Skip to content

GitLab

Switch branch/tag
Find file Normal viewHistoryPermalink
test_tilelang_language_alloc.py 3.58 KB
Newer Older
root's avatar
init  
root 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
154
155
156
157
158
159
160
161
162
163
164
165 
import tilelang.testing


def alloc_var(
    N,
    block_N,
    dtype,
):
    import tilelang.language as T

    @T.prim_func
    def main(
            A: T.Tensor((N,), dtype),
            B: T.Tensor((N,), dtype),
    ):
        with T.Kernel(T.ceildiv(N, block_N), threads=block_N) as bx:
            A_shared = T.alloc_shared([block_N], dtype)
            tmp = T.alloc_var(dtype)
            tmp = 1  # noqa: F841
            T.copy(A[bx * block_N], A_shared)
            T.copy(A_shared, B[bx * block_N])

    return main


def run_alloc_var(
    N,
    block_N,
    dtype,
    min=None,
    max=None,
):
    program = alloc_var(N, block_N, dtype)

    kernel = tilelang.compile(program, out_idx=[1])
    code = kernel.get_kernel_source()
    assert "tmp =" in code


def test_alloc_var():
    run_alloc_var(1024, 128, "float16")


def alloc_var_add(
    N,
    block_N,
    dtype,
):
    import tilelang.language as T

    @T.prim_func
    def main(
            A: T.Tensor((N,), dtype),
            B: T.Tensor((N,), dtype),
    ):
        with T.Kernel(T.ceildiv(N, block_N), threads=block_N) as bx:
            A_shared = T.alloc_shared([block_N], dtype)
            tmp = T.alloc_var(dtype)
            tmp = 1  # noqa: F841
            T.copy(A[bx * block_N], A_shared)
            for i in T.Parallel(block_N):
                A_shared[i] = A_shared[i] + tmp
            T.copy(A_shared, B[bx * block_N])

    return main


def run_alloc_var_add(
    N,
    block_N,
    dtype,
):
    program = alloc_var_add(N, block_N, dtype)

    kernel = tilelang.compile(program, out_idx=[1])
    code = kernel.get_kernel_source()
    assert "tmp =" in code


def test_alloc_var_add():
    run_alloc_var_add(1024, 128, "float16")


def alloc_var_with_initializer(
    N,
    block_N,
    dtype,
    init_value,
):
    import tilelang.language as T

    @T.prim_func
    def main(
            A: T.Tensor((N,), dtype),
            B: T.Tensor((N,), dtype),
    ):
        with T.Kernel(T.ceildiv(N, block_N), threads=block_N) as bx:
            tmp = T.alloc_var(dtype, init_value)
            T.copy(A[bx * block_N], B[bx * block_N])
            for i in T.Parallel(block_N):
                B[bx * block_N + i] = tmp

    return main


def run_alloc_var_with_initializer(
    N,
    block_N,
    dtype,
    init_value,
):
    program = alloc_var_with_initializer(N, block_N, dtype, init_value)

    kernel = tilelang.compile(program, out_idx=[1])
    code = kernel.get_kernel_source()
    print(code)
    assert f"= {init_value};" in code


def test_alloc_var_with_initializer():
    run_alloc_var_with_initializer(256, 64, "int32", 5)


def alloc_multi_vars_with_initializer(
    N,
    block_N,
    dtype,
):
    import tilelang.language as T

    @T.prim_func
    def main(
            A: T.Tensor((N,), dtype),
            B: T.Tensor((N,), dtype),
    ):
        with T.Kernel(T.ceildiv(N, block_N), threads=block_N) as bx:
            tmp0 = T.alloc_var(dtype, 1)
            tmp1 = T.alloc_var(dtype, 2)
            T.copy(A[bx * block_N], B[bx * block_N])
            for i in T.Parallel(block_N):
                B[bx * block_N + i] = tmp0 + tmp1

    return main


def run_alloc_multi_vars_with_initializer(
    N,
    block_N,
    dtype,
):
    program = alloc_multi_vars_with_initializer(N, block_N, dtype)

    kernel = tilelang.compile(program, out_idx=[1])
    code = kernel.get_kernel_source()
    print(code)
    assert code.count("= 1;") == 1
    assert code.count("= 2;") == 1


def test_alloc_multi_vars_with_initializer():
    run_alloc_multi_vars_with_initializer(256, 64, "int32")


if __name__ == "__main__":
    tilelang.testing.main()