1 //===----------X86InstrFragments - X86 Pattern fragments. --*- tablegen -*-===//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 // X86-specific DAG node.
10 def SDTX86CmpTest : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
12 def SDTX86FCmp : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisFP<1>,
15 def SDTX86Ccmp : SDTypeProfile<1, 5,
16 [SDTCisVT<3, i8>, SDTCisVT<4, i8>, SDTCisVT<5, i32>]>;
18 // RES = op PTR, PASSTHRU, COND, EFLAGS
19 def SDTX86Cload : SDTypeProfile<1, 4,
20 [SDTCisInt<0>, SDTCisPtrTy<1>, SDTCisSameAs<0, 2>,
21 SDTCisVT<3, i8>, SDTCisVT<4, i32>]>;
22 // op VAL, PTR, COND, EFLAGS
23 def SDTX86Cstore : SDTypeProfile<0, 4,
24 [SDTCisInt<0>, SDTCisPtrTy<1>,
25 SDTCisVT<2, i8>, SDTCisVT<3, i32>]>;
27 def SDTX86Cmov : SDTypeProfile<1, 4,
28 [SDTCisSameAs<0, 1>, SDTCisSameAs<1, 2>,
29 SDTCisVT<3, i8>, SDTCisVT<4, i32>]>;
31 // Unary and binary operator instructions that set EFLAGS as a side-effect.
32 def SDTUnaryArithWithFlags : SDTypeProfile<2, 1,
34 SDTCisInt<0>, SDTCisVT<1, i32>]>;
36 def SDTBinaryArithWithFlags : SDTypeProfile<2, 2,
39 SDTCisInt<0>, SDTCisVT<1, i32>]>;
41 // SDTBinaryArithWithFlagsInOut - RES1, EFLAGS = op LHS, RHS, EFLAGS
42 def SDTBinaryArithWithFlagsInOut : SDTypeProfile<2, 3,
49 def SDTX86BrCond : SDTypeProfile<0, 3,
50 [SDTCisVT<0, OtherVT>,
51 SDTCisVT<1, i8>, SDTCisVT<2, i32>]>;
53 def SDTX86SetCC : SDTypeProfile<1, 2,
55 SDTCisVT<1, i8>, SDTCisVT<2, i32>]>;
56 def SDTX86SetCC_C : SDTypeProfile<1, 2,
58 SDTCisVT<1, i8>, SDTCisVT<2, i32>]>;
60 def SDTX86sahf : SDTypeProfile<1, 1, [SDTCisVT<0, i32>, SDTCisVT<1, i8>]>;
62 def SDTX86rdrand : SDTypeProfile<2, 0, [SDTCisInt<0>, SDTCisVT<1, i32>]>;
64 def SDTX86rdpkru : SDTypeProfile<1, 1, [SDTCisVT<0, i32>, SDTCisVT<1, i32>]>;
65 def SDTX86wrpkru : SDTypeProfile<0, 3, [SDTCisVT<0, i32>, SDTCisVT<1, i32>,
68 def SDTX86cas : SDTypeProfile<0, 3, [SDTCisPtrTy<0>, SDTCisInt<1>,
70 def SDTX86cas8pair : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
71 def SDTX86cas16pair : SDTypeProfile<0, 2, [SDTCisPtrTy<0>, SDTCisVT<1, i64>]>;
73 def SDTLockBinaryArithWithFlags : SDTypeProfile<1, 2, [SDTCisVT<0, i32>,
77 def SDTLockUnaryArithWithFlags : SDTypeProfile<1, 1, [SDTCisVT<0, i32>,
80 def SDTX86Ret : SDTypeProfile<0, -1, [SDTCisVT<0, i32>]>;
82 def SDT_X86CallSeqStart : SDCallSeqStart<[SDTCisVT<0, i32>,
84 def SDT_X86CallSeqEnd : SDCallSeqEnd<[SDTCisVT<0, i32>,
87 def SDT_X86Call : SDTypeProfile<0, -1, [SDTCisVT<0, iPTR>]>;
89 def SDT_X86NtBrind : SDTypeProfile<0, -1, [SDTCisVT<0, iPTR>]>;
91 def SDT_X86VASTART_SAVE_XMM_REGS : SDTypeProfile<0, -1, [SDTCisVT<0, i8>,
94 def SDT_X86VAARG : SDTypeProfile<1, -1, [SDTCisPtrTy<0>,
100 def SDTX86RepStr : SDTypeProfile<0, 1, [SDTCisVT<0, OtherVT>]>;
102 def SDTX86Void : SDTypeProfile<0, 0, []>;
104 def SDTX86Wrapper : SDTypeProfile<1, 1, [SDTCisSameAs<0, 1>, SDTCisPtrTy<0>]>;
106 def SDT_X86TLSADDR : SDTypeProfile<0, 1, [SDTCisInt<0>]>;
108 def SDT_X86TLSBASEADDR : SDTypeProfile<0, 1, [SDTCisInt<0>]>;
110 def SDT_X86TLSCALL : SDTypeProfile<0, 1, [SDTCisInt<0>]>;
112 def SDT_X86DYN_ALLOCA : SDTypeProfile<0, 1, [SDTCisVT<0, iPTR>]>;
114 def SDT_X86SEG_ALLOCA : SDTypeProfile<1, 1, [SDTCisVT<0, iPTR>, SDTCisVT<1, iPTR>]>;
116 def SDT_X86PROBED_ALLOCA : SDTypeProfile<1, 1, [SDTCisVT<0, iPTR>, SDTCisVT<1, iPTR>]>;
118 def SDT_X86EHRET : SDTypeProfile<0, 1, [SDTCisInt<0>]>;
120 def SDT_X86TCRET : SDTypeProfile<0, 2, [SDTCisPtrTy<0>, SDTCisVT<1, i32>]>;
122 def SDT_X86ENQCMD : SDTypeProfile<1, 2, [SDTCisVT<0, i32>,
123 SDTCisPtrTy<1>, SDTCisSameAs<1, 2>]>;
125 def SDT_X86AESENCDECKL : SDTypeProfile<2, 2, [SDTCisVT<0, v2i64>,
130 def SDTX86Cmpccxadd : SDTypeProfile<1, 4, [SDTCisSameAs<0, 2>,
131 SDTCisPtrTy<1>, SDTCisSameAs<2, 3>,
134 def X86MFence : SDNode<"X86ISD::MFENCE", SDTNone, [SDNPHasChain]>;
137 def X86bsf : SDNode<"X86ISD::BSF", SDTUnaryArithWithFlags>;
138 def X86bsr : SDNode<"X86ISD::BSR", SDTUnaryArithWithFlags>;
139 def X86fshl : SDNode<"X86ISD::FSHL", SDTIntShiftDOp>;
140 def X86fshr : SDNode<"X86ISD::FSHR", SDTIntShiftDOp>;
142 def X86cmp : SDNode<"X86ISD::CMP" , SDTX86CmpTest>;
143 def X86fcmp : SDNode<"X86ISD::FCMP", SDTX86FCmp>;
144 def X86strict_fcmp : SDNode<"X86ISD::STRICT_FCMP", SDTX86FCmp, [SDNPHasChain]>;
145 def X86strict_fcmps : SDNode<"X86ISD::STRICT_FCMPS", SDTX86FCmp, [SDNPHasChain]>;
146 def X86bt : SDNode<"X86ISD::BT", SDTX86CmpTest>;
148 def X86ccmp : SDNode<"X86ISD::CCMP", SDTX86Ccmp>;
149 def X86ctest : SDNode<"X86ISD::CTEST", SDTX86Ccmp>;
151 def X86cload : SDNode<"X86ISD::CLOAD", SDTX86Cload, [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
152 def X86cstore : SDNode<"X86ISD::CSTORE", SDTX86Cstore, [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
154 def X86cmov : SDNode<"X86ISD::CMOV", SDTX86Cmov>;
155 def X86brcond : SDNode<"X86ISD::BRCOND", SDTX86BrCond,
157 def X86setcc : SDNode<"X86ISD::SETCC", SDTX86SetCC>;
158 def X86setcc_c : SDNode<"X86ISD::SETCC_CARRY", SDTX86SetCC_C>;
160 def X86rdrand : SDNode<"X86ISD::RDRAND", SDTX86rdrand,
161 [SDNPHasChain, SDNPSideEffect]>;
163 def X86rdseed : SDNode<"X86ISD::RDSEED", SDTX86rdrand,
164 [SDNPHasChain, SDNPSideEffect]>;
166 def X86rdpkru : SDNode<"X86ISD::RDPKRU", SDTX86rdpkru,
167 [SDNPHasChain, SDNPSideEffect]>;
168 def X86wrpkru : SDNode<"X86ISD::WRPKRU", SDTX86wrpkru,
169 [SDNPHasChain, SDNPSideEffect]>;
171 def X86cas : SDNode<"X86ISD::LCMPXCHG_DAG", SDTX86cas,
172 [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore,
173 SDNPMayLoad, SDNPMemOperand]>;
174 def X86cas8 : SDNode<"X86ISD::LCMPXCHG8_DAG", SDTX86cas8pair,
175 [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore,
176 SDNPMayLoad, SDNPMemOperand]>;
177 def X86cas16 : SDNode<"X86ISD::LCMPXCHG16_DAG", SDTX86cas16pair,
178 [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore,
179 SDNPMayLoad, SDNPMemOperand]>;
181 def X86retglue : SDNode<"X86ISD::RET_GLUE", SDTX86Ret,
182 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
183 def X86iret : SDNode<"X86ISD::IRET", SDTX86Ret,
184 [SDNPHasChain, SDNPOptInGlue]>;
186 def X86vastart_save_xmm_regs :
187 SDNode<"X86ISD::VASTART_SAVE_XMM_REGS",
188 SDT_X86VASTART_SAVE_XMM_REGS,
189 [SDNPHasChain, SDNPMayStore, SDNPMemOperand, SDNPVariadic]>;
191 SDNode<"X86ISD::VAARG_64", SDT_X86VAARG,
192 [SDNPHasChain, SDNPMayLoad, SDNPMayStore,
195 SDNode<"X86ISD::VAARG_X32", SDT_X86VAARG,
196 [SDNPHasChain, SDNPMayLoad, SDNPMayStore,
198 def X86callseq_start :
199 SDNode<"ISD::CALLSEQ_START", SDT_X86CallSeqStart,
200 [SDNPHasChain, SDNPOutGlue]>;
202 SDNode<"ISD::CALLSEQ_END", SDT_X86CallSeqEnd,
203 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
205 def X86call : SDNode<"X86ISD::CALL", SDT_X86Call,
206 [SDNPHasChain, SDNPOutGlue, SDNPOptInGlue,
209 def X86call_rvmarker : SDNode<"X86ISD::CALL_RVMARKER", SDT_X86Call,
210 [SDNPHasChain, SDNPOutGlue, SDNPOptInGlue,
214 def X86NoTrackCall : SDNode<"X86ISD::NT_CALL", SDT_X86Call,
215 [SDNPHasChain, SDNPOutGlue, SDNPOptInGlue,
217 def X86NoTrackBrind : SDNode<"X86ISD::NT_BRIND", SDT_X86NtBrind,
220 def X86rep_stos: SDNode<"X86ISD::REP_STOS", SDTX86RepStr,
221 [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore]>;
222 def X86rep_movs: SDNode<"X86ISD::REP_MOVS", SDTX86RepStr,
223 [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore,
226 def X86Wrapper : SDNode<"X86ISD::Wrapper", SDTX86Wrapper>;
227 def X86WrapperRIP : SDNode<"X86ISD::WrapperRIP", SDTX86Wrapper>;
229 def X86RecoverFrameAlloc : SDNode<"ISD::LOCAL_RECOVER",
230 SDTypeProfile<1, 1, [SDTCisSameAs<0, 1>,
233 def X86tlsaddr : SDNode<"X86ISD::TLSADDR", SDT_X86TLSADDR,
234 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
236 def X86tlsbaseaddr : SDNode<"X86ISD::TLSBASEADDR", SDT_X86TLSBASEADDR,
237 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
239 def X86tlsdesc : SDNode<"X86ISD::TLSDESC", SDT_X86TLSADDR,
240 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
242 def X86ehret : SDNode<"X86ISD::EH_RETURN", SDT_X86EHRET,
245 def X86eh_sjlj_setjmp : SDNode<"X86ISD::EH_SJLJ_SETJMP",
246 SDTypeProfile<1, 1, [SDTCisInt<0>,
248 [SDNPHasChain, SDNPSideEffect]>;
249 def X86eh_sjlj_longjmp : SDNode<"X86ISD::EH_SJLJ_LONGJMP",
250 SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>,
251 [SDNPHasChain, SDNPSideEffect]>;
252 def X86eh_sjlj_setup_dispatch : SDNode<"X86ISD::EH_SJLJ_SETUP_DISPATCH",
253 SDTypeProfile<0, 0, []>,
254 [SDNPHasChain, SDNPSideEffect]>;
256 def X86tcret : SDNode<"X86ISD::TC_RETURN", SDT_X86TCRET,
257 [SDNPHasChain, SDNPOptInGlue, SDNPVariadic]>;
259 def X86add_flag : SDNode<"X86ISD::ADD", SDTBinaryArithWithFlags,
261 def X86sub_flag : SDNode<"X86ISD::SUB", SDTBinaryArithWithFlags>;
262 def X86smul_flag : SDNode<"X86ISD::SMUL", SDTBinaryArithWithFlags,
264 def X86umul_flag : SDNode<"X86ISD::UMUL", SDTBinaryArithWithFlags,
266 def X86adc_flag : SDNode<"X86ISD::ADC", SDTBinaryArithWithFlagsInOut>;
267 def X86sbb_flag : SDNode<"X86ISD::SBB", SDTBinaryArithWithFlagsInOut>;
269 def X86or_flag : SDNode<"X86ISD::OR", SDTBinaryArithWithFlags,
271 def X86xor_flag : SDNode<"X86ISD::XOR", SDTBinaryArithWithFlags,
273 def X86and_flag : SDNode<"X86ISD::AND", SDTBinaryArithWithFlags,
276 def X86lock_add : SDNode<"X86ISD::LADD", SDTLockBinaryArithWithFlags,
277 [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
279 def X86lock_sub : SDNode<"X86ISD::LSUB", SDTLockBinaryArithWithFlags,
280 [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
282 def X86lock_or : SDNode<"X86ISD::LOR", SDTLockBinaryArithWithFlags,
283 [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
285 def X86lock_xor : SDNode<"X86ISD::LXOR", SDTLockBinaryArithWithFlags,
286 [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
288 def X86lock_and : SDNode<"X86ISD::LAND", SDTLockBinaryArithWithFlags,
289 [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
292 def X86bextr : SDNode<"X86ISD::BEXTR", SDTIntBinOp>;
293 def X86bextri : SDNode<"X86ISD::BEXTRI", SDTIntBinOp>;
295 def X86bzhi : SDNode<"X86ISD::BZHI", SDTIntBinOp>;
297 def X86pdep : SDNode<"X86ISD::PDEP", SDTIntBinOp>;
298 def X86pext : SDNode<"X86ISD::PEXT", SDTIntBinOp>;
300 def X86mul_imm : SDNode<"X86ISD::MUL_IMM", SDTIntBinOp>;
302 def X86DynAlloca : SDNode<"X86ISD::DYN_ALLOCA", SDT_X86DYN_ALLOCA,
303 [SDNPHasChain, SDNPOutGlue]>;
305 def X86SegAlloca : SDNode<"X86ISD::SEG_ALLOCA", SDT_X86SEG_ALLOCA,
308 def X86ProbedAlloca : SDNode<"X86ISD::PROBED_ALLOCA", SDT_X86PROBED_ALLOCA,
311 def X86TLSCall : SDNode<"X86ISD::TLSCALL", SDT_X86TLSCALL,
312 [SDNPHasChain, SDNPOptInGlue, SDNPOutGlue]>;
314 def X86lwpins : SDNode<"X86ISD::LWPINS",
315 SDTypeProfile<1, 3, [SDTCisVT<0, i32>, SDTCisInt<1>,
316 SDTCisVT<2, i32>, SDTCisVT<3, i32>]>,
317 [SDNPHasChain, SDNPMayStore, SDNPMayLoad, SDNPSideEffect]>;
319 def X86umwait : SDNode<"X86ISD::UMWAIT",
320 SDTypeProfile<1, 3, [SDTCisVT<0, i32>, SDTCisInt<1>,
321 SDTCisVT<2, i32>, SDTCisVT<3, i32>]>,
322 [SDNPHasChain, SDNPSideEffect]>;
324 def X86tpause : SDNode<"X86ISD::TPAUSE",
325 SDTypeProfile<1, 3, [SDTCisVT<0, i32>, SDTCisInt<1>,
326 SDTCisVT<2, i32>, SDTCisVT<3, i32>]>,
327 [SDNPHasChain, SDNPSideEffect]>;
329 def X86enqcmd : SDNode<"X86ISD::ENQCMD", SDT_X86ENQCMD,
330 [SDNPHasChain, SDNPSideEffect]>;
331 def X86enqcmds : SDNode<"X86ISD::ENQCMDS", SDT_X86ENQCMD,
332 [SDNPHasChain, SDNPSideEffect]>;
333 def X86testui : SDNode<"X86ISD::TESTUI",
334 SDTypeProfile<1, 0, [SDTCisVT<0, i32>]>,
335 [SDNPHasChain, SDNPSideEffect]>;
337 def X86aesenc128kl : SDNode<"X86ISD::AESENC128KL", SDT_X86AESENCDECKL,
338 [SDNPHasChain, SDNPMayLoad, SDNPSideEffect,
340 def X86aesdec128kl : SDNode<"X86ISD::AESDEC128KL", SDT_X86AESENCDECKL,
341 [SDNPHasChain, SDNPMayLoad, SDNPSideEffect,
343 def X86aesenc256kl : SDNode<"X86ISD::AESENC256KL", SDT_X86AESENCDECKL,
344 [SDNPHasChain, SDNPMayLoad, SDNPSideEffect,
346 def X86aesdec256kl : SDNode<"X86ISD::AESDEC256KL", SDT_X86AESENCDECKL,
347 [SDNPHasChain, SDNPMayLoad, SDNPSideEffect,
350 def X86cmpccxadd : SDNode<"X86ISD::CMPCCXADD", SDTX86Cmpccxadd,
351 [SDNPHasChain, SDNPMayLoad, SDNPMayStore,
354 // Define X86-specific addressing mode.
355 let WantsParent = true in
356 def addr : ComplexPattern<iPTR, 5, "selectAddr">;
357 def lea32addr : ComplexPattern<i32, 5, "selectLEAAddr",
358 [add, sub, mul, X86mul_imm, shl, or, xor, frameindex],
360 // In 64-bit mode 32-bit LEAs can use RIP-relative addressing.
361 def lea64_32addr : ComplexPattern<i32, 5, "selectLEA64_32Addr",
362 [add, sub, mul, X86mul_imm, shl, or, xor,
363 frameindex, X86WrapperRIP],
366 def tls32addr : ComplexPattern<i32, 5, "selectTLSADDRAddr",
367 [tglobaltlsaddr], []>;
369 def tls32baseaddr : ComplexPattern<i32, 5, "selectTLSADDRAddr",
370 [tglobaltlsaddr], []>;
372 def lea64addr : ComplexPattern<i64, 5, "selectLEAAddr",
373 [add, sub, mul, X86mul_imm, shl, or, xor, frameindex,
376 def tls64addr : ComplexPattern<i64, 5, "selectTLSADDRAddr",
377 [tglobaltlsaddr], []>;
379 def tls64baseaddr : ComplexPattern<i64, 5, "selectTLSADDRAddr",
380 [tglobaltlsaddr], []>;
382 let WantsParent = true in
383 def vectoraddr : ComplexPattern<iPTR, 5, "selectVectorAddr">;
385 // A relocatable immediate is an operand that can be relocated by the linker to
386 // an immediate, such as a regular symbol in non-PIC code.
387 def relocImm : ComplexPattern<iAny, 1, "selectRelocImm",
388 [X86Wrapper], [], 0>;
390 // X86 specific condition code. These correspond to CondCode in
391 // X86InstrInfo.h. They must be kept in synch.
392 def X86_COND_O : PatLeaf<(i8 0)>;
393 def X86_COND_NO : PatLeaf<(i8 1)>;
394 def X86_COND_B : PatLeaf<(i8 2)>; // alt. COND_C
395 def X86_COND_AE : PatLeaf<(i8 3)>; // alt. COND_NC
396 def X86_COND_E : PatLeaf<(i8 4)>; // alt. COND_Z
397 def X86_COND_NE : PatLeaf<(i8 5)>; // alt. COND_NZ
398 def X86_COND_BE : PatLeaf<(i8 6)>; // alt. COND_NA
399 def X86_COND_A : PatLeaf<(i8 7)>; // alt. COND_NBE
400 def X86_COND_S : PatLeaf<(i8 8)>;
401 def X86_COND_NS : PatLeaf<(i8 9)>;
402 def X86_COND_P : PatLeaf<(i8 10)>; // alt. COND_PE
403 def X86_COND_NP : PatLeaf<(i8 11)>; // alt. COND_PO
404 def X86_COND_L : PatLeaf<(i8 12)>; // alt. COND_NGE
405 def X86_COND_GE : PatLeaf<(i8 13)>; // alt. COND_NL
406 def X86_COND_LE : PatLeaf<(i8 14)>; // alt. COND_NG
407 def X86_COND_G : PatLeaf<(i8 15)>; // alt. COND_NLE
409 def i16immSExt8 : ImmLeaf<i16, [{ return isInt<8>(Imm); }]>;
410 def i32immSExt8 : ImmLeaf<i32, [{ return isInt<8>(Imm); }]>;
411 def i64immSExt8 : ImmLeaf<i64, [{ return isInt<8>(Imm); }]>;
412 def i64immSExt32 : ImmLeaf<i64, [{ return isInt<32>(Imm); }]>;
413 def i64timmSExt32 : TImmLeaf<i64, [{ return isInt<32>(Imm); }]>;
415 def i16relocImmSExt8 : PatLeaf<(i16 relocImm), [{
416 return isSExtAbsoluteSymbolRef(8, N);
418 def i32relocImmSExt8 : PatLeaf<(i32 relocImm), [{
419 return isSExtAbsoluteSymbolRef(8, N);
421 def i64relocImmSExt8 : PatLeaf<(i64 relocImm), [{
422 return isSExtAbsoluteSymbolRef(8, N);
424 def i64relocImmSExt32 : PatLeaf<(i64 relocImm), [{
425 return isSExtAbsoluteSymbolRef(32, N);
428 // If we have multiple users of an immediate, it's much smaller to reuse
429 // the register, rather than encode the immediate in every instruction.
430 // This has the risk of increasing register pressure from stretched live
431 // ranges, however, the immediates should be trivial to rematerialize by
432 // the RA in the event of high register pressure.
433 // TODO : This is currently enabled for stores and binary ops. There are more
434 // cases for which this can be enabled, though this catches the bulk of the
436 // TODO2 : This should really also be enabled under O2, but there's currently
437 // an issue with RA where we don't pull the constants into their users
438 // when we rematerialize them. I'll follow-up on enabling O2 after we fix that
440 // TODO3 : This is currently limited to single basic blocks (DAG creation
441 // pulls block immediates to the top and merges them if necessary).
442 // Eventually, it would be nice to allow ConstantHoisting to merge constants
443 // globally for potentially added savings.
445 def imm_su : PatLeaf<(imm), [{
446 return !shouldAvoidImmediateInstFormsForSize(N);
448 def i64immSExt32_su : PatLeaf<(i64immSExt32), [{
449 return !shouldAvoidImmediateInstFormsForSize(N);
452 def relocImm8_su : PatLeaf<(i8 relocImm), [{
453 return !shouldAvoidImmediateInstFormsForSize(N);
455 def relocImm16_su : PatLeaf<(i16 relocImm), [{
456 return !shouldAvoidImmediateInstFormsForSize(N);
458 def relocImm32_su : PatLeaf<(i32 relocImm), [{
459 return !shouldAvoidImmediateInstFormsForSize(N);
462 def i16relocImmSExt8_su : PatLeaf<(i16relocImmSExt8), [{
463 return !shouldAvoidImmediateInstFormsForSize(N);
465 def i32relocImmSExt8_su : PatLeaf<(i32relocImmSExt8), [{
466 return !shouldAvoidImmediateInstFormsForSize(N);
468 def i64relocImmSExt8_su : PatLeaf<(i64relocImmSExt8), [{
469 return !shouldAvoidImmediateInstFormsForSize(N);
471 def i64relocImmSExt32_su : PatLeaf<(i64relocImmSExt32), [{
472 return !shouldAvoidImmediateInstFormsForSize(N);
475 def i16immSExt8_su : PatLeaf<(i16immSExt8), [{
476 return !shouldAvoidImmediateInstFormsForSize(N);
478 def i32immSExt8_su : PatLeaf<(i32immSExt8), [{
479 return !shouldAvoidImmediateInstFormsForSize(N);
481 def i64immSExt8_su : PatLeaf<(i64immSExt8), [{
482 return !shouldAvoidImmediateInstFormsForSize(N);
485 // i64immZExt32 predicate - True if the 64-bit immediate fits in a 32-bit
487 def i64immZExt32 : ImmLeaf<i64, [{ return isUInt<32>(Imm); }]>;
489 def i64immZExt32SExt8 : ImmLeaf<i64, [{
490 return isUInt<32>(Imm) && isInt<8>(static_cast<int32_t>(Imm));
493 // Helper fragments for loads.
495 // It's safe to fold a zextload/extload from i1 as a regular i8 load. The
496 // upper bits are guaranteed to be zero and we were going to emit a MOV8rm
497 // which might get folded during peephole anyway.
498 def loadi8 : PatFrag<(ops node:$ptr), (i8 (unindexedload node:$ptr)), [{
499 LoadSDNode *LD = cast<LoadSDNode>(N);
500 ISD::LoadExtType ExtType = LD->getExtensionType();
501 return ExtType == ISD::NON_EXTLOAD || ExtType == ISD::EXTLOAD ||
502 ExtType == ISD::ZEXTLOAD;
505 // It's always safe to treat a anyext i16 load as a i32 load if the i16 is
506 // known to be 32-bit aligned or better. Ditto for i8 to i16.
507 def loadi16 : PatFrag<(ops node:$ptr), (i16 (unindexedload node:$ptr)), [{
508 LoadSDNode *LD = cast<LoadSDNode>(N);
509 ISD::LoadExtType ExtType = LD->getExtensionType();
510 if (ExtType == ISD::NON_EXTLOAD)
512 if (ExtType == ISD::EXTLOAD && EnablePromoteAnyextLoad)
513 return LD->getAlign() >= 2 && LD->isSimple();
517 def loadi32 : PatFrag<(ops node:$ptr), (i32 (unindexedload node:$ptr)), [{
518 LoadSDNode *LD = cast<LoadSDNode>(N);
519 ISD::LoadExtType ExtType = LD->getExtensionType();
520 if (ExtType == ISD::NON_EXTLOAD)
522 if (ExtType == ISD::EXTLOAD && EnablePromoteAnyextLoad)
523 return LD->getAlign() >= 4 && LD->isSimple();
527 def loadi64 : PatFrag<(ops node:$ptr), (i64 (load node:$ptr))>;
528 def loadf16 : PatFrag<(ops node:$ptr), (f16 (load node:$ptr))>;
529 def loadf32 : PatFrag<(ops node:$ptr), (f32 (load node:$ptr))>;
530 def loadf64 : PatFrag<(ops node:$ptr), (f64 (load node:$ptr))>;
531 def loadf80 : PatFrag<(ops node:$ptr), (f80 (load node:$ptr))>;
532 def loadf128 : PatFrag<(ops node:$ptr), (f128 (load node:$ptr))>;
533 def alignedloadf128 : PatFrag<(ops node:$ptr), (f128 (load node:$ptr)), [{
534 LoadSDNode *Ld = cast<LoadSDNode>(N);
535 return Ld->getAlign() >= Ld->getMemoryVT().getStoreSize();
537 def memopf128 : PatFrag<(ops node:$ptr), (f128 (load node:$ptr)), [{
538 LoadSDNode *Ld = cast<LoadSDNode>(N);
539 return Subtarget->hasSSEUnalignedMem() ||
540 Ld->getAlign() >= Ld->getMemoryVT().getStoreSize();
543 def sextloadi16i8 : PatFrag<(ops node:$ptr), (i16 (sextloadi8 node:$ptr))>;
544 def sextloadi32i8 : PatFrag<(ops node:$ptr), (i32 (sextloadi8 node:$ptr))>;
545 def sextloadi32i16 : PatFrag<(ops node:$ptr), (i32 (sextloadi16 node:$ptr))>;
546 def sextloadi64i8 : PatFrag<(ops node:$ptr), (i64 (sextloadi8 node:$ptr))>;
547 def sextloadi64i16 : PatFrag<(ops node:$ptr), (i64 (sextloadi16 node:$ptr))>;
548 def sextloadi64i32 : PatFrag<(ops node:$ptr), (i64 (sextloadi32 node:$ptr))>;
550 def zextloadi8i1 : PatFrag<(ops node:$ptr), (i8 (zextloadi1 node:$ptr))>;
551 def zextloadi16i1 : PatFrag<(ops node:$ptr), (i16 (zextloadi1 node:$ptr))>;
552 def zextloadi32i1 : PatFrag<(ops node:$ptr), (i32 (zextloadi1 node:$ptr))>;
553 def zextloadi16i8 : PatFrag<(ops node:$ptr), (i16 (zextloadi8 node:$ptr))>;
554 def zextloadi32i8 : PatFrag<(ops node:$ptr), (i32 (zextloadi8 node:$ptr))>;
555 def zextloadi32i16 : PatFrag<(ops node:$ptr), (i32 (zextloadi16 node:$ptr))>;
556 def zextloadi64i1 : PatFrag<(ops node:$ptr), (i64 (zextloadi1 node:$ptr))>;
557 def zextloadi64i8 : PatFrag<(ops node:$ptr), (i64 (zextloadi8 node:$ptr))>;
558 def zextloadi64i16 : PatFrag<(ops node:$ptr), (i64 (zextloadi16 node:$ptr))>;
559 def zextloadi64i32 : PatFrag<(ops node:$ptr), (i64 (zextloadi32 node:$ptr))>;
561 def extloadi8i1 : PatFrag<(ops node:$ptr), (i8 (extloadi1 node:$ptr))>;
562 def extloadi16i1 : PatFrag<(ops node:$ptr), (i16 (extloadi1 node:$ptr))>;
563 def extloadi32i1 : PatFrag<(ops node:$ptr), (i32 (extloadi1 node:$ptr))>;
564 def extloadi16i8 : PatFrag<(ops node:$ptr), (i16 (extloadi8 node:$ptr))>;
565 def extloadi32i8 : PatFrag<(ops node:$ptr), (i32 (extloadi8 node:$ptr))>;
566 def extloadi32i16 : PatFrag<(ops node:$ptr), (i32 (extloadi16 node:$ptr))>;
567 def extloadi64i1 : PatFrag<(ops node:$ptr), (i64 (extloadi1 node:$ptr))>;
568 def extloadi64i8 : PatFrag<(ops node:$ptr), (i64 (extloadi8 node:$ptr))>;
569 def extloadi64i16 : PatFrag<(ops node:$ptr), (i64 (extloadi16 node:$ptr))>;
571 // We can treat an i8/i16 extending load to i64 as a 32 bit load if its known
572 // to be 4 byte aligned or better.
573 def extloadi64i32 : PatFrag<(ops node:$ptr), (i64 (unindexedload node:$ptr)), [{
574 LoadSDNode *LD = cast<LoadSDNode>(N);
575 ISD::LoadExtType ExtType = LD->getExtensionType();
576 if (ExtType != ISD::EXTLOAD)
578 if (LD->getMemoryVT() == MVT::i32)
581 return LD->getAlign() >= 4 && LD->isSimple();
584 // binary op with only one user
585 class binop_oneuse<SDPatternOperator operator>
586 : PatFrag<(ops node:$A, node:$B),
587 (operator node:$A, node:$B), [{
588 return N->hasOneUse();
591 def add_su : binop_oneuse<add>;
592 def and_su : binop_oneuse<and>;
593 def srl_su : binop_oneuse<srl>;
595 class binop_twouses<SDPatternOperator operator>
596 : PatFrag<(ops node:$A, node:$B),
597 (operator node:$A, node:$B), [{
598 return N->hasNUsesOfValue(2, 0);
601 def and_du : binop_twouses<and>;
603 // unary op with only one user
604 class unop_oneuse<SDPatternOperator operator>
605 : PatFrag<(ops node:$A),
606 (operator node:$A), [{
607 return N->hasOneUse();
611 def ineg_su : unop_oneuse<ineg>;
612 def trunc_su : unop_oneuse<trunc>;
614 def X86add_flag_nocf : PatFrag<(ops node:$lhs, node:$rhs),
615 (X86add_flag node:$lhs, node:$rhs), [{
616 return hasNoCarryFlagUses(SDValue(N, 1));
619 def X86sub_flag_nocf : PatFrag<(ops node:$lhs, node:$rhs),
620 (X86sub_flag node:$lhs, node:$rhs), [{
621 // Only use DEC if the result is used.
622 return !SDValue(N, 0).use_empty() && hasNoCarryFlagUses(SDValue(N, 1));
625 def X86testpat : PatFrag<(ops node:$lhs, node:$rhs),
626 (X86cmp (and_su node:$lhs, node:$rhs), 0)>;
627 def X86ctestpat : PatFrag<(ops node:$lhs, node:$rhs, node:$dcf, node:$cond),
628 (X86ctest (and_du node:$lhs, node:$rhs),
629 (and_du node:$lhs, node:$rhs), node:$dcf,
630 node:$cond, EFLAGS)>;
632 def X86any_fcmp : PatFrags<(ops node:$lhs, node:$rhs),
633 [(X86strict_fcmp node:$lhs, node:$rhs),
634 (X86fcmp node:$lhs, node:$rhs)]>;
636 def PrefetchWLevel : PatFrag<(ops), (i32 timm), [{
637 return N->getSExtValue() <= 3;
640 def X86lock_add_nocf : PatFrag<(ops node:$lhs, node:$rhs),
641 (X86lock_add node:$lhs, node:$rhs), [{
642 return hasNoCarryFlagUses(SDValue(N, 0));
645 def X86lock_sub_nocf : PatFrag<(ops node:$lhs, node:$rhs),
646 (X86lock_sub node:$lhs, node:$rhs), [{
647 return hasNoCarryFlagUses(SDValue(N, 0));
650 def X86tcret_6regs : PatFrag<(ops node:$ptr, node:$off),
651 (X86tcret node:$ptr, node:$off), [{
652 // X86tcret args: (*chain, ptr, imm, regs..., glue)
653 unsigned NumRegs = 0;
654 for (unsigned i = 3, e = N->getNumOperands(); i != e; ++i)
655 if (isa<RegisterSDNode>(N->getOperand(i)) && ++NumRegs > 6)
660 def X86tcret_1reg : PatFrag<(ops node:$ptr, node:$off),
661 (X86tcret node:$ptr, node:$off), [{
662 // X86tcret args: (*chain, ptr, imm, regs..., glue)
663 unsigned NumRegs = 1;
664 const SDValue& BasePtr = cast<LoadSDNode>(N->getOperand(1))->getBasePtr();
665 if (isa<FrameIndexSDNode>(BasePtr))
667 else if (BasePtr->getNumOperands() && isa<GlobalAddressSDNode>(BasePtr->getOperand(0)))
669 for (unsigned i = 3, e = N->getNumOperands(); i != e; ++i)
670 if (isa<RegisterSDNode>(N->getOperand(i)) && ( NumRegs-- == 0))
675 // If this is an anyext of the remainder of an 8-bit sdivrem, use a MOVSX
676 // instead of a MOVZX. The sdivrem lowering will emit emit a MOVSX to move
677 // %ah to the lower byte of a register. By using a MOVSX here we allow a
678 // post-isel peephole to merge the two MOVSX instructions into one.
679 def anyext_sdiv : PatFrag<(ops node:$lhs), (anyext node:$lhs),[{
680 return (N->getOperand(0).getOpcode() == ISD::SDIVREM &&
681 N->getOperand(0).getResNo() == 1);
684 // Any instruction that defines a 32-bit result leaves the high half of the
685 // register. Truncate can be lowered to EXTRACT_SUBREG. CopyFromReg may
686 // be copying from a truncate. AssertSext/AssertZext/AssertAlign aren't saying
687 // anything about the upper 32 bits, they're probably just qualifying a
688 // CopyFromReg. FREEZE may be coming from a a truncate. Any other 32-bit
689 // operation will zero-extend up to 64 bits.
690 def def32 : PatLeaf<(i32 GR32:$src), [{
691 return N->getOpcode() != ISD::TRUNCATE &&
692 N->getOpcode() != TargetOpcode::EXTRACT_SUBREG &&
693 N->getOpcode() != ISD::CopyFromReg &&
694 N->getOpcode() != ISD::AssertSext &&
695 N->getOpcode() != ISD::AssertZext &&
696 N->getOpcode() != ISD::AssertAlign &&
697 N->getOpcode() != ISD::FREEZE;
700 // Treat an 'or' node is as an 'add' if the or'ed bits are known to be zero.
701 def or_is_add : PatFrag<(ops node:$lhs, node:$rhs), (or node:$lhs, node:$rhs),[{
702 if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N->getOperand(1)))
703 return CurDAG->MaskedValueIsZero(N->getOperand(0), CN->getAPIntValue());
705 KnownBits Known0 = CurDAG->computeKnownBits(N->getOperand(0), 0);
706 KnownBits Known1 = CurDAG->computeKnownBits(N->getOperand(1), 0);
707 return (~Known0.Zero & ~Known1.Zero) == 0;
710 def shiftMask8 : PatFrag<(ops node:$lhs), (and node:$lhs, imm), [{
711 return isUnneededShiftMask(N, 3);
714 def shiftMask16 : PatFrag<(ops node:$lhs), (and node:$lhs, imm), [{
715 return isUnneededShiftMask(N, 4);
718 def shiftMask32 : PatFrag<(ops node:$lhs), (and node:$lhs, imm), [{
719 return isUnneededShiftMask(N, 5);
722 def shiftMask64 : PatFrag<(ops node:$lhs), (and node:$lhs, imm), [{
723 return isUnneededShiftMask(N, 6);
726 //===----------------------------------------------------------------------===//
727 // Pattern fragments to auto generate BMI instructions.
728 //===----------------------------------------------------------------------===//
730 def or_flag_nocf : PatFrag<(ops node:$lhs, node:$rhs),
731 (X86or_flag node:$lhs, node:$rhs), [{
732 return hasNoCarryFlagUses(SDValue(N, 1));
735 def xor_flag_nocf : PatFrag<(ops node:$lhs, node:$rhs),
736 (X86xor_flag node:$lhs, node:$rhs), [{
737 return hasNoCarryFlagUses(SDValue(N, 1));
740 def and_flag_nocf : PatFrag<(ops node:$lhs, node:$rhs),
741 (X86and_flag node:$lhs, node:$rhs), [{
742 return hasNoCarryFlagUses(SDValue(N, 1));
745 //===----------------------------------------------------------------------===//
746 // FPStack specific DAG Nodes.
747 //===----------------------------------------------------------------------===//
749 def SDTX86Fld : SDTypeProfile<1, 1, [SDTCisFP<0>,
751 def SDTX86Fst : SDTypeProfile<0, 2, [SDTCisFP<0>,
753 def SDTX86Fild : SDTypeProfile<1, 1, [SDTCisFP<0>, SDTCisPtrTy<1>]>;
754 def SDTX86Fist : SDTypeProfile<0, 2, [SDTCisFP<0>, SDTCisPtrTy<1>]>;
756 def SDTX86CwdStore : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
757 def SDTX86CwdLoad : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
758 def SDTX86FPEnv : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
760 def X86fp80_add : SDNode<"X86ISD::FP80_ADD", SDTFPBinOp, [SDNPCommutative]>;
761 def X86strict_fp80_add : SDNode<"X86ISD::STRICT_FP80_ADD", SDTFPBinOp,
762 [SDNPHasChain,SDNPCommutative]>;
763 def any_X86fp80_add : PatFrags<(ops node:$lhs, node:$rhs),
764 [(X86strict_fp80_add node:$lhs, node:$rhs),
765 (X86fp80_add node:$lhs, node:$rhs)]>;
767 def X86fld : SDNode<"X86ISD::FLD", SDTX86Fld,
768 [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
769 def X86fst : SDNode<"X86ISD::FST", SDTX86Fst,
770 [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
771 def X86fild : SDNode<"X86ISD::FILD", SDTX86Fild,
772 [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
773 def X86fist : SDNode<"X86ISD::FIST", SDTX86Fist,
774 [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
775 def X86fp_to_mem : SDNode<"X86ISD::FP_TO_INT_IN_MEM", SDTX86Fst,
776 [SDNPHasChain, SDNPMayStore, SDNPMemOperand]>;
777 def X86fp_cwd_get16 : SDNode<"X86ISD::FNSTCW16m", SDTX86CwdStore,
778 [SDNPHasChain, SDNPMayStore, SDNPSideEffect,
780 def X86fp_cwd_set16 : SDNode<"X86ISD::FLDCW16m", SDTX86CwdLoad,
781 [SDNPHasChain, SDNPMayLoad, SDNPSideEffect,
783 def X86fpenv_get : SDNode<"X86ISD::FNSTENVm", SDTX86FPEnv,
784 [SDNPHasChain, SDNPMayStore, SDNPSideEffect,
786 def X86fpenv_set : SDNode<"X86ISD::FLDENVm", SDTX86FPEnv,
787 [SDNPHasChain, SDNPMayLoad, SDNPSideEffect,
790 def X86fstf32 : PatFrag<(ops node:$val, node:$ptr),
791 (X86fst node:$val, node:$ptr), [{
792 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::f32;
794 def X86fstf64 : PatFrag<(ops node:$val, node:$ptr),
795 (X86fst node:$val, node:$ptr), [{
796 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::f64;
798 def X86fstf80 : PatFrag<(ops node:$val, node:$ptr),
799 (X86fst node:$val, node:$ptr), [{
800 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::f80;
803 def X86fldf32 : PatFrag<(ops node:$ptr), (X86fld node:$ptr), [{
804 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::f32;
806 def X86fldf64 : PatFrag<(ops node:$ptr), (X86fld node:$ptr), [{
807 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::f64;
809 def X86fldf80 : PatFrag<(ops node:$ptr), (X86fld node:$ptr), [{
810 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::f80;
813 def X86fild16 : PatFrag<(ops node:$ptr), (X86fild node:$ptr), [{
814 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i16;
816 def X86fild32 : PatFrag<(ops node:$ptr), (X86fild node:$ptr), [{
817 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i32;
819 def X86fild64 : PatFrag<(ops node:$ptr), (X86fild node:$ptr), [{
820 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i64;
823 def X86fist32 : PatFrag<(ops node:$val, node:$ptr),
824 (X86fist node:$val, node:$ptr), [{
825 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i32;
828 def X86fist64 : PatFrag<(ops node:$val, node:$ptr),
829 (X86fist node:$val, node:$ptr), [{
830 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i64;
833 def X86fp_to_i16mem : PatFrag<(ops node:$val, node:$ptr),
834 (X86fp_to_mem node:$val, node:$ptr), [{
835 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i16;
837 def X86fp_to_i32mem : PatFrag<(ops node:$val, node:$ptr),
838 (X86fp_to_mem node:$val, node:$ptr), [{
839 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i32;
841 def X86fp_to_i64mem : PatFrag<(ops node:$val, node:$ptr),
842 (X86fp_to_mem node:$val, node:$ptr), [{
843 return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i64;
846 //===----------------------------------------------------------------------===//
847 // FPStack pattern fragments
848 //===----------------------------------------------------------------------===//
850 def fpimm0 : FPImmLeaf<fAny, [{
851 return Imm.isExactlyValue(+0.0);
854 def fpimmneg0 : FPImmLeaf<fAny, [{
855 return Imm.isExactlyValue(-0.0);
858 def fpimm1 : FPImmLeaf<fAny, [{
859 return Imm.isExactlyValue(+1.0);
862 def fpimmneg1 : FPImmLeaf<fAny, [{
863 return Imm.isExactlyValue(-1.0);