --git a/pkgs/top-level/all-packages.nix b/pkgs/top-level/all-packages.nix index ecc9681d53daa..fc0332ab8ca31 100644 --- a/pkgs/top-level/all-packages.nix +++ b/pkgs/top-level/all-packages.nix @@ -7304,7 +7304,7 @@ with pkgs; cudaPackages_10_0 = callPackage ./cuda-packages.nix { cudaVersion = "10.0"; }; cudaPackages_10_1 = callPackage ./cuda-packages.nix { cudaVersion = "10.1"; }; cudaPackages_10_2 = callPackage ./cuda-packages.nix { cudaVersion = "10.2"; }; - cudaPackages_10 = cudaPackages_10_2; + cudaPackages_10 = recurseIntoAttrs cudaPackages_10_2; cudaPackages_11_0 = callPackage ./cuda-packages.nix { cudaVersion = "11.0"; }; cudaPackages_11_1 = callPackage ./cuda-packages.nix { cudaVersion = "11.1"; }; @@ -7315,12 +7315,13 @@ with pkgs; cudaPackages_11_6 = callPackage ./cuda-packages.nix { cudaVersion = "11.6"; }; cudaPackages_11_7 = callPackage ./cuda-packages.nix { cudaVersion = "11.7"; }; cudaPackages_11_8 = callPackage ./cuda-packages.nix { cudaVersion = "11.8"; }; - cudaPackages_11 = cudaPackages_11_8; + cudaPackages_11 = recurseIntoAttrs cudaPackages_11_8; cudaPackages_12_0 = callPackage ./cuda-packages.nix { cudaVersion = "12.0"; }; cudaPackages_12_1 = callPackage ./cuda-packages.nix { cudaVersion = "12.1"; }; cudaPackages_12_2 = callPackage ./cuda-packages.nix { cudaVersion = "12.2"; }; - cudaPackages_12 = cudaPackages_12_0; + cudaPackages_12_3 = callPackage ./cuda-packages.nix { cudaVersion = "12.3"; }; + cudaPackages_12 = recurseIntoAttrs cudaPackages_12_0; # Use the older cudaPackages for tensorflow and jax, as determined by cudnn # compatibility: https://www.tensorflow.org/install/source#gpu @@ -7328,7 +7329,7 @@ with pkgs; # TODO: try upgrading once there is a cuDNN release supporting CUDA 12. No # such cuDNN release as of 2023-01-10. - cudaPackages = recurseIntoAttrs cudaPackages_11; + cudaPackages = cudaPackages_11; # TODO: move to alias cudatoolkit = cudaPackages.cudatoolkit; diff --git a/pkgs/top-level/cuda-packages.nix b/pkgs/top-level/cuda-packages.nix index 3912422785bc4..5d4d2fcf29247 100644 --- a/pkgs/top-level/cuda-packages.nix +++ b/pkgs/top-level/cuda-packages.nix @@ -1,88 +1,118 @@ -{ lib -, pkgs -, cudaVersion +# Notes: +# +# Silvan (Tweag) covered some things on recursive attribute sets in the Nix Hour: +# https://www.youtube.com/watch?v=BgnUFtd1Ivs +# +# I (@connorbaker) highly recommend watching it. +# +# Most helpful comment regarding recursive attribute sets: +# +# https://github.com/NixOS/nixpkgs/pull/256324#issuecomment-1749935979 +# +# To summarize: +# +# - `prev` should only be used to access attributes which are going to be overriden. +# - `final` should only be used to access `callPackage` to build new packages. +# - Attribute names should be computable without relying on `final`. +# - Extensions should take arguments to build attribute names before relying on `final`. +# +# Silvan's recommendation then is to explicitly use `callPackage` to provide everything our extensions need +# to compute the attribute names, without relying on `final`. +# +# I've (@connorbaker) attempted to do that, though I'm unsure of how this will interact with overrides. +{ + callPackage, + cudaVersion, + lib, + newScope, + pkgs, }: - -with lib; - let - - scope = makeScope pkgs.newScope (final: { - # Here we put package set configuration and utility functions. - inherit cudaVersion; - cudaMajorVersion = versions.major final.cudaVersion; - cudaMajorMinorVersion = lib.versions.majorMinor final.cudaVersion; - inherit lib pkgs; - - addBuildInputs = drv: buildInputs: drv.overrideAttrs (oldAttrs: { - buildInputs = (oldAttrs.buildInputs or []) ++ buildInputs; - }); - }); - - cutensorExtension = final: prev: let - ### CuTensor - - buildCuTensorPackage = final.callPackage ../development/libraries/science/math/cutensor/generic.nix; - - # FIXME: Include non-x86_64 platforms - cuTensorVersions = { - "1.2.2.5" = { - hash = "sha256-lU7iK4DWuC/U3s1Ct/rq2Gr3w4F2U7RYYgpmF05bibY="; - }; - "1.5.0.3" = { - hash = "sha256-T96+lPC6OTOkIs/z3QWg73oYVSyidN0SVkBWmT9VRx0="; - }; - "2.0.0.7" = { - hash = "sha256-32M4rtGOW2rgxJUhBT0WBtKkHhh9f17M+RgK9rvE72g="; - }; - }; - - inherit (final) cudaMajorMinorVersion cudaMajorVersion; - - cudaToCutensor = { - "10" = "1.2.25"; - "11" = "1.5.0.3"; - "12" = "2.0.0.7"; - }; - - versionNewer = lib.flip lib.versionOlder; - latestVersion = (builtins.head (lib.sort versionNewer (builtins.attrNames cuTensorVersions))); - - cutensor = buildCuTensorPackage rec { - version = cudaToCutensor.${cudaMajorVersion} or latestVersion; - inherit (cuTensorVersions.${version}) hash; - # This can go into generic.nix - libPath = "lib/${if cudaMajorVersion == "10" then cudaMajorMinorVersion else cudaMajorVersion}"; - }; - in { inherit cutensor; }; - - extraPackagesExtension = final: prev: { - - nccl = final.callPackage ../development/libraries/science/math/nccl { }; - - nccl-tests = final.callPackage ../development/libraries/science/math/nccl/tests.nix { }; - - autoAddOpenGLRunpathHook = final.callPackage ( { makeSetupHook, addOpenGLRunpath }: - makeSetupHook { - name = "auto-add-opengl-runpath-hook"; - propagatedBuildInputs = [ - addOpenGLRunpath - ]; - } ../development/compilers/cudatoolkit/auto-add-opengl-runpath-hook.sh - ) {}; - - }; - - composedExtension = composeManyExtensions ([ - extraPackagesExtension - (import ../development/compilers/cudatoolkit/extension.nix) - (import ../development/compilers/cudatoolkit/redist/extension.nix) - (import ../development/compilers/cudatoolkit/redist/overrides.nix) - (import ../development/libraries/science/math/cudnn/extension.nix) - (import ../development/libraries/science/math/tensorrt/extension.nix) - (import ../test/cuda/cuda-samples/extension.nix) - (import ../test/cuda/cuda-library-samples/extension.nix) - cutensorExtension - ]); - -in (scope.overrideScope composedExtension) + inherit (lib) + attrsets + customisation + fixedPoints + strings + versions + ; + # Backbone + gpus = builtins.import ../development/cuda-modules/gpus.nix; + nvccCompatibilities = builtins.import ../development/cuda-modules/nvcc-compatibilities.nix; + flags = callPackage ../development/cuda-modules/flags.nix {inherit cudaVersion gpus;}; + passthruFunction = + final: + ( + { + inherit cudaVersion lib pkgs; + inherit gpus nvccCompatibilities flags; + cudaMajorVersion = versions.major cudaVersion; + cudaMajorMinorVersion = versions.majorMinor cudaVersion; + + # Maintain a reference to the final cudaPackages. + # Without this, if we use `final.callPackage` and a package accepts `cudaPackages` as an argument, + # it's provided with `cudaPackages` from the top-level scope, which is not what we want. We want to + # provide the `cudaPackages` from the final scope -- that is, the *current* scope. + cudaPackages = final; + + # TODO(@connorbaker): `cudaFlags` is an alias for `flags` which should be removed in the future. + cudaFlags = flags; + + # Exposed as cudaPackages.backendStdenv. + # This is what nvcc uses as a backend, + # and it has to be an officially supported one (e.g. gcc11 for cuda11). + # + # It, however, propagates current stdenv's libstdc++ to avoid "GLIBCXX_* not found errors" + # when linked with other C++ libraries. + # E.g. for cudaPackages_11_8 we use gcc11 with gcc12's libstdc++ + # Cf. https://github.com/NixOS/nixpkgs/pull/218265 for context + backendStdenv = final.callPackage ../development/cuda-modules/backend-stdenv.nix {}; + + # Loose packages + cudatoolkit = final.callPackage ../development/cuda-modules/cudatoolkit {}; + saxpy = final.callPackage ../development/cuda-modules/saxpy {}; + } + # NCCL is not supported on Jetson, because it does not use NVLink or PCI-e for inter-GPU communication. + # https://forums.developer.nvidia.com/t/can-jetson-orin-support-nccl/232845/9 + // attrsets.optionalAttrs (!flags.isJetsonBuild) { + nccl = final.callPackage ../development/cuda-modules/nccl {}; + nccl-tests = final.callPackage ../development/cuda-modules/nccl-tests {}; + } + ); + + mkVersionedPackageName = + name: version: + strings.concatStringsSep "_" [ + name + (strings.replaceStrings ["."] ["_"] (versions.majorMinor version)) + ]; + + composedExtension = fixedPoints.composeManyExtensions [ + (import ../development/cuda-modules/setup-hooks/extension.nix) + (callPackage ../development/cuda-modules/cuda/extension.nix {inherit cudaVersion;}) + (callPackage ../development/cuda-modules/cuda/overrides.nix {inherit cudaVersion;}) + (callPackage ../development/cuda-modules/generic-builders/multiplex.nix { + inherit cudaVersion flags mkVersionedPackageName; + pname = "cudnn"; + releasesModule = ../development/cuda-modules/cudnn/releases.nix; + shimsFn = ../development/cuda-modules/cudnn/shims.nix; + fixupFn = ../development/cuda-modules/cudnn/fixup.nix; + }) + (callPackage ../development/cuda-modules/cutensor/extension.nix { + inherit cudaVersion flags mkVersionedPackageName; + }) + (callPackage ../development/cuda-modules/generic-builders/multiplex.nix { + inherit cudaVersion flags mkVersionedPackageName; + pname = "tensorrt"; + releasesModule = ../development/cuda-modules/tensorrt/releases.nix; + shimsFn = ../development/cuda-modules/tensorrt/shims.nix; + fixupFn = ../development/cuda-modules/tensorrt/fixup.nix; + }) + (callPackage ../test/cuda/cuda-samples/extension.nix {inherit cudaVersion;}) + (callPackage ../test/cuda/cuda-library-samples/extension.nix {}) + ]; + + cudaPackages = customisation.makeScope newScope ( + fixedPoints.extends composedExtension passthruFunction + ); +in +cudaPackages -- cgit 1.4.1