| File: | nnc/ccv_cnnp_model_addons.c |
| Warning: | line 400, column 3 Declared variable-length array (VLA) has negative size |
Press '?' to see keyboard shortcuts
Keyboard shortcuts:
| 1 | #include "ccv_nnc.h" | |||
| 2 | #include "ccv_nnc_easy.h" | |||
| 3 | #include "ccv_nnc_internal.h" | |||
| 4 | #include "ccv_internal.h" | |||
| 5 | #include "_ccv_cnnp_model.h" | |||
| 6 | ||||
| 7 | // MARK - Add-on Functions | |||
| 8 | ||||
| 9 | static int _ccv_cnnp_model_clip_grad_norm_reduce_norm2(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) | |||
| 10 | { | |||
| 11 | const int device_id = CCV_TENSOR_GET_DEVICE_ID(inputs[0]->info.type)(((inputs[0]->info.type) & 0xfff00) >> 8); | |||
| 12 | ccv_nnc_tensor_t* const old_norm2 = outputs[1 + device_id * 2]; | |||
| 13 | ccv_nnc_tensor_t* const norm2 = outputs[1 + device_id * 2 + 1]; | |||
| 14 | const int tensor_count = ccv_nnc_tensor_count(inputs[0]->info); | |||
| 15 | if (tensor_count == 1) | |||
| 16 | ccv_nnc_cmd_exec(CMD_MUL_FORWARD(1)ccv_nnc_cmd(CCV_NNC_MUL_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={1,}}}, 0), hint, flags, TENSOR_LIST(inputs[0], inputs[0])(ccv_nnc_tensor_t* []){inputs[0], inputs[0]}, (1 +1 +1 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(norm2)(ccv_nnc_tensor_t* []){norm2}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 17 | else { | |||
| 18 | ccv_nnc_cmd_exec(CMD_REDUCE_NORM2_FORWARD()ccv_nnc_cmd(CCV_NNC_REDUCE_NORM2_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}} ), 0), hint, flags, TENSOR_LIST(inputs[0])(ccv_nnc_tensor_t* []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(norm2)(ccv_nnc_tensor_t* []){norm2}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 19 | ccv_nnc_cmd_exec(CMD_MUL_FORWARD(1)ccv_nnc_cmd(CCV_NNC_MUL_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={1,}}}, 0), hint, flags, TENSOR_LIST(norm2, norm2)(ccv_nnc_tensor_t* []){norm2, norm2}, (1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(norm2)(ccv_nnc_tensor_t* []){norm2}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 20 | } | |||
| 21 | ccv_nnc_cmd_exec(CMD_ADD_FORWARD(1, 1)ccv_nnc_cmd(CCV_NNC_ADD_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={1, 1}}}, 0), hint, flags, TENSOR_LIST(old_norm2, norm2)(ccv_nnc_tensor_t* []){old_norm2, norm2}, (1 +1 +1 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(old_norm2)(ccv_nnc_tensor_t* []){old_norm2}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 22 | return CCV_NNC_EXEC_SUCCESS; | |||
| 23 | } | |||
| 24 | ||||
| 25 | static ccv_nnc_cmd_vtab_t clip_grad_norm_reduce_norm2_vtab = { | |||
| 26 | .exec = _ccv_cnnp_model_clip_grad_norm_reduce_norm2 | |||
| 27 | }; | |||
| 28 | ||||
| 29 | static int _ccv_cnnp_model_clip_grad_norm_scatter_norm2(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) | |||
| 30 | { | |||
| 31 | const int device_id = CCV_TENSOR_GET_DEVICE_ID(inputs[0]->info.type)(((inputs[0]->info.type) & 0xfff00) >> 8); | |||
| 32 | ccv_nnc_tensor_t* const norm2 = inputs[1 + device_id * 2]; | |||
| 33 | ccv_nnc_cmd_exec(CMD_MUL_FORWARD(1)ccv_nnc_cmd(CCV_NNC_MUL_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={1,}}}, 0), hint, flags, TENSOR_LIST(inputs[0], norm2)(ccv_nnc_tensor_t* []){inputs[0], norm2}, (1 +1 +1 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(outputs[0])(ccv_nnc_tensor_t* []){outputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 34 | return CCV_NNC_EXEC_SUCCESS; | |||
| 35 | } | |||
| 36 | ||||
| 37 | static ccv_nnc_cmd_vtab_t clip_grad_norm_scatter_norm2_vtab = { | |||
| 38 | .exec = _ccv_cnnp_model_clip_grad_norm_scatter_norm2 | |||
| 39 | }; | |||
| 40 | ||||
| 41 | void ccv_cnnp_model_parameters_clip_grad_norm(ccv_cnnp_model_t* const model, const ccv_cnnp_model_io_t parameters, int norm_type, float max_norm, ccv_nnc_stream_context_t* const stream_context) | |||
| 42 | { | |||
| 43 | assert(norm_type == 2)((void) sizeof ((norm_type == 2) ? 1 : 0), __extension__ ({ if (norm_type == 2) ; else __assert_fail ("norm_type == 2", "ccv_cnnp_model_addons.c" , 43, __extension__ __PRETTY_FUNCTION__); })); | |||
| 44 | ccv_cnnp_compiled_data_t* const compiled_data = model->compiled_data; | |||
| 45 | assert(compiled_data)((void) sizeof ((compiled_data) ? 1 : 0), __extension__ ({ if (compiled_data) ; else __assert_fail ("compiled_data", "ccv_cnnp_model_addons.c" , 45, __extension__ __PRETTY_FUNCTION__); })); | |||
| 46 | const int parallel_count = ccv_max(model->parallel_count, 1)({ typeof (model->parallel_count) _a = (model->parallel_count ); typeof (1) _b = (1); (_a > _b) ? _a : _b; }); | |||
| 47 | ccv_nnc_tensor_t* norm2[parallel_count * 2]; | |||
| 48 | ccv_nnc_tensor_t* max_normt[parallel_count]; | |||
| 49 | const int stream_type = model->compiled_data->stream_type; | |||
| 50 | int i; | |||
| 51 | if (stream_type == CCV_STREAM_CONTEXT_GPU) | |||
| 52 | { | |||
| 53 | for (i = 0; i < parallel_count; i++) | |||
| 54 | { | |||
| 55 | ccv_nnc_tensor_param_t info = { | |||
| 56 | .type = CCV_TENSOR_GPU_MEMORY, | |||
| 57 | .format = CCV_TENSOR_FORMAT_NHWC, | |||
| 58 | .datatype = CCV_32F, | |||
| 59 | .dim = {1}, | |||
| 60 | }; | |||
| 61 | CCV_TENSOR_SET_DEVICE_ID(info.type, i)(info.type) = (((info.type) & ~0xfff00) | (((i) & 0xfff ) << 8)); | |||
| 62 | norm2[i * 2] = ccv_nnc_tensor_new(ccv_nnc_xpu_alloc(&compiled_data->xpu_alloc, i, stream_context, ccv_nnc_tensor_data_size(info)), info, 0); | |||
| 63 | norm2[i * 2 + 1] = ccv_nnc_tensor_new(ccv_nnc_xpu_alloc(&compiled_data->xpu_alloc, i, stream_context, ccv_nnc_tensor_data_size(info)), info, 0); | |||
| 64 | max_normt[i] = ccv_nnc_tensor_new(ccv_nnc_xpu_alloc(&compiled_data->xpu_alloc, i, stream_context, ccv_nnc_tensor_data_size(info)), info, 0); | |||
| 65 | } | |||
| 66 | } else { | |||
| 67 | for (i = 0; i < parallel_count; i++) | |||
| 68 | { | |||
| 69 | ccv_nnc_tensor_param_t info = { | |||
| 70 | .type = CCV_TENSOR_CPU_MEMORY, | |||
| 71 | .format = CCV_TENSOR_FORMAT_NHWC, | |||
| 72 | .datatype = CCV_32F, | |||
| 73 | .dim = {1}, | |||
| 74 | }; | |||
| 75 | norm2[i * 2] = ccv_nnc_tensor_new(0, info, 0); | |||
| 76 | norm2[i * 2 + 1] = ccv_nnc_tensor_new(0, info, 0); | |||
| 77 | max_normt[i] = ccv_nnc_tensor_new(0, info, 0); | |||
| 78 | } | |||
| 79 | } | |||
| 80 | // zero out old norm2. | |||
| 81 | if (parallel_count > 1) | |||
| 82 | { | |||
| 83 | ccv_nnc_stream_context_t* streams[parallel_count]; | |||
| 84 | ccv_nnc_stream_signal_t* signal; | |||
| 85 | if (stream_context) | |||
| 86 | signal = ccv_nnc_stream_context_emit_signal_new(stream_context); | |||
| 87 | for (i = 0; i < parallel_count; i++) | |||
| 88 | { | |||
| 89 | const int stream_type = CCV_TENSOR_GET_MEMORY(norm2[i * 2]->info.type)((norm2[i * 2]->info.type) & 0x3) == CCV_TENSOR_GPU_MEMORY ? CCV_STREAM_CONTEXT_GPU : CCV_STREAM_CONTEXT_CPU; | |||
| 90 | const int device_id = CCV_TENSOR_GET_DEVICE_ID(norm2[i * 2]->info.type)(((norm2[i * 2]->info.type) & 0xfff00) >> 8); | |||
| 91 | int type = stream_type; | |||
| 92 | CCV_STREAM_SET_DEVICE_ID(type, device_id)(type) = (((type) & ~0xfff00) | (((device_id) & 0xfff ) << 8)); | |||
| 93 | ccv_nnc_stream_context_t* const stream_0 = ccv_cnnp_compiled_data_get_stream(compiled_data, type); | |||
| 94 | // Wait signal to finish. | |||
| 95 | if (stream_context) | |||
| 96 | ccv_nnc_stream_context_wait_signal(stream_0, signal); | |||
| 97 | ccv_nnc_cmd_exec(CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, 0, TENSOR_LIST(norm2[i * 2])(ccv_nnc_tensor_t* []){norm2[i * 2]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_0); | |||
| 98 | if (stream_context) | |||
| 99 | { | |||
| 100 | ccv_nnc_stream_signal_t* const signal = ccv_nnc_stream_context_emit_signal_new(stream_0); | |||
| 101 | ccv_nnc_stream_context_wait_signal(stream_context, signal); | |||
| 102 | } | |||
| 103 | streams[i] = stream_0; | |||
| 104 | } | |||
| 105 | // If this should be blocking, blocking it. | |||
| 106 | if (!stream_context) | |||
| 107 | for (i = 0; i < parallel_count; i++) | |||
| 108 | if (streams[i]) | |||
| 109 | ccv_nnc_stream_context_wait(streams[i]); | |||
| 110 | } else { | |||
| 111 | ccv_nnc_cmd_exec(CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, 0, TENSOR_LIST(norm2[0])(ccv_nnc_tensor_t* []){norm2[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 112 | } | |||
| 113 | // Gather norm2. | |||
| 114 | ccv_nnc_cmd_t reduce_cmd = { | |||
| 115 | .cmd = CCV_NNC_CUSTOM_FORWARD, | |||
| 116 | .isa = &clip_grad_norm_reduce_norm2_vtab, | |||
| 117 | }; | |||
| 118 | ccv_cnnp_model_parameter_gradients_map(model, parameters, reduce_cmd, ccv_nnc_no_hint, 0, 0, 0, norm2, parallel_count * 2, stream_context); | |||
| 119 | // Now compute max(max_norm / norm2, 1.0). | |||
| 120 | if (parallel_count > 1) | |||
| 121 | { | |||
| 122 | ccv_nnc_stream_context_t* streams[parallel_count]; | |||
| 123 | ccv_nnc_stream_signal_t* signal; | |||
| 124 | if (stream_context) | |||
| 125 | signal = ccv_nnc_stream_context_emit_signal_new(stream_context); | |||
| 126 | for (i = 0; i < parallel_count; i++) | |||
| 127 | { | |||
| 128 | const int stream_type = CCV_TENSOR_GET_MEMORY(norm2[i * 2]->info.type)((norm2[i * 2]->info.type) & 0x3) == CCV_TENSOR_GPU_MEMORY ? CCV_STREAM_CONTEXT_GPU : CCV_STREAM_CONTEXT_CPU; | |||
| 129 | const int device_id = CCV_TENSOR_GET_DEVICE_ID(norm2[i * 2]->info.type)(((norm2[i * 2]->info.type) & 0xfff00) >> 8); | |||
| 130 | int type = stream_type; | |||
| 131 | CCV_STREAM_SET_DEVICE_ID(type, device_id)(type) = (((type) & ~0xfff00) | (((device_id) & 0xfff ) << 8)); | |||
| 132 | ccv_nnc_stream_context_t* const stream_0 = ccv_cnnp_compiled_data_get_stream(compiled_data, type); | |||
| 133 | // Wait signal to finish. | |||
| 134 | if (stream_context) | |||
| 135 | ccv_nnc_stream_context_wait_signal(stream_0, signal); | |||
| 136 | ccv_nnc_cmd_exec(CMD_EWSQRT_FORWARD()ccv_nnc_cmd(CCV_NNC_EWSQRT_FORWARD, 0, ccv_nnc_cmd_auto, 0), ccv_nnc_no_hint, 0, TENSOR_LIST(norm2[i * 2])(ccv_nnc_tensor_t* []){norm2[i * 2]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(norm2[i * 2])(ccv_nnc_tensor_t* []){norm2[i * 2]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_0); | |||
| 137 | ccv_nnc_cmd_exec(CMD_SET_FORWARD(max_norm)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={max_norm,}}}, 0), ccv_nnc_no_hint, 0, 0, 0, TENSOR_LIST(max_normt[i])(ccv_nnc_tensor_t* []){max_normt[i]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_0); | |||
| 138 | ccv_nnc_cmd_exec(CMD_EWDIV_FORWARD()ccv_nnc_cmd(CCV_NNC_EWDIV_FORWARD, 0, ccv_nnc_cmd_auto, 0), ccv_nnc_no_hint, 0, TENSOR_LIST(max_normt[i], norm2[i * 2])(ccv_nnc_tensor_t* []){max_normt[i], norm2[i * 2]}, (1 +1 +1 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1 ), TENSOR_LIST(norm2[i * 2])(ccv_nnc_tensor_t* []){norm2[i * 2]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_0); | |||
| 139 | ccv_nnc_cmd_exec(CMD_CLAMP_FORWARD(NAN, 1)ccv_nnc_cmd(CCV_NNC_CLAMP_FORWARD, 0, (ccv_nnc_cmd_param_t){. size={.dim={1,1,1}},.clamp={.min=(__builtin_nanf ("")),.max=1 }}, 0), ccv_nnc_no_hint, 0, TENSOR_LIST(norm2[i * 2])(ccv_nnc_tensor_t* []){norm2[i * 2]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(norm2[i * 2])(ccv_nnc_tensor_t* []){norm2[i * 2]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_0); | |||
| 140 | if (stream_context) | |||
| 141 | { | |||
| 142 | ccv_nnc_stream_signal_t* const signal = ccv_nnc_stream_context_emit_signal_new(stream_0); | |||
| 143 | ccv_nnc_stream_context_wait_signal(stream_context, signal); | |||
| 144 | } | |||
| 145 | streams[i] = stream_0; | |||
| 146 | } | |||
| 147 | // If this should be blocking, blocking it. | |||
| 148 | if (!stream_context) | |||
| 149 | for (i = 0; i < parallel_count; i++) | |||
| 150 | if (streams[i]) | |||
| 151 | ccv_nnc_stream_context_wait(streams[i]); | |||
| 152 | } else { | |||
| 153 | ccv_nnc_cmd_exec(CMD_EWSQRT_FORWARD()ccv_nnc_cmd(CCV_NNC_EWSQRT_FORWARD, 0, ccv_nnc_cmd_auto, 0), ccv_nnc_no_hint, 0, TENSOR_LIST(norm2[0])(ccv_nnc_tensor_t* []){norm2[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(norm2[0])(ccv_nnc_tensor_t* []){norm2[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 154 | ccv_nnc_cmd_exec(CMD_SET_FORWARD(max_norm)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={max_norm,}}}, 0), ccv_nnc_no_hint, 0, 0, 0, TENSOR_LIST(max_normt[0])(ccv_nnc_tensor_t* []){max_normt[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 155 | ccv_nnc_cmd_exec(CMD_EWDIV_FORWARD()ccv_nnc_cmd(CCV_NNC_EWDIV_FORWARD, 0, ccv_nnc_cmd_auto, 0), ccv_nnc_no_hint, 0, TENSOR_LIST(max_normt[0], norm2[0])(ccv_nnc_tensor_t* []){max_normt[0], norm2[0]}, (1 +1 +1 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(norm2[0])(ccv_nnc_tensor_t* []){norm2[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 156 | ccv_nnc_cmd_exec(CMD_CLAMP_FORWARD(NAN, 1)ccv_nnc_cmd(CCV_NNC_CLAMP_FORWARD, 0, (ccv_nnc_cmd_param_t){. size={.dim={1,1,1}},.clamp={.min=(__builtin_nanf ("")),.max=1 }}, 0), ccv_nnc_no_hint, 0, TENSOR_LIST(norm2[0])(ccv_nnc_tensor_t* []){norm2[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(norm2[0])(ccv_nnc_tensor_t* []){norm2[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 157 | } | |||
| 158 | ccv_nnc_cmd_t scatter_cmd = { | |||
| 159 | .cmd = CCV_NNC_CUSTOM_FORWARD, | |||
| 160 | .isa = &clip_grad_norm_scatter_norm2_vtab, | |||
| 161 | }; | |||
| 162 | ccv_cnnp_model_parameter_gradients_map(model, parameters, scatter_cmd, ccv_nnc_no_hint, 0, norm2, parallel_count * 2, 0, 0, stream_context); | |||
| 163 | if (stream_type == CCV_STREAM_CONTEXT_GPU) | |||
| 164 | for (i = 0; i < parallel_count; i++) | |||
| 165 | { | |||
| 166 | ccv_nnc_xpu_free(&compiled_data->xpu_alloc, norm2[i * 2]->data.u8); | |||
| 167 | ccv_nnc_xpu_free(&compiled_data->xpu_alloc, norm2[i * 2 + 1]->data.u8); | |||
| 168 | ccv_nnc_xpu_free(&compiled_data->xpu_alloc, max_normt[i]->data.u8); | |||
| 169 | } | |||
| 170 | for (i = 0; i < parallel_count; i++) | |||
| 171 | { | |||
| 172 | ccv_nnc_tensor_free(norm2[i * 2]); | |||
| 173 | ccv_nnc_tensor_free(norm2[i * 2 + 1]); | |||
| 174 | ccv_nnc_tensor_free(max_normt[i]); | |||
| 175 | } | |||
| 176 | } | |||
| 177 | ||||
| 178 | // MARK - Add-on Functions | |||
| 179 | ||||
| 180 | static int _ccv_cnnp_model_isnan(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) | |||
| 181 | { | |||
| 182 | const int device_id = CCV_TENSOR_GET_DEVICE_ID(inputs[0]->info.type)(((inputs[0]->info.type) & 0xfff00) >> 8); | |||
| 183 | ccv_nnc_tensor_t* const old_isnanr = outputs[1 + device_id * 2]; | |||
| 184 | ccv_nnc_tensor_t* const isnanr = outputs[1 + device_id * 2 + 1]; | |||
| 185 | ccv_nnc_cmd_t reduce_cmd = CMD_REDUCE_ISNAN_FORWARD()ccv_nnc_cmd(CCV_NNC_REDUCE_ISNAN_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}} ), 0); | |||
| 186 | reduce_cmd.info.reduce.count = ccv_nnc_tensor_nd(inputs[0]->info.dim); | |||
| 187 | int i; | |||
| 188 | for (i = 0; i < cmd.info.reduce.count; i++) | |||
| 189 | reduce_cmd.info.reduce.axis[i] = i; | |||
| 190 | ccv_nnc_cmd_exec(reduce_cmd, hint, flags, TENSOR_LIST(inputs[0])(ccv_nnc_tensor_t* []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(isnanr)(ccv_nnc_tensor_t* []){isnanr}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 191 | ccv_nnc_cmd_exec(CMD_EWSUM_FORWARD()ccv_nnc_cmd(CCV_NNC_EWSUM_FORWARD, 0, ccv_nnc_cmd_auto, 0), hint, flags, TENSOR_LIST(old_isnanr, isnanr)(ccv_nnc_tensor_t* []){old_isnanr, isnanr}, (1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(old_isnanr)(ccv_nnc_tensor_t* []){old_isnanr}, (1 +1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 192 | return CCV_NNC_EXEC_SUCCESS; | |||
| 193 | } | |||
| 194 | ||||
| 195 | static ccv_nnc_cmd_vtab_t reduce_isnan_vtab = { | |||
| 196 | .exec = _ccv_cnnp_model_isnan | |||
| 197 | }; | |||
| 198 | ||||
| 199 | int ccv_cnnp_model_parameter_gradients_isnan(ccv_cnnp_model_t* const model, const ccv_cnnp_model_io_t parameters, ccv_nnc_stream_context_t* const stream_context) | |||
| 200 | { | |||
| 201 | ccv_cnnp_compiled_data_t* const compiled_data = model->compiled_data; | |||
| 202 | assert(compiled_data)((void) sizeof ((compiled_data) ? 1 : 0), __extension__ ({ if (compiled_data) ; else __assert_fail ("compiled_data", "ccv_cnnp_model_addons.c" , 202, __extension__ __PRETTY_FUNCTION__); })); | |||
| 203 | const int parallel_count = ccv_max(model->parallel_count, 1)({ typeof (model->parallel_count) _a = (model->parallel_count ); typeof (1) _b = (1); (_a > _b) ? _a : _b; }); | |||
| 204 | ccv_nnc_tensor_t* isnanr[parallel_count * 2]; | |||
| 205 | const int stream_type = model->compiled_data->stream_type; | |||
| 206 | int i; | |||
| 207 | if (stream_type == CCV_STREAM_CONTEXT_GPU) | |||
| 208 | { | |||
| 209 | for (i = 0; i < parallel_count; i++) | |||
| 210 | { | |||
| 211 | ccv_nnc_tensor_param_t info = { | |||
| 212 | .type = CCV_TENSOR_GPU_MEMORY, | |||
| 213 | .format = CCV_TENSOR_FORMAT_NHWC, | |||
| 214 | .datatype = CCV_32S, | |||
| 215 | .dim = {1}, | |||
| 216 | }; | |||
| 217 | CCV_TENSOR_SET_DEVICE_ID(info.type, i)(info.type) = (((info.type) & ~0xfff00) | (((i) & 0xfff ) << 8)); | |||
| 218 | isnanr[i * 2] = ccv_nnc_tensor_new(ccv_nnc_xpu_alloc(&compiled_data->xpu_alloc, i, stream_context, ccv_nnc_tensor_data_size(info)), info, 0); | |||
| 219 | isnanr[i * 2 + 1] = ccv_nnc_tensor_new(ccv_nnc_xpu_alloc(&compiled_data->xpu_alloc, i, stream_context, ccv_nnc_tensor_data_size(info)), info, 0); | |||
| 220 | } | |||
| 221 | } else { | |||
| 222 | for (i = 0; i < parallel_count; i++) | |||
| 223 | { | |||
| 224 | ccv_nnc_tensor_param_t info = { | |||
| 225 | .type = CCV_TENSOR_CPU_MEMORY, | |||
| 226 | .format = CCV_TENSOR_FORMAT_NHWC, | |||
| 227 | .datatype = CCV_32S, | |||
| 228 | .dim = {1}, | |||
| 229 | }; | |||
| 230 | isnanr[i * 2] = ccv_nnc_tensor_new(0, info, 0); | |||
| 231 | isnanr[i * 2 + 1] = ccv_nnc_tensor_new(0, info, 0); | |||
| 232 | } | |||
| 233 | } | |||
| 234 | // zero out old isnanr. | |||
| 235 | if (parallel_count > 1) | |||
| 236 | { | |||
| 237 | ccv_nnc_stream_context_t* streams[parallel_count]; | |||
| 238 | ccv_nnc_stream_signal_t* signal; | |||
| 239 | if (stream_context) | |||
| 240 | signal = ccv_nnc_stream_context_emit_signal_new(stream_context); | |||
| 241 | for (i = 0; i < parallel_count; i++) | |||
| 242 | { | |||
| 243 | const int stream_type = CCV_TENSOR_GET_MEMORY(isnanr[i * 2]->info.type)((isnanr[i * 2]->info.type) & 0x3) == CCV_TENSOR_GPU_MEMORY ? CCV_STREAM_CONTEXT_GPU : CCV_STREAM_CONTEXT_CPU; | |||
| 244 | const int device_id = CCV_TENSOR_GET_DEVICE_ID(isnanr[i * 2]->info.type)(((isnanr[i * 2]->info.type) & 0xfff00) >> 8); | |||
| 245 | int type = stream_type; | |||
| 246 | CCV_STREAM_SET_DEVICE_ID(type, device_id)(type) = (((type) & ~0xfff00) | (((device_id) & 0xfff ) << 8)); | |||
| 247 | ccv_nnc_stream_context_t* const stream_0 = ccv_cnnp_compiled_data_get_stream(compiled_data, type); | |||
| 248 | // Wait signal to finish. | |||
| 249 | if (stream_context) | |||
| 250 | ccv_nnc_stream_context_wait_signal(stream_0, signal); | |||
| 251 | ccv_nnc_cmd_exec(CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, 0, TENSOR_LIST(isnanr[i * 2])(ccv_nnc_tensor_t* []){isnanr[i * 2]}, (1 +1 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_0); | |||
| 252 | if (stream_context) | |||
| 253 | { | |||
| 254 | ccv_nnc_stream_signal_t* const signal = ccv_nnc_stream_context_emit_signal_new(stream_0); | |||
| 255 | ccv_nnc_stream_context_wait_signal(stream_context, signal); | |||
| 256 | } | |||
| 257 | streams[i] = stream_0; | |||
| 258 | } | |||
| 259 | // If this should be blocking, blocking it. | |||
| 260 | if (!stream_context) | |||
| 261 | for (i = 0; i < parallel_count; i++) | |||
| 262 | if (streams[i]) | |||
| 263 | ccv_nnc_stream_context_wait(streams[i]); | |||
| 264 | } else | |||
| 265 | ccv_nnc_cmd_exec(CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, 0, TENSOR_LIST(isnanr[0])(ccv_nnc_tensor_t* []){isnanr[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), stream_context); | |||
| 266 | // Gather isnanr. | |||
| 267 | ccv_nnc_cmd_t reduce_cmd = { | |||
| 268 | .cmd = CCV_NNC_CUSTOM_FORWARD, | |||
| 269 | .isa = &reduce_isnan_vtab, | |||
| 270 | }; | |||
| 271 | ccv_cnnp_model_parameter_gradients_map(model, parameters, reduce_cmd, ccv_nnc_no_hint, 0, 0, 0, isnanr, parallel_count * 2, stream_context); | |||
| 272 | for (i = 0; i < parallel_count; i++) | |||
| 273 | ccv_nnc_tensor_free(isnanr[i * 2 + 1]); | |||
| 274 | int retval = 0; | |||
| 275 | if (stream_type == CCV_TENSOR_GPU_MEMORY) | |||
| 276 | { | |||
| 277 | ccv_nnc_tensor_param_t info = { | |||
| 278 | .type = CCV_TENSOR_CPU_MEMORY, | |||
| 279 | .format = CCV_TENSOR_FORMAT_NHWC, | |||
| 280 | .datatype = CCV_32S, | |||
| 281 | .dim = {1}, | |||
| 282 | }; | |||
| 283 | ccv_nnc_tensor_t* checknan = ccv_nnc_tensor_new(0, info, 0); | |||
| 284 | for (i = 0; i < parallel_count; i++) | |||
| 285 | { | |||
| 286 | ccv_nnc_cmd_exec(CMD_DATA_TRANSFER_FORWARD()ccv_nnc_cmd(CCV_NNC_DATA_TRANSFER_FORWARD, 0, ccv_nnc_cmd_auto , 0), ccv_nnc_no_hint, 0, TENSOR_LIST(isnanr[i * 2])(ccv_nnc_tensor_t* []){isnanr[i * 2]}, (1 +1 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_LIST(checknan)(ccv_nnc_tensor_t* []){checknan}, (1 +1 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), 0); | |||
| 287 | if (checknan->data.i32[0] > 0) | |||
| 288 | { | |||
| 289 | retval = 1; | |||
| 290 | break; | |||
| 291 | } | |||
| 292 | } | |||
| 293 | ccv_nnc_tensor_free(checknan); | |||
| 294 | } else { | |||
| 295 | for (i = 0; i < parallel_count; i++) | |||
| 296 | if (isnanr[i * 2]->data.i32[0] > 0) | |||
| 297 | { | |||
| 298 | retval = 1; | |||
| 299 | break; | |||
| 300 | } | |||
| 301 | } | |||
| 302 | for (i = 0; i < parallel_count; i++) | |||
| 303 | ccv_nnc_tensor_free(isnanr[i * 2]); | |||
| 304 | return retval; | |||
| 305 | } | |||
| 306 | ||||
| 307 | // MARK - Core Layers | |||
| 308 | ||||
| 309 | static void _ccv_cnnp_sum_build(ccv_cnnp_model_t* const self, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 310 | { | |||
| 311 | PRINT(CCV_CLI_VERBOSE, "[cnnp_sum_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_sum_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 312 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 312, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 313 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, ccv_nnc_tensor_symbol_params(graph, inputs[0]), 0); | |||
| 314 | ccv_nnc_graph_exec_symbol_new(graph, CMD_EWSUM_FORWARD()ccv_nnc_cmd(CCV_NNC_EWSUM_FORWARD, 0, ccv_nnc_cmd_auto, 0), inputs, input_size, outputs, output_size, 0); | |||
| 315 | } | |||
| 316 | ||||
| 317 | static ccv_cnnp_model_t* _ccv_cnnp_sum_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 318 | ||||
| 319 | static const ccv_cnnp_model_vtab_t ccv_cnnp_sum_isa = { | |||
| 320 | .build = _ccv_cnnp_sum_build, | |||
| 321 | .copy = _ccv_cnnp_sum_copy, | |||
| 322 | }; | |||
| 323 | ||||
| 324 | typedef struct { | |||
| 325 | ccv_cnnp_model_t super; | |||
| 326 | ccv_nnc_tensor_symbol_t output; | |||
| 327 | } ccv_cnnp_model_sum_t; | |||
| 328 | ||||
| 329 | ccv_cnnp_model_t* ccv_cnnp_sum(const char* const name) | |||
| 330 | { | |||
| 331 | ccv_cnnp_model_sum_t* const model_sum = (ccv_cnnp_model_sum_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_sum_t)); | |||
| 332 | model_sum->super.isa = &ccv_cnnp_sum_isa; | |||
| 333 | model_sum->super.input_size = 0; | |||
| 334 | model_sum->super.outputs = &model_sum->output; | |||
| 335 | model_sum->super.output_size = 1; | |||
| 336 | ccv_cnnp_model_copy_name(&model_sum->super, name); | |||
| 337 | return (ccv_cnnp_model_t*)model_sum; | |||
| 338 | } | |||
| 339 | ||||
| 340 | static ccv_cnnp_model_t* _ccv_cnnp_sum_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 341 | { | |||
| 342 | return ccv_cnnp_sum(self->name); | |||
| 343 | } | |||
| 344 | ||||
| 345 | typedef struct { | |||
| 346 | ccv_cnnp_model_t super; | |||
| 347 | int axis; | |||
| 348 | ccv_nnc_tensor_symbol_t output; | |||
| 349 | } ccv_cnnp_model_concat_t; | |||
| 350 | ||||
| 351 | static void _ccv_cnnp_concat_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 352 | { | |||
| 353 | const ccv_cnnp_model_concat_t* const self = (const ccv_cnnp_model_concat_t*)super; | |||
| 354 | PRINT(CCV_CLI_VERBOSE, "[cnnp_concat_build] 1. -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_concat_build] 1. -\n"); fflush(stdout); } } while (0); | |||
| ||||
| 355 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 355, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 356 | ccv_nnc_tensor_param_t output_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 357 | int i, j; | |||
| 358 | if (output_params.dim[0] == 0) | |||
| 359 | for (i = 1; i < input_size; i++) | |||
| 360 | { | |||
| 361 | output_params = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 362 | if (output_params.dim[0] != 0) | |||
| 363 | break; | |||
| 364 | } | |||
| 365 | const int nd = ccv_nnc_tensor_nd(output_params.dim); | |||
| 366 | const int axis = self->axis; | |||
| 367 | assert(axis < nd)((void) sizeof ((axis < nd) ? 1 : 0), __extension__ ({ if ( axis < nd) ; else __assert_fail ("axis < nd", "ccv_cnnp_model_addons.c" , 367, __extension__ __PRETTY_FUNCTION__); })); | |||
| 368 | output_params.dim[axis] = 0; | |||
| 369 | int input_is_contiguous = 1; | |||
| 370 | for (i = 0; i < input_size; i++) | |||
| 371 | { | |||
| 372 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 373 | const int input_nd = ccv_nnc_tensor_nd(input_params.dim); | |||
| 374 | if (input_nd == 0) | |||
| 375 | { | |||
| 376 | PRINT(CCV_CLI_VERBOSE, "[cnnp_concat_build] %d. input[%d]: -\n", i + 2, i)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_concat_build] %d. input[%d]: -\n", i + 2, i) ; fflush(stdout); } } while (0); | |||
| 377 | input_is_contiguous = 0; | |||
| 378 | continue; | |||
| 379 | } | |||
| 380 | if (CCV_CLI_OUTPUT_LEVEL_IS(CCV_CLI_VERBOSE)(CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) | |||
| 381 | { | |||
| 382 | PRINT(CCV_CLI_VERBOSE, "[cnnp_concat_build] %d. input[%d]: (%d", i + 2, i, input_params.dim[0])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_concat_build] %d. input[%d]: (%d", i + 2, i, input_params.dim[0]); fflush(stdout); } } while (0); | |||
| 383 | int i; | |||
| 384 | for (i = 1; i < CCV_NNC_MAX_DIM_ALLOC(12) && input_params.dim[i] > 0; i++) | |||
| 385 | PRINT(CCV_CLI_VERBOSE, ", %d", input_params.dim[i])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(", %d", input_params.dim[i]); fflush(stdout); } } while (0); | |||
| 386 | PRINT(CCV_CLI_VERBOSE, ")\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(")\n"); fflush(stdout); } } while (0); | |||
| 387 | } | |||
| 388 | assert(input_nd == nd)((void) sizeof ((input_nd == nd) ? 1 : 0), __extension__ ({ if (input_nd == nd) ; else __assert_fail ("input_nd == nd", "ccv_cnnp_model_addons.c" , 388, __extension__ __PRETTY_FUNCTION__); })); | |||
| 389 | for (j = 0; j < nd; j++) | |||
| 390 | if (j != axis) | |||
| 391 | { assert(input_params.dim[j] == output_params.dim[j])((void) sizeof ((input_params.dim[j] == output_params.dim[j]) ? 1 : 0), __extension__ ({ if (input_params.dim[j] == output_params .dim[j]) ; else __assert_fail ("input_params.dim[j] == output_params.dim[j]" , "ccv_cnnp_model_addons.c", 391, __extension__ __PRETTY_FUNCTION__ ); })); } | |||
| 392 | output_params.dim[axis] += input_params.dim[axis]; | |||
| 393 | } | |||
| 394 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 395 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 396 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 397 | ccv_nnc_tensor_get_stride(output_params.dim, stride); | |||
| 398 | if (input_is_contiguous
| |||
| 399 | { | |||
| 400 | ccv_nnc_tensor_symbol_t aliases[input_size]; | |||
| ||||
| 401 | for (i = 0; i < input_size; i++) | |||
| 402 | { | |||
| 403 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 404 | aliases[i] = ccv_nnc_tensor_symbol_alias_new(graph, outputs[0], ofs, stride, input_params, 0); | |||
| 405 | ofs[axis] += input_params.dim[axis]; | |||
| 406 | } | |||
| 407 | // Format transform is more flexible. | |||
| 408 | ccv_nnc_graph_exec_symbol_new(graph, CMD_FORMAT_TRANSFORM_FORWARD()ccv_nnc_cmd(CCV_NNC_FORMAT_TRANSFORM_FORWARD, 0, ccv_nnc_cmd_auto , 0), inputs, input_size, aliases, input_size, "concat"); | |||
| 409 | } else { | |||
| 410 | ccv_nnc_tensor_symbol_t aliases[input_size]; | |||
| 411 | for (i = 0; i < input_size; i++) | |||
| 412 | { | |||
| 413 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 414 | if (input_params.dim[0] == 0) | |||
| 415 | { | |||
| 416 | // Create a new alias anyway, but not going to use it, in this way, the alias count will match during absorb. | |||
| 417 | aliases[i] = ccv_nnc_tensor_symbol_alias_new(graph, outputs[0], ofs, stride, input_params, 0); | |||
| 418 | continue; | |||
| 419 | } | |||
| 420 | aliases[i] = ccv_nnc_tensor_symbol_alias_new(graph, outputs[0], ofs, stride, input_params, 0); | |||
| 421 | ofs[axis] += input_params.dim[axis]; | |||
| 422 | } | |||
| 423 | // Format transform is more flexible. | |||
| 424 | ccv_nnc_graph_exec_symbol_new(graph, CMD_FORMAT_TRANSFORM_FORWARD()ccv_nnc_cmd(CCV_NNC_FORMAT_TRANSFORM_FORWARD, 0, ccv_nnc_cmd_auto , 0), inputs, input_size, aliases, input_size, "concat"); | |||
| 425 | } | |||
| 426 | } | |||
| 427 | ||||
| 428 | static ccv_cnnp_model_t* _ccv_cnnp_concat_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 429 | ||||
| 430 | static const ccv_cnnp_model_vtab_t ccv_cnnp_concat_isa = { | |||
| 431 | .build = _ccv_cnnp_concat_build, | |||
| 432 | .copy = _ccv_cnnp_concat_copy, | |||
| 433 | }; | |||
| 434 | ||||
| 435 | ccv_cnnp_model_t* ccv_cnnp_concat(const int axis, const char* const name) | |||
| 436 | { | |||
| 437 | ccv_cnnp_model_concat_t* const model_concat = (ccv_cnnp_model_concat_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_concat_t)); | |||
| 438 | model_concat->super.isa = &ccv_cnnp_concat_isa; | |||
| 439 | model_concat->super.input_size = 0; | |||
| 440 | model_concat->super.outputs = &model_concat->output; | |||
| 441 | model_concat->super.output_size = 1; | |||
| 442 | model_concat->axis = axis; | |||
| 443 | ccv_cnnp_model_copy_name(&model_concat->super, name); | |||
| 444 | return (ccv_cnnp_model_t*)model_concat; | |||
| 445 | } | |||
| 446 | ||||
| 447 | static ccv_cnnp_model_t* _ccv_cnnp_concat_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 448 | { | |||
| 449 | const ccv_cnnp_model_concat_t* const self = (const ccv_cnnp_model_concat_t*)super; | |||
| 450 | return ccv_cnnp_concat(self->axis, self->super.name); | |||
| 451 | } | |||
| 452 | ||||
| 453 | typedef struct { | |||
| 454 | ccv_cnnp_model_t super; | |||
| 455 | int axis; | |||
| 456 | ccv_nnc_tensor_symbol_t outputs[1]; | |||
| 457 | } ccv_cnnp_model_chunk_t; | |||
| 458 | ||||
| 459 | static void _ccv_cnnp_chunk_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 460 | { | |||
| 461 | const ccv_cnnp_model_concat_t* const self = (const ccv_cnnp_model_concat_t*)super; | |||
| 462 | PRINT(CCV_CLI_VERBOSE, "[cnnp_chunk_build] 1. axis: %d\n", self->axis)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_chunk_build] 1. axis: %d\n", self->axis); fflush(stdout); } } while (0); | |||
| 463 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 463, __extension__ __PRETTY_FUNCTION__); })); | |||
| 464 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 465 | if (CCV_CLI_OUTPUT_LEVEL_IS(CCV_CLI_VERBOSE)(CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) | |||
| 466 | { | |||
| 467 | PRINT(CCV_CLI_VERBOSE, "[cnnp_chunk_build] 2. input: (%d", input_params.dim[0])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_chunk_build] 2. input: (%d", input_params.dim [0]); fflush(stdout); } } while (0); | |||
| 468 | int i; | |||
| 469 | for (i = 1; i < CCV_NNC_MAX_DIM_ALLOC(12) && input_params.dim[i] > 0; i++) | |||
| 470 | PRINT(CCV_CLI_VERBOSE, ", %d", input_params.dim[i])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(", %d", input_params.dim[i]); fflush(stdout); } } while (0); | |||
| 471 | PRINT(CCV_CLI_VERBOSE, ")\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(")\n"); fflush(stdout); } } while (0); | |||
| 472 | } | |||
| 473 | ccv_nnc_tensor_param_t output_params = input_params; | |||
| 474 | int i; | |||
| 475 | const int nd = ccv_nnc_tensor_nd(output_params.dim); | |||
| 476 | const int axis = self->axis; | |||
| 477 | assert(axis < nd)((void) sizeof ((axis < nd) ? 1 : 0), __extension__ ({ if ( axis < nd) ; else __assert_fail ("axis < nd", "ccv_cnnp_model_addons.c" , 477, __extension__ __PRETTY_FUNCTION__); })); | |||
| 478 | const int n = self->super.output_size; | |||
| 479 | assert(n == output_size)((void) sizeof ((n == output_size) ? 1 : 0), __extension__ ({ if (n == output_size) ; else __assert_fail ("n == output_size" , "ccv_cnnp_model_addons.c", 479, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 480 | assert(output_params.dim[axis] % n == 0)((void) sizeof ((output_params.dim[axis] % n == 0) ? 1 : 0), __extension__ ({ if (output_params.dim[axis] % n == 0) ; else __assert_fail ("output_params.dim[axis] % n == 0", "ccv_cnnp_model_addons.c" , 480, __extension__ __PRETTY_FUNCTION__); })); | |||
| 481 | output_params.dim[axis] = output_params.dim[axis] / n; | |||
| 482 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 483 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 484 | ccv_nnc_tensor_get_stride(input_params.dim, stride); | |||
| 485 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 486 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 487 | { | |||
| 488 | for (i = 0; i < output_size; i++) | |||
| 489 | { | |||
| 490 | outputs[i] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ofs, stride, output_params, 0); | |||
| 491 | ofs[axis] += output_params.dim[axis]; | |||
| 492 | } | |||
| 493 | } else { | |||
| 494 | // Otherwise, we need to check if it is permute. For permute, we cannot do alias directly. | |||
| 495 | // We need to first materialize the permute and then run reshape on top of it, otherwise it will be wrong. | |||
| 496 | int old_stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 497 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], 0, old_stride); | |||
| 498 | // We identify permute by checking if the stride is not in descending order. | |||
| 499 | // This also covered "permute" through reshape, rather than using ccv_cnnp_permute directly. | |||
| 500 | int i, no_permute = 1; | |||
| 501 | for (i = 1; no_permute && i < nd; i++) | |||
| 502 | if (old_stride[i - 1] < old_stride[i]) | |||
| 503 | no_permute = 0; | |||
| 504 | if (no_permute) | |||
| 505 | { // Just straightforward reshape if there is no no permute. | |||
| 506 | for (i = 0; i < output_size; i++) | |||
| 507 | { | |||
| 508 | outputs[i] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ofs, old_stride, output_params, 0); | |||
| 509 | ofs[axis] += output_params.dim[axis]; | |||
| 510 | } | |||
| 511 | } else { | |||
| 512 | // Otherwise, we first do format transform to plain tensor and then do reshape. | |||
| 513 | ccv_nnc_tensor_symbol_t permuted = ccv_nnc_tensor_symbol_new(graph, input_params, 0); | |||
| 514 | ccv_nnc_graph_exec_symbol_new(graph, CMD_FORMAT_TRANSFORM_FORWARD()ccv_nnc_cmd(CCV_NNC_FORMAT_TRANSFORM_FORWARD, 0, ccv_nnc_cmd_auto , 0), TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(permuted)(const ccv_nnc_tensor_symbol_t []){permuted}, (1 +1 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "reshape"); | |||
| 515 | for (i = 0; i < output_size; i++) | |||
| 516 | { | |||
| 517 | outputs[i] = ccv_nnc_tensor_symbol_alias_new(graph, permuted, ofs, stride, output_params, 0); | |||
| 518 | ofs[axis] += output_params.dim[axis]; | |||
| 519 | } | |||
| 520 | } | |||
| 521 | } | |||
| 522 | } | |||
| 523 | ||||
| 524 | static ccv_cnnp_model_t* _ccv_cnnp_chunk_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 525 | ||||
| 526 | static const ccv_cnnp_model_vtab_t ccv_cnnp_chunk_isa = { | |||
| 527 | .build = _ccv_cnnp_chunk_build, | |||
| 528 | .copy = _ccv_cnnp_chunk_copy, | |||
| 529 | }; | |||
| 530 | ||||
| 531 | ccv_cnnp_model_t* ccv_cnnp_chunk(const int n, const int axis, const char* const name) | |||
| 532 | { | |||
| 533 | assert(n >= 1)((void) sizeof ((n >= 1) ? 1 : 0), __extension__ ({ if (n >= 1) ; else __assert_fail ("n >= 1", "ccv_cnnp_model_addons.c" , 533, __extension__ __PRETTY_FUNCTION__); })); | |||
| 534 | ccv_cnnp_model_chunk_t* const model_chunk = (ccv_cnnp_model_chunk_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_chunk_t) + sizeof(ccv_nnc_tensor_symbol_t) * (n - 1)); | |||
| 535 | model_chunk->super.isa = &ccv_cnnp_chunk_isa; | |||
| 536 | model_chunk->super.input_size = 1; | |||
| 537 | model_chunk->super.outputs = model_chunk->outputs; | |||
| 538 | model_chunk->super.output_size = n; | |||
| 539 | model_chunk->axis = axis; | |||
| 540 | ccv_cnnp_model_copy_name(&model_chunk->super, name); | |||
| 541 | return (ccv_cnnp_model_t*)model_chunk; | |||
| 542 | } | |||
| 543 | ||||
| 544 | static ccv_cnnp_model_t* _ccv_cnnp_chunk_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 545 | { | |||
| 546 | const ccv_cnnp_model_chunk_t* const self = (const ccv_cnnp_model_chunk_t*)super; | |||
| 547 | return ccv_cnnp_chunk(self->super.output_size, self->axis, self->super.name); | |||
| 548 | } | |||
| 549 | ||||
| 550 | typedef struct { | |||
| 551 | ccv_cnnp_model_t super; | |||
| 552 | ccv_nnc_tensor_symbol_t output; | |||
| 553 | int format; | |||
| 554 | int dim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 555 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 556 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 557 | } ccv_cnnp_model_reshape_t; | |||
| 558 | ||||
| 559 | static void _ccv_cnnp_reshape_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 560 | { | |||
| 561 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 561, __extension__ __PRETTY_FUNCTION__); })); | |||
| 562 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 562, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 563 | ccv_cnnp_model_reshape_t* const self = (ccv_cnnp_model_reshape_t*)super; | |||
| 564 | if (CCV_CLI_OUTPUT_LEVEL_IS(CCV_CLI_VERBOSE)(CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) | |||
| 565 | { | |||
| 566 | PRINT(CCV_CLI_VERBOSE, "[cnnp_reshape_build] 1. dim: (%d", self->dim[0])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_reshape_build] 1. dim: (%d", self->dim[0] ); fflush(stdout); } } while (0); | |||
| 567 | int i; | |||
| 568 | for (i = 1; i < CCV_NNC_MAX_DIM_ALLOC(12) && self->dim[i] > 0; i++) | |||
| 569 | PRINT(CCV_CLI_VERBOSE, ", %d", self->dim[i])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(", %d", self->dim[i]); fflush(stdout); } } while (0); | |||
| 570 | const int count = i; | |||
| 571 | PRINT(CCV_CLI_VERBOSE, "), ofs: (%d", self->ofs[0])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("), ofs: (%d", self->ofs[0]); fflush(stdout); } } while (0); | |||
| 572 | for (i = 1; i < count; i++) | |||
| 573 | PRINT(CCV_CLI_VERBOSE, ", %d", self->ofs[i])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(", %d", self->ofs[i]); fflush(stdout); } } while (0); | |||
| 574 | PRINT(CCV_CLI_VERBOSE, "), stride: (%d", self->stride[0])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("), stride: (%d", self->stride[0]); fflush(stdout ); } } while (0); | |||
| 575 | for (i = 1; i < count; i++) | |||
| 576 | PRINT(CCV_CLI_VERBOSE, ", %d", self->stride[i])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(", %d", self->stride[i]); fflush(stdout); } } while (0); | |||
| 577 | PRINT(CCV_CLI_VERBOSE, ")\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(")\n"); fflush(stdout); } } while (0); | |||
| 578 | } | |||
| 579 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 580 | if (CCV_CLI_OUTPUT_LEVEL_IS(CCV_CLI_VERBOSE)(CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) | |||
| 581 | { | |||
| 582 | PRINT(CCV_CLI_VERBOSE, "[cnnp_reshape_build] 2. input: (%d", params.dim[0])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_reshape_build] 2. input: (%d", params.dim[0] ); fflush(stdout); } } while (0); | |||
| 583 | int i; | |||
| 584 | for (i = 1; i < CCV_NNC_MAX_DIM_ALLOC(12) && params.dim[i] > 0; i++) | |||
| 585 | PRINT(CCV_CLI_VERBOSE, ", %d", params.dim[i])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(", %d", params.dim[i]); fflush(stdout); } } while ( 0); | |||
| 586 | PRINT(CCV_CLI_VERBOSE, ")\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(")\n"); fflush(stdout); } } while (0); | |||
| 587 | } | |||
| 588 | if (self->format > 0) | |||
| 589 | params.format = self->format; | |||
| 590 | assert(ccv_nnc_dimension_count(self->dim) <= ccv_nnc_tensor_count(params))((void) sizeof ((ccv_nnc_dimension_count(self->dim) <= ccv_nnc_tensor_count (params)) ? 1 : 0), __extension__ ({ if (ccv_nnc_dimension_count (self->dim) <= ccv_nnc_tensor_count(params)) ; else __assert_fail ("ccv_nnc_dimension_count(self->dim) <= ccv_nnc_tensor_count(params)" , "ccv_cnnp_model_addons.c", 590, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 591 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 592 | int stride_from_dim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 593 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 594 | { | |||
| 595 | memcpy(params.dim, self->dim, sizeof(params.dim)); | |||
| 596 | int* stride; | |||
| 597 | if (self->stride[0] == 0) | |||
| 598 | { | |||
| 599 | ccv_nnc_tensor_get_stride(self->dim, stride_from_dim); | |||
| 600 | stride = stride_from_dim; | |||
| 601 | } else | |||
| 602 | stride = self->stride; | |||
| 603 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], self->ofs, stride, params, 0); | |||
| 604 | } else { | |||
| 605 | // Otherwise, we need to check if it is permute. For permute, we cannot do alias directly. | |||
| 606 | // We need to first materialize the permute and then run reshape on top of it, otherwise it will be wrong. | |||
| 607 | int old_stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 608 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], 0, old_stride); | |||
| 609 | // We identify permute by checking if the stride is not in descending order. | |||
| 610 | // This also covered "permute" through reshape, rather than using ccv_cnnp_permute directly. | |||
| 611 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 612 | const int new_nd = ccv_nnc_tensor_nd(self->dim); | |||
| 613 | int i, no_permute = 1; | |||
| 614 | // If the new dim has different nd, or we actually have a stride, we need to check if it is no permute or not. | |||
| 615 | if (new_nd != nd || (self->stride[0] != 0 && memcmp(self->stride, old_stride, sizeof(self->stride)) != 0)) | |||
| 616 | for (i = 1; no_permute && i < nd; i++) | |||
| 617 | if (old_stride[i - 1] < old_stride[i]) | |||
| 618 | no_permute = 0; | |||
| 619 | if (no_permute) | |||
| 620 | { // Just straightforward reshape if there is no no permute. | |||
| 621 | memcpy(params.dim, self->dim, sizeof(params.dim)); | |||
| 622 | int* stride; | |||
| 623 | if (self->stride[0] == 0) | |||
| 624 | { | |||
| 625 | if (new_nd != nd) // Cannot use old stride. | |||
| 626 | { | |||
| 627 | ccv_nnc_tensor_get_stride(self->dim, stride_from_dim); | |||
| 628 | stride = stride_from_dim; | |||
| 629 | } else | |||
| 630 | stride = old_stride; | |||
| 631 | } else | |||
| 632 | stride = self->stride; | |||
| 633 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], self->ofs, stride, params, 0); | |||
| 634 | } else { | |||
| 635 | // Otherwise, we first do format transform to plain tensor and then do reshape. | |||
| 636 | ccv_nnc_tensor_symbol_t permuted = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 637 | ccv_nnc_graph_exec_symbol_new(graph, CMD_FORMAT_TRANSFORM_FORWARD()ccv_nnc_cmd(CCV_NNC_FORMAT_TRANSFORM_FORWARD, 0, ccv_nnc_cmd_auto , 0), TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(permuted)(const ccv_nnc_tensor_symbol_t []){permuted}, (1 +1 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "reshape"); | |||
| 638 | memcpy(params.dim, self->dim, sizeof(params.dim)); | |||
| 639 | int* stride; | |||
| 640 | if (self->stride[0] == 0) | |||
| 641 | { | |||
| 642 | ccv_nnc_tensor_get_stride(self->dim, stride_from_dim); | |||
| 643 | stride = stride_from_dim; | |||
| 644 | } else | |||
| 645 | stride = self->stride; | |||
| 646 | // And then we create alias against the permuted one. | |||
| 647 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, permuted, self->ofs, stride, params, 0); | |||
| 648 | } | |||
| 649 | } | |||
| 650 | } | |||
| 651 | ||||
| 652 | static ccv_cnnp_model_t* _ccv_cnnp_reshape_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 653 | ||||
| 654 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reshape_isa = { | |||
| 655 | .build = _ccv_cnnp_reshape_build, | |||
| 656 | .copy = _ccv_cnnp_reshape_copy, | |||
| 657 | }; | |||
| 658 | ||||
| 659 | ccv_cnnp_model_t* ccv_cnnp_reshape(const int format, const int dim[CCV_NNC_MAX_DIM_ALLOC(12)], const int ofs[CCV_NNC_MAX_DIM_ALLOC(12)], const int stride[CCV_NNC_MAX_DIM_ALLOC(12)], const char* const name) | |||
| 660 | { | |||
| 661 | ccv_cnnp_model_reshape_t* const model_reshape = (ccv_cnnp_model_reshape_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_reshape_t)); | |||
| 662 | model_reshape->super.isa = &ccv_cnnp_reshape_isa; | |||
| 663 | model_reshape->super.input_size = 1; | |||
| 664 | model_reshape->super.outputs = &model_reshape->output; | |||
| 665 | model_reshape->super.output_size = 1; | |||
| 666 | ccv_cnnp_model_copy_name(&model_reshape->super, name); | |||
| 667 | model_reshape->format = format; | |||
| 668 | memcpy(model_reshape->dim, dim, sizeof(model_reshape->dim)); | |||
| 669 | memcpy(model_reshape->ofs, ofs, sizeof(model_reshape->ofs)); | |||
| 670 | if (stride[0] != 0) | |||
| 671 | memcpy(model_reshape->stride, stride, sizeof(model_reshape->stride)); | |||
| 672 | return (ccv_cnnp_model_t*)model_reshape; | |||
| 673 | } | |||
| 674 | ||||
| 675 | static ccv_cnnp_model_t* _ccv_cnnp_reshape_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 676 | { | |||
| 677 | const ccv_cnnp_model_reshape_t* const self = (const ccv_cnnp_model_reshape_t*)super; | |||
| 678 | return ccv_cnnp_reshape(self->format, self->dim, self->ofs, self->stride, self->super.name); | |||
| 679 | } | |||
| 680 | ||||
| 681 | typedef struct { | |||
| 682 | ccv_cnnp_model_t super; | |||
| 683 | ccv_nnc_tensor_symbol_t output; | |||
| 684 | int type; | |||
| 685 | int begin[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 686 | int end[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 687 | } ccv_cnnp_model_pad_t; | |||
| 688 | ||||
| 689 | static void _ccv_cnnp_pad_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 690 | { | |||
| 691 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 691, __extension__ __PRETTY_FUNCTION__); })); | |||
| 692 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 692, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 693 | ccv_cnnp_model_pad_t* const self = (ccv_cnnp_model_pad_t*)super; | |||
| 694 | PRINT(CCV_CLI_VERBOSE, "[cnnp_pad_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_pad_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 695 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 696 | const int nd = ccv_nnc_tensor_nd(input_params.dim); | |||
| 697 | ccv_nnc_tensor_param_t params = input_params; | |||
| 698 | int i; | |||
| 699 | for (i = 0 ; i < nd; i++) | |||
| 700 | params.dim[i] += self->begin[i] + self->end[i]; | |||
| 701 | const ccv_nnc_tensor_symbol_t padded = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 702 | ccv_nnc_cmd_t pad = CMD_PAD_FORWARD(self->type, (), ())ccv_nnc_cmd(CCV_NNC_PAD_FORWARD, 0, ((ccv_nnc_cmd_param_t){.size ={.dim={}},.pad={.type=self->type,.end={}}}), 0); | |||
| 703 | memcpy(pad.info.size.dim, self->begin, sizeof(pad.info.size.dim)); | |||
| 704 | memcpy(pad.info.pad.end, self->end, sizeof(pad.info.pad.end)); | |||
| 705 | ccv_nnc_graph_exec_symbol_new(graph, pad, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(padded)(const ccv_nnc_tensor_symbol_t []){padded}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "pad"); | |||
| 706 | outputs[0] = padded; | |||
| 707 | } | |||
| 708 | ||||
| 709 | static ccv_cnnp_model_t* _ccv_cnnp_pad_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 710 | ||||
| 711 | static const ccv_cnnp_model_vtab_t ccv_cnnp_pad_isa = { | |||
| 712 | .build = _ccv_cnnp_pad_build, | |||
| 713 | .copy = _ccv_cnnp_pad_copy, | |||
| 714 | }; | |||
| 715 | ||||
| 716 | ccv_cnnp_model_t* ccv_cnnp_pad(const int type, const int begin[CCV_NNC_MAX_DIM_ALLOC(12)], const int end[CCV_NNC_MAX_DIM_ALLOC(12)], const char* const name) | |||
| 717 | { | |||
| 718 | ccv_cnnp_model_pad_t* const model_pad = (ccv_cnnp_model_pad_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_pad_t)); | |||
| 719 | model_pad->super.isa = &ccv_cnnp_pad_isa; | |||
| 720 | model_pad->super.input_size = 1; | |||
| 721 | model_pad->super.outputs = &model_pad->output; | |||
| 722 | model_pad->super.output_size = 1; | |||
| 723 | ccv_cnnp_model_copy_name(&model_pad->super, name); | |||
| 724 | model_pad->type = type; | |||
| 725 | memcpy(model_pad->begin, begin, sizeof(model_pad->begin)); | |||
| 726 | memcpy(model_pad->end, end, sizeof(model_pad->end)); | |||
| 727 | return (ccv_cnnp_model_t*)model_pad; | |||
| 728 | } | |||
| 729 | ||||
| 730 | static ccv_cnnp_model_t* _ccv_cnnp_pad_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 731 | { | |||
| 732 | const ccv_cnnp_model_pad_t* const self = (const ccv_cnnp_model_pad_t*)super; | |||
| 733 | return ccv_cnnp_pad(self->type, self->begin, self->end, self->super.name); | |||
| 734 | } | |||
| 735 | ||||
| 736 | typedef struct { | |||
| 737 | ccv_cnnp_model_t super; | |||
| 738 | ccv_nnc_tensor_symbol_t output; | |||
| 739 | } ccv_cnnp_model_identity_t; | |||
| 740 | ||||
| 741 | static void _ccv_cnnp_identity_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 742 | { | |||
| 743 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 743, __extension__ __PRETTY_FUNCTION__); })); | |||
| 744 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 744, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 745 | PRINT(CCV_CLI_VERBOSE, "[cnnp_identity_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_identity_build] -\n"); fflush(stdout); } } while (0); | |||
| 746 | outputs[0] = inputs[0]; | |||
| 747 | } | |||
| 748 | ||||
| 749 | static ccv_cnnp_model_t* _ccv_cnnp_identity_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 750 | ||||
| 751 | static const ccv_cnnp_model_vtab_t ccv_cnnp_identity_isa = { | |||
| 752 | .build = _ccv_cnnp_identity_build, | |||
| 753 | .copy = _ccv_cnnp_identity_copy, | |||
| 754 | }; | |||
| 755 | ||||
| 756 | ccv_cnnp_model_t* ccv_cnnp_identity(const char* const name) | |||
| 757 | { | |||
| 758 | ccv_cnnp_model_identity_t* const model_identity = (ccv_cnnp_model_identity_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_identity_t)); | |||
| 759 | model_identity->super.isa = &ccv_cnnp_identity_isa; | |||
| 760 | model_identity->super.input_size = 1; | |||
| 761 | model_identity->super.outputs = &model_identity->output; | |||
| 762 | model_identity->super.output_size = 1; | |||
| 763 | ccv_cnnp_model_copy_name(&model_identity->super, name); | |||
| 764 | return (ccv_cnnp_model_t*)model_identity; | |||
| 765 | } | |||
| 766 | ||||
| 767 | static ccv_cnnp_model_t* _ccv_cnnp_identity_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 768 | { | |||
| 769 | const ccv_cnnp_model_identity_t* const self = (const ccv_cnnp_model_identity_t*)super; | |||
| 770 | return ccv_cnnp_identity(self->super.name); | |||
| 771 | } | |||
| 772 | ||||
| 773 | typedef struct { | |||
| 774 | ccv_cnnp_model_t super; | |||
| 775 | ccv_nnc_tensor_symbol_t output; | |||
| 776 | int index[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 777 | } ccv_cnnp_model_permute_t; | |||
| 778 | ||||
| 779 | static void _ccv_cnnp_permute_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 780 | { | |||
| 781 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 781, __extension__ __PRETTY_FUNCTION__); })); | |||
| 782 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 782, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 783 | ccv_cnnp_model_permute_t* const self = (ccv_cnnp_model_permute_t*)super; | |||
| 784 | PRINT(CCV_CLI_VERBOSE, "[cnnp_permute_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_permute_build] -\n"); fflush(stdout); } } while (0); | |||
| 785 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 786 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 787 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 788 | int input_dim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 789 | memcpy(input_dim, params.dim, sizeof(params.dim)); | |||
| 790 | int input_stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 791 | int output_stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 792 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If it is not an alias. Find stride and permute. | |||
| 793 | { | |||
| 794 | ccv_nnc_tensor_get_stride(input_dim, input_stride); | |||
| 795 | int i; | |||
| 796 | for (i = 0; i < nd; i++) | |||
| 797 | { | |||
| 798 | const int idx = self->index[i]; | |||
| 799 | assert(idx >= 0 && idx < nd)((void) sizeof ((idx >= 0 && idx < nd) ? 1 : 0) , __extension__ ({ if (idx >= 0 && idx < nd) ; else __assert_fail ("idx >= 0 && idx < nd", "ccv_cnnp_model_addons.c" , 799, __extension__ __PRETTY_FUNCTION__); })); | |||
| 800 | params.dim[i] = input_dim[idx]; | |||
| 801 | output_stride[i] = input_stride[idx]; | |||
| 802 | } | |||
| 803 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ccv_nnc_no_ofs, output_stride, params, 0); | |||
| 804 | } else { | |||
| 805 | // if it is an alias, we can get the stride from it and use that. | |||
| 806 | int input_ofs[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 807 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], input_ofs, input_stride); | |||
| 808 | assert(input_stride[0] != 0)((void) sizeof ((input_stride[0] != 0) ? 1 : 0), __extension__ ({ if (input_stride[0] != 0) ; else __assert_fail ("input_stride[0] != 0" , "ccv_cnnp_model_addons.c", 808, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 809 | int output_ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 810 | int i; | |||
| 811 | for (i = 0; i < nd; i++) | |||
| 812 | { | |||
| 813 | const int idx = self->index[i]; | |||
| 814 | assert(idx >= 0 && idx < nd)((void) sizeof ((idx >= 0 && idx < nd) ? 1 : 0) , __extension__ ({ if (idx >= 0 && idx < nd) ; else __assert_fail ("idx >= 0 && idx < nd", "ccv_cnnp_model_addons.c" , 814, __extension__ __PRETTY_FUNCTION__); })); | |||
| 815 | params.dim[i] = input_dim[idx]; | |||
| 816 | output_stride[i] = input_stride[idx]; | |||
| 817 | output_ofs[i] = input_ofs[idx]; | |||
| 818 | } | |||
| 819 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], output_ofs, output_stride, params, 0); | |||
| 820 | } | |||
| 821 | } | |||
| 822 | ||||
| 823 | static ccv_cnnp_model_t* _ccv_cnnp_permute_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 824 | ||||
| 825 | static const ccv_cnnp_model_vtab_t ccv_cnnp_permute_isa = { | |||
| 826 | .build = _ccv_cnnp_permute_build, | |||
| 827 | .copy = _ccv_cnnp_permute_copy, | |||
| 828 | }; | |||
| 829 | ||||
| 830 | ccv_cnnp_model_t* ccv_cnnp_permute(const int index[CCV_NNC_MAX_DIM_ALLOC(12)], const char* const name) | |||
| 831 | { | |||
| 832 | ccv_cnnp_model_permute_t* const model_permute = (ccv_cnnp_model_permute_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_permute_t)); | |||
| 833 | model_permute->super.isa = &ccv_cnnp_permute_isa; | |||
| 834 | model_permute->super.input_size = 1; | |||
| 835 | model_permute->super.outputs = &model_permute->output; | |||
| 836 | model_permute->super.output_size = 1; | |||
| 837 | ccv_cnnp_model_copy_name(&model_permute->super, name); | |||
| 838 | memcpy(model_permute->index, index, sizeof(model_permute->index)); | |||
| 839 | return (ccv_cnnp_model_t*)model_permute; | |||
| 840 | } | |||
| 841 | ||||
| 842 | static ccv_cnnp_model_t* _ccv_cnnp_permute_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 843 | { | |||
| 844 | const ccv_cnnp_model_permute_t* const self = (const ccv_cnnp_model_permute_t*)super; | |||
| 845 | return ccv_cnnp_permute(self->index, self->super.name); | |||
| 846 | } | |||
| 847 | ||||
| 848 | typedef struct { | |||
| 849 | ccv_cnnp_model_t super; | |||
| 850 | int index; | |||
| 851 | ccv_nnc_tensor_symbol_t output; | |||
| 852 | } ccv_cnnp_model_extract_t; | |||
| 853 | ||||
| 854 | static void _ccv_cnnp_extract_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 855 | { | |||
| 856 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 856, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 857 | ccv_cnnp_model_extract_t* const self = (ccv_cnnp_model_extract_t*)super; | |||
| 858 | PRINT(CCV_CLI_VERBOSE, "[cnnp_extract_build] index: %d\n", self->index)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_extract_build] index: %d\n", self->index) ; fflush(stdout); } } while (0); | |||
| 859 | outputs[0] = inputs[self->index]; | |||
| 860 | } | |||
| 861 | ||||
| 862 | static ccv_cnnp_model_t* _ccv_cnnp_extract_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 863 | ||||
| 864 | static const ccv_cnnp_model_vtab_t ccv_cnnp_extract_isa = { | |||
| 865 | .build = _ccv_cnnp_extract_build, | |||
| 866 | .copy = _ccv_cnnp_extract_copy, | |||
| 867 | }; | |||
| 868 | ||||
| 869 | ccv_cnnp_model_t* ccv_cnnp_extract(const int index, const char* const name) | |||
| 870 | { | |||
| 871 | ccv_cnnp_model_extract_t* const model_extract = (ccv_cnnp_model_extract_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_extract_t)); | |||
| 872 | model_extract->index = index; | |||
| 873 | model_extract->super.isa = &ccv_cnnp_extract_isa; | |||
| 874 | model_extract->super.input_size = 0; | |||
| 875 | model_extract->super.outputs = &model_extract->output; | |||
| 876 | model_extract->super.output_size = 1; | |||
| 877 | ccv_cnnp_model_copy_name(&model_extract->super, name); | |||
| 878 | return (ccv_cnnp_model_t*)model_extract; | |||
| 879 | } | |||
| 880 | ||||
| 881 | static ccv_cnnp_model_t* _ccv_cnnp_extract_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 882 | { | |||
| 883 | ccv_cnnp_model_extract_t* const self = (ccv_cnnp_model_extract_t*)super; | |||
| 884 | return ccv_cnnp_extract(self->index, self->super.name); | |||
| 885 | } | |||
| 886 | ||||
| 887 | typedef struct { | |||
| 888 | ccv_cnnp_model_t super; | |||
| 889 | ccv_nnc_tensor_symbol_t output; | |||
| 890 | } ccv_cnnp_model_flatten_t; | |||
| 891 | ||||
| 892 | static void _ccv_cnnp_flatten_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 893 | { | |||
| 894 | PRINT(CCV_CLI_VERBOSE, "[cnnp_flatten_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_flatten_build] -\n"); fflush(stdout); } } while (0); | |||
| 895 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 895, __extension__ __PRETTY_FUNCTION__); })); | |||
| 896 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 896, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 897 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 898 | ccv_nnc_tensor_param_t output_params = params; | |||
| 899 | memset(output_params.dim, 0, sizeof(output_params.dim)); | |||
| 900 | output_params.dim[0] = ccv_nnc_tensor_get_n(params); | |||
| 901 | assert(output_params.dim[0] > 0)((void) sizeof ((output_params.dim[0] > 0) ? 1 : 0), __extension__ ({ if (output_params.dim[0] > 0) ; else __assert_fail ("output_params.dim[0] > 0" , "ccv_cnnp_model_addons.c", 901, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 902 | output_params.dim[1] = ccv_nnc_tensor_count(params) / output_params.dim[0]; | |||
| 903 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 904 | ccv_nnc_tensor_get_stride(output_params.dim, stride); | |||
| 905 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], DIM_ALLOC()(int [(12)]){}, stride, output_params, 0); | |||
| 906 | } | |||
| 907 | ||||
| 908 | static ccv_cnnp_model_t* _ccv_cnnp_flatten_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 909 | ||||
| 910 | static const ccv_cnnp_model_vtab_t ccv_cnnp_flatten_isa = { | |||
| 911 | .build = _ccv_cnnp_flatten_build, | |||
| 912 | .copy = _ccv_cnnp_flatten_copy, | |||
| 913 | }; | |||
| 914 | ||||
| 915 | ccv_cnnp_model_t* ccv_cnnp_flatten(const char* const name) | |||
| 916 | { | |||
| 917 | ccv_cnnp_model_flatten_t* const model_flatten = (ccv_cnnp_model_flatten_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_flatten_t)); | |||
| 918 | model_flatten->super.isa = &ccv_cnnp_flatten_isa; | |||
| 919 | model_flatten->super.input_size = 1; | |||
| 920 | model_flatten->super.outputs = &model_flatten->output; | |||
| 921 | model_flatten->super.output_size = 1; | |||
| 922 | ccv_cnnp_model_copy_name(&model_flatten->super, name); | |||
| 923 | return (ccv_cnnp_model_t*)model_flatten; | |||
| 924 | } | |||
| 925 | ||||
| 926 | static ccv_cnnp_model_t* _ccv_cnnp_flatten_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 927 | { | |||
| 928 | return ccv_cnnp_flatten(self->name); | |||
| 929 | } | |||
| 930 | ||||
| 931 | // MARK - Batch Norm Layer | |||
| 932 | ||||
| 933 | typedef struct { | |||
| 934 | ccv_cnnp_model_t super; | |||
| 935 | ccv_nnc_tensor_symbol_t output; | |||
| 936 | ccv_nnc_tensor_symbol_t bias; | |||
| 937 | ccv_nnc_tensor_symbol_t scale; | |||
| 938 | ccv_nnc_graph_exec_symbol_t batch_norm; | |||
| 939 | ccv_nnc_cmd_param_t params; | |||
| 940 | ccv_array_t* zero_inits; | |||
| 941 | ccv_array_t* retainables; | |||
| 942 | } ccv_cnnp_model_batch_norm_t; | |||
| 943 | ||||
| 944 | static void _ccv_cnnp_batch_norm_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 945 | { | |||
| 946 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 946, __extension__ __PRETTY_FUNCTION__); })); | |||
| 947 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 947, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 948 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 949 | PRINT(CCV_CLI_VERBOSE, "[cnnp_batch_norm_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_batch_norm_build] -\n"); fflush(stdout); } } while (0); | |||
| 950 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 951 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 952 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 953 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 954 | // If the accuracy is not enough, bump it to 32-bit floating point. | |||
| 955 | if (bias_params.datatype != CCV_32F && bias_params.datatype != CCV_64F) | |||
| 956 | bias_params.datatype = CCV_32F; | |||
| 957 | bias_params.dim[0] = nd > 1 ? ccv_nnc_tensor_get_c(params) : params.dim[0]; | |||
| 958 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 959 | // Both scale and bias are shared between if this model is reused. | |||
| 960 | if (!self->scale.graph) | |||
| 961 | self->scale = ccv_nnc_tensor_symbol_new(graph, bias_params, "scale"); | |||
| 962 | if (!self->bias.graph) | |||
| 963 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 964 | const ccv_nnc_tensor_symbol_t mean = ccv_nnc_tensor_symbol_new(graph, bias_params, "mean"); | |||
| 965 | const ccv_nnc_tensor_symbol_t var = ccv_nnc_tensor_symbol_new(graph, bias_params, "var"); | |||
| 966 | // Otherwise, notice mean, var, saved_mean, saved_inv_std are not reused. | |||
| 967 | if (!self->zero_inits) | |||
| 968 | self->zero_inits = ccv_array_new(sizeof(ccv_nnc_tensor_symbol_t), 0, 0); | |||
| 969 | ccv_array_push(self->zero_inits, &mean); | |||
| 970 | ccv_array_push(self->zero_inits, &var); | |||
| 971 | const ccv_nnc_tensor_symbol_t out_mean = ccv_nnc_tensor_symbol_new(graph, bias_params, "out_mean"); | |||
| 972 | const ccv_nnc_tensor_symbol_t out_var = ccv_nnc_tensor_symbol_new(graph, bias_params, "out_var"); | |||
| 973 | if (!self->retainables) | |||
| 974 | self->retainables = ccv_array_new(sizeof(ccv_nnc_tensor_symbol_t), 0, 0); | |||
| 975 | ccv_array_push(self->retainables, &out_mean); | |||
| 976 | ccv_array_push(self->retainables, &out_var); | |||
| 977 | const ccv_nnc_tensor_symbol_t saved_mean = ccv_nnc_tensor_symbol_new(graph, bias_params, "saved_mean"); | |||
| 978 | const ccv_nnc_tensor_symbol_t saved_inv_std = ccv_nnc_tensor_symbol_new(graph, bias_params, "saved_inv_std"); | |||
| 979 | const int hw = ccv_nnc_tensor_hw(params, ccv_nnc_tensor_nd(params.dim)); | |||
| 980 | ccv_nnc_cmd_param_t batch_norm = self->params; | |||
| 981 | batch_norm.bnorm.count = hw >= 0 ? CCV_NNC_MAX_DIM(2) + 1 : 1; | |||
| 982 | int i; | |||
| 983 | batch_norm.bnorm.axis[0] = (params.format == CCV_TENSOR_FORMAT_CHWN) ? 3 : 0; | |||
| 984 | if (hw >= 0) | |||
| 985 | for (i = 0; i < CCV_NNC_MAX_DIM(2); i++) | |||
| 986 | batch_norm.bnorm.axis[i + 1] = i + hw; | |||
| 987 | self->params = batch_norm; | |||
| 988 | self->batch_norm = ccv_nnc_graph_exec_symbol_new(graph, ccv_nnc_cmd(CCV_NNC_BATCH_NORM_FORWARD, 0, batch_norm, 0), TENSOR_SYMBOL_LIST(inputs[0], self->scale, self->bias, mean, var)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->scale, self->bias, mean, var}, (1 +1 +1 +1 +1 +1 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output, out_mean, out_var, saved_mean, saved_inv_std)(const ccv_nnc_tensor_symbol_t []){output, out_mean, out_var, saved_mean, saved_inv_std}, (1 +1 +1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "batch_norm"); | |||
| 989 | outputs[0] = output; | |||
| 990 | } | |||
| 991 | ||||
| 992 | static void _ccv_cnnp_batch_norm_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 993 | { | |||
| 994 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 995 | if (self->scale.graph) | |||
| 996 | initializer(context, CMD_RANDOM_UNIFORM_FORWARD(0, 1)ccv_nnc_cmd(CCV_NNC_RANDOM_UNIFORM_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={0, 1}}}, 0), ccv_nnc_no_hint, 0, 0, self->scale); | |||
| 997 | if (self->bias.graph) | |||
| 998 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, self->bias); | |||
| 999 | int i; | |||
| 1000 | if (self->zero_inits) | |||
| 1001 | for (i = 0; i < self->zero_inits->rnum; i++) | |||
| 1002 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, *(ccv_nnc_tensor_symbol_t*)ccv_array_get(self->zero_inits, i)((void*)(((char*)((self->zero_inits)->data)) + (size_t) (self->zero_inits)->rsize * (size_t)(i)))); | |||
| 1003 | } | |||
| 1004 | ||||
| 1005 | static void _ccv_cnnp_batch_norm_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 1006 | { | |||
| 1007 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1008 | if (self->scale.graph) | |||
| 1009 | add_to_array(parameters, self->scale, is_trainable); | |||
| 1010 | if (self->bias.graph) | |||
| 1011 | add_to_array(parameters, self->bias, is_trainable); | |||
| 1012 | } | |||
| 1013 | ||||
| 1014 | static void _ccv_cnnp_batch_norm_add_to_output(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const outputs) | |||
| 1015 | { | |||
| 1016 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1017 | int i; | |||
| 1018 | if (self->retainables) | |||
| 1019 | for (i = 0; i < self->retainables->rnum; i++) | |||
| 1020 | { | |||
| 1021 | const ccv_nnc_tensor_symbol_t symbol = *(ccv_nnc_tensor_symbol_t*)ccv_array_get(self->retainables, i)((void*)(((char*)((self->retainables)->data)) + (size_t )(self->retainables)->rsize * (size_t)(i))); | |||
| 1022 | add_to_array(outputs, symbol, 0); | |||
| 1023 | } | |||
| 1024 | } | |||
| 1025 | ||||
| 1026 | static void _ccv_cnnp_batch_norm_set_is_test(ccv_cnnp_model_t* const super, const int is_test, const ccv_cnnp_cmd_updater_f updater, void* const context) | |||
| 1027 | { | |||
| 1028 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1029 | if (self->batch_norm.graph) | |||
| 1030 | { | |||
| 1031 | self->params.bnorm.is_test = is_test; | |||
| 1032 | updater(context, self->batch_norm, ccv_nnc_cmd(CCV_NNC_BATCH_NORM_FORWARD, 0, self->params, 0), ccv_nnc_no_hint); | |||
| 1033 | } | |||
| 1034 | } | |||
| 1035 | ||||
| 1036 | static void _ccv_cnnp_batch_norm_deinit(ccv_cnnp_model_t* const super) | |||
| 1037 | { | |||
| 1038 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1039 | if (self->zero_inits) | |||
| 1040 | ccv_array_free(self->zero_inits); | |||
| 1041 | if (self->retainables) | |||
| 1042 | ccv_array_free(self->retainables); | |||
| 1043 | } | |||
| 1044 | ||||
| 1045 | static ccv_cnnp_model_t* _ccv_cnnp_batch_norm_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1046 | ||||
| 1047 | static const ccv_cnnp_model_vtab_t ccv_cnnp_batch_norm_isa = { | |||
| 1048 | .build = _ccv_cnnp_batch_norm_build, | |||
| 1049 | .init_states = _ccv_cnnp_batch_norm_init_states, | |||
| 1050 | .add_to_parameter = _ccv_cnnp_batch_norm_add_to_parameter, | |||
| 1051 | .add_to_output = _ccv_cnnp_batch_norm_add_to_output, | |||
| 1052 | .copy = _ccv_cnnp_batch_norm_copy, | |||
| 1053 | .set_is_test = _ccv_cnnp_batch_norm_set_is_test, | |||
| 1054 | .deinit = _ccv_cnnp_batch_norm_deinit, | |||
| 1055 | }; | |||
| 1056 | ||||
| 1057 | ccv_cnnp_model_t* ccv_cnnp_batch_norm(const float momentum, const float epsilon, const int is_trainable, const char* const name) | |||
| 1058 | { | |||
| 1059 | ccv_cnnp_model_batch_norm_t* const model_batch_norm = (ccv_cnnp_model_batch_norm_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_batch_norm_t)); | |||
| 1060 | model_batch_norm->super.isa = &ccv_cnnp_batch_norm_isa; | |||
| 1061 | model_batch_norm->super.input_size = 1; | |||
| 1062 | model_batch_norm->super.outputs = &model_batch_norm->output; | |||
| 1063 | model_batch_norm->super.output_size = 1; | |||
| 1064 | model_batch_norm->super.is_trainable = is_trainable; | |||
| 1065 | ccv_cnnp_model_copy_name(&model_batch_norm->super, name); | |||
| 1066 | model_batch_norm->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1067 | model_batch_norm->scale.graph = 0; | |||
| 1068 | model_batch_norm->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1069 | model_batch_norm->bias.graph = 0; | |||
| 1070 | model_batch_norm->params.bnorm.momentum = momentum; | |||
| 1071 | model_batch_norm->params.bnorm.epsilon = epsilon; | |||
| 1072 | return (ccv_cnnp_model_t*)model_batch_norm; | |||
| 1073 | } | |||
| 1074 | ||||
| 1075 | static ccv_cnnp_model_t* _ccv_cnnp_batch_norm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1076 | { | |||
| 1077 | const ccv_cnnp_model_batch_norm_t* const self = (const ccv_cnnp_model_batch_norm_t*)super; | |||
| 1078 | return ccv_cnnp_batch_norm(self->params.bnorm.momentum, self->params.bnorm.epsilon, self->super.is_trainable, self->super.name); | |||
| 1079 | } | |||
| 1080 | ||||
| 1081 | // MARK - Convolution Layer | |||
| 1082 | ||||
| 1083 | typedef struct { | |||
| 1084 | ccv_cnnp_model_t super; | |||
| 1085 | ccv_nnc_tensor_symbol_t output; | |||
| 1086 | ccv_nnc_tensor_symbol_t weights; | |||
| 1087 | ccv_nnc_tensor_symbol_t bias; | |||
| 1088 | int groups; | |||
| 1089 | int filters; | |||
| 1090 | int kdim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1091 | int dilation[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1092 | int no_bias; | |||
| 1093 | int format; | |||
| 1094 | ccv_nnc_hint_t hint; | |||
| 1095 | } ccv_cnnp_model_convolution_t; | |||
| 1096 | ||||
| 1097 | static void _ccv_cnnp_convolution_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1098 | { | |||
| 1099 | ccv_cnnp_model_convolution_t* const self = (ccv_cnnp_model_convolution_t*)super; | |||
| 1100 | PRINT(CCV_CLI_VERBOSE, "[cnnp_convolution_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_convolution_build] -\n"); fflush(stdout); } } while (0); | |||
| 1101 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1101, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1102 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1102, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1103 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1104 | int i; | |||
| 1105 | const int k_nd = ccv_nnc_tensor_nd(self->kdim); | |||
| 1106 | const int nd = k_nd + 2; | |||
| 1107 | ccv_nnc_tensor_param_t weights_params = params; | |||
| 1108 | if (self->format) | |||
| 1109 | weights_params.format = self->format; | |||
| 1110 | ccv_nnc_tensor_set_n(&weights_params, self->filters); | |||
| 1111 | const int c = ccv_nnc_tensor_get_c(params); | |||
| 1112 | assert(c % self->groups == 0)((void) sizeof ((c % self->groups == 0) ? 1 : 0), __extension__ ({ if (c % self->groups == 0) ; else __assert_fail ("c % self->groups == 0" , "ccv_cnnp_model_addons.c", 1112, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1113 | ccv_nnc_tensor_set_c(&weights_params, nd, c / self->groups); | |||
| 1114 | const int hw = ccv_nnc_tensor_hw(weights_params, nd); | |||
| 1115 | assert(hw >= 0)((void) sizeof ((hw >= 0) ? 1 : 0), __extension__ ({ if (hw >= 0) ; else __assert_fail ("hw >= 0", "ccv_cnnp_model_addons.c" , 1115, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1116 | for (i = 0; i < k_nd; i++) | |||
| 1117 | weights_params.dim[i + hw] = self->kdim[i]; | |||
| 1118 | if (!self->weights.graph) | |||
| 1119 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 1120 | assert(self->weights.graph == graph)((void) sizeof ((self->weights.graph == graph) ? 1 : 0), __extension__ ({ if (self->weights.graph == graph) ; else __assert_fail ("self->weights.graph == graph", "ccv_cnnp_model_addons.c" , 1120, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1121 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 1122 | if (self->format) | |||
| 1123 | bias_params.format = self->format; | |||
| 1124 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 1125 | bias_params.dim[0] = self->filters; | |||
| 1126 | ccv_nnc_cmd_t cmd = CMD_CONVOLUTION_FORWARD(self->groups, self->filters)ccv_nnc_cmd(CCV_NNC_CONVOLUTION_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={}},.convolution={.count=self->filters,.groups =self->groups}}), 0); | |||
| 1127 | for (i = 0; i < k_nd; i++) | |||
| 1128 | cmd.info.size.dim[i] = self->kdim[i]; | |||
| 1129 | cmd.info.size.dim[k_nd] = c; | |||
| 1130 | memcpy(cmd.info.convolution.dilation, self->dilation, sizeof(self->dilation)); | |||
| 1131 | ccv_nnc_tensor_param_t output_params; | |||
| 1132 | // Dilate weight size based on the dilation factor. | |||
| 1133 | for (i = 0; i < k_nd; i++) | |||
| 1134 | weights_params.dim[i + hw] = (self->kdim[i] - 1) * ccv_max(self->dilation[i], 1)({ typeof (self->dilation[i]) _a = (self->dilation[i]); typeof (1) _b = (1); (_a > _b) ? _a : _b; }) + 1; | |||
| 1135 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 1136 | params, | |||
| 1137 | weights_params, | |||
| 1138 | bias_params, | |||
| 1139 | }, 3, self->hint, &output_params, 1); | |||
| 1140 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1141 | ccv_nnc_graph_exec_symbol_t convolution; | |||
| 1142 | if (self->no_bias) | |||
| 1143 | convolution = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], self->weights)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->weights }, (1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "convolution"); | |||
| 1144 | else { | |||
| 1145 | if (!self->bias.graph) | |||
| 1146 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 1147 | convolution = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], self->weights, self->bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->weights , self->bias}, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "convolution"); | |||
| 1148 | } | |||
| 1149 | ccv_nnc_graph_exec_symbol_set_hint(graph, convolution, self->hint); | |||
| 1150 | outputs[0] = output; | |||
| 1151 | } | |||
| 1152 | ||||
| 1153 | static void _ccv_cnnp_convolution_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 1154 | { | |||
| 1155 | ccv_cnnp_model_convolution_t* const self = (ccv_cnnp_model_convolution_t*)super; | |||
| 1156 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 1157 | const int n = ccv_max(ccv_nnc_tensor_get_n(weight_params), 1)({ typeof (ccv_nnc_tensor_get_n(weight_params)) _a = (ccv_nnc_tensor_get_n (weight_params)); typeof (1) _b = (1); (_a > _b) ? _a : _b ; }); | |||
| 1158 | const int count = ccv_nnc_tensor_count(weight_params); | |||
| 1159 | const float std = sqrtf(2) / sqrtf(count / n); | |||
| 1160 | const float bound = sqrtf(3) * std; | |||
| 1161 | initializer(context, CMD_RANDOM_UNIFORM_FORWARD(-bound, bound)ccv_nnc_cmd(CCV_NNC_RANDOM_UNIFORM_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={-bound, bound}}}, 0), ccv_nnc_no_hint, 0, 0, self->weights); | |||
| 1162 | if (self->bias.graph) | |||
| 1163 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, self->bias); | |||
| 1164 | } | |||
| 1165 | ||||
| 1166 | static void _ccv_cnnp_convolution_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 1167 | { | |||
| 1168 | ccv_cnnp_model_convolution_t* const self = (ccv_cnnp_model_convolution_t*)super; | |||
| 1169 | add_to_array(parameters, self->weights, is_trainable); | |||
| 1170 | if (self->bias.graph) | |||
| 1171 | add_to_array(parameters, self->bias, is_trainable); | |||
| 1172 | } | |||
| 1173 | ||||
| 1174 | static ccv_cnnp_model_t* _ccv_cnnp_convolution_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1175 | ||||
| 1176 | static const ccv_cnnp_model_vtab_t ccv_cnnp_convolution_isa = { | |||
| 1177 | .build = _ccv_cnnp_convolution_build, | |||
| 1178 | .init_states = _ccv_cnnp_convolution_init_states, | |||
| 1179 | .add_to_parameter = _ccv_cnnp_convolution_add_to_parameter, | |||
| 1180 | .copy = _ccv_cnnp_convolution_copy, | |||
| 1181 | }; | |||
| 1182 | ||||
| 1183 | ccv_cnnp_model_t* ccv_cnnp_convolution(const int groups, const int filters, const int kdim[CCV_NNC_MAX_DIM_ALLOC(12)], const int dilation[CCV_NNC_MAX_DIM_ALLOC(12)], const int no_bias, ccv_nnc_hint_t hint, const int format, const int is_trainable, const char* const name) | |||
| 1184 | { | |||
| 1185 | ccv_cnnp_model_convolution_t* const model_convolution = (ccv_cnnp_model_convolution_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_convolution_t)); | |||
| 1186 | model_convolution->super.isa = &ccv_cnnp_convolution_isa; | |||
| 1187 | model_convolution->super.input_size = 1; | |||
| 1188 | model_convolution->super.outputs = &model_convolution->output; | |||
| 1189 | model_convolution->super.output_size = 1; | |||
| 1190 | model_convolution->super.is_trainable = is_trainable; | |||
| 1191 | ccv_cnnp_model_copy_name(&model_convolution->super, name); | |||
| 1192 | model_convolution->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1193 | model_convolution->weights.graph = 0; | |||
| 1194 | model_convolution->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1195 | model_convolution->bias.graph = 0; | |||
| 1196 | model_convolution->groups = groups; | |||
| 1197 | model_convolution->filters = filters; | |||
| 1198 | memcpy(model_convolution->kdim, kdim, sizeof(model_convolution->kdim)); | |||
| 1199 | memcpy(model_convolution->dilation, dilation, sizeof(model_convolution->dilation)); | |||
| 1200 | model_convolution->no_bias = no_bias; | |||
| 1201 | model_convolution->hint = hint; | |||
| 1202 | model_convolution->format = format; | |||
| 1203 | return (ccv_cnnp_model_t*)model_convolution; | |||
| 1204 | } | |||
| 1205 | ||||
| 1206 | static ccv_cnnp_model_t* _ccv_cnnp_convolution_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1207 | { | |||
| 1208 | ccv_cnnp_model_convolution_t* const self = (ccv_cnnp_model_convolution_t*)super; | |||
| 1209 | return ccv_cnnp_convolution(self->groups, self->filters, self->kdim, self->dilation, self->no_bias, self->hint, self->format, self->super.is_trainable, self->super.name); | |||
| 1210 | } | |||
| 1211 | ||||
| 1212 | // MARK - Convolution Transpose Layer | |||
| 1213 | ||||
| 1214 | typedef struct { | |||
| 1215 | ccv_cnnp_model_t super; | |||
| 1216 | ccv_nnc_tensor_symbol_t output; | |||
| 1217 | ccv_nnc_tensor_symbol_t weights; | |||
| 1218 | ccv_nnc_tensor_symbol_t bias; | |||
| 1219 | int groups; | |||
| 1220 | int filters; | |||
| 1221 | int kdim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1222 | int dilation[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1223 | int output_padding; | |||
| 1224 | int no_bias; | |||
| 1225 | int format; | |||
| 1226 | ccv_nnc_hint_t hint; | |||
| 1227 | } ccv_cnnp_model_convolution_transpose_t; | |||
| 1228 | ||||
| 1229 | static void _ccv_cnnp_convolution_transpose_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1230 | { | |||
| 1231 | ccv_cnnp_model_convolution_transpose_t* const self = (ccv_cnnp_model_convolution_transpose_t*)super; | |||
| 1232 | PRINT(CCV_CLI_VERBOSE, "[cnnp_convolution_transpose_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_convolution_transpose_build] -\n"); fflush(stdout ); } } while (0); | |||
| 1233 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1233, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1234 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1234, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1235 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1236 | int i; | |||
| 1237 | const int nd = CCV_NNC_MAX_DIM(2) + 2; | |||
| 1238 | ccv_nnc_tensor_param_t weights_params = params; | |||
| 1239 | if (self->format) | |||
| 1240 | weights_params.format = self->format; | |||
| 1241 | const int c = ccv_nnc_tensor_get_c(params); | |||
| 1242 | ccv_nnc_tensor_set_n(&weights_params, c); | |||
| 1243 | assert(c % self->groups == 0)((void) sizeof ((c % self->groups == 0) ? 1 : 0), __extension__ ({ if (c % self->groups == 0) ; else __assert_fail ("c % self->groups == 0" , "ccv_cnnp_model_addons.c", 1243, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1244 | ccv_nnc_tensor_set_c(&weights_params, nd, self->filters / self->groups); | |||
| 1245 | const int hw = ccv_nnc_tensor_hw(weights_params, nd); | |||
| 1246 | assert(hw >= 0)((void) sizeof ((hw >= 0) ? 1 : 0), __extension__ ({ if (hw >= 0) ; else __assert_fail ("hw >= 0", "ccv_cnnp_model_addons.c" , 1246, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1247 | for (i = 0; i < CCV_NNC_MAX_DIM(2); i++) | |||
| 1248 | weights_params.dim[i + hw] = self->kdim[i]; | |||
| 1249 | if (!self->weights.graph) | |||
| 1250 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 1251 | assert(self->weights.graph == graph)((void) sizeof ((self->weights.graph == graph) ? 1 : 0), __extension__ ({ if (self->weights.graph == graph) ; else __assert_fail ("self->weights.graph == graph", "ccv_cnnp_model_addons.c" , 1251, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1252 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 1253 | if (self->format) | |||
| 1254 | bias_params.format = self->format; | |||
| 1255 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 1256 | bias_params.dim[0] = self->filters; | |||
| 1257 | ccv_nnc_cmd_t cmd = CMD_CONVOLUTION_TRANSPOSE_FORWARD(self->groups, self->filters, self->output_padding)ccv_nnc_cmd(CCV_NNC_CONVOLUTION_TRANSPOSE_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={}},.convolution_transpose={.count=self->filters ,.groups=self->groups,.output_padding=self->output_padding }}), 0); | |||
| 1258 | for (i = 0; i < CCV_NNC_MAX_DIM(2); i++) | |||
| 1259 | cmd.info.size.dim[i] = self->kdim[i]; | |||
| 1260 | cmd.info.size.dim[CCV_NNC_MAX_DIM(2)] = c; | |||
| 1261 | memcpy(cmd.info.convolution_transpose.dilation, self->dilation, sizeof(self->dilation)); | |||
| 1262 | ccv_nnc_tensor_param_t output_params; | |||
| 1263 | // Dilate weight size based on the dilation factor. | |||
| 1264 | for (i = 0; i < CCV_NNC_MAX_DIM(2); i++) | |||
| 1265 | weights_params.dim[i + hw] = (self->kdim[i] - 1) * ccv_max(self->dilation[i], 1)({ typeof (self->dilation[i]) _a = (self->dilation[i]); typeof (1) _b = (1); (_a > _b) ? _a : _b; }) + 1; | |||
| 1266 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 1267 | params, | |||
| 1268 | weights_params, | |||
| 1269 | bias_params, | |||
| 1270 | }, 3, self->hint, &output_params, 1); | |||
| 1271 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1272 | ccv_nnc_graph_exec_symbol_t convolution_transpose; | |||
| 1273 | if (self->no_bias) | |||
| 1274 | convolution_transpose = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], self->weights)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->weights }, (1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "convolution_transpose"); | |||
| 1275 | else { | |||
| 1276 | if (!self->bias.graph) | |||
| 1277 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 1278 | convolution_transpose = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], self->weights, self->bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->weights , self->bias}, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "convolution_transpose"); | |||
| 1279 | } | |||
| 1280 | ccv_nnc_graph_exec_symbol_set_hint(graph, convolution_transpose, self->hint); | |||
| 1281 | outputs[0] = output; | |||
| 1282 | } | |||
| 1283 | ||||
| 1284 | static void _ccv_cnnp_convolution_transpose_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 1285 | { | |||
| 1286 | ccv_cnnp_model_convolution_transpose_t* const self = (ccv_cnnp_model_convolution_transpose_t*)super; | |||
| 1287 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 1288 | const int n = ccv_max(ccv_nnc_tensor_get_n(weight_params), 1)({ typeof (ccv_nnc_tensor_get_n(weight_params)) _a = (ccv_nnc_tensor_get_n (weight_params)); typeof (1) _b = (1); (_a > _b) ? _a : _b ; }); | |||
| 1289 | const int count = ccv_nnc_tensor_count(weight_params); | |||
| 1290 | const float std = sqrtf(2) / sqrtf(count / n); | |||
| 1291 | const float bound = sqrtf(3) * std; | |||
| 1292 | initializer(context, CMD_RANDOM_UNIFORM_FORWARD(-bound, bound)ccv_nnc_cmd(CCV_NNC_RANDOM_UNIFORM_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={-bound, bound}}}, 0), ccv_nnc_no_hint, 0, 0, self->weights); | |||
| 1293 | if (self->bias.graph) | |||
| 1294 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, self->bias); | |||
| 1295 | } | |||
| 1296 | ||||
| 1297 | static void _ccv_cnnp_convolution_transpose_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 1298 | { | |||
| 1299 | ccv_cnnp_model_convolution_transpose_t* const self = (ccv_cnnp_model_convolution_transpose_t*)super; | |||
| 1300 | add_to_array(parameters, self->weights, is_trainable); | |||
| 1301 | if (self->bias.graph) | |||
| 1302 | add_to_array(parameters, self->bias, is_trainable); | |||
| 1303 | } | |||
| 1304 | ||||
| 1305 | static ccv_cnnp_model_t* _ccv_cnnp_convolution_transpose_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1306 | ||||
| 1307 | static const ccv_cnnp_model_vtab_t ccv_cnnp_convolution_transpose_isa = { | |||
| 1308 | .build = _ccv_cnnp_convolution_transpose_build, | |||
| 1309 | .init_states = _ccv_cnnp_convolution_transpose_init_states, | |||
| 1310 | .add_to_parameter = _ccv_cnnp_convolution_transpose_add_to_parameter, | |||
| 1311 | .copy = _ccv_cnnp_convolution_transpose_copy, | |||
| 1312 | }; | |||
| 1313 | ||||
| 1314 | ccv_cnnp_model_t* ccv_cnnp_convolution_transpose(const int groups, const int filters, const int kdim[CCV_NNC_MAX_DIM_ALLOC(12)], const int dilation[CCV_NNC_MAX_DIM_ALLOC(12)], const int output_padding, const int no_bias, ccv_nnc_hint_t hint, const int format, const int is_trainable, const char* const name) | |||
| 1315 | { | |||
| 1316 | ccv_cnnp_model_convolution_transpose_t* const model_convolution_transpose = (ccv_cnnp_model_convolution_transpose_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_convolution_transpose_t)); | |||
| 1317 | model_convolution_transpose->super.isa = &ccv_cnnp_convolution_transpose_isa; | |||
| 1318 | model_convolution_transpose->super.input_size = 1; | |||
| 1319 | model_convolution_transpose->super.outputs = &model_convolution_transpose->output; | |||
| 1320 | model_convolution_transpose->super.output_size = 1; | |||
| 1321 | model_convolution_transpose->super.is_trainable = is_trainable; | |||
| 1322 | ccv_cnnp_model_copy_name(&model_convolution_transpose->super, name); | |||
| 1323 | model_convolution_transpose->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1324 | model_convolution_transpose->weights.graph = 0; | |||
| 1325 | model_convolution_transpose->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1326 | model_convolution_transpose->bias.graph = 0; | |||
| 1327 | model_convolution_transpose->groups = groups; | |||
| 1328 | model_convolution_transpose->filters = filters; | |||
| 1329 | memcpy(model_convolution_transpose->kdim, kdim, sizeof(model_convolution_transpose->kdim)); | |||
| 1330 | memcpy(model_convolution_transpose->dilation, dilation, sizeof(model_convolution_transpose->dilation)); | |||
| 1331 | model_convolution_transpose->output_padding = output_padding; | |||
| 1332 | model_convolution_transpose->no_bias = no_bias; | |||
| 1333 | model_convolution_transpose->hint = hint; | |||
| 1334 | model_convolution_transpose->format = format; | |||
| 1335 | return (ccv_cnnp_model_t*)model_convolution_transpose; | |||
| 1336 | } | |||
| 1337 | ||||
| 1338 | static ccv_cnnp_model_t* _ccv_cnnp_convolution_transpose_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1339 | { | |||
| 1340 | ccv_cnnp_model_convolution_transpose_t* const self = (ccv_cnnp_model_convolution_transpose_t*)super; | |||
| 1341 | return ccv_cnnp_convolution_transpose(self->groups, self->filters, self->kdim, self->dilation, self->output_padding, self->no_bias, self->hint, self->format, self->super.is_trainable, self->super.name); | |||
| 1342 | } | |||
| 1343 | ||||
| 1344 | // MARK - Dense Layer | |||
| 1345 | ||||
| 1346 | typedef struct { | |||
| 1347 | ccv_cnnp_model_t super; | |||
| 1348 | ccv_nnc_tensor_symbol_t output; | |||
| 1349 | ccv_nnc_tensor_symbol_t weights; | |||
| 1350 | ccv_nnc_tensor_symbol_t bias; | |||
| 1351 | int count; | |||
| 1352 | int no_bias; | |||
| 1353 | int flags; | |||
| 1354 | } ccv_cnnp_model_dense_t; | |||
| 1355 | ||||
| 1356 | static void _ccv_cnnp_dense_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1357 | { | |||
| 1358 | ccv_cnnp_model_dense_t* const self = (ccv_cnnp_model_dense_t*)super; | |||
| 1359 | PRINT(CCV_CLI_VERBOSE, "[cnnp_dense_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_dense_build] -\n"); fflush(stdout); } } while (0); | |||
| 1360 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1360, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1361 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1361, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1362 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1363 | ccv_nnc_tensor_param_t weights_params = params; | |||
| 1364 | memset(weights_params.dim, 0, sizeof(weights_params.dim)); | |||
| 1365 | weights_params.dim[0] = self->count; | |||
| 1366 | weights_params.dim[1] = params.dim[ccv_nnc_tensor_nd(params.dim) - 1]; | |||
| 1367 | if (!self->weights.graph) | |||
| 1368 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 1369 | assert(self->weights.graph == graph)((void) sizeof ((self->weights.graph == graph) ? 1 : 0), __extension__ ({ if (self->weights.graph == graph) ; else __assert_fail ("self->weights.graph == graph", "ccv_cnnp_model_addons.c" , 1369, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1370 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 1371 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 1372 | bias_params.dim[0] = self->count; | |||
| 1373 | ccv_nnc_cmd_t cmd = {0}; | |||
| 1374 | cmd.cmd = CCV_NNC_GEMM_FORWARD; | |||
| 1375 | cmd.info.blas.a[0] = 1; | |||
| 1376 | cmd.info.blas.a[1] = 1; | |||
| 1377 | cmd.info.blas.transpose_b[0] = 0; | |||
| 1378 | cmd.info.blas.transpose_b[1] = 1; | |||
| 1379 | cmd.info.blas.flags = self->flags; | |||
| 1380 | ccv_nnc_tensor_param_t output_params; | |||
| 1381 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 1382 | params, | |||
| 1383 | weights_params, | |||
| 1384 | bias_params, | |||
| 1385 | }, 3, ccv_nnc_no_hint, &output_params, 1); | |||
| 1386 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1387 | if (self->no_bias) | |||
| 1388 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], self->weights)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->weights }, (1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "dense"); | |||
| 1389 | else { | |||
| 1390 | if (!self->bias.graph) | |||
| 1391 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 1392 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], self->weights, self->bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->weights , self->bias}, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "dense"); | |||
| 1393 | } | |||
| 1394 | outputs[0] = output; | |||
| 1395 | } | |||
| 1396 | ||||
| 1397 | static void _ccv_cnnp_dense_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 1398 | { | |||
| 1399 | ccv_cnnp_model_dense_t* const self = (ccv_cnnp_model_dense_t*)super; | |||
| 1400 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 1401 | const int c = weight_params.dim[1]; | |||
| 1402 | const float std = sqrtf(2) / sqrtf(c); | |||
| 1403 | const float bound = sqrtf(3) * std; | |||
| 1404 | initializer(context, CMD_RANDOM_UNIFORM_FORWARD(-bound, bound)ccv_nnc_cmd(CCV_NNC_RANDOM_UNIFORM_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={-bound, bound}}}, 0), ccv_nnc_no_hint, 0, 0, self->weights); | |||
| 1405 | if (self->bias.graph) | |||
| 1406 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, self->bias); | |||
| 1407 | } | |||
| 1408 | ||||
| 1409 | static void _ccv_cnnp_dense_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 1410 | { | |||
| 1411 | ccv_cnnp_model_dense_t* const self = (ccv_cnnp_model_dense_t*)super; | |||
| 1412 | add_to_array(parameters, self->weights, is_trainable); | |||
| 1413 | if (self->bias.graph) | |||
| 1414 | add_to_array(parameters, self->bias, is_trainable); | |||
| 1415 | } | |||
| 1416 | ||||
| 1417 | static ccv_cnnp_model_t* _ccv_cnnp_dense_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1418 | ||||
| 1419 | static const ccv_cnnp_model_vtab_t ccv_cnnp_dense_isa = { | |||
| 1420 | .build = _ccv_cnnp_dense_build, | |||
| 1421 | .init_states = _ccv_cnnp_dense_init_states, | |||
| 1422 | .add_to_parameter = _ccv_cnnp_dense_add_to_parameter, | |||
| 1423 | .copy = _ccv_cnnp_dense_copy, | |||
| 1424 | }; | |||
| 1425 | ||||
| 1426 | ccv_cnnp_model_t* ccv_cnnp_dense(const int count, const int no_bias, const int flags, const int is_trainable, const char* const name) | |||
| 1427 | { | |||
| 1428 | ccv_cnnp_model_dense_t* const model_dense = (ccv_cnnp_model_dense_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_dense_t)); | |||
| 1429 | model_dense->super.isa = &ccv_cnnp_dense_isa; | |||
| 1430 | model_dense->super.input_size = 1; | |||
| 1431 | model_dense->super.outputs = &model_dense->output; | |||
| 1432 | model_dense->super.output_size = 1; | |||
| 1433 | model_dense->super.is_trainable = is_trainable; | |||
| 1434 | ccv_cnnp_model_copy_name(&model_dense->super, name); | |||
| 1435 | model_dense->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1436 | model_dense->weights.graph = 0; | |||
| 1437 | model_dense->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1438 | model_dense->bias.graph = 0; | |||
| 1439 | model_dense->count = count; | |||
| 1440 | model_dense->no_bias = no_bias; | |||
| 1441 | model_dense->flags = flags; | |||
| 1442 | return (ccv_cnnp_model_t*)model_dense; | |||
| 1443 | } | |||
| 1444 | ||||
| 1445 | static ccv_cnnp_model_t* _ccv_cnnp_dense_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1446 | { | |||
| 1447 | const ccv_cnnp_model_dense_t* const self = (const ccv_cnnp_model_dense_t*)super; | |||
| 1448 | return ccv_cnnp_dense(self->count, self->no_bias, self->flags, self->super.is_trainable, self->super.name); | |||
| 1449 | } | |||
| 1450 | ||||
| 1451 | // MARK - Pool Layers | |||
| 1452 | ||||
| 1453 | typedef struct { | |||
| 1454 | ccv_cnnp_model_t super; | |||
| 1455 | ccv_nnc_tensor_symbol_t output; | |||
| 1456 | int kdim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1457 | ccv_nnc_hint_t hint; | |||
| 1458 | } ccv_cnnp_model_pool_t; | |||
| 1459 | ||||
| 1460 | static void _ccv_cnnp_max_pool_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1461 | { | |||
| 1462 | ccv_cnnp_model_pool_t* const self = (ccv_cnnp_model_pool_t*)super; | |||
| 1463 | PRINT(CCV_CLI_VERBOSE, "[cnnp_max_pool_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_max_pool_build] -\n"); fflush(stdout); } } while (0); | |||
| 1464 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1464, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1465 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1465, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1466 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1467 | const int hw = ccv_nnc_tensor_hw(params, ccv_nnc_tensor_nd(params.dim)); | |||
| 1468 | ccv_nnc_cmd_t cmd; | |||
| 1469 | if (hw >= 0 && self->kdim[0] == 0 && self->kdim[1] == 0) | |||
| 1470 | cmd = CMD_MAX_POOL_FORWARD(params.dim[hw], params.dim[hw + 1])ccv_nnc_cmd(CCV_NNC_MAX_POOL_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={params.dim[hw], params.dim[hw + 1],1}}}), 0); | |||
| 1471 | else | |||
| 1472 | cmd = CMD_MAX_POOL_FORWARD(self->kdim[0], self->kdim[1])ccv_nnc_cmd(CCV_NNC_MAX_POOL_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={self->kdim[0], self->kdim[1],1}}}), 0); | |||
| 1473 | ccv_nnc_tensor_param_t output_params; | |||
| 1474 | ccv_nnc_hint_tensor_auto(cmd, ¶ms, 1, self->hint, &output_params, 1); | |||
| 1475 | const ccv_nnc_tensor_symbol_t pool_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1476 | const ccv_nnc_graph_exec_symbol_t exec = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(pool_output)(const ccv_nnc_tensor_symbol_t []){pool_output}, (1 +1 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "max_pool"); | |||
| 1477 | ccv_nnc_graph_exec_symbol_set_hint(graph, exec, self->hint); | |||
| 1478 | outputs[0] = pool_output; | |||
| 1479 | } | |||
| 1480 | ||||
| 1481 | static ccv_cnnp_model_t* _ccv_cnnp_max_pool_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1482 | ||||
| 1483 | static const ccv_cnnp_model_vtab_t ccv_cnnp_max_pool_isa = { | |||
| 1484 | .build = _ccv_cnnp_max_pool_build, | |||
| 1485 | .copy = _ccv_cnnp_max_pool_copy, | |||
| 1486 | }; | |||
| 1487 | ||||
| 1488 | ccv_cnnp_model_t* ccv_cnnp_max_pool(const int kdim[CCV_NNC_MAX_DIM_ALLOC(12)], const ccv_nnc_hint_t hint, const char* const name) | |||
| 1489 | { | |||
| 1490 | ccv_cnnp_model_pool_t* const model_pool = (ccv_cnnp_model_pool_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_pool_t)); | |||
| 1491 | model_pool->super.isa = &ccv_cnnp_max_pool_isa; | |||
| 1492 | model_pool->super.input_size = 1; | |||
| 1493 | model_pool->super.outputs = &model_pool->output; | |||
| 1494 | model_pool->super.output_size = 1; | |||
| 1495 | ccv_cnnp_model_copy_name(&model_pool->super, name); | |||
| 1496 | memcpy(model_pool->kdim, kdim, sizeof(model_pool->kdim)); | |||
| 1497 | model_pool->hint = hint; | |||
| 1498 | return (ccv_cnnp_model_t*)model_pool; | |||
| 1499 | } | |||
| 1500 | ||||
| 1501 | static ccv_cnnp_model_t* _ccv_cnnp_max_pool_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1502 | { | |||
| 1503 | const ccv_cnnp_model_pool_t* const self = (const ccv_cnnp_model_pool_t*)super; | |||
| 1504 | return ccv_cnnp_max_pool(self->kdim, self->hint, self->super.name); | |||
| 1505 | } | |||
| 1506 | ||||
| 1507 | static void _ccv_cnnp_average_pool_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1508 | { | |||
| 1509 | ccv_cnnp_model_pool_t* const self = (ccv_cnnp_model_pool_t*)super; | |||
| 1510 | PRINT(CCV_CLI_VERBOSE, "[cnnp_average_pool_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_average_pool_build] -\n"); fflush(stdout); } } while (0); | |||
| 1511 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1511, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1512 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1512, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1513 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1514 | const int hw = ccv_nnc_tensor_hw(params, ccv_nnc_tensor_nd(params.dim)); | |||
| 1515 | ccv_nnc_cmd_t cmd; | |||
| 1516 | if (hw >= 0 && self->kdim[0] == 0 && self->kdim[1] == 0) | |||
| 1517 | cmd = CMD_AVERAGE_POOL_FORWARD(params.dim[hw], params.dim[hw + 1])ccv_nnc_cmd(CCV_NNC_AVERAGE_POOL_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={params.dim[hw], params.dim[hw + 1],1}}}), 0); | |||
| 1518 | else | |||
| 1519 | cmd = CMD_AVERAGE_POOL_FORWARD(self->kdim[0], self->kdim[1])ccv_nnc_cmd(CCV_NNC_AVERAGE_POOL_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={self->kdim[0], self->kdim[1],1}}}), 0); | |||
| 1520 | ccv_nnc_tensor_param_t output_params; | |||
| 1521 | ccv_nnc_hint_tensor_auto(cmd, ¶ms, 1, self->hint, &output_params, 1); | |||
| 1522 | const ccv_nnc_tensor_symbol_t pool_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1523 | const ccv_nnc_graph_exec_symbol_t exec = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(pool_output)(const ccv_nnc_tensor_symbol_t []){pool_output}, (1 +1 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "average_pool"); | |||
| 1524 | ccv_nnc_graph_exec_symbol_set_hint(graph, exec, self->hint); | |||
| 1525 | outputs[0] = pool_output; | |||
| 1526 | } | |||
| 1527 | ||||
| 1528 | static ccv_cnnp_model_t* _ccv_cnnp_average_pool_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1529 | ||||
| 1530 | static const ccv_cnnp_model_vtab_t ccv_cnnp_average_pool_isa = { | |||
| 1531 | .build = _ccv_cnnp_average_pool_build, | |||
| 1532 | .copy = _ccv_cnnp_average_pool_copy, | |||
| 1533 | }; | |||
| 1534 | ||||
| 1535 | ccv_cnnp_model_t* ccv_cnnp_average_pool(const int kdim[CCV_NNC_MAX_DIM_ALLOC(12)], const ccv_nnc_hint_t hint, const char* const name) | |||
| 1536 | { | |||
| 1537 | ccv_cnnp_model_pool_t* const model_pool = (ccv_cnnp_model_pool_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_pool_t)); | |||
| 1538 | model_pool->super.isa = &ccv_cnnp_average_pool_isa; | |||
| 1539 | model_pool->super.input_size = 1; | |||
| 1540 | model_pool->super.outputs = &model_pool->output; | |||
| 1541 | model_pool->super.output_size = 1; | |||
| 1542 | ccv_cnnp_model_copy_name(&model_pool->super, name); | |||
| 1543 | memcpy(model_pool->kdim, kdim, sizeof(model_pool->kdim)); | |||
| 1544 | model_pool->hint = hint; | |||
| 1545 | return (ccv_cnnp_model_t*)model_pool; | |||
| 1546 | } | |||
| 1547 | ||||
| 1548 | static ccv_cnnp_model_t* _ccv_cnnp_average_pool_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1549 | { | |||
| 1550 | const ccv_cnnp_model_pool_t* const self = (const ccv_cnnp_model_pool_t*)super; | |||
| 1551 | return ccv_cnnp_average_pool(self->kdim, self->hint, self->super.name); | |||
| 1552 | } | |||
| 1553 | ||||
| 1554 | // MARK - RELU Layer | |||
| 1555 | ||||
| 1556 | typedef struct { | |||
| 1557 | ccv_cnnp_model_t super; | |||
| 1558 | ccv_nnc_tensor_symbol_t output; | |||
| 1559 | } ccv_cnnp_model_relu_t; | |||
| 1560 | ||||
| 1561 | static void _ccv_cnnp_relu_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1562 | { | |||
| 1563 | PRINT(CCV_CLI_VERBOSE, "[cnnp_relu_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_relu_build] -\n"); fflush(stdout); } } while (0); | |||
| 1564 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1564, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1565 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1565, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1566 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1567 | ccv_nnc_tensor_param_t output_params; | |||
| 1568 | const ccv_nnc_cmd_t relu = CMD_RELU_FORWARD()ccv_nnc_cmd(CCV_NNC_RELU_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1569 | ccv_nnc_hint_tensor_auto(relu, (ccv_nnc_tensor_param_t []){ | |||
| 1570 | params, | |||
| 1571 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1572 | const ccv_nnc_tensor_symbol_t relu_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1573 | ccv_nnc_graph_exec_symbol_new(graph, relu, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(relu_output)(const ccv_nnc_tensor_symbol_t []){relu_output}, (1 +1 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "relu"); | |||
| 1574 | outputs[0] = relu_output; | |||
| 1575 | } | |||
| 1576 | ||||
| 1577 | static ccv_cnnp_model_t* _ccv_cnnp_relu_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1578 | ||||
| 1579 | static const ccv_cnnp_model_vtab_t ccv_cnnp_relu_isa = { | |||
| 1580 | .build = _ccv_cnnp_relu_build, | |||
| 1581 | .copy = _ccv_cnnp_relu_copy, | |||
| 1582 | }; | |||
| 1583 | ||||
| 1584 | ccv_cnnp_model_t* ccv_cnnp_relu(const char* const name) | |||
| 1585 | { | |||
| 1586 | ccv_cnnp_model_relu_t* const model_relu = (ccv_cnnp_model_relu_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_relu_t)); | |||
| 1587 | model_relu->super.isa = &ccv_cnnp_relu_isa; | |||
| 1588 | model_relu->super.input_size = 1; | |||
| 1589 | model_relu->super.outputs = &model_relu->output; | |||
| 1590 | model_relu->super.output_size = 1; | |||
| 1591 | ccv_cnnp_model_copy_name(&model_relu->super, name); | |||
| 1592 | return (ccv_cnnp_model_t*)model_relu; | |||
| 1593 | } | |||
| 1594 | ||||
| 1595 | static ccv_cnnp_model_t* _ccv_cnnp_relu_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1596 | { | |||
| 1597 | return ccv_cnnp_relu(self->name); | |||
| 1598 | } | |||
| 1599 | ||||
| 1600 | // MARK - Sigmoid Layer | |||
| 1601 | ||||
| 1602 | typedef struct { | |||
| 1603 | ccv_cnnp_model_t super; | |||
| 1604 | ccv_nnc_tensor_symbol_t output; | |||
| 1605 | } ccv_cnnp_model_sigmoid_t; | |||
| 1606 | ||||
| 1607 | static void _ccv_cnnp_sigmoid_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1608 | { | |||
| 1609 | PRINT(CCV_CLI_VERBOSE, "[cnnp_sigmoid_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_sigmoid_build] -\n"); fflush(stdout); } } while (0); | |||
| 1610 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1610, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1611 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1611, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1612 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1613 | ccv_nnc_tensor_param_t output_params; | |||
| 1614 | const ccv_nnc_cmd_t sigmoid = CMD_SIGMOID_FORWARD()ccv_nnc_cmd(CCV_NNC_SIGMOID_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1615 | ccv_nnc_hint_tensor_auto(sigmoid, (ccv_nnc_tensor_param_t []){ | |||
| 1616 | params, | |||
| 1617 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1618 | const ccv_nnc_tensor_symbol_t sigmoid_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1619 | ccv_nnc_graph_exec_symbol_new(graph, sigmoid, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(sigmoid_output)(const ccv_nnc_tensor_symbol_t []){sigmoid_output}, (1 +1 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1 ), "sigmoid"); | |||
| 1620 | outputs[0] = sigmoid_output; | |||
| 1621 | } | |||
| 1622 | ||||
| 1623 | static ccv_cnnp_model_t* _ccv_cnnp_sigmoid_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1624 | ||||
| 1625 | static const ccv_cnnp_model_vtab_t ccv_cnnp_sigmoid_isa = { | |||
| 1626 | .build = _ccv_cnnp_sigmoid_build, | |||
| 1627 | .copy = _ccv_cnnp_sigmoid_copy, | |||
| 1628 | }; | |||
| 1629 | ||||
| 1630 | ccv_cnnp_model_t* ccv_cnnp_sigmoid(const char* const name) | |||
| 1631 | { | |||
| 1632 | ccv_cnnp_model_sigmoid_t* const model_sigmoid = (ccv_cnnp_model_sigmoid_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_sigmoid_t)); | |||
| 1633 | model_sigmoid->super.isa = &ccv_cnnp_sigmoid_isa; | |||
| 1634 | model_sigmoid->super.input_size = 1; | |||
| 1635 | model_sigmoid->super.outputs = &model_sigmoid->output; | |||
| 1636 | model_sigmoid->super.output_size = 1; | |||
| 1637 | ccv_cnnp_model_copy_name(&model_sigmoid->super, name); | |||
| 1638 | return (ccv_cnnp_model_t*)model_sigmoid; | |||
| 1639 | } | |||
| 1640 | ||||
| 1641 | static ccv_cnnp_model_t* _ccv_cnnp_sigmoid_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1642 | { | |||
| 1643 | return ccv_cnnp_sigmoid(self->name); | |||
| 1644 | } | |||
| 1645 | ||||
| 1646 | // MARK - Tanh Layer | |||
| 1647 | ||||
| 1648 | typedef struct { | |||
| 1649 | ccv_cnnp_model_t super; | |||
| 1650 | ccv_nnc_tensor_symbol_t output; | |||
| 1651 | } ccv_cnnp_model_tanh_t; | |||
| 1652 | ||||
| 1653 | static void _ccv_cnnp_tanh_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1654 | { | |||
| 1655 | PRINT(CCV_CLI_VERBOSE, "[cnnp_tanh_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_tanh_build] -\n"); fflush(stdout); } } while (0); | |||
| 1656 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1656, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1657 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1657, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1658 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1659 | ccv_nnc_tensor_param_t output_params; | |||
| 1660 | const ccv_nnc_cmd_t tanh = CMD_TANH_FORWARD()ccv_nnc_cmd(CCV_NNC_TANH_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1661 | ccv_nnc_hint_tensor_auto(tanh, (ccv_nnc_tensor_param_t []){ | |||
| 1662 | params, | |||
| 1663 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1664 | const ccv_nnc_tensor_symbol_t tanh_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1665 | ccv_nnc_graph_exec_symbol_new(graph, tanh, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(tanh_output)(const ccv_nnc_tensor_symbol_t []){tanh_output}, (1 +1 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "tanh"); | |||
| 1666 | outputs[0] = tanh_output; | |||
| 1667 | } | |||
| 1668 | ||||
| 1669 | static ccv_cnnp_model_t* _ccv_cnnp_tanh_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1670 | ||||
| 1671 | static const ccv_cnnp_model_vtab_t ccv_cnnp_tanh_isa = { | |||
| 1672 | .build = _ccv_cnnp_tanh_build, | |||
| 1673 | .copy = _ccv_cnnp_tanh_copy, | |||
| 1674 | }; | |||
| 1675 | ||||
| 1676 | ccv_cnnp_model_t* ccv_cnnp_tanh(const char* const name) | |||
| 1677 | { | |||
| 1678 | ccv_cnnp_model_tanh_t* const model_tanh = (ccv_cnnp_model_tanh_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_tanh_t)); | |||
| 1679 | model_tanh->super.isa = &ccv_cnnp_tanh_isa; | |||
| 1680 | model_tanh->super.input_size = 1; | |||
| 1681 | model_tanh->super.outputs = &model_tanh->output; | |||
| 1682 | model_tanh->super.output_size = 1; | |||
| 1683 | ccv_cnnp_model_copy_name(&model_tanh->super, name); | |||
| 1684 | return (ccv_cnnp_model_t*)model_tanh; | |||
| 1685 | } | |||
| 1686 | ||||
| 1687 | static ccv_cnnp_model_t* _ccv_cnnp_tanh_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1688 | { | |||
| 1689 | return ccv_cnnp_tanh(self->name); | |||
| 1690 | } | |||
| 1691 | ||||
| 1692 | // MARK - Swish Layer | |||
| 1693 | ||||
| 1694 | typedef struct { | |||
| 1695 | ccv_cnnp_model_t super; | |||
| 1696 | ccv_nnc_tensor_symbol_t output; | |||
| 1697 | } ccv_cnnp_model_swish_t; | |||
| 1698 | ||||
| 1699 | static void _ccv_cnnp_swish_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1700 | { | |||
| 1701 | PRINT(CCV_CLI_VERBOSE, "[cnnp_swish_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_swish_build] -\n"); fflush(stdout); } } while (0); | |||
| 1702 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1702, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1703 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1703, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1704 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1705 | ccv_nnc_tensor_param_t output_params; | |||
| 1706 | const ccv_nnc_cmd_t swish = CMD_SWISH_FORWARD()ccv_nnc_cmd(CCV_NNC_SWISH_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1707 | ccv_nnc_hint_tensor_auto(swish, (ccv_nnc_tensor_param_t []){ | |||
| 1708 | params, | |||
| 1709 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1710 | const ccv_nnc_tensor_symbol_t swish_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1711 | ccv_nnc_graph_exec_symbol_new(graph, swish, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(swish_output)(const ccv_nnc_tensor_symbol_t []){swish_output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "swish"); | |||
| 1712 | outputs[0] = swish_output; | |||
| 1713 | } | |||
| 1714 | ||||
| 1715 | static ccv_cnnp_model_t* _ccv_cnnp_swish_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1716 | ||||
| 1717 | static const ccv_cnnp_model_vtab_t ccv_cnnp_swish_isa = { | |||
| 1718 | .build = _ccv_cnnp_swish_build, | |||
| 1719 | .copy = _ccv_cnnp_swish_copy, | |||
| 1720 | }; | |||
| 1721 | ||||
| 1722 | ccv_cnnp_model_t* ccv_cnnp_swish(const char* const name) | |||
| 1723 | { | |||
| 1724 | ccv_cnnp_model_swish_t* const model_swish = (ccv_cnnp_model_swish_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_swish_t)); | |||
| 1725 | model_swish->super.isa = &ccv_cnnp_swish_isa; | |||
| 1726 | model_swish->super.input_size = 1; | |||
| 1727 | model_swish->super.outputs = &model_swish->output; | |||
| 1728 | model_swish->super.output_size = 1; | |||
| 1729 | ccv_cnnp_model_copy_name(&model_swish->super, name); | |||
| 1730 | return (ccv_cnnp_model_t*)model_swish; | |||
| 1731 | } | |||
| 1732 | ||||
| 1733 | static ccv_cnnp_model_t* _ccv_cnnp_swish_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1734 | { | |||
| 1735 | return ccv_cnnp_swish(self->name); | |||
| 1736 | } | |||
| 1737 | ||||
| 1738 | // MARK - GELU Layer | |||
| 1739 | ||||
| 1740 | typedef struct { | |||
| 1741 | ccv_cnnp_model_t super; | |||
| 1742 | ccv_nnc_tensor_symbol_t output; | |||
| 1743 | int tanh; | |||
| 1744 | } ccv_cnnp_model_gelu_t; | |||
| 1745 | ||||
| 1746 | static void _ccv_cnnp_gelu_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1747 | { | |||
| 1748 | PRINT(CCV_CLI_VERBOSE, "[cnnp_gelu_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_gelu_build] -\n"); fflush(stdout); } } while (0); | |||
| 1749 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1749, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1750 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1750, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1751 | ccv_cnnp_model_gelu_t* const self = (ccv_cnnp_model_gelu_t*)super; | |||
| 1752 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1753 | ccv_nnc_tensor_param_t output_params; | |||
| 1754 | const ccv_nnc_cmd_t gelu = CMD_GELU_FORWARD(self->tanh)ccv_nnc_cmd(CCV_NNC_GELU_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.gelu={.tanh=self->tanh}}, 0); | |||
| 1755 | ccv_nnc_hint_tensor_auto(gelu, (ccv_nnc_tensor_param_t []){ | |||
| 1756 | params, | |||
| 1757 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1758 | const ccv_nnc_tensor_symbol_t gelu_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1759 | ccv_nnc_graph_exec_symbol_new(graph, gelu, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(gelu_output)(const ccv_nnc_tensor_symbol_t []){gelu_output}, (1 +1 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "gelu"); | |||
| 1760 | outputs[0] = gelu_output; | |||
| 1761 | } | |||
| 1762 | ||||
| 1763 | static ccv_cnnp_model_t* _ccv_cnnp_gelu_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1764 | ||||
| 1765 | static const ccv_cnnp_model_vtab_t ccv_cnnp_gelu_isa = { | |||
| 1766 | .build = _ccv_cnnp_gelu_build, | |||
| 1767 | .copy = _ccv_cnnp_gelu_copy, | |||
| 1768 | }; | |||
| 1769 | ||||
| 1770 | ccv_cnnp_model_t* ccv_cnnp_gelu(const int tanh, const char* const name) | |||
| 1771 | { | |||
| 1772 | ccv_cnnp_model_gelu_t* const model_gelu = (ccv_cnnp_model_gelu_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_gelu_t)); | |||
| 1773 | model_gelu->super.isa = &ccv_cnnp_gelu_isa; | |||
| 1774 | model_gelu->super.input_size = 1; | |||
| 1775 | model_gelu->super.outputs = &model_gelu->output; | |||
| 1776 | model_gelu->super.output_size = 1; | |||
| 1777 | model_gelu->tanh = tanh; | |||
| 1778 | ccv_cnnp_model_copy_name(&model_gelu->super, name); | |||
| 1779 | return (ccv_cnnp_model_t*)model_gelu; | |||
| 1780 | } | |||
| 1781 | ||||
| 1782 | static ccv_cnnp_model_t* _ccv_cnnp_gelu_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1783 | { | |||
| 1784 | ccv_cnnp_model_gelu_t* const self = (ccv_cnnp_model_gelu_t*)super; | |||
| 1785 | return ccv_cnnp_gelu(self->tanh, self->super.name); | |||
| 1786 | } | |||
| 1787 | ||||
| 1788 | // MARK - Leaky ReLU Layer | |||
| 1789 | ||||
| 1790 | typedef struct { | |||
| 1791 | ccv_cnnp_model_t super; | |||
| 1792 | ccv_nnc_tensor_symbol_t output; | |||
| 1793 | float negative_slope; | |||
| 1794 | } ccv_cnnp_model_leaky_relu_t; | |||
| 1795 | ||||
| 1796 | static void _ccv_cnnp_leaky_relu_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1797 | { | |||
| 1798 | PRINT(CCV_CLI_VERBOSE, "[cnnp_leaky_relu_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_leaky_relu_build] -\n"); fflush(stdout); } } while (0); | |||
| 1799 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1799, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1800 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1800, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1801 | ccv_cnnp_model_leaky_relu_t* const self = (ccv_cnnp_model_leaky_relu_t*)super; | |||
| 1802 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1803 | ccv_nnc_tensor_param_t output_params; | |||
| 1804 | const ccv_nnc_cmd_t leaky_relu = CMD_LEAKY_RELU_FORWARD(self->negative_slope)ccv_nnc_cmd(CCV_NNC_LEAKY_RELU_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.leaky_relu={.negative_slope=self-> negative_slope}}, 0); | |||
| 1805 | ccv_nnc_hint_tensor_auto(leaky_relu, (ccv_nnc_tensor_param_t []){ | |||
| 1806 | params, | |||
| 1807 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1808 | const ccv_nnc_tensor_symbol_t leaky_relu_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1809 | ccv_nnc_graph_exec_symbol_new(graph, leaky_relu, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(leaky_relu_output)(const ccv_nnc_tensor_symbol_t []){leaky_relu_output}, (1 +1 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "leaky_relu"); | |||
| 1810 | outputs[0] = leaky_relu_output; | |||
| 1811 | } | |||
| 1812 | ||||
| 1813 | static ccv_cnnp_model_t* _ccv_cnnp_leaky_relu_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1814 | ||||
| 1815 | static const ccv_cnnp_model_vtab_t ccv_cnnp_leaky_relu_isa = { | |||
| 1816 | .build = _ccv_cnnp_leaky_relu_build, | |||
| 1817 | .copy = _ccv_cnnp_leaky_relu_copy, | |||
| 1818 | }; | |||
| 1819 | ||||
| 1820 | ccv_cnnp_model_t* ccv_cnnp_leaky_relu(const float negative_slope, const char* const name) | |||
| 1821 | { | |||
| 1822 | ccv_cnnp_model_leaky_relu_t* const model_leaky_relu = (ccv_cnnp_model_leaky_relu_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_leaky_relu_t)); | |||
| 1823 | model_leaky_relu->super.isa = &ccv_cnnp_leaky_relu_isa; | |||
| 1824 | model_leaky_relu->super.input_size = 1; | |||
| 1825 | model_leaky_relu->super.outputs = &model_leaky_relu->output; | |||
| 1826 | model_leaky_relu->super.output_size = 1; | |||
| 1827 | model_leaky_relu->negative_slope = negative_slope; | |||
| 1828 | ccv_cnnp_model_copy_name(&model_leaky_relu->super, name); | |||
| 1829 | return (ccv_cnnp_model_t*)model_leaky_relu; | |||
| 1830 | } | |||
| 1831 | ||||
| 1832 | static ccv_cnnp_model_t* _ccv_cnnp_leaky_relu_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1833 | { | |||
| 1834 | ccv_cnnp_model_leaky_relu_t* const self = (ccv_cnnp_model_leaky_relu_t*)super; | |||
| 1835 | return ccv_cnnp_leaky_relu(self->negative_slope, self->super.name); | |||
| 1836 | } | |||
| 1837 | ||||
| 1838 | // MARK - Softmax Layer | |||
| 1839 | ||||
| 1840 | typedef struct { | |||
| 1841 | ccv_cnnp_model_t super; | |||
| 1842 | ccv_nnc_tensor_symbol_t output; | |||
| 1843 | } ccv_cnnp_model_softmax_t; | |||
| 1844 | ||||
| 1845 | static void _ccv_cnnp_softmax_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1846 | { | |||
| 1847 | PRINT(CCV_CLI_VERBOSE, "[cnnp_softmax_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_softmax_build] -\n"); fflush(stdout); } } while (0); | |||
| 1848 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1848, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1849 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1849, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1850 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1851 | ccv_nnc_tensor_param_t output_params; | |||
| 1852 | const ccv_nnc_cmd_t softmax = CMD_SOFTMAX_FORWARD()ccv_nnc_cmd(CCV_NNC_SOFTMAX_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1853 | ccv_nnc_hint_tensor_auto(softmax, (ccv_nnc_tensor_param_t []){ | |||
| 1854 | params, | |||
| 1855 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1856 | const ccv_nnc_tensor_symbol_t softmax_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1857 | ccv_nnc_graph_exec_symbol_new(graph, softmax, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(softmax_output)(const ccv_nnc_tensor_symbol_t []){softmax_output}, (1 +1 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1 ), "softmax"); | |||
| 1858 | outputs[0] = softmax_output; | |||
| 1859 | } | |||
| 1860 | ||||
| 1861 | static ccv_cnnp_model_t* _ccv_cnnp_softmax_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1862 | ||||
| 1863 | static const ccv_cnnp_model_vtab_t ccv_cnnp_softmax_isa = { | |||
| 1864 | .build = _ccv_cnnp_softmax_build, | |||
| 1865 | .copy = _ccv_cnnp_softmax_copy, | |||
| 1866 | }; | |||
| 1867 | ||||
| 1868 | ccv_cnnp_model_t* ccv_cnnp_softmax(const char* const name) | |||
| 1869 | { | |||
| 1870 | ccv_cnnp_model_softmax_t* const model_softmax = (ccv_cnnp_model_softmax_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_softmax_t)); | |||
| 1871 | model_softmax->super.isa = &ccv_cnnp_softmax_isa; | |||
| 1872 | model_softmax->super.input_size = 1; | |||
| 1873 | model_softmax->super.outputs = &model_softmax->output; | |||
| 1874 | model_softmax->super.output_size = 1; | |||
| 1875 | ccv_cnnp_model_copy_name(&model_softmax->super, name); | |||
| 1876 | return (ccv_cnnp_model_t*)model_softmax; | |||
| 1877 | } | |||
| 1878 | ||||
| 1879 | static ccv_cnnp_model_t* _ccv_cnnp_softmax_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1880 | { | |||
| 1881 | return ccv_cnnp_softmax(self->name); | |||
| 1882 | } | |||
| 1883 | ||||
| 1884 | // MARK - Add Layer | |||
| 1885 | ||||
| 1886 | typedef struct { | |||
| 1887 | ccv_cnnp_model_t super; | |||
| 1888 | float p; | |||
| 1889 | float q; | |||
| 1890 | ccv_nnc_tensor_symbol_t output; | |||
| 1891 | } ccv_cnnp_model_add_t; | |||
| 1892 | ||||
| 1893 | static void _ccv_cnnp_add_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1894 | { | |||
| 1895 | PRINT(CCV_CLI_VERBOSE, "[cnnp_add_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_add_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 1896 | const ccv_cnnp_model_add_t* const self = (const ccv_cnnp_model_add_t*)super; | |||
| 1897 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 1897, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1898 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1898, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1899 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 1900 | int i; | |||
| 1901 | for (i = 0; i < 2; i++) | |||
| 1902 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 1903 | ccv_nnc_tensor_param_t output_params; | |||
| 1904 | const ccv_nnc_cmd_t add = CMD_ADD_FORWARD(self->p, self->q)ccv_nnc_cmd(CCV_NNC_ADD_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={self->p, self->q}}}, 0); | |||
| 1905 | ccv_nnc_hint_tensor_auto(add, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 1906 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1907 | ccv_nnc_graph_exec_symbol_new(graph, add, inputs, input_size, outputs, output_size, "add"); | |||
| 1908 | } | |||
| 1909 | ||||
| 1910 | static ccv_cnnp_model_t* _ccv_cnnp_add_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1911 | ||||
| 1912 | static const ccv_cnnp_model_vtab_t ccv_cnnp_add_isa = { | |||
| 1913 | .build = _ccv_cnnp_add_build, | |||
| 1914 | .copy = _ccv_cnnp_add_copy, | |||
| 1915 | }; | |||
| 1916 | ||||
| 1917 | ccv_cnnp_model_t* ccv_cnnp_add(const float p, const float q, const char* const name) | |||
| 1918 | { | |||
| 1919 | ccv_cnnp_model_add_t* const model_add = (ccv_cnnp_model_add_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_add_t)); | |||
| 1920 | model_add->super.isa = &ccv_cnnp_add_isa; | |||
| 1921 | model_add->super.input_size = 2; | |||
| 1922 | model_add->super.outputs = &model_add->output; | |||
| 1923 | model_add->super.output_size = 1; | |||
| 1924 | model_add->p = p; | |||
| 1925 | model_add->q = q; | |||
| 1926 | ccv_cnnp_model_copy_name(&model_add->super, name); | |||
| 1927 | return (ccv_cnnp_model_t*)model_add; | |||
| 1928 | } | |||
| 1929 | ||||
| 1930 | static ccv_cnnp_model_t* _ccv_cnnp_add_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1931 | { | |||
| 1932 | const ccv_cnnp_model_add_t* const self = (const ccv_cnnp_model_add_t*)super; | |||
| 1933 | return ccv_cnnp_add(self->p, self->q, self->super.name); | |||
| 1934 | } | |||
| 1935 | ||||
| 1936 | // MARK - Mul Layer | |||
| 1937 | ||||
| 1938 | typedef struct { | |||
| 1939 | ccv_cnnp_model_t super; | |||
| 1940 | ccv_nnc_tensor_symbol_t output; | |||
| 1941 | float p; | |||
| 1942 | } ccv_cnnp_model_mul_t; | |||
| 1943 | ||||
| 1944 | static void _ccv_cnnp_mul_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1945 | { | |||
| 1946 | PRINT(CCV_CLI_VERBOSE, "[cnnp_mul_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_mul_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 1947 | const ccv_cnnp_model_mul_t* const self = (const ccv_cnnp_model_mul_t*)super; | |||
| 1948 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 1948, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1949 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1949, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1950 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 1951 | int i; | |||
| 1952 | for (i = 0; i < 2; i++) | |||
| 1953 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 1954 | ccv_nnc_tensor_param_t output_params; | |||
| 1955 | const ccv_nnc_cmd_t mul = CMD_MUL_FORWARD(self->p)ccv_nnc_cmd(CCV_NNC_MUL_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={self->p,}}}, 0); | |||
| 1956 | ccv_nnc_hint_tensor_auto(mul, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 1957 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1958 | ccv_nnc_graph_exec_symbol_new(graph, mul, inputs, input_size, outputs, output_size, "mul"); | |||
| 1959 | } | |||
| 1960 | ||||
| 1961 | static ccv_cnnp_model_t* _ccv_cnnp_mul_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1962 | ||||
| 1963 | static const ccv_cnnp_model_vtab_t ccv_cnnp_mul_isa = { | |||
| 1964 | .build = _ccv_cnnp_mul_build, | |||
| 1965 | .copy = _ccv_cnnp_mul_copy, | |||
| 1966 | }; | |||
| 1967 | ||||
| 1968 | ccv_cnnp_model_t* ccv_cnnp_mul(const float p, const char* const name) | |||
| 1969 | { | |||
| 1970 | ccv_cnnp_model_mul_t* const model_mul = (ccv_cnnp_model_mul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_mul_t)); | |||
| 1971 | model_mul->super.isa = &ccv_cnnp_mul_isa; | |||
| 1972 | model_mul->super.input_size = 2; | |||
| 1973 | model_mul->super.outputs = &model_mul->output; | |||
| 1974 | model_mul->super.output_size = 1; | |||
| 1975 | model_mul->p = p; | |||
| 1976 | ccv_cnnp_model_copy_name(&model_mul->super, name); | |||
| 1977 | return (ccv_cnnp_model_t*)model_mul; | |||
| 1978 | } | |||
| 1979 | ||||
| 1980 | static ccv_cnnp_model_t* _ccv_cnnp_mul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1981 | { | |||
| 1982 | const ccv_cnnp_model_mul_t* const self = (const ccv_cnnp_model_mul_t*)super; | |||
| 1983 | return ccv_cnnp_mul(self->p, self->super.name); | |||
| 1984 | } | |||
| 1985 | ||||
| 1986 | // MARK - Scalar Mul Layer | |||
| 1987 | ||||
| 1988 | typedef struct { | |||
| 1989 | ccv_cnnp_model_t super; | |||
| 1990 | ccv_nnc_tensor_symbol_t output; | |||
| 1991 | float a; | |||
| 1992 | } ccv_cnnp_model_scalar_mul_t; | |||
| 1993 | ||||
| 1994 | static void _ccv_cnnp_scalar_mul_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 1995 | { | |||
| 1996 | PRINT(CCV_CLI_VERBOSE, "[cnnp_scalar_mul_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_scalar_mul_build] -\n"); fflush(stdout); } } while (0); | |||
| 1997 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 1997, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1998 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 1998, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1999 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2000 | ccv_nnc_tensor_param_t output_params; | |||
| 2001 | ccv_cnnp_model_scalar_mul_t* const self = (ccv_cnnp_model_scalar_mul_t*)super; | |||
| 2002 | const ccv_nnc_cmd_t scalar_mul = CMD_SCALAR_MUL_FORWARD(self->a)ccv_nnc_cmd(CCV_NNC_SCALAR_MUL_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={self->a,}}}, 0); | |||
| 2003 | ccv_nnc_hint_tensor_auto(scalar_mul, (ccv_nnc_tensor_param_t []){ | |||
| 2004 | params, | |||
| 2005 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2006 | const ccv_nnc_tensor_symbol_t scalar_mul_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2007 | ccv_nnc_graph_exec_symbol_new(graph, scalar_mul, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(scalar_mul_output)(const ccv_nnc_tensor_symbol_t []){scalar_mul_output}, (1 +1 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "scalar_mul"); | |||
| 2008 | outputs[0] = scalar_mul_output; | |||
| 2009 | } | |||
| 2010 | ||||
| 2011 | static ccv_cnnp_model_t* _ccv_cnnp_scalar_mul_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2012 | ||||
| 2013 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scalar_mul_isa = { | |||
| 2014 | .build = _ccv_cnnp_scalar_mul_build, | |||
| 2015 | .copy = _ccv_cnnp_scalar_mul_copy, | |||
| 2016 | }; | |||
| 2017 | ||||
| 2018 | ccv_cnnp_model_t* ccv_cnnp_scalar_mul(const float a, const char* const name) | |||
| 2019 | { | |||
| 2020 | ccv_cnnp_model_scalar_mul_t* const model_scalar_mul = (ccv_cnnp_model_scalar_mul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_scalar_mul_t)); | |||
| 2021 | model_scalar_mul->super.isa = &ccv_cnnp_scalar_mul_isa; | |||
| 2022 | model_scalar_mul->super.input_size = 1; | |||
| 2023 | model_scalar_mul->super.outputs = &model_scalar_mul->output; | |||
| 2024 | model_scalar_mul->super.output_size = 1; | |||
| 2025 | model_scalar_mul->a = a; | |||
| 2026 | ccv_cnnp_model_copy_name(&model_scalar_mul->super, name); | |||
| 2027 | return (ccv_cnnp_model_t*)model_scalar_mul; | |||
| 2028 | } | |||
| 2029 | ||||
| 2030 | static ccv_cnnp_model_t* _ccv_cnnp_scalar_mul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2031 | { | |||
| 2032 | const ccv_cnnp_model_scalar_mul_t* const self = (const ccv_cnnp_model_scalar_mul_t*)super; | |||
| 2033 | return ccv_cnnp_scalar_mul(self->a, self->super.name); | |||
| 2034 | } | |||
| 2035 | ||||
| 2036 | // MARK - Div Layer | |||
| 2037 | ||||
| 2038 | typedef struct { | |||
| 2039 | ccv_cnnp_model_t super; | |||
| 2040 | ccv_nnc_tensor_symbol_t output; | |||
| 2041 | int reciprocal; | |||
| 2042 | } ccv_cnnp_model_div_t; | |||
| 2043 | ||||
| 2044 | static void _ccv_cnnp_div_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2045 | { | |||
| 2046 | const ccv_cnnp_model_div_t* const self = (const ccv_cnnp_model_div_t*)super; | |||
| 2047 | PRINT(CCV_CLI_VERBOSE, "[cnnp_div_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_div_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 2048 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2048, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2049 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 2050 | int i; | |||
| 2051 | ccv_nnc_tensor_param_t output_params; | |||
| 2052 | const ccv_nnc_cmd_t div = CMD_EWDIV_FORWARD()ccv_nnc_cmd(CCV_NNC_EWDIV_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 2053 | if (self->reciprocal) | |||
| 2054 | { | |||
| 2055 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2055, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2056 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2057 | input_params[1] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2058 | ccv_nnc_hint_tensor_auto(div, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2059 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2060 | ccv_nnc_graph_exec_symbol_new(graph, div, TENSOR_SYMBOL_LIST(NO_TENSOR_SYMBOL, inputs[0])(const ccv_nnc_tensor_symbol_t []){(const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, inputs[0]}, (1 +1 +1 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), outputs, output_size, "div"); | |||
| 2061 | } else { | |||
| 2062 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 2062, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2063 | for (i = 0; i < 2; i++) | |||
| 2064 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 2065 | ccv_nnc_hint_tensor_auto(div, input_params, input_size, ccv_nnc_no_hint, &output_params, 1); | |||
| 2066 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2067 | ccv_nnc_graph_exec_symbol_new(graph, div, inputs, input_size, outputs, output_size, "div"); | |||
| 2068 | } | |||
| 2069 | } | |||
| 2070 | ||||
| 2071 | static ccv_cnnp_model_t* _ccv_cnnp_div_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2072 | ||||
| 2073 | static const ccv_cnnp_model_vtab_t ccv_cnnp_div_isa = { | |||
| 2074 | .build = _ccv_cnnp_div_build, | |||
| 2075 | .copy = _ccv_cnnp_div_copy, | |||
| 2076 | }; | |||
| 2077 | ||||
| 2078 | ccv_cnnp_model_t* ccv_cnnp_div(const int reciprocal, const char* const name) | |||
| 2079 | { | |||
| 2080 | ccv_cnnp_model_div_t* const model_div = (ccv_cnnp_model_div_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_div_t)); | |||
| 2081 | model_div->super.isa = &ccv_cnnp_div_isa; | |||
| 2082 | model_div->super.input_size = reciprocal ? 1 : 2; | |||
| 2083 | model_div->super.outputs = &model_div->output; | |||
| 2084 | model_div->super.output_size = 1; | |||
| 2085 | model_div->reciprocal = reciprocal; | |||
| 2086 | ccv_cnnp_model_copy_name(&model_div->super, name); | |||
| 2087 | return (ccv_cnnp_model_t*)model_div; | |||
| 2088 | } | |||
| 2089 | ||||
| 2090 | static ccv_cnnp_model_t* _ccv_cnnp_div_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2091 | { | |||
| 2092 | const ccv_cnnp_model_div_t* const self = (const ccv_cnnp_model_div_t*)super; | |||
| 2093 | return ccv_cnnp_div(self->reciprocal, self->super.name); | |||
| 2094 | } | |||
| 2095 | ||||
| 2096 | // MARK - Sqrt Layer | |||
| 2097 | ||||
| 2098 | typedef struct { | |||
| 2099 | ccv_cnnp_model_t super; | |||
| 2100 | ccv_nnc_tensor_symbol_t output; | |||
| 2101 | } ccv_cnnp_model_sqrt_t; | |||
| 2102 | ||||
| 2103 | static void _ccv_cnnp_sqrt_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2104 | { | |||
| 2105 | PRINT(CCV_CLI_VERBOSE, "[cnnp_sqrt_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_sqrt_build] -\n"); fflush(stdout); } } while (0); | |||
| 2106 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2106, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2107 | ccv_nnc_tensor_param_t input_params[1]; | |||
| 2108 | ccv_nnc_tensor_param_t output_params; | |||
| 2109 | const ccv_nnc_cmd_t sqrt = CMD_EWSQRT_FORWARD()ccv_nnc_cmd(CCV_NNC_EWSQRT_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 2110 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2110, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2111 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2112 | ccv_nnc_hint_tensor_auto(sqrt, input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2113 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2114 | ccv_nnc_graph_exec_symbol_new(graph, sqrt, inputs, 1, outputs, output_size, "sqrt"); | |||
| 2115 | } | |||
| 2116 | ||||
| 2117 | static ccv_cnnp_model_t* _ccv_cnnp_sqrt_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2118 | ||||
| 2119 | static const ccv_cnnp_model_vtab_t ccv_cnnp_sqrt_isa = { | |||
| 2120 | .build = _ccv_cnnp_sqrt_build, | |||
| 2121 | .copy = _ccv_cnnp_sqrt_copy, | |||
| 2122 | }; | |||
| 2123 | ||||
| 2124 | ccv_cnnp_model_t* ccv_cnnp_sqrt(const char* const name) | |||
| 2125 | { | |||
| 2126 | ccv_cnnp_model_sqrt_t* const model_sqrt = (ccv_cnnp_model_sqrt_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_sqrt_t)); | |||
| 2127 | model_sqrt->super.isa = &ccv_cnnp_sqrt_isa; | |||
| 2128 | model_sqrt->super.input_size = 1; | |||
| 2129 | model_sqrt->super.outputs = &model_sqrt->output; | |||
| 2130 | model_sqrt->super.output_size = 1; | |||
| 2131 | ccv_cnnp_model_copy_name(&model_sqrt->super, name); | |||
| 2132 | return (ccv_cnnp_model_t*)model_sqrt; | |||
| 2133 | } | |||
| 2134 | ||||
| 2135 | static ccv_cnnp_model_t* _ccv_cnnp_sqrt_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2136 | { | |||
| 2137 | const ccv_cnnp_model_sqrt_t* const self = (const ccv_cnnp_model_sqrt_t*)super; | |||
| 2138 | return ccv_cnnp_sqrt(self->super.name); | |||
| 2139 | } | |||
| 2140 | ||||
| 2141 | // MARK - Cmul Layer | |||
| 2142 | ||||
| 2143 | typedef struct { | |||
| 2144 | ccv_cnnp_model_t super; | |||
| 2145 | ccv_nnc_tensor_symbol_t output; | |||
| 2146 | } ccv_cnnp_model_cmul_t; | |||
| 2147 | ||||
| 2148 | static void _ccv_cnnp_cmul_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2149 | { | |||
| 2150 | PRINT(CCV_CLI_VERBOSE, "[cnnp_cmul_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_cmul_build] -\n"); fflush(stdout); } } while (0); | |||
| 2151 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 2151, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2152 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2152, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2153 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 2154 | int i; | |||
| 2155 | for (i = 0; i < 2; i++) | |||
| 2156 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 2157 | ccv_nnc_tensor_param_t output_params; | |||
| 2158 | const ccv_nnc_cmd_t mul = CMD_CMUL_FORWARD()ccv_nnc_cmd(CCV_NNC_CMUL_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}}}, 0); | |||
| 2159 | ccv_nnc_hint_tensor_auto(mul, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2160 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2161 | ccv_nnc_graph_exec_symbol_new(graph, mul, inputs, input_size, outputs, output_size, "cmul"); | |||
| 2162 | } | |||
| 2163 | ||||
| 2164 | static ccv_cnnp_model_t* _ccv_cnnp_cmul_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2165 | ||||
| 2166 | static const ccv_cnnp_model_vtab_t ccv_cnnp_cmul_isa = { | |||
| 2167 | .build = _ccv_cnnp_cmul_build, | |||
| 2168 | .copy = _ccv_cnnp_cmul_copy, | |||
| 2169 | }; | |||
| 2170 | ||||
| 2171 | ccv_cnnp_model_t* ccv_cnnp_cmul(const char* const name) | |||
| 2172 | { | |||
| 2173 | ccv_cnnp_model_cmul_t* const model_cmul = (ccv_cnnp_model_cmul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_cmul_t)); | |||
| 2174 | model_cmul->super.isa = &ccv_cnnp_cmul_isa; | |||
| 2175 | model_cmul->super.input_size = 2; | |||
| 2176 | model_cmul->super.outputs = &model_cmul->output; | |||
| 2177 | model_cmul->super.output_size = 1; | |||
| 2178 | ccv_cnnp_model_copy_name(&model_cmul->super, name); | |||
| 2179 | return (ccv_cnnp_model_t*)model_cmul; | |||
| 2180 | } | |||
| 2181 | ||||
| 2182 | static ccv_cnnp_model_t* _ccv_cnnp_cmul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2183 | { | |||
| 2184 | return ccv_cnnp_cmul(super->name); | |||
| 2185 | } | |||
| 2186 | ||||
| 2187 | // MARK - Transpose Layer | |||
| 2188 | ||||
| 2189 | typedef struct { | |||
| 2190 | ccv_cnnp_model_t super; | |||
| 2191 | ccv_nnc_tensor_symbol_t output; | |||
| 2192 | int transpose[2]; | |||
| 2193 | } ccv_cnnp_model_transpose_t; | |||
| 2194 | ||||
| 2195 | static void _ccv_cnnp_transpose_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2196 | { | |||
| 2197 | ccv_cnnp_model_transpose_t* const self = (ccv_cnnp_model_transpose_t*)super; | |||
| 2198 | PRINT(CCV_CLI_VERBOSE, "[cnnp_transpose_build] (%d, %d)\n", self->transpose[0], self->transpose[1])do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_transpose_build] (%d, %d)\n", self->transpose [0], self->transpose[1]); fflush(stdout); } } while (0); | |||
| 2199 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2199, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2200 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2200, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2201 | if (self->transpose[0] == self->transpose[1]) | |||
| 2202 | { | |||
| 2203 | outputs[0] = inputs[0]; | |||
| 2204 | return; | |||
| 2205 | } | |||
| 2206 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2207 | ccv_nnc_tensor_param_t output_params; | |||
| 2208 | const ccv_nnc_cmd_t transpose = CMD_TRANSPOSE_FORWARD(self->transpose[0], self->transpose[1])ccv_nnc_cmd(CCV_NNC_TRANSPOSE_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.transpose={.axis={self->transpose[ 0], self->transpose[1]}}}), 0); | |||
| 2209 | ccv_nnc_hint_tensor_auto(transpose, (ccv_nnc_tensor_param_t []){ | |||
| 2210 | params, | |||
| 2211 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2212 | const ccv_nnc_tensor_symbol_t transpose_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2213 | ccv_nnc_graph_exec_symbol_new(graph, transpose, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(transpose_output)(const ccv_nnc_tensor_symbol_t []){transpose_output}, (1 +1 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "transpose"); | |||
| 2214 | outputs[0] = transpose_output; | |||
| 2215 | } | |||
| 2216 | ||||
| 2217 | static ccv_cnnp_model_t* _ccv_cnnp_transpose_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2218 | ||||
| 2219 | static const ccv_cnnp_model_vtab_t ccv_cnnp_transpose_isa = { | |||
| 2220 | .build = _ccv_cnnp_transpose_build, | |||
| 2221 | .copy = _ccv_cnnp_transpose_copy, | |||
| 2222 | }; | |||
| 2223 | ||||
| 2224 | ccv_cnnp_model_t* ccv_cnnp_transpose(const int axis_a, const int axis_b, const char* const name) | |||
| 2225 | { | |||
| 2226 | ccv_cnnp_model_transpose_t* const model_transpose = (ccv_cnnp_model_transpose_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_transpose_t)); | |||
| 2227 | model_transpose->super.isa = &ccv_cnnp_transpose_isa; | |||
| 2228 | model_transpose->super.input_size = 1; | |||
| 2229 | model_transpose->super.outputs = &model_transpose->output; | |||
| 2230 | model_transpose->super.output_size = 1; | |||
| 2231 | model_transpose->transpose[0] = axis_a; | |||
| 2232 | model_transpose->transpose[1] = axis_b; | |||
| 2233 | ccv_cnnp_model_copy_name(&model_transpose->super, name); | |||
| 2234 | return (ccv_cnnp_model_t*)model_transpose; | |||
| 2235 | } | |||
| 2236 | ||||
| 2237 | static ccv_cnnp_model_t* _ccv_cnnp_transpose_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2238 | { | |||
| 2239 | const ccv_cnnp_model_transpose_t* const self = (const ccv_cnnp_model_transpose_t*)super; | |||
| 2240 | return ccv_cnnp_transpose(self->transpose[0], self->transpose[1], self->super.name); | |||
| 2241 | } | |||
| 2242 | ||||
| 2243 | // MARK - Layer Norm Layer | |||
| 2244 | ||||
| 2245 | typedef struct { | |||
| 2246 | ccv_cnnp_model_t super; | |||
| 2247 | ccv_nnc_tensor_symbol_t output; | |||
| 2248 | ccv_nnc_tensor_symbol_t bias; | |||
| 2249 | ccv_nnc_tensor_symbol_t scale; | |||
| 2250 | ccv_nnc_cmd_param_t params; | |||
| 2251 | } ccv_cnnp_model_layer_norm_t; | |||
| 2252 | ||||
| 2253 | static void _ccv_cnnp_layer_norm_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2254 | { | |||
| 2255 | PRINT(CCV_CLI_VERBOSE, "[cnnp_layer_norm_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_layer_norm_build] -\n"); fflush(stdout); } } while (0); | |||
| 2256 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2256, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2257 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2257, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2258 | ccv_cnnp_model_layer_norm_t* const self = (ccv_cnnp_model_layer_norm_t*)super; | |||
| 2259 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2260 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 2261 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 2262 | int i; | |||
| 2263 | for (i = 0; i < nd; i++) | |||
| 2264 | bias_params.dim[i] = 1; | |||
| 2265 | for (i = 0; i < self->params.lnorm.count; i++) | |||
| 2266 | bias_params.dim[self->params.lnorm.axis[i]] = params.dim[self->params.lnorm.axis[i]]; | |||
| 2267 | if (self->params.lnorm.elementwise_affine) | |||
| 2268 | { | |||
| 2269 | // Both scale and bias are shared between if this model is reused. | |||
| 2270 | if (!self->scale.graph) | |||
| 2271 | self->scale = ccv_nnc_tensor_symbol_new(graph, bias_params, "scale"); | |||
| 2272 | if (!self->bias.graph) | |||
| 2273 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 2274 | } | |||
| 2275 | const ccv_nnc_cmd_t layer_norm = ccv_nnc_cmd(CCV_NNC_LAYER_NORM_FORWARD, 0, self->params, 0); | |||
| 2276 | ccv_nnc_tensor_param_t output_params[3]; | |||
| 2277 | if (self->params.lnorm.elementwise_affine) | |||
| 2278 | ccv_nnc_hint_tensor_auto(layer_norm, (ccv_nnc_tensor_param_t []){ | |||
| 2279 | params, | |||
| 2280 | bias_params, | |||
| 2281 | bias_params, | |||
| 2282 | }, 3, ccv_nnc_no_hint, output_params, 3); | |||
| 2283 | else | |||
| 2284 | ccv_nnc_hint_tensor_auto(layer_norm, (ccv_nnc_tensor_param_t []){ | |||
| 2285 | params, | |||
| 2286 | }, 1, ccv_nnc_no_hint, output_params, 3); | |||
| 2287 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 2288 | const ccv_nnc_tensor_symbol_t saved_mean = ccv_nnc_tensor_symbol_new(graph, output_params[1], "saved_mean"); | |||
| 2289 | const ccv_nnc_tensor_symbol_t saved_inv_std = ccv_nnc_tensor_symbol_new(graph, output_params[2], "saved_inv_std"); | |||
| 2290 | if (self->params.lnorm.elementwise_affine) | |||
| 2291 | ccv_nnc_graph_exec_symbol_new(graph, layer_norm, TENSOR_SYMBOL_LIST(inputs[0], self->scale, self->bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->scale, self->bias}, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output, saved_mean, saved_inv_std)(const ccv_nnc_tensor_symbol_t []){output, saved_mean, saved_inv_std }, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 -1), "layer_norm"); | |||
| 2292 | else | |||
| 2293 | ccv_nnc_graph_exec_symbol_new(graph, layer_norm, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output, saved_mean, saved_inv_std)(const ccv_nnc_tensor_symbol_t []){output, saved_mean, saved_inv_std }, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 -1), "layer_norm"); | |||
| 2294 | outputs[0] = output; | |||
| 2295 | } | |||
| 2296 | ||||
| 2297 | static void _ccv_cnnp_layer_norm_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 2298 | { | |||
| 2299 | ccv_cnnp_model_layer_norm_t* const self = (ccv_cnnp_model_layer_norm_t*)super; | |||
| 2300 | if (self->scale.graph) | |||
| 2301 | initializer(context, CMD_SET_FORWARD(1)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={1,}}}, 0), ccv_nnc_no_hint, 0, 0, self->scale); | |||
| 2302 | if (self->bias.graph) | |||
| 2303 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, self->bias); | |||
| 2304 | } | |||
| 2305 | ||||
| 2306 | static void _ccv_cnnp_layer_norm_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 2307 | { | |||
| 2308 | ccv_cnnp_model_layer_norm_t* const self = (ccv_cnnp_model_layer_norm_t*)super; | |||
| 2309 | if (self->scale.graph) | |||
| 2310 | add_to_array(parameters, self->scale, is_trainable); | |||
| 2311 | if (self->bias.graph) | |||
| 2312 | add_to_array(parameters, self->bias, is_trainable); | |||
| 2313 | } | |||
| 2314 | ||||
| 2315 | static ccv_cnnp_model_t* _ccv_cnnp_layer_norm_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2316 | ||||
| 2317 | static const ccv_cnnp_model_vtab_t ccv_cnnp_layer_norm_isa = { | |||
| 2318 | .build = _ccv_cnnp_layer_norm_build, | |||
| 2319 | .init_states = _ccv_cnnp_layer_norm_init_states, | |||
| 2320 | .add_to_parameter = _ccv_cnnp_layer_norm_add_to_parameter, | |||
| 2321 | .copy = _ccv_cnnp_layer_norm_copy, | |||
| 2322 | }; | |||
| 2323 | ||||
| 2324 | ccv_cnnp_model_t* ccv_cnnp_layer_norm(const float epsilon, const int axis[CCV_NNC_MAX_DIM_ALLOC(12)], const int axis_count, const int elementwise_affine, const int is_trainable, const char* const name) | |||
| 2325 | { | |||
| 2326 | ccv_cnnp_model_layer_norm_t* const model_layer_norm = (ccv_cnnp_model_layer_norm_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_layer_norm_t)); | |||
| 2327 | model_layer_norm->super.isa = &ccv_cnnp_layer_norm_isa; | |||
| 2328 | model_layer_norm->super.input_size = 1; | |||
| 2329 | model_layer_norm->super.outputs = &model_layer_norm->output; | |||
| 2330 | model_layer_norm->super.output_size = 1; | |||
| 2331 | model_layer_norm->super.is_trainable = is_trainable; | |||
| 2332 | ccv_cnnp_model_copy_name(&model_layer_norm->super, name); | |||
| 2333 | model_layer_norm->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 2334 | model_layer_norm->scale.graph = 0; | |||
| 2335 | model_layer_norm->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 2336 | model_layer_norm->bias.graph = 0; | |||
| 2337 | model_layer_norm->params.lnorm.epsilon = epsilon; | |||
| 2338 | model_layer_norm->params.lnorm.count = axis_count; | |||
| 2339 | model_layer_norm->params.lnorm.elementwise_affine = elementwise_affine; | |||
| 2340 | memcpy(model_layer_norm->params.lnorm.axis, axis, sizeof(int) * axis_count); | |||
| 2341 | return (ccv_cnnp_model_t*)model_layer_norm; | |||
| 2342 | } | |||
| 2343 | ||||
| 2344 | static ccv_cnnp_model_t* _ccv_cnnp_layer_norm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2345 | { | |||
| 2346 | const ccv_cnnp_model_layer_norm_t* const self = (const ccv_cnnp_model_layer_norm_t*)super; | |||
| 2347 | return ccv_cnnp_layer_norm(self->params.lnorm.epsilon, self->params.lnorm.axis, self->params.lnorm.count, self->params.lnorm.elementwise_affine, self->super.is_trainable, self->super.name); | |||
| 2348 | } | |||
| 2349 | ||||
| 2350 | // MARK - Group Norm Layer | |||
| 2351 | ||||
| 2352 | typedef struct { | |||
| 2353 | ccv_cnnp_model_t super; | |||
| 2354 | ccv_nnc_tensor_symbol_t output; | |||
| 2355 | ccv_nnc_tensor_symbol_t bias; | |||
| 2356 | ccv_nnc_tensor_symbol_t scale; | |||
| 2357 | ccv_nnc_cmd_param_t params; | |||
| 2358 | } ccv_cnnp_model_group_norm_t; | |||
| 2359 | ||||
| 2360 | static void _ccv_cnnp_group_norm_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2361 | { | |||
| 2362 | PRINT(CCV_CLI_VERBOSE, "[cnnp_group_norm_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_group_norm_build] -\n"); fflush(stdout); } } while (0); | |||
| 2363 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2363, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2364 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2364, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2365 | ccv_cnnp_model_group_norm_t* const self = (ccv_cnnp_model_group_norm_t*)super; | |||
| 2366 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2367 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 2368 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 2369 | int i; | |||
| 2370 | for (i = 0; i < nd; i++) | |||
| 2371 | bias_params.dim[i] = 1; | |||
| 2372 | bias_params.dim[self->params.gnorm.group_axis] = params.dim[self->params.gnorm.group_axis]; | |||
| 2373 | if (self->params.gnorm.elementwise_affine) | |||
| 2374 | { | |||
| 2375 | // Both scale and bias are shared between if this model is reused. | |||
| 2376 | if (!self->scale.graph) | |||
| 2377 | self->scale = ccv_nnc_tensor_symbol_new(graph, bias_params, "scale"); | |||
| 2378 | if (!self->bias.graph) | |||
| 2379 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 2380 | } | |||
| 2381 | const ccv_nnc_cmd_t group_norm = ccv_nnc_cmd(CCV_NNC_GROUP_NORM_FORWARD, 0, self->params, 0); | |||
| 2382 | ccv_nnc_tensor_param_t output_params[3]; | |||
| 2383 | if (self->params.gnorm.elementwise_affine) | |||
| 2384 | ccv_nnc_hint_tensor_auto(group_norm, (ccv_nnc_tensor_param_t []){ | |||
| 2385 | params, | |||
| 2386 | bias_params, | |||
| 2387 | bias_params, | |||
| 2388 | }, 3, ccv_nnc_no_hint, output_params, 3); | |||
| 2389 | else | |||
| 2390 | ccv_nnc_hint_tensor_auto(group_norm, (ccv_nnc_tensor_param_t []){ | |||
| 2391 | params, | |||
| 2392 | }, 1, ccv_nnc_no_hint, output_params, 3); | |||
| 2393 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 2394 | const ccv_nnc_tensor_symbol_t saved_mean = ccv_nnc_tensor_symbol_new(graph, output_params[1], "saved_mean"); | |||
| 2395 | const ccv_nnc_tensor_symbol_t saved_inv_std = ccv_nnc_tensor_symbol_new(graph, output_params[2], "saved_inv_std"); | |||
| 2396 | if (self->params.gnorm.elementwise_affine) | |||
| 2397 | ccv_nnc_graph_exec_symbol_new(graph, group_norm, TENSOR_SYMBOL_LIST(inputs[0], self->scale, self->bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->scale, self->bias}, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output, saved_mean, saved_inv_std)(const ccv_nnc_tensor_symbol_t []){output, saved_mean, saved_inv_std }, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 -1), "group_norm"); | |||
| 2398 | else | |||
| 2399 | ccv_nnc_graph_exec_symbol_new(graph, group_norm, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output, saved_mean, saved_inv_std)(const ccv_nnc_tensor_symbol_t []){output, saved_mean, saved_inv_std }, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 -1), "group_norm"); | |||
| 2400 | outputs[0] = output; | |||
| 2401 | } | |||
| 2402 | ||||
| 2403 | static void _ccv_cnnp_group_norm_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 2404 | { | |||
| 2405 | ccv_cnnp_model_group_norm_t* const self = (ccv_cnnp_model_group_norm_t*)super; | |||
| 2406 | if (self->scale.graph) | |||
| 2407 | initializer(context, CMD_SET_FORWARD(1)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={1,}}}, 0), ccv_nnc_no_hint, 0, 0, self->scale); | |||
| 2408 | if (self->bias.graph) | |||
| 2409 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, self->bias); | |||
| 2410 | } | |||
| 2411 | ||||
| 2412 | static void _ccv_cnnp_group_norm_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 2413 | { | |||
| 2414 | ccv_cnnp_model_group_norm_t* const self = (ccv_cnnp_model_group_norm_t*)super; | |||
| 2415 | if (self->scale.graph) | |||
| 2416 | add_to_array(parameters, self->scale, is_trainable); | |||
| 2417 | if (self->bias.graph) | |||
| 2418 | add_to_array(parameters, self->bias, is_trainable); | |||
| 2419 | } | |||
| 2420 | ||||
| 2421 | static ccv_cnnp_model_t* _ccv_cnnp_group_norm_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2422 | ||||
| 2423 | static const ccv_cnnp_model_vtab_t ccv_cnnp_group_norm_isa = { | |||
| 2424 | .build = _ccv_cnnp_group_norm_build, | |||
| 2425 | .init_states = _ccv_cnnp_group_norm_init_states, | |||
| 2426 | .add_to_parameter = _ccv_cnnp_group_norm_add_to_parameter, | |||
| 2427 | .copy = _ccv_cnnp_group_norm_copy, | |||
| 2428 | }; | |||
| 2429 | ||||
| 2430 | ccv_cnnp_model_t* ccv_cnnp_group_norm(const int group_axis, const int groups, const float epsilon, const int reduce_axis[CCV_NNC_MAX_DIM_ALLOC(12)], const int axis_count, const int elementwise_affine, const int is_trainable, const char* const name) | |||
| 2431 | { | |||
| 2432 | ccv_cnnp_model_group_norm_t* const model_group_norm = (ccv_cnnp_model_group_norm_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_group_norm_t)); | |||
| 2433 | model_group_norm->super.isa = &ccv_cnnp_group_norm_isa; | |||
| 2434 | model_group_norm->super.input_size = 1; | |||
| 2435 | model_group_norm->super.outputs = &model_group_norm->output; | |||
| 2436 | model_group_norm->super.output_size = 1; | |||
| 2437 | model_group_norm->super.is_trainable = is_trainable; | |||
| 2438 | ccv_cnnp_model_copy_name(&model_group_norm->super, name); | |||
| 2439 | model_group_norm->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 2440 | model_group_norm->scale.graph = 0; | |||
| 2441 | model_group_norm->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 2442 | model_group_norm->bias.graph = 0; | |||
| 2443 | model_group_norm->params.gnorm.group_axis = group_axis; | |||
| 2444 | model_group_norm->params.gnorm.groups = groups; | |||
| 2445 | model_group_norm->params.gnorm.epsilon = epsilon; | |||
| 2446 | model_group_norm->params.gnorm.reduce_count = axis_count; | |||
| 2447 | model_group_norm->params.gnorm.elementwise_affine = elementwise_affine; | |||
| 2448 | memcpy(model_group_norm->params.gnorm.reduce_axis, reduce_axis, sizeof(int) * axis_count); | |||
| 2449 | return (ccv_cnnp_model_t*)model_group_norm; | |||
| 2450 | } | |||
| 2451 | ||||
| 2452 | static ccv_cnnp_model_t* _ccv_cnnp_group_norm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2453 | { | |||
| 2454 | const ccv_cnnp_model_group_norm_t* const self = (const ccv_cnnp_model_group_norm_t*)super; | |||
| 2455 | return ccv_cnnp_group_norm(self->params.gnorm.group_axis, self->params.gnorm.groups, self->params.gnorm.epsilon, self->params.gnorm.reduce_axis, self->params.gnorm.reduce_count, self->params.gnorm.elementwise_affine, self->super.is_trainable, self->super.name); | |||
| 2456 | } | |||
| 2457 | ||||
| 2458 | // MARK - RMSNorm Layer | |||
| 2459 | ||||
| 2460 | typedef struct { | |||
| 2461 | ccv_cnnp_model_t super; | |||
| 2462 | ccv_nnc_tensor_symbol_t output; | |||
| 2463 | ccv_nnc_tensor_symbol_t scale; | |||
| 2464 | ccv_nnc_cmd_param_t params; | |||
| 2465 | } ccv_cnnp_model_rmsnorm_t; | |||
| 2466 | ||||
| 2467 | static void _ccv_cnnp_rmsnorm_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2468 | { | |||
| 2469 | PRINT(CCV_CLI_VERBOSE, "[cnnp_rmsnorm_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_rmsnorm_build] -\n"); fflush(stdout); } } while (0); | |||
| 2470 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2470, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2471 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2471, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2472 | ccv_cnnp_model_rmsnorm_t* const self = (ccv_cnnp_model_rmsnorm_t*)super; | |||
| 2473 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2474 | ccv_nnc_tensor_param_t scale_params = params; | |||
| 2475 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 2476 | int i; | |||
| 2477 | for (i = 0; i < nd; i++) | |||
| 2478 | scale_params.dim[i] = 1; | |||
| 2479 | for (i = 0; i < self->params.rmsnorm.count; i++) | |||
| 2480 | scale_params.dim[self->params.rmsnorm.axis[i]] = params.dim[self->params.rmsnorm.axis[i]]; | |||
| 2481 | // Both scale and bias are shared between if this model is reused. | |||
| 2482 | if (!self->scale.graph) | |||
| 2483 | self->scale = ccv_nnc_tensor_symbol_new(graph, scale_params, "scale"); | |||
| 2484 | const ccv_nnc_cmd_t rmsnorm = ccv_nnc_cmd(CCV_NNC_RMSNORM_FORWARD, 0, self->params, 0); | |||
| 2485 | ccv_nnc_tensor_param_t output_params[2]; | |||
| 2486 | ccv_nnc_hint_tensor_auto(rmsnorm, (ccv_nnc_tensor_param_t []){ | |||
| 2487 | params, | |||
| 2488 | scale_params, | |||
| 2489 | }, 2, ccv_nnc_no_hint, output_params, 2); | |||
| 2490 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 2491 | const ccv_nnc_tensor_symbol_t saved_inv_std = ccv_nnc_tensor_symbol_new(graph, output_params[1], "saved_inv_std"); | |||
| 2492 | ccv_nnc_graph_exec_symbol_new(graph, rmsnorm, TENSOR_SYMBOL_LIST(inputs[0], self->scale)(const ccv_nnc_tensor_symbol_t []){inputs[0], self->scale} , (1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output, saved_inv_std)(const ccv_nnc_tensor_symbol_t []){output, saved_inv_std}, (1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 -1), "rmsnorm"); | |||
| 2493 | outputs[0] = output; | |||
| 2494 | } | |||
| 2495 | ||||
| 2496 | static void _ccv_cnnp_rmsnorm_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 2497 | { | |||
| 2498 | ccv_cnnp_model_rmsnorm_t* const self = (ccv_cnnp_model_rmsnorm_t*)super; | |||
| 2499 | if (self->scale.graph) | |||
| 2500 | initializer(context, CMD_SET_FORWARD(1)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={1,}}}, 0), ccv_nnc_no_hint, 0, 0, self->scale); | |||
| 2501 | } | |||
| 2502 | ||||
| 2503 | static void _ccv_cnnp_rmsnorm_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 2504 | { | |||
| 2505 | ccv_cnnp_model_rmsnorm_t* const self = (ccv_cnnp_model_rmsnorm_t*)super; | |||
| 2506 | if (self->scale.graph) | |||
| 2507 | add_to_array(parameters, self->scale, is_trainable); | |||
| 2508 | } | |||
| 2509 | ||||
| 2510 | static ccv_cnnp_model_t* _ccv_cnnp_rmsnorm_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2511 | ||||
| 2512 | static const ccv_cnnp_model_vtab_t ccv_cnnp_rmsnorm_isa = { | |||
| 2513 | .build = _ccv_cnnp_rmsnorm_build, | |||
| 2514 | .init_states = _ccv_cnnp_rmsnorm_init_states, | |||
| 2515 | .add_to_parameter = _ccv_cnnp_rmsnorm_add_to_parameter, | |||
| 2516 | .copy = _ccv_cnnp_rmsnorm_copy, | |||
| 2517 | }; | |||
| 2518 | ||||
| 2519 | ccv_cnnp_model_t* ccv_cnnp_rmsnorm(const float epsilon, const int axis[CCV_NNC_MAX_DIM_ALLOC(12)], const int axis_count, const int is_trainable, const char* const name) | |||
| 2520 | { | |||
| 2521 | ccv_cnnp_model_rmsnorm_t* const model_rmsnorm = (ccv_cnnp_model_rmsnorm_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_rmsnorm_t)); | |||
| 2522 | model_rmsnorm->super.isa = &ccv_cnnp_rmsnorm_isa; | |||
| 2523 | model_rmsnorm->super.input_size = 1; | |||
| 2524 | model_rmsnorm->super.outputs = &model_rmsnorm->output; | |||
| 2525 | model_rmsnorm->super.output_size = 1; | |||
| 2526 | model_rmsnorm->super.is_trainable = is_trainable; | |||
| 2527 | ccv_cnnp_model_copy_name(&model_rmsnorm->super, name); | |||
| 2528 | model_rmsnorm->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 2529 | model_rmsnorm->scale.graph = 0; | |||
| 2530 | model_rmsnorm->params.rmsnorm.epsilon = epsilon; | |||
| 2531 | model_rmsnorm->params.rmsnorm.count = axis_count; | |||
| 2532 | memcpy(model_rmsnorm->params.lnorm.axis, axis, sizeof(int) * axis_count); | |||
| 2533 | return (ccv_cnnp_model_t*)model_rmsnorm; | |||
| 2534 | } | |||
| 2535 | ||||
| 2536 | static ccv_cnnp_model_t* _ccv_cnnp_rmsnorm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2537 | { | |||
| 2538 | const ccv_cnnp_model_rmsnorm_t* const self = (const ccv_cnnp_model_rmsnorm_t*)super; | |||
| 2539 | return ccv_cnnp_rmsnorm(self->params.rmsnorm.epsilon, self->params.rmsnorm.axis, self->params.rmsnorm.count, self->super.is_trainable, self->super.name); | |||
| 2540 | } | |||
| 2541 | ||||
| 2542 | // MARK - Batched Matrix Mul Layer | |||
| 2543 | ||||
| 2544 | typedef struct { | |||
| 2545 | ccv_cnnp_model_t super; | |||
| 2546 | ccv_nnc_tensor_symbol_t output; | |||
| 2547 | int transpose_a[2]; | |||
| 2548 | int transpose_b[2]; | |||
| 2549 | int flags; | |||
| 2550 | } ccv_cnnp_model_matmul_t; | |||
| 2551 | ||||
| 2552 | static void _ccv_cnnp_matmul_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2553 | { | |||
| 2554 | PRINT(CCV_CLI_VERBOSE, "[cnnp_matmul_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_matmul_build] -\n"); fflush(stdout); } } while (0); | |||
| 2555 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 2555, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2556 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2556, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2557 | ccv_cnnp_model_matmul_t* const self = (ccv_cnnp_model_matmul_t*)super; | |||
| 2558 | ccv_nnc_tensor_param_t a_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2559 | ccv_nnc_tensor_param_t b_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 2560 | ccv_nnc_tensor_param_t output_params; | |||
| 2561 | ccv_nnc_cmd_t matmul = CMD_GEMM_FORWARD(self->transpose_a, self->transpose_b)ccv_nnc_cmd(CCV_NNC_GEMM_FORWARD, 0, ((ccv_nnc_cmd_param_t){. size={.dim={1,1,1}},.blas={.a={1,1},.transpose_a={self->transpose_a [0],self->transpose_a[1]},.transpose_b={self->transpose_b [0],self->transpose_b[1]},}}), 0); | |||
| 2562 | matmul.info.blas.flags = self->flags; | |||
| 2563 | ccv_nnc_hint_tensor_auto(matmul, (ccv_nnc_tensor_param_t []){ | |||
| 2564 | a_params, | |||
| 2565 | b_params, | |||
| 2566 | }, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2567 | const ccv_nnc_tensor_symbol_t matmul_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2568 | ccv_nnc_graph_exec_symbol_new(graph, matmul, inputs, input_size, TENSOR_SYMBOL_LIST(matmul_output)(const ccv_nnc_tensor_symbol_t []){matmul_output}, (1 +1 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1 ), "matmul"); | |||
| 2569 | outputs[0] = matmul_output; | |||
| 2570 | } | |||
| 2571 | ||||
| 2572 | static ccv_cnnp_model_t* _ccv_cnnp_matmul_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2573 | ||||
| 2574 | static const ccv_cnnp_model_vtab_t ccv_cnnp_matmul_isa = { | |||
| 2575 | .build = _ccv_cnnp_matmul_build, | |||
| 2576 | .copy = _ccv_cnnp_matmul_copy, | |||
| 2577 | }; | |||
| 2578 | ||||
| 2579 | ccv_cnnp_model_t* ccv_cnnp_matmul(const int transpose_a[2], const int transpose_b[2], const int flags, const char* const name) | |||
| 2580 | { | |||
| 2581 | ccv_cnnp_model_matmul_t* const model_matmul = (ccv_cnnp_model_matmul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_matmul_t)); | |||
| 2582 | model_matmul->super.isa = &ccv_cnnp_matmul_isa; | |||
| 2583 | model_matmul->super.input_size = 2; | |||
| 2584 | model_matmul->super.outputs = &model_matmul->output; | |||
| 2585 | model_matmul->super.output_size = 1; | |||
| 2586 | model_matmul->transpose_a[0] = transpose_a[0]; | |||
| 2587 | model_matmul->transpose_a[1] = transpose_a[1]; | |||
| 2588 | model_matmul->transpose_b[0] = transpose_b[0]; | |||
| 2589 | model_matmul->transpose_b[1] = transpose_b[1]; | |||
| 2590 | model_matmul->flags = flags; | |||
| 2591 | ccv_cnnp_model_copy_name(&model_matmul->super, name); | |||
| 2592 | return (ccv_cnnp_model_t*)model_matmul; | |||
| 2593 | } | |||
| 2594 | ||||
| 2595 | static ccv_cnnp_model_t* _ccv_cnnp_matmul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2596 | { | |||
| 2597 | const ccv_cnnp_model_matmul_t* const self = (const ccv_cnnp_model_matmul_t*)super; | |||
| 2598 | return ccv_cnnp_matmul(self->transpose_a, self->transpose_b, self->flags, self->super.name); | |||
| 2599 | } | |||
| 2600 | ||||
| 2601 | // MARK - Dropout Layer | |||
| 2602 | ||||
| 2603 | typedef struct { | |||
| 2604 | ccv_cnnp_model_t super; | |||
| 2605 | ccv_nnc_tensor_symbol_t output; | |||
| 2606 | ccv_nnc_graph_exec_symbol_t dropout; | |||
| 2607 | float p; | |||
| 2608 | int entirety; | |||
| 2609 | } ccv_cnnp_model_dropout_t; | |||
| 2610 | ||||
| 2611 | static void _ccv_cnnp_dropout_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2612 | { | |||
| 2613 | PRINT(CCV_CLI_VERBOSE, "[cnnp_dropout_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_dropout_build] -\n"); fflush(stdout); } } while (0); | |||
| 2614 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2614, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2615 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2615, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2616 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2617 | ccv_nnc_tensor_param_t output_params[2]; | |||
| 2618 | ccv_cnnp_model_dropout_t* const self = (ccv_cnnp_model_dropout_t*)super; | |||
| 2619 | const ccv_nnc_cmd_t dropout = CMD_DROPOUT_FORWARD(self->p, self->entirety)ccv_nnc_cmd(CCV_NNC_DROPOUT_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.dropout={.p=self->p,.entirety=self ->entirety}}), 0); | |||
| 2620 | ccv_nnc_hint_tensor_auto(dropout, (ccv_nnc_tensor_param_t []){ | |||
| 2621 | params, | |||
| 2622 | }, 1, ccv_nnc_no_hint, output_params, 2); | |||
| 2623 | const ccv_nnc_tensor_symbol_t dropout_output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 2624 | const ccv_nnc_tensor_symbol_t mask = ccv_nnc_tensor_symbol_new(graph, output_params[1], "mask"); | |||
| 2625 | self->dropout = ccv_nnc_graph_exec_symbol_new(graph, dropout, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(dropout_output, mask)(const ccv_nnc_tensor_symbol_t []){dropout_output, mask}, (1 + 1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "dropout"); | |||
| 2626 | outputs[0] = dropout_output; | |||
| 2627 | } | |||
| 2628 | ||||
| 2629 | static void _ccv_cnnp_dropout_set_is_test(ccv_cnnp_model_t* const super, const int is_test, const ccv_cnnp_cmd_updater_f updater, void* const context) | |||
| 2630 | { | |||
| 2631 | ccv_cnnp_model_dropout_t* const self = (ccv_cnnp_model_dropout_t*)super; | |||
| 2632 | if (self->dropout.graph) | |||
| 2633 | { | |||
| 2634 | if (is_test) | |||
| 2635 | // During test, the dropout is not applied. Data transfer is perfect because if these are the same tensor, it will skip. | |||
| 2636 | updater(context, self->dropout, CMD_DATA_TRANSFER_FORWARD()ccv_nnc_cmd(CCV_NNC_DATA_TRANSFER_FORWARD, 0, ccv_nnc_cmd_auto , 0), ccv_nnc_no_hint); | |||
| 2637 | else | |||
| 2638 | updater(context, self->dropout, CMD_DROPOUT_FORWARD(self->p, self->entirety)ccv_nnc_cmd(CCV_NNC_DROPOUT_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.dropout={.p=self->p,.entirety=self ->entirety}}), 0), ccv_nnc_no_hint); | |||
| 2639 | } | |||
| 2640 | } | |||
| 2641 | ||||
| 2642 | static ccv_cnnp_model_t* _ccv_cnnp_dropout_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2643 | ||||
| 2644 | static const ccv_cnnp_model_vtab_t ccv_cnnp_dropout_isa = { | |||
| 2645 | .build = _ccv_cnnp_dropout_build, | |||
| 2646 | .set_is_test = _ccv_cnnp_dropout_set_is_test, | |||
| 2647 | .copy = _ccv_cnnp_dropout_copy, | |||
| 2648 | }; | |||
| 2649 | ||||
| 2650 | ccv_cnnp_model_t* ccv_cnnp_dropout(const float p, const int entirety, const char* const name) | |||
| 2651 | { | |||
| 2652 | ccv_cnnp_model_dropout_t* const model_dropout = (ccv_cnnp_model_dropout_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_dropout_t)); | |||
| 2653 | model_dropout->super.isa = &ccv_cnnp_dropout_isa; | |||
| 2654 | model_dropout->super.input_size = 1; | |||
| 2655 | model_dropout->super.outputs = &model_dropout->output; | |||
| 2656 | model_dropout->super.output_size = 1; | |||
| 2657 | model_dropout->p = p; | |||
| 2658 | model_dropout->entirety = entirety; | |||
| 2659 | ccv_cnnp_model_copy_name(&model_dropout->super, name); | |||
| 2660 | return (ccv_cnnp_model_t*)model_dropout; | |||
| 2661 | } | |||
| 2662 | ||||
| 2663 | static ccv_cnnp_model_t* _ccv_cnnp_dropout_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2664 | { | |||
| 2665 | const ccv_cnnp_model_dropout_t* const self = (const ccv_cnnp_model_dropout_t*)super; | |||
| 2666 | return ccv_cnnp_dropout(self->p, self->entirety, self->super.name); | |||
| 2667 | } | |||
| 2668 | ||||
| 2669 | // MARK - Masked Fill Layer | |||
| 2670 | ||||
| 2671 | typedef struct { | |||
| 2672 | ccv_cnnp_model_t super; | |||
| 2673 | ccv_nnc_tensor_symbol_t output; | |||
| 2674 | float eq; | |||
| 2675 | float fill; | |||
| 2676 | } ccv_cnnp_model_masked_fill_t; | |||
| 2677 | ||||
| 2678 | static void _ccv_cnnp_masked_fill_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2679 | { | |||
| 2680 | PRINT(CCV_CLI_VERBOSE, "[cnnp_masked_fill_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_masked_fill_build] -\n"); fflush(stdout); } } while (0); | |||
| 2681 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 2681, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2682 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2682, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2683 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2684 | ccv_cnnp_model_masked_fill_t* const self = (ccv_cnnp_model_masked_fill_t*)super; | |||
| 2685 | const ccv_nnc_tensor_symbol_t masked_fill_output = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 2686 | ccv_nnc_graph_exec_symbol_new(graph, CMD_MASKED_FILL_FORWARD(self->eq, self->fill)ccv_nnc_cmd(CCV_NNC_MASKED_FILL_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={self->eq, self->fill} }}, 0), TENSOR_SYMBOL_LIST(inputs[0], inputs[1])(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1]}, (1 + 1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(masked_fill_output)(const ccv_nnc_tensor_symbol_t []){masked_fill_output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 -1), "masked_fill"); | |||
| 2687 | outputs[0] = masked_fill_output; | |||
| 2688 | } | |||
| 2689 | ||||
| 2690 | static ccv_cnnp_model_t* _ccv_cnnp_masked_fill_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2691 | ||||
| 2692 | static const ccv_cnnp_model_vtab_t ccv_cnnp_masked_fill_isa = { | |||
| 2693 | .build = _ccv_cnnp_masked_fill_build, | |||
| 2694 | .copy = _ccv_cnnp_masked_fill_copy, | |||
| 2695 | }; | |||
| 2696 | ||||
| 2697 | ccv_cnnp_model_t* ccv_cnnp_masked_fill(const float eq, const float fill, const char* const name) | |||
| 2698 | { | |||
| 2699 | ccv_cnnp_model_masked_fill_t* const model_masked_fill = (ccv_cnnp_model_masked_fill_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_masked_fill_t)); | |||
| 2700 | model_masked_fill->super.isa = &ccv_cnnp_masked_fill_isa; | |||
| 2701 | model_masked_fill->super.input_size = 2; | |||
| 2702 | model_masked_fill->super.outputs = &model_masked_fill->output; | |||
| 2703 | model_masked_fill->super.output_size = 1; | |||
| 2704 | model_masked_fill->eq = eq; | |||
| 2705 | model_masked_fill->fill = fill; | |||
| 2706 | ccv_cnnp_model_copy_name(&model_masked_fill->super, name); | |||
| 2707 | return (ccv_cnnp_model_t*)model_masked_fill; | |||
| 2708 | } | |||
| 2709 | ||||
| 2710 | static ccv_cnnp_model_t* _ccv_cnnp_masked_fill_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2711 | { | |||
| 2712 | const ccv_cnnp_model_masked_fill_t* const self = (const ccv_cnnp_model_masked_fill_t*)super; | |||
| 2713 | return ccv_cnnp_masked_fill(self->eq, self->fill, self->super.name); | |||
| 2714 | } | |||
| 2715 | ||||
| 2716 | // MARK - Index Select Layer | |||
| 2717 | ||||
| 2718 | typedef struct { | |||
| 2719 | ccv_cnnp_model_t super; | |||
| 2720 | ccv_nnc_tensor_symbol_t output; | |||
| 2721 | } ccv_cnnp_model_index_select_t; | |||
| 2722 | ||||
| 2723 | static void _ccv_cnnp_index_select_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2724 | { | |||
| 2725 | PRINT(CCV_CLI_VERBOSE, "[cnnp_index_select_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_index_select_build] -\n"); fflush(stdout); } } while (0); | |||
| 2726 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 2726, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2727 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2727, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2728 | const ccv_nnc_tensor_param_t vocab_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2729 | const ccv_nnc_tensor_param_t index_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 2730 | ccv_nnc_tensor_param_t output_params; | |||
| 2731 | const ccv_nnc_cmd_t index_select = CMD_INDEX_SELECT_FORWARD()ccv_nnc_cmd(CCV_NNC_INDEX_SELECT_FORWARD, 0, ccv_nnc_cmd_auto , 0); | |||
| 2732 | ccv_nnc_hint_tensor_auto(index_select, (ccv_nnc_tensor_param_t []){ | |||
| 2733 | vocab_params, | |||
| 2734 | index_params, | |||
| 2735 | }, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2736 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2737 | ccv_nnc_graph_exec_symbol_new(graph, index_select, TENSOR_SYMBOL_LIST(inputs[0], inputs[1])(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1]}, (1 + 1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "index_select"); | |||
| 2738 | outputs[0] = output; | |||
| 2739 | } | |||
| 2740 | ||||
| 2741 | static ccv_cnnp_model_t* _ccv_cnnp_index_select_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2742 | ||||
| 2743 | static const ccv_cnnp_model_vtab_t ccv_cnnp_index_select_isa = { | |||
| 2744 | .build = _ccv_cnnp_index_select_build, | |||
| 2745 | .copy = _ccv_cnnp_index_select_copy, | |||
| 2746 | }; | |||
| 2747 | ||||
| 2748 | ccv_cnnp_model_t* ccv_cnnp_index_select(const char* const name) | |||
| 2749 | { | |||
| 2750 | ccv_cnnp_model_index_select_t* const model_index_select = (ccv_cnnp_model_index_select_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_index_select_t)); | |||
| 2751 | model_index_select->super.isa = &ccv_cnnp_index_select_isa; | |||
| 2752 | model_index_select->super.input_size = 2; | |||
| 2753 | model_index_select->super.outputs = &model_index_select->output; | |||
| 2754 | model_index_select->super.output_size = 1; | |||
| 2755 | ccv_cnnp_model_copy_name(&model_index_select->super, name); | |||
| 2756 | return (ccv_cnnp_model_t*)model_index_select; | |||
| 2757 | } | |||
| 2758 | ||||
| 2759 | static ccv_cnnp_model_t* _ccv_cnnp_index_select_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2760 | { | |||
| 2761 | ccv_cnnp_model_index_select_t* const self = (ccv_cnnp_model_index_select_t*)super; | |||
| 2762 | return ccv_cnnp_index_select(self->super.name); | |||
| 2763 | } | |||
| 2764 | ||||
| 2765 | // MARK - Embedding Layer | |||
| 2766 | ||||
| 2767 | typedef struct { | |||
| 2768 | ccv_cnnp_model_t super; | |||
| 2769 | ccv_nnc_tensor_symbol_t output; | |||
| 2770 | ccv_nnc_tensor_symbol_t vocab; | |||
| 2771 | int datatype; | |||
| 2772 | int vocab_size; | |||
| 2773 | int embed_size; | |||
| 2774 | } ccv_cnnp_model_embedding_t; | |||
| 2775 | ||||
| 2776 | static void _ccv_cnnp_embedding_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2777 | { | |||
| 2778 | ccv_cnnp_model_embedding_t* const self = (ccv_cnnp_model_embedding_t*)super; | |||
| 2779 | PRINT(CCV_CLI_VERBOSE, "[cnnp_embedding_build] vocab_size: %d, embed_size: %d\n", self->vocab_size, self->embed_size)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_embedding_build] vocab_size: %d, embed_size: %d\n" , self->vocab_size, self->embed_size); fflush(stdout); } } while (0); | |||
| 2780 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2780, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2781 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2781, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2782 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2783 | ccv_nnc_tensor_param_t vocab_params = params; | |||
| 2784 | memset(vocab_params.dim, 0, sizeof(vocab_params.dim)); | |||
| 2785 | vocab_params.datatype = self->datatype; | |||
| 2786 | vocab_params.dim[0] = self->vocab_size; | |||
| 2787 | vocab_params.dim[1] = self->embed_size; | |||
| 2788 | if (!self->vocab.graph) | |||
| 2789 | self->vocab = ccv_nnc_tensor_symbol_new(graph, vocab_params, "vocab"); | |||
| 2790 | assert(self->vocab.graph == graph)((void) sizeof ((self->vocab.graph == graph) ? 1 : 0), __extension__ ({ if (self->vocab.graph == graph) ; else __assert_fail ( "self->vocab.graph == graph", "ccv_cnnp_model_addons.c", 2790 , __extension__ __PRETTY_FUNCTION__); })); | |||
| 2791 | ccv_nnc_tensor_param_t output_params; | |||
| 2792 | const ccv_nnc_cmd_t embedding = CMD_INDEX_SELECT_FORWARD()ccv_nnc_cmd(CCV_NNC_INDEX_SELECT_FORWARD, 0, ccv_nnc_cmd_auto , 0); | |||
| 2793 | ccv_nnc_hint_tensor_auto(embedding, (ccv_nnc_tensor_param_t []){ | |||
| 2794 | vocab_params, | |||
| 2795 | params, | |||
| 2796 | }, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2797 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2798 | ccv_nnc_graph_exec_symbol_new(graph, embedding, TENSOR_SYMBOL_LIST(self->vocab, inputs[0])(const ccv_nnc_tensor_symbol_t []){self->vocab, inputs[0]} , (1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "embedding"); | |||
| 2799 | outputs[0] = output; | |||
| 2800 | } | |||
| 2801 | ||||
| 2802 | static void _ccv_cnnp_embedding_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 2803 | { | |||
| 2804 | ccv_cnnp_model_embedding_t* const self = (ccv_cnnp_model_embedding_t*)super; | |||
| 2805 | const float std = sqrtf(2) / sqrtf(self->vocab_size + self->embed_size); | |||
| 2806 | const float bound = sqrtf(3) * std; | |||
| 2807 | initializer(context, CMD_RANDOM_UNIFORM_FORWARD(-bound, bound)ccv_nnc_cmd(CCV_NNC_RANDOM_UNIFORM_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={-bound, bound}}}, 0), ccv_nnc_no_hint, 0, 0, self->vocab); | |||
| 2808 | } | |||
| 2809 | ||||
| 2810 | static void _ccv_cnnp_embedding_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 2811 | { | |||
| 2812 | ccv_cnnp_model_embedding_t* const self = (ccv_cnnp_model_embedding_t*)super; | |||
| 2813 | add_to_array(parameters, self->vocab, is_trainable); | |||
| 2814 | } | |||
| 2815 | ||||
| 2816 | static ccv_cnnp_model_t* _ccv_cnnp_embedding_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2817 | ||||
| 2818 | static const ccv_cnnp_model_vtab_t ccv_cnnp_embedding_isa = { | |||
| 2819 | .build = _ccv_cnnp_embedding_build, | |||
| 2820 | .init_states = _ccv_cnnp_embedding_init_states, | |||
| 2821 | .add_to_parameter = _ccv_cnnp_embedding_add_to_parameter, | |||
| 2822 | .copy = _ccv_cnnp_embedding_copy, | |||
| 2823 | }; | |||
| 2824 | ||||
| 2825 | ccv_cnnp_model_t* ccv_cnnp_embedding(const int datatype, const int vocab_size, const int embed_size, const int is_trainable, const char* const name) | |||
| 2826 | { | |||
| 2827 | ccv_cnnp_model_embedding_t* const model_embedding = (ccv_cnnp_model_embedding_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_embedding_t)); | |||
| 2828 | model_embedding->super.isa = &ccv_cnnp_embedding_isa; | |||
| 2829 | model_embedding->super.input_size = 1; | |||
| 2830 | model_embedding->super.outputs = &model_embedding->output; | |||
| 2831 | model_embedding->super.output_size = 1; | |||
| 2832 | model_embedding->super.is_trainable = is_trainable; | |||
| 2833 | ccv_cnnp_model_copy_name(&model_embedding->super, name); | |||
| 2834 | model_embedding->vocab.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 2835 | model_embedding->vocab.graph = 0; | |||
| 2836 | assert(datatype == CCV_32F || datatype == CCV_16F)((void) sizeof ((datatype == CCV_32F || datatype == CCV_16F) ? 1 : 0), __extension__ ({ if (datatype == CCV_32F || datatype == CCV_16F) ; else __assert_fail ("datatype == CCV_32F || datatype == CCV_16F" , "ccv_cnnp_model_addons.c", 2836, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2837 | model_embedding->datatype = datatype; | |||
| 2838 | assert(vocab_size > 0)((void) sizeof ((vocab_size > 0) ? 1 : 0), __extension__ ( { if (vocab_size > 0) ; else __assert_fail ("vocab_size > 0" , "ccv_cnnp_model_addons.c", 2838, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2839 | model_embedding->vocab_size = vocab_size; | |||
| 2840 | assert(embed_size > 0)((void) sizeof ((embed_size > 0) ? 1 : 0), __extension__ ( { if (embed_size > 0) ; else __assert_fail ("embed_size > 0" , "ccv_cnnp_model_addons.c", 2840, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2841 | model_embedding->embed_size = embed_size; | |||
| 2842 | return (ccv_cnnp_model_t*)model_embedding; | |||
| 2843 | } | |||
| 2844 | ||||
| 2845 | static ccv_cnnp_model_t* _ccv_cnnp_embedding_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2846 | { | |||
| 2847 | ccv_cnnp_model_embedding_t* const self = (ccv_cnnp_model_embedding_t*)super; | |||
| 2848 | return ccv_cnnp_embedding(self->datatype, self->vocab_size, self->embed_size, self->super.is_trainable, self->super.name); | |||
| 2849 | } | |||
| 2850 | ||||
| 2851 | // MARK - Pool Layers | |||
| 2852 | ||||
| 2853 | typedef struct { | |||
| 2854 | ccv_cnnp_model_t super; | |||
| 2855 | ccv_nnc_tensor_symbol_t output; | |||
| 2856 | int type; | |||
| 2857 | float width_scale; | |||
| 2858 | float height_scale; | |||
| 2859 | int align_corners; | |||
| 2860 | } ccv_cnnp_model_upsample_t; | |||
| 2861 | ||||
| 2862 | static void _ccv_cnnp_upsample_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2863 | { | |||
| 2864 | PRINT(CCV_CLI_VERBOSE, "[cnnp_upsample_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_upsample_build] -\n"); fflush(stdout); } } while (0); | |||
| 2865 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2865, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2866 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2866, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2867 | ccv_cnnp_model_upsample_t* const self = (ccv_cnnp_model_upsample_t*)super; | |||
| 2868 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2869 | ccv_nnc_cmd_t cmd = CMD_UPSAMPLE_FORWARD(self->type, self->width_scale, self->height_scale, self->align_corners)ccv_nnc_cmd(CCV_NNC_UPSAMPLE_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.upsample={.type=self->type,.width_scale =self->width_scale,.height_scale=self->height_scale,.align_corners =self->align_corners}}), 0); | |||
| 2870 | ccv_nnc_tensor_param_t output_params; | |||
| 2871 | ccv_nnc_hint_tensor_auto(cmd, ¶ms, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2872 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2873 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0])(const ccv_nnc_tensor_symbol_t []){inputs[0]}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output)(const ccv_nnc_tensor_symbol_t []){output}, (1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "upsample"); | |||
| 2874 | outputs[0] = output; | |||
| 2875 | } | |||
| 2876 | ||||
| 2877 | static ccv_cnnp_model_t* _ccv_cnnp_upsample_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2878 | ||||
| 2879 | static const ccv_cnnp_model_vtab_t ccv_cnnp_upsample_isa = { | |||
| 2880 | .build = _ccv_cnnp_upsample_build, | |||
| 2881 | .copy = _ccv_cnnp_upsample_copy, | |||
| 2882 | }; | |||
| 2883 | ||||
| 2884 | ccv_cnnp_model_t* ccv_cnnp_upsample(const int type, const float width_scale, const float height_scale, const int align_corners, const char* const name) | |||
| 2885 | { | |||
| 2886 | ccv_cnnp_model_upsample_t* const model_upsample = (ccv_cnnp_model_upsample_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_upsample_t)); | |||
| 2887 | model_upsample->super.isa = &ccv_cnnp_upsample_isa; | |||
| 2888 | model_upsample->super.input_size = 1; | |||
| 2889 | model_upsample->super.outputs = &model_upsample->output; | |||
| 2890 | model_upsample->super.output_size = 1; | |||
| 2891 | ccv_cnnp_model_copy_name(&model_upsample->super, name); | |||
| 2892 | assert(type == CCV_NNC_UPSAMPLE_NEAREST || type == CCV_NNC_UPSAMPLE_BILINEAR)((void) sizeof ((type == CCV_NNC_UPSAMPLE_NEAREST || type == CCV_NNC_UPSAMPLE_BILINEAR ) ? 1 : 0), __extension__ ({ if (type == CCV_NNC_UPSAMPLE_NEAREST || type == CCV_NNC_UPSAMPLE_BILINEAR) ; else __assert_fail ( "type == CCV_NNC_UPSAMPLE_NEAREST || type == CCV_NNC_UPSAMPLE_BILINEAR" , "ccv_cnnp_model_addons.c", 2892, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2893 | model_upsample->type = type; | |||
| 2894 | model_upsample->width_scale = width_scale; | |||
| 2895 | model_upsample->height_scale = height_scale; | |||
| 2896 | model_upsample->align_corners = align_corners; | |||
| 2897 | return (ccv_cnnp_model_t*)model_upsample; | |||
| 2898 | } | |||
| 2899 | ||||
| 2900 | static ccv_cnnp_model_t* _ccv_cnnp_upsample_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2901 | { | |||
| 2902 | const ccv_cnnp_model_upsample_t* const self = (const ccv_cnnp_model_upsample_t*)super; | |||
| 2903 | return ccv_cnnp_upsample(self->type, self->width_scale, self->height_scale, self->align_corners, self->super.name); | |||
| 2904 | } | |||
| 2905 | ||||
| 2906 | // MARK - Reduce Sum Layer | |||
| 2907 | ||||
| 2908 | typedef struct { | |||
| 2909 | ccv_cnnp_model_t super; | |||
| 2910 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 2911 | int count; | |||
| 2912 | ccv_nnc_tensor_symbol_t output; | |||
| 2913 | } ccv_cnnp_model_reduce_sum_t; | |||
| 2914 | ||||
| 2915 | static void _ccv_cnnp_reduce_sum_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2916 | { | |||
| 2917 | PRINT(CCV_CLI_VERBOSE, "[cnnp_reduce_sum_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_reduce_sum_build] -\n"); fflush(stdout); } } while (0); | |||
| 2918 | const ccv_cnnp_model_reduce_sum_t* const self = (const ccv_cnnp_model_reduce_sum_t*)super; | |||
| 2919 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2919, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2920 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2920, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2921 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2922 | ccv_nnc_tensor_param_t output_params; | |||
| 2923 | ccv_nnc_cmd_t reduce_sum = CMD_REDUCE_SUM_FORWARD()ccv_nnc_cmd(CCV_NNC_REDUCE_SUM_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}} ), 0); | |||
| 2924 | int i; | |||
| 2925 | for (i = 0; i < self->count; i++) | |||
| 2926 | reduce_sum.info.reduce.axis[i] = self->axis[i]; | |||
| 2927 | reduce_sum.info.reduce.count = self->count; | |||
| 2928 | ccv_nnc_hint_tensor_auto(reduce_sum, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2929 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2930 | ccv_nnc_graph_exec_symbol_new(graph, reduce_sum, inputs, input_size, outputs, output_size, "reduce_sum"); | |||
| 2931 | } | |||
| 2932 | ||||
| 2933 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_sum_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2934 | ||||
| 2935 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_sum_isa = { | |||
| 2936 | .build = _ccv_cnnp_reduce_sum_build, | |||
| 2937 | .copy = _ccv_cnnp_reduce_sum_copy, | |||
| 2938 | }; | |||
| 2939 | ||||
| 2940 | ccv_cnnp_model_t* ccv_cnnp_reduce_sum(const int* const axis, const int axis_count, const char* const name) | |||
| 2941 | { | |||
| 2942 | ccv_cnnp_model_reduce_sum_t* const model_reduce_sum = (ccv_cnnp_model_reduce_sum_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_reduce_sum_t)); | |||
| 2943 | model_reduce_sum->super.isa = &ccv_cnnp_reduce_sum_isa; | |||
| 2944 | model_reduce_sum->super.input_size = 1; | |||
| 2945 | model_reduce_sum->super.outputs = &model_reduce_sum->output; | |||
| 2946 | model_reduce_sum->super.output_size = 1; | |||
| 2947 | ccv_cnnp_model_copy_name(&model_reduce_sum->super, name); | |||
| 2948 | assert(axis_count <= CCV_NNC_MAX_DIM_ALLOC)((void) sizeof ((axis_count <= (12)) ? 1 : 0), __extension__ ({ if (axis_count <= (12)) ; else __assert_fail ("axis_count <= CCV_NNC_MAX_DIM_ALLOC" , "ccv_cnnp_model_addons.c", 2948, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2949 | int i; | |||
| 2950 | for (i = 0; i < axis_count; i++) | |||
| 2951 | model_reduce_sum->axis[i] = axis[i]; | |||
| 2952 | model_reduce_sum->count = axis_count; | |||
| 2953 | return (ccv_cnnp_model_t*)model_reduce_sum; | |||
| 2954 | } | |||
| 2955 | ||||
| 2956 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_sum_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2957 | { | |||
| 2958 | const ccv_cnnp_model_reduce_sum_t* const self = (const ccv_cnnp_model_reduce_sum_t*)super; | |||
| 2959 | return ccv_cnnp_reduce_sum(self->axis, self->count, self->super.name); | |||
| 2960 | } | |||
| 2961 | ||||
| 2962 | // MARK - Reduce Mean Layer | |||
| 2963 | ||||
| 2964 | typedef struct { | |||
| 2965 | ccv_cnnp_model_t super; | |||
| 2966 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 2967 | int count; | |||
| 2968 | ccv_nnc_tensor_symbol_t output; | |||
| 2969 | } ccv_cnnp_model_reduce_mean_t; | |||
| 2970 | ||||
| 2971 | static void _ccv_cnnp_reduce_mean_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 2972 | { | |||
| 2973 | PRINT(CCV_CLI_VERBOSE, "[cnnp_reduce_mean_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_reduce_mean_build] -\n"); fflush(stdout); } } while (0); | |||
| 2974 | const ccv_cnnp_model_reduce_mean_t* const self = (const ccv_cnnp_model_reduce_mean_t*)super; | |||
| 2975 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 2975, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2976 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 2976, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2977 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2978 | ccv_nnc_tensor_param_t output_params; | |||
| 2979 | ccv_nnc_cmd_t reduce_mean = CMD_REDUCE_MEAN_FORWARD()ccv_nnc_cmd(CCV_NNC_REDUCE_MEAN_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}} ), 0); | |||
| 2980 | int i; | |||
| 2981 | for (i = 0; i < self->count; i++) | |||
| 2982 | reduce_mean.info.reduce.axis[i] = self->axis[i]; | |||
| 2983 | reduce_mean.info.reduce.count = self->count; | |||
| 2984 | ccv_nnc_hint_tensor_auto(reduce_mean, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2985 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2986 | ccv_nnc_graph_exec_symbol_new(graph, reduce_mean, inputs, input_size, outputs, output_size, "reduce_mean"); | |||
| 2987 | } | |||
| 2988 | ||||
| 2989 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_mean_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2990 | ||||
| 2991 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_mean_isa = { | |||
| 2992 | .build = _ccv_cnnp_reduce_mean_build, | |||
| 2993 | .copy = _ccv_cnnp_reduce_mean_copy, | |||
| 2994 | }; | |||
| 2995 | ||||
| 2996 | ccv_cnnp_model_t* ccv_cnnp_reduce_mean(const int* const axis, const int axis_count, const char* const name) | |||
| 2997 | { | |||
| 2998 | ccv_cnnp_model_reduce_mean_t* const model_reduce_mean = (ccv_cnnp_model_reduce_mean_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_reduce_mean_t)); | |||
| 2999 | model_reduce_mean->super.isa = &ccv_cnnp_reduce_mean_isa; | |||
| 3000 | model_reduce_mean->super.input_size = 1; | |||
| 3001 | model_reduce_mean->super.outputs = &model_reduce_mean->output; | |||
| 3002 | model_reduce_mean->super.output_size = 1; | |||
| 3003 | ccv_cnnp_model_copy_name(&model_reduce_mean->super, name); | |||
| 3004 | assert(axis_count <= CCV_NNC_MAX_DIM_ALLOC)((void) sizeof ((axis_count <= (12)) ? 1 : 0), __extension__ ({ if (axis_count <= (12)) ; else __assert_fail ("axis_count <= CCV_NNC_MAX_DIM_ALLOC" , "ccv_cnnp_model_addons.c", 3004, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3005 | int i; | |||
| 3006 | for (i = 0; i < axis_count; i++) | |||
| 3007 | model_reduce_mean->axis[i] = axis[i]; | |||
| 3008 | model_reduce_mean->count = axis_count; | |||
| 3009 | return (ccv_cnnp_model_t*)model_reduce_mean; | |||
| 3010 | } | |||
| 3011 | ||||
| 3012 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_mean_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3013 | { | |||
| 3014 | const ccv_cnnp_model_reduce_mean_t* const self = (const ccv_cnnp_model_reduce_mean_t*)super; | |||
| 3015 | return ccv_cnnp_reduce_mean(self->axis, self->count, self->super.name); | |||
| 3016 | } | |||
| 3017 | ||||
| 3018 | // MARK - Reduce Max Layer | |||
| 3019 | ||||
| 3020 | typedef struct { | |||
| 3021 | ccv_cnnp_model_t super; | |||
| 3022 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 3023 | int count; | |||
| 3024 | ccv_nnc_tensor_symbol_t output; | |||
| 3025 | } ccv_cnnp_model_reduce_max_t; | |||
| 3026 | ||||
| 3027 | static void _ccv_cnnp_reduce_max_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3028 | { | |||
| 3029 | PRINT(CCV_CLI_VERBOSE, "[cnnp_reduce_max_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_reduce_max_build] -\n"); fflush(stdout); } } while (0); | |||
| 3030 | const ccv_cnnp_model_reduce_max_t* const self = (const ccv_cnnp_model_reduce_max_t*)super; | |||
| 3031 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 3031, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3032 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3032, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3033 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3034 | ccv_nnc_tensor_param_t output_params; | |||
| 3035 | ccv_nnc_cmd_t reduce_max = CMD_REDUCE_MAX_FORWARD()ccv_nnc_cmd(CCV_NNC_REDUCE_MAX_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}} ), 0); | |||
| 3036 | int i; | |||
| 3037 | for (i = 0; i < self->count; i++) | |||
| 3038 | reduce_max.info.reduce.axis[i] = self->axis[i]; | |||
| 3039 | reduce_max.info.reduce.count = self->count; | |||
| 3040 | ccv_nnc_hint_tensor_auto(reduce_max, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3041 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3042 | ccv_nnc_graph_exec_symbol_new(graph, reduce_max, inputs, input_size, outputs, output_size, "reduce_max"); | |||
| 3043 | } | |||
| 3044 | ||||
| 3045 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_max_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3046 | ||||
| 3047 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_max_isa = { | |||
| 3048 | .build = _ccv_cnnp_reduce_max_build, | |||
| 3049 | .copy = _ccv_cnnp_reduce_max_copy, | |||
| 3050 | }; | |||
| 3051 | ||||
| 3052 | ccv_cnnp_model_t* ccv_cnnp_reduce_max(const int* const axis, const int axis_count, const char* const name) | |||
| 3053 | { | |||
| 3054 | ccv_cnnp_model_reduce_max_t* const model_reduce_max = (ccv_cnnp_model_reduce_max_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_reduce_max_t)); | |||
| 3055 | model_reduce_max->super.isa = &ccv_cnnp_reduce_max_isa; | |||
| 3056 | model_reduce_max->super.input_size = 1; | |||
| 3057 | model_reduce_max->super.outputs = &model_reduce_max->output; | |||
| 3058 | model_reduce_max->super.output_size = 1; | |||
| 3059 | ccv_cnnp_model_copy_name(&model_reduce_max->super, name); | |||
| 3060 | assert(axis_count <= CCV_NNC_MAX_DIM_ALLOC)((void) sizeof ((axis_count <= (12)) ? 1 : 0), __extension__ ({ if (axis_count <= (12)) ; else __assert_fail ("axis_count <= CCV_NNC_MAX_DIM_ALLOC" , "ccv_cnnp_model_addons.c", 3060, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3061 | int i; | |||
| 3062 | for (i = 0; i < axis_count; i++) | |||
| 3063 | model_reduce_max->axis[i] = axis[i]; | |||
| 3064 | model_reduce_max->count = axis_count; | |||
| 3065 | return (ccv_cnnp_model_t*)model_reduce_max; | |||
| 3066 | } | |||
| 3067 | ||||
| 3068 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_max_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3069 | { | |||
| 3070 | const ccv_cnnp_model_reduce_max_t* const self = (const ccv_cnnp_model_reduce_max_t*)super; | |||
| 3071 | return ccv_cnnp_reduce_max(self->axis, self->count, self->super.name); | |||
| 3072 | } | |||
| 3073 | ||||
| 3074 | // MARK - Reduce Min Layer | |||
| 3075 | ||||
| 3076 | typedef struct { | |||
| 3077 | ccv_cnnp_model_t super; | |||
| 3078 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 3079 | int count; | |||
| 3080 | ccv_nnc_tensor_symbol_t output; | |||
| 3081 | } ccv_cnnp_model_reduce_min_t; | |||
| 3082 | ||||
| 3083 | static void _ccv_cnnp_reduce_min_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3084 | { | |||
| 3085 | PRINT(CCV_CLI_VERBOSE, "[cnnp_reduce_min_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_reduce_min_build] -\n"); fflush(stdout); } } while (0); | |||
| 3086 | const ccv_cnnp_model_reduce_min_t* const self = (const ccv_cnnp_model_reduce_min_t*)super; | |||
| 3087 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 3087, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3088 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3088, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3089 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3090 | ccv_nnc_tensor_param_t output_params; | |||
| 3091 | ccv_nnc_cmd_t reduce_min = CMD_REDUCE_MIN_FORWARD()ccv_nnc_cmd(CCV_NNC_REDUCE_MIN_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}} ), 0); | |||
| 3092 | int i; | |||
| 3093 | for (i = 0; i < self->count; i++) | |||
| 3094 | reduce_min.info.reduce.axis[i] = self->axis[i]; | |||
| 3095 | reduce_min.info.reduce.count = self->count; | |||
| 3096 | ccv_nnc_hint_tensor_auto(reduce_min, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3097 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3098 | ccv_nnc_graph_exec_symbol_new(graph, reduce_min, inputs, input_size, outputs, output_size, "reduce_min"); | |||
| 3099 | } | |||
| 3100 | ||||
| 3101 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_min_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3102 | ||||
| 3103 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_min_isa = { | |||
| 3104 | .build = _ccv_cnnp_reduce_min_build, | |||
| 3105 | .copy = _ccv_cnnp_reduce_min_copy, | |||
| 3106 | }; | |||
| 3107 | ||||
| 3108 | ccv_cnnp_model_t* ccv_cnnp_reduce_min(const int* const axis, const int axis_count, const char* const name) | |||
| 3109 | { | |||
| 3110 | ccv_cnnp_model_reduce_min_t* const model_reduce_min = (ccv_cnnp_model_reduce_min_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_reduce_min_t)); | |||
| 3111 | model_reduce_min->super.isa = &ccv_cnnp_reduce_min_isa; | |||
| 3112 | model_reduce_min->super.input_size = 1; | |||
| 3113 | model_reduce_min->super.outputs = &model_reduce_min->output; | |||
| 3114 | model_reduce_min->super.output_size = 1; | |||
| 3115 | ccv_cnnp_model_copy_name(&model_reduce_min->super, name); | |||
| 3116 | assert(axis_count <= CCV_NNC_MAX_DIM_ALLOC)((void) sizeof ((axis_count <= (12)) ? 1 : 0), __extension__ ({ if (axis_count <= (12)) ; else __assert_fail ("axis_count <= CCV_NNC_MAX_DIM_ALLOC" , "ccv_cnnp_model_addons.c", 3116, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3117 | int i; | |||
| 3118 | for (i = 0; i < axis_count; i++) | |||
| 3119 | model_reduce_min->axis[i] = axis[i]; | |||
| 3120 | model_reduce_min->count = axis_count; | |||
| 3121 | return (ccv_cnnp_model_t*)model_reduce_min; | |||
| 3122 | } | |||
| 3123 | ||||
| 3124 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_min_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3125 | { | |||
| 3126 | const ccv_cnnp_model_reduce_min_t* const self = (const ccv_cnnp_model_reduce_min_t*)super; | |||
| 3127 | return ccv_cnnp_reduce_min(self->axis, self->count, self->super.name); | |||
| 3128 | } | |||
| 3129 | ||||
| 3130 | // MARK - Reduce Norm2 Layer | |||
| 3131 | ||||
| 3132 | typedef struct { | |||
| 3133 | ccv_cnnp_model_t super; | |||
| 3134 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 3135 | int count; | |||
| 3136 | ccv_nnc_tensor_symbol_t output; | |||
| 3137 | } ccv_cnnp_model_reduce_norm2_t; | |||
| 3138 | ||||
| 3139 | static void _ccv_cnnp_reduce_norm2_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3140 | { | |||
| 3141 | const ccv_cnnp_model_reduce_norm2_t* const self = (const ccv_cnnp_model_reduce_norm2_t*)super; | |||
| 3142 | PRINT(CCV_CLI_VERBOSE, "[cnnp_reduce_norm2_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_reduce_norm2_build] -\n"); fflush(stdout); } } while (0); | |||
| 3143 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 3143, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3144 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3144, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3145 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3146 | ccv_nnc_tensor_param_t output_params; | |||
| 3147 | ccv_nnc_cmd_t reduce_norm2 = CMD_REDUCE_NORM2_FORWARD()ccv_nnc_cmd(CCV_NNC_REDUCE_NORM2_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}} ), 0); | |||
| 3148 | int i; | |||
| 3149 | for (i = 0; i < self->count; i++) | |||
| 3150 | reduce_norm2.info.reduce.axis[i] = self->axis[i]; | |||
| 3151 | reduce_norm2.info.reduce.count = self->count; | |||
| 3152 | ccv_nnc_hint_tensor_auto(reduce_norm2, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3153 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3154 | ccv_nnc_graph_exec_symbol_new(graph, reduce_norm2, inputs, input_size, outputs, output_size, "reduce_norm2"); | |||
| 3155 | } | |||
| 3156 | ||||
| 3157 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_norm2_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3158 | ||||
| 3159 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_norm2_isa = { | |||
| 3160 | .build = _ccv_cnnp_reduce_norm2_build, | |||
| 3161 | .copy = _ccv_cnnp_reduce_norm2_copy, | |||
| 3162 | }; | |||
| 3163 | ||||
| 3164 | ccv_cnnp_model_t* ccv_cnnp_reduce_norm2(const int* const axis, const int axis_count, const char* const name) | |||
| 3165 | { | |||
| 3166 | ccv_cnnp_model_reduce_norm2_t* const model_reduce_norm2 = (ccv_cnnp_model_reduce_norm2_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_reduce_norm2_t)); | |||
| 3167 | model_reduce_norm2->super.isa = &ccv_cnnp_reduce_norm2_isa; | |||
| 3168 | model_reduce_norm2->super.input_size = 1; | |||
| 3169 | model_reduce_norm2->super.outputs = &model_reduce_norm2->output; | |||
| 3170 | model_reduce_norm2->super.output_size = 1; | |||
| 3171 | ccv_cnnp_model_copy_name(&model_reduce_norm2->super, name); | |||
| 3172 | assert(axis_count <= CCV_NNC_MAX_DIM_ALLOC)((void) sizeof ((axis_count <= (12)) ? 1 : 0), __extension__ ({ if (axis_count <= (12)) ; else __assert_fail ("axis_count <= CCV_NNC_MAX_DIM_ALLOC" , "ccv_cnnp_model_addons.c", 3172, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3173 | int i; | |||
| 3174 | for (i = 0; i < axis_count; i++) | |||
| 3175 | model_reduce_norm2->axis[i] = axis[i]; | |||
| 3176 | model_reduce_norm2->count = axis_count; | |||
| 3177 | return (ccv_cnnp_model_t*)model_reduce_norm2; | |||
| 3178 | } | |||
| 3179 | ||||
| 3180 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_norm2_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3181 | { | |||
| 3182 | const ccv_cnnp_model_reduce_norm2_t* const self = (const ccv_cnnp_model_reduce_norm2_t*)super; | |||
| 3183 | return ccv_cnnp_reduce_norm2(self->axis, self->count, self->super.name); | |||
| 3184 | } | |||
| 3185 | ||||
| 3186 | // MARK - Argmax Layer | |||
| 3187 | ||||
| 3188 | typedef struct { | |||
| 3189 | ccv_cnnp_model_t super; | |||
| 3190 | int axis; | |||
| 3191 | ccv_nnc_tensor_symbol_t output; | |||
| 3192 | } ccv_cnnp_model_argmax_t; | |||
| 3193 | ||||
| 3194 | static void _ccv_cnnp_argmax_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3195 | { | |||
| 3196 | const ccv_cnnp_model_argmax_t* const self = (const ccv_cnnp_model_argmax_t*)super; | |||
| 3197 | PRINT(CCV_CLI_VERBOSE, "[cnnp_argmax_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_argmax_build] -\n"); fflush(stdout); } } while (0); | |||
| 3198 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 3198, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3199 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3199, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3200 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3201 | ccv_nnc_tensor_param_t output_params; | |||
| 3202 | ccv_nnc_cmd_t argmax = CMD_ARGMAX_FORWARD()ccv_nnc_cmd(CCV_NNC_ARGMAX_FORWARD, 0, ((ccv_nnc_cmd_param_t) {.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}}) , 0); | |||
| 3203 | argmax.info.reduce.axis[0] = self->axis; | |||
| 3204 | argmax.info.reduce.count = 1; | |||
| 3205 | ccv_nnc_hint_tensor_auto(argmax, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3206 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3207 | ccv_nnc_graph_exec_symbol_new(graph, argmax, inputs, input_size, outputs, output_size, "argmax"); | |||
| 3208 | } | |||
| 3209 | ||||
| 3210 | static ccv_cnnp_model_t* _ccv_cnnp_argmax_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3211 | ||||
| 3212 | static const ccv_cnnp_model_vtab_t ccv_cnnp_argmax_isa = { | |||
| 3213 | .build = _ccv_cnnp_argmax_build, | |||
| 3214 | .copy = _ccv_cnnp_argmax_copy, | |||
| 3215 | }; | |||
| 3216 | ||||
| 3217 | ccv_cnnp_model_t* ccv_cnnp_argmax(const int axis, const char* const name) | |||
| 3218 | { | |||
| 3219 | ccv_cnnp_model_argmax_t* const model_argmax = (ccv_cnnp_model_argmax_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_argmax_t)); | |||
| 3220 | model_argmax->super.isa = &ccv_cnnp_argmax_isa; | |||
| 3221 | model_argmax->super.input_size = 1; | |||
| 3222 | model_argmax->super.outputs = &model_argmax->output; | |||
| 3223 | model_argmax->super.output_size = 1; | |||
| 3224 | ccv_cnnp_model_copy_name(&model_argmax->super, name); | |||
| 3225 | model_argmax->axis = axis; | |||
| 3226 | return (ccv_cnnp_model_t*)model_argmax; | |||
| 3227 | } | |||
| 3228 | ||||
| 3229 | static ccv_cnnp_model_t* _ccv_cnnp_argmax_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3230 | { | |||
| 3231 | const ccv_cnnp_model_argmax_t* const self = (const ccv_cnnp_model_argmax_t*)super; | |||
| 3232 | return ccv_cnnp_argmax(self->axis, self->super.name); | |||
| 3233 | } | |||
| 3234 | ||||
| 3235 | // MARK - Argmin Layer | |||
| 3236 | ||||
| 3237 | typedef struct { | |||
| 3238 | ccv_cnnp_model_t super; | |||
| 3239 | int axis; | |||
| 3240 | ccv_nnc_tensor_symbol_t output; | |||
| 3241 | } ccv_cnnp_model_argmin_t; | |||
| 3242 | ||||
| 3243 | static void _ccv_cnnp_argmin_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3244 | { | |||
| 3245 | const ccv_cnnp_model_argmin_t* const self = (const ccv_cnnp_model_argmin_t*)super; | |||
| 3246 | PRINT(CCV_CLI_VERBOSE, "[cnnp_argmin_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_argmin_build] -\n"); fflush(stdout); } } while (0); | |||
| 3247 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 3247, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3248 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3248, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3249 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3250 | ccv_nnc_tensor_param_t output_params; | |||
| 3251 | ccv_nnc_cmd_t argmin = CMD_ARGMIN_FORWARD()ccv_nnc_cmd(CCV_NNC_ARGMIN_FORWARD, 0, ((ccv_nnc_cmd_param_t) {.size={.dim={1,1,1}},.reduce={.count=(1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1),.axis={}}}) , 0); | |||
| 3252 | argmin.info.reduce.axis[0] = self->axis; | |||
| 3253 | argmin.info.reduce.count = 1; | |||
| 3254 | ccv_nnc_hint_tensor_auto(argmin, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3255 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3256 | ccv_nnc_graph_exec_symbol_new(graph, argmin, inputs, input_size, outputs, output_size, "argmin"); | |||
| 3257 | } | |||
| 3258 | ||||
| 3259 | static ccv_cnnp_model_t* _ccv_cnnp_argmin_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3260 | ||||
| 3261 | static const ccv_cnnp_model_vtab_t ccv_cnnp_argmin_isa = { | |||
| 3262 | .build = _ccv_cnnp_argmin_build, | |||
| 3263 | .copy = _ccv_cnnp_argmin_copy, | |||
| 3264 | }; | |||
| 3265 | ||||
| 3266 | ccv_cnnp_model_t* ccv_cnnp_argmin(const int axis, const char* const name) | |||
| 3267 | { | |||
| 3268 | ccv_cnnp_model_argmin_t* const model_argmin = (ccv_cnnp_model_argmin_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_argmin_t)); | |||
| 3269 | model_argmin->super.isa = &ccv_cnnp_argmin_isa; | |||
| 3270 | model_argmin->super.input_size = 1; | |||
| 3271 | model_argmin->super.outputs = &model_argmin->output; | |||
| 3272 | model_argmin->super.output_size = 1; | |||
| 3273 | ccv_cnnp_model_copy_name(&model_argmin->super, name); | |||
| 3274 | model_argmin->axis = axis; | |||
| 3275 | return (ccv_cnnp_model_t*)model_argmin; | |||
| 3276 | } | |||
| 3277 | ||||
| 3278 | static ccv_cnnp_model_t* _ccv_cnnp_argmin_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3279 | { | |||
| 3280 | const ccv_cnnp_model_argmin_t* const self = (const ccv_cnnp_model_argmin_t*)super; | |||
| 3281 | return ccv_cnnp_argmin(self->axis, self->super.name); | |||
| 3282 | } | |||
| 3283 | ||||
| 3284 | // MARK - Min Layer | |||
| 3285 | ||||
| 3286 | typedef struct { | |||
| 3287 | ccv_cnnp_model_t super; | |||
| 3288 | ccv_nnc_tensor_symbol_t output; | |||
| 3289 | } ccv_cnnp_model_min_t; | |||
| 3290 | ||||
| 3291 | static void _ccv_cnnp_min_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3292 | { | |||
| 3293 | PRINT(CCV_CLI_VERBOSE, "[cnnp_min_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_min_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 3294 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 3294, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3295 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3295, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3296 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 3297 | int i; | |||
| 3298 | for (i = 0; i < 2; i++) | |||
| 3299 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 3300 | ccv_nnc_tensor_param_t output_params; | |||
| 3301 | const ccv_nnc_cmd_t min = CMD_MIN_FORWARD()ccv_nnc_cmd(CCV_NNC_MIN_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}}}, 0); | |||
| 3302 | ccv_nnc_hint_tensor_auto(min, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 3303 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3304 | ccv_nnc_graph_exec_symbol_new(graph, min, inputs, input_size, outputs, output_size, "min"); | |||
| 3305 | } | |||
| 3306 | ||||
| 3307 | static ccv_cnnp_model_t* _ccv_cnnp_min_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3308 | ||||
| 3309 | static const ccv_cnnp_model_vtab_t ccv_cnnp_min_isa = { | |||
| 3310 | .build = _ccv_cnnp_min_build, | |||
| 3311 | .copy = _ccv_cnnp_min_copy, | |||
| 3312 | }; | |||
| 3313 | ||||
| 3314 | ccv_cnnp_model_t* ccv_cnnp_min(const char* const name) | |||
| 3315 | { | |||
| 3316 | ccv_cnnp_model_min_t* const model_min = (ccv_cnnp_model_min_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_min_t)); | |||
| 3317 | model_min->super.isa = &ccv_cnnp_min_isa; | |||
| 3318 | model_min->super.input_size = 2; | |||
| 3319 | model_min->super.outputs = &model_min->output; | |||
| 3320 | model_min->super.output_size = 1; | |||
| 3321 | ccv_cnnp_model_copy_name(&model_min->super, name); | |||
| 3322 | return (ccv_cnnp_model_t*)model_min; | |||
| 3323 | } | |||
| 3324 | ||||
| 3325 | static ccv_cnnp_model_t* _ccv_cnnp_min_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3326 | { | |||
| 3327 | const ccv_cnnp_model_min_t* const self = (const ccv_cnnp_model_min_t*)super; | |||
| 3328 | return ccv_cnnp_min(self->super.name); | |||
| 3329 | } | |||
| 3330 | ||||
| 3331 | // MARK - Max Layer | |||
| 3332 | ||||
| 3333 | typedef struct { | |||
| 3334 | ccv_cnnp_model_t super; | |||
| 3335 | ccv_nnc_tensor_symbol_t output; | |||
| 3336 | } ccv_cnnp_model_max_t; | |||
| 3337 | ||||
| 3338 | static void _ccv_cnnp_max_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3339 | { | |||
| 3340 | PRINT(CCV_CLI_VERBOSE, "[cnnp_max_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_max_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 3341 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 3341, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3342 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3342, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3343 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 3344 | int i; | |||
| 3345 | for (i = 0; i < 2; i++) | |||
| 3346 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 3347 | ccv_nnc_tensor_param_t output_params; | |||
| 3348 | const ccv_nnc_cmd_t max = CMD_MAX_FORWARD()ccv_nnc_cmd(CCV_NNC_MAX_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}}}, 0); | |||
| 3349 | ccv_nnc_hint_tensor_auto(max, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 3350 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3351 | ccv_nnc_graph_exec_symbol_new(graph, max, inputs, input_size, outputs, output_size, "max"); | |||
| 3352 | } | |||
| 3353 | ||||
| 3354 | static ccv_cnnp_model_t* _ccv_cnnp_max_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3355 | ||||
| 3356 | static const ccv_cnnp_model_vtab_t ccv_cnnp_max_isa = { | |||
| 3357 | .build = _ccv_cnnp_max_build, | |||
| 3358 | .copy = _ccv_cnnp_max_copy, | |||
| 3359 | }; | |||
| 3360 | ||||
| 3361 | ccv_cnnp_model_t* ccv_cnnp_max(const char* const name) | |||
| 3362 | { | |||
| 3363 | ccv_cnnp_model_max_t* const model_max = (ccv_cnnp_model_max_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_max_t)); | |||
| 3364 | model_max->super.isa = &ccv_cnnp_max_isa; | |||
| 3365 | model_max->super.input_size = 2; | |||
| 3366 | model_max->super.outputs = &model_max->output; | |||
| 3367 | model_max->super.output_size = 1; | |||
| 3368 | ccv_cnnp_model_copy_name(&model_max->super, name); | |||
| 3369 | return (ccv_cnnp_model_t*)model_max; | |||
| 3370 | } | |||
| 3371 | ||||
| 3372 | static ccv_cnnp_model_t* _ccv_cnnp_max_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3373 | { | |||
| 3374 | const ccv_cnnp_model_max_t* const self = (const ccv_cnnp_model_max_t*)super; | |||
| 3375 | return ccv_cnnp_max(self->super.name); | |||
| 3376 | } | |||
| 3377 | ||||
| 3378 | // MARK - LSTM Layer | |||
| 3379 | ||||
| 3380 | typedef struct { | |||
| 3381 | ccv_cnnp_model_t super; | |||
| 3382 | int masked; | |||
| 3383 | ccv_nnc_tensor_symbol_t output; | |||
| 3384 | ccv_nnc_tensor_symbol_t weights; | |||
| 3385 | ccv_nnc_tensor_symbol_t reserves; | |||
| 3386 | ccv_nnc_cmd_param_t params; | |||
| 3387 | ccv_nnc_graph_exec_symbol_t lstm; | |||
| 3388 | } ccv_cnnp_model_lstm_t; | |||
| 3389 | ||||
| 3390 | static int _ccv_cnnp_lstm_weight_dim(int bidirectional, int num_layers, int input_size, int hidden_size, int proj_size, int bias) | |||
| 3391 | { | |||
| 3392 | const int D = !!bidirectional + 1; | |||
| 3393 | if (hidden_size == proj_size) | |||
| 3394 | return (num_layers * (bias ? 8 : 0) + (num_layers - 1) * (hidden_size * 4 * D + hidden_size * 4) + input_size * 4 + hidden_size * 4) * D; | |||
| 3395 | else | |||
| 3396 | return (num_layers * (bias ? 8 : 0) + (num_layers - 1) * (proj_size * 4 * D + proj_size * 4) + (proj_size * 4 + input_size * 4) + num_layers * proj_size) * D; | |||
| 3397 | } | |||
| 3398 | ||||
| 3399 | static void _ccv_cnnp_lstm_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3400 | { | |||
| 3401 | ccv_cnnp_model_lstm_t* const self = (ccv_cnnp_model_lstm_t*)super; | |||
| 3402 | PRINT(CCV_CLI_VERBOSE, "[cnnp_lstm_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_lstm_build] -\n"); fflush(stdout); } } while (0); | |||
| 3403 | assert(input_size == self->super.input_size)((void) sizeof ((input_size == self->super.input_size) ? 1 : 0), __extension__ ({ if (input_size == self->super.input_size ) ; else __assert_fail ("input_size == self->super.input_size" , "ccv_cnnp_model_addons.c", 3403, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3404 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3404, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3405 | const int proj_size = self->params.rnn.proj_size == 0 ? self->params.rnn.hidden_size : self->params.rnn.proj_size; | |||
| 3406 | ccv_nnc_tensor_param_t input_params[5]; | |||
| 3407 | input_params[0]= ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3408 | if (input_size == 2) | |||
| 3409 | input_params[1] = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 3410 | input_params[4] = input_params[0]; | |||
| 3411 | memset(input_params[4].dim, 0, sizeof(input_params[4].dim)); | |||
| 3412 | const int x_nd = ccv_nnc_tensor_nd(input_params[0].dim); | |||
| 3413 | const int feature_count = input_params[0].dim[x_nd - 1]; | |||
| 3414 | input_params[4].dim[0] = _ccv_cnnp_lstm_weight_dim(self->params.rnn.bidirectional, self->params.rnn.num_layers, feature_count, self->params.rnn.hidden_size, proj_size, self->params.rnn.bias); | |||
| 3415 | input_params[4].dim[1] = self->params.rnn.hidden_size; | |||
| 3416 | const ccv_nnc_cmd_t lstm = ccv_nnc_cmd(CCV_NNC_LSTM_FORWARD, 0, self->params, 0); | |||
| 3417 | ccv_nnc_tensor_param_t output_params[4]; | |||
| 3418 | ccv_nnc_hint_tensor_auto(lstm, input_params, 5, ccv_nnc_no_hint, output_params, 4); | |||
| 3419 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 3420 | if (!self->weights.graph) | |||
| 3421 | self->weights = ccv_nnc_tensor_symbol_new(graph, input_params[4], "weights"); | |||
| 3422 | if (!self->reserves.graph) | |||
| 3423 | self->reserves = ccv_nnc_tensor_symbol_new(graph, output_params[3], "reserves"); | |||
| 3424 | const ccv_nnc_tensor_symbol_t mask = input_size == 2 ? inputs[1] : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3425 | self->lstm = ccv_nnc_graph_exec_symbol_new(graph, lstm, TENSOR_SYMBOL_LIST(inputs[0], mask, NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, self->weights)(const ccv_nnc_tensor_symbol_t []){inputs[0], mask, (const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, (const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, self->weights}, (1 +1 +1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 - 1), TENSOR_SYMBOL_LIST(outputs[0], NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, self->reserves)(const ccv_nnc_tensor_symbol_t []){outputs[0], (const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, (const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, self->reserves}, (1 +1 + 1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), "lstm"); | |||
| 3426 | } | |||
| 3427 | ||||
| 3428 | static void _ccv_cnnp_lstm_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 3429 | { | |||
| 3430 | ccv_cnnp_model_lstm_t* const self = (ccv_cnnp_model_lstm_t*)super; | |||
| 3431 | if (self->weights.graph) | |||
| 3432 | { | |||
| 3433 | const float stdv = 1.0 / sqrt(self->params.rnn.hidden_size); | |||
| 3434 | initializer(context, CMD_RANDOM_UNIFORM_FORWARD(-stdv, stdv)ccv_nnc_cmd(CCV_NNC_RANDOM_UNIFORM_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={-stdv, stdv}}}, 0), ccv_nnc_no_hint, 0, 0, self->weights); | |||
| 3435 | } | |||
| 3436 | } | |||
| 3437 | ||||
| 3438 | static void _ccv_cnnp_lstm_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 3439 | { | |||
| 3440 | ccv_cnnp_model_lstm_t* const self = (ccv_cnnp_model_lstm_t*)super; | |||
| 3441 | if (self->weights.graph) | |||
| 3442 | add_to_array(parameters, self->weights, is_trainable); | |||
| 3443 | } | |||
| 3444 | ||||
| 3445 | static void _ccv_cnnp_lstm_set_is_test(ccv_cnnp_model_t* const super, const int is_test, const ccv_cnnp_cmd_updater_f updater, void* const context) | |||
| 3446 | { | |||
| 3447 | ccv_cnnp_model_lstm_t* const self = (ccv_cnnp_model_lstm_t*)super; | |||
| 3448 | if (self->lstm.graph) | |||
| 3449 | { | |||
| 3450 | self->params.rnn.is_test = is_test; | |||
| 3451 | updater(context, self->lstm, ccv_nnc_cmd(CCV_NNC_LSTM_FORWARD, 0, self->params, 0), ccv_nnc_no_hint); | |||
| 3452 | } | |||
| 3453 | } | |||
| 3454 | ||||
| 3455 | static ccv_cnnp_model_t* _ccv_cnnp_lstm_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3456 | ||||
| 3457 | static const ccv_cnnp_model_vtab_t ccv_cnnp_lstm_isa = { | |||
| 3458 | .build = _ccv_cnnp_lstm_build, | |||
| 3459 | .init_states = _ccv_cnnp_lstm_init_states, | |||
| 3460 | .add_to_parameter = _ccv_cnnp_lstm_add_to_parameter, | |||
| 3461 | .copy = _ccv_cnnp_lstm_copy, | |||
| 3462 | .set_is_test = _ccv_cnnp_lstm_set_is_test, | |||
| 3463 | }; | |||
| 3464 | ||||
| 3465 | ccv_cnnp_model_t* ccv_cnnp_lstm(const int masked, const int hidden_size, const int proj_size, const int num_layers, const int bias, const int batch_first, const int bidirectional, const float dropout, const int is_trainable, const char* const name) | |||
| 3466 | { | |||
| 3467 | ccv_cnnp_model_lstm_t* const model_lstm = (ccv_cnnp_model_lstm_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_lstm_t)); | |||
| 3468 | model_lstm->super.isa = &ccv_cnnp_lstm_isa; | |||
| 3469 | model_lstm->super.input_size = masked ? 2 : 1; | |||
| 3470 | model_lstm->super.outputs = &model_lstm->output; | |||
| 3471 | model_lstm->super.output_size = 1; | |||
| 3472 | model_lstm->super.is_trainable = is_trainable; | |||
| 3473 | ccv_cnnp_model_copy_name(&model_lstm->super, name); | |||
| 3474 | model_lstm->masked = masked; | |||
| 3475 | model_lstm->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3476 | model_lstm->weights.graph = 0; | |||
| 3477 | model_lstm->params.rnn.hidden_size = hidden_size; | |||
| 3478 | model_lstm->params.rnn.proj_size = proj_size; | |||
| 3479 | model_lstm->params.rnn.num_layers = num_layers; | |||
| 3480 | model_lstm->params.rnn.bias = bias; | |||
| 3481 | model_lstm->params.rnn.batch_first = batch_first; | |||
| 3482 | model_lstm->params.rnn.bidirectional = bidirectional; | |||
| 3483 | model_lstm->params.rnn.dropout = dropout; | |||
| 3484 | return (ccv_cnnp_model_t*)model_lstm; | |||
| 3485 | } | |||
| 3486 | ||||
| 3487 | static ccv_cnnp_model_t* _ccv_cnnp_lstm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3488 | { | |||
| 3489 | const ccv_cnnp_model_lstm_t* const self = (const ccv_cnnp_model_lstm_t*)super; | |||
| 3490 | return ccv_cnnp_lstm(self->masked, self->params.rnn.hidden_size, self->params.rnn.proj_size, self->params.rnn.num_layers, self->params.rnn.bias, self->params.rnn.batch_first, self->params.rnn.bidirectional, self->params.rnn.dropout, self->super.is_trainable, self->super.name); | |||
| 3491 | } | |||
| 3492 | ||||
| 3493 | /// MARK - Datatype conversion layer. | |||
| 3494 | ||||
| 3495 | typedef struct { | |||
| 3496 | ccv_cnnp_model_t super; | |||
| 3497 | ccv_nnc_tensor_symbol_t output; | |||
| 3498 | int datatype; | |||
| 3499 | int ref_to_last; | |||
| 3500 | } ccv_cnnp_model_datatype_conversion_t; | |||
| 3501 | ||||
| 3502 | static void _ccv_cnnp_datatype_conversion_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3503 | { | |||
| 3504 | ccv_cnnp_model_datatype_conversion_t* const self = (ccv_cnnp_model_datatype_conversion_t*)super; | |||
| 3505 | PRINT(CCV_CLI_VERBOSE, "[cnnp_datatype_conversion_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_datatype_conversion_build] -\n"); fflush(stdout ); } } while (0); | |||
| 3506 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3507 | if (self->ref_to_last) | |||
| 3508 | { | |||
| 3509 | assert(input_size > 1)((void) sizeof ((input_size > 1) ? 1 : 0), __extension__ ( { if (input_size > 1) ; else __assert_fail ("input_size > 1" , "ccv_cnnp_model_addons.c", 3509, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3510 | const ccv_nnc_tensor_param_t last_params = ccv_nnc_tensor_symbol_params(graph, inputs[input_size - 1]); | |||
| 3511 | params.datatype = last_params.datatype; | |||
| 3512 | } else | |||
| 3513 | params.datatype = self->datatype; | |||
| 3514 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3514, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3515 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 3516 | ccv_nnc_graph_exec_symbol_new(graph, CMD_DATATYPE_CONVERSION_FORWARD()ccv_nnc_cmd(CCV_NNC_DATATYPE_CONVERSION_FORWARD, 0, ccv_nnc_cmd_auto , 0), inputs, output_size, outputs, output_size, 0); | |||
| 3517 | } | |||
| 3518 | ||||
| 3519 | static ccv_cnnp_model_t* _ccv_cnnp_datatype_conversion_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3520 | ||||
| 3521 | static const ccv_cnnp_model_vtab_t ccv_cnnp_datatype_conversion_isa = { | |||
| 3522 | .build = _ccv_cnnp_datatype_conversion_build, | |||
| 3523 | .copy = _ccv_cnnp_datatype_conversion_copy, | |||
| 3524 | }; | |||
| 3525 | ||||
| 3526 | ccv_cnnp_model_t* ccv_cnnp_datatype_conversion(const int datatype, const int ref_to_last, const char* const name) | |||
| 3527 | { | |||
| 3528 | ccv_cnnp_model_datatype_conversion_t* const model_datatype_conversion = (ccv_cnnp_model_datatype_conversion_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_datatype_conversion_t)); | |||
| 3529 | model_datatype_conversion->super.isa = &ccv_cnnp_datatype_conversion_isa; | |||
| 3530 | model_datatype_conversion->super.input_size = 0; | |||
| 3531 | model_datatype_conversion->super.outputs = &model_datatype_conversion->output; | |||
| 3532 | model_datatype_conversion->super.output_size = 1; | |||
| 3533 | model_datatype_conversion->datatype = datatype; | |||
| 3534 | model_datatype_conversion->ref_to_last = ref_to_last; | |||
| 3535 | ccv_cnnp_model_copy_name(&model_datatype_conversion->super, name); | |||
| 3536 | return (ccv_cnnp_model_t*)model_datatype_conversion; | |||
| 3537 | } | |||
| 3538 | ||||
| 3539 | static ccv_cnnp_model_t* _ccv_cnnp_datatype_conversion_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3540 | { | |||
| 3541 | ccv_cnnp_model_datatype_conversion_t* const self = (ccv_cnnp_model_datatype_conversion_t*)super; | |||
| 3542 | return ccv_cnnp_datatype_conversion(self->datatype, self->ref_to_last, self->super.name); | |||
| 3543 | } | |||
| 3544 | ||||
| 3545 | /// MARK - Clamp layer. | |||
| 3546 | ||||
| 3547 | typedef struct { | |||
| 3548 | ccv_cnnp_model_t super; | |||
| 3549 | ccv_nnc_tensor_symbol_t output; | |||
| 3550 | float min; | |||
| 3551 | float max; | |||
| 3552 | } ccv_cnnp_model_clamp_t; | |||
| 3553 | ||||
| 3554 | static void _ccv_cnnp_clamp_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3555 | { | |||
| 3556 | ccv_cnnp_model_clamp_t* const self = (ccv_cnnp_model_clamp_t*)super; | |||
| 3557 | PRINT(CCV_CLI_VERBOSE, "[cnnp_clamp_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_clamp_build] -\n"); fflush(stdout); } } while (0); | |||
| 3558 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3559 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3559, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3560 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 3561 | ccv_nnc_graph_exec_symbol_new(graph, CMD_CLAMP_FORWARD(self->min, self->max)ccv_nnc_cmd(CCV_NNC_CLAMP_FORWARD, 0, (ccv_nnc_cmd_param_t){. size={.dim={1,1,1}},.clamp={.min=self->min,.max=self->max }}, 0), inputs, output_size, outputs, output_size, 0); | |||
| 3562 | } | |||
| 3563 | ||||
| 3564 | static ccv_cnnp_model_t* _ccv_cnnp_clamp_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3565 | ||||
| 3566 | static const ccv_cnnp_model_vtab_t ccv_cnnp_clamp_isa = { | |||
| 3567 | .build = _ccv_cnnp_clamp_build, | |||
| 3568 | .copy = _ccv_cnnp_clamp_copy, | |||
| 3569 | }; | |||
| 3570 | ||||
| 3571 | ccv_cnnp_model_t* ccv_cnnp_clamp(const float min, const float max, const char* const name) | |||
| 3572 | { | |||
| 3573 | ccv_cnnp_model_clamp_t* const model_clamp = (ccv_cnnp_model_clamp_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_clamp_t)); | |||
| 3574 | model_clamp->super.isa = &ccv_cnnp_clamp_isa; | |||
| 3575 | model_clamp->super.input_size = 0; | |||
| 3576 | model_clamp->super.outputs = &model_clamp->output; | |||
| 3577 | model_clamp->super.output_size = 1; | |||
| 3578 | model_clamp->min = min; | |||
| 3579 | model_clamp->max = max; | |||
| 3580 | ccv_cnnp_model_copy_name(&model_clamp->super, name); | |||
| 3581 | return (ccv_cnnp_model_t*)model_clamp; | |||
| 3582 | } | |||
| 3583 | ||||
| 3584 | static ccv_cnnp_model_t* _ccv_cnnp_clamp_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3585 | { | |||
| 3586 | ccv_cnnp_model_clamp_t* const self = (ccv_cnnp_model_clamp_t*)super; | |||
| 3587 | return ccv_cnnp_clamp(self->min, self->max, self->super.name); | |||
| 3588 | } | |||
| 3589 | ||||
| 3590 | // MARK - Parameter Layer | |||
| 3591 | ||||
| 3592 | typedef struct { | |||
| 3593 | ccv_cnnp_model_t super; | |||
| 3594 | float init_bound; | |||
| 3595 | ccv_nnc_tensor_symbol_t weights; | |||
| 3596 | ccv_nnc_tensor_param_t weights_params; | |||
| 3597 | ccv_nnc_tensor_symbol_t output; | |||
| 3598 | } ccv_cnnp_model_parameter_t; | |||
| 3599 | ||||
| 3600 | static void _ccv_cnnp_parameter_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3601 | { | |||
| 3602 | PRINT(CCV_CLI_VERBOSE, "[cnnp_parameter_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_parameter_build] -\n"); fflush(stdout); } } while (0); | |||
| 3603 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3603, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3604 | ccv_cnnp_model_parameter_t* const self = (ccv_cnnp_model_parameter_t*)super; | |||
| 3605 | if (!self->weights.graph) | |||
| 3606 | self->weights = ccv_nnc_tensor_symbol_new(graph, self->weights_params, "weights"); | |||
| 3607 | assert(self->weights.graph == graph)((void) sizeof ((self->weights.graph == graph) ? 1 : 0), __extension__ ({ if (self->weights.graph == graph) ; else __assert_fail ("self->weights.graph == graph", "ccv_cnnp_model_addons.c" , 3607, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3608 | outputs[0] = self->weights; | |||
| 3609 | } | |||
| 3610 | ||||
| 3611 | static void _ccv_cnnp_parameter_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 3612 | { | |||
| 3613 | ccv_cnnp_model_parameter_t* const self = (ccv_cnnp_model_parameter_t*)super; | |||
| 3614 | if (self->init_bound > 0) | |||
| 3615 | initializer(context, CMD_RANDOM_UNIFORM_FORWARD(-self->init_bound, self->init_bound)ccv_nnc_cmd(CCV_NNC_RANDOM_UNIFORM_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={-self->init_bound, self-> init_bound}}}, 0), ccv_nnc_no_hint, 0, 0, self->weights); | |||
| 3616 | else | |||
| 3617 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, self->weights); | |||
| 3618 | } | |||
| 3619 | ||||
| 3620 | static void _ccv_cnnp_parameter_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 3621 | { | |||
| 3622 | ccv_cnnp_model_parameter_t* const self = (ccv_cnnp_model_parameter_t*)super; | |||
| 3623 | add_to_array(parameters, self->weights, is_trainable); | |||
| 3624 | } | |||
| 3625 | ||||
| 3626 | static ccv_cnnp_model_t* _ccv_cnnp_parameter_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3627 | ||||
| 3628 | static const ccv_cnnp_model_vtab_t ccv_cnnp_parameter_isa = { | |||
| 3629 | .build = _ccv_cnnp_parameter_build, | |||
| 3630 | .init_states = _ccv_cnnp_parameter_init_states, | |||
| 3631 | .add_to_parameter = _ccv_cnnp_parameter_add_to_parameter, | |||
| 3632 | .copy = _ccv_cnnp_parameter_copy, | |||
| 3633 | }; | |||
| 3634 | ||||
| 3635 | ccv_cnnp_model_t* ccv_cnnp_parameter(const ccv_nnc_tensor_param_t params, const float init_bound, const int is_trainable, const char* const name) | |||
| 3636 | { | |||
| 3637 | ccv_cnnp_model_parameter_t* const model_parameter = (ccv_cnnp_model_parameter_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_parameter_t)); | |||
| 3638 | model_parameter->super.isa = &ccv_cnnp_parameter_isa; | |||
| 3639 | model_parameter->super.input_size = 0; | |||
| 3640 | model_parameter->super.outputs = &model_parameter->output; | |||
| 3641 | model_parameter->super.output_size = 1; | |||
| 3642 | model_parameter->super.is_trainable = is_trainable; | |||
| 3643 | ccv_cnnp_model_copy_name(&model_parameter->super, name); | |||
| 3644 | model_parameter->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3645 | model_parameter->weights.graph = 0; | |||
| 3646 | model_parameter->weights_params = params; | |||
| 3647 | return (ccv_cnnp_model_t*)model_parameter; | |||
| 3648 | } | |||
| 3649 | ||||
| 3650 | static ccv_cnnp_model_t* _ccv_cnnp_parameter_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3651 | { | |||
| 3652 | const ccv_cnnp_model_parameter_t* const self = (const ccv_cnnp_model_parameter_t*)super; | |||
| 3653 | return ccv_cnnp_parameter(self->weights_params, self->init_bound, self->super.is_trainable, self->super.name); | |||
| 3654 | } | |||
| 3655 | ||||
| 3656 | // MARK - Scalar Layer | |||
| 3657 | ||||
| 3658 | typedef struct { | |||
| 3659 | ccv_cnnp_model_t super; | |||
| 3660 | int type; | |||
| 3661 | int format; | |||
| 3662 | int datatype; | |||
| 3663 | float value; | |||
| 3664 | ccv_nnc_tensor_symbol_t output; | |||
| 3665 | } ccv_cnnp_model_scalar_t; | |||
| 3666 | ||||
| 3667 | static void _ccv_cnnp_scalar_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3668 | { | |||
| 3669 | PRINT(CCV_CLI_VERBOSE, "[cnnp_scalar_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_scalar_build] -\n"); fflush(stdout); } } while (0); | |||
| 3670 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3670, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3671 | ccv_cnnp_model_scalar_t* const self = (ccv_cnnp_model_scalar_t*)super; | |||
| 3672 | ccv_nnc_tensor_param_t params = { | |||
| 3673 | .type = self->type, | |||
| 3674 | .format = self->format, | |||
| 3675 | .datatype = self->datatype, | |||
| 3676 | .dim = { | |||
| 3677 | 1 | |||
| 3678 | } | |||
| 3679 | }; | |||
| 3680 | if (input_size > 0) | |||
| 3681 | { | |||
| 3682 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3683 | params.type = input_params.type; | |||
| 3684 | params.format = input_params.format; | |||
| 3685 | params.datatype = input_params.datatype; | |||
| 3686 | } | |||
| 3687 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 3688 | ccv_nnc_graph_exec_symbol_new(graph, CMD_SET_FORWARD(self->value)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={self->value,}}}, 0), 0, 0, outputs, 1, 0); | |||
| 3689 | } | |||
| 3690 | ||||
| 3691 | static ccv_cnnp_model_t* _ccv_cnnp_scalar_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3692 | ||||
| 3693 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scalar_isa = { | |||
| 3694 | .build = _ccv_cnnp_scalar_build, | |||
| 3695 | .copy = _ccv_cnnp_scalar_copy, | |||
| 3696 | }; | |||
| 3697 | ||||
| 3698 | ccv_cnnp_model_t* ccv_cnnp_scalar(const int type, const int format, const int datatype, const float value, const char* const name) | |||
| 3699 | { | |||
| 3700 | ccv_cnnp_model_scalar_t* const model_scalar = (ccv_cnnp_model_scalar_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_scalar_t)); | |||
| 3701 | model_scalar->super.isa = &ccv_cnnp_scalar_isa; | |||
| 3702 | model_scalar->super.input_size = 0; | |||
| 3703 | model_scalar->super.outputs = &model_scalar->output; | |||
| 3704 | model_scalar->super.output_size = 1; | |||
| 3705 | ccv_cnnp_model_copy_name(&model_scalar->super, name); | |||
| 3706 | model_scalar->type = type; | |||
| 3707 | model_scalar->format = format; | |||
| 3708 | model_scalar->datatype = datatype; | |||
| 3709 | model_scalar->value = value; | |||
| 3710 | return (ccv_cnnp_model_t*)model_scalar; | |||
| 3711 | } | |||
| 3712 | ||||
| 3713 | static ccv_cnnp_model_t* _ccv_cnnp_scalar_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3714 | { | |||
| 3715 | const ccv_cnnp_model_scalar_t* const self = (const ccv_cnnp_model_scalar_t*)super; | |||
| 3716 | return ccv_cnnp_scalar(self->type, self->format, self->datatype, self->value, self->super.name); | |||
| 3717 | } | |||
| 3718 | ||||
| 3719 | // MARK - Variable Layer | |||
| 3720 | ||||
| 3721 | typedef struct { | |||
| 3722 | ccv_cnnp_model_t super; | |||
| 3723 | ccv_nnc_tensor_param_t params; | |||
| 3724 | ccv_nnc_tensor_symbol_t output; | |||
| 3725 | } ccv_cnnp_model_variable_t; | |||
| 3726 | ||||
| 3727 | static void _ccv_cnnp_variable_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3728 | { | |||
| 3729 | PRINT(CCV_CLI_VERBOSE, "[cnnp_variable_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_variable_build] -\n"); fflush(stdout); } } while (0); | |||
| 3730 | assert(input_size == 0)((void) sizeof ((input_size == 0) ? 1 : 0), __extension__ ({ if (input_size == 0) ; else __assert_fail ("input_size == 0", "ccv_cnnp_model_addons.c" , 3730, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3731 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3731, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3732 | ccv_cnnp_model_variable_t* const self = (ccv_cnnp_model_variable_t*)super; | |||
| 3733 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, self->params, 0); | |||
| 3734 | } | |||
| 3735 | ||||
| 3736 | static ccv_cnnp_model_t* _ccv_cnnp_variable_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3737 | ||||
| 3738 | static const ccv_cnnp_model_vtab_t ccv_cnnp_variable_isa = { | |||
| 3739 | .build = _ccv_cnnp_variable_build, | |||
| 3740 | .copy = _ccv_cnnp_variable_copy, | |||
| 3741 | }; | |||
| 3742 | ||||
| 3743 | ccv_cnnp_model_t* ccv_cnnp_variable(const ccv_nnc_tensor_param_t params, const char* const name) | |||
| 3744 | { | |||
| 3745 | ccv_cnnp_model_variable_t* const model_variable = (ccv_cnnp_model_variable_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_variable_t)); | |||
| 3746 | model_variable->super.isa = &ccv_cnnp_variable_isa; | |||
| 3747 | model_variable->super.input_size = 0; | |||
| 3748 | model_variable->super.outputs = &model_variable->output; | |||
| 3749 | model_variable->super.output_size = 1; | |||
| 3750 | ccv_cnnp_model_copy_name(&model_variable->super, name); | |||
| 3751 | model_variable->params = params; | |||
| 3752 | return (ccv_cnnp_model_t*)model_variable; | |||
| 3753 | } | |||
| 3754 | ||||
| 3755 | static ccv_cnnp_model_t* _ccv_cnnp_variable_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3756 | { | |||
| 3757 | const ccv_cnnp_model_variable_t* const self = (const ccv_cnnp_model_variable_t*)super; | |||
| 3758 | return ccv_cnnp_variable(self->params, self->super.name); | |||
| 3759 | } | |||
| 3760 | ||||
| 3761 | // MARK - Move Layer | |||
| 3762 | ||||
| 3763 | typedef struct { | |||
| 3764 | ccv_cnnp_model_t super; | |||
| 3765 | ccv_nnc_tensor_symbol_t output; | |||
| 3766 | } ccv_cnnp_model_move_t; | |||
| 3767 | ||||
| 3768 | static void _ccv_cnnp_move_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3769 | { | |||
| 3770 | PRINT(CCV_CLI_VERBOSE, "[cnnp_move_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_move_build] -\n"); fflush(stdout); } } while (0); | |||
| 3771 | assert(input_size == 2)((void) sizeof ((input_size == 2) ? 1 : 0), __extension__ ({ if (input_size == 2) ; else __assert_fail ("input_size == 2", "ccv_cnnp_model_addons.c" , 3771, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3772 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3772, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3773 | outputs[0] = inputs[1]; | |||
| 3774 | ccv_nnc_graph_exec_symbol_new(graph, CMD_FORMAT_TRANSFORM_FORWARD()ccv_nnc_cmd(CCV_NNC_FORMAT_TRANSFORM_FORWARD, 0, ccv_nnc_cmd_auto , 0), inputs, 1, outputs, 1, "move"); | |||
| 3775 | } | |||
| 3776 | ||||
| 3777 | static ccv_cnnp_model_t* _ccv_cnnp_move_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3778 | ||||
| 3779 | static const ccv_cnnp_model_vtab_t ccv_cnnp_move_isa = { | |||
| 3780 | .build = _ccv_cnnp_move_build, | |||
| 3781 | .copy = _ccv_cnnp_move_copy, | |||
| 3782 | }; | |||
| 3783 | ||||
| 3784 | ccv_cnnp_model_t* ccv_cnnp_move(const char* const name) | |||
| 3785 | { | |||
| 3786 | ccv_cnnp_model_move_t* const model_move = (ccv_cnnp_model_move_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_move_t)); | |||
| 3787 | model_move->super.isa = &ccv_cnnp_move_isa; | |||
| 3788 | model_move->super.input_size = 2; | |||
| 3789 | model_move->super.outputs = &model_move->output; | |||
| 3790 | model_move->super.output_size = 1; | |||
| 3791 | ccv_cnnp_model_copy_name(&model_move->super, name); | |||
| 3792 | return (ccv_cnnp_model_t*)model_move; | |||
| 3793 | } | |||
| 3794 | ||||
| 3795 | static ccv_cnnp_model_t* _ccv_cnnp_move_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3796 | { | |||
| 3797 | const ccv_cnnp_model_move_t* const self = (const ccv_cnnp_model_move_t*)super; | |||
| 3798 | return ccv_cnnp_move(self->super.name); | |||
| 3799 | } | |||
| 3800 | ||||
| 3801 | // MARK - "Making" Contiguous Layer | |||
| 3802 | ||||
| 3803 | typedef struct { | |||
| 3804 | ccv_cnnp_model_t super; | |||
| 3805 | ccv_nnc_tensor_symbol_t output; | |||
| 3806 | } ccv_cnnp_model_contiguous_t; | |||
| 3807 | ||||
| 3808 | static void _ccv_cnnp_contiguous_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3809 | { | |||
| 3810 | PRINT(CCV_CLI_VERBOSE, "[cnnp_contiguous_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_contiguous_build] -\n"); fflush(stdout); } } while (0); | |||
| 3811 | assert(input_size == 1)((void) sizeof ((input_size == 1) ? 1 : 0), __extension__ ({ if (input_size == 1) ; else __assert_fail ("input_size == 1", "ccv_cnnp_model_addons.c" , 3811, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3812 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3812, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3813 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3814 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 3815 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 3816 | { | |||
| 3817 | outputs[0] = inputs[0]; | |||
| 3818 | return; | |||
| 3819 | } | |||
| 3820 | // Otherwise, we need to check its stride to know if it is contiguous. | |||
| 3821 | int old_stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 3822 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], 0, old_stride); | |||
| 3823 | // We identify permute by checking if the stride is not in descending order. | |||
| 3824 | // This also covered "permute" through reshape, rather than using ccv_cnnp_permute directly. | |||
| 3825 | if (ccv_nnc_is_tensor_stride_packed(old_stride, params.dim)) | |||
| 3826 | { | |||
| 3827 | outputs[0] = inputs[0]; | |||
| 3828 | return; | |||
| 3829 | } | |||
| 3830 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 3831 | ccv_nnc_graph_exec_symbol_t make_contiguous = ccv_nnc_graph_exec_symbol_new(graph, CMD_FORMAT_TRANSFORM_FORWARD()ccv_nnc_cmd(CCV_NNC_FORMAT_TRANSFORM_FORWARD, 0, ccv_nnc_cmd_auto , 0), inputs, 1, outputs, 1, "contiguous"); | |||
| 3832 | ccv_nnc_graph_exec_symbol_set_flags(graph, make_contiguous, CCV_NNC_GRAPH_EXEC_DISABLE_OPT); | |||
| 3833 | } | |||
| 3834 | ||||
| 3835 | static ccv_cnnp_model_t* _ccv_cnnp_contiguous_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3836 | ||||
| 3837 | static const ccv_cnnp_model_vtab_t ccv_cnnp_contiguous_isa = { | |||
| 3838 | .build = _ccv_cnnp_contiguous_build, | |||
| 3839 | .copy = _ccv_cnnp_contiguous_copy, | |||
| 3840 | }; | |||
| 3841 | ||||
| 3842 | ccv_cnnp_model_t* ccv_cnnp_contiguous(const char* const name) | |||
| 3843 | { | |||
| 3844 | ccv_cnnp_model_contiguous_t* const model_contiguous = (ccv_cnnp_model_contiguous_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_contiguous_t)); | |||
| 3845 | model_contiguous->super.isa = &ccv_cnnp_contiguous_isa; | |||
| 3846 | model_contiguous->super.input_size = 1; | |||
| 3847 | model_contiguous->super.outputs = &model_contiguous->output; | |||
| 3848 | model_contiguous->super.output_size = 1; | |||
| 3849 | ccv_cnnp_model_copy_name(&model_contiguous->super, name); | |||
| 3850 | return (ccv_cnnp_model_t*)model_contiguous; | |||
| 3851 | } | |||
| 3852 | ||||
| 3853 | static ccv_cnnp_model_t* _ccv_cnnp_contiguous_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3854 | { | |||
| 3855 | const ccv_cnnp_model_contiguous_t* const self = (const ccv_cnnp_model_contiguous_t*)super; | |||
| 3856 | return ccv_cnnp_contiguous(self->super.name); | |||
| 3857 | } | |||
| 3858 | ||||
| 3859 | // MARK - Scaled-Dot Product Attention Layer | |||
| 3860 | ||||
| 3861 | typedef struct { | |||
| 3862 | ccv_cnnp_model_t super; | |||
| 3863 | ccv_nnc_tensor_symbol_t output; | |||
| 3864 | ccv_nnc_tensor_symbol_t weights; | |||
| 3865 | ccv_nnc_tensor_symbol_t bias; | |||
| 3866 | float scale; | |||
| 3867 | int is_causal; | |||
| 3868 | int has_attn_mask; | |||
| 3869 | int flags; | |||
| 3870 | int fused_unify_head_weights; | |||
| 3871 | int no_bias; | |||
| 3872 | } ccv_cnnp_model_scaled_dot_product_attention_t; | |||
| 3873 | ||||
| 3874 | static void _ccv_cnnp_scaled_dot_product_attention_build(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 3875 | { | |||
| 3876 | PRINT(CCV_CLI_VERBOSE, "[cnnp_scaled_dot_product_attention_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_scaled_dot_product_attention_build] -\n"); fflush (stdout); } } while (0); | |||
| 3877 | assert(input_size == 3 || input_size == 4)((void) sizeof ((input_size == 3 || input_size == 4) ? 1 : 0) , __extension__ ({ if (input_size == 3 || input_size == 4) ; else __assert_fail ("input_size == 3 || input_size == 4", "ccv_cnnp_model_addons.c" , 3877, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3878 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 3878, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3879 | ccv_cnnp_model_scaled_dot_product_attention_t* const self = (ccv_cnnp_model_scaled_dot_product_attention_t*)super; | |||
| 3880 | const ccv_nnc_tensor_param_t q_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3881 | const ccv_nnc_tensor_param_t k_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 3882 | const ccv_nnc_tensor_param_t v_params = ccv_nnc_tensor_symbol_params(graph, inputs[2]); | |||
| 3883 | const int v_nd = ccv_nnc_tensor_nd(v_params.dim); | |||
| 3884 | assert(v_nd == 3 || v_nd == 4)((void) sizeof ((v_nd == 3 || v_nd == 4) ? 1 : 0), __extension__ ({ if (v_nd == 3 || v_nd == 4) ; else __assert_fail ("v_nd == 3 || v_nd == 4" , "ccv_cnnp_model_addons.c", 3884, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3885 | const int hEv = (v_nd == 3 ? 1 : v_params.dim[2]) * v_params.dim[v_nd - 1]; | |||
| 3886 | ccv_nnc_tensor_param_t weights_params = q_params; | |||
| 3887 | memset(weights_params.dim, 0, sizeof(weights_params.dim)); | |||
| 3888 | weights_params.dim[0] = hEv; | |||
| 3889 | weights_params.dim[1] = hEv; | |||
| 3890 | ccv_nnc_tensor_param_t bias_params = q_params; | |||
| 3891 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 3892 | bias_params.dim[0] = hEv; | |||
| 3893 | ccv_nnc_cmd_t cmd = {0}; | |||
| 3894 | cmd.cmd = CCV_NNC_SCALED_DOT_PRODUCT_ATTENTION_FORWARD; | |||
| 3895 | cmd.info.scaled_dot_product_attention.scale = self->scale; | |||
| 3896 | cmd.info.scaled_dot_product_attention.is_causal = self->is_causal; | |||
| 3897 | cmd.info.scaled_dot_product_attention.flags = self->flags; | |||
| 3898 | ccv_nnc_tensor_param_t output_params[3]; | |||
| 3899 | ccv_nnc_tensor_symbol_t output; | |||
| 3900 | ccv_nnc_tensor_symbol_t saved_softmax_lse; | |||
| 3901 | ccv_nnc_tensor_symbol_t saved_v_proj = NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3902 | ccv_nnc_tensor_symbol_t attn_mask = NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3903 | ccv_nnc_tensor_symbol_t weights = NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3904 | ccv_nnc_tensor_symbol_t bias = NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3905 | if (self->has_attn_mask) | |||
| 3906 | attn_mask = inputs[3]; | |||
| 3907 | if (self->fused_unify_head_weights) | |||
| 3908 | { | |||
| 3909 | if (!self->weights.graph) | |||
| 3910 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 3911 | weights = self->weights; | |||
| 3912 | if (!self->no_bias) | |||
| 3913 | { | |||
| 3914 | if (!self->bias.graph) | |||
| 3915 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 3916 | bias = self->bias; | |||
| 3917 | } | |||
| 3918 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 3919 | q_params, | |||
| 3920 | k_params, | |||
| 3921 | v_params, | |||
| 3922 | (ccv_nnc_tensor_param_t){}, | |||
| 3923 | weights_params, | |||
| 3924 | bias_params, | |||
| 3925 | }, 6, ccv_nnc_no_hint, output_params, 3); | |||
| 3926 | output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 3927 | saved_softmax_lse = ccv_nnc_tensor_symbol_new(graph, output_params[1], 0); | |||
| 3928 | saved_v_proj = ccv_nnc_tensor_symbol_new(graph, output_params[2], 0); | |||
| 3929 | } else { | |||
| 3930 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 3931 | q_params, | |||
| 3932 | k_params, | |||
| 3933 | v_params, | |||
| 3934 | }, 3, ccv_nnc_no_hint, output_params, 2); | |||
| 3935 | output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 3936 | saved_softmax_lse = ccv_nnc_tensor_symbol_new(graph, output_params[1], 0); | |||
| 3937 | } | |||
| 3938 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], inputs[2], attn_mask, weights, bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], inputs [2], attn_mask, weights, bias}, (1 +1 +1 +1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 -1), TENSOR_SYMBOL_LIST(output, saved_softmax_lse, saved_v_proj)(const ccv_nnc_tensor_symbol_t []){output, saved_softmax_lse, saved_v_proj}, (1 +1 +1 +1 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 +0 + 0 +0 +0 +0 +0 +0 +0 +0 -1), "scaled_dot_product_attention"); | |||
| 3939 | outputs[0] = output; | |||
| 3940 | } | |||
| 3941 | ||||
| 3942 | static void _ccv_cnnp_scaled_dot_product_attention_init_states(ccv_cnnp_model_t* const super, ccv_nnc_symbolic_graph_t* const graph, const ccv_cnnp_state_initializer_f initializer, void* const context) | |||
| 3943 | { | |||
| 3944 | ccv_cnnp_model_scaled_dot_product_attention_t* const self = (ccv_cnnp_model_scaled_dot_product_attention_t*)super; | |||
| 3945 | if (self->weights.graph) | |||
| 3946 | { | |||
| 3947 | assert(self->fused_unify_head_weights)((void) sizeof ((self->fused_unify_head_weights) ? 1 : 0), __extension__ ({ if (self->fused_unify_head_weights) ; else __assert_fail ("self->fused_unify_head_weights", "ccv_cnnp_model_addons.c" , 3947, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3948 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 3949 | const int c = weight_params.dim[1]; | |||
| 3950 | const float std = sqrtf(2) / sqrtf(c); | |||
| 3951 | const float bound = sqrtf(3) * std; | |||
| 3952 | initializer(context, CMD_RANDOM_UNIFORM_FORWARD(-bound, bound)ccv_nnc_cmd(CCV_NNC_RANDOM_UNIFORM_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.blas={.a={-bound, bound}}}, 0), ccv_nnc_no_hint, 0, 0, self->weights); | |||
| 3953 | if (self->bias.graph) | |||
| 3954 | initializer(context, CMD_SET_FORWARD(0)ccv_nnc_cmd(CCV_NNC_SET_FORWARD, 0, (ccv_nnc_cmd_param_t){.size ={.dim={1,1,1}},.blas={.a={0,}}}, 0), ccv_nnc_no_hint, 0, 0, self->bias); | |||
| 3955 | } | |||
| 3956 | } | |||
| 3957 | ||||
| 3958 | static void _ccv_cnnp_scaled_dot_product_attention_add_to_parameter(ccv_cnnp_model_t* const super, const ccv_cnnp_add_to_array_f add_to_array, void* const parameters, const int is_trainable) | |||
| 3959 | { | |||
| 3960 | ccv_cnnp_model_scaled_dot_product_attention_t* const self = (ccv_cnnp_model_scaled_dot_product_attention_t*)super; | |||
| 3961 | if (self->weights.graph) | |||
| 3962 | { | |||
| 3963 | assert(self->fused_unify_head_weights)((void) sizeof ((self->fused_unify_head_weights) ? 1 : 0), __extension__ ({ if (self->fused_unify_head_weights) ; else __assert_fail ("self->fused_unify_head_weights", "ccv_cnnp_model_addons.c" , 3963, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3964 | add_to_array(parameters, self->weights, is_trainable); | |||
| 3965 | if (self->bias.graph) | |||
| 3966 | add_to_array(parameters, self->bias, is_trainable); | |||
| 3967 | } | |||
| 3968 | } | |||
| 3969 | ||||
| 3970 | static ccv_cnnp_model_t* _ccv_cnnp_scaled_dot_product_attention_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3971 | ||||
| 3972 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scaled_dot_product_attention_isa = { | |||
| 3973 | .build = _ccv_cnnp_scaled_dot_product_attention_build, | |||
| 3974 | .copy = _ccv_cnnp_scaled_dot_product_attention_copy, | |||
| 3975 | }; | |||
| 3976 | ||||
| 3977 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scaled_dot_product_attention_fused_isa = { | |||
| 3978 | .build = _ccv_cnnp_scaled_dot_product_attention_build, | |||
| 3979 | .init_states = _ccv_cnnp_scaled_dot_product_attention_init_states, | |||
| 3980 | .add_to_parameter = _ccv_cnnp_scaled_dot_product_attention_add_to_parameter, | |||
| 3981 | .copy = _ccv_cnnp_scaled_dot_product_attention_copy, | |||
| 3982 | }; | |||
| 3983 | ||||
| 3984 | ccv_cnnp_model_t* ccv_cnnp_scaled_dot_product_attention(const float scale, const int is_causal, const int has_attn_mask, const int flags, const int fused_unify_head_weights, const int no_bias, const int is_trainable, const char* const name) | |||
| 3985 | { | |||
| 3986 | ccv_cnnp_model_scaled_dot_product_attention_t* const model_scaled_dot_product_attention = (ccv_cnnp_model_scaled_dot_product_attention_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_scaled_dot_product_attention_t)); | |||
| 3987 | model_scaled_dot_product_attention->super.isa = fused_unify_head_weights ? &ccv_cnnp_scaled_dot_product_attention_fused_isa : &ccv_cnnp_scaled_dot_product_attention_isa; | |||
| 3988 | model_scaled_dot_product_attention->super.input_size = has_attn_mask ? 4 : 3; | |||
| 3989 | model_scaled_dot_product_attention->super.outputs = &model_scaled_dot_product_attention->output; | |||
| 3990 | model_scaled_dot_product_attention->super.output_size = 1; | |||
| 3991 | model_scaled_dot_product_attention->super.is_trainable = is_trainable; | |||
| 3992 | ccv_cnnp_model_copy_name(&model_scaled_dot_product_attention->super, name); | |||
| 3993 | model_scaled_dot_product_attention->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3994 | model_scaled_dot_product_attention->weights.graph = 0; | |||
| 3995 | model_scaled_dot_product_attention->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3996 | model_scaled_dot_product_attention->bias.graph = 0; | |||
| 3997 | model_scaled_dot_product_attention->scale = scale; | |||
| 3998 | model_scaled_dot_product_attention->is_causal = is_causal; | |||
| 3999 | model_scaled_dot_product_attention->has_attn_mask = has_attn_mask; | |||
| 4000 | model_scaled_dot_product_attention->flags = flags; | |||
| 4001 | model_scaled_dot_product_attention->fused_unify_head_weights = fused_unify_head_weights; | |||
| 4002 | model_scaled_dot_product_attention->no_bias = no_bias; | |||
| 4003 | return (ccv_cnnp_model_t*)model_scaled_dot_product_attention; | |||
| 4004 | } | |||
| 4005 | ||||
| 4006 | static ccv_cnnp_model_t* _ccv_cnnp_scaled_dot_product_attention_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4007 | { | |||
| 4008 | const ccv_cnnp_model_scaled_dot_product_attention_t* const self = (const ccv_cnnp_model_scaled_dot_product_attention_t*)super; | |||
| 4009 | return ccv_cnnp_scaled_dot_product_attention(self->scale, self->is_causal, self->has_attn_mask, self->flags, self->fused_unify_head_weights, self->no_bias, self->super.is_trainable, self->super.name); | |||
| 4010 | } | |||
| 4011 | ||||
| 4012 | // MARK - Debug Layer | |||
| 4013 | ||||
| 4014 | typedef struct { | |||
| 4015 | ccv_cnnp_model_t super; | |||
| 4016 | ccv_nnc_tensor_symbol_t output; | |||
| 4017 | ccv_cnnp_model_debug_f debugger; | |||
| 4018 | ccv_cnnp_model_debug_context_deinit_f debug_deinit; | |||
| 4019 | ccv_cnnp_model_debug_context_copy_f debug_copy; | |||
| 4020 | void* debug_context; | |||
| 4021 | } ccv_cnnp_model_debug_t; | |||
| 4022 | ||||
| 4023 | static int _ccv_cnnp_debug_exec(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) | |||
| 4024 | { | |||
| 4025 | if (cmd.cmd == CCV_NNC_CUSTOM_BACKWARD) | |||
| 4026 | { | |||
| 4027 | assert(0 && "don't support debug backward pass yet")((void) sizeof ((0 && "don't support debug backward pass yet" ) ? 1 : 0), __extension__ ({ if (0 && "don't support debug backward pass yet" ) ; else __assert_fail ("0 && \"don't support debug backward pass yet\"" , "ccv_cnnp_model_addons.c", 4027, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4028 | } | |||
| 4029 | ccv_cnnp_model_debug_t* const self = (ccv_cnnp_model_debug_t*)cmd.data; | |||
| 4030 | self->debugger(inputs, input_size, stream_context, self->debug_context); | |||
| 4031 | return CCV_NNC_EXEC_SUCCESS; | |||
| 4032 | } | |||
| 4033 | ||||
| 4034 | static ccv_nnc_cmd_vtab_t ccv_cnnp_debug_exec_isa = { | |||
| 4035 | .exec = _ccv_cnnp_debug_exec | |||
| 4036 | }; | |||
| 4037 | ||||
| 4038 | static void _ccv_cnnp_debug_build(ccv_cnnp_model_t* const self, ccv_nnc_symbolic_graph_t* const graph, const ccv_nnc_tensor_symbol_t* const inputs, const int input_size, ccv_nnc_tensor_symbol_t* const outputs, const int output_size) | |||
| 4039 | { | |||
| 4040 | PRINT(CCV_CLI_VERBOSE, "[cnnp_debug_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_debug_build] -\n"); fflush(stdout); } } while (0); | |||
| 4041 | assert(input_size >= 1)((void) sizeof ((input_size >= 1) ? 1 : 0), __extension__ ( { if (input_size >= 1) ; else __assert_fail ("input_size >= 1" , "ccv_cnnp_model_addons.c", 4041, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4042 | assert(output_size == 1)((void) sizeof ((output_size == 1) ? 1 : 0), __extension__ ({ if (output_size == 1) ; else __assert_fail ("output_size == 1" , "ccv_cnnp_model_addons.c", 4042, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4043 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 4044 | ccv_nnc_tensor_param_t output_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4045 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 4046 | { | |||
| 4047 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {0}; | |||
| 4048 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 4049 | ccv_nnc_tensor_get_stride(output_params.dim, stride); | |||
| 4050 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ofs, stride, output_params, 0); | |||
| 4051 | } else { | |||
| 4052 | int old_ofs[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 4053 | int old_stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 4054 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], old_ofs, old_stride); | |||
| 4055 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, to, old_ofs, old_stride, output_params, 0); | |||
| 4056 | } | |||
| 4057 | ccv_nnc_cmd_t cmd = ccv_nnc_cmd(CCV_NNC_CUSTOM_FORWARD, (ccv_nnc_cmd_vtab_t*)&ccv_cnnp_debug_exec_isa, (ccv_nnc_cmd_param_t){}, 0); | |||
| 4058 | cmd.data = self; | |||
| 4059 | ccv_nnc_graph_exec_symbol_t make_debug = ccv_nnc_graph_exec_symbol_new(graph, cmd, inputs, input_size, outputs, 1, "debug"); | |||
| 4060 | // Disable any optimizations. | |||
| 4061 | ccv_nnc_graph_exec_symbol_set_flags(graph, make_debug, CCV_NNC_GRAPH_EXEC_DISABLE_OPT); | |||
| 4062 | } | |||
| 4063 | ||||
| 4064 | static void _ccv_cnnp_debug_deinit(ccv_cnnp_model_t* const super) | |||
| 4065 | { | |||
| 4066 | const ccv_cnnp_model_debug_t* const self = (const ccv_cnnp_model_debug_t*)super; | |||
| 4067 | if (self->debug_deinit && self->debug_context) | |||
| 4068 | self->debug_deinit(self->debug_context); | |||
| 4069 | } | |||
| 4070 | ||||
| 4071 | static ccv_cnnp_model_t* _ccv_cnnp_debug_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4072 | ||||
| 4073 | static const ccv_cnnp_model_vtab_t ccv_cnnp_debug_isa = { | |||
| 4074 | .build = _ccv_cnnp_debug_build, | |||
| 4075 | .deinit = _ccv_cnnp_debug_deinit, | |||
| 4076 | .copy = _ccv_cnnp_debug_copy, | |||
| 4077 | }; | |||
| 4078 | ||||
| 4079 | ccv_cnnp_model_t* ccv_cnnp_debug(ccv_cnnp_model_debug_f func, void* const context, ccv_cnnp_model_debug_context_deinit_f deinit, ccv_cnnp_model_debug_context_copy_f copy, const char* const name) | |||
| 4080 | { | |||
| 4081 | ccv_cnnp_model_debug_t* const model_debug = (ccv_cnnp_model_debug_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_debug_t)); | |||
| 4082 | model_debug->super.isa = &ccv_cnnp_debug_isa; | |||
| 4083 | model_debug->super.input_size = 0; | |||
| 4084 | model_debug->super.outputs = &model_debug->output; | |||
| 4085 | model_debug->super.output_size = 1; | |||
| 4086 | model_debug->debugger = func; | |||
| 4087 | model_debug->debug_context = context; | |||
| 4088 | model_debug->debug_deinit = deinit; | |||
| 4089 | model_debug->debug_copy = copy; | |||
| 4090 | ccv_cnnp_model_copy_name(&model_debug->super, name); | |||
| 4091 | return (ccv_cnnp_model_t*)model_debug; | |||
| 4092 | } | |||
| 4093 | ||||
| 4094 | static ccv_cnnp_model_t* _ccv_cnnp_debug_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4095 | { | |||
| 4096 | const ccv_cnnp_model_debug_t* const self = (const ccv_cnnp_model_debug_t*)super; | |||
| 4097 | void* debug_context = self->debug_context; | |||
| 4098 | if (self->debug_copy && self->debug_context) | |||
| 4099 | debug_context = self->debug_copy(self->debug_context); | |||
| 4100 | return ccv_cnnp_debug(self->debugger, debug_context, self->debug_deinit, self->debug_copy, self->super.name); | |||
| 4101 | } |