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
|
/*
* Vector addition with multiple tasklets
*
*/
#include <stdint.h>
#include <stdio.h>
#include <defs.h>
#include <mram.h>
#include <alloc.h>
#include <perfcounter.h>
#include <barrier.h>
#include "../support/common.h"
__host dpu_arguments_t DPU_INPUT_ARGUMENTS;
// vector_addition: Computes the vector addition of a cached block
static void vector_addition(T *bufferB, T *bufferA, unsigned int l_size)
{
for (unsigned int i = 0; i < l_size; i++) {
bufferB[i] += bufferA[i];
}
}
// Barrier
BARRIER_INIT(my_barrier, NR_TASKLETS);
extern int main_kernel1(void);
int (*kernels[nr_kernels])(void) = { main_kernel1 };
int main(void)
{
// Kernel
return kernels[DPU_INPUT_ARGUMENTS.kernel] ();
}
// main_kernel1
int main_kernel1()
{
unsigned int tasklet_id = me();
#if PRINT
printf("tasklet_id = %u\n", tasklet_id);
#endif
if (tasklet_id == 0) { // Initialize once the cycle counter
mem_reset(); // Reset the heap
}
// Barrier
barrier_wait(&my_barrier);
uint32_t input_size_dpu_bytes = DPU_INPUT_ARGUMENTS.size; // Input size per DPU in bytes
uint32_t input_size_dpu_bytes_transfer = DPU_INPUT_ARGUMENTS.transfer_size; // Transfer input size per DPU in bytes
// Address of the current processing block in MRAM
uint32_t base_tasklet = tasklet_id << BLOCK_SIZE_LOG2;
uint32_t mram_base_addr_A = (uint32_t) DPU_MRAM_HEAP_POINTER;
uint32_t mram_base_addr_B =
(uint32_t) (DPU_MRAM_HEAP_POINTER + input_size_dpu_bytes_transfer);
// Initialize a local cache to store the MRAM block
T *cache_A = (T *) mem_alloc(BLOCK_SIZE);
T *cache_B = (T *) mem_alloc(BLOCK_SIZE);
for (unsigned int byte_index = base_tasklet;
byte_index < input_size_dpu_bytes;
byte_index += BLOCK_SIZE * NR_TASKLETS) {
// Bound checking
uint32_t l_size_bytes =
(byte_index + BLOCK_SIZE >=
input_size_dpu_bytes) ? (input_size_dpu_bytes -
byte_index) : BLOCK_SIZE;
// Load cache with current MRAM block
mram_read((__mram_ptr void const *)(mram_base_addr_A +
byte_index), cache_A,
l_size_bytes);
mram_read((__mram_ptr void const *)(mram_base_addr_B +
byte_index), cache_B,
l_size_bytes);
// Computer vector addition
vector_addition(cache_B, cache_A, l_size_bytes >> DIV);
// Write cache to current MRAM block
mram_write(cache_B,
(__mram_ptr void *)(mram_base_addr_B + byte_index),
l_size_bytes);
}
return 0;
}
|
