onnx.cpp 27.4 KB
Newer Older
Paul's avatar
Paul committed
1
2
3
4
5
6
7
8
#include <google/protobuf/text_format.h>
#include <google/protobuf/io/zero_copy_stream_impl.h>
#include <onnx.pb.h>
#include <iostream>
#include <fstream>
#include <unordered_map>
#include <functional>
#include <array>
Paul's avatar
Paul committed
9
#include <utility>
10
#include <vector>
Paul's avatar
Paul committed
11

Paul's avatar
Paul committed
12
13
14
15
16
17
18
19
#include <migraphx/fallthrough.hpp>
#include <migraphx/program.hpp>
#include <migraphx/operators.hpp>
#include <migraphx/ranges.hpp>
#include <migraphx/instruction.hpp>
#include <migraphx/config.hpp>

namespace migraphx {
20
inline namespace MIGRAPH_INLINE_NS {
Paul's avatar
Paul committed
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
struct unknown
{
    std::string op;
    std::string name() const { return "unknown:" + op; }
    shape compute_shape(std::vector<shape> input) const
    {
        if(input.empty())
            return {};
        else
            return input.front();
    }
    friend std::ostream& operator<<(std::ostream& os, const unknown& x)
    {
        os << x.name();
        return os;
    }
};

struct onnx_parser
{
    using attribute_map = std::unordered_map<std::string, onnx::AttributeProto>;
    using node_map      = std::unordered_map<std::string, onnx::NodeProto>;
Paul's avatar
Paul committed
43
    using op_func = std::function<instruction_ref(attribute_map, std::vector<instruction_ref>)>;
Paul's avatar
Paul committed
44
45
46
47
48
49
50
51
    node_map nodes;
    std::unordered_map<std::string, instruction_ref> instructions;
    program prog = program();

    std::unordered_map<std::string, op_func> ops;

    onnx_parser()
    {
Shucai Xiao's avatar
Shucai Xiao committed
52
        add_generic_op("MatMul", op::dot{});
Khalique's avatar
Khalique committed
53
        add_generic_op("Relu", op::relu{});
Khalique's avatar
Khalique committed
54
55
56
        add_generic_op("Sigmoid", op::sigmoid{});
        add_generic_op("Tanh", op::tanh{});
        add_generic_op("Abs", op::abs{});
Khalique's avatar
Khalique committed
57
58
        // disable dropout for inference
        add_generic_op("Dropout", op::identity{});
Khalique's avatar
Khalique committed
59
        add_generic_op("Identity", op::identity{});
Shucai Xiao's avatar
Shucai Xiao committed
60
61
62
        add_generic_op("Sin", op::sin{});
        add_generic_op("Cos", op::cos{});
        add_generic_op("Tan", op::tan{});
Paul's avatar
Paul committed
63

64
65
66
67
68
        add_broadcastable_binary_op("Add", op::add{});
        add_broadcastable_binary_op("Div", op::div{});
        add_broadcastable_binary_op("Mul", op::mul{});
        add_broadcastable_binary_op("Sub", op::sub{});
        add_broadcastable_binary_op("Sum", op::add{});
Paul's avatar
Paul committed
69

Khalique's avatar
Khalique committed
70
        add_mem_op("ImageScaler", &onnx_parser::parse_imagescaler);
71
        add_mem_op("LeakyRelu", &onnx_parser::parse_leaky_relu);
Khalique's avatar
Khalique committed
72
        add_mem_op("Elu", &onnx_parser::parse_elu);
Paul's avatar
Paul committed
73
74
        add_mem_op("Constant", &onnx_parser::parse_constant);
        add_mem_op("Conv", &onnx_parser::parse_conv);
Paul's avatar
Paul committed
75
76
        add_mem_op("MaxPool", &onnx_parser::parse_pooling);
        add_mem_op("AveragePool", &onnx_parser::parse_pooling);
77
78
        add_mem_op("GlobalMaxPool", &onnx_parser::parse_pooling);
        add_mem_op("GlobalAveragePool", &onnx_parser::parse_pooling);
Paul's avatar
Paul committed
79
        add_mem_op("Reshape", &onnx_parser::parse_reshape);
Paul's avatar
Paul committed
80
81
        add_mem_op("Flatten", &onnx_parser::parse_flatten);
        add_mem_op("Gemm", &onnx_parser::parse_gemm);
82
        add_mem_op("BatchNormalization", &onnx_parser::parse_batchnorm);
Paul's avatar
Paul committed
83
        add_mem_op("Softmax", &onnx_parser::parse_softmax);
84
85
86
        add_mem_op("Squeeze", &onnx_parser::parse_squeeze);
        add_mem_op("Unsqueeze", &onnx_parser::parse_unsqueeze);
        add_mem_op("Slice", &onnx_parser::parse_slice);
Scott Thornton's avatar
Scott Thornton committed
87
        add_mem_op("Concat", &onnx_parser::parse_concat);
Khalique's avatar
Khalique committed
88
        add_mem_op("Transpose", &onnx_parser::parse_transpose);
Paul's avatar
Paul committed
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
    }

    template <class F>
    void add_op(std::string name, F f)
    {
        ops.emplace(name, f);
    }

    template <class F>
    void add_mem_op(std::string name, F f)
    {
        ops.emplace(name, [=](auto&&... xs) {
            return std::mem_fn(f)(*this, name, std::forward<decltype(xs)>(xs)...);
        });
    }
104
105
106
107
    template <class T>
    void add_broadcastable_binary_op(std::string name, T x)
    {
        ops.emplace(name, [this, x](attribute_map attributes, std::vector<instruction_ref> args) {
Scott Thornton's avatar
Scott Thornton committed
108
            if(args.size() != 2)
Paul's avatar
Paul committed
109
                MIGRAPH_THROW("binary operators should have 2 operands");
110
111
112
113
114
115
116
117
118
119
120
121
            if(contains(attributes, "broadcast"))
            {
                uint64_t broadcasted = parse_value(attributes.at("broadcast")).at<uint64_t>();
                if(broadcasted != 0)
                {
                    uint64_t axis = (contains(attributes, "axis"))
                                        ? parse_value(attributes.at("axis")).at<uint64_t>()
                                        : 0;
                    auto l =
                        prog.add_instruction(op::broadcast{axis, args[0]->get_shape()}, args[1]);
                    return prog.add_instruction(x, args[0], l);
                }
122
                return prog.add_instruction(x, args);
123
            }
124
            else if(args[0]->get_shape() != args[1]->get_shape())
125
126
127
128
            {
                // Example:
                // s0 = (3,2,4,5) and s1 = (2,1,1)
                //
Scott Thornton's avatar
Scott Thornton committed
129
130
                // In this case we need to broadcast (:,1,1) portion of
                // s1 plus broadcast the 1st dimension of s1
131
132
133
134
135
136
137
138
139
                // giving output_lens = (3,2,4,5)
                //
                // Another example:
                // s0 = (3,2,1,5) and s1 = (2,7,5)
                // In this case we need to broadcast the (:,:,1:,:) axis
                // of s0 plus the 1st dimension of s1 giving
                // output_lens = (3,2,7,5)
                //
                // Get lengths for both arguments
Paul's avatar
Paul committed
140
141
142
143
144
145
                const std::vector<std::size_t>* s0 = &args[0]->get_shape().lens();
                const std::vector<std::size_t>* s1 = &args[1]->get_shape().lens();

                // Make sure s0 is the smaller size
                if(s0->size() > s1->size())
                    std::swap(s0, s1);
146
147

                // Copy the larger vector to output_lens
Paul's avatar
Paul committed
148
149
                std::vector<std::size_t> output_lens(s1->size());
                auto offset = s1->size() - s0->size();
Paul's avatar
Paul committed
150
151
152
153
154
155
                std::transform(s0->begin(),
                               s0->end(),
                               s1->begin() + offset,
                               output_lens.begin() + offset,
                               [](auto a, auto b) { return std::max(a, b); });

156
157
158
                auto l0 = prog.add_instruction(op::multibroadcast{output_lens}, args[0]);
                auto l1 = prog.add_instruction(op::multibroadcast{output_lens}, args[1]);
                return prog.add_instruction(x, l0, l1);
Paul's avatar
Paul committed
159
160
            }
            else
161
162
            {
                return prog.add_instruction(x, args);
163
164
165
166
            }
        });
    }

Paul's avatar
Paul committed
167
    template <class T>
Paul's avatar
Paul committed
168
169
    void add_generic_op(std::string name, T x)
    {
Paul's avatar
Paul committed
170
        ops.emplace(name, [this, x](attribute_map, std::vector<instruction_ref> args) {
Paul's avatar
Paul committed
171
172
173
174
            return prog.add_instruction(x, args);
        });
    }

Paul's avatar
Paul committed
175
    instruction_ref
Paul's avatar
Paul committed
176
    parse_softmax(const std::string&, const attribute_map&, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
177
178
    {
        auto dims = args.front()->get_shape().lens();
Scott Thornton's avatar
Scott Thornton committed
179
180
        auto r =
            prog.add_instruction(op::reshape{{long(dims[0]), long(dims[1]), 1, 1}}, args.front());
181
182
        auto s = prog.add_instruction(op::softmax{}, r);
        return prog.add_instruction(op::reshape{{long(dims[0]), long(dims[1])}}, s);
Paul's avatar
Paul committed
183
184
    }

Paul's avatar
Paul committed
185
    instruction_ref
Paul's avatar
Paul committed
186
    parse_conv(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
187
    {
188
        op::convolution op;
Paul's avatar
Paul committed
189
190
191
        if(contains(attributes, "pads"))
        {
            copy(attributes["pads"].ints(), op.padding.begin());
Paul's avatar
Paul committed
192
        }
Paul's avatar
Paul committed
193
194
195
196
197
198
199
200
201
202
203
204
        if(contains(attributes, "strides"))
        {
            copy(attributes["strides"].ints(), op.stride.begin());
        }
        if(contains(attributes, "dilations"))
        {
            copy(attributes["dilations"].ints(), op.dilation.begin());
        }
        if(args.size() == 3)
        {
            uint64_t axis = 1;
            auto l1       = prog.add_instruction(op, args[0], args[1]);
Scott Thornton's avatar
Scott Thornton committed
205
            auto l2       = prog.add_instruction(op::broadcast{axis, l1->get_shape()}, args[2]);
206
            return prog.add_instruction(op::add{}, l1, l2);
Paul's avatar
Paul committed
207
        }
Paul's avatar
Paul committed
208
209
        return prog.add_instruction(op, args);
    }
Paul's avatar
Paul committed
210

Paul's avatar
Paul committed
211
212
213
    instruction_ref parse_pooling(const std::string& name,
                                  attribute_map attributes,
                                  std::vector<instruction_ref> args)
Paul's avatar
Paul committed
214
    {
Khalique's avatar
Khalique committed
215
216
        op::pooling op{ends_with(name, "MaxPool") ? "max" : "average"};
        if(starts_with(name, "Global"))
217
        {
Khalique's avatar
Khalique committed
218
219
            auto lens  = args.front()->get_shape().lens();
            op.lengths = {lens[2], lens[3]};
220
        }
Paul's avatar
Paul committed
221
222
223
224
225
226
227
228
229
230
231
232
        if(contains(attributes, "pads"))
        {
            copy(attributes["pads"].ints(), op.padding.begin());
        }
        if(contains(attributes, "strides"))
        {
            copy(attributes["strides"].ints(), op.stride.begin());
        }
        if(contains(attributes, "kernel_shape"))
        {
            copy(attributes["kernel_shape"].ints(), op.lengths.begin());
        }
Paul's avatar
Paul committed
233
        return prog.add_instruction(op, std::move(args));
Paul's avatar
Paul committed
234
235
    }

Paul's avatar
Paul committed
236
    instruction_ref
Paul's avatar
Paul committed
237
    parse_reshape(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
238
    {
239
        op::reshape op;
Paul's avatar
Paul committed
240
241
242
243
244
245
246
        if(args.size() == 1)
        {
            literal s = parse_value(attributes.at("shape"));
            s.visit([&](auto v) { copy(v, std::back_inserter(op.dims)); });
        }
        if(args.size() == 2)
        {
Paul's avatar
Paul committed
247
            literal s = args[1]->get_literal();
Paul's avatar
Paul committed
248
            s.visit([&](auto v) { copy(v, std::back_inserter(op.dims)); });
Paul's avatar
Paul committed
249
        }
Paul's avatar
Paul committed
250
251
252
        return prog.add_instruction(op, args[0]);
    }

Paul's avatar
Paul committed
253
    instruction_ref
Paul's avatar
Paul committed
254
    parse_flatten(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
255
256
    {
        uint64_t axis = 0;
Paul's avatar
Paul committed
257
258
259
260
        if(contains(attributes, "axis"))
        {
            axis = parse_value(attributes.at("axis")).at<int>();
        }
261
        return prog.add_instruction(op::flatten{axis}, args[0]);
Paul's avatar
Paul committed
262
263
    }

264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
    instruction_ref
    parse_squeeze(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
        op::squeeze op;
        literal s = parse_value(attributes.at("axes"));
        s.visit([&](auto v) { copy(v, std::back_inserter(op.axes)); });
        return prog.add_instruction(op, args[0]);
    }

    instruction_ref
    parse_unsqueeze(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
        op::unsqueeze op;
        literal s = parse_value(attributes.at("axes"));
        s.visit([&](auto v) { copy(v, std::back_inserter(op.axes)); });
        return prog.add_instruction(op, args[0]);
    }

Scott Thornton's avatar
Scott Thornton committed
282
283
284
285
286
287
288
    instruction_ref
    parse_concat(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
        std::size_t axis = parse_value(attributes.at("axis")).at<int>();
        op::concat op{axis};
        return prog.add_instruction(op, std::move(args));
    }
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309

    instruction_ref
    parse_slice(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
        op::slice op;
        if(contains(attributes, "axes"))
        {
            literal s = parse_value(attributes.at("axes"));
            s.visit([&](auto v) { copy(v, std::back_inserter(op.axes)); });
        }
        {
            literal s = parse_value(attributes.at("ends"));
            s.visit([&](auto v) { copy(v, std::back_inserter(op.ends)); });
        }
        {
            literal s = parse_value(attributes.at("starts"));
            s.visit([&](auto v) { copy(v, std::back_inserter(op.starts)); });
        }
        return prog.add_instruction(op, args[0]);
    }

Paul's avatar
Paul committed
310
311
312
    instruction_ref parse_constant(const std::string&,
                                   attribute_map attributes,
                                   const std::vector<instruction_ref>&)
Paul's avatar
Paul committed
313
314
315
316
    {
        literal v = parse_value(attributes.at("value"));
        return prog.add_literal(v);
    }
Paul's avatar
Paul committed
317

Paul's avatar
Paul committed
318
    instruction_ref
Paul's avatar
Paul committed
319
    parse_gemm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
    {
        float alpha = 1.0f;
        float beta  = 0.0f;
        bool transa = false;
        bool transb = false;
        if(contains(attributes, "alpha"))
        {
            alpha = parse_value(attributes.at("alpha")).at<float>();
        }
        if(contains(attributes, "beta"))
        {
            alpha = parse_value(attributes.at("beta")).at<float>();
        }
        if(contains(attributes, "transA"))
        {
            transa = parse_value(attributes.at("transA")).at<bool>();
        }
        if(contains(attributes, "transB"))
        {
            transb = parse_value(attributes.at("transB")).at<bool>();
        }
        std::vector<int64_t> perm = {1, 0};
342
343
        auto l1 = (transa) ? prog.add_instruction(op::transpose{perm}, args[0]) : args[0];
        auto l2 = (transb) ? prog.add_instruction(op::transpose{perm}, args[1]) : args[1];
Paul's avatar
Paul committed
344
345
346
        if(args.size() == 3)
        {
            uint64_t axis = 1;
Shucai Xiao's avatar
Shucai Xiao committed
347
            auto l3       = prog.add_instruction(op::dot{alpha, beta}, l1, l2);
Scott Thornton's avatar
Scott Thornton committed
348
            auto l4       = prog.add_instruction(op::broadcast{axis, l3->get_shape()}, args[2]);
349
            return prog.add_instruction(op::add{}, l3, l4);
Paul's avatar
Paul committed
350
        }
Shucai Xiao's avatar
Shucai Xiao committed
351
        return prog.add_instruction(op::dot{alpha, beta}, l1, l2);
Paul's avatar
Paul committed
352
353
    }

354
    instruction_ref
Paul's avatar
Paul committed
355
    parse_batchnorm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
356
    {
Scott Thornton's avatar
Scott Thornton committed
357
358
        float epsilon                                     = 1e-5f;
        float momentum                                    = 0.9f;
359
        op::batch_norm_inference::bn_infer_mode_t bn_mode = op::batch_norm_inference::spatial;
Scott Thornton's avatar
Scott Thornton committed
360
        bool is_test                                      = false;
361
362
363
364
365
366
        if(contains(attributes, "epsilon"))
        {
            epsilon = parse_value(attributes.at("epsilon")).at<float>();
        }
        if(contains(attributes, "momentum"))
        {
367
            momentum = parse_value(attributes.at("momentum")).at<float>();
368
369
370
        }
        if(contains(attributes, "is_test"))
        {
wsttiger's avatar
wsttiger committed
371
            is_test = parse_value(attributes.at("is_test")).at<uint64_t>() > 0;
372
373
374
        }
        if(contains(attributes, "spatial"))
        {
375
            bn_mode = (parse_value(attributes.at("spatial")).at<uint64_t>() > 0)
376
377
                          ? op::batch_norm_inference::spatial
                          : op::batch_norm_inference::per_activation;
378
        }
Paul's avatar
Paul committed
379
        (void)is_test;
Paul's avatar
Paul committed
380
        op::batch_norm_inference op{epsilon, momentum, bn_mode};
Paul's avatar
Paul committed
381
        return prog.add_instruction(op, std::move(args));
382
383
    }

384
385
386
387
    instruction_ref parse_leaky_relu(const std::string&,
                                     attribute_map attributes,
                                     std::vector<instruction_ref> args)
    {
Khalique's avatar
Khalique committed
388
        float alpha = 0.01; // default alpha val for leaky relu
389
390
391
392
393
394
395
396
        if(contains(attributes, "alpha"))
        {
            alpha = parse_value(attributes.at("alpha")).at<float>();
        }
        op::leaky_relu op{alpha};
        return prog.add_instruction(op, args.front());
    }

Khalique's avatar
Khalique committed
397
398
    instruction_ref
    parse_elu(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
399
400
401
402
403
404
405
406
407
408
    {
        float alpha = 1.0; // default alpha val for elu
        if(contains(attributes, "alpha"))
        {
            alpha = parse_value(attributes.at("alpha")).at<float>();
        }
        op::elu op{alpha};
        return prog.add_instruction(op, args.front());
    }

Khalique's avatar
Khalique committed
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
    instruction_ref parse_imagescaler(const std::string&,
                                      attribute_map attributes,
                                      std::vector<instruction_ref> args)
    {
        float scale = 1.0;
        std::vector<float> bias{};
        if(contains(attributes, "scale"))
        {
            scale = parse_value(attributes.at("scale")).at<float>();
        }

        if(contains(attributes, "bias"))
        {
            auto&& bias_floats = attributes["bias"].floats();
            bias               = std::vector<float>(bias_floats.begin(), bias_floats.end());
        }
        auto input_shape = args.front()->get_shape();
Khalique's avatar
Khalique committed
426

Khalique's avatar
Khalique committed
427
428
        auto scale_val = prog.add_literal(scale);
        auto bias_vals = prog.add_literal(
Paul's avatar
Paul committed
429
            migraphx::literal{migraphx::shape{migraphx::shape::float_type, {bias.size()}}, bias});
Khalique's avatar
Khalique committed
430

Paul's avatar
Paul committed
431
432
        auto scale_tensor = prog.add_instruction(migraphx::op::scalar{input_shape}, scale_val);
        auto img_scaled   = prog.add_instruction(migraphx::op::mul{}, args.front(), scale_tensor);
Paul's avatar
Paul committed
433
        auto bias_bcast = prog.add_instruction(migraphx::op::broadcast{1, input_shape}, bias_vals);
Paul's avatar
Paul committed
434
        return prog.add_instruction(migraphx::op::add{}, img_scaled, bias_bcast);
Khalique's avatar
Khalique committed
435
    }
Khalique's avatar
Khalique committed
436

Khalique's avatar
Khalique committed
437
438
    instruction_ref
    parse_transpose(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
439
440
441
442
443
444
445
    {
        std::vector<int64_t> perm{};
        if(contains(attributes, "perm"))
        {
            auto&& perm_vals = attributes["perm"].ints();
            perm             = std::vector<int64_t>(perm_vals.begin(), perm_vals.end());
        }
Paul's avatar
Paul committed
446
        return prog.add_instruction(migraphx::op::transpose{perm}, args.front());
Khalique's avatar
Khalique committed
447
448
    }

Paul's avatar
Paul committed
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
    void parse_from(std::istream& is)
    {
        onnx::ModelProto model;
        if(model.ParseFromIstream(&is))
        {
            if(model.has_graph())
            {
                this->parse_graph(model.graph());
            }
        }
        else
        {
            throw std::runtime_error("Failed reading");
        }
    }

    void parse_graph(const onnx::GraphProto& graph)
    {
        nodes = get_nodes(graph);
468
469
470
471
472
        std::unordered_map<std::string, onnx::TensorProto> initializer_data;
        for(auto&& f : graph.initializer())
        {
            initializer_data[f.name()] = f;
        }
Paul's avatar
Paul committed
473
474
475
        for(auto&& input : graph.input())
        {
            const std::string& name = input.name();
476
477
478
479
480
481
482
483
484
485
486
487
            // Does the input have an initializer?
            if(contains(initializer_data, name))
            {
                auto t             = initializer_data[name];
                instructions[name] = prog.add_literal(parse_tensor(t));
            }
            else
            {
                // TODO: Get shape of input parameter
                shape s            = parse_type(input.type());
                instructions[name] = prog.add_parameter(name, s);
            }
Paul's avatar
Paul committed
488
489
490
        }
        for(auto&& p : nodes)
        {
491
            this->parse_node(get_name(p.second));
Paul's avatar
Paul committed
492
493
494
        }
    }

Paul's avatar
Paul committed
495
    void parse_node(const std::string& name)
Paul's avatar
Paul committed
496
    {
Paul's avatar
Paul committed
497
        if(name.empty())
Paul's avatar
Paul committed
498
            MIGRAPH_THROW("Onnx node must have a name");
Paul's avatar
Paul committed
499
500
501
502
503
504
505
506
        if(instructions.count(name) == 0)
        {
            auto&& node = nodes.at(name);
            std::vector<instruction_ref> args;
            for(auto&& input : node.input())
            {
                if(nodes.count(input) > 0)
                {
507
                    auto&& iname = get_name(nodes.at(input));
Paul's avatar
Paul committed
508
                    assert(name != iname);
Paul's avatar
Paul committed
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
                    this->parse_node(iname);
                    args.push_back(instructions.at(iname));
                }
                else
                {
                    args.push_back(instructions.at(input));
                }
            }
            if(ops.count(node.op_type()) == 0)
            {
                instructions[name] = prog.add_instruction(unknown{node.op_type()}, args);
            }
            else
            {
                instructions[name] = ops[node.op_type()](get_attributes(node), args);
            }
        }
    }

    static attribute_map get_attributes(const onnx::NodeProto& node)
    {
        std::unordered_map<std::string, onnx::AttributeProto> result;
        for(auto&& attr : node.attribute())
        {
            result[attr.name()] = attr;
        }
        return result;
    }

538
539
540
541
    static std::string get_name(const onnx::NodeProto& node)
    {
        if(node.name().empty())
        {
Paul's avatar
Paul committed
542
            std::string generated = "migraphx_unnamed_node";
Paul's avatar
Paul committed
543
544
545
546
            return std::accumulate(node.output().begin(),
                                   node.output().end(),
                                   generated,
                                   [](auto x, auto y) { return x + "_" + y; });
547
548
549
550
        }
        return node.name();
    }

Paul's avatar
Paul committed
551
552
553
554
555
    static node_map get_nodes(const onnx::GraphProto& graph)
    {
        std::unordered_map<std::string, onnx::NodeProto> result;
        for(auto&& node : graph.node())
        {
556
            result[get_name(node)] = node;
Paul's avatar
Paul committed
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
            for(auto&& output : node.output())
            {
                result[output] = node;
            }
        }
        return result;
    }

    template <class T>
    static literal from_repeated(shape::type_t t, const T& r)
    {
        std::size_t size = r.size();
        return literal{{t, {size}}, r.begin(), r.end()};
    }

    static literal parse_value(const onnx::AttributeProto& attr)
    {
        switch(attr.type())
        {
        case onnx::AttributeProto::UNDEFINED: return {};
        case onnx::AttributeProto::FLOAT: return literal{attr.f()};
        case onnx::AttributeProto::INT: return literal{attr.i()};
        case onnx::AttributeProto::STRING: return {};
        case onnx::AttributeProto::TENSOR: return parse_tensor(attr.t());
        case onnx::AttributeProto::GRAPH: return {};
Paul's avatar
Paul committed
582
        case onnx::AttributeProto::FLOATS: return from_repeated(shape::float_type, attr.floats());
Paul's avatar
Paul committed
583
584
585
586
587
        case onnx::AttributeProto::INTS: return from_repeated(shape::int64_type, attr.ints());
        case onnx::AttributeProto::STRINGS: return {};
        case onnx::AttributeProto::TENSORS: return {};
        case onnx::AttributeProto::GRAPHS: return {};
        }
Paul's avatar
Paul committed
588
        MIGRAPH_THROW("Invalid attribute type");
Paul's avatar
Paul committed
589
590
591
592
593
    }

    static literal parse_tensor(const onnx::TensorProto& t)
    {
        std::vector<std::size_t> dims(t.dims().begin(), t.dims().end());
594
595
        if(t.has_raw_data())
        {
wsttiger's avatar
wsttiger committed
596
            const std::string& s = t.raw_data();
Scott Thornton's avatar
Scott Thornton committed
597
598
599
600
601
602
603
604
605
606
607
608
            switch(t.data_type())
            {
            case onnx::TensorProto::UNDEFINED: throw std::runtime_error("");
            case onnx::TensorProto::FLOAT: return literal{{shape::float_type, dims}, s.data()};
            case onnx::TensorProto::UINT8: throw std::runtime_error("");
            case onnx::TensorProto::INT8: return literal{{shape::int32_type, dims}, s.data()};
            case onnx::TensorProto::UINT16: return literal{{shape::int32_type, dims}, s.data()};
            case onnx::TensorProto::INT16: return literal{{shape::int32_type, dims}, s.data()};
            case onnx::TensorProto::INT32: return literal{{shape::int32_type, dims}, s.data()};
            case onnx::TensorProto::INT64: return literal{{shape::int64_type, dims}, s.data()};
            case onnx::TensorProto::STRING: throw std::runtime_error("");
            case onnx::TensorProto::BOOL: return literal{{shape::int32_type, dims}, s.data()};
Paul's avatar
Paul committed
609
            case onnx::TensorProto::FLOAT16: return literal{{shape::half_type, dims}, s.data()};
Scott Thornton's avatar
Scott Thornton committed
610
611
612
613
614
615
616
            case onnx::TensorProto::DOUBLE: return literal{{shape::double_type, dims}, s.data()};
            case onnx::TensorProto::UINT32: throw std::runtime_error("");
            case onnx::TensorProto::UINT64: throw std::runtime_error("");
            case onnx::TensorProto::COMPLEX64: throw std::runtime_error("");
            case onnx::TensorProto::COMPLEX128: throw std::runtime_error("");
            }
            MIGRAPH_THROW("Invalid tensor type");
617
        }
Paul's avatar
Paul committed
618
619
620
621
        switch(t.data_type())
        {
        case onnx::TensorProto::UNDEFINED: throw std::runtime_error("");
        case onnx::TensorProto::FLOAT:
Paul's avatar
Paul committed
622
            return literal{{shape::float_type, dims}, t.float_data().begin(), t.float_data().end()};
Paul's avatar
Paul committed
623
624
        case onnx::TensorProto::UINT8: throw std::runtime_error("");
        case onnx::TensorProto::INT8:
Paul's avatar
Paul committed
625
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
626
        case onnx::TensorProto::UINT16:
Paul's avatar
Paul committed
627
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
628
        case onnx::TensorProto::INT16:
Paul's avatar
Paul committed
629
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
630
        case onnx::TensorProto::INT32:
Paul's avatar
Paul committed
631
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
632
        case onnx::TensorProto::INT64:
Paul's avatar
Paul committed
633
            return literal{{shape::int64_type, dims}, t.int64_data().begin(), t.int64_data().end()};
Paul's avatar
Paul committed
634
635
        case onnx::TensorProto::STRING: throw std::runtime_error("");
        case onnx::TensorProto::BOOL:
Paul's avatar
Paul committed
636
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
637
638
        case onnx::TensorProto::FLOAT16:
            return literal{{shape::half_type, dims}, t.float_data().begin(), t.float_data().end()};
Paul's avatar
Paul committed
639
640
641
642
643
644
645
646
        case onnx::TensorProto::DOUBLE:
            return literal{
                {shape::double_type, dims}, t.double_data().begin(), t.double_data().end()};
        case onnx::TensorProto::UINT32: throw std::runtime_error("");
        case onnx::TensorProto::UINT64: throw std::runtime_error("");
        case onnx::TensorProto::COMPLEX64: throw std::runtime_error("");
        case onnx::TensorProto::COMPLEX128: throw std::runtime_error("");
        }
Paul's avatar
Paul committed
647
        MIGRAPH_THROW("Invalid tensor type");
Paul's avatar
Paul committed
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
    }

    static shape parse_type(const onnx::TypeProto& t)
    {
        shape::type_t shape_type{};
        switch(t.tensor_type().elem_type())
        {
        case onnx::TensorProto::UNDEFINED:
            break; // throw std::runtime_error("Unsupported type UNDEFINED");
        case onnx::TensorProto::FLOAT: shape_type = shape::float_type; break;
        case onnx::TensorProto::UINT8:
            break; // throw std::runtime_error("Unsupported type UINT8");
        case onnx::TensorProto::INT8: shape_type = shape::int8_type; break;
        case onnx::TensorProto::UINT16: shape_type = shape::uint16_type; break;
        case onnx::TensorProto::INT16: shape_type = shape::int16_type; break;
        case onnx::TensorProto::INT32: shape_type = shape::int32_type; break;
        case onnx::TensorProto::INT64: shape_type = shape::int64_type; break;
        case onnx::TensorProto::STRING:
            break; // throw std::runtime_error("Unsupported type STRING");
        case onnx::TensorProto::BOOL:
            break; // throw std::runtime_error("Unsupported type BOOL");
Paul's avatar
Paul committed
669
        case onnx::TensorProto::FLOAT16: shape_type = shape::half_type; break;
Paul's avatar
Paul committed
670
671
672
673
674
675
676
677
678
        case onnx::TensorProto::DOUBLE: shape_type = shape::double_type; break;
        case onnx::TensorProto::UINT32: shape_type = shape::uint32_type; break;
        case onnx::TensorProto::UINT64: shape_type = shape::uint64_type; break;
        case onnx::TensorProto::COMPLEX64:
            break; // throw std::runtime_error("Unsupported type COMPLEX64");
        case onnx::TensorProto::COMPLEX128:
            break; // throw std::runtime_error("Unsupported type COMPLEX128");
        }
        std::vector<std::size_t> dims;
Paul's avatar
Paul committed
679
        auto&& tensor_dims = t.tensor_type().shape().dim();
680
681
682
683
684
685
686
687
688
689
690
        std::transform(tensor_dims.begin(),
                       tensor_dims.end(),
                       std::back_inserter(dims),
                       [](auto&& d) -> std::size_t {
                           if(not d.has_dim_value())
                           {
                               long default_batch_size = 1; // FIXME
                               return default_batch_size;
                           }
                           return d.dim_value();
                       });
Paul's avatar
Paul committed
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
        return {shape_type, dims};
    }
};

program parse_onnx(const std::string& name)
{
    std::fstream input(name.c_str(), std::ios::in | std::ios::binary);
    onnx_parser parser;
#ifndef NDEBUG
    // Log the program when it can't be parsed
    try
    {
        parser.parse_from(input);
    }
    catch(...)
    {
        std::cerr << parser.prog << std::endl;
        throw;
    }
#else
    parser.parse_from(input);
#endif
    return std::move(parser.prog);
}

716
} // namespace MIGRAPH_INLINE_NS
Paul's avatar
Paul committed
717
} // namespace migraphx