From 8c902df20de93b73ad0e69a1ea7022a908f60b02 Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Thu, 18 Sep 2025 15:43:02 -0400
Subject: [PATCH 01/10] ci: build free-threaded wheels

---
 .github/workflows/build-wheel.yml |  5 +++++
 ci/test-matrix.json               | 16 ++++++++++++++++
 2 files changed, 21 insertions(+)

diff --git a/.github/workflows/build-wheel.yml b/.github/workflows/build-wheel.yml
index d015c49fa8..5d4a7310e9 100644
--- a/.github/workflows/build-wheel.yml
+++ b/.github/workflows/build-wheel.yml
@@ -30,6 +30,7 @@ jobs:
           - "3.11"
           - "3.12"
           - "3.13"
+          - "3.13t"
     name: py${{ matrix.python-version }}
     runs-on: ${{ (inputs.host-platform == 'linux-64' && 'linux-amd64-cpu8') ||
                  (inputs.host-platform == 'linux-aarch64' && 'linux-arm64-cpu8') ||
@@ -223,6 +224,10 @@ jobs:
           # workaround for actions/runner-images#12377 (the cached 3.13.4 is buggy on Windows)
           python-version: ${{ matrix.python-version == '3.13' && '3.13.5' || matrix.python-version }}
 
+      - name: verify free-threaded build
+        if: matrix.python-verison == '3.13t'
+        run: python -c 'import sys; assert not sys._is_gil_enabled()'
+
       - name: Set up Python include paths
         run: |
           if [[ "${{ inputs.host-platform }}" == linux* ]]; then
diff --git a/ci/test-matrix.json b/ci/test-matrix.json
index 9311df909d..0f7b6501a5 100644
--- a/ci/test-matrix.json
+++ b/ci/test-matrix.json
@@ -14,6 +14,8 @@
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.9",  "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.9",  "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.10", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
@@ -24,6 +26,8 @@
       { "ARCH": "arm64", "PY_VER": "3.12", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
+      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
+      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
     ],
     "nightly": [
       { "ARCH": "amd64", "PY_VER": "3.9",  "CUDA_VER": "11.8.0", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "earliest" },
@@ -51,6 +55,11 @@
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "11.8.0", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "earliest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.9",  "CUDA_VER": "11.8.0", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "earliest" },
       { "ARCH": "arm64", "PY_VER": "3.9",  "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.9",  "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
@@ -76,6 +85,11 @@
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
+      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "11.8.0", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "earliest" },
+      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
+      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
+      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
+      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
     ],
     "special_runners": {
       "amd64": [
@@ -95,6 +109,8 @@
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" }
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" }
     ]
   }
 }

From 99ccb1f2b24d1b436f306b40414f486bf36731a7 Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Thu, 18 Sep 2025 15:57:10 -0400
Subject: [PATCH 02/10] build(deps): bump cython lower bound

---
 cuda_bindings/pyproject.toml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/cuda_bindings/pyproject.toml b/cuda_bindings/pyproject.toml
index 459050038f..b7dbd7a665 100644
--- a/cuda_bindings/pyproject.toml
+++ b/cuda_bindings/pyproject.toml
@@ -2,7 +2,7 @@
 # SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
 
 [build-system]
-requires = ["setuptools>=77.0.0", "cython>=3.0,<3.1.0", "pyclibrary>=0.1.7"]
+requires = ["setuptools>=77.0.0", "cython>=3.1,<3.2", "pyclibrary>=0.1.7"]
 build-backend = "setuptools.build_meta"
 
 [project]
@@ -37,7 +37,7 @@ all = [
 ]
 
 test = [
-    "cython>=3.0,<3.1.0",
+    "cython>=3.1,<3.2",
     "setuptools>=77.0.0",
     "numpy>=1.21.1",
     "pytest>=6.2.4",

From 81279493a23b400c179c3a2c49d81346b2eb16de Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Thu, 18 Sep 2025 16:52:20 -0400
Subject: [PATCH 03/10] ci: enable freethreading for 3.13 in cibuildwheel

---
 .github/workflows/build-wheel.yml | 2 ++
 ci/tools/env-vars                 | 1 +
 2 files changed, 3 insertions(+)

diff --git a/.github/workflows/build-wheel.yml b/.github/workflows/build-wheel.yml
index 5d4a7310e9..918849c590 100644
--- a/.github/workflows/build-wheel.yml
+++ b/.github/workflows/build-wheel.yml
@@ -117,6 +117,7 @@ jobs:
           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 }}
+          CIBW_ENABLE: "cpython-freethreading"
         with:
           package-dir: ./cuda_core/
           output-dir: ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
@@ -164,6 +165,7 @@ jobs:
             CUDA_PYTHON_PARALLEL_LEVEL=${{ 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}"
+          CIBW_ENABLE: "cpython-freethreading"
         with:
           package-dir: ./cuda_bindings/
           output-dir: ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}
diff --git a/ci/tools/env-vars b/ci/tools/env-vars
index 19126cd133..104b86a16b 100755
--- a/ci/tools/env-vars
+++ b/ci/tools/env-vars
@@ -38,6 +38,7 @@ echo "PYTHON_VERSION_FORMATTED=${PYTHON_VERSION_FORMATTED}" >> $GITHUB_ENV
 
 if [[ "${1}" == "build" ]]; then
   echo "CIBW_BUILD=${CIBW_BUILD}" >> $GITHUB_ENV
+  echo "CIBW_ENABLE=cpython-freethreading" >> $GITHUB_ENV
   CUDA_BINDINGS_ARTIFACT_BASENAME="cuda-bindings-python${PYTHON_VERSION_FORMATTED}-cuda${CUDA_VER}-${HOST_PLATFORM}"
 elif [[ "${1}" == "test" ]]; then
   BUILD_CUDA_MAJOR="$(cut -d '.' -f 1 <<< ${BUILD_CUDA_VER})"

From f57ff36588291f9cdf69317d5e1c2fd9c8804da8 Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Thu, 18 Sep 2025 16:55:41 -0400
Subject: [PATCH 04/10] ci: bump cython lower bound in cuda_core

---
 cuda_core/pyproject.toml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/cuda_core/pyproject.toml b/cuda_core/pyproject.toml
index 3609396617..b833ba343f 100644
--- a/cuda_core/pyproject.toml
+++ b/cuda_core/pyproject.toml
@@ -3,7 +3,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 [build-system]
-requires = ["setuptools>=77.0.0", "Cython>=3.0"]
+requires = ["setuptools>=77.0.0", "Cython>=3.1"]
 build-backend = "setuptools.build_meta"
 
 
@@ -49,7 +49,7 @@ dependencies = [
 cu11 = ["cuda-bindings[all]==11.8.*"]
 cu12 = ["cuda-bindings[all]==12.*"]
 cu13 = ["cuda-bindings[all]==13.*"]
-test = ["cython>=3.0", "setuptools", "pytest>=6.2.4"]
+test = ["cython>=3.1", "setuptools", "pytest>=6.2.4"]
 test-cu11 = ["cuda-core[test]", "cupy-cuda11x", "cuda-toolkit[cudart]==11.*"]  # runtime headers needed by CuPy
 test-cu12 = ["cuda-core[test]", "cupy-cuda12x", "cuda-toolkit[cudart]==12.*"]  # runtime headers needed by CuPy
 test-cu13 = ["cuda-core[test]", "cupy-cuda13x", "cuda-toolkit[cudart]==13.*"]  # runtime headers needed by CuPy

From be5c2bb47eff20031429e753eac4b4d95e44c63f Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Thu, 18 Sep 2025 17:12:31 -0400
Subject: [PATCH 05/10] ci: skip tests that require llvmlite because it does
 not ship a free-threading build wheel

---
 .github/workflows/build-wheel.yml | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/build-wheel.yml b/.github/workflows/build-wheel.yml
index 918849c590..a4c4ed70d0 100644
--- a/.github/workflows/build-wheel.yml
+++ b/.github/workflows/build-wheel.yml
@@ -227,7 +227,7 @@ jobs:
           python-version: ${{ matrix.python-version == '3.13' && '3.13.5' || matrix.python-version }}
 
       - name: verify free-threaded build
-        if: matrix.python-verison == '3.13t'
+        if: endsWith(matrix.python-verison, 't')
         run: python -c 'import sys; assert not sys._is_gil_enabled()'
 
       - name: Set up Python include paths
@@ -245,6 +245,7 @@ jobs:
           pip install cuda_pathfinder/*.whl
 
       - name: Build cuda.bindings Cython tests
+        if: ${{ !endsWith(matrix.python-version, 't') }}
         run: |
           pip install $(ls ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}/*.whl)[test]
           pushd ${{ env.CUDA_BINDINGS_CYTHON_TESTS_DIR }}
@@ -252,6 +253,7 @@ jobs:
           popd
 
       - name: Upload cuda.bindings Cython tests
+        if: ${{ !endsWith(matrix.python-version, 't') }}
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02  # v4.6.2
         with:
           name: ${{ env.CUDA_BINDINGS_ARTIFACT_NAME }}-tests

From 013b588e3477a00006e570f83a65d6f1da4c76b4 Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Fri, 19 Sep 2025 09:21:58 -0400
Subject: [PATCH 06/10] ci: remove filtering out of cuda-bindings artifacts

---
 .github/workflows/build-wheel.yml | 2 --
 1 file changed, 2 deletions(-)

diff --git a/.github/workflows/build-wheel.yml b/.github/workflows/build-wheel.yml
index a4c4ed70d0..b2be4af40b 100644
--- a/.github/workflows/build-wheel.yml
+++ b/.github/workflows/build-wheel.yml
@@ -245,7 +245,6 @@ jobs:
           pip install cuda_pathfinder/*.whl
 
       - name: Build cuda.bindings Cython tests
-        if: ${{ !endsWith(matrix.python-version, 't') }}
         run: |
           pip install $(ls ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}/*.whl)[test]
           pushd ${{ env.CUDA_BINDINGS_CYTHON_TESTS_DIR }}
@@ -253,7 +252,6 @@ jobs:
           popd
 
       - name: Upload cuda.bindings Cython tests
-        if: ${{ !endsWith(matrix.python-version, 't') }}
         uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02  # v4.6.2
         with:
           name: ${{ env.CUDA_BINDINGS_ARTIFACT_NAME }}-tests

From 3310ebf09af7716b1083267b03cfd6e267d6aede Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Fri, 19 Sep 2025 09:22:10 -0400
Subject: [PATCH 07/10] ci: remove redundant environment variable setting

---
 ci/tools/env-vars | 1 -
 1 file changed, 1 deletion(-)

diff --git a/ci/tools/env-vars b/ci/tools/env-vars
index 104b86a16b..19126cd133 100755
--- a/ci/tools/env-vars
+++ b/ci/tools/env-vars
@@ -38,7 +38,6 @@ echo "PYTHON_VERSION_FORMATTED=${PYTHON_VERSION_FORMATTED}" >> $GITHUB_ENV
 
 if [[ "${1}" == "build" ]]; then
   echo "CIBW_BUILD=${CIBW_BUILD}" >> $GITHUB_ENV
-  echo "CIBW_ENABLE=cpython-freethreading" >> $GITHUB_ENV
   CUDA_BINDINGS_ARTIFACT_BASENAME="cuda-bindings-python${PYTHON_VERSION_FORMATTED}-cuda${CUDA_VER}-${HOST_PLATFORM}"
 elif [[ "${1}" == "test" ]]; then
   BUILD_CUDA_MAJOR="$(cut -d '.' -f 1 <<< ${BUILD_CUDA_VER})"

From 1cffc268504caf8b56cf63df6a0a5f208abc3382 Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Fri, 19 Sep 2025 10:24:21 -0400
Subject: [PATCH 08/10] ci: fix syntax errors in ci/test-matrix.json

---
 ci/test-matrix.json | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/ci/test-matrix.json b/ci/test-matrix.json
index 0f7b6501a5..c3524a81e0 100644
--- a/ci/test-matrix.json
+++ b/ci/test-matrix.json
@@ -25,7 +25,7 @@
       { "ARCH": "arm64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.12", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
+      { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
     ],
@@ -84,7 +84,7 @@
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
+      { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "11.8.0", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "earliest" },
       { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
@@ -102,13 +102,15 @@
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" }
+      { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
+      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" }
     ],
     "nightly": [
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "11.8.0", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" }
+      { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "t4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" }
     ]

From 96ee1c22ef73b8e0f191329e29af8ea9e44cfbff Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Fri, 19 Sep 2025 12:25:10 -0400
Subject: [PATCH 09/10] ci: add optional dependency group to avoid installation
 of cupy when testing the FT builds

---
 ci/tools/run-tests       | 10 ++++++++--
 cuda_core/pyproject.toml |  6 ++++++
 2 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/ci/tools/run-tests b/ci/tools/run-tests
index 22d6bd07c6..8686590974 100755
--- a/ci/tools/run-tests
+++ b/ci/tools/run-tests
@@ -77,12 +77,18 @@ elif [[ "${test_module}" == "core" ]]; then
   echo "Installing core wheel"
   pwd
   ls
+
+  FREE_THREADING=""
+  if python -c 'import sys; assert not sys._is_gil_enabled()' 2> /dev/null; then
+    FREE_THREADING+="-ft"
+  fi
+
   if [[ "${LOCAL_CTK}" == 1 ]]; then
     # We already installed cuda-bindings, and all CTK components exist locally,
     # so just install the test dependencies.
-    pip install $(ls *.whl)["test-cu${TEST_CUDA_MAJOR}"]
+    pip install $(ls *.whl)["test-cu${TEST_CUDA_MAJOR}${FREE_THREADING}"]
   else
-    pip install $(ls *.whl)["cu${TEST_CUDA_MAJOR}","test-cu${TEST_CUDA_MAJOR}"]
+    pip install $(ls *.whl)["cu${TEST_CUDA_MAJOR}","test-cu${TEST_CUDA_MAJOR}${FREE_THREADING}"]
   fi
   popd
   pushd ./cuda_core
diff --git a/cuda_core/pyproject.toml b/cuda_core/pyproject.toml
index b833ba343f..690a49a05f 100644
--- a/cuda_core/pyproject.toml
+++ b/cuda_core/pyproject.toml
@@ -49,10 +49,16 @@ dependencies = [
 cu11 = ["cuda-bindings[all]==11.8.*"]
 cu12 = ["cuda-bindings[all]==12.*"]
 cu13 = ["cuda-bindings[all]==13.*"]
+# TODO: these should all be in development dependencies; optional dependencies
+# are for features exposed to *users*, not a dumping ground for all tooling
+# needed to build and test the project
 test = ["cython>=3.1", "setuptools", "pytest>=6.2.4"]
 test-cu11 = ["cuda-core[test]", "cupy-cuda11x", "cuda-toolkit[cudart]==11.*"]  # runtime headers needed by CuPy
 test-cu12 = ["cuda-core[test]", "cupy-cuda12x", "cuda-toolkit[cudart]==12.*"]  # runtime headers needed by CuPy
 test-cu13 = ["cuda-core[test]", "cupy-cuda13x", "cuda-toolkit[cudart]==13.*"]  # runtime headers needed by CuPy
+# free threaded build, cupy doesn't support free-threaded builds yet, so avoid installing it for now
+# TODO: cupy should support free threaded builds
+test-cu13-ft = ["cuda-core[test]", "cuda-toolkit[cudart]==13.*"]
 
 [project.urls]
 homepage = "https://nvidia.github.io/cuda-python/"

From 86fc8f4583af1a08744e2625c763dc324bf5cd77 Mon Sep 17 00:00:00 2001
From: Phillip Cloud <417981+cpcloud@users.noreply.github.com>
Date: Fri, 19 Sep 2025 13:33:07 -0400
Subject: [PATCH 10/10] ci: remove 12.9.x from builds for now

---
 ci/test-matrix.json | 9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/ci/test-matrix.json b/ci/test-matrix.json
index c3524a81e0..0d6fdb7e71 100644
--- a/ci/test-matrix.json
+++ b/ci/test-matrix.json
@@ -14,7 +14,6 @@
       { "ARCH": "amd64", "PY_VER": "3.12", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
-      { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "amd64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "l4", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.9",  "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.9",  "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
@@ -26,7 +25,6 @@
       { "ARCH": "arm64", "PY_VER": "3.12", "CUDA_VER": "13.0.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "13.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
     ],
     "nightly": [
@@ -84,12 +82,7 @@
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
       { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "11.8.0", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "earliest" },
-      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "11.8.0", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.0.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "0", "GPU": "a100", "DRIVER": "latest" },
-      { "ARCH": "arm64", "PY_VER": "3.13t", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
+      { "ARCH": "arm64", "PY_VER": "3.13", "CUDA_VER": "12.9.1", "LOCAL_CTK": "1", "GPU": "a100", "DRIVER": "latest" }
     ],
     "special_runners": {
       "amd64": [