std1.0.0-beta-7 SafeC docs

std :: ml::gpu_cuda

Function std::cuda_half_gemm_strided

#include <std/ml/gpu_cuda.h> source · definition

int cuda_half_gemm_strided(
    struct CudaHalfBuf a,
    int ta,
    unsigned long strideA,
    struct CudaHalfBuf b,
    int tb,
    unsigned long strideB,
    struct CudaHalfBuf c,
    unsigned long strideC,
    unsigned long m,
    unsigned long n,
    unsigned long k,
    unsigned long batch,
)

'batch' products c_i = op(a_i) op(b_i) in one cuBLAS call, a_i starting at element i * strideA of a (likewise b, c); a stride of 0 reuses one matrix for every product (e.g. a weight shared by a batch). One dtype.