blob: ea94871d931555f17cee9d80eb656046176a65df [file] [log] [blame]
Andrew Scull5e1ddfa2018-08-14 10:06:54 +01001//===-- llvm/CodeGen/ISDOpcodes.h - CodeGen opcodes -------------*- C++ -*-===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file declares codegen opcodes and related utilities.
11//
12//===----------------------------------------------------------------------===//
13
14#ifndef LLVM_CODEGEN_ISDOPCODES_H
15#define LLVM_CODEGEN_ISDOPCODES_H
16
17namespace llvm {
18
19/// ISD namespace - This namespace contains an enum which represents all of the
20/// SelectionDAG node types and value types.
21///
22namespace ISD {
23
24 //===--------------------------------------------------------------------===//
25 /// ISD::NodeType enum - This enum defines the target-independent operators
26 /// for a SelectionDAG.
27 ///
28 /// Targets may also define target-dependent operator codes for SDNodes. For
29 /// example, on x86, these are the enum values in the X86ISD namespace.
30 /// Targets should aim to use target-independent operators to model their
31 /// instruction sets as much as possible, and only use target-dependent
32 /// operators when they have special requirements.
33 ///
34 /// Finally, during and after selection proper, SNodes may use special
35 /// operator codes that correspond directly with MachineInstr opcodes. These
36 /// are used to represent selected instructions. See the isMachineOpcode()
37 /// and getMachineOpcode() member functions of SDNode.
38 ///
39 enum NodeType {
40 /// DELETED_NODE - This is an illegal value that is used to catch
41 /// errors. This opcode is not a legal opcode for any node.
42 DELETED_NODE,
43
44 /// EntryToken - This is the marker used to indicate the start of a region.
45 EntryToken,
46
47 /// TokenFactor - This node takes multiple tokens as input and produces a
48 /// single token result. This is used to represent the fact that the operand
49 /// operators are independent of each other.
50 TokenFactor,
51
52 /// AssertSext, AssertZext - These nodes record if a register contains a
53 /// value that has already been zero or sign extended from a narrower type.
54 /// These nodes take two operands. The first is the node that has already
55 /// been extended, and the second is a value type node indicating the width
56 /// of the extension
57 AssertSext, AssertZext,
58
59 /// Various leaf nodes.
60 BasicBlock, VALUETYPE, CONDCODE, Register, RegisterMask,
61 Constant, ConstantFP,
62 GlobalAddress, GlobalTLSAddress, FrameIndex,
63 JumpTable, ConstantPool, ExternalSymbol, BlockAddress,
64
65 /// The address of the GOT
66 GLOBAL_OFFSET_TABLE,
67
68 /// FRAMEADDR, RETURNADDR - These nodes represent llvm.frameaddress and
69 /// llvm.returnaddress on the DAG. These nodes take one operand, the index
70 /// of the frame or return address to return. An index of zero corresponds
71 /// to the current function's frame or return address, an index of one to
72 /// the parent's frame or return address, and so on.
73 FRAMEADDR, RETURNADDR, ADDROFRETURNADDR,
74
75 /// LOCAL_RECOVER - Represents the llvm.localrecover intrinsic.
76 /// Materializes the offset from the local object pointer of another
77 /// function to a particular local object passed to llvm.localescape. The
78 /// operand is the MCSymbol label used to represent this offset, since
79 /// typically the offset is not known until after code generation of the
80 /// parent.
81 LOCAL_RECOVER,
82
83 /// READ_REGISTER, WRITE_REGISTER - This node represents llvm.register on
84 /// the DAG, which implements the named register global variables extension.
85 READ_REGISTER,
86 WRITE_REGISTER,
87
88 /// FRAME_TO_ARGS_OFFSET - This node represents offset from frame pointer to
89 /// first (possible) on-stack argument. This is needed for correct stack
90 /// adjustment during unwind.
91 FRAME_TO_ARGS_OFFSET,
92
93 /// EH_DWARF_CFA - This node represents the pointer to the DWARF Canonical
94 /// Frame Address (CFA), generally the value of the stack pointer at the
95 /// call site in the previous frame.
96 EH_DWARF_CFA,
97
98 /// OUTCHAIN = EH_RETURN(INCHAIN, OFFSET, HANDLER) - This node represents
99 /// 'eh_return' gcc dwarf builtin, which is used to return from
100 /// exception. The general meaning is: adjust stack by OFFSET and pass
101 /// execution to HANDLER. Many platform-related details also :)
102 EH_RETURN,
103
104 /// RESULT, OUTCHAIN = EH_SJLJ_SETJMP(INCHAIN, buffer)
105 /// This corresponds to the eh.sjlj.setjmp intrinsic.
106 /// It takes an input chain and a pointer to the jump buffer as inputs
107 /// and returns an outchain.
108 EH_SJLJ_SETJMP,
109
110 /// OUTCHAIN = EH_SJLJ_LONGJMP(INCHAIN, buffer)
111 /// This corresponds to the eh.sjlj.longjmp intrinsic.
112 /// It takes an input chain and a pointer to the jump buffer as inputs
113 /// and returns an outchain.
114 EH_SJLJ_LONGJMP,
115
116 /// OUTCHAIN = EH_SJLJ_SETUP_DISPATCH(INCHAIN)
117 /// The target initializes the dispatch table here.
118 EH_SJLJ_SETUP_DISPATCH,
119
120 /// TargetConstant* - Like Constant*, but the DAG does not do any folding,
121 /// simplification, or lowering of the constant. They are used for constants
122 /// which are known to fit in the immediate fields of their users, or for
123 /// carrying magic numbers which are not values which need to be
124 /// materialized in registers.
125 TargetConstant,
126 TargetConstantFP,
127
128 /// TargetGlobalAddress - Like GlobalAddress, but the DAG does no folding or
129 /// anything else with this node, and this is valid in the target-specific
130 /// dag, turning into a GlobalAddress operand.
131 TargetGlobalAddress,
132 TargetGlobalTLSAddress,
133 TargetFrameIndex,
134 TargetJumpTable,
135 TargetConstantPool,
136 TargetExternalSymbol,
137 TargetBlockAddress,
138
139 MCSymbol,
140
141 /// TargetIndex - Like a constant pool entry, but with completely
142 /// target-dependent semantics. Holds target flags, a 32-bit index, and a
143 /// 64-bit index. Targets can use this however they like.
144 TargetIndex,
145
146 /// RESULT = INTRINSIC_WO_CHAIN(INTRINSICID, arg1, arg2, ...)
147 /// This node represents a target intrinsic function with no side effects.
148 /// The first operand is the ID number of the intrinsic from the
149 /// llvm::Intrinsic namespace. The operands to the intrinsic follow. The
150 /// node returns the result of the intrinsic.
151 INTRINSIC_WO_CHAIN,
152
153 /// RESULT,OUTCHAIN = INTRINSIC_W_CHAIN(INCHAIN, INTRINSICID, arg1, ...)
154 /// This node represents a target intrinsic function with side effects that
155 /// returns a result. The first operand is a chain pointer. The second is
156 /// the ID number of the intrinsic from the llvm::Intrinsic namespace. The
157 /// operands to the intrinsic follow. The node has two results, the result
158 /// of the intrinsic and an output chain.
159 INTRINSIC_W_CHAIN,
160
161 /// OUTCHAIN = INTRINSIC_VOID(INCHAIN, INTRINSICID, arg1, arg2, ...)
162 /// This node represents a target intrinsic function with side effects that
163 /// does not return a result. The first operand is a chain pointer. The
164 /// second is the ID number of the intrinsic from the llvm::Intrinsic
165 /// namespace. The operands to the intrinsic follow.
166 INTRINSIC_VOID,
167
168 /// CopyToReg - This node has three operands: a chain, a register number to
169 /// set to this value, and a value.
170 CopyToReg,
171
172 /// CopyFromReg - This node indicates that the input value is a virtual or
173 /// physical register that is defined outside of the scope of this
174 /// SelectionDAG. The register is available from the RegisterSDNode object.
175 CopyFromReg,
176
177 /// UNDEF - An undefined node.
178 UNDEF,
179
180 /// EXTRACT_ELEMENT - This is used to get the lower or upper (determined by
181 /// a Constant, which is required to be operand #1) half of the integer or
182 /// float value specified as operand #0. This is only for use before
183 /// legalization, for values that will be broken into multiple registers.
184 EXTRACT_ELEMENT,
185
186 /// BUILD_PAIR - This is the opposite of EXTRACT_ELEMENT in some ways.
187 /// Given two values of the same integer value type, this produces a value
188 /// twice as big. Like EXTRACT_ELEMENT, this can only be used before
189 /// legalization. The lower part of the composite value should be in
190 /// element 0 and the upper part should be in element 1.
191 BUILD_PAIR,
192
193 /// MERGE_VALUES - This node takes multiple discrete operands and returns
194 /// them all as its individual results. This nodes has exactly the same
195 /// number of inputs and outputs. This node is useful for some pieces of the
196 /// code generator that want to think about a single node with multiple
197 /// results, not multiple nodes.
198 MERGE_VALUES,
199
200 /// Simple integer binary arithmetic operators.
201 ADD, SUB, MUL, SDIV, UDIV, SREM, UREM,
202
203 /// SMUL_LOHI/UMUL_LOHI - Multiply two integers of type iN, producing
204 /// a signed/unsigned value of type i[2*N], and return the full value as
205 /// two results, each of type iN.
206 SMUL_LOHI, UMUL_LOHI,
207
208 /// SDIVREM/UDIVREM - Divide two integers and produce both a quotient and
209 /// remainder result.
210 SDIVREM, UDIVREM,
211
212 /// CARRY_FALSE - This node is used when folding other nodes,
213 /// like ADDC/SUBC, which indicate the carry result is always false.
214 CARRY_FALSE,
215
216 /// Carry-setting nodes for multiple precision addition and subtraction.
217 /// These nodes take two operands of the same value type, and produce two
218 /// results. The first result is the normal add or sub result, the second
219 /// result is the carry flag result.
220 /// FIXME: These nodes are deprecated in favor of ADDCARRY and SUBCARRY.
221 /// They are kept around for now to provide a smooth transition path
222 /// toward the use of ADDCARRY/SUBCARRY and will eventually be removed.
223 ADDC, SUBC,
224
225 /// Carry-using nodes for multiple precision addition and subtraction. These
226 /// nodes take three operands: The first two are the normal lhs and rhs to
227 /// the add or sub, and the third is the input carry flag. These nodes
228 /// produce two results; the normal result of the add or sub, and the output
229 /// carry flag. These nodes both read and write a carry flag to allow them
230 /// to them to be chained together for add and sub of arbitrarily large
231 /// values.
232 ADDE, SUBE,
233
234 /// Carry-using nodes for multiple precision addition and subtraction.
235 /// These nodes take three operands: The first two are the normal lhs and
236 /// rhs to the add or sub, and the third is a boolean indicating if there
237 /// is an incoming carry. These nodes produce two results: the normal
238 /// result of the add or sub, and the output carry so they can be chained
239 /// together. The use of this opcode is preferable to adde/sube if the
240 /// target supports it, as the carry is a regular value rather than a
241 /// glue, which allows further optimisation.
242 ADDCARRY, SUBCARRY,
243
244 /// RESULT, BOOL = [SU]ADDO(LHS, RHS) - Overflow-aware nodes for addition.
245 /// These nodes take two operands: the normal LHS and RHS to the add. They
246 /// produce two results: the normal result of the add, and a boolean that
247 /// indicates if an overflow occurred (*not* a flag, because it may be store
248 /// to memory, etc.). If the type of the boolean is not i1 then the high
249 /// bits conform to getBooleanContents.
250 /// These nodes are generated from llvm.[su]add.with.overflow intrinsics.
251 SADDO, UADDO,
252
253 /// Same for subtraction.
254 SSUBO, USUBO,
255
256 /// Same for multiplication.
257 SMULO, UMULO,
258
259 /// Simple binary floating point operators.
260 FADD, FSUB, FMUL, FDIV, FREM,
261
262 /// Constrained versions of the binary floating point operators.
263 /// These will be lowered to the simple operators before final selection.
264 /// They are used to limit optimizations while the DAG is being
265 /// optimized.
266 STRICT_FADD, STRICT_FSUB, STRICT_FMUL, STRICT_FDIV, STRICT_FREM,
267 STRICT_FMA,
268
269 /// Constrained versions of libm-equivalent floating point intrinsics.
270 /// These will be lowered to the equivalent non-constrained pseudo-op
271 /// (or expanded to the equivalent library call) before final selection.
272 /// They are used to limit optimizations while the DAG is being optimized.
273 STRICT_FSQRT, STRICT_FPOW, STRICT_FPOWI, STRICT_FSIN, STRICT_FCOS,
274 STRICT_FEXP, STRICT_FEXP2, STRICT_FLOG, STRICT_FLOG10, STRICT_FLOG2,
275 STRICT_FRINT, STRICT_FNEARBYINT,
276
277 /// FMA - Perform a * b + c with no intermediate rounding step.
278 FMA,
279
280 /// FMAD - Perform a * b + c, while getting the same result as the
281 /// separately rounded operations.
282 FMAD,
283
284 /// FCOPYSIGN(X, Y) - Return the value of X with the sign of Y. NOTE: This
285 /// DAG node does not require that X and Y have the same type, just that
286 /// they are both floating point. X and the result must have the same type.
287 /// FCOPYSIGN(f32, f64) is allowed.
288 FCOPYSIGN,
289
290 /// INT = FGETSIGN(FP) - Return the sign bit of the specified floating point
291 /// value as an integer 0/1 value.
292 FGETSIGN,
293
294 /// Returns platform specific canonical encoding of a floating point number.
295 FCANONICALIZE,
296
297 /// BUILD_VECTOR(ELT0, ELT1, ELT2, ELT3,...) - Return a vector with the
298 /// specified, possibly variable, elements. The number of elements is
299 /// required to be a power of two. The types of the operands must all be
300 /// the same and must match the vector element type, except that integer
301 /// types are allowed to be larger than the element type, in which case
302 /// the operands are implicitly truncated.
303 BUILD_VECTOR,
304
305 /// INSERT_VECTOR_ELT(VECTOR, VAL, IDX) - Returns VECTOR with the element
306 /// at IDX replaced with VAL. If the type of VAL is larger than the vector
307 /// element type then VAL is truncated before replacement.
308 INSERT_VECTOR_ELT,
309
310 /// EXTRACT_VECTOR_ELT(VECTOR, IDX) - Returns a single element from VECTOR
311 /// identified by the (potentially variable) element number IDX. If the
312 /// return type is an integer type larger than the element type of the
313 /// vector, the result is extended to the width of the return type. In
314 /// that case, the high bits are undefined.
315 EXTRACT_VECTOR_ELT,
316
317 /// CONCAT_VECTORS(VECTOR0, VECTOR1, ...) - Given a number of values of
318 /// vector type with the same length and element type, this produces a
319 /// concatenated vector result value, with length equal to the sum of the
320 /// lengths of the input vectors.
321 CONCAT_VECTORS,
322
323 /// INSERT_SUBVECTOR(VECTOR1, VECTOR2, IDX) - Returns a vector
324 /// with VECTOR2 inserted into VECTOR1 at the (potentially
325 /// variable) element number IDX, which must be a multiple of the
326 /// VECTOR2 vector length. The elements of VECTOR1 starting at
327 /// IDX are overwritten with VECTOR2. Elements IDX through
328 /// vector_length(VECTOR2) must be valid VECTOR1 indices.
329 INSERT_SUBVECTOR,
330
331 /// EXTRACT_SUBVECTOR(VECTOR, IDX) - Returns a subvector from VECTOR (an
332 /// vector value) starting with the element number IDX, which must be a
333 /// constant multiple of the result vector length.
334 EXTRACT_SUBVECTOR,
335
336 /// VECTOR_SHUFFLE(VEC1, VEC2) - Returns a vector, of the same type as
337 /// VEC1/VEC2. A VECTOR_SHUFFLE node also contains an array of constant int
338 /// values that indicate which value (or undef) each result element will
339 /// get. These constant ints are accessible through the
340 /// ShuffleVectorSDNode class. This is quite similar to the Altivec
341 /// 'vperm' instruction, except that the indices must be constants and are
342 /// in terms of the element size of VEC1/VEC2, not in terms of bytes.
343 VECTOR_SHUFFLE,
344
345 /// SCALAR_TO_VECTOR(VAL) - This represents the operation of loading a
346 /// scalar value into element 0 of the resultant vector type. The top
347 /// elements 1 to N-1 of the N-element vector are undefined. The type
348 /// of the operand must match the vector element type, except when they
349 /// are integer types. In this case the operand is allowed to be wider
350 /// than the vector element type, and is implicitly truncated to it.
351 SCALAR_TO_VECTOR,
352
353 /// MULHU/MULHS - Multiply high - Multiply two integers of type iN,
354 /// producing an unsigned/signed value of type i[2*N], then return the top
355 /// part.
356 MULHU, MULHS,
357
358 /// [US]{MIN/MAX} - Binary minimum or maximum or signed or unsigned
359 /// integers.
360 SMIN, SMAX, UMIN, UMAX,
361
362 /// Bitwise operators - logical and, logical or, logical xor.
363 AND, OR, XOR,
364
365 /// ABS - Determine the unsigned absolute value of a signed integer value of
366 /// the same bitwidth.
367 /// Note: A value of INT_MIN will return INT_MIN, no saturation or overflow
368 /// is performed.
369 ABS,
370
371 /// Shift and rotation operations. After legalization, the type of the
372 /// shift amount is known to be TLI.getShiftAmountTy(). Before legalization
373 /// the shift amount can be any type, but care must be taken to ensure it is
374 /// large enough. TLI.getShiftAmountTy() is i8 on some targets, but before
375 /// legalization, types like i1024 can occur and i8 doesn't have enough bits
376 /// to represent the shift amount.
377 /// When the 1st operand is a vector, the shift amount must be in the same
378 /// type. (TLI.getShiftAmountTy() will return the same type when the input
379 /// type is a vector.)
380 SHL, SRA, SRL, ROTL, ROTR,
381
382 /// Byte Swap and Counting operators.
383 BSWAP, CTTZ, CTLZ, CTPOP, BITREVERSE,
384
385 /// Bit counting operators with an undefined result for zero inputs.
386 CTTZ_ZERO_UNDEF, CTLZ_ZERO_UNDEF,
387
388 /// Select(COND, TRUEVAL, FALSEVAL). If the type of the boolean COND is not
389 /// i1 then the high bits must conform to getBooleanContents.
390 SELECT,
391
392 /// Select with a vector condition (op #0) and two vector operands (ops #1
393 /// and #2), returning a vector result. All vectors have the same length.
394 /// Much like the scalar select and setcc, each bit in the condition selects
395 /// whether the corresponding result element is taken from op #1 or op #2.
396 /// At first, the VSELECT condition is of vXi1 type. Later, targets may
397 /// change the condition type in order to match the VSELECT node using a
398 /// pattern. The condition follows the BooleanContent format of the target.
399 VSELECT,
400
401 /// Select with condition operator - This selects between a true value and
402 /// a false value (ops #2 and #3) based on the boolean result of comparing
403 /// the lhs and rhs (ops #0 and #1) of a conditional expression with the
404 /// condition code in op #4, a CondCodeSDNode.
405 SELECT_CC,
406
407 /// SetCC operator - This evaluates to a true value iff the condition is
408 /// true. If the result value type is not i1 then the high bits conform
409 /// to getBooleanContents. The operands to this are the left and right
410 /// operands to compare (ops #0, and #1) and the condition code to compare
411 /// them with (op #2) as a CondCodeSDNode. If the operands are vector types
412 /// then the result type must also be a vector type.
413 SETCC,
414
415 /// Like SetCC, ops #0 and #1 are the LHS and RHS operands to compare, and
416 /// op #2 is a *carry value*. This operator checks the result of
417 /// "LHS - RHS - Carry", and can be used to compare two wide integers:
418 /// (setcce lhshi rhshi (subc lhslo rhslo) cc). Only valid for integers.
419 /// FIXME: This node is deprecated in favor of SETCCCARRY.
420 /// It is kept around for now to provide a smooth transition path
421 /// toward the use of SETCCCARRY and will eventually be removed.
422 SETCCE,
423
424 /// Like SetCC, ops #0 and #1 are the LHS and RHS operands to compare, but
425 /// op #2 is a boolean indicating if there is an incoming carry. This
426 /// operator checks the result of "LHS - RHS - Carry", and can be used to
427 /// compare two wide integers: (setcce lhshi rhshi (subc lhslo rhslo) cc).
428 /// Only valid for integers.
429 SETCCCARRY,
430
431 /// SHL_PARTS/SRA_PARTS/SRL_PARTS - These operators are used for expanded
432 /// integer shift operations. The operation ordering is:
433 /// [Lo,Hi] = op [LoLHS,HiLHS], Amt
434 SHL_PARTS, SRA_PARTS, SRL_PARTS,
435
436 /// Conversion operators. These are all single input single output
437 /// operations. For all of these, the result type must be strictly
438 /// wider or narrower (depending on the operation) than the source
439 /// type.
440
441 /// SIGN_EXTEND - Used for integer types, replicating the sign bit
442 /// into new bits.
443 SIGN_EXTEND,
444
445 /// ZERO_EXTEND - Used for integer types, zeroing the new bits.
446 ZERO_EXTEND,
447
448 /// ANY_EXTEND - Used for integer types. The high bits are undefined.
449 ANY_EXTEND,
450
451 /// TRUNCATE - Completely drop the high bits.
452 TRUNCATE,
453
454 /// [SU]INT_TO_FP - These operators convert integers (whose interpreted sign
455 /// depends on the first letter) to floating point.
456 SINT_TO_FP,
457 UINT_TO_FP,
458
459 /// SIGN_EXTEND_INREG - This operator atomically performs a SHL/SRA pair to
460 /// sign extend a small value in a large integer register (e.g. sign
461 /// extending the low 8 bits of a 32-bit register to fill the top 24 bits
462 /// with the 7th bit). The size of the smaller type is indicated by the 1th
463 /// operand, a ValueType node.
464 SIGN_EXTEND_INREG,
465
466 /// ANY_EXTEND_VECTOR_INREG(Vector) - This operator represents an
467 /// in-register any-extension of the low lanes of an integer vector. The
468 /// result type must have fewer elements than the operand type, and those
469 /// elements must be larger integer types such that the total size of the
470 /// operand type and the result type match. Each of the low operand
471 /// elements is any-extended into the corresponding, wider result
472 /// elements with the high bits becoming undef.
473 ANY_EXTEND_VECTOR_INREG,
474
475 /// SIGN_EXTEND_VECTOR_INREG(Vector) - This operator represents an
476 /// in-register sign-extension of the low lanes of an integer vector. The
477 /// result type must have fewer elements than the operand type, and those
478 /// elements must be larger integer types such that the total size of the
479 /// operand type and the result type match. Each of the low operand
480 /// elements is sign-extended into the corresponding, wider result
481 /// elements.
482 // FIXME: The SIGN_EXTEND_INREG node isn't specifically limited to
483 // scalars, but it also doesn't handle vectors well. Either it should be
484 // restricted to scalars or this node (and its handling) should be merged
485 // into it.
486 SIGN_EXTEND_VECTOR_INREG,
487
488 /// ZERO_EXTEND_VECTOR_INREG(Vector) - This operator represents an
489 /// in-register zero-extension of the low lanes of an integer vector. The
490 /// result type must have fewer elements than the operand type, and those
491 /// elements must be larger integer types such that the total size of the
492 /// operand type and the result type match. Each of the low operand
493 /// elements is zero-extended into the corresponding, wider result
494 /// elements.
495 ZERO_EXTEND_VECTOR_INREG,
496
497 /// FP_TO_[US]INT - Convert a floating point value to a signed or unsigned
498 /// integer.
499 FP_TO_SINT,
500 FP_TO_UINT,
501
502 /// X = FP_ROUND(Y, TRUNC) - Rounding 'Y' from a larger floating point type
503 /// down to the precision of the destination VT. TRUNC is a flag, which is
504 /// always an integer that is zero or one. If TRUNC is 0, this is a
505 /// normal rounding, if it is 1, this FP_ROUND is known to not change the
506 /// value of Y.
507 ///
508 /// The TRUNC = 1 case is used in cases where we know that the value will
509 /// not be modified by the node, because Y is not using any of the extra
510 /// precision of source type. This allows certain transformations like
511 /// FP_EXTEND(FP_ROUND(X,1)) -> X which are not safe for
512 /// FP_EXTEND(FP_ROUND(X,0)) because the extra bits aren't removed.
513 FP_ROUND,
514
515 /// FLT_ROUNDS_ - Returns current rounding mode:
516 /// -1 Undefined
517 /// 0 Round to 0
518 /// 1 Round to nearest
519 /// 2 Round to +inf
520 /// 3 Round to -inf
521 FLT_ROUNDS_,
522
523 /// X = FP_ROUND_INREG(Y, VT) - This operator takes an FP register, and
524 /// rounds it to a floating point value. It then promotes it and returns it
525 /// in a register of the same size. This operation effectively just
526 /// discards excess precision. The type to round down to is specified by
527 /// the VT operand, a VTSDNode.
528 FP_ROUND_INREG,
529
530 /// X = FP_EXTEND(Y) - Extend a smaller FP type into a larger FP type.
531 FP_EXTEND,
532
533 /// BITCAST - This operator converts between integer, vector and FP
534 /// values, as if the value was stored to memory with one type and loaded
535 /// from the same address with the other type (or equivalently for vector
536 /// format conversions, etc). The source and result are required to have
537 /// the same bit size (e.g. f32 <-> i32). This can also be used for
538 /// int-to-int or fp-to-fp conversions, but that is a noop, deleted by
539 /// getNode().
540 ///
541 /// This operator is subtly different from the bitcast instruction from
542 /// LLVM-IR since this node may change the bits in the register. For
543 /// example, this occurs on big-endian NEON and big-endian MSA where the
544 /// layout of the bits in the register depends on the vector type and this
545 /// operator acts as a shuffle operation for some vector type combinations.
546 BITCAST,
547
548 /// ADDRSPACECAST - This operator converts between pointers of different
549 /// address spaces.
550 ADDRSPACECAST,
551
552 /// FP16_TO_FP, FP_TO_FP16 - These operators are used to perform promotions
553 /// and truncation for half-precision (16 bit) floating numbers. These nodes
554 /// form a semi-softened interface for dealing with f16 (as an i16), which
555 /// is often a storage-only type but has native conversions.
556 FP16_TO_FP, FP_TO_FP16,
557
558 /// FNEG, FABS, FSQRT, FSIN, FCOS, FPOWI, FPOW,
559 /// FLOG, FLOG2, FLOG10, FEXP, FEXP2,
560 /// FCEIL, FTRUNC, FRINT, FNEARBYINT, FROUND, FFLOOR - Perform various unary
561 /// floating point operations. These are inspired by libm.
562 FNEG, FABS, FSQRT, FSIN, FCOS, FPOWI, FPOW,
563 FLOG, FLOG2, FLOG10, FEXP, FEXP2,
564 FCEIL, FTRUNC, FRINT, FNEARBYINT, FROUND, FFLOOR,
565 /// FMINNUM/FMAXNUM - Perform floating-point minimum or maximum on two
566 /// values.
567 /// In the case where a single input is NaN, the non-NaN input is returned.
568 ///
569 /// The return value of (FMINNUM 0.0, -0.0) could be either 0.0 or -0.0.
570 FMINNUM, FMAXNUM,
571 /// FMINNAN/FMAXNAN - Behave identically to FMINNUM/FMAXNUM, except that
572 /// when a single input is NaN, NaN is returned.
573 FMINNAN, FMAXNAN,
574
575 /// FSINCOS - Compute both fsin and fcos as a single operation.
576 FSINCOS,
577
578 /// LOAD and STORE have token chains as their first operand, then the same
579 /// operands as an LLVM load/store instruction, then an offset node that
580 /// is added / subtracted from the base pointer to form the address (for
581 /// indexed memory ops).
582 LOAD, STORE,
583
584 /// DYNAMIC_STACKALLOC - Allocate some number of bytes on the stack aligned
585 /// to a specified boundary. This node always has two return values: a new
586 /// stack pointer value and a chain. The first operand is the token chain,
587 /// the second is the number of bytes to allocate, and the third is the
588 /// alignment boundary. The size is guaranteed to be a multiple of the
589 /// stack alignment, and the alignment is guaranteed to be bigger than the
590 /// stack alignment (if required) or 0 to get standard stack alignment.
591 DYNAMIC_STACKALLOC,
592
593 /// Control flow instructions. These all have token chains.
594
595 /// BR - Unconditional branch. The first operand is the chain
596 /// operand, the second is the MBB to branch to.
597 BR,
598
599 /// BRIND - Indirect branch. The first operand is the chain, the second
600 /// is the value to branch to, which must be of the same type as the
601 /// target's pointer type.
602 BRIND,
603
604 /// BR_JT - Jumptable branch. The first operand is the chain, the second
605 /// is the jumptable index, the last one is the jumptable entry index.
606 BR_JT,
607
608 /// BRCOND - Conditional branch. The first operand is the chain, the
609 /// second is the condition, the third is the block to branch to if the
610 /// condition is true. If the type of the condition is not i1, then the
611 /// high bits must conform to getBooleanContents.
612 BRCOND,
613
614 /// BR_CC - Conditional branch. The behavior is like that of SELECT_CC, in
615 /// that the condition is represented as condition code, and two nodes to
616 /// compare, rather than as a combined SetCC node. The operands in order
617 /// are chain, cc, lhs, rhs, block to branch to if condition is true.
618 BR_CC,
619
620 /// INLINEASM - Represents an inline asm block. This node always has two
621 /// return values: a chain and a flag result. The inputs are as follows:
622 /// Operand #0 : Input chain.
623 /// Operand #1 : a ExternalSymbolSDNode with a pointer to the asm string.
624 /// Operand #2 : a MDNodeSDNode with the !srcloc metadata.
625 /// Operand #3 : HasSideEffect, IsAlignStack bits.
626 /// After this, it is followed by a list of operands with this format:
627 /// ConstantSDNode: Flags that encode whether it is a mem or not, the
628 /// of operands that follow, etc. See InlineAsm.h.
629 /// ... however many operands ...
630 /// Operand #last: Optional, an incoming flag.
631 ///
632 /// The variable width operands are required to represent target addressing
633 /// modes as a single "operand", even though they may have multiple
634 /// SDOperands.
635 INLINEASM,
636
637 /// EH_LABEL - Represents a label in mid basic block used to track
638 /// locations needed for debug and exception handling tables. These nodes
639 /// take a chain as input and return a chain.
640 EH_LABEL,
641
642 /// ANNOTATION_LABEL - Represents a mid basic block label used by
643 /// annotations. This should remain within the basic block and be ordered
644 /// with respect to other call instructions, but loads and stores may float
645 /// past it.
646 ANNOTATION_LABEL,
647
648 /// CATCHPAD - Represents a catchpad instruction.
649 CATCHPAD,
650
651 /// CATCHRET - Represents a return from a catch block funclet. Used for
652 /// MSVC compatible exception handling. Takes a chain operand and a
653 /// destination basic block operand.
654 CATCHRET,
655
656 /// CLEANUPRET - Represents a return from a cleanup block funclet. Used for
657 /// MSVC compatible exception handling. Takes only a chain operand.
658 CLEANUPRET,
659
660 /// STACKSAVE - STACKSAVE has one operand, an input chain. It produces a
661 /// value, the same type as the pointer type for the system, and an output
662 /// chain.
663 STACKSAVE,
664
665 /// STACKRESTORE has two operands, an input chain and a pointer to restore
666 /// to it returns an output chain.
667 STACKRESTORE,
668
669 /// CALLSEQ_START/CALLSEQ_END - These operators mark the beginning and end
670 /// of a call sequence, and carry arbitrary information that target might
671 /// want to know. The first operand is a chain, the rest are specified by
672 /// the target and not touched by the DAG optimizers.
673 /// Targets that may use stack to pass call arguments define additional
674 /// operands:
675 /// - size of the call frame part that must be set up within the
676 /// CALLSEQ_START..CALLSEQ_END pair,
677 /// - part of the call frame prepared prior to CALLSEQ_START.
678 /// Both these parameters must be constants, their sum is the total call
679 /// frame size.
680 /// CALLSEQ_START..CALLSEQ_END pairs may not be nested.
681 CALLSEQ_START, // Beginning of a call sequence
682 CALLSEQ_END, // End of a call sequence
683
684 /// VAARG - VAARG has four operands: an input chain, a pointer, a SRCVALUE,
685 /// and the alignment. It returns a pair of values: the vaarg value and a
686 /// new chain.
687 VAARG,
688
689 /// VACOPY - VACOPY has 5 operands: an input chain, a destination pointer,
690 /// a source pointer, a SRCVALUE for the destination, and a SRCVALUE for the
691 /// source.
692 VACOPY,
693
694 /// VAEND, VASTART - VAEND and VASTART have three operands: an input chain,
695 /// pointer, and a SRCVALUE.
696 VAEND, VASTART,
697
698 /// SRCVALUE - This is a node type that holds a Value* that is used to
699 /// make reference to a value in the LLVM IR.
700 SRCVALUE,
701
702 /// MDNODE_SDNODE - This is a node that holdes an MDNode*, which is used to
703 /// reference metadata in the IR.
704 MDNODE_SDNODE,
705
706 /// PCMARKER - This corresponds to the pcmarker intrinsic.
707 PCMARKER,
708
709 /// READCYCLECOUNTER - This corresponds to the readcyclecounter intrinsic.
710 /// It produces a chain and one i64 value. The only operand is a chain.
711 /// If i64 is not legal, the result will be expanded into smaller values.
712 /// Still, it returns an i64, so targets should set legality for i64.
713 /// The result is the content of the architecture-specific cycle
714 /// counter-like register (or other high accuracy low latency clock source).
715 READCYCLECOUNTER,
716
717 /// HANDLENODE node - Used as a handle for various purposes.
718 HANDLENODE,
719
720 /// INIT_TRAMPOLINE - This corresponds to the init_trampoline intrinsic. It
721 /// takes as input a token chain, the pointer to the trampoline, the pointer
722 /// to the nested function, the pointer to pass for the 'nest' parameter, a
723 /// SRCVALUE for the trampoline and another for the nested function
724 /// (allowing targets to access the original Function*).
725 /// It produces a token chain as output.
726 INIT_TRAMPOLINE,
727
728 /// ADJUST_TRAMPOLINE - This corresponds to the adjust_trampoline intrinsic.
729 /// It takes a pointer to the trampoline and produces a (possibly) new
730 /// pointer to the same trampoline with platform-specific adjustments
731 /// applied. The pointer it returns points to an executable block of code.
732 ADJUST_TRAMPOLINE,
733
734 /// TRAP - Trapping instruction
735 TRAP,
736
737 /// DEBUGTRAP - Trap intended to get the attention of a debugger.
738 DEBUGTRAP,
739
740 /// PREFETCH - This corresponds to a prefetch intrinsic. The first operand
741 /// is the chain. The other operands are the address to prefetch,
742 /// read / write specifier, locality specifier and instruction / data cache
743 /// specifier.
744 PREFETCH,
745
746 /// OUTCHAIN = ATOMIC_FENCE(INCHAIN, ordering, scope)
747 /// This corresponds to the fence instruction. It takes an input chain, and
748 /// two integer constants: an AtomicOrdering and a SynchronizationScope.
749 ATOMIC_FENCE,
750
751 /// Val, OUTCHAIN = ATOMIC_LOAD(INCHAIN, ptr)
752 /// This corresponds to "load atomic" instruction.
753 ATOMIC_LOAD,
754
755 /// OUTCHAIN = ATOMIC_STORE(INCHAIN, ptr, val)
756 /// This corresponds to "store atomic" instruction.
757 ATOMIC_STORE,
758
759 /// Val, OUTCHAIN = ATOMIC_CMP_SWAP(INCHAIN, ptr, cmp, swap)
760 /// For double-word atomic operations:
761 /// ValLo, ValHi, OUTCHAIN = ATOMIC_CMP_SWAP(INCHAIN, ptr, cmpLo, cmpHi,
762 /// swapLo, swapHi)
763 /// This corresponds to the cmpxchg instruction.
764 ATOMIC_CMP_SWAP,
765
766 /// Val, Success, OUTCHAIN
767 /// = ATOMIC_CMP_SWAP_WITH_SUCCESS(INCHAIN, ptr, cmp, swap)
768 /// N.b. this is still a strong cmpxchg operation, so
769 /// Success == "Val == cmp".
770 ATOMIC_CMP_SWAP_WITH_SUCCESS,
771
772 /// Val, OUTCHAIN = ATOMIC_SWAP(INCHAIN, ptr, amt)
773 /// Val, OUTCHAIN = ATOMIC_LOAD_[OpName](INCHAIN, ptr, amt)
774 /// For double-word atomic operations:
775 /// ValLo, ValHi, OUTCHAIN = ATOMIC_SWAP(INCHAIN, ptr, amtLo, amtHi)
776 /// ValLo, ValHi, OUTCHAIN = ATOMIC_LOAD_[OpName](INCHAIN, ptr, amtLo, amtHi)
777 /// These correspond to the atomicrmw instruction.
778 ATOMIC_SWAP,
779 ATOMIC_LOAD_ADD,
780 ATOMIC_LOAD_SUB,
781 ATOMIC_LOAD_AND,
782 ATOMIC_LOAD_CLR,
783 ATOMIC_LOAD_OR,
784 ATOMIC_LOAD_XOR,
785 ATOMIC_LOAD_NAND,
786 ATOMIC_LOAD_MIN,
787 ATOMIC_LOAD_MAX,
788 ATOMIC_LOAD_UMIN,
789 ATOMIC_LOAD_UMAX,
790
791 // Masked load and store - consecutive vector load and store operations
792 // with additional mask operand that prevents memory accesses to the
793 // masked-off lanes.
794 MLOAD, MSTORE,
795
796 // Masked gather and scatter - load and store operations for a vector of
797 // random addresses with additional mask operand that prevents memory
798 // accesses to the masked-off lanes.
799 MGATHER, MSCATTER,
800
801 /// This corresponds to the llvm.lifetime.* intrinsics. The first operand
802 /// is the chain and the second operand is the alloca pointer.
803 LIFETIME_START, LIFETIME_END,
804
805 /// GC_TRANSITION_START/GC_TRANSITION_END - These operators mark the
806 /// beginning and end of GC transition sequence, and carry arbitrary
807 /// information that target might need for lowering. The first operand is
808 /// a chain, the rest are specified by the target and not touched by the DAG
809 /// optimizers. GC_TRANSITION_START..GC_TRANSITION_END pairs may not be
810 /// nested.
811 GC_TRANSITION_START,
812 GC_TRANSITION_END,
813
814 /// GET_DYNAMIC_AREA_OFFSET - get offset from native SP to the address of
815 /// the most recent dynamic alloca. For most targets that would be 0, but
816 /// for some others (e.g. PowerPC, PowerPC64) that would be compile-time
817 /// known nonzero constant. The only operand here is the chain.
818 GET_DYNAMIC_AREA_OFFSET,
819
820 /// Generic reduction nodes. These nodes represent horizontal vector
821 /// reduction operations, producing a scalar result.
822 /// The STRICT variants perform reductions in sequential order. The first
823 /// operand is an initial scalar accumulator value, and the second operand
824 /// is the vector to reduce.
825 VECREDUCE_STRICT_FADD, VECREDUCE_STRICT_FMUL,
826 /// These reductions are non-strict, and have a single vector operand.
827 VECREDUCE_FADD, VECREDUCE_FMUL,
828 VECREDUCE_ADD, VECREDUCE_MUL,
829 VECREDUCE_AND, VECREDUCE_OR, VECREDUCE_XOR,
830 VECREDUCE_SMAX, VECREDUCE_SMIN, VECREDUCE_UMAX, VECREDUCE_UMIN,
831 /// FMIN/FMAX nodes can have flags, for NaN/NoNaN variants.
832 VECREDUCE_FMAX, VECREDUCE_FMIN,
833
834 /// BUILTIN_OP_END - This must be the last enum value in this list.
835 /// The target-specific pre-isel opcode values start here.
836 BUILTIN_OP_END
837 };
838
839 /// FIRST_TARGET_MEMORY_OPCODE - Target-specific pre-isel operations
840 /// which do not reference a specific memory location should be less than
841 /// this value. Those that do must not be less than this value, and can
842 /// be used with SelectionDAG::getMemIntrinsicNode.
843 static const int FIRST_TARGET_MEMORY_OPCODE = BUILTIN_OP_END+400;
844
845 //===--------------------------------------------------------------------===//
846 /// MemIndexedMode enum - This enum defines the load / store indexed
847 /// addressing modes.
848 ///
849 /// UNINDEXED "Normal" load / store. The effective address is already
850 /// computed and is available in the base pointer. The offset
851 /// operand is always undefined. In addition to producing a
852 /// chain, an unindexed load produces one value (result of the
853 /// load); an unindexed store does not produce a value.
854 ///
855 /// PRE_INC Similar to the unindexed mode where the effective address is
856 /// PRE_DEC the value of the base pointer add / subtract the offset.
857 /// It considers the computation as being folded into the load /
858 /// store operation (i.e. the load / store does the address
859 /// computation as well as performing the memory transaction).
860 /// The base operand is always undefined. In addition to
861 /// producing a chain, pre-indexed load produces two values
862 /// (result of the load and the result of the address
863 /// computation); a pre-indexed store produces one value (result
864 /// of the address computation).
865 ///
866 /// POST_INC The effective address is the value of the base pointer. The
867 /// POST_DEC value of the offset operand is then added to / subtracted
868 /// from the base after memory transaction. In addition to
869 /// producing a chain, post-indexed load produces two values
870 /// (the result of the load and the result of the base +/- offset
871 /// computation); a post-indexed store produces one value (the
872 /// the result of the base +/- offset computation).
873 enum MemIndexedMode {
874 UNINDEXED = 0,
875 PRE_INC,
876 PRE_DEC,
877 POST_INC,
878 POST_DEC
879 };
880
881 static const int LAST_INDEXED_MODE = POST_DEC + 1;
882
883 //===--------------------------------------------------------------------===//
884 /// LoadExtType enum - This enum defines the three variants of LOADEXT
885 /// (load with extension).
886 ///
887 /// SEXTLOAD loads the integer operand and sign extends it to a larger
888 /// integer result type.
889 /// ZEXTLOAD loads the integer operand and zero extends it to a larger
890 /// integer result type.
891 /// EXTLOAD is used for two things: floating point extending loads and
892 /// integer extending loads [the top bits are undefined].
893 enum LoadExtType {
894 NON_EXTLOAD = 0,
895 EXTLOAD,
896 SEXTLOAD,
897 ZEXTLOAD
898 };
899
900 static const int LAST_LOADEXT_TYPE = ZEXTLOAD + 1;
901
902 NodeType getExtForLoadExtType(bool IsFP, LoadExtType);
903
904 //===--------------------------------------------------------------------===//
905 /// ISD::CondCode enum - These are ordered carefully to make the bitfields
906 /// below work out, when considering SETFALSE (something that never exists
907 /// dynamically) as 0. "U" -> Unsigned (for integer operands) or Unordered
908 /// (for floating point), "L" -> Less than, "G" -> Greater than, "E" -> Equal
909 /// to. If the "N" column is 1, the result of the comparison is undefined if
910 /// the input is a NAN.
911 ///
912 /// All of these (except for the 'always folded ops') should be handled for
913 /// floating point. For integer, only the SETEQ,SETNE,SETLT,SETLE,SETGT,
914 /// SETGE,SETULT,SETULE,SETUGT, and SETUGE opcodes are used.
915 ///
916 /// Note that these are laid out in a specific order to allow bit-twiddling
917 /// to transform conditions.
918 enum CondCode {
919 // Opcode N U L G E Intuitive operation
920 SETFALSE, // 0 0 0 0 Always false (always folded)
921 SETOEQ, // 0 0 0 1 True if ordered and equal
922 SETOGT, // 0 0 1 0 True if ordered and greater than
923 SETOGE, // 0 0 1 1 True if ordered and greater than or equal
924 SETOLT, // 0 1 0 0 True if ordered and less than
925 SETOLE, // 0 1 0 1 True if ordered and less than or equal
926 SETONE, // 0 1 1 0 True if ordered and operands are unequal
927 SETO, // 0 1 1 1 True if ordered (no nans)
928 SETUO, // 1 0 0 0 True if unordered: isnan(X) | isnan(Y)
929 SETUEQ, // 1 0 0 1 True if unordered or equal
930 SETUGT, // 1 0 1 0 True if unordered or greater than
931 SETUGE, // 1 0 1 1 True if unordered, greater than, or equal
932 SETULT, // 1 1 0 0 True if unordered or less than
933 SETULE, // 1 1 0 1 True if unordered, less than, or equal
934 SETUNE, // 1 1 1 0 True if unordered or not equal
935 SETTRUE, // 1 1 1 1 Always true (always folded)
936 // Don't care operations: undefined if the input is a nan.
937 SETFALSE2, // 1 X 0 0 0 Always false (always folded)
938 SETEQ, // 1 X 0 0 1 True if equal
939 SETGT, // 1 X 0 1 0 True if greater than
940 SETGE, // 1 X 0 1 1 True if greater than or equal
941 SETLT, // 1 X 1 0 0 True if less than
942 SETLE, // 1 X 1 0 1 True if less than or equal
943 SETNE, // 1 X 1 1 0 True if not equal
944 SETTRUE2, // 1 X 1 1 1 Always true (always folded)
945
946 SETCC_INVALID // Marker value.
947 };
948
949 /// Return true if this is a setcc instruction that performs a signed
950 /// comparison when used with integer operands.
951 inline bool isSignedIntSetCC(CondCode Code) {
952 return Code == SETGT || Code == SETGE || Code == SETLT || Code == SETLE;
953 }
954
955 /// Return true if this is a setcc instruction that performs an unsigned
956 /// comparison when used with integer operands.
957 inline bool isUnsignedIntSetCC(CondCode Code) {
958 return Code == SETUGT || Code == SETUGE || Code == SETULT || Code == SETULE;
959 }
960
961 /// Return true if the specified condition returns true if the two operands to
962 /// the condition are equal. Note that if one of the two operands is a NaN,
963 /// this value is meaningless.
964 inline bool isTrueWhenEqual(CondCode Cond) {
965 return ((int)Cond & 1) != 0;
966 }
967
968 /// This function returns 0 if the condition is always false if an operand is
969 /// a NaN, 1 if the condition is always true if the operand is a NaN, and 2 if
970 /// the condition is undefined if the operand is a NaN.
971 inline unsigned getUnorderedFlavor(CondCode Cond) {
972 return ((int)Cond >> 3) & 3;
973 }
974
975 /// Return the operation corresponding to !(X op Y), where 'op' is a valid
976 /// SetCC operation.
977 CondCode getSetCCInverse(CondCode Operation, bool isInteger);
978
979 /// Return the operation corresponding to (Y op X) when given the operation
980 /// for (X op Y).
981 CondCode getSetCCSwappedOperands(CondCode Operation);
982
983 /// Return the result of a logical OR between different comparisons of
984 /// identical values: ((X op1 Y) | (X op2 Y)). This function returns
985 /// SETCC_INVALID if it is not possible to represent the resultant comparison.
986 CondCode getSetCCOrOperation(CondCode Op1, CondCode Op2, bool isInteger);
987
988 /// Return the result of a logical AND between different comparisons of
989 /// identical values: ((X op1 Y) & (X op2 Y)). This function returns
990 /// SETCC_INVALID if it is not possible to represent the resultant comparison.
991 CondCode getSetCCAndOperation(CondCode Op1, CondCode Op2, bool isInteger);
992
993} // end llvm::ISD namespace
994
995} // end llvm namespace
996
997#endif