1 //=- WebAssemblyMCCodeEmitter.cpp - Convert WebAssembly code to machine code -//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 ///
10 /// \file
11 /// This file implements the WebAssemblyMCCodeEmitter class.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "MCTargetDesc/WebAssemblyFixupKinds.h"
16 #include "MCTargetDesc/WebAssemblyMCTargetDesc.h"
17 #include "llvm/ADT/STLExtras.h"
18 #include "llvm/ADT/Statistic.h"
19 #include "llvm/MC/MCCodeEmitter.h"
20 #include "llvm/MC/MCFixup.h"
21 #include "llvm/MC/MCInst.h"
22 #include "llvm/MC/MCInstrInfo.h"
23 #include "llvm/MC/MCRegisterInfo.h"
24 #include "llvm/MC/MCSubtargetInfo.h"
25 #include "llvm/MC/MCSymbol.h"
26 #include "llvm/Support/Debug.h"
27 #include "llvm/Support/EndianStream.h"
28 #include "llvm/Support/LEB128.h"
29 #include "llvm/Support/raw_ostream.h"
30 
31 using namespace llvm;
32 
33 #define DEBUG_TYPE "mccodeemitter"
34 
35 STATISTIC(MCNumEmitted, "Number of MC instructions emitted.");
36 STATISTIC(MCNumFixups, "Number of MC fixups created.");
37 
38 namespace {
39 class WebAssemblyMCCodeEmitter final : public MCCodeEmitter {
40   const MCInstrInfo &MCII;
41 
42   // Implementation generated by tablegen.
43   uint64_t getBinaryCodeForInstr(const MCInst &MI,
44                                  SmallVectorImpl<MCFixup> &Fixups,
45                                  const MCSubtargetInfo &STI) const;
46 
47   void encodeInstruction(const MCInst &MI, raw_ostream &OS,
48                          SmallVectorImpl<MCFixup> &Fixups,
49                          const MCSubtargetInfo &STI) const override;
50 
51 public:
52   WebAssemblyMCCodeEmitter(const MCInstrInfo &mcii) : MCII(mcii) {}
53 };
54 } // end anonymous namespace
55 
56 MCCodeEmitter *llvm::createWebAssemblyMCCodeEmitter(const MCInstrInfo &MCII) {
57   return new WebAssemblyMCCodeEmitter(MCII);
58 }
59 
60 void WebAssemblyMCCodeEmitter::encodeInstruction(
61     const MCInst &MI, raw_ostream &OS, SmallVectorImpl<MCFixup> &Fixups,
62     const MCSubtargetInfo &STI) const {
63   uint64_t Start = OS.tell();
64 
65   uint64_t Binary = getBinaryCodeForInstr(MI, Fixups, STI);
66   if (Binary <= UINT8_MAX) {
67     OS << uint8_t(Binary);
68   } else {
69     assert(Binary <= UINT16_MAX && "Several-byte opcodes not supported yet");
70     OS << uint8_t(Binary >> 8)
71        << uint8_t(Binary);
72   }
73 
74   // For br_table instructions, encode the size of the table. In the MCInst,
75   // there's an index operand, one operand for each table entry, and the
76   // default operand.
77   if (MI.getOpcode() == WebAssembly::BR_TABLE_I32 ||
78       MI.getOpcode() == WebAssembly::BR_TABLE_I64)
79     encodeULEB128(MI.getNumOperands() - 2, OS);
80 
81   const MCInstrDesc &Desc = MCII.get(MI.getOpcode());
82   for (unsigned i = 0, e = MI.getNumOperands(); i < e; ++i) {
83     const MCOperand &MO = MI.getOperand(i);
84     if (MO.isReg()) {
85       /* nothing to encode */
86     } else if (MO.isImm()) {
87       if (i < Desc.getNumOperands()) {
88         assert(Desc.TSFlags == 0 &&
89                "WebAssembly non-variable_ops don't use TSFlags");
90         const MCOperandInfo &Info = Desc.OpInfo[i];
91         LLVM_DEBUG(dbgs() << "Encoding immediate: type="
92                           << int(Info.OperandType) << "\n");
93         switch (Info.OperandType) {
94         case WebAssembly::OPERAND_I32IMM:
95           encodeSLEB128(int32_t(MO.getImm()), OS);
96           break;
97         case WebAssembly::OPERAND_OFFSET32:
98           encodeULEB128(uint32_t(MO.getImm()), OS);
99           break;
100         case WebAssembly::OPERAND_I64IMM:
101           encodeSLEB128(int64_t(MO.getImm()), OS);
102           break;
103         case WebAssembly::OPERAND_SIGNATURE:
104           OS << uint8_t(MO.getImm());
105           break;
106         case WebAssembly::OPERAND_VEC_I8IMM:
107           support::endian::write<uint8_t>(OS, MO.getImm(), support::little);
108           break;
109         case WebAssembly::OPERAND_VEC_I16IMM:
110           support::endian::write<uint16_t>(OS, MO.getImm(), support::little);
111           break;
112         case WebAssembly::OPERAND_VEC_I32IMM:
113           support::endian::write<uint32_t>(OS, MO.getImm(), support::little);
114           break;
115         case WebAssembly::OPERAND_VEC_I64IMM:
116           support::endian::write<uint64_t>(OS, MO.getImm(), support::little);
117           break;
118         case WebAssembly::OPERAND_GLOBAL:
119           llvm_unreachable("wasm globals should only be accessed symbolicly");
120         default:
121           encodeULEB128(uint64_t(MO.getImm()), OS);
122         }
123       } else {
124         assert(Desc.TSFlags == (WebAssemblyII::VariableOpIsImmediate |
125                                 WebAssemblyII::VariableOpImmediateIsLabel));
126         encodeULEB128(uint64_t(MO.getImm()), OS);
127       }
128     } else if (MO.isFPImm()) {
129       assert(i < Desc.getNumOperands() &&
130              "Unexpected floating-point immediate as a non-fixed operand");
131       assert(Desc.TSFlags == 0 &&
132              "WebAssembly variable_ops floating point ops don't use TSFlags");
133       const MCOperandInfo &Info = Desc.OpInfo[i];
134       if (Info.OperandType == WebAssembly::OPERAND_F32IMM) {
135         // TODO: MC converts all floating point immediate operands to double.
136         // This is fine for numeric values, but may cause NaNs to change bits.
137         float f = float(MO.getFPImm());
138         support::endian::write<float>(OS, f, support::little);
139       } else {
140         assert(Info.OperandType == WebAssembly::OPERAND_F64IMM);
141         double d = MO.getFPImm();
142         support::endian::write<double>(OS, d, support::little);
143       }
144     } else if (MO.isExpr()) {
145       const MCOperandInfo &Info = Desc.OpInfo[i];
146       llvm::MCFixupKind FixupKind;
147       size_t PaddedSize = 5;
148       if (Info.OperandType == WebAssembly::OPERAND_I32IMM) {
149         FixupKind = MCFixupKind(WebAssembly::fixup_code_sleb128_i32);
150       } else if (Info.OperandType == WebAssembly::OPERAND_I64IMM) {
151         FixupKind = MCFixupKind(WebAssembly::fixup_code_sleb128_i64);
152         PaddedSize = 10;
153       } else if (Info.OperandType == WebAssembly::OPERAND_FUNCTION32 ||
154                  Info.OperandType == WebAssembly::OPERAND_OFFSET32 ||
155                  Info.OperandType == WebAssembly::OPERAND_TYPEINDEX) {
156         FixupKind = MCFixupKind(WebAssembly::fixup_code_uleb128_i32);
157       } else if (Info.OperandType == WebAssembly::OPERAND_GLOBAL) {
158         FixupKind = MCFixupKind(WebAssembly::fixup_code_uleb128_i32);
159       } else {
160         llvm_unreachable("unexpected symbolic operand kind");
161       }
162       Fixups.push_back(MCFixup::create(
163           OS.tell() - Start, MO.getExpr(),
164           FixupKind, MI.getLoc()));
165       ++MCNumFixups;
166       encodeULEB128(0, OS, PaddedSize);
167     } else {
168       llvm_unreachable("unexpected operand kind");
169     }
170   }
171 
172   ++MCNumEmitted; // Keep track of the # of mi's emitted.
173 }
174 
175 #include "WebAssemblyGenMCCodeEmitter.inc"
176