1 | /* |
2 | * Copyright (C) 2017 Apple Inc. All rights reserved. |
3 | * |
4 | * Redistribution and use in source and binary forms, with or without |
5 | * modification, are permitted provided that the following conditions |
6 | * are met: |
7 | * 1. Redistributions of source code must retain the above copyright |
8 | * notice, this list of conditions and the following disclaimer. |
9 | * 2. Redistributions in binary form must reproduce the above copyright |
10 | * notice, this list of conditions and the following disclaimer in the |
11 | * documentation and/or other materials provided with the distribution. |
12 | * |
13 | * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY |
14 | * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
15 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
16 | * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR |
17 | * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
18 | * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
19 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
20 | * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY |
21 | * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
22 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
23 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
24 | */ |
25 | |
26 | #include "config.h" |
27 | #include "AirLowerStackArgs.h" |
28 | |
29 | #if ENABLE(B3_JIT) |
30 | |
31 | #include "AirArgInlines.h" |
32 | #include "AirCode.h" |
33 | #include "AirInsertionSet.h" |
34 | #include "AirInstInlines.h" |
35 | #include "AirPhaseScope.h" |
36 | #include "StackAlignment.h" |
37 | #include <wtf/ListDump.h> |
38 | |
39 | namespace JSC { namespace B3 { namespace Air { |
40 | |
41 | void lowerStackArgs(Code& code) |
42 | { |
43 | PhaseScope phaseScope(code, "lowerStackArgs" ); |
44 | |
45 | // Now we need to deduce how much argument area we need. |
46 | for (BasicBlock* block : code) { |
47 | for (Inst& inst : *block) { |
48 | for (Arg& arg : inst.args) { |
49 | if (arg.isCallArg()) { |
50 | // For now, we assume that we use 8 bytes of the call arg. But that's not |
51 | // such an awesome assumption. |
52 | // FIXME: https://bugs.webkit.org/show_bug.cgi?id=150454 |
53 | ASSERT(arg.offset() >= 0); |
54 | code.requestCallArgAreaSizeInBytes(arg.offset() + 8); |
55 | } |
56 | } |
57 | } |
58 | } |
59 | |
60 | code.setFrameSize(code.frameSize() + code.callArgAreaSizeInBytes()); |
61 | |
62 | // Finally, transform the code to use Addr's instead of StackSlot's. This is a lossless |
63 | // transformation since we can search the StackSlots array to figure out which StackSlot any |
64 | // offset-from-FP refers to. |
65 | |
66 | InsertionSet insertionSet(code); |
67 | for (BasicBlock* block : code) { |
68 | // FIXME We can keep track of the last large offset which was materialized in this block, and reuse the register |
69 | // if it hasn't been clobbered instead of renetating imm+add+addr every time. https://bugs.webkit.org/show_bug.cgi?id=171387 |
70 | |
71 | for (unsigned instIndex = 0; instIndex < block->size(); ++instIndex) { |
72 | Inst& inst = block->at(instIndex); |
73 | |
74 | if (isARM64() && (inst.kind.opcode == Lea32 || inst.kind.opcode == Lea64)) { |
75 | // On ARM64, Lea is just an add. We can't handle this below because |
76 | // taking into account the Width to see if we can compute the immediate |
77 | // is wrong. |
78 | auto lowerArmLea = [&] (Value::OffsetType offset, Tmp base) { |
79 | ASSERT(inst.args[1].isTmp()); |
80 | |
81 | if (Arg::isValidImmForm(offset)) |
82 | inst = Inst(inst.kind.opcode == Lea32 ? Add32 : Add64, inst.origin, Arg::imm(offset), base, inst.args[1]); |
83 | else { |
84 | ASSERT(pinnedExtendedOffsetAddrRegister()); |
85 | Air::Tmp tmp = Air::Tmp(*pinnedExtendedOffsetAddrRegister()); |
86 | Arg offsetArg = Arg::bigImm(offset); |
87 | insertionSet.insert(instIndex, Move, inst.origin, offsetArg, tmp); |
88 | inst = Inst(inst.kind.opcode == Lea32 ? Add32 : Add64, inst.origin, tmp, base, inst.args[1]); |
89 | } |
90 | }; |
91 | |
92 | switch (inst.args[0].kind()) { |
93 | case Arg::Stack: { |
94 | StackSlot* slot = inst.args[0].stackSlot(); |
95 | lowerArmLea(inst.args[0].offset() + slot->offsetFromFP(), Tmp(GPRInfo::callFrameRegister)); |
96 | break; |
97 | } |
98 | case Arg::CallArg: |
99 | lowerArmLea(inst.args[0].offset() - code.frameSize(), Tmp(GPRInfo::callFrameRegister)); |
100 | break; |
101 | case Arg::Addr: |
102 | lowerArmLea(inst.args[0].offset(), inst.args[0].base()); |
103 | break; |
104 | case Arg::ExtendedOffsetAddr: |
105 | ASSERT_NOT_REACHED(); |
106 | break; |
107 | default: |
108 | break; |
109 | } |
110 | |
111 | continue; |
112 | } |
113 | |
114 | inst.forEachArg( |
115 | [&] (Arg& arg, Arg::Role role, Bank, Width width) { |
116 | auto stackAddr = [&] (Value::OffsetType offsetFromFP) -> Arg { |
117 | int32_t offsetFromSP = offsetFromFP + code.frameSize(); |
118 | |
119 | if (inst.admitsExtendedOffsetAddr(arg)) { |
120 | // Stackmaps and patchpoints expect addr inputs relative to SP or FP only. We might as well |
121 | // not even bother generating an addr with valid form for these opcodes since extended offset |
122 | // addr is always valid. |
123 | return Arg::extendedOffsetAddr(offsetFromFP); |
124 | } |
125 | |
126 | Arg result = Arg::addr(Air::Tmp(GPRInfo::callFrameRegister), offsetFromFP); |
127 | if (result.isValidForm(width)) |
128 | return result; |
129 | |
130 | result = Arg::addr(Air::Tmp(MacroAssembler::stackPointerRegister), offsetFromSP); |
131 | if (result.isValidForm(width)) |
132 | return result; |
133 | #if CPU(ARM64) |
134 | ASSERT(pinnedExtendedOffsetAddrRegister()); |
135 | Air::Tmp tmp = Air::Tmp(*pinnedExtendedOffsetAddrRegister()); |
136 | |
137 | Arg largeOffset = Arg::isValidImmForm(offsetFromSP) ? Arg::imm(offsetFromSP) : Arg::bigImm(offsetFromSP); |
138 | insertionSet.insert(instIndex, Move, inst.origin, largeOffset, tmp); |
139 | insertionSet.insert(instIndex, Add64, inst.origin, Air::Tmp(MacroAssembler::stackPointerRegister), tmp); |
140 | result = Arg::addr(tmp, 0); |
141 | return result; |
142 | #elif CPU(X86_64) |
143 | // Can't happen on x86: immediates are always big enough for frame size. |
144 | RELEASE_ASSERT_NOT_REACHED(); |
145 | #else |
146 | #error Unhandled architecture. |
147 | #endif |
148 | }; |
149 | |
150 | switch (arg.kind()) { |
151 | case Arg::Stack: { |
152 | StackSlot* slot = arg.stackSlot(); |
153 | if (Arg::isZDef(role) |
154 | && slot->kind() == StackSlotKind::Spill |
155 | && slot->byteSize() > bytes(width)) { |
156 | // Currently we only handle this simple case because it's the only one |
157 | // that arises: ZDef's are only 32-bit right now. So, when we hit these |
158 | // assertions it means that we need to implement those other kinds of |
159 | // zero fills. |
160 | RELEASE_ASSERT(slot->byteSize() == 8); |
161 | RELEASE_ASSERT(width == Width32); |
162 | |
163 | RELEASE_ASSERT(isValidForm(StoreZero32, Arg::Stack)); |
164 | insertionSet.insert( |
165 | instIndex + 1, StoreZero32, inst.origin, |
166 | stackAddr(arg.offset() + 4 + slot->offsetFromFP())); |
167 | } |
168 | arg = stackAddr(arg.offset() + slot->offsetFromFP()); |
169 | break; |
170 | } |
171 | case Arg::CallArg: |
172 | arg = stackAddr(arg.offset() - code.frameSize()); |
173 | break; |
174 | default: |
175 | break; |
176 | } |
177 | } |
178 | ); |
179 | } |
180 | insertionSet.execute(block); |
181 | } |
182 | } |
183 | |
184 | } } } // namespace JSC::B3::Air |
185 | |
186 | #endif // ENABLE(B3_JIT) |
187 | |
188 | |