forked from soumith/cudnn.torch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
VolumetricConvolution.lua
97 lines (82 loc) · 3.39 KB
/
VolumetricConvolution.lua
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
local VolumetricConvolution, parent
= torch.class('cudnn.VolumetricConvolution', 'nn.VolumetricConvolution')
local ffi = require 'ffi'
local find = require 'cudnn.find'
local errcheck = cudnn.errcheck
local Convolution = cudnn.SpatialConvolution
-- if you change the configuration of the module manually, call this
function VolumetricConvolution:resetWeightDescriptors()
local desc = {self.nOutputPlane, self.nInputPlane,
self.kT, self.kH, self.kW}
return Convolution.resetWeightDescriptors(self,desc)
end
function VolumetricConvolution:fastest(mode)
return Convolution.fastest(self,mode)
end
function VolumetricConvolution:setMode(fmode, bdmode, bwmode)
return Convolution.setMode(self, fmode, bdmode, bwmode)
end
function VolumetricConvolution:resetMode()
return Convolution.resetMode(self)
end
function VolumetricConvolution:createIODescriptors(input)
if input:dim() == 4 then
input = input:view(1, input:size(1), input:size(2),
input:size(3), input:size(4))
batch = false
end
if Convolution.checkInputChanged(self, input) then
-- create input descriptor
self.iDesc = cudnn.toDescriptor(input)
-- create conv descriptor
self.pad = {self.padT, self.padH, self.padW}
self.stride = {self.dT, self.dH, self.dW}
local mathtype=cudnn.configmap(torch.type(self.weight))
-- 3D convolutions do not work in 16 bits
if mathtype == 'CUDNN_DATA_HALF' then
mathtype = 'CUDNN_DATA_FLOAT'
end
self.convDescData = { padA = self.pad, filterStrideA = self.stride,
dataType = mathtype }
self.convDesc = cudnn.setConvolutionDescriptor(self.convDescData)
local oSize = torch.IntTensor(5)
errcheck('cudnnGetConvolutionNdForwardOutputDim',
self.convDesc[0], self.iDesc[0],
self.weightDesc[0], 5, oSize:data())
self.output:resize(oSize:long():storage())
-- create descriptor for output
self.oDesc = cudnn.toDescriptor(self.output)
self.oDescForBias = cudnn.toDescriptor(
self.output:view(self.output:size(1),
self.output:size(2),
self.output:size(3)*self.output:size(4),
self.output:size(5)))
self.input_offset = 0
self.output_offset = 0
self.weight_offset = 0
-- next two lines are so that input does not get wiped out in clearState
-- otherwise, tests do not pass
local input_slice = input:narrow(2,1,self.nInputPlane)
local output_slice = self.output:narrow(2,1,self.nOutputPlane)
find:prepare(self, input_slice, output_slice)
end
end
function VolumetricConvolution:updateOutput(input)
return Convolution.updateOutput(self, input)
end
function VolumetricConvolution:updateGradInput(input, gradOutput)
return Convolution.updateGradInput(self, input, gradOutput)
end
function VolumetricConvolution:accGradParameters(input, gradOutput, scale)
return Convolution.accGradParameters(self, input, gradOutput, scale)
end
function VolumetricConvolution:clearDesc()
return Convolution.clearDesc(self)
end
function VolumetricConvolution:write(f)
return Convolution.write(self, f)
end
function VolumetricConvolution:clearState()
return Convolution.clearState(self)
end
return VolumetricConvolution