From c4e806dc80c492a2b620c83a4fea0dfeee23541d Mon Sep 17 00:00:00 2001
From: Leo Fang <leof@nvidia.com>
Date: Wed, 2 Jul 2025 03:06:42 +0000
Subject: [PATCH 1/3] fix CI parallelism

---
 .github/workflows/build-wheel.yml | 4 ++--
 ci/tools/env-vars                 | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/build-wheel.yml b/.github/workflows/build-wheel.yml
index 3bd427256..86eef7f39 100644
--- a/.github/workflows/build-wheel.yml
+++ b/.github/workflows/build-wheel.yml
@@ -121,11 +121,11 @@ jobs:
           CIBW_ENVIRONMENT_LINUX: >
             CUDA_PATH=/host/${{ env.CUDA_PATH }}
             LIBRARY_PATH=/host/${{ env.CUDA_PATH }}/lib
-            CUDA_BINDINGS_PARALLEL_LEVEL=${{ env.CUDA_BINDINGS_PARALLEL_LEVEL }}
+            CUDA_PYTHON_PARALLEL_LEVEL=${{ env.CUDA_PYTHON_PARALLEL_LEVEL }}
           CIBW_ENVIRONMENT_WINDOWS: >
             CUDA_HOME="$(cygpath -w ${{ env.CUDA_PATH }})"
             LIB="${CUDA_HOME}\\lib\\x64;${LIB}"
-            CUDA_BINDINGS_PARALLEL_LEVEL=${{ env.CUDA_BINDINGS_PARALLEL_LEVEL }}
+            CUDA_PYTHON_PARALLEL_LEVEL=${{ env.CUDA_PYTHON_PARALLEL_LEVEL }}
           CIBW_BEFORE_BUILD_WINDOWS: "pip install delvewheel"
           CIBW_REPAIR_WHEEL_COMMAND_WINDOWS: "delvewheel repair --namespace-pkg cuda -w {dest_dir} {wheel}"
         with:
diff --git a/ci/tools/env-vars b/ci/tools/env-vars
index 7211dc08f..f1882a990 100755
--- a/ci/tools/env-vars
+++ b/ci/tools/env-vars
@@ -28,7 +28,7 @@ elif [[ "${HOST_PLATFORM}" == win* ]]; then
 fi
 
 echo "${TOOLS_PATH}" >> $GITHUB_PATH
-echo "CUDA_BINDINGS_PARALLEL_LEVEL=$(nproc)" >> $GITHUB_ENV
+echo "CUDA_PYTHON_PARALLEL_LEVEL=$(nproc)" >> $GITHUB_ENV
 CUDA_CORE_ARTIFACT_BASENAME="cuda-core-python${PYTHON_VERSION_FORMATTED}-${HOST_PLATFORM}"
 echo "CUDA_CORE_ARTIFACT_BASENAME=${CUDA_CORE_ARTIFACT_BASENAME}" >> $GITHUB_ENV
 echo "CUDA_CORE_ARTIFACT_NAME=${CUDA_CORE_ARTIFACT_BASENAME}-${SHA}" >> $GITHUB_ENV

From 6834001556937540d48f213cc540823b18fa26f9 Mon Sep 17 00:00:00 2001
From: Leo Fang <leof@nvidia.com>
Date: Wed, 2 Jul 2025 03:13:37 +0000
Subject: [PATCH 2/3] make cuda.core also support CUDA_PYTHON_PARALLEL_LEVEL

---
 .github/workflows/build-wheel.yml | 2 ++
 cuda_core/setup.py                | 6 +++++-
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/build-wheel.yml b/.github/workflows/build-wheel.yml
index 86eef7f39..c8a163c97 100644
--- a/.github/workflows/build-wheel.yml
+++ b/.github/workflows/build-wheel.yml
@@ -78,6 +78,8 @@ jobs:
           CIBW_BUILD_VERBOSITY: 1
           CIBW_BEFORE_BUILD_WINDOWS: "pip install delvewheel"
           CIBW_REPAIR_WHEEL_COMMAND_WINDOWS: "delvewheel repair --namespace-pkg cuda -w {dest_dir} {wheel}"
+          CIBW_ENVIRONMENT: >
+            CUDA_PYTHON_PARALLEL_LEVEL=${{ env.CUDA_PYTHON_PARALLEL_LEVEL }}
         with:
           package-dir: ./cuda_core/
           output-dir: ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
diff --git a/cuda_core/setup.py b/cuda_core/setup.py
index 26b9a5ab0..7a5cf3bf0 100644
--- a/cuda_core/setup.py
+++ b/cuda_core/setup.py
@@ -9,6 +9,10 @@
 from setuptools import Extension, setup
 from setuptools.command.build_ext import build_ext as _build_ext
 
+
+nthreads = int(os.environ.get("CUDA_PYTHON_PARALLEL_LEVEL", os.cpu_count() // 2))
+
+
 # It seems setuptools' wildcard support has problems for namespace packages,
 # so we explicitly spell out all Extension instances.
 root_module = "cuda.core.experimental"
@@ -33,7 +37,7 @@ def strip_prefix_suffix(filename):
 
 class build_ext(_build_ext):
     def build_extensions(self):
-        self.parallel = os.cpu_count() // 2
+        self.parallel = nthreads
         super().build_extensions()
 
 

From 11796bcfe613dbbab0e1fe49f4badd6418319d27 Mon Sep 17 00:00:00 2001
From: "pre-commit-ci[bot]"
 <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Date: Wed, 2 Jul 2025 03:18:11 +0000
Subject: [PATCH 3/3] [pre-commit.ci] auto code formatting

---
 cuda_core/setup.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/cuda_core/setup.py b/cuda_core/setup.py
index 7a5cf3bf0..3e93e6e9a 100644
--- a/cuda_core/setup.py
+++ b/cuda_core/setup.py
@@ -9,7 +9,6 @@
 from setuptools import Extension, setup
 from setuptools.command.build_ext import build_ext as _build_ext
 
-
 nthreads = int(os.environ.get("CUDA_PYTHON_PARALLEL_LEVEL", os.cpu_count() // 2))