aboutsummaryrefslogtreecommitdiff
path: root/deps/cudnn-frontend/include/cudnn_frontend_find_plan.h
blob: b90790385c571e1060f7f2c59252741edb7290b3 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
/*
 * Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice shall be included in
 * all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 * DEALINGS IN THE SOFTWARE.
 */

#pragma once

#include <iomanip>
#include <set>
#include "cudnn_frontend_EngineConfigGenerator.h"

namespace cudnn_frontend {

/// Sorts the execution plans by their run time.
/// The run time of plan may not trivial and hence we
/// run it multiple times till we get a stable value.
/// We have an additional dry-run which helps stabilize the
/// time further.
template <CudnnFindSamplingTechnique samplingTechnique>
auto
time_sorted_plan(cudnnHandle_t handle,
                 executionPlans_t plans,
                 VariantPack const &variantPack,
                 uint64_t max_plans_to_evaluate = 1000) -> executionPlans_t {
    executionPlans_t time_sorted_plans;

    auto plan_cmp = [](const ExecutionPlan &a, const ExecutionPlan &b) {
        return a.getExecutionTime() < b.getExecutionTime();
    };
    std::set<std::reference_wrapper<ExecutionPlan>, decltype(plan_cmp)> timed_execution_plans(plan_cmp);

    const int maxIterCount         = (samplingTechnique == CudnnFindSamplingTechnique::CUDNN_FIND_SAMPLE_ONCE) ? 1
                                     : (samplingTechnique == CudnnFindSamplingTechnique::CUDNN_FIND_SAMPLE_MEDIAN_OF_THREE)
                                         ? 3
                                         : 100;
    const float threshhold         = 0.95f;
    uint64_t successful_plan_count = 0;
    cudaEvent_t start, stop;
    detail::cuda_event_create(&start);
    detail::cuda_event_create(&stop);
    detail::cuda_device_synchronize();

    cudaStream_t stream = nullptr;
    detail::get_stream(handle, &stream);

    for (auto &plan : plans) {
        float time_ms       = 0.0f;
        float final_time_ms = 0.0f;
        float min_time_ms   = std::numeric_limits<float>::max();

        // Warm-up run
        auto warmup_status = detail::execute(handle, plan.get_raw_desc(), variantPack.get_raw_desc());
        if (warmup_status != CUDNN_STATUS_SUCCESS) {
            CUDNN_FE_LOG_LABEL_ENDL("Plan " << plan.getTag() << " failed with " << to_string(warmup_status));
            continue;
        }
        successful_plan_count++;
        detail::cuda_device_synchronize();

        float time_run_ms[3] = {0.0f, 0.0f, 0.0f};
        for (int i = 0; i < maxIterCount; i++) {
            detail::cuda_event_record(start, stream);

            detail::execute(handle, plan.get_raw_desc(), variantPack.get_raw_desc());

            detail::cuda_event_record(stop, stream);
            detail::cuda_event_synchronize(stop);
            detail::cuda_event_elapsed_time(&time_ms, start, stop);

            if constexpr (samplingTechnique == CudnnFindSamplingTechnique::CUDNN_FIND_SAMPLE_TILL_STABLE) {
                final_time_ms = std::min(min_time_ms, time_ms);
                if (time_ms / min_time_ms < threshhold) {
                    min_time_ms = final_time_ms;
                } else {
                    break;
                }
            } else if constexpr (samplingTechnique == CudnnFindSamplingTechnique::CUDNN_FIND_SAMPLE_MEDIAN_OF_THREE) {
                time_run_ms[i] = time_ms;
                if (i == maxIterCount - 1) {
                    auto min_of_first_two = std::min(time_run_ms[0], time_run_ms[1]);
                    auto max_of_first_two = std::max(time_run_ms[0], time_run_ms[1]);
                    final_time_ms         = std::max(min_of_first_two, std::min(max_of_first_two, time_run_ms[2]));
                }
            } else {
                final_time_ms = time_ms;
            }
        }
        CUDNN_FE_LOG_LABEL_ENDL("Plan " << plan.getTag() << " took " << std::setw(10) << final_time_ms);
        plan.setExecutionTime(final_time_ms);
        timed_execution_plans.insert(plan);
        if (successful_plan_count >= max_plans_to_evaluate) {
            CUDNN_FE_LOG_LABEL_ENDL("Successfully profiled " << max_plans_to_evaluate << "plans.");
            break;
        }
    }

    for (ExecutionPlan &plan : timed_execution_plans) {
        time_sorted_plans.emplace_back(std::move(plan));
    }

    detail::cuda_event_destroy(start);
    detail::cuda_event_destroy(stop);

    CUDNN_FE_LOG_LABEL_ENDL("Auto-tuning returns " << time_sorted_plans.size() << " plans.");

    return time_sorted_plans;
}

template <CudnnFindSamplingTechnique samplingTechnique>
auto
EngineConfigGenerator::cudnnFindPlan(cudnnHandle_t handle,
                                     cudnn_frontend::OperationGraph &opGraph,
                                     cudnn_frontend::VariantPack const &variantPack) -> executionPlans_t {
    /// Creating a set of execution plans that are supported.
    executionPlans_t plans = cudnnGetPlan(handle, opGraph);
    return time_sorted_plan<samplingTechnique>(handle, std::move(plans), variantPack);
}

template <CudnnFindSamplingTechnique samplingTechnique>
auto
EngineConfigGenerator::cudnnFindPlan(cudnnHandle_t handle,
                                     cudnn_frontend::OperationGraph &opGraph,
                                     cudnn_frontend::VariantPack const &variantPack,
                                     Predicate pred) -> executionPlans_t {
    /// Creating a set of execution plans that are supported.
    executionPlans_t plans = cudnnGetPlan(handle, opGraph, pred);
    return time_sorted_plan<samplingTechnique>(handle, std::move(plans), variantPack);
}

template <CudnnFindSamplingTechnique samplingTechnique>
auto
EngineConfigGenerator::cudnnFindPlanAndCache(cudnnHandle_t handle,
                                             cudnn_frontend::OperationGraph &opGraph,
                                             cudnn_frontend::VariantPack const &variantPack,
                                             cudnn_frontend::ExecutionPlanCache &cache,
                                             Predicate pred) -> cudnn_frontend::ExecutionPlan {
    /// Creating a set of execution plans that are supported.
    auto sorted_plans = cudnnFindPlan<samplingTechnique>(handle, opGraph, variantPack, pred);
    /// Check if the fastest plan is stable enough to be added to the plan cache
    if (cache.is_fastest_plan_stable(opGraph, sorted_plans.front().getTag())) {
        cache.add_plan_to_cache(opGraph, sorted_plans.front());
    }
    return sorted_plans.front();
}

}  // namespace cudnn_frontend