Line data Source code
1 : // Copyright 2014 the V8 project authors. All rights reserved.
2 : // Use of this source code is governed by a BSD-style license that can be
3 : // found in the LICENSE file.
4 :
5 : #include "src/compiler/backend/gap-resolver.h"
6 :
7 : #include <algorithm>
8 : #include <set>
9 :
10 : #include "src/base/enum-set.h"
11 : #include "src/register-configuration.h"
12 :
13 : namespace v8 {
14 : namespace internal {
15 : namespace compiler {
16 :
17 : namespace {
18 :
19 : // Splits a FP move between two location operands into the equivalent series of
20 : // moves between smaller sub-operands, e.g. a double move to two single moves.
21 : // This helps reduce the number of cycles that would normally occur under FP
22 : // aliasing, and makes swaps much easier to implement.
23 : MoveOperands* Split(MoveOperands* move, MachineRepresentation smaller_rep,
24 : ParallelMove* moves) {
25 : DCHECK(!kSimpleFPAliasing);
26 : // Splitting is only possible when the slot size is the same as float size.
27 : DCHECK_EQ(kSystemPointerSize, kFloatSize);
28 : const LocationOperand& src_loc = LocationOperand::cast(move->source());
29 : const LocationOperand& dst_loc = LocationOperand::cast(move->destination());
30 : MachineRepresentation dst_rep = dst_loc.representation();
31 : DCHECK_NE(smaller_rep, dst_rep);
32 : auto src_kind = src_loc.location_kind();
33 : auto dst_kind = dst_loc.location_kind();
34 :
35 : int aliases =
36 : 1 << (ElementSizeLog2Of(dst_rep) - ElementSizeLog2Of(smaller_rep));
37 : int base = -1;
38 : USE(base);
39 : DCHECK_EQ(aliases, RegisterConfiguration::Default()->GetAliases(
40 : dst_rep, 0, smaller_rep, &base));
41 :
42 : int src_index = -1;
43 : int slot_size = (1 << ElementSizeLog2Of(smaller_rep)) / kSystemPointerSize;
44 : int src_step = 1;
45 : if (src_kind == LocationOperand::REGISTER) {
46 : src_index = src_loc.register_code() * aliases;
47 : } else {
48 : src_index = src_loc.index();
49 : // For operands that occupy multiple slots, the index refers to the last
50 : // slot. On little-endian architectures, we start at the high slot and use a
51 : // negative step so that register-to-slot moves are in the correct order.
52 : src_step = -slot_size;
53 : }
54 : int dst_index = -1;
55 : int dst_step = 1;
56 : if (dst_kind == LocationOperand::REGISTER) {
57 : dst_index = dst_loc.register_code() * aliases;
58 : } else {
59 : dst_index = dst_loc.index();
60 : dst_step = -slot_size;
61 : }
62 :
63 : // Reuse 'move' for the first fragment. It is not pending.
64 : move->set_source(AllocatedOperand(src_kind, smaller_rep, src_index));
65 : move->set_destination(AllocatedOperand(dst_kind, smaller_rep, dst_index));
66 : // Add the remaining fragment moves.
67 : for (int i = 1; i < aliases; ++i) {
68 : src_index += src_step;
69 : dst_index += dst_step;
70 : moves->AddMove(AllocatedOperand(src_kind, smaller_rep, src_index),
71 : AllocatedOperand(dst_kind, smaller_rep, dst_index));
72 : }
73 : // Return the first fragment.
74 : return move;
75 : }
76 :
77 : enum MoveOperandKind : uint8_t { kConstant, kGpReg, kFpReg, kStack };
78 :
79 : MoveOperandKind GetKind(const InstructionOperand& move) {
80 77847642 : if (move.IsConstant()) return kConstant;
81 : LocationOperand loc_op = LocationOperand::cast(move);
82 58281826 : if (loc_op.location_kind() != LocationOperand::REGISTER) return kStack;
83 43020745 : return IsFloatingPoint(loc_op.representation()) ? kFpReg : kGpReg;
84 : }
85 :
86 : } // namespace
87 :
88 49638657 : void GapResolver::Resolve(ParallelMove* moves) {
89 : base::EnumSet<MoveOperandKind, uint8_t> source_kinds;
90 : base::EnumSet<MoveOperandKind, uint8_t> destination_kinds;
91 :
92 : // Remove redundant moves, collect source kinds and destination kinds to
93 : // detect simple non-overlapping moves, and collect FP move representations if
94 : // aliasing is non-simple.
95 : int fp_reps = 0;
96 107975378 : for (auto it = moves->begin(); it != moves->end();) {
97 58335587 : MoveOperands* move = *it;
98 58335587 : if (move->IsRedundant()) {
99 19412980 : it = moves->erase(it);
100 19412900 : continue;
101 : }
102 : source_kinds.Add(GetKind(move->source()));
103 : destination_kinds.Add(GetKind(move->destination()));
104 : if (!kSimpleFPAliasing && move->destination().IsFPRegister()) {
105 : fp_reps |= RepresentationBit(
106 : LocationOperand::cast(move->destination()).representation());
107 : }
108 : ++it;
109 : }
110 :
111 55526462 : if ((source_kinds & destination_kinds).empty() || moves->size() < 2) {
112 : // Fast path for non-conflicting parallel moves.
113 70324727 : for (MoveOperands* move : *moves) {
114 49328672 : assembler_->AssembleMove(&move->source(), &move->destination());
115 : }
116 : return;
117 : }
118 :
119 : if (!kSimpleFPAliasing) {
120 : if (fp_reps && !base::bits::IsPowerOfTwo(fp_reps)) {
121 : // Start with the smallest FP moves, so we never encounter smaller moves
122 : // in the middle of a cycle of larger moves.
123 : if ((fp_reps & RepresentationBit(MachineRepresentation::kFloat32)) != 0) {
124 : split_rep_ = MachineRepresentation::kFloat32;
125 : for (size_t i = 0; i < moves->size(); ++i) {
126 : auto move = (*moves)[i];
127 : if (!move->IsEliminated() && move->destination().IsFloatRegister())
128 : PerformMove(moves, move);
129 : }
130 : }
131 : if ((fp_reps & RepresentationBit(MachineRepresentation::kFloat64)) != 0) {
132 : split_rep_ = MachineRepresentation::kFloat64;
133 : for (size_t i = 0; i < moves->size(); ++i) {
134 : auto move = (*moves)[i];
135 : if (!move->IsEliminated() && move->destination().IsDoubleRegister())
136 : PerformMove(moves, move);
137 : }
138 : }
139 : }
140 : split_rep_ = MachineRepresentation::kSimd128;
141 : }
142 :
143 32497178 : for (size_t i = 0; i < moves->size(); ++i) {
144 14259464 : auto move = (*moves)[i];
145 14259464 : if (!move->IsEliminated()) PerformMove(moves, move);
146 : }
147 : }
148 :
149 14259469 : void GapResolver::PerformMove(ParallelMove* moves, MoveOperands* move) {
150 : // Each call to this function performs a move and deletes it from the move
151 : // graph. We first recursively perform any move blocking this one. We mark a
152 : // move as "pending" on entry to PerformMove in order to detect cycles in the
153 : // move graph. We use operand swaps to resolve cycles, which means that a
154 : // call to PerformMove could change any source operand in the move graph.
155 : DCHECK(!move->IsPending());
156 : DCHECK(!move->IsRedundant());
157 :
158 : // Clear this move's destination to indicate a pending move. The actual
159 : // destination is saved on the side.
160 14259469 : InstructionOperand source = move->source();
161 : DCHECK(!source.IsInvalid()); // Or else it will look eliminated.
162 14259469 : InstructionOperand destination = move->destination();
163 : move->SetPending();
164 :
165 : // We may need to split moves between FP locations differently.
166 : const bool is_fp_loc_move =
167 : !kSimpleFPAliasing && destination.IsFPLocationOperand();
168 :
169 : // Perform a depth-first traversal of the move graph to resolve dependencies.
170 : // Any unperformed, unpending move with a source the same as this one's
171 : // destination blocks this one so recursively perform all such moves.
172 145391399 : for (size_t i = 0; i < moves->size(); ++i) {
173 65565937 : auto other = (*moves)[i];
174 65565937 : if (other->IsEliminated()) continue;
175 40536288 : if (other->IsPending()) continue;
176 25609840 : if (other->source().InterferesWith(destination)) {
177 : if (is_fp_loc_move &&
178 : LocationOperand::cast(other->source()).representation() >
179 : split_rep_) {
180 : // 'other' must also be an FP location move. Break it into fragments
181 : // of the same size as 'move'. 'other' is set to one of the fragments,
182 : // and the rest are appended to 'moves'.
183 : other = Split(other, split_rep_, moves);
184 : // 'other' may not block destination now.
185 : if (!other->source().InterferesWith(destination)) continue;
186 : }
187 : // Though PerformMove can change any source operand in the move graph,
188 : // this call cannot create a blocking move via a swap (this loop does not
189 : // miss any). Assume there is a non-blocking move with source A and this
190 : // move is blocked on source B and there is a swap of A and B. Then A and
191 : // B must be involved in the same cycle (or they would not be swapped).
192 : // Since this move's destination is B and there is only a single incoming
193 : // edge to an operand, this move must also be involved in the same cycle.
194 : // In that case, the blocking move will be created but will be "pending"
195 : // when we return from PerformMove.
196 587816 : PerformMove(moves, other);
197 : }
198 : }
199 :
200 : // This move's source may have changed due to swaps to resolve cycles and so
201 : // it may now be the last move in the cycle. If so remove it.
202 14259497 : source = move->source();
203 14259492 : if (source.EqualsCanonicalized(destination)) {
204 : move->Eliminate();
205 14190806 : return;
206 : }
207 :
208 : // We are about to resolve this move and don't need it marked as pending, so
209 : // restore its destination.
210 : move->set_destination(destination);
211 :
212 : // The move may be blocked on a (at most one) pending move, in which case we
213 : // have a cycle. Search for such a blocking move and perform a swap to
214 : // resolve it.
215 : auto blocker =
216 : std::find_if(moves->begin(), moves->end(), [&](MoveOperands* move) {
217 104220788 : return !move->IsEliminated() &&
218 78842350 : move->source().InterferesWith(destination);
219 : });
220 14201926 : if (blocker == moves->end()) {
221 : // The easy case: This move is not blocked.
222 14133267 : assembler_->AssembleMove(&source, &destination);
223 : move->Eliminate();
224 : return;
225 : }
226 :
227 : // Ensure source is a register or both are stack slots, to limit swap cases.
228 132139 : if (source.IsStackSlot() || source.IsFPStackSlot()) {
229 : std::swap(source, destination);
230 : }
231 68659 : assembler_->AssembleSwap(&source, &destination);
232 : move->Eliminate();
233 :
234 : // Update outstanding moves whose source may now have been moved.
235 : if (is_fp_loc_move) {
236 : // We may have to split larger moves.
237 : for (size_t i = 0; i < moves->size(); ++i) {
238 : auto other = (*moves)[i];
239 : if (other->IsEliminated()) continue;
240 : if (source.InterferesWith(other->source())) {
241 : if (LocationOperand::cast(other->source()).representation() >
242 : split_rep_) {
243 : other = Split(other, split_rep_, moves);
244 : if (!source.InterferesWith(other->source())) continue;
245 : }
246 : other->set_source(destination);
247 : } else if (destination.InterferesWith(other->source())) {
248 : if (LocationOperand::cast(other->source()).representation() >
249 : split_rep_) {
250 : other = Split(other, split_rep_, moves);
251 : if (!destination.InterferesWith(other->source())) continue;
252 : }
253 : other->set_source(source);
254 : }
255 : }
256 : } else {
257 604127 : for (auto other : *moves) {
258 535469 : if (other->IsEliminated()) continue;
259 284736 : if (source.EqualsCanonicalized(other->source())) {
260 : other->set_source(destination);
261 273053 : } else if (destination.EqualsCanonicalized(other->source())) {
262 : other->set_source(source);
263 : }
264 : }
265 : }
266 : }
267 : } // namespace compiler
268 : } // namespace internal
269 120216 : } // namespace v8
|
