/
init_opencl.nim
81 lines (67 loc) · 2.59 KB
/
init_opencl.nim
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
# Copyright 2017 the Arraymancer contributors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import ../private/sequninit,
./private/p_init_opencl,
./backend/opencl_backend,
./data_structure,
./init_cpu
proc opencl*[T:SomeReal](t: Tensor[T]): ClTensor[T] {.noInit.}=
## Convert a tensor on Cpu to a tensor on an OpenCL device.
result = newClTensor[T](t.shape)
let contig_t = t.asContiguous(rowMajor, force = true)
let size = csize(result.size * sizeof(T))
check enqueueWriteBuffer(
clQueue0,
result.get_data_ptr.toClpointer,
CL_true, # Blocking copy, we don't want contig_t to disappear while copy is pending
0,
size,
contig_t.get_data_ptr.toClpointer,
0, nil, nil
)
proc cpu*[T:SomeReal](t: ClTensor[T]): Tensor[T] {.noInit.}=
## Convert a tensor on an OpenCL device to a tensor on Cpu.
# We use blocking copy in this case to make sure
# all data is available for future computation
result.shape = t.shape
result.strides = t.strides
result.offset = t.offset
result.data = newSeqUninit[T](t.storage.Flen) # We copy over all the memory allocated (without prior asContiguous)
let size = t.storage.Flen * sizeof(T)
check enqueueReadBuffer(
clQueue0,
t.get_data_ptr.toClpointer,
CL_true, # Blocking copy, we don't want computation to continue while copy is still pending
0,
size,
result.get_data_ptr.toClpointer,
0, nil, nil
)
proc zeros_like*[T: SomeReal](t: ClTensor[T]): ClTensor[T] {.noInit, inline.} =
## Creates a new ClTensor filled with 0 with the same shape as the input
## Input:
## - Shape of the CudaTensor
## - Type of its elements
## Result:
## - A zero-ed ClTensor of the same shape
# TODO use clEnqueueFillBuffer (OpenCL 1.2 only)
result = zeros[T](t.shape).opencl
proc ones_like*[T: SomeReal](t: ClTensor[T]): ClTensor[T] {.noInit, inline.} =
## Creates a new ClTensor filled with 1 with the same shape as the input
## and filled with 1
## Input:
## - A CudaTensor
## Result:
## - A one-ed ClTensor of the same shape
result = ones[T](t.shape).opencl