Coverage Report

Created: 2024-08-19 11:27

/home/liu/actions-runner/_work/ccv/ccv/lib/nnc/cmd/lamb/ccv_nnc_lamb.c
Line
Count
Source (jump to first uncovered line)
1
#include "ccv.h"
2
#include "nnc/ccv_nnc.h"
3
#include "nnc/ccv_nnc_internal.h"
4
5
static int _ccv_nnc_lamb_forw_bitmask(const ccv_nnc_cmd_param_t cmd, const int input_size, const int output_size, const uint64_t* const input_bitmasks, const int input_bitmask_size, const uint64_t* const output_bitmasks, const int output_bitmask_size)
6
15
{
7
  // 3 inputs (gradient, x, momentum, velocity)
8
  // 2 outputs (y, new momentum, new velocity)
9
15
  if (input_bitmasks[0] == 15u && 
output_bitmasks[0] == 7u5
)
10
5
    return 1;
11
10
  return 0;
12
15
}
13
14
static int _ccv_nnc_lamb_allow_inplace(const ccv_nnc_cmd_param_t cmd, const int input_idx, const int input_size, const int output_idx, const int output_size)
15
18
{
16
18
  if (input_idx == output_idx + 1)
17
6
    return 1;
18
12
  return 0;
19
18
}
20
21
static int _ccv_nnc_lamb_back_bitmask(const ccv_nnc_cmd_param_t cmd, const int input_size, const int output_size, const uint64_t* const input_bitmasks, const int input_bitmask_size, const uint64_t* const output_bitmasks, const int output_bitmask_size)
22
0
{
23
  // Doesn't support.
24
0
  return 0;
25
0
}
26
27
static void _ccv_nnc_lamb_tensor_auto_forw(const ccv_nnc_cmd_param_t cmd, const ccv_nnc_tensor_param_t* const inputs, const int input_size, const ccv_nnc_hint_t hint, ccv_nnc_tensor_param_t* const outputs, const int output_size)
28
17
{
29
17
  int i;
30
68
  for (i = 0; i < output_size; 
i++51
)
31
51
    outputs[i] = inputs[0];
32
17
}
33
34
static void _ccv_nnc_lamb_tensor_auto_back(const ccv_nnc_cmd_param_t cmd, const ccv_nnc_tensor_param_t* const inputs, const int input_size, const ccv_nnc_hint_t hint, ccv_nnc_tensor_param_t* const outputs, const int output_size)
35
0
{
36
  // Doesn't support.
37
0
}
38
39
REGISTER_COMMAND(CCV_NNC_LAMB_FORWARD)(ccv_nnc_cmd_registry_t* const registry)
40
  FIND_BACKEND(ccv_nnc_lamb_cpu_ref.c, gpu/ccv_nnc_lamb_gpu_ref.cu)
41
1
{
42
1
  registry->bitmask = _ccv_nnc_lamb_forw_bitmask;
43
1
  registry->tensor_auto = _ccv_nnc_lamb_tensor_auto_forw;
44
1
  registry->allow_inplace = _ccv_nnc_lamb_allow_inplace;
45
1
}
46
47
REGISTER_COMMAND(CCV_NNC_LAMB_BACKWARD)(ccv_nnc_cmd_registry_t* const registry)
48
  FIND_BACKEND(ccv_nnc_lamb_cpu_ref.c, gpu/ccv_nnc_lamb_gpu_ref.cu)
49
1
{
50
1
  registry->bitmask = _ccv_nnc_lamb_back_bitmask;
51
1
  registry->tensor_auto = _ccv_nnc_lamb_tensor_auto_back;
52
1
}
53
54
//@REGISTER_EASY_COMMAND_MACRO(CCV_NNC_LAMB_FORWARD)
55
#define CMD_LAMB_FORWARD(_step, _rate, _beta1, _beta2, _decay, _epsilon) ccv_nnc_cmd(CCV_NNC_LAMB_FORWARD, 0, ((ccv_nnc_cmd_param_t){.size={.dim={1,1,1}},.lamb={.step=_step,.rate=_rate,.scale=1,.beta1=_beta1,.beta2=_beta2,.decay=_decay,.epsilon=_epsilon}}), 0)