profile_grouped_gemm.cpp 8 KB
Newer Older
zjing14's avatar
zjing14 committed
1
2
3
4
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
Chao Liu's avatar
Chao Liu committed
5
6

#include "profiler/include/profile_grouped_gemm_impl.hpp"
zjing14's avatar
zjing14 committed
7

8
enum struct GemmMatrixLayout
zjing14's avatar
zjing14 committed
9
10
11
12
13
14
15
16
17
18
19
{
    MK_KN_MN, // 0
    MK_NK_MN, // 1
    KM_KN_MN, // 2
    KM_NK_MN, // 3
    MK_KN_NM, // 4
    MK_NK_NM, // 5
    KM_KN_NM, // 6
    KM_NK_NM, // 7
};

20
enum struct GemmDataType
zjing14's avatar
zjing14 committed
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
{
    F32_F32_F32,    // 0
    F16_F16_F16,    // 1
    BF16_BF16_BF16, // 2
    INT8_INT8_INT8, // 3
};

std::vector<int> argToIntArray(char* input)
{
    std::vector<int> out;

    std::istringstream in(input);

    std::string item;

    while(std::getline(in, item, ','))
    {
        out.push_back(std::stoi(item));
    }

    return out;
}

int profile_grouped_gemm(int argc, char* argv[])
{
    if(!(argc == 14))
    {
        printf("arg1: tensor operation (grouped_gemm: Grouped GEMM)\n");
        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n");
        printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
        printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
        printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
        printf("                     3: A[k, m] * B[n, k] = C[m, n])\n");
        printf("arg4: verification (0: no; 1: yes)\n");
        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
JD's avatar
JD committed
56
57
        printf("arg6: print tensor value (0: no; 1: yes)\n");
        printf("arg7: time kernel (0=n0, 1=yes)\n");
zjing14's avatar
zjing14 committed
58
59
60
61
62
        printf("arg8 to 13: Ms, Ns, Ks, StrideAs, StrideBs, StrideCs (e.g., 256,256 128,128 64,64 "
               "64,64 64,64 128,128)\n");
        exit(1);
    }

63
64
    const auto data_type       = static_cast<GemmDataType>(std::stoi(argv[2]));
    const auto layout          = static_cast<GemmMatrixLayout>(std::stoi(argv[3]));
zjing14's avatar
zjing14 committed
65
66
67
    const bool do_verification = std::stoi(argv[4]);
    const int init_method      = std::stoi(argv[5]);
    const bool do_log          = std::stoi(argv[6]);
JD's avatar
JD committed
68
    const bool time_kernel     = std::stoi(argv[7]);
zjing14's avatar
zjing14 committed
69
70
71
72
73
74
75
76
77
78
79
80

    const auto Ms = argToIntArray(argv[8]);
    const auto Ns = argToIntArray(argv[9]);
    const auto Ks = argToIntArray(argv[10]);

    const auto StrideAs = argToIntArray(argv[11]);
    const auto StrideBs = argToIntArray(argv[12]);
    const auto StrideCs = argToIntArray(argv[13]);

    if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
    {
        ck::profiler::profile_grouped_gemm_impl<ck::half_t,
81
                                                ck::half_t,
zjing14's avatar
zjing14 committed
82
83
84
85
86
87
88
                                                ck::half_t,
                                                ck::half_t,
                                                ck::tensor_layout::gemm::RowMajor,
                                                ck::tensor_layout::gemm::RowMajor,
                                                ck::tensor_layout::gemm::RowMajor>(do_verification,
                                                                                   init_method,
                                                                                   do_log,
JD's avatar
JD committed
89
                                                                                   time_kernel,
zjing14's avatar
zjing14 committed
90
91
92
93
94
95
96
97
98
99
                                                                                   Ms,
                                                                                   Ns,
                                                                                   Ks,
                                                                                   StrideAs,
                                                                                   StrideBs,
                                                                                   StrideCs);
    }
    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
    {
        ck::profiler::profile_grouped_gemm_impl<ck::half_t,
100
                                                ck::half_t,
zjing14's avatar
zjing14 committed
101
102
103
104
105
106
107
                                                ck::half_t,
                                                ck::half_t,
                                                ck::tensor_layout::gemm::RowMajor,
                                                ck::tensor_layout::gemm::ColumnMajor,
                                                ck::tensor_layout::gemm::RowMajor>(do_verification,
                                                                                   init_method,
                                                                                   do_log,
JD's avatar
JD committed
108
                                                                                   time_kernel,
zjing14's avatar
zjing14 committed
109
110
111
112
113
114
115
116
117
118
                                                                                   Ms,
                                                                                   Ns,
                                                                                   Ks,
                                                                                   StrideAs,
                                                                                   StrideBs,
                                                                                   StrideCs);
    }
    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
    {
        ck::profiler::profile_grouped_gemm_impl<ck::half_t,
119
                                                ck::half_t,
zjing14's avatar
zjing14 committed
120
121
122
123
124
125
126
                                                ck::half_t,
                                                ck::half_t,
                                                ck::tensor_layout::gemm::ColumnMajor,
                                                ck::tensor_layout::gemm::RowMajor,
                                                ck::tensor_layout::gemm::RowMajor>(do_verification,
                                                                                   init_method,
                                                                                   do_log,
JD's avatar
JD committed
127
                                                                                   time_kernel,
zjing14's avatar
zjing14 committed
128
129
130
131
132
133
134
135
136
137
                                                                                   Ms,
                                                                                   Ns,
                                                                                   Ks,
                                                                                   StrideAs,
                                                                                   StrideBs,
                                                                                   StrideCs);
    }
    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
    {
        ck::profiler::profile_grouped_gemm_impl<ck::half_t,
138
                                                ck::half_t,
zjing14's avatar
zjing14 committed
139
140
141
142
143
144
145
                                                ck::half_t,
                                                ck::half_t,
                                                ck::tensor_layout::gemm::ColumnMajor,
                                                ck::tensor_layout::gemm::ColumnMajor,
                                                ck::tensor_layout::gemm::RowMajor>(do_verification,
                                                                                   init_method,
                                                                                   do_log,
JD's avatar
JD committed
146
                                                                                   time_kernel,
zjing14's avatar
zjing14 committed
147
148
149
150
151
152
153
154
155
156
157
158
                                                                                   Ms,
                                                                                   Ns,
                                                                                   Ks,
                                                                                   StrideAs,
                                                                                   StrideBs,
                                                                                   StrideCs);
    }
    else
    {
        throw std::runtime_error("wrong! this GEMM data_type & layout is not implemented");
    }

159
    return 0;
zjing14's avatar
zjing14 committed
160
}