Support function_order and add implement flag into nn_function_t.

build-tools/code_generator/generate.py

@@ -14,13 +14,27 @@
 
 import io
 import os
-from os.path import abspath, dirname, join
+from os.path import abspath, dirname, join, exists
 
 here = abspath(dirname(abspath(__file__)))
 base = abspath(join(here, '../..'))
 
 import code_generator_utils as utils
-
+import yaml
+
+def type_to_pack_format(typestring):
+    fmt = None
+    if typestring == 'bool':
+        fmt = 'B'
+    elif typestring == 'double' or typestring == 'float':
+        fmt = 'f'
+    elif typestring == 'int64':
+        fmt = 'i'
+    elif typestring == 'repeated int64' or typestring == 'Shape':
+        fmt = 'iI'
+    elif typestring == 'string':
+        fmt = 'i'
+    return fmt
 
 def generate_cpp_utils(function_info):
     function_list = utils.info_to_list(function_info)
@@ -55,6 +69,53 @@ def generate_solver_python_intereface(solver_info):
     utils.generate_from_template(
         join(base, 'python/src/nnabla/solver.pxd.tmpl'), solver_info=solver_info)
 
+def generate_function_order(function_info):
+    with open(join(base, 'python/src/nnabla/utils/converter/utils.py')) as f:
+        exec(f.read())
+    order_yaml = join(base, 'python/src/nnabla/utils/converter/function_order.yaml')
+    order_info= None
+    if exists(order_yaml):
+        with open(order_yaml, 'r') as f:
+            order_info = utils.load_yaml_ordered(f)
+
+    if order_info is None:
+        order_info= OrderedDict()    
+
+    func_id = 0
+    for func, func_info in function_info.items():
+        fmt = ''
+        if 'arguments' in func_info:
+            fmt = '_'
+            for arg, arg_info in func_info['arguments'].items():
+                fmt += type_to_pack_format(arg_info['type'])
+
+        name = func+fmt
+        if name not in order_info:
+            while func_id in order_info.values():
+                func_id += 1
+            order_info[name] = func_id
+
+        func_id += 1
+    with open(order_yaml, 'w') as f:
+        f.write('# Copyright (c) 2017 Sony Corporation. All Rights Reserved.\n')
+        f.write('#\n')
+        f.write('# Licensed under the Apache License, Version 2.0 (the "License");\n')
+        f.write('# you may not use this file except in compliance with the License.\n')
+        f.write('# You may obtain a copy of the License at\n')
+        f.write('#\n')
+        f.write('#     http://www.apache.org/licenses/LICENSE-2.0\n')
+        f.write('#\n')
+        f.write('# Unless required by applicable law or agreed to in writing, software\n')
+        f.write('# distributed under the License is distributed on an "AS IS" BASIS,\n')
+        f.write('# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n')
+        f.write('# See the License for the specific language governing permissions and\n')
+        f.write('# limitations under the License.\n')
+        f.write('#\n')
+        f.write('# DO NOT EDIT THIS FILE!\n')
+        f.write('# THIS FILE IS GENERATED BY CODE GENERATOR BUT ITS COMMITED INTO\n')
+        f.write('# SOURCE TREE TO MAKE FUNCTION ID PERSIST.\n')
+        f.write('\n')
+        f.write(yaml.dump(order_info, default_flow_style=False))
 
 def generate():
     function_info = utils.load_function_info(flatten=True)
@@ -73,6 +134,7 @@ def generate():
     generate_python_utils(function_info)
     generate_proto(function_info, solver_info)
     generate_cpp_utils(function_info)
+    generate_function_order(function_info)
 
     # Generate function skeltons if new ones are added to functions.yaml and function_types.yaml.
     utils.generate_skelton_function_impl(

python/setup.py

@@ -185,7 +185,9 @@ def extopts(library_name, library_dir):
 
     shutil.copyfile(library_path, os.path.join(path_pkg, library_file_name))
     package_data = {"nnabla": [
-        library_file_name, 'nnabla.conf', 'utils/converter/functions.yaml']}
+        library_file_name, 'nnabla.conf',
+                           'utils/converter/functions.yaml',
+                           'utils/converter/function_order.yaml']}
 
     for root, dirs, files in os.walk(os.path.join(build_dir, 'bin')):
         for fn in files:

python/src/nnabla/utils/converter/__init__.py

@@ -15,6 +15,7 @@
 from .supported_info import formats
 from .supported_info import extensions
 from .convert_files import convert_files
+from .utils import type_to_pack_format
 from .utils import get_category_info_string
 from .utils import get_category_info_version
 from .utils import get_category_info

python/src/nnabla/utils/converter/function_order.yaml

@@ -0,0 +1,139 @@
+# Copyright (c) 2017 Sony Corporation. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+# DO NOT EDIT THIS FILE!
+# THIS FILE IS GENERATED BY CODE GENERATOR BUT ITS COMMITED INTO
+# SOURCE TREE TO MAKE FUNCTION ID PERSIST.
+
+Affine_i: 0
+Convolution_iiIiIiIi: 1
+DepthwiseConvolution_iiIiIiIi: 2
+Deconvolution_iiIiIiIi: 3
+DepthwiseDeconvolution_iiIiIiIi: 4
+MaxPooling_iIiIBiI: 5
+AveragePooling_iIiIBiIB: 6
+GlobalAveragePooling: 7
+SumPooling_iIiIBiI: 8
+Unpooling_iI: 9
+Embed: 10
+Sigmoid: 11
+Swish: 12
+Tanh: 13
+ReLU_B: 14
+LeakyReLU_f: 15
+Softmax_i: 16
+ELU_f: 17
+SELU_ff: 18
+CReLU_i: 19
+CELU_fi: 20
+PReLU_i: 21
+BatchNormalization_iIffB: 22
+MeanSubtraction_iB: 23
+Sum_iIB: 24
+Mean_iIB: 25
+Max_iIB: 26
+Min_iIB: 27
+Prod_iIB: 28
+ReduceSum: 29
+ReduceMean: 30
+Add2_B: 31
+BcAdd2: 32
+Sub2: 33
+Mul2: 34
+Div2: 35
+Pow2: 36
+AddScalar_f: 37
+MulScalar_f: 38
+PowScalar_f: 39
+RSubScalar_f: 40
+RDivScalar_f: 41
+RPowScalar_f: 42
+Sign_f: 43
+Minimum2: 44
+Maximum2: 45
+MinimumScalar_f: 46
+MaximumScalar_f: 47
+LogicalAnd: 48
+LogicalOr: 49
+LogicalXor: 50
+Equal: 51
+NotEqual: 52
+GreaterEqual: 53
+Greater: 54
+LessEqual: 55
+Less: 56
+LogicalAndScalar_B: 57
+LogicalOrScalar_B: 58
+LogicalXorScalar_B: 59
+EqualScalar_f: 60
+NotEqualScalar_f: 61
+GreaterEqualScalar_f: 62
+GreaterScalar_f: 63
+LessEqualScalar_f: 64
+LessScalar_f: 65
+LogicalNot: 66
+Constant_fiI: 67
+Abs: 68
+Exp: 69
+Log: 70
+Identity: 71
+BatchMatmul_BB: 72
+Round_: 73
+Concatenate_i: 74
+Split_i: 75
+Stack_i: 76
+Slice_iIiIiI: 77
+Transpose_iI: 78
+Broadcast_iI: 79
+OneHot_iI: 80
+Flip_iI: 81
+Shift_iIi: 82
+Reshape_iI: 83
+MatrixDiag: 84
+MatrixDiagPart: 85
+Dropout_fi: 86
+TopKData_iBBi: 87
+TopKGrad_iBi: 88
+Rand_ffiIi: 89
+Randint_iiiIi: 90
+Randn_ffiIi: 91
+RandomCrop_iIii: 92
+RandomFlip_iIii: 93
+RandomShift_iIiii: 94
+ImageAugmentation_iIiIfffffBBfBffBfi: 95
+SigmoidCrossEntropy: 96
+BinaryCrossEntropy: 97
+SoftmaxCrossEntropy_i: 98
+CategoricalCrossEntropy_i: 99
+SquaredError: 100
+AbsoluteError: 101
+HuberLoss_f: 102
+EpsilonInsensitiveLoss_f: 103
+KLMultinomial_i: 104
+BinarySigmoid: 105
+BinaryTanh: 106
+BinaryConnectAffine_i: 107
+BinaryConnectConvolution_iiIiIiIi: 108
+BinaryWeightAffine_i: 109
+BinaryWeightConvolution_iiIiIiIi: 110
+INQAffine_iiiIii: 111
+INQConvolution_iiIiIiIiiiIii: 112
+FixedPointQuantize_BifB: 113
+Pow2Quantize_BBiiB: 114
+TopNError_ii: 115
+BinaryError: 116
+ConfusionMatrix_i: 117
+VATNoise_if: 118
+Unlink: 119
+Sink_B: 120

python/src/nnabla/utils/converter/nnablart/nnb.py

@@ -50,16 +50,8 @@ def __init__(self, nnp, batch_size):
             if 'arguments' in func and len(func['arguments']) > 0:
                 argfmt = ''
                 for an, arg in func['arguments'].items():
-                    if arg['type'] == 'bool':
-                        argfmt += 'B'
-                    elif arg['type'] == 'double' or arg['type'] == 'float':
-                        argfmt += 'f'
-                    elif arg['type'] == 'int64':
-                        argfmt += 'i'
-                    elif arg['type'] == 'repeated int64' or arg['type'] == 'Shape':
-                        argfmt += 'iI'
-                    elif arg['type'] == 'string':
-                        argfmt += 'i'
+                    argfmt += nnabla.utils.converter.type_to_pack_format(
+                        arg['type'])
                 self._argument_formats[fn] = argfmt
 
     def export(self, *args):
@@ -133,9 +125,13 @@ def export(self, *args):
             # Functions
             findexes = []
             for n, f in enumerate(self._info._network.function):
+
                 function_data = struct.pack(
                     'I', list(self._info._function_info.keys()).index(f.type))
 
+                # Default function implementation is 0(float)
+                function_data += struct.pack('I', 0)
+
                 finfo = self._info._function_info[f.type]
 
                 inputs = [vindexes_by_name[i] for i in f.input]
@@ -154,29 +150,26 @@ def export(self, *args):
                     for an, arg in finfo['arguments'].items():
                         val = eval('f.{}_param.{}'.format(
                             finfo['snake_name'], an))
+
+                        argfmt += nnabla.utils.converter.type_to_pack_format(
+                            arg['type'])
                         if arg['type'] == 'bool':
-                            argfmt += 'B'
                             values.append(val)
                         elif arg['type'] == 'double' or arg['type'] == 'float':
-                            argfmt += 'f'
                             values.append(val)
                         elif arg['type'] == 'int64':
-                            argfmt += 'i'
                             values.append(val)
                         elif arg['type'] == 'repeated int64':
                             index, pointer = self._alloc(
                                 data=struct.pack('{}i'.format(len(val)), *val))
                             values.append(len(val))
                             values.append(index)
-                            argfmt += 'iI'
                         elif arg['type'] == 'Shape':
-                            argfmt += 'iI'
                             index, pointer = self._alloc(data=struct.pack(
                                 '{}i'.format(len(val.dim)), *val.dim))
                             values.append(len(val.dim))
                             values.append(index)
                         elif arg['type'] == 'string':
-                            argfmt += 'i'
                             val = arg['available_values'].index(val)
                             values.append(val)
                     function_data += struct.pack(argfmt, *values)

python/src/nnabla/utils/converter/utils.py

@@ -15,11 +15,27 @@
 
 from collections import OrderedDict
 import json
+from os.path import abspath, join, dirname
 import os
 import yaml
 import zlib
 
 
+def type_to_pack_format(typestring):
+    fmt = None
+    if typestring == 'bool':
+        fmt = 'B'
+    elif typestring == 'double' or typestring == 'float':
+        fmt = 'f'
+    elif typestring == 'int64':
+        fmt = 'i'
+    elif typestring == 'repeated int64' or typestring == 'Shape':
+        fmt = 'iI'
+    elif typestring == 'string':
+        fmt = 'i'
+    return fmt
+
+
 def represent_odict(dumper, instance):
     return dumper.represent_mapping('tag:yaml.org,2002:map', instance.items())
 
@@ -44,8 +60,37 @@ def construct_mapping(loader, node):
 
 
 def get_category_info_string():
-    with open(os.path.join(os.path.dirname(os.path.abspath(__file__)), 'functions.yaml')) as f:
-        return f.read()
+    order = load_yaml_ordered(
+        open(join(dirname(abspath(__file__)), 'function_order.yaml'), 'r'))
+    string = open(join(dirname(abspath(__file__)),
+                       'functions.yaml'), 'r').read()
+    info = load_yaml_ordered(string)
+    for cat, cat_info in info.items():
+        for func, func_info in cat_info.items():
+            if 'arguments' in func_info:
+                for a, a_info in func_info['arguments'].items():
+                    if 'default' in a_info:
+                        a_info.pop('default')
+                    if 'doc' in a_info:
+                        a_info.pop('doc')
+            for n, n_info in func_info['inputs'].items():
+                if 'doc' in n_info:
+                    n_info.pop('doc')
+            for n, n_info in func_info['outputs'].items():
+                if 'doc' in n_info:
+                    n_info.pop('doc')
+            func_info.pop('doc')
+
+    for cat, cat_info in info.items():
+        for func, func_info in cat_info.items():
+            fmt = ''
+            if 'arguments' in func_info:
+                fmt = '_'
+                for a, a_info in func_info['arguments'].items():
+                    fmt += type_to_pack_format(a_info['type'])
+            func_info['uniq_name'] = func + fmt
+            func_info['id'] = order[func+fmt]
+    return yaml.dump(info, default_flow_style=False)
 
 
 def get_category_info_version():