| | 1 | == Supported Features == |
| | 2 | |
| | 3 | * CUDA kernels with the `__global__` specifier |
| | 4 | * The dim3 struct type |
| | 5 | * Use of the CUDA variables `threadIdx`, `blockIdx`, `gridDim`, and `blockDim` |
| | 6 | * `__syncthreads` |
| | 7 | * Enqueuing multiple kernel calls with the null stream |
| | 8 | * `cudaMalloc` |
| | 9 | * `cudaMemcpy` |
| | 10 | * `cudaFree` |
| | 11 | * `cudaDeviceSynchronize` |
| | 12 | |
| | 13 | == Major Limitations == |
| | 14 | |
| | 15 | * Lacks support for other stream types |
| | 16 | |
| | 17 | === Missing Features === |
| | 18 | |
| | 19 | * Use of the `warpSize` variable |
| | 20 | * Atomic functions (e.g. `atomicAdd`) |
