1 //===-- ARMInstrMVE.td - MVE support for ARM ---------------*- tablegen -*-===//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 // This file describes the ARM MVE instruction set.
11 //===----------------------------------------------------------------------===//
13 class ExpandImmAsmOp<string shift> : AsmOperandClass {
14 let Name = !strconcat("ExpandImm", shift);
15 let PredicateMethod = !strconcat("isExpImm<", shift, ">");
16 let RenderMethod = "addImmOperands";
18 class InvertedExpandImmAsmOp<string shift, string size> : AsmOperandClass {
19 let Name = !strconcat("InvertedExpandImm", shift, "_", size);
20 let PredicateMethod = !strconcat("isInvertedExpImm<", shift, ",", size, ">");
21 let RenderMethod = "addImmOperands";
24 class ExpandImm<string shift> : Operand<i32> {
25 let ParserMatchClass = ExpandImmAsmOp<shift>;
26 let EncoderMethod = !strconcat("getExpandedImmOpValue<",shift,",false>");
27 let DecoderMethod = !strconcat("DecodeExpandedImmOperand<",shift,">");
28 let PrintMethod = "printExpandedImmOperand";
30 class InvertedExpandImm<string shift, string size> : Operand<i32> {
31 let ParserMatchClass = InvertedExpandImmAsmOp<shift, size>;
32 let EncoderMethod = !strconcat("getExpandedImmOpValue<",shift,",true>");
33 let PrintMethod = "printExpandedImmOperand";
34 // No decoder method needed, because this operand type is only used
35 // by aliases (VAND and VORN)
38 def expzero00 : ExpandImm<"0">;
39 def expzero08 : ExpandImm<"8">;
40 def expzero16 : ExpandImm<"16">;
41 def expzero24 : ExpandImm<"24">;
43 def expzero00inv16 : InvertedExpandImm<"0", "16">;
44 def expzero08inv16 : InvertedExpandImm<"8", "16">;
46 def expzero00inv32 : InvertedExpandImm<"0", "32">;
47 def expzero08inv32 : InvertedExpandImm<"8", "32">;
48 def expzero16inv32 : InvertedExpandImm<"16", "32">;
49 def expzero24inv32 : InvertedExpandImm<"24", "32">;
52 def vpt_mask : Operand<i32> {
53 let PrintMethod = "printVPTMask";
54 let ParserMatchClass = it_mask_asmoperand;
55 let EncoderMethod = "getVPTMaskOpValue";
56 let DecoderMethod = "DecodeVPTMaskOperand";
59 // VPT/VCMP restricted predicate for sign invariant types
60 def pred_restricted_i_asmoperand : AsmOperandClass {
61 let Name = "CondCodeRestrictedI";
62 let RenderMethod = "addITCondCodeOperands";
63 let PredicateMethod = "isITCondCodeRestrictedI";
64 let ParserMethod = "parseITCondCode";
65 let DiagnosticString = "condition code for sign-independent integer "#
66 "comparison must be EQ or NE";
69 // VPT/VCMP restricted predicate for signed types
70 def pred_restricted_s_asmoperand : AsmOperandClass {
71 let Name = "CondCodeRestrictedS";
72 let RenderMethod = "addITCondCodeOperands";
73 let PredicateMethod = "isITCondCodeRestrictedS";
74 let ParserMethod = "parseITCondCode";
75 let DiagnosticString = "condition code for signed integer "#
76 "comparison must be EQ, NE, LT, GT, LE or GE";
79 // VPT/VCMP restricted predicate for unsigned types
80 def pred_restricted_u_asmoperand : AsmOperandClass {
81 let Name = "CondCodeRestrictedU";
82 let RenderMethod = "addITCondCodeOperands";
83 let PredicateMethod = "isITCondCodeRestrictedU";
84 let ParserMethod = "parseITCondCode";
85 let DiagnosticString = "condition code for unsigned integer "#
86 "comparison must be EQ, NE, HS or HI";
89 // VPT/VCMP restricted predicate for floating point
90 def pred_restricted_fp_asmoperand : AsmOperandClass {
91 let Name = "CondCodeRestrictedFP";
92 let RenderMethod = "addITCondCodeOperands";
93 let PredicateMethod = "isITCondCodeRestrictedFP";
94 let ParserMethod = "parseITCondCode";
95 let DiagnosticString = "condition code for floating-point "#
96 "comparison must be EQ, NE, LT, GT, LE or GE";
99 class VCMPPredicateOperand : Operand<i32>;
101 def pred_basic_i : VCMPPredicateOperand {
102 let PrintMethod = "printMandatoryRestrictedPredicateOperand";
103 let ParserMatchClass = pred_restricted_i_asmoperand;
104 let DecoderMethod = "DecodeRestrictedIPredicateOperand";
105 let EncoderMethod = "getRestrictedCondCodeOpValue";
108 def pred_basic_u : VCMPPredicateOperand {
109 let PrintMethod = "printMandatoryRestrictedPredicateOperand";
110 let ParserMatchClass = pred_restricted_u_asmoperand;
111 let DecoderMethod = "DecodeRestrictedUPredicateOperand";
112 let EncoderMethod = "getRestrictedCondCodeOpValue";
115 def pred_basic_s : VCMPPredicateOperand {
116 let PrintMethod = "printMandatoryRestrictedPredicateOperand";
117 let ParserMatchClass = pred_restricted_s_asmoperand;
118 let DecoderMethod = "DecodeRestrictedSPredicateOperand";
119 let EncoderMethod = "getRestrictedCondCodeOpValue";
122 def pred_basic_fp : VCMPPredicateOperand {
123 let PrintMethod = "printMandatoryRestrictedPredicateOperand";
124 let ParserMatchClass = pred_restricted_fp_asmoperand;
125 let DecoderMethod = "DecodeRestrictedFPPredicateOperand";
126 let EncoderMethod = "getRestrictedCondCodeOpValue";
129 // Register list operands for interleaving load/stores
130 def VecList2QAsmOperand : AsmOperandClass {
131 let Name = "VecListTwoMQ";
132 let ParserMethod = "parseVectorList";
133 let RenderMethod = "addMVEVecListOperands";
134 let DiagnosticString = "operand must be a list of two consecutive "#
135 "q-registers in range [q0,q7]";
138 def VecList2Q : RegisterOperand<QQPR, "printMVEVectorListTwoQ"> {
139 let ParserMatchClass = VecList2QAsmOperand;
140 let PrintMethod = "printMVEVectorList<2>";
143 def VecList4QAsmOperand : AsmOperandClass {
144 let Name = "VecListFourMQ";
145 let ParserMethod = "parseVectorList";
146 let RenderMethod = "addMVEVecListOperands";
147 let DiagnosticString = "operand must be a list of four consecutive "#
148 "q-registers in range [q0,q7]";
151 def VecList4Q : RegisterOperand<QQQQPR, "printMVEVectorListFourQ"> {
152 let ParserMatchClass = VecList4QAsmOperand;
153 let PrintMethod = "printMVEVectorList<4>";
156 // taddrmode_imm7 := reg[r0-r7] +/- (imm7 << shift)
157 class TMemImm7ShiftOffsetAsmOperand<int shift> : AsmOperandClass {
158 let Name = "TMemImm7Shift"#shift#"Offset";
159 let PredicateMethod = "isMemImm7ShiftedOffset<"#shift#",ARM::tGPRRegClassID>";
160 let RenderMethod = "addMemImmOffsetOperands";
163 class taddrmode_imm7<int shift> : MemOperand {
164 let ParserMatchClass = TMemImm7ShiftOffsetAsmOperand<shift>;
165 // They are printed the same way as the T2 imm8 version
166 let PrintMethod = "printT2AddrModeImm8Operand<false>";
167 // This can also be the same as the T2 version.
168 let EncoderMethod = "getT2AddrModeImmOpValue<7,"#shift#">";
169 let DecoderMethod = "DecodeTAddrModeImm7<"#shift#">";
170 let MIOperandInfo = (ops tGPR:$base, i32imm:$offsimm);
173 // t2addrmode_imm7 := reg +/- (imm7)
174 class MemImm7ShiftOffsetAsmOperand<int shift> : AsmOperandClass {
175 let Name = "MemImm7Shift"#shift#"Offset";
176 let PredicateMethod = "isMemImm7ShiftedOffset<" # shift #
177 ",ARM::GPRnopcRegClassID>";
178 let RenderMethod = "addMemImmOffsetOperands";
181 def MemImm7Shift0OffsetAsmOperand : MemImm7ShiftOffsetAsmOperand<0>;
182 def MemImm7Shift1OffsetAsmOperand : MemImm7ShiftOffsetAsmOperand<1>;
183 def MemImm7Shift2OffsetAsmOperand : MemImm7ShiftOffsetAsmOperand<2>;
184 class T2AddrMode_Imm7<int shift> : MemOperand,
185 ComplexPattern<i32, 2, "SelectT2AddrModeImm7<"#shift#">", []> {
186 let EncoderMethod = "getT2AddrModeImmOpValue<7,"#shift#">";
187 let DecoderMethod = "DecodeT2AddrModeImm7<"#shift#", 0>";
188 let ParserMatchClass =
189 !cast<AsmOperandClass>("MemImm7Shift"#shift#"OffsetAsmOperand");
190 let MIOperandInfo = (ops GPRnopc:$base, i32imm:$offsimm);
193 class t2addrmode_imm7<int shift> : T2AddrMode_Imm7<shift> {
194 // They are printed the same way as the imm8 version
195 let PrintMethod = "printT2AddrModeImm8Operand<false>";
198 class MemImm7ShiftOffsetWBAsmOperand<int shift> : AsmOperandClass {
199 let Name = "MemImm7Shift"#shift#"OffsetWB";
200 let PredicateMethod = "isMemImm7ShiftedOffset<" # shift #
201 ",ARM::rGPRRegClassID>";
202 let RenderMethod = "addMemImmOffsetOperands";
205 def MemImm7Shift0OffsetWBAsmOperand : MemImm7ShiftOffsetWBAsmOperand<0>;
206 def MemImm7Shift1OffsetWBAsmOperand : MemImm7ShiftOffsetWBAsmOperand<1>;
207 def MemImm7Shift2OffsetWBAsmOperand : MemImm7ShiftOffsetWBAsmOperand<2>;
209 class t2addrmode_imm7_pre<int shift> : T2AddrMode_Imm7<shift> {
210 // They are printed the same way as the imm8 version
211 let PrintMethod = "printT2AddrModeImm8Operand<true>";
212 let ParserMatchClass =
213 !cast<AsmOperandClass>("MemImm7Shift"#shift#"OffsetWBAsmOperand");
214 let DecoderMethod = "DecodeT2AddrModeImm7<"#shift#", 1>";
215 let MIOperandInfo = (ops rGPR:$base, i32imm:$offsim);
218 class t2am_imm7shiftOffsetAsmOperand<int shift>
219 : AsmOperandClass { let Name = "Imm7Shift"#shift; }
220 def t2am_imm7shift0OffsetAsmOperand : t2am_imm7shiftOffsetAsmOperand<0>;
221 def t2am_imm7shift1OffsetAsmOperand : t2am_imm7shiftOffsetAsmOperand<1>;
222 def t2am_imm7shift2OffsetAsmOperand : t2am_imm7shiftOffsetAsmOperand<2>;
224 class t2am_imm7_offset<int shift> : MemOperand {
225 // They are printed the same way as the imm8 version
226 let PrintMethod = "printT2AddrModeImm8OffsetOperand";
227 let ParserMatchClass =
228 !cast<AsmOperandClass>("t2am_imm7shift"#shift#"OffsetAsmOperand");
229 let EncoderMethod = "getT2ScaledImmOpValue<7,"#shift#">";
230 let DecoderMethod = "DecodeT2Imm7<"#shift#">";
233 // Operands for gather/scatter loads of the form [Rbase, Qoffsets]
234 class MemRegRQOffsetAsmOperand<int shift> : AsmOperandClass {
235 let Name = "MemRegRQS"#shift#"Offset";
236 let PredicateMethod = "isMemRegRQOffset<"#shift#">";
237 let RenderMethod = "addMemRegRQOffsetOperands";
240 def MemRegRQS0OffsetAsmOperand : MemRegRQOffsetAsmOperand<0>;
241 def MemRegRQS1OffsetAsmOperand : MemRegRQOffsetAsmOperand<1>;
242 def MemRegRQS2OffsetAsmOperand : MemRegRQOffsetAsmOperand<2>;
243 def MemRegRQS3OffsetAsmOperand : MemRegRQOffsetAsmOperand<3>;
245 // mve_addr_rq_shift := reg + vreg{ << UXTW #shift}
246 class mve_addr_rq_shift<int shift> : MemOperand {
247 let EncoderMethod = "getMveAddrModeRQOpValue";
248 let PrintMethod = "printMveAddrModeRQOperand<"#shift#">";
249 let ParserMatchClass =
250 !cast<AsmOperandClass>("MemRegRQS"#shift#"OffsetAsmOperand");
251 let DecoderMethod = "DecodeMveAddrModeRQ";
252 let MIOperandInfo = (ops GPRnopc:$base, MQPR:$offsreg);
255 class MemRegQOffsetAsmOperand<int shift> : AsmOperandClass {
256 let Name = "MemRegQS"#shift#"Offset";
257 let PredicateMethod = "isMemRegQOffset<"#shift#">";
258 let RenderMethod = "addMemImmOffsetOperands";
261 def MemRegQS2OffsetAsmOperand : MemRegQOffsetAsmOperand<2>;
262 def MemRegQS3OffsetAsmOperand : MemRegQOffsetAsmOperand<3>;
264 // mve_addr_q_shift := vreg {+ #imm7s2/4}
265 class mve_addr_q_shift<int shift> : MemOperand {
266 let EncoderMethod = "getMveAddrModeQOpValue<"#shift#">";
267 // Can be printed same way as other reg + imm operands
268 let PrintMethod = "printT2AddrModeImm8Operand<false>";
269 let ParserMatchClass =
270 !cast<AsmOperandClass>("MemRegQS"#shift#"OffsetAsmOperand");
271 let DecoderMethod = "DecodeMveAddrModeQ<"#shift#">";
272 let MIOperandInfo = (ops MQPR:$base, i32imm:$imm);
275 // --------- Start of base classes for the instructions themselves
277 class MVE_MI<dag oops, dag iops, InstrItinClass itin, string asm,
278 string ops, string cstr, list<dag> pattern>
279 : Thumb2XI<oops, iops, AddrModeNone, 4, itin, !strconcat(asm, "\t", ops), cstr,
281 Requires<[HasMVEInt]> {
283 let DecoderNamespace = "MVE";
286 // MVE_p is used for most predicated instructions, to add the cluster
287 // of input operands that provides the VPT suffix (none, T or E) and
288 // the input predicate register.
289 class MVE_p<dag oops, dag iops, InstrItinClass itin, string iname,
290 string suffix, string ops, vpred_ops vpred, string cstr,
291 list<dag> pattern=[]>
292 : MVE_MI<oops, !con(iops, (ins vpred:$vp)), itin,
293 // If the instruction has a suffix, like vadd.f32, then the
294 // VPT predication suffix goes before the dot, so the full
295 // name has to be "vadd${vp}.f32".
296 !strconcat(iname, "${vp}",
297 !if(!eq(suffix, ""), "", !strconcat(".", suffix))),
298 ops, !strconcat(cstr, vpred.vpred_constraint), pattern> {
299 let Inst{31-29} = 0b111;
300 let Inst{27-26} = 0b11;
303 class MVE_f<dag oops, dag iops, InstrItinClass itin, string iname,
304 string suffix, string ops, vpred_ops vpred, string cstr,
305 list<dag> pattern=[]>
306 : MVE_p<oops, iops, itin, iname, suffix, ops, vpred, cstr, pattern> {
307 let Predicates = [HasMVEFloat];
310 class MVE_MI_with_pred<dag oops, dag iops, InstrItinClass itin, string asm,
311 string ops, string cstr, list<dag> pattern>
312 : Thumb2I<oops, iops, AddrModeNone, 4, itin, asm, !strconcat("\t", ops), cstr,
314 Requires<[HasV8_1MMainline, HasMVEInt]> {
316 let DecoderNamespace = "MVE";
319 class MVE_VMOV_lane_base<dag oops, dag iops, InstrItinClass itin, string asm,
320 string suffix, string ops, string cstr,
322 : Thumb2I<oops, iops, AddrModeNone, 4, itin, asm,
323 !if(!eq(suffix, ""), "", "." # suffix) # "\t" # ops,
325 Requires<[HasV8_1MMainline, HasMVEInt]> {
327 let DecoderNamespace = "MVE";
330 class MVE_ScalarShift<string iname, dag oops, dag iops, string asm, string cstr,
331 list<dag> pattern=[]>
332 : MVE_MI_with_pred<oops, iops, NoItinerary, iname, asm, cstr, pattern> {
333 let Inst{31-20} = 0b111010100101;
338 class MVE_ScalarShiftSingleReg<string iname, dag iops, string asm, string cstr,
339 list<dag> pattern=[]>
340 : MVE_ScalarShift<iname, (outs rGPR:$RdaDest), iops, asm, cstr, pattern> {
343 let Inst{19-16} = RdaDest{3-0};
346 class MVE_ScalarShiftSRegImm<string iname, bits<2> op5_4, list<dag> pattern=[]>
347 : MVE_ScalarShiftSingleReg<iname, (ins rGPR:$RdaSrc, long_shift:$imm),
348 "$RdaSrc, $imm", "$RdaDest = $RdaSrc", pattern> {
352 let Inst{14-12} = imm{4-2};
353 let Inst{11-8} = 0b1111;
354 let Inst{7-6} = imm{1-0};
355 let Inst{5-4} = op5_4{1-0};
356 let Inst{3-0} = 0b1111;
359 def MVE_SQSHL : MVE_ScalarShiftSRegImm<"sqshl", 0b11>;
360 def MVE_SRSHR : MVE_ScalarShiftSRegImm<"srshr", 0b10>;
361 def MVE_UQSHL : MVE_ScalarShiftSRegImm<"uqshl", 0b00>;
362 def MVE_URSHR : MVE_ScalarShiftSRegImm<"urshr", 0b01>;
364 class MVE_ScalarShiftSRegReg<string iname, bits<2> op5_4, list<dag> pattern=[]>
365 : MVE_ScalarShiftSingleReg<iname, (ins rGPR:$RdaSrc, rGPR:$Rm),
366 "$RdaSrc, $Rm", "$RdaDest = $RdaSrc", pattern> {
369 let Inst{15-12} = Rm{3-0};
370 let Inst{11-8} = 0b1111;
371 let Inst{7-6} = 0b00;
372 let Inst{5-4} = op5_4{1-0};
373 let Inst{3-0} = 0b1101;
376 def MVE_SQRSHR : MVE_ScalarShiftSRegReg<"sqrshr", 0b10>;
377 def MVE_UQRSHL : MVE_ScalarShiftSRegReg<"uqrshl", 0b00>;
379 class MVE_ScalarShiftDoubleReg<string iname, dag iops, string asm,
380 string cstr, list<dag> pattern=[]>
381 : MVE_ScalarShift<iname, (outs tGPREven:$RdaLo, tGPROdd:$RdaHi),
382 iops, asm, cstr, pattern> {
386 let Inst{19-17} = RdaLo{3-1};
387 let Inst{11-9} = RdaHi{3-1};
390 class MVE_ScalarShiftDRegImm<string iname, bits<2> op5_4, bit op16,
391 list<dag> pattern=[]>
392 : MVE_ScalarShiftDoubleReg<
393 iname, (ins tGPREven:$RdaLo_src, tGPROdd:$RdaHi_src, long_shift:$imm),
394 "$RdaLo, $RdaHi, $imm", "$RdaLo = $RdaLo_src,$RdaHi = $RdaHi_src",
400 let Inst{14-12} = imm{4-2};
401 let Inst{7-6} = imm{1-0};
402 let Inst{5-4} = op5_4{1-0};
403 let Inst{3-0} = 0b1111;
406 class MVE_ScalarShiftDRegReg<string iname, bit op5, bit op16,
407 list<dag> pattern=[]>
408 : MVE_ScalarShiftDoubleReg<
409 iname, (ins tGPREven:$RdaLo_src, tGPROdd:$RdaHi_src, rGPR:$Rm),
410 "$RdaLo, $RdaHi, $Rm", "@earlyclobber $RdaHi,@earlyclobber $RdaLo,"
411 "$RdaLo = $RdaLo_src,$RdaHi = $RdaHi_src",
416 let Inst{15-12} = Rm{3-0};
417 let Inst{7-6} = 0b00;
420 let Inst{3-0} = 0b1101;
422 // Custom decoder method because of the following overlapping encodings:
425 // SQRSHRL and SQRSHR
426 // UQRSHLL and UQRSHL
427 let DecoderMethod = "DecodeMVEOverlappingLongShift";
430 def MVE_ASRLr : MVE_ScalarShiftDRegReg<"asrl", 0b1, 0b0, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
431 (ARMasrl tGPREven:$RdaLo_src,
432 tGPROdd:$RdaHi_src, rGPR:$Rm))]>;
433 def MVE_ASRLi : MVE_ScalarShiftDRegImm<"asrl", 0b10, ?, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
434 (ARMasrl tGPREven:$RdaLo_src,
435 tGPROdd:$RdaHi_src, (i32 imm:$imm)))]>;
436 def MVE_LSLLr : MVE_ScalarShiftDRegReg<"lsll", 0b0, 0b0, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
437 (ARMlsll tGPREven:$RdaLo_src,
438 tGPROdd:$RdaHi_src, rGPR:$Rm))]>;
439 def MVE_LSLLi : MVE_ScalarShiftDRegImm<"lsll", 0b00, ?, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
440 (ARMlsll tGPREven:$RdaLo_src,
441 tGPROdd:$RdaHi_src, (i32 imm:$imm)))]>;
442 def MVE_LSRL : MVE_ScalarShiftDRegImm<"lsrl", 0b01, ?, [(set tGPREven:$RdaLo, tGPROdd:$RdaHi,
443 (ARMlsrl tGPREven:$RdaLo_src,
444 tGPROdd:$RdaHi_src, (i32 imm:$imm)))]>;
446 def MVE_SQRSHRL : MVE_ScalarShiftDRegReg<"sqrshrl", 0b1, 0b1>;
447 def MVE_SQSHLL : MVE_ScalarShiftDRegImm<"sqshll", 0b11, 0b1>;
448 def MVE_SRSHRL : MVE_ScalarShiftDRegImm<"srshrl", 0b10, 0b1>;
450 def MVE_UQRSHLL : MVE_ScalarShiftDRegReg<"uqrshll", 0b0, 0b1>;
451 def MVE_UQSHLL : MVE_ScalarShiftDRegImm<"uqshll", 0b00, 0b1>;
452 def MVE_URSHRL : MVE_ScalarShiftDRegImm<"urshrl", 0b01, 0b1>;
454 // start of mve_rDest instructions
456 class MVE_rDest<dag oops, dag iops, InstrItinClass itin,
457 string iname, string suffix,
458 string ops, string cstr, list<dag> pattern=[]>
459 // Always use vpred_n and not vpred_r: with the output register being
460 // a GPR and not a vector register, there can't be any question of
461 // what to put in its inactive lanes.
462 : MVE_p<oops, iops, itin, iname, suffix, ops, vpred_n, cstr, pattern> {
464 let Inst{25-23} = 0b101;
465 let Inst{11-9} = 0b111;
469 class MVE_VABAV<string suffix, bit U, bits<2> size, list<dag> pattern=[]>
470 : MVE_rDest<(outs rGPR:$Rda), (ins rGPR:$Rda_src, MQPR:$Qn, MQPR:$Qm),
471 NoItinerary, "vabav", suffix, "$Rda, $Qn, $Qm", "$Rda = $Rda_src",
479 let Inst{21-20} = size{1-0};
480 let Inst{19-17} = Qn{2-0};
482 let Inst{15-12} = Rda{3-0};
487 let Inst{3-1} = Qm{2-0};
491 def MVE_VABAVs8 : MVE_VABAV<"s8", 0b0, 0b00>;
492 def MVE_VABAVs16 : MVE_VABAV<"s16", 0b0, 0b01>;
493 def MVE_VABAVs32 : MVE_VABAV<"s32", 0b0, 0b10>;
494 def MVE_VABAVu8 : MVE_VABAV<"u8", 0b1, 0b00>;
495 def MVE_VABAVu16 : MVE_VABAV<"u16", 0b1, 0b01>;
496 def MVE_VABAVu32 : MVE_VABAV<"u32", 0b1, 0b10>;
498 class MVE_VADDV<string iname, string suffix, dag iops, string cstr,
499 bit A, bit U, bits<2> size, list<dag> pattern=[]>
500 : MVE_rDest<(outs tGPREven:$Rda), iops, NoItinerary,
501 iname, suffix, "$Rda, $Qm", cstr, pattern> {
506 let Inst{22-20} = 0b111;
507 let Inst{19-18} = size{1-0};
508 let Inst{17-16} = 0b01;
509 let Inst{15-13} = Rda{3-1};
511 let Inst{8-6} = 0b100;
513 let Inst{3-1} = Qm{2-0};
517 multiclass MVE_VADDV_A<string suffix, bit U, bits<2> size,
518 list<dag> pattern=[]> {
519 def acc : MVE_VADDV<"vaddva", suffix,
520 (ins tGPREven:$Rda_src, MQPR:$Qm), "$Rda = $Rda_src",
521 0b1, U, size, pattern>;
522 def no_acc : MVE_VADDV<"vaddv", suffix,
524 0b0, U, size, pattern>;
527 defm MVE_VADDVs8 : MVE_VADDV_A<"s8", 0b0, 0b00>;
528 defm MVE_VADDVs16 : MVE_VADDV_A<"s16", 0b0, 0b01>;
529 defm MVE_VADDVs32 : MVE_VADDV_A<"s32", 0b0, 0b10>;
530 defm MVE_VADDVu8 : MVE_VADDV_A<"u8", 0b1, 0b00>;
531 defm MVE_VADDVu16 : MVE_VADDV_A<"u16", 0b1, 0b01>;
532 defm MVE_VADDVu32 : MVE_VADDV_A<"u32", 0b1, 0b10>;
534 class MVE_VADDLV<string iname, string suffix, dag iops, string cstr,
535 bit A, bit U, list<dag> pattern=[]>
536 : MVE_rDest<(outs tGPREven:$RdaLo, tGPROdd:$RdaHi), iops, NoItinerary, iname,
537 suffix, "$RdaLo, $RdaHi, $Qm", cstr, pattern> {
543 let Inst{22-20} = RdaHi{3-1};
544 let Inst{19-18} = 0b10;
545 let Inst{17-16} = 0b01;
546 let Inst{15-13} = RdaLo{3-1};
548 let Inst{8-6} = 0b100;
550 let Inst{3-1} = Qm{2-0};
554 multiclass MVE_VADDLV_A<string suffix, bit U, list<dag> pattern=[]> {
555 def acc : MVE_VADDLV<"vaddlva", suffix,
556 (ins tGPREven:$RdaLo_src, tGPROdd:$RdaHi_src, MQPR:$Qm),
557 "$RdaLo = $RdaLo_src,$RdaHi = $RdaHi_src",
559 def no_acc : MVE_VADDLV<"vaddlv", suffix,
565 defm MVE_VADDLVs32 : MVE_VADDLV_A<"s32", 0b0>;
566 defm MVE_VADDLVu32 : MVE_VADDLV_A<"u32", 0b1>;
568 class MVE_VMINMAXNMV<string iname, string suffix, bit sz,
569 bit bit_17, bit bit_7, list<dag> pattern=[]>
570 : MVE_rDest<(outs rGPR:$RdaDest), (ins rGPR:$RdaSrc, MQPR:$Qm),
571 NoItinerary, iname, suffix, "$RdaSrc, $Qm",
572 "$RdaDest = $RdaSrc", pattern> {
577 let Inst{22-20} = 0b110;
578 let Inst{19-18} = 0b11;
579 let Inst{17} = bit_17;
581 let Inst{15-12} = RdaDest{3-0};
584 let Inst{6-5} = 0b00;
585 let Inst{3-1} = Qm{2-0};
588 let Predicates = [HasMVEFloat];
591 multiclass MVE_VMINMAXNMV_fty<string iname, bit bit_7, list<dag> pattern=[]> {
592 def f32 : MVE_VMINMAXNMV<iname, "f32", 0b0, 0b1, bit_7, pattern>;
593 def f16 : MVE_VMINMAXNMV<iname, "f16", 0b1, 0b1, bit_7, pattern>;
596 defm MVE_VMINNMV : MVE_VMINMAXNMV_fty<"vminnmv", 0b1>;
597 defm MVE_VMAXNMV : MVE_VMINMAXNMV_fty<"vmaxnmv", 0b0>;
599 multiclass MVE_VMINMAXNMAV_fty<string iname, bit bit_7, list<dag> pattern=[]> {
600 def f32 : MVE_VMINMAXNMV<iname, "f32", 0b0, 0b0, bit_7, pattern>;
601 def f16 : MVE_VMINMAXNMV<iname, "f16", 0b1, 0b0, bit_7, pattern>;
604 defm MVE_VMINNMAV : MVE_VMINMAXNMAV_fty<"vminnmav", 0b1>;
605 defm MVE_VMAXNMAV : MVE_VMINMAXNMAV_fty<"vmaxnmav", 0b0>;
607 class MVE_VMINMAXV<string iname, string suffix, bit U, bits<2> size,
608 bit bit_17, bit bit_7, list<dag> pattern=[]>
609 : MVE_rDest<(outs rGPR:$RdaDest), (ins rGPR:$RdaSrc, MQPR:$Qm), NoItinerary,
610 iname, suffix, "$RdaSrc, $Qm", "$RdaDest = $RdaSrc", pattern> {
615 let Inst{22-20} = 0b110;
616 let Inst{19-18} = size{1-0};
617 let Inst{17} = bit_17;
619 let Inst{15-12} = RdaDest{3-0};
622 let Inst{6-5} = 0b00;
623 let Inst{3-1} = Qm{2-0};
627 multiclass MVE_VMINMAXV_ty<string iname, bit bit_7, list<dag> pattern=[]> {
628 def s8 : MVE_VMINMAXV<iname, "s8", 0b0, 0b00, 0b1, bit_7>;
629 def s16 : MVE_VMINMAXV<iname, "s16", 0b0, 0b01, 0b1, bit_7>;
630 def s32 : MVE_VMINMAXV<iname, "s32", 0b0, 0b10, 0b1, bit_7>;
631 def u8 : MVE_VMINMAXV<iname, "u8", 0b1, 0b00, 0b1, bit_7>;
632 def u16 : MVE_VMINMAXV<iname, "u16", 0b1, 0b01, 0b1, bit_7>;
633 def u32 : MVE_VMINMAXV<iname, "u32", 0b1, 0b10, 0b1, bit_7>;
636 defm MVE_VMINV : MVE_VMINMAXV_ty<"vminv", 0b1>;
637 defm MVE_VMAXV : MVE_VMINMAXV_ty<"vmaxv", 0b0>;
639 multiclass MVE_VMINMAXAV_ty<string iname, bit bit_7, list<dag> pattern=[]> {
640 def s8 : MVE_VMINMAXV<iname, "s8", 0b0, 0b00, 0b0, bit_7>;
641 def s16 : MVE_VMINMAXV<iname, "s16", 0b0, 0b01, 0b0, bit_7>;
642 def s32 : MVE_VMINMAXV<iname, "s32", 0b0, 0b10, 0b0, bit_7>;
645 defm MVE_VMINAV : MVE_VMINMAXAV_ty<"vminav", 0b1>;
646 defm MVE_VMAXAV : MVE_VMINMAXAV_ty<"vmaxav", 0b0>;
648 class MVE_VMLAMLSDAV<string iname, string suffix, dag iops, string cstr,
649 bit sz, bit bit_28, bit A, bit X, bit bit_8, bit bit_0,
650 list<dag> pattern=[]>
651 : MVE_rDest<(outs tGPREven:$RdaDest), iops, NoItinerary, iname, suffix,
652 "$RdaDest, $Qn, $Qm", cstr, pattern> {
657 let Inst{28} = bit_28;
658 let Inst{22-20} = 0b111;
659 let Inst{19-17} = Qn{2-0};
661 let Inst{15-13} = RdaDest{3-1};
664 let Inst{7-6} = 0b00;
666 let Inst{3-1} = Qm{2-0};
670 multiclass MVE_VMLAMLSDAV_X<string iname, string suffix, dag iops, string cstr,
671 bit sz, bit bit_28, bit A, bit bit_8, bit bit_0,
672 list<dag> pattern=[]> {
673 def _noexch : MVE_VMLAMLSDAV<iname, suffix, iops, cstr, sz,
674 bit_28, A, 0b0, bit_8, bit_0, pattern>;
675 def _exch : MVE_VMLAMLSDAV<iname # "x", suffix, iops, cstr, sz,
676 bit_28, A, 0b1, bit_8, bit_0, pattern>;
679 multiclass MVE_VMLAMLSDAV_XA<string iname, string suffix, bit sz, bit bit_28,
680 bit bit_8, bit bit_0, list<dag> pattern=[]> {
681 defm _noacc : MVE_VMLAMLSDAV_X<iname, suffix, (ins MQPR:$Qn, MQPR:$Qm), "",
682 sz, bit_28, 0b0, bit_8, bit_0, pattern>;
683 defm _acc : MVE_VMLAMLSDAV_X<iname # "a", suffix,
684 (ins tGPREven:$RdaSrc, MQPR:$Qn, MQPR:$Qm),
685 "$RdaDest = $RdaSrc",
686 sz, bit_28, 0b1, bit_8, bit_0, pattern>;
689 multiclass MVE_VMLADAV_multi<string suffix, bit sz, bit U, bit bit_8,
690 list<dag> pattern=[]> {
691 defm "" : MVE_VMLAMLSDAV_XA<"vmladav", suffix, sz, U, bit_8, 0b0, pattern>;
694 defm MVE_VMLADAVs16 : MVE_VMLADAV_multi<"s16", 0b0, 0b0, 0b0>;
695 defm MVE_VMLADAVs32 : MVE_VMLADAV_multi<"s32", 0b1, 0b0, 0b0>;
696 defm MVE_VMLADAVu16 : MVE_VMLADAV_multi<"u16", 0b0, 0b1, 0b0>;
697 defm MVE_VMLADAVu32 : MVE_VMLADAV_multi<"u32", 0b1, 0b1, 0b0>;
699 defm MVE_VMLADAVs8 : MVE_VMLADAV_multi<"s8", 0b0, 0b0, 0b1>;
700 defm MVE_VMLADAVu8 : MVE_VMLADAV_multi<"u8", 0b0, 0b1, 0b1>;
702 // vmlav aliases vmladav
703 foreach acc = ["_acc", "_noacc"] in {
704 foreach suffix = ["s8", "s16", "s32", "u8", "u16", "u32"] in {
705 def : MVEInstAlias<!strconcat("vmlav", !if(!eq(acc, "_acc"), "a", ""),
706 "${vp}.", suffix, "\t$RdaDest, $Qn, $Qm"),
707 (!cast<Instruction>("MVE_VMLADAV"#suffix#acc#"_noexch")
708 tGPREven:$RdaDest, MQPR:$Qn, MQPR:$Qm, vpred_n:$vp)>;
712 multiclass MVE_VMLSDAV_multi<string suffix, bit sz, bit bit_28,
713 list<dag> pattern=[]> {
714 defm "" : MVE_VMLAMLSDAV_XA<"vmlsdav", suffix, sz, bit_28, 0b0, 0b1, pattern>;
717 defm MVE_VMLSDAVs8 : MVE_VMLSDAV_multi<"s8", 0, 0b1>;
718 defm MVE_VMLSDAVs16 : MVE_VMLSDAV_multi<"s16", 0, 0b0>;
719 defm MVE_VMLSDAVs32 : MVE_VMLSDAV_multi<"s32", 1, 0b0>;
721 // Base class for VMLALDAV and VMLSLDAV, VRMLALDAVH, VRMLSLDAVH
722 class MVE_VMLALDAVBase<string iname, string suffix, dag iops, string cstr,
723 bit sz, bit bit_28, bit A, bit X, bit bit_8, bit bit_0,
724 list<dag> pattern=[]>
725 : MVE_rDest<(outs tGPREven:$RdaLoDest, tGPROdd:$RdaHiDest), iops, NoItinerary,
726 iname, suffix, "$RdaLoDest, $RdaHiDest, $Qn, $Qm", cstr, pattern> {
732 let Inst{28} = bit_28;
733 let Inst{22-20} = RdaHiDest{3-1};
734 let Inst{19-17} = Qn{2-0};
736 let Inst{15-13} = RdaLoDest{3-1};
739 let Inst{7-6} = 0b00;
741 let Inst{3-1} = Qm{2-0};
745 multiclass MVE_VMLALDAVBase_X<string iname, string suffix, dag iops,
746 string cstr, bit sz, bit bit_28, bit A,
747 bit bit_8, bit bit_0, list<dag> pattern=[]> {
748 def _noexch : MVE_VMLALDAVBase<iname, suffix, iops, cstr, sz,
749 bit_28, A, 0b0, bit_8, bit_0, pattern>;
750 def _exch : MVE_VMLALDAVBase<iname # "x", suffix, iops, cstr, sz,
751 bit_28, A, 0b1, bit_8, bit_0, pattern>;
754 multiclass MVE_VMLALDAVBase_XA<string iname, string suffix, bit sz, bit bit_28,
755 bit bit_8, bit bit_0, list<dag> pattern=[]> {
756 defm _noacc : MVE_VMLALDAVBase_X<
757 iname, suffix, (ins MQPR:$Qn, MQPR:$Qm), "",
758 sz, bit_28, 0b0, bit_8, bit_0, pattern>;
759 defm _acc : MVE_VMLALDAVBase_X<
760 iname # "a", suffix, (ins tGPREven:$RdaLoSrc, tGPROdd:$RdaHiSrc,
762 "$RdaLoDest = $RdaLoSrc,$RdaHiDest = $RdaHiSrc",
763 sz, bit_28, 0b1, bit_8, bit_0, pattern>;
766 multiclass MVE_VRMLALDAVH_multi<string suffix, bit U, list<dag> pattern=[]> {
767 defm "" : MVE_VMLALDAVBase_XA<
768 "vrmlaldavh", suffix, 0b0, U, 0b1, 0b0, pattern>;
771 defm MVE_VRMLALDAVHs32 : MVE_VRMLALDAVH_multi<"s32", 0>;
772 defm MVE_VRMLALDAVHu32 : MVE_VRMLALDAVH_multi<"u32", 1>;
774 // vrmlalvh aliases for vrmlaldavh
775 def : MVEInstAlias<"vrmlalvh${vp}.s32\t$RdaLo, $RdaHi, $Qn, $Qm",
776 (MVE_VRMLALDAVHs32_noacc_noexch
777 tGPREven:$RdaLo, tGPROdd:$RdaHi,
778 MQPR:$Qn, MQPR:$Qm, vpred_n:$vp)>;
779 def : MVEInstAlias<"vrmlalvha${vp}.s32\t$RdaLo, $RdaHi, $Qn, $Qm",
780 (MVE_VRMLALDAVHs32_acc_noexch
781 tGPREven:$RdaLo, tGPROdd:$RdaHi,
782 MQPR:$Qn, MQPR:$Qm, vpred_n:$vp)>;
783 def : MVEInstAlias<"vrmlalvh${vp}.u32\t$RdaLo, $RdaHi, $Qn, $Qm",
784 (MVE_VRMLALDAVHu32_noacc_noexch
785 tGPREven:$RdaLo, tGPROdd:$RdaHi,
786 MQPR:$Qn, MQPR:$Qm, vpred_n:$vp)>;
787 def : MVEInstAlias<"vrmlalvha${vp}.u32\t$RdaLo, $RdaHi, $Qn, $Qm",
788 (MVE_VRMLALDAVHu32_acc_noexch
789 tGPREven:$RdaLo, tGPROdd:$RdaHi,
790 MQPR:$Qn, MQPR:$Qm, vpred_n:$vp)>;
792 multiclass MVE_VMLALDAV_multi<string suffix, bit sz, bit U,
793 list<dag> pattern=[]> {
794 defm "" : MVE_VMLALDAVBase_XA<"vmlaldav", suffix, sz, U, 0b0, 0b0, pattern>;
797 defm MVE_VMLALDAVs16 : MVE_VMLALDAV_multi<"s16", 0b0, 0b0>;
798 defm MVE_VMLALDAVs32 : MVE_VMLALDAV_multi<"s32", 0b1, 0b0>;
799 defm MVE_VMLALDAVu16 : MVE_VMLALDAV_multi<"u16", 0b0, 0b1>;
800 defm MVE_VMLALDAVu32 : MVE_VMLALDAV_multi<"u32", 0b1, 0b1>;
802 // vmlalv aliases vmlaldav
803 foreach acc = ["_acc", "_noacc"] in {
804 foreach suffix = ["s16", "s32", "u16", "u32"] in {
805 def : MVEInstAlias<!strconcat("vmlalv", !if(!eq(acc, "_acc"), "a", ""),
806 "${vp}.", suffix, "\t$RdaLoDest, $RdaHiDest, $Qn, $Qm"),
807 (!cast<Instruction>("MVE_VMLALDAV"#suffix#acc#"_noexch")
808 tGPREven:$RdaLoDest, tGPROdd:$RdaHiDest,
809 MQPR:$Qn, MQPR:$Qm, vpred_n:$vp)>;
813 multiclass MVE_VMLSLDAV_multi<string iname, string suffix, bit sz,
814 bit bit_28, list<dag> pattern=[]> {
815 defm "" : MVE_VMLALDAVBase_XA<iname, suffix, sz, bit_28, 0b0, 0b1, pattern>;
818 defm MVE_VMLSLDAVs16 : MVE_VMLSLDAV_multi<"vmlsldav", "s16", 0b0, 0b0>;
819 defm MVE_VMLSLDAVs32 : MVE_VMLSLDAV_multi<"vmlsldav", "s32", 0b1, 0b0>;
820 defm MVE_VRMLSLDAVHs32 : MVE_VMLSLDAV_multi<"vrmlsldavh", "s32", 0b0, 0b1>;
822 // end of mve_rDest instructions
824 // start of mve_comp instructions
826 class MVE_comp<InstrItinClass itin, string iname, string suffix,
827 string cstr, list<dag> pattern=[]>
828 : MVE_p<(outs MQPR:$Qd), (ins MQPR:$Qn, MQPR:$Qm), itin, iname, suffix,
829 "$Qd, $Qn, $Qm", vpred_r, cstr, pattern> {
834 let Inst{22} = Qd{3};
835 let Inst{19-17} = Qn{2-0};
837 let Inst{15-13} = Qd{2-0};
839 let Inst{10-9} = 0b11;
842 let Inst{3-1} = Qm{2-0};
846 class MVE_VMINMAXNM<string iname, string suffix, bit sz, bit bit_21,
847 list<dag> pattern=[]>
848 : MVE_comp<NoItinerary, iname, suffix, "", pattern> {
851 let Inst{25-24} = 0b11;
853 let Inst{21} = bit_21;
860 let Predicates = [HasMVEFloat];
863 def MVE_VMAXNMf32 : MVE_VMINMAXNM<"vmaxnm", "f32", 0b0, 0b0>;
864 def MVE_VMAXNMf16 : MVE_VMINMAXNM<"vmaxnm", "f16", 0b1, 0b0>;
866 let Predicates = [HasMVEFloat] in {
867 def : Pat<(v4f32 (fmaxnum (v4f32 MQPR:$val1), (v4f32 MQPR:$val2))),
868 (v4f32 (MVE_VMAXNMf32 (v4f32 MQPR:$val1), (v4f32 MQPR:$val2)))>;
869 def : Pat<(v8f16 (fmaxnum (v8f16 MQPR:$val1), (v8f16 MQPR:$val2))),
870 (v8f16 (MVE_VMAXNMf16 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2)))>;
873 def MVE_VMINNMf32 : MVE_VMINMAXNM<"vminnm", "f32", 0b0, 0b1>;
874 def MVE_VMINNMf16 : MVE_VMINMAXNM<"vminnm", "f16", 0b1, 0b1>;
876 let Predicates = [HasMVEFloat] in {
877 def : Pat<(v4f32 (fminnum (v4f32 MQPR:$val1), (v4f32 MQPR:$val2))),
878 (v4f32 (MVE_VMINNMf32 (v4f32 MQPR:$val1), (v4f32 MQPR:$val2)))>;
879 def : Pat<(v8f16 (fminnum (v8f16 MQPR:$val1), (v8f16 MQPR:$val2))),
880 (v8f16 (MVE_VMINNMf16 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2)))>;
884 class MVE_VMINMAX<string iname, string suffix, bit U, bits<2> size,
885 bit bit_4, list<dag> pattern=[]>
886 : MVE_comp<NoItinerary, iname, suffix, "", pattern> {
889 let Inst{25-24} = 0b11;
891 let Inst{21-20} = size{1-0};
898 multiclass MVE_VMINMAX_all_sizes<string iname, bit bit_4> {
899 def s8 : MVE_VMINMAX<iname, "s8", 0b0, 0b00, bit_4>;
900 def s16 : MVE_VMINMAX<iname, "s16", 0b0, 0b01, bit_4>;
901 def s32 : MVE_VMINMAX<iname, "s32", 0b0, 0b10, bit_4>;
902 def u8 : MVE_VMINMAX<iname, "u8", 0b1, 0b00, bit_4>;
903 def u16 : MVE_VMINMAX<iname, "u16", 0b1, 0b01, bit_4>;
904 def u32 : MVE_VMINMAX<iname, "u32", 0b1, 0b10, bit_4>;
907 defm MVE_VMAX : MVE_VMINMAX_all_sizes<"vmax", 0b0>;
908 defm MVE_VMIN : MVE_VMINMAX_all_sizes<"vmin", 0b1>;
910 let Predicates = [HasMVEInt] in {
911 def : Pat<(v16i8 (smin (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
912 (v16i8 (MVE_VMINs8 (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
913 def : Pat<(v8i16 (smin (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
914 (v8i16 (MVE_VMINs16 (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
915 def : Pat<(v4i32 (smin (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
916 (v4i32 (MVE_VMINs32 (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
918 def : Pat<(v16i8 (smax (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
919 (v16i8 (MVE_VMAXs8 (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
920 def : Pat<(v8i16 (smax (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
921 (v8i16 (MVE_VMAXs16 (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
922 def : Pat<(v4i32 (smax (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
923 (v4i32 (MVE_VMAXs32 (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
925 def : Pat<(v16i8 (umin (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
926 (v16i8 (MVE_VMINu8 (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
927 def : Pat<(v8i16 (umin (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
928 (v8i16 (MVE_VMINu16 (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
929 def : Pat<(v4i32 (umin (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
930 (v4i32 (MVE_VMINu32 (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
932 def : Pat<(v16i8 (umax (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
933 (v16i8 (MVE_VMAXu8 (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
934 def : Pat<(v8i16 (umax (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
935 (v8i16 (MVE_VMAXu16 (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
936 def : Pat<(v4i32 (umax (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
937 (v4i32 (MVE_VMAXu32 (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
940 // end of mve_comp instructions
942 // start of mve_bit instructions
944 class MVE_bit_arith<dag oops, dag iops, string iname, string suffix,
945 string ops, string cstr, list<dag> pattern=[]>
946 : MVE_p<oops, iops, NoItinerary, iname, suffix, ops, vpred_r, cstr, pattern> {
950 let Inst{22} = Qd{3};
951 let Inst{15-13} = Qd{2-0};
953 let Inst{3-1} = Qm{2-0};
956 def MVE_VBIC : MVE_bit_arith<(outs MQPR:$Qd), (ins MQPR:$Qn, MQPR:$Qm),
957 "vbic", "", "$Qd, $Qn, $Qm", ""> {
961 let Inst{25-23} = 0b110;
962 let Inst{21-20} = 0b01;
963 let Inst{19-17} = Qn{2-0};
965 let Inst{12-8} = 0b00001;
972 class MVE_VREV<string iname, string suffix, bits<2> size, bits<2> bit_8_7>
973 : MVE_bit_arith<(outs MQPR:$Qd), (ins MQPR:$Qm), iname,
974 suffix, "$Qd, $Qm", ""> {
977 let Inst{25-23} = 0b111;
978 let Inst{21-20} = 0b11;
979 let Inst{19-18} = size;
980 let Inst{17-16} = 0b00;
981 let Inst{12-9} = 0b0000;
982 let Inst{8-7} = bit_8_7;
988 def MVE_VREV64_8 : MVE_VREV<"vrev64", "8", 0b00, 0b00>;
989 def MVE_VREV64_16 : MVE_VREV<"vrev64", "16", 0b01, 0b00>;
990 def MVE_VREV64_32 : MVE_VREV<"vrev64", "32", 0b10, 0b00>;
992 def MVE_VREV32_8 : MVE_VREV<"vrev32", "8", 0b00, 0b01>;
993 def MVE_VREV32_16 : MVE_VREV<"vrev32", "16", 0b01, 0b01>;
995 def MVE_VREV16_8 : MVE_VREV<"vrev16", "8", 0b00, 0b10>;
997 let Predicates = [HasMVEInt] in {
998 def : Pat<(v4i32 (ARMvrev64 (v4i32 MQPR:$src))),
999 (v4i32 (MVE_VREV64_32 (v4i32 MQPR:$src)))>;
1000 def : Pat<(v8i16 (ARMvrev64 (v8i16 MQPR:$src))),
1001 (v8i16 (MVE_VREV64_16 (v8i16 MQPR:$src)))>;
1002 def : Pat<(v16i8 (ARMvrev64 (v16i8 MQPR:$src))),
1003 (v16i8 (MVE_VREV64_8 (v16i8 MQPR:$src)))>;
1005 def : Pat<(v8i16 (ARMvrev32 (v8i16 MQPR:$src))),
1006 (v8i16 (MVE_VREV32_16 (v8i16 MQPR:$src)))>;
1007 def : Pat<(v16i8 (ARMvrev32 (v16i8 MQPR:$src))),
1008 (v16i8 (MVE_VREV32_8 (v16i8 MQPR:$src)))>;
1010 def : Pat<(v16i8 (ARMvrev16 (v16i8 MQPR:$src))),
1011 (v16i8 (MVE_VREV16_8 (v16i8 MQPR:$src)))>;
1013 def : Pat<(v4f32 (ARMvrev64 (v4f32 MQPR:$src))),
1014 (v4f32 (MVE_VREV64_32 (v4f32 MQPR:$src)))>;
1015 def : Pat<(v8f16 (ARMvrev64 (v8f16 MQPR:$src))),
1016 (v8f16 (MVE_VREV64_16 (v8f16 MQPR:$src)))>;
1017 def : Pat<(v8f16 (ARMvrev32 (v8f16 MQPR:$src))),
1018 (v8f16 (MVE_VREV32_16 (v8f16 MQPR:$src)))>;
1021 def MVE_VMVN : MVE_bit_arith<(outs MQPR:$Qd), (ins MQPR:$Qm),
1022 "vmvn", "", "$Qd, $Qm", ""> {
1024 let Inst{25-23} = 0b111;
1025 let Inst{21-16} = 0b110000;
1026 let Inst{12-6} = 0b0010111;
1031 let Predicates = [HasMVEInt] in {
1032 def : Pat<(v16i8 (vnotq (v16i8 MQPR:$val1))),
1033 (v16i8 (MVE_VMVN (v16i8 MQPR:$val1)))>;
1034 def : Pat<(v8i16 (vnotq (v8i16 MQPR:$val1))),
1035 (v8i16 (MVE_VMVN (v8i16 MQPR:$val1)))>;
1036 def : Pat<(v4i32 (vnotq (v4i32 MQPR:$val1))),
1037 (v4i32 (MVE_VMVN (v4i32 MQPR:$val1)))>;
1038 def : Pat<(v2i64 (vnotq (v2i64 MQPR:$val1))),
1039 (v2i64 (MVE_VMVN (v2i64 MQPR:$val1)))>;
1042 class MVE_bit_ops<string iname, bits<2> bit_21_20, bit bit_28>
1043 : MVE_bit_arith<(outs MQPR:$Qd), (ins MQPR:$Qn, MQPR:$Qm),
1044 iname, "", "$Qd, $Qn, $Qm", ""> {
1047 let Inst{28} = bit_28;
1048 let Inst{25-23} = 0b110;
1049 let Inst{21-20} = bit_21_20;
1050 let Inst{19-17} = Qn{2-0};
1052 let Inst{12-8} = 0b00001;
1053 let Inst{7} = Qn{3};
1059 def MVE_VEOR : MVE_bit_ops<"veor", 0b00, 0b1>;
1060 def MVE_VORN : MVE_bit_ops<"vorn", 0b11, 0b0>;
1061 def MVE_VORR : MVE_bit_ops<"vorr", 0b10, 0b0>;
1062 def MVE_VAND : MVE_bit_ops<"vand", 0b00, 0b0>;
1064 // add ignored suffixes as aliases
1066 foreach s=["s8", "s16", "s32", "u8", "u16", "u32", "i8", "i16", "i32", "f16", "f32"] in {
1067 def : MVEInstAlias<"vbic${vp}." # s # "\t$QdSrc, $QnSrc, $QmSrc",
1068 (MVE_VBIC MQPR:$QdSrc, MQPR:$QnSrc, MQPR:$QmSrc, vpred_r:$vp)>;
1069 def : MVEInstAlias<"veor${vp}." # s # "\t$QdSrc, $QnSrc, $QmSrc",
1070 (MVE_VEOR MQPR:$QdSrc, MQPR:$QnSrc, MQPR:$QmSrc, vpred_r:$vp)>;
1071 def : MVEInstAlias<"vorn${vp}." # s # "\t$QdSrc, $QnSrc, $QmSrc",
1072 (MVE_VORN MQPR:$QdSrc, MQPR:$QnSrc, MQPR:$QmSrc, vpred_r:$vp)>;
1073 def : MVEInstAlias<"vorr${vp}." # s # "\t$QdSrc, $QnSrc, $QmSrc",
1074 (MVE_VORR MQPR:$QdSrc, MQPR:$QnSrc, MQPR:$QmSrc, vpred_r:$vp)>;
1075 def : MVEInstAlias<"vand${vp}." # s # "\t$QdSrc, $QnSrc, $QmSrc",
1076 (MVE_VAND MQPR:$QdSrc, MQPR:$QnSrc, MQPR:$QmSrc, vpred_r:$vp)>;
1079 let Predicates = [HasMVEInt] in {
1080 def : Pat<(v16i8 (and (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
1081 (v16i8 (MVE_VAND (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
1082 def : Pat<(v8i16 (and (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
1083 (v8i16 (MVE_VAND (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
1084 def : Pat<(v4i32 (and (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
1085 (v4i32 (MVE_VAND (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
1086 def : Pat<(v2i64 (and (v2i64 MQPR:$val1), (v2i64 MQPR:$val2))),
1087 (v2i64 (MVE_VAND (v2i64 MQPR:$val1), (v2i64 MQPR:$val2)))>;
1089 def : Pat<(v16i8 (or (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
1090 (v16i8 (MVE_VORR (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
1091 def : Pat<(v8i16 (or (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
1092 (v8i16 (MVE_VORR (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
1093 def : Pat<(v4i32 (or (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
1094 (v4i32 (MVE_VORR (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
1095 def : Pat<(v2i64 (or (v2i64 MQPR:$val1), (v2i64 MQPR:$val2))),
1096 (v2i64 (MVE_VORR (v2i64 MQPR:$val1), (v2i64 MQPR:$val2)))>;
1098 def : Pat<(v16i8 (xor (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
1099 (v16i8 (MVE_VEOR (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
1100 def : Pat<(v8i16 (xor (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
1101 (v8i16 (MVE_VEOR (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
1102 def : Pat<(v4i32 (xor (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
1103 (v4i32 (MVE_VEOR (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
1104 def : Pat<(v2i64 (xor (v2i64 MQPR:$val1), (v2i64 MQPR:$val2))),
1105 (v2i64 (MVE_VEOR (v2i64 MQPR:$val1), (v2i64 MQPR:$val2)))>;
1107 def : Pat<(v16i8 (and (v16i8 MQPR:$val1), (vnotq MQPR:$val2))),
1108 (v16i8 (MVE_VBIC (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
1109 def : Pat<(v8i16 (and (v8i16 MQPR:$val1), (vnotq MQPR:$val2))),
1110 (v8i16 (MVE_VBIC (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
1111 def : Pat<(v4i32 (and (v4i32 MQPR:$val1), (vnotq MQPR:$val2))),
1112 (v4i32 (MVE_VBIC (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
1113 def : Pat<(v2i64 (and (v2i64 MQPR:$val1), (vnotq MQPR:$val2))),
1114 (v2i64 (MVE_VBIC (v2i64 MQPR:$val1), (v2i64 MQPR:$val2)))>;
1116 def : Pat<(v16i8 (or (v16i8 MQPR:$val1), (vnotq MQPR:$val2))),
1117 (v16i8 (MVE_VORN (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
1118 def : Pat<(v8i16 (or (v8i16 MQPR:$val1), (vnotq MQPR:$val2))),
1119 (v8i16 (MVE_VORN (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
1120 def : Pat<(v4i32 (or (v4i32 MQPR:$val1), (vnotq MQPR:$val2))),
1121 (v4i32 (MVE_VORN (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
1122 def : Pat<(v2i64 (or (v2i64 MQPR:$val1), (vnotq MQPR:$val2))),
1123 (v2i64 (MVE_VORN (v2i64 MQPR:$val1), (v2i64 MQPR:$val2)))>;
1126 class MVE_bit_cmode<string iname, string suffix, bits<4> cmode, dag inOps>
1127 : MVE_p<(outs MQPR:$Qd), inOps, NoItinerary,
1128 iname, suffix, "$Qd, $imm", vpred_n, "$Qd = $Qd_src"> {
1132 let Inst{28} = imm{7};
1133 let Inst{27-23} = 0b11111;
1134 let Inst{22} = Qd{3};
1135 let Inst{21-19} = 0b000;
1136 let Inst{18-16} = imm{6-4};
1137 let Inst{15-13} = Qd{2-0};
1139 let Inst{11-8} = cmode;
1140 let Inst{7-6} = 0b01;
1142 let Inst{3-0} = imm{3-0};
1145 class MVE_VORR<string suffix, bits<4> cmode, ExpandImm imm_type>
1146 : MVE_bit_cmode<"vorr", suffix, cmode, (ins MQPR:$Qd_src, imm_type:$imm)> {
1150 def MVE_VORRIZ0v4i32 : MVE_VORR<"i32", 0b0001, expzero00>;
1151 def MVE_VORRIZ0v8i16 : MVE_VORR<"i16", 0b1001, expzero00>;
1152 def MVE_VORRIZ8v4i32 : MVE_VORR<"i32", 0b0011, expzero08>;
1153 def MVE_VORRIZ8v8i16 : MVE_VORR<"i16", 0b1011, expzero08>;
1154 def MVE_VORRIZ16v4i32 : MVE_VORR<"i32", 0b0101, expzero16>;
1155 def MVE_VORRIZ24v4i32 : MVE_VORR<"i32", 0b0111, expzero24>;
1157 def MVE_VORNIZ0v4i32 : MVEAsmPseudo<"vorn${vp}.i32\t$Qd, $imm",
1158 (ins MQPR:$Qd_src, expzero00inv32:$imm, vpred_n:$vp), (outs MQPR:$Qd)>;
1159 def MVE_VORNIZ0v8i16 : MVEAsmPseudo<"vorn${vp}.i16\t$Qd, $imm",
1160 (ins MQPR:$Qd_src, expzero00inv16:$imm, vpred_n:$vp), (outs MQPR:$Qd)>;
1161 def MVE_VORNIZ8v4i32 : MVEAsmPseudo<"vorn${vp}.i32\t$Qd, $imm",
1162 (ins MQPR:$Qd_src, expzero08inv32:$imm, vpred_n:$vp), (outs MQPR:$Qd)>;
1163 def MVE_VORNIZ8v8i16 : MVEAsmPseudo<"vorn${vp}.i16\t$Qd, $imm",
1164 (ins MQPR:$Qd_src, expzero08inv16:$imm, vpred_n:$vp), (outs MQPR:$Qd)>;
1165 def MVE_VORNIZ16v4i32 : MVEAsmPseudo<"vorn${vp}.i32\t$Qd, $imm",
1166 (ins MQPR:$Qd_src, expzero16inv32:$imm, vpred_n:$vp), (outs MQPR:$Qd)>;
1167 def MVE_VORNIZ24v4i32 : MVEAsmPseudo<"vorn${vp}.i32\t$Qd, $imm",
1168 (ins MQPR:$Qd_src, expzero24inv32:$imm, vpred_n:$vp), (outs MQPR:$Qd)>;
1170 def MVE_VMOV : MVEInstAlias<"vmov${vp}\t$Qd, $Qm",
1171 (MVE_VORR MQPR:$Qd, MQPR:$Qm, MQPR:$Qm, vpred_r:$vp)>;
1173 class MVE_VBIC<string suffix, bits<4> cmode, ExpandImm imm_type>
1174 : MVE_bit_cmode<"vbic", suffix, cmode, (ins MQPR:$Qd_src, imm_type:$imm)> {
1178 def MVE_VBICIZ0v4i32 : MVE_VBIC<"i32", 0b0001, expzero00>;
1179 def MVE_VBICIZ0v8i16 : MVE_VBIC<"i16", 0b1001, expzero00>;
1180 def MVE_VBICIZ8v4i32 : MVE_VBIC<"i32", 0b0011, expzero08>;
1181 def MVE_VBICIZ8v8i16 : MVE_VBIC<"i16", 0b1011, expzero08>;
1182 def MVE_VBICIZ16v4i32 : MVE_VBIC<"i32", 0b0101, expzero16>;
1183 def MVE_VBICIZ24v4i32 : MVE_VBIC<"i32", 0b0111, expzero24>;
1185 def MVE_VANDIZ0v4i32 : MVEAsmPseudo<"vand${vp}.i32\t$Qda, $imm",
1186 (ins MQPR:$Qda_src, expzero00inv32:$imm, vpred_n:$vp), (outs MQPR:$Qda)>;
1187 def MVE_VANDIZ0v8i16 : MVEAsmPseudo<"vand${vp}.i16\t$Qda, $imm",
1188 (ins MQPR:$Qda_src, expzero00inv16:$imm, vpred_n:$vp), (outs MQPR:$Qda)>;
1189 def MVE_VANDIZ8v4i32 : MVEAsmPseudo<"vand${vp}.i32\t$Qda, $imm",
1190 (ins MQPR:$Qda_src, expzero08inv32:$imm, vpred_n:$vp), (outs MQPR:$Qda)>;
1191 def MVE_VANDIZ8v8i16 : MVEAsmPseudo<"vand${vp}.i16\t$Qda, $imm",
1192 (ins MQPR:$Qda_src, expzero08inv16:$imm, vpred_n:$vp), (outs MQPR:$Qda)>;
1193 def MVE_VANDIZ16v4i32 : MVEAsmPseudo<"vand${vp}.i32\t$Qda, $imm",
1194 (ins MQPR:$Qda_src, expzero16inv32:$imm, vpred_n:$vp), (outs MQPR:$Qda)>;
1195 def MVE_VANDIZ24v4i32 : MVEAsmPseudo<"vand${vp}.i32\t$Qda, $imm",
1196 (ins MQPR:$Qda_src, expzero24inv32:$imm, vpred_n:$vp), (outs MQPR:$Qda)>;
1198 class MVE_VMOV_lane_direction {
1205 def MVE_VMOV_from_lane : MVE_VMOV_lane_direction {
1207 let oops = (outs rGPR:$Rt);
1208 let iops = (ins MQPR:$Qd);
1209 let ops = "$Rt, $Qd$Idx";
1212 def MVE_VMOV_to_lane : MVE_VMOV_lane_direction {
1214 let oops = (outs MQPR:$Qd);
1215 let iops = (ins MQPR:$Qd_src, rGPR:$Rt);
1216 let ops = "$Qd$Idx, $Rt";
1217 let cstr = "$Qd = $Qd_src";
1220 class MVE_VMOV_lane<string suffix, bit U, dag indexop,
1221 MVE_VMOV_lane_direction dir>
1222 : MVE_VMOV_lane_base<dir.oops, !con(dir.iops, indexop), NoItinerary,
1223 "vmov", suffix, dir.ops, dir.cstr, []> {
1227 let Inst{31-24} = 0b11101110;
1229 let Inst{20} = dir.bit_20;
1230 let Inst{19-17} = Qd{2-0};
1231 let Inst{15-12} = Rt{3-0};
1232 let Inst{11-8} = 0b1011;
1233 let Inst{7} = Qd{3};
1234 let Inst{4-0} = 0b10000;
1237 class MVE_VMOV_lane_32<MVE_VMOV_lane_direction dir>
1238 : MVE_VMOV_lane<"32", 0b0, (ins MVEVectorIndex<4>:$Idx), dir> {
1241 let Inst{6-5} = 0b00;
1242 let Inst{16} = Idx{1};
1243 let Inst{21} = Idx{0};
1245 let Predicates = [HasFPRegsV8_1M];
1248 class MVE_VMOV_lane_16<string suffix, bit U, MVE_VMOV_lane_direction dir>
1249 : MVE_VMOV_lane<suffix, U, (ins MVEVectorIndex<8>:$Idx), dir> {
1253 let Inst{16} = Idx{2};
1254 let Inst{21} = Idx{1};
1255 let Inst{6} = Idx{0};
1258 class MVE_VMOV_lane_8<string suffix, bit U, MVE_VMOV_lane_direction dir>
1259 : MVE_VMOV_lane<suffix, U, (ins MVEVectorIndex<16>:$Idx), dir> {
1262 let Inst{16} = Idx{3};
1263 let Inst{21} = Idx{2};
1264 let Inst{6} = Idx{1};
1265 let Inst{5} = Idx{0};
1268 def MVE_VMOV_from_lane_32 : MVE_VMOV_lane_32< MVE_VMOV_from_lane>;
1269 def MVE_VMOV_to_lane_32 : MVE_VMOV_lane_32< MVE_VMOV_to_lane>;
1270 def MVE_VMOV_from_lane_s16 : MVE_VMOV_lane_16<"s16", 0b0, MVE_VMOV_from_lane>;
1271 def MVE_VMOV_from_lane_u16 : MVE_VMOV_lane_16<"u16", 0b1, MVE_VMOV_from_lane>;
1272 def MVE_VMOV_to_lane_16 : MVE_VMOV_lane_16< "16", 0b0, MVE_VMOV_to_lane>;
1273 def MVE_VMOV_from_lane_s8 : MVE_VMOV_lane_8 < "s8", 0b0, MVE_VMOV_from_lane>;
1274 def MVE_VMOV_from_lane_u8 : MVE_VMOV_lane_8 < "u8", 0b1, MVE_VMOV_from_lane>;
1275 def MVE_VMOV_to_lane_8 : MVE_VMOV_lane_8 < "8", 0b0, MVE_VMOV_to_lane>;
1277 let Predicates = [HasMVEInt] in {
1278 def : Pat<(extractelt (v2f64 MQPR:$src), imm:$lane),
1279 (f64 (EXTRACT_SUBREG MQPR:$src, (DSubReg_f64_reg imm:$lane)))>;
1280 def : Pat<(insertelt (v2f64 MQPR:$src1), DPR:$src2, imm:$lane),
1281 (INSERT_SUBREG (v2f64 (COPY_TO_REGCLASS MQPR:$src1, MQPR)), DPR:$src2, (DSubReg_f64_reg imm:$lane))>;
1283 def : Pat<(extractelt (v4i32 MQPR:$src), imm:$lane),
1285 (i32 (EXTRACT_SUBREG MQPR:$src, (SSubReg_f32_reg imm:$lane))), rGPR)>;
1286 def : Pat<(insertelt (v4i32 MQPR:$src1), rGPR:$src2, imm:$lane),
1287 (MVE_VMOV_to_lane_32 MQPR:$src1, rGPR:$src2, imm:$lane)>;
1289 def : Pat<(vector_insert (v16i8 MQPR:$src1), rGPR:$src2, imm:$lane),
1290 (MVE_VMOV_to_lane_8 MQPR:$src1, rGPR:$src2, imm:$lane)>;
1291 def : Pat<(vector_insert (v8i16 MQPR:$src1), rGPR:$src2, imm:$lane),
1292 (MVE_VMOV_to_lane_16 MQPR:$src1, rGPR:$src2, imm:$lane)>;
1294 def : Pat<(ARMvgetlanes (v16i8 MQPR:$src), imm:$lane),
1295 (MVE_VMOV_from_lane_s8 MQPR:$src, imm:$lane)>;
1296 def : Pat<(ARMvgetlanes (v8i16 MQPR:$src), imm:$lane),
1297 (MVE_VMOV_from_lane_s16 MQPR:$src, imm:$lane)>;
1298 def : Pat<(ARMvgetlaneu (v16i8 MQPR:$src), imm:$lane),
1299 (MVE_VMOV_from_lane_u8 MQPR:$src, imm:$lane)>;
1300 def : Pat<(ARMvgetlaneu (v8i16 MQPR:$src), imm:$lane),
1301 (MVE_VMOV_from_lane_u16 MQPR:$src, imm:$lane)>;
1303 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
1304 (MVE_VMOV_to_lane_8 (v16i8 (IMPLICIT_DEF)), rGPR:$src, (i32 0))>;
1305 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
1306 (MVE_VMOV_to_lane_16 (v8i16 (IMPLICIT_DEF)), rGPR:$src, (i32 0))>;
1307 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
1308 (MVE_VMOV_to_lane_32 (v4i32 (IMPLICIT_DEF)), rGPR:$src, (i32 0))>;
1310 // Floating point patterns, still enabled under HasMVEInt
1311 def : Pat<(extractelt (v4f32 MQPR:$src), imm:$lane),
1312 (COPY_TO_REGCLASS (f32 (EXTRACT_SUBREG MQPR:$src, (SSubReg_f32_reg imm:$lane))), SPR)>;
1313 def : Pat<(insertelt (v4f32 MQPR:$src1), (f32 SPR:$src2), imm:$lane),
1314 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS MQPR:$src1, MQPR)), SPR:$src2, (SSubReg_f32_reg imm:$lane))>;
1316 def : Pat<(insertelt (v8f16 MQPR:$src1), HPR:$src2, imm:$lane),
1317 (MVE_VMOV_to_lane_16 MQPR:$src1, (COPY_TO_REGCLASS HPR:$src2, rGPR), imm:$lane)>;
1318 def : Pat<(extractelt (v8f16 MQPR:$src), imm:$lane),
1319 (COPY_TO_REGCLASS (MVE_VMOV_from_lane_u16 MQPR:$src, imm:$lane), HPR)>;
1321 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
1322 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
1323 def : Pat<(v4f32 (scalar_to_vector GPR:$src)),
1324 (MVE_VMOV_to_lane_32 (v4f32 (IMPLICIT_DEF)), rGPR:$src, (i32 0))>;
1325 def : Pat<(v8f16 (scalar_to_vector HPR:$src)),
1326 (INSERT_SUBREG (v8f16 (IMPLICIT_DEF)), HPR:$src, ssub_0)>;
1327 def : Pat<(v8f16 (scalar_to_vector GPR:$src)),
1328 (MVE_VMOV_to_lane_16 (v8f16 (IMPLICIT_DEF)), rGPR:$src, (i32 0))>;
1331 // end of mve_bit instructions
1333 // start of MVE Integer instructions
1335 class MVE_int<string iname, string suffix, bits<2> size, list<dag> pattern=[]>
1336 : MVE_p<(outs MQPR:$Qd), (ins MQPR:$Qn, MQPR:$Qm), NoItinerary,
1337 iname, suffix, "$Qd, $Qn, $Qm", vpred_r, "", pattern> {
1342 let Inst{22} = Qd{3};
1343 let Inst{21-20} = size;
1344 let Inst{19-17} = Qn{2-0};
1345 let Inst{15-13} = Qd{2-0};
1346 let Inst{7} = Qn{3};
1348 let Inst{5} = Qm{3};
1349 let Inst{3-1} = Qm{2-0};
1352 class MVE_VMULt1<string suffix, bits<2> size, list<dag> pattern=[]>
1353 : MVE_int<"vmul", suffix, size, pattern> {
1356 let Inst{25-23} = 0b110;
1358 let Inst{12-8} = 0b01001;
1363 def MVE_VMULt1i8 : MVE_VMULt1<"i8", 0b00>;
1364 def MVE_VMULt1i16 : MVE_VMULt1<"i16", 0b01>;
1365 def MVE_VMULt1i32 : MVE_VMULt1<"i32", 0b10>;
1367 let Predicates = [HasMVEInt] in {
1368 def : Pat<(v16i8 (mul (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
1369 (v16i8 (MVE_VMULt1i8 (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
1370 def : Pat<(v8i16 (mul (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
1371 (v8i16 (MVE_VMULt1i16 (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
1372 def : Pat<(v4i32 (mul (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
1373 (v4i32 (MVE_VMULt1i32 (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
1376 class MVE_VQxDMULH<string iname, string suffix, bits<2> size, bit rounding,
1377 list<dag> pattern=[]>
1378 : MVE_int<iname, suffix, size, pattern> {
1380 let Inst{28} = rounding;
1381 let Inst{25-23} = 0b110;
1383 let Inst{12-8} = 0b01011;
1388 class MVE_VQDMULH<string suffix, bits<2> size, list<dag> pattern=[]>
1389 : MVE_VQxDMULH<"vqdmulh", suffix, size, 0b0, pattern>;
1390 class MVE_VQRDMULH<string suffix, bits<2> size, list<dag> pattern=[]>
1391 : MVE_VQxDMULH<"vqrdmulh", suffix, size, 0b1, pattern>;
1393 def MVE_VQDMULHi8 : MVE_VQDMULH<"s8", 0b00>;
1394 def MVE_VQDMULHi16 : MVE_VQDMULH<"s16", 0b01>;
1395 def MVE_VQDMULHi32 : MVE_VQDMULH<"s32", 0b10>;
1397 def MVE_VQRDMULHi8 : MVE_VQRDMULH<"s8", 0b00>;
1398 def MVE_VQRDMULHi16 : MVE_VQRDMULH<"s16", 0b01>;
1399 def MVE_VQRDMULHi32 : MVE_VQRDMULH<"s32", 0b10>;
1401 class MVE_VADDSUB<string iname, string suffix, bits<2> size, bit subtract,
1402 list<dag> pattern=[]>
1403 : MVE_int<iname, suffix, size, pattern> {
1405 let Inst{28} = subtract;
1406 let Inst{25-23} = 0b110;
1408 let Inst{12-8} = 0b01000;
1413 class MVE_VADD<string suffix, bits<2> size, list<dag> pattern=[]>
1414 : MVE_VADDSUB<"vadd", suffix, size, 0b0, pattern>;
1415 class MVE_VSUB<string suffix, bits<2> size, list<dag> pattern=[]>
1416 : MVE_VADDSUB<"vsub", suffix, size, 0b1, pattern>;
1418 def MVE_VADDi8 : MVE_VADD<"i8", 0b00>;
1419 def MVE_VADDi16 : MVE_VADD<"i16", 0b01>;
1420 def MVE_VADDi32 : MVE_VADD<"i32", 0b10>;
1422 let Predicates = [HasMVEInt] in {
1423 def : Pat<(v16i8 (add (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
1424 (v16i8 (MVE_VADDi8 (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
1425 def : Pat<(v8i16 (add (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
1426 (v8i16 (MVE_VADDi16 (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
1427 def : Pat<(v4i32 (add (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
1428 (v4i32 (MVE_VADDi32 (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
1431 def MVE_VSUBi8 : MVE_VSUB<"i8", 0b00>;
1432 def MVE_VSUBi16 : MVE_VSUB<"i16", 0b01>;
1433 def MVE_VSUBi32 : MVE_VSUB<"i32", 0b10>;
1435 let Predicates = [HasMVEInt] in {
1436 def : Pat<(v16i8 (sub (v16i8 MQPR:$val1), (v16i8 MQPR:$val2))),
1437 (v16i8 (MVE_VSUBi8 (v16i8 MQPR:$val1), (v16i8 MQPR:$val2)))>;
1438 def : Pat<(v8i16 (sub (v8i16 MQPR:$val1), (v8i16 MQPR:$val2))),
1439 (v8i16 (MVE_VSUBi16 (v8i16 MQPR:$val1), (v8i16 MQPR:$val2)))>;
1440 def : Pat<(v4i32 (sub (v4i32 MQPR:$val1), (v4i32 MQPR:$val2))),
1441 (v4i32 (MVE_VSUBi32 (v4i32 MQPR:$val1), (v4i32 MQPR:$val2)))>;
1444 class MVE_VQADDSUB<string iname, string suffix, bit U, bit subtract,
1445 bits<2> size, list<dag> pattern=[]>
1446 : MVE_int<iname, suffix, size, pattern> {
1449 let Inst{25-23} = 0b110;
1451 let Inst{12-10} = 0b000;
1452 let Inst{9} = subtract;
1458 class MVE_VQADD<string suffix, bit U, bits<2> size, list<dag> pattern=[]>
1459 : MVE_VQADDSUB<"vqadd", suffix, U, 0b0, size, pattern>;
1460 class MVE_VQSUB<string suffix, bit U, bits<2> size, list<dag> pattern=[]>
1461 : MVE_VQADDSUB<"vqsub", suffix, U, 0b1, size, pattern>;
1463 def MVE_VQADDs8 : MVE_VQADD<"s8", 0b0, 0b00>;
1464 def MVE_VQADDs16 : MVE_VQADD<"s16", 0b0, 0b01>;
1465 def MVE_VQADDs32 : MVE_VQADD<"s32", 0b0, 0b10>;
1466 def MVE_VQADDu8 : MVE_VQADD<"u8", 0b1, 0b00>;
1467 def MVE_VQADDu16 : MVE_VQADD<"u16", 0b1, 0b01>;
1468 def MVE_VQADDu32 : MVE_VQADD<"u32", 0b1, 0b10>;
1470 def MVE_VQSUBs8 : MVE_VQSUB<"s8", 0b0, 0b00>;
1471 def MVE_VQSUBs16 : MVE_VQSUB<"s16", 0b0, 0b01>;
1472 def MVE_VQSUBs32 : MVE_VQSUB<"s32", 0b0, 0b10>;
1473 def MVE_VQSUBu8 : MVE_VQSUB<"u8", 0b1, 0b00>;
1474 def MVE_VQSUBu16 : MVE_VQSUB<"u16", 0b1, 0b01>;
1475 def MVE_VQSUBu32 : MVE_VQSUB<"u32", 0b1, 0b10>;
1477 class MVE_VABD_int<string suffix, bit U, bits<2> size, list<dag> pattern=[]>
1478 : MVE_int<"vabd", suffix, size, pattern> {
1481 let Inst{25-23} = 0b110;
1483 let Inst{12-8} = 0b00111;
1488 def MVE_VABDs8 : MVE_VABD_int<"s8", 0b0, 0b00>;
1489 def MVE_VABDs16 : MVE_VABD_int<"s16", 0b0, 0b01>;
1490 def MVE_VABDs32 : MVE_VABD_int<"s32", 0b0, 0b10>;
1491 def MVE_VABDu8 : MVE_VABD_int<"u8", 0b1, 0b00>;
1492 def MVE_VABDu16 : MVE_VABD_int<"u16", 0b1, 0b01>;
1493 def MVE_VABDu32 : MVE_VABD_int<"u32", 0b1, 0b10>;
1495 class MVE_VRHADD<string suffix, bit U, bits<2> size, list<dag> pattern=[]>
1496 : MVE_int<"vrhadd", suffix, size, pattern> {
1499 let Inst{25-23} = 0b110;
1501 let Inst{12-8} = 0b00001;
1506 def MVE_VRHADDs8 : MVE_VRHADD<"s8", 0b0, 0b00>;
1507 def MVE_VRHADDs16 : MVE_VRHADD<"s16", 0b0, 0b01>;
1508 def MVE_VRHADDs32 : MVE_VRHADD<"s32", 0b0, 0b10>;
1509 def MVE_VRHADDu8 : MVE_VRHADD<"u8", 0b1, 0b00>;
1510 def MVE_VRHADDu16 : MVE_VRHADD<"u16", 0b1, 0b01>;
1511 def MVE_VRHADDu32 : MVE_VRHADD<"u32", 0b1, 0b10>;
1513 class MVE_VHADDSUB<string iname, string suffix, bit U, bit subtract,
1514 bits<2> size, list<dag> pattern=[]>
1515 : MVE_int<iname, suffix, size, pattern> {
1518 let Inst{25-23} = 0b110;
1520 let Inst{12-10} = 0b000;
1521 let Inst{9} = subtract;
1527 class MVE_VHADD<string suffix, bit U, bits<2> size,
1528 list<dag> pattern=[]>
1529 : MVE_VHADDSUB<"vhadd", suffix, U, 0b0, size, pattern>;
1530 class MVE_VHSUB<string suffix, bit U, bits<2> size,
1531 list<dag> pattern=[]>
1532 : MVE_VHADDSUB<"vhsub", suffix, U, 0b1, size, pattern>;
1534 def MVE_VHADDs8 : MVE_VHADD<"s8", 0b0, 0b00>;
1535 def MVE_VHADDs16 : MVE_VHADD<"s16", 0b0, 0b01>;
1536 def MVE_VHADDs32 : MVE_VHADD<"s32", 0b0, 0b10>;
1537 def MVE_VHADDu8 : MVE_VHADD<"u8", 0b1, 0b00>;
1538 def MVE_VHADDu16 : MVE_VHADD<"u16", 0b1, 0b01>;
1539 def MVE_VHADDu32 : MVE_VHADD<"u32", 0b1, 0b10>;
1541 def MVE_VHSUBs8 : MVE_VHSUB<"s8", 0b0, 0b00>;
1542 def MVE_VHSUBs16 : MVE_VHSUB<"s16", 0b0, 0b01>;
1543 def MVE_VHSUBs32 : MVE_VHSUB<"s32", 0b0, 0b10>;
1544 def MVE_VHSUBu8 : MVE_VHSUB<"u8", 0b1, 0b00>;
1545 def MVE_VHSUBu16 : MVE_VHSUB<"u16", 0b1, 0b01>;
1546 def MVE_VHSUBu32 : MVE_VHSUB<"u32", 0b1, 0b10>;
1548 class MVE_VDUP<string suffix, bit B, bit E, list<dag> pattern=[]>
1549 : MVE_p<(outs MQPR:$Qd), (ins rGPR:$Rt), NoItinerary,
1550 "vdup", suffix, "$Qd, $Rt", vpred_r, "", pattern> {
1555 let Inst{25-23} = 0b101;
1557 let Inst{21-20} = 0b10;
1558 let Inst{19-17} = Qd{2-0};
1560 let Inst{15-12} = Rt;
1561 let Inst{11-8} = 0b1011;
1562 let Inst{7} = Qd{3};
1565 let Inst{4-0} = 0b10000;
1568 def MVE_VDUP32 : MVE_VDUP<"32", 0b0, 0b0>;
1569 def MVE_VDUP16 : MVE_VDUP<"16", 0b0, 0b1>;
1570 def MVE_VDUP8 : MVE_VDUP<"8", 0b1, 0b0>;
1572 let Predicates = [HasMVEInt] in {
1573 def : Pat<(v16i8 (ARMvdup (i32 rGPR:$elem))),
1574 (MVE_VDUP8 rGPR:$elem)>;
1575 def : Pat<(v8i16 (ARMvdup (i32 rGPR:$elem))),
1576 (MVE_VDUP16 rGPR:$elem)>;
1577 def : Pat<(v4i32 (ARMvdup (i32 rGPR:$elem))),
1578 (MVE_VDUP32 rGPR:$elem)>;
1580 def : Pat<(v4i32 (ARMvduplane (v4i32 MQPR:$src), imm:$lane)),
1581 (MVE_VDUP32 (MVE_VMOV_from_lane_32 MQPR:$src, imm:$lane))>;
1582 // For the 16-bit and 8-bit vduplanes we don't care about the signedness
1583 // of the lane move operation as we only want the lowest 8/16 bits anyway.
1584 def : Pat<(v8i16 (ARMvduplane (v8i16 MQPR:$src), imm:$lane)),
1585 (MVE_VDUP16 (MVE_VMOV_from_lane_u16 MQPR:$src, imm:$lane))>;
1586 def : Pat<(v16i8 (ARMvduplane (v16i8 MQPR:$src), imm:$lane)),
1587 (MVE_VDUP8 (MVE_VMOV_from_lane_u8 MQPR:$src, imm:$lane))>;
1589 def : Pat<(v4f32 (ARMvdup (f32 SPR:$elem))),
1590 (v4f32 (MVE_VDUP32 (i32 (COPY_TO_REGCLASS (f32 SPR:$elem), rGPR))))>;
1591 def : Pat<(v8f16 (ARMvdup (f16 HPR:$elem))),
1592 (v8f16 (MVE_VDUP16 (i32 (COPY_TO_REGCLASS (f16 HPR:$elem), rGPR))))>;
1594 def : Pat<(v4f32 (ARMvduplane (v4f32 MQPR:$src), imm:$lane)),
1595 (MVE_VDUP32 (MVE_VMOV_from_lane_32 MQPR:$src, imm:$lane))>;
1596 def : Pat<(v8f16 (ARMvduplane (v8f16 MQPR:$src), imm:$lane)),
1597 (MVE_VDUP16 (MVE_VMOV_from_lane_u16 MQPR:$src, imm:$lane))>;
1601 class MVEIntSingleSrc<string iname, string suffix, bits<2> size,
1602 list<dag> pattern=[]>
1603 : MVE_p<(outs MQPR:$Qd), (ins MQPR:$Qm), NoItinerary,
1604 iname, suffix, "$Qd, $Qm", vpred_r, "", pattern> {
1608 let Inst{22} = Qd{3};
1609 let Inst{19-18} = size{1-0};
1610 let Inst{15-13} = Qd{2-0};
1611 let Inst{5} = Qm{3};
1612 let Inst{3-1} = Qm{2-0};
1615 class MVE_VCLSCLZ<string iname, string suffix, bits<2> size,
1616 bit count_zeroes, list<dag> pattern=[]>
1617 : MVEIntSingleSrc<iname, suffix, size, pattern> {
1620 let Inst{25-23} = 0b111;
1621 let Inst{21-20} = 0b11;
1622 let Inst{17-16} = 0b00;
1623 let Inst{12-8} = 0b00100;
1624 let Inst{7} = count_zeroes;
1630 def MVE_VCLSs8 : MVE_VCLSCLZ<"vcls", "s8", 0b00, 0b0>;
1631 def MVE_VCLSs16 : MVE_VCLSCLZ<"vcls", "s16", 0b01, 0b0>;
1632 def MVE_VCLSs32 : MVE_VCLSCLZ<"vcls", "s32", 0b10, 0b0>;
1634 def MVE_VCLZs8 : MVE_VCLSCLZ<"vclz", "i8", 0b00, 0b1>;
1635 def MVE_VCLZs16 : MVE_VCLSCLZ<"vclz", "i16", 0b01, 0b1>;
1636 def MVE_VCLZs32 : MVE_VCLSCLZ<"vclz", "i32", 0b10, 0b1>;
1638 class MVE_VABSNEG_int<string iname, string suffix, bits<2> size, bit negate,
1639 list<dag> pattern=[]>
1640 : MVEIntSingleSrc<iname, suffix, size, pattern> {
1643 let Inst{25-23} = 0b111;
1644 let Inst{21-20} = 0b11;
1645 let Inst{17-16} = 0b01;
1646 let Inst{12-8} = 0b00011;
1647 let Inst{7} = negate;
1653 def MVE_VABSs8 : MVE_VABSNEG_int<"vabs", "s8", 0b00, 0b0>;
1654 def MVE_VABSs16 : MVE_VABSNEG_int<"vabs", "s16", 0b01, 0b0>;
1655 def MVE_VABSs32 : MVE_VABSNEG_int<"vabs", "s32", 0b10, 0b0>;
1657 let Predicates = [HasMVEInt] in {
1658 def : Pat<(v16i8 (abs (v16i8 MQPR:$v))),
1659 (v16i8 (MVE_VABSs8 $v))>;
1660 def : Pat<(v8i16 (abs (v8i16 MQPR:$v))),
1661 (v8i16 (MVE_VABSs16 $v))>;
1662 def : Pat<(v4i32 (abs (v4i32 MQPR:$v))),
1663 (v4i32 (MVE_VABSs32 $v))>;
1666 def MVE_VNEGs8 : MVE_VABSNEG_int<"vneg", "s8", 0b00, 0b1>;
1667 def MVE_VNEGs16 : MVE_VABSNEG_int<"vneg", "s16", 0b01, 0b1>;
1668 def MVE_VNEGs32 : MVE_VABSNEG_int<"vneg", "s32", 0b10, 0b1>;
1670 let Predicates = [HasMVEInt] in {
1671 def : Pat<(v16i8 (vnegq (v16i8 MQPR:$v))),
1672 (v16i8 (MVE_VNEGs8 $v))>;
1673 def : Pat<(v8i16 (vnegq (v8i16 MQPR:$v))),
1674 (v8i16 (MVE_VNEGs16 $v))>;
1675 def : Pat<(v4i32 (vnegq (v4i32 MQPR:$v))),
1676 (v4i32 (MVE_VNEGs32 $v))>;
1679 class MVE_VQABSNEG<string iname, string suffix, bits<2> size,
1680 bit negate, list<dag> pattern=[]>
1681 : MVEIntSingleSrc<iname, suffix, size, pattern> {
1684 let Inst{25-23} = 0b111;
1685 let Inst{21-20} = 0b11;
1686 let Inst{17-16} = 0b00;
1687 let Inst{12-8} = 0b00111;
1688 let Inst{7} = negate;
1694 def MVE_VQABSs8 : MVE_VQABSNEG<"vqabs", "s8", 0b00, 0b0>;
1695 def MVE_VQABSs16 : MVE_VQABSNEG<"vqabs", "s16", 0b01, 0b0>;
1696 def MVE_VQABSs32 : MVE_VQABSNEG<"vqabs", "s32", 0b10, 0b0>;
1698 def MVE_VQNEGs8 : MVE_VQABSNEG<"vqneg", "s8", 0b00, 0b1>;
1699 def MVE_VQNEGs16 : MVE_VQABSNEG<"vqneg", "s16", 0b01, 0b1>;
1700 def MVE_VQNEGs32 : MVE_VQABSNEG<"vqneg", "s32", 0b10, 0b1>;
1702 class MVE_mod_imm<string iname, string suffix, bits<4> cmode, bit op,
1703 dag iops, list<dag> pattern=[]>
1704 : MVE_p<(outs MQPR:$Qd), iops, NoItinerary, iname, suffix, "$Qd, $imm",
1705 vpred_r, "", pattern> {
1709 let Inst{28} = imm{7};
1710 let Inst{25-23} = 0b111;
1711 let Inst{22} = Qd{3};
1712 let Inst{21-19} = 0b000;
1713 let Inst{18-16} = imm{6-4};
1714 let Inst{15-13} = Qd{2-0};
1716 let Inst{11-8} = cmode{3-0};
1717 let Inst{7-6} = 0b01;
1720 let Inst{3-0} = imm{3-0};
1722 let DecoderMethod = "DecodeMVEModImmInstruction";
1725 let isReMaterializable = 1 in {
1726 let isAsCheapAsAMove = 1 in {
1727 def MVE_VMOVimmi8 : MVE_mod_imm<"vmov", "i8", {1,1,1,0}, 0b0, (ins nImmSplatI8:$imm)>;
1728 def MVE_VMOVimmi16 : MVE_mod_imm<"vmov", "i16", {1,0,?,0}, 0b0, (ins nImmSplatI16:$imm)> {
1729 let Inst{9} = imm{9};
1731 def MVE_VMOVimmi32 : MVE_mod_imm<"vmov", "i32", {?,?,?,?}, 0b0, (ins nImmVMOVI32:$imm)> {
1732 let Inst{11-8} = imm{11-8};
1734 def MVE_VMOVimmi64 : MVE_mod_imm<"vmov", "i64", {1,1,1,0}, 0b1, (ins nImmSplatI64:$imm)>;
1735 def MVE_VMOVimmf32 : MVE_mod_imm<"vmov", "f32", {1,1,1,1}, 0b0, (ins nImmVMOVF32:$imm)>;
1736 } // let isAsCheapAsAMove = 1
1738 def MVE_VMVNimmi16 : MVE_mod_imm<"vmvn", "i16", {1,0,?,0}, 0b1, (ins nImmSplatI16:$imm)> {
1739 let Inst{9} = imm{9};
1741 def MVE_VMVNimmi32 : MVE_mod_imm<"vmvn", "i32", {?,?,?,?}, 0b1, (ins nImmVMOVI32:$imm)> {
1742 let Inst{11-8} = imm{11-8};
1744 } // let isReMaterializable = 1
1746 let Predicates = [HasMVEInt] in {
1747 def : Pat<(v16i8 (ARMvmovImm timm:$simm)),
1748 (v16i8 (MVE_VMOVimmi8 nImmSplatI8:$simm))>;
1749 def : Pat<(v8i16 (ARMvmovImm timm:$simm)),
1750 (v8i16 (MVE_VMOVimmi16 nImmSplatI16:$simm))>;
1751 def : Pat<(v4i32 (ARMvmovImm timm:$simm)),
1752 (v4i32 (MVE_VMOVimmi32 nImmVMOVI32:$simm))>;
1754 def : Pat<(v8i16 (ARMvmvnImm timm:$simm)),
1755 (v8i16 (MVE_VMVNimmi16 nImmSplatI16:$simm))>;
1756 def : Pat<(v4i32 (ARMvmvnImm timm:$simm)),
1757 (v4i32 (MVE_VMVNimmi32 nImmVMOVI32:$simm))>;
1759 def : Pat<(v4f32 (ARMvmovFPImm timm:$simm)),
1760 (v4f32 (MVE_VMOVimmf32 nImmVMOVF32:$simm))>;
1763 class MVE_VMINMAXA<string iname, string suffix, bits<2> size,
1764 bit bit_12, list<dag> pattern=[]>
1765 : MVE_p<(outs MQPR:$Qd), (ins MQPR:$Qd_src, MQPR:$Qm),
1766 NoItinerary, iname, suffix, "$Qd, $Qm", vpred_n, "$Qd = $Qd_src",
1772 let Inst{25-23} = 0b100;
1773 let Inst{22} = Qd{3};
1774 let Inst{21-20} = 0b11;
1775 let Inst{19-18} = size;
1776 let Inst{17-16} = 0b11;
1777 let Inst{15-13} = Qd{2-0};
1778 let Inst{12} = bit_12;
1779 let Inst{11-6} = 0b111010;
1780 let Inst{5} = Qm{3};
1782 let Inst{3-1} = Qm{2-0};
1786 def MVE_VMAXAs8 : MVE_VMINMAXA<"vmaxa", "s8", 0b00, 0b0>;
1787 def MVE_VMAXAs16 : MVE_VMINMAXA<"vmaxa", "s16", 0b01, 0b0>;
1788 def MVE_VMAXAs32 : MVE_VMINMAXA<"vmaxa", "s32", 0b10, 0b0>;
1790 def MVE_VMINAs8 : MVE_VMINMAXA<"vmina", "s8", 0b00, 0b1>;
1791 def MVE_VMINAs16 : MVE_VMINMAXA<"vmina", "s16", 0b01, 0b1>;
1792 def MVE_VMINAs32 : MVE_VMINMAXA<"vmina", "s32", 0b10, 0b1>;
1794 // end of MVE Integer instructions
1796 // start of mve_imm_shift instructions
1798 def MVE_VSHLC : MVE_p<(outs rGPR:$RdmDest, MQPR:$Qd),
1799 (ins MQPR:$QdSrc, rGPR:$RdmSrc, long_shift:$imm),
1800 NoItinerary, "vshlc", "", "$QdSrc, $RdmSrc, $imm",
1801 vpred_n, "$RdmDest = $RdmSrc,$Qd = $QdSrc"> {
1807 let Inst{25-23} = 0b101;
1808 let Inst{22} = Qd{3};
1810 let Inst{20-16} = imm{4-0};
1811 let Inst{15-13} = Qd{2-0};
1812 let Inst{12-4} = 0b011111100;
1813 let Inst{3-0} = RdmDest{3-0};
1816 class MVE_shift_imm<dag oops, dag iops, string iname, string suffix,
1817 string ops, vpred_ops vpred, string cstr,
1818 list<dag> pattern=[]>
1819 : MVE_p<oops, iops, NoItinerary, iname, suffix, ops, vpred, cstr, pattern> {
1823 let Inst{22} = Qd{3};
1824 let Inst{15-13} = Qd{2-0};
1825 let Inst{5} = Qm{3};
1826 let Inst{3-1} = Qm{2-0};
1829 class MVE_VMOVL<string iname, string suffix, bits<2> sz, bit U,
1830 list<dag> pattern=[]>
1831 : MVE_shift_imm<(outs MQPR:$Qd), (ins MQPR:$Qm),
1832 iname, suffix, "$Qd, $Qm", vpred_r, "",
1835 let Inst{25-23} = 0b101;
1837 let Inst{20-19} = sz{1-0};
1838 let Inst{18-16} = 0b000;
1839 let Inst{11-6} = 0b111101;
1844 multiclass MVE_VMOVL_shift_half<string iname, string suffix, bits<2> sz, bit U,
1845 list<dag> pattern=[]> {
1846 def bh : MVE_VMOVL<!strconcat(iname, "b"), suffix, sz, U, pattern> {
1849 def th : MVE_VMOVL<!strconcat(iname, "t"), suffix, sz, U, pattern> {
1854 defm MVE_VMOVLs8 : MVE_VMOVL_shift_half<"vmovl", "s8", 0b01, 0b0>;
1855 defm MVE_VMOVLu8 : MVE_VMOVL_shift_half<"vmovl", "u8", 0b01, 0b1>;
1856 defm MVE_VMOVLs16 : MVE_VMOVL_shift_half<"vmovl", "s16", 0b10, 0b0>;
1857 defm MVE_VMOVLu16 : MVE_VMOVL_shift_half<"vmovl", "u16", 0b10, 0b1>;
1859 let Predicates = [HasMVEInt] in {
1860 def : Pat<(sext_inreg (v4i32 MQPR:$src), v4i16),
1861 (MVE_VMOVLs16bh MQPR:$src)>;
1862 def : Pat<(sext_inreg (v8i16 MQPR:$src), v8i8),
1863 (MVE_VMOVLs8bh MQPR:$src)>;
1864 def : Pat<(sext_inreg (v4i32 MQPR:$src), v4i8),
1865 (MVE_VMOVLs16bh (MVE_VMOVLs8bh MQPR:$src))>;
1867 // zext_inreg 16 -> 32
1868 def : Pat<(and (v4i32 MQPR:$src), (v4i32 (ARMvmovImm (i32 0xCFF)))),
1869 (MVE_VMOVLu16bh MQPR:$src)>;
1870 // zext_inreg 8 -> 16
1871 def : Pat<(and (v8i16 MQPR:$src), (v8i16 (ARMvmovImm (i32 0x8FF)))),
1872 (MVE_VMOVLu8bh MQPR:$src)>;
1876 class MVE_VSHLL_imm<string iname, string suffix, bit U, bit th,
1877 dag immops, list<dag> pattern=[]>
1878 : MVE_shift_imm<(outs MQPR:$Qd), !con((ins MQPR:$Qm), immops),
1879 iname, suffix, "$Qd, $Qm, $imm", vpred_r, "", pattern> {
1881 let Inst{25-23} = 0b101;
1884 let Inst{11-6} = 0b111101;
1889 // The immediate VSHLL instructions accept shift counts from 1 up to
1890 // the lane width (8 or 16), but the full-width shifts have an
1891 // entirely separate encoding, given below with 'lw' in the name.
1893 class MVE_VSHLL_imm8<string iname, string suffix,
1894 bit U, bit th, list<dag> pattern=[]>
1895 : MVE_VSHLL_imm<iname, suffix, U, th, (ins mve_shift_imm1_7:$imm), pattern> {
1897 let Inst{20-19} = 0b01;
1898 let Inst{18-16} = imm;
1901 class MVE_VSHLL_imm16<string iname, string suffix,
1902 bit U, bit th, list<dag> pattern=[]>
1903 : MVE_VSHLL_imm<iname, suffix, U, th, (ins mve_shift_imm1_15:$imm), pattern> {
1906 let Inst{19-16} = imm;
1909 def MVE_VSHLL_imms8bh : MVE_VSHLL_imm8 <"vshllb", "s8", 0b0, 0b0>;
1910 def MVE_VSHLL_imms8th : MVE_VSHLL_imm8 <"vshllt", "s8", 0b0, 0b1>;
1911 def MVE_VSHLL_immu8bh : MVE_VSHLL_imm8 <"vshllb", "u8", 0b1, 0b0>;
1912 def MVE_VSHLL_immu8th : MVE_VSHLL_imm8 <"vshllt", "u8", 0b1, 0b1>;
1913 def MVE_VSHLL_imms16bh : MVE_VSHLL_imm16<"vshllb", "s16", 0b0, 0b0>;
1914 def MVE_VSHLL_imms16th : MVE_VSHLL_imm16<"vshllt", "s16", 0b0, 0b1>;
1915 def MVE_VSHLL_immu16bh : MVE_VSHLL_imm16<"vshllb", "u16", 0b1, 0b0>;
1916 def MVE_VSHLL_immu16th : MVE_VSHLL_imm16<"vshllt", "u16", 0b1, 0b1>;
1918 class MVE_VSHLL_by_lane_width<string iname, string suffix, bits<2> size,
1919 bit U, string ops, list<dag> pattern=[]>
1920 : MVE_shift_imm<(outs MQPR:$Qd), (ins MQPR:$Qm),
1921 iname, suffix, ops, vpred_r, "", pattern> {
1923 let Inst{25-23} = 0b100;
1924 let Inst{21-20} = 0b11;
1925 let Inst{19-18} = size{1-0};
1926 let Inst{17-16} = 0b01;
1927 let Inst{11-6} = 0b111000;
1932 multiclass MVE_VSHLL_lw<string iname, string suffix, bits<2> sz, bit U,
1933 string ops, list<dag> pattern=[]> {
1934 def bh : MVE_VSHLL_by_lane_width<iname#"b", suffix, sz, U, ops, pattern> {
1937 def th : MVE_VSHLL_by_lane_width<iname#"t", suffix, sz, U, ops, pattern> {
1942 defm MVE_VSHLL_lws8 : MVE_VSHLL_lw<"vshll", "s8", 0b00, 0b0, "$Qd, $Qm, #8">;
1943 defm MVE_VSHLL_lws16 : MVE_VSHLL_lw<"vshll", "s16", 0b01, 0b0, "$Qd, $Qm, #16">;
1944 defm MVE_VSHLL_lwu8 : MVE_VSHLL_lw<"vshll", "u8", 0b00, 0b1, "$Qd, $Qm, #8">;
1945 defm MVE_VSHLL_lwu16 : MVE_VSHLL_lw<"vshll", "u16", 0b01, 0b1, "$Qd, $Qm, #16">;
1947 class MVE_VxSHRN<string iname, string suffix, bit bit_12, bit bit_28,
1948 dag immops, list<dag> pattern=[]>
1949 : MVE_shift_imm<(outs MQPR:$Qd), !con((ins MQPR:$QdSrc, MQPR:$Qm), immops),
1950 iname, suffix, "$Qd, $Qm, $imm", vpred_n, "$Qd = $QdSrc",
1954 let Inst{28} = bit_28;
1955 let Inst{25-23} = 0b101;
1957 let Inst{20-16} = imm{4-0};
1958 let Inst{12} = bit_12;
1959 let Inst{11-6} = 0b111111;
1964 def MVE_VRSHRNi16bh : MVE_VxSHRN<
1965 "vrshrnb", "i16", 0b0, 0b1, (ins shr_imm8:$imm)> {
1966 let Inst{20-19} = 0b01;
1968 def MVE_VRSHRNi16th : MVE_VxSHRN<
1969 "vrshrnt", "i16", 0b1, 0b1,(ins shr_imm8:$imm)> {
1970 let Inst{20-19} = 0b01;
1972 def MVE_VRSHRNi32bh : MVE_VxSHRN<
1973 "vrshrnb", "i32", 0b0, 0b1, (ins shr_imm16:$imm)> {
1976 def MVE_VRSHRNi32th : MVE_VxSHRN<
1977 "vrshrnt", "i32", 0b1, 0b1, (ins shr_imm16:$imm)> {
1981 def MVE_VSHRNi16bh : MVE_VxSHRN<
1982 "vshrnb", "i16", 0b0, 0b0, (ins shr_imm8:$imm)> {
1983 let Inst{20-19} = 0b01;
1985 def MVE_VSHRNi16th : MVE_VxSHRN<
1986 "vshrnt", "i16", 0b1, 0b0, (ins shr_imm8:$imm)> {
1987 let Inst{20-19} = 0b01;
1989 def MVE_VSHRNi32bh : MVE_VxSHRN<
1990 "vshrnb", "i32", 0b0, 0b0, (ins shr_imm16:$imm)> {
1993 def MVE_VSHRNi32th : MVE_VxSHRN<
1994 "vshrnt", "i32", 0b1, 0b0, (ins shr_imm16:$imm)> {
1998 class MVE_VxQRSHRUN<string iname, string suffix, bit bit_28, bit bit_12, dag immops,
1999 list<dag> pattern=[]>
2000 : MVE_shift_imm<(outs MQPR:$Qd), !con((ins MQPR:$QdSrc, MQPR:$Qm), immops),
2001 iname, suffix, "$Qd, $Qm, $imm", vpred_n, "$Qd = $QdSrc",
2005 let Inst{28} = bit_28;
2006 let Inst{25-23} = 0b101;
2008 let Inst{20-16} = imm{4-0};
2009 let Inst{12} = bit_12;
2010 let Inst{11-6} = 0b111111;
2015 def MVE_VQRSHRUNs16bh : MVE_VxQRSHRUN<
2016 "vqrshrunb", "s16", 0b1, 0b0, (ins shr_imm8:$imm)> {
2017 let Inst{20-19} = 0b01;
2019 def MVE_VQRSHRUNs16th : MVE_VxQRSHRUN<
2020 "vqrshrunt", "s16", 0b1, 0b1, (ins shr_imm8:$imm)> {
2021 let Inst{20-19} = 0b01;
2023 def MVE_VQRSHRUNs32bh : MVE_VxQRSHRUN<
2024 "vqrshrunb", "s32", 0b1, 0b0, (ins shr_imm16:$imm)> {
2027 def MVE_VQRSHRUNs32th : MVE_VxQRSHRUN<
2028 "vqrshrunt", "s32", 0b1, 0b1, (ins shr_imm16:$imm)> {
2032 def MVE_VQSHRUNs16bh : MVE_VxQRSHRUN<
2033 "vqshrunb", "s16", 0b0, 0b0, (ins shr_imm8:$imm)> {
2034 let Inst{20-19} = 0b01;
2036 def MVE_VQSHRUNs16th : MVE_VxQRSHRUN<
2037 "vqshrunt", "s16", 0b0, 0b1, (ins shr_imm8:$imm)> {
2038 let Inst{20-19} = 0b01;
2040 def MVE_VQSHRUNs32bh : MVE_VxQRSHRUN<
2041 "vqshrunb", "s32", 0b0, 0b0, (ins shr_imm16:$imm)> {
2044 def MVE_VQSHRUNs32th : MVE_VxQRSHRUN<
2045 "vqshrunt", "s32", 0b0, 0b1, (ins shr_imm16:$imm)> {
2049 class MVE_VxQRSHRN<string iname, string suffix, bit bit_0, bit bit_12,
2050 dag immops, list<dag> pattern=[]>
2051 : MVE_shift_imm<(outs MQPR:$Qd), !con((ins MQPR:$QdSrc, MQPR:$Qm), immops),
2052 iname, suffix, "$Qd, $Qm, $imm", vpred_n, "$Qd = $QdSrc",
2056 let Inst{25-23} = 0b101;
2058 let Inst{20-16} = imm{4-0};
2059 let Inst{12} = bit_12;
2060 let Inst{11-6} = 0b111101;
2062 let Inst{0} = bit_0;
2065 multiclass MVE_VxQRSHRN_types<string iname, bit bit_0, bit bit_12> {
2066 def s16 : MVE_VxQRSHRN<iname, "s16", bit_0, bit_12, (ins shr_imm8:$imm)> {
2068 let Inst{20-19} = 0b01;
2070 def u16 : MVE_VxQRSHRN<iname, "u16", bit_0, bit_12, (ins shr_imm8:$imm)> {
2072 let Inst{20-19} = 0b01;
2074 def s32 : MVE_VxQRSHRN<iname, "s32", bit_0, bit_12, (ins shr_imm16:$imm)> {
2078 def u32 : MVE_VxQRSHRN<iname, "u32", bit_0, bit_12, (ins shr_imm16:$imm)> {
2084 defm MVE_VQRSHRNbh : MVE_VxQRSHRN_types<"vqrshrnb", 0b1, 0b0>;
2085 defm MVE_VQRSHRNth : MVE_VxQRSHRN_types<"vqrshrnt", 0b1, 0b1>;
2086 defm MVE_VQSHRNbh : MVE_VxQRSHRN_types<"vqshrnb", 0b0, 0b0>;
2087 defm MVE_VQSHRNth : MVE_VxQRSHRN_types<"vqshrnt", 0b0, 0b1>;
2089 // end of mve_imm_shift instructions
2091 // start of mve_shift instructions
2093 class MVE_shift_by_vec<string iname, string suffix, bit U,
2094 bits<2> size, bit bit_4, bit bit_8>
2095 : MVE_p<(outs MQPR:$Qd), (ins MQPR:$Qm, MQPR:$Qn), NoItinerary,
2096 iname, suffix, "$Qd, $Qm, $Qn", vpred_r, "", []> {
2097 // Shift instructions which take a vector of shift counts
2103 let Inst{25-24} = 0b11;
2105 let Inst{22} = Qd{3};
2106 let Inst{21-20} = size;
2107 let Inst{19-17} = Qn{2-0};
2109 let Inst{15-13} = Qd{2-0};
2110 let Inst{12-9} = 0b0010;
2111 let Inst{8} = bit_8;
2112 let Inst{7} = Qn{3};
2114 let Inst{5} = Qm{3};
2115 let Inst{4} = bit_4;
2116 let Inst{3-1} = Qm{2-0};
2120 multiclass mve_shift_by_vec_multi<string iname, bit bit_4, bit bit_8> {
2121 def s8 : MVE_shift_by_vec<iname, "s8", 0b0, 0b00, bit_4, bit_8>;
2122 def s16 : MVE_shift_by_vec<iname, "s16", 0b0, 0b01, bit_4, bit_8>;
2123 def s32 : MVE_shift_by_vec<iname, "s32", 0b0, 0b10, bit_4, bit_8>;
2124 def u8 : MVE_shift_by_vec<iname, "u8", 0b1, 0b00, bit_4, bit_8>;
2125 def u16 : MVE_shift_by_vec<iname, "u16", 0b1, 0b01, bit_4, bit_8>;
2126 def u32 : MVE_shift_by_vec<iname, "u32", 0b1, 0b10, bit_4, bit_8>;
2129 defm MVE_VSHL_by_vec : mve_shift_by_vec_multi<"vshl", 0b0, 0b0>;
2130 defm MVE_VQSHL_by_vec : mve_shift_by_vec_multi<"vqshl", 0b1, 0b0>;
2131 defm MVE_VQRSHL_by_vec : mve_shift_by_vec_multi<"vqrshl", 0b1, 0b1>;
2132 defm MVE_VRSHL_by_vec : mve_shift_by_vec_multi<"vrshl", 0b0, 0b1>;
2134 let Predicates = [HasMVEInt] in {
2135 def : Pat<(v4i32 (ARMvshlu (v4i32 MQPR:$Qm), (v4i32 MQPR:$Qn))),
2136 (v4i32 (MVE_VSHL_by_vecu32 (v4i32 MQPR:$Qm), (v4i32 MQPR:$Qn)))>;
2137 def : Pat<(v8i16 (ARMvshlu (v8i16 MQPR:$Qm), (v8i16 MQPR:$Qn))),
2138 (v8i16 (MVE_VSHL_by_vecu16 (v8i16 MQPR:$Qm), (v8i16 MQPR:$Qn)))>;
2139 def : Pat<(v16i8 (ARMvshlu (v16i8 MQPR:$Qm), (v16i8 MQPR:$Qn))),
2140 (v16i8 (MVE_VSHL_by_vecu8 (v16i8 MQPR:$Qm), (v16i8 MQPR:$Qn)))>;
2142 def : Pat<(v4i32 (ARMvshls (v4i32 MQPR:$Qm), (v4i32 MQPR:$Qn))),
2143 (v4i32 (MVE_VSHL_by_vecs32 (v4i32 MQPR:$Qm), (v4i32 MQPR:$Qn)))>;
2144 def : Pat<(v8i16 (ARMvshls (v8i16 MQPR:$Qm), (v8i16 MQPR:$Qn))),
2145 (v8i16 (MVE_VSHL_by_vecs16 (v8i16 MQPR:$Qm), (v8i16 MQPR:$Qn)))>;
2146 def : Pat<(v16i8 (ARMvshls (v16i8 MQPR:$Qm), (v16i8 MQPR:$Qn))),
2147 (v16i8 (MVE_VSHL_by_vecs8 (v16i8 MQPR:$Qm), (v16i8 MQPR:$Qn)))>;
2150 class MVE_shift_with_imm<string iname, string suffix, dag oops, dag iops,
2151 string ops, vpred_ops vpred, string cstr,
2152 list<dag> pattern=[]>
2153 : MVE_p<oops, iops, NoItinerary, iname, suffix, ops, vpred, cstr, pattern> {
2158 let Inst{22} = Qd{3};
2159 let Inst{15-13} = Qd{2-0};
2160 let Inst{12-11} = 0b00;
2161 let Inst{7-6} = 0b01;
2162 let Inst{5} = Qm{3};
2164 let Inst{3-1} = Qm{2-0};
2168 class MVE_VSxI_imm<string iname, string suffix, bit bit_8, dag imm>
2169 : MVE_shift_with_imm<iname, suffix, (outs MQPR:$Qd),
2170 !con((ins MQPR:$Qd_src, MQPR:$Qm), imm),
2171 "$Qd, $Qm, $imm", vpred_n, "$Qd = $Qd_src"> {
2174 let Inst{25-24} = 0b11;
2175 let Inst{21-16} = imm;
2176 let Inst{10-9} = 0b10;
2177 let Inst{8} = bit_8;
2180 def MVE_VSRIimm8 : MVE_VSxI_imm<"vsri", "8", 0b0, (ins shr_imm8:$imm)> {
2181 let Inst{21-19} = 0b001;
2184 def MVE_VSRIimm16 : MVE_VSxI_imm<"vsri", "16", 0b0, (ins shr_imm16:$imm)> {
2185 let Inst{21-20} = 0b01;
2188 def MVE_VSRIimm32 : MVE_VSxI_imm<"vsri", "32", 0b0, (ins shr_imm32:$imm)> {
2192 def MVE_VSLIimm8 : MVE_VSxI_imm<"vsli", "8", 0b1, (ins imm0_7:$imm)> {
2193 let Inst{21-19} = 0b001;
2196 def MVE_VSLIimm16 : MVE_VSxI_imm<"vsli", "16", 0b1, (ins imm0_15:$imm)> {
2197 let Inst{21-20} = 0b01;
2200 def MVE_VSLIimm32 : MVE_VSxI_imm<"vsli", "32", 0b1,(ins imm0_31:$imm)> {
2204 class MVE_VQSHL_imm<string suffix, dag imm>
2205 : MVE_shift_with_imm<"vqshl", suffix, (outs MQPR:$Qd),
2206 !con((ins MQPR:$Qm), imm), "$Qd, $Qm, $imm",
2210 let Inst{25-24} = 0b11;
2211 let Inst{21-16} = imm;
2212 let Inst{10-8} = 0b111;
2215 def MVE_VSLIimms8 : MVE_VQSHL_imm<"s8", (ins imm0_7:$imm)> {
2217 let Inst{21-19} = 0b001;
2220 def MVE_VSLIimmu8 : MVE_VQSHL_imm<"u8", (ins imm0_7:$imm)> {
2222 let Inst{21-19} = 0b001;
2225 def MVE_VSLIimms16 : MVE_VQSHL_imm<"s16", (ins imm0_15:$imm)> {
2227 let Inst{21-20} = 0b01;
2230 def MVE_VSLIimmu16 : MVE_VQSHL_imm<"u16", (ins imm0_15:$imm)> {
2232 let Inst{21-20} = 0b01;
2235 def MVE_VSLIimms32 : MVE_VQSHL_imm<"s32", (ins imm0_31:$imm)> {
2240 def MVE_VSLIimmu32 : MVE_VQSHL_imm<"u32", (ins imm0_31:$imm)> {
2245 class MVE_VQSHLU_imm<string suffix, dag imm>
2246 : MVE_shift_with_imm<"vqshlu", suffix, (outs MQPR:$Qd),
2247 !con((ins MQPR:$Qm), imm), "$Qd, $Qm, $imm",
2252 let Inst{25-24} = 0b11;
2253 let Inst{21-16} = imm;
2254 let Inst{10-8} = 0b110;
2257 def MVE_VQSHLU_imms8 : MVE_VQSHLU_imm<"s8", (ins imm0_7:$imm)> {
2258 let Inst{21-19} = 0b001;
2261 def MVE_VQSHLU_imms16 : MVE_VQSHLU_imm<"s16", (ins imm0_15:$imm)> {
2262 let Inst{21-20} = 0b01;
2265 def MVE_VQSHLU_imms32 : MVE_VQSHLU_imm<"s32", (ins imm0_31:$imm)> {
2269 class MVE_VRSHR_imm<string suffix, dag imm>
2270 : MVE_shift_with_imm<"vrshr", suffix, (outs MQPR:$Qd),
2271 !con((ins MQPR:$Qm), imm), "$Qd, $Qm, $imm",
2275 let Inst{25-24} = 0b11;
2276 let Inst{21-16} = imm;
2277 let Inst{10-8} = 0b010;
2280 def MVE_VRSHR_imms8 : MVE_VRSHR_imm<"s8", (ins shr_imm8:$imm)> {
2282 let Inst{21-19} = 0b001;
2285 def MVE_VRSHR_immu8 : MVE_VRSHR_imm<"u8", (ins shr_imm8:$imm)> {
2287 let Inst{21-19} = 0b001;
2290 def MVE_VRSHR_imms16 : MVE_VRSHR_imm<"s16", (ins shr_imm16:$imm)> {
2292 let Inst{21-20} = 0b01;
2295 def MVE_VRSHR_immu16 : MVE_VRSHR_imm<"u16", (ins shr_imm16:$imm)> {
2297 let Inst{21-20} = 0b01;
2300 def MVE_VRSHR_imms32 : MVE_VRSHR_imm<"s32", (ins shr_imm32:$imm)> {
2305 def MVE_VRSHR_immu32 : MVE_VRSHR_imm<"u32", (ins shr_imm32:$imm)> {
2310 class MVE_VSHR_imm<string suffix, dag imm>
2311 : MVE_shift_with_imm<"vshr", suffix, (outs MQPR:$Qd),
2312 !con((ins MQPR:$Qm), imm), "$Qd, $Qm, $imm",
2316 let Inst{25-24} = 0b11;
2317 let Inst{21-16} = imm;
2318 let Inst{10-8} = 0b000;
2321 def MVE_VSHR_imms8 : MVE_VSHR_imm<"s8", (ins shr_imm8:$imm)> {
2323 let Inst{21-19} = 0b001;
2326 def MVE_VSHR_immu8 : MVE_VSHR_imm<"u8", (ins shr_imm8:$imm)> {
2328 let Inst{21-19} = 0b001;
2331 def MVE_VSHR_imms16 : MVE_VSHR_imm<"s16", (ins shr_imm16:$imm)> {
2333 let Inst{21-20} = 0b01;
2336 def MVE_VSHR_immu16 : MVE_VSHR_imm<"u16", (ins shr_imm16:$imm)> {
2338 let Inst{21-20} = 0b01;
2341 def MVE_VSHR_imms32 : MVE_VSHR_imm<"s32", (ins shr_imm32:$imm)> {
2346 def MVE_VSHR_immu32 : MVE_VSHR_imm<"u32", (ins shr_imm32:$imm)> {
2351 class MVE_VSHL_imm<string suffix, dag imm>
2352 : MVE_shift_with_imm<"vshl", suffix, (outs MQPR:$Qd),
2353 !con((ins MQPR:$Qm), imm), "$Qd, $Qm, $imm",
2358 let Inst{25-24} = 0b11;
2359 let Inst{21-16} = imm;
2360 let Inst{10-8} = 0b101;
2363 def MVE_VSHL_immi8 : MVE_VSHL_imm<"i8", (ins imm0_7:$imm)> {
2364 let Inst{21-19} = 0b001;
2367 def MVE_VSHL_immi16 : MVE_VSHL_imm<"i16", (ins imm0_15:$imm)> {
2368 let Inst{21-20} = 0b01;
2371 def MVE_VSHL_immi32 : MVE_VSHL_imm<"i32", (ins imm0_31:$imm)> {
2375 let Predicates = [HasMVEInt] in {
2376 def : Pat<(v4i32 (ARMvshlImm (v4i32 MQPR:$src), imm0_31:$imm)),
2377 (v4i32 (MVE_VSHL_immi32 (v4i32 MQPR:$src), imm0_31:$imm))>;
2378 def : Pat<(v8i16 (ARMvshlImm (v8i16 MQPR:$src), imm0_15:$imm)),
2379 (v8i16 (MVE_VSHL_immi16 (v8i16 MQPR:$src), imm0_15:$imm))>;
2380 def : Pat<(v16i8 (ARMvshlImm (v16i8 MQPR:$src), imm0_7:$imm)),
2381 (v16i8 (MVE_VSHL_immi8 (v16i8 MQPR:$src), imm0_7:$imm))>;
2383 def : Pat<(v4i32 (ARMvshruImm (v4i32 MQPR:$src), imm0_31:$imm)),
2384 (v4i32 (MVE_VSHR_immu32 (v4i32 MQPR:$src), imm0_31:$imm))>;
2385 def : Pat<(v8i16 (ARMvshruImm (v8i16 MQPR:$src), imm0_15:$imm)),
2386 (v8i16 (MVE_VSHR_immu16 (v8i16 MQPR:$src), imm0_15:$imm))>;
2387 def : Pat<(v16i8 (ARMvshruImm (v16i8 MQPR:$src), imm0_7:$imm)),
2388 (v16i8 (MVE_VSHR_immu8 (v16i8 MQPR:$src), imm0_7:$imm))>;
2390 def : Pat<(v4i32 (ARMvshrsImm (v4i32 MQPR:$src), imm0_31:$imm)),
2391 (v4i32 (MVE_VSHR_imms32 (v4i32 MQPR:$src), imm0_31:$imm))>;
2392 def : Pat<(v8i16 (ARMvshrsImm (v8i16 MQPR:$src), imm0_15:$imm)),
2393 (v8i16 (MVE_VSHR_imms16 (v8i16 MQPR:$src), imm0_15:$imm))>;
2394 def : Pat<(v16i8 (ARMvshrsImm (v16i8 MQPR:$src), imm0_7:$imm)),
2395 (v16i8 (MVE_VSHR_imms8 (v16i8 MQPR:$src), imm0_7:$imm))>;
2398 // end of mve_shift instructions
2400 // start of MVE Floating Point instructions
2402 class MVE_float<string iname, string suffix, dag oops, dag iops, string ops,
2403 vpred_ops vpred, string cstr, list<dag> pattern=[]>
2404 : MVE_f<oops, iops, NoItinerary, iname, suffix, ops, vpred, cstr, pattern> {
2409 let Inst{5} = Qm{3};
2410 let Inst{3-1} = Qm{2-0};
2414 class MVE_VRINT<string rmode, bits<3> op, string suffix, bits<2> size,
2415 list<dag> pattern=[]>
2416 : MVE_float<!strconcat("vrint", rmode), suffix, (outs MQPR:$Qd),
2417 (ins MQPR:$Qm), "$Qd, $Qm", vpred_r, "", pattern> {
2421 let Inst{25-23} = 0b111;
2422 let Inst{22} = Qd{3};
2423 let Inst{21-20} = 0b11;
2424 let Inst{19-18} = size;
2425 let Inst{17-16} = 0b10;
2426 let Inst{15-13} = Qd{2-0};
2427 let Inst{11-10} = 0b01;
2428 let Inst{9-7} = op{2-0};
2433 multiclass MVE_VRINT_ops<string suffix, bits<2> size, list<dag> pattern=[]> {
2434 def N : MVE_VRINT<"n", 0b000, suffix, size, pattern>;
2435 def X : MVE_VRINT<"x", 0b001, suffix, size, pattern>;
2436 def A : MVE_VRINT<"a", 0b010, suffix, size, pattern>;
2437 def Z : MVE_VRINT<"z", 0b011, suffix, size, pattern>;
2438 def M : MVE_VRINT<"m", 0b101, suffix, size, pattern>;
2439 def P : MVE_VRINT<"p", 0b111, suffix, size, pattern>;
2442 defm MVE_VRINTf16 : MVE_VRINT_ops<"f16", 0b01>;
2443 defm MVE_VRINTf32 : MVE_VRINT_ops<"f32", 0b10>;
2445 let Predicates = [HasMVEFloat] in {
2446 def : Pat<(v4f32 (frint (v4f32 MQPR:$val1))),
2447 (v4f32 (MVE_VRINTf32X (v4f32 MQPR:$val1)))>;
2448 def : Pat<(v8f16 (frint (v8f16 MQPR:$val1))),
2449 (v8f16 (MVE_VRINTf16X (v8f16 MQPR:$val1)))>;
2450 def : Pat<(v4f32 (fround (v4f32 MQPR:$val1))),
2451 (v4f32 (MVE_VRINTf32A (v4f32 MQPR:$val1)))>;
2452 def : Pat<(v8f16 (fround (v8f16 MQPR:$val1))),
2453 (v8f16 (MVE_VRINTf16A (v8f16 MQPR:$val1)))>;
2454 def : Pat<(v4f32 (ftrunc (v4f32 MQPR:$val1))),
2455 (v4f32 (MVE_VRINTf32Z (v4f32 MQPR:$val1)))>;
2456 def : Pat<(v8f16 (ftrunc (v8f16 MQPR:$val1))),
2457 (v8f16 (MVE_VRINTf16Z (v8f16 MQPR:$val1)))>;
2458 def : Pat<(v4f32 (ffloor (v4f32 MQPR:$val1))),
2459 (v4f32 (MVE_VRINTf32M (v4f32 MQPR:$val1)))>;
2460 def : Pat<(v8f16 (ffloor (v8f16 MQPR:$val1))),
2461 (v8f16 (MVE_VRINTf16M (v8f16 MQPR:$val1)))>;
2462 def : Pat<(v4f32 (fceil (v4f32 MQPR:$val1))),
2463 (v4f32 (MVE_VRINTf32P (v4f32 MQPR:$val1)))>;
2464 def : Pat<(v8f16 (fceil (v8f16 MQPR:$val1))),
2465 (v8f16 (MVE_VRINTf16P (v8f16 MQPR:$val1)))>;
2468 class MVEFloatArithNeon<string iname, string suffix, bit size,
2469 dag oops, dag iops, string ops,
2470 vpred_ops vpred, string cstr, list<dag> pattern=[]>
2471 : MVE_float<iname, suffix, oops, iops, ops, vpred, cstr, pattern> {
2472 let Inst{20} = size;
2476 class MVE_VMUL_fp<string suffix, bit size, list<dag> pattern=[]>
2477 : MVEFloatArithNeon<"vmul", suffix, size, (outs MQPR:$Qd),
2478 (ins MQPR:$Qn, MQPR:$Qm), "$Qd, $Qn, $Qm", vpred_r, "",
2484 let Inst{25-23} = 0b110;
2485 let Inst{22} = Qd{3};
2487 let Inst{19-17} = Qn{2-0};
2488 let Inst{15-13} = Qd{2-0};
2489 let Inst{12-8} = 0b01101;
2490 let Inst{7} = Qn{3};
2494 def MVE_VMULf32 : MVE_VMUL_fp<"f32", 0b0>;
2495 def MVE_VMULf16 : MVE_VMUL_fp<"f16", 0b1>;
2497 let Predicates = [HasMVEFloat] in {
2498 def : Pat<(v4f32 (fmul (v4f32 MQPR:$val1), (v4f32 MQPR:$val2))),
2499 (v4f32 (MVE_VMULf32 (v4f32 MQPR:$val1), (v4f32 MQPR:$val2)))>;
2500 def : Pat<(v8f16 (fmul (v8f16 MQPR:$val1), (v8f16 MQPR:$val2))),
2501 (v8f16 (MVE_VMULf16 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2)))>;
2504 class MVE_VCMLA<string suffix, bit size, list<dag> pattern=[]>
2505 : MVEFloatArithNeon<"vcmla", suffix, size, (outs MQPR:$Qd),
2506 (ins MQPR:$Qd_src, MQPR:$Qn, MQPR:$Qm, complexrotateop:$rot),
2507 "$Qd, $Qn, $Qm, $rot", vpred_n, "$Qd = $Qd_src", pattern> {
2514 let Inst{24-23} = rot;
2515 let Inst{22} = Qd{3};
2517 let Inst{19-17} = Qn{2-0};
2518 let Inst{15-13} = Qd{2-0};
2519 let Inst{12-8} = 0b01000;
2520 let Inst{7} = Qn{3};
2524 def MVE_VCMLAf16 : MVE_VCMLA<"f16", 0b0>;
2525 def MVE_VCMLAf32 : MVE_VCMLA<"f32", 0b1>;
2527 class MVE_VADDSUBFMA_fp<string iname, string suffix, bit size, bit bit_4,
2528 bit bit_8, bit bit_21, dag iops=(ins),
2529 vpred_ops vpred=vpred_r, string cstr="",
2530 list<dag> pattern=[]>
2531 : MVEFloatArithNeon<iname, suffix, size, (outs MQPR:$Qd),
2532 !con(iops, (ins MQPR:$Qn, MQPR:$Qm)), "$Qd, $Qn, $Qm",
2533 vpred, cstr, pattern> {
2538 let Inst{25-23} = 0b110;
2539 let Inst{22} = Qd{3};
2540 let Inst{21} = bit_21;
2541 let Inst{19-17} = Qn{2-0};
2542 let Inst{15-13} = Qd{2-0};
2543 let Inst{11-9} = 0b110;
2544 let Inst{8} = bit_8;
2545 let Inst{7} = Qn{3};
2546 let Inst{4} = bit_4;
2549 def MVE_VFMAf32 : MVE_VADDSUBFMA_fp<"vfma", "f32", 0b0, 0b1, 0b0, 0b0,
2550 (ins MQPR:$Qd_src), vpred_n, "$Qd = $Qd_src">;
2551 def MVE_VFMAf16 : MVE_VADDSUBFMA_fp<"vfma", "f16", 0b1, 0b1, 0b0, 0b0,
2552 (ins MQPR:$Qd_src), vpred_n, "$Qd = $Qd_src">;
2554 def MVE_VFMSf32 : MVE_VADDSUBFMA_fp<"vfms", "f32", 0b0, 0b1, 0b0, 0b1,
2555 (ins MQPR:$Qd_src), vpred_n, "$Qd = $Qd_src">;
2556 def MVE_VFMSf16 : MVE_VADDSUBFMA_fp<"vfms", "f16", 0b1, 0b1, 0b0, 0b1,
2557 (ins MQPR:$Qd_src), vpred_n, "$Qd = $Qd_src">;
2559 def MVE_VADDf32 : MVE_VADDSUBFMA_fp<"vadd", "f32", 0b0, 0b0, 0b1, 0b0>;
2560 def MVE_VADDf16 : MVE_VADDSUBFMA_fp<"vadd", "f16", 0b1, 0b0, 0b1, 0b0>;
2562 let Predicates = [HasMVEFloat] in {
2563 def : Pat<(v4f32 (fadd (v4f32 MQPR:$val1), (v4f32 MQPR:$val2))),
2564 (v4f32 (MVE_VADDf32 (v4f32 MQPR:$val1), (v4f32 MQPR:$val2)))>;
2565 def : Pat<(v8f16 (fadd (v8f16 MQPR:$val1), (v8f16 MQPR:$val2))),
2566 (v8f16 (MVE_VADDf16 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2)))>;
2569 def MVE_VSUBf32 : MVE_VADDSUBFMA_fp<"vsub", "f32", 0b0, 0b0, 0b1, 0b1>;
2570 def MVE_VSUBf16 : MVE_VADDSUBFMA_fp<"vsub", "f16", 0b1, 0b0, 0b1, 0b1>;
2572 let Predicates = [HasMVEFloat] in {
2573 def : Pat<(v4f32 (fsub (v4f32 MQPR:$val1), (v4f32 MQPR:$val2))),
2574 (v4f32 (MVE_VSUBf32 (v4f32 MQPR:$val1), (v4f32 MQPR:$val2)))>;
2575 def : Pat<(v8f16 (fsub (v8f16 MQPR:$val1), (v8f16 MQPR:$val2))),
2576 (v8f16 (MVE_VSUBf16 (v8f16 MQPR:$val1), (v8f16 MQPR:$val2)))>;
2579 class MVE_VCADD<string suffix, bit size, list<dag> pattern=[]>
2580 : MVEFloatArithNeon<"vcadd", suffix, size, (outs MQPR:$Qd),
2581 (ins MQPR:$Qn, MQPR:$Qm, complexrotateopodd:$rot),
2582 "$Qd, $Qn, $Qm, $rot", vpred_r, "", pattern> {
2591 let Inst{22} = Qd{3};
2593 let Inst{19-17} = Qn{2-0};
2594 let Inst{15-13} = Qd{2-0};
2595 let Inst{12-8} = 0b01000;
2596 let Inst{7} = Qn{3};
2600 def MVE_VCADDf16 : MVE_VCADD<"f16", 0b0>;
2601 def MVE_VCADDf32 : MVE_VCADD<"f32", 0b1>;
2603 class MVE_VABD_fp<string suffix, bit size>
2604 : MVE_float<"vabd", suffix, (outs MQPR:$Qd), (ins MQPR:$Qn, MQPR:$Qm),
2605 "$Qd, $Qn, $Qm", vpred_r, ""> {
2610 let Inst{25-23} = 0b110;
2611 let Inst{22} = Qd{3};
2613 let Inst{20} = size;
2614 let Inst{19-17} = Qn{2-0};
2616 let Inst{15-13} = Qd{2-0};
2617 let Inst{11-8} = 0b1101;
2618 let Inst{7} = Qn{3};
2622 def MVE_VABDf32 : MVE_VABD_fp<"f32", 0b0>;
2623 def MVE_VABDf16 : MVE_VABD_fp<"f16", 0b1>;
2625 class MVE_VCVT_fix<string suffix, bit fsi, bit U, bit op,
2626 Operand imm_operand_type, list<dag> pattern=[]>
2627 : MVE_float<"vcvt", suffix,
2628 (outs MQPR:$Qd), (ins MQPR:$Qm, imm_operand_type:$imm6),
2629 "$Qd, $Qm, $imm6", vpred_r, "", pattern> {
2634 let Inst{25-23} = 0b111;
2635 let Inst{22} = Qd{3};
2637 let Inst{19-16} = imm6{3-0};
2638 let Inst{15-13} = Qd{2-0};
2639 let Inst{11-10} = 0b11;
2645 let DecoderMethod = "DecodeMVEVCVTt1fp";
2648 class MVE_VCVT_imm_asmop<int Bits> : AsmOperandClass {
2649 let PredicateMethod = "isImmediate<1," # Bits # ">";
2650 let DiagnosticString =
2651 "MVE fixed-point immediate operand must be between 1 and " # Bits;
2652 let Name = "MVEVcvtImm" # Bits;
2653 let RenderMethod = "addImmOperands";
2655 class MVE_VCVT_imm<int Bits>: Operand<i32> {
2656 let ParserMatchClass = MVE_VCVT_imm_asmop<Bits>;
2657 let EncoderMethod = "getNEONVcvtImm32OpValue";
2658 let DecoderMethod = "DecodeVCVTImmOperand";
2661 class MVE_VCVT_fix_f32<string suffix, bit U, bit op>
2662 : MVE_VCVT_fix<suffix, 0b1, U, op, MVE_VCVT_imm<32>> {
2663 let Inst{20} = imm6{4};
2665 class MVE_VCVT_fix_f16<string suffix, bit U, bit op>
2666 : MVE_VCVT_fix<suffix, 0b0, U, op, MVE_VCVT_imm<16>> {
2670 def MVE_VCVTf16s16_fix : MVE_VCVT_fix_f16<"f16.s16", 0b0, 0b0>;
2671 def MVE_VCVTs16f16_fix : MVE_VCVT_fix_f16<"s16.f16", 0b0, 0b1>;
2672 def MVE_VCVTf16u16_fix : MVE_VCVT_fix_f16<"f16.u16", 0b1, 0b0>;
2673 def MVE_VCVTu16f16_fix : MVE_VCVT_fix_f16<"u16.f16", 0b1, 0b1>;
2674 def MVE_VCVTf32s32_fix : MVE_VCVT_fix_f32<"f32.s32", 0b0, 0b0>;
2675 def MVE_VCVTs32f32_fix : MVE_VCVT_fix_f32<"s32.f32", 0b0, 0b1>;
2676 def MVE_VCVTf32u32_fix : MVE_VCVT_fix_f32<"f32.u32", 0b1, 0b0>;
2677 def MVE_VCVTu32f32_fix : MVE_VCVT_fix_f32<"u32.f32", 0b1, 0b1>;
2679 class MVE_VCVT_fp_int_anpm<string suffix, bits<2> size, bit op, string anpm,
2680 bits<2> rm, list<dag> pattern=[]>
2681 : MVE_float<!strconcat("vcvt", anpm), suffix, (outs MQPR:$Qd),
2682 (ins MQPR:$Qm), "$Qd, $Qm", vpred_r, "", pattern> {
2686 let Inst{25-23} = 0b111;
2687 let Inst{22} = Qd{3};
2688 let Inst{21-20} = 0b11;
2689 let Inst{19-18} = size;
2690 let Inst{17-16} = 0b11;
2691 let Inst{15-13} = Qd{2-0};
2692 let Inst{12-10} = 0b000;
2698 multiclass MVE_VCVT_fp_int_anpm_multi<string suffix, bits<2> size, bit op,
2699 list<dag> pattern=[]> {
2700 def a : MVE_VCVT_fp_int_anpm<suffix, size, op, "a", 0b00>;
2701 def n : MVE_VCVT_fp_int_anpm<suffix, size, op, "n", 0b01>;
2702 def p : MVE_VCVT_fp_int_anpm<suffix, size, op, "p", 0b10>;
2703 def m : MVE_VCVT_fp_int_anpm<suffix, size, op, "m", 0b11>;
2706 // This defines instructions such as MVE_VCVTu16f16a, with an explicit
2707 // rounding-mode suffix on the mnemonic. The class below will define
2708 // the bare MVE_VCVTu16f16 (with implied rounding toward zero).
2709 defm MVE_VCVTs16f16 : MVE_VCVT_fp_int_anpm_multi<"s16.f16", 0b01, 0b0>;
2710 defm MVE_VCVTu16f16 : MVE_VCVT_fp_int_anpm_multi<"u16.f16", 0b01, 0b1>;
2711 defm MVE_VCVTs32f32 : MVE_VCVT_fp_int_anpm_multi<"s32.f32", 0b10, 0b0>;
2712 defm MVE_VCVTu32f32 : MVE_VCVT_fp_int_anpm_multi<"u32.f32", 0b10, 0b1>;
2714 class MVE_VCVT_fp_int<string suffix, bits<2> size, bits<2> op,
2715 list<dag> pattern=[]>
2716 : MVE_float<"vcvt", suffix, (outs MQPR:$Qd),
2717 (ins MQPR:$Qm), "$Qd, $Qm", vpred_r, "", pattern> {
2721 let Inst{25-23} = 0b111;
2722 let Inst{22} = Qd{3};
2723 let Inst{21-20} = 0b11;
2724 let Inst{19-18} = size;
2725 let Inst{17-16} = 0b11;
2726 let Inst{15-13} = Qd{2-0};
2727 let Inst{12-9} = 0b0011;
2732 // The unsuffixed VCVT for float->int implicitly rounds toward zero,
2733 // which I reflect here in the llvm instruction names
2734 def MVE_VCVTs16f16z : MVE_VCVT_fp_int<"s16.f16", 0b01, 0b10>;
2735 def MVE_VCVTu16f16z : MVE_VCVT_fp_int<"u16.f16", 0b01, 0b11>;
2736 def MVE_VCVTs32f32z : MVE_VCVT_fp_int<"s32.f32", 0b10, 0b10>;
2737 def MVE_VCVTu32f32z : MVE_VCVT_fp_int<"u32.f32", 0b10, 0b11>;
2738 // Whereas VCVT for int->float rounds to nearest
2739 def MVE_VCVTf16s16n : MVE_VCVT_fp_int<"f16.s16", 0b01, 0b00>;
2740 def MVE_VCVTf16u16n : MVE_VCVT_fp_int<"f16.u16", 0b01, 0b01>;
2741 def MVE_VCVTf32s32n : MVE_VCVT_fp_int<"f32.s32", 0b10, 0b00>;
2742 def MVE_VCVTf32u32n : MVE_VCVT_fp_int<"f32.u32", 0b10, 0b01>;
2744 let Predicates = [HasMVEFloat] in {
2745 def : Pat<(v4i32 (fp_to_sint (v4f32 MQPR:$src))),
2746 (v4i32 (MVE_VCVTs32f32z (v4f32 MQPR:$src)))>;
2747 def : Pat<(v4i32 (fp_to_uint (v4f32 MQPR:$src))),
2748 (v4i32 (MVE_VCVTu32f32z (v4f32 MQPR:$src)))>;
2749 def : Pat<(v8i16 (fp_to_sint (v8f16 MQPR:$src))),
2750 (v8i16 (MVE_VCVTs16f16z (v8f16 MQPR:$src)))>;
2751 def : Pat<(v8i16 (fp_to_uint (v8f16 MQPR:$src))),
2752 (v8i16 (MVE_VCVTu16f16z (v8f16 MQPR:$src)))>;
2753 def : Pat<(v4f32 (sint_to_fp (v4i32 MQPR:$src))),
2754 (v4f32 (MVE_VCVTf32s32n (v4i32 MQPR:$src)))>;
2755 def : Pat<(v4f32 (uint_to_fp (v4i32 MQPR:$src))),
2756 (v4f32 (MVE_VCVTf32u32n (v4i32 MQPR:$src)))>;
2757 def : Pat<(v8f16 (sint_to_fp (v8i16 MQPR:$src))),
2758 (v8f16 (MVE_VCVTf16s16n (v8i16 MQPR:$src)))>;
2759 def : Pat<(v8f16 (uint_to_fp (v8i16 MQPR:$src))),
2760 (v8f16 (MVE_VCVTf16u16n (v8i16 MQPR:$src)))>;
2763 class MVE_VABSNEG_fp<string iname, string suffix, bits<2> size, bit negate,
2764 list<dag> pattern=[]>
2765 : MVE_float<iname, suffix, (outs MQPR:$Qd),
2766 (ins MQPR:$Qm), "$Qd, $Qm", vpred_r, "", pattern> {
2770 let Inst{25-23} = 0b111;
2771 let Inst{22} = Qd{3};
2772 let Inst{21-20} = 0b11;
2773 let Inst{19-18} = size;
2774 let Inst{17-16} = 0b01;
2775 let Inst{15-13} = Qd{2-0};
2776 let Inst{11-8} = 0b0111;
2777 let Inst{7} = negate;
2781 def MVE_VABSf16 : MVE_VABSNEG_fp<"vabs", "f16", 0b01, 0b0>;
2782 def MVE_VABSf32 : MVE_VABSNEG_fp<"vabs", "f32", 0b10, 0b0>;
2784 let Predicates = [HasMVEFloat] in {
2785 def : Pat<(v8f16 (fabs MQPR:$src)),
2786 (MVE_VABSf16 MQPR:$src)>;
2787 def : Pat<(v4f32 (fabs MQPR:$src)),
2788 (MVE_VABSf32 MQPR:$src)>;
2791 def MVE_VNEGf16 : MVE_VABSNEG_fp<"vneg", "f16", 0b01, 0b1>;
2792 def MVE_VNEGf32 : MVE_VABSNEG_fp<"vneg", "f32", 0b10, 0b1>;
2794 let Predicates = [HasMVEFloat] in {
2795 def : Pat<(v8f16 (fneg MQPR:$src)),
2796 (MVE_VNEGf16 MQPR:$src)>;
2797 def : Pat<(v4f32 (fneg MQPR:$src)),
2798 (MVE_VNEGf32 MQPR:$src)>;
2801 class MVE_VMAXMINNMA<string iname, string suffix, bit size, bit bit_12,
2802 list<dag> pattern=[]>
2803 : MVE_f<(outs MQPR:$Qd), (ins MQPR:$Qd_src, MQPR:$Qm),
2804 NoItinerary, iname, suffix, "$Qd, $Qm", vpred_n, "$Qd = $Qd_src",
2809 let Inst{28} = size;
2810 let Inst{25-23} = 0b100;
2811 let Inst{22} = Qd{3};
2812 let Inst{21-16} = 0b111111;
2813 let Inst{15-13} = Qd{2-0};
2814 let Inst{12} = bit_12;
2815 let Inst{11-6} = 0b111010;
2816 let Inst{5} = Qm{3};
2818 let Inst{3-1} = Qm{2-0};
2822 def MVE_VMAXNMAf32 : MVE_VMAXMINNMA<"vmaxnma", "f32", 0b0, 0b0>;
2823 def MVE_VMAXNMAf16 : MVE_VMAXMINNMA<"vmaxnma", "f16", 0b1, 0b0>;
2825 def MVE_VMINNMAf32 : MVE_VMAXMINNMA<"vminnma", "f32", 0b0, 0b1>;
2826 def MVE_VMINNMAf16 : MVE_VMAXMINNMA<"vminnma", "f16", 0b1, 0b1>;
2828 // end of MVE Floating Point instructions
2830 // start of MVE compares
2832 class MVE_VCMPqq<string suffix, bit bit_28, bits<2> bits_21_20,
2833 VCMPPredicateOperand predtype, list<dag> pattern=[]>
2834 : MVE_p<(outs VCCR:$P0), (ins MQPR:$Qn, MQPR:$Qm, predtype:$fc),
2835 NoItinerary, "vcmp", suffix, "$fc, $Qn, $Qm", vpred_n, "", pattern> {
2836 // Base class for comparing two vector registers
2841 let Inst{28} = bit_28;
2842 let Inst{25-22} = 0b1000;
2843 let Inst{21-20} = bits_21_20;
2844 let Inst{19-17} = Qn{2-0};
2845 let Inst{16-13} = 0b1000;
2846 let Inst{12} = fc{2};
2847 let Inst{11-8} = 0b1111;
2848 let Inst{7} = fc{0};
2850 let Inst{5} = Qm{3};
2852 let Inst{3-1} = Qm{2-0};
2853 let Inst{0} = fc{1};
2855 let Constraints = "";
2857 // We need a custom decoder method for these instructions because of
2858 // the output VCCR operand, which isn't encoded in the instruction
2859 // bits anywhere (there is only one choice for it) but has to be
2860 // included in the MC operands so that codegen will be able to track
2861 // its data flow between instructions, spill/reload it when
2862 // necessary, etc. There seems to be no way to get the Tablegen
2863 // decoder to emit an operand that isn't affected by any instruction
2865 let DecoderMethod = "DecodeMVEVCMP<false," # predtype.DecoderMethod # ">";
2868 class MVE_VCMPqqf<string suffix, bit size>
2869 : MVE_VCMPqq<suffix, size, 0b11, pred_basic_fp> {
2870 let Predicates = [HasMVEFloat];
2873 class MVE_VCMPqqi<string suffix, bits<2> size>
2874 : MVE_VCMPqq<suffix, 0b1, size, pred_basic_i> {
2879 class MVE_VCMPqqu<string suffix, bits<2> size>
2880 : MVE_VCMPqq<suffix, 0b1, size, pred_basic_u> {
2885 class MVE_VCMPqqs<string suffix, bits<2> size>
2886 : MVE_VCMPqq<suffix, 0b1, size, pred_basic_s> {
2890 def MVE_VCMPf32 : MVE_VCMPqqf<"f32", 0b0>;
2891 def MVE_VCMPf16 : MVE_VCMPqqf<"f16", 0b1>;
2893 def MVE_VCMPi8 : MVE_VCMPqqi<"i8", 0b00>;
2894 def MVE_VCMPi16 : MVE_VCMPqqi<"i16", 0b01>;
2895 def MVE_VCMPi32 : MVE_VCMPqqi<"i32", 0b10>;
2897 def MVE_VCMPu8 : MVE_VCMPqqu<"u8", 0b00>;
2898 def MVE_VCMPu16 : MVE_VCMPqqu<"u16", 0b01>;
2899 def MVE_VCMPu32 : MVE_VCMPqqu<"u32", 0b10>;
2901 def MVE_VCMPs8 : MVE_VCMPqqs<"s8", 0b00>;
2902 def MVE_VCMPs16 : MVE_VCMPqqs<"s16", 0b01>;
2903 def MVE_VCMPs32 : MVE_VCMPqqs<"s32", 0b10>;
2905 class MVE_VCMPqr<string suffix, bit bit_28, bits<2> bits_21_20,
2906 VCMPPredicateOperand predtype, list<dag> pattern=[]>
2907 : MVE_p<(outs VCCR:$P0), (ins MQPR:$Qn, GPRwithZR:$Rm, predtype:$fc),
2908 NoItinerary, "vcmp", suffix, "$fc, $Qn, $Rm", vpred_n, "", pattern> {
2909 // Base class for comparing a vector register with a scalar
2914 let Inst{28} = bit_28;
2915 let Inst{25-22} = 0b1000;
2916 let Inst{21-20} = bits_21_20;
2917 let Inst{19-17} = Qn{2-0};
2918 let Inst{16-13} = 0b1000;
2919 let Inst{12} = fc{2};
2920 let Inst{11-8} = 0b1111;
2921 let Inst{7} = fc{0};
2923 let Inst{5} = fc{1};
2925 let Inst{3-0} = Rm{3-0};
2927 let Constraints = "";
2928 // Custom decoder method, for the same reason as MVE_VCMPqq
2929 let DecoderMethod = "DecodeMVEVCMP<true," # predtype.DecoderMethod # ">";
2932 class MVE_VCMPqrf<string suffix, bit size>
2933 : MVE_VCMPqr<suffix, size, 0b11, pred_basic_fp> {
2934 let Predicates = [HasMVEFloat];
2937 class MVE_VCMPqri<string suffix, bits<2> size>
2938 : MVE_VCMPqr<suffix, 0b1, size, pred_basic_i> {
2943 class MVE_VCMPqru<string suffix, bits<2> size>
2944 : MVE_VCMPqr<suffix, 0b1, size, pred_basic_u> {
2949 class MVE_VCMPqrs<string suffix, bits<2> size>
2950 : MVE_VCMPqr<suffix, 0b1, size, pred_basic_s> {
2954 def MVE_VCMPf32r : MVE_VCMPqrf<"f32", 0b0>;
2955 def MVE_VCMPf16r : MVE_VCMPqrf<"f16", 0b1>;
2957 def MVE_VCMPi8r : MVE_VCMPqri<"i8", 0b00>;
2958 def MVE_VCMPi16r : MVE_VCMPqri<"i16", 0b01>;
2959 def MVE_VCMPi32r : MVE_VCMPqri<"i32", 0b10>;
2961 def MVE_VCMPu8r : MVE_VCMPqru<"u8", 0b00>;
2962 def MVE_VCMPu16r : MVE_VCMPqru<"u16", 0b01>;
2963 def MVE_VCMPu32r : MVE_VCMPqru<"u32", 0b10>;
2965 def MVE_VCMPs8r : MVE_VCMPqrs<"s8", 0b00>;
2966 def MVE_VCMPs16r : MVE_VCMPqrs<"s16", 0b01>;
2967 def MVE_VCMPs32r : MVE_VCMPqrs<"s32", 0b10>;
2969 // end of MVE compares
2971 // start of MVE_qDest_qSrc
2973 class MVE_qDest_qSrc<string iname, string suffix, dag oops, dag iops,
2974 string ops, vpred_ops vpred, string cstr,
2975 list<dag> pattern=[]>
2976 : MVE_p<oops, iops, NoItinerary, iname, suffix,
2977 ops, vpred, cstr, pattern> {
2981 let Inst{25-23} = 0b100;
2982 let Inst{22} = Qd{3};
2983 let Inst{15-13} = Qd{2-0};
2984 let Inst{11-9} = 0b111;
2986 let Inst{5} = Qm{3};
2988 let Inst{3-1} = Qm{2-0};
2991 class MVE_VQxDMLxDH<string iname, bit exch, bit round, bit subtract,
2992 string suffix, bits<2> size, list<dag> pattern=[]>
2993 : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd),
2994 (ins MQPR:$Qd_src, MQPR:$Qn, MQPR:$Qm), "$Qd, $Qn, $Qm",
2995 vpred_n, "$Qd = $Qd_src", pattern> {
2998 let Inst{28} = subtract;
2999 let Inst{21-20} = size;
3000 let Inst{19-17} = Qn{2-0};
3002 let Inst{12} = exch;
3004 let Inst{7} = Qn{3};
3005 let Inst{0} = round;
3008 multiclass MVE_VQxDMLxDH_multi<string iname, bit exch,
3009 bit round, bit subtract> {
3010 def s8 : MVE_VQxDMLxDH<iname, exch, round, subtract, "s8", 0b00>;
3011 def s16 : MVE_VQxDMLxDH<iname, exch, round, subtract, "s16", 0b01>;
3012 def s32 : MVE_VQxDMLxDH<iname, exch, round, subtract, "s32", 0b10>;
3015 defm MVE_VQDMLADH : MVE_VQxDMLxDH_multi<"vqdmladh", 0b0, 0b0, 0b0>;
3016 defm MVE_VQDMLADHX : MVE_VQxDMLxDH_multi<"vqdmladhx", 0b1, 0b0, 0b0>;
3017 defm MVE_VQRDMLADH : MVE_VQxDMLxDH_multi<"vqrdmladh", 0b0, 0b1, 0b0>;
3018 defm MVE_VQRDMLADHX : MVE_VQxDMLxDH_multi<"vqrdmladhx", 0b1, 0b1, 0b0>;
3019 defm MVE_VQDMLSDH : MVE_VQxDMLxDH_multi<"vqdmlsdh", 0b0, 0b0, 0b1>;
3020 defm MVE_VQDMLSDHX : MVE_VQxDMLxDH_multi<"vqdmlsdhx", 0b1, 0b0, 0b1>;
3021 defm MVE_VQRDMLSDH : MVE_VQxDMLxDH_multi<"vqrdmlsdh", 0b0, 0b1, 0b1>;
3022 defm MVE_VQRDMLSDHX : MVE_VQxDMLxDH_multi<"vqrdmlsdhx", 0b1, 0b1, 0b1>;
3024 class MVE_VCMUL<string iname, string suffix, bit size, list<dag> pattern=[]>
3025 : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd),
3026 (ins MQPR:$Qn, MQPR:$Qm, complexrotateop:$rot),
3027 "$Qd, $Qn, $Qm, $rot", vpred_r, "", pattern> {
3031 let Inst{28} = size;
3032 let Inst{21-20} = 0b11;
3033 let Inst{19-17} = Qn{2-0};
3035 let Inst{12} = rot{1};
3037 let Inst{7} = Qn{3};
3038 let Inst{0} = rot{0};
3040 let Predicates = [HasMVEFloat];
3043 def MVE_VCMULf16 : MVE_VCMUL<"vcmul", "f16", 0b0>;
3044 def MVE_VCMULf32 : MVE_VCMUL<"vcmul", "f32", 0b1>;
3046 class MVE_VMULL<string iname, string suffix, bit bit_28, bits<2> bits_21_20,
3047 bit T, list<dag> pattern=[]>
3048 : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd),
3049 (ins MQPR:$Qn, MQPR:$Qm), "$Qd, $Qn, $Qm",
3050 vpred_r, "", pattern> {
3055 let Inst{28} = bit_28;
3056 let Inst{21-20} = bits_21_20;
3057 let Inst{19-17} = Qn{2-0};
3061 let Inst{7} = Qn{3};
3065 multiclass MVE_VMULL_multi<string iname, string suffix,
3066 bit bit_28, bits<2> bits_21_20> {
3067 def bh : MVE_VMULL<iname # "b", suffix, bit_28, bits_21_20, 0b0>;
3068 def th : MVE_VMULL<iname # "t", suffix, bit_28, bits_21_20, 0b1>;
3071 // For integer multiplies, bits 21:20 encode size, and bit 28 signedness.
3072 // For polynomial multiplies, bits 21:20 take the unused value 0b11, and
3073 // bit 28 switches to encoding the size.
3075 defm MVE_VMULLs8 : MVE_VMULL_multi<"vmull", "s8", 0b0, 0b00>;
3076 defm MVE_VMULLs16 : MVE_VMULL_multi<"vmull", "s16", 0b0, 0b01>;
3077 defm MVE_VMULLs32 : MVE_VMULL_multi<"vmull", "s32", 0b0, 0b10>;
3078 defm MVE_VMULLu8 : MVE_VMULL_multi<"vmull", "u8", 0b1, 0b00>;
3079 defm MVE_VMULLu16 : MVE_VMULL_multi<"vmull", "u16", 0b1, 0b01>;
3080 defm MVE_VMULLu32 : MVE_VMULL_multi<"vmull", "u32", 0b1, 0b10>;
3081 defm MVE_VMULLp8 : MVE_VMULL_multi<"vmull", "p8", 0b0, 0b11>;
3082 defm MVE_VMULLp16 : MVE_VMULL_multi<"vmull", "p16", 0b1, 0b11>;
3084 class MVE_VxMULH<string iname, string suffix, bit U, bits<2> size,
3085 bit round, list<dag> pattern=[]>
3086 : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd),
3087 (ins MQPR:$Qn, MQPR:$Qm), "$Qd, $Qn, $Qm",
3088 vpred_r, "", pattern> {
3092 let Inst{21-20} = size;
3093 let Inst{19-17} = Qn{2-0};
3095 let Inst{12} = round;
3097 let Inst{7} = Qn{3};
3101 def MVE_VMULHs8 : MVE_VxMULH<"vmulh", "s8", 0b0, 0b00, 0b0>;
3102 def MVE_VMULHs16 : MVE_VxMULH<"vmulh", "s16", 0b0, 0b01, 0b0>;
3103 def MVE_VMULHs32 : MVE_VxMULH<"vmulh", "s32", 0b0, 0b10, 0b0>;
3104 def MVE_VMULHu8 : MVE_VxMULH<"vmulh", "u8", 0b1, 0b00, 0b0>;
3105 def MVE_VMULHu16 : MVE_VxMULH<"vmulh", "u16", 0b1, 0b01, 0b0>;
3106 def MVE_VMULHu32 : MVE_VxMULH<"vmulh", "u32", 0b1, 0b10, 0b0>;
3108 def MVE_VRMULHs8 : MVE_VxMULH<"vrmulh", "s8", 0b0, 0b00, 0b1>;
3109 def MVE_VRMULHs16 : MVE_VxMULH<"vrmulh", "s16", 0b0, 0b01, 0b1>;
3110 def MVE_VRMULHs32 : MVE_VxMULH<"vrmulh", "s32", 0b0, 0b10, 0b1>;
3111 def MVE_VRMULHu8 : MVE_VxMULH<"vrmulh", "u8", 0b1, 0b00, 0b1>;
3112 def MVE_VRMULHu16 : MVE_VxMULH<"vrmulh", "u16", 0b1, 0b01, 0b1>;
3113 def MVE_VRMULHu32 : MVE_VxMULH<"vrmulh", "u32", 0b1, 0b10, 0b1>;
3115 class MVE_VxMOVxN<string iname, string suffix, bit bit_28, bit bit_17,
3116 bits<2> size, bit T, list<dag> pattern=[]>
3117 : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd),
3118 (ins MQPR:$Qd_src, MQPR:$Qm), "$Qd, $Qm",
3119 vpred_n, "$Qd = $Qd_src", pattern> {
3121 let Inst{28} = bit_28;
3122 let Inst{21-20} = 0b11;
3123 let Inst{19-18} = size;
3124 let Inst{17} = bit_17;
3128 let Inst{7} = !if(!eq(bit_17, 0), 1, 0);
3132 multiclass MVE_VxMOVxN_halves<string iname, string suffix,
3133 bit bit_28, bit bit_17, bits<2> size> {
3134 def bh : MVE_VxMOVxN<iname # "b", suffix, bit_28, bit_17, size, 0b0>;
3135 def th : MVE_VxMOVxN<iname # "t", suffix, bit_28, bit_17, size, 0b1>;
3138 defm MVE_VMOVNi16 : MVE_VxMOVxN_halves<"vmovn", "i16", 0b1, 0b0, 0b00>;
3139 defm MVE_VMOVNi32 : MVE_VxMOVxN_halves<"vmovn", "i32", 0b1, 0b0, 0b01>;
3140 defm MVE_VQMOVNs16 : MVE_VxMOVxN_halves<"vqmovn", "s16", 0b0, 0b1, 0b00>;
3141 defm MVE_VQMOVNs32 : MVE_VxMOVxN_halves<"vqmovn", "s32", 0b0, 0b1, 0b01>;
3142 defm MVE_VQMOVNu16 : MVE_VxMOVxN_halves<"vqmovn", "u16", 0b1, 0b1, 0b00>;
3143 defm MVE_VQMOVNu32 : MVE_VxMOVxN_halves<"vqmovn", "u32", 0b1, 0b1, 0b01>;
3144 defm MVE_VQMOVUNs16 : MVE_VxMOVxN_halves<"vqmovun", "s16", 0b0, 0b0, 0b00>;
3145 defm MVE_VQMOVUNs32 : MVE_VxMOVxN_halves<"vqmovun", "s32", 0b0, 0b0, 0b01>;
3147 class MVE_VCVT_ff<string iname, string suffix, bit op, bit T,
3148 list<dag> pattern=[]>
3149 : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd), (ins MQPR:$Qd_src, MQPR:$Qm),
3150 "$Qd, $Qm", vpred_n, "$Qd = $Qd_src", pattern> {
3152 let Inst{21-16} = 0b111111;
3154 let Inst{8-7} = 0b00;
3157 let Predicates = [HasMVEFloat];
3160 multiclass MVE_VCVT_ff_halves<string suffix, bit op> {
3161 def bh : MVE_VCVT_ff<"vcvtb", suffix, op, 0b0>;
3162 def th : MVE_VCVT_ff<"vcvtt", suffix, op, 0b1>;
3165 defm MVE_VCVTf16f32 : MVE_VCVT_ff_halves<"f16.f32", 0b0>;
3166 defm MVE_VCVTf32f16 : MVE_VCVT_ff_halves<"f32.f16", 0b1>;
3168 class MVE_VxCADD<string iname, string suffix, bits<2> size, bit halve,
3169 list<dag> pattern=[]>
3170 : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd),
3171 (ins MQPR:$Qn, MQPR:$Qm, complexrotateopodd:$rot),
3172 "$Qd, $Qn, $Qm, $rot", vpred_r, "",
3177 let Inst{28} = halve;
3178 let Inst{21-20} = size;
3179 let Inst{19-17} = Qn{2-0};
3183 let Inst{7} = Qn{3};
3187 def MVE_VCADDi8 : MVE_VxCADD<"vcadd", "i8", 0b00, 0b1>;
3188 def MVE_VCADDi16 : MVE_VxCADD<"vcadd", "i16", 0b01, 0b1>;
3189 def MVE_VCADDi32 : MVE_VxCADD<"vcadd", "i32", 0b10, 0b1>;
3191 def MVE_VHCADDs8 : MVE_VxCADD<"vhcadd", "s8", 0b00, 0b0>;
3192 def MVE_VHCADDs16 : MVE_VxCADD<"vhcadd", "s16", 0b01, 0b0>;
3193 def MVE_VHCADDs32 : MVE_VxCADD<"vhcadd", "s32", 0b10, 0b0>;
3195 class MVE_VADCSBC<string iname, bit I, bit subtract,
3196 dag carryin, list<dag> pattern=[]>
3197 : MVE_qDest_qSrc<iname, "i32", (outs MQPR:$Qd, cl_FPSCR_NZCV:$carryout),
3198 !con((ins MQPR:$Qn, MQPR:$Qm), carryin),
3199 "$Qd, $Qn, $Qm", vpred_r, "", pattern> {
3202 let Inst{28} = subtract;
3203 let Inst{21-20} = 0b11;
3204 let Inst{19-17} = Qn{2-0};
3208 let Inst{7} = Qn{3};
3211 // Custom decoder method in order to add the FPSCR operand(s), which
3212 // Tablegen won't do right
3213 let DecoderMethod = "DecodeMVEVADCInstruction";
3216 def MVE_VADC : MVE_VADCSBC<"vadc", 0b0, 0b0, (ins cl_FPSCR_NZCV:$carryin)>;
3217 def MVE_VADCI : MVE_VADCSBC<"vadci", 0b1, 0b0, (ins)>;
3219 def MVE_VSBC : MVE_VADCSBC<"vsbc", 0b0, 0b1, (ins cl_FPSCR_NZCV:$carryin)>;
3220 def MVE_VSBCI : MVE_VADCSBC<"vsbci", 0b1, 0b1, (ins)>;
3222 class MVE_VQDMULL<string iname, string suffix, bit size, bit T,
3223 list<dag> pattern=[]>
3224 : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd),
3225 (ins MQPR:$Qn, MQPR:$Qm), "$Qd, $Qn, $Qm",
3226 vpred_r, "", pattern> {
3229 let Inst{28} = size;
3230 let Inst{21-20} = 0b11;
3231 let Inst{19-17} = Qn{2-0};
3235 let Inst{7} = Qn{3};
3239 multiclass MVE_VQDMULL_halves<string suffix, bit size> {
3240 def bh : MVE_VQDMULL<"vqdmullb", suffix, size, 0b0>;
3241 def th : MVE_VQDMULL<"vqdmullt", suffix, size, 0b1>;
3244 defm MVE_VQDMULLs16 : MVE_VQDMULL_halves<"s16", 0b0>;
3245 defm MVE_VQDMULLs32 : MVE_VQDMULL_halves<"s32", 0b1>;
3247 // end of mve_qDest_qSrc
3249 // start of mve_qDest_rSrc
3251 class MVE_qr_base<dag oops, dag iops, InstrItinClass itin, string iname,
3252 string suffix, string ops, vpred_ops vpred, string cstr,
3253 list<dag> pattern=[]>
3254 : MVE_p<oops, iops, NoItinerary, iname, suffix, ops, vpred, cstr, pattern> {
3259 let Inst{25-23} = 0b100;
3260 let Inst{22} = Qd{3};
3261 let Inst{19-17} = Qn{2-0};
3262 let Inst{15-13} = Qd{2-0};
3263 let Inst{11-9} = 0b111;
3264 let Inst{7} = Qn{3};
3267 let Inst{3-0} = Rm{3-0};
3270 class MVE_qDest_rSrc<string iname, string suffix, list<dag> pattern=[]>
3271 : MVE_qr_base<(outs MQPR:$Qd), (ins MQPR:$Qn, rGPR:$Rm),
3272 NoItinerary, iname, suffix, "$Qd, $Qn, $Rm", vpred_r, "",
3275 class MVE_qDestSrc_rSrc<string iname, string suffix, list<dag> pattern=[]>
3276 : MVE_qr_base<(outs MQPR:$Qd), (ins MQPR:$Qd_src, MQPR:$Qn, rGPR:$Rm),
3277 NoItinerary, iname, suffix, "$Qd, $Qn, $Rm", vpred_n, "$Qd = $Qd_src",
3280 class MVE_qDest_single_rSrc<string iname, string suffix, list<dag> pattern=[]>
3281 : MVE_p<(outs MQPR:$Qd), (ins MQPR:$Qd_src, rGPR:$Rm), NoItinerary, iname,
3282 suffix, "$Qd, $Rm", vpred_n, "$Qd = $Qd_src", pattern> {
3286 let Inst{22} = Qd{3};
3287 let Inst{15-13} = Qd{2-0};
3288 let Inst{3-0} = Rm{3-0};
3291 class MVE_VADDSUB_qr<string iname, string suffix, bits<2> size,
3292 bit bit_5, bit bit_12, bit bit_16,
3293 bit bit_28, list<dag> pattern=[]>
3294 : MVE_qDest_rSrc<iname, suffix, pattern> {
3296 let Inst{28} = bit_28;
3297 let Inst{21-20} = size;
3298 let Inst{16} = bit_16;
3299 let Inst{12} = bit_12;
3301 let Inst{5} = bit_5;
3304 multiclass MVE_VADDSUB_qr_sizes<string iname, string suffix,
3305 bit bit_5, bit bit_12, bit bit_16,
3306 bit bit_28, list<dag> pattern=[]> {
3307 def "8" : MVE_VADDSUB_qr<iname, suffix#"8", 0b00,
3308 bit_5, bit_12, bit_16, bit_28>;
3309 def "16" : MVE_VADDSUB_qr<iname, suffix#"16", 0b01,
3310 bit_5, bit_12, bit_16, bit_28>;
3311 def "32" : MVE_VADDSUB_qr<iname, suffix#"32", 0b10,
3312 bit_5, bit_12, bit_16, bit_28>;
3315 defm MVE_VADD_qr_i : MVE_VADDSUB_qr_sizes<"vadd", "i", 0b0, 0b0, 0b1, 0b0>;
3316 defm MVE_VQADD_qr_s : MVE_VADDSUB_qr_sizes<"vqadd", "s", 0b1, 0b0, 0b0, 0b0>;
3317 defm MVE_VQADD_qr_u : MVE_VADDSUB_qr_sizes<"vqadd", "u", 0b1, 0b0, 0b0, 0b1>;
3319 defm MVE_VSUB_qr_i : MVE_VADDSUB_qr_sizes<"vsub", "i", 0b0, 0b1, 0b1, 0b0>;
3320 defm MVE_VQSUB_qr_s : MVE_VADDSUB_qr_sizes<"vqsub", "s", 0b1, 0b1, 0b0, 0b0>;
3321 defm MVE_VQSUB_qr_u : MVE_VADDSUB_qr_sizes<"vqsub", "u", 0b1, 0b1, 0b0, 0b1>;
3323 class MVE_VQDMULL_qr<string iname, string suffix, bit size,
3324 bit T, list<dag> pattern=[]>
3325 : MVE_qDest_rSrc<iname, suffix, pattern> {
3327 let Inst{28} = size;
3328 let Inst{21-20} = 0b11;
3335 multiclass MVE_VQDMULL_qr_halves<string suffix, bit size> {
3336 def bh : MVE_VQDMULL_qr<"vqdmullb", suffix, size, 0b0>;
3337 def th : MVE_VQDMULL_qr<"vqdmullt", suffix, size, 0b1>;
3340 defm MVE_VQDMULL_qr_s16 : MVE_VQDMULL_qr_halves<"s16", 0b0>;
3341 defm MVE_VQDMULL_qr_s32 : MVE_VQDMULL_qr_halves<"s32", 0b1>;
3343 class MVE_VxADDSUB_qr<string iname, string suffix,
3344 bit bit_28, bits<2> bits_21_20, bit subtract,
3345 list<dag> pattern=[]>
3346 : MVE_qDest_rSrc<iname, suffix, pattern> {
3348 let Inst{28} = bit_28;
3349 let Inst{21-20} = bits_21_20;
3351 let Inst{12} = subtract;
3356 def MVE_VHADD_qr_s8 : MVE_VxADDSUB_qr<"vhadd", "s8", 0b0, 0b00, 0b0>;
3357 def MVE_VHADD_qr_s16 : MVE_VxADDSUB_qr<"vhadd", "s16", 0b0, 0b01, 0b0>;
3358 def MVE_VHADD_qr_s32 : MVE_VxADDSUB_qr<"vhadd", "s32", 0b0, 0b10, 0b0>;
3359 def MVE_VHADD_qr_u8 : MVE_VxADDSUB_qr<"vhadd", "u8", 0b1, 0b00, 0b0>;
3360 def MVE_VHADD_qr_u16 : MVE_VxADDSUB_qr<"vhadd", "u16", 0b1, 0b01, 0b0>;
3361 def MVE_VHADD_qr_u32 : MVE_VxADDSUB_qr<"vhadd", "u32", 0b1, 0b10, 0b0>;
3363 def MVE_VHSUB_qr_s8 : MVE_VxADDSUB_qr<"vhsub", "s8", 0b0, 0b00, 0b1>;
3364 def MVE_VHSUB_qr_s16 : MVE_VxADDSUB_qr<"vhsub", "s16", 0b0, 0b01, 0b1>;
3365 def MVE_VHSUB_qr_s32 : MVE_VxADDSUB_qr<"vhsub", "s32", 0b0, 0b10, 0b1>;
3366 def MVE_VHSUB_qr_u8 : MVE_VxADDSUB_qr<"vhsub", "u8", 0b1, 0b00, 0b1>;
3367 def MVE_VHSUB_qr_u16 : MVE_VxADDSUB_qr<"vhsub", "u16", 0b1, 0b01, 0b1>;
3368 def MVE_VHSUB_qr_u32 : MVE_VxADDSUB_qr<"vhsub", "u32", 0b1, 0b10, 0b1>;
3370 let Predicates = [HasMVEFloat] in {
3371 def MVE_VADD_qr_f32 : MVE_VxADDSUB_qr<"vadd", "f32", 0b0, 0b11, 0b0>;
3372 def MVE_VADD_qr_f16 : MVE_VxADDSUB_qr<"vadd", "f16", 0b1, 0b11, 0b0>;
3374 def MVE_VSUB_qr_f32 : MVE_VxADDSUB_qr<"vsub", "f32", 0b0, 0b11, 0b1>;
3375 def MVE_VSUB_qr_f16 : MVE_VxADDSUB_qr<"vsub", "f16", 0b1, 0b11, 0b1>;
3378 class MVE_VxSHL_qr<string iname, string suffix, bit U, bits<2> size,
3379 bit bit_7, bit bit_17, list<dag> pattern=[]>
3380 : MVE_qDest_single_rSrc<iname, suffix, pattern> {
3383 let Inst{25-23} = 0b100;
3384 let Inst{21-20} = 0b11;
3385 let Inst{19-18} = size;
3386 let Inst{17} = bit_17;
3388 let Inst{12-8} = 0b11110;
3389 let Inst{7} = bit_7;
3390 let Inst{6-4} = 0b110;
3393 multiclass MVE_VxSHL_qr_types<string iname, bit bit_7, bit bit_17> {
3394 def s8 : MVE_VxSHL_qr<iname, "s8", 0b0, 0b00, bit_7, bit_17>;
3395 def s16 : MVE_VxSHL_qr<iname, "s16", 0b0, 0b01, bit_7, bit_17>;
3396 def s32 : MVE_VxSHL_qr<iname, "s32", 0b0, 0b10, bit_7, bit_17>;
3397 def u8 : MVE_VxSHL_qr<iname, "u8", 0b1, 0b00, bit_7, bit_17>;
3398 def u16 : MVE_VxSHL_qr<iname, "u16", 0b1, 0b01, bit_7, bit_17>;
3399 def u32 : MVE_VxSHL_qr<iname, "u32", 0b1, 0b10, bit_7, bit_17>;
3402 defm MVE_VSHL_qr : MVE_VxSHL_qr_types<"vshl", 0b0, 0b0>;
3403 defm MVE_VRSHL_qr : MVE_VxSHL_qr_types<"vrshl", 0b0, 0b1>;
3404 defm MVE_VQSHL_qr : MVE_VxSHL_qr_types<"vqshl", 0b1, 0b0>;
3405 defm MVE_VQRSHL_qr : MVE_VxSHL_qr_types<"vqrshl", 0b1, 0b1>;
3407 let Predicates = [HasMVEInt] in {
3408 def : Pat<(v4i32 (ARMvshlu (v4i32 MQPR:$Qm), (v4i32 (ARMvdup GPR:$Rm)))),
3409 (v4i32 (MVE_VSHL_qru32 (v4i32 MQPR:$Qm), GPR:$Rm))>;
3410 def : Pat<(v8i16 (ARMvshlu (v8i16 MQPR:$Qm), (v8i16 (ARMvdup GPR:$Rm)))),
3411 (v8i16 (MVE_VSHL_qru16 (v8i16 MQPR:$Qm), GPR:$Rm))>;
3412 def : Pat<(v16i8 (ARMvshlu (v16i8 MQPR:$Qm), (v16i8 (ARMvdup GPR:$Rm)))),
3413 (v16i8 (MVE_VSHL_qru8 (v16i8 MQPR:$Qm), GPR:$Rm))>;
3415 def : Pat<(v4i32 (ARMvshls (v4i32 MQPR:$Qm), (v4i32 (ARMvdup GPR:$Rm)))),
3416 (v4i32 (MVE_VSHL_qrs32 (v4i32 MQPR:$Qm), GPR:$Rm))>;
3417 def : Pat<(v8i16 (ARMvshls (v8i16 MQPR:$Qm), (v8i16 (ARMvdup GPR:$Rm)))),
3418 (v8i16 (MVE_VSHL_qrs16 (v8i16 MQPR:$Qm), GPR:$Rm))>;
3419 def : Pat<(v16i8 (ARMvshls (v16i8 MQPR:$Qm), (v16i8 (ARMvdup GPR:$Rm)))),
3420 (v16i8 (MVE_VSHL_qrs8 (v16i8 MQPR:$Qm), GPR:$Rm))>;
3423 class MVE_VBRSR<string iname, string suffix, bits<2> size, list<dag> pattern=[]>
3424 : MVE_qDest_rSrc<iname, suffix, pattern> {
3427 let Inst{21-20} = size;
3434 def MVE_VBRSR8 : MVE_VBRSR<"vbrsr", "8", 0b00>;
3435 def MVE_VBRSR16 : MVE_VBRSR<"vbrsr", "16", 0b01>;
3436 def MVE_VBRSR32 : MVE_VBRSR<"vbrsr", "32", 0b10>;
3438 class MVE_VMUL_qr_int<string iname, string suffix,
3439 bits<2> size, list<dag> pattern=[]>
3440 : MVE_qDest_rSrc<iname, suffix, pattern> {
3443 let Inst{21-20} = size;
3450 def MVE_VMUL_qr_i8 : MVE_VMUL_qr_int<"vmul", "i8", 0b00>;
3451 def MVE_VMUL_qr_i16 : MVE_VMUL_qr_int<"vmul", "i16", 0b01>;
3452 def MVE_VMUL_qr_i32 : MVE_VMUL_qr_int<"vmul", "i32", 0b10>;
3454 class MVE_VxxMUL_qr<string iname, string suffix,
3455 bit bit_28, bits<2> bits_21_20, list<dag> pattern=[]>
3456 : MVE_qDest_rSrc<iname, suffix, pattern> {
3458 let Inst{28} = bit_28;
3459 let Inst{21-20} = bits_21_20;
3466 def MVE_VQDMULH_qr_s8 : MVE_VxxMUL_qr<"vqdmulh", "s8", 0b0, 0b00>;
3467 def MVE_VQDMULH_qr_s16 : MVE_VxxMUL_qr<"vqdmulh", "s16", 0b0, 0b01>;
3468 def MVE_VQDMULH_qr_s32 : MVE_VxxMUL_qr<"vqdmulh", "s32", 0b0, 0b10>;
3470 def MVE_VQRDMULH_qr_s8 : MVE_VxxMUL_qr<"vqrdmulh", "s8", 0b1, 0b00>;
3471 def MVE_VQRDMULH_qr_s16 : MVE_VxxMUL_qr<"vqrdmulh", "s16", 0b1, 0b01>;
3472 def MVE_VQRDMULH_qr_s32 : MVE_VxxMUL_qr<"vqrdmulh", "s32", 0b1, 0b10>;
3474 let Predicates = [HasMVEFloat] in {
3475 def MVE_VMUL_qr_f16 : MVE_VxxMUL_qr<"vmul", "f16", 0b1, 0b11>;
3476 def MVE_VMUL_qr_f32 : MVE_VxxMUL_qr<"vmul", "f32", 0b0, 0b11>;
3479 class MVE_VFMAMLA_qr<string iname, string suffix,
3480 bit bit_28, bits<2> bits_21_20, bit S,
3481 list<dag> pattern=[]>
3482 : MVE_qDestSrc_rSrc<iname, suffix, pattern> {
3484 let Inst{28} = bit_28;
3485 let Inst{21-20} = bits_21_20;
3492 def MVE_VMLA_qr_s8 : MVE_VFMAMLA_qr<"vmla", "s8", 0b0, 0b00, 0b0>;
3493 def MVE_VMLA_qr_s16 : MVE_VFMAMLA_qr<"vmla", "s16", 0b0, 0b01, 0b0>;
3494 def MVE_VMLA_qr_s32 : MVE_VFMAMLA_qr<"vmla", "s32", 0b0, 0b10, 0b0>;
3495 def MVE_VMLA_qr_u8 : MVE_VFMAMLA_qr<"vmla", "u8", 0b1, 0b00, 0b0>;
3496 def MVE_VMLA_qr_u16 : MVE_VFMAMLA_qr<"vmla", "u16", 0b1, 0b01, 0b0>;
3497 def MVE_VMLA_qr_u32 : MVE_VFMAMLA_qr<"vmla", "u32", 0b1, 0b10, 0b0>;
3499 def MVE_VMLAS_qr_s8 : MVE_VFMAMLA_qr<"vmlas", "s8", 0b0, 0b00, 0b1>;
3500 def MVE_VMLAS_qr_s16 : MVE_VFMAMLA_qr<"vmlas", "s16", 0b0, 0b01, 0b1>;
3501 def MVE_VMLAS_qr_s32 : MVE_VFMAMLA_qr<"vmlas", "s32", 0b0, 0b10, 0b1>;
3502 def MVE_VMLAS_qr_u8 : MVE_VFMAMLA_qr<"vmlas", "u8", 0b1, 0b00, 0b1>;
3503 def MVE_VMLAS_qr_u16 : MVE_VFMAMLA_qr<"vmlas", "u16", 0b1, 0b01, 0b1>;
3504 def MVE_VMLAS_qr_u32 : MVE_VFMAMLA_qr<"vmlas", "u32", 0b1, 0b10, 0b1>;
3506 let Predicates = [HasMVEFloat] in {
3507 def MVE_VFMA_qr_f16 : MVE_VFMAMLA_qr<"vfma", "f16", 0b1, 0b11, 0b0>;
3508 def MVE_VFMA_qr_f32 : MVE_VFMAMLA_qr<"vfma", "f32", 0b0, 0b11, 0b0>;
3509 def MVE_VFMA_qr_Sf16 : MVE_VFMAMLA_qr<"vfmas", "f16", 0b1, 0b11, 0b1>;
3510 def MVE_VFMA_qr_Sf32 : MVE_VFMAMLA_qr<"vfmas", "f32", 0b0, 0b11, 0b1>;
3513 class MVE_VQDMLAH_qr<string iname, string suffix, bit U, bits<2> size,
3514 bit bit_5, bit bit_12, list<dag> pattern=[]>
3515 : MVE_qDestSrc_rSrc<iname, suffix, pattern> {
3518 let Inst{21-20} = size;
3520 let Inst{12} = bit_12;
3522 let Inst{5} = bit_5;
3525 multiclass MVE_VQDMLAH_qr_types<string iname, bit bit_5, bit bit_12> {
3526 def s8 : MVE_VQDMLAH_qr<iname, "s8", 0b0, 0b00, bit_5, bit_12>;
3527 def s16 : MVE_VQDMLAH_qr<iname, "s16", 0b0, 0b01, bit_5, bit_12>;
3528 def s32 : MVE_VQDMLAH_qr<iname, "s32", 0b0, 0b10, bit_5, bit_12>;
3531 defm MVE_VQDMLAH_qr : MVE_VQDMLAH_qr_types<"vqdmlah", 0b1, 0b0>;
3532 defm MVE_VQRDMLAH_qr : MVE_VQDMLAH_qr_types<"vqrdmlah", 0b0, 0b0>;
3533 defm MVE_VQDMLASH_qr : MVE_VQDMLAH_qr_types<"vqdmlash", 0b1, 0b1>;
3534 defm MVE_VQRDMLASH_qr : MVE_VQDMLAH_qr_types<"vqrdmlash", 0b0, 0b1>;
3536 class MVE_VxDUP<string iname, string suffix, bits<2> size, bit bit_12,
3537 list<dag> pattern=[]>
3538 : MVE_p<(outs MQPR:$Qd, tGPREven:$Rn),
3539 (ins tGPREven:$Rn_src, MVE_VIDUP_imm:$imm), NoItinerary,
3540 iname, suffix, "$Qd, $Rn, $imm", vpred_r, "$Rn = $Rn_src",
3547 let Inst{25-23} = 0b100;
3548 let Inst{22} = Qd{3};
3549 let Inst{21-20} = size;
3550 let Inst{19-17} = Rn{3-1};
3552 let Inst{15-13} = Qd{2-0};
3553 let Inst{12} = bit_12;
3554 let Inst{11-8} = 0b1111;
3555 let Inst{7} = imm{1};
3556 let Inst{6-1} = 0b110111;
3557 let Inst{0} = imm{0};
3560 def MVE_VIDUPu8 : MVE_VxDUP<"vidup", "u8", 0b00, 0b0>;
3561 def MVE_VIDUPu16 : MVE_VxDUP<"vidup", "u16", 0b01, 0b0>;
3562 def MVE_VIDUPu32 : MVE_VxDUP<"vidup", "u32", 0b10, 0b0>;
3564 def MVE_VDDUPu8 : MVE_VxDUP<"vddup", "u8", 0b00, 0b1>;
3565 def MVE_VDDUPu16 : MVE_VxDUP<"vddup", "u16", 0b01, 0b1>;
3566 def MVE_VDDUPu32 : MVE_VxDUP<"vddup", "u32", 0b10, 0b1>;
3568 class MVE_VxWDUP<string iname, string suffix, bits<2> size, bit bit_12,
3569 list<dag> pattern=[]>
3570 : MVE_p<(outs MQPR:$Qd, tGPREven:$Rn),
3571 (ins tGPREven:$Rn_src, tGPROdd:$Rm, MVE_VIDUP_imm:$imm), NoItinerary,
3572 iname, suffix, "$Qd, $Rn, $Rm, $imm", vpred_r, "$Rn = $Rn_src",
3580 let Inst{25-23} = 0b100;
3581 let Inst{22} = Qd{3};
3582 let Inst{21-20} = size;
3583 let Inst{19-17} = Rn{3-1};
3585 let Inst{15-13} = Qd{2-0};
3586 let Inst{12} = bit_12;
3587 let Inst{11-8} = 0b1111;
3588 let Inst{7} = imm{1};
3589 let Inst{6-4} = 0b110;
3590 let Inst{3-1} = Rm{3-1};
3591 let Inst{0} = imm{0};
3594 def MVE_VIWDUPu8 : MVE_VxWDUP<"viwdup", "u8", 0b00, 0b0>;
3595 def MVE_VIWDUPu16 : MVE_VxWDUP<"viwdup", "u16", 0b01, 0b0>;
3596 def MVE_VIWDUPu32 : MVE_VxWDUP<"viwdup", "u32", 0b10, 0b0>;
3598 def MVE_VDWDUPu8 : MVE_VxWDUP<"vdwdup", "u8", 0b00, 0b1>;
3599 def MVE_VDWDUPu16 : MVE_VxWDUP<"vdwdup", "u16", 0b01, 0b1>;
3600 def MVE_VDWDUPu32 : MVE_VxWDUP<"vdwdup", "u32", 0b10, 0b1>;
3602 class MVE_VCTP<string suffix, bits<2> size, list<dag> pattern=[]>
3603 : MVE_p<(outs VCCR:$P0), (ins rGPR:$Rn), NoItinerary, "vctp", suffix,
3604 "$Rn", vpred_n, "", pattern> {
3607 let Inst{28-27} = 0b10;
3608 let Inst{26-22} = 0b00000;
3609 let Inst{21-20} = size;
3610 let Inst{19-16} = Rn{3-0};
3611 let Inst{15-11} = 0b11101;
3612 let Inst{10-0} = 0b00000000001;
3613 let Unpredictable{10-0} = 0b11111111111;
3615 let Constraints = "";
3616 let DecoderMethod = "DecodeMveVCTP";
3619 def MVE_VCTP8 : MVE_VCTP<"8", 0b00>;
3620 def MVE_VCTP16 : MVE_VCTP<"16", 0b01>;
3621 def MVE_VCTP32 : MVE_VCTP<"32", 0b10>;
3622 def MVE_VCTP64 : MVE_VCTP<"64", 0b11>;
3624 // end of mve_qDest_rSrc
3626 // start of coproc mov
3628 class MVE_VMOV_64bit<dag oops, dag iops, bit to_qreg, string ops, string cstr>
3629 : MVE_VMOV_lane_base<oops, !con(iops, (ins MVEPairVectorIndex2:$idx,
3630 MVEPairVectorIndex0:$idx2)),
3631 NoItinerary, "vmov", "", ops, cstr, []> {
3638 let Inst{31-23} = 0b111011000;
3639 let Inst{22} = Qd{3};
3641 let Inst{20} = to_qreg;
3642 let Inst{19-16} = Rt2{3-0};
3643 let Inst{15-13} = Qd{2-0};
3644 let Inst{12-5} = 0b01111000;
3646 let Inst{3-0} = Rt{3-0};
3649 // The assembly syntax for these instructions mentions the vector
3650 // register name twice, e.g.
3652 // vmov q2[2], q2[0], r0, r1
3653 // vmov r0, r1, q2[2], q2[0]
3655 // which needs a bit of juggling with MC operand handling.
3657 // For the move _into_ a vector register, the MC operand list also has
3658 // to mention the register name twice: once as the output, and once as
3659 // an extra input to represent where the unchanged half of the output
3660 // register comes from (when this instruction is used in code
3661 // generation). So we arrange that the first mention of the vector reg
3662 // in the instruction is considered by the AsmMatcher to be the output
3663 // ($Qd), and the second one is the input ($QdSrc). Binding them
3664 // together with the existing 'tie' constraint is enough to enforce at
3665 // register allocation time that they have to be the same register.
3667 // For the move _from_ a vector register, there's no way to get round
3668 // the fact that both instances of that register name have to be
3669 // inputs. They have to be the same register again, but this time, we
3670 // can't use a tie constraint, because that has to be between an
3671 // output and an input operand. So this time, we have to arrange that
3672 // the q-reg appears just once in the MC operand list, in spite of
3673 // being mentioned twice in the asm syntax - which needs a custom
3674 // AsmMatchConverter.
3676 def MVE_VMOV_q_rr : MVE_VMOV_64bit<(outs MQPR:$Qd),
3677 (ins MQPR:$QdSrc, rGPR:$Rt, rGPR:$Rt2),
3678 0b1, "$Qd$idx, $QdSrc$idx2, $Rt, $Rt2",
3680 let DecoderMethod = "DecodeMVEVMOVDRegtoQ";
3683 def MVE_VMOV_rr_q : MVE_VMOV_64bit<(outs rGPR:$Rt, rGPR:$Rt2), (ins MQPR:$Qd),
3684 0b0, "$Rt, $Rt2, $Qd$idx, $Qd$idx2", ""> {
3685 let DecoderMethod = "DecodeMVEVMOVQtoDReg";
3686 let AsmMatchConverter = "cvtMVEVMOVQtoDReg";
3689 // end of coproc mov
3691 // start of MVE interleaving load/store
3693 // Base class for the family of interleaving/deinterleaving
3694 // load/stores with names like VLD20.8 and VST43.32.
3695 class MVE_vldst24_base<bit writeback, bit fourregs, bits<2> stage, bits<2> size,
3696 bit load, dag Oops, dag loadIops, dag wbIops,
3697 string iname, string ops,
3698 string cstr, list<dag> pattern=[]>
3699 : MVE_MI<Oops, !con(loadIops, wbIops), NoItinerary, iname, ops, cstr, pattern> {
3703 let Inst{31-22} = 0b1111110010;
3704 let Inst{21} = writeback;
3705 let Inst{20} = load;
3706 let Inst{19-16} = Rn;
3707 let Inst{15-13} = VQd{2-0};
3708 let Inst{12-9} = 0b1111;
3709 let Inst{8-7} = size;
3710 let Inst{6-5} = stage;
3711 let Inst{4-1} = 0b0000;
3712 let Inst{0} = fourregs;
3715 let mayStore = !eq(load,0);
3718 // A parameter class used to encapsulate all the ways the writeback
3719 // variants of VLD20 and friends differ from the non-writeback ones.
3720 class MVE_vldst24_writeback<bit b, dag Oo, dag Io,
3721 string sy="", string c="", string n=""> {
3727 string id_suffix = n;
3730 // Another parameter class that encapsulates the differences between VLD2x
3732 class MVE_vldst24_nvecs<int n, list<int> s, bit b, RegisterOperand vl> {
3734 list<int> stages = s;
3736 RegisterOperand VecList = vl;
3739 // A third parameter class that distinguishes VLDnn.8 from .16 from .32.
3740 class MVE_vldst24_lanesize<int i, bits<2> b> {
3742 bits<2> sizebits = b;
3745 // A base class for each direction of transfer: one for load, one for
3746 // store. I can't make these a fourth independent parametric tuple
3747 // class, because they have to take the nvecs tuple class as a
3748 // parameter, in order to find the right VecList operand type.
3750 class MVE_vld24_base<MVE_vldst24_nvecs n, bits<2> pat, bits<2> size,
3751 MVE_vldst24_writeback wb, string iname,
3752 list<dag> pattern=[]>
3753 : MVE_vldst24_base<wb.writeback, n.bit0, pat, size, 1,
3754 !con((outs n.VecList:$VQd), wb.Oops),
3755 (ins n.VecList:$VQdSrc), wb.Iops,
3756 iname, "$VQd, $Rn" # wb.syntax,
3757 wb.cstr # ",$VQdSrc = $VQd", pattern>;
3759 class MVE_vst24_base<MVE_vldst24_nvecs n, bits<2> pat, bits<2> size,
3760 MVE_vldst24_writeback wb, string iname,
3761 list<dag> pattern=[]>
3762 : MVE_vldst24_base<wb.writeback, n.bit0, pat, size, 0,
3763 wb.Oops, (ins n.VecList:$VQd), wb.Iops,
3764 iname, "$VQd, $Rn" # wb.syntax,
3767 // Actually define all the interleaving loads and stores, by a series
3768 // of nested foreaches over number of vectors (VLD2/VLD4); stage
3769 // within one of those series (VLDx0/VLDx1/VLDx2/VLDx3); size of
3770 // vector lane; writeback or no writeback.
3771 foreach n = [MVE_vldst24_nvecs<2, [0,1], 0, VecList2Q>,
3772 MVE_vldst24_nvecs<4, [0,1,2,3], 1, VecList4Q>] in
3773 foreach stage = n.stages in
3774 foreach s = [MVE_vldst24_lanesize< 8, 0b00>,
3775 MVE_vldst24_lanesize<16, 0b01>,
3776 MVE_vldst24_lanesize<32, 0b10>] in
3777 foreach wb = [MVE_vldst24_writeback<
3778 1, (outs rGPR:$wb), (ins t2_nosp_addr_offset_none:$Rn),
3779 "!", "$Rn.base = $wb", "_wb">,
3780 MVE_vldst24_writeback<0, (outs), (ins t2_addr_offset_none:$Rn)>] in {
3782 // For each case within all of those foreaches, define the actual
3783 // instructions. The def names are made by gluing together pieces
3784 // from all the parameter classes, and will end up being things like
3785 // MVE_VLD20_8 and MVE_VST43_16_wb.
3787 def "MVE_VLD" # n.nvecs # stage # "_" # s.lanesize # wb.id_suffix
3788 : MVE_vld24_base<n, stage, s.sizebits, wb,
3789 "vld" # n.nvecs # stage # "." # s.lanesize>;
3791 def "MVE_VST" # n.nvecs # stage # "_" # s.lanesize # wb.id_suffix
3792 : MVE_vst24_base<n, stage, s.sizebits, wb,
3793 "vst" # n.nvecs # stage # "." # s.lanesize>;
3796 // end of MVE interleaving load/store
3798 // start of MVE predicable load/store
3800 // A parameter class for the direction of transfer.
3801 class MVE_ldst_direction<bit b, dag Oo, dag Io, string c=""> {
3807 def MVE_ld: MVE_ldst_direction<1, (outs MQPR:$Qd), (ins), ",@earlyclobber $Qd">;
3808 def MVE_st: MVE_ldst_direction<0, (outs), (ins MQPR:$Qd)>;
3810 // A parameter class for the size of memory access in a load.
3811 class MVE_memsz<bits<2> e, int s, AddrMode m, string mn, list<string> types> {
3812 bits<2> encoding = e; // opcode bit(s) for encoding
3813 int shift = s; // shift applied to immediate load offset
3816 // For instruction aliases: define the complete list of type
3817 // suffixes at this size, and the canonical ones for loads and
3819 string MnemonicLetter = mn;
3820 int TypeBits = !shl(8, s);
3821 string CanonLoadSuffix = ".u" # TypeBits;
3822 string CanonStoreSuffix = "." # TypeBits;
3823 list<string> suffixes = !foreach(letter, types, "." # letter # TypeBits);
3826 // Instances of MVE_memsz.
3828 // (memD doesn't need an AddrMode, because those are only for
3829 // contiguous loads, and memD is only used by gather/scatters.)
3830 def MVE_memB: MVE_memsz<0b00, 0, AddrModeT2_i7, "b", ["", "u", "s"]>;
3831 def MVE_memH: MVE_memsz<0b01, 1, AddrModeT2_i7s2, "h", ["", "u", "s", "f"]>;
3832 def MVE_memW: MVE_memsz<0b10, 2, AddrModeT2_i7s4, "w", ["", "u", "s", "f"]>;
3833 def MVE_memD: MVE_memsz<0b11, 3, ?, "d", ["", "u", "s", "f"]>;
3835 // This is the base class for all the MVE loads and stores other than
3836 // the interleaving ones. All the non-interleaving loads/stores share
3837 // the characteristic that they operate on just one vector register,
3838 // so they are VPT-predicable.
3840 // The predication operand is vpred_n, for both loads and stores. For
3841 // store instructions, the reason is obvious: if there is no output
3842 // register, there can't be a need for an input parameter giving the
3843 // output register's previous value. Load instructions also don't need
3844 // that input parameter, because unlike MVE data processing
3845 // instructions, predicated loads are defined to set the inactive
3846 // lanes of the output register to zero, instead of preserving their
3848 class MVE_VLDRSTR_base<MVE_ldst_direction dir, bit U, bit P, bit W, bit opc,
3849 dag oops, dag iops, string asm, string suffix,
3850 string ops, string cstr, list<dag> pattern=[]>
3851 : MVE_p<oops, iops, NoItinerary, asm, suffix, ops, vpred_n, cstr, pattern> {
3859 let Inst{20} = dir.load;
3860 let Inst{15-13} = Qd{2-0};
3862 let Inst{11-9} = 0b111;
3864 let mayLoad = dir.load;
3865 let mayStore = !eq(dir.load,0);
3868 // Contiguous load and store instructions. These come in two main
3869 // categories: same-size loads/stores in which 128 bits of vector
3870 // register is transferred to or from 128 bits of memory in the most
3871 // obvious way, and widening loads / narrowing stores, in which the
3872 // size of memory accessed is less than the size of a vector register,
3873 // so the load instructions sign- or zero-extend each memory value
3874 // into a wider vector lane, and the store instructions truncate
3877 // The instruction mnemonics for these two classes look reasonably
3878 // similar, but the actual encodings are different enough to need two
3879 // separate base classes.
3881 // Contiguous, same size
3882 class MVE_VLDRSTR_cs<MVE_ldst_direction dir, MVE_memsz memsz, bit P, bit W,
3883 dag oops, dag iops, string asm, string suffix,
3884 IndexMode im, string ops, string cstr>
3885 : MVE_VLDRSTR_base<dir, 0, P, W, 1, oops, iops, asm, suffix, ops, cstr> {
3887 let Inst{23} = addr{7};
3888 let Inst{19-16} = addr{11-8};
3889 let Inst{8-7} = memsz.encoding;
3890 let Inst{6-0} = addr{6-0};
3893 // Contiguous, widening/narrowing
3894 class MVE_VLDRSTR_cw<MVE_ldst_direction dir, MVE_memsz memsz, bit U,
3895 bit P, bit W, bits<2> size, dag oops, dag iops,
3896 string asm, string suffix, IndexMode im,
3897 string ops, string cstr>
3898 : MVE_VLDRSTR_base<dir, U, P, W, 0, oops, iops, asm, suffix, ops, cstr> {
3900 let Inst{23} = addr{7};
3901 let Inst{19} = memsz.encoding{0}; // enough to tell 16- from 32-bit
3902 let Inst{18-16} = addr{10-8};
3903 let Inst{8-7} = size;
3904 let Inst{6-0} = addr{6-0};
3909 // Multiclass wrapper on each of the _cw and _cs base classes, to
3910 // generate three writeback modes (none, preindex, postindex).
3912 multiclass MVE_VLDRSTR_cw_m<MVE_ldst_direction dir, MVE_memsz memsz,
3913 string asm, string suffix, bit U, bits<2> size> {
3914 let AM = memsz.AM in {
3915 def "" : MVE_VLDRSTR_cw<
3916 dir, memsz, U, 1, 0, size,
3917 dir.Oops, !con(dir.Iops, (ins taddrmode_imm7<memsz.shift>:$addr)),
3918 asm, suffix, IndexModeNone, "$Qd, $addr", "">;
3920 def _pre : MVE_VLDRSTR_cw<
3921 dir, memsz, U, 1, 1, size,
3922 !con((outs tGPR:$wb), dir.Oops),
3923 !con(dir.Iops, (ins taddrmode_imm7<memsz.shift>:$addr)),
3924 asm, suffix, IndexModePre, "$Qd, $addr!", "$addr.base = $wb"> {
3925 let DecoderMethod = "DecodeMVE_MEM_1_pre<"#memsz.shift#">";
3928 def _post : MVE_VLDRSTR_cw<
3929 dir, memsz, U, 0, 1, size,
3930 !con((outs tGPR:$wb), dir.Oops),
3931 !con(dir.Iops, (ins t_addr_offset_none:$Rn,
3932 t2am_imm7_offset<memsz.shift>:$addr)),
3933 asm, suffix, IndexModePost, "$Qd, $Rn$addr", "$Rn.base = $wb"> {
3935 let Inst{18-16} = Rn{2-0};
3940 multiclass MVE_VLDRSTR_cs_m<MVE_ldst_direction dir, MVE_memsz memsz,
3941 string asm, string suffix> {
3942 let AM = memsz.AM in {
3943 def "" : MVE_VLDRSTR_cs<
3945 dir.Oops, !con(dir.Iops, (ins t2addrmode_imm7<memsz.shift>:$addr)),
3946 asm, suffix, IndexModeNone, "$Qd, $addr", "">;
3948 def _pre : MVE_VLDRSTR_cs<
3950 !con((outs rGPR:$wb), dir.Oops),
3951 !con(dir.Iops, (ins t2addrmode_imm7_pre<memsz.shift>:$addr)),
3952 asm, suffix, IndexModePre, "$Qd, $addr!", "$addr.base = $wb"> {
3953 let DecoderMethod = "DecodeMVE_MEM_2_pre<"#memsz.shift#">";
3956 def _post : MVE_VLDRSTR_cs<
3958 !con((outs rGPR:$wb), dir.Oops),
3959 // We need an !if here to select the base register class,
3960 // because it's legal to write back to SP in a load of this
3961 // type, but not in a store.
3962 !con(dir.Iops, (ins !if(dir.load, t2_addr_offset_none,
3963 t2_nosp_addr_offset_none):$Rn,
3964 t2am_imm7_offset<memsz.shift>:$addr)),
3965 asm, suffix, IndexModePost, "$Qd, $Rn$addr", "$Rn.base = $wb"> {
3967 let Inst{19-16} = Rn{3-0};
3972 // Now actually declare all the contiguous load/stores, via those
3973 // multiclasses. The instruction ids coming out of this are the bare
3974 // names shown in the defm, with _pre or _post appended for writeback,
3975 // e.g. MVE_VLDRBS16, MVE_VSTRB16_pre, MVE_VSTRHU16_post.
3977 defm MVE_VLDRBS16: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memB, "vldrb", "s16", 0, 0b01>;
3978 defm MVE_VLDRBS32: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memB, "vldrb", "s32", 0, 0b10>;
3979 defm MVE_VLDRBU16: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memB, "vldrb", "u16", 1, 0b01>;
3980 defm MVE_VLDRBU32: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memB, "vldrb", "u32", 1, 0b10>;
3981 defm MVE_VLDRHS32: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memH, "vldrh", "s32", 0, 0b10>;
3982 defm MVE_VLDRHU32: MVE_VLDRSTR_cw_m<MVE_ld, MVE_memH, "vldrh", "u32", 1, 0b10>;
3984 defm MVE_VLDRBU8: MVE_VLDRSTR_cs_m<MVE_ld, MVE_memB, "vldrb", "u8">;
3985 defm MVE_VLDRHU16: MVE_VLDRSTR_cs_m<MVE_ld, MVE_memH, "vldrh", "u16">;
3986 defm MVE_VLDRWU32: MVE_VLDRSTR_cs_m<MVE_ld, MVE_memW, "vldrw", "u32">;
3988 defm MVE_VSTRB16: MVE_VLDRSTR_cw_m<MVE_st, MVE_memB, "vstrb", "16", 0, 0b01>;
3989 defm MVE_VSTRB32: MVE_VLDRSTR_cw_m<MVE_st, MVE_memB, "vstrb", "32", 0, 0b10>;
3990 defm MVE_VSTRH32: MVE_VLDRSTR_cw_m<MVE_st, MVE_memH, "vstrh", "32", 0, 0b10>;
3992 defm MVE_VSTRBU8 : MVE_VLDRSTR_cs_m<MVE_st, MVE_memB, "vstrb", "8">;
3993 defm MVE_VSTRHU16: MVE_VLDRSTR_cs_m<MVE_st, MVE_memH, "vstrh", "16">;
3994 defm MVE_VSTRWU32: MVE_VLDRSTR_cs_m<MVE_st, MVE_memW, "vstrw", "32">;
3996 // Gather loads / scatter stores whose address operand is of the form
3997 // [Rn,Qm], i.e. a single GPR as the common base address, plus a
3998 // vector of offset from it. ('Load/store this sequence of elements of
3999 // the same array.')
4001 // Like the contiguous family, these loads and stores can widen the
4002 // loaded values / truncate the stored ones, or they can just
4003 // load/store the same size of memory and vector lane. But unlike the
4004 // contiguous family, there's no particular difference in encoding
4005 // between those two cases.
4007 // This family also comes with the option to scale the offset values
4008 // in Qm by the size of the loaded memory (i.e. to treat them as array
4009 // indices), or not to scale them (to treat them as plain byte offsets
4010 // in memory, so that perhaps the loaded values are unaligned). The
4011 // scaled instructions' address operand in assembly looks like
4012 // [Rn,Qm,UXTW #2] or similar.
4015 class MVE_VLDRSTR_rq<MVE_ldst_direction dir, MVE_memsz memsz, bit U,
4016 bits<2> size, bit os, string asm, string suffix, int shift>
4017 : MVE_VLDRSTR_base<dir, U, 0b0, 0b0, 0, dir.Oops,
4018 !con(dir.Iops, (ins mve_addr_rq_shift<shift>:$addr)),
4019 asm, suffix, "$Qd, $addr", dir.cstr> {
4022 let Inst{19-16} = addr{6-3};
4023 let Inst{8-7} = size;
4024 let Inst{6} = memsz.encoding{1};
4026 let Inst{4} = memsz.encoding{0};
4027 let Inst{3-1} = addr{2-0};
4031 // Multiclass that defines the scaled and unscaled versions of an
4032 // instruction, when the memory size is wider than a byte. The scaled
4033 // version gets the default name like MVE_VLDRBU16_rq; the unscaled /
4034 // potentially unaligned version gets a "_u" suffix, e.g.
4035 // MVE_VLDRBU16_rq_u.
4036 multiclass MVE_VLDRSTR_rq_w<MVE_ldst_direction dir, MVE_memsz memsz,
4037 string asm, string suffix, bit U, bits<2> size> {
4038 def _u : MVE_VLDRSTR_rq<dir, memsz, U, size, 0, asm, suffix, 0>;
4039 def "" : MVE_VLDRSTR_rq<dir, memsz, U, size, 1, asm, suffix, memsz.shift>;
4042 // Subclass of MVE_VLDRSTR_rq with the same API as that multiclass,
4043 // for use when the memory size is one byte, so there's no 'scaled'
4044 // version of the instruction at all. (This is encoded as if it were
4045 // unscaled, but named in the default way with no _u suffix.)
4046 class MVE_VLDRSTR_rq_b<MVE_ldst_direction dir, MVE_memsz memsz,
4047 string asm, string suffix, bit U, bits<2> size>
4048 : MVE_VLDRSTR_rq<dir, memsz, U, size, 0, asm, suffix, 0>;
4050 // Actually define all the loads and stores in this family.
4052 def MVE_VLDRBU8_rq : MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","u8", 1,0b00>;
4053 def MVE_VLDRBU16_rq: MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","u16", 1,0b01>;
4054 def MVE_VLDRBS16_rq: MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","s16", 0,0b01>;
4055 def MVE_VLDRBU32_rq: MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","u32", 1,0b10>;
4056 def MVE_VLDRBS32_rq: MVE_VLDRSTR_rq_b<MVE_ld, MVE_memB, "vldrb","s32", 0,0b10>;
4058 defm MVE_VLDRHU16_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memH, "vldrh","u16", 1,0b01>;
4059 defm MVE_VLDRHU32_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memH, "vldrh","u32", 1,0b10>;
4060 defm MVE_VLDRHS32_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memH, "vldrh","s32", 0,0b10>;
4061 defm MVE_VLDRWU32_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memW, "vldrw","u32", 1,0b10>;
4062 defm MVE_VLDRDU64_rq: MVE_VLDRSTR_rq_w<MVE_ld, MVE_memD, "vldrd","u64", 1,0b11>;
4064 def MVE_VSTRB8_rq : MVE_VLDRSTR_rq_b<MVE_st, MVE_memB, "vstrb","8", 0,0b00>;
4065 def MVE_VSTRB16_rq : MVE_VLDRSTR_rq_b<MVE_st, MVE_memB, "vstrb","16", 0,0b01>;
4066 def MVE_VSTRB32_rq : MVE_VLDRSTR_rq_b<MVE_st, MVE_memB, "vstrb","32", 0,0b10>;
4068 defm MVE_VSTRH16_rq : MVE_VLDRSTR_rq_w<MVE_st, MVE_memH, "vstrh","16", 0,0b01>;
4069 defm MVE_VSTRH32_rq : MVE_VLDRSTR_rq_w<MVE_st, MVE_memH, "vstrh","32", 0,0b10>;
4070 defm MVE_VSTRW32_rq : MVE_VLDRSTR_rq_w<MVE_st, MVE_memW, "vstrw","32", 0,0b10>;
4071 defm MVE_VSTRD64_rq : MVE_VLDRSTR_rq_w<MVE_st, MVE_memD, "vstrd","64", 0,0b11>;
4073 // Gather loads / scatter stores whose address operand is of the form
4074 // [Qm,#imm], i.e. a vector containing a full base address for each
4075 // loaded item, plus an immediate offset applied consistently to all
4076 // of them. ('Load/store the same field from this vector of pointers
4077 // to a structure type.')
4079 // This family requires the vector lane size to be at least 32 bits
4080 // (so there's room for an address in each lane at all). It has no
4081 // widening/narrowing variants. But it does support preindex
4082 // writeback, in which the address vector is updated to hold the
4083 // addresses actually loaded from.
4086 class MVE_VLDRSTR_qi<MVE_ldst_direction dir, MVE_memsz memsz, bit W, dag wbops,
4087 string asm, string wbAsm, string suffix, string cstr = "">
4088 : MVE_VLDRSTR_base<dir, 1, 1, W, 1, !con(wbops, dir.Oops),
4089 !con(dir.Iops, (ins mve_addr_q_shift<memsz.shift>:$addr)),
4090 asm, suffix, "$Qd, $addr" # wbAsm, cstr # dir.cstr> {
4092 let Inst{23} = addr{7};
4093 let Inst{19-17} = addr{10-8};
4095 let Inst{8} = memsz.encoding{0}; // enough to distinguish 32- from 64-bit
4097 let Inst{6-0} = addr{6-0};
4100 // Multiclass that generates the non-writeback and writeback variants.
4101 multiclass MVE_VLDRSTR_qi_m<MVE_ldst_direction dir, MVE_memsz memsz,
4102 string asm, string suffix> {
4103 def "" : MVE_VLDRSTR_qi<dir, memsz, 0, (outs), asm, "", suffix>;
4104 def _pre : MVE_VLDRSTR_qi<dir, memsz, 1, (outs MQPR:$wb), asm, "!", suffix,
4105 "$addr.base = $wb"> {
4106 let DecoderMethod="DecodeMVE_MEM_3_pre<"#memsz.shift#">";
4110 // Actual instruction definitions.
4111 defm MVE_VLDRWU32_qi: MVE_VLDRSTR_qi_m<MVE_ld, MVE_memW, "vldrw", "u32">;
4112 defm MVE_VLDRDU64_qi: MVE_VLDRSTR_qi_m<MVE_ld, MVE_memD, "vldrd", "u64">;
4113 defm MVE_VSTRW32_qi: MVE_VLDRSTR_qi_m<MVE_st, MVE_memW, "vstrw", "32">;
4114 defm MVE_VSTRD64_qi: MVE_VLDRSTR_qi_m<MVE_st, MVE_memD, "vstrd", "64">;
4116 // Define aliases for all the instructions where memory size and
4117 // vector lane size are the same. These are mnemonic aliases, so they
4118 // apply consistently across all of the above families - contiguous
4119 // loads, and both the rq and qi types of gather/scatter.
4121 // Rationale: As long as you're loading (for example) 16-bit memory
4122 // values into 16-bit vector lanes, you can think of them as signed or
4123 // unsigned integers, fp16 or just raw 16-bit blobs and it makes no
4124 // difference. So we permit all of vldrh.16, vldrh.u16, vldrh.s16,
4125 // vldrh.f16 and treat them all as equivalent to the canonical
4126 // spelling (which happens to be .u16 for loads, and just .16 for
4129 foreach vpt_cond = ["", "t", "e"] in
4130 foreach memsz = [MVE_memB, MVE_memH, MVE_memW, MVE_memD] in
4131 foreach suffix = memsz.suffixes in {
4133 // These foreaches are conceptually ifs, implemented by iterating a
4134 // dummy variable over a list with 0 or 1 elements depending on the
4135 // condition. The idea is to iterate over _nearly_ all the suffixes
4136 // in memsz.suffixes, but omit the one we want all the others to alias.
4138 foreach _ = !if(!ne(suffix, memsz.CanonLoadSuffix), [1], []<int>) in
4139 def : MnemonicAlias<
4140 "vldr" # memsz.MnemonicLetter # vpt_cond # suffix,
4141 "vldr" # memsz.MnemonicLetter # vpt_cond # memsz.CanonLoadSuffix>;
4143 foreach _ = !if(!ne(suffix, memsz.CanonStoreSuffix), [1], []<int>) in
4144 def : MnemonicAlias<
4145 "vstr" # memsz.MnemonicLetter # vpt_cond # suffix,
4146 "vstr" # memsz.MnemonicLetter # vpt_cond # memsz.CanonStoreSuffix>;
4149 // end of MVE predicable load/store
4151 class MVE_VPT<string suffix, bits<2> size, dag iops, string asm, list<dag> pattern=[]>
4152 : MVE_MI<(outs ), iops, NoItinerary, !strconcat("vpt", "${Mk}", ".", suffix), asm, "", pattern> {
4157 let Inst{31-23} = 0b111111100;
4158 let Inst{22} = Mk{3};
4159 let Inst{21-20} = size;
4160 let Inst{19-17} = Qn{2-0};
4162 let Inst{15-13} = Mk{2-0};
4163 let Inst{12} = fc{2};
4164 let Inst{11-8} = 0b1111;
4165 let Inst{7} = fc{0};
4168 let Defs = [VPR, P0];
4171 class MVE_VPTt1<string suffix, bits<2> size, dag iops>
4172 : MVE_VPT<suffix, size, iops, "$fc, $Qn, $Qm"> {
4177 let Inst{5} = Qm{3};
4178 let Inst{3-1} = Qm{2-0};
4179 let Inst{0} = fc{1};
4182 class MVE_VPTt1i<string suffix, bits<2> size>
4183 : MVE_VPTt1<suffix, size,
4184 (ins vpt_mask:$Mk, pred_basic_i:$fc, MQPR:$Qn, MQPR:$Qm)> {
4189 def MVE_VPTv4i32 : MVE_VPTt1i<"i32", 0b10>;
4190 def MVE_VPTv8i16 : MVE_VPTt1i<"i16", 0b01>;
4191 def MVE_VPTv16i8 : MVE_VPTt1i<"i8", 0b00>;
4193 class MVE_VPTt1u<string suffix, bits<2> size>
4194 : MVE_VPTt1<suffix, size,
4195 (ins vpt_mask:$Mk, pred_basic_u:$fc, MQPR:$Qn, MQPR:$Qm)> {
4200 def MVE_VPTv4u32 : MVE_VPTt1u<"u32", 0b10>;
4201 def MVE_VPTv8u16 : MVE_VPTt1u<"u16", 0b01>;
4202 def MVE_VPTv16u8 : MVE_VPTt1u<"u8", 0b00>;
4204 class MVE_VPTt1s<string suffix, bits<2> size>
4205 : MVE_VPTt1<suffix, size,
4206 (ins vpt_mask:$Mk, pred_basic_s:$fc, MQPR:$Qn, MQPR:$Qm)> {
4210 def MVE_VPTv4s32 : MVE_VPTt1s<"s32", 0b10>;
4211 def MVE_VPTv8s16 : MVE_VPTt1s<"s16", 0b01>;
4212 def MVE_VPTv16s8 : MVE_VPTt1s<"s8", 0b00>;
4214 class MVE_VPTt2<string suffix, bits<2> size, dag iops>
4215 : MVE_VPT<suffix, size, iops,
4222 let Inst{5} = fc{1};
4223 let Inst{3-0} = Rm{3-0};
4226 class MVE_VPTt2i<string suffix, bits<2> size>
4227 : MVE_VPTt2<suffix, size,
4228 (ins vpt_mask:$Mk, pred_basic_i:$fc, MQPR:$Qn, GPRwithZR:$Rm)> {
4233 def MVE_VPTv4i32r : MVE_VPTt2i<"i32", 0b10>;
4234 def MVE_VPTv8i16r : MVE_VPTt2i<"i16", 0b01>;
4235 def MVE_VPTv16i8r : MVE_VPTt2i<"i8", 0b00>;
4237 class MVE_VPTt2u<string suffix, bits<2> size>
4238 : MVE_VPTt2<suffix, size,
4239 (ins vpt_mask:$Mk, pred_basic_u:$fc, MQPR:$Qn, GPRwithZR:$Rm)> {
4244 def MVE_VPTv4u32r : MVE_VPTt2u<"u32", 0b10>;
4245 def MVE_VPTv8u16r : MVE_VPTt2u<"u16", 0b01>;
4246 def MVE_VPTv16u8r : MVE_VPTt2u<"u8", 0b00>;
4248 class MVE_VPTt2s<string suffix, bits<2> size>
4249 : MVE_VPTt2<suffix, size,
4250 (ins vpt_mask:$Mk, pred_basic_s:$fc, MQPR:$Qn, GPRwithZR:$Rm)> {
4254 def MVE_VPTv4s32r : MVE_VPTt2s<"s32", 0b10>;
4255 def MVE_VPTv8s16r : MVE_VPTt2s<"s16", 0b01>;
4256 def MVE_VPTv16s8r : MVE_VPTt2s<"s8", 0b00>;
4259 class MVE_VPTf<string suffix, bit size, dag iops, string asm, list<dag> pattern=[]>
4260 : MVE_MI<(outs ), iops, NoItinerary, !strconcat("vpt", "${Mk}", ".", suffix), asm,
4266 let Inst{31-29} = 0b111;
4267 let Inst{28} = size;
4268 let Inst{27-23} = 0b11100;
4269 let Inst{22} = Mk{3};
4270 let Inst{21-20} = 0b11;
4271 let Inst{19-17} = Qn{2-0};
4273 let Inst{15-13} = Mk{2-0};
4274 let Inst{12} = fc{2};
4275 let Inst{11-8} = 0b1111;
4276 let Inst{7} = fc{0};
4280 let Predicates = [HasMVEFloat];
4283 class MVE_VPTft1<string suffix, bit size>
4284 : MVE_VPTf<suffix, size, (ins vpt_mask:$Mk, pred_basic_fp:$fc, MQPR:$Qn, MQPR:$Qm),
4290 let Inst{5} = Qm{3};
4291 let Inst{3-1} = Qm{2-0};
4292 let Inst{0} = fc{1};
4295 def MVE_VPTv4f32 : MVE_VPTft1<"f32", 0b0>;
4296 def MVE_VPTv8f16 : MVE_VPTft1<"f16", 0b1>;
4298 class MVE_VPTft2<string suffix, bit size>
4299 : MVE_VPTf<suffix, size, (ins vpt_mask:$Mk, pred_basic_fp:$fc, MQPR:$Qn, GPRwithZR:$Rm),
4305 let Inst{5} = fc{1};
4306 let Inst{3-0} = Rm{3-0};
4309 def MVE_VPTv4f32r : MVE_VPTft2<"f32", 0b0>;
4310 def MVE_VPTv8f16r : MVE_VPTft2<"f16", 0b1>;
4312 def MVE_VPST : MVE_MI<(outs ), (ins vpt_mask:$Mk), NoItinerary,
4313 !strconcat("vpst", "${Mk}"), "", "", []> {
4316 let Inst{31-23} = 0b111111100;
4317 let Inst{22} = Mk{3};
4318 let Inst{21-16} = 0b110001;
4319 let Inst{15-13} = Mk{2-0};
4320 let Inst{12-0} = 0b0111101001101;
4321 let Unpredictable{12} = 0b1;
4322 let Unpredictable{7} = 0b1;
4323 let Unpredictable{5} = 0b1;
4328 def MVE_VPSEL : MVE_p<(outs MQPR:$Qd), (ins MQPR:$Qn, MQPR:$Qm), NoItinerary,
4329 "vpsel", "", "$Qd, $Qn, $Qm", vpred_n, "", []> {
4335 let Inst{25-23} = 0b100;
4336 let Inst{22} = Qd{3};
4337 let Inst{21-20} = 0b11;
4338 let Inst{19-17} = Qn{2-0};
4340 let Inst{15-13} = Qd{2-0};
4341 let Inst{12-9} = 0b0111;
4343 let Inst{7} = Qn{3};
4345 let Inst{5} = Qm{3};
4347 let Inst{3-1} = Qm{2-0};
4351 foreach suffix = ["s8", "s16", "s32", "u8", "u16", "u32",
4352 "i8", "i16", "i32", "f16", "f32"] in
4353 def : MVEInstAlias<"vpsel${vp}." # suffix # "\t$Qd, $Qn, $Qm",
4354 (MVE_VPSEL MQPR:$Qd, MQPR:$Qn, MQPR:$Qm, vpred_n:$vp)>;
4356 def MVE_VPNOT : MVE_p<(outs), (ins), NoItinerary,
4357 "vpnot", "", "", vpred_n, "", []> {
4358 let Inst{31-0} = 0b11111110001100010000111101001101;
4359 let Unpredictable{19-17} = 0b111;
4360 let Unpredictable{12} = 0b1;
4361 let Unpredictable{7} = 0b1;
4362 let Unpredictable{5} = 0b1;
4366 let Constraints = "";
4369 class MVE_loltp_start<dag iops, string asm, string ops, bits<2> size>
4370 : t2LOL<(outs GPRlr:$LR), iops, asm, ops> {
4372 let Predicates = [HasMVEInt];
4374 let Inst{21-20} = size;
4375 let Inst{19-16} = Rn{3-0};
4379 class MVE_DLSTP<string asm, bits<2> size>
4380 : MVE_loltp_start<(ins rGPR:$Rn), asm, "$LR, $Rn", size> {
4382 let Inst{11-1} = 0b00000000000;
4383 let Unpredictable{10-1} = 0b1111111111;
4386 class MVE_WLSTP<string asm, bits<2> size>
4387 : MVE_loltp_start<(ins rGPR:$Rn, wlslabel_u11:$label),
4388 asm, "$LR, $Rn, $label", size> {
4391 let Inst{11} = label{0};
4392 let Inst{10-1} = label{10-1};
4395 def MVE_DLSTP_8 : MVE_DLSTP<"dlstp.8", 0b00>;
4396 def MVE_DLSTP_16 : MVE_DLSTP<"dlstp.16", 0b01>;
4397 def MVE_DLSTP_32 : MVE_DLSTP<"dlstp.32", 0b10>;
4398 def MVE_DLSTP_64 : MVE_DLSTP<"dlstp.64", 0b11>;
4400 def MVE_WLSTP_8 : MVE_WLSTP<"wlstp.8", 0b00>;
4401 def MVE_WLSTP_16 : MVE_WLSTP<"wlstp.16", 0b01>;
4402 def MVE_WLSTP_32 : MVE_WLSTP<"wlstp.32", 0b10>;
4403 def MVE_WLSTP_64 : MVE_WLSTP<"wlstp.64", 0b11>;
4405 class MVE_loltp_end<dag oops, dag iops, string asm, string ops>
4406 : t2LOL<oops, iops, asm, ops> {
4407 let Predicates = [HasMVEInt];
4408 let Inst{22-21} = 0b00;
4409 let Inst{19-16} = 0b1111;
4413 def MVE_LETP : MVE_loltp_end<(outs GPRlr:$LRout),
4414 (ins GPRlr:$LRin, lelabel_u11:$label),
4415 "letp", "$LRin, $label"> {
4419 let Inst{11} = label{0};
4420 let Inst{10-1} = label{10-1};
4423 def MVE_LCTP : MVE_loltp_end<(outs), (ins pred:$p), "lctp${p}", ""> {
4426 let Inst{11-1} = 0b00000000000;
4427 let Unpredictable{21-20} = 0b11;
4428 let Unpredictable{11-1} = 0b11111111111;
4432 //===----------------------------------------------------------------------===//
4434 //===----------------------------------------------------------------------===//
4436 class MVE_unpred_vector_store_typed<ValueType Ty, Instruction RegImmInst,
4437 PatFrag StoreKind, int shift>
4438 : Pat<(StoreKind (Ty MQPR:$val), t2addrmode_imm7<shift>:$addr),
4439 (RegImmInst (Ty MQPR:$val), t2addrmode_imm7<shift>:$addr)>;
4441 multiclass MVE_unpred_vector_store<Instruction RegImmInst, PatFrag StoreKind,
4443 def : MVE_unpred_vector_store_typed<v16i8, RegImmInst, StoreKind, shift>;
4444 def : MVE_unpred_vector_store_typed<v8i16, RegImmInst, StoreKind, shift>;
4445 def : MVE_unpred_vector_store_typed<v8f16, RegImmInst, StoreKind, shift>;
4446 def : MVE_unpred_vector_store_typed<v4i32, RegImmInst, StoreKind, shift>;
4447 def : MVE_unpred_vector_store_typed<v4f32, RegImmInst, StoreKind, shift>;
4448 def : MVE_unpred_vector_store_typed<v2i64, RegImmInst, StoreKind, shift>;
4449 def : MVE_unpred_vector_store_typed<v2f64, RegImmInst, StoreKind, shift>;
4452 class MVE_unpred_vector_load_typed<ValueType Ty, Instruction RegImmInst,
4453 PatFrag LoadKind, int shift>
4454 : Pat<(Ty (LoadKind t2addrmode_imm7<shift>:$addr)),
4455 (Ty (RegImmInst t2addrmode_imm7<shift>:$addr))>;
4457 multiclass MVE_unpred_vector_load<Instruction RegImmInst, PatFrag LoadKind,
4459 def : MVE_unpred_vector_load_typed<v16i8, RegImmInst, LoadKind, shift>;
4460 def : MVE_unpred_vector_load_typed<v8i16, RegImmInst, LoadKind, shift>;
4461 def : MVE_unpred_vector_load_typed<v8f16, RegImmInst, LoadKind, shift>;
4462 def : MVE_unpred_vector_load_typed<v4i32, RegImmInst, LoadKind, shift>;
4463 def : MVE_unpred_vector_load_typed<v4f32, RegImmInst, LoadKind, shift>;
4464 def : MVE_unpred_vector_load_typed<v2i64, RegImmInst, LoadKind, shift>;
4465 def : MVE_unpred_vector_load_typed<v2f64, RegImmInst, LoadKind, shift>;
4468 let Predicates = [HasMVEInt, IsLE] in {
4469 defm : MVE_unpred_vector_store<MVE_VSTRBU8, byte_alignedstore, 0>;
4470 defm : MVE_unpred_vector_store<MVE_VSTRHU16, hword_alignedstore, 1>;
4471 defm : MVE_unpred_vector_store<MVE_VSTRWU32, alignedstore32, 2>;
4473 defm : MVE_unpred_vector_load<MVE_VLDRBU8, byte_alignedload, 0>;
4474 defm : MVE_unpred_vector_load<MVE_VLDRHU16, hword_alignedload, 1>;
4475 defm : MVE_unpred_vector_load<MVE_VLDRWU32, alignedload32, 2>;
4477 def : Pat<(v16i1 (load t2addrmode_imm7<2>:$addr)),
4478 (v16i1 (VLDR_P0_off t2addrmode_imm7<2>:$addr))>;
4479 def : Pat<(v8i1 (load t2addrmode_imm7<2>:$addr)),
4480 (v8i1 (VLDR_P0_off t2addrmode_imm7<2>:$addr))>;
4481 def : Pat<(v4i1 (load t2addrmode_imm7<2>:$addr)),
4482 (v4i1 (VLDR_P0_off t2addrmode_imm7<2>:$addr))>;
4485 let Predicates = [HasMVEInt, IsBE] in {
4486 def : MVE_unpred_vector_store_typed<v16i8, MVE_VSTRBU8, store, 0>;
4487 def : MVE_unpred_vector_store_typed<v8i16, MVE_VSTRHU16, alignedstore16, 1>;
4488 def : MVE_unpred_vector_store_typed<v8f16, MVE_VSTRHU16, alignedstore16, 1>;
4489 def : MVE_unpred_vector_store_typed<v4i32, MVE_VSTRWU32, alignedstore32, 2>;
4490 def : MVE_unpred_vector_store_typed<v4f32, MVE_VSTRWU32, alignedstore32, 2>;
4492 def : MVE_unpred_vector_load_typed<v16i8, MVE_VLDRBU8, load, 0>;
4493 def : MVE_unpred_vector_load_typed<v8i16, MVE_VLDRHU16, alignedload16, 1>;
4494 def : MVE_unpred_vector_load_typed<v8f16, MVE_VLDRHU16, alignedload16, 1>;
4495 def : MVE_unpred_vector_load_typed<v4i32, MVE_VLDRWU32, alignedload32, 2>;
4496 def : MVE_unpred_vector_load_typed<v4f32, MVE_VLDRWU32, alignedload32, 2>;
4500 // Widening/Narrowing Loads/Stores
4502 let Predicates = [HasMVEInt] in {
4503 def : Pat<(truncstorevi8 (v8i16 MQPR:$val), t2addrmode_imm7<1>:$addr),
4504 (MVE_VSTRB16 MQPR:$val, t2addrmode_imm7<1>:$addr)>;
4505 def : Pat<(truncstorevi8 (v4i32 MQPR:$val), t2addrmode_imm7<1>:$addr),
4506 (MVE_VSTRB32 MQPR:$val, t2addrmode_imm7<1>:$addr)>;
4507 def : Pat<(truncstorevi16 (v4i32 MQPR:$val), t2addrmode_imm7<2>:$addr),
4508 (MVE_VSTRH32 MQPR:$val, t2addrmode_imm7<2>:$addr)>;
4511 multiclass MVEExtLoad<string DestLanes, string DestElemBits,
4512 string SrcElemBits, string SrcElemType,
4514 def _Any : Pat<(!cast<ValueType>("v" # DestLanes # "i" # DestElemBits)
4515 (!cast<PatFrag>("extloadvi" # SrcElemBits) am:$addr)),
4516 (!cast<Instruction>("MVE_VLDR" # SrcElemType # "U" # DestElemBits)
4518 def _Z : Pat<(!cast<ValueType>("v" # DestLanes # "i" # DestElemBits)
4519 (!cast<PatFrag>("zextloadvi" # SrcElemBits) am:$addr)),
4520 (!cast<Instruction>("MVE_VLDR" # SrcElemType # "U" # DestElemBits)
4522 def _S : Pat<(!cast<ValueType>("v" # DestLanes # "i" # DestElemBits)
4523 (!cast<PatFrag>("sextloadvi" # SrcElemBits) am:$addr)),
4524 (!cast<Instruction>("MVE_VLDR" # SrcElemType # "S" # DestElemBits)
4528 let Predicates = [HasMVEInt] in {
4529 defm : MVEExtLoad<"4", "32", "8", "B", t2addrmode_imm7<1>>;
4530 defm : MVEExtLoad<"8", "16", "8", "B", t2addrmode_imm7<1>>;
4531 defm : MVEExtLoad<"4", "32", "16", "H", t2addrmode_imm7<2>>;
4535 // Bit convert patterns
4537 let Predicates = [HasMVEInt] in {
4538 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4539 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4541 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4542 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4544 def : Pat<(v8i16 (bitconvert (v8f16 QPR:$src))), (v8i16 QPR:$src)>;
4545 def : Pat<(v8f16 (bitconvert (v8i16 QPR:$src))), (v8f16 QPR:$src)>;
4548 let Predicates = [IsLE,HasMVEInt] in {
4549 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
4550 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4551 def : Pat<(v2f64 (bitconvert (v8f16 QPR:$src))), (v2f64 QPR:$src)>;
4552 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4553 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4555 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4556 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4557 def : Pat<(v2i64 (bitconvert (v8f16 QPR:$src))), (v2i64 QPR:$src)>;
4558 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4559 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4561 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4562 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4563 def : Pat<(v4f32 (bitconvert (v8f16 QPR:$src))), (v4f32 QPR:$src)>;
4564 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4565 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4567 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4568 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4569 def : Pat<(v4i32 (bitconvert (v8f16 QPR:$src))), (v4i32 QPR:$src)>;
4570 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4571 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4573 def : Pat<(v8f16 (bitconvert (v2f64 QPR:$src))), (v8f16 QPR:$src)>;
4574 def : Pat<(v8f16 (bitconvert (v2i64 QPR:$src))), (v8f16 QPR:$src)>;
4575 def : Pat<(v8f16 (bitconvert (v4f32 QPR:$src))), (v8f16 QPR:$src)>;
4576 def : Pat<(v8f16 (bitconvert (v4i32 QPR:$src))), (v8f16 QPR:$src)>;
4577 def : Pat<(v8f16 (bitconvert (v16i8 QPR:$src))), (v8f16 QPR:$src)>;
4579 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4580 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4581 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4582 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4583 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4585 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4586 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4587 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4588 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4589 def : Pat<(v16i8 (bitconvert (v8f16 QPR:$src))), (v16i8 QPR:$src)>;
4590 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;