xref: /freebsd/contrib/llvm-project/llvm/include/llvm/IR/IntrinsicsAArch64.td (revision 4b50c451720d8b427757a6da1dd2bb4c52cd9e35)
1//===- IntrinsicsAARCH64.td - Defines AARCH64 intrinsics ---*- tablegen -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file defines all of the AARCH64-specific intrinsics.
10//
11//===----------------------------------------------------------------------===//
12
13let TargetPrefix = "aarch64" in {
14
15def int_aarch64_ldxr : Intrinsic<[llvm_i64_ty], [llvm_anyptr_ty]>;
16def int_aarch64_ldaxr : Intrinsic<[llvm_i64_ty], [llvm_anyptr_ty]>;
17def int_aarch64_stxr : Intrinsic<[llvm_i32_ty], [llvm_i64_ty, llvm_anyptr_ty]>;
18def int_aarch64_stlxr : Intrinsic<[llvm_i32_ty], [llvm_i64_ty, llvm_anyptr_ty]>;
19
20def int_aarch64_ldxp : Intrinsic<[llvm_i64_ty, llvm_i64_ty], [llvm_ptr_ty]>;
21def int_aarch64_ldaxp : Intrinsic<[llvm_i64_ty, llvm_i64_ty], [llvm_ptr_ty]>;
22def int_aarch64_stxp : Intrinsic<[llvm_i32_ty],
23                               [llvm_i64_ty, llvm_i64_ty, llvm_ptr_ty]>;
24def int_aarch64_stlxp : Intrinsic<[llvm_i32_ty],
25                                [llvm_i64_ty, llvm_i64_ty, llvm_ptr_ty]>;
26
27def int_aarch64_clrex : Intrinsic<[]>;
28
29def int_aarch64_sdiv : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
30                                LLVMMatchType<0>], [IntrNoMem]>;
31def int_aarch64_udiv : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>,
32                                LLVMMatchType<0>], [IntrNoMem]>;
33
34def int_aarch64_fjcvtzs : Intrinsic<[llvm_i32_ty], [llvm_double_ty], [IntrNoMem]>;
35
36//===----------------------------------------------------------------------===//
37// HINT
38
39def int_aarch64_hint : Intrinsic<[], [llvm_i32_ty]>;
40
41//===----------------------------------------------------------------------===//
42// Data Barrier Instructions
43
44def int_aarch64_dmb : GCCBuiltin<"__builtin_arm_dmb">, MSBuiltin<"__dmb">, Intrinsic<[], [llvm_i32_ty]>;
45def int_aarch64_dsb : GCCBuiltin<"__builtin_arm_dsb">, MSBuiltin<"__dsb">, Intrinsic<[], [llvm_i32_ty]>;
46def int_aarch64_isb : GCCBuiltin<"__builtin_arm_isb">, MSBuiltin<"__isb">, Intrinsic<[], [llvm_i32_ty]>;
47
48// A space-consuming intrinsic primarily for testing block and jump table
49// placements. The first argument is the number of bytes this "instruction"
50// takes up, the second and return value are essentially chains, used to force
51// ordering during ISel.
52def int_aarch64_space : Intrinsic<[llvm_i64_ty], [llvm_i32_ty, llvm_i64_ty], []>;
53
54}
55
56//===----------------------------------------------------------------------===//
57// Advanced SIMD (NEON)
58
59let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
60  class AdvSIMD_2Scalar_Float_Intrinsic
61    : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>],
62                [IntrNoMem]>;
63
64  class AdvSIMD_FPToIntRounding_Intrinsic
65    : Intrinsic<[llvm_anyint_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
66
67  class AdvSIMD_1IntArg_Intrinsic
68    : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>], [IntrNoMem]>;
69  class AdvSIMD_1FloatArg_Intrinsic
70    : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
71  class AdvSIMD_1VectorArg_Intrinsic
72    : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>], [IntrNoMem]>;
73  class AdvSIMD_1VectorArg_Expand_Intrinsic
74    : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
75  class AdvSIMD_1VectorArg_Long_Intrinsic
76    : Intrinsic<[llvm_anyvector_ty], [LLVMTruncatedType<0>], [IntrNoMem]>;
77  class AdvSIMD_1IntArg_Narrow_Intrinsic
78    : Intrinsic<[llvm_anyint_ty], [llvm_anyint_ty], [IntrNoMem]>;
79  class AdvSIMD_1VectorArg_Narrow_Intrinsic
80    : Intrinsic<[llvm_anyint_ty], [LLVMExtendedType<0>], [IntrNoMem]>;
81  class AdvSIMD_1VectorArg_Int_Across_Intrinsic
82    : Intrinsic<[llvm_anyint_ty], [llvm_anyvector_ty], [IntrNoMem]>;
83  class AdvSIMD_1VectorArg_Float_Across_Intrinsic
84    : Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>;
85
86  class AdvSIMD_2IntArg_Intrinsic
87    : Intrinsic<[llvm_anyint_ty], [LLVMMatchType<0>, LLVMMatchType<0>],
88                [IntrNoMem]>;
89  class AdvSIMD_2FloatArg_Intrinsic
90    : Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>],
91                [IntrNoMem]>;
92  class AdvSIMD_2VectorArg_Intrinsic
93    : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>, LLVMMatchType<0>],
94                [IntrNoMem]>;
95  class AdvSIMD_2VectorArg_Compare_Intrinsic
96    : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty, LLVMMatchType<1>],
97                [IntrNoMem]>;
98  class AdvSIMD_2Arg_FloatCompare_Intrinsic
99    : Intrinsic<[llvm_anyint_ty], [llvm_anyfloat_ty, LLVMMatchType<1>],
100                [IntrNoMem]>;
101  class AdvSIMD_2VectorArg_Long_Intrinsic
102    : Intrinsic<[llvm_anyvector_ty],
103                [LLVMTruncatedType<0>, LLVMTruncatedType<0>],
104                [IntrNoMem]>;
105  class AdvSIMD_2VectorArg_Wide_Intrinsic
106    : Intrinsic<[llvm_anyvector_ty],
107                [LLVMMatchType<0>, LLVMTruncatedType<0>],
108                [IntrNoMem]>;
109  class AdvSIMD_2VectorArg_Narrow_Intrinsic
110    : Intrinsic<[llvm_anyvector_ty],
111                [LLVMExtendedType<0>, LLVMExtendedType<0>],
112                [IntrNoMem]>;
113  class AdvSIMD_2Arg_Scalar_Narrow_Intrinsic
114    : Intrinsic<[llvm_anyint_ty],
115                [LLVMExtendedType<0>, llvm_i32_ty],
116                [IntrNoMem]>;
117  class AdvSIMD_2VectorArg_Scalar_Expand_BySize_Intrinsic
118    : Intrinsic<[llvm_anyvector_ty],
119                [llvm_anyvector_ty],
120                [IntrNoMem]>;
121  class AdvSIMD_2VectorArg_Scalar_Wide_BySize_Intrinsic
122    : Intrinsic<[llvm_anyvector_ty],
123                [LLVMTruncatedType<0>],
124                [IntrNoMem]>;
125  class AdvSIMD_2VectorArg_Scalar_Wide_Intrinsic
126    : Intrinsic<[llvm_anyvector_ty],
127                [LLVMTruncatedType<0>, llvm_i32_ty],
128                [IntrNoMem]>;
129  class AdvSIMD_2VectorArg_Tied_Narrow_Intrinsic
130    : Intrinsic<[llvm_anyvector_ty],
131                [LLVMHalfElementsVectorType<0>, llvm_anyvector_ty],
132                [IntrNoMem]>;
133
134  class AdvSIMD_3VectorArg_Intrinsic
135      : Intrinsic<[llvm_anyvector_ty],
136               [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
137               [IntrNoMem]>;
138  class AdvSIMD_3VectorArg_Scalar_Intrinsic
139      : Intrinsic<[llvm_anyvector_ty],
140               [LLVMMatchType<0>, LLVMMatchType<0>, llvm_i32_ty],
141               [IntrNoMem]>;
142  class AdvSIMD_3VectorArg_Tied_Narrow_Intrinsic
143      : Intrinsic<[llvm_anyvector_ty],
144               [LLVMHalfElementsVectorType<0>, llvm_anyvector_ty,
145                LLVMMatchType<1>], [IntrNoMem]>;
146  class AdvSIMD_3VectorArg_Scalar_Tied_Narrow_Intrinsic
147    : Intrinsic<[llvm_anyvector_ty],
148                [LLVMHalfElementsVectorType<0>, llvm_anyvector_ty, llvm_i32_ty],
149                [IntrNoMem]>;
150  class AdvSIMD_CvtFxToFP_Intrinsic
151    : Intrinsic<[llvm_anyfloat_ty], [llvm_anyint_ty, llvm_i32_ty],
152                [IntrNoMem]>;
153  class AdvSIMD_CvtFPToFx_Intrinsic
154    : Intrinsic<[llvm_anyint_ty], [llvm_anyfloat_ty, llvm_i32_ty],
155                [IntrNoMem]>;
156
157  class AdvSIMD_1Arg_Intrinsic
158    : Intrinsic<[llvm_any_ty], [LLVMMatchType<0>], [IntrNoMem]>;
159
160  class AdvSIMD_Dot_Intrinsic
161    : Intrinsic<[llvm_anyvector_ty],
162                [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>],
163                [IntrNoMem]>;
164
165  class AdvSIMD_FP16FML_Intrinsic
166    : Intrinsic<[llvm_anyvector_ty],
167                [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>],
168                [IntrNoMem]>;
169}
170
171// Arithmetic ops
172
173let TargetPrefix = "aarch64", IntrProperties = [IntrNoMem] in {
174  // Vector Add Across Lanes
175  def int_aarch64_neon_saddv : AdvSIMD_1VectorArg_Int_Across_Intrinsic;
176  def int_aarch64_neon_uaddv : AdvSIMD_1VectorArg_Int_Across_Intrinsic;
177  def int_aarch64_neon_faddv : AdvSIMD_1VectorArg_Float_Across_Intrinsic;
178
179  // Vector Long Add Across Lanes
180  def int_aarch64_neon_saddlv : AdvSIMD_1VectorArg_Int_Across_Intrinsic;
181  def int_aarch64_neon_uaddlv : AdvSIMD_1VectorArg_Int_Across_Intrinsic;
182
183  // Vector Halving Add
184  def int_aarch64_neon_shadd : AdvSIMD_2VectorArg_Intrinsic;
185  def int_aarch64_neon_uhadd : AdvSIMD_2VectorArg_Intrinsic;
186
187  // Vector Rounding Halving Add
188  def int_aarch64_neon_srhadd : AdvSIMD_2VectorArg_Intrinsic;
189  def int_aarch64_neon_urhadd : AdvSIMD_2VectorArg_Intrinsic;
190
191  // Vector Saturating Add
192  def int_aarch64_neon_sqadd : AdvSIMD_2IntArg_Intrinsic;
193  def int_aarch64_neon_suqadd : AdvSIMD_2IntArg_Intrinsic;
194  def int_aarch64_neon_usqadd : AdvSIMD_2IntArg_Intrinsic;
195  def int_aarch64_neon_uqadd : AdvSIMD_2IntArg_Intrinsic;
196
197  // Vector Add High-Half
198  // FIXME: this is a legacy intrinsic for aarch64_simd.h. Remove it when that
199  // header is no longer supported.
200  def int_aarch64_neon_addhn : AdvSIMD_2VectorArg_Narrow_Intrinsic;
201
202  // Vector Rounding Add High-Half
203  def int_aarch64_neon_raddhn : AdvSIMD_2VectorArg_Narrow_Intrinsic;
204
205  // Vector Saturating Doubling Multiply High
206  def int_aarch64_neon_sqdmulh : AdvSIMD_2IntArg_Intrinsic;
207
208  // Vector Saturating Rounding Doubling Multiply High
209  def int_aarch64_neon_sqrdmulh : AdvSIMD_2IntArg_Intrinsic;
210
211  // Vector Polynominal Multiply
212  def int_aarch64_neon_pmul : AdvSIMD_2VectorArg_Intrinsic;
213
214  // Vector Long Multiply
215  def int_aarch64_neon_smull : AdvSIMD_2VectorArg_Long_Intrinsic;
216  def int_aarch64_neon_umull : AdvSIMD_2VectorArg_Long_Intrinsic;
217  def int_aarch64_neon_pmull : AdvSIMD_2VectorArg_Long_Intrinsic;
218
219  // 64-bit polynomial multiply really returns an i128, which is not legal. Fake
220  // it with a v16i8.
221  def int_aarch64_neon_pmull64 :
222        Intrinsic<[llvm_v16i8_ty], [llvm_i64_ty, llvm_i64_ty], [IntrNoMem]>;
223
224  // Vector Extending Multiply
225  def int_aarch64_neon_fmulx : AdvSIMD_2FloatArg_Intrinsic {
226    let IntrProperties = [IntrNoMem, Commutative];
227  }
228
229  // Vector Saturating Doubling Long Multiply
230  def int_aarch64_neon_sqdmull : AdvSIMD_2VectorArg_Long_Intrinsic;
231  def int_aarch64_neon_sqdmulls_scalar
232    : Intrinsic<[llvm_i64_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
233
234  // Vector Halving Subtract
235  def int_aarch64_neon_shsub : AdvSIMD_2VectorArg_Intrinsic;
236  def int_aarch64_neon_uhsub : AdvSIMD_2VectorArg_Intrinsic;
237
238  // Vector Saturating Subtract
239  def int_aarch64_neon_sqsub : AdvSIMD_2IntArg_Intrinsic;
240  def int_aarch64_neon_uqsub : AdvSIMD_2IntArg_Intrinsic;
241
242  // Vector Subtract High-Half
243  // FIXME: this is a legacy intrinsic for aarch64_simd.h. Remove it when that
244  // header is no longer supported.
245  def int_aarch64_neon_subhn : AdvSIMD_2VectorArg_Narrow_Intrinsic;
246
247  // Vector Rounding Subtract High-Half
248  def int_aarch64_neon_rsubhn : AdvSIMD_2VectorArg_Narrow_Intrinsic;
249
250  // Vector Compare Absolute Greater-than-or-equal
251  def int_aarch64_neon_facge : AdvSIMD_2Arg_FloatCompare_Intrinsic;
252
253  // Vector Compare Absolute Greater-than
254  def int_aarch64_neon_facgt : AdvSIMD_2Arg_FloatCompare_Intrinsic;
255
256  // Vector Absolute Difference
257  def int_aarch64_neon_sabd : AdvSIMD_2VectorArg_Intrinsic;
258  def int_aarch64_neon_uabd : AdvSIMD_2VectorArg_Intrinsic;
259  def int_aarch64_neon_fabd : AdvSIMD_2VectorArg_Intrinsic;
260
261  // Scalar Absolute Difference
262  def int_aarch64_sisd_fabd : AdvSIMD_2Scalar_Float_Intrinsic;
263
264  // Vector Max
265  def int_aarch64_neon_smax : AdvSIMD_2VectorArg_Intrinsic;
266  def int_aarch64_neon_umax : AdvSIMD_2VectorArg_Intrinsic;
267  def int_aarch64_neon_fmax : AdvSIMD_2FloatArg_Intrinsic;
268  def int_aarch64_neon_fmaxnmp : AdvSIMD_2VectorArg_Intrinsic;
269
270  // Vector Max Across Lanes
271  def int_aarch64_neon_smaxv : AdvSIMD_1VectorArg_Int_Across_Intrinsic;
272  def int_aarch64_neon_umaxv : AdvSIMD_1VectorArg_Int_Across_Intrinsic;
273  def int_aarch64_neon_fmaxv : AdvSIMD_1VectorArg_Float_Across_Intrinsic;
274  def int_aarch64_neon_fmaxnmv : AdvSIMD_1VectorArg_Float_Across_Intrinsic;
275
276  // Vector Min
277  def int_aarch64_neon_smin : AdvSIMD_2VectorArg_Intrinsic;
278  def int_aarch64_neon_umin : AdvSIMD_2VectorArg_Intrinsic;
279  def int_aarch64_neon_fmin : AdvSIMD_2FloatArg_Intrinsic;
280  def int_aarch64_neon_fminnmp : AdvSIMD_2VectorArg_Intrinsic;
281
282  // Vector Min/Max Number
283  def int_aarch64_neon_fminnm : AdvSIMD_2FloatArg_Intrinsic;
284  def int_aarch64_neon_fmaxnm : AdvSIMD_2FloatArg_Intrinsic;
285
286  // Vector Min Across Lanes
287  def int_aarch64_neon_sminv : AdvSIMD_1VectorArg_Int_Across_Intrinsic;
288  def int_aarch64_neon_uminv : AdvSIMD_1VectorArg_Int_Across_Intrinsic;
289  def int_aarch64_neon_fminv : AdvSIMD_1VectorArg_Float_Across_Intrinsic;
290  def int_aarch64_neon_fminnmv : AdvSIMD_1VectorArg_Float_Across_Intrinsic;
291
292  // Pairwise Add
293  def int_aarch64_neon_addp : AdvSIMD_2VectorArg_Intrinsic;
294  def int_aarch64_neon_faddp : AdvSIMD_2VectorArg_Intrinsic;
295
296  // Long Pairwise Add
297  // FIXME: In theory, we shouldn't need intrinsics for saddlp or
298  // uaddlp, but tblgen's type inference currently can't handle the
299  // pattern fragments this ends up generating.
300  def int_aarch64_neon_saddlp : AdvSIMD_1VectorArg_Expand_Intrinsic;
301  def int_aarch64_neon_uaddlp : AdvSIMD_1VectorArg_Expand_Intrinsic;
302
303  // Folding Maximum
304  def int_aarch64_neon_smaxp : AdvSIMD_2VectorArg_Intrinsic;
305  def int_aarch64_neon_umaxp : AdvSIMD_2VectorArg_Intrinsic;
306  def int_aarch64_neon_fmaxp : AdvSIMD_2VectorArg_Intrinsic;
307
308  // Folding Minimum
309  def int_aarch64_neon_sminp : AdvSIMD_2VectorArg_Intrinsic;
310  def int_aarch64_neon_uminp : AdvSIMD_2VectorArg_Intrinsic;
311  def int_aarch64_neon_fminp : AdvSIMD_2VectorArg_Intrinsic;
312
313  // Reciprocal Estimate/Step
314  def int_aarch64_neon_frecps : AdvSIMD_2FloatArg_Intrinsic;
315  def int_aarch64_neon_frsqrts : AdvSIMD_2FloatArg_Intrinsic;
316
317  // Reciprocal Exponent
318  def int_aarch64_neon_frecpx : AdvSIMD_1FloatArg_Intrinsic;
319
320  // Vector Saturating Shift Left
321  def int_aarch64_neon_sqshl : AdvSIMD_2IntArg_Intrinsic;
322  def int_aarch64_neon_uqshl : AdvSIMD_2IntArg_Intrinsic;
323
324  // Vector Rounding Shift Left
325  def int_aarch64_neon_srshl : AdvSIMD_2IntArg_Intrinsic;
326  def int_aarch64_neon_urshl : AdvSIMD_2IntArg_Intrinsic;
327
328  // Vector Saturating Rounding Shift Left
329  def int_aarch64_neon_sqrshl : AdvSIMD_2IntArg_Intrinsic;
330  def int_aarch64_neon_uqrshl : AdvSIMD_2IntArg_Intrinsic;
331
332  // Vector Signed->Unsigned Shift Left by Constant
333  def int_aarch64_neon_sqshlu : AdvSIMD_2IntArg_Intrinsic;
334
335  // Vector Signed->Unsigned Narrowing Saturating Shift Right by Constant
336  def int_aarch64_neon_sqshrun : AdvSIMD_2Arg_Scalar_Narrow_Intrinsic;
337
338  // Vector Signed->Unsigned Rounding Narrowing Saturating Shift Right by Const
339  def int_aarch64_neon_sqrshrun : AdvSIMD_2Arg_Scalar_Narrow_Intrinsic;
340
341  // Vector Narrowing Shift Right by Constant
342  def int_aarch64_neon_sqshrn : AdvSIMD_2Arg_Scalar_Narrow_Intrinsic;
343  def int_aarch64_neon_uqshrn : AdvSIMD_2Arg_Scalar_Narrow_Intrinsic;
344
345  // Vector Rounding Narrowing Shift Right by Constant
346  def int_aarch64_neon_rshrn : AdvSIMD_2Arg_Scalar_Narrow_Intrinsic;
347
348  // Vector Rounding Narrowing Saturating Shift Right by Constant
349  def int_aarch64_neon_sqrshrn : AdvSIMD_2Arg_Scalar_Narrow_Intrinsic;
350  def int_aarch64_neon_uqrshrn : AdvSIMD_2Arg_Scalar_Narrow_Intrinsic;
351
352  // Vector Shift Left
353  def int_aarch64_neon_sshl : AdvSIMD_2IntArg_Intrinsic;
354  def int_aarch64_neon_ushl : AdvSIMD_2IntArg_Intrinsic;
355
356  // Vector Widening Shift Left by Constant
357  def int_aarch64_neon_shll : AdvSIMD_2VectorArg_Scalar_Wide_BySize_Intrinsic;
358  def int_aarch64_neon_sshll : AdvSIMD_2VectorArg_Scalar_Wide_Intrinsic;
359  def int_aarch64_neon_ushll : AdvSIMD_2VectorArg_Scalar_Wide_Intrinsic;
360
361  // Vector Shift Right by Constant and Insert
362  def int_aarch64_neon_vsri : AdvSIMD_3VectorArg_Scalar_Intrinsic;
363
364  // Vector Shift Left by Constant and Insert
365  def int_aarch64_neon_vsli : AdvSIMD_3VectorArg_Scalar_Intrinsic;
366
367  // Vector Saturating Narrow
368  def int_aarch64_neon_scalar_sqxtn: AdvSIMD_1IntArg_Narrow_Intrinsic;
369  def int_aarch64_neon_scalar_uqxtn : AdvSIMD_1IntArg_Narrow_Intrinsic;
370  def int_aarch64_neon_sqxtn : AdvSIMD_1VectorArg_Narrow_Intrinsic;
371  def int_aarch64_neon_uqxtn : AdvSIMD_1VectorArg_Narrow_Intrinsic;
372
373  // Vector Saturating Extract and Unsigned Narrow
374  def int_aarch64_neon_scalar_sqxtun : AdvSIMD_1IntArg_Narrow_Intrinsic;
375  def int_aarch64_neon_sqxtun : AdvSIMD_1VectorArg_Narrow_Intrinsic;
376
377  // Vector Absolute Value
378  def int_aarch64_neon_abs : AdvSIMD_1Arg_Intrinsic;
379
380  // Vector Saturating Absolute Value
381  def int_aarch64_neon_sqabs : AdvSIMD_1IntArg_Intrinsic;
382
383  // Vector Saturating Negation
384  def int_aarch64_neon_sqneg : AdvSIMD_1IntArg_Intrinsic;
385
386  // Vector Count Leading Sign Bits
387  def int_aarch64_neon_cls : AdvSIMD_1VectorArg_Intrinsic;
388
389  // Vector Reciprocal Estimate
390  def int_aarch64_neon_urecpe : AdvSIMD_1VectorArg_Intrinsic;
391  def int_aarch64_neon_frecpe : AdvSIMD_1FloatArg_Intrinsic;
392
393  // Vector Square Root Estimate
394  def int_aarch64_neon_ursqrte : AdvSIMD_1VectorArg_Intrinsic;
395  def int_aarch64_neon_frsqrte : AdvSIMD_1FloatArg_Intrinsic;
396
397  // Vector Bitwise Reverse
398  def int_aarch64_neon_rbit : AdvSIMD_1VectorArg_Intrinsic;
399
400  // Vector Conversions Between Half-Precision and Single-Precision.
401  def int_aarch64_neon_vcvtfp2hf
402    : Intrinsic<[llvm_v4i16_ty], [llvm_v4f32_ty], [IntrNoMem]>;
403  def int_aarch64_neon_vcvthf2fp
404    : Intrinsic<[llvm_v4f32_ty], [llvm_v4i16_ty], [IntrNoMem]>;
405
406  // Vector Conversions Between Floating-point and Fixed-point.
407  def int_aarch64_neon_vcvtfp2fxs : AdvSIMD_CvtFPToFx_Intrinsic;
408  def int_aarch64_neon_vcvtfp2fxu : AdvSIMD_CvtFPToFx_Intrinsic;
409  def int_aarch64_neon_vcvtfxs2fp : AdvSIMD_CvtFxToFP_Intrinsic;
410  def int_aarch64_neon_vcvtfxu2fp : AdvSIMD_CvtFxToFP_Intrinsic;
411
412  // Vector FP->Int Conversions
413  def int_aarch64_neon_fcvtas : AdvSIMD_FPToIntRounding_Intrinsic;
414  def int_aarch64_neon_fcvtau : AdvSIMD_FPToIntRounding_Intrinsic;
415  def int_aarch64_neon_fcvtms : AdvSIMD_FPToIntRounding_Intrinsic;
416  def int_aarch64_neon_fcvtmu : AdvSIMD_FPToIntRounding_Intrinsic;
417  def int_aarch64_neon_fcvtns : AdvSIMD_FPToIntRounding_Intrinsic;
418  def int_aarch64_neon_fcvtnu : AdvSIMD_FPToIntRounding_Intrinsic;
419  def int_aarch64_neon_fcvtps : AdvSIMD_FPToIntRounding_Intrinsic;
420  def int_aarch64_neon_fcvtpu : AdvSIMD_FPToIntRounding_Intrinsic;
421  def int_aarch64_neon_fcvtzs : AdvSIMD_FPToIntRounding_Intrinsic;
422  def int_aarch64_neon_fcvtzu : AdvSIMD_FPToIntRounding_Intrinsic;
423
424  // Vector FP Rounding: only ties to even is unrepresented by a normal
425  // intrinsic.
426  def int_aarch64_neon_frintn : AdvSIMD_1FloatArg_Intrinsic;
427
428  // Scalar FP->Int conversions
429
430  // Vector FP Inexact Narrowing
431  def int_aarch64_neon_fcvtxn : AdvSIMD_1VectorArg_Expand_Intrinsic;
432
433  // Scalar FP Inexact Narrowing
434  def int_aarch64_sisd_fcvtxn : Intrinsic<[llvm_float_ty], [llvm_double_ty],
435                                        [IntrNoMem]>;
436
437  // v8.2-A Dot Product
438  def int_aarch64_neon_udot : AdvSIMD_Dot_Intrinsic;
439  def int_aarch64_neon_sdot : AdvSIMD_Dot_Intrinsic;
440
441  // v8.2-A FP16 Fused Multiply-Add Long
442  def int_aarch64_neon_fmlal : AdvSIMD_FP16FML_Intrinsic;
443  def int_aarch64_neon_fmlsl : AdvSIMD_FP16FML_Intrinsic;
444  def int_aarch64_neon_fmlal2 : AdvSIMD_FP16FML_Intrinsic;
445  def int_aarch64_neon_fmlsl2 : AdvSIMD_FP16FML_Intrinsic;
446}
447
448let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
449  class AdvSIMD_2Vector2Index_Intrinsic
450    : Intrinsic<[llvm_anyvector_ty],
451                [llvm_anyvector_ty, llvm_i64_ty, LLVMMatchType<0>, llvm_i64_ty],
452                [IntrNoMem]>;
453}
454
455// Vector element to element moves
456def int_aarch64_neon_vcopy_lane: AdvSIMD_2Vector2Index_Intrinsic;
457
458let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
459  class AdvSIMD_1Vec_Load_Intrinsic
460      : Intrinsic<[llvm_anyvector_ty], [LLVMAnyPointerType<LLVMMatchType<0>>],
461                  [IntrReadMem, IntrArgMemOnly]>;
462  class AdvSIMD_1Vec_Store_Lane_Intrinsic
463    : Intrinsic<[], [llvm_anyvector_ty, llvm_i64_ty, llvm_anyptr_ty],
464                [IntrArgMemOnly, NoCapture<2>]>;
465
466  class AdvSIMD_2Vec_Load_Intrinsic
467    : Intrinsic<[LLVMMatchType<0>, llvm_anyvector_ty],
468                [LLVMAnyPointerType<LLVMMatchType<0>>],
469                [IntrReadMem, IntrArgMemOnly]>;
470  class AdvSIMD_2Vec_Load_Lane_Intrinsic
471    : Intrinsic<[LLVMMatchType<0>, LLVMMatchType<0>],
472                [LLVMMatchType<0>, llvm_anyvector_ty,
473                 llvm_i64_ty, llvm_anyptr_ty],
474                [IntrReadMem, IntrArgMemOnly]>;
475  class AdvSIMD_2Vec_Store_Intrinsic
476    : Intrinsic<[], [llvm_anyvector_ty, LLVMMatchType<0>,
477                     LLVMAnyPointerType<LLVMMatchType<0>>],
478                [IntrArgMemOnly, NoCapture<2>]>;
479  class AdvSIMD_2Vec_Store_Lane_Intrinsic
480    : Intrinsic<[], [llvm_anyvector_ty, LLVMMatchType<0>,
481                 llvm_i64_ty, llvm_anyptr_ty],
482                [IntrArgMemOnly, NoCapture<3>]>;
483
484  class AdvSIMD_3Vec_Load_Intrinsic
485    : Intrinsic<[LLVMMatchType<0>, LLVMMatchType<0>, llvm_anyvector_ty],
486                [LLVMAnyPointerType<LLVMMatchType<0>>],
487                [IntrReadMem, IntrArgMemOnly]>;
488  class AdvSIMD_3Vec_Load_Lane_Intrinsic
489    : Intrinsic<[LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
490                [LLVMMatchType<0>, LLVMMatchType<0>, llvm_anyvector_ty,
491                 llvm_i64_ty, llvm_anyptr_ty],
492                [IntrReadMem, IntrArgMemOnly]>;
493  class AdvSIMD_3Vec_Store_Intrinsic
494    : Intrinsic<[], [llvm_anyvector_ty, LLVMMatchType<0>,
495                     LLVMMatchType<0>, LLVMAnyPointerType<LLVMMatchType<0>>],
496                [IntrArgMemOnly, NoCapture<3>]>;
497  class AdvSIMD_3Vec_Store_Lane_Intrinsic
498    : Intrinsic<[], [llvm_anyvector_ty,
499                 LLVMMatchType<0>, LLVMMatchType<0>,
500                 llvm_i64_ty, llvm_anyptr_ty],
501                [IntrArgMemOnly, NoCapture<4>]>;
502
503  class AdvSIMD_4Vec_Load_Intrinsic
504    : Intrinsic<[LLVMMatchType<0>, LLVMMatchType<0>,
505                 LLVMMatchType<0>, llvm_anyvector_ty],
506                [LLVMAnyPointerType<LLVMMatchType<0>>],
507                [IntrReadMem, IntrArgMemOnly]>;
508  class AdvSIMD_4Vec_Load_Lane_Intrinsic
509    : Intrinsic<[LLVMMatchType<0>, LLVMMatchType<0>,
510                 LLVMMatchType<0>, LLVMMatchType<0>],
511                [LLVMMatchType<0>, LLVMMatchType<0>,
512                 LLVMMatchType<0>, llvm_anyvector_ty,
513                 llvm_i64_ty, llvm_anyptr_ty],
514                [IntrReadMem, IntrArgMemOnly]>;
515  class AdvSIMD_4Vec_Store_Intrinsic
516    : Intrinsic<[], [llvm_anyvector_ty, LLVMMatchType<0>,
517                 LLVMMatchType<0>, LLVMMatchType<0>,
518                 LLVMAnyPointerType<LLVMMatchType<0>>],
519                [IntrArgMemOnly, NoCapture<4>]>;
520  class AdvSIMD_4Vec_Store_Lane_Intrinsic
521    : Intrinsic<[], [llvm_anyvector_ty, LLVMMatchType<0>,
522                 LLVMMatchType<0>, LLVMMatchType<0>,
523                 llvm_i64_ty, llvm_anyptr_ty],
524                [IntrArgMemOnly, NoCapture<5>]>;
525}
526
527// Memory ops
528
529def int_aarch64_neon_ld1x2 : AdvSIMD_2Vec_Load_Intrinsic;
530def int_aarch64_neon_ld1x3 : AdvSIMD_3Vec_Load_Intrinsic;
531def int_aarch64_neon_ld1x4 : AdvSIMD_4Vec_Load_Intrinsic;
532
533def int_aarch64_neon_st1x2 : AdvSIMD_2Vec_Store_Intrinsic;
534def int_aarch64_neon_st1x3 : AdvSIMD_3Vec_Store_Intrinsic;
535def int_aarch64_neon_st1x4 : AdvSIMD_4Vec_Store_Intrinsic;
536
537def int_aarch64_neon_ld2 : AdvSIMD_2Vec_Load_Intrinsic;
538def int_aarch64_neon_ld3 : AdvSIMD_3Vec_Load_Intrinsic;
539def int_aarch64_neon_ld4 : AdvSIMD_4Vec_Load_Intrinsic;
540
541def int_aarch64_neon_ld2lane : AdvSIMD_2Vec_Load_Lane_Intrinsic;
542def int_aarch64_neon_ld3lane : AdvSIMD_3Vec_Load_Lane_Intrinsic;
543def int_aarch64_neon_ld4lane : AdvSIMD_4Vec_Load_Lane_Intrinsic;
544
545def int_aarch64_neon_ld2r : AdvSIMD_2Vec_Load_Intrinsic;
546def int_aarch64_neon_ld3r : AdvSIMD_3Vec_Load_Intrinsic;
547def int_aarch64_neon_ld4r : AdvSIMD_4Vec_Load_Intrinsic;
548
549def int_aarch64_neon_st2  : AdvSIMD_2Vec_Store_Intrinsic;
550def int_aarch64_neon_st3  : AdvSIMD_3Vec_Store_Intrinsic;
551def int_aarch64_neon_st4  : AdvSIMD_4Vec_Store_Intrinsic;
552
553def int_aarch64_neon_st2lane  : AdvSIMD_2Vec_Store_Lane_Intrinsic;
554def int_aarch64_neon_st3lane  : AdvSIMD_3Vec_Store_Lane_Intrinsic;
555def int_aarch64_neon_st4lane  : AdvSIMD_4Vec_Store_Lane_Intrinsic;
556
557let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
558  class AdvSIMD_Tbl1_Intrinsic
559    : Intrinsic<[llvm_anyvector_ty], [llvm_v16i8_ty, LLVMMatchType<0>],
560                [IntrNoMem]>;
561  class AdvSIMD_Tbl2_Intrinsic
562    : Intrinsic<[llvm_anyvector_ty],
563                [llvm_v16i8_ty, llvm_v16i8_ty, LLVMMatchType<0>], [IntrNoMem]>;
564  class AdvSIMD_Tbl3_Intrinsic
565    : Intrinsic<[llvm_anyvector_ty],
566                [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty,
567                 LLVMMatchType<0>],
568                [IntrNoMem]>;
569  class AdvSIMD_Tbl4_Intrinsic
570    : Intrinsic<[llvm_anyvector_ty],
571                [llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty, llvm_v16i8_ty,
572                 LLVMMatchType<0>],
573                [IntrNoMem]>;
574
575  class AdvSIMD_Tbx1_Intrinsic
576    : Intrinsic<[llvm_anyvector_ty],
577                [LLVMMatchType<0>, llvm_v16i8_ty, LLVMMatchType<0>],
578                [IntrNoMem]>;
579  class AdvSIMD_Tbx2_Intrinsic
580    : Intrinsic<[llvm_anyvector_ty],
581                [LLVMMatchType<0>, llvm_v16i8_ty, llvm_v16i8_ty,
582                 LLVMMatchType<0>],
583                [IntrNoMem]>;
584  class AdvSIMD_Tbx3_Intrinsic
585    : Intrinsic<[llvm_anyvector_ty],
586                [LLVMMatchType<0>, llvm_v16i8_ty, llvm_v16i8_ty,
587                 llvm_v16i8_ty, LLVMMatchType<0>],
588                [IntrNoMem]>;
589  class AdvSIMD_Tbx4_Intrinsic
590    : Intrinsic<[llvm_anyvector_ty],
591                [LLVMMatchType<0>, llvm_v16i8_ty, llvm_v16i8_ty,
592                 llvm_v16i8_ty, llvm_v16i8_ty, LLVMMatchType<0>],
593                [IntrNoMem]>;
594}
595def int_aarch64_neon_tbl1 : AdvSIMD_Tbl1_Intrinsic;
596def int_aarch64_neon_tbl2 : AdvSIMD_Tbl2_Intrinsic;
597def int_aarch64_neon_tbl3 : AdvSIMD_Tbl3_Intrinsic;
598def int_aarch64_neon_tbl4 : AdvSIMD_Tbl4_Intrinsic;
599
600def int_aarch64_neon_tbx1 : AdvSIMD_Tbx1_Intrinsic;
601def int_aarch64_neon_tbx2 : AdvSIMD_Tbx2_Intrinsic;
602def int_aarch64_neon_tbx3 : AdvSIMD_Tbx3_Intrinsic;
603def int_aarch64_neon_tbx4 : AdvSIMD_Tbx4_Intrinsic;
604
605let TargetPrefix = "aarch64" in {
606  class FPCR_Get_Intrinsic
607    : Intrinsic<[llvm_i64_ty], [], [IntrNoMem]>;
608}
609
610// FPCR
611def int_aarch64_get_fpcr : FPCR_Get_Intrinsic;
612
613let TargetPrefix = "aarch64" in {
614  class Crypto_AES_DataKey_Intrinsic
615    : Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty, llvm_v16i8_ty], [IntrNoMem]>;
616
617  class Crypto_AES_Data_Intrinsic
618    : Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty], [IntrNoMem]>;
619
620  // SHA intrinsic taking 5 words of the hash (v4i32, i32) and 4 of the schedule
621  // (v4i32).
622  class Crypto_SHA_5Hash4Schedule_Intrinsic
623    : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_i32_ty, llvm_v4i32_ty],
624                [IntrNoMem]>;
625
626  // SHA intrinsic taking 5 words of the hash (v4i32, i32) and 4 of the schedule
627  // (v4i32).
628  class Crypto_SHA_1Hash_Intrinsic
629    : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem]>;
630
631  // SHA intrinsic taking 8 words of the schedule
632  class Crypto_SHA_8Schedule_Intrinsic
633    : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty], [IntrNoMem]>;
634
635  // SHA intrinsic taking 12 words of the schedule
636  class Crypto_SHA_12Schedule_Intrinsic
637    : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
638                [IntrNoMem]>;
639
640  // SHA intrinsic taking 8 words of the hash and 4 of the schedule.
641  class Crypto_SHA_8Hash4Schedule_Intrinsic
642    : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty, llvm_v4i32_ty],
643                [IntrNoMem]>;
644}
645
646// AES
647def int_aarch64_crypto_aese   : Crypto_AES_DataKey_Intrinsic;
648def int_aarch64_crypto_aesd   : Crypto_AES_DataKey_Intrinsic;
649def int_aarch64_crypto_aesmc  : Crypto_AES_Data_Intrinsic;
650def int_aarch64_crypto_aesimc : Crypto_AES_Data_Intrinsic;
651
652// SHA1
653def int_aarch64_crypto_sha1c  : Crypto_SHA_5Hash4Schedule_Intrinsic;
654def int_aarch64_crypto_sha1p  : Crypto_SHA_5Hash4Schedule_Intrinsic;
655def int_aarch64_crypto_sha1m  : Crypto_SHA_5Hash4Schedule_Intrinsic;
656def int_aarch64_crypto_sha1h  : Crypto_SHA_1Hash_Intrinsic;
657
658def int_aarch64_crypto_sha1su0 : Crypto_SHA_12Schedule_Intrinsic;
659def int_aarch64_crypto_sha1su1 : Crypto_SHA_8Schedule_Intrinsic;
660
661// SHA256
662def int_aarch64_crypto_sha256h   : Crypto_SHA_8Hash4Schedule_Intrinsic;
663def int_aarch64_crypto_sha256h2  : Crypto_SHA_8Hash4Schedule_Intrinsic;
664def int_aarch64_crypto_sha256su0 : Crypto_SHA_8Schedule_Intrinsic;
665def int_aarch64_crypto_sha256su1 : Crypto_SHA_12Schedule_Intrinsic;
666
667//===----------------------------------------------------------------------===//
668// CRC32
669
670let TargetPrefix = "aarch64" in {
671
672def int_aarch64_crc32b  : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
673    [IntrNoMem]>;
674def int_aarch64_crc32cb : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
675    [IntrNoMem]>;
676def int_aarch64_crc32h  : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
677    [IntrNoMem]>;
678def int_aarch64_crc32ch : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
679    [IntrNoMem]>;
680def int_aarch64_crc32w  : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
681    [IntrNoMem]>;
682def int_aarch64_crc32cw : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
683    [IntrNoMem]>;
684def int_aarch64_crc32x  : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i64_ty],
685    [IntrNoMem]>;
686def int_aarch64_crc32cx : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i64_ty],
687    [IntrNoMem]>;
688}
689
690//===----------------------------------------------------------------------===//
691// Memory Tagging Extensions (MTE) Intrinsics
692let TargetPrefix = "aarch64" in {
693def int_aarch64_irg   : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty, llvm_i64_ty],
694    [IntrInaccessibleMemOnly]>;
695def int_aarch64_addg  : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty, llvm_i64_ty],
696    [IntrNoMem]>;
697def int_aarch64_gmi   : Intrinsic<[llvm_i64_ty], [llvm_ptr_ty, llvm_i64_ty],
698    [IntrNoMem]>;
699def int_aarch64_ldg   : Intrinsic<[llvm_ptr_ty], [llvm_ptr_ty, llvm_ptr_ty],
700    [IntrReadMem]>;
701def int_aarch64_stg   : Intrinsic<[], [llvm_ptr_ty, llvm_ptr_ty],
702    [IntrWriteMem]>;
703def int_aarch64_subp :  Intrinsic<[llvm_i64_ty], [llvm_ptr_ty, llvm_ptr_ty],
704    [IntrNoMem]>;
705
706// The following are codegen-only intrinsics for stack instrumentation.
707
708// Generate a randomly tagged stack base pointer.
709def int_aarch64_irg_sp   : Intrinsic<[llvm_ptr_ty], [llvm_i64_ty],
710    [IntrInaccessibleMemOnly]>;
711
712// Transfer pointer tag with offset.
713// ptr1 = tagp(ptr0, baseptr, tag_offset) returns a pointer where
714// * address is the address in ptr0
715// * tag is a function of (tag in baseptr, tag_offset).
716// Address bits in baseptr and tag bits in ptr0 are ignored.
717// When offset between ptr0 and baseptr is a compile time constant, this can be emitted as
718//   ADDG ptr1, baseptr, (ptr0 - baseptr), tag_offset
719// It is intended that ptr0 is an alloca address, and baseptr is the direct output of llvm.aarch64.irg.sp.
720def int_aarch64_tagp : Intrinsic<[llvm_anyptr_ty], [LLVMMatchType<0>, llvm_ptr_ty, llvm_i64_ty],
721    [IntrNoMem, ImmArg<2>]>;
722
723// Update allocation tags for the memory range to match the tag in the pointer argument.
724def int_aarch64_settag  : Intrinsic<[], [llvm_ptr_ty, llvm_i64_ty],
725    [IntrWriteMem, IntrArgMemOnly, NoCapture<0>, WriteOnly<0>]>;
726
727// Update allocation tags for the memory range to match the tag in the pointer argument,
728// and set memory contents to zero.
729def int_aarch64_settag_zero  : Intrinsic<[], [llvm_ptr_ty, llvm_i64_ty],
730    [IntrWriteMem, IntrArgMemOnly, NoCapture<0>, WriteOnly<0>]>;
731
732// Update allocation tags for 16-aligned, 16-sized memory region, and store a pair 8-byte values.
733def int_aarch64_stgp  : Intrinsic<[], [llvm_ptr_ty, llvm_i64_ty, llvm_i64_ty],
734    [IntrWriteMem, IntrArgMemOnly, NoCapture<0>, WriteOnly<0>]>;
735}
736