{devel}[foss/2024a] Triton v3.1.0 w/ CUDA 12.6.0

easybuild/easyconfigs/c/Clang/Clang-18.1.6_10dc3a8-GCCcore-13.3.0-CUDA-12.6.0.eb

@@ -0,0 +1,71 @@
+##
+# This file is an EasyBuild reciPY as per https://github.com/easybuilders/easybuild
+#
+# Copyright:: Copyright 2013-2015 Dmitri Gribenko, Ward Poelmans
+# Authors:: Dmitri Gribenko <[email protected]>
+# Authors:: Ward Poelmans <[email protected]>
+# License:: GPLv2 or later, MIT, three-clause BSD.
+# $Id$
+##
+
+# thembl: according to cmake/llvm-hash.txt Triton 3.1.0 (5fe38ffd73c2ac6ed6323b554205186696631c6f) requires 
+# this commit 10dc3a8e916d73291269e5e2b82dd22681489aa1 of LLVM
+
+name = 'Clang'
+# local_date = '20240523'
+local_commit = '10dc3a8e916d73291269e5e2b82dd22681489aa1'
+version = '18.1.6_%s' % local_commit[:7]
+versionsuffix = '-CUDA-%(cudaver)s'
+
+homepage = 'https://clang.llvm.org/'
+description = """C, C++, Objective-C compiler, based on LLVM.  Does not
+ include C++ standard library -- use libstdc++ from GCC."""
+
+# Clang also depends on libstdc++ during runtime, but this dependency is
+# already specified as the toolchain.
+toolchain = {'name': 'GCCcore', 'version': '13.3.0'}
+
+source_urls = ['https://github.com/llvm/llvm-project/archive/']
+sources = [{
+    'download_filename': '%s.tar.gz' % local_commit,
+    'filename': 'llvm-project-%s.tar.gz' % version,
+}]
+checksums = ['6ee5e0f9a49d41b5f48ebc4613ce3371f686bf70fcece9f849aba3c37bdeb3e8']
+
+builddependencies = [
+    ('CMake', '3.29.3'),
+    ('Perl', '5.38.2'),
+    # Including Python bindings would require this as a runtime dep
+    # and SWIG as an additional build dep
+    ('Python', '3.12.3'),
+]
+dependencies = [
+    # since Clang is a compiler, binutils is a runtime dependency too
+    ('binutils', '2.42'),
+    ('hwloc', '2.10.0'),
+    ('libxml2', '2.12.7'),
+    ('ncurses', '6.5'),
+    ('GMP', '6.3.0'),
+    ('Z3', '4.13.0'),
+    ('CUDA', '12.6.0', '', SYSTEM),
+
+]
+configopts = '-DUSE_DEPRECATED_GCC_INSTALL_PREFIX=1 '
+# clang easyblock requires an update:
+#  GCC_INSTALL_PREFIX is deprecated and will be removed.  Use configuration
+#  files (https://clang.llvm.org/docs/UsersManual.html#configuration-files)to
+#  specify the default --gcc-install-dir= or --gcc-triple=.  --gcc-toolchain=
+#  is discouraged.  See https://github.com/llvm/llvm-project/pull/77537 for
+#  detail.
+
+# If True, Flang does not currently support building with LLVM exceptions enabled.
+enable_rtti = False
+
+assertions = True
+python_bindings = False
+skip_all_tests = True
+
+llvm_runtimes = ['libunwind', 'libcxx', 'libcxxabi']
+llvm_projects = ['polly', 'lld', 'lldb', 'clang-tools-extra', 'flang']
+
+moduleclass = 'compiler'

easybuild/easyconfigs/t/Triton/Triton-3.1.0-foss-2024a-CUDA-12.6.0.eb

@@ -0,0 +1,87 @@
+# Update 3.1.0: Thomas Hoffmann, EMBL Heidelberg, [email protected], 2024/12
+# TODO: run unittests
+easyblock = 'PythonBundle'
+
+name = 'Triton'
+version = '3.1.0'
+versionsuffix = '-CUDA-%(cudaver)s'
+# There is no 3.1 in pypi and no 3.1-tag at github. However, 5fe38ffd is version bump 3.1 in the release_3.1.x branch:
+local_commit = '5fe38ffd73c2ac6ed6323b554205186696631c6f'
+# Triton 3.1.0 commit 5fe38ffd73c2ac6ed6323b554205186696631c6f requires a particular commit of LLVM 
+# (acc. to cmake/llvm-hash.txt: 10dc3a8e916d73291269e5e2b82dd22681489aa1, 2024/05/23):
+local_clang_commit = '10dc3a8e916d73291269e5e2b82dd22681489aa1'
+
+homepage = 'https://triton-lang.org/'
+
+description = """Triton is a language and compiler for parallel programming. It aims to provide a
+Python-based programming environment for productively writing custom DNN compute
+kernels capable of running at maximal throughput on modern GPU hardware."""
+
+toolchain = {'name': 'foss', 'version': '2024a'}
+
+github_account = 'openai'
+
+builddependencies = [
+    ('Clang', '18.1.6_%s' % local_clang_commit[:7], versionsuffix),
+    ('CMake', '3.29.3'),
+    ('Ninja', '1.12.1'),
+    ('pybind11', '2.12.0'),
+    ('poetry', '1.8.3'),
+    ('nlohmann_json', '3.11.3'),
+    ('googletest', '1.15.2'),
+]
+
+dependencies = [
+    ('CUDA', '12.6.0', '', SYSTEM),
+    ('Python', '3.12.3'),
+    ('Z3', '4.13.0'),
+]
+
+use_pip = True
+
+_tr_start_dir = 'python'
+
+_tr_preinstallopts = 'export PYBIND11_SYSPATH=$EBROOTPYBIND11 && '
+_tr_preinstallopts += 'export JSON_SYSPATH=$EBROOTNLOHMANN_JSON && '
+_tr_preinstallopts += 'export LLVM_INCLUDE_DIRS=$EBROOTCLANG/include && '
+_tr_preinstallopts += 'export LLVM_LIBRARY_DIR=$EBROOTCLANG/lib && '
+_tr_preinstallopts += 'export LLVM_SYSPATH=$EBROOTCLANG && '
+_tr_preinstallopts += 'export TRITON_BUILD_WITH_CLANG_LLD=1 && '
+_tr_preinstallopts += 'export TRITON_HOME=%(builddir)s && '
+_tr_installopts = "-v "
+
+exts_list = [
+    (name, version, {
+        'installopts': "-v ",
+        'patches': [
+            # disable AMD GPU, disable all downloads, use pybind11 and json from EB:
+            'Triton-3.1.0_5fe38ff_eb_env_python_build.patch'
+        ],
+        'postinstallcmds': [
+            'rm -rf %(installdir)s/lib/python%(pyshortver)s/site-packages/triton/backends/nvidia/lib/libdevice.10.bc'
+        ],
+        'preinstallopts': _tr_preinstallopts,
+        'source_urls': ['https://github.com/triton-lang/triton/archive/'],
+        'sources': [{
+            'filename': 'v%%(version)s-%s.tar.gz' % local_commit[:7],
+            'download_filename': '%s.tar.gz' % local_commit,
+        }],
+        'start_dir': 'python',
+        'checksums': [
+            {'v3.1.0-5fe38ff.tar.gz': '933babc32b69872efbce05fe8be61129fecf52c724fadea42d8c7b2d10e16ad9'},
+            {'Triton-3.1.0_5fe38ff_eb_env_python_build.patch':
+             '6b46064b4892c7df340b6afd7ffb4abb2ea4486df9406626cd9b2c92a748705d'},
+        ],
+    }),
+    ('filelock', '3.15.1', {
+        'checksums': ['58a2549afdf9e02e10720eaa4d4470f56386d7a6f72edd7d0596337af8ed7ad8'],
+    }),
+]
+
+sanity_pip_check = True
+
+modluafooter = 'setenv("TRITON_PTXAS_PATH", pathJoin(os.getenv("CUDA_HOME"), "bin", "ptxas"));'
+modluafooter += 'setenv("TRITON_LIBDEVICE_PATH", pathJoin(os.getenv("CUDA_HOME"), "nvvm", '
+modluafooter += '"libdevice", "libdevice.10.bc")) '
+
+moduleclass = 'devel'

easybuild/easyconfigs/t/Triton/Triton-3.1.0_5fe38ff_eb_env_python_build.patch

@@ -0,0 +1,221 @@
+# Thomas Hoffmann, EMBL Heidelberg, [email protected], 2024/12
+# (adapted from Triton-2.1.0-use_eb_env_python_build.patch and Triton-2.1.0-disable_rocm_support.patch)
+# - disable support for AMD GPUs
+# - disable all downloads at build time.
+
+diff -ru triton-5fe38ffd73c2ac6ed6323b554205186696631c6f/CMakeLists.txt triton-5fe38ffd73c2ac6ed6323b554205186696631c6f_eb_env_python_build/CMakeLists.txt
+--- triton-5fe38ffd73c2ac6ed6323b554205186696631c6f/CMakeLists.txt	2024-09-10 23:44:54.000000000 +0200
++++ triton-5fe38ffd73c2ac6ed6323b554205186696631c6f_eb_env_python_build/CMakeLists.txt	2024-12-19 11:17:38.959269261 +0100
+@@ -184,7 +184,7 @@
+     ${triton_plugins}
+
+     # mlir
+-    MLIRAMDGPUDialect
++    #MLIRAMDGPUDialect
+     MLIRNVVMDialect
+     MLIRNVVMToLLVMIRTranslation
+     MLIRGPUToNVVMTransforms
+@@ -208,9 +208,9 @@
+     # LLVM
+     LLVMPasses
+     LLVMNVPTXCodeGen
+-    # LLVMNVPTXAsmPrinter
+-    LLVMAMDGPUCodeGen
+-    LLVMAMDGPUAsmParser
++    #LLVMNVPTXAsmPrinter
++    #LLVMAMDGPUCodeGen
++    #LLVMAMDGPUAsmParser
+
+   )
+   if(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64" OR # Linux arm64
+diff -ru triton-5fe38ffd73c2ac6ed6323b554205186696631c6f/python/setup.py triton-5fe38ffd73c2ac6ed6323b554205186696631c6f_eb_env_python_build/python/setup.py
+--- triton-5fe38ffd73c2ac6ed6323b554205186696631c6f/python/setup.py	2024-09-10 23:44:54.000000000 +0200
++++ triton-5fe38ffd73c2ac6ed6323b554205186696631c6f_eb_env_python_build/python/setup.py	2024-12-19 12:36:55.358107277 +0100
+@@ -130,13 +130,15 @@
+         version = pybind11_version_file.read().strip()
+     name = f"pybind11-{version}"
+     url = f"https://github.com/pybind/pybind11/archive/refs/tags/v{version}.tar.gz"
+-    return Package("pybind11", name, url, "PYBIND11_INCLUDE_DIR", "", "PYBIND11_SYSPATH")
++    #return Package("pybind11", name, url, "PYBIND11_INCLUDE_DIR", "", "PYBIND11_SYSPATH")
++    return Package("pybind11", name, '', "PYBIND11_INCLUDE_DIR", "", "PYBIND11_SYSPATH")
+
+
+ # json
+ def get_json_package_info():
+     url = "https://github.com/nlohmann/json/releases/download/v3.11.3/include.zip"
+-    return Package("json", "", url, "JSON_INCLUDE_DIR", "", "JSON_SYSPATH")
++    #return Package("json", "", url, "JSON_INCLUDE_DIR", "", "JSON_SYSPATH")
++    return Package("json", "", '', "JSON_INCLUDE_DIR", "", "JSON_SYSPATH")
+
+
+ # llvm
+@@ -201,7 +203,9 @@
+
+
+ def get_triton_cache_path():
+-    user_home = os.getenv("HOME") or os.getenv("USERPROFILE") or os.getenv("HOMEPATH") or None
++    user_home = os.getenv("TRITON_HOME")
++    if not user_home:
++        user_home = os.getenv("HOME") or os.getenv("USERPROFILE") or os.getenv("HOMEPATH") or None
+     if not user_home:
+         raise RuntimeError("Could not find user home directory")
+     return os.path.join(user_home, ".triton")
+@@ -222,14 +226,14 @@
+                 shutil.rmtree(package_root_dir)
+             os.makedirs(package_root_dir, exist_ok=True)
+             print(f'downloading and extracting {p.url} ...')
+-            with open_url(p.url) as response:
+-                if p.url.endswith(".zip"):
+-                    file_bytes = BytesIO(response.read())
+-                    with zipfile.ZipFile(file_bytes, "r") as file:
+-                        file.extractall(path=package_root_dir)
+-                else:
+-                    with tarfile.open(fileobj=response, mode="r|*") as file:
+-                        file.extractall(path=package_root_dir)
++            #with open_url(p.url) as response:
++            #    if p.url.endswith(".zip"):
++            #        file_bytes = BytesIO(response.read())
++            #        with zipfile.ZipFile(file_bytes, "r") as file:
++            #            file.extractall(path=package_root_dir)
++            #    else:
++            #        with tarfile.open(fileobj=response, mode="r|*") as file:
++            #            file.extractall(path=package_root_dir)
+             # write version url to package_dir
+             with open(os.path.join(package_dir, "version.txt"), "w") as f:
+                 f.write(p.url)
+@@ -363,12 +367,14 @@
+             "-G", "Ninja",  # Ninja is much faster than make
+             "-DCMAKE_MAKE_PROGRAM=" +
+             ninja_dir,  # Pass explicit path to ninja otherwise cmake may cache a temporary path
+-            "-DCMAKE_EXPORT_COMPILE_COMMANDS=ON", "-DLLVM_ENABLE_WERROR=ON",
++            "-DCMAKE_EXPORT_COMPILE_COMMANDS=ON",
++            # "-DLLVM_ENABLE_WERROR=ON",
+             "-DCMAKE_LIBRARY_OUTPUT_DIRECTORY=" + extdir, "-DTRITON_BUILD_TUTORIALS=OFF",
+             "-DTRITON_BUILD_PYTHON_MODULE=ON", "-DPython3_EXECUTABLE:FILEPATH=" + sys.executable,
+             "-DCMAKE_VERBOSE_MAKEFILE:BOOL=ON", "-DPYTHON_INCLUDE_DIRS=" + python_include_dir,
+             "-DTRITON_CODEGEN_BACKENDS=" + ';'.join([b.name for b in backends if not b.is_external]),
+-            "-DTRITON_PLUGIN_DIRS=" + ';'.join([b.src_dir for b in backends if b.is_external])
++            "-DTRITON_PLUGIN_DIRS=" + ';'.join([b.src_dir for b in backends if b.is_external]),
++            "-DFETCHCONTENT_FULLY_DISCONNECTED=1",  
+         ]
+         if lit_dir is not None:
+             cmake_args.append("-DLLVM_EXTERNAL_LIT=" + lit_dir)
+@@ -432,54 +438,54 @@
+ with open(nvidia_version_path, "r") as nvidia_version_file:
+     NVIDIA_TOOLCHAIN_VERSION = nvidia_version_file.read().strip()
+
+-download_and_copy(
+-    name="ptxas",
+-    src_path="bin/ptxas",
+-    variable="TRITON_PTXAS_PATH",
+-    version=NVIDIA_TOOLCHAIN_VERSION,
+-    url_func=lambda arch, version:
+-    f"https://anaconda.org/nvidia/cuda-nvcc/{version}/download/linux-{arch}/cuda-nvcc-{version}-0.tar.bz2",
+-)
+-download_and_copy(
+-    name="cuobjdump",
+-    src_path="bin/cuobjdump",
+-    variable="TRITON_CUOBJDUMP_PATH",
+-    version=NVIDIA_TOOLCHAIN_VERSION,
+-    url_func=lambda arch, version:
+-    f"https://anaconda.org/nvidia/cuda-cuobjdump/{version}/download/linux-{arch}/cuda-cuobjdump-{version}-0.tar.bz2",
+-)
+-download_and_copy(
+-    name="nvdisasm",
+-    src_path="bin/nvdisasm",
+-    variable="TRITON_NVDISASM_PATH",
+-    version=NVIDIA_TOOLCHAIN_VERSION,
+-    url_func=lambda arch, version:
+-    f"https://anaconda.org/nvidia/cuda-nvdisasm/{version}/download/linux-{arch}/cuda-nvdisasm-{version}-0.tar.bz2",
+-)
+-download_and_copy(
+-    name="cudacrt",
+-    src_path="include",
+-    variable="TRITON_CUDACRT_PATH",
+-    version=NVIDIA_TOOLCHAIN_VERSION,
+-    url_func=lambda arch, version:
+-    f"https://anaconda.org/nvidia/cuda-nvcc/{version}/download/linux-{arch}/cuda-nvcc-{version}-0.tar.bz2",
+-)
+-download_and_copy(
+-    name="cudart",
+-    src_path="include",
+-    variable="TRITON_CUDART_PATH",
+-    version=NVIDIA_TOOLCHAIN_VERSION,
+-    url_func=lambda arch, version:
+-    f"https://anaconda.org/nvidia/cuda-cudart-dev/{version}/download/linux-{arch}/cuda-cudart-dev-{version}-0.tar.bz2",
+-)
+-download_and_copy(
+-    name="cupti",
+-    src_path="include",
+-    variable="TRITON_CUPTI_PATH",
+-    version=NVIDIA_TOOLCHAIN_VERSION,
+-    url_func=lambda arch, version:
+-    f"https://anaconda.org/nvidia/cuda-cupti/{version}/download/linux-{arch}/cuda-cupti-{version}-0.tar.bz2",
+-)
++#download_and_copy(
++#    name="ptxas",
++#    src_path="bin/ptxas",
++#    variable="TRITON_PTXAS_PATH",
++#    version=NVIDIA_TOOLCHAIN_VERSION,
++#    url_func=lambda arch, version:
++#    f"https://anaconda.org/nvidia/cuda-nvcc/{version}/download/linux-{arch}/cuda-nvcc-{version}-0.tar.bz2",
++#)
++#download_and_copy(
++#    name="cuobjdump",
++#    src_path="bin/cuobjdump",
++#    variable="TRITON_CUOBJDUMP_PATH",
++#    version=NVIDIA_TOOLCHAIN_VERSION,
++#    url_func=lambda arch, version:
++#    f"https://anaconda.org/nvidia/cuda-cuobjdump/{version}/download/linux-{arch}/cuda-cuobjdump-{version}-0.tar.bz2",
++#)
++#download_and_copy(
++#    name="nvdisasm",
++#    src_path="bin/nvdisasm",
++#    variable="TRITON_NVDISASM_PATH",
++#    version=NVIDIA_TOOLCHAIN_VERSION,
++#    url_func=lambda arch, version:
++#    f"https://anaconda.org/nvidia/cuda-nvdisasm/{version}/download/linux-{arch}/cuda-nvdisasm-{version}-0.tar.bz2",
++#)
++#download_and_copy(
++#    name="cudacrt",
++#    src_path="include",
++#    variable="TRITON_CUDACRT_PATH",
++#    version=NVIDIA_TOOLCHAIN_VERSION,
++#    url_func=lambda arch, version:
++#    f"https://anaconda.org/nvidia/cuda-nvcc/{version}/download/linux-{arch}/cuda-nvcc-{version}-0.tar.bz2",
++#)
++#download_and_copy(
++#    name="cudart",
++#    src_path="include",
++#    variable="TRITON_CUDART_PATH",
++#    version=NVIDIA_TOOLCHAIN_VERSION,
++#    url_func=lambda arch, version:
++#    f"https://anaconda.org/nvidia/cuda-cudart-dev/{version}/download/linux-{arch}/cuda-cudart-dev-{version}-0.tar.bz2",
++#)
++#download_and_copy(
++#    name="cupti",
++#    src_path="include",
++#    variable="TRITON_CUPTI_PATH",
++#    version=NVIDIA_TOOLCHAIN_VERSION,
++#    url_func=lambda arch, version:
++#    f"https://anaconda.org/nvidia/cuda-cupti/{version}/download/linux-{arch}/cuda-cupti-{version}-0.tar.bz2",
++#)
+
+ backends = [*BackendInstaller.copy(["nvidia", "amd"]), *BackendInstaller.copy_externals()]
+
+diff -ru triton-5fe38ffd73c2ac6ed6323b554205186696631c6f/unittest/CMakeLists.txt triton-5fe38ffd73c2ac6ed6323b554205186696631c6f_eb_env_python_build/unittest/CMakeLists.txt
+--- triton-5fe38ffd73c2ac6ed6323b554205186696631c6f/unittest/CMakeLists.txt	2024-09-10 23:44:54.000000000 +0200
++++ triton-5fe38ffd73c2ac6ed6323b554205186696631c6f_eb_env_python_build/unittest/CMakeLists.txt	2024-12-19 13:43:42.815629305 +0100
+@@ -1,8 +1,11 @@
+-include (${CMAKE_CURRENT_SOURCE_DIR}/googletest.cmake)
++#include (${CMAKE_CURRENT_SOURCE_DIR}/googletest.cmake)
+
+ include(GoogleTest)
+ enable_testing()
+
++find_package(GTest REQUIRED)
++include_directories(${GTEST_INCLUDE_DIR})
++
+ get_property(dialect_libs GLOBAL PROPERTY MLIR_DIALECT_LIBS)
+ get_property(conversion_libs GLOBAL PROPERTY MLIR_CONVERSION_LIBS)
+ get_property(triton_libs GLOBAL PROPERTY TRITON_LIBS)