-
Notifications
You must be signed in to change notification settings - Fork 1.5k
Expand file tree
/
Copy pathconv_bias_relu.py
More file actions
102 lines (78 loc) · 3.35 KB
/
conv_bias_relu.py
File metadata and controls
102 lines (78 loc) · 3.35 KB
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
import torch
from apex import check_cudnn_version_and_warn
import fused_conv_bias_relu
check_cudnn_version_and_warn(__name__, 8400)
class ConvBiasReLU_(torch.autograd.Function):
@staticmethod
@torch.amp.custom_fwd(cast_inputs=torch.half, device_type="cuda")
def forward(ctx, x, weight, bias, padding, stride):
outputs = fused_conv_bias_relu.forward([x, weight, bias], padding, stride)
ctx.save_for_backward(x, weight, outputs[0])
ctx.padding = padding
ctx.stride = stride
return outputs[0]
@staticmethod
@torch.amp.custom_bwd(device_type="cuda")
def backward(ctx, grad_output):
bwd_args = [*ctx.saved_tensors, grad_output]
padding = ctx.padding
stride = ctx.stride
grads = fused_conv_bias_relu.backward(bwd_args, padding, stride)
return grads[0], grads[1], grads[2], None, None
class ConvBiasMaskReLU_(torch.autograd.Function):
@staticmethod
@torch.amp.custom_fwd(cast_inputs=torch.half, device_type="cuda")
def forward(ctx, x, weight, bias, mask, padding, stride):
outputs = fused_conv_bias_relu.forward_mask([x, weight, bias, mask], padding, stride)
ctx.save_for_backward(x, weight, outputs[0])
ctx.padding = padding
ctx.stride = stride
return outputs[0]
@staticmethod
@torch.amp.custom_bwd(device_type="cuda")
def backward(ctx, grad_output):
bwd_args = [*ctx.saved_tensors, grad_output]
padding = ctx.padding
stride = ctx.stride
grads = fused_conv_bias_relu.backward(bwd_args, padding, stride)
return grads[0], grads[1], grads[2], None, None, None
class ConvBias_(torch.autograd.Function):
@staticmethod
@torch.amp.custom_fwd(cast_inputs=torch.half, device_type="cuda")
def forward(ctx, x, weight, bias, padding, stride):
outputs = fused_conv_bias_relu.forward_no_relu([x, weight, bias], padding, stride)
ctx.save_for_backward(x, weight)
ctx.padding = padding
ctx.stride = stride
return outputs[0]
@staticmethod
@torch.amp.custom_bwd(device_type="cuda")
def backward(ctx, grad_output):
bwd_args = [*ctx.saved_tensors, grad_output]
padding = ctx.padding
stride = ctx.stride
grads = fused_conv_bias_relu.backward_no_relu(bwd_args, padding, stride)
return grads[0], grads[1], grads[2], None, None
class ConvFrozenScaleBiasReLU_(torch.autograd.Function):
@staticmethod
@torch.amp.custom_fwd(cast_inputs=torch.half, device_type="cuda")
def forward(ctx, x, weight, scale, bias, padding, stride):
output = fused_conv_bias_relu.forward_cscale_cbias_relu(
[x, weight, scale, bias], padding, stride
)
ctx.save_for_backward(x, weight, scale, output)
ctx.padding = padding
ctx.stride = stride
return output
@staticmethod
@torch.amp.custom_bwd(device_type="cuda")
def backward(ctx, grad_output):
bwd_args = [*ctx.saved_tensors, grad_output]
padding = ctx.padding
stride = ctx.stride
grads = fused_conv_bias_relu.backward_cscale_cbias_relu(bwd_args, padding, stride)
return grads[0], grads[1], None, None, None, None
ConvBiasReLU = ConvBiasReLU_.apply
ConvBiasMaskReLU = ConvBiasMaskReLU_.apply
ConvBias = ConvBias_.apply
ConvFrozenScaleBiasReLU = ConvFrozenScaleBiasReLU_.apply