[SLP] Add cost model for `llvm.powi.*` intrinsics (REAPPLIED)
[llvm-project.git] / lld / MachO / Arch / X86_64.cpp
blobd675356b9ffb968e657b4fbfd312eb5600bad365
1 //===- X86_64.cpp ---------------------------------------------------------===//
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 //===----------------------------------------------------------------------===//
9 #include "InputFiles.h"
10 #include "Symbols.h"
11 #include "SyntheticSections.h"
12 #include "Target.h"
14 #include "lld/Common/ErrorHandler.h"
15 #include "mach-o/compact_unwind_encoding.h"
16 #include "llvm/BinaryFormat/MachO.h"
17 #include "llvm/Support/Endian.h"
19 using namespace llvm::MachO;
20 using namespace llvm::support::endian;
21 using namespace lld;
22 using namespace lld::macho;
24 namespace {
26 struct X86_64 : TargetInfo {
27 X86_64();
29 int64_t getEmbeddedAddend(MemoryBufferRef, uint64_t offset,
30 const relocation_info) const override;
31 void relocateOne(uint8_t *loc, const Reloc &, uint64_t va,
32 uint64_t relocVA) const override;
34 void writeStub(uint8_t *buf, const Symbol &) const override;
35 void writeStubHelperHeader(uint8_t *buf) const override;
36 void writeStubHelperEntry(uint8_t *buf, const Symbol &,
37 uint64_t entryAddr) const override;
39 void relaxGotLoad(uint8_t *loc, uint8_t type) const override;
40 const RelocAttrs &getRelocAttrs(uint8_t type) const override;
41 uint64_t getPageSize() const override { return 4 * 1024; }
44 } // namespace
46 const RelocAttrs &X86_64::getRelocAttrs(uint8_t type) const {
47 static const std::array<RelocAttrs, 10> relocAttrsArray{{
48 #define B(x) RelocAttrBits::x
49 {"UNSIGNED",
50 B(UNSIGNED) | B(ABSOLUTE) | B(EXTERN) | B(LOCAL) | B(BYTE4) | B(BYTE8)},
51 {"SIGNED", B(PCREL) | B(EXTERN) | B(LOCAL) | B(BYTE4)},
52 {"BRANCH", B(PCREL) | B(EXTERN) | B(BRANCH) | B(BYTE4)},
53 {"GOT_LOAD", B(PCREL) | B(EXTERN) | B(GOT) | B(LOAD) | B(BYTE4)},
54 {"GOT", B(PCREL) | B(EXTERN) | B(GOT) | B(POINTER) | B(BYTE4)},
55 {"SUBTRACTOR", B(SUBTRAHEND) | B(EXTERN) | B(BYTE4) | B(BYTE8)},
56 {"SIGNED_1", B(PCREL) | B(EXTERN) | B(LOCAL) | B(BYTE4)},
57 {"SIGNED_2", B(PCREL) | B(EXTERN) | B(LOCAL) | B(BYTE4)},
58 {"SIGNED_4", B(PCREL) | B(EXTERN) | B(LOCAL) | B(BYTE4)},
59 {"TLV", B(PCREL) | B(EXTERN) | B(TLV) | B(LOAD) | B(BYTE4)},
60 #undef B
61 }};
62 assert(type < relocAttrsArray.size() && "invalid relocation type");
63 if (type >= relocAttrsArray.size())
64 return invalidRelocAttrs;
65 return relocAttrsArray[type];
68 static int pcrelOffset(uint8_t type) {
69 switch (type) {
70 case X86_64_RELOC_SIGNED_1:
71 return 1;
72 case X86_64_RELOC_SIGNED_2:
73 return 2;
74 case X86_64_RELOC_SIGNED_4:
75 return 4;
76 default:
77 return 0;
81 int64_t X86_64::getEmbeddedAddend(MemoryBufferRef mb, uint64_t offset,
82 relocation_info rel) const {
83 auto *buf = reinterpret_cast<const uint8_t *>(mb.getBufferStart());
84 const uint8_t *loc = buf + offset + rel.r_address;
86 switch (rel.r_length) {
87 case 2:
88 return static_cast<int32_t>(read32le(loc)) + pcrelOffset(rel.r_type);
89 case 3:
90 return read64le(loc) + pcrelOffset(rel.r_type);
91 default:
92 llvm_unreachable("invalid r_length");
96 void X86_64::relocateOne(uint8_t *loc, const Reloc &r, uint64_t value,
97 uint64_t relocVA) const {
98 if (r.pcrel) {
99 uint64_t pc = relocVA + 4 + pcrelOffset(r.type);
100 value -= pc;
103 switch (r.length) {
104 case 2:
105 if (r.type == X86_64_RELOC_UNSIGNED)
106 checkUInt(loc, r, value, 32);
107 else
108 checkInt(loc, r, value, 32);
109 write32le(loc, value);
110 break;
111 case 3:
112 write64le(loc, value);
113 break;
114 default:
115 llvm_unreachable("invalid r_length");
119 // The following methods emit a number of assembly sequences with RIP-relative
120 // addressing. Note that RIP-relative addressing on X86-64 has the RIP pointing
121 // to the next instruction, not the current instruction, so we always have to
122 // account for the current instruction's size when calculating offsets.
123 // writeRipRelative helps with that.
125 // bufAddr: The virtual address corresponding to buf[0].
126 // bufOff: The offset within buf of the next instruction.
127 // destAddr: The destination address that the current instruction references.
128 static void writeRipRelative(SymbolDiagnostic d, uint8_t *buf, uint64_t bufAddr,
129 uint64_t bufOff, uint64_t destAddr) {
130 uint64_t rip = bufAddr + bufOff;
131 checkInt(buf, d, destAddr - rip, 32);
132 // For the instructions we care about, the RIP-relative address is always
133 // stored in the last 4 bytes of the instruction.
134 write32le(buf + bufOff - 4, destAddr - rip);
137 static constexpr uint8_t stub[] = {
138 0xff, 0x25, 0, 0, 0, 0, // jmpq *__la_symbol_ptr(%rip)
141 void X86_64::writeStub(uint8_t *buf, const Symbol &sym) const {
142 memcpy(buf, stub, 2); // just copy the two nonzero bytes
143 uint64_t stubAddr = in.stubs->addr + sym.stubsIndex * sizeof(stub);
144 writeRipRelative({&sym, "stub"}, buf, stubAddr, sizeof(stub),
145 in.lazyPointers->addr + sym.stubsIndex * LP64::wordSize);
148 static constexpr uint8_t stubHelperHeader[] = {
149 0x4c, 0x8d, 0x1d, 0, 0, 0, 0, // 0x0: leaq ImageLoaderCache(%rip), %r11
150 0x41, 0x53, // 0x7: pushq %r11
151 0xff, 0x25, 0, 0, 0, 0, // 0x9: jmpq *dyld_stub_binder@GOT(%rip)
152 0x90, // 0xf: nop
155 void X86_64::writeStubHelperHeader(uint8_t *buf) const {
156 memcpy(buf, stubHelperHeader, sizeof(stubHelperHeader));
157 SymbolDiagnostic d = {nullptr, "stub helper header"};
158 writeRipRelative(d, buf, in.stubHelper->addr, 7,
159 in.imageLoaderCache->getVA());
160 writeRipRelative(d, buf, in.stubHelper->addr, 0xf,
161 in.got->addr +
162 in.stubHelper->stubBinder->gotIndex * LP64::wordSize);
165 static constexpr uint8_t stubHelperEntry[] = {
166 0x68, 0, 0, 0, 0, // 0x0: pushq <bind offset>
167 0xe9, 0, 0, 0, 0, // 0x5: jmp <__stub_helper>
170 void X86_64::writeStubHelperEntry(uint8_t *buf, const Symbol &sym,
171 uint64_t entryAddr) const {
172 memcpy(buf, stubHelperEntry, sizeof(stubHelperEntry));
173 write32le(buf + 1, sym.lazyBindOffset);
174 writeRipRelative({&sym, "stub helper"}, buf, entryAddr,
175 sizeof(stubHelperEntry), in.stubHelper->addr);
178 void X86_64::relaxGotLoad(uint8_t *loc, uint8_t type) const {
179 // Convert MOVQ to LEAQ
180 if (loc[-2] != 0x8b)
181 error(getRelocAttrs(type).name + " reloc requires MOVQ instruction");
182 loc[-2] = 0x8d;
185 X86_64::X86_64() : TargetInfo(LP64()) {
186 cpuType = CPU_TYPE_X86_64;
187 cpuSubtype = CPU_SUBTYPE_X86_64_ALL;
189 modeDwarfEncoding = UNWIND_X86_MODE_DWARF;
190 subtractorRelocType = X86_64_RELOC_SUBTRACTOR;
191 unsignedRelocType = X86_64_RELOC_UNSIGNED;
193 stubSize = sizeof(stub);
194 stubHelperHeaderSize = sizeof(stubHelperHeader);
195 stubHelperEntrySize = sizeof(stubHelperEntry);
198 TargetInfo *macho::createX86_64TargetInfo() {
199 static X86_64 t;
200 return &t;