std1.0.0-beta-7 SafeC docs

std :: ml::gpu_cuda

Function std::cuda_mse_grad

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

int cuda_mse_grad(
    struct CudaBuf y,
    struct CudaBuf t,
    struct CudaBuf dy,
    struct CudaBuf loss1,
)

Fused MSE gradient: dy = 2 (y - t) and loss1[0] = sum((y - t)^2), in one pass.