1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
20 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
21 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
22 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
23 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
24 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
25 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
27 // Types for vector shift by immediates. The "SHX" version is for long and
28 // narrow operations where the source and destination vectors have different
29 // types. The "SHINS" version is for shift and insert operations.
30 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
32 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
34 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
35 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
37 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
38 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
39 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
40 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
41 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
42 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
43 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
45 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
46 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
47 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
49 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
50 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
51 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
52 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
53 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
54 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
56 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
57 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
58 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
60 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
61 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
63 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
65 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
66 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
68 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
69 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
70 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
72 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
74 // VDUPLANE can produce a quad-register result from a double-register source,
75 // so the result is not constrained to match the source.
76 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
77 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
80 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
81 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
82 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
84 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
85 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
86 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
87 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
89 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
92 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
93 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
94 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
96 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
98 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
99 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
101 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
102 SDTCisSameAs<0, 2>]>;
103 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
104 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
106 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
107 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
108 unsigned EltBits = 0;
109 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
110 return (EltBits == 32 && EltVal == 0);
113 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
114 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
115 unsigned EltBits = 0;
116 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
117 return (EltBits == 8 && EltVal == 0xff);
120 //===----------------------------------------------------------------------===//
121 // NEON operand definitions
122 //===----------------------------------------------------------------------===//
124 def nModImm : Operand<i32> {
125 let PrintMethod = "printNEONModImmOperand";
128 //===----------------------------------------------------------------------===//
129 // NEON load / store instructions
130 //===----------------------------------------------------------------------===//
132 // Use VLDM to load a Q register as a D register pair.
133 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
135 : PseudoVFPLdStM<(outs QPR:$dst), (ins addrmode4:$addr), IIC_fpLoad_m, "",
136 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]>;
138 // Use VSTM to store a Q register as a D register pair.
139 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
141 : PseudoVFPLdStM<(outs), (ins QPR:$src, addrmode4:$addr), IIC_fpStore_m, "",
142 [(store (v2f64 QPR:$src), addrmode4:$addr)]>;
144 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
146 // Classes for VLD* pseudo-instructions with multi-register operands.
147 // These are expanded to real instructions after register allocation.
148 class VLDQPseudo<InstrItinClass itin>
149 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
150 class VLDQWBPseudo<InstrItinClass itin>
151 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
152 (ins addrmode6:$addr, am6offset:$offset), itin,
154 class VLDQQPseudo<InstrItinClass itin>
155 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
156 class VLDQQWBPseudo<InstrItinClass itin>
157 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
158 (ins addrmode6:$addr, am6offset:$offset), itin,
160 class VLDQQQQWBPseudo<InstrItinClass itin>
161 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
162 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
163 "$addr.addr = $wb, $src = $dst">;
165 // VLD1 : Vector Load (multiple single elements)
166 class VLD1D<bits<4> op7_4, string Dt>
167 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
168 (ins addrmode6:$Rn), IIC_VLD1,
169 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
173 class VLD1Q<bits<4> op7_4, string Dt>
174 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
175 (ins addrmode6:$Rn), IIC_VLD1x2,
176 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
178 let Inst{5-4} = Rn{5-4};
181 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
182 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
183 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
184 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
186 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
187 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
188 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
189 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
191 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
192 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
193 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
194 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
196 // ...with address register writeback:
197 class VLD1DWB<bits<4> op7_4, string Dt>
198 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
199 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
200 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
201 "$Rn.addr = $wb", []> {
204 class VLD1QWB<bits<4> op7_4, string Dt>
205 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
206 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
207 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
208 "$Rn.addr = $wb", []> {
209 let Inst{5-4} = Rn{5-4};
212 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
213 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
214 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
215 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
217 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
218 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
219 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
220 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
222 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
223 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
224 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
225 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
227 // ...with 3 registers (some of these are only for the disassembler):
228 class VLD1D3<bits<4> op7_4, string Dt>
229 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
230 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
231 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
235 class VLD1D3WB<bits<4> op7_4, string Dt>
236 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
237 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
238 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
242 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
243 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
244 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
245 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
247 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
248 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
249 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
250 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
252 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
253 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
255 // ...with 4 registers (some of these are only for the disassembler):
256 class VLD1D4<bits<4> op7_4, string Dt>
257 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
258 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
259 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
261 let Inst{5-4} = Rn{5-4};
263 class VLD1D4WB<bits<4> op7_4, string Dt>
264 : NLdSt<0,0b10,0b0010,op7_4,
265 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
266 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4, "vld1", Dt,
267 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
269 let Inst{5-4} = Rn{5-4};
272 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
273 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
274 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
275 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
277 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
278 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
279 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
280 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
282 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
283 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
285 // VLD2 : Vector Load (multiple 2-element structures)
286 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
287 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
288 (ins addrmode6:$Rn), IIC_VLD2,
289 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
291 let Inst{5-4} = Rn{5-4};
293 class VLD2Q<bits<4> op7_4, string Dt>
294 : NLdSt<0, 0b10, 0b0011, op7_4,
295 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
296 (ins addrmode6:$Rn), IIC_VLD2x2,
297 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
299 let Inst{5-4} = Rn{5-4};
302 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
303 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
304 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
306 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
307 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
308 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
310 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
311 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
312 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
314 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
315 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
316 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
318 // ...with address register writeback:
319 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
320 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
321 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
322 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
323 "$Rn.addr = $wb", []> {
324 let Inst{5-4} = Rn{5-4};
326 class VLD2QWB<bits<4> op7_4, string Dt>
327 : NLdSt<0, 0b10, 0b0011, op7_4,
328 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
329 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
330 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
331 "$Rn.addr = $wb", []> {
332 let Inst{5-4} = Rn{5-4};
335 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
336 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
337 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
339 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
340 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
341 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
343 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
344 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
345 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
347 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
348 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
349 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
351 // ...with double-spaced registers (for disassembly only):
352 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
353 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
354 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
355 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
356 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
357 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
359 // VLD3 : Vector Load (multiple 3-element structures)
360 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
361 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
362 (ins addrmode6:$Rn), IIC_VLD3,
363 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
368 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
369 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
370 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
372 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
373 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
374 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
376 // ...with address register writeback:
377 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
378 : NLdSt<0, 0b10, op11_8, op7_4,
379 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
380 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
381 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
382 "$Rn.addr = $wb", []> {
386 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
387 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
388 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
390 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
391 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
392 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
394 // ...with double-spaced registers (non-updating versions for disassembly only):
395 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
396 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
397 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
398 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
399 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
400 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
402 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
403 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
404 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
406 // ...alternate versions to be allocated odd register numbers:
407 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
408 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
409 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
411 // VLD4 : Vector Load (multiple 4-element structures)
412 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
413 : NLdSt<0, 0b10, op11_8, op7_4,
414 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
415 (ins addrmode6:$Rn), IIC_VLD4,
416 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
418 let Inst{5-4} = Rn{5-4};
421 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
422 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
423 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
425 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
426 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
427 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
429 // ...with address register writeback:
430 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
431 : NLdSt<0, 0b10, op11_8, op7_4,
432 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
433 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4,
434 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
435 "$Rn.addr = $wb", []> {
436 let Inst{5-4} = Rn{5-4};
439 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
440 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
441 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
443 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
444 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
445 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
447 // ...with double-spaced registers (non-updating versions for disassembly only):
448 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
449 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
450 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
451 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
452 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
453 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
455 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
456 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
457 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
459 // ...alternate versions to be allocated odd register numbers:
460 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
461 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
462 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
464 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
466 // Classes for VLD*LN pseudo-instructions with multi-register operands.
467 // These are expanded to real instructions after register allocation.
468 class VLDQLNPseudo<InstrItinClass itin>
469 : PseudoNLdSt<(outs QPR:$dst),
470 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
471 itin, "$src = $dst">;
472 class VLDQLNWBPseudo<InstrItinClass itin>
473 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
474 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
475 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
476 class VLDQQLNPseudo<InstrItinClass itin>
477 : PseudoNLdSt<(outs QQPR:$dst),
478 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
479 itin, "$src = $dst">;
480 class VLDQQLNWBPseudo<InstrItinClass itin>
481 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
482 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
483 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
484 class VLDQQQQLNPseudo<InstrItinClass itin>
485 : PseudoNLdSt<(outs QQQQPR:$dst),
486 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
487 itin, "$src = $dst">;
488 class VLDQQQQLNWBPseudo<InstrItinClass itin>
489 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
490 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
491 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
493 // VLD1LN : Vector Load (single element to one lane)
494 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
496 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
497 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
498 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
500 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
501 (i32 (LoadOp addrmode6:$Rn)),
506 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
507 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
508 (i32 (LoadOp addrmode6:$addr)),
512 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
513 let Inst{7-5} = lane{2-0};
515 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
516 let Inst{7-6} = lane{1-0};
519 def VLD1LNd32 : VLD1LN<0b1000, {?,0,?,?}, "32", v2i32, load> {
520 let Inst{7} = lane{0};
525 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
526 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
527 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
529 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
531 // ...with address register writeback:
532 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
533 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
534 (ins addrmode6:$Rn, am6offset:$Rm,
535 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
536 "\\{$Vd[$lane]\\}, $Rn$Rm",
537 "$src = $Vd, $Rn.addr = $wb", []> {
541 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
542 let Inst{7-5} = lane{2-0};
544 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
545 let Inst{7-6} = lane{1-0};
548 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
549 let Inst{7} = lane{0};
554 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
555 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
556 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
558 // VLD2LN : Vector Load (single 2-element structure to one lane)
559 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
560 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
561 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
562 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
563 "$src1 = $Vd, $src2 = $dst2", []> {
570 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
571 let Inst{7-5} = lane{2-0};
573 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
574 let Inst{7-6} = lane{1-0};
576 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
577 let Inst{7} = lane{0};
580 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
581 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
582 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
584 // ...with double-spaced registers:
585 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
586 let Inst{7-6} = lane{1-0};
588 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
589 let Inst{7} = lane{0};
592 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
593 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
595 // ...with address register writeback:
596 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
597 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
598 (ins addrmode6:$Rn, am6offset:$Rm,
599 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
600 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
601 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
606 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
607 let Inst{7-5} = lane{2-0};
609 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
610 let Inst{7-6} = lane{1-0};
612 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
613 let Inst{7} = lane{0};
616 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
617 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
618 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
620 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
621 let Inst{7-6} = lane{1-0};
623 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
624 let Inst{7} = lane{0};
627 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
628 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
630 // VLD3LN : Vector Load (single 3-element structure to one lane)
631 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
632 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
633 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
634 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
635 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
636 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
641 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
642 let Inst{7-5} = lane{2-0};
644 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
645 let Inst{7-6} = lane{1-0};
647 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
648 let Inst{7} = lane{0};
651 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
652 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
653 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
655 // ...with double-spaced registers:
656 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
657 let Inst{7-6} = lane{1-0};
659 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
660 let Inst{7} = lane{0};
663 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
664 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
666 // ...with address register writeback:
667 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
668 : NLdSt<1, 0b10, op11_8, op7_4,
669 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
670 (ins addrmode6:$Rn, am6offset:$Rm,
671 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
672 IIC_VLD3lnu, "vld3", Dt,
673 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
674 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
679 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
680 let Inst{7-5} = lane{2-0};
682 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
683 let Inst{7-6} = lane{1-0};
685 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
686 let Inst{7} = lane{0};
689 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
690 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
691 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
693 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
694 let Inst{7-6} = lane{1-0};
696 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
697 let Inst{7} = lane{0};
700 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
701 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
703 // VLD4LN : Vector Load (single 4-element structure to one lane)
704 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
705 : NLdSt<1, 0b10, op11_8, op7_4,
706 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
707 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
708 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
709 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
710 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
717 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
718 let Inst{7-5} = lane{2-0};
720 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
721 let Inst{7-6} = lane{1-0};
723 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
724 let Inst{7} = lane{0};
728 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
729 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
730 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
732 // ...with double-spaced registers:
733 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
734 let Inst{7-6} = lane{1-0};
736 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
737 let Inst{7} = lane{0};
741 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
742 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
744 // ...with address register writeback:
745 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
746 : NLdSt<1, 0b10, op11_8, op7_4,
747 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
748 (ins addrmode6:$Rn, am6offset:$Rm,
749 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
750 IIC_VLD4ln, "vld4", Dt,
751 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
752 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
758 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
759 let Inst{7-5} = lane{2-0};
761 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
762 let Inst{7-6} = lane{1-0};
764 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
765 let Inst{7} = lane{0};
769 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
770 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
771 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
773 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
774 let Inst{7-6} = lane{1-0};
776 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
777 let Inst{7} = lane{0};
781 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
782 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
784 // VLD1DUP : Vector Load (single element to all lanes)
785 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
786 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
787 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
788 // FIXME: Not yet implemented.
789 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
791 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
793 // Classes for VST* pseudo-instructions with multi-register operands.
794 // These are expanded to real instructions after register allocation.
795 class VSTQPseudo<InstrItinClass itin>
796 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
797 class VSTQWBPseudo<InstrItinClass itin>
798 : PseudoNLdSt<(outs GPR:$wb),
799 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
801 class VSTQQPseudo<InstrItinClass itin>
802 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
803 class VSTQQWBPseudo<InstrItinClass itin>
804 : PseudoNLdSt<(outs GPR:$wb),
805 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
807 class VSTQQQQWBPseudo<InstrItinClass itin>
808 : PseudoNLdSt<(outs GPR:$wb),
809 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
812 // VST1 : Vector Store (multiple single elements)
813 class VST1D<bits<4> op7_4, string Dt>
814 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src),
815 IIC_VST1, "vst1", Dt, "\\{$src\\}, $addr", "", []>;
816 class VST1Q<bits<4> op7_4, string Dt>
817 : NLdSt<0,0b00,0b1010,op7_4, (outs),
818 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST1x2,
819 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
821 def VST1d8 : VST1D<0b0000, "8">;
822 def VST1d16 : VST1D<0b0100, "16">;
823 def VST1d32 : VST1D<0b1000, "32">;
824 def VST1d64 : VST1D<0b1100, "64">;
826 def VST1q8 : VST1Q<0b0000, "8">;
827 def VST1q16 : VST1Q<0b0100, "16">;
828 def VST1q32 : VST1Q<0b1000, "32">;
829 def VST1q64 : VST1Q<0b1100, "64">;
831 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
832 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
833 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
834 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
836 // ...with address register writeback:
837 class VST1DWB<bits<4> op7_4, string Dt>
838 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
839 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST1u,
840 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
841 class VST1QWB<bits<4> op7_4, string Dt>
842 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
843 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
844 IIC_VST1x2u, "vst1", Dt, "\\{$src1, $src2\\}, $addr$offset",
845 "$addr.addr = $wb", []>;
847 def VST1d8_UPD : VST1DWB<0b0000, "8">;
848 def VST1d16_UPD : VST1DWB<0b0100, "16">;
849 def VST1d32_UPD : VST1DWB<0b1000, "32">;
850 def VST1d64_UPD : VST1DWB<0b1100, "64">;
852 def VST1q8_UPD : VST1QWB<0b0000, "8">;
853 def VST1q16_UPD : VST1QWB<0b0100, "16">;
854 def VST1q32_UPD : VST1QWB<0b1000, "32">;
855 def VST1q64_UPD : VST1QWB<0b1100, "64">;
857 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
858 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
859 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
860 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
862 // ...with 3 registers (some of these are only for the disassembler):
863 class VST1D3<bits<4> op7_4, string Dt>
864 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
865 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
866 IIC_VST1x3, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
867 class VST1D3WB<bits<4> op7_4, string Dt>
868 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
869 (ins addrmode6:$addr, am6offset:$offset,
870 DPR:$src1, DPR:$src2, DPR:$src3),
871 IIC_VST1x3u, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
872 "$addr.addr = $wb", []>;
874 def VST1d8T : VST1D3<0b0000, "8">;
875 def VST1d16T : VST1D3<0b0100, "16">;
876 def VST1d32T : VST1D3<0b1000, "32">;
877 def VST1d64T : VST1D3<0b1100, "64">;
879 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
880 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
881 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
882 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
884 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
885 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
887 // ...with 4 registers (some of these are only for the disassembler):
888 class VST1D4<bits<4> op7_4, string Dt>
889 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
890 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
891 IIC_VST1x4, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
893 class VST1D4WB<bits<4> op7_4, string Dt>
894 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
895 (ins addrmode6:$addr, am6offset:$offset,
896 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
897 "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
898 "$addr.addr = $wb", []>;
900 def VST1d8Q : VST1D4<0b0000, "8">;
901 def VST1d16Q : VST1D4<0b0100, "16">;
902 def VST1d32Q : VST1D4<0b1000, "32">;
903 def VST1d64Q : VST1D4<0b1100, "64">;
905 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
906 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
907 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
908 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
910 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
911 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
913 // VST2 : Vector Store (multiple 2-element structures)
914 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
915 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
916 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
917 IIC_VST2, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
918 class VST2Q<bits<4> op7_4, string Dt>
919 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
920 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
921 IIC_VST2x2, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
924 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
925 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
926 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
928 def VST2q8 : VST2Q<0b0000, "8">;
929 def VST2q16 : VST2Q<0b0100, "16">;
930 def VST2q32 : VST2Q<0b1000, "32">;
932 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
933 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
934 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
936 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
937 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
938 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
940 // ...with address register writeback:
941 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
942 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
943 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
944 IIC_VST2u, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
945 "$addr.addr = $wb", []>;
946 class VST2QWB<bits<4> op7_4, string Dt>
947 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
948 (ins addrmode6:$addr, am6offset:$offset,
949 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
950 "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
951 "$addr.addr = $wb", []>;
953 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
954 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
955 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
957 def VST2q8_UPD : VST2QWB<0b0000, "8">;
958 def VST2q16_UPD : VST2QWB<0b0100, "16">;
959 def VST2q32_UPD : VST2QWB<0b1000, "32">;
961 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
962 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
963 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
965 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
966 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
967 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
969 // ...with double-spaced registers (for disassembly only):
970 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
971 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
972 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
973 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
974 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
975 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
977 // VST3 : Vector Store (multiple 3-element structures)
978 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
979 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
980 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST3,
981 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
983 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
984 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
985 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
987 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
988 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
989 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
991 // ...with address register writeback:
992 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
993 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
994 (ins addrmode6:$addr, am6offset:$offset,
995 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST3u,
996 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
997 "$addr.addr = $wb", []>;
999 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
1000 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
1001 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
1003 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1004 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1005 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1007 // ...with double-spaced registers (non-updating versions for disassembly only):
1008 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
1009 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
1010 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
1011 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
1012 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
1013 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
1015 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1016 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1017 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1019 // ...alternate versions to be allocated odd register numbers:
1020 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1021 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1022 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1024 // VST4 : Vector Store (multiple 4-element structures)
1025 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1026 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1027 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
1028 IIC_VST4, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
1031 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
1032 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
1033 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
1035 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1036 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1037 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1039 // ...with address register writeback:
1040 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1041 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1042 (ins addrmode6:$addr, am6offset:$offset,
1043 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1044 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
1045 "$addr.addr = $wb", []>;
1047 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
1048 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
1049 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
1051 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1052 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1053 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1055 // ...with double-spaced registers (non-updating versions for disassembly only):
1056 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
1057 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
1058 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
1059 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
1060 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
1061 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
1063 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1064 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1065 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1067 // ...alternate versions to be allocated odd register numbers:
1068 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1069 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1070 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1072 // Classes for VST*LN pseudo-instructions with multi-register operands.
1073 // These are expanded to real instructions after register allocation.
1074 class VSTQLNPseudo<InstrItinClass itin>
1075 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1077 class VSTQLNWBPseudo<InstrItinClass itin>
1078 : PseudoNLdSt<(outs GPR:$wb),
1079 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1080 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1081 class VSTQQLNPseudo<InstrItinClass itin>
1082 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1084 class VSTQQLNWBPseudo<InstrItinClass itin>
1085 : PseudoNLdSt<(outs GPR:$wb),
1086 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1087 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1088 class VSTQQQQLNPseudo<InstrItinClass itin>
1089 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1091 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1092 : PseudoNLdSt<(outs GPR:$wb),
1093 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1094 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1096 // VST1LN : Vector Store (single element from one lane)
1097 // FIXME: Not yet implemented.
1099 // VST2LN : Vector Store (single 2-element structure from one lane)
1100 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1101 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
1102 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1103 IIC_VST2ln, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
1106 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
1107 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
1108 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
1110 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1111 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1112 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1114 // ...with double-spaced registers:
1115 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
1116 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
1118 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1119 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1121 // ...with address register writeback:
1122 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1123 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1124 (ins addrmode6:$addr, am6offset:$offset,
1125 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1126 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1127 "$addr.addr = $wb", []>;
1129 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
1130 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
1131 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
1133 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1134 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1135 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1137 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
1138 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
1140 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1141 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1143 // VST3LN : Vector Store (single 3-element structure from one lane)
1144 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1145 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
1146 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
1147 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1148 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
1150 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
1151 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
1152 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
1154 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1155 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1156 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1158 // ...with double-spaced registers:
1159 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
1160 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
1162 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1163 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1165 // ...with address register writeback:
1166 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1167 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1168 (ins addrmode6:$addr, am6offset:$offset,
1169 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1170 IIC_VST3lnu, "vst3", Dt,
1171 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
1172 "$addr.addr = $wb", []>;
1174 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
1175 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
1176 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
1178 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1179 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1180 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1182 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
1183 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
1185 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1186 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1188 // VST4LN : Vector Store (single 4-element structure from one lane)
1189 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1190 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
1191 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1192 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1193 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
1196 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
1197 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
1198 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
1200 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1201 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1202 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1204 // ...with double-spaced registers:
1205 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
1206 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
1208 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1209 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1211 // ...with address register writeback:
1212 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1213 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1214 (ins addrmode6:$addr, am6offset:$offset,
1215 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1216 IIC_VST4lnu, "vst4", Dt,
1217 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
1218 "$addr.addr = $wb", []>;
1220 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
1221 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
1222 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
1224 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1225 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1226 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1228 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
1229 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
1231 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1232 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1234 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1237 //===----------------------------------------------------------------------===//
1238 // NEON pattern fragments
1239 //===----------------------------------------------------------------------===//
1241 // Extract D sub-registers of Q registers.
1242 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1243 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1244 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1246 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1247 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1248 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1250 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1251 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1252 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1254 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1255 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1256 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1259 // Extract S sub-registers of Q/D registers.
1260 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1261 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1262 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1265 // Translate lane numbers from Q registers to D subregs.
1266 def SubReg_i8_lane : SDNodeXForm<imm, [{
1267 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1269 def SubReg_i16_lane : SDNodeXForm<imm, [{
1270 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1272 def SubReg_i32_lane : SDNodeXForm<imm, [{
1273 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1276 //===----------------------------------------------------------------------===//
1277 // Instruction Classes
1278 //===----------------------------------------------------------------------===//
1280 // Basic 2-register operations: single-, double- and quad-register.
1281 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1282 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1283 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1284 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1285 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1286 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1287 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1288 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1289 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1290 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1291 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1292 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1293 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1294 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1295 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1296 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1297 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1298 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1300 // Basic 2-register intrinsics, both double- and quad-register.
1301 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1302 bits<2> op17_16, bits<5> op11_7, bit op4,
1303 InstrItinClass itin, string OpcodeStr, string Dt,
1304 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1305 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1306 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1307 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1308 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1309 bits<2> op17_16, bits<5> op11_7, bit op4,
1310 InstrItinClass itin, string OpcodeStr, string Dt,
1311 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1312 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1313 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1314 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1316 // Narrow 2-register operations.
1317 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1318 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1319 InstrItinClass itin, string OpcodeStr, string Dt,
1320 ValueType TyD, ValueType TyQ, SDNode OpNode>
1321 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1322 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1323 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1325 // Narrow 2-register intrinsics.
1326 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1327 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1328 InstrItinClass itin, string OpcodeStr, string Dt,
1329 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1330 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1331 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1332 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1334 // Long 2-register operations (currently only used for VMOVL).
1335 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1336 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1337 InstrItinClass itin, string OpcodeStr, string Dt,
1338 ValueType TyQ, ValueType TyD, SDNode OpNode>
1339 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1340 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1341 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1343 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1344 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1345 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1346 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1347 OpcodeStr, Dt, "$dst1, $dst2",
1348 "$src1 = $dst1, $src2 = $dst2", []>;
1349 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1350 InstrItinClass itin, string OpcodeStr, string Dt>
1351 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1352 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1353 "$src1 = $dst1, $src2 = $dst2", []>;
1355 // Basic 3-register operations: single-, double- and quad-register.
1356 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1357 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1358 SDNode OpNode, bit Commutable>
1359 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1360 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1361 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1362 let isCommutable = Commutable;
1365 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1366 InstrItinClass itin, string OpcodeStr, string Dt,
1367 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1368 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1369 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1370 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1371 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1372 let isCommutable = Commutable;
1374 // Same as N3VD but no data type.
1375 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1376 InstrItinClass itin, string OpcodeStr,
1377 ValueType ResTy, ValueType OpTy,
1378 SDNode OpNode, bit Commutable>
1379 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1380 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1381 OpcodeStr, "$dst, $src1, $src2", "",
1382 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1383 let isCommutable = Commutable;
1386 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1387 InstrItinClass itin, string OpcodeStr, string Dt,
1388 ValueType Ty, SDNode ShOp>
1389 : N3V<0, 1, op21_20, op11_8, 1, 0,
1390 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1391 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1392 [(set (Ty DPR:$dst),
1393 (Ty (ShOp (Ty DPR:$src1),
1394 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1395 let isCommutable = 0;
1397 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1398 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1399 : N3V<0, 1, op21_20, op11_8, 1, 0,
1400 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1401 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1402 [(set (Ty DPR:$dst),
1403 (Ty (ShOp (Ty DPR:$src1),
1404 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1405 let isCommutable = 0;
1408 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1409 InstrItinClass itin, string OpcodeStr, string Dt,
1410 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1411 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1412 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1413 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1414 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1415 let isCommutable = Commutable;
1417 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1418 InstrItinClass itin, string OpcodeStr,
1419 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1420 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1421 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1422 OpcodeStr, "$dst, $src1, $src2", "",
1423 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1424 let isCommutable = Commutable;
1426 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1427 InstrItinClass itin, string OpcodeStr, string Dt,
1428 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1429 : N3V<1, 1, op21_20, op11_8, 1, 0,
1430 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1431 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1432 [(set (ResTy QPR:$dst),
1433 (ResTy (ShOp (ResTy QPR:$src1),
1434 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1436 let isCommutable = 0;
1438 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1439 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1440 : N3V<1, 1, op21_20, op11_8, 1, 0,
1441 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1442 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1443 [(set (ResTy QPR:$dst),
1444 (ResTy (ShOp (ResTy QPR:$src1),
1445 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1447 let isCommutable = 0;
1450 // Basic 3-register intrinsics, both double- and quad-register.
1451 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1452 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1453 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1454 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1455 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1456 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1457 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1458 let isCommutable = Commutable;
1460 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1461 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1462 : N3V<0, 1, op21_20, op11_8, 1, 0,
1463 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1464 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1465 [(set (Ty DPR:$dst),
1466 (Ty (IntOp (Ty DPR:$src1),
1467 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1469 let isCommutable = 0;
1471 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1472 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1473 : N3V<0, 1, op21_20, op11_8, 1, 0,
1474 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1475 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1476 [(set (Ty DPR:$dst),
1477 (Ty (IntOp (Ty DPR:$src1),
1478 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1479 let isCommutable = 0;
1481 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1482 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1483 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1484 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1485 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1486 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1487 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1488 let isCommutable = 0;
1491 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1492 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1493 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1494 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1495 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1496 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1497 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1498 let isCommutable = Commutable;
1500 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1501 string OpcodeStr, string Dt,
1502 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1503 : N3V<1, 1, op21_20, op11_8, 1, 0,
1504 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1505 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1506 [(set (ResTy QPR:$dst),
1507 (ResTy (IntOp (ResTy QPR:$src1),
1508 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1510 let isCommutable = 0;
1512 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1513 string OpcodeStr, string Dt,
1514 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1515 : N3V<1, 1, op21_20, op11_8, 1, 0,
1516 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1517 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1518 [(set (ResTy QPR:$dst),
1519 (ResTy (IntOp (ResTy QPR:$src1),
1520 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1522 let isCommutable = 0;
1524 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1525 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1526 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1527 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1528 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1529 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1530 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1531 let isCommutable = 0;
1534 // Multiply-Add/Sub operations: single-, double- and quad-register.
1535 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1536 InstrItinClass itin, string OpcodeStr, string Dt,
1537 ValueType Ty, SDNode MulOp, SDNode OpNode>
1538 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1539 (outs DPR_VFP2:$dst),
1540 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1541 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1543 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1544 InstrItinClass itin, string OpcodeStr, string Dt,
1545 ValueType Ty, SDNode MulOp, SDNode OpNode>
1546 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1547 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1548 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1549 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1550 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1552 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1553 string OpcodeStr, string Dt,
1554 ValueType Ty, SDNode MulOp, SDNode ShOp>
1555 : N3V<0, 1, op21_20, op11_8, 1, 0,
1557 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1559 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1560 [(set (Ty DPR:$dst),
1561 (Ty (ShOp (Ty DPR:$src1),
1562 (Ty (MulOp DPR:$src2,
1563 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1565 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1566 string OpcodeStr, string Dt,
1567 ValueType Ty, SDNode MulOp, SDNode ShOp>
1568 : N3V<0, 1, op21_20, op11_8, 1, 0,
1570 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1572 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1574 (Ty (ShOp (Ty DPR:$src1),
1576 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1579 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1580 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1581 SDNode MulOp, SDNode OpNode>
1582 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1583 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1584 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1585 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1586 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1587 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1588 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1589 SDNode MulOp, SDNode ShOp>
1590 : N3V<1, 1, op21_20, op11_8, 1, 0,
1592 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1594 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1595 [(set (ResTy QPR:$dst),
1596 (ResTy (ShOp (ResTy QPR:$src1),
1597 (ResTy (MulOp QPR:$src2,
1598 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1600 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1601 string OpcodeStr, string Dt,
1602 ValueType ResTy, ValueType OpTy,
1603 SDNode MulOp, SDNode ShOp>
1604 : N3V<1, 1, op21_20, op11_8, 1, 0,
1606 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1608 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1609 [(set (ResTy QPR:$dst),
1610 (ResTy (ShOp (ResTy QPR:$src1),
1611 (ResTy (MulOp QPR:$src2,
1612 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1615 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1616 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1617 InstrItinClass itin, string OpcodeStr, string Dt,
1618 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1619 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1620 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1621 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1622 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1623 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1624 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1625 InstrItinClass itin, string OpcodeStr, string Dt,
1626 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1627 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1628 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1629 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1630 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1631 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1633 // Neon 3-argument intrinsics, both double- and quad-register.
1634 // The destination register is also used as the first source operand register.
1635 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1636 InstrItinClass itin, string OpcodeStr, string Dt,
1637 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1638 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1639 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1640 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1641 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1642 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1643 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1644 InstrItinClass itin, string OpcodeStr, string Dt,
1645 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1646 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1647 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1648 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1649 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1650 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1652 // Long Multiply-Add/Sub operations.
1653 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1654 InstrItinClass itin, string OpcodeStr, string Dt,
1655 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1656 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1657 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1658 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1659 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1660 (TyQ (MulOp (TyD DPR:$Vn),
1661 (TyD DPR:$Vm)))))]>;
1662 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1663 InstrItinClass itin, string OpcodeStr, string Dt,
1664 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1665 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1666 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1668 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1670 (OpNode (TyQ QPR:$src1),
1671 (TyQ (MulOp (TyD DPR:$src2),
1672 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1674 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1675 InstrItinClass itin, string OpcodeStr, string Dt,
1676 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1677 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1678 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1680 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1682 (OpNode (TyQ QPR:$src1),
1683 (TyQ (MulOp (TyD DPR:$src2),
1684 (TyD (NEONvduplane (TyD DPR_8:$src3),
1687 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1688 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1689 InstrItinClass itin, string OpcodeStr, string Dt,
1690 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1692 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1693 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1694 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1695 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1696 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
1697 (TyD DPR:$Vm)))))))]>;
1699 // Neon Long 3-argument intrinsic. The destination register is
1700 // a quad-register and is also used as the first source operand register.
1701 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1702 InstrItinClass itin, string OpcodeStr, string Dt,
1703 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1704 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1705 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1706 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1708 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
1709 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1710 string OpcodeStr, string Dt,
1711 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1712 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1714 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1716 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1717 [(set (ResTy QPR:$dst),
1718 (ResTy (IntOp (ResTy QPR:$src1),
1720 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1722 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1723 InstrItinClass itin, string OpcodeStr, string Dt,
1724 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1725 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1727 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1729 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1730 [(set (ResTy QPR:$dst),
1731 (ResTy (IntOp (ResTy QPR:$src1),
1733 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1736 // Narrowing 3-register intrinsics.
1737 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1738 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1739 Intrinsic IntOp, bit Commutable>
1740 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1741 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1742 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1743 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1744 let isCommutable = Commutable;
1747 // Long 3-register operations.
1748 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1749 InstrItinClass itin, string OpcodeStr, string Dt,
1750 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1751 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1752 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1753 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1754 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1755 let isCommutable = Commutable;
1757 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1758 InstrItinClass itin, string OpcodeStr, string Dt,
1759 ValueType TyQ, ValueType TyD, SDNode OpNode>
1760 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1761 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1762 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1764 (TyQ (OpNode (TyD DPR:$src1),
1765 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1766 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1767 InstrItinClass itin, string OpcodeStr, string Dt,
1768 ValueType TyQ, ValueType TyD, SDNode OpNode>
1769 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1770 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1771 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1773 (TyQ (OpNode (TyD DPR:$src1),
1774 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1776 // Long 3-register operations with explicitly extended operands.
1777 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1778 InstrItinClass itin, string OpcodeStr, string Dt,
1779 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1781 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1782 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
1783 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
1784 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
1785 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1786 let isCommutable = Commutable;
1789 // Long 3-register intrinsics with explicit extend (VABDL).
1790 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1791 InstrItinClass itin, string OpcodeStr, string Dt,
1792 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1794 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1795 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1796 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1797 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
1798 (TyD DPR:$src2))))))]> {
1799 let isCommutable = Commutable;
1802 // Long 3-register intrinsics.
1803 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1804 InstrItinClass itin, string OpcodeStr, string Dt,
1805 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1806 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1807 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1808 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1809 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1810 let isCommutable = Commutable;
1812 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1813 string OpcodeStr, string Dt,
1814 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1815 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1816 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1817 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1818 [(set (ResTy QPR:$dst),
1819 (ResTy (IntOp (OpTy DPR:$src1),
1820 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1822 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1823 InstrItinClass itin, string OpcodeStr, string Dt,
1824 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1825 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1826 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1827 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1828 [(set (ResTy QPR:$dst),
1829 (ResTy (IntOp (OpTy DPR:$src1),
1830 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1833 // Wide 3-register operations.
1834 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1835 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1836 SDNode OpNode, SDNode ExtOp, bit Commutable>
1837 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1838 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
1839 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
1840 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
1841 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1842 let isCommutable = Commutable;
1845 // Pairwise long 2-register intrinsics, both double- and quad-register.
1846 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1847 bits<2> op17_16, bits<5> op11_7, bit op4,
1848 string OpcodeStr, string Dt,
1849 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1850 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1851 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1852 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1853 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1854 bits<2> op17_16, bits<5> op11_7, bit op4,
1855 string OpcodeStr, string Dt,
1856 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1857 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1858 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1859 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1861 // Pairwise long 2-register accumulate intrinsics,
1862 // both double- and quad-register.
1863 // The destination register is also used as the first source operand register.
1864 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1865 bits<2> op17_16, bits<5> op11_7, bit op4,
1866 string OpcodeStr, string Dt,
1867 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1868 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1869 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
1870 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
1871 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
1872 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1873 bits<2> op17_16, bits<5> op11_7, bit op4,
1874 string OpcodeStr, string Dt,
1875 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1876 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1877 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
1878 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
1879 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
1881 // Shift by immediate,
1882 // both double- and quad-register.
1883 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1884 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1885 ValueType Ty, SDNode OpNode>
1886 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1887 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1888 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1889 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1890 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1891 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1892 ValueType Ty, SDNode OpNode>
1893 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1894 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1895 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1896 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1898 // Long shift by immediate.
1899 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1900 string OpcodeStr, string Dt,
1901 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1902 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1903 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1904 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1905 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1906 (i32 imm:$SIMM))))]>;
1908 // Narrow shift by immediate.
1909 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1910 InstrItinClass itin, string OpcodeStr, string Dt,
1911 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1912 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1913 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1914 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1915 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1916 (i32 imm:$SIMM))))]>;
1918 // Shift right by immediate and accumulate,
1919 // both double- and quad-register.
1920 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1921 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1922 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
1923 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1924 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
1925 [(set DPR:$Vd, (Ty (add DPR:$src1,
1926 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
1927 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1928 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1929 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
1930 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1931 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
1932 [(set QPR:$Vd, (Ty (add QPR:$src1,
1933 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
1935 // Shift by immediate and insert,
1936 // both double- and quad-register.
1937 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1938 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1939 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
1940 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
1941 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
1942 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
1943 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1944 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1945 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
1946 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
1947 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
1948 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
1950 // Convert, with fractional bits immediate,
1951 // both double- and quad-register.
1952 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1953 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1955 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1956 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
1957 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
1958 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
1959 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1960 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1962 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1963 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
1964 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
1965 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
1967 //===----------------------------------------------------------------------===//
1969 //===----------------------------------------------------------------------===//
1971 // Abbreviations used in multiclass suffixes:
1972 // Q = quarter int (8 bit) elements
1973 // H = half int (16 bit) elements
1974 // S = single int (32 bit) elements
1975 // D = double int (64 bit) elements
1977 // Neon 2-register vector operations -- for disassembly only.
1979 // First with only element sizes of 8, 16 and 32 bits:
1980 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1981 bits<5> op11_7, bit op4, string opc, string Dt,
1983 // 64-bit vector types.
1984 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1985 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1986 opc, !strconcat(Dt, "8"), asm, "", []>;
1987 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1988 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1989 opc, !strconcat(Dt, "16"), asm, "", []>;
1990 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1991 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1992 opc, !strconcat(Dt, "32"), asm, "", []>;
1993 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1994 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1995 opc, "f32", asm, "", []> {
1996 let Inst{10} = 1; // overwrite F = 1
1999 // 128-bit vector types.
2000 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2001 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2002 opc, !strconcat(Dt, "8"), asm, "", []>;
2003 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2004 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2005 opc, !strconcat(Dt, "16"), asm, "", []>;
2006 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2007 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2008 opc, !strconcat(Dt, "32"), asm, "", []>;
2009 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2010 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2011 opc, "f32", asm, "", []> {
2012 let Inst{10} = 1; // overwrite F = 1
2016 // Neon 3-register vector operations.
2018 // First with only element sizes of 8, 16 and 32 bits:
2019 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2020 InstrItinClass itinD16, InstrItinClass itinD32,
2021 InstrItinClass itinQ16, InstrItinClass itinQ32,
2022 string OpcodeStr, string Dt,
2023 SDNode OpNode, bit Commutable = 0> {
2024 // 64-bit vector types.
2025 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2026 OpcodeStr, !strconcat(Dt, "8"),
2027 v8i8, v8i8, OpNode, Commutable>;
2028 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2029 OpcodeStr, !strconcat(Dt, "16"),
2030 v4i16, v4i16, OpNode, Commutable>;
2031 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2032 OpcodeStr, !strconcat(Dt, "32"),
2033 v2i32, v2i32, OpNode, Commutable>;
2035 // 128-bit vector types.
2036 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2037 OpcodeStr, !strconcat(Dt, "8"),
2038 v16i8, v16i8, OpNode, Commutable>;
2039 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2040 OpcodeStr, !strconcat(Dt, "16"),
2041 v8i16, v8i16, OpNode, Commutable>;
2042 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2043 OpcodeStr, !strconcat(Dt, "32"),
2044 v4i32, v4i32, OpNode, Commutable>;
2047 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2048 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2050 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2052 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2053 v8i16, v4i16, ShOp>;
2054 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2055 v4i32, v2i32, ShOp>;
2058 // ....then also with element size 64 bits:
2059 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2060 InstrItinClass itinD, InstrItinClass itinQ,
2061 string OpcodeStr, string Dt,
2062 SDNode OpNode, bit Commutable = 0>
2063 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2064 OpcodeStr, Dt, OpNode, Commutable> {
2065 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2066 OpcodeStr, !strconcat(Dt, "64"),
2067 v1i64, v1i64, OpNode, Commutable>;
2068 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2069 OpcodeStr, !strconcat(Dt, "64"),
2070 v2i64, v2i64, OpNode, Commutable>;
2074 // Neon Narrowing 2-register vector operations,
2075 // source operand element sizes of 16, 32 and 64 bits:
2076 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2077 bits<5> op11_7, bit op6, bit op4,
2078 InstrItinClass itin, string OpcodeStr, string Dt,
2080 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2081 itin, OpcodeStr, !strconcat(Dt, "16"),
2082 v8i8, v8i16, OpNode>;
2083 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2084 itin, OpcodeStr, !strconcat(Dt, "32"),
2085 v4i16, v4i32, OpNode>;
2086 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2087 itin, OpcodeStr, !strconcat(Dt, "64"),
2088 v2i32, v2i64, OpNode>;
2091 // Neon Narrowing 2-register vector intrinsics,
2092 // source operand element sizes of 16, 32 and 64 bits:
2093 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2094 bits<5> op11_7, bit op6, bit op4,
2095 InstrItinClass itin, string OpcodeStr, string Dt,
2097 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2098 itin, OpcodeStr, !strconcat(Dt, "16"),
2099 v8i8, v8i16, IntOp>;
2100 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2101 itin, OpcodeStr, !strconcat(Dt, "32"),
2102 v4i16, v4i32, IntOp>;
2103 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2104 itin, OpcodeStr, !strconcat(Dt, "64"),
2105 v2i32, v2i64, IntOp>;
2109 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2110 // source operand element sizes of 16, 32 and 64 bits:
2111 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2112 string OpcodeStr, string Dt, SDNode OpNode> {
2113 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2114 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2115 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2116 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2117 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2118 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2122 // Neon 3-register vector intrinsics.
2124 // First with only element sizes of 16 and 32 bits:
2125 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2126 InstrItinClass itinD16, InstrItinClass itinD32,
2127 InstrItinClass itinQ16, InstrItinClass itinQ32,
2128 string OpcodeStr, string Dt,
2129 Intrinsic IntOp, bit Commutable = 0> {
2130 // 64-bit vector types.
2131 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2132 OpcodeStr, !strconcat(Dt, "16"),
2133 v4i16, v4i16, IntOp, Commutable>;
2134 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2135 OpcodeStr, !strconcat(Dt, "32"),
2136 v2i32, v2i32, IntOp, Commutable>;
2138 // 128-bit vector types.
2139 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2140 OpcodeStr, !strconcat(Dt, "16"),
2141 v8i16, v8i16, IntOp, Commutable>;
2142 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2143 OpcodeStr, !strconcat(Dt, "32"),
2144 v4i32, v4i32, IntOp, Commutable>;
2146 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2147 InstrItinClass itinD16, InstrItinClass itinD32,
2148 InstrItinClass itinQ16, InstrItinClass itinQ32,
2149 string OpcodeStr, string Dt,
2151 // 64-bit vector types.
2152 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2153 OpcodeStr, !strconcat(Dt, "16"),
2154 v4i16, v4i16, IntOp>;
2155 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2156 OpcodeStr, !strconcat(Dt, "32"),
2157 v2i32, v2i32, IntOp>;
2159 // 128-bit vector types.
2160 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2161 OpcodeStr, !strconcat(Dt, "16"),
2162 v8i16, v8i16, IntOp>;
2163 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2164 OpcodeStr, !strconcat(Dt, "32"),
2165 v4i32, v4i32, IntOp>;
2168 multiclass N3VIntSL_HS<bits<4> op11_8,
2169 InstrItinClass itinD16, InstrItinClass itinD32,
2170 InstrItinClass itinQ16, InstrItinClass itinQ32,
2171 string OpcodeStr, string Dt, Intrinsic IntOp> {
2172 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2173 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2174 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2175 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2176 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2177 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2178 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2179 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2182 // ....then also with element size of 8 bits:
2183 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2184 InstrItinClass itinD16, InstrItinClass itinD32,
2185 InstrItinClass itinQ16, InstrItinClass itinQ32,
2186 string OpcodeStr, string Dt,
2187 Intrinsic IntOp, bit Commutable = 0>
2188 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2189 OpcodeStr, Dt, IntOp, Commutable> {
2190 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2191 OpcodeStr, !strconcat(Dt, "8"),
2192 v8i8, v8i8, IntOp, Commutable>;
2193 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2194 OpcodeStr, !strconcat(Dt, "8"),
2195 v16i8, v16i8, IntOp, Commutable>;
2197 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2198 InstrItinClass itinD16, InstrItinClass itinD32,
2199 InstrItinClass itinQ16, InstrItinClass itinQ32,
2200 string OpcodeStr, string Dt,
2202 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2203 OpcodeStr, Dt, IntOp> {
2204 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2205 OpcodeStr, !strconcat(Dt, "8"),
2207 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2208 OpcodeStr, !strconcat(Dt, "8"),
2209 v16i8, v16i8, IntOp>;
2213 // ....then also with element size of 64 bits:
2214 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2215 InstrItinClass itinD16, InstrItinClass itinD32,
2216 InstrItinClass itinQ16, InstrItinClass itinQ32,
2217 string OpcodeStr, string Dt,
2218 Intrinsic IntOp, bit Commutable = 0>
2219 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2220 OpcodeStr, Dt, IntOp, Commutable> {
2221 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2222 OpcodeStr, !strconcat(Dt, "64"),
2223 v1i64, v1i64, IntOp, Commutable>;
2224 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2225 OpcodeStr, !strconcat(Dt, "64"),
2226 v2i64, v2i64, IntOp, Commutable>;
2228 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2229 InstrItinClass itinD16, InstrItinClass itinD32,
2230 InstrItinClass itinQ16, InstrItinClass itinQ32,
2231 string OpcodeStr, string Dt,
2233 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2234 OpcodeStr, Dt, IntOp> {
2235 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2236 OpcodeStr, !strconcat(Dt, "64"),
2237 v1i64, v1i64, IntOp>;
2238 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2239 OpcodeStr, !strconcat(Dt, "64"),
2240 v2i64, v2i64, IntOp>;
2243 // Neon Narrowing 3-register vector intrinsics,
2244 // source operand element sizes of 16, 32 and 64 bits:
2245 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2246 string OpcodeStr, string Dt,
2247 Intrinsic IntOp, bit Commutable = 0> {
2248 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2249 OpcodeStr, !strconcat(Dt, "16"),
2250 v8i8, v8i16, IntOp, Commutable>;
2251 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2252 OpcodeStr, !strconcat(Dt, "32"),
2253 v4i16, v4i32, IntOp, Commutable>;
2254 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2255 OpcodeStr, !strconcat(Dt, "64"),
2256 v2i32, v2i64, IntOp, Commutable>;
2260 // Neon Long 3-register vector operations.
2262 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2263 InstrItinClass itin16, InstrItinClass itin32,
2264 string OpcodeStr, string Dt,
2265 SDNode OpNode, bit Commutable = 0> {
2266 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2267 OpcodeStr, !strconcat(Dt, "8"),
2268 v8i16, v8i8, OpNode, Commutable>;
2269 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2270 OpcodeStr, !strconcat(Dt, "16"),
2271 v4i32, v4i16, OpNode, Commutable>;
2272 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2273 OpcodeStr, !strconcat(Dt, "32"),
2274 v2i64, v2i32, OpNode, Commutable>;
2277 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2278 InstrItinClass itin, string OpcodeStr, string Dt,
2280 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2281 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2282 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2283 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2286 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2287 InstrItinClass itin16, InstrItinClass itin32,
2288 string OpcodeStr, string Dt,
2289 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2290 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2291 OpcodeStr, !strconcat(Dt, "8"),
2292 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2293 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2294 OpcodeStr, !strconcat(Dt, "16"),
2295 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2296 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2297 OpcodeStr, !strconcat(Dt, "32"),
2298 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2301 // Neon Long 3-register vector intrinsics.
2303 // First with only element sizes of 16 and 32 bits:
2304 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2305 InstrItinClass itin16, InstrItinClass itin32,
2306 string OpcodeStr, string Dt,
2307 Intrinsic IntOp, bit Commutable = 0> {
2308 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2309 OpcodeStr, !strconcat(Dt, "16"),
2310 v4i32, v4i16, IntOp, Commutable>;
2311 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2312 OpcodeStr, !strconcat(Dt, "32"),
2313 v2i64, v2i32, IntOp, Commutable>;
2316 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2317 InstrItinClass itin, string OpcodeStr, string Dt,
2319 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2320 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2321 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2322 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2325 // ....then also with element size of 8 bits:
2326 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2327 InstrItinClass itin16, InstrItinClass itin32,
2328 string OpcodeStr, string Dt,
2329 Intrinsic IntOp, bit Commutable = 0>
2330 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2331 IntOp, Commutable> {
2332 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2333 OpcodeStr, !strconcat(Dt, "8"),
2334 v8i16, v8i8, IntOp, Commutable>;
2337 // ....with explicit extend (VABDL).
2338 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2339 InstrItinClass itin, string OpcodeStr, string Dt,
2340 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2341 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2342 OpcodeStr, !strconcat(Dt, "8"),
2343 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2344 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2345 OpcodeStr, !strconcat(Dt, "16"),
2346 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2347 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2348 OpcodeStr, !strconcat(Dt, "32"),
2349 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2353 // Neon Wide 3-register vector intrinsics,
2354 // source operand element sizes of 8, 16 and 32 bits:
2355 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2356 string OpcodeStr, string Dt,
2357 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2358 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2359 OpcodeStr, !strconcat(Dt, "8"),
2360 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2361 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2362 OpcodeStr, !strconcat(Dt, "16"),
2363 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2364 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2365 OpcodeStr, !strconcat(Dt, "32"),
2366 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2370 // Neon Multiply-Op vector operations,
2371 // element sizes of 8, 16 and 32 bits:
2372 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2373 InstrItinClass itinD16, InstrItinClass itinD32,
2374 InstrItinClass itinQ16, InstrItinClass itinQ32,
2375 string OpcodeStr, string Dt, SDNode OpNode> {
2376 // 64-bit vector types.
2377 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2378 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2379 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2380 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2381 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2382 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2384 // 128-bit vector types.
2385 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2386 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2387 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2388 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2389 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2390 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2393 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2394 InstrItinClass itinD16, InstrItinClass itinD32,
2395 InstrItinClass itinQ16, InstrItinClass itinQ32,
2396 string OpcodeStr, string Dt, SDNode ShOp> {
2397 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2398 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2399 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2400 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2401 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2402 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2404 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2405 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2409 // Neon Intrinsic-Op vector operations,
2410 // element sizes of 8, 16 and 32 bits:
2411 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2412 InstrItinClass itinD, InstrItinClass itinQ,
2413 string OpcodeStr, string Dt, Intrinsic IntOp,
2415 // 64-bit vector types.
2416 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2417 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2418 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2419 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2420 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2421 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2423 // 128-bit vector types.
2424 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2425 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2426 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2427 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2428 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2429 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2432 // Neon 3-argument intrinsics,
2433 // element sizes of 8, 16 and 32 bits:
2434 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2435 InstrItinClass itinD, InstrItinClass itinQ,
2436 string OpcodeStr, string Dt, Intrinsic IntOp> {
2437 // 64-bit vector types.
2438 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2439 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2440 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2441 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2442 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2443 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2445 // 128-bit vector types.
2446 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2447 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2448 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2449 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2450 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2451 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2455 // Neon Long Multiply-Op vector operations,
2456 // element sizes of 8, 16 and 32 bits:
2457 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2458 InstrItinClass itin16, InstrItinClass itin32,
2459 string OpcodeStr, string Dt, SDNode MulOp,
2461 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2462 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2463 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2464 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2465 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2466 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2469 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2470 string Dt, SDNode MulOp, SDNode OpNode> {
2471 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2472 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2473 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2474 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2478 // Neon Long 3-argument intrinsics.
2480 // First with only element sizes of 16 and 32 bits:
2481 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2482 InstrItinClass itin16, InstrItinClass itin32,
2483 string OpcodeStr, string Dt, Intrinsic IntOp> {
2484 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2485 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2486 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2487 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2490 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2491 string OpcodeStr, string Dt, Intrinsic IntOp> {
2492 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2493 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2494 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2495 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2498 // ....then also with element size of 8 bits:
2499 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2500 InstrItinClass itin16, InstrItinClass itin32,
2501 string OpcodeStr, string Dt, Intrinsic IntOp>
2502 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2503 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2504 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2507 // ....with explicit extend (VABAL).
2508 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2509 InstrItinClass itin, string OpcodeStr, string Dt,
2510 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2511 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2512 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2513 IntOp, ExtOp, OpNode>;
2514 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2515 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2516 IntOp, ExtOp, OpNode>;
2517 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2518 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2519 IntOp, ExtOp, OpNode>;
2523 // Neon 2-register vector intrinsics,
2524 // element sizes of 8, 16 and 32 bits:
2525 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2526 bits<5> op11_7, bit op4,
2527 InstrItinClass itinD, InstrItinClass itinQ,
2528 string OpcodeStr, string Dt, Intrinsic IntOp> {
2529 // 64-bit vector types.
2530 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2531 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2532 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2533 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2534 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2535 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2537 // 128-bit vector types.
2538 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2539 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2540 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2541 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2542 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2543 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2547 // Neon Pairwise long 2-register intrinsics,
2548 // element sizes of 8, 16 and 32 bits:
2549 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2550 bits<5> op11_7, bit op4,
2551 string OpcodeStr, string Dt, Intrinsic IntOp> {
2552 // 64-bit vector types.
2553 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2554 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2555 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2556 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2557 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2558 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2560 // 128-bit vector types.
2561 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2562 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2563 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2564 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2565 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2566 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2570 // Neon Pairwise long 2-register accumulate intrinsics,
2571 // element sizes of 8, 16 and 32 bits:
2572 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2573 bits<5> op11_7, bit op4,
2574 string OpcodeStr, string Dt, Intrinsic IntOp> {
2575 // 64-bit vector types.
2576 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2577 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2578 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2579 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2580 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2581 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2583 // 128-bit vector types.
2584 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2585 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2586 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2587 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2588 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2589 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2593 // Neon 2-register vector shift by immediate,
2594 // with f of either N2RegVShLFrm or N2RegVShRFrm
2595 // element sizes of 8, 16, 32 and 64 bits:
2596 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2597 InstrItinClass itin, string OpcodeStr, string Dt,
2598 SDNode OpNode, Format f> {
2599 // 64-bit vector types.
2600 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2601 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2602 let Inst{21-19} = 0b001; // imm6 = 001xxx
2604 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2605 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2606 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2608 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2609 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2610 let Inst{21} = 0b1; // imm6 = 1xxxxx
2612 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2613 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2616 // 128-bit vector types.
2617 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2618 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2619 let Inst{21-19} = 0b001; // imm6 = 001xxx
2621 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2622 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2623 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2625 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2626 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2627 let Inst{21} = 0b1; // imm6 = 1xxxxx
2629 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2630 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2634 // Neon Shift-Accumulate vector operations,
2635 // element sizes of 8, 16, 32 and 64 bits:
2636 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2637 string OpcodeStr, string Dt, SDNode ShOp> {
2638 // 64-bit vector types.
2639 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2640 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2641 let Inst{21-19} = 0b001; // imm6 = 001xxx
2643 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2644 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2645 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2647 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2648 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2649 let Inst{21} = 0b1; // imm6 = 1xxxxx
2651 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2652 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2655 // 128-bit vector types.
2656 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2657 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2658 let Inst{21-19} = 0b001; // imm6 = 001xxx
2660 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2661 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2662 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2664 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2665 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2666 let Inst{21} = 0b1; // imm6 = 1xxxxx
2668 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2669 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2674 // Neon Shift-Insert vector operations,
2675 // with f of either N2RegVShLFrm or N2RegVShRFrm
2676 // element sizes of 8, 16, 32 and 64 bits:
2677 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2678 string OpcodeStr, SDNode ShOp,
2680 // 64-bit vector types.
2681 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2682 f, OpcodeStr, "8", v8i8, ShOp> {
2683 let Inst{21-19} = 0b001; // imm6 = 001xxx
2685 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2686 f, OpcodeStr, "16", v4i16, ShOp> {
2687 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2689 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2690 f, OpcodeStr, "32", v2i32, ShOp> {
2691 let Inst{21} = 0b1; // imm6 = 1xxxxx
2693 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2694 f, OpcodeStr, "64", v1i64, ShOp>;
2697 // 128-bit vector types.
2698 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2699 f, OpcodeStr, "8", v16i8, ShOp> {
2700 let Inst{21-19} = 0b001; // imm6 = 001xxx
2702 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2703 f, OpcodeStr, "16", v8i16, ShOp> {
2704 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2706 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2707 f, OpcodeStr, "32", v4i32, ShOp> {
2708 let Inst{21} = 0b1; // imm6 = 1xxxxx
2710 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2711 f, OpcodeStr, "64", v2i64, ShOp>;
2715 // Neon Shift Long operations,
2716 // element sizes of 8, 16, 32 bits:
2717 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2718 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2719 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2720 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2721 let Inst{21-19} = 0b001; // imm6 = 001xxx
2723 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2724 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2725 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2727 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2728 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2729 let Inst{21} = 0b1; // imm6 = 1xxxxx
2733 // Neon Shift Narrow operations,
2734 // element sizes of 16, 32, 64 bits:
2735 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2736 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2738 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2739 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2740 let Inst{21-19} = 0b001; // imm6 = 001xxx
2742 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2743 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2744 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2746 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2747 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2748 let Inst{21} = 0b1; // imm6 = 1xxxxx
2752 //===----------------------------------------------------------------------===//
2753 // Instruction Definitions.
2754 //===----------------------------------------------------------------------===//
2756 // Vector Add Operations.
2758 // VADD : Vector Add (integer and floating-point)
2759 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2761 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2762 v2f32, v2f32, fadd, 1>;
2763 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2764 v4f32, v4f32, fadd, 1>;
2765 // VADDL : Vector Add Long (Q = D + D)
2766 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2767 "vaddl", "s", add, sext, 1>;
2768 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2769 "vaddl", "u", add, zext, 1>;
2770 // VADDW : Vector Add Wide (Q = Q + D)
2771 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2772 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2773 // VHADD : Vector Halving Add
2774 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2775 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2776 "vhadd", "s", int_arm_neon_vhadds, 1>;
2777 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2778 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2779 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2780 // VRHADD : Vector Rounding Halving Add
2781 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2782 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2783 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2784 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2785 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2786 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2787 // VQADD : Vector Saturating Add
2788 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2789 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2790 "vqadd", "s", int_arm_neon_vqadds, 1>;
2791 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2792 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2793 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2794 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2795 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2796 int_arm_neon_vaddhn, 1>;
2797 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2798 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2799 int_arm_neon_vraddhn, 1>;
2801 // Vector Multiply Operations.
2803 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2804 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2805 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2806 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2807 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2808 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2809 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2810 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
2811 v2f32, v2f32, fmul, 1>;
2812 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
2813 v4f32, v4f32, fmul, 1>;
2814 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2815 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2816 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2819 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2820 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2821 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2822 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2823 (DSubReg_i16_reg imm:$lane))),
2824 (SubReg_i16_lane imm:$lane)))>;
2825 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2826 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2827 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2828 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2829 (DSubReg_i32_reg imm:$lane))),
2830 (SubReg_i32_lane imm:$lane)))>;
2831 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2832 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2833 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2834 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2835 (DSubReg_i32_reg imm:$lane))),
2836 (SubReg_i32_lane imm:$lane)))>;
2838 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2839 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2840 IIC_VMULi16Q, IIC_VMULi32Q,
2841 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2842 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2843 IIC_VMULi16Q, IIC_VMULi32Q,
2844 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2845 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2846 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2848 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2849 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2850 (DSubReg_i16_reg imm:$lane))),
2851 (SubReg_i16_lane imm:$lane)))>;
2852 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2853 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2855 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2856 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2857 (DSubReg_i32_reg imm:$lane))),
2858 (SubReg_i32_lane imm:$lane)))>;
2860 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2861 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2862 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2863 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2864 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2865 IIC_VMULi16Q, IIC_VMULi32Q,
2866 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2867 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2868 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2870 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2871 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2872 (DSubReg_i16_reg imm:$lane))),
2873 (SubReg_i16_lane imm:$lane)))>;
2874 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2875 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2877 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2878 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2879 (DSubReg_i32_reg imm:$lane))),
2880 (SubReg_i32_lane imm:$lane)))>;
2882 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2883 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2884 "vmull", "s", NEONvmulls, 1>;
2885 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2886 "vmull", "u", NEONvmullu, 1>;
2887 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2888 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2889 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
2890 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
2892 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2893 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2894 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2895 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2896 "vqdmull", "s", int_arm_neon_vqdmull>;
2898 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2900 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2901 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2902 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2903 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2905 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2907 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2908 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2909 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2911 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2912 v4f32, v2f32, fmul, fadd>;
2914 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2915 (mul (v8i16 QPR:$src2),
2916 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2917 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2918 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2919 (DSubReg_i16_reg imm:$lane))),
2920 (SubReg_i16_lane imm:$lane)))>;
2922 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2923 (mul (v4i32 QPR:$src2),
2924 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2925 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2926 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2927 (DSubReg_i32_reg imm:$lane))),
2928 (SubReg_i32_lane imm:$lane)))>;
2930 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2931 (fmul (v4f32 QPR:$src2),
2932 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2933 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2935 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2936 (DSubReg_i32_reg imm:$lane))),
2937 (SubReg_i32_lane imm:$lane)))>;
2939 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2940 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2941 "vmlal", "s", NEONvmulls, add>;
2942 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2943 "vmlal", "u", NEONvmullu, add>;
2945 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
2946 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
2948 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2949 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2950 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2951 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2953 // VMLS : Vector Multiply Subtract (integer and floating-point)
2954 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2955 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2956 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2958 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2960 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2961 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2962 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2964 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2965 v4f32, v2f32, fmul, fsub>;
2967 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2968 (mul (v8i16 QPR:$src2),
2969 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2970 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2971 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2972 (DSubReg_i16_reg imm:$lane))),
2973 (SubReg_i16_lane imm:$lane)))>;
2975 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2976 (mul (v4i32 QPR:$src2),
2977 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2978 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2979 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2980 (DSubReg_i32_reg imm:$lane))),
2981 (SubReg_i32_lane imm:$lane)))>;
2983 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2984 (fmul (v4f32 QPR:$src2),
2985 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2986 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2987 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2988 (DSubReg_i32_reg imm:$lane))),
2989 (SubReg_i32_lane imm:$lane)))>;
2991 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2992 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2993 "vmlsl", "s", NEONvmulls, sub>;
2994 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2995 "vmlsl", "u", NEONvmullu, sub>;
2997 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
2998 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3000 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3001 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3002 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3003 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3005 // Vector Subtract Operations.
3007 // VSUB : Vector Subtract (integer and floating-point)
3008 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3009 "vsub", "i", sub, 0>;
3010 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3011 v2f32, v2f32, fsub, 0>;
3012 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3013 v4f32, v4f32, fsub, 0>;
3014 // VSUBL : Vector Subtract Long (Q = D - D)
3015 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3016 "vsubl", "s", sub, sext, 0>;
3017 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3018 "vsubl", "u", sub, zext, 0>;
3019 // VSUBW : Vector Subtract Wide (Q = Q - D)
3020 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3021 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3022 // VHSUB : Vector Halving Subtract
3023 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3024 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3025 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3026 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3027 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3028 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3029 // VQSUB : Vector Saturing Subtract
3030 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3031 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3032 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3033 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3034 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3035 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3036 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3037 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3038 int_arm_neon_vsubhn, 0>;
3039 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3040 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3041 int_arm_neon_vrsubhn, 0>;
3043 // Vector Comparisons.
3045 // VCEQ : Vector Compare Equal
3046 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3047 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3048 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3050 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3052 // For disassembly only.
3053 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3056 // VCGE : Vector Compare Greater Than or Equal
3057 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3058 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3059 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3060 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3061 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3063 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3065 // For disassembly only.
3066 // FIXME: This instruction's encoding MAY NOT BE correct.
3067 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3069 // For disassembly only.
3070 // FIXME: This instruction's encoding MAY NOT BE correct.
3071 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3074 // VCGT : Vector Compare Greater Than
3075 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3076 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3077 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3078 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3079 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3081 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3083 // For disassembly only.
3084 // FIXME: This instruction's encoding MAY NOT BE correct.
3085 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3087 // For disassembly only.
3088 // FIXME: This instruction's encoding MAY NOT BE correct.
3089 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3092 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3093 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3094 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3095 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3096 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3097 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3098 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3099 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3100 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3101 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3102 // VTST : Vector Test Bits
3103 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3104 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3106 // Vector Bitwise Operations.
3108 def vnotd : PatFrag<(ops node:$in),
3109 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3110 def vnotq : PatFrag<(ops node:$in),
3111 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3114 // VAND : Vector Bitwise AND
3115 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3116 v2i32, v2i32, and, 1>;
3117 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3118 v4i32, v4i32, and, 1>;
3120 // VEOR : Vector Bitwise Exclusive OR
3121 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3122 v2i32, v2i32, xor, 1>;
3123 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3124 v4i32, v4i32, xor, 1>;
3126 // VORR : Vector Bitwise OR
3127 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3128 v2i32, v2i32, or, 1>;
3129 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3130 v4i32, v4i32, or, 1>;
3132 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3133 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3134 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3135 "vbic", "$dst, $src1, $src2", "",
3136 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3137 (vnotd DPR:$src2))))]>;
3138 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3139 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3140 "vbic", "$dst, $src1, $src2", "",
3141 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3142 (vnotq QPR:$src2))))]>;
3144 // VORN : Vector Bitwise OR NOT
3145 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3146 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3147 "vorn", "$dst, $src1, $src2", "",
3148 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3149 (vnotd DPR:$src2))))]>;
3150 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3151 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3152 "vorn", "$dst, $src1, $src2", "",
3153 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3154 (vnotq QPR:$src2))))]>;
3156 // VMVN : Vector Bitwise NOT (Immediate)
3158 let isReMaterializable = 1 in {
3160 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3161 (ins nModImm:$SIMM), IIC_VMOVImm,
3162 "vmvn", "i16", "$dst, $SIMM", "",
3163 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3164 let Inst{9} = SIMM{9};
3167 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3168 (ins nModImm:$SIMM), IIC_VMOVImm,
3169 "vmvn", "i16", "$dst, $SIMM", "",
3170 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3171 let Inst{9} = SIMM{9};
3174 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3175 (ins nModImm:$SIMM), IIC_VMOVImm,
3176 "vmvn", "i32", "$dst, $SIMM", "",
3177 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3178 let Inst{11-8} = SIMM{11-8};
3181 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3182 (ins nModImm:$SIMM), IIC_VMOVImm,
3183 "vmvn", "i32", "$dst, $SIMM", "",
3184 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3185 let Inst{11-8} = SIMM{11-8};
3189 // VMVN : Vector Bitwise NOT
3190 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3191 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3192 "vmvn", "$dst, $src", "",
3193 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3194 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3195 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3196 "vmvn", "$dst, $src", "",
3197 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3198 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3199 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3201 // VBSL : Vector Bitwise Select
3202 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3203 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3204 N3RegFrm, IIC_VCNTiD,
3205 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3207 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3208 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3209 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3210 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3211 N3RegFrm, IIC_VCNTiQ,
3212 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3214 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3215 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3217 // VBIF : Vector Bitwise Insert if False
3218 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3219 // FIXME: This instruction's encoding MAY NOT BE correct.
3220 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3221 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3222 N3RegFrm, IIC_VBINiD,
3223 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3224 [/* For disassembly only; pattern left blank */]>;
3225 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3226 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3227 N3RegFrm, IIC_VBINiQ,
3228 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3229 [/* For disassembly only; pattern left blank */]>;
3231 // VBIT : Vector Bitwise Insert if True
3232 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3233 // FIXME: This instruction's encoding MAY NOT BE correct.
3234 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3235 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3236 N3RegFrm, IIC_VBINiD,
3237 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3238 [/* For disassembly only; pattern left blank */]>;
3239 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3240 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3241 N3RegFrm, IIC_VBINiQ,
3242 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3243 [/* For disassembly only; pattern left blank */]>;
3245 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3246 // for equivalent operations with different register constraints; it just
3249 // Vector Absolute Differences.
3251 // VABD : Vector Absolute Difference
3252 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3253 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3254 "vabd", "s", int_arm_neon_vabds, 1>;
3255 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3256 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3257 "vabd", "u", int_arm_neon_vabdu, 1>;
3258 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3259 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3260 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3261 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3263 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3264 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3265 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3266 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3267 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3269 // VABA : Vector Absolute Difference and Accumulate
3270 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3271 "vaba", "s", int_arm_neon_vabds, add>;
3272 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3273 "vaba", "u", int_arm_neon_vabdu, add>;
3275 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3276 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3277 "vabal", "s", int_arm_neon_vabds, zext, add>;
3278 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3279 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3281 // Vector Maximum and Minimum.
3283 // VMAX : Vector Maximum
3284 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3285 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3286 "vmax", "s", int_arm_neon_vmaxs, 1>;
3287 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3288 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3289 "vmax", "u", int_arm_neon_vmaxu, 1>;
3290 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3292 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3293 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3295 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3297 // VMIN : Vector Minimum
3298 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3299 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3300 "vmin", "s", int_arm_neon_vmins, 1>;
3301 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3302 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3303 "vmin", "u", int_arm_neon_vminu, 1>;
3304 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3306 v2f32, v2f32, int_arm_neon_vmins, 1>;
3307 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3309 v4f32, v4f32, int_arm_neon_vmins, 1>;
3311 // Vector Pairwise Operations.
3313 // VPADD : Vector Pairwise Add
3314 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3316 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3317 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3319 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3320 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3322 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3323 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3324 IIC_VPBIND, "vpadd", "f32",
3325 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3327 // VPADDL : Vector Pairwise Add Long
3328 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3329 int_arm_neon_vpaddls>;
3330 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3331 int_arm_neon_vpaddlu>;
3333 // VPADAL : Vector Pairwise Add and Accumulate Long
3334 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3335 int_arm_neon_vpadals>;
3336 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3337 int_arm_neon_vpadalu>;
3339 // VPMAX : Vector Pairwise Maximum
3340 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3341 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3342 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3343 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3344 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3345 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3346 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3347 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3348 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3349 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3350 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3351 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3352 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3353 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3355 // VPMIN : Vector Pairwise Minimum
3356 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3357 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3358 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3359 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3360 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3361 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3362 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3363 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3364 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3365 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3366 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3367 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3368 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3369 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3371 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3373 // VRECPE : Vector Reciprocal Estimate
3374 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3375 IIC_VUNAD, "vrecpe", "u32",
3376 v2i32, v2i32, int_arm_neon_vrecpe>;
3377 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3378 IIC_VUNAQ, "vrecpe", "u32",
3379 v4i32, v4i32, int_arm_neon_vrecpe>;
3380 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3381 IIC_VUNAD, "vrecpe", "f32",
3382 v2f32, v2f32, int_arm_neon_vrecpe>;
3383 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3384 IIC_VUNAQ, "vrecpe", "f32",
3385 v4f32, v4f32, int_arm_neon_vrecpe>;
3387 // VRECPS : Vector Reciprocal Step
3388 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3389 IIC_VRECSD, "vrecps", "f32",
3390 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3391 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3392 IIC_VRECSQ, "vrecps", "f32",
3393 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3395 // VRSQRTE : Vector Reciprocal Square Root Estimate
3396 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3397 IIC_VUNAD, "vrsqrte", "u32",
3398 v2i32, v2i32, int_arm_neon_vrsqrte>;
3399 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3400 IIC_VUNAQ, "vrsqrte", "u32",
3401 v4i32, v4i32, int_arm_neon_vrsqrte>;
3402 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3403 IIC_VUNAD, "vrsqrte", "f32",
3404 v2f32, v2f32, int_arm_neon_vrsqrte>;
3405 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3406 IIC_VUNAQ, "vrsqrte", "f32",
3407 v4f32, v4f32, int_arm_neon_vrsqrte>;
3409 // VRSQRTS : Vector Reciprocal Square Root Step
3410 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3411 IIC_VRECSD, "vrsqrts", "f32",
3412 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3413 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3414 IIC_VRECSQ, "vrsqrts", "f32",
3415 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3419 // VSHL : Vector Shift
3420 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3421 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3422 "vshl", "s", int_arm_neon_vshifts>;
3423 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3424 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3425 "vshl", "u", int_arm_neon_vshiftu>;
3426 // VSHL : Vector Shift Left (Immediate)
3427 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3429 // VSHR : Vector Shift Right (Immediate)
3430 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3432 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3435 // VSHLL : Vector Shift Left Long
3436 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3437 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3439 // VSHLL : Vector Shift Left Long (with maximum shift count)
3440 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3441 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3442 ValueType OpTy, SDNode OpNode>
3443 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3444 ResTy, OpTy, OpNode> {
3445 let Inst{21-16} = op21_16;
3447 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3448 v8i16, v8i8, NEONvshlli>;
3449 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3450 v4i32, v4i16, NEONvshlli>;
3451 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3452 v2i64, v2i32, NEONvshlli>;
3454 // VSHRN : Vector Shift Right and Narrow
3455 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3458 // VRSHL : Vector Rounding Shift
3459 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3460 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3461 "vrshl", "s", int_arm_neon_vrshifts>;
3462 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3463 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3464 "vrshl", "u", int_arm_neon_vrshiftu>;
3465 // VRSHR : Vector Rounding Shift Right
3466 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3468 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3471 // VRSHRN : Vector Rounding Shift Right and Narrow
3472 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3475 // VQSHL : Vector Saturating Shift
3476 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3477 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3478 "vqshl", "s", int_arm_neon_vqshifts>;
3479 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3480 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3481 "vqshl", "u", int_arm_neon_vqshiftu>;
3482 // VQSHL : Vector Saturating Shift Left (Immediate)
3483 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3485 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3487 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3488 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3491 // VQSHRN : Vector Saturating Shift Right and Narrow
3492 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3494 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3497 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3498 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3501 // VQRSHL : Vector Saturating Rounding Shift
3502 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3503 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3504 "vqrshl", "s", int_arm_neon_vqrshifts>;
3505 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3506 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3507 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3509 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3510 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3512 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3515 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3516 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3519 // VSRA : Vector Shift Right and Accumulate
3520 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3521 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3522 // VRSRA : Vector Rounding Shift Right and Accumulate
3523 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3524 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3526 // VSLI : Vector Shift Left and Insert
3527 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3528 // VSRI : Vector Shift Right and Insert
3529 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3531 // Vector Absolute and Saturating Absolute.
3533 // VABS : Vector Absolute Value
3534 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3535 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3537 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3538 IIC_VUNAD, "vabs", "f32",
3539 v2f32, v2f32, int_arm_neon_vabs>;
3540 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3541 IIC_VUNAQ, "vabs", "f32",
3542 v4f32, v4f32, int_arm_neon_vabs>;
3544 // VQABS : Vector Saturating Absolute Value
3545 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3546 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3547 int_arm_neon_vqabs>;
3551 def vnegd : PatFrag<(ops node:$in),
3552 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3553 def vnegq : PatFrag<(ops node:$in),
3554 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3556 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3557 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3558 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3559 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3560 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3561 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3562 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3563 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3565 // VNEG : Vector Negate (integer)
3566 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3567 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3568 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3569 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3570 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3571 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3573 // VNEG : Vector Negate (floating-point)
3574 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3575 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3576 "vneg", "f32", "$dst, $src", "",
3577 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3578 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3579 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3580 "vneg", "f32", "$dst, $src", "",
3581 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3583 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3584 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3585 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3586 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3587 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3588 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3590 // VQNEG : Vector Saturating Negate
3591 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3592 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3593 int_arm_neon_vqneg>;
3595 // Vector Bit Counting Operations.
3597 // VCLS : Vector Count Leading Sign Bits
3598 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3599 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3601 // VCLZ : Vector Count Leading Zeros
3602 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3603 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3605 // VCNT : Vector Count One Bits
3606 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3607 IIC_VCNTiD, "vcnt", "8",
3608 v8i8, v8i8, int_arm_neon_vcnt>;
3609 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3610 IIC_VCNTiQ, "vcnt", "8",
3611 v16i8, v16i8, int_arm_neon_vcnt>;
3613 // Vector Swap -- for disassembly only.
3614 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3615 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3616 "vswp", "$dst, $src", "", []>;
3617 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3618 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3619 "vswp", "$dst, $src", "", []>;
3621 // Vector Move Operations.
3623 // VMOV : Vector Move (Register)
3625 let neverHasSideEffects = 1 in {
3626 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3627 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3628 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3629 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3631 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3632 // be expanded after register allocation is completed.
3633 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3634 NoItinerary, "", []>;
3636 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3637 NoItinerary, "", []>;
3638 } // neverHasSideEffects
3640 // VMOV : Vector Move (Immediate)
3642 let isReMaterializable = 1 in {
3643 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3644 (ins nModImm:$SIMM), IIC_VMOVImm,
3645 "vmov", "i8", "$dst, $SIMM", "",
3646 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3647 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3648 (ins nModImm:$SIMM), IIC_VMOVImm,
3649 "vmov", "i8", "$dst, $SIMM", "",
3650 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3652 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3653 (ins nModImm:$SIMM), IIC_VMOVImm,
3654 "vmov", "i16", "$dst, $SIMM", "",
3655 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
3656 let Inst{9} = SIMM{9};
3659 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3660 (ins nModImm:$SIMM), IIC_VMOVImm,
3661 "vmov", "i16", "$dst, $SIMM", "",
3662 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
3663 let Inst{9} = SIMM{9};
3666 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3667 (ins nModImm:$SIMM), IIC_VMOVImm,
3668 "vmov", "i32", "$dst, $SIMM", "",
3669 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
3670 let Inst{11-8} = SIMM{11-8};
3673 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3674 (ins nModImm:$SIMM), IIC_VMOVImm,
3675 "vmov", "i32", "$dst, $SIMM", "",
3676 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
3677 let Inst{11-8} = SIMM{11-8};
3680 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3681 (ins nModImm:$SIMM), IIC_VMOVImm,
3682 "vmov", "i64", "$dst, $SIMM", "",
3683 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3684 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3685 (ins nModImm:$SIMM), IIC_VMOVImm,
3686 "vmov", "i64", "$dst, $SIMM", "",
3687 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3688 } // isReMaterializable
3690 // VMOV : Vector Get Lane (move scalar to ARM core register)
3692 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3693 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3694 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
3695 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
3697 let Inst{21} = lane{2};
3698 let Inst{6-5} = lane{1-0};
3700 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3701 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3702 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
3703 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
3705 let Inst{21} = lane{1};
3706 let Inst{6} = lane{0};
3708 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3709 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3710 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
3711 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
3713 let Inst{21} = lane{2};
3714 let Inst{6-5} = lane{1-0};
3716 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3717 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3718 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
3719 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
3721 let Inst{21} = lane{1};
3722 let Inst{6} = lane{0};
3724 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3725 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3726 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
3727 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
3729 let Inst{21} = lane{0};
3731 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3732 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3733 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3734 (DSubReg_i8_reg imm:$lane))),
3735 (SubReg_i8_lane imm:$lane))>;
3736 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3737 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3738 (DSubReg_i16_reg imm:$lane))),
3739 (SubReg_i16_lane imm:$lane))>;
3740 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3741 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3742 (DSubReg_i8_reg imm:$lane))),
3743 (SubReg_i8_lane imm:$lane))>;
3744 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3745 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3746 (DSubReg_i16_reg imm:$lane))),
3747 (SubReg_i16_lane imm:$lane))>;
3748 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3749 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3750 (DSubReg_i32_reg imm:$lane))),
3751 (SubReg_i32_lane imm:$lane))>;
3752 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3753 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3754 (SSubReg_f32_reg imm:$src2))>;
3755 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3756 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3757 (SSubReg_f32_reg imm:$src2))>;
3758 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3759 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3760 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3761 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3764 // VMOV : Vector Set Lane (move ARM core register to scalar)
3766 let Constraints = "$src1 = $V" in {
3767 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
3768 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3769 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
3770 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
3771 GPR:$R, imm:$lane))]> {
3772 let Inst{21} = lane{2};
3773 let Inst{6-5} = lane{1-0};
3775 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
3776 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3777 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
3778 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
3779 GPR:$R, imm:$lane))]> {
3780 let Inst{21} = lane{1};
3781 let Inst{6} = lane{0};
3783 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
3784 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3785 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
3786 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
3787 GPR:$R, imm:$lane))]> {
3788 let Inst{21} = lane{0};
3791 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3792 (v16i8 (INSERT_SUBREG QPR:$src1,
3793 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3794 (DSubReg_i8_reg imm:$lane))),
3795 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3796 (DSubReg_i8_reg imm:$lane)))>;
3797 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3798 (v8i16 (INSERT_SUBREG QPR:$src1,
3799 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3800 (DSubReg_i16_reg imm:$lane))),
3801 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3802 (DSubReg_i16_reg imm:$lane)))>;
3803 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3804 (v4i32 (INSERT_SUBREG QPR:$src1,
3805 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3806 (DSubReg_i32_reg imm:$lane))),
3807 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3808 (DSubReg_i32_reg imm:$lane)))>;
3810 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3811 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3812 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3813 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3814 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3815 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3817 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3818 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3819 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3820 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3822 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3823 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3824 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3825 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3826 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3827 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3829 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3830 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3831 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3832 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3833 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3834 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3836 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3837 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3838 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3840 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3841 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3842 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3844 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3845 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3846 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3849 // VDUP : Vector Duplicate (from ARM core register to all elements)
3851 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3852 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3853 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3854 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3855 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3856 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3857 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3858 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3860 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3861 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3862 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3863 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3864 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3865 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3867 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3868 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3869 [(set DPR:$dst, (v2f32 (NEONvdup
3870 (f32 (bitconvert GPR:$src)))))]>;
3871 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3872 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3873 [(set QPR:$dst, (v4f32 (NEONvdup
3874 (f32 (bitconvert GPR:$src)))))]>;
3876 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3878 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3880 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3881 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3882 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3884 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3885 ValueType ResTy, ValueType OpTy>
3886 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3887 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
3888 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3891 // Inst{19-16} is partially specified depending on the element size.
3893 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
3894 let Inst{19-17} = lane{2-0};
3896 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
3897 let Inst{19-18} = lane{1-0};
3899 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
3900 let Inst{19} = lane{0};
3902 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
3903 let Inst{19} = lane{0};
3905 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
3906 let Inst{19-17} = lane{2-0};
3908 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
3909 let Inst{19-18} = lane{1-0};
3911 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
3912 let Inst{19} = lane{0};
3914 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
3915 let Inst{19} = lane{0};
3918 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3919 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3920 (DSubReg_i8_reg imm:$lane))),
3921 (SubReg_i8_lane imm:$lane)))>;
3922 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3923 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3924 (DSubReg_i16_reg imm:$lane))),
3925 (SubReg_i16_lane imm:$lane)))>;
3926 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3927 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3928 (DSubReg_i32_reg imm:$lane))),
3929 (SubReg_i32_lane imm:$lane)))>;
3930 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3931 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3932 (DSubReg_i32_reg imm:$lane))),
3933 (SubReg_i32_lane imm:$lane)))>;
3935 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
3936 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3937 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
3938 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3940 // VMOVN : Vector Narrowing Move
3941 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
3942 "vmovn", "i", trunc>;
3943 // VQMOVN : Vector Saturating Narrowing Move
3944 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3945 "vqmovn", "s", int_arm_neon_vqmovns>;
3946 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3947 "vqmovn", "u", int_arm_neon_vqmovnu>;
3948 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3949 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3950 // VMOVL : Vector Lengthening Move
3951 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
3952 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
3954 // Vector Conversions.
3956 // VCVT : Vector Convert Between Floating-Point and Integers
3957 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3958 v2i32, v2f32, fp_to_sint>;
3959 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3960 v2i32, v2f32, fp_to_uint>;
3961 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3962 v2f32, v2i32, sint_to_fp>;
3963 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3964 v2f32, v2i32, uint_to_fp>;
3966 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3967 v4i32, v4f32, fp_to_sint>;
3968 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3969 v4i32, v4f32, fp_to_uint>;
3970 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3971 v4f32, v4i32, sint_to_fp>;
3972 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3973 v4f32, v4i32, uint_to_fp>;
3975 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3976 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3977 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3978 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3979 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3980 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3981 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3982 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3983 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3985 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3986 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3987 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3988 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3989 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3990 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3991 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3992 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3996 // VREV64 : Vector Reverse elements within 64-bit doublewords
3998 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3999 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
4000 (ins DPR:$src), IIC_VMOVD,
4001 OpcodeStr, Dt, "$dst, $src", "",
4002 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
4003 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4004 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
4005 (ins QPR:$src), IIC_VMOVQ,
4006 OpcodeStr, Dt, "$dst, $src", "",
4007 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
4009 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4010 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4011 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4012 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4014 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4015 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4016 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4017 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4019 // VREV32 : Vector Reverse elements within 32-bit words
4021 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4022 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
4023 (ins DPR:$src), IIC_VMOVD,
4024 OpcodeStr, Dt, "$dst, $src", "",
4025 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
4026 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4027 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
4028 (ins QPR:$src), IIC_VMOVQ,
4029 OpcodeStr, Dt, "$dst, $src", "",
4030 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
4032 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4033 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4035 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4036 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4038 // VREV16 : Vector Reverse elements within 16-bit halfwords
4040 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4041 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
4042 (ins DPR:$src), IIC_VMOVD,
4043 OpcodeStr, Dt, "$dst, $src", "",
4044 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
4045 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4046 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
4047 (ins QPR:$src), IIC_VMOVQ,
4048 OpcodeStr, Dt, "$dst, $src", "",
4049 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
4051 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4052 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4054 // Other Vector Shuffles.
4056 // VEXT : Vector Extract
4058 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4059 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
4060 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
4061 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4062 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
4063 (Ty DPR:$rhs), imm:$index)))]> {
4065 let Inst{11-8} = index{3-0};
4068 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4069 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
4070 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
4071 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4072 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
4073 (Ty QPR:$rhs), imm:$index)))]> {
4075 let Inst{11-8} = index{3-0};
4078 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
4079 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
4080 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
4081 def VEXTdf : VEXTd<"vext", "32", v2f32>;
4083 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
4084 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
4085 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
4086 def VEXTqf : VEXTq<"vext", "32", v4f32>;
4088 // VTRN : Vector Transpose
4090 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4091 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4092 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4094 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4095 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4096 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4098 // VUZP : Vector Unzip (Deinterleave)
4100 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4101 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4102 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4104 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4105 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4106 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4108 // VZIP : Vector Zip (Interleave)
4110 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4111 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4112 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4114 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4115 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4116 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4118 // Vector Table Lookup and Table Extension.
4120 // VTBL : Vector Table Lookup
4122 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4123 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4124 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4125 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4126 let hasExtraSrcRegAllocReq = 1 in {
4128 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4129 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4130 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4132 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4133 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4134 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4136 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4137 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4139 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4140 } // hasExtraSrcRegAllocReq = 1
4143 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4145 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4147 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4149 // VTBX : Vector Table Extension
4151 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4152 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4153 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4154 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4155 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4156 let hasExtraSrcRegAllocReq = 1 in {
4158 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4159 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4160 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4162 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4163 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4164 NVTBLFrm, IIC_VTBX3,
4165 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4168 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4169 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4170 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4172 } // hasExtraSrcRegAllocReq = 1
4175 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4176 IIC_VTBX2, "$orig = $dst", []>;
4178 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4179 IIC_VTBX3, "$orig = $dst", []>;
4181 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4182 IIC_VTBX4, "$orig = $dst", []>;
4184 //===----------------------------------------------------------------------===//
4185 // NEON instructions for single-precision FP math
4186 //===----------------------------------------------------------------------===//
4188 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4189 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4190 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4194 class N3VSPat<SDNode OpNode, NeonI Inst>
4195 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4196 (EXTRACT_SUBREG (v2f32
4197 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4199 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4203 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4204 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4205 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4207 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4209 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4213 // These need separate instructions because they must use DPR_VFP2 register
4214 // class which have SPR sub-registers.
4216 // Vector Add Operations used for single-precision FP
4217 let neverHasSideEffects = 1 in
4218 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4219 def : N3VSPat<fadd, VADDfd_sfp>;
4221 // Vector Sub Operations used for single-precision FP
4222 let neverHasSideEffects = 1 in
4223 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4224 def : N3VSPat<fsub, VSUBfd_sfp>;
4226 // Vector Multiply Operations used for single-precision FP
4227 let neverHasSideEffects = 1 in
4228 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4229 def : N3VSPat<fmul, VMULfd_sfp>;
4231 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4232 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4233 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4235 //let neverHasSideEffects = 1 in
4236 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4237 // v2f32, fmul, fadd>;
4238 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4240 //let neverHasSideEffects = 1 in
4241 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4242 // v2f32, fmul, fsub>;
4243 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4245 // Vector Absolute used for single-precision FP
4246 let neverHasSideEffects = 1 in
4247 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4248 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4249 "vabs", "f32", "$dst, $src", "", []>;
4250 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4252 // Vector Negate used for single-precision FP
4253 let neverHasSideEffects = 1 in
4254 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4255 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4256 "vneg", "f32", "$dst, $src", "", []>;
4257 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4259 // Vector Maximum used for single-precision FP
4260 let neverHasSideEffects = 1 in
4261 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4262 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4263 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4264 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4266 // Vector Minimum used for single-precision FP
4267 let neverHasSideEffects = 1 in
4268 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4269 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4270 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4271 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4273 // Vector Convert between single-precision FP and integer
4274 let neverHasSideEffects = 1 in
4275 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4276 v2i32, v2f32, fp_to_sint>;
4277 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4279 let neverHasSideEffects = 1 in
4280 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4281 v2i32, v2f32, fp_to_uint>;
4282 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4284 let neverHasSideEffects = 1 in
4285 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4286 v2f32, v2i32, sint_to_fp>;
4287 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4289 let neverHasSideEffects = 1 in
4290 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4291 v2f32, v2i32, uint_to_fp>;
4292 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4294 //===----------------------------------------------------------------------===//
4295 // Non-Instruction Patterns
4296 //===----------------------------------------------------------------------===//
4299 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4300 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4301 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4302 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4303 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4304 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4305 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4306 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4307 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4308 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4309 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4310 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4311 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4312 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4313 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4314 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4315 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4316 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4317 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4318 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4319 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4320 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4321 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4322 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4323 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4324 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4325 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4326 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4327 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4328 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4330 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4331 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4332 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4333 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4334 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4335 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4336 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4337 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4338 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4339 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4340 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4341 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4342 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4343 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4344 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4345 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4346 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4347 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4348 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4349 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4350 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4351 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4352 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4353 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4354 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4355 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4356 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4357 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4358 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4359 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;