std1.0.0-beta-7 SafeC docs

std :: ml::gpu_rocm

Function std::rocm_to_half_many

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

int rocm_to_half_many(const &Vec<struct RocmBuf> src, const &Vec<struct RocmHalfBuf> dst)

Every src[i] -> dst[i] (lengths dst[i].len / src[i].len as in rocm_to_half / rocm_from_half), 32 buffers per launch: a model's parameters in one or a few launches instead of one per tensor.