program.cpp 30.4 KB
Newer Older
Paul's avatar
Paul committed
1
2
3
#include <migraphx/program.hpp>
#include <migraphx/stringutils.hpp>
#include <migraphx/instruction.hpp>
4
#include <migraphx/op/identity.hpp>
Paul's avatar
Paul committed
5
#include <migraphx/target.hpp>
Paul's avatar
Paul committed
6
7
8
#include <migraphx/env.hpp>
#include <migraphx/ranges.hpp>
#include <migraphx/time.hpp>
9
#include <migraphx/pass_manager.hpp>
10
#include <migraphx/register_target.hpp>
Shucai Xiao's avatar
Shucai Xiao committed
11
#include <migraphx/iterator_for.hpp>
12
#include <migraphx/iterator.hpp>
13
#include <migraphx/algorithm.hpp>
14
#include <migraphx/output_iterator.hpp>
Shucai Xiao's avatar
Shucai Xiao committed
15
#include <migraphx/make_op.hpp>
16
#include <migraphx/marker.hpp>
Paul's avatar
Paul committed
17
#include <iostream>
Paul's avatar
Paul committed
18
#include <sstream>
Paul's avatar
Paul committed
19
#include <algorithm>
20
#include <set>
Paul's avatar
Paul committed
21
#include <utility>
22

23
#include <unordered_set>
Shucai Xiao's avatar
Shucai Xiao committed
24
25
#include <map>
#include <cassert>
Paul's avatar
Paul committed
26

Paul's avatar
Paul committed
27
namespace migraphx {
Paul's avatar
Paul committed
28
inline namespace MIGRAPHX_INLINE_NS {
Paul's avatar
Paul committed
29

30
31
using milliseconds = std::chrono::duration<double, std::milli>;

Paul's avatar
Paul committed
32
33
struct program_impl
{
Shucai Xiao's avatar
Shucai Xiao committed
34
    // A map is used to keep references to modules of the program
35
    std::unordered_map<std::string, module> modules;
Paul's avatar
Paul committed
36
    context ctx;
37
    std::string target_name;
Paul's avatar
Paul committed
38
39
};

40
program::program() : impl(std::make_unique<program_impl>()) { this->create_module("main"); }
Paul's avatar
Paul committed
41

Paul's avatar
Paul committed
42
program::program(program&&) noexcept = default;
Shucai Xiao's avatar
Shucai Xiao committed
43
program::~program() noexcept         = default;
Paul's avatar
Paul committed
44

45
// copy constructor
Shucai Xiao's avatar
Shucai Xiao committed
46
program::program(const program& p) { assign(p); }
47
48

// copy assignment operator
Shucai Xiao's avatar
Shucai Xiao committed
49
program& program::operator=(program p)
50
{
Shucai Xiao's avatar
Shucai Xiao committed
51
    std::swap(p.impl, this->impl);
52
53
54
    return *this;
}

Shucai Xiao's avatar
Shucai Xiao committed
55
void program::assign(const program& p)
56
{
Shucai Xiao's avatar
Shucai Xiao committed
57
    if(!impl)
58
59
60
    {
        impl = std::make_unique<program_impl>();
    }
Shucai Xiao's avatar
Shucai Xiao committed
61
    else if(!impl->modules.empty())
62
    {
Shucai Xiao's avatar
Shucai Xiao committed
63
        impl->modules.clear();
64
    }
Shucai Xiao's avatar
Shucai Xiao committed
65

66
    impl->ctx         = p.impl->ctx;
Shucai Xiao's avatar
Shucai Xiao committed
67
68
    impl->target_name = p.impl->target_name;
    impl->modules     = p.impl->modules;
Shucai Xiao's avatar
Shucai Xiao committed
69
70
71
72

    // build a map from old ins to new ins
    // Build a map from old module to new module
    std::unordered_map<module_ref, module_ref> mod_map;
Paul's avatar
Paul committed
73
74
75
76
77
    std::transform(
        impl->modules.begin(),
        impl->modules.end(),
        std::inserter(mod_map, mod_map.begin()),
        [&](auto&& xp) { return std::make_pair(&p.impl->modules.at(xp.first), &xp.second); });
Shucai Xiao's avatar
Shucai Xiao committed
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93

    std::unordered_map<instruction_ref, instruction_ref> ins_map;
    for(auto&& pp : mod_map)
    {
        auto old_ins = iterator_for(*pp.first);
        auto new_ins = iterator_for(*pp.second);
        std::transform(old_ins.begin(),
                       old_ins.end(),
                       new_ins.begin(),
                       std::inserter(ins_map, ins_map.begin()),
                       [](auto x, auto y) { return std::make_pair(x, y); });
    }

    // Update all references from all modules
    for(auto&& mp : impl->modules)
    {
94
        for(auto ins : iterator_for(mp.second))
Shucai Xiao's avatar
Shucai Xiao committed
95
96
            instruction::replace_refs(ins, ins_map, mod_map);
    }
97
98
}

Paul's avatar
Paul committed
99
shape program::get_parameter_shape(std::string name) const
Paul's avatar
Paul committed
100
{
Shucai Xiao's avatar
Shucai Xiao committed
101
102
    const auto* mm = this->get_main_module();
    return mm->get_parameter_shape(std::move(name));
Paul's avatar
Paul committed
103
104
}

105
106
std::vector<std::string> program::get_parameter_names() const
{
Shucai Xiao's avatar
Shucai Xiao committed
107
108
    const auto* mm = this->get_main_module();
    return mm->get_parameter_names();
109
110
}

mei-ye's avatar
mei-ye committed
111
112
instruction_ref program::get_parameter(std::string name) const
{
Shucai Xiao's avatar
Shucai Xiao committed
113
114
    const auto* mm = this->get_main_module();
    return mm->get_parameter(std::move(name));
mei-ye's avatar
mei-ye committed
115
116
}

Paul's avatar
Paul committed
117
118
std::unordered_map<std::string, shape> program::get_parameter_shapes() const
{
Shucai Xiao's avatar
Shucai Xiao committed
119
120
    const auto* mm = this->get_main_module();
    return mm->get_parameter_shapes();
Paul's avatar
Paul committed
121
122
}

Shucai Xiao's avatar
Shucai Xiao committed
123
std::size_t program::size() const { return impl->modules.size(); }
124

125
126
std::vector<shape> program::get_output_shapes() const
{
Shucai Xiao's avatar
Shucai Xiao committed
127
128
    const auto* mm = this->get_main_module();
    return mm->get_output_shapes();
129
}
Paul's avatar
Paul committed
130

Paul's avatar
Paul committed
131
132
context& program::get_context() const { return impl->ctx; }

Paul's avatar
Paul committed
133
134
instruction_ref program::validate() const
{
Shucai Xiao's avatar
Shucai Xiao committed
135
136
    const auto* mm = this->get_main_module();
    return mm->validate();
Paul's avatar
Paul committed
137
138
}

139
140
bool program::is_compiled() const { return not this->impl->target_name.empty(); }

141
void program::compile(const target& t, compile_options options)
Paul's avatar
Paul committed
142
{
143
144
145
    assert(not this->is_compiled());
    this->impl->target_name = t.name();
    this->impl->ctx         = t.get_context();
Paul's avatar
Paul committed
146
    if(enabled(MIGRAPHX_TRACE_COMPILE{}))
147
        options.trace = tracer{std::cout};
Shucai Xiao's avatar
Shucai Xiao committed
148

149
150
    options.trace(*this);
    options.trace();
Shucai Xiao's avatar
Shucai Xiao committed
151

Shucai Xiao's avatar
Shucai Xiao committed
152
    auto&& passes = t.get_passes(this->impl->ctx, options);
153
154
155
    run_passes(*this, passes, options.trace);

    auto mods = this->get_modules();
Shucai Xiao's avatar
Shucai Xiao committed
156

157
158
    // Validate and finalize
    for(const auto& mod : reverse(mods))
Paul's avatar
Paul committed
159
    {
Shucai Xiao's avatar
Shucai Xiao committed
160
161
162
163
164
165
        auto invalid = mod->validate();
        if(invalid != mod->end())
        {
            MIGRAPHX_THROW("Invalid module " + mod->name() + " from compilation at instruction " +
                           std::to_string(std::distance(mod->begin(), invalid)));
        }
166
167
168
169
170
171
172
        auto dangling = mod->find_dangling_reference();
        if(dangling != mod->end())
        {
            auto index = std::distance(mod->begin(), dangling);
            MIGRAPHX_THROW("Dangling reference in module " + mod->name() + " from instruction " +
                           std::to_string(index));
        }
Shucai Xiao's avatar
Shucai Xiao committed
173
        mod->finalize(this->impl->ctx);
Paul's avatar
Paul committed
174
    }
Paul's avatar
Paul committed
175
176
177
178
}

void program::finalize()
{
Shucai Xiao's avatar
Shucai Xiao committed
179
180
    auto* mm = this->get_main_module();
    mm->finalize(this->impl->ctx);
Paul's avatar
Paul committed
181
182
}

183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
template <class T>
std::string classify(T x)
{
    switch(std::fpclassify(x))
    {
    case FP_INFINITE: return "inf";
    case FP_NAN: return "nan";
    case FP_NORMAL: return "normal";
    case FP_SUBNORMAL: return "subnormal";
    case FP_ZERO: return "zero";
    default: return "unknown";
    }
}

std::unordered_set<std::string> classify_argument(const argument& a)
{
    std::unordered_set<std::string> result;
    a.visit(
        [&](auto t) {
            for(const auto& x : t)
                result.insert(classify(x));
        },
        [&](const auto& xs) {
            for(const auto& x : xs)
            {
                auto r = classify_argument(x);
                result.insert(r.begin(), r.end());
            }
        });
    return result;
}

void preview_argument(std::ostream& os, const argument& a)
{
    a.visit(
        [&](auto t) {
            if(t.size() <= 10)
            {
                os << t;
            }
            else
            {
                os << to_string_range(t.begin(), t.begin() + 5);
                os << ", ..., ";
                os << to_string_range(t.end() - 5, t.end());
            }
        },
        [&](const auto& xs) {
            for(const auto& x : xs)
            {
                os << '{';
                preview_argument(os, x);
                os << '}';
            }
        });
}

Paul's avatar
Paul committed
240
template <class F>
Shucai Xiao's avatar
Shucai Xiao committed
241
std::vector<argument> generic_eval(const module* mod,
242
243
                                   context& ctx,
                                   std::unordered_map<std::string, argument> params,
Shucai Xiao's avatar
Shucai Xiao committed
244
                                   std::unordered_map<instruction_ref, argument> results,
245
                                   F make_trace)
Paul's avatar
Paul committed
246
{
Shucai Xiao's avatar
Shucai Xiao committed
247
248
    assert(mod->validate() == mod->end());
    results.reserve(mod->size() * 2);
Paul's avatar
Paul committed
249
250
    std::vector<argument> values;
    values.reserve(16);
251
    auto trace = make_trace(mod);
Shucai Xiao's avatar
Shucai Xiao committed
252
    for(auto ins : iterator_for(*mod))
Paul's avatar
Paul committed
253
    {
254
        assert(results.find(ins) == results.end());
255
256
        const auto& name = ins->name();
        if(name == "@literal")
Paul's avatar
Paul committed
257
        {
Paul's avatar
Paul committed
258
            results.emplace(ins, trace(ins, [&] { return ins->get_literal().get_argument(); }));
Paul's avatar
Paul committed
259
        }
260
        else if(name == "@param")
Paul's avatar
Paul committed
261
        {
Paul's avatar
Paul committed
262
263
264
265
266
            results.emplace(
                ins, trace(ins, [&] {
                    auto param_name = any_cast<builtin::param>(ins->get_operator()).parameter;
                    if(not contains(params, param_name))
                        MIGRAPHX_THROW("Parameter not found: " + param_name);
267
                    auto param = params[param_name];
Paul's avatar
Paul committed
268
269
270
271
272
                    if(param.get_shape() != ins->get_shape())
                        MIGRAPHX_THROW("Incorrect shape {" + to_string(param.get_shape()) +
                                       "} for parameter: " + param_name);
                    return param;
                }));
Paul's avatar
Paul committed
273
        }
274
        else if(name == "@outline")
Paul's avatar
Paul committed
275
        {
Paul's avatar
Paul committed
276
            results.emplace(ins, trace(ins, [&] { return argument{ins->get_shape(), nullptr}; }));
Paul's avatar
Paul committed
277
        }
278
279
280
281
282
283
284
285
286
287
288
289
290
        else if(name == "@return")
        {
            std::vector<argument> prog_outputs;
            std::transform(ins->inputs().begin(),
                           ins->inputs().end(),
                           std::back_inserter(prog_outputs),
                           [&](instruction_ref i) {
                               assert(results.find(i) != results.end());
                               return results[i];
                           });

            return prog_outputs;
        }
Paul's avatar
Paul committed
291
292
        else
        {
Paul's avatar
Paul committed
293
            values.resize(ins->inputs().size());
Paul's avatar
Paul committed
294
295
296
297
298
            std::transform(
                ins->inputs().begin(), ins->inputs().end(), values.begin(), [&](instruction_ref i) {
                    assert(results.find(i) != results.end());
                    return results[i];
                });
Shucai Xiao's avatar
Shucai Xiao committed
299
300
301
302

            const auto& mod_args = ins->module_inputs();
            auto module_eval     = [&](module_ref smod,
                                   const std::unordered_map<std::string, argument>& inputs) {
Shucai Xiao's avatar
Shucai Xiao committed
303
304
                auto ssctx = ctx;
                return generic_eval(smod, ssctx, inputs, results, make_trace);
Shucai Xiao's avatar
Shucai Xiao committed
305
306
            };

Shucai Xiao's avatar
Shucai Xiao committed
307
308
309
310
            results.emplace(ins, trace(ins, [&] {
                                return ins->normalized_operator().compute(
                                    ctx, ins->get_shape(), values, mod_args, module_eval);
                            }));
Paul's avatar
Paul committed
311
        }
312
        assert(results.find(ins) != results.end());
313
        assert(results.at(ins).get_shape() == ins->get_shape());
Paul's avatar
Paul committed
314
    }
Shucai Xiao's avatar
Shucai Xiao committed
315
    return {results.at(std::prev(mod->end()))};
Paul's avatar
Paul committed
316
317
}

Shucai Xiao's avatar
Shucai Xiao committed
318
319
320
321
template <class F>
std::vector<argument> generic_eval(const program& p,
                                   context& ctx,
                                   std::unordered_map<std::string, argument> params,
322
                                   F make_trace)
Shucai Xiao's avatar
Shucai Xiao committed
323
{
Shucai Xiao's avatar
Shucai Xiao committed
324
    const module* mm = p.get_main_module();
325
    return generic_eval(mm, ctx, params, {}, make_trace);
Shucai Xiao's avatar
Shucai Xiao committed
326
327
}

328
std::vector<argument> program::eval(parameter_map params) const
Paul's avatar
Paul committed
329
{
Paul's avatar
Paul committed
330
331
    auto& ctx = this->impl->ctx;
#ifndef NDEBUG
332
333
334
335
336
337
338
339
340
341
342
    auto with_check_context = [&](auto f) {
        return [=, &ctx](auto&&) {
            auto sctx          = std::make_shared<context>(ctx);
            auto check_context = [=, &ctx](auto g) {
                assert(is_shared(ctx, *sctx));
                auto x = g();
                *sctx  = ctx;
                return x;
            };
            return [=](auto&&... xs) { return f(xs..., check_context); };
        };
Paul's avatar
Paul committed
343
344
    };
#else
345
346
347
348
349
    auto with_check_context = [](auto f) {
        return [=](auto&&) {
            return [=](auto&&... xs) { return f(xs..., [](auto g) { return g(); }); };
        };
    };
Paul's avatar
Paul committed
350
#endif
Paul's avatar
Paul committed
351
352
353
354

    auto trace_level = value_of(MIGRAPHX_TRACE_EVAL{});

    if(trace_level > 0)
Paul's avatar
Paul committed
355
    {
Shucai Xiao's avatar
Shucai Xiao committed
356
357
358
        std::unordered_map<instruction_ref, std::string> ins_names;
        // get instruction names
        this->print(ins_names, [](auto, auto) {});
359
360
361
362
363
364
        return generic_eval(*this,
                            ctx,
                            std::move(params),
                            with_check_context([&](auto& ins, auto f, auto&& check_context) {
                                ctx.finish();
                                std::cout << "Run instruction: ";
Shucai Xiao's avatar
Shucai Xiao committed
365
366
                                this->debug_print(ins, ins_names);
                                std::cout << std::endl;
367
368
369
370
371
372
373
                                timer t{};
                                auto result = check_context(f);
                                double t1   = t.record<milliseconds>();
                                ctx.finish();
                                double t2 = t.record<milliseconds>();
                                std::cout << "Time: " << t1 << "ms, " << t2 << "ms" << std::endl;
                                if(trace_level > 1 and ins->name().front() != '@' and
Shucai Xiao's avatar
Shucai Xiao committed
374
375
                                   ins->name() != "load" and not result.empty())
                                {
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
                                    target tgt  = make_target(this->impl->target_name);
                                    auto buffer = tgt.copy_from(result);
                                    if(trace_level == 2)
                                    {
                                        std::cout << "Output has "
                                                  << to_string_range(classify_argument(buffer))
                                                  << std::endl;
                                        std::cout << "Output: ";
                                        preview_argument(std::cout, buffer);
                                        std::cout << std::endl;
                                    }
                                    else
                                    {
                                        std::cout << "Output: " << buffer << std::endl;
                                    }
Shucai Xiao's avatar
Shucai Xiao committed
391
                                }
392
393
                                return result;
                            }));
Paul's avatar
Paul committed
394
395
396
    }
    else
    {
397
398
399
400
401
402
        return generic_eval(*this,
                            ctx,
                            std::move(params),
                            with_check_context([&](auto&, auto f, auto&& check_context) {
                                return check_context(f);
                            }));
Paul's avatar
Paul committed
403
    }
Paul's avatar
Paul committed
404
405
}

406
const int program_file_version = 5;
407
408
409
410
411
412
413
414

value program::to_value() const
{
    value result;
    result["version"] = program_file_version;
    result["target"]  = this->impl->target_name;
    if(not this->impl->target_name.empty())
        result["context"] = this->impl->ctx.to_value();
Shucai Xiao's avatar
Shucai Xiao committed
415

416
    value module_vals = value::object{};
Shucai Xiao's avatar
Shucai Xiao committed
417
    std::unordered_map<instruction_ref, std::string> names;
418
    for(auto& mod : this->get_modules())
Shucai Xiao's avatar
Shucai Xiao committed
419
    {
Shucai Xiao's avatar
Shucai Xiao committed
420
421
        value mod_val;
        value nodes;
422
423
        mod_val["name"] = mod->name();
        names           = mod->print(
Shucai Xiao's avatar
Shucai Xiao committed
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
            [&](auto ins, auto ins_names) {
                value node;
                node["output"]     = ins_names.at(ins);
                node["name"]       = ins->name();
                node["shape"]      = migraphx::to_value(ins->get_shape());
                node["normalized"] = ins->is_normalized();
                if(ins->name() == "@literal")
                    node["literal"] = migraphx::to_value(ins->get_literal());
                node["operator"] = ins->get_operator().to_value();
                std::vector<std::string> inputs;
                std::transform(ins->inputs().begin(),
                               ins->inputs().end(),
                               std::back_inserter(inputs),
                               [&](auto i) {
                                   assert(contains(ins_names, i));
                                   return ins_names.at(i);
                               });
                node["inputs"]   = inputs;
                auto module_args = ins->module_inputs();
                if(not module_args.empty())
                {
                    std::vector<std::string> module_inputs;
                    std::transform(module_args.begin(),
                                   module_args.end(),
                                   std::back_inserter(module_inputs),
                                   [&](auto mod_ref) { return mod_ref->name(); });
                    node["module_inputs"] = module_inputs;
                }

                nodes.push_back(node);
            },
            names);
        mod_val["nodes"] = nodes;

458
        module_vals[mod->name()] = mod_val;
Shucai Xiao's avatar
Shucai Xiao committed
459
    }
Shucai Xiao's avatar
Shucai Xiao committed
460
461
462

    result["modules"] = module_vals;

463
464
    return result;
}
Shucai Xiao's avatar
Shucai Xiao committed
465

Shucai Xiao's avatar
Shucai Xiao committed
466
467
468
469
470
static void mod_from_val(module_ref mod,
                         const value& v,
                         std::unordered_map<std::string, instruction_ref>& instructions,
                         const std::unordered_map<std::string, module_ref>& map_mods)
{
471
    const auto& module_val = v.at(mod->name());
Shucai Xiao's avatar
Shucai Xiao committed
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
    for(const value& node : module_val.at("nodes"))
    {
        instruction_ref output;
        auto name       = node.at("name").to<std::string>();
        auto fields     = node.at("operator");
        auto normalized = node.at("normalized").to<bool>();

        if(name == "@param")
        {
            output = mod->add_parameter(fields["parameter"].to<std::string>(),
                                        migraphx::from_value<shape>(node.at("shape")));
        }
        else if(name == "@literal")
        {
            output = mod->add_literal(migraphx::from_value<literal>(node.at("literal")));
        }
        else
        {
            auto op = make_op(name, fields);
            std::vector<instruction_ref> inputs;
            std::transform(node.at("inputs").begin(),
                           node.at("inputs").end(),
                           std::back_inserter(inputs),
                           [&](const value& i) {
                               auto i_name = i.to<std::string>();
                               assert(contains(instructions, i_name));
                               return instructions.at(i_name);
                           });

            std::vector<module_ref> module_inputs;
            if(node.contains("module_inputs"))
            {
                std::transform(node.at("module_inputs").begin(),
                               node.at("module_inputs").end(),
                               std::back_inserter(module_inputs),
                               [&](const value& i) { return map_mods.at(i.to<std::string>()); });

                for(auto& smod : module_inputs)
                {
                    mod_from_val(smod, v, instructions, map_mods);
                }
            }

            if(name == "@return")
            {
                output = mod->add_return(inputs);
            }
            else if(module_inputs.empty())
            {
                output = mod->add_instruction(op, inputs);
            }
            else
            {
                output = mod->add_instruction(op, inputs, module_inputs);
            }
        }
        output->set_normalized(normalized);
        instructions[node.at("output").to<std::string>()] = output;
    }
}

533
534
535
536
void program::from_value(const value& v)
{
    auto version = v.at("version").to<int>();
    if(version != program_file_version)
Shucai Xiao's avatar
Shucai Xiao committed
537
538
539
540
    {
        MIGRAPHX_THROW("Warning: Program version mismatch");
    }

541
542
543
544
545
546
547
548
    this->impl->target_name = v.at("target").to<std::string>();
    if(not this->impl->target_name.empty())
    {
        target t        = make_target(this->impl->target_name);
        this->impl->ctx = t.get_context();
        this->impl->ctx.from_value(v.at("context"));
    }

Shucai Xiao's avatar
Shucai Xiao committed
549
550
    auto module_vals = v.at("modules");
    for(const auto& vv : module_vals)
551
    {
552
        const auto& name = vv.get_key();
Shucai Xiao's avatar
Shucai Xiao committed
553
554
        if(name == "main")
            continue;
555
        impl->modules.emplace(name, name);
556
    }
557
    std::unordered_map<std::string, module_ref> map_mods;
Paul's avatar
Paul committed
558
559
560
561
    std::transform(impl->modules.begin(),
                   impl->modules.end(),
                   std::inserter(map_mods, map_mods.end()),
                   [&](auto&& pp) { return std::make_pair(pp.first, &pp.second); });
Shucai Xiao's avatar
Shucai Xiao committed
562
563
564
565
566

    std::unordered_map<std::string, instruction_ref> map_insts;
    auto* mm = get_main_module();
    mod_from_val(mm, module_vals, map_insts, map_mods);

567
568
569
    this->finalize();
}

Paul's avatar
Paul committed
570
571
572
double common_average(const std::vector<double>& v)
{
    std::size_t n = v.size() / 4;
Paul's avatar
Paul committed
573
574
    double total  = std::accumulate(v.begin() + n, v.end() - n, 0.0);
    return total / std::distance(v.begin() + n, v.end() - n);
Paul's avatar
Paul committed
575
576
}

Paul Fultz II's avatar
Paul Fultz II committed
577
578
579
580
581
582
583
584
std::string perf_group(const operation& op)
{
    auto attr = op.attributes();
    if(attr.contains("group"))
        return attr.at("group").to<std::string>();
    return op.name();
}

585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
void program::mark(const parameter_map& params, marker&& m)
{
    auto& ctx = this->impl->ctx;
    // Run once by itself
    eval(params);
    ctx.finish();
    // Start marking
    m.mark_start(*this);
    generic_eval(*this, ctx, params, always([&](auto ins, auto f) {
        argument result;
        m.mark_start(ins);
        result = f();
        m.mark_stop(ins);
        return result;
    }));
    m.mark_stop(*this);
}

603
604
605
606
void program::perf_report(std::ostream& os,
                          std::size_t n,
                          parameter_map params,
                          std::size_t batch) const
Paul's avatar
Paul committed
607
{
608
    auto& ctx = this->impl->ctx;
Paul's avatar
Paul committed
609
610
    // Run once by itself
    eval(params);
Paul's avatar
Paul committed
611
    ctx.finish();
Paul's avatar
Paul committed
612
    // Run and time entire program
Paul's avatar
Paul committed
613
614
    std::vector<double> total_vec;
    total_vec.reserve(n);
Paul's avatar
Paul committed
615
    for(std::size_t i = 0; i < n; i++)
Paul's avatar
Paul committed
616
    {
Paul's avatar
Paul committed
617
618
619
620
        total_vec.push_back(time<milliseconds>([&] {
            eval(params);
            ctx.finish();
        }));
Paul's avatar
Paul committed
621
    }
Paul's avatar
Paul committed
622
623
    std::sort(total_vec.begin(), total_vec.end());
    std::unordered_map<instruction_ref, std::vector<double>> ins_vec;
Paul's avatar
Paul committed
624
    // Fill the map
625
    generic_eval(*this, ctx, params, always([&](auto ins, auto) {
Paul's avatar
Paul committed
626
        ins_vec[ins].reserve(n);
627
        return argument{ins->get_shape(), nullptr};
628
    }));
629

Paul's avatar
Paul committed
630
    // Run and time each instruction
Paul's avatar
Paul committed
631
    for(std::size_t i = 0; i < n; i++)
Paul's avatar
Paul committed
632
    {
633
        generic_eval(*this, ctx, params, always([&](auto ins, auto f) {
634
            argument result;
Paul's avatar
Paul committed
635
636
637
638
            ins_vec[ins].push_back(time<milliseconds>([&] {
                result = f();
                ctx.finish();
            }));
639
            return result;
640
        }));
Paul's avatar
Paul committed
641
    }
Paul's avatar
Paul committed
642
643
    for(auto&& p : ins_vec)
        std::sort(p.second.begin(), p.second.end());
Paul's avatar
Paul committed
644
    // Run and time implicit overhead
Paul's avatar
Paul committed
645
646
    std::vector<double> overhead_vec;
    overhead_vec.reserve(n);
Paul's avatar
Paul committed
647
    for(std::size_t i = 0; i < n; i++)
Paul's avatar
Paul committed
648
    {
Paul's avatar
Paul committed
649
        overhead_vec.push_back(time<milliseconds>([&] { dry_run(params); }));
Paul's avatar
Paul committed
650
651
    }

Paul's avatar
Paul committed
652
    double total_time             = common_average(total_vec);
Paul's avatar
Paul committed
653
    double rate                   = 1000.0 / total_time;
Paul's avatar
Paul committed
654
    double overhead_time          = common_average(overhead_vec);
Paul's avatar
Paul committed
655
    double overhead_percent       = overhead_time * 100.0 / total_time;
Paul's avatar
Paul committed
656
    double total_instruction_time = 0.0;
Paul's avatar
Paul committed
657
    std::unordered_map<std::string, double> op_times;
Paul's avatar
Paul committed
658
    for(auto&& p : ins_vec)
Paul's avatar
Paul committed
659
660
    {
        double avg = common_average(p.second);
Paul Fultz II's avatar
Paul Fultz II committed
661
        op_times[perf_group(p.first->get_operator())] += avg;
Paul's avatar
Paul committed
662
663
        total_instruction_time += avg;
    }
Paul's avatar
Paul committed
664
665
    double calculate_overhead_time    = total_time - total_instruction_time;
    double calculate_overhead_percent = calculate_overhead_time * 100.0 / total_time;
Paul's avatar
Paul committed
666

Shucai Xiao's avatar
Shucai Xiao committed
667
668
669
    std::unordered_map<instruction_ref, std::string> names;
    this->print(names, [&](auto ins, auto ins_names) {
        instruction::print(std::cout, ins, ins_names);
670
671
672
673
674

        // skip return instruction
        if(ins->name() == "@return")
            return;

Paul's avatar
Paul committed
675
676
677
        double avg     = common_average(ins_vec[ins]);
        double percent = std::ceil(100.0 * avg / total_instruction_time);
        os << ": " << avg << "ms, " << percent << "%";
678
        os << std::endl;
Paul's avatar
Paul committed
679
    });
Paul's avatar
Paul committed
680
681
682

    os << std::endl;
    os << "Summary:" << std::endl;
683
684
685
686
687
688
689
    std::vector<std::pair<double, std::string>> op_times_sorted;
    std::transform(op_times.begin(),
                   op_times.end(),
                   std::back_inserter(op_times_sorted),
                   [](auto p) { return std::make_pair(p.second, p.first); });
    std::sort(op_times_sorted.begin(), op_times_sorted.end(), std::greater<>{});
    for(auto&& p : op_times_sorted)
Paul's avatar
Paul committed
690
    {
691
692
        auto&& name    = p.second;
        double avg     = p.first;
Paul's avatar
Paul committed
693
694
695
696
697
        double percent = std::ceil(100.0 * avg / total_instruction_time);
        os << name << ": " << avg << "ms, " << percent << "%" << std::endl;
    }

    os << std::endl;
Paul's avatar
Paul committed
698

699
700
    os << "Batch size: " << batch << std::endl;
    os << "Rate: " << rate * batch << "/sec" << std::endl;
Paul's avatar
Paul committed
701
702
    os << "Total time: " << total_time << "ms" << std::endl;
    os << "Total instructions time: " << total_instruction_time << "ms" << std::endl;
Paul's avatar
Paul committed
703
704
705
706
    os << "Overhead time: " << overhead_time << "ms"
       << ", " << calculate_overhead_time << "ms" << std::endl;
    os << "Overhead: " << std::round(overhead_percent) << "%"
       << ", " << std::round(calculate_overhead_percent) << "%" << std::endl;
Paul's avatar
Paul committed
707
708
}

Paul's avatar
Paul committed
709
void program::debug_print() const { std::cout << *this << std::endl; }
Shucai Xiao's avatar
Shucai Xiao committed
710
711
void program::debug_print(instruction_ref ins, 
                          const std::unordered_map<instruction_ref, std::string>& ins_names) const
Paul's avatar
Paul committed
712
{
713
    if(std::any_of(this->impl->modules.begin(), this->impl->modules.end(), [&](const auto& pp) {
714
           return is_end(pp.second.end(), ins);
Shucai Xiao's avatar
Shucai Xiao committed
715
       }))
Paul's avatar
Paul committed
716
717
718
719
    {
        std::cout << "End instruction" << std::endl;
        return;
    }
720
721
    else if(std::none_of(this->impl->modules.begin(),
                         this->impl->modules.end(),
722
                         [&](const auto& pp) { return pp.second.has_instruction(ins); }))
Paul's avatar
Paul committed
723
724
725
726
    {
        std::cout << "Instruction not part of program" << std::endl;
        return;
    }
Shucai Xiao's avatar
Shucai Xiao committed
727

Shucai Xiao's avatar
Shucai Xiao committed
728
729
730
731
    if (contains(ins_names, ins))
    {
        instruction::print(std::cout, ins, ins_names);
    }
Paul's avatar
Paul committed
732
733
}

Shucai Xiao's avatar
Shucai Xiao committed
734
735
736
737
void program::print(
    std::unordered_map<instruction_ref, std::string>& names,
    const std::function<void(instruction_ref, std::unordered_map<instruction_ref, std::string>)>&
        print_func) const
738
{
739
    for(const auto& pp : this->impl->modules)
740
    {
741
        names = pp.second.print(print_func, names);
742
743
744
    }
}

Shucai Xiao's avatar
Shucai Xiao committed
745
void program::print_graph(std::ostream& os, bool brief) const
746
{
Shucai Xiao's avatar
Shucai Xiao committed
747
748
    const auto* mm = this->get_main_module();
    mm->print_graph(os, brief);
749
750
751
752
}

void program::print_cpp(std::ostream& os) const
{
Shucai Xiao's avatar
Shucai Xiao committed
753
754
755
756
757
758
759
760
    auto vec_modules = this->get_modules();
    std::unordered_map<instruction_ref, std::string> names;
    for(auto& mod : vec_modules)
    {
        os << "module: \"" << mod->name() << "\"" << std::endl;
        names = mod->print_cpp(os, names);
        os << std::endl;
    }
761
762
}

Paul's avatar
Paul committed
763
764
void program::dry_run(std::unordered_map<std::string, argument> params) const
{
Paul's avatar
Paul committed
765
    auto& ctx = this->impl->ctx;
766
767
768
    generic_eval(*this, ctx, std::move(params), always([](auto ins, auto&&...) {
        return argument{ins->get_shape(), nullptr};
    }));
Paul's avatar
Paul committed
769
770
}

Shucai Xiao's avatar
Shucai Xiao committed
771
void program::annotate(std::ostream& os, const std::function<void(instruction_ref)>& a) const
Paul's avatar
Paul committed
772
{
773
    for(auto& pp : this->impl->modules)
Shucai Xiao's avatar
Shucai Xiao committed
774
    {
775
776
        std::cout << pp.first << ":" << std::endl;
        pp.second.annotate(os, a);
Shucai Xiao's avatar
Shucai Xiao committed
777
    }
Paul's avatar
Paul committed
778
779
}

Paul's avatar
Paul committed
780
const module* program::get_module(const std::string& name) const { return &impl->modules.at(name); }
Shucai Xiao's avatar
Shucai Xiao committed
781
782
783

module* program::create_module(const std::string& name)
{
784
    assert(not contains(impl->modules, name));
785
786
    auto r = impl->modules.emplace(name, name);
    return &(r.first->second);
Shucai Xiao's avatar
Shucai Xiao committed
787
788
}

Paul's avatar
Paul committed
789
module* program::get_module(const std::string& name) { return &impl->modules.at(name); }
Shucai Xiao's avatar
Shucai Xiao committed
790
791
792
793
794

module* program::get_main_module() { return get_module("main"); }

const module* program::get_main_module() const { return get_module("main"); }

Paul's avatar
Paul committed
795
template <class T>
796
std::vector<T*> generic_get_modules(T* mm)
Shucai Xiao's avatar
Shucai Xiao committed
797
{
798
    std::vector<T*> vec_modules;
Shucai Xiao's avatar
Shucai Xiao committed
799
800
801
802
803
    vec_modules.push_back(mm);
    auto sub_modules = mm->get_sub_modules();
    vec_modules.insert(vec_modules.end(), sub_modules.begin(), sub_modules.end());
    return vec_modules;
}
Shucai Xiao's avatar
Shucai Xiao committed
804

Paul's avatar
Paul committed
805
template <class Map, class T, class OutputIterator>
806
void generic_get_unused_modules(Map& m, const std::vector<T*>& mods, OutputIterator out)
Shucai Xiao's avatar
Shucai Xiao committed
807
{
808
809
810
811
    std::unordered_set<std::string> used;
    std::transform(mods.begin(), mods.end(), std::inserter(used, used.end()), [](auto&& mod) {
        return mod->name();
    });
Paul's avatar
Paul committed
812
813
814
815
816
    transform_if(m.begin(),
                 m.end(),
                 out,
                 [&](auto&& pp) { return not contains(used, pp.first); },
                 [](auto&& pp) { return &pp.second; });
817
}
Shucai Xiao's avatar
Shucai Xiao committed
818

819
820
821
822
823
824
825
826
827
828
829
830
std::vector<const module*> program::get_modules() const
{
    auto result = generic_get_modules(this->get_main_module());
    generic_get_unused_modules(impl->modules, result, std::back_inserter(result));
    return result;
}

std::vector<module*> program::get_modules()
{
    auto result = generic_get_modules(this->get_main_module());
    generic_get_unused_modules(impl->modules, result, std::back_inserter(result));
    return result;
Shucai Xiao's avatar
Shucai Xiao committed
831
832
}

833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
template <class Map, class T>
bool is_unused_module(Map& m, const std::vector<T*>& mods, const std::string& name)
{
    bool is_unused = false;
    generic_get_unused_modules(m, mods, make_function_output_iterator([&](auto* mod) {
                                   if(mod->name() == name)
                                       is_unused = true;
                               }));
    return is_unused;
}

template <class Map>
bool references_instruction(Map& m, const instruction& ins, const std::string& name)
{
    return std::any_of(m.begin(), m.end(), [&](auto&& p) {
        if(p.first == name)
            return false;
        return std::any_of(p.second.begin(), p.second.end(), [&](auto&& i) {
            return std::any_of(i.inputs().begin(), i.inputs().end(), [&](auto&& j) {
                return std::addressof(*j) == std::addressof(ins);
            });
        });
    });
}

void program::remove_module(const std::string& name)
{
    // cppcheck-suppress assertWithSideEffect
    assert(is_unused_module(impl->modules, generic_get_modules(this->get_main_module()), name) &&
           "Module used in program");
    assert(std::none_of(
               impl->modules.at(name).begin(),
               impl->modules.at(name).end(),
               [&](auto&& ins) { return references_instruction(impl->modules, ins, name); }) &&
           "Instruction referenced in another module");
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883

    // if an instruction has an input out side of the current module, need to remove
    // the instruction from its input's outputs
    auto& mod = impl->modules.at(name);
    for(auto ins : iterator_for(mod))
    {
        auto inputs = ins->inputs();
        for(auto in : inputs)
        {
            if(not mod.has_instruction(in))
            {
                in->remove_output(ins);
            }
        }
    }

884
885
886
887
888
889
890
891
892
893
894
895
    impl->modules.erase(name);
}

void program::remove_unused_modules()
{
    std::vector<module*> unused;
    generic_get_unused_modules(
        impl->modules, generic_get_modules(this->get_main_module()), std::back_inserter(unused));
    for(auto* m : unused)
        this->remove_module(m->name());
}

896
897
program& program::sort()
{
898
    for(auto& pp : this->impl->modules)
Shucai Xiao's avatar
Shucai Xiao committed
899
    {
900
        pp.second.sort();
Shucai Xiao's avatar
Shucai Xiao committed
901
902
    }

903
904
905
    return *this;
}

Paul's avatar
Paul committed
906
bool operator==(const program& x, const program& y) { return to_string(x) == to_string(y); }
Paul's avatar
Paul committed
907

Paul's avatar
Paul committed
908
std::ostream& operator<<(std::ostream& os, const program& p)
Paul's avatar
Paul committed
909
{
Shucai Xiao's avatar
Shucai Xiao committed
910
911
912
    auto vec_modules = p.get_modules();
    std::unordered_map<instruction_ref, std::string> names;
    for(auto& mod : vec_modules)
Shucai Xiao's avatar
Shucai Xiao committed
913
    {
Shucai Xiao's avatar
Shucai Xiao committed
914
915
916
917
918
919
920
        os << "module: \"" << mod->name() << "\"" << std::endl;
        names = mod->print(
            [&](auto ins, auto ins_names) {
                instruction::print(os, ins, ins_names);
                os << std::endl;
            },
            names);
921
        os << std::endl;
Shucai Xiao's avatar
Shucai Xiao committed
922
923
    }

Paul's avatar
Paul committed
924
    return os;
Paul's avatar
Paul committed
925
}
Paul's avatar
Paul committed
926

Paul's avatar
Paul committed
927
} // namespace MIGRAPHX_INLINE_NS
Paul's avatar
Paul committed
928
} // namespace migraphx