aiter/op_tests/test_aiter_addInp.py at main · drivenets/aiter · 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
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
166
167
168
169
170
171
172
173
174
175
176
# SPDX-License-Identifier: MIT
# Copyright (C) 2024-2026, Advanced Micro Devices, Inc. All rights reserved.

import torch
import aiter
from torch.profiler import profile, ProfilerActivity
from aiter import dtypes
import argparse

parser = argparse.ArgumentParser(
    formatter_class=argparse.RawTextHelpFormatter,
    description="config input of test",
)
parser.add_argument(
    "-s",
    "--shape",
    nargs="*",
    type=dtypes.str2tuple,
    choices=[
        (512,),
        (1280, 232, 256),
        (256, 256),
        (256, 8192),
        (256,),
        (1280, 32, 256),
        (384, 256),
        (384,),
        (65536,),
        (65536, 256),
        (1, 8, 256),
        (512, 256),
        (1280, 532, 256),
    ],
    default=[
        (512,),
        (1280, 232, 256),
        (256, 256),
        (256, 8192),
        (256,),
        (1280, 32, 256),
        (384, 256),
        (384,),
        (65536,),
        (65536, 256),
        (1, 8, 256),
        (512, 256),
        (1280, 532, 256),
    ],
    help="""Input shapes.  Dimensionality must match dimensionality of stride (-st).
    e.g.: -s 1280,232,256""",
)
parser.add_argument(
    "-st",
    "--stride",
    nargs="*",
    type=dtypes.str2tuple,
    choices=[
        (1,),
        (59392, 256, 1),
        (256, 1),
        (8192, 1),
        (1,),
        (8192, 256, 1),
        (256, 1),
        (1,),
        (1,),
        (256, 1),
        (2048, 256, 1),
        (256, 1),
        (136192, 256, 1),
    ],
    default=[
        (1,),
        (59392, 256, 1),
        (256, 1),
        (8192, 1),
        (1,),
        (8192, 256, 1),
        (256, 1),
        (1,),
        (1,),
        (256, 1),
        (2048, 256, 1),
        (256, 1),
        (136192, 256, 1),
    ],
    help="""Input strides. Dimensionality must match dimensionality of shape (-s).
    e.g.: -st 59392,256,1""",
)
# input shape: torch.Size([4096, 64, 160]) (20480, 1, 128)
# other shape: torch.Size([4096, 64, 160]) (10240, 160, 1)

# input shape: torch.Size([4096, 64, 160]) (47360, 1, 296)
# other shape: torch.Size([4096, 64, 160]) (10240, 160, 1)

# shape0 = (4096, 64, 160)
# shape1 = (4096, 64, 160)
# stride0 = (47360, 1, 296)
# # stride1 = (20480, 1, 128)
# stride1 = (10240, 160, 1)

# shape1 = (4096, 200, 64)
# shape0 = (1, 200, 64)
# stride0 = (12800, 64, 1)
# stride1 = (12800, 64, 1)

# shape1 = (144, 1, 160)
# shape0 = (144, 4096, 160)
# stride1 = (160, 160, 1)
# stride0 = (655360, 160, 1)

args = parser.parse_args()

tensors0 = [
    torch.empty_strided(shape, stride, dtype=dtypes.bf16, device="cuda")
    for shape, stride in zip(args.shape, args.stride)
]
tensors1 = [
    torch.empty_strided(shape, stride, dtype=dtypes.bf16, device="cuda")
    for shape, stride in zip(args.shape, args.stride)
]
for tensor in tensors0:
    tensor.copy_(torch.rand_like(tensor))
    # tensor.fill_(1)
for tensor in tensors1:
    tensor.copy_(torch.rand_like(tensor))
    # tensor.fill_(1)

# tensor0 = torch.empty_strided(shape0, stride0, dtype=dtypes.bf16, device='cuda')
# tensor1 = torch.empty_strided(shape1, stride1, dtype=dtypes.bf16, device='cuda')
# # tensor0 = torch.empty_strided(shape0, stride0, dtype=dtypes.fp32, device='cuda')
# # tensor1 = torch.empty_strided(shape1, stride1, dtype=dtypes.fp32, device='cuda')
# random_data0 = torch.rand(shape0)
# # tensor0.copy_(random_data0)
# tensor0.fill_(0)
# random_data1 = torch.rand(shape1)
# # tensor1.copy_(random_data1)
# tensor1.fill_(2)

for tensor0, tensor1 in zip(tensors0, tensors1):
    print("shape:", tensor0.size())
    with profile(
        activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
        profile_memory=True,
        with_stack=True,
        with_modules=True,
        record_shapes=True,
    ) as prof:
        for j in range(100):
            # cache_flush1 = torch.randn(10000, 10000, requires_grad=True, device="cuda", dtype=dtypes.fp32).to(dtypes.i32)
            # result = torch.add_(tensor0, tensor1)
            torch_add_a_ = tensor0.clone()
            torch_add_b_ = tensor1.clone()
            torch_add_a_.add_(torch_add_b_)
            result = torch_add_a_
            # result_con = result.contiguous()
    print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=10))

    with profile(
        activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA],
        profile_memory=True,
        with_stack=True,
        with_modules=True,
        record_shapes=True,
    ) as prof:
        for j in range(100):
            # cache_flush1 = torch.randn(10000, 10000, requires_grad=True, device="cuda", dtype=dtypes.fp32).to(dtypes.i32)
            # output = torch.empty_like(tensor1)
            aiter_add_a_ = tensor0.clone()
            aiter_add_b_ = tensor1.clone()
            output = aiter.add_(aiter_add_a_, aiter_add_b_)
    print(prof.key_averages().table(sort_by="cuda_time_total", row_limit=10))

    print(torch.equal(result, output))
    # print("result:", result)
    # print("output:", output)