Merge pull request #287914 from GaetanLepage/gpuctypes

python3Packages.tinygrad: init at 0.8.0

pkgs/development/python-modules/gpuctypes/0001-fix-dlopen-cuda.patch

@@ -0,0 +1,44 @@
+From d448321436e8314d3e2a6a09d4017c4bc10f612d Mon Sep 17 00:00:00 2001
+From: Gaetan Lepage <gaetan@glepage.com>
+Date: Sat, 17 Feb 2024 17:37:22 +0100
+Subject: [PATCH] fix-dlopen-cuda
+
+---
+ gpuctypes/cuda.py | 20 ++++++++++++++++++--
+ 1 file changed, 18 insertions(+), 2 deletions(-)
+
+diff --git a/gpuctypes/cuda.py b/gpuctypes/cuda.py
+index acba81c..091f7f7 100644
+--- a/gpuctypes/cuda.py
++++ b/gpuctypes/cuda.py
+@@ -143,9 +143,25 @@ def char_pointer_cast(string, encoding='utf-8'):
+ 
+ 
+ 
++NAME_TO_PATHS = {
++    "libcuda.so": ["@driverLink@/lib/libcuda.so"],
++    "libnvrtc.so": ["@libnvrtc@"],
++}
++def _try_dlopen(name):
++    try:
++        return ctypes.CDLL(name)
++    except OSError:
++        pass
++    for candidate in NAME_TO_PATHS.get(name, []):
++        try:
++            return ctypes.CDLL(candidate)
++        except OSError:
++	        pass
++    raise RuntimeError(f"{name} not found")
++
+ _libraries = {}
+-_libraries['libcuda.so'] = ctypes.CDLL(ctypes.util.find_library('cuda'))
+-_libraries['libnvrtc.so'] = ctypes.CDLL(ctypes.util.find_library('nvrtc'))
++_libraries['libcuda.so'] = _try_dlopen('libcuda.so')
++_libraries['libnvrtc.so'] = _try_dlopen('libnvrtc.so')
+ 
+ 
+ cuuint32_t = ctypes.c_uint32
+-- 
+2.43.0
+

pkgs/development/python-modules/gpuctypes/default.nix

@@ -0,0 +1,127 @@
+{ lib
+, config
+, buildPythonPackage
+, fetchFromGitHub
+, substituteAll
+, addDriverRunpath
+, cudaSupport ? config.cudaSupport
+, rocmSupport ? config.rocmSupport
+, cudaPackages
+, setuptools
+, ocl-icd
+, rocmPackages
+, pytestCheckHook
+, gpuctypes
+, testCudaRuntime ? false
+, testOpenclRuntime ? false
+, testRocmRuntime ? false
+}:
+assert testCudaRuntime -> cudaSupport;
+assert testRocmRuntime -> rocmSupport;
+
+buildPythonPackage rec {
+  pname = "gpuctypes";
+  version = "0.3.0";
+  pyproject = true;
+
+  src = fetchFromGitHub {
+    repo = "gpuctypes";
+    owner = "tinygrad";
+    rev = "refs/tags/${version}";
+    hash = "sha256-xUMvMBK1UhZaMZfik0Ia6+siyZGpCkBV+LTnQvzt/rw=";
+  };
+
+  patches = [
+    (substituteAll {
+      src = ./0001-fix-dlopen-cuda.patch;
+      inherit (addDriverRunpath) driverLink;
+      libnvrtc =
+        if cudaSupport
+        then "${lib.getLib cudaPackages.cuda_nvrtc}/lib/libnvrtc.so"
+        else "Please import nixpkgs with `config.cudaSupport = true`";
+    })
+  ];
+
+  nativeBuildInputs = [
+    setuptools
+  ];
+
+  postPatch = ''
+    substituteInPlace gpuctypes/opencl.py \
+      --replace "ctypes.util.find_library('OpenCL')" "'${ocl-icd}/lib/libOpenCL.so'"
+  ''
+  # hipGetDevicePropertiesR0600 is a symbol from rocm-6. We are currently at rocm-5.
+  # We are not sure that this works. Remove when rocm gets updated to version 6.
+  + lib.optionalString rocmSupport ''
+    substituteInPlace gpuctypes/hip.py \
+      --replace "/opt/rocm/lib/libamdhip64.so" "${rocmPackages.clr}/lib/libamdhip64.so" \
+      --replace "/opt/rocm/lib/libhiprtc.so" "${rocmPackages.clr}/lib/libhiprtc.so" \
+      --replace "hipGetDevicePropertiesR0600" "hipGetDeviceProperties"
+
+    substituteInPlace gpuctypes/comgr.py \
+      --replace "/opt/rocm/lib/libamd_comgr.so" "${rocmPackages.rocm-comgr}/lib/libamd_comgr.so"
+  '';
+
+  pythonImportsCheck = [ "gpuctypes" ];
+
+  nativeCheckInputs = [
+    pytestCheckHook
+  ];
+
+  disabledTestPaths = lib.optionals (!testOpenclRuntime) [
+    "test/test_opencl.py"
+  ] ++ lib.optionals (!rocmSupport) [
+    "test/test_hip.py"
+  ] ++ lib.optionals (!cudaSupport) [
+    "test/test_cuda.py"
+  ];
+
+  # Require GPU access to run (not available in the sandbox)
+  pytestFlagsArray = lib.optionals (!testCudaRuntime) [
+    "-k" "'not TestCUDADevice'"
+  ] ++ lib.optionals (!testRocmRuntime) [
+    "-k" "'not TestHIPDevice'"
+  ] ++ lib.optionals (testCudaRuntime || testOpenclRuntime || testRocmRuntime) [
+    "-v"
+  ];
+
+  # Running these tests requires special configuration on the builder.
+  # e.g. https://github.com/NixOS/nixpkgs/pull/256230 implements a nix
+  # pre-build hook which exposes the devices and the drivers in the sandbox
+  # based on requiredSystemFeatures:
+  requiredSystemFeatures = lib.optionals testCudaRuntime [
+    "cuda"
+  ] ++ lib.optionals testOpenclRuntime [
+    "opencl"
+  ] ++ lib.optionals testRocmRuntime [
+    "rocm"
+  ];
+
+  passthru.gpuChecks = {
+    cuda = gpuctypes.override {
+      cudaSupport = true;
+      testCudaRuntime = true;
+    };
+    opencl = gpuctypes.override {
+      testOpenclRuntime = true;
+    };
+    rocm = gpuctypes.override {
+      rocmSupport = true;
+      testRocmRuntime = true;
+    };
+  };
+
+  preCheck = lib.optionalString (cudaSupport && !testCudaRuntime) ''
+    addToSearchPath LD_LIBRARY_PATH ${lib.getLib cudaPackages.cuda_cudart}/lib/stubs
+  '';
+
+  # If neither rocmSupport or cudaSupport is enabled, no tests are selected
+  dontUsePytestCheck = !(rocmSupport || cudaSupport) && (!testOpenclRuntime);
+
+  meta = with lib; {
+    description = "Ctypes wrappers for HIP, CUDA, and OpenCL";
+    homepage = "https://github.com/tinygrad/gpuctypes";
+    license = licenses.mit;
+    maintainers = with maintainers; [ GaetanLepage matthewcroughan wozeparrot ];
+  };
+}

pkgs/development/python-modules/gpuctypes/fix-dlopen-cuda.patch

@@ -0,0 +1,32 @@
+diff --git a/gpuctypes/cuda.py b/gpuctypes/cuda.py
+index acba81c..aac5fc7 100644
+--- a/gpuctypes/cuda.py
++++ b/gpuctypes/cuda.py
+@@ -143,9 +143,25 @@ def char_pointer_cast(string, encoding='utf-8'):
+ 
+ 
+ 
++NAME_TO_PATHS = {
++    "libcuda.so": ["@driverLink@/lib/libcuda.so"],
++    "libnvrtc.so": ["@libnvrtc@"],
++}
++def _try_dlopen(name):
++    try:
++        return ctypes.CDLL(name)
++    except OSError:
++        pass
++    for candidate in NAME_TO_PATHS.get(name, []):
++        try:
++            return ctypes.CDLL(candidate)
++        except OSError:
++	     pass
++    raise RuntimeError(f"{name} not found")
++
+ _libraries = {}
+-_libraries['libcuda.so'] = ctypes.CDLL(ctypes.util.find_library('cuda'))
+-_libraries['libnvrtc.so'] = ctypes.CDLL(ctypes.util.find_library('nvrtc'))
++_libraries['libcuda.so'] = _try_dlopen('libcuda.so')
++_libraries['libnvrtc.so'] = _try_dlopen('libnvrtc.so')
+ 
+ 
+ cuuint32_t = ctypes.c_uint32

pkgs/development/python-modules/tinygrad/default.nix

@@ -0,0 +1,111 @@
+{ lib
+, buildPythonPackage
+, fetchFromGitHub
+, setuptools
+, wheel
+, gpuctypes
+, numpy
+, tqdm
+, hypothesis
+, librosa
+, onnx
+, pillow
+, pytest-xdist
+, pytestCheckHook
+, safetensors
+, sentencepiece
+, tiktoken
+, torch
+, transformers
+}:
+
+buildPythonPackage rec {
+  pname = "tinygrad";
+  version = "0.8.0";
+  pyproject = true;
+
+  src = fetchFromGitHub {
+    owner = "tinygrad";
+    repo = "tinygrad";
+    rev = "refs/tags/v${version}";
+    hash = "sha256-QAccZ79qUbe27yUykIf22WdkxYUlOffnMlShakKfp60=";
+  };
+
+  nativeBuildInputs = [
+    setuptools
+    wheel
+  ];
+
+  propagatedBuildInputs = [
+    gpuctypes
+    numpy
+    tqdm
+  ];
+
+  pythonImportsCheck = [ "tinygrad" ];
+
+  nativeCheckInputs = [
+    hypothesis
+    librosa
+    onnx
+    pillow
+    pytest-xdist
+    pytestCheckHook
+    safetensors
+    sentencepiece
+    tiktoken
+    torch
+    transformers
+  ];
+
+  preCheck = ''
+    export HOME=$(mktemp -d)
+  '';
+
+  disabledTests = [
+    # Require internet access
+    "test_benchmark_openpilot_model"
+    "test_bn_alone"
+    "test_bn_linear"
+    "test_bn_mnist"
+    "test_car"
+    "test_chicken"
+    "test_chicken_bigbatch"
+    "test_conv_mnist"
+    "testCopySHMtoDefault"
+    "test_data_parallel_resnet"
+    "test_e2e_big"
+    "test_fetch_small"
+    "test_huggingface_enet_safetensors"
+    "test_linear_mnist"
+    "test_load_convnext"
+    "test_load_enet"
+    "test_load_enet_alt"
+    "test_load_llama2bfloat"
+    "test_load_resnet"
+    "test_openpilot_model"
+    "test_resnet"
+    "test_shufflenet"
+    "test_transcribe_batch12"
+    "test_transcribe_batch21"
+    "test_transcribe_file1"
+    "test_transcribe_file2"
+    "test_transcribe_long"
+    "test_transcribe_long_no_batch"
+    "test_vgg7"
+  ];
+
+  disabledTestPaths = [
+    "test/extra/test_lr_scheduler.py"
+    "test/models/test_mnist.py"
+    "test/models/test_real_world.py"
+  ];
+
+  meta = with lib; {
+    description = "A simple and powerful neural network framework";
+    homepage = "https://github.com/tinygrad/tinygrad";
+    changelog = "https://github.com/tinygrad/tinygrad/releases/tag/v${version}";
+    license = licenses.mit;
+    maintainers = with maintainers; [ GaetanLepage ];
+  };
+}

pkgs/top-level/python-packages.nix

@@ -4939,6 +4939,8 @@ self: super: with self; {
 
   gpsoauth = callPackage ../development/python-modules/gpsoauth { };
 
+  gpuctypes = callPackage ../development/python-modules/gpuctypes { };
+
   gpustat = callPackage ../development/python-modules/gpustat { };
 
   gpxpy = callPackage ../development/python-modules/gpxpy { };
@@ -14763,6 +14765,8 @@ self: super: with self; {
 
   tinydb = callPackage ../development/python-modules/tinydb { };
 
+  tinygrad = callPackage ../development/python-modules/tinygrad { };
+
   tinyobjloader-py = callPackage ../development/python-modules/tinyobjloader-py { };
 
   tinyrecord = callPackage ../development/python-modules/tinyrecord { };