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_FMS_LN   : Op<(call "vfma_lane", $p0, $p1, (op "-", $p2), $p3)>;
377def OP_FMS_LNQ  : Op<(call "vfma_laneq", $p0, $p1, (op "-", $p2), $p3)>;
378def OP_TRN1     : Op<(shuffle $p0, $p1, (interleave (decimate mask0, 2),
379                                                    (decimate mask1, 2)))>;
380def OP_ZIP1     : Op<(shuffle $p0, $p1, (lowhalf (interleave mask0, mask1)))>;
381def OP_UZP1     : Op<(shuffle $p0, $p1, (add (decimate mask0, 2),
382                                             (decimate mask1, 2)))>;
383def OP_TRN2     : Op<(shuffle $p0, $p1, (interleave
384                                          (decimate (rotl mask0, 1), 2),
385                                          (decimate (rotl mask1, 1), 2)))>;
386def OP_ZIP2     : Op<(shuffle $p0, $p1, (highhalf (interleave mask0, mask1)))>;
387def OP_UZP2     : Op<(shuffle $p0, $p1, (add (decimate (rotl mask0, 1), 2),
388                                             (decimate (rotl mask1, 1), 2)))>;
389def OP_EQ       : Op<(cast "R", (op "==", $p0, $p1))>;
390def OP_GE       : Op<(cast "R", (op ">=", $p0, $p1))>;
391def OP_LE       : Op<(cast "R", (op "<=", $p0, $p1))>;
392def OP_GT       : Op<(cast "R", (op ">", $p0, $p1))>;
393def OP_LT       : Op<(cast "R", (op "<", $p0, $p1))>;
394def OP_NEG      : Op<(op "-", $p0)>;
395def OP_NOT      : Op<(op "~", $p0)>;
396def OP_AND      : Op<(op "&", $p0, $p1)>;
397def OP_OR       : Op<(op "|", $p0, $p1)>;
398def OP_XOR      : Op<(op "^", $p0, $p1)>;
399def OP_ANDN     : Op<(op "&", $p0, (op "~", $p1))>;
400def OP_ORN      : Op<(op "|", $p0, (op "~", $p1))>;
401def OP_CAST     : Op<(cast "R", $p0)>;
402def OP_HI       : Op<(shuffle $p0, $p0, (highhalf mask0))>;
403def OP_LO       : Op<(shuffle $p0, $p0, (lowhalf mask0))>;
404def OP_CONC     : Op<(shuffle $p0, $p1, (add mask0, mask1))>;
405def OP_DUP      : Op<(dup $p0)>;
406def OP_DUP_LN   : Op<(splat $p0, $p1)>;
407def OP_SEL      : Op<(cast "R", (op "|",
408                                    (op "&", $p0, (cast $p0, $p1)),
409                                    (op "&", (op "~", $p0), (cast $p0, $p2))))>;
410def OP_REV16    : Op<(shuffle $p0, $p0, (rev 16, mask0))>;
411def OP_REV32    : Op<(shuffle $p0, $p0, (rev 32, mask0))>;
412def OP_REV64    : Op<(shuffle $p0, $p0, (rev 64, mask0))>;
413def OP_XTN      : Op<(call "vcombine", $p0, (call "vmovn", $p1))>;
414def OP_SQXTUN   : Op<(call "vcombine", (cast $p0, "U", $p0),
415                                       (call "vqmovun", $p1))>;
416def OP_QXTN     : Op<(call "vcombine", $p0, (call "vqmovn", $p1))>;
417def OP_VCVT_NA_HI_F16 : Op<(call "vcombine", $p0, (call "vcvt_f16", $p1))>;
418def OP_VCVT_NA_HI_F32 : Op<(call "vcombine", $p0, (call "vcvt_f32_f64", $p1))>;
419def OP_VCVT_EX_HI_F32 : Op<(call "vcvt_f32_f16", (call "vget_high", $p0))>;
420def OP_VCVT_EX_HI_F64 : Op<(call "vcvt_f64_f32", (call "vget_high", $p0))>;
421def OP_VCVTX_HI : Op<(call "vcombine", $p0, (call "vcvtx_f32", $p1))>;
422def OP_REINT    : Op<(cast "R", $p0)>;
423def OP_ADDHNHi  : Op<(call "vcombine", $p0, (call "vaddhn", $p1, $p2))>;
424def OP_RADDHNHi : Op<(call "vcombine", $p0, (call "vraddhn", $p1, $p2))>;
425def OP_SUBHNHi  : Op<(call "vcombine", $p0, (call "vsubhn", $p1, $p2))>;
426def OP_RSUBHNHi : Op<(call "vcombine", $p0, (call "vrsubhn", $p1, $p2))>;
427def OP_ABDL     : Op<(cast "R", (call "vmovl", (cast $p0, "U",
428                                                     (call "vabd", $p0, $p1))))>;
429def OP_ABDLHi   : Op<(call "vabdl", (call "vget_high", $p0),
430                                    (call "vget_high", $p1))>;
431def OP_ABA      : Op<(op "+", $p0, (call "vabd", $p1, $p2))>;
432def OP_ABAL     : Op<(op "+", $p0, (call "vabdl", $p1, $p2))>;
433def OP_ABALHi   : Op<(call "vabal", $p0, (call "vget_high", $p1),
434                                       (call "vget_high", $p2))>;
435def OP_QDMULLHi : Op<(call "vqdmull", (call "vget_high", $p0),
436                                      (call "vget_high", $p1))>;
437def OP_QDMULLHi_N : Op<(call "vqdmull_n", (call "vget_high", $p0), $p1)>;
438def OP_QDMLALHi : Op<(call "vqdmlal", $p0, (call "vget_high", $p1),
439                                           (call "vget_high", $p2))>;
440def OP_QDMLALHi_N : Op<(call "vqdmlal_n", $p0, (call "vget_high", $p1), $p2)>;
441def OP_QDMLSLHi : Op<(call "vqdmlsl", $p0, (call "vget_high", $p1),
442                                           (call "vget_high", $p2))>;
443def OP_QDMLSLHi_N : Op<(call "vqdmlsl_n", $p0, (call "vget_high", $p1), $p2)>;
444def OP_DIV  : Op<(op "/", $p0, $p1)>;
445def OP_LONG_HI : Op<(cast "R", (call (name_replace "_high_", "_"),
446                                                (call "vget_high", $p0), $p1))>;
447def OP_NARROW_HI : Op<(cast "R", (call "vcombine",
448                                       (cast "R", "H", $p0),
449                                       (cast "R", "H",
450                                           (call (name_replace "_high_", "_"),
451                                                 $p1, $p2))))>;
452def OP_MOVL_HI  : LOp<[(save_temp $a1, (call "vget_high", $p0)),
453                       (cast "R",
454                            (call "vshll_n", $a1, (literal "int32_t", "0")))]>;
455def OP_COPY_LN : Op<(call "vset_lane", (call "vget_lane", $p2, $p3), $p0, $p1)>;
456def OP_SCALAR_MUL_LN : Op<(op "*", $p0, (call "vget_lane", $p1, $p2))>;
457def OP_SCALAR_MULX_LN : Op<(call "vmulx", $p0, (call "vget_lane", $p1, $p2))>;
458def OP_SCALAR_VMULX_LN : LOp<[(save_temp $x, (call "vget_lane", $p0,
459                                                    (literal "int32_t", "0"))),
460                              (save_temp $y, (call "vget_lane", $p1, $p2)),
461                              (save_temp $z, (call "vmulx", $x, $y)),
462                              (call "vset_lane", $z, $p0, $p2)]>;
463def OP_SCALAR_VMULX_LNQ : LOp<[(save_temp $x, (call "vget_lane", $p0,
464                                                     (literal "int32_t", "0"))),
465                               (save_temp $y, (call "vget_lane", $p1, $p2)),
466                               (save_temp $z, (call "vmulx", $x, $y)),
467                               (call "vset_lane", $z, $p0, (literal "int32_t",
468                                                                     "0"))]>;
469class ScalarMulOp<string opname> :
470  Op<(call opname, $p0, (call "vget_lane", $p1, $p2))>;
471
472def OP_SCALAR_QDMULL_LN : ScalarMulOp<"vqdmull">;
473def OP_SCALAR_QDMULH_LN : ScalarMulOp<"vqdmulh">;
474def OP_SCALAR_QRDMULH_LN : ScalarMulOp<"vqrdmulh">;
475
476def OP_SCALAR_HALF_GET_LN : Op<(bitcast "float16_t",
477                                   (call "vget_lane",
478                                         (bitcast "int16x4_t", $p0), $p1))>;
479def OP_SCALAR_HALF_GET_LNQ : Op<(bitcast "float16_t",
480                                    (call "vget_lane",
481                                          (bitcast "int16x8_t", $p0), $p1))>;
482def OP_SCALAR_HALF_SET_LN : Op<(bitcast "float16x4_t",
483                                   (call "vset_lane",
484                                         (bitcast "int16_t", $p0),
485                                         (bitcast "int16x4_t", $p1), $p2))>;
486def OP_SCALAR_HALF_SET_LNQ : Op<(bitcast "float16x8_t",
487                                    (call "vset_lane",
488                                          (bitcast "int16_t", $p0),
489                                          (bitcast "int16x8_t", $p1), $p2))>;
490
491//===----------------------------------------------------------------------===//
492// Instructions
493//===----------------------------------------------------------------------===//
494
495////////////////////////////////////////////////////////////////////////////////
496// E.3.1 Addition
497def VADD    : IOpInst<"vadd", "ddd",
498                      "csilfUcUsUiUlQcQsQiQlQfQUcQUsQUiQUl", OP_ADD>;
499def VADDL   : SOpInst<"vaddl", "wdd", "csiUcUsUi", OP_ADDL>;
500def VADDW   : SOpInst<"vaddw", "wwd", "csiUcUsUi", OP_ADDW>;
501def VHADD   : SInst<"vhadd", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
502def VRHADD  : SInst<"vrhadd", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
503def VQADD   : SInst<"vqadd", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
504def VADDHN  : IInst<"vaddhn", "hkk", "silUsUiUl">;
505def VRADDHN : IInst<"vraddhn", "hkk", "silUsUiUl">;
506
507////////////////////////////////////////////////////////////////////////////////
508// E.3.2 Multiplication
509def VMUL     : IOpInst<"vmul", "ddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MUL>;
510def VMULP    : SInst<"vmul", "ddd", "PcQPc">;
511def VMLA     : IOpInst<"vmla", "dddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MLA>;
512def VMLAL    : SOpInst<"vmlal", "wwdd", "csiUcUsUi", OP_MLAL>;
513def VMLS     : IOpInst<"vmls", "dddd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_MLS>;
514def VMLSL    : SOpInst<"vmlsl", "wwdd", "csiUcUsUi", OP_MLSL>;
515def VQDMULH  : SInst<"vqdmulh", "ddd", "siQsQi">;
516def VQRDMULH : SInst<"vqrdmulh", "ddd", "siQsQi">;
517def VQDMLAL  : SInst<"vqdmlal", "wwdd", "si">;
518def VQDMLSL  : SInst<"vqdmlsl", "wwdd", "si">;
519def VMULL    : SInst<"vmull", "wdd", "csiUcUsUiPc">;
520def VQDMULL  : SInst<"vqdmull", "wdd", "si">;
521
522////////////////////////////////////////////////////////////////////////////////
523// E.3.3 Subtraction
524def VSUB    : IOpInst<"vsub", "ddd",
525                      "csilfUcUsUiUlQcQsQiQlQfQUcQUsQUiQUl", OP_SUB>;
526def VSUBL   : SOpInst<"vsubl", "wdd", "csiUcUsUi", OP_SUBL>;
527def VSUBW   : SOpInst<"vsubw", "wwd", "csiUcUsUi", OP_SUBW>;
528def VQSUB   : SInst<"vqsub", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
529def VHSUB   : SInst<"vhsub", "ddd", "csiUcUsUiQcQsQiQUcQUsQUi">;
530def VSUBHN  : IInst<"vsubhn", "hkk", "silUsUiUl">;
531def VRSUBHN : IInst<"vrsubhn", "hkk", "silUsUiUl">;
532
533////////////////////////////////////////////////////////////////////////////////
534// E.3.4 Comparison
535def VCEQ  : IOpInst<"vceq", "udd", "csifUcUsUiPcQcQsQiQfQUcQUsQUiQPc", OP_EQ>;
536def VCGE  : SOpInst<"vcge", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_GE>;
537let InstName = "vcge" in
538def VCLE  : SOpInst<"vcle", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_LE>;
539def VCGT  : SOpInst<"vcgt", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_GT>;
540let InstName = "vcgt" in
541def VCLT  : SOpInst<"vclt", "udd", "csifUcUsUiQcQsQiQfQUcQUsQUi", OP_LT>;
542let InstName = "vacge" in {
543def VCAGE : IInst<"vcage", "udd", "fQf">;
544def VCALE : IInst<"vcale", "udd", "fQf">;
545}
546let InstName = "vacgt" in {
547def VCAGT : IInst<"vcagt", "udd", "fQf">;
548def VCALT : IInst<"vcalt", "udd", "fQf">;
549}
550def VTST  : WInst<"vtst", "udd", "csiUcUsUiPcPsQcQsQiQUcQUsQUiQPcQPs">;
551
552////////////////////////////////////////////////////////////////////////////////
553// E.3.5 Absolute Difference
554def VABD  : SInst<"vabd", "ddd",  "csiUcUsUifQcQsQiQUcQUsQUiQf">;
555def VABDL : SOpInst<"vabdl", "wdd",  "csiUcUsUi", OP_ABDL>;
556def VABA  : SOpInst<"vaba", "dddd", "csiUcUsUiQcQsQiQUcQUsQUi", OP_ABA>;
557def VABAL : SOpInst<"vabal", "wwdd", "csiUcUsUi", OP_ABAL>;
558
559////////////////////////////////////////////////////////////////////////////////
560// E.3.6 Max/Min
561def VMAX : SInst<"vmax", "ddd", "csiUcUsUifQcQsQiQUcQUsQUiQf">;
562def VMIN : SInst<"vmin", "ddd", "csiUcUsUifQcQsQiQUcQUsQUiQf">;
563
564////////////////////////////////////////////////////////////////////////////////
565// E.3.7 Pairwise Addition
566def VPADD  : IInst<"vpadd", "ddd", "csiUcUsUif">;
567def VPADDL : SInst<"vpaddl", "nd",  "csiUcUsUiQcQsQiQUcQUsQUi">;
568def VPADAL : SInst<"vpadal", "nnd", "csiUcUsUiQcQsQiQUcQUsQUi">;
569
570////////////////////////////////////////////////////////////////////////////////
571// E.3.8-9 Folding Max/Min
572def VPMAX : SInst<"vpmax", "ddd", "csiUcUsUif">;
573def VPMIN : SInst<"vpmin", "ddd", "csiUcUsUif">;
574
575////////////////////////////////////////////////////////////////////////////////
576// E.3.10 Reciprocal/Sqrt
577def VRECPS  : IInst<"vrecps", "ddd", "fQf">;
578def VRSQRTS : IInst<"vrsqrts", "ddd", "fQf">;
579
580////////////////////////////////////////////////////////////////////////////////
581// E.3.11 Shifts by signed variable
582def VSHL   : SInst<"vshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
583def VQSHL  : SInst<"vqshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
584def VRSHL  : SInst<"vrshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
585def VQRSHL : SInst<"vqrshl", "ddx", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
586
587////////////////////////////////////////////////////////////////////////////////
588// E.3.12 Shifts by constant
589let isShift = 1 in {
590def VSHR_N     : SInst<"vshr_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
591def VSHL_N     : IInst<"vshl_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
592def VRSHR_N    : SInst<"vrshr_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
593def VSRA_N     : SInst<"vsra_n", "dddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
594def VRSRA_N    : SInst<"vrsra_n", "dddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
595def VQSHL_N    : SInst<"vqshl_n", "ddi", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl">;
596def VQSHLU_N   : SInst<"vqshlu_n", "udi", "csilQcQsQiQl">;
597def VSHRN_N    : IInst<"vshrn_n", "hki", "silUsUiUl">;
598def VQSHRUN_N  : SInst<"vqshrun_n", "eki", "sil">;
599def VQRSHRUN_N : SInst<"vqrshrun_n", "eki", "sil">;
600def VQSHRN_N   : SInst<"vqshrn_n", "hki", "silUsUiUl">;
601def VRSHRN_N   : IInst<"vrshrn_n", "hki", "silUsUiUl">;
602def VQRSHRN_N  : SInst<"vqrshrn_n", "hki", "silUsUiUl">;
603def VSHLL_N    : SInst<"vshll_n", "wdi", "csiUcUsUi">;
604
605////////////////////////////////////////////////////////////////////////////////
606// E.3.13 Shifts with insert
607def VSRI_N : WInst<"vsri_n", "dddi",
608                   "csilUcUsUiUlPcPsQcQsQiQlQUcQUsQUiQUlQPcQPs">;
609def VSLI_N : WInst<"vsli_n", "dddi",
610                   "csilUcUsUiUlPcPsQcQsQiQlQUcQUsQUiQUlQPcQPs">;
611}
612
613////////////////////////////////////////////////////////////////////////////////
614// E.3.14 Loads and stores of a single vector
615def VLD1      : WInst<"vld1", "dc",
616                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
617def VLD1_LANE : WInst<"vld1_lane", "dcdi",
618                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
619def VLD1_DUP  : WInst<"vld1_dup", "dc",
620                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
621def VST1      : WInst<"vst1", "vpd",
622                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
623def VST1_LANE : WInst<"vst1_lane", "vpdi",
624                      "QUcQUsQUiQUlQcQsQiQlQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
625
626////////////////////////////////////////////////////////////////////////////////
627// E.3.15 Loads and stores of an N-element structure
628def VLD2 : WInst<"vld2", "2c", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
629def VLD3 : WInst<"vld3", "3c", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
630def VLD4 : WInst<"vld4", "4c", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
631def VLD2_DUP  : WInst<"vld2_dup", "2c", "UcUsUiUlcsilhfPcPs">;
632def VLD3_DUP  : WInst<"vld3_dup", "3c", "UcUsUiUlcsilhfPcPs">;
633def VLD4_DUP  : WInst<"vld4_dup", "4c", "UcUsUiUlcsilhfPcPs">;
634def VLD2_LANE : WInst<"vld2_lane", "2c2i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
635def VLD3_LANE : WInst<"vld3_lane", "3c3i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
636def VLD4_LANE : WInst<"vld4_lane", "4c4i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
637def VST2 : WInst<"vst2", "vp2", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
638def VST3 : WInst<"vst3", "vp3", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
639def VST4 : WInst<"vst4", "vp4", "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPs">;
640def VST2_LANE : WInst<"vst2_lane", "vp2i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
641def VST3_LANE : WInst<"vst3_lane", "vp3i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
642def VST4_LANE : WInst<"vst4_lane", "vp4i", "QUsQUiQsQiQhQfQPsUcUsUicsihfPcPs">;
643
644////////////////////////////////////////////////////////////////////////////////
645// E.3.16 Extract lanes from a vector
646let InstName = "vmov" in
647def VGET_LANE : IInst<"vget_lane", "sdi",
648                      "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl">;
649
650////////////////////////////////////////////////////////////////////////////////
651// E.3.17 Set lanes within a vector
652let InstName = "vmov" in
653def VSET_LANE : IInst<"vset_lane", "dsdi",
654                      "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl">;
655
656////////////////////////////////////////////////////////////////////////////////
657// E.3.18 Initialize a vector from bit pattern
658def VCREATE : NoTestOpInst<"vcreate", "dl", "csihfUcUsUiUlPcPsl", OP_CAST> {
659  let BigEndianSafe = 1;
660}
661
662////////////////////////////////////////////////////////////////////////////////
663// E.3.19 Set all lanes to same value
664let InstName = "vmov" in {
665def VDUP_N   : WOpInst<"vdup_n", "ds",
666                       "UcUsUicsiPcPshfQUcQUsQUiQcQsQiQPcQPsQhQflUlQlQUl",
667                       OP_DUP>;
668def VMOV_N   : WOpInst<"vmov_n", "ds",
669                       "UcUsUicsiPcPshfQUcQUsQUiQcQsQiQPcQPsQhQflUlQlQUl",
670                       OP_DUP>;
671}
672let InstName = "" in
673def VDUP_LANE: WOpInst<"vdup_lane", "dgi",
674                       "UcUsUicsiPcPsfQUcQUsQUiQcQsQiQPcQPsQflUlQlQUl",
675                       OP_DUP_LN>;
676
677////////////////////////////////////////////////////////////////////////////////
678// E.3.20 Combining vectors
679def VCOMBINE : NoTestOpInst<"vcombine", "kdd", "csilhfUcUsUiUlPcPs", OP_CONC>;
680
681////////////////////////////////////////////////////////////////////////////////
682// E.3.21 Splitting vectors
683let InstName = "vmov" in {
684def VGET_HIGH : NoTestOpInst<"vget_high", "dk", "csilhfUcUsUiUlPcPs", OP_HI>;
685def VGET_LOW  : NoTestOpInst<"vget_low", "dk", "csilhfUcUsUiUlPcPs", OP_LO>;
686}
687
688////////////////////////////////////////////////////////////////////////////////
689// E.3.22 Converting vectors
690def VCVT_S32     : SInst<"vcvt_s32", "xd",  "fQf">;
691def VCVT_U32     : SInst<"vcvt_u32", "ud",  "fQf">;
692def VCVT_F16     : SInst<"vcvt_f16", "hk",  "f">;
693def VCVT_F32     : SInst<"vcvt_f32", "fd",  "iUiQiQUi">;
694def VCVT_F32_F16 : SInst<"vcvt_f32_f16", "fd",  "h">;
695let isVCVT_N = 1 in {
696def VCVT_N_S32   : SInst<"vcvt_n_s32", "xdi", "fQf">;
697def VCVT_N_U32   : SInst<"vcvt_n_u32", "udi", "fQf">;
698def VCVT_N_F32   : SInst<"vcvt_n_f32", "fdi", "iUiQiQUi">;
699}
700def VMOVN        : IInst<"vmovn", "hk",  "silUsUiUl">;
701def VMOVL        : SInst<"vmovl", "wd",  "csiUcUsUi">;
702def VQMOVN       : SInst<"vqmovn", "hk",  "silUsUiUl">;
703def VQMOVUN      : SInst<"vqmovun", "ek",  "sil">;
704
705////////////////////////////////////////////////////////////////////////////////
706// E.3.23-24 Table lookup, Extended table lookup
707let InstName = "vtbl" in {
708def VTBL1 : WInst<"vtbl1", "ddt",  "UccPc">;
709def VTBL2 : WInst<"vtbl2", "d2t",  "UccPc">;
710def VTBL3 : WInst<"vtbl3", "d3t",  "UccPc">;
711def VTBL4 : WInst<"vtbl4", "d4t",  "UccPc">;
712}
713let InstName = "vtbx" in {
714def VTBX1 : WInst<"vtbx1", "dddt", "UccPc">;
715def VTBX2 : WInst<"vtbx2", "dd2t", "UccPc">;
716def VTBX3 : WInst<"vtbx3", "dd3t", "UccPc">;
717def VTBX4 : WInst<"vtbx4", "dd4t", "UccPc">;
718}
719
720////////////////////////////////////////////////////////////////////////////////
721// E.3.25 Operations with a scalar value
722def VMLA_LANE     : IOpInst<"vmla_lane", "dddgi",
723                            "siUsUifQsQiQUsQUiQf", OP_MLA_LN>;
724def VMLAL_LANE    : SOpInst<"vmlal_lane", "wwddi", "siUsUi", OP_MLAL_LN>;
725def VQDMLAL_LANE  : SOpInst<"vqdmlal_lane", "wwddi", "si", OP_QDMLAL_LN>;
726def VMLS_LANE     : IOpInst<"vmls_lane", "dddgi",
727                            "siUsUifQsQiQUsQUiQf", OP_MLS_LN>;
728def VMLSL_LANE    : SOpInst<"vmlsl_lane", "wwddi", "siUsUi", OP_MLSL_LN>;
729def VQDMLSL_LANE  : SOpInst<"vqdmlsl_lane", "wwddi", "si", OP_QDMLSL_LN>;
730def VMUL_N        : IOpInst<"vmul_n", "dds", "sifUsUiQsQiQfQUsQUi", OP_MUL_N>;
731def VMUL_LANE     : IOpInst<"vmul_lane", "ddgi",
732                            "sifUsUiQsQiQfQUsQUi", OP_MUL_LN>;
733def VMULL_N       : SInst<"vmull_n", "wda", "siUsUi">;
734def VMULL_LANE    : SOpInst<"vmull_lane", "wddi", "siUsUi", OP_MULL_LN>;
735def VQDMULL_N     : SInst<"vqdmull_n", "wda", "si">;
736def VQDMULL_LANE  : SOpInst<"vqdmull_lane", "wddi", "si", OP_QDMULL_LN>;
737def VQDMULH_N     : SInst<"vqdmulh_n", "dda", "siQsQi">;
738def VQDMULH_LANE  : SOpInst<"vqdmulh_lane", "ddgi", "siQsQi", OP_QDMULH_LN>;
739def VQRDMULH_N    : SInst<"vqrdmulh_n", "dda", "siQsQi">;
740def VQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "ddgi", "siQsQi", OP_QRDMULH_LN>;
741def VMLA_N        : IOpInst<"vmla_n", "ddda", "siUsUifQsQiQUsQUiQf", OP_MLA_N>;
742def VMLAL_N       : SOpInst<"vmlal_n", "wwda", "siUsUi", OP_MLAL_N>;
743def VQDMLAL_N     : SInst<"vqdmlal_n", "wwda", "si">;
744def VMLS_N        : IOpInst<"vmls_n", "ddds", "siUsUifQsQiQUsQUiQf", OP_MLS_N>;
745def VMLSL_N       : SOpInst<"vmlsl_n", "wwda", "siUsUi", OP_MLSL_N>;
746def VQDMLSL_N     : SInst<"vqdmlsl_n", "wwda", "si">;
747
748////////////////////////////////////////////////////////////////////////////////
749// E.3.26 Vector Extract
750def VEXT : WInst<"vext", "dddi",
751                 "cUcPcsUsPsiUilUlfQcQUcQPcQsQUsQPsQiQUiQlQUlQf">;
752
753////////////////////////////////////////////////////////////////////////////////
754// E.3.27 Reverse vector elements
755def VREV64 : WOpInst<"vrev64", "dd", "csiUcUsUiPcPsfQcQsQiQUcQUsQUiQPcQPsQf",
756                  OP_REV64>;
757def VREV32 : WOpInst<"vrev32", "dd", "csUcUsPcPsQcQsQUcQUsQPcQPs", OP_REV32>;
758def VREV16 : WOpInst<"vrev16", "dd", "cUcPcQcQUcQPc", OP_REV16>;
759
760////////////////////////////////////////////////////////////////////////////////
761// E.3.28 Other single operand arithmetic
762def VABS    : SInst<"vabs", "dd", "csifQcQsQiQf">;
763def VQABS   : SInst<"vqabs", "dd", "csiQcQsQi">;
764def VNEG    : SOpInst<"vneg", "dd", "csifQcQsQiQf", OP_NEG>;
765def VQNEG   : SInst<"vqneg", "dd", "csiQcQsQi">;
766def VCLS    : SInst<"vcls", "dd", "csiQcQsQi">;
767def VCLZ    : IInst<"vclz", "dd", "csiUcUsUiQcQsQiQUcQUsQUi">;
768def VCNT    : WInst<"vcnt", "dd", "UccPcQUcQcQPc">;
769def VRECPE  : SInst<"vrecpe", "dd", "fUiQfQUi">;
770def VRSQRTE : SInst<"vrsqrte", "dd", "fUiQfQUi">;
771
772////////////////////////////////////////////////////////////////////////////////
773// E.3.29 Logical operations
774def VMVN : LOpInst<"vmvn", "dd", "csiUcUsUiPcQcQsQiQUcQUsQUiQPc", OP_NOT>;
775def VAND : LOpInst<"vand", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_AND>;
776def VORR : LOpInst<"vorr", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_OR>;
777def VEOR : LOpInst<"veor", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_XOR>;
778def VBIC : LOpInst<"vbic", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_ANDN>;
779def VORN : LOpInst<"vorn", "ddd", "csilUcUsUiUlQcQsQiQlQUcQUsQUiQUl", OP_ORN>;
780let isHiddenLInst = 1 in
781def VBSL : SInst<"vbsl", "dudd",
782                "csilUcUsUiUlfPcPsQcQsQiQlQUcQUsQUiQUlQfQPcQPs">;
783
784////////////////////////////////////////////////////////////////////////////////
785// E.3.30 Transposition operations
786def VTRN : WInst<"vtrn", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
787def VZIP : WInst<"vzip", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
788def VUZP : WInst<"vuzp", "2dd", "csiUcUsUifPcPsQcQsQiQUcQUsQUiQfQPcQPs">;
789
790////////////////////////////////////////////////////////////////////////////////
791// E.3.31 Vector reinterpret cast operations
792def VREINTERPRET
793  : NoTestOpInst<"vreinterpret", "dd",
794         "csilUcUsUiUlhfPcPsQcQsQiQlQUcQUsQUiQUlQhQfQPcQPs", OP_REINT> {
795  let CartesianProductOfTypes = 1;
796  let ArchGuard = "!defined(__aarch64__)";
797  let BigEndianSafe = 1;
798}
799
800////////////////////////////////////////////////////////////////////////////////
801// Vector fused multiply-add operations
802
803def VFMA : SInst<"vfma", "dddd", "fQf">;
804
805////////////////////////////////////////////////////////////////////////////////
806// AArch64 Intrinsics
807
808let ArchGuard = "defined(__aarch64__)" in {
809
810////////////////////////////////////////////////////////////////////////////////
811// Load/Store
812def LD1 : WInst<"vld1", "dc", "dQdPlQPl">;
813def LD2 : WInst<"vld2", "2c", "QUlQldQdPlQPl">;
814def LD3 : WInst<"vld3", "3c", "QUlQldQdPlQPl">;
815def LD4 : WInst<"vld4", "4c", "QUlQldQdPlQPl">;
816def ST1 : WInst<"vst1", "vpd", "dQdPlQPl">;
817def ST2 : WInst<"vst2", "vp2", "QUlQldQdPlQPl">;
818def ST3 : WInst<"vst3", "vp3", "QUlQldQdPlQPl">;
819def ST4 : WInst<"vst4", "vp4", "QUlQldQdPlQPl">;
820
821def LD1_X2 : WInst<"vld1_x2", "2c",
822                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
823def LD3_x3 : WInst<"vld1_x3", "3c",
824                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
825def LD4_x4 : WInst<"vld1_x4", "4c",
826                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
827
828def ST1_X2 : WInst<"vst1_x2", "vp2",
829                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
830def ST1_X3 : WInst<"vst1_x3", "vp3",
831                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
832def ST1_X4 : WInst<"vst1_x4", "vp4",
833                   "QUcQUsQUiQcQsQiQhQfQPcQPsUcUsUiUlcsilhfPcPsQUlQldQdPlQPl">;
834
835def LD1_LANE : WInst<"vld1_lane", "dcdi", "dQdPlQPl">;
836def LD2_LANE : WInst<"vld2_lane", "2c2i", "lUlQcQUcQPcQlQUldQdPlQPl">;
837def LD3_LANE : WInst<"vld3_lane", "3c3i", "lUlQcQUcQPcQlQUldQdPlQPl">;
838def LD4_LANE : WInst<"vld4_lane", "4c4i", "lUlQcQUcQPcQlQUldQdPlQPl">;
839def ST1_LANE : WInst<"vst1_lane", "vpdi", "dQdPlQPl">;
840def ST2_LANE : WInst<"vst2_lane", "vp2i", "lUlQcQUcQPcQlQUldQdPlQPl">;
841def ST3_LANE : WInst<"vst3_lane", "vp3i", "lUlQcQUcQPcQlQUldQdPlQPl">;
842def ST4_LANE : WInst<"vst4_lane", "vp4i", "lUlQcQUcQPcQlQUldQdPlQPl">;
843
844def LD1_DUP  : WInst<"vld1_dup", "dc", "dQdPlQPl">;
845def LD2_DUP  : WInst<"vld2_dup", "2c",
846                     "QUcQUsQUiQUlQcQsQiQlQhQfQdQPcQPsQPldPl">;
847def LD3_DUP  : WInst<"vld3_dup", "3c",
848                     "QUcQUsQUiQUlQcQsQiQlQhQfQdQPcQPsQPldPl">;
849def LD4_DUP  : WInst<"vld4_dup", "4c",
850                     "QUcQUsQUiQUlQcQsQiQlQhQfQdQPcQPsQPldPl">;
851
852def VLDRQ : WInst<"vldrq", "sc", "Pk">;
853def VSTRQ : WInst<"vstrq", "vps", "Pk">;
854
855////////////////////////////////////////////////////////////////////////////////
856// Addition
857def ADD : IOpInst<"vadd", "ddd", "dQd", OP_ADD>;
858
859////////////////////////////////////////////////////////////////////////////////
860// Subtraction
861def SUB : IOpInst<"vsub", "ddd", "dQd", OP_SUB>;
862
863////////////////////////////////////////////////////////////////////////////////
864// Multiplication
865def MUL     : IOpInst<"vmul", "ddd", "dQd", OP_MUL>;
866def MLA     : IOpInst<"vmla", "dddd", "dQd", OP_MLA>;
867def MLS     : IOpInst<"vmls", "dddd", "dQd", OP_MLS>;
868
869////////////////////////////////////////////////////////////////////////////////
870// Multiplication Extended
871def MULX : SInst<"vmulx", "ddd", "fdQfQd">;
872
873////////////////////////////////////////////////////////////////////////////////
874// Division
875def FDIV : IOpInst<"vdiv", "ddd",  "fdQfQd", OP_DIV>;
876
877////////////////////////////////////////////////////////////////////////////////
878// Vector fused multiply-add operations
879def FMLA : SInst<"vfma", "dddd", "dQd">;
880def FMLS : SInst<"vfms", "dddd", "fdQfQd">;
881
882////////////////////////////////////////////////////////////////////////////////
883// MUL, MLA, MLS, FMA, FMS definitions with scalar argument
884def VMUL_N_A64 : IOpInst<"vmul_n", "dds", "Qd", OP_MUL_N>;
885
886def FMLA_N : SOpInst<"vfma_n", "ddds", "fQfQd", OP_FMLA_N>;
887def FMLS_N : SOpInst<"vfms_n", "ddds", "fQfQd", OP_FMLS_N>;
888
889def MLA_N : SOpInst<"vmla_n", "ddds", "Qd", OP_MLA_N>;
890def MLS_N : SOpInst<"vmls_n", "ddds", "Qd", OP_MLS_N>;
891
892////////////////////////////////////////////////////////////////////////////////
893// Logical operations
894def BSL : SInst<"vbsl", "dudd", "dPlQdQPl">;
895
896////////////////////////////////////////////////////////////////////////////////
897// Absolute Difference
898def ABD  : SInst<"vabd", "ddd",  "dQd">;
899
900////////////////////////////////////////////////////////////////////////////////
901// saturating absolute/negate
902def ABS    : SInst<"vabs", "dd", "dQdlQl">;
903def QABS   : SInst<"vqabs", "dd", "lQl">;
904def NEG    : SOpInst<"vneg", "dd", "dlQdQl", OP_NEG>;
905def QNEG   : SInst<"vqneg", "dd", "lQl">;
906
907////////////////////////////////////////////////////////////////////////////////
908// Signed Saturating Accumulated of Unsigned Value
909def SUQADD : SInst<"vuqadd", "ddd", "csilQcQsQiQl">;
910
911////////////////////////////////////////////////////////////////////////////////
912// Unsigned Saturating Accumulated of Signed Value
913def USQADD : SInst<"vsqadd", "ddd", "UcUsUiUlQUcQUsQUiQUl">;
914
915////////////////////////////////////////////////////////////////////////////////
916// Reciprocal/Sqrt
917def FRECPS  : IInst<"vrecps", "ddd", "dQd">;
918def FRSQRTS : IInst<"vrsqrts", "ddd", "dQd">;
919
920////////////////////////////////////////////////////////////////////////////////
921// bitwise reverse
922def RBIT : IInst<"vrbit", "dd", "cUcPcQcQUcQPc">;
923
924////////////////////////////////////////////////////////////////////////////////
925// Integer extract and narrow to high
926def XTN2 : SOpInst<"vmovn_high", "qhk", "silUsUiUl", OP_XTN>;
927
928////////////////////////////////////////////////////////////////////////////////
929// Signed integer saturating extract and unsigned narrow to high
930def SQXTUN2 : SOpInst<"vqmovun_high", "qhk", "sil", OP_SQXTUN>;
931
932////////////////////////////////////////////////////////////////////////////////
933// Integer saturating extract and narrow to high
934def QXTN2 : SOpInst<"vqmovn_high", "qhk", "silUsUiUl", OP_QXTN>;
935
936////////////////////////////////////////////////////////////////////////////////
937// Converting vectors
938def VCVT_HIGH_F16 : SOpInst<"vcvt_high_f16", "qhj", "f", OP_VCVT_NA_HI_F16>;
939def VCVT_HIGH_F32_F16 : SOpInst<"vcvt_high_f32", "wk", "h", OP_VCVT_EX_HI_F32>;
940def VCVT_F32_F64 : SInst<"vcvt_f32_f64", "md", "Qd">;
941def VCVT_HIGH_F32_F64 : SOpInst<"vcvt_high_f32", "qfj", "d", OP_VCVT_NA_HI_F32>;
942def VCVT_F64_F32 : SInst<"vcvt_f64_f32", "wd", "f">;
943def VCVT_F64 : SInst<"vcvt_f64", "Fd",  "lUlQlQUl">;
944def VCVT_HIGH_F64_F32 : SOpInst<"vcvt_high_f64", "wj", "f", OP_VCVT_EX_HI_F64>;
945def VCVTX_F32_F64 : SInst<"vcvtx_f32", "fj",  "d">;
946def VCVTX_HIGH_F32_F64 : SOpInst<"vcvtx_high_f32", "qfj", "d", OP_VCVTX_HI>;
947def VCVT_S64 : SInst<"vcvt_s64", "xd",  "dQd">;
948def VCVT_U64 : SInst<"vcvt_u64", "ud",  "dQd">;
949def FRECPE  : SInst<"vrecpe", "dd", "dQd">;
950def FRSQRTE : SInst<"vrsqrte", "dd", "dQd">;
951def FSQRT   : SInst<"vsqrt", "dd", "fdQfQd">;
952
953////////////////////////////////////////////////////////////////////////////////
954// Comparison
955def FCAGE : IInst<"vcage", "udd", "dQd">;
956def FCAGT : IInst<"vcagt", "udd", "dQd">;
957def FCALE : IInst<"vcale", "udd", "dQd">;
958def FCALT : IInst<"vcalt", "udd", "dQd">;
959def CMTST  : WInst<"vtst", "udd", "lUlPlQlQUlQPl">;
960def CFMEQ  : SOpInst<"vceq", "udd", "lUldQdQlQUlPlQPl", OP_EQ>;
961def CFMGE  : SOpInst<"vcge", "udd", "lUldQdQlQUl", OP_GE>;
962def CFMLE  : SOpInst<"vcle", "udd", "lUldQdQlQUl", OP_LE>;
963def CFMGT  : SOpInst<"vcgt", "udd", "lUldQdQlQUl", OP_GT>;
964def CFMLT  : SOpInst<"vclt", "udd", "lUldQdQlQUl", OP_LT>;
965
966def CMEQ  : SInst<"vceqz", "ud",
967                  "csilfUcUsUiUlPcPsPlQcQsQiQlQfQUcQUsQUiQUlQPcQPsdQdQPl">;
968def CMGE  : SInst<"vcgez", "ud", "csilfdQcQsQiQlQfQd">;
969def CMLE  : SInst<"vclez", "ud", "csilfdQcQsQiQlQfQd">;
970def CMGT  : SInst<"vcgtz", "ud", "csilfdQcQsQiQlQfQd">;
971def CMLT  : SInst<"vcltz", "ud", "csilfdQcQsQiQlQfQd">;
972
973////////////////////////////////////////////////////////////////////////////////
974// Max/Min Integer
975def MAX : SInst<"vmax", "ddd", "dQd">;
976def MIN : SInst<"vmin", "ddd", "dQd">;
977
978////////////////////////////////////////////////////////////////////////////////
979// Pairwise Max/Min
980def MAXP : SInst<"vpmax", "ddd", "QcQsQiQUcQUsQUiQfQd">;
981def MINP : SInst<"vpmin", "ddd", "QcQsQiQUcQUsQUiQfQd">;
982
983////////////////////////////////////////////////////////////////////////////////
984// Pairwise MaxNum/MinNum Floating Point
985def FMAXNMP : SInst<"vpmaxnm", "ddd", "fQfQd">;
986def FMINNMP : SInst<"vpminnm", "ddd", "fQfQd">;
987
988////////////////////////////////////////////////////////////////////////////////
989// Pairwise Addition
990def ADDP  : IInst<"vpadd", "ddd", "QcQsQiQlQUcQUsQUiQUlQfQd">;
991
992////////////////////////////////////////////////////////////////////////////////
993// Shifts by constant
994let isShift = 1 in {
995// Left shift long high
996def SHLL_HIGH_N    : SOpInst<"vshll_high_n", "ndi", "HcHsHiHUcHUsHUi",
997                             OP_LONG_HI>;
998
999////////////////////////////////////////////////////////////////////////////////
1000def SRI_N : WInst<"vsri_n", "dddi", "PlQPl">;
1001def SLI_N : WInst<"vsli_n", "dddi", "PlQPl">;
1002
1003// Right shift narrow high
1004def SHRN_HIGH_N    : IOpInst<"vshrn_high_n", "hmdi",
1005                             "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
1006def QSHRUN_HIGH_N  : SOpInst<"vqshrun_high_n", "hmdi",
1007                             "HsHiHl", OP_NARROW_HI>;
1008def RSHRN_HIGH_N   : IOpInst<"vrshrn_high_n", "hmdi",
1009                             "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
1010def QRSHRUN_HIGH_N : SOpInst<"vqrshrun_high_n", "hmdi",
1011                             "HsHiHl", OP_NARROW_HI>;
1012def QSHRN_HIGH_N   : SOpInst<"vqshrn_high_n", "hmdi",
1013                             "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
1014def QRSHRN_HIGH_N  : SOpInst<"vqrshrn_high_n", "hmdi",
1015                             "HsHiHlHUsHUiHUl", OP_NARROW_HI>;
1016}
1017
1018////////////////////////////////////////////////////////////////////////////////
1019// Converting vectors
1020def VMOVL_HIGH   : SOpInst<"vmovl_high", "nd", "HcHsHiHUcHUsHUi", OP_MOVL_HI>;
1021
1022let isVCVT_N = 1 in {
1023def CVTF_N_F64   : SInst<"vcvt_n_f64", "Fdi", "lUlQlQUl">;
1024def FCVTZS_N_S64 : SInst<"vcvt_n_s64", "xdi", "dQd">;
1025def FCVTZS_N_U64 : SInst<"vcvt_n_u64", "udi", "dQd">;
1026}
1027
1028////////////////////////////////////////////////////////////////////////////////
1029// 3VDiff class using high 64-bit in operands
1030def VADDL_HIGH   : SOpInst<"vaddl_high", "wkk", "csiUcUsUi", OP_ADDLHi>;
1031def VADDW_HIGH   : SOpInst<"vaddw_high", "wwk", "csiUcUsUi", OP_ADDWHi>;
1032def VSUBL_HIGH   : SOpInst<"vsubl_high", "wkk", "csiUcUsUi", OP_SUBLHi>;
1033def VSUBW_HIGH   : SOpInst<"vsubw_high", "wwk", "csiUcUsUi", OP_SUBWHi>;
1034
1035def VABDL_HIGH   : SOpInst<"vabdl_high", "wkk",  "csiUcUsUi", OP_ABDLHi>;
1036def VABAL_HIGH   : SOpInst<"vabal_high", "wwkk", "csiUcUsUi", OP_ABALHi>;
1037
1038def VMULL_HIGH   : SOpInst<"vmull_high", "wkk", "csiUcUsUiPc", OP_MULLHi>;
1039def VMULL_HIGH_N : SOpInst<"vmull_high_n", "wks", "siUsUi", OP_MULLHi_N>;
1040def VMLAL_HIGH   : SOpInst<"vmlal_high", "wwkk", "csiUcUsUi", OP_MLALHi>;
1041def VMLAL_HIGH_N : SOpInst<"vmlal_high_n", "wwks", "siUsUi", OP_MLALHi_N>;
1042def VMLSL_HIGH   : SOpInst<"vmlsl_high", "wwkk", "csiUcUsUi", OP_MLSLHi>;
1043def VMLSL_HIGH_N : SOpInst<"vmlsl_high_n", "wwks", "siUsUi", OP_MLSLHi_N>;
1044
1045def VADDHN_HIGH  : SOpInst<"vaddhn_high", "qhkk", "silUsUiUl", OP_ADDHNHi>;
1046def VRADDHN_HIGH : SOpInst<"vraddhn_high", "qhkk", "silUsUiUl", OP_RADDHNHi>;
1047def VSUBHN_HIGH  : SOpInst<"vsubhn_high", "qhkk", "silUsUiUl", OP_SUBHNHi>;
1048def VRSUBHN_HIGH : SOpInst<"vrsubhn_high", "qhkk", "silUsUiUl", OP_RSUBHNHi>;
1049
1050def VQDMULL_HIGH : SOpInst<"vqdmull_high", "wkk", "si", OP_QDMULLHi>;
1051def VQDMULL_HIGH_N : SOpInst<"vqdmull_high_n", "wks", "si", OP_QDMULLHi_N>;
1052def VQDMLAL_HIGH : SOpInst<"vqdmlal_high", "wwkk", "si", OP_QDMLALHi>;
1053def VQDMLAL_HIGH_N : SOpInst<"vqdmlal_high_n", "wwks", "si", OP_QDMLALHi_N>;
1054def VQDMLSL_HIGH : SOpInst<"vqdmlsl_high", "wwkk", "si", OP_QDMLSLHi>;
1055def VQDMLSL_HIGH_N : SOpInst<"vqdmlsl_high_n", "wwks", "si", OP_QDMLSLHi_N>;
1056def VMULL_P64    : SInst<"vmull", "rss", "Pl">;
1057def VMULL_HIGH_P64 : SOpInst<"vmull_high", "rdd", "HPl", OP_MULLHi_P64>;
1058
1059
1060////////////////////////////////////////////////////////////////////////////////
1061// Extract or insert element from vector
1062def GET_LANE : IInst<"vget_lane", "sdi", "dQdPlQPl">;
1063def SET_LANE : IInst<"vset_lane", "dsdi", "dQdPlQPl">;
1064def COPY_LANE : IOpInst<"vcopy_lane", "ddidi",
1065                        "csilUcUsUiUlPcPsPlfd", OP_COPY_LN>;
1066def COPYQ_LANE : IOpInst<"vcopy_lane", "ddigi",
1067                        "QcQsQiQlQUcQUsQUiQUlQPcQPsQfQdQPl", OP_COPY_LN>;
1068def COPY_LANEQ : IOpInst<"vcopy_laneq", "ddiki",
1069                     "csilPcPsPlUcUsUiUlfd", OP_COPY_LN>;
1070def COPYQ_LANEQ : IOpInst<"vcopy_laneq", "ddidi",
1071                     "QcQsQiQlQUcQUsQUiQUlQPcQPsQfQdQPl", OP_COPY_LN>;
1072
1073////////////////////////////////////////////////////////////////////////////////
1074// Set all lanes to same value
1075def VDUP_LANE1: WOpInst<"vdup_lane", "dgi", "hdQhQdPlQPl", OP_DUP_LN>;
1076def VDUP_LANE2: WOpInst<"vdup_laneq", "dji",
1077                  "csilUcUsUiUlPcPshfdQcQsQiQlQPcQPsQUcQUsQUiQUlQhQfQdPlQPl",
1078                        OP_DUP_LN>;
1079def DUP_N   : WOpInst<"vdup_n", "ds", "dQdPlQPl", OP_DUP>;
1080def MOV_N   : WOpInst<"vmov_n", "ds", "dQdPlQPl", OP_DUP>;
1081
1082////////////////////////////////////////////////////////////////////////////////
1083def COMBINE : NoTestOpInst<"vcombine", "kdd", "dPl", OP_CONC>;
1084
1085////////////////////////////////////////////////////////////////////////////////
1086//Initialize a vector from bit pattern
1087def CREATE : NoTestOpInst<"vcreate", "dl", "dPl", OP_CAST> {
1088  let BigEndianSafe = 1;
1089}
1090
1091////////////////////////////////////////////////////////////////////////////////
1092
1093def VMLA_LANEQ   : IOpInst<"vmla_laneq", "dddji",
1094                           "siUsUifQsQiQUsQUiQf", OP_MLA_LN>;
1095def VMLS_LANEQ   : IOpInst<"vmls_laneq", "dddji",
1096                           "siUsUifQsQiQUsQUiQf", OP_MLS_LN>;
1097
1098def VFMA_LANE    : IInst<"vfma_lane", "dddgi", "fdQfQd">;
1099def VFMA_LANEQ   : IInst<"vfma_laneq", "dddji", "fdQfQd"> {
1100  let isLaneQ = 1;
1101}
1102def VFMS_LANE    : IOpInst<"vfms_lane", "dddgi", "fdQfQd", OP_FMS_LN>;
1103def VFMS_LANEQ   : IOpInst<"vfms_laneq", "dddji", "fdQfQd", OP_FMS_LNQ>;
1104
1105def VMLAL_LANEQ  : SOpInst<"vmlal_laneq", "wwdki", "siUsUi", OP_MLAL_LN>;
1106def VMLAL_HIGH_LANE   : SOpInst<"vmlal_high_lane", "wwkdi", "siUsUi",
1107                                OP_MLALHi_LN>;
1108def VMLAL_HIGH_LANEQ  : SOpInst<"vmlal_high_laneq", "wwkki", "siUsUi",
1109                                OP_MLALHi_LN>;
1110def VMLSL_LANEQ  : SOpInst<"vmlsl_laneq", "wwdki", "siUsUi", OP_MLSL_LN>;
1111def VMLSL_HIGH_LANE   : SOpInst<"vmlsl_high_lane", "wwkdi", "siUsUi",
1112                                OP_MLSLHi_LN>;
1113def VMLSL_HIGH_LANEQ  : SOpInst<"vmlsl_high_laneq", "wwkki", "siUsUi",
1114                                OP_MLSLHi_LN>;
1115
1116def VQDMLAL_LANEQ  : SOpInst<"vqdmlal_laneq", "wwdki", "si", OP_QDMLAL_LN>;
1117def VQDMLAL_HIGH_LANE   : SOpInst<"vqdmlal_high_lane", "wwkdi", "si",
1118                                OP_QDMLALHi_LN>;
1119def VQDMLAL_HIGH_LANEQ  : SOpInst<"vqdmlal_high_laneq", "wwkki", "si",
1120                                OP_QDMLALHi_LN>;
1121def VQDMLSL_LANEQ  : SOpInst<"vqdmlsl_laneq", "wwdki", "si", OP_QDMLSL_LN>;
1122def VQDMLSL_HIGH_LANE   : SOpInst<"vqdmlsl_high_lane", "wwkdi", "si",
1123                                OP_QDMLSLHi_LN>;
1124def VQDMLSL_HIGH_LANEQ  : SOpInst<"vqdmlsl_high_laneq", "wwkki", "si",
1125                                OP_QDMLSLHi_LN>;
1126
1127// Newly add double parameter for vmul_lane in aarch64
1128// Note: d type is handled by SCALAR_VMUL_LANE
1129def VMUL_LANE_A64 : IOpInst<"vmul_lane", "ddgi", "Qd", OP_MUL_LN>;
1130
1131// Note: d type is handled by SCALAR_VMUL_LANEQ
1132def VMUL_LANEQ   : IOpInst<"vmul_laneq", "ddji",
1133                           "sifUsUiQsQiQUsQUiQfQd", OP_MUL_LN>;
1134def VMULL_LANEQ  : SOpInst<"vmull_laneq", "wdki", "siUsUi", OP_MULL_LN>;
1135def VMULL_HIGH_LANE   : SOpInst<"vmull_high_lane", "wkdi", "siUsUi",
1136                                OP_MULLHi_LN>;
1137def VMULL_HIGH_LANEQ  : SOpInst<"vmull_high_laneq", "wkki", "siUsUi",
1138                                OP_MULLHi_LN>;
1139
1140def VQDMULL_LANEQ  : SOpInst<"vqdmull_laneq", "wdki", "si", OP_QDMULL_LN>;
1141def VQDMULL_HIGH_LANE   : SOpInst<"vqdmull_high_lane", "wkdi", "si",
1142                                  OP_QDMULLHi_LN>;
1143def VQDMULL_HIGH_LANEQ  : SOpInst<"vqdmull_high_laneq", "wkki", "si",
1144                                  OP_QDMULLHi_LN>;
1145
1146def VQDMULH_LANEQ  : SOpInst<"vqdmulh_laneq", "ddji", "siQsQi", OP_QDMULH_LN>;
1147def VQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "ddji", "siQsQi", OP_QRDMULH_LN>;
1148
1149// Note: d type implemented by SCALAR_VMULX_LANE
1150def VMULX_LANE : IOpInst<"vmulx_lane", "ddgi", "fQfQd", OP_MULX_LN>;
1151// Note: d type is implemented by SCALAR_VMULX_LANEQ
1152def VMULX_LANEQ : IOpInst<"vmulx_laneq", "ddji", "fQfQd", OP_MULX_LN>;
1153
1154////////////////////////////////////////////////////////////////////////////////
1155// Across vectors class
1156def VADDLV  : SInst<"vaddlv", "rd", "csiUcUsUiQcQsQiQUcQUsQUi">;
1157def VMAXV   : SInst<"vmaxv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
1158def VMINV   : SInst<"vminv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQd">;
1159def VADDV   : SInst<"vaddv", "sd", "csifUcUsUiQcQsQiQUcQUsQUiQfQdQlQUl">;
1160def FMAXNMV : SInst<"vmaxnmv", "sd", "fQfQd">;
1161def FMINNMV : SInst<"vminnmv", "sd", "fQfQd">;
1162
1163////////////////////////////////////////////////////////////////////////////////
1164// Newly added Vector Extract for f64
1165def VEXT_A64 : WInst<"vext", "dddi", "dQdPlQPl">;
1166
1167////////////////////////////////////////////////////////////////////////////////
1168// Crypto
1169let ArchGuard = "__ARM_FEATURE_CRYPTO" in {
1170def AESE : SInst<"vaese", "ddd", "QUc">;
1171def AESD : SInst<"vaesd", "ddd", "QUc">;
1172def AESMC : SInst<"vaesmc", "dd", "QUc">;
1173def AESIMC : SInst<"vaesimc", "dd", "QUc">;
1174
1175def SHA1H : SInst<"vsha1h", "ss", "Ui">;
1176def SHA1SU1 : SInst<"vsha1su1", "ddd", "QUi">;
1177def SHA256SU0 : SInst<"vsha256su0", "ddd", "QUi">;
1178
1179def SHA1C : SInst<"vsha1c", "ddsd", "QUi">;
1180def SHA1P : SInst<"vsha1p", "ddsd", "QUi">;
1181def SHA1M : SInst<"vsha1m", "ddsd", "QUi">;
1182def SHA1SU0 : SInst<"vsha1su0", "dddd", "QUi">;
1183def SHA256H : SInst<"vsha256h", "dddd", "QUi">;
1184def SHA256H2 : SInst<"vsha256h2", "dddd", "QUi">;
1185def SHA256SU1 : SInst<"vsha256su1", "dddd", "QUi">;
1186}
1187
1188////////////////////////////////////////////////////////////////////////////////
1189// Float -> Int conversions with explicit rounding mode
1190
1191let ArchGuard = "__ARM_ARCH >= 8" in {
1192def FCVTNS_S32 : SInst<"vcvtn_s32", "xd", "fQf">;
1193def FCVTNU_S32 : SInst<"vcvtn_u32", "ud", "fQf">;
1194def FCVTPS_S32 : SInst<"vcvtp_s32", "xd", "fQf">;
1195def FCVTPU_S32 : SInst<"vcvtp_u32", "ud", "fQf">;
1196def FCVTMS_S32 : SInst<"vcvtm_s32", "xd", "fQf">;
1197def FCVTMU_S32 : SInst<"vcvtm_u32", "ud", "fQf">;
1198def FCVTAS_S32 : SInst<"vcvta_s32", "xd", "fQf">;
1199def FCVTAU_S32 : SInst<"vcvta_u32", "ud", "fQf">;
1200}
1201
1202let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__)" in {
1203def FCVTNS_S64 : SInst<"vcvtn_s64", "xd", "dQd">;
1204def FCVTNU_S64 : SInst<"vcvtn_u64", "ud", "dQd">;
1205def FCVTPS_S64 : SInst<"vcvtp_s64", "xd", "dQd">;
1206def FCVTPU_S64 : SInst<"vcvtp_u64", "ud", "dQd">;
1207def FCVTMS_S64 : SInst<"vcvtm_s64", "xd", "dQd">;
1208def FCVTMU_S64 : SInst<"vcvtm_u64", "ud", "dQd">;
1209def FCVTAS_S64 : SInst<"vcvta_s64", "xd", "dQd">;
1210def FCVTAU_S64 : SInst<"vcvta_u64", "ud", "dQd">;
1211}
1212
1213////////////////////////////////////////////////////////////////////////////////
1214// Round to Integral
1215
1216let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_DIRECTED_ROUNDING)" in {
1217def FRINTN_S32 : SInst<"vrndn", "dd", "fQf">;
1218def FRINTA_S32 : SInst<"vrnda", "dd", "fQf">;
1219def FRINTP_S32 : SInst<"vrndp", "dd", "fQf">;
1220def FRINTM_S32 : SInst<"vrndm", "dd", "fQf">;
1221def FRINTX_S32 : SInst<"vrndx", "dd", "fQf">;
1222def FRINTZ_S32 : SInst<"vrnd", "dd", "fQf">;
1223}
1224
1225let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__) && defined(__ARM_FEATURE_DIRECTED_ROUNDING)" in {
1226def FRINTN_S64 : SInst<"vrndn", "dd", "dQd">;
1227def FRINTA_S64 : SInst<"vrnda", "dd", "dQd">;
1228def FRINTP_S64 : SInst<"vrndp", "dd", "dQd">;
1229def FRINTM_S64 : SInst<"vrndm", "dd", "dQd">;
1230def FRINTX_S64 : SInst<"vrndx", "dd", "dQd">;
1231def FRINTZ_S64 : SInst<"vrnd", "dd", "dQd">;
1232def FRINTI_S64 : SInst<"vrndi", "dd", "fdQfQd">;
1233}
1234
1235////////////////////////////////////////////////////////////////////////////////
1236// MaxNum/MinNum Floating Point
1237
1238let ArchGuard = "__ARM_ARCH >= 8 && defined(__ARM_FEATURE_NUMERIC_MAXMIN)" in {
1239def FMAXNM_S32 : SInst<"vmaxnm", "ddd", "fQf">;
1240def FMINNM_S32 : SInst<"vminnm", "ddd", "fQf">;
1241}
1242
1243let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__) && defined(__ARM_FEATURE_NUMERIC_MAXMIN)" in {
1244def FMAXNM_S64 : SInst<"vmaxnm", "ddd", "dQd">;
1245def FMINNM_S64 : SInst<"vminnm", "ddd", "dQd">;
1246}
1247
1248////////////////////////////////////////////////////////////////////////////////
1249// Permutation
1250def VTRN1 : SOpInst<"vtrn1", "ddd",
1251                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_TRN1>;
1252def VZIP1 : SOpInst<"vzip1", "ddd",
1253                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_ZIP1>;
1254def VUZP1 : SOpInst<"vuzp1", "ddd",
1255                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_UZP1>;
1256def VTRN2 : SOpInst<"vtrn2", "ddd",
1257                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_TRN2>;
1258def VZIP2 : SOpInst<"vzip2", "ddd",
1259                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_ZIP2>;
1260def VUZP2 : SOpInst<"vuzp2", "ddd",
1261                    "csiUcUsUifPcPsQcQsQiQlQUcQUsQUiQUlQfQdQPcQPsQPl", OP_UZP2>;
1262
1263////////////////////////////////////////////////////////////////////////////////
1264// Table lookup
1265let InstName = "vtbl" in {
1266def VQTBL1_A64 : WInst<"vqtbl1", "djt",  "UccPcQUcQcQPc">;
1267def VQTBL2_A64 : WInst<"vqtbl2", "dBt",  "UccPcQUcQcQPc">;
1268def VQTBL3_A64 : WInst<"vqtbl3", "dCt",  "UccPcQUcQcQPc">;
1269def VQTBL4_A64 : WInst<"vqtbl4", "dDt",  "UccPcQUcQcQPc">;
1270}
1271let InstName = "vtbx" in {
1272def VQTBX1_A64 : WInst<"vqtbx1", "ddjt", "UccPcQUcQcQPc">;
1273def VQTBX2_A64 : WInst<"vqtbx2", "ddBt", "UccPcQUcQcQPc">;
1274def VQTBX3_A64 : WInst<"vqtbx3", "ddCt", "UccPcQUcQcQPc">;
1275def VQTBX4_A64 : WInst<"vqtbx4", "ddDt", "UccPcQUcQcQPc">;
1276}
1277
1278////////////////////////////////////////////////////////////////////////////////
1279// Vector reinterpret cast operations
1280
1281// NeonEmitter implicitly takes the cartesian product of the type string with
1282// itself during generation so, unlike all other intrinsics, this one should
1283// include *all* types, not just additional ones.
1284def VVREINTERPRET
1285  : NoTestOpInst<"vreinterpret", "dd",
1286       "csilUcUsUiUlhfdPcPsPlQcQsQiQlQUcQUsQUiQUlQhQfQdQPcQPsQPlQPk", OP_REINT> {
1287  let CartesianProductOfTypes = 1;
1288  let BigEndianSafe = 1;
1289  let ArchGuard = "__ARM_ARCH >= 8 && defined(__aarch64__)";
1290}
1291
1292////////////////////////////////////////////////////////////////////////////////
1293// Scalar Intrinsics
1294// Scalar Arithmetic
1295
1296// Scalar Addition
1297def SCALAR_ADD : SInst<"vadd", "sss",  "SlSUl">;
1298// Scalar  Saturating Add
1299def SCALAR_QADD   : SInst<"vqadd", "sss", "ScSsSiSlSUcSUsSUiSUl">;
1300
1301// Scalar Subtraction
1302def SCALAR_SUB : SInst<"vsub", "sss",  "SlSUl">;
1303// Scalar  Saturating Sub
1304def SCALAR_QSUB   : SInst<"vqsub", "sss", "ScSsSiSlSUcSUsSUiSUl">;
1305
1306let InstName = "vmov" in {
1307def VGET_HIGH_A64 : NoTestOpInst<"vget_high", "dk", "dPl", OP_HI>;
1308def VGET_LOW_A64  : NoTestOpInst<"vget_low", "dk", "dPl", OP_LO>;
1309}
1310
1311////////////////////////////////////////////////////////////////////////////////
1312// Scalar Shift
1313// Scalar Shift Left
1314def SCALAR_SHL: SInst<"vshl", "sss", "SlSUl">;
1315// Scalar Saturating Shift Left
1316def SCALAR_QSHL: SInst<"vqshl", "sss", "ScSsSiSlSUcSUsSUiSUl">;
1317// Scalar Saturating Rounding Shift Left
1318def SCALAR_QRSHL: SInst<"vqrshl", "sss", "ScSsSiSlSUcSUsSUiSUl">;
1319// Scalar Shift Rouding Left
1320def SCALAR_RSHL: SInst<"vrshl", "sss", "SlSUl">;
1321
1322////////////////////////////////////////////////////////////////////////////////
1323// Scalar Shift (Immediate)
1324let isScalarShift = 1 in {
1325// Signed/Unsigned Shift Right (Immediate)
1326def SCALAR_SSHR_N: SInst<"vshr_n", "ssi", "SlSUl">;
1327// Signed/Unsigned Rounding Shift Right (Immediate)
1328def SCALAR_SRSHR_N: SInst<"vrshr_n", "ssi", "SlSUl">;
1329
1330// Signed/Unsigned Shift Right and Accumulate (Immediate)
1331def SCALAR_SSRA_N: SInst<"vsra_n", "sssi", "SlSUl">;
1332// Signed/Unsigned Rounding Shift Right and Accumulate (Immediate)
1333def SCALAR_SRSRA_N: SInst<"vrsra_n", "sssi", "SlSUl">;
1334
1335// Shift Left (Immediate)
1336def SCALAR_SHL_N: SInst<"vshl_n", "ssi", "SlSUl">;
1337// Signed/Unsigned Saturating Shift Left (Immediate)
1338def SCALAR_SQSHL_N: SInst<"vqshl_n", "ssi", "ScSsSiSlSUcSUsSUiSUl">;
1339// Signed Saturating Shift Left Unsigned (Immediate)
1340def SCALAR_SQSHLU_N: SInst<"vqshlu_n", "ssi", "ScSsSiSl">;
1341
1342// Shift Right And Insert (Immediate)
1343def SCALAR_SRI_N: SInst<"vsri_n", "sssi", "SlSUl">;
1344// Shift Left And Insert (Immediate)
1345def SCALAR_SLI_N: SInst<"vsli_n", "sssi", "SlSUl">;
1346
1347let isScalarNarrowShift = 1 in {
1348  // Signed/Unsigned Saturating Shift Right Narrow (Immediate)
1349  def SCALAR_SQSHRN_N: SInst<"vqshrn_n", "zsi", "SsSiSlSUsSUiSUl">;
1350  // Signed/Unsigned Saturating Rounded Shift Right Narrow (Immediate)
1351  def SCALAR_SQRSHRN_N: SInst<"vqrshrn_n", "zsi", "SsSiSlSUsSUiSUl">;
1352  // Signed Saturating Shift Right Unsigned Narrow (Immediate)
1353  def SCALAR_SQSHRUN_N: SInst<"vqshrun_n", "zsi", "SsSiSl">;
1354  // Signed Saturating Rounded Shift Right Unsigned Narrow (Immediate)
1355  def SCALAR_SQRSHRUN_N: SInst<"vqrshrun_n", "zsi", "SsSiSl">;
1356}
1357
1358////////////////////////////////////////////////////////////////////////////////
1359// Scalar Signed/Unsigned Fixed-point Convert To Floating-Point (Immediate)
1360def SCALAR_SCVTF_N_F32: SInst<"vcvt_n_f32", "ysi", "SiSUi">;
1361def SCALAR_SCVTF_N_F64: SInst<"vcvt_n_f64", "osi", "SlSUl">;
1362
1363////////////////////////////////////////////////////////////////////////////////
1364// Scalar Floating-point Convert To Signed/Unsigned Fixed-point (Immediate)
1365def SCALAR_FCVTZS_N_S32 : SInst<"vcvt_n_s32", "$si", "Sf">;
1366def SCALAR_FCVTZU_N_U32 : SInst<"vcvt_n_u32", "bsi", "Sf">;
1367def SCALAR_FCVTZS_N_S64 : SInst<"vcvt_n_s64", "$si", "Sd">;
1368def SCALAR_FCVTZU_N_U64 : SInst<"vcvt_n_u64", "bsi", "Sd">;
1369}
1370
1371////////////////////////////////////////////////////////////////////////////////
1372// Scalar Reduce Pairwise Addition (Scalar and Floating Point)
1373def SCALAR_ADDP  : SInst<"vpadd", "sd", "SfSHlSHdSHUl">;
1374
1375////////////////////////////////////////////////////////////////////////////////
1376// Scalar Reduce Floating Point Pairwise Max/Min
1377def SCALAR_FMAXP : SInst<"vpmax", "sd", "SfSQd">;
1378
1379def SCALAR_FMINP : SInst<"vpmin", "sd", "SfSQd">;
1380
1381////////////////////////////////////////////////////////////////////////////////
1382// Scalar Reduce Floating Point Pairwise maxNum/minNum
1383def SCALAR_FMAXNMP : SInst<"vpmaxnm", "sd", "SfSQd">;
1384def SCALAR_FMINNMP : SInst<"vpminnm", "sd", "SfSQd">;
1385
1386////////////////////////////////////////////////////////////////////////////////
1387// Scalar Integer Saturating Doubling Multiply Half High
1388def SCALAR_SQDMULH : SInst<"vqdmulh", "sss", "SsSi">;
1389
1390////////////////////////////////////////////////////////////////////////////////
1391// Scalar Integer Saturating Rounding Doubling Multiply Half High
1392def SCALAR_SQRDMULH : SInst<"vqrdmulh", "sss", "SsSi">;
1393
1394////////////////////////////////////////////////////////////////////////////////
1395// Scalar Floating-point Multiply Extended
1396def SCALAR_FMULX : IInst<"vmulx", "sss", "SfSd">;
1397
1398////////////////////////////////////////////////////////////////////////////////
1399// Scalar Floating-point Reciprocal Step
1400def SCALAR_FRECPS : IInst<"vrecps", "sss", "SfSd">;
1401
1402////////////////////////////////////////////////////////////////////////////////
1403// Scalar Floating-point Reciprocal Square Root Step
1404def SCALAR_FRSQRTS : IInst<"vrsqrts", "sss", "SfSd">;
1405
1406////////////////////////////////////////////////////////////////////////////////
1407// Scalar Signed Integer Convert To Floating-point
1408def SCALAR_SCVTFS : SInst<"vcvt_f32", "ys", "Si">;
1409def SCALAR_SCVTFD : SInst<"vcvt_f64", "os", "Sl">;
1410
1411////////////////////////////////////////////////////////////////////////////////
1412// Scalar Unsigned Integer Convert To Floating-point
1413def SCALAR_UCVTFS : SInst<"vcvt_f32", "ys", "SUi">;
1414def SCALAR_UCVTFD : SInst<"vcvt_f64", "os", "SUl">;
1415
1416////////////////////////////////////////////////////////////////////////////////
1417// Scalar Floating-point Converts
1418def SCALAR_FCVTXN  : IInst<"vcvtx_f32", "ys", "Sd">;
1419def SCALAR_FCVTNSS : SInst<"vcvtn_s32", "$s", "Sf">;
1420def SCALAR_FCVTNUS : SInst<"vcvtn_u32", "bs", "Sf">;
1421def SCALAR_FCVTNSD : SInst<"vcvtn_s64", "$s", "Sd">;
1422def SCALAR_FCVTNUD : SInst<"vcvtn_u64", "bs", "Sd">;
1423def SCALAR_FCVTMSS : SInst<"vcvtm_s32", "$s", "Sf">;
1424def SCALAR_FCVTMUS : SInst<"vcvtm_u32", "bs", "Sf">;
1425def SCALAR_FCVTMSD : SInst<"vcvtm_s64", "$s", "Sd">;
1426def SCALAR_FCVTMUD : SInst<"vcvtm_u64", "bs", "Sd">;
1427def SCALAR_FCVTASS : SInst<"vcvta_s32", "$s", "Sf">;
1428def SCALAR_FCVTAUS : SInst<"vcvta_u32", "bs", "Sf">;
1429def SCALAR_FCVTASD : SInst<"vcvta_s64", "$s", "Sd">;
1430def SCALAR_FCVTAUD : SInst<"vcvta_u64", "bs", "Sd">;
1431def SCALAR_FCVTPSS : SInst<"vcvtp_s32", "$s", "Sf">;
1432def SCALAR_FCVTPUS : SInst<"vcvtp_u32", "bs", "Sf">;
1433def SCALAR_FCVTPSD : SInst<"vcvtp_s64", "$s", "Sd">;
1434def SCALAR_FCVTPUD : SInst<"vcvtp_u64", "bs", "Sd">;
1435def SCALAR_FCVTZSS : SInst<"vcvt_s32", "$s", "Sf">;
1436def SCALAR_FCVTZUS : SInst<"vcvt_u32", "bs", "Sf">;
1437def SCALAR_FCVTZSD : SInst<"vcvt_s64", "$s", "Sd">;
1438def SCALAR_FCVTZUD : SInst<"vcvt_u64", "bs", "Sd">;
1439
1440////////////////////////////////////////////////////////////////////////////////
1441// Scalar Floating-point Reciprocal Estimate
1442def SCALAR_FRECPE : IInst<"vrecpe", "ss", "SfSd">;
1443
1444////////////////////////////////////////////////////////////////////////////////
1445// Scalar Floating-point Reciprocal Exponent
1446def SCALAR_FRECPX : IInst<"vrecpx", "ss", "SfSd">;
1447
1448////////////////////////////////////////////////////////////////////////////////
1449// Scalar Floating-point Reciprocal Square Root Estimate
1450def SCALAR_FRSQRTE : IInst<"vrsqrte", "ss", "SfSd">;
1451
1452////////////////////////////////////////////////////////////////////////////////
1453// Scalar Integer Comparison
1454def SCALAR_CMEQ : SInst<"vceq", "sss", "SlSUl">;
1455def SCALAR_CMEQZ : SInst<"vceqz", "ss", "SlSUl">;
1456def SCALAR_CMGE : SInst<"vcge", "sss", "Sl">;
1457def SCALAR_CMGEZ : SInst<"vcgez", "ss", "Sl">;
1458def SCALAR_CMHS : SInst<"vcge", "sss", "SUl">;
1459def SCALAR_CMLE : SInst<"vcle", "sss", "SlSUl">;
1460def SCALAR_CMLEZ : SInst<"vclez", "ss", "Sl">;
1461def SCALAR_CMLT : SInst<"vclt", "sss", "SlSUl">;
1462def SCALAR_CMLTZ : SInst<"vcltz", "ss", "Sl">;
1463def SCALAR_CMGT : SInst<"vcgt", "sss", "Sl">;
1464def SCALAR_CMGTZ : SInst<"vcgtz", "ss", "Sl">;
1465def SCALAR_CMHI : SInst<"vcgt", "sss", "SUl">;
1466def SCALAR_CMTST : SInst<"vtst", "sss", "SlSUl">;
1467
1468////////////////////////////////////////////////////////////////////////////////
1469// Scalar Floating-point Comparison
1470def SCALAR_FCMEQ : IInst<"vceq", "bss", "SfSd">;
1471def SCALAR_FCMEQZ : IInst<"vceqz", "bs", "SfSd">;
1472def SCALAR_FCMGE : IInst<"vcge", "bss", "SfSd">;
1473def SCALAR_FCMGEZ : IInst<"vcgez", "bs", "SfSd">;
1474def SCALAR_FCMGT : IInst<"vcgt", "bss", "SfSd">;
1475def SCALAR_FCMGTZ : IInst<"vcgtz", "bs", "SfSd">;
1476def SCALAR_FCMLE : IInst<"vcle", "bss", "SfSd">;
1477def SCALAR_FCMLEZ : IInst<"vclez", "bs", "SfSd">;
1478def SCALAR_FCMLT : IInst<"vclt", "bss", "SfSd">;
1479def SCALAR_FCMLTZ : IInst<"vcltz", "bs", "SfSd">;
1480
1481////////////////////////////////////////////////////////////////////////////////
1482// Scalar Floating-point Absolute Compare Mask Greater Than Or Equal
1483def SCALAR_FACGE : IInst<"vcage", "bss", "SfSd">;
1484def SCALAR_FACLE : IInst<"vcale", "bss", "SfSd">;
1485
1486////////////////////////////////////////////////////////////////////////////////
1487// Scalar Floating-point Absolute Compare Mask Greater Than
1488def SCALAR_FACGT : IInst<"vcagt", "bss", "SfSd">;
1489def SCALAR_FACLT : IInst<"vcalt", "bss", "SfSd">;
1490
1491////////////////////////////////////////////////////////////////////////////////
1492// Scalar Absolute Value
1493def SCALAR_ABS : SInst<"vabs", "ss", "Sl">;
1494
1495////////////////////////////////////////////////////////////////////////////////
1496// Scalar Absolute Difference
1497def SCALAR_ABD : IInst<"vabd", "sss", "SfSd">;
1498
1499////////////////////////////////////////////////////////////////////////////////
1500// Scalar Signed Saturating Absolute Value
1501def SCALAR_SQABS : SInst<"vqabs", "ss", "ScSsSiSl">;
1502
1503////////////////////////////////////////////////////////////////////////////////
1504// Scalar Negate
1505def SCALAR_NEG : SInst<"vneg", "ss", "Sl">;
1506
1507////////////////////////////////////////////////////////////////////////////////
1508// Scalar Signed Saturating Negate
1509def SCALAR_SQNEG : SInst<"vqneg", "ss", "ScSsSiSl">;
1510
1511////////////////////////////////////////////////////////////////////////////////
1512// Scalar Signed Saturating Accumulated of Unsigned Value
1513def SCALAR_SUQADD : SInst<"vuqadd", "sss", "ScSsSiSl">;
1514
1515////////////////////////////////////////////////////////////////////////////////
1516// Scalar Unsigned Saturating Accumulated of Signed Value
1517def SCALAR_USQADD : SInst<"vsqadd", "sss", "SUcSUsSUiSUl">;
1518
1519////////////////////////////////////////////////////////////////////////////////
1520// Signed Saturating Doubling Multiply-Add Long
1521def SCALAR_SQDMLAL : SInst<"vqdmlal", "rrss", "SsSi">;
1522
1523////////////////////////////////////////////////////////////////////////////////
1524// Signed Saturating Doubling Multiply-Subtract Long
1525def SCALAR_SQDMLSL : SInst<"vqdmlsl", "rrss", "SsSi">;
1526
1527////////////////////////////////////////////////////////////////////////////////
1528// Signed Saturating Doubling Multiply Long
1529def SCALAR_SQDMULL : SInst<"vqdmull", "rss", "SsSi">;
1530
1531////////////////////////////////////////////////////////////////////////////////
1532// Scalar Signed Saturating Extract Unsigned Narrow
1533def SCALAR_SQXTUN : SInst<"vqmovun", "zs", "SsSiSl">;
1534
1535////////////////////////////////////////////////////////////////////////////////
1536// Scalar Signed Saturating Extract Narrow
1537def SCALAR_SQXTN : SInst<"vqmovn", "zs", "SsSiSl">;
1538
1539////////////////////////////////////////////////////////////////////////////////
1540// Scalar Unsigned Saturating Extract Narrow
1541def SCALAR_UQXTN : SInst<"vqmovn", "zs", "SUsSUiSUl">;
1542
1543// Scalar Floating Point  multiply (scalar, by element)
1544def SCALAR_FMUL_LANE : IOpInst<"vmul_lane", "ssdi", "SfSd", OP_SCALAR_MUL_LN>;
1545def SCALAR_FMUL_LANEQ : IOpInst<"vmul_laneq", "ssji", "SfSd", OP_SCALAR_MUL_LN>;
1546
1547// Scalar Floating Point  multiply extended (scalar, by element)
1548def SCALAR_FMULX_LANE : IOpInst<"vmulx_lane", "ssdi", "SfSd", OP_SCALAR_MULX_LN>;
1549def SCALAR_FMULX_LANEQ : IOpInst<"vmulx_laneq", "ssji", "SfSd", OP_SCALAR_MULX_LN>;
1550
1551def SCALAR_VMUL_N : IInst<"vmul_n", "dds", "d">;
1552
1553// VMUL_LANE_A64 d type implemented using scalar mul lane
1554def SCALAR_VMUL_LANE : IInst<"vmul_lane", "ddgi", "d">;
1555
1556// VMUL_LANEQ d type implemented using scalar mul lane
1557def SCALAR_VMUL_LANEQ   : IInst<"vmul_laneq", "ddji", "d"> {
1558  let isLaneQ = 1;
1559}
1560
1561// VMULX_LANE d type implemented using scalar vmulx_lane
1562def SCALAR_VMULX_LANE : IOpInst<"vmulx_lane", "ddgi", "d", OP_SCALAR_VMULX_LN>;
1563
1564// VMULX_LANEQ d type implemented using scalar vmulx_laneq
1565def SCALAR_VMULX_LANEQ : IOpInst<"vmulx_laneq", "ddji", "d", OP_SCALAR_VMULX_LNQ>;
1566
1567// Scalar Floating Point fused multiply-add (scalar, by element)
1568def SCALAR_FMLA_LANE : IInst<"vfma_lane", "sssdi", "SfSd">;
1569def SCALAR_FMLA_LANEQ : IInst<"vfma_laneq", "sssji", "SfSd">;
1570
1571// Scalar Floating Point fused multiply-subtract (scalar, by element)
1572def SCALAR_FMLS_LANE : IOpInst<"vfms_lane", "sssdi", "SfSd", OP_FMS_LN>;
1573def SCALAR_FMLS_LANEQ : IOpInst<"vfms_laneq", "sssji", "SfSd", OP_FMS_LNQ>;
1574
1575// Signed Saturating Doubling Multiply Long (scalar by element)
1576def SCALAR_SQDMULL_LANE : SOpInst<"vqdmull_lane", "rsdi", "SsSi", OP_SCALAR_QDMULL_LN>;
1577def SCALAR_SQDMULL_LANEQ : SOpInst<"vqdmull_laneq", "rsji", "SsSi", OP_SCALAR_QDMULL_LN>;
1578
1579// Signed Saturating Doubling Multiply-Add Long (scalar by element)
1580def SCALAR_SQDMLAL_LANE : SInst<"vqdmlal_lane", "rrsdi", "SsSi">;
1581def SCALAR_SQDMLAL_LANEQ : SInst<"vqdmlal_laneq", "rrsji", "SsSi">;
1582
1583// Signed Saturating Doubling Multiply-Subtract Long (scalar by element)
1584def SCALAR_SQDMLS_LANE : SInst<"vqdmlsl_lane", "rrsdi", "SsSi">;
1585def SCALAR_SQDMLS_LANEQ : SInst<"vqdmlsl_laneq", "rrsji", "SsSi">;
1586
1587// Scalar Integer Saturating Doubling Multiply Half High (scalar by element)
1588def SCALAR_SQDMULH_LANE : SOpInst<"vqdmulh_lane", "ssdi", "SsSi", OP_SCALAR_QDMULH_LN>;
1589def SCALAR_SQDMULH_LANEQ : SOpInst<"vqdmulh_laneq", "ssji", "SsSi", OP_SCALAR_QDMULH_LN>;
1590
1591// Scalar Integer Saturating Rounding Doubling Multiply Half High
1592def SCALAR_SQRDMULH_LANE : SOpInst<"vqrdmulh_lane", "ssdi", "SsSi", OP_SCALAR_QRDMULH_LN>;
1593def SCALAR_SQRDMULH_LANEQ : SOpInst<"vqrdmulh_laneq", "ssji", "SsSi", OP_SCALAR_QRDMULH_LN>;
1594
1595def SCALAR_VDUP_LANE : IInst<"vdup_lane", "sdi", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
1596def SCALAR_VDUP_LANEQ : IInst<"vdup_laneq", "sji", "ScSsSiSlSfSdSUcSUsSUiSUlSPcSPs">;
1597
1598// FIXME: Rename so it is obvious this only applies to halfs.
1599def SCALAR_HALF_GET_LANE : IOpInst<"vget_lane", "sdi", "h", OP_SCALAR_HALF_GET_LN>;
1600def SCALAR_HALF_SET_LANE : IOpInst<"vset_lane", "dsdi", "h", OP_SCALAR_HALF_SET_LN>;
1601def SCALAR_HALF_GET_LANEQ : IOpInst<"vget_lane", "sdi", "Qh", OP_SCALAR_HALF_GET_LNQ>;
1602def SCALAR_HALF_SET_LANEQ : IOpInst<"vset_lane", "dsdi", "Qh", OP_SCALAR_HALF_SET_LNQ>;
1603}
1604