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>
71 def PPClxvd2x : SDNode<"PPCISD::LXVD2X", SDT_PPClxvd2x,
72 [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
73 def PPCstxvd2x : SDNode<"PPCISD::STXVD2X", SDT_PPCstxvd2x,
74 [SDNPHasChain, SDNPMayStore]>;
75 def PPCxxswapd : SDNode<"PPCISD::XXSWAPD", SDT_PPCxxswapd, [SDNPHasChain]>;
76 def PPCmfvsr : SDNode<"PPCISD::MFVSR", SDTUnaryOp, []>;
77 def PPCmtvsra : SDNode<"PPCISD::MTVSRA", SDTUnaryOp, []>;
78 def PPCmtvsrz : SDNode<"PPCISD::MTVSRZ", SDTUnaryOp, []>;
79 def PPCsvec2fp : SDNode<"PPCISD::SINT_VEC_TO_FP", SDTVecConv, []>;
80 def PPCuvec2fp: SDNode<"PPCISD::UINT_VEC_TO_FP", SDTVecConv, []>;
81 def PPCswapNoChain : SDNode<"PPCISD::SWAP_NO_CHAIN", SDT_PPCxxswapd>;
83 multiclass XX3Form_Rcr<bits<6> opcode, bits<7> xo, string asmbase,
84 string asmstr, InstrItinClass itin, Intrinsic Int,
85 ValueType OutTy, ValueType InTy> {
86 let BaseName = asmbase in {
87 def NAME : XX3Form_Rc<opcode, xo, (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
88 !strconcat(asmbase, !strconcat(" ", asmstr)), itin,
89 [(set OutTy:$XT, (Int InTy:$XA, InTy:$XB))]>;
91 def o : XX3Form_Rc<opcode, xo, (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
92 !strconcat(asmbase, !strconcat(". ", asmstr)), itin,
94 (InTy (PPCvcmp_o InTy:$XA, InTy:$XB, xo)))]>,
99 // Instruction form with a single input register for instructions such as
100 // XXPERMDI. The reason for defining this is that specifying multiple chained
101 // operands (such as loads) to an instruction will perform both chained
102 // operations rather than coalescing them into a single register - even though
103 // the source memory location is the same. This simply forces the instruction
104 // to use the same register for both inputs.
105 // For example, an output DAG such as this:
106 // (XXPERMDI (LXSIBZX xoaddr:$src), (LXSIBZX xoaddr:$src ), 0))
107 // would result in two load instructions emitted and used as separate inputs
108 // to the XXPERMDI instruction.
109 class XX3Form_2s<bits<6> opcode, bits<5> xo, dag OOL, dag IOL, string asmstr,
110 InstrItinClass itin, list<dag> pattern>
111 : XX3Form_2<opcode, xo, OOL, IOL, asmstr, itin, pattern> {
115 def HasVSX : Predicate<"PPCSubTarget->hasVSX()">;
116 def IsLittleEndian : Predicate<"PPCSubTarget->isLittleEndian()">;
117 def IsBigEndian : Predicate<"!PPCSubTarget->isLittleEndian()">;
118 def HasOnlySwappingMemOps : Predicate<"!PPCSubTarget->hasP9Vector()">;
120 let Predicates = [HasVSX] in {
121 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
122 let UseVSXReg = 1 in {
123 let hasSideEffects = 0 in { // VSX instructions don't have side effects.
126 // Load indexed instructions
127 let mayLoad = 1, mayStore = 0 in {
129 def LXSDX : XX1Form_memOp<31, 588,
130 (outs vsfrc:$XT), (ins memrr:$src),
131 "lxsdx $XT, $src", IIC_LdStLFD,
134 // Pseudo instruction XFLOADf64 will be expanded to LXSDX or LFDX later
135 let isPseudo = 1, CodeSize = 3 in
136 def XFLOADf64 : PseudoXFormMemOp<(outs vsfrc:$XT), (ins memrr:$src),
138 [(set f64:$XT, (load xoaddr:$src))]>;
140 let Predicates = [HasVSX, HasOnlySwappingMemOps] in
141 def LXVD2X : XX1Form_memOp<31, 844,
142 (outs vsrc:$XT), (ins memrr:$src),
143 "lxvd2x $XT, $src", IIC_LdStLFD,
144 [(set v2f64:$XT, (int_ppc_vsx_lxvd2x xoaddr:$src))]>;
146 def LXVDSX : XX1Form_memOp<31, 332,
147 (outs vsrc:$XT), (ins memrr:$src),
148 "lxvdsx $XT, $src", IIC_LdStLFD, []>;
150 let Predicates = [HasVSX, HasOnlySwappingMemOps] in
151 def LXVW4X : XX1Form_memOp<31, 780,
152 (outs vsrc:$XT), (ins memrr:$src),
153 "lxvw4x $XT, $src", IIC_LdStLFD,
157 // Store indexed instructions
158 let mayStore = 1, mayLoad = 0 in {
160 def STXSDX : XX1Form_memOp<31, 716,
161 (outs), (ins vsfrc:$XT, memrr:$dst),
162 "stxsdx $XT, $dst", IIC_LdStSTFD,
165 // Pseudo instruction XFSTOREf64 will be expanded to STXSDX or STFDX later
166 let isPseudo = 1, CodeSize = 3 in
167 def XFSTOREf64 : PseudoXFormMemOp<(outs), (ins vsfrc:$XT, memrr:$dst),
169 [(store f64:$XT, xoaddr:$dst)]>;
171 let Predicates = [HasVSX, HasOnlySwappingMemOps] in {
172 // The behaviour of this instruction is endianness-specific so we provide no
173 // pattern to match it without considering endianness.
174 def STXVD2X : XX1Form_memOp<31, 972,
175 (outs), (ins vsrc:$XT, memrr:$dst),
176 "stxvd2x $XT, $dst", IIC_LdStSTFD,
179 def STXVW4X : XX1Form_memOp<31, 908,
180 (outs), (ins vsrc:$XT, memrr:$dst),
181 "stxvw4x $XT, $dst", IIC_LdStSTFD,
186 // Add/Mul Instructions
187 let isCommutable = 1 in {
188 def XSADDDP : XX3Form<60, 32,
189 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
190 "xsadddp $XT, $XA, $XB", IIC_VecFP,
191 [(set f64:$XT, (fadd f64:$XA, f64:$XB))]>;
192 def XSMULDP : XX3Form<60, 48,
193 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
194 "xsmuldp $XT, $XA, $XB", IIC_VecFP,
195 [(set f64:$XT, (fmul f64:$XA, f64:$XB))]>;
197 def XVADDDP : XX3Form<60, 96,
198 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
199 "xvadddp $XT, $XA, $XB", IIC_VecFP,
200 [(set v2f64:$XT, (fadd v2f64:$XA, v2f64:$XB))]>;
202 def XVADDSP : XX3Form<60, 64,
203 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
204 "xvaddsp $XT, $XA, $XB", IIC_VecFP,
205 [(set v4f32:$XT, (fadd v4f32:$XA, v4f32:$XB))]>;
207 def XVMULDP : XX3Form<60, 112,
208 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
209 "xvmuldp $XT, $XA, $XB", IIC_VecFP,
210 [(set v2f64:$XT, (fmul v2f64:$XA, v2f64:$XB))]>;
212 def XVMULSP : XX3Form<60, 80,
213 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
214 "xvmulsp $XT, $XA, $XB", IIC_VecFP,
215 [(set v4f32:$XT, (fmul v4f32:$XA, v4f32:$XB))]>;
218 // Subtract Instructions
219 def XSSUBDP : XX3Form<60, 40,
220 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
221 "xssubdp $XT, $XA, $XB", IIC_VecFP,
222 [(set f64:$XT, (fsub f64:$XA, f64:$XB))]>;
224 def XVSUBDP : XX3Form<60, 104,
225 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
226 "xvsubdp $XT, $XA, $XB", IIC_VecFP,
227 [(set v2f64:$XT, (fsub v2f64:$XA, v2f64:$XB))]>;
228 def XVSUBSP : XX3Form<60, 72,
229 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
230 "xvsubsp $XT, $XA, $XB", IIC_VecFP,
231 [(set v4f32:$XT, (fsub v4f32:$XA, v4f32:$XB))]>;
234 let BaseName = "XSMADDADP" in {
235 let isCommutable = 1 in
236 def XSMADDADP : XX3Form<60, 33,
237 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
238 "xsmaddadp $XT, $XA, $XB", IIC_VecFP,
239 [(set f64:$XT, (fma f64:$XA, f64:$XB, f64:$XTi))]>,
240 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
242 let IsVSXFMAAlt = 1 in
243 def XSMADDMDP : XX3Form<60, 41,
244 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
245 "xsmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
246 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
250 let BaseName = "XSMSUBADP" in {
251 let isCommutable = 1 in
252 def XSMSUBADP : XX3Form<60, 49,
253 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
254 "xsmsubadp $XT, $XA, $XB", IIC_VecFP,
255 [(set f64:$XT, (fma f64:$XA, f64:$XB, (fneg f64:$XTi)))]>,
256 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
258 let IsVSXFMAAlt = 1 in
259 def XSMSUBMDP : XX3Form<60, 57,
260 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
261 "xsmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
262 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
266 let BaseName = "XSNMADDADP" in {
267 let isCommutable = 1 in
268 def XSNMADDADP : XX3Form<60, 161,
269 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
270 "xsnmaddadp $XT, $XA, $XB", IIC_VecFP,
271 [(set f64:$XT, (fneg (fma f64:$XA, f64:$XB, f64:$XTi)))]>,
272 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
274 let IsVSXFMAAlt = 1 in
275 def XSNMADDMDP : XX3Form<60, 169,
276 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
277 "xsnmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
278 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
282 let BaseName = "XSNMSUBADP" in {
283 let isCommutable = 1 in
284 def XSNMSUBADP : XX3Form<60, 177,
285 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
286 "xsnmsubadp $XT, $XA, $XB", IIC_VecFP,
287 [(set f64:$XT, (fneg (fma f64:$XA, f64:$XB, (fneg f64:$XTi))))]>,
288 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
290 let IsVSXFMAAlt = 1 in
291 def XSNMSUBMDP : XX3Form<60, 185,
292 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
293 "xsnmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
294 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
298 let BaseName = "XVMADDADP" in {
299 let isCommutable = 1 in
300 def XVMADDADP : XX3Form<60, 97,
301 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
302 "xvmaddadp $XT, $XA, $XB", IIC_VecFP,
303 [(set v2f64:$XT, (fma v2f64:$XA, v2f64:$XB, v2f64:$XTi))]>,
304 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
306 let IsVSXFMAAlt = 1 in
307 def XVMADDMDP : XX3Form<60, 105,
308 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
309 "xvmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
310 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
314 let BaseName = "XVMADDASP" in {
315 let isCommutable = 1 in
316 def XVMADDASP : XX3Form<60, 65,
317 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
318 "xvmaddasp $XT, $XA, $XB", IIC_VecFP,
319 [(set v4f32:$XT, (fma v4f32:$XA, v4f32:$XB, v4f32:$XTi))]>,
320 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
322 let IsVSXFMAAlt = 1 in
323 def XVMADDMSP : XX3Form<60, 73,
324 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
325 "xvmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
326 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
330 let BaseName = "XVMSUBADP" in {
331 let isCommutable = 1 in
332 def XVMSUBADP : XX3Form<60, 113,
333 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
334 "xvmsubadp $XT, $XA, $XB", IIC_VecFP,
335 [(set v2f64:$XT, (fma v2f64:$XA, v2f64:$XB, (fneg v2f64:$XTi)))]>,
336 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
338 let IsVSXFMAAlt = 1 in
339 def XVMSUBMDP : XX3Form<60, 121,
340 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
341 "xvmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
342 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
346 let BaseName = "XVMSUBASP" in {
347 let isCommutable = 1 in
348 def XVMSUBASP : XX3Form<60, 81,
349 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
350 "xvmsubasp $XT, $XA, $XB", IIC_VecFP,
351 [(set v4f32:$XT, (fma v4f32:$XA, v4f32:$XB, (fneg v4f32:$XTi)))]>,
352 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
354 let IsVSXFMAAlt = 1 in
355 def XVMSUBMSP : XX3Form<60, 89,
356 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
357 "xvmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
358 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
362 let BaseName = "XVNMADDADP" in {
363 let isCommutable = 1 in
364 def XVNMADDADP : XX3Form<60, 225,
365 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
366 "xvnmaddadp $XT, $XA, $XB", IIC_VecFP,
367 [(set v2f64:$XT, (fneg (fma v2f64:$XA, v2f64:$XB, v2f64:$XTi)))]>,
368 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
370 let IsVSXFMAAlt = 1 in
371 def XVNMADDMDP : XX3Form<60, 233,
372 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
373 "xvnmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
374 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
378 let BaseName = "XVNMADDASP" in {
379 let isCommutable = 1 in
380 def XVNMADDASP : XX3Form<60, 193,
381 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
382 "xvnmaddasp $XT, $XA, $XB", IIC_VecFP,
383 [(set v4f32:$XT, (fneg (fma v4f32:$XA, v4f32:$XB, v4f32:$XTi)))]>,
384 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
386 let IsVSXFMAAlt = 1 in
387 def XVNMADDMSP : XX3Form<60, 201,
388 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
389 "xvnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
390 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
394 let BaseName = "XVNMSUBADP" in {
395 let isCommutable = 1 in
396 def XVNMSUBADP : XX3Form<60, 241,
397 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
398 "xvnmsubadp $XT, $XA, $XB", IIC_VecFP,
399 [(set v2f64:$XT, (fneg (fma v2f64:$XA, v2f64:$XB, (fneg v2f64:$XTi))))]>,
400 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
402 let IsVSXFMAAlt = 1 in
403 def XVNMSUBMDP : XX3Form<60, 249,
404 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
405 "xvnmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
406 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
410 let BaseName = "XVNMSUBASP" in {
411 let isCommutable = 1 in
412 def XVNMSUBASP : XX3Form<60, 209,
413 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
414 "xvnmsubasp $XT, $XA, $XB", IIC_VecFP,
415 [(set v4f32:$XT, (fneg (fma v4f32:$XA, v4f32:$XB, (fneg v4f32:$XTi))))]>,
416 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
418 let IsVSXFMAAlt = 1 in
419 def XVNMSUBMSP : XX3Form<60, 217,
420 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
421 "xvnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
422 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
426 // Division Instructions
427 def XSDIVDP : XX3Form<60, 56,
428 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
429 "xsdivdp $XT, $XA, $XB", IIC_FPDivD,
430 [(set f64:$XT, (fdiv f64:$XA, f64:$XB))]>;
431 def XSSQRTDP : XX2Form<60, 75,
432 (outs vsfrc:$XT), (ins vsfrc:$XB),
433 "xssqrtdp $XT, $XB", IIC_FPSqrtD,
434 [(set f64:$XT, (fsqrt f64:$XB))]>;
436 def XSREDP : XX2Form<60, 90,
437 (outs vsfrc:$XT), (ins vsfrc:$XB),
438 "xsredp $XT, $XB", IIC_VecFP,
439 [(set f64:$XT, (PPCfre f64:$XB))]>;
440 def XSRSQRTEDP : XX2Form<60, 74,
441 (outs vsfrc:$XT), (ins vsfrc:$XB),
442 "xsrsqrtedp $XT, $XB", IIC_VecFP,
443 [(set f64:$XT, (PPCfrsqrte f64:$XB))]>;
445 def XSTDIVDP : XX3Form_1<60, 61,
446 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
447 "xstdivdp $crD, $XA, $XB", IIC_FPCompare, []>;
448 def XSTSQRTDP : XX2Form_1<60, 106,
449 (outs crrc:$crD), (ins vsfrc:$XB),
450 "xstsqrtdp $crD, $XB", IIC_FPCompare, []>;
452 def XVDIVDP : XX3Form<60, 120,
453 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
454 "xvdivdp $XT, $XA, $XB", IIC_FPDivD,
455 [(set v2f64:$XT, (fdiv v2f64:$XA, v2f64:$XB))]>;
456 def XVDIVSP : XX3Form<60, 88,
457 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
458 "xvdivsp $XT, $XA, $XB", IIC_FPDivS,
459 [(set v4f32:$XT, (fdiv v4f32:$XA, v4f32:$XB))]>;
461 def XVSQRTDP : XX2Form<60, 203,
462 (outs vsrc:$XT), (ins vsrc:$XB),
463 "xvsqrtdp $XT, $XB", IIC_FPSqrtD,
464 [(set v2f64:$XT, (fsqrt v2f64:$XB))]>;
465 def XVSQRTSP : XX2Form<60, 139,
466 (outs vsrc:$XT), (ins vsrc:$XB),
467 "xvsqrtsp $XT, $XB", IIC_FPSqrtS,
468 [(set v4f32:$XT, (fsqrt v4f32:$XB))]>;
470 def XVTDIVDP : XX3Form_1<60, 125,
471 (outs crrc:$crD), (ins vsrc:$XA, vsrc:$XB),
472 "xvtdivdp $crD, $XA, $XB", IIC_FPCompare, []>;
473 def XVTDIVSP : XX3Form_1<60, 93,
474 (outs crrc:$crD), (ins vsrc:$XA, vsrc:$XB),
475 "xvtdivsp $crD, $XA, $XB", IIC_FPCompare, []>;
477 def XVTSQRTDP : XX2Form_1<60, 234,
478 (outs crrc:$crD), (ins vsrc:$XB),
479 "xvtsqrtdp $crD, $XB", IIC_FPCompare, []>;
480 def XVTSQRTSP : XX2Form_1<60, 170,
481 (outs crrc:$crD), (ins vsrc:$XB),
482 "xvtsqrtsp $crD, $XB", IIC_FPCompare, []>;
484 def XVREDP : XX2Form<60, 218,
485 (outs vsrc:$XT), (ins vsrc:$XB),
486 "xvredp $XT, $XB", IIC_VecFP,
487 [(set v2f64:$XT, (PPCfre v2f64:$XB))]>;
488 def XVRESP : XX2Form<60, 154,
489 (outs vsrc:$XT), (ins vsrc:$XB),
490 "xvresp $XT, $XB", IIC_VecFP,
491 [(set v4f32:$XT, (PPCfre v4f32:$XB))]>;
493 def XVRSQRTEDP : XX2Form<60, 202,
494 (outs vsrc:$XT), (ins vsrc:$XB),
495 "xvrsqrtedp $XT, $XB", IIC_VecFP,
496 [(set v2f64:$XT, (PPCfrsqrte v2f64:$XB))]>;
497 def XVRSQRTESP : XX2Form<60, 138,
498 (outs vsrc:$XT), (ins vsrc:$XB),
499 "xvrsqrtesp $XT, $XB", IIC_VecFP,
500 [(set v4f32:$XT, (PPCfrsqrte v4f32:$XB))]>;
502 // Compare Instructions
503 def XSCMPODP : XX3Form_1<60, 43,
504 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
505 "xscmpodp $crD, $XA, $XB", IIC_FPCompare, []>;
506 def XSCMPUDP : XX3Form_1<60, 35,
507 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
508 "xscmpudp $crD, $XA, $XB", IIC_FPCompare, []>;
510 defm XVCMPEQDP : XX3Form_Rcr<60, 99,
511 "xvcmpeqdp", "$XT, $XA, $XB", IIC_VecFPCompare,
512 int_ppc_vsx_xvcmpeqdp, v2i64, v2f64>;
513 defm XVCMPEQSP : XX3Form_Rcr<60, 67,
514 "xvcmpeqsp", "$XT, $XA, $XB", IIC_VecFPCompare,
515 int_ppc_vsx_xvcmpeqsp, v4i32, v4f32>;
516 defm XVCMPGEDP : XX3Form_Rcr<60, 115,
517 "xvcmpgedp", "$XT, $XA, $XB", IIC_VecFPCompare,
518 int_ppc_vsx_xvcmpgedp, v2i64, v2f64>;
519 defm XVCMPGESP : XX3Form_Rcr<60, 83,
520 "xvcmpgesp", "$XT, $XA, $XB", IIC_VecFPCompare,
521 int_ppc_vsx_xvcmpgesp, v4i32, v4f32>;
522 defm XVCMPGTDP : XX3Form_Rcr<60, 107,
523 "xvcmpgtdp", "$XT, $XA, $XB", IIC_VecFPCompare,
524 int_ppc_vsx_xvcmpgtdp, v2i64, v2f64>;
525 defm XVCMPGTSP : XX3Form_Rcr<60, 75,
526 "xvcmpgtsp", "$XT, $XA, $XB", IIC_VecFPCompare,
527 int_ppc_vsx_xvcmpgtsp, v4i32, v4f32>;
530 def XSABSDP : XX2Form<60, 345,
531 (outs vsfrc:$XT), (ins vsfrc:$XB),
532 "xsabsdp $XT, $XB", IIC_VecFP,
533 [(set f64:$XT, (fabs f64:$XB))]>;
534 def XSNABSDP : XX2Form<60, 361,
535 (outs vsfrc:$XT), (ins vsfrc:$XB),
536 "xsnabsdp $XT, $XB", IIC_VecFP,
537 [(set f64:$XT, (fneg (fabs f64:$XB)))]>;
538 def XSNEGDP : XX2Form<60, 377,
539 (outs vsfrc:$XT), (ins vsfrc:$XB),
540 "xsnegdp $XT, $XB", IIC_VecFP,
541 [(set f64:$XT, (fneg f64:$XB))]>;
542 def XSCPSGNDP : XX3Form<60, 176,
543 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
544 "xscpsgndp $XT, $XA, $XB", IIC_VecFP,
545 [(set f64:$XT, (fcopysign f64:$XB, f64:$XA))]>;
547 def XVABSDP : XX2Form<60, 473,
548 (outs vsrc:$XT), (ins vsrc:$XB),
549 "xvabsdp $XT, $XB", IIC_VecFP,
550 [(set v2f64:$XT, (fabs v2f64:$XB))]>;
552 def XVABSSP : XX2Form<60, 409,
553 (outs vsrc:$XT), (ins vsrc:$XB),
554 "xvabssp $XT, $XB", IIC_VecFP,
555 [(set v4f32:$XT, (fabs v4f32:$XB))]>;
557 def XVCPSGNDP : XX3Form<60, 240,
558 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
559 "xvcpsgndp $XT, $XA, $XB", IIC_VecFP,
560 [(set v2f64:$XT, (fcopysign v2f64:$XB, v2f64:$XA))]>;
561 def XVCPSGNSP : XX3Form<60, 208,
562 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
563 "xvcpsgnsp $XT, $XA, $XB", IIC_VecFP,
564 [(set v4f32:$XT, (fcopysign v4f32:$XB, v4f32:$XA))]>;
566 def XVNABSDP : XX2Form<60, 489,
567 (outs vsrc:$XT), (ins vsrc:$XB),
568 "xvnabsdp $XT, $XB", IIC_VecFP,
569 [(set v2f64:$XT, (fneg (fabs v2f64:$XB)))]>;
570 def XVNABSSP : XX2Form<60, 425,
571 (outs vsrc:$XT), (ins vsrc:$XB),
572 "xvnabssp $XT, $XB", IIC_VecFP,
573 [(set v4f32:$XT, (fneg (fabs v4f32:$XB)))]>;
575 def XVNEGDP : XX2Form<60, 505,
576 (outs vsrc:$XT), (ins vsrc:$XB),
577 "xvnegdp $XT, $XB", IIC_VecFP,
578 [(set v2f64:$XT, (fneg v2f64:$XB))]>;
579 def XVNEGSP : XX2Form<60, 441,
580 (outs vsrc:$XT), (ins vsrc:$XB),
581 "xvnegsp $XT, $XB", IIC_VecFP,
582 [(set v4f32:$XT, (fneg v4f32:$XB))]>;
584 // Conversion Instructions
585 def XSCVDPSP : XX2Form<60, 265,
586 (outs vsfrc:$XT), (ins vsfrc:$XB),
587 "xscvdpsp $XT, $XB", IIC_VecFP, []>;
588 def XSCVDPSXDS : XX2Form<60, 344,
589 (outs vsfrc:$XT), (ins vsfrc:$XB),
590 "xscvdpsxds $XT, $XB", IIC_VecFP,
591 [(set f64:$XT, (PPCfctidz f64:$XB))]>;
592 let isCodeGenOnly = 1 in
593 def XSCVDPSXDSs : XX2Form<60, 344,
594 (outs vssrc:$XT), (ins vssrc:$XB),
595 "xscvdpsxds $XT, $XB", IIC_VecFP,
596 [(set f32:$XT, (PPCfctidz f32:$XB))]>;
597 def XSCVDPSXWS : XX2Form<60, 88,
598 (outs vsfrc:$XT), (ins vsfrc:$XB),
599 "xscvdpsxws $XT, $XB", IIC_VecFP,
600 [(set f64:$XT, (PPCfctiwz f64:$XB))]>;
601 let isCodeGenOnly = 1 in
602 def XSCVDPSXWSs : XX2Form<60, 88,
603 (outs vssrc:$XT), (ins vssrc:$XB),
604 "xscvdpsxws $XT, $XB", IIC_VecFP,
605 [(set f32:$XT, (PPCfctiwz f32:$XB))]>;
606 def XSCVDPUXDS : XX2Form<60, 328,
607 (outs vsfrc:$XT), (ins vsfrc:$XB),
608 "xscvdpuxds $XT, $XB", IIC_VecFP,
609 [(set f64:$XT, (PPCfctiduz f64:$XB))]>;
610 let isCodeGenOnly = 1 in
611 def XSCVDPUXDSs : XX2Form<60, 328,
612 (outs vssrc:$XT), (ins vssrc:$XB),
613 "xscvdpuxds $XT, $XB", IIC_VecFP,
614 [(set f32:$XT, (PPCfctiduz f32:$XB))]>;
615 def XSCVDPUXWS : XX2Form<60, 72,
616 (outs vsfrc:$XT), (ins vsfrc:$XB),
617 "xscvdpuxws $XT, $XB", IIC_VecFP,
618 [(set f64:$XT, (PPCfctiwuz f64:$XB))]>;
619 let isCodeGenOnly = 1 in
620 def XSCVDPUXWSs : XX2Form<60, 72,
621 (outs vssrc:$XT), (ins vssrc:$XB),
622 "xscvdpuxws $XT, $XB", IIC_VecFP,
623 [(set f32:$XT, (PPCfctiwuz f32:$XB))]>;
624 def XSCVSPDP : XX2Form<60, 329,
625 (outs vsfrc:$XT), (ins vsfrc:$XB),
626 "xscvspdp $XT, $XB", IIC_VecFP, []>;
627 def XSCVSXDDP : XX2Form<60, 376,
628 (outs vsfrc:$XT), (ins vsfrc:$XB),
629 "xscvsxddp $XT, $XB", IIC_VecFP,
630 [(set f64:$XT, (PPCfcfid f64:$XB))]>;
631 def XSCVUXDDP : XX2Form<60, 360,
632 (outs vsfrc:$XT), (ins vsfrc:$XB),
633 "xscvuxddp $XT, $XB", IIC_VecFP,
634 [(set f64:$XT, (PPCfcfidu f64:$XB))]>;
636 def XVCVDPSP : XX2Form<60, 393,
637 (outs vsrc:$XT), (ins vsrc:$XB),
638 "xvcvdpsp $XT, $XB", IIC_VecFP,
639 [(set v4f32:$XT, (int_ppc_vsx_xvcvdpsp v2f64:$XB))]>;
640 def XVCVDPSXDS : XX2Form<60, 472,
641 (outs vsrc:$XT), (ins vsrc:$XB),
642 "xvcvdpsxds $XT, $XB", IIC_VecFP,
643 [(set v2i64:$XT, (fp_to_sint v2f64:$XB))]>;
644 def XVCVDPSXWS : XX2Form<60, 216,
645 (outs vsrc:$XT), (ins vsrc:$XB),
646 "xvcvdpsxws $XT, $XB", IIC_VecFP,
647 [(set v4i32:$XT, (int_ppc_vsx_xvcvdpsxws v2f64:$XB))]>;
648 def XVCVDPUXDS : XX2Form<60, 456,
649 (outs vsrc:$XT), (ins vsrc:$XB),
650 "xvcvdpuxds $XT, $XB", IIC_VecFP,
651 [(set v2i64:$XT, (fp_to_uint v2f64:$XB))]>;
652 def XVCVDPUXWS : XX2Form<60, 200,
653 (outs vsrc:$XT), (ins vsrc:$XB),
654 "xvcvdpuxws $XT, $XB", IIC_VecFP,
655 [(set v4i32:$XT, (int_ppc_vsx_xvcvdpuxws v2f64:$XB))]>;
657 def XVCVSPDP : XX2Form<60, 457,
658 (outs vsrc:$XT), (ins vsrc:$XB),
659 "xvcvspdp $XT, $XB", IIC_VecFP,
660 [(set v2f64:$XT, (int_ppc_vsx_xvcvspdp v4f32:$XB))]>;
661 def XVCVSPSXDS : XX2Form<60, 408,
662 (outs vsrc:$XT), (ins vsrc:$XB),
663 "xvcvspsxds $XT, $XB", IIC_VecFP, []>;
664 def XVCVSPSXWS : XX2Form<60, 152,
665 (outs vsrc:$XT), (ins vsrc:$XB),
666 "xvcvspsxws $XT, $XB", IIC_VecFP,
667 [(set v4i32:$XT, (fp_to_sint v4f32:$XB))]>;
668 def XVCVSPUXDS : XX2Form<60, 392,
669 (outs vsrc:$XT), (ins vsrc:$XB),
670 "xvcvspuxds $XT, $XB", IIC_VecFP, []>;
671 def XVCVSPUXWS : XX2Form<60, 136,
672 (outs vsrc:$XT), (ins vsrc:$XB),
673 "xvcvspuxws $XT, $XB", IIC_VecFP,
674 [(set v4i32:$XT, (fp_to_uint v4f32:$XB))]>;
675 def XVCVSXDDP : XX2Form<60, 504,
676 (outs vsrc:$XT), (ins vsrc:$XB),
677 "xvcvsxddp $XT, $XB", IIC_VecFP,
678 [(set v2f64:$XT, (sint_to_fp v2i64:$XB))]>;
679 def XVCVSXDSP : XX2Form<60, 440,
680 (outs vsrc:$XT), (ins vsrc:$XB),
681 "xvcvsxdsp $XT, $XB", IIC_VecFP,
682 [(set v4f32:$XT, (int_ppc_vsx_xvcvsxdsp v2i64:$XB))]>;
683 def XVCVSXWDP : XX2Form<60, 248,
684 (outs vsrc:$XT), (ins vsrc:$XB),
685 "xvcvsxwdp $XT, $XB", IIC_VecFP,
686 [(set v2f64:$XT, (int_ppc_vsx_xvcvsxwdp v4i32:$XB))]>;
687 def XVCVSXWSP : XX2Form<60, 184,
688 (outs vsrc:$XT), (ins vsrc:$XB),
689 "xvcvsxwsp $XT, $XB", IIC_VecFP,
690 [(set v4f32:$XT, (sint_to_fp v4i32:$XB))]>;
691 def XVCVUXDDP : XX2Form<60, 488,
692 (outs vsrc:$XT), (ins vsrc:$XB),
693 "xvcvuxddp $XT, $XB", IIC_VecFP,
694 [(set v2f64:$XT, (uint_to_fp v2i64:$XB))]>;
695 def XVCVUXDSP : XX2Form<60, 424,
696 (outs vsrc:$XT), (ins vsrc:$XB),
697 "xvcvuxdsp $XT, $XB", IIC_VecFP,
698 [(set v4f32:$XT, (int_ppc_vsx_xvcvuxdsp v2i64:$XB))]>;
699 def XVCVUXWDP : XX2Form<60, 232,
700 (outs vsrc:$XT), (ins vsrc:$XB),
701 "xvcvuxwdp $XT, $XB", IIC_VecFP,
702 [(set v2f64:$XT, (int_ppc_vsx_xvcvuxwdp v4i32:$XB))]>;
703 def XVCVUXWSP : XX2Form<60, 168,
704 (outs vsrc:$XT), (ins vsrc:$XB),
705 "xvcvuxwsp $XT, $XB", IIC_VecFP,
706 [(set v4f32:$XT, (uint_to_fp v4i32:$XB))]>;
708 // Rounding Instructions
709 def XSRDPI : XX2Form<60, 73,
710 (outs vsfrc:$XT), (ins vsfrc:$XB),
711 "xsrdpi $XT, $XB", IIC_VecFP,
712 [(set f64:$XT, (fround f64:$XB))]>;
713 def XSRDPIC : XX2Form<60, 107,
714 (outs vsfrc:$XT), (ins vsfrc:$XB),
715 "xsrdpic $XT, $XB", IIC_VecFP,
716 [(set f64:$XT, (fnearbyint f64:$XB))]>;
717 def XSRDPIM : XX2Form<60, 121,
718 (outs vsfrc:$XT), (ins vsfrc:$XB),
719 "xsrdpim $XT, $XB", IIC_VecFP,
720 [(set f64:$XT, (ffloor f64:$XB))]>;
721 def XSRDPIP : XX2Form<60, 105,
722 (outs vsfrc:$XT), (ins vsfrc:$XB),
723 "xsrdpip $XT, $XB", IIC_VecFP,
724 [(set f64:$XT, (fceil f64:$XB))]>;
725 def XSRDPIZ : XX2Form<60, 89,
726 (outs vsfrc:$XT), (ins vsfrc:$XB),
727 "xsrdpiz $XT, $XB", IIC_VecFP,
728 [(set f64:$XT, (ftrunc f64:$XB))]>;
730 def XVRDPI : XX2Form<60, 201,
731 (outs vsrc:$XT), (ins vsrc:$XB),
732 "xvrdpi $XT, $XB", IIC_VecFP,
733 [(set v2f64:$XT, (fround v2f64:$XB))]>;
734 def XVRDPIC : XX2Form<60, 235,
735 (outs vsrc:$XT), (ins vsrc:$XB),
736 "xvrdpic $XT, $XB", IIC_VecFP,
737 [(set v2f64:$XT, (fnearbyint v2f64:$XB))]>;
738 def XVRDPIM : XX2Form<60, 249,
739 (outs vsrc:$XT), (ins vsrc:$XB),
740 "xvrdpim $XT, $XB", IIC_VecFP,
741 [(set v2f64:$XT, (ffloor v2f64:$XB))]>;
742 def XVRDPIP : XX2Form<60, 233,
743 (outs vsrc:$XT), (ins vsrc:$XB),
744 "xvrdpip $XT, $XB", IIC_VecFP,
745 [(set v2f64:$XT, (fceil v2f64:$XB))]>;
746 def XVRDPIZ : XX2Form<60, 217,
747 (outs vsrc:$XT), (ins vsrc:$XB),
748 "xvrdpiz $XT, $XB", IIC_VecFP,
749 [(set v2f64:$XT, (ftrunc v2f64:$XB))]>;
751 def XVRSPI : XX2Form<60, 137,
752 (outs vsrc:$XT), (ins vsrc:$XB),
753 "xvrspi $XT, $XB", IIC_VecFP,
754 [(set v4f32:$XT, (fround v4f32:$XB))]>;
755 def XVRSPIC : XX2Form<60, 171,
756 (outs vsrc:$XT), (ins vsrc:$XB),
757 "xvrspic $XT, $XB", IIC_VecFP,
758 [(set v4f32:$XT, (fnearbyint v4f32:$XB))]>;
759 def XVRSPIM : XX2Form<60, 185,
760 (outs vsrc:$XT), (ins vsrc:$XB),
761 "xvrspim $XT, $XB", IIC_VecFP,
762 [(set v4f32:$XT, (ffloor v4f32:$XB))]>;
763 def XVRSPIP : XX2Form<60, 169,
764 (outs vsrc:$XT), (ins vsrc:$XB),
765 "xvrspip $XT, $XB", IIC_VecFP,
766 [(set v4f32:$XT, (fceil v4f32:$XB))]>;
767 def XVRSPIZ : XX2Form<60, 153,
768 (outs vsrc:$XT), (ins vsrc:$XB),
769 "xvrspiz $XT, $XB", IIC_VecFP,
770 [(set v4f32:$XT, (ftrunc v4f32:$XB))]>;
772 // Max/Min Instructions
773 let isCommutable = 1 in {
774 def XSMAXDP : XX3Form<60, 160,
775 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
776 "xsmaxdp $XT, $XA, $XB", IIC_VecFP,
778 (int_ppc_vsx_xsmaxdp vsfrc:$XA, vsfrc:$XB))]>;
779 def XSMINDP : XX3Form<60, 168,
780 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
781 "xsmindp $XT, $XA, $XB", IIC_VecFP,
783 (int_ppc_vsx_xsmindp vsfrc:$XA, vsfrc:$XB))]>;
785 def XVMAXDP : XX3Form<60, 224,
786 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
787 "xvmaxdp $XT, $XA, $XB", IIC_VecFP,
789 (int_ppc_vsx_xvmaxdp vsrc:$XA, vsrc:$XB))]>;
790 def XVMINDP : XX3Form<60, 232,
791 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
792 "xvmindp $XT, $XA, $XB", IIC_VecFP,
794 (int_ppc_vsx_xvmindp vsrc:$XA, vsrc:$XB))]>;
796 def XVMAXSP : XX3Form<60, 192,
797 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
798 "xvmaxsp $XT, $XA, $XB", IIC_VecFP,
800 (int_ppc_vsx_xvmaxsp vsrc:$XA, vsrc:$XB))]>;
801 def XVMINSP : XX3Form<60, 200,
802 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
803 "xvminsp $XT, $XA, $XB", IIC_VecFP,
805 (int_ppc_vsx_xvminsp vsrc:$XA, vsrc:$XB))]>;
809 // Logical Instructions
810 let isCommutable = 1 in
811 def XXLAND : XX3Form<60, 130,
812 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
813 "xxland $XT, $XA, $XB", IIC_VecGeneral,
814 [(set v4i32:$XT, (and v4i32:$XA, v4i32:$XB))]>;
815 def XXLANDC : XX3Form<60, 138,
816 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
817 "xxlandc $XT, $XA, $XB", IIC_VecGeneral,
818 [(set v4i32:$XT, (and v4i32:$XA,
819 (vnot_ppc v4i32:$XB)))]>;
820 let isCommutable = 1 in {
821 def XXLNOR : XX3Form<60, 162,
822 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
823 "xxlnor $XT, $XA, $XB", IIC_VecGeneral,
824 [(set v4i32:$XT, (vnot_ppc (or v4i32:$XA,
826 def XXLOR : XX3Form<60, 146,
827 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
828 "xxlor $XT, $XA, $XB", IIC_VecGeneral,
829 [(set v4i32:$XT, (or v4i32:$XA, v4i32:$XB))]>;
830 let isCodeGenOnly = 1 in
831 def XXLORf: XX3Form<60, 146,
832 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
833 "xxlor $XT, $XA, $XB", IIC_VecGeneral, []>;
834 def XXLXOR : XX3Form<60, 154,
835 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
836 "xxlxor $XT, $XA, $XB", IIC_VecGeneral,
837 [(set v4i32:$XT, (xor v4i32:$XA, v4i32:$XB))]>;
839 let isCodeGenOnly = 1 in
840 def XXLXORz : XX3Form_Zero<60, 154, (outs vsrc:$XT), (ins),
841 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
842 [(set v4i32:$XT, (v4i32 immAllZerosV))]>;
844 let isCodeGenOnly = 1 in {
845 def XXLXORdpz : XX3Form_SetZero<60, 154,
846 (outs vsfrc:$XT), (ins),
847 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
848 [(set f64:$XT, (fpimm0))]>;
849 def XXLXORspz : XX3Form_SetZero<60, 154,
850 (outs vssrc:$XT), (ins),
851 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
852 [(set f32:$XT, (fpimm0))]>;
855 // Permutation Instructions
856 def XXMRGHW : XX3Form<60, 18,
857 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
858 "xxmrghw $XT, $XA, $XB", IIC_VecPerm, []>;
859 def XXMRGLW : XX3Form<60, 50,
860 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
861 "xxmrglw $XT, $XA, $XB", IIC_VecPerm, []>;
863 def XXPERMDI : XX3Form_2<60, 10,
864 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, u2imm:$DM),
865 "xxpermdi $XT, $XA, $XB, $DM", IIC_VecPerm,
866 [(set v2i64:$XT, (PPCxxpermdi v2i64:$XA, v2i64:$XB,
868 let isCodeGenOnly = 1 in
869 def XXPERMDIs : XX3Form_2s<60, 10, (outs vsrc:$XT), (ins vsfrc:$XA, u2imm:$DM),
870 "xxpermdi $XT, $XA, $XA, $DM", IIC_VecPerm, []>;
871 def XXSEL : XX4Form<60, 3,
872 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, vsrc:$XC),
873 "xxsel $XT, $XA, $XB, $XC", IIC_VecPerm, []>;
875 def XXSLDWI : XX3Form_2<60, 2,
876 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, u2imm:$SHW),
877 "xxsldwi $XT, $XA, $XB, $SHW", IIC_VecPerm,
878 [(set v4i32:$XT, (PPCvecshl v4i32:$XA, v4i32:$XB,
879 imm32SExt16:$SHW))]>;
881 let isCodeGenOnly = 1 in
882 def XXSLDWIs : XX3Form_2s<60, 2,
883 (outs vsrc:$XT), (ins vsfrc:$XA, u2imm:$SHW),
884 "xxsldwi $XT, $XA, $XA, $SHW", IIC_VecPerm, []>;
886 def XXSPLTW : XX2Form_2<60, 164,
887 (outs vsrc:$XT), (ins vsrc:$XB, u2imm:$UIM),
888 "xxspltw $XT, $XB, $UIM", IIC_VecPerm,
890 (PPCxxsplt v4i32:$XB, imm32SExt16:$UIM))]>;
891 let isCodeGenOnly = 1 in
892 def XXSPLTWs : XX2Form_2<60, 164,
893 (outs vsrc:$XT), (ins vfrc:$XB, u2imm:$UIM),
894 "xxspltw $XT, $XB, $UIM", IIC_VecPerm, []>;
899 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation. Expanded after
900 // instruction selection into a branch sequence.
901 let usesCustomInserter = 1, // Expanded after instruction selection.
902 PPC970_Single = 1 in {
904 def SELECT_CC_VSRC: Pseudo<(outs vsrc:$dst),
905 (ins crrc:$cond, vsrc:$T, vsrc:$F, i32imm:$BROPC),
908 def SELECT_VSRC: Pseudo<(outs vsrc:$dst),
909 (ins crbitrc:$cond, vsrc:$T, vsrc:$F),
912 (select i1:$cond, v2f64:$T, v2f64:$F))]>;
913 def SELECT_CC_VSFRC: Pseudo<(outs f8rc:$dst),
914 (ins crrc:$cond, f8rc:$T, f8rc:$F,
915 i32imm:$BROPC), "#SELECT_CC_VSFRC",
917 def SELECT_VSFRC: Pseudo<(outs f8rc:$dst),
918 (ins crbitrc:$cond, f8rc:$T, f8rc:$F),
921 (select i1:$cond, f64:$T, f64:$F))]>;
922 def SELECT_CC_VSSRC: Pseudo<(outs f4rc:$dst),
923 (ins crrc:$cond, f4rc:$T, f4rc:$F,
924 i32imm:$BROPC), "#SELECT_CC_VSSRC",
926 def SELECT_VSSRC: Pseudo<(outs f4rc:$dst),
927 (ins crbitrc:$cond, f4rc:$T, f4rc:$F),
930 (select i1:$cond, f32:$T, f32:$F))]>;
931 } // usesCustomInserter
934 def : InstAlias<"xvmovdp $XT, $XB",
935 (XVCPSGNDP vsrc:$XT, vsrc:$XB, vsrc:$XB)>;
936 def : InstAlias<"xvmovsp $XT, $XB",
937 (XVCPSGNSP vsrc:$XT, vsrc:$XB, vsrc:$XB)>;
939 def : InstAlias<"xxspltd $XT, $XB, 0",
940 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 0)>;
941 def : InstAlias<"xxspltd $XT, $XB, 1",
942 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 3)>;
943 def : InstAlias<"xxmrghd $XT, $XA, $XB",
944 (XXPERMDI vsrc:$XT, vsrc:$XA, vsrc:$XB, 0)>;
945 def : InstAlias<"xxmrgld $XT, $XA, $XB",
946 (XXPERMDI vsrc:$XT, vsrc:$XA, vsrc:$XB, 3)>;
947 def : InstAlias<"xxswapd $XT, $XB",
948 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 2)>;
949 def : InstAlias<"xxspltd $XT, $XB, 0",
950 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 0)>;
951 def : InstAlias<"xxspltd $XT, $XB, 1",
952 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 3)>;
953 def : InstAlias<"xxswapd $XT, $XB",
954 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 2)>;
956 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
958 def : Pat<(v4i32 (vnot_ppc v4i32:$A)),
959 (v4i32 (XXLNOR $A, $A))>;
960 let Predicates = [IsBigEndian] in {
961 def : Pat<(v2f64 (scalar_to_vector f64:$A)),
962 (v2f64 (SUBREG_TO_REG (i64 1), $A, sub_64))>;
964 def : Pat<(f64 (extractelt v2f64:$S, 0)),
965 (f64 (EXTRACT_SUBREG $S, sub_64))>;
966 def : Pat<(f64 (extractelt v2f64:$S, 1)),
967 (f64 (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64))>;
970 let Predicates = [IsLittleEndian] in {
971 def : Pat<(v2f64 (scalar_to_vector f64:$A)),
972 (v2f64 (XXPERMDI (SUBREG_TO_REG (i64 1), $A, sub_64),
973 (SUBREG_TO_REG (i64 1), $A, sub_64), 0))>;
975 def : Pat<(f64 (extractelt v2f64:$S, 0)),
976 (f64 (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64))>;
977 def : Pat<(f64 (extractelt v2f64:$S, 1)),
978 (f64 (EXTRACT_SUBREG $S, sub_64))>;
981 // Additional fnmsub patterns: -a*c + b == -(a*c - b)
982 def : Pat<(fma (fneg f64:$A), f64:$C, f64:$B),
983 (XSNMSUBADP $B, $C, $A)>;
984 def : Pat<(fma f64:$A, (fneg f64:$C), f64:$B),
985 (XSNMSUBADP $B, $C, $A)>;
987 def : Pat<(fma (fneg v2f64:$A), v2f64:$C, v2f64:$B),
988 (XVNMSUBADP $B, $C, $A)>;
989 def : Pat<(fma v2f64:$A, (fneg v2f64:$C), v2f64:$B),
990 (XVNMSUBADP $B, $C, $A)>;
992 def : Pat<(fma (fneg v4f32:$A), v4f32:$C, v4f32:$B),
993 (XVNMSUBASP $B, $C, $A)>;
994 def : Pat<(fma v4f32:$A, (fneg v4f32:$C), v4f32:$B),
995 (XVNMSUBASP $B, $C, $A)>;
997 def : Pat<(v2f64 (bitconvert v4f32:$A)),
998 (COPY_TO_REGCLASS $A, VSRC)>;
999 def : Pat<(v2f64 (bitconvert v4i32:$A)),
1000 (COPY_TO_REGCLASS $A, VSRC)>;
1001 def : Pat<(v2f64 (bitconvert v8i16:$A)),
1002 (COPY_TO_REGCLASS $A, VSRC)>;
1003 def : Pat<(v2f64 (bitconvert v16i8:$A)),
1004 (COPY_TO_REGCLASS $A, VSRC)>;
1006 def : Pat<(v4f32 (bitconvert v2f64:$A)),
1007 (COPY_TO_REGCLASS $A, VRRC)>;
1008 def : Pat<(v4i32 (bitconvert v2f64:$A)),
1009 (COPY_TO_REGCLASS $A, VRRC)>;
1010 def : Pat<(v8i16 (bitconvert v2f64:$A)),
1011 (COPY_TO_REGCLASS $A, VRRC)>;
1012 def : Pat<(v16i8 (bitconvert v2f64:$A)),
1013 (COPY_TO_REGCLASS $A, VRRC)>;
1015 def : Pat<(v2i64 (bitconvert v4f32:$A)),
1016 (COPY_TO_REGCLASS $A, VSRC)>;
1017 def : Pat<(v2i64 (bitconvert v4i32:$A)),
1018 (COPY_TO_REGCLASS $A, VSRC)>;
1019 def : Pat<(v2i64 (bitconvert v8i16:$A)),
1020 (COPY_TO_REGCLASS $A, VSRC)>;
1021 def : Pat<(v2i64 (bitconvert v16i8:$A)),
1022 (COPY_TO_REGCLASS $A, VSRC)>;
1024 def : Pat<(v4f32 (bitconvert v2i64:$A)),
1025 (COPY_TO_REGCLASS $A, VRRC)>;
1026 def : Pat<(v4i32 (bitconvert v2i64:$A)),
1027 (COPY_TO_REGCLASS $A, VRRC)>;
1028 def : Pat<(v8i16 (bitconvert v2i64:$A)),
1029 (COPY_TO_REGCLASS $A, VRRC)>;
1030 def : Pat<(v16i8 (bitconvert v2i64:$A)),
1031 (COPY_TO_REGCLASS $A, VRRC)>;
1033 def : Pat<(v2f64 (bitconvert v2i64:$A)),
1034 (COPY_TO_REGCLASS $A, VRRC)>;
1035 def : Pat<(v2i64 (bitconvert v2f64:$A)),
1036 (COPY_TO_REGCLASS $A, VRRC)>;
1038 def : Pat<(v2f64 (bitconvert v1i128:$A)),
1039 (COPY_TO_REGCLASS $A, VRRC)>;
1040 def : Pat<(v1i128 (bitconvert v2f64:$A)),
1041 (COPY_TO_REGCLASS $A, VRRC)>;
1043 // sign extension patterns
1044 // To extend "in place" from v2i32 to v2i64, we have input data like:
1045 // | undef | i32 | undef | i32 |
1046 // but xvcvsxwdp expects the input in big-Endian format:
1047 // | i32 | undef | i32 | undef |
1048 // so we need to shift everything to the left by one i32 (word) before
1050 def : Pat<(sext_inreg v2i64:$C, v2i32),
1051 (XVCVDPSXDS (XVCVSXWDP (XXSLDWI $C, $C, 1)))>;
1052 def : Pat<(v2f64 (sint_to_fp (sext_inreg v2i64:$C, v2i32))),
1053 (XVCVSXWDP (XXSLDWI $C, $C, 1))>;
1055 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 0)),
1056 (v2f64 (XVCVSXWDP (v2i64 (XXMRGHW $C, $C))))>;
1057 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 1)),
1058 (v2f64 (XVCVSXWDP (v2i64 (XXMRGLW $C, $C))))>;
1060 def : Pat<(v2f64 (PPCuvec2fp v4i32:$C, 0)),
1061 (v2f64 (XVCVUXWDP (v2i64 (XXMRGHW $C, $C))))>;
1062 def : Pat<(v2f64 (PPCuvec2fp v4i32:$C, 1)),
1063 (v2f64 (XVCVUXWDP (v2i64 (XXMRGLW $C, $C))))>;
1066 let Predicates = [HasVSX, HasOnlySwappingMemOps] in {
1067 def : Pat<(v2f64 (PPClxvd2x xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1070 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, xoaddr:$dst),
1071 (STXVD2X $rS, xoaddr:$dst)>;
1072 def : Pat<(int_ppc_vsx_stxvd2x_be v2f64:$rS, xoaddr:$dst),
1073 (STXVD2X $rS, xoaddr:$dst)>;
1074 def : Pat<(int_ppc_vsx_stxvw4x_be v4i32:$rS, xoaddr:$dst),
1075 (STXVW4X $rS, xoaddr:$dst)>;
1076 def : Pat<(PPCstxvd2x v2f64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1078 let Predicates = [IsBigEndian, HasVSX, HasOnlySwappingMemOps] in {
1079 def : Pat<(v2f64 (load xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1080 def : Pat<(v2i64 (load xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1081 def : Pat<(v4i32 (load xoaddr:$src)), (LXVW4X xoaddr:$src)>;
1082 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x xoaddr:$src)), (LXVW4X xoaddr:$src)>;
1083 def : Pat<(store v2f64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1084 def : Pat<(store v2i64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1085 def : Pat<(store v4i32:$XT, xoaddr:$dst), (STXVW4X $XT, xoaddr:$dst)>;
1086 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, xoaddr:$dst),
1087 (STXVW4X $rS, xoaddr:$dst)>;
1091 def : Pat<(v2f64 (PPCxxswapd v2f64:$src)), (XXPERMDI $src, $src, 2)>;
1092 def : Pat<(v2i64 (PPCxxswapd v2i64:$src)), (XXPERMDI $src, $src, 2)>;
1093 def : Pat<(v4f32 (PPCxxswapd v4f32:$src)), (XXPERMDI $src, $src, 2)>;
1094 def : Pat<(v4i32 (PPCxxswapd v4i32:$src)), (XXPERMDI $src, $src, 2)>;
1095 def : Pat<(v2f64 (PPCswapNoChain v2f64:$src)), (XXPERMDI $src, $src, 2)>;
1097 // PPCvecshl XT, XA, XA, 2 can be selected to both XXSLDWI XT,XA,XA,2 and
1098 // XXSWAPD XT,XA (i.e. XXPERMDI XT,XA,XA,2), the later one is more profitable.
1099 def : Pat<(v4i32 (PPCvecshl v4i32:$src, v4i32:$src, 2)), (XXPERMDI $src, $src, 2)>;
1102 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETLT)),
1103 (SELECT_VSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1104 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETULT)),
1105 (SELECT_VSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1106 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETLE)),
1107 (SELECT_VSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1108 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETULE)),
1109 (SELECT_VSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1110 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETEQ)),
1111 (SELECT_VSRC (CREQV $lhs, $rhs), $tval, $fval)>;
1112 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETGE)),
1113 (SELECT_VSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1114 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETUGE)),
1115 (SELECT_VSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1116 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETGT)),
1117 (SELECT_VSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1118 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETUGT)),
1119 (SELECT_VSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1120 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETNE)),
1121 (SELECT_VSRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1123 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETLT)),
1124 (SELECT_VSFRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1125 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETULT)),
1126 (SELECT_VSFRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1127 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETLE)),
1128 (SELECT_VSFRC (CRORC $lhs, $rhs), $tval, $fval)>;
1129 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETULE)),
1130 (SELECT_VSFRC (CRORC $rhs, $lhs), $tval, $fval)>;
1131 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETEQ)),
1132 (SELECT_VSFRC (CREQV $lhs, $rhs), $tval, $fval)>;
1133 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETGE)),
1134 (SELECT_VSFRC (CRORC $rhs, $lhs), $tval, $fval)>;
1135 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGE)),
1136 (SELECT_VSFRC (CRORC $lhs, $rhs), $tval, $fval)>;
1137 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETGT)),
1138 (SELECT_VSFRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1139 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGT)),
1140 (SELECT_VSFRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1141 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETNE)),
1142 (SELECT_VSFRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1145 def : Pat<(int_ppc_vsx_xvdivsp v4f32:$A, v4f32:$B),
1147 def : Pat<(int_ppc_vsx_xvdivdp v2f64:$A, v2f64:$B),
1150 // Reciprocal estimate
1151 def : Pat<(int_ppc_vsx_xvresp v4f32:$A),
1153 def : Pat<(int_ppc_vsx_xvredp v2f64:$A),
1156 // Recip. square root estimate
1157 def : Pat<(int_ppc_vsx_xvrsqrtesp v4f32:$A),
1159 def : Pat<(int_ppc_vsx_xvrsqrtedp v2f64:$A),
1162 let Predicates = [IsLittleEndian] in {
1163 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1164 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1165 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1166 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1167 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1168 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1169 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1170 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1173 let Predicates = [IsBigEndian] in {
1174 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1175 (f64 (XSCVSXDDP (COPY_TO_REGCLASS $S, VSFRC)))>;
1176 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1177 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1178 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1179 (f64 (XSCVUXDDP (COPY_TO_REGCLASS $S, VSFRC)))>;
1180 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1181 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1184 } // AddedComplexity
1188 dag Li8 = (i32 (extloadi8 xoaddr:$src));
1189 dag ZELi8 = (i32 (zextloadi8 xoaddr:$src));
1190 dag ZELi8i64 = (i64 (zextloadi8 xoaddr:$src));
1191 dag SELi8 = (i32 (sext_inreg (extloadi8 xoaddr:$src), i8));
1192 dag SELi8i64 = (i64 (sext_inreg (extloadi8 xoaddr:$src), i8));
1194 dag Li16 = (i32 (extloadi16 xoaddr:$src));
1195 dag ZELi16 = (i32 (zextloadi16 xoaddr:$src));
1196 dag ZELi16i64 = (i64 (zextloadi16 xoaddr:$src));
1197 dag SELi16 = (i32 (sextloadi16 xoaddr:$src));
1198 dag SELi16i64 = (i64 (sextloadi16 xoaddr:$src));
1200 dag Li32 = (i32 (load xoaddr:$src));
1203 // The following VSX instructions were introduced in Power ISA 2.07
1204 /* FIXME: if the operands are v2i64, these patterns will not match.
1205 we should define new patterns or otherwise match the same patterns
1206 when the elements are larger than i32.
1208 def HasP8Vector : Predicate<"PPCSubTarget->hasP8Vector()">;
1209 def HasDirectMove : Predicate<"PPCSubTarget->hasDirectMove()">;
1210 def NoP9Vector : Predicate<"!PPCSubTarget->hasP9Vector()">;
1211 let Predicates = [HasP8Vector] in {
1212 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
1213 let isCommutable = 1, UseVSXReg = 1 in {
1214 def XXLEQV : XX3Form<60, 186,
1215 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1216 "xxleqv $XT, $XA, $XB", IIC_VecGeneral,
1217 [(set v4i32:$XT, (vnot_ppc (xor v4i32:$XA, v4i32:$XB)))]>;
1218 def XXLNAND : XX3Form<60, 178,
1219 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1220 "xxlnand $XT, $XA, $XB", IIC_VecGeneral,
1221 [(set v4i32:$XT, (vnot_ppc (and v4i32:$XA,
1223 } // isCommutable, UseVSXReg
1225 def : Pat<(int_ppc_vsx_xxleqv v4i32:$A, v4i32:$B),
1228 let UseVSXReg = 1 in {
1229 def XXLORC : XX3Form<60, 170,
1230 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1231 "xxlorc $XT, $XA, $XB", IIC_VecGeneral,
1232 [(set v4i32:$XT, (or v4i32:$XA, (vnot_ppc v4i32:$XB)))]>;
1234 // VSX scalar loads introduced in ISA 2.07
1235 let mayLoad = 1, mayStore = 0 in {
1237 def LXSSPX : XX1Form_memOp<31, 524, (outs vssrc:$XT), (ins memrr:$src),
1238 "lxsspx $XT, $src", IIC_LdStLFD, []>;
1239 def LXSIWAX : XX1Form_memOp<31, 76, (outs vsfrc:$XT), (ins memrr:$src),
1240 "lxsiwax $XT, $src", IIC_LdStLFD, []>;
1241 def LXSIWZX : XX1Form_memOp<31, 12, (outs vsfrc:$XT), (ins memrr:$src),
1242 "lxsiwzx $XT, $src", IIC_LdStLFD, []>;
1244 // Please note let isPseudo = 1 is not part of class Pseudo<>. Missing it
1245 // would cause these Pseudos are not expanded in expandPostRAPseudos()
1246 let isPseudo = 1 in {
1247 // Pseudo instruction XFLOADf32 will be expanded to LXSSPX or LFSX later
1249 def XFLOADf32 : PseudoXFormMemOp<(outs vssrc:$XT), (ins memrr:$src),
1251 [(set f32:$XT, (load xoaddr:$src))]>;
1252 // Pseudo instruction LIWAX will be expanded to LXSIWAX or LFIWAX later
1253 def LIWAX : PseudoXFormMemOp<(outs vsfrc:$XT), (ins memrr:$src),
1255 [(set f64:$XT, (PPClfiwax xoaddr:$src))]>;
1256 // Pseudo instruction LIWZX will be expanded to LXSIWZX or LFIWZX later
1257 def LIWZX : PseudoXFormMemOp<(outs vsfrc:$XT), (ins memrr:$src),
1259 [(set f64:$XT, (PPClfiwzx xoaddr:$src))]>;
1263 // VSX scalar stores introduced in ISA 2.07
1264 let mayStore = 1, mayLoad = 0 in {
1266 def STXSSPX : XX1Form_memOp<31, 652, (outs), (ins vssrc:$XT, memrr:$dst),
1267 "stxsspx $XT, $dst", IIC_LdStSTFD, []>;
1268 def STXSIWX : XX1Form_memOp<31, 140, (outs), (ins vsfrc:$XT, memrr:$dst),
1269 "stxsiwx $XT, $dst", IIC_LdStSTFD, []>;
1271 // Please note let isPseudo = 1 is not part of class Pseudo<>. Missing it
1272 // would cause these Pseudos are not expanded in expandPostRAPseudos()
1273 let isPseudo = 1 in {
1274 // Pseudo instruction XFSTOREf32 will be expanded to STXSSPX or STFSX later
1276 def XFSTOREf32 : PseudoXFormMemOp<(outs), (ins vssrc:$XT, memrr:$dst),
1278 [(store f32:$XT, xoaddr:$dst)]>;
1279 // Pseudo instruction STIWX will be expanded to STXSIWX or STFIWX later
1280 def STIWX : PseudoXFormMemOp<(outs), (ins vsfrc:$XT, memrr:$dst),
1282 [(PPCstfiwx f64:$XT, xoaddr:$dst)]>;
1287 def : Pat<(f64 (extloadf32 xoaddr:$src)),
1288 (COPY_TO_REGCLASS (XFLOADf32 xoaddr:$src), VSFRC)>;
1289 def : Pat<(f32 (fpround (f64 (extloadf32 xoaddr:$src)))),
1290 (f32 (XFLOADf32 xoaddr:$src))>;
1291 def : Pat<(f64 (fpextend f32:$src)),
1292 (COPY_TO_REGCLASS $src, VSFRC)>;
1294 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLT)),
1295 (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1296 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULT)),
1297 (SELECT_VSSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1298 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLE)),
1299 (SELECT_VSSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1300 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULE)),
1301 (SELECT_VSSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1302 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETEQ)),
1303 (SELECT_VSSRC (CREQV $lhs, $rhs), $tval, $fval)>;
1304 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETGE)),
1305 (SELECT_VSSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1306 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETUGE)),
1307 (SELECT_VSSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1308 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETGT)),
1309 (SELECT_VSSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1310 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETUGT)),
1311 (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1312 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETNE)),
1313 (SELECT_VSSRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1315 let UseVSXReg = 1 in {
1316 // VSX Elementary Scalar FP arithmetic (SP)
1317 let isCommutable = 1 in {
1318 def XSADDSP : XX3Form<60, 0,
1319 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1320 "xsaddsp $XT, $XA, $XB", IIC_VecFP,
1321 [(set f32:$XT, (fadd f32:$XA, f32:$XB))]>;
1322 def XSMULSP : XX3Form<60, 16,
1323 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1324 "xsmulsp $XT, $XA, $XB", IIC_VecFP,
1325 [(set f32:$XT, (fmul f32:$XA, f32:$XB))]>;
1328 def XSDIVSP : XX3Form<60, 24,
1329 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1330 "xsdivsp $XT, $XA, $XB", IIC_FPDivS,
1331 [(set f32:$XT, (fdiv f32:$XA, f32:$XB))]>;
1332 def XSRESP : XX2Form<60, 26,
1333 (outs vssrc:$XT), (ins vssrc:$XB),
1334 "xsresp $XT, $XB", IIC_VecFP,
1335 [(set f32:$XT, (PPCfre f32:$XB))]>;
1336 def XSRSP : XX2Form<60, 281,
1337 (outs vssrc:$XT), (ins vsfrc:$XB),
1338 "xsrsp $XT, $XB", IIC_VecFP, []>;
1339 def XSSQRTSP : XX2Form<60, 11,
1340 (outs vssrc:$XT), (ins vssrc:$XB),
1341 "xssqrtsp $XT, $XB", IIC_FPSqrtS,
1342 [(set f32:$XT, (fsqrt f32:$XB))]>;
1343 def XSRSQRTESP : XX2Form<60, 10,
1344 (outs vssrc:$XT), (ins vssrc:$XB),
1345 "xsrsqrtesp $XT, $XB", IIC_VecFP,
1346 [(set f32:$XT, (PPCfrsqrte f32:$XB))]>;
1347 def XSSUBSP : XX3Form<60, 8,
1348 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1349 "xssubsp $XT, $XA, $XB", IIC_VecFP,
1350 [(set f32:$XT, (fsub f32:$XA, f32:$XB))]>;
1353 let BaseName = "XSMADDASP" in {
1354 let isCommutable = 1 in
1355 def XSMADDASP : XX3Form<60, 1,
1357 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1358 "xsmaddasp $XT, $XA, $XB", IIC_VecFP,
1359 [(set f32:$XT, (fma f32:$XA, f32:$XB, f32:$XTi))]>,
1360 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1362 let IsVSXFMAAlt = 1 in
1363 def XSMADDMSP : XX3Form<60, 9,
1365 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1366 "xsmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
1367 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1371 let BaseName = "XSMSUBASP" in {
1372 let isCommutable = 1 in
1373 def XSMSUBASP : XX3Form<60, 17,
1375 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1376 "xsmsubasp $XT, $XA, $XB", IIC_VecFP,
1377 [(set f32:$XT, (fma f32:$XA, f32:$XB,
1378 (fneg f32:$XTi)))]>,
1379 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1381 let IsVSXFMAAlt = 1 in
1382 def XSMSUBMSP : XX3Form<60, 25,
1384 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1385 "xsmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
1386 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1390 let BaseName = "XSNMADDASP" in {
1391 let isCommutable = 1 in
1392 def XSNMADDASP : XX3Form<60, 129,
1394 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1395 "xsnmaddasp $XT, $XA, $XB", IIC_VecFP,
1396 [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
1398 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1400 let IsVSXFMAAlt = 1 in
1401 def XSNMADDMSP : XX3Form<60, 137,
1403 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1404 "xsnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
1405 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1409 let BaseName = "XSNMSUBASP" in {
1410 let isCommutable = 1 in
1411 def XSNMSUBASP : XX3Form<60, 145,
1413 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1414 "xsnmsubasp $XT, $XA, $XB", IIC_VecFP,
1415 [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
1416 (fneg f32:$XTi))))]>,
1417 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1419 let IsVSXFMAAlt = 1 in
1420 def XSNMSUBMSP : XX3Form<60, 153,
1422 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1423 "xsnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
1424 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1428 // Single Precision Conversions (FP <-> INT)
1429 def XSCVSXDSP : XX2Form<60, 312,
1430 (outs vssrc:$XT), (ins vsfrc:$XB),
1431 "xscvsxdsp $XT, $XB", IIC_VecFP,
1432 [(set f32:$XT, (PPCfcfids f64:$XB))]>;
1433 def XSCVUXDSP : XX2Form<60, 296,
1434 (outs vssrc:$XT), (ins vsfrc:$XB),
1435 "xscvuxdsp $XT, $XB", IIC_VecFP,
1436 [(set f32:$XT, (PPCfcfidus f64:$XB))]>;
1438 // Conversions between vector and scalar single precision
1439 def XSCVDPSPN : XX2Form<60, 267, (outs vsrc:$XT), (ins vssrc:$XB),
1440 "xscvdpspn $XT, $XB", IIC_VecFP, []>;
1441 def XSCVSPDPN : XX2Form<60, 331, (outs vssrc:$XT), (ins vsrc:$XB),
1442 "xscvspdpn $XT, $XB", IIC_VecFP, []>;
1445 let Predicates = [IsLittleEndian] in {
1446 def : Pat<(f32 (PPCfcfids
1447 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S, 0)))))),
1448 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1449 def : Pat<(f32 (PPCfcfids
1450 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S, 1)))))),
1451 (f32 (XSCVSXDSP (COPY_TO_REGCLASS
1452 (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1453 def : Pat<(f32 (PPCfcfidus
1454 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S, 0)))))),
1455 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1456 def : Pat<(f32 (PPCfcfidus
1457 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S, 1)))))),
1458 (f32 (XSCVUXDSP (COPY_TO_REGCLASS
1459 (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1462 let Predicates = [IsBigEndian] in {
1463 def : Pat<(f32 (PPCfcfids
1464 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S, 0)))))),
1465 (f32 (XSCVSXDSP (COPY_TO_REGCLASS $S, VSFRC)))>;
1466 def : Pat<(f32 (PPCfcfids
1467 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S, 1)))))),
1468 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1469 def : Pat<(f32 (PPCfcfidus
1470 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S, 0)))))),
1471 (f32 (XSCVUXDSP (COPY_TO_REGCLASS $S, VSFRC)))>;
1472 def : Pat<(f32 (PPCfcfidus
1473 (f64 (PPCmtvsra (i64 (vector_extract v2i64:$S, 1)))))),
1474 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1477 // Instructions for converting float to i64 feeding a store.
1478 let Predicates = [NoP9Vector] in {
1479 def : Pat<(PPCstore_scal_int_from_vsr
1480 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xoaddr:$dst, 8),
1481 (STXSDX (XSCVDPSXDS f64:$src), xoaddr:$dst)>;
1482 def : Pat<(PPCstore_scal_int_from_vsr
1483 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xoaddr:$dst, 8),
1484 (STXSDX (XSCVDPUXDS f64:$src), xoaddr:$dst)>;
1487 // Instructions for converting float to i32 feeding a store.
1488 def : Pat<(PPCstore_scal_int_from_vsr
1489 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xoaddr:$dst, 4),
1490 (STIWX (XSCVDPSXWS f64:$src), xoaddr:$dst)>;
1491 def : Pat<(PPCstore_scal_int_from_vsr
1492 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xoaddr:$dst, 4),
1493 (STIWX (XSCVDPUXWS f64:$src), xoaddr:$dst)>;
1495 } // AddedComplexity = 400
1498 let UseVSXReg = 1, AddedComplexity = 400 in {
1499 let Predicates = [HasDirectMove] in {
1500 // VSX direct move instructions
1501 def MFVSRD : XX1_RS6_RD5_XO<31, 51, (outs g8rc:$rA), (ins vsfrc:$XT),
1502 "mfvsrd $rA, $XT", IIC_VecGeneral,
1503 [(set i64:$rA, (PPCmfvsr f64:$XT))]>,
1504 Requires<[In64BitMode]>;
1505 let isCodeGenOnly = 1 in
1506 def MFVRD : XX1_RS6_RD5_XO<31, 51, (outs g8rc:$rA), (ins vrrc:$XT),
1507 "mfvsrd $rA, $XT", IIC_VecGeneral,
1509 Requires<[In64BitMode]>;
1510 def MFVSRWZ : XX1_RS6_RD5_XO<31, 115, (outs gprc:$rA), (ins vsfrc:$XT),
1511 "mfvsrwz $rA, $XT", IIC_VecGeneral,
1512 [(set i32:$rA, (PPCmfvsr f64:$XT))]>;
1513 def MTVSRD : XX1_RS6_RD5_XO<31, 179, (outs vsfrc:$XT), (ins g8rc:$rA),
1514 "mtvsrd $XT, $rA", IIC_VecGeneral,
1515 [(set f64:$XT, (PPCmtvsra i64:$rA))]>,
1516 Requires<[In64BitMode]>;
1517 def MTVSRWA : XX1_RS6_RD5_XO<31, 211, (outs vsfrc:$XT), (ins gprc:$rA),
1518 "mtvsrwa $XT, $rA", IIC_VecGeneral,
1519 [(set f64:$XT, (PPCmtvsra i32:$rA))]>;
1520 def MTVSRWZ : XX1_RS6_RD5_XO<31, 243, (outs vsfrc:$XT), (ins gprc:$rA),
1521 "mtvsrwz $XT, $rA", IIC_VecGeneral,
1522 [(set f64:$XT, (PPCmtvsrz i32:$rA))]>;
1525 let Predicates = [IsISA3_0, HasDirectMove] in {
1526 def MTVSRWS: XX1_RS6_RD5_XO<31, 403, (outs vsrc:$XT), (ins gprc:$rA),
1527 "mtvsrws $XT, $rA", IIC_VecGeneral, []>;
1529 def MTVSRDD: XX1Form<31, 435, (outs vsrc:$XT), (ins g8rc_nox0:$rA, g8rc:$rB),
1530 "mtvsrdd $XT, $rA, $rB", IIC_VecGeneral,
1531 []>, Requires<[In64BitMode]>;
1533 def MFVSRLD: XX1_RS6_RD5_XO<31, 307, (outs g8rc:$rA), (ins vsrc:$XT),
1534 "mfvsrld $rA, $XT", IIC_VecGeneral,
1535 []>, Requires<[In64BitMode]>;
1537 } // IsISA3_0, HasDirectMove
1540 // We want to parse this from asm, but we don't want to emit this as it would
1541 // be emitted with a VSX reg. So leave Emit = 0 here.
1542 def : InstAlias<"mfvrd $rA, $XT",
1543 (MFVRD g8rc:$rA, vrrc:$XT), 0>;
1544 def : InstAlias<"mffprd $rA, $src",
1545 (MFVSRD g8rc:$rA, f8rc:$src)>;
1547 /* Direct moves of various widths from GPR's into VSR's. Each move lines
1548 the value up into element 0 (both BE and LE). Namely, entities smaller than
1549 a doubleword are shifted left and moved for BE. For LE, they're moved, then
1550 swapped to go into the least significant element of the VSR.
1556 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 56, 7));
1560 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 48, 15));
1564 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 32, 31));
1565 dag BE_DWORD_0 = (MTVSRD $A);
1567 dag LE_MTVSRW = (MTVSRD (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32));
1568 dag LE_WORD_1 = (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1569 LE_MTVSRW, sub_64));
1570 dag LE_WORD_0 = (XXPERMDI LE_WORD_1, LE_WORD_1, 2);
1571 dag LE_DWORD_1 = (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1572 BE_DWORD_0, sub_64));
1573 dag LE_DWORD_0 = (XXPERMDI LE_DWORD_1, LE_DWORD_1, 2);
1576 /* Patterns for extracting elements out of vectors. Integer elements are
1577 extracted using direct move operations. Patterns for extracting elements
1578 whose indices are not available at compile time are also provided with
1579 various _VARIABLE_ patterns.
1580 The numbering for the DAG's is for LE, but when used on BE, the correct
1581 LE element can just be used (i.e. LE_BYTE_2 == BE_BYTE_13).
1583 def VectorExtractions {
1584 // Doubleword extraction
1588 (XXPERMDI (COPY_TO_REGCLASS $S, VSRC),
1589 (COPY_TO_REGCLASS $S, VSRC), 2), sub_64));
1590 dag LE_DWORD_1 = (MFVSRD
1592 (v2i64 (COPY_TO_REGCLASS $S, VSRC)), sub_64));
1595 dag LE_WORD_0 = (MFVSRWZ (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64));
1596 dag LE_WORD_1 = (MFVSRWZ (EXTRACT_SUBREG (XXSLDWI $S, $S, 1), sub_64));
1597 dag LE_WORD_2 = (MFVSRWZ (EXTRACT_SUBREG
1598 (v2i64 (COPY_TO_REGCLASS $S, VSRC)), sub_64));
1599 dag LE_WORD_3 = (MFVSRWZ (EXTRACT_SUBREG (XXSLDWI $S, $S, 3), sub_64));
1601 // Halfword extraction
1602 dag LE_HALF_0 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 0, 48), sub_32));
1603 dag LE_HALF_1 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 48, 48), sub_32));
1604 dag LE_HALF_2 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 32, 48), sub_32));
1605 dag LE_HALF_3 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 16, 48), sub_32));
1606 dag LE_HALF_4 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 0, 48), sub_32));
1607 dag LE_HALF_5 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 48, 48), sub_32));
1608 dag LE_HALF_6 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 32, 48), sub_32));
1609 dag LE_HALF_7 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 16, 48), sub_32));
1612 dag LE_BYTE_0 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 0, 56), sub_32));
1613 dag LE_BYTE_1 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 56, 56), sub_32));
1614 dag LE_BYTE_2 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 48, 56), sub_32));
1615 dag LE_BYTE_3 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 40, 56), sub_32));
1616 dag LE_BYTE_4 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 32, 56), sub_32));
1617 dag LE_BYTE_5 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 24, 56), sub_32));
1618 dag LE_BYTE_6 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 16, 56), sub_32));
1619 dag LE_BYTE_7 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 8, 56), sub_32));
1620 dag LE_BYTE_8 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 0, 56), sub_32));
1621 dag LE_BYTE_9 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 56, 56), sub_32));
1622 dag LE_BYTE_10 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 48, 56), sub_32));
1623 dag LE_BYTE_11 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 40, 56), sub_32));
1624 dag LE_BYTE_12 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 32, 56), sub_32));
1625 dag LE_BYTE_13 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 24, 56), sub_32));
1626 dag LE_BYTE_14 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 16, 56), sub_32));
1627 dag LE_BYTE_15 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 8, 56), sub_32));
1629 /* Variable element number (BE and LE patterns must be specified separately)
1630 This is a rather involved process.
1632 Conceptually, this is how the move is accomplished:
1633 1. Identify which doubleword contains the element
1634 2. Shift in the VMX register so that the correct doubleword is correctly
1635 lined up for the MFVSRD
1636 3. Perform the move so that the element (along with some extra stuff)
1638 4. Right shift within the GPR so that the element is right-justified
1640 Of course, the index is an element number which has a different meaning
1641 on LE/BE so the patterns have to be specified separately.
1643 Note: The final result will be the element right-justified with high
1644 order bits being arbitrarily defined (namely, whatever was in the
1645 vector register to the left of the value originally).
1650 - For elements 0-7, we shift left by 8 bytes since they're on the right
1651 - For elements 8-15, we need not shift (shift left by zero bytes)
1652 This is accomplished by inverting the bits of the index and AND-ing
1653 with 0x8 (i.e. clearing all bits of the index and inverting bit 60).
1655 dag LE_VBYTE_PERM_VEC = (v16i8 (LVSL ZERO8, (ANDC8 (LI8 8), $Idx)));
1658 // - Now that we set up the shift amount, we shift in the VMX register
1659 dag LE_VBYTE_PERMUTE = (v16i8 (VPERM $S, $S, LE_VBYTE_PERM_VEC));
1662 // - The doubleword containing our element is moved to a GPR
1663 dag LE_MV_VBYTE = (MFVSRD
1665 (v2i64 (COPY_TO_REGCLASS LE_VBYTE_PERMUTE, VSRC)),
1669 - Truncate the element number to the range 0-7 (8-15 are symmetrical
1670 and out of range values are truncated accordingly)
1671 - Multiply by 8 as we need to shift right by the number of bits, not bytes
1672 - Shift right in the GPR by the calculated value
1674 dag LE_VBYTE_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 7), $Idx), 3, 60),
1676 dag LE_VARIABLE_BYTE = (EXTRACT_SUBREG (SRD LE_MV_VBYTE, LE_VBYTE_SHIFT),
1679 /* LE variable halfword
1681 - For elements 0-3, we shift left by 8 since they're on the right
1682 - For elements 4-7, we need not shift (shift left by zero bytes)
1683 Similarly to the byte pattern, we invert the bits of the index, but we
1684 AND with 0x4 (i.e. clear all bits of the index and invert bit 61).
1685 Of course, the shift is still by 8 bytes, so we must multiply by 2.
1687 dag LE_VHALF_PERM_VEC =
1688 (v16i8 (LVSL ZERO8, (RLDICR (ANDC8 (LI8 4), $Idx), 1, 62)));
1691 // - Now that we set up the shift amount, we shift in the VMX register
1692 dag LE_VHALF_PERMUTE = (v16i8 (VPERM $S, $S, LE_VHALF_PERM_VEC));
1695 // - The doubleword containing our element is moved to a GPR
1696 dag LE_MV_VHALF = (MFVSRD
1698 (v2i64 (COPY_TO_REGCLASS LE_VHALF_PERMUTE, VSRC)),
1702 - Truncate the element number to the range 0-3 (4-7 are symmetrical
1703 and out of range values are truncated accordingly)
1704 - Multiply by 16 as we need to shift right by the number of bits
1705 - Shift right in the GPR by the calculated value
1707 dag LE_VHALF_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 3), $Idx), 4, 59),
1709 dag LE_VARIABLE_HALF = (EXTRACT_SUBREG (SRD LE_MV_VHALF, LE_VHALF_SHIFT),
1714 - For elements 0-1, we shift left by 8 since they're on the right
1715 - For elements 2-3, we need not shift
1717 dag LE_VWORD_PERM_VEC = (v16i8 (LVSL ZERO8,
1718 (RLDICR (ANDC8 (LI8 2), $Idx), 2, 61)));
1721 // - Now that we set up the shift amount, we shift in the VMX register
1722 dag LE_VWORD_PERMUTE = (v16i8 (VPERM $S, $S, LE_VWORD_PERM_VEC));
1725 // - The doubleword containing our element is moved to a GPR
1726 dag LE_MV_VWORD = (MFVSRD
1728 (v2i64 (COPY_TO_REGCLASS LE_VWORD_PERMUTE, VSRC)),
1732 - Truncate the element number to the range 0-1 (2-3 are symmetrical
1733 and out of range values are truncated accordingly)
1734 - Multiply by 32 as we need to shift right by the number of bits
1735 - Shift right in the GPR by the calculated value
1737 dag LE_VWORD_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 1), $Idx), 5, 58),
1739 dag LE_VARIABLE_WORD = (EXTRACT_SUBREG (SRD LE_MV_VWORD, LE_VWORD_SHIFT),
1742 /* LE variable doubleword
1744 - For element 0, we shift left by 8 since it's on the right
1745 - For element 1, we need not shift
1747 dag LE_VDWORD_PERM_VEC = (v16i8 (LVSL ZERO8,
1748 (RLDICR (ANDC8 (LI8 1), $Idx), 3, 60)));
1751 // - Now that we set up the shift amount, we shift in the VMX register
1752 dag LE_VDWORD_PERMUTE = (v16i8 (VPERM $S, $S, LE_VDWORD_PERM_VEC));
1755 // - The doubleword containing our element is moved to a GPR
1756 // - Number 4. is not needed for the doubleword as the value is 64-bits
1757 dag LE_VARIABLE_DWORD =
1758 (MFVSRD (EXTRACT_SUBREG
1759 (v2i64 (COPY_TO_REGCLASS LE_VDWORD_PERMUTE, VSRC)),
1762 /* LE variable float
1763 - Shift the vector to line up the desired element to BE Word 0
1764 - Convert 32-bit float to a 64-bit single precision float
1766 dag LE_VFLOAT_PERM_VEC = (v16i8 (LVSL ZERO8,
1767 (RLDICR (XOR8 (LI8 3), $Idx), 2, 61)));
1768 dag LE_VFLOAT_PERMUTE = (VPERM $S, $S, LE_VFLOAT_PERM_VEC);
1769 dag LE_VARIABLE_FLOAT = (XSCVSPDPN LE_VFLOAT_PERMUTE);
1771 /* LE variable double
1772 Same as the LE doubleword except there is no move.
1774 dag LE_VDOUBLE_PERMUTE = (v16i8 (VPERM (v16i8 (COPY_TO_REGCLASS $S, VRRC)),
1775 (v16i8 (COPY_TO_REGCLASS $S, VRRC)),
1776 LE_VDWORD_PERM_VEC));
1777 dag LE_VARIABLE_DOUBLE = (COPY_TO_REGCLASS LE_VDOUBLE_PERMUTE, VSRC);
1780 The algorithm here is the same as the LE variable byte except:
1781 - The shift in the VMX register is by 0/8 for opposite element numbers so
1782 we simply AND the element number with 0x8
1783 - The order of elements after the move to GPR is reversed, so we invert
1784 the bits of the index prior to truncating to the range 0-7
1786 dag BE_VBYTE_PERM_VEC = (v16i8 (LVSL ZERO8, (ANDIo8 $Idx, 8)));
1787 dag BE_VBYTE_PERMUTE = (v16i8 (VPERM $S, $S, BE_VBYTE_PERM_VEC));
1788 dag BE_MV_VBYTE = (MFVSRD
1790 (v2i64 (COPY_TO_REGCLASS BE_VBYTE_PERMUTE, VSRC)),
1792 dag BE_VBYTE_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 7), $Idx), 3, 60),
1794 dag BE_VARIABLE_BYTE = (EXTRACT_SUBREG (SRD BE_MV_VBYTE, BE_VBYTE_SHIFT),
1797 /* BE variable halfword
1798 The algorithm here is the same as the LE variable halfword except:
1799 - The shift in the VMX register is by 0/8 for opposite element numbers so
1800 we simply AND the element number with 0x4 and multiply by 2
1801 - The order of elements after the move to GPR is reversed, so we invert
1802 the bits of the index prior to truncating to the range 0-3
1804 dag BE_VHALF_PERM_VEC = (v16i8 (LVSL ZERO8,
1805 (RLDICR (ANDIo8 $Idx, 4), 1, 62)));
1806 dag BE_VHALF_PERMUTE = (v16i8 (VPERM $S, $S, BE_VHALF_PERM_VEC));
1807 dag BE_MV_VHALF = (MFVSRD
1809 (v2i64 (COPY_TO_REGCLASS BE_VHALF_PERMUTE, VSRC)),
1811 dag BE_VHALF_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 3), $Idx), 4, 59),
1813 dag BE_VARIABLE_HALF = (EXTRACT_SUBREG (SRD BE_MV_VHALF, BE_VHALF_SHIFT),
1817 The algorithm is the same as the LE variable word except:
1818 - The shift in the VMX register happens for opposite element numbers
1819 - The order of elements after the move to GPR is reversed, so we invert
1820 the bits of the index prior to truncating to the range 0-1
1822 dag BE_VWORD_PERM_VEC = (v16i8 (LVSL ZERO8,
1823 (RLDICR (ANDIo8 $Idx, 2), 2, 61)));
1824 dag BE_VWORD_PERMUTE = (v16i8 (VPERM $S, $S, BE_VWORD_PERM_VEC));
1825 dag BE_MV_VWORD = (MFVSRD
1827 (v2i64 (COPY_TO_REGCLASS BE_VWORD_PERMUTE, VSRC)),
1829 dag BE_VWORD_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 1), $Idx), 5, 58),
1831 dag BE_VARIABLE_WORD = (EXTRACT_SUBREG (SRD BE_MV_VWORD, BE_VWORD_SHIFT),
1834 /* BE variable doubleword
1835 Same as the LE doubleword except we shift in the VMX register for opposite
1838 dag BE_VDWORD_PERM_VEC = (v16i8 (LVSL ZERO8,
1839 (RLDICR (ANDIo8 $Idx, 1), 3, 60)));
1840 dag BE_VDWORD_PERMUTE = (v16i8 (VPERM $S, $S, BE_VDWORD_PERM_VEC));
1841 dag BE_VARIABLE_DWORD =
1842 (MFVSRD (EXTRACT_SUBREG
1843 (v2i64 (COPY_TO_REGCLASS BE_VDWORD_PERMUTE, VSRC)),
1846 /* BE variable float
1847 - Shift the vector to line up the desired element to BE Word 0
1848 - Convert 32-bit float to a 64-bit single precision float
1850 dag BE_VFLOAT_PERM_VEC = (v16i8 (LVSL ZERO8, (RLDICR $Idx, 2, 61)));
1851 dag BE_VFLOAT_PERMUTE = (VPERM $S, $S, BE_VFLOAT_PERM_VEC);
1852 dag BE_VARIABLE_FLOAT = (XSCVSPDPN BE_VFLOAT_PERMUTE);
1854 /* BE variable double
1855 Same as the BE doubleword except there is no move.
1857 dag BE_VDOUBLE_PERMUTE = (v16i8 (VPERM (v16i8 (COPY_TO_REGCLASS $S, VRRC)),
1858 (v16i8 (COPY_TO_REGCLASS $S, VRRC)),
1859 BE_VDWORD_PERM_VEC));
1860 dag BE_VARIABLE_DOUBLE = (COPY_TO_REGCLASS BE_VDOUBLE_PERMUTE, VSRC);
1863 def NoP9Altivec : Predicate<"!PPCSubTarget->hasP9Altivec()">;
1864 let AddedComplexity = 400 in {
1865 // v4f32 scalar <-> vector conversions (BE)
1866 let Predicates = [IsBigEndian, HasP8Vector] in {
1867 def : Pat<(v4f32 (scalar_to_vector f32:$A)),
1868 (v4f32 (XSCVDPSPN $A))>;
1869 def : Pat<(f32 (vector_extract v4f32:$S, 0)),
1870 (f32 (XSCVSPDPN $S))>;
1871 def : Pat<(f32 (vector_extract v4f32:$S, 1)),
1872 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 1)))>;
1873 def : Pat<(f32 (vector_extract v4f32:$S, 2)),
1874 (f32 (XSCVSPDPN (XXPERMDI $S, $S, 2)))>;
1875 def : Pat<(f32 (vector_extract v4f32:$S, 3)),
1876 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 3)))>;
1877 def : Pat<(f32 (vector_extract v4f32:$S, i64:$Idx)),
1878 (f32 VectorExtractions.BE_VARIABLE_FLOAT)>;
1879 } // IsBigEndian, HasP8Vector
1881 // Variable index vector_extract for v2f64 does not require P8Vector
1882 let Predicates = [IsBigEndian, HasVSX] in
1883 def : Pat<(f64 (vector_extract v2f64:$S, i64:$Idx)),
1884 (f64 VectorExtractions.BE_VARIABLE_DOUBLE)>;
1886 let Predicates = [IsBigEndian, HasDirectMove] in {
1887 // v16i8 scalar <-> vector conversions (BE)
1888 def : Pat<(v16i8 (scalar_to_vector i32:$A)),
1889 (v16i8 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_BYTE_0, sub_64))>;
1890 def : Pat<(v8i16 (scalar_to_vector i32:$A)),
1891 (v8i16 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_HALF_0, sub_64))>;
1892 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
1893 (v4i32 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_WORD_0, sub_64))>;
1894 def : Pat<(v2i64 (scalar_to_vector i64:$A)),
1895 (v2i64 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_DWORD_0, sub_64))>;
1897 // v2i64 scalar <-> vector conversions (BE)
1898 def : Pat<(i64 (vector_extract v2i64:$S, 0)),
1899 (i64 VectorExtractions.LE_DWORD_1)>;
1900 def : Pat<(i64 (vector_extract v2i64:$S, 1)),
1901 (i64 VectorExtractions.LE_DWORD_0)>;
1902 def : Pat<(i64 (vector_extract v2i64:$S, i64:$Idx)),
1903 (i64 VectorExtractions.BE_VARIABLE_DWORD)>;
1904 } // IsBigEndian, HasDirectMove
1906 let Predicates = [IsBigEndian, HasDirectMove, NoP9Altivec] in {
1907 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
1908 (i32 VectorExtractions.LE_BYTE_15)>;
1909 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
1910 (i32 VectorExtractions.LE_BYTE_14)>;
1911 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
1912 (i32 VectorExtractions.LE_BYTE_13)>;
1913 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
1914 (i32 VectorExtractions.LE_BYTE_12)>;
1915 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
1916 (i32 VectorExtractions.LE_BYTE_11)>;
1917 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
1918 (i32 VectorExtractions.LE_BYTE_10)>;
1919 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
1920 (i32 VectorExtractions.LE_BYTE_9)>;
1921 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
1922 (i32 VectorExtractions.LE_BYTE_8)>;
1923 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
1924 (i32 VectorExtractions.LE_BYTE_7)>;
1925 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
1926 (i32 VectorExtractions.LE_BYTE_6)>;
1927 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
1928 (i32 VectorExtractions.LE_BYTE_5)>;
1929 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
1930 (i32 VectorExtractions.LE_BYTE_4)>;
1931 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
1932 (i32 VectorExtractions.LE_BYTE_3)>;
1933 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
1934 (i32 VectorExtractions.LE_BYTE_2)>;
1935 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
1936 (i32 VectorExtractions.LE_BYTE_1)>;
1937 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
1938 (i32 VectorExtractions.LE_BYTE_0)>;
1939 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
1940 (i32 VectorExtractions.BE_VARIABLE_BYTE)>;
1942 // v8i16 scalar <-> vector conversions (BE)
1943 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
1944 (i32 VectorExtractions.LE_HALF_7)>;
1945 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
1946 (i32 VectorExtractions.LE_HALF_6)>;
1947 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
1948 (i32 VectorExtractions.LE_HALF_5)>;
1949 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
1950 (i32 VectorExtractions.LE_HALF_4)>;
1951 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
1952 (i32 VectorExtractions.LE_HALF_3)>;
1953 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
1954 (i32 VectorExtractions.LE_HALF_2)>;
1955 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
1956 (i32 VectorExtractions.LE_HALF_1)>;
1957 def : Pat<(i32 (vector_extract v8i16:$S, 7)),
1958 (i32 VectorExtractions.LE_HALF_0)>;
1959 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
1960 (i32 VectorExtractions.BE_VARIABLE_HALF)>;
1962 // v4i32 scalar <-> vector conversions (BE)
1963 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
1964 (i32 VectorExtractions.LE_WORD_3)>;
1965 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
1966 (i32 VectorExtractions.LE_WORD_2)>;
1967 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
1968 (i32 VectorExtractions.LE_WORD_1)>;
1969 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
1970 (i32 VectorExtractions.LE_WORD_0)>;
1971 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
1972 (i32 VectorExtractions.BE_VARIABLE_WORD)>;
1973 } // IsBigEndian, HasDirectMove, NoP9Altivec
1975 // v4f32 scalar <-> vector conversions (LE)
1976 let Predicates = [IsLittleEndian, HasP8Vector] in {
1977 def : Pat<(v4f32 (scalar_to_vector f32:$A)),
1978 (v4f32 (XXSLDWI (XSCVDPSPN $A), (XSCVDPSPN $A), 1))>;
1979 def : Pat<(f32 (vector_extract v4f32:$S, 0)),
1980 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 3)))>;
1981 def : Pat<(f32 (vector_extract v4f32:$S, 1)),
1982 (f32 (XSCVSPDPN (XXPERMDI $S, $S, 2)))>;
1983 def : Pat<(f32 (vector_extract v4f32:$S, 2)),
1984 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 1)))>;
1985 def : Pat<(f32 (vector_extract v4f32:$S, 3)),
1986 (f32 (XSCVSPDPN $S))>;
1987 def : Pat<(f32 (vector_extract v4f32:$S, i64:$Idx)),
1988 (f32 VectorExtractions.LE_VARIABLE_FLOAT)>;
1989 } // IsLittleEndian, HasP8Vector
1991 // Variable index vector_extract for v2f64 does not require P8Vector
1992 let Predicates = [IsLittleEndian, HasVSX] in
1993 def : Pat<(f64 (vector_extract v2f64:$S, i64:$Idx)),
1994 (f64 VectorExtractions.LE_VARIABLE_DOUBLE)>;
1996 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x_be xoaddr:$src)), (LXVW4X xoaddr:$src)>;
1997 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x_be xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1999 // Variable index unsigned vector_extract on Power9
2000 let Predicates = [HasP9Altivec, IsLittleEndian] in {
2001 def : Pat<(i64 (anyext (i32 (vector_extract v16i8:$S, i64:$Idx)))),
2002 (VEXTUBRX $Idx, $S)>;
2004 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, i64:$Idx)))),
2005 (VEXTUHRX (RLWINM8 $Idx, 1, 28, 30), $S)>;
2006 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 0)))),
2007 (VEXTUHRX (LI8 0), $S)>;
2008 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 1)))),
2009 (VEXTUHRX (LI8 2), $S)>;
2010 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 2)))),
2011 (VEXTUHRX (LI8 4), $S)>;
2012 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 3)))),
2013 (VEXTUHRX (LI8 6), $S)>;
2014 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 4)))),
2015 (VEXTUHRX (LI8 8), $S)>;
2016 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 5)))),
2017 (VEXTUHRX (LI8 10), $S)>;
2018 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 6)))),
2019 (VEXTUHRX (LI8 12), $S)>;
2020 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 7)))),
2021 (VEXTUHRX (LI8 14), $S)>;
2023 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, i64:$Idx)))),
2024 (VEXTUWRX (RLWINM8 $Idx, 2, 28, 29), $S)>;
2025 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 0)))),
2026 (VEXTUWRX (LI8 0), $S)>;
2027 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 1)))),
2028 (VEXTUWRX (LI8 4), $S)>;
2029 // For extracting LE word 2, MFVSRWZ is better than VEXTUWRX
2030 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 2)))),
2031 (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
2032 (i32 VectorExtractions.LE_WORD_2), sub_32)>;
2033 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 3)))),
2034 (VEXTUWRX (LI8 12), $S)>;
2036 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, i64:$Idx)))),
2037 (EXTSW (VEXTUWRX (RLWINM8 $Idx, 2, 28, 29), $S))>;
2038 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 0)))),
2039 (EXTSW (VEXTUWRX (LI8 0), $S))>;
2040 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 1)))),
2041 (EXTSW (VEXTUWRX (LI8 4), $S))>;
2042 // For extracting LE word 2, MFVSRWZ is better than VEXTUWRX
2043 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 2)))),
2044 (EXTSW (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
2045 (i32 VectorExtractions.LE_WORD_2), sub_32))>;
2046 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 3)))),
2047 (EXTSW (VEXTUWRX (LI8 12), $S))>;
2049 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
2050 (i32 (EXTRACT_SUBREG (VEXTUBRX $Idx, $S), sub_32))>;
2051 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
2052 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 0), $S), sub_32))>;
2053 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
2054 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 1), $S), sub_32))>;
2055 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
2056 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 2), $S), sub_32))>;
2057 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
2058 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 3), $S), sub_32))>;
2059 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
2060 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 4), $S), sub_32))>;
2061 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
2062 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 5), $S), sub_32))>;
2063 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
2064 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 6), $S), sub_32))>;
2065 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
2066 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 7), $S), sub_32))>;
2067 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
2068 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 8), $S), sub_32))>;
2069 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
2070 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 9), $S), sub_32))>;
2071 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
2072 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 10), $S), sub_32))>;
2073 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
2074 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 11), $S), sub_32))>;
2075 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
2076 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 12), $S), sub_32))>;
2077 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
2078 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 13), $S), sub_32))>;
2079 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
2080 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 14), $S), sub_32))>;
2081 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
2082 (i32 (EXTRACT_SUBREG (VEXTUBRX (LI8 15), $S), sub_32))>;
2084 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
2085 (i32 (EXTRACT_SUBREG (VEXTUHRX
2086 (RLWINM8 $Idx, 1, 28, 30), $S), sub_32))>;
2087 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
2088 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 0), $S), sub_32))>;
2089 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
2090 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 2), $S), sub_32))>;
2091 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
2092 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 4), $S), sub_32))>;
2093 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
2094 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 6), $S), sub_32))>;
2095 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
2096 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 8), $S), sub_32))>;
2097 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
2098 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 10), $S), sub_32))>;
2099 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2100 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 12), $S), sub_32))>;
2101 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2102 (i32 (EXTRACT_SUBREG (VEXTUHRX (LI8 14), $S), sub_32))>;
2104 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
2105 (i32 (EXTRACT_SUBREG (VEXTUWRX
2106 (RLWINM8 $Idx, 2, 28, 29), $S), sub_32))>;
2107 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
2108 (i32 (EXTRACT_SUBREG (VEXTUWRX (LI8 0), $S), sub_32))>;
2109 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
2110 (i32 (EXTRACT_SUBREG (VEXTUWRX (LI8 4), $S), sub_32))>;
2111 // For extracting LE word 2, MFVSRWZ is better than VEXTUWRX
2112 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
2113 (i32 VectorExtractions.LE_WORD_2)>;
2114 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
2115 (i32 (EXTRACT_SUBREG (VEXTUWRX (LI8 12), $S), sub_32))>;
2118 let Predicates = [HasP9Altivec, IsBigEndian] in {
2119 def : Pat<(i64 (anyext (i32 (vector_extract v16i8:$S, i64:$Idx)))),
2120 (VEXTUBLX $Idx, $S)>;
2122 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, i64:$Idx)))),
2123 (VEXTUHLX (RLWINM8 $Idx, 1, 28, 30), $S)>;
2124 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 0)))),
2125 (VEXTUHLX (LI8 0), $S)>;
2126 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 1)))),
2127 (VEXTUHLX (LI8 2), $S)>;
2128 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 2)))),
2129 (VEXTUHLX (LI8 4), $S)>;
2130 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 3)))),
2131 (VEXTUHLX (LI8 6), $S)>;
2132 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 4)))),
2133 (VEXTUHLX (LI8 8), $S)>;
2134 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 5)))),
2135 (VEXTUHLX (LI8 10), $S)>;
2136 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 6)))),
2137 (VEXTUHLX (LI8 12), $S)>;
2138 def : Pat<(i64 (anyext (i32 (vector_extract v8i16:$S, 7)))),
2139 (VEXTUHLX (LI8 14), $S)>;
2141 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, i64:$Idx)))),
2142 (VEXTUWLX (RLWINM8 $Idx, 2, 28, 29), $S)>;
2143 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 0)))),
2144 (VEXTUWLX (LI8 0), $S)>;
2146 // For extracting BE word 1, MFVSRWZ is better than VEXTUWLX
2147 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 1)))),
2148 (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
2149 (i32 VectorExtractions.LE_WORD_2), sub_32)>;
2150 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 2)))),
2151 (VEXTUWLX (LI8 8), $S)>;
2152 def : Pat<(i64 (zext (i32 (vector_extract v4i32:$S, 3)))),
2153 (VEXTUWLX (LI8 12), $S)>;
2155 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, i64:$Idx)))),
2156 (EXTSW (VEXTUWLX (RLWINM8 $Idx, 2, 28, 29), $S))>;
2157 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 0)))),
2158 (EXTSW (VEXTUWLX (LI8 0), $S))>;
2159 // For extracting BE word 1, MFVSRWZ is better than VEXTUWLX
2160 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 1)))),
2161 (EXTSW (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
2162 (i32 VectorExtractions.LE_WORD_2), sub_32))>;
2163 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 2)))),
2164 (EXTSW (VEXTUWLX (LI8 8), $S))>;
2165 def : Pat<(i64 (sext (i32 (vector_extract v4i32:$S, 3)))),
2166 (EXTSW (VEXTUWLX (LI8 12), $S))>;
2168 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
2169 (i32 (EXTRACT_SUBREG (VEXTUBLX $Idx, $S), sub_32))>;
2170 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
2171 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 0), $S), sub_32))>;
2172 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
2173 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 1), $S), sub_32))>;
2174 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
2175 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 2), $S), sub_32))>;
2176 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
2177 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 3), $S), sub_32))>;
2178 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
2179 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 4), $S), sub_32))>;
2180 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
2181 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 5), $S), sub_32))>;
2182 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
2183 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 6), $S), sub_32))>;
2184 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
2185 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 7), $S), sub_32))>;
2186 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
2187 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 8), $S), sub_32))>;
2188 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
2189 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 9), $S), sub_32))>;
2190 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
2191 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 10), $S), sub_32))>;
2192 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
2193 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 11), $S), sub_32))>;
2194 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
2195 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 12), $S), sub_32))>;
2196 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
2197 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 13), $S), sub_32))>;
2198 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
2199 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 14), $S), sub_32))>;
2200 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
2201 (i32 (EXTRACT_SUBREG (VEXTUBLX (LI8 15), $S), sub_32))>;
2203 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
2204 (i32 (EXTRACT_SUBREG (VEXTUHLX
2205 (RLWINM8 $Idx, 1, 28, 30), $S), sub_32))>;
2206 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
2207 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 0), $S), sub_32))>;
2208 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
2209 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 2), $S), sub_32))>;
2210 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
2211 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 4), $S), sub_32))>;
2212 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
2213 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 6), $S), sub_32))>;
2214 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
2215 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 8), $S), sub_32))>;
2216 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
2217 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 10), $S), sub_32))>;
2218 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2219 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 12), $S), sub_32))>;
2220 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2221 (i32 (EXTRACT_SUBREG (VEXTUHLX (LI8 14), $S), sub_32))>;
2223 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
2224 (i32 (EXTRACT_SUBREG (VEXTUWLX
2225 (RLWINM8 $Idx, 2, 28, 29), $S), sub_32))>;
2226 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
2227 (i32 (EXTRACT_SUBREG (VEXTUWLX (LI8 0), $S), sub_32))>;
2228 // For extracting BE word 1, MFVSRWZ is better than VEXTUWLX
2229 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
2230 (i32 VectorExtractions.LE_WORD_2)>;
2231 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
2232 (i32 (EXTRACT_SUBREG (VEXTUWLX (LI8 8), $S), sub_32))>;
2233 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
2234 (i32 (EXTRACT_SUBREG (VEXTUWLX (LI8 12), $S), sub_32))>;
2237 let Predicates = [IsLittleEndian, HasDirectMove] in {
2238 // v16i8 scalar <-> vector conversions (LE)
2239 def : Pat<(v16i8 (scalar_to_vector i32:$A)),
2240 (v16i8 (COPY_TO_REGCLASS MovesToVSR.LE_WORD_0, VSRC))>;
2241 def : Pat<(v8i16 (scalar_to_vector i32:$A)),
2242 (v8i16 (COPY_TO_REGCLASS MovesToVSR.LE_WORD_0, VSRC))>;
2243 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
2244 (v4i32 MovesToVSR.LE_WORD_0)>;
2245 def : Pat<(v2i64 (scalar_to_vector i64:$A)),
2246 (v2i64 MovesToVSR.LE_DWORD_0)>;
2247 // v2i64 scalar <-> vector conversions (LE)
2248 def : Pat<(i64 (vector_extract v2i64:$S, 0)),
2249 (i64 VectorExtractions.LE_DWORD_0)>;
2250 def : Pat<(i64 (vector_extract v2i64:$S, 1)),
2251 (i64 VectorExtractions.LE_DWORD_1)>;
2252 def : Pat<(i64 (vector_extract v2i64:$S, i64:$Idx)),
2253 (i64 VectorExtractions.LE_VARIABLE_DWORD)>;
2254 } // IsLittleEndian, HasDirectMove
2256 let Predicates = [IsLittleEndian, HasDirectMove, NoP9Altivec] in {
2257 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
2258 (i32 VectorExtractions.LE_BYTE_0)>;
2259 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
2260 (i32 VectorExtractions.LE_BYTE_1)>;
2261 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
2262 (i32 VectorExtractions.LE_BYTE_2)>;
2263 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
2264 (i32 VectorExtractions.LE_BYTE_3)>;
2265 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
2266 (i32 VectorExtractions.LE_BYTE_4)>;
2267 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
2268 (i32 VectorExtractions.LE_BYTE_5)>;
2269 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
2270 (i32 VectorExtractions.LE_BYTE_6)>;
2271 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
2272 (i32 VectorExtractions.LE_BYTE_7)>;
2273 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
2274 (i32 VectorExtractions.LE_BYTE_8)>;
2275 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
2276 (i32 VectorExtractions.LE_BYTE_9)>;
2277 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
2278 (i32 VectorExtractions.LE_BYTE_10)>;
2279 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
2280 (i32 VectorExtractions.LE_BYTE_11)>;
2281 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
2282 (i32 VectorExtractions.LE_BYTE_12)>;
2283 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
2284 (i32 VectorExtractions.LE_BYTE_13)>;
2285 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
2286 (i32 VectorExtractions.LE_BYTE_14)>;
2287 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
2288 (i32 VectorExtractions.LE_BYTE_15)>;
2289 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
2290 (i32 VectorExtractions.LE_VARIABLE_BYTE)>;
2292 // v8i16 scalar <-> vector conversions (LE)
2293 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
2294 (i32 VectorExtractions.LE_HALF_0)>;
2295 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
2296 (i32 VectorExtractions.LE_HALF_1)>;
2297 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
2298 (i32 VectorExtractions.LE_HALF_2)>;
2299 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
2300 (i32 VectorExtractions.LE_HALF_3)>;
2301 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
2302 (i32 VectorExtractions.LE_HALF_4)>;
2303 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
2304 (i32 VectorExtractions.LE_HALF_5)>;
2305 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
2306 (i32 VectorExtractions.LE_HALF_6)>;
2307 def : Pat<(i32 (vector_extract v8i16:$S, 7)),
2308 (i32 VectorExtractions.LE_HALF_7)>;
2309 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
2310 (i32 VectorExtractions.LE_VARIABLE_HALF)>;
2312 // v4i32 scalar <-> vector conversions (LE)
2313 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
2314 (i32 VectorExtractions.LE_WORD_0)>;
2315 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
2316 (i32 VectorExtractions.LE_WORD_1)>;
2317 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
2318 (i32 VectorExtractions.LE_WORD_2)>;
2319 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
2320 (i32 VectorExtractions.LE_WORD_3)>;
2321 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
2322 (i32 VectorExtractions.LE_VARIABLE_WORD)>;
2323 } // IsLittleEndian, HasDirectMove, NoP9Altivec
2325 let Predicates = [HasDirectMove, HasVSX] in {
2326 // bitconvert f32 -> i32
2327 // (convert to 32-bit fp single, shift right 1 word, move to GPR)
2328 def : Pat<(i32 (bitconvert f32:$S)),
2329 (i32 (MFVSRWZ (EXTRACT_SUBREG
2330 (XXSLDWI (XSCVDPSPN $S), (XSCVDPSPN $S), 3),
2332 // bitconvert i32 -> f32
2333 // (move to FPR, shift left 1 word, convert to 64-bit fp single)
2334 def : Pat<(f32 (bitconvert i32:$A)),
2336 (XXSLDWI MovesToVSR.LE_WORD_1, MovesToVSR.LE_WORD_1, 1)))>;
2338 // bitconvert f64 -> i64
2339 // (move to GPR, nothing else needed)
2340 def : Pat<(i64 (bitconvert f64:$S)),
2343 // bitconvert i64 -> f64
2344 // (move to FPR, nothing else needed)
2345 def : Pat<(f64 (bitconvert i64:$S)),
2349 // Materialize a zero-vector of long long
2350 def : Pat<(v2i64 immAllZerosV),
2355 dag F32_TO_BE_WORD1 = (v4f32 (XXSLDWI (XSCVDPSPN $B), (XSCVDPSPN $B), 3));
2356 dag I32_TO_BE_WORD1 = (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC);
2359 // The following VSX instructions were introduced in Power ISA 3.0
2360 def HasP9Vector : Predicate<"PPCSubTarget->hasP9Vector()">;
2361 let AddedComplexity = 400, Predicates = [HasP9Vector] in {
2363 // [PO VRT XO VRB XO /]
2364 class X_VT5_XO5_VB5<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2366 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vrrc:$vT), (ins vrrc:$vB),
2367 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
2369 // [PO VRT XO VRB XO RO], Round to Odd version of [PO VRT XO VRB XO /]
2370 class X_VT5_XO5_VB5_Ro<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2372 : X_VT5_XO5_VB5<opcode, xo2, xo, opc, pattern>, isDOT;
2374 // [PO VRT XO VRB XO /], but the VRB is only used the left 64 bits (or less),
2375 // So we use different operand class for VRB
2376 class X_VT5_XO5_VB5_TyVB<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2377 RegisterOperand vbtype, list<dag> pattern>
2378 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vrrc:$vT), (ins vbtype:$vB),
2379 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
2381 // [PO VRT XO VRB XO /]
2382 class X_VT5_XO5_VB5_VSFR<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2384 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vfrc:$vT), (ins vrrc:$vB),
2385 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
2387 // [PO VRT XO VRB XO RO], Round to Odd version of [PO VRT XO VRB XO /]
2388 class X_VT5_XO5_VB5_VSFR_Ro<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2390 : X_VT5_XO5_VB5_VSFR<opcode, xo2, xo, opc, pattern>, isDOT;
2392 let UseVSXReg = 1 in {
2393 // [PO T XO B XO BX /]
2394 class XX2_RT5_XO5_XB6<bits<6> opcode, bits<5> xo2, bits<9> xo, string opc,
2396 : XX2_RD5_XO5_RS6<opcode, xo2, xo, (outs g8rc:$rT), (ins vsfrc:$XB),
2397 !strconcat(opc, " $rT, $XB"), IIC_VecFP, pattern>;
2399 // [PO T XO B XO BX TX]
2400 class XX2_XT6_XO5_XB6<bits<6> opcode, bits<5> xo2, bits<9> xo, string opc,
2401 RegisterOperand vtype, list<dag> pattern>
2402 : XX2_RD6_XO5_RS6<opcode, xo2, xo, (outs vtype:$XT), (ins vtype:$XB),
2403 !strconcat(opc, " $XT, $XB"), IIC_VecFP, pattern>;
2405 // [PO T A B XO AX BX TX], src and dest register use different operand class
2406 class XX3_XT5_XA5_XB5<bits<6> opcode, bits<8> xo, string opc,
2407 RegisterOperand xty, RegisterOperand aty, RegisterOperand bty,
2408 InstrItinClass itin, list<dag> pattern>
2409 : XX3Form<opcode, xo, (outs xty:$XT), (ins aty:$XA, bty:$XB),
2410 !strconcat(opc, " $XT, $XA, $XB"), itin, pattern>;
2413 // [PO VRT VRA VRB XO /]
2414 class X_VT5_VA5_VB5<bits<6> opcode, bits<10> xo, string opc,
2416 : XForm_1<opcode, xo, (outs vrrc:$vT), (ins vrrc:$vA, vrrc:$vB),
2417 !strconcat(opc, " $vT, $vA, $vB"), IIC_VecFP, pattern>;
2419 // [PO VRT VRA VRB XO RO], Round to Odd version of [PO VRT VRA VRB XO /]
2420 class X_VT5_VA5_VB5_Ro<bits<6> opcode, bits<10> xo, string opc,
2422 : X_VT5_VA5_VB5<opcode, xo, opc, pattern>, isDOT;
2424 // [PO VRT VRA VRB XO /]
2425 class X_VT5_VA5_VB5_FMA<bits<6> opcode, bits<10> xo, string opc,
2427 : XForm_1<opcode, xo, (outs vrrc:$vT), (ins vrrc:$vTi, vrrc:$vA, vrrc:$vB),
2428 !strconcat(opc, " $vT, $vA, $vB"), IIC_VecFP, pattern>,
2429 RegConstraint<"$vTi = $vT">, NoEncode<"$vTi">;
2431 // [PO VRT VRA VRB XO RO], Round to Odd version of [PO VRT VRA VRB XO /]
2432 class X_VT5_VA5_VB5_FMA_Ro<bits<6> opcode, bits<10> xo, string opc,
2434 : X_VT5_VA5_VB5_FMA<opcode, xo, opc, pattern>, isDOT;
2436 //===--------------------------------------------------------------------===//
2437 // Quad-Precision Scalar Move Instructions:
2440 def XSCPSGNQP : X_VT5_VA5_VB5<63, 100, "xscpsgnqp",
2442 (fcopysign f128:$vB, f128:$vA))]>;
2444 // Absolute/Negative-Absolute/Negate
2445 def XSABSQP : X_VT5_XO5_VB5<63, 0, 804, "xsabsqp",
2446 [(set f128:$vT, (fabs f128:$vB))]>;
2447 def XSNABSQP : X_VT5_XO5_VB5<63, 8, 804, "xsnabsqp",
2448 [(set f128:$vT, (fneg (fabs f128:$vB)))]>;
2449 def XSNEGQP : X_VT5_XO5_VB5<63, 16, 804, "xsnegqp",
2450 [(set f128:$vT, (fneg f128:$vB))]>;
2452 //===--------------------------------------------------------------------===//
2453 // Quad-Precision Scalar Floating-Point Arithmetic Instructions:
2455 // Add/Divide/Multiply/Subtract
2456 let isCommutable = 1 in {
2457 def XSADDQP : X_VT5_VA5_VB5 <63, 4, "xsaddqp",
2458 [(set f128:$vT, (fadd f128:$vA, f128:$vB))]>;
2459 def XSADDQPO : X_VT5_VA5_VB5_Ro<63, 4, "xsaddqpo",
2461 (int_ppc_addf128_round_to_odd
2462 f128:$vA, f128:$vB))]>;
2463 def XSMULQP : X_VT5_VA5_VB5 <63, 36, "xsmulqp",
2464 [(set f128:$vT, (fmul f128:$vA, f128:$vB))]>;
2465 def XSMULQPO : X_VT5_VA5_VB5_Ro<63, 36, "xsmulqpo",
2467 (int_ppc_mulf128_round_to_odd
2468 f128:$vA, f128:$vB))]>;
2471 def XSSUBQP : X_VT5_VA5_VB5 <63, 516, "xssubqp" ,
2472 [(set f128:$vT, (fsub f128:$vA, f128:$vB))]>;
2473 def XSSUBQPO : X_VT5_VA5_VB5_Ro<63, 516, "xssubqpo",
2475 (int_ppc_subf128_round_to_odd
2476 f128:$vA, f128:$vB))]>;
2477 def XSDIVQP : X_VT5_VA5_VB5 <63, 548, "xsdivqp",
2478 [(set f128:$vT, (fdiv f128:$vA, f128:$vB))]>;
2479 def XSDIVQPO : X_VT5_VA5_VB5_Ro<63, 548, "xsdivqpo",
2481 (int_ppc_divf128_round_to_odd
2482 f128:$vA, f128:$vB))]>;
2485 def XSSQRTQP : X_VT5_XO5_VB5 <63, 27, 804, "xssqrtqp",
2486 [(set f128:$vT, (fsqrt f128:$vB))]>;
2487 def XSSQRTQPO : X_VT5_XO5_VB5_Ro<63, 27, 804, "xssqrtqpo",
2489 (int_ppc_sqrtf128_round_to_odd f128:$vB))]>;
2491 // (Negative) Multiply-{Add/Subtract}
2492 def XSMADDQP : X_VT5_VA5_VB5_FMA <63, 388, "xsmaddqp",
2494 (fma f128:$vA, f128:$vB,
2497 def XSMADDQPO : X_VT5_VA5_VB5_FMA_Ro<63, 388, "xsmaddqpo",
2499 (int_ppc_fmaf128_round_to_odd
2500 f128:$vA,f128:$vB,f128:$vTi))]>;
2502 def XSMSUBQP : X_VT5_VA5_VB5_FMA <63, 420, "xsmsubqp" ,
2504 (fma f128:$vA, f128:$vB,
2505 (fneg f128:$vTi)))]>;
2506 def XSMSUBQPO : X_VT5_VA5_VB5_FMA_Ro<63, 420, "xsmsubqpo" ,
2508 (int_ppc_fmaf128_round_to_odd
2509 f128:$vA, f128:$vB, (fneg f128:$vTi)))]>;
2510 def XSNMADDQP : X_VT5_VA5_VB5_FMA <63, 452, "xsnmaddqp",
2512 (fneg (fma f128:$vA, f128:$vB,
2514 def XSNMADDQPO: X_VT5_VA5_VB5_FMA_Ro<63, 452, "xsnmaddqpo",
2516 (fneg (int_ppc_fmaf128_round_to_odd
2517 f128:$vA, f128:$vB, f128:$vTi)))]>;
2518 def XSNMSUBQP : X_VT5_VA5_VB5_FMA <63, 484, "xsnmsubqp",
2520 (fneg (fma f128:$vA, f128:$vB,
2521 (fneg f128:$vTi))))]>;
2522 def XSNMSUBQPO: X_VT5_VA5_VB5_FMA_Ro<63, 484, "xsnmsubqpo",
2524 (fneg (int_ppc_fmaf128_round_to_odd
2525 f128:$vA, f128:$vB, (fneg f128:$vTi))))]>;
2527 // Additional fnmsub patterns: -a*c + b == -(a*c - b)
2528 def : Pat<(fma (fneg f128:$A), f128:$C, f128:$B), (XSNMSUBQP $B, $C, $A)>;
2529 def : Pat<(fma f128:$A, (fneg f128:$C), f128:$B), (XSNMSUBQP $B, $C, $A)>;
2531 //===--------------------------------------------------------------------===//
2532 // Quad/Double-Precision Compare Instructions:
2534 // [PO BF // VRA VRB XO /]
2535 class X_BF3_VA5_VB5<bits<6> opcode, bits<10> xo, string opc,
2537 : XForm_17<opcode, xo, (outs crrc:$crD), (ins vrrc:$VA, vrrc:$VB),
2538 !strconcat(opc, " $crD, $VA, $VB"), IIC_FPCompare> {
2539 let Pattern = pattern;
2542 // QP Compare Ordered/Unordered
2543 def XSCMPOQP : X_BF3_VA5_VB5<63, 132, "xscmpoqp", []>;
2544 def XSCMPUQP : X_BF3_VA5_VB5<63, 644, "xscmpuqp", []>;
2546 // DP/QP Compare Exponents
2547 def XSCMPEXPDP : XX3Form_1<60, 59,
2548 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
2549 "xscmpexpdp $crD, $XA, $XB", IIC_FPCompare, []>,
2551 def XSCMPEXPQP : X_BF3_VA5_VB5<63, 164, "xscmpexpqp", []>;
2553 // DP Compare ==, >=, >, !=
2554 // Use vsrc for XT, because the entire register of XT is set.
2555 // XT.dword[1] = 0x0000_0000_0000_0000
2556 def XSCMPEQDP : XX3_XT5_XA5_XB5<60, 3, "xscmpeqdp", vsrc, vsfrc, vsfrc,
2558 def XSCMPGEDP : XX3_XT5_XA5_XB5<60, 19, "xscmpgedp", vsrc, vsfrc, vsfrc,
2560 def XSCMPGTDP : XX3_XT5_XA5_XB5<60, 11, "xscmpgtdp", vsrc, vsfrc, vsfrc,
2563 //===--------------------------------------------------------------------===//
2564 // Quad-Precision Floating-Point Conversion Instructions:
2567 def XSCVDPQP : X_VT5_XO5_VB5_TyVB<63, 22, 836, "xscvdpqp", vfrc,
2568 [(set f128:$vT, (fpextend f64:$vB))]>;
2570 // Round & Convert QP -> DP (dword[1] is set to zero)
2571 def XSCVQPDP : X_VT5_XO5_VB5_VSFR<63, 20, 836, "xscvqpdp" , []>;
2572 def XSCVQPDPO : X_VT5_XO5_VB5_VSFR_Ro<63, 20, 836, "xscvqpdpo",
2574 (int_ppc_truncf128_round_to_odd
2577 // Truncate & Convert QP -> (Un)Signed (D)Word (dword[1] is set to zero)
2578 def XSCVQPSDZ : X_VT5_XO5_VB5<63, 25, 836, "xscvqpsdz", []>;
2579 def XSCVQPSWZ : X_VT5_XO5_VB5<63, 9, 836, "xscvqpswz", []>;
2580 def XSCVQPUDZ : X_VT5_XO5_VB5<63, 17, 836, "xscvqpudz", []>;
2581 def XSCVQPUWZ : X_VT5_XO5_VB5<63, 1, 836, "xscvqpuwz", []>;
2583 // Convert (Un)Signed DWord -> QP.
2584 def XSCVSDQP : X_VT5_XO5_VB5_TyVB<63, 10, 836, "xscvsdqp", vfrc, []>;
2585 def : Pat<(f128 (sint_to_fp i64:$src)),
2586 (f128 (XSCVSDQP (COPY_TO_REGCLASS $src, VFRC)))>;
2587 def : Pat<(f128 (sint_to_fp (i64 (PPCmfvsr f64:$src)))),
2588 (f128 (XSCVSDQP $src))>;
2589 def : Pat<(f128 (sint_to_fp (i32 (PPCmfvsr f64:$src)))),
2590 (f128 (XSCVSDQP (VEXTSW2Ds $src)))>;
2592 def XSCVUDQP : X_VT5_XO5_VB5_TyVB<63, 2, 836, "xscvudqp", vfrc, []>;
2593 def : Pat<(f128 (uint_to_fp i64:$src)),
2594 (f128 (XSCVUDQP (COPY_TO_REGCLASS $src, VFRC)))>;
2595 def : Pat<(f128 (uint_to_fp (i64 (PPCmfvsr f64:$src)))),
2596 (f128 (XSCVUDQP $src))>;
2598 // Convert (Un)Signed Word -> QP.
2599 def : Pat<(f128 (sint_to_fp i32:$src)),
2600 (f128 (XSCVSDQP (MTVSRWA $src)))>;
2601 def : Pat<(f128 (sint_to_fp (i32 (load xoaddr:$src)))),
2602 (f128 (XSCVSDQP (LIWAX xoaddr:$src)))>;
2603 def : Pat<(f128 (uint_to_fp i32:$src)),
2604 (f128 (XSCVUDQP (MTVSRWZ $src)))>;
2605 def : Pat<(f128 (uint_to_fp (i32 (load xoaddr:$src)))),
2606 (f128 (XSCVUDQP (LIWZX xoaddr:$src)))>;
2608 let UseVSXReg = 1 in {
2609 //===--------------------------------------------------------------------===//
2610 // Round to Floating-Point Integer Instructions
2612 // (Round &) Convert DP <-> HP
2613 // Note! xscvdphp's src and dest register both use the left 64 bits, so we use
2614 // vsfrc for src and dest register. xscvhpdp's src only use the left 16 bits,
2615 // but we still use vsfrc for it.
2616 def XSCVDPHP : XX2_XT6_XO5_XB6<60, 17, 347, "xscvdphp", vsfrc, []>;
2617 def XSCVHPDP : XX2_XT6_XO5_XB6<60, 16, 347, "xscvhpdp", vsfrc, []>;
2620 def XVCVHPSP : XX2_XT6_XO5_XB6<60, 24, 475, "xvcvhpsp", vsrc, []>;
2621 def XVCVSPHP : XX2_XT6_XO5_XB6<60, 25, 475, "xvcvsphp", vsrc,
2623 (int_ppc_vsx_xvcvsphp v4f32:$XB))]>;
2627 // Pattern for matching Vector HP -> Vector SP intrinsic. Defined as a
2628 // separate pattern so that it can convert the input register class from
2629 // VRRC(v8i16) to VSRC.
2630 def : Pat<(v4f32 (int_ppc_vsx_xvcvhpsp v8i16:$A)),
2631 (v4f32 (XVCVHPSP (COPY_TO_REGCLASS $A, VSRC)))>;
2633 class Z23_VT5_R1_VB5_RMC2_EX1<bits<6> opcode, bits<8> xo, bit ex, string opc,
2635 : Z23Form_8<opcode, xo,
2636 (outs vrrc:$vT), (ins u1imm:$r, vrrc:$vB, u2imm:$rmc),
2637 !strconcat(opc, " $r, $vT, $vB, $rmc"), IIC_VecFP, pattern> {
2641 // Round to Quad-Precision Integer [with Inexact]
2642 def XSRQPI : Z23_VT5_R1_VB5_RMC2_EX1<63, 5, 0, "xsrqpi" , []>;
2643 def XSRQPIX : Z23_VT5_R1_VB5_RMC2_EX1<63, 5, 1, "xsrqpix", []>;
2645 // Use current rounding mode
2646 def : Pat<(f128 (fnearbyint f128:$vB)), (f128 (XSRQPI 0, $vB, 3))>;
2647 // Round to nearest, ties away from zero
2648 def : Pat<(f128 (fround f128:$vB)), (f128 (XSRQPI 0, $vB, 0))>;
2649 // Round towards Zero
2650 def : Pat<(f128 (ftrunc f128:$vB)), (f128 (XSRQPI 1, $vB, 1))>;
2651 // Round towards +Inf
2652 def : Pat<(f128 (fceil f128:$vB)), (f128 (XSRQPI 1, $vB, 2))>;
2653 // Round towards -Inf
2654 def : Pat<(f128 (ffloor f128:$vB)), (f128 (XSRQPI 1, $vB, 3))>;
2656 // Use current rounding mode, [with Inexact]
2657 def : Pat<(f128 (frint f128:$vB)), (f128 (XSRQPIX 0, $vB, 3))>;
2659 // Round Quad-Precision to Double-Extended Precision (fp80)
2660 def XSRQPXP : Z23_VT5_R1_VB5_RMC2_EX1<63, 37, 0, "xsrqpxp", []>;
2662 //===--------------------------------------------------------------------===//
2663 // Insert/Extract Instructions
2665 // Insert Exponent DP/QP
2666 // XT NOTE: XT.dword[1] = 0xUUUU_UUUU_UUUU_UUUU
2667 def XSIEXPDP : XX1Form <60, 918, (outs vsrc:$XT), (ins g8rc:$rA, g8rc:$rB),
2668 "xsiexpdp $XT, $rA, $rB", IIC_VecFP, []>, UseVSXReg;
2669 // vB NOTE: only vB.dword[0] is used, that's why we don't use
2670 // X_VT5_VA5_VB5 form
2671 def XSIEXPQP : XForm_18<63, 868, (outs vrrc:$vT), (ins vrrc:$vA, vsfrc:$vB),
2672 "xsiexpqp $vT, $vA, $vB", IIC_VecFP, []>;
2674 // Extract Exponent/Significand DP/QP
2675 def XSXEXPDP : XX2_RT5_XO5_XB6<60, 0, 347, "xsxexpdp", []>;
2676 def XSXSIGDP : XX2_RT5_XO5_XB6<60, 1, 347, "xsxsigdp", []>;
2678 def XSXEXPQP : X_VT5_XO5_VB5 <63, 2, 804, "xsxexpqp", []>;
2679 def XSXSIGQP : X_VT5_XO5_VB5 <63, 18, 804, "xsxsigqp", []>;
2681 // Vector Insert Word
2682 let UseVSXReg = 1 in {
2683 // XB NOTE: Only XB.dword[1] is used, but we use vsrc on XB.
2685 XX2_RD6_UIM5_RS6<60, 181, (outs vsrc:$XT),
2686 (ins vsrc:$XTi, vsrc:$XB, u4imm:$UIM),
2687 "xxinsertw $XT, $XB, $UIM", IIC_VecFP,
2688 [(set v4i32:$XT, (PPCvecinsert v4i32:$XTi, v4i32:$XB,
2689 imm32SExt16:$UIM))]>,
2690 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">;
2692 // Vector Extract Unsigned Word
2693 def XXEXTRACTUW : XX2_RD6_UIM5_RS6<60, 165,
2694 (outs vsfrc:$XT), (ins vsrc:$XB, u4imm:$UIMM),
2695 "xxextractuw $XT, $XB, $UIMM", IIC_VecFP, []>;
2698 // Vector Insert Exponent DP/SP
2699 def XVIEXPDP : XX3_XT5_XA5_XB5<60, 248, "xviexpdp", vsrc, vsrc, vsrc,
2700 IIC_VecFP, [(set v2f64: $XT,(int_ppc_vsx_xviexpdp v2i64:$XA, v2i64:$XB))]>;
2701 def XVIEXPSP : XX3_XT5_XA5_XB5<60, 216, "xviexpsp", vsrc, vsrc, vsrc,
2702 IIC_VecFP, [(set v4f32: $XT,(int_ppc_vsx_xviexpsp v4i32:$XA, v4i32:$XB))]>;
2704 // Vector Extract Exponent/Significand DP/SP
2705 def XVXEXPDP : XX2_XT6_XO5_XB6<60, 0, 475, "xvxexpdp", vsrc,
2707 (int_ppc_vsx_xvxexpdp v2f64:$XB))]>;
2708 def XVXEXPSP : XX2_XT6_XO5_XB6<60, 8, 475, "xvxexpsp", vsrc,
2710 (int_ppc_vsx_xvxexpsp v4f32:$XB))]>;
2711 def XVXSIGDP : XX2_XT6_XO5_XB6<60, 1, 475, "xvxsigdp", vsrc,
2713 (int_ppc_vsx_xvxsigdp v2f64:$XB))]>;
2714 def XVXSIGSP : XX2_XT6_XO5_XB6<60, 9, 475, "xvxsigsp", vsrc,
2716 (int_ppc_vsx_xvxsigsp v4f32:$XB))]>;
2718 let AddedComplexity = 400, Predicates = [HasP9Vector] in {
2719 // Extra patterns expanding to vector Extract Word/Insert Word
2720 def : Pat<(v4i32 (int_ppc_vsx_xxinsertw v4i32:$A, v2i64:$B, imm:$IMM)),
2721 (v4i32 (XXINSERTW $A, $B, imm:$IMM))>;
2722 def : Pat<(v2i64 (int_ppc_vsx_xxextractuw v2i64:$A, imm:$IMM)),
2723 (v2i64 (COPY_TO_REGCLASS (XXEXTRACTUW $A, imm:$IMM), VSRC))>;
2724 } // AddedComplexity = 400, HasP9Vector
2726 //===--------------------------------------------------------------------===//
2728 // Test Data Class SP/DP/QP
2729 let UseVSXReg = 1 in {
2730 def XSTSTDCSP : XX2_BF3_DCMX7_RS6<60, 298,
2731 (outs crrc:$BF), (ins u7imm:$DCMX, vsfrc:$XB),
2732 "xststdcsp $BF, $XB, $DCMX", IIC_VecFP, []>;
2733 def XSTSTDCDP : XX2_BF3_DCMX7_RS6<60, 362,
2734 (outs crrc:$BF), (ins u7imm:$DCMX, vsfrc:$XB),
2735 "xststdcdp $BF, $XB, $DCMX", IIC_VecFP, []>;
2737 def XSTSTDCQP : X_BF3_DCMX7_RS5 <63, 708,
2738 (outs crrc:$BF), (ins u7imm:$DCMX, vrrc:$vB),
2739 "xststdcqp $BF, $vB, $DCMX", IIC_VecFP, []>;
2741 // Vector Test Data Class SP/DP
2742 let UseVSXReg = 1 in {
2743 def XVTSTDCSP : XX2_RD6_DCMX7_RS6<60, 13, 5,
2744 (outs vsrc:$XT), (ins u7imm:$DCMX, vsrc:$XB),
2745 "xvtstdcsp $XT, $XB, $DCMX", IIC_VecFP,
2747 (int_ppc_vsx_xvtstdcsp v4f32:$XB, imm:$DCMX))]>;
2748 def XVTSTDCDP : XX2_RD6_DCMX7_RS6<60, 15, 5,
2749 (outs vsrc:$XT), (ins u7imm:$DCMX, vsrc:$XB),
2750 "xvtstdcdp $XT, $XB, $DCMX", IIC_VecFP,
2752 (int_ppc_vsx_xvtstdcdp v2f64:$XB, imm:$DCMX))]>;
2755 //===--------------------------------------------------------------------===//
2757 // Maximum/Minimum Type-C/Type-J DP
2758 // XT.dword[1] = 0xUUUU_UUUU_UUUU_UUUU, so we use vsrc for XT
2759 def XSMAXCDP : XX3_XT5_XA5_XB5<60, 128, "xsmaxcdp", vsrc, vsfrc, vsfrc,
2761 def XSMAXJDP : XX3_XT5_XA5_XB5<60, 144, "xsmaxjdp", vsrc, vsfrc, vsfrc,
2763 def XSMINCDP : XX3_XT5_XA5_XB5<60, 136, "xsmincdp", vsrc, vsfrc, vsfrc,
2765 def XSMINJDP : XX3_XT5_XA5_XB5<60, 152, "xsminjdp", vsrc, vsfrc, vsfrc,
2768 //===--------------------------------------------------------------------===//
2770 // Vector Byte-Reverse H/W/D/Q Word
2771 def XXBRH : XX2_XT6_XO5_XB6<60, 7, 475, "xxbrh", vsrc, []>;
2772 def XXBRW : XX2_XT6_XO5_XB6<60, 15, 475, "xxbrw", vsrc, []>;
2773 def XXBRD : XX2_XT6_XO5_XB6<60, 23, 475, "xxbrd", vsrc, []>;
2774 def XXBRQ : XX2_XT6_XO5_XB6<60, 31, 475, "xxbrq", vsrc, []>;
2777 def : Pat<(v8i16 (PPCxxreverse v8i16 :$A)),
2778 (v8i16 (COPY_TO_REGCLASS (XXBRH (COPY_TO_REGCLASS $A, VSRC)), VRRC))>;
2779 def : Pat<(v4i32 (PPCxxreverse v4i32 :$A)),
2780 (v4i32 (XXBRW $A))>;
2781 def : Pat<(v2i64 (PPCxxreverse v2i64 :$A)),
2782 (v2i64 (XXBRD $A))>;
2783 def : Pat<(v1i128 (PPCxxreverse v1i128 :$A)),
2784 (v1i128 (COPY_TO_REGCLASS (XXBRQ (COPY_TO_REGCLASS $A, VSRC)), VRRC))>;
2787 def XXPERM : XX3_XT5_XA5_XB5<60, 26, "xxperm" , vsrc, vsrc, vsrc,
2789 def XXPERMR : XX3_XT5_XA5_XB5<60, 58, "xxpermr", vsrc, vsrc, vsrc,
2792 // Vector Splat Immediate Byte
2793 def XXSPLTIB : X_RD6_IMM8<60, 360, (outs vsrc:$XT), (ins u8imm:$IMM8),
2794 "xxspltib $XT, $IMM8", IIC_VecPerm, []>, UseVSXReg;
2796 //===--------------------------------------------------------------------===//
2797 // Vector/Scalar Load/Store Instructions
2799 // When adding new D-Form loads/stores, be sure to update the ImmToIdxMap in
2800 // PPCRegisterInfo::PPCRegisterInfo and maybe save yourself some debugging.
2801 let mayLoad = 1, mayStore = 0 in {
2803 def LXV : DQ_RD6_RS5_DQ12<61, 1, (outs vsrc:$XT), (ins memrix16:$src),
2804 "lxv $XT, $src", IIC_LdStLFD, []>, UseVSXReg;
2806 def LXSD : DSForm_1<57, 2, (outs vfrc:$vD), (ins memrix:$src),
2807 "lxsd $vD, $src", IIC_LdStLFD, []>;
2808 // Load SP from src, convert it to DP, and place in dword[0]
2809 def LXSSP : DSForm_1<57, 3, (outs vfrc:$vD), (ins memrix:$src),
2810 "lxssp $vD, $src", IIC_LdStLFD, []>;
2812 // [PO T RA RB XO TX] almost equal to [PO S RA RB XO SX], but has different
2813 // "out" and "in" dag
2814 class X_XT6_RA5_RB5<bits<6> opcode, bits<10> xo, string opc,
2815 RegisterOperand vtype, list<dag> pattern>
2816 : XX1Form_memOp<opcode, xo, (outs vtype:$XT), (ins memrr:$src),
2817 !strconcat(opc, " $XT, $src"), IIC_LdStLFD, pattern>, UseVSXReg;
2819 // Load as Integer Byte/Halfword & Zero Indexed
2820 def LXSIBZX : X_XT6_RA5_RB5<31, 781, "lxsibzx", vsfrc,
2821 [(set f64:$XT, (PPClxsizx xoaddr:$src, 1))]>;
2822 def LXSIHZX : X_XT6_RA5_RB5<31, 813, "lxsihzx", vsfrc,
2823 [(set f64:$XT, (PPClxsizx xoaddr:$src, 2))]>;
2825 // Load Vector Halfword*8/Byte*16 Indexed
2826 def LXVH8X : X_XT6_RA5_RB5<31, 812, "lxvh8x" , vsrc, []>;
2827 def LXVB16X : X_XT6_RA5_RB5<31, 876, "lxvb16x", vsrc, []>;
2829 // Load Vector Indexed
2830 def LXVX : X_XT6_RA5_RB5<31, 268, "lxvx" , vsrc,
2831 [(set v2f64:$XT, (load xaddr:$src))]>;
2832 // Load Vector (Left-justified) with Length
2833 def LXVL : XX1Form_memOp<31, 269, (outs vsrc:$XT), (ins memr:$src, g8rc:$rB),
2834 "lxvl $XT, $src, $rB", IIC_LdStLoad,
2835 [(set v4i32:$XT, (int_ppc_vsx_lxvl addr:$src, i64:$rB))]>,
2837 def LXVLL : XX1Form_memOp<31,301, (outs vsrc:$XT), (ins memr:$src, g8rc:$rB),
2838 "lxvll $XT, $src, $rB", IIC_LdStLoad,
2839 [(set v4i32:$XT, (int_ppc_vsx_lxvll addr:$src, i64:$rB))]>,
2842 // Load Vector Word & Splat Indexed
2843 def LXVWSX : X_XT6_RA5_RB5<31, 364, "lxvwsx" , vsrc, []>;
2846 // When adding new D-Form loads/stores, be sure to update the ImmToIdxMap in
2847 // PPCRegisterInfo::PPCRegisterInfo and maybe save yourself some debugging.
2848 let mayStore = 1, mayLoad = 0 in {
2850 def STXV : DQ_RD6_RS5_DQ12<61, 5, (outs), (ins vsrc:$XT, memrix16:$dst),
2851 "stxv $XT, $dst", IIC_LdStSTFD, []>, UseVSXReg;
2853 def STXSD : DSForm_1<61, 2, (outs), (ins vfrc:$vS, memrix:$dst),
2854 "stxsd $vS, $dst", IIC_LdStSTFD, []>;
2855 // Convert DP of dword[0] to SP, and Store to dst
2856 def STXSSP : DSForm_1<61, 3, (outs), (ins vfrc:$vS, memrix:$dst),
2857 "stxssp $vS, $dst", IIC_LdStSTFD, []>;
2859 // [PO S RA RB XO SX]
2860 class X_XS6_RA5_RB5<bits<6> opcode, bits<10> xo, string opc,
2861 RegisterOperand vtype, list<dag> pattern>
2862 : XX1Form_memOp<opcode, xo, (outs), (ins vtype:$XT, memrr:$dst),
2863 !strconcat(opc, " $XT, $dst"), IIC_LdStSTFD, pattern>, UseVSXReg;
2865 // Store as Integer Byte/Halfword Indexed
2866 def STXSIBX : X_XS6_RA5_RB5<31, 909, "stxsibx" , vsfrc,
2867 [(PPCstxsix f64:$XT, xoaddr:$dst, 1)]>;
2868 def STXSIHX : X_XS6_RA5_RB5<31, 941, "stxsihx" , vsfrc,
2869 [(PPCstxsix f64:$XT, xoaddr:$dst, 2)]>;
2870 let isCodeGenOnly = 1 in {
2871 def STXSIBXv : X_XS6_RA5_RB5<31, 909, "stxsibx" , vrrc, []>;
2872 def STXSIHXv : X_XS6_RA5_RB5<31, 941, "stxsihx" , vrrc, []>;
2875 // Store Vector Halfword*8/Byte*16 Indexed
2876 def STXVH8X : X_XS6_RA5_RB5<31, 940, "stxvh8x" , vsrc, []>;
2877 def STXVB16X : X_XS6_RA5_RB5<31, 1004, "stxvb16x", vsrc, []>;
2879 // Store Vector Indexed
2880 def STXVX : X_XS6_RA5_RB5<31, 396, "stxvx" , vsrc,
2881 [(store v2f64:$XT, xaddr:$dst)]>;
2883 // Store Vector (Left-justified) with Length
2884 def STXVL : XX1Form_memOp<31, 397, (outs),
2885 (ins vsrc:$XT, memr:$dst, g8rc:$rB),
2886 "stxvl $XT, $dst, $rB", IIC_LdStLoad,
2887 [(int_ppc_vsx_stxvl v4i32:$XT, addr:$dst,
2890 def STXVLL : XX1Form_memOp<31, 429, (outs),
2891 (ins vsrc:$XT, memr:$dst, g8rc:$rB),
2892 "stxvll $XT, $dst, $rB", IIC_LdStLoad,
2893 [(int_ppc_vsx_stxvll v4i32:$XT, addr:$dst,
2898 let Predicates = [IsLittleEndian] in {
2899 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 0)))))),
2900 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 3))))>;
2901 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 1)))))),
2902 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 2))))>;
2903 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 2)))))),
2904 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 1))))>;
2905 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 3)))))),
2906 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 0))))>;
2907 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 0)))))),
2908 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 3)), VSFRC))>;
2909 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 1)))))),
2910 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 2)), VSFRC))>;
2911 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 2)))))),
2912 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 1)), VSFRC))>;
2913 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 3)))))),
2914 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 0)), VSFRC))>;
2917 let Predicates = [IsBigEndian] in {
2918 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 0)))))),
2919 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 0))))>;
2920 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 1)))))),
2921 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 1))))>;
2922 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 2)))))),
2923 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 2))))>;
2924 def: Pat<(f32 (PPCfcfids (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 3)))))),
2925 (f32 (XSCVSPDPN (XVCVSXWSP (XXSPLTW $A, 3))))>;
2926 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 0)))))),
2927 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 0)), VSFRC))>;
2928 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 1)))))),
2929 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 1)), VSFRC))>;
2930 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 2)))))),
2931 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 2)), VSFRC))>;
2932 def: Pat<(f64 (PPCfcfid (f64 (PPCmtvsra (i32 (extractelt v4i32:$A, 3)))))),
2933 (f64 (COPY_TO_REGCLASS (XVCVSXWDP (XXSPLTW $A, 3)), VSFRC))>;
2936 // Alternate patterns for PPCmtvsrz where the output is v8i16 or v16i8 instead
2938 def : Pat<(v8i16 (PPCmtvsrz i32:$A)),
2939 (v8i16 (SUBREG_TO_REG (i64 1), (MTVSRWZ $A), sub_64))>;
2940 def : Pat<(v16i8 (PPCmtvsrz i32:$A)),
2941 (v16i8 (SUBREG_TO_REG (i64 1), (MTVSRWZ $A), sub_64))>;
2943 // Patterns for which instructions from ISA 3.0 are a better match
2944 let Predicates = [IsLittleEndian, HasP9Vector] in {
2945 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 0)))))),
2946 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 12)))>;
2947 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 1)))))),
2948 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 8)))>;
2949 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 2)))))),
2950 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 4)))>;
2951 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 3)))))),
2952 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 0)))>;
2953 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 0)))))),
2954 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 12)))>;
2955 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 1)))))),
2956 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 8)))>;
2957 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 2)))))),
2958 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 4)))>;
2959 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 3)))))),
2960 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 0)))>;
2961 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 0)),
2962 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 12))>;
2963 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 1)),
2964 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 8))>;
2965 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 2)),
2966 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 4))>;
2967 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 3)),
2968 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 0))>;
2969 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 0)),
2970 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 12))>;
2971 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 1)),
2972 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 8))>;
2973 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 2)),
2974 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 4))>;
2975 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 3)),
2976 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 0))>;
2977 } // IsLittleEndian, HasP9Vector
2979 let Predicates = [IsBigEndian, HasP9Vector] in {
2980 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 0)))))),
2981 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 0)))>;
2982 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 1)))))),
2983 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 4)))>;
2984 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 2)))))),
2985 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 8)))>;
2986 def : Pat<(f32 (PPCfcfidus (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 3)))))),
2987 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 12)))>;
2988 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 0)))))),
2989 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 0)))>;
2990 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 1)))))),
2991 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 4)))>;
2992 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 2)))))),
2993 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 8)))>;
2994 def : Pat<(f64 (PPCfcfidu (f64 (PPCmtvsrz (i32 (extractelt v4i32:$A, 3)))))),
2995 (f64 (XSCVUXDDP (XXEXTRACTUW $A, 12)))>;
2996 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 0)),
2997 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 0))>;
2998 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 1)),
2999 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 4))>;
3000 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 2)),
3001 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 8))>;
3002 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 3)),
3003 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 12))>;
3004 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 0)),
3005 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 0))>;
3006 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 1)),
3007 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 4))>;
3008 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 2)),
3009 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 8))>;
3010 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 3)),
3011 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 12))>;
3012 } // IsLittleEndian, HasP9Vector
3014 // D-Form Load/Store
3015 def : Pat<(v4i32 (quadwOffsetLoad iqaddr:$src)), (LXV memrix16:$src)>;
3016 def : Pat<(v4f32 (quadwOffsetLoad iqaddr:$src)), (LXV memrix16:$src)>;
3017 def : Pat<(v2i64 (quadwOffsetLoad iqaddr:$src)), (LXV memrix16:$src)>;
3018 def : Pat<(v2f64 (quadwOffsetLoad iqaddr:$src)), (LXV memrix16:$src)>;
3019 def : Pat<(f128 (quadwOffsetLoad iqaddr:$src)),
3020 (COPY_TO_REGCLASS (LXV memrix16:$src), VRRC)>;
3021 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x iqaddr:$src)), (LXV memrix16:$src)>;
3022 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x iqaddr:$src)), (LXV memrix16:$src)>;
3024 def : Pat<(quadwOffsetStore v4f32:$rS, iqaddr:$dst), (STXV $rS, memrix16:$dst)>;
3025 def : Pat<(quadwOffsetStore v4i32:$rS, iqaddr:$dst), (STXV $rS, memrix16:$dst)>;
3026 def : Pat<(quadwOffsetStore v2f64:$rS, iqaddr:$dst), (STXV $rS, memrix16:$dst)>;
3027 def : Pat<(quadwOffsetStore f128:$rS, iqaddr:$dst),
3028 (STXV (COPY_TO_REGCLASS $rS, VSRC), memrix16:$dst)>;
3029 def : Pat<(quadwOffsetStore v2i64:$rS, iqaddr:$dst), (STXV $rS, memrix16:$dst)>;
3030 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, iqaddr:$dst),
3031 (STXV $rS, memrix16:$dst)>;
3032 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, iqaddr:$dst),
3033 (STXV $rS, memrix16:$dst)>;
3036 def : Pat<(v2f64 (nonQuadwOffsetLoad xoaddr:$src)), (LXVX xoaddr:$src)>;
3037 def : Pat<(v2i64 (nonQuadwOffsetLoad xoaddr:$src)), (LXVX xoaddr:$src)>;
3038 def : Pat<(v4f32 (nonQuadwOffsetLoad xoaddr:$src)), (LXVX xoaddr:$src)>;
3039 def : Pat<(v4i32 (nonQuadwOffsetLoad xoaddr:$src)), (LXVX xoaddr:$src)>;
3040 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x xoaddr:$src)), (LXVX xoaddr:$src)>;
3041 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x xoaddr:$src)), (LXVX xoaddr:$src)>;
3042 def : Pat<(f128 (nonQuadwOffsetLoad xoaddr:$src)),
3043 (COPY_TO_REGCLASS (LXVX xoaddr:$src), VRRC)>;
3044 def : Pat<(nonQuadwOffsetStore f128:$rS, xoaddr:$dst),
3045 (STXVX (COPY_TO_REGCLASS $rS, VSRC), xoaddr:$dst)>;
3046 def : Pat<(nonQuadwOffsetStore v2f64:$rS, xoaddr:$dst),
3047 (STXVX $rS, xoaddr:$dst)>;
3048 def : Pat<(nonQuadwOffsetStore v2i64:$rS, xoaddr:$dst),
3049 (STXVX $rS, xoaddr:$dst)>;
3050 def : Pat<(nonQuadwOffsetStore v4f32:$rS, xoaddr:$dst),
3051 (STXVX $rS, xoaddr:$dst)>;
3052 def : Pat<(nonQuadwOffsetStore v4i32:$rS, xoaddr:$dst),
3053 (STXVX $rS, xoaddr:$dst)>;
3054 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, xoaddr:$dst),
3055 (STXVX $rS, xoaddr:$dst)>;
3056 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, xoaddr:$dst),
3057 (STXVX $rS, xoaddr:$dst)>;
3059 let AddedComplexity = 400 in {
3060 // LIWAX - This instruction is used for sign extending i32 -> i64.
3061 // LIWZX - This instruction will be emitted for i32, f32, and when
3062 // zero-extending i32 to i64 (zext i32 -> i64).
3063 let Predicates = [IsLittleEndian] in {
3065 def : Pat<(v2i64 (scalar_to_vector (i64 (sextloadi32 xoaddr:$src)))),
3067 (COPY_TO_REGCLASS (LIWAX xoaddr:$src), VSRC), 2))>;
3069 def : Pat<(v2i64 (scalar_to_vector (i64 (zextloadi32 xoaddr:$src)))),
3071 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 2))>;
3073 def : Pat<(v4i32 (scalar_to_vector (i32 (load xoaddr:$src)))),
3075 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 2))>;
3077 def : Pat<(v4f32 (scalar_to_vector (f32 (load xoaddr:$src)))),
3079 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 2))>;
3082 let Predicates = [IsBigEndian] in {
3083 def : Pat<(v2i64 (scalar_to_vector (i64 (sextloadi32 xoaddr:$src)))),
3084 (v2i64 (COPY_TO_REGCLASS (LIWAX xoaddr:$src), VSRC))>;
3086 def : Pat<(v2i64 (scalar_to_vector (i64 (zextloadi32 xoaddr:$src)))),
3087 (v2i64 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC))>;
3089 def : Pat<(v4i32 (scalar_to_vector (i32 (load xoaddr:$src)))),
3091 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 1))>;
3093 def : Pat<(v4f32 (scalar_to_vector (f32 (load xoaddr:$src)))),
3095 (COPY_TO_REGCLASS (LIWZX xoaddr:$src), VSRC), 1))>;
3100 // Build vectors from i8 loads
3101 def : Pat<(v16i8 (scalar_to_vector ScalarLoads.Li8)),
3102 (v16i8 (VSPLTBs 7, (LXSIBZX xoaddr:$src)))>;
3103 def : Pat<(v8i16 (scalar_to_vector ScalarLoads.ZELi8)),
3104 (v8i16 (VSPLTHs 3, (LXSIBZX xoaddr:$src)))>;
3105 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.ZELi8)),
3106 (v4i32 (XXSPLTWs (LXSIBZX xoaddr:$src), 1))>;
3107 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.ZELi8i64)),
3108 (v2i64 (XXPERMDIs (LXSIBZX xoaddr:$src), 0))>;
3109 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.SELi8)),
3110 (v4i32 (XXSPLTWs (VEXTSB2Ws (LXSIBZX xoaddr:$src)), 1))>;
3111 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.SELi8i64)),
3112 (v2i64 (XXPERMDIs (VEXTSB2Ds (LXSIBZX xoaddr:$src)), 0))>;
3114 // Build vectors from i16 loads
3115 def : Pat<(v8i16 (scalar_to_vector ScalarLoads.Li16)),
3116 (v8i16 (VSPLTHs 3, (LXSIHZX xoaddr:$src)))>;
3117 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.ZELi16)),
3118 (v4i32 (XXSPLTWs (LXSIHZX xoaddr:$src), 1))>;
3119 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.ZELi16i64)),
3120 (v2i64 (XXPERMDIs (LXSIHZX xoaddr:$src), 0))>;
3121 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.SELi16)),
3122 (v4i32 (XXSPLTWs (VEXTSH2Ws (LXSIHZX xoaddr:$src)), 1))>;
3123 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.SELi16i64)),
3124 (v2i64 (XXPERMDIs (VEXTSH2Ds (LXSIHZX xoaddr:$src)), 0))>;
3126 let Predicates = [IsBigEndian, HasP9Vector] in {
3127 // Scalar stores of i8
3128 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 0)), xoaddr:$dst),
3129 (STXSIBXv (v16i8 (VSLDOI $S, $S, 9)), xoaddr:$dst)>;
3130 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 1)), xoaddr:$dst),
3131 (STXSIBXv (v16i8 (VSLDOI $S, $S, 10)), xoaddr:$dst)>;
3132 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 2)), xoaddr:$dst),
3133 (STXSIBXv (v16i8 (VSLDOI $S, $S, 11)), xoaddr:$dst)>;
3134 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 3)), xoaddr:$dst),
3135 (STXSIBXv (v16i8 (VSLDOI $S, $S, 12)), xoaddr:$dst)>;
3136 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 4)), xoaddr:$dst),
3137 (STXSIBXv (v16i8 (VSLDOI $S, $S, 13)), xoaddr:$dst)>;
3138 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 5)), xoaddr:$dst),
3139 (STXSIBXv (v16i8 (VSLDOI $S, $S, 14)), xoaddr:$dst)>;
3140 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 6)), xoaddr:$dst),
3141 (STXSIBXv (v16i8 (VSLDOI $S, $S, 15)), xoaddr:$dst)>;
3142 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 7)), xoaddr:$dst),
3143 (STXSIBXv $S, xoaddr:$dst)>;
3144 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 8)), xoaddr:$dst),
3145 (STXSIBXv (v16i8 (VSLDOI $S, $S, 1)), xoaddr:$dst)>;
3146 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 9)), xoaddr:$dst),
3147 (STXSIBXv (v16i8 (VSLDOI $S, $S, 2)), xoaddr:$dst)>;
3148 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 10)), xoaddr:$dst),
3149 (STXSIBXv (v16i8 (VSLDOI $S, $S, 3)), xoaddr:$dst)>;
3150 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 11)), xoaddr:$dst),
3151 (STXSIBXv (v16i8 (VSLDOI $S, $S, 4)), xoaddr:$dst)>;
3152 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 12)), xoaddr:$dst),
3153 (STXSIBXv (v16i8 (VSLDOI $S, $S, 5)), xoaddr:$dst)>;
3154 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 13)), xoaddr:$dst),
3155 (STXSIBXv (v16i8 (VSLDOI $S, $S, 6)), xoaddr:$dst)>;
3156 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 14)), xoaddr:$dst),
3157 (STXSIBXv (v16i8 (VSLDOI $S, $S, 7)), xoaddr:$dst)>;
3158 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 15)), xoaddr:$dst),
3159 (STXSIBXv (v16i8 (VSLDOI $S, $S, 8)), xoaddr:$dst)>;
3161 // Scalar stores of i16
3162 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 0)), xoaddr:$dst),
3163 (STXSIHXv (v16i8 (VSLDOI $S, $S, 10)), xoaddr:$dst)>;
3164 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 1)), xoaddr:$dst),
3165 (STXSIHXv (v16i8 (VSLDOI $S, $S, 12)), xoaddr:$dst)>;
3166 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 2)), xoaddr:$dst),
3167 (STXSIHXv (v16i8 (VSLDOI $S, $S, 14)), xoaddr:$dst)>;
3168 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 3)), xoaddr:$dst),
3169 (STXSIHXv $S, xoaddr:$dst)>;
3170 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 4)), xoaddr:$dst),
3171 (STXSIHXv (v16i8 (VSLDOI $S, $S, 2)), xoaddr:$dst)>;
3172 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 5)), xoaddr:$dst),
3173 (STXSIHXv (v16i8 (VSLDOI $S, $S, 4)), xoaddr:$dst)>;
3174 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 6)), xoaddr:$dst),
3175 (STXSIHXv (v16i8 (VSLDOI $S, $S, 6)), xoaddr:$dst)>;
3176 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 7)), xoaddr:$dst),
3177 (STXSIHXv (v16i8 (VSLDOI $S, $S, 8)), xoaddr:$dst)>;
3178 } // IsBigEndian, HasP9Vector
3180 let Predicates = [IsLittleEndian, HasP9Vector] in {
3181 // Scalar stores of i8
3182 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 0)), xoaddr:$dst),
3183 (STXSIBXv (v16i8 (VSLDOI $S, $S, 8)), xoaddr:$dst)>;
3184 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 1)), xoaddr:$dst),
3185 (STXSIBXv (v16i8 (VSLDOI $S, $S, 7)), xoaddr:$dst)>;
3186 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 2)), xoaddr:$dst),
3187 (STXSIBXv (v16i8 (VSLDOI $S, $S, 6)), xoaddr:$dst)>;
3188 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 3)), xoaddr:$dst),
3189 (STXSIBXv (v16i8 (VSLDOI $S, $S, 5)), xoaddr:$dst)>;
3190 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 4)), xoaddr:$dst),
3191 (STXSIBXv (v16i8 (VSLDOI $S, $S, 4)), xoaddr:$dst)>;
3192 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 5)), xoaddr:$dst),
3193 (STXSIBXv (v16i8 (VSLDOI $S, $S, 3)), xoaddr:$dst)>;
3194 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 6)), xoaddr:$dst),
3195 (STXSIBXv (v16i8 (VSLDOI $S, $S, 2)), xoaddr:$dst)>;
3196 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 7)), xoaddr:$dst),
3197 (STXSIBXv (v16i8 (VSLDOI $S, $S, 1)), xoaddr:$dst)>;
3198 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 8)), xoaddr:$dst),
3199 (STXSIBXv $S, xoaddr:$dst)>;
3200 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 9)), xoaddr:$dst),
3201 (STXSIBXv (v16i8 (VSLDOI $S, $S, 15)), xoaddr:$dst)>;
3202 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 10)), xoaddr:$dst),
3203 (STXSIBXv (v16i8 (VSLDOI $S, $S, 14)), xoaddr:$dst)>;
3204 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 11)), xoaddr:$dst),
3205 (STXSIBXv (v16i8 (VSLDOI $S, $S, 13)), xoaddr:$dst)>;
3206 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 12)), xoaddr:$dst),
3207 (STXSIBXv (v16i8 (VSLDOI $S, $S, 12)), xoaddr:$dst)>;
3208 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 13)), xoaddr:$dst),
3209 (STXSIBXv (v16i8 (VSLDOI $S, $S, 11)), xoaddr:$dst)>;
3210 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 14)), xoaddr:$dst),
3211 (STXSIBXv (v16i8 (VSLDOI $S, $S, 10)), xoaddr:$dst)>;
3212 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 15)), xoaddr:$dst),
3213 (STXSIBXv (v16i8 (VSLDOI $S, $S, 9)), xoaddr:$dst)>;
3215 // Scalar stores of i16
3216 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 0)), xoaddr:$dst),
3217 (STXSIHXv (v16i8 (VSLDOI $S, $S, 8)), xoaddr:$dst)>;
3218 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 1)), xoaddr:$dst),
3219 (STXSIHXv (v16i8 (VSLDOI $S, $S, 6)), xoaddr:$dst)>;
3220 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 2)), xoaddr:$dst),
3221 (STXSIHXv (v16i8 (VSLDOI $S, $S, 4)), xoaddr:$dst)>;
3222 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 3)), xoaddr:$dst),
3223 (STXSIHXv (v16i8 (VSLDOI $S, $S, 2)), xoaddr:$dst)>;
3224 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 4)), xoaddr:$dst),
3225 (STXSIHXv $S, xoaddr:$dst)>;
3226 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 5)), xoaddr:$dst),
3227 (STXSIHXv (v16i8 (VSLDOI $S, $S, 14)), xoaddr:$dst)>;
3228 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 6)), xoaddr:$dst),
3229 (STXSIHXv (v16i8 (VSLDOI $S, $S, 12)), xoaddr:$dst)>;
3230 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 7)), xoaddr:$dst),
3231 (STXSIHXv (v16i8 (VSLDOI $S, $S, 10)), xoaddr:$dst)>;
3232 } // IsLittleEndian, HasP9Vector
3235 // Vector sign extensions
3236 def : Pat<(f64 (PPCVexts f64:$A, 1)),
3237 (f64 (COPY_TO_REGCLASS (VEXTSB2Ds $A), VSFRC))>;
3238 def : Pat<(f64 (PPCVexts f64:$A, 2)),
3239 (f64 (COPY_TO_REGCLASS (VEXTSH2Ds $A), VSFRC))>;
3241 let isPseudo = 1 in {
3242 def DFLOADf32 : Pseudo<(outs vssrc:$XT), (ins memrix:$src),
3244 [(set f32:$XT, (load ixaddr:$src))]>;
3245 def DFLOADf64 : Pseudo<(outs vsfrc:$XT), (ins memrix:$src),
3247 [(set f64:$XT, (load ixaddr:$src))]>;
3248 def DFSTOREf32 : Pseudo<(outs), (ins vssrc:$XT, memrix:$dst),
3250 [(store f32:$XT, ixaddr:$dst)]>;
3251 def DFSTOREf64 : Pseudo<(outs), (ins vsfrc:$XT, memrix:$dst),
3253 [(store f64:$XT, ixaddr:$dst)]>;
3255 def : Pat<(f64 (extloadf32 ixaddr:$src)),
3256 (COPY_TO_REGCLASS (DFLOADf32 ixaddr:$src), VSFRC)>;
3257 def : Pat<(f32 (fpround (f64 (extloadf32 ixaddr:$src)))),
3258 (f32 (DFLOADf32 ixaddr:$src))>;
3261 let AddedComplexity = 400 in {
3262 // The following pseudoinstructions are used to ensure the utilization
3263 // of all 64 VSX registers.
3264 let Predicates = [IsLittleEndian, HasP9Vector] in {
3265 def : Pat<(v2i64 (scalar_to_vector (i64 (load ixaddr:$src)))),
3267 (COPY_TO_REGCLASS (DFLOADf64 ixaddr:$src), VSRC), 2))>;
3268 def : Pat<(v2i64 (scalar_to_vector (i64 (load xaddr:$src)))),
3270 (COPY_TO_REGCLASS (XFLOADf64 xaddr:$src), VSRC), 2))>;
3272 def : Pat<(v2f64 (scalar_to_vector (f64 (load ixaddr:$src)))),
3274 (COPY_TO_REGCLASS (DFLOADf64 ixaddr:$src), VSRC), 2))>;
3275 def : Pat<(v2f64 (scalar_to_vector (f64 (load xaddr:$src)))),
3277 (COPY_TO_REGCLASS (XFLOADf64 xaddr:$src), VSRC), 2))>;
3280 let Predicates = [IsBigEndian, HasP9Vector] in {
3281 def : Pat<(v2i64 (scalar_to_vector (i64 (load ixaddr:$src)))),
3282 (v2i64 (COPY_TO_REGCLASS (DFLOADf64 ixaddr:$src), VSRC))>;
3283 def : Pat<(v2i64 (scalar_to_vector (i64 (load xaddr:$src)))),
3284 (v2i64 (COPY_TO_REGCLASS (XFLOADf64 xaddr:$src), VSRC))>;
3286 def : Pat<(v2f64 (scalar_to_vector (f64 (load ixaddr:$src)))),
3287 (v2f64 (COPY_TO_REGCLASS (DFLOADf64 ixaddr:$src), VSRC))>;
3288 def : Pat<(v2f64 (scalar_to_vector (f64 (load xaddr:$src)))),
3289 (v2f64 (COPY_TO_REGCLASS (XFLOADf64 xaddr:$src), VSRC))>;
3293 let Predicates = [IsBigEndian, HasP9Vector] in {
3295 // (Un)Signed DWord vector extract -> QP
3296 def : Pat<(f128 (sint_to_fp (i64 (extractelt v2i64:$src, 0)))),
3297 (f128 (XSCVSDQP (COPY_TO_REGCLASS $src, VFRC)))>;
3298 def : Pat<(f128 (sint_to_fp (i64 (extractelt v2i64:$src, 1)))),
3300 (EXTRACT_SUBREG (XXPERMDI $src, $src, 3), sub_64)))>;
3301 def : Pat<(f128 (uint_to_fp (i64 (extractelt v2i64:$src, 0)))),
3302 (f128 (XSCVUDQP (COPY_TO_REGCLASS $src, VFRC)))>;
3303 def : Pat<(f128 (uint_to_fp (i64 (extractelt v2i64:$src, 1)))),
3305 (EXTRACT_SUBREG (XXPERMDI $src, $src, 3), sub_64)))>;
3307 // (Un)Signed Word vector extract -> QP
3308 def : Pat<(f128 (sint_to_fp (i32 (extractelt v4i32:$src, 1)))),
3309 (f128 (XSCVSDQP (EXTRACT_SUBREG (VEXTSW2D $src), sub_64)))>;
3310 foreach Idx = [0,2,3] in {
3311 def : Pat<(f128 (sint_to_fp (i32 (extractelt v4i32:$src, Idx)))),
3312 (f128 (XSCVSDQP (EXTRACT_SUBREG
3313 (VEXTSW2D (VSPLTW Idx, $src)), sub_64)))>;
3315 foreach Idx = 0-3 in {
3316 def : Pat<(f128 (uint_to_fp (i32 (extractelt v4i32:$src, Idx)))),
3317 (f128 (XSCVUDQP (XXEXTRACTUW $src, !shl(Idx, 2))))>;
3320 // (Un)Signed HWord vector extract -> QP
3321 foreach Idx = 0-7 in {
3322 def : Pat<(f128 (sint_to_fp
3324 (vector_extract v8i16:$src, Idx), i16)))),
3325 (f128 (XSCVSDQP (EXTRACT_SUBREG
3326 (VEXTSH2D (VEXTRACTUH !add(Idx, Idx), $src)),
3328 // The SDAG adds the `and` since an `i16` is being extracted as an `i32`.
3329 def : Pat<(f128 (uint_to_fp
3330 (and (i32 (vector_extract v8i16:$src, Idx)), 65535))),
3331 (f128 (XSCVUDQP (EXTRACT_SUBREG
3332 (VEXTRACTUH !add(Idx, Idx), $src), sub_64)))>;
3335 // (Un)Signed Byte vector extract -> QP
3336 foreach Idx = 0-15 in {
3337 def : Pat<(f128 (sint_to_fp
3338 (i32 (sext_inreg (vector_extract v16i8:$src, Idx),
3340 (f128 (XSCVSDQP (EXTRACT_SUBREG
3341 (VEXTSB2D (VEXTRACTUB Idx, $src)), sub_64)))>;
3342 def : Pat<(f128 (uint_to_fp
3343 (and (i32 (vector_extract v16i8:$src, Idx)), 255))),
3345 (EXTRACT_SUBREG (VEXTRACTUB Idx, $src), sub_64)))>;
3348 // Unsiged int in vsx register -> QP
3349 def : Pat<(f128 (uint_to_fp (i32 (PPCmfvsr f64:$src)))),
3351 (XXEXTRACTUW (SUBREG_TO_REG (i64 1), $src, sub_64), 4)))>;
3352 } // IsBigEndian, HasP9Vector
3354 let Predicates = [IsLittleEndian, HasP9Vector] in {
3356 // (Un)Signed DWord vector extract -> QP
3357 def : Pat<(f128 (sint_to_fp (i64 (extractelt v2i64:$src, 0)))),
3359 (EXTRACT_SUBREG (XXPERMDI $src, $src, 3), sub_64)))>;
3360 def : Pat<(f128 (sint_to_fp (i64 (extractelt v2i64:$src, 1)))),
3361 (f128 (XSCVSDQP (COPY_TO_REGCLASS $src, VFRC)))>;
3362 def : Pat<(f128 (uint_to_fp (i64 (extractelt v2i64:$src, 0)))),
3364 (EXTRACT_SUBREG (XXPERMDI $src, $src, 3), sub_64)))>;
3365 def : Pat<(f128 (uint_to_fp (i64 (extractelt v2i64:$src, 1)))),
3366 (f128 (XSCVUDQP (COPY_TO_REGCLASS $src, VFRC)))>;
3368 // (Un)Signed Word vector extract -> QP
3369 foreach Idx = [[0,3],[1,2],[3,0]] in {
3370 def : Pat<(f128 (sint_to_fp (i32 (extractelt v4i32:$src, !head(Idx))))),
3371 (f128 (XSCVSDQP (EXTRACT_SUBREG
3372 (VEXTSW2D (VSPLTW !head(!tail(Idx)), $src)),
3375 def : Pat<(f128 (sint_to_fp (i32 (extractelt v4i32:$src, 2)))),
3376 (f128 (XSCVSDQP (EXTRACT_SUBREG (VEXTSW2D $src), sub_64)))>;
3378 foreach Idx = [[0,12],[1,8],[2,4],[3,0]] in {
3379 def : Pat<(f128 (uint_to_fp (i32 (extractelt v4i32:$src, !head(Idx))))),
3380 (f128 (XSCVUDQP (XXEXTRACTUW $src, !head(!tail(Idx)))))>;
3383 // (Un)Signed HWord vector extract -> QP
3384 // The Nested foreach lists identifies the vector element and corresponding
3385 // register byte location.
3386 foreach Idx = [[0,14],[1,12],[2,10],[3,8],[4,6],[5,4],[6,2],[7,0]] in {
3387 def : Pat<(f128 (sint_to_fp
3389 (vector_extract v8i16:$src, !head(Idx)), i16)))),
3391 (EXTRACT_SUBREG (VEXTSH2D
3392 (VEXTRACTUH !head(!tail(Idx)), $src)),
3394 def : Pat<(f128 (uint_to_fp
3395 (and (i32 (vector_extract v8i16:$src, !head(Idx))),
3397 (f128 (XSCVUDQP (EXTRACT_SUBREG
3398 (VEXTRACTUH !head(!tail(Idx)), $src), sub_64)))>;
3401 // (Un)Signed Byte vector extract -> QP
3402 foreach Idx = [[0,15],[1,14],[2,13],[3,12],[4,11],[5,10],[6,9],[7,8],[8,7],
3403 [9,6],[10,5],[11,4],[12,3],[13,2],[14,1],[15,0]] in {
3404 def : Pat<(f128 (sint_to_fp
3406 (vector_extract v16i8:$src, !head(Idx)), i8)))),
3409 (VEXTSB2D (VEXTRACTUB !head(!tail(Idx)), $src)),
3411 def : Pat<(f128 (uint_to_fp
3412 (and (i32 (vector_extract v16i8:$src, !head(Idx))),
3416 (VEXTRACTUB !head(!tail(Idx)), $src), sub_64)))>;
3419 // Unsiged int in vsx register -> QP
3420 def : Pat<(f128 (uint_to_fp (i32 (PPCmfvsr f64:$src)))),
3422 (XXEXTRACTUW (SUBREG_TO_REG (i64 1), $src, sub_64), 8)))>;
3423 } // IsLittleEndian, HasP9Vector
3425 // Convert (Un)Signed DWord in memory -> QP
3426 def : Pat<(f128 (sint_to_fp (i64 (load xaddr:$src)))),
3427 (f128 (XSCVSDQP (LXSDX xaddr:$src)))>;
3428 def : Pat<(f128 (sint_to_fp (i64 (load ixaddr:$src)))),
3429 (f128 (XSCVSDQP (LXSD ixaddr:$src)))>;
3430 def : Pat<(f128 (uint_to_fp (i64 (load xaddr:$src)))),
3431 (f128 (XSCVUDQP (LXSDX xaddr:$src)))>;
3432 def : Pat<(f128 (uint_to_fp (i64 (load ixaddr:$src)))),
3433 (f128 (XSCVUDQP (LXSD ixaddr:$src)))>;
3435 // Convert Unsigned HWord in memory -> QP
3436 def : Pat<(f128 (uint_to_fp ScalarLoads.ZELi16)),
3437 (f128 (XSCVUDQP (LXSIHZX xaddr:$src)))>;
3439 // Convert Unsigned Byte in memory -> QP
3440 def : Pat<(f128 (uint_to_fp ScalarLoads.ZELi8)),
3441 (f128 (XSCVUDQP (LXSIBZX xoaddr:$src)))>;
3443 // Truncate & Convert QP -> (Un)Signed (D)Word.
3444 def : Pat<(i64 (fp_to_sint f128:$src)), (i64 (MFVRD (XSCVQPSDZ $src)))>;
3445 def : Pat<(i64 (fp_to_uint f128:$src)), (i64 (MFVRD (XSCVQPUDZ $src)))>;
3446 def : Pat<(i32 (fp_to_sint f128:$src)),
3447 (i32 (MFVSRWZ (COPY_TO_REGCLASS (XSCVQPSWZ $src), VFRC)))>;
3448 def : Pat<(i32 (fp_to_uint f128:$src)),
3449 (i32 (MFVSRWZ (COPY_TO_REGCLASS (XSCVQPUWZ $src), VFRC)))>;
3451 // Instructions for store(fptosi).
3452 // The 8-byte version is repeated here due to availability of D-Form STXSD.
3453 def : Pat<(PPCstore_scal_int_from_vsr
3454 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), xaddr:$dst, 8),
3455 (STXSDX (COPY_TO_REGCLASS (XSCVQPSDZ f128:$src), VFRC),
3457 def : Pat<(PPCstore_scal_int_from_vsr
3458 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), ixaddr:$dst, 8),
3459 (STXSD (COPY_TO_REGCLASS (XSCVQPSDZ f128:$src), VFRC),
3461 def : Pat<(PPCstore_scal_int_from_vsr
3462 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), xoaddr:$dst, 4),
3463 (STXSIWX (COPY_TO_REGCLASS (XSCVQPSWZ $src), VFRC), xoaddr:$dst)>;
3464 def : Pat<(PPCstore_scal_int_from_vsr
3465 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), xoaddr:$dst, 2),
3466 (STXSIHX (COPY_TO_REGCLASS (XSCVQPSWZ $src), VFRC), xoaddr:$dst)>;
3467 def : Pat<(PPCstore_scal_int_from_vsr
3468 (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)), xoaddr:$dst, 1),
3469 (STXSIBX (COPY_TO_REGCLASS (XSCVQPSWZ $src), VFRC), xoaddr:$dst)>;
3470 def : Pat<(PPCstore_scal_int_from_vsr
3471 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xaddr:$dst, 8),
3472 (STXSDX (XSCVDPSXDS f64:$src), xaddr:$dst)>;
3473 def : Pat<(PPCstore_scal_int_from_vsr
3474 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), ixaddr:$dst, 8),
3475 (STXSD (XSCVDPSXDS f64:$src), ixaddr:$dst)>;
3476 def : Pat<(PPCstore_scal_int_from_vsr
3477 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xoaddr:$dst, 2),
3478 (STXSIHX (XSCVDPSXWS f64:$src), xoaddr:$dst)>;
3479 def : Pat<(PPCstore_scal_int_from_vsr
3480 (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)), xoaddr:$dst, 1),
3481 (STXSIBX (XSCVDPSXWS f64:$src), xoaddr:$dst)>;
3483 // Instructions for store(fptoui).
3484 def : Pat<(PPCstore_scal_int_from_vsr
3485 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), xaddr:$dst, 8),
3486 (STXSDX (COPY_TO_REGCLASS (XSCVQPUDZ f128:$src), VFRC),
3488 def : Pat<(PPCstore_scal_int_from_vsr
3489 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), ixaddr:$dst, 8),
3490 (STXSD (COPY_TO_REGCLASS (XSCVQPUDZ f128:$src), VFRC),
3492 def : Pat<(PPCstore_scal_int_from_vsr
3493 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), xoaddr:$dst, 4),
3494 (STXSIWX (COPY_TO_REGCLASS (XSCVQPUWZ $src), VFRC), xoaddr:$dst)>;
3495 def : Pat<(PPCstore_scal_int_from_vsr
3496 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), xoaddr:$dst, 2),
3497 (STXSIHX (COPY_TO_REGCLASS (XSCVQPUWZ $src), VFRC), xoaddr:$dst)>;
3498 def : Pat<(PPCstore_scal_int_from_vsr
3499 (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)), xoaddr:$dst, 1),
3500 (STXSIBX (COPY_TO_REGCLASS (XSCVQPUWZ $src), VFRC), xoaddr:$dst)>;
3501 def : Pat<(PPCstore_scal_int_from_vsr
3502 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xaddr:$dst, 8),
3503 (STXSDX (XSCVDPUXDS f64:$src), xaddr:$dst)>;
3504 def : Pat<(PPCstore_scal_int_from_vsr
3505 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), ixaddr:$dst, 8),
3506 (STXSD (XSCVDPUXDS f64:$src), ixaddr:$dst)>;
3507 def : Pat<(PPCstore_scal_int_from_vsr
3508 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xoaddr:$dst, 2),
3509 (STXSIHX (XSCVDPUXWS f64:$src), xoaddr:$dst)>;
3510 def : Pat<(PPCstore_scal_int_from_vsr
3511 (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)), xoaddr:$dst, 1),
3512 (STXSIBX (XSCVDPUXWS f64:$src), xoaddr:$dst)>;
3514 // Round & Convert QP -> DP/SP
3515 def : Pat<(f64 (fpround f128:$src)), (f64 (XSCVQPDP $src))>;
3516 def : Pat<(f32 (fpround f128:$src)), (f32 (XSRSP (XSCVQPDPO $src)))>;
3519 def : Pat<(f128 (fpextend f32:$src)),
3520 (f128 (XSCVDPQP (COPY_TO_REGCLASS $src, VFRC)))>;
3522 } // end HasP9Vector, AddedComplexity
3524 let AddedComplexity = 400 in {
3525 let Predicates = [IsISA3_0, HasP9Vector, HasDirectMove, IsBigEndian] in {
3526 def : Pat<(f128 (PPCbuild_fp128 i64:$rB, i64:$rA)),
3527 (f128 (COPY_TO_REGCLASS (MTVSRDD $rB, $rA), VRRC))>;
3529 let Predicates = [IsISA3_0, HasP9Vector, HasDirectMove, IsLittleEndian] in {
3530 def : Pat<(f128 (PPCbuild_fp128 i64:$rA, i64:$rB)),
3531 (f128 (COPY_TO_REGCLASS (MTVSRDD $rB, $rA), VRRC))>;
3535 let Predicates = [HasP9Vector] in {
3536 let isPseudo = 1 in {
3537 let mayStore = 1 in {
3538 def SPILLTOVSR_STX : PseudoXFormMemOp<(outs),
3539 (ins spilltovsrrc:$XT, memrr:$dst),
3540 "#SPILLTOVSR_STX", []>;
3541 def SPILLTOVSR_ST : Pseudo<(outs), (ins spilltovsrrc:$XT, memrix:$dst),
3542 "#SPILLTOVSR_ST", []>;
3544 let mayLoad = 1 in {
3545 def SPILLTOVSR_LDX : PseudoXFormMemOp<(outs spilltovsrrc:$XT),
3547 "#SPILLTOVSR_LDX", []>;
3548 def SPILLTOVSR_LD : Pseudo<(outs spilltovsrrc:$XT), (ins memrix:$src),
3549 "#SPILLTOVSR_LD", []>;
3554 // Integer extend helper dags 32 -> 64
3556 dag A = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32);
3557 dag B = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $B, sub_32);
3558 dag C = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $C, sub_32);
3559 dag D = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $D, sub_32);
3563 dag A0 = (f32 (fpround (f64 (extractelt v2f64:$A, 0))));
3564 dag A1 = (f32 (fpround (f64 (extractelt v2f64:$A, 1))));
3565 dag B0 = (f32 (fpround (f64 (extractelt v2f64:$B, 0))));
3566 dag B1 = (f32 (fpround (f64 (extractelt v2f64:$B, 1))));
3570 dag A0S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$A, 0))))));
3571 dag A1S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$A, 1))))));
3572 dag B0S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$B, 0))))));
3573 dag B1S = (i32 (PPCmfvsr (f64 (PPCfctiwz (f64 (extractelt v2f64:$B, 1))))));
3574 dag A0U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$A, 0))))));
3575 dag A1U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$A, 1))))));
3576 dag B0U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$B, 0))))));
3577 dag B1U = (i32 (PPCmfvsr (f64 (PPCfctiwuz (f64 (extractelt v2f64:$B, 1))))));
3581 dag LE_A0 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 0)), i8));
3582 dag LE_A1 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 4)), i8));
3583 dag LE_A2 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 8)), i8));
3584 dag LE_A3 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 12)), i8));
3585 dag BE_A0 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 3)), i8));
3586 dag BE_A1 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 7)), i8));
3587 dag BE_A2 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 11)), i8));
3588 dag BE_A3 = (i32 (sext_inreg (i32 (vector_extract v16i8:$A, 15)), i8));
3592 dag LE_A0 = (i64 (sext_inreg
3593 (i64 (anyext (i32 (vector_extract v16i8:$A, 0)))), i8));
3594 dag LE_A1 = (i64 (sext_inreg
3595 (i64 (anyext (i32 (vector_extract v16i8:$A, 8)))), i8));
3596 dag BE_A0 = (i64 (sext_inreg
3597 (i64 (anyext (i32 (vector_extract v16i8:$A, 7)))), i8));
3598 dag BE_A1 = (i64 (sext_inreg
3599 (i64 (anyext (i32 (vector_extract v16i8:$A, 15)))), i8));
3603 dag LE_A0 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 0)), i16));
3604 dag LE_A1 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 2)), i16));
3605 dag LE_A2 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 4)), i16));
3606 dag LE_A3 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 6)), i16));
3607 dag BE_A0 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 1)), i16));
3608 dag BE_A1 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 3)), i16));
3609 dag BE_A2 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 5)), i16));
3610 dag BE_A3 = (i32 (sext_inreg (i32 (vector_extract v8i16:$A, 7)), i16));
3614 dag LE_A0 = (i64 (sext_inreg
3615 (i64 (anyext (i32 (vector_extract v8i16:$A, 0)))), i16));
3616 dag LE_A1 = (i64 (sext_inreg
3617 (i64 (anyext (i32 (vector_extract v8i16:$A, 4)))), i16));
3618 dag BE_A0 = (i64 (sext_inreg
3619 (i64 (anyext (i32 (vector_extract v8i16:$A, 3)))), i16));
3620 dag BE_A1 = (i64 (sext_inreg
3621 (i64 (anyext (i32 (vector_extract v8i16:$A, 7)))), i16));
3625 dag LE_A0 = (i64 (sext (i32 (vector_extract v4i32:$A, 0))));
3626 dag LE_A1 = (i64 (sext (i32 (vector_extract v4i32:$A, 2))));
3627 dag BE_A0 = (i64 (sext (i32 (vector_extract v4i32:$A, 1))));
3628 dag BE_A1 = (i64 (sext (i32 (vector_extract v4i32:$A, 3))));
3632 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (extloadf32 xoaddr:$A)))));
3635 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (extloadf32 xoaddr:$A)))));
3638 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (extloadf32 xoaddr:$A)))));
3640 def FltToLongLoadP9 {
3641 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (extloadf32 ixaddr:$A)))));
3643 def FltToULongLoad {
3644 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (extloadf32 xoaddr:$A)))));
3646 def FltToULongLoadP9 {
3647 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (extloadf32 ixaddr:$A)))));
3650 dag A = (i64 (PPCmfvsr (f64 (PPCfctidz (fpextend f32:$A)))));
3653 dag A = (i64 (PPCmfvsr (f64 (PPCfctiduz (fpextend f32:$A)))));
3656 dag A = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$A))));
3657 dag B = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$B))));
3658 dag C = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$C))));
3659 dag D = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$D))));
3662 dag A = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$A))));
3663 dag B = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$B))));
3664 dag C = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$C))));
3665 dag D = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$D))));
3668 dag A = (i64 (PPCmfvsr (f64 (PPCfctidz f64:$A))));
3671 dag A = (i64 (PPCmfvsr (f64 (PPCfctiduz f64:$A))));
3674 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (load xoaddr:$A)))));
3676 def DblToIntLoadP9 {
3677 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (load ixaddr:$A)))));
3680 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (load xoaddr:$A)))));
3682 def DblToUIntLoadP9 {
3683 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (load ixaddr:$A)))));
3686 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (load xoaddr:$A)))));
3688 def DblToULongLoad {
3689 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (load xoaddr:$A)))));
3692 // FP merge dags (for f32 -> v4f32)
3694 dag AC = (XVCVDPSP (XXPERMDI (COPY_TO_REGCLASS $A, VSRC),
3695 (COPY_TO_REGCLASS $C, VSRC), 0));
3696 dag BD = (XVCVDPSP (XXPERMDI (COPY_TO_REGCLASS $B, VSRC),
3697 (COPY_TO_REGCLASS $D, VSRC), 0));
3698 dag ABhToFlt = (XVCVDPSP (XXPERMDI $A, $B, 0));
3699 dag ABlToFlt = (XVCVDPSP (XXPERMDI $A, $B, 3));
3700 dag BAhToFlt = (XVCVDPSP (XXPERMDI $B, $A, 0));
3701 dag BAlToFlt = (XVCVDPSP (XXPERMDI $B, $A, 3));
3704 // Word-element merge dags - conversions from f64 to i32 merged into vectors.
3706 // For big endian, we merge low and hi doublewords (A, B).
3707 dag A0B0 = (v2f64 (XXPERMDI v2f64:$A, v2f64:$B, 0));
3708 dag A1B1 = (v2f64 (XXPERMDI v2f64:$A, v2f64:$B, 3));
3709 dag CVA1B1S = (v4i32 (XVCVDPSXWS A1B1));
3710 dag CVA0B0S = (v4i32 (XVCVDPSXWS A0B0));
3711 dag CVA1B1U = (v4i32 (XVCVDPUXWS A1B1));
3712 dag CVA0B0U = (v4i32 (XVCVDPUXWS A0B0));
3714 // For little endian, we merge low and hi doublewords (B, A).
3715 dag B1A1 = (v2f64 (XXPERMDI v2f64:$B, v2f64:$A, 0));
3716 dag B0A0 = (v2f64 (XXPERMDI v2f64:$B, v2f64:$A, 3));
3717 dag CVB1A1S = (v4i32 (XVCVDPSXWS B1A1));
3718 dag CVB0A0S = (v4i32 (XVCVDPSXWS B0A0));
3719 dag CVB1A1U = (v4i32 (XVCVDPUXWS B1A1));
3720 dag CVB0A0U = (v4i32 (XVCVDPUXWS B0A0));
3722 // For big endian, we merge hi doublewords of (A, C) and (B, D), convert
3724 dag AC = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$A, VSRC),
3725 (COPY_TO_REGCLASS f64:$C, VSRC), 0));
3726 dag BD = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$B, VSRC),
3727 (COPY_TO_REGCLASS f64:$D, VSRC), 0));
3728 dag CVACS = (v4i32 (XVCVDPSXWS AC));
3729 dag CVBDS = (v4i32 (XVCVDPSXWS BD));
3730 dag CVACU = (v4i32 (XVCVDPUXWS AC));
3731 dag CVBDU = (v4i32 (XVCVDPUXWS BD));
3733 // For little endian, we merge hi doublewords of (D, B) and (C, A), convert
3735 dag DB = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$D, VSRC),
3736 (COPY_TO_REGCLASS f64:$B, VSRC), 0));
3737 dag CA = (v2f64 (XXPERMDI (COPY_TO_REGCLASS f64:$C, VSRC),
3738 (COPY_TO_REGCLASS f64:$A, VSRC), 0));
3739 dag CVDBS = (v4i32 (XVCVDPSXWS DB));
3740 dag CVCAS = (v4i32 (XVCVDPSXWS CA));
3741 dag CVDBU = (v4i32 (XVCVDPUXWS DB));
3742 dag CVCAU = (v4i32 (XVCVDPUXWS CA));
3745 // Patterns for BUILD_VECTOR nodes.
3746 let AddedComplexity = 400 in {
3748 let Predicates = [HasVSX] in {
3749 // Build vectors of floating point converted to i32.
3750 def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.A,
3751 DblToInt.A, DblToInt.A)),
3752 (v4i32 (XXSPLTW (COPY_TO_REGCLASS (XSCVDPSXWS $A), VSRC), 1))>;
3753 def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.A,
3754 DblToUInt.A, DblToUInt.A)),
3755 (v4i32 (XXSPLTW (COPY_TO_REGCLASS (XSCVDPUXWS $A), VSRC), 1))>;
3756 def : Pat<(v2i64 (build_vector DblToLong.A, DblToLong.A)),
3757 (v2i64 (XXPERMDI (COPY_TO_REGCLASS (XSCVDPSXDS $A), VSRC),
3758 (COPY_TO_REGCLASS (XSCVDPSXDS $A), VSRC), 0))>;
3759 def : Pat<(v2i64 (build_vector DblToULong.A, DblToULong.A)),
3760 (v2i64 (XXPERMDI (COPY_TO_REGCLASS (XSCVDPUXDS $A), VSRC),
3761 (COPY_TO_REGCLASS (XSCVDPUXDS $A), VSRC), 0))>;
3762 def : Pat<(v4i32 (scalar_to_vector FltToIntLoad.A)),
3763 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3764 (XSCVDPSXWSs (XFLOADf32 xoaddr:$A)), VSRC), 1))>;
3765 def : Pat<(v4i32 (scalar_to_vector FltToUIntLoad.A)),
3766 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3767 (XSCVDPUXWSs (XFLOADf32 xoaddr:$A)), VSRC), 1))>;
3768 def : Pat<(v4f32 (build_vector f32:$A, f32:$A, f32:$A, f32:$A)),
3769 (v4f32 (XXSPLTW (v4f32 (XSCVDPSPN $A)), 0))>;
3771 // Build vectors of floating point converted to i64.
3772 def : Pat<(v2i64 (build_vector FltToLong.A, FltToLong.A)),
3774 (COPY_TO_REGCLASS (XSCVDPSXDSs $A), VSFRC), 0))>;
3775 def : Pat<(v2i64 (build_vector FltToULong.A, FltToULong.A)),
3777 (COPY_TO_REGCLASS (XSCVDPUXDSs $A), VSFRC), 0))>;
3778 def : Pat<(v2i64 (scalar_to_vector DblToLongLoad.A)),
3779 (v2i64 (XVCVDPSXDS (LXVDSX xoaddr:$A)))>;
3780 def : Pat<(v2i64 (scalar_to_vector DblToULongLoad.A)),
3781 (v2i64 (XVCVDPUXDS (LXVDSX xoaddr:$A)))>;
3784 let Predicates = [HasVSX, NoP9Vector] in {
3785 // Load-and-splat with fp-to-int conversion (using X-Form VSX/FP loads).
3786 def : Pat<(v4i32 (scalar_to_vector DblToIntLoad.A)),
3787 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3788 (XSCVDPSXWS (XFLOADf64 xoaddr:$A)), VSRC), 1))>;
3789 def : Pat<(v4i32 (scalar_to_vector DblToUIntLoad.A)),
3790 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3791 (XSCVDPUXWS (XFLOADf64 xoaddr:$A)), VSRC), 1))>;
3792 def : Pat<(v2i64 (scalar_to_vector FltToLongLoad.A)),
3793 (v2i64 (XXPERMDIs (XSCVDPSXDS (COPY_TO_REGCLASS
3794 (XFLOADf32 xoaddr:$A), VSFRC)), 0))>;
3795 def : Pat<(v2i64 (scalar_to_vector FltToULongLoad.A)),
3796 (v2i64 (XXPERMDIs (XSCVDPUXDS (COPY_TO_REGCLASS
3797 (XFLOADf32 xoaddr:$A), VSFRC)), 0))>;
3800 // Big endian, available on all targets with VSX
3801 let Predicates = [IsBigEndian, HasVSX] in {
3802 def : Pat<(v2f64 (build_vector f64:$A, f64:$B)),
3804 (COPY_TO_REGCLASS $A, VSRC),
3805 (COPY_TO_REGCLASS $B, VSRC), 0))>;
3807 def : Pat<(v4f32 (build_vector f32:$A, f32:$B, f32:$C, f32:$D)),
3808 (VMRGEW MrgFP.AC, MrgFP.BD)>;
3809 def : Pat<(v4f32 (build_vector DblToFlt.A0, DblToFlt.A1,
3810 DblToFlt.B0, DblToFlt.B1)),
3811 (v4f32 (VMRGEW MrgFP.ABhToFlt, MrgFP.ABlToFlt))>;
3813 // Convert 4 doubles to a vector of ints.
3814 def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.B,
3815 DblToInt.C, DblToInt.D)),
3816 (v4i32 (VMRGEW MrgWords.CVACS, MrgWords.CVBDS))>;
3817 def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.B,
3818 DblToUInt.C, DblToUInt.D)),
3819 (v4i32 (VMRGEW MrgWords.CVACU, MrgWords.CVBDU))>;
3820 def : Pat<(v4i32 (build_vector ExtDbl.A0S, ExtDbl.A1S,
3821 ExtDbl.B0S, ExtDbl.B1S)),
3822 (v4i32 (VMRGEW MrgWords.CVA0B0S, MrgWords.CVA1B1S))>;
3823 def : Pat<(v4i32 (build_vector ExtDbl.A0U, ExtDbl.A1U,
3824 ExtDbl.B0U, ExtDbl.B1U)),
3825 (v4i32 (VMRGEW MrgWords.CVA0B0U, MrgWords.CVA1B1U))>;
3828 let Predicates = [IsLittleEndian, HasVSX] in {
3829 // Little endian, available on all targets with VSX
3830 def : Pat<(v2f64 (build_vector f64:$A, f64:$B)),
3832 (COPY_TO_REGCLASS $B, VSRC),
3833 (COPY_TO_REGCLASS $A, VSRC), 0))>;
3835 def : Pat<(v4f32 (build_vector f32:$D, f32:$C, f32:$B, f32:$A)),
3836 (VMRGEW MrgFP.AC, MrgFP.BD)>;
3837 def : Pat<(v4f32 (build_vector DblToFlt.A0, DblToFlt.A1,
3838 DblToFlt.B0, DblToFlt.B1)),
3839 (v4f32 (VMRGEW MrgFP.BAhToFlt, MrgFP.BAlToFlt))>;
3841 // Convert 4 doubles to a vector of ints.
3842 def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.B,
3843 DblToInt.C, DblToInt.D)),
3844 (v4i32 (VMRGEW MrgWords.CVDBS, MrgWords.CVCAS))>;
3845 def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.B,
3846 DblToUInt.C, DblToUInt.D)),
3847 (v4i32 (VMRGEW MrgWords.CVDBU, MrgWords.CVCAU))>;
3848 def : Pat<(v4i32 (build_vector ExtDbl.A0S, ExtDbl.A1S,
3849 ExtDbl.B0S, ExtDbl.B1S)),
3850 (v4i32 (VMRGEW MrgWords.CVB1A1S, MrgWords.CVB0A0S))>;
3851 def : Pat<(v4i32 (build_vector ExtDbl.A0U, ExtDbl.A1U,
3852 ExtDbl.B0U, ExtDbl.B1U)),
3853 (v4i32 (VMRGEW MrgWords.CVB1A1U, MrgWords.CVB0A0U))>;
3856 let Predicates = [HasDirectMove] in {
3857 // Endianness-neutral constant splat on P8 and newer targets. The reason
3858 // for this pattern is that on targets with direct moves, we don't expand
3859 // BUILD_VECTOR nodes for v4i32.
3860 def : Pat<(v4i32 (build_vector immSExt5NonZero:$A, immSExt5NonZero:$A,
3861 immSExt5NonZero:$A, immSExt5NonZero:$A)),
3862 (v4i32 (VSPLTISW imm:$A))>;
3865 let Predicates = [IsBigEndian, HasDirectMove, NoP9Vector] in {
3866 // Big endian integer vectors using direct moves.
3867 def : Pat<(v2i64 (build_vector i64:$A, i64:$B)),
3869 (COPY_TO_REGCLASS (MTVSRD $A), VSRC),
3870 (COPY_TO_REGCLASS (MTVSRD $B), VSRC), 0))>;
3871 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
3872 (VMRGOW (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC),
3873 (COPY_TO_REGCLASS (MTVSRWZ $C), VSRC), 0),
3874 (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC),
3875 (COPY_TO_REGCLASS (MTVSRWZ $D), VSRC), 0))>;
3876 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
3877 (XXSPLTW (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 1)>;
3880 let Predicates = [IsLittleEndian, HasDirectMove, NoP9Vector] in {
3881 // Little endian integer vectors using direct moves.
3882 def : Pat<(v2i64 (build_vector i64:$A, i64:$B)),
3884 (COPY_TO_REGCLASS (MTVSRD $B), VSRC),
3885 (COPY_TO_REGCLASS (MTVSRD $A), VSRC), 0))>;
3886 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
3887 (VMRGOW (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $D), VSRC),
3888 (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC), 0),
3889 (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $C), VSRC),
3890 (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 0))>;
3891 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
3892 (XXSPLTW (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 1)>;
3895 let Predicates = [HasP9Vector] in {
3896 // Endianness-neutral patterns for const splats with ISA 3.0 instructions.
3897 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
3898 (v4i32 (MTVSRWS $A))>;
3899 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
3900 (v4i32 (MTVSRWS $A))>;
3901 def : Pat<(v16i8 (build_vector immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3902 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3903 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3904 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3905 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
3907 (v16i8 (COPY_TO_REGCLASS (XXSPLTIB imm:$A), VSRC))>;
3908 def : Pat<(v16i8 immAllOnesV),
3909 (v16i8 (COPY_TO_REGCLASS (XXSPLTIB 255), VSRC))>;
3910 def : Pat<(v8i16 immAllOnesV),
3911 (v8i16 (COPY_TO_REGCLASS (XXSPLTIB 255), VSRC))>;
3912 def : Pat<(v4i32 immAllOnesV),
3913 (v4i32 (XXSPLTIB 255))>;
3914 def : Pat<(v2i64 immAllOnesV),
3915 (v2i64 (XXSPLTIB 255))>;
3916 def : Pat<(v4i32 (scalar_to_vector FltToIntLoad.A)),
3917 (v4i32 (XVCVSPSXWS (LXVWSX xoaddr:$A)))>;
3918 def : Pat<(v4i32 (scalar_to_vector FltToUIntLoad.A)),
3919 (v4i32 (XVCVSPUXWS (LXVWSX xoaddr:$A)))>;
3920 def : Pat<(v4i32 (scalar_to_vector DblToIntLoadP9.A)),
3921 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3922 (XSCVDPSXWS (DFLOADf64 ixaddr:$A)), VSRC), 1))>;
3923 def : Pat<(v4i32 (scalar_to_vector DblToUIntLoadP9.A)),
3924 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
3925 (XSCVDPUXWS (DFLOADf64 ixaddr:$A)), VSRC), 1))>;
3926 def : Pat<(v2i64 (scalar_to_vector FltToLongLoadP9.A)),
3927 (v2i64 (XXPERMDIs (XSCVDPSXDS (COPY_TO_REGCLASS
3928 (DFLOADf32 ixaddr:$A),
3930 def : Pat<(v2i64 (scalar_to_vector FltToULongLoadP9.A)),
3931 (v2i64 (XXPERMDIs (XSCVDPUXDS (COPY_TO_REGCLASS
3932 (DFLOADf32 ixaddr:$A),
3936 let Predicates = [IsISA3_0, HasDirectMove, IsBigEndian] in {
3937 def : Pat<(i64 (extractelt v2i64:$A, 1)),
3938 (i64 (MFVSRLD $A))>;
3939 // Better way to build integer vectors if we have MTVSRDD. Big endian.
3940 def : Pat<(v2i64 (build_vector i64:$rB, i64:$rA)),
3941 (v2i64 (MTVSRDD $rB, $rA))>;
3942 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
3944 (v4i32 (COPY_TO_REGCLASS (MTVSRDD AnyExts.A, AnyExts.C), VSRC)),
3946 (COPY_TO_REGCLASS (MTVSRDD AnyExts.B, AnyExts.D), VSRC)))>;
3949 let Predicates = [IsISA3_0, HasDirectMove, IsLittleEndian] in {
3950 def : Pat<(i64 (extractelt v2i64:$A, 0)),
3951 (i64 (MFVSRLD $A))>;
3952 // Better way to build integer vectors if we have MTVSRDD. Little endian.
3953 def : Pat<(v2i64 (build_vector i64:$rA, i64:$rB)),
3954 (v2i64 (MTVSRDD $rB, $rA))>;
3955 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
3957 (v4i32 (COPY_TO_REGCLASS (MTVSRDD AnyExts.D, AnyExts.B), VSRC)),
3959 (COPY_TO_REGCLASS (MTVSRDD AnyExts.C, AnyExts.A), VSRC)))>;
3961 // P9 Altivec instructions that can be used to build vectors.
3962 // Adding them to PPCInstrVSX.td rather than PPCAltivecVSX.td to compete
3963 // with complexities of existing build vector patterns in this file.
3964 let Predicates = [HasP9Altivec, IsLittleEndian] in {
3965 def : Pat<(v2i64 (build_vector WordToDWord.LE_A0, WordToDWord.LE_A1)),
3966 (v2i64 (VEXTSW2D $A))>;
3967 def : Pat<(v2i64 (build_vector HWordToDWord.LE_A0, HWordToDWord.LE_A1)),
3968 (v2i64 (VEXTSH2D $A))>;
3969 def : Pat<(v4i32 (build_vector HWordToWord.LE_A0, HWordToWord.LE_A1,
3970 HWordToWord.LE_A2, HWordToWord.LE_A3)),
3971 (v4i32 (VEXTSH2W $A))>;
3972 def : Pat<(v4i32 (build_vector ByteToWord.LE_A0, ByteToWord.LE_A1,
3973 ByteToWord.LE_A2, ByteToWord.LE_A3)),
3974 (v4i32 (VEXTSB2W $A))>;
3975 def : Pat<(v2i64 (build_vector ByteToDWord.LE_A0, ByteToDWord.LE_A1)),
3976 (v2i64 (VEXTSB2D $A))>;
3979 let Predicates = [HasP9Altivec, IsBigEndian] in {
3980 def : Pat<(v2i64 (build_vector WordToDWord.BE_A0, WordToDWord.BE_A1)),
3981 (v2i64 (VEXTSW2D $A))>;
3982 def : Pat<(v2i64 (build_vector HWordToDWord.BE_A0, HWordToDWord.BE_A1)),
3983 (v2i64 (VEXTSH2D $A))>;
3984 def : Pat<(v4i32 (build_vector HWordToWord.BE_A0, HWordToWord.BE_A1,
3985 HWordToWord.BE_A2, HWordToWord.BE_A3)),
3986 (v4i32 (VEXTSH2W $A))>;
3987 def : Pat<(v4i32 (build_vector ByteToWord.BE_A0, ByteToWord.BE_A1,
3988 ByteToWord.BE_A2, ByteToWord.BE_A3)),
3989 (v4i32 (VEXTSB2W $A))>;
3990 def : Pat<(v2i64 (build_vector ByteToDWord.BE_A0, ByteToDWord.BE_A1)),
3991 (v2i64 (VEXTSB2D $A))>;
3994 let Predicates = [HasP9Altivec] in {
3995 def: Pat<(v2i64 (PPCSExtVElems v16i8:$A)),
3996 (v2i64 (VEXTSB2D $A))>;
3997 def: Pat<(v2i64 (PPCSExtVElems v8i16:$A)),
3998 (v2i64 (VEXTSH2D $A))>;
3999 def: Pat<(v2i64 (PPCSExtVElems v4i32:$A)),
4000 (v2i64 (VEXTSW2D $A))>;
4001 def: Pat<(v4i32 (PPCSExtVElems v16i8:$A)),
4002 (v4i32 (VEXTSB2W $A))>;
4003 def: Pat<(v4i32 (PPCSExtVElems v8i16:$A)),
4004 (v4i32 (VEXTSH2W $A))>;