1 //===-- X86Subtarget.cpp - X86 Subtarget Information ------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by Nate Begeman and is distributed under the
6 // University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the X86 specific subclass of TargetSubtarget.
12 //===----------------------------------------------------------------------===//
14 #include "X86Subtarget.h"
15 #include "X86GenSubtarget.inc"
16 #include "llvm/Module.h"
17 #include "llvm/Support/CommandLine.h"
18 #include "llvm/Target/TargetMachine.h"
21 cl::opt
<X86Subtarget::AsmWriterFlavorTy
>
22 AsmWriterFlavor("x86-asm-syntax", cl::init(X86Subtarget::Unset
),
23 cl::desc("Choose style of code to emit from X86 backend:"),
25 clEnumValN(X86Subtarget::ATT
, "att", " Emit AT&T-style assembly"),
26 clEnumValN(X86Subtarget::Intel
, "intel", " Emit Intel-style assembly"),
30 /// True if accessing the GV requires an extra load. For Windows, dllimported
31 /// symbols are indirect, loading the value at address GV rather then the
32 /// value of GV itself. This means that the GlobalAddress must be in the base
33 /// or index register of the address, not the GV offset field.
34 bool X86Subtarget::GVRequiresExtraLoad(const GlobalValue
* GV
,
35 const TargetMachine
& TM
,
36 bool isDirectCall
) const
39 if (TM
.getRelocationModel() != Reloc::Static
)
40 if (isTargetDarwin()) {
41 return (!isDirectCall
&&
42 (GV
->hasWeakLinkage() || GV
->hasLinkOnceLinkage() ||
43 (GV
->isDeclaration() && !GV
->hasNotBeenReadFromBitcode())));
44 } else if (TM
.getRelocationModel() == Reloc::PIC_
&& isPICStyleGOT()) {
45 // Extra load is needed for all non-statics.
46 return (!isDirectCall
&&
47 (GV
->isDeclaration() || !GV
->hasInternalLinkage()));
48 } else if (isTargetCygMing() || isTargetWindows()) {
49 return (GV
->hasDLLImportLinkage());
55 /// GetCpuIDAndInfo - Execute the specified cpuid and return the 4 values in the
56 /// specified arguments. If we can't run cpuid on the host, return true.
57 bool X86::GetCpuIDAndInfo(unsigned value
, unsigned *rEAX
, unsigned *rEBX
,
58 unsigned *rECX
, unsigned *rEDX
) {
59 #if defined(__x86_64__)
60 // gcc doesn't know cpuid would clobber ebx/rbx. Preseve it manually.
61 asm ("movq\t%%rbx, %%rsi\n\t"
63 "xchgq\t%%rbx, %%rsi\n\t"
70 #elif defined(i386) || defined(__i386__) || defined(__x86__) || defined(_M_IX86)
72 asm ("movl\t%%ebx, %%esi\n\t"
74 "xchgl\t%%ebx, %%esi\n\t"
81 #elif defined(_MSC_VER)
86 mov dword ptr
[esi
],eax
88 mov dword ptr
[esi
],ebx
90 mov dword ptr
[esi
],ecx
92 mov dword ptr
[esi
],edx
100 void X86Subtarget::AutoDetectSubtargetFeatures() {
101 unsigned EAX
= 0, EBX
= 0, ECX
= 0, EDX
= 0;
107 if (X86::GetCpuIDAndInfo(0, &EAX
, text
.u
+0, text
.u
+2, text
.u
+1))
110 X86::GetCpuIDAndInfo(0x1, &EAX
, &EBX
, &ECX
, &EDX
);
112 if ((EDX
>> 23) & 0x1) X86SSELevel
= MMX
;
113 if ((EDX
>> 25) & 0x1) X86SSELevel
= SSE1
;
114 if ((EDX
>> 26) & 0x1) X86SSELevel
= SSE2
;
115 if (ECX
& 0x1) X86SSELevel
= SSE3
;
116 if ((ECX
>> 9) & 0x1) X86SSELevel
= SSSE3
;
118 if (memcmp(text
.c
, "GenuineIntel", 12) == 0 ||
119 memcmp(text
.c
, "AuthenticAMD", 12) == 0) {
120 X86::GetCpuIDAndInfo(0x80000001, &EAX
, &EBX
, &ECX
, &EDX
);
121 HasX86_64
= (EDX
>> 29) & 0x1;
125 static const char *GetCurrentX86CPU() {
126 unsigned EAX
= 0, EBX
= 0, ECX
= 0, EDX
= 0;
127 if (X86::GetCpuIDAndInfo(0x1, &EAX
, &EBX
, &ECX
, &EDX
))
129 unsigned Family
= (EAX
>> 8) & 0xf; // Bits 8 - 11
130 unsigned Model
= (EAX
>> 4) & 0xf; // Bits 4 - 7
131 X86::GetCpuIDAndInfo(0x80000001, &EAX
, &EBX
, &ECX
, &EDX
);
132 bool Em64T
= (EDX
>> 29) & 0x1;
139 X86::GetCpuIDAndInfo(0, &EAX
, text
.u
+0, text
.u
+2, text
.u
+1);
140 if (memcmp(text
.c
, "GenuineIntel", 12) == 0) {
148 case 4: return "pentium-mmx";
149 default: return "pentium";
153 case 1: return "pentiumpro";
156 case 6: return "pentium2";
160 case 11: return "pentium3";
162 case 13: return "pentium-m";
163 case 14: return "yonah";
164 case 15: return "core2";
165 default: return "i686";
171 return (Em64T
) ? "nocona" : "prescott";
173 return (Em64T
) ? "x86-64" : "pentium4";
180 } else if (memcmp(text
.c
, "AuthenticAMD", 12) == 0) {
181 // FIXME: this poorly matches the generated SubtargetFeatureKV table. There
182 // appears to be no way to generate the wide variety of AMD-specific targets
183 // from the information returned from CPUID.
191 case 8: return "k6-2";
193 case 13: return "k6-3";
194 default: return "pentium";
198 case 4: return "athlon-tbird";
201 case 8: return "athlon-mp";
202 case 10: return "athlon-xp";
203 default: return "athlon";
207 case 1: return "opteron";
208 case 5: return "athlon-fx"; // also opteron
209 default: return "athlon64";
219 X86Subtarget::X86Subtarget(const Module
&M
, const std::string
&FS
, bool is64Bit
)
220 : AsmFlavor(AsmWriterFlavor
)
221 , PICStyle(PICStyle::None
)
222 , X86SSELevel(NoMMXSSE
)
225 // FIXME: this is a known good value for Yonah. How about others?
226 , MinRepStrSizeThreshold(128)
228 , HasLow4GUserAddress(true)
229 , TargetType(isELF
) { // Default to ELF unless otherwise specified.
231 // Determine default and user specified characteristics
233 // If feature string is not empty, parse features string.
234 std::string CPU
= GetCurrentX86CPU();
235 ParseSubtargetFeatures(FS
, CPU
);
237 if (Is64Bit
&& !HasX86_64
)
238 cerr
<< "Warning: Generation of 64-bit code for a 32-bit processor "
240 if (Is64Bit
&& X86SSELevel
< SSE2
)
241 cerr
<< "Warning: 64-bit processors all have at least SSE2.\n";
243 // Otherwise, use CPUID to auto-detect feature set.
244 AutoDetectSubtargetFeatures();
247 // If requesting codegen for X86-64, make sure that 64-bit and SSE2 features
248 // are enabled. These are available on all x86-64 CPUs.
251 if (X86SSELevel
< SSE2
)
255 // Set the boolean corresponding to the current target triple, or the default
256 // if one cannot be determined, to true.
257 const std::string
& TT
= M
.getTargetTriple();
258 if (TT
.length() > 5) {
259 if (TT
.find("cygwin") != std::string::npos
)
260 TargetType
= isCygwin
;
261 else if (TT
.find("mingw") != std::string::npos
)
262 TargetType
= isMingw
;
263 else if (TT
.find("darwin") != std::string::npos
)
264 TargetType
= isDarwin
;
265 else if (TT
.find("win32") != std::string::npos
)
266 TargetType
= isWindows
;
267 } else if (TT
.empty()) {
268 #if defined(__CYGWIN__)
269 TargetType
= isCygwin
;
270 #elif defined(__MINGW32__)
271 TargetType
= isMingw
;
272 #elif defined(__APPLE__)
273 TargetType
= isDarwin
;
274 #elif defined(_WIN32)
275 TargetType
= isWindows
;
279 // If the asm syntax hasn't been overridden on the command line, use whatever
281 if (AsmFlavor
== X86Subtarget::Unset
) {
282 if (TargetType
== isWindows
) {
283 AsmFlavor
= X86Subtarget::Intel
;
285 AsmFlavor
= X86Subtarget::ATT
;
289 if (TargetType
== isDarwin
&& Is64Bit
)
290 HasLow4GUserAddress
= false;
292 if (TargetType
== isDarwin
||
293 TargetType
== isCygwin
||
294 TargetType
== isMingw
||
295 (TargetType
== isELF
&& Is64Bit
))