1//===--- arm_neon.td - ARM NEON compiler interface ------------------------===//
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 defines the TableGen definitions from which the ARM NEON header
11//  file will be generated.  See ARM document DUI0348B.
12//
13//===----------------------------------------------------------------------===//
14//
15// Each intrinsic is a subclass of the Inst class. An intrinsic can either
16// generate a __builtin_* call or it can expand to a set of generic operations.
17//
18// The operations are subclasses of Operation providing a list of DAGs, the
19// last of which is the return value. The available DAG nodes are documented
20// below.
21//
22//===----------------------------------------------------------------------===//
23
24// The base Operation class. All operations must subclass this.
25class Operation<list<dag> ops=[]> {
26  list<dag> Ops = ops;
27  bit Unavailable = 0;
28}
29// An operation that only contains a single DAG.
30class Op<dag op> : Operation<[op]>;
31// A shorter version of Operation - takes a list of DAGs. The last of these will
32// be the return value.
33class LOp<list<dag> ops> : Operation<ops>;
34
35// These defs and classes are used internally to implement the SetTheory
36// expansion and should be ignored.
37foreach Index = 0-63 in
38  def sv##Index;
39class MaskExpand;
40
41//===----------------------------------------------------------------------===//
42// Available operations
43//===----------------------------------------------------------------------===//
44
45// DAG arguments can either be operations (documented below) or variables.
46// Variables are prefixed with '$'. There are variables for each input argument,
47// with the name $pN, where N starts at zero. So the zero'th argument will be
48// $p0, the first $p1 etc.
49
50// op - Binary or unary operator, depending on the number of arguments. The
51//      operator itself is just treated as a raw string and is not checked.
52// example: (op "+", $p0, $p1) -> "__p0 + __p1".
53//          (op "-", $p0)      -> "-__p0"
54def op;
55// call - Invoke another intrinsic. The input types are type checked and
56//        disambiguated. If there is no intrinsic defined that takes
57//        the given types (or if there is a type ambiguity) an error is
58//        generated at tblgen time. The name of the intrinsic is the raw
59//        name as given to the Inst class (not mangled).
60// example: (call "vget_high", $p0) -> "vgetq_high_s16(__p0)"
61//            (assuming $p0 has type int16x8_t).
62def call;
63// cast - Perform a cast to a different type. This gets emitted as a static
64//        C-style cast. For a pure reinterpret cast (T x = *(T*)&y), use
65//        "bitcast".
66//
67//        The syntax is (cast MOD* VAL). The last argument is the value to
68//        cast, preceded by a sequence of type modifiers. The target type
69//        starts off as the type of VAL, and is modified by MOD in sequence.
70//        The available modifiers are:
71//          - $X  - Take the type of parameter/variable X. For example:
72//                  (cast $p0, $p1) would cast $p1 to the type of $p0.
73//          - "R" - The type of the return type.
74//          - A typedef string - A NEON or stdint.h type that is then parsed.
75//                               for example: (cast "uint32x4_t", $p0).
76//          - "U" - Make the type unsigned.
77//          - "S" - Make the type signed.
78//          - "H" - Halve the number of lanes in the type.
79//          - "D" - Double the number of lanes in the type.
80//          - "8" - Convert type to an equivalent vector of 8-bit signed
81//                  integers.
82// example: (cast "R", "U", $p0) -> "(uint32x4_t)__p0" (assuming the return
83//           value is of type "int32x4_t".
84//          (cast $p0, "D", "8", $p1) -> "(int8x16_t)__p1" (assuming __p0
85//           has type float64x1_t or any other vector type of 64 bits).
86//          (cast "int32_t", $p2) -> "(int32_t)__p2"
87def cast;
88// bitcast - Same as "cast", except a reinterpret-cast is produced:
89//             (bitcast "T", $p0) -> "*(T*)&__p0".
90//           The VAL argument is saved to a temporary so it can be used
91//           as an l-value.
92def bitcast;
93// dup - Take a scalar argument and create a vector by duplicating it into
94//       all lanes. The type of the vector is the base type of the intrinsic.
95// example: (dup $p1) -> "(uint32x2_t) {__p1, __p1}" (assuming the base type
96//          is uint32x2_t).
97def dup;
98// splat - Take a vector and a lane index, and return a vector of the same type
99//         containing repeated instances of the source vector at the lane index.
100// example: (splat $p0, $p1) ->
101//            "__builtin_shufflevector(__p0, __p0, __p1, __p1, __p1, __p1)"
102//          (assuming __p0 has four elements).
103def splat;
104// save_temp - Create a temporary (local) variable. The variable takes a name
105//             based on the zero'th parameter and can be referenced using
106//             using that name in subsequent DAGs in the same
107//             operation. The scope of a temp is the operation. If a variable
108//             with the given name already exists, an error will be given at
109//             tblgen time.
110// example: [(save_temp $var, (call "foo", $p0)),
111//           (op "+", $var, $p1)] ->
112//              "int32x2_t __var = foo(__p0); return __var + __p1;"
113def save_temp;
114// name_replace - Return the name of the current intrinsic with the first
115//                argument replaced by the second argument. Raises an error if
116//                the first argument does not exist in the intrinsic name.
117// example: (call (name_replace "_high_", "_"), $p0) (to call the non-high
118//            version of this intrinsic).
119def name_replace;
120// literal - Create a literal piece of code. The code is treated as a raw
121//           string, and must be given a type. The type is a stdint.h or
122//           NEON intrinsic type as given to (cast).
123// example: (literal "int32_t", "0")
124def literal;
125// shuffle - Create a vector shuffle. The syntax is (shuffle ARG0, ARG1, MASK).
126//           The MASK argument is a set of elements. The elements are generated
127//           from the two special defs "mask0" and "mask1". "mask0" expands to
128//           the lane indices in sequence for ARG0, and "mask1" expands to
129//           the lane indices in sequence for ARG1. They can be used as-is, e.g.
130//
131//             (shuffle $p0, $p1, mask0) -> $p0
132//             (shuffle $p0, $p1, mask1) -> $p1
133//
134//           or, more usefully, they can be manipulated using the SetTheory
135//           operators plus some extra operators defined in the NEON emitter.
136//           The operators are described below.
137// example: (shuffle $p0, $p1, (add (highhalf mask0), (highhalf mask1))) ->
138//            A concatenation of the high halves of the input vectors.
139def shuffle;
140
141// add, interleave, decimate: These set operators are vanilla SetTheory
142// operators and take their normal definition.
143def add;
144def interleave;
145def decimate;
146// rotl - Rotate set left by a number of elements.
147// example: (rotl mask0, 3) -> [3, 4, 5, 6, 0, 1, 2]
148def rotl;
149// rotl - Rotate set right by a number of elements.
150// example: (rotr mask0, 3) -> [4, 5, 6, 0, 1, 2, 3]
151def rotr;
152// highhalf - Take only the high half of the input.
153// example: (highhalf mask0) -> [4, 5, 6, 7] (assuming mask0 had 8 elements)
154def highhalf;
155// highhalf - Take only the low half of the input.
156// example: (lowhalf mask0) -> [0, 1, 2, 3] (assuming mask0 had 8 elements)
157def lowhalf;
158// rev - Perform a variable-width reversal of the elements. The zero'th argument
159//       is a width in bits to reverse. The lanes this maps to is determined
160//       based on the element width of the underlying type.
161// example: (rev 32, mask0) -> [3, 2, 1, 0, 7, 6, 5, 4] (if 8-bit elements)
162// example: (rev 32, mask0) -> [1, 0, 3, 2]             (if 16-bit elements)
163def rev;
164// mask0 - The initial sequence of lanes for shuffle ARG0
165def mask0 : MaskExpand;
166// mask0 - The initial sequence of lanes for shuffle ARG1
167def mask1 : MaskExpand;
168
169def OP_NONE  : Operation;
170def OP_UNAVAILABLE : Operation {
171  let Unavailable = 1;
172}
173
174//===----------------------------------------------------------------------===//
175// Instruction definitions
176//===----------------------------------------------------------------------===//
177
178// Every intrinsic subclasses "Inst". An intrinsic has a name, a prototype and
179// a sequence of typespecs.
180//
181// The name is the base name of the intrinsic, for example "vget_lane". This is
182// then mangled by the tblgen backend to add type information ("vget_lane_s16").
183//
184// A typespec is a sequence of uppercase characters (modifiers) followed by one
185// lowercase character. A typespec encodes a particular "base type" of the
186// intrinsic.
187//
188// An example typespec is "Qs" - quad-size short - uint16x8_t. The available
189// typespec codes are given below.
190//
191// The string given to an Inst class is a sequence of typespecs. The intrinsic
192// is instantiated for every typespec in the sequence. For example "sdQsQd".
193//
194// The prototype is a string that defines the return type of the intrinsic
195// and the type of each argument. The return type and every argument gets a
196// "modifier" that can change in some way the "base type" of the intrinsic.
197//
198// The modifier 'd' means "default" and does not modify the base type in any
199// way. The available modifiers are given below.
200//
201// Typespecs
202// ---------
203// c: char
204// s: short
205// i: int
206// l: long
207// k: 128-bit long
208// f: float
209// h: half-float
210// d: double
211//
212// Typespec modifiers
213// ------------------
214// S: scalar, only used for function mangling.
215// U: unsigned
216// Q: 128b
217// H: 128b without mangling 'q'
218// P: polynomial
219//
220// Prototype modifiers
221// -------------------
222// prototype: return (arg, arg, ...)
223//
224// v: void
225// t: best-fit integer (int/poly args)
226// x: signed integer   (int/float args)
227// u: unsigned integer (int/float args)
228// f: float (int args)
229// F: double (int args)
230// d: default
231// g: default, ignore 'Q' size modifier.
232// j: default, force 'Q' size modifier.
233// w: double width elements, same num elts
234// n: double width elements, half num elts
235// h: half width elements, double num elts
236// q: half width elements, quad num elts
237// e: half width elements, double num elts, unsigned
238// m: half width elements, same num elts
239// i: constant int
240// l: constant uint64
241// s: scalar of element type
242// z: scalar of half width element type, signed
243// r: scalar of double width element type, signed
244// a: scalar of element type (splat to vector type)
245// b: scalar of unsigned integer/long type (int/float args)
246// $: scalar of signed integer/long type (int/float args)
247// y: scalar of float
248// o: scalar of double
249// k: default elt width, double num elts
250// 2,3,4: array of default vectors
251// B,C,D: array of default elts, force 'Q' size modifier.
252// p: pointer type
253// c: const pointer type
254
255// Every intrinsic subclasses Inst.
256class Inst <string n, string p, string t, Operation o> {
257  string Name = n;
258  string Prototype = p;
259  string Types = t;
260  string ArchGuard = "";
261
262  Operation Operation = o;
263  bit CartesianProductOfTypes = 0;
264  bit BigEndianSafe = 0;
265  bit isShift = 0;
266  bit isScalarShift = 0;
267  bit isScalarNarrowShift = 0;
268  bit isVCVT_N = 0;
269  // For immediate checks: the immediate will be assumed to specify the lane of
270  // a Q register. Only used for intrinsics which end up calling polymorphic
271  // builtins.
272  bit isLaneQ = 0;
273
274  // Certain intrinsics have different names than their representative
275  // instructions. This field allows us to handle this correctly when we
276  // are generating tests.
277  string InstName = "";
278
279  // Certain intrinsics even though they are not a WOpInst or LOpInst,
280  // generate a WOpInst/LOpInst instruction (see below for definition
281  // of a WOpInst/LOpInst). For testing purposes we need to know
282  // this. Ex: vset_lane which outputs vmov instructions.
283  bit isHiddenWInst = 0;
284  bit isHiddenLInst = 0;
285}
286
287// The following instruction classes are implemented via builtins.
288// These declarations are used to generate Builtins.def:
289//
290// SInst: Instruction with signed/unsigned suffix (e.g., "s8", "u8", "p8")
291// IInst: Instruction with generic integer suffix (e.g., "i8")
292// WInst: Instruction with only bit size suffix (e.g., "8")
293class SInst<string n, string p, string t> : Inst<n, p, t, OP_NONE> {}
294class IInst<string n, string p, string t> : Inst<n, p, t, OP_NONE> {}
295class WInst<string n, string p, string t> : Inst<n, p, t, OP_NONE> {}
296
297// The following instruction classes are implemented via operators
298// instead of builtins. As such these declarations are only used for
299// the purpose of generating tests.
300//
301// SOpInst:       Instruction with signed/unsigned suffix (e.g., "s8",
302//                "u8", "p8").
303// IOpInst:       Instruction with generic integer suffix (e.g., "i8").
304// WOpInst:       Instruction with bit size only suffix (e.g., "8").
305// LOpInst:       Logical instruction with no bit size suffix.
306// NoTestOpInst:  Intrinsic that has no corresponding instruction.
307class SOpInst<string n, string p, string t, Operation o> : Inst<n, p, t, o> {}
308class IOpInst<string n, string p, string t, Operation o> : Inst<n, p, t, o> {}
309class WOpInst<string n, string p, string t, Operation o> : Inst<n, p, t, o> {}
310class LOpInst<string n, string p, string t, Operation o> : Inst<n, p, t, o> {}
311class NoTestOpInst<string n, string p, string t, Operation o> : Inst<n, p, t, o> {}
312
313//===----------------------------------------------------------------------===//
314// Operations
315//===----------------------------------------------------------------------===//
316
317def OP_ADD      : Op<(op "+", $p0, $p1)>;
318def OP_ADDL     : Op<(op "+", (call "vmovl", $p0), (call "vmovl", $p1))>;
319def OP_ADDLHi   : Op<(op "+", (call "vmovl_high", $p0),
320                              (call "vmovl_high", $p1))>;
321def OP_ADDW     : Op<(op "+", $p0, (call "vmovl", $p1))>;
322def OP_ADDWHi   : Op<(op "+", $p0, (call "vmovl_high", $p1))>;
323def OP_SUB      : Op<(op "-", $p0, $p1)>;
324def OP_SUBL     : Op<(op "-", (call "vmovl", $p0), (call "vmovl", $p1))>;
325def OP_SUBLHi   : Op<(op "-", (call "vmovl_high", $p0),
326                              (call "vmovl_high", $p1))>;
327def OP_SUBW     : Op<(op "-", $p0, (call "vmovl", $p1))>;
328def OP_SUBWHi   : Op<(op "-", $p0, (call "vmovl_high", $p1))>;
329def OP_MUL      : Op<(op "*", $p0, $p1)>;
330def OP_MLA      : Op<(op "+", $p0, (op "*", $p1, $p2))>;
331def OP_MLAL     : Op<(op "+", $p0, (call "vmull", $p1, $p2))>;
332def OP_MULLHi   : Op<(call "vmull", (call "vget_high", $p0),
333                                    (call "vget_high", $p1))>;
334def OP_MULLHi_P64 : Op<(call "vmull",
335                         (cast "poly64_t", (call "vget_high", $p0)),
336                         (cast "poly64_t", (call "vget_high", $p1)))>;
337def OP_MULLHi_N : Op<(call "vmull_n", (call "vget_high", $p0), $p1)>;
338def OP_MLALHi   : Op<(call "vmlal", $p0, (call "vget_high", $p1),
339                                         (call "vget_high", $p2))>;
340def OP_MLALHi_N : Op<(call "vmlal_n", $p0, (call "vget_high", $p1), $p2)>;
341def OP_MLS      : Op<(op "-", $p0, (op "*", $p1, $p2))>;
342def OP_MLSL     : Op<(op "-", $p0, (call "vmull", $p1, $p2))>;
343def OP_MLSLHi   : Op<(call "vmlsl", $p0, (call "vget_high", $p1),
344                                         (call "vget_high", $p2))>;
345def OP_MLSLHi_N : Op<(call "vmlsl_n", $p0, (call "vget_high", $p1), $p2)>;
346def OP_MUL_N    : Op<(op "*", $p0, (dup $p1))>;
347def OP_MLA_N    : Op<(op "+", $p0, (op "*", $p1, (dup $p2)))>;
348def OP_MLS_N    : Op<(op "-", $p0, (op "*", $p1, (dup $p2)))>;
349def OP_FMLA_N   : Op<(call "vfma", $p0, $p1, (dup $p2))>;
350def OP_FMLS_N   : Op<(call "vfms", $p0, $p1, (dup $p2))>;
351def OP_MLAL_N   : Op<(op "+", $p0, (call "vmull", $p1, (dup $p2)))>;
352def OP_MLSL_N   : Op<(op "-", $p0, (call "vmull", $p1, (dup $p2)))>;
353def OP_MUL_LN   : Op<(op "*", $p0, (splat $p1, $p2))>;
354def OP_MULX_LN  : Op<(call "vmulx", $p0, (splat $p1, $p2))>;
355def OP_MULL_LN  : Op<(call "vmull", $p0, (splat $p1, $p2))>;
356def OP_MULLHi_LN: Op<(call "vmull", (call "vget_high", $p0), (splat $p1, $p2))>;
357def OP_MLA_LN   : Op<(op "+", $p0, (op "*", $p1, (splat $p2, $p3)))>;
358def OP_MLS_LN   : Op<(op "-", $p0, (op "*", $p1, (splat $p2, $p3)))>;
359def OP_MLAL_LN  : Op<(op "+", $p0, (call "vmull", $p1, (splat $p2, $p3)))>;
360def OP_MLALHi_LN: Op<(op "+", $p0, (call "vmull", (call "vget_high", $p1),
361                                                  (splat $p2, $p3)))>;
362def OP_MLSL_LN  : Op<(op "-", $p0, (call "vmull", $p1, (splat $p2, $p3)))>;
363def OP_MLSLHi_LN : Op<(op "-", $p0, (call "vmull", (call "vget_high", $p1),
364                                                   (splat $p2, $p3)))>;
365def OP_QDMULL_LN : Op<(call "vqdmull", $p0, (splat $p1, $p2))>;
366def OP_QDMULLHi_LN : Op<(call "vqdmull", (call "vget_high", $p0),
367                                         (splat $p1, $p2))>;
368def OP_QDMLAL_LN : Op<(call "vqdmlal", $p0, $p1, (splat $p2, $p3))>;
369def OP_QDMLALHi_LN : Op<(call "vqdmlal", $p0, (call "vget_high", $p1),
370                                              (splat $p2, $p3))>;
371def OP_QDMLSL_LN : Op<(call "vqdmlsl", $p0, $p1, (splat $p2, $p3))>;
372def OP_QDMLSLHi_LN : Op<(call "vqdmlsl", $p0, (call "vget_high", $p1),
373                                              (splat $p2, $p3))>;
374def OP_QDMULH_LN : Op<(call "vqdmulh", $p0, (splat $p1, $p2))>;
375def OP_QRDMULH_LN : Op<(call "vqrdmulh", $p0, (splat $p1, $p2))>;
376def OP_QRDMLAH : Op<(call "vqadd", $p0, (call "vqrdmulh", $p1, $p2))>;
377def OP_QRDMLSH : Op<(call "vqsub", $p0, (call "vqrdmulh", $p1, $p2))>;
378def OP_QRDMLAH_LN : Op<(call "vqadd", $p0, (call "vqrdmulh", $p1, (splat $p2, $p3)))>;
379def OP_QRDMLSH_LN : Op<(call "vqsub", $p0, (call "vqrdmulh", $p1, (splat $p2, $p3)))>;
380def OP_FMS_LN   : Op<(call "vfma_lane", $p0, $p1, (op "-", $p2), $p3)>;
381def OP_FMS_LNQ  : Op<(call "vfma_laneq", $p0, $p1, (op "-", $p2), $p3)>;
382def OP_TRN1     : Op<(shuffle $p0, $p1, (interleave (decimate mask0, 2),
383                                                    (decimate mask1, 2)))>;
384def OP_ZIP1     : Op<(shuffle $p0, $p1, (lowhalf (interleave mask0, mask1)))>;
385def OP_UZP1     : Op<(shuffle $p0, $p1, (add (decimate mask0, 2),
386                                             (decimate mask1, 2)))>;
387def OP_TRN2     : Op<(shuffle $p0, $p1, (interleave
388                                          (decimate (rotl mask0, 1), 2),
389                                          (decimate (rotl mask1, 1), 2)))>;
390def OP_ZIP2     : Op<(shuffle $p0, $p1, (highhalf (interleave mask0, mask1)))>;
391def OP_UZP2     : Op<(shuffle $p0, $p1, (add (decimate (rotl mask0, 1), 2),
392                                             (decimate (rotl mask1, 1), 2)))>;
393def OP_EQ       : Op<(cast "R", (op "==", $p0, $p1))>;
394def OP_GE       : Op<(cast "R", (op ">=", $p0, $p1))>;
395def OP_LE       : Op<(cast "R", (op "<=", $p0, $p1))>;
396def OP_GT       : Op<(cast "R", (op ">", $p0, $p1))>;
397def OP_LT       : Op<(cast "R", (op "<", $p0, $p1))>;
398def OP_NEG      : Op<(op "-", $p0)>;
399def OP_NOT      : Op<(op "~", $p0)>;
400def OP_AND      : Op<(op "&", $p0, $p1)>;
401def OP_OR       : Op<(op "|", $p0, $p1)>;
402def OP_XOR      : Op<(op "^", $p0, $p1)>;
403def OP_ANDN     : Op<(op "&", $p0, (op "~", $p1))>;
404def OP_ORN      : Op<(op "|", $p0, (op "~", $p1))>;
405def OP_CAST     : Op<(cast "R", $p0)>;
406def OP_HI       : Op<(shuffle $p0, $p0, (highhalf mask0))>;
407def OP_LO       : Op<(shuffle $p0, $p0, (lowhalf mask0))>;
408def OP_CONC     : Op<(shuffle $p0, $p1, (add mask0, mask1))>;
409def OP_DUP      : Op<(dup $p0)>;
410def OP_DUP_LN   : Op<(splat $p0, $p1)>;
411def OP_SEL      : Op<(cast "R", (op "|",
412                                    (op "&", $p0, (cast $p0, $p1)),
413                                    (op "&", (op "~", $p0), (cast $p0, $p2))))>;
414def OP_REV16    : Op<(shuffle $p0, $p0, (rev 16, mask0))>;
415def OP_REV32    : Op<(shuffle $p0, $p0, (rev 32, mask0))>;
416def OP_REV64    : Op<(shuffle $p0, $p0, (rev 64, mask0))>;
417def OP_XTN      : Op<(call "vcombine", $p0, (call "vmovn", $p1))>;
418def OP_SQXTUN   : Op<(call "vcombine", (cast $p0, "U", $p0),
419                                       (call "vqmovun", $p1))>;
420def OP_QXTN     : Op<(call "vcombine", $p0, (call "vqmovn", $p1))>;
421def OP_VCVT_NA_HI_F16 : Op<(call "vcombine", $p0, (call "vcvt_f16_f32", $p1))>;
422def OP_VCVT_NA_HI_F32 : Op<(call "vcombine", $p0, (call "vcvt_f32_f64", $p1))>;
423def OP_VCVT_EX_HI_F32 : Op<(call "vcvt_f32_f16", (call "vget_high", $p0))>;
424def OP_VCVT_EX_HI_F64 : Op<(call "vcvt_f64_f32", (call "vget_high", $p0))>;
425def OP_VCVTX_HI : Op<(call "vcombine", $p0, (call "vcvtx_f32", $p1))>;
426def OP_REINT    : Op<(cast "R", $p0)>;
427def OP_ADDHNHi  : Op<(call "vcombine", $p0, (call "vaddhn", $p1, $p2))>;
428def OP_RADDHNHi : Op<(call "vcombine", $p0, (call "vraddhn", $p1, $p2))>;
429def OP_SUBHNHi  : Op<(call "vcombine", $p0, (call "vsubhn", $p1, $p2))>;
430def OP_RSUBHNHi : Op<(call "vcombine", $p0, (call "vrsubhn", $p1, $p2))>;
431def OP_ABDL     : Op<(cast "R", (call "vmovl", (cast $p0, "U",
432                                                     (call "vabd", $p0, $p1))))>;
433def OP_ABDLHi   : Op<(call "vabdl", (call "vget_high", $p0),
434                                    (call "vget_high", $p1))>;
435def OP_ABA      : Op<(op "+", $p0, (call "vabd", $p1, $p2))>;
436def OP_ABAL     : Op<(op "+", $p0, (call "vabdl", $p1, $p2))>;
437def OP_ABALHi   : Op<(call "vabal", $p0, (call "vget_high", $p1),
438                                       (call "vget_high", $p2))>;
439def OP_QDMULLHi : Op<(call "vqdmull", (call "vget_high", $p0),
440                                      (call "vget_high", $p1))>;
441def OP_QDMULLHi_N : Op<(call "vqdmull_n", (call "vget_high", $p0), $p1)>;
442def OP_QDMLALHi : Op<(call "vqdmlal", $p0, (call "vget_high", $p1),
443                                           (call "vget_high", $p2))>;
444def OP_QDMLALHi_N : Op<(call "vqdmlal_n", $p0, (call "vget_high", $p1), $p2)>;
445def OP_QDMLSLHi : Op<(call "vqdmlsl", $p0, (call "vget_high", $p1),
446                                           (call "vget_high", $p2))>;
447def OP_QDMLSLHi_N : Op<(call "vqdmlsl_n", $p0, (call "vget_high", $p1), $p2)>;
448def OP_DIV  : Op<(op "/", $p0, $p1)>;
449def OP_LONG_HI : Op<(cast "R", (call (name_replace "_high_", "_"),
450                                                (call "vget_high", $p0), $p1))>;
451def OP_NARROW_HI : Op<(cast "R", (call "vcombine",
452                                       (cast "R", "H", $p0),
453                                       (cast "R", "H",
454                                           (call (name_replace "_high_", "_"),
455                                                 $p1, $p2))))>;
456def OP_MOVL_HI  : LOp<[(save_temp $a1, (call "vget_high", $p0)),
457                       (cast "R",
458                            (call "vshll_n", $a1, (literal "int32_t", "0")))]>;
459def OP_COPY_LN : Op<(call "vset_lane", (call "vget_lane", $p2, $p3), $p0, $p1)>;
460def OP_SCALAR_MUL_LN : Op<(op "*", $p0, (call "vget_lane", $p1, $p2))>;
461def OP_SCALAR_MULX_LN : Op<(call "vmulx", $p0, (call "vget_lane", $p1, $p2))>;
462def OP_SCALAR_VMULX_LN : LOp<[(save_temp $x, (call "vget_lane", $p0,
463                                                    (literal "int32_t", "0"))),
464                              (save_temp $y, (call "vget_lane", $p1, $p2)),
465                              (save_temp $z, (call "vmulx", $x, $y)),
466                              (call "vset_lane", $z, $p0, $p2)]>;
467def OP_SCALAR_VMULX_LNQ : LOp<[(save_temp $x, (call "vget_lane", $p0,
468                                                     (literal "int32_t", "0"))),
469                               (save_temp $y, (call "vget_lane", $p1, $p2)),
470                               (save_temp $z, (call "vmulx", $x, $y)),
471                               (call "vset_lane", $z, $p0, (literal "int32_t",
472                                                                     "0"))]>;
473class ScalarMulOp<string opname> :
474  Op<(call opname, $p0, (call "vget_lane", $p1, $p2))>;
475
476def OP_SCALAR_QDMULL_LN : ScalarMulOp<"vqdmull">;
477def OP_SCALAR_QDMULH_LN : ScalarMulOp<"vqdmulh">;
478def OP_SCALAR_QRDMULH_LN : ScalarMulOp<"vqrdmulh">;
479
480def OP_SCALAR_QRDMLAH_LN : Op<(call "vqadd", $p0, (call "vqrdmulh", $p1,
481                              (call "vget_lane", $p2, $p3)))>;
482def OP_SCALAR_QRDMLSH_LN : Op<(call "vqsub", $p0, (call "vqrdmulh", $p1,
483                              (call "vget_lane", $p2, $p3)))>;
484
485def OP_SCALAR_HALF_GET_LN : Op<(bitcast "float16_t",
486                                   (call "vget_lane",
487                                         (bitcast "int16x4_t", $p0), $p1))>;
488def OP_SCALAR_HALF_GET_LNQ : Op<(bitcast "float16_t",
489                                    (call "vget_lane",
490                                          (bitcast "int16x8_t", $p0), $p1))>;
491def OP_SCALAR_HALF_SET_LN : Op<(bitcast "float16x4_t",
492                                   (call "vset_lane",
493                                         (bitcast "int16_t", $p0),
494                                         (bitcast "int16x4_t", $p1), $p2))>;
495def OP_SCALAR_HALF_SET_LNQ : Op<(bitcast "float16x8_t",
496                                    (call "vset_lane",
497                                          (bitcast "int16_t", $p0),
498                                          (bitcast "int16x8_t", $p1), $p2))>;
499
500//===----------------------------------------------------------------------===//
501// Instructions
502//===----------------------------------------------------------------------===//
503
504////////////////////////////////////////////////////////////////////////////////
505// E.3.1 Addition
506def VADD    : IOpInst<"vadd", "ddd",
507                      "csilfUcUsUiUlQcQsQiQlQfQUcQUsQUiQUl", OP_ADD>;
508def VADDL   : SOpInst<"vaddl", "wdd", "csiUcUsUi", OP_ADDL>;
509def VADDW   : SOpInst<"vaddw", "wwd", "csiUcUsUi", OP_ADDW>;
510def VHADD   : SInst<"vhadd", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
511def VRHADD  : SInst<"vrhadd", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
512def VQADD   : SInst<"vqadd", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
513def VADDHN  : IInst<"vaddhn", "hkk", "silUsUiUl">;
514def VRADDHN : IInst<"vraddhn", "hkk", "silUsUiUl">;
515
516////////////////////////////////////////////////////////////////////////////////
517// E.3.2 Multiplication
518def VMUL     : IOpInst<"vmul", "ddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MUL>;
519def VMULP    : SInst<"vmul", "ddd", "PcQPc">;
520def VMLA     : IOpInst<"vmla", "dddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MLA>;
521def VMLAL    : SOpInst<"vmlal", "wwdd", "csiUcUsUi", OP_MLAL>;
522def VMLS     : IOpInst<"vmls", "dddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MLS>;
523def VMLSL    : SOpInst<"vmlsl", "wwdd", "csiUcUsUi", OP_MLSL>;
524def VQDMULH  : SInst<"vqdmulh", "ddd", "siQsQi">;
525def VQRDMULH : SInst<"vqrdmulh", "ddd", "siQsQi">;
526
527let ArchGuard = "defined(__ARM_FEATURE_QRDMX)" in {
528def VQRDMLAH : SOpInst<"vqrdmlah", "dddd", "siQsQi", OP_QRDMLAH>;
529def VQRDMLSH : SOpInst<"vqrdmlsh", "dddd", "siQsQi", OP_QRDMLSH>;
530}
531
532def VQDMLAL  : SInst<"vqdmlal", "wwdd", "si">;
533def VQDMLSL  : SInst<"vqdmlsl", "wwdd", "si">;
534def VMULL    : SInst<"vmull", "wdd", "csiUcUsUiPc">;
535def VQDMULL  : SInst<"vqdmull", "wdd", "si">;
536
537////////////////////////////////////////////////////////////////////////////////
538// E.3.3 Subtraction
539def VSUB    : IOpInst<"vsub", "ddd",
540                      "csilfUcUsUiUlQcQsQiQlQfQUcQUsQUiQUl", OP_SUB>;
541def VSUBL   : SOpInst<"vsubl", "wdd", "csiUcUsUi", OP_SUBL>;
542def VSUBW   : SOpInst<"vsubw", "wwd", "csiUcUsUi", OP_SUBW>;
543def VQSUB   : SInst<"vqsub", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
544def VHSUB   : SInst<"vhsub", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
545def VSUBHN  : IInst<"vsubhn", "hkk", "silUsUiUl">;
546def VRSUBHN : IInst<"vrsubhn", "hkk", "silUsUiUl">;
547
548////////////////////////////////////////////////////////////////////////////////
549// E.3.4 Comparison
550def VCEQ  : IOpInst<"vceq", "udd", "csifUcUsUiPcQcQsQiQfQUcQUsQUiQPc", OP_EQ>;
551def VCGE  : SOpInst<"vcge", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_GE>;
552let InstName = "vcge" in
553def VCLE  : SOpInst<"vcle", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_LE>;
554def VCGT  : SOpInst<"vcgt", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_GT>;
555let InstName = "vcgt" in
556def VCLT  : SOpInst<"vclt", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_LT>;
557let InstName = "vacge" in {
558def VCAGE : IInst<"vcage", "udd", "fQf">;
559def VCALE : IInst<"vcale", "udd", "fQf">;
560}
561let InstName = "vacgt" in {
562def VCAGT : IInst<"vcagt", "udd", "fQf">;
563def VCALT : IInst<"vcalt", "udd", "fQf">;
564}
565def VTST  : WInst<"vtst", "udd", "csiUcUsUiPcPsQcQsQiQUcQUsQUiQPcQPs">;
566
567////////////////////////////////////////////////////////////////////////////////
568// E.3.5 Absolute Difference
569def VABD  : SInst<"vabd", "ddd",  "csiUcUsUifQcQsQiQUcQUsQUiQf">;
570def VABDL : SOpInst<"vabdl", "wdd",  "csiUcUsUi", OP_ABDL>;
571def VABA  : SOpInst<"vaba", "dddd", "csiUcUsUiQcQsQiQUcQUsQUi", OP_ABA>;
572def VABAL : SOpInst<"vabal", "wwdd", "csiUcUsUi", OP_ABAL>;
573
574////////////////////////////////////////////////////////////////////////////////
575// E.3.6 Max/Min
576def VMAX : SInst<"vmax", "ddd", "csiUcUsUifQcQsQiQUcQUsQUiQf">;
577def VMIN : SInst<"vmin", "ddd", "csiUcUsUifQcQsQiQUcQUsQUiQf">;
578
579////////////////////////////////////////////////////////////////////////////////
580// E.3.7 Pairwise Addition
581def VPADD  : IInst<"vpadd", "ddd", "csiUcUsUif">;
582def VPADDL : SInst<"vpaddl", "nd",  "csiUcUsUiQcQsQiQUcQUsQUi">;
583def VPADAL : SInst<"vpadal", "nnd", "csiUcUsUiQcQsQiQUcQUsQUi">;
584
585////////////////////////////////////////////////////////////////////////////////
586// E.3.8-9 Folding Max/Min
587def VPMAX : SInst<"vpmax", "ddd", "csiUcUsUif">;
588def VPMIN : SInst<"vpmin", "ddd", "csiUcUsUif">;
589
590////////////////////////////////////////////////////////////////////////////////
591// E.3.10 Reciprocal/Sqrt
592def VRECPS  : IInst<"vrecps", "ddd", "fQf">;
593def VRSQRTS : IInst<"vrsqrts", "ddd", "fQf">;
594
595////////////////////////////////////////////////////////////////////////////////
596// E.3.11 Shifts by signed variable
597def VSHL   : SInst<"vshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
598def VQSHL  : SInst<"vqshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
599def VRSHL  : SInst<"vrshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
600def VQRSHL : SInst<"vqrshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
601
602////////////////////////////////////////////////////////////////////////////////
603// E.3.12 Shifts by constant
604let isShift = 1 in {
605def VSHR_N     : SInst<"vshr_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
606def VSHL_N     : IInst<"vshl_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
607def VRSHR_N    : SInst<"vrshr_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
608def VSRA_N     : SInst<"vsra_n", "dddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
609def VRSRA_N    : SInst<"vrsra_n", "dddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
610def VQSHL_N    : SInst<"vqshl_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
611def VQSHLU_N   : SInst<"vqshlu_n", "udi", "csilQcQsQiQl">;
612def VSHRN_N    : IInst<"vshrn_n", "hki", "silUsUiUl">;
613def VQSHRUN_N  : SInst<"vqshrun_n", "eki", "sil">;
614def VQRSHRUN_N : SInst<"vqrshrun_n", "eki", "sil">;
615def VQSHRN_N   : SInst<"vqshrn_n", "hki", "silUsUiUl">;
616def VRSHRN_N   : IInst<"vrshrn_n", "hki", "silUsUiUl">;
617def VQRSHRN_N  : SInst<"vqrshrn_n", "hki", "silUsUiUl">;
618def VSHLL_N    : SInst<"vshll_n", "wdi", "csiUcUsUi">;
619
620////////////////////////////////////////////////////////////////////////////////
621// E.3.13 Shifts with insert
622def VSRI_N : WInst<"vsri_n", "dddi",
623                   "csilUcUsUiUlPcPsQcQsQiQlQUcQUsQUiQUlQPcQPs">;
624def VSLI_N : WInst<"vsli_n", "dddi",
625                   "csilUcUsUiUlPcPsQcQsQiQlQUcQUsQUiQUlQPcQPs">;
626}
627
628////////////////////////////////////////////////////////////////////////////////
629// E.3.14 Loads and stores of a single vector
630def VLD1      : WInst<"vld1", "dc",
631                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
632def VLD1_LANE : WInst<"vld1_lane", "dcdi",
633                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
634def VLD1_DUP  : WInst<"vld1_dup", "dc",
635                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
636def VST1      : WInst<"vst1", "vpd",
637                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
638def VST1_LANE : WInst<"vst1_lane", "vpdi",
639                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
640
641////////////////////////////////////////////////////////////////////////////////
642// E.3.15 Loads and stores of an N-element structure
643def VLD2 : WInst<"vld2", "2c", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
644def VLD3 : WInst<"vld3", "3c", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
645def VLD4 : WInst<"vld4", "4c", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
646def VLD2_DUP  : WInst<"vld2_dup", "2c", "UcUsUiUlcsilhfPcPs">;
647def VLD3_DUP  : WInst<"vld3_dup", "3c", "UcUsUiUlcsilhfPcPs">;
648def VLD4_DUP  : WInst<"vld4_dup", "4c", "UcUsUiUlcsilhfPcPs">;
649def VLD2_LANE : WInst<"vld2_lane", "2c2i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
650def VLD3_LANE : WInst<"vld3_lane", "3c3i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
651def VLD4_LANE : WInst<"vld4_lane", "4c4i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
652def VST2 : WInst<"vst2", "vp2", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
653def VST3 : WInst<"vst3", "vp3", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
654def VST4 : WInst<"vst4", "vp4", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
655def VST2_LANE : WInst<"vst2_lane", "vp2i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
656def VST3_LANE : WInst<"vst3_lane", "vp3i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
657def VST4_LANE : WInst<"vst4_lane", "vp4i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
658
659////////////////////////////////////////////////////////////////////////////////
660// E.3.16 Extract lanes from a vector
661let InstName = "vmov" in
662def VGET_LANE : IInst<"vget_lane", "sdi",
663                      "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl">;
664
665////////////////////////////////////////////////////////////////////////////////
666// E.3.17 Set lanes within a vector
667let InstName = "vmov" in
668def VSET_LANE : IInst<"vset_lane", "dsdi",
669                      "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl">;
670
671////////////////////////////////////////////////////////////////////////////////
672// E.3.18 Initialize a vector from bit pattern
673def VCREATE : NoTestOpInst<"vcreate", "dl", "csihfUcUsUiUlPcPsl", OP_CAST> {
674  let BigEndianSafe = 1;
675}
676
677////////////////////////////////////////////////////////////////////////////////
678// E.3.19 Set all lanes to same value
679let InstName = "vmov" in {
680def VDUP_N   : WOpInst<"vdup_n", "ds",
681                       "UcUsUicsiPcPshfQUcQUsQUiQcQsQiQPcQPsQhQflUlQlQUl",
682                       OP_DUP>;
683def VMOV_N   : WOpInst<"vmov_n", "ds",
684                       "UcUsUicsiPcPshfQUcQUsQUiQcQsQiQPcQPsQhQflUlQlQUl",
685                       OP_DUP>;
686}
687let InstName = "" in
688def VDUP_LANE: WOpInst<"vdup_lane", "dgi",
689                       "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl",
690                       OP_DUP_LN>;
691
692////////////////////////////////////////////////////////////////////////////////
693// E.3.20 Combining vectors
694def VCOMBINE : NoTestOpInst<"vcombine", "kdd", "csilhfUcUsUiUlPcPs", OP_CONC>;
695
696////////////////////////////////////////////////////////////////////////////////
697// E.3.21 Splitting vectors
698let InstName = "vmov" in {
699def VGET_HIGH : NoTestOpInst<"vget_high", "dk", "csilhfUcUsUiUlPcPs", OP_HI>;
700def VGET_LOW  : NoTestOpInst<"vget_low", "dk", "csilhfUcUsUiUlPcPs", OP_LO>;
701}
702
703////////////////////////////////////////////////////////////////////////////////
704// E.3.22 Converting vectors
705
706def VCVT_F16_F32 : SInst<"vcvt_f16_f32", "md", "Hf">;
707def VCVT_F32_F16 : SInst<"vcvt_f32_f16", "wd", "h">;
708
709def VCVT_S32     : SInst<"vcvt_s32", "xd",  "fQf">;
710def VCVT_U32     : SInst<"vcvt_u32", "ud",  "fQf">;
711def VCVT_F32     : SInst<"vcvt_f32", "fd",  "iUiQiQUi">;
712let isVCVT_N = 1 in {
713def VCVT_N_S32   : SInst<"vcvt_n_s32", "xdi", "fQf">;
714def VCVT_N_U32   : SInst<"vcvt_n_u32", "udi", "fQf">;
715def VCVT_N_F32   : SInst<"vcvt_n_f32", "fdi", "iUiQiQUi">;
716}
717
718def VMOVN        : IInst<"vmovn", "hk",  "silUsUiUl">;
719def VMOVL        : SInst<"vmovl", "wd",  "csiUcUsUi">;
720def VQMOVN       : SInst<"vqmovn", "hk",  "silUsUiUl">;
721def VQMOVUN      : SInst<"vqmovun", "ek",  "sil">;
722
723////////////////////////////////////////////////////////////////////////////////
724// E.3.23-24 Table lookup, Extended table lookup
725let InstName = "vtbl" in {
726def VTBL1 : WInst<"vtbl1", "ddt",  "UccPc">;
727def VTBL2 : WInst<"vtbl2", "d2t",  "UccPc">;
728def VTBL3 : WInst<"vtbl3", "d3t",  "UccPc">;
729def VTBL4 : WInst<"vtbl4", "d4t",  "UccPc">;
730}
731let InstName = "vtbx" in {
732def VTBX1 : WInst<"vtbx1", "dddt", "UccPc">;
733def VTBX2 : WInst<"vtbx2", "dd2t", "UccPc">;
734def VTBX3 : WInst<"vtbx3", "dd3t", "UccPc">;
735def VTBX4 : WInst<"vtbx4", "dd4t", "UccPc">;
736}
737
738////////////////////////////////////////////////////////////////////////////////
739// E.3.25 Operations with a scalar value
740def VMLA_LANE     : IOpInst<"vmla_lane", "dddgi",
741                            "siUsUifQsQiQUsQUiQf", OP_MLA_LN>;
742def VMLAL_LANE    : SOpInst<"vmlal_lane", "wwddi", "siUsUi", OP_MLAL_LN>;
743def VQDMLAL_LANE  : SOpInst<"vqdmlal_lane", "wwddi", "si", OP_QDMLAL_LN>;
744def VMLS_LANE     : IOpInst<"vmls_lane", "dddgi",
745                            "siUsUifQsQiQUsQUiQf", OP_MLS_LN>;
746def VMLSL_LANE    : SOpInst<"vmlsl_lane", "wwddi", "siUsUi", OP_MLSL_LN>;
747def VQDMLSL_LANE  : SOpInst<"vqdmlsl_lane", "wwddi", "si", OP_QDMLSL_LN>;
748def VMUL_N        : IOpInst<"vmul_n", "dds", "sifUsUiQsQiQfQUsQUi", OP_MUL_N>;
749def VMUL_LANE     : IOpInst<"vmul_lane", "ddgi",
750                            "sifUsUiQsQiQfQUsQUi", OP_MUL_LN>;
751def VMULL_N       : SInst<"vmull_n", "wda", "siUsUi">;
752def VMULL_LANE    : SOpInst<"vmull_lane", "wddi", "siUsUi", OP_MULL_LN>;
753def VQDMULL_N     : SInst<"vqdmull_n", "wda", "si">;
754def VQDMULL_LANE  : SOpInst<"vqdmull_lane", "wddi", "si", OP_QDMULL_LN>;
755def VQDMULH_N     : SInst<"vqdmulh_n", "dda", "siQsQi">;
756def VQDMULH_LANE  : SOpInst<"vqdmulh_lane", "ddgi", "siQsQi", OP_QDMULH_LN>;
757def VQRDMULH_N    : SInst<"vqrdmulh_n", "dda", "siQsQi">;
758def VQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "ddgi", "siQsQi", OP_QRDMULH_LN>;
759
760let ArchGuard = "defined(__ARM_FEATURE_QRDMX)" in {
761def VQRDMLAH_LANE : SOpInst<"vqrdmlah_lane", "dddgi", "siQsQi", OP_QRDMLAH_LN>;
762def VQRDMLSH_LANE : SOpInst<"vqrdmlsh_lane", "dddgi", "siQsQi", OP_QRDMLSH_LN>;
763}
764
765def VMLA_N        : IOpInst<"vmla_n", "ddda", "siUsUifQsQiQUsQUiQf", OP_MLA_N>;
766def VMLAL_N       : SOpInst<"vmlal_n", "wwda", "siUsUi", OP_MLAL_N>;
767def VQDMLAL_N     : SInst<"vqdmlal_n", "wwda", "si">;
768def VMLS_N        : IOpInst<"vmls_n", "ddds", "siUsUifQsQiQUsQUiQf", OP_MLS_N>;
769def VMLSL_N       : SOpInst<"vmlsl_n", "wwda", "siUsUi", OP_MLSL_N>;
770def VQDMLSL_N     : SInst<"vqdmlsl_n", "wwda", "si">;
771
772////////////////////////////////////////////////////////////////////////////////
773// E.3.26 Vector Extract
774def VEXT : WInst<"vext", "dddi",
775                 "cUcPcsUsPsiUilUlfQcQUcQPcQsQUsQPsQiQUiQlQUlQf">;
776
777////////////////////////////////////////////////////////////////////////////////
778// E.3.27 Reverse vector elements
779def VREV64 : WOpInst<"vrev64", "dd", "csiUcUsUiPcPsfQcQsQiQUcQUsQUiQPcQPsQf",
780                  OP_REV64>;
781def VREV32 : WOpInst<"vrev32", "dd", "csUcUsPcPsQcQsQUcQUsQPcQPs", OP_REV32>;
782def VREV16 : WOpInst<"vrev16", "dd", "cUcPcQcQUcQPc", OP_REV16>;
783
784////////////////////////////////////////////////////////////////////////////////
785// E.3.28 Other single operand arithmetic
786def VABS    : SInst<"vabs", "dd", "csifQcQsQiQf">;
787def VQABS   : SInst<"vqabs", "dd", "csiQcQsQi">;
788def VNEG    : SOpInst<"vneg", "dd", "csifQcQsQiQf", OP_NEG>;
789def VQNEG   : SInst<"vqneg", "dd", "csiQcQsQi">;
790def VCLS    : SInst<"vcls", "dd", "csiQcQsQi">;
791def VCLZ    : IInst<"vclz", "dd", "csiUcUsUiQcQsQiQUcQUsQUi">;
792def VCNT    : WInst<"vcnt", "dd", "UccPcQUcQcQPc">;
793def VRECPE  : SInst<"vrecpe", "dd", "fUiQfQUi">;
794def VRSQRTE : SInst<"vrsqrte", "dd", "fUiQfQUi">;
795
796////////////////////////////////////////////////////////////////////////////////
797// E.3.29 Logical operations
798def VMVN : LOpInst<"vmvn", "dd", "csiUcUsUiPcQcQsQiQUcQUsQUiQPc", OP_NOT>;
799def VAND : LOpInst<"vand", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_AND>;
800def VORR : LOpInst<"vorr", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_OR>;
801def VEOR : LOpInst<"veor", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_XOR>;
802def VBIC : LOpInst<"vbic", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_ANDN>;
803def VORN : LOpInst<"vorn", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_ORN>;
804let isHiddenLInst = 1 in
805def VBSL : SInst<"vbsl", "dudd",
806                "csilUcUsUiUlfPcPsQcQsQiQlQUcQUsQUiQUlQfQPcQPs">;
807
808////////////////////////////////////////////////////////////////////////////////
809// E.3.30 Transposition operations
810def VTRN : WInst<"vtrn", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
811def VZIP : WInst<"vzip", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
812def VUZP : WInst<"vuzp", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
813
814////////////////////////////////////////////////////////////////////////////////
815// E.3.31 Vector reinterpret cast operations
816def VREINTERPRET
817  : NoTestOpInst<"vreinterpret", "dd",
818         "csilUcUsUiUlhfPcPsQcQsQiQlQUcQUsQUiQUlQhQfQPcQPs", OP_REINT> {
819  let CartesianProductOfTypes = 1;
820  let ArchGuard = "!defined(__aarch64__)";
821  let BigEndianSafe = 1;
822}
823
824////////////////////////////////////////////////////////////////////////////////
825// Vector fused multiply-add operations
826
827def VFMA : SInst<"vfma", "dddd", "fQf">;
828
829////////////////////////////////////////////////////////////////////////////////
830// fp16 vector operations
831def SCALAR_HALF_GET_LANE : IOpInst<"vget_lane", "sdi", "h", OP_SCALAR_HALF_GET_LN>;
832def SCALAR_HALF_SET_LANE : IOpInst<"vset_lane", "dsdi", "h", OP_SCALAR_HALF_SET_LN>;
833def SCALAR_HALF_GET_LANEQ : IOpInst<"vget_lane", "sdi", "Qh", OP_SCALAR_HALF_GET_LNQ>;
834def SCALAR_HALF_SET_LANEQ : IOpInst<"vset_lane", "dsdi", "Qh", OP_SCALAR_HALF_SET_LNQ>;
835
836////////////////////////////////////////////////////////////////////////////////
837// AArch64 Intrinsics
838
839let ArchGuard = "defined(__aarch64__)" in {
840
841////////////////////////////////////////////////////////////////////////////////
842// Load/Store
843def LD1 : WInst<"vld1", "dc", "dQdPlQPl">;
844def LD2 : WInst<"vld2", "2c", "QUlQldQdPlQPl">;
845def LD3 : WInst<"vld3", "3c", "QUlQldQdPlQPl">;
846def LD4 : WInst<"vld4", "4c", "QUlQldQdPlQPl">;
847def ST1 : WInst<"vst1", "vpd", "dQdPlQPl">;
848def ST2 : WInst<"vst2", "vp2", "QUlQldQdPlQPl">;
849def ST3 : WInst<"vst3", "vp3", "QUlQldQdPlQPl">;
850def ST4 : WInst<"vst4", "vp4", "QUlQldQdPlQPl">;
851
852def LD1_X2 : WInst<"vld1_x2", "2c",
853                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
854def LD3_x3 : WInst<"vld1_x3", "3c",
855                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
856def LD4_x4 : WInst<"vld1_x4", "4c",
857                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
858
859def ST1_X2 : WInst<"vst1_x2", "vp2",
860                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
861def ST1_X3 : WInst<"vst1_x3", "vp3",
862                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
863def ST1_X4 : WInst<"vst1_x4", "vp4",
864                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
865
866def LD1_LANE : WInst<"vld1_lane", "dcdi", "dQdPlQPl">;
867def LD2_LANE : WInst<"vld2_lane", "2c2i", "lUlQcQUcQPcQlQUldQdPlQPl">;
868def LD3_LANE : WInst<"vld3_lane", "3c3i", "lUlQcQUcQPcQlQUldQdPlQPl">;
869def LD4_LANE : WInst<"vld4_lane", "4c4i", "lUlQcQUcQPcQlQUldQdPlQPl">;
870def ST1_LANE : WInst<"vst1_lane", "vpdi", "dQdPlQPl">;
871def ST2_LANE : WInst<"vst2_lane", "vp2i", "lUlQcQUcQPcQlQUldQdPlQPl">;
872def ST3_LANE : WInst<"vst3_lane", "vp3i", "lUlQcQUcQPcQlQUldQdPlQPl">;
873def ST4_LANE : WInst<"vst4_lane", "vp4i", "lUlQcQUcQPcQlQUldQdPlQPl">;
874
875def LD1_DUP  : WInst<"vld1_dup", "dc", "dQdPlQPl">;
876def LD2_DUP  : WInst<"vld2_dup", "2c",
877                     "QUcQUsQUiQUlQcQsQiQlQhQfQdQPcQPsQPldPl">;
878def LD3_DUP  : WInst<"vld3_dup", "3c",
879                     "QUcQUsQUiQUlQcQsQiQlQhQfQdQPcQPsQPldPl">;
880def LD4_DUP  : WInst<"vld4_dup", "4c",
881                     "QUcQUsQUiQUlQcQsQiQlQhQfQdQPcQPsQPldPl">;
882
883def VLDRQ : WInst<"vldrq", "sc", "Pk">;
884def VSTRQ : WInst<"vstrq", "vps", "Pk">;
885
886////////////////////////////////////////////////////////////////////////////////
887// Addition
888def ADD : IOpInst<"vadd", "ddd", "dQd", OP_ADD>;
889
890////////////////////////////////////////////////////////////////////////////////
891// Subtraction
892def SUB : IOpInst<"vsub", "ddd", "dQd", OP_SUB>;
893
894////////////////////////////////////////////////////////////////////////////////
895// Multiplication
896def MUL     : IOpInst<"vmul", "ddd", "dQd", OP_MUL>;
897def MLA     : IOpInst<"vmla", "dddd", "dQd", OP_MLA>;
898def MLS     : IOpInst<"vmls", "dddd", "dQd", OP_MLS>;
899
900////////////////////////////////////////////////////////////////////////////////
901// Multiplication Extended
902def MULX : SInst<"vmulx", "ddd", "fdQfQd">;
903
904////////////////////////////////////////////////////////////////////////////////
905// Division
906def FDIV : IOpInst<"vdiv", "ddd",  "fdQfQd", OP_DIV>;
907
908////////////////////////////////////////////////////////////////////////////////
909// Vector fused multiply-add operations
910def FMLA : SInst<"vfma", "dddd", "dQd">;
911def FMLS : SInst<"vfms", "dddd", "fdQfQd">;
912
913////////////////////////////////////////////////////////////////////////////////
914// MUL, MLA, MLS, FMA, FMS definitions with scalar argument
915def VMUL_N_A64 : IOpInst<"vmul_n", "dds", "Qd", OP_MUL_N>;
916
917def FMLA_N : SOpInst<"vfma_n", "ddds", "fQfQd", OP_FMLA_N>;
918def FMLS_N : SOpInst<"vfms_n", "ddds", "fQfQd", OP_FMLS_N>;
919
920def MLA_N : SOpInst<"vmla_n", "ddds", "Qd", OP_MLA_N>;
921def MLS_N : SOpInst<"vmls_n", "ddds", "Qd", OP_MLS_N>;
922
923////////////////////////////////////////////////////////////////////////////////
924// Logical operations
925def BSL : SInst<"vbsl", "dudd", "dPlQdQPl">;
926
927////////////////////////////////////////////////////////////////////////////////
928// Absolute Difference
929def ABD  : SInst<"vabd", "ddd",  "dQd">;
930
931////////////////////////////////////////////////////////////////////////////////
932// saturating absolute/negate
933def ABS    : SInst<"vabs", "dd", "dQdlQl">;
934def QABS   : SInst<"vqabs", "dd", "lQl">;
935def NEG    : SOpInst<"vneg", "dd", "dlQdQl", OP_NEG>;
936def QNEG   : SInst<"vqneg", "dd", "lQl">;
937
938////////////////////////////////////////////////////////////////////////////////
939// Signed Saturating Accumulated of Unsigned Value
940def SUQADD : SInst<"vuqadd", "ddd", "csilQcQsQiQl">;
941
942////////////////////////////////////////////////////////////////////////////////
943// Unsigned Saturating Accumulated of Signed Value
944def USQADD : SInst<"vsqadd", "ddd", "UcUsUiUlQUcQUsQUiQUl">;
945
946////////////////////////////////////////////////////////////////////////////////
947// Reciprocal/Sqrt
948def FRECPS  : IInst<"vrecps", "ddd", "dQd">;
949def FRSQRTS : IInst<"vrsqrts", "ddd", "dQd">;
950def FRECPE  : SInst<"vrecpe", "dd", "dQd">;
951def FRSQRTE : SInst<"vrsqrte", "dd", "dQd">;
952def FSQRT   : SInst<"vsqrt", "dd", "fdQfQd">;
953
954////////////////////////////////////////////////////////////////////////////////
955// bitwise reverse
956def RBIT : IInst<"vrbit", "dd", "cUcPcQcQUcQPc">;
957
958////////////////////////////////////////////////////////////////////////////////
959// Integer extract and narrow to high
960def XTN2 : SOpInst<"vmovn_high", "qhk", "silUsUiUl", OP_XTN>;
961
962////////////////////////////////////////////////////////////////////////////////
963// Signed integer saturating extract and unsigned narrow to high
964def SQXTUN2 : SOpInst<"vqmovun_high", "qhk", "sil", OP_SQXTUN>;
965
966////////////////////////////////////////////////////////////////////////////////
967// Integer saturating extract and narrow to high
968def QXTN2 : SOpInst<"vqmovn_high", "qhk", "silUsUiUl", OP_QXTN>;
969
970////////////////////////////////////////////////////////////////////////////////
971// Converting vectors
972
973def VCVT_F32_F64 : SInst<"vcvt_f32_f64", "md", "Qd">;
974def VCVT_F64_F32 : SInst<"vcvt_f64_f32", "wd", "f">;
975
976def VCVT_S64 : SInst<"vcvt_s64", "xd",  "dQd">;
977def VCVT_U64 : SInst<"vcvt_u64", "ud",  "dQd">;
978def VCVT_F64 : SInst<"vcvt_f64", "Fd",  "lUlQlQUl">;
979
980def VCVT_HIGH_F16_F32 : SOpInst<"vcvt_high_f16", "hmj", "Hf", OP_VCVT_NA_HI_F16>;
981def VCVT_HIGH_F32_F16 : SOpInst<"vcvt_high_f32", "wk", "h", OP_VCVT_EX_HI_F32>;
982def VCVT_HIGH_F32_F64 : SOpInst<"vcvt_high_f32", "qfj", "d", OP_VCVT_NA_HI_F32>;
983def VCVT_HIGH_F64_F32 : SOpInst<"vcvt_high_f64", "wj", "f", OP_VCVT_EX_HI_F64>;
984
985def VCVTX_F32_F64      : SInst<"vcvtx_f32", "fj",  "d">;
986def VCVTX_HIGH_F32_F64 : SOpInst<"vcvtx_high_f32", "qfj", "d", OP_VCVTX_HI>;
987
988////////////////////////////////////////////////////////////////////////////////
989// Comparison
990def FCAGE : IInst<"vcage", "udd", "dQd">;
991def FCAGT : IInst<"vcagt", "udd", "dQd">;
992def FCALE : IInst<"vcale", "udd", "dQd">;
993def FCALT : IInst<"vcalt", "udd", "dQd">;
994def CMTST  : WInst<"vtst", "udd", "lUlPlQlQUlQPl">;
995def CFMEQ  : SOpInst<"vceq", "udd", "lUldQdQlQUlPlQPl", OP_EQ>;
996def CFMGE  : SOpInst<"vcge", "udd", "lUldQdQlQUl", OP_GE>;
997def CFMLE  : SOpInst<"vcle", "udd", "lUldQdQlQUl", OP_LE>;
998def CFMGT  : SOpInst<"vcgt", "udd", "lUldQdQlQUl", OP_GT>;
999def CFMLT  : SOpInst<"vclt", "udd", "lUldQdQlQUl", OP_LT>;
1000
1001def CMEQ  : SInst<"vceqz", "ud",
1002                  "csilfUcUsUiUlPcPsPlQcQsQiQlQfQUcQUsQUiQUlQPcQPsdQdQPl">;
1003def CMGE  : SInst<"vcgez", "ud", "csilfdQcQsQiQlQfQd">;
1004def CMLE  : SInst<"vclez", "ud", "csilfdQcQsQiQlQfQd">;
1005def CMGT  : SInst<"vcgtz", "ud", "csilfdQcQsQiQlQfQd">;
1006def CMLT  : SInst<"vcltz", "ud", "csilfdQcQsQiQlQfQd">;
1007
1008////////////////////////////////////////////////////////////////////////////////
1009// Max/Min Integer
1010def MAX : SInst<"vmax", "ddd", "dQd">;
1011def MIN : SInst<"vmin", "ddd", "dQd">;
1012
1013////////////////////////////////////////////////////////////////////////////////
1014// Pairwise Max/Min
1015def MAXP : SInst<"vpmax", "ddd", "QcQsQiQUcQUsQUiQfQd">;
1016def MINP : SInst<"vpmin", "ddd", "QcQsQiQUcQUsQUiQfQd">;
1017
1018////////////////////////////////////////////////////////////////////////////////
1019// Pairwise MaxNum/MinNum Floating Point
1020def FMAXNMP : SInst<"vpmaxnm", "ddd", "fQfQd">;
1021def FMINNMP : SInst<"vpminnm", "ddd", "fQfQd">;
1022
1023////////////////////////////////////////////////////////////////////////////////
1024// Pairwise Addition
1025def ADDP  : IInst<"vpadd", "ddd", "QcQsQiQlQUcQUsQUiQUlQfQd">;
1026
1027////////////////////////////////////////////////////////////////////////////////
1028// Shifts by constant
1029let isShift = 1 in {
1030// Left shift long high
1031def SHLL_HIGH_N    : SOpInst<"vshll_high_n", "ndi", "HcHsHiHUcHUsHUi",
1032                             OP_LONG_HI>;
1033
1034////////////////////////////////////////////////////////////////////////////////
1035def SRI_N : WInst<"vsri_n", "dddi", "PlQPl">;
1036def SLI_N : WInst<"vsli_n", "dddi", "PlQPl">;
1037
1038// Right shift narrow high
1039def SHRN_HIGH_N    : IOpInst<"vshrn_high_n", "hmdi",
1040                             "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
1041def QSHRUN_HIGH_N  : SOpInst<"vqshrun_high_n", "hmdi",
1042                             "HsHiHl", OP_NARROW_HI>;
1043def RSHRN_HIGH_N   : IOpInst<"vrshrn_high_n", "hmdi",
1044                             "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
1045def QRSHRUN_HIGH_N : SOpInst<"vqrshrun_high_n", "hmdi",
1046                             "HsHiHl", OP_NARROW_HI>;
1047def QSHRN_HIGH_N   : SOpInst<"vqshrn_high_n", "hmdi", 
1048                             "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
1049def QRSHRN_HIGH_N  : SOpInst<"vqrshrn_high_n", "hmdi", 
1050                             "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
1051}
1052
1053////////////////////////////////////////////////////////////////////////////////
1054// Converting vectors
1055def VMOVL_HIGH   : SOpInst<"vmovl_high", "nd", "HcHsHiHUcHUsHUi", OP_MOVL_HI>;
1056
1057let isVCVT_N = 1 in {
1058def CVTF_N_F64   : SInst<"vcvt_n_f64", "Fdi", "lUlQlQUl">;
1059def FCVTZS_N_S64 : SInst<"vcvt_n_s64", "xdi", "dQd">;
1060def FCVTZS_N_U64 : SInst<"vcvt_n_u64", "udi", "dQd">;
1061}
1062
1063////////////////////////////////////////////////////////////////////////////////
1064// 3VDiff class using high 64-bit in operands
1065def VADDL_HIGH   : SOpInst<"vaddl_high", "wkk", "csiUcUsUi", OP_ADDLHi>;
1066def VADDW_HIGH   : SOpInst<"vaddw_high", "wwk", "csiUcUsUi", OP_ADDWHi>;
1067def VSUBL_HIGH   : SOpInst<"vsubl_high", "wkk", "csiUcUsUi", OP_SUBLHi>;
1068def VSUBW_HIGH   : SOpInst<"vsubw_high", "wwk", "csiUcUsUi", OP_SUBWHi>;
1069
1070def VABDL_HIGH   : SOpInst<"vabdl_high", "wkk",  "csiUcUsUi", OP_ABDLHi>;
1071def VABAL_HIGH   : SOpInst<"vabal_high", "wwkk", "csiUcUsUi", OP_ABALHi>;
1072
1073def VMULL_HIGH   : SOpInst<"vmull_high", "wkk", "csiUcUsUiPc", OP_MULLHi>;
1074def VMULL_HIGH_N : SOpInst<"vmull_high_n", "wks", "siUsUi", OP_MULLHi_N>;
1075def VMLAL_HIGH   : SOpInst<"vmlal_high", "wwkk", "csiUcUsUi", OP_MLALHi>;
1076def VMLAL_HIGH_N : SOpInst<"vmlal_high_n", "wwks", "siUsUi", OP_MLALHi_N>;
1077def VMLSL_HIGH   : SOpInst<"vmlsl_high", "wwkk", "csiUcUsUi", OP_MLSLHi>;
1078def VMLSL_HIGH_N : SOpInst<"vmlsl_high_n", "wwks", "siUsUi", OP_MLSLHi_N>;
1079
1080def VADDHN_HIGH  : SOpInst<"vaddhn_high", "qhkk", "silUsUiUl", OP_ADDHNHi>;
1081def VRADDHN_HIGH : SOpInst<"vraddhn_high", "qhkk", "silUsUiUl", OP_RADDHNHi>;
1082def VSUBHN_HIGH  : SOpInst<"vsubhn_high", "qhkk", "silUsUiUl", OP_SUBHNHi>;
1083def VRSUBHN_HIGH : SOpInst<"vrsubhn_high", "qhkk", "silUsUiUl", OP_RSUBHNHi>;
1084
1085def VQDMULL_HIGH : SOpInst<"vqdmull_high", "wkk", "si", OP_QDMULLHi>;
1086def VQDMULL_HIGH_N : SOpInst<"vqdmull_high_n", "wks", "si", OP_QDMULLHi_N>;
1087def VQDMLAL_HIGH : SOpInst<"vqdmlal_high", "wwkk", "si", OP_QDMLALHi>;
1088def VQDMLAL_HIGH_N : SOpInst<"vqdmlal_high_n", "wwks", "si", OP_QDMLALHi_N>;
1089def VQDMLSL_HIGH : SOpInst<"vqdmlsl_high", "wwkk", "si", OP_QDMLSLHi>;
1090def VQDMLSL_HIGH_N : SOpInst<"vqdmlsl_high_n", "wwks", "si", OP_QDMLSLHi_N>;
1091def VMULL_P64    : SInst<"vmull", "rss", "Pl">;
1092def VMULL_HIGH_P64 : SOpInst<"vmull_high", "rdd", "HPl", OP_MULLHi_P64>;
1093
1094
1095////////////////////////////////////////////////////////////////////////////////
1096// Extract or insert element from vector
1097def GET_LANE : IInst<"vget_lane", "sdi", "dQdPlQPl">;
1098def SET_LANE : IInst<"vset_lane", "dsdi", "dQdPlQPl">;
1099def COPY_LANE : IOpInst<"vcopy_lane", "ddidi",
1100                        "csilUcUsUiUlPcPsPlfd", OP_COPY_LN>;
1101def COPYQ_LANE : IOpInst<"vcopy_lane", "ddigi",
1102                        "QcQsQiQlQUcQUsQUiQUlQPcQPsQfQdQPl", OP_COPY_LN>;
1103def COPY_LANEQ : IOpInst<"vcopy_laneq", "ddiki",
1104                     "csilPcPsPlUcUsUiUlfd", OP_COPY_LN>;
1105def COPYQ_LANEQ : IOpInst<"vcopy_laneq", "ddidi",
1106                     "QcQsQiQlQUcQUsQUiQUlQPcQPsQfQdQPl", OP_COPY_LN>;
1107
1108////////////////////////////////////////////////////////////////////////////////
1109// Set all lanes to same value
1110def VDUP_LANE1: WOpInst<"vdup_lane", "dgi", "hdQhQdPlQPl", OP_DUP_LN>;
1111def VDUP_LANE2: WOpInst<"vdup_laneq", "dji",
1112                  "csilUcUsUiUlPcPshfdQcQsQiQlQPcQPsQUcQUsQUiQUlQhQfQdPlQPl",
1113                        OP_DUP_LN>;
1114def DUP_N   : WOpInst<"vdup_n", "ds", "dQdPlQPl", OP_DUP>;
1115def MOV_N   : WOpInst<"vmov_n", "ds", "dQdPlQPl", OP_DUP>;
1116
1117////////////////////////////////////////////////////////////////////////////////
1118def COMBINE : NoTestOpInst<"vcombine", "kdd", "dPl", OP_CONC>;
1119
1120////////////////////////////////////////////////////////////////////////////////
1121//Initialize a vector from bit pattern
1122def CREATE : NoTestOpInst<"vcreate", "dl", "dPl", OP_CAST> {
1123  let BigEndianSafe = 1;
1124}
1125
1126////////////////////////////////////////////////////////////////////////////////
1127
1128def VMLA_LANEQ   : IOpInst<"vmla_laneq", "dddji",
1129                           "siUsUifQsQiQUsQUiQf", OP_MLA_LN>;
1130def VMLS_LANEQ   : IOpInst<"vmls_laneq", "dddji",
1131                           "siUsUifQsQiQUsQUiQf", OP_MLS_LN>;
1132
1133def VFMA_LANE    : IInst<"vfma_lane", "dddgi", "fdQfQd">;
1134def VFMA_LANEQ   : IInst<"vfma_laneq", "dddji", "fdQfQd"> {
1135  let isLaneQ = 1;
1136}
1137def VFMS_LANE    : IOpInst<"vfms_lane", "dddgi", "fdQfQd", OP_FMS_LN>;
1138def VFMS_LANEQ   : IOpInst<"vfms_laneq", "dddji", "fdQfQd", OP_FMS_LNQ>;
1139
1140def VMLAL_LANEQ  : SOpInst<"vmlal_laneq", "wwdki", "siUsUi", OP_MLAL_LN>;
1141def VMLAL_HIGH_LANE   : SOpInst<"vmlal_high_lane", "wwkdi", "siUsUi",
1142                                OP_MLALHi_LN>;
1143def VMLAL_HIGH_LANEQ  : SOpInst<"vmlal_high_laneq", "wwkki", "siUsUi",
1144                                OP_MLALHi_LN>;
1145def VMLSL_LANEQ  : SOpInst<"vmlsl_laneq", "wwdki", "siUsUi", OP_MLSL_LN>;
1146def VMLSL_HIGH_LANE   : SOpInst<"vmlsl_high_lane", "wwkdi", "siUsUi",
1147                                OP_MLSLHi_LN>;
1148def VMLSL_HIGH_LANEQ  : SOpInst<"vmlsl_high_laneq", "wwkki", "siUsUi",
1149                                OP_MLSLHi_LN>;
1150
1151def VQDMLAL_LANEQ  : SOpInst<"vqdmlal_laneq", "wwdki", "si", OP_QDMLAL_LN>;
1152def VQDMLAL_HIGH_LANE   : SOpInst<"vqdmlal_high_lane", "wwkdi", "si",
1153                                OP_QDMLALHi_LN>;
1154def VQDMLAL_HIGH_LANEQ  : SOpInst<"vqdmlal_high_laneq", "wwkki", "si",
1155                                OP_QDMLALHi_LN>;
1156def VQDMLSL_LANEQ  : SOpInst<"vqdmlsl_laneq", "wwdki", "si", OP_QDMLSL_LN>;
1157def VQDMLSL_HIGH_LANE   : SOpInst<"vqdmlsl_high_lane", "wwkdi", "si",
1158                                OP_QDMLSLHi_LN>;
1159def VQDMLSL_HIGH_LANEQ  : SOpInst<"vqdmlsl_high_laneq", "wwkki", "si",
1160                                OP_QDMLSLHi_LN>;
1161
1162// Newly add double parameter for vmul_lane in aarch64
1163// Note: d type is handled by SCALAR_VMUL_LANE
1164def VMUL_LANE_A64 : IOpInst<"vmul_lane", "ddgi", "Qd", OP_MUL_LN>;
1165
1166// Note: d type is handled by SCALAR_VMUL_LANEQ
1167def VMUL_LANEQ   : IOpInst<"vmul_laneq", "ddji",
1168                           "sifUsUiQsQiQUsQUiQfQd", OP_MUL_LN>;
1169def VMULL_LANEQ  : SOpInst<"vmull_laneq", "wdki", "siUsUi", OP_MULL_LN>;
1170def VMULL_HIGH_LANE   : SOpInst<"vmull_high_lane", "wkdi", "siUsUi",
1171                                OP_MULLHi_LN>;
1172def VMULL_HIGH_LANEQ  : SOpInst<"vmull_high_laneq", "wkki", "siUsUi",
1173                                OP_MULLHi_LN>;
1174
1175def VQDMULL_LANEQ  : SOpInst<"vqdmull_laneq", "wdki", "si", OP_QDMULL_LN>;
1176def VQDMULL_HIGH_LANE   : SOpInst<"vqdmull_high_lane", "wkdi", "si",
1177                                  OP_QDMULLHi_LN>;
1178def VQDMULL_HIGH_LANEQ  : SOpInst<"vqdmull_high_laneq", "wkki", "si",
1179                                  OP_QDMULLHi_LN>;
1180
1181def VQDMULH_LANEQ  : SOpInst<"vqdmulh_laneq", "ddji", "siQsQi", OP_QDMULH_LN>;
1182def VQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "ddji", "siQsQi", OP_QRDMULH_LN>;
1183
1184let ArchGuard = "defined(__ARM_FEATURE_QRDMX) && defined(__aarch64__)" in {
1185def VQRDMLAH_LANEQ : SOpInst<"vqrdmlah_laneq", "dddji", "siQsQi", OP_QRDMLAH_LN>;
1186def VQRDMLSH_LANEQ : SOpInst<"vqrdmlsh_laneq", "dddji", "siQsQi", OP_QRDMLSH_LN>;
1187}
1188
1189// Note: d type implemented by SCALAR_VMULX_LANE
1190def VMULX_LANE : IOpInst<"vmulx_lane", "ddgi", "fQfQd", OP_MULX_LN>;
1191// Note: d type is implemented by SCALAR_VMULX_LANEQ
1192def VMULX_LANEQ : IOpInst<"vmulx_laneq", "ddji", "fQfQd", OP_MULX_LN>;
1193
1194////////////////////////////////////////////////////////////////////////////////
1195// Across vectors class
1196def VADDLV  : SInst<"vaddlv", "rd", "csiUcUsUiQcQsQiQUcQUsQUi">;
1197def VMAXV   : SInst<"vmaxv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
1198def VMINV   : SInst<"vminv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
1199def VADDV   : SInst<"vaddv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQdQlQUl">;
1200def FMAXNMV : SInst<"vmaxnmv", "sd", "fQfQd">;
1201def FMINNMV : SInst<"vminnmv", "sd", "fQfQd">;
1202 
1203////////////////////////////////////////////////////////////////////////////////
1204// Newly added Vector Extract for f64
1205def VEXT_A64 : WInst<"vext", "dddi", "dQdPlQPl">;
1206
1207////////////////////////////////////////////////////////////////////////////////
1208// Crypto
1209let ArchGuard = "__ARM_FEATURE_CRYPTO" in {
1210def AESE : SInst<"vaese", "ddd", "QUc">;
1211def AESD : SInst<"vaesd", "ddd", "QUc">;
1212def AESMC : SInst<"vaesmc", "dd", "QUc">;
1213def AESIMC : SInst<"vaesimc", "dd", "QUc">;
1214
1215def SHA1H : SInst<"vsha1h", "ss", "Ui">;
1216def SHA1SU1 : SInst<"vsha1su1", "ddd", "QUi">;
1217def SHA256SU0 : SInst<"vsha256su0", "ddd", "QUi">;
1218
1219def SHA1C : SInst<"vsha1c", "ddsd", "QUi">;
1220def SHA1P : SInst<"vsha1p", "ddsd", "QUi">;
1221def SHA1M : SInst<"vsha1m", "ddsd", "QUi">;
1222def SHA1SU0 : SInst<"vsha1su0", "dddd", "QUi">;
1223def SHA256H : SInst<"vsha256h", "dddd", "QUi">;
1224def SHA256H2 : SInst<"vsha256h2", "dddd", "QUi">;
1225def SHA256SU1 : SInst<"vsha256su1", "dddd", "QUi">;
1226}
1227
1228////////////////////////////////////////////////////////////////////////////////
1229// Float -> Int conversions with explicit rounding mode
1230
1231let ArchGuard = "__ARM_ARCH >= 8" in {
1232def FCVTNS_S32 : SInst<"vcvtn_s32", "xd", "fQf">;
1233def FCVTNU_S32 : SInst<"vcvtn_u32", "ud", "fQf">;
1234def FCVTPS_S32 : SInst<"vcvtp_s32", "xd", "fQf">;
1235def FCVTPU_S32 : SInst<"vcvtp_u32", "ud", "fQf">;
1236def FCVTMS_S32 : SInst<"vcvtm_s32", "xd", "fQf">;
1237def FCVTMU_S32 : SInst<"vcvtm_u32", "ud", "fQf">;
1238def FCVTAS_S32 : SInst<"vcvta_s32", "xd", "fQf">;
1239def FCVTAU_S32 : SInst<"vcvta_u32", "ud", "fQf">;
1240}
1241
1242let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__)" in {
1243def FCVTNS_S64 : SInst<"vcvtn_s64", "xd", "dQd">;
1244def FCVTNU_S64 : SInst<"vcvtn_u64", "ud", "dQd">;
1245def FCVTPS_S64 : SInst<"vcvtp_s64", "xd", "dQd">;
1246def FCVTPU_S64 : SInst<"vcvtp_u64", "ud", "dQd">;
1247def FCVTMS_S64 : SInst<"vcvtm_s64", "xd", "dQd">;
1248def FCVTMU_S64 : SInst<"vcvtm_u64", "ud", "dQd">;
1249def FCVTAS_S64 : SInst<"vcvta_s64", "xd", "dQd">;
1250def FCVTAU_S64 : SInst<"vcvta_u64", "ud", "dQd">;
1251}
1252
1253////////////////////////////////////////////////////////////////////////////////
1254// Round to Integral
1255
1256let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_DIRECTED_ROUNDING)" in {
1257def FRINTN_S32 : SInst<"vrndn", "dd", "fQf">;
1258def FRINTA_S32 : SInst<"vrnda", "dd", "fQf">;
1259def FRINTP_S32 : SInst<"vrndp", "dd", "fQf">;
1260def FRINTM_S32 : SInst<"vrndm", "dd", "fQf">;
1261def FRINTX_S32 : SInst<"vrndx", "dd", "fQf">;
1262def FRINTZ_S32 : SInst<"vrnd", "dd", "fQf">;
1263}
1264
1265let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__) && defined(__ARM_FEATURE_DIRECTED_ROUNDING)" in {
1266def FRINTN_S64 : SInst<"vrndn", "dd", "dQd">;
1267def FRINTA_S64 : SInst<"vrnda", "dd", "dQd">;
1268def FRINTP_S64 : SInst<"vrndp", "dd", "dQd">;
1269def FRINTM_S64 : SInst<"vrndm", "dd", "dQd">;
1270def FRINTX_S64 : SInst<"vrndx", "dd", "dQd">;
1271def FRINTZ_S64 : SInst<"vrnd", "dd", "dQd">;
1272def FRINTI_S64 : SInst<"vrndi", "dd", "fdQfQd">;
1273}
1274
1275////////////////////////////////////////////////////////////////////////////////
1276// MaxNum/MinNum Floating Point
1277
1278let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_NUMERIC_MAXMIN)" in {
1279def FMAXNM_S32 : SInst<"vmaxnm", "ddd", "fQf">;
1280def FMINNM_S32 : SInst<"vminnm", "ddd", "fQf">;
1281}
1282
1283let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__) && defined(__ARM_FEATURE_NUMERIC_MAXMIN)" in {
1284def FMAXNM_S64 : SInst<"vmaxnm", "ddd", "dQd">;
1285def FMINNM_S64 : SInst<"vminnm", "ddd", "dQd">;
1286}
1287
1288////////////////////////////////////////////////////////////////////////////////
1289// Permutation
1290def VTRN1 : SOpInst<"vtrn1", "ddd",
1291                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_TRN1>;
1292def VZIP1 : SOpInst<"vzip1", "ddd",
1293                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_ZIP1>;
1294def VUZP1 : SOpInst<"vuzp1", "ddd",
1295                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_UZP1>;
1296def VTRN2 : SOpInst<"vtrn2", "ddd",
1297                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_TRN2>;
1298def VZIP2 : SOpInst<"vzip2", "ddd",
1299                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_ZIP2>;
1300def VUZP2 : SOpInst<"vuzp2", "ddd",
1301                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_UZP2>;
1302
1303////////////////////////////////////////////////////////////////////////////////
1304// Table lookup
1305let InstName = "vtbl" in {
1306def VQTBL1_A64 : WInst<"vqtbl1", "djt",  "UccPcQUcQcQPc">;
1307def VQTBL2_A64 : WInst<"vqtbl2", "dBt",  "UccPcQUcQcQPc">;
1308def VQTBL3_A64 : WInst<"vqtbl3", "dCt",  "UccPcQUcQcQPc">;
1309def VQTBL4_A64 : WInst<"vqtbl4", "dDt",  "UccPcQUcQcQPc">;
1310}
1311let InstName = "vtbx" in {
1312def VQTBX1_A64 : WInst<"vqtbx1", "ddjt", "UccPcQUcQcQPc">;
1313def VQTBX2_A64 : WInst<"vqtbx2", "ddBt", "UccPcQUcQcQPc">;
1314def VQTBX3_A64 : WInst<"vqtbx3", "ddCt", "UccPcQUcQcQPc">;
1315def VQTBX4_A64 : WInst<"vqtbx4", "ddDt", "UccPcQUcQcQPc">;
1316}
1317
1318////////////////////////////////////////////////////////////////////////////////
1319// Vector reinterpret cast operations
1320
1321// NeonEmitter implicitly takes the cartesian product of the type string with
1322// itself during generation so, unlike all other intrinsics, this one should
1323// include *all* types, not just additional ones.
1324def VVREINTERPRET
1325  : NoTestOpInst<"vreinterpret", "dd",
1326       "csilUcUsUiUlhfdPcPsPlQcQsQiQlQUcQUsQUiQUlQhQfQdQPcQPsQPlQPk", OP_REINT> {
1327  let CartesianProductOfTypes = 1;
1328  let BigEndianSafe = 1;
1329  let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__)";
1330}
1331
1332////////////////////////////////////////////////////////////////////////////////
1333// Scalar Intrinsics
1334// Scalar Arithmetic
1335
1336// Scalar Addition
1337def SCALAR_ADD : SInst<"vadd", "sss",  "SlSUl">;
1338// Scalar  Saturating Add
1339def SCALAR_QADD   : SInst<"vqadd", "sss", "ScSsSiSlSUcSUsSUiSUl">;
1340
1341// Scalar Subtraction
1342def SCALAR_SUB : SInst<"vsub", "sss",  "SlSUl">;
1343// Scalar  Saturating Sub
1344def SCALAR_QSUB   : SInst<"vqsub", "sss", "ScSsSiSlSUcSUsSUiSUl">;
1345
1346let InstName = "vmov" in {
1347def VGET_HIGH_A64 : NoTestOpInst<"vget_high", "dk", "dPl", OP_HI>;
1348def VGET_LOW_A64  : NoTestOpInst<"vget_low", "dk", "dPl", OP_LO>;
1349}
1350
1351////////////////////////////////////////////////////////////////////////////////
1352// Scalar Shift
1353// Scalar Shift Left
1354def SCALAR_SHL: SInst<"vshl", "sss", "SlSUl">;
1355// Scalar Saturating Shift Left
1356def SCALAR_QSHL: SInst<"vqshl", "sss", "ScSsSiSlSUcSUsSUiSUl">;
1357// Scalar Saturating Rounding Shift Left
1358def SCALAR_QRSHL: SInst<"vqrshl", "sss", "ScSsSiSlSUcSUsSUiSUl">;
1359// Scalar Shift Rouding Left
1360def SCALAR_RSHL: SInst<"vrshl", "sss", "SlSUl">;
1361
1362////////////////////////////////////////////////////////////////////////////////
1363// Scalar Shift (Immediate)
1364let isScalarShift = 1 in {
1365// Signed/Unsigned Shift Right (Immediate)
1366def SCALAR_SSHR_N: SInst<"vshr_n", "ssi", "SlSUl">;
1367// Signed/Unsigned Rounding Shift Right (Immediate)
1368def SCALAR_SRSHR_N: SInst<"vrshr_n", "ssi", "SlSUl">;
1369
1370// Signed/Unsigned Shift Right and Accumulate (Immediate)
1371def SCALAR_SSRA_N: SInst<"vsra_n", "sssi", "SlSUl">;
1372// Signed/Unsigned Rounding Shift Right and Accumulate (Immediate)
1373def SCALAR_SRSRA_N: SInst<"vrsra_n", "sssi", "SlSUl">;
1374
1375// Shift Left (Immediate)
1376def SCALAR_SHL_N: SInst<"vshl_n", "ssi", "SlSUl">;
1377// Signed/Unsigned Saturating Shift Left (Immediate)
1378def SCALAR_SQSHL_N: SInst<"vqshl_n", "ssi", "ScSsSiSlSUcSUsSUiSUl">;
1379// Signed Saturating Shift Left Unsigned (Immediate)
1380def SCALAR_SQSHLU_N: SInst<"vqshlu_n", "ssi", "ScSsSiSl">;
1381
1382// Shift Right And Insert (Immediate)
1383def SCALAR_SRI_N: SInst<"vsri_n", "sssi", "SlSUl">;
1384// Shift Left And Insert (Immediate)
1385def SCALAR_SLI_N: SInst<"vsli_n", "sssi", "SlSUl">;
1386
1387let isScalarNarrowShift = 1 in {
1388  // Signed/Unsigned Saturating Shift Right Narrow (Immediate)
1389  def SCALAR_SQSHRN_N: SInst<"vqshrn_n", "zsi", "SsSiSlSUsSUiSUl">;
1390  // Signed/Unsigned Saturating Rounded Shift Right Narrow (Immediate)
1391  def SCALAR_SQRSHRN_N: SInst<"vqrshrn_n", "zsi", "SsSiSlSUsSUiSUl">;
1392  // Signed Saturating Shift Right Unsigned Narrow (Immediate)
1393  def SCALAR_SQSHRUN_N: SInst<"vqshrun_n", "zsi", "SsSiSl">;
1394  // Signed Saturating Rounded Shift Right Unsigned Narrow (Immediate)
1395  def SCALAR_SQRSHRUN_N: SInst<"vqrshrun_n", "zsi", "SsSiSl">;
1396}
1397
1398////////////////////////////////////////////////////////////////////////////////
1399// Scalar Signed/Unsigned Fixed-point Convert To Floating-Point (Immediate)
1400def SCALAR_SCVTF_N_F32: SInst<"vcvt_n_f32", "ysi", "SiSUi">;
1401def SCALAR_SCVTF_N_F64: SInst<"vcvt_n_f64", "osi", "SlSUl">;
1402
1403////////////////////////////////////////////////////////////////////////////////
1404// Scalar Floating-point Convert To Signed/Unsigned Fixed-point (Immediate)
1405def SCALAR_FCVTZS_N_S32 : SInst<"vcvt_n_s32", "$si", "Sf">;
1406def SCALAR_FCVTZU_N_U32 : SInst<"vcvt_n_u32", "bsi", "Sf">;
1407def SCALAR_FCVTZS_N_S64 : SInst<"vcvt_n_s64", "$si", "Sd">;
1408def SCALAR_FCVTZU_N_U64 : SInst<"vcvt_n_u64", "bsi", "Sd">;
1409}
1410
1411////////////////////////////////////////////////////////////////////////////////
1412// Scalar Reduce Pairwise Addition (Scalar and Floating Point)
1413def SCALAR_ADDP  : SInst<"vpadd", "sd", "SfSHlSHdSHUl">;
1414
1415////////////////////////////////////////////////////////////////////////////////
1416// Scalar Reduce Floating Point Pairwise Max/Min
1417def SCALAR_FMAXP : SInst<"vpmax", "sd", "SfSQd">;
1418
1419def SCALAR_FMINP : SInst<"vpmin", "sd", "SfSQd">;
1420
1421////////////////////////////////////////////////////////////////////////////////
1422// Scalar Reduce Floating Point Pairwise maxNum/minNum
1423def SCALAR_FMAXNMP : SInst<"vpmaxnm", "sd", "SfSQd">;
1424def SCALAR_FMINNMP : SInst<"vpminnm", "sd", "SfSQd">;
1425
1426////////////////////////////////////////////////////////////////////////////////
1427// Scalar Integer Saturating Doubling Multiply Half High
1428def SCALAR_SQDMULH : SInst<"vqdmulh", "sss", "SsSi">;
1429
1430////////////////////////////////////////////////////////////////////////////////
1431// Scalar Integer Saturating Rounding Doubling Multiply Half High
1432def SCALAR_SQRDMULH : SInst<"vqrdmulh", "sss", "SsSi">;
1433
1434let ArchGuard = "defined(__ARM_FEATURE_QRDMX) && defined(__aarch64__)" in {
1435////////////////////////////////////////////////////////////////////////////////
1436// Signed Saturating Rounding Doubling Multiply Accumulate Returning High Half
1437def SCALAR_SQRDMLAH : SOpInst<"vqrdmlah", "ssss", "SsSi", OP_QRDMLAH>;
1438
1439////////////////////////////////////////////////////////////////////////////////
1440// Signed Saturating Rounding Doubling Multiply Subtract Returning High Half
1441def SCALAR_SQRDMLSH : SOpInst<"vqrdmlsh", "ssss", "SsSi", OP_QRDMLSH>;
1442}
1443
1444////////////////////////////////////////////////////////////////////////////////
1445// Scalar Floating-point Multiply Extended
1446def SCALAR_FMULX : IInst<"vmulx", "sss", "SfSd">;
1447
1448////////////////////////////////////////////////////////////////////////////////
1449// Scalar Floating-point Reciprocal Step
1450def SCALAR_FRECPS : IInst<"vrecps", "sss", "SfSd">;
1451
1452////////////////////////////////////////////////////////////////////////////////
1453// Scalar Floating-point Reciprocal Square Root Step
1454def SCALAR_FRSQRTS : IInst<"vrsqrts", "sss", "SfSd">;
1455
1456////////////////////////////////////////////////////////////////////////////////
1457// Scalar Signed Integer Convert To Floating-point
1458def SCALAR_SCVTFS : SInst<"vcvt_f32", "ys", "Si">;
1459def SCALAR_SCVTFD : SInst<"vcvt_f64", "os", "Sl">;
1460
1461////////////////////////////////////////////////////////////////////////////////
1462// Scalar Unsigned Integer Convert To Floating-point
1463def SCALAR_UCVTFS : SInst<"vcvt_f32", "ys", "SUi">;
1464def SCALAR_UCVTFD : SInst<"vcvt_f64", "os", "SUl">;
1465
1466////////////////////////////////////////////////////////////////////////////////
1467// Scalar Floating-point Converts
1468def SCALAR_FCVTXN  : IInst<"vcvtx_f32", "ys", "Sd">;
1469def SCALAR_FCVTNSS : SInst<"vcvtn_s32", "$s", "Sf">;
1470def SCALAR_FCVTNUS : SInst<"vcvtn_u32", "bs", "Sf">;
1471def SCALAR_FCVTNSD : SInst<"vcvtn_s64", "$s", "Sd">;
1472def SCALAR_FCVTNUD : SInst<"vcvtn_u64", "bs", "Sd">;
1473def SCALAR_FCVTMSS : SInst<"vcvtm_s32", "$s", "Sf">;
1474def SCALAR_FCVTMUS : SInst<"vcvtm_u32", "bs", "Sf">;
1475def SCALAR_FCVTMSD : SInst<"vcvtm_s64", "$s", "Sd">;
1476def SCALAR_FCVTMUD : SInst<"vcvtm_u64", "bs", "Sd">;
1477def SCALAR_FCVTASS : SInst<"vcvta_s32", "$s", "Sf">;
1478def SCALAR_FCVTAUS : SInst<"vcvta_u32", "bs", "Sf">;
1479def SCALAR_FCVTASD : SInst<"vcvta_s64", "$s", "Sd">;
1480def SCALAR_FCVTAUD : SInst<"vcvta_u64", "bs", "Sd">;
1481def SCALAR_FCVTPSS : SInst<"vcvtp_s32", "$s", "Sf">;
1482def SCALAR_FCVTPUS : SInst<"vcvtp_u32", "bs", "Sf">;
1483def SCALAR_FCVTPSD : SInst<"vcvtp_s64", "$s", "Sd">;
1484def SCALAR_FCVTPUD : SInst<"vcvtp_u64", "bs", "Sd">;
1485def SCALAR_FCVTZSS : SInst<"vcvt_s32", "$s", "Sf">;
1486def SCALAR_FCVTZUS : SInst<"vcvt_u32", "bs", "Sf">;
1487def SCALAR_FCVTZSD : SInst<"vcvt_s64", "$s", "Sd">;
1488def SCALAR_FCVTZUD : SInst<"vcvt_u64", "bs", "Sd">;
1489
1490////////////////////////////////////////////////////////////////////////////////
1491// Scalar Floating-point Reciprocal Estimate
1492def SCALAR_FRECPE : IInst<"vrecpe", "ss", "SfSd">;
1493
1494////////////////////////////////////////////////////////////////////////////////
1495// Scalar Floating-point Reciprocal Exponent
1496def SCALAR_FRECPX : IInst<"vrecpx", "ss", "SfSd">;
1497
1498////////////////////////////////////////////////////////////////////////////////
1499// Scalar Floating-point Reciprocal Square Root Estimate
1500def SCALAR_FRSQRTE : IInst<"vrsqrte", "ss", "SfSd">;
1501
1502////////////////////////////////////////////////////////////////////////////////
1503// Scalar Integer Comparison
1504def SCALAR_CMEQ : SInst<"vceq", "sss", "SlSUl">;
1505def SCALAR_CMEQZ : SInst<"vceqz", "ss", "SlSUl">;
1506def SCALAR_CMGE : SInst<"vcge", "sss", "Sl">;
1507def SCALAR_CMGEZ : SInst<"vcgez", "ss", "Sl">;
1508def SCALAR_CMHS : SInst<"vcge", "sss", "SUl">;
1509def SCALAR_CMLE : SInst<"vcle", "sss", "SlSUl">;
1510def SCALAR_CMLEZ : SInst<"vclez", "ss", "Sl">;
1511def SCALAR_CMLT : SInst<"vclt", "sss", "SlSUl">;
1512def SCALAR_CMLTZ : SInst<"vcltz", "ss", "Sl">;
1513def SCALAR_CMGT : SInst<"vcgt", "sss", "Sl">;
1514def SCALAR_CMGTZ : SInst<"vcgtz", "ss", "Sl">;
1515def SCALAR_CMHI : SInst<"vcgt", "sss", "SUl">;
1516def SCALAR_CMTST : SInst<"vtst", "sss", "SlSUl">;
1517
1518////////////////////////////////////////////////////////////////////////////////
1519// Scalar Floating-point Comparison
1520def SCALAR_FCMEQ : IInst<"vceq", "bss", "SfSd">;
1521def SCALAR_FCMEQZ : IInst<"vceqz", "bs", "SfSd">;
1522def SCALAR_FCMGE : IInst<"vcge", "bss", "SfSd">;
1523def SCALAR_FCMGEZ : IInst<"vcgez", "bs", "SfSd">;
1524def SCALAR_FCMGT : IInst<"vcgt", "bss", "SfSd">;
1525def SCALAR_FCMGTZ : IInst<"vcgtz", "bs", "SfSd">;
1526def SCALAR_FCMLE : IInst<"vcle", "bss", "SfSd">;
1527def SCALAR_FCMLEZ : IInst<"vclez", "bs", "SfSd">;
1528def SCALAR_FCMLT : IInst<"vclt", "bss", "SfSd">;
1529def SCALAR_FCMLTZ : IInst<"vcltz", "bs", "SfSd">;
1530
1531////////////////////////////////////////////////////////////////////////////////
1532// Scalar Floating-point Absolute Compare Mask Greater Than Or Equal
1533def SCALAR_FACGE : IInst<"vcage", "bss", "SfSd">;
1534def SCALAR_FACLE : IInst<"vcale", "bss", "SfSd">;
1535
1536////////////////////////////////////////////////////////////////////////////////
1537// Scalar Floating-point Absolute Compare Mask Greater Than
1538def SCALAR_FACGT : IInst<"vcagt", "bss", "SfSd">;
1539def SCALAR_FACLT : IInst<"vcalt", "bss", "SfSd">;
1540
1541////////////////////////////////////////////////////////////////////////////////
1542// Scalar Absolute Value
1543def SCALAR_ABS : SInst<"vabs", "ss", "Sl">;
1544
1545////////////////////////////////////////////////////////////////////////////////
1546// Scalar Absolute Difference
1547def SCALAR_ABD : IInst<"vabd", "sss", "SfSd">;
1548
1549////////////////////////////////////////////////////////////////////////////////
1550// Scalar Signed Saturating Absolute Value
1551def SCALAR_SQABS : SInst<"vqabs", "ss", "ScSsSiSl">;
1552
1553////////////////////////////////////////////////////////////////////////////////
1554// Scalar Negate
1555def SCALAR_NEG : SInst<"vneg", "ss", "Sl">;
1556
1557////////////////////////////////////////////////////////////////////////////////
1558// Scalar Signed Saturating Negate
1559def SCALAR_SQNEG : SInst<"vqneg", "ss", "ScSsSiSl">;
1560
1561////////////////////////////////////////////////////////////////////////////////
1562// Scalar Signed Saturating Accumulated of Unsigned Value
1563def SCALAR_SUQADD : SInst<"vuqadd", "sss", "ScSsSiSl">;
1564
1565////////////////////////////////////////////////////////////////////////////////
1566// Scalar Unsigned Saturating Accumulated of Signed Value
1567def SCALAR_USQADD : SInst<"vsqadd", "sss", "SUcSUsSUiSUl">;
1568
1569////////////////////////////////////////////////////////////////////////////////
1570// Signed Saturating Doubling Multiply-Add Long
1571def SCALAR_SQDMLAL : SInst<"vqdmlal", "rrss", "SsSi">;
1572
1573////////////////////////////////////////////////////////////////////////////////
1574// Signed Saturating Doubling Multiply-Subtract Long
1575def SCALAR_SQDMLSL : SInst<"vqdmlsl", "rrss", "SsSi">;
1576
1577////////////////////////////////////////////////////////////////////////////////
1578// Signed Saturating Doubling Multiply Long
1579def SCALAR_SQDMULL : SInst<"vqdmull", "rss", "SsSi">;
1580
1581////////////////////////////////////////////////////////////////////////////////
1582// Scalar Signed Saturating Extract Unsigned Narrow
1583def SCALAR_SQXTUN : SInst<"vqmovun", "zs", "SsSiSl">;
1584
1585////////////////////////////////////////////////////////////////////////////////
1586// Scalar Signed Saturating Extract Narrow
1587def SCALAR_SQXTN : SInst<"vqmovn", "zs", "SsSiSl">;
1588
1589////////////////////////////////////////////////////////////////////////////////
1590// Scalar Unsigned Saturating Extract Narrow
1591def SCALAR_UQXTN : SInst<"vqmovn", "zs", "SUsSUiSUl">;
1592
1593// Scalar Floating Point  multiply (scalar, by element)
1594def SCALAR_FMUL_LANE : IOpInst<"vmul_lane", "ssdi", "SfSd", OP_SCALAR_MUL_LN>;
1595def SCALAR_FMUL_LANEQ : IOpInst<"vmul_laneq", "ssji", "SfSd", OP_SCALAR_MUL_LN>;
1596
1597// Scalar Floating Point  multiply extended (scalar, by element)
1598def SCALAR_FMULX_LANE : IOpInst<"vmulx_lane", "ssdi", "SfSd", OP_SCALAR_MULX_LN>;
1599def SCALAR_FMULX_LANEQ : IOpInst<"vmulx_laneq", "ssji", "SfSd", OP_SCALAR_MULX_LN>;
1600
1601def SCALAR_VMUL_N : IInst<"vmul_n", "dds", "d">;
1602
1603// VMUL_LANE_A64 d type implemented using scalar mul lane
1604def SCALAR_VMUL_LANE : IInst<"vmul_lane", "ddgi", "d">;
1605
1606// VMUL_LANEQ d type implemented using scalar mul lane
1607def SCALAR_VMUL_LANEQ   : IInst<"vmul_laneq", "ddji", "d"> {
1608  let isLaneQ = 1;
1609}
1610
1611// VMULX_LANE d type implemented using scalar vmulx_lane
1612def SCALAR_VMULX_LANE : IOpInst<"vmulx_lane", "ddgi", "d", OP_SCALAR_VMULX_LN>;
1613
1614// VMULX_LANEQ d type implemented using scalar vmulx_laneq
1615def SCALAR_VMULX_LANEQ : IOpInst<"vmulx_laneq", "ddji", "d", OP_SCALAR_VMULX_LNQ>;
1616
1617// Scalar Floating Point fused multiply-add (scalar, by element)
1618def SCALAR_FMLA_LANE : IInst<"vfma_lane", "sssdi", "SfSd">;
1619def SCALAR_FMLA_LANEQ : IInst<"vfma_laneq", "sssji", "SfSd">;
1620
1621// Scalar Floating Point fused multiply-subtract (scalar, by element)
1622def SCALAR_FMLS_LANE : IOpInst<"vfms_lane", "sssdi", "SfSd", OP_FMS_LN>;
1623def SCALAR_FMLS_LANEQ : IOpInst<"vfms_laneq", "sssji", "SfSd", OP_FMS_LNQ>;
1624
1625// Signed Saturating Doubling Multiply Long (scalar by element)
1626def SCALAR_SQDMULL_LANE : SOpInst<"vqdmull_lane", "rsdi", "SsSi", OP_SCALAR_QDMULL_LN>;
1627def SCALAR_SQDMULL_LANEQ : SOpInst<"vqdmull_laneq", "rsji", "SsSi", OP_SCALAR_QDMULL_LN>;
1628
1629// Signed Saturating Doubling Multiply-Add Long (scalar by element)
1630def SCALAR_SQDMLAL_LANE : SInst<"vqdmlal_lane", "rrsdi", "SsSi">;
1631def SCALAR_SQDMLAL_LANEQ : SInst<"vqdmlal_laneq", "rrsji", "SsSi">;
1632
1633// Signed Saturating Doubling Multiply-Subtract Long (scalar by element)
1634def SCALAR_SQDMLS_LANE : SInst<"vqdmlsl_lane", "rrsdi", "SsSi">;
1635def SCALAR_SQDMLS_LANEQ : SInst<"vqdmlsl_laneq", "rrsji", "SsSi">;
1636
1637// Scalar Integer Saturating Doubling Multiply Half High (scalar by element)
1638def SCALAR_SQDMULH_LANE : SOpInst<"vqdmulh_lane", "ssdi", "SsSi", OP_SCALAR_QDMULH_LN>;
1639def SCALAR_SQDMULH_LANEQ : SOpInst<"vqdmulh_laneq", "ssji", "SsSi", OP_SCALAR_QDMULH_LN>;
1640
1641// Scalar Integer Saturating Rounding Doubling Multiply Half High
1642def SCALAR_SQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "ssdi", "SsSi", OP_SCALAR_QRDMULH_LN>;
1643def SCALAR_SQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "ssji", "SsSi", OP_SCALAR_QRDMULH_LN>;
1644
1645let ArchGuard = "defined(__ARM_FEATURE_QRDMX) && defined(__aarch64__)" in {
1646// Signed Saturating Rounding Doubling Multiply Accumulate Returning High Half
1647def SCALAR_SQRDMLAH_LANE : SOpInst<"vqrdmlah_lane", "sssdi", "SsSi", OP_SCALAR_QRDMLAH_LN>;
1648def SCALAR_SQRDMLAH_LANEQ : SOpInst<"vqrdmlah_laneq", "sssji", "SsSi", OP_SCALAR_QRDMLAH_LN>;
1649
1650// Signed Saturating Rounding Doubling Multiply Subtract Returning High Half
1651def SCALAR_SQRDMLSH_LANE : SOpInst<"vqrdmlsh_lane", "sssdi", "SsSi", OP_SCALAR_QRDMLSH_LN>;
1652def SCALAR_SQRDMLSH_LANEQ : SOpInst<"vqrdmlsh_laneq", "sssji", "SsSi", OP_SCALAR_QRDMLSH_LN>;
1653}
1654
1655def SCALAR_VDUP_LANE : IInst<"vdup_lane", "sdi", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
1656def SCALAR_VDUP_LANEQ : IInst<"vdup_laneq", "sji", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
1657}
1658