cgemm_bf16.cpp 4.4 KB
Newer Older
myamlak's avatar
myamlak committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
#include <algorithm>
#include <cstdlib>
#include <half.hpp>
#include <iostream>
#include <numeric>
#include <tuple>
#include <vector>

#include "cgemm_util.hpp"
#include "config.hpp"
#include "print.hpp"
#include "device.hpp"
#include "host_tensor.hpp"
#include "host_tensor_generator.hpp"
#include "device_tensor.hpp"
#include "device_cgemm_4gemm_xdl_cshuffle.hpp"
#include "element_wise_operation.hpp"
#include "reference_cgemm.hpp"
#include "gemm_specialization.hpp"

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

using DeviceCGemmNoOpPtr =
myamlak's avatar
myamlak committed
24
25
26
    ck::tensor_operation::device::DeviceCGemmPtr<ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough>;
myamlak's avatar
myamlak committed
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50

namespace ck {
namespace tensor_operation {
namespace device {
namespace device_cgemm_instance {
void add_device_cgemm_4gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(
    std::vector<DeviceCGemmNoOpPtr>&);
void add_device_cgemm_4gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(
    std::vector<DeviceCGemmNoOpPtr>&);
void add_device_cgemm_4gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(
    std::vector<DeviceCGemmNoOpPtr>&);
void add_device_cgemm_4gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(
    std::vector<DeviceCGemmNoOpPtr>&);
} // namespace device_cgemm_instance
} // namespace device
} // namespace tensor_operation
} // namespace ck

int main()
{
    using RowMajor    = ck::tensor_layout::gemm::RowMajor;
    using ColumnMajor = ck::tensor_layout::gemm::ColumnMajor;

    bool res = true;
myamlak's avatar
myamlak committed
51
    std::vector<DeviceCGemmNoOpPtr> cgemmPtrs;
myamlak's avatar
myamlak committed
52

myamlak's avatar
myamlak committed
53
    ck::tensor_operation::device::device_cgemm_instance::
myamlak's avatar
myamlak committed
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
        add_device_cgemm_4gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(cgemmPtrs);

    for(auto& cgemmPtr : cgemmPtrs)
    {
        res &= ck::cgemm_util::TestCGemmBF16<DeviceCGemmNoOpPtr,
                                             ColumnMajor,
                                             RowMajor,
                                             RowMajor,
                                             PassThrough,
                                             PassThrough,
                                             PassThrough>{}(cgemmPtr);
    }

    cgemmPtrs.clear();
    ck::tensor_operation::device::device_cgemm_instance::
        add_device_cgemm_4gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(cgemmPtrs);

    for(auto& cgemmPtr : cgemmPtrs)
    {
        res &= ck::cgemm_util::TestCGemmBF16<DeviceCGemmNoOpPtr,
                                             ColumnMajor,
                                             ColumnMajor,
                                             RowMajor,
                                             PassThrough,
                                             PassThrough,
myamlak's avatar
myamlak committed
79
                                             PassThrough>{}(cgemmPtr);
myamlak's avatar
myamlak committed
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
    }

    cgemmPtrs.clear();
    ck::tensor_operation::device::device_cgemm_instance::
        add_device_cgemm_4gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(cgemmPtrs);

    for(auto& cgemmPtr : cgemmPtrs)
    {
        res &= ck::cgemm_util::TestCGemmBF16<DeviceCGemmNoOpPtr,
                                             RowMajor,
                                             RowMajor,
                                             RowMajor,
                                             PassThrough,
                                             PassThrough,
                                             PassThrough>{}(cgemmPtr);
    }

    cgemmPtrs.clear();
    ck::tensor_operation::device::device_cgemm_instance::
        add_device_cgemm_4gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(cgemmPtrs);

    for(auto& cgemmPtr : cgemmPtrs)
    {
        res &= ck::cgemm_util::TestCGemmBF16<DeviceCGemmNoOpPtr,
                                             RowMajor,
                                             ColumnMajor,
                                             RowMajor,
                                             PassThrough,
                                             PassThrough,
                                             PassThrough>{}(cgemmPtr);
    }

    std::cout << "TestCGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
    return res ? 0 : 1;
}