onnx.cpp 30.1 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
        add_generic_op("Sigmoid", op::sigmoid{});
        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{});
63
64
        add_generic_op("Sinh", op::sinh{});
        add_generic_op("Cosh", op::cosh{});
65
        add_generic_op("Tanh", op::tanh{});
66
67
68
        add_generic_op("Asin", op::asin{});
        add_generic_op("Acos", op::acos{});
        add_generic_op("Atan", op::atan{});
Paul's avatar
Paul committed
69

Khalique's avatar
Khalique committed
70
71
72
73
74
        add_binary_op("Add", op::add{});
        add_binary_op("Div", op::div{});
        add_binary_op("Mul", op::mul{});
        add_binary_op("Sub", op::sub{});

Khalique's avatar
Khalique committed
75
76
77
        add_variadic_op("Sum", op::add{});
        add_variadic_op("Max", op::max{});
        add_variadic_op("Min", op::min{});
Paul's avatar
Paul committed
78

Khalique's avatar
Khalique committed
79
        add_mem_op("ImageScaler", &onnx_parser::parse_imagescaler);
80
        add_mem_op("LeakyRelu", &onnx_parser::parse_leaky_relu);
Khalique's avatar
Khalique committed
81
        add_mem_op("Elu", &onnx_parser::parse_elu);
Paul's avatar
Paul committed
82
83
        add_mem_op("Constant", &onnx_parser::parse_constant);
        add_mem_op("Conv", &onnx_parser::parse_conv);
Paul's avatar
Paul committed
84
85
        add_mem_op("MaxPool", &onnx_parser::parse_pooling);
        add_mem_op("AveragePool", &onnx_parser::parse_pooling);
86
87
        add_mem_op("GlobalMaxPool", &onnx_parser::parse_pooling);
        add_mem_op("GlobalAveragePool", &onnx_parser::parse_pooling);
Paul's avatar
Paul committed
88
        add_mem_op("Reshape", &onnx_parser::parse_reshape);
Paul's avatar
Paul committed
89
90
        add_mem_op("Flatten", &onnx_parser::parse_flatten);
        add_mem_op("Gemm", &onnx_parser::parse_gemm);
91
        add_mem_op("BatchNormalization", &onnx_parser::parse_batchnorm);
Paul's avatar
Paul committed
92
        add_mem_op("Softmax", &onnx_parser::parse_softmax);
93
94
95
        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
96
        add_mem_op("Concat", &onnx_parser::parse_concat);
Khalique's avatar
Khalique committed
97
        add_mem_op("Transpose", &onnx_parser::parse_transpose);
Paul's avatar
Paul committed
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
    }

    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
113

114
    template <class T>
Khalique's avatar
Khalique committed
115
    void add_binary_op(std::string name, T x)
116
117
    {
        ops.emplace(name, [this, x](attribute_map attributes, std::vector<instruction_ref> args) {
Scott Thornton's avatar
Scott Thornton committed
118
            if(args.size() != 2)
Paul's avatar
Paul committed
119
                MIGRAPHX_THROW("binary operators should have 2 operands");
120
121
122
123
124
125
126
127
128
129
130
131
            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);
                }
132
                return prog.add_instruction(x, args);
133
            }
Khalique's avatar
Khalique committed
134
            else
135
            {
Khalique's avatar
Khalique committed
136
137
138
139
140
141
142
143
144
145
                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
146
147
148
149
150
151
152
153
154
155
156
157
158
            // 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
159
160
161
162
163
164
165
166
167
168
169
            // 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
170
171
172
173
                           s0->end(),
                           s1->begin() + offset,
                           output_lens.begin() + offset,
                           [](auto a, auto b) { return std::max(a, b); });
Khalique's avatar
Khalique committed
174
175
176
177
178
179
180
181
182

            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});
        }
183
184
    }

Paul's avatar
Paul committed
185
    template <class T>
Paul's avatar
Paul committed
186
187
    void add_generic_op(std::string name, T x)
    {
Paul's avatar
Paul committed
188
        ops.emplace(name, [this, x](attribute_map, std::vector<instruction_ref> args) {
Paul's avatar
Paul committed
189
190
191
192
            return prog.add_instruction(x, args);
        });
    }

Khalique's avatar
Khalique committed
193
    template <class T>
Khalique's avatar
Khalique committed
194
    void add_variadic_op(std::string name, T x)
Khalique's avatar
Khalique committed
195
    {
Khalique's avatar
Khalique committed
196
197
        ops.emplace(name, [this, x](attribute_map, std::vector<instruction_ref> args) {
            return std::accumulate(std::next(args.begin()),
Khalique's avatar
Khalique committed
198
199
200
201
202
                                   args.end(),
                                   args.front(),
                                   [this, x](instruction_ref a, instruction_ref b) {
                                       return add_broadcastable_binary_op(a, b, x);
                                   });
Khalique's avatar
Khalique committed
203
        });
Khalique's avatar
Khalique committed
204
205
    }

Paul's avatar
Paul committed
206
    instruction_ref
Paul's avatar
Paul committed
207
    parse_softmax(const std::string&, const attribute_map&, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
208
209
    {
        auto dims = args.front()->get_shape().lens();
Scott Thornton's avatar
Scott Thornton committed
210
211
        auto r =
            prog.add_instruction(op::reshape{{long(dims[0]), long(dims[1]), 1, 1}}, args.front());
212
213
        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
214
215
    }

Paul's avatar
Paul committed
216
    instruction_ref
Paul's avatar
Paul committed
217
    parse_conv(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
218
    {
219
        op::convolution op;
Paul's avatar
Paul committed
220
221
        if(contains(attributes, "pads"))
        {
Scott Thornton's avatar
Scott Thornton committed
222
            if(contains(attributes, "auto_pad"))
223
            {
Paul's avatar
Paul committed
224
                MIGRAPHX_THROW("auto_pad and padding cannot be specified simultaneously");
225
226
227
            }
            std::vector<std::size_t> padding(4);
            copy(attributes["pads"].ints(), padding.begin());
Scott Thornton's avatar
Scott Thornton committed
228
            if(padding.size() != 4)
229
            {
Paul's avatar
Paul committed
230
                MIGRAPHX_THROW("padding should have 4 values");
231
            }
Scott Thornton's avatar
Scott Thornton committed
232
            if(padding[0] != padding[2] || padding[1] != padding[3])
233
            {
Paul's avatar
Paul committed
234
                MIGRAPHX_THROW("migraphx does not support asymetric padding");
235
236
237
            }
            op.padding[0] = padding[0];
            op.padding[1] = padding[1];
Paul's avatar
Paul committed
238
        }
Paul's avatar
Paul committed
239
240
241
242
243
244
245
246
        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
247
        if(contains(attributes, "auto_pad"))
248
249
        {
            auto s = attributes["auto_pad"].s();
Scott Thornton's avatar
Scott Thornton committed
250
            if(contains(attributes, "pads") and to_upper(s) != "NOTSET")
251
            {
Paul's avatar
Paul committed
252
                MIGRAPHX_THROW("auto_pad and padding cannot be specified simultaneously");
253
254
            }

wsttiger's avatar
fixes  
wsttiger committed
255
            if(s.find("SAME") != std::string::npos)
256
257
258
259
            {
                op.padding_mode = op::convolution::same;
            }
        }
Paul's avatar
Paul committed
260
261
262
263
        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
264
            auto l2       = prog.add_instruction(op::broadcast{axis, l1->get_shape()}, args[2]);
265
            return prog.add_instruction(op::add{}, l1, l2);
Paul's avatar
Paul committed
266
        }
Paul's avatar
Paul committed
267
268
        return prog.add_instruction(op, args);
    }
Paul's avatar
Paul committed
269

Paul's avatar
Paul committed
270
271
272
    instruction_ref parse_pooling(const std::string& name,
                                  attribute_map attributes,
                                  std::vector<instruction_ref> args)
Paul's avatar
Paul committed
273
    {
Khalique's avatar
Khalique committed
274
275
        op::pooling op{ends_with(name, "MaxPool") ? "max" : "average"};
        if(starts_with(name, "Global"))
276
        {
Khalique's avatar
Khalique committed
277
278
            auto lens  = args.front()->get_shape().lens();
            op.lengths = {lens[2], lens[3]};
279
        }
Paul's avatar
Paul committed
280
281
        if(contains(attributes, "pads"))
        {
282
283
            std::vector<std::size_t> padding(4);
            copy(attributes["pads"].ints(), padding.begin());
Scott Thornton's avatar
Scott Thornton committed
284
            if(padding.size() != 4)
285
            {
Paul's avatar
Paul committed
286
                MIGRAPHX_THROW("padding should have 4 values");
287
            }
Scott Thornton's avatar
Scott Thornton committed
288
            if(padding[0] != padding[2] || padding[1] != padding[3])
289
            {
Paul's avatar
Paul committed
290
                MIGRAPHX_THROW("migraphx does not support asymetric padding");
291
292
293
            }
            op.padding[0] = padding[0];
            op.padding[1] = padding[1];
Paul's avatar
Paul committed
294
295
296
297
298
299
300
301
302
        }
        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
303
        if(contains(attributes, "auto_pad"))
304
305
        {
            auto s = attributes["auto_pad"].s();
Scott Thornton's avatar
Scott Thornton committed
306
            if(to_upper(s) != "NOTSET")
307
            {
Paul's avatar
Paul committed
308
                MIGRAPHX_THROW("auto_pad is not supported for pooling");
309
310
311
            }
        }

Paul's avatar
Paul committed
312
        return prog.add_instruction(op, std::move(args));
Paul's avatar
Paul committed
313
314
    }

Paul's avatar
Paul committed
315
    instruction_ref
Paul's avatar
Paul committed
316
    parse_reshape(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
317
    {
318
        op::reshape op;
Paul's avatar
Paul committed
319
320
321
322
323
324
325
        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
326
            literal s = args[1]->get_literal();
Paul's avatar
Paul committed
327
            s.visit([&](auto v) { copy(v, std::back_inserter(op.dims)); });
Paul's avatar
Paul committed
328
        }
Paul's avatar
Paul committed
329
330
331
        return prog.add_instruction(op, args[0]);
    }

Paul's avatar
Paul committed
332
    instruction_ref
Paul's avatar
Paul committed
333
    parse_flatten(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
334
335
    {
        uint64_t axis = 0;
Paul's avatar
Paul committed
336
337
338
339
        if(contains(attributes, "axis"))
        {
            axis = parse_value(attributes.at("axis")).at<int>();
        }
340
        return prog.add_instruction(op::flatten{axis}, args[0]);
Paul's avatar
Paul committed
341
342
    }

343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
    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
361
362
363
364
365
366
367
    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));
    }
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388

    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
389
390
391
    instruction_ref parse_constant(const std::string&,
                                   attribute_map attributes,
                                   const std::vector<instruction_ref>&)
Paul's avatar
Paul committed
392
393
394
395
    {
        literal v = parse_value(attributes.at("value"));
        return prog.add_literal(v);
    }
Paul's avatar
Paul committed
396

Paul's avatar
Paul committed
397
    instruction_ref
Paul's avatar
Paul committed
398
    parse_gemm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
    {
        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};
421
422
        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
423
424
425
        if(args.size() == 3)
        {
            uint64_t axis = 1;
Shucai Xiao's avatar
Shucai Xiao committed
426
            auto l3       = prog.add_instruction(op::dot{alpha, beta}, l1, l2);
Scott Thornton's avatar
Scott Thornton committed
427
            auto l4       = prog.add_instruction(op::broadcast{axis, l3->get_shape()}, args[2]);
428
            return prog.add_instruction(op::add{}, l3, l4);
Paul's avatar
Paul committed
429
        }
Shucai Xiao's avatar
Shucai Xiao committed
430
        return prog.add_instruction(op::dot{alpha, beta}, l1, l2);
Paul's avatar
Paul committed
431
432
    }

433
    instruction_ref
Paul's avatar
Paul committed
434
    parse_batchnorm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
435
    {
Scott Thornton's avatar
Scott Thornton committed
436
437
        float epsilon                                     = 1e-5f;
        float momentum                                    = 0.9f;
438
        op::batch_norm_inference::bn_infer_mode_t bn_mode = op::batch_norm_inference::spatial;
Scott Thornton's avatar
Scott Thornton committed
439
        bool is_test                                      = false;
440
441
442
443
444
445
        if(contains(attributes, "epsilon"))
        {
            epsilon = parse_value(attributes.at("epsilon")).at<float>();
        }
        if(contains(attributes, "momentum"))
        {
446
            momentum = parse_value(attributes.at("momentum")).at<float>();
447
448
449
        }
        if(contains(attributes, "is_test"))
        {
wsttiger's avatar
wsttiger committed
450
            is_test = parse_value(attributes.at("is_test")).at<uint64_t>() > 0;
451
452
453
        }
        if(contains(attributes, "spatial"))
        {
454
            bn_mode = (parse_value(attributes.at("spatial")).at<uint64_t>() > 0)
455
456
                          ? op::batch_norm_inference::spatial
                          : op::batch_norm_inference::per_activation;
457
        }
Paul's avatar
Paul committed
458
        (void)is_test;
Paul's avatar
Paul committed
459
        op::batch_norm_inference op{epsilon, momentum, bn_mode};
Paul's avatar
Paul committed
460
        return prog.add_instruction(op, std::move(args));
461
462
    }

463
464
465
466
    instruction_ref parse_leaky_relu(const std::string&,
                                     attribute_map attributes,
                                     std::vector<instruction_ref> args)
    {
Khalique's avatar
Khalique committed
467
        float alpha = 0.01; // default alpha val for leaky relu
468
469
470
471
472
473
474
475
        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
476
477
    instruction_ref
    parse_elu(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
478
479
480
481
482
483
484
485
486
487
    {
        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
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
    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
505

Khalique's avatar
Khalique committed
506
507
        auto scale_val = prog.add_literal(scale);
        auto bias_vals = prog.add_literal(
Paul's avatar
Paul committed
508
            migraphx::literal{migraphx::shape{migraphx::shape::float_type, {bias.size()}}, bias});
Khalique's avatar
Khalique committed
509

Paul's avatar
Paul committed
510
511
        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
512
        auto bias_bcast = prog.add_instruction(migraphx::op::broadcast{1, input_shape}, bias_vals);
Paul's avatar
Paul committed
513
        return prog.add_instruction(migraphx::op::add{}, img_scaled, bias_bcast);
Khalique's avatar
Khalique committed
514
    }
Khalique's avatar
Khalique committed
515

Khalique's avatar
Khalique committed
516
517
    instruction_ref
    parse_transpose(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
518
519
520
521
522
523
524
    {
        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
525
        return prog.add_instruction(migraphx::op::transpose{perm}, args.front());
Khalique's avatar
Khalique committed
526
527
    }

Paul's avatar
Paul committed
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
    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);
547
548
549
550
551
        std::unordered_map<std::string, onnx::TensorProto> initializer_data;
        for(auto&& f : graph.initializer())
        {
            initializer_data[f.name()] = f;
        }
Paul's avatar
Paul committed
552
553
554
        for(auto&& input : graph.input())
        {
            const std::string& name = input.name();
555
556
557
558
559
560
561
562
563
564
565
566
            // 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
567
568
569
        }
        for(auto&& p : nodes)
        {
570
            this->parse_node(get_name(p.second));
Paul's avatar
Paul committed
571
572
573
        }
    }

Paul's avatar
Paul committed
574
    void parse_node(const std::string& name)
Paul's avatar
Paul committed
575
    {
Paul's avatar
Paul committed
576
        if(name.empty())
Paul's avatar
Paul committed
577
            MIGRAPHX_THROW("Onnx node must have a name");
Paul's avatar
Paul committed
578
579
580
581
582
583
584
585
        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)
                {
586
                    auto&& iname = get_name(nodes.at(input));
Paul's avatar
Paul committed
587
                    assert(name != iname);
Paul's avatar
Paul committed
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
                    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;
    }

617
618
619
620
    static std::string get_name(const onnx::NodeProto& node)
    {
        if(node.name().empty())
        {
Paul's avatar
Paul committed
621
            std::string generated = "migraphx_unnamed_node";
Paul's avatar
Paul committed
622
623
624
625
            return std::accumulate(node.output().begin(),
                                   node.output().end(),
                                   generated,
                                   [](auto x, auto y) { return x + "_" + y; });
626
627
628
629
        }
        return node.name();
    }

Paul's avatar
Paul committed
630
631
632
633
634
    static node_map get_nodes(const onnx::GraphProto& graph)
    {
        std::unordered_map<std::string, onnx::NodeProto> result;
        for(auto&& node : graph.node())
        {
635
            result[get_name(node)] = node;
Paul's avatar
Paul committed
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
            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
661
        case onnx::AttributeProto::FLOATS: return from_repeated(shape::float_type, attr.floats());
Paul's avatar
Paul committed
662
663
664
665
666
        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
667
        MIGRAPHX_THROW("Invalid attribute type");
Paul's avatar
Paul committed
668
669
670
671
672
    }

    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
673
        // in case of scalar constants in onnx file, use dims=1 to fill initializer data
674
        if(dims.empty())
Khalique's avatar
Khalique committed
675
676
677
        {
            dims = {1};
        }
678
679
        if(t.has_raw_data())
        {
wsttiger's avatar
wsttiger committed
680
            const std::string& s = t.raw_data();
Scott Thornton's avatar
Scott Thornton committed
681
682
683
684
685
686
687
688
689
690
691
692
            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
693
            case onnx::TensorProto::FLOAT16: return literal{{shape::half_type, dims}, s.data()};
Scott Thornton's avatar
Scott Thornton committed
694
695
696
697
698
699
            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
700
            MIGRAPHX_THROW("Invalid tensor type");
701
        }
Paul's avatar
Paul committed
702
703
704
705
        switch(t.data_type())
        {
        case onnx::TensorProto::UNDEFINED: throw std::runtime_error("");
        case onnx::TensorProto::FLOAT:
Paul's avatar
Paul committed
706
            return literal{{shape::float_type, dims}, t.float_data().begin(), t.float_data().end()};
Paul's avatar
Paul committed
707
708
        case onnx::TensorProto::UINT8: throw std::runtime_error("");
        case onnx::TensorProto::INT8:
Paul's avatar
Paul committed
709
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
710
        case onnx::TensorProto::UINT16:
Paul's avatar
Paul committed
711
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
712
        case onnx::TensorProto::INT16:
Paul's avatar
Paul committed
713
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
714
        case onnx::TensorProto::INT32:
Paul's avatar
Paul committed
715
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
716
        case onnx::TensorProto::INT64:
Paul's avatar
Paul committed
717
            return literal{{shape::int64_type, dims}, t.int64_data().begin(), t.int64_data().end()};
Paul's avatar
Paul committed
718
719
        case onnx::TensorProto::STRING: throw std::runtime_error("");
        case onnx::TensorProto::BOOL:
Paul's avatar
Paul committed
720
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
721
722
        case onnx::TensorProto::FLOAT16:
            return literal{{shape::half_type, dims}, t.float_data().begin(), t.float_data().end()};
Paul's avatar
Paul committed
723
724
725
726
727
728
729
730
        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
731
        MIGRAPHX_THROW("Invalid tensor type");
Paul's avatar
Paul committed
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
    }

    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
753
        case onnx::TensorProto::FLOAT16: shape_type = shape::half_type; break;
Paul's avatar
Paul committed
754
755
756
757
758
759
760
761
762
        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
763
        auto&& tensor_dims = t.tensor_type().shape().dim();
764
765
766
767
768
769
770
771
772
773
774
        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
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
        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
800
} // namespace MIGRAPHX_INLINE_NS
Paul's avatar
Paul committed
801
} // namespace migraphx