1//===- TargetLoweringBase.cpp - Implement the TargetLoweringBase class ----===// 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// This implements the TargetLoweringBase class. 11//===----------------------------------------------------------------------===// 70cl::desc(
"Do not create extra branches to split comparison logic."),
75cl::desc(
"Set minimum number of entries to use a jump table."));
79cl::desc(
"Set maximum size of jump tables."));
81/// Minimum jump table density for normal functions. 84cl::desc(
"Minimum density for building a jump table in " 87/// Minimum jump table density for -Os or -Oz functions. 90cl::desc(
"Minimum density for building a jump table in " 91"an optsize function"));
93// FIXME: This option is only to test if the strict fp operation processed 94// correctly by preventing mutating strict fp operation to normal fp operation 95// during development. When the backend supports strict float operation, this 96// option will be meaningless. 98cl::desc(
"Don't mutate strict-float node to a legalize node"),
101/// GetFPLibCall - Helper to return the right libcall for the given floating 102/// point type, or UNKNOWN_LIBCALL if there is none. 110 VT == MVT::f32 ? Call_F32 :
111 VT == MVT::f64 ? Call_F64 :
112 VT == MVT::f80 ? Call_F80 :
113 VT == MVT::f128 ? Call_F128 :
114 VT == MVT::ppcf128 ? Call_PPCF128 :
115 RTLIB::UNKNOWN_LIBCALL;
118/// getFPEXT - Return the FPEXT_*_* value for the given types, or 119/// UNKNOWN_LIBCALL if there is none. 121if (OpVT == MVT::f16) {
122if (RetVT == MVT::f32)
124if (RetVT == MVT::f64)
126if (RetVT == MVT::f80)
128if (RetVT == MVT::f128)
129return FPEXT_F16_F128;
130 }
elseif (OpVT == MVT::f32) {
131if (RetVT == MVT::f64)
133if (RetVT == MVT::f128)
134return FPEXT_F32_F128;
135if (RetVT == MVT::ppcf128)
136return FPEXT_F32_PPCF128;
137 }
elseif (OpVT == MVT::f64) {
138if (RetVT == MVT::f128)
139return FPEXT_F64_F128;
140elseif (RetVT == MVT::ppcf128)
141return FPEXT_F64_PPCF128;
142 }
elseif (OpVT == MVT::f80) {
143if (RetVT == MVT::f128)
144return FPEXT_F80_F128;
145 }
elseif (OpVT == MVT::bf16) {
146if (RetVT == MVT::f32)
147return FPEXT_BF16_F32;
150return UNKNOWN_LIBCALL;
153/// getFPROUND - Return the FPROUND_*_* value for the given types, or 154/// UNKNOWN_LIBCALL if there is none. 156if (RetVT == MVT::f16) {
158return FPROUND_F32_F16;
160return FPROUND_F64_F16;
162return FPROUND_F80_F16;
163if (OpVT == MVT::f128)
164return FPROUND_F128_F16;
165if (OpVT == MVT::ppcf128)
166return FPROUND_PPCF128_F16;
167 }
elseif (RetVT == MVT::bf16) {
169return FPROUND_F32_BF16;
171return FPROUND_F64_BF16;
173return FPROUND_F80_BF16;
174if (OpVT == MVT::f128)
175return FPROUND_F128_BF16;
176 }
elseif (RetVT == MVT::f32) {
178return FPROUND_F64_F32;
180return FPROUND_F80_F32;
181if (OpVT == MVT::f128)
182return FPROUND_F128_F32;
183if (OpVT == MVT::ppcf128)
184return FPROUND_PPCF128_F32;
185 }
elseif (RetVT == MVT::f64) {
187return FPROUND_F80_F64;
188if (OpVT == MVT::f128)
189return FPROUND_F128_F64;
190if (OpVT == MVT::ppcf128)
191return FPROUND_PPCF128_F64;
192 }
elseif (RetVT == MVT::f80) {
193if (OpVT == MVT::f128)
194return FPROUND_F128_F80;
197return UNKNOWN_LIBCALL;
200/// getFPTOSINT - Return the FPTOSINT_*_* value for the given types, or 201/// UNKNOWN_LIBCALL if there is none. 203if (OpVT == MVT::f16) {
204if (RetVT == MVT::i32)
205return FPTOSINT_F16_I32;
206if (RetVT == MVT::i64)
207return FPTOSINT_F16_I64;
208if (RetVT == MVT::i128)
209return FPTOSINT_F16_I128;
210 }
elseif (OpVT == MVT::f32) {
211if (RetVT == MVT::i32)
212return FPTOSINT_F32_I32;
213if (RetVT == MVT::i64)
214return FPTOSINT_F32_I64;
215if (RetVT == MVT::i128)
216return FPTOSINT_F32_I128;
217 }
elseif (OpVT == MVT::f64) {
218if (RetVT == MVT::i32)
219return FPTOSINT_F64_I32;
220if (RetVT == MVT::i64)
221return FPTOSINT_F64_I64;
222if (RetVT == MVT::i128)
223return FPTOSINT_F64_I128;
224 }
elseif (OpVT == MVT::f80) {
225if (RetVT == MVT::i32)
226return FPTOSINT_F80_I32;
227if (RetVT == MVT::i64)
228return FPTOSINT_F80_I64;
229if (RetVT == MVT::i128)
230return FPTOSINT_F80_I128;
231 }
elseif (OpVT == MVT::f128) {
232if (RetVT == MVT::i32)
233return FPTOSINT_F128_I32;
234if (RetVT == MVT::i64)
235return FPTOSINT_F128_I64;
236if (RetVT == MVT::i128)
237return FPTOSINT_F128_I128;
238 }
elseif (OpVT == MVT::ppcf128) {
239if (RetVT == MVT::i32)
240return FPTOSINT_PPCF128_I32;
241if (RetVT == MVT::i64)
242return FPTOSINT_PPCF128_I64;
243if (RetVT == MVT::i128)
244return FPTOSINT_PPCF128_I128;
246return UNKNOWN_LIBCALL;
249/// getFPTOUINT - Return the FPTOUINT_*_* value for the given types, or 250/// UNKNOWN_LIBCALL if there is none. 252if (OpVT == MVT::f16) {
253if (RetVT == MVT::i32)
254return FPTOUINT_F16_I32;
255if (RetVT == MVT::i64)
256return FPTOUINT_F16_I64;
257if (RetVT == MVT::i128)
258return FPTOUINT_F16_I128;
259 }
elseif (OpVT == MVT::f32) {
260if (RetVT == MVT::i32)
261return FPTOUINT_F32_I32;
262if (RetVT == MVT::i64)
263return FPTOUINT_F32_I64;
264if (RetVT == MVT::i128)
265return FPTOUINT_F32_I128;
266 }
elseif (OpVT == MVT::f64) {
267if (RetVT == MVT::i32)
268return FPTOUINT_F64_I32;
269if (RetVT == MVT::i64)
270return FPTOUINT_F64_I64;
271if (RetVT == MVT::i128)
272return FPTOUINT_F64_I128;
273 }
elseif (OpVT == MVT::f80) {
274if (RetVT == MVT::i32)
275return FPTOUINT_F80_I32;
276if (RetVT == MVT::i64)
277return FPTOUINT_F80_I64;
278if (RetVT == MVT::i128)
279return FPTOUINT_F80_I128;
280 }
elseif (OpVT == MVT::f128) {
281if (RetVT == MVT::i32)
282return FPTOUINT_F128_I32;
283if (RetVT == MVT::i64)
284return FPTOUINT_F128_I64;
285if (RetVT == MVT::i128)
286return FPTOUINT_F128_I128;
287 }
elseif (OpVT == MVT::ppcf128) {
288if (RetVT == MVT::i32)
289return FPTOUINT_PPCF128_I32;
290if (RetVT == MVT::i64)
291return FPTOUINT_PPCF128_I64;
292if (RetVT == MVT::i128)
293return FPTOUINT_PPCF128_I128;
295return UNKNOWN_LIBCALL;
298/// getSINTTOFP - Return the SINTTOFP_*_* value for the given types, or 299/// UNKNOWN_LIBCALL if there is none. 301if (OpVT == MVT::i32) {
302if (RetVT == MVT::f16)
303return SINTTOFP_I32_F16;
304if (RetVT == MVT::f32)
305return SINTTOFP_I32_F32;
306if (RetVT == MVT::f64)
307return SINTTOFP_I32_F64;
308if (RetVT == MVT::f80)
309return SINTTOFP_I32_F80;
310if (RetVT == MVT::f128)
311return SINTTOFP_I32_F128;
312if (RetVT == MVT::ppcf128)
313return SINTTOFP_I32_PPCF128;
314 }
elseif (OpVT == MVT::i64) {
315if (RetVT == MVT::f16)
316return SINTTOFP_I64_F16;
317if (RetVT == MVT::f32)
318return SINTTOFP_I64_F32;
319if (RetVT == MVT::f64)
320return SINTTOFP_I64_F64;
321if (RetVT == MVT::f80)
322return SINTTOFP_I64_F80;
323if (RetVT == MVT::f128)
324return SINTTOFP_I64_F128;
325if (RetVT == MVT::ppcf128)
326return SINTTOFP_I64_PPCF128;
327 }
elseif (OpVT == MVT::i128) {
328if (RetVT == MVT::f16)
329return SINTTOFP_I128_F16;
330if (RetVT == MVT::f32)
331return SINTTOFP_I128_F32;
332if (RetVT == MVT::f64)
333return SINTTOFP_I128_F64;
334if (RetVT == MVT::f80)
335return SINTTOFP_I128_F80;
336if (RetVT == MVT::f128)
337return SINTTOFP_I128_F128;
338if (RetVT == MVT::ppcf128)
339return SINTTOFP_I128_PPCF128;
341return UNKNOWN_LIBCALL;
344/// getUINTTOFP - Return the UINTTOFP_*_* value for the given types, or 345/// UNKNOWN_LIBCALL if there is none. 347if (OpVT == MVT::i32) {
348if (RetVT == MVT::f16)
349return UINTTOFP_I32_F16;
350if (RetVT == MVT::f32)
351return UINTTOFP_I32_F32;
352if (RetVT == MVT::f64)
353return UINTTOFP_I32_F64;
354if (RetVT == MVT::f80)
355return UINTTOFP_I32_F80;
356if (RetVT == MVT::f128)
357return UINTTOFP_I32_F128;
358if (RetVT == MVT::ppcf128)
359return UINTTOFP_I32_PPCF128;
360 }
elseif (OpVT == MVT::i64) {
361if (RetVT == MVT::f16)
362return UINTTOFP_I64_F16;
363if (RetVT == MVT::f32)
364return UINTTOFP_I64_F32;
365if (RetVT == MVT::f64)
366return UINTTOFP_I64_F64;
367if (RetVT == MVT::f80)
368return UINTTOFP_I64_F80;
369if (RetVT == MVT::f128)
370return UINTTOFP_I64_F128;
371if (RetVT == MVT::ppcf128)
372return UINTTOFP_I64_PPCF128;
373 }
elseif (OpVT == MVT::i128) {
374if (RetVT == MVT::f16)
375return UINTTOFP_I128_F16;
376if (RetVT == MVT::f32)
377return UINTTOFP_I128_F32;
378if (RetVT == MVT::f64)
379return UINTTOFP_I128_F64;
380if (RetVT == MVT::f80)
381return UINTTOFP_I128_F80;
382if (RetVT == MVT::f128)
383return UINTTOFP_I128_F128;
384if (RetVT == MVT::ppcf128)
385return UINTTOFP_I128_PPCF128;
387return UNKNOWN_LIBCALL;
391returngetFPLibCall(RetVT, POWI_F32, POWI_F64, POWI_F80, POWI_F128,
396returngetFPLibCall(RetVT, LDEXP_F32, LDEXP_F64, LDEXP_F80, LDEXP_F128,
401returngetFPLibCall(RetVT, FREXP_F32, FREXP_F64, FREXP_F80, FREXP_F128,
406returngetFPLibCall(RetVT, SINCOS_F32, SINCOS_F64, SINCOS_F80, SINCOS_F128,
413unsigned ModeN, ModelN;
431return RTLIB::UNKNOWN_LIBCALL;
435case AtomicOrdering::Monotonic:
438case AtomicOrdering::Acquire:
441case AtomicOrdering::Release:
444case AtomicOrdering::AcquireRelease:
445case AtomicOrdering::SequentiallyConsistent:
449return UNKNOWN_LIBCALL;
452return LC[ModeN][ModelN];
458return UNKNOWN_LIBCALL;
461#define LCALLS(A, B) \ 462 { A##B##_RELAX, A##B##_ACQ, A##B##_REL, A##B##_ACQ_REL } 464 LCALLS(A, 1), LCALLS(A, 2), LCALLS(A, 4), LCALLS(A, 8), LCALLS(A, 16) 491return UNKNOWN_LIBCALL;
498#define OP_TO_LIBCALL(Name, Enum) \ 500 switch (VT.SimpleTy) { \ 502 return UNKNOWN_LIBCALL; \ 532return UNKNOWN_LIBCALL;
536switch (ElementSize) {
538return MEMCPY_ELEMENT_UNORDERED_ATOMIC_1;
540return MEMCPY_ELEMENT_UNORDERED_ATOMIC_2;
542return MEMCPY_ELEMENT_UNORDERED_ATOMIC_4;
544return MEMCPY_ELEMENT_UNORDERED_ATOMIC_8;
546return MEMCPY_ELEMENT_UNORDERED_ATOMIC_16;
548return UNKNOWN_LIBCALL;
553switch (ElementSize) {
555return MEMMOVE_ELEMENT_UNORDERED_ATOMIC_1;
557return MEMMOVE_ELEMENT_UNORDERED_ATOMIC_2;
559return MEMMOVE_ELEMENT_UNORDERED_ATOMIC_4;
561return MEMMOVE_ELEMENT_UNORDERED_ATOMIC_8;
563return MEMMOVE_ELEMENT_UNORDERED_ATOMIC_16;
565return UNKNOWN_LIBCALL;
570switch (ElementSize) {
572return MEMSET_ELEMENT_UNORDERED_ATOMIC_1;
574return MEMSET_ELEMENT_UNORDERED_ATOMIC_2;
576return MEMSET_ELEMENT_UNORDERED_ATOMIC_4;
578return MEMSET_ELEMENT_UNORDERED_ATOMIC_8;
580return MEMSET_ELEMENT_UNORDERED_ATOMIC_16;
582return UNKNOWN_LIBCALL;
587 std::fill(CmpLibcallCCs, CmpLibcallCCs + RTLIB::UNKNOWN_LIBCALL,
592 CmpLibcallCCs[RTLIB::OEQ_PPCF128] =
ISD::SETEQ;
596 CmpLibcallCCs[RTLIB::UNE_PPCF128] =
ISD::SETNE;
600 CmpLibcallCCs[RTLIB::OGE_PPCF128] =
ISD::SETGE;
604 CmpLibcallCCs[RTLIB::OLT_PPCF128] =
ISD::SETLT;
608 CmpLibcallCCs[RTLIB::OLE_PPCF128] =
ISD::SETLE;
612 CmpLibcallCCs[RTLIB::OGT_PPCF128] =
ISD::SETGT;
619/// NOTE: The TargetMachine owns TLOF. 621 : TM(tm), Libcalls(TM.getTargetTriple()) {
624// Perform these initializations only once. 630 HasMultipleConditionRegisters =
false;
631 HasExtractBitsInsn =
false;
635 StackPointerRegisterToSaveRestore = 0;
642 MaxBytesForAlignment = 0;
643 MaxAtomicSizeInBitsSupported = 0;
645// Assume that even with libcalls, no target supports wider than 128 bit 647 MaxDivRemBitWidthSupported = 128;
651 MinCmpXchgSizeInBits = 0;
652 SupportsUnalignedAtomics =
false;
658// All operations default to being supported. 659 memset(OpActions, 0,
sizeof(OpActions));
660 memset(LoadExtActions, 0,
sizeof(LoadExtActions));
661 memset(TruncStoreActions, 0,
sizeof(TruncStoreActions));
662 memset(IndexedModeActions, 0,
sizeof(IndexedModeActions));
663 memset(CondCodeActions, 0,
sizeof(CondCodeActions));
664 std::fill(std::begin(RegClassForVT), std::end(RegClassForVT),
nullptr);
665 std::fill(std::begin(TargetDAGCombineArray),
666 std::end(TargetDAGCombineArray), 0);
668// Let extending atomic loads be unsupported by default. 674// We're somewhat special casing MVT::i2 and MVT::i4. Ideally we want to 675// remove this and targets should individually set these types if not legal. 678for (
MVT VT : {MVT::i2, MVT::i4})
679 OpActions[(
unsigned)VT.SimpleTy][NT] =
Expand;
682for (
MVT VT : {MVT::i2, MVT::i4, MVT::v128i2, MVT::v64i4}) {
690for (
MVT VT : {MVT::i2, MVT::i4}) {
706// Set default actions for various operations. 708// Default all indexed load / store to expand. 717// Most backends expect to see the node which just returns the value loaded. 720// These operations default to expand. 741// Overflow operations default to expand 746// Carry-using overflow operations default to expand. 751// ADDC/ADDE/SUBC/SUBE default to expand. 755// [US]CMP default to expand 763// Absolute difference 771// These default to Expand so they will be expanded to CTLZ/CTTZ by default. 777// These library functions default to expand. 782// These operations default to expand for vector types. 792// Constrained floating-point operations default to expand. 793#define DAG_INSTRUCTION(NAME, NARG, ROUND_MODE, INTRINSIC, DAGN) \ 794 setOperationAction(ISD::STRICT_##DAGN, VT, Expand); 795#include "llvm/IR/ConstrainedOps.def" 797// For most targets @llvm.get.dynamic.area.offset just returns 0. 800// Vector reduction default to expand. 810// Named vector shuffles default to expand. 813// Only some target support this vector operation. Most need to expand it. 816// VP operations default to expand. 817#define BEGIN_REGISTER_VP_SDNODE(SDOPC, ...) \ 818 setOperationAction(ISD::SDOPC, VT, Expand); 819#include "llvm/IR/VPIntrinsics.def" 821// Masked vector extracts default to expand. 824// FP environment operations default to expand. 830// Most targets ignore the @llvm.prefetch intrinsic. 833// Most targets also ignore the @llvm.readcyclecounter intrinsic. 836// Most targets also ignore the @llvm.readsteadycounter intrinsic. 839// ConstantFP nodes default to expand. Targets can either change this to 840// Legal, in which case all fp constants are legal, or use isFPImmLegal() 841// to optimize expansions for certain constants. 843 {MVT::bf16, MVT::f16, MVT::f32, MVT::f64, MVT::f80, MVT::f128},
846// These library functions default to expand. 853 {MVT::f32, MVT::f64, MVT::f128},
Expand);
855// FIXME: Query RuntimeLibCalls to make the decision. 857 {MVT::f32, MVT::f64, MVT::f128},
LibCall);
862// Default ISD::TRAP to expand (which turns it into abort). 865// On most systems, DEBUGTRAP and TRAP have no difference. The "Expand" 866// here is to inform DAG Legalizer to replace DEBUGTRAP with TRAP. 874for (
MVT VT : {MVT::i8, MVT::i16, MVT::i32, MVT::i64}) {
880// This one by default will call __clear_cache unless the target 881// wants something different. 896// If any possible shift value won't fit in the prefered type, just use 897// something safe. Assume it will be legalized when the shift is expanded. 901"ShiftVT is still too small!");
919unsigned DestAS)
const{
926// Find the smallest "sensible" element type to use for the expansion. 934unsigned EltWidth =
RetTy->getScalarSizeInBits();
942// If the command-line option was specified, ignore this request. 944 JumpIsExpensive = isExpensive;
949// If this is a simple type, use the ComputeRegisterProp mechanism. 960"Promote may not follow Expand or Promote");
969// Handle Extended Scalar Types. 973// First promote to a power-of-two size, then expand if necessary. 976assert(NVT != VT &&
"Unable to round integer VT");
978// Avoid multi-step promotion. 981// Return rounded integer type. 989// Handle vector types. 993// Vectors with only one element are always scalarized. 997// Try to widen vector elements until the element type is a power of two and 998// promote it to a legal type later on, for example: 999// <3 x i8> -> <4 x i8> -> <4 x i32> 1001// Vectors with a number of elements that is not a power of two are always 1002// widened, for example <3 x i8> -> <4 x i8>. 1009// Examine the element type. 1012// If type is to be expanded, split the vector. 1013// <4 x i140> -> <2 x i140> 1021// Promote the integer element types until a legal vector type is found 1022// or until the element integer type is too big. If a legal type was not 1023// found, fallback to the usual mechanism of widening/splitting the 1025EVT OldEltVT = EltVT;
1027// Increase the bitwidth of the element to the next pow-of-two 1028// (which is greater than 8 bits). 1032// Stop trying when getting a non-simple element type. 1033// Note that vector elements may be greater than legal vector element 1034// types. Example: X86 XMM registers hold 64bit element on 32bit 1039// Build a new vector type and check if it is legal. 1041// Found a legal promoted vector type. 1047// Reset the type to the unexpanded type if we did not find a legal vector 1048// type with a promoted vector element type. 1052// Try to widen the vector until a legal type is found. 1053// If there is no wider legal type, split the vector. 1055// Round up to the next power of 2. 1058// If there is no simple vector type with this many elements then there 1059// cannot be a larger legal vector type. Note that this assumes that 1060// there are no skipped intermediate vector types in the simple types. 1064if (LargerVector ==
MVT())
1067// If this type is legal then widen the vector. 1072// Widen odd vectors to next power of two. 1081// Vectors with illegal element types are expanded. 1088unsigned &NumIntermediates,
1091// Figure out the right, legal destination reg to copy into. 1095unsigned NumVectorRegs = 1;
1097// Scalable vectors cannot be scalarized, so splitting or widening is 1101"Splitting or widening of non-power-of-2 MVTs is not implemented.");
1103// FIXME: We don't support non-power-of-2-sized vectors for now. 1104// Ideally we could break down into LHS/RHS like LegalizeDAG does. 1106// Split EC to unit size (scalable property is preserved). 1107 NumVectorRegs = EC.getKnownMinValue();
1111// Divide the input until we get to a supported size. This will 1112// always end up with an EC that represent a scalar or a scalable 1114while (EC.getKnownMinValue() > 1 &&
1116 EC = EC.divideCoefficientBy(2);
1117 NumVectorRegs <<= 1;
1120 NumIntermediates = NumVectorRegs;
1125 IntermediateVT = NewVT;
1129// Convert sizes such as i33 to i64. 1133 RegisterVT = DestVT;
1134if (
EVT(DestVT).bitsLT(NewVT))
// Value is expanded, e.g. i64 -> i16. 1137// Otherwise, promotion or legal types use the same number of registers as 1138// the vector decimated to the appropriate level. 1139return NumVectorRegs;
1142/// isLegalRC - Return true if the value types that can be represented by the 1143/// specified register class are all legal. 1146for (
constauto *
I =
TRI.legalclasstypes_begin(RC); *
I != MVT::Other; ++
I)
1152/// Replace/modify any TargetFrameIndex operands with a targte-dependent 1153/// sequence of memory operands that is recognized by PrologEpilogInserter. 1161// We're handling multiple types of operands here: 1162// PATCHPOINT MetaArgs - live-in, read only, direct 1163// STATEPOINT Deopt Spill - live-through, read only, indirect 1164// STATEPOINT Deopt Alloca - live-through, read only, direct 1165// (We're currently conservative and mark the deopt slots read/write in 1167// STATEPOINT GC Spill - live-through, read/write, indirect 1168// STATEPOINT GC Alloca - live-through, read/write, direct 1169// The live-in vs live-through is handled already (the live through ones are 1170// all stack slots), but we need to handle the different type of stackmap 1171// operands and memory effects here. 1179// Inherit previous memory operands. 1182for (
unsigned i = 0; i <
MI->getNumOperands(); ++i) {
1185// Index of Def operand this Use it tied to. 1186// Since Defs are coming before Uses, if Use is tied, then 1187// index of Def must be smaller that index of that Use. 1188// Also, Defs preserve their position in new MI. 1191 TiedTo =
MI->findTiedOperandIdx(i);
1198// foldMemoryOperand builds a new MI after replacing a single FI operand 1199// with the canonical set of five x86 addressing-mode operands. 1202// Add frame index operands recognized by stackmaps.cpp 1204// indirect-mem-ref tag, size, #FI, offset. 1205// Used for spills inserted by StatepointLowering. This codepath is not 1206// used for patchpoints/stackmaps at all, for these spilling is done via 1207// foldMemoryOperand callback only. 1208assert(
MI->getOpcode() == TargetOpcode::STATEPOINT &&
"sanity");
1209 MIB.
addImm(StackMaps::IndirectMemRefOp);
1214// direct-mem-ref tag, #FI, offset. 1215// Used by patchpoint, and direct alloca arguments to statepoints 1216 MIB.
addImm(StackMaps::DirectMemRefOp);
1221assert(MIB->
mayLoad() &&
"Folded a stackmap use to a non-load!");
1223// Add a new memory operand for this FI. 1226// Note: STATEPOINT MMOs are added during SelectionDAG. STACKMAP, and 1227// PATCHPOINT should be updated to do the same. (TODO) 1228if (
MI->getOpcode() != TargetOpcode::STATEPOINT) {
1237MI->eraseFromParent();
1241/// findRepresentativeClass - Return the largest legal super-reg register class 1242/// of the register class for the specified type and its associated "cost". 1243// This function is in TargetLowering because it uses RegClassForVT which would 1244// need to be moved to TargetRegisterInfo and would necessitate moving 1245// isTypeLegal over as well - a massive change that would just require 1246// TargetLowering having a TargetRegisterInfo class member that it would use. 1247std::pair<const TargetRegisterClass *, uint8_t>
1252return std::make_pair(RC, 0);
1254// Compute the set of all super-register classes. 1259// Find the first legal register class with the largest spill size. 1261for (
unsigned i : SuperRegRC.
set_bits()) {
1263// We want the largest possible spill size. 1264if (
TRI->getSpillSize(*SuperRC) <=
TRI->getSpillSize(*BestRC))
1270return std::make_pair(BestRC, 1);
1273/// computeRegisterProperties - Once all of the register classes are added, 1274/// this allows us to compute derived properties we expose. 1277// Everything defaults to needing one register. 1279 NumRegistersForVT[i] = 1;
1282// ...except isVoid, which doesn't need any registers. 1283 NumRegistersForVT[MVT::isVoid] = 0;
1285// Find the largest integer register class. 1286unsigned LargestIntReg = MVT::LAST_INTEGER_VALUETYPE;
1287for (; RegClassForVT[LargestIntReg] ==
nullptr; --LargestIntReg)
1288assert(LargestIntReg != MVT::i1 &&
"No integer registers defined!");
1290// Every integer value type larger than this largest register takes twice as 1291// many registers to represent as the previous ValueType. 1292for (
unsigned ExpandedReg = LargestIntReg + 1;
1293 ExpandedReg <= MVT::LAST_INTEGER_VALUETYPE; ++ExpandedReg) {
1294 NumRegistersForVT[ExpandedReg] = 2*NumRegistersForVT[ExpandedReg-1];
1301// Inspect all of the ValueType's smaller than the largest integer 1302// register to see which ones need promotion. 1303unsigned LegalIntReg = LargestIntReg;
1304for (
unsigned IntReg = LargestIntReg - 1;
1305 IntReg >= (
unsigned)MVT::i1; --IntReg) {
1308 LegalIntReg = IntReg;
1310 RegisterTypeForVT[IntReg] = TransformToType[IntReg] =
1316// ppcf128 type is really two f64's. 1319 NumRegistersForVT[MVT::ppcf128] = 2*NumRegistersForVT[MVT::f64];
1320 RegisterTypeForVT[MVT::ppcf128] = MVT::f64;
1321 TransformToType[MVT::ppcf128] = MVT::f64;
1324 NumRegistersForVT[MVT::ppcf128] = NumRegistersForVT[MVT::i128];
1325 RegisterTypeForVT[MVT::ppcf128] = RegisterTypeForVT[MVT::i128];
1326 TransformToType[MVT::ppcf128] = MVT::i128;
1331// Decide how to handle f128. If the target does not have native f128 support, 1332// expand it to i128 and we will be generating soft float library calls. 1334 NumRegistersForVT[MVT::f128] = NumRegistersForVT[MVT::i128];
1335 RegisterTypeForVT[MVT::f128] = RegisterTypeForVT[MVT::i128];
1336 TransformToType[MVT::f128] = MVT::i128;
1340// Decide how to handle f80. If the target does not have native f80 support, 1341// expand it to i96 and we will be generating soft float library calls. 1343 NumRegistersForVT[MVT::f80] = 3*NumRegistersForVT[MVT::i32];
1344 RegisterTypeForVT[MVT::f80] = RegisterTypeForVT[MVT::i32];
1345 TransformToType[MVT::f80] = MVT::i32;
1349// Decide how to handle f64. If the target does not have native f64 support, 1350// expand it to i64 and we will be generating soft float library calls. 1352 NumRegistersForVT[MVT::f64] = NumRegistersForVT[MVT::i64];
1353 RegisterTypeForVT[MVT::f64] = RegisterTypeForVT[MVT::i64];
1354 TransformToType[MVT::f64] = MVT::i64;
1358// Decide how to handle f32. If the target does not have native f32 support, 1359// expand it to i32 and we will be generating soft float library calls. 1361 NumRegistersForVT[MVT::f32] = NumRegistersForVT[MVT::i32];
1362 RegisterTypeForVT[MVT::f32] = RegisterTypeForVT[MVT::i32];
1363 TransformToType[MVT::f32] = MVT::i32;
1367// Decide how to handle f16. If the target does not have native f16 support, 1368// promote it to f32, because there are no f16 library calls (except for 1371// Allow targets to control how we legalize half. 1375if (!UseFPRegsForHalfType) {
1376 NumRegistersForVT[MVT::f16] = NumRegistersForVT[MVT::i16];
1377 RegisterTypeForVT[MVT::f16] = RegisterTypeForVT[MVT::i16];
1379 NumRegistersForVT[MVT::f16] = NumRegistersForVT[MVT::f32];
1380 RegisterTypeForVT[MVT::f16] = RegisterTypeForVT[MVT::f32];
1382 TransformToType[MVT::f16] = MVT::f32;
1383if (SoftPromoteHalfType) {
1390// Decide how to handle bf16. If the target does not have native bf16 support, 1391// promote it to f32, because there are no bf16 library calls (except for 1392// converting from f32 to bf16). 1394 NumRegistersForVT[MVT::bf16] = NumRegistersForVT[MVT::f32];
1395 RegisterTypeForVT[MVT::bf16] = RegisterTypeForVT[MVT::f32];
1396 TransformToType[MVT::bf16] = MVT::f32;
1400// Loop over all of the vector value types to see which need transformations. 1401for (
unsigned i = MVT::FIRST_VECTOR_VALUETYPE;
1402 i <= (
unsigned)MVT::LAST_VECTOR_VALUETYPE; ++i) {
1409bool IsLegalWiderType =
false;
1412switch (PreferredAction) {
1415 MVT::LAST_INTEGER_SCALABLE_VECTOR_VALUETYPE :
1416 MVT::LAST_INTEGER_FIXEDLEN_VECTOR_VALUETYPE;
1417// Try to promote the elements of integer vectors. If no legal 1418// promotion was found, fall through to the widen-vector method. 1419for (
unsigned nVT = i + 1;
1422// Promote vectors of integers to vectors with the same number 1423// of elements, with a wider element type. 1426 TransformToType[i] = SVT;
1427 RegisterTypeForVT[i] = SVT;
1428 NumRegistersForVT[i] = 1;
1430 IsLegalWiderType =
true;
1434if (IsLegalWiderType)
1441// Try to widen the vector. 1442for (
unsigned nVT = i + 1; nVT <= MVT::LAST_VECTOR_VALUETYPE; ++nVT) {
1447 EC.getKnownMinValue() &&
1449 TransformToType[i] = SVT;
1450 RegisterTypeForVT[i] = SVT;
1451 NumRegistersForVT[i] = 1;
1453 IsLegalWiderType =
true;
1457if (IsLegalWiderType)
1460// Only widen to the next power of 2 to keep consistency with EVT. 1463 TransformToType[i] = NVT;
1465 RegisterTypeForVT[i] = NVT;
1466 NumRegistersForVT[i] = 1;
1476unsigned NumIntermediates;
1478 NumIntermediates, RegisterVT,
this);
1479 NumRegistersForVT[i] = NumRegisters;
1480assert(NumRegistersForVT[i] == NumRegisters &&
1481"NumRegistersForVT size cannot represent NumRegisters!");
1482 RegisterTypeForVT[i] = RegisterVT;
1486// Type is already a power of 2. The default action is to split. 1487 TransformToType[i] = MVT::Other;
1492elseif (EC.getKnownMinValue() > 1)
1499 TransformToType[i] = NVT;
1509// Determine the 'representative' register class for each value type. 1510// An representative register class is the largest (meaning one which is 1511// not a sub-register class / subreg register class) legal register class for 1512// a group of value types. For example, on i386, i8, i16, and i32 1513// representative would be GR32; while on x86_64 it's GR64. 1518 RepRegClassForVT[i] = RRC;
1519 RepRegClassCostForVT[i] =
Cost;
1530return MVT::i32;
// return the default value 1533/// getVectorTypeBreakdown - Vector types are broken down into some number of 1534/// legal first class types. For example, MVT::v8f32 maps to 2 MVT::v4f32 1535/// with Altivec or SSE1, or 8 promoted MVT::f64 values with the X86 FP stack. 1536/// Similarly, MVT::v2i64 turns into 4 MVT::i32 values with both PPC and X86. 1538/// This method returns the number of registers needed, and the VT for each 1539/// register. It also returns the VT and quantity of the intermediate values 1540/// before they are promoted/expanded. 1542EVT VT,
EVT &IntermediateVT,
1543unsigned &NumIntermediates,
1544MVT &RegisterVT)
const{
1547// If there is a wider vector type with the same element type as this one, 1548// or a promoted vector type that has the same number of elements which 1549// are wider, then we should convert to that legal vector type. 1550// This handles things like <2 x float> -> <4 x float> and 1551// <4 x i1> -> <4 x i32>. 1557 IntermediateVT = RegisterEVT;
1559 NumIntermediates = 1;
1564// Figure out the right, legal destination reg to copy into. 1567unsigned NumVectorRegs = 1;
1569// Scalable vectors cannot be scalarized, so handle the legalisation of the 1570// types like done elsewhere in SelectionDAG. 1575// Iterate until we've found a legal (part) type to hold VT. 1582"Don't know how to legalize this scalable vector type");
1588 IntermediateVT = PartVT;
1590return NumIntermediates;
1593// FIXME: We don't support non-power-of-2-sized vectors for now. Ideally 1594// we could break down into LHS/RHS like LegalizeDAG does. 1600// Divide the input until we get to a supported size. This will always 1601// end with a scalar if the target doesn't support vectors. 1605 NumVectorRegs <<= 1;
1608 NumIntermediates = NumVectorRegs;
1613 IntermediateVT = NewVT;
1616 RegisterVT = DestVT;
1618if (
EVT(DestVT).bitsLT(NewVT)) {
// Value is expanded, e.g. i64 -> i16. 1620// Convert sizes such as i33 to i64. 1626// Otherwise, promotion or legal types use the same number of registers as 1627// the vector decimated to the appropriate level. 1628return NumVectorRegs;
1636// FIXME: This function check the maximum table size and density, but the 1637// minimum size is not checked. It would be nice if the minimum size is 1638// also combined within this function. Currently, the minimum size check is 1639// performed in findJumpTable() in SelectionDAGBuiler and 1640// getEstimatedNumberOfCaseClusters() in BasicTTIImpl. 1641constbool OptForSize =
1646// Check whether the number of cases is small enough and 1647// the range is dense enough for a jump table. 1648return (OptForSize ||
Range <= MaxJumpTableSize) &&
1649 (NumCases * 100 >=
Range * MinDensity);
1653EVT ConditionVT)
const{
1657/// Get the EVTs and ArgFlags collections that represent the legalized return 1658/// type of the given function. This does not require a DAG or a return value, 1659/// and is suitable for use before any DAGs for the function are constructed. 1660/// TODO: Move this out of TargetLowering.cpp. 1667unsigned NumValues = ValueVTs.
size();
1668if (NumValues == 0)
return;
1670for (
unsigned j = 0, f = NumValues; j != f; ++j) {
1671EVT VT = ValueVTs[j];
1687// 'inreg' on function refers to return value 1692// Propagate extension type if any 1698for (
unsigned i = 0; i < NumParts; ++i)
1705returnDL.getABITypeAlign(Ty);
1711// Check if the specified alignment is sufficient based on the data layout. 1712// TODO: While using the data layout works in practice, a better solution 1713// would be to implement this check directly (make this a virtual function). 1714// For example, the ABI alignment may change based on software platform while 1715// this function should only be affected by hardware implementation. 1717if (VT.
isZeroSized() || Alignment >=
DL.getABITypeAlign(Ty)) {
1718// Assume that an access that meets the ABI-specified alignment is fast. 1724// This is a misaligned access. 1737unsigned AddrSpace,
Align Alignment,
1739unsigned *
Fast)
const{
1747unsigned *
Fast)
const{
1755unsigned *
Fast)
const{
1761//===----------------------------------------------------------------------===// 1762// TargetTransformInfo Helpers 1763//===----------------------------------------------------------------------===// 1766enum InstructionOpcodes {
1767#define HANDLE_INST(NUM, OPCODE, CLASS) OPCODE = NUM, 1768#define LAST_OTHER_INST(NUM) InstructionOpcodesCount = NUM 1769#include "llvm/IR/Instruction.def" 1771switch (
static_cast<InstructionOpcodes
>(Opcode)) {
1774case Switch:
return 0;
1775case IndirectBr:
return 0;
1776case Invoke:
return 0;
1777case CallBr:
return 0;
1778case Resume:
return 0;
1779case Unreachable:
return 0;
1780case CleanupRet:
return 0;
1781case CatchRet:
return 0;
1782case CatchPad:
return 0;
1783case CatchSwitch:
return 0;
1784case CleanupPad:
return 0;
1804case Alloca:
return 0;
1807case GetElementPtr:
return 0;
1809case AtomicCmpXchg:
return 0;
1810case AtomicRMW:
return 0;
1829case UserOp1:
return 0;
1830case UserOp2:
return 0;
1837case LandingPad:
return 0;
1848case Intrinsic::exp2:
1858// compiler-rt provides a variable with a magic name. Targets that do not 1859// link with compiler-rt may also provide such a variable. 1861constchar *UnsafeStackPtrVar =
"__safestack_unsafe_stack_ptr";
1862auto UnsafeStackPtr =
1863 dyn_cast_or_null<GlobalVariable>(M->getNamedValue(UnsafeStackPtrVar));
1866PointerType *StackPtrTy =
DL.getAllocaPtrType(M->getContext());
1868if (!UnsafeStackPtr) {
1869auto TLSModel = UseTLS ?
1872// The global variable is not defined yet, define it ourselves. 1873// We use the initial-exec TLS model because we do not support the 1874// variable living anywhere other than in the main executable. 1877 UnsafeStackPtrVar,
nullptr, TLSModel);
1879// The variable exists, check its type and attributes. 1881// FIXME: Move to IR verifier. 1882if (UnsafeStackPtr->getValueType() != StackPtrTy)
1884if (UseTLS != UnsafeStackPtr->isThreadLocal())
1886 (UseTLS ?
"" :
"not ") +
"be thread-local");
1888return UnsafeStackPtr;
1896// Android provides a libc function to retrieve the address of the current 1897// thread's unsafe stack pointer. 1901 M->getOrInsertFunction(
"__safestack_pointer_address", PtrTy);
1905//===----------------------------------------------------------------------===// 1906// Loop Strength Reduction hooks 1907//===----------------------------------------------------------------------===// 1909/// isLegalAddressingMode - Return true if the addressing mode represented 1910/// by AM is legal for this target, for a load/store of the specified type. 1914// The default implementation of this implements a conservative RISCy, r+r and 1917// Scalable offsets not supported 1921// Allows a sign-extended 16-bit immediate field. 1925// No global is ever allowed as a base. 1931case 0:
// "r+i" or just "i", depending on HasBaseReg. 1936// Otherwise we have r+r or r+i. 1943default:
// Don't allow n * r 1950//===----------------------------------------------------------------------===// 1952//===----------------------------------------------------------------------===// 1954// For OpenBSD return its special guard variable. Otherwise return nullptr, 1955// so that SelectionDAG handle SSP. 1960Constant *
C = M.getOrInsertGlobal(
"__guard_local", PtrTy);
1968// Currently only support "standard" __stack_chk_guard. 1969// TODO: add LOAD_STACK_GUARD support. 1971if (!M.getNamedValue(
"__stack_chk_guard")) {
1974nullptr,
"__stack_chk_guard");
1976// FreeBSD has "__stack_chk_guard" defined externally on libc.so 1977if (M.getDirectAccessExternalData() &&
1983 GV->setDSOLocal(
true);
1987// Currently only support "standard" __stack_chk_guard. 1988// TODO: add LOAD_STACK_GUARD support. 1990return M.getNamedValue(
"__stack_chk_guard");
2024return PrefLoopAlignment;
2029return MaxBytesForAlignment;
2032//===----------------------------------------------------------------------===// 2033// Reciprocal Estimates 2034//===----------------------------------------------------------------------===// 2036/// Get the reciprocal estimate attribute string for a function that will 2037/// override the target defaults. 2040returnF.getFnAttribute(
"reciprocal-estimates").getValueAsString();
2043/// Construct a string for the given reciprocal operation of the given type. 2044/// This string should match the corresponding option to the front-end's 2045/// "-mrecip" flag assuming those strings have been passed through in an 2046/// attribute string. For example, "vec-divf" for a division of a vXf32. 2050Name += IsSqrt ?
"sqrt" :
"div";
2052// TODO: Handle other float types? 2059"Unexpected FP type for reciprocal estimate");
2066/// Return the character position and value (a single numeric character) of a 2067/// customized refinement operation in the input string if it exists. Return 2068/// false if there is no customized refinement step count. 2071constchar RefStepToken =
':';
2072 Position = In.find(RefStepToken);
2076StringRef RefStepString = In.substr(Position + 1);
2077// Allow exactly one numeric character for the additional refinement 2079if (RefStepString.
size() == 1) {
2080char RefStepChar = RefStepString[0];
2082Value = RefStepChar -
'0';
2089/// For the input attribute string, return one of the ReciprocalEstimate enum 2090/// status values (enabled, disabled, or not specified) for this operation on 2091/// the specified data type. 2093if (Override.
empty())
2097 Override.
split(OverrideVector,
',');
2098unsigned NumArgs = OverrideVector.
size();
2100// Check if "all", "none", or "default" was specified. 2102// Look for an optional setting of the number of refinement steps needed 2103// for this type of reciprocal operation. 2107// Split the string for further processing. 2108 Override = Override.
substr(0, RefPos);
2111// All reciprocal types are enabled. 2112if (Override ==
"all")
2115// All reciprocal types are disabled. 2116if (Override ==
"none")
2119// Target defaults for enablement are used. 2120if (Override ==
"default")
2124// The attribute string may omit the size suffix ('f'/'d'). 2126 std::string VTNameNoSize = VTName;
2127 VTNameNoSize.pop_back();
2128staticconstchar DisabledPrefix =
'!';
2130for (
StringRef RecipType : OverrideVector) {
2134 RecipType = RecipType.substr(0, RefPos);
2136// Ignore the disablement token for string matching. 2137bool IsDisabled = RecipType[0] == DisabledPrefix;
2139 RecipType = RecipType.substr(1);
2141if (RecipType == VTName || RecipType == VTNameNoSize)
2149/// For the input attribute string, return the customized refinement step count 2150/// for this operation on the specified data type. If the step count does not 2151/// exist, return the ReciprocalEstimate enum value for unspecified. 2153if (Override.
empty())
2157 Override.
split(OverrideVector,
',');
2158unsigned NumArgs = OverrideVector.
size();
2160// Check if "all", "default", or "none" was specified. 2162// Look for an optional setting of the number of refinement steps needed 2163// for this type of reciprocal operation. 2169// Split the string for further processing. 2170 Override = Override.
substr(0, RefPos);
2171assert(Override !=
"none" &&
2172"Disabled reciprocals, but specifed refinement steps?");
2174// If this is a general override, return the specified number of steps. 2175if (Override ==
"all" || Override ==
"default")
2179// The attribute string may omit the size suffix ('f'/'d'). 2181 std::string VTNameNoSize = VTName;
2182 VTNameNoSize.pop_back();
2184for (
StringRef RecipType : OverrideVector) {
2190 RecipType = RecipType.substr(0, RefPos);
2191if (RecipType == VTName || RecipType == VTNameNoSize)
2221// Single-element vectors are scalarized, so we should generally avoid having 2222// any memory operations on such types, as they would get scalarized too. 2227// Don't do if we could do an indexed load on the original type, but not on 2234// Don't bother doing this if it's just going to be promoted again later, as 2235// doing so might interfere with other combines. 2260if (LI.
hasMetadata(LLVMContext::MD_invariant_load))
2265/*DT=*/nullptr, LibInfo))
2280if (SI.hasMetadata(LLVMContext::MD_nontemporal))
2283// FIXME: Not preserving dereferenceable 2293if (
constAtomicRMWInst *RMW = dyn_cast<AtomicRMWInst>(&AI)) {
2294if (RMW->isVolatile())
2297if (CmpX->isVolatile())
2302// FIXME: Not preserving dereferenceable 2325//===----------------------------------------------------------------------===// 2327//===----------------------------------------------------------------------===// 2331auto &MF = *
MI.getMF();
2332auto &
MRI = MF.getRegInfo();
2333// Assuming a spill and reload of a value has a cost of 1 instruction each, 2334// this helper function computes the maximum number of uses we should consider 2335// for remat. E.g. on arm64 global addresses take 2 insts to materialize. We 2336// break even in terms of code size when the original MI has 2 users vs 2337// choosing to potentially spill. Any more than 2 users we we have a net code 2338// size increase. This doesn't take into account register pressure though. 2339auto maxUses = [](
unsigned RematCost) {
2340// A cost of 1 means remats are basically free. 2342return std::numeric_limits<unsigned>::max();
2346// Remat is too expensive, only sink if there's one user. 2352switch (
MI.getOpcode()) {
2355// Constants-like instructions should be close to their users. 2356// We don't want long live-ranges for them. 2357case TargetOpcode::G_CONSTANT:
2358case TargetOpcode::G_FCONSTANT:
2359case TargetOpcode::G_FRAME_INDEX:
2360case TargetOpcode::G_INTTOPTR:
2362case TargetOpcode::G_GLOBAL_VALUE: {
2365unsigned MaxUses = maxUses(RematCost);
2366if (MaxUses == UINT_MAX)
2367returntrue;
// Remats are "free" so always localize. 2368returnMRI.hasAtMostUserInstrs(Reg, MaxUses);
unsigned const MachineRegisterInfo * MRI
AMDGPU Register Bank Select
MachineBasicBlock MachineBasicBlock::iterator DebugLoc DL
This file contains the simple types necessary to represent the attributes associated with functions a...
This file implements the BitVector class.
Module.h This file contains the declarations for the Module class.
unsigned const TargetRegisterInfo * TRI
ConstantRange Range(APInt(BitWidth, Low), APInt(BitWidth, High))
static bool isDigit(const char C)
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
This file contains some templates that are useful if you are working with the STL at all.
This file defines the SmallVector class.
This file contains some functions that are useful when dealing with strings.
static cl::opt< bool > JumpIsExpensiveOverride("jump-is-expensive", cl::init(false), cl::desc("Do not create extra branches to split comparison logic."), cl::Hidden)
#define OP_TO_LIBCALL(Name, Enum)
static cl::opt< unsigned > MinimumJumpTableEntries("min-jump-table-entries", cl::init(4), cl::Hidden, cl::desc("Set minimum number of entries to use a jump table."))
static cl::opt< bool > DisableStrictNodeMutation("disable-strictnode-mutation", cl::desc("Don't mutate strict-float node to a legalize node"), cl::init(false), cl::Hidden)
static bool parseRefinementStep(StringRef In, size_t &Position, uint8_t &Value)
Return the character position and value (a single numeric character) of a customized refinement opera...
static cl::opt< unsigned > MaximumJumpTableSize("max-jump-table-size", cl::init(UINT_MAX), cl::Hidden, cl::desc("Set maximum size of jump tables."))
static cl::opt< unsigned > JumpTableDensity("jump-table-density", cl::init(10), cl::Hidden, cl::desc("Minimum density for building a jump table in " "a normal function"))
Minimum jump table density for normal functions.
static unsigned getVectorTypeBreakdownMVT(MVT VT, MVT &IntermediateVT, unsigned &NumIntermediates, MVT &RegisterVT, TargetLoweringBase *TLI)
static std::string getReciprocalOpName(bool IsSqrt, EVT VT)
Construct a string for the given reciprocal operation of the given type.
static int getOpRefinementSteps(bool IsSqrt, EVT VT, StringRef Override)
For the input attribute string, return the customized refinement step count for this operation on the...
static int getOpEnabled(bool IsSqrt, EVT VT, StringRef Override)
For the input attribute string, return one of the ReciprocalEstimate enum status values (enabled,...
static StringRef getRecipEstimateForFunc(MachineFunction &MF)
Get the reciprocal estimate attribute string for a function that will override the target defaults.
static cl::opt< unsigned > OptsizeJumpTableDensity("optsize-jump-table-density", cl::init(40), cl::Hidden, cl::desc("Minimum density for building a jump table in " "an optsize function"))
Minimum jump table density for -Os or -Oz functions.
This file describes how to lower LLVM code to machine code.
This pass exposes codegen information to IR-level passes.
Class for arbitrary precision integers.
A cache of @llvm.assume calls within a function.
An instruction that atomically checks whether a specified value is in a memory location,...
an instruction that atomically reads a memory location, combines it with another value,...
bool hasRetAttr(Attribute::AttrKind Kind) const
Return true if the attribute exists for the return value.
const Function * getParent() const
Return the enclosing method, or null if none.
void setBitsInMask(const uint32_t *Mask, unsigned MaskWords=~0u)
setBitsInMask - Add '1' bits from Mask to this vector.
iterator_range< const_set_bits_iterator > set_bits() const
BlockFrequencyInfo pass uses BlockFrequencyInfoImpl implementation to estimate IR basic block frequen...
This class represents a range of values.
unsigned getActiveBits() const
Compute the maximal number of active bits needed to represent every value in this range.
ConstantRange umul_sat(const ConstantRange &Other) const
Perform an unsigned saturating multiplication of two constant ranges.
ConstantRange subtract(const APInt &CI) const
Subtract the specified constant from the endpoints of this constant range.
This is an important base class in LLVM.
This class represents an Operation in the Expression.
A parsed version of the target data layout string in and methods for querying it.
unsigned getPointerSize(unsigned AS=0) const
Layout pointer size in bytes, rounded up to a whole number of bytes.
static constexpr ElementCount getScalable(ScalarTy MinVal)
static constexpr ElementCount getFixed(ScalarTy MinVal)
constexpr bool isScalar() const
Exactly one element.
A handy container for a FunctionType+Callee-pointer pair, which can be passed around as a single enti...
Module * getParent()
Get the module that this global value is contained inside of...
@ HiddenVisibility
The GV is hidden.
@ ExternalLinkage
Externally visible function.
Common base class shared among various IRBuilders.
FenceInst * CreateFence(AtomicOrdering Ordering, SyncScope::ID SSID=SyncScope::System, const Twine &Name="")
BasicBlock * GetInsertBlock() const
CallInst * CreateCall(FunctionType *FTy, Value *Callee, ArrayRef< Value * > Args={}, const Twine &Name="", MDNode *FPMathTag=nullptr)
bool hasAtomicStore() const LLVM_READONLY
Return true if this atomic instruction stores to memory.
bool hasMetadata() const
Return true if this instruction has any metadata attached to it.
@ MAX_INT_BITS
Maximum number of bits that can be specified.
This is an important class for using LLVM in a threaded context.
An instruction for reading from memory.
Value * getPointerOperand()
bool isVolatile() const
Return true if this is a load from a volatile memory location.
Align getAlign() const
Return the alignment of the access that is being performed.
uint64_t getScalarSizeInBits() const
bool isVector() const
Return true if this is a vector value type.
bool isScalableVector() const
Return true if this is a vector value type where the runtime length is machine dependent.
static auto all_valuetypes()
SimpleValueType Iteration.
TypeSize getSizeInBits() const
Returns the size of the specified MVT in bits.
uint64_t getFixedSizeInBits() const
Return the size of the specified fixed width value type in bits.
ElementCount getVectorElementCount() const
bool isScalarInteger() const
Return true if this is an integer, not including vectors.
static MVT getVectorVT(MVT VT, unsigned NumElements)
MVT getVectorElementType() const
bool isValid() const
Return true if this is a valid simple valuetype.
static MVT getIntegerVT(unsigned BitWidth)
static auto fp_valuetypes()
MVT getPow2VectorType() const
Widens the length of the given vector MVT up to the nearest power of 2 and returns that type.
instr_iterator insert(instr_iterator I, MachineInstr *M)
Insert MI into the instruction list before I, possibly inside a bundle.
The MachineFrameInfo class represents an abstract stack frame until prolog/epilog code is inserted.
bool isStatepointSpillSlotObjectIndex(int ObjectIdx) const
Align getObjectAlign(int ObjectIdx) const
Return the alignment of the specified stack object.
int64_t getObjectSize(int ObjectIdx) const
Return the size of the specified object.
int64_t getObjectOffset(int ObjectIdx) const
Return the assigned stack offset of the specified object from the incoming stack pointer.
MachineMemOperand * getMachineMemOperand(MachinePointerInfo PtrInfo, MachineMemOperand::Flags f, LLT MemTy, Align base_alignment, const AAMDNodes &AAInfo=AAMDNodes(), const MDNode *Ranges=nullptr, SyncScope::ID SSID=SyncScope::System, AtomicOrdering Ordering=AtomicOrdering::NotAtomic, AtomicOrdering FailureOrdering=AtomicOrdering::NotAtomic)
getMachineMemOperand - Allocate a new MachineMemOperand.
MachineFrameInfo & getFrameInfo()
getFrameInfo - Return the frame info object for the current function.
MachineRegisterInfo & getRegInfo()
getRegInfo - Return information about the registers currently in use.
const DataLayout & getDataLayout() const
Return the DataLayout attached to the Module associated to this MF.
Function & getFunction()
Return the LLVM function that this machine code represents.
const MachineInstrBuilder & addImm(int64_t Val) const
Add a new immediate operand.
const MachineInstrBuilder & add(const MachineOperand &MO) const
const MachineInstrBuilder & cloneMemRefs(const MachineInstr &OtherMI) const
Representation of each machine instruction.
unsigned getNumOperands() const
Retuns the total number of operands.
bool mayLoad(QueryType Type=AnyInBundle) const
Return true if this instruction could possibly read memory.
void tieOperands(unsigned DefIdx, unsigned UseIdx)
Add a tie between the register operands at DefIdx and UseIdx.
void addMemOperand(MachineFunction &MF, MachineMemOperand *MO)
Add a MachineMemOperand to the machine instruction.
A description of a memory reference used in the backend.
unsigned getAddrSpace() const
Flags
Flags values. These may be or'd together.
@ MOVolatile
The memory access is volatile.
@ MODereferenceable
The memory access is dereferenceable (i.e., doesn't trap).
@ MOLoad
The memory access reads data.
@ MONonTemporal
The memory access is non-temporal.
@ MOInvariant
The memory access always returns the same value (or traps).
@ MOStore
The memory access writes data.
Flags getFlags() const
Return the raw flags of the source value,.
Align getAlign() const
Return the minimum known alignment in bytes of the actual memory reference.
MachineOperand class - Representation of each machine instruction operand.
bool isReg() const
isReg - Tests if this is a MO_Register operand.
bool isFI() const
isFI - Tests if this is a MO_FrameIndex operand.
void freezeReservedRegs()
freezeReservedRegs - Called by the register allocator to freeze the set of reserved registers before ...
A Module instance is used to store all the information related to an LLVM module.
Class to represent pointers.
static PointerType * getUnqual(Type *ElementType)
This constructs a pointer to an object of the specified type in the default address space (address sp...
Analysis providing profile information.
Wrapper class representing virtual and physical registers.
This is used to represent a portion of an LLVM function in a low-level Data Dependence DAG representa...
const DataLayout & getDataLayout() const
LLVMContext * getContext() const
This class consists of common code factored out of the SmallVector class to reduce code duplication b...
void push_back(const T &Elt)
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
An instruction for storing to memory.
StringRef - Represent a constant reference to a string, i.e.
std::pair< StringRef, StringRef > split(char Separator) const
Split into two substrings around the first occurrence of a separator character.
constexpr StringRef substr(size_t Start, size_t N=npos) const
Return a reference to the substring from [Start, Start + N).
constexpr bool empty() const
empty - Check if the string is empty.
constexpr size_t size() const
size - Get the string size.
static constexpr size_t npos
bool isValid() const
Returns true if this iterator is still pointing at a valid entry.
Provides information about what library functions are available for the current target.
LegalizeTypeAction getTypeAction(MVT VT) const
void setTypeAction(MVT VT, LegalizeTypeAction Action)
This base class for TargetLowering contains the SelectionDAG-independent parts that can be used from ...
virtual Align getByValTypeAlignment(Type *Ty, const DataLayout &DL) const
Returns the desired alignment for ByVal or InAlloca aggregate function arguments in the caller parame...
int InstructionOpcodeToISD(unsigned Opcode) const
Get the ISD node that corresponds to the Instruction class opcode.
void setOperationAction(unsigned Op, MVT VT, LegalizeAction Action)
Indicate that the specified operation does not work with the specified type and indicate what to do a...
virtual void finalizeLowering(MachineFunction &MF) const
Execute target specific actions to finalize target lowering.
void initActions()
Initialize all of the actions to default values.
bool PredictableSelectIsExpensive
Tells the code generator that select is more expensive than a branch if the branch is usually predict...
unsigned MaxStoresPerMemcpyOptSize
Likewise for functions with the OptSize attribute.
MachineBasicBlock * emitPatchPoint(MachineInstr &MI, MachineBasicBlock *MBB) const
Replace/modify any TargetFrameIndex operands with a targte-dependent sequence of memory operands that...
virtual Value * getSafeStackPointerLocation(IRBuilderBase &IRB) const
Returns the target-specific address of the unsafe stack pointer.
int getRecipEstimateSqrtEnabled(EVT VT, MachineFunction &MF) const
Return a ReciprocalEstimate enum value for a square root of the given type based on the function's at...
virtual bool canOpTrap(unsigned Op, EVT VT) const
Returns true if the operation can trap for the value type.
virtual bool shouldLocalize(const MachineInstr &MI, const TargetTransformInfo *TTI) const
Check whether or not MI needs to be moved close to its uses.
virtual unsigned getMaxPermittedBytesForAlignment(MachineBasicBlock *MBB) const
Return the maximum amount of bytes allowed to be emitted when padding for alignment.
void setMaximumJumpTableSize(unsigned)
Indicate the maximum number of entries in jump tables.
virtual unsigned getMinimumJumpTableEntries() const
Return lower limit for number of blocks in a jump table.
const TargetMachine & getTargetMachine() const
unsigned MaxLoadsPerMemcmp
Specify maximum number of load instructions per memcmp call.
virtual unsigned getNumRegistersForCallingConv(LLVMContext &Context, CallingConv::ID CC, EVT VT) const
Certain targets require unusual breakdowns of certain types.
virtual MachineMemOperand::Flags getTargetMMOFlags(const Instruction &I) const
This callback is used to inspect load/store instructions and add target-specific MachineMemOperand fl...
unsigned MaxGluedStoresPerMemcpy
Specify max number of store instructions to glue in inlined memcpy.
virtual MVT getRegisterTypeForCallingConv(LLVMContext &Context, CallingConv::ID CC, EVT VT) const
Certain combinations of ABIs, Targets and features require that types are legal for some operations a...
LegalizeTypeAction
This enum indicates whether a types are legal for a target, and if not, what action should be used to...
@ TypeScalarizeScalableVector
virtual bool isSuitableForJumpTable(const SwitchInst *SI, uint64_t NumCases, uint64_t Range, ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI) const
Return true if lowering to a jump table is suitable for a set of case clusters which may contain NumC...
void setIndexedMaskedLoadAction(unsigned IdxMode, MVT VT, LegalizeAction Action)
Indicate that the specified indexed masked load does or does not work with the specified type and ind...
virtual Value * getSDagStackGuard(const Module &M) const
Return the variable that's previously inserted by insertSSPDeclarations, if any, otherwise return nul...
virtual bool useFPRegsForHalfType() const
virtual bool isLoadBitCastBeneficial(EVT LoadVT, EVT BitcastVT, const SelectionDAG &DAG, const MachineMemOperand &MMO) const
Return true if the following transform is beneficial: fold (conv (load x)) -> (load (conv*)x) On arch...
void setIndexedLoadAction(ArrayRef< unsigned > IdxModes, MVT VT, LegalizeAction Action)
Indicate that the specified indexed load does or does not work with the specified type and indicate w...
virtual bool softPromoteHalfType() const
unsigned getMaximumJumpTableSize() const
Return upper limit for number of entries in a jump table.
virtual MVT::SimpleValueType getCmpLibcallReturnType() const
Return the ValueType for comparison libcalls.
unsigned getBitWidthForCttzElements(Type *RetTy, ElementCount EC, bool ZeroIsPoison, const ConstantRange *VScaleRange) const
Return the minimum number of bits required to hold the maximum possible number of trailing zero vecto...
bool isLegalRC(const TargetRegisterInfo &TRI, const TargetRegisterClass &RC) const
Return true if the value types that can be represented by the specified register class are all legal.
virtual TargetLoweringBase::LegalizeTypeAction getPreferredVectorAction(MVT VT) const
Return the preferred vector type legalization action.
void setAtomicLoadExtAction(unsigned ExtType, MVT ValVT, MVT MemVT, LegalizeAction Action)
Let target indicate that an extending atomic load of the specified type is legal.
Value * getDefaultSafeStackPointerLocation(IRBuilderBase &IRB, bool UseTLS) const
virtual Function * getSSPStackGuardCheck(const Module &M) const
If the target has a standard stack protection check function that performs validation and error handl...
MachineMemOperand::Flags getAtomicMemOperandFlags(const Instruction &AI, const DataLayout &DL) const
virtual bool allowsMisalignedMemoryAccesses(EVT, unsigned AddrSpace=0, Align Alignment=Align(1), MachineMemOperand::Flags Flags=MachineMemOperand::MONone, unsigned *=nullptr) const
Determine if the target supports unaligned memory accesses.
unsigned MaxStoresPerMemsetOptSize
Likewise for functions with the OptSize attribute.
EVT getShiftAmountTy(EVT LHSTy, const DataLayout &DL) const
Returns the type for the shift amount of a shift opcode.
unsigned MaxStoresPerMemmove
Specify maximum number of store instructions per memmove call.
virtual Align getPrefLoopAlignment(MachineLoop *ML=nullptr) const
Return the preferred loop alignment.
void computeRegisterProperties(const TargetRegisterInfo *TRI)
Once all of the register classes are added, this allows us to compute derived properties we expose.
int getDivRefinementSteps(EVT VT, MachineFunction &MF) const
Return the refinement step count for a division of the given type based on the function's attributes.
virtual EVT getSetCCResultType(const DataLayout &DL, LLVMContext &Context, EVT VT) const
Return the ValueType of the result of SETCC operations.
MachineMemOperand::Flags getLoadMemOperandFlags(const LoadInst &LI, const DataLayout &DL, AssumptionCache *AC=nullptr, const TargetLibraryInfo *LibInfo=nullptr) const
virtual EVT getTypeToTransformTo(LLVMContext &Context, EVT VT) const
For types supported by the target, this is an identity function.
unsigned MaxStoresPerMemmoveOptSize
Likewise for functions with the OptSize attribute.
virtual Value * getIRStackGuard(IRBuilderBase &IRB) const
If the target has a standard location for the stack protector guard, returns the address of that loca...
virtual MVT getPreferredSwitchConditionType(LLVMContext &Context, EVT ConditionVT) const
Returns preferred type for switch condition.
bool isTypeLegal(EVT VT) const
Return true if the target has native support for the specified value type.
bool EnableExtLdPromotion
int getRecipEstimateDivEnabled(EVT VT, MachineFunction &MF) const
Return a ReciprocalEstimate enum value for a division of the given type based on the function's attri...
void setIndexedStoreAction(ArrayRef< unsigned > IdxModes, MVT VT, LegalizeAction Action)
Indicate that the specified indexed store does or does not work with the specified type and indicate ...
virtual bool isJumpTableRelative() const
virtual MVT getScalarShiftAmountTy(const DataLayout &, EVT) const
Return the type to use for a scalar shift opcode, given the shifted amount type.
virtual MVT getPointerTy(const DataLayout &DL, uint32_t AS=0) const
Return the pointer type for the given address space, defaults to the pointer type from the data layou...
virtual bool isFreeAddrSpaceCast(unsigned SrcAS, unsigned DestAS) const
Returns true if a cast from SrcAS to DestAS is "cheap", such that e.g.
void setIndexedMaskedStoreAction(unsigned IdxMode, MVT VT, LegalizeAction Action)
Indicate that the specified indexed masked store does or does not work with the specified type and in...
unsigned MaxStoresPerMemset
Specify maximum number of store instructions per memset call.
void setMinimumJumpTableEntries(unsigned Val)
Indicate the minimum number of blocks to generate jump tables.
void setTruncStoreAction(MVT ValVT, MVT MemVT, LegalizeAction Action)
Indicate that the specified truncating store does not work with the specified type and indicate what ...
@ UndefinedBooleanContent
virtual bool allowsMemoryAccess(LLVMContext &Context, const DataLayout &DL, EVT VT, unsigned AddrSpace=0, Align Alignment=Align(1), MachineMemOperand::Flags Flags=MachineMemOperand::MONone, unsigned *Fast=nullptr) const
Return true if the target supports a memory access of this type for the given address space and align...
unsigned MaxLoadsPerMemcmpOptSize
Likewise for functions with the OptSize attribute.
MachineMemOperand::Flags getStoreMemOperandFlags(const StoreInst &SI, const DataLayout &DL) const
void AddPromotedToType(unsigned Opc, MVT OrigVT, MVT DestVT)
If Opc/OrigVT is specified as being promoted, the promotion code defaults to trying a larger integer/...
unsigned getMinimumJumpTableDensity(bool OptForSize) const
Return lower limit of the density in a jump table.
virtual std::pair< const TargetRegisterClass *, uint8_t > findRepresentativeClass(const TargetRegisterInfo *TRI, MVT VT) const
Return the largest legal super-reg register class of the register class for the specified type and it...
TargetLoweringBase(const TargetMachine &TM)
NOTE: The TargetMachine owns TLOF.
LegalizeKind getTypeConversion(LLVMContext &Context, EVT VT) const
Return pair that represents the legalization kind (first) that needs to happen to EVT (second) in ord...
void setLoadExtAction(unsigned ExtType, MVT ValVT, MVT MemVT, LegalizeAction Action)
Indicate that the specified load with extension does not work with the specified type and indicate wh...
unsigned GatherAllAliasesMaxDepth
Depth that GatherAllAliases should continue looking for chain dependencies when trying to find a more...
int IntrinsicIDToISD(Intrinsic::ID ID) const
Get the ISD node that corresponds to the Intrinsic ID.
LegalizeTypeAction getTypeAction(LLVMContext &Context, EVT VT) const
Return how we should legalize values of this type, either it is already legal (return 'Legal') or we ...
int getSqrtRefinementSteps(EVT VT, MachineFunction &MF) const
Return the refinement step count for a square root of the given type based on the function's attribut...
bool allowsMemoryAccessForAlignment(LLVMContext &Context, const DataLayout &DL, EVT VT, unsigned AddrSpace=0, Align Alignment=Align(1), MachineMemOperand::Flags Flags=MachineMemOperand::MONone, unsigned *Fast=nullptr) const
This function returns true if the memory access is aligned or if the target allows this specific unal...
virtual Instruction * emitTrailingFence(IRBuilderBase &Builder, Instruction *Inst, AtomicOrdering Ord) const
virtual Instruction * emitLeadingFence(IRBuilderBase &Builder, Instruction *Inst, AtomicOrdering Ord) const
Inserts in the IR a target-specific intrinsic specifying a fence.
unsigned MaxStoresPerMemcpy
Specify maximum number of store instructions per memcpy call.
MVT getRegisterType(MVT VT) const
Return the type of registers that this ValueType will eventually require.
virtual void insertSSPDeclarations(Module &M) const
Inserts necessary declarations for SSP (stack protection) purpose.
void setJumpIsExpensive(bool isExpensive=true)
Tells the code generator not to expand logic operations on comparison predicates into separate sequen...
LegalizeAction getOperationAction(unsigned Op, EVT VT) const
Return how this operation should be treated: either it is legal, needs to be promoted to a larger siz...
MVT getTypeToPromoteTo(unsigned Op, MVT VT) const
If the action for this operation is to promote, this method returns the ValueType to promote to.
virtual bool isLegalAddressingMode(const DataLayout &DL, const AddrMode &AM, Type *Ty, unsigned AddrSpace, Instruction *I=nullptr) const
Return true if the addressing mode represented by AM is legal for this target, for a load/store of th...
unsigned getVectorTypeBreakdown(LLVMContext &Context, EVT VT, EVT &IntermediateVT, unsigned &NumIntermediates, MVT &RegisterVT) const
Vector types are broken down into some number of legal first class types.
std::pair< LegalizeTypeAction, EVT > LegalizeKind
LegalizeKind holds the legalization kind that needs to happen to EVT in order to type-legalize it.
This class defines information used to lower LLVM code to legal SelectionDAG operators that the targe...
virtual EVT getTypeForExtReturn(LLVMContext &Context, EVT VT, ISD::NodeType) const
Return the type that should be used to zero or sign extend a zeroext/signext integer return value.
Primary interface to the complete machine description for the target machine.
bool isPositionIndependent() const
virtual bool isNoopAddrSpaceCast(unsigned SrcAS, unsigned DestAS) const
Returns true if a cast between SrcAS and DestAS is a noop.
const Triple & getTargetTriple() const
Reloc::Model getRelocationModel() const
Returns the code generation relocation model.
unsigned LoopAlignment
If greater than 0, override TargetLoweringBase::PrefLoopAlignment.
TargetRegisterInfo base class - We assume that the target defines a static array of TargetRegisterDes...
This pass provides access to the codegen interfaces that are needed for IR-level transformations.
unsigned getGISelRematGlobalCost() const
bool isWindowsGNUEnvironment() const
bool isAndroid() const
Tests whether the target is Android.
bool isPPC64() const
Tests whether the target is 64-bit PowerPC (little and big endian).
bool isOSDarwin() const
Is this a "Darwin" OS (macOS, iOS, tvOS, watchOS, XROS, or DriverKit).
Twine - A lightweight data structure for efficiently representing the concatenation of temporary valu...
The instances of the Type class are immutable: once they are created, they are never changed.
LLVM Value Representation.
Type * getType() const
All values are typed, get the type of this value.
int getNumOccurrences() const
constexpr LeafTy coefficientNextPowerOf2() const
constexpr bool isScalable() const
Returns whether the quantity is scaled by a runtime quantity (vscale).
constexpr ScalarTy getKnownMinValue() const
Returns the minimum value this quantity can represent.
constexpr LeafTy divideCoefficientBy(ScalarTy RHS) const
We do not provide the '/' operator here because division for polynomial types does not work in the sa...
#define llvm_unreachable(msg)
Marks that the current location is not supposed to be reachable.
@ Fast
Attempts to make calls as fast as possible (e.g.
@ C
The default llvm calling convention, compatible with C.
NodeType
ISD::NodeType enum - This enum defines the target-independent operators for a SelectionDAG.
@ SETCC
SetCC operator - This evaluates to a true value iff the condition is true.
@ MERGE_VALUES
MERGE_VALUES - This node takes multiple discrete operands and returns them all as its individual resu...
@ DELETED_NODE
DELETED_NODE - This is an illegal value that is used to catch errors.
@ SET_FPENV
Sets the current floating-point environment.
@ VECREDUCE_SEQ_FADD
Generic reduction nodes.
@ FGETSIGN
INT = FGETSIGN(FP) - Return the sign bit of the specified floating point value as an integer 0/1 valu...
@ SMULFIX
RESULT = [US]MULFIX(LHS, RHS, SCALE) - Perform fixed point multiplication on 2 integers with the same...
@ ADDC
Carry-setting nodes for multiple precision addition and subtraction.
@ RESET_FPENV
Set floating-point environment to default state.
@ FMAD
FMAD - Perform a * b + c, while getting the same result as the separately rounded operations.
@ ADD
Simple integer binary arithmetic operators.
@ LOAD
LOAD and STORE have token chains as their first operand, then the same operands as an LLVM load/store...
@ SMULFIXSAT
Same as the corresponding unsaturated fixed point instructions, but the result is clamped between the...
@ SET_FPMODE
Sets the current dynamic floating-point control modes.
@ ANY_EXTEND
ANY_EXTEND - Used for integer types. The high bits are undefined.
@ VECTOR_FIND_LAST_ACTIVE
@ FATAN2
FATAN2 - atan2, inspired by libm.
@ ATOMIC_CMP_SWAP_WITH_SUCCESS
Val, Success, OUTCHAIN = ATOMIC_CMP_SWAP_WITH_SUCCESS(INCHAIN, ptr, cmp, swap) N.b.
@ SINT_TO_FP
[SU]INT_TO_FP - These operators convert integers (whose interpreted sign depends on the first letter)...
@ CONCAT_VECTORS
CONCAT_VECTORS(VECTOR0, VECTOR1, ...) - Given a number of values of vector type with the same length ...
@ VECREDUCE_FMAX
FMIN/FMAX nodes can have flags, for NaN/NoNaN variants.
@ FADD
Simple binary floating point operators.
@ VECREDUCE_FMAXIMUM
FMINIMUM/FMAXIMUM nodes propatate NaNs and signed zeroes using the llvm.minimum and llvm....
@ ABS
ABS - Determine the unsigned absolute value of a signed integer value of the same bitwidth.
@ RESET_FPMODE
Sets default dynamic floating-point control modes.
@ SIGN_EXTEND_VECTOR_INREG
SIGN_EXTEND_VECTOR_INREG(Vector) - This operator represents an in-register sign-extension of the low ...
@ BITCAST
BITCAST - This operator converts between integer, vector and FP values, as if the value was stored to...
@ FLDEXP
FLDEXP - ldexp, inspired by libm (op0 * 2**op1).
@ SDIVFIX
RESULT = [US]DIVFIX(LHS, RHS, SCALE) - Perform fixed point division on 2 integers with the same width...
@ BUILTIN_OP_END
BUILTIN_OP_END - This must be the last enum value in this list.
@ SIGN_EXTEND
Conversion operators.
@ AVGCEILS
AVGCEILS/AVGCEILU - Rounding averaging add - Add two integers using an integer of type i[N+2],...
@ READSTEADYCOUNTER
READSTEADYCOUNTER - This corresponds to the readfixedcounter intrinsic.
@ VECREDUCE_FADD
These reductions have relaxed evaluation order semantics, and have a single vector operand.
@ CTTZ_ZERO_UNDEF
Bit counting operators with an undefined result for zero inputs.
@ PREFETCH
PREFETCH - This corresponds to a prefetch intrinsic.
@ FSINCOS
FSINCOS - Compute both fsin and fcos as a single operation.
@ SETCCCARRY
Like SetCC, ops #0 and #1 are the LHS and RHS operands to compare, but op #2 is a boolean indicating ...
@ FNEG
Perform various unary floating-point operations inspired by libm.
@ SSUBO
Same for subtraction.
@ IS_FPCLASS
Performs a check of floating point class property, defined by IEEE-754.
@ SSUBSAT
RESULT = [US]SUBSAT(LHS, RHS) - Perform saturation subtraction on 2 integers with the same bit width ...
@ SELECT
Select(COND, TRUEVAL, FALSEVAL).
@ SPLAT_VECTOR
SPLAT_VECTOR(VAL) - Returns a vector with the scalar value VAL duplicated in all lanes.
@ SADDO
RESULT, BOOL = [SU]ADDO(LHS, RHS) - Overflow-aware nodes for addition.
@ VECREDUCE_ADD
Integer reductions may have a result type larger than the vector element type.
@ GET_FPMODE
Reads the current dynamic floating-point control modes.
@ GET_FPENV
Gets the current floating-point environment.
@ SHL
Shift and rotation operations.
@ VECTOR_SHUFFLE
VECTOR_SHUFFLE(VEC1, VEC2) - Returns a vector, of the same type as VEC1/VEC2.
@ FMINNUM_IEEE
FMINNUM_IEEE/FMAXNUM_IEEE - Perform floating-point minimumNumber or maximumNumber on two values,...
@ EXTRACT_VECTOR_ELT
EXTRACT_VECTOR_ELT(VECTOR, IDX) - Returns a single element from VECTOR identified by the (potentially...
@ ZERO_EXTEND
ZERO_EXTEND - Used for integer types, zeroing the new bits.
@ DEBUGTRAP
DEBUGTRAP - Trap intended to get the attention of a debugger.
@ ATOMIC_CMP_SWAP
Val, OUTCHAIN = ATOMIC_CMP_SWAP(INCHAIN, ptr, cmp, swap) For double-word atomic operations: ValLo,...
@ FMINNUM
FMINNUM/FMAXNUM - Perform floating-point minimum or maximum on two values.
@ UBSANTRAP
UBSANTRAP - Trap with an immediate describing the kind of sanitizer failure.
@ SSHLSAT
RESULT = [US]SHLSAT(LHS, RHS) - Perform saturation left shift.
@ SMULO
Same for multiplication.
@ ANY_EXTEND_VECTOR_INREG
ANY_EXTEND_VECTOR_INREG(Vector) - This operator represents an in-register any-extension of the low la...
@ SIGN_EXTEND_INREG
SIGN_EXTEND_INREG - This operator atomically performs a SHL/SRA pair to sign extend a small value in ...
@ SMIN
[US]{MIN/MAX} - Binary minimum or maximum of signed or unsigned integers.
@ SDIVFIXSAT
Same as the corresponding unsaturated fixed point instructions, but the result is clamped between the...
@ FP_EXTEND
X = FP_EXTEND(Y) - Extend a smaller FP type into a larger FP type.
@ UADDO_CARRY
Carry-using nodes for multiple precision addition and subtraction.
@ FMINIMUM
FMINIMUM/FMAXIMUM - NaN-propagating minimum/maximum that also treat -0.0 as less than 0....
@ FP_TO_SINT
FP_TO_[US]INT - Convert a floating point value to a signed or unsigned integer.
@ READCYCLECOUNTER
READCYCLECOUNTER - This corresponds to the readcyclecounter intrinsic.
@ AND
Bitwise operators - logical and, logical or, logical xor.
@ TRAP
TRAP - Trapping instruction.
@ GET_FPENV_MEM
Gets the current floating-point environment.
@ SCMP
[US]CMP - 3-way comparison of signed or unsigned integers.
@ AVGFLOORS
AVGFLOORS/AVGFLOORU - Averaging add - Add two integers using an integer of type i[N+1],...
@ ADDE
Carry-using nodes for multiple precision addition and subtraction.
@ FREEZE
FREEZE - FREEZE(VAL) returns an arbitrary value if VAL is UNDEF (or is evaluated to UNDEF),...
@ INSERT_VECTOR_ELT
INSERT_VECTOR_ELT(VECTOR, VAL, IDX) - Returns VECTOR with the element at IDX replaced with VAL.
@ VECTOR_SPLICE
VECTOR_SPLICE(VEC1, VEC2, IMM) - Returns a subvector of the same type as VEC1/VEC2 from CONCAT_VECTOR...
@ ATOMIC_SWAP
Val, OUTCHAIN = ATOMIC_SWAP(INCHAIN, ptr, amt) Val, OUTCHAIN = ATOMIC_LOAD_[OpName](INCHAIN,...
@ FFREXP
FFREXP - frexp, extract fractional and exponent component of a floating-point value.
@ FP_ROUND
X = FP_ROUND(Y, TRUNC) - Rounding 'Y' from a larger floating point type down to the precision of the ...
@ VECTOR_COMPRESS
VECTOR_COMPRESS(Vec, Mask, Passthru) consecutively place vector elements based on mask e....
@ ZERO_EXTEND_VECTOR_INREG
ZERO_EXTEND_VECTOR_INREG(Vector) - This operator represents an in-register zero-extension of the low ...
@ ADDRSPACECAST
ADDRSPACECAST - This operator converts between pointers of different address spaces.
@ FP_TO_SINT_SAT
FP_TO_[US]INT_SAT - Convert floating point value in operand 0 to a signed or unsigned scalar integer ...
@ TRUNCATE
TRUNCATE - Completely drop the high bits.
@ FCOPYSIGN
FCOPYSIGN(X, Y) - Return the value of X with the sign of Y.
@ SADDSAT
RESULT = [US]ADDSAT(LHS, RHS) - Perform saturation addition on 2 integers with the same bit width (W)...
@ GET_DYNAMIC_AREA_OFFSET
GET_DYNAMIC_AREA_OFFSET - get offset from native SP to the address of the most recent dynamic alloca.
@ SET_FPENV_MEM
Sets the current floating point environment.
@ FMINIMUMNUM
FMINIMUMNUM/FMAXIMUMNUM - minimumnum/maximumnum that is same with FMINNUM_IEEE and FMAXNUM_IEEE besid...
@ TRUNCATE_SSAT_S
TRUNCATE_[SU]SAT_[SU] - Truncate for saturated operand [SU] located in middle, prefix for SAT means i...
@ ABDS
ABDS/ABDU - Absolute difference - Return the absolute difference between two numbers interpreted as s...
@ SADDO_CARRY
Carry-using overflow-aware nodes for multiple precision addition and subtraction.
CondCode
ISD::CondCode enum - These are ordered carefully to make the bitfields below work out,...
static const int LAST_INDEXED_MODE
Libcall getFSINCOS(EVT RetVT)
getFSINCOS - Return the FSINCOS_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall getPOWI(EVT RetVT)
getPOWI - Return the POWI_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall getSINTTOFP(EVT OpVT, EVT RetVT)
getSINTTOFP - Return the SINTTOFP_*_* value for the given types, or UNKNOWN_LIBCALL if there is none.
void initCmpLibcallCCs(ISD::CondCode *CmpLibcallCCs)
Initialize the default condition code on the libcalls.
Libcall getSYNC(unsigned Opc, MVT VT)
Return the SYNC_FETCH_AND_* value for the given opcode and type, or UNKNOWN_LIBCALL if there is none.
Libcall getLDEXP(EVT RetVT)
getLDEXP - Return the LDEXP_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall getUINTTOFP(EVT OpVT, EVT RetVT)
getUINTTOFP - Return the UINTTOFP_*_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall getFREXP(EVT RetVT)
getFREXP - Return the FREXP_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall
RTLIB::Libcall enum - This enum defines all of the runtime library calls the backend can emit.
Libcall getMEMCPY_ELEMENT_UNORDERED_ATOMIC(uint64_t ElementSize)
getMEMCPY_ELEMENT_UNORDERED_ATOMIC - Return MEMCPY_ELEMENT_UNORDERED_ATOMIC_* value for the given ele...
Libcall getFPTOUINT(EVT OpVT, EVT RetVT)
getFPTOUINT - Return the FPTOUINT_*_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall getFPTOSINT(EVT OpVT, EVT RetVT)
getFPTOSINT - Return the FPTOSINT_*_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall getOUTLINE_ATOMIC(unsigned Opc, AtomicOrdering Order, MVT VT)
Return the outline atomics value for the given opcode, atomic ordering and type, or UNKNOWN_LIBCALL i...
Libcall getFPEXT(EVT OpVT, EVT RetVT)
getFPEXT - Return the FPEXT_*_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall getFPROUND(EVT OpVT, EVT RetVT)
getFPROUND - Return the FPROUND_*_* value for the given types, or UNKNOWN_LIBCALL if there is none.
Libcall getMEMSET_ELEMENT_UNORDERED_ATOMIC(uint64_t ElementSize)
getMEMSET_ELEMENT_UNORDERED_ATOMIC - Return MEMSET_ELEMENT_UNORDERED_ATOMIC_* value for the given ele...
Libcall getOutlineAtomicHelper(const Libcall(&LC)[5][4], AtomicOrdering Order, uint64_t MemSize)
Return the outline atomics value for the given atomic ordering, access size and set of libcalls for a...
Libcall getFPLibCall(EVT VT, Libcall Call_F32, Libcall Call_F64, Libcall Call_F80, Libcall Call_F128, Libcall Call_PPCF128)
GetFPLibCall - Helper to return the right libcall for the given floating point type,...
Libcall getMEMMOVE_ELEMENT_UNORDERED_ATOMIC(uint64_t ElementSize)
getMEMMOVE_ELEMENT_UNORDERED_ATOMIC - Return MEMMOVE_ELEMENT_UNORDERED_ATOMIC_* value for the given e...
initializer< Ty > init(const Ty &Val)
This is an optimization pass for GlobalISel generic memory operations.
unsigned Log2_32_Ceil(uint32_t Value)
Return the ceil log base 2 of the specified value, 32 if the value is zero.
void GetReturnInfo(CallingConv::ID CC, Type *ReturnType, AttributeList attr, SmallVectorImpl< ISD::OutputArg > &Outs, const TargetLowering &TLI, const DataLayout &DL)
Given an LLVM IR type and return type attributes, compute the return value EVTs and flags,...
MachineInstrBuilder BuildMI(MachineFunction &MF, const MIMetadata &MIMD, const MCInstrDesc &MCID)
Builder interface. Specify how to create the initial instruction itself.
auto enum_seq(EnumT Begin, EnumT End)
Iterate over an enum type from Begin up to - but not including - End.
bool isDereferenceableAndAlignedPointer(const Value *V, Type *Ty, Align Alignment, const DataLayout &DL, const Instruction *CtxI=nullptr, AssumptionCache *AC=nullptr, const DominatorTree *DT=nullptr, const TargetLibraryInfo *TLI=nullptr)
Returns true if V is always a dereferenceable pointer with alignment greater or equal than requested.
bool shouldOptimizeForSize(const MachineFunction *MF, ProfileSummaryInfo *PSI, const MachineBlockFrequencyInfo *BFI, PGSOQueryType QueryType=PGSOQueryType::Other)
Returns true if machine function MF is suggested to be size-optimized based on the profile.
constexpr force_iteration_on_noniterable_enum_t force_iteration_on_noniterable_enum
T bit_ceil(T Value)
Returns the smallest integral power of two no smaller than Value if Value is nonzero.
bool isReleaseOrStronger(AtomicOrdering AO)
constexpr bool isPowerOf2_32(uint32_t Value)
Return true if the argument is a power of two > 0.
bool none_of(R &&Range, UnaryPredicate P)
Provide wrappers to std::none_of which take ranges instead of having to pass begin/end explicitly.
void report_fatal_error(Error Err, bool gen_crash_diag=true)
Report a serious error, calling any installed error handler.
AtomicOrdering
Atomic ordering for LLVM's memory model.
EVT getApproximateEVTForLLT(LLT Ty, LLVMContext &Ctx)
constexpr T divideCeil(U Numerator, V Denominator)
Returns the integer ceil(Numerator / Denominator).
@ Or
Bitwise or logical OR of integers.
@ Mul
Product of integers.
@ Xor
Bitwise or logical XOR of integers.
@ And
Bitwise or logical AND of integers.
void ComputeValueVTs(const TargetLowering &TLI, const DataLayout &DL, Type *Ty, SmallVectorImpl< EVT > &ValueVTs, SmallVectorImpl< EVT > *MemVTs, SmallVectorImpl< TypeSize > *Offsets=nullptr, TypeSize StartingOffset=TypeSize::getZero())
ComputeValueVTs - Given an LLVM IR type, compute a sequence of EVTs that represent all the individual...
bool isAcquireOrStronger(AtomicOrdering AO)
This struct is a compact representation of a valid (non-zero power of two) alignment.
EVT getPow2VectorType(LLVMContext &Context) const
Widens the length of the given vector EVT up to the nearest power of 2 and returns that type.
bool isSimple() const
Test if the given EVT is simple (as opposed to being extended).
static EVT getVectorVT(LLVMContext &Context, EVT VT, unsigned NumElements, bool IsScalable=false)
Returns the EVT that represents a vector NumElements in length, where each element is of type VT.
ElementCount getVectorElementCount() const
TypeSize getSizeInBits() const
Return the size of the specified value type in bits.
bool isPow2VectorType() const
Returns true if the given vector is a power of 2.
MVT getSimpleVT() const
Return the SimpleValueType held in the specified simple EVT.
static EVT getIntegerVT(LLVMContext &Context, unsigned BitWidth)
Returns the EVT that represents an integer with the given number of bits.
bool isFixedLengthVector() const
EVT getRoundIntegerType(LLVMContext &Context) const
Rounds the bit-width of the given integer EVT up to the nearest power of two (and at least to eight),...
bool isVector() const
Return true if this is a vector value type.
EVT getScalarType() const
If this is a vector type, return the element type, otherwise return this.
Type * getTypeForEVT(LLVMContext &Context) const
This method returns an LLVM type corresponding to the specified EVT.
EVT getVectorElementType() const
Given a vector type, return the type of each element.
unsigned getVectorNumElements() const
Given a vector type, return the number of elements it contains.
bool isZeroSized() const
Test if the given EVT has zero size, this will fail if called on a scalable type.
EVT getHalfNumVectorElementsVT(LLVMContext &Context) const
bool isInteger() const
Return true if this is an integer or a vector integer type.
OutputArg - This struct carries flags and a value for a single outgoing (actual) argument or outgoing...
static MachinePointerInfo getFixedStack(MachineFunction &MF, int FI, int64_t Offset=0)
Return a MachinePointerInfo record that refers to the specified FrameIndex.
This represents an addressing mode of: BaseGV + BaseOffs + BaseReg + Scale*ScaleReg + ScalableOffset*...