python3Packages.triton: fix pytorch.compile

This commit is contained in:
Connor Baker
2025-01-03 22:08:46 +00:00
parent 171dfafe38
commit 2586c3adf7
@@ -53,7 +53,7 @@ buildPythonPackage {
}
// lib.optionalAttrs rocmSupport { libhipDir = "${lib.getLib rocmPackages.clr}/lib"; }
// lib.optionalAttrs cudaSupport {
libcudaStubsDir = "${lib.getLib cudaPackages.cuda_cudart}/lib/stubs";
libcudaStubsDir = "${lib.getOutput "stubs" cudaPackages.cuda_cudart}/lib/stubs";
ccCmdExtraFlags = "-Wl,-rpath,${addDriverRunpath.driverLink}/lib";
}
))
@@ -82,6 +82,15 @@ buildPythonPackage {
substituteInPlace unittest/CMakeLists.txt \
--replace-fail "include (\''${CMAKE_CURRENT_SOURCE_DIR}/googletest.cmake)" ""\
--replace-fail "include(GoogleTest)" "find_package(GTest REQUIRED)"
# Patch the source code to make sure it doesn't specify a non-existent PTXAS version.
# CUDA 12.6 (the current default/max) tops out at PTXAS version 8.5.
# NOTE: This is fixed in `master`:
# https://github.com/triton-lang/triton/commit/f48dbc1b106c93144c198fbf3c4f30b2aab9d242
substituteInPlace "$NIX_BUILD_TOP/$sourceRoot/third_party/nvidia/backend/compiler.py" \
--replace-fail \
'return 80 + minor' \
'return 80 + min(minor, 5)'
'';
build-system = [ setuptools ];