1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175
|
//===- WebAssemblyDisassemblerEmitter.cpp - Disassembler tables -*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file is part of the WebAssembly Disassembler Emitter.
// It contains the implementation of the disassembler tables.
// Documentation for the disassembler emitter in general can be found in
// WebAssemblyDisassemblerEmitter.h.
//
//===----------------------------------------------------------------------===//
#include "WebAssemblyDisassemblerEmitter.h"
#include "CodeGenInstruction.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/TableGen/Record.h"
namespace llvm {
static constexpr int WebAssemblyInstructionTableSize = 256;
void emitWebAssemblyDisassemblerTables(
raw_ostream &OS,
const ArrayRef<const CodeGenInstruction *> &NumberedInstructions) {
// First lets organize all opcodes by (prefix) byte. Prefix 0 is the
// starting table.
std::map<unsigned,
std::map<unsigned, std::pair<unsigned, const CodeGenInstruction *>>>
OpcodeTable;
for (unsigned I = 0; I != NumberedInstructions.size(); ++I) {
auto &CGI = *NumberedInstructions[I];
auto &Def = *CGI.TheDef;
if (!Def.getValue("Inst"))
continue;
auto &Inst = *Def.getValueAsBitsInit("Inst");
RecordKeeper &RK = Inst.getRecordKeeper();
unsigned Opc = static_cast<unsigned>(
cast<IntInit>(Inst.convertInitializerTo(IntRecTy::get(RK)))
->getValue());
if (Opc == 0xFFFFFFFF)
continue; // No opcode defined.
assert(Opc <= 0xFFFFFF);
unsigned Prefix;
if (Opc <= 0xFFFF) {
Prefix = Opc >> 8;
Opc = Opc & 0xFF;
} else {
Prefix = Opc >> 16;
Opc = Opc & 0xFFFF;
}
auto &CGIP = OpcodeTable[Prefix][Opc];
// All wasm instructions have a StackBased field of type string, we only
// want the instructions for which this is "true".
bool IsStackBased = Def.getValueAsBit("StackBased");
if (!IsStackBased)
continue;
if (CGIP.second) {
// We already have an instruction for this slot, so decide which one
// should be the canonical one. This determines which variant gets
// printed in a disassembly. We want e.g. "call" not "i32.call", and
// "end" when we don't know if its "end_loop" or "end_block" etc.
bool IsCanonicalExisting = CGIP.second->TheDef->getValueAsBit("IsCanonical");
// We already have one marked explicitly as canonical, so keep it.
if (IsCanonicalExisting)
continue;
bool IsCanonicalNew = Def.getValueAsBit("IsCanonical");
// If the new one is explicitly marked as canonical, take it.
if (!IsCanonicalNew) {
// Neither the existing or new instruction is canonical.
// Pick the one with the shortest name as heuristic.
// Though ideally IsCanonical is always defined for at least one
// variant so this never has to apply.
if (CGIP.second->AsmString.size() <= CGI.AsmString.size())
continue;
}
}
// Set this instruction as the one to use.
CGIP = std::make_pair(I, &CGI);
}
OS << "#include \"MCTargetDesc/WebAssemblyMCTargetDesc.h\"\n";
OS << "\n";
OS << "namespace llvm {\n\n";
OS << "static constexpr int WebAssemblyInstructionTableSize = ";
OS << WebAssemblyInstructionTableSize << ";\n\n";
OS << "enum EntryType : uint8_t { ";
OS << "ET_Unused, ET_Prefix, ET_Instruction };\n\n";
OS << "struct WebAssemblyInstruction {\n";
OS << " uint16_t Opcode;\n";
OS << " EntryType ET;\n";
OS << " uint8_t NumOperands;\n";
OS << " uint16_t OperandStart;\n";
OS << "};\n\n";
std::vector<std::string> OperandTable, CurOperandList;
// Output one table per prefix.
for (auto &PrefixPair : OpcodeTable) {
if (PrefixPair.second.empty())
continue;
OS << "WebAssemblyInstruction InstructionTable" << PrefixPair.first;
OS << "[] = {\n";
for (unsigned I = 0; I < WebAssemblyInstructionTableSize; I++) {
auto InstIt = PrefixPair.second.find(I);
if (InstIt != PrefixPair.second.end()) {
// Regular instruction.
assert(InstIt->second.second);
auto &CGI = *InstIt->second.second;
OS << " // 0x";
OS.write_hex(static_cast<unsigned long long>(I));
OS << ": " << CGI.AsmString << "\n";
OS << " { " << InstIt->second.first << ", ET_Instruction, ";
OS << CGI.Operands.OperandList.size() << ", ";
// Collect operand types for storage in a shared list.
CurOperandList.clear();
for (auto &Op : CGI.Operands.OperandList) {
assert(Op.OperandType != "MCOI::OPERAND_UNKNOWN");
CurOperandList.push_back(Op.OperandType);
}
// See if we already have stored this sequence before. This is not
// strictly necessary but makes the table really small.
size_t OperandStart = OperandTable.size();
if (CurOperandList.size() <= OperandTable.size()) {
for (size_t J = 0; J <= OperandTable.size() - CurOperandList.size();
++J) {
size_t K = 0;
for (; K < CurOperandList.size(); ++K) {
if (OperandTable[J + K] != CurOperandList[K]) break;
}
if (K == CurOperandList.size()) {
OperandStart = J;
break;
}
}
}
// Store operands if no prior occurrence.
if (OperandStart == OperandTable.size()) {
llvm::append_range(OperandTable, CurOperandList);
}
OS << OperandStart;
} else {
auto PrefixIt = OpcodeTable.find(I);
// If we have a non-empty table for it that's not 0, this is a prefix.
if (PrefixIt != OpcodeTable.end() && I && !PrefixPair.first) {
OS << " { 0, ET_Prefix, 0, 0";
} else {
OS << " { 0, ET_Unused, 0, 0";
}
}
OS << " },\n";
}
OS << "};\n\n";
}
// Create a table of all operands:
OS << "const uint8_t OperandTable[] = {\n";
for (auto &Op : OperandTable) {
OS << " " << Op << ",\n";
}
OS << "};\n\n";
// Create a table of all extension tables:
OS << "struct { uint8_t Prefix; const WebAssemblyInstruction *Table; }\n";
OS << "PrefixTable[] = {\n";
for (auto &PrefixPair : OpcodeTable) {
if (PrefixPair.second.empty() || !PrefixPair.first)
continue;
OS << " { " << PrefixPair.first << ", InstructionTable"
<< PrefixPair.first;
OS << " },\n";
}
OS << " { 0, nullptr }\n};\n\n";
OS << "} // end namespace llvm\n";
}
} // namespace llvm
|