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
|
#include <sycl/ext/intel/fpga_extensions.hpp>
#include <chrono>
#include "kernel.hpp"
template<typename T>
SYCL_EXTERNAL bool checkOptimality(device_ptr<T> C, int size) {
bool isOptimal = false;
int positveValueCount = 0;
//check if the coefficients of the objective function are negative
for(int i=0; i<size;i++){
float value = C[i];
if(value >= 0){
positveValueCount++;
}
}
//if all the constraints are positive now,the table is optimal
if(positveValueCount == size){
isOptimal = true;
}
return isOptimal;
}
template<typename T>
SYCL_EXTERNAL int findPivotColumn(device_ptr<T> C, int size) {
int location = 0;
float minimum = C[0];
for(int i=1; i<size; ++i) {
if(C[i]<minimum) {
minimum = C[i];
location = i;
}
}
return location;
}
//find the row with the pivot value.The least value item's row in the B array
template<typename T>
SYCL_EXTERNAL int findPivotRow(device_ptr<T> A, device_ptr<T> B, device_ptr<T> C, int pivotColumn, int rows, int cols, bool *isUnbounded) {
int negativeValueCount = 0;
for (int i = 0; i < rows; i++) {
// 2d to 1d array index mapping
int pivotColumnIndex = (i*cols)+pivotColumn;
if (A[pivotColumnIndex] <= 0) {
negativeValueCount += 1;
}
}
int location = 0;
//checking the unbound condition if all the values are negative ones
if (negativeValueCount == rows) {
*isUnbounded = true;
} else {
float minimum = 99999999.0;
for (int i = 0; i < rows; ++i) {
// 2d to 1d array index mapping
int pivotColumnIndex = (i*cols)+pivotColumn;
float tmpACols = A[pivotColumnIndex];
if (tmpACols > 0) {
float result = B[i] / tmpACols;
if (result > 0 && result < minimum) {
minimum = result;
location = i;
}
}
}
}
return location;
}
template<typename T>
SYCL_EXTERNAL void doPivotting(device_ptr<T> A, device_ptr<T> B, device_ptr<T> C, int pivotRow, int pivotColumn, int rows, int cols) {
int columnIndex = (pivotRow*cols)+pivotColumn;
float pivetValue = A[columnIndex];
float pivotRowVals[6]; //the column with the pivot
float pivotColVals[3]; //the row with the pivot
float rowNew[6]; //the row after processing the pivot value
float maximum = 0;
maximum = maximum-(C[pivotColumn]*(B[pivotRow]/pivetValue)); //set the maximum step by step
//get the row that has the pivot value
for (int i = 0; i < cols; ++i) {
int pivotRowIndex = (pivotRow*cols)+i;
pivotRowVals[i] = A[pivotRowIndex];
}
//get the column that has the pivot value
for (int j = 0; j < rows; ++j) {
int pivotColIndex = (j*cols)+pivotColumn;
pivotColVals[j] = A[pivotColIndex];
}
//set the row values that has the pivot value divided by the pivot value and put into new row
for (int k = 0; k < cols; ++k) {
rowNew[k] = pivotRowVals[k]/pivetValue;
}
B[pivotRow] = B[pivotRow]/pivetValue;
//process the other coefficients in the A array by subtracting
for (int m=0; m < rows; ++m) {
//ignore the pivot row as we already calculated that
if (m != pivotRow) {
for (int p = 0; p<cols; ++p) {
float multiplyValue = pivotColVals[m];
int indexA_M_P = (m*cols)+p;
A[indexA_M_P] = A[indexA_M_P] - (multiplyValue * rowNew[p]);
//C[p] = C[p] - (multiplyValue*C[pivotRow]);
//B[i] = B[i] - (multiplyValue*B[pivotRow]);
}
}
}
//process the values of the B array
for (int i = 0; i<rows; ++i) { // rows = B.size()
if (i != pivotRow) {
float multiplyValue = pivotColVals[i];
B[i] = B[i]-(multiplyValue*B[pivotRow]);
}
}
//the least coefficient of the constraints of the objective function
float multiplyValue = C[pivotColumn];
//process the C array
for (int i = 0; i < C.size(); i++) {
C[i] = C[i]-(multiplyValue * rowNew[i]);
}
//replacing the pivot row in the new calculated A array
for (int i = 0; i<cols; ++i) {
int indexA_pivotRow_i = (pivotRow*cols)+i;
A[indexA_pivotRow_i] = rowNew[i];
}
}
// Forward declare the kernel names in the global scope. This FPGA best practice
// reduces compiler name mangling in the optimization reports.
class SimplexCalc;
double RunKernel(queue &q, std::vector<T> &inAHost, std::vector<T> &inBHost,
std::vector<T> &inCHost, std::vector<int>& resultFlags) {
int rowSizeA = inBHost.size();
int colSizeA = inCHost.size();
T *inADevice = malloc_device<T> (inAHost.size(), q);
T *inBDevice = malloc_device<T> (inBHost.size(), q);
T *inCDevice = malloc_device<T> (inCHost.size(), q);
int *inResultFlagsDevice = malloc_device<int> (resultFlags.size(), q);
if (inADevice == nullptr) {
std::cerr << "ERROR: failed to allocate space for 'inADevice'\n";
std::terminate();
}
if (inBDevice == nullptr) {
std::cerr << "ERROR: failed to allocate space for 'inBDevice'\n";
std::terminate();
}
if (inCDevice == nullptr) {
std::cerr << "ERROR: failed to allocate space for 'inCDevice'\n";
std::terminate();
}
auto start = std::chrono::high_resolution_clock::now();
q.memcpy(inADevice, inAHost.data(), inAHost.size()*sizeof(T)).wait();
q.memcpy(inBDevice, inBHost.data(), inBHost.size() * sizeof(T)).wait();
q.memcpy(inCDevice, inCHost.data(), inCHost.size() * sizeof(T)).wait();
q.memcpy(inResultFlagsDevice, resultFlags.data(), resultFlags.size()*sizeof(int)).wait();
q.submit([&](handler &h) {
h.single_task < SimplexCalc > ([=]()[[intel::kernel_args_restrict]] {
device_ptr<T> inA(inADevice);
device_ptr<T> inB(inBDevice);
device_ptr<T> inC(inCDevice);
device_ptr<int> inResultFlags(inResultFlagsDevice);
bool tempIsOptimizal = checkOptimality(inC, colSizeA);
if (tempIsOptimizal) {
inResultFlags[0] = 1;
return;
} else {
inResultFlags[0] = 0;
}
int pivotColumn = findPivotColumn(inC, colSizeA);
inResultFlags[1] = pivotColumn;
inA[0] = 43.0; //test only
bool isUnbounded = true;
int pivotRow = findPivotRow(inA, inB, inC, pivotColumn, rowSizeA, colSizeA, &isUnbounded);
inResultFlags[2] = pivotRow;
});
}).wait();
q.memcpy(inAHost.data(), inADevice, inAHost.size()*sizeof(T)).wait();
q.memcpy(inBHost.data(), inBDevice, inBHost.size()*sizeof(T)).wait();
q.memcpy(inCHost.data(), inCDevice, inCHost.size()*sizeof(T)).wait();
q.memcpy(resultFlags.data(), inResultFlagsDevice, resultFlags.size()*sizeof(int)).wait();
auto end = std::chrono::high_resolution_clock::now();
std::chrono::duration<double, std::milli> diff = end - start;
sycl::free(inADevice, q);
sycl::free(inBDevice, q);
sycl::free(inCDevice, q);
sycl::free(inResultFlagsDevice, q);
return diff.count();
}
|
