1 | /* Copyright 2019 Google LLC. All Rights Reserved. |
2 | |
3 | Licensed under the Apache License, Version 2.0 (the "License"); |
4 | you may not use this file except in compliance with the License. |
5 | You may obtain a copy of the License at |
6 | |
7 | http://www.apache.org/licenses/LICENSE-2.0 |
8 | |
9 | Unless required by applicable law or agreed to in writing, software |
10 | distributed under the License is distributed on an "AS IS" BASIS, |
11 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
12 | See the License for the specific language governing permissions and |
13 | limitations under the License. |
14 | ==============================================================================*/ |
15 | |
16 | // Internal implementation details for Ctx. Drags in the entire world. Avoid |
17 | // #including this, use "ctx.h" instead. |
18 | |
19 | #ifndef RUY_RUY_CTX_IMPL_H_ |
20 | #define RUY_RUY_CTX_IMPL_H_ |
21 | |
22 | #include <cstddef> |
23 | #include <memory> |
24 | #include <vector> |
25 | |
26 | #include "ruy/allocator.h" |
27 | #include "ruy/cpuinfo.h" |
28 | #include "ruy/ctx.h" |
29 | #include "ruy/path.h" |
30 | #include "ruy/performance_advisory.h" |
31 | #include "ruy/prepacked_cache.h" |
32 | #include "ruy/thread_pool.h" |
33 | #include "ruy/tune.h" |
34 | |
35 | namespace ruy { |
36 | |
37 | // The resources private to each Ruy thread. |
38 | struct ThreadSpecificResource final { |
39 | // Each thread may be running on a different microarchitecture. For example, |
40 | // some threads may be on big cores, while others are on little cores. Thus, |
41 | // it's best for the tuning to be per-thread. |
42 | TuningResolver tuning_resolver; |
43 | // Each thread has its own local allocator. |
44 | Allocator allocator; |
45 | }; |
46 | |
47 | // CtxImpl is what actually holds all the data members in a context. |
48 | // It is a subclass of Ctx, which provides the interface that is what most |
49 | // of ruy's code needs. |
50 | // |
51 | // A key requirement is that since many ruy files, including public headers, |
52 | // need a definition of Ctx, the "ctx.h" header defining it must minimize how |
53 | // many other ruy internal headers it includes. That is achieved by putting data |
54 | // members in the CtxImpl subclass, and ensuring that only a few .cc files, not |
55 | // header files, need a definition of CtxImpl. |
56 | class CtxImpl final : public Ctx { |
57 | private: |
58 | friend class Ctx; |
59 | |
60 | // Single Path bit indicating which Path was used last. |
61 | Path last_used_path_ = Path::kNone; |
62 | PerformanceAdvisory performance_advisory_ = PerformanceAdvisory::kNone; |
63 | Tuning explicit_tuning_ = Tuning::kAuto; |
64 | ThreadPool thread_pool_; |
65 | int max_num_threads_ = 1; |
66 | // Allocator for main thread work before invoking the threadpool. |
67 | // Our simple Allocator does not allow reserving/allocating more blocks |
68 | // while it's already in committed state, so the main thread needs both |
69 | // this allocator, and its per-thread allocator. |
70 | std::unique_ptr<Allocator> main_allocator_; |
71 | std::unique_ptr<PrepackedCache> prepacked_cache_; |
72 | // Set of Paths enabled at runtime. By default, that is based on runtime |
73 | // detection, but may be overridden. The initial value kNone |
74 | // means that detection has not yet been performed. |
75 | Path runtime_enabled_paths_ = Path::kNone; |
76 | CpuInfo cpuinfo_; |
77 | // State for each thread in the thread pool. Entry 0 is the main thread. |
78 | std::vector<std::unique_ptr<ThreadSpecificResource>> |
79 | thread_specific_resources_; |
80 | }; |
81 | |
82 | } // namespace ruy |
83 | |
84 | #endif // RUY_RUY_CTX_IMPL_H_ |
85 | |