Expand description
Kernel launch API.
Implements cudaLaunchKernel, cudaFuncGetAttributes, and
cudaFuncSetAttribute on top of the CUDA Driver API.
§Design
In the CUDA Runtime, kernels are typically invoked via <<<...>>> syntax
which the NVCC compiler rewrites into cudaLaunchKernel calls. Since
OxiCUDA never uses NVCC, callers must use the driver-level module/function
handle pair directly. This module therefore exposes a slightly lower-level
surface that accepts a CudaFunction instead of a raw symbol pointer.
Structs§
- Dim3
- 3-D grid / block dimensions for kernel launches.
- Func
Attributes - Attributes of a compiled kernel function.
Enums§
- Func
Attribute - Attribute selector for
cudaFuncSetAttribute.
Functions§
- func_
get_ attributes - Query attributes of a compiled kernel.
- func_
set_ attribute - Set a kernel attribute.
- launch_
kernel ⚠ - Launch a CUDA kernel.
- module_
get_ function - Get a function handle by name from a loaded module.
- module_
load_ ptx - Load a PTX module from a null-terminated byte string.
- module_
unload - Unload a previously loaded module.
Type Aliases§
- Cuda
Function - A compiled GPU kernel function (alias for the driver’s
CUfunction). - Cuda
Module - A compiled GPU module (alias for the driver’s
CUmodule).