Matmul_benchmark/kernel.hpp at master · JosephChenHub/Matmul_benchmark · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
#pragma once


void cublas_sgemm(const float* d_A, const float *d_B, float* d_C, const int A_ROW, const int A_COL, const int B_COL);

template <typename Dtype>
void matmul_naive_kernel(const Dtype*, const Dtype* , Dtype*,
        const int, const int, const int );

template <typename Dtype>
void matmul_tile_kernel(const Dtype*, const Dtype* , Dtype*,
        const int, const int, const int );

template <typename Dtype>
void matmul_unroll_kernel(const Dtype*, const Dtype* , Dtype*,
        const int, const int, const int );

template <typename Dtype>
void matmul_comopt_kernel(const Dtype*, const Dtype* , Dtype*,
        const int, const int, const int );


template <typename Dtype>
void cuda_transpose(const Dtype*, Dtype*, const int, const int);

//template <typename Dtype>
//void matmul_vectorize_kernel(const Dtype*, const Dtype*, Dtype*,
//        const int, const int, const int );