grain.cuda

CUDA wrapper module

Members

CudaElementType alias CudaElementType(M : CuPtr!T, T) = T
alias CudaElementType(M : CuArray!T, T) = T
alias CudaElementType(M : RefCounted!(CuPtr!T), T) = T
alias CudaElementType(M : RefCounted!(CuArray!T), T) = T: alias to element type of cuda storage

axpy void axpy(const ref CuArray!T x, ref CuArray!T y, T alpha = 1, int incx = 1, int incy = 1): high-level axpy (y = alpha * x + y) wrapper for CuPtr
checkCUDNN void checkCUDNN(cudnnStatus_t err): cudnn error checker
checkCublasErrors void checkCublasErrors(cublasStatus_t err): cublas error checker
checkCudaErrors void checkCudaErrors(CUresult err): cuda error checker
copy void copy(ref CuPtr!T src, ref CuPtr!T dst): deep copy inter device memory without allocation
dup auto dup(ref M m): duplicate cuda memory (deep copy)
empty bool empty(M m): true if length == 0
fill_ ref fill_(ref S storage, V v, size_t N): fill value for N elements from the first position TODO use cudnnSetTensor
fill_ ref fill_(ref S storage, V value): fill value for all the element in device array
global auto global(): global accessor for the cuda module in grain
sum float sum(ref S a): test sum
toHost ref toHost(ref M m, scope ref T[] host): copy device memory to host (maybe reallocate in host)
toHost auto toHost(ref M m, T* host): copy device memory to host (CAUTION: no reallocation here)
toHost auto toHost(ref M m): allocate host memory and copy device memory content
zero_ ref zero_(ref S storage): fill zero for all the element in device array
zeros auto zeros(size_t N): create zero filled N elements array

CuArray struct CuArray(T): sub-region on CuPtr!T
CuModule struct CuModule: cuda module compiled from ptx string
CuPtr struct CuPtr(T): fat pointer in CUDA
Kernel struct Kernel(alias F): cuda function object called by mangled name of C++/D device function F
Launcher struct Launcher(Args...): cuda kernel function launcher with runtime numbers of blocks/threads

isDeviceMemory enum bool isDeviceMemory(T);: trait to identify cuda storage