| // Copyright 2021 Google LLC |
| // |
| // This source code is licensed under the BSD-style license found in the |
| // LICENSE file in the root directory of this source tree. |
| |
| #include <assert.h> |
| #include <math.h> |
| #include <stddef.h> |
| #include <stdint.h> |
| |
| #include <xnnpack.h> |
| #include <xnnpack/log.h> |
| #include <xnnpack/operator.h> |
| #include <xnnpack/params.h> |
| #include <xnnpack/subgraph.h> |
| #include <xnnpack/subgraph-validation.h> |
| |
| |
| |
| static enum xnn_status create_convert_operator( |
| const struct xnn_node* node, |
| const struct xnn_value* values, |
| size_t num_values, |
| struct xnn_operator_data* opdata, |
| const struct xnn_caches* caches) |
| { |
| assert(node->num_inputs == 1); |
| const uint32_t input_id = node->inputs[0]; |
| assert(input_id != XNN_INVALID_VALUE_ID); |
| assert(input_id < num_values); |
| |
| assert(node->num_outputs == 1); |
| const uint32_t output_id = node->outputs[0]; |
| assert(output_id != XNN_INVALID_VALUE_ID); |
| assert(output_id < num_values); |
| |
| const size_t num_input_dims = values[input_id].shape.num_dims; |
| const size_t channel_dim = num_input_dims == 0 ? 1 : values[input_id].shape.dim[num_input_dims - 1]; |
| |
| enum xnn_status status = xnn_status_uninitialized; |
| switch (node->compute_type) { |
| case xnn_compute_type_fp32_to_fp16: |
| status = xnn_create_convert_nc_f32_f16( |
| channel_dim /* channels */, channel_dim /* input stride */, channel_dim /* output stride */, |
| node->flags, |
| &opdata->operator_objects[0]); |
| break; |
| case xnn_compute_type_fp32_to_qs8: |
| status = xnn_create_convert_nc_f32_qs8( |
| channel_dim /* channels */, channel_dim /* input stride */, channel_dim /* output stride */, |
| values[output_id].quantization.scale, |
| (int8_t) values[output_id].quantization.zero_point, |
| INT8_MIN, INT8_MAX, |
| node->flags, |
| &opdata->operator_objects[0]); |
| break; |
| case xnn_compute_type_fp32_to_qu8: |
| status = xnn_create_convert_nc_f32_qu8( |
| channel_dim /* channels */, channel_dim /* input stride */, channel_dim /* output stride */, |
| values[output_id].quantization.scale, |
| (uint8_t) values[output_id].quantization.zero_point, |
| 0, UINT8_MAX, |
| node->flags, |
| &opdata->operator_objects[0]); |
| break; |
| case xnn_compute_type_fp16_to_fp32: |
| status = xnn_create_convert_nc_f16_f32( |
| channel_dim /* channels */, channel_dim /* input stride */, channel_dim /* output stride */, |
| node->flags, |
| &opdata->operator_objects[0]); |
| break; |
| case xnn_compute_type_qs8: |
| status = xnn_create_convert_nc_qs8( |
| channel_dim /* channels */, channel_dim /* input stride */, channel_dim /* output stride */, |
| values[input_id].quantization.scale, |
| (int8_t) values[input_id].quantization.zero_point, |
| values[output_id].quantization.scale, |
| (int8_t) values[output_id].quantization.zero_point, |
| node->flags, |
| &opdata->operator_objects[0]); |
| break; |
| case xnn_compute_type_qs8_to_fp32: |
| status = xnn_create_convert_nc_qs8_f32( |
| channel_dim /* channels */, channel_dim /* input stride */, channel_dim /* output stride */, |
| values[input_id].quantization.scale, |
| (int8_t) values[input_id].quantization.zero_point, |
| node->flags, |
| &opdata->operator_objects[0]); |
| break; |
| case xnn_compute_type_qu8: |
| status = xnn_create_convert_nc_qu8( |
| channel_dim /* channels */, channel_dim /* input stride */, channel_dim /* output stride */, |
| values[input_id].quantization.scale, |
| (uint8_t) values[input_id].quantization.zero_point, |
| values[output_id].quantization.scale, |
| (uint8_t) values[output_id].quantization.zero_point, |
| node->flags, |
| &opdata->operator_objects[0]); |
| break; |
| case xnn_compute_type_qu8_to_fp32: |
| status = xnn_create_convert_nc_qu8_f32( |
| channel_dim /* channels */, channel_dim /* input stride */, channel_dim /* output stride */, |
| values[input_id].quantization.scale, |
| (uint8_t) values[input_id].quantization.zero_point, |
| node->flags, |
| &opdata->operator_objects[0]); |
| break; |
| default: |
| XNN_UNREACHABLE; |
| } |
| if (status == xnn_status_success) { |
| opdata->batch_size = xnn_shape_multiply_non_channel_dims(&values[input_id].shape); |
| opdata->inputs[0] = input_id; |
| opdata->outputs[0] = output_id; |
| } |
| return status; |
| } |
| |
| static enum xnn_status setup_convert_operator( |
| const struct xnn_operator_data* opdata, |
| const struct xnn_blob* blobs, |
| size_t num_blobs, |
| pthreadpool_t threadpool) |
| { |
| const uint32_t input_id = opdata->inputs[0]; |
| assert(input_id != XNN_INVALID_VALUE_ID); |
| assert(input_id < num_blobs); |
| |
| const uint32_t output_id = opdata->outputs[0]; |
| assert(output_id != XNN_INVALID_VALUE_ID); |
| assert(output_id < num_blobs); |
| |
| const struct xnn_blob* input_blob = blobs + input_id; |
| const void* input_data = input_blob->data; |
| assert(input_data != NULL); |
| |
| const struct xnn_blob* output_blob = blobs + output_id; |
| void* output_data = output_blob->data; |
| assert(output_data != NULL); |
| |
| switch (opdata->operator_objects[0]->type) { |
| case xnn_operator_type_convert_nc_f32_f16: |
| return xnn_setup_convert_nc_f32_f16( |
| opdata->operator_objects[0], |
| opdata->batch_size, |
| input_data, |
| output_data, |
| threadpool); |
| case xnn_operator_type_convert_nc_f32_qs8: |
| return xnn_setup_convert_nc_f32_qs8( |
| opdata->operator_objects[0], |
| opdata->batch_size, |
| input_data, |
| output_data, |
| threadpool); |
| case xnn_operator_type_convert_nc_f32_qu8: |
| return xnn_setup_convert_nc_f32_qu8( |
| opdata->operator_objects[0], |
| opdata->batch_size, |
| input_data, |
| output_data, |
| threadpool); |
| case xnn_operator_type_convert_nc_f16_f32: |
| return xnn_setup_convert_nc_f16_f32( |
| opdata->operator_objects[0], |
| opdata->batch_size, |
| input_data, |
| output_data, |
| threadpool); |
| case xnn_operator_type_convert_nc_qs8: |
| return xnn_setup_convert_nc_qs8( |
| opdata->operator_objects[0], |
| opdata->batch_size, |
| input_data, |
| output_data, |
| threadpool); |
| case xnn_operator_type_convert_nc_qs8_f32: |
| return xnn_setup_convert_nc_qs8_f32( |
| opdata->operator_objects[0], |
| opdata->batch_size, |
| input_data, |
| output_data, |
| threadpool); |
| case xnn_operator_type_convert_nc_qu8: |
| return xnn_setup_convert_nc_qu8( |
| opdata->operator_objects[0], |
| opdata->batch_size, |
| input_data, |
| output_data, |
| threadpool); |
| case xnn_operator_type_convert_nc_qu8_f32: |
| return xnn_setup_convert_nc_qu8_f32( |
| opdata->operator_objects[0], |
| opdata->batch_size, |
| input_data, |
| output_data, |
| threadpool); |
| default: |
| XNN_UNREACHABLE; |
| } |
| } |
| |
| static inline enum xnn_compute_type validate_datatypes( |
| enum xnn_datatype input_datatype, |
| enum xnn_datatype output_datatype) |
| { |
| switch (input_datatype) { |
| case xnn_datatype_fp32: |
| switch (output_datatype) { |
| case xnn_datatype_fp16: |
| return xnn_compute_type_fp32_to_fp16; |
| case xnn_datatype_qint8: |
| return xnn_compute_type_fp32_to_qs8; |
| case xnn_datatype_quint8: |
| return xnn_compute_type_fp32_to_qu8; |
| default: |
| break; |
| } |
| break; |
| case xnn_datatype_fp16: |
| if (output_datatype == xnn_datatype_fp32) { |
| return xnn_compute_type_fp16_to_fp32; |
| } |
| break; |
| case xnn_datatype_qint8: |
| switch (output_datatype) { |
| case xnn_datatype_fp32: |
| return xnn_compute_type_qs8_to_fp32; |
| case xnn_datatype_qint8: |
| return xnn_compute_type_qs8; |
| default: |
| break; |
| } |
| break; |
| case xnn_datatype_quint8: |
| switch (output_datatype) { |
| case xnn_datatype_fp32: |
| return xnn_compute_type_qu8_to_fp32; |
| case xnn_datatype_quint8: |
| return xnn_compute_type_qu8; |
| default: |
| break; |
| } |
| break; |
| default: |
| XNN_UNREACHABLE; |
| } |
| return xnn_compute_type_invalid; |
| } |
| |
| void xnn_init_convert_node( |
| struct xnn_node* node, |
| enum xnn_compute_type compute_type, |
| uint32_t input_id, |
| uint32_t output_id, |
| uint32_t flags) |
| { |
| node->type = xnn_node_type_convert; |
| node->compute_type = compute_type; |
| node->num_inputs = 1; |
| node->inputs[0] = input_id; |
| node->num_outputs = 1; |
| node->outputs[0] = output_id; |
| node->flags = flags; |
| |
| node->create = create_convert_operator; |
| node->setup = setup_convert_operator; |
| } |
| |
| enum xnn_status xnn_define_convert( |
| xnn_subgraph_t subgraph, |
| uint32_t input_id, |
| uint32_t output_id, |
| uint32_t flags) |
| { |
| enum xnn_status status; |
| if ((status = xnn_subgraph_check_xnnpack_initialized(xnn_node_type_convert)) != xnn_status_success) { |
| return status; |
| } |
| |
| if ((status = xnn_subgraph_check_input_node_id(xnn_node_type_convert, input_id, subgraph->num_values)) != |
| xnn_status_success) { |
| return status; |
| } |
| |
| const struct xnn_value* input_value = &subgraph->values[input_id]; |
| status = xnn_subgraph_check_input_type_dense(xnn_node_type_convert, input_id, input_value); |
| if (status != xnn_status_success) { |
| return status; |
| } |
| |
| switch (input_value->datatype) { |
| case xnn_datatype_fp16: |
| case xnn_datatype_fp32: |
| case xnn_datatype_qint8: |
| case xnn_datatype_quint8: |
| break; |
| default: |
| xnn_log_error( |
| "failed to define %s operator with input ID #%" PRIu32 ": unsupported Value datatype %s (%d)", |
| xnn_node_type_to_string(xnn_node_type_convert), input_id, |
| xnn_datatype_to_string(input_value->datatype), input_value->datatype); |
| return xnn_status_invalid_parameter; |
| } |
| |
| status = xnn_subgraph_check_output_node_id(xnn_node_type_convert, output_id, subgraph->num_values); |
| if (status != xnn_status_success) { |
| return status; |
| } |
| |
| const struct xnn_value* output_value = &subgraph->values[output_id]; |
| status = xnn_subgraph_check_output_type_dense(xnn_node_type_convert, output_id, output_value); |
| if (status != xnn_status_success) { |
| return status; |
| } |
| |
| switch (output_value->datatype) { |
| case xnn_datatype_fp16: |
| case xnn_datatype_fp32: |
| case xnn_datatype_qint8: |
| case xnn_datatype_quint8: |
| break; |
| default: |
| xnn_log_error( |
| "failed to define %s operator with output ID #%" PRIu32 ": unsupported Value datatype %s (%d)", |
| xnn_node_type_to_string(xnn_node_type_convert), output_id, |
| xnn_datatype_to_string(output_value->datatype), output_value->datatype); |
| return xnn_status_invalid_parameter; |
| } |
| |
| enum xnn_compute_type compute_type = validate_datatypes(input_value->datatype, output_value->datatype); |
| if (compute_type == xnn_compute_type_invalid) { |
| xnn_log_error( |
| "failed to define %s operator with input ID #%" PRIu32 " and output ID #%" PRIu32 |
| ": mismatching datatypes across input (%s) and output (%s)", |
| xnn_node_type_to_string(xnn_node_type_convert), input_id, output_id, |
| xnn_datatype_to_string(input_value->datatype), |
| xnn_datatype_to_string(output_value->datatype)); |
| return xnn_status_invalid_parameter; |
| } |
| |
| switch (compute_type) { |
| case xnn_compute_type_invalid: |
| xnn_log_error( |
| "failed to define %s operator with input ID #%" PRIu32 " and output ID #%" PRIu32 |
| ": mismatching datatypes across input (%s) and output (%s)", |
| xnn_node_type_to_string(xnn_node_type_convert), input_id, output_id, |
| xnn_datatype_to_string(input_value->datatype), |
| xnn_datatype_to_string(output_value->datatype)); |
| return xnn_status_invalid_parameter; |
| case xnn_compute_type_qs8: |
| case xnn_compute_type_qu8: |
| { |
| const float input_output_scale = input_value->quantization.scale / output_value->quantization.scale; |
| if (input_output_scale < 0x1.0p-8f || input_output_scale > 0x1.0p+7f) { |
| xnn_log_error( |
| "failed to define %s operator with %.7g input-to-output scale ratio (input #%"PRIu32" scale %.7g, output #%"PRIu32" scale %.7g): " |
| "scale ratio must be in [2**-8, 2**7] range", |
| xnn_node_type_to_string(xnn_node_type_convert), input_output_scale, |
| input_id, input_value->quantization.scale, output_id, output_value->quantization.scale); |
| return xnn_status_invalid_parameter; |
| } |
| break; |
| } |
| default: |
| break; |
| } |
| |
| struct xnn_node* node = xnn_subgraph_new_node(subgraph); |
| if (node == NULL) { |
| return xnn_status_out_of_memory; |
| } |
| |
| xnn_init_convert_node(node, compute_type, input_id, output_id, flags); |
| return xnn_status_success; |
| } |
