annotate lib/CodeGen/InterleavedAccessPass.cpp @ 134:3a76565eade5 LLVM5.0.1

update 5.0.1
author mir3636
date Sat, 17 Feb 2018 09:57:20 +0900
parents 803732b1fca8
children c2174574ed3a
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
1 //===- InterleavedAccessPass.cpp ------------------------------------------===//
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
2 //
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
3 // The LLVM Compiler Infrastructure
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
4 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
5 // This file is distributed under the University of Illinois Open Source
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
6 // License. See LICENSE.TXT for details.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
7 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
8 //===----------------------------------------------------------------------===//
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
9 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
10 // This file implements the Interleaved Access pass, which identifies
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
11 // interleaved memory accesses and transforms them into target specific
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
12 // intrinsics.
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
13 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
14 // An interleaved load reads data from memory into several vectors, with
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
15 // DE-interleaving the data on a factor. An interleaved store writes several
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
16 // vectors to memory with RE-interleaving the data on a factor.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
17 //
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
18 // As interleaved accesses are difficult to identified in CodeGen (mainly
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
19 // because the VECTOR_SHUFFLE DAG node is quite different from the shufflevector
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
20 // IR), we identify and transform them to intrinsics in this pass so the
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
21 // intrinsics can be easily matched into target specific instructions later in
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
22 // CodeGen.
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
23 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
24 // E.g. An interleaved load (Factor = 2):
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
25 // %wide.vec = load <8 x i32>, <8 x i32>* %ptr
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
26 // %v0 = shuffle <8 x i32> %wide.vec, <8 x i32> undef, <0, 2, 4, 6>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
27 // %v1 = shuffle <8 x i32> %wide.vec, <8 x i32> undef, <1, 3, 5, 7>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
28 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
29 // It could be transformed into a ld2 intrinsic in AArch64 backend or a vld2
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
30 // intrinsic in ARM backend.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
31 //
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
32 // In X86, this can be further optimized into a set of target
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
33 // specific loads followed by an optimized sequence of shuffles.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
34 //
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
35 // E.g. An interleaved store (Factor = 3):
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
36 // %i.vec = shuffle <8 x i32> %v0, <8 x i32> %v1,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
37 // <0, 4, 8, 1, 5, 9, 2, 6, 10, 3, 7, 11>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
38 // store <12 x i32> %i.vec, <12 x i32>* %ptr
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
39 //
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
40 // It could be transformed into a st3 intrinsic in AArch64 backend or a vst3
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
41 // intrinsic in ARM backend.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
42 //
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
43 // Similarly, a set of interleaved stores can be transformed into an optimized
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
44 // sequence of shuffles followed by a set of target specific stores for X86.
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
45 //
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
46 //===----------------------------------------------------------------------===//
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
47
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
48 #include "llvm/ADT/ArrayRef.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
49 #include "llvm/ADT/DenseMap.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
50 #include "llvm/ADT/SmallVector.h"
134
3a76565eade5 update 5.0.1
mir3636
parents: 121
diff changeset
51 #include "llvm/CodeGen/TargetLowering.h"
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
52 #include "llvm/CodeGen/TargetPassConfig.h"
134
3a76565eade5 update 5.0.1
mir3636
parents: 121
diff changeset
53 #include "llvm/CodeGen/TargetSubtargetInfo.h"
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
54 #include "llvm/IR/Constants.h"
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
55 #include "llvm/IR/Dominators.h"
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
56 #include "llvm/IR/Function.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
57 #include "llvm/IR/IRBuilder.h"
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
58 #include "llvm/IR/InstIterator.h"
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
59 #include "llvm/IR/Instruction.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
60 #include "llvm/IR/Instructions.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
61 #include "llvm/IR/Type.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
62 #include "llvm/Pass.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
63 #include "llvm/Support/Casting.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
64 #include "llvm/Support/CommandLine.h"
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
65 #include "llvm/Support/Debug.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
66 #include "llvm/Support/MathExtras.h"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
67 #include "llvm/Support/raw_ostream.h"
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
68 #include "llvm/Target/TargetMachine.h"
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
69 #include <cassert>
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
70 #include <utility>
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
71
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
72 using namespace llvm;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
73
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
74 #define DEBUG_TYPE "interleaved-access"
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
75
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
76 static cl::opt<bool> LowerInterleavedAccesses(
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
77 "lower-interleaved-accesses",
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
78 cl::desc("Enable lowering interleaved accesses to intrinsics"),
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
79 cl::init(true), cl::Hidden);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
80
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
81 namespace {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
82
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
83 class InterleavedAccess : public FunctionPass {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
84 public:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
85 static char ID;
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
86
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
87 InterleavedAccess() : FunctionPass(ID) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
88 initializeInterleavedAccessPass(*PassRegistry::getPassRegistry());
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
89 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
90
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
91 StringRef getPassName() const override { return "Interleaved Access Pass"; }
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
92
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
93 bool runOnFunction(Function &F) override;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
94
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
95 void getAnalysisUsage(AnalysisUsage &AU) const override {
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
96 AU.addRequired<DominatorTreeWrapperPass>();
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
97 AU.addPreserved<DominatorTreeWrapperPass>();
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
98 }
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
99
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
100 private:
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
101 DominatorTree *DT = nullptr;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
102 const TargetLowering *TLI = nullptr;
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
103
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
104 /// The maximum supported interleave factor.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
105 unsigned MaxFactor;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
106
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
107 /// \brief Transform an interleaved load into target specific intrinsics.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
108 bool lowerInterleavedLoad(LoadInst *LI,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
109 SmallVector<Instruction *, 32> &DeadInsts);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
110
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
111 /// \brief Transform an interleaved store into target specific intrinsics.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
112 bool lowerInterleavedStore(StoreInst *SI,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
113 SmallVector<Instruction *, 32> &DeadInsts);
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
114
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
115 /// \brief Returns true if the uses of an interleaved load by the
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
116 /// extractelement instructions in \p Extracts can be replaced by uses of the
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
117 /// shufflevector instructions in \p Shuffles instead. If so, the necessary
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
118 /// replacements are also performed.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
119 bool tryReplaceExtracts(ArrayRef<ExtractElementInst *> Extracts,
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
120 ArrayRef<ShuffleVectorInst *> Shuffles);
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
121 };
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
122
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
123 } // end anonymous namespace.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
124
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
125 char InterleavedAccess::ID = 0;
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
126
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
127 INITIALIZE_PASS_BEGIN(InterleavedAccess, DEBUG_TYPE,
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
128 "Lower interleaved memory accesses to target specific intrinsics", false,
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
129 false)
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
130 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
131 INITIALIZE_PASS_END(InterleavedAccess, DEBUG_TYPE,
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
132 "Lower interleaved memory accesses to target specific intrinsics", false,
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
133 false)
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
134
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
135 FunctionPass *llvm::createInterleavedAccessPass() {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
136 return new InterleavedAccess();
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
137 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
138
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
139 /// \brief Check if the mask is a DE-interleave mask of the given factor
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
140 /// \p Factor like:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
141 /// <Index, Index+Factor, ..., Index+(NumElts-1)*Factor>
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
142 static bool isDeInterleaveMaskOfFactor(ArrayRef<int> Mask, unsigned Factor,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
143 unsigned &Index) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
144 // Check all potential start indices from 0 to (Factor - 1).
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
145 for (Index = 0; Index < Factor; Index++) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
146 unsigned i = 0;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
147
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
148 // Check that elements are in ascending order by Factor. Ignore undef
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
149 // elements.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
150 for (; i < Mask.size(); i++)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
151 if (Mask[i] >= 0 && static_cast<unsigned>(Mask[i]) != Index + i * Factor)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
152 break;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
153
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
154 if (i == Mask.size())
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
155 return true;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
156 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
157
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
158 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
159 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
160
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
161 /// \brief Check if the mask is a DE-interleave mask for an interleaved load.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
162 ///
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
163 /// E.g. DE-interleave masks (Factor = 2) could be:
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
164 /// <0, 2, 4, 6> (mask of index 0 to extract even elements)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
165 /// <1, 3, 5, 7> (mask of index 1 to extract odd elements)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
166 static bool isDeInterleaveMask(ArrayRef<int> Mask, unsigned &Factor,
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
167 unsigned &Index, unsigned MaxFactor) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
168 if (Mask.size() < 2)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
169 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
170
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
171 // Check potential Factors.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
172 for (Factor = 2; Factor <= MaxFactor; Factor++)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
173 if (isDeInterleaveMaskOfFactor(Mask, Factor, Index))
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
174 return true;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
175
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
176 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
177 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
178
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
179 /// \brief Check if the mask can be used in an interleaved store.
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
180 //
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
181 /// It checks for a more general pattern than the RE-interleave mask.
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
182 /// I.e. <x, y, ... z, x+1, y+1, ...z+1, x+2, y+2, ...z+2, ...>
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
183 /// E.g. For a Factor of 2 (LaneLen=4): <4, 32, 5, 33, 6, 34, 7, 35>
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
184 /// E.g. For a Factor of 3 (LaneLen=4): <4, 32, 16, 5, 33, 17, 6, 34, 18, 7, 35, 19>
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
185 /// E.g. For a Factor of 4 (LaneLen=2): <8, 2, 12, 4, 9, 3, 13, 5>
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
186 ///
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
187 /// The particular case of an RE-interleave mask is:
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
188 /// I.e. <0, LaneLen, ... , LaneLen*(Factor - 1), 1, LaneLen + 1, ...>
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
189 /// E.g. For a Factor of 2 (LaneLen=4): <0, 4, 1, 5, 2, 6, 3, 7>
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
190 static bool isReInterleaveMask(ArrayRef<int> Mask, unsigned &Factor,
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
191 unsigned MaxFactor, unsigned OpNumElts) {
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
192 unsigned NumElts = Mask.size();
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
193 if (NumElts < 4)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
194 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
195
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
196 // Check potential Factors.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
197 for (Factor = 2; Factor <= MaxFactor; Factor++) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
198 if (NumElts % Factor)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
199 continue;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
200
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
201 unsigned LaneLen = NumElts / Factor;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
202 if (!isPowerOf2_32(LaneLen))
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
203 continue;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
204
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
205 // Check whether each element matches the general interleaved rule.
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
206 // Ignore undef elements, as long as the defined elements match the rule.
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
207 // Outer loop processes all factors (x, y, z in the above example)
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
208 unsigned I = 0, J;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
209 for (; I < Factor; I++) {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
210 unsigned SavedLaneValue;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
211 unsigned SavedNoUndefs = 0;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
212
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
213 // Inner loop processes consecutive accesses (x, x+1... in the example)
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
214 for (J = 0; J < LaneLen - 1; J++) {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
215 // Lane computes x's position in the Mask
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
216 unsigned Lane = J * Factor + I;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
217 unsigned NextLane = Lane + Factor;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
218 int LaneValue = Mask[Lane];
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
219 int NextLaneValue = Mask[NextLane];
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
220
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
221 // If both are defined, values must be sequential
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
222 if (LaneValue >= 0 && NextLaneValue >= 0 &&
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
223 LaneValue + 1 != NextLaneValue)
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
224 break;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
225
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
226 // If the next value is undef, save the current one as reference
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
227 if (LaneValue >= 0 && NextLaneValue < 0) {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
228 SavedLaneValue = LaneValue;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
229 SavedNoUndefs = 1;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
230 }
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
231
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
232 // Undefs are allowed, but defined elements must still be consecutive:
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
233 // i.e.: x,..., undef,..., x + 2,..., undef,..., undef,..., x + 5, ....
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
234 // Verify this by storing the last non-undef followed by an undef
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
235 // Check that following non-undef masks are incremented with the
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
236 // corresponding distance.
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
237 if (SavedNoUndefs > 0 && LaneValue < 0) {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
238 SavedNoUndefs++;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
239 if (NextLaneValue >= 0 &&
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
240 SavedLaneValue + SavedNoUndefs != (unsigned)NextLaneValue)
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
241 break;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
242 }
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
243 }
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
244
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
245 if (J < LaneLen - 1)
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
246 break;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
247
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
248 int StartMask = 0;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
249 if (Mask[I] >= 0) {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
250 // Check that the start of the I range (J=0) is greater than 0
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
251 StartMask = Mask[I];
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
252 } else if (Mask[(LaneLen - 1) * Factor + I] >= 0) {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
253 // StartMask defined by the last value in lane
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
254 StartMask = Mask[(LaneLen - 1) * Factor + I] - J;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
255 } else if (SavedNoUndefs > 0) {
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
256 // StartMask defined by some non-zero value in the j loop
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
257 StartMask = SavedLaneValue - (LaneLen - 1 - SavedNoUndefs);
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
258 }
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
259 // else StartMask remains set to 0, i.e. all elements are undefs
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
260
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
261 if (StartMask < 0)
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
262 break;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
263 // We must stay within the vectors; This case can happen with undefs.
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
264 if (StartMask + LaneLen > OpNumElts*2)
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
265 break;
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
266 }
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
267
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
268 // Found an interleaved mask of current factor.
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
269 if (I == Factor)
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
270 return true;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
271 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
272
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
273 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
274 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
275
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
276 bool InterleavedAccess::lowerInterleavedLoad(
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
277 LoadInst *LI, SmallVector<Instruction *, 32> &DeadInsts) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
278 if (!LI->isSimple())
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
279 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
280
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
281 SmallVector<ShuffleVectorInst *, 4> Shuffles;
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
282 SmallVector<ExtractElementInst *, 4> Extracts;
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
283
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
284 // Check if all users of this load are shufflevectors. If we encounter any
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
285 // users that are extractelement instructions, we save them to later check if
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
286 // they can be modifed to extract from one of the shufflevectors instead of
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
287 // the load.
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
288 for (auto UI = LI->user_begin(), E = LI->user_end(); UI != E; UI++) {
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
289 auto *Extract = dyn_cast<ExtractElementInst>(*UI);
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
290 if (Extract && isa<ConstantInt>(Extract->getIndexOperand())) {
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
291 Extracts.push_back(Extract);
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
292 continue;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
293 }
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
294 ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(*UI);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
295 if (!SVI || !isa<UndefValue>(SVI->getOperand(1)))
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
296 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
297
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
298 Shuffles.push_back(SVI);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
299 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
300
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
301 if (Shuffles.empty())
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
302 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
303
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
304 unsigned Factor, Index;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
305
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
306 // Check if the first shufflevector is DE-interleave shuffle.
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
307 if (!isDeInterleaveMask(Shuffles[0]->getShuffleMask(), Factor, Index,
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
308 MaxFactor))
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
309 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
310
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
311 // Holds the corresponding index for each DE-interleave shuffle.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
312 SmallVector<unsigned, 4> Indices;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
313 Indices.push_back(Index);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
314
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
315 Type *VecTy = Shuffles[0]->getType();
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
316
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
317 // Check if other shufflevectors are also DE-interleaved of the same type
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
318 // and factor as the first shufflevector.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
319 for (unsigned i = 1; i < Shuffles.size(); i++) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
320 if (Shuffles[i]->getType() != VecTy)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
321 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
322
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
323 if (!isDeInterleaveMaskOfFactor(Shuffles[i]->getShuffleMask(), Factor,
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
324 Index))
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
325 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
326
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
327 Indices.push_back(Index);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
328 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
329
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
330 // Try and modify users of the load that are extractelement instructions to
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
331 // use the shufflevector instructions instead of the load.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
332 if (!tryReplaceExtracts(Extracts, Shuffles))
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
333 return false;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
334
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
335 DEBUG(dbgs() << "IA: Found an interleaved load: " << *LI << "\n");
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
336
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
337 // Try to create target specific intrinsics to replace the load and shuffles.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
338 if (!TLI->lowerInterleavedLoad(LI, Shuffles, Indices, Factor))
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
339 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
340
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
341 for (auto SVI : Shuffles)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
342 DeadInsts.push_back(SVI);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
343
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
344 DeadInsts.push_back(LI);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
345 return true;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
346 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
347
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
348 bool InterleavedAccess::tryReplaceExtracts(
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
349 ArrayRef<ExtractElementInst *> Extracts,
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
350 ArrayRef<ShuffleVectorInst *> Shuffles) {
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
351 // If there aren't any extractelement instructions to modify, there's nothing
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
352 // to do.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
353 if (Extracts.empty())
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
354 return true;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
355
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
356 // Maps extractelement instructions to vector-index pairs. The extractlement
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
357 // instructions will be modified to use the new vector and index operands.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
358 DenseMap<ExtractElementInst *, std::pair<Value *, int>> ReplacementMap;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
359
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
360 for (auto *Extract : Extracts) {
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
361 // The vector index that is extracted.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
362 auto *IndexOperand = cast<ConstantInt>(Extract->getIndexOperand());
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
363 auto Index = IndexOperand->getSExtValue();
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
364
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
365 // Look for a suitable shufflevector instruction. The goal is to modify the
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
366 // extractelement instruction (which uses an interleaved load) to use one
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
367 // of the shufflevector instructions instead of the load.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
368 for (auto *Shuffle : Shuffles) {
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
369 // If the shufflevector instruction doesn't dominate the extract, we
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
370 // can't create a use of it.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
371 if (!DT->dominates(Shuffle, Extract))
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
372 continue;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
373
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
374 // Inspect the indices of the shufflevector instruction. If the shuffle
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
375 // selects the same index that is extracted, we can modify the
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
376 // extractelement instruction.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
377 SmallVector<int, 4> Indices;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
378 Shuffle->getShuffleMask(Indices);
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
379 for (unsigned I = 0; I < Indices.size(); ++I)
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
380 if (Indices[I] == Index) {
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
381 assert(Extract->getOperand(0) == Shuffle->getOperand(0) &&
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
382 "Vector operations do not match");
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
383 ReplacementMap[Extract] = std::make_pair(Shuffle, I);
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
384 break;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
385 }
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
386
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
387 // If we found a suitable shufflevector instruction, stop looking.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
388 if (ReplacementMap.count(Extract))
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
389 break;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
390 }
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
391
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
392 // If we did not find a suitable shufflevector instruction, the
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
393 // extractelement instruction cannot be modified, so we must give up.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
394 if (!ReplacementMap.count(Extract))
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
395 return false;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
396 }
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
397
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
398 // Finally, perform the replacements.
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
399 IRBuilder<> Builder(Extracts[0]->getContext());
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
400 for (auto &Replacement : ReplacementMap) {
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
401 auto *Extract = Replacement.first;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
402 auto *Vector = Replacement.second.first;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
403 auto Index = Replacement.second.second;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
404 Builder.SetInsertPoint(Extract);
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
405 Extract->replaceAllUsesWith(Builder.CreateExtractElement(Vector, Index));
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
406 Extract->eraseFromParent();
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
407 }
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
408
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
409 return true;
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
410 }
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
411
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
412 bool InterleavedAccess::lowerInterleavedStore(
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
413 StoreInst *SI, SmallVector<Instruction *, 32> &DeadInsts) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
414 if (!SI->isSimple())
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
415 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
416
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
417 ShuffleVectorInst *SVI = dyn_cast<ShuffleVectorInst>(SI->getValueOperand());
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
418 if (!SVI || !SVI->hasOneUse())
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
419 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
420
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
421 // Check if the shufflevector is RE-interleave shuffle.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
422 unsigned Factor;
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
423 unsigned OpNumElts = SVI->getOperand(0)->getType()->getVectorNumElements();
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
424 if (!isReInterleaveMask(SVI->getShuffleMask(), Factor, MaxFactor, OpNumElts))
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
425 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
426
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
427 DEBUG(dbgs() << "IA: Found an interleaved store: " << *SI << "\n");
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
428
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
429 // Try to create target specific intrinsics to replace the store and shuffle.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
430 if (!TLI->lowerInterleavedStore(SI, SVI, Factor))
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
431 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
432
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
433 // Already have a new target specific interleaved store. Erase the old store.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
434 DeadInsts.push_back(SI);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
435 DeadInsts.push_back(SVI);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
436 return true;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
437 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
438
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
439 bool InterleavedAccess::runOnFunction(Function &F) {
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
440 auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
441 if (!TPC || !LowerInterleavedAccesses)
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
442 return false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
443
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
444 DEBUG(dbgs() << "*** " << getPassName() << ": " << F.getName() << "\n");
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
445
120
1172e4bd9c6f update 4.0.0
mir3636
parents: 95
diff changeset
446 DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
121
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
447 auto &TM = TPC->getTM<TargetMachine>();
803732b1fca8 LLVM 5.0
kono
parents: 120
diff changeset
448 TLI = TM.getSubtargetImpl(F)->getTargetLowering();
95
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
449 MaxFactor = TLI->getMaxSupportedInterleaveFactor();
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
450
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
451 // Holds dead instructions that will be erased later.
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
452 SmallVector<Instruction *, 32> DeadInsts;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
453 bool Changed = false;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
454
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
455 for (auto &I : instructions(F)) {
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
456 if (LoadInst *LI = dyn_cast<LoadInst>(&I))
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
457 Changed |= lowerInterleavedLoad(LI, DeadInsts);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
458
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
459 if (StoreInst *SI = dyn_cast<StoreInst>(&I))
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
460 Changed |= lowerInterleavedStore(SI, DeadInsts);
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
461 }
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
462
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
463 for (auto I : DeadInsts)
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
464 I->eraseFromParent();
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
465
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
466 return Changed;
afa8332a0e37 LLVM 3.8
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff changeset
467 }