/home/liu/actions-runner/_work/ccv/ccv/lib/nnc/cmd/nms/ccv_nnc_nms.c
Line | Count | Source (jump to first uncovered line) |
1 | | #include "ccv.h" |
2 | | #include "nnc/ccv_nnc.h" |
3 | | #include "nnc/ccv_nnc_easy.h" |
4 | | #include "nnc/ccv_nnc_internal.h" |
5 | | |
6 | | static int _ccv_nnc_nms_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) |
7 | 0 | { |
8 | 0 | if ((input_bitmasks[0] & 1u) == 1u && output_bitmasks[0] == 3u) |
9 | 0 | return 1; |
10 | 0 | return 0; |
11 | 0 | } |
12 | | |
13 | | static int _ccv_nnc_nms_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) |
14 | 0 | { |
15 | | // gradient of sorted, gradient of sorting index, input, output of sorted, output of sorting index. |
16 | 0 | if ((input_bitmasks[0] & 17u) == ((1u << 0) | (0u << 1) | (0u << 2) | (0u << 3) | (1u << 4)) && output_bitmasks[0] == 1u) |
17 | 0 | return 1; |
18 | 0 | return 0; |
19 | 0 | } |
20 | | |
21 | | static void _ccv_nnc_nms_tensor_auto_forw(const ccv_nnc_cmd_param_t cmd, const ccv_nnc_tensor_param_t* inputs, const int input_size, const ccv_nnc_hint_t hint, ccv_nnc_tensor_param_t* outputs, const int output_size) |
22 | 0 | { |
23 | 0 | assert(output_size == 2); |
24 | 0 | outputs[0] = inputs[0]; |
25 | 0 | const int nd = ccv_nnc_tensor_nd(inputs[0].dim); |
26 | 0 | assert(nd >= 1); |
27 | 0 | outputs[1] = inputs[0]; |
28 | 0 | outputs[1].datatype = CCV_32S; |
29 | 0 | memset(outputs[1].dim, 0, sizeof(outputs[1].dim)); |
30 | 0 | outputs[1].dim[0] = inputs[0].dim[0]; // How many to rank (or batch size). |
31 | 0 | outputs[1].dim[1] = (nd <= 2) ? 0 : inputs[0].dim[1]; // How many to rank. |
32 | 0 | } |
33 | | |
34 | | static void _ccv_nnc_nms_tensor_auto_back(const ccv_nnc_cmd_param_t cmd, const ccv_nnc_tensor_param_t* inputs, const int input_size, const ccv_nnc_hint_t hint, ccv_nnc_tensor_param_t* outputs, const int output_size) |
35 | 0 | { |
36 | 0 | assert(output_size == 1); |
37 | 0 | outputs[0] = inputs[2]; |
38 | 0 | } |
39 | | |
40 | | REGISTER_COMMAND(CCV_NNC_NMS_FORWARD)(ccv_nnc_cmd_registry_t* const registry) |
41 | | FIND_BACKEND(ccv_nnc_nms_cpu_ref.c, gpu/ccv_nnc_nms_gpu_ref.cu) |
42 | 1 | { |
43 | 1 | registry->bitmask = _ccv_nnc_nms_forw_bitmask; |
44 | 1 | registry->tensor_auto = _ccv_nnc_nms_tensor_auto_forw; |
45 | 1 | } |
46 | | |
47 | | REGISTER_COMMAND(CCV_NNC_NMS_BACKWARD)(ccv_nnc_cmd_registry_t* const registry) |
48 | | FIND_BACKEND(ccv_nnc_nms_cpu_ref.c, gpu/ccv_nnc_nms_gpu_ref.cu) |
49 | 1 | { |
50 | 1 | registry->bitmask = _ccv_nnc_nms_back_bitmask; |
51 | 1 | registry->tensor_auto = _ccv_nnc_nms_tensor_auto_back; |
52 | 1 | } |
53 | | |
54 | | //@REGISTER_EASY_COMMAND_MACRO(CCV_NNC_NMS_FORWARD) |
55 | | #define CMD_NMS_FORWARD(_iou_threshold) ccv_nnc_cmd(CCV_NNC_NMS_FORWARD, 0, ((ccv_nnc_cmd_param_t){.nms={.iou_threshold=_iou_threshold}}), 0) |
56 | | //@REGISTER_EASY_COMMAND_MACRO(CCV_NNC_NMS_BACKWARD) |
57 | | #define CMD_NMS_BACKWARD(_iou_threshold) ccv_nnc_cmd(CCV_NNC_NMS_BACKWARD, 0, ((ccv_nnc_cmd_param_t){.nms={.iou_threshold=_iou_threshold}}), 0) |