Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
gemm_fp16.cpp 6.79 KB
Newer Older
rocking5566's avatar
Gemm_c_shuffle (4 layouts) X (fp32 bf16 int8) (#131)  
rocking5566 committed
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
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
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153 
#include <algorithm>
#include <cstdlib>
#include <half.hpp>
#include <iostream>
#include <numeric>
#include <tuple>
#include <vector>

#include "gemm_util.hpp"
#include "config.hpp"
#include "print.hpp"
#include "device.hpp"
#include "host_gemm.hpp"
#include "device_tensor.hpp"
#include "device_gemm_xdl.hpp"
#include "device_gemm_xdl_c_shuffle.hpp"
#include "element_wise_operation.hpp"
#include "gemm_specialization.hpp"

using PassThrough = ck::tensor_operation::element_wise::PassThrough;

using DeviceGemmNoOpPtr =
    ck::tensor_operation::device::DeviceGemmPtr<ck::tensor_operation::element_wise::PassThrough,
                                                ck::tensor_operation::element_wise::PassThrough,
                                                ck::tensor_operation::element_wise::PassThrough>;

namespace ck {
namespace tensor_operation {
namespace device {
namespace device_gemm_instance {
void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);

void add_device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);

void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(std::vector<DeviceGemmNoOpPtr>&);
void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(std::vector<DeviceGemmNoOpPtr>&);

void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
    std::vector<DeviceGemmNoOpPtr>&);
} // namespace device_gemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

int main()
{
    using ADataType = ck::half_t;
    using BDataType = ck::half_t;
    using CDataType = ck::half_t;

    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;

    bool res = true;
    std::vector<DeviceGemmNoOpPtr> gemmPtrs;
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(gemmPtrs);

    for(auto& gemmPtr : gemmPtrs)
    {
        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
                                       ADataType,
                                       BDataType,
                                       CDataType,
                                       ColumnMajor,
                                       RowMajor,
                                       RowMajor,
                                       PassThrough,
                                       PassThrough,
                                       PassThrough>{}(gemmPtr);
    }

    gemmPtrs.clear();
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(gemmPtrs);

    for(auto& gemmPtr : gemmPtrs)
    {
        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
                                       ADataType,
                                       BDataType,
                                       CDataType,
                                       ColumnMajor,
                                       ColumnMajor,
                                       RowMajor,
                                       PassThrough,
                                       PassThrough,
                                       PassThrough>{}(gemmPtr);
    }

    gemmPtrs.clear();
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(gemmPtrs);

    for(auto& gemmPtr : gemmPtrs)
    {
        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
                                       ADataType,
                                       BDataType,
                                       CDataType,
                                       RowMajor,
                                       RowMajor,
                                       RowMajor,
                                       PassThrough,
                                       PassThrough,
                                       PassThrough>{}(gemmPtr);
    }

    gemmPtrs.clear();
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);
    ck::tensor_operation::device::device_gemm_instance::
        add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(gemmPtrs);

    for(auto& gemmPtr : gemmPtrs)
    {
        res &= ck::gemm_util::TestGemm<DeviceGemmNoOpPtr,
                                       ADataType,
                                       BDataType,
                                       CDataType,
                                       RowMajor,
                                       ColumnMajor,
                                       RowMajor,
                                       PassThrough,
                                       PassThrough,
                                       PassThrough>{}(gemmPtr);
    }

    std::cout << "TestGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
rocking5566's avatar
Gemm test return value (#148)  
rocking5566 committed
154 
    return res ? 0 : 1;
rocking5566's avatar
Gemm_c_shuffle (4 layouts) X (fp32 bf16 int8) (#131)  
rocking5566 committed
155 
}