Remove libcugraphops dependency and add CUDA 12 CI support

ci/release/update-version.sh

@@ -66,7 +66,6 @@ NEXT_SHORT_TAG_PEP440=$(python -c "from setuptools.extern import packaging; prin
 
 DEPENDENCIES=(
 
-  libcugraphops
   libraft
   libraft-headers
   librmm

ci/test_python.sh

@@ -6,12 +6,18 @@ set -euo pipefail
 . /opt/conda/etc/profile.d/conda.sh
 
 ARCH=$(arch)
+EXITCODE=0
 
 if [ "${ARCH}" = "aarch64" ]; then
   rapids-logger "Exiting aarch64 due to no pytorch-cuda"
   exit ${EXITCODE}
 fi
 
+if [ "${RAPIDS_CUDA_VERSION:0:2}" == "12" ]; then
+  rapids-logger "Exiting CUDA 12 due to no pytorch stable yet"
+  exit ${EXITCODE}
+fi
+
 rapids-logger "Generate Python testing dependencies"
 rapids-dependency-file-generator \
   --output conda \
@@ -44,7 +50,6 @@ rapids-mamba-retry install \
 
 rapids-logger "Check GPU usage"
 nvidia-smi
-EXITCODE=0
 trap "EXITCODE=1" ERR
 set +e
 

conda/environments/all_cuda-118_arch-x86_64.yaml

@@ -21,14 +21,10 @@ dependencies:
 - gcc_linux-64=11.*
 - gitpython
 - graphviz
-- gtest>=1.13.0
-- gmock>=1.13.0
 - ipykernel
 - ipython
-- libcugraphops=23.8.*
-- libraft-headers=23.8.*
-- libraft=23.8.*
-- librmm=23.8.*
+- libraft-headers==23.08.*
+- librmm==23.08.*
 - nanobind>=0.2.0
 - nbsphinx
 - nccl
@@ -51,6 +47,4 @@ dependencies:
 - sphinx<6
 - sphinxcontrib-websupport
 - sysroot_linux-64=2.17
-- unzip
-- wget
 name: all_cuda-118_arch-x86_64

conda/environments/all_cuda-120_arch-x86_64.yaml

@@ -0,0 +1,49 @@
+# This file is generated by `rapids-dependency-file-generator`.
+# To make changes, edit ../../dependencies.yaml and run `rapids-dependency-file-generator`.
+channels:
+- rapidsai
+- rapidsai-nightly
+- pytorch
+- conda-forge
+- nvidia
+dependencies:
+- breathe
+- c-compiler
+- clang-tools=16.0.0
+- clangxx=16.0.0
+- cmake>=3.23.1,!=3.25.0
+- cuda-nvcc
+- cuda-nvtx
+- cuda-version=12.0
+- cudatoolkit
+- cudnn=8.4
+- cxx-compiler
+- cython
+- doxygen=1.8.20
+- gcc_linux-64=11.*
+- gitpython
+- graphviz
+- ipykernel
+- ipython
+- libraft-headers==23.08.*
+- librmm==23.08.*
+- nanobind>=0.2.0
+- nbsphinx
+- nccl
+- ninja
+- numpy>=1.17
+- numpydoc
+- pre-commit
+- pydata-sphinx-theme
+- pytest
+- pytest-forked
+- pytest-xdist
+- python>=3.9,<3.11
+- recommonmark
+- scikit-build
+- sphinx-copybutton
+- sphinx-markdown-tables
+- sphinx<6
+- sphinxcontrib-websupport
+- sysroot_linux-64=2.17
+name: all_cuda-120_arch-x86_64

conda/recipes/libwholegraph/conda_build_config.yaml

@@ -5,10 +5,13 @@ cxx_compiler_version:
   - 11
 
 cuda_compiler:
+  - cuda-nvcc
+
+cuda11_compiler:
   - nvcc
 
 cmake_version:
-  - ">=3.23.1,!=3.25.0"
+  - ">=3.26.4"
 
 doxygen_version:
   - ">=1.8.11"

conda/recipes/libwholegraph/meta.yaml

@@ -41,19 +41,28 @@ build:
 requirements:
   build:
     - {{ compiler('c') }}
-    - {{ compiler('cuda') }} {{ cuda_version }}
     - {{ compiler('cxx') }}
+    {% if cuda_major == "11" %}
+    - {{ compiler('cuda11') }} ={{ cuda_version }}
+    {% else %}
+    - {{ compiler('cuda') }}
+    {% endif %}
+    - cuda-version ={{ cuda_version }}
+    {% if cuda_major == "11" %}
+    - cudatoolkit
+    {% else %}
+    - cuda-cudart-dev
+    - cuda-driver-dev
+    {% endif %}
     - cmake {{ cmake_version }}
     - ninja
     - sysroot_{{ target_platform }} {{ sysroot_version }}
   host:
     - cmake {{ cmake_version }}
-    - cuda-nvtx ={{ cuda_version }}
-    - cudatoolkit ={{ cuda_version }}
+    - cuda-version ={{ cuda_version }}
     - doxygen {{ doxygen_version }}
     - gmock {{ gtest_version }}
     - gtest {{ gtest_version }}
-    - libcugraphops ={{ minor_version }}
     - libraft ={{ minor_version }}
     - libraft-headers ={{ minor_version }}
     - librmm ={{ minor_version }}
@@ -72,9 +81,19 @@ outputs:
     requirements:
       build:
         - cmake {{ cmake_version }}
+      host:
+        - cuda-version ={{ cuda_version }}
+        {% if cuda_major == "11" %}
+        - cudatoolkit
+        {% else %}
+        - cuda-cudart-dev
+        - cuda-driver-dev
+        {% endif %}
       run:
-        - cudatoolkit {{ cuda_spec }}
-        - libcugraphops ={{ minor_version }}
+        - cuda-version ={{ cuda_version }}
+        {% if cuda_major == "11" %}
+        - cudatoolkit
+        {% endif %}
         - libraft ={{ minor_version }}
         - libraft-headers ={{ minor_version }}
         - librmm ={{ minor_version }}
@@ -91,13 +110,29 @@ outputs:
       number: {{ GIT_DESCRIBE_NUMBER }}
       string: cuda{{ cuda_major }}_{{ date_string }}_{{ GIT_DESCRIBE_HASH }}_{{ GIT_DESCRIBE_NUMBER }}
       ignore_run_exports_from:
+        {% if cuda_major == "11" %}
+        - {{ compiler('cuda11') }} ={{ cuda_version }}
+        {% else %}
         - {{ compiler('cuda') }}
+        {% endif %}
+        - cuda-version ={{ cuda_version }}
     requirements:
       build:
         - cmake {{ cmake_version }}
+      host:
+        - cuda-version ={{ cuda_version }}
+        {% if cuda_major == "11" %}
+        - cudatoolkit
+        {% else %}
+        - cuda-cudart-dev
+        - cuda-driver-dev
+        {% endif %}
       run:
         - {{ pin_subpackage('libwholegraph', exact=True) }}
-        - cudatoolkit {{ cuda_spec }}
+        - cuda-version ={{ cuda_version }}
+        {% if cuda_major == "11" %}
+        - cudatoolkit
+        {% endif %}
         - gmock {{ gtest_version }}
         - gtest {{ gtest_version }}
     about:

conda/recipes/pylibwholegraph/conda_build_config.yaml

@@ -5,10 +5,13 @@ cxx_compiler_version:
   - 11
 
 cuda_compiler:
+  - cuda-nvcc
+
+cuda11_compiler:
   - nvcc
 
 cmake_version:
-  - ">=3.23.1,!=3.25.0"
+  - ">=3.26.4"
 
 scikit_build_version:
   - ">=0.13.1"

conda/recipes/pylibwholegraph/meta.yaml

@@ -35,13 +35,23 @@ build:
     - SCCACHE_S3_USE_SSL
     - SCCACHE_S3_NO_CREDENTIALS
   ignore_run_exports_from:
+    {% if cuda_major == "11" %}
+    - {{ compiler('cuda11') }} ={{ cuda_version }}
+    {% else %}
     - {{ compiler('cuda') }}
+    {% endif %}
+    - cuda-version ={{ cuda_version }}
 
 requirements:
   build:
     - {{ compiler('c') }}
-    - {{ compiler('cuda') }} {{ cuda_version }}
     - {{ compiler('cxx') }}
+    {% if cuda_major == "11" %}
+    - {{ compiler('cuda11') }} ={{ cuda_version }}
+    {% else %}
+    - {{ compiler('cuda') }}
+    {% endif %}
+    - cuda-version ={{ cuda_version }}
     - cmake {{ cmake_version }}
     - ninja
     - nccl
@@ -51,18 +61,22 @@ requirements:
     - doxygen =1.8.20
     - sysroot_{{ target_platform }} {{ sysroot_version }}
   host:
-    - cudatoolkit ={{ cuda_version }}
+    - cuda-version ={{ cuda_version }}
     - libwholegraph ={{ version }}
     - nanobind >=0.2.0
     - python
     - scikit-build {{ scikit_build_version }}
   run:
     - {{ compiler('c') }}
-    - {{ compiler('cuda') }} {{ cuda_version }}
     - {{ compiler('cxx') }}
+    {% if cuda_major == "11" %}
+    - {{ compiler('cuda11') }} ={{ cuda_version }}
+    {% else %}
+    - {{ compiler('cuda') }}
+    {% endif %}
+    - cuda-version ={{ cuda_version }}
     - cmake {{ cmake_version }}
     - ninja
-    - {{ pin_compatible('cudatoolkit', max_pin='x', min_pin='x') }}
     - libwholegraph ={{ version }}
     - python
 

dependencies.yaml

@@ -3,7 +3,7 @@ files:
   all:
     output: [conda]
     matrix:
-      cuda: ["11.5", "11.8"]
+      cuda: ["11.8", "12.0"]
       arch: [x86_64]
     includes:
       - checks
@@ -98,6 +98,11 @@ dependencies:
               cuda: "11.8"
             packages:
               - nvcc_linux-aarch64=11.8
+          - matrix:
+              cuda: "12.0"
+            packages:
+              - cuda-version=12.0
+              - cuda-nvcc
   cudatoolkit:
     specific:
       - output_types: conda
@@ -122,6 +127,11 @@ dependencies:
             packages:
               - cudatoolkit=11.8
               - cuda-nvtx=11.8
+          - matrix:
+              cuda: "12.0"
+            packages:
+              - cuda-version=12.0
+              - cuda-nvtx
   checks:
     common:
       - output_types: [conda, requirements]

python/pylibwholegraph/pylibwholegraph/binding/wholememory_binding.pyx

@@ -351,13 +351,13 @@ cdef class GlobalContextWrapper:
         if output_global_context:
             self.output_global_context = <PyObject *> output_global_context
             Py_INCREF(self.output_global_context)
-        self.env_func.temporary_fns.create_memory_context_fn = &python_cb_wrapper_temp_create_context
-        self.env_func.temporary_fns.destroy_memory_context_fn = &python_cb_wrapper_temp_destroy_context
-        self.env_func.temporary_fns.malloc_fn = &python_cb_wrapper_temp_malloc
-        self.env_func.temporary_fns.free_fn = &python_cb_wrapper_temp_free
+        self.env_func.temporary_fns.create_memory_context_fn = <wholememory_create_memory_context_func_t>&python_cb_wrapper_temp_create_context
+        self.env_func.temporary_fns.destroy_memory_context_fn = <wholememory_destroy_memory_context_func_t>&python_cb_wrapper_temp_destroy_context
+        self.env_func.temporary_fns.malloc_fn = <wholememory_malloc_func_t>&python_cb_wrapper_temp_malloc
+        self.env_func.temporary_fns.free_fn = <wholememory_free_func_t>&python_cb_wrapper_temp_free
         self.env_func.temporary_fns.global_context = <PyObject *> self
-        self.env_func.output_fns.malloc_fn = &python_cb_wrapper_output_malloc
-        self.env_func.output_fns.free_fn = &python_cb_wrapper_output_free
+        self.env_func.output_fns.malloc_fn = <wholememory_malloc_func_t>&python_cb_wrapper_output_malloc
+        self.env_func.output_fns.free_fn = <wholememory_free_func_t>&python_cb_wrapper_output_free
         self.env_func.output_fns.global_context = <PyObject *> self
 
     cpdef int64_t get_env_fns(self):
@@ -952,7 +952,7 @@ cdef class PyWholeMemoryUniqueID:
         dlm_tensor.manager_ctx = <void *> self
         cpython.Py_INCREF(self)
         dlm_tensor.deleter = deleter
-        return cpython.PyCapsule_New(dlm_tensor, 'dltensor', pycapsule_deleter)
+        return cpython.PyCapsule_New(dlm_tensor, 'dltensor', <cpython.PyCapsule_Destructor>&pycapsule_deleter)
 
     def __dlpack_device__(self):
         return (kDLCPU, 0)
@@ -1173,7 +1173,7 @@ cdef class PyWholeMemoryFlattenDlpack:
         dlm_tensor.manager_ctx = <void *> self
         cpython.Py_INCREF(self)
         dlm_tensor.deleter = deleter
-        return cpython.PyCapsule_New(dlm_tensor, 'dltensor', pycapsule_deleter)
+        return cpython.PyCapsule_New(dlm_tensor, 'dltensor', <cpython.PyCapsule_Destructor> &pycapsule_deleter)
 
     def __dlpack_device__(self):
         if self.device_type == MlHost: