The CUDA Driver API can be used immediately, without any configuration.
LWJGL does not bundle CUDA Toolkit libraries. For the bindings of such a library to work, LWJGL must be able to find the corresponding shared libraries
that export the library's API. By default, LWJGL scans the standard CUDA Toolkit installation directory and loads the shared libraries of the CUDA Toolkit
with the greatest version it can detect. This process can be overridden with the Configuration#CUDA_TOOLKIT_VERSION and
Configuration#CUDA_TOOLKIT_PATH options.
In the case of a custom CUDA Toolkit library distribution:
-Djava.library.path or -Dorg.lwjgl.librarypath.
Note that, on Windows, the shared libraries should be renamed to drop the <bits>_<version> suffix. For example
nvgraph64_100.dll should be renamed to just nvgraph.dll. On Linux & macOS, either rename the shared libraries or include the unversioned
symbolic links.
Configuration options to set the name or path of each shared library, e.g. Configuration#CUDA_NVRTC_LIBRARY_NAME.| Modifier and Type | Field and Description |
|---|---|
| private static final String | |
| private static final String | |
| private static final Pattern |
| Access | Constructor and Description |
|---|---|
| private |
| Modifier and Type | Method and Description |
|---|---|
| pack-priv static String | |
| pack-priv static String | |
| pack-priv static String | |
| private static String | |
| pack-priv static SharedLibrary | |
| private static Path | |
| public static boolean | Returns: true if per-thread synchronization is supported, false if only the legacy default stream is supportedReturns true if the CUDA driver supports per-thread synchronization for the default stream. |
| TOOLKIT_DEFAULT_ROOT | back to summary |
|---|---|
| private static final String TOOLKIT_DEFAULT_ROOT | |
| TOOLKIT_PREFIX | back to summary |
|---|---|
| private static final String TOOLKIT_PREFIX | |
| TOOLKIT_REGEX | back to summary |
|---|---|
| private static final Pattern TOOLKIT_REGEX | |
| CUDA | back to summary |
|---|---|
| private CUDA() | |
| __CUDA_API_PTDS | back to summary |
|---|---|
| pack-priv static String __CUDA_API_PTDS(String api) | |
| __CUDA_API_PTSZ | back to summary |
|---|---|
| pack-priv static String __CUDA_API_PTSZ(String api) | |
| __CUDA_API_VERSION | back to summary |
|---|---|
| pack-priv static String __CUDA_API_VERSION(String api, int version) | |
| getToolkitLibrary | back to summary |
|---|---|
| private static String getToolkitLibrary(String name) | |
| getToolkitLibrary | back to summary |
|---|---|
| pack-priv static SharedLibrary getToolkitLibrary(Class<?> context, Configuration<String> option, String name) | |
| getToolkitPath | back to summary |
|---|---|
| private static Path getToolkitPath() | |
| isPerThreadDefaultStreamSupported | back to summary |
|---|---|
| public static boolean isPerThreadDefaultStreamSupported() Returns true if the CUDA driver supports per-thread synchronization for the default stream. CUDA defaults to the legacy default stream. If this method returns
| |