1 | //===--- CtxInstrContextNode.h - Contextual Profile Node --------*- C++ -*-===// |
2 | // |
3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | // See https://llvm.org/LICENSE.txt for license information. |
5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | // |
7 | //===----------------------------------------------------------------------===// |
8 | //============================================================================== |
9 | // |
10 | // NOTE! |
11 | // llvm/include/llvm/ProfileData/CtxInstrContextNode.h and |
12 | // compiler-rt/lib/ctx_profile/CtxInstrContextNode.h |
13 | // must be exact copies of each other. |
14 | // |
15 | // compiler-rt creates these objects as part of the instrumentation runtime for |
16 | // contextual profiling. LLVM only consumes them to convert a contextual tree |
17 | // to a bitstream. |
18 | // |
19 | //============================================================================== |
20 | |
21 | /// The contextual profile is a directed tree where each node has one parent. A |
22 | /// node (ContextNode) corresponds to a function activation. The root of the |
23 | /// tree is at a function that was marked as entrypoint to the compiler. A node |
24 | /// stores counter values for edges and a vector of subcontexts. These are the |
25 | /// contexts of callees. The index in the subcontext vector corresponds to the |
26 | /// index of the callsite (as was instrumented via llvm.instrprof.callsite). At |
27 | /// that index we find a linked list, potentially empty, of ContextNodes. Direct |
28 | /// calls will have 0 or 1 values in the linked list, but indirect callsites may |
29 | /// have more. |
30 | /// |
31 | /// The ContextNode has a fixed sized header describing it - the GUID of the |
32 | /// function, the size of the counter and callsite vectors. It is also an |
33 | /// (intrusive) linked list for the purposes of the indirect call case above. |
34 | /// |
35 | /// Allocation is expected to happen on an Arena. The allocation lays out inline |
36 | /// the counter and subcontexts vectors. The class offers APIs to correctly |
37 | /// reference the latter. |
38 | /// |
39 | /// The layout is as follows: |
40 | /// |
41 | /// [[declared fields][counters vector][vector of ptrs to subcontexts]] |
42 | /// |
43 | /// See also documentation on the counters and subContexts members below. |
44 | /// |
45 | /// The structure of the ContextNode is known to LLVM, because LLVM needs to: |
46 | /// (1) increment counts, and |
47 | /// (2) form a GEP for the position in the subcontext list of a callsite |
48 | /// This means changes to LLVM contextual profile lowering and changes here |
49 | /// must be coupled. |
50 | /// Note: the header content isn't interesting to LLVM (other than its size) |
51 | /// |
52 | /// Part of contextual collection is the notion of "scratch contexts". These are |
53 | /// buffers that are "large enough" to allow for memory-safe acceses during |
54 | /// counter increments - meaning the counter increment code in LLVM doesn't need |
55 | /// to be concerned with memory safety. Their subcontexts never get populated, |
56 | /// though. The runtime code here produces and recognizes them. |
57 | |
58 | #ifndef LLVM_PROFILEDATA_CTXINSTRCONTEXTNODE_H |
59 | #define LLVM_PROFILEDATA_CTXINSTRCONTEXTNODE_H |
60 | |
61 | #include <stdint.h> |
62 | #include <stdlib.h> |
63 | |
64 | namespace llvm { |
65 | namespace ctx_profile { |
66 | using GUID = uint64_t; |
67 | |
68 | class ContextNode final { |
69 | const GUID Guid; |
70 | ContextNode *const Next; |
71 | const uint32_t NumCounters; |
72 | const uint32_t NumCallsites; |
73 | |
74 | public: |
75 | ContextNode(GUID Guid, uint32_t NumCounters, uint32_t NumCallsites, |
76 | ContextNode *Next = nullptr) |
77 | : Guid(Guid), Next(Next), NumCounters(NumCounters), |
78 | NumCallsites(NumCallsites) {} |
79 | |
80 | static inline size_t getAllocSize(uint32_t NumCounters, |
81 | uint32_t NumCallsites) { |
82 | return sizeof(ContextNode) + sizeof(uint64_t) * NumCounters + |
83 | sizeof(ContextNode *) * NumCallsites; |
84 | } |
85 | |
86 | // The counters vector starts right after the static header. |
87 | uint64_t *counters() { |
88 | ContextNode *addr_after = &(this[1]); |
89 | return reinterpret_cast<uint64_t *>(addr_after); |
90 | } |
91 | |
92 | uint32_t counters_size() const { return NumCounters; } |
93 | uint32_t callsites_size() const { return NumCallsites; } |
94 | |
95 | const uint64_t *counters() const { |
96 | return const_cast<ContextNode *>(this)->counters(); |
97 | } |
98 | |
99 | // The subcontexts vector starts right after the end of the counters vector. |
100 | ContextNode **subContexts() { |
101 | return reinterpret_cast<ContextNode **>(&(counters()[NumCounters])); |
102 | } |
103 | |
104 | ContextNode *const *subContexts() const { |
105 | return const_cast<ContextNode *>(this)->subContexts(); |
106 | } |
107 | |
108 | GUID guid() const { return Guid; } |
109 | ContextNode *next() const { return Next; } |
110 | |
111 | size_t size() const { return getAllocSize(NumCounters, NumCallsites); } |
112 | |
113 | uint64_t entrycount() const { return counters()[0]; } |
114 | }; |
115 | |
116 | /// The internal structure of FunctionData. This makes sure that changes to |
117 | /// the fields of FunctionData either get automatically captured on the llvm |
118 | /// side, or force a manual corresponding update. |
119 | /// |
120 | /// The macro arguments (see CtxInstrProfiling.h for example): |
121 | /// |
122 | /// PTRDECL is a macro taking 2 parameters: a type and the name of the field. |
123 | /// The field is a pointer of that type; |
124 | /// |
125 | /// VOLATILE_PTRDECL is the same as above, but for volatile pointers; |
126 | /// |
127 | /// MUTEXDECL takes one parameter, the name of a field that is a mutex. |
128 | #define CTXPROF_FUNCTION_DATA(PTRDECL, CONTEXT_PTR, VOLATILE_PTRDECL, \ |
129 | MUTEXDECL) \ |
130 | PTRDECL(FunctionData, Next) \ |
131 | VOLATILE_PTRDECL(void, EntryAddress) \ |
132 | CONTEXT_PTR \ |
133 | VOLATILE_PTRDECL(ContextNode, FlatCtx) \ |
134 | MUTEXDECL(Mutex) |
135 | |
136 | /// Abstraction for the parameter passed to `__llvm_ctx_profile_fetch`. |
137 | /// `startContextSection` is called before any context roots are sent for |
138 | /// writing. Then one or more `writeContextual` calls are made; finally, |
139 | /// `endContextSection` is called. |
140 | class ProfileWriter { |
141 | public: |
142 | virtual void startContextSection() = 0; |
143 | virtual void writeContextual(const ctx_profile::ContextNode &RootNode, |
144 | const ctx_profile::ContextNode *Unhandled, |
145 | uint64_t TotalRootEntryCount) = 0; |
146 | virtual void endContextSection() = 0; |
147 | |
148 | virtual void startFlatSection() = 0; |
149 | virtual void writeFlat(ctx_profile::GUID Guid, const uint64_t *Buffer, |
150 | size_t BufferSize) = 0; |
151 | virtual void endFlatSection() = 0; |
152 | |
153 | virtual ~ProfileWriter() = default; |
154 | }; |
155 | } // namespace ctx_profile |
156 | } // namespace llvm |
157 | #endif |
158 | |