| File: | nnc/ccv_cnnp_model_addons.c |
| Warning: | line 405, 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 | if (nd > 0) | |||
| 368 | { | |||
| 369 | assert(axis < nd)((void) sizeof ((axis < nd) ? 1 : 0), __extension__ ({ if ( axis < nd) ; else __assert_fail ("axis < nd", "ccv_cnnp_model_addons.c" , 369, __extension__ __PRETTY_FUNCTION__); })); | |||
| 370 | output_params.dim[axis] = 0; | |||
| 371 | } else | |||
| 372 | for (i = 0; i < CCV_NNC_MAX_DIM_ALLOC(12); i++) | |||
| 373 | output_params.dim[i] = 0; | |||
| 374 | int input_is_contiguous = 1; | |||
| 375 | for (i = 0; i < input_size; i++) | |||
| 376 | { | |||
| 377 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 378 | const int input_nd = ccv_nnc_tensor_nd(input_params.dim); | |||
| 379 | if (input_nd == 0) | |||
| 380 | { | |||
| 381 | 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); | |||
| 382 | input_is_contiguous = 0; | |||
| 383 | continue; | |||
| 384 | } | |||
| 385 | if (CCV_CLI_OUTPUT_LEVEL_IS(CCV_CLI_VERBOSE)(CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) | |||
| 386 | { | |||
| 387 | 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); | |||
| 388 | int i; | |||
| 389 | for (i = 1; i < CCV_NNC_MAX_DIM_ALLOC(12) && input_params.dim[i] > 0; i++) | |||
| 390 | 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); | |||
| 391 | PRINT(CCV_CLI_VERBOSE, ")\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(")\n"); fflush(stdout); } } while (0); | |||
| 392 | } | |||
| 393 | 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" , 393, __extension__ __PRETTY_FUNCTION__); })); | |||
| 394 | for (j = 0; j < nd; j++) | |||
| 395 | if (j != axis) | |||
| 396 | { 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", 396, __extension__ __PRETTY_FUNCTION__ ); })); } | |||
| 397 | output_params.dim[axis] += input_params.dim[axis]; | |||
| 398 | } | |||
| 399 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 400 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 401 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 402 | ccv_nnc_tensor_get_stride(output_params.dim, stride); | |||
| 403 | if (input_is_contiguous
| |||
| 404 | { | |||
| 405 | ccv_nnc_tensor_symbol_t aliases[input_size]; | |||
| ||||
| 406 | for (i = 0; i < input_size; i++) | |||
| 407 | { | |||
| 408 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 409 | aliases[i] = ccv_nnc_tensor_symbol_alias_new(graph, outputs[0], ofs, stride, input_params, 0); | |||
| 410 | ofs[axis] += input_params.dim[axis]; | |||
| 411 | } | |||
| 412 | // Format transform is more flexible. | |||
| 413 | 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"); | |||
| 414 | } else { | |||
| 415 | ccv_nnc_tensor_symbol_t aliases[input_size]; | |||
| 416 | for (i = 0; i < input_size; i++) | |||
| 417 | { | |||
| 418 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 419 | if (input_params.dim[0] == 0) | |||
| 420 | { | |||
| 421 | // Create a new alias anyway, but not going to use it, in this way, the alias count will match during absorb. | |||
| 422 | aliases[i] = ccv_nnc_tensor_symbol_alias_new(graph, outputs[0], ofs, stride, input_params, 0); | |||
| 423 | continue; | |||
| 424 | } | |||
| 425 | aliases[i] = ccv_nnc_tensor_symbol_alias_new(graph, outputs[0], ofs, stride, input_params, 0); | |||
| 426 | ofs[axis] += input_params.dim[axis]; | |||
| 427 | } | |||
| 428 | // Format transform is more flexible. | |||
| 429 | 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"); | |||
| 430 | } | |||
| 431 | } | |||
| 432 | ||||
| 433 | static ccv_cnnp_model_t* _ccv_cnnp_concat_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 434 | ||||
| 435 | static const ccv_cnnp_model_vtab_t ccv_cnnp_concat_isa = { | |||
| 436 | .build = _ccv_cnnp_concat_build, | |||
| 437 | .copy = _ccv_cnnp_concat_copy, | |||
| 438 | }; | |||
| 439 | ||||
| 440 | ccv_cnnp_model_t* ccv_cnnp_concat(const int axis, const char* const name) | |||
| 441 | { | |||
| 442 | ccv_cnnp_model_concat_t* const model_concat = (ccv_cnnp_model_concat_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_concat_t)); | |||
| 443 | model_concat->super.isa = &ccv_cnnp_concat_isa; | |||
| 444 | model_concat->super.input_size = 0; | |||
| 445 | model_concat->super.outputs = &model_concat->output; | |||
| 446 | model_concat->super.output_size = 1; | |||
| 447 | model_concat->axis = axis; | |||
| 448 | ccv_cnnp_model_copy_name(&model_concat->super, name); | |||
| 449 | return (ccv_cnnp_model_t*)model_concat; | |||
| 450 | } | |||
| 451 | ||||
| 452 | static ccv_cnnp_model_t* _ccv_cnnp_concat_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 453 | { | |||
| 454 | const ccv_cnnp_model_concat_t* const self = (const ccv_cnnp_model_concat_t*)super; | |||
| 455 | return ccv_cnnp_concat(self->axis, self->super.name); | |||
| 456 | } | |||
| 457 | ||||
| 458 | typedef struct { | |||
| 459 | ccv_cnnp_model_t super; | |||
| 460 | int axis; | |||
| 461 | ccv_nnc_tensor_symbol_t outputs[1]; | |||
| 462 | } ccv_cnnp_model_chunk_t; | |||
| 463 | ||||
| 464 | 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) | |||
| 465 | { | |||
| 466 | const ccv_cnnp_model_concat_t* const self = (const ccv_cnnp_model_concat_t*)super; | |||
| 467 | 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); | |||
| 468 | 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" , 468, __extension__ __PRETTY_FUNCTION__); })); | |||
| 469 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 470 | if (CCV_CLI_OUTPUT_LEVEL_IS(CCV_CLI_VERBOSE)(CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) | |||
| 471 | { | |||
| 472 | 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); | |||
| 473 | int i; | |||
| 474 | for (i = 1; i < CCV_NNC_MAX_DIM_ALLOC(12) && input_params.dim[i] > 0; i++) | |||
| 475 | 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); | |||
| 476 | PRINT(CCV_CLI_VERBOSE, ")\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(")\n"); fflush(stdout); } } while (0); | |||
| 477 | } | |||
| 478 | ccv_nnc_tensor_param_t output_params = input_params; | |||
| 479 | int i; | |||
| 480 | const int nd = ccv_nnc_tensor_nd(output_params.dim); | |||
| 481 | const int axis = self->axis; | |||
| 482 | assert(axis < nd)((void) sizeof ((axis < nd) ? 1 : 0), __extension__ ({ if ( axis < nd) ; else __assert_fail ("axis < nd", "ccv_cnnp_model_addons.c" , 482, __extension__ __PRETTY_FUNCTION__); })); | |||
| 483 | const int n = self->super.output_size; | |||
| 484 | 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", 484, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 485 | 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" , 485, __extension__ __PRETTY_FUNCTION__); })); | |||
| 486 | output_params.dim[axis] = output_params.dim[axis] / n; | |||
| 487 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 488 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 489 | ccv_nnc_tensor_get_stride(input_params.dim, stride); | |||
| 490 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 491 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 492 | { | |||
| 493 | for (i = 0; i < output_size; i++) | |||
| 494 | { | |||
| 495 | outputs[i] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ofs, stride, output_params, 0); | |||
| 496 | ofs[axis] += output_params.dim[axis]; | |||
| 497 | } | |||
| 498 | } else { | |||
| 499 | // Otherwise, we need to check if it is permute. For permute, we cannot do alias directly. | |||
| 500 | // We need to first materialize the permute and then run reshape on top of it, otherwise it will be wrong. | |||
| 501 | int old_stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 502 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], 0, old_stride); | |||
| 503 | // We identify permute by checking if the stride is not in descending order. | |||
| 504 | // This also covered "permute" through reshape, rather than using ccv_cnnp_permute directly. | |||
| 505 | int i, no_permute = 1; | |||
| 506 | for (i = 1; no_permute && i < nd; i++) | |||
| 507 | if (old_stride[i - 1] < old_stride[i]) | |||
| 508 | no_permute = 0; | |||
| 509 | if (no_permute) | |||
| 510 | { // Just straightforward reshape if there is no no permute. | |||
| 511 | for (i = 0; i < output_size; i++) | |||
| 512 | { | |||
| 513 | outputs[i] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ofs, old_stride, output_params, 0); | |||
| 514 | ofs[axis] += output_params.dim[axis]; | |||
| 515 | } | |||
| 516 | } else { | |||
| 517 | // Otherwise, we first do format transform to plain tensor and then do reshape. | |||
| 518 | ccv_nnc_tensor_symbol_t permuted = ccv_nnc_tensor_symbol_new(graph, input_params, 0); | |||
| 519 | 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"); | |||
| 520 | for (i = 0; i < output_size; i++) | |||
| 521 | { | |||
| 522 | outputs[i] = ccv_nnc_tensor_symbol_alias_new(graph, permuted, ofs, stride, output_params, 0); | |||
| 523 | ofs[axis] += output_params.dim[axis]; | |||
| 524 | } | |||
| 525 | } | |||
| 526 | } | |||
| 527 | } | |||
| 528 | ||||
| 529 | static ccv_cnnp_model_t* _ccv_cnnp_chunk_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 530 | ||||
| 531 | static const ccv_cnnp_model_vtab_t ccv_cnnp_chunk_isa = { | |||
| 532 | .build = _ccv_cnnp_chunk_build, | |||
| 533 | .copy = _ccv_cnnp_chunk_copy, | |||
| 534 | }; | |||
| 535 | ||||
| 536 | ccv_cnnp_model_t* ccv_cnnp_chunk(const int n, const int axis, const char* const name) | |||
| 537 | { | |||
| 538 | assert(n >= 1)((void) sizeof ((n >= 1) ? 1 : 0), __extension__ ({ if (n >= 1) ; else __assert_fail ("n >= 1", "ccv_cnnp_model_addons.c" , 538, __extension__ __PRETTY_FUNCTION__); })); | |||
| 539 | 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)); | |||
| 540 | model_chunk->super.isa = &ccv_cnnp_chunk_isa; | |||
| 541 | model_chunk->super.input_size = 1; | |||
| 542 | model_chunk->super.outputs = model_chunk->outputs; | |||
| 543 | model_chunk->super.output_size = n; | |||
| 544 | model_chunk->axis = axis; | |||
| 545 | ccv_cnnp_model_copy_name(&model_chunk->super, name); | |||
| 546 | return (ccv_cnnp_model_t*)model_chunk; | |||
| 547 | } | |||
| 548 | ||||
| 549 | static ccv_cnnp_model_t* _ccv_cnnp_chunk_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 550 | { | |||
| 551 | const ccv_cnnp_model_chunk_t* const self = (const ccv_cnnp_model_chunk_t*)super; | |||
| 552 | return ccv_cnnp_chunk(self->super.output_size, self->axis, self->super.name); | |||
| 553 | } | |||
| 554 | ||||
| 555 | typedef struct { | |||
| 556 | ccv_cnnp_model_t super; | |||
| 557 | ccv_nnc_tensor_symbol_t output; | |||
| 558 | int format; | |||
| 559 | int dim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 560 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 561 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 562 | } ccv_cnnp_model_reshape_t; | |||
| 563 | ||||
| 564 | 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) | |||
| 565 | { | |||
| 566 | 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" , 566, __extension__ __PRETTY_FUNCTION__); })); | |||
| 567 | 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", 567, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 568 | ccv_cnnp_model_reshape_t* const self = (ccv_cnnp_model_reshape_t*)super; | |||
| 569 | if (CCV_CLI_OUTPUT_LEVEL_IS(CCV_CLI_VERBOSE)(CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) | |||
| 570 | { | |||
| 571 | 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); | |||
| 572 | int i; | |||
| 573 | for (i = 1; i < CCV_NNC_MAX_DIM_ALLOC(12) && self->dim[i] > 0; i++) | |||
| 574 | 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); | |||
| 575 | const int count = i; | |||
| 576 | 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); | |||
| 577 | for (i = 1; i < count; i++) | |||
| 578 | 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); | |||
| 579 | 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); | |||
| 580 | for (i = 1; i < count; i++) | |||
| 581 | 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); | |||
| 582 | PRINT(CCV_CLI_VERBOSE, ")\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(")\n"); fflush(stdout); } } while (0); | |||
| 583 | } | |||
| 584 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 585 | if (params.dim[0] == 0) | |||
| 586 | { | |||
| 587 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 588 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 589 | if (ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], ofs, stride) < 0) | |||
| 590 | ccv_nnc_tensor_get_stride(params.dim, stride); | |||
| 591 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ofs, stride, params, 0); | |||
| 592 | return; | |||
| 593 | } | |||
| 594 | int dim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 595 | memcpy(dim, self->dim, sizeof(dim)); | |||
| 596 | int i, auto_idx = -1; | |||
| 597 | size_t known = 1; | |||
| 598 | const size_t tensor_count = ccv_nnc_tensor_count(params); | |||
| 599 | for (i = 0; i < CCV_NNC_MAX_DIM_ALLOC(12) && dim[i]; i++) | |||
| 600 | if (dim[i] == -1) | |||
| 601 | auto_idx = i; | |||
| 602 | else | |||
| 603 | known *= dim[i]; | |||
| 604 | if (auto_idx >= 0) | |||
| 605 | { | |||
| 606 | assert(known > 0 && tensor_count % known == 0)((void) sizeof ((known > 0 && tensor_count % known == 0) ? 1 : 0), __extension__ ({ if (known > 0 && tensor_count % known == 0) ; else __assert_fail ("known > 0 && tensor_count % known == 0" , "ccv_cnnp_model_addons.c", 606, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 607 | dim[auto_idx] = tensor_count / known; | |||
| 608 | } | |||
| 609 | if (CCV_CLI_OUTPUT_LEVEL_IS(CCV_CLI_VERBOSE)(CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) | |||
| 610 | { | |||
| 611 | 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); | |||
| 612 | int i; | |||
| 613 | for (i = 1; i < CCV_NNC_MAX_DIM_ALLOC(12) && params.dim[i] > 0; i++) | |||
| 614 | 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); | |||
| 615 | PRINT(CCV_CLI_VERBOSE, ")\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf(")\n"); fflush(stdout); } } while (0); | |||
| 616 | } | |||
| 617 | assert(ccv_nnc_dimension_count(dim) <= ccv_nnc_tensor_count(params))((void) sizeof ((ccv_nnc_dimension_count(dim) <= ccv_nnc_tensor_count (params)) ? 1 : 0), __extension__ ({ if (ccv_nnc_dimension_count (dim) <= ccv_nnc_tensor_count(params)) ; else __assert_fail ("ccv_nnc_dimension_count(dim) <= ccv_nnc_tensor_count(params)" , "ccv_cnnp_model_addons.c", 617, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 618 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 619 | int stride_from_dim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 620 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 621 | { | |||
| 622 | memcpy(params.dim, dim, sizeof(params.dim)); | |||
| 623 | int* stride; | |||
| 624 | if (self->stride[0] == 0) | |||
| 625 | { | |||
| 626 | ccv_nnc_tensor_get_stride(dim, stride_from_dim); | |||
| 627 | stride = stride_from_dim; | |||
| 628 | } else | |||
| 629 | stride = self->stride; | |||
| 630 | if (self->format > 0) | |||
| 631 | params.format = self->format; | |||
| 632 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], self->ofs, stride, params, 0); | |||
| 633 | } else { | |||
| 634 | // Otherwise, we need to check if it is permute. For permute, we cannot do alias directly. | |||
| 635 | // We need to first materialize the permute and then run reshape on top of it, otherwise it will be wrong. | |||
| 636 | int old_ofs[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 637 | int old_stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 638 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], old_ofs, old_stride); | |||
| 639 | // We identify permute by checking if the stride is not in descending order. | |||
| 640 | // This also covered "permute" through reshape, rather than using ccv_cnnp_permute directly. | |||
| 641 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 642 | const int new_nd = ccv_nnc_tensor_nd(dim); | |||
| 643 | int i, no_permute = 1; | |||
| 644 | // If the new dim has different nd, or we actually have a stride, we need to check if it is no permute or not. | |||
| 645 | if (new_nd != nd || (self->stride[0] != 0 && memcmp(self->stride, old_stride, sizeof(self->stride)))) | |||
| 646 | for (i = 1; no_permute && i < nd; i++) | |||
| 647 | if (old_stride[i - 1] < old_stride[i]) | |||
| 648 | no_permute = 0; | |||
| 649 | if (no_permute) | |||
| 650 | { // Just straightforward reshape if there is no no permute. | |||
| 651 | memcpy(params.dim, dim, sizeof(params.dim)); | |||
| 652 | int* stride; | |||
| 653 | if (self->stride[0] == 0) | |||
| 654 | { | |||
| 655 | if (new_nd != nd) // Cannot use old stride. | |||
| 656 | { | |||
| 657 | ccv_nnc_tensor_get_stride(dim, stride_from_dim); | |||
| 658 | stride = stride_from_dim; | |||
| 659 | } else | |||
| 660 | stride = old_stride; | |||
| 661 | } else | |||
| 662 | stride = self->stride; | |||
| 663 | if (self->format > 0) | |||
| 664 | params.format = self->format; | |||
| 665 | // tensor_symbol_alias_new flattens an alias to its root tensor. Preserve the | |||
| 666 | // parent view's starting position when this reshape keeps the same axes and | |||
| 667 | // strides; otherwise a chained reshape silently resets the offset to zero. | |||
| 668 | const int* ofs = self->ofs; | |||
| 669 | int new_ofs[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 670 | if (new_nd == nd && memcmp(stride, old_stride, sizeof(old_stride)) == 0) | |||
| 671 | { | |||
| 672 | for (i = 0; i < CCV_NNC_MAX_DIM_ALLOC(12); i++) | |||
| 673 | new_ofs[i] = old_ofs[i] + self->ofs[i]; | |||
| 674 | ofs = new_ofs; | |||
| 675 | } else { | |||
| 676 | for (i = 0; i < nd; i++) | |||
| 677 | { assert(old_ofs[i] == 0)((void) sizeof ((old_ofs[i] == 0) ? 1 : 0), __extension__ ({ if (old_ofs[i] == 0) ; else __assert_fail ("old_ofs[i] == 0", "ccv_cnnp_model_addons.c" , 677, __extension__ __PRETTY_FUNCTION__); })); } | |||
| 678 | } | |||
| 679 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ofs, stride, params, 0); | |||
| 680 | } else { | |||
| 681 | // Otherwise, we first do format transform to plain tensor and then do reshape. | |||
| 682 | ccv_nnc_tensor_symbol_t permuted = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 683 | 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"); | |||
| 684 | memcpy(params.dim, dim, sizeof(params.dim)); | |||
| 685 | int* stride; | |||
| 686 | if (self->stride[0] == 0) | |||
| 687 | { | |||
| 688 | ccv_nnc_tensor_get_stride(dim, stride_from_dim); | |||
| 689 | stride = stride_from_dim; | |||
| 690 | } else | |||
| 691 | stride = self->stride; | |||
| 692 | if (self->format > 0) | |||
| 693 | params.format = self->format; | |||
| 694 | // And then we create alias against the permuted one. | |||
| 695 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, permuted, self->ofs, stride, params, 0); | |||
| 696 | } | |||
| 697 | } | |||
| 698 | } | |||
| 699 | ||||
| 700 | static ccv_cnnp_model_t* _ccv_cnnp_reshape_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 701 | ||||
| 702 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reshape_isa = { | |||
| 703 | .build = _ccv_cnnp_reshape_build, | |||
| 704 | .copy = _ccv_cnnp_reshape_copy, | |||
| 705 | }; | |||
| 706 | ||||
| 707 | 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) | |||
| 708 | { | |||
| 709 | ccv_cnnp_model_reshape_t* const model_reshape = (ccv_cnnp_model_reshape_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_reshape_t)); | |||
| 710 | model_reshape->super.isa = &ccv_cnnp_reshape_isa; | |||
| 711 | model_reshape->super.input_size = 1; | |||
| 712 | model_reshape->super.outputs = &model_reshape->output; | |||
| 713 | model_reshape->super.output_size = 1; | |||
| 714 | ccv_cnnp_model_copy_name(&model_reshape->super, name); | |||
| 715 | model_reshape->format = format; | |||
| 716 | memcpy(model_reshape->dim, dim, sizeof(model_reshape->dim)); | |||
| 717 | memcpy(model_reshape->ofs, ofs, sizeof(model_reshape->ofs)); | |||
| 718 | if (stride[0] != 0) | |||
| 719 | memcpy(model_reshape->stride, stride, sizeof(model_reshape->stride)); | |||
| 720 | return (ccv_cnnp_model_t*)model_reshape; | |||
| 721 | } | |||
| 722 | ||||
| 723 | static ccv_cnnp_model_t* _ccv_cnnp_reshape_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 724 | { | |||
| 725 | const ccv_cnnp_model_reshape_t* const self = (const ccv_cnnp_model_reshape_t*)super; | |||
| 726 | return ccv_cnnp_reshape(self->format, self->dim, self->ofs, self->stride, self->super.name); | |||
| 727 | } | |||
| 728 | ||||
| 729 | typedef struct { | |||
| 730 | ccv_cnnp_model_t super; | |||
| 731 | ccv_nnc_tensor_symbol_t output; | |||
| 732 | int type; | |||
| 733 | int begin[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 734 | int end[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 735 | } ccv_cnnp_model_pad_t; | |||
| 736 | ||||
| 737 | 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) | |||
| 738 | { | |||
| 739 | 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" , 739, __extension__ __PRETTY_FUNCTION__); })); | |||
| 740 | 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", 740, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 741 | ccv_cnnp_model_pad_t* const self = (ccv_cnnp_model_pad_t*)super; | |||
| 742 | 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); | |||
| 743 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 744 | const int nd = ccv_nnc_tensor_nd(input_params.dim); | |||
| 745 | ccv_nnc_tensor_param_t params = input_params; | |||
| 746 | int i; | |||
| 747 | for (i = 0 ; i < nd; i++) | |||
| 748 | params.dim[i] += self->begin[i] + self->end[i]; | |||
| 749 | const ccv_nnc_tensor_symbol_t padded = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 750 | 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); | |||
| 751 | memcpy(pad.info.size.dim, self->begin, sizeof(pad.info.size.dim)); | |||
| 752 | memcpy(pad.info.pad.end, self->end, sizeof(pad.info.pad.end)); | |||
| 753 | 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"); | |||
| 754 | outputs[0] = padded; | |||
| 755 | } | |||
| 756 | ||||
| 757 | static ccv_cnnp_model_t* _ccv_cnnp_pad_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 758 | ||||
| 759 | static const ccv_cnnp_model_vtab_t ccv_cnnp_pad_isa = { | |||
| 760 | .build = _ccv_cnnp_pad_build, | |||
| 761 | .copy = _ccv_cnnp_pad_copy, | |||
| 762 | }; | |||
| 763 | ||||
| 764 | 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) | |||
| 765 | { | |||
| 766 | ccv_cnnp_model_pad_t* const model_pad = (ccv_cnnp_model_pad_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_pad_t)); | |||
| 767 | model_pad->super.isa = &ccv_cnnp_pad_isa; | |||
| 768 | model_pad->super.input_size = 1; | |||
| 769 | model_pad->super.outputs = &model_pad->output; | |||
| 770 | model_pad->super.output_size = 1; | |||
| 771 | ccv_cnnp_model_copy_name(&model_pad->super, name); | |||
| 772 | model_pad->type = type; | |||
| 773 | memcpy(model_pad->begin, begin, sizeof(model_pad->begin)); | |||
| 774 | memcpy(model_pad->end, end, sizeof(model_pad->end)); | |||
| 775 | return (ccv_cnnp_model_t*)model_pad; | |||
| 776 | } | |||
| 777 | ||||
| 778 | static ccv_cnnp_model_t* _ccv_cnnp_pad_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 779 | { | |||
| 780 | const ccv_cnnp_model_pad_t* const self = (const ccv_cnnp_model_pad_t*)super; | |||
| 781 | return ccv_cnnp_pad(self->type, self->begin, self->end, self->super.name); | |||
| 782 | } | |||
| 783 | ||||
| 784 | typedef struct { | |||
| 785 | ccv_cnnp_model_t super; | |||
| 786 | ccv_nnc_tensor_symbol_t output; | |||
| 787 | } ccv_cnnp_model_identity_t; | |||
| 788 | ||||
| 789 | 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) | |||
| 790 | { | |||
| 791 | 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" , 791, __extension__ __PRETTY_FUNCTION__); })); | |||
| 792 | 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", 792, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 793 | 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); | |||
| 794 | outputs[0] = inputs[0]; | |||
| 795 | } | |||
| 796 | ||||
| 797 | static ccv_cnnp_model_t* _ccv_cnnp_identity_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 798 | ||||
| 799 | static const ccv_cnnp_model_vtab_t ccv_cnnp_identity_isa = { | |||
| 800 | .build = _ccv_cnnp_identity_build, | |||
| 801 | .copy = _ccv_cnnp_identity_copy, | |||
| 802 | }; | |||
| 803 | ||||
| 804 | ccv_cnnp_model_t* ccv_cnnp_identity(const char* const name) | |||
| 805 | { | |||
| 806 | ccv_cnnp_model_identity_t* const model_identity = (ccv_cnnp_model_identity_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_identity_t)); | |||
| 807 | model_identity->super.isa = &ccv_cnnp_identity_isa; | |||
| 808 | model_identity->super.input_size = 1; | |||
| 809 | model_identity->super.outputs = &model_identity->output; | |||
| 810 | model_identity->super.output_size = 1; | |||
| 811 | ccv_cnnp_model_copy_name(&model_identity->super, name); | |||
| 812 | return (ccv_cnnp_model_t*)model_identity; | |||
| 813 | } | |||
| 814 | ||||
| 815 | static ccv_cnnp_model_t* _ccv_cnnp_identity_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 816 | { | |||
| 817 | const ccv_cnnp_model_identity_t* const self = (const ccv_cnnp_model_identity_t*)super; | |||
| 818 | return ccv_cnnp_identity(self->super.name); | |||
| 819 | } | |||
| 820 | ||||
| 821 | typedef struct { | |||
| 822 | ccv_cnnp_model_t super; | |||
| 823 | ccv_nnc_tensor_symbol_t output; | |||
| 824 | int index[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 825 | } ccv_cnnp_model_permute_t; | |||
| 826 | ||||
| 827 | 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) | |||
| 828 | { | |||
| 829 | 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" , 829, __extension__ __PRETTY_FUNCTION__); })); | |||
| 830 | 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", 830, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 831 | ccv_cnnp_model_permute_t* const self = (ccv_cnnp_model_permute_t*)super; | |||
| 832 | 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); | |||
| 833 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 834 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 835 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 836 | int input_dim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 837 | memcpy(input_dim, params.dim, sizeof(params.dim)); | |||
| 838 | int input_stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 839 | int output_stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 840 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If it is not an alias. Find stride and permute. | |||
| 841 | { | |||
| 842 | ccv_nnc_tensor_get_stride(input_dim, input_stride); | |||
| 843 | int i; | |||
| 844 | for (i = 0; i < nd; i++) | |||
| 845 | { | |||
| 846 | const int idx = self->index[i]; | |||
| 847 | 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" , 847, __extension__ __PRETTY_FUNCTION__); })); | |||
| 848 | params.dim[i] = input_dim[idx]; | |||
| 849 | output_stride[i] = input_stride[idx]; | |||
| 850 | } | |||
| 851 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ccv_nnc_no_ofs, output_stride, params, 0); | |||
| 852 | } else { | |||
| 853 | // if it is an alias, we can get the stride from it and use that. | |||
| 854 | int input_ofs[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 855 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], input_ofs, input_stride); | |||
| 856 | 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", 856, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 857 | int output_ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 858 | int i; | |||
| 859 | for (i = 0; i < nd; i++) | |||
| 860 | { | |||
| 861 | const int idx = self->index[i]; | |||
| 862 | 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" , 862, __extension__ __PRETTY_FUNCTION__); })); | |||
| 863 | params.dim[i] = input_dim[idx]; | |||
| 864 | output_stride[i] = input_stride[idx]; | |||
| 865 | output_ofs[i] = input_ofs[idx]; | |||
| 866 | } | |||
| 867 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], output_ofs, output_stride, params, 0); | |||
| 868 | } | |||
| 869 | } | |||
| 870 | ||||
| 871 | static ccv_cnnp_model_t* _ccv_cnnp_permute_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 872 | ||||
| 873 | static const ccv_cnnp_model_vtab_t ccv_cnnp_permute_isa = { | |||
| 874 | .build = _ccv_cnnp_permute_build, | |||
| 875 | .copy = _ccv_cnnp_permute_copy, | |||
| 876 | }; | |||
| 877 | ||||
| 878 | ccv_cnnp_model_t* ccv_cnnp_permute(const int index[CCV_NNC_MAX_DIM_ALLOC(12)], const char* const name) | |||
| 879 | { | |||
| 880 | ccv_cnnp_model_permute_t* const model_permute = (ccv_cnnp_model_permute_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_permute_t)); | |||
| 881 | model_permute->super.isa = &ccv_cnnp_permute_isa; | |||
| 882 | model_permute->super.input_size = 1; | |||
| 883 | model_permute->super.outputs = &model_permute->output; | |||
| 884 | model_permute->super.output_size = 1; | |||
| 885 | ccv_cnnp_model_copy_name(&model_permute->super, name); | |||
| 886 | memcpy(model_permute->index, index, sizeof(model_permute->index)); | |||
| 887 | return (ccv_cnnp_model_t*)model_permute; | |||
| 888 | } | |||
| 889 | ||||
| 890 | static ccv_cnnp_model_t* _ccv_cnnp_permute_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 891 | { | |||
| 892 | const ccv_cnnp_model_permute_t* const self = (const ccv_cnnp_model_permute_t*)super; | |||
| 893 | return ccv_cnnp_permute(self->index, self->super.name); | |||
| 894 | } | |||
| 895 | ||||
| 896 | typedef struct { | |||
| 897 | ccv_cnnp_model_t super; | |||
| 898 | int index; | |||
| 899 | ccv_nnc_tensor_symbol_t output; | |||
| 900 | } ccv_cnnp_model_extract_t; | |||
| 901 | ||||
| 902 | 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) | |||
| 903 | { | |||
| 904 | 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", 904, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 905 | ccv_cnnp_model_extract_t* const self = (ccv_cnnp_model_extract_t*)super; | |||
| 906 | 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); | |||
| 907 | outputs[0] = inputs[self->index]; | |||
| 908 | } | |||
| 909 | ||||
| 910 | static ccv_cnnp_model_t* _ccv_cnnp_extract_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 911 | ||||
| 912 | static const ccv_cnnp_model_vtab_t ccv_cnnp_extract_isa = { | |||
| 913 | .build = _ccv_cnnp_extract_build, | |||
| 914 | .copy = _ccv_cnnp_extract_copy, | |||
| 915 | }; | |||
| 916 | ||||
| 917 | ccv_cnnp_model_t* ccv_cnnp_extract(const int index, const char* const name) | |||
| 918 | { | |||
| 919 | ccv_cnnp_model_extract_t* const model_extract = (ccv_cnnp_model_extract_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_extract_t)); | |||
| 920 | model_extract->index = index; | |||
| 921 | model_extract->super.isa = &ccv_cnnp_extract_isa; | |||
| 922 | model_extract->super.input_size = 0; | |||
| 923 | model_extract->super.outputs = &model_extract->output; | |||
| 924 | model_extract->super.output_size = 1; | |||
| 925 | ccv_cnnp_model_copy_name(&model_extract->super, name); | |||
| 926 | return (ccv_cnnp_model_t*)model_extract; | |||
| 927 | } | |||
| 928 | ||||
| 929 | static ccv_cnnp_model_t* _ccv_cnnp_extract_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 930 | { | |||
| 931 | ccv_cnnp_model_extract_t* const self = (ccv_cnnp_model_extract_t*)super; | |||
| 932 | return ccv_cnnp_extract(self->index, self->super.name); | |||
| 933 | } | |||
| 934 | ||||
| 935 | typedef struct { | |||
| 936 | ccv_cnnp_model_t super; | |||
| 937 | ccv_nnc_tensor_symbol_t output; | |||
| 938 | } ccv_cnnp_model_flatten_t; | |||
| 939 | ||||
| 940 | 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) | |||
| 941 | { | |||
| 942 | 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); | |||
| 943 | 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" , 943, __extension__ __PRETTY_FUNCTION__); })); | |||
| 944 | 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", 944, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 945 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 946 | ccv_nnc_tensor_param_t output_params = params; | |||
| 947 | memset(output_params.dim, 0, sizeof(output_params.dim)); | |||
| 948 | output_params.dim[0] = ccv_nnc_tensor_get_n(params); | |||
| 949 | 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", 949, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 950 | output_params.dim[1] = ccv_nnc_tensor_count(params) / output_params.dim[0]; | |||
| 951 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)] = {}; | |||
| 952 | ccv_nnc_tensor_get_stride(output_params.dim, stride); | |||
| 953 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], DIM_ALLOC()(int [(12)]){}, stride, output_params, 0); | |||
| 954 | } | |||
| 955 | ||||
| 956 | static ccv_cnnp_model_t* _ccv_cnnp_flatten_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 957 | ||||
| 958 | static const ccv_cnnp_model_vtab_t ccv_cnnp_flatten_isa = { | |||
| 959 | .build = _ccv_cnnp_flatten_build, | |||
| 960 | .copy = _ccv_cnnp_flatten_copy, | |||
| 961 | }; | |||
| 962 | ||||
| 963 | ccv_cnnp_model_t* ccv_cnnp_flatten(const char* const name) | |||
| 964 | { | |||
| 965 | ccv_cnnp_model_flatten_t* const model_flatten = (ccv_cnnp_model_flatten_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_flatten_t)); | |||
| 966 | model_flatten->super.isa = &ccv_cnnp_flatten_isa; | |||
| 967 | model_flatten->super.input_size = 1; | |||
| 968 | model_flatten->super.outputs = &model_flatten->output; | |||
| 969 | model_flatten->super.output_size = 1; | |||
| 970 | ccv_cnnp_model_copy_name(&model_flatten->super, name); | |||
| 971 | return (ccv_cnnp_model_t*)model_flatten; | |||
| 972 | } | |||
| 973 | ||||
| 974 | static ccv_cnnp_model_t* _ccv_cnnp_flatten_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 975 | { | |||
| 976 | return ccv_cnnp_flatten(self->name); | |||
| 977 | } | |||
| 978 | ||||
| 979 | // MARK - Batch Norm Layer | |||
| 980 | ||||
| 981 | typedef struct { | |||
| 982 | ccv_cnnp_model_t super; | |||
| 983 | ccv_nnc_tensor_symbol_t output; | |||
| 984 | ccv_nnc_tensor_symbol_t bias; | |||
| 985 | ccv_nnc_tensor_symbol_t scale; | |||
| 986 | ccv_nnc_graph_exec_symbol_t batch_norm; | |||
| 987 | ccv_nnc_cmd_param_t params; | |||
| 988 | ccv_array_t* zero_inits; | |||
| 989 | ccv_array_t* retainables; | |||
| 990 | } ccv_cnnp_model_batch_norm_t; | |||
| 991 | ||||
| 992 | 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) | |||
| 993 | { | |||
| 994 | 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" , 994, __extension__ __PRETTY_FUNCTION__); })); | |||
| 995 | 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", 995, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 996 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 997 | 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); | |||
| 998 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 999 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 1000 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 1001 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 1002 | // If the accuracy is not enough, bump it to 32-bit floating point. | |||
| 1003 | if (bias_params.datatype != CCV_32F && bias_params.datatype != CCV_64F) | |||
| 1004 | bias_params.datatype = CCV_32F; | |||
| 1005 | bias_params.dim[0] = nd > 1 ? ccv_nnc_tensor_get_c(params) : params.dim[0]; | |||
| 1006 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 1007 | // Both scale and bias are shared between if this model is reused. | |||
| 1008 | if (!self->scale.graph) | |||
| 1009 | self->scale = ccv_nnc_tensor_symbol_new(graph, bias_params, "scale"); | |||
| 1010 | if (!self->bias.graph) | |||
| 1011 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 1012 | const ccv_nnc_tensor_symbol_t scale = ccv_cnnp_model_get_symbol(super, self->scale); | |||
| 1013 | const ccv_nnc_tensor_symbol_t bias = ccv_cnnp_model_get_symbol(super, self->bias); | |||
| 1014 | const ccv_nnc_tensor_symbol_t mean = ccv_nnc_tensor_symbol_new(graph, bias_params, "mean"); | |||
| 1015 | const ccv_nnc_tensor_symbol_t var = ccv_nnc_tensor_symbol_new(graph, bias_params, "var"); | |||
| 1016 | // Otherwise, notice mean, var, saved_mean, saved_inv_std are not reused. | |||
| 1017 | if (!self->zero_inits) | |||
| 1018 | self->zero_inits = ccv_array_new(sizeof(ccv_nnc_tensor_symbol_t), 0, 0); | |||
| 1019 | ccv_array_push(self->zero_inits, &mean); | |||
| 1020 | ccv_array_push(self->zero_inits, &var); | |||
| 1021 | const ccv_nnc_tensor_symbol_t out_mean = ccv_nnc_tensor_symbol_new(graph, bias_params, "out_mean"); | |||
| 1022 | const ccv_nnc_tensor_symbol_t out_var = ccv_nnc_tensor_symbol_new(graph, bias_params, "out_var"); | |||
| 1023 | if (!self->retainables) | |||
| 1024 | self->retainables = ccv_array_new(sizeof(ccv_nnc_tensor_symbol_t), 0, 0); | |||
| 1025 | ccv_array_push(self->retainables, &out_mean); | |||
| 1026 | ccv_array_push(self->retainables, &out_var); | |||
| 1027 | const ccv_nnc_tensor_symbol_t saved_mean = ccv_nnc_tensor_symbol_new(graph, bias_params, "saved_mean"); | |||
| 1028 | const ccv_nnc_tensor_symbol_t saved_inv_std = ccv_nnc_tensor_symbol_new(graph, bias_params, "saved_inv_std"); | |||
| 1029 | const int hw = ccv_nnc_tensor_hw(params, ccv_nnc_tensor_nd(params.dim), CCV_NNC_MAX_DIM(2)); | |||
| 1030 | ccv_nnc_cmd_param_t batch_norm = self->params; | |||
| 1031 | batch_norm.bnorm.count = hw >= 0 ? CCV_NNC_MAX_DIM(2) + 1 : 1; | |||
| 1032 | int i; | |||
| 1033 | batch_norm.bnorm.axis[0] = (params.format == CCV_TENSOR_FORMAT_CHWN) ? 3 : 0; | |||
| 1034 | if (hw >= 0) | |||
| 1035 | for (i = 0; i < CCV_NNC_MAX_DIM(2); i++) | |||
| 1036 | batch_norm.bnorm.axis[i + 1] = i + hw; | |||
| 1037 | self->params = batch_norm; | |||
| 1038 | 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], scale, bias, mean, var)(const ccv_nnc_tensor_symbol_t []){inputs[0], scale, 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"); | |||
| 1039 | outputs[0] = output; | |||
| 1040 | } | |||
| 1041 | ||||
| 1042 | 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) | |||
| 1043 | { | |||
| 1044 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1045 | if (self->scale.graph) | |||
| 1046 | 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); | |||
| 1047 | if (self->bias.graph) | |||
| 1048 | 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); | |||
| 1049 | int i; | |||
| 1050 | if (self->zero_inits) | |||
| 1051 | for (i = 0; i < self->zero_inits->rnum; i++) | |||
| 1052 | 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)))); | |||
| 1053 | } | |||
| 1054 | ||||
| 1055 | 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) | |||
| 1056 | { | |||
| 1057 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1058 | if (self->scale.graph) | |||
| 1059 | add_to_array(parameters, self->scale, is_trainable); | |||
| 1060 | if (self->bias.graph) | |||
| 1061 | add_to_array(parameters, self->bias, is_trainable); | |||
| 1062 | } | |||
| 1063 | ||||
| 1064 | 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) | |||
| 1065 | { | |||
| 1066 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1067 | int i; | |||
| 1068 | if (self->retainables) | |||
| 1069 | for (i = 0; i < self->retainables->rnum; i++) | |||
| 1070 | { | |||
| 1071 | 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))); | |||
| 1072 | add_to_array(outputs, symbol, 0); | |||
| 1073 | } | |||
| 1074 | } | |||
| 1075 | ||||
| 1076 | 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) | |||
| 1077 | { | |||
| 1078 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1079 | if (self->batch_norm.graph) | |||
| 1080 | { | |||
| 1081 | self->params.bnorm.is_test = is_test; | |||
| 1082 | updater(context, self->batch_norm, ccv_nnc_cmd(CCV_NNC_BATCH_NORM_FORWARD, 0, self->params, 0), ccv_nnc_no_hint); | |||
| 1083 | } | |||
| 1084 | } | |||
| 1085 | ||||
| 1086 | static void _ccv_cnnp_batch_norm_deinit(ccv_cnnp_model_t* const super) | |||
| 1087 | { | |||
| 1088 | ccv_cnnp_model_batch_norm_t* const self = (ccv_cnnp_model_batch_norm_t*)super; | |||
| 1089 | if (self->zero_inits) | |||
| 1090 | ccv_array_free(self->zero_inits); | |||
| 1091 | if (self->retainables) | |||
| 1092 | ccv_array_free(self->retainables); | |||
| 1093 | } | |||
| 1094 | ||||
| 1095 | static ccv_cnnp_model_t* _ccv_cnnp_batch_norm_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1096 | ||||
| 1097 | static const ccv_cnnp_model_vtab_t ccv_cnnp_batch_norm_isa = { | |||
| 1098 | .build = _ccv_cnnp_batch_norm_build, | |||
| 1099 | .init_states = _ccv_cnnp_batch_norm_init_states, | |||
| 1100 | .add_to_parameter = _ccv_cnnp_batch_norm_add_to_parameter, | |||
| 1101 | .add_to_output = _ccv_cnnp_batch_norm_add_to_output, | |||
| 1102 | .copy = _ccv_cnnp_batch_norm_copy, | |||
| 1103 | .set_is_test = _ccv_cnnp_batch_norm_set_is_test, | |||
| 1104 | .deinit = _ccv_cnnp_batch_norm_deinit, | |||
| 1105 | }; | |||
| 1106 | ||||
| 1107 | ccv_cnnp_model_t* ccv_cnnp_batch_norm(const float momentum, const float epsilon, const int is_trainable, const char* const name) | |||
| 1108 | { | |||
| 1109 | 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)); | |||
| 1110 | model_batch_norm->super.isa = &ccv_cnnp_batch_norm_isa; | |||
| 1111 | model_batch_norm->super.input_size = 1; | |||
| 1112 | model_batch_norm->super.outputs = &model_batch_norm->output; | |||
| 1113 | model_batch_norm->super.output_size = 1; | |||
| 1114 | model_batch_norm->super.is_trainable = is_trainable; | |||
| 1115 | ccv_cnnp_model_copy_name(&model_batch_norm->super, name); | |||
| 1116 | model_batch_norm->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1117 | model_batch_norm->scale.graph = 0; | |||
| 1118 | model_batch_norm->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1119 | model_batch_norm->bias.graph = 0; | |||
| 1120 | model_batch_norm->params.bnorm.momentum = momentum; | |||
| 1121 | model_batch_norm->params.bnorm.epsilon = epsilon; | |||
| 1122 | return (ccv_cnnp_model_t*)model_batch_norm; | |||
| 1123 | } | |||
| 1124 | ||||
| 1125 | static ccv_cnnp_model_t* _ccv_cnnp_batch_norm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1126 | { | |||
| 1127 | const ccv_cnnp_model_batch_norm_t* const self = (const ccv_cnnp_model_batch_norm_t*)super; | |||
| 1128 | return ccv_cnnp_batch_norm(self->params.bnorm.momentum, self->params.bnorm.epsilon, self->super.is_trainable, self->super.name); | |||
| 1129 | } | |||
| 1130 | ||||
| 1131 | // MARK - Convolution Layer | |||
| 1132 | ||||
| 1133 | typedef struct { | |||
| 1134 | ccv_cnnp_model_t super; | |||
| 1135 | ccv_nnc_tensor_symbol_t output; | |||
| 1136 | ccv_nnc_tensor_symbol_t weights; | |||
| 1137 | ccv_nnc_tensor_symbol_t bias; | |||
| 1138 | int groups; | |||
| 1139 | int filters; | |||
| 1140 | int kdim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1141 | int dilation[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1142 | int no_bias; | |||
| 1143 | int format; | |||
| 1144 | ccv_nnc_hint_t hint; | |||
| 1145 | } ccv_cnnp_model_convolution_t; | |||
| 1146 | ||||
| 1147 | 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) | |||
| 1148 | { | |||
| 1149 | ccv_cnnp_model_convolution_t* const self = (ccv_cnnp_model_convolution_t*)super; | |||
| 1150 | 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); | |||
| 1151 | 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" , 1151, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1152 | 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", 1152, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1153 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1154 | int i; | |||
| 1155 | const int k_nd = ccv_nnc_tensor_nd(self->kdim); | |||
| 1156 | const int nd = k_nd + 2; | |||
| 1157 | ccv_nnc_tensor_param_t weights_params = params; | |||
| 1158 | if (self->format) | |||
| 1159 | weights_params.format = self->format; | |||
| 1160 | ccv_nnc_tensor_set_n(&weights_params, self->filters); | |||
| 1161 | const int a_nd = ccv_nnc_tensor_nd(params.dim); | |||
| 1162 | int c; | |||
| 1163 | switch (params.format) | |||
| 1164 | { | |||
| 1165 | case CCV_TENSOR_FORMAT_NHWC: | |||
| 1166 | c = params.dim[a_nd - 1]; | |||
| 1167 | break; | |||
| 1168 | case CCV_TENSOR_FORMAT_NCHW: | |||
| 1169 | if (a_nd == k_nd + 1) | |||
| 1170 | c = params.dim[0]; | |||
| 1171 | else | |||
| 1172 | c = params.dim[a_nd <= 1 ? 0 : 1]; | |||
| 1173 | break; | |||
| 1174 | case CCV_TENSOR_FORMAT_CHWN: | |||
| 1175 | c = params.dim[0]; | |||
| 1176 | break; | |||
| 1177 | } | |||
| 1178 | 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", 1178, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1179 | ccv_nnc_tensor_set_c(&weights_params, nd, c / self->groups); | |||
| 1180 | int hw = -1; | |||
| 1181 | if (weights_params.format == CCV_TENSOR_FORMAT_NHWC || weights_params.format == CCV_TENSOR_FORMAT_CHWN) | |||
| 1182 | hw = 1; | |||
| 1183 | else if (weights_params.format == CCV_TENSOR_FORMAT_NCHW) | |||
| 1184 | hw = 2; | |||
| 1185 | assert(hw >= 0)((void) sizeof ((hw >= 0) ? 1 : 0), __extension__ ({ if (hw >= 0) ; else __assert_fail ("hw >= 0", "ccv_cnnp_model_addons.c" , 1185, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1186 | for (i = 0; i < k_nd; i++) | |||
| 1187 | weights_params.dim[i + hw] = self->kdim[i]; | |||
| 1188 | if (!self->weights.graph) | |||
| 1189 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 1190 | 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" , 1190, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1191 | const ccv_nnc_tensor_symbol_t weights = ccv_cnnp_model_get_symbol(super, self->weights); | |||
| 1192 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 1193 | if (self->format) | |||
| 1194 | bias_params.format = self->format; | |||
| 1195 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 1196 | bias_params.dim[0] = self->filters; | |||
| 1197 | 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); | |||
| 1198 | for (i = 0; i < k_nd; i++) | |||
| 1199 | cmd.info.size.dim[i] = self->kdim[i]; | |||
| 1200 | cmd.info.size.dim[k_nd] = c; | |||
| 1201 | memcpy(cmd.info.convolution.dilation, self->dilation, sizeof(self->dilation)); | |||
| 1202 | ccv_nnc_tensor_param_t output_params; | |||
| 1203 | // Dilate weight size based on the dilation factor. | |||
| 1204 | for (i = 0; i < k_nd; i++) | |||
| 1205 | 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; | |||
| 1206 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 1207 | params, | |||
| 1208 | weights_params, | |||
| 1209 | bias_params, | |||
| 1210 | }, 3, self->hint, &output_params, 1); | |||
| 1211 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1212 | ccv_nnc_graph_exec_symbol_t convolution; | |||
| 1213 | if (self->no_bias) | |||
| 1214 | convolution = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], weights)(const ccv_nnc_tensor_symbol_t []){inputs[0], 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"); | |||
| 1215 | else { | |||
| 1216 | if (!self->bias.graph) | |||
| 1217 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 1218 | const ccv_nnc_tensor_symbol_t bias = ccv_cnnp_model_get_symbol(super, self->bias); | |||
| 1219 | convolution = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], weights, bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], weights, 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"); | |||
| 1220 | } | |||
| 1221 | ccv_nnc_graph_exec_symbol_set_hint(graph, convolution, self->hint); | |||
| 1222 | outputs[0] = output; | |||
| 1223 | } | |||
| 1224 | ||||
| 1225 | 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) | |||
| 1226 | { | |||
| 1227 | ccv_cnnp_model_convolution_t* const self = (ccv_cnnp_model_convolution_t*)super; | |||
| 1228 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 1229 | 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 ; }); | |||
| 1230 | const int count = ccv_nnc_tensor_count(weight_params); | |||
| 1231 | const float std = sqrtf(2) / sqrtf(count / n); | |||
| 1232 | const float bound = sqrtf(3) * std; | |||
| 1233 | 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); | |||
| 1234 | if (self->bias.graph) | |||
| 1235 | 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); | |||
| 1236 | } | |||
| 1237 | ||||
| 1238 | 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) | |||
| 1239 | { | |||
| 1240 | ccv_cnnp_model_convolution_t* const self = (ccv_cnnp_model_convolution_t*)super; | |||
| 1241 | add_to_array(parameters, self->weights, is_trainable); | |||
| 1242 | if (self->bias.graph) | |||
| 1243 | add_to_array(parameters, self->bias, is_trainable); | |||
| 1244 | } | |||
| 1245 | ||||
| 1246 | static ccv_cnnp_model_t* _ccv_cnnp_convolution_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1247 | ||||
| 1248 | static const ccv_cnnp_model_vtab_t ccv_cnnp_convolution_isa = { | |||
| 1249 | .build = _ccv_cnnp_convolution_build, | |||
| 1250 | .init_states = _ccv_cnnp_convolution_init_states, | |||
| 1251 | .add_to_parameter = _ccv_cnnp_convolution_add_to_parameter, | |||
| 1252 | .copy = _ccv_cnnp_convolution_copy, | |||
| 1253 | }; | |||
| 1254 | ||||
| 1255 | 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) | |||
| 1256 | { | |||
| 1257 | ccv_cnnp_model_convolution_t* const model_convolution = (ccv_cnnp_model_convolution_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_convolution_t)); | |||
| 1258 | model_convolution->super.isa = &ccv_cnnp_convolution_isa; | |||
| 1259 | model_convolution->super.input_size = 1; | |||
| 1260 | model_convolution->super.outputs = &model_convolution->output; | |||
| 1261 | model_convolution->super.output_size = 1; | |||
| 1262 | model_convolution->super.is_trainable = is_trainable; | |||
| 1263 | ccv_cnnp_model_copy_name(&model_convolution->super, name); | |||
| 1264 | model_convolution->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1265 | model_convolution->weights.graph = 0; | |||
| 1266 | model_convolution->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1267 | model_convolution->bias.graph = 0; | |||
| 1268 | model_convolution->groups = groups; | |||
| 1269 | model_convolution->filters = filters; | |||
| 1270 | memcpy(model_convolution->kdim, kdim, sizeof(model_convolution->kdim)); | |||
| 1271 | memcpy(model_convolution->dilation, dilation, sizeof(model_convolution->dilation)); | |||
| 1272 | model_convolution->no_bias = no_bias; | |||
| 1273 | model_convolution->hint = hint; | |||
| 1274 | model_convolution->format = format; | |||
| 1275 | return (ccv_cnnp_model_t*)model_convolution; | |||
| 1276 | } | |||
| 1277 | ||||
| 1278 | static ccv_cnnp_model_t* _ccv_cnnp_convolution_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1279 | { | |||
| 1280 | ccv_cnnp_model_convolution_t* const self = (ccv_cnnp_model_convolution_t*)super; | |||
| 1281 | 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); | |||
| 1282 | } | |||
| 1283 | ||||
| 1284 | // MARK - Convolution Transpose Layer | |||
| 1285 | ||||
| 1286 | typedef struct { | |||
| 1287 | ccv_cnnp_model_t super; | |||
| 1288 | ccv_nnc_tensor_symbol_t output; | |||
| 1289 | ccv_nnc_tensor_symbol_t weights; | |||
| 1290 | ccv_nnc_tensor_symbol_t bias; | |||
| 1291 | int groups; | |||
| 1292 | int filters; | |||
| 1293 | int kdim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1294 | int dilation[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1295 | int output_padding; | |||
| 1296 | int no_bias; | |||
| 1297 | int format; | |||
| 1298 | ccv_nnc_hint_t hint; | |||
| 1299 | } ccv_cnnp_model_convolution_transpose_t; | |||
| 1300 | ||||
| 1301 | 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) | |||
| 1302 | { | |||
| 1303 | ccv_cnnp_model_convolution_transpose_t* const self = (ccv_cnnp_model_convolution_transpose_t*)super; | |||
| 1304 | 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); | |||
| 1305 | 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" , 1305, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1306 | 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", 1306, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1307 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1308 | int i; | |||
| 1309 | const int nd = CCV_NNC_MAX_DIM(2) + 2; | |||
| 1310 | ccv_nnc_tensor_param_t weights_params = params; | |||
| 1311 | if (self->format) | |||
| 1312 | weights_params.format = self->format; | |||
| 1313 | const int c = ccv_nnc_tensor_get_c(params); | |||
| 1314 | ccv_nnc_tensor_set_n(&weights_params, c); | |||
| 1315 | 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", 1315, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1316 | ccv_nnc_tensor_set_c(&weights_params, nd, self->filters / self->groups); | |||
| 1317 | const int hw = ccv_nnc_tensor_hw(weights_params, nd, CCV_NNC_MAX_DIM(2)); | |||
| 1318 | assert(hw >= 0)((void) sizeof ((hw >= 0) ? 1 : 0), __extension__ ({ if (hw >= 0) ; else __assert_fail ("hw >= 0", "ccv_cnnp_model_addons.c" , 1318, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1319 | for (i = 0; i < CCV_NNC_MAX_DIM(2); i++) | |||
| 1320 | weights_params.dim[i + hw] = self->kdim[i]; | |||
| 1321 | if (!self->weights.graph) | |||
| 1322 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 1323 | 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" , 1323, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1324 | const ccv_nnc_tensor_symbol_t weights = ccv_cnnp_model_get_symbol(super, self->weights); | |||
| 1325 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 1326 | if (self->format) | |||
| 1327 | bias_params.format = self->format; | |||
| 1328 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 1329 | bias_params.dim[0] = self->filters; | |||
| 1330 | 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); | |||
| 1331 | for (i = 0; i < CCV_NNC_MAX_DIM(2); i++) | |||
| 1332 | cmd.info.size.dim[i] = self->kdim[i]; | |||
| 1333 | cmd.info.size.dim[CCV_NNC_MAX_DIM(2)] = c; | |||
| 1334 | memcpy(cmd.info.convolution_transpose.dilation, self->dilation, sizeof(self->dilation)); | |||
| 1335 | ccv_nnc_tensor_param_t output_params; | |||
| 1336 | // Dilate weight size based on the dilation factor. | |||
| 1337 | for (i = 0; i < CCV_NNC_MAX_DIM(2); i++) | |||
| 1338 | 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; | |||
| 1339 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 1340 | params, | |||
| 1341 | weights_params, | |||
| 1342 | bias_params, | |||
| 1343 | }, 3, self->hint, &output_params, 1); | |||
| 1344 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1345 | ccv_nnc_graph_exec_symbol_t convolution_transpose; | |||
| 1346 | if (self->no_bias) | |||
| 1347 | convolution_transpose = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], weights)(const ccv_nnc_tensor_symbol_t []){inputs[0], 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"); | |||
| 1348 | else { | |||
| 1349 | if (!self->bias.graph) | |||
| 1350 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 1351 | const ccv_nnc_tensor_symbol_t bias = ccv_cnnp_model_get_symbol(super, self->bias); | |||
| 1352 | convolution_transpose = ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], weights, bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], weights, 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"); | |||
| 1353 | } | |||
| 1354 | ccv_nnc_graph_exec_symbol_set_hint(graph, convolution_transpose, self->hint); | |||
| 1355 | outputs[0] = output; | |||
| 1356 | } | |||
| 1357 | ||||
| 1358 | 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) | |||
| 1359 | { | |||
| 1360 | ccv_cnnp_model_convolution_transpose_t* const self = (ccv_cnnp_model_convolution_transpose_t*)super; | |||
| 1361 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 1362 | 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 ; }); | |||
| 1363 | const int count = ccv_nnc_tensor_count(weight_params); | |||
| 1364 | const float std = sqrtf(2) / sqrtf(count / n); | |||
| 1365 | const float bound = sqrtf(3) * std; | |||
| 1366 | 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); | |||
| 1367 | if (self->bias.graph) | |||
| 1368 | 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); | |||
| 1369 | } | |||
| 1370 | ||||
| 1371 | 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) | |||
| 1372 | { | |||
| 1373 | ccv_cnnp_model_convolution_transpose_t* const self = (ccv_cnnp_model_convolution_transpose_t*)super; | |||
| 1374 | add_to_array(parameters, self->weights, is_trainable); | |||
| 1375 | if (self->bias.graph) | |||
| 1376 | add_to_array(parameters, self->bias, is_trainable); | |||
| 1377 | } | |||
| 1378 | ||||
| 1379 | static ccv_cnnp_model_t* _ccv_cnnp_convolution_transpose_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1380 | ||||
| 1381 | static const ccv_cnnp_model_vtab_t ccv_cnnp_convolution_transpose_isa = { | |||
| 1382 | .build = _ccv_cnnp_convolution_transpose_build, | |||
| 1383 | .init_states = _ccv_cnnp_convolution_transpose_init_states, | |||
| 1384 | .add_to_parameter = _ccv_cnnp_convolution_transpose_add_to_parameter, | |||
| 1385 | .copy = _ccv_cnnp_convolution_transpose_copy, | |||
| 1386 | }; | |||
| 1387 | ||||
| 1388 | 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) | |||
| 1389 | { | |||
| 1390 | 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)); | |||
| 1391 | model_convolution_transpose->super.isa = &ccv_cnnp_convolution_transpose_isa; | |||
| 1392 | model_convolution_transpose->super.input_size = 1; | |||
| 1393 | model_convolution_transpose->super.outputs = &model_convolution_transpose->output; | |||
| 1394 | model_convolution_transpose->super.output_size = 1; | |||
| 1395 | model_convolution_transpose->super.is_trainable = is_trainable; | |||
| 1396 | ccv_cnnp_model_copy_name(&model_convolution_transpose->super, name); | |||
| 1397 | model_convolution_transpose->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1398 | model_convolution_transpose->weights.graph = 0; | |||
| 1399 | model_convolution_transpose->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1400 | model_convolution_transpose->bias.graph = 0; | |||
| 1401 | model_convolution_transpose->groups = groups; | |||
| 1402 | model_convolution_transpose->filters = filters; | |||
| 1403 | memcpy(model_convolution_transpose->kdim, kdim, sizeof(model_convolution_transpose->kdim)); | |||
| 1404 | memcpy(model_convolution_transpose->dilation, dilation, sizeof(model_convolution_transpose->dilation)); | |||
| 1405 | model_convolution_transpose->output_padding = output_padding; | |||
| 1406 | model_convolution_transpose->no_bias = no_bias; | |||
| 1407 | model_convolution_transpose->hint = hint; | |||
| 1408 | model_convolution_transpose->format = format; | |||
| 1409 | return (ccv_cnnp_model_t*)model_convolution_transpose; | |||
| 1410 | } | |||
| 1411 | ||||
| 1412 | static ccv_cnnp_model_t* _ccv_cnnp_convolution_transpose_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1413 | { | |||
| 1414 | ccv_cnnp_model_convolution_transpose_t* const self = (ccv_cnnp_model_convolution_transpose_t*)super; | |||
| 1415 | 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); | |||
| 1416 | } | |||
| 1417 | ||||
| 1418 | // MARK - Dense Layer | |||
| 1419 | ||||
| 1420 | typedef struct { | |||
| 1421 | ccv_cnnp_model_t super; | |||
| 1422 | ccv_nnc_tensor_symbol_t output; | |||
| 1423 | ccv_nnc_tensor_symbol_t weights; | |||
| 1424 | ccv_nnc_tensor_symbol_t bias; | |||
| 1425 | int count; | |||
| 1426 | int no_bias; | |||
| 1427 | int flags; | |||
| 1428 | } ccv_cnnp_model_dense_t; | |||
| 1429 | ||||
| 1430 | 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) | |||
| 1431 | { | |||
| 1432 | ccv_cnnp_model_dense_t* const self = (ccv_cnnp_model_dense_t*)super; | |||
| 1433 | 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); | |||
| 1434 | 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" , 1434, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1435 | 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", 1435, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1436 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1437 | ccv_nnc_tensor_param_t weights_params = params; | |||
| 1438 | memset(weights_params.dim, 0, sizeof(weights_params.dim)); | |||
| 1439 | if (params.dim[0] != 0) | |||
| 1440 | { | |||
| 1441 | weights_params.dim[0] = self->count; | |||
| 1442 | weights_params.dim[1] = params.dim[ccv_nnc_tensor_nd(params.dim) - 1]; | |||
| 1443 | } | |||
| 1444 | if (!self->weights.graph) | |||
| 1445 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 1446 | 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" , 1446, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1447 | const ccv_nnc_tensor_symbol_t weights = ccv_cnnp_model_get_symbol(super, self->weights); | |||
| 1448 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 1449 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 1450 | bias_params.dim[0] = self->count; | |||
| 1451 | ccv_nnc_cmd_t cmd = {0}; | |||
| 1452 | cmd.cmd = CCV_NNC_GEMM_FORWARD; | |||
| 1453 | cmd.info.blas.a[0] = 1; | |||
| 1454 | cmd.info.blas.a[1] = 1; | |||
| 1455 | cmd.info.blas.transpose_b[0] = 0; | |||
| 1456 | cmd.info.blas.transpose_b[1] = 1; | |||
| 1457 | cmd.info.blas.flags = self->flags; | |||
| 1458 | ccv_nnc_tensor_param_t output_params; | |||
| 1459 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 1460 | params, | |||
| 1461 | weights_params, | |||
| 1462 | bias_params, | |||
| 1463 | }, 3, ccv_nnc_no_hint, &output_params, 1); | |||
| 1464 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1465 | if (self->no_bias) | |||
| 1466 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], weights)(const ccv_nnc_tensor_symbol_t []){inputs[0], 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"); | |||
| 1467 | else { | |||
| 1468 | if (!self->bias.graph) | |||
| 1469 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 1470 | const ccv_nnc_tensor_symbol_t bias = ccv_cnnp_model_get_symbol(super, self->bias); | |||
| 1471 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], weights, bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], weights, 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"); | |||
| 1472 | } | |||
| 1473 | outputs[0] = output; | |||
| 1474 | } | |||
| 1475 | ||||
| 1476 | 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) | |||
| 1477 | { | |||
| 1478 | ccv_cnnp_model_dense_t* const self = (ccv_cnnp_model_dense_t*)super; | |||
| 1479 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 1480 | const int c = weight_params.dim[1]; | |||
| 1481 | const float std = sqrtf(2) / sqrtf(c); | |||
| 1482 | const float bound = sqrtf(3) * std; | |||
| 1483 | 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); | |||
| 1484 | if (self->bias.graph) | |||
| 1485 | 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); | |||
| 1486 | } | |||
| 1487 | ||||
| 1488 | 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) | |||
| 1489 | { | |||
| 1490 | ccv_cnnp_model_dense_t* const self = (ccv_cnnp_model_dense_t*)super; | |||
| 1491 | add_to_array(parameters, self->weights, is_trainable); | |||
| 1492 | if (self->bias.graph) | |||
| 1493 | add_to_array(parameters, self->bias, is_trainable); | |||
| 1494 | } | |||
| 1495 | ||||
| 1496 | static ccv_cnnp_model_t* _ccv_cnnp_dense_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1497 | ||||
| 1498 | static const ccv_cnnp_model_vtab_t ccv_cnnp_dense_isa = { | |||
| 1499 | .build = _ccv_cnnp_dense_build, | |||
| 1500 | .init_states = _ccv_cnnp_dense_init_states, | |||
| 1501 | .add_to_parameter = _ccv_cnnp_dense_add_to_parameter, | |||
| 1502 | .copy = _ccv_cnnp_dense_copy, | |||
| 1503 | }; | |||
| 1504 | ||||
| 1505 | 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) | |||
| 1506 | { | |||
| 1507 | ccv_cnnp_model_dense_t* const model_dense = (ccv_cnnp_model_dense_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_dense_t)); | |||
| 1508 | model_dense->super.isa = &ccv_cnnp_dense_isa; | |||
| 1509 | model_dense->super.input_size = 1; | |||
| 1510 | model_dense->super.outputs = &model_dense->output; | |||
| 1511 | model_dense->super.output_size = 1; | |||
| 1512 | model_dense->super.is_trainable = is_trainable; | |||
| 1513 | ccv_cnnp_model_copy_name(&model_dense->super, name); | |||
| 1514 | model_dense->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1515 | model_dense->weights.graph = 0; | |||
| 1516 | model_dense->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 1517 | model_dense->bias.graph = 0; | |||
| 1518 | model_dense->count = count; | |||
| 1519 | model_dense->no_bias = no_bias; | |||
| 1520 | model_dense->flags = flags; | |||
| 1521 | return (ccv_cnnp_model_t*)model_dense; | |||
| 1522 | } | |||
| 1523 | ||||
| 1524 | static ccv_cnnp_model_t* _ccv_cnnp_dense_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1525 | { | |||
| 1526 | const ccv_cnnp_model_dense_t* const self = (const ccv_cnnp_model_dense_t*)super; | |||
| 1527 | return ccv_cnnp_dense(self->count, self->no_bias, self->flags, self->super.is_trainable, self->super.name); | |||
| 1528 | } | |||
| 1529 | ||||
| 1530 | // MARK - Pool Layers | |||
| 1531 | ||||
| 1532 | typedef struct { | |||
| 1533 | ccv_cnnp_model_t super; | |||
| 1534 | ccv_nnc_tensor_symbol_t output; | |||
| 1535 | int kdim[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 1536 | ccv_nnc_hint_t hint; | |||
| 1537 | } ccv_cnnp_model_pool_t; | |||
| 1538 | ||||
| 1539 | 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) | |||
| 1540 | { | |||
| 1541 | ccv_cnnp_model_pool_t* const self = (ccv_cnnp_model_pool_t*)super; | |||
| 1542 | 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); | |||
| 1543 | 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" , 1543, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1544 | 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", 1544, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1545 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1546 | const int hw = ccv_nnc_tensor_hw(params, ccv_nnc_tensor_nd(params.dim), CCV_NNC_MAX_DIM(2)); | |||
| 1547 | ccv_nnc_cmd_t cmd; | |||
| 1548 | if (hw >= 0 && self->kdim[0] == 0 && self->kdim[1] == 0) | |||
| 1549 | 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); | |||
| 1550 | else | |||
| 1551 | 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); | |||
| 1552 | ccv_nnc_tensor_param_t output_params; | |||
| 1553 | ccv_nnc_hint_tensor_auto(cmd, ¶ms, 1, self->hint, &output_params, 1); | |||
| 1554 | const ccv_nnc_tensor_symbol_t pool_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1555 | 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"); | |||
| 1556 | ccv_nnc_graph_exec_symbol_set_hint(graph, exec, self->hint); | |||
| 1557 | outputs[0] = pool_output; | |||
| 1558 | } | |||
| 1559 | ||||
| 1560 | static ccv_cnnp_model_t* _ccv_cnnp_max_pool_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1561 | ||||
| 1562 | static const ccv_cnnp_model_vtab_t ccv_cnnp_max_pool_isa = { | |||
| 1563 | .build = _ccv_cnnp_max_pool_build, | |||
| 1564 | .copy = _ccv_cnnp_max_pool_copy, | |||
| 1565 | }; | |||
| 1566 | ||||
| 1567 | 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) | |||
| 1568 | { | |||
| 1569 | ccv_cnnp_model_pool_t* const model_pool = (ccv_cnnp_model_pool_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_pool_t)); | |||
| 1570 | model_pool->super.isa = &ccv_cnnp_max_pool_isa; | |||
| 1571 | model_pool->super.input_size = 1; | |||
| 1572 | model_pool->super.outputs = &model_pool->output; | |||
| 1573 | model_pool->super.output_size = 1; | |||
| 1574 | ccv_cnnp_model_copy_name(&model_pool->super, name); | |||
| 1575 | memcpy(model_pool->kdim, kdim, sizeof(model_pool->kdim)); | |||
| 1576 | model_pool->hint = hint; | |||
| 1577 | return (ccv_cnnp_model_t*)model_pool; | |||
| 1578 | } | |||
| 1579 | ||||
| 1580 | static ccv_cnnp_model_t* _ccv_cnnp_max_pool_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1581 | { | |||
| 1582 | const ccv_cnnp_model_pool_t* const self = (const ccv_cnnp_model_pool_t*)super; | |||
| 1583 | return ccv_cnnp_max_pool(self->kdim, self->hint, self->super.name); | |||
| 1584 | } | |||
| 1585 | ||||
| 1586 | 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) | |||
| 1587 | { | |||
| 1588 | ccv_cnnp_model_pool_t* const self = (ccv_cnnp_model_pool_t*)super; | |||
| 1589 | 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); | |||
| 1590 | 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" , 1590, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1591 | 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", 1591, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1592 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1593 | const int hw = ccv_nnc_tensor_hw(params, ccv_nnc_tensor_nd(params.dim), CCV_NNC_MAX_DIM(2)); | |||
| 1594 | ccv_nnc_cmd_t cmd; | |||
| 1595 | if (hw >= 0 && self->kdim[0] == 0 && self->kdim[1] == 0) | |||
| 1596 | 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); | |||
| 1597 | else | |||
| 1598 | 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); | |||
| 1599 | ccv_nnc_tensor_param_t output_params; | |||
| 1600 | ccv_nnc_hint_tensor_auto(cmd, ¶ms, 1, self->hint, &output_params, 1); | |||
| 1601 | const ccv_nnc_tensor_symbol_t pool_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1602 | 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"); | |||
| 1603 | ccv_nnc_graph_exec_symbol_set_hint(graph, exec, self->hint); | |||
| 1604 | outputs[0] = pool_output; | |||
| 1605 | } | |||
| 1606 | ||||
| 1607 | static ccv_cnnp_model_t* _ccv_cnnp_average_pool_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 1608 | ||||
| 1609 | static const ccv_cnnp_model_vtab_t ccv_cnnp_average_pool_isa = { | |||
| 1610 | .build = _ccv_cnnp_average_pool_build, | |||
| 1611 | .copy = _ccv_cnnp_average_pool_copy, | |||
| 1612 | }; | |||
| 1613 | ||||
| 1614 | 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) | |||
| 1615 | { | |||
| 1616 | ccv_cnnp_model_pool_t* const model_pool = (ccv_cnnp_model_pool_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_pool_t)); | |||
| 1617 | model_pool->super.isa = &ccv_cnnp_average_pool_isa; | |||
| 1618 | model_pool->super.input_size = 1; | |||
| 1619 | model_pool->super.outputs = &model_pool->output; | |||
| 1620 | model_pool->super.output_size = 1; | |||
| 1621 | ccv_cnnp_model_copy_name(&model_pool->super, name); | |||
| 1622 | memcpy(model_pool->kdim, kdim, sizeof(model_pool->kdim)); | |||
| 1623 | model_pool->hint = hint; | |||
| 1624 | return (ccv_cnnp_model_t*)model_pool; | |||
| 1625 | } | |||
| 1626 | ||||
| 1627 | static ccv_cnnp_model_t* _ccv_cnnp_average_pool_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1628 | { | |||
| 1629 | const ccv_cnnp_model_pool_t* const self = (const ccv_cnnp_model_pool_t*)super; | |||
| 1630 | return ccv_cnnp_average_pool(self->kdim, self->hint, self->super.name); | |||
| 1631 | } | |||
| 1632 | ||||
| 1633 | // MARK - RELU Layer | |||
| 1634 | ||||
| 1635 | typedef struct { | |||
| 1636 | ccv_cnnp_model_t super; | |||
| 1637 | ccv_nnc_tensor_symbol_t output; | |||
| 1638 | } ccv_cnnp_model_relu_t; | |||
| 1639 | ||||
| 1640 | 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) | |||
| 1641 | { | |||
| 1642 | 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); | |||
| 1643 | 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" , 1643, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1644 | 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", 1644, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1645 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1646 | ccv_nnc_tensor_param_t output_params; | |||
| 1647 | const ccv_nnc_cmd_t relu = CMD_RELU_FORWARD()ccv_nnc_cmd(CCV_NNC_RELU_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1648 | ccv_nnc_hint_tensor_auto(relu, (ccv_nnc_tensor_param_t []){ | |||
| 1649 | params, | |||
| 1650 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1651 | const ccv_nnc_tensor_symbol_t relu_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1652 | 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"); | |||
| 1653 | outputs[0] = relu_output; | |||
| 1654 | } | |||
| 1655 | ||||
| 1656 | static ccv_cnnp_model_t* _ccv_cnnp_relu_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1657 | ||||
| 1658 | static const ccv_cnnp_model_vtab_t ccv_cnnp_relu_isa = { | |||
| 1659 | .build = _ccv_cnnp_relu_build, | |||
| 1660 | .copy = _ccv_cnnp_relu_copy, | |||
| 1661 | }; | |||
| 1662 | ||||
| 1663 | ccv_cnnp_model_t* ccv_cnnp_relu(const char* const name) | |||
| 1664 | { | |||
| 1665 | ccv_cnnp_model_relu_t* const model_relu = (ccv_cnnp_model_relu_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_relu_t)); | |||
| 1666 | model_relu->super.isa = &ccv_cnnp_relu_isa; | |||
| 1667 | model_relu->super.input_size = 1; | |||
| 1668 | model_relu->super.outputs = &model_relu->output; | |||
| 1669 | model_relu->super.output_size = 1; | |||
| 1670 | ccv_cnnp_model_copy_name(&model_relu->super, name); | |||
| 1671 | return (ccv_cnnp_model_t*)model_relu; | |||
| 1672 | } | |||
| 1673 | ||||
| 1674 | static ccv_cnnp_model_t* _ccv_cnnp_relu_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1675 | { | |||
| 1676 | return ccv_cnnp_relu(self->name); | |||
| 1677 | } | |||
| 1678 | ||||
| 1679 | // MARK - Sigmoid Layer | |||
| 1680 | ||||
| 1681 | typedef struct { | |||
| 1682 | ccv_cnnp_model_t super; | |||
| 1683 | ccv_nnc_tensor_symbol_t output; | |||
| 1684 | } ccv_cnnp_model_sigmoid_t; | |||
| 1685 | ||||
| 1686 | 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) | |||
| 1687 | { | |||
| 1688 | 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); | |||
| 1689 | 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" , 1689, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1690 | 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", 1690, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1691 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1692 | ccv_nnc_tensor_param_t output_params; | |||
| 1693 | const ccv_nnc_cmd_t sigmoid = CMD_SIGMOID_FORWARD()ccv_nnc_cmd(CCV_NNC_SIGMOID_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1694 | ccv_nnc_hint_tensor_auto(sigmoid, (ccv_nnc_tensor_param_t []){ | |||
| 1695 | params, | |||
| 1696 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1697 | const ccv_nnc_tensor_symbol_t sigmoid_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1698 | 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"); | |||
| 1699 | outputs[0] = sigmoid_output; | |||
| 1700 | } | |||
| 1701 | ||||
| 1702 | static ccv_cnnp_model_t* _ccv_cnnp_sigmoid_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1703 | ||||
| 1704 | static const ccv_cnnp_model_vtab_t ccv_cnnp_sigmoid_isa = { | |||
| 1705 | .build = _ccv_cnnp_sigmoid_build, | |||
| 1706 | .copy = _ccv_cnnp_sigmoid_copy, | |||
| 1707 | }; | |||
| 1708 | ||||
| 1709 | ccv_cnnp_model_t* ccv_cnnp_sigmoid(const char* const name) | |||
| 1710 | { | |||
| 1711 | ccv_cnnp_model_sigmoid_t* const model_sigmoid = (ccv_cnnp_model_sigmoid_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_sigmoid_t)); | |||
| 1712 | model_sigmoid->super.isa = &ccv_cnnp_sigmoid_isa; | |||
| 1713 | model_sigmoid->super.input_size = 1; | |||
| 1714 | model_sigmoid->super.outputs = &model_sigmoid->output; | |||
| 1715 | model_sigmoid->super.output_size = 1; | |||
| 1716 | ccv_cnnp_model_copy_name(&model_sigmoid->super, name); | |||
| 1717 | return (ccv_cnnp_model_t*)model_sigmoid; | |||
| 1718 | } | |||
| 1719 | ||||
| 1720 | static ccv_cnnp_model_t* _ccv_cnnp_sigmoid_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1721 | { | |||
| 1722 | return ccv_cnnp_sigmoid(self->name); | |||
| 1723 | } | |||
| 1724 | ||||
| 1725 | // MARK - Tanh Layer | |||
| 1726 | ||||
| 1727 | typedef struct { | |||
| 1728 | ccv_cnnp_model_t super; | |||
| 1729 | ccv_nnc_tensor_symbol_t output; | |||
| 1730 | } ccv_cnnp_model_tanh_t; | |||
| 1731 | ||||
| 1732 | 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) | |||
| 1733 | { | |||
| 1734 | 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); | |||
| 1735 | 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" , 1735, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1736 | 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", 1736, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1737 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1738 | ccv_nnc_tensor_param_t output_params; | |||
| 1739 | const ccv_nnc_cmd_t tanh = CMD_TANH_FORWARD()ccv_nnc_cmd(CCV_NNC_TANH_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1740 | ccv_nnc_hint_tensor_auto(tanh, (ccv_nnc_tensor_param_t []){ | |||
| 1741 | params, | |||
| 1742 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1743 | const ccv_nnc_tensor_symbol_t tanh_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1744 | 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"); | |||
| 1745 | outputs[0] = tanh_output; | |||
| 1746 | } | |||
| 1747 | ||||
| 1748 | static ccv_cnnp_model_t* _ccv_cnnp_tanh_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1749 | ||||
| 1750 | static const ccv_cnnp_model_vtab_t ccv_cnnp_tanh_isa = { | |||
| 1751 | .build = _ccv_cnnp_tanh_build, | |||
| 1752 | .copy = _ccv_cnnp_tanh_copy, | |||
| 1753 | }; | |||
| 1754 | ||||
| 1755 | ccv_cnnp_model_t* ccv_cnnp_tanh(const char* const name) | |||
| 1756 | { | |||
| 1757 | ccv_cnnp_model_tanh_t* const model_tanh = (ccv_cnnp_model_tanh_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_tanh_t)); | |||
| 1758 | model_tanh->super.isa = &ccv_cnnp_tanh_isa; | |||
| 1759 | model_tanh->super.input_size = 1; | |||
| 1760 | model_tanh->super.outputs = &model_tanh->output; | |||
| 1761 | model_tanh->super.output_size = 1; | |||
| 1762 | ccv_cnnp_model_copy_name(&model_tanh->super, name); | |||
| 1763 | return (ccv_cnnp_model_t*)model_tanh; | |||
| 1764 | } | |||
| 1765 | ||||
| 1766 | static ccv_cnnp_model_t* _ccv_cnnp_tanh_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1767 | { | |||
| 1768 | return ccv_cnnp_tanh(self->name); | |||
| 1769 | } | |||
| 1770 | ||||
| 1771 | // MARK - Exp Layer | |||
| 1772 | ||||
| 1773 | typedef struct { | |||
| 1774 | ccv_cnnp_model_t super; | |||
| 1775 | ccv_nnc_tensor_symbol_t output; | |||
| 1776 | } ccv_cnnp_model_exp_t; | |||
| 1777 | ||||
| 1778 | static void _ccv_cnnp_exp_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) | |||
| 1779 | { | |||
| 1780 | PRINT(CCV_CLI_VERBOSE, "[cnnp_exp_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_exp_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 1781 | 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" , 1781, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1782 | 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", 1782, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1783 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1784 | ccv_nnc_tensor_param_t output_params; | |||
| 1785 | const ccv_nnc_cmd_t exp = CMD_EWEXP_FORWARD()ccv_nnc_cmd(CCV_NNC_EWEXP_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 1786 | ccv_nnc_hint_tensor_auto(exp, (ccv_nnc_tensor_param_t []){ | |||
| 1787 | params, | |||
| 1788 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1789 | const ccv_nnc_tensor_symbol_t exp_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1790 | ccv_nnc_graph_exec_symbol_new(graph, exp, 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(exp_output)(const ccv_nnc_tensor_symbol_t []){exp_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), "exp"); | |||
| 1791 | outputs[0] = exp_output; | |||
| 1792 | } | |||
| 1793 | ||||
| 1794 | static ccv_cnnp_model_t* _ccv_cnnp_exp_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1795 | ||||
| 1796 | static const ccv_cnnp_model_vtab_t ccv_cnnp_exp_isa = { | |||
| 1797 | .build = _ccv_cnnp_exp_build, | |||
| 1798 | .copy = _ccv_cnnp_exp_copy, | |||
| 1799 | }; | |||
| 1800 | ||||
| 1801 | ccv_cnnp_model_t* ccv_cnnp_exp(const char* const name) | |||
| 1802 | { | |||
| 1803 | ccv_cnnp_model_exp_t* const model_exp = (ccv_cnnp_model_exp_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_exp_t)); | |||
| 1804 | model_exp->super.isa = &ccv_cnnp_exp_isa; | |||
| 1805 | model_exp->super.input_size = 1; | |||
| 1806 | model_exp->super.outputs = &model_exp->output; | |||
| 1807 | model_exp->super.output_size = 1; | |||
| 1808 | ccv_cnnp_model_copy_name(&model_exp->super, name); | |||
| 1809 | return (ccv_cnnp_model_t*)model_exp; | |||
| 1810 | } | |||
| 1811 | ||||
| 1812 | static ccv_cnnp_model_t* _ccv_cnnp_exp_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1813 | { | |||
| 1814 | return ccv_cnnp_exp(self->name); | |||
| 1815 | } | |||
| 1816 | ||||
| 1817 | // MARK - Softplus Layer | |||
| 1818 | ||||
| 1819 | typedef struct { | |||
| 1820 | ccv_cnnp_model_t super; | |||
| 1821 | ccv_nnc_tensor_symbol_t output; | |||
| 1822 | } ccv_cnnp_model_softplus_t; | |||
| 1823 | ||||
| 1824 | static void _ccv_cnnp_softplus_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) | |||
| 1825 | { | |||
| 1826 | PRINT(CCV_CLI_VERBOSE, "[cnnp_softplus_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_softplus_build] -\n"); fflush(stdout); } } while (0); | |||
| 1827 | 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" , 1827, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1828 | 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", 1828, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1829 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1830 | ccv_nnc_tensor_param_t output_params; | |||
| 1831 | const ccv_nnc_cmd_t softplus = CMD_EWSOFTPLUS_FORWARD()ccv_nnc_cmd(CCV_NNC_EWSOFTPLUS_FORWARD, 0, ccv_nnc_cmd_auto, 0 ); | |||
| 1832 | ccv_nnc_hint_tensor_auto(softplus, (ccv_nnc_tensor_param_t []){ | |||
| 1833 | params, | |||
| 1834 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1835 | const ccv_nnc_tensor_symbol_t softplus_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1836 | ccv_nnc_graph_exec_symbol_new(graph, softplus, 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(softplus_output)(const ccv_nnc_tensor_symbol_t []){softplus_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), "softplus"); | |||
| 1837 | outputs[0] = softplus_output; | |||
| 1838 | } | |||
| 1839 | ||||
| 1840 | static ccv_cnnp_model_t* _ccv_cnnp_softplus_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1841 | ||||
| 1842 | static const ccv_cnnp_model_vtab_t ccv_cnnp_softplus_isa = { | |||
| 1843 | .build = _ccv_cnnp_softplus_build, | |||
| 1844 | .copy = _ccv_cnnp_softplus_copy, | |||
| 1845 | }; | |||
| 1846 | ||||
| 1847 | ccv_cnnp_model_t* ccv_cnnp_softplus(const char* const name) | |||
| 1848 | { | |||
| 1849 | ccv_cnnp_model_softplus_t* const model_softplus = (ccv_cnnp_model_softplus_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_softplus_t)); | |||
| 1850 | model_softplus->super.isa = &ccv_cnnp_softplus_isa; | |||
| 1851 | model_softplus->super.input_size = 1; | |||
| 1852 | model_softplus->super.outputs = &model_softplus->output; | |||
| 1853 | model_softplus->super.output_size = 1; | |||
| 1854 | ccv_cnnp_model_copy_name(&model_softplus->super, name); | |||
| 1855 | return (ccv_cnnp_model_t*)model_softplus; | |||
| 1856 | } | |||
| 1857 | ||||
| 1858 | static ccv_cnnp_model_t* _ccv_cnnp_softplus_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1859 | { | |||
| 1860 | return ccv_cnnp_softplus(self->name); | |||
| 1861 | } | |||
| 1862 | ||||
| 1863 | // MARK - Swish Layer | |||
| 1864 | ||||
| 1865 | typedef struct { | |||
| 1866 | ccv_cnnp_model_t super; | |||
| 1867 | ccv_nnc_tensor_symbol_t output; | |||
| 1868 | float beta; | |||
| 1869 | } ccv_cnnp_model_swish_t; | |||
| 1870 | ||||
| 1871 | 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) | |||
| 1872 | { | |||
| 1873 | 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); | |||
| 1874 | 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" , 1874, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1875 | 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", 1875, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1876 | ccv_cnnp_model_swish_t* const self = (ccv_cnnp_model_swish_t*)super; | |||
| 1877 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1878 | ccv_nnc_tensor_param_t output_params; | |||
| 1879 | const ccv_nnc_cmd_t swish = CMD_SWISH_FORWARD(self->beta)ccv_nnc_cmd(CCV_NNC_SWISH_FORWARD, 0, ((ccv_nnc_cmd_param_t){ .size={.dim={1,1,1}},.swish={.beta=self->beta}}), 0); | |||
| 1880 | ccv_nnc_hint_tensor_auto(swish, (ccv_nnc_tensor_param_t []){ | |||
| 1881 | params, | |||
| 1882 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1883 | const ccv_nnc_tensor_symbol_t swish_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1884 | 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"); | |||
| 1885 | outputs[0] = swish_output; | |||
| 1886 | } | |||
| 1887 | ||||
| 1888 | static ccv_cnnp_model_t* _ccv_cnnp_swish_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1889 | ||||
| 1890 | static const ccv_cnnp_model_vtab_t ccv_cnnp_swish_isa = { | |||
| 1891 | .build = _ccv_cnnp_swish_build, | |||
| 1892 | .copy = _ccv_cnnp_swish_copy, | |||
| 1893 | }; | |||
| 1894 | ||||
| 1895 | ccv_cnnp_model_t* ccv_cnnp_swish(const float beta, const char* const name) | |||
| 1896 | { | |||
| 1897 | ccv_cnnp_model_swish_t* const model_swish = (ccv_cnnp_model_swish_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_swish_t)); | |||
| 1898 | model_swish->super.isa = &ccv_cnnp_swish_isa; | |||
| 1899 | model_swish->super.input_size = 1; | |||
| 1900 | model_swish->super.outputs = &model_swish->output; | |||
| 1901 | model_swish->super.output_size = 1; | |||
| 1902 | model_swish->beta = beta; | |||
| 1903 | ccv_cnnp_model_copy_name(&model_swish->super, name); | |||
| 1904 | return (ccv_cnnp_model_t*)model_swish; | |||
| 1905 | } | |||
| 1906 | ||||
| 1907 | static ccv_cnnp_model_t* _ccv_cnnp_swish_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 1908 | { | |||
| 1909 | const ccv_cnnp_model_swish_t* const swish = (const ccv_cnnp_model_swish_t*)self; | |||
| 1910 | return ccv_cnnp_swish(swish->beta, self->name); | |||
| 1911 | } | |||
| 1912 | ||||
| 1913 | // MARK - Swish Mul Layer | |||
| 1914 | ||||
| 1915 | typedef struct { | |||
| 1916 | ccv_cnnp_model_t super; | |||
| 1917 | ccv_nnc_tensor_symbol_t output; | |||
| 1918 | float beta; | |||
| 1919 | float scale; | |||
| 1920 | } ccv_cnnp_model_swish_mul_t; | |||
| 1921 | ||||
| 1922 | static void _ccv_cnnp_swish_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) | |||
| 1923 | { | |||
| 1924 | PRINT(CCV_CLI_VERBOSE, "[cnnp_swish_mul_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_swish_mul_build] -\n"); fflush(stdout); } } while (0); | |||
| 1925 | 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" , 1925, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1926 | 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", 1926, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1927 | const ccv_cnnp_model_swish_mul_t* const self = (const ccv_cnnp_model_swish_mul_t*)super; | |||
| 1928 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 1929 | int i; | |||
| 1930 | for (i = 0; i < 2; i++) | |||
| 1931 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 1932 | ccv_nnc_tensor_param_t output_params; | |||
| 1933 | const ccv_nnc_cmd_t swish_mul = CMD_SWISH_MUL_FORWARD(self->beta, self->scale)ccv_nnc_cmd(CCV_NNC_SWISH_MUL_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.swish_mul={.beta=self->beta,.scale =self->scale}}), 0); | |||
| 1934 | ccv_nnc_hint_tensor_auto(swish_mul, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 1935 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1936 | ccv_nnc_graph_exec_symbol_new(graph, swish_mul, inputs, input_size, outputs, output_size, "swish_mul"); | |||
| 1937 | } | |||
| 1938 | ||||
| 1939 | static ccv_cnnp_model_t* _ccv_cnnp_swish_mul_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1940 | ||||
| 1941 | static const ccv_cnnp_model_vtab_t ccv_cnnp_swish_mul_isa = { | |||
| 1942 | .build = _ccv_cnnp_swish_mul_build, | |||
| 1943 | .copy = _ccv_cnnp_swish_mul_copy, | |||
| 1944 | }; | |||
| 1945 | ||||
| 1946 | ccv_cnnp_model_t* ccv_cnnp_swish_mul(const float beta, const float scale, const char* const name) | |||
| 1947 | { | |||
| 1948 | ccv_cnnp_model_swish_mul_t* const model_swish_mul = (ccv_cnnp_model_swish_mul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_swish_mul_t)); | |||
| 1949 | model_swish_mul->super.isa = &ccv_cnnp_swish_mul_isa; | |||
| 1950 | model_swish_mul->super.input_size = 2; | |||
| 1951 | model_swish_mul->super.outputs = &model_swish_mul->output; | |||
| 1952 | model_swish_mul->super.output_size = 1; | |||
| 1953 | model_swish_mul->beta = beta; | |||
| 1954 | model_swish_mul->scale = scale; | |||
| 1955 | ccv_cnnp_model_copy_name(&model_swish_mul->super, name); | |||
| 1956 | return (ccv_cnnp_model_t*)model_swish_mul; | |||
| 1957 | } | |||
| 1958 | ||||
| 1959 | static ccv_cnnp_model_t* _ccv_cnnp_swish_mul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 1960 | { | |||
| 1961 | const ccv_cnnp_model_swish_mul_t* const self = (const ccv_cnnp_model_swish_mul_t*)super; | |||
| 1962 | return ccv_cnnp_swish_mul(self->beta, self->scale, self->super.name); | |||
| 1963 | } | |||
| 1964 | ||||
| 1965 | // MARK - GELU Layer | |||
| 1966 | ||||
| 1967 | typedef struct { | |||
| 1968 | ccv_cnnp_model_t super; | |||
| 1969 | ccv_nnc_tensor_symbol_t output; | |||
| 1970 | int tanh; | |||
| 1971 | } ccv_cnnp_model_gelu_t; | |||
| 1972 | ||||
| 1973 | 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) | |||
| 1974 | { | |||
| 1975 | 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); | |||
| 1976 | 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" , 1976, __extension__ __PRETTY_FUNCTION__); })); | |||
| 1977 | 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", 1977, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 1978 | ccv_cnnp_model_gelu_t* const self = (ccv_cnnp_model_gelu_t*)super; | |||
| 1979 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 1980 | ccv_nnc_tensor_param_t output_params; | |||
| 1981 | 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); | |||
| 1982 | ccv_nnc_hint_tensor_auto(gelu, (ccv_nnc_tensor_param_t []){ | |||
| 1983 | params, | |||
| 1984 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 1985 | const ccv_nnc_tensor_symbol_t gelu_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 1986 | 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"); | |||
| 1987 | outputs[0] = gelu_output; | |||
| 1988 | } | |||
| 1989 | ||||
| 1990 | static ccv_cnnp_model_t* _ccv_cnnp_gelu_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 1991 | ||||
| 1992 | static const ccv_cnnp_model_vtab_t ccv_cnnp_gelu_isa = { | |||
| 1993 | .build = _ccv_cnnp_gelu_build, | |||
| 1994 | .copy = _ccv_cnnp_gelu_copy, | |||
| 1995 | }; | |||
| 1996 | ||||
| 1997 | ccv_cnnp_model_t* ccv_cnnp_gelu(const int tanh, const char* const name) | |||
| 1998 | { | |||
| 1999 | ccv_cnnp_model_gelu_t* const model_gelu = (ccv_cnnp_model_gelu_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_gelu_t)); | |||
| 2000 | model_gelu->super.isa = &ccv_cnnp_gelu_isa; | |||
| 2001 | model_gelu->super.input_size = 1; | |||
| 2002 | model_gelu->super.outputs = &model_gelu->output; | |||
| 2003 | model_gelu->super.output_size = 1; | |||
| 2004 | model_gelu->tanh = tanh; | |||
| 2005 | ccv_cnnp_model_copy_name(&model_gelu->super, name); | |||
| 2006 | return (ccv_cnnp_model_t*)model_gelu; | |||
| 2007 | } | |||
| 2008 | ||||
| 2009 | static ccv_cnnp_model_t* _ccv_cnnp_gelu_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2010 | { | |||
| 2011 | ccv_cnnp_model_gelu_t* const self = (ccv_cnnp_model_gelu_t*)super; | |||
| 2012 | return ccv_cnnp_gelu(self->tanh, self->super.name); | |||
| 2013 | } | |||
| 2014 | ||||
| 2015 | // MARK - Leaky ReLU Layer | |||
| 2016 | ||||
| 2017 | typedef struct { | |||
| 2018 | ccv_cnnp_model_t super; | |||
| 2019 | ccv_nnc_tensor_symbol_t output; | |||
| 2020 | float negative_slope; | |||
| 2021 | } ccv_cnnp_model_leaky_relu_t; | |||
| 2022 | ||||
| 2023 | 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) | |||
| 2024 | { | |||
| 2025 | 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); | |||
| 2026 | 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" , 2026, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2027 | 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", 2027, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2028 | ccv_cnnp_model_leaky_relu_t* const self = (ccv_cnnp_model_leaky_relu_t*)super; | |||
| 2029 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2030 | ccv_nnc_tensor_param_t output_params; | |||
| 2031 | 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); | |||
| 2032 | ccv_nnc_hint_tensor_auto(leaky_relu, (ccv_nnc_tensor_param_t []){ | |||
| 2033 | params, | |||
| 2034 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2035 | const ccv_nnc_tensor_symbol_t leaky_relu_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2036 | 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"); | |||
| 2037 | outputs[0] = leaky_relu_output; | |||
| 2038 | } | |||
| 2039 | ||||
| 2040 | static ccv_cnnp_model_t* _ccv_cnnp_leaky_relu_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2041 | ||||
| 2042 | static const ccv_cnnp_model_vtab_t ccv_cnnp_leaky_relu_isa = { | |||
| 2043 | .build = _ccv_cnnp_leaky_relu_build, | |||
| 2044 | .copy = _ccv_cnnp_leaky_relu_copy, | |||
| 2045 | }; | |||
| 2046 | ||||
| 2047 | ccv_cnnp_model_t* ccv_cnnp_leaky_relu(const float negative_slope, const char* const name) | |||
| 2048 | { | |||
| 2049 | 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)); | |||
| 2050 | model_leaky_relu->super.isa = &ccv_cnnp_leaky_relu_isa; | |||
| 2051 | model_leaky_relu->super.input_size = 1; | |||
| 2052 | model_leaky_relu->super.outputs = &model_leaky_relu->output; | |||
| 2053 | model_leaky_relu->super.output_size = 1; | |||
| 2054 | model_leaky_relu->negative_slope = negative_slope; | |||
| 2055 | ccv_cnnp_model_copy_name(&model_leaky_relu->super, name); | |||
| 2056 | return (ccv_cnnp_model_t*)model_leaky_relu; | |||
| 2057 | } | |||
| 2058 | ||||
| 2059 | static ccv_cnnp_model_t* _ccv_cnnp_leaky_relu_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2060 | { | |||
| 2061 | ccv_cnnp_model_leaky_relu_t* const self = (ccv_cnnp_model_leaky_relu_t*)super; | |||
| 2062 | return ccv_cnnp_leaky_relu(self->negative_slope, self->super.name); | |||
| 2063 | } | |||
| 2064 | ||||
| 2065 | // MARK - Softmax Layer | |||
| 2066 | ||||
| 2067 | typedef struct { | |||
| 2068 | ccv_cnnp_model_t super; | |||
| 2069 | ccv_nnc_tensor_symbol_t output; | |||
| 2070 | } ccv_cnnp_model_softmax_t; | |||
| 2071 | ||||
| 2072 | 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) | |||
| 2073 | { | |||
| 2074 | 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); | |||
| 2075 | 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" , 2075, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2076 | 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", 2076, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2077 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2078 | ccv_nnc_tensor_param_t output_params; | |||
| 2079 | const ccv_nnc_cmd_t softmax = CMD_SOFTMAX_FORWARD()ccv_nnc_cmd(CCV_NNC_SOFTMAX_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 2080 | ccv_nnc_hint_tensor_auto(softmax, (ccv_nnc_tensor_param_t []){ | |||
| 2081 | params, | |||
| 2082 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2083 | const ccv_nnc_tensor_symbol_t softmax_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2084 | 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"); | |||
| 2085 | outputs[0] = softmax_output; | |||
| 2086 | } | |||
| 2087 | ||||
| 2088 | static ccv_cnnp_model_t* _ccv_cnnp_softmax_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2089 | ||||
| 2090 | static const ccv_cnnp_model_vtab_t ccv_cnnp_softmax_isa = { | |||
| 2091 | .build = _ccv_cnnp_softmax_build, | |||
| 2092 | .copy = _ccv_cnnp_softmax_copy, | |||
| 2093 | }; | |||
| 2094 | ||||
| 2095 | ccv_cnnp_model_t* ccv_cnnp_softmax(const char* const name) | |||
| 2096 | { | |||
| 2097 | ccv_cnnp_model_softmax_t* const model_softmax = (ccv_cnnp_model_softmax_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_softmax_t)); | |||
| 2098 | model_softmax->super.isa = &ccv_cnnp_softmax_isa; | |||
| 2099 | model_softmax->super.input_size = 1; | |||
| 2100 | model_softmax->super.outputs = &model_softmax->output; | |||
| 2101 | model_softmax->super.output_size = 1; | |||
| 2102 | ccv_cnnp_model_copy_name(&model_softmax->super, name); | |||
| 2103 | return (ccv_cnnp_model_t*)model_softmax; | |||
| 2104 | } | |||
| 2105 | ||||
| 2106 | static ccv_cnnp_model_t* _ccv_cnnp_softmax_copy(const ccv_cnnp_model_t* const self, void* const context) | |||
| 2107 | { | |||
| 2108 | return ccv_cnnp_softmax(self->name); | |||
| 2109 | } | |||
| 2110 | ||||
| 2111 | // MARK - Add Layer | |||
| 2112 | ||||
| 2113 | typedef struct { | |||
| 2114 | ccv_cnnp_model_t super; | |||
| 2115 | float p; | |||
| 2116 | float q; | |||
| 2117 | ccv_nnc_tensor_symbol_t output; | |||
| 2118 | } ccv_cnnp_model_add_t; | |||
| 2119 | ||||
| 2120 | 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) | |||
| 2121 | { | |||
| 2122 | 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); | |||
| 2123 | const ccv_cnnp_model_add_t* const self = (const ccv_cnnp_model_add_t*)super; | |||
| 2124 | 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" , 2124, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2125 | 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", 2125, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2126 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 2127 | int i; | |||
| 2128 | for (i = 0; i < 2; i++) | |||
| 2129 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 2130 | ccv_nnc_tensor_param_t output_params; | |||
| 2131 | 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); | |||
| 2132 | ccv_nnc_hint_tensor_auto(add, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2133 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2134 | ccv_nnc_graph_exec_symbol_new(graph, add, inputs, input_size, outputs, output_size, "add"); | |||
| 2135 | } | |||
| 2136 | ||||
| 2137 | static ccv_cnnp_model_t* _ccv_cnnp_add_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2138 | ||||
| 2139 | static const ccv_cnnp_model_vtab_t ccv_cnnp_add_isa = { | |||
| 2140 | .build = _ccv_cnnp_add_build, | |||
| 2141 | .copy = _ccv_cnnp_add_copy, | |||
| 2142 | }; | |||
| 2143 | ||||
| 2144 | ccv_cnnp_model_t* ccv_cnnp_add(const float p, const float q, const char* const name) | |||
| 2145 | { | |||
| 2146 | ccv_cnnp_model_add_t* const model_add = (ccv_cnnp_model_add_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_add_t)); | |||
| 2147 | model_add->super.isa = &ccv_cnnp_add_isa; | |||
| 2148 | model_add->super.input_size = 2; | |||
| 2149 | model_add->super.outputs = &model_add->output; | |||
| 2150 | model_add->super.output_size = 1; | |||
| 2151 | model_add->p = p; | |||
| 2152 | model_add->q = q; | |||
| 2153 | ccv_cnnp_model_copy_name(&model_add->super, name); | |||
| 2154 | return (ccv_cnnp_model_t*)model_add; | |||
| 2155 | } | |||
| 2156 | ||||
| 2157 | static ccv_cnnp_model_t* _ccv_cnnp_add_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2158 | { | |||
| 2159 | const ccv_cnnp_model_add_t* const self = (const ccv_cnnp_model_add_t*)super; | |||
| 2160 | return ccv_cnnp_add(self->p, self->q, self->super.name); | |||
| 2161 | } | |||
| 2162 | ||||
| 2163 | // MARK - Mul Layer | |||
| 2164 | ||||
| 2165 | typedef struct { | |||
| 2166 | ccv_cnnp_model_t super; | |||
| 2167 | ccv_nnc_tensor_symbol_t output; | |||
| 2168 | float p; | |||
| 2169 | } ccv_cnnp_model_mul_t; | |||
| 2170 | ||||
| 2171 | 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) | |||
| 2172 | { | |||
| 2173 | 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); | |||
| 2174 | const ccv_cnnp_model_mul_t* const self = (const ccv_cnnp_model_mul_t*)super; | |||
| 2175 | 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" , 2175, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2176 | 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", 2176, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2177 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 2178 | int i; | |||
| 2179 | for (i = 0; i < 2; i++) | |||
| 2180 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 2181 | ccv_nnc_tensor_param_t output_params; | |||
| 2182 | 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); | |||
| 2183 | ccv_nnc_hint_tensor_auto(mul, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2184 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2185 | ccv_nnc_graph_exec_symbol_new(graph, mul, inputs, input_size, outputs, output_size, "mul"); | |||
| 2186 | } | |||
| 2187 | ||||
| 2188 | static ccv_cnnp_model_t* _ccv_cnnp_mul_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2189 | ||||
| 2190 | static const ccv_cnnp_model_vtab_t ccv_cnnp_mul_isa = { | |||
| 2191 | .build = _ccv_cnnp_mul_build, | |||
| 2192 | .copy = _ccv_cnnp_mul_copy, | |||
| 2193 | }; | |||
| 2194 | ||||
| 2195 | ccv_cnnp_model_t* ccv_cnnp_mul(const float p, const char* const name) | |||
| 2196 | { | |||
| 2197 | ccv_cnnp_model_mul_t* const model_mul = (ccv_cnnp_model_mul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_mul_t)); | |||
| 2198 | model_mul->super.isa = &ccv_cnnp_mul_isa; | |||
| 2199 | model_mul->super.input_size = 2; | |||
| 2200 | model_mul->super.outputs = &model_mul->output; | |||
| 2201 | model_mul->super.output_size = 1; | |||
| 2202 | model_mul->p = p; | |||
| 2203 | ccv_cnnp_model_copy_name(&model_mul->super, name); | |||
| 2204 | return (ccv_cnnp_model_t*)model_mul; | |||
| 2205 | } | |||
| 2206 | ||||
| 2207 | static ccv_cnnp_model_t* _ccv_cnnp_mul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2208 | { | |||
| 2209 | const ccv_cnnp_model_mul_t* const self = (const ccv_cnnp_model_mul_t*)super; | |||
| 2210 | return ccv_cnnp_mul(self->p, self->super.name); | |||
| 2211 | } | |||
| 2212 | ||||
| 2213 | // MARK - Scalar Mul Layer | |||
| 2214 | ||||
| 2215 | typedef struct { | |||
| 2216 | ccv_cnnp_model_t super; | |||
| 2217 | ccv_nnc_tensor_symbol_t output; | |||
| 2218 | float a; | |||
| 2219 | } ccv_cnnp_model_scalar_mul_t; | |||
| 2220 | ||||
| 2221 | 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) | |||
| 2222 | { | |||
| 2223 | 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); | |||
| 2224 | 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" , 2224, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2225 | 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", 2225, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2226 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2227 | ccv_nnc_tensor_param_t output_params; | |||
| 2228 | ccv_cnnp_model_scalar_mul_t* const self = (ccv_cnnp_model_scalar_mul_t*)super; | |||
| 2229 | 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); | |||
| 2230 | ccv_nnc_hint_tensor_auto(scalar_mul, (ccv_nnc_tensor_param_t []){ | |||
| 2231 | params, | |||
| 2232 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2233 | const ccv_nnc_tensor_symbol_t scalar_mul_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2234 | 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"); | |||
| 2235 | outputs[0] = scalar_mul_output; | |||
| 2236 | } | |||
| 2237 | ||||
| 2238 | static ccv_cnnp_model_t* _ccv_cnnp_scalar_mul_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2239 | ||||
| 2240 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scalar_mul_isa = { | |||
| 2241 | .build = _ccv_cnnp_scalar_mul_build, | |||
| 2242 | .copy = _ccv_cnnp_scalar_mul_copy, | |||
| 2243 | }; | |||
| 2244 | ||||
| 2245 | ccv_cnnp_model_t* ccv_cnnp_scalar_mul(const float a, const char* const name) | |||
| 2246 | { | |||
| 2247 | 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)); | |||
| 2248 | model_scalar_mul->super.isa = &ccv_cnnp_scalar_mul_isa; | |||
| 2249 | model_scalar_mul->super.input_size = 1; | |||
| 2250 | model_scalar_mul->super.outputs = &model_scalar_mul->output; | |||
| 2251 | model_scalar_mul->super.output_size = 1; | |||
| 2252 | model_scalar_mul->a = a; | |||
| 2253 | ccv_cnnp_model_copy_name(&model_scalar_mul->super, name); | |||
| 2254 | return (ccv_cnnp_model_t*)model_scalar_mul; | |||
| 2255 | } | |||
| 2256 | ||||
| 2257 | static ccv_cnnp_model_t* _ccv_cnnp_scalar_mul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2258 | { | |||
| 2259 | const ccv_cnnp_model_scalar_mul_t* const self = (const ccv_cnnp_model_scalar_mul_t*)super; | |||
| 2260 | return ccv_cnnp_scalar_mul(self->a, self->super.name); | |||
| 2261 | } | |||
| 2262 | ||||
| 2263 | // MARK - Div Layer | |||
| 2264 | ||||
| 2265 | typedef struct { | |||
| 2266 | ccv_cnnp_model_t super; | |||
| 2267 | ccv_nnc_tensor_symbol_t output; | |||
| 2268 | int reciprocal; | |||
| 2269 | } ccv_cnnp_model_div_t; | |||
| 2270 | ||||
| 2271 | 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) | |||
| 2272 | { | |||
| 2273 | const ccv_cnnp_model_div_t* const self = (const ccv_cnnp_model_div_t*)super; | |||
| 2274 | 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); | |||
| 2275 | 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", 2275, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2276 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 2277 | int i; | |||
| 2278 | ccv_nnc_tensor_param_t output_params; | |||
| 2279 | const ccv_nnc_cmd_t div = CMD_EWDIV_FORWARD()ccv_nnc_cmd(CCV_NNC_EWDIV_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 2280 | if (self->reciprocal) | |||
| 2281 | { | |||
| 2282 | 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" , 2282, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2283 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2284 | input_params[1] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2285 | ccv_nnc_hint_tensor_auto(div, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2286 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2287 | 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"); | |||
| 2288 | } else { | |||
| 2289 | 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" , 2289, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2290 | for (i = 0; i < 2; i++) | |||
| 2291 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 2292 | ccv_nnc_hint_tensor_auto(div, input_params, input_size, ccv_nnc_no_hint, &output_params, 1); | |||
| 2293 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2294 | ccv_nnc_graph_exec_symbol_new(graph, div, inputs, input_size, outputs, output_size, "div"); | |||
| 2295 | } | |||
| 2296 | } | |||
| 2297 | ||||
| 2298 | static ccv_cnnp_model_t* _ccv_cnnp_div_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2299 | ||||
| 2300 | static const ccv_cnnp_model_vtab_t ccv_cnnp_div_isa = { | |||
| 2301 | .build = _ccv_cnnp_div_build, | |||
| 2302 | .copy = _ccv_cnnp_div_copy, | |||
| 2303 | }; | |||
| 2304 | ||||
| 2305 | ccv_cnnp_model_t* ccv_cnnp_div(const int reciprocal, const char* const name) | |||
| 2306 | { | |||
| 2307 | ccv_cnnp_model_div_t* const model_div = (ccv_cnnp_model_div_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_div_t)); | |||
| 2308 | model_div->super.isa = &ccv_cnnp_div_isa; | |||
| 2309 | model_div->super.input_size = reciprocal ? 1 : 2; | |||
| 2310 | model_div->super.outputs = &model_div->output; | |||
| 2311 | model_div->super.output_size = 1; | |||
| 2312 | model_div->reciprocal = reciprocal; | |||
| 2313 | ccv_cnnp_model_copy_name(&model_div->super, name); | |||
| 2314 | return (ccv_cnnp_model_t*)model_div; | |||
| 2315 | } | |||
| 2316 | ||||
| 2317 | static ccv_cnnp_model_t* _ccv_cnnp_div_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2318 | { | |||
| 2319 | const ccv_cnnp_model_div_t* const self = (const ccv_cnnp_model_div_t*)super; | |||
| 2320 | return ccv_cnnp_div(self->reciprocal, self->super.name); | |||
| 2321 | } | |||
| 2322 | ||||
| 2323 | // MARK - Sqrt Layer | |||
| 2324 | ||||
| 2325 | typedef struct { | |||
| 2326 | ccv_cnnp_model_t super; | |||
| 2327 | ccv_nnc_tensor_symbol_t output; | |||
| 2328 | } ccv_cnnp_model_sqrt_t; | |||
| 2329 | ||||
| 2330 | 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) | |||
| 2331 | { | |||
| 2332 | 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); | |||
| 2333 | 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", 2333, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2334 | ccv_nnc_tensor_param_t input_params[1]; | |||
| 2335 | ccv_nnc_tensor_param_t output_params; | |||
| 2336 | const ccv_nnc_cmd_t sqrt = CMD_EWSQRT_FORWARD()ccv_nnc_cmd(CCV_NNC_EWSQRT_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 2337 | 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" , 2337, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2338 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2339 | ccv_nnc_hint_tensor_auto(sqrt, input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2340 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2341 | ccv_nnc_graph_exec_symbol_new(graph, sqrt, inputs, 1, outputs, output_size, "sqrt"); | |||
| 2342 | } | |||
| 2343 | ||||
| 2344 | static ccv_cnnp_model_t* _ccv_cnnp_sqrt_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2345 | ||||
| 2346 | static const ccv_cnnp_model_vtab_t ccv_cnnp_sqrt_isa = { | |||
| 2347 | .build = _ccv_cnnp_sqrt_build, | |||
| 2348 | .copy = _ccv_cnnp_sqrt_copy, | |||
| 2349 | }; | |||
| 2350 | ||||
| 2351 | ccv_cnnp_model_t* ccv_cnnp_sqrt(const char* const name) | |||
| 2352 | { | |||
| 2353 | ccv_cnnp_model_sqrt_t* const model_sqrt = (ccv_cnnp_model_sqrt_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_sqrt_t)); | |||
| 2354 | model_sqrt->super.isa = &ccv_cnnp_sqrt_isa; | |||
| 2355 | model_sqrt->super.input_size = 1; | |||
| 2356 | model_sqrt->super.outputs = &model_sqrt->output; | |||
| 2357 | model_sqrt->super.output_size = 1; | |||
| 2358 | ccv_cnnp_model_copy_name(&model_sqrt->super, name); | |||
| 2359 | return (ccv_cnnp_model_t*)model_sqrt; | |||
| 2360 | } | |||
| 2361 | ||||
| 2362 | static ccv_cnnp_model_t* _ccv_cnnp_sqrt_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2363 | { | |||
| 2364 | const ccv_cnnp_model_sqrt_t* const self = (const ccv_cnnp_model_sqrt_t*)super; | |||
| 2365 | return ccv_cnnp_sqrt(self->super.name); | |||
| 2366 | } | |||
| 2367 | ||||
| 2368 | // MARK - Signed Sqrt Layer | |||
| 2369 | ||||
| 2370 | typedef struct { | |||
| 2371 | ccv_cnnp_model_t super; | |||
| 2372 | ccv_nnc_tensor_symbol_t output; | |||
| 2373 | float minimum_magnitude; | |||
| 2374 | } ccv_cnnp_model_signed_sqrt_t; | |||
| 2375 | ||||
| 2376 | static void _ccv_cnnp_signed_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) | |||
| 2377 | { | |||
| 2378 | PRINT(CCV_CLI_VERBOSE, "[cnnp_signed_sqrt_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_signed_sqrt_build] -\n"); fflush(stdout); } } while (0); | |||
| 2379 | 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" , 2379, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2380 | 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", 2380, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2381 | const ccv_cnnp_model_signed_sqrt_t* const self = (const ccv_cnnp_model_signed_sqrt_t*)super; | |||
| 2382 | const ccv_nnc_cmd_t cmd = CMD_SIGNED_SQRT_FORWARD(self->minimum_magnitude)ccv_nnc_cmd(CCV_NNC_SIGNED_SQRT_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.signed_sqrt={.minimum_magnitude=(self ->minimum_magnitude)}}), 0); | |||
| 2383 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2384 | ccv_nnc_tensor_param_t output_params; | |||
| 2385 | ccv_nnc_hint_tensor_auto(cmd, ¶ms, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2386 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2387 | ccv_nnc_graph_exec_symbol_new(graph, cmd, inputs, input_size, outputs, output_size, "signed_sqrt"); | |||
| 2388 | } | |||
| 2389 | ||||
| 2390 | static ccv_cnnp_model_t* _ccv_cnnp_signed_sqrt_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2391 | ||||
| 2392 | static const ccv_cnnp_model_vtab_t ccv_cnnp_signed_sqrt_isa = { | |||
| 2393 | .build = _ccv_cnnp_signed_sqrt_build, | |||
| 2394 | .copy = _ccv_cnnp_signed_sqrt_copy, | |||
| 2395 | }; | |||
| 2396 | ||||
| 2397 | ccv_cnnp_model_t* ccv_cnnp_signed_sqrt(const float minimum_magnitude, const char* const name) | |||
| 2398 | { | |||
| 2399 | assert(minimum_magnitude > 0 && isfinite(minimum_magnitude))((void) sizeof ((minimum_magnitude > 0 && __builtin_isfinite (minimum_magnitude)) ? 1 : 0), __extension__ ({ if (minimum_magnitude > 0 && __builtin_isfinite (minimum_magnitude)) ; else __assert_fail ("minimum_magnitude > 0 && isfinite(minimum_magnitude)" , "ccv_cnnp_model_addons.c", 2399, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2400 | ccv_cnnp_model_signed_sqrt_t* const model = (ccv_cnnp_model_signed_sqrt_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_signed_sqrt_t)); | |||
| 2401 | model->super.isa = &ccv_cnnp_signed_sqrt_isa; | |||
| 2402 | model->super.input_size = 1; | |||
| 2403 | model->super.outputs = &model->output; | |||
| 2404 | model->super.output_size = 1; | |||
| 2405 | model->minimum_magnitude = minimum_magnitude; | |||
| 2406 | ccv_cnnp_model_copy_name(&model->super, name); | |||
| 2407 | return (ccv_cnnp_model_t*)model; | |||
| 2408 | } | |||
| 2409 | ||||
| 2410 | static ccv_cnnp_model_t* _ccv_cnnp_signed_sqrt_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2411 | { | |||
| 2412 | const ccv_cnnp_model_signed_sqrt_t* const self = (const ccv_cnnp_model_signed_sqrt_t*)super; | |||
| 2413 | return ccv_cnnp_signed_sqrt(self->minimum_magnitude, self->super.name); | |||
| 2414 | } | |||
| 2415 | ||||
| 2416 | // MARK - Log Layer | |||
| 2417 | ||||
| 2418 | typedef struct { | |||
| 2419 | ccv_cnnp_model_t super; | |||
| 2420 | ccv_nnc_tensor_symbol_t output; | |||
| 2421 | } ccv_cnnp_model_log_t; | |||
| 2422 | ||||
| 2423 | static void _ccv_cnnp_log_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) | |||
| 2424 | { | |||
| 2425 | PRINT(CCV_CLI_VERBOSE, "[cnnp_log_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_log_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 2426 | 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", 2426, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2427 | ccv_nnc_tensor_param_t input_params[1]; | |||
| 2428 | ccv_nnc_tensor_param_t output_params; | |||
| 2429 | const ccv_nnc_cmd_t log = CMD_EWLOG_FORWARD()ccv_nnc_cmd(CCV_NNC_EWLOG_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 2430 | 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" , 2430, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2431 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2432 | ccv_nnc_hint_tensor_auto(log, input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2433 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2434 | ccv_nnc_graph_exec_symbol_new(graph, log, inputs, 1, outputs, output_size, "log"); | |||
| 2435 | } | |||
| 2436 | ||||
| 2437 | static ccv_cnnp_model_t* _ccv_cnnp_log_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2438 | ||||
| 2439 | static const ccv_cnnp_model_vtab_t ccv_cnnp_log_isa = { | |||
| 2440 | .build = _ccv_cnnp_log_build, | |||
| 2441 | .copy = _ccv_cnnp_log_copy, | |||
| 2442 | }; | |||
| 2443 | ||||
| 2444 | ccv_cnnp_model_t* ccv_cnnp_log(const char* const name) | |||
| 2445 | { | |||
| 2446 | ccv_cnnp_model_log_t* const model_log = (ccv_cnnp_model_log_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_log_t)); | |||
| 2447 | model_log->super.isa = &ccv_cnnp_log_isa; | |||
| 2448 | model_log->super.input_size = 1; | |||
| 2449 | model_log->super.outputs = &model_log->output; | |||
| 2450 | model_log->super.output_size = 1; | |||
| 2451 | ccv_cnnp_model_copy_name(&model_log->super, name); | |||
| 2452 | return (ccv_cnnp_model_t*)model_log; | |||
| 2453 | } | |||
| 2454 | ||||
| 2455 | static ccv_cnnp_model_t* _ccv_cnnp_log_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2456 | { | |||
| 2457 | return ccv_cnnp_log(super->name); | |||
| 2458 | } | |||
| 2459 | ||||
| 2460 | // MARK - Pow Layer | |||
| 2461 | ||||
| 2462 | typedef struct { | |||
| 2463 | ccv_cnnp_model_t super; | |||
| 2464 | ccv_nnc_tensor_symbol_t output; | |||
| 2465 | ccv_nnc_cmd_param_t params; | |||
| 2466 | } ccv_cnnp_model_pow_t; | |||
| 2467 | ||||
| 2468 | static void _ccv_cnnp_pow_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) | |||
| 2469 | { | |||
| 2470 | ccv_cnnp_model_pow_t* const self = (ccv_cnnp_model_pow_t*)super; | |||
| 2471 | PRINT(CCV_CLI_VERBOSE, "[cnnp_pow_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_pow_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 2472 | 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" , 2472, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2473 | 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", 2473, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2474 | ccv_nnc_tensor_param_t input_params[1]; | |||
| 2475 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2476 | ccv_nnc_tensor_param_t output_params; | |||
| 2477 | const ccv_nnc_cmd_t pow = ccv_nnc_cmd(CCV_NNC_EWPOW_FORWARD, 0, self->params, 0); | |||
| 2478 | ccv_nnc_hint_tensor_auto(pow, input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2479 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2480 | ccv_nnc_graph_exec_symbol_new(graph, pow, inputs, input_size, outputs, output_size, "pow"); | |||
| 2481 | } | |||
| 2482 | ||||
| 2483 | static ccv_cnnp_model_t* _ccv_cnnp_pow_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2484 | ||||
| 2485 | static const ccv_cnnp_model_vtab_t ccv_cnnp_pow_isa = { | |||
| 2486 | .build = _ccv_cnnp_pow_build, | |||
| 2487 | .copy = _ccv_cnnp_pow_copy, | |||
| 2488 | }; | |||
| 2489 | ||||
| 2490 | ccv_cnnp_model_t* ccv_cnnp_pow(const float exponent, const char* const name) | |||
| 2491 | { | |||
| 2492 | ccv_cnnp_model_pow_t* const model_pow = (ccv_cnnp_model_pow_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_pow_t)); | |||
| 2493 | model_pow->super.isa = &ccv_cnnp_pow_isa; | |||
| 2494 | model_pow->super.input_size = 1; | |||
| 2495 | model_pow->super.outputs = &model_pow->output; | |||
| 2496 | model_pow->super.output_size = 1; | |||
| 2497 | model_pow->params = (ccv_nnc_cmd_param_t){ | |||
| 2498 | .size = { | |||
| 2499 | .dim = { 1, 1, 1 } | |||
| 2500 | }, | |||
| 2501 | .pow = { | |||
| 2502 | .exponent = exponent, | |||
| 2503 | }, | |||
| 2504 | }; | |||
| 2505 | ccv_cnnp_model_copy_name(&model_pow->super, name); | |||
| 2506 | return (ccv_cnnp_model_t*)model_pow; | |||
| 2507 | } | |||
| 2508 | ||||
| 2509 | static ccv_cnnp_model_t* _ccv_cnnp_pow_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2510 | { | |||
| 2511 | const ccv_cnnp_model_pow_t* const self = (const ccv_cnnp_model_pow_t*)super; | |||
| 2512 | return ccv_cnnp_pow(self->params.pow.exponent, super->name); | |||
| 2513 | } | |||
| 2514 | ||||
| 2515 | // MARK - Sin Layer | |||
| 2516 | ||||
| 2517 | typedef struct { | |||
| 2518 | ccv_cnnp_model_t super; | |||
| 2519 | ccv_nnc_tensor_symbol_t output; | |||
| 2520 | } ccv_cnnp_model_sin_t; | |||
| 2521 | ||||
| 2522 | static void _ccv_cnnp_sin_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) | |||
| 2523 | { | |||
| 2524 | PRINT(CCV_CLI_VERBOSE, "[cnnp_sin_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_sin_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 2525 | 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", 2525, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2526 | 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" , 2526, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2527 | ccv_nnc_tensor_param_t input_params[1]; | |||
| 2528 | ccv_nnc_tensor_param_t output_params; | |||
| 2529 | const ccv_nnc_cmd_t sin = CMD_EWSIN_FORWARD()ccv_nnc_cmd(CCV_NNC_EWSIN_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 2530 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2531 | ccv_nnc_hint_tensor_auto(sin, input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2532 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2533 | ccv_nnc_graph_exec_symbol_new(graph, sin, inputs, 1, outputs, output_size, "sin"); | |||
| 2534 | } | |||
| 2535 | ||||
| 2536 | static ccv_cnnp_model_t* _ccv_cnnp_sin_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2537 | ||||
| 2538 | static const ccv_cnnp_model_vtab_t ccv_cnnp_sin_isa = { | |||
| 2539 | .build = _ccv_cnnp_sin_build, | |||
| 2540 | .copy = _ccv_cnnp_sin_copy, | |||
| 2541 | }; | |||
| 2542 | ||||
| 2543 | ccv_cnnp_model_t* ccv_cnnp_sin(const char* const name) | |||
| 2544 | { | |||
| 2545 | ccv_cnnp_model_sin_t* const model_sin = (ccv_cnnp_model_sin_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_sin_t)); | |||
| 2546 | model_sin->super.isa = &ccv_cnnp_sin_isa; | |||
| 2547 | model_sin->super.input_size = 1; | |||
| 2548 | model_sin->super.outputs = &model_sin->output; | |||
| 2549 | model_sin->super.output_size = 1; | |||
| 2550 | ccv_cnnp_model_copy_name(&model_sin->super, name); | |||
| 2551 | return (ccv_cnnp_model_t*)model_sin; | |||
| 2552 | } | |||
| 2553 | ||||
| 2554 | static ccv_cnnp_model_t* _ccv_cnnp_sin_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2555 | { | |||
| 2556 | return ccv_cnnp_sin(super->name); | |||
| 2557 | } | |||
| 2558 | ||||
| 2559 | // MARK - Cos Layer | |||
| 2560 | ||||
| 2561 | typedef struct { | |||
| 2562 | ccv_cnnp_model_t super; | |||
| 2563 | ccv_nnc_tensor_symbol_t output; | |||
| 2564 | } ccv_cnnp_model_cos_t; | |||
| 2565 | ||||
| 2566 | static void _ccv_cnnp_cos_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) | |||
| 2567 | { | |||
| 2568 | PRINT(CCV_CLI_VERBOSE, "[cnnp_cos_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_cos_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 2569 | 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", 2569, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2570 | 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" , 2570, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2571 | ccv_nnc_tensor_param_t input_params[1]; | |||
| 2572 | ccv_nnc_tensor_param_t output_params; | |||
| 2573 | const ccv_nnc_cmd_t cos = CMD_EWCOS_FORWARD()ccv_nnc_cmd(CCV_NNC_EWCOS_FORWARD, 0, ccv_nnc_cmd_auto, 0); | |||
| 2574 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2575 | ccv_nnc_hint_tensor_auto(cos, input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2576 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2577 | ccv_nnc_graph_exec_symbol_new(graph, cos, inputs, 1, outputs, output_size, "cos"); | |||
| 2578 | } | |||
| 2579 | ||||
| 2580 | static ccv_cnnp_model_t* _ccv_cnnp_cos_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2581 | ||||
| 2582 | static const ccv_cnnp_model_vtab_t ccv_cnnp_cos_isa = { | |||
| 2583 | .build = _ccv_cnnp_cos_build, | |||
| 2584 | .copy = _ccv_cnnp_cos_copy, | |||
| 2585 | }; | |||
| 2586 | ||||
| 2587 | ccv_cnnp_model_t* ccv_cnnp_cos(const char* const name) | |||
| 2588 | { | |||
| 2589 | ccv_cnnp_model_cos_t* const model_cos = (ccv_cnnp_model_cos_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_cos_t)); | |||
| 2590 | model_cos->super.isa = &ccv_cnnp_cos_isa; | |||
| 2591 | model_cos->super.input_size = 1; | |||
| 2592 | model_cos->super.outputs = &model_cos->output; | |||
| 2593 | model_cos->super.output_size = 1; | |||
| 2594 | ccv_cnnp_model_copy_name(&model_cos->super, name); | |||
| 2595 | return (ccv_cnnp_model_t*)model_cos; | |||
| 2596 | } | |||
| 2597 | ||||
| 2598 | static ccv_cnnp_model_t* _ccv_cnnp_cos_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2599 | { | |||
| 2600 | return ccv_cnnp_cos(super->name); | |||
| 2601 | } | |||
| 2602 | ||||
| 2603 | // MARK - Rotate Half Layer | |||
| 2604 | ||||
| 2605 | typedef struct { | |||
| 2606 | ccv_cnnp_model_t super; | |||
| 2607 | ccv_nnc_tensor_symbol_t output; | |||
| 2608 | } ccv_cnnp_model_rotate_half_t; | |||
| 2609 | ||||
| 2610 | static void _ccv_cnnp_rotate_half_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) | |||
| 2611 | { | |||
| 2612 | PRINT(CCV_CLI_VERBOSE, "[cnnp_rotate_half_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_rotate_half_build] -\n"); fflush(stdout); } } while (0); | |||
| 2613 | 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" , 2613, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2614 | 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", 2614, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2615 | ccv_nnc_tensor_param_t input_params[1]; | |||
| 2616 | ccv_nnc_tensor_param_t output_params; | |||
| 2617 | const ccv_nnc_cmd_t rotate_half = CMD_ROTATE_HALF_FORWARD()ccv_nnc_cmd(CCV_NNC_ROTATE_HALF_FORWARD, 0, (ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}}}, 0); | |||
| 2618 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2619 | ccv_nnc_hint_tensor_auto(rotate_half, input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2620 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2621 | ccv_nnc_graph_exec_symbol_new(graph, rotate_half, inputs, 1, outputs, output_size, "rotate_half"); | |||
| 2622 | } | |||
| 2623 | ||||
| 2624 | static ccv_cnnp_model_t* _ccv_cnnp_rotate_half_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2625 | ||||
| 2626 | static const ccv_cnnp_model_vtab_t ccv_cnnp_rotate_half_isa = { | |||
| 2627 | .build = _ccv_cnnp_rotate_half_build, | |||
| 2628 | .copy = _ccv_cnnp_rotate_half_copy, | |||
| 2629 | }; | |||
| 2630 | ||||
| 2631 | ccv_cnnp_model_t* ccv_cnnp_rotate_half(const char* const name) | |||
| 2632 | { | |||
| 2633 | ccv_cnnp_model_rotate_half_t* const model_rotate_half = (ccv_cnnp_model_rotate_half_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_rotate_half_t)); | |||
| 2634 | model_rotate_half->super.isa = &ccv_cnnp_rotate_half_isa; | |||
| 2635 | model_rotate_half->super.input_size = 1; | |||
| 2636 | model_rotate_half->super.outputs = &model_rotate_half->output; | |||
| 2637 | model_rotate_half->super.output_size = 1; | |||
| 2638 | ccv_cnnp_model_copy_name(&model_rotate_half->super, name); | |||
| 2639 | return (ccv_cnnp_model_t*)model_rotate_half; | |||
| 2640 | } | |||
| 2641 | ||||
| 2642 | static ccv_cnnp_model_t* _ccv_cnnp_rotate_half_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2643 | { | |||
| 2644 | return ccv_cnnp_rotate_half(super->name); | |||
| 2645 | } | |||
| 2646 | ||||
| 2647 | // MARK - Walsh-Hadamard Transform Layer | |||
| 2648 | ||||
| 2649 | typedef struct { | |||
| 2650 | ccv_cnnp_model_t super; | |||
| 2651 | ccv_nnc_tensor_symbol_t output; | |||
| 2652 | float scale; | |||
| 2653 | } ccv_cnnp_model_walsh_hadamard_transform_t; | |||
| 2654 | ||||
| 2655 | static void _ccv_cnnp_walsh_hadamard_transform_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) | |||
| 2656 | { | |||
| 2657 | PRINT(CCV_CLI_VERBOSE, "[cnnp_walsh_hadamard_transform_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_walsh_hadamard_transform_build] -\n"); fflush (stdout); } } while (0); | |||
| 2658 | 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" , 2658, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2659 | 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", 2659, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2660 | const ccv_cnnp_model_walsh_hadamard_transform_t* const self = (const ccv_cnnp_model_walsh_hadamard_transform_t*)super; | |||
| 2661 | ccv_nnc_tensor_param_t input_params[1]; | |||
| 2662 | ccv_nnc_tensor_param_t output_params; | |||
| 2663 | const ccv_nnc_cmd_t walsh_hadamard_transform = CMD_WALSH_HADAMARD_TRANSFORM_FORWARD(self->scale)ccv_nnc_cmd(CCV_NNC_WALSH_HADAMARD_TRANSFORM_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.walsh_hadamard_transform={.scale=(self ->scale)}}), 0); | |||
| 2664 | input_params[0] = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2665 | ccv_nnc_hint_tensor_auto(walsh_hadamard_transform, input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2666 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2667 | ccv_nnc_graph_exec_symbol_new(graph, walsh_hadamard_transform, inputs, 1, outputs, output_size, "walsh_hadamard_transform"); | |||
| 2668 | } | |||
| 2669 | ||||
| 2670 | static ccv_cnnp_model_t* _ccv_cnnp_walsh_hadamard_transform_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2671 | ||||
| 2672 | static const ccv_cnnp_model_vtab_t ccv_cnnp_walsh_hadamard_transform_isa = { | |||
| 2673 | .build = _ccv_cnnp_walsh_hadamard_transform_build, | |||
| 2674 | .copy = _ccv_cnnp_walsh_hadamard_transform_copy, | |||
| 2675 | }; | |||
| 2676 | ||||
| 2677 | ccv_cnnp_model_t* ccv_cnnp_walsh_hadamard_transform(const float scale, const char* const name) | |||
| 2678 | { | |||
| 2679 | ccv_cnnp_model_walsh_hadamard_transform_t* const model_walsh_hadamard_transform = (ccv_cnnp_model_walsh_hadamard_transform_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_walsh_hadamard_transform_t)); | |||
| 2680 | model_walsh_hadamard_transform->super.isa = &ccv_cnnp_walsh_hadamard_transform_isa; | |||
| 2681 | model_walsh_hadamard_transform->super.input_size = 1; | |||
| 2682 | model_walsh_hadamard_transform->super.outputs = &model_walsh_hadamard_transform->output; | |||
| 2683 | model_walsh_hadamard_transform->super.output_size = 1; | |||
| 2684 | model_walsh_hadamard_transform->scale = scale; | |||
| 2685 | ccv_cnnp_model_copy_name(&model_walsh_hadamard_transform->super, name); | |||
| 2686 | return (ccv_cnnp_model_t*)model_walsh_hadamard_transform; | |||
| 2687 | } | |||
| 2688 | ||||
| 2689 | static ccv_cnnp_model_t* _ccv_cnnp_walsh_hadamard_transform_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2690 | { | |||
| 2691 | const ccv_cnnp_model_walsh_hadamard_transform_t* const self = (const ccv_cnnp_model_walsh_hadamard_transform_t*)super; | |||
| 2692 | return ccv_cnnp_walsh_hadamard_transform(self->scale, self->super.name); | |||
| 2693 | } | |||
| 2694 | ||||
| 2695 | // MARK - Hyper Connection Layer | |||
| 2696 | ||||
| 2697 | typedef struct { | |||
| 2698 | ccv_cnnp_model_t super; | |||
| 2699 | ccv_nnc_tensor_symbol_t outputs[3]; | |||
| 2700 | int count; | |||
| 2701 | int sinkhorn_iterations; | |||
| 2702 | float epsilon; | |||
| 2703 | int operation; | |||
| 2704 | } ccv_cnnp_model_hyper_connection_t; | |||
| 2705 | ||||
| 2706 | static void _ccv_cnnp_hyper_connection_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) | |||
| 2707 | { | |||
| 2708 | PRINT(CCV_CLI_VERBOSE, "[cnnp_hyper_connection_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_hyper_connection_build] -\n"); fflush(stdout ); } } while (0); | |||
| 2709 | const ccv_cnnp_model_hyper_connection_t* const self = (const ccv_cnnp_model_hyper_connection_t*)super; | |||
| 2710 | assert(input_size == (self->operation == 0 ? 3 : 4))((void) sizeof ((input_size == (self->operation == 0 ? 3 : 4)) ? 1 : 0), __extension__ ({ if (input_size == (self->operation == 0 ? 3 : 4)) ; else __assert_fail ("input_size == (self->operation == 0 ? 3 : 4)" , "ccv_cnnp_model_addons.c", 2710, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2711 | assert(output_size == (self->operation == 2 ? 1 : 3))((void) sizeof ((output_size == (self->operation == 2 ? 1 : 3)) ? 1 : 0), __extension__ ({ if (output_size == (self-> operation == 2 ? 1 : 3)) ; else __assert_fail ("output_size == (self->operation == 2 ? 1 : 3)" , "ccv_cnnp_model_addons.c", 2711, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2712 | ccv_nnc_tensor_param_t input_params[4]; | |||
| 2713 | int i; | |||
| 2714 | for (i = 0; i < input_size; i++) | |||
| 2715 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 2716 | ccv_nnc_tensor_param_t output_params[3]; | |||
| 2717 | const ccv_nnc_cmd_t hyper_connection = CMD_HYPER_CONNECTION_FORWARD(self->count, self->sinkhorn_iterations, self->epsilon)ccv_nnc_cmd(CCV_NNC_HYPER_CONNECTION_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.hyper_connection={.count=(self->count ),.sinkhorn_iterations=(self->sinkhorn_iterations),.epsilon =(self->epsilon)}}), 0); | |||
| 2718 | ccv_nnc_hint_tensor_auto(hyper_connection, input_params, input_size, ccv_nnc_no_hint, output_params, output_size); | |||
| 2719 | for (i = 0; i < output_size; i++) | |||
| 2720 | outputs[i] = ccv_nnc_tensor_symbol_new(graph, output_params[i], 0); | |||
| 2721 | ccv_nnc_graph_exec_symbol_new(graph, hyper_connection, inputs, input_size, outputs, output_size, "hyper_connection"); | |||
| 2722 | } | |||
| 2723 | ||||
| 2724 | static ccv_cnnp_model_t* _ccv_cnnp_hyper_connection_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2725 | ||||
| 2726 | static const ccv_cnnp_model_vtab_t ccv_cnnp_hyper_connection_isa = { | |||
| 2727 | .build = _ccv_cnnp_hyper_connection_build, | |||
| 2728 | .copy = _ccv_cnnp_hyper_connection_copy, | |||
| 2729 | }; | |||
| 2730 | ||||
| 2731 | ccv_cnnp_model_t* ccv_cnnp_hyper_connection(const int count, const int sinkhorn_iterations, const float epsilon, const int operation, const char* const name) | |||
| 2732 | { | |||
| 2733 | assert(count > 0 && count <= 16)((void) sizeof ((count > 0 && count <= 16) ? 1 : 0), __extension__ ({ if (count > 0 && count <= 16) ; else __assert_fail ("count > 0 && count <= 16" , "ccv_cnnp_model_addons.c", 2733, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2734 | assert(sinkhorn_iterations > 0)((void) sizeof ((sinkhorn_iterations > 0) ? 1 : 0), __extension__ ({ if (sinkhorn_iterations > 0) ; else __assert_fail ("sinkhorn_iterations > 0" , "ccv_cnnp_model_addons.c", 2734, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2735 | assert(epsilon >= 0)((void) sizeof ((epsilon >= 0) ? 1 : 0), __extension__ ({ if (epsilon >= 0) ; else __assert_fail ("epsilon >= 0", "ccv_cnnp_model_addons.c" , 2735, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2736 | assert(operation >= 0 && operation <= 2)((void) sizeof ((operation >= 0 && operation <= 2) ? 1 : 0), __extension__ ({ if (operation >= 0 && operation <= 2) ; else __assert_fail ("operation >= 0 && operation <= 2" , "ccv_cnnp_model_addons.c", 2736, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2737 | ccv_cnnp_model_hyper_connection_t* const model_hyper_connection = (ccv_cnnp_model_hyper_connection_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_hyper_connection_t)); | |||
| 2738 | model_hyper_connection->super.isa = &ccv_cnnp_hyper_connection_isa; | |||
| 2739 | model_hyper_connection->super.input_size = operation == 0 ? 3 : 4; | |||
| 2740 | model_hyper_connection->super.outputs = model_hyper_connection->outputs; | |||
| 2741 | model_hyper_connection->super.output_size = operation == 2 ? 1 : 3; | |||
| 2742 | model_hyper_connection->count = count; | |||
| 2743 | model_hyper_connection->sinkhorn_iterations = sinkhorn_iterations; | |||
| 2744 | model_hyper_connection->epsilon = epsilon; | |||
| 2745 | model_hyper_connection->operation = operation; | |||
| 2746 | ccv_cnnp_model_copy_name(&model_hyper_connection->super, name); | |||
| 2747 | return (ccv_cnnp_model_t*)model_hyper_connection; | |||
| 2748 | } | |||
| 2749 | ||||
| 2750 | static ccv_cnnp_model_t* _ccv_cnnp_hyper_connection_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2751 | { | |||
| 2752 | const ccv_cnnp_model_hyper_connection_t* const self = (const ccv_cnnp_model_hyper_connection_t*)super; | |||
| 2753 | return ccv_cnnp_hyper_connection(self->count, self->sinkhorn_iterations, self->epsilon, self->operation, self->super.name); | |||
| 2754 | } | |||
| 2755 | ||||
| 2756 | // MARK - Gated Delta Layer | |||
| 2757 | ||||
| 2758 | typedef struct { | |||
| 2759 | ccv_cnnp_model_t super; | |||
| 2760 | ccv_nnc_tensor_symbol_t outputs[2]; | |||
| 2761 | int log_decay; | |||
| 2762 | int state_checkpoint_count; | |||
| 2763 | } ccv_cnnp_model_gated_delta_t; | |||
| 2764 | ||||
| 2765 | static void _ccv_cnnp_gated_delta_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) | |||
| 2766 | { | |||
| 2767 | PRINT(CCV_CLI_VERBOSE, "[cnnp_gated_delta_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_gated_delta_build] -\n"); fflush(stdout); } } while (0); | |||
| 2768 | ccv_cnnp_model_gated_delta_t* const self = (ccv_cnnp_model_gated_delta_t*)super; | |||
| 2769 | assert(input_size == 6)((void) sizeof ((input_size == 6) ? 1 : 0), __extension__ ({ if (input_size == 6) ; else __assert_fail ("input_size == 6", "ccv_cnnp_model_addons.c" , 2769, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2770 | assert(output_size == 2)((void) sizeof ((output_size == 2) ? 1 : 0), __extension__ ({ if (output_size == 2) ; else __assert_fail ("output_size == 2" , "ccv_cnnp_model_addons.c", 2770, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2771 | ccv_nnc_tensor_param_t input_params[6]; | |||
| 2772 | int i; | |||
| 2773 | for (i = 0; i < 6; i++) | |||
| 2774 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 2775 | ccv_nnc_tensor_param_t output_params[2]; | |||
| 2776 | const ccv_nnc_cmd_t gated_delta = CMD_GATED_DELTA_FORWARD(self->log_decay, self->state_checkpoint_count)ccv_nnc_cmd(CCV_NNC_GATED_DELTA_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.gated_delta={.log_decay=(self->log_decay ),.state_checkpoint_count=(self->state_checkpoint_count)}} ), 0); | |||
| 2777 | ccv_nnc_hint_tensor_auto(gated_delta, input_params, 6, ccv_nnc_no_hint, output_params, 2); | |||
| 2778 | for (i = 0; i < 2; i++) | |||
| 2779 | outputs[i] = ccv_nnc_tensor_symbol_new(graph, output_params[i], 0); | |||
| 2780 | ccv_nnc_graph_exec_symbol_new(graph, gated_delta, inputs, input_size, outputs, output_size, "gated_delta"); | |||
| 2781 | } | |||
| 2782 | ||||
| 2783 | static ccv_cnnp_model_t* _ccv_cnnp_gated_delta_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2784 | ||||
| 2785 | static const ccv_cnnp_model_vtab_t ccv_cnnp_gated_delta_isa = { | |||
| 2786 | .build = _ccv_cnnp_gated_delta_build, | |||
| 2787 | .copy = _ccv_cnnp_gated_delta_copy, | |||
| 2788 | }; | |||
| 2789 | ||||
| 2790 | ccv_cnnp_model_t* ccv_cnnp_gated_delta(const int log_decay, const int state_checkpoint_count, const char* const name) | |||
| 2791 | { | |||
| 2792 | ccv_cnnp_model_gated_delta_t* const model_gated_delta = (ccv_cnnp_model_gated_delta_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_gated_delta_t)); | |||
| 2793 | model_gated_delta->super.isa = &ccv_cnnp_gated_delta_isa; | |||
| 2794 | model_gated_delta->super.input_size = 6; | |||
| 2795 | model_gated_delta->super.outputs = model_gated_delta->outputs; | |||
| 2796 | model_gated_delta->super.output_size = 2; | |||
| 2797 | model_gated_delta->log_decay = log_decay; | |||
| 2798 | model_gated_delta->state_checkpoint_count = state_checkpoint_count; | |||
| 2799 | ccv_cnnp_model_copy_name(&model_gated_delta->super, name); | |||
| 2800 | return (ccv_cnnp_model_t*)model_gated_delta; | |||
| 2801 | } | |||
| 2802 | ||||
| 2803 | static ccv_cnnp_model_t* _ccv_cnnp_gated_delta_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2804 | { | |||
| 2805 | const ccv_cnnp_model_gated_delta_t* const self = (const ccv_cnnp_model_gated_delta_t*)super; | |||
| 2806 | return ccv_cnnp_gated_delta(self->log_decay, self->state_checkpoint_count, super->name); | |||
| 2807 | } | |||
| 2808 | ||||
| 2809 | // MARK - Cmul Layer | |||
| 2810 | ||||
| 2811 | typedef struct { | |||
| 2812 | ccv_cnnp_model_t super; | |||
| 2813 | ccv_nnc_tensor_symbol_t output; | |||
| 2814 | int conjugate; | |||
| 2815 | } ccv_cnnp_model_cmul_t; | |||
| 2816 | ||||
| 2817 | 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) | |||
| 2818 | { | |||
| 2819 | 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); | |||
| 2820 | 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" , 2820, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2821 | 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", 2821, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2822 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 2823 | int i; | |||
| 2824 | for (i = 0; i < 2; i++) | |||
| 2825 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 2826 | ccv_nnc_tensor_param_t output_params; | |||
| 2827 | 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); | |||
| 2828 | mul.info.cmul.conjugate = ((ccv_cnnp_model_cmul_t*)super)->conjugate; | |||
| 2829 | ccv_nnc_hint_tensor_auto(mul, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 2830 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2831 | ccv_nnc_graph_exec_symbol_new(graph, mul, inputs, input_size, outputs, output_size, "cmul"); | |||
| 2832 | } | |||
| 2833 | ||||
| 2834 | static ccv_cnnp_model_t* _ccv_cnnp_cmul_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 2835 | ||||
| 2836 | static const ccv_cnnp_model_vtab_t ccv_cnnp_cmul_isa = { | |||
| 2837 | .build = _ccv_cnnp_cmul_build, | |||
| 2838 | .copy = _ccv_cnnp_cmul_copy, | |||
| 2839 | }; | |||
| 2840 | ||||
| 2841 | ccv_cnnp_model_t* ccv_cnnp_cmul(const int conjugate, const char* const name) | |||
| 2842 | { | |||
| 2843 | assert(conjugate == 0 || conjugate == 1)((void) sizeof ((conjugate == 0 || conjugate == 1) ? 1 : 0), __extension__ ({ if (conjugate == 0 || conjugate == 1) ; else __assert_fail ("conjugate == 0 || conjugate == 1", "ccv_cnnp_model_addons.c" , 2843, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2844 | ccv_cnnp_model_cmul_t* const model_cmul = (ccv_cnnp_model_cmul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_cmul_t)); | |||
| 2845 | model_cmul->super.isa = &ccv_cnnp_cmul_isa; | |||
| 2846 | model_cmul->super.input_size = 2; | |||
| 2847 | model_cmul->super.outputs = &model_cmul->output; | |||
| 2848 | model_cmul->super.output_size = 1; | |||
| 2849 | model_cmul->conjugate = conjugate; | |||
| 2850 | ccv_cnnp_model_copy_name(&model_cmul->super, name); | |||
| 2851 | return (ccv_cnnp_model_t*)model_cmul; | |||
| 2852 | } | |||
| 2853 | ||||
| 2854 | static ccv_cnnp_model_t* _ccv_cnnp_cmul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2855 | { | |||
| 2856 | const ccv_cnnp_model_cmul_t* const self = (const ccv_cnnp_model_cmul_t*)super; | |||
| 2857 | return ccv_cnnp_cmul(self->conjugate, super->name); | |||
| 2858 | } | |||
| 2859 | ||||
| 2860 | // MARK - Transpose Layer | |||
| 2861 | ||||
| 2862 | typedef struct { | |||
| 2863 | ccv_cnnp_model_t super; | |||
| 2864 | ccv_nnc_tensor_symbol_t output; | |||
| 2865 | int transpose[2]; | |||
| 2866 | } ccv_cnnp_model_transpose_t; | |||
| 2867 | ||||
| 2868 | 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) | |||
| 2869 | { | |||
| 2870 | ccv_cnnp_model_transpose_t* const self = (ccv_cnnp_model_transpose_t*)super; | |||
| 2871 | 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); | |||
| 2872 | 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" , 2872, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2873 | 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", 2873, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2874 | if (self->transpose[0] == self->transpose[1]) | |||
| 2875 | { | |||
| 2876 | outputs[0] = inputs[0]; | |||
| 2877 | return; | |||
| 2878 | } | |||
| 2879 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2880 | ccv_nnc_tensor_param_t output_params; | |||
| 2881 | 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); | |||
| 2882 | ccv_nnc_hint_tensor_auto(transpose, (ccv_nnc_tensor_param_t []){ | |||
| 2883 | params, | |||
| 2884 | }, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 2885 | const ccv_nnc_tensor_symbol_t transpose_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 2886 | 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"); | |||
| 2887 | outputs[0] = transpose_output; | |||
| 2888 | } | |||
| 2889 | ||||
| 2890 | static ccv_cnnp_model_t* _ccv_cnnp_transpose_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2891 | ||||
| 2892 | static const ccv_cnnp_model_vtab_t ccv_cnnp_transpose_isa = { | |||
| 2893 | .build = _ccv_cnnp_transpose_build, | |||
| 2894 | .copy = _ccv_cnnp_transpose_copy, | |||
| 2895 | }; | |||
| 2896 | ||||
| 2897 | ccv_cnnp_model_t* ccv_cnnp_transpose(const int axis_a, const int axis_b, const char* const name) | |||
| 2898 | { | |||
| 2899 | ccv_cnnp_model_transpose_t* const model_transpose = (ccv_cnnp_model_transpose_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_transpose_t)); | |||
| 2900 | model_transpose->super.isa = &ccv_cnnp_transpose_isa; | |||
| 2901 | model_transpose->super.input_size = 1; | |||
| 2902 | model_transpose->super.outputs = &model_transpose->output; | |||
| 2903 | model_transpose->super.output_size = 1; | |||
| 2904 | model_transpose->transpose[0] = axis_a; | |||
| 2905 | model_transpose->transpose[1] = axis_b; | |||
| 2906 | ccv_cnnp_model_copy_name(&model_transpose->super, name); | |||
| 2907 | return (ccv_cnnp_model_t*)model_transpose; | |||
| 2908 | } | |||
| 2909 | ||||
| 2910 | static ccv_cnnp_model_t* _ccv_cnnp_transpose_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 2911 | { | |||
| 2912 | const ccv_cnnp_model_transpose_t* const self = (const ccv_cnnp_model_transpose_t*)super; | |||
| 2913 | return ccv_cnnp_transpose(self->transpose[0], self->transpose[1], self->super.name); | |||
| 2914 | } | |||
| 2915 | ||||
| 2916 | // MARK - Layer Norm Layer | |||
| 2917 | ||||
| 2918 | typedef struct { | |||
| 2919 | ccv_cnnp_model_t super; | |||
| 2920 | ccv_nnc_tensor_symbol_t output; | |||
| 2921 | ccv_nnc_tensor_symbol_t bias; | |||
| 2922 | ccv_nnc_tensor_symbol_t scale; | |||
| 2923 | ccv_nnc_cmd_param_t params; | |||
| 2924 | } ccv_cnnp_model_layer_norm_t; | |||
| 2925 | ||||
| 2926 | 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) | |||
| 2927 | { | |||
| 2928 | 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); | |||
| 2929 | 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" , 2929, __extension__ __PRETTY_FUNCTION__); })); | |||
| 2930 | 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", 2930, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 2931 | ccv_cnnp_model_layer_norm_t* const self = (ccv_cnnp_model_layer_norm_t*)super; | |||
| 2932 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 2933 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 2934 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 2935 | int i; | |||
| 2936 | for (i = 0; i < nd; i++) | |||
| 2937 | bias_params.dim[i] = 1; | |||
| 2938 | for (i = 0; i < self->params.lnorm.count; i++) | |||
| 2939 | bias_params.dim[self->params.lnorm.axis[i]] = params.dim[self->params.lnorm.axis[i]]; | |||
| 2940 | if (self->params.lnorm.elementwise_affine) | |||
| 2941 | { | |||
| 2942 | // Both scale and bias are shared between if this model is reused. | |||
| 2943 | if (!self->scale.graph) | |||
| 2944 | self->scale = ccv_nnc_tensor_symbol_new(graph, bias_params, "scale"); | |||
| 2945 | if (!self->bias.graph) | |||
| 2946 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 2947 | } | |||
| 2948 | const ccv_nnc_tensor_symbol_t scale = self->params.lnorm.elementwise_affine ? ccv_cnnp_model_get_symbol(super, self->scale) : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 2949 | const ccv_nnc_tensor_symbol_t bias = self->params.lnorm.elementwise_affine ? ccv_cnnp_model_get_symbol(super, self->bias) : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 2950 | const ccv_nnc_cmd_t layer_norm = ccv_nnc_cmd(CCV_NNC_LAYER_NORM_FORWARD, 0, self->params, 0); | |||
| 2951 | ccv_nnc_tensor_param_t output_params[3]; | |||
| 2952 | if (self->params.lnorm.elementwise_affine) | |||
| 2953 | ccv_nnc_hint_tensor_auto(layer_norm, (ccv_nnc_tensor_param_t []){ | |||
| 2954 | params, | |||
| 2955 | bias_params, | |||
| 2956 | bias_params, | |||
| 2957 | }, 3, ccv_nnc_no_hint, output_params, 3); | |||
| 2958 | else | |||
| 2959 | ccv_nnc_hint_tensor_auto(layer_norm, (ccv_nnc_tensor_param_t []){ | |||
| 2960 | params, | |||
| 2961 | }, 1, ccv_nnc_no_hint, output_params, 3); | |||
| 2962 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 2963 | const ccv_nnc_tensor_symbol_t saved_mean = ccv_nnc_tensor_symbol_new(graph, output_params[1], "saved_mean"); | |||
| 2964 | const ccv_nnc_tensor_symbol_t saved_inv_std = ccv_nnc_tensor_symbol_new(graph, output_params[2], "saved_inv_std"); | |||
| 2965 | if (self->params.lnorm.elementwise_affine) | |||
| 2966 | ccv_nnc_graph_exec_symbol_new(graph, layer_norm, TENSOR_SYMBOL_LIST(inputs[0], scale, bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], scale, 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"); | |||
| 2967 | else | |||
| 2968 | 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"); | |||
| 2969 | outputs[0] = output; | |||
| 2970 | } | |||
| 2971 | ||||
| 2972 | 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) | |||
| 2973 | { | |||
| 2974 | ccv_cnnp_model_layer_norm_t* const self = (ccv_cnnp_model_layer_norm_t*)super; | |||
| 2975 | if (self->scale.graph) | |||
| 2976 | 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); | |||
| 2977 | if (self->bias.graph) | |||
| 2978 | 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); | |||
| 2979 | } | |||
| 2980 | ||||
| 2981 | 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) | |||
| 2982 | { | |||
| 2983 | ccv_cnnp_model_layer_norm_t* const self = (ccv_cnnp_model_layer_norm_t*)super; | |||
| 2984 | if (self->scale.graph) | |||
| 2985 | add_to_array(parameters, self->scale, is_trainable); | |||
| 2986 | if (self->bias.graph) | |||
| 2987 | add_to_array(parameters, self->bias, is_trainable); | |||
| 2988 | } | |||
| 2989 | ||||
| 2990 | static ccv_cnnp_model_t* _ccv_cnnp_layer_norm_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 2991 | ||||
| 2992 | static const ccv_cnnp_model_vtab_t ccv_cnnp_layer_norm_isa = { | |||
| 2993 | .build = _ccv_cnnp_layer_norm_build, | |||
| 2994 | .init_states = _ccv_cnnp_layer_norm_init_states, | |||
| 2995 | .add_to_parameter = _ccv_cnnp_layer_norm_add_to_parameter, | |||
| 2996 | .copy = _ccv_cnnp_layer_norm_copy, | |||
| 2997 | }; | |||
| 2998 | ||||
| 2999 | 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 float scale, const int is_trainable, const char* const name) | |||
| 3000 | { | |||
| 3001 | 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)); | |||
| 3002 | model_layer_norm->super.isa = &ccv_cnnp_layer_norm_isa; | |||
| 3003 | model_layer_norm->super.input_size = 1; | |||
| 3004 | model_layer_norm->super.outputs = &model_layer_norm->output; | |||
| 3005 | model_layer_norm->super.output_size = 1; | |||
| 3006 | model_layer_norm->super.is_trainable = is_trainable; | |||
| 3007 | ccv_cnnp_model_copy_name(&model_layer_norm->super, name); | |||
| 3008 | model_layer_norm->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3009 | model_layer_norm->scale.graph = 0; | |||
| 3010 | model_layer_norm->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3011 | model_layer_norm->bias.graph = 0; | |||
| 3012 | model_layer_norm->params.lnorm.epsilon = epsilon; | |||
| 3013 | model_layer_norm->params.lnorm.scale = scale; | |||
| 3014 | model_layer_norm->params.lnorm.count = axis_count; | |||
| 3015 | model_layer_norm->params.lnorm.elementwise_affine = elementwise_affine; | |||
| 3016 | memcpy(model_layer_norm->params.lnorm.axis, axis, sizeof(int) * axis_count); | |||
| 3017 | return (ccv_cnnp_model_t*)model_layer_norm; | |||
| 3018 | } | |||
| 3019 | ||||
| 3020 | static ccv_cnnp_model_t* _ccv_cnnp_layer_norm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3021 | { | |||
| 3022 | const ccv_cnnp_model_layer_norm_t* const self = (const ccv_cnnp_model_layer_norm_t*)super; | |||
| 3023 | return ccv_cnnp_layer_norm(self->params.lnorm.epsilon, self->params.lnorm.axis, self->params.lnorm.count, self->params.lnorm.elementwise_affine, self->params.lnorm.scale, self->super.is_trainable, self->super.name); | |||
| 3024 | } | |||
| 3025 | ||||
| 3026 | // MARK - Group Norm Layer | |||
| 3027 | ||||
| 3028 | typedef struct { | |||
| 3029 | ccv_cnnp_model_t super; | |||
| 3030 | ccv_nnc_tensor_symbol_t output; | |||
| 3031 | ccv_nnc_tensor_symbol_t bias; | |||
| 3032 | ccv_nnc_tensor_symbol_t scale; | |||
| 3033 | ccv_nnc_cmd_param_t params; | |||
| 3034 | } ccv_cnnp_model_group_norm_t; | |||
| 3035 | ||||
| 3036 | 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) | |||
| 3037 | { | |||
| 3038 | 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); | |||
| 3039 | 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" , 3039, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3040 | 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", 3040, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3041 | ccv_cnnp_model_group_norm_t* const self = (ccv_cnnp_model_group_norm_t*)super; | |||
| 3042 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3043 | ccv_nnc_tensor_param_t bias_params = params; | |||
| 3044 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 3045 | int i; | |||
| 3046 | for (i = 0; i < nd; i++) | |||
| 3047 | bias_params.dim[i] = 1; | |||
| 3048 | bias_params.dim[self->params.gnorm.group_axis] = params.dim[self->params.gnorm.group_axis]; | |||
| 3049 | if (self->params.gnorm.elementwise_affine) | |||
| 3050 | { | |||
| 3051 | // Both scale and bias are shared between if this model is reused. | |||
| 3052 | if (!self->scale.graph) | |||
| 3053 | self->scale = ccv_nnc_tensor_symbol_new(graph, bias_params, "scale"); | |||
| 3054 | if (!self->bias.graph) | |||
| 3055 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 3056 | } | |||
| 3057 | const ccv_nnc_tensor_symbol_t scale = self->params.gnorm.elementwise_affine ? ccv_cnnp_model_get_symbol(super, self->scale) : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3058 | const ccv_nnc_tensor_symbol_t bias = self->params.gnorm.elementwise_affine ? ccv_cnnp_model_get_symbol(super, self->bias) : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3059 | const ccv_nnc_cmd_t group_norm = ccv_nnc_cmd(CCV_NNC_GROUP_NORM_FORWARD, 0, self->params, 0); | |||
| 3060 | ccv_nnc_tensor_param_t output_params[3]; | |||
| 3061 | if (self->params.gnorm.elementwise_affine) | |||
| 3062 | ccv_nnc_hint_tensor_auto(group_norm, (ccv_nnc_tensor_param_t []){ | |||
| 3063 | params, | |||
| 3064 | bias_params, | |||
| 3065 | bias_params, | |||
| 3066 | }, 3, ccv_nnc_no_hint, output_params, 3); | |||
| 3067 | else | |||
| 3068 | ccv_nnc_hint_tensor_auto(group_norm, (ccv_nnc_tensor_param_t []){ | |||
| 3069 | params, | |||
| 3070 | }, 1, ccv_nnc_no_hint, output_params, 3); | |||
| 3071 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 3072 | const ccv_nnc_tensor_symbol_t saved_mean = ccv_nnc_tensor_symbol_new(graph, output_params[1], "saved_mean"); | |||
| 3073 | const ccv_nnc_tensor_symbol_t saved_inv_std = ccv_nnc_tensor_symbol_new(graph, output_params[2], "saved_inv_std"); | |||
| 3074 | if (self->params.gnorm.elementwise_affine) | |||
| 3075 | ccv_nnc_graph_exec_symbol_new(graph, group_norm, TENSOR_SYMBOL_LIST(inputs[0], scale, bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], scale, 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"); | |||
| 3076 | else | |||
| 3077 | 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"); | |||
| 3078 | outputs[0] = output; | |||
| 3079 | } | |||
| 3080 | ||||
| 3081 | 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) | |||
| 3082 | { | |||
| 3083 | ccv_cnnp_model_group_norm_t* const self = (ccv_cnnp_model_group_norm_t*)super; | |||
| 3084 | if (self->scale.graph) | |||
| 3085 | 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); | |||
| 3086 | if (self->bias.graph) | |||
| 3087 | 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); | |||
| 3088 | } | |||
| 3089 | ||||
| 3090 | 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) | |||
| 3091 | { | |||
| 3092 | ccv_cnnp_model_group_norm_t* const self = (ccv_cnnp_model_group_norm_t*)super; | |||
| 3093 | if (self->scale.graph) | |||
| 3094 | add_to_array(parameters, self->scale, is_trainable); | |||
| 3095 | if (self->bias.graph) | |||
| 3096 | add_to_array(parameters, self->bias, is_trainable); | |||
| 3097 | } | |||
| 3098 | ||||
| 3099 | static ccv_cnnp_model_t* _ccv_cnnp_group_norm_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3100 | ||||
| 3101 | static const ccv_cnnp_model_vtab_t ccv_cnnp_group_norm_isa = { | |||
| 3102 | .build = _ccv_cnnp_group_norm_build, | |||
| 3103 | .init_states = _ccv_cnnp_group_norm_init_states, | |||
| 3104 | .add_to_parameter = _ccv_cnnp_group_norm_add_to_parameter, | |||
| 3105 | .copy = _ccv_cnnp_group_norm_copy, | |||
| 3106 | }; | |||
| 3107 | ||||
| 3108 | 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) | |||
| 3109 | { | |||
| 3110 | 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)); | |||
| 3111 | model_group_norm->super.isa = &ccv_cnnp_group_norm_isa; | |||
| 3112 | model_group_norm->super.input_size = 1; | |||
| 3113 | model_group_norm->super.outputs = &model_group_norm->output; | |||
| 3114 | model_group_norm->super.output_size = 1; | |||
| 3115 | model_group_norm->super.is_trainable = is_trainable; | |||
| 3116 | ccv_cnnp_model_copy_name(&model_group_norm->super, name); | |||
| 3117 | model_group_norm->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3118 | model_group_norm->scale.graph = 0; | |||
| 3119 | model_group_norm->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3120 | model_group_norm->bias.graph = 0; | |||
| 3121 | model_group_norm->params.gnorm.group_axis = group_axis; | |||
| 3122 | model_group_norm->params.gnorm.groups = groups; | |||
| 3123 | model_group_norm->params.gnorm.epsilon = epsilon; | |||
| 3124 | model_group_norm->params.gnorm.reduce_count = axis_count; | |||
| 3125 | model_group_norm->params.gnorm.elementwise_affine = elementwise_affine; | |||
| 3126 | memcpy(model_group_norm->params.gnorm.reduce_axis, reduce_axis, sizeof(int) * axis_count); | |||
| 3127 | return (ccv_cnnp_model_t*)model_group_norm; | |||
| 3128 | } | |||
| 3129 | ||||
| 3130 | static ccv_cnnp_model_t* _ccv_cnnp_group_norm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3131 | { | |||
| 3132 | const ccv_cnnp_model_group_norm_t* const self = (const ccv_cnnp_model_group_norm_t*)super; | |||
| 3133 | 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); | |||
| 3134 | } | |||
| 3135 | ||||
| 3136 | // MARK - RMSNorm Layer | |||
| 3137 | ||||
| 3138 | typedef struct { | |||
| 3139 | ccv_cnnp_model_t super; | |||
| 3140 | ccv_nnc_tensor_symbol_t output; | |||
| 3141 | ccv_nnc_tensor_symbol_t scale; | |||
| 3142 | ccv_nnc_cmd_param_t params; | |||
| 3143 | } ccv_cnnp_model_rmsnorm_t; | |||
| 3144 | ||||
| 3145 | 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) | |||
| 3146 | { | |||
| 3147 | 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); | |||
| 3148 | 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" , 3148, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3149 | 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", 3149, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3150 | ccv_cnnp_model_rmsnorm_t* const self = (ccv_cnnp_model_rmsnorm_t*)super; | |||
| 3151 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3152 | ccv_nnc_tensor_param_t scale_params = params; | |||
| 3153 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 3154 | int i; | |||
| 3155 | for (i = 0; i < nd; i++) | |||
| 3156 | scale_params.dim[i] = 1; | |||
| 3157 | for (i = 0; i < self->params.rmsnorm.count; i++) | |||
| 3158 | scale_params.dim[self->params.rmsnorm.axis[i]] = params.dim[self->params.rmsnorm.axis[i]]; | |||
| 3159 | // Both scale and bias are shared between if this model is reused. | |||
| 3160 | if (self->params.rmsnorm.elementwise_affine) | |||
| 3161 | { | |||
| 3162 | if (!self->scale.graph) | |||
| 3163 | self->scale = ccv_nnc_tensor_symbol_new(graph, scale_params, "scale"); | |||
| 3164 | } | |||
| 3165 | const ccv_nnc_tensor_symbol_t scale = self->params.rmsnorm.elementwise_affine ? ccv_cnnp_model_get_symbol(super, self->scale) : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3166 | const ccv_nnc_cmd_t rmsnorm = ccv_nnc_cmd(CCV_NNC_RMSNORM_FORWARD, 0, self->params, 0); | |||
| 3167 | ccv_nnc_tensor_param_t output_params[2]; | |||
| 3168 | if (self->params.rmsnorm.elementwise_affine) | |||
| 3169 | ccv_nnc_hint_tensor_auto(rmsnorm, (ccv_nnc_tensor_param_t []){ | |||
| 3170 | params, | |||
| 3171 | scale_params, | |||
| 3172 | }, 2, ccv_nnc_no_hint, output_params, 2); | |||
| 3173 | else | |||
| 3174 | ccv_nnc_hint_tensor_auto(rmsnorm, (ccv_nnc_tensor_param_t []){ | |||
| 3175 | params, | |||
| 3176 | }, 1, ccv_nnc_no_hint, output_params, 2); | |||
| 3177 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 3178 | const ccv_nnc_tensor_symbol_t saved_inv_std = ccv_nnc_tensor_symbol_new(graph, output_params[1], "saved_inv_std"); | |||
| 3179 | if (self->params.rmsnorm.elementwise_affine) | |||
| 3180 | ccv_nnc_graph_exec_symbol_new(graph, rmsnorm, TENSOR_SYMBOL_LIST(inputs[0], scale)(const ccv_nnc_tensor_symbol_t []){inputs[0], 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"); | |||
| 3181 | else | |||
| 3182 | ccv_nnc_graph_exec_symbol_new(graph, rmsnorm, 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_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"); | |||
| 3183 | outputs[0] = output; | |||
| 3184 | } | |||
| 3185 | ||||
| 3186 | 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) | |||
| 3187 | { | |||
| 3188 | ccv_cnnp_model_rmsnorm_t* const self = (ccv_cnnp_model_rmsnorm_t*)super; | |||
| 3189 | if (self->scale.graph) | |||
| 3190 | 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); | |||
| 3191 | } | |||
| 3192 | ||||
| 3193 | 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) | |||
| 3194 | { | |||
| 3195 | ccv_cnnp_model_rmsnorm_t* const self = (ccv_cnnp_model_rmsnorm_t*)super; | |||
| 3196 | if (self->scale.graph) | |||
| 3197 | add_to_array(parameters, self->scale, is_trainable); | |||
| 3198 | } | |||
| 3199 | ||||
| 3200 | static ccv_cnnp_model_t* _ccv_cnnp_rmsnorm_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3201 | ||||
| 3202 | static const ccv_cnnp_model_vtab_t ccv_cnnp_rmsnorm_isa = { | |||
| 3203 | .build = _ccv_cnnp_rmsnorm_build, | |||
| 3204 | .init_states = _ccv_cnnp_rmsnorm_init_states, | |||
| 3205 | .add_to_parameter = _ccv_cnnp_rmsnorm_add_to_parameter, | |||
| 3206 | .copy = _ccv_cnnp_rmsnorm_copy, | |||
| 3207 | }; | |||
| 3208 | ||||
| 3209 | 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 elementwise_affine, const float scale, const int is_trainable, const char* const name) | |||
| 3210 | { | |||
| 3211 | ccv_cnnp_model_rmsnorm_t* const model_rmsnorm = (ccv_cnnp_model_rmsnorm_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_rmsnorm_t)); | |||
| 3212 | model_rmsnorm->super.isa = &ccv_cnnp_rmsnorm_isa; | |||
| 3213 | model_rmsnorm->super.input_size = 1; | |||
| 3214 | model_rmsnorm->super.outputs = &model_rmsnorm->output; | |||
| 3215 | model_rmsnorm->super.output_size = 1; | |||
| 3216 | model_rmsnorm->super.is_trainable = is_trainable; | |||
| 3217 | ccv_cnnp_model_copy_name(&model_rmsnorm->super, name); | |||
| 3218 | model_rmsnorm->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3219 | model_rmsnorm->scale.graph = 0; | |||
| 3220 | model_rmsnorm->params.rmsnorm.epsilon = epsilon; | |||
| 3221 | model_rmsnorm->params.rmsnorm.scale = scale; | |||
| 3222 | model_rmsnorm->params.rmsnorm.count = axis_count; | |||
| 3223 | model_rmsnorm->params.rmsnorm.elementwise_affine = elementwise_affine; | |||
| 3224 | memcpy(model_rmsnorm->params.lnorm.axis, axis, sizeof(int) * axis_count); | |||
| 3225 | return (ccv_cnnp_model_t*)model_rmsnorm; | |||
| 3226 | } | |||
| 3227 | ||||
| 3228 | static ccv_cnnp_model_t* _ccv_cnnp_rmsnorm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3229 | { | |||
| 3230 | const ccv_cnnp_model_rmsnorm_t* const self = (const ccv_cnnp_model_rmsnorm_t*)super; | |||
| 3231 | return ccv_cnnp_rmsnorm(self->params.rmsnorm.epsilon, self->params.rmsnorm.axis, self->params.rmsnorm.count, self->params.rmsnorm.elementwise_affine, self->params.rmsnorm.scale, self->super.is_trainable, self->super.name); | |||
| 3232 | } | |||
| 3233 | ||||
| 3234 | // MARK - RMSNorm CMul Layer | |||
| 3235 | ||||
| 3236 | typedef struct { | |||
| 3237 | ccv_cnnp_model_t super; | |||
| 3238 | ccv_nnc_tensor_symbol_t output; | |||
| 3239 | ccv_nnc_tensor_symbol_t scale; | |||
| 3240 | ccv_nnc_cmd_param_t params; | |||
| 3241 | } ccv_cnnp_model_rmsnorm_cmul_t; | |||
| 3242 | ||||
| 3243 | static void _ccv_cnnp_rmsnorm_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) | |||
| 3244 | { | |||
| 3245 | PRINT(CCV_CLI_VERBOSE, "[cnnp_rmsnorm_cmul_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_rmsnorm_cmul_build] -\n"); fflush(stdout); } } while (0); | |||
| 3246 | 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" , 3246, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3247 | 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", 3247, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3248 | ccv_cnnp_model_rmsnorm_cmul_t* const self = (ccv_cnnp_model_rmsnorm_cmul_t*)super; | |||
| 3249 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3250 | const ccv_nnc_tensor_param_t rotation_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 3251 | ccv_nnc_tensor_param_t scale_params = params; | |||
| 3252 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 3253 | int i; | |||
| 3254 | for (i = 0; i < nd; i++) | |||
| 3255 | scale_params.dim[i] = 1; | |||
| 3256 | for (i = 0; i < self->params.rmsnorm_cmul.count; i++) | |||
| 3257 | scale_params.dim[self->params.rmsnorm_cmul.axis[i]] = params.dim[self->params.rmsnorm_cmul.axis[i]]; | |||
| 3258 | if (self->params.rmsnorm_cmul.elementwise_affine) | |||
| 3259 | { | |||
| 3260 | if (!self->scale.graph) | |||
| 3261 | self->scale = ccv_nnc_tensor_symbol_new(graph, scale_params, "scale"); | |||
| 3262 | } | |||
| 3263 | const ccv_nnc_tensor_symbol_t scale = self->params.rmsnorm_cmul.elementwise_affine ? ccv_cnnp_model_get_symbol(super, self->scale) : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3264 | const ccv_nnc_cmd_t rmsnorm_cmul = ccv_nnc_cmd(CCV_NNC_RMSNORM_CMUL_FORWARD, 0, self->params, 0); | |||
| 3265 | ccv_nnc_tensor_param_t output_params; | |||
| 3266 | if (self->params.rmsnorm_cmul.elementwise_affine) | |||
| 3267 | ccv_nnc_hint_tensor_auto(rmsnorm_cmul, (ccv_nnc_tensor_param_t []){ | |||
| 3268 | params, | |||
| 3269 | rotation_params, | |||
| 3270 | scale_params, | |||
| 3271 | }, 3, ccv_nnc_no_hint, &output_params, 1); | |||
| 3272 | else | |||
| 3273 | ccv_nnc_hint_tensor_auto(rmsnorm_cmul, (ccv_nnc_tensor_param_t []){ | |||
| 3274 | params, | |||
| 3275 | rotation_params, | |||
| 3276 | }, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 3277 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3278 | if (self->params.rmsnorm_cmul.elementwise_affine) | |||
| 3279 | ccv_nnc_graph_exec_symbol_new(graph, rmsnorm_cmul, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], scale)(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], scale }, (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), outputs, output_size, "rmsnorm_cmul"); | |||
| 3280 | else | |||
| 3281 | ccv_nnc_graph_exec_symbol_new(graph, rmsnorm_cmul, inputs, input_size, outputs, output_size, "rmsnorm_cmul"); | |||
| 3282 | } | |||
| 3283 | ||||
| 3284 | static void _ccv_cnnp_rmsnorm_cmul_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) | |||
| 3285 | { | |||
| 3286 | ccv_cnnp_model_rmsnorm_cmul_t* const self = (ccv_cnnp_model_rmsnorm_cmul_t*)super; | |||
| 3287 | if (self->scale.graph) | |||
| 3288 | 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); | |||
| 3289 | } | |||
| 3290 | ||||
| 3291 | static void _ccv_cnnp_rmsnorm_cmul_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) | |||
| 3292 | { | |||
| 3293 | ccv_cnnp_model_rmsnorm_cmul_t* const self = (ccv_cnnp_model_rmsnorm_cmul_t*)super; | |||
| 3294 | if (self->scale.graph) | |||
| 3295 | add_to_array(parameters, self->scale, is_trainable); | |||
| 3296 | } | |||
| 3297 | ||||
| 3298 | static ccv_cnnp_model_t* _ccv_cnnp_rmsnorm_cmul_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3299 | ||||
| 3300 | static const ccv_cnnp_model_vtab_t ccv_cnnp_rmsnorm_cmul_isa = { | |||
| 3301 | .build = _ccv_cnnp_rmsnorm_cmul_build, | |||
| 3302 | .init_states = _ccv_cnnp_rmsnorm_cmul_init_states, | |||
| 3303 | .add_to_parameter = _ccv_cnnp_rmsnorm_cmul_add_to_parameter, | |||
| 3304 | .copy = _ccv_cnnp_rmsnorm_cmul_copy, | |||
| 3305 | }; | |||
| 3306 | ||||
| 3307 | ccv_cnnp_model_t* ccv_cnnp_rmsnorm_cmul(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) | |||
| 3308 | { | |||
| 3309 | ccv_cnnp_model_rmsnorm_cmul_t* const model_rmsnorm_cmul = (ccv_cnnp_model_rmsnorm_cmul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_rmsnorm_cmul_t)); | |||
| 3310 | model_rmsnorm_cmul->super.isa = &ccv_cnnp_rmsnorm_cmul_isa; | |||
| 3311 | model_rmsnorm_cmul->super.input_size = 2; | |||
| 3312 | model_rmsnorm_cmul->super.outputs = &model_rmsnorm_cmul->output; | |||
| 3313 | model_rmsnorm_cmul->super.output_size = 1; | |||
| 3314 | model_rmsnorm_cmul->super.is_trainable = is_trainable; | |||
| 3315 | ccv_cnnp_model_copy_name(&model_rmsnorm_cmul->super, name); | |||
| 3316 | model_rmsnorm_cmul->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3317 | model_rmsnorm_cmul->scale.graph = 0; | |||
| 3318 | model_rmsnorm_cmul->params.rmsnorm_cmul.epsilon = epsilon; | |||
| 3319 | model_rmsnorm_cmul->params.rmsnorm_cmul.count = axis_count; | |||
| 3320 | model_rmsnorm_cmul->params.rmsnorm_cmul.elementwise_affine = elementwise_affine; | |||
| 3321 | memcpy(model_rmsnorm_cmul->params.rmsnorm_cmul.axis, axis, sizeof(int) * axis_count); | |||
| 3322 | return (ccv_cnnp_model_t*)model_rmsnorm_cmul; | |||
| 3323 | } | |||
| 3324 | ||||
| 3325 | static ccv_cnnp_model_t* _ccv_cnnp_rmsnorm_cmul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3326 | { | |||
| 3327 | const ccv_cnnp_model_rmsnorm_cmul_t* const self = (const ccv_cnnp_model_rmsnorm_cmul_t*)super; | |||
| 3328 | return ccv_cnnp_rmsnorm_cmul(self->params.rmsnorm_cmul.epsilon, self->params.rmsnorm_cmul.axis, self->params.rmsnorm_cmul.count, self->params.rmsnorm_cmul.elementwise_affine, self->super.is_trainable, self->super.name); | |||
| 3329 | } | |||
| 3330 | ||||
| 3331 | // MARK - RMSNorm Gated Layer | |||
| 3332 | ||||
| 3333 | typedef struct { | |||
| 3334 | ccv_cnnp_model_t super; | |||
| 3335 | ccv_nnc_tensor_symbol_t output; | |||
| 3336 | ccv_nnc_tensor_symbol_t scale; | |||
| 3337 | ccv_nnc_cmd_param_t params; | |||
| 3338 | } ccv_cnnp_model_rmsnorm_gated_t; | |||
| 3339 | ||||
| 3340 | static void _ccv_cnnp_rmsnorm_gated_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) | |||
| 3341 | { | |||
| 3342 | PRINT(CCV_CLI_VERBOSE, "[cnnp_rmsnorm_gated_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_rmsnorm_gated_build] -\n"); fflush(stdout); } } while (0); | |||
| 3343 | 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" , 3343, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3344 | 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", 3344, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3345 | ccv_cnnp_model_rmsnorm_gated_t* const self = (ccv_cnnp_model_rmsnorm_gated_t*)super; | |||
| 3346 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3347 | const ccv_nnc_tensor_param_t gate_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 3348 | ccv_nnc_tensor_param_t scale_params = params; | |||
| 3349 | const int nd = ccv_nnc_tensor_nd(params.dim); | |||
| 3350 | int i; | |||
| 3351 | for (i = 0; i < nd; i++) | |||
| 3352 | scale_params.dim[i] = 1; | |||
| 3353 | for (i = 0; i < self->params.rmsnorm_gated.count; i++) | |||
| 3354 | scale_params.dim[self->params.rmsnorm_gated.axis[i]] = params.dim[self->params.rmsnorm_gated.axis[i]]; | |||
| 3355 | if (self->params.rmsnorm_gated.elementwise_affine) | |||
| 3356 | { | |||
| 3357 | if (!self->scale.graph) | |||
| 3358 | self->scale = ccv_nnc_tensor_symbol_new(graph, scale_params, "scale"); | |||
| 3359 | } | |||
| 3360 | const ccv_nnc_tensor_symbol_t scale = self->params.rmsnorm_gated.elementwise_affine ? ccv_cnnp_model_get_symbol(super, self->scale) : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 3361 | const ccv_nnc_cmd_t rmsnorm_gated = ccv_nnc_cmd(CCV_NNC_RMSNORM_GATED_FORWARD, 0, self->params, 0); | |||
| 3362 | ccv_nnc_tensor_param_t output_params; | |||
| 3363 | if (self->params.rmsnorm_gated.elementwise_affine) | |||
| 3364 | ccv_nnc_hint_tensor_auto(rmsnorm_gated, (ccv_nnc_tensor_param_t []){ | |||
| 3365 | params, | |||
| 3366 | gate_params, | |||
| 3367 | scale_params, | |||
| 3368 | }, 3, ccv_nnc_no_hint, &output_params, 1); | |||
| 3369 | else | |||
| 3370 | ccv_nnc_hint_tensor_auto(rmsnorm_gated, (ccv_nnc_tensor_param_t []){ | |||
| 3371 | params, | |||
| 3372 | gate_params, | |||
| 3373 | }, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 3374 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3375 | if (self->params.rmsnorm_gated.elementwise_affine) | |||
| 3376 | ccv_nnc_graph_exec_symbol_new(graph, rmsnorm_gated, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], scale)(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], scale }, (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), outputs, output_size, "rmsnorm_gated"); | |||
| 3377 | else | |||
| 3378 | ccv_nnc_graph_exec_symbol_new(graph, rmsnorm_gated, inputs, input_size, outputs, output_size, "rmsnorm_gated"); | |||
| 3379 | } | |||
| 3380 | ||||
| 3381 | static void _ccv_cnnp_rmsnorm_gated_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) | |||
| 3382 | { | |||
| 3383 | ccv_cnnp_model_rmsnorm_gated_t* const self = (ccv_cnnp_model_rmsnorm_gated_t*)super; | |||
| 3384 | if (self->scale.graph) | |||
| 3385 | 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); | |||
| 3386 | } | |||
| 3387 | ||||
| 3388 | static void _ccv_cnnp_rmsnorm_gated_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) | |||
| 3389 | { | |||
| 3390 | ccv_cnnp_model_rmsnorm_gated_t* const self = (ccv_cnnp_model_rmsnorm_gated_t*)super; | |||
| 3391 | if (self->scale.graph) | |||
| 3392 | add_to_array(parameters, self->scale, is_trainable); | |||
| 3393 | } | |||
| 3394 | ||||
| 3395 | static ccv_cnnp_model_t* _ccv_cnnp_rmsnorm_gated_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3396 | ||||
| 3397 | static const ccv_cnnp_model_vtab_t ccv_cnnp_rmsnorm_gated_isa = { | |||
| 3398 | .build = _ccv_cnnp_rmsnorm_gated_build, | |||
| 3399 | .init_states = _ccv_cnnp_rmsnorm_gated_init_states, | |||
| 3400 | .add_to_parameter = _ccv_cnnp_rmsnorm_gated_add_to_parameter, | |||
| 3401 | .copy = _ccv_cnnp_rmsnorm_gated_copy, | |||
| 3402 | }; | |||
| 3403 | ||||
| 3404 | ccv_cnnp_model_t* ccv_cnnp_rmsnorm_gated(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) | |||
| 3405 | { | |||
| 3406 | ccv_cnnp_model_rmsnorm_gated_t* const model_rmsnorm_gated = (ccv_cnnp_model_rmsnorm_gated_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_rmsnorm_gated_t)); | |||
| 3407 | model_rmsnorm_gated->super.isa = &ccv_cnnp_rmsnorm_gated_isa; | |||
| 3408 | model_rmsnorm_gated->super.input_size = 2; | |||
| 3409 | model_rmsnorm_gated->super.outputs = &model_rmsnorm_gated->output; | |||
| 3410 | model_rmsnorm_gated->super.output_size = 1; | |||
| 3411 | model_rmsnorm_gated->super.is_trainable = is_trainable; | |||
| 3412 | ccv_cnnp_model_copy_name(&model_rmsnorm_gated->super, name); | |||
| 3413 | model_rmsnorm_gated->scale.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3414 | model_rmsnorm_gated->scale.graph = 0; | |||
| 3415 | model_rmsnorm_gated->params.rmsnorm_gated.epsilon = epsilon; | |||
| 3416 | model_rmsnorm_gated->params.rmsnorm_gated.count = axis_count; | |||
| 3417 | model_rmsnorm_gated->params.rmsnorm_gated.elementwise_affine = elementwise_affine; | |||
| 3418 | memcpy(model_rmsnorm_gated->params.rmsnorm_gated.axis, axis, sizeof(int) * axis_count); | |||
| 3419 | return (ccv_cnnp_model_t*)model_rmsnorm_gated; | |||
| 3420 | } | |||
| 3421 | ||||
| 3422 | static ccv_cnnp_model_t* _ccv_cnnp_rmsnorm_gated_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3423 | { | |||
| 3424 | const ccv_cnnp_model_rmsnorm_gated_t* const self = (const ccv_cnnp_model_rmsnorm_gated_t*)super; | |||
| 3425 | return ccv_cnnp_rmsnorm_gated(self->params.rmsnorm_gated.epsilon, self->params.rmsnorm_gated.axis, self->params.rmsnorm_gated.count, self->params.rmsnorm_gated.elementwise_affine, self->super.is_trainable, self->super.name); | |||
| 3426 | } | |||
| 3427 | ||||
| 3428 | // MARK - Batched Matrix Mul Layer | |||
| 3429 | ||||
| 3430 | typedef struct { | |||
| 3431 | ccv_cnnp_model_t super; | |||
| 3432 | ccv_nnc_tensor_symbol_t output; | |||
| 3433 | int transpose_a[2]; | |||
| 3434 | int transpose_b[2]; | |||
| 3435 | int flags; | |||
| 3436 | } ccv_cnnp_model_matmul_t; | |||
| 3437 | ||||
| 3438 | 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) | |||
| 3439 | { | |||
| 3440 | 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); | |||
| 3441 | 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" , 3441, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3442 | 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", 3442, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3443 | ccv_cnnp_model_matmul_t* const self = (ccv_cnnp_model_matmul_t*)super; | |||
| 3444 | ccv_nnc_tensor_param_t a_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3445 | ccv_nnc_tensor_param_t b_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 3446 | ccv_nnc_tensor_param_t output_params; | |||
| 3447 | 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); | |||
| 3448 | matmul.info.blas.flags = self->flags; | |||
| 3449 | ccv_nnc_hint_tensor_auto(matmul, (ccv_nnc_tensor_param_t []){ | |||
| 3450 | a_params, | |||
| 3451 | b_params, | |||
| 3452 | }, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 3453 | const ccv_nnc_tensor_symbol_t matmul_output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3454 | 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"); | |||
| 3455 | outputs[0] = matmul_output; | |||
| 3456 | } | |||
| 3457 | ||||
| 3458 | static ccv_cnnp_model_t* _ccv_cnnp_matmul_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3459 | ||||
| 3460 | static const ccv_cnnp_model_vtab_t ccv_cnnp_matmul_isa = { | |||
| 3461 | .build = _ccv_cnnp_matmul_build, | |||
| 3462 | .copy = _ccv_cnnp_matmul_copy, | |||
| 3463 | }; | |||
| 3464 | ||||
| 3465 | ccv_cnnp_model_t* ccv_cnnp_matmul(const int transpose_a[2], const int transpose_b[2], const int flags, const char* const name) | |||
| 3466 | { | |||
| 3467 | ccv_cnnp_model_matmul_t* const model_matmul = (ccv_cnnp_model_matmul_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_matmul_t)); | |||
| 3468 | model_matmul->super.isa = &ccv_cnnp_matmul_isa; | |||
| 3469 | model_matmul->super.input_size = 2; | |||
| 3470 | model_matmul->super.outputs = &model_matmul->output; | |||
| 3471 | model_matmul->super.output_size = 1; | |||
| 3472 | model_matmul->transpose_a[0] = transpose_a[0]; | |||
| 3473 | model_matmul->transpose_a[1] = transpose_a[1]; | |||
| 3474 | model_matmul->transpose_b[0] = transpose_b[0]; | |||
| 3475 | model_matmul->transpose_b[1] = transpose_b[1]; | |||
| 3476 | model_matmul->flags = flags; | |||
| 3477 | ccv_cnnp_model_copy_name(&model_matmul->super, name); | |||
| 3478 | return (ccv_cnnp_model_t*)model_matmul; | |||
| 3479 | } | |||
| 3480 | ||||
| 3481 | static ccv_cnnp_model_t* _ccv_cnnp_matmul_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3482 | { | |||
| 3483 | const ccv_cnnp_model_matmul_t* const self = (const ccv_cnnp_model_matmul_t*)super; | |||
| 3484 | return ccv_cnnp_matmul(self->transpose_a, self->transpose_b, self->flags, self->super.name); | |||
| 3485 | } | |||
| 3486 | ||||
| 3487 | // MARK - Dropout Layer | |||
| 3488 | ||||
| 3489 | typedef struct { | |||
| 3490 | ccv_cnnp_model_t super; | |||
| 3491 | ccv_nnc_tensor_symbol_t output; | |||
| 3492 | ccv_nnc_graph_exec_symbol_t dropout; | |||
| 3493 | float p; | |||
| 3494 | int entirety; | |||
| 3495 | } ccv_cnnp_model_dropout_t; | |||
| 3496 | ||||
| 3497 | 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) | |||
| 3498 | { | |||
| 3499 | 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); | |||
| 3500 | 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" , 3500, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3501 | 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", 3501, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3502 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3503 | ccv_nnc_tensor_param_t output_params[2]; | |||
| 3504 | ccv_cnnp_model_dropout_t* const self = (ccv_cnnp_model_dropout_t*)super; | |||
| 3505 | 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); | |||
| 3506 | ccv_nnc_hint_tensor_auto(dropout, (ccv_nnc_tensor_param_t []){ | |||
| 3507 | params, | |||
| 3508 | }, 1, ccv_nnc_no_hint, output_params, 2); | |||
| 3509 | const ccv_nnc_tensor_symbol_t dropout_output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 3510 | const ccv_nnc_tensor_symbol_t mask = ccv_nnc_tensor_symbol_new(graph, output_params[1], "mask"); | |||
| 3511 | 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"); | |||
| 3512 | outputs[0] = dropout_output; | |||
| 3513 | } | |||
| 3514 | ||||
| 3515 | 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) | |||
| 3516 | { | |||
| 3517 | ccv_cnnp_model_dropout_t* const self = (ccv_cnnp_model_dropout_t*)super; | |||
| 3518 | if (self->dropout.graph) | |||
| 3519 | { | |||
| 3520 | if (is_test) | |||
| 3521 | // During test, the dropout is not applied. Data transfer is perfect because if these are the same tensor, it will skip. | |||
| 3522 | 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); | |||
| 3523 | else | |||
| 3524 | 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); | |||
| 3525 | } | |||
| 3526 | } | |||
| 3527 | ||||
| 3528 | static ccv_cnnp_model_t* _ccv_cnnp_dropout_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3529 | ||||
| 3530 | static const ccv_cnnp_model_vtab_t ccv_cnnp_dropout_isa = { | |||
| 3531 | .build = _ccv_cnnp_dropout_build, | |||
| 3532 | .set_is_test = _ccv_cnnp_dropout_set_is_test, | |||
| 3533 | .copy = _ccv_cnnp_dropout_copy, | |||
| 3534 | }; | |||
| 3535 | ||||
| 3536 | ccv_cnnp_model_t* ccv_cnnp_dropout(const float p, const int entirety, const char* const name) | |||
| 3537 | { | |||
| 3538 | ccv_cnnp_model_dropout_t* const model_dropout = (ccv_cnnp_model_dropout_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_dropout_t)); | |||
| 3539 | model_dropout->super.isa = &ccv_cnnp_dropout_isa; | |||
| 3540 | model_dropout->super.input_size = 1; | |||
| 3541 | model_dropout->super.outputs = &model_dropout->output; | |||
| 3542 | model_dropout->super.output_size = 1; | |||
| 3543 | model_dropout->p = p; | |||
| 3544 | model_dropout->entirety = entirety; | |||
| 3545 | ccv_cnnp_model_copy_name(&model_dropout->super, name); | |||
| 3546 | return (ccv_cnnp_model_t*)model_dropout; | |||
| 3547 | } | |||
| 3548 | ||||
| 3549 | static ccv_cnnp_model_t* _ccv_cnnp_dropout_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3550 | { | |||
| 3551 | const ccv_cnnp_model_dropout_t* const self = (const ccv_cnnp_model_dropout_t*)super; | |||
| 3552 | return ccv_cnnp_dropout(self->p, self->entirety, self->super.name); | |||
| 3553 | } | |||
| 3554 | ||||
| 3555 | // MARK - Masked Fill Layer | |||
| 3556 | ||||
| 3557 | typedef struct { | |||
| 3558 | ccv_cnnp_model_t super; | |||
| 3559 | ccv_nnc_tensor_symbol_t output; | |||
| 3560 | float eq; | |||
| 3561 | float fill; | |||
| 3562 | } ccv_cnnp_model_masked_fill_t; | |||
| 3563 | ||||
| 3564 | 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) | |||
| 3565 | { | |||
| 3566 | 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); | |||
| 3567 | 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" , 3567, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3568 | 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", 3568, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3569 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3570 | ccv_cnnp_model_masked_fill_t* const self = (ccv_cnnp_model_masked_fill_t*)super; | |||
| 3571 | const ccv_nnc_tensor_symbol_t masked_fill_output = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 3572 | 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"); | |||
| 3573 | outputs[0] = masked_fill_output; | |||
| 3574 | } | |||
| 3575 | ||||
| 3576 | static ccv_cnnp_model_t* _ccv_cnnp_masked_fill_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3577 | ||||
| 3578 | static const ccv_cnnp_model_vtab_t ccv_cnnp_masked_fill_isa = { | |||
| 3579 | .build = _ccv_cnnp_masked_fill_build, | |||
| 3580 | .copy = _ccv_cnnp_masked_fill_copy, | |||
| 3581 | }; | |||
| 3582 | ||||
| 3583 | ccv_cnnp_model_t* ccv_cnnp_masked_fill(const float eq, const float fill, const char* const name) | |||
| 3584 | { | |||
| 3585 | 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)); | |||
| 3586 | model_masked_fill->super.isa = &ccv_cnnp_masked_fill_isa; | |||
| 3587 | model_masked_fill->super.input_size = 2; | |||
| 3588 | model_masked_fill->super.outputs = &model_masked_fill->output; | |||
| 3589 | model_masked_fill->super.output_size = 1; | |||
| 3590 | model_masked_fill->eq = eq; | |||
| 3591 | model_masked_fill->fill = fill; | |||
| 3592 | ccv_cnnp_model_copy_name(&model_masked_fill->super, name); | |||
| 3593 | return (ccv_cnnp_model_t*)model_masked_fill; | |||
| 3594 | } | |||
| 3595 | ||||
| 3596 | static ccv_cnnp_model_t* _ccv_cnnp_masked_fill_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3597 | { | |||
| 3598 | const ccv_cnnp_model_masked_fill_t* const self = (const ccv_cnnp_model_masked_fill_t*)super; | |||
| 3599 | return ccv_cnnp_masked_fill(self->eq, self->fill, self->super.name); | |||
| 3600 | } | |||
| 3601 | ||||
| 3602 | // MARK - Index Select Layer | |||
| 3603 | ||||
| 3604 | typedef struct { | |||
| 3605 | ccv_cnnp_model_t super; | |||
| 3606 | ccv_nnc_tensor_symbol_t output; | |||
| 3607 | } ccv_cnnp_model_index_select_t; | |||
| 3608 | ||||
| 3609 | 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) | |||
| 3610 | { | |||
| 3611 | 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); | |||
| 3612 | 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" , 3612, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3613 | 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", 3613, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3614 | const ccv_nnc_tensor_param_t vocab_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3615 | const ccv_nnc_tensor_param_t index_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 3616 | ccv_nnc_tensor_param_t output_params; | |||
| 3617 | 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); | |||
| 3618 | ccv_nnc_hint_tensor_auto(index_select, (ccv_nnc_tensor_param_t []){ | |||
| 3619 | vocab_params, | |||
| 3620 | index_params, | |||
| 3621 | }, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 3622 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3623 | 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"); | |||
| 3624 | outputs[0] = output; | |||
| 3625 | } | |||
| 3626 | ||||
| 3627 | static ccv_cnnp_model_t* _ccv_cnnp_index_select_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3628 | ||||
| 3629 | static const ccv_cnnp_model_vtab_t ccv_cnnp_index_select_isa = { | |||
| 3630 | .build = _ccv_cnnp_index_select_build, | |||
| 3631 | .copy = _ccv_cnnp_index_select_copy, | |||
| 3632 | }; | |||
| 3633 | ||||
| 3634 | ccv_cnnp_model_t* ccv_cnnp_index_select(const char* const name) | |||
| 3635 | { | |||
| 3636 | 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)); | |||
| 3637 | model_index_select->super.isa = &ccv_cnnp_index_select_isa; | |||
| 3638 | model_index_select->super.input_size = 2; | |||
| 3639 | model_index_select->super.outputs = &model_index_select->output; | |||
| 3640 | model_index_select->super.output_size = 1; | |||
| 3641 | ccv_cnnp_model_copy_name(&model_index_select->super, name); | |||
| 3642 | return (ccv_cnnp_model_t*)model_index_select; | |||
| 3643 | } | |||
| 3644 | ||||
| 3645 | static ccv_cnnp_model_t* _ccv_cnnp_index_select_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3646 | { | |||
| 3647 | ccv_cnnp_model_index_select_t* const self = (ccv_cnnp_model_index_select_t*)super; | |||
| 3648 | return ccv_cnnp_index_select(self->super.name); | |||
| 3649 | } | |||
| 3650 | ||||
| 3651 | // MARK - Embedding Layer | |||
| 3652 | ||||
| 3653 | typedef struct { | |||
| 3654 | ccv_cnnp_model_t super; | |||
| 3655 | ccv_nnc_tensor_symbol_t output; | |||
| 3656 | ccv_nnc_tensor_symbol_t vocab; | |||
| 3657 | int datatype; | |||
| 3658 | int vocab_size; | |||
| 3659 | int embed_size; | |||
| 3660 | } ccv_cnnp_model_embedding_t; | |||
| 3661 | ||||
| 3662 | 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) | |||
| 3663 | { | |||
| 3664 | ccv_cnnp_model_embedding_t* const self = (ccv_cnnp_model_embedding_t*)super; | |||
| 3665 | 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); | |||
| 3666 | 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" , 3666, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3667 | 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", 3667, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3668 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3669 | ccv_nnc_tensor_param_t vocab_params = params; | |||
| 3670 | memset(vocab_params.dim, 0, sizeof(vocab_params.dim)); | |||
| 3671 | vocab_params.datatype = self->datatype; | |||
| 3672 | vocab_params.dim[0] = self->vocab_size; | |||
| 3673 | vocab_params.dim[1] = self->embed_size; | |||
| 3674 | if (!self->vocab.graph) | |||
| 3675 | self->vocab = ccv_nnc_tensor_symbol_new(graph, vocab_params, "vocab"); | |||
| 3676 | 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", 3676 , __extension__ __PRETTY_FUNCTION__); })); | |||
| 3677 | const ccv_nnc_tensor_symbol_t vocab = ccv_cnnp_model_get_symbol(super, self->vocab); | |||
| 3678 | ccv_nnc_tensor_param_t output_params; | |||
| 3679 | const ccv_nnc_cmd_t embedding = CMD_INDEX_SELECT_FORWARD()ccv_nnc_cmd(CCV_NNC_INDEX_SELECT_FORWARD, 0, ccv_nnc_cmd_auto , 0); | |||
| 3680 | ccv_nnc_hint_tensor_auto(embedding, (ccv_nnc_tensor_param_t []){ | |||
| 3681 | vocab_params, | |||
| 3682 | params, | |||
| 3683 | }, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 3684 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3685 | ccv_nnc_graph_exec_symbol_new(graph, embedding, TENSOR_SYMBOL_LIST(vocab, inputs[0])(const ccv_nnc_tensor_symbol_t []){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"); | |||
| 3686 | outputs[0] = output; | |||
| 3687 | } | |||
| 3688 | ||||
| 3689 | 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) | |||
| 3690 | { | |||
| 3691 | ccv_cnnp_model_embedding_t* const self = (ccv_cnnp_model_embedding_t*)super; | |||
| 3692 | const float std = sqrtf(2) / sqrtf(self->vocab_size + self->embed_size); | |||
| 3693 | const float bound = sqrtf(3) * std; | |||
| 3694 | 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); | |||
| 3695 | } | |||
| 3696 | ||||
| 3697 | 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) | |||
| 3698 | { | |||
| 3699 | ccv_cnnp_model_embedding_t* const self = (ccv_cnnp_model_embedding_t*)super; | |||
| 3700 | add_to_array(parameters, self->vocab, is_trainable); | |||
| 3701 | } | |||
| 3702 | ||||
| 3703 | static ccv_cnnp_model_t* _ccv_cnnp_embedding_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3704 | ||||
| 3705 | static const ccv_cnnp_model_vtab_t ccv_cnnp_embedding_isa = { | |||
| 3706 | .build = _ccv_cnnp_embedding_build, | |||
| 3707 | .init_states = _ccv_cnnp_embedding_init_states, | |||
| 3708 | .add_to_parameter = _ccv_cnnp_embedding_add_to_parameter, | |||
| 3709 | .copy = _ccv_cnnp_embedding_copy, | |||
| 3710 | }; | |||
| 3711 | ||||
| 3712 | 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) | |||
| 3713 | { | |||
| 3714 | ccv_cnnp_model_embedding_t* const model_embedding = (ccv_cnnp_model_embedding_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_embedding_t)); | |||
| 3715 | model_embedding->super.isa = &ccv_cnnp_embedding_isa; | |||
| 3716 | model_embedding->super.input_size = 1; | |||
| 3717 | model_embedding->super.outputs = &model_embedding->output; | |||
| 3718 | model_embedding->super.output_size = 1; | |||
| 3719 | model_embedding->super.is_trainable = is_trainable; | |||
| 3720 | ccv_cnnp_model_copy_name(&model_embedding->super, name); | |||
| 3721 | model_embedding->vocab.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 3722 | model_embedding->vocab.graph = 0; | |||
| 3723 | assert(datatype == CCV_32F || datatype == CCV_16F || datatype == CCV_16BF || datatype == CCV_32S)((void) sizeof ((datatype == CCV_32F || datatype == CCV_16F || datatype == CCV_16BF || datatype == CCV_32S) ? 1 : 0), __extension__ ({ if (datatype == CCV_32F || datatype == CCV_16F || datatype == CCV_16BF || datatype == CCV_32S) ; else __assert_fail ("datatype == CCV_32F || datatype == CCV_16F || datatype == CCV_16BF || datatype == CCV_32S" , "ccv_cnnp_model_addons.c", 3723, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3724 | model_embedding->datatype = datatype; | |||
| 3725 | 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", 3725, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3726 | model_embedding->vocab_size = vocab_size; | |||
| 3727 | 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", 3727, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3728 | model_embedding->embed_size = embed_size; | |||
| 3729 | return (ccv_cnnp_model_t*)model_embedding; | |||
| 3730 | } | |||
| 3731 | ||||
| 3732 | static ccv_cnnp_model_t* _ccv_cnnp_embedding_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3733 | { | |||
| 3734 | ccv_cnnp_model_embedding_t* const self = (ccv_cnnp_model_embedding_t*)super; | |||
| 3735 | return ccv_cnnp_embedding(self->datatype, self->vocab_size, self->embed_size, self->super.is_trainable, self->super.name); | |||
| 3736 | } | |||
| 3737 | ||||
| 3738 | // MARK - Pool Layers | |||
| 3739 | ||||
| 3740 | typedef struct { | |||
| 3741 | ccv_cnnp_model_t super; | |||
| 3742 | ccv_nnc_tensor_symbol_t output; | |||
| 3743 | int type; | |||
| 3744 | float width_scale; | |||
| 3745 | float height_scale; | |||
| 3746 | int align_corners; | |||
| 3747 | } ccv_cnnp_model_upsample_t; | |||
| 3748 | ||||
| 3749 | 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) | |||
| 3750 | { | |||
| 3751 | 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); | |||
| 3752 | 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" , 3752, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3753 | 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", 3753, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3754 | ccv_cnnp_model_upsample_t* const self = (ccv_cnnp_model_upsample_t*)super; | |||
| 3755 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3756 | 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); | |||
| 3757 | ccv_nnc_tensor_param_t output_params; | |||
| 3758 | ccv_nnc_hint_tensor_auto(cmd, ¶ms, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3759 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3760 | 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"); | |||
| 3761 | outputs[0] = output; | |||
| 3762 | } | |||
| 3763 | ||||
| 3764 | static ccv_cnnp_model_t* _ccv_cnnp_upsample_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 3765 | ||||
| 3766 | static const ccv_cnnp_model_vtab_t ccv_cnnp_upsample_isa = { | |||
| 3767 | .build = _ccv_cnnp_upsample_build, | |||
| 3768 | .copy = _ccv_cnnp_upsample_copy, | |||
| 3769 | }; | |||
| 3770 | ||||
| 3771 | 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) | |||
| 3772 | { | |||
| 3773 | ccv_cnnp_model_upsample_t* const model_upsample = (ccv_cnnp_model_upsample_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_upsample_t)); | |||
| 3774 | model_upsample->super.isa = &ccv_cnnp_upsample_isa; | |||
| 3775 | model_upsample->super.input_size = 1; | |||
| 3776 | model_upsample->super.outputs = &model_upsample->output; | |||
| 3777 | model_upsample->super.output_size = 1; | |||
| 3778 | ccv_cnnp_model_copy_name(&model_upsample->super, name); | |||
| 3779 | 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", 3779, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3780 | model_upsample->type = type; | |||
| 3781 | model_upsample->width_scale = width_scale; | |||
| 3782 | model_upsample->height_scale = height_scale; | |||
| 3783 | model_upsample->align_corners = align_corners; | |||
| 3784 | return (ccv_cnnp_model_t*)model_upsample; | |||
| 3785 | } | |||
| 3786 | ||||
| 3787 | static ccv_cnnp_model_t* _ccv_cnnp_upsample_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3788 | { | |||
| 3789 | const ccv_cnnp_model_upsample_t* const self = (const ccv_cnnp_model_upsample_t*)super; | |||
| 3790 | return ccv_cnnp_upsample(self->type, self->width_scale, self->height_scale, self->align_corners, self->super.name); | |||
| 3791 | } | |||
| 3792 | ||||
| 3793 | // MARK - Reduce Sum Layer | |||
| 3794 | ||||
| 3795 | typedef struct { | |||
| 3796 | ccv_cnnp_model_t super; | |||
| 3797 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 3798 | int count; | |||
| 3799 | ccv_nnc_tensor_symbol_t output; | |||
| 3800 | } ccv_cnnp_model_reduce_sum_t; | |||
| 3801 | ||||
| 3802 | 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) | |||
| 3803 | { | |||
| 3804 | 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); | |||
| 3805 | const ccv_cnnp_model_reduce_sum_t* const self = (const ccv_cnnp_model_reduce_sum_t*)super; | |||
| 3806 | 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" , 3806, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3807 | 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", 3807, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3808 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3809 | ccv_nnc_tensor_param_t output_params; | |||
| 3810 | 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); | |||
| 3811 | int i; | |||
| 3812 | for (i = 0; i < self->count; i++) | |||
| 3813 | reduce_sum.info.reduce.axis[i] = self->axis[i]; | |||
| 3814 | reduce_sum.info.reduce.count = self->count; | |||
| 3815 | ccv_nnc_hint_tensor_auto(reduce_sum, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3816 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3817 | ccv_nnc_graph_exec_symbol_new(graph, reduce_sum, inputs, input_size, outputs, output_size, "reduce_sum"); | |||
| 3818 | } | |||
| 3819 | ||||
| 3820 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_sum_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3821 | ||||
| 3822 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_sum_isa = { | |||
| 3823 | .build = _ccv_cnnp_reduce_sum_build, | |||
| 3824 | .copy = _ccv_cnnp_reduce_sum_copy, | |||
| 3825 | }; | |||
| 3826 | ||||
| 3827 | ccv_cnnp_model_t* ccv_cnnp_reduce_sum(const int* const axis, const int axis_count, const char* const name) | |||
| 3828 | { | |||
| 3829 | 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)); | |||
| 3830 | model_reduce_sum->super.isa = &ccv_cnnp_reduce_sum_isa; | |||
| 3831 | model_reduce_sum->super.input_size = 1; | |||
| 3832 | model_reduce_sum->super.outputs = &model_reduce_sum->output; | |||
| 3833 | model_reduce_sum->super.output_size = 1; | |||
| 3834 | ccv_cnnp_model_copy_name(&model_reduce_sum->super, name); | |||
| 3835 | 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", 3835, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3836 | int i; | |||
| 3837 | for (i = 0; i < axis_count; i++) | |||
| 3838 | model_reduce_sum->axis[i] = axis[i]; | |||
| 3839 | model_reduce_sum->count = axis_count; | |||
| 3840 | return (ccv_cnnp_model_t*)model_reduce_sum; | |||
| 3841 | } | |||
| 3842 | ||||
| 3843 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_sum_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3844 | { | |||
| 3845 | const ccv_cnnp_model_reduce_sum_t* const self = (const ccv_cnnp_model_reduce_sum_t*)super; | |||
| 3846 | return ccv_cnnp_reduce_sum(self->axis, self->count, self->super.name); | |||
| 3847 | } | |||
| 3848 | ||||
| 3849 | // MARK - Reduce Mean Layer | |||
| 3850 | ||||
| 3851 | typedef struct { | |||
| 3852 | ccv_cnnp_model_t super; | |||
| 3853 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 3854 | int count; | |||
| 3855 | ccv_nnc_tensor_symbol_t output; | |||
| 3856 | } ccv_cnnp_model_reduce_mean_t; | |||
| 3857 | ||||
| 3858 | 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) | |||
| 3859 | { | |||
| 3860 | 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); | |||
| 3861 | const ccv_cnnp_model_reduce_mean_t* const self = (const ccv_cnnp_model_reduce_mean_t*)super; | |||
| 3862 | 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" , 3862, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3863 | 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", 3863, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3864 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3865 | ccv_nnc_tensor_param_t output_params; | |||
| 3866 | 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); | |||
| 3867 | int i; | |||
| 3868 | for (i = 0; i < self->count; i++) | |||
| 3869 | reduce_mean.info.reduce.axis[i] = self->axis[i]; | |||
| 3870 | reduce_mean.info.reduce.count = self->count; | |||
| 3871 | ccv_nnc_hint_tensor_auto(reduce_mean, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3872 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3873 | ccv_nnc_graph_exec_symbol_new(graph, reduce_mean, inputs, input_size, outputs, output_size, "reduce_mean"); | |||
| 3874 | } | |||
| 3875 | ||||
| 3876 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_mean_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3877 | ||||
| 3878 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_mean_isa = { | |||
| 3879 | .build = _ccv_cnnp_reduce_mean_build, | |||
| 3880 | .copy = _ccv_cnnp_reduce_mean_copy, | |||
| 3881 | }; | |||
| 3882 | ||||
| 3883 | ccv_cnnp_model_t* ccv_cnnp_reduce_mean(const int* const axis, const int axis_count, const char* const name) | |||
| 3884 | { | |||
| 3885 | 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)); | |||
| 3886 | model_reduce_mean->super.isa = &ccv_cnnp_reduce_mean_isa; | |||
| 3887 | model_reduce_mean->super.input_size = 1; | |||
| 3888 | model_reduce_mean->super.outputs = &model_reduce_mean->output; | |||
| 3889 | model_reduce_mean->super.output_size = 1; | |||
| 3890 | ccv_cnnp_model_copy_name(&model_reduce_mean->super, name); | |||
| 3891 | 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", 3891, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3892 | int i; | |||
| 3893 | for (i = 0; i < axis_count; i++) | |||
| 3894 | model_reduce_mean->axis[i] = axis[i]; | |||
| 3895 | model_reduce_mean->count = axis_count; | |||
| 3896 | return (ccv_cnnp_model_t*)model_reduce_mean; | |||
| 3897 | } | |||
| 3898 | ||||
| 3899 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_mean_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3900 | { | |||
| 3901 | const ccv_cnnp_model_reduce_mean_t* const self = (const ccv_cnnp_model_reduce_mean_t*)super; | |||
| 3902 | return ccv_cnnp_reduce_mean(self->axis, self->count, self->super.name); | |||
| 3903 | } | |||
| 3904 | ||||
| 3905 | // MARK - Reduce Max Layer | |||
| 3906 | ||||
| 3907 | typedef struct { | |||
| 3908 | ccv_cnnp_model_t super; | |||
| 3909 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 3910 | int count; | |||
| 3911 | ccv_nnc_tensor_symbol_t output; | |||
| 3912 | } ccv_cnnp_model_reduce_max_t; | |||
| 3913 | ||||
| 3914 | 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) | |||
| 3915 | { | |||
| 3916 | 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); | |||
| 3917 | const ccv_cnnp_model_reduce_max_t* const self = (const ccv_cnnp_model_reduce_max_t*)super; | |||
| 3918 | 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" , 3918, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3919 | 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", 3919, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3920 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3921 | ccv_nnc_tensor_param_t output_params; | |||
| 3922 | 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); | |||
| 3923 | int i; | |||
| 3924 | for (i = 0; i < self->count; i++) | |||
| 3925 | reduce_max.info.reduce.axis[i] = self->axis[i]; | |||
| 3926 | reduce_max.info.reduce.count = self->count; | |||
| 3927 | ccv_nnc_hint_tensor_auto(reduce_max, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3928 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3929 | ccv_nnc_graph_exec_symbol_new(graph, reduce_max, inputs, input_size, outputs, output_size, "reduce_max"); | |||
| 3930 | } | |||
| 3931 | ||||
| 3932 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_max_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3933 | ||||
| 3934 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_max_isa = { | |||
| 3935 | .build = _ccv_cnnp_reduce_max_build, | |||
| 3936 | .copy = _ccv_cnnp_reduce_max_copy, | |||
| 3937 | }; | |||
| 3938 | ||||
| 3939 | ccv_cnnp_model_t* ccv_cnnp_reduce_max(const int* const axis, const int axis_count, const char* const name) | |||
| 3940 | { | |||
| 3941 | 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)); | |||
| 3942 | model_reduce_max->super.isa = &ccv_cnnp_reduce_max_isa; | |||
| 3943 | model_reduce_max->super.input_size = 1; | |||
| 3944 | model_reduce_max->super.outputs = &model_reduce_max->output; | |||
| 3945 | model_reduce_max->super.output_size = 1; | |||
| 3946 | ccv_cnnp_model_copy_name(&model_reduce_max->super, name); | |||
| 3947 | 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", 3947, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3948 | int i; | |||
| 3949 | for (i = 0; i < axis_count; i++) | |||
| 3950 | model_reduce_max->axis[i] = axis[i]; | |||
| 3951 | model_reduce_max->count = axis_count; | |||
| 3952 | return (ccv_cnnp_model_t*)model_reduce_max; | |||
| 3953 | } | |||
| 3954 | ||||
| 3955 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_max_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 3956 | { | |||
| 3957 | const ccv_cnnp_model_reduce_max_t* const self = (const ccv_cnnp_model_reduce_max_t*)super; | |||
| 3958 | return ccv_cnnp_reduce_max(self->axis, self->count, self->super.name); | |||
| 3959 | } | |||
| 3960 | ||||
| 3961 | // MARK - Reduce Min Layer | |||
| 3962 | ||||
| 3963 | typedef struct { | |||
| 3964 | ccv_cnnp_model_t super; | |||
| 3965 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 3966 | int count; | |||
| 3967 | ccv_nnc_tensor_symbol_t output; | |||
| 3968 | } ccv_cnnp_model_reduce_min_t; | |||
| 3969 | ||||
| 3970 | 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) | |||
| 3971 | { | |||
| 3972 | 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); | |||
| 3973 | const ccv_cnnp_model_reduce_min_t* const self = (const ccv_cnnp_model_reduce_min_t*)super; | |||
| 3974 | 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" , 3974, __extension__ __PRETTY_FUNCTION__); })); | |||
| 3975 | 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", 3975, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 3976 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 3977 | ccv_nnc_tensor_param_t output_params; | |||
| 3978 | 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); | |||
| 3979 | int i; | |||
| 3980 | for (i = 0; i < self->count; i++) | |||
| 3981 | reduce_min.info.reduce.axis[i] = self->axis[i]; | |||
| 3982 | reduce_min.info.reduce.count = self->count; | |||
| 3983 | ccv_nnc_hint_tensor_auto(reduce_min, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 3984 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 3985 | ccv_nnc_graph_exec_symbol_new(graph, reduce_min, inputs, input_size, outputs, output_size, "reduce_min"); | |||
| 3986 | } | |||
| 3987 | ||||
| 3988 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_min_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 3989 | ||||
| 3990 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_min_isa = { | |||
| 3991 | .build = _ccv_cnnp_reduce_min_build, | |||
| 3992 | .copy = _ccv_cnnp_reduce_min_copy, | |||
| 3993 | }; | |||
| 3994 | ||||
| 3995 | ccv_cnnp_model_t* ccv_cnnp_reduce_min(const int* const axis, const int axis_count, const char* const name) | |||
| 3996 | { | |||
| 3997 | 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)); | |||
| 3998 | model_reduce_min->super.isa = &ccv_cnnp_reduce_min_isa; | |||
| 3999 | model_reduce_min->super.input_size = 1; | |||
| 4000 | model_reduce_min->super.outputs = &model_reduce_min->output; | |||
| 4001 | model_reduce_min->super.output_size = 1; | |||
| 4002 | ccv_cnnp_model_copy_name(&model_reduce_min->super, name); | |||
| 4003 | 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", 4003, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4004 | int i; | |||
| 4005 | for (i = 0; i < axis_count; i++) | |||
| 4006 | model_reduce_min->axis[i] = axis[i]; | |||
| 4007 | model_reduce_min->count = axis_count; | |||
| 4008 | return (ccv_cnnp_model_t*)model_reduce_min; | |||
| 4009 | } | |||
| 4010 | ||||
| 4011 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_min_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4012 | { | |||
| 4013 | const ccv_cnnp_model_reduce_min_t* const self = (const ccv_cnnp_model_reduce_min_t*)super; | |||
| 4014 | return ccv_cnnp_reduce_min(self->axis, self->count, self->super.name); | |||
| 4015 | } | |||
| 4016 | ||||
| 4017 | // MARK - Reduce Norm2 Layer | |||
| 4018 | ||||
| 4019 | typedef struct { | |||
| 4020 | ccv_cnnp_model_t super; | |||
| 4021 | int axis[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 4022 | int count; | |||
| 4023 | ccv_nnc_tensor_symbol_t output; | |||
| 4024 | } ccv_cnnp_model_reduce_norm2_t; | |||
| 4025 | ||||
| 4026 | 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) | |||
| 4027 | { | |||
| 4028 | const ccv_cnnp_model_reduce_norm2_t* const self = (const ccv_cnnp_model_reduce_norm2_t*)super; | |||
| 4029 | 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); | |||
| 4030 | 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" , 4030, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4031 | 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", 4031, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4032 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4033 | ccv_nnc_tensor_param_t output_params; | |||
| 4034 | 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); | |||
| 4035 | int i; | |||
| 4036 | for (i = 0; i < self->count; i++) | |||
| 4037 | reduce_norm2.info.reduce.axis[i] = self->axis[i]; | |||
| 4038 | reduce_norm2.info.reduce.count = self->count; | |||
| 4039 | ccv_nnc_hint_tensor_auto(reduce_norm2, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 4040 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 4041 | ccv_nnc_graph_exec_symbol_new(graph, reduce_norm2, inputs, input_size, outputs, output_size, "reduce_norm2"); | |||
| 4042 | } | |||
| 4043 | ||||
| 4044 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_norm2_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 4045 | ||||
| 4046 | static const ccv_cnnp_model_vtab_t ccv_cnnp_reduce_norm2_isa = { | |||
| 4047 | .build = _ccv_cnnp_reduce_norm2_build, | |||
| 4048 | .copy = _ccv_cnnp_reduce_norm2_copy, | |||
| 4049 | }; | |||
| 4050 | ||||
| 4051 | ccv_cnnp_model_t* ccv_cnnp_reduce_norm2(const int* const axis, const int axis_count, const char* const name) | |||
| 4052 | { | |||
| 4053 | 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)); | |||
| 4054 | model_reduce_norm2->super.isa = &ccv_cnnp_reduce_norm2_isa; | |||
| 4055 | model_reduce_norm2->super.input_size = 1; | |||
| 4056 | model_reduce_norm2->super.outputs = &model_reduce_norm2->output; | |||
| 4057 | model_reduce_norm2->super.output_size = 1; | |||
| 4058 | ccv_cnnp_model_copy_name(&model_reduce_norm2->super, name); | |||
| 4059 | 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", 4059, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4060 | int i; | |||
| 4061 | for (i = 0; i < axis_count; i++) | |||
| 4062 | model_reduce_norm2->axis[i] = axis[i]; | |||
| 4063 | model_reduce_norm2->count = axis_count; | |||
| 4064 | return (ccv_cnnp_model_t*)model_reduce_norm2; | |||
| 4065 | } | |||
| 4066 | ||||
| 4067 | static ccv_cnnp_model_t* _ccv_cnnp_reduce_norm2_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4068 | { | |||
| 4069 | const ccv_cnnp_model_reduce_norm2_t* const self = (const ccv_cnnp_model_reduce_norm2_t*)super; | |||
| 4070 | return ccv_cnnp_reduce_norm2(self->axis, self->count, self->super.name); | |||
| 4071 | } | |||
| 4072 | ||||
| 4073 | // MARK - Argmax Layer | |||
| 4074 | ||||
| 4075 | typedef struct { | |||
| 4076 | ccv_cnnp_model_t super; | |||
| 4077 | int axis; | |||
| 4078 | ccv_nnc_tensor_symbol_t output; | |||
| 4079 | } ccv_cnnp_model_argmax_t; | |||
| 4080 | ||||
| 4081 | 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) | |||
| 4082 | { | |||
| 4083 | const ccv_cnnp_model_argmax_t* const self = (const ccv_cnnp_model_argmax_t*)super; | |||
| 4084 | 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); | |||
| 4085 | 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" , 4085, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4086 | 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", 4086, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4087 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4088 | ccv_nnc_tensor_param_t output_params; | |||
| 4089 | 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); | |||
| 4090 | argmax.info.reduce.axis[0] = self->axis; | |||
| 4091 | argmax.info.reduce.count = 1; | |||
| 4092 | ccv_nnc_hint_tensor_auto(argmax, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 4093 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 4094 | ccv_nnc_graph_exec_symbol_new(graph, argmax, inputs, input_size, outputs, output_size, "argmax"); | |||
| 4095 | } | |||
| 4096 | ||||
| 4097 | static ccv_cnnp_model_t* _ccv_cnnp_argmax_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 4098 | ||||
| 4099 | static const ccv_cnnp_model_vtab_t ccv_cnnp_argmax_isa = { | |||
| 4100 | .build = _ccv_cnnp_argmax_build, | |||
| 4101 | .copy = _ccv_cnnp_argmax_copy, | |||
| 4102 | }; | |||
| 4103 | ||||
| 4104 | ccv_cnnp_model_t* ccv_cnnp_argmax(const int axis, const char* const name) | |||
| 4105 | { | |||
| 4106 | ccv_cnnp_model_argmax_t* const model_argmax = (ccv_cnnp_model_argmax_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_argmax_t)); | |||
| 4107 | model_argmax->super.isa = &ccv_cnnp_argmax_isa; | |||
| 4108 | model_argmax->super.input_size = 1; | |||
| 4109 | model_argmax->super.outputs = &model_argmax->output; | |||
| 4110 | model_argmax->super.output_size = 1; | |||
| 4111 | ccv_cnnp_model_copy_name(&model_argmax->super, name); | |||
| 4112 | model_argmax->axis = axis; | |||
| 4113 | return (ccv_cnnp_model_t*)model_argmax; | |||
| 4114 | } | |||
| 4115 | ||||
| 4116 | static ccv_cnnp_model_t* _ccv_cnnp_argmax_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4117 | { | |||
| 4118 | const ccv_cnnp_model_argmax_t* const self = (const ccv_cnnp_model_argmax_t*)super; | |||
| 4119 | return ccv_cnnp_argmax(self->axis, self->super.name); | |||
| 4120 | } | |||
| 4121 | ||||
| 4122 | // MARK - Argmin Layer | |||
| 4123 | ||||
| 4124 | typedef struct { | |||
| 4125 | ccv_cnnp_model_t super; | |||
| 4126 | int axis; | |||
| 4127 | ccv_nnc_tensor_symbol_t output; | |||
| 4128 | } ccv_cnnp_model_argmin_t; | |||
| 4129 | ||||
| 4130 | 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) | |||
| 4131 | { | |||
| 4132 | const ccv_cnnp_model_argmin_t* const self = (const ccv_cnnp_model_argmin_t*)super; | |||
| 4133 | 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); | |||
| 4134 | 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" , 4134, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4135 | 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", 4135, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4136 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4137 | ccv_nnc_tensor_param_t output_params; | |||
| 4138 | 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); | |||
| 4139 | argmin.info.reduce.axis[0] = self->axis; | |||
| 4140 | argmin.info.reduce.count = 1; | |||
| 4141 | ccv_nnc_hint_tensor_auto(argmin, &input_params, 1, ccv_nnc_no_hint, &output_params, 1); | |||
| 4142 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 4143 | ccv_nnc_graph_exec_symbol_new(graph, argmin, inputs, input_size, outputs, output_size, "argmin"); | |||
| 4144 | } | |||
| 4145 | ||||
| 4146 | static ccv_cnnp_model_t* _ccv_cnnp_argmin_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 4147 | ||||
| 4148 | static const ccv_cnnp_model_vtab_t ccv_cnnp_argmin_isa = { | |||
| 4149 | .build = _ccv_cnnp_argmin_build, | |||
| 4150 | .copy = _ccv_cnnp_argmin_copy, | |||
| 4151 | }; | |||
| 4152 | ||||
| 4153 | ccv_cnnp_model_t* ccv_cnnp_argmin(const int axis, const char* const name) | |||
| 4154 | { | |||
| 4155 | ccv_cnnp_model_argmin_t* const model_argmin = (ccv_cnnp_model_argmin_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_argmin_t)); | |||
| 4156 | model_argmin->super.isa = &ccv_cnnp_argmin_isa; | |||
| 4157 | model_argmin->super.input_size = 1; | |||
| 4158 | model_argmin->super.outputs = &model_argmin->output; | |||
| 4159 | model_argmin->super.output_size = 1; | |||
| 4160 | ccv_cnnp_model_copy_name(&model_argmin->super, name); | |||
| 4161 | model_argmin->axis = axis; | |||
| 4162 | return (ccv_cnnp_model_t*)model_argmin; | |||
| 4163 | } | |||
| 4164 | ||||
| 4165 | static ccv_cnnp_model_t* _ccv_cnnp_argmin_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4166 | { | |||
| 4167 | const ccv_cnnp_model_argmin_t* const self = (const ccv_cnnp_model_argmin_t*)super; | |||
| 4168 | return ccv_cnnp_argmin(self->axis, self->super.name); | |||
| 4169 | } | |||
| 4170 | ||||
| 4171 | // MARK - Min Layer | |||
| 4172 | ||||
| 4173 | typedef struct { | |||
| 4174 | ccv_cnnp_model_t super; | |||
| 4175 | ccv_nnc_tensor_symbol_t output; | |||
| 4176 | } ccv_cnnp_model_min_t; | |||
| 4177 | ||||
| 4178 | 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) | |||
| 4179 | { | |||
| 4180 | 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); | |||
| 4181 | 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" , 4181, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4182 | 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", 4182, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4183 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 4184 | int i; | |||
| 4185 | for (i = 0; i < 2; i++) | |||
| 4186 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 4187 | ccv_nnc_tensor_param_t output_params; | |||
| 4188 | 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); | |||
| 4189 | ccv_nnc_hint_tensor_auto(min, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 4190 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 4191 | ccv_nnc_graph_exec_symbol_new(graph, min, inputs, input_size, outputs, output_size, "min"); | |||
| 4192 | } | |||
| 4193 | ||||
| 4194 | static ccv_cnnp_model_t* _ccv_cnnp_min_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 4195 | ||||
| 4196 | static const ccv_cnnp_model_vtab_t ccv_cnnp_min_isa = { | |||
| 4197 | .build = _ccv_cnnp_min_build, | |||
| 4198 | .copy = _ccv_cnnp_min_copy, | |||
| 4199 | }; | |||
| 4200 | ||||
| 4201 | ccv_cnnp_model_t* ccv_cnnp_min(const char* const name) | |||
| 4202 | { | |||
| 4203 | ccv_cnnp_model_min_t* const model_min = (ccv_cnnp_model_min_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_min_t)); | |||
| 4204 | model_min->super.isa = &ccv_cnnp_min_isa; | |||
| 4205 | model_min->super.input_size = 2; | |||
| 4206 | model_min->super.outputs = &model_min->output; | |||
| 4207 | model_min->super.output_size = 1; | |||
| 4208 | ccv_cnnp_model_copy_name(&model_min->super, name); | |||
| 4209 | return (ccv_cnnp_model_t*)model_min; | |||
| 4210 | } | |||
| 4211 | ||||
| 4212 | static ccv_cnnp_model_t* _ccv_cnnp_min_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4213 | { | |||
| 4214 | const ccv_cnnp_model_min_t* const self = (const ccv_cnnp_model_min_t*)super; | |||
| 4215 | return ccv_cnnp_min(self->super.name); | |||
| 4216 | } | |||
| 4217 | ||||
| 4218 | // MARK - Max Layer | |||
| 4219 | ||||
| 4220 | typedef struct { | |||
| 4221 | ccv_cnnp_model_t super; | |||
| 4222 | ccv_nnc_tensor_symbol_t output; | |||
| 4223 | } ccv_cnnp_model_max_t; | |||
| 4224 | ||||
| 4225 | 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) | |||
| 4226 | { | |||
| 4227 | 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); | |||
| 4228 | 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" , 4228, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4229 | 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", 4229, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4230 | ccv_nnc_tensor_param_t input_params[2]; | |||
| 4231 | int i; | |||
| 4232 | for (i = 0; i < 2; i++) | |||
| 4233 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 4234 | ccv_nnc_tensor_param_t output_params; | |||
| 4235 | 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); | |||
| 4236 | ccv_nnc_hint_tensor_auto(max, input_params, 2, ccv_nnc_no_hint, &output_params, 1); | |||
| 4237 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 4238 | ccv_nnc_graph_exec_symbol_new(graph, max, inputs, input_size, outputs, output_size, "max"); | |||
| 4239 | } | |||
| 4240 | ||||
| 4241 | static ccv_cnnp_model_t* _ccv_cnnp_max_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 4242 | ||||
| 4243 | static const ccv_cnnp_model_vtab_t ccv_cnnp_max_isa = { | |||
| 4244 | .build = _ccv_cnnp_max_build, | |||
| 4245 | .copy = _ccv_cnnp_max_copy, | |||
| 4246 | }; | |||
| 4247 | ||||
| 4248 | ccv_cnnp_model_t* ccv_cnnp_max(const char* const name) | |||
| 4249 | { | |||
| 4250 | ccv_cnnp_model_max_t* const model_max = (ccv_cnnp_model_max_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_max_t)); | |||
| 4251 | model_max->super.isa = &ccv_cnnp_max_isa; | |||
| 4252 | model_max->super.input_size = 2; | |||
| 4253 | model_max->super.outputs = &model_max->output; | |||
| 4254 | model_max->super.output_size = 1; | |||
| 4255 | ccv_cnnp_model_copy_name(&model_max->super, name); | |||
| 4256 | return (ccv_cnnp_model_t*)model_max; | |||
| 4257 | } | |||
| 4258 | ||||
| 4259 | static ccv_cnnp_model_t* _ccv_cnnp_max_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4260 | { | |||
| 4261 | const ccv_cnnp_model_max_t* const self = (const ccv_cnnp_model_max_t*)super; | |||
| 4262 | return ccv_cnnp_max(self->super.name); | |||
| 4263 | } | |||
| 4264 | ||||
| 4265 | // MARK - LSTM Layer | |||
| 4266 | ||||
| 4267 | typedef struct { | |||
| 4268 | ccv_cnnp_model_t super; | |||
| 4269 | int masked; | |||
| 4270 | ccv_nnc_tensor_symbol_t output; | |||
| 4271 | ccv_nnc_tensor_symbol_t weights; | |||
| 4272 | ccv_nnc_tensor_symbol_t reserves; | |||
| 4273 | ccv_nnc_cmd_param_t params; | |||
| 4274 | ccv_nnc_graph_exec_symbol_t lstm; | |||
| 4275 | } ccv_cnnp_model_lstm_t; | |||
| 4276 | ||||
| 4277 | static int _ccv_cnnp_lstm_weight_dim(int bidirectional, int num_layers, int input_size, int hidden_size, int proj_size, int bias) | |||
| 4278 | { | |||
| 4279 | const int D = !!bidirectional + 1; | |||
| 4280 | if (hidden_size == proj_size) | |||
| 4281 | return (num_layers * (bias ? 8 : 0) + (num_layers - 1) * (hidden_size * 4 * D + hidden_size * 4) + input_size * 4 + hidden_size * 4) * D; | |||
| 4282 | else | |||
| 4283 | 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; | |||
| 4284 | } | |||
| 4285 | ||||
| 4286 | 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) | |||
| 4287 | { | |||
| 4288 | ccv_cnnp_model_lstm_t* const self = (ccv_cnnp_model_lstm_t*)super; | |||
| 4289 | 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); | |||
| 4290 | 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", 4290, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4291 | 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", 4291, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4292 | const int proj_size = self->params.rnn.proj_size == 0 ? self->params.rnn.hidden_size : self->params.rnn.proj_size; | |||
| 4293 | ccv_nnc_tensor_param_t input_params[5]; | |||
| 4294 | input_params[0]= ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4295 | if (input_size == 2) | |||
| 4296 | input_params[1] = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 4297 | input_params[4] = input_params[0]; | |||
| 4298 | memset(input_params[4].dim, 0, sizeof(input_params[4].dim)); | |||
| 4299 | const int x_nd = ccv_nnc_tensor_nd(input_params[0].dim); | |||
| 4300 | const int feature_count = input_params[0].dim[x_nd - 1]; | |||
| 4301 | 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); | |||
| 4302 | input_params[4].dim[1] = self->params.rnn.hidden_size; | |||
| 4303 | const ccv_nnc_cmd_t lstm = ccv_nnc_cmd(CCV_NNC_LSTM_FORWARD, 0, self->params, 0); | |||
| 4304 | ccv_nnc_tensor_param_t output_params[4]; | |||
| 4305 | ccv_nnc_hint_tensor_auto(lstm, input_params, 5, ccv_nnc_no_hint, output_params, 4); | |||
| 4306 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 4307 | if (!self->weights.graph) | |||
| 4308 | self->weights = ccv_nnc_tensor_symbol_new(graph, input_params[4], "weights"); | |||
| 4309 | 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" , 4309, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4310 | const ccv_nnc_tensor_symbol_t weights = ccv_cnnp_model_get_symbol(super, self->weights); | |||
| 4311 | if (!self->reserves.graph) | |||
| 4312 | self->reserves = ccv_nnc_tensor_symbol_new(graph, output_params[3], "reserves"); | |||
| 4313 | assert(self->reserves.graph == graph)((void) sizeof ((self->reserves.graph == graph) ? 1 : 0), __extension__ ({ if (self->reserves.graph == graph) ; else __assert_fail ("self->reserves.graph == graph", "ccv_cnnp_model_addons.c" , 4313, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4314 | const ccv_nnc_tensor_symbol_t reserves = ccv_cnnp_model_get_symbol(super, self->reserves); | |||
| 4315 | 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 }; | |||
| 4316 | self->lstm = ccv_nnc_graph_exec_symbol_new(graph, lstm, TENSOR_SYMBOL_LIST(inputs[0], mask, NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, 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}, 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, 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}, 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"); | |||
| 4317 | } | |||
| 4318 | ||||
| 4319 | 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) | |||
| 4320 | { | |||
| 4321 | ccv_cnnp_model_lstm_t* const self = (ccv_cnnp_model_lstm_t*)super; | |||
| 4322 | if (self->weights.graph) | |||
| 4323 | { | |||
| 4324 | const float stdv = 1.0 / sqrt(self->params.rnn.hidden_size); | |||
| 4325 | 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); | |||
| 4326 | } | |||
| 4327 | } | |||
| 4328 | ||||
| 4329 | 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) | |||
| 4330 | { | |||
| 4331 | ccv_cnnp_model_lstm_t* const self = (ccv_cnnp_model_lstm_t*)super; | |||
| 4332 | if (self->weights.graph) | |||
| 4333 | add_to_array(parameters, self->weights, is_trainable); | |||
| 4334 | } | |||
| 4335 | ||||
| 4336 | 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) | |||
| 4337 | { | |||
| 4338 | ccv_cnnp_model_lstm_t* const self = (ccv_cnnp_model_lstm_t*)super; | |||
| 4339 | if (self->lstm.graph) | |||
| 4340 | { | |||
| 4341 | self->params.rnn.is_test = is_test; | |||
| 4342 | updater(context, self->lstm, ccv_nnc_cmd(CCV_NNC_LSTM_FORWARD, 0, self->params, 0), ccv_nnc_no_hint); | |||
| 4343 | } | |||
| 4344 | } | |||
| 4345 | ||||
| 4346 | static ccv_cnnp_model_t* _ccv_cnnp_lstm_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 4347 | ||||
| 4348 | static const ccv_cnnp_model_vtab_t ccv_cnnp_lstm_isa = { | |||
| 4349 | .build = _ccv_cnnp_lstm_build, | |||
| 4350 | .init_states = _ccv_cnnp_lstm_init_states, | |||
| 4351 | .add_to_parameter = _ccv_cnnp_lstm_add_to_parameter, | |||
| 4352 | .copy = _ccv_cnnp_lstm_copy, | |||
| 4353 | .set_is_test = _ccv_cnnp_lstm_set_is_test, | |||
| 4354 | }; | |||
| 4355 | ||||
| 4356 | 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) | |||
| 4357 | { | |||
| 4358 | ccv_cnnp_model_lstm_t* const model_lstm = (ccv_cnnp_model_lstm_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_lstm_t)); | |||
| 4359 | model_lstm->super.isa = &ccv_cnnp_lstm_isa; | |||
| 4360 | model_lstm->super.input_size = masked ? 2 : 1; | |||
| 4361 | model_lstm->super.outputs = &model_lstm->output; | |||
| 4362 | model_lstm->super.output_size = 1; | |||
| 4363 | model_lstm->super.is_trainable = is_trainable; | |||
| 4364 | ccv_cnnp_model_copy_name(&model_lstm->super, name); | |||
| 4365 | model_lstm->masked = masked; | |||
| 4366 | model_lstm->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 4367 | model_lstm->weights.graph = 0; | |||
| 4368 | model_lstm->params.rnn.hidden_size = hidden_size; | |||
| 4369 | model_lstm->params.rnn.proj_size = proj_size; | |||
| 4370 | model_lstm->params.rnn.num_layers = num_layers; | |||
| 4371 | model_lstm->params.rnn.bias = bias; | |||
| 4372 | model_lstm->params.rnn.batch_first = batch_first; | |||
| 4373 | model_lstm->params.rnn.bidirectional = bidirectional; | |||
| 4374 | model_lstm->params.rnn.dropout = dropout; | |||
| 4375 | return (ccv_cnnp_model_t*)model_lstm; | |||
| 4376 | } | |||
| 4377 | ||||
| 4378 | static ccv_cnnp_model_t* _ccv_cnnp_lstm_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4379 | { | |||
| 4380 | const ccv_cnnp_model_lstm_t* const self = (const ccv_cnnp_model_lstm_t*)super; | |||
| 4381 | 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); | |||
| 4382 | } | |||
| 4383 | ||||
| 4384 | /// MARK - Datatype conversion layer. | |||
| 4385 | ||||
| 4386 | typedef struct { | |||
| 4387 | ccv_cnnp_model_t super; | |||
| 4388 | ccv_nnc_tensor_symbol_t output; | |||
| 4389 | int datatype; | |||
| 4390 | int ref_to_last; | |||
| 4391 | } ccv_cnnp_model_datatype_conversion_t; | |||
| 4392 | ||||
| 4393 | 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) | |||
| 4394 | { | |||
| 4395 | ccv_cnnp_model_datatype_conversion_t* const self = (ccv_cnnp_model_datatype_conversion_t*)super; | |||
| 4396 | 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); | |||
| 4397 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4398 | if (self->ref_to_last) | |||
| 4399 | { | |||
| 4400 | 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", 4400, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4401 | const ccv_nnc_tensor_param_t last_params = ccv_nnc_tensor_symbol_params(graph, inputs[input_size - 1]); | |||
| 4402 | params.datatype = last_params.datatype; | |||
| 4403 | } else | |||
| 4404 | params.datatype = self->datatype; | |||
| 4405 | 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", 4405, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4406 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 4407 | 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 /* intentional */, outputs, output_size, 0); | |||
| 4408 | } | |||
| 4409 | ||||
| 4410 | static ccv_cnnp_model_t* _ccv_cnnp_datatype_conversion_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 4411 | ||||
| 4412 | static const ccv_cnnp_model_vtab_t ccv_cnnp_datatype_conversion_isa = { | |||
| 4413 | .build = _ccv_cnnp_datatype_conversion_build, | |||
| 4414 | .copy = _ccv_cnnp_datatype_conversion_copy, | |||
| 4415 | }; | |||
| 4416 | ||||
| 4417 | ccv_cnnp_model_t* ccv_cnnp_datatype_conversion(const int datatype, const int ref_to_last, const char* const name) | |||
| 4418 | { | |||
| 4419 | 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)); | |||
| 4420 | model_datatype_conversion->super.isa = &ccv_cnnp_datatype_conversion_isa; | |||
| 4421 | model_datatype_conversion->super.input_size = 0; | |||
| 4422 | model_datatype_conversion->super.outputs = &model_datatype_conversion->output; | |||
| 4423 | model_datatype_conversion->super.output_size = 1; | |||
| 4424 | model_datatype_conversion->datatype = datatype; | |||
| 4425 | model_datatype_conversion->ref_to_last = ref_to_last; | |||
| 4426 | ccv_cnnp_model_copy_name(&model_datatype_conversion->super, name); | |||
| 4427 | return (ccv_cnnp_model_t*)model_datatype_conversion; | |||
| 4428 | } | |||
| 4429 | ||||
| 4430 | static ccv_cnnp_model_t* _ccv_cnnp_datatype_conversion_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4431 | { | |||
| 4432 | ccv_cnnp_model_datatype_conversion_t* const self = (ccv_cnnp_model_datatype_conversion_t*)super; | |||
| 4433 | return ccv_cnnp_datatype_conversion(self->datatype, self->ref_to_last, self->super.name); | |||
| 4434 | } | |||
| 4435 | ||||
| 4436 | /// MARK - Clamp layer. | |||
| 4437 | ||||
| 4438 | typedef struct { | |||
| 4439 | ccv_cnnp_model_t super; | |||
| 4440 | ccv_nnc_tensor_symbol_t output; | |||
| 4441 | float min; | |||
| 4442 | float max; | |||
| 4443 | } ccv_cnnp_model_clamp_t; | |||
| 4444 | ||||
| 4445 | 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) | |||
| 4446 | { | |||
| 4447 | ccv_cnnp_model_clamp_t* const self = (ccv_cnnp_model_clamp_t*)super; | |||
| 4448 | 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); | |||
| 4449 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4450 | 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", 4450, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4451 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 4452 | 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 /* intentional */, outputs, output_size, 0); | |||
| 4453 | } | |||
| 4454 | ||||
| 4455 | static ccv_cnnp_model_t* _ccv_cnnp_clamp_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 4456 | ||||
| 4457 | static const ccv_cnnp_model_vtab_t ccv_cnnp_clamp_isa = { | |||
| 4458 | .build = _ccv_cnnp_clamp_build, | |||
| 4459 | .copy = _ccv_cnnp_clamp_copy, | |||
| 4460 | }; | |||
| 4461 | ||||
| 4462 | ccv_cnnp_model_t* ccv_cnnp_clamp(const float min, const float max, const char* const name) | |||
| 4463 | { | |||
| 4464 | ccv_cnnp_model_clamp_t* const model_clamp = (ccv_cnnp_model_clamp_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_clamp_t)); | |||
| 4465 | model_clamp->super.isa = &ccv_cnnp_clamp_isa; | |||
| 4466 | model_clamp->super.input_size = 0; | |||
| 4467 | model_clamp->super.outputs = &model_clamp->output; | |||
| 4468 | model_clamp->super.output_size = 1; | |||
| 4469 | model_clamp->min = min; | |||
| 4470 | model_clamp->max = max; | |||
| 4471 | ccv_cnnp_model_copy_name(&model_clamp->super, name); | |||
| 4472 | return (ccv_cnnp_model_t*)model_clamp; | |||
| 4473 | } | |||
| 4474 | ||||
| 4475 | static ccv_cnnp_model_t* _ccv_cnnp_clamp_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4476 | { | |||
| 4477 | ccv_cnnp_model_clamp_t* const self = (ccv_cnnp_model_clamp_t*)super; | |||
| 4478 | return ccv_cnnp_clamp(self->min, self->max, self->super.name); | |||
| 4479 | } | |||
| 4480 | ||||
| 4481 | // MARK - Parameter Layer | |||
| 4482 | ||||
| 4483 | typedef struct { | |||
| 4484 | ccv_cnnp_model_t super; | |||
| 4485 | float init_bound; | |||
| 4486 | ccv_nnc_tensor_symbol_t weights; | |||
| 4487 | ccv_nnc_tensor_param_t weights_params; | |||
| 4488 | ccv_nnc_tensor_symbol_t output; | |||
| 4489 | } ccv_cnnp_model_parameter_t; | |||
| 4490 | ||||
| 4491 | 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) | |||
| 4492 | { | |||
| 4493 | 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); | |||
| 4494 | 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", 4494, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4495 | ccv_cnnp_model_parameter_t* const self = (ccv_cnnp_model_parameter_t*)super; | |||
| 4496 | if (!self->weights.graph) | |||
| 4497 | self->weights = ccv_nnc_tensor_symbol_new(graph, self->weights_params, "weights"); | |||
| 4498 | 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" , 4498, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4499 | outputs[0] = ccv_cnnp_model_get_symbol(super, self->weights); | |||
| 4500 | } | |||
| 4501 | ||||
| 4502 | 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) | |||
| 4503 | { | |||
| 4504 | ccv_cnnp_model_parameter_t* const self = (ccv_cnnp_model_parameter_t*)super; | |||
| 4505 | if (self->init_bound > 0) | |||
| 4506 | 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); | |||
| 4507 | else | |||
| 4508 | 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); | |||
| 4509 | } | |||
| 4510 | ||||
| 4511 | 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) | |||
| 4512 | { | |||
| 4513 | ccv_cnnp_model_parameter_t* const self = (ccv_cnnp_model_parameter_t*)super; | |||
| 4514 | add_to_array(parameters, self->weights, is_trainable); | |||
| 4515 | } | |||
| 4516 | ||||
| 4517 | static ccv_cnnp_model_t* _ccv_cnnp_parameter_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4518 | ||||
| 4519 | static const ccv_cnnp_model_vtab_t ccv_cnnp_parameter_isa = { | |||
| 4520 | .build = _ccv_cnnp_parameter_build, | |||
| 4521 | .init_states = _ccv_cnnp_parameter_init_states, | |||
| 4522 | .add_to_parameter = _ccv_cnnp_parameter_add_to_parameter, | |||
| 4523 | .copy = _ccv_cnnp_parameter_copy, | |||
| 4524 | }; | |||
| 4525 | ||||
| 4526 | 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) | |||
| 4527 | { | |||
| 4528 | ccv_cnnp_model_parameter_t* const model_parameter = (ccv_cnnp_model_parameter_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_parameter_t)); | |||
| 4529 | model_parameter->super.isa = &ccv_cnnp_parameter_isa; | |||
| 4530 | model_parameter->super.input_size = 0; | |||
| 4531 | model_parameter->super.outputs = &model_parameter->output; | |||
| 4532 | model_parameter->super.output_size = 1; | |||
| 4533 | model_parameter->super.is_trainable = is_trainable; | |||
| 4534 | ccv_cnnp_model_copy_name(&model_parameter->super, name); | |||
| 4535 | model_parameter->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 4536 | model_parameter->weights.graph = 0; | |||
| 4537 | model_parameter->weights_params = params; | |||
| 4538 | return (ccv_cnnp_model_t*)model_parameter; | |||
| 4539 | } | |||
| 4540 | ||||
| 4541 | static ccv_cnnp_model_t* _ccv_cnnp_parameter_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4542 | { | |||
| 4543 | const ccv_cnnp_model_parameter_t* const self = (const ccv_cnnp_model_parameter_t*)super; | |||
| 4544 | return ccv_cnnp_parameter(self->weights_params, self->init_bound, self->super.is_trainable, self->super.name); | |||
| 4545 | } | |||
| 4546 | ||||
| 4547 | // MARK - Scalar Layer | |||
| 4548 | ||||
| 4549 | typedef struct { | |||
| 4550 | ccv_cnnp_model_t super; | |||
| 4551 | int type; | |||
| 4552 | int format; | |||
| 4553 | int datatype; | |||
| 4554 | float value; | |||
| 4555 | ccv_nnc_tensor_symbol_t output; | |||
| 4556 | } ccv_cnnp_model_scalar_t; | |||
| 4557 | ||||
| 4558 | 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) | |||
| 4559 | { | |||
| 4560 | 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); | |||
| 4561 | 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", 4561, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4562 | ccv_cnnp_model_scalar_t* const self = (ccv_cnnp_model_scalar_t*)super; | |||
| 4563 | ccv_nnc_tensor_param_t params = { | |||
| 4564 | .type = self->type, | |||
| 4565 | .format = self->format, | |||
| 4566 | .datatype = self->datatype, | |||
| 4567 | .dim = { | |||
| 4568 | 1 | |||
| 4569 | } | |||
| 4570 | }; | |||
| 4571 | if (input_size > 0) | |||
| 4572 | { | |||
| 4573 | ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4574 | params.type = input_params.type; | |||
| 4575 | params.format = input_params.format; | |||
| 4576 | params.datatype = input_params.datatype; | |||
| 4577 | } | |||
| 4578 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 4579 | 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); | |||
| 4580 | } | |||
| 4581 | ||||
| 4582 | static ccv_cnnp_model_t* _ccv_cnnp_scalar_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4583 | ||||
| 4584 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scalar_isa = { | |||
| 4585 | .build = _ccv_cnnp_scalar_build, | |||
| 4586 | .copy = _ccv_cnnp_scalar_copy, | |||
| 4587 | }; | |||
| 4588 | ||||
| 4589 | ccv_cnnp_model_t* ccv_cnnp_scalar(const int type, const int format, const int datatype, const float value, const char* const name) | |||
| 4590 | { | |||
| 4591 | ccv_cnnp_model_scalar_t* const model_scalar = (ccv_cnnp_model_scalar_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_scalar_t)); | |||
| 4592 | model_scalar->super.isa = &ccv_cnnp_scalar_isa; | |||
| 4593 | model_scalar->super.input_size = 0; | |||
| 4594 | model_scalar->super.outputs = &model_scalar->output; | |||
| 4595 | model_scalar->super.output_size = 1; | |||
| 4596 | ccv_cnnp_model_copy_name(&model_scalar->super, name); | |||
| 4597 | model_scalar->type = type; | |||
| 4598 | model_scalar->format = format; | |||
| 4599 | model_scalar->datatype = datatype; | |||
| 4600 | model_scalar->value = value; | |||
| 4601 | return (ccv_cnnp_model_t*)model_scalar; | |||
| 4602 | } | |||
| 4603 | ||||
| 4604 | static ccv_cnnp_model_t* _ccv_cnnp_scalar_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4605 | { | |||
| 4606 | const ccv_cnnp_model_scalar_t* const self = (const ccv_cnnp_model_scalar_t*)super; | |||
| 4607 | return ccv_cnnp_scalar(self->type, self->format, self->datatype, self->value, self->super.name); | |||
| 4608 | } | |||
| 4609 | ||||
| 4610 | // MARK - Variable Layer | |||
| 4611 | ||||
| 4612 | typedef struct { | |||
| 4613 | ccv_cnnp_model_t super; | |||
| 4614 | ccv_nnc_tensor_param_t params; | |||
| 4615 | ccv_nnc_tensor_symbol_t output; | |||
| 4616 | } ccv_cnnp_model_variable_t; | |||
| 4617 | ||||
| 4618 | 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) | |||
| 4619 | { | |||
| 4620 | 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); | |||
| 4621 | 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" , 4621, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4622 | 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", 4622, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4623 | ccv_cnnp_model_variable_t* const self = (ccv_cnnp_model_variable_t*)super; | |||
| 4624 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, self->params, 0); | |||
| 4625 | } | |||
| 4626 | ||||
| 4627 | static ccv_cnnp_model_t* _ccv_cnnp_variable_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4628 | ||||
| 4629 | static const ccv_cnnp_model_vtab_t ccv_cnnp_variable_isa = { | |||
| 4630 | .build = _ccv_cnnp_variable_build, | |||
| 4631 | .copy = _ccv_cnnp_variable_copy, | |||
| 4632 | }; | |||
| 4633 | ||||
| 4634 | ccv_cnnp_model_t* ccv_cnnp_variable(const ccv_nnc_tensor_param_t params, const char* const name) | |||
| 4635 | { | |||
| 4636 | ccv_cnnp_model_variable_t* const model_variable = (ccv_cnnp_model_variable_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_variable_t)); | |||
| 4637 | model_variable->super.isa = &ccv_cnnp_variable_isa; | |||
| 4638 | model_variable->super.input_size = 0; | |||
| 4639 | model_variable->super.outputs = &model_variable->output; | |||
| 4640 | model_variable->super.output_size = 1; | |||
| 4641 | ccv_cnnp_model_copy_name(&model_variable->super, name); | |||
| 4642 | model_variable->params = params; | |||
| 4643 | return (ccv_cnnp_model_t*)model_variable; | |||
| 4644 | } | |||
| 4645 | ||||
| 4646 | static ccv_cnnp_model_t* _ccv_cnnp_variable_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4647 | { | |||
| 4648 | const ccv_cnnp_model_variable_t* const self = (const ccv_cnnp_model_variable_t*)super; | |||
| 4649 | return ccv_cnnp_variable(self->params, self->super.name); | |||
| 4650 | } | |||
| 4651 | ||||
| 4652 | // MARK - Set Layer | |||
| 4653 | ||||
| 4654 | typedef struct { | |||
| 4655 | ccv_cnnp_model_t super; | |||
| 4656 | float value; | |||
| 4657 | ccv_nnc_tensor_symbol_t output; | |||
| 4658 | } ccv_cnnp_model_set_t; | |||
| 4659 | ||||
| 4660 | static void _ccv_cnnp_set_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) | |||
| 4661 | { | |||
| 4662 | PRINT(CCV_CLI_VERBOSE, "[cnnp_set_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_set_build] -\n"); fflush(stdout); } } while ( 0); | |||
| 4663 | 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" , 4663, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4664 | 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", 4664, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4665 | ccv_cnnp_model_set_t* const self = (ccv_cnnp_model_set_t*)super; | |||
| 4666 | outputs[0] = inputs[0]; | |||
| 4667 | 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, "set"); | |||
| 4668 | } | |||
| 4669 | ||||
| 4670 | static ccv_cnnp_model_t* _ccv_cnnp_set_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4671 | ||||
| 4672 | static const ccv_cnnp_model_vtab_t ccv_cnnp_set_isa = { | |||
| 4673 | .build = _ccv_cnnp_set_build, | |||
| 4674 | .copy = _ccv_cnnp_set_copy, | |||
| 4675 | }; | |||
| 4676 | ||||
| 4677 | ccv_cnnp_model_t* ccv_cnnp_set(const float value, const char* const name) | |||
| 4678 | { | |||
| 4679 | ccv_cnnp_model_set_t* const model_set = (ccv_cnnp_model_set_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_set_t)); | |||
| 4680 | model_set->super.isa = &ccv_cnnp_set_isa; | |||
| 4681 | model_set->super.input_size = 1; | |||
| 4682 | model_set->super.outputs = &model_set->output; | |||
| 4683 | model_set->super.output_size = 1; | |||
| 4684 | ccv_cnnp_model_copy_name(&model_set->super, name); | |||
| 4685 | model_set->value = value; | |||
| 4686 | return (ccv_cnnp_model_t*)model_set; | |||
| 4687 | } | |||
| 4688 | ||||
| 4689 | static ccv_cnnp_model_t* _ccv_cnnp_set_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4690 | { | |||
| 4691 | const ccv_cnnp_model_set_t* const self = (const ccv_cnnp_model_set_t*)super; | |||
| 4692 | return ccv_cnnp_set(self->value, self->super.name); | |||
| 4693 | } | |||
| 4694 | ||||
| 4695 | // MARK - Send Layer | |||
| 4696 | ||||
| 4697 | typedef struct { | |||
| 4698 | ccv_cnnp_model_t super; | |||
| 4699 | int device_id; | |||
| 4700 | ccv_nnc_tensor_symbol_t output; | |||
| 4701 | } ccv_cnnp_model_send_t; | |||
| 4702 | ||||
| 4703 | static void _ccv_cnnp_send_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) | |||
| 4704 | { | |||
| 4705 | ccv_cnnp_model_send_t* const self = (ccv_cnnp_model_send_t*)super; | |||
| 4706 | PRINT(CCV_CLI_VERBOSE, "[cnnp_send_build] - device_id: %d\n", self->device_id)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_send_build] - device_id: %d\n", self->device_id ); fflush(stdout); } } while (0); | |||
| 4707 | 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" , 4707, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4708 | 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", 4708, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4709 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4710 | params.type = (params.type & ~0x3) | CCV_TENSOR_GPU_MEMORY; | |||
| 4711 | CCV_TENSOR_SET_DEVICE_ID(params.type, self->device_id)(params.type) = (((params.type) & ~0xfff00) | (((self-> device_id) & 0xfff) << 8)); | |||
| 4712 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 4713 | ccv_nnc_graph_exec_symbol_new(graph, CMD_DATA_TRANSFER_FORWARD()ccv_nnc_cmd(CCV_NNC_DATA_TRANSFER_FORWARD, 0, ccv_nnc_cmd_auto , 0), inputs, 1, outputs, 1, "send"); | |||
| 4714 | } | |||
| 4715 | ||||
| 4716 | static ccv_cnnp_model_t* _ccv_cnnp_send_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4717 | ||||
| 4718 | static const ccv_cnnp_model_vtab_t ccv_cnnp_send_isa = { | |||
| 4719 | .build = _ccv_cnnp_send_build, | |||
| 4720 | .copy = _ccv_cnnp_send_copy, | |||
| 4721 | }; | |||
| 4722 | ||||
| 4723 | ccv_cnnp_model_t* ccv_cnnp_send(const int device_id, const char* const name) | |||
| 4724 | { | |||
| 4725 | assert(device_id >= 0)((void) sizeof ((device_id >= 0) ? 1 : 0), __extension__ ( { if (device_id >= 0) ; else __assert_fail ("device_id >= 0" , "ccv_cnnp_model_addons.c", 4725, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4726 | ccv_cnnp_model_send_t* const model_send = (ccv_cnnp_model_send_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_send_t)); | |||
| 4727 | model_send->super.isa = &ccv_cnnp_send_isa; | |||
| 4728 | model_send->super.input_size = 1; | |||
| 4729 | model_send->super.outputs = &model_send->output; | |||
| 4730 | model_send->super.output_size = 1; | |||
| 4731 | ccv_cnnp_model_copy_name(&model_send->super, name); | |||
| 4732 | model_send->device_id = device_id; | |||
| 4733 | return (ccv_cnnp_model_t*)model_send; | |||
| 4734 | } | |||
| 4735 | ||||
| 4736 | static ccv_cnnp_model_t* _ccv_cnnp_send_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4737 | { | |||
| 4738 | const ccv_cnnp_model_send_t* const self = (const ccv_cnnp_model_send_t*)super; | |||
| 4739 | return ccv_cnnp_send(self->device_id, self->super.name); | |||
| 4740 | } | |||
| 4741 | ||||
| 4742 | // MARK - Move Layer | |||
| 4743 | ||||
| 4744 | typedef struct { | |||
| 4745 | ccv_cnnp_model_t super; | |||
| 4746 | ccv_nnc_tensor_symbol_t output; | |||
| 4747 | } ccv_cnnp_model_move_t; | |||
| 4748 | ||||
| 4749 | 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) | |||
| 4750 | { | |||
| 4751 | 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); | |||
| 4752 | 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" , 4752, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4753 | 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", 4753, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4754 | outputs[0] = inputs[1]; | |||
| 4755 | 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"); | |||
| 4756 | } | |||
| 4757 | ||||
| 4758 | static ccv_cnnp_model_t* _ccv_cnnp_move_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4759 | ||||
| 4760 | static const ccv_cnnp_model_vtab_t ccv_cnnp_move_isa = { | |||
| 4761 | .build = _ccv_cnnp_move_build, | |||
| 4762 | .copy = _ccv_cnnp_move_copy, | |||
| 4763 | }; | |||
| 4764 | ||||
| 4765 | ccv_cnnp_model_t* ccv_cnnp_move(const char* const name) | |||
| 4766 | { | |||
| 4767 | ccv_cnnp_model_move_t* const model_move = (ccv_cnnp_model_move_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_move_t)); | |||
| 4768 | model_move->super.isa = &ccv_cnnp_move_isa; | |||
| 4769 | model_move->super.input_size = 2; | |||
| 4770 | model_move->super.outputs = &model_move->output; | |||
| 4771 | model_move->super.output_size = 1; | |||
| 4772 | ccv_cnnp_model_copy_name(&model_move->super, name); | |||
| 4773 | return (ccv_cnnp_model_t*)model_move; | |||
| 4774 | } | |||
| 4775 | ||||
| 4776 | static ccv_cnnp_model_t* _ccv_cnnp_move_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4777 | { | |||
| 4778 | const ccv_cnnp_model_move_t* const self = (const ccv_cnnp_model_move_t*)super; | |||
| 4779 | return ccv_cnnp_move(self->super.name); | |||
| 4780 | } | |||
| 4781 | ||||
| 4782 | // MARK - "Making" Contiguous Layer | |||
| 4783 | ||||
| 4784 | typedef struct { | |||
| 4785 | ccv_cnnp_model_t super; | |||
| 4786 | ccv_nnc_tensor_symbol_t output; | |||
| 4787 | } ccv_cnnp_model_contiguous_t; | |||
| 4788 | ||||
| 4789 | 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) | |||
| 4790 | { | |||
| 4791 | 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); | |||
| 4792 | 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" , 4792, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4793 | 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", 4793, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4794 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4795 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 4796 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 4797 | { | |||
| 4798 | outputs[0] = inputs[0]; | |||
| 4799 | return; | |||
| 4800 | } | |||
| 4801 | // Otherwise, we need to check its stride to know if it is contiguous. | |||
| 4802 | int old_stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 4803 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], 0, old_stride); | |||
| 4804 | // We identify permute by checking if the stride is not in descending order. | |||
| 4805 | // This also covered "permute" through reshape, rather than using ccv_cnnp_permute directly. | |||
| 4806 | if (ccv_nnc_is_tensor_stride_packed(old_stride, params.dim)) | |||
| 4807 | { | |||
| 4808 | outputs[0] = inputs[0]; | |||
| 4809 | return; | |||
| 4810 | } | |||
| 4811 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 4812 | 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"); | |||
| 4813 | ccv_nnc_graph_exec_symbol_set_flags(graph, make_contiguous, CCV_NNC_GRAPH_EXEC_DISABLE_OPT); | |||
| 4814 | } | |||
| 4815 | ||||
| 4816 | static ccv_cnnp_model_t* _ccv_cnnp_contiguous_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4817 | ||||
| 4818 | static const ccv_cnnp_model_vtab_t ccv_cnnp_contiguous_isa = { | |||
| 4819 | .build = _ccv_cnnp_contiguous_build, | |||
| 4820 | .copy = _ccv_cnnp_contiguous_copy, | |||
| 4821 | }; | |||
| 4822 | ||||
| 4823 | ccv_cnnp_model_t* ccv_cnnp_contiguous(const char* const name) | |||
| 4824 | { | |||
| 4825 | ccv_cnnp_model_contiguous_t* const model_contiguous = (ccv_cnnp_model_contiguous_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_contiguous_t)); | |||
| 4826 | model_contiguous->super.isa = &ccv_cnnp_contiguous_isa; | |||
| 4827 | model_contiguous->super.input_size = 1; | |||
| 4828 | model_contiguous->super.outputs = &model_contiguous->output; | |||
| 4829 | model_contiguous->super.output_size = 1; | |||
| 4830 | ccv_cnnp_model_copy_name(&model_contiguous->super, name); | |||
| 4831 | return (ccv_cnnp_model_t*)model_contiguous; | |||
| 4832 | } | |||
| 4833 | ||||
| 4834 | static ccv_cnnp_model_t* _ccv_cnnp_contiguous_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4835 | { | |||
| 4836 | const ccv_cnnp_model_contiguous_t* const self = (const ccv_cnnp_model_contiguous_t*)super; | |||
| 4837 | return ccv_cnnp_contiguous(self->super.name); | |||
| 4838 | } | |||
| 4839 | ||||
| 4840 | // MARK - "Making" Copy Layer | |||
| 4841 | ||||
| 4842 | typedef struct { | |||
| 4843 | ccv_cnnp_model_t super; | |||
| 4844 | ccv_nnc_tensor_symbol_t output; | |||
| 4845 | } ccv_cnnp_model_copy_t; | |||
| 4846 | ||||
| 4847 | static void _ccv_cnnp_copy_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) | |||
| 4848 | { | |||
| 4849 | PRINT(CCV_CLI_VERBOSE, "[cnnp_copy_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_copy_build] -\n"); fflush(stdout); } } while (0); | |||
| 4850 | 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" , 4850, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4851 | 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", 4851, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4852 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4853 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 4854 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 4855 | { | |||
| 4856 | outputs[0] = inputs[0]; | |||
| 4857 | return; | |||
| 4858 | } | |||
| 4859 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 4860 | 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"); | |||
| 4861 | ccv_nnc_graph_exec_symbol_set_flags(graph, make_contiguous, CCV_NNC_GRAPH_EXEC_DISABLE_OPT); | |||
| 4862 | } | |||
| 4863 | ||||
| 4864 | static ccv_cnnp_model_t* _ccv_cnnp_copy_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4865 | ||||
| 4866 | static const ccv_cnnp_model_vtab_t ccv_cnnp_copy_isa = { | |||
| 4867 | .build = _ccv_cnnp_copy_build, | |||
| 4868 | .copy = _ccv_cnnp_copy_copy, | |||
| 4869 | }; | |||
| 4870 | ||||
| 4871 | ccv_cnnp_model_t* ccv_cnnp_copy(const char* const name) | |||
| 4872 | { | |||
| 4873 | ccv_cnnp_model_copy_t* const model_copy = (ccv_cnnp_model_copy_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_copy_t)); | |||
| 4874 | model_copy->super.isa = &ccv_cnnp_copy_isa; | |||
| 4875 | model_copy->super.input_size = 1; | |||
| 4876 | model_copy->super.outputs = &model_copy->output; | |||
| 4877 | model_copy->super.output_size = 1; | |||
| 4878 | ccv_cnnp_model_copy_name(&model_copy->super, name); | |||
| 4879 | return (ccv_cnnp_model_t*)model_copy; | |||
| 4880 | } | |||
| 4881 | ||||
| 4882 | static ccv_cnnp_model_t* _ccv_cnnp_copy_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4883 | { | |||
| 4884 | const ccv_cnnp_model_copy_t* const self = (const ccv_cnnp_model_copy_t*)super; | |||
| 4885 | return ccv_cnnp_copy(self->super.name); | |||
| 4886 | } | |||
| 4887 | ||||
| 4888 | // MARK - All-To-All Layer | |||
| 4889 | ||||
| 4890 | typedef struct { | |||
| 4891 | ccv_cnnp_model_t super; | |||
| 4892 | int axis; | |||
| 4893 | ccv_nnc_tensor_symbol_t outputs[1]; | |||
| 4894 | } ccv_cnnp_model_all_to_all_t; | |||
| 4895 | ||||
| 4896 | static void _ccv_cnnp_all_to_all_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) | |||
| 4897 | { | |||
| 4898 | ccv_cnnp_model_all_to_all_t* const self = (ccv_cnnp_model_all_to_all_t*)super; | |||
| 4899 | PRINT(CCV_CLI_VERBOSE, "[cnnp_all_to_all_build] - axis: %d\n", self->axis)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_all_to_all_build] - axis: %d\n", self->axis ); fflush(stdout); } } while (0); | |||
| 4900 | 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", 4900, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4901 | assert(output_size == self->super.output_size)((void) sizeof ((output_size == self->super.output_size) ? 1 : 0), __extension__ ({ if (output_size == self->super.output_size ) ; else __assert_fail ("output_size == self->super.output_size" , "ccv_cnnp_model_addons.c", 4901, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4902 | assert(input_size == output_size)((void) sizeof ((input_size == output_size) ? 1 : 0), __extension__ ({ if (input_size == output_size) ; else __assert_fail ("input_size == output_size" , "ccv_cnnp_model_addons.c", 4902, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4903 | 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", 4903, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4904 | const ccv_nnc_tensor_param_t input_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4905 | const int nd = ccv_nnc_tensor_nd(input_params.dim); | |||
| 4906 | assert(self->axis >= 0 && self->axis < nd)((void) sizeof ((self->axis >= 0 && self->axis < nd) ? 1 : 0), __extension__ ({ if (self->axis >= 0 && self->axis < nd) ; else __assert_fail ("self->axis >= 0 && self->axis < nd" , "ccv_cnnp_model_addons.c", 4906, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4907 | assert(input_params.dim[self->axis] % input_size == 0)((void) sizeof ((input_params.dim[self->axis] % input_size == 0) ? 1 : 0), __extension__ ({ if (input_params.dim[self-> axis] % input_size == 0) ; else __assert_fail ("input_params.dim[self->axis] % input_size == 0" , "ccv_cnnp_model_addons.c", 4907, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4908 | int i; | |||
| 4909 | for (i = 0; i < input_size; i++) | |||
| 4910 | { | |||
| 4911 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 4912 | assert(params.format == input_params.format)((void) sizeof ((params.format == input_params.format) ? 1 : 0 ), __extension__ ({ if (params.format == input_params.format) ; else __assert_fail ("params.format == input_params.format" , "ccv_cnnp_model_addons.c", 4912, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4913 | assert(params.datatype == input_params.datatype)((void) sizeof ((params.datatype == input_params.datatype) ? 1 : 0), __extension__ ({ if (params.datatype == input_params.datatype ) ; else __assert_fail ("params.datatype == input_params.datatype" , "ccv_cnnp_model_addons.c", 4913, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4914 | assert(CCV_TENSOR_GET_MEMORY(params.type) == CCV_TENSOR_GET_MEMORY(input_params.type))((void) sizeof ((((params.type) & 0x3) == ((input_params. type) & 0x3)) ? 1 : 0), __extension__ ({ if (((params.type ) & 0x3) == ((input_params.type) & 0x3)) ; else __assert_fail ("CCV_TENSOR_GET_MEMORY(params.type) == CCV_TENSOR_GET_MEMORY(input_params.type)" , "ccv_cnnp_model_addons.c", 4914, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4915 | assert(memcmp(params.dim, input_params.dim, sizeof(input_params.dim)) == 0)((void) sizeof ((memcmp(params.dim, input_params.dim, sizeof( input_params.dim)) == 0) ? 1 : 0), __extension__ ({ if (memcmp (params.dim, input_params.dim, sizeof(input_params.dim)) == 0 ) ; else __assert_fail ("memcmp(params.dim, input_params.dim, sizeof(input_params.dim)) == 0" , "ccv_cnnp_model_addons.c", 4915, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4916 | outputs[i] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 4917 | } | |||
| 4918 | ccv_nnc_graph_exec_symbol_new(graph, CMD_COMM_ALL_TO_ALL_FORWARD(self->axis)ccv_nnc_cmd(CCV_NNC_COMM_ALL_TO_ALL_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.all_to_all={.axis=self->axis}}), 0 ), inputs, input_size, outputs, output_size, "all_to_all"); | |||
| 4919 | } | |||
| 4920 | ||||
| 4921 | static ccv_cnnp_model_t* _ccv_cnnp_all_to_all_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 4922 | ||||
| 4923 | static const ccv_cnnp_model_vtab_t ccv_cnnp_all_to_all_isa = { | |||
| 4924 | .build = _ccv_cnnp_all_to_all_build, | |||
| 4925 | .copy = _ccv_cnnp_all_to_all_copy, | |||
| 4926 | }; | |||
| 4927 | ||||
| 4928 | ccv_cnnp_model_t* ccv_cnnp_all_to_all(const int count, const int axis, const char* const name) | |||
| 4929 | { | |||
| 4930 | assert(count > 0)((void) sizeof ((count > 0) ? 1 : 0), __extension__ ({ if ( count > 0) ; else __assert_fail ("count > 0", "ccv_cnnp_model_addons.c" , 4930, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4931 | assert(axis >= 0)((void) sizeof ((axis >= 0) ? 1 : 0), __extension__ ({ if ( axis >= 0) ; else __assert_fail ("axis >= 0", "ccv_cnnp_model_addons.c" , 4931, __extension__ __PRETTY_FUNCTION__); })); | |||
| 4932 | ccv_cnnp_model_all_to_all_t* const model_all_to_all = (ccv_cnnp_model_all_to_all_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_all_to_all_t) + sizeof(ccv_nnc_tensor_symbol_t) * (count - 1)); | |||
| 4933 | model_all_to_all->super.isa = &ccv_cnnp_all_to_all_isa; | |||
| 4934 | model_all_to_all->super.input_size = count; | |||
| 4935 | model_all_to_all->super.outputs = model_all_to_all->outputs; | |||
| 4936 | model_all_to_all->super.output_size = count; | |||
| 4937 | model_all_to_all->axis = axis; | |||
| 4938 | ccv_cnnp_model_copy_name(&model_all_to_all->super, name); | |||
| 4939 | return (ccv_cnnp_model_t*)model_all_to_all; | |||
| 4940 | } | |||
| 4941 | ||||
| 4942 | static ccv_cnnp_model_t* _ccv_cnnp_all_to_all_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 4943 | { | |||
| 4944 | const ccv_cnnp_model_all_to_all_t* const self = (const ccv_cnnp_model_all_to_all_t*)super; | |||
| 4945 | return ccv_cnnp_all_to_all(self->super.output_size, self->axis, self->super.name); | |||
| 4946 | } | |||
| 4947 | ||||
| 4948 | // MARK - Scaled-Dot Product Attention Layer | |||
| 4949 | ||||
| 4950 | typedef struct { | |||
| 4951 | ccv_cnnp_model_t super; | |||
| 4952 | ccv_nnc_tensor_symbol_t output; | |||
| 4953 | ccv_nnc_tensor_symbol_t weights; | |||
| 4954 | ccv_nnc_tensor_symbol_t bias; | |||
| 4955 | float scale; | |||
| 4956 | int is_causal; | |||
| 4957 | int has_attn_mask; | |||
| 4958 | int is_varlen; | |||
| 4959 | int max_seqlen_q; | |||
| 4960 | int max_seqlen_kv; | |||
| 4961 | int flags; | |||
| 4962 | int attention_sinks; | |||
| 4963 | int sliding_window; | |||
| 4964 | int fused_unify_head_weights; | |||
| 4965 | int no_bias; | |||
| 4966 | } ccv_cnnp_model_scaled_dot_product_attention_t; | |||
| 4967 | ||||
| 4968 | 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) | |||
| 4969 | { | |||
| 4970 | 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); | |||
| 4971 | 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", 4971, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4972 | ccv_cnnp_model_scaled_dot_product_attention_t* const self = (ccv_cnnp_model_scaled_dot_product_attention_t*)super; | |||
| 4973 | assert(input_size == (self->is_varlen ? 5 : (self->has_attn_mask ? 4 : 3)) + self->attention_sinks)((void) sizeof ((input_size == (self->is_varlen ? 5 : (self ->has_attn_mask ? 4 : 3)) + self->attention_sinks) ? 1 : 0), __extension__ ({ if (input_size == (self->is_varlen ? 5 : (self->has_attn_mask ? 4 : 3)) + self->attention_sinks ) ; else __assert_fail ("input_size == (self->is_varlen ? 5 : (self->has_attn_mask ? 4 : 3)) + self->attention_sinks" , "ccv_cnnp_model_addons.c", 4973, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4974 | assert(!self->is_varlen || !self->has_attn_mask)((void) sizeof ((!self->is_varlen || !self->has_attn_mask ) ? 1 : 0), __extension__ ({ if (!self->is_varlen || !self ->has_attn_mask) ; else __assert_fail ("!self->is_varlen || !self->has_attn_mask" , "ccv_cnnp_model_addons.c", 4974, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4975 | assert(!self->is_varlen || !self->fused_unify_head_weights)((void) sizeof ((!self->is_varlen || !self->fused_unify_head_weights ) ? 1 : 0), __extension__ ({ if (!self->is_varlen || !self ->fused_unify_head_weights) ; else __assert_fail ("!self->is_varlen || !self->fused_unify_head_weights" , "ccv_cnnp_model_addons.c", 4975, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4976 | const ccv_nnc_tensor_param_t q_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 4977 | const ccv_nnc_tensor_param_t k_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 4978 | const ccv_nnc_tensor_param_t v_params = ccv_nnc_tensor_symbol_params(graph, inputs[2]); | |||
| 4979 | const int sink_idx = self->is_varlen ? 5 : (3 + self->has_attn_mask); | |||
| 4980 | const ccv_nnc_tensor_param_t sinks_params = self->attention_sinks ? ccv_nnc_tensor_symbol_params(graph, inputs[sink_idx]) : (ccv_nnc_tensor_param_t){}; | |||
| 4981 | const ccv_nnc_tensor_param_t attn_mask_params = self->has_attn_mask ? ccv_nnc_tensor_symbol_params(graph, inputs[3]) : (ccv_nnc_tensor_param_t){}; | |||
| 4982 | const ccv_nnc_tensor_param_t q_seq_offsets_params = self->is_varlen ? ccv_nnc_tensor_symbol_params(graph, inputs[3]) : (ccv_nnc_tensor_param_t){}; | |||
| 4983 | const ccv_nnc_tensor_param_t kv_seq_offsets_params = self->is_varlen ? ccv_nnc_tensor_symbol_params(graph, inputs[4]) : (ccv_nnc_tensor_param_t){}; | |||
| 4984 | const int v_nd = ccv_nnc_tensor_nd(v_params.dim); | |||
| 4985 | 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", 4985, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 4986 | const int hEv = (v_nd == 3 ? 1 : v_params.dim[2]) * v_params.dim[v_nd - 1]; | |||
| 4987 | ccv_nnc_tensor_param_t weights_params = q_params; | |||
| 4988 | memset(weights_params.dim, 0, sizeof(weights_params.dim)); | |||
| 4989 | weights_params.dim[0] = hEv; | |||
| 4990 | weights_params.dim[1] = hEv; | |||
| 4991 | ccv_nnc_tensor_param_t bias_params = q_params; | |||
| 4992 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 4993 | bias_params.dim[0] = hEv; | |||
| 4994 | ccv_nnc_cmd_t cmd = {0}; | |||
| 4995 | cmd.cmd = CCV_NNC_SCALED_DOT_PRODUCT_ATTENTION_FORWARD; | |||
| 4996 | cmd.info.scaled_dot_product_attention.scale = self->scale; | |||
| 4997 | cmd.info.scaled_dot_product_attention.is_causal = self->is_causal; | |||
| 4998 | cmd.info.scaled_dot_product_attention.is_varlen = self->is_varlen; | |||
| 4999 | cmd.info.scaled_dot_product_attention.max_seqlen_q = self->max_seqlen_q; | |||
| 5000 | cmd.info.scaled_dot_product_attention.max_seqlen_kv = self->max_seqlen_kv; | |||
| 5001 | cmd.info.scaled_dot_product_attention.flags = self->flags; | |||
| 5002 | cmd.info.scaled_dot_product_attention.attention_sinks = self->attention_sinks; | |||
| 5003 | cmd.info.scaled_dot_product_attention.sliding_window = self->sliding_window; | |||
| 5004 | ccv_nnc_tensor_param_t output_params[3]; | |||
| 5005 | ccv_nnc_tensor_symbol_t output; | |||
| 5006 | ccv_nnc_tensor_symbol_t saved_softmax_lse; | |||
| 5007 | ccv_nnc_tensor_symbol_t saved_v_proj = NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 5008 | ccv_nnc_tensor_symbol_t attn_mask = NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 5009 | ccv_nnc_tensor_symbol_t weights = NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 5010 | ccv_nnc_tensor_symbol_t bias = NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 5011 | ccv_nnc_tensor_symbol_t sinks = self->attention_sinks ? inputs[sink_idx] : NO_TENSOR_SYMBOL(const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }; | |||
| 5012 | if (self->has_attn_mask) | |||
| 5013 | attn_mask = inputs[3]; | |||
| 5014 | if (self->is_varlen) | |||
| 5015 | { | |||
| 5016 | ccv_nnc_tensor_param_t input_params[9] = { | |||
| 5017 | q_params, | |||
| 5018 | k_params, | |||
| 5019 | v_params, | |||
| 5020 | (ccv_nnc_tensor_param_t){}, | |||
| 5021 | (ccv_nnc_tensor_param_t){}, | |||
| 5022 | (ccv_nnc_tensor_param_t){}, | |||
| 5023 | q_seq_offsets_params, | |||
| 5024 | kv_seq_offsets_params, | |||
| 5025 | sinks_params, | |||
| 5026 | }; | |||
| 5027 | ccv_nnc_hint_tensor_auto(cmd, input_params, self->attention_sinks ? 9 : 8, ccv_nnc_no_hint, output_params, 2); | |||
| 5028 | output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 5029 | saved_softmax_lse = ccv_nnc_tensor_symbol_new(graph, output_params[1], 0); | |||
| 5030 | if (self->attention_sinks) | |||
| 5031 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], inputs[2], NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, inputs[3], inputs[4], sinks)(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], inputs [2], (const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }, (const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }, (const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }, inputs[3], inputs[4], sinks}, (1 +1 +1 +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 -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"); | |||
| 5032 | else | |||
| 5033 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], inputs[2], NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, inputs[3], inputs[4])(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], inputs [2], (const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }, (const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }, (const ccv_nnc_tensor_symbol_t){.d = CCV_NNC_NO_TENSOR_SYMBOL }, inputs[3], inputs[4]}, (1 +1 +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 -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"); | |||
| 5034 | } else if (self->fused_unify_head_weights) | |||
| 5035 | { | |||
| 5036 | if (!self->weights.graph) | |||
| 5037 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 5038 | 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" , 5038, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5039 | weights = ccv_cnnp_model_get_symbol(super, self->weights); | |||
| 5040 | if (!self->no_bias) | |||
| 5041 | { | |||
| 5042 | if (!self->bias.graph) | |||
| 5043 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 5044 | assert(self->bias.graph == graph)((void) sizeof ((self->bias.graph == graph) ? 1 : 0), __extension__ ({ if (self->bias.graph == graph) ; else __assert_fail ("self->bias.graph == graph" , "ccv_cnnp_model_addons.c", 5044, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5045 | bias = ccv_cnnp_model_get_symbol(super, self->bias); | |||
| 5046 | } | |||
| 5047 | ccv_nnc_tensor_param_t input_params[9] = { | |||
| 5048 | q_params, | |||
| 5049 | k_params, | |||
| 5050 | v_params, | |||
| 5051 | attn_mask_params, | |||
| 5052 | weights_params, | |||
| 5053 | bias_params, | |||
| 5054 | (ccv_nnc_tensor_param_t){}, | |||
| 5055 | (ccv_nnc_tensor_param_t){}, | |||
| 5056 | sinks_params, | |||
| 5057 | }; | |||
| 5058 | ccv_nnc_hint_tensor_auto(cmd, input_params, self->attention_sinks ? 9 : 6, ccv_nnc_no_hint, output_params, 3); | |||
| 5059 | output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 5060 | saved_softmax_lse = ccv_nnc_tensor_symbol_new(graph, output_params[1], 0); | |||
| 5061 | saved_v_proj = ccv_nnc_tensor_symbol_new(graph, output_params[2], 0); | |||
| 5062 | if (self->attention_sinks) | |||
| 5063 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], inputs[2], attn_mask, weights, bias, NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, sinks)(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], inputs [2], attn_mask, weights, bias, (const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, (const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, sinks}, (1 +1 +1 +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 -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"); | |||
| 5064 | else | |||
| 5065 | 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"); | |||
| 5066 | } else { | |||
| 5067 | ccv_nnc_tensor_param_t input_params[9] = { | |||
| 5068 | q_params, | |||
| 5069 | k_params, | |||
| 5070 | v_params, | |||
| 5071 | attn_mask_params, | |||
| 5072 | (ccv_nnc_tensor_param_t){}, | |||
| 5073 | (ccv_nnc_tensor_param_t){}, | |||
| 5074 | (ccv_nnc_tensor_param_t){}, | |||
| 5075 | (ccv_nnc_tensor_param_t){}, | |||
| 5076 | sinks_params, | |||
| 5077 | }; | |||
| 5078 | ccv_nnc_hint_tensor_auto(cmd, input_params, self->attention_sinks ? 9 : 3, ccv_nnc_no_hint, output_params, 2); | |||
| 5079 | output = ccv_nnc_tensor_symbol_new(graph, output_params[0], 0); | |||
| 5080 | saved_softmax_lse = ccv_nnc_tensor_symbol_new(graph, output_params[1], 0); | |||
| 5081 | if (self->attention_sinks) | |||
| 5082 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], inputs[2], attn_mask, weights, bias, NO_TENSOR_SYMBOL, NO_TENSOR_SYMBOL, sinks)(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], inputs [2], attn_mask, weights, bias, (const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, (const ccv_nnc_tensor_symbol_t ){.d = CCV_NNC_NO_TENSOR_SYMBOL}, sinks}, (1 +1 +1 +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 -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"); | |||
| 5083 | else | |||
| 5084 | 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"); | |||
| 5085 | } | |||
| 5086 | outputs[0] = output; | |||
| 5087 | } | |||
| 5088 | ||||
| 5089 | 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) | |||
| 5090 | { | |||
| 5091 | ccv_cnnp_model_scaled_dot_product_attention_t* const self = (ccv_cnnp_model_scaled_dot_product_attention_t*)super; | |||
| 5092 | if (self->weights.graph) | |||
| 5093 | { | |||
| 5094 | 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" , 5094, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5095 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 5096 | const int c = weight_params.dim[1]; | |||
| 5097 | const float std = sqrtf(2) / sqrtf(c); | |||
| 5098 | const float bound = sqrtf(3) * std; | |||
| 5099 | 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); | |||
| 5100 | if (self->bias.graph) | |||
| 5101 | 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); | |||
| 5102 | } | |||
| 5103 | } | |||
| 5104 | ||||
| 5105 | 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) | |||
| 5106 | { | |||
| 5107 | ccv_cnnp_model_scaled_dot_product_attention_t* const self = (ccv_cnnp_model_scaled_dot_product_attention_t*)super; | |||
| 5108 | if (self->weights.graph) | |||
| 5109 | { | |||
| 5110 | 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" , 5110, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5111 | add_to_array(parameters, self->weights, is_trainable); | |||
| 5112 | if (self->bias.graph) | |||
| 5113 | add_to_array(parameters, self->bias, is_trainable); | |||
| 5114 | } | |||
| 5115 | } | |||
| 5116 | ||||
| 5117 | static ccv_cnnp_model_t* _ccv_cnnp_scaled_dot_product_attention_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 5118 | ||||
| 5119 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scaled_dot_product_attention_isa = { | |||
| 5120 | .build = _ccv_cnnp_scaled_dot_product_attention_build, | |||
| 5121 | .copy = _ccv_cnnp_scaled_dot_product_attention_copy, | |||
| 5122 | }; | |||
| 5123 | ||||
| 5124 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scaled_dot_product_attention_fused_isa = { | |||
| 5125 | .build = _ccv_cnnp_scaled_dot_product_attention_build, | |||
| 5126 | .init_states = _ccv_cnnp_scaled_dot_product_attention_init_states, | |||
| 5127 | .add_to_parameter = _ccv_cnnp_scaled_dot_product_attention_add_to_parameter, | |||
| 5128 | .copy = _ccv_cnnp_scaled_dot_product_attention_copy, | |||
| 5129 | }; | |||
| 5130 | ||||
| 5131 | ccv_cnnp_model_t* ccv_cnnp_scaled_dot_product_attention(const float scale, const int is_causal, const int has_attn_mask, const int is_varlen, const int max_seqlen_q, const int max_seqlen_kv, const int flags, const int attention_sinks, const int sliding_window, const int fused_unify_head_weights, const int no_bias, const int is_trainable, const char* const name) | |||
| 5132 | { | |||
| 5133 | assert(!is_varlen || !has_attn_mask)((void) sizeof ((!is_varlen || !has_attn_mask) ? 1 : 0), __extension__ ({ if (!is_varlen || !has_attn_mask) ; else __assert_fail ("!is_varlen || !has_attn_mask" , "ccv_cnnp_model_addons.c", 5133, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5134 | assert(!is_varlen || !fused_unify_head_weights)((void) sizeof ((!is_varlen || !fused_unify_head_weights) ? 1 : 0), __extension__ ({ if (!is_varlen || !fused_unify_head_weights ) ; else __assert_fail ("!is_varlen || !fused_unify_head_weights" , "ccv_cnnp_model_addons.c", 5134, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5135 | assert(sliding_window == 0 || is_causal)((void) sizeof ((sliding_window == 0 || is_causal) ? 1 : 0), __extension__ ({ if (sliding_window == 0 || is_causal) ; else __assert_fail ("sliding_window == 0 || is_causal", "ccv_cnnp_model_addons.c" , 5135, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5136 | assert(sliding_window == 0 || !is_varlen)((void) sizeof ((sliding_window == 0 || !is_varlen) ? 1 : 0), __extension__ ({ if (sliding_window == 0 || !is_varlen) ; else __assert_fail ("sliding_window == 0 || !is_varlen", "ccv_cnnp_model_addons.c" , 5136, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5137 | 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)); | |||
| 5138 | 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; | |||
| 5139 | const int has_attention_sinks = attention_sinks ? 1 : 0; | |||
| 5140 | model_scaled_dot_product_attention->super.input_size = (is_varlen ? 5 : (has_attn_mask ? 4 : 3)) + has_attention_sinks; | |||
| 5141 | model_scaled_dot_product_attention->super.outputs = &model_scaled_dot_product_attention->output; | |||
| 5142 | model_scaled_dot_product_attention->super.output_size = 1; | |||
| 5143 | model_scaled_dot_product_attention->super.is_trainable = is_trainable; | |||
| 5144 | ccv_cnnp_model_copy_name(&model_scaled_dot_product_attention->super, name); | |||
| 5145 | model_scaled_dot_product_attention->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 5146 | model_scaled_dot_product_attention->weights.graph = 0; | |||
| 5147 | model_scaled_dot_product_attention->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 5148 | model_scaled_dot_product_attention->bias.graph = 0; | |||
| 5149 | model_scaled_dot_product_attention->scale = scale; | |||
| 5150 | model_scaled_dot_product_attention->is_causal = is_causal; | |||
| 5151 | model_scaled_dot_product_attention->has_attn_mask = has_attn_mask; | |||
| 5152 | model_scaled_dot_product_attention->is_varlen = is_varlen; | |||
| 5153 | model_scaled_dot_product_attention->max_seqlen_q = max_seqlen_q; | |||
| 5154 | model_scaled_dot_product_attention->max_seqlen_kv = max_seqlen_kv; | |||
| 5155 | model_scaled_dot_product_attention->flags = flags; | |||
| 5156 | model_scaled_dot_product_attention->attention_sinks = has_attention_sinks; | |||
| 5157 | model_scaled_dot_product_attention->sliding_window = sliding_window; | |||
| 5158 | model_scaled_dot_product_attention->fused_unify_head_weights = fused_unify_head_weights; | |||
| 5159 | model_scaled_dot_product_attention->no_bias = no_bias; | |||
| 5160 | return (ccv_cnnp_model_t*)model_scaled_dot_product_attention; | |||
| 5161 | } | |||
| 5162 | ||||
| 5163 | static ccv_cnnp_model_t* _ccv_cnnp_scaled_dot_product_attention_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5164 | { | |||
| 5165 | const ccv_cnnp_model_scaled_dot_product_attention_t* const self = (const ccv_cnnp_model_scaled_dot_product_attention_t*)super; | |||
| 5166 | return ccv_cnnp_scaled_dot_product_attention(self->scale, self->is_causal, self->has_attn_mask, self->is_varlen, self->max_seqlen_q, self->max_seqlen_kv, self->flags, self->attention_sinks, self->sliding_window, self->fused_unify_head_weights, self->no_bias, self->super.is_trainable, self->super.name); | |||
| 5167 | } | |||
| 5168 | ||||
| 5169 | // MARK - Debug Layer | |||
| 5170 | ||||
| 5171 | typedef struct { | |||
| 5172 | ccv_cnnp_model_t super; | |||
| 5173 | ccv_nnc_tensor_symbol_t output; | |||
| 5174 | ccv_cnnp_model_debug_f debugger; | |||
| 5175 | ccv_cnnp_model_debug_context_deinit_f debug_deinit; | |||
| 5176 | ccv_cnnp_model_debug_context_copy_f debug_copy; | |||
| 5177 | void* debug_context; | |||
| 5178 | } ccv_cnnp_model_debug_t; | |||
| 5179 | ||||
| 5180 | 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) | |||
| 5181 | { | |||
| 5182 | if (cmd.cmd == CCV_NNC_CUSTOM_BACKWARD) | |||
| 5183 | { | |||
| 5184 | 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", 5184, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5185 | } | |||
| 5186 | ccv_cnnp_model_debug_t* const self = (ccv_cnnp_model_debug_t*)cmd.data; | |||
| 5187 | self->debugger(inputs, input_size, stream_context, self->debug_context); | |||
| 5188 | return CCV_NNC_EXEC_SUCCESS; | |||
| 5189 | } | |||
| 5190 | ||||
| 5191 | static ccv_nnc_cmd_vtab_t ccv_cnnp_debug_exec_isa = { | |||
| 5192 | .exec = _ccv_cnnp_debug_exec | |||
| 5193 | }; | |||
| 5194 | ||||
| 5195 | 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) | |||
| 5196 | { | |||
| 5197 | 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); | |||
| 5198 | 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", 5198, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5199 | 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", 5199, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5200 | ccv_nnc_tensor_symbol_t to = ccv_nnc_tensor_symbol_alias_to(graph, inputs[0]); | |||
| 5201 | ccv_nnc_tensor_param_t output_params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 5202 | if (to.d == CCV_NNC_NO_TENSOR_SYMBOL) // If we are not reshape an alias, it is straightforward. | |||
| 5203 | { | |||
| 5204 | int ofs[CCV_NNC_MAX_DIM_ALLOC(12)] = {0}; | |||
| 5205 | int stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 5206 | ccv_nnc_tensor_get_stride(output_params.dim, stride); | |||
| 5207 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, inputs[0], ofs, stride, output_params, 0); | |||
| 5208 | } else { | |||
| 5209 | int old_ofs[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 5210 | int old_stride[CCV_NNC_MAX_DIM_ALLOC(12)]; | |||
| 5211 | ccv_nnc_tensor_symbol_alias_params(graph, inputs[0], old_ofs, old_stride); | |||
| 5212 | outputs[0] = ccv_nnc_tensor_symbol_alias_new(graph, to, old_ofs, old_stride, output_params, 0); | |||
| 5213 | } | |||
| 5214 | 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); | |||
| 5215 | cmd.data = self; | |||
| 5216 | ccv_nnc_graph_exec_symbol_t make_debug = ccv_nnc_graph_exec_symbol_new(graph, cmd, inputs, input_size, outputs, 1, "debug"); | |||
| 5217 | // Disable any optimizations. | |||
| 5218 | ccv_nnc_graph_exec_symbol_set_flags(graph, make_debug, CCV_NNC_GRAPH_EXEC_DISABLE_OPT); | |||
| 5219 | } | |||
| 5220 | ||||
| 5221 | static void _ccv_cnnp_debug_deinit(ccv_cnnp_model_t* const super) | |||
| 5222 | { | |||
| 5223 | const ccv_cnnp_model_debug_t* const self = (const ccv_cnnp_model_debug_t*)super; | |||
| 5224 | if (self->debug_deinit && self->debug_context) | |||
| 5225 | self->debug_deinit(self->debug_context); | |||
| 5226 | } | |||
| 5227 | ||||
| 5228 | static ccv_cnnp_model_t* _ccv_cnnp_debug_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 5229 | ||||
| 5230 | static const ccv_cnnp_model_vtab_t ccv_cnnp_debug_isa = { | |||
| 5231 | .build = _ccv_cnnp_debug_build, | |||
| 5232 | .deinit = _ccv_cnnp_debug_deinit, | |||
| 5233 | .copy = _ccv_cnnp_debug_copy, | |||
| 5234 | }; | |||
| 5235 | ||||
| 5236 | 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) | |||
| 5237 | { | |||
| 5238 | ccv_cnnp_model_debug_t* const model_debug = (ccv_cnnp_model_debug_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_debug_t)); | |||
| 5239 | model_debug->super.isa = &ccv_cnnp_debug_isa; | |||
| 5240 | model_debug->super.input_size = 0; | |||
| 5241 | model_debug->super.outputs = &model_debug->output; | |||
| 5242 | model_debug->super.output_size = 1; | |||
| 5243 | model_debug->debugger = func; | |||
| 5244 | model_debug->debug_context = context; | |||
| 5245 | model_debug->debug_deinit = deinit; | |||
| 5246 | model_debug->debug_copy = copy; | |||
| 5247 | ccv_cnnp_model_copy_name(&model_debug->super, name); | |||
| 5248 | return (ccv_cnnp_model_t*)model_debug; | |||
| 5249 | } | |||
| 5250 | ||||
| 5251 | static ccv_cnnp_model_t* _ccv_cnnp_debug_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5252 | { | |||
| 5253 | const ccv_cnnp_model_debug_t* const self = (const ccv_cnnp_model_debug_t*)super; | |||
| 5254 | void* debug_context = self->debug_context; | |||
| 5255 | if (self->debug_copy && self->debug_context) | |||
| 5256 | debug_context = self->debug_copy(self->debug_context); | |||
| 5257 | return ccv_cnnp_debug(self->debugger, debug_context, self->debug_deinit, self->debug_copy, self->super.name); | |||
| 5258 | } | |||
| 5259 | ||||
| 5260 | /// MARK - Sort layer. | |||
| 5261 | ||||
| 5262 | typedef struct { | |||
| 5263 | ccv_cnnp_model_t super; | |||
| 5264 | ccv_nnc_tensor_symbol_t outputs[2]; | |||
| 5265 | int along_axis; | |||
| 5266 | int descending; | |||
| 5267 | } ccv_cnnp_model_sort_t; | |||
| 5268 | ||||
| 5269 | static void _ccv_cnnp_sort_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) | |||
| 5270 | { | |||
| 5271 | ccv_cnnp_model_sort_t* const self = (ccv_cnnp_model_sort_t*)super; | |||
| 5272 | PRINT(CCV_CLI_VERBOSE, "[cnnp_sort_build] - along_axis: %d, descending: %d\n", self->along_axis, self->descending)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_sort_build] - along_axis: %d, descending: %d\n" , self->along_axis, self->descending); fflush(stdout); } } while (0); | |||
| 5273 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 5274 | assert(output_size == 2)((void) sizeof ((output_size == 2) ? 1 : 0), __extension__ ({ if (output_size == 2) ; else __assert_fail ("output_size == 2" , "ccv_cnnp_model_addons.c", 5274, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5275 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 5276 | params.datatype = CCV_32S; | |||
| 5277 | outputs[1] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 5278 | ccv_nnc_graph_exec_symbol_new(graph, CMD_SORT_FORWARD(self->along_axis, self->descending)ccv_nnc_cmd(CCV_NNC_SORT_FORWARD, 0, ((ccv_nnc_cmd_param_t){. size={.dim={1,1,1}},.sort={.along_axis=self->along_axis,.descending =self->descending}}), 0), inputs, input_size, outputs, output_size, "sort"); | |||
| 5279 | } | |||
| 5280 | ||||
| 5281 | static ccv_cnnp_model_t* _ccv_cnnp_sort_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 5282 | ||||
| 5283 | static const ccv_cnnp_model_vtab_t ccv_cnnp_sort_isa = { | |||
| 5284 | .build = _ccv_cnnp_sort_build, | |||
| 5285 | .copy = _ccv_cnnp_sort_copy, | |||
| 5286 | }; | |||
| 5287 | ||||
| 5288 | ccv_cnnp_model_t* ccv_cnnp_sort(const int along_axis, const int descending, const char* const name) | |||
| 5289 | { | |||
| 5290 | ccv_cnnp_model_sort_t* const model_sort = (ccv_cnnp_model_sort_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_sort_t)); | |||
| 5291 | model_sort->super.isa = &ccv_cnnp_sort_isa; | |||
| 5292 | model_sort->super.input_size = 0; | |||
| 5293 | model_sort->super.outputs = model_sort->outputs; | |||
| 5294 | model_sort->super.output_size = 2; | |||
| 5295 | model_sort->along_axis = along_axis; | |||
| 5296 | model_sort->descending = descending; | |||
| 5297 | ccv_cnnp_model_copy_name(&model_sort->super, name); | |||
| 5298 | return (ccv_cnnp_model_t*)model_sort; | |||
| 5299 | } | |||
| 5300 | ||||
| 5301 | static ccv_cnnp_model_t* _ccv_cnnp_sort_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5302 | { | |||
| 5303 | ccv_cnnp_model_sort_t* const self = (ccv_cnnp_model_sort_t*)super; | |||
| 5304 | return ccv_cnnp_sort(self->along_axis, self->descending, self->super.name); | |||
| 5305 | } | |||
| 5306 | ||||
| 5307 | /// MARK - Partition layer. | |||
| 5308 | ||||
| 5309 | typedef struct { | |||
| 5310 | ccv_cnnp_model_t super; | |||
| 5311 | ccv_nnc_tensor_symbol_t outputs[2]; | |||
| 5312 | int kth; | |||
| 5313 | int along_axis; | |||
| 5314 | int descending; | |||
| 5315 | } ccv_cnnp_model_partition_t; | |||
| 5316 | ||||
| 5317 | static void _ccv_cnnp_partition_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) | |||
| 5318 | { | |||
| 5319 | ccv_cnnp_model_partition_t* const self = (ccv_cnnp_model_partition_t*)super; | |||
| 5320 | PRINT(CCV_CLI_VERBOSE, "[cnnp_partition_build] - kth: %d, along_axis: %d, descending: %d\n", self->kth, self->along_axis, self->descending)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_partition_build] - kth: %d, along_axis: %d, descending: %d\n" , self->kth, self->along_axis, self->descending); fflush (stdout); } } while (0); | |||
| 5321 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 5322 | assert(output_size == 2)((void) sizeof ((output_size == 2) ? 1 : 0), __extension__ ({ if (output_size == 2) ; else __assert_fail ("output_size == 2" , "ccv_cnnp_model_addons.c", 5322, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5323 | if (self->kth > 0) | |||
| 5324 | params.dim[self->along_axis] = self->kth; | |||
| 5325 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 5326 | params.datatype = CCV_32S; | |||
| 5327 | outputs[1] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 5328 | ccv_nnc_graph_exec_symbol_new(graph, CMD_PARTITION_FORWARD(self->kth, self->along_axis, self->descending)ccv_nnc_cmd(CCV_NNC_PARTITION_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.partition={.kth=self->kth,.along_axis =self->along_axis,.descending=self->descending}}), 0), inputs, input_size, outputs, output_size, "partition"); | |||
| 5329 | } | |||
| 5330 | ||||
| 5331 | static ccv_cnnp_model_t* _ccv_cnnp_partition_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 5332 | ||||
| 5333 | static const ccv_cnnp_model_vtab_t ccv_cnnp_partition_isa = { | |||
| 5334 | .build = _ccv_cnnp_partition_build, | |||
| 5335 | .copy = _ccv_cnnp_partition_copy, | |||
| 5336 | }; | |||
| 5337 | ||||
| 5338 | ccv_cnnp_model_t* ccv_cnnp_partition(const int kth, const int along_axis, const int descending, const char* const name) | |||
| 5339 | { | |||
| 5340 | ccv_cnnp_model_partition_t* const model_partition = (ccv_cnnp_model_partition_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_partition_t)); | |||
| 5341 | model_partition->super.isa = &ccv_cnnp_partition_isa; | |||
| 5342 | model_partition->super.input_size = 0; | |||
| 5343 | model_partition->super.outputs = model_partition->outputs; | |||
| 5344 | model_partition->super.output_size = 2; | |||
| 5345 | model_partition->kth = kth; | |||
| 5346 | model_partition->along_axis = along_axis; | |||
| 5347 | model_partition->descending = descending; | |||
| 5348 | ccv_cnnp_model_copy_name(&model_partition->super, name); | |||
| 5349 | return (ccv_cnnp_model_t*)model_partition; | |||
| 5350 | } | |||
| 5351 | ||||
| 5352 | static ccv_cnnp_model_t* _ccv_cnnp_partition_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5353 | { | |||
| 5354 | ccv_cnnp_model_partition_t* const self = (ccv_cnnp_model_partition_t*)super; | |||
| 5355 | return ccv_cnnp_partition(self->kth, self->along_axis, self->descending, self->super.name); | |||
| 5356 | } | |||
| 5357 | ||||
| 5358 | /// MARK - Unique consecutive layer. | |||
| 5359 | ||||
| 5360 | typedef struct { | |||
| 5361 | ccv_cnnp_model_t super; | |||
| 5362 | ccv_nnc_tensor_symbol_t outputs[2]; | |||
| 5363 | int bincount; | |||
| 5364 | } ccv_cnnp_model_unique_consecutive_t; | |||
| 5365 | ||||
| 5366 | static void _ccv_cnnp_unique_consecutive_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) | |||
| 5367 | { | |||
| 5368 | ccv_cnnp_model_unique_consecutive_t* const self = (ccv_cnnp_model_unique_consecutive_t*)super; | |||
| 5369 | PRINT(CCV_CLI_VERBOSE, "[cnnp_unique_consecutive_build] - bincount: %d\n", self->bincount)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_unique_consecutive_build] - bincount: %d\n", self->bincount); fflush(stdout); } } while (0); | |||
| 5370 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 5371 | assert(output_size == 2)((void) sizeof ((output_size == 2) ? 1 : 0), __extension__ ({ if (output_size == 2) ; else __assert_fail ("output_size == 2" , "ccv_cnnp_model_addons.c", 5371, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5372 | if (self->bincount > 0) | |||
| 5373 | params.dim[0] = ccv_min(params.dim[0], self->bincount)({ typeof (params.dim[0]) _a = (params.dim[0]); typeof (self-> bincount) _b = (self->bincount); (_a < _b) ? _a : _b; } ); | |||
| 5374 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 5375 | params.datatype = CCV_32S; | |||
| 5376 | outputs[1] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 5377 | ccv_nnc_graph_exec_symbol_new(graph, CMD_UNIQUE_CONSECUTIVE_FORWARD(self->bincount)ccv_nnc_cmd(CCV_NNC_UNIQUE_CONSECUTIVE_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.unique_consecutive={.bincount=self-> bincount}}), 0), inputs, input_size, outputs, output_size, "unique_consecutive"); | |||
| 5378 | } | |||
| 5379 | ||||
| 5380 | static ccv_cnnp_model_t* _ccv_cnnp_unique_consecutive_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 5381 | ||||
| 5382 | static const ccv_cnnp_model_vtab_t ccv_cnnp_unique_consecutive_isa = { | |||
| 5383 | .build = _ccv_cnnp_unique_consecutive_build, | |||
| 5384 | .copy = _ccv_cnnp_unique_consecutive_copy, | |||
| 5385 | }; | |||
| 5386 | ||||
| 5387 | ccv_cnnp_model_t* ccv_cnnp_unique_consecutive(const int bincount, const char* const name) | |||
| 5388 | { | |||
| 5389 | ccv_cnnp_model_unique_consecutive_t* const model_unique_consecutive = (ccv_cnnp_model_unique_consecutive_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_unique_consecutive_t)); | |||
| 5390 | model_unique_consecutive->super.isa = &ccv_cnnp_unique_consecutive_isa; | |||
| 5391 | model_unique_consecutive->super.input_size = 0; | |||
| 5392 | model_unique_consecutive->super.outputs = model_unique_consecutive->outputs; | |||
| 5393 | model_unique_consecutive->super.output_size = 2; | |||
| 5394 | model_unique_consecutive->bincount = bincount; | |||
| 5395 | ccv_cnnp_model_copy_name(&model_unique_consecutive->super, name); | |||
| 5396 | return (ccv_cnnp_model_t*)model_unique_consecutive; | |||
| 5397 | } | |||
| 5398 | ||||
| 5399 | static ccv_cnnp_model_t* _ccv_cnnp_unique_consecutive_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5400 | { | |||
| 5401 | ccv_cnnp_model_unique_consecutive_t* const self = (ccv_cnnp_model_unique_consecutive_t*)super; | |||
| 5402 | return ccv_cnnp_unique_consecutive(self->bincount, self->super.name); | |||
| 5403 | } | |||
| 5404 | ||||
| 5405 | /// MARK - Scatter add layer. | |||
| 5406 | ||||
| 5407 | typedef struct { | |||
| 5408 | ccv_cnnp_model_t super; | |||
| 5409 | ccv_nnc_tensor_symbol_t output; | |||
| 5410 | int bincount; | |||
| 5411 | int count_per_output; | |||
| 5412 | } ccv_cnnp_model_scatter_add_t; | |||
| 5413 | ||||
| 5414 | static void _ccv_cnnp_scatter_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) | |||
| 5415 | { | |||
| 5416 | ccv_cnnp_model_scatter_add_t* const self = (ccv_cnnp_model_scatter_add_t*)super; | |||
| 5417 | PRINT(CCV_CLI_VERBOSE, "[cnnp_scatter_add_build] - bincount: %d, count_per_output: %d\n", self->bincount, self->count_per_output)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_scatter_add_build] - bincount: %d, count_per_output: %d\n" , self->bincount, self->count_per_output); fflush(stdout ); } } while (0); | |||
| 5418 | ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 5419 | 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", 5419, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5420 | assert(self->bincount > 0)((void) sizeof ((self->bincount > 0) ? 1 : 0), __extension__ ({ if (self->bincount > 0) ; else __assert_fail ("self->bincount > 0" , "ccv_cnnp_model_addons.c", 5420, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5421 | params.dim[0] = self->bincount; | |||
| 5422 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, params, 0); | |||
| 5423 | ccv_nnc_graph_exec_symbol_new(graph, CMD_SCATTER_ADD_FORWARD(self->bincount, self->count_per_output)ccv_nnc_cmd(CCV_NNC_SCATTER_ADD_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.scatter_add={.bincount=self->bincount ,.count_per_output=self->count_per_output}}), 0), inputs, input_size, outputs, output_size, "scatter_add"); | |||
| 5424 | } | |||
| 5425 | ||||
| 5426 | static ccv_cnnp_model_t* _ccv_cnnp_scatter_add_copy(const ccv_cnnp_model_t* const self, void* const context); | |||
| 5427 | ||||
| 5428 | static const ccv_cnnp_model_vtab_t ccv_cnnp_scatter_add_isa = { | |||
| 5429 | .build = _ccv_cnnp_scatter_add_build, | |||
| 5430 | .copy = _ccv_cnnp_scatter_add_copy, | |||
| 5431 | }; | |||
| 5432 | ||||
| 5433 | ccv_cnnp_model_t* ccv_cnnp_scatter_add(const int bincount, const int count_per_output, const char* const name) | |||
| 5434 | { | |||
| 5435 | assert(bincount > 0)((void) sizeof ((bincount > 0) ? 1 : 0), __extension__ ({ if (bincount > 0) ; else __assert_fail ("bincount > 0", "ccv_cnnp_model_addons.c" , 5435, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5436 | assert(count_per_output >= 0)((void) sizeof ((count_per_output >= 0) ? 1 : 0), __extension__ ({ if (count_per_output >= 0) ; else __assert_fail ("count_per_output >= 0" , "ccv_cnnp_model_addons.c", 5436, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5437 | ccv_cnnp_model_scatter_add_t* const model_scatter_add = (ccv_cnnp_model_scatter_add_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_scatter_add_t)); | |||
| 5438 | model_scatter_add->super.isa = &ccv_cnnp_scatter_add_isa; | |||
| 5439 | model_scatter_add->super.input_size = 0; | |||
| 5440 | model_scatter_add->super.outputs = &model_scatter_add->output; | |||
| 5441 | model_scatter_add->super.output_size = 1; | |||
| 5442 | model_scatter_add->bincount = bincount; | |||
| 5443 | model_scatter_add->count_per_output = count_per_output; | |||
| 5444 | ccv_cnnp_model_copy_name(&model_scatter_add->super, name); | |||
| 5445 | return (ccv_cnnp_model_t*)model_scatter_add; | |||
| 5446 | } | |||
| 5447 | ||||
| 5448 | static ccv_cnnp_model_t* _ccv_cnnp_scatter_add_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5449 | { | |||
| 5450 | ccv_cnnp_model_scatter_add_t* const self = (ccv_cnnp_model_scatter_add_t*)super; | |||
| 5451 | return ccv_cnnp_scatter_add(self->bincount, self->count_per_output, self->super.name); | |||
| 5452 | } | |||
| 5453 | ||||
| 5454 | // MARK - Segmented Dense Layer | |||
| 5455 | ||||
| 5456 | typedef struct { | |||
| 5457 | ccv_cnnp_model_t super; | |||
| 5458 | ccv_nnc_tensor_symbol_t output; | |||
| 5459 | ccv_nnc_tensor_symbol_t weights; | |||
| 5460 | ccv_nnc_tensor_symbol_t bias; | |||
| 5461 | int segments; | |||
| 5462 | int count; | |||
| 5463 | int no_bias; | |||
| 5464 | int flags; | |||
| 5465 | int functional; | |||
| 5466 | } ccv_cnnp_model_segmented_dense_t; | |||
| 5467 | ||||
| 5468 | static void _ccv_cnnp_segmented_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) | |||
| 5469 | { | |||
| 5470 | ccv_cnnp_model_segmented_dense_t* const self = (ccv_cnnp_model_segmented_dense_t*)super; | |||
| 5471 | PRINT(CCV_CLI_VERBOSE, "[cnnp_segmented_dense_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_segmented_dense_build] -\n"); fflush(stdout) ; } } while (0); | |||
| 5472 | assert(input_size == (self->functional ? 4 + !self->no_bias : 3))((void) sizeof ((input_size == (self->functional ? 4 + !self ->no_bias : 3)) ? 1 : 0), __extension__ ({ if (input_size == (self->functional ? 4 + !self->no_bias : 3)) ; else __assert_fail ("input_size == (self->functional ? 4 + !self->no_bias : 3)" , "ccv_cnnp_model_addons.c", 5472, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5473 | 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", 5473, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5474 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 5475 | const ccv_nnc_tensor_param_t indices_params = ccv_nnc_tensor_symbol_params(graph, inputs[1]); | |||
| 5476 | const ccv_nnc_tensor_param_t counts_params = ccv_nnc_tensor_symbol_params(graph, inputs[2]); | |||
| 5477 | ccv_nnc_tensor_param_t weights_params; | |||
| 5478 | ccv_nnc_tensor_symbol_t weights; | |||
| 5479 | if (self->functional) | |||
| 5480 | { | |||
| 5481 | weights = inputs[3]; | |||
| 5482 | weights_params = ccv_nnc_tensor_symbol_params(graph, weights); | |||
| 5483 | } else { | |||
| 5484 | weights_params = params; | |||
| 5485 | memset(weights_params.dim, 0, sizeof(weights_params.dim)); | |||
| 5486 | weights_params.dim[0] = self->segments; | |||
| 5487 | weights_params.dim[1] = self->count; | |||
| 5488 | weights_params.dim[2] = params.dim[ccv_nnc_tensor_nd(params.dim) - 1]; | |||
| 5489 | if (!self->weights.graph) | |||
| 5490 | self->weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "weights"); | |||
| 5491 | 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" , 5491, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5492 | weights = ccv_cnnp_model_get_symbol(super, self->weights); | |||
| 5493 | } | |||
| 5494 | ccv_nnc_tensor_param_t bias_params = {}; | |||
| 5495 | ccv_nnc_tensor_symbol_t bias = {}; | |||
| 5496 | if (!self->no_bias) | |||
| 5497 | { | |||
| 5498 | if (self->functional) | |||
| 5499 | { | |||
| 5500 | bias = inputs[4]; | |||
| 5501 | bias_params = ccv_nnc_tensor_symbol_params(graph, bias); | |||
| 5502 | } else { | |||
| 5503 | bias_params = params; | |||
| 5504 | memset(bias_params.dim, 0, sizeof(bias_params.dim)); | |||
| 5505 | bias_params.dim[0] = self->segments; | |||
| 5506 | bias_params.dim[1] = self->count; | |||
| 5507 | if (!self->bias.graph) | |||
| 5508 | self->bias = ccv_nnc_tensor_symbol_new(graph, bias_params, "bias"); | |||
| 5509 | bias = ccv_cnnp_model_get_symbol(super, self->bias); | |||
| 5510 | } | |||
| 5511 | } | |||
| 5512 | ccv_nnc_cmd_t cmd = {0}; | |||
| 5513 | cmd.cmd = CCV_NNC_SEGMENTED_GEMM_FORWARD; | |||
| 5514 | cmd.info.blas.a[0] = 1; | |||
| 5515 | cmd.info.blas.a[1] = 1; | |||
| 5516 | cmd.info.blas.transpose_b[0] = 1; | |||
| 5517 | cmd.info.blas.transpose_b[1] = 2; | |||
| 5518 | cmd.info.blas.flags = self->flags; | |||
| 5519 | ccv_nnc_tensor_param_t output_params; | |||
| 5520 | ccv_nnc_hint_tensor_auto(cmd, (ccv_nnc_tensor_param_t []){ | |||
| 5521 | params, indices_params, counts_params, | |||
| 5522 | weights_params, | |||
| 5523 | bias_params, | |||
| 5524 | }, 4 + !self->no_bias, ccv_nnc_no_hint, &output_params, 1); | |||
| 5525 | const ccv_nnc_tensor_symbol_t output = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 5526 | if (self->no_bias) | |||
| 5527 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], inputs[2], weights)(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], inputs [2], weights}, (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), 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), "segmented_dense"); | |||
| 5528 | else | |||
| 5529 | ccv_nnc_graph_exec_symbol_new(graph, cmd, TENSOR_SYMBOL_LIST(inputs[0], inputs[1], inputs[2], weights, bias)(const ccv_nnc_tensor_symbol_t []){inputs[0], inputs[1], inputs [2], weights, bias}, (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)(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), "segmented_dense"); | |||
| 5530 | outputs[0] = output; | |||
| 5531 | } | |||
| 5532 | ||||
| 5533 | static void _ccv_cnnp_segmented_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) | |||
| 5534 | { | |||
| 5535 | ccv_cnnp_model_segmented_dense_t* const self = (ccv_cnnp_model_segmented_dense_t*)super; | |||
| 5536 | if (self->functional) | |||
| 5537 | return; | |||
| 5538 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->weights); | |||
| 5539 | const int c = weight_params.dim[1]; | |||
| 5540 | const float std = sqrtf(2) / sqrtf(c); | |||
| 5541 | const float bound = sqrtf(3) * std; | |||
| 5542 | 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); | |||
| 5543 | if (self->bias.graph) | |||
| 5544 | 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); | |||
| 5545 | } | |||
| 5546 | ||||
| 5547 | static void _ccv_cnnp_segmented_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) | |||
| 5548 | { | |||
| 5549 | ccv_cnnp_model_segmented_dense_t* const self = (ccv_cnnp_model_segmented_dense_t*)super; | |||
| 5550 | if (self->functional) | |||
| 5551 | return; | |||
| 5552 | add_to_array(parameters, self->weights, is_trainable); | |||
| 5553 | if (self->bias.graph) | |||
| 5554 | add_to_array(parameters, self->bias, is_trainable); | |||
| 5555 | } | |||
| 5556 | ||||
| 5557 | static ccv_cnnp_model_t* _ccv_cnnp_segmented_dense_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 5558 | ||||
| 5559 | static const ccv_cnnp_model_vtab_t ccv_cnnp_segmented_dense_isa = { | |||
| 5560 | .build = _ccv_cnnp_segmented_dense_build, | |||
| 5561 | .init_states = _ccv_cnnp_segmented_dense_init_states, | |||
| 5562 | .add_to_parameter = _ccv_cnnp_segmented_dense_add_to_parameter, | |||
| 5563 | .copy = _ccv_cnnp_segmented_dense_copy, | |||
| 5564 | }; | |||
| 5565 | ||||
| 5566 | ccv_cnnp_model_t* ccv_cnnp_segmented_dense(const int segments, const int count, const int no_bias, const int flags, const int functional, const int is_trainable, const char* const name) | |||
| 5567 | { | |||
| 5568 | ccv_cnnp_model_segmented_dense_t* const model_segmented_dense = (ccv_cnnp_model_segmented_dense_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_segmented_dense_t)); | |||
| 5569 | model_segmented_dense->super.isa = &ccv_cnnp_segmented_dense_isa; | |||
| 5570 | model_segmented_dense->super.input_size = functional ? 4 + !no_bias : 3; | |||
| 5571 | model_segmented_dense->super.outputs = &model_segmented_dense->output; | |||
| 5572 | model_segmented_dense->super.output_size = 1; | |||
| 5573 | model_segmented_dense->super.is_trainable = is_trainable; | |||
| 5574 | ccv_cnnp_model_copy_name(&model_segmented_dense->super, name); | |||
| 5575 | model_segmented_dense->weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 5576 | model_segmented_dense->weights.graph = 0; | |||
| 5577 | model_segmented_dense->bias.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 5578 | model_segmented_dense->bias.graph = 0; | |||
| 5579 | model_segmented_dense->segments = segments; | |||
| 5580 | model_segmented_dense->count = count; | |||
| 5581 | model_segmented_dense->no_bias = no_bias; | |||
| 5582 | model_segmented_dense->flags = flags; | |||
| 5583 | model_segmented_dense->functional = functional; | |||
| 5584 | return (ccv_cnnp_model_t*)model_segmented_dense; | |||
| 5585 | } | |||
| 5586 | ||||
| 5587 | static ccv_cnnp_model_t* _ccv_cnnp_segmented_dense_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5588 | { | |||
| 5589 | const ccv_cnnp_model_segmented_dense_t* const self = (const ccv_cnnp_model_segmented_dense_t*)super; | |||
| 5590 | return ccv_cnnp_segmented_dense(self->segments, self->count, self->no_bias, self->flags, self->functional, self->super.is_trainable, self->super.name); | |||
| 5591 | } | |||
| 5592 | ||||
| 5593 | // MARK - SwiGLU Layer | |||
| 5594 | ||||
| 5595 | typedef struct { | |||
| 5596 | ccv_cnnp_model_t super; | |||
| 5597 | ccv_nnc_tensor_symbol_t output; | |||
| 5598 | ccv_nnc_tensor_symbol_t gate_weights; | |||
| 5599 | ccv_nnc_tensor_symbol_t up_weights; | |||
| 5600 | int count; | |||
| 5601 | float clamp; | |||
| 5602 | } ccv_cnnp_model_swiglu_t; | |||
| 5603 | ||||
| 5604 | static void _ccv_cnnp_swiglu_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) | |||
| 5605 | { | |||
| 5606 | ccv_cnnp_model_swiglu_t* const self = (ccv_cnnp_model_swiglu_t*)super; | |||
| 5607 | PRINT(CCV_CLI_VERBOSE, "[cnnp_swiglu_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_swiglu_build] -\n"); fflush(stdout); } } while (0); | |||
| 5608 | 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" , 5608, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5609 | 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", 5609, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5610 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 5611 | const int params_nd = ccv_nnc_tensor_nd(params.dim); | |||
| 5612 | assert(params_nd >= 1)((void) sizeof ((params_nd >= 1) ? 1 : 0), __extension__ ( { if (params_nd >= 1) ; else __assert_fail ("params_nd >= 1" , "ccv_cnnp_model_addons.c", 5612, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5613 | ccv_nnc_tensor_param_t weights_params = params; | |||
| 5614 | memset(weights_params.dim, 0, sizeof(weights_params.dim)); | |||
| 5615 | weights_params.dim[0] = self->count; | |||
| 5616 | weights_params.dim[1] = params.dim[params_nd - 1]; | |||
| 5617 | if (!self->gate_weights.graph) | |||
| 5618 | self->gate_weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "gate.weight"); | |||
| 5619 | if (!self->up_weights.graph) | |||
| 5620 | self->up_weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "up.weight"); | |||
| 5621 | assert(self->gate_weights.graph == graph)((void) sizeof ((self->gate_weights.graph == graph) ? 1 : 0 ), __extension__ ({ if (self->gate_weights.graph == graph) ; else __assert_fail ("self->gate_weights.graph == graph" , "ccv_cnnp_model_addons.c", 5621, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5622 | assert(self->up_weights.graph == graph)((void) sizeof ((self->up_weights.graph == graph) ? 1 : 0) , __extension__ ({ if (self->up_weights.graph == graph) ; else __assert_fail ("self->up_weights.graph == graph", "ccv_cnnp_model_addons.c" , 5622, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5623 | const ccv_nnc_tensor_symbol_t gate_weights = ccv_cnnp_model_get_symbol(super, self->gate_weights); | |||
| 5624 | const ccv_nnc_tensor_symbol_t up_weights = ccv_cnnp_model_get_symbol(super, self->up_weights); | |||
| 5625 | const ccv_nnc_cmd_t cmd = CMD_SWIGLU_FORWARD(self->clamp)ccv_nnc_cmd(CCV_NNC_SWIGLU_FORWARD, 0, ((ccv_nnc_cmd_param_t) {.size={.dim={1,1,1}},.swiglu={.clamp=self->clamp}}), 0); | |||
| 5626 | const ccv_nnc_tensor_symbol_t command_inputs[] = { | |||
| 5627 | inputs[0], gate_weights, up_weights, | |||
| 5628 | }; | |||
| 5629 | ccv_nnc_tensor_param_t command_input_params[3]; | |||
| 5630 | int i; | |||
| 5631 | for (i = 0; i < 3; i++) | |||
| 5632 | command_input_params[i] = ccv_nnc_tensor_symbol_params(graph, command_inputs[i]); | |||
| 5633 | ccv_nnc_tensor_param_t output_params; | |||
| 5634 | ccv_nnc_hint_tensor_auto(cmd, command_input_params, 3, ccv_nnc_no_hint, &output_params, 1); | |||
| 5635 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 5636 | ccv_nnc_graph_exec_symbol_new(graph, cmd, command_inputs, 3, outputs, 1, "swiglu"); | |||
| 5637 | } | |||
| 5638 | ||||
| 5639 | static void _ccv_cnnp_swiglu_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) | |||
| 5640 | { | |||
| 5641 | ccv_cnnp_model_swiglu_t* const self = (ccv_cnnp_model_swiglu_t*)super; | |||
| 5642 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->gate_weights); | |||
| 5643 | const float bound = sqrtf(6.f / weight_params.dim[1]); | |||
| 5644 | 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->gate_weights); | |||
| 5645 | 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->up_weights); | |||
| 5646 | } | |||
| 5647 | ||||
| 5648 | static void _ccv_cnnp_swiglu_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) | |||
| 5649 | { | |||
| 5650 | ccv_cnnp_model_swiglu_t* const self = (ccv_cnnp_model_swiglu_t*)super; | |||
| 5651 | add_to_array(parameters, self->gate_weights, is_trainable); | |||
| 5652 | add_to_array(parameters, self->up_weights, is_trainable); | |||
| 5653 | } | |||
| 5654 | ||||
| 5655 | static ccv_cnnp_model_t* _ccv_cnnp_swiglu_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 5656 | ||||
| 5657 | static const ccv_cnnp_model_vtab_t ccv_cnnp_swiglu_isa = { | |||
| 5658 | .build = _ccv_cnnp_swiglu_build, | |||
| 5659 | .init_states = _ccv_cnnp_swiglu_init_states, | |||
| 5660 | .add_to_parameter = _ccv_cnnp_swiglu_add_to_parameter, | |||
| 5661 | .copy = _ccv_cnnp_swiglu_copy, | |||
| 5662 | }; | |||
| 5663 | ||||
| 5664 | ccv_cnnp_model_t* ccv_cnnp_swiglu(const int count, const float clamp, const int is_trainable, const char* const name) | |||
| 5665 | { | |||
| 5666 | ccv_cnnp_model_swiglu_t* const model = (ccv_cnnp_model_swiglu_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_swiglu_t)); | |||
| 5667 | model->super.isa = &ccv_cnnp_swiglu_isa; | |||
| 5668 | model->super.input_size = 1; | |||
| 5669 | model->super.outputs = &model->output; | |||
| 5670 | model->super.output_size = 1; | |||
| 5671 | model->super.is_trainable = is_trainable; | |||
| 5672 | ccv_cnnp_model_copy_name(&model->super, name); | |||
| 5673 | model->gate_weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 5674 | model->up_weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 5675 | model->count = count; | |||
| 5676 | model->clamp = clamp; | |||
| 5677 | return (ccv_cnnp_model_t*)model; | |||
| 5678 | } | |||
| 5679 | ||||
| 5680 | static ccv_cnnp_model_t* _ccv_cnnp_swiglu_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5681 | { | |||
| 5682 | const ccv_cnnp_model_swiglu_t* const self = (const ccv_cnnp_model_swiglu_t*)super; | |||
| 5683 | return ccv_cnnp_swiglu(self->count, self->clamp, self->super.is_trainable, self->super.name); | |||
| 5684 | } | |||
| 5685 | ||||
| 5686 | // MARK - Segmented SwiGLU Layer | |||
| 5687 | ||||
| 5688 | typedef struct { | |||
| 5689 | ccv_cnnp_model_t super; | |||
| 5690 | ccv_nnc_tensor_symbol_t output; | |||
| 5691 | ccv_nnc_tensor_symbol_t gate_weights; | |||
| 5692 | ccv_nnc_tensor_symbol_t up_weights; | |||
| 5693 | int segments; | |||
| 5694 | int count; | |||
| 5695 | float clamp; | |||
| 5696 | int functional; | |||
| 5697 | } ccv_cnnp_model_segmented_swiglu_t; | |||
| 5698 | ||||
| 5699 | static void _ccv_cnnp_segmented_swiglu_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) | |||
| 5700 | { | |||
| 5701 | ccv_cnnp_model_segmented_swiglu_t* const self = (ccv_cnnp_model_segmented_swiglu_t*)super; | |||
| 5702 | PRINT(CCV_CLI_VERBOSE, "[cnnp_segmented_swiglu_build] -\n")do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_segmented_swiglu_build] -\n"); fflush(stdout ); } } while (0); | |||
| 5703 | assert(input_size == (self->functional ? 6 : 4))((void) sizeof ((input_size == (self->functional ? 6 : 4)) ? 1 : 0), __extension__ ({ if (input_size == (self->functional ? 6 : 4)) ; else __assert_fail ("input_size == (self->functional ? 6 : 4)" , "ccv_cnnp_model_addons.c", 5703, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5704 | 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", 5704, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5705 | const ccv_nnc_tensor_param_t params = ccv_nnc_tensor_symbol_params(graph, inputs[0]); | |||
| 5706 | ccv_nnc_tensor_symbol_t gate_weights; | |||
| 5707 | ccv_nnc_tensor_symbol_t up_weights; | |||
| 5708 | if (self->functional) | |||
| 5709 | { | |||
| 5710 | gate_weights = inputs[3]; | |||
| 5711 | up_weights = inputs[4]; | |||
| 5712 | } else { | |||
| 5713 | const int params_nd = ccv_nnc_tensor_nd(params.dim); | |||
| 5714 | assert(params_nd >= 1)((void) sizeof ((params_nd >= 1) ? 1 : 0), __extension__ ( { if (params_nd >= 1) ; else __assert_fail ("params_nd >= 1" , "ccv_cnnp_model_addons.c", 5714, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5715 | ccv_nnc_tensor_param_t weights_params = params; | |||
| 5716 | memset(weights_params.dim, 0, sizeof(weights_params.dim)); | |||
| 5717 | weights_params.dim[0] = self->segments; | |||
| 5718 | weights_params.dim[1] = self->count; | |||
| 5719 | weights_params.dim[2] = params.dim[params_nd - 1]; | |||
| 5720 | if (!self->gate_weights.graph) | |||
| 5721 | self->gate_weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "gate.weight"); | |||
| 5722 | if (!self->up_weights.graph) | |||
| 5723 | self->up_weights = ccv_nnc_tensor_symbol_new(graph, weights_params, "up.weight"); | |||
| 5724 | assert(self->gate_weights.graph == graph)((void) sizeof ((self->gate_weights.graph == graph) ? 1 : 0 ), __extension__ ({ if (self->gate_weights.graph == graph) ; else __assert_fail ("self->gate_weights.graph == graph" , "ccv_cnnp_model_addons.c", 5724, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5725 | assert(self->up_weights.graph == graph)((void) sizeof ((self->up_weights.graph == graph) ? 1 : 0) , __extension__ ({ if (self->up_weights.graph == graph) ; else __assert_fail ("self->up_weights.graph == graph", "ccv_cnnp_model_addons.c" , 5725, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5726 | gate_weights = ccv_cnnp_model_get_symbol(super, self->gate_weights); | |||
| 5727 | up_weights = ccv_cnnp_model_get_symbol(super, self->up_weights); | |||
| 5728 | } | |||
| 5729 | const ccv_nnc_cmd_t cmd = CMD_SEGMENTED_SWIGLU_FORWARD(self->clamp)ccv_nnc_cmd(CCV_NNC_SEGMENTED_SWIGLU_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.segmented_swiglu={.clamp=self->clamp }}), 0); | |||
| 5730 | const ccv_nnc_tensor_symbol_t command_inputs[] = { | |||
| 5731 | inputs[0], inputs[1], inputs[2], gate_weights, up_weights, inputs[self->functional ? 5 : 3], | |||
| 5732 | }; | |||
| 5733 | ccv_nnc_tensor_param_t command_input_params[6]; | |||
| 5734 | int i; | |||
| 5735 | for (i = 0; i < 6; i++) | |||
| 5736 | command_input_params[i] = ccv_nnc_tensor_symbol_params(graph, command_inputs[i]); | |||
| 5737 | ccv_nnc_tensor_param_t output_params; | |||
| 5738 | ccv_nnc_hint_tensor_auto(cmd, command_input_params, 6, ccv_nnc_no_hint, &output_params, 1); | |||
| 5739 | outputs[0] = ccv_nnc_tensor_symbol_new(graph, output_params, 0); | |||
| 5740 | ccv_nnc_graph_exec_symbol_new(graph, cmd, command_inputs, 6, outputs, 1, "segmented_swiglu"); | |||
| 5741 | } | |||
| 5742 | ||||
| 5743 | static void _ccv_cnnp_segmented_swiglu_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) | |||
| 5744 | { | |||
| 5745 | ccv_cnnp_model_segmented_swiglu_t* const self = (ccv_cnnp_model_segmented_swiglu_t*)super; | |||
| 5746 | if (self->functional) | |||
| 5747 | return; | |||
| 5748 | const ccv_nnc_tensor_param_t weight_params = ccv_nnc_tensor_symbol_params(graph, self->gate_weights); | |||
| 5749 | const float bound = sqrtf(6.f / weight_params.dim[2]); | |||
| 5750 | 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->gate_weights); | |||
| 5751 | 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->up_weights); | |||
| 5752 | } | |||
| 5753 | ||||
| 5754 | static void _ccv_cnnp_segmented_swiglu_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) | |||
| 5755 | { | |||
| 5756 | ccv_cnnp_model_segmented_swiglu_t* const self = (ccv_cnnp_model_segmented_swiglu_t*)super; | |||
| 5757 | if (self->functional) | |||
| 5758 | return; | |||
| 5759 | add_to_array(parameters, self->gate_weights, is_trainable); | |||
| 5760 | add_to_array(parameters, self->up_weights, is_trainable); | |||
| 5761 | } | |||
| 5762 | ||||
| 5763 | static ccv_cnnp_model_t* _ccv_cnnp_segmented_swiglu_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 5764 | ||||
| 5765 | static const ccv_cnnp_model_vtab_t ccv_cnnp_segmented_swiglu_isa = { | |||
| 5766 | .build = _ccv_cnnp_segmented_swiglu_build, | |||
| 5767 | .init_states = _ccv_cnnp_segmented_swiglu_init_states, | |||
| 5768 | .add_to_parameter = _ccv_cnnp_segmented_swiglu_add_to_parameter, | |||
| 5769 | .copy = _ccv_cnnp_segmented_swiglu_copy, | |||
| 5770 | }; | |||
| 5771 | ||||
| 5772 | ccv_cnnp_model_t* ccv_cnnp_segmented_swiglu(const int segments, const int count, const float clamp, const int functional, const int is_trainable, const char* const name) | |||
| 5773 | { | |||
| 5774 | ccv_cnnp_model_segmented_swiglu_t* const model = (ccv_cnnp_model_segmented_swiglu_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_segmented_swiglu_t)); | |||
| 5775 | model->super.isa = &ccv_cnnp_segmented_swiglu_isa; | |||
| 5776 | model->super.input_size = functional ? 6 : 4; | |||
| 5777 | model->super.outputs = &model->output; | |||
| 5778 | model->super.output_size = 1; | |||
| 5779 | model->super.is_trainable = is_trainable; | |||
| 5780 | ccv_cnnp_model_copy_name(&model->super, name); | |||
| 5781 | model->gate_weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 5782 | model->up_weights.d = CCV_NNC_NO_TENSOR_SYMBOL; | |||
| 5783 | model->segments = segments; | |||
| 5784 | model->count = count; | |||
| 5785 | model->clamp = clamp; | |||
| 5786 | model->functional = functional; | |||
| 5787 | return (ccv_cnnp_model_t*)model; | |||
| 5788 | } | |||
| 5789 | ||||
| 5790 | static ccv_cnnp_model_t* _ccv_cnnp_segmented_swiglu_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5791 | { | |||
| 5792 | const ccv_cnnp_model_segmented_swiglu_t* const self = (const ccv_cnnp_model_segmented_swiglu_t*)super; | |||
| 5793 | return ccv_cnnp_segmented_swiglu(self->segments, self->count, self->clamp, self->functional, self->super.is_trainable, self->super.name); | |||
| 5794 | } | |||
| 5795 | ||||
| 5796 | // MARK - MoE Weights Streaming | |||
| 5797 | ||||
| 5798 | typedef struct { | |||
| 5799 | ccv_cnnp_model_t super; | |||
| 5800 | ccv_nnc_tensor_symbol_t outputs[6]; | |||
| 5801 | int resident_slots; | |||
| 5802 | int routing_width; | |||
| 5803 | } ccv_cnnp_model_moe_weights_streaming_t; | |||
| 5804 | ||||
| 5805 | static void _ccv_cnnp_moe_weights_streaming_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) | |||
| 5806 | { | |||
| 5807 | ccv_cnnp_model_moe_weights_streaming_t* const self = (ccv_cnnp_model_moe_weights_streaming_t*)super; | |||
| 5808 | PRINT(CCV_CLI_VERBOSE, "[cnnp_moe_weights_streaming_build] - resident_slots: %d, routing_width: %d\n", self->resident_slots, self->routing_width)do { if ((CCV_CLI_VERBOSE & ccv_cli_get_output_levels())) { printf("[cnnp_moe_weights_streaming_build] - resident_slots: %d, routing_width: %d\n" , self->resident_slots, self->routing_width); fflush(stdout ); } } while (0); | |||
| 5809 | assert(input_size == 6)((void) sizeof ((input_size == 6) ? 1 : 0), __extension__ ({ if (input_size == 6) ; else __assert_fail ("input_size == 6", "ccv_cnnp_model_addons.c" , 5809, __extension__ __PRETTY_FUNCTION__); })); | |||
| 5810 | assert(output_size == 6)((void) sizeof ((output_size == 6) ? 1 : 0), __extension__ ({ if (output_size == 6) ; else __assert_fail ("output_size == 6" , "ccv_cnnp_model_addons.c", 5810, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5811 | const ccv_nnc_cmd_t cmd = CMD_MOE_WEIGHTS_STREAMING_FORWARD(self->resident_slots, self->routing_width)ccv_nnc_cmd(CCV_NNC_MOE_WEIGHTS_STREAMING_FORWARD, 0, ((ccv_nnc_cmd_param_t ){.size={.dim={1,1,1}},.moe_weights_streaming={.resident_slots =(self->resident_slots),.routing_width=(self->routing_width )}}), 0); | |||
| 5812 | ccv_nnc_tensor_param_t input_params[6]; | |||
| 5813 | ccv_nnc_tensor_param_t output_params[6]; | |||
| 5814 | int i; | |||
| 5815 | for (i = 0; i < 6; i++) | |||
| 5816 | input_params[i] = ccv_nnc_tensor_symbol_params(graph, inputs[i]); | |||
| 5817 | ccv_nnc_hint_tensor_auto(cmd, input_params, 6, ccv_nnc_no_hint, output_params, 6); | |||
| 5818 | for (i = 0; i < 6; i++) | |||
| 5819 | outputs[i] = ccv_nnc_tensor_symbol_new(graph, output_params[i], 0); | |||
| 5820 | ccv_nnc_graph_exec_symbol_new(graph, cmd, inputs, 6, outputs, 6, "moe_weights_streaming"); | |||
| 5821 | } | |||
| 5822 | ||||
| 5823 | static ccv_cnnp_model_t* _ccv_cnnp_moe_weights_streaming_copy(const ccv_cnnp_model_t* const super, void* const context); | |||
| 5824 | ||||
| 5825 | static const ccv_cnnp_model_vtab_t ccv_cnnp_moe_weights_streaming_isa = { | |||
| 5826 | .build = _ccv_cnnp_moe_weights_streaming_build, | |||
| 5827 | .copy = _ccv_cnnp_moe_weights_streaming_copy, | |||
| 5828 | }; | |||
| 5829 | ||||
| 5830 | ccv_cnnp_model_t* ccv_cnnp_moe_weights_streaming(const int resident_slots, const int routing_width, const char* const name) | |||
| 5831 | { | |||
| 5832 | assert(resident_slots > 0)((void) sizeof ((resident_slots > 0) ? 1 : 0), __extension__ ({ if (resident_slots > 0) ; else __assert_fail ("resident_slots > 0" , "ccv_cnnp_model_addons.c", 5832, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5833 | assert(routing_width > 0)((void) sizeof ((routing_width > 0) ? 1 : 0), __extension__ ({ if (routing_width > 0) ; else __assert_fail ("routing_width > 0" , "ccv_cnnp_model_addons.c", 5833, __extension__ __PRETTY_FUNCTION__ ); })); | |||
| 5834 | ccv_cnnp_model_moe_weights_streaming_t* const model = (ccv_cnnp_model_moe_weights_streaming_t*)cccalloccalloc(1, sizeof(ccv_cnnp_model_moe_weights_streaming_t)); | |||
| 5835 | model->super.isa = &ccv_cnnp_moe_weights_streaming_isa; | |||
| 5836 | model->super.input_size = 6; | |||
| 5837 | model->super.outputs = model->outputs; | |||
| 5838 | model->super.output_size = 6; | |||
| 5839 | model->resident_slots = resident_slots; | |||
| 5840 | model->routing_width = routing_width; | |||
| 5841 | ccv_cnnp_model_copy_name(&model->super, name); | |||
| 5842 | return (ccv_cnnp_model_t*)model; | |||
| 5843 | } | |||
| 5844 | ||||
| 5845 | static ccv_cnnp_model_t* _ccv_cnnp_moe_weights_streaming_copy(const ccv_cnnp_model_t* const super, void* const context) | |||
| 5846 | { | |||
| 5847 | const ccv_cnnp_model_moe_weights_streaming_t* const self = (const ccv_cnnp_model_moe_weights_streaming_t*)super; | |||
| 5848 | return ccv_cnnp_moe_weights_streaming(self->resident_slots, self->routing_width, self->super.name); | |||
| 5849 | } |