bump version, merge pull request #21 from AMYPAD/simplify-build

.github/workflows/test.yml

@@ -44,7 +44,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python: [3.6, 3.9]
+        python: [3.6, '3.10']
     steps:
     - uses: actions/checkout@v2
       with:
@@ -69,7 +69,7 @@ jobs:
     runs-on: [self-hosted, cuda, python]
     strategy:
       matrix:
-        python: [3.6, 3.9]
+        python: [3.6, '3.10']
     steps:
     - uses: actions/checkout@v2
       with:

cuvec/CMakeLists.txt

@@ -46,6 +46,7 @@ if("${CMAKE_BUILD_TYPE}" STREQUAL "")
   set(CMAKE_BUILD_TYPE Release CACHE STRING "Build type" FORCE)
 endif()
 message(STATUS "Build type: ${CMAKE_BUILD_TYPE}")
+
 option(CUVEC_DEBUG "Print out CUDA malloc & free operations" OFF)
 if(CUVEC_DEBUG)
   add_compile_definitions(CUVEC_DEBUG)

cuvec/__init__.py

@@ -21,7 +21,7 @@
     # classes
     'CuVec',
     # functions
-    'dev_sync', 'cu_copy', 'cu_zeros',
+    'dev_set', 'dev_sync', 'cu_copy', 'cu_zeros',
     'copy', 'asarray',
     'zeros', 'ones', 'zeros_like', 'ones_like',
     # data
@@ -33,7 +33,7 @@
     from importlib import resources
 
 try:
-    from .cuvec import dev_sync
+    from .cuvec import dev_set, dev_sync
 except ImportError as err: # pragma: no cover
     from warnings import warn
     warn(str(err), UserWarning)

cuvec/include/pycuvec.cuh

@@ -53,7 +53,27 @@ _PYCUVEC_TPCHR(_CUVEC_HALF, "e", "e");
 #endif
 _PYCUVEC_TPCHR(float, "f", "f");
 _PYCUVEC_TPCHR(double, "d", "d");
+#ifndef CUVEC_DISABLE_CUDA
+/// returns `0` iff last CUDA operation was a success; `-1` and sets exception otherwise
+static int PyHandleError(cudaError_t err, const char *file, int line) {
+  if (err != cudaSuccess) {
+    std::stringstream ss;
+    ss << file << ':' << line << ": " << cudaGetErrorString(err);
+    std::string s = ss.str();
+    PyErr_SetString(PyExc_ValueError, s.c_str());
+    return -1;
+  }
+  return 0;
+}
+#endif // CUVEC_DISABLE_CUDA
 } // namespace cuvec
+#ifndef CUDA_PyErr
+#ifdef CUVEC_DISABLE_CUDA
+#define CUDA_PyErr() (0)
+#else // CUVEC_DISABLE_CUDA
+#define CUDA_PyErr() (cuvec::PyHandleError(cudaGetLastError(), __FILE__, __LINE__))
+#endif // CUVEC_DISABLE_CUDA
+#endif // CUDA_PyErr
 
 /** classes */
 /// class PyCuVec<T>
@@ -86,7 +106,7 @@ template <class T> int PyCuVec_init(PyCuVec<T> *self, PyObject *args, PyObject *
   self->strides[ndim - 1] = (Py_ssize_t)sizeof(T);
   for (int i = ndim - 2; i >= 0; i--) self->strides[i] = self->shape[i + 1] * self->strides[i + 1];
   self->vec.resize(self->shape[0] * (self->strides[0] / sizeof(T)));
-  return 0;
+  return CUDA_PyErr();
 }
 /// __del__
 template <class T> void PyCuVec_dealloc(PyCuVec<T> *self) {
@@ -213,12 +233,13 @@ template <class T> PyCuVec<T> *PyCuVec_zeros(std::vector<Py_ssize_t> shape) {
   self->strides[ndim - 1] = (Py_ssize_t)sizeof(T);
   for (int i = ndim - 2; i >= 0; i--) self->strides[i] = self->shape[i + 1] * self->strides[i + 1];
   self->vec.resize(self->shape[0] * (self->strides[0] / sizeof(T)));
-  return self;
+  return CUDA_PyErr() ? NULL : self;
 }
 template <class T> PyCuVec<T> *PyCuVec_zeros_like(PyCuVec<T> *other) {
   PyCuVec<T> *self = PyCuVec_new<T>();
   if (!self) return NULL;
   self->vec.resize(other->vec.size());
+  if (CUDA_PyErr()) return NULL;
   self->shape = other->shape;
   self->strides = other->strides;
   return self;
@@ -227,6 +248,7 @@ template <class T> PyCuVec<T> *PyCuVec_deepcopy(PyCuVec<T> *other) {
   PyCuVec<T> *self = PyCuVec_new<T>();
   if (!self) return NULL;
   self->vec = other->vec;
+  if (CUDA_PyErr()) return NULL;
   self->shape = other->shape;
   self->strides = other->strides;
   return self;

cuvec/src/cuvec.cu

@@ -12,13 +12,25 @@
 static PyObject *dev_sync(PyObject *self, PyObject *args) {
 #ifndef CUVEC_DISABLE_CUDA
   cudaDeviceSynchronize();
+  if (CUDA_PyErr()) return NULL;
+#endif
+  Py_INCREF(Py_None);
+  return Py_None;
+}
+/// select CUDA device
+static PyObject *dev_set(PyObject *self, PyObject *args) {
+#ifndef CUVEC_DISABLE_CUDA
+  int DEVID = 0;
+  if (!PyArg_ParseTuple(args, "i", &DEVID)) return NULL;
+  cudaSetDevice(DEVID);
+  if (CUDA_PyErr()) return NULL;
 #endif
-
   Py_INCREF(Py_None);
   return Py_None;
 }
 static PyMethodDef cuvec_methods[] = {
     {"dev_sync", dev_sync, METH_NOARGS, "Required before accessing cuvec on host."},
+    {"dev_set", dev_set, METH_VARARGS, "Select CUDA device."},
     {NULL, NULL, 0, NULL} // Sentinel
 };
 

setup.cfg

@@ -13,7 +13,7 @@ author=Casper da Costa-Luis
 author_email[email protected]
 keywords=Python, C, C++, buffer, vector, array, CUDA, CPython, SWIG, extensions, API
 classifiers=
-    Development Status :: 4 - Beta
+    Development Status :: 5 - Production/Stable
     Intended Audience :: Developers
     Environment :: GPU
     Environment :: GPU :: NVIDIA CUDA
@@ -29,6 +29,7 @@ classifiers=
     Programming Language :: Python :: 3.7
     Programming Language :: Python :: 3.8
     Programming Language :: Python :: 3.9
+    Programming Language :: Python :: 3.10
     Programming Language :: Python :: 3 :: Only
     Topic :: Software Development :: Libraries
     Topic :: Software Development :: Libraries :: Python Modules

setup.py

@@ -4,34 +4,28 @@
 import sys
 from pathlib import Path
 
-from setuptools import find_packages, setup
+from setuptools import find_packages
 from setuptools_scm import get_version
+from skbuild import setup
 
 __version__ = get_version(root=".", relative_to=__file__)
 
 logging.basicConfig(level=logging.INFO)
 log = logging.getLogger("cuvec.setup")
 
 build_ver = ".".join(__version__.split('.')[:3]).split(".dev")[0]
-setup_kwargs = {"use_scm_version": True, "packages": find_packages(exclude=["docs", "tests"])}
 cmake_args = [f"-DCUVEC_BUILD_VERSION={build_ver}", f"-DPython3_ROOT_DIR={sys.prefix}"]
-
 try:
     from miutil import cuinfo
     nvcc_arch_raw = map(cuinfo.compute_capability, range(cuinfo.num_devices()))
     nvcc_arches = {"%d%d" % i for i in nvcc_arch_raw if i >= (3, 5)}
     if nvcc_arches:
         cmake_args.append("-DCMAKE_CUDA_ARCHITECTURES=" + ";".join(sorted(nvcc_arches)))
 except Exception as exc:
-    log.warning("Import or CUDA device detection error:\n%s", exc)
-
-try:
-    from skbuild import setup as sksetup
-except ImportError:
-    log.warning("`skbuild.setup` not found: Using `setuptools.setup`")
-    setup(**setup_kwargs) # type: ignore # yapf: disable
-else:
-    for i in (Path(__file__).resolve().parent / "_skbuild").rglob("CMakeCache.txt"):
-        i.write_text(re.sub("^//.*$\n^[^#].*pip-build-env.*$", "", i.read_text(), flags=re.M))
-    sksetup(cmake_source_dir="cuvec", cmake_minimum_required_version="3.18", cmake_args=cmake_args,
-            **setup_kwargs)
+    if "sdist" not in sys.argv or any(i in sys.argv for i in ["build", "bdist", "wheel"]):
+        log.warning("Import or CUDA device detection error:\n%s", exc)
+for i in (Path(__file__).resolve().parent / "_skbuild").rglob("CMakeCache.txt"):
+    i.write_text(re.sub("^//.*$\n^[^#].*pip-build-env.*$", "", i.read_text(), flags=re.M))
+setup(use_scm_version=True, packages=find_packages(exclude=["docs", "tests"]),
+      cmake_source_dir="cuvec", cmake_languages=("C", "CXX"),
+      cmake_minimum_required_version="3.18", cmake_args=cmake_args)