1// ======================================================================== //
2// Copyright 2009-2019 Intel Corporation //
3// //
4// Licensed under the Apache License, Version 2.0 (the "License"); //
5// you may not use this file except in compliance with the License. //
6// You may obtain a copy of the License at //
7// //
8// http://www.apache.org/licenses/LICENSE-2.0 //
9// //
10// Unless required by applicable law or agreed to in writing, software //
11// distributed under the License is distributed on an "AS IS" BASIS, //
12// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. //
13// See the License for the specific language governing permissions and //
14// limitations under the License. //
15// ======================================================================== //
16
17#pragma once
18
19#include "node.h"
20
21namespace oidn {
22
23 // Reorders weights from oihw to padded oihw format
24 template<int K>
25 class WeightsReorderNode : public Node
26 {
27 private:
28 std::shared_ptr<memory> src;
29 std::shared_ptr<memory> dst;
30
31 public:
32 WeightsReorderNode(const std::shared_ptr<memory>& src,
33 const std::shared_ptr<memory>& dst)
34 : src(src),
35 dst(dst)
36 {
37 const mkldnn_memory_desc_t& srcDesc = src->get_desc().data;
38 const mkldnn_memory_desc_t& dstDesc = dst->get_desc().data;
39 MAYBE_UNUSED(srcDesc);
40 MAYBE_UNUSED(dstDesc);
41 assert(memory_desc_matches_tag(srcDesc, mkldnn_format_tag_t(memory::format_tag::oihw)));
42 assert(memory_desc_matches_tag(dstDesc, mkldnn_format_tag_t(memory::format_tag::oihw)));
43 assert(srcDesc.ndims == 4);
44 assert(dstDesc.ndims == 4);
45 assert(srcDesc.data_type == memory::data_type::f32);
46 assert(dstDesc.data_type == memory::data_type::f32);
47 assert(getPadded<K>(srcDesc.dims[0]) == dstDesc.dims[0]); // OC
48 assert(getPadded<K>(srcDesc.dims[1]) == dstDesc.dims[1]); // IC
49 assert(srcDesc.dims[2] == dstDesc.dims[2]);
50 assert(srcDesc.dims[3] == dstDesc.dims[3]);
51 }
52
53 void execute(stream& sm) override
54 {
55 const mkldnn_memory_desc_t& srcDesc = src->get_desc().data;
56 const mkldnn_memory_desc_t& dstDesc = dst->get_desc().data;
57
58 const float* srcPtr = (float*)src->get_data_handle();
59 float* dstPtr = (float*)dst->get_data_handle();
60
61 const int OC1 = srcDesc.dims[0];
62 const int OC2 = dstDesc.dims[0];
63 const int IC1 = srcDesc.dims[1];
64 const int IC2 = dstDesc.dims[1];
65 const int H = dstDesc.dims[2];
66 const int W = dstDesc.dims[3];
67
68 for (int oc = 0; oc < OC2; ++oc)
69 {
70 for (int ic = 0; ic < IC2; ++ic)
71 {
72 for (int h = 0; h < H; ++h)
73 {
74 for (int w = 0; w < W; ++w)
75 {
76 // Output is in oihw format
77 float* dstPtr_c = dstPtr + oc*IC2*H*W + ic*H*W + h*W + w;
78
79 if (oc < OC1 && ic < IC1)
80 {
81 // Input is in oihw format
82 const float* srcPtr_c = srcPtr + oc*IC1*H*W + ic*H*W + h*W + w;
83 *dstPtr_c = *srcPtr_c;
84 }
85 else
86 {
87 // padding
88 *dstPtr_c = 0;
89 }
90 }
91 }
92 }
93 }
94 }
95
96 std::shared_ptr<memory> getDst() const override { return dst; }
97 };
98
99} // namespace oidn
100