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
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
|
//===-- X86TargetMachine.cpp - Define TargetMachine for the X86 -----------===//
//
// The LLVM Compiler Infrastructure
//
// This file was developed by the LLVM research group and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the X86 specific subclass of TargetMachine.
//
//===----------------------------------------------------------------------===//
#include "X86TargetMachine.h"
#include "X86.h"
#include "llvm/Module.h"
#include "llvm/PassManager.h"
#include "llvm/CodeGen/IntrinsicLowering.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/Passes.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Target/TargetMachineRegistry.h"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/ADT/Statistic.h"
using namespace llvm;
X86VectorEnum llvm::X86Vector = NoSSE;
bool llvm::X86ScalarSSE = false;
/// X86TargetMachineModule - Note that this is used on hosts that cannot link
/// in a library unless there are references into the library. In particular,
/// it seems that it is not possible to get things to work on Win32 without
/// this. Though it is unused, do not remove it.
extern "C" int X86TargetMachineModule;
int X86TargetMachineModule = 0;
namespace {
cl::opt<bool> NoSSAPeephole("disable-ssa-peephole", cl::init(true),
cl::desc("Disable the ssa-based peephole optimizer "
"(defaults to disabled)"));
cl::opt<bool> DisableOutput("disable-x86-llc-output", cl::Hidden,
cl::desc("Disable the X86 asm printer, for use "
"when profiling the code generator."));
cl::opt<bool, true> EnableSSEFP("enable-sse-scalar-fp",
cl::desc("Perform FP math in SSE regs instead of the FP stack"),
cl::location(X86ScalarSSE),
cl::init(false));
// FIXME: This should eventually be handled with target triples and
// subtarget support!
cl::opt<X86VectorEnum, true>
SSEArg(
cl::desc("Enable SSE support in the X86 target:"),
cl::values(
clEnumValN(SSE, "sse", " Enable SSE support"),
clEnumValN(SSE2, "sse2", " Enable SSE and SSE2 support"),
clEnumValN(SSE3, "sse3", " Enable SSE, SSE2, and SSE3 support"),
clEnumValEnd),
cl::location(X86Vector), cl::init(NoSSE));
// Register the target.
RegisterTarget<X86TargetMachine> X("x86", " IA-32 (Pentium and above)");
}
unsigned X86TargetMachine::getJITMatchQuality() {
#if defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86)
return 10;
#else
return 0;
#endif
}
unsigned X86TargetMachine::getModuleMatchQuality(const Module &M) {
// We strongly match "i[3-9]86-*".
std::string TT = M.getTargetTriple();
if (TT.size() >= 5 && TT[0] == 'i' && TT[2] == '8' && TT[3] == '6' &&
TT[4] == '-' && TT[1] - '3' < 6)
return 20;
if (M.getEndianness() == Module::LittleEndian &&
M.getPointerSize() == Module::Pointer32)
return 10; // Weak match
else if (M.getEndianness() != Module::AnyEndianness ||
M.getPointerSize() != Module::AnyPointerSize)
return 0; // Match for some other target
return getJITMatchQuality()/2;
}
/// X86TargetMachine ctor - Create an ILP32 architecture model
///
X86TargetMachine::X86TargetMachine(const Module &M,
IntrinsicLowering *IL,
const std::string &FS)
: TargetMachine("X86", IL, true, 4, 4, 4, 4, 4),
Subtarget(M, FS),
FrameInfo(TargetFrameInfo::StackGrowsDown,
Subtarget.getStackAlignment(), -4),
JITInfo(*this) {
// Scalar SSE FP requires at least SSE2
X86ScalarSSE &= X86Vector >= SSE2;
}
// addPassesToEmitFile - We currently use all of the same passes as the JIT
// does to emit statically compiled machine code.
bool X86TargetMachine::addPassesToEmitFile(PassManager &PM, std::ostream &Out,
CodeGenFileType FileType) {
if (FileType != TargetMachine::AssemblyFile &&
FileType != TargetMachine::ObjectFile) return true;
// FIXME: Implement efficient support for garbage collection intrinsics.
PM.add(createLowerGCPass());
// FIXME: Implement the invoke/unwind instructions!
PM.add(createLowerInvokePass());
// FIXME: Implement the switch instruction in the instruction selector!
PM.add(createLowerSwitchPass());
// Make sure that no unreachable blocks are instruction selected.
PM.add(createUnreachableBlockEliminationPass());
// Install an instruction selector.
PM.add(createX86PatternInstructionSelector(*this));
// Run optional SSA-based machine code optimizations next...
if (!NoSSAPeephole)
PM.add(createX86SSAPeepholeOptimizerPass());
// Print the instruction selected machine code...
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(&std::cerr));
// Perform register allocation to convert to a concrete x86 representation
PM.add(createRegisterAllocator());
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(&std::cerr));
PM.add(createX86FloatingPointStackifierPass());
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(&std::cerr));
// Insert prolog/epilog code. Eliminate abstract frame index references...
PM.add(createPrologEpilogCodeInserter());
PM.add(createX86PeepholeOptimizerPass());
if (PrintMachineCode) // Print the register-allocated code
PM.add(createX86CodePrinterPass(std::cerr, *this));
if (!DisableOutput)
switch (FileType) {
default:
assert(0 && "Unexpected filetype here!");
case TargetMachine::AssemblyFile:
PM.add(createX86CodePrinterPass(Out, *this));
break;
case TargetMachine::ObjectFile:
// FIXME: We only support emission of ELF files for now, this should check
// the target triple and decide on the format to write (e.g. COFF on
// win32).
addX86ELFObjectWriterPass(PM, Out, *this);
break;
}
// Delete machine code for this function
PM.add(createMachineCodeDeleter());
return false; // success!
}
/// addPassesToJITCompile - Add passes to the specified pass manager to
/// implement a fast dynamic compiler for this target. Return true if this is
/// not supported for this target.
///
void X86JITInfo::addPassesToJITCompile(FunctionPassManager &PM) {
// FIXME: Implement efficient support for garbage collection intrinsics.
PM.add(createLowerGCPass());
// FIXME: Implement the invoke/unwind instructions!
PM.add(createLowerInvokePass());
// FIXME: Implement the switch instruction in the instruction selector!
PM.add(createLowerSwitchPass());
// Make sure that no unreachable blocks are instruction selected.
PM.add(createUnreachableBlockEliminationPass());
// Install an instruction selector.
PM.add(createX86PatternInstructionSelector(TM));
// Run optional SSA-based machine code optimizations next...
if (!NoSSAPeephole)
PM.add(createX86SSAPeepholeOptimizerPass());
// FIXME: Add SSA based peephole optimizer here.
// Print the instruction selected machine code...
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(&std::cerr));
// Perform register allocation to convert to a concrete x86 representation
PM.add(createRegisterAllocator());
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(&std::cerr));
PM.add(createX86FloatingPointStackifierPass());
if (PrintMachineCode)
PM.add(createMachineFunctionPrinterPass(&std::cerr));
// Insert prolog/epilog code. Eliminate abstract frame index references...
PM.add(createPrologEpilogCodeInserter());
PM.add(createX86PeepholeOptimizerPass());
if (PrintMachineCode) // Print the register-allocated code
PM.add(createX86CodePrinterPass(std::cerr, TM));
}
bool X86TargetMachine::addPassesToEmitMachineCode(FunctionPassManager &PM,
MachineCodeEmitter &MCE) {
PM.add(createX86CodeEmitterPass(MCE));
// Delete machine code for this function
PM.add(createMachineCodeDeleter());
return false;
}
|
