xref: /llvm-project/llvm/include/llvm/Transforms/Vectorize/LoopVectorize.h (revision 13107cb09441dfeab24fcbcae9f4d3ba4cfc2703)
1 //===- LoopVectorize.h ------------------------------------------*- 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 // This is the LLVM loop vectorizer. This pass modifies 'vectorizable' loops
10 // and generates target-independent LLVM-IR.
11 // The vectorizer uses the TargetTransformInfo analysis to estimate the costs
12 // of instructions in order to estimate the profitability of vectorization.
13 //
14 // The loop vectorizer combines consecutive loop iterations into a single
15 // 'wide' iteration. After this transformation the index is incremented
16 // by the SIMD vector width, and not by one.
17 //
18 // This pass has four parts:
19 // 1. The main loop pass that drives the different parts.
20 // 2. LoopVectorizationLegality - A unit that checks for the legality
21 //    of the vectorization.
22 // 3. InnerLoopVectorizer - A unit that performs the actual
23 //    widening of instructions.
24 // 4. LoopVectorizationCostModel - A unit that checks for the profitability
25 //    of vectorization. It decides on the optimal vector width, which
26 //    can be one, if vectorization is not profitable.
27 //
28 // There is a development effort going on to migrate loop vectorizer to the
29 // VPlan infrastructure and to introduce outer loop vectorization support (see
30 // docs/VectorizationPlan.rst and
31 // http://lists.llvm.org/pipermail/llvm-dev/2017-December/119523.html). For this
32 // purpose, we temporarily introduced the VPlan-native vectorization path: an
33 // alternative vectorization path that is natively implemented on top of the
34 // VPlan infrastructure. See EnableVPlanNativePath for enabling.
35 //
36 //===----------------------------------------------------------------------===//
37 //
38 // The reduction-variable vectorization is based on the paper:
39 //  D. Nuzman and R. Henderson. Multi-platform Auto-vectorization.
40 //
41 // Variable uniformity checks are inspired by:
42 //  Karrenberg, R. and Hack, S. Whole Function Vectorization.
43 //
44 // The interleaved access vectorization is based on the paper:
45 //  Dorit Nuzman, Ira Rosen and Ayal Zaks.  Auto-Vectorization of Interleaved
46 //  Data for SIMD
47 //
48 // Other ideas/concepts are from:
49 //  A. Zaks and D. Nuzman. Autovectorization in GCC-two years later.
50 //
51 //  S. Maleki, Y. Gao, M. Garzaran, T. Wong and D. Padua.  An Evaluation of
52 //  Vectorizing Compilers.
53 //
54 //===----------------------------------------------------------------------===//
55 
56 #ifndef LLVM_TRANSFORMS_VECTORIZE_LOOPVECTORIZE_H
57 #define LLVM_TRANSFORMS_VECTORIZE_LOOPVECTORIZE_H
58 
59 #include "llvm/IR/PassManager.h"
60 #include "llvm/Support/CommandLine.h"
61 #include "llvm/Transforms/Utils/ExtraPassManager.h"
62 #include <functional>
63 
64 namespace llvm {
65 
66 class AssumptionCache;
67 class BlockFrequencyInfo;
68 class DemandedBits;
69 class DominatorTree;
70 class Function;
71 class Instruction;
72 class Loop;
73 class LoopAccessInfoManager;
74 class LoopInfo;
75 class OptimizationRemarkEmitter;
76 class ProfileSummaryInfo;
77 class ScalarEvolution;
78 class TargetLibraryInfo;
79 class TargetTransformInfo;
80 
81 extern cl::opt<bool> EnableLoopInterleaving;
82 extern cl::opt<bool> EnableLoopVectorization;
83 
84 struct LoopVectorizeOptions {
85   /// If false, consider all loops for interleaving.
86   /// If true, only loops that explicitly request interleaving are considered.
87   bool InterleaveOnlyWhenForced;
88 
89   /// If false, consider all loops for vectorization.
90   /// If true, only loops that explicitly request vectorization are considered.
91   bool VectorizeOnlyWhenForced;
92 
93   /// The current defaults when creating the pass with no arguments are:
94   /// EnableLoopInterleaving = true and EnableLoopVectorization = true. This
95   /// means that interleaving default is consistent with the cl::opt flag, while
96   /// vectorization is not.
97   /// FIXME: The default for EnableLoopVectorization in the cl::opt should be
98   /// set to true, and the corresponding change to account for this be made in
99   /// opt.cpp. The initializations below will become:
100   /// InterleaveOnlyWhenForced(!EnableLoopInterleaving)
101   /// VectorizeOnlyWhenForced(!EnableLoopVectorization).
102   LoopVectorizeOptions()
103       : InterleaveOnlyWhenForced(false), VectorizeOnlyWhenForced(false) {}
104   LoopVectorizeOptions(bool InterleaveOnlyWhenForced,
105                        bool VectorizeOnlyWhenForced)
106       : InterleaveOnlyWhenForced(InterleaveOnlyWhenForced),
107         VectorizeOnlyWhenForced(VectorizeOnlyWhenForced) {}
108 
109   LoopVectorizeOptions &setInterleaveOnlyWhenForced(bool Value) {
110     InterleaveOnlyWhenForced = Value;
111     return *this;
112   }
113 
114   LoopVectorizeOptions &setVectorizeOnlyWhenForced(bool Value) {
115     VectorizeOnlyWhenForced = Value;
116     return *this;
117   }
118 };
119 
120 /// Storage for information about made changes.
121 struct LoopVectorizeResult {
122   bool MadeAnyChange;
123   bool MadeCFGChange;
124 
125   LoopVectorizeResult(bool MadeAnyChange, bool MadeCFGChange)
126       : MadeAnyChange(MadeAnyChange), MadeCFGChange(MadeCFGChange) {}
127 };
128 
129 /// The LoopVectorize Pass.
130 struct LoopVectorizePass : public PassInfoMixin<LoopVectorizePass> {
131 private:
132   /// If false, consider all loops for interleaving.
133   /// If true, only loops that explicitly request interleaving are considered.
134   bool InterleaveOnlyWhenForced;
135 
136   /// If false, consider all loops for vectorization.
137   /// If true, only loops that explicitly request vectorization are considered.
138   bool VectorizeOnlyWhenForced;
139 
140 public:
141   LoopVectorizePass(LoopVectorizeOptions Opts = {});
142 
143   ScalarEvolution *SE;
144   LoopInfo *LI;
145   TargetTransformInfo *TTI;
146   DominatorTree *DT;
147   BlockFrequencyInfo *BFI;
148   TargetLibraryInfo *TLI;
149   DemandedBits *DB;
150   AssumptionCache *AC;
151   LoopAccessInfoManager *LAIs;
152   OptimizationRemarkEmitter *ORE;
153   ProfileSummaryInfo *PSI;
154 
155   PreservedAnalyses run(Function &F, FunctionAnalysisManager &AM);
156   void printPipeline(raw_ostream &OS,
157                      function_ref<StringRef(StringRef)> MapClassName2PassName);
158 
159   // Shim for old PM.
160   LoopVectorizeResult runImpl(Function &F);
161 
162   bool processLoop(Loop *L);
163 };
164 
165 /// Reports a vectorization failure: print \p DebugMsg for debugging
166 /// purposes along with the corresponding optimization remark \p RemarkName.
167 /// If \p I is passed, it is an instruction that prevents vectorization.
168 /// Otherwise, the loop \p TheLoop is used for the location of the remark.
169 void reportVectorizationFailure(const StringRef DebugMsg,
170     const StringRef OREMsg, const StringRef ORETag,
171     OptimizationRemarkEmitter *ORE, Loop *TheLoop, Instruction *I = nullptr);
172 
173 /// Same as above, but the debug message and optimization remark are identical
174 inline void reportVectorizationFailure(const StringRef DebugMsg,
175                                        const StringRef ORETag,
176                                        OptimizationRemarkEmitter *ORE,
177                                        Loop *TheLoop,
178                                        Instruction *I = nullptr) {
179   reportVectorizationFailure(DebugMsg, DebugMsg, ORETag, ORE, TheLoop, I);
180 }
181 
182 /// A marker analysis to determine if extra passes should be run after loop
183 /// vectorization.
184 struct ShouldRunExtraVectorPasses
185     : public ShouldRunExtraPasses<ShouldRunExtraVectorPasses>,
186       public AnalysisInfoMixin<ShouldRunExtraVectorPasses> {
187   static AnalysisKey Key;
188 };
189 } // end namespace llvm
190 
191 #endif // LLVM_TRANSFORMS_VECTORIZE_LOOPVECTORIZE_H
192