onnx.cpp 30.6 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 {
Paul's avatar
Paul committed
20
inline namespace MIGRAPHX_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
    node_map nodes;
    std::unordered_map<std::string, instruction_ref> instructions;
Scott Thornton's avatar
Scott Thornton committed
46
    program prog    = program();
47
    bool is_pytorch = false;
Paul's avatar
Paul committed
48
49
50
51
52

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

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

Khalique's avatar
Khalique committed
62
63
64
65
66
67
68
69
        add_binary_op("Add", op::add{});
        add_binary_op("Div", op::div{});
        add_binary_op("Mul", op::mul{});
        add_binary_op("Sub", op::sub{});

        add_mem_op("Sum", &onnx_parser::parse_sum);
        add_mem_op("Max", &onnx_parser::parse_max);
        add_mem_op("Min", &onnx_parser::parse_min);
Paul's avatar
Paul committed
70

Khalique's avatar
Khalique committed
71
        add_mem_op("ImageScaler", &onnx_parser::parse_imagescaler);
72
        add_mem_op("LeakyRelu", &onnx_parser::parse_leaky_relu);
Khalique's avatar
Khalique committed
73
        add_mem_op("Elu", &onnx_parser::parse_elu);
Paul's avatar
Paul committed
74
75
        add_mem_op("Constant", &onnx_parser::parse_constant);
        add_mem_op("Conv", &onnx_parser::parse_conv);
Paul's avatar
Paul committed
76
77
        add_mem_op("MaxPool", &onnx_parser::parse_pooling);
        add_mem_op("AveragePool", &onnx_parser::parse_pooling);
78
79
        add_mem_op("GlobalMaxPool", &onnx_parser::parse_pooling);
        add_mem_op("GlobalAveragePool", &onnx_parser::parse_pooling);
Paul's avatar
Paul committed
80
        add_mem_op("Reshape", &onnx_parser::parse_reshape);
Paul's avatar
Paul committed
81
82
        add_mem_op("Flatten", &onnx_parser::parse_flatten);
        add_mem_op("Gemm", &onnx_parser::parse_gemm);
83
        add_mem_op("BatchNormalization", &onnx_parser::parse_batchnorm);
Paul's avatar
Paul committed
84
        add_mem_op("Softmax", &onnx_parser::parse_softmax);
85
86
87
        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
88
        add_mem_op("Concat", &onnx_parser::parse_concat);
Khalique's avatar
Khalique committed
89
        add_mem_op("Transpose", &onnx_parser::parse_transpose);
Paul's avatar
Paul committed
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
    }

    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)...);
        });
    }
Khalique's avatar
Khalique committed
105

106
    template <class T>
Khalique's avatar
Khalique committed
107
    void add_binary_op(std::string name, T x)
108
109
    {
        ops.emplace(name, [this, x](attribute_map attributes, std::vector<instruction_ref> args) {
Scott Thornton's avatar
Scott Thornton committed
110
            if(args.size() != 2)
Paul's avatar
Paul committed
111
                MIGRAPHX_THROW("binary operators should have 2 operands");
112
113
114
115
116
117
118
119
120
121
122
123
            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);
                }
124
                return prog.add_instruction(x, args);
125
            }
Khalique's avatar
Khalique committed
126
            else
127
            {
Khalique's avatar
Khalique committed
128
129
130
131
132
133
134
135
136
137
                return add_broadcastable_binary_op(args[0], args[1], x);
            }
        });
    }

    template <class T>
    instruction_ref add_broadcastable_binary_op(instruction_ref arg0, instruction_ref arg1, T x)
    {
        if(arg0->get_shape() != arg1->get_shape())
        {
Khalique's avatar
Khalique committed
138
139
140
141
142
143
144
145
146
147
148
149
150
            // Example:
            // s0 = (3,2,4,5) and s1 = (2,1,1)
            //
            // In this case we need to broadcast (:,1,1) portion of
            // s1 plus broadcast the 1st dimension of s1
            // 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)
            //
Khalique's avatar
Khalique committed
151
152
153
154
155
156
157
158
159
160
161
            // Get lengths for both arguments
            const std::vector<std::size_t>* s0 = &arg0->get_shape().lens();
            const std::vector<std::size_t>* s1 = &arg1->get_shape().lens();

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

            std::vector<std::size_t> output_lens(s1->size());
            auto offset = s1->size() - s0->size();
            std::transform(s0->begin(),
Khalique's avatar
Khalique committed
162
163
164
165
                           s0->end(),
                           s1->begin() + offset,
                           output_lens.begin() + offset,
                           [](auto a, auto b) { return std::max(a, b); });
Khalique's avatar
Khalique committed
166
167
168
169
170
171
172
173
174

            auto l0 = prog.add_instruction(op::multibroadcast{output_lens}, arg0);
            auto l1 = prog.add_instruction(op::multibroadcast{output_lens}, arg1);
            return prog.add_instruction(x, l0, l1);
        }
        else
        {
            return prog.add_instruction(x, {arg0, arg1});
        }
175
176
    }

Paul's avatar
Paul committed
177
    template <class T>
Paul's avatar
Paul committed
178
179
    void add_generic_op(std::string name, T x)
    {
Paul's avatar
Paul committed
180
        ops.emplace(name, [this, x](attribute_map, std::vector<instruction_ref> args) {
Paul's avatar
Paul committed
181
182
183
184
            return prog.add_instruction(x, args);
        });
    }

Khalique's avatar
Khalique committed
185
186
187
    instruction_ref
    parse_sum(const std::string&, const attribute_map&, std::vector<instruction_ref> args)
    {
Khalique's avatar
Khalique committed
188
189
190
191
192
193
        return std::accumulate(std::next(args.begin()),
                               args.end(),
                               args.front(),
                               [this](instruction_ref a, instruction_ref b) {
                                   return add_broadcastable_binary_op(a, b, op::add{});
                               });
Khalique's avatar
Khalique committed
194
195
196
197
198
    }

    instruction_ref
    parse_max(const std::string&, const attribute_map&, std::vector<instruction_ref> args)
    {
Khalique's avatar
Khalique committed
199
200
201
202
203
204
        return std::accumulate(std::next(args.begin()),
                               args.end(),
                               args.front(),
                               [this](instruction_ref a, instruction_ref b) {
                                   return add_broadcastable_binary_op(a, b, op::max{});
                               });
Khalique's avatar
Khalique committed
205
206
207
208
209
    }

    instruction_ref
    parse_min(const std::string&, const attribute_map&, std::vector<instruction_ref> args)
    {
Khalique's avatar
Khalique committed
210
211
212
213
214
215
        return std::accumulate(std::next(args.begin()),
                               args.end(),
                               args.front(),
                               [this](instruction_ref a, instruction_ref b) {
                                   return add_broadcastable_binary_op(a, b, op::min{});
                               });
Khalique's avatar
Khalique committed
216
217
    }

Paul's avatar
Paul committed
218
    instruction_ref
Paul's avatar
Paul committed
219
    parse_softmax(const std::string&, const attribute_map&, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
220
221
    {
        auto dims = args.front()->get_shape().lens();
Scott Thornton's avatar
Scott Thornton committed
222
223
        auto r =
            prog.add_instruction(op::reshape{{long(dims[0]), long(dims[1]), 1, 1}}, args.front());
224
225
        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
226
227
    }

Paul's avatar
Paul committed
228
    instruction_ref
Paul's avatar
Paul committed
229
    parse_conv(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
230
    {
231
        op::convolution op;
Paul's avatar
Paul committed
232
233
        if(contains(attributes, "pads"))
        {
Scott Thornton's avatar
Scott Thornton committed
234
            if(contains(attributes, "auto_pad"))
235
            {
Paul's avatar
Paul committed
236
                MIGRAPHX_THROW("auto_pad and padding cannot be specified simultaneously");
237
238
239
            }
            std::vector<std::size_t> padding(4);
            copy(attributes["pads"].ints(), padding.begin());
Scott Thornton's avatar
Scott Thornton committed
240
            if(padding.size() != 4)
241
            {
Paul's avatar
Paul committed
242
                MIGRAPHX_THROW("padding should have 4 values");
243
            }
Scott Thornton's avatar
Scott Thornton committed
244
            if(padding[0] != padding[2] || padding[1] != padding[3])
245
            {
Paul's avatar
Paul committed
246
                MIGRAPHX_THROW("migraphx does not support asymetric padding");
247
248
249
            }
            op.padding[0] = padding[0];
            op.padding[1] = padding[1];
Paul's avatar
Paul committed
250
        }
Paul's avatar
Paul committed
251
252
253
254
255
256
257
258
        if(contains(attributes, "strides"))
        {
            copy(attributes["strides"].ints(), op.stride.begin());
        }
        if(contains(attributes, "dilations"))
        {
            copy(attributes["dilations"].ints(), op.dilation.begin());
        }
Scott Thornton's avatar
Scott Thornton committed
259
        if(contains(attributes, "auto_pad"))
260
261
        {
            auto s = attributes["auto_pad"].s();
Scott Thornton's avatar
Scott Thornton committed
262
            if(contains(attributes, "pads") and to_upper(s) != "NOTSET")
263
            {
Paul's avatar
Paul committed
264
                MIGRAPHX_THROW("auto_pad and padding cannot be specified simultaneously");
265
266
            }

wsttiger's avatar
fixes  
wsttiger committed
267
            if(s.find("SAME") != std::string::npos)
268
269
270
271
            {
                op.padding_mode = op::convolution::same;
            }
        }
Paul's avatar
Paul committed
272
273
274
275
        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
276
            auto l2       = prog.add_instruction(op::broadcast{axis, l1->get_shape()}, args[2]);
277
            return prog.add_instruction(op::add{}, l1, l2);
Paul's avatar
Paul committed
278
        }
Paul's avatar
Paul committed
279
280
        return prog.add_instruction(op, args);
    }
Paul's avatar
Paul committed
281

Paul's avatar
Paul committed
282
283
284
    instruction_ref parse_pooling(const std::string& name,
                                  attribute_map attributes,
                                  std::vector<instruction_ref> args)
Paul's avatar
Paul committed
285
    {
Khalique's avatar
Khalique committed
286
287
        op::pooling op{ends_with(name, "MaxPool") ? "max" : "average"};
        if(starts_with(name, "Global"))
288
        {
Khalique's avatar
Khalique committed
289
290
            auto lens  = args.front()->get_shape().lens();
            op.lengths = {lens[2], lens[3]};
291
        }
Paul's avatar
Paul committed
292
293
        if(contains(attributes, "pads"))
        {
294
295
            std::vector<std::size_t> padding(4);
            copy(attributes["pads"].ints(), padding.begin());
Scott Thornton's avatar
Scott Thornton committed
296
            if(padding.size() != 4)
297
            {
Paul's avatar
Paul committed
298
                MIGRAPHX_THROW("padding should have 4 values");
299
            }
Scott Thornton's avatar
Scott Thornton committed
300
            if(padding[0] != padding[2] || padding[1] != padding[3])
301
            {
Paul's avatar
Paul committed
302
                MIGRAPHX_THROW("migraphx does not support asymetric padding");
303
304
305
            }
            op.padding[0] = padding[0];
            op.padding[1] = padding[1];
Paul's avatar
Paul committed
306
307
308
309
310
311
312
313
314
        }
        if(contains(attributes, "strides"))
        {
            copy(attributes["strides"].ints(), op.stride.begin());
        }
        if(contains(attributes, "kernel_shape"))
        {
            copy(attributes["kernel_shape"].ints(), op.lengths.begin());
        }
Scott Thornton's avatar
Scott Thornton committed
315
        if(contains(attributes, "auto_pad"))
316
317
        {
            auto s = attributes["auto_pad"].s();
Scott Thornton's avatar
Scott Thornton committed
318
            if(to_upper(s) != "NOTSET")
319
            {
Paul's avatar
Paul committed
320
                MIGRAPHX_THROW("auto_pad is not supported for pooling");
321
322
323
            }
        }

Paul's avatar
Paul committed
324
        return prog.add_instruction(op, std::move(args));
Paul's avatar
Paul committed
325
326
    }

Paul's avatar
Paul committed
327
    instruction_ref
Paul's avatar
Paul committed
328
    parse_reshape(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
329
    {
330
        op::reshape op;
Paul's avatar
Paul committed
331
332
333
334
335
336
337
        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
338
            literal s = args[1]->get_literal();
Paul's avatar
Paul committed
339
            s.visit([&](auto v) { copy(v, std::back_inserter(op.dims)); });
Paul's avatar
Paul committed
340
        }
Paul's avatar
Paul committed
341
342
343
        return prog.add_instruction(op, args[0]);
    }

Paul's avatar
Paul committed
344
    instruction_ref
Paul's avatar
Paul committed
345
    parse_flatten(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
346
347
    {
        uint64_t axis = 0;
Paul's avatar
Paul committed
348
349
350
351
        if(contains(attributes, "axis"))
        {
            axis = parse_value(attributes.at("axis")).at<int>();
        }
352
        return prog.add_instruction(op::flatten{axis}, args[0]);
Paul's avatar
Paul committed
353
354
    }

355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
    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
373
374
375
376
377
378
379
    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));
    }
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400

    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
401
402
403
    instruction_ref parse_constant(const std::string&,
                                   attribute_map attributes,
                                   const std::vector<instruction_ref>&)
Paul's avatar
Paul committed
404
405
406
407
    {
        literal v = parse_value(attributes.at("value"));
        return prog.add_literal(v);
    }
Paul's avatar
Paul committed
408

Paul's avatar
Paul committed
409
    instruction_ref
Paul's avatar
Paul committed
410
    parse_gemm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
    {
        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};
433
434
        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
435
436
437
        if(args.size() == 3)
        {
            uint64_t axis = 1;
Shucai Xiao's avatar
Shucai Xiao committed
438
            auto l3       = prog.add_instruction(op::dot{alpha, beta}, l1, l2);
Scott Thornton's avatar
Scott Thornton committed
439
            auto l4       = prog.add_instruction(op::broadcast{axis, l3->get_shape()}, args[2]);
440
            return prog.add_instruction(op::add{}, l3, l4);
Paul's avatar
Paul committed
441
        }
Shucai Xiao's avatar
Shucai Xiao committed
442
        return prog.add_instruction(op::dot{alpha, beta}, l1, l2);
Paul's avatar
Paul committed
443
444
    }

445
    instruction_ref
Paul's avatar
Paul committed
446
    parse_batchnorm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
447
    {
Scott Thornton's avatar
Scott Thornton committed
448
449
        float epsilon                                     = 1e-5f;
        float momentum                                    = 0.9f;
450
        op::batch_norm_inference::bn_infer_mode_t bn_mode = op::batch_norm_inference::spatial;
Scott Thornton's avatar
Scott Thornton committed
451
        bool is_test                                      = false;
452
453
454
455
456
457
        if(contains(attributes, "epsilon"))
        {
            epsilon = parse_value(attributes.at("epsilon")).at<float>();
        }
        if(contains(attributes, "momentum"))
        {
458
            momentum = parse_value(attributes.at("momentum")).at<float>();
459
460
461
        }
        if(contains(attributes, "is_test"))
        {
wsttiger's avatar
wsttiger committed
462
            is_test = parse_value(attributes.at("is_test")).at<uint64_t>() > 0;
463
464
465
        }
        if(contains(attributes, "spatial"))
        {
466
            bn_mode = (parse_value(attributes.at("spatial")).at<uint64_t>() > 0)
467
468
                          ? op::batch_norm_inference::spatial
                          : op::batch_norm_inference::per_activation;
469
        }
Paul's avatar
Paul committed
470
        (void)is_test;
Paul's avatar
Paul committed
471
        op::batch_norm_inference op{epsilon, momentum, bn_mode};
Paul's avatar
Paul committed
472
        return prog.add_instruction(op, std::move(args));
473
474
    }

475
476
477
478
    instruction_ref parse_leaky_relu(const std::string&,
                                     attribute_map attributes,
                                     std::vector<instruction_ref> args)
    {
Khalique's avatar
Khalique committed
479
        float alpha = 0.01; // default alpha val for leaky relu
480
481
482
483
484
485
486
487
        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
488
489
    instruction_ref
    parse_elu(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
490
491
492
493
494
495
496
497
498
499
    {
        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
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
    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
517

Khalique's avatar
Khalique committed
518
519
        auto scale_val = prog.add_literal(scale);
        auto bias_vals = prog.add_literal(
Paul's avatar
Paul committed
520
            migraphx::literal{migraphx::shape{migraphx::shape::float_type, {bias.size()}}, bias});
Khalique's avatar
Khalique committed
521

Paul's avatar
Paul committed
522
523
        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
524
        auto bias_bcast = prog.add_instruction(migraphx::op::broadcast{1, input_shape}, bias_vals);
Paul's avatar
Paul committed
525
        return prog.add_instruction(migraphx::op::add{}, img_scaled, bias_bcast);
Khalique's avatar
Khalique committed
526
    }
Khalique's avatar
Khalique committed
527

Khalique's avatar
Khalique committed
528
529
    instruction_ref
    parse_transpose(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
530
531
532
533
534
535
536
    {
        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
537
        return prog.add_instruction(migraphx::op::transpose{perm}, args.front());
Khalique's avatar
Khalique committed
538
539
    }

Paul's avatar
Paul committed
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
    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);
559
560
561
562
563
        std::unordered_map<std::string, onnx::TensorProto> initializer_data;
        for(auto&& f : graph.initializer())
        {
            initializer_data[f.name()] = f;
        }
Paul's avatar
Paul committed
564
565
566
        for(auto&& input : graph.input())
        {
            const std::string& name = input.name();
567
568
569
570
571
572
573
574
575
576
577
578
            // 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
579
580
581
        }
        for(auto&& p : nodes)
        {
582
            this->parse_node(get_name(p.second));
Paul's avatar
Paul committed
583
584
585
        }
    }

Paul's avatar
Paul committed
586
    void parse_node(const std::string& name)
Paul's avatar
Paul committed
587
    {
Paul's avatar
Paul committed
588
        if(name.empty())
Paul's avatar
Paul committed
589
            MIGRAPHX_THROW("Onnx node must have a name");
Paul's avatar
Paul committed
590
591
592
593
594
595
596
597
        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)
                {
598
                    auto&& iname = get_name(nodes.at(input));
Paul's avatar
Paul committed
599
                    assert(name != iname);
Paul's avatar
Paul committed
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
                    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;
    }

629
630
631
632
    static std::string get_name(const onnx::NodeProto& node)
    {
        if(node.name().empty())
        {
Paul's avatar
Paul committed
633
            std::string generated = "migraphx_unnamed_node";
Paul's avatar
Paul committed
634
635
636
637
            return std::accumulate(node.output().begin(),
                                   node.output().end(),
                                   generated,
                                   [](auto x, auto y) { return x + "_" + y; });
638
639
640
641
        }
        return node.name();
    }

Paul's avatar
Paul committed
642
643
644
645
646
    static node_map get_nodes(const onnx::GraphProto& graph)
    {
        std::unordered_map<std::string, onnx::NodeProto> result;
        for(auto&& node : graph.node())
        {
647
            result[get_name(node)] = node;
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
669
670
671
672
            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
673
        case onnx::AttributeProto::FLOATS: return from_repeated(shape::float_type, attr.floats());
Paul's avatar
Paul committed
674
675
676
677
678
        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
679
        MIGRAPHX_THROW("Invalid attribute type");
Paul's avatar
Paul committed
680
681
682
683
684
    }

    static literal parse_tensor(const onnx::TensorProto& t)
    {
        std::vector<std::size_t> dims(t.dims().begin(), t.dims().end());
Khalique's avatar
Khalique committed
685
        // in case of scalar constants in onnx file, use dims=1 to fill initializer data
686
        if(dims.empty())
Khalique's avatar
Khalique committed
687
688
689
        {
            dims = {1};
        }
690
691
        if(t.has_raw_data())
        {
wsttiger's avatar
wsttiger committed
692
            const std::string& s = t.raw_data();
Scott Thornton's avatar
Scott Thornton committed
693
694
695
696
697
698
699
700
701
702
703
704
            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
705
            case onnx::TensorProto::FLOAT16: return literal{{shape::half_type, dims}, s.data()};
Scott Thornton's avatar
Scott Thornton committed
706
707
708
709
710
711
            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("");
            }
Paul's avatar
Paul committed
712
            MIGRAPHX_THROW("Invalid tensor type");
713
        }
Paul's avatar
Paul committed
714
715
716
717
        switch(t.data_type())
        {
        case onnx::TensorProto::UNDEFINED: throw std::runtime_error("");
        case onnx::TensorProto::FLOAT:
Paul's avatar
Paul committed
718
            return literal{{shape::float_type, dims}, t.float_data().begin(), t.float_data().end()};
Paul's avatar
Paul committed
719
720
        case onnx::TensorProto::UINT8: throw std::runtime_error("");
        case onnx::TensorProto::INT8:
Paul's avatar
Paul committed
721
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
722
        case onnx::TensorProto::UINT16:
Paul's avatar
Paul committed
723
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
724
        case onnx::TensorProto::INT16:
Paul's avatar
Paul committed
725
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
726
        case onnx::TensorProto::INT32:
Paul's avatar
Paul committed
727
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
728
        case onnx::TensorProto::INT64:
Paul's avatar
Paul committed
729
            return literal{{shape::int64_type, dims}, t.int64_data().begin(), t.int64_data().end()};
Paul's avatar
Paul committed
730
731
        case onnx::TensorProto::STRING: throw std::runtime_error("");
        case onnx::TensorProto::BOOL:
Paul's avatar
Paul committed
732
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
733
734
        case onnx::TensorProto::FLOAT16:
            return literal{{shape::half_type, dims}, t.float_data().begin(), t.float_data().end()};
Paul's avatar
Paul committed
735
736
737
738
739
740
741
742
        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
743
        MIGRAPHX_THROW("Invalid tensor type");
Paul's avatar
Paul committed
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
    }

    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
765
        case onnx::TensorProto::FLOAT16: shape_type = shape::half_type; break;
Paul's avatar
Paul committed
766
767
768
769
770
771
772
773
774
        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
775
        auto&& tensor_dims = t.tensor_type().shape().dim();
776
777
778
779
780
781
782
783
784
785
786
        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
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
        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);
}

Paul's avatar
Paul committed
812
} // namespace MIGRAPHX_INLINE_NS
Paul's avatar
Paul committed
813
} // namespace migraphx