1 //===- PPCInstrVSX.td - The PowerPC VSX Extension --*- tablegen -*-===//
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 VSX extension to the PowerPC instruction set.
12 //===----------------------------------------------------------------------===//
14 // *********************************** NOTE ***********************************
15 // ** For POWER8 Little Endian, the VSX swap optimization relies on knowing **
16 // ** which VMX and VSX instructions are lane-sensitive and which are not. **
17 // ** A lane-sensitive instruction relies, implicitly or explicitly, on **
18 // ** whether lanes are numbered from left to right. An instruction like **
19 // ** VADDFP is not lane-sensitive, because each lane of the result vector **
20 // ** relies only on the corresponding lane of the source vectors. However, **
21 // ** an instruction like VMULESB is lane-sensitive, because "even" and **
22 // ** "odd" lanes are different for big-endian and little-endian numbering. **
24 // ** When adding new VMX and VSX instructions, please consider whether they **
25 // ** are lane-sensitive. If so, they must be added to a switch statement **
26 // ** in PPCVSXSwapRemoval::gatherVectorInstructions(). **
27 // ****************************************************************************
29 def PPCRegVSRCAsmOperand : AsmOperandClass {
30 let Name = "RegVSRC"; let PredicateMethod = "isVSRegNumber";
32 def vsrc : RegisterOperand<VSRC> {
33 let ParserMatchClass = PPCRegVSRCAsmOperand;
36 def PPCRegVSFRCAsmOperand : AsmOperandClass {
37 let Name = "RegVSFRC"; let PredicateMethod = "isVSRegNumber";
39 def vsfrc : RegisterOperand<VSFRC> {
40 let ParserMatchClass = PPCRegVSFRCAsmOperand;
43 def PPCRegVSSRCAsmOperand : AsmOperandClass {
44 let Name = "RegVSSRC"; let PredicateMethod = "isVSRegNumber";
46 def vssrc : RegisterOperand<VSSRC> {
47 let ParserMatchClass = PPCRegVSSRCAsmOperand;
50 def PPCRegSPILLTOVSRRCAsmOperand : AsmOperandClass {
51 let Name = "RegSPILLTOVSRRC"; let PredicateMethod = "isVSRegNumber";
54 def spilltovsrrc : RegisterOperand<SPILLTOVSRRC> {
55 let ParserMatchClass = PPCRegSPILLTOVSRRCAsmOperand;
57 // Little-endian-specific nodes.
58 def SDT_PPClxvd2x : SDTypeProfile<1, 1, [
59 SDTCisVT<0, v2f64>, SDTCisPtrTy<1>
61 def SDT_PPCstxvd2x : SDTypeProfile<0, 2, [
62 SDTCisVT<0, v2f64>, SDTCisPtrTy<1>
64 def SDT_PPCxxswapd : SDTypeProfile<1, 1, [
67 def SDTVecConv : SDTypeProfile<1, 2, [
68 SDTCisVec<0>, SDTCisVec<1>, SDTCisPtrTy<2>
70 def SDTVabsd : SDTypeProfile<1, 3, [
71 SDTCisVec<0>, SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisVT<3, i32>
75 def PPClxvd2x : SDNode<"PPCISD::LXVD2X", SDT_PPClxvd2x,
76 [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
77 def PPCstxvd2x : SDNode<"PPCISD::STXVD2X", SDT_PPCstxvd2x,
78 [SDNPHasChain, SDNPMayStore]>;
79 def PPCxxswapd : SDNode<"PPCISD::XXSWAPD", SDT_PPCxxswapd, [SDNPHasChain]>;
80 def PPCmfvsr : SDNode<"PPCISD::MFVSR", SDTUnaryOp, []>;
81 def PPCmtvsra : SDNode<"PPCISD::MTVSRA", SDTUnaryOp, []>;
82 def PPCmtvsrz : SDNode<"PPCISD::MTVSRZ", SDTUnaryOp, []>;
83 def PPCsvec2fp : SDNode<"PPCISD::SINT_VEC_TO_FP", SDTVecConv, []>;
84 def PPCuvec2fp: SDNode<"PPCISD::UINT_VEC_TO_FP", SDTVecConv, []>;
85 def PPCswapNoChain : SDNode<"PPCISD::SWAP_NO_CHAIN", SDT_PPCxxswapd>;
86 def PPCvabsd : SDNode<"PPCISD::VABSD", SDTVabsd, []>;
88 multiclass XX3Form_Rcr<bits<6> opcode, bits<7> xo, string asmbase,
89 string asmstr, InstrItinClass itin, Intrinsic Int,
90 ValueType OutTy, ValueType InTy> {
91 let BaseName = asmbase in {
92 def NAME : XX3Form_Rc<opcode, xo, (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
93 !strconcat(asmbase, !strconcat(" ", asmstr)), itin,
94 [(set OutTy:$XT, (Int InTy:$XA, InTy:$XB))]>;
96 def o : XX3Form_Rc<opcode, xo, (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
97 !strconcat(asmbase, !strconcat(". ", asmstr)), itin,
99 (InTy (PPCvcmp_o InTy:$XA, InTy:$XB, xo)))]>,
104 // Instruction form with a single input register for instructions such as
105 // XXPERMDI. The reason for defining this is that specifying multiple chained
106 // operands (such as loads) to an instruction will perform both chained
107 // operations rather than coalescing them into a single register - even though
108 // the source memory location is the same. This simply forces the instruction
109 // to use the same register for both inputs.
110 // For example, an output DAG such as this:
111 // (XXPERMDI (LXSIBZX xoaddr:$src), (LXSIBZX xoaddr:$src ), 0))
112 // would result in two load instructions emitted and used as separate inputs
113 // to the XXPERMDI instruction.
114 class XX3Form_2s<bits<6> opcode, bits<5> xo, dag OOL, dag IOL, string asmstr,
115 InstrItinClass itin, list<dag> pattern>
116 : XX3Form_2<opcode, xo, OOL, IOL, asmstr, itin, pattern> {
120 def HasVSX : Predicate<"PPCSubTarget->hasVSX()">;
121 def IsLittleEndian : Predicate<"PPCSubTarget->isLittleEndian()">;
122 def IsBigEndian : Predicate<"!PPCSubTarget->isLittleEndian()">;
123 def HasOnlySwappingMemOps : Predicate<"!PPCSubTarget->hasP9Vector()">;
125 let Predicates = [HasVSX] in {
126 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
127 let UseVSXReg = 1 in {
128 let hasSideEffects = 0 in { // VSX instructions don't have side effects.
131 // Load indexed instructions
132 let mayLoad = 1, mayStore = 0 in {
134 def LXSDX : XX1Form_memOp<31, 588,
135 (outs vsfrc:$XT), (ins memrr:$src),
136 "lxsdx $XT, $src", IIC_LdStLFD,
139 // Pseudo instruction XFLOADf64 will be expanded to LXSDX or LFDX later
141 def XFLOADf64 : PseudoXFormMemOp<(outs vsfrc:$XT), (ins memrr:$src),
143 [(set f64:$XT, (load xoaddr:$src))]>;
145 let Predicates = [HasVSX, HasOnlySwappingMemOps] in
146 def LXVD2X : XX1Form_memOp<31, 844,
147 (outs vsrc:$XT), (ins memrr:$src),
148 "lxvd2x $XT, $src", IIC_LdStLFD,
149 [(set v2f64:$XT, (int_ppc_vsx_lxvd2x xoaddr:$src))]>;
151 def LXVDSX : XX1Form_memOp<31, 332,
152 (outs vsrc:$XT), (ins memrr:$src),
153 "lxvdsx $XT, $src", IIC_LdStLFD, []>;
155 let Predicates = [HasVSX, HasOnlySwappingMemOps] in
156 def LXVW4X : XX1Form_memOp<31, 780,
157 (outs vsrc:$XT), (ins memrr:$src),
158 "lxvw4x $XT, $src", IIC_LdStLFD,
162 // Store indexed instructions
163 let mayStore = 1, mayLoad = 0 in {
165 def STXSDX : XX1Form_memOp<31, 716,
166 (outs), (ins vsfrc:$XT, memrr:$dst),
167 "stxsdx $XT, $dst", IIC_LdStSTFD,
170 // Pseudo instruction XFSTOREf64 will be expanded to STXSDX or STFDX later
172 def XFSTOREf64 : PseudoXFormMemOp<(outs), (ins vsfrc:$XT, memrr:$dst),
174 [(store f64:$XT, xoaddr:$dst)]>;
176 let Predicates = [HasVSX, HasOnlySwappingMemOps] in {
177 // The behaviour of this instruction is endianness-specific so we provide no
178 // pattern to match it without considering endianness.
179 def STXVD2X : XX1Form_memOp<31, 972,
180 (outs), (ins vsrc:$XT, memrr:$dst),
181 "stxvd2x $XT, $dst", IIC_LdStSTFD,
184 def STXVW4X : XX1Form_memOp<31, 908,
185 (outs), (ins vsrc:$XT, memrr:$dst),
186 "stxvw4x $XT, $dst", IIC_LdStSTFD,
191 // Add/Mul Instructions
192 let isCommutable = 1 in {
193 def XSADDDP : XX3Form<60, 32,
194 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
195 "xsadddp $XT, $XA, $XB", IIC_VecFP,
196 [(set f64:$XT, (fadd f64:$XA, f64:$XB))]>;
197 def XSMULDP : XX3Form<60, 48,
198 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
199 "xsmuldp $XT, $XA, $XB", IIC_VecFP,
200 [(set f64:$XT, (fmul f64:$XA, f64:$XB))]>;
202 def XVADDDP : XX3Form<60, 96,
203 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
204 "xvadddp $XT, $XA, $XB", IIC_VecFP,
205 [(set v2f64:$XT, (fadd v2f64:$XA, v2f64:$XB))]>;
207 def XVADDSP : XX3Form<60, 64,
208 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
209 "xvaddsp $XT, $XA, $XB", IIC_VecFP,
210 [(set v4f32:$XT, (fadd v4f32:$XA, v4f32:$XB))]>;
212 def XVMULDP : XX3Form<60, 112,
213 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
214 "xvmuldp $XT, $XA, $XB", IIC_VecFP,
215 [(set v2f64:$XT, (fmul v2f64:$XA, v2f64:$XB))]>;
217 def XVMULSP : XX3Form<60, 80,
218 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
219 "xvmulsp $XT, $XA, $XB", IIC_VecFP,
220 [(set v4f32:$XT, (fmul v4f32:$XA, v4f32:$XB))]>;
223 // Subtract Instructions
224 def XSSUBDP : XX3Form<60, 40,
225 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
226 "xssubdp $XT, $XA, $XB", IIC_VecFP,
227 [(set f64:$XT, (fsub f64:$XA, f64:$XB))]>;
229 def XVSUBDP : XX3Form<60, 104,
230 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
231 "xvsubdp $XT, $XA, $XB", IIC_VecFP,
232 [(set v2f64:$XT, (fsub v2f64:$XA, v2f64:$XB))]>;
233 def XVSUBSP : XX3Form<60, 72,
234 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
235 "xvsubsp $XT, $XA, $XB", IIC_VecFP,
236 [(set v4f32:$XT, (fsub v4f32:$XA, v4f32:$XB))]>;
239 let BaseName = "XSMADDADP" in {
240 let isCommutable = 1 in
241 def XSMADDADP : XX3Form<60, 33,
242 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
243 "xsmaddadp $XT, $XA, $XB", IIC_VecFP,
244 [(set f64:$XT, (fma f64:$XA, f64:$XB, f64:$XTi))]>,
245 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
247 let IsVSXFMAAlt = 1 in
248 def XSMADDMDP : XX3Form<60, 41,
249 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
250 "xsmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
251 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
255 let BaseName = "XSMSUBADP" in {
256 let isCommutable = 1 in
257 def XSMSUBADP : XX3Form<60, 49,
258 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
259 "xsmsubadp $XT, $XA, $XB", IIC_VecFP,
260 [(set f64:$XT, (fma f64:$XA, f64:$XB, (fneg f64:$XTi)))]>,
261 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
263 let IsVSXFMAAlt = 1 in
264 def XSMSUBMDP : XX3Form<60, 57,
265 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
266 "xsmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
267 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
271 let BaseName = "XSNMADDADP" in {
272 let isCommutable = 1 in
273 def XSNMADDADP : XX3Form<60, 161,
274 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
275 "xsnmaddadp $XT, $XA, $XB", IIC_VecFP,
276 [(set f64:$XT, (fneg (fma f64:$XA, f64:$XB, f64:$XTi)))]>,
277 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
279 let IsVSXFMAAlt = 1 in
280 def XSNMADDMDP : XX3Form<60, 169,
281 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
282 "xsnmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
283 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
287 let BaseName = "XSNMSUBADP" in {
288 let isCommutable = 1 in
289 def XSNMSUBADP : XX3Form<60, 177,
290 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
291 "xsnmsubadp $XT, $XA, $XB", IIC_VecFP,
292 [(set f64:$XT, (fneg (fma f64:$XA, f64:$XB, (fneg f64:$XTi))))]>,
293 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
295 let IsVSXFMAAlt = 1 in
296 def XSNMSUBMDP : XX3Form<60, 185,
297 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
298 "xsnmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
299 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
303 let BaseName = "XVMADDADP" in {
304 let isCommutable = 1 in
305 def XVMADDADP : XX3Form<60, 97,
306 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
307 "xvmaddadp $XT, $XA, $XB", IIC_VecFP,
308 [(set v2f64:$XT, (fma v2f64:$XA, v2f64:$XB, v2f64:$XTi))]>,
309 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
311 let IsVSXFMAAlt = 1 in
312 def XVMADDMDP : XX3Form<60, 105,
313 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
314 "xvmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
315 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
319 let BaseName = "XVMADDASP" in {
320 let isCommutable = 1 in
321 def XVMADDASP : XX3Form<60, 65,
322 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
323 "xvmaddasp $XT, $XA, $XB", IIC_VecFP,
324 [(set v4f32:$XT, (fma v4f32:$XA, v4f32:$XB, v4f32:$XTi))]>,
325 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
327 let IsVSXFMAAlt = 1 in
328 def XVMADDMSP : XX3Form<60, 73,
329 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
330 "xvmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
331 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
335 let BaseName = "XVMSUBADP" in {
336 let isCommutable = 1 in
337 def XVMSUBADP : XX3Form<60, 113,
338 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
339 "xvmsubadp $XT, $XA, $XB", IIC_VecFP,
340 [(set v2f64:$XT, (fma v2f64:$XA, v2f64:$XB, (fneg v2f64:$XTi)))]>,
341 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
343 let IsVSXFMAAlt = 1 in
344 def XVMSUBMDP : XX3Form<60, 121,
345 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
346 "xvmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
347 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
351 let BaseName = "XVMSUBASP" in {
352 let isCommutable = 1 in
353 def XVMSUBASP : XX3Form<60, 81,
354 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
355 "xvmsubasp $XT, $XA, $XB", IIC_VecFP,
356 [(set v4f32:$XT, (fma v4f32:$XA, v4f32:$XB, (fneg v4f32:$XTi)))]>,
357 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
359 let IsVSXFMAAlt = 1 in
360 def XVMSUBMSP : XX3Form<60, 89,
361 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
362 "xvmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
363 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
367 let BaseName = "XVNMADDADP" in {
368 let isCommutable = 1 in
369 def XVNMADDADP : XX3Form<60, 225,
370 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
371 "xvnmaddadp $XT, $XA, $XB", IIC_VecFP,
372 [(set v2f64:$XT, (fneg (fma v2f64:$XA, v2f64:$XB, v2f64:$XTi)))]>,
373 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
375 let IsVSXFMAAlt = 1 in
376 def XVNMADDMDP : XX3Form<60, 233,
377 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
378 "xvnmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
379 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
383 let BaseName = "XVNMADDASP" in {
384 let isCommutable = 1 in
385 def XVNMADDASP : XX3Form<60, 193,
386 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
387 "xvnmaddasp $XT, $XA, $XB", IIC_VecFP,
388 [(set v4f32:$XT, (fneg (fma v4f32:$XA, v4f32:$XB, v4f32:$XTi)))]>,
389 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
391 let IsVSXFMAAlt = 1 in
392 def XVNMADDMSP : XX3Form<60, 201,
393 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
394 "xvnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
395 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
399 let BaseName = "XVNMSUBADP" in {
400 let isCommutable = 1 in
401 def XVNMSUBADP : XX3Form<60, 241,
402 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
403 "xvnmsubadp $XT, $XA, $XB", IIC_VecFP,
404 [(set v2f64:$XT, (fneg (fma v2f64:$XA, v2f64:$XB, (fneg v2f64:$XTi))))]>,
405 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
407 let IsVSXFMAAlt = 1 in
408 def XVNMSUBMDP : XX3Form<60, 249,
409 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
410 "xvnmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
411 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
415 let BaseName = "XVNMSUBASP" in {
416 let isCommutable = 1 in
417 def XVNMSUBASP : XX3Form<60, 209,
418 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
419 "xvnmsubasp $XT, $XA, $XB", IIC_VecFP,
420 [(set v4f32:$XT, (fneg (fma v4f32:$XA, v4f32:$XB, (fneg v4f32:$XTi))))]>,
421 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
423 let IsVSXFMAAlt = 1 in
424 def XVNMSUBMSP : XX3Form<60, 217,
425 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
426 "xvnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
427 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
431 // Division Instructions
432 def XSDIVDP : XX3Form<60, 56,
433 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
434 "xsdivdp $XT, $XA, $XB", IIC_FPDivD,
435 [(set f64:$XT, (fdiv f64:$XA, f64:$XB))]>;
436 def XSSQRTDP : XX2Form<60, 75,
437 (outs vsfrc:$XT), (ins vsfrc:$XB),
438 "xssqrtdp $XT, $XB", IIC_FPSqrtD,
439 [(set f64:$XT, (fsqrt f64:$XB))]>;
441 def XSREDP : XX2Form<60, 90,
442 (outs vsfrc:$XT), (ins vsfrc:$XB),
443 "xsredp $XT, $XB", IIC_VecFP,
444 [(set f64:$XT, (PPCfre f64:$XB))]>;
445 def XSRSQRTEDP : XX2Form<60, 74,
446 (outs vsfrc:$XT), (ins vsfrc:$XB),
447 "xsrsqrtedp $XT, $XB", IIC_VecFP,
448 [(set f64:$XT, (PPCfrsqrte f64:$XB))]>;
450 def XSTDIVDP : XX3Form_1<60, 61,
451 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
452 "xstdivdp $crD, $XA, $XB", IIC_FPCompare, []>;
453 def XSTSQRTDP : XX2Form_1<60, 106,
454 (outs crrc:$crD), (ins vsfrc:$XB),
455 "xstsqrtdp $crD, $XB", IIC_FPCompare, []>;
457 def XVDIVDP : XX3Form<60, 120,
458 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
459 "xvdivdp $XT, $XA, $XB", IIC_FPDivD,
460 [(set v2f64:$XT, (fdiv v2f64:$XA, v2f64:$XB))]>;
461 def XVDIVSP : XX3Form<60, 88,
462 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
463 "xvdivsp $XT, $XA, $XB", IIC_FPDivS,
464 [(set v4f32:$XT, (fdiv v4f32:$XA, v4f32:$XB))]>;
466 def XVSQRTDP : XX2Form<60, 203,
467 (outs vsrc:$XT), (ins vsrc:$XB),
468 "xvsqrtdp $XT, $XB", IIC_FPSqrtD,
469 [(set v2f64:$XT, (fsqrt v2f64:$XB))]>;
470 def XVSQRTSP : XX2Form<60, 139,
471 (outs vsrc:$XT), (ins vsrc:$XB),
472 "xvsqrtsp $XT, $XB", IIC_FPSqrtS,
473 [(set v4f32:$XT, (fsqrt v4f32:$XB))]>;
475 def XVTDIVDP : XX3Form_1<60, 125,
476 (outs crrc:$crD), (ins vsrc:$XA, vsrc:$XB),
477 "xvtdivdp $crD, $XA, $XB", IIC_FPCompare, []>;
478 def XVTDIVSP : XX3Form_1<60, 93,
479 (outs crrc:$crD), (ins vsrc:$XA, vsrc:$XB),
480 "xvtdivsp $crD, $XA, $XB", IIC_FPCompare, []>;
482 def XVTSQRTDP : XX2Form_1<60, 234,
483 (outs crrc:$crD), (ins vsrc:$XB),
484 "xvtsqrtdp $crD, $XB", IIC_FPCompare, []>;
485 def XVTSQRTSP : XX2Form_1<60, 170,
486 (outs crrc:$crD), (ins vsrc:$XB),
487 "xvtsqrtsp $crD, $XB", IIC_FPCompare, []>;
489 def XVREDP : XX2Form<60, 218,
490 (outs vsrc:$XT), (ins vsrc:$XB),
491 "xvredp $XT, $XB", IIC_VecFP,
492 [(set v2f64:$XT, (PPCfre v2f64:$XB))]>;
493 def XVRESP : XX2Form<60, 154,
494 (outs vsrc:$XT), (ins vsrc:$XB),
495 "xvresp $XT, $XB", IIC_VecFP,
496 [(set v4f32:$XT, (PPCfre v4f32:$XB))]>;
498 def XVRSQRTEDP : XX2Form<60, 202,
499 (outs vsrc:$XT), (ins vsrc:$XB),
500 "xvrsqrtedp $XT, $XB", IIC_VecFP,
501 [(set v2f64:$XT, (PPCfrsqrte v2f64:$XB))]>;
502 def XVRSQRTESP : XX2Form<60, 138,
503 (outs vsrc:$XT), (ins vsrc:$XB),
504 "xvrsqrtesp $XT, $XB", IIC_VecFP,
505 [(set v4f32:$XT, (PPCfrsqrte v4f32:$XB))]>;
507 // Compare Instructions
508 def XSCMPODP : XX3Form_1<60, 43,
509 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
510 "xscmpodp $crD, $XA, $XB", IIC_FPCompare, []>;
511 def XSCMPUDP : XX3Form_1<60, 35,
512 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
513 "xscmpudp $crD, $XA, $XB", IIC_FPCompare, []>;
515 defm XVCMPEQDP : XX3Form_Rcr<60, 99,
516 "xvcmpeqdp", "$XT, $XA, $XB", IIC_VecFPCompare,
517 int_ppc_vsx_xvcmpeqdp, v2i64, v2f64>;
518 defm XVCMPEQSP : XX3Form_Rcr<60, 67,
519 "xvcmpeqsp", "$XT, $XA, $XB", IIC_VecFPCompare,
520 int_ppc_vsx_xvcmpeqsp, v4i32, v4f32>;
521 defm XVCMPGEDP : XX3Form_Rcr<60, 115,
522 "xvcmpgedp", "$XT, $XA, $XB", IIC_VecFPCompare,
523 int_ppc_vsx_xvcmpgedp, v2i64, v2f64>;
524 defm XVCMPGESP : XX3Form_Rcr<60, 83,
525 "xvcmpgesp", "$XT, $XA, $XB", IIC_VecFPCompare,
526 int_ppc_vsx_xvcmpgesp, v4i32, v4f32>;
527 defm XVCMPGTDP : XX3Form_Rcr<60, 107,
528 "xvcmpgtdp", "$XT, $XA, $XB", IIC_VecFPCompare,
529 int_ppc_vsx_xvcmpgtdp, v2i64, v2f64>;
530 defm XVCMPGTSP : XX3Form_Rcr<60, 75,
531 "xvcmpgtsp", "$XT, $XA, $XB", IIC_VecFPCompare,
532 int_ppc_vsx_xvcmpgtsp, v4i32, v4f32>;
535 def XSABSDP : XX2Form<60, 345,
536 (outs vsfrc:$XT), (ins vsfrc:$XB),
537 "xsabsdp $XT, $XB", IIC_VecFP,
538 [(set f64:$XT, (fabs f64:$XB))]>;
539 def XSNABSDP : XX2Form<60, 361,
540 (outs vsfrc:$XT), (ins vsfrc:$XB),
541 "xsnabsdp $XT, $XB", IIC_VecFP,
542 [(set f64:$XT, (fneg (fabs f64:$XB)))]>;
543 def XSNEGDP : XX2Form<60, 377,
544 (outs vsfrc:$XT), (ins vsfrc:$XB),
545 "xsnegdp $XT, $XB", IIC_VecFP,
546 [(set f64:$XT, (fneg f64:$XB))]>;
547 def XSCPSGNDP : XX3Form<60, 176,
548 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
549 "xscpsgndp $XT, $XA, $XB", IIC_VecFP,
550 [(set f64:$XT, (fcopysign f64:$XB, f64:$XA))]>;
552 def XVABSDP : XX2Form<60, 473,
553 (outs vsrc:$XT), (ins vsrc:$XB),
554 "xvabsdp $XT, $XB", IIC_VecFP,
555 [(set v2f64:$XT, (fabs v2f64:$XB))]>;
557 def XVABSSP : XX2Form<60, 409,
558 (outs vsrc:$XT), (ins vsrc:$XB),
559 "xvabssp $XT, $XB", IIC_VecFP,
560 [(set v4f32:$XT, (fabs v4f32:$XB))]>;
562 def XVCPSGNDP : XX3Form<60, 240,
563 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
564 "xvcpsgndp $XT, $XA, $XB", IIC_VecFP,
565 [(set v2f64:$XT, (fcopysign v2f64:$XB, v2f64:$XA))]>;
566 def XVCPSGNSP : XX3Form<60, 208,
567 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
568 "xvcpsgnsp $XT, $XA, $XB", IIC_VecFP,
569 [(set v4f32:$XT, (fcopysign v4f32:$XB, v4f32:$XA))]>;
571 def XVNABSDP : XX2Form<60, 489,
572 (outs vsrc:$XT), (ins vsrc:$XB),
573 "xvnabsdp $XT, $XB", IIC_VecFP,
574 [(set v2f64:$XT, (fneg (fabs v2f64:$XB)))]>;
575 def XVNABSSP : XX2Form<60, 425,
576 (outs vsrc:$XT), (ins vsrc:$XB),
577 "xvnabssp $XT, $XB", IIC_VecFP,
578 [(set v4f32:$XT, (fneg (fabs v4f32:$XB)))]>;
580 def XVNEGDP : XX2Form<60, 505,
581 (outs vsrc:$XT), (ins vsrc:$XB),
582 "xvnegdp $XT, $XB", IIC_VecFP,
583 [(set v2f64:$XT, (fneg v2f64:$XB))]>;
584 def XVNEGSP : XX2Form<60, 441,
585 (outs vsrc:$XT), (ins vsrc:$XB),
586 "xvnegsp $XT, $XB", IIC_VecFP,
587 [(set v4f32:$XT, (fneg v4f32:$XB))]>;
589 // Conversion Instructions
590 def XSCVDPSP : XX2Form<60, 265,
591 (outs vsfrc:$XT), (ins vsfrc:$XB),
592 "xscvdpsp $XT, $XB", IIC_VecFP, []>;
593 def XSCVDPSXDS : XX2Form<60, 344,
594 (outs vsfrc:$XT), (ins vsfrc:$XB),
595 "xscvdpsxds $XT, $XB", IIC_VecFP,
596 [(set f64:$XT, (PPCfctidz f64:$XB))]>;
597 let isCodeGenOnly = 1 in
598 def XSCVDPSXDSs : XX2Form<60, 344,
599 (outs vssrc:$XT), (ins vssrc:$XB),
600 "xscvdpsxds $XT, $XB", IIC_VecFP,
601 [(set f32:$XT, (PPCfctidz f32:$XB))]>;
602 def XSCVDPSXWS : XX2Form<60, 88,
603 (outs vsfrc:$XT), (ins vsfrc:$XB),
604 "xscvdpsxws $XT, $XB", IIC_VecFP,
605 [(set f64:$XT, (PPCfctiwz f64:$XB))]>;
606 let isCodeGenOnly = 1 in
607 def XSCVDPSXWSs : XX2Form<60, 88,
608 (outs vssrc:$XT), (ins vssrc:$XB),
609 "xscvdpsxws $XT, $XB", IIC_VecFP,
610 [(set f32:$XT, (PPCfctiwz f32:$XB))]>;
611 def XSCVDPUXDS : XX2Form<60, 328,
612 (outs vsfrc:$XT), (ins vsfrc:$XB),
613 "xscvdpuxds $XT, $XB", IIC_VecFP,
614 [(set f64:$XT, (PPCfctiduz f64:$XB))]>;
615 let isCodeGenOnly = 1 in
616 def XSCVDPUXDSs : XX2Form<60, 328,
617 (outs vssrc:$XT), (ins vssrc:$XB),
618 "xscvdpuxds $XT, $XB", IIC_VecFP,
619 [(set f32:$XT, (PPCfctiduz f32:$XB))]>;
620 def XSCVDPUXWS : XX2Form<60, 72,
621 (outs vsfrc:$XT), (ins vsfrc:$XB),
622 "xscvdpuxws $XT, $XB", IIC_VecFP,
623 [(set f64:$XT, (PPCfctiwuz f64:$XB))]>;
624 let isCodeGenOnly = 1 in
625 def XSCVDPUXWSs : XX2Form<60, 72,
626 (outs vssrc:$XT), (ins vssrc:$XB),
627 "xscvdpuxws $XT, $XB", IIC_VecFP,
628 [(set f32:$XT, (PPCfctiwuz f32:$XB))]>;
629 def XSCVSPDP : XX2Form<60, 329,
630 (outs vsfrc:$XT), (ins vsfrc:$XB),
631 "xscvspdp $XT, $XB", IIC_VecFP, []>;
632 def XSCVSXDDP : XX2Form<60, 376,
633 (outs vsfrc:$XT), (ins vsfrc:$XB),
634 "xscvsxddp $XT, $XB", IIC_VecFP,
635 [(set f64:$XT, (PPCfcfid f64:$XB))]>;
636 def XSCVUXDDP : XX2Form<60, 360,
637 (outs vsfrc:$XT), (ins vsfrc:$XB),
638 "xscvuxddp $XT, $XB", IIC_VecFP,
639 [(set f64:$XT, (PPCfcfidu f64:$XB))]>;
641 def XVCVDPSP : XX2Form<60, 393,
642 (outs vsrc:$XT), (ins vsrc:$XB),
643 "xvcvdpsp $XT, $XB", IIC_VecFP,
644 [(set v4f32:$XT, (int_ppc_vsx_xvcvdpsp v2f64:$XB))]>;
645 def XVCVDPSXDS : XX2Form<60, 472,
646 (outs vsrc:$XT), (ins vsrc:$XB),
647 "xvcvdpsxds $XT, $XB", IIC_VecFP,
648 [(set v2i64:$XT, (fp_to_sint v2f64:$XB))]>;
649 def XVCVDPSXWS : XX2Form<60, 216,
650 (outs vsrc:$XT), (ins vsrc:$XB),
651 "xvcvdpsxws $XT, $XB", IIC_VecFP,
652 [(set v4i32:$XT, (int_ppc_vsx_xvcvdpsxws v2f64:$XB))]>;
653 def XVCVDPUXDS : XX2Form<60, 456,
654 (outs vsrc:$XT), (ins vsrc:$XB),
655 "xvcvdpuxds $XT, $XB", IIC_VecFP,
656 [(set v2i64:$XT, (fp_to_uint v2f64:$XB))]>;
657 def XVCVDPUXWS : XX2Form<60, 200,
658 (outs vsrc:$XT), (ins vsrc:$XB),
659 "xvcvdpuxws $XT, $XB", IIC_VecFP,
660 [(set v4i32:$XT, (int_ppc_vsx_xvcvdpuxws v2f64:$XB))]>;
662 def XVCVSPDP : XX2Form<60, 457,
663 (outs vsrc:$XT), (ins vsrc:$XB),
664 "xvcvspdp $XT, $XB", IIC_VecFP,
665 [(set v2f64:$XT, (int_ppc_vsx_xvcvspdp v4f32:$XB))]>;
666 def XVCVSPSXDS : XX2Form<60, 408,
667 (outs vsrc:$XT), (ins vsrc:$XB),
668 "xvcvspsxds $XT, $XB", IIC_VecFP, []>;
669 def XVCVSPSXWS : XX2Form<60, 152,
670 (outs vsrc:$XT), (ins vsrc:$XB),
671 "xvcvspsxws $XT, $XB", IIC_VecFP,
672 [(set v4i32:$XT, (fp_to_sint v4f32:$XB))]>;
673 def XVCVSPUXDS : XX2Form<60, 392,
674 (outs vsrc:$XT), (ins vsrc:$XB),
675 "xvcvspuxds $XT, $XB", IIC_VecFP, []>;
676 def XVCVSPUXWS : XX2Form<60, 136,
677 (outs vsrc:$XT), (ins vsrc:$XB),
678 "xvcvspuxws $XT, $XB", IIC_VecFP,
679 [(set v4i32:$XT, (fp_to_uint v4f32:$XB))]>;
680 def XVCVSXDDP : XX2Form<60, 504,
681 (outs vsrc:$XT), (ins vsrc:$XB),
682 "xvcvsxddp $XT, $XB", IIC_VecFP,
683 [(set v2f64:$XT, (sint_to_fp v2i64:$XB))]>;
684 def XVCVSXDSP : XX2Form<60, 440,
685 (outs vsrc:$XT), (ins vsrc:$XB),
686 "xvcvsxdsp $XT, $XB", IIC_VecFP,
687 [(set v4f32:$XT, (int_ppc_vsx_xvcvsxdsp v2i64:$XB))]>;
688 def XVCVSXWDP : XX2Form<60, 248,
689 (outs vsrc:$XT), (ins vsrc:$XB),
690 "xvcvsxwdp $XT, $XB", IIC_VecFP,
691 [(set v2f64:$XT, (int_ppc_vsx_xvcvsxwdp v4i32:$XB))]>;
692 def XVCVSXWSP : XX2Form<60, 184,
693 (outs vsrc:$XT), (ins vsrc:$XB),
694 "xvcvsxwsp $XT, $XB", IIC_VecFP,
695 [(set v4f32:$XT, (sint_to_fp v4i32:$XB))]>;
696 def XVCVUXDDP : XX2Form<60, 488,
697 (outs vsrc:$XT), (ins vsrc:$XB),
698 "xvcvuxddp $XT, $XB", IIC_VecFP,
699 [(set v2f64:$XT, (uint_to_fp v2i64:$XB))]>;
700 def XVCVUXDSP : XX2Form<60, 424,
701 (outs vsrc:$XT), (ins vsrc:$XB),
702 "xvcvuxdsp $XT, $XB", IIC_VecFP,
703 [(set v4f32:$XT, (int_ppc_vsx_xvcvuxdsp v2i64:$XB))]>;
704 def XVCVUXWDP : XX2Form<60, 232,
705 (outs vsrc:$XT), (ins vsrc:$XB),
706 "xvcvuxwdp $XT, $XB", IIC_VecFP,
707 [(set v2f64:$XT, (int_ppc_vsx_xvcvuxwdp v4i32:$XB))]>;
708 def XVCVUXWSP : XX2Form<60, 168,
709 (outs vsrc:$XT), (ins vsrc:$XB),
710 "xvcvuxwsp $XT, $XB", IIC_VecFP,
711 [(set v4f32:$XT, (uint_to_fp v4i32:$XB))]>;
713 // Rounding Instructions
714 def XSRDPI : XX2Form<60, 73,
715 (outs vsfrc:$XT), (ins vsfrc:$XB),
716 "xsrdpi $XT, $XB", IIC_VecFP,
717 [(set f64:$XT, (fround f64:$XB))]>;
718 def XSRDPIC : XX2Form<60, 107,
719 (outs vsfrc:$XT), (ins vsfrc:$XB),
720 "xsrdpic $XT, $XB", IIC_VecFP,
721 [(set f64:$XT, (fnearbyint f64:$XB))]>;
722 def XSRDPIM : XX2Form<60, 121,
723 (outs vsfrc:$XT), (ins vsfrc:$XB),
724 "xsrdpim $XT, $XB", IIC_VecFP,
725 [(set f64:$XT, (ffloor f64:$XB))]>;
726 def XSRDPIP : XX2Form<60, 105,
727 (outs vsfrc:$XT), (ins vsfrc:$XB),
728 "xsrdpip $XT, $XB", IIC_VecFP,
729 [(set f64:$XT, (fceil f64:$XB))]>;
730 def XSRDPIZ : XX2Form<60, 89,
731 (outs vsfrc:$XT), (ins vsfrc:$XB),
732 "xsrdpiz $XT, $XB", IIC_VecFP,
733 [(set f64:$XT, (ftrunc f64:$XB))]>;
735 def XVRDPI : XX2Form<60, 201,
736 (outs vsrc:$XT), (ins vsrc:$XB),
737 "xvrdpi $XT, $XB", IIC_VecFP,
738 [(set v2f64:$XT, (fround v2f64:$XB))]>;
739 def XVRDPIC : XX2Form<60, 235,
740 (outs vsrc:$XT), (ins vsrc:$XB),
741 "xvrdpic $XT, $XB", IIC_VecFP,
742 [(set v2f64:$XT, (fnearbyint v2f64:$XB))]>;
743 def XVRDPIM : XX2Form<60, 249,
744 (outs vsrc:$XT), (ins vsrc:$XB),
745 "xvrdpim $XT, $XB", IIC_VecFP,
746 [(set v2f64:$XT, (ffloor v2f64:$XB))]>;
747 def XVRDPIP : XX2Form<60, 233,
748 (outs vsrc:$XT), (ins vsrc:$XB),
749 "xvrdpip $XT, $XB", IIC_VecFP,
750 [(set v2f64:$XT, (fceil v2f64:$XB))]>;
751 def XVRDPIZ : XX2Form<60, 217,
752 (outs vsrc:$XT), (ins vsrc:$XB),
753 "xvrdpiz $XT, $XB", IIC_VecFP,
754 [(set v2f64:$XT, (ftrunc v2f64:$XB))]>;
756 def XVRSPI : XX2Form<60, 137,
757 (outs vsrc:$XT), (ins vsrc:$XB),
758 "xvrspi $XT, $XB", IIC_VecFP,
759 [(set v4f32:$XT, (fround v4f32:$XB))]>;
760 def XVRSPIC : XX2Form<60, 171,
761 (outs vsrc:$XT), (ins vsrc:$XB),
762 "xvrspic $XT, $XB", IIC_VecFP,
763 [(set v4f32:$XT, (fnearbyint v4f32:$XB))]>;
764 def XVRSPIM : XX2Form<60, 185,
765 (outs vsrc:$XT), (ins vsrc:$XB),
766 "xvrspim $XT, $XB", IIC_VecFP,
767 [(set v4f32:$XT, (ffloor v4f32:$XB))]>;
768 def XVRSPIP : XX2Form<60, 169,
769 (outs vsrc:$XT), (ins vsrc:$XB),
770 "xvrspip $XT, $XB", IIC_VecFP,
771 [(set v4f32:$XT, (fceil v4f32:$XB))]>;
772 def XVRSPIZ : XX2Form<60, 153,
773 (outs vsrc:$XT), (ins vsrc:$XB),
774 "xvrspiz $XT, $XB", IIC_VecFP,
775 [(set v4f32:$XT, (ftrunc v4f32:$XB))]>;
777 // Max/Min Instructions
778 let isCommutable = 1 in {
779 def XSMAXDP : XX3Form<60, 160,
780 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
781 "xsmaxdp $XT, $XA, $XB", IIC_VecFP,
783 (int_ppc_vsx_xsmaxdp vsfrc:$XA, vsfrc:$XB))]>;
784 def XSMINDP : XX3Form<60, 168,
785 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
786 "xsmindp $XT, $XA, $XB", IIC_VecFP,
788 (int_ppc_vsx_xsmindp vsfrc:$XA, vsfrc:$XB))]>;
790 def XVMAXDP : XX3Form<60, 224,
791 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
792 "xvmaxdp $XT, $XA, $XB", IIC_VecFP,
794 (int_ppc_vsx_xvmaxdp vsrc:$XA, vsrc:$XB))]>;
795 def XVMINDP : XX3Form<60, 232,
796 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
797 "xvmindp $XT, $XA, $XB", IIC_VecFP,
799 (int_ppc_vsx_xvmindp vsrc:$XA, vsrc:$XB))]>;
801 def XVMAXSP : XX3Form<60, 192,
802 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
803 "xvmaxsp $XT, $XA, $XB", IIC_VecFP,
805 (int_ppc_vsx_xvmaxsp vsrc:$XA, vsrc:$XB))]>;
806 def XVMINSP : XX3Form<60, 200,
807 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
808 "xvminsp $XT, $XA, $XB", IIC_VecFP,
810 (int_ppc_vsx_xvminsp vsrc:$XA, vsrc:$XB))]>;
814 // Logical Instructions
815 let isCommutable = 1 in
816 def XXLAND : XX3Form<60, 130,
817 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
818 "xxland $XT, $XA, $XB", IIC_VecGeneral,
819 [(set v4i32:$XT, (and v4i32:$XA, v4i32:$XB))]>;
820 def XXLANDC : XX3Form<60, 138,
821 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
822 "xxlandc $XT, $XA, $XB", IIC_VecGeneral,
823 [(set v4i32:$XT, (and v4i32:$XA,
824 (vnot_ppc v4i32:$XB)))]>;
825 let isCommutable = 1 in {
826 def XXLNOR : XX3Form<60, 162,
827 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
828 "xxlnor $XT, $XA, $XB", IIC_VecGeneral,
829 [(set v4i32:$XT, (vnot_ppc (or v4i32:$XA,
831 def XXLOR : XX3Form<60, 146,
832 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
833 "xxlor $XT, $XA, $XB", IIC_VecGeneral,
834 [(set v4i32:$XT, (or v4i32:$XA, v4i32:$XB))]>;
835 let isCodeGenOnly = 1 in
836 def XXLORf: XX3Form<60, 146,
837 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
838 "xxlor $XT, $XA, $XB", IIC_VecGeneral, []>;
839 def XXLXOR : XX3Form<60, 154,
840 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
841 "xxlxor $XT, $XA, $XB", IIC_VecGeneral,
842 [(set v4i32:$XT, (xor v4i32:$XA, v4i32:$XB))]>;
844 let isCodeGenOnly = 1 in
845 def XXLXORz : XX3Form_Zero<60, 154, (outs vsrc:$XT), (ins),
846 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
847 [(set v4i32:$XT, (v4i32 immAllZerosV))]>;
849 let isCodeGenOnly = 1 in {
850 def XXLXORdpz : XX3Form_SetZero<60, 154,
851 (outs vsfrc:$XT), (ins),
852 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
853 [(set f64:$XT, (fpimm0))]>;
854 def XXLXORspz : XX3Form_SetZero<60, 154,
855 (outs vssrc:$XT), (ins),
856 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
857 [(set f32:$XT, (fpimm0))]>;
860 // Permutation Instructions
861 def XXMRGHW : XX3Form<60, 18,
862 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
863 "xxmrghw $XT, $XA, $XB", IIC_VecPerm, []>;
864 def XXMRGLW : XX3Form<60, 50,
865 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
866 "xxmrglw $XT, $XA, $XB", IIC_VecPerm, []>;
868 def XXPERMDI : XX3Form_2<60, 10,
869 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, u2imm:$DM),
870 "xxpermdi $XT, $XA, $XB, $DM", IIC_VecPerm,
871 [(set v2i64:$XT, (PPCxxpermdi v2i64:$XA, v2i64:$XB,
873 let isCodeGenOnly = 1 in
874 def XXPERMDIs : XX3Form_2s<60, 10, (outs vsrc:$XT), (ins vsfrc:$XA, u2imm:$DM),
875 "xxpermdi $XT, $XA, $XA, $DM", IIC_VecPerm, []>;
876 def XXSEL : XX4Form<60, 3,
877 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, vsrc:$XC),
878 "xxsel $XT, $XA, $XB, $XC", IIC_VecPerm, []>;
880 def XXSLDWI : XX3Form_2<60, 2,
881 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, u2imm:$SHW),
882 "xxsldwi $XT, $XA, $XB, $SHW", IIC_VecPerm,
883 [(set v4i32:$XT, (PPCvecshl v4i32:$XA, v4i32:$XB,
884 imm32SExt16:$SHW))]>;
886 let isCodeGenOnly = 1 in
887 def XXSLDWIs : XX3Form_2s<60, 2,
888 (outs vsrc:$XT), (ins vsfrc:$XA, u2imm:$SHW),
889 "xxsldwi $XT, $XA, $XA, $SHW", IIC_VecPerm, []>;
891 def XXSPLTW : XX2Form_2<60, 164,
892 (outs vsrc:$XT), (ins vsrc:$XB, u2imm:$UIM),
893 "xxspltw $XT, $XB, $UIM", IIC_VecPerm,
895 (PPCxxsplt v4i32:$XB, imm32SExt16:$UIM))]>;
896 let isCodeGenOnly = 1 in
897 def XXSPLTWs : XX2Form_2<60, 164,
898 (outs vsrc:$XT), (ins vfrc:$XB, u2imm:$UIM),
899 "xxspltw $XT, $XB, $UIM", IIC_VecPerm, []>;
904 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation. Expanded after
905 // instruction selection into a branch sequence.
906 let PPC970_Single = 1 in {
908 def SELECT_CC_VSRC: PPCCustomInserterPseudo<(outs vsrc:$dst),
909 (ins crrc:$cond, vsrc:$T, vsrc:$F, i32imm:$BROPC),
912 def SELECT_VSRC: PPCCustomInserterPseudo<(outs vsrc:$dst),
913 (ins crbitrc:$cond, vsrc:$T, vsrc:$F),
916 (select i1:$cond, v2f64:$T, v2f64:$F))]>;
917 def SELECT_CC_VSFRC: PPCCustomInserterPseudo<(outs f8rc:$dst),
918 (ins crrc:$cond, f8rc:$T, f8rc:$F,
919 i32imm:$BROPC), "#SELECT_CC_VSFRC",
921 def SELECT_VSFRC: PPCCustomInserterPseudo<(outs f8rc:$dst),
922 (ins crbitrc:$cond, f8rc:$T, f8rc:$F),
925 (select i1:$cond, f64:$T, f64:$F))]>;
926 def SELECT_CC_VSSRC: PPCCustomInserterPseudo<(outs f4rc:$dst),
927 (ins crrc:$cond, f4rc:$T, f4rc:$F,
928 i32imm:$BROPC), "#SELECT_CC_VSSRC",
930 def SELECT_VSSRC: PPCCustomInserterPseudo<(outs f4rc:$dst),
931 (ins crbitrc:$cond, f4rc:$T, f4rc:$F),
934 (select i1:$cond, f32:$T, f32:$F))]>;
938 def : InstAlias<"xvmovdp $XT, $XB",
939 (XVCPSGNDP vsrc:$XT, vsrc:$XB, vsrc:$XB)>;
940 def : InstAlias<"xvmovsp $XT, $XB",
941 (XVCPSGNSP vsrc:$XT, vsrc:$XB, vsrc:$XB)>;
943 def : InstAlias<"xxspltd $XT, $XB, 0",
944 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 0)>;
945 def : InstAlias<"xxspltd $XT, $XB, 1",
946 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 3)>;
947 def : InstAlias<"xxmrghd $XT, $XA, $XB",
948 (XXPERMDI vsrc:$XT, vsrc:$XA, vsrc:$XB, 0)>;
949 def : InstAlias<"xxmrgld $XT, $XA, $XB",
950 (XXPERMDI vsrc:$XT, vsrc:$XA, vsrc:$XB, 3)>;
951 def : InstAlias<"xxswapd $XT, $XB",
952 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 2)>;
953 def : InstAlias<"xxspltd $XT, $XB, 0",
954 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 0)>;
955 def : InstAlias<"xxspltd $XT, $XB, 1",
956 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 3)>;
957 def : InstAlias<"xxswapd $XT, $XB",
958 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 2)>;
960 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
962 def : Pat<(v4i32 (vnot_ppc v4i32:$A)),
963 (v4i32 (XXLNOR $A, $A))>;
964 let Predicates = [IsBigEndian] in {
965 def : Pat<(v2f64 (scalar_to_vector f64:$A)),
966 (v2f64 (SUBREG_TO_REG (i64 1), $A, sub_64))>;
968 def : Pat<(f64 (extractelt v2f64:$S, 0)),
969 (f64 (EXTRACT_SUBREG $S, sub_64))>;
970 def : Pat<(f64 (extractelt v2f64:$S, 1)),
971 (f64 (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64))>;
974 let Predicates = [IsLittleEndian] in {
975 def : Pat<(v2f64 (scalar_to_vector f64:$A)),
976 (v2f64 (XXPERMDI (SUBREG_TO_REG (i64 1), $A, sub_64),
977 (SUBREG_TO_REG (i64 1), $A, sub_64), 0))>;
979 def : Pat<(f64 (extractelt v2f64:$S, 0)),
980 (f64 (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64))>;
981 def : Pat<(f64 (extractelt v2f64:$S, 1)),
982 (f64 (EXTRACT_SUBREG $S, sub_64))>;
985 // Additional fnmsub patterns: -a*c + b == -(a*c - b)
986 def : Pat<(fma (fneg f64:$A), f64:$C, f64:$B),
987 (XSNMSUBADP $B, $C, $A)>;
988 def : Pat<(fma f64:$A, (fneg f64:$C), f64:$B),
989 (XSNMSUBADP $B, $C, $A)>;
991 def : Pat<(fma (fneg v2f64:$A), v2f64:$C, v2f64:$B),
992 (XVNMSUBADP $B, $C, $A)>;
993 def : Pat<(fma v2f64:$A, (fneg v2f64:$C), v2f64:$B),
994 (XVNMSUBADP $B, $C, $A)>;
996 def : Pat<(fma (fneg v4f32:$A), v4f32:$C, v4f32:$B),
997 (XVNMSUBASP $B, $C, $A)>;
998 def : Pat<(fma v4f32:$A, (fneg v4f32:$C), v4f32:$B),
999 (XVNMSUBASP $B, $C, $A)>;
1001 def : Pat<(v2f64 (bitconvert v4f32:$A)),
1002 (COPY_TO_REGCLASS $A, VSRC)>;
1003 def : Pat<(v2f64 (bitconvert v4i32:$A)),
1004 (COPY_TO_REGCLASS $A, VSRC)>;
1005 def : Pat<(v2f64 (bitconvert v8i16:$A)),
1006 (COPY_TO_REGCLASS $A, VSRC)>;
1007 def : Pat<(v2f64 (bitconvert v16i8:$A)),
1008 (COPY_TO_REGCLASS $A, VSRC)>;
1010 def : Pat<(v4f32 (bitconvert v2f64:$A)),
1011 (COPY_TO_REGCLASS $A, VRRC)>;
1012 def : Pat<(v4i32 (bitconvert v2f64:$A)),
1013 (COPY_TO_REGCLASS $A, VRRC)>;
1014 def : Pat<(v8i16 (bitconvert v2f64:$A)),
1015 (COPY_TO_REGCLASS $A, VRRC)>;
1016 def : Pat<(v16i8 (bitconvert v2f64:$A)),
1017 (COPY_TO_REGCLASS $A, VRRC)>;
1019 def : Pat<(v2i64 (bitconvert v4f32:$A)),
1020 (COPY_TO_REGCLASS $A, VSRC)>;
1021 def : Pat<(v2i64 (bitconvert v4i32:$A)),
1022 (COPY_TO_REGCLASS $A, VSRC)>;
1023 def : Pat<(v2i64 (bitconvert v8i16:$A)),
1024 (COPY_TO_REGCLASS $A, VSRC)>;
1025 def : Pat<(v2i64 (bitconvert v16i8:$A)),
1026 (COPY_TO_REGCLASS $A, VSRC)>;
1028 def : Pat<(v4f32 (bitconvert v2i64:$A)),
1029 (COPY_TO_REGCLASS $A, VRRC)>;
1030 def : Pat<(v4i32 (bitconvert v2i64:$A)),
1031 (COPY_TO_REGCLASS $A, VRRC)>;
1032 def : Pat<(v8i16 (bitconvert v2i64:$A)),
1033 (COPY_TO_REGCLASS $A, VRRC)>;
1034 def : Pat<(v16i8 (bitconvert v2i64:$A)),
1035 (COPY_TO_REGCLASS $A, VRRC)>;
1037 def : Pat<(v2f64 (bitconvert v2i64:$A)),
1038 (COPY_TO_REGCLASS $A, VRRC)>;
1039 def : Pat<(v2i64 (bitconvert v2f64:$A)),
1040 (COPY_TO_REGCLASS $A, VRRC)>;
1042 def : Pat<(v2f64 (bitconvert v1i128:$A)),
1043 (COPY_TO_REGCLASS $A, VRRC)>;
1044 def : Pat<(v1i128 (bitconvert v2f64:$A)),
1045 (COPY_TO_REGCLASS $A, VRRC)>;
1047 def : Pat<(v2i64 (bitconvert f128:$A)),
1048 (COPY_TO_REGCLASS $A, VRRC)>;
1049 def : Pat<(v4i32 (bitconvert f128:$A)),
1050 (COPY_TO_REGCLASS $A, VRRC)>;
1051 def : Pat<(v8i16 (bitconvert f128:$A)),
1052 (COPY_TO_REGCLASS $A, VRRC)>;
1053 def : Pat<(v16i8 (bitconvert f128:$A)),
1054 (COPY_TO_REGCLASS $A, VRRC)>;
1056 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 0)),
1057 (v2f64 (XVCVSXWDP (v2i64 (XXMRGHW $C, $C))))>;
1058 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 1)),
1059 (v2f64 (XVCVSXWDP (v2i64 (XXMRGLW $C, $C))))>;
1061 def : Pat<(v2f64 (PPCuvec2fp v4i32:$C, 0)),
1062 (v2f64 (XVCVUXWDP (v2i64 (XXMRGHW $C, $C))))>;
1063 def : Pat<(v2f64 (PPCuvec2fp v4i32:$C, 1)),
1064 (v2f64 (XVCVUXWDP (v2i64 (XXMRGLW $C, $C))))>;
1067 let Predicates = [HasVSX, HasOnlySwappingMemOps] in {
1068 def : Pat<(v2f64 (PPClxvd2x xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1071 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, xoaddr:$dst),
1072 (STXVD2X $rS, xoaddr:$dst)>;
1073 def : Pat<(PPCstxvd2x v2f64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1075 let Predicates = [IsBigEndian, HasVSX, HasOnlySwappingMemOps] in {
1076 def : Pat<(v2f64 (load xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1077 def : Pat<(v2i64 (load xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1078 def : Pat<(v4i32 (load xoaddr:$src)), (LXVW4X xoaddr:$src)>;
1079 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x xoaddr:$src)), (LXVW4X xoaddr:$src)>;
1080 def : Pat<(store v2f64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1081 def : Pat<(store v2i64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1082 def : Pat<(store v4i32:$XT, xoaddr:$dst), (STXVW4X $XT, xoaddr:$dst)>;
1083 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, xoaddr:$dst),
1084 (STXVW4X $rS, xoaddr:$dst)>;
1088 def : Pat<(v2f64 (PPCxxswapd v2f64:$src)), (XXPERMDI $src, $src, 2)>;
1089 def : Pat<(v2i64 (PPCxxswapd v2i64:$src)), (XXPERMDI $src, $src, 2)>;
1090 def : Pat<(v4f32 (PPCxxswapd v4f32:$src)), (XXPERMDI $src, $src, 2)>;
1091 def : Pat<(v4i32 (PPCxxswapd v4i32:$src)), (XXPERMDI $src, $src, 2)>;
1092 def : Pat<(v2f64 (PPCswapNoChain v2f64:$src)), (XXPERMDI $src, $src, 2)>;
1094 // PPCvecshl XT, XA, XA, 2 can be selected to both XXSLDWI XT,XA,XA,2 and
1095 // XXSWAPD XT,XA (i.e. XXPERMDI XT,XA,XA,2), the later one is more profitable.
1096 def : Pat<(v4i32 (PPCvecshl v4i32:$src, v4i32:$src, 2)), (XXPERMDI $src, $src, 2)>;
1099 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETLT)),
1100 (SELECT_VSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1101 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETULT)),
1102 (SELECT_VSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1103 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETLE)),
1104 (SELECT_VSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1105 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETULE)),
1106 (SELECT_VSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1107 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETEQ)),
1108 (SELECT_VSRC (CREQV $lhs, $rhs), $tval, $fval)>;
1109 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETGE)),
1110 (SELECT_VSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1111 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETUGE)),
1112 (SELECT_VSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1113 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETGT)),
1114 (SELECT_VSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1115 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETUGT)),
1116 (SELECT_VSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1117 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETNE)),
1118 (SELECT_VSRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1120 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETLT)),
1121 (SELECT_VSFRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1122 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETULT)),
1123 (SELECT_VSFRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1124 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETLE)),
1125 (SELECT_VSFRC (CRORC $lhs, $rhs), $tval, $fval)>;
1126 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETULE)),
1127 (SELECT_VSFRC (CRORC $rhs, $lhs), $tval, $fval)>;
1128 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETEQ)),
1129 (SELECT_VSFRC (CREQV $lhs, $rhs), $tval, $fval)>;
1130 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETGE)),
1131 (SELECT_VSFRC (CRORC $rhs, $lhs), $tval, $fval)>;
1132 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGE)),
1133 (SELECT_VSFRC (CRORC $lhs, $rhs), $tval, $fval)>;
1134 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETGT)),
1135 (SELECT_VSFRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1136 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGT)),
1137 (SELECT_VSFRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1138 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETNE)),
1139 (SELECT_VSFRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1142 def : Pat<(int_ppc_vsx_xvdivsp v4f32:$A, v4f32:$B),
1144 def : Pat<(int_ppc_vsx_xvdivdp v2f64:$A, v2f64:$B),
1147 // Reciprocal estimate
1148 def : Pat<(int_ppc_vsx_xvresp v4f32:$A),
1150 def : Pat<(int_ppc_vsx_xvredp v2f64:$A),
1153 // Recip. square root estimate
1154 def : Pat<(int_ppc_vsx_xvrsqrtesp v4f32:$A),
1156 def : Pat<(int_ppc_vsx_xvrsqrtedp v2f64:$A),
1160 def : Pat<(v16i8 (vselect v16i8:$vA, v16i8:$vB, v16i8:$vC)),
1162 (XXSEL (COPY_TO_REGCLASS $vC, VSRC),
1163 (COPY_TO_REGCLASS $vB, VSRC),
1164 (COPY_TO_REGCLASS $vA, VSRC)), VRRC)>;
1165 def : Pat<(v8i16 (vselect v8i16:$vA, v8i16:$vB, v8i16:$vC)),
1167 (XXSEL (COPY_TO_REGCLASS $vC, VSRC),
1168 (COPY_TO_REGCLASS $vB, VSRC),
1169 (COPY_TO_REGCLASS $vA, VSRC)), VRRC)>;
1170 def : Pat<(vselect v4i32:$vA, v4i32:$vB, v4i32:$vC),
1171 (XXSEL $vC, $vB, $vA)>;
1172 def : Pat<(vselect v2i64:$vA, v2i64:$vB, v2i64:$vC),
1173 (XXSEL $vC, $vB, $vA)>;
1174 def : Pat<(vselect v4i32:$vA, v4f32:$vB, v4f32:$vC),
1175 (XXSEL $vC, $vB, $vA)>;
1176 def : Pat<(vselect v2i64:$vA, v2f64:$vB, v2f64:$vC),
1177 (XXSEL $vC, $vB, $vA)>;
1179 let Predicates = [IsLittleEndian] in {
1180 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1181 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1182 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1183 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1184 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1185 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1186 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1187 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1190 let Predicates = [IsBigEndian] in {
1191 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1192 (f64 (XSCVSXDDP (COPY_TO_REGCLASS $S, VSFRC)))>;
1193 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1194 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1195 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1196 (f64 (XSCVUXDDP (COPY_TO_REGCLASS $S, VSFRC)))>;
1197 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1198 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1201 } // AddedComplexity
1205 dag Li8 = (i32 (extloadi8 xoaddr:$src));
1206 dag ZELi8 = (i32 (zextloadi8 xoaddr:$src));
1207 dag ZELi8i64 = (i64 (zextloadi8 xoaddr:$src));
1208 dag SELi8 = (i32 (sext_inreg (extloadi8 xoaddr:$src), i8));
1209 dag SELi8i64 = (i64 (sext_inreg (extloadi8 xoaddr:$src), i8));
1211 dag Li16 = (i32 (extloadi16 xoaddr:$src));
1212 dag ZELi16 = (i32 (zextloadi16 xoaddr:$src));
1213 dag ZELi16i64 = (i64 (zextloadi16 xoaddr:$src));
1214 dag SELi16 = (i32 (sextloadi16 xoaddr:$src));
1215 dag SELi16i64 = (i64 (sextloadi16 xoaddr:$src));
1217 dag Li32 = (i32 (load xoaddr:$src));
1220 def DWToSPExtractConv {
1221 dag El0US1 = (f32 (PPCfcfidus
1222 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S1, 0))))));
1223 dag El1US1 = (f32 (PPCfcfidus
1224 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S1, 1))))));
1225 dag El0US2 = (f32 (PPCfcfidus
1226 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S2, 0))))));
1227 dag El1US2 = (f32 (PPCfcfidus
1228 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S2, 1))))));
1229 dag El0SS1 = (f32 (PPCfcfids
1230 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S1, 0))))));
1231 dag El1SS1 = (f32 (PPCfcfids
1232 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S1, 1))))));
1233 dag El0SS2 = (f32 (PPCfcfids
1234 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S2, 0))))));
1235 dag El1SS2 = (f32 (PPCfcfids
1236 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S2, 1))))));
1237 dag BVU = (v4f32 (build_vector El0US1, El1US1, El0US2, El1US2));
1238 dag BVS = (v4f32 (build_vector El0SS1, El1SS1, El0SS2, El1SS2));
1241 // The following VSX instructions were introduced in Power ISA 2.07
1242 /* FIXME: if the operands are v2i64, these patterns will not match.
1243 we should define new patterns or otherwise match the same patterns
1244 when the elements are larger than i32.
1246 def HasP8Vector : Predicate<"PPCSubTarget->hasP8Vector()">;
1247 def HasDirectMove : Predicate<"PPCSubTarget->hasDirectMove()">;
1248 def NoP9Vector : Predicate<"!PPCSubTarget->hasP9Vector()">;
1249 let Predicates = [HasP8Vector] in {
1250 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
1251 let isCommutable = 1, UseVSXReg = 1 in {
1252 def XXLEQV : XX3Form<60, 186,
1253 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1254 "xxleqv $XT, $XA, $XB", IIC_VecGeneral,
1255 [(set v4i32:$XT, (vnot_ppc (xor v4i32:$XA, v4i32:$XB)))]>;
1256 def XXLNAND : XX3Form<60, 178,
1257 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1258 "xxlnand $XT, $XA, $XB", IIC_VecGeneral,
1259 [(set v4i32:$XT, (vnot_ppc (and v4i32:$XA,
1261 } // isCommutable, UseVSXReg
1263 def : Pat<(int_ppc_vsx_xxleqv v4i32:$A, v4i32:$B),
1266 let UseVSXReg = 1 in {
1267 def XXLORC : XX3Form<60, 170,
1268 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1269 "xxlorc $XT, $XA, $XB", IIC_VecGeneral,
1270 [(set v4i32:$XT, (or v4i32:$XA, (vnot_ppc v4i32:$XB)))]>;
1272 // VSX scalar loads introduced in ISA 2.07
1273 let mayLoad = 1, mayStore = 0 in {
1275 def LXSSPX : XX1Form_memOp<31, 524, (outs vssrc:$XT), (ins memrr:$src),
1276 "lxsspx $XT, $src", IIC_LdStLFD, []>;
1277 def LXSIWAX : XX1Form_memOp<31, 76, (outs vsfrc:$XT), (ins memrr:$src),
1278 "lxsiwax $XT, $src", IIC_LdStLFD, []>;
1279 def LXSIWZX : XX1Form_memOp<31, 12, (outs vsfrc:$XT), (ins memrr:$src),
1280 "lxsiwzx $XT, $src", IIC_LdStLFD, []>;
1282 // Pseudo instruction XFLOADf32 will be expanded to LXSSPX or LFSX later
1284 def XFLOADf32 : PseudoXFormMemOp<(outs vssrc:$XT), (ins memrr:$src),
1286 [(set f32:$XT, (load xoaddr:$src))]>;
1287 // Pseudo instruction LIWAX will be expanded to LXSIWAX or LFIWAX later
1288 def LIWAX : PseudoXFormMemOp<(outs vsfrc:$XT), (ins memrr:$src),
1290 [(set f64:$XT, (PPClfiwax xoaddr:$src))]>;
1291 // Pseudo instruction LIWZX will be expanded to LXSIWZX or LFIWZX later
1292 def LIWZX : PseudoXFormMemOp<(outs vsfrc:$XT), (ins memrr:$src),
1294 [(set f64:$XT, (PPClfiwzx xoaddr:$src))]>;
1297 // VSX scalar stores introduced in ISA 2.07
1298 let mayStore = 1, mayLoad = 0 in {
1300 def STXSSPX : XX1Form_memOp<31, 652, (outs), (ins vssrc:$XT, memrr:$dst),
1301 "stxsspx $XT, $dst", IIC_LdStSTFD, []>;
1302 def STXSIWX : XX1Form_memOp<31, 140, (outs), (ins vsfrc:$XT, memrr:$dst),
1303 "stxsiwx $XT, $dst", IIC_LdStSTFD, []>;
1305 // Pseudo instruction XFSTOREf32 will be expanded to STXSSPX or STFSX later
1307 def XFSTOREf32 : PseudoXFormMemOp<(outs), (ins vssrc:$XT, memrr:$dst),
1309 [(store f32:$XT, xoaddr:$dst)]>;
1310 // Pseudo instruction STIWX will be expanded to STXSIWX or STFIWX later
1311 def STIWX : PseudoXFormMemOp<(outs), (ins vsfrc:$XT, memrr:$dst),
1313 [(PPCstfiwx f64:$XT, xoaddr:$dst)]>;
1317 def : Pat<(f64 (extloadf32 xoaddr:$src)),
1318 (COPY_TO_REGCLASS (XFLOADf32 xoaddr:$src), VSFRC)>;
1319 def : Pat<(f32 (fpround (f64 (extloadf32 xoaddr:$src)))),
1320 (f32 (XFLOADf32 xoaddr:$src))>;
1321 def : Pat<(f64 (fpextend f32:$src)),
1322 (COPY_TO_REGCLASS $src, VSFRC)>;
1324 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLT)),
1325 (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1326 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULT)),
1327 (SELECT_VSSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1328 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLE)),
1329 (SELECT_VSSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1330 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULE)),
1331 (SELECT_VSSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1332 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETEQ)),
1333 (SELECT_VSSRC (CREQV $lhs, $rhs), $tval, $fval)>;
1334 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETGE)),
1335 (SELECT_VSSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1336 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETUGE)),
1337 (SELECT_VSSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1338 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETGT)),
1339 (SELECT_VSSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1340 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETUGT)),
1341 (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1342 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETNE)),
1343 (SELECT_VSSRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1345 let UseVSXReg = 1 in {
1346 // VSX Elementary Scalar FP arithmetic (SP)
1347 let isCommutable = 1 in {
1348 def XSADDSP : XX3Form<60, 0,
1349 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1350 "xsaddsp $XT, $XA, $XB", IIC_VecFP,
1351 [(set f32:$XT, (fadd f32:$XA, f32:$XB))]>;
1352 def XSMULSP : XX3Form<60, 16,
1353 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1354 "xsmulsp $XT, $XA, $XB", IIC_VecFP,
1355 [(set f32:$XT, (fmul f32:$XA, f32:$XB))]>;
1358 def XSDIVSP : XX3Form<60, 24,
1359 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1360 "xsdivsp $XT, $XA, $XB", IIC_FPDivS,
1361 [(set f32:$XT, (fdiv f32:$XA, f32:$XB))]>;
1362 def XSRESP : XX2Form<60, 26,
1363 (outs vssrc:$XT), (ins vssrc:$XB),
1364 "xsresp $XT, $XB", IIC_VecFP,
1365 [(set f32:$XT, (PPCfre f32:$XB))]>;
1366 def XSRSP : XX2Form<60, 281,
1367 (outs vssrc:$XT), (ins vsfrc:$XB),
1368 "xsrsp $XT, $XB", IIC_VecFP, []>;
1369 def XSSQRTSP : XX2Form<60, 11,
1370 (outs vssrc:$XT), (ins vssrc:$XB),
1371 "xssqrtsp $XT, $XB", IIC_FPSqrtS,
1372 [(set f32:$XT, (fsqrt f32:$XB))]>;
1373 def XSRSQRTESP : XX2Form<60, 10,
1374 (outs vssrc:$XT), (ins vssrc:$XB),
1375 "xsrsqrtesp $XT, $XB", IIC_VecFP,
1376 [(set f32:$XT, (PPCfrsqrte f32:$XB))]>;
1377 def XSSUBSP : XX3Form<60, 8,
1378 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1379 "xssubsp $XT, $XA, $XB", IIC_VecFP,
1380 [(set f32:$XT, (fsub f32:$XA, f32:$XB))]>;
1383 let BaseName = "XSMADDASP" in {
1384 let isCommutable = 1 in
1385 def XSMADDASP : XX3Form<60, 1,
1387 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1388 "xsmaddasp $XT, $XA, $XB", IIC_VecFP,
1389 [(set f32:$XT, (fma f32:$XA, f32:$XB, f32:$XTi))]>,
1390 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1392 let IsVSXFMAAlt = 1 in
1393 def XSMADDMSP : XX3Form<60, 9,
1395 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1396 "xsmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
1397 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1401 let BaseName = "XSMSUBASP" in {
1402 let isCommutable = 1 in
1403 def XSMSUBASP : XX3Form<60, 17,
1405 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1406 "xsmsubasp $XT, $XA, $XB", IIC_VecFP,
1407 [(set f32:$XT, (fma f32:$XA, f32:$XB,
1408 (fneg f32:$XTi)))]>,
1409 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1411 let IsVSXFMAAlt = 1 in
1412 def XSMSUBMSP : XX3Form<60, 25,
1414 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1415 "xsmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
1416 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1420 let BaseName = "XSNMADDASP" in {
1421 let isCommutable = 1 in
1422 def XSNMADDASP : XX3Form<60, 129,
1424 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1425 "xsnmaddasp $XT, $XA, $XB", IIC_VecFP,
1426 [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
1428 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1430 let IsVSXFMAAlt = 1 in
1431 def XSNMADDMSP : XX3Form<60, 137,
1433 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1434 "xsnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
1435 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1439 let BaseName = "XSNMSUBASP" in {
1440 let isCommutable = 1 in
1441 def XSNMSUBASP : XX3Form<60, 145,
1443 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1444 "xsnmsubasp $XT, $XA, $XB", IIC_VecFP,
1445 [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
1446 (fneg f32:$XTi))))]>,
1447 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1449 let IsVSXFMAAlt = 1 in
1450 def XSNMSUBMSP : XX3Form<60, 153,
1452 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1453 "xsnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
1454 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1458 // Single Precision Conversions (FP <-> INT)
1459 def XSCVSXDSP : XX2Form<60, 312,
1460 (outs vssrc:$XT), (ins vsfrc:$XB),
1461 "xscvsxdsp $XT, $XB", IIC_VecFP,
1462 [(set f32:$XT, (PPCfcfids f64:$XB))]>;
1463 def XSCVUXDSP : XX2Form<60, 296,
1464 (outs vssrc:$XT), (ins vsfrc:$XB),
1465 "xscvuxdsp $XT, $XB", IIC_VecFP,
1466 [(set f32:$XT, (PPCfcfidus f64:$XB))]>;
1468 // Conversions between vector and scalar single precision
1469 def XSCVDPSPN : XX2Form<60, 267, (outs vsrc:$XT), (ins vssrc:$XB),
1470 "xscvdpspn $XT, $XB", IIC_VecFP, []>;
1471 def XSCVSPDPN : XX2Form<60, 331, (outs vssrc:$XT), (ins vsrc:$XB),
1472 "xscvspdpn $XT, $XB", IIC_VecFP, []>;
1475 let Predicates = [IsLittleEndian] in {
1476 def : Pat<DWToSPExtractConv.El0SS1,
1477 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (XXPERMDI $S1, $S1, 2), VSFRC)))>;
1478 def : Pat<DWToSPExtractConv.El1SS1,
1479 (f32 (XSCVSXDSP (COPY_TO_REGCLASS
1480 (f64 (COPY_TO_REGCLASS $S1, VSRC)), VSFRC)))>;
1481 def : Pat<DWToSPExtractConv.El0US1,
1482 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (XXPERMDI $S1, $S1, 2), VSFRC)))>;
1483 def : Pat<DWToSPExtractConv.El1US1,
1484 (f32 (XSCVUXDSP (COPY_TO_REGCLASS
1485 (f64 (COPY_TO_REGCLASS $S1, VSRC)), VSFRC)))>;
1488 let Predicates = [IsBigEndian] in {
1489 def : Pat<DWToSPExtractConv.El0SS1,
1490 (f32 (XSCVSXDSP (COPY_TO_REGCLASS $S1, VSFRC)))>;
1491 def : Pat<DWToSPExtractConv.El1SS1,
1492 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (XXPERMDI $S1, $S1, 2), VSFRC)))>;
1493 def : Pat<DWToSPExtractConv.El0US1,
1494 (f32 (XSCVUXDSP (COPY_TO_REGCLASS $S1, VSFRC)))>;
1495 def : Pat<DWToSPExtractConv.El1US1,
1496 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (XXPERMDI $S1, $S1, 2), VSFRC)))>;
1499 // Instructions for converting float to i64 feeding a store.
1500 let Predicates = [NoP9Vector] in {
1501 def : Pat<(PPCstore_scal_int_from_vsr
1502 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xoaddr:$dst, 8),
1503 (STXSDX (XSCVDPSXDS f64:$src), xoaddr:$dst)>;
1504 def : Pat<(PPCstore_scal_int_from_vsr
1505 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xoaddr:$dst, 8),
1506 (STXSDX (XSCVDPUXDS f64:$src), xoaddr:$dst)>;
1509 // Instructions for converting float to i32 feeding a store.
1510 def : Pat<(PPCstore_scal_int_from_vsr
1511 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xoaddr:$dst, 4),
1512 (STIWX (XSCVDPSXWS f64:$src), xoaddr:$dst)>;
1513 def : Pat<(PPCstore_scal_int_from_vsr
1514 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xoaddr:$dst, 4),
1515 (STIWX (XSCVDPUXWS f64:$src), xoaddr:$dst)>;
1517 } // AddedComplexity = 400
1520 let UseVSXReg = 1, AddedComplexity = 400 in {
1521 let Predicates = [HasDirectMove] in {
1522 // VSX direct move instructions
1523 def MFVSRD : XX1_RS6_RD5_XO<31, 51, (outs g8rc:$rA), (ins vsfrc:$XT),
1524 "mfvsrd $rA, $XT", IIC_VecGeneral,
1525 [(set i64:$rA, (PPCmfvsr f64:$XT))]>,
1526 Requires<[In64BitMode]>;
1527 let isCodeGenOnly = 1 in
1528 def MFVRD : XX1_RS6_RD5_XO<31, 51, (outs g8rc:$rA), (ins vrrc:$XT),
1529 "mfvsrd $rA, $XT", IIC_VecGeneral,
1531 Requires<[In64BitMode]>;
1532 def MFVSRWZ : XX1_RS6_RD5_XO<31, 115, (outs gprc:$rA), (ins vsfrc:$XT),
1533 "mfvsrwz $rA, $XT", IIC_VecGeneral,
1534 [(set i32:$rA, (PPCmfvsr f64:$XT))]>;
1535 def MTVSRD : XX1_RS6_RD5_XO<31, 179, (outs vsfrc:$XT), (ins g8rc:$rA),
1536 "mtvsrd $XT, $rA", IIC_VecGeneral,
1537 [(set f64:$XT, (PPCmtvsra i64:$rA))]>,
1538 Requires<[In64BitMode]>;
1539 def MTVSRWA : XX1_RS6_RD5_XO<31, 211, (outs vsfrc:$XT), (ins gprc:$rA),
1540 "mtvsrwa $XT, $rA", IIC_VecGeneral,
1541 [(set f64:$XT, (PPCmtvsra i32:$rA))]>;
1542 def MTVSRWZ : XX1_RS6_RD5_XO<31, 243, (outs vsfrc:$XT), (ins gprc:$rA),
1543 "mtvsrwz $XT, $rA", IIC_VecGeneral,
1544 [(set f64:$XT, (PPCmtvsrz i32:$rA))]>;
1547 let Predicates = [IsISA3_0, HasDirectMove] in {
1548 def MTVSRWS: XX1_RS6_RD5_XO<31, 403, (outs vsrc:$XT), (ins gprc:$rA),
1549 "mtvsrws $XT, $rA", IIC_VecGeneral, []>;
1551 def MTVSRDD: XX1Form<31, 435, (outs vsrc:$XT), (ins g8rc_nox0:$rA, g8rc:$rB),
1552 "mtvsrdd $XT, $rA, $rB", IIC_VecGeneral,
1553 []>, Requires<[In64BitMode]>;
1555 def MFVSRLD: XX1_RS6_RD5_XO<31, 307, (outs g8rc:$rA), (ins vsrc:$XT),
1556 "mfvsrld $rA, $XT", IIC_VecGeneral,
1557 []>, Requires<[In64BitMode]>;
1559 } // IsISA3_0, HasDirectMove
1562 // We want to parse this from asm, but we don't want to emit this as it would
1563 // be emitted with a VSX reg. So leave Emit = 0 here.
1564 def : InstAlias<"mfvrd $rA, $XT",
1565 (MFVRD g8rc:$rA, vrrc:$XT), 0>;
1566 def : InstAlias<"mffprd $rA, $src",
1567 (MFVSRD g8rc:$rA, f8rc:$src)>;
1569 /* Direct moves of various widths from GPR's into VSR's. Each move lines
1570 the value up into element 0 (both BE and LE). Namely, entities smaller than
1571 a doubleword are shifted left and moved for BE. For LE, they're moved, then
1572 swapped to go into the least significant element of the VSR.
1578 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 56, 7));
1582 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 48, 15));
1586 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 32, 31));
1587 dag BE_DWORD_0 = (MTVSRD $A);
1589 dag LE_MTVSRW = (MTVSRD (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32));
1590 dag LE_WORD_1 = (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1591 LE_MTVSRW, sub_64));
1592 dag LE_WORD_0 = (XXPERMDI LE_WORD_1, LE_WORD_1, 2);
1593 dag LE_DWORD_1 = (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1594 BE_DWORD_0, sub_64));
1595 dag LE_DWORD_0 = (XXPERMDI LE_DWORD_1, LE_DWORD_1, 2);
1598 /* Patterns for extracting elements out of vectors. Integer elements are
1599 extracted using direct move operations. Patterns for extracting elements
1600 whose indices are not available at compile time are also provided with
1601 various _VARIABLE_ patterns.
1602 The numbering for the DAG's is for LE, but when used on BE, the correct
1603 LE element can just be used (i.e. LE_BYTE_2 == BE_BYTE_13).
1605 def VectorExtractions {
1606 // Doubleword extraction
1610 (XXPERMDI (COPY_TO_REGCLASS $S, VSRC),
1611 (COPY_TO_REGCLASS $S, VSRC), 2), sub_64));
1612 dag LE_DWORD_1 = (MFVSRD
1614 (v2i64 (COPY_TO_REGCLASS $S, VSRC)), sub_64));
1617 dag LE_WORD_0 = (MFVSRWZ (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64));
1618 dag LE_WORD_1 = (MFVSRWZ (EXTRACT_SUBREG (XXSLDWI $S, $S, 1), sub_64));
1619 dag LE_WORD_2 = (MFVSRWZ (EXTRACT_SUBREG
1620 (v2i64 (COPY_TO_REGCLASS $S, VSRC)), sub_64));
1621 dag LE_WORD_3 = (MFVSRWZ (EXTRACT_SUBREG (XXSLDWI $S, $S, 3), sub_64));
1623 // Halfword extraction
1624 dag LE_HALF_0 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 0, 48), sub_32));
1625 dag LE_HALF_1 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 48, 48), sub_32));
1626 dag LE_HALF_2 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 32, 48), sub_32));
1627 dag LE_HALF_3 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 16, 48), sub_32));
1628 dag LE_HALF_4 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 0, 48), sub_32));
1629 dag LE_HALF_5 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 48, 48), sub_32));
1630 dag LE_HALF_6 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 32, 48), sub_32));
1631 dag LE_HALF_7 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 16, 48), sub_32));
1634 dag LE_BYTE_0 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 0, 56), sub_32));
1635 dag LE_BYTE_1 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 56, 56), sub_32));
1636 dag LE_BYTE_2 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 48, 56), sub_32));
1637 dag LE_BYTE_3 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 40, 56), sub_32));
1638 dag LE_BYTE_4 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 32, 56), sub_32));
1639 dag LE_BYTE_5 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 24, 56), sub_32));
1640 dag LE_BYTE_6 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 16, 56), sub_32));
1641 dag LE_BYTE_7 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 8, 56), sub_32));
1642 dag LE_BYTE_8 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 0, 56), sub_32));
1643 dag LE_BYTE_9 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 56, 56), sub_32));
1644 dag LE_BYTE_10 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 48, 56), sub_32));
1645 dag LE_BYTE_11 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 40, 56), sub_32));
1646 dag LE_BYTE_12 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 32, 56), sub_32));
1647 dag LE_BYTE_13 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 24, 56), sub_32));
1648 dag LE_BYTE_14 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 16, 56), sub_32));
1649 dag LE_BYTE_15 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 8, 56), sub_32));
1651 /* Variable element number (BE and LE patterns must be specified separately)
1652 This is a rather involved process.
1654 Conceptually, this is how the move is accomplished:
1655 1. Identify which doubleword contains the element
1656 2. Shift in the VMX register so that the correct doubleword is correctly
1657 lined up for the MFVSRD
1658 3. Perform the move so that the element (along with some extra stuff)
1660 4. Right shift within the GPR so that the element is right-justified
1662 Of course, the index is an element number which has a different meaning
1663 on LE/BE so the patterns have to be specified separately.
1665 Note: The final result will be the element right-justified with high
1666 order bits being arbitrarily defined (namely, whatever was in the
1667 vector register to the left of the value originally).
1672 - For elements 0-7, we shift left by 8 bytes since they're on the right
1673 - For elements 8-15, we need not shift (shift left by zero bytes)
1674 This is accomplished by inverting the bits of the index and AND-ing
1675 with 0x8 (i.e. clearing all bits of the index and inverting bit 60).
1677 dag LE_VBYTE_PERM_VEC = (v16i8 (LVSL ZERO8, (ANDC8 (LI8 8), $Idx)));
1680 // - Now that we set up the shift amount, we shift in the VMX register
1681 dag LE_VBYTE_PERMUTE = (v16i8 (VPERM $S, $S, LE_VBYTE_PERM_VEC));
1684 // - The doubleword containing our element is moved to a GPR
1685 dag LE_MV_VBYTE = (MFVSRD
1687 (v2i64 (COPY_TO_REGCLASS LE_VBYTE_PERMUTE, VSRC)),
1691 - Truncate the element number to the range 0-7 (8-15 are symmetrical
1692 and out of range values are truncated accordingly)
1693 - Multiply by 8 as we need to shift right by the number of bits, not bytes
1694 - Shift right in the GPR by the calculated value
1696 dag LE_VBYTE_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 7), $Idx), 3, 60),
1698 dag LE_VARIABLE_BYTE = (EXTRACT_SUBREG (SRD LE_MV_VBYTE, LE_VBYTE_SHIFT),
1701 /* LE variable halfword
1703 - For elements 0-3, we shift left by 8 since they're on the right
1704 - For elements 4-7, we need not shift (shift left by zero bytes)
1705 Similarly to the byte pattern, we invert the bits of the index, but we
1706 AND with 0x4 (i.e. clear all bits of the index and invert bit 61).
1707 Of course, the shift is still by 8 bytes, so we must multiply by 2.
1709 dag LE_VHALF_PERM_VEC =
1710 (v16i8 (LVSL ZERO8, (RLDICR (ANDC8 (LI8 4), $Idx), 1, 62)));
1713 // - Now that we set up the shift amount, we shift in the VMX register
1714 dag LE_VHALF_PERMUTE = (v16i8 (VPERM $S, $S, LE_VHALF_PERM_VEC));
1717 // - The doubleword containing our element is moved to a GPR
1718 dag LE_MV_VHALF = (MFVSRD
1720 (v2i64 (COPY_TO_REGCLASS LE_VHALF_PERMUTE, VSRC)),
1724 - Truncate the element number to the range 0-3 (4-7 are symmetrical
1725 and out of range values are truncated accordingly)
1726 - Multiply by 16 as we need to shift right by the number of bits
1727 - Shift right in the GPR by the calculated value
1729 dag LE_VHALF_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 3), $Idx), 4, 59),
1731 dag LE_VARIABLE_HALF = (EXTRACT_SUBREG (SRD LE_MV_VHALF, LE_VHALF_SHIFT),
1736 - For elements 0-1, we shift left by 8 since they're on the right
1737 - For elements 2-3, we need not shift
1739 dag LE_VWORD_PERM_VEC = (v16i8 (LVSL ZERO8,
1740 (RLDICR (ANDC8 (LI8 2), $Idx), 2, 61)));
1743 // - Now that we set up the shift amount, we shift in the VMX register
1744 dag LE_VWORD_PERMUTE = (v16i8 (VPERM $S, $S, LE_VWORD_PERM_VEC));
1747 // - The doubleword containing our element is moved to a GPR
1748 dag LE_MV_VWORD = (MFVSRD
1750 (v2i64 (COPY_TO_REGCLASS LE_VWORD_PERMUTE, VSRC)),
1754 - Truncate the element number to the range 0-1 (2-3 are symmetrical
1755 and out of range values are truncated accordingly)
1756 - Multiply by 32 as we need to shift right by the number of bits
1757 - Shift right in the GPR by the calculated value
1759 dag LE_VWORD_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 1), $Idx), 5, 58),
1761 dag LE_VARIABLE_WORD = (EXTRACT_SUBREG (SRD LE_MV_VWORD, LE_VWORD_SHIFT),
1764 /* LE variable doubleword
1766 - For element 0, we shift left by 8 since it's on the right
1767 - For element 1, we need not shift
1769 dag LE_VDWORD_PERM_VEC = (v16i8 (LVSL ZERO8,
1770 (RLDICR (ANDC8 (LI8 1), $Idx), 3, 60)));
1773 // - Now that we set up the shift amount, we shift in the VMX register
1774 dag LE_VDWORD_PERMUTE = (v16i8 (VPERM $S, $S, LE_VDWORD_PERM_VEC));
1777 // - The doubleword containing our element is moved to a GPR
1778 // - Number 4. is not needed for the doubleword as the value is 64-bits
1779 dag LE_VARIABLE_DWORD =
1780 (MFVSRD (EXTRACT_SUBREG
1781 (v2i64 (COPY_TO_REGCLASS LE_VDWORD_PERMUTE, VSRC)),
1784 /* LE variable float
1785 - Shift the vector to line up the desired element to BE Word 0
1786 - Convert 32-bit float to a 64-bit single precision float
1788 dag LE_VFLOAT_PERM_VEC = (v16i8 (LVSL ZERO8,
1789 (RLDICR (XOR8 (LI8 3), $Idx), 2, 61)));
1790 dag LE_VFLOAT_PERMUTE = (VPERM $S, $S, LE_VFLOAT_PERM_VEC);
1791 dag LE_VARIABLE_FLOAT = (XSCVSPDPN LE_VFLOAT_PERMUTE);
1793 /* LE variable double
1794 Same as the LE doubleword except there is no move.
1796 dag LE_VDOUBLE_PERMUTE = (v16i8 (VPERM (v16i8 (COPY_TO_REGCLASS $S, VRRC)),
1797 (v16i8 (COPY_TO_REGCLASS $S, VRRC)),
1798 LE_VDWORD_PERM_VEC));
1799 dag LE_VARIABLE_DOUBLE = (COPY_TO_REGCLASS LE_VDOUBLE_PERMUTE, VSRC);
1802 The algorithm here is the same as the LE variable byte except:
1803 - The shift in the VMX register is by 0/8 for opposite element numbers so
1804 we simply AND the element number with 0x8
1805 - The order of elements after the move to GPR is reversed, so we invert
1806 the bits of the index prior to truncating to the range 0-7
1808 dag BE_VBYTE_PERM_VEC = (v16i8 (LVSL ZERO8, (ANDIo8 $Idx, 8)));
1809 dag BE_VBYTE_PERMUTE = (v16i8 (VPERM $S, $S, BE_VBYTE_PERM_VEC));
1810 dag BE_MV_VBYTE = (MFVSRD
1812 (v2i64 (COPY_TO_REGCLASS BE_VBYTE_PERMUTE, VSRC)),
1814 dag BE_VBYTE_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 7), $Idx), 3, 60),
1816 dag BE_VARIABLE_BYTE = (EXTRACT_SUBREG (SRD BE_MV_VBYTE, BE_VBYTE_SHIFT),
1819 /* BE variable halfword
1820 The algorithm here is the same as the LE variable halfword except:
1821 - The shift in the VMX register is by 0/8 for opposite element numbers so
1822 we simply AND the element number with 0x4 and multiply by 2
1823 - The order of elements after the move to GPR is reversed, so we invert
1824 the bits of the index prior to truncating to the range 0-3
1826 dag BE_VHALF_PERM_VEC = (v16i8 (LVSL ZERO8,
1827 (RLDICR (ANDIo8 $Idx, 4), 1, 62)));
1828 dag BE_VHALF_PERMUTE = (v16i8 (VPERM $S, $S, BE_VHALF_PERM_VEC));
1829 dag BE_MV_VHALF = (MFVSRD
1831 (v2i64 (COPY_TO_REGCLASS BE_VHALF_PERMUTE, VSRC)),
1833 dag BE_VHALF_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 3), $Idx), 4, 59),
1835 dag BE_VARIABLE_HALF = (EXTRACT_SUBREG (SRD BE_MV_VHALF, BE_VHALF_SHIFT),
1839 The algorithm is the same as the LE variable word except:
1840 - The shift in the VMX register happens for opposite element numbers
1841 - The order of elements after the move to GPR is reversed, so we invert
1842 the bits of the index prior to truncating to the range 0-1
1844 dag BE_VWORD_PERM_VEC = (v16i8 (LVSL ZERO8,
1845 (RLDICR (ANDIo8 $Idx, 2), 2, 61)));
1846 dag BE_VWORD_PERMUTE = (v16i8 (VPERM $S, $S, BE_VWORD_PERM_VEC));
1847 dag BE_MV_VWORD = (MFVSRD
1849 (v2i64 (COPY_TO_REGCLASS BE_VWORD_PERMUTE, VSRC)),
1851 dag BE_VWORD_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 1), $Idx), 5, 58),
1853 dag BE_VARIABLE_WORD = (EXTRACT_SUBREG (SRD BE_MV_VWORD, BE_VWORD_SHIFT),
1856 /* BE variable doubleword
1857 Same as the LE doubleword except we shift in the VMX register for opposite
1860 dag BE_VDWORD_PERM_VEC = (v16i8 (LVSL ZERO8,
1861 (RLDICR (ANDIo8 $Idx, 1), 3, 60)));
1862 dag BE_VDWORD_PERMUTE = (v16i8 (VPERM $S, $S, BE_VDWORD_PERM_VEC));
1863 dag BE_VARIABLE_DWORD =
1864 (MFVSRD (EXTRACT_SUBREG
1865 (v2i64 (COPY_TO_REGCLASS BE_VDWORD_PERMUTE, VSRC)),
1868 /* BE variable float
1869 - Shift the vector to line up the desired element to BE Word 0
1870 - Convert 32-bit float to a 64-bit single precision float
1872 dag BE_VFLOAT_PERM_VEC = (v16i8 (LVSL ZERO8, (RLDICR $Idx, 2, 61)));
1873 dag BE_VFLOAT_PERMUTE = (VPERM $S, $S, BE_VFLOAT_PERM_VEC);
1874 dag BE_VARIABLE_FLOAT = (XSCVSPDPN BE_VFLOAT_PERMUTE);
1876 /* BE variable double
1877 Same as the BE doubleword except there is no move.
1879 dag BE_VDOUBLE_PERMUTE = (v16i8 (VPERM (v16i8 (COPY_TO_REGCLASS $S, VRRC)),
1880 (v16i8 (COPY_TO_REGCLASS $S, VRRC)),
1881 BE_VDWORD_PERM_VEC));
1882 dag BE_VARIABLE_DOUBLE = (COPY_TO_REGCLASS BE_VDOUBLE_PERMUTE, VSRC);
1885 def NoP9Altivec : Predicate<"!PPCSubTarget->hasP9Altivec()">;
1886 let AddedComplexity = 400 in {
1887 // v4f32 scalar <-> vector conversions (BE)
1888 let Predicates = [IsBigEndian, HasP8Vector] in {
1889 def : Pat<(v4f32 (scalar_to_vector f32:$A)),
1890 (v4f32 (XSCVDPSPN $A))>;
1891 def : Pat<(f32 (vector_extract v4f32:$S, 0)),
1892 (f32 (XSCVSPDPN $S))>;
1893 def : Pat<(f32 (vector_extract v4f32:$S, 1)),
1894 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 1)))>;
1895 def : Pat<(f32 (vector_extract v4f32:$S, 2)),
1896 (f32 (XSCVSPDPN (XXPERMDI $S, $S, 2)))>;
1897 def : Pat<(f32 (vector_extract v4f32:$S, 3)),
1898 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 3)))>;
1899 def : Pat<(f32 (vector_extract v4f32:$S, i64:$Idx)),
1900 (f32 VectorExtractions.BE_VARIABLE_FLOAT)>;
1901 } // IsBigEndian, HasP8Vector
1903 // Variable index vector_extract for v2f64 does not require P8Vector
1904 let Predicates = [IsBigEndian, HasVSX] in
1905 def : Pat<(f64 (vector_extract v2f64:$S, i64:$Idx)),
1906 (f64 VectorExtractions.BE_VARIABLE_DOUBLE)>;
1908 let Predicates = [IsBigEndian, HasDirectMove] in {
1909 // v16i8 scalar <-> vector conversions (BE)
1910 def : Pat<(v16i8 (scalar_to_vector i32:$A)),
1911 (v16i8 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_BYTE_0, sub_64))>;
1912 def : Pat<(v8i16 (scalar_to_vector i32:$A)),
1913 (v8i16 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_HALF_0, sub_64))>;
1914 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
1915 (v4i32 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_WORD_0, sub_64))>;
1916 def : Pat<(v2i64 (scalar_to_vector i64:$A)),
1917 (v2i64 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_DWORD_0, sub_64))>;
1919 // v2i64 scalar <-> vector conversions (BE)
1920 def : Pat<(i64 (vector_extract v2i64:$S, 0)),
1921 (i64 VectorExtractions.LE_DWORD_1)>;
1922 def : Pat<(i64 (vector_extract v2i64:$S, 1)),
1923 (i64 VectorExtractions.LE_DWORD_0)>;
1924 def : Pat<(i64 (vector_extract v2i64:$S, i64:$Idx)),
1925 (i64 VectorExtractions.BE_VARIABLE_DWORD)>;
1926 } // IsBigEndian, HasDirectMove
1928 let Predicates = [IsBigEndian, HasDirectMove, NoP9Altivec] in {
1929 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
1930 (i32 VectorExtractions.LE_BYTE_15)>;
1931 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
1932 (i32 VectorExtractions.LE_BYTE_14)>;
1933 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
1934 (i32 VectorExtractions.LE_BYTE_13)>;
1935 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
1936 (i32 VectorExtractions.LE_BYTE_12)>;
1937 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
1938 (i32 VectorExtractions.LE_BYTE_11)>;
1939 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
1940 (i32 VectorExtractions.LE_BYTE_10)>;
1941 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
1942 (i32 VectorExtractions.LE_BYTE_9)>;
1943 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
1944 (i32 VectorExtractions.LE_BYTE_8)>;
1945 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
1946 (i32 VectorExtractions.LE_BYTE_7)>;
1947 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
1948 (i32 VectorExtractions.LE_BYTE_6)>;
1949 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
1950 (i32 VectorExtractions.LE_BYTE_5)>;
1951 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
1952 (i32 VectorExtractions.LE_BYTE_4)>;
1953 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
1954 (i32 VectorExtractions.LE_BYTE_3)>;
1955 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
1956 (i32 VectorExtractions.LE_BYTE_2)>;
1957 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
1958 (i32 VectorExtractions.LE_BYTE_1)>;
1959 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
1960 (i32 VectorExtractions.LE_BYTE_0)>;
1961 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
1962 (i32 VectorExtractions.BE_VARIABLE_BYTE)>;
1964 // v8i16 scalar <-> vector conversions (BE)
1965 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
1966 (i32 VectorExtractions.LE_HALF_7)>;
1967 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
1968 (i32 VectorExtractions.LE_HALF_6)>;
1969 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
1970 (i32 VectorExtractions.LE_HALF_5)>;
1971 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
1972 (i32 VectorExtractions.LE_HALF_4)>;
1973 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
1974 (i32 VectorExtractions.LE_HALF_3)>;
1975 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
1976 (i32 VectorExtractions.LE_HALF_2)>;
1977 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
1978 (i32 VectorExtractions.LE_HALF_1)>;
1979 def : Pat<(i32 (vector_extract v8i16:$S, 7)),
1980 (i32 VectorExtractions.LE_HALF_0)>;
1981 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
1982 (i32 VectorExtractions.BE_VARIABLE_HALF)>;
1984 // v4i32 scalar <-> vector conversions (BE)
1985 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
1986 (i32 VectorExtractions.LE_WORD_3)>;
1987 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
1988 (i32 VectorExtractions.LE_WORD_2)>;
1989 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
1990 (i32 VectorExtractions.LE_WORD_1)>;
1991 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
1992 (i32 VectorExtractions.LE_WORD_0)>;
1993 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
1994 (i32 VectorExtractions.BE_VARIABLE_WORD)>;
1995 } // IsBigEndian, HasDirectMove, NoP9Altivec
1997 // v4f32 scalar <-> vector conversions (LE)
1998 let Predicates = [IsLittleEndian, HasP8Vector] in {
1999 def : Pat<(v4f32 (scalar_to_vector f32:$A)),
2000 (v4f32 (XXSLDWI (XSCVDPSPN $A), (XSCVDPSPN $A), 1))>;
2001 def : Pat<(f32 (vector_extract v4f32:$S, 0)),
2002 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 3)))>;
2003 def : Pat<(f32 (vector_extract v4f32:$S, 1)),
2004 (f32 (XSCVSPDPN (XXPERMDI $S, $S, 2)))>;
2005 def : Pat<(f32 (vector_extract v4f32:$S, 2)),
2006 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 1)))>;
2007 def : Pat<(f32 (vector_extract v4f32:$S, 3)),
2008 (f32 (XSCVSPDPN $S))>;
2009 def : Pat<(f32 (vector_extract v4f32:$S, i64:$Idx)),
2010 (f32 VectorExtractions.LE_VARIABLE_FLOAT)>;
2011 } // IsLittleEndian, HasP8Vector
2013 // Variable index vector_extract for v2f64 does not require P8Vector
2014 let Predicates = [IsLittleEndian, HasVSX] in
2015 def : Pat<(f64 (vector_extract v2f64:$S, i64:$Idx)),
2016 (f64 VectorExtractions.LE_VARIABLE_DOUBLE)>;
2018 def : Pat<(int_ppc_vsx_stxvd2x_be v2f64:$rS, xoaddr:$dst),
2019 (STXVD2X $rS, xoaddr:$dst)>;
2020 def : Pat<(int_ppc_vsx_stxvw4x_be v4i32:$rS, xoaddr:$dst),
2021 (STXVW4X $rS, xoaddr:$dst)>;
2022 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x_be xoaddr:$src)), (LXVW4X xoaddr:$src)>;
2023 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x_be xoaddr:$src)), (LXVD2X xoaddr:$src)>;
2025 // Variable index unsigned vector_extract on Power9
2026 let Predicates = [HasP9Altivec, IsLittleEndian] in {
2027 def : Pat<(i64 (anyext (i32 (vector_extract v16i8:$S, i64:$Idx)))),
2028 (VEXTUBRX $Idx, $S)>;
2030 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, i64:$Idx)))),
2031 (VEXTUHRX (RLWINM8 $Idx, 1, 28, 30), $S)>;
2032 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 0)))),
2033 (VEXTUHRX (LI8 0), $S)>;
2034 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 1)))),
2035 (VEXTUHRX (LI8 2), $S)>;
2036 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 2)))),
2037 (VEXTUHRX (LI8 4), $S)>;
2038 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 3)))),
2039 (VEXTUHRX (LI8 6), $S)>;
2040 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 4)))),
2041 (VEXTUHRX (LI8 8), $S)>;
2042 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 5)))),
2043 (VEXTUHRX (LI8 10), $S)>;
2044 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 6)))),
2045 (VEXTUHRX (LI8 12), $S)>;
2046 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 7)))),
2047 (VEXTUHRX (LI8 14), $S)>;
2049 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, i64:$Idx)))),
2050 (VEXTUWRX (RLWINM8 $Idx, 2, 28, 29), $S)>;
2051 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 0)))),
2052 (VEXTUWRX (LI8 0), $S)>;
2053 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 1)))),
2054 (VEXTUWRX (LI8 4), $S)>;
2055 // For extracting LE word 2, MFVSRWZ is better than VEXTUWRX
2056 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 2)))),
2057 (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
2058 (i32 VectorExtractions.LE_WORD_2), sub_32)>;
2059 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 3)))),
2060 (VEXTUWRX (LI8 12), $S)>;
2062 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, i64:$Idx)))),
2063 (EXTSW (VEXTUWRX (RLWINM8 $Idx, 2, 28, 29), $S))>;
2064 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 0)))),
2065 (EXTSW (VEXTUWRX (LI8 0), $S))>;
2066 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 1)))),
2067 (EXTSW (VEXTUWRX (LI8 4), $S))>;
2068 // For extracting LE word 2, MFVSRWZ is better than VEXTUWRX
2069 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 2)))),
2070 (EXTSW (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
2071 (i32 VectorExtractions.LE_WORD_2), sub_32))>;
2072 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 3)))),
2073 (EXTSW (VEXTUWRX (LI8 12), $S))>;
2075 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
2076 (i32 (EXTRACT_SUBREG (VEXTUBRX $Idx, $S), sub_32))>;
2077 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
2078 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 0), $S), sub_32))>;
2079 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
2080 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 1), $S), sub_32))>;
2081 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
2082 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 2), $S), sub_32))>;
2083 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
2084 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 3), $S), sub_32))>;
2085 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
2086 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 4), $S), sub_32))>;
2087 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
2088 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 5), $S), sub_32))>;
2089 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
2090 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 6), $S), sub_32))>;
2091 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
2092 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 7), $S), sub_32))>;
2093 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
2094 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 8), $S), sub_32))>;
2095 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
2096 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 9), $S), sub_32))>;
2097 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
2098 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 10), $S), sub_32))>;
2099 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
2100 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 11), $S), sub_32))>;
2101 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
2102 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 12), $S), sub_32))>;
2103 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
2104 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 13), $S), sub_32))>;
2105 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
2106 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 14), $S), sub_32))>;
2107 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
2108 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 15), $S), sub_32))>;
2110 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
2111 (i32 (EXTRACT_SUBREG (VEXTUHRX
2112 (RLWINM8 $Idx, 1, 28, 30), $S), sub_32))>;
2113 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
2114 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 0), $S), sub_32))>;
2115 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
2116 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 2), $S), sub_32))>;
2117 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
2118 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 4), $S), sub_32))>;
2119 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
2120 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 6), $S), sub_32))>;
2121 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
2122 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 8), $S), sub_32))>;
2123 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
2124 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 10), $S), sub_32))>;
2125 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2126 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 12), $S), sub_32))>;
2127 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2128 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 14), $S), sub_32))>;
2130 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
2131 (i32 (EXTRACT_SUBREG (VEXTUWRX
2132 (RLWINM8 $Idx, 2, 28, 29), $S), sub_32))>;
2133 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
2134 (i32 (EXTRACT_SUBREG (VEXTUWRX (LI8 0), $S), sub_32))>;
2135 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
2136 (i32 (EXTRACT_SUBREG (VEXTUWRX (LI8 4), $S), sub_32))>;
2137 // For extracting LE word 2, MFVSRWZ is better than VEXTUWRX
2138 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
2139 (i32 VectorExtractions.LE_WORD_2)>;
2140 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
2141 (i32 (EXTRACT_SUBREG (VEXTUWRX (LI8 12), $S), sub_32))>;
2144 let Predicates = [HasP9Altivec, IsBigEndian] in {
2145 def : Pat<(i64 (anyext (i32 (vector_extract v16i8:$S, i64:$Idx)))),
2146 (VEXTUBLX $Idx, $S)>;
2148 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, i64:$Idx)))),
2149 (VEXTUHLX (RLWINM8 $Idx, 1, 28, 30), $S)>;
2150 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 0)))),
2151 (VEXTUHLX (LI8 0), $S)>;
2152 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 1)))),
2153 (VEXTUHLX (LI8 2), $S)>;
2154 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 2)))),
2155 (VEXTUHLX (LI8 4), $S)>;
2156 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 3)))),
2157 (VEXTUHLX (LI8 6), $S)>;
2158 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 4)))),
2159 (VEXTUHLX (LI8 8), $S)>;
2160 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 5)))),
2161 (VEXTUHLX (LI8 10), $S)>;
2162 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 6)))),
2163 (VEXTUHLX (LI8 12), $S)>;
2164 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 7)))),
2165 (VEXTUHLX (LI8 14), $S)>;
2167 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, i64:$Idx)))),
2168 (VEXTUWLX (RLWINM8 $Idx, 2, 28, 29), $S)>;
2169 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 0)))),
2170 (VEXTUWLX (LI8 0), $S)>;
2172 // For extracting BE word 1, MFVSRWZ is better than VEXTUWLX
2173 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 1)))),
2174 (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
2175 (i32 VectorExtractions.LE_WORD_2), sub_32)>;
2176 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 2)))),
2177 (VEXTUWLX (LI8 8), $S)>;
2178 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 3)))),
2179 (VEXTUWLX (LI8 12), $S)>;
2181 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, i64:$Idx)))),
2182 (EXTSW (VEXTUWLX (RLWINM8 $Idx, 2, 28, 29), $S))>;
2183 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 0)))),
2184 (EXTSW (VEXTUWLX (LI8 0), $S))>;
2185 // For extracting BE word 1, MFVSRWZ is better than VEXTUWLX
2186 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 1)))),
2187 (EXTSW (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
2188 (i32 VectorExtractions.LE_WORD_2), sub_32))>;
2189 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 2)))),
2190 (EXTSW (VEXTUWLX (LI8 8), $S))>;
2191 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 3)))),
2192 (EXTSW (VEXTUWLX (LI8 12), $S))>;
2194 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
2195 (i32 (EXTRACT_SUBREG (VEXTUBLX $Idx, $S), sub_32))>;
2196 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
2197 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 0), $S), sub_32))>;
2198 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
2199 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 1), $S), sub_32))>;
2200 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
2201 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 2), $S), sub_32))>;
2202 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
2203 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 3), $S), sub_32))>;
2204 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
2205 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 4), $S), sub_32))>;
2206 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
2207 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 5), $S), sub_32))>;
2208 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
2209 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 6), $S), sub_32))>;
2210 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
2211 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 7), $S), sub_32))>;
2212 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
2213 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 8), $S), sub_32))>;
2214 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
2215 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 9), $S), sub_32))>;
2216 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
2217 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 10), $S), sub_32))>;
2218 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
2219 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 11), $S), sub_32))>;
2220 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
2221 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 12), $S), sub_32))>;
2222 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
2223 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 13), $S), sub_32))>;
2224 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
2225 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 14), $S), sub_32))>;
2226 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
2227 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 15), $S), sub_32))>;
2229 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
2230 (i32 (EXTRACT_SUBREG (VEXTUHLX
2231 (RLWINM8 $Idx, 1, 28, 30), $S), sub_32))>;
2232 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
2233 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 0), $S), sub_32))>;
2234 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
2235 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 2), $S), sub_32))>;
2236 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
2237 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 4), $S), sub_32))>;
2238 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
2239 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 6), $S), sub_32))>;
2240 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
2241 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 8), $S), sub_32))>;
2242 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
2243 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 10), $S), sub_32))>;
2244 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2245 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 12), $S), sub_32))>;
2246 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2247 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 14), $S), sub_32))>;
2249 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
2250 (i32 (EXTRACT_SUBREG (VEXTUWLX
2251 (RLWINM8 $Idx, 2, 28, 29), $S), sub_32))>;
2252 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
2253 (i32 (EXTRACT_SUBREG (VEXTUWLX (LI8 0), $S), sub_32))>;
2254 // For extracting BE word 1, MFVSRWZ is better than VEXTUWLX
2255 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
2256 (i32 VectorExtractions.LE_WORD_2)>;
2257 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
2258 (i32 (EXTRACT_SUBREG (VEXTUWLX (LI8 8), $S), sub_32))>;
2259 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
2260 (i32 (EXTRACT_SUBREG (VEXTUWLX (LI8 12), $S), sub_32))>;
2263 let Predicates = [IsLittleEndian, HasDirectMove] in {
2264 // v16i8 scalar <-> vector conversions (LE)
2265 def : Pat<(v16i8 (scalar_to_vector i32:$A)),
2266 (v16i8 (COPY_TO_REGCLASS MovesToVSR.LE_WORD_0, VSRC))>;
2267 def : Pat<(v8i16 (scalar_to_vector i32:$A)),
2268 (v8i16 (COPY_TO_REGCLASS MovesToVSR.LE_WORD_0, VSRC))>;
2269 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
2270 (v4i32 MovesToVSR.LE_WORD_0)>;
2271 def : Pat<(v2i64 (scalar_to_vector i64:$A)),
2272 (v2i64 MovesToVSR.LE_DWORD_0)>;
2273 // v2i64 scalar <-> vector conversions (LE)
2274 def : Pat<(i64 (vector_extract v2i64:$S, 0)),
2275 (i64 VectorExtractions.LE_DWORD_0)>;
2276 def : Pat<(i64 (vector_extract v2i64:$S, 1)),
2277 (i64 VectorExtractions.LE_DWORD_1)>;
2278 def : Pat<(i64 (vector_extract v2i64:$S, i64:$Idx)),
2279 (i64 VectorExtractions.LE_VARIABLE_DWORD)>;
2280 } // IsLittleEndian, HasDirectMove
2282 let Predicates = [IsLittleEndian, HasDirectMove, NoP9Altivec] in {
2283 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
2284 (i32 VectorExtractions.LE_BYTE_0)>;
2285 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
2286 (i32 VectorExtractions.LE_BYTE_1)>;
2287 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
2288 (i32 VectorExtractions.LE_BYTE_2)>;
2289 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
2290 (i32 VectorExtractions.LE_BYTE_3)>;
2291 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
2292 (i32 VectorExtractions.LE_BYTE_4)>;
2293 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
2294 (i32 VectorExtractions.LE_BYTE_5)>;
2295 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
2296 (i32 VectorExtractions.LE_BYTE_6)>;
2297 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
2298 (i32 VectorExtractions.LE_BYTE_7)>;
2299 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
2300 (i32 VectorExtractions.LE_BYTE_8)>;
2301 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
2302 (i32 VectorExtractions.LE_BYTE_9)>;
2303 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
2304 (i32 VectorExtractions.LE_BYTE_10)>;
2305 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
2306 (i32 VectorExtractions.LE_BYTE_11)>;
2307 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
2308 (i32 VectorExtractions.LE_BYTE_12)>;
2309 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
2310 (i32 VectorExtractions.LE_BYTE_13)>;
2311 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
2312 (i32 VectorExtractions.LE_BYTE_14)>;
2313 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
2314 (i32 VectorExtractions.LE_BYTE_15)>;
2315 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
2316 (i32 VectorExtractions.LE_VARIABLE_BYTE)>;
2318 // v8i16 scalar <-> vector conversions (LE)
2319 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
2320 (i32 VectorExtractions.LE_HALF_0)>;
2321 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
2322 (i32 VectorExtractions.LE_HALF_1)>;
2323 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
2324 (i32 VectorExtractions.LE_HALF_2)>;
2325 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
2326 (i32 VectorExtractions.LE_HALF_3)>;
2327 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
2328 (i32 VectorExtractions.LE_HALF_4)>;
2329 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
2330 (i32 VectorExtractions.LE_HALF_5)>;
2331 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2332 (i32 VectorExtractions.LE_HALF_6)>;
2333 def : Pat<(i32 (vector_extract v8i16:$S, 7)),
2334 (i32 VectorExtractions.LE_HALF_7)>;
2335 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
2336 (i32 VectorExtractions.LE_VARIABLE_HALF)>;
2338 // v4i32 scalar <-> vector conversions (LE)
2339 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
2340 (i32 VectorExtractions.LE_WORD_0)>;
2341 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
2342 (i32 VectorExtractions.LE_WORD_1)>;
2343 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
2344 (i32 VectorExtractions.LE_WORD_2)>;
2345 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
2346 (i32 VectorExtractions.LE_WORD_3)>;
2347 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
2348 (i32 VectorExtractions.LE_VARIABLE_WORD)>;
2349 } // IsLittleEndian, HasDirectMove, NoP9Altivec
2351 let Predicates = [HasDirectMove, HasVSX] in {
2352 // bitconvert f32 -> i32
2353 // (convert to 32-bit fp single, shift right 1 word, move to GPR)
2354 def : Pat<(i32 (bitconvert f32:$S)),
2355 (i32 (MFVSRWZ (EXTRACT_SUBREG
2356 (XXSLDWI (XSCVDPSPN $S), (XSCVDPSPN $S), 3),
2358 // bitconvert i32 -> f32
2359 // (move to FPR, shift left 1 word, convert to 64-bit fp single)
2360 def : Pat<(f32 (bitconvert i32:$A)),
2362 (XXSLDWI MovesToVSR.LE_WORD_1, MovesToVSR.LE_WORD_1, 1)))>;
2364 // bitconvert f64 -> i64
2365 // (move to GPR, nothing else needed)
2366 def : Pat<(i64 (bitconvert f64:$S)),
2369 // bitconvert i64 -> f64
2370 // (move to FPR, nothing else needed)
2371 def : Pat<(f64 (bitconvert i64:$S)),
2375 // Materialize a zero-vector of long long
2376 def : Pat<(v2i64 immAllZerosV),
2381 dag F32_TO_BE_WORD1 = (v4f32 (XXSLDWI (XSCVDPSPN $B), (XSCVDPSPN $B), 3));
2382 dag I32_TO_BE_WORD1 = (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC);
2385 // The following VSX instructions were introduced in Power ISA 3.0
2386 def HasP9Vector : Predicate<"PPCSubTarget->hasP9Vector()">;
2387 let AddedComplexity = 400, Predicates = [HasP9Vector] in {
2389 // [PO VRT XO VRB XO /]
2390 class X_VT5_XO5_VB5<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2392 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vrrc:$vT), (ins vrrc:$vB),
2393 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
2395 // [PO VRT XO VRB XO RO], Round to Odd version of [PO VRT XO VRB XO /]
2396 class X_VT5_XO5_VB5_Ro<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2398 : X_VT5_XO5_VB5<opcode, xo2, xo, opc, pattern>, isDOT;
2400 // [PO VRT XO VRB XO /], but the VRB is only used the left 64 bits (or less),
2401 // So we use different operand class for VRB
2402 class X_VT5_XO5_VB5_TyVB<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2403 RegisterOperand vbtype, list<dag> pattern>
2404 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vrrc:$vT), (ins vbtype:$vB),
2405 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
2407 // [PO VRT XO VRB XO /]
2408 class X_VT5_XO5_VB5_VSFR<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2410 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vfrc:$vT), (ins vrrc:$vB),
2411 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
2413 // [PO VRT XO VRB XO RO], Round to Odd version of [PO VRT XO VRB XO /]
2414 class X_VT5_XO5_VB5_VSFR_Ro<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2416 : X_VT5_XO5_VB5_VSFR<opcode, xo2, xo, opc, pattern>, isDOT;
2418 let UseVSXReg = 1 in {
2419 // [PO T XO B XO BX /]
2420 class XX2_RT5_XO5_XB6<bits<6> opcode, bits<5> xo2, bits<9> xo, string opc,
2422 : XX2_RD5_XO5_RS6<opcode, xo2, xo, (outs g8rc:$rT), (ins vsfrc:$XB),
2423 !strconcat(opc, " $rT, $XB"), IIC_VecFP, pattern>;
2425 // [PO T XO B XO BX TX]
2426 class XX2_XT6_XO5_XB6<bits<6> opcode, bits<5> xo2, bits<9> xo, string opc,
2427 RegisterOperand vtype, list<dag> pattern>
2428 : XX2_RD6_XO5_RS6<opcode, xo2, xo, (outs vtype:$XT), (ins vtype:$XB),
2429 !strconcat(opc, " $XT, $XB"), IIC_VecFP, pattern>;
2431 // [PO T A B XO AX BX TX], src and dest register use different operand class
2432 class XX3_XT5_XA5_XB5<bits<6> opcode, bits<8> xo, string opc,
2433 RegisterOperand xty, RegisterOperand aty, RegisterOperand bty,
2434 InstrItinClass itin, list<dag> pattern>
2435 : XX3Form<opcode, xo, (outs xty:$XT), (ins aty:$XA, bty:$XB),
2436 !strconcat(opc, " $XT, $XA, $XB"), itin, pattern>;
2439 // [PO VRT VRA VRB XO /]
2440 class X_VT5_VA5_VB5<bits<6> opcode, bits<10> xo, string opc,
2442 : XForm_1<opcode, xo, (outs vrrc:$vT), (ins vrrc:$vA, vrrc:$vB),
2443 !strconcat(opc, " $vT, $vA, $vB"), IIC_VecFP, pattern>;
2445 // [PO VRT VRA VRB XO RO], Round to Odd version of [PO VRT VRA VRB XO /]
2446 class X_VT5_VA5_VB5_Ro<bits<6> opcode, bits<10> xo, string opc,
2448 : X_VT5_VA5_VB5<opcode, xo, opc, pattern>, isDOT;
2450 // [PO VRT VRA VRB XO /]
2451 class X_VT5_VA5_VB5_FMA<bits<6> opcode, bits<10> xo, string opc,
2453 : XForm_1<opcode, xo, (outs vrrc:$vT), (ins vrrc:$vTi, vrrc:$vA, vrrc:$vB),
2454 !strconcat(opc, " $vT, $vA, $vB"), IIC_VecFP, pattern>,
2455 RegConstraint<"$vTi = $vT">, NoEncode<"$vTi">;
2457 // [PO VRT VRA VRB XO RO], Round to Odd version of [PO VRT VRA VRB XO /]
2458 class X_VT5_VA5_VB5_FMA_Ro<bits<6> opcode, bits<10> xo, string opc,
2460 : X_VT5_VA5_VB5_FMA<opcode, xo, opc, pattern>, isDOT;
2462 //===--------------------------------------------------------------------===//
2463 // Quad-Precision Scalar Move Instructions:
2466 def XSCPSGNQP : X_VT5_VA5_VB5<63, 100, "xscpsgnqp",
2468 (fcopysign f128:$vB, f128:$vA))]>;
2470 // Absolute/Negative-Absolute/Negate
2471 def XSABSQP : X_VT5_XO5_VB5<63, 0, 804, "xsabsqp",
2472 [(set f128:$vT, (fabs f128:$vB))]>;
2473 def XSNABSQP : X_VT5_XO5_VB5<63, 8, 804, "xsnabsqp",
2474 [(set f128:$vT, (fneg (fabs f128:$vB)))]>;
2475 def XSNEGQP : X_VT5_XO5_VB5<63, 16, 804, "xsnegqp",
2476 [(set f128:$vT, (fneg f128:$vB))]>;
2478 //===--------------------------------------------------------------------===//
2479 // Quad-Precision Scalar Floating-Point Arithmetic Instructions:
2481 // Add/Divide/Multiply/Subtract
2482 let isCommutable = 1 in {
2483 def XSADDQP : X_VT5_VA5_VB5 <63, 4, "xsaddqp",
2484 [(set f128:$vT, (fadd f128:$vA, f128:$vB))]>;
2485 def XSADDQPO : X_VT5_VA5_VB5_Ro<63, 4, "xsaddqpo",
2487 (int_ppc_addf128_round_to_odd
2488 f128:$vA, f128:$vB))]>;
2489 def XSMULQP : X_VT5_VA5_VB5 <63, 36, "xsmulqp",
2490 [(set f128:$vT, (fmul f128:$vA, f128:$vB))]>;
2491 def XSMULQPO : X_VT5_VA5_VB5_Ro<63, 36, "xsmulqpo",
2493 (int_ppc_mulf128_round_to_odd
2494 f128:$vA, f128:$vB))]>;
2497 def XSSUBQP : X_VT5_VA5_VB5 <63, 516, "xssubqp" ,
2498 [(set f128:$vT, (fsub f128:$vA, f128:$vB))]>;
2499 def XSSUBQPO : X_VT5_VA5_VB5_Ro<63, 516, "xssubqpo",
2501 (int_ppc_subf128_round_to_odd
2502 f128:$vA, f128:$vB))]>;
2503 def XSDIVQP : X_VT5_VA5_VB5 <63, 548, "xsdivqp",
2504 [(set f128:$vT, (fdiv f128:$vA, f128:$vB))]>;
2505 def XSDIVQPO : X_VT5_VA5_VB5_Ro<63, 548, "xsdivqpo",
2507 (int_ppc_divf128_round_to_odd
2508 f128:$vA, f128:$vB))]>;
2511 def XSSQRTQP : X_VT5_XO5_VB5 <63, 27, 804, "xssqrtqp",
2512 [(set f128:$vT, (fsqrt f128:$vB))]>;
2513 def XSSQRTQPO : X_VT5_XO5_VB5_Ro<63, 27, 804, "xssqrtqpo",
2515 (int_ppc_sqrtf128_round_to_odd f128:$vB))]>;
2517 // (Negative) Multiply-{Add/Subtract}
2518 def XSMADDQP : X_VT5_VA5_VB5_FMA <63, 388, "xsmaddqp",
2520 (fma f128:$vA, f128:$vB,
2523 def XSMADDQPO : X_VT5_VA5_VB5_FMA_Ro<63, 388, "xsmaddqpo",
2525 (int_ppc_fmaf128_round_to_odd
2526 f128:$vA,f128:$vB,f128:$vTi))]>;
2528 def XSMSUBQP : X_VT5_VA5_VB5_FMA <63, 420, "xsmsubqp" ,
2530 (fma f128:$vA, f128:$vB,
2531 (fneg f128:$vTi)))]>;
2532 def XSMSUBQPO : X_VT5_VA5_VB5_FMA_Ro<63, 420, "xsmsubqpo" ,
2534 (int_ppc_fmaf128_round_to_odd
2535 f128:$vA, f128:$vB, (fneg f128:$vTi)))]>;
2536 def XSNMADDQP : X_VT5_VA5_VB5_FMA <63, 452, "xsnmaddqp",
2538 (fneg (fma f128:$vA, f128:$vB,
2540 def XSNMADDQPO: X_VT5_VA5_VB5_FMA_Ro<63, 452, "xsnmaddqpo",
2542 (fneg (int_ppc_fmaf128_round_to_odd
2543 f128:$vA, f128:$vB, f128:$vTi)))]>;
2544 def XSNMSUBQP : X_VT5_VA5_VB5_FMA <63, 484, "xsnmsubqp",
2546 (fneg (fma f128:$vA, f128:$vB,
2547 (fneg f128:$vTi))))]>;
2548 def XSNMSUBQPO: X_VT5_VA5_VB5_FMA_Ro<63, 484, "xsnmsubqpo",
2550 (fneg (int_ppc_fmaf128_round_to_odd
2551 f128:$vA, f128:$vB, (fneg f128:$vTi))))]>;
2553 // Additional fnmsub patterns: -a*c + b == -(a*c - b)
2554 def : Pat<(fma (fneg f128:$A), f128:$C, f128:$B), (XSNMSUBQP $B, $C, $A)>;
2555 def : Pat<(fma f128:$A, (fneg f128:$C), f128:$B), (XSNMSUBQP $B, $C, $A)>;
2557 //===--------------------------------------------------------------------===//
2558 // Quad/Double-Precision Compare Instructions:
2560 // [PO BF // VRA VRB XO /]
2561 class X_BF3_VA5_VB5<bits<6> opcode, bits<10> xo, string opc,
2563 : XForm_17<opcode, xo, (outs crrc:$crD), (ins vrrc:$VA, vrrc:$VB),
2564 !strconcat(opc, " $crD, $VA, $VB"), IIC_FPCompare> {
2565 let Pattern = pattern;
2568 // QP Compare Ordered/Unordered
2569 def XSCMPOQP : X_BF3_VA5_VB5<63, 132, "xscmpoqp", []>;
2570 def XSCMPUQP : X_BF3_VA5_VB5<63, 644, "xscmpuqp", []>;
2572 // DP/QP Compare Exponents
2573 def XSCMPEXPDP : XX3Form_1<60, 59,
2574 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
2575 "xscmpexpdp $crD, $XA, $XB", IIC_FPCompare, []>,
2577 def XSCMPEXPQP : X_BF3_VA5_VB5<63, 164, "xscmpexpqp", []>;
2579 // DP Compare ==, >=, >, !=
2580 // Use vsrc for XT, because the entire register of XT is set.
2581 // XT.dword[1] = 0x0000_0000_0000_0000
2582 def XSCMPEQDP : XX3_XT5_XA5_XB5<60, 3, "xscmpeqdp", vsrc, vsfrc, vsfrc,
2584 def XSCMPGEDP : XX3_XT5_XA5_XB5<60, 19, "xscmpgedp", vsrc, vsfrc, vsfrc,
2586 def XSCMPGTDP : XX3_XT5_XA5_XB5<60, 11, "xscmpgtdp", vsrc, vsfrc, vsfrc,
2589 //===--------------------------------------------------------------------===//
2590 // Quad-Precision Floating-Point Conversion Instructions:
2593 def XSCVDPQP : X_VT5_XO5_VB5_TyVB<63, 22, 836, "xscvdpqp", vfrc,
2594 [(set f128:$vT, (fpextend f64:$vB))]>;
2596 // Round & Convert QP -> DP (dword[1] is set to zero)
2597 def XSCVQPDP : X_VT5_XO5_VB5_VSFR<63, 20, 836, "xscvqpdp" , []>;
2598 def XSCVQPDPO : X_VT5_XO5_VB5_VSFR_Ro<63, 20, 836, "xscvqpdpo",
2600 (int_ppc_truncf128_round_to_odd
2603 // Truncate & Convert QP -> (Un)Signed (D)Word (dword[1] is set to zero)
2604 def XSCVQPSDZ : X_VT5_XO5_VB5<63, 25, 836, "xscvqpsdz", []>;
2605 def XSCVQPSWZ : X_VT5_XO5_VB5<63, 9, 836, "xscvqpswz", []>;
2606 def XSCVQPUDZ : X_VT5_XO5_VB5<63, 17, 836, "xscvqpudz", []>;
2607 def XSCVQPUWZ : X_VT5_XO5_VB5<63, 1, 836, "xscvqpuwz", []>;
2609 // Convert (Un)Signed DWord -> QP.
2610 def XSCVSDQP : X_VT5_XO5_VB5_TyVB<63, 10, 836, "xscvsdqp", vfrc, []>;
2611 def : Pat<(f128 (sint_to_fp i64:$src)),
2612 (f128 (XSCVSDQP (COPY_TO_REGCLASS $src, VFRC)))>;
2613 def : Pat<(f128 (sint_to_fp (i64 (PPCmfvsr f64:$src)))),
2614 (f128 (XSCVSDQP $src))>;
2615 def : Pat<(f128 (sint_to_fp (i32 (PPCmfvsr f64:$src)))),
2616 (f128 (XSCVSDQP (VEXTSW2Ds $src)))>;
2618 def XSCVUDQP : X_VT5_XO5_VB5_TyVB<63, 2, 836, "xscvudqp", vfrc, []>;
2619 def : Pat<(f128 (uint_to_fp i64:$src)),
2620 (f128 (XSCVUDQP (COPY_TO_REGCLASS $src, VFRC)))>;
2621 def : Pat<(f128 (uint_to_fp (i64 (PPCmfvsr f64:$src)))),
2622 (f128 (XSCVUDQP $src))>;
2624 // Convert (Un)Signed Word -> QP.
2625 def : Pat<(f128 (sint_to_fp i32:$src)),
2626 (f128 (XSCVSDQP (MTVSRWA $src)))>;
2627 def : Pat<(f128 (sint_to_fp (i32 (load xoaddr:$src)))),
2628 (f128 (XSCVSDQP (LIWAX xoaddr:$src)))>;
2629 def : Pat<(f128 (uint_to_fp i32:$src)),
2630 (f128 (XSCVUDQP (MTVSRWZ $src)))>;
2631 def : Pat<(f128 (uint_to_fp (i32 (load xoaddr:$src)))),
2632 (f128 (XSCVUDQP (LIWZX xoaddr:$src)))>;
2634 let UseVSXReg = 1 in {
2635 //===--------------------------------------------------------------------===//
2636 // Round to Floating-Point Integer Instructions
2638 // (Round &) Convert DP <-> HP
2639 // Note! xscvdphp's src and dest register both use the left 64 bits, so we use
2640 // vsfrc for src and dest register. xscvhpdp's src only use the left 16 bits,
2641 // but we still use vsfrc for it.
2642 def XSCVDPHP : XX2_XT6_XO5_XB6<60, 17, 347, "xscvdphp", vsfrc, []>;
2643 def XSCVHPDP : XX2_XT6_XO5_XB6<60, 16, 347, "xscvhpdp", vsfrc, []>;
2646 def XVCVHPSP : XX2_XT6_XO5_XB6<60, 24, 475, "xvcvhpsp", vsrc, []>;
2647 def XVCVSPHP : XX2_XT6_XO5_XB6<60, 25, 475, "xvcvsphp", vsrc,
2649 (int_ppc_vsx_xvcvsphp v4f32:$XB))]>;
2653 // Pattern for matching Vector HP -> Vector SP intrinsic. Defined as a
2654 // separate pattern so that it can convert the input register class from
2655 // VRRC(v8i16) to VSRC.
2656 def : Pat<(v4f32 (int_ppc_vsx_xvcvhpsp v8i16:$A)),
2657 (v4f32 (XVCVHPSP (COPY_TO_REGCLASS $A, VSRC)))>;
2659 class Z23_VT5_R1_VB5_RMC2_EX1<bits<6> opcode, bits<8> xo, bit ex, string opc,
2661 : Z23Form_8<opcode, xo,
2662 (outs vrrc:$vT), (ins u1imm:$r, vrrc:$vB, u2imm:$rmc),
2663 !strconcat(opc, " $r, $vT, $vB, $rmc"), IIC_VecFP, pattern> {
2667 // Round to Quad-Precision Integer [with Inexact]
2668 def XSRQPI : Z23_VT5_R1_VB5_RMC2_EX1<63, 5, 0, "xsrqpi" , []>;
2669 def XSRQPIX : Z23_VT5_R1_VB5_RMC2_EX1<63, 5, 1, "xsrqpix", []>;
2671 // Use current rounding mode
2672 def : Pat<(f128 (fnearbyint f128:$vB)), (f128 (XSRQPI 0, $vB, 3))>;
2673 // Round to nearest, ties away from zero
2674 def : Pat<(f128 (fround f128:$vB)), (f128 (XSRQPI 0, $vB, 0))>;
2675 // Round towards Zero
2676 def : Pat<(f128 (ftrunc f128:$vB)), (f128 (XSRQPI 1, $vB, 1))>;
2677 // Round towards +Inf
2678 def : Pat<(f128 (fceil f128:$vB)), (f128 (XSRQPI 1, $vB, 2))>;
2679 // Round towards -Inf
2680 def : Pat<(f128 (ffloor f128:$vB)), (f128 (XSRQPI 1, $vB, 3))>;
2682 // Use current rounding mode, [with Inexact]
2683 def : Pat<(f128 (frint f128:$vB)), (f128 (XSRQPIX 0, $vB, 3))>;
2685 // Round Quad-Precision to Double-Extended Precision (fp80)
2686 def XSRQPXP : Z23_VT5_R1_VB5_RMC2_EX1<63, 37, 0, "xsrqpxp", []>;
2688 //===--------------------------------------------------------------------===//
2689 // Insert/Extract Instructions
2691 // Insert Exponent DP/QP
2692 // XT NOTE: XT.dword[1] = 0xUUUU_UUUU_UUUU_UUUU
2693 def XSIEXPDP : XX1Form <60, 918, (outs vsrc:$XT), (ins g8rc:$rA, g8rc:$rB),
2694 "xsiexpdp $XT, $rA, $rB", IIC_VecFP, []>, UseVSXReg;
2695 // vB NOTE: only vB.dword[0] is used, that's why we don't use
2696 // X_VT5_VA5_VB5 form
2697 def XSIEXPQP : XForm_18<63, 868, (outs vrrc:$vT), (ins vrrc:$vA, vsfrc:$vB),
2698 "xsiexpqp $vT, $vA, $vB", IIC_VecFP, []>;
2700 def : Pat<(f128 (int_ppc_scalar_insert_exp_qp f128:$vA, i64:$vB)),
2701 (f128 (XSIEXPQP $vA, (MTVSRD $vB)))>;
2703 // Extract Exponent/Significand DP/QP
2704 def XSXEXPDP : XX2_RT5_XO5_XB6<60, 0, 347, "xsxexpdp", []>;
2705 def XSXSIGDP : XX2_RT5_XO5_XB6<60, 1, 347, "xsxsigdp", []>;
2707 def XSXEXPQP : X_VT5_XO5_VB5 <63, 2, 804, "xsxexpqp", []>;
2708 def XSXSIGQP : X_VT5_XO5_VB5 <63, 18, 804, "xsxsigqp", []>;
2710 def : Pat<(i64 (int_ppc_scalar_extract_expq f128:$vA)),
2711 (i64 (MFVSRD (EXTRACT_SUBREG
2712 (v2i64 (XSXEXPQP $vA)), sub_64)))>;
2714 // Vector Insert Word
2715 let UseVSXReg = 1 in {
2716 // XB NOTE: Only XB.dword[1] is used, but we use vsrc on XB.
2718 XX2_RD6_UIM5_RS6<60, 181, (outs vsrc:$XT),
2719 (ins vsrc:$XTi, vsrc:$XB, u4imm:$UIM),
2720 "xxinsertw $XT, $XB, $UIM", IIC_VecFP,
2721 [(set v4i32:$XT, (PPCvecinsert v4i32:$XTi, v4i32:$XB,
2722 imm32SExt16:$UIM))]>,
2723 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">;
2725 // Vector Extract Unsigned Word
2726 def XXEXTRACTUW : XX2_RD6_UIM5_RS6<60, 165,
2727 (outs vsfrc:$XT), (ins vsrc:$XB, u4imm:$UIMM),
2728 "xxextractuw $XT, $XB, $UIMM", IIC_VecFP, []>;
2731 // Vector Insert Exponent DP/SP
2732 def XVIEXPDP : XX3_XT5_XA5_XB5<60, 248, "xviexpdp", vsrc, vsrc, vsrc,
2733 IIC_VecFP, [(set v2f64: $XT,(int_ppc_vsx_xviexpdp v2i64:$XA, v2i64:$XB))]>;
2734 def XVIEXPSP : XX3_XT5_XA5_XB5<60, 216, "xviexpsp", vsrc, vsrc, vsrc,
2735 IIC_VecFP, [(set v4f32: $XT,(int_ppc_vsx_xviexpsp v4i32:$XA, v4i32:$XB))]>;
2737 // Vector Extract Exponent/Significand DP/SP
2738 def XVXEXPDP : XX2_XT6_XO5_XB6<60, 0, 475, "xvxexpdp", vsrc,
2740 (int_ppc_vsx_xvxexpdp v2f64:$XB))]>;
2741 def XVXEXPSP : XX2_XT6_XO5_XB6<60, 8, 475, "xvxexpsp", vsrc,
2743 (int_ppc_vsx_xvxexpsp v4f32:$XB))]>;
2744 def XVXSIGDP : XX2_XT6_XO5_XB6<60, 1, 475, "xvxsigdp", vsrc,
2746 (int_ppc_vsx_xvxsigdp v2f64:$XB))]>;
2747 def XVXSIGSP : XX2_XT6_XO5_XB6<60, 9, 475, "xvxsigsp", vsrc,
2749 (int_ppc_vsx_xvxsigsp v4f32:$XB))]>;
2751 let AddedComplexity = 400, Predicates = [HasP9Vector] in {
2752 // Extra patterns expanding to vector Extract Word/Insert Word
2753 def : Pat<(v4i32 (int_ppc_vsx_xxinsertw v4i32:$A, v2i64:$B, imm:$IMM)),
2754 (v4i32 (XXINSERTW $A, $B, imm:$IMM))>;
2755 def : Pat<(v2i64 (int_ppc_vsx_xxextractuw v2i64:$A, imm:$IMM)),
2756 (v2i64 (COPY_TO_REGCLASS (XXEXTRACTUW $A, imm:$IMM), VSRC))>;
2757 } // AddedComplexity = 400, HasP9Vector
2759 //===--------------------------------------------------------------------===//
2761 // Test Data Class SP/DP/QP
2762 let UseVSXReg = 1 in {
2763 def XSTSTDCSP : XX2_BF3_DCMX7_RS6<60, 298,
2764 (outs crrc:$BF), (ins u7imm:$DCMX, vsfrc:$XB),
2765 "xststdcsp $BF, $XB, $DCMX", IIC_VecFP, []>;
2766 def XSTSTDCDP : XX2_BF3_DCMX7_RS6<60, 362,
2767 (outs crrc:$BF), (ins u7imm:$DCMX, vsfrc:$XB),
2768 "xststdcdp $BF, $XB, $DCMX", IIC_VecFP, []>;
2770 def XSTSTDCQP : X_BF3_DCMX7_RS5 <63, 708,
2771 (outs crrc:$BF), (ins u7imm:$DCMX, vrrc:$vB),
2772 "xststdcqp $BF, $vB, $DCMX", IIC_VecFP, []>;
2774 // Vector Test Data Class SP/DP
2775 let UseVSXReg = 1 in {
2776 def XVTSTDCSP : XX2_RD6_DCMX7_RS6<60, 13, 5,
2777 (outs vsrc:$XT), (ins u7imm:$DCMX, vsrc:$XB),
2778 "xvtstdcsp $XT, $XB, $DCMX", IIC_VecFP,
2780 (int_ppc_vsx_xvtstdcsp v4f32:$XB, imm:$DCMX))]>;
2781 def XVTSTDCDP : XX2_RD6_DCMX7_RS6<60, 15, 5,
2782 (outs vsrc:$XT), (ins u7imm:$DCMX, vsrc:$XB),
2783 "xvtstdcdp $XT, $XB, $DCMX", IIC_VecFP,
2785 (int_ppc_vsx_xvtstdcdp v2f64:$XB, imm:$DCMX))]>;
2788 //===--------------------------------------------------------------------===//
2790 // Maximum/Minimum Type-C/Type-J DP
2791 // XT.dword[1] = 0xUUUU_UUUU_UUUU_UUUU, so we use vsrc for XT
2792 def XSMAXCDP : XX3_XT5_XA5_XB5<60, 128, "xsmaxcdp", vsrc, vsfrc, vsfrc,
2794 def XSMAXJDP : XX3_XT5_XA5_XB5<60, 144, "xsmaxjdp", vsrc, vsfrc, vsfrc,
2796 def XSMINCDP : XX3_XT5_XA5_XB5<60, 136, "xsmincdp", vsrc, vsfrc, vsfrc,
2798 def XSMINJDP : XX3_XT5_XA5_XB5<60, 152, "xsminjdp", vsrc, vsfrc, vsfrc,
2801 //===--------------------------------------------------------------------===//
2803 // Vector Byte-Reverse H/W/D/Q Word
2804 def XXBRH : XX2_XT6_XO5_XB6<60, 7, 475, "xxbrh", vsrc, []>;
2805 def XXBRW : XX2_XT6_XO5_XB6<60, 15, 475, "xxbrw", vsrc, []>;
2806 def XXBRD : XX2_XT6_XO5_XB6<60, 23, 475, "xxbrd", vsrc, []>;
2807 def XXBRQ : XX2_XT6_XO5_XB6<60, 31, 475, "xxbrq", vsrc, []>;
2810 def : Pat<(v8i16 (PPCxxreverse v8i16 :$A)),
2811 (v8i16 (COPY_TO_REGCLASS (XXBRH (COPY_TO_REGCLASS $A, VSRC)), VRRC))>;
2812 def : Pat<(v4i32 (PPCxxreverse v4i32 :$A)),
2813 (v4i32 (XXBRW $A))>;
2814 def : Pat<(v2i64 (PPCxxreverse v2i64 :$A)),
2815 (v2i64 (XXBRD $A))>;
2816 def : Pat<(v1i128 (PPCxxreverse v1i128 :$A)),
2817 (v1i128 (COPY_TO_REGCLASS (XXBRQ (COPY_TO_REGCLASS $A, VSRC)), VRRC))>;
2820 def XXPERM : XX3_XT5_XA5_XB5<60, 26, "xxperm" , vsrc, vsrc, vsrc,
2822 def XXPERMR : XX3_XT5_XA5_XB5<60, 58, "xxpermr", vsrc, vsrc, vsrc,
2825 // Vector Splat Immediate Byte
2826 def XXSPLTIB : X_RD6_IMM8<60, 360, (outs vsrc:$XT), (ins u8imm:$IMM8),
2827 "xxspltib $XT, $IMM8", IIC_VecPerm, []>, UseVSXReg;
2829 //===--------------------------------------------------------------------===//
2830 // Vector/Scalar Load/Store Instructions
2832 // When adding new D-Form loads/stores, be sure to update the ImmToIdxMap in
2833 // PPCRegisterInfo::PPCRegisterInfo and maybe save yourself some debugging.
2834 let mayLoad = 1, mayStore = 0 in {
2836 def LXV : DQ_RD6_RS5_DQ12<61, 1, (outs vsrc:$XT), (ins memrix16:$src),
2837 "lxv $XT, $src", IIC_LdStLFD, []>, UseVSXReg;
2839 def LXSD : DSForm_1<57, 2, (outs vfrc:$vD), (ins memrix:$src),
2840 "lxsd $vD, $src", IIC_LdStLFD, []>;
2841 // Load SP from src, convert it to DP, and place in dword[0]
2842 def LXSSP : DSForm_1<57, 3, (outs vfrc:$vD), (ins memrix:$src),
2843 "lxssp $vD, $src", IIC_LdStLFD, []>;
2845 // [PO T RA RB XO TX] almost equal to [PO S RA RB XO SX], but has different
2846 // "out" and "in" dag
2847 class X_XT6_RA5_RB5<bits<6> opcode, bits<10> xo, string opc,
2848 RegisterOperand vtype, list<dag> pattern>
2849 : XX1Form_memOp<opcode, xo, (outs vtype:$XT), (ins memrr:$src),
2850 !strconcat(opc, " $XT, $src"), IIC_LdStLFD, pattern>, UseVSXReg;
2852 // Load as Integer Byte/Halfword & Zero Indexed
2853 def LXSIBZX : X_XT6_RA5_RB5<31, 781, "lxsibzx", vsfrc,
2854 [(set f64:$XT, (PPClxsizx xoaddr:$src, 1))]>;
2855 def LXSIHZX : X_XT6_RA5_RB5<31, 813, "lxsihzx", vsfrc,
2856 [(set f64:$XT, (PPClxsizx xoaddr:$src, 2))]>;
2858 // Load Vector Halfword*8/Byte*16 Indexed
2859 def LXVH8X : X_XT6_RA5_RB5<31, 812, "lxvh8x" , vsrc, []>;
2860 def LXVB16X : X_XT6_RA5_RB5<31, 876, "lxvb16x", vsrc, []>;
2862 // Load Vector Indexed
2863 def LXVX : X_XT6_RA5_RB5<31, 268, "lxvx" , vsrc,
2864 [(set v2f64:$XT, (load xaddr:$src))]>;
2865 // Load Vector (Left-justified) with Length
2866 def LXVL : XX1Form_memOp<31, 269, (outs vsrc:$XT), (ins memr:$src, g8rc:$rB),
2867 "lxvl $XT, $src, $rB", IIC_LdStLoad,
2868 [(set v4i32:$XT, (int_ppc_vsx_lxvl addr:$src, i64:$rB))]>,
2870 def LXVLL : XX1Form_memOp<31,301, (outs vsrc:$XT), (ins memr:$src, g8rc:$rB),
2871 "lxvll $XT, $src, $rB", IIC_LdStLoad,
2872 [(set v4i32:$XT, (int_ppc_vsx_lxvll addr:$src, i64:$rB))]>,
2875 // Load Vector Word & Splat Indexed
2876 def LXVWSX : X_XT6_RA5_RB5<31, 364, "lxvwsx" , vsrc, []>;
2879 // When adding new D-Form loads/stores, be sure to update the ImmToIdxMap in
2880 // PPCRegisterInfo::PPCRegisterInfo and maybe save yourself some debugging.
2881 let mayStore = 1, mayLoad = 0 in {
2883 def STXV : DQ_RD6_RS5_DQ12<61, 5, (outs), (ins vsrc:$XT, memrix16:$dst),
2884 "stxv $XT, $dst", IIC_LdStSTFD, []>, UseVSXReg;
2886 def STXSD : DSForm_1<61, 2, (outs), (ins vfrc:$vS, memrix:$dst),
2887 "stxsd $vS, $dst", IIC_LdStSTFD, []>;
2888 // Convert DP of dword[0] to SP, and Store to dst
2889 def STXSSP : DSForm_1<61, 3, (outs), (ins vfrc:$vS, memrix:$dst),
2890 "stxssp $vS, $dst", IIC_LdStSTFD, []>;
2892 // [PO S RA RB XO SX]
2893 class X_XS6_RA5_RB5<bits<6> opcode, bits<10> xo, string opc,
2894 RegisterOperand vtype, list<dag> pattern>
2895 : XX1Form_memOp<opcode, xo, (outs), (ins vtype:$XT, memrr:$dst),
2896 !strconcat(opc, " $XT, $dst"), IIC_LdStSTFD, pattern>, UseVSXReg;
2898 // Store as Integer Byte/Halfword Indexed
2899 def STXSIBX : X_XS6_RA5_RB5<31, 909, "stxsibx" , vsfrc,
2900 [(PPCstxsix f64:$XT, xoaddr:$dst, 1)]>;
2901 def STXSIHX : X_XS6_RA5_RB5<31, 941, "stxsihx" , vsfrc,
2902 [(PPCstxsix f64:$XT, xoaddr:$dst, 2)]>;
2903 let isCodeGenOnly = 1 in {
2904 def STXSIBXv : X_XS6_RA5_RB5<31, 909, "stxsibx" , vrrc, []>;
2905 def STXSIHXv : X_XS6_RA5_RB5<31, 941, "stxsihx" , vrrc, []>;
2908 // Store Vector Halfword*8/Byte*16 Indexed
2909 def STXVH8X : X_XS6_RA5_RB5<31, 940, "stxvh8x" , vsrc, []>;
2910 def STXVB16X : X_XS6_RA5_RB5<31, 1004, "stxvb16x", vsrc, []>;
2912 // Store Vector Indexed
2913 def STXVX : X_XS6_RA5_RB5<31, 396, "stxvx" , vsrc,
2914 [(store v2f64:$XT, xaddr:$dst)]>;
2916 // Store Vector (Left-justified) with Length
2917 def STXVL : XX1Form_memOp<31, 397, (outs),
2918 (ins vsrc:$XT, memr:$dst, g8rc:$rB),
2919 "stxvl $XT, $dst, $rB", IIC_LdStLoad,
2920 [(int_ppc_vsx_stxvl v4i32:$XT, addr:$dst,
2923 def STXVLL : XX1Form_memOp<31, 429, (outs),
2924 (ins vsrc:$XT, memr:$dst, g8rc:$rB),
2925 "stxvll $XT, $dst, $rB", IIC_LdStLoad,
2926 [(int_ppc_vsx_stxvll v4i32:$XT, addr:$dst,
2931 let Predicates = [IsLittleEndian] in {
2932 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 0)))))),
2933 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 3))))>;
2934 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 1)))))),
2935 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 2))))>;
2936 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 2)))))),
2937 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 1))))>;
2938 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 3)))))),
2939 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 0))))>;
2940 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 0)))))),
2941 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 3)), VSFRC))>;
2942 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 1)))))),
2943 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 2)), VSFRC))>;
2944 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 2)))))),
2945 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 1)), VSFRC))>;
2946 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 3)))))),
2947 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 0)), VSFRC))>;
2950 let Predicates = [IsBigEndian] in {
2951 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 0)))))),
2952 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 0))))>;
2953 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 1)))))),
2954 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 1))))>;
2955 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 2)))))),
2956 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 2))))>;
2957 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 3)))))),
2958 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 3))))>;
2959 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 0)))))),
2960 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 0)), VSFRC))>;
2961 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 1)))))),
2962 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 1)), VSFRC))>;
2963 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 2)))))),
2964 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 2)), VSFRC))>;
2965 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 3)))))),
2966 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 3)), VSFRC))>;
2969 // Alternate patterns for PPCmtvsrz where the output is v8i16 or v16i8 instead
2971 def : Pat<(v8i16 (PPCmtvsrz i32:$A)),
2972 (v8i16 (SUBREG_TO_REG (i64 1), (MTVSRWZ $A), sub_64))>;
2973 def : Pat<(v16i8 (PPCmtvsrz i32:$A)),
2974 (v16i8 (SUBREG_TO_REG (i64 1), (MTVSRWZ $A), sub_64))>;
2976 // Patterns for which instructions from ISA 3.0 are a better match
2977 let Predicates = [IsLittleEndian, HasP9Vector] in {
2978 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 0)))))),
2979 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 12)))>;
2980 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 1)))))),
2981 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 8)))>;
2982 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 2)))))),
2983 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 4)))>;
2984 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 3)))))),
2985 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 0)))>;
2986 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 0)))))),
2987 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 12)))>;
2988 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 1)))))),
2989 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 8)))>;
2990 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 2)))))),
2991 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 4)))>;
2992 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 3)))))),
2993 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 0)))>;
2994 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 0)),
2995 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 12))>;
2996 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 1)),
2997 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 8))>;
2998 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 2)),
2999 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 4))>;
3000 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 3)),
3001 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 0))>;
3002 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 0)),
3003 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 12))>;
3004 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 1)),
3005 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 8))>;
3006 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 2)),
3007 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 4))>;
3008 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 3)),
3009 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 0))>;
3010 } // IsLittleEndian, HasP9Vector
3012 let Predicates = [IsBigEndian, HasP9Vector] in {
3013 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 0)))))),
3014 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 0)))>;
3015 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 1)))))),
3016 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 4)))>;
3017 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 2)))))),
3018 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 8)))>;
3019 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 3)))))),
3020 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 12)))>;
3021 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 0)))))),
3022 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 0)))>;
3023 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 1)))))),
3024 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 4)))>;
3025 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 2)))))),
3026 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 8)))>;
3027 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 3)))))),
3028 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 12)))>;
3029 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 0)),
3030 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 0))>;
3031 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 1)),
3032 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 4))>;
3033 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 2)),
3034 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 8))>;
3035 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 3)),
3036 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 12))>;
3037 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 0)),
3038 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 0))>;
3039 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 1)),
3040 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 4))>;
3041 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 2)),
3042 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 8))>;
3043 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 3)),
3044 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 12))>;
3045 } // IsLittleEndian, HasP9Vector
3047 // D-Form Load/Store
3048 def : Pat<(v4i32 (quadwOffsetLoad iqaddr:$src)), (LXV memrix16:$src)>;
3049 def : Pat<(v4f32 (quadwOffsetLoad iqaddr:$src)), (LXV memrix16:$src)>;
3050 def : Pat<(v2i64 (quadwOffsetLoad iqaddr:$src)), (LXV memrix16:$src)>;
3051 def : Pat<(v2f64 (quadwOffsetLoad iqaddr:$src)), (LXV memrix16:$src)>;
3052 def : Pat<(f128 (quadwOffsetLoad iqaddr:$src)),
3053 (COPY_TO_REGCLASS (LXV memrix16:$src), VRRC)>;
3054 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x iqaddr:$src)), (LXV memrix16:$src)>;
3055 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x iqaddr:$src)), (LXV memrix16:$src)>;
3057 def : Pat<(quadwOffsetStore v4f32:$rS, iqaddr:$dst), (STXV $rS, memrix16:$dst)>;
3058 def : Pat<(quadwOffsetStore v4i32:$rS, iqaddr:$dst), (STXV $rS, memrix16:$dst)>;
3059 def : Pat<(quadwOffsetStore v2f64:$rS, iqaddr:$dst), (STXV $rS, memrix16:$dst)>;
3060 def : Pat<(quadwOffsetStore f128:$rS, iqaddr:$dst),
3061 (STXV (COPY_TO_REGCLASS $rS, VSRC), memrix16:$dst)>;
3062 def : Pat<(quadwOffsetStore v2i64:$rS, iqaddr:$dst), (STXV $rS, memrix16:$dst)>;
3063 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, iqaddr:$dst),
3064 (STXV $rS, memrix16:$dst)>;
3065 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, iqaddr:$dst),
3066 (STXV $rS, memrix16:$dst)>;
3069 def : Pat<(v2f64 (nonQuadwOffsetLoad xoaddr:$src)), (LXVX xoaddr:$src)>;
3070 def : Pat<(v2i64 (nonQuadwOffsetLoad xoaddr:$src)), (LXVX xoaddr:$src)>;
3071 def : Pat<(v4f32 (nonQuadwOffsetLoad xoaddr:$src)), (LXVX xoaddr:$src)>;
3072 def : Pat<(v4i32 (nonQuadwOffsetLoad xoaddr:$src)), (LXVX xoaddr:$src)>;
3073 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x xoaddr:$src)), (LXVX xoaddr:$src)>;
3074 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x xoaddr:$src)), (LXVX xoaddr:$src)>;
3075 def : Pat<(f128 (nonQuadwOffsetLoad xoaddr:$src)),
3076 (COPY_TO_REGCLASS (LXVX xoaddr:$src), VRRC)>;
3077 def : Pat<(nonQuadwOffsetStore f128:$rS, xoaddr:$dst),
3078 (STXVX (COPY_TO_REGCLASS $rS, VSRC), xoaddr:$dst)>;
3079 def : Pat<(nonQuadwOffsetStore v2f64:$rS, xoaddr:$dst),
3080 (STXVX $rS, xoaddr:$dst)>;
3081 def : Pat<(nonQuadwOffsetStore v2i64:$rS, xoaddr:$dst),
3082 (STXVX $rS, xoaddr:$dst)>;
3083 def : Pat<(nonQuadwOffsetStore v4f32:$rS, xoaddr:$dst),
3084 (STXVX $rS, xoaddr:$dst)>;
3085 def : Pat<(nonQuadwOffsetStore v4i32:$rS, xoaddr:$dst),
3086 (STXVX $rS, xoaddr:$dst)>;
3087 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, xoaddr:$dst),
3088 (STXVX $rS, xoaddr:$dst)>;
3089 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, xoaddr:$dst),
3090 (STXVX $rS, xoaddr:$dst)>;
3092 let AddedComplexity = 400 in {
3093 // LIWAX - This instruction is used for sign extending i32 -> i64.
3094 // LIWZX - This instruction will be emitted for i32, f32, and when
3095 // zero-extending i32 to i64 (zext i32 -> i64).
3096 let Predicates = [IsLittleEndian] in {
3098 def : Pat<(v2i64 (scalar_to_vector (i64 (sextloadi32 xoaddr:$src)))),
3100 (COPY_TO_REGCLASS (LIWAX xoaddr:$src), VSRC), 2))>;
3102 def : Pat<(v2i64 (scalar_to_vector (i64 (zextloadi32 xoaddr:$src)))),
3104 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 2))>;
3106 def : Pat<(v4i32 (scalar_to_vector (i32 (load xoaddr:$src)))),
3108 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 2))>;
3110 def : Pat<(v4f32 (scalar_to_vector (f32 (load xoaddr:$src)))),
3112 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 2))>;
3115 let Predicates = [IsBigEndian] in {
3116 def : Pat<(v2i64 (scalar_to_vector (i64 (sextloadi32 xoaddr:$src)))),
3117 (v2i64 (COPY_TO_REGCLASS (LIWAX xoaddr:$src), VSRC))>;
3119 def : Pat<(v2i64 (scalar_to_vector (i64 (zextloadi32 xoaddr:$src)))),
3120 (v2i64 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC))>;
3122 def : Pat<(v4i32 (scalar_to_vector (i32 (load xoaddr:$src)))),
3124 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 1))>;
3126 def : Pat<(v4f32 (scalar_to_vector (f32 (load xoaddr:$src)))),
3128 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 1))>;
3133 // Build vectors from i8 loads
3134 def : Pat<(v16i8 (scalar_to_vector ScalarLoads.Li8)),
3135 (v16i8 (VSPLTBs 7, (LXSIBZX xoaddr:$src)))>;
3136 def : Pat<(v8i16 (scalar_to_vector ScalarLoads.ZELi8)),
3137 (v8i16 (VSPLTHs 3, (LXSIBZX xoaddr:$src)))>;
3138 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.ZELi8)),
3139 (v4i32 (XXSPLTWs (LXSIBZX xoaddr:$src), 1))>;
3140 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.ZELi8i64)),
3141 (v2i64 (XXPERMDIs (LXSIBZX xoaddr:$src), 0))>;
3142 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.SELi8)),
3143 (v4i32 (XXSPLTWs (VEXTSB2Ws (LXSIBZX xoaddr:$src)), 1))>;
3144 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.SELi8i64)),
3145 (v2i64 (XXPERMDIs (VEXTSB2Ds (LXSIBZX xoaddr:$src)), 0))>;
3147 // Build vectors from i16 loads
3148 def : Pat<(v8i16 (scalar_to_vector ScalarLoads.Li16)),
3149 (v8i16 (VSPLTHs 3, (LXSIHZX xoaddr:$src)))>;
3150 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.ZELi16)),
3151 (v4i32 (XXSPLTWs (LXSIHZX xoaddr:$src), 1))>;
3152 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.ZELi16i64)),
3153 (v2i64 (XXPERMDIs (LXSIHZX xoaddr:$src), 0))>;
3154 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.SELi16)),
3155 (v4i32 (XXSPLTWs (VEXTSH2Ws (LXSIHZX xoaddr:$src)), 1))>;
3156 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.SELi16i64)),
3157 (v2i64 (XXPERMDIs (VEXTSH2Ds (LXSIHZX xoaddr:$src)), 0))>;
3159 let Predicates = [IsBigEndian, HasP9Vector] in {
3160 // Scalar stores of i8
3161 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 0)), xoaddr:$dst),
3162 (STXSIBXv (v16i8 (VSLDOI $S, $S, 9)), xoaddr:$dst)>;
3163 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 1)), xoaddr:$dst),
3164 (STXSIBXv (v16i8 (VSLDOI $S, $S, 10)), xoaddr:$dst)>;
3165 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 2)), xoaddr:$dst),
3166 (STXSIBXv (v16i8 (VSLDOI $S, $S, 11)), xoaddr:$dst)>;
3167 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 3)), xoaddr:$dst),
3168 (STXSIBXv (v16i8 (VSLDOI $S, $S, 12)), xoaddr:$dst)>;
3169 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 4)), xoaddr:$dst),
3170 (STXSIBXv (v16i8 (VSLDOI $S, $S, 13)), xoaddr:$dst)>;
3171 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 5)), xoaddr:$dst),
3172 (STXSIBXv (v16i8 (VSLDOI $S, $S, 14)), xoaddr:$dst)>;
3173 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 6)), xoaddr:$dst),
3174 (STXSIBXv (v16i8 (VSLDOI $S, $S, 15)), xoaddr:$dst)>;
3175 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 7)), xoaddr:$dst),
3176 (STXSIBXv $S, xoaddr:$dst)>;
3177 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 8)), xoaddr:$dst),
3178 (STXSIBXv (v16i8 (VSLDOI $S, $S, 1)), xoaddr:$dst)>;
3179 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 9)), xoaddr:$dst),
3180 (STXSIBXv (v16i8 (VSLDOI $S, $S, 2)), xoaddr:$dst)>;
3181 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 10)), xoaddr:$dst),
3182 (STXSIBXv (v16i8 (VSLDOI $S, $S, 3)), xoaddr:$dst)>;
3183 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 11)), xoaddr:$dst),
3184 (STXSIBXv (v16i8 (VSLDOI $S, $S, 4)), xoaddr:$dst)>;
3185 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 12)), xoaddr:$dst),
3186 (STXSIBXv (v16i8 (VSLDOI $S, $S, 5)), xoaddr:$dst)>;
3187 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 13)), xoaddr:$dst),
3188 (STXSIBXv (v16i8 (VSLDOI $S, $S, 6)), xoaddr:$dst)>;
3189 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 14)), xoaddr:$dst),
3190 (STXSIBXv (v16i8 (VSLDOI $S, $S, 7)), xoaddr:$dst)>;
3191 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 15)), xoaddr:$dst),
3192 (STXSIBXv (v16i8 (VSLDOI $S, $S, 8)), xoaddr:$dst)>;
3194 // Scalar stores of i16
3195 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 0)), xoaddr:$dst),
3196 (STXSIHXv (v16i8 (VSLDOI $S, $S, 10)), xoaddr:$dst)>;
3197 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 1)), xoaddr:$dst),
3198 (STXSIHXv (v16i8 (VSLDOI $S, $S, 12)), xoaddr:$dst)>;
3199 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 2)), xoaddr:$dst),
3200 (STXSIHXv (v16i8 (VSLDOI $S, $S, 14)), xoaddr:$dst)>;
3201 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 3)), xoaddr:$dst),
3202 (STXSIHXv $S, xoaddr:$dst)>;
3203 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 4)), xoaddr:$dst),
3204 (STXSIHXv (v16i8 (VSLDOI $S, $S, 2)), xoaddr:$dst)>;
3205 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 5)), xoaddr:$dst),
3206 (STXSIHXv (v16i8 (VSLDOI $S, $S, 4)), xoaddr:$dst)>;
3207 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 6)), xoaddr:$dst),
3208 (STXSIHXv (v16i8 (VSLDOI $S, $S, 6)), xoaddr:$dst)>;
3209 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 7)), xoaddr:$dst),
3210 (STXSIHXv (v16i8 (VSLDOI $S, $S, 8)), xoaddr:$dst)>;
3211 } // IsBigEndian, HasP9Vector
3213 let Predicates = [IsLittleEndian, HasP9Vector] in {
3214 // Scalar stores of i8
3215 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 0)), xoaddr:$dst),
3216 (STXSIBXv (v16i8 (VSLDOI $S, $S, 8)), xoaddr:$dst)>;
3217 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 1)), xoaddr:$dst),
3218 (STXSIBXv (v16i8 (VSLDOI $S, $S, 7)), xoaddr:$dst)>;
3219 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 2)), xoaddr:$dst),
3220 (STXSIBXv (v16i8 (VSLDOI $S, $S, 6)), xoaddr:$dst)>;
3221 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 3)), xoaddr:$dst),
3222 (STXSIBXv (v16i8 (VSLDOI $S, $S, 5)), xoaddr:$dst)>;
3223 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 4)), xoaddr:$dst),
3224 (STXSIBXv (v16i8 (VSLDOI $S, $S, 4)), xoaddr:$dst)>;
3225 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 5)), xoaddr:$dst),
3226 (STXSIBXv (v16i8 (VSLDOI $S, $S, 3)), xoaddr:$dst)>;
3227 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 6)), xoaddr:$dst),
3228 (STXSIBXv (v16i8 (VSLDOI $S, $S, 2)), xoaddr:$dst)>;
3229 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 7)), xoaddr:$dst),
3230 (STXSIBXv (v16i8 (VSLDOI $S, $S, 1)), xoaddr:$dst)>;
3231 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 8)), xoaddr:$dst),
3232 (STXSIBXv $S, xoaddr:$dst)>;
3233 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 9)), xoaddr:$dst),
3234 (STXSIBXv (v16i8 (VSLDOI $S, $S, 15)), xoaddr:$dst)>;
3235 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 10)), xoaddr:$dst),
3236 (STXSIBXv (v16i8 (VSLDOI $S, $S, 14)), xoaddr:$dst)>;
3237 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 11)), xoaddr:$dst),
3238 (STXSIBXv (v16i8 (VSLDOI $S, $S, 13)), xoaddr:$dst)>;
3239 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 12)), xoaddr:$dst),
3240 (STXSIBXv (v16i8 (VSLDOI $S, $S, 12)), xoaddr:$dst)>;
3241 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 13)), xoaddr:$dst),
3242 (STXSIBXv (v16i8 (VSLDOI $S, $S, 11)), xoaddr:$dst)>;
3243 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 14)), xoaddr:$dst),
3244 (STXSIBXv (v16i8 (VSLDOI $S, $S, 10)), xoaddr:$dst)>;
3245 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 15)), xoaddr:$dst),
3246 (STXSIBXv (v16i8 (VSLDOI $S, $S, 9)), xoaddr:$dst)>;
3248 // Scalar stores of i16
3249 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 0)), xoaddr:$dst),
3250 (STXSIHXv (v16i8 (VSLDOI $S, $S, 8)), xoaddr:$dst)>;
3251 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 1)), xoaddr:$dst),
3252 (STXSIHXv (v16i8 (VSLDOI $S, $S, 6)), xoaddr:$dst)>;
3253 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 2)), xoaddr:$dst),
3254 (STXSIHXv (v16i8 (VSLDOI $S, $S, 4)), xoaddr:$dst)>;
3255 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 3)), xoaddr:$dst),
3256 (STXSIHXv (v16i8 (VSLDOI $S, $S, 2)), xoaddr:$dst)>;
3257 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 4)), xoaddr:$dst),
3258 (STXSIHXv $S, xoaddr:$dst)>;
3259 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 5)), xoaddr:$dst),
3260 (STXSIHXv (v16i8 (VSLDOI $S, $S, 14)), xoaddr:$dst)>;
3261 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 6)), xoaddr:$dst),
3262 (STXSIHXv (v16i8 (VSLDOI $S, $S, 12)), xoaddr:$dst)>;
3263 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 7)), xoaddr:$dst),
3264 (STXSIHXv (v16i8 (VSLDOI $S, $S, 10)), xoaddr:$dst)>;
3265 } // IsLittleEndian, HasP9Vector
3268 // Vector sign extensions
3269 def : Pat<(f64 (PPCVexts f64:$A, 1)),
3270 (f64 (COPY_TO_REGCLASS (VEXTSB2Ds $A), VSFRC))>;
3271 def : Pat<(f64 (PPCVexts f64:$A, 2)),
3272 (f64 (COPY_TO_REGCLASS (VEXTSH2Ds $A), VSFRC))>;
3274 def DFLOADf32 : PPCPostRAExpPseudo<(outs vssrc:$XT), (ins memrix:$src),
3276 [(set f32:$XT, (load ixaddr:$src))]>;
3277 def DFLOADf64 : PPCPostRAExpPseudo<(outs vsfrc:$XT), (ins memrix:$src),
3279 [(set f64:$XT, (load ixaddr:$src))]>;
3280 def DFSTOREf32 : PPCPostRAExpPseudo<(outs), (ins vssrc:$XT, memrix:$dst),
3282 [(store f32:$XT, ixaddr:$dst)]>;
3283 def DFSTOREf64 : PPCPostRAExpPseudo<(outs), (ins vsfrc:$XT, memrix:$dst),
3285 [(store f64:$XT, ixaddr:$dst)]>;
3287 def : Pat<(f64 (extloadf32 ixaddr:$src)),
3288 (COPY_TO_REGCLASS (DFLOADf32 ixaddr:$src), VSFRC)>;
3289 def : Pat<(f32 (fpround (f64 (extloadf32 ixaddr:$src)))),
3290 (f32 (DFLOADf32 ixaddr:$src))>;
3293 let AddedComplexity = 400 in {
3294 // The following pseudoinstructions are used to ensure the utilization
3295 // of all 64 VSX registers.
3296 let Predicates = [IsLittleEndian, HasP9Vector] in {
3297 def : Pat<(v2i64 (scalar_to_vector (i64 (load ixaddr:$src)))),
3299 (COPY_TO_REGCLASS (DFLOADf64 ixaddr:$src), VSRC), 2))>;
3300 def : Pat<(v2i64 (scalar_to_vector (i64 (load xaddr:$src)))),
3302 (COPY_TO_REGCLASS (XFLOADf64 xaddr:$src), VSRC), 2))>;
3304 def : Pat<(v2f64 (scalar_to_vector (f64 (load ixaddr:$src)))),
3306 (COPY_TO_REGCLASS (DFLOADf64 ixaddr:$src), VSRC), 2))>;
3307 def : Pat<(v2f64 (scalar_to_vector (f64 (load xaddr:$src)))),
3309 (COPY_TO_REGCLASS (XFLOADf64 xaddr:$src), VSRC), 2))>;
3312 let Predicates = [IsBigEndian, HasP9Vector] in {
3313 def : Pat<(v2i64 (scalar_to_vector (i64 (load ixaddr:$src)))),
3314 (v2i64 (COPY_TO_REGCLASS (DFLOADf64 ixaddr:$src), VSRC))>;
3315 def : Pat<(v2i64 (scalar_to_vector (i64 (load xaddr:$src)))),
3316 (v2i64 (COPY_TO_REGCLASS (XFLOADf64 xaddr:$src), VSRC))>;
3318 def : Pat<(v2f64 (scalar_to_vector (f64 (load ixaddr:$src)))),
3319 (v2f64 (COPY_TO_REGCLASS (DFLOADf64 ixaddr:$src), VSRC))>;
3320 def : Pat<(v2f64 (scalar_to_vector (f64 (load xaddr:$src)))),
3321 (v2f64 (COPY_TO_REGCLASS (XFLOADf64 xaddr:$src), VSRC))>;
3325 let Predicates = [IsBigEndian, HasP9Vector] in {
3327 // (Un)Signed DWord vector extract -> QP
3328 def : Pat<(f128 (sint_to_fp (i64 (extractelt v2i64:$src, 0)))),
3329 (f128 (XSCVSDQP (COPY_TO_REGCLASS $src, VFRC)))>;
3330 def : Pat<(f128 (sint_to_fp (i64 (extractelt v2i64:$src, 1)))),
3332 (EXTRACT_SUBREG (XXPERMDI $src, $src, 3), sub_64)))>;
3333 def : Pat<(f128 (uint_to_fp (i64 (extractelt v2i64:$src, 0)))),
3334 (f128 (XSCVUDQP (COPY_TO_REGCLASS $src, VFRC)))>;
3335 def : Pat<(f128 (uint_to_fp (i64 (extractelt v2i64:$src, 1)))),
3337 (EXTRACT_SUBREG (XXPERMDI $src, $src, 3), sub_64)))>;
3339 // (Un)Signed Word vector extract -> QP
3340 def : Pat<(f128 (sint_to_fp (i32 (extractelt v4i32:$src, 1)))),
3341 (f128 (XSCVSDQP (EXTRACT_SUBREG (VEXTSW2D $src), sub_64)))>;
3342 foreach Idx = [0,2,3] in {
3343 def : Pat<(f128 (sint_to_fp (i32 (extractelt v4i32:$src, Idx)))),
3344 (f128 (XSCVSDQP (EXTRACT_SUBREG
3345 (VEXTSW2D (VSPLTW Idx, $src)), sub_64)))>;
3347 foreach Idx = 0-3 in {
3348 def : Pat<(f128 (uint_to_fp (i32 (extractelt v4i32:$src, Idx)))),
3349 (f128 (XSCVUDQP (XXEXTRACTUW $src, !shl(Idx, 2))))>;
3352 // (Un)Signed HWord vector extract -> QP
3353 foreach Idx = 0-7 in {
3354 def : Pat<(f128 (sint_to_fp
3356 (vector_extract v8i16:$src, Idx), i16)))),
3357 (f128 (XSCVSDQP (EXTRACT_SUBREG
3358 (VEXTSH2D (VEXTRACTUH !add(Idx, Idx), $src)),
3360 // The SDAG adds the `and` since an `i16` is being extracted as an `i32`.
3361 def : Pat<(f128 (uint_to_fp
3362 (and (i32 (vector_extract v8i16:$src, Idx)), 65535))),
3363 (f128 (XSCVUDQP (EXTRACT_SUBREG
3364 (VEXTRACTUH !add(Idx, Idx), $src), sub_64)))>;
3367 // (Un)Signed Byte vector extract -> QP
3368 foreach Idx = 0-15 in {
3369 def : Pat<(f128 (sint_to_fp
3370 (i32 (sext_inreg (vector_extract v16i8:$src, Idx),
3372 (f128 (XSCVSDQP (EXTRACT_SUBREG
3373 (VEXTSB2D (VEXTRACTUB Idx, $src)), sub_64)))>;
3374 def : Pat<(f128 (uint_to_fp
3375 (and (i32 (vector_extract v16i8:$src, Idx)), 255))),
3377 (EXTRACT_SUBREG (VEXTRACTUB Idx, $src), sub_64)))>;
3380 // Unsiged int in vsx register -> QP
3381 def : Pat<(f128 (uint_to_fp (i32 (PPCmfvsr f64:$src)))),
3383 (XXEXTRACTUW (SUBREG_TO_REG (i64 1), $src, sub_64), 4)))>;
3384 } // IsBigEndian, HasP9Vector
3386 let Predicates = [IsLittleEndian, HasP9Vector] in {
3388 // (Un)Signed DWord vector extract -> QP
3389 def : Pat<(f128 (sint_to_fp (i64 (extractelt v2i64:$src, 0)))),
3391 (EXTRACT_SUBREG (XXPERMDI $src, $src, 3), sub_64)))>;
3392 def : Pat<(f128 (sint_to_fp (i64 (extractelt v2i64:$src, 1)))),
3393 (f128 (XSCVSDQP (COPY_TO_REGCLASS $src, VFRC)))>;
3394 def : Pat<(f128 (uint_to_fp (i64 (extractelt v2i64:$src, 0)))),
3396 (EXTRACT_SUBREG (XXPERMDI $src, $src, 3), sub_64)))>;
3397 def : Pat<(f128 (uint_to_fp (i64 (extractelt v2i64:$src, 1)))),
3398 (f128 (XSCVUDQP (COPY_TO_REGCLASS $src, VFRC)))>;
3400 // (Un)Signed Word vector extract -> QP
3401 foreach Idx = [[0,3],[1,2],[3,0]] in {
3402 def : Pat<(f128 (sint_to_fp (i32 (extractelt v4i32:$src, !head(Idx))))),
3403 (f128 (XSCVSDQP (EXTRACT_SUBREG
3404 (VEXTSW2D (VSPLTW !head(!tail(Idx)), $src)),
3407 def : Pat<(f128 (sint_to_fp (i32 (extractelt v4i32:$src, 2)))),
3408 (f128 (XSCVSDQP (EXTRACT_SUBREG (VEXTSW2D $src), sub_64)))>;
3410 foreach Idx = [[0,12],[1,8],[2,4],[3,0]] in {
3411 def : Pat<(f128 (uint_to_fp (i32 (extractelt v4i32:$src, !head(Idx))))),
3412 (f128 (XSCVUDQP (XXEXTRACTUW $src, !head(!tail(Idx)))))>;
3415 // (Un)Signed HWord vector extract -> QP
3416 // The Nested foreach lists identifies the vector element and corresponding
3417 // register byte location.
3418 foreach Idx = [[0,14],[1,12],[2,10],[3,8],[4,6],[5,4],[6,2],[7,0]] in {
3419 def : Pat<(f128 (sint_to_fp
3421 (vector_extract v8i16:$src, !head(Idx)), i16)))),
3423 (EXTRACT_SUBREG (VEXTSH2D
3424 (VEXTRACTUH !head(!tail(Idx)), $src)),
3426 def : Pat<(f128 (uint_to_fp
3427 (and (i32 (vector_extract v8i16:$src, !head(Idx))),
3429 (f128 (XSCVUDQP (EXTRACT_SUBREG
3430 (VEXTRACTUH !head(!tail(Idx)), $src), sub_64)))>;
3433 // (Un)Signed Byte vector extract -> QP
3434 foreach Idx = [[0,15],[1,14],[2,13],[3,12],[4,11],[5,10],[6,9],[7,8],[8,7],
3435 [9,6],[10,5],[11,4],[12,3],[13,2],[14,1],[15,0]] in {
3436 def : Pat<(f128 (sint_to_fp
3438 (vector_extract v16i8:$src, !head(Idx)), i8)))),
3441 (VEXTSB2D (VEXTRACTUB !head(!tail(Idx)), $src)),
3443 def : Pat<(f128 (uint_to_fp
3444 (and (i32 (vector_extract v16i8:$src, !head(Idx))),
3448 (VEXTRACTUB !head(!tail(Idx)), $src), sub_64)))>;
3451 // Unsiged int in vsx register -> QP
3452 def : Pat<(f128 (uint_to_fp (i32 (PPCmfvsr f64:$src)))),
3454 (XXEXTRACTUW (SUBREG_TO_REG (i64 1), $src, sub_64), 8)))>;
3455 } // IsLittleEndian, HasP9Vector
3457 // Convert (Un)Signed DWord in memory -> QP
3458 def : Pat<(f128 (sint_to_fp (i64 (load xaddr:$src)))),
3459 (f128 (XSCVSDQP (LXSDX xaddr:$src)))>;
3460 def : Pat<(f128 (sint_to_fp (i64 (load ixaddr:$src)))),
3461 (f128 (XSCVSDQP (LXSD ixaddr:$src)))>;
3462 def : Pat<(f128 (uint_to_fp (i64 (load xaddr:$src)))),
3463 (f128 (XSCVUDQP (LXSDX xaddr:$src)))>;
3464 def : Pat<(f128 (uint_to_fp (i64 (load ixaddr:$src)))),
3465 (f128 (XSCVUDQP (LXSD ixaddr:$src)))>;
3467 // Convert Unsigned HWord in memory -> QP
3468 def : Pat<(f128 (uint_to_fp ScalarLoads.ZELi16)),
3469 (f128 (XSCVUDQP (LXSIHZX xaddr:$src)))>;
3471 // Convert Unsigned Byte in memory -> QP
3472 def : Pat<(f128 (uint_to_fp ScalarLoads.ZELi8)),
3473 (f128 (XSCVUDQP (LXSIBZX xoaddr:$src)))>;
3475 // Truncate & Convert QP -> (Un)Signed (D)Word.
3476 def : Pat<(i64 (fp_to_sint f128:$src)), (i64 (MFVRD (XSCVQPSDZ $src)))>;
3477 def : Pat<(i64 (fp_to_uint f128:$src)), (i64 (MFVRD (XSCVQPUDZ $src)))>;
3478 def : Pat<(i32 (fp_to_sint f128:$src)),
3479 (i32 (MFVSRWZ (COPY_TO_REGCLASS (XSCVQPSWZ $src), VFRC)))>;
3480 def : Pat<(i32 (fp_to_uint f128:$src)),
3481 (i32 (MFVSRWZ (COPY_TO_REGCLASS (XSCVQPUWZ $src), VFRC)))>;
3483 // Instructions for store(fptosi).
3484 // The 8-byte version is repeated here due to availability of D-Form STXSD.
3485 def : Pat<(PPCstore_scal_int_from_vsr
3486 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), xaddr:$dst, 8),
3487 (STXSDX (COPY_TO_REGCLASS (XSCVQPSDZ f128:$src), VFRC),
3489 def : Pat<(PPCstore_scal_int_from_vsr
3490 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), ixaddr:$dst, 8),
3491 (STXSD (COPY_TO_REGCLASS (XSCVQPSDZ f128:$src), VFRC),
3493 def : Pat<(PPCstore_scal_int_from_vsr
3494 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), xoaddr:$dst, 4),
3495 (STXSIWX (COPY_TO_REGCLASS (XSCVQPSWZ $src), VFRC), xoaddr:$dst)>;
3496 def : Pat<(PPCstore_scal_int_from_vsr
3497 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), xoaddr:$dst, 2),
3498 (STXSIHX (COPY_TO_REGCLASS (XSCVQPSWZ $src), VFRC), xoaddr:$dst)>;
3499 def : Pat<(PPCstore_scal_int_from_vsr
3500 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), xoaddr:$dst, 1),
3501 (STXSIBX (COPY_TO_REGCLASS (XSCVQPSWZ $src), VFRC), xoaddr:$dst)>;
3502 def : Pat<(PPCstore_scal_int_from_vsr
3503 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xaddr:$dst, 8),
3504 (STXSDX (XSCVDPSXDS f64:$src), xaddr:$dst)>;
3505 def : Pat<(PPCstore_scal_int_from_vsr
3506 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), ixaddr:$dst, 8),
3507 (STXSD (XSCVDPSXDS f64:$src), ixaddr:$dst)>;
3508 def : Pat<(PPCstore_scal_int_from_vsr
3509 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xoaddr:$dst, 2),
3510 (STXSIHX (XSCVDPSXWS f64:$src), xoaddr:$dst)>;
3511 def : Pat<(PPCstore_scal_int_from_vsr
3512 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xoaddr:$dst, 1),
3513 (STXSIBX (XSCVDPSXWS f64:$src), xoaddr:$dst)>;
3515 // Instructions for store(fptoui).
3516 def : Pat<(PPCstore_scal_int_from_vsr
3517 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), xaddr:$dst, 8),
3518 (STXSDX (COPY_TO_REGCLASS (XSCVQPUDZ f128:$src), VFRC),
3520 def : Pat<(PPCstore_scal_int_from_vsr
3521 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), ixaddr:$dst, 8),
3522 (STXSD (COPY_TO_REGCLASS (XSCVQPUDZ f128:$src), VFRC),
3524 def : Pat<(PPCstore_scal_int_from_vsr
3525 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), xoaddr:$dst, 4),
3526 (STXSIWX (COPY_TO_REGCLASS (XSCVQPUWZ $src), VFRC), xoaddr:$dst)>;
3527 def : Pat<(PPCstore_scal_int_from_vsr
3528 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), xoaddr:$dst, 2),
3529 (STXSIHX (COPY_TO_REGCLASS (XSCVQPUWZ $src), VFRC), xoaddr:$dst)>;
3530 def : Pat<(PPCstore_scal_int_from_vsr
3531 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), xoaddr:$dst, 1),
3532 (STXSIBX (COPY_TO_REGCLASS (XSCVQPUWZ $src), VFRC), xoaddr:$dst)>;
3533 def : Pat<(PPCstore_scal_int_from_vsr
3534 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xaddr:$dst, 8),
3535 (STXSDX (XSCVDPUXDS f64:$src), xaddr:$dst)>;
3536 def : Pat<(PPCstore_scal_int_from_vsr
3537 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), ixaddr:$dst, 8),
3538 (STXSD (XSCVDPUXDS f64:$src), ixaddr:$dst)>;
3539 def : Pat<(PPCstore_scal_int_from_vsr
3540 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xoaddr:$dst, 2),
3541 (STXSIHX (XSCVDPUXWS f64:$src), xoaddr:$dst)>;
3542 def : Pat<(PPCstore_scal_int_from_vsr
3543 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xoaddr:$dst, 1),
3544 (STXSIBX (XSCVDPUXWS f64:$src), xoaddr:$dst)>;
3546 // Round & Convert QP -> DP/SP
3547 def : Pat<(f64 (fpround f128:$src)), (f64 (XSCVQPDP $src))>;
3548 def : Pat<(f32 (fpround f128:$src)), (f32 (XSRSP (XSCVQPDPO $src)))>;
3551 def : Pat<(f128 (fpextend f32:$src)),
3552 (f128 (XSCVDPQP (COPY_TO_REGCLASS $src, VFRC)))>;
3554 } // end HasP9Vector, AddedComplexity
3556 let AddedComplexity = 400 in {
3557 let Predicates = [IsISA3_0, HasP9Vector, HasDirectMove, IsBigEndian] in {
3558 def : Pat<(f128 (PPCbuild_fp128 i64:$rB, i64:$rA)),
3559 (f128 (COPY_TO_REGCLASS (MTVSRDD $rB, $rA), VRRC))>;
3561 let Predicates = [IsISA3_0, HasP9Vector, HasDirectMove, IsLittleEndian] in {
3562 def : Pat<(f128 (PPCbuild_fp128 i64:$rA, i64:$rB)),
3563 (f128 (COPY_TO_REGCLASS (MTVSRDD $rB, $rA), VRRC))>;
3567 let Predicates = [HasP9Vector] in {
3568 let mayStore = 1 in {
3569 def SPILLTOVSR_STX : PseudoXFormMemOp<(outs),
3570 (ins spilltovsrrc:$XT, memrr:$dst),
3571 "#SPILLTOVSR_STX", []>;
3572 def SPILLTOVSR_ST : PPCPostRAExpPseudo<(outs), (ins spilltovsrrc:$XT, memrix:$dst),
3573 "#SPILLTOVSR_ST", []>;
3575 let mayLoad = 1 in {
3576 def SPILLTOVSR_LDX : PseudoXFormMemOp<(outs spilltovsrrc:$XT),
3578 "#SPILLTOVSR_LDX", []>;
3579 def SPILLTOVSR_LD : PPCPostRAExpPseudo<(outs spilltovsrrc:$XT), (ins memrix:$src),
3580 "#SPILLTOVSR_LD", []>;
3584 // Integer extend helper dags 32 -> 64
3586 dag A = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32);
3587 dag B = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $B, sub_32);
3588 dag C = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $C, sub_32);
3589 dag D = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $D, sub_32);
3593 dag A0 = (f32 (fpround (f64 (extractelt v2f64:$A, 0))));
3594 dag A1 = (f32 (fpround (f64 (extractelt v2f64:$A, 1))));
3595 dag B0 = (f32 (fpround (f64 (extractelt v2f64:$B, 0))));
3596 dag B1 = (f32 (fpround (f64 (extractelt v2f64:$B, 1))));
3600 dag A0S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$A, 0))))));
3601 dag A1S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$A, 1))))));
3602 dag B0S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$B, 0))))));
3603 dag B1S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$B, 1))))));
3604 dag A0U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$A, 0))))));
3605 dag A1U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$A, 1))))));
3606 dag B0U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$B, 0))))));
3607 dag B1U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$B, 1))))));
3611 dag LE_A0 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 0)), i8));
3612 dag LE_A1 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 4)), i8));
3613 dag LE_A2 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 8)), i8));
3614 dag LE_A3 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 12)), i8));
3615 dag BE_A0 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 3)), i8));
3616 dag BE_A1 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 7)), i8));
3617 dag BE_A2 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 11)), i8));
3618 dag BE_A3 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 15)), i8));
3622 dag LE_A0 = (i64 (sext_inreg
3623 (i64 (anyext (i32 (vector_extract v16i8:$A, 0)))), i8));
3624 dag LE_A1 = (i64 (sext_inreg
3625 (i64 (anyext (i32 (vector_extract v16i8:$A, 8)))), i8));
3626 dag BE_A0 = (i64 (sext_inreg
3627 (i64 (anyext (i32 (vector_extract v16i8:$A, 7)))), i8));
3628 dag BE_A1 = (i64 (sext_inreg
3629 (i64 (anyext (i32 (vector_extract v16i8:$A, 15)))), i8));
3633 dag LE_A0 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 0)), i16));
3634 dag LE_A1 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 2)), i16));
3635 dag LE_A2 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 4)), i16));
3636 dag LE_A3 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 6)), i16));
3637 dag BE_A0 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 1)), i16));
3638 dag BE_A1 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 3)), i16));
3639 dag BE_A2 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 5)), i16));
3640 dag BE_A3 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 7)), i16));
3644 dag LE_A0 = (i64 (sext_inreg
3645 (i64 (anyext (i32 (vector_extract v8i16:$A, 0)))), i16));
3646 dag LE_A1 = (i64 (sext_inreg
3647 (i64 (anyext (i32 (vector_extract v8i16:$A, 4)))), i16));
3648 dag BE_A0 = (i64 (sext_inreg
3649 (i64 (anyext (i32 (vector_extract v8i16:$A, 3)))), i16));
3650 dag BE_A1 = (i64 (sext_inreg
3651 (i64 (anyext (i32 (vector_extract v8i16:$A, 7)))), i16));
3655 dag LE_A0 = (i64 (sext (i32 (vector_extract v4i32:$A, 0))));
3656 dag LE_A1 = (i64 (sext (i32 (vector_extract v4i32:$A, 2))));
3657 dag BE_A0 = (i64 (sext (i32 (vector_extract v4i32:$A, 1))));
3658 dag BE_A1 = (i64 (sext (i32 (vector_extract v4i32:$A, 3))));
3662 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (extloadf32 xoaddr:$A)))));
3665 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (extloadf32 xoaddr:$A)))));
3668 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (extloadf32 xoaddr:$A)))));
3670 def FltToLongLoadP9 {
3671 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (extloadf32 ixaddr:$A)))));
3673 def FltToULongLoad {
3674 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (extloadf32 xoaddr:$A)))));
3676 def FltToULongLoadP9 {
3677 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (extloadf32 ixaddr:$A)))));
3680 dag A = (i64 (PPCmfvsr (f64 (PPCfctidz (fpextend f32:$A)))));
3683 dag A = (i64 (PPCmfvsr (f64 (PPCfctiduz (fpextend f32:$A)))));
3686 dag A = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$A))));
3687 dag B = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$B))));
3688 dag C = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$C))));
3689 dag D = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$D))));
3692 dag A = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$A))));
3693 dag B = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$B))));
3694 dag C = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$C))));
3695 dag D = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$D))));
3698 dag A = (i64 (PPCmfvsr (f64 (PPCfctidz f64:$A))));
3701 dag A = (i64 (PPCmfvsr (f64 (PPCfctiduz f64:$A))));
3704 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (load xoaddr:$A)))));
3706 def DblToIntLoadP9 {
3707 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (load ixaddr:$A)))));
3710 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (load xoaddr:$A)))));
3712 def DblToUIntLoadP9 {
3713 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (load ixaddr:$A)))));
3716 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (load xoaddr:$A)))));
3718 def DblToULongLoad {
3719 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (load xoaddr:$A)))));
3722 // FP merge dags (for f32 -> v4f32)
3724 dag AC = (XVCVDPSP (XXPERMDI (COPY_TO_REGCLASS $A, VSRC),
3725 (COPY_TO_REGCLASS $C, VSRC), 0));
3726 dag BD = (XVCVDPSP (XXPERMDI (COPY_TO_REGCLASS $B, VSRC),
3727 (COPY_TO_REGCLASS $D, VSRC), 0));
3728 dag ABhToFlt = (XVCVDPSP (XXPERMDI $A, $B, 0));
3729 dag ABlToFlt = (XVCVDPSP (XXPERMDI $A, $B, 3));
3730 dag BAhToFlt = (XVCVDPSP (XXPERMDI $B, $A, 0));
3731 dag BAlToFlt = (XVCVDPSP (XXPERMDI $B, $A, 3));
3734 // Word-element merge dags - conversions from f64 to i32 merged into vectors.
3736 // For big endian, we merge low and hi doublewords (A, B).
3737 dag A0B0 = (v2f64 (XXPERMDI v2f64:$A, v2f64:$B, 0));
3738 dag A1B1 = (v2f64 (XXPERMDI v2f64:$A, v2f64:$B, 3));
3739 dag CVA1B1S = (v4i32 (XVCVDPSXWS A1B1));
3740 dag CVA0B0S = (v4i32 (XVCVDPSXWS A0B0));
3741 dag CVA1B1U = (v4i32 (XVCVDPUXWS A1B1));
3742 dag CVA0B0U = (v4i32 (XVCVDPUXWS A0B0));
3744 // For little endian, we merge low and hi doublewords (B, A).
3745 dag B1A1 = (v2f64 (XXPERMDI v2f64:$B, v2f64:$A, 0));
3746 dag B0A0 = (v2f64 (XXPERMDI v2f64:$B, v2f64:$A, 3));
3747 dag CVB1A1S = (v4i32 (XVCVDPSXWS B1A1));
3748 dag CVB0A0S = (v4i32 (XVCVDPSXWS B0A0));
3749 dag CVB1A1U = (v4i32 (XVCVDPUXWS B1A1));
3750 dag CVB0A0U = (v4i32 (XVCVDPUXWS B0A0));
3752 // For big endian, we merge hi doublewords of (A, C) and (B, D), convert
3754 dag AC = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$A, VSRC),
3755 (COPY_TO_REGCLASS f64:$C, VSRC), 0));
3756 dag BD = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$B, VSRC),
3757 (COPY_TO_REGCLASS f64:$D, VSRC), 0));
3758 dag CVACS = (v4i32 (XVCVDPSXWS AC));
3759 dag CVBDS = (v4i32 (XVCVDPSXWS BD));
3760 dag CVACU = (v4i32 (XVCVDPUXWS AC));
3761 dag CVBDU = (v4i32 (XVCVDPUXWS BD));
3763 // For little endian, we merge hi doublewords of (D, B) and (C, A), convert
3765 dag DB = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$D, VSRC),
3766 (COPY_TO_REGCLASS f64:$B, VSRC), 0));
3767 dag CA = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$C, VSRC),
3768 (COPY_TO_REGCLASS f64:$A, VSRC), 0));
3769 dag CVDBS = (v4i32 (XVCVDPSXWS DB));
3770 dag CVCAS = (v4i32 (XVCVDPSXWS CA));
3771 dag CVDBU = (v4i32 (XVCVDPUXWS DB));
3772 dag CVCAU = (v4i32 (XVCVDPUXWS CA));
3775 // Patterns for BUILD_VECTOR nodes.
3776 let AddedComplexity = 400 in {
3778 let Predicates = [HasVSX] in {
3779 // Build vectors of floating point converted to i32.
3780 def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.A,
3781 DblToInt.A, DblToInt.A)),
3782 (v4i32 (XXSPLTW (COPY_TO_REGCLASS (XSCVDPSXWS $A), VSRC), 1))>;
3783 def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.A,
3784 DblToUInt.A, DblToUInt.A)),
3785 (v4i32 (XXSPLTW (COPY_TO_REGCLASS (XSCVDPUXWS $A), VSRC), 1))>;
3786 def : Pat<(v2i64 (build_vector DblToLong.A, DblToLong.A)),
3787 (v2i64 (XXPERMDI (COPY_TO_REGCLASS (XSCVDPSXDS $A), VSRC),
3788 (COPY_TO_REGCLASS (XSCVDPSXDS $A), VSRC), 0))>;
3789 def : Pat<(v2i64 (build_vector DblToULong.A, DblToULong.A)),
3790 (v2i64 (XXPERMDI (COPY_TO_REGCLASS (XSCVDPUXDS $A), VSRC),
3791 (COPY_TO_REGCLASS (XSCVDPUXDS $A), VSRC), 0))>;
3792 def : Pat<(v4i32 (scalar_to_vector FltToIntLoad.A)),
3793 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3794 (XSCVDPSXWSs (XFLOADf32 xoaddr:$A)), VSRC), 1))>;
3795 def : Pat<(v4i32 (scalar_to_vector FltToUIntLoad.A)),
3796 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3797 (XSCVDPUXWSs (XFLOADf32 xoaddr:$A)), VSRC), 1))>;
3798 def : Pat<(v4f32 (build_vector f32:$A, f32:$A, f32:$A, f32:$A)),
3799 (v4f32 (XXSPLTW (v4f32 (XSCVDPSPN $A)), 0))>;
3801 // Build vectors of floating point converted to i64.
3802 def : Pat<(v2i64 (build_vector FltToLong.A, FltToLong.A)),
3804 (COPY_TO_REGCLASS (XSCVDPSXDSs $A), VSFRC), 0))>;
3805 def : Pat<(v2i64 (build_vector FltToULong.A, FltToULong.A)),
3807 (COPY_TO_REGCLASS (XSCVDPUXDSs $A), VSFRC), 0))>;
3808 def : Pat<(v2i64 (scalar_to_vector DblToLongLoad.A)),
3809 (v2i64 (XVCVDPSXDS (LXVDSX xoaddr:$A)))>;
3810 def : Pat<(v2i64 (scalar_to_vector DblToULongLoad.A)),
3811 (v2i64 (XVCVDPUXDS (LXVDSX xoaddr:$A)))>;
3814 let Predicates = [HasVSX, NoP9Vector] in {
3815 // Load-and-splat with fp-to-int conversion (using X-Form VSX/FP loads).
3816 def : Pat<(v4i32 (scalar_to_vector DblToIntLoad.A)),
3817 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3818 (XSCVDPSXWS (XFLOADf64 xoaddr:$A)), VSRC), 1))>;
3819 def : Pat<(v4i32 (scalar_to_vector DblToUIntLoad.A)),
3820 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3821 (XSCVDPUXWS (XFLOADf64 xoaddr:$A)), VSRC), 1))>;
3822 def : Pat<(v2i64 (scalar_to_vector FltToLongLoad.A)),
3823 (v2i64 (XXPERMDIs (XSCVDPSXDS (COPY_TO_REGCLASS
3824 (XFLOADf32 xoaddr:$A), VSFRC)), 0))>;
3825 def : Pat<(v2i64 (scalar_to_vector FltToULongLoad.A)),
3826 (v2i64 (XXPERMDIs (XSCVDPUXDS (COPY_TO_REGCLASS
3827 (XFLOADf32 xoaddr:$A), VSFRC)), 0))>;
3830 let Predicates = [IsBigEndian, HasP8Vector] in {
3831 def : Pat<DWToSPExtractConv.BVU,
3832 (v4f32 (VPKUDUM (XXSLDWI (XVCVUXDSP $S1), (XVCVUXDSP $S1), 3),
3833 (XXSLDWI (XVCVUXDSP $S2), (XVCVUXDSP $S2), 3)))>;
3834 def : Pat<DWToSPExtractConv.BVS,
3835 (v4f32 (VPKUDUM (XXSLDWI (XVCVSXDSP $S1), (XVCVSXDSP $S1), 3),
3836 (XXSLDWI (XVCVSXDSP $S2), (XVCVSXDSP $S2), 3)))>;
3839 // Big endian, available on all targets with VSX
3840 let Predicates = [IsBigEndian, HasVSX] in {
3841 def : Pat<(v2f64 (build_vector f64:$A, f64:$B)),
3843 (COPY_TO_REGCLASS $A, VSRC),
3844 (COPY_TO_REGCLASS $B, VSRC), 0))>;
3846 def : Pat<(v4f32 (build_vector f32:$A, f32:$B, f32:$C, f32:$D)),
3847 (VMRGEW MrgFP.AC, MrgFP.BD)>;
3848 def : Pat<(v4f32 (build_vector DblToFlt.A0, DblToFlt.A1,
3849 DblToFlt.B0, DblToFlt.B1)),
3850 (v4f32 (VMRGEW MrgFP.ABhToFlt, MrgFP.ABlToFlt))>;
3852 // Convert 4 doubles to a vector of ints.
3853 def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.B,
3854 DblToInt.C, DblToInt.D)),
3855 (v4i32 (VMRGEW MrgWords.CVACS, MrgWords.CVBDS))>;
3856 def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.B,
3857 DblToUInt.C, DblToUInt.D)),
3858 (v4i32 (VMRGEW MrgWords.CVACU, MrgWords.CVBDU))>;
3859 def : Pat<(v4i32 (build_vector ExtDbl.A0S, ExtDbl.A1S,
3860 ExtDbl.B0S, ExtDbl.B1S)),
3861 (v4i32 (VMRGEW MrgWords.CVA0B0S, MrgWords.CVA1B1S))>;
3862 def : Pat<(v4i32 (build_vector ExtDbl.A0U, ExtDbl.A1U,
3863 ExtDbl.B0U, ExtDbl.B1U)),
3864 (v4i32 (VMRGEW MrgWords.CVA0B0U, MrgWords.CVA1B1U))>;
3867 let Predicates = [IsLittleEndian, HasP8Vector] in {
3868 def : Pat<DWToSPExtractConv.BVU,
3869 (v4f32 (VPKUDUM (XXSLDWI (XVCVUXDSP $S2), (XVCVUXDSP $S2), 3),
3870 (XXSLDWI (XVCVUXDSP $S1), (XVCVUXDSP $S1), 3)))>;
3871 def : Pat<DWToSPExtractConv.BVS,
3872 (v4f32 (VPKUDUM (XXSLDWI (XVCVSXDSP $S2), (XVCVSXDSP $S2), 3),
3873 (XXSLDWI (XVCVSXDSP $S1), (XVCVSXDSP $S1), 3)))>;
3876 let Predicates = [IsLittleEndian, HasVSX] in {
3877 // Little endian, available on all targets with VSX
3878 def : Pat<(v2f64 (build_vector f64:$A, f64:$B)),
3880 (COPY_TO_REGCLASS $B, VSRC),
3881 (COPY_TO_REGCLASS $A, VSRC), 0))>;
3883 def : Pat<(v4f32 (build_vector f32:$D, f32:$C, f32:$B, f32:$A)),
3884 (VMRGEW MrgFP.AC, MrgFP.BD)>;
3885 def : Pat<(v4f32 (build_vector DblToFlt.A0, DblToFlt.A1,
3886 DblToFlt.B0, DblToFlt.B1)),
3887 (v4f32 (VMRGEW MrgFP.BAhToFlt, MrgFP.BAlToFlt))>;
3889 // Convert 4 doubles to a vector of ints.
3890 def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.B,
3891 DblToInt.C, DblToInt.D)),
3892 (v4i32 (VMRGEW MrgWords.CVDBS, MrgWords.CVCAS))>;
3893 def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.B,
3894 DblToUInt.C, DblToUInt.D)),
3895 (v4i32 (VMRGEW MrgWords.CVDBU, MrgWords.CVCAU))>;
3896 def : Pat<(v4i32 (build_vector ExtDbl.A0S, ExtDbl.A1S,
3897 ExtDbl.B0S, ExtDbl.B1S)),
3898 (v4i32 (VMRGEW MrgWords.CVB1A1S, MrgWords.CVB0A0S))>;
3899 def : Pat<(v4i32 (build_vector ExtDbl.A0U, ExtDbl.A1U,
3900 ExtDbl.B0U, ExtDbl.B1U)),
3901 (v4i32 (VMRGEW MrgWords.CVB1A1U, MrgWords.CVB0A0U))>;
3904 let Predicates = [HasDirectMove] in {
3905 // Endianness-neutral constant splat on P8 and newer targets. The reason
3906 // for this pattern is that on targets with direct moves, we don't expand
3907 // BUILD_VECTOR nodes for v4i32.
3908 def : Pat<(v4i32 (build_vector immSExt5NonZero:$A, immSExt5NonZero:$A,
3909 immSExt5NonZero:$A, immSExt5NonZero:$A)),
3910 (v4i32 (VSPLTISW imm:$A))>;
3913 let Predicates = [IsBigEndian, HasDirectMove, NoP9Vector] in {
3914 // Big endian integer vectors using direct moves.
3915 def : Pat<(v2i64 (build_vector i64:$A, i64:$B)),
3917 (COPY_TO_REGCLASS (MTVSRD $A), VSRC),
3918 (COPY_TO_REGCLASS (MTVSRD $B), VSRC), 0))>;
3919 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
3922 (MTVSRD (RLDIMI AnyExts.B, AnyExts.A, 32, 0)), VSRC),
3924 (MTVSRD (RLDIMI AnyExts.D, AnyExts.C, 32, 0)), VSRC), 0)>;
3925 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
3926 (XXSPLTW (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 1)>;
3929 let Predicates = [IsLittleEndian, HasDirectMove, NoP9Vector] in {
3930 // Little endian integer vectors using direct moves.
3931 def : Pat<(v2i64 (build_vector i64:$A, i64:$B)),
3933 (COPY_TO_REGCLASS (MTVSRD $B), VSRC),
3934 (COPY_TO_REGCLASS (MTVSRD $A), VSRC), 0))>;
3935 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
3938 (MTVSRD (RLDIMI AnyExts.C, AnyExts.D, 32, 0)), VSRC),
3940 (MTVSRD (RLDIMI AnyExts.A, AnyExts.B, 32, 0)), VSRC), 0)>;
3941 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
3942 (XXSPLTW (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 1)>;
3945 let Predicates = [HasP9Vector] in {
3946 // Endianness-neutral patterns for const splats with ISA 3.0 instructions.
3947 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
3948 (v4i32 (MTVSRWS $A))>;
3949 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
3950 (v4i32 (MTVSRWS $A))>;
3951 def : Pat<(v16i8 (build_vector immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3952 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3953 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3954 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3955 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3957 (v16i8 (COPY_TO_REGCLASS (XXSPLTIB imm:$A), VSRC))>;
3958 def : Pat<(v16i8 immAllOnesV),
3959 (v16i8 (COPY_TO_REGCLASS (XXSPLTIB 255), VSRC))>;
3960 def : Pat<(v8i16 immAllOnesV),
3961 (v8i16 (COPY_TO_REGCLASS (XXSPLTIB 255), VSRC))>;
3962 def : Pat<(v4i32 immAllOnesV),
3963 (v4i32 (XXSPLTIB 255))>;
3964 def : Pat<(v2i64 immAllOnesV),
3965 (v2i64 (XXSPLTIB 255))>;
3966 def : Pat<(v4i32 (scalar_to_vector FltToIntLoad.A)),
3967 (v4i32 (XVCVSPSXWS (LXVWSX xoaddr:$A)))>;
3968 def : Pat<(v4i32 (scalar_to_vector FltToUIntLoad.A)),
3969 (v4i32 (XVCVSPUXWS (LXVWSX xoaddr:$A)))>;
3970 def : Pat<(v4i32 (scalar_to_vector DblToIntLoadP9.A)),
3971 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3972 (XSCVDPSXWS (DFLOADf64 ixaddr:$A)), VSRC), 1))>;
3973 def : Pat<(v4i32 (scalar_to_vector DblToUIntLoadP9.A)),
3974 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3975 (XSCVDPUXWS (DFLOADf64 ixaddr:$A)), VSRC), 1))>;
3976 def : Pat<(v2i64 (scalar_to_vector FltToLongLoadP9.A)),
3977 (v2i64 (XXPERMDIs (XSCVDPSXDS (COPY_TO_REGCLASS
3978 (DFLOADf32 ixaddr:$A),
3980 def : Pat<(v2i64 (scalar_to_vector FltToULongLoadP9.A)),
3981 (v2i64 (XXPERMDIs (XSCVDPUXDS (COPY_TO_REGCLASS
3982 (DFLOADf32 ixaddr:$A),
3986 let Predicates = [IsISA3_0, HasDirectMove, IsBigEndian] in {
3987 def : Pat<(i64 (extractelt v2i64:$A, 1)),
3988 (i64 (MFVSRLD $A))>;
3989 // Better way to build integer vectors if we have MTVSRDD. Big endian.
3990 def : Pat<(v2i64 (build_vector i64:$rB, i64:$rA)),
3991 (v2i64 (MTVSRDD $rB, $rA))>;
3992 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
3994 (RLDIMI AnyExts.B, AnyExts.A, 32, 0),
3995 (RLDIMI AnyExts.D, AnyExts.C, 32, 0))>;
3998 let Predicates = [IsISA3_0, HasDirectMove, IsLittleEndian] in {
3999 def : Pat<(i64 (extractelt v2i64:$A, 0)),
4000 (i64 (MFVSRLD $A))>;
4001 // Better way to build integer vectors if we have MTVSRDD. Little endian.
4002 def : Pat<(v2i64 (build_vector i64:$rA, i64:$rB)),
4003 (v2i64 (MTVSRDD $rB, $rA))>;
4004 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
4006 (RLDIMI AnyExts.C, AnyExts.D, 32, 0),
4007 (RLDIMI AnyExts.A, AnyExts.B, 32, 0))>;
4009 // P9 Altivec instructions that can be used to build vectors.
4010 // Adding them to PPCInstrVSX.td rather than PPCAltivecVSX.td to compete
4011 // with complexities of existing build vector patterns in this file.
4012 let Predicates = [HasP9Altivec, IsLittleEndian] in {
4013 def : Pat<(v2i64 (build_vector WordToDWord.LE_A0, WordToDWord.LE_A1)),
4014 (v2i64 (VEXTSW2D $A))>;
4015 def : Pat<(v2i64 (build_vector HWordToDWord.LE_A0, HWordToDWord.LE_A1)),
4016 (v2i64 (VEXTSH2D $A))>;
4017 def : Pat<(v4i32 (build_vector HWordToWord.LE_A0, HWordToWord.LE_A1,
4018 HWordToWord.LE_A2, HWordToWord.LE_A3)),
4019 (v4i32 (VEXTSH2W $A))>;
4020 def : Pat<(v4i32 (build_vector ByteToWord.LE_A0, ByteToWord.LE_A1,
4021 ByteToWord.LE_A2, ByteToWord.LE_A3)),
4022 (v4i32 (VEXTSB2W $A))>;
4023 def : Pat<(v2i64 (build_vector ByteToDWord.LE_A0, ByteToDWord.LE_A1)),
4024 (v2i64 (VEXTSB2D $A))>;
4027 let Predicates = [HasP9Altivec, IsBigEndian] in {
4028 def : Pat<(v2i64 (build_vector WordToDWord.BE_A0, WordToDWord.BE_A1)),
4029 (v2i64 (VEXTSW2D $A))>;
4030 def : Pat<(v2i64 (build_vector HWordToDWord.BE_A0, HWordToDWord.BE_A1)),
4031 (v2i64 (VEXTSH2D $A))>;
4032 def : Pat<(v4i32 (build_vector HWordToWord.BE_A0, HWordToWord.BE_A1,
4033 HWordToWord.BE_A2, HWordToWord.BE_A3)),
4034 (v4i32 (VEXTSH2W $A))>;
4035 def : Pat<(v4i32 (build_vector ByteToWord.BE_A0, ByteToWord.BE_A1,
4036 ByteToWord.BE_A2, ByteToWord.BE_A3)),
4037 (v4i32 (VEXTSB2W $A))>;
4038 def : Pat<(v2i64 (build_vector ByteToDWord.BE_A0, ByteToDWord.BE_A1)),
4039 (v2i64 (VEXTSB2D $A))>;
4042 let Predicates = [HasP9Altivec] in {
4043 def: Pat<(v2i64 (PPCSExtVElems v16i8:$A)),
4044 (v2i64 (VEXTSB2D $A))>;
4045 def: Pat<(v2i64 (PPCSExtVElems v8i16:$A)),
4046 (v2i64 (VEXTSH2D $A))>;
4047 def: Pat<(v2i64 (PPCSExtVElems v4i32:$A)),
4048 (v2i64 (VEXTSW2D $A))>;
4049 def: Pat<(v4i32 (PPCSExtVElems v16i8:$A)),
4050 (v4i32 (VEXTSB2W $A))>;
4051 def: Pat<(v4i32 (PPCSExtVElems v8i16:$A)),
4052 (v4i32 (VEXTSH2W $A))>;
4056 // Put this P9Altivec related definition here since it's possible to be
4057 // selected to VSX instruction xvnegsp, avoid possible undef.
4058 let Predicates = [HasP9Altivec] in {
4060 def : Pat<(v4i32 (PPCvabsd v4i32:$A, v4i32:$B, (i32 0))),
4061 (v4i32 (VABSDUW $A, $B))>;
4063 def : Pat<(v8i16 (PPCvabsd v8i16:$A, v8i16:$B, (i32 0))),
4064 (v8i16 (VABSDUH $A, $B))>;
4066 def : Pat<(v16i8 (PPCvabsd v16i8:$A, v16i8:$B, (i32 0))),
4067 (v16i8 (VABSDUB $A, $B))>;
4069 // As PPCVABSD description, the last operand indicates whether do the
4071 def : Pat<(v4i32 (PPCvabsd v4i32:$A, v4i32:$B, (i32 1))),
4072 (v4i32 (VABSDUW (XVNEGSP $A), (XVNEGSP $B)))>;