wangwei
/
parser
Not watched
1
0
Fork 0
Code
Releases 0 Wiki Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain
Browse Source

fix

pull/242/head
wjm 5 years ago
parent
cc5425b0a0
commit
ff3aabb6f2
7 changed files with 11 additions and 1072 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +8
    -0
      CMakeLists.txt
  2. +3
    -3
      atc/main.cc
  3. +0
    -164
      atc/module.mk
  4. +0
    -191
      atc/proto/ge_ir.proto
  5. +0
    -139
      atc/proto/insert_op.proto
  6. +0
    -396
      atc/proto/om.proto
  7. +0
    -179
      atc/proto/task.proto

+ 8
- 0
CMakeLists.txt View File

@@ -16,6 +16,7 @@ if (ENABLE_OPEN_SRC)
include(cmake/external_libs/protoc.cmake)
include(cmake/external_libs/securec.cmake)
include(cmake/external_libs/json.cmake)
include(cmake/external_libs/gflags.cmake)
include(cmake/FindModule.cmake)
include(cmake/intf_pub_linux.cmake)

@@ -36,6 +37,9 @@ if (ENABLE_OPEN_SRC)
find_module(slog libalog.so ${GE_LIB_PATH})
find_module(static_mmpa libmmpa.a ${GE_LIB_PATH})
find_module(error_manager liberror_manager.so ${GE_LIB_PATH})
find_module(ge_compiler libge_compiler.so ${GE_LIB_PATH})
find_module(ge_runner libge_runner.so ${GE_LIB_PATH})
find_module(ge_common libge_common.so ${GE_LIB_PATH})
elseif(ENABLE_GE_COV OR ENABLE_GE_UT)
message(STATUS "Runing on llt mode, no need to depend other component")
else()
@@ -50,6 +54,9 @@ if (ENABLE_OPEN_SRC)
find_module(slog libalog.so ${ASCEND_ATC_DIR})
find_module(static_mmpa libmmpa.a ${ASCEND_ATC_DIR})
find_module(error_manager liberror_manager.so ${ASCEND_ATC_DIR})
find_module(ge_compiler libge_compiler.so ${ASCEND_ATC_DIR})
find_module(ge_runner libge_runner.so ${ASCEND_ATC_DIR})
find_module(ge_common libge_common.so ${ASCEND_ATC_DIR})
endif()

if (NOT DEFINED METADEF_DIR)
@@ -68,3 +75,4 @@ add_subdirectory(parser/common)
add_subdirectory(parser/func_to_graph)
add_subdirectory(parser/onnx)
add_subdirectory(parser/proto/caffe)
#add_subdirectory(atc)

+ 3
- 3
atc/main.cc View File

@@ -1359,10 +1359,10 @@ bool CheckMemInfo() {
}
// only check current available mem when auto_tune_mode is set.
long current_mem_available = GetMemInfo("MemAvailable");
GELOGI("Get mem available [%lu].", current_mem_available);
GELOGI("Get mem available [%lu kB].", current_mem_available);
std::cout << "Current available mem is " << current_mem_available << "kB." << std::endl;
if ((current_mem_available > 0) && (current_mem_available < kMinAvailableMem)) {
GELOGE(ge::PARAM_INVALID, "Current available mem [%lu] can not be smaller than [%lu] .",
GELOGE(ge::PARAM_INVALID, "Current available mem [%lu kB] can not be smaller than [%lu kB] .",
current_mem_available, kMinAvailableMem);
ErrorManager::GetInstance().ATCReportErrMessage("E10044", {"value", "min_value"},
{to_string(current_mem_available), to_string(kMinAvailableMem)});
@@ -1418,7 +1418,7 @@ int main(int argc, char* argv[]) {
if (result != 0) {
GELOGE(ge::FAILED, "ErrorManager outputErrMessage fail !");
}
GELOGI("Current mem available mem is [%lu]", GetMemInfo("MemAvailable"));
GELOGI("Current mem available mem is [%lu kB]", GetMemInfo("MemAvailable"));
return ret;
} else {
std::cout << "ATC run success, welcome to the next use." << std::endl;


+ 0
- 164
atc/module.mk View File

@@ -1,164 +0,0 @@

LOCAL_PATH := $(call my-dir)

include $(CLEAR_VARS)

LOCAL_MODULE := atc

LOCAL_CFLAGS += -Werror -Wno-deprecated-declarations
LOCAL_CFLAGS += -DPROTOBUF_INLINE_NOT_IN_HEADERS=0 -DCOMPILE_OMG_PACKAGE -O2 -Dgoogle=ascend_private

LOCAL_SRC_FILES := \
main.cc \
keep_dtype_option.cc \
single_op_parser.cc \
../session/omg.cc \
../ir_build/atc_ir_common.cc \

LOCAL_C_INCLUDES := \
$(LOCAL_PATH)/../ ./ \
$(TOPDIR)inc \
$(TOPDIR)metadef/inc \
$(TOPDIR)graphengine/inc \
$(TOPDIR)inc/external \
$(TOPDIR)metadef/inc/external \
$(TOPDIR)graphengine/inc/external \
$(TOPDIR)metadef/inc/external/graph \
$(TOPDIR)graphengine/inc/framework \
$(TOPDIR)libc_sec/include \
$(TOPDIR)metadef/inc/common/util \
$(TOPDIR)parser \
third_party/json/include \
third_party/gflags/include \
third_party/protobuf/include \
proto/om.proto \
proto/ge_ir.proto \
proto/task.proto \
proto/insert_op.proto \

LOCAL_SHARED_LIBRARIES := \
libc_sec \
libge_common \
libascend_protobuf \
libslog \
libgraph \
libregister \
liberror_manager \
libge_compiler \
libruntime_compile \
libparser_common \
liberror_manager \

LOCAL_STATIC_LIBRARIES := libgflags

LOCAL_LDFLAGS := -lrt -ldl

include $(BUILD_HOST_EXECUTABLE)

include $(CLEAR_VARS)

LOCAL_MODULE := atclib/atc.bin

LOCAL_CFLAGS += -Werror -Wno-deprecated-declarations
LOCAL_CFLAGS += -DPROTOBUF_INLINE_NOT_IN_HEADERS=0 -DCOMPILE_OMG_PACKAGE -O2 -Dgoogle=ascend_private

LOCAL_SRC_FILES := \
main.cc \
keep_dtype_option.cc \
single_op_parser.cc \
../session/omg.cc \
../ir_build/atc_ir_common.cc \

LOCAL_C_INCLUDES := \
$(LOCAL_PATH)/../ ./ \
$(TOPDIR)inc \
$(TOPDIR)metadef/inc \
$(TOPDIR)graphengine/inc \
$(TOPDIR)inc/external \
$(TOPDIR)metadef/inc/external \
$(TOPDIR)graphengine/inc/external \
$(TOPDIR)metadef/inc/external/graph \
$(TOPDIR)graphengine/inc/framework \
$(TOPDIR)libc_sec/include \
$(TOPDIR)metadef/inc/common/util \
$(TOPDIR)parser \
third_party/json/include \
third_party/gflags/include \
third_party/protobuf/include \
proto/om.proto \
proto/ge_ir.proto \
proto/task.proto \
proto/insert_op.proto \

LOCAL_SHARED_LIBRARIES := \
libc_sec \
libge_common \
libascend_protobuf \
libslog \
libgraph \
libregister \
liberror_manager \
libge_compiler \
libruntime_compile \
libparser_common \
liberror_manager \

LOCAL_STATIC_LIBRARIES := libgflags

LOCAL_LDFLAGS := -lrt -ldl

include $(BUILD_HOST_EXECUTABLE)

include $(CLEAR_VARS)

LOCAL_MODULE := fwkacl/atc.bin

LOCAL_CFLAGS += -Werror -Wno-deprecated-declarations
LOCAL_CFLAGS += -DPROTOBUF_INLINE_NOT_IN_HEADERS=0 -DCOMPILE_OMG_PACKAGE -O2 -Dgoogle=ascend_private

LOCAL_SRC_FILES := \
main.cc \
keep_dtype_option.cc \
single_op_parser.cc \
../session/omg.cc \
../ir_build/atc_ir_common.cc \

LOCAL_C_INCLUDES := \
$(LOCAL_PATH)/../ ./ \
$(TOPDIR)inc \
$(TOPDIR)metadef/inc \
$(TOPDIR)graphengine/inc \
$(TOPDIR)inc/external \
$(TOPDIR)metadef/inc/external \
$(TOPDIR)graphengine/inc/external \
$(TOPDIR)metadef/inc/external/graph \
$(TOPDIR)graphengine/inc/framework \
$(TOPDIR)libc_sec/include \
$(TOPDIR)metadef/inc/common/util \
$(TOPDIR)parser \
third_party/json/include \
third_party/gflags/include \
third_party/protobuf/include \
proto/om.proto \
proto/ge_ir.proto \
proto/task.proto \
proto/insert_op.proto \

LOCAL_SHARED_LIBRARIES := \
libc_sec \
libge_common \
libascend_protobuf \
libslog \
libgraph \
libregister \
liberror_manager \
libge_runner \
libruntime \
libparser_common \
liberror_manager \

LOCAL_STATIC_LIBRARIES := libgflags

LOCAL_LDFLAGS := -lrt -ldl

include $(BUILD_HOST_EXECUTABLE)

+ 0
- 191
atc/proto/ge_ir.proto View File

@@ -1,191 +0,0 @@
syntax = "proto3";

package ge.proto;

enum DataType
{
DT_UNDEFINED = 0; // Used to indicate a DataType field has not been set.
DT_FLOAT = 1; // float type
DT_FLOAT16 = 2; // fp16 type
DT_INT8 = 3; // int8 type
DT_UINT8 = 4; // uint8 type
DT_INT16 = 5; // int16 type
DT_UINT16 = 6; // uint16 type
DT_INT32 = 7; //
DT_INT64 = 8; // int64 type
DT_UINT32 = 9; // unsigned int32
DT_UINT64 = 10; // unsigned int64
DT_BOOL = 11; // bool type
DT_DOUBLE = 12; // double type
DT_STRING = 13; // string type
DT_DUAL_SUB_INT8 = 14; /**< dual output int8 type */
DT_DUAL_SUB_UINT8 = 15; /**< dual output uint8 type */
DT_COMPLEX64 = 16; // complex64 type
DT_COMPLEX128 = 17; // complex128 type
DT_QINT8 = 18; // qint8 type
DT_QINT16 = 19; // qint16 type
DT_QINT32 = 20; // qint32 type
DT_QUINT8 = 21; // quint8 type
DT_QUINT16 = 22; // quint16 type
DT_RESOURCE = 23; // resource type
DT_STRING_REF = 24; // string_ref type
DT_DUAL = 25; /**< dual output type */
DT_VARIANT = 26; // variant type
}

message AttrDef
{
message ListValue
{
enum ListValueType{
VT_LIST_NONE = 0;
VT_LIST_STRING = 1;
VT_LIST_INT = 2;
VT_LIST_FLOAT = 3;
VT_LIST_BOOL = 4;
VT_LIST_BYTES = 5;
VT_LIST_TENSOR_DESC = 6;
VT_LIST_TENSOR = 7;
VT_LIST_GRAPH = 8;
VT_LIST_NAMED_ATTRS = 9;
VT_LIST_DATA_TYPE = 10;
}
repeated bytes s = 2; // "list(string)"
repeated int64 i = 3; // "list(int)"
repeated float f = 4; // "list(float)"
repeated bool b = 5; // "list(bool)"
repeated bytes bt = 7;
repeated TensorDescriptor td = 8;
repeated TensorDef t = 9;
repeated GraphDef g = 10;
repeated NamedAttrs na = 11;
repeated int64 dt = 12; // list ge::DataType

ListValueType val_type = 20;
}

message ListListInt{
message ListInt{
repeated int64 list_i = 1; // list int
}
repeated ListInt list_list_i = 1; // list list int
}

oneof value
{
bytes s = 2; // "string"
int64 i = 3; // "int"
float f = 4; // "float"
bool b = 5; // "bool"
bytes bt = 7;
ListValue list = 1; // any "list(...)"
NamedAttrs func = 10; // Used to support attr nesting
TensorDescriptor td = 11; // GeTensorDesc type
TensorDef t = 12; // GeTensor type
GraphDef g = 13; // Graph type
ListListInt list_list_int = 14; // List List Int type
int64 dt = 15; // ge::DataType
}
}

// A list of attr names and their values. The whole list is attached
// with a string name. E.g., MatMul[T=float].
message NamedAttrs
{
string name = 1;
map<string, AttrDef> attr = 2;
}

// Shape / dimension description, using row-major order
message ShapeDef
{
repeated int64 dim = 1; // Size of each dimension
}

// Multidimensional data description
message TensorDescriptor
{
string name = 1; // Optional parameter, tensor name

DataType dtype = 2; // tensor datatype
ShapeDef shape = 3; // Shape / dimension
string layout = 4; // Tensor format, eg: "NCHW", "NHWC", "CHW", "ND"

bool has_out_attr = 9;
int64 size = 10;
int64 weight_size = 11;
bool reuse_input = 12;
bool output_tensor = 13;
string device_type = 14;
bool input_tensor =15;
int64 real_dim_cnt = 16;
int64 reuse_input_index = 17;
int64 data_offset = 18;
int64 cmps_size = 19;
string cmps_tab = 20;
int64 cmps_tab_offset = 21;

map<string, AttrDef> attr = 5; // Set of extra parameter fields
}

// GeTensor definition
message TensorDef
{
TensorDescriptor desc = 1; // Tensor description
bytes data = 2; // Tensor data
}


// Operator description
message OpDef
{
string name = 1; // name
string type = 2; // type

repeated string input = 5; // input original op name + outgoing index. op_name：index

map<string, AttrDef> attr = 10; // Set of operator parameter fields

bool has_out_attr = 20;
int64 id = 21;
int64 stream_id =22;
repeated string input_name = 23;
repeated string src_name = 24;
repeated int64 src_index = 25;
repeated string dst_name = 26;
repeated int64 dst_index = 27;
repeated int64 input_i = 28;
repeated int64 output_i = 29;
repeated int64 workspace = 30;
repeated int64 workspace_bytes = 31;
repeated bool is_input_const = 32;
repeated TensorDescriptor input_desc = 33;
repeated TensorDescriptor output_desc = 34;
repeated string subgraph_name = 35;
}

// Graph definition
message GraphDef
{
string name = 1; // name

repeated string input = 4; // Graph input
repeated string output = 5; // Graph output

repeated OpDef op = 6; // List of operators

map<string, AttrDef> attr = 11; // Extended field
}

// model definition
message ModelDef
{
string name = 1; // name
uint32 version = 2; // IR Proto verion
string custom_version = 3; // User model version number, passed in by user

repeated GraphDef graph = 7; // Graph definition，graph[0] represents the main diagram in modeldef

map<string, AttrDef> attr = 11; // Extended field
}


+ 0
- 139
atc/proto/insert_op.proto View File

@@ -1,139 +0,0 @@
syntax = "proto3";

package domi;

message InsertNewOps {
repeated AippOpParams aipp_op = 1;
repeated MultiShapeOpParams multi_shape_op = 2;
}

message AippOpParams {
enum InputFormat {
UNDEFINED = 0;
YUV420SP_U8 = 1;
XRGB8888_U8 = 2;
RGB888_U8 = 3;
YUV400_U8 = 4;
NC1HWC0DI_FP16 = 5;
NC1HWC0DI_S8 = 6;
ARGB8888_U8 = 7;
YUYV_U8 = 8;
YUV422SP_U8 = 9;
AYUV444_U8 = 10;
RAW10 = 11;
RAW12 = 12;
RAW16 = 13;
RAW24 = 14;
RGB16 = 15;
RGB20 = 16;
RGB24 = 17;
RGB8_IR = 18;
RGB16_IR = 19;
RGB24_IR = 20;
}

enum AippMode {
undefined = 0;
static = 1;
dynamic = 2;
}

// AIPP模式，区分静态AIPP和动态AIPP
AippMode aipp_mode = 1;

// related_input_rank参数为必填，类型为整型，配置范围>=0, <=输入Data算子的个数，默认值为0。
// 标识对模型的第几个输入做AIPP处理，例如模型有两个输入，需要对第2个输入做AIPP，则配置related_input_rank为1。
uint32 related_input_rank = 2;

// related_input_name is optional and the top name of data node which inserts aipp
string related_input_name = 6;

// input_edge_idx参数为可选，类型为整型，配置范围为>=0。
// 配置该参数的作用，在于对Data算子不同的输出做不同的AIPP处理，如果该参数没有配置，默认对related_input_rank指定的模型输入的所有输出边做AIPP。
// 配置值 <= Data算子输出边的个数。
repeated uint32 input_edge_idx = 3;

// [Begin] 动态AIPP参数，配置静态AIPP时无效
uint32 max_src_image_size = 4;

// 是否支持旋转。默认不支持，开启支持旋转时，会有额外的空间和性能损失
bool support_rotation = 5;

// [End] 动态AIPP参数


// [Begin] 静态AIPP参数，配置动态AIPP时无效
InputFormat input_format = 51;
bool csc_switch = 52;
float cpadding_value = 53;
bool rbuv_swap_switch = 54;
bool ax_swap_switch = 55;
bool single_line_mode = 56;

int32 src_image_size_w = 57;
int32 src_image_size_h = 58;

bool crop = 59;
int32 load_start_pos_w = 60;
int32 load_start_pos_h = 61;
int32 crop_size_w = 62;
int32 crop_size_h = 63;

bool resize = 64;
int32 resize_output_w = 65;
int32 resize_output_h = 66;

bool padding = 67;
int32 left_padding_size = 68;
int32 right_padding_size = 69;
int32 top_padding_size = 70;
int32 bottom_padding_size = 71;

int32 mean_chn_0 = 10;
int32 mean_chn_1 = 11;
int32 mean_chn_2 = 12;
int32 mean_chn_3 = 19;
float min_chn_0 = 13;
float min_chn_1 = 14;
float min_chn_2 = 15;
float min_chn_3 = 20;
repeated float var_reci_chn_0 = 16;
repeated float var_reci_chn_1 = 17;
repeated float var_reci_chn_2 = 18;
repeated float var_reci_chn_3 = 21;

repeated int32 matrix_r0c0 = 30;
repeated int32 matrix_r0c1 = 31;
repeated int32 matrix_r0c2 = 32;
repeated int32 matrix_r1c0 = 33;
repeated int32 matrix_r1c1 = 34;
repeated int32 matrix_r1c2 = 35;
repeated int32 matrix_r2c0 = 36;
repeated int32 matrix_r2c1 = 37;
repeated int32 matrix_r2c2 = 38;
repeated int32 output_bias_0 = 39;
repeated int32 output_bias_1 = 40;
repeated int32 output_bias_2 = 41;
repeated int32 input_bias_0 = 42;
repeated int32 input_bias_1 = 43;
repeated int32 input_bias_2 = 44;

// [End] 静态AIPP参数

// The n number that is used for raw/rgbir data into f16 transformation.
// The transformation equation is x/(2^n). If set to 0, no transform is performed.
uint32 raw_rgbir_to_f16_n = 45;
}

message MultiShapeOpParams {
enum MultiShapeMode {
batch = 0; //动态batch
resolution = 1; //动态分辨率，扩展用
}

MultiShapeMode mode = 1; //算子模式
uint32 related_input_rank = 2; //新增算子插入到哪个输入


repeated uint32 batch_list = 11; //batch_list值，batch_list的个数是2到8之间
}

+ 0
- 396
atc/proto/om.proto View File

@@ -1,396 +0,0 @@
/* Copyright (C) 2018. Huawei Technologies Co., Ltd. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the Apache License Version 2.0.You may not use this file except in compliance with the License.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* Apache License for more details at
* http://www.apache.org/licenses/LICENSE-2.0
*/
syntax = "proto3";

package domi;

enum TargetType
{
MINI = 0;
TINY = 1;
LITE = 2;
}

// offline model
message ModelDef {
string name = 1;
uint32 version = 2;

uint64 memory_size = 10;
uint32 stream_num = 11;
uint32 event_num = 12;
uint64 weight_size = 13;
uint32 label_num = 15;
repeated OpDef op = 20;
TargetType target_type = 23;

map<string, AttrDef> attr = 30;
};

// operator define
message OpDef {
string name = 1;
string type = 2;

uint32 id = 3;
uint32 stream_id = 4;

repeated string input_name = 5;

repeated string src_name = 8;
repeated int32 src_index = 9;
repeated int64 input = 10;
repeated int64 output = 11;
repeated TensorDescriptor input_desc = 12;
repeated TensorDescriptor output_desc = 13;
repeated WeightDef weights = 14;
repeated string dst_name = 15;
repeated int32 dst_index = 16;

repeated int64 workspace = 20;
repeated uint32 workspace_bytes = 21;

repeated string weight_name = 22;
repeated bool is_input_const = 23;

map<string, AttrDef> attr = 30;

QuantizeFactorParams quantize_factor = 31;

oneof op_params {
// start at 100 here
SendOpParams sender_param = 100;
RecvOpParams receiver_param = 200;
ConvolutionOpParams convolution_param = 300;
PoolingOpParams pooling_param = 400;
EltwiseOpParams eltwise_param = 500;
BatchNormOpParams batchnorm_param = 600;
ScaleOpParams scale_param = 700;
FullConnectionOpParams full_connection_param = 800;
SoftmaxOpParams softmax_param = 900;
ActivationOpParams activation_param = 1000;
ReshapeOpParams reshape_param = 1100;
}
};

message SendOpParams {
uint32 event_id = 1;
};

message RecvOpParams {
uint32 event_id = 1;
};

enum QuantizeScaleType
{
VECTOR_SCALE = 0;
SCALAR_SCALE = 1;
}

enum QuantizeScaleMode
{
NORMAL_MODE = 0;
SQRT_MODE = 1;
}

enum QuantizeAlgorithm
{
NON_OFFSET_ALGO = 0;
HALF_OFFSET_ALGO = 1;
ALL_OFFSET_ALGO = 2;
}
message QuantizeFactor
{
QuantizeScaleMode scale_mode = 1;
bytes scale_value = 2;
int64 scale_offset = 3;
bytes offset_data_value = 4;
int64 offset_data_offset = 5;
bytes offset_weight_value = 6;
int64 offset_weight_offset = 7;
bytes offset_pad_value = 8;
int64 offset_pad_offset = 9;
};

message QuantizeCalcFactor
{
bytes offsetw = 1;
int64 offsetw_offset = 2;
bytes offsetd = 3;
int64 offsetd_offset = 4;
bytes scalereq = 5;
int64 scaledreq_offset = 6;
bytes offsetdnext = 7;
int64 offsetdnext_offset = 8;
}

message QuantizeFactorParams
{
QuantizeAlgorithm quantize_algo = 1;
QuantizeScaleType scale_type = 2;
QuantizeFactor quantize_param = 3;
QuantizeFactor dequantize_param = 4;
QuantizeFactor requantize_param = 5;
QuantizeCalcFactor quantizecalc_param = 6;
};

message ConvolutionOpParams {
int32 mode = 1;
int32 algo = 2;
int32 pad_mode = 3;
uint32 group = 4;
uint32 num_output = 5;

repeated uint32 pad = 10;
repeated uint32 stride = 11;
repeated uint32 dilation = 12;
repeated uint32 kernel = 13;

float alpha = 20;
float beta = 21;

WeightDef filter = 40;
WeightDef bias = 41;

bool relu_flag = 62;
repeated uint32 adj = 70;
repeated uint32 target_shape = 71;
repeated uint32 before_pad = 72;
};

message PoolingOpParams {
int32 mode = 1;
int32 nan_opt = 2;
int32 pad_mode = 3;
bool global_pooling = 4;

repeated uint32 window = 10;
repeated uint32 pad = 11;
repeated uint32 stride = 12;
bool ceil_mode = 13;
int32 data_mode = 14;

float alpha = 20;
float beta = 21;
repeated uint32 before_pad = 22;
};

message EltwiseOpParams {
int32 mode = 1;
repeated float coeff = 2;
float alpha = 3;
float beta = 4;
repeated WeightDef weight = 5;
bool relu_flag = 6;
};

message ActivationOpParams {
int32 mode = 1;
float coef = 2;
float alpha = 3;
float beta = 4;
};

message BatchNormOpParams {
int32 mode = 1;

float alpha = 2;
float beta = 3;
double epsilon = 4;//optinal,[default = 1e-5]
bool use_global_stats = 5; //optinal,by default true,testing mode
float moving_average_fraction = 6; //optinal,[default = .999];

WeightDef estimated_mean = 7;
WeightDef estimated_variance = 8;

WeightDef scale = 9;
WeightDef bias = 10;
};

message ScaleOpParams {
WeightDef scale = 1;
WeightDef bias = 2;
};

message ReshapeOpParams {
float alpha = 1;
float beta = 2;
ShapeDef shape = 3;
int32 axis = 4;
int32 num_axes = 5;
int32 format = 6;
};

message SoftmaxOpParams {
int32 algo = 1;
int32 mode = 2;
float alpha = 3;
float beta = 4;
};

message FullConnectionOpParams {
WeightDef filter = 1;
WeightDef bias = 2;
uint32 num_output = 3;
bool relu_flag = 12;
};

message FlattenOpParams {
float alpha = 1;
float beta = 2;
int32 start_axis = 3;
int32 end_axis = 4;
}

message AddLimitedOpParams {
float alpha = 1;
float beta = 2;
int32 axis = 3;
bool broadcast = 4;

repeated WeightDef weight = 10;
};

message MulLimitedOpParams {
float alpha = 1;
float beta = 2;
int32 axis = 3;
bool broadcast = 4;

repeated WeightDef weight = 10;
};

message AddOpParams {
float alpha = 1;
float beta = 2;

repeated WeightDef weight = 10;
};

message MulOpParams {
float alpha = 1;
float beta = 2;

repeated WeightDef weight = 10;
};

message SubOpParams {
float alpha = 1;
float beta = 2;

repeated WeightDef weight = 10;
};

message BiasAddOpParams {
float alpha = 1;
float beta = 2;

WeightDef bias = 10;
};

message MatMulOpParams {
float alpha = 1;
float beta = 2;
bool transposeX = 3;
bool transposeW = 4;

WeightDef filter = 10;
WeightDef bias = 12;
};

message RsqrtOpParams {
float alpha = 1;
float beta = 2;
};


message WeightDef {
int32 format = 1;
int32 data_type = 2;
ShapeDef shape = 3;
bytes data = 4;
int64 data_offset = 5;
uint32 cmps_size = 6;
bytes cmps_tab = 7;
int64 cmps_tab_offset = 10;
CompressInfo cmps_info = 8;
AllOffsetQuantizeInfo alloffset_quantize_info = 11;
}

message ShapeDef {
repeated int64 dim = 1;
}

enum DeviceType {
NPU = 0; // In default, we will use NPU.
CPU = 1; // CPU
}

message AllOffsetQuantizeInfo {
float scale = 1;
int32 offset = 2;
}

message TensorDescriptor {
int32 format = 1;
int32 data_type = 2;
repeated int64 dim = 3;
uint32 size = 4;
bool reuse_input = 5;
bool output_tensor = 7;
DeviceType device_type = 8;
bool input_tensor = 9;
uint32 real_dim_cnt = 10;
uint32 reuse_input_index = 11;
AllOffsetQuantizeInfo alloffset_quantize_info = 12;
}

message CompressInfo {
int32 blockRow = 1; // block row
int32 blockCol = 2; // block col
int32 fractalK = 3; // fractal K
int32 fractalN = 4; // fractal N
int32 lastFractalK = 5; // K of last fractal
int32 lastFractalN = 6; // N of last fractal
int32 cubeSize = 7; // cube's length
int32 loadDir = 8; // data load directtiono 0:col load 1:row load
}

message AttrDef {
message ListValue {
repeated string s = 2; // "list(string)"
repeated int64 i = 3 [packed = true]; // "list(int)"
repeated float f = 4 [packed = true]; // "list(float)"
repeated bool b = 5 [packed = true]; // "list(bool)"
repeated uint32 u = 6 [packed = true]; // "list(uint)"
repeated bytes bt = 7;
}

oneof value {
string s = 2; // "string"
int64 i = 3; // "int"
float f = 4; // "float"
bool b = 5; // "bool"
uint32 u = 6; // "uint32"
bytes bt = 7;
ListValue list = 1; // any "list(...)"
NamedAttrs func = 10;
}
}

// A list of attr names and their values. The whole list is attached
// with a string name. E.g., MatMul[T=float].
message NamedAttrs {
string name = 1;
map<string, AttrDef> attr = 2;
}


+ 0
- 179
atc/proto/task.proto View File

@@ -1,179 +0,0 @@
/* Copyright (C) 2018. Huawei Technologies Co., Ltd. All rights reserved.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the Apache License Version 2.0.You may not use this file except in compliance with the License.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* Apache License for more details at
* http://www.apache.org/licenses/LICENSE-2.0
*/
syntax = "proto3";

package domi;

message ModelTaskDef {
string version = 1;

map<string, string> attr = 9; // Extended field
repeated TaskDef task = 10;

uint64 memory_size = 11;
uint32 stream_num = 12;
uint32 event_num = 13;
uint64 weight_size = 14;

repeated bytes op = 15; // input/output opdef in bytes

uint64 base_addr = 16; // base addr
uint64 weight_addr = 17; // weight addr
uint32 batch_num = 18;
}


message TaskDef {
uint32 id = 1;
uint32 type = 2;

uint32 stream_id = 10;
uint32 event_id = 11;

KernelDef kernel = 20;
KernelExDef kernel_ex = 21;
KernelHcclDef kernel_hccl = 25;
EventExDef event_ex = 26;
LogTimeStampDef log_timestamp = 28;

uint32 label_id = 30;

MemcpyAsyncDef memcpy_async = 31;
StreamSwitchDef stream_switch = 32;
StreamActiveDef stream_active = 33;
bytes private_def = 34;
uint64 ops_kernel_store_ptr = 35; // adjustments to other fields in the future
StreamSwitchNDef stream_switch_n = 36;

LabelSetDef label_set = 37;
LabelGotoExDef label_goto_ex = 38;
LabelSwitchByIndexDef label_switch_by_index = 39;
KernelDefWithHandle kernel_with_handle = 40;
}

message KernelDef {
KernelContext context = 1;

string stub_func = 10;
uint32 block_dim = 11;
uint32 args_size = 12;
bytes args = 13;
bytes sm_desc = 14;
bytes flowtable = 15;
string so_name = 16;
string kernel_name = 17;
bytes kernel_ext_info = 18;
uint32 kernel_ext_info_size = 19;
}

message KernelDefWithHandle {
KernelContext context = 1;

uint64 handle = 10;
string dev_func = 11;
uint32 block_dim = 12;
uint32 args_size = 13;
bytes args = 14;
bytes sm_desc = 15;
string original_kernel_key = 16;
string node_info = 17;
}

message KernelContext {
uint32 kernel_type = 1;
uint32 op_id = 2; // OP type in CCE
uint32 kernel_func_id = 3;
uint32 op_index = 4; // TE/Custom operator
bool is_flowtable = 5; // Identify whether args is a flowtable structure
bytes args_offset = 6; // args offset information
uint32 args_count = 7; // args count
repeated uint32 origin_op_index = 8;
}


message KernelExDef {
uint32 flags = 1;

uint32 op_index = 4;
uint32 args_size = 12;
bytes args = 13;
bytes task_info = 14; // serialized nodeDef, funcDef, inputoutput
uint32 task_info_size = 15;
bytes kernel_ext_info = 16;
uint32 kernel_ext_info_size = 17;
}


message KernelHcclDef {
uint32 op_index = 8;
string hccl_type = 9;
}


message EventExDef {
uint32 op_index = 1;
uint32 event_type = 2;
}

message LogTimeStampDef {
uint64 logid = 1;
bool notify = 2;
uint32 flat = 3;
}

message MemcpyAsyncDef {
uint64 dst = 1;
uint64 dst_max = 2;
uint64 src = 3;
uint64 count = 4;
uint32 kind = 5;
uint32 op_index = 6;
}

message StreamSwitchDef {
uint32 op_index = 1;
uint32 true_stream_id = 2;
int64 value = 3;
uint64 value_ptr = 4;
uint32 data_type = 5;
}

message StreamActiveDef {
uint32 op_index = 1;
uint32 active_stream_id = 2;
}

message StreamSwitchNDef {
uint32 op_index = 1;
uint32 size = 2;
repeated int64 target_value = 3;
repeated uint32 true_stream_id = 4;
uint32 element_size = 5;
uint32 data_type = 6;
}

message LabelSetDef {
uint32 op_index = 1;
uint32 label_id = 2;
uint32 model_id = 3;
}

message LabelGotoExDef {
uint32 op_index = 1;
uint32 label_id = 2;
uint32 model_id = 3;
}

message LabelSwitchByIndexDef {
uint32 op_index = 1;
uint32 label_max = 2;
}

Write Preview
Loading…
Cancel
Save