Add X86 BZHI instruction as well as BMI2 feature detection.
[oota-llvm.git] / lib / Target / X86 / X86.td
1 //===- X86.td - Target definition file for the Intel X86 ---*- tablegen -*-===//
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 // This is a target description file for the Intel i386 architecture, referred
11 // to here as the "X86" architecture.
12 //
13 //===----------------------------------------------------------------------===//
14
15 // Get the target-independent interfaces which we are implementing...
16 //
17 include "llvm/Target/Target.td"
18
19 //===----------------------------------------------------------------------===//
20 // X86 Subtarget state.
21 //
22
23 def Mode64Bit : SubtargetFeature<"64bit-mode", "In64BitMode", "true",
24                                   "64-bit mode (x86_64)">;
25
26 def ModeNaCl  : SubtargetFeature<"nacl-mode", "InNaClMode", "true",
27                                  "Native Client mode">;
28
29 //===----------------------------------------------------------------------===//
30 // X86 Subtarget features.
31 //===----------------------------------------------------------------------===//
32
33 def FeatureCMOV    : SubtargetFeature<"cmov","HasCMov", "true",
34                                       "Enable conditional move instructions">;
35
36 def FeaturePOPCNT   : SubtargetFeature<"popcnt", "HasPOPCNT", "true",
37                                        "Support POPCNT instruction">;
38
39
40 def FeatureMMX     : SubtargetFeature<"mmx","X86SSELevel", "MMX",
41                                       "Enable MMX instructions">;
42 def FeatureSSE1    : SubtargetFeature<"sse", "X86SSELevel", "SSE1",
43                                       "Enable SSE instructions",
44                                       // SSE codegen depends on cmovs, and all
45                                       // SSE1+ processors support them.
46                                       [FeatureMMX, FeatureCMOV]>;
47 def FeatureSSE2    : SubtargetFeature<"sse2", "X86SSELevel", "SSE2",
48                                       "Enable SSE2 instructions",
49                                       [FeatureSSE1]>;
50 def FeatureSSE3    : SubtargetFeature<"sse3", "X86SSELevel", "SSE3",
51                                       "Enable SSE3 instructions",
52                                       [FeatureSSE2]>;
53 def FeatureSSSE3   : SubtargetFeature<"ssse3", "X86SSELevel", "SSSE3",
54                                       "Enable SSSE3 instructions",
55                                       [FeatureSSE3]>;
56 def FeatureSSE41   : SubtargetFeature<"sse41", "X86SSELevel", "SSE41",
57                                       "Enable SSE 4.1 instructions",
58                                       [FeatureSSSE3]>;
59 def FeatureSSE42   : SubtargetFeature<"sse42", "X86SSELevel", "SSE42",
60                                       "Enable SSE 4.2 instructions",
61                                       [FeatureSSE41, FeaturePOPCNT]>;
62 def Feature3DNow   : SubtargetFeature<"3dnow", "X863DNowLevel", "ThreeDNow",
63                                       "Enable 3DNow! instructions",
64                                       [FeatureMMX]>;
65 def Feature3DNowA  : SubtargetFeature<"3dnowa", "X863DNowLevel", "ThreeDNowA",
66                                       "Enable 3DNow! Athlon instructions",
67                                       [Feature3DNow]>;
68 // All x86-64 hardware has SSE2, but we don't mark SSE2 as an implied
69 // feature, because SSE2 can be disabled (e.g. for compiling OS kernels)
70 // without disabling 64-bit mode.
71 def Feature64Bit   : SubtargetFeature<"64bit", "HasX86_64", "true",
72                                       "Support 64-bit instructions",
73                                       [FeatureCMOV]>;
74 def FeatureCMPXCHG16B : SubtargetFeature<"cmpxchg16b", "HasCmpxchg16b", "true",
75                                       "64-bit with cmpxchg16b",
76                                       [Feature64Bit]>;
77 def FeatureSlowBTMem : SubtargetFeature<"slow-bt-mem", "IsBTMemSlow", "true",
78                                        "Bit testing of memory is slow">;
79 def FeatureFastUAMem : SubtargetFeature<"fast-unaligned-mem",
80                                         "IsUAMemFast", "true",
81                                         "Fast unaligned memory access">;
82 def FeatureSSE4A   : SubtargetFeature<"sse4a", "HasSSE4A", "true",
83                                       "Support SSE 4a instructions",
84                                       [FeaturePOPCNT]>;
85
86 def FeatureAVX     : SubtargetFeature<"avx", "HasAVX", "true",
87                                       "Enable AVX instructions">;
88 def FeatureCLMUL   : SubtargetFeature<"clmul", "HasCLMUL", "true",
89                                "Enable carry-less multiplication instructions">;
90 def FeatureFMA3    : SubtargetFeature<"fma3", "HasFMA3", "true",
91                                      "Enable three-operand fused multiple-add">;
92 def FeatureFMA4    : SubtargetFeature<"fma4", "HasFMA4", "true",
93                                       "Enable four-operand fused multiple-add">;
94 def FeatureVectorUAMem : SubtargetFeature<"vector-unaligned-mem",
95                                           "HasVectorUAMem", "true",
96                  "Allow unaligned memory operands on vector/SIMD instructions">;
97 def FeatureAES     : SubtargetFeature<"aes", "HasAES", "true",
98                                       "Enable AES instructions">;
99 def FeatureMOVBE   : SubtargetFeature<"movbe", "HasMOVBE", "true",
100                                       "Support MOVBE instruction">;
101 def FeatureRDRAND  : SubtargetFeature<"rdrand", "HasRDRAND", "true",
102                                       "Support RDRAND instruction">;
103 def FeatureF16C    : SubtargetFeature<"f16c", "HasF16C", "true",
104                        "Support 16-bit floating point conversion instructions">;
105 def FeatureLZCNT   : SubtargetFeature<"lzcnt", "HasLZCNT", "true",
106                                       "Support LZCNT instruction">;
107 def FeatureBMI     : SubtargetFeature<"bmi", "HasBMI", "true",
108                                       "Support BMI instructions">;
109 def FeatureBMI2    : SubtargetFeature<"bmi2", "HasBMI2", "true",
110                                       "Support BMI2 instructions">;
111
112 //===----------------------------------------------------------------------===//
113 // X86 processors supported.
114 //===----------------------------------------------------------------------===//
115
116 class Proc<string Name, list<SubtargetFeature> Features>
117  : Processor<Name, NoItineraries, Features>;
118
119 def : Proc<"generic",         []>;
120 def : Proc<"i386",            []>;
121 def : Proc<"i486",            []>;
122 def : Proc<"i586",            []>;
123 def : Proc<"pentium",         []>;
124 def : Proc<"pentium-mmx",     [FeatureMMX]>;
125 def : Proc<"i686",            []>;
126 def : Proc<"pentiumpro",      [FeatureCMOV]>;
127 def : Proc<"pentium2",        [FeatureMMX, FeatureCMOV]>;
128 def : Proc<"pentium3",        [FeatureSSE1]>;
129 def : Proc<"pentium3m",       [FeatureSSE1, FeatureSlowBTMem]>;
130 def : Proc<"pentium-m",       [FeatureSSE2, FeatureSlowBTMem]>;
131 def : Proc<"pentium4",        [FeatureSSE2]>;
132 def : Proc<"pentium4m",       [FeatureSSE2, FeatureSlowBTMem]>;
133 def : Proc<"x86-64",          [FeatureSSE2, Feature64Bit, FeatureSlowBTMem]>;
134 def : Proc<"yonah",           [FeatureSSE3, FeatureSlowBTMem]>;
135 def : Proc<"prescott",        [FeatureSSE3, FeatureSlowBTMem]>;
136 def : Proc<"nocona",          [FeatureSSE3, FeatureCMPXCHG16B,
137                                FeatureSlowBTMem]>;
138 def : Proc<"core2",           [FeatureSSSE3, FeatureCMPXCHG16B,
139                                FeatureSlowBTMem]>;
140 def : Proc<"penryn",          [FeatureSSE41, FeatureCMPXCHG16B,
141                                FeatureSlowBTMem]>;
142 def : Proc<"atom",            [FeatureSSE3, FeatureCMPXCHG16B, FeatureMOVBE,
143                                FeatureSlowBTMem]>;
144 // "Arrandale" along with corei3 and corei5
145 def : Proc<"corei7",          [FeatureSSE42, FeatureCMPXCHG16B,
146                                FeatureSlowBTMem, FeatureFastUAMem, FeatureAES]>;
147 def : Proc<"nehalem",         [FeatureSSE42,  FeatureCMPXCHG16B,
148                                FeatureSlowBTMem, FeatureFastUAMem]>;
149 // Westmere is a similar machine to nehalem with some additional features.
150 // Westmere is the corei3/i5/i7 path from nehalem to sandybridge
151 def : Proc<"westmere",        [FeatureSSE42, FeatureCMPXCHG16B,
152                                FeatureSlowBTMem, FeatureFastUAMem, FeatureAES,
153                                FeatureCLMUL]>;
154 // Sandy Bridge
155 // SSE is not listed here since llvm treats AVX as a reimplementation of SSE,
156 // rather than a superset.
157 // FIXME: Disabling AVX for now since it's not ready.
158 def : Proc<"corei7-avx",      [FeatureSSE42, FeatureCMPXCHG16B,
159                                FeatureAES, FeatureCLMUL]>;
160 // Ivy Bridge
161 def : Proc<"core-avx-i",      [FeatureSSE42, FeatureCMPXCHG16B,
162                                FeatureAES, FeatureCLMUL,
163                                FeatureRDRAND, FeatureF16C]>;
164
165 // Haswell
166 def : Proc<"core-avx2",       [FeatureSSE42, FeatureCMPXCHG16B, FeatureAES,
167                                FeatureCLMUL, FeatureRDRAND, FeatureF16C,
168                                FeatureFMA3, FeatureMOVBE, FeatureLZCNT,
169                                FeatureBMI, FeatureBMI2]>;
170
171 def : Proc<"k6",              [FeatureMMX]>;
172 def : Proc<"k6-2",            [Feature3DNow]>;
173 def : Proc<"k6-3",            [Feature3DNow]>;
174 def : Proc<"athlon",          [Feature3DNowA, FeatureSlowBTMem]>;
175 def : Proc<"athlon-tbird",    [Feature3DNowA, FeatureSlowBTMem]>;
176 def : Proc<"athlon-4",        [FeatureSSE1,   Feature3DNowA, FeatureSlowBTMem]>;
177 def : Proc<"athlon-xp",       [FeatureSSE1,   Feature3DNowA, FeatureSlowBTMem]>;
178 def : Proc<"athlon-mp",       [FeatureSSE1,   Feature3DNowA, FeatureSlowBTMem]>;
179 def : Proc<"k8",              [FeatureSSE2,   Feature3DNowA, Feature64Bit,
180                                FeatureSlowBTMem]>;
181 def : Proc<"opteron",         [FeatureSSE2,   Feature3DNowA, Feature64Bit,
182                                FeatureSlowBTMem]>;
183 def : Proc<"athlon64",        [FeatureSSE2,   Feature3DNowA, Feature64Bit,
184                                FeatureSlowBTMem]>;
185 def : Proc<"athlon-fx",       [FeatureSSE2,   Feature3DNowA, Feature64Bit,
186                                FeatureSlowBTMem]>;
187 def : Proc<"k8-sse3",         [FeatureSSE3,   Feature3DNowA, FeatureCMPXCHG16B,
188                                FeatureSlowBTMem]>;
189 def : Proc<"opteron-sse3",    [FeatureSSE3,   Feature3DNowA, FeatureCMPXCHG16B,
190                                FeatureSlowBTMem]>;
191 def : Proc<"athlon64-sse3",   [FeatureSSE3,   Feature3DNowA, FeatureCMPXCHG16B,
192                                FeatureSlowBTMem]>;
193 def : Proc<"amdfam10",        [FeatureSSE3,   FeatureSSE4A,
194                                Feature3DNowA, FeatureCMPXCHG16B,
195                                FeatureSlowBTMem]>;
196 def : Proc<"barcelona",       [FeatureSSE3,   FeatureSSE4A,
197                                Feature3DNowA, FeatureCMPXCHG16B,
198                                FeatureSlowBTMem]>;
199 def : Proc<"istanbul",        [Feature3DNowA, FeatureCMPXCHG16B,
200                                FeatureSSE4A, Feature3DNowA]>;
201 def : Proc<"shanghai",        [Feature3DNowA, FeatureCMPXCHG16B, FeatureSSE4A,
202                                Feature3DNowA]>;
203
204 def : Proc<"winchip-c6",      [FeatureMMX]>;
205 def : Proc<"winchip2",        [Feature3DNow]>;
206 def : Proc<"c3",              [Feature3DNow]>;
207 def : Proc<"c3-2",            [FeatureSSE1]>;
208
209 //===----------------------------------------------------------------------===//
210 // Register File Description
211 //===----------------------------------------------------------------------===//
212
213 include "X86RegisterInfo.td"
214
215 //===----------------------------------------------------------------------===//
216 // Instruction Descriptions
217 //===----------------------------------------------------------------------===//
218
219 include "X86InstrInfo.td"
220
221 def X86InstrInfo : InstrInfo;
222
223 //===----------------------------------------------------------------------===//
224 // Calling Conventions
225 //===----------------------------------------------------------------------===//
226
227 include "X86CallingConv.td"
228
229
230 //===----------------------------------------------------------------------===//
231 // Assembly Parser
232 //===----------------------------------------------------------------------===//
233
234 // Currently the X86 assembly parser only supports ATT syntax.
235 def ATTAsmParser : AsmParser {
236   string AsmParserClassName = "ATTAsmParser";
237   int Variant = 0;
238
239   // Discard comments in assembly strings.
240   string CommentDelimiter = "#";
241
242   // Recognize hard coded registers.
243   string RegisterPrefix = "%";
244 }
245
246 //===----------------------------------------------------------------------===//
247 // Assembly Printers
248 //===----------------------------------------------------------------------===//
249
250 // The X86 target supports two different syntaxes for emitting machine code.
251 // This is controlled by the -x86-asm-syntax={att|intel}
252 def ATTAsmWriter : AsmWriter {
253   string AsmWriterClassName  = "ATTInstPrinter";
254   int Variant = 0;
255   bit isMCAsmWriter = 1;
256 }
257 def IntelAsmWriter : AsmWriter {
258   string AsmWriterClassName  = "IntelInstPrinter";
259   int Variant = 1;
260   bit isMCAsmWriter = 1;
261 }
262
263 def X86 : Target {
264   // Information about the instructions...
265   let InstructionSet = X86InstrInfo;
266
267   let AssemblyParsers = [ATTAsmParser];
268
269   let AssemblyWriters = [ATTAsmWriter, IntelAsmWriter];
270 }