Coverage Report

Created: 2019-07-03 22:50

/home/liu/buildslave/linux-x64-runtests/build/lib/nnc/cmd/relu/ccv_nnc_relu_cpu_ref.c
Line
Count
Source
1
#include <ccv.h>
2
#include <ccv_internal.h>
3
#include <nnc/ccv_nnc.h>
4
#include <nnc/ccv_nnc_easy.h>
5
#include <nnc/ccv_nnc_internal.h>
6
#ifdef USE_OPENMP
7
#include <omp.h>
8
#endif
9
#ifdef USE_DISPATCH
10
#include <dispatch/dispatch.h>
11
#endif
12
13
static int _ccv_nnc_relu_forw(const ccv_nnc_cmd_t cmd, const ccv_nnc_hint_t hint, const int flags, ccv_nnc_tensor_t* const* const inputs, const int input_size, ccv_nnc_tensor_t* const* const outputs, const int output_size, ccv_nnc_stream_context_t* const stream_context)
14
1.26k
{
15
1.26k
  assert(input_size == 1);
16
1.26k
  const ccv_nnc_tensor_t* a = inputs[0];
17
1.26k
  assert(!CCV_IS_TENSOR_VIEW(a));
18
1.26k
  assert(output_size == 1);
19
1.26k
  ccv_nnc_tensor_t* b = outputs[0];
20
1.26k
  assert(!CCV_IS_TENSOR_VIEW(b));
21
1.26k
  int i, count = ccv_nnc_tensor_count(a->info);
22
5.65k
  for (i = 0; i < CCV_NNC_MAX_DIM_ALLOC && a->info.dim[i] > 0; 
i++4.39k
)
23
4.39k
  {
24
4.39k
    assert(a->info.dim[i] == b->info.dim[i]);
25
4.39k
  }
26
1.26k
  float* ap = a->data.f32;
27
1.26k
  float* bp = b->data.f32;
28
39.1M
  for (i = 0; i < count; 
i++39.1M
)
29
39.1M
    bp[i] = ccv_max(ap[i], 0);
30
1.26k
  return CCV_NNC_EXEC_SUCCESS;
31
1.26k
}
32
33
static int _ccv_nnc_relu_back(const ccv_nnc_cmd_t cmd, const ccv_nnc_hint_t hint, const int flags, ccv_nnc_tensor_t* const* const inputs, const int input_size, ccv_nnc_tensor_t* const* const outputs, const int output_size, ccv_nnc_stream_context_t* const stream_context)
34
1.47k
{
35
1.47k
  assert(input_size == 3);
36
1.47k
  const ccv_nnc_tensor_t* g = inputs[0]; // gradient
37
1.47k
  assert(!CCV_IS_TENSOR_VIEW(g));
38
1.47k
  const ccv_nnc_tensor_t* b = inputs[2];
39
1.47k
  assert(!CCV_IS_TENSOR_VIEW(b));
40
1.47k
  assert(output_size == 1);
41
1.47k
  ccv_nnc_tensor_t* h = outputs[0];
42
1.47k
  assert(!CCV_IS_TENSOR_VIEW(h));
43
1.47k
  int i, count = ccv_nnc_tensor_count(g->info);
44
6.61k
  for (i = 0; i < CCV_NNC_MAX_DIM_ALLOC && g->info.dim[i] > 0; 
i++5.14k
)
45
5.14k
  {
46
5.14k
    assert(b->info.dim[i] == g->info.dim[i]);
47
5.14k
    assert(g->info.dim[i] == h->info.dim[i]);
48
5.14k
  }
49
1.47k
  float* bp = b->data.f32;
50
1.47k
  float* gp = g->data.f32;
51
1.47k
  float* hp = h->data.f32;
52
15.1M
  for (i = 0; i < count; 
i++15.1M
)
53
15.1M
    hp[i] = (bp[i] > 0) ? 
gp[i]7.30M
:
07.87M
;
54
1.47k
  return CCV_NNC_EXEC_SUCCESS;
55
1.47k
}
56
57
REGISTER_COMMAND_BACKEND(CCV_NNC_RELU_FORWARD, CCV_NNC_BACKEND_CPU_REF)(ccv_nnc_cmd_backend_registry_t* const registry)
58
1
{
59
1
  registry->tensor_formats = CCV_TENSOR_FORMAT_NHWC | CCV_TENSOR_FORMAT_NCHW | CCV_TENSOR_FORMAT_CHWN;
60
1
  registry->tensor_datatypes = CCV_32F;
61
1
  registry->tensor_memory = CCV_TENSOR_CPU_MEMORY;
62
1
  registry->algorithms = 1;
63
1
  registry->exec = _ccv_nnc_relu_forw;
64
1
}
65
66
REGISTER_COMMAND_BACKEND(CCV_NNC_RELU_BACKWARD, CCV_NNC_BACKEND_CPU_REF)(ccv_nnc_cmd_backend_registry_t* const registry)
67
1
{
68
1
  registry->tensor_formats = CCV_TENSOR_FORMAT_NHWC | CCV_TENSOR_FORMAT_NCHW | CCV_TENSOR_FORMAT_CHWN;
69
1
  registry->tensor_datatypes = CCV_32F;
70
1
  registry->tensor_memory = CCV_TENSOR_CPU_MEMORY;
71
1
  registry->algorithms = 1;
72
1
  registry->exec = _ccv_nnc_relu_back;
73
1
}