-
Notifications
You must be signed in to change notification settings - Fork 67
/
dma.h
238 lines (220 loc) · 9.24 KB
/
dma.h
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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
/******************************************************************************
* Copyright (c) 2019, Xilinx, Inc.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
* THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
* OR BUSINESS INTERRUPTION). HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
* ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
******************************************************************************/
/******************************************************************************
*
* Authors: Giulio Gambardella <[email protected]>
* Thomas B. Preusser <[email protected]>
* Marie-Curie Fellow, Xilinx Ireland, Grant Agreement No. 751339
* Christoph Doehring <[email protected]>
*
* \file dma.h
*
* Library of templated HLS functions for BNN deployment.
* This file lists a set of functions to access memory mapped values into
* streams.
*
*****************************************************************************/
#ifndef DMA_HPP
#define DMA_HPP
#include <ap_int.h>
#include <hls_stream.h>
/*!
* \brief DMA block accessing AXI4 memory and output HLS streams
*
*
* \tparam DataWidth Width, in number of bits, of the AXI4 memory pointer and the output HLS stream
* \tparam numBytes Number of bytes to be read from the memory
*
* \param in Input memory pointer
* \param out Output HLS stream
*/
template<unsigned int DataWidth, unsigned int numBytes>
void Mem2Stream(ap_uint<DataWidth> * in, hls::stream<ap_uint<DataWidth> > & out);
/*!
* \brief DMA block accessing AXI4 memory and output HLS streams multiple times
*
* It basically calls Mem2Stream function multiple times, possibly with bigger sizes so to increase
* the burst size
*
* \tparam DataWidth Width, in number of bits, of the AXI4 memory pointer and the output HLS stream
* \tparam numBytes Number of bytes to be read from the memory
*
* \param in Input memory pointer
* \param out Output HLS stream
* \param numReps Number of times the Stream2Mem function has to be called
*/
template<unsigned int DataWidth, unsigned int numBytes>
void Mem2Stream_Batch(ap_uint<DataWidth> * in, hls::stream<ap_uint<DataWidth> > & out, const unsigned int numReps);
/*!
* \brief DMA block writing HLS streams content in AXI4 pointed memory
*
*
* \tparam DataWidth Width, in number of bits, of the AXI4 memory pointer and the output HLS stream
* \tparam numBytes Number of bytes to be read from the memory
*
* \param in Input HLS stream
* \param out Output memory pointer
*/
template<unsigned int DataWidth, unsigned int numBytes>
void Stream2Mem(hls::stream<ap_uint<DataWidth> > & in, ap_uint<DataWidth> * out);
/*!
* \brief DMA block that accesses the external memory and outputs HLS streams multiple times
*
* It basically calls Mem2Stream function multiple times, possibly with bigger sizes so to increase
* the burst size
*
* \tparam DataWidth Width, in number of bits, of the AXI4 memory pointer and the output HLS stream
* \tparam numBytes Number of bytes to be read from the memory
*
* \param in Input pointer to external memory
* \param out Output the generated HLS sream
* \param numReps Number of times the Mem2Stream function has to be called
*/
template<unsigned int DataWidth, unsigned int numBytes>
void Mem2Stream_Batch_external_wmem(ap_uint<DataWidth> * in,
hls::stream<ap_uint<DataWidth> > & out, const unsigned int numReps) {
unsigned int rep = 0;
while (rep != numReps) {
Mem2Stream<DataWidth, numBytes>(&in[0], out);
rep += 1;
}
}
/*!
* \brief DMA block writing HLS streams content in AXI4 pointed memory multiple times
*
* It basically calls Stream2Mem function multiple times, possibly with bigger sizes so to increase
* the burst size
*
* \tparam DataWidth Width, in number of bits, of the AXI4 memory pointer and the output HLS stream
* \tparam numBytes Number of bytes to be read from the memory
*
* \param in Input HLS stream
* \param out Output memory pointer
* \param numReps Number of times the Stream2Mem function has to be called
*/
template<unsigned int DataWidth, unsigned int numBytes>
void Stream2Mem_Batch(hls::stream<ap_uint<DataWidth> > & in, ap_uint<DataWidth> * out, const unsigned int numReps);
template<unsigned int DataWidth, unsigned int numBytes>
void Mem2Stream(ap_uint<DataWidth> * in, hls::stream<ap_uint<DataWidth> > & out) {
static_assert(DataWidth % 8 == 0, "");
const unsigned int numWords = numBytes / (DataWidth / 8);
static_assert(numWords != 0, "");
for (unsigned int i = 0; i < numWords; i++) {
#pragma HLS pipeline style=flp II=1
ap_uint<DataWidth> e = in[i];
out.write(e);
}
}
template<unsigned int DataWidth, unsigned int numBytes>
void Stream2Mem(hls::stream<ap_uint<DataWidth> > & in, ap_uint<DataWidth> * out) {
static_assert(DataWidth % 8 == 0, "");
const unsigned int numWords = numBytes / (DataWidth / 8);
static_assert(numWords != 0, "");
for (unsigned int i = 0; i < numWords; i++) {
#pragma HLS pipeline style=flp II=1
ap_uint<DataWidth> e = in.read();
out[i] = e;
}
}
template<unsigned int DataWidth, unsigned int numBytes>
void Mem2Stream_Batch(ap_uint<DataWidth> * in, hls::stream<ap_uint<DataWidth> > & out, const unsigned int numReps) {
const unsigned int indsPerRep = numBytes / (DataWidth / 8);
unsigned int rep = 0;
// make sure Mem2Stream does not get inlined here
// we lose burst inference otherwise
while (rep != numReps) {
unsigned int repsLeft = numReps - rep;
if ((repsLeft & 0xF) == 0) {
// repsLeft divisable by 16, read 16 images
Mem2Stream<DataWidth, numBytes * 16>(&in[rep * indsPerRep], out);
rep += 16;
} else {
// fallback, read single image
Mem2Stream<DataWidth, numBytes>(&in[rep * indsPerRep], out);
rep += 1;
}
}
}
template<unsigned int DataWidth, unsigned int numBytes>
void Stream2Mem_Batch(hls::stream<ap_uint<DataWidth> > & in, ap_uint<DataWidth> * out, const unsigned int numReps) {
const unsigned int indsPerRep = numBytes / (DataWidth / 8);
unsigned int rep = 0;
// make sure Stream2Mem does not get inlined here
// we lose burst inference otherwise
while (rep != numReps) {
unsigned int repsLeft = numReps - rep;
if ((repsLeft & 0xF) == 0) {
// repsLeft divisable by 16, write 16 images
Stream2Mem<DataWidth, numBytes * 16>(in, &out[rep * indsPerRep]);
rep += 16;
} else {
// fallback, write single image
Stream2Mem<DataWidth, numBytes>(in, &out[rep * indsPerRep]);
rep += 1;
}
}
}
/*!
* \brief Streaming block that fetches parameters from internal memory and presents them to the MVAU
*
* \tparam TILES Total folding factor of the neural network mapped on the fabric (Neuron Fold * Synapse Fold)
* \tparam SIMD Number of input columns computed in parallel
* \tparam PE Number of output rows computed in parallel
* \tparam WP Precision of the weights in the network
* \tparam TW DataType of the weights matrix - safely deducible from the paramaters
*
* \param W_in Pointer to the weight memory
* \param paramStreamOut Parameter stream that contains SIMD * PE * WP long words to digest by the MVAU
* \param numReps Number of time the function has to be repeatedly executed (e.g. number of images)
*/
template<
unsigned int TILES,
unsigned int SIMD,
unsigned int PE,
unsigned int WP,
typename TW
>
void GenParamStream(TW const &W_in, hls::stream<ap_uint<SIMD * PE * WP>> ¶mStreamOut, int const numReps) {
for (unsigned rep = 0; rep < numReps; rep++) {
for (unsigned tile = 0; tile < TILES; tile++) {
#pragma HLS pipeline style=flp II=1
ap_uint<SIMD * PE * WP> strMem;
for (unsigned pe = 0; pe < PE; pe++) {
// Concatenate the weights within the tile into a large SIMD * PE * WP wide word
// Using Little Endian PE order
strMem((SIMD * WP)*(pe+1)-1,(SIMD * WP)*pe) = W_in.m_weights[pe][tile];
}
paramStreamOut.write(strMem);
}
}
}
#endif