onnx.cpp 56 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
#include <migraphx/fallthrough.hpp>
#include <migraphx/program.hpp>
#include <migraphx/operators.hpp>
#include <migraphx/ranges.hpp>
#include <migraphx/instruction.hpp>
#include <migraphx/config.hpp>
18
#include <migraphx/onnx.hpp>
Paul's avatar
Paul committed
19
20

namespace migraphx {
Paul's avatar
Paul committed
21
inline namespace MIGRAPHX_INLINE_NS {
Paul's avatar
Paul committed
22
23
24
25
26

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
27
28
    using op_func =
        std::function<std::vector<instruction_ref>(attribute_map, std::vector<instruction_ref>)>;
Paul's avatar
Paul committed
29
30
    node_map nodes;
    std::unordered_map<std::string, instruction_ref> instructions;
Scott Thornton's avatar
Scott Thornton committed
31
    program prog    = program();
32
    bool is_pytorch = false;
Paul's avatar
Paul committed
33
34

    std::unordered_map<std::string, op_func> ops;
35
    std::unordered_map<std::string, operation> map_actv_funcs;
Paul's avatar
Paul committed
36
37
38

    onnx_parser()
    {
Khalique's avatar
Khalique committed
39
        add_generic_op("Relu", op::relu{});
Khalique's avatar
Khalique committed
40
41
        add_generic_op("Sigmoid", op::sigmoid{});
        add_generic_op("Abs", op::abs{});
Shucai Xiao's avatar
Shucai Xiao committed
42
43
        add_generic_op("Exp", op::exp{});
        add_generic_op("Log", op::log{});
Khalique's avatar
Khalique committed
44
45
        // disable dropout for inference
        add_generic_op("Dropout", op::identity{});
Khalique's avatar
Khalique committed
46
        add_generic_op("Identity", op::identity{});
Shucai Xiao's avatar
Shucai Xiao committed
47
48
49
        add_generic_op("Sin", op::sin{});
        add_generic_op("Cos", op::cos{});
        add_generic_op("Tan", op::tan{});
50
51
        add_generic_op("Sinh", op::sinh{});
        add_generic_op("Cosh", op::cosh{});
52
        add_generic_op("Tanh", op::tanh{});
53
54
55
        add_generic_op("Asin", op::asin{});
        add_generic_op("Acos", op::acos{});
        add_generic_op("Atan", op::atan{});
Paul's avatar
Paul committed
56

Khalique's avatar
Khalique committed
57
58
59
60
61
        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
62
63
64
        add_variadic_op("Sum", op::add{});
        add_variadic_op("Max", op::max{});
        add_variadic_op("Min", op::min{});
Paul's avatar
Paul committed
65

Khalique's avatar
Khalique committed
66
        add_mem_op("LRN", &onnx_parser::parse_lrn);
Khalique's avatar
Khalique committed
67
        add_mem_op("ImageScaler", &onnx_parser::parse_imagescaler);
68
        add_mem_op("LeakyRelu", &onnx_parser::parse_leaky_relu);
Khalique's avatar
Khalique committed
69
        add_mem_op("Elu", &onnx_parser::parse_elu);
Paul's avatar
Paul committed
70
71
        add_mem_op("Constant", &onnx_parser::parse_constant);
        add_mem_op("Conv", &onnx_parser::parse_conv);
Paul's avatar
Paul committed
72
73
        add_mem_op("MaxPool", &onnx_parser::parse_pooling);
        add_mem_op("AveragePool", &onnx_parser::parse_pooling);
74
75
        add_mem_op("GlobalMaxPool", &onnx_parser::parse_pooling);
        add_mem_op("GlobalAveragePool", &onnx_parser::parse_pooling);
Paul's avatar
Paul committed
76
        add_mem_op("Reshape", &onnx_parser::parse_reshape);
Paul's avatar
Paul committed
77
78
        add_mem_op("Flatten", &onnx_parser::parse_flatten);
        add_mem_op("Gemm", &onnx_parser::parse_gemm);
79
        add_mem_op("MatMul", &onnx_parser::parse_matmul);
80
        add_mem_op("BatchNormalization", &onnx_parser::parse_batchnorm);
Paul's avatar
Paul committed
81
        add_mem_op("Softmax", &onnx_parser::parse_softmax);
Shucai Xiao's avatar
Shucai Xiao committed
82
        add_mem_op("LogSoftmax", &onnx_parser::parse_logsoftmax);
83
84
85
        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
86
        add_mem_op("Concat", &onnx_parser::parse_concat);
87
88
89
        add_mem_op("Gather", &onnx_parser::parse_gather);
        add_mem_op("Shape", &onnx_parser::parse_shape);
        add_mem_op("ConstantFill", &onnx_parser::parse_constant_fill);
Khalique's avatar
Khalique committed
90
        add_mem_op("Transpose", &onnx_parser::parse_transpose);
Shucai Xiao's avatar
Shucai Xiao committed
91
        add_mem_op("RNN", &onnx_parser::parse_rnn);
92
        add_mem_op("GRU", &onnx_parser::parse_gru);
Shucai Xiao's avatar
Shucai Xiao committed
93
        add_mem_op("LSTM", &onnx_parser::parse_lstm);
Khalique's avatar
Khalique committed
94
        add_mem_op("Pad", &onnx_parser::parse_pad);
95
96
97
98
99
100
101

        // init the activation function map
        init_actv_func();
    }

    void init_actv_func()
    {
102
103
104
105
106
        map_actv_funcs.insert(std::make_pair("tanh", op::tanh{}));
        map_actv_funcs.insert(std::make_pair("relu", op::relu{}));
        map_actv_funcs.insert(std::make_pair("sigmoid", op::sigmoid{}));
        map_actv_funcs.insert(std::make_pair("leakyrelu", op::leaky_relu{}));
        map_actv_funcs.insert(std::make_pair("elu", op::elu{}));
Paul's avatar
Paul committed
107
108
109
110
    }

    template <class F>
    void add_op(std::string name, F f)
Paul's avatar
Paul committed
111
112
113
114
115
116
117
118
119
    {
        ops.emplace(name, [=](auto&&... xs) {
            return std::vector<instruction_ref>{f(std::forward<decltype(xs)>(xs)...)};
        });
    }

    // Multi output op
    template <class F>
    void add_multi_op(std::string name, F f)
Paul's avatar
Paul committed
120
121
122
123
124
125
126
    {
        ops.emplace(name, f);
    }

    template <class F>
    void add_mem_op(std::string name, F f)
    {
Paul's avatar
Paul committed
127
        add_op(name, [=](auto&&... xs) {
Paul's avatar
Paul committed
128
129
130
            return std::mem_fn(f)(*this, name, std::forward<decltype(xs)>(xs)...);
        });
    }
Khalique's avatar
Khalique committed
131

132
    template <class T>
Khalique's avatar
Khalique committed
133
    void add_binary_op(std::string name, T x)
134
    {
Paul's avatar
Paul committed
135
        add_op(name, [this, x](attribute_map attributes, std::vector<instruction_ref> args) {
Scott Thornton's avatar
Scott Thornton committed
136
            if(args.size() != 2)
Paul's avatar
Paul committed
137
                MIGRAPHX_THROW("binary operators should have 2 operands");
138
            if(contains(attributes, "broadcast") and contains(attributes, "axis"))
139
140
141
142
            {
                uint64_t broadcasted = parse_value(attributes.at("broadcast")).at<uint64_t>();
                if(broadcasted != 0)
                {
143
                    uint64_t axis = parse_value(attributes.at("axis")).at<uint64_t>();
144
145
146
147
                    auto l =
                        prog.add_instruction(op::broadcast{axis, args[0]->get_shape()}, args[1]);
                    return prog.add_instruction(x, args[0], l);
                }
148
                return prog.add_instruction(x, args);
149
            }
Paul's avatar
Paul committed
150
            else
151
            {
Khalique's avatar
Khalique committed
152
                return add_broadcastable_binary_op(args[0], args[1], x);
153
154
155
156
            }
        });
    }

157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
    std::vector<std::size_t> compute_broadcasted_lens(std::vector<std::size_t> s0, std::vector<std::size_t> s1)
    {
        // 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)
        if (s0.size() > s1.size())
        {
            s0.swap(s1);
        }

        std::vector<std::size_t> out_lens(s1);
        auto offset = s1.size() - s0.size();
        std::transform(s0.begin(), s0.end(), s1.begin() + offset, out_lens.begin() + offset,
        [](auto a, auto b) { return std::max(a, b); });

        return out_lens;
    }

Khalique's avatar
Khalique committed
184
185
186
    template <class T>
    instruction_ref add_broadcastable_binary_op(instruction_ref arg0, instruction_ref arg1, T x)
    {
Khalique's avatar
Khalique committed
187
        if(arg0->get_shape().lens() != arg1->get_shape().lens())
Khalique's avatar
Khalique committed
188
189
        {
            // Get lengths for both arguments
190
191
192
193
194
            auto s0 = arg0->get_shape().lens();
            auto s1 = arg1->get_shape().lens();
            auto out_lens = compute_broadcasted_lens(s0, s1);
            auto l0 = prog.add_instruction(op::multibroadcast{out_lens}, arg0);
            auto l1 = prog.add_instruction(op::multibroadcast{out_lens}, arg1);
Khalique's avatar
Khalique committed
195
196
197
198
199
200
            return prog.add_instruction(x, l0, l1);
        }
        else
        {
            return prog.add_instruction(x, {arg0, arg1});
        }
201
202
    }

Paul's avatar
Paul committed
203
    template <class T>
Paul's avatar
Paul committed
204
205
    void add_generic_op(std::string name, T x)
    {
Paul's avatar
Paul committed
206
        add_op(name, [this, x](attribute_map, std::vector<instruction_ref> args) {
Paul's avatar
Paul committed
207
208
209
210
            return prog.add_instruction(x, args);
        });
    }

Khalique's avatar
Khalique committed
211
    template <class T>
Khalique's avatar
Khalique committed
212
    void add_variadic_op(std::string name, T x)
Khalique's avatar
Khalique committed
213
    {
Paul's avatar
Paul committed
214
        add_op(name, [this, x](attribute_map, std::vector<instruction_ref> args) {
Khalique's avatar
Khalique committed
215
            return std::accumulate(std::next(args.begin()),
Khalique's avatar
Khalique committed
216
217
218
219
220
                                   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
221
        });
Khalique's avatar
Khalique committed
222
223
    }

Paul's avatar
Paul committed
224
    instruction_ref
Paul's avatar
Paul committed
225
    parse_softmax(const std::string&, const attribute_map&, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
226
227
    {
        auto dims = args.front()->get_shape().lens();
Scott Thornton's avatar
Scott Thornton committed
228
229
        auto r =
            prog.add_instruction(op::reshape{{long(dims[0]), long(dims[1]), 1, 1}}, args.front());
230
231
        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
232
233
    }

Shucai Xiao's avatar
Shucai Xiao committed
234
235
236
    instruction_ref parse_logsoftmax(const std::string&,
                                     const attribute_map& attributes,
                                     std::vector<instruction_ref> args)
Shucai Xiao's avatar
Shucai Xiao committed
237
238
239
240
241
242
243
244
245
246
    {
        int axis = 1;
        if(contains(attributes, "axis"))
        {
            axis = parse_value(attributes.at("axis")).at<int>();
        }

        return prog.add_instruction(op::logsoftmax{axis}, std::move(args));
    }

Paul's avatar
Paul committed
247
    instruction_ref
Paul's avatar
Paul committed
248
    parse_conv(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
249
    {
250
        op::convolution op;
251
        auto l0 = args[0];
Paul's avatar
Paul committed
252
253
        if(contains(attributes, "pads"))
        {
Scott Thornton's avatar
Scott Thornton committed
254
            if(contains(attributes, "auto_pad"))
255
            {
Paul's avatar
Paul committed
256
                MIGRAPHX_THROW("auto_pad and padding cannot be specified simultaneously");
257
            }
258
259
            std::vector<std::int64_t> padding;
            copy(attributes["pads"].ints(), std::back_inserter(padding));
Scott Thornton's avatar
Scott Thornton committed
260
            if(padding.size() != 4)
261
            {
Paul's avatar
Paul committed
262
                MIGRAPHX_THROW("padding should have 4 values");
263
            }
Scott Thornton's avatar
Scott Thornton committed
264
            if(padding[0] != padding[2] || padding[1] != padding[3])
265
            {
266
267
                // insert zeros for pad op (args[0] has 4 dims)
                padding = {0, 0, padding[0], padding[1], 0, 0, padding[2], padding[3]};
Khalique's avatar
Khalique committed
268
                l0      = prog.add_instruction(op::pad{padding}, l0);
269
            }
270
271
272
273
            else
            {
                op.padding[0] = padding[0];
                op.padding[1] = padding[1];
274
            }
Paul's avatar
Paul committed
275
        }
Paul's avatar
Paul committed
276
277
278
279
280
281
282
283
        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
284
        if(contains(attributes, "auto_pad"))
285
286
        {
            auto s = attributes["auto_pad"].s();
Scott Thornton's avatar
Scott Thornton committed
287
            if(contains(attributes, "pads") and to_upper(s) != "NOTSET")
288
            {
Paul's avatar
Paul committed
289
                MIGRAPHX_THROW("auto_pad and padding cannot be specified simultaneously");
290
291
            }

wsttiger's avatar
fixes  
wsttiger committed
292
            if(s.find("SAME") != std::string::npos)
293
            {
294
                op.padding_mode = op::padding_mode_t::same;
295
296
            }
        }
Khalique's avatar
Khalique committed
297
298
299
300
        if(contains(attributes, "group"))
        {
            op.group = parse_value(attributes.at("group")).at<int>();
        }
Paul's avatar
Paul committed
301
302
303
304
        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
305
            auto l2       = prog.add_instruction(op::broadcast{axis, l1->get_shape()}, args[2]);
306
            return prog.add_instruction(op::add{}, l1, l2);
Paul's avatar
Paul committed
307
        }
308
        return prog.add_instruction(op, l0, args[1]);
Paul's avatar
Paul committed
309
    }
Paul's avatar
Paul committed
310

Paul's avatar
Paul committed
311
312
313
    instruction_ref parse_pooling(const std::string& name,
                                  attribute_map attributes,
                                  std::vector<instruction_ref> args)
Paul's avatar
Paul committed
314
    {
Khalique's avatar
Khalique committed
315
        op::pooling op{ends_with(name, "MaxPool") ? "max" : "average"};
316
        auto l0 = args[0];
Khalique's avatar
Khalique committed
317
        if(starts_with(name, "Global"))
318
        {
Khalique's avatar
Khalique committed
319
320
            auto lens  = args.front()->get_shape().lens();
            op.lengths = {lens[2], lens[3]};
321
        }
Paul's avatar
Paul committed
322
323
        if(contains(attributes, "pads"))
        {
324
325
            std::vector<std::int64_t> padding;
            copy(attributes["pads"].ints(), std::back_inserter(padding));
Scott Thornton's avatar
Scott Thornton committed
326
            if(padding.size() != 4)
327
            {
Paul's avatar
Paul committed
328
                MIGRAPHX_THROW("padding should have 4 values");
329
            }
Scott Thornton's avatar
Scott Thornton committed
330
            if(padding[0] != padding[2] || padding[1] != padding[3])
331
            {
332
333
                // insert zeros for pad op (args[0] has 4 dims)
                padding = {0, 0, padding[0], padding[1], 0, 0, padding[2], padding[3]};
Khalique's avatar
Khalique committed
334
                l0      = prog.add_instruction(op::pad{padding}, l0);
335
336
337
338
339
            }
            else
            {
                op.padding[0] = padding[0];
                op.padding[1] = padding[1];
340
            }
Paul's avatar
Paul committed
341
342
343
344
345
346
347
348
349
        }
        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
350
        if(contains(attributes, "auto_pad"))
351
352
        {
            auto s = attributes["auto_pad"].s();
353
            if(s.find("SAME_UPPER") == std::string::npos)
354
            {
355
                MIGRAPHX_THROW("auto_pad only supports SAME_UPPER for pooling");
356
            }
357
            op.padding_mode = op::padding_mode_t::same;
358
359
        }

360
        return prog.add_instruction(op, l0);
Paul's avatar
Paul committed
361
362
    }

Paul's avatar
Paul committed
363
    instruction_ref
Paul's avatar
Paul committed
364
    parse_reshape(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
365
    {
366
        op::reshape op;
Paul's avatar
Paul committed
367
368
369
370
371
372
373
        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
374
            auto s = args[1]->eval();
Paul's avatar
Paul committed
375
            if(s.empty())
Paul's avatar
Paul committed
376
                MIGRAPHX_THROW("Dynamic shape is not supported.");
Paul's avatar
Paul committed
377
            s.visit([&](auto v) { copy(v, std::back_inserter(op.dims)); });
Paul's avatar
Paul committed
378
        }
Paul's avatar
Paul committed
379
380
381
        return prog.add_instruction(op, args[0]);
    }

Paul's avatar
Paul committed
382
    instruction_ref
Paul's avatar
Paul committed
383
    parse_flatten(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
384
    {
385
        uint64_t axis = 1;
Paul's avatar
Paul committed
386
387
388
389
        if(contains(attributes, "axis"))
        {
            axis = parse_value(attributes.at("axis")).at<int>();
        }
390
        return prog.add_instruction(op::flatten{axis}, args[0]);
Paul's avatar
Paul committed
391
392
    }

393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
    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
411
412
413
414
415
416
417
    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));
    }
418

419
420
421
    instruction_ref
    parse_gather(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
422
        int axis = 0;
423
424
425
426
        if(contains(attributes, "axis"))
        {
            axis = parse_value(attributes.at("axis")).at<int>();
        }
427
        op::gather op{axis};
428
429
430
        return prog.add_instruction(op, std::move(args));
    }

431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
    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
451
452
453
    instruction_ref parse_constant(const std::string&,
                                   attribute_map attributes,
                                   const std::vector<instruction_ref>&)
Paul's avatar
Paul committed
454
    {
Shucai Xiao's avatar
Shucai Xiao committed
455
        literal v     = parse_value(attributes.at("value"));
456
457
458
        auto dim_size = attributes.at("value").t().dims_size();
        // if dim_size is 0, it is a scalar
        if(dim_size == 0)
459
        {
460
            migraphx::shape scalar_shape{v.get_shape().type()};
461
462
463
            return prog.add_literal(migraphx::literal{scalar_shape, v.data()});
        }

Paul's avatar
Paul committed
464
465
        return prog.add_literal(v);
    }
Paul's avatar
Paul committed
466

Paul's avatar
Paul committed
467
    instruction_ref
Paul's avatar
Paul committed
468
    parse_gemm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Paul's avatar
Paul committed
469
470
    {
        float alpha = 1.0f;
Khalique's avatar
Khalique committed
471
        float beta  = 1.0f;
Paul's avatar
Paul committed
472
473
474
475
476
477
478
479
        bool transa = false;
        bool transb = false;
        if(contains(attributes, "alpha"))
        {
            alpha = parse_value(attributes.at("alpha")).at<float>();
        }
        if(contains(attributes, "beta"))
        {
480
            beta = parse_value(attributes.at("beta")).at<float>();
Paul's avatar
Paul committed
481
482
483
484
485
486
487
488
489
        }
        if(contains(attributes, "transA"))
        {
            transa = parse_value(attributes.at("transA")).at<bool>();
        }
        if(contains(attributes, "transB"))
        {
            transb = parse_value(attributes.at("transB")).at<bool>();
        }
490

491
492
493
494
495
        std::vector<int64_t> perm(args[0]->get_shape().lens().size());
        std::iota(perm.begin(), perm.end(), int64_t{0});
        // swap the last two elements
        std::swap(*perm.rbegin(), *(perm.rbegin() + 1));

496
497
        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
498
499
        if(args.size() == 3)
        {
500
            if(beta != 0.f && args[2]->get_shape().elements() > 0)
501
            {
502
503
504
505
506
                auto out_lens = l1->get_shape().lens();
                out_lens.back() = l2->get_shape().lens().back();
                auto l3 = args[2];
                auto l3_lens = l3->get_shape().lens();
                if (!std::equal(out_lens.begin(), out_lens.end(), l3_lens.begin(), l3_lens.end()))
Khalique's avatar
Khalique committed
507
                {
508
                    l3 = prog.add_instruction(op::multibroadcast{out_lens}, args[2]);
Khalique's avatar
Khalique committed
509
                }
510
                return prog.add_instruction(op::dot{alpha, beta}, l1, l2, l3);
511
            }
Paul's avatar
Paul committed
512
        }
513

Shucai Xiao's avatar
Shucai Xiao committed
514
        return prog.add_instruction(op::dot{alpha, beta}, l1, l2);
Paul's avatar
Paul committed
515
516
    }

517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
    instruction_ref
    parse_matmul(const std::string&, attribute_map, std::vector<instruction_ref> args)
    {
        auto l0 = args[0];
        auto l1 = args[1];
        auto l0_lens = l0->get_shape().lens();
        auto l1_lens = l1->get_shape().lens();

        // args[0] is a vector, prepend 1 to the shape
        bool is_a_prepended = false;
        if (l0_lens.size() == 1)
        {
            is_a_prepended = true;
            l0_lens.insert(l0_lens.begin(), 1);
            l0 = prog.add_instruction(op::unsqueeze{{0}}, args[0]);
        }

        bool is_b_appended = false;
        if (l1_lens.size() == 1)
        {
            is_b_appended = true;
            l1_lens.push_back(1);
            l1 = prog.add_instruction(op::unsqueeze{{1}}, args[1]);
        }

        instruction_ref bl0 = l0;
        instruction_ref bl1 = l1;
        if (!std::equal(l0_lens.rbegin() + 2, l0_lens.rend(), l1_lens.rbegin() + 2, l1_lens.rend()))
        {
            auto l0_it = l0_lens.begin() + l0_lens.size() - 2;
            std::vector<std::size_t> l0_broadcasted_lens(l0_lens.begin(), l0_it);
            auto l1_it = l1_lens.begin() + l1_lens.size() - 2;
            std::vector<std::size_t> l1_broadcasted_lens(l1_lens.begin(), l1_it);
            auto output_lens = compute_broadcasted_lens(l0_broadcasted_lens, l1_broadcasted_lens);
            l0_broadcasted_lens.insert(l0_broadcasted_lens.end(), l0_it, l0_lens.end());
            l1_broadcasted_lens.insert(l1_broadcasted_lens.end(), l1_it, l1_lens.end());
            if (l0_lens != l0_broadcasted_lens)
            {
                bl0 = prog.add_instruction(op::multibroadcast{l0_broadcasted_lens}, l0);
            }
            if (l1_lens != l1_broadcasted_lens)
            {
                bl1 = prog.add_instruction(op::multibroadcast{l1_broadcasted_lens}, l1);
            }
        }

        auto dot_res = prog.add_instruction(op::dot{1.0f, 0.0f}, bl0, bl1);
        int64_t num_axis = static_cast<int64_t>(dot_res->get_shape().lens().size());
        if (is_a_prepended)
        {
            dot_res = prog.add_instruction(op::squeeze{{num_axis - 2}}, dot_res);
            --num_axis;
        }
        if (is_b_appended)
        {
            dot_res = prog.add_instruction(op::squeeze{{num_axis - 1}}, dot_res);
        }
        
        return dot_res;
    }

578
    instruction_ref
Paul's avatar
Paul committed
579
    parse_batchnorm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
580
    {
Scott Thornton's avatar
Scott Thornton committed
581
582
        float epsilon                                     = 1e-5f;
        float momentum                                    = 0.9f;
583
        op::batch_norm_inference::bn_infer_mode_t bn_mode = op::batch_norm_inference::spatial;
Scott Thornton's avatar
Scott Thornton committed
584
        bool is_test                                      = false;
585
586
587
588
589
590
        if(contains(attributes, "epsilon"))
        {
            epsilon = parse_value(attributes.at("epsilon")).at<float>();
        }
        if(contains(attributes, "momentum"))
        {
591
            momentum = parse_value(attributes.at("momentum")).at<float>();
592
593
594
        }
        if(contains(attributes, "is_test"))
        {
wsttiger's avatar
wsttiger committed
595
            is_test = parse_value(attributes.at("is_test")).at<uint64_t>() > 0;
596
597
598
        }
        if(contains(attributes, "spatial"))
        {
599
            bn_mode = (parse_value(attributes.at("spatial")).at<uint64_t>() > 0)
600
601
                          ? op::batch_norm_inference::spatial
                          : op::batch_norm_inference::per_activation;
602
        }
Paul's avatar
Paul committed
603
        (void)is_test;
Paul's avatar
Paul committed
604
        op::batch_norm_inference op{epsilon, momentum, bn_mode};
Paul's avatar
Paul committed
605
        return prog.add_instruction(op, std::move(args));
606
607
    }

608
609
610
611
    instruction_ref parse_leaky_relu(const std::string&,
                                     attribute_map attributes,
                                     std::vector<instruction_ref> args)
    {
Khalique's avatar
Khalique committed
612
        float alpha = 0.01; // default alpha val for leaky relu
613
614
615
616
617
618
619
620
        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
621
622
    instruction_ref
    parse_elu(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
623
624
625
626
627
628
629
630
631
632
    {
        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
633
634
    instruction_ref
    parse_lrn(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
635
636
    {
        float alpha = 0.0001;
Khalique's avatar
Khalique committed
637
638
639
        float beta  = 0.75;
        float bias  = 1.0;
        int size    = 1;
Khalique's avatar
Khalique committed
640
641
642
643
644
645
646
647
648
649
650
651
        if(contains(attributes, "alpha"))
            alpha = parse_value(attributes.at("alpha")).at<float>();
        if(contains(attributes, "beta"))
            beta = parse_value(attributes.at("beta")).at<float>();
        if(contains(attributes, "bias"))
            bias = parse_value(attributes.at("bias")).at<float>();
        if(contains(attributes, "size"))
            size = parse_value(attributes.at("size")).at<int>();
        op::lrn op{alpha, beta, bias, size};
        return prog.add_instruction(op, args.front());
    }

Khalique's avatar
Khalique committed
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
    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
669

Khalique's avatar
Khalique committed
670
671
        auto scale_val = prog.add_literal(scale);
        auto bias_vals = prog.add_literal(
Paul's avatar
Paul committed
672
            migraphx::literal{migraphx::shape{migraphx::shape::float_type, {bias.size()}}, bias});
Khalique's avatar
Khalique committed
673

Paul's avatar
Paul committed
674
675
        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
676
        auto bias_bcast = prog.add_instruction(migraphx::op::broadcast{1, input_shape}, bias_vals);
Paul's avatar
Paul committed
677
        return prog.add_instruction(migraphx::op::add{}, img_scaled, bias_bcast);
Khalique's avatar
Khalique committed
678
    }
Khalique's avatar
Khalique committed
679

Khalique's avatar
Khalique committed
680
681
    instruction_ref
    parse_transpose(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
Khalique's avatar
Khalique committed
682
683
684
685
686
687
688
    {
        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
689
        return prog.add_instruction(migraphx::op::transpose{perm}, args.front());
Khalique's avatar
Khalique committed
690
691
    }

Khalique's avatar
Khalique committed
692
693
694
695
696
697
698
699
700
701
    instruction_ref
    parse_pad(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
        std::vector<int64_t> pads{};
        float value = 0.0f;
        if(contains(attributes, "pads"))
        {
            auto&& pad_vals = attributes["pads"].ints();
            pads            = std::vector<int64_t>(pad_vals.begin(), pad_vals.end());
        }
702
        // check if padding is actually being done (at least one value is nonzero)
Khalique's avatar
Khalique committed
703
        if(std::all_of(pads.begin(), pads.end(), [](const int& i) { return i == 0; }))
704
705
706
        {
            return prog.add_instruction(migraphx::op::identity{}, args.front());
        }
Khalique's avatar
Khalique committed
707
708
709
710
711
712
713
714
715
716
717
718
        if(contains(attributes, "value"))
        {
            value = parse_value(attributes.at("value")).at<float>();
        }
        if(contains(attributes, "mode"))
        {
            auto mode = attributes.at("mode").s();
            if(mode != "constant")
                MIGRAPHX_THROW("migraphx currently only supports constant padding");
        }
        return prog.add_instruction(migraphx::op::pad{pads, value}, args.front());
    }
719
720
721
    // Use a literal instruction to replace the shape since, output of
    // shape operator are literals in migraphx
    instruction_ref
Shucai Xiao's avatar
Shucai Xiao committed
722
    parse_shape(const std::string&, const attribute_map&, std::vector<instruction_ref> args)
723
724
    {
        if(args.size() != 1)
725
            MIGRAPHX_THROW("Shape: operator should have 1 operand");
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
        std::vector<std::size_t> arg_shape = args[0]->get_shape().lens();
        std::vector<int64_t> vec_shape(arg_shape.size());
        migraphx::shape s(migraphx::shape::int64_type, {arg_shape.size()});
        std::transform(arg_shape.begin(), arg_shape.end(), vec_shape.begin(), [](auto i) {
            return int64_t(i);
        });
        return prog.add_literal(migraphx::literal{s, vec_shape});
    }

    // Use a literal instruction to replace the constantFill operator. In RNN, input shape
    // and value are fixed, so no need to do the actual computation for the constantFill
    // operator
    instruction_ref parse_constant_fill(const std::string&,
                                        attribute_map attributes,
                                        std::vector<instruction_ref> args)
    {
        int input_as_shape = 0;
        int dtype          = 1;
        float value        = 0.0f;

        if(contains(attributes, "dtype"))
        {
            dtype = parse_value(attributes.at("dtype")).at<int>();
        }
        migraphx::shape::type_t type = get_type(dtype);

        if(contains(attributes, "input_as_shape"))
        {
            input_as_shape = parse_value(attributes.at("input_as_shape")).at<int>();
        }

        if(contains(attributes, "value"))
        {
            value = parse_value(attributes.at("value")).at<float>();
        }

Shucai Xiao's avatar
Shucai Xiao committed
762
763
        if(contains(attributes, "extra_shape"))
        {
764
            MIGRAPHX_THROW("ConstantFill: cannot handle extra shape attribute");
765
766
        }

767
768
        if(input_as_shape == 1)
        {
Shucai Xiao's avatar
Shucai Xiao committed
769
            if(args.size() != 1)
770
            {
771
                MIGRAPHX_THROW("ConstantFill: need an input argument as output shape");
772
773
            }

Shucai Xiao's avatar
Shucai Xiao committed
774
775
            if(contains(attributes, "shape"))
            {
776
                MIGRAPHX_THROW("ConstantFill: cannot set the shape argument and pass in an input "
Shucai Xiao's avatar
Shucai Xiao committed
777
                               "at the same time");
778
779
            }

780
781
782
            migraphx::argument in = args[0]->eval();
            if(in.empty())
            {
783
                MIGRAPHX_THROW("ConstantFill: cannot handle dynamic shape as input");
784
            }
785

786
787
788
            std::vector<std::size_t> dims;
            in.visit([&](auto input) { dims.assign(input.begin(), input.end()); });
            migraphx::shape s(type, dims);
789
790
            std::vector<float> values(s.elements(), value);
            return prog.add_literal(migraphx::literal(s, values));
791
792
793
        }
        else if(input_as_shape == 0)
        {
Shucai Xiao's avatar
Shucai Xiao committed
794
795
            if(!contains(attributes, "shape"))
            {
796
                MIGRAPHX_THROW("ConstantFill: attribute output shape is needed");
797
798
799
            }

            literal ls = parse_value(attributes.at("shape"));
800
            std::vector<std::size_t> dims;
Shucai Xiao's avatar
Shucai Xiao committed
801
            ls.visit([&](auto s) { dims.assign(s.begin(), s.end()); });
802
            migraphx::shape s{type, dims};
803
804
            std::vector<float> values(s.elements(), value);
            return prog.add_literal(migraphx::literal(s, values));
805
806
807
        }
        else
        {
808
            MIGRAPHX_THROW("ConstantFill: wrong value of attribute input_as_shape");
809
810
811
        }
    }

Shucai Xiao's avatar
Shucai Xiao committed
812
    std::vector<instruction_ref>
Shucai Xiao's avatar
Shucai Xiao committed
813
814
815
    parse_rnn(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
        migraphx::shape input_shape = args[0]->get_shape();
816
        std::size_t hidden_size     = args[1]->get_shape().lens()[1];
Shucai Xiao's avatar
Shucai Xiao committed
817
818
819

        if(contains(attributes, "hidden_size"))
        {
Shucai Xiao's avatar
Shucai Xiao committed
820
            std::size_t hidden_size_att = parse_value(attributes.at("hidden_size")).at<int>();
Shucai Xiao's avatar
Shucai Xiao committed
821
            if(hidden_size != hidden_size_att)
Shucai Xiao's avatar
Shucai Xiao committed
822
823
824
            {
                MIGRAPHX_THROW("RNN: hidden size mismatch in input and attribute");
            }
Shucai Xiao's avatar
Shucai Xiao committed
825
826
827
828
829
830
831
832
833
        }

        // Handling of direction to be added later
        std::string direction{"forward"};
        if(contains(attributes, "direction"))
        {
            direction = attributes.at("direction").s();
        }

834
        op::rnn_direction dirct = op::rnn_direction::forward;
Shucai Xiao's avatar
Shucai Xiao committed
835
836
        if(direction == "bidirectional")
        {
837
            dirct = op::rnn_direction::bidirectional;
Shucai Xiao's avatar
Shucai Xiao committed
838
839
840
        }
        else if(direction == "reverse")
        {
841
            dirct = op::rnn_direction::reverse;
Shucai Xiao's avatar
Shucai Xiao committed
842
843
        }

844
845
846
847
848
        std::vector<std::string> vec_names{"tanh"};
        if(contains(attributes, "activations"))
        {
            auto names = attributes.at("activations").strings();
            vec_names.clear();
849
            vec_names.resize(names.size());
850
            std::copy(names.begin(), names.end(), vec_names.begin());
851
852
        }

853
854
        auto name_it = std::find_if(vec_names.begin(), vec_names.end(), [&](auto& name) {
            return (map_actv_funcs.count(name) == 0);
855
        });
Shucai Xiao's avatar
Shucai Xiao committed
856
        if(name_it != vec_names.end())
857
858
859
        {
            MIGRAPHX_THROW("RNN: activation function " + std::string(*name_it) + " not supported");
        }
860

Shucai Xiao's avatar
Shucai Xiao committed
861
        // bidirectional case should have two activation functions.
Shucai Xiao's avatar
Shucai Xiao committed
862
        // one is for forward, and the other is for reverse.
Shucai Xiao's avatar
Shucai Xiao committed
863
        // if only one actv function is provided, we use it in both
864
        // forward and reverse direction
865
        if(dirct == op::rnn_direction::bidirectional)
866
        {
Shucai Xiao's avatar
Shucai Xiao committed
867
            if(vec_names.size() == 1)
868
869
870
871
872
            {
                vec_names.push_back(vec_names.at(0));
            }
        }

Shucai Xiao's avatar
Shucai Xiao committed
873
874
875
        std::vector<operation> vec_actv_funcs(vec_names.size());
        std::transform(vec_names.begin(), vec_names.end(), vec_actv_funcs.begin(), [&](auto& fn) {
            return map_actv_funcs[fn];
876
        });
Shucai Xiao's avatar
Shucai Xiao committed
877

Shucai Xiao's avatar
Shucai Xiao committed
878
879
880
881
882
883
884
        // To be added later
        float clip = 0.0;
        if(contains(attributes, "clip"))
        {
            clip = parse_value(attributes.at("clip")).at<float>();
        }

885
886
        // if the number of arguments is less than 6, append
        // undefined operator to have 6 arguments
Shucai Xiao's avatar
Shucai Xiao committed
887
        if(args.size() < 6)
888
889
890
891
892
        {
            auto ins = prog.add_instruction(op::undefined{});
            args.insert(args.end(), (6 - args.size()), ins);
        }

Shucai Xiao's avatar
Shucai Xiao committed
893
894
        // first output for the concatenation of hidden states
        auto hidden_states = prog.add_instruction(op::rnn{hidden_size, vec_actv_funcs, dirct, clip},
Shucai Xiao's avatar
Shucai Xiao committed
895
                                                  std::move(args));
Shucai Xiao's avatar
Shucai Xiao committed
896

897
        // second output for the last hidden state
Shucai Xiao's avatar
Shucai Xiao committed
898
        auto last_output = prog.add_instruction(op::rnn_last_output{}, hidden_states);
Shucai Xiao's avatar
Shucai Xiao committed
899

Shucai Xiao's avatar
Shucai Xiao committed
900
        return {hidden_states, last_output};
Shucai Xiao's avatar
Shucai Xiao committed
901
902
    }

903
    std::vector<instruction_ref>
904
905
906
907
908
909
910
    parse_gru(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
        migraphx::shape input_shape = args[0]->get_shape();
        std::size_t hidden_size     = args[2]->get_shape().lens()[2];

        if(contains(attributes, "hidden_size"))
        {
Shucai Xiao's avatar
Shucai Xiao committed
911
            std::size_t hidden_size_att = parse_value(attributes.at("hidden_size")).at<int>();
Shucai Xiao's avatar
Shucai Xiao committed
912
            if(hidden_size != hidden_size_att)
Shucai Xiao's avatar
Shucai Xiao committed
913
914
915
            {
                MIGRAPHX_THROW("GRU: hidden size mismatch in input and attribute");
            }
916
917
918
919
920
921
922
923
924
        }

        // Handling of direction to be added later
        std::string direction{"forward"};
        if(contains(attributes, "direction"))
        {
            direction = attributes.at("direction").s();
        }

925
        op::rnn_direction dirct = op::rnn_direction::forward;
926
927
        if(direction == "bidirectional")
        {
928
            dirct = op::rnn_direction::bidirectional;
929
930
931
        }
        else if(direction == "reverse")
        {
932
            dirct = op::rnn_direction::reverse;
933
934
        }

935
        std::vector<std::string> vec_names = {"sigmoid", "tanh"};
936
937
        if(contains(attributes, "activations"))
        {
938
            auto names = attributes.at("activations").strings();
939
            vec_names.clear();
Shucai Xiao's avatar
Shucai Xiao committed
940
            vec_names.resize(names.size());
941
            std::copy(names.begin(), names.end(), vec_names.begin());
942
943
        }

944
        // need 4 activation functions
945
        if(dirct == op::rnn_direction::bidirectional)
946
        {
Shucai Xiao's avatar
Shucai Xiao committed
947
            // 4 activation functions are used in the bidirectional
948
            // scenario. No spec is provided in onnx::operator. we
Shucai Xiao's avatar
Shucai Xiao committed
949
950
            // use the algorithm that: if 1 actv function is provided,
            // repeat 1 four times. If 2 actv functins are provided,
951
952
            // assume forward and reverse use the same pair of actv
            // functions. For the case of 3 actv functions provided,
Shucai Xiao's avatar
Shucai Xiao committed
953
954
955
            // assume the 3rd one is repeated once and used by the
            // reverse direction.
            // This may need change later
956
            if(vec_names.size() == 1)
957
            {
958
                vec_names.insert(vec_names.end(), 3, vec_names.at(0));
959
            }
960
            else if(vec_names.size() == 2)
961
            {
962
963
964
                // repeat the activation functions
                vec_names.push_back(vec_names.at(0));
                vec_names.push_back(vec_names.at(1));
965
            }
966
            else if(vec_names.size() == 3)
967
            {
968
                vec_names.push_back(vec_names.at(2));
969
970
            }
        }
Shucai Xiao's avatar
Shucai Xiao committed
971
        else
972
        {
973
            if(vec_names.size() == 1)
974
            {
975
                vec_names.push_back(vec_names.at(0));
976
977
978
            }
        }

979
980
        auto name_it = std::find_if(vec_names.begin(), vec_names.end(), [&](auto& name) {
            return (map_actv_funcs.count(name) == 0);
Shucai Xiao's avatar
Shucai Xiao committed
981
        });
Shucai Xiao's avatar
Shucai Xiao committed
982
983
        if(name_it != vec_names.end())
        {
984
985
            MIGRAPHX_THROW("GRU: activation function " + std::string(*name_it) + " not supported");
        }
986

Shucai Xiao's avatar
Shucai Xiao committed
987
988
989
        std::vector<operation> vec_actv_funcs(vec_names.size());
        std::transform(vec_names.begin(), vec_names.end(), vec_actv_funcs.begin(), [&](auto& name) {
            return map_actv_funcs[name];
Shucai Xiao's avatar
Shucai Xiao committed
990
        });
991
992
993
994
995
996
997
998

        float clip = 0.0;
        if(contains(attributes, "clip"))
        {
            clip = parse_value(attributes.at("clip")).at<float>();
        }

        int linear_before_reset = 0;
Shucai Xiao's avatar
Shucai Xiao committed
999
        if(contains(attributes, "linear_before_reset"))
1000
1001
1002
1003
        {
            linear_before_reset = parse_value(attributes.at("linear_before_reset")).at<int>();
        }

Shucai Xiao's avatar
Shucai Xiao committed
1004
        // append undefined opeator to make 6 arguments
Shucai Xiao's avatar
Shucai Xiao committed
1005
        if(args.size() < 6)
Shucai Xiao's avatar
Shucai Xiao committed
1006
1007
1008
1009
1010
        {
            auto ins = prog.add_instruction(op::undefined{});
            args.insert(args.end(), 6 - args.size(), ins);
        }

1011
1012
        // first output for concatenation of hidden states
        auto hidden_states = prog.add_instruction(
Shucai Xiao's avatar
Shucai Xiao committed
1013
            op::gru{hidden_size, vec_actv_funcs, dirct, clip, linear_before_reset},
Shucai Xiao's avatar
Shucai Xiao committed
1014
            std::move(args));
1015
1016

        // second output for last gru output
1017
        auto last_output = prog.add_instruction(op::rnn_last_output{}, hidden_states);
1018

Shucai Xiao's avatar
Shucai Xiao committed
1019
        return {hidden_states, last_output};
1020
1021
    }

Shucai Xiao's avatar
Shucai Xiao committed
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
    std::vector<instruction_ref>
    parse_lstm(const std::string&, attribute_map attributes, std::vector<instruction_ref> args)
    {
        migraphx::shape input_shape = args[0]->get_shape();
        std::size_t hidden_size     = args[2]->get_shape().lens()[2];

        if(contains(attributes, "hidden_size"))
        {
            std::size_t hidden_size_att = parse_value(attributes.at("hidden_size")).at<int>();
            if(hidden_size != hidden_size_att)
            {
                MIGRAPHX_THROW("LSTM: hidden size mismatch in input and attribute");
            }
        }

        // Handling of direction to be added later
        std::string direction{"forward"};
        if(contains(attributes, "direction"))
        {
            direction = attributes.at("direction").s();
        }

Shucai Xiao's avatar
Shucai Xiao committed
1044
        op::rnn_direction dirct = op::rnn_direction::forward;
Shucai Xiao's avatar
Shucai Xiao committed
1045
1046
        if(direction == "bidirectional")
        {
Shucai Xiao's avatar
Shucai Xiao committed
1047
            dirct = op::rnn_direction::bidirectional;
Shucai Xiao's avatar
Shucai Xiao committed
1048
1049
1050
        }
        else if(direction == "reverse")
        {
Shucai Xiao's avatar
Shucai Xiao committed
1051
            dirct = op::rnn_direction::reverse;
Shucai Xiao's avatar
Shucai Xiao committed
1052
        }
Shucai Xiao's avatar
Shucai Xiao committed
1053
        else if(direction == "forward")
Shucai Xiao's avatar
Shucai Xiao committed
1054
        {
Shucai Xiao's avatar
Shucai Xiao committed
1055
            dirct = op::rnn_direction::forward;
Shucai Xiao's avatar
Shucai Xiao committed
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
        }
        else
        {
            MIGRAPHX_THROW("LSTM: incorrect direction attribute");
        }

        std::vector<std::string> vec_names = {"sigmoid", "tanh", "tanh"};
        if(contains(attributes, "activations"))
        {
            auto names = attributes.at("activations").strings();
            vec_names.clear();
            vec_names.resize(names.size());
1068
            std::copy(names.begin(), names.end(), vec_names.begin());
Shucai Xiao's avatar
Shucai Xiao committed
1069
1070
1071
        }

        // need 6 activation functions for bidirectional directions
Shucai Xiao's avatar
Shucai Xiao committed
1072
        if(dirct == op::rnn_direction::bidirectional)
Shucai Xiao's avatar
Shucai Xiao committed
1073
1074
1075
1076
1077
1078
        {
            // 6 activation functions are used in the bidirectional
            // scenario. No spec is provided in onnx::operator. we
            // use the algorithm that: if 1 actv function is provided,
            // repeat 1st six times. If 2 actv functins are provided,
            // repeat 2nd once, then repeat all three once
Shucai Xiao's avatar
Shucai Xiao committed
1079
            // if 3 actv funcs are provide, repeat all three once.
Shucai Xiao's avatar
Shucai Xiao committed
1080
1081
1082
1083
            // the same algorithm is used for 4, 5, and 6 actv funcions
            // provided. This may need change later
            switch(vec_names.size())
            {
1084
            case 1:
Shucai Xiao's avatar
Shucai Xiao committed
1085
1086
1087
1088
1089
1090
                vec_names = {vec_names.at(0),
                             vec_names.at(0),
                             vec_names.at(0),
                             vec_names.at(0),
                             vec_names.at(0),
                             vec_names.at(0)};
1091
                break;
Shucai Xiao's avatar
Shucai Xiao committed
1092
1093
1094

            case 2:
                // repeat the 2nd actv func once, then repeat all three another time
Shucai Xiao's avatar
Shucai Xiao committed
1095
1096
1097
1098
1099
1100
                vec_names = {vec_names.at(0),
                             vec_names.at(1),
                             vec_names.at(1),
                             vec_names.at(0),
                             vec_names.at(1),
                             vec_names.at(1)};
Shucai Xiao's avatar
Shucai Xiao committed
1101
1102
1103
1104
                break;

            case 3:
                // repeat all three actv funcs once
Shucai Xiao's avatar
Shucai Xiao committed
1105
1106
1107
1108
1109
1110
                vec_names = {vec_names.at(0),
                             vec_names.at(1),
                             vec_names.at(2),
                             vec_names.at(0),
                             vec_names.at(1),
                             vec_names.at(2)};
Shucai Xiao's avatar
Shucai Xiao committed
1111
1112
                break;

Shucai Xiao's avatar
Shucai Xiao committed
1113
1114
1115
1116
1117
1118
1119
            case 4:
                vec_names = {vec_names.at(0),
                             vec_names.at(1),
                             vec_names.at(2),
                             vec_names.at(3),
                             vec_names.at(3),
                             vec_names.at(3)};
1120
                break;
Shucai Xiao's avatar
Shucai Xiao committed
1121

Shucai Xiao's avatar
Shucai Xiao committed
1122
1123
1124
1125
1126
1127
1128
            case 5:
                vec_names = {vec_names.at(0),
                             vec_names.at(1),
                             vec_names.at(2),
                             vec_names.at(3),
                             vec_names.at(4),
                             vec_names.at(4)};
1129
                break;
Shucai Xiao's avatar
Shucai Xiao committed
1130

Shucai Xiao's avatar
Shucai Xiao committed
1131
            default: break;
Shucai Xiao's avatar
Shucai Xiao committed
1132
1133
1134
1135
1136
1137
            }
        }
        else
        {
            switch(vec_names.size())
            {
Shucai Xiao's avatar
Shucai Xiao committed
1138
            case 1: vec_names = {vec_names.at(0), vec_names.at(0), vec_names.at(0)}; break;
Shucai Xiao's avatar
Shucai Xiao committed
1139
1140
1141

            case 2:
                // repeat the 2nd actv func once, so we have 3 actv funcs
Shucai Xiao's avatar
Shucai Xiao committed
1142
                vec_names = {vec_names.at(0), vec_names.at(1), vec_names.at(1)};
Shucai Xiao's avatar
Shucai Xiao committed
1143
1144
                break;

Shucai Xiao's avatar
Shucai Xiao committed
1145
            default: break;
Shucai Xiao's avatar
Shucai Xiao committed
1146
1147
1148
            }
        }

1149
1150
1151
        auto name_it = std::find_if(vec_names.begin(), vec_names.end(), [&](auto& name) {
            return (map_actv_funcs.count(name) == 0);
        });
Shucai Xiao's avatar
Shucai Xiao committed
1152
        if(name_it != vec_names.end())
1153
1154
1155
        {
            MIGRAPHX_THROW("LSTM: activation function " + std::string(*name_it) + " not supported");
        }
Shucai Xiao's avatar
Shucai Xiao committed
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177

        std::vector<operation> vec_actv_funcs(vec_names.size());
        std::transform(vec_names.begin(), vec_names.end(), vec_actv_funcs.begin(), [&](auto& name) {
            return map_actv_funcs[name];
        });

        float clip = 0.0;
        if(contains(attributes, "clip"))
        {
            clip = parse_value(attributes.at("clip")).at<float>();
        }

        int input_forget = 0;
        if(contains(attributes, "input_forget"))
        {
            input_forget = parse_value(attributes.at("input_forget")).at<int>();
        }

        // append undefined opeator to make 6 arguments
        if(args.size() < 8)
        {
            auto ins = prog.add_instruction(op::undefined{});
Shucai Xiao's avatar
Shucai Xiao committed
1178
            args.insert(args.end(), 8 - args.size(), ins);
Shucai Xiao's avatar
Shucai Xiao committed
1179
1180
1181
1182
        }

        // first output for concatenation of hidden states
        auto hidden_states = prog.add_instruction(
Shucai Xiao's avatar
Shucai Xiao committed
1183
            op::lstm{hidden_size, vec_actv_funcs, dirct, clip, input_forget}, std::move(args));
Shucai Xiao's avatar
Shucai Xiao committed
1184
1185

        // second output for last lstm output
Shucai Xiao's avatar
Shucai Xiao committed
1186
        auto last_output = prog.add_instruction(op::rnn_last_output{}, hidden_states);
Shucai Xiao's avatar
Shucai Xiao committed
1187
1188
1189
1190
1191
1192
1193

        // third output for last cell output
        auto last_cell_output = prog.add_instruction(op::lstm_last_cell_output{}, hidden_states);

        return {hidden_states, last_output, last_cell_output};
    }

Paul's avatar
Paul committed
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
    void parse_from(std::istream& is)
    {
        onnx::ModelProto model;
        if(model.ParseFromIstream(&is))
        {
            if(model.has_graph())
            {
                this->parse_graph(model.graph());
            }
        }
        else
        {
Paul's avatar
Paul committed
1206
            MIGRAPHX_THROW("Failed reading onnx file.");
Paul's avatar
Paul committed
1207
1208
1209
1210
1211
1212
        }
    }

    void parse_graph(const onnx::GraphProto& graph)
    {
        nodes = get_nodes(graph);
1213
1214
1215
1216
1217
        std::unordered_map<std::string, onnx::TensorProto> initializer_data;
        for(auto&& f : graph.initializer())
        {
            initializer_data[f.name()] = f;
        }
Paul's avatar
Paul committed
1218
1219
1220
        for(auto&& input : graph.input())
        {
            const std::string& name = input.name();
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
            // 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
1233
        }
Paul's avatar
Paul committed
1234
        for(auto&& output : graph.output())
Paul's avatar
Paul committed
1235
        {
Paul's avatar
Paul committed
1236
            this->parse_node(output.name());
Paul's avatar
Paul committed
1237
1238
1239
        }
    }

Shucai Xiao's avatar
Shucai Xiao committed
1240
    void parse_undefined(const std::string& name)
1241
    {
Shucai Xiao's avatar
Shucai Xiao committed
1242
        auto ins           = prog.add_instruction(op::undefined{});
1243
1244
1245
        instructions[name] = ins;
    }

Paul's avatar
Paul committed
1246
    void parse_node(const std::string& name)
Paul's avatar
Paul committed
1247
    {
Paul's avatar
Paul committed
1248
        if(name.empty())
Paul's avatar
Paul committed
1249
            MIGRAPHX_THROW("Onnx node must have a name");
Paul's avatar
Paul committed
1250
1251
1252
1253
1254
1255
1256
1257
        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)
                {
Paul's avatar
Paul committed
1258
1259
                    assert(name != input);
                    this->parse_node(input);
Paul's avatar
Paul committed
1260
                }
Shucai Xiao's avatar
Shucai Xiao committed
1261
                else if(input.empty())
Paul's avatar
Paul committed
1262
                {
1263
                    this->parse_undefined(input);
Paul's avatar
Paul committed
1264
                }
1265
                args.push_back(instructions.at(input));
Paul's avatar
Paul committed
1266
            }
Paul's avatar
Paul committed
1267
            std::vector<instruction_ref> result;
Paul's avatar
Paul committed
1268
1269
            if(ops.count(node.op_type()) == 0)
            {
1270
                result.push_back(prog.add_instruction(op::unknown{node.op_type()}, args));
Paul's avatar
Paul committed
1271
1272
1273
            }
            else
            {
Paul's avatar
Paul committed
1274
                result = ops[node.op_type()](get_attributes(node), args);
Paul's avatar
Paul committed
1275
            }
Paul's avatar
Paul committed
1276
            // Even no output nodes produce output in migraphx
Paul's avatar
Paul committed
1277
            if(node.output().empty() and result.size() == 1)
Paul's avatar
Paul committed
1278
1279
            {
                instructions[name] = result.front();
Paul's avatar
Paul committed
1280
1281
1282
            }
            else
            {
Paul's avatar
Paul committed
1283
1284
1285
1286
1287
1288
                assert(node.output().size() >= result.size());
                std::transform(result.begin(),
                               result.end(),
                               node.output().begin(),
                               std::inserter(instructions, instructions.end()),
                               [](auto&& x, auto&& y) { return std::make_pair(y, x); });
Paul's avatar
Paul committed
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
            }
        }
    }

    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;
    }

    static node_map get_nodes(const onnx::GraphProto& graph)
    {
        std::unordered_map<std::string, onnx::NodeProto> result;
Paul's avatar
Paul committed
1306
        std::size_t n = 0;
Paul's avatar
Paul committed
1307
1308
        for(auto&& node : graph.node())
        {
Paul's avatar
Paul committed
1309
            if(node.output().empty())
Paul's avatar
Paul committed
1310
            {
Paul's avatar
Paul committed
1311
                if(node.name().empty())
Paul's avatar
Paul committed
1312
1313
1314
1315
1316
1317
1318
1319
1320
                {
                    result["migraphx_unamed_node_" + std::to_string(n)] = node;
                    n++;
                }
                else
                {
                    result[node.name()] = node;
                }
            }
Paul's avatar
Paul committed
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
            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
1346
        case onnx::AttributeProto::FLOATS: return from_repeated(shape::float_type, attr.floats());
Paul's avatar
Paul committed
1347
1348
1349
1350
1351
        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
1352
        MIGRAPHX_THROW("Invalid attribute type");
Paul's avatar
Paul committed
1353
1354
1355
1356
1357
    }

    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
1358
        // in case of scalar constants in onnx file, use dims=1 to fill initializer data
1359
        if(dims.empty())
Khalique's avatar
Khalique committed
1360
1361
1362
        {
            dims = {1};
        }
1363
1364
        if(t.has_raw_data())
        {
wsttiger's avatar
wsttiger committed
1365
            const std::string& s = t.raw_data();
Scott Thornton's avatar
Scott Thornton committed
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
            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
1378
            case onnx::TensorProto::FLOAT16: return literal{{shape::half_type, dims}, s.data()};
Scott Thornton's avatar
Scott Thornton committed
1379
1380
1381
1382
1383
1384
            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
1385
            MIGRAPHX_THROW("Invalid tensor type");
1386
        }
Paul's avatar
Paul committed
1387
1388
1389
1390
        switch(t.data_type())
        {
        case onnx::TensorProto::UNDEFINED: throw std::runtime_error("");
        case onnx::TensorProto::FLOAT:
Paul's avatar
Paul committed
1391
            return literal{{shape::float_type, dims}, t.float_data().begin(), t.float_data().end()};
Paul's avatar
Paul committed
1392
1393
        case onnx::TensorProto::UINT8: throw std::runtime_error("");
        case onnx::TensorProto::INT8:
Paul's avatar
Paul committed
1394
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
1395
        case onnx::TensorProto::UINT16:
Paul's avatar
Paul committed
1396
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
1397
        case onnx::TensorProto::INT16:
Paul's avatar
Paul committed
1398
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
1399
        case onnx::TensorProto::INT32:
Paul's avatar
Paul committed
1400
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
1401
        case onnx::TensorProto::INT64:
Paul's avatar
Paul committed
1402
            return literal{{shape::int64_type, dims}, t.int64_data().begin(), t.int64_data().end()};
Paul's avatar
Paul committed
1403
1404
        case onnx::TensorProto::STRING: throw std::runtime_error("");
        case onnx::TensorProto::BOOL:
Paul's avatar
Paul committed
1405
            return literal{{shape::int32_type, dims}, t.int32_data().begin(), t.int32_data().end()};
Paul's avatar
Paul committed
1406
        case onnx::TensorProto::FLOAT16:
Khalique's avatar
Khalique committed
1407
        {
Khalique's avatar
Khalique committed
1408
            std::vector<uint16_t> data_uint16(t.int32_data().begin(), t.int32_data().end());
1409
            std::vector<half> data_half;
Khalique's avatar
Khalique committed
1410
1411
1412
            std::transform(data_uint16.begin(),
                           data_uint16.end(),
                           std::back_inserter(data_half),
1413
                           [](uint16_t raw_val) { return *reinterpret_cast<half*>(&raw_val); });
1414
            return literal{{shape::half_type, dims}, data_half.begin(), data_half.end()};
Khalique's avatar
Khalique committed
1415
        }
Paul's avatar
Paul committed
1416
1417
1418
1419
1420
1421
1422
1423
        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
1424
        MIGRAPHX_THROW("Invalid tensor type");
Paul's avatar
Paul committed
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
    }

    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
1446
        case onnx::TensorProto::FLOAT16: shape_type = shape::half_type; break;
Paul's avatar
Paul committed
1447
1448
1449
1450
1451
1452
1453
1454
1455
        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
1456
        auto&& tensor_dims = t.tensor_type().shape().dim();
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
        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
1468
1469
        return {shape_type, dims};
    }
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491

    shape::type_t get_type(int dtype)
    {
        switch(dtype)
        {
        case 1: return shape::float_type;
        case 2: return shape::uint8_type;
        case 3: return shape::int8_type;
        case 4: return shape::uint16_type;
        case 5: return shape::int16_type;
        case 6: return shape::int32_type;
        case 7: return shape::int64_type;
        case 10: return shape::half_type;
        case 11: return shape::double_type;
        case 12: return shape::uint32_type;
        case 13: return shape::uint64_type;
        default:
        {
            MIGRAPHX_THROW("Prototensor data type " + std::to_string(dtype) + " not supported");
        }
        }
    }
Paul's avatar
Paul committed
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
};

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
1515
} // namespace MIGRAPHX_INLINE_NS
Paul's avatar
Paul committed
1516
} // namespace migraphx