From 2aca1ce3c56aa90b6f7c93d43085e25f50670838 Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Thu, 20 Jul 2023 12:43:46 -0500
Subject: [PATCH 01/27] skip tests that can corrupt memory (#3)

---
 nvjitlink/tests/test_nvjitlink.py     |  8 ++++----
 nvjitlink/tests/test_nvjitlink_api.py | 16 ++++++++--------
 2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/nvjitlink/tests/test_nvjitlink.py b/nvjitlink/tests/test_nvjitlink.py
index 39910b4a..72d43192 100644
--- a/nvjitlink/tests/test_nvjitlink.py
+++ b/nvjitlink/tests/test_nvjitlink.py
@@ -48,7 +48,7 @@ def device_functions_ptx():
 def undefined_extern_cubin():
     return read_test_file('undefined_extern.cubin')
 
-
+@pytest.mark.skip
 def test_create_no_arch_error():
     # nvjitlink expects at least the architecture to be specified.
     with pytest.raises(RuntimeError,
@@ -63,13 +63,13 @@ def test_invalid_arch_error():
                        match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
         _nvjitlinklib.create('-arch=sm_XX')
 
-
+@pytest.mark.skip
 def test_unrecognized_option_error():
     with pytest.raises(RuntimeError,
                        match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
         _nvjitlinklib.create('-fictitious_option')
 
-
+@pytest.mark.skip
 def test_invalid_option_type_error():
     with pytest.raises(TypeError,
                        match='Expecting only strings'):
@@ -108,7 +108,7 @@ def test_add_file(input_file, input_type, request):
     _nvjitlinklib.add_data(handle, input_type.value, data, filename)
     _nvjitlinklib.destroy(handle)
 
-
+@pytest.mark.skip
 def test_get_error_log(undefined_extern_cubin):
     handle = _nvjitlinklib.create('-arch=sm_75')
     filename, data = undefined_extern_cubin
diff --git a/nvjitlink/tests/test_nvjitlink_api.py b/nvjitlink/tests/test_nvjitlink_api.py
index d012403a..85aae91b 100644
--- a/nvjitlink/tests/test_nvjitlink_api.py
+++ b/nvjitlink/tests/test_nvjitlink_api.py
@@ -30,21 +30,21 @@ def undefined_extern_cubin():
     with open(fatbin_path, 'rb') as f:
         return f.read()
 
-
+@pytest.mark.skip
 def test_create_no_arch_error():
     # nvlink expects at least the architecture to be specified.
     with pytest.raises(NvJitLinkError,
                        match='NVJITLINK_ERROR_MISSING_ARCH error'):
         NvJitLinker()
 
-
+@pytest.mark.skip
 def test_invalid_arch_error():
     # sm_XX is not a valid architecture
     with pytest.raises(NvJitLinkError,
                        match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
         NvJitLinker('-arch=sm_XX')
 
-
+@pytest.mark.skip
 def test_invalid_option_type_error():
     with pytest.raises(TypeError,
                        match='Expecting only strings'):
@@ -61,7 +61,7 @@ def test_add_cubin(device_functions_cubin):
     name = 'test_device_functions.cubin'
     nvjitlinker.add_cubin(device_functions_cubin, name)
 
-
+@pytest.mark.skip
 def test_add_incompatible_cubin_arch_error(device_functions_cubin):
     nvjitlinker = NvJitLinker('-arch=sm_70')
     name = 'test_device_functions.cubin'
@@ -81,7 +81,7 @@ def test_add_fatbin_sm70(device_functions_fatbin):
     name = 'test_device_functions.fatbin'
     nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
-
+@pytest.mark.skip
 def test_add_incompatible_fatbin_arch_error(device_functions_fatbin):
     nvjitlinker = NvJitLinker('-arch=sm_80')
     name = 'test_device_functions.fatbin'
@@ -89,7 +89,7 @@ def test_add_incompatible_fatbin_arch_error(device_functions_fatbin):
                        match='NVJITLINK_ERROR_INVALID_INPUT error'):
         nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
-
+@pytest.mark.skip
 def test_add_cubin_with_fatbin_error(device_functions_fatbin):
     nvjitlinker = NvJitLinker('-arch=sm_75')
     name = 'test_device_functions.fatbin'
@@ -105,7 +105,7 @@ def test_add_fatbin_with_cubin(device_functions_cubin):
     name = 'test_device_functions.cubin'
     nvjitlinker.add_fatbin(device_functions_cubin, name)
 
-
+@pytest.mark.skip
 def test_duplicate_symbols_cubin_and_fatbin(device_functions_cubin,
                                             device_functions_fatbin):
     # This link errors because the cubin and the fatbin contain the same
@@ -136,7 +136,7 @@ def test_get_linked_cubin(device_functions_cubin):
     # Just check we got something that looks like an ELF
     assert cubin[:4] == b'\x7fELF'
 
-
+@pytest.mark.skip
 def test_get_error_log(undefined_extern_cubin):
     nvjitlinker = NvJitLinker('-arch=sm_75')
     name = 'undefined_extern.cubin'

From e489f502eefaf006d7b6dddf546e6a68154d950f Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Tue, 25 Jul 2023 15:14:50 -0500
Subject: [PATCH 02/27] nvjitlink->pynvjitlink (#5)

* nvjitlink->pynvjitlink

* update codeowners file

* lint codeowners
---
 .github/CODEOWNERS                                 |  13 ++++---------
 .github/ISSUE_TEMPLATE/bug_report.md               |   4 ++--
 .github/ISSUE_TEMPLATE/feature-request.md          |   4 ++--
 .github/ISSUE_TEMPLATE/submit-question.md          |   2 +-
 .github/PULL_REQUEST_TEMPLATE.md                   |   2 +-
 CHANGELOG.md                                       |   2 +-
 CMakeLists.txt                                     |   6 +++---
 README.md                                          |  10 +++++-----
 {nvjitlink => pynvjitlink}/__init__.py             |   2 +-
 {nvjitlink => pynvjitlink}/_nvjitlinklib.cpp       |   2 +-
 {nvjitlink => pynvjitlink}/api.py                  |   2 +-
 {nvjitlink => pynvjitlink}/patch.py                |   2 +-
 .../tests/test_device_functions.a                  | Bin
 .../tests/test_device_functions.cubin              | Bin
 .../tests/test_device_functions.fatbin             | Bin
 .../tests/test_device_functions.ltoir              | Bin
 .../tests/test_device_functions.o                  | Bin
 .../tests/test_device_functions.ptx                |   0
 .../tests/test_pynvjitlink.py                      |   4 ++--
 .../tests/test_pynvjitlink_api.py                  |   2 +-
 .../tests/undefined_extern.cubin                   | Bin
 pyproject.toml                                     |   2 +-
 test_binary_generation/Makefile                    |   4 ++--
 23 files changed, 29 insertions(+), 34 deletions(-)
 rename {nvjitlink => pynvjitlink}/__init__.py (65%)
 rename {nvjitlink => pynvjitlink}/_nvjitlinklib.cpp (99%)
 rename {nvjitlink => pynvjitlink}/api.py (98%)
 rename {nvjitlink => pynvjitlink}/patch.py (98%)
 rename {nvjitlink => pynvjitlink}/tests/test_device_functions.a (100%)
 rename {nvjitlink => pynvjitlink}/tests/test_device_functions.cubin (100%)
 rename {nvjitlink => pynvjitlink}/tests/test_device_functions.fatbin (100%)
 rename {nvjitlink => pynvjitlink}/tests/test_device_functions.ltoir (100%)
 rename {nvjitlink => pynvjitlink}/tests/test_device_functions.o (100%)
 rename {nvjitlink => pynvjitlink}/tests/test_device_functions.ptx (100%)
 rename nvjitlink/tests/test_nvjitlink.py => pynvjitlink/tests/test_pynvjitlink.py (98%)
 rename nvjitlink/tests/test_nvjitlink_api.py => pynvjitlink/tests/test_pynvjitlink_api.py (99%)
 rename {nvjitlink => pynvjitlink}/tests/undefined_extern.cubin (100%)

diff --git a/.github/CODEOWNERS b/.github/CODEOWNERS
index c0463368..b7ac431a 100644
--- a/.github/CODEOWNERS
+++ b/.github/CODEOWNERS
@@ -1,14 +1,9 @@
 #python code owners
-nvjitlink/            @rapidsai/nvjitlink-python-codeowners
+pynvjitlink/            @rapidsai/pynvjitlink-write
 
 #cmake code owners
-**/CMakeLists.txt  @rapidsai/nvjitlink-cmake-codeowners
-**/cmake/          @rapidsai/nvjitlink-cmake-codeowners
+**/CMakeLists.txt  @rapidsai/pynvjitlink-write
+**/cmake/          @rapidsai/pynvjitlink-write
 
 #build/ops code owners
-.github/           @rapidsai/ops-codeowners 
-ci/                @rapidsai/ops-codeowners
-conda/             @rapidsai/ops-codeowners
-**/Dockerfile      @rapidsai/ops-codeowners
-**/.dockerignore   @rapidsai/ops-codeowners
-docker/            @rapidsai/ops-codeowners
+.github/           @rapidsai/ops-codeowners
diff --git a/.github/ISSUE_TEMPLATE/bug_report.md b/.github/ISSUE_TEMPLATE/bug_report.md
index 85bcc1e0..34df795e 100644
--- a/.github/ISSUE_TEMPLATE/bug_report.md
+++ b/.github/ISSUE_TEMPLATE/bug_report.md
@@ -1,6 +1,6 @@
 ---
 name: Bug report
-about: Create a bug report to help us improve nvjitlink
+about: Create a bug report to help us improve pynvjitlink
 title: "[BUG]"
 labels: "? - Needs Triage, bug"
 assignees: ''
@@ -18,7 +18,7 @@ A clear and concise description of what you expected to happen.
 
 **Environment details (please complete the following information):**
  - Environment location: [Bare-metal, Docker, Cloud(specify cloud provider)]
- - Method of nvjitlink install: [conda, Docker, or from source]
+ - Method of pynvjitlink install: [conda, Docker, or from source]
    - If method of install is [Docker], provide `docker pull` & `docker run` commands used
  
 
diff --git a/.github/ISSUE_TEMPLATE/feature-request.md b/.github/ISSUE_TEMPLATE/feature-request.md
index 3457b336..d8efc8f9 100644
--- a/.github/ISSUE_TEMPLATE/feature-request.md
+++ b/.github/ISSUE_TEMPLATE/feature-request.md
@@ -1,6 +1,6 @@
 ---
 name: Feature request
-about: Suggest an idea for nvjitlink
+about: Suggest an idea for pynvjitlink
 title: "[FEA]"
 labels: "? - Needs Triage, feature request"
 assignees: ''
@@ -8,7 +8,7 @@ assignees: ''
 ---
 
 **Is your feature request related to a problem? Please describe.**
-A clear and concise description of what the problem is. Ex. I wish I could use nvjitlink to do [...]
+A clear and concise description of what the problem is. Ex. I wish I could use pynvjitlink to do [...]
 
 **Describe the solution you'd like**
 A clear and concise description of what you want to happen.
diff --git a/.github/ISSUE_TEMPLATE/submit-question.md b/.github/ISSUE_TEMPLATE/submit-question.md
index c1890912..b3b86d1d 100644
--- a/.github/ISSUE_TEMPLATE/submit-question.md
+++ b/.github/ISSUE_TEMPLATE/submit-question.md
@@ -1,6 +1,6 @@
 ---
 name: Submit question
-about: Ask a general question about nvjitlink
+about: Ask a general question about pynvjitlink
 title: "[QST]"
 labels: "? - Needs Triage, question"
 assignees: ''
diff --git a/.github/PULL_REQUEST_TEMPLATE.md b/.github/PULL_REQUEST_TEMPLATE.md
index 4e930c42..03d36a01 100644
--- a/.github/PULL_REQUEST_TEMPLATE.md
+++ b/.github/PULL_REQUEST_TEMPLATE.md
@@ -1,6 +1,6 @@
 <!--
 
-Thank you for contributing to nvjitlink :)
+Thank you for contributing to pynvjitlink :)
 
 Here are some guidelines to help the review process go smoothly.
 
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 54e30c07..3038ebd3 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,4 +1,4 @@
-# nvjitlink 0.0.0 (DD Mon YYYY)
+# pynvjitlink 0.0.0 (DD Mon YYYY)
 
 ## New Features
 
diff --git a/CMakeLists.txt b/CMakeLists.txt
index c5aef5a2..e222b100 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,14 +1,14 @@
 cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 
 project(
-  nvjitlink
+  pynvjitlink
   VERSION 0.1.0
   LANGUAGES CXX
 )
 
 find_package(Python COMPONENTS Interpreter Development REQUIRED)
 
-Python_add_library(_nvjitlinklib MODULE nvjitlink/_nvjitlinklib.cpp WITH_SOABI)
+Python_add_library(_nvjitlinklib MODULE pynvjitlink/_nvjitlinklib.cpp WITH_SOABI)
 
 find_package(
   CUDAToolkit 12.0 REQUIRED
@@ -21,4 +21,4 @@ if(WIN32)
 endif()
 target_compile_options(_nvjitlinklib PRIVATE -Werror -Wall ${opt_option})
 
-install(TARGETS _nvjitlinklib LIBRARY DESTINATION nvjitlink)
+install(TARGETS _nvjitlinklib LIBRARY DESTINATION pynvjitlink)
diff --git a/README.md b/README.md
index cfaeba09..a18c78b8 100644
--- a/README.md
+++ b/README.md
@@ -1,15 +1,15 @@
-# <div align="left"><img src="https://rapids.ai/assets/images/rapids_logo.png" width="90px"/>&nbsp;nvjitlink</div>
+# <div align="left"><img src="https://rapids.ai/assets/images/rapids_logo.png" width="90px"/>&nbsp;pynvjitlink</div>
 
-The [RAPIDS](https://rapids.ai) nvjitlink library provides a Python binding for the 
+The [RAPIDS](https://rapids.ai) pynvjitlink library provides a Python binding for the 
 [nvJitLink library](https://docs.nvidia.com/cuda/nvJitLink/index.html). It is presently a
 work-in-progress.
 
-**NOTE:** For the latest stable [README.md](https://github.com/rapidsai/nvjitlink/blob/main/README.md) ensure you are on the `main` branch.
+**NOTE:** For the latest stable [README.md](https://github.com/rapidsai/pynvjitlink/blob/main/README.md) ensure you are on the `main` branch.
 
 ## Quick Start
 TODO
 
-## Install nvjitlink
+## Install pynvjitlink
 Install with either:
 
 ```
@@ -30,4 +30,4 @@ TODO
 
 ## Contributing Guide
 
-Review the [CONTRIBUTING.md](https://github.com/rapidsai/nvjitlink/blob/main/CONTRIBUTING.md) file for information on how to contribute code and issues to the project.
+Review the [CONTRIBUTING.md](https://github.com/rapidsai/pynvjitlink/blob/main/CONTRIBUTING.md) file for information on how to contribute code and issues to the project.
diff --git a/nvjitlink/__init__.py b/pynvjitlink/__init__.py
similarity index 65%
rename from nvjitlink/__init__.py
rename to pynvjitlink/__init__.py
index 7b99f87d..9ad447f1 100644
--- a/nvjitlink/__init__.py
+++ b/pynvjitlink/__init__.py
@@ -1,6 +1,6 @@
 # Copyright (c) 2023, NVIDIA CORPORATION.
 
-from nvjitlink.api import NvJitLinker, NvJitLinkError
+from pynvjitlink.api import NvJitLinker, NvJitLinkError
 
 __all__ = (NvJitLinker, NvJitLinkError)
 __version__ = '0.0.0'
diff --git a/nvjitlink/_nvjitlinklib.cpp b/pynvjitlink/_nvjitlinklib.cpp
similarity index 99%
rename from nvjitlink/_nvjitlinklib.cpp
rename to pynvjitlink/_nvjitlinklib.cpp
index 83488166..421731d9 100644
--- a/nvjitlink/_nvjitlinklib.cpp
+++ b/pynvjitlink/_nvjitlinklib.cpp
@@ -323,7 +323,7 @@ static PyMethodDef ext_methods[] = {
     {nullptr}};
 
 static struct PyModuleDef moduledef = {
-    PyModuleDef_HEAD_INIT, "nvjitlink",
+    PyModuleDef_HEAD_INIT, "pynvjitlink",
     "Provides access to nvJitLink API methods", -1, ext_methods};
 
 PyMODINIT_FUNC PyInit__nvjitlinklib(void) {
diff --git a/nvjitlink/api.py b/pynvjitlink/api.py
similarity index 98%
rename from nvjitlink/api.py
rename to pynvjitlink/api.py
index 20cac5ed..f8aff6d0 100644
--- a/nvjitlink/api.py
+++ b/pynvjitlink/api.py
@@ -1,7 +1,7 @@
 # Copyright (c) 2023, NVIDIA CORPORATION.
 
 from enum import Enum
-from nvjitlink import _nvjitlinklib
+from pynvjitlink import _nvjitlinklib
 
 import weakref
 
diff --git a/nvjitlink/patch.py b/pynvjitlink/patch.py
similarity index 98%
rename from nvjitlink/patch.py
rename to pynvjitlink/patch.py
index 784ef27b..2a4565e6 100644
--- a/nvjitlink/patch.py
+++ b/pynvjitlink/patch.py
@@ -1,5 +1,5 @@
 # Copyright (c) 2023, NVIDIA CORPORATION.
-from nvjitlink.api import NvJitLinker, NvJitLinkError
+from pynvjitlink.api import NvJitLinker, NvJitLinkError
 
 import os
 import pathlib
diff --git a/nvjitlink/tests/test_device_functions.a b/pynvjitlink/tests/test_device_functions.a
similarity index 100%
rename from nvjitlink/tests/test_device_functions.a
rename to pynvjitlink/tests/test_device_functions.a
diff --git a/nvjitlink/tests/test_device_functions.cubin b/pynvjitlink/tests/test_device_functions.cubin
similarity index 100%
rename from nvjitlink/tests/test_device_functions.cubin
rename to pynvjitlink/tests/test_device_functions.cubin
diff --git a/nvjitlink/tests/test_device_functions.fatbin b/pynvjitlink/tests/test_device_functions.fatbin
similarity index 100%
rename from nvjitlink/tests/test_device_functions.fatbin
rename to pynvjitlink/tests/test_device_functions.fatbin
diff --git a/nvjitlink/tests/test_device_functions.ltoir b/pynvjitlink/tests/test_device_functions.ltoir
similarity index 100%
rename from nvjitlink/tests/test_device_functions.ltoir
rename to pynvjitlink/tests/test_device_functions.ltoir
diff --git a/nvjitlink/tests/test_device_functions.o b/pynvjitlink/tests/test_device_functions.o
similarity index 100%
rename from nvjitlink/tests/test_device_functions.o
rename to pynvjitlink/tests/test_device_functions.o
diff --git a/nvjitlink/tests/test_device_functions.ptx b/pynvjitlink/tests/test_device_functions.ptx
similarity index 100%
rename from nvjitlink/tests/test_device_functions.ptx
rename to pynvjitlink/tests/test_device_functions.ptx
diff --git a/nvjitlink/tests/test_nvjitlink.py b/pynvjitlink/tests/test_pynvjitlink.py
similarity index 98%
rename from nvjitlink/tests/test_nvjitlink.py
rename to pynvjitlink/tests/test_pynvjitlink.py
index 72d43192..4fa53f90 100644
--- a/nvjitlink/tests/test_nvjitlink.py
+++ b/pynvjitlink/tests/test_pynvjitlink.py
@@ -3,8 +3,8 @@
 import os
 import pytest
 
-from nvjitlink import _nvjitlinklib
-from nvjitlink.api import InputType
+from pynvjitlink import _nvjitlinklib
+from pynvjitlink.api import InputType
 
 
 def read_test_file(filename):
diff --git a/nvjitlink/tests/test_nvjitlink_api.py b/pynvjitlink/tests/test_pynvjitlink_api.py
similarity index 99%
rename from nvjitlink/tests/test_nvjitlink_api.py
rename to pynvjitlink/tests/test_pynvjitlink_api.py
index 85aae91b..8c74d814 100644
--- a/nvjitlink/tests/test_nvjitlink_api.py
+++ b/pynvjitlink/tests/test_pynvjitlink_api.py
@@ -4,7 +4,7 @@
 import sys
 import os
 
-from nvjitlink import NvJitLinker, NvJitLinkError
+from pynvjitlink import NvJitLinker, NvJitLinkError
 
 
 @pytest.fixture
diff --git a/nvjitlink/tests/undefined_extern.cubin b/pynvjitlink/tests/undefined_extern.cubin
similarity index 100%
rename from nvjitlink/tests/undefined_extern.cubin
rename to pynvjitlink/tests/undefined_extern.cubin
diff --git a/pyproject.toml b/pyproject.toml
index fcf1792d..d766e7d2 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -11,7 +11,7 @@ requires = ["scikit-build-core"]
 build-backend = "scikit_build_core.build"
 
 [project]
-name = "nvjitlink"
+name = "pynvjitlink"
 version = "0.1.0"
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }
diff --git a/test_binary_generation/Makefile b/test_binary_generation/Makefile
index 4d772556..3671d566 100644
--- a/test_binary_generation/Makefile
+++ b/test_binary_generation/Makefile
@@ -1,4 +1,4 @@
-# Generates the input files used by the nvjitlink binding test suite
+# Generates the input files used by the pynvjitlink binding test suite
 
 NVCC_FLAGS := -O3 -rdc true
 CUBIN_FLAGS := -gencode arch=compute_75,code=sm_75 --cubin
@@ -9,7 +9,7 @@ OBJECT_FLAGS := -gencode arch=compute_75,code=compute_75 -dc
 LIBRARY_FLAGS := -gencode arch=compute_75,code=compute_75 -lib
 
 
-OUTPUT_DIR := ../nvjitlink/tests
+OUTPUT_DIR := ../pynvjitlink/tests
 
 all:
 	nvcc $(NVCC_FLAGS) $(CUBIN_FLAGS) -o $(OUTPUT_DIR)/undefined_extern.cubin undefined_extern.cu

From d4a45cc3e94ca44a8036917ce66b0da735f04700 Mon Sep 17 00:00:00 2001
From: Vyas Ramasubramani <vyasr@nvidia.com>
Date: Mon, 7 Aug 2023 15:53:10 -0700
Subject: [PATCH 03/27] Remove exclude needed to patch scikit-build-core bug
 (#6)

---
 pyproject.toml | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/pyproject.toml b/pyproject.toml
index d766e7d2..692b3cdd 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -2,9 +2,6 @@
 cmake.minimum-version = "3.26.4"
 ninja.make-fallback = true
 build-dir = "build/{wheel_tag}"
-# Workaround for
-# https://github.com/scikit-build/scikit-build-core/issues/437
-sdist.exclude = [".gitignore"]
 
 [build-system]
 requires = ["scikit-build-core"]

From 888ed76ee107adb5710e387d82d33bcd9d621b54 Mon Sep 17 00:00:00 2001
From: Graham Markall <535640+gmarkall@users.noreply.github.com>
Date: Mon, 16 Oct 2023 14:32:23 +0100
Subject: [PATCH 04/27] Support Numba 0.58 (#7)

The `NvrtcProgram` class moved to a new module,
`numba.cuda.cudadrv.nvrtc` in Numba 0.58 (in PR numba/numba#9086), so we
need to import it from there for that version onwards.
---
 pynvjitlink/patch.py | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/pynvjitlink/patch.py b/pynvjitlink/patch.py
index 2a4565e6..1928e050 100644
--- a/pynvjitlink/patch.py
+++ b/pynvjitlink/patch.py
@@ -25,8 +25,12 @@
 
 if _numba_version_ok:
     from numba.core import config
-    from numba.cuda.cudadrv.driver import (FILE_EXTENSION_MAP, NvrtcProgram,
-                                           Linker, LinkerError)
+    from numba.cuda.cudadrv.driver import (FILE_EXTENSION_MAP, Linker,
+                                           LinkerError)
+    if ver < (0, 58):
+        from numba.cuda.cudadrv.driver import NvrtcProgram
+    else:
+        from numba.cuda.cudadrv.nvrtc import NvrtcProgram
 else:
     # Prevent the definition of PatchedLinker failing if we have no Numba
     # Linker - it won't be used anyway.

From eadefb42b9130f39cbc1f18c64c9fcd32e468657 Mon Sep 17 00:00:00 2001
From: Bradley Dice <bdice@bradleydice.com>
Date: Fri, 20 Oct 2023 16:11:33 -0500
Subject: [PATCH 05/27] Remove extraneous build flags. (#8)

* Remove extraneous build flags.

* Update CMakeLists.txt

Co-authored-by: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>

---------

Co-authored-by: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
---
 CMakeLists.txt | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index e222b100..07d0975c 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -15,10 +15,6 @@ find_package(
 )
 target_link_libraries(_nvjitlinklib PRIVATE CUDA::nvJitLink_static CUDA::nvptxcompiler_static)
 
-set(opt_option -O0)
-if(WIN32)
-  set(opt_option -/Od)
-endif()
-target_compile_options(_nvjitlinklib PRIVATE -Werror -Wall ${opt_option})
+target_compile_options(_nvjitlinklib PRIVATE -Werror -Wall)
 
 install(TARGETS _nvjitlinklib LIBRARY DESTINATION pynvjitlink)

From 65e110951d0bd53a91c341fd637aa01c3f24adba Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Thu, 26 Oct 2023 14:59:24 -0500
Subject: [PATCH 06/27] Enable `copy-pr-bot` (#9)

---
 .github/copy-pr-bot.yaml | 6 ++++++
 1 file changed, 6 insertions(+)
 create mode 100644 .github/copy-pr-bot.yaml

diff --git a/.github/copy-pr-bot.yaml b/.github/copy-pr-bot.yaml
new file mode 100644
index 00000000..07ad9ea5
--- /dev/null
+++ b/.github/copy-pr-bot.yaml
@@ -0,0 +1,6 @@
+# Copyright (c) 2023, NVIDIA CORPORATION.
+
+# Configuration file for `copy-pr-bot` GitHub App
+# https://docs.gha-runners.nvidia.com/apps/copy-pr-bot/
+
+enabled: true

From c5403581cec04250b45633f9bc35b922ff84be6c Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Fri, 17 Nov 2023 08:12:32 -0600
Subject: [PATCH 07/27] Add CI jobs (#10)

* add initial check

* move pr.yaml

* add empty check_style.sh

* basic check_style.sh that does nothing

* add +x

* try adding a wheel-build job

* fix spacing

* add runs-on

* install ctk

* add -q for now

* switch to apt-get

* minor fixes

* update package name

* add install options

* expanding build_wheel.sh

* add tests and logging

* more test build steps

* fix path

* Apply suggestions from code review

Co-authored-by: Bradley Dice <bdice@bradleydice.com>

* fix style

* add wheel-build to needs

* apt-get -> yum due to switching to centos

* fix centos cuda install

* set c++11

* remove testing phase for now, will test on downstream jobs instead later

* try adding a matrix

* Update .github/workflows/pr.yaml

Co-authored-by: Bradley Dice <bdice@bradleydice.com>

* install python via conda

* missed endquote

* try adding test steps in separate jobs

* fix syntax

* switch to * for now

* formatting

* add missing job

* try uploading a directory

* update paths

* drop 3.12 for now

* add matrix to download

* +x to  test_wheel.sh

* expand test_wheel.sh

* try reading artifact name

* a debugging commit

* update wheel install command

* missing ..

* fix more paths

* build in verbose mode

* cmake.verbose=True in pyproject.toml

* require cuda 12.3

* explicitly install gcc

* switch containers, remove extra gcc install

* endquote

* base image name on python version, dont install python in test container

* adjust matrix

* install pytest using pip

* use gpu runners

* install gcc again

* enforce compiler versions in cmake

* try installing system compilers

* set and unset e

* start expanding test matrix

* do not set e for now

* use SCCACHE_S3_NO_CREDENTIALS=1

* use ci-wheel containers to get arm images

* get python from the image instead of conda

* use inputs.node_type

* build on cpu runners

* Update .github/workflows/pr.yaml

* remove python version from path

* separate testing into arm64 and amd64

* add to needs section

* use citestwheel test images

* skip bad test

* Apply suggestions from code review

* Update pynvjitlink/tests/test_pynvjitlink.py

Co-authored-by: Bradley Dice <bdice@bradleydice.com>

* build amd and arm in two separate jobs

* don't test ubuntu 18.04 with arm

* address reviews

* fix cmake

* try removing cache again

* Revert "try removing cache again"

This reverts commit 78b0b4119d7b04c5fb945b620ae323c990a551af.

---------

Co-authored-by: Bradley Dice <bdice@bradleydice.com>
---
 .github/workflows/pr.yaml                 | 120 ++++++++++++++++++++++
 CMakeLists.txt                            |   6 +-
 ci/build_wheel.sh                         |  17 +++
 ci/check_style.sh                         |   9 ++
 ci/test_wheel.sh                          |  18 ++++
 pynvjitlink/tests/test_pynvjitlink.py     |   6 +-
 pynvjitlink/tests/test_pynvjitlink_api.py |   9 ++
 pyproject.toml                            |   1 +
 8 files changed, 183 insertions(+), 3 deletions(-)
 create mode 100644 .github/workflows/pr.yaml
 create mode 100755 ci/build_wheel.sh
 create mode 100755 ci/check_style.sh
 create mode 100755 ci/test_wheel.sh

diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
new file mode 100644
index 00000000..9abbcd4b
--- /dev/null
+++ b/.github/workflows/pr.yaml
@@ -0,0 +1,120 @@
+# Copyright (c) 2023, NVIDIA CORPORATION.
+name: pr
+
+on:
+  push:
+    branches:
+      - "pull-request/[0-9]+"
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  pr-builder:
+    needs:
+      - checks
+      - wheel-build-amd64
+      - wheel-build-arm64
+      - wheel-test-amd64
+      - wheel-test-arm64
+    secrets: inherit
+    uses: rapidsai/shared-action-workflows/.github/workflows/pr-builder.yaml@branch-23.12
+  checks:
+    secrets: inherit
+    uses: rapidsai/shared-action-workflows/.github/workflows/checks.yaml@branch-23.12
+    with:
+      enable_check_generated_files: false
+  wheel-build-amd64:
+    runs-on: linux-amd64-cpu4
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10"]
+    container:
+      image: "rapidsai/ci-wheel:cuda12.0.1-centos7-py${{ matrix.python-version }}"
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
+    - name: Get PR Info
+      id: get-pr-info
+      uses: rapidsai/shared-actions/get-pr-info@main
+    - name: Run build_wheel.sh
+      run: ci/build_wheel.sh
+    - name: Upload Python Wheel
+      uses: actions/upload-artifact@v2
+      with:
+        path: ./wheel-build
+  wheel-build-arm64:
+    runs-on: linux-arm64-cpu4
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10"]
+    container:
+      image: "rapidsai/ci-wheel:cuda12.0.1-rockylinux8-py${{ matrix.python-version }}"
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
+    - name: Get PR Info
+      id: get-pr-info
+      uses: rapidsai/shared-actions/get-pr-info@main
+    - name: Run build_wheel.sh
+      run: ci/build_wheel.sh
+    - name: Upload Python Wheel
+      uses: actions/upload-artifact@v2
+      with:
+        path: ./wheel-build
+  wheel-test-amd64:
+    needs:
+      - wheel-build-amd64
+    runs-on: linux-amd64-gpu-v100-latest-1
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10"]
+        linux-version: ["ubuntu20.04", "ubuntu18.04"]
+    container:
+      image: "rapidsai/citestwheel:cuda12.0.1-${{ matrix.linux-version }}-py${{ matrix.python-version }}"
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
+    - name: Get PR Info
+      id: get-pr-info
+      uses: rapidsai/shared-actions/get-pr-info@main
+    - name: Download Python Wheel
+      uses: actions/download-artifact@v2
+      with:
+        path: ./
+    - name: Run test_wheel.sh
+      run: ci/test_wheel.sh 
+  wheel-test-arm64:
+    needs:
+      - wheel-build-arm64
+    runs-on: linux-arm64-gpu-a100-latest-1
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10"]
+    container:
+      image: "rapidsai/citestwheel:cuda12.0.1-ubuntu20.04-py${{ matrix.python-version }}"
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
+    - name: Get PR Info
+      id: get-pr-info
+      uses: rapidsai/shared-actions/get-pr-info@main
+    - name: Download Python Wheel
+      uses: actions/download-artifact@v2
+      with:
+        path: ./
+    - name: run test_wheel.sh
+      run: ci/test_wheel.sh 
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 07d0975c..18d03db2 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -3,7 +3,7 @@ cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 project(
   pynvjitlink
   VERSION 0.1.0
-  LANGUAGES CXX
+  LANGUAGES CXX CUDA
 )
 
 find_package(Python COMPONENTS Interpreter Development REQUIRED)
@@ -11,10 +11,12 @@ find_package(Python COMPONENTS Interpreter Development REQUIRED)
 Python_add_library(_nvjitlinklib MODULE pynvjitlink/_nvjitlinklib.cpp WITH_SOABI)
 
 find_package(
-  CUDAToolkit 12.0 REQUIRED
+  CUDAToolkit 12.3 REQUIRED
 )
 target_link_libraries(_nvjitlinklib PRIVATE CUDA::nvJitLink_static CUDA::nvptxcompiler_static)
 
 target_compile_options(_nvjitlinklib PRIVATE -Werror -Wall)
 
+target_compile_features(_nvjitlinklib PRIVATE cxx_std_11)
+
 install(TARGETS _nvjitlinklib LIBRARY DESTINATION pynvjitlink)
diff --git a/ci/build_wheel.sh b/ci/build_wheel.sh
new file mode 100755
index 00000000..c15c3131
--- /dev/null
+++ b/ci/build_wheel.sh
@@ -0,0 +1,17 @@
+#!/bin/bash
+# Copyright (c) 2023, NVIDIA CORPORATION
+
+
+rapids-logger "Install CUDA Toolkit"
+yum update
+yum install -y epel-release
+yum install -y nvidia-driver-latest-dkms
+yum install -y cuda-toolkit-12-3
+
+rapids-logger "Check conda environment"
+conda list
+
+rapids-logger "Build wheel"
+export SCCACHE_S3_NO_CREDENTIALS=1
+mkdir -p ./wheel-build
+pip wheel . --wheel-dir=./wheel-build -vvv
diff --git a/ci/check_style.sh b/ci/check_style.sh
new file mode 100755
index 00000000..84822c4b
--- /dev/null
+++ b/ci/check_style.sh
@@ -0,0 +1,9 @@
+#!/bin/bash
+# Copyright (c) 2023, NVIDIA CORPORATION
+
+set -euo pipefail
+
+pip install pre-commit
+
+# Run pre-commit checks
+pre-commit run --all-files --show-diff-on-failure
diff --git a/ci/test_wheel.sh b/ci/test_wheel.sh
new file mode 100755
index 00000000..1b8c530c
--- /dev/null
+++ b/ci/test_wheel.sh
@@ -0,0 +1,18 @@
+#!/bin/bash
+# Copyright (c) 2023, NVIDIA CORPORATION
+
+set -e
+
+rapids-logger "Search artifact directory"
+ls ./artifact
+
+rapids-logger "Install wheel"
+pip install --find-links ./artifact pynvjitlink
+
+rapids-logger "Build Tests"
+cd test_binary_generation && make
+
+rapids-logger "Run Tests"
+cd ..
+pip install pytest
+pytest pynvjitlink/tests
diff --git a/pynvjitlink/tests/test_pynvjitlink.py b/pynvjitlink/tests/test_pynvjitlink.py
index 4fa53f90..f08eb4ab 100644
--- a/pynvjitlink/tests/test_pynvjitlink.py
+++ b/pynvjitlink/tests/test_pynvjitlink.py
@@ -48,6 +48,7 @@ def device_functions_ptx():
 def undefined_extern_cubin():
     return read_test_file('undefined_extern.cubin')
 
+
 @pytest.mark.skip
 def test_create_no_arch_error():
     # nvjitlink expects at least the architecture to be specified.
@@ -63,12 +64,14 @@ def test_invalid_arch_error():
                        match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
         _nvjitlinklib.create('-arch=sm_XX')
 
+
 @pytest.mark.skip
 def test_unrecognized_option_error():
     with pytest.raises(RuntimeError,
                        match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
         _nvjitlinklib.create('-fictitious_option')
 
+
 @pytest.mark.skip
 def test_invalid_option_type_error():
     with pytest.raises(TypeError,
@@ -108,6 +111,7 @@ def test_add_file(input_file, input_type, request):
     _nvjitlinklib.add_data(handle, input_type.value, data, filename)
     _nvjitlinklib.destroy(handle)
 
+
 @pytest.mark.skip
 def test_get_error_log(undefined_extern_cubin):
     handle = _nvjitlinklib.create('-arch=sm_75')
@@ -188,7 +192,7 @@ def test_get_linked_ptx_from_lto(device_functions_ltoir):
     handle = _nvjitlinklib.create('-arch=sm_75', '-lto', '-ptx')
     _nvjitlinklib.add_data(handle, input_type, data, filename)
     _nvjitlinklib.complete(handle)
-    ptx = _nvjitlinklib.get_linked_ptx(handle)
+    _nvjitlinklib.get_linked_ptx(handle)
     _nvjitlinklib.destroy(handle)
 
 
diff --git a/pynvjitlink/tests/test_pynvjitlink_api.py b/pynvjitlink/tests/test_pynvjitlink_api.py
index 8c74d814..ab7e615d 100644
--- a/pynvjitlink/tests/test_pynvjitlink_api.py
+++ b/pynvjitlink/tests/test_pynvjitlink_api.py
@@ -30,6 +30,7 @@ def undefined_extern_cubin():
     with open(fatbin_path, 'rb') as f:
         return f.read()
 
+
 @pytest.mark.skip
 def test_create_no_arch_error():
     # nvlink expects at least the architecture to be specified.
@@ -37,6 +38,7 @@ def test_create_no_arch_error():
                        match='NVJITLINK_ERROR_MISSING_ARCH error'):
         NvJitLinker()
 
+
 @pytest.mark.skip
 def test_invalid_arch_error():
     # sm_XX is not a valid architecture
@@ -44,6 +46,7 @@ def test_invalid_arch_error():
                        match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
         NvJitLinker('-arch=sm_XX')
 
+
 @pytest.mark.skip
 def test_invalid_option_type_error():
     with pytest.raises(TypeError,
@@ -61,6 +64,7 @@ def test_add_cubin(device_functions_cubin):
     name = 'test_device_functions.cubin'
     nvjitlinker.add_cubin(device_functions_cubin, name)
 
+
 @pytest.mark.skip
 def test_add_incompatible_cubin_arch_error(device_functions_cubin):
     nvjitlinker = NvJitLinker('-arch=sm_70')
@@ -81,6 +85,7 @@ def test_add_fatbin_sm70(device_functions_fatbin):
     name = 'test_device_functions.fatbin'
     nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
+
 @pytest.mark.skip
 def test_add_incompatible_fatbin_arch_error(device_functions_fatbin):
     nvjitlinker = NvJitLinker('-arch=sm_80')
@@ -89,6 +94,7 @@ def test_add_incompatible_fatbin_arch_error(device_functions_fatbin):
                        match='NVJITLINK_ERROR_INVALID_INPUT error'):
         nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
+
 @pytest.mark.skip
 def test_add_cubin_with_fatbin_error(device_functions_fatbin):
     nvjitlinker = NvJitLinker('-arch=sm_75')
@@ -98,6 +104,7 @@ def test_add_cubin_with_fatbin_error(device_functions_fatbin):
         nvjitlinker.add_cubin(device_functions_fatbin, name)
 
 
+@pytest.mark.skip
 def test_add_fatbin_with_cubin(device_functions_cubin):
     # Adding a cubin with add_fatbin seems to work - this may be expected
     # behaviour.
@@ -105,6 +112,7 @@ def test_add_fatbin_with_cubin(device_functions_cubin):
     name = 'test_device_functions.cubin'
     nvjitlinker.add_fatbin(device_functions_cubin, name)
 
+
 @pytest.mark.skip
 def test_duplicate_symbols_cubin_and_fatbin(device_functions_cubin,
                                             device_functions_fatbin):
@@ -136,6 +144,7 @@ def test_get_linked_cubin(device_functions_cubin):
     # Just check we got something that looks like an ELF
     assert cubin[:4] == b'\x7fELF'
 
+
 @pytest.mark.skip
 def test_get_error_log(undefined_extern_cubin):
     nvjitlinker = NvJitLinker('-arch=sm_75')
diff --git a/pyproject.toml b/pyproject.toml
index 692b3cdd..f11ded5f 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,5 +1,6 @@
 [tool.scikit-build]
 cmake.minimum-version = "3.26.4"
+cmake.verbose = true
 ninja.make-fallback = true
 build-dir = "build/{wheel_tag}"
 

From e583cc7ce4e0873e9824b03fde5f7c33d1b6718a Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Mon, 20 Nov 2023 10:56:02 -0600
Subject: [PATCH 08/27] Remove compiled files (#13)

* remove compiled files

* update .gitignore
---
 .gitignore                                    |   4 ++
 pynvjitlink/tests/test_device_functions.a     | Bin 5008 -> 0 bytes
 pynvjitlink/tests/test_device_functions.cubin | Bin 2560 -> 0 bytes
 .../tests/test_device_functions.fatbin        | Bin 1936 -> 0 bytes
 pynvjitlink/tests/test_device_functions.ltoir | Bin 8984 -> 0 bytes
 pynvjitlink/tests/test_device_functions.o     | Bin 4440 -> 0 bytes
 pynvjitlink/tests/test_device_functions.ptx   |  57 ------------------
 pynvjitlink/tests/undefined_extern.cubin      | Bin 2880 -> 0 bytes
 8 files changed, 4 insertions(+), 57 deletions(-)
 delete mode 100644 pynvjitlink/tests/test_device_functions.a
 delete mode 100644 pynvjitlink/tests/test_device_functions.cubin
 delete mode 100644 pynvjitlink/tests/test_device_functions.fatbin
 delete mode 100644 pynvjitlink/tests/test_device_functions.ltoir
 delete mode 100644 pynvjitlink/tests/test_device_functions.o
 delete mode 100644 pynvjitlink/tests/test_device_functions.ptx
 delete mode 100644 pynvjitlink/tests/undefined_extern.cubin

diff --git a/.gitignore b/.gitignore
index fb95f3b5..1ee3bad8 100644
--- a/.gitignore
+++ b/.gitignore
@@ -3,3 +3,7 @@ build
 *.so
 *.egg-info
 .*.swp
+
+# compiled test files
+pynvjitlink/tests/test_device_functions.*
+pynvjitlink/tests/undefined_extern.cubin
diff --git a/pynvjitlink/tests/test_device_functions.a b/pynvjitlink/tests/test_device_functions.a
deleted file mode 100644
index ced3797d40efdecbe57ba305ecb2379e6baaf218..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 5008
zcmds5-ESOM6+g3furp1)b`sK%G-LvCl)!pCv%B%etQ*$#I(3!MS_x?bQ8b?2y=#v$
zJ3F13HFm$HR=m(iQPfIBsA?bjfRI3d6d`zNRUZ0OfslCUA5aMpm1xUD)m94U+_~2t
zk7rFp-#FU2=bqpBxOdLE_s-QGo3VrD%2{<%+e$`4jQhz}(@N#Cuvk(7U^jsB#eaLO
zADX3zg^^{qT2|XPr=~1x#dh15EZ9&rJAPVawe6_kcs~kk-)fjfbDC8uR>k6xnmx;$
zE|n`rr2;G6FuKUvb6qctY%iJ(0(*UaEn<OZyB8eqy1mRYZ65YAZR%{*X|kQ7BrN*G
zIFExfcc&N$i*yW_Ocy0lldNuON}@|@)bZEak!2vAu9S+rL&C%vo;bZbsj8F7BBN_k
zOjdhJ)Vf?Q3zH@Fx%ms{a*Bebk4b^EuqjCt;8j{3f@+-C;4mD5Px4>??2qlXo<I5`
zN3-O`D+@3D^WeMn<9jIbPo6%$=Wk@2KUxRw-$&2mEq-|U4+@Nemg6GMO;MJx`?i?m
zTgYo%H4NuDkDD<qAz=5jfmmKZHj2;0M6=uR9hXgXJ$GHxVlBWvRe_K1*H>8(I$cjY
zqZ_Io*}*c4w6J4UPOCaD`hbO@6*^BcZK^y8|Jvt-PQzuI-tKu#?WFG8f!)z`+jW*b
zt*GgZGuk1zbI1x<v}(J?)#9~N+QvrNwuertv!LV+7#g#57_}aUqlGGb^W#wy{nS(H
zQ>q@YWejRgY2!gSQ$AZ&Rn={MAHJdY?t@2*1?}n$$WL6WPQ%gCG>pAtUPFuBPvJm%
zy8_3|LJwdTUZ|?iARI<l0DcY!ul_)p{kCzfP=eR1Dozw#o_S3fnr@VeuP8&el<(q(
zzNH+x*MN~*O1=t5?zyGB<p->#jr%jjv(+~h7}yKXm5Mm2nc{6FKk{qH-&XdI{0es8
zR_@b&2_NMw>v+UEeyOU4Eb?{s#GjPG1F)~~GUWcM)I;X_KK$4$X!tV<n1oLhs6)+T
zdN>Xx-z<C!_8T*xd=9&+@YuL-R@D=^$x`t|E;l%yTcp>C*VDp{d#RiYH@+IP(=mI1
z*x#wipx?|j&WR<pMZ4ok1rN>5eMdWaxzY2Yo@SbQNjE0U9v97Li$&cipAxVPI*1)w
zf4lH`=#k5S%ld7?#PWpw*e6n+ya>wLsB(DE(C|&182Q}KHL)*s2{TgIL6mV@aXL)x
z{K)9+$Z&l`10dZJK1699jr%m$0E`R16Qf)W!bf+qYz8t<5PC;?>#a1w#qIbR#B+%_
z(&90J)4LN%ocaUs6=GZHS(I`Kd|YT>L$1Ky4cj9dJ&Lo5OZd0g*^3^U0C@he{~kVa
zE<(lGU$7<nngFH!hbWIFJg6o_2SklyX|w$v;>Td~@TI&HJ;CvNVgXx;zryi7+wk9U
zobL;+llJd(d}N#b=N!*(!=dj7X)698HG3_)%}l+Cn>qgA*X_v0i@}9r(L(Z8tyZto
zS-1EpiDzPaah0?slk3QR-iw@Q{UTd-Lj0o7*{<8Lo7Wd(yJFe0!pNaeWGk&^U6+S3
zPg_=}+v>T@a$0B@Y+gyWowUH#oM==3C9W)6=R%hh)_TzK>|lMz^W*>SrUOtBD~8TA
zT7_TSiRtk^_>jQy9Hg^w!(OgP!f5r&IQ&Q#wqzr5c{a*8at9nk8zPzyt!8OY7!mCQ
zIR}!5sX-!Kp2;L@)F8?Du>?>S9f`}cfbJPY64$r{&=OqYH6cHrz?1U<JzVcZb_017
z#C%2e0_NI!#MUCvTeQdYMi>I$MW~7Q9^hp`7-0ZAh<d(&9M8cEJl(B9_f{7!&dm${
zaLnsMt_%6Rkk6ycV|uIS><Ab6CAQpQ9_}OB+8bLBbi7}2T=%*W)A3t3ffHz6$LJ(4
zdp+F`x;_gcoB$8VR;)H&oS4YOmtz{bQw`H!gc@Q*S7Wx(%DPE_&TCpTH3<m5C>Dqz
zWIDfT9Jy}fd459R$u&eSKcgmtC`OVBC{WD*)!#&4ib?A%NpA?jNGas}rR_X+vhw$5
z;U~YN(mtKP8Y;5#_q-TrO*m4%s77*pnZGND(f&@q|EFSnX^$qHq%8R*M6<^KR2ax{
zh!Z6JBu@J@D}QoA>3XGQw~xOfUMOv!mZLuYxx4hIyBx(#l^F)^(*MOi{)2bv|56|S
zU3cmKQ{gYnQHp6;0J2USxa<i2GV-kYkK#WZMADx)t=)9}e}Pz5{xmUu^1g_si5neB
z0KdnF{N*`87|{d5pZ0f>Wc?+72bHp)v~$u45z4eJ-9PC&lCHlm{59bxnS6gK{{uS|
zGwpv<_}`R<f~WmIKs+n|Cq@3(gdgSq06x-R)+K9fa?vO#(x{l&*RW5w;Iuytla(q%
Zd=QSLUM!uzu|EE9i245;^+`(m{|6~J&^Q19

diff --git a/pynvjitlink/tests/test_device_functions.cubin b/pynvjitlink/tests/test_device_functions.cubin
deleted file mode 100644
index 0b6eae941bed2df852734403b41b3e770f71e2c6..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 2560
zcmcIlO=}ZT6uoI)>=di1ACxW&-BggVA5y!s3qLAtL8uFHGfbMK5z;nHMomC)ihn>v
zp-caOJC`C|xDi}*DXxmR6e;c$HQqCG-*ob3+CtO|FZZ1@=ic|;eKR+Yt}b5d%jKr~
zDY6{Bp_a^(E;tsUg~9^O(JX^GI!QSqyIE`cp6`};rhR)Y!l8GXwe@m~?DhNh-Sz4Q
z*?y(v+s>WIX}{9+ot3qQQ*BJn%s5WXtyh<wSb3<XJBeE^ADJXWQ?8WmtvXfDU8|7o
zRqD1$xZMmNVa2W2S3S2;3%A+uHvA30-Kg|Xhr=q+eY)OC6z9)4cz6B`fn3)96{Iy|
zcS}>HDH_8s`|e8pWigX)UiW;7VV(xNNk%V2{zQt;@hh4qmEtEu-b(S8Lw+E|Pcd^k
zv2GxWry;+;uI4*z=S71#jQ8>ISrCj5Zt@RVoy}m}!qft3JqJn9={O0Goz5MeCQ71)
z5^Lx<<beH;^>pRc1&h!#MtG>PXpXI@Q;9ii&^EM2T1cpg=Q9djgw!P!_8nzl<iuIv
ze~bSm{wI6n`Nji#`Of1t4_F@$aEkzJz|dX+U<1zccniRGgNN>cz_0Mbrge=7(k=1O
z>oW+|lN)X+HS~bbi0IEn8g3N|F@9g9iyo;0(I|b`%hOJg`dg}ldWH6>NH^PbHUhhO
z`dXyxZK6G{OQ-WVPuoQrVH{fqK5G35PetqFQ~3w$nfr<N*5d@Vz&)BaCxjIjD<TX?
z$C=KGpBLeh5nw<%!|`}WOxqH2TOI)7ebF5U-vkWrtC<u0ZQ0Z=K*CJzx#*dGJ!9pp
zw7hv?^cjVYb;}3D2O`w{1@ZXy&G@YY;-6&1qf46cUzp=FrS|_qgriaZ_>OeEuHz4W
z&b3X&Z)dEJRA&4;<_7S<upJRj2#@oF((CHFd=y?}`;I?aShD`fczyr)U%{Q4y(EzM
cfS<vzK5Iq-fP``D$c)zs1JNR$`&%mh51=h@mH+?%

diff --git a/pynvjitlink/tests/test_device_functions.fatbin b/pynvjitlink/tests/test_device_functions.fatbin
deleted file mode 100644
index d61b9e9f32b3b8e6b81dbf04d7c2d40bda88fba0..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 1936
zcmeH|O=ufO6vzLwJ1ec_m3F1IBN1p;wiU+~?#hybE7A{dp;laMgCPkqB_O0&iUa;o
zWhGfbp=;_x3x$T#KyGn+D7_SNXfLHLD)bTvtqVDn9D*r6w!v-4A*HGM#&%+*F*Iqf
z9hjN_d;8|`W@mo8Q@@<}g2902$5xZTm_y2yx9#&gvmXrO+peuK9B(m%3~bvUy*WCO
zlNcKk3l<o@hD)N$^Z9W%gmGmY4zd*LAVhxHOS4P0YOPR&t>RF9Ud7$6{L<{gj0=0=
zynS||ya>Bia%*<}we(P}v{cKVp0DJ~mGtm%K0jNSD<8`zp(hXsjAEoPGc$@1S-8(;
zO2u<$^5tq_z685knhRrpA%brx>FL7U+?i^jG8@8`N_DZeSgTh`DvRs45Y@YY;4Op_
z2;GDJ5_|r`p0mYFF@v6BMuk4fj_3^jcPD8FGBnyj_|q5$f+XUMk>bLYCZP&3LZXet
z$qdCuTy%TF`+f}11Y;;SsSQvP@@*<I_!}R1UZ1v11JvfS*Qd6_P&*M<FLk0V3yp!F
z0{afqC^FojReA*{xWpWuQgN6YF%6NPt#pTYSi@2B`VxpzS2V<jd5$lFKgZ!30AB-C
zn152gufq9*N*P1?H=u^%Ju3D~N^eRX=$)XufdTZsLUU2vg>S`^R|mlvlAoau-4$~5
z9>*Y3K2oXWeJ<z51>gsq+t9JkRTKkv0@rmMu7WE-96xT#SkvKmd6UtS*MHJ6RY!-0
z2G(WV)^W5BY^K@A&9`JUbj0curBfZchhS!~$~{<W1w)zvI*n&dw#w+`bUPIFqROZe
z8mx8)+#KtNHR(Vz&4ffD=H*T#8}KAv<>-)2T$OB5fjNB%sdad05BK|^T#*zyrtIf(
z8W{7zw}GTMSL>Ka>GdOOIIuJaur8Cwx+qkThRDx1K(o2pkKQ=l@F;aY#v;F;V&t|O
z9_@gN9JvPERxkvtQQ)8m<XP}(L~kBP8^Zy`V-|-5GjoG!6FliDD@sI8;7Cl|AL{~T
zwc&HH-|t`w+wB3jZ}0HB(!y<SCq_)+c#9$A6LFhUa#ZI3-2QuR4gS3Vw~HGQ!R^VX
z;`U|1t;<Jvn#T~}>yNQJy~Xaze`MF{e&^0t_?0#3+YEwomt<qZz`>^ld2?@&Qw_rS
zz%Ir;4>7*@&lq>KU|e~K@#>x!zb7y@vne_JcZic7|3v(PS|BEpABFhpo)DXatVQ^1
cyAZEG4l#XyY=bCP3GLXl&tfldJQ2mfUxCk3y#N3J

diff --git a/pynvjitlink/tests/test_device_functions.ltoir b/pynvjitlink/tests/test_device_functions.ltoir
deleted file mode 100644
index c4838663e77f46f3d4bc1d8a5e3db5b8bcfc3959..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 8984
zcmbt(2{@GN`~Qr+>`U3BkbU1p_I=+I8Dosy*v49vghXUliY#%gA!~NBlq5n5p;QQ2
zvqiu69W&?jb)0jZ>;J#6>wWI~{yg{Re)ebH>z%9WdK!38C<G@^$a%<L9u9>-F0lM%
zD`89}2s4BUvPOAor^C@vh?4CF20h9*x}X)lz|gJ6Nd{()xxz^{3&v8isX;LOf2hag
zGg>r%3WZQXMrkoFgaiVG1_GuX)`tOF3GiisUh5u&U`lHC-yFcy!^Ud?HV^O!I--LW
zMA!wre2@s3gNwg{DAxA%_CbS=qX*mp&_F0gR%sVi4X#3E@x<|=#1ICcH86_+@IWC;
zcz6&z2m-K{fBju}Cf{z;Z-$mr)~Ua=UOm&m%)_6^{@^3k=NnGMjVbdx-W3IeIaJr5
zmF=_=5=`?K&Q|ulTc*~3z)sT^r(3Nj)~-ta$wIw#`^`h>>*LW;2qU?H6rv>64PKdS
zW)otsNBa~b!vZcbl;6He{1Kb>>lOJ{*;X-gbKEBb=6pIX!8+TGJvJ`FLZ57;uU`hg
zjyg%qsw=6mALWAa`vLqpRD4j8-@_-`bI&Z=Odsg2<QbA}IP-e7%4L1bH*Pd5@cn94
z<g{YWs%R9fSzsRc*}uzt{OT_IGnpaTuJX0rrjM~7Sx~G2tMkjJ2ddwia^N*Ct-<NP
z3$oqfylWYKdiHG9XFdV0dh$A*RTtSU1~op$)Hed8Mh(<vX}vDTE+UL;{%BqzKWQ7O
zYeQ1klQBJgzWz~uyIa8y5$xuRW4@R1?kFT1$Qd#Ua>sZ)R}onoLs%IouRi)=gV^G`
z_13E!N|r7Y)Z7_t{=~o1{0j|>il1POy1KZfkvh%f{y@h{K7p%YZ;Tu5F4N8Tan|ZO
z)r6l#;C}X3LW=KZPGtPSl?<!&iQkbGwqq=*7kFt#TST}N5X4;SEm9_}Af3yIA6%o>
zxc=^aU$Oxe&jy~eYIX~CPE)Q)#F~?RkMk&F*Z4|s+e~Qk9e(Ob#viE<ayy1Hr?|Mr
z*Qg!}bvTiZk(v~mj~}lcdv!yfp1Z?SLXnO4<>KN{>4=4fT}h2W0_-_k`a+^a=xvT>
z^Ir+mOljKIas$~Ox1P!o$2gW6<Qiw5Z{EGT<a``ItC%1!`(@gl7skD(XqnSb7T7oY
z6!R81i6b`Z-vzN>HxioVw0Ka}>rmRzG5jdrM!O&xN;th}bm@xzDWRKE)q)j0hD>#H
zmOmH8?T;f#ryG)~lZr(AOXIk{o1jyiJX3^g^!mRN%lF=KPPc&ua`q_vWaJ9`KIon-
zG_~BQsBPn+xxR&mxTHqSKTTuT=#U#%AKuN>*HZfF%+16LBk*`ixqbd!k>Hv-sa8p*
z0Ew=jq>rfr&OixgMv%dYcxn`y=~`vKSn5<}1aC1<)!dAh4#1ZyDp`%4^k2OcQQbh-
zF+5+BvUe3qWWzH|rwXO0O3G2PewZIqm$xrQ9vJzVlK7a9)zc9f$)R#$@|P@JTB#nQ
zDXn(Mxmj-g?w6}H^N&ehdAOw4+k4o)O<R`c7`io*)4$(sAsT3*vHab-r@+`{-ORSI
zJCte95Y{2ZeYJm~^a=hA#prgm%NeX^8oFYlx5VliG<_I&$V5MIw3}wXTMO3ozPcLp
zbei#;CE2sGJwm}+Z?fk|)y2=AF^PNR6)Hv(=+bQx`|IPWgK+C}tUEo4WxoifB&$5}
z>VJ4m@fNn8xo5hP#!EYVF<`}9!}X7-*yCHyzF}{RdZAlvo&3RlBvPA5;byC^-NI4$
zI}vV6Ps+sRhT-(YMx2l+#Yu?bRIh{-T(PxI_yURKlb{`F_4B=G>)3amT;Fes1k_6@
zL7XMO+{bt5i--=^MV^6-#T!RvT-W<?&#QicAd)U9X<8~_m^p}Zi1}6TidQ15{@a~X
z9Ks(5QqJ4ST_I^RHet(AbX+K#6|^y%Q@ukGA4q8;T}j;I8_M_5oBrx9HB(ebn#Zdh
zzZa@CmU`v})poK$vGPzY^pDSt{b(Ty^Zt+)yoi}dM`Nqg(5w$4>svvy0{N6?HQUMC
z1kZj&O2y#uY;(t36z>Ug)JXd+)`hO@8Oz7pI~Cj`k&%88wfNOkj;3eD$pO;fkstiB
zkTaRG?dCkkJ9SaHhR$k6`4)ChUCI|6tH#-mc-11I)dBcb{?d1z-`wYGUE*$_3~1t-
z`{9t4*REJ>MY+A+{`G9acHGBAl6JzdU4hE0^k1}|D9Z?HJR!_0ZQ2-G7t-?ITMZ4I
z<<C2X>LrU&NAD0HKZX|hrcv<fCq<r0?-R3UL|ez{b{a)EI=#OOyAF9f%dc)<f3)Tr
z<f1f_Ccdm-usCj;MM)aSwIs}rx8NErrx{B<D9cgPywkYz=xhmH{M7E1`+LVjmp`OU
z#3_f>5nK7wrNl<{H@_Ne%T~KrV`+D+rRhqd)&9&g{<kx^<UWB}#8WpXY&pGS`m#RA
zy!vi&gDy$k^_46Ba`3OxcA^yOd@l1C5qA^UHcpiybHicx4_~Tch$>-^Tr4kYdo$Zc
zCBS1%E32agSO$nUGJcG(T+59S<FsUft2#o>D~bsb%M}x^EvF_=bc(+*jNV(><$$e{
zkJqsK>&5x*j;Q8SUM!Q4adwfo7&6gQ)ZoqSjn;+J^L(7>dGjQbuFI~|bE`M6O4*va
zpW|B%v^e9N(>a&$UkZMDqKaN5{cg$Kfy4?hjkdTDoBJ0^=WAMr=$%H3e<ql&Sa2DC
zr&e)R?9Y23JlU?;=Sg`hC_FMLbHpT)6W({b&+gopcTbD_E_W$TOlQ*TeI}E=6!yZu
zRTFkE0e-T(>kiL`$H<Mqr;hdXo04KCb&OxitWKb9cn4hM5Z}curL6qEqW&O&$eY;-
z=2m&WwsN87;^0R8@u(@w9-sU6a0|_65{ngHZsbac`eV|NLfHgLha%F)gSR*%YJ<Yh
z*}QEpj2P9I&~~SpH$4>{);-`E#qexz{c~AbdzHj8o$lt=vcs9@eA{blUeOI{Q6u3G
zV_S~Jg-YIg{|Mb4A*~o{6){d7npICy@}OF~`7ByDgSS;*Z_Fh!^R{RrD|LSd+aq7A
zicIKjGK=!}4}Z{(`ifr@OD*Bj_T!D4i|Kj0?j!Y_mRRtPA;%kXE`>g}hPj?mSP+x{
zimfYrZC=!3go;s*u*mDAvj*SFrXzwn!)s`_HR__Sy4e$n4~=LEqRz<yqxVTwJHjqH
zck|*H2y{=DrMj>0jT=h2-A~6K;@FfN_2}KX%F{Vay?nD<R<SxP|GqTkJLNXo?OpQ-
zYu}XHbs|5klM_0=F`8z&g<nRHKiDj)Q{f&jKe2t&k4vLsa<6zojmg86LHi<c|MPCW
zmkEjIUV66GHTHUV#f13Hruxf_O{uQ3%-Mf;tUT%e#cg$-#BWmfc|{H4>vsIFl9b5p
z2AUzK*Yl<g6M}SALd&h@il@GiT`9uTb-$)C^$Gn>AZ6p(6%95{qRdi0eNpYOLJprW
z3ddaMbDk}~gqx{(uOn9DIe8mamDNl=XOwx*Y5e?l-Dks4>;2n6kGnj#h_km@)-u(v
zWU956b9RUrrWvmfE{WOD<e5Y0gu?f~e1&MfoiE;&4jM6M5o~@$yzhTPwj15y&s}?-
zRim9ihx^sOu;!wR@RiXlm1C%xNagkW%pF5=7r2*>Wyi)d<vw-|7g2<rsx{+liq>@W
z9DAcx?cf*67@Vq-N;fZh?dNLY3;yfL;}VN}^glTwJZ7lV_pEgMciR{P!@1{+`}VV8
z>d%v3s*5026}(=3$t*72;()Zj7;~Z;3N<OXXM7jstSZ=@v&$M7RctIkqHu|*Q0rPT
z^VQ^QstUok!_U`!W!@YpNg_#6dTU3vu>U&ttAVkU4sRSs?oZW@dqJAt7G(ls)1g}8
z3k-}D=`l6b89SHc+4fu_s8U7LQD?Vowfz06f9PCEgFtS&@}EbV29K`ZtF0t!JpHSP
zBcgX|en38?jIMUjWpS#ySJ5gpjH~9<ezP};Ud#{K*I(-1<QY7;*VRCE_0#FE<t$Lm
zrk(|sqC1uokE6_gv279sK<r$7DE9d7*T%l^$}B9jEXw$Spi79jns+~ICq6Eq)KEm{
zk_C)9Rd@Db-Qxsr2d3JN>q=P<gPJPl)1)X>8aF%%Pt>XDPam5gZN71u?$-#PD{+jM
zU{W&O^-JNIjV;a)rkGLDDw*mBF6%jWc#V$jO@E+&*tpR>7!kyf>I@%wBbC%jMzr^M
zpGEwZ)rYe9d`5Trz-IEvyWC{OI<Q+_hInGBta_$+)BU;TL%PXX!w(5!sg3`TOi1bL
z9i+o=qvYcuy|v7TFW&EH5;Fs(;Np_&Qh3nbarPBiwovtg$h=nR#Z4N-Rz<Gc++#<1
zJw4}pCOzTzX`3|oA?Z%@TcQ3X-s}v};{=y!K3Uzp<GL?(k?W*|M$Fex4x*+;=1uOT
zQDR9o+1d+c?Y8O4-kP6$_ld0wSy4m#R2yf?xk7ES*d6>>@Zzr#3NG{*!q9ZFA)XxT
zK9drekvgL=gP>ve95*Dy__R|QM_W5;2$A=iTd951Zkmn_rtf;bRLf7MygLl(^w26&
zAkK91(}N56_b(1Sev|TKAVI}(Mo@4%rI*NZGfKKxrQufY?lEs>XS;?F;d?Wg3`5`b
z;bYCzogaU-8_62Ky&DA&ZW-R|)e4UCHgca_;-bJGz1~Dz8L{sWvE`_F^;<BSo=16*
zV$?1+o>%PrvINnDZs6c)ORY~E-%7-K6s~dKh}Zq2jyL}G*rLQwe^k1`XNq%43QrmH
z-o5G&C+uPuXT^_Q4OAOc%gYoxbtC_F(nB%>Xh$18;m;QgL?dLRPZ#&kmv5D&if7V6
zk}Pi%y?4%hpSE~h_}1u<=vZ<m@?%f>nP|p28+_Ny(>Z)w0vF$;yy$onTKeIpa6A#O
z^l(aVSc~KNa{JCP=-at@<K?SRZ8JU-@?jx6&o|sb0(d+_Pno=0e+{FFPbemE%6&{K
zPv}N{sR`gYeSZrUSg|S5&>O?_BV>1N<dU?Cc)+9eGmE~I>pgMFmpf=Et7%Sc<nN?B
zc9^}}T3)sl??B)t#V%97tY*z<2`_4*%PHZZDJU0vk$k+c*6Y;sqzV0Np0&)*WQ-;B
zRd5xdL}os&zIA()(7Xtq9+P(Bx+ZO8Qt)J?W{2hX(>s@aqsP^s*uS}_rjvI@MVdl#
zy?sV|>5?*S{qyjpvTWE7rMVEi${)_Q;jyZre4mvU2IBOUP2AtGedx@I77bW-6sMJ)
z(J*Zjp}#4vn|)H}k5l_`oBX*nO}<GXwNuuYVJ}}W;g9m!i<vs{)f!%GA4y*`uPV3l
z(hrCn49^(ht))!L)A3eSe?)DSBYBy8S(V?#;Ht~hvq2<=W~~xds#14i4L+01t`TVy
z#arCJdBuCKtGqYx;q12O?XZCmINgw!z(QV@TVYb~>&}?A5RDDW0$w^tnT_#p-j&|A
z^+mz3E5_XOi%F7o=x-Io@?Ggr<+?Fl3$l+@ulF8X)P%-p!ev>Op8Kk;GRkiB$<;YM
zJK5y1<dDpsQ?XCYccx=OR;gHA@p3~fGm5e3W4Kzp*5!0&eS(nwd@2X`td;vr8-5p|
z4HWe{cr2qov%|FEF<eaoGV~gP`l0L}zr+Qf!w>FLl-}d(>m{zKV&9qJVkr6PtlIVT
z^>G$C)ra0@+%UKGs<}35ZRb8}zGeGyw~R_p(V`2989CVxPALyePTPgYq{kpGDUZm9
zweyxO=bIIGxmq^1Z=HXovT>a24eZ3A<!n$_K=sT#bl>LUuua<j9hJICcy^YN{F#VZ
zH~Qg1*3olq)Vd`!*GQvlR7YEMPnY+&@YjZ2Rg1V?U2m>HE&PaTMIie_3yD1hYMWM^
zQ(wQ7RfF<oCALm$IIEIhW22jlytEd|ICcJKWzMGl<qE;OiT(7mG9UIW$jl^ol1fJ?
z5_k)4n4Zn%9&@TrY0w@z%TY3$oNm7C-<{L)iI9L_MEyFQ+S_a;Wj58QIsd`eH;Ves
z$`zALW-=@sv)1{-T_c2gx&QUVq)oO>+J#&yYktGwy$m%XFWc`amqtcQ(-P>%qCzMx
zO-!G?e9W`ehV%042Sm?ptq9qU^)na~ib*UF&CpWj4eu#NQ`7Dv$b$n-g8QaUZqIz0
zi(@3-S-yGp8cnc3v9L77#BJINy2(>r?5jT!foIM&o2KmAcyxT>WLbW2V}dv<e$>pk
zWoV&*^{W8c2Y7|oLY#?<_@L*=%2aR?+308MYIXUdcceE$gkQEN2J2<e*eaecYnQFu
zh4wlcNK%EZ_a@~=oNP+Qlgnlsk`${yzUNjeO5M5fD3mbXV^xI?;cb{NnEPRDn82e#
z`Eox(I<hhNZOe#0q3ZCa+uUoFy7SA|(#*@WS{Jr*h?$OYcNCrzzW1({e1-roBOXzU
zR@8%N@Lo|X=#9v%i#|ajlgtLZZ!a^blq{;!K0(b>kbPGw_A7l6m$xl%#xJs)8}@jw
zm+Dli!b_cHEk2D_D4)WtW?0qga2SdsrgNRXQ0KZzmt@vsmSAH1#8vHkmBs!`_zdns
z<7B5DRMU`5YLj6#?3Xef^#k<n9OKLGZ)+KyX(9a{C_OxG8`@ffxNzay6M3Rq&$QJ5
zIi>cUZq|~()9{vZr@Q1&ptpkL7X=G$^ORp0XR&3}s&(RvWbGdv{eU+{L^ERrx80zt
zQ~k2{RJ@pcCL7f*7#Q2*r~Ui{#eS*{PtLqm(Ra!wwr_g%^|QomdZwhyZ4L>+x0^`V
zKMnn`QDc_osWRcNp&p;%?sQs{lKJ$~jqJT}1et)lNL%LWC}Y5u2P%5j7@<{pm%HIn
zsXLt3Gx65rlRxh!PG6UhN0rlkTP^Gey<jGBDKNG_I3O&_U9te(UCfcKrFZa=4Jg6y
zhg!GEL0{|ki325hpcFtg2Np+2g{r;AQQs|L_M##PS){ZyOd5tp`l4Y7KbWYjxP-Kb
zv@`^wsj4c+&SP%xheG?Yi;4<~3yBDd`e9trP%$wf5eZ%_jBAnrFsOis5O_2V%z^KI
zkX8`G2Od!94ip3$YyyD>QbC!?h)M1M9RoSIPpBXl6c7`^SO@t)lL2^u&!QM%21|7c
zDrE{1H465FIy(>p<ZvC>?iEZ81j2>Y|E9pXB7m&?Iw}Sq_>y^);A2Mpzcqpw7dN86
z>SO>D@3%a-2n#F@jvW*n4%+w-B2fL0gaRJkZ#Zsj4*(nrq4_hG^-l}T^B0bTD*=NB
zV6bBW%=`g%AMoJ%f~~@+L7)WVm9QW#ei87ge`A2_1<Fb68XtInmHGgF8gl$k05`MV
z7@YKPcnSt5`x{=1!7=9oQ-{m*4ueztEq)h+Q~nKypgp_-9nmlmKqRFc5f}q>1f?+N
zLkQvL0Cz-+3L(6_AdYafy$ebWj)p^E7xl!%V1S_fVGe#C9zifKxDVU|=7V%J@&{|h
zMKN_CT^)sXK?j*2om_m;NFP<WySqIc;bwFYCnf=d`J!FGRA4MFO4;2VlSWh$2J`TA
z@N-APTpWNvA0W4=IOgQS9UNeeUZPS`Fqkvk-O&^VK?b^@|EfoLcpa*T!45_c-~;#i
z*NfnP`=aN9azi?(xS-%ZLH|l~_Vh)YARSdbQNCz63a#wp0}oOU1jdYlyMsyKPDsd6
zG1{mjW7LrTE(qkmW%(`QA9CmdjU(OvEyr&WhjReC1-KM30^VuhF8qo8L&Nd_2JVdi
z#fh;va1Vla;h*ui9SLIK?#Jc9T<O?YFb0R??nYc5+zB`h#DD^}3+`rIJcxk;o&#JT
zK%WFG00D8_oecWyU;zj&|A7aANMH>dj=KxA4&m%r-{BChg!QlehX39lh>&lP2O4e<
zXtd9PcE(vgNO!mp8X1U&2swafOvv8X7lJuOK+VC<g$Vr_1SEj_p#8kCkPFHM4TFO_
z4kCo|*D_I6$7<~Zt%mj0u)Z$V*9Ck`PL#g}91U!8Q>2pz5(S(i@M!;e^dLeAPY(<(
zgz`iqg@E5WL7)Q&Hy}IcJE8o9ynH;pkUnUj1DJ$JXP6^!B_Odg@&BjT0Xn#bhkp^2
z01OoC1NCQ#TQ?j4-q+xsQNjUOO~rIzh+uv2{s!B@?Hk;Eein=WzK1|c23Q;+V2<BK
z0C2#Z|F3*efD7gX_vP>M5bMlh|IftL;rfsB)dAzE`MZOCLBaiw;^Gg_pAt}U)ch%7
z=PwXT1m_D}BV7Bq`MUsM;Q2m0{yJ>?xOgB0BRC(YO#$?%_Uo|i<Jti=$ieY&ICxHv
zn!i?T`-j)-u<!m4{I>tdNB;xA{Xg=-ce$hb-|-*$H~xYD>_75j|AF86ANhB&{KM<t
zgXQDa2?)aoZvAk6GGH9F{;!W%e{k>O__(}90Cv>;O<~*r{e2N^^YH$!$MSLa1Q-L#
z9+nTD@81Qte{i}TsKnX7$MNER0DlLMhy~w2!QYX?=U)WNX9wzj7u@)9*gRkyHGU&3
z{|+t?s}J)(1Nc$%KYPUaKMBO*__+B$s%@~TDQx>x*nu$u82A=^n12kw4(}ZRf-KX6
dB?Q8HXkre}H8l`(6yFZ}BEOEw@>g+~|9?KGVe9|^

diff --git a/pynvjitlink/tests/test_device_functions.o b/pynvjitlink/tests/test_device_functions.o
deleted file mode 100644
index ea6efe1116abddb03eb5b9b5b3a646970c906899..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 4440
zcmbtW-ES0C6hE^qwk*(X0Rj1tK`8;<4m-Q0OI^UN-9ibd4akQOGR*GYEfZ#^!^~`L
z`A#6dC<!5uXbh1DAB>4cO^DGK2tN2^G$uay2bidlM8F3ECV0-Bd)w`>ZHS)i-gD3I
zeB7^d&)hQuJGaIZ1tcl31tukl0=yTW?B{tm56fUF+)F(B;M+<?ORPN4(MIyn{=sv9
zF8QvodM-tN=j7_SzmaWXWdXSVG<qIy==Pp(6le!E#zmZ)q8!HVO)<$E$n#v)3OhIt
zn+~;3!0v;lP##3qj{mN%(n!^{Y}Pg6*yE}iY5^9+6}Y=x8)cqvjX3H?Efv=S(;H@i
z>Q{~IhPY;y%O3N6!?%tzwWoVM{I$UHt)k6Ttx|JJ>RQb;J+rE6rfm&7YFgEb8`Y)o
zXsO|`VAQlz2h)evsm1|46EHt8Fq~0w^`0KXIAYqB-NxE@A1H|wv~(C6v>W?iWikiP
z-)+~?FTO5*BCdIC7=!9PYNzLK>fVxz$K!VS4R}tgEr!?AN%i0bNOT>_^}@<bFLd0}
z523~E6F87o$-*i<Sp(>UbGi5_g#BPYz{jxU;G0U{i>X7&41Ac2<3!Qr)K#UWx0p#^
zQCco3FPEzBC1vTn6trDZ5;<5g=aO>6^;lW$bT_59<UUfMX+E6Eq;XQ4($|$l+vkwD
zt}JZ(3}#<f7OS7aotR;)4p`OA<l;UHT#X(3L1|tD3zC;1_Onv(neDppww_e+mrCL$
z+*6<cd539!CuCeb`2s9VZ35*X%*w&uPFK&x*TmLm(raR|=FZp<Eho&U!3*=~I0r90
z9kLA}+f3|tY8ucN{l%>!rKV_&qMj=CvTWAK8vMLUHAC;sbZ1l9EWoz@{+HCXdx|wD
zsHwWHWwcb6UgM&EI-S;1-RlG_gEnJ_@^2RYx4pIpFs*MACejlYVxLGkz6+GGc4gVz
zmez|nG4ff?HK8wc2}?z=B`D*y;<Sv~fwuO(w$?(M3P8HU_$P|$Xxx)r1JEh>OpNxb
z8SXr4WYg5Jg3uai)l;d9i;v?Q5s%fykrsOePHQKUIQ5(0DPmLTG)l2Lyi;heB3EGk
zgzbun9>tl&CHy6J=A(xy0KR|N{~rHxE<(lGPuP-uUVzg6Hp(4!9&{!|i^Lg+(q#K8
z;;Ue?_|m=;9pm_%P{3B=S2#ZRG5jlz^LxQ{(*75Yw>@V6kmHHRa0sgISS2u0NE@<x
zg?C8k>f!f8E7i(oh3Q%eFLT8Vik4F_0~3q`JJV?c$r&}uwQ6<T;8hZ9Vr6KQv>Bc2
z$bG;GtYCZ>8@Bv_dHtqs7tPYqq0lbfZ5V!FQ7E#NR(-b3!{{3fqdHQq+03xYXy{?=
z=_)O-F)NtVV}+YqiNP?!6};t{?u><C>RfhO&QVr=#d1t<d`7n;BYv=pRr*I9KQNu3
z&-2XjfiYaOW7-tN9A?nSX1mkSrocw65}PSXoy9aUcHrR5o+(D1MKlIx3l>FARMRL6
z|9KSBz7WQmF%@qSIMzX`g%kENBMGDImvMMW7q(<0aakK>9C-s)p$!p@OW7>#2_vF=
zAm>1GD>X=j%bHBGN)3{XA4&k-q9bux3+SCeByp8X0A;}?o)_}jI=p^9p!ws~z$_vU
zypSJ}J&)O@7O=4Zv@+c>t?2uhvv|W$6J{>pV?h{U0MiR<u7E7Z!p8??9B9ty;I94w
z!MBIJAmoCOcL;e0$~>ksy44IY$#=8iDs%82(ZCZ|4>WvVaa?mo0;b`&t_vqnI*QRr
z9(HP)>y5a~3vdEFAUk4I@ZrS7PW(TnqC1^o^ox*3jObv<CR%xJ5}<lbIg^)w;6oxo
zQbMNsP2<SCk@b9y!0U5}Tt1+t0a1*4N}@n9|Es@_z7&)4te&n3Ar(=``AgdY>@?cH
zABCU%j!OIJ{^e28X#dWNfyRU*?H8Sq9AEC=e#Gd0N7sK}j4$ocgp-sd-;HRa@$U-*
zISz4xq@Tp;o;KP)IicuzMP+*$e?=@PY9Ez@Y5ZeP=udAsiWxa(XnI2b^V9e@KcWA-
z)A-MNLjU)Lzc5EBq*ejQbDF?qNASzY8=Zgq?9f2^6Q|sbp8qF^HQGN_j9>p=q=?b{
z|62IVIzcu>n}t8!?|PEwFZnH0%6`3_72Ob_jM~!slfEO-^LK^6D*Pmq>zDE$*rAwF
z|BJ%^qBImd>i;X^jrM=7*#9x%NBh4B|I%NcOQW&LMWvuf?P6ljV4q&WQGXhy(N?9z
Y2Vq5IVbT5Tn8yE_nE&6XucxU0ze}nyz5oCK

diff --git a/pynvjitlink/tests/test_device_functions.ptx b/pynvjitlink/tests/test_device_functions.ptx
deleted file mode 100644
index 08d3e3f6..00000000
--- a/pynvjitlink/tests/test_device_functions.ptx
+++ /dev/null
@@ -1,57 +0,0 @@
-//
-// Generated by NVIDIA NVVM Compiler
-//
-// Compiler Build ID: CL-31968024
-// Cuda compilation tools, release 12.0, V12.0.76
-// Based on NVVM 7.0.1
-//
-
-.version 8.0
-.target sm_75
-.address_size 64
-
-	// .globl	_Z13test_add_fp166__halfS_
-
-.visible .func  (.param .align 2 .b8 func_retval0[2]) _Z13test_add_fp166__halfS_(
-	.param .align 2 .b8 _Z13test_add_fp166__halfS__param_0[2],
-	.param .align 2 .b8 _Z13test_add_fp166__halfS__param_1[2]
-)
-{
-	.reg .b16 	%rs<4>;
-
-
-	ld.param.u16 	%rs2, [_Z13test_add_fp166__halfS__param_0];
-	ld.param.u16 	%rs3, [_Z13test_add_fp166__halfS__param_1];
-	// begin inline asm
-	{add.f16 %rs1,%rs2,%rs3;
-}
-	// end inline asm
-	st.param.b16 	[func_retval0+0], %rs1;
-	ret;
-
-}
-	// .globl	_Z13test_cmp_fp166__halfS_
-.visible .func  (.param .b32 func_retval0) _Z13test_cmp_fp166__halfS_(
-	.param .align 2 .b8 _Z13test_cmp_fp166__halfS__param_0[2],
-	.param .align 2 .b8 _Z13test_cmp_fp166__halfS__param_1[2]
-)
-{
-	.reg .pred 	%p<2>;
-	.reg .b16 	%rs<4>;
-	.reg .b32 	%r<2>;
-
-
-	ld.param.u16 	%rs2, [_Z13test_cmp_fp166__halfS__param_0];
-	ld.param.u16 	%rs3, [_Z13test_cmp_fp166__halfS__param_1];
-	// begin inline asm
-	{ .reg .pred __$temp3;
-  setp.eq.f16  __$temp3, %rs2, %rs3;
-  selp.u16 %rs1, 1, 0, __$temp3;}
-	// end inline asm
-	setp.ne.s16 	%p1, %rs1, 0;
-	selp.u32 	%r1, 1, 0, %p1;
-	st.param.b32 	[func_retval0+0], %r1;
-	ret;
-
-}
-
diff --git a/pynvjitlink/tests/undefined_extern.cubin b/pynvjitlink/tests/undefined_extern.cubin
deleted file mode 100644
index 00c220e015813952ce95f207727dfb0d8dcd3522..0000000000000000000000000000000000000000
GIT binary patch
literal 0
HcmV?d00001

literal 2880
zcmeHI&1(};5T9gUHP+Nkt71g)gZ5A_v3^&vP^ceBD-?Q>A|zXP(*)WiBwI;};D&-{
zj~;vUBzW{vdJw!6J$UI=@lr4L53sH?@6Ag#yXhr82@br>%x~s5@4eagUS8gxc@U4q
zt{ny?G1!1Q+d<`V!gSv>Ou{&ZCc%UlK&DoxxmDL%Kuhb!vht<Q)(R!N4w=&POtF+N
zL&kOLZYDc_B|n>=%g$w6kuX*(SXIaFK&#G@73Cuu1#!A(<x<VHO72)k4cl2*S<L3E
z*0KW`qJ{yh6QM%Z($Zqpsubi}E7h`Fb{iF^hs_7K-FqLh^EXyXwv*52<qi(ol5EAo
zZoqJ{ScJ~KAFw^a0~E3+&^O>%2nBd2@qq~b36X*30Oat<k))M^uc!?W?6sAFc?RfH
z-Npwco@fJXT}3^V3ap3cp}Q~?$p-LIMi?iX1lRF`w3%#r-Y6verZ<|9rpmwGrH+Jn
zE<xK6Dg03d#ECXRTu#W{(7wg6vMDMC1NKcpHxk4xA@8v0Pe2^~Tr3U7=`@H_NunTz
zbL?Jh`7ywL1bmr<zH!C}VG4?ZI4fXX2Gm}tv$~6aaL*LlWBnU7#XjT6;UKw<gE$@z
zl3AtXOA^E@jNd^`HSV!b`=GqL0iW76!~P@G6#p;(RtMs%cw+l;%_feICVW`MuR4!5
zVL!+}1n`vo?<O=G!l&;JJLi&6AivMY4CUJ<#Or)%s2=j;^^+7;^|wrzYA6?A(}165
zT+a6dF`n;L17v-%lS%^4_qdhsQ9e9Ge;3cU@2?rKX+pXc#~MBOdNsXjy6;98@+4z>
zp%_0kE%Yc_-S`uncs@i3xo{l$bT`A2V{n@T`u^%SgWfTc_c+2e2_NT0>Gyyzl5p-N
z1&7n%E%QfIytPMpb{HcyrR)fQi}^O^l1u*%Vg5J7a7_{Z7$=Sf_+)Cn?!zy{us0F@
z6He55yqpMsIZ%`8YsoQxi~+rVT~GIa4dX{NDWG37KeZ2E;~V?%*O`AIz$Zx0ulZjP
z<5bR6YLh{|PFjDEukkmC+t|PG8Qf-mJ#b*UeLwo&rxf{ypw{APfcZ2({=cL>oxu7p
Dq3Sx6


From 0b5df09d4abc178e041ad1b16204dc364a3a12ae Mon Sep 17 00:00:00 2001
From: Benjamin Zaitlen <quasiben@users.noreply.github.com>
Date: Mon, 20 Nov 2023 12:16:06 -0500
Subject: [PATCH 09/27] Additional cleanup (#12)

* add missing license metadata and remove skips in tests

* lint

* lint
---
 pynvjitlink/VERSION                       |  1 +
 pynvjitlink/tests/test_pynvjitlink.py     | 16 +++-------------
 pynvjitlink/tests/test_pynvjitlink_api.py | 17 ++++-------------
 pyproject.toml                            |  6 ++++++
 4 files changed, 14 insertions(+), 26 deletions(-)
 create mode 100644 pynvjitlink/VERSION

diff --git a/pynvjitlink/VERSION b/pynvjitlink/VERSION
new file mode 100644
index 00000000..6e8bf73a
--- /dev/null
+++ b/pynvjitlink/VERSION
@@ -0,0 +1 @@
+0.1.0
diff --git a/pynvjitlink/tests/test_pynvjitlink.py b/pynvjitlink/tests/test_pynvjitlink.py
index f08eb4ab..bad1ceea 100644
--- a/pynvjitlink/tests/test_pynvjitlink.py
+++ b/pynvjitlink/tests/test_pynvjitlink.py
@@ -49,7 +49,6 @@ def undefined_extern_cubin():
     return read_test_file('undefined_extern.cubin')
 
 
-@pytest.mark.skip
 def test_create_no_arch_error():
     # nvjitlink expects at least the architecture to be specified.
     with pytest.raises(RuntimeError,
@@ -57,7 +56,6 @@ def test_create_no_arch_error():
         _nvjitlinklib.create()
 
 
-@pytest.mark.skip('Causes fatal error and exit(1)')
 def test_invalid_arch_error():
     # sm_XX is not a valid architecture
     with pytest.raises(RuntimeError,
@@ -65,14 +63,12 @@ def test_invalid_arch_error():
         _nvjitlinklib.create('-arch=sm_XX')
 
 
-@pytest.mark.skip
 def test_unrecognized_option_error():
     with pytest.raises(RuntimeError,
                        match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
         _nvjitlinklib.create('-fictitious_option')
 
 
-@pytest.mark.skip
 def test_invalid_option_type_error():
     with pytest.raises(TypeError,
                        match='Expecting only strings'):
@@ -100,9 +96,7 @@ def test_complete_empty():
                  marks=pytest.mark.xfail),
     ('device_functions_ptx', InputType.PTX),
     ('device_functions_object', InputType.OBJECT),
-    # XXX: Archive type needs debugging - results in a segfault.
-    pytest.param('device_functions_archive', InputType.LIBRARY,
-                 marks=pytest.mark.skip),
+    ('device_functions_archive', InputType.LIBRARY),
 ])
 def test_add_file(input_file, input_type, request):
     filename, data = request.getfixturevalue(input_file)
@@ -112,7 +106,6 @@ def test_add_file(input_file, input_type, request):
     _nvjitlinklib.destroy(handle)
 
 
-@pytest.mark.skip
 def test_get_error_log(undefined_extern_cubin):
     handle = _nvjitlinklib.create('-arch=sm_75')
     filename, data = undefined_extern_cubin
@@ -122,11 +115,8 @@ def test_get_error_log(undefined_extern_cubin):
         _nvjitlinklib.complete(handle)
     error_log = _nvjitlinklib.get_error_log(handle)
     _nvjitlinklib.destroy(handle)
-    # XXX: The error message in the log is strange. The actual expected error
-    # message appears on the terminal:
-    #     error   : Undefined reference to '_Z5undefff' in
-    #               'undefined_extern.cubin'
-    assert "ERROR 9: finish" in error_log
+    assert "Undefined reference to '_Z5undefff' " \
+           "in 'undefined_extern.cubin'" in error_log
 
 
 def test_get_info_log(device_functions_cubin):
diff --git a/pynvjitlink/tests/test_pynvjitlink_api.py b/pynvjitlink/tests/test_pynvjitlink_api.py
index ab7e615d..a7c8650b 100644
--- a/pynvjitlink/tests/test_pynvjitlink_api.py
+++ b/pynvjitlink/tests/test_pynvjitlink_api.py
@@ -31,7 +31,6 @@ def undefined_extern_cubin():
         return f.read()
 
 
-@pytest.mark.skip
 def test_create_no_arch_error():
     # nvlink expects at least the architecture to be specified.
     with pytest.raises(NvJitLinkError,
@@ -39,7 +38,6 @@ def test_create_no_arch_error():
         NvJitLinker()
 
 
-@pytest.mark.skip
 def test_invalid_arch_error():
     # sm_XX is not a valid architecture
     with pytest.raises(NvJitLinkError,
@@ -47,7 +45,6 @@ def test_invalid_arch_error():
         NvJitLinker('-arch=sm_XX')
 
 
-@pytest.mark.skip
 def test_invalid_option_type_error():
     with pytest.raises(TypeError,
                        match='Expecting only strings'):
@@ -65,7 +62,6 @@ def test_add_cubin(device_functions_cubin):
     nvjitlinker.add_cubin(device_functions_cubin, name)
 
 
-@pytest.mark.skip
 def test_add_incompatible_cubin_arch_error(device_functions_cubin):
     nvjitlinker = NvJitLinker('-arch=sm_70')
     name = 'test_device_functions.cubin'
@@ -86,7 +82,6 @@ def test_add_fatbin_sm70(device_functions_fatbin):
     nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
 
-@pytest.mark.skip
 def test_add_incompatible_fatbin_arch_error(device_functions_fatbin):
     nvjitlinker = NvJitLinker('-arch=sm_80')
     name = 'test_device_functions.fatbin'
@@ -95,7 +90,6 @@ def test_add_incompatible_fatbin_arch_error(device_functions_fatbin):
         nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
 
-@pytest.mark.skip
 def test_add_cubin_with_fatbin_error(device_functions_fatbin):
     nvjitlinker = NvJitLinker('-arch=sm_75')
     name = 'test_device_functions.fatbin'
@@ -104,16 +98,14 @@ def test_add_cubin_with_fatbin_error(device_functions_fatbin):
         nvjitlinker.add_cubin(device_functions_fatbin, name)
 
 
-@pytest.mark.skip
-def test_add_fatbin_with_cubin(device_functions_cubin):
-    # Adding a cubin with add_fatbin seems to work - this may be expected
-    # behaviour.
+def test_add_fatbin_with_cubin_error(device_functions_cubin):
     nvjitlinker = NvJitLinker('-arch=sm_75')
     name = 'test_device_functions.cubin'
-    nvjitlinker.add_fatbin(device_functions_cubin, name)
+    with pytest.raises(NvJitLinkError,
+                       match='NVJITLINK_ERROR_INVALID_INPUT error'):
+        nvjitlinker.add_fatbin(device_functions_cubin, name)
 
 
-@pytest.mark.skip
 def test_duplicate_symbols_cubin_and_fatbin(device_functions_cubin,
                                             device_functions_fatbin):
     # This link errors because the cubin and the fatbin contain the same
@@ -145,7 +137,6 @@ def test_get_linked_cubin(device_functions_cubin):
     assert cubin[:4] == b'\x7fELF'
 
 
-@pytest.mark.skip
 def test_get_error_log(undefined_extern_cubin):
     nvjitlinker = NvJitLinker('-arch=sm_75')
     name = 'undefined_extern.cubin'
diff --git a/pyproject.toml b/pyproject.toml
index f11ded5f..245e7509 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -18,3 +18,9 @@ authors = [
 ]
 license = { text = "Apache 2.0" }
 requires-python = ">=3.9"
+
+[tool.setuptools]
+license-files = ["LICENSE"]
+
+[tool.setuptools.dynamic]
+version = {file = "pynvjitlink/VERSION"}

From f5affdbb6a9d8eb6636723047ff7176da63bb4f2 Mon Sep 17 00:00:00 2001
From: Graham Markall <535640+gmarkall@users.noreply.github.com>
Date: Mon, 20 Nov 2023 19:42:37 +0000
Subject: [PATCH 10/27] Add support for LTO in Numba linker (#11)

* Add support for LTO and linking object files

The `lto` kwarg provides a convenience option for enabling LTO. Other
flags might also be required by users; rather than attempting to
maintain a mapping of all flags, we enable arbitrary additional flags to
be passed in for now.

* Add patch tests based on CubinLinker patch tests

This also requires the `PatchedLinker` to store its options for
inspection (which feels like it should be generally handy for a user to
be able to check anyway).

* Bump CUDA toolkit version requirement to 12.3
---
 pynvjitlink/patch.py            | 33 +++++++++---
 pynvjitlink/tests/test_patch.py | 95 +++++++++++++++++++++++++++++++++
 2 files changed, 121 insertions(+), 7 deletions(-)
 create mode 100644 pynvjitlink/tests/test_patch.py

diff --git a/pynvjitlink/patch.py b/pynvjitlink/patch.py
index 1928e050..f979f58f 100644
--- a/pynvjitlink/patch.py
+++ b/pynvjitlink/patch.py
@@ -38,7 +38,8 @@
 
 
 class PatchedLinker(Linker):
-    def __init__(self, max_registers=None, lineinfo=False, cc=None):
+    def __init__(self, max_registers=None, lineinfo=False, cc=None,
+                 lto=False, additional_flags=None):
         if cc is None:
             raise RuntimeError("PatchedLinker requires CC to be specified")
         if not any(isinstance(cc, t) for t in [list, tuple]):
@@ -46,13 +47,18 @@ def __init__(self, max_registers=None, lineinfo=False, cc=None):
 
         sm_ver = f"{cc[0] * 10 + cc[1]}"
         arch = f"-arch=sm_{sm_ver}"
-        opts = [arch]
+        options = [arch]
         if max_registers:
-            opts.append(f"-maxrregcount={max_registers}")
+            options.append(f"-maxrregcount={max_registers}")
         if lineinfo:
-            opts.append("-lineinfo")
+            options.append("-lineinfo")
+        if lto:
+            options.append('-lto')
+        if additional_flags is not None:
+            options.extend(additional_flags)
 
-        self._linker = NvJitLinker(*opts)
+        self._linker = NvJitLinker(*options)
+        self.options = options
 
     @property
     def info_log(self):
@@ -65,6 +71,15 @@ def error_log(self):
     def add_ptx(self, ptx, name='<cudapy-ptx>'):
         self._linker.add_ptx(ptx, name)
 
+    def add_fatbin(self, fatbin, name='<external-fatbin>'):
+        self._linker.add_fatbin(fatbin, name)
+
+    def add_ltoir(self, ltoir, name='<external-ltoir>'):
+        self._linker.add_ltoir(ltoir, name)
+
+    def add_object(self, obj, name='<external-object>'):
+        self._linker.add_object(obj, name)
+
     def add_file(self, path, kind):
         try:
             with open(path, 'rb') as f:
@@ -81,6 +96,8 @@ def add_file(self, path, kind):
             raise LinkerError("Don't know how to link archives")
         elif kind == FILE_EXTENSION_MAP['ptx']:
             return self.add_ptx(data, name)
+        elif kind == FILE_EXTENSION_MAP['o']:
+            fn = self._linker.add_object
         else:
             raise LinkerError(f"Don't know how to link {kind}")
 
@@ -110,8 +127,10 @@ def complete(self):
             raise LinkerError from e
 
 
-def new_patched_linker(max_registers=0, lineinfo=False, cc=None):
-    return PatchedLinker(max_registers, lineinfo, cc)
+def new_patched_linker(max_registers=0, lineinfo=False, cc=None, lto=False,
+                       additional_flags=None):
+    return PatchedLinker(max_registers=max_registers, lineinfo=lineinfo, cc=cc,
+                         lto=lto, additional_flags=additional_flags)
 
 
 def patch_numba_linker():
diff --git a/pynvjitlink/tests/test_patch.py b/pynvjitlink/tests/test_patch.py
new file mode 100644
index 00000000..6d76000d
--- /dev/null
+++ b/pynvjitlink/tests/test_patch.py
@@ -0,0 +1,95 @@
+# Copyright (c) 2023, NVIDIA CORPORATION.
+
+import pytest
+import sys
+
+from pynvjitlink import patch, NvJitLinkError
+from pynvjitlink.patch import (PatchedLinker, patch_numba_linker,
+                               new_patched_linker, required_numba_ver,
+                               _numba_version_ok)
+from unittest.mock import patch as mock_patch
+
+
+def test_numba_patching_numba_not_ok():
+    with mock_patch.multiple(
+            patch,
+            _numba_version_ok=False,
+            _numba_error='<error>'):
+        with pytest.raises(RuntimeError, match='Cannot patch Numba: <error>'):
+            patch_numba_linker()
+
+
+@pytest.mark.skipif(
+    not _numba_version_ok,
+    reason=f"Requires Numba == {required_numba_ver[0]}.{required_numba_ver[1]}"
+)
+def test_numba_patching():
+    # We import the linker here rather than at the top level because the import
+    # may fail if if Numba is not present or an unsupported version.
+    from numba.cuda.cudadrv.driver import Linker
+    patch_numba_linker()
+    assert Linker.new is new_patched_linker
+
+
+def test_create():
+    patched_linker = PatchedLinker(cc=(7, 5))
+    assert "-arch=sm_75" in patched_linker.options
+
+
+def test_create_no_cc_error():
+    # nvJitLink expects at least the architecture to be specified.
+    with pytest.raises(RuntimeError,
+                       match='PatchedLinker requires CC to be specified'):
+        PatchedLinker()
+
+
+def test_invalid_arch_error():
+    # CC 0.0 is not a valid compute capability
+    with pytest.raises(NvJitLinkError,
+                       match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
+        PatchedLinker(cc=(0, 0))
+
+
+def test_invalid_cc_type_error():
+    with pytest.raises(TypeError,
+                       match='`cc` must be a list or tuple of length 2'):
+        PatchedLinker(cc=0)
+
+
+@pytest.mark.parametrize('max_registers', (None, 32))
+@pytest.mark.parametrize('lineinfo', (False, True))
+@pytest.mark.parametrize('lto', (False, True))
+@pytest.mark.parametrize('additional_flags', (None, ('-g',), ('-g', '-time')))
+def test_ptx_compile_options(max_registers, lineinfo, lto, additional_flags):
+    patched_linker = PatchedLinker(
+        cc=(7, 5),
+        max_registers=max_registers,
+        lineinfo=lineinfo,
+        lto=lto,
+        additional_flags=additional_flags
+    )
+
+    assert "-arch=sm_75" in patched_linker.options
+
+    if max_registers:
+        assert f"-maxrregcount={max_registers}" in patched_linker.options
+    else:
+        assert "-maxrregcount" not in patched_linker.options
+
+    if lineinfo:
+        assert "-lineinfo" in patched_linker.options
+    else:
+        assert "-lineinfo" not in patched_linker.options
+
+    if lto:
+        assert "-lto" in patched_linker.options
+    else:
+        assert "-lto" not in patched_linker.options
+
+    if additional_flags:
+        for flag in additional_flags:
+            assert flag in patched_linker.options
+
+
+if __name__ == '__main__':
+    sys.exit(pytest.main())

From 2a4bf86a61e32be4ab2a80474d04c413ada34fe0 Mon Sep 17 00:00:00 2001
From: Lawrence Mitchell <wence@gmx.li>
Date: Mon, 20 Nov 2023 22:53:15 +0000
Subject: [PATCH 11/27] Update to most recent actions versions (#14)

This squashes warnings about old node versions for upload/download of
artifacts.
---
 .github/workflows/pr.yaml | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index 9abbcd4b..02f12cc5 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -43,7 +43,7 @@ jobs:
     - name: Run build_wheel.sh
       run: ci/build_wheel.sh
     - name: Upload Python Wheel
-      uses: actions/upload-artifact@v2
+      uses: actions/upload-artifact@v3
       with:
         path: ./wheel-build
   wheel-build-arm64:
@@ -64,7 +64,7 @@ jobs:
     - name: Run build_wheel.sh
       run: ci/build_wheel.sh
     - name: Upload Python Wheel
-      uses: actions/upload-artifact@v2
+      uses: actions/upload-artifact@v3
       with:
         path: ./wheel-build
   wheel-test-amd64:
@@ -88,7 +88,7 @@ jobs:
       id: get-pr-info
       uses: rapidsai/shared-actions/get-pr-info@main
     - name: Download Python Wheel
-      uses: actions/download-artifact@v2
+      uses: actions/download-artifact@v3
       with:
         path: ./
     - name: Run test_wheel.sh
@@ -113,7 +113,7 @@ jobs:
       id: get-pr-info
       uses: rapidsai/shared-actions/get-pr-info@main
     - name: Download Python Wheel
-      uses: actions/download-artifact@v2
+      uses: actions/download-artifact@v3
       with:
         path: ./
     - name: run test_wheel.sh

From 63fc33cf7404b4959630d4084381b6433222d780 Mon Sep 17 00:00:00 2001
From: Bradley Dice <bdice@bradleydice.com>
Date: Mon, 27 Nov 2023 08:25:04 -0600
Subject: [PATCH 12/27] Clean up style checks. (#16)

This PR updates style checks to trim trailing whitespace and run a basic
suite of Python linting/formatting with ruff.
---
 .github/ISSUE_TEMPLATE/bug_report.md       |   2 +-
 .github/workflows/pr.yaml                  |   4 +-
 .pre-commit-config.yaml                    |  22 +++--
 .ycm_extra_conf.py                         |  24 ++---
 README.md                                  |   2 +-
 pynvjitlink/__init__.py                    |   2 +-
 pynvjitlink/api.py                         |  12 +--
 pynvjitlink/patch.py                       |  70 ++++++++------
 pynvjitlink/tests/test_patch.py            |  43 ++++-----
 pynvjitlink/tests/test_pynvjitlink.py      |  96 +++++++++----------
 pynvjitlink/tests/test_pynvjitlink_api.py  | 103 ++++++++++-----------
 test_binary_generation/test.cu             |   8 +-
 test_binary_generation/undefined_extern.cu |   5 +-
 13 files changed, 202 insertions(+), 191 deletions(-)

diff --git a/.github/ISSUE_TEMPLATE/bug_report.md b/.github/ISSUE_TEMPLATE/bug_report.md
index 34df795e..d47e0704 100644
--- a/.github/ISSUE_TEMPLATE/bug_report.md
+++ b/.github/ISSUE_TEMPLATE/bug_report.md
@@ -20,7 +20,7 @@ A clear and concise description of what you expected to happen.
  - Environment location: [Bare-metal, Docker, Cloud(specify cloud provider)]
  - Method of pynvjitlink install: [conda, Docker, or from source]
    - If method of install is [Docker], provide `docker pull` & `docker run` commands used
- 
+
 
 **Additional context**
 Add any other context about the problem here.
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index 02f12cc5..ab296847 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -92,7 +92,7 @@ jobs:
       with:
         path: ./
     - name: Run test_wheel.sh
-      run: ci/test_wheel.sh 
+      run: ci/test_wheel.sh
   wheel-test-arm64:
     needs:
       - wheel-build-arm64
@@ -117,4 +117,4 @@ jobs:
       with:
         path: ./
     - name: run test_wheel.sh
-      run: ci/test_wheel.sh 
+      run: ci/test_wheel.sh
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index c55fc173..0e557d65 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -1,9 +1,15 @@
 repos:
-- repo: https://github.com/PyCQA/flake8
-  rev: 3.7.8
-  hooks:
-  - id: flake8
-- repo: https://github.com/pre-commit/mirrors-clang-format
-  rev: v11.1.0
-  hooks:
-  - id: clang-format
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v4.5.0
+    hooks:
+      - id: trailing-whitespace
+      - id: end-of-file-fixer
+  - repo: https://github.com/pre-commit/mirrors-clang-format
+    rev: v16.0.6
+    hooks:
+      - id: clang-format
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.1.6
+    hooks:
+      - id: ruff
+      - id: ruff-format
diff --git a/.ycm_extra_conf.py b/.ycm_extra_conf.py
index c7483920..f14a0da5 100644
--- a/.ycm_extra_conf.py
+++ b/.ycm_extra_conf.py
@@ -3,28 +3,28 @@
 
 from pathlib import Path
 
-CONDA_PREFIX = os.environ['CONDA_PREFIX']
+CONDA_PREFIX = os.environ["CONDA_PREFIX"]
 NVJITLINK_DIR = Path(__file__).parent
 VER = sys.version_info
-PYTHON_INCLUDE_NAME = 'python3.10'  # Not ideal
+PYTHON_INCLUDE_NAME = "python3.10"  # Not ideal
 
-CONDA_INCLUDE_DIR = Path(CONDA_PREFIX, 'include')
+CONDA_INCLUDE_DIR = Path(CONDA_PREFIX, "include")
 PYTHON_INCLUDE_DIR = Path(CONDA_INCLUDE_DIR, PYTHON_INCLUDE_NAME)
-NVJITLINK_INCLUDE_DIR = Path(NVJITLINK_DIR, 'include')
-CUDA_INCLUDE_DIR = '/usr/local/cuda-12.0/include'
+NVJITLINK_INCLUDE_DIR = Path(NVJITLINK_DIR, "include")
+CUDA_INCLUDE_DIR = "/usr/local/cuda-12.0/include"
 
 flags = [
-    '--cuda-gpu-arch=sm_50',
-    f'-I{CONDA_INCLUDE_DIR}',
-    f'-I{CUDA_INCLUDE_DIR}',
-    f'-I{NVJITLINK_INCLUDE_DIR}',
-    f'-I{PYTHON_INCLUDE_DIR}',
+    "--cuda-gpu-arch=sm_50",
+    f"-I{CONDA_INCLUDE_DIR}",
+    f"-I{CUDA_INCLUDE_DIR}",
+    f"-I{NVJITLINK_INCLUDE_DIR}",
+    f"-I{PYTHON_INCLUDE_DIR}",
 ]
 
 
 def Settings(**kwargs):
-    return {'flags': flags}
+    return {"flags": flags}
 
 
-if __name__ == '__main__':
+if __name__ == "__main__":
     print(Settings())
diff --git a/README.md b/README.md
index a18c78b8..7f31947d 100644
--- a/README.md
+++ b/README.md
@@ -1,6 +1,6 @@
 # <div align="left"><img src="https://rapids.ai/assets/images/rapids_logo.png" width="90px"/>&nbsp;pynvjitlink</div>
 
-The [RAPIDS](https://rapids.ai) pynvjitlink library provides a Python binding for the 
+The [RAPIDS](https://rapids.ai) pynvjitlink library provides a Python binding for the
 [nvJitLink library](https://docs.nvidia.com/cuda/nvJitLink/index.html). It is presently a
 work-in-progress.
 
diff --git a/pynvjitlink/__init__.py b/pynvjitlink/__init__.py
index 9ad447f1..597f9626 100644
--- a/pynvjitlink/__init__.py
+++ b/pynvjitlink/__init__.py
@@ -3,4 +3,4 @@
 from pynvjitlink.api import NvJitLinker, NvJitLinkError
 
 __all__ = (NvJitLinker, NvJitLinkError)
-__version__ = '0.0.0'
+__version__ = "0.0.0"
diff --git a/pynvjitlink/api.py b/pynvjitlink/api.py
index f8aff6d0..34da29ee 100644
--- a/pynvjitlink/api.py
+++ b/pynvjitlink/api.py
@@ -43,7 +43,7 @@ def error_log(self):
 
     def add_data(self, input_type, data, name):
         if self._complete:
-            raise NvJitLinkError('Cannot add data to already-completeted link')
+            raise NvJitLinkError("Cannot add data to already-completeted link")
 
         try:
             _nvjitlinklib.add_data(self.handle, input_type.value, data, name)
@@ -53,23 +53,23 @@ def add_data(self, input_type, data, name):
             raise NvJitLinkError(f"{e}\n{self.error_log}")
 
     def add_cubin(self, cubin, name=None):
-        name = name or 'unnamed-cubin'
+        name = name or "unnamed-cubin"
         self.add_data(InputType.CUBIN, cubin, name)
 
     def add_ptx(self, ptx, name=None):
-        name = name or 'unnamed-ptx'
+        name = name or "unnamed-ptx"
         self.add_data(InputType.PTX, ptx, name)
 
     def add_ltoir(self, ltoir, name=None):
-        name = name or 'unnamed-ltoir'
+        name = name or "unnamed-ltoir"
         self.add_data(InputType.LTOIR, ltoir, name)
 
     def add_object(self, object_, name=None):
-        name = name or 'unnamed-object'
+        name = name or "unnamed-object"
         self.add_data(InputType.OBJECT, object_, name)
 
     def add_fatbin(self, fatbin, name=None):
-        name = name or 'unnamed-fatbin'
+        name = name or "unnamed-fatbin"
         self.add_data(InputType.FATBIN, fatbin, name)
 
     def add_library(self, library, name=None):
diff --git a/pynvjitlink/patch.py b/pynvjitlink/patch.py
index f979f58f..e5f6cb3c 100644
--- a/pynvjitlink/patch.py
+++ b/pynvjitlink/patch.py
@@ -9,15 +9,19 @@
 
 required_numba_ver = (0, 57)
 
-mvc_docs_url = ("https://numba.readthedocs.io/en/stable/cuda/"
-                "minor_version_compatibility.html")
+mvc_docs_url = (
+    "https://numba.readthedocs.io/en/stable/cuda/" "minor_version_compatibility.html"
+)
 
 try:
     import numba
+
     ver = numba.version_info.short
     if ver < required_numba_ver:
-        _numba_error = (f"version {numba.__version__} is insufficient for "
-                        "patching - %s.%s is needed." % required_numba_ver)
+        _numba_error = (
+            f"version {numba.__version__} is insufficient for "
+            "patching - %s.%s is needed." % required_numba_ver
+        )
     else:
         _numba_version_ok = True
 except ImportError as ie:
@@ -25,8 +29,8 @@
 
 if _numba_version_ok:
     from numba.core import config
-    from numba.cuda.cudadrv.driver import (FILE_EXTENSION_MAP, Linker,
-                                           LinkerError)
+    from numba.cuda.cudadrv.driver import FILE_EXTENSION_MAP, Linker, LinkerError
+
     if ver < (0, 58):
         from numba.cuda.cudadrv.driver import NvrtcProgram
     else:
@@ -38,8 +42,14 @@
 
 
 class PatchedLinker(Linker):
-    def __init__(self, max_registers=None, lineinfo=False, cc=None,
-                 lto=False, additional_flags=None):
+    def __init__(
+        self,
+        max_registers=None,
+        lineinfo=False,
+        cc=None,
+        lto=False,
+        additional_flags=None,
+    ):
         if cc is None:
             raise RuntimeError("PatchedLinker requires CC to be specified")
         if not any(isinstance(cc, t) for t in [list, tuple]):
@@ -53,7 +63,7 @@ def __init__(self, max_registers=None, lineinfo=False, cc=None,
         if lineinfo:
             options.append("-lineinfo")
         if lto:
-            options.append('-lto')
+            options.append("-lto")
         if additional_flags is not None:
             options.extend(additional_flags)
 
@@ -68,35 +78,35 @@ def info_log(self):
     def error_log(self):
         return self._linker.error_log
 
-    def add_ptx(self, ptx, name='<cudapy-ptx>'):
+    def add_ptx(self, ptx, name="<cudapy-ptx>"):
         self._linker.add_ptx(ptx, name)
 
-    def add_fatbin(self, fatbin, name='<external-fatbin>'):
+    def add_fatbin(self, fatbin, name="<external-fatbin>"):
         self._linker.add_fatbin(fatbin, name)
 
-    def add_ltoir(self, ltoir, name='<external-ltoir>'):
+    def add_ltoir(self, ltoir, name="<external-ltoir>"):
         self._linker.add_ltoir(ltoir, name)
 
-    def add_object(self, obj, name='<external-object>'):
+    def add_object(self, obj, name="<external-object>"):
         self._linker.add_object(obj, name)
 
     def add_file(self, path, kind):
         try:
-            with open(path, 'rb') as f:
+            with open(path, "rb") as f:
                 data = f.read()
         except FileNotFoundError:
-            raise LinkerError(f'{path} not found')
+            raise LinkerError(f"{path} not found")
 
         name = pathlib.Path(path).name
-        if kind == FILE_EXTENSION_MAP['cubin']:
+        if kind == FILE_EXTENSION_MAP["cubin"]:
             fn = self._linker.add_cubin
-        elif kind == FILE_EXTENSION_MAP['fatbin']:
+        elif kind == FILE_EXTENSION_MAP["fatbin"]:
             fn = self._linker.add_fatbin
-        elif kind == FILE_EXTENSION_MAP['a']:
+        elif kind == FILE_EXTENSION_MAP["a"]:
             raise LinkerError("Don't know how to link archives")
-        elif kind == FILE_EXTENSION_MAP['ptx']:
+        elif kind == FILE_EXTENSION_MAP["ptx"]:
             return self.add_ptx(data, name)
-        elif kind == FILE_EXTENSION_MAP['o']:
+        elif kind == FILE_EXTENSION_MAP["o"]:
             fn = self._linker.add_object
         else:
             raise LinkerError(f"Don't know how to link {kind}")
@@ -110,13 +120,13 @@ def add_cu(self, cu, name):
         program = NvrtcProgram(cu, name)
 
         if config.DUMP_ASSEMBLY:
-            print(("ASSEMBLY %s" % name).center(80, '-'))
+            print(("ASSEMBLY %s" % name).center(80, "-"))
             print(program.ptx.decode())
-            print('=' * 80)
+            print("=" * 80)
 
         # Link the program's PTX using the normal linker mechanism
         ptx_name = os.path.splitext(name)[0] + ".ptx"
-        self.add_ptx(program.ptx.rstrip(b'\x00'), ptx_name)
+        self.add_ptx(program.ptx.rstrip(b"\x00"), ptx_name)
 
     def complete(self):
         try:
@@ -127,10 +137,16 @@ def complete(self):
             raise LinkerError from e
 
 
-def new_patched_linker(max_registers=0, lineinfo=False, cc=None, lto=False,
-                       additional_flags=None):
-    return PatchedLinker(max_registers=max_registers, lineinfo=lineinfo, cc=cc,
-                         lto=lto, additional_flags=additional_flags)
+def new_patched_linker(
+    max_registers=0, lineinfo=False, cc=None, lto=False, additional_flags=None
+):
+    return PatchedLinker(
+        max_registers=max_registers,
+        lineinfo=lineinfo,
+        cc=cc,
+        lto=lto,
+        additional_flags=additional_flags,
+    )
 
 
 def patch_numba_linker():
diff --git a/pynvjitlink/tests/test_patch.py b/pynvjitlink/tests/test_patch.py
index 6d76000d..14a57a5b 100644
--- a/pynvjitlink/tests/test_patch.py
+++ b/pynvjitlink/tests/test_patch.py
@@ -4,29 +4,31 @@
 import sys
 
 from pynvjitlink import patch, NvJitLinkError
-from pynvjitlink.patch import (PatchedLinker, patch_numba_linker,
-                               new_patched_linker, required_numba_ver,
-                               _numba_version_ok)
+from pynvjitlink.patch import (
+    PatchedLinker,
+    patch_numba_linker,
+    new_patched_linker,
+    required_numba_ver,
+    _numba_version_ok,
+)
 from unittest.mock import patch as mock_patch
 
 
 def test_numba_patching_numba_not_ok():
-    with mock_patch.multiple(
-            patch,
-            _numba_version_ok=False,
-            _numba_error='<error>'):
-        with pytest.raises(RuntimeError, match='Cannot patch Numba: <error>'):
+    with mock_patch.multiple(patch, _numba_version_ok=False, _numba_error="<error>"):
+        with pytest.raises(RuntimeError, match="Cannot patch Numba: <error>"):
             patch_numba_linker()
 
 
 @pytest.mark.skipif(
     not _numba_version_ok,
-    reason=f"Requires Numba == {required_numba_ver[0]}.{required_numba_ver[1]}"
+    reason=f"Requires Numba == {required_numba_ver[0]}.{required_numba_ver[1]}",
 )
 def test_numba_patching():
     # We import the linker here rather than at the top level because the import
     # may fail if if Numba is not present or an unsupported version.
     from numba.cuda.cudadrv.driver import Linker
+
     patch_numba_linker()
     assert Linker.new is new_patched_linker
 
@@ -38,35 +40,34 @@ def test_create():
 
 def test_create_no_cc_error():
     # nvJitLink expects at least the architecture to be specified.
-    with pytest.raises(RuntimeError,
-                       match='PatchedLinker requires CC to be specified'):
+    with pytest.raises(RuntimeError, match="PatchedLinker requires CC to be specified"):
         PatchedLinker()
 
 
 def test_invalid_arch_error():
     # CC 0.0 is not a valid compute capability
-    with pytest.raises(NvJitLinkError,
-                       match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
+    with pytest.raises(
+        NvJitLinkError, match="NVJITLINK_ERROR_UNRECOGNIZED_OPTION error"
+    ):
         PatchedLinker(cc=(0, 0))
 
 
 def test_invalid_cc_type_error():
-    with pytest.raises(TypeError,
-                       match='`cc` must be a list or tuple of length 2'):
+    with pytest.raises(TypeError, match="`cc` must be a list or tuple of length 2"):
         PatchedLinker(cc=0)
 
 
-@pytest.mark.parametrize('max_registers', (None, 32))
-@pytest.mark.parametrize('lineinfo', (False, True))
-@pytest.mark.parametrize('lto', (False, True))
-@pytest.mark.parametrize('additional_flags', (None, ('-g',), ('-g', '-time')))
+@pytest.mark.parametrize("max_registers", (None, 32))
+@pytest.mark.parametrize("lineinfo", (False, True))
+@pytest.mark.parametrize("lto", (False, True))
+@pytest.mark.parametrize("additional_flags", (None, ("-g",), ("-g", "-time")))
 def test_ptx_compile_options(max_registers, lineinfo, lto, additional_flags):
     patched_linker = PatchedLinker(
         cc=(7, 5),
         max_registers=max_registers,
         lineinfo=lineinfo,
         lto=lto,
-        additional_flags=additional_flags
+        additional_flags=additional_flags,
     )
 
     assert "-arch=sm_75" in patched_linker.options
@@ -91,5 +92,5 @@ def test_ptx_compile_options(max_registers, lineinfo, lto, additional_flags):
             assert flag in patched_linker.options
 
 
-if __name__ == '__main__':
+if __name__ == "__main__":
     sys.exit(pytest.main())
diff --git a/pynvjitlink/tests/test_pynvjitlink.py b/pynvjitlink/tests/test_pynvjitlink.py
index bad1ceea..e4c7e2fc 100644
--- a/pynvjitlink/tests/test_pynvjitlink.py
+++ b/pynvjitlink/tests/test_pynvjitlink.py
@@ -10,104 +10,104 @@
 def read_test_file(filename):
     test_dir = os.path.dirname(os.path.abspath(__file__))
     path = os.path.join(test_dir, filename)
-    with open(path, 'rb') as f:
+    with open(path, "rb") as f:
         return filename, f.read()
 
 
 @pytest.fixture
 def device_functions_cubin():
-    return read_test_file('test_device_functions.cubin')
+    return read_test_file("test_device_functions.cubin")
 
 
 @pytest.fixture
 def device_functions_fatbin():
-    return read_test_file('test_device_functions.fatbin')
+    return read_test_file("test_device_functions.fatbin")
 
 
 @pytest.fixture
 def device_functions_ltoir():
-    return read_test_file('test_device_functions.ltoir')
+    return read_test_file("test_device_functions.ltoir")
 
 
 @pytest.fixture
 def device_functions_object():
-    return read_test_file('test_device_functions.o')
+    return read_test_file("test_device_functions.o")
 
 
 @pytest.fixture
 def device_functions_archive():
-    return read_test_file('test_device_functions.a')
+    return read_test_file("test_device_functions.a")
 
 
 @pytest.fixture
 def device_functions_ptx():
-    return read_test_file('test_device_functions.ptx')
+    return read_test_file("test_device_functions.ptx")
 
 
 @pytest.fixture
 def undefined_extern_cubin():
-    return read_test_file('undefined_extern.cubin')
+    return read_test_file("undefined_extern.cubin")
 
 
 def test_create_no_arch_error():
     # nvjitlink expects at least the architecture to be specified.
-    with pytest.raises(RuntimeError,
-                       match='NVJITLINK_ERROR_MISSING_ARCH error'):
+    with pytest.raises(RuntimeError, match="NVJITLINK_ERROR_MISSING_ARCH error"):
         _nvjitlinklib.create()
 
 
 def test_invalid_arch_error():
     # sm_XX is not a valid architecture
-    with pytest.raises(RuntimeError,
-                       match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
-        _nvjitlinklib.create('-arch=sm_XX')
+    with pytest.raises(RuntimeError, match="NVJITLINK_ERROR_UNRECOGNIZED_OPTION error"):
+        _nvjitlinklib.create("-arch=sm_XX")
 
 
 def test_unrecognized_option_error():
-    with pytest.raises(RuntimeError,
-                       match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
-        _nvjitlinklib.create('-fictitious_option')
+    with pytest.raises(RuntimeError, match="NVJITLINK_ERROR_UNRECOGNIZED_OPTION error"):
+        _nvjitlinklib.create("-fictitious_option")
 
 
 def test_invalid_option_type_error():
-    with pytest.raises(TypeError,
-                       match='Expecting only strings'):
-        _nvjitlinklib.create('-arch', 53)
+    with pytest.raises(TypeError, match="Expecting only strings"):
+        _nvjitlinklib.create("-arch", 53)
 
 
 def test_create_and_destroy():
-    handle = _nvjitlinklib.create('-arch=sm_53')
+    handle = _nvjitlinklib.create("-arch=sm_53")
     assert handle != 0
     _nvjitlinklib.destroy(handle)
 
 
 def test_complete_empty():
-    handle = _nvjitlinklib.create('-arch=sm_75')
+    handle = _nvjitlinklib.create("-arch=sm_75")
     _nvjitlinklib.complete(handle)
     _nvjitlinklib.destroy(handle)
 
 
-@pytest.mark.parametrize('input_file,input_type', [
-    ('device_functions_cubin', InputType.CUBIN),
-    ('device_functions_fatbin', InputType.FATBIN),
-    # XXX: LTOIR input type needs debugging - results in
-    # NVJITLINK_ERROR_INTERNAL.
-    pytest.param('device_functions_ltoir', InputType.LTOIR,
-                 marks=pytest.mark.xfail),
-    ('device_functions_ptx', InputType.PTX),
-    ('device_functions_object', InputType.OBJECT),
-    ('device_functions_archive', InputType.LIBRARY),
-])
+@pytest.mark.parametrize(
+    "input_file,input_type",
+    [
+        ("device_functions_cubin", InputType.CUBIN),
+        ("device_functions_fatbin", InputType.FATBIN),
+        # XXX: LTOIR input type needs debugging - results in
+        # NVJITLINK_ERROR_INTERNAL.
+        pytest.param(
+            "device_functions_ltoir", InputType.LTOIR, marks=pytest.mark.xfail
+        ),
+        ("device_functions_ptx", InputType.PTX),
+        ("device_functions_object", InputType.OBJECT),
+        ("device_functions_archive", InputType.LIBRARY),
+    ],
+)
 def test_add_file(input_file, input_type, request):
     filename, data = request.getfixturevalue(input_file)
 
-    handle = _nvjitlinklib.create('-arch=sm_75')
+    handle = _nvjitlinklib.create("-arch=sm_75")
     _nvjitlinklib.add_data(handle, input_type.value, data, filename)
     _nvjitlinklib.destroy(handle)
 
 
 def test_get_error_log(undefined_extern_cubin):
-    handle = _nvjitlinklib.create('-arch=sm_75')
+    handle = _nvjitlinklib.create("-arch=sm_75")
     filename, data = undefined_extern_cubin
     input_type = InputType.CUBIN.value
     _nvjitlinklib.add_data(handle, input_type, data, filename)
@@ -115,12 +115,14 @@ def test_get_error_log(undefined_extern_cubin):
         _nvjitlinklib.complete(handle)
     error_log = _nvjitlinklib.get_error_log(handle)
     _nvjitlinklib.destroy(handle)
-    assert "Undefined reference to '_Z5undefff' " \
-           "in 'undefined_extern.cubin'" in error_log
+    assert (
+        "Undefined reference to '_Z5undefff' "
+        "in 'undefined_extern.cubin'" in error_log
+    )
 
 
 def test_get_info_log(device_functions_cubin):
-    handle = _nvjitlinklib.create('-arch=sm_75')
+    handle = _nvjitlinklib.create("-arch=sm_75")
     filename, data = device_functions_cubin
     input_type = InputType.CUBIN.value
     _nvjitlinklib.add_data(handle, input_type, data, filename)
@@ -132,7 +134,7 @@ def test_get_info_log(device_functions_cubin):
 
 
 def test_get_linked_cubin(device_functions_cubin):
-    handle = _nvjitlinklib.create('-arch=sm_75')
+    handle = _nvjitlinklib.create("-arch=sm_75")
     filename, data = device_functions_cubin
     input_type = InputType.CUBIN.value
     _nvjitlinklib.add_data(handle, input_type, data, filename)
@@ -141,16 +143,15 @@ def test_get_linked_cubin(device_functions_cubin):
     _nvjitlinklib.destroy(handle)
 
     # Just check we got something that looks like an ELF
-    assert cubin[:4] == b'\x7fELF'
+    assert cubin[:4] == b"\x7fELF"
 
 
 def test_get_linked_cubin_link_not_complete_error(device_functions_cubin):
-    handle = _nvjitlinklib.create('-arch=sm_75')
+    handle = _nvjitlinklib.create("-arch=sm_75")
     filename, data = device_functions_cubin
     input_type = InputType.CUBIN.value
     _nvjitlinklib.add_data(handle, input_type, data, filename)
-    with pytest.raises(RuntimeError,
-                       match="NVJITLINK_ERROR_INTERNAL error"):
+    with pytest.raises(RuntimeError, match="NVJITLINK_ERROR_INTERNAL error"):
         _nvjitlinklib.get_linked_cubin(handle)
     _nvjitlinklib.destroy(handle)
 
@@ -162,14 +163,14 @@ def test_get_linked_cubin_from_lto(device_functions_ltoir):
     # requested. So we need to use the OBJECT input type, and the linker
     # retrieves the LTO IR from it because we passed the -lto flag.
     input_type = InputType.OBJECT.value
-    handle = _nvjitlinklib.create('-arch=sm_75', '-lto')
+    handle = _nvjitlinklib.create("-arch=sm_75", "-lto")
     _nvjitlinklib.add_data(handle, input_type, data, filename)
     _nvjitlinklib.complete(handle)
     cubin = _nvjitlinklib.get_linked_cubin(handle)
     _nvjitlinklib.destroy(handle)
 
     # Just check we got something that looks like an ELF
-    assert cubin[:4] == b'\x7fELF'
+    assert cubin[:4] == b"\x7fELF"
 
 
 def test_get_linked_ptx_from_lto(device_functions_ltoir):
@@ -179,7 +180,7 @@ def test_get_linked_ptx_from_lto(device_functions_ltoir):
     # requested. So we need to use the OBJECT input type, and the linker
     # retrieves the LTO IR from it because we passed the -lto flag.
     input_type = InputType.OBJECT.value
-    handle = _nvjitlinklib.create('-arch=sm_75', '-lto', '-ptx')
+    handle = _nvjitlinklib.create("-arch=sm_75", "-lto", "-ptx")
     _nvjitlinklib.add_data(handle, input_type, data, filename)
     _nvjitlinklib.complete(handle)
     _nvjitlinklib.get_linked_ptx(handle)
@@ -187,11 +188,10 @@ def test_get_linked_ptx_from_lto(device_functions_ltoir):
 
 
 def test_get_linked_ptx_link_not_complete_error(device_functions_ltoir):
-    handle = _nvjitlinklib.create('-arch=sm_75', '-lto', '-ptx')
+    handle = _nvjitlinklib.create("-arch=sm_75", "-lto", "-ptx")
     filename, data = device_functions_ltoir
     input_type = InputType.OBJECT.value
     _nvjitlinklib.add_data(handle, input_type, data, filename)
-    with pytest.raises(RuntimeError,
-                       match="NVJITLINK_ERROR_INTERNAL error"):
+    with pytest.raises(RuntimeError, match="NVJITLINK_ERROR_INTERNAL error"):
         _nvjitlinklib.get_linked_ptx(handle)
     _nvjitlinklib.destroy(handle)
diff --git a/pynvjitlink/tests/test_pynvjitlink_api.py b/pynvjitlink/tests/test_pynvjitlink_api.py
index a7c8650b..807a0711 100644
--- a/pynvjitlink/tests/test_pynvjitlink_api.py
+++ b/pynvjitlink/tests/test_pynvjitlink_api.py
@@ -10,136 +10,131 @@
 @pytest.fixture
 def device_functions_cubin():
     test_dir = os.path.dirname(os.path.abspath(__file__))
-    cubin_path = os.path.join(test_dir, 'test_device_functions.cubin')
-    with open(cubin_path, 'rb') as f:
+    cubin_path = os.path.join(test_dir, "test_device_functions.cubin")
+    with open(cubin_path, "rb") as f:
         return f.read()
 
 
 @pytest.fixture
 def device_functions_fatbin():
     test_dir = os.path.dirname(os.path.abspath(__file__))
-    fatbin_path = os.path.join(test_dir, 'test_device_functions.fatbin')
-    with open(fatbin_path, 'rb') as f:
+    fatbin_path = os.path.join(test_dir, "test_device_functions.fatbin")
+    with open(fatbin_path, "rb") as f:
         return f.read()
 
 
 @pytest.fixture
 def undefined_extern_cubin():
     test_dir = os.path.dirname(os.path.abspath(__file__))
-    fatbin_path = os.path.join(test_dir, 'undefined_extern.cubin')
-    with open(fatbin_path, 'rb') as f:
+    fatbin_path = os.path.join(test_dir, "undefined_extern.cubin")
+    with open(fatbin_path, "rb") as f:
         return f.read()
 
 
 def test_create_no_arch_error():
     # nvlink expects at least the architecture to be specified.
-    with pytest.raises(NvJitLinkError,
-                       match='NVJITLINK_ERROR_MISSING_ARCH error'):
+    with pytest.raises(NvJitLinkError, match="NVJITLINK_ERROR_MISSING_ARCH error"):
         NvJitLinker()
 
 
 def test_invalid_arch_error():
     # sm_XX is not a valid architecture
-    with pytest.raises(NvJitLinkError,
-                       match='NVJITLINK_ERROR_UNRECOGNIZED_OPTION error'):
-        NvJitLinker('-arch=sm_XX')
+    with pytest.raises(
+        NvJitLinkError, match="NVJITLINK_ERROR_UNRECOGNIZED_OPTION error"
+    ):
+        NvJitLinker("-arch=sm_XX")
 
 
 def test_invalid_option_type_error():
-    with pytest.raises(TypeError,
-                       match='Expecting only strings'):
-        NvJitLinker('-arch', 53)
+    with pytest.raises(TypeError, match="Expecting only strings"):
+        NvJitLinker("-arch", 53)
 
 
 def test_create_and_destroy():
-    nvjitlinker = NvJitLinker('-arch=sm_53')
+    nvjitlinker = NvJitLinker("-arch=sm_53")
     assert nvjitlinker.handle != 0
 
 
 def test_add_cubin(device_functions_cubin):
-    nvjitlinker = NvJitLinker('-arch=sm_75')
-    name = 'test_device_functions.cubin'
+    nvjitlinker = NvJitLinker("-arch=sm_75")
+    name = "test_device_functions.cubin"
     nvjitlinker.add_cubin(device_functions_cubin, name)
 
 
 def test_add_incompatible_cubin_arch_error(device_functions_cubin):
-    nvjitlinker = NvJitLinker('-arch=sm_70')
-    name = 'test_device_functions.cubin'
-    with pytest.raises(NvJitLinkError,
-                       match='NVJITLINK_ERROR_INVALID_INPUT error'):
+    nvjitlinker = NvJitLinker("-arch=sm_70")
+    name = "test_device_functions.cubin"
+    with pytest.raises(NvJitLinkError, match="NVJITLINK_ERROR_INVALID_INPUT error"):
         nvjitlinker.add_cubin(device_functions_cubin, name)
 
 
 def test_add_fatbin_sm75(device_functions_fatbin):
-    nvjitlinker = NvJitLinker('-arch=sm_75')
-    name = 'test_device_functions.fatbin'
+    nvjitlinker = NvJitLinker("-arch=sm_75")
+    name = "test_device_functions.fatbin"
     nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
 
 def test_add_fatbin_sm70(device_functions_fatbin):
-    nvjitlinker = NvJitLinker('-arch=sm_70')
-    name = 'test_device_functions.fatbin'
+    nvjitlinker = NvJitLinker("-arch=sm_70")
+    name = "test_device_functions.fatbin"
     nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
 
 def test_add_incompatible_fatbin_arch_error(device_functions_fatbin):
-    nvjitlinker = NvJitLinker('-arch=sm_80')
-    name = 'test_device_functions.fatbin'
-    with pytest.raises(NvJitLinkError,
-                       match='NVJITLINK_ERROR_INVALID_INPUT error'):
+    nvjitlinker = NvJitLinker("-arch=sm_80")
+    name = "test_device_functions.fatbin"
+    with pytest.raises(NvJitLinkError, match="NVJITLINK_ERROR_INVALID_INPUT error"):
         nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
 
 def test_add_cubin_with_fatbin_error(device_functions_fatbin):
-    nvjitlinker = NvJitLinker('-arch=sm_75')
-    name = 'test_device_functions.fatbin'
-    with pytest.raises(NvJitLinkError,
-                       match='NVJITLINK_ERROR_INVALID_INPUT error'):
+    nvjitlinker = NvJitLinker("-arch=sm_75")
+    name = "test_device_functions.fatbin"
+    with pytest.raises(NvJitLinkError, match="NVJITLINK_ERROR_INVALID_INPUT error"):
         nvjitlinker.add_cubin(device_functions_fatbin, name)
 
 
 def test_add_fatbin_with_cubin_error(device_functions_cubin):
-    nvjitlinker = NvJitLinker('-arch=sm_75')
-    name = 'test_device_functions.cubin'
-    with pytest.raises(NvJitLinkError,
-                       match='NVJITLINK_ERROR_INVALID_INPUT error'):
+    nvjitlinker = NvJitLinker("-arch=sm_75")
+    name = "test_device_functions.cubin"
+    with pytest.raises(NvJitLinkError, match="NVJITLINK_ERROR_INVALID_INPUT error"):
         nvjitlinker.add_fatbin(device_functions_cubin, name)
 
 
-def test_duplicate_symbols_cubin_and_fatbin(device_functions_cubin,
-                                            device_functions_fatbin):
+def test_duplicate_symbols_cubin_and_fatbin(
+    device_functions_cubin, device_functions_fatbin
+):
     # This link errors because the cubin and the fatbin contain the same
     # symbols.
-    nvjitlinker = NvJitLinker('-arch=sm_75')
-    name = 'test_device_functions.cubin'
+    nvjitlinker = NvJitLinker("-arch=sm_75")
+    name = "test_device_functions.cubin"
     nvjitlinker.add_cubin(device_functions_cubin, name)
-    name = 'test_device_functions.fatbin'
-    with pytest.raises(NvJitLinkError,
-                       match="NVJITLINK_ERROR_INVALID_INPUT error"):
+    name = "test_device_functions.fatbin"
+    with pytest.raises(NvJitLinkError, match="NVJITLINK_ERROR_INVALID_INPUT error"):
         nvjitlinker.add_fatbin(device_functions_fatbin, name)
 
 
 def test_get_linked_cubin_complete_empty_error():
-    nvjitlinker = NvJitLinker('-arch=sm_75')
+    nvjitlinker = NvJitLinker("-arch=sm_75")
     cubin = nvjitlinker.get_linked_cubin()
 
     # Linking nothing still gives us an empty ELF back
-    assert cubin[:4] == b'\x7fELF'
+    assert cubin[:4] == b"\x7fELF"
 
 
 def test_get_linked_cubin(device_functions_cubin):
-    nvjitlinker = NvJitLinker('-arch=sm_75')
-    name = 'test_device_functions.cubin'
+    nvjitlinker = NvJitLinker("-arch=sm_75")
+    name = "test_device_functions.cubin"
     nvjitlinker.add_cubin(device_functions_cubin, name)
     cubin = nvjitlinker.get_linked_cubin()
 
     # Just check we got something that looks like an ELF
-    assert cubin[:4] == b'\x7fELF'
+    assert cubin[:4] == b"\x7fELF"
 
 
 def test_get_error_log(undefined_extern_cubin):
-    nvjitlinker = NvJitLinker('-arch=sm_75')
-    name = 'undefined_extern.cubin'
+    nvjitlinker = NvJitLinker("-arch=sm_75")
+    name = "undefined_extern.cubin"
     nvjitlinker.add_cubin(undefined_extern_cubin, name)
     with pytest.raises(NvJitLinkError):
         nvjitlinker.get_linked_cubin()
@@ -148,8 +143,8 @@ def test_get_error_log(undefined_extern_cubin):
 
 
 def test_get_info_log(device_functions_cubin):
-    nvjitlinker = NvJitLinker('-arch=sm_75')
-    name = 'test_device_functions.cubin'
+    nvjitlinker = NvJitLinker("-arch=sm_75")
+    name = "test_device_functions.cubin"
     nvjitlinker.add_cubin(device_functions_cubin, name)
     nvjitlinker.get_linked_cubin()
     info_log = nvjitlinker.info_log
@@ -157,5 +152,5 @@ def test_get_info_log(device_functions_cubin):
     assert "" == info_log
 
 
-if __name__ == '__main__':
+if __name__ == "__main__":
     sys.exit(pytest.main())
diff --git a/test_binary_generation/test.cu b/test_binary_generation/test.cu
index c7a37953..00b52061 100644
--- a/test_binary_generation/test.cu
+++ b/test_binary_generation/test.cu
@@ -2,14 +2,10 @@
 
 extern __device__ bool __heq(__half arg1, __half arg2);
 
-__device__
-__half test_add_fp16(__half arg1, __half arg2)
-{
+__device__ __half test_add_fp16(__half arg1, __half arg2) {
   return __hadd(arg1, arg2);
 }
 
-__device__
-bool test_cmp_fp16(__half arg1, __half arg2)
-{
+__device__ bool test_cmp_fp16(__half arg1, __half arg2) {
   return __heq(arg1, arg2);
 }
diff --git a/test_binary_generation/undefined_extern.cu b/test_binary_generation/undefined_extern.cu
index 86b07d85..9e0a0cad 100644
--- a/test_binary_generation/undefined_extern.cu
+++ b/test_binary_generation/undefined_extern.cu
@@ -1,6 +1,3 @@
 extern __device__ float undef(float a, float b);
 
-__global__ void f(float* r, float *a, float *b)
-{
-  r[0] = undef(a[0], b[0]);
-}
+__global__ void f(float *r, float *a, float *b) { r[0] = undef(a[0], b[0]); }

From 2903ee3bdbf73f63e334def32874168c09779e0d Mon Sep 17 00:00:00 2001
From: Bradley Dice <bdice@bradleydice.com>
Date: Mon, 27 Nov 2023 10:42:20 -0800
Subject: [PATCH 13/27] Create releases when tags are pushed. (#17)

Co-authored-by: brandon-b-miller <brmiller@nvidia.com>
Co-authored-by: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Co-authored-by: Lawrence Mitchell <wence@gmx.li>
---
 .github/workflows/pr.yaml | 21 +++++++++++++++++++--
 1 file changed, 19 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index ab296847..533fa9f8 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -1,13 +1,15 @@
 # Copyright (c) 2023, NVIDIA CORPORATION.
-name: pr
+name: Build and test
 
 on:
   push:
     branches:
       - "pull-request/[0-9]+"
+    tags:
+      - 'v*'
 
 concurrency:
-  group: ${{ github.workflow }}-${{ github.ref }}
+  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 jobs:
@@ -25,6 +27,7 @@ jobs:
     uses: rapidsai/shared-action-workflows/.github/workflows/checks.yaml@branch-23.12
     with:
       enable_check_generated_files: false
+      enable_check_pr_job_dependencies: false
   wheel-build-amd64:
     runs-on: linux-amd64-cpu4
     strategy:
@@ -118,3 +121,17 @@ jobs:
         path: ./
     - name: run test_wheel.sh
       run: ci/test_wheel.sh
+  wheel-publish-pynvjitlink:
+    needs:
+      - wheel-test-amd64
+      - wheel-test-arm64
+    # Only trigger this step for tags
+    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/')
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-23.12
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: pynvjitlink

From 064052f2b8aca36d044ba09012b882b0e187214d Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Mon, 27 Nov 2023 16:58:53 -0600
Subject: [PATCH 14/27] Disambiguate cuda major versions in wheel names (#18)

This PR labels the wheel name with a `-cu12` suffix.
---
 ci/test_wheel.sh | 2 +-
 pyproject.toml   | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/ci/test_wheel.sh b/ci/test_wheel.sh
index 1b8c530c..49e8149e 100755
--- a/ci/test_wheel.sh
+++ b/ci/test_wheel.sh
@@ -7,7 +7,7 @@ rapids-logger "Search artifact directory"
 ls ./artifact
 
 rapids-logger "Install wheel"
-pip install --find-links ./artifact pynvjitlink
+pip install --find-links ./artifact pynvjitlink-cu12
 
 rapids-logger "Build Tests"
 cd test_binary_generation && make
diff --git a/pyproject.toml b/pyproject.toml
index 245e7509..c6f530c6 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -3,13 +3,14 @@ cmake.minimum-version = "3.26.4"
 cmake.verbose = true
 ninja.make-fallback = true
 build-dir = "build/{wheel_tag}"
+wheel.packages = ["pynvjitlink"]
 
 [build-system]
 requires = ["scikit-build-core"]
 build-backend = "scikit_build_core.build"
 
 [project]
-name = "pynvjitlink"
+name = "pynvjitlink-cu12"
 version = "0.1.0"
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }

From 10722696bbfb1c2e681634a54a05ac8444edcf5d Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Mon, 27 Nov 2023 17:29:27 -0600
Subject: [PATCH 15/27] Update version to 0.1.1 (#19)

This PR updates the package version to `0.1.1` everywhere.
---
 CMakeLists.txt      | 2 +-
 pynvjitlink/VERSION | 2 +-
 pyproject.toml      | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 18d03db2..5be343c3 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 
 project(
   pynvjitlink
-  VERSION 0.1.0
+  VERSION 0.1.1
   LANGUAGES CXX CUDA
 )
 
diff --git a/pynvjitlink/VERSION b/pynvjitlink/VERSION
index 6e8bf73a..17e51c38 100644
--- a/pynvjitlink/VERSION
+++ b/pynvjitlink/VERSION
@@ -1 +1 @@
-0.1.0
+0.1.1
diff --git a/pyproject.toml b/pyproject.toml
index c6f530c6..36fc2222 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -11,7 +11,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "pynvjitlink-cu12"
-version = "0.1.0"
+version = "0.1.1"
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }
 authors = [

From 032cac5eac183ea544417fd09e93310f51d6e5a3 Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Tue, 28 Nov 2023 11:14:42 -0600
Subject: [PATCH 16/27] Separate pull request and tag workflows (#20)

This PR implements a two-workflow approach where one workflow is
responsible for pull requests and another is responsible for tags and
uploading wheels.

---------

Co-authored-by: Bradley Dice <bdice@bradleydice.com>
---
 .github/workflows/build.yaml                 |  37 ++++++
 .github/workflows/build_and_test_wheels.yaml |  94 +++++++++++++++
 .github/workflows/pr.yaml                    | 120 +------------------
 3 files changed, 136 insertions(+), 115 deletions(-)
 create mode 100644 .github/workflows/build.yaml
 create mode 100644 .github/workflows/build_and_test_wheels.yaml

diff --git a/.github/workflows/build.yaml b/.github/workflows/build.yaml
new file mode 100644
index 00000000..4b28ca63
--- /dev/null
+++ b/.github/workflows/build.yaml
@@ -0,0 +1,37 @@
+# Copyright (c) 2023, NVIDIA CORPORATION.
+name: Release
+
+on:
+  push:
+    tags:
+      - 'v*'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}-release
+  cancel-in-progress: true
+
+jobs:
+  pr-builder:
+    needs:
+      - checks
+      - build-and-test-wheels
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-24.02
+  checks:
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/checks.yaml@branch-24.02
+    with:
+      enable_check_generated_files: false
+  build-and-test-wheels:
+    uses: ./.github/workflows/build_and_test_wheels.yaml
+  wheel-publish-pynvjitlink:
+    needs:
+      - build-and-test-wheels
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-24.02
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: pynvjitlink-cu12
diff --git a/.github/workflows/build_and_test_wheels.yaml b/.github/workflows/build_and_test_wheels.yaml
new file mode 100644
index 00000000..2d63b390
--- /dev/null
+++ b/.github/workflows/build_and_test_wheels.yaml
@@ -0,0 +1,94 @@
+# Copyright (c) 2023, NVIDIA CORPORATION.
+on:
+  workflow_call:
+
+jobs:
+  build-and-test-wheels:
+    needs:
+      - wheel-build-amd64
+      - wheel-build-arm64
+      - wheel-test-amd64
+      - wheel-test-arm64
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-24.02
+  wheel-build-amd64:
+    runs-on: linux-amd64-cpu4
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10"]
+    container:
+      image: "rapidsai/ci-wheel:cuda12.0.1-centos7-py${{ matrix.python-version }}"
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
+    - name: Run build_wheel.sh
+      run: ci/build_wheel.sh
+    - name: Upload Python Wheel
+      uses: actions/upload-artifact@v3
+      with:
+        path: ./wheel-build
+  wheel-build-arm64:
+    runs-on: linux-arm64-cpu4
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10"]
+    container:
+      image: "rapidsai/ci-wheel:cuda12.0.1-rockylinux8-py${{ matrix.python-version }}"
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
+    - name: Run build_wheel.sh
+      run: ci/build_wheel.sh
+    - name: Upload Python Wheel
+      uses: actions/upload-artifact@v3
+      with:
+        path: ./wheel-build
+  wheel-test-amd64:
+    needs:
+      - wheel-build-amd64
+    runs-on: linux-amd64-gpu-v100-latest-1
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10"]
+        linux-version: ["ubuntu20.04", "ubuntu18.04"]
+    container:
+      image: "rapidsai/citestwheel:cuda12.0.1-${{ matrix.linux-version }}-py${{ matrix.python-version }}"
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
+    - name: Download Python Wheel
+      uses: actions/download-artifact@v3
+      with:
+        path: ./
+    - name: Run test_wheel.sh
+      run: ci/test_wheel.sh
+  wheel-test-arm64:
+    needs:
+      - wheel-build-arm64
+    runs-on: linux-arm64-gpu-a100-latest-1
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10"]
+    container:
+      image: "rapidsai/citestwheel:cuda12.0.1-ubuntu20.04-py${{ matrix.python-version }}"
+      env:
+        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v3
+      with:
+        fetch-depth: 0
+    - name: Download Python Wheel
+      uses: actions/download-artifact@v3
+      with:
+        path: ./
+    - name: Run test_wheel.sh
+      run: ci/test_wheel.sh
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index 533fa9f8..b2709fd0 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -5,8 +5,6 @@ on:
   push:
     branches:
       - "pull-request/[0-9]+"
-    tags:
-      - 'v*'
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name }}
@@ -16,122 +14,14 @@ jobs:
   pr-builder:
     needs:
       - checks
-      - wheel-build-amd64
-      - wheel-build-arm64
-      - wheel-test-amd64
-      - wheel-test-arm64
+      - build-and-test-wheels
     secrets: inherit
-    uses: rapidsai/shared-action-workflows/.github/workflows/pr-builder.yaml@branch-23.12
+    uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-24.02
   checks:
     secrets: inherit
-    uses: rapidsai/shared-action-workflows/.github/workflows/checks.yaml@branch-23.12
+    uses: rapidsai/shared-workflows/.github/workflows/checks.yaml@branch-24.02
     with:
       enable_check_generated_files: false
       enable_check_pr_job_dependencies: false
-  wheel-build-amd64:
-    runs-on: linux-amd64-cpu4
-    strategy:
-      matrix:
-        python-version: ["3.9", "3.10"]
-    container:
-      image: "rapidsai/ci-wheel:cuda12.0.1-centos7-py${{ matrix.python-version }}"
-    steps:
-    - name: Checkout code
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-    - name: Get PR Info
-      id: get-pr-info
-      uses: rapidsai/shared-actions/get-pr-info@main
-    - name: Run build_wheel.sh
-      run: ci/build_wheel.sh
-    - name: Upload Python Wheel
-      uses: actions/upload-artifact@v3
-      with:
-        path: ./wheel-build
-  wheel-build-arm64:
-    runs-on: linux-arm64-cpu4
-    strategy:
-      matrix:
-        python-version: ["3.9", "3.10"]
-    container:
-      image: "rapidsai/ci-wheel:cuda12.0.1-rockylinux8-py${{ matrix.python-version }}"
-    steps:
-    - name: Checkout code
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-    - name: Get PR Info
-      id: get-pr-info
-      uses: rapidsai/shared-actions/get-pr-info@main
-    - name: Run build_wheel.sh
-      run: ci/build_wheel.sh
-    - name: Upload Python Wheel
-      uses: actions/upload-artifact@v3
-      with:
-        path: ./wheel-build
-  wheel-test-amd64:
-    needs:
-      - wheel-build-amd64
-    runs-on: linux-amd64-gpu-v100-latest-1
-    strategy:
-      matrix:
-        python-version: ["3.9", "3.10"]
-        linux-version: ["ubuntu20.04", "ubuntu18.04"]
-    container:
-      image: "rapidsai/citestwheel:cuda12.0.1-${{ matrix.linux-version }}-py${{ matrix.python-version }}"
-      env:
-        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
-    steps:
-    - name: Checkout code
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-    - name: Get PR Info
-      id: get-pr-info
-      uses: rapidsai/shared-actions/get-pr-info@main
-    - name: Download Python Wheel
-      uses: actions/download-artifact@v3
-      with:
-        path: ./
-    - name: Run test_wheel.sh
-      run: ci/test_wheel.sh
-  wheel-test-arm64:
-    needs:
-      - wheel-build-arm64
-    runs-on: linux-arm64-gpu-a100-latest-1
-    strategy:
-      matrix:
-        python-version: ["3.9", "3.10"]
-    container:
-      image: "rapidsai/citestwheel:cuda12.0.1-ubuntu20.04-py${{ matrix.python-version }}"
-      env:
-        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
-    steps:
-    - name: Checkout code
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-    - name: Get PR Info
-      id: get-pr-info
-      uses: rapidsai/shared-actions/get-pr-info@main
-    - name: Download Python Wheel
-      uses: actions/download-artifact@v3
-      with:
-        path: ./
-    - name: run test_wheel.sh
-      run: ci/test_wheel.sh
-  wheel-publish-pynvjitlink:
-    needs:
-      - wheel-test-amd64
-      - wheel-test-arm64
-    # Only trigger this step for tags
-    if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/')
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-23.12
-    with:
-      build_type: ${{ inputs.build_type || 'branch' }}
-      branch: ${{ inputs.branch }}
-      sha: ${{ inputs.sha }}
-      date: ${{ inputs.date }}
-      package-name: pynvjitlink
+  build-and-test-wheels:
+    uses: ./.github/workflows/build_and_test_wheels.yaml

From 89919834eb2e6dfe8ea1334b9386d7ddc6533603 Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Tue, 28 Nov 2023 11:37:29 -0600
Subject: [PATCH 17/27] Update to version 0.1.2 (#22)

This PR bumps the package version to 0.1.2.
---
 CMakeLists.txt      | 2 +-
 pynvjitlink/VERSION | 2 +-
 pyproject.toml      | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 5be343c3..c56e6643 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 
 project(
   pynvjitlink
-  VERSION 0.1.1
+  VERSION 0.1.2
   LANGUAGES CXX CUDA
 )
 
diff --git a/pynvjitlink/VERSION b/pynvjitlink/VERSION
index 17e51c38..d917d3e2 100644
--- a/pynvjitlink/VERSION
+++ b/pynvjitlink/VERSION
@@ -1 +1 @@
-0.1.1
+0.1.2
diff --git a/pyproject.toml b/pyproject.toml
index 36fc2222..bb89f048 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -11,7 +11,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "pynvjitlink-cu12"
-version = "0.1.1"
+version = "0.1.2"
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }
 authors = [

From 32c96b777d655c096a6c61863b27367d5493e1bf Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Thu, 30 Nov 2023 15:51:33 -0600
Subject: [PATCH 18/27] Use RAPIDS workflows for PRs and Releases (#23)

Co-authored-by: AJ Schmidt <ajschmidt8@users.noreply.github.com>
Co-authored-by: AJ Schmidt <aschmidt@nvidia.com>
---
 .github/actions/compute-matrix/action.yaml   | 40 +++++++++
 .github/workflows/build.yaml                 | 39 ++++----
 .github/workflows/build_and_test_wheels.yaml | 94 --------------------
 .github/workflows/pr.yaml                    | 36 ++++++--
 ci/build_wheel.sh                            |  4 +-
 ci/test_wheel.sh                             |  6 +-
 6 files changed, 93 insertions(+), 126 deletions(-)
 create mode 100644 .github/actions/compute-matrix/action.yaml
 delete mode 100644 .github/workflows/build_and_test_wheels.yaml

diff --git a/.github/actions/compute-matrix/action.yaml b/.github/actions/compute-matrix/action.yaml
new file mode 100644
index 00000000..cbdc58c7
--- /dev/null
+++ b/.github/actions/compute-matrix/action.yaml
@@ -0,0 +1,40 @@
+name: Compute matrix
+
+outputs:
+  BUILD_MATRIX:
+    value: ${{ steps.compute-matrix.outputs.BUILD_MATRIX }}
+  TEST_MATRIX:
+    value: ${{ steps.compute-matrix.outputs.TEST_MATRIX }}
+
+runs:
+  using: composite
+  steps:
+    - name: Compute matrix
+      id: compute-matrix
+      shell: bash
+      run: |
+        set -eo pipefail
+
+        export BUILD_MATRIX="
+        - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.9', LINUX_VER: 'centos7' }
+        - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.10', LINUX_VER: 'centos7' }
+        - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.9', LINUX_VER: 'rockylinux8' }
+        - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.10', LINUX_VER: 'rockylinux8' }
+        "
+
+        export TEST_MATRIX="
+          - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.9', LINUX_VER: 'ubuntu18.04', gpu: 'v100', driver: 'latest' }
+          - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.10', LINUX_VER: 'ubuntu18.04', gpu: 'v100', driver: 'latest' }
+          - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.9', LINUX_VER: 'ubuntu20.04', gpu: 'a100', driver: 'latest' }
+          - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.10', LINUX_VER: 'ubuntu20.04', gpu: 'a100', driver: 'latest' }
+        "
+
+        echo "BUILD_MATRIX=$(
+          yq -n -o json 'env(BUILD_MATRIX)' | \
+          jq -c '.' \
+        )" | tee --append "${GITHUB_OUTPUT}"
+
+        echo "TEST_MATRIX=$(
+          yq -n -o json 'env(TEST_MATRIX)' | \
+          jq -c '.' \
+        )" | tee --append "${GITHUB_OUTPUT}"
diff --git a/.github/workflows/build.yaml b/.github/workflows/build.yaml
index 4b28ca63..5b050666 100644
--- a/.github/workflows/build.yaml
+++ b/.github/workflows/build.yaml
@@ -1,5 +1,5 @@
 # Copyright (c) 2023, NVIDIA CORPORATION.
-name: Release
+name: build
 
 on:
   push:
@@ -11,27 +11,20 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  pr-builder:
+  compute-matrix:
+    runs-on: ubuntu-latest
+    outputs:
+      BUILD_MATRIX: ${{ steps.compute-matrix.outputs.BUILD_MATRIX }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Compute Build Matrix
+        id: compute-matrix
+        uses: ./.github/actions/compute-matrix
+  build-wheels:
     needs:
-      - checks
-      - build-and-test-wheels
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-24.02
-  checks:
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/checks.yaml@branch-24.02
+      - compute-matrix
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-24.02
     with:
-      enable_check_generated_files: false
-  build-and-test-wheels:
-    uses: ./.github/workflows/build_and_test_wheels.yaml
-  wheel-publish-pynvjitlink:
-    needs:
-      - build-and-test-wheels
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-24.02
-    with:
-      build_type: ${{ inputs.build_type || 'branch' }}
-      branch: ${{ inputs.branch }}
-      sha: ${{ inputs.sha }}
-      date: ${{ inputs.date }}
-      package-name: pynvjitlink-cu12
+      build_type: branch
+      script: "ci/build_wheel.sh"
+      matrix_filter: ${{ needs.compute-matrix.outputs.BUILD_MATRIX }}
diff --git a/.github/workflows/build_and_test_wheels.yaml b/.github/workflows/build_and_test_wheels.yaml
deleted file mode 100644
index 2d63b390..00000000
--- a/.github/workflows/build_and_test_wheels.yaml
+++ /dev/null
@@ -1,94 +0,0 @@
-# Copyright (c) 2023, NVIDIA CORPORATION.
-on:
-  workflow_call:
-
-jobs:
-  build-and-test-wheels:
-    needs:
-      - wheel-build-amd64
-      - wheel-build-arm64
-      - wheel-test-amd64
-      - wheel-test-arm64
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-24.02
-  wheel-build-amd64:
-    runs-on: linux-amd64-cpu4
-    strategy:
-      matrix:
-        python-version: ["3.9", "3.10"]
-    container:
-      image: "rapidsai/ci-wheel:cuda12.0.1-centos7-py${{ matrix.python-version }}"
-    steps:
-    - name: Checkout code
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-    - name: Run build_wheel.sh
-      run: ci/build_wheel.sh
-    - name: Upload Python Wheel
-      uses: actions/upload-artifact@v3
-      with:
-        path: ./wheel-build
-  wheel-build-arm64:
-    runs-on: linux-arm64-cpu4
-    strategy:
-      matrix:
-        python-version: ["3.9", "3.10"]
-    container:
-      image: "rapidsai/ci-wheel:cuda12.0.1-rockylinux8-py${{ matrix.python-version }}"
-    steps:
-    - name: Checkout code
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-    - name: Run build_wheel.sh
-      run: ci/build_wheel.sh
-    - name: Upload Python Wheel
-      uses: actions/upload-artifact@v3
-      with:
-        path: ./wheel-build
-  wheel-test-amd64:
-    needs:
-      - wheel-build-amd64
-    runs-on: linux-amd64-gpu-v100-latest-1
-    strategy:
-      matrix:
-        python-version: ["3.9", "3.10"]
-        linux-version: ["ubuntu20.04", "ubuntu18.04"]
-    container:
-      image: "rapidsai/citestwheel:cuda12.0.1-${{ matrix.linux-version }}-py${{ matrix.python-version }}"
-      env:
-        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
-    steps:
-    - name: Checkout code
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-    - name: Download Python Wheel
-      uses: actions/download-artifact@v3
-      with:
-        path: ./
-    - name: Run test_wheel.sh
-      run: ci/test_wheel.sh
-  wheel-test-arm64:
-    needs:
-      - wheel-build-arm64
-    runs-on: linux-arm64-gpu-a100-latest-1
-    strategy:
-      matrix:
-        python-version: ["3.9", "3.10"]
-    container:
-      image: "rapidsai/citestwheel:cuda12.0.1-ubuntu20.04-py${{ matrix.python-version }}"
-      env:
-        NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
-    steps:
-    - name: Checkout code
-      uses: actions/checkout@v3
-      with:
-        fetch-depth: 0
-    - name: Download Python Wheel
-      uses: actions/download-artifact@v3
-      with:
-        path: ./
-    - name: Run test_wheel.sh
-      run: ci/test_wheel.sh
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index b2709fd0..0fe57264 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -1,5 +1,5 @@
 # Copyright (c) 2023, NVIDIA CORPORATION.
-name: Build and test
+name: pr
 
 on:
   push:
@@ -14,7 +14,9 @@ jobs:
   pr-builder:
     needs:
       - checks
-      - build-and-test-wheels
+      - compute-matrix
+      - build-wheels
+      - test-wheels
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-24.02
   checks:
@@ -22,6 +24,30 @@ jobs:
     uses: rapidsai/shared-workflows/.github/workflows/checks.yaml@branch-24.02
     with:
       enable_check_generated_files: false
-      enable_check_pr_job_dependencies: false
-  build-and-test-wheels:
-    uses: ./.github/workflows/build_and_test_wheels.yaml
+  compute-matrix:
+    runs-on: ubuntu-latest
+    outputs:
+      BUILD_MATRIX: ${{ steps.compute-matrix.outputs.BUILD_MATRIX }}
+      TEST_MATRIX: ${{ steps.compute-matrix.outputs.TEST_MATRIX }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Compute Build Matrix
+        id: compute-matrix
+        uses: ./.github/actions/compute-matrix
+  build-wheels:
+    needs:
+      - compute-matrix
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-24.02
+    with:
+      build_type: pull-request
+      script: "ci/build_wheel.sh"
+      matrix_filter: ${{ needs.compute-matrix.outputs.BUILD_MATRIX }}
+  test-wheels:
+    needs:
+      - build-wheels
+      - compute-matrix
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@branch-24.02
+    with:
+      build_type: pull-request
+      script: "ci/test_wheel.sh"
+      matrix_filter: ${{ needs.compute-matrix.outputs.TEST_MATRIX }}
diff --git a/ci/build_wheel.sh b/ci/build_wheel.sh
index c15c3131..f1ce9d74 100755
--- a/ci/build_wheel.sh
+++ b/ci/build_wheel.sh
@@ -12,6 +12,8 @@ rapids-logger "Check conda environment"
 conda list
 
 rapids-logger "Build wheel"
-export SCCACHE_S3_NO_CREDENTIALS=1
 mkdir -p ./wheel-build
 pip wheel . --wheel-dir=./wheel-build -vvv
+
+rapids-logger "Upload Wheel"
+rapids-upload-wheels-to-s3 ./wheel-build
diff --git a/ci/test_wheel.sh b/ci/test_wheel.sh
index 49e8149e..53f07346 100755
--- a/ci/test_wheel.sh
+++ b/ci/test_wheel.sh
@@ -3,11 +3,11 @@
 
 set -e
 
-rapids-logger "Search artifact directory"
-ls ./artifact
+rapids-logger "Download Wheel"
+rapids-download-wheels-from-s3 ./wheel-build/
 
 rapids-logger "Install wheel"
-pip install --find-links ./artifact pynvjitlink-cu12
+pip install --find-links ./wheel-build pynvjitlink-cu12
 
 rapids-logger "Build Tests"
 cd test_binary_generation && make

From 0e69e7240a9a21341c595e7fd92be60f985f653e Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Thu, 30 Nov 2023 18:45:22 -0600
Subject: [PATCH 19/27] Update version to `0.1.3` (#24)

This PR bumps the project version to `0.1.3`.
---
 CMakeLists.txt      | 2 +-
 pynvjitlink/VERSION | 2 +-
 pyproject.toml      | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index c56e6643..d900de74 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 
 project(
   pynvjitlink
-  VERSION 0.1.2
+  VERSION 0.1.3
   LANGUAGES CXX CUDA
 )
 
diff --git a/pynvjitlink/VERSION b/pynvjitlink/VERSION
index d917d3e2..b1e80bb2 100644
--- a/pynvjitlink/VERSION
+++ b/pynvjitlink/VERSION
@@ -1 +1 @@
-0.1.2
+0.1.3
diff --git a/pyproject.toml b/pyproject.toml
index bb89f048..44362b54 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -11,7 +11,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "pynvjitlink-cu12"
-version = "0.1.2"
+version = "0.1.3"
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }
 authors = [

From 9fbc64254edcaecec8c17b719edc5246c7dc990a Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Tue, 5 Dec 2023 09:03:54 -0600
Subject: [PATCH 20/27] Add a `wheels-publish` job (#25)

This PR adds a job to our release workflow which uploads release wheels
to the nightly rapids pypi index.

---------

Co-authored-by: Vyas Ramasubramani <vyas.ramasubramani@gmail.com>
---
 .github/workflows/build.yaml | 13 +++++++++++++
 ci/build_wheel.sh            |  2 +-
 ci/test_wheel.sh             |  2 +-
 3 files changed, 15 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/build.yaml b/.github/workflows/build.yaml
index 5b050666..6cf21476 100644
--- a/.github/workflows/build.yaml
+++ b/.github/workflows/build.yaml
@@ -3,6 +3,8 @@ name: build
 
 on:
   push:
+    branches:
+      - main
     tags:
       - 'v*'
 
@@ -28,3 +30,14 @@ jobs:
       build_type: branch
       script: "ci/build_wheel.sh"
       matrix_filter: ${{ needs.compute-matrix.outputs.BUILD_MATRIX }}
+  publish-wheels:
+    needs:
+      - build-wheels
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-24.02
+    with:
+      build_type: ${{ inputs.build_type || 'branch' }}
+      branch: ${{ inputs.branch }}
+      sha: ${{ inputs.sha }}
+      date: ${{ inputs.date }}
+      package-name: pynvjitlink
diff --git a/ci/build_wheel.sh b/ci/build_wheel.sh
index f1ce9d74..4604e5ef 100755
--- a/ci/build_wheel.sh
+++ b/ci/build_wheel.sh
@@ -16,4 +16,4 @@ mkdir -p ./wheel-build
 pip wheel . --wheel-dir=./wheel-build -vvv
 
 rapids-logger "Upload Wheel"
-rapids-upload-wheels-to-s3 ./wheel-build
+RAPIDS_PY_WHEEL_NAME="pynvjitlink-cu12" rapids-upload-wheels-to-s3 ./wheel-build
diff --git a/ci/test_wheel.sh b/ci/test_wheel.sh
index 53f07346..489208df 100755
--- a/ci/test_wheel.sh
+++ b/ci/test_wheel.sh
@@ -4,7 +4,7 @@
 set -e
 
 rapids-logger "Download Wheel"
-rapids-download-wheels-from-s3 ./wheel-build/
+RAPIDS_PY_WHEEL_NAME="pynvjitlink-cu12" rapids-download-wheels-from-s3 ./wheel-build/
 
 rapids-logger "Install wheel"
 pip install --find-links ./wheel-build pynvjitlink-cu12

From 94e0a1e89d0c6aa25617c4beb7330b23e21c6ee4 Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Tue, 5 Dec 2023 09:56:49 -0600
Subject: [PATCH 21/27] Update version to `0.1.4` (#27)

This PR bumps the project version to `0.1.4`.
---
 CMakeLists.txt      | 2 +-
 pynvjitlink/VERSION | 2 +-
 pyproject.toml      | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index d900de74..15ad4520 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 
 project(
   pynvjitlink
-  VERSION 0.1.3
+  VERSION 0.1.4
   LANGUAGES CXX CUDA
 )
 
diff --git a/pynvjitlink/VERSION b/pynvjitlink/VERSION
index b1e80bb2..845639ee 100644
--- a/pynvjitlink/VERSION
+++ b/pynvjitlink/VERSION
@@ -1 +1 @@
-0.1.3
+0.1.4
diff --git a/pyproject.toml b/pyproject.toml
index 44362b54..1d23f61b 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -11,7 +11,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "pynvjitlink-cu12"
-version = "0.1.3"
+version = "0.1.4"
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }
 authors = [

From 988ecf24d8cc26d4f6cb96abc3efdd1633829b80 Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Tue, 5 Dec 2023 10:43:46 -0600
Subject: [PATCH 22/27] Don't upload wheels when merging to main (#28)

This PR changes the release workflow to only upload wheels when a tag is
pushed.
---
 .github/workflows/build.yaml | 2 --
 1 file changed, 2 deletions(-)

diff --git a/.github/workflows/build.yaml b/.github/workflows/build.yaml
index 6cf21476..a8f251c2 100644
--- a/.github/workflows/build.yaml
+++ b/.github/workflows/build.yaml
@@ -3,8 +3,6 @@ name: build
 
 on:
   push:
-    branches:
-      - main
     tags:
       - 'v*'
 

From 670fb678854c619f87386e14d1d58571828aaccf Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Tue, 5 Dec 2023 11:01:58 -0600
Subject: [PATCH 23/27] Update version to `0.1.5` (#29)

This PR bumps the project version to `0.1.5`.
---
 CMakeLists.txt      | 2 +-
 pynvjitlink/VERSION | 2 +-
 pyproject.toml      | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 15ad4520..ee577734 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 
 project(
   pynvjitlink
-  VERSION 0.1.4
+  VERSION 0.1.5
   LANGUAGES CXX CUDA
 )
 
diff --git a/pynvjitlink/VERSION b/pynvjitlink/VERSION
index 845639ee..9faa1b7a 100644
--- a/pynvjitlink/VERSION
+++ b/pynvjitlink/VERSION
@@ -1 +1 @@
-0.1.4
+0.1.5
diff --git a/pyproject.toml b/pyproject.toml
index 1d23f61b..459d4eff 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -11,7 +11,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "pynvjitlink-cu12"
-version = "0.1.4"
+version = "0.1.5"
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }
 authors = [

From d2cd8a2c3f66a24e93fbc6890f492cfa6466d768 Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Wed, 13 Dec 2023 13:49:39 -0600
Subject: [PATCH 24/27] Support Python 3.11 (#30)

With the merge of https://github.com/rapidsai/ci-imgs/pull/96, we might
now be able to build python 3.11 packages using our existing pipelines.
---
 .github/actions/compute-matrix/action.yaml | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/.github/actions/compute-matrix/action.yaml b/.github/actions/compute-matrix/action.yaml
index cbdc58c7..4672fba3 100644
--- a/.github/actions/compute-matrix/action.yaml
+++ b/.github/actions/compute-matrix/action.yaml
@@ -18,15 +18,19 @@ runs:
         export BUILD_MATRIX="
         - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.9', LINUX_VER: 'centos7' }
         - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.10', LINUX_VER: 'centos7' }
+        - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.11', LINUX_VER: 'centos7' }
         - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.9', LINUX_VER: 'rockylinux8' }
         - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.10', LINUX_VER: 'rockylinux8' }
+        - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.11', LINUX_VER: 'rockylinux8' }
         "
 
         export TEST_MATRIX="
           - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.9', LINUX_VER: 'ubuntu18.04', gpu: 'v100', driver: 'latest' }
           - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.10', LINUX_VER: 'ubuntu18.04', gpu: 'v100', driver: 'latest' }
+          - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.11', LINUX_VER: 'ubuntu18.04', gpu: 'v100', driver: 'latest' }
           - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.9', LINUX_VER: 'ubuntu20.04', gpu: 'a100', driver: 'latest' }
           - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.10', LINUX_VER: 'ubuntu20.04', gpu: 'a100', driver: 'latest' }
+          - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.11', LINUX_VER: 'ubuntu20.04', gpu: 'a100', driver: 'latest' }
         "
 
         echo "BUILD_MATRIX=$(

From 56a6d74f365c3e601a503dfd6a58ad4e3fc532c7 Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Tue, 2 Jan 2024 10:38:53 -0600
Subject: [PATCH 25/27] Update version to `0.1.6` (#31)

This PR bumps the project version to `0.1.6`.
---
 CMakeLists.txt      | 2 +-
 pynvjitlink/VERSION | 2 +-
 pyproject.toml      | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index ee577734..47cd7ad0 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -2,7 +2,7 @@ cmake_minimum_required(VERSION 3.26.4 FATAL_ERROR)
 
 project(
   pynvjitlink
-  VERSION 0.1.5
+  VERSION 0.1.6
   LANGUAGES CXX CUDA
 )
 
diff --git a/pynvjitlink/VERSION b/pynvjitlink/VERSION
index 9faa1b7a..c946ee61 100644
--- a/pynvjitlink/VERSION
+++ b/pynvjitlink/VERSION
@@ -1 +1 @@
-0.1.5
+0.1.6
diff --git a/pyproject.toml b/pyproject.toml
index 459d4eff..5c380377 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -11,7 +11,7 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "pynvjitlink-cu12"
-version = "0.1.5"
+version = "0.1.6"
 description = "nvJitLink Python binding"
 readme = { file = "README.md", content-type = "text/markdown" }
 authors = [

From 2e7e967a92ed5e55455452dc4f078c7232848428 Mon Sep 17 00:00:00 2001
From: James Lamb <jaylamb20@gmail.com>
Date: Fri, 5 Jan 2024 15:25:08 -0600
Subject: [PATCH 26/27] [REVIEW] remove references to setup.py in docs (#36)

Proposes removing references to `setup.py` in this project's docs. This
project doesn't use a `setup.py` file.

### How I tested this

On a machine with a couple V100s, ran the following (based on what I see
in this project's CI):

```shell
docker run \
    --rm \
    --gpus all \
    -v $(pwd):/usr/local/pynvjitlink \
    -w /usr/local/pynvjitlink \
    -it rapidsai/ci-wheel:cuda12.0.1-centos7-py3.9 \
    bash

# installed CTK following this:
# https://github.com/rapidsai/pynvjitlink/blob/56a6d74f365c3e601a503dfd6a58ad4e3fc532c7/ci/build_wheel.sh
yum update -y
yum install -y \
    cuda-toolkit-12-3 \
    epel-release \
    nvidia-driver-latest-dkms

# skipped using sccache, just to avoid having to set up credentials
unset \
    CMAKE_C_COMPILER_LAUNCHER \
    CMAKE_CUDA_COMPILER_LAUNCHER \
    CMAKE_CXX_COMPILER_LAUNCHER

# built a wheel and installed it
python -m pip install -e .

# tested
make -C ./test_binary_generation
pip install pytest
pytest -v ./pynvjitlink/tests
```

I saw that build succeed and all tests pass.

### Notes for Reviewers

In case you're wondering "what is this random contribution"... I
recently joined NVIDIA and am working on RAPIDS packaging stuff. Was
just poking around this repo and noticed in this in the README.

Thanks for your time and consideration.
---
 README.md | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/README.md b/README.md
index 7f31947d..1890c51b 100644
--- a/README.md
+++ b/README.md
@@ -10,16 +10,17 @@ work-in-progress.
 TODO
 
 ## Install pynvjitlink
+
 Install with either:
 
-```
-python setup.py develop
+```shell
+python -m pip install -e .
 ```
 
 or
 
-```
-python setup.py install
+```shell
+python -m pip install .
 ```
 
 ### Conda

From ce20b818b7503688fa4a334305ad3e19c5266f90 Mon Sep 17 00:00:00 2001
From: brandon-b-miller <53796099+brandon-b-miller@users.noreply.github.com>
Date: Fri, 5 Jan 2024 16:26:27 -0600
Subject: [PATCH 27/27] Set pipefail in wheel build scripts (#34)

This PR fixes up some CI scripts to actually fail if something goes
wrong, as well as some other minor cleanup.

Closes https://github.com/rapidsai/pynvjitlink/issues/32

---------

Co-authored-by: Bradley Dice <bdice@bradleydice.com>
---
 ci/build_wheel.sh                 | 11 +++--------
 ci/install_latest_cuda_toolkit.sh | 20 ++++++++++++++++++++
 ci/test_wheel.sh                  |  4 ++--
 3 files changed, 25 insertions(+), 10 deletions(-)
 create mode 100755 ci/install_latest_cuda_toolkit.sh

diff --git a/ci/build_wheel.sh b/ci/build_wheel.sh
index 4604e5ef..79d4a923 100755
--- a/ci/build_wheel.sh
+++ b/ci/build_wheel.sh
@@ -1,15 +1,10 @@
 #!/bin/bash
-# Copyright (c) 2023, NVIDIA CORPORATION
+# Copyright (c) 2023-2024, NVIDIA CORPORATION
 
+set -euo pipefail
 
 rapids-logger "Install CUDA Toolkit"
-yum update
-yum install -y epel-release
-yum install -y nvidia-driver-latest-dkms
-yum install -y cuda-toolkit-12-3
-
-rapids-logger "Check conda environment"
-conda list
+source "$(dirname "$0")/install_latest_cuda_toolkit.sh"
 
 rapids-logger "Build wheel"
 mkdir -p ./wheel-build
diff --git a/ci/install_latest_cuda_toolkit.sh b/ci/install_latest_cuda_toolkit.sh
new file mode 100755
index 00000000..ca4a7813
--- /dev/null
+++ b/ci/install_latest_cuda_toolkit.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+# Copyright (c) 2024, NVIDIA CORPORATION
+
+# Installs the latest CUDA Toolkit.
+# Supports CentOS 7 and Rocky Linux 8.
+
+yum update -y
+yum install -y epel-release
+
+OS_ID=$(. /etc/os-release; echo $ID)
+if [ "${OS_ID}" == "centos" ]; then
+    yum install -y nvidia-driver-latest-dkms
+elif [ "${OS_ID}" == "rocky" ]; then
+    yum install -y nvidia-driver
+else
+    echo "Error: OS not detected as CentOS or Rocky Linux. Exiting."
+    exit 1
+fi
+
+yum install -y cuda-toolkit-12-3
diff --git a/ci/test_wheel.sh b/ci/test_wheel.sh
index 489208df..ed329e16 100755
--- a/ci/test_wheel.sh
+++ b/ci/test_wheel.sh
@@ -1,7 +1,7 @@
 #!/bin/bash
-# Copyright (c) 2023, NVIDIA CORPORATION
+# Copyright (c) 2023-2024, NVIDIA CORPORATION
 
-set -e
+set -euo pipefail
 
 rapids-logger "Download Wheel"
 RAPIDS_PY_WHEEL_NAME="pynvjitlink-cu12" rapids-download-wheels-from-s3 ./wheel-build/