-
Notifications
You must be signed in to change notification settings - Fork 1
/
gemmini_dev_a.cc
383 lines (332 loc) · 12.2 KB
/
gemmini_dev_a.cc
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
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
#include "gemmini_dev_a/gemmini_dev_a.hh"
namespace gem5
{
GemminiDevA::GemminiDevA(const GemminiDevAParams ¶ms) :
NDP(params)
{
}
uint64_t
GemminiDevA::readPI(uint64_t ridx)
{
switch (ridx)
{
case 7: return pi_status;
default:
panic("GemminiDevA does not have readable r[%lu] register!\n", ridx);
}
}
void
GemminiDevA::writePI(uint64_t ridx, uint64_t data)
{
DPRINTF(GemminiDevAPI, "Gemmini device PI: %lu -> r[%lu]\n", data, ridx);
if (!pi_status)
{
panic("Tried to started workload when previous one is not finished!\n");
}
switch (ridx)
{
case 0: pi_addr_m = data; break;
case 1: pi_addr_k = data; break;
case 2: pi_addr_o = data; break;
case 3: pi_size_m = data; break;
case 4: pi_size_k = data; break;
case 5: pi_opcode = data; break;
case 6:
// Print PI current status
DPRINTF(
GemminiDevAPI,
"GemminiDevA started processing...\n"
"===== pi_addr_m = %lu\n"
"===== pi_addr_k = %lu\n"
"===== pi_addr_o = %lu\n"
"===== pi_size_m = %lu\n"
"===== pi_size_k = %lu\n"
"===== pi_opcode = %lu\n"
"===== pi_status = %lu\n",
pi_addr_m,
pi_addr_k,
pi_addr_o,
pi_size_m,
pi_size_k,
pi_opcode,
pi_status
);
pi_status = 0;
// Allocate memory for operands and result
switch (pi_opcode)
{
case op_conv2d: m_size =
o_size = pi_size_m * pi_size_m;
k_size = pi_size_k * pi_size_k;
break;
case op_conv2d_gemm: m_size = pi_size_m * pi_size_m * pi_size_k * pi_size_k;
k_size = pi_size_k * pi_size_k;
o_size = pi_size_m * pi_size_m;
break;
case op_conv3d: m_size =
o_size = pi_size_m * pi_size_m * pi_size_m;
k_size = pi_size_k * pi_size_k * pi_size_k;
break;
case op_conv3d_gemm: m_size = pi_size_m * pi_size_m * pi_size_m * pi_size_k * pi_size_k * pi_size_k;
k_size = pi_size_k * pi_size_k * pi_size_k;
o_size = pi_size_m * pi_size_m * pi_size_m;
break;
case op_maxpool_gemm:
case op_maxpool: m_size = pi_size_m * pi_size_m;
k_size = 0;
o_size = pi_size_m * pi_size_m / pi_size_k / pi_size_k;
break;
case op_relu: m_size =
o_size = pi_size_m * pi_size_m;
k_size = 0;
break;
case op_mm_gemm:
case op_mm: m_size =
k_size =
o_size = pi_size_m * pi_size_m;
break;
default: panic("An invalid opcode was issued!\n");
}
m_size *= sizeof(float);
k_size *= sizeof(float);
o_size *= sizeof(float);
// Allocate memory for operands and result
m = (float *) malloc(m_size);
o = (float *) malloc(o_size);
if (k_size > 0)
k = (float *) malloc(k_size);
// Start finite state machine
process_fsm();
break;
default:
panic("GemminiDevA does not have writable r[%lu] register!\n", ridx);
}
}
void
GemminiDevA::recvData(Addr addr, uint8_t *data, size_t size)
{
DPRINTF(GemminiDevAMem, "GemminiDevA received %u bytes from %p\n", size, addr);
if (data)
{
has_operands++;
process_fsm();
}
else
{
pi_status = 1;
has_operands = 0;
free(m);
free(o);
if (k_size > 0)
free(k);
}
}
void
GemminiDevA::process_fsm()
{
// Retrieve m from memory
if (has_operands == 0)
{
DPRINTF(
GemminiDevA,
"Retrieving m from memory\n"
);
accessMemory(
Addr(pi_addr_m),
m_size,
false,
(uint8_t *) m
);
}
// Retrieve k from memory
else if(has_operands == 1 && k_size > 0)
{
DPRINTF(
GemminiDevA,
"Retrieving k from memory\n"
);
accessMemory(
Addr(pi_addr_k),
k_size,
false,
(uint8_t *) k
);
}
else
{
DPRINTF(
GemminiDevA,
"Processing operation...\n"
);
switch (pi_opcode)
{
case op_conv2d: baseConv_2D(m, k, o, pi_size_m, pi_size_k); break;
case op_conv2d_gemm: baseConvGemm(m, k, o, pi_size_m * pi_size_m, pi_size_k * pi_size_k); break;
case op_conv3d: baseConv_3D(m, k, o, pi_size_m, pi_size_k); break;
case op_conv3d_gemm: baseConvGemm(m, k, o, pi_size_m * pi_size_m * pi_size_m, pi_size_k * pi_size_k * pi_size_k); break;
case op_maxpool: baseMaxPool(m, o, pi_size_m, pi_size_k); break;
case op_maxpool_gemm: baseMaxPoolGemm(m, o, pi_size_m, pi_size_k); break;
case op_relu: baseRelu(m, o, pi_size_m); break;
case op_mm: baseMM(m, k, o, pi_size_m); break;
case op_mm_gemm: baseMMGemm(m, k, o, pi_size_m); break;
default: panic("Not implemented yet!\n");
}
accessMemory(
Addr(pi_addr_o),
o_size,
true,
(uint8_t *) o
);
}
}
/* ==================== BEGIN SPECIFIC METHODS ==================== */
void
GemminiDevA::baseConv_2D(float *m, float *k, float *o, unsigned int m_size, unsigned int k_size)
{
for (int i = 0; i < m_size; i++)
for (int j = 0; j < m_size; j++)
{
o[i * m_size + j] = 0;
for (int w = 0; w < k_size; w++)
for (int x = 0; x < k_size; x++)
{
int m_i = i + w - k_size / 2,
m_j = j + x - k_size / 2,
m_1d_idx;
if (m_i < 0 || m_i >= m_size ||
m_j < 0 || m_j >= m_size)
m_1d_idx = -1;
else
m_1d_idx = m_i * m_size + m_j;
o[i * m_size + j] += (m_1d_idx == -1) ? 0 : m[m_1d_idx] * k[w * k_size + x];
}
}
}
void
GemminiDevA::baseConv_3D(float *m, float *k, float *o, unsigned int m_size, unsigned int k_size)
{
for (int i = 0; i < m_size; i++)
for (int j = 0; j < m_size; j++)
for (int w = 0; w < m_size; w++)
{
o[i * m_size * m_size + j * m_size + w] = 0;
for (int x = 0; x < k_size; x++)
for (int y = 0; y < k_size; y++)
for (int z = 0; z < k_size; z++)
{
int m_i = i + x - k_size / 2,
m_j = j + y - k_size / 2,
m_w = w + z - k_size / 2,
m_1d_idx;
if (m_i < 0 || m_i >= m_size ||
m_j < 0 || m_j >= m_size ||
m_w < 0 || m_w >= m_size)
m_1d_idx = -1;
else
m_1d_idx = m_i * m_size * m_size + m_j * m_size + m_w;
o[i * m_size * m_size + j * m_size + w] +=
(m_1d_idx == -1) ? 0 : m[m_1d_idx] * k[x * k_size * k_size + y * k_size + z];
}
}
}
void
GemminiDevA::baseConvGemm(float *a, float *k, float *o, unsigned int o_size, unsigned int k_size)
{
for (int i = 0; i < o_size; i++)
{
o[i] = 0;
for (int j = 0; j < k_size; j++)
o[i] += a[i * k_size + j] * k[j];
}
}
void
GemminiDevA::baseMaxPool(float *m, float *o, unsigned int m_size, unsigned int k_size)
{
assert(m_size % k_size == 0);
for (int i = 0; i < m_size / k_size; i++)
for (int j = 0; j < m_size / k_size; j++)
{
o[i * m_size / k_size + j] = m[i * k_size * m_size + j * k_size];
for (int w = 0; w < k_size; w++)
for (int x = 0; x < k_size; x++)
{
unsigned int m_1d_idx = (i * k_size + w) * m_size + j * k_size + x;
if (m[m_1d_idx] > o[i * m_size / k_size + j])
o[i * m_size / k_size + j] = m[m_1d_idx];
}
}
}
void
GemminiDevA::baseMaxPoolGemm(float *a, float *o, unsigned int m_size, unsigned int k_size)
{
for (int i = 0; i < m_size / k_size; i++)
for (int j = 0; j < m_size / k_size; j++)
{
unsigned int offset = (i * m_size / k_size + j) * k_size * k_size;
float max = a[offset];
for (int w = 0; w < k_size; w++)
for (int x = 0; x < k_size; x++)
{
unsigned int idx = offset + w * k_size + x;
if (a[idx] > max)
max = a[idx];
}
o[i * m_size / 2 + j] = max;
}
}
void
GemminiDevA::baseRelu(float *m, float *o, unsigned int m_size)
{
for (int i = 0; i < m_size; i++)
for (int j = 0; j < m_size; j++)
o[i * m_size + j] = (m[i * m_size + j] < 0) ? 0 : m[i * m_size + j];
}
void
GemminiDevA::baseMM(float *a, float *b, float *c, unsigned int m_size)
{
for (int i = 0; i < m_size; i++)
for (int j = 0; j < m_size; j++)
{
c[i * m_size + j] = 0;
for (int w = 0; w < m_size; w++)
c[i * m_size + j] += a[i * m_size + w] * b[w * m_size + j];
}
}
void
GemminiDevA::baseMMGemm(float *a, float *b, float *c, unsigned int m_size)
{
for (int i = 0; i < m_size; i++)
for (int j = 0; j < m_size; j++)
{
c[i * m_size + j] = 0;
for (int w = 0; w < m_size; w++)
c[i * m_size + j] += a[i * m_size + w] * b[j * m_size + w];
}
}
void
GemminiDevA::printMatrix_2D(float *m, unsigned int m_size)
{
for (int i = 0; i < m_size; i++)
{
for (int j = 0; j < m_size; j++)
printf("%3.0f ", m[i * m_size + j]);
printf("\n");
}
}
void
GemminiDevA::printMatrix_3D(float *m, unsigned int m_size)
{
for (int i = 0; i < m_size; i++)
{
for (int w = 0; w < m_size; w++)
{
printf("[ ");
for (int j = 0; j < m_size; j++)
printf("%3.0f ", m[i * m_size * m_size + j * m_size + w]);
printf("] ");
}
printf("\n");
}
}
/* ===================== END SPECIFIC METHODS ==================== */
} // namespace gem5