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 // Little-endian-specific nodes.
51 def SDT_PPClxvd2x : SDTypeProfile<1, 1, [
52 SDTCisVT<0, v2f64>, SDTCisPtrTy<1>
54 def SDT_PPCstxvd2x : SDTypeProfile<0, 2, [
55 SDTCisVT<0, v2f64>, SDTCisPtrTy<1>
57 def SDT_PPCxxswapd : SDTypeProfile<1, 1, [
60 def SDTVecConv : SDTypeProfile<1, 2, [
61 SDTCisVec<0>, SDTCisVec<1>, SDTCisPtrTy<2>
64 def PPClxvd2x : SDNode<"PPCISD::LXVD2X", SDT_PPClxvd2x,
65 [SDNPHasChain, SDNPMayLoad, SDNPMemOperand]>;
66 def PPCstxvd2x : SDNode<"PPCISD::STXVD2X", SDT_PPCstxvd2x,
67 [SDNPHasChain, SDNPMayStore]>;
68 def PPCxxswapd : SDNode<"PPCISD::XXSWAPD", SDT_PPCxxswapd, [SDNPHasChain]>;
69 def PPCmfvsr : SDNode<"PPCISD::MFVSR", SDTUnaryOp, []>;
70 def PPCmtvsra : SDNode<"PPCISD::MTVSRA", SDTUnaryOp, []>;
71 def PPCmtvsrz : SDNode<"PPCISD::MTVSRZ", SDTUnaryOp, []>;
72 def PPCsvec2fp : SDNode<"PPCISD::SINT_VEC_TO_FP", SDTVecConv, []>;
73 def PPCuvec2fp: SDNode<"PPCISD::UINT_VEC_TO_FP", SDTVecConv, []>;
74 def PPCswapNoChain : SDNode<"PPCISD::SWAP_NO_CHAIN", SDT_PPCxxswapd>;
76 multiclass XX3Form_Rcr<bits<6> opcode, bits<7> xo, string asmbase,
77 string asmstr, InstrItinClass itin, Intrinsic Int,
78 ValueType OutTy, ValueType InTy> {
79 let BaseName = asmbase in {
80 def NAME : XX3Form_Rc<opcode, xo, (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
81 !strconcat(asmbase, !strconcat(" ", asmstr)), itin,
82 [(set OutTy:$XT, (Int InTy:$XA, InTy:$XB))]>;
84 def o : XX3Form_Rc<opcode, xo, (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
85 !strconcat(asmbase, !strconcat(". ", asmstr)), itin,
87 (InTy (PPCvcmp_o InTy:$XA, InTy:$XB, xo)))]>,
92 // Instruction form with a single input register for instructions such as
93 // XXPERMDI. The reason for defining this is that specifying multiple chained
94 // operands (such as loads) to an instruction will perform both chained
95 // operations rather than coalescing them into a single register - even though
96 // the source memory location is the same. This simply forces the instruction
97 // to use the same register for both inputs.
98 // For example, an output DAG such as this:
99 // (XXPERMDI (LXSIBZX xoaddr:$src), (LXSIBZX xoaddr:$src ), 0))
100 // would result in two load instructions emitted and used as separate inputs
101 // to the XXPERMDI instruction.
102 class XX3Form_2s<bits<6> opcode, bits<5> xo, dag OOL, dag IOL, string asmstr,
103 InstrItinClass itin, list<dag> pattern>
104 : XX3Form_2<opcode, xo, OOL, IOL, asmstr, itin, pattern> {
108 def HasVSX : Predicate<"PPCSubTarget->hasVSX()">;
109 def IsLittleEndian : Predicate<"PPCSubTarget->isLittleEndian()">;
110 def IsBigEndian : Predicate<"!PPCSubTarget->isLittleEndian()">;
111 def HasOnlySwappingMemOps : Predicate<"!PPCSubTarget->hasP9Vector()">;
113 let Predicates = [HasVSX] in {
114 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
115 let UseVSXReg = 1 in {
116 let hasSideEffects = 0 in { // VSX instructions don't have side effects.
119 // Load indexed instructions
120 let mayLoad = 1, mayStore = 0 in {
122 def LXSDX : XX1Form<31, 588,
123 (outs vsfrc:$XT), (ins memrr:$src),
124 "lxsdx $XT, $src", IIC_LdStLFD,
125 [(set f64:$XT, (load xoaddr:$src))]>;
127 let Predicates = [HasVSX, HasOnlySwappingMemOps] in
128 def LXVD2X : XX1Form<31, 844,
129 (outs vsrc:$XT), (ins memrr:$src),
130 "lxvd2x $XT, $src", IIC_LdStLFD,
131 [(set v2f64:$XT, (int_ppc_vsx_lxvd2x xoaddr:$src))]>;
133 def LXVDSX : XX1Form<31, 332,
134 (outs vsrc:$XT), (ins memrr:$src),
135 "lxvdsx $XT, $src", IIC_LdStLFD, []>;
137 let Predicates = [HasVSX, HasOnlySwappingMemOps] in
138 def LXVW4X : XX1Form<31, 780,
139 (outs vsrc:$XT), (ins memrr:$src),
140 "lxvw4x $XT, $src", IIC_LdStLFD,
144 // Store indexed instructions
145 let mayStore = 1, mayLoad = 0 in {
147 def STXSDX : XX1Form<31, 716,
148 (outs), (ins vsfrc:$XT, memrr:$dst),
149 "stxsdx $XT, $dst", IIC_LdStSTFD,
150 [(store f64:$XT, xoaddr:$dst)]>;
152 let Predicates = [HasVSX, HasOnlySwappingMemOps] in {
153 // The behaviour of this instruction is endianness-specific so we provide no
154 // pattern to match it without considering endianness.
155 def STXVD2X : XX1Form<31, 972,
156 (outs), (ins vsrc:$XT, memrr:$dst),
157 "stxvd2x $XT, $dst", IIC_LdStSTFD,
160 def STXVW4X : XX1Form<31, 908,
161 (outs), (ins vsrc:$XT, memrr:$dst),
162 "stxvw4x $XT, $dst", IIC_LdStSTFD,
167 // Add/Mul Instructions
168 let isCommutable = 1 in {
169 def XSADDDP : XX3Form<60, 32,
170 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
171 "xsadddp $XT, $XA, $XB", IIC_VecFP,
172 [(set f64:$XT, (fadd f64:$XA, f64:$XB))]>;
173 def XSMULDP : XX3Form<60, 48,
174 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
175 "xsmuldp $XT, $XA, $XB", IIC_VecFP,
176 [(set f64:$XT, (fmul f64:$XA, f64:$XB))]>;
178 def XVADDDP : XX3Form<60, 96,
179 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
180 "xvadddp $XT, $XA, $XB", IIC_VecFP,
181 [(set v2f64:$XT, (fadd v2f64:$XA, v2f64:$XB))]>;
183 def XVADDSP : XX3Form<60, 64,
184 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
185 "xvaddsp $XT, $XA, $XB", IIC_VecFP,
186 [(set v4f32:$XT, (fadd v4f32:$XA, v4f32:$XB))]>;
188 def XVMULDP : XX3Form<60, 112,
189 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
190 "xvmuldp $XT, $XA, $XB", IIC_VecFP,
191 [(set v2f64:$XT, (fmul v2f64:$XA, v2f64:$XB))]>;
193 def XVMULSP : XX3Form<60, 80,
194 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
195 "xvmulsp $XT, $XA, $XB", IIC_VecFP,
196 [(set v4f32:$XT, (fmul v4f32:$XA, v4f32:$XB))]>;
199 // Subtract Instructions
200 def XSSUBDP : XX3Form<60, 40,
201 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
202 "xssubdp $XT, $XA, $XB", IIC_VecFP,
203 [(set f64:$XT, (fsub f64:$XA, f64:$XB))]>;
205 def XVSUBDP : XX3Form<60, 104,
206 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
207 "xvsubdp $XT, $XA, $XB", IIC_VecFP,
208 [(set v2f64:$XT, (fsub v2f64:$XA, v2f64:$XB))]>;
209 def XVSUBSP : XX3Form<60, 72,
210 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
211 "xvsubsp $XT, $XA, $XB", IIC_VecFP,
212 [(set v4f32:$XT, (fsub v4f32:$XA, v4f32:$XB))]>;
215 let BaseName = "XSMADDADP" in {
216 let isCommutable = 1 in
217 def XSMADDADP : XX3Form<60, 33,
218 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
219 "xsmaddadp $XT, $XA, $XB", IIC_VecFP,
220 [(set f64:$XT, (fma f64:$XA, f64:$XB, f64:$XTi))]>,
221 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
223 let IsVSXFMAAlt = 1 in
224 def XSMADDMDP : XX3Form<60, 41,
225 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
226 "xsmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
227 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
231 let BaseName = "XSMSUBADP" in {
232 let isCommutable = 1 in
233 def XSMSUBADP : XX3Form<60, 49,
234 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
235 "xsmsubadp $XT, $XA, $XB", IIC_VecFP,
236 [(set f64:$XT, (fma f64:$XA, f64:$XB, (fneg f64:$XTi)))]>,
237 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
239 let IsVSXFMAAlt = 1 in
240 def XSMSUBMDP : XX3Form<60, 57,
241 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
242 "xsmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
243 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
247 let BaseName = "XSNMADDADP" in {
248 let isCommutable = 1 in
249 def XSNMADDADP : XX3Form<60, 161,
250 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
251 "xsnmaddadp $XT, $XA, $XB", IIC_VecFP,
252 [(set f64:$XT, (fneg (fma f64:$XA, f64:$XB, f64:$XTi)))]>,
253 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
255 let IsVSXFMAAlt = 1 in
256 def XSNMADDMDP : XX3Form<60, 169,
257 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
258 "xsnmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
259 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
263 let BaseName = "XSNMSUBADP" in {
264 let isCommutable = 1 in
265 def XSNMSUBADP : XX3Form<60, 177,
266 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
267 "xsnmsubadp $XT, $XA, $XB", IIC_VecFP,
268 [(set f64:$XT, (fneg (fma f64:$XA, f64:$XB, (fneg f64:$XTi))))]>,
269 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
271 let IsVSXFMAAlt = 1 in
272 def XSNMSUBMDP : XX3Form<60, 185,
273 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
274 "xsnmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
275 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
279 let BaseName = "XVMADDADP" in {
280 let isCommutable = 1 in
281 def XVMADDADP : XX3Form<60, 97,
282 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
283 "xvmaddadp $XT, $XA, $XB", IIC_VecFP,
284 [(set v2f64:$XT, (fma v2f64:$XA, v2f64:$XB, v2f64:$XTi))]>,
285 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
287 let IsVSXFMAAlt = 1 in
288 def XVMADDMDP : XX3Form<60, 105,
289 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
290 "xvmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
291 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
295 let BaseName = "XVMADDASP" in {
296 let isCommutable = 1 in
297 def XVMADDASP : XX3Form<60, 65,
298 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
299 "xvmaddasp $XT, $XA, $XB", IIC_VecFP,
300 [(set v4f32:$XT, (fma v4f32:$XA, v4f32:$XB, v4f32:$XTi))]>,
301 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
303 let IsVSXFMAAlt = 1 in
304 def XVMADDMSP : XX3Form<60, 73,
305 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
306 "xvmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
307 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
311 let BaseName = "XVMSUBADP" in {
312 let isCommutable = 1 in
313 def XVMSUBADP : XX3Form<60, 113,
314 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
315 "xvmsubadp $XT, $XA, $XB", IIC_VecFP,
316 [(set v2f64:$XT, (fma v2f64:$XA, v2f64:$XB, (fneg v2f64:$XTi)))]>,
317 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
319 let IsVSXFMAAlt = 1 in
320 def XVMSUBMDP : XX3Form<60, 121,
321 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
322 "xvmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
323 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
327 let BaseName = "XVMSUBASP" in {
328 let isCommutable = 1 in
329 def XVMSUBASP : XX3Form<60, 81,
330 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
331 "xvmsubasp $XT, $XA, $XB", IIC_VecFP,
332 [(set v4f32:$XT, (fma v4f32:$XA, v4f32:$XB, (fneg v4f32:$XTi)))]>,
333 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
335 let IsVSXFMAAlt = 1 in
336 def XVMSUBMSP : XX3Form<60, 89,
337 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
338 "xvmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
339 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
343 let BaseName = "XVNMADDADP" in {
344 let isCommutable = 1 in
345 def XVNMADDADP : XX3Form<60, 225,
346 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
347 "xvnmaddadp $XT, $XA, $XB", IIC_VecFP,
348 [(set v2f64:$XT, (fneg (fma v2f64:$XA, v2f64:$XB, v2f64:$XTi)))]>,
349 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
351 let IsVSXFMAAlt = 1 in
352 def XVNMADDMDP : XX3Form<60, 233,
353 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
354 "xvnmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
355 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
359 let BaseName = "XVNMADDASP" in {
360 let isCommutable = 1 in
361 def XVNMADDASP : XX3Form<60, 193,
362 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
363 "xvnmaddasp $XT, $XA, $XB", IIC_VecFP,
364 [(set v4f32:$XT, (fneg (fma v4f32:$XA, v4f32:$XB, v4f32:$XTi)))]>,
365 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
367 let IsVSXFMAAlt = 1 in
368 def XVNMADDMSP : XX3Form<60, 201,
369 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
370 "xvnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
371 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
375 let BaseName = "XVNMSUBADP" in {
376 let isCommutable = 1 in
377 def XVNMSUBADP : XX3Form<60, 241,
378 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
379 "xvnmsubadp $XT, $XA, $XB", IIC_VecFP,
380 [(set v2f64:$XT, (fneg (fma v2f64:$XA, v2f64:$XB, (fneg v2f64:$XTi))))]>,
381 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
383 let IsVSXFMAAlt = 1 in
384 def XVNMSUBMDP : XX3Form<60, 249,
385 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
386 "xvnmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
387 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
391 let BaseName = "XVNMSUBASP" in {
392 let isCommutable = 1 in
393 def XVNMSUBASP : XX3Form<60, 209,
394 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
395 "xvnmsubasp $XT, $XA, $XB", IIC_VecFP,
396 [(set v4f32:$XT, (fneg (fma v4f32:$XA, v4f32:$XB, (fneg v4f32:$XTi))))]>,
397 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
399 let IsVSXFMAAlt = 1 in
400 def XVNMSUBMSP : XX3Form<60, 217,
401 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
402 "xvnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
403 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
407 // Division Instructions
408 def XSDIVDP : XX3Form<60, 56,
409 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
410 "xsdivdp $XT, $XA, $XB", IIC_FPDivD,
411 [(set f64:$XT, (fdiv f64:$XA, f64:$XB))]>;
412 def XSSQRTDP : XX2Form<60, 75,
413 (outs vsfrc:$XT), (ins vsfrc:$XB),
414 "xssqrtdp $XT, $XB", IIC_FPSqrtD,
415 [(set f64:$XT, (fsqrt f64:$XB))]>;
417 def XSREDP : XX2Form<60, 90,
418 (outs vsfrc:$XT), (ins vsfrc:$XB),
419 "xsredp $XT, $XB", IIC_VecFP,
420 [(set f64:$XT, (PPCfre f64:$XB))]>;
421 def XSRSQRTEDP : XX2Form<60, 74,
422 (outs vsfrc:$XT), (ins vsfrc:$XB),
423 "xsrsqrtedp $XT, $XB", IIC_VecFP,
424 [(set f64:$XT, (PPCfrsqrte f64:$XB))]>;
426 def XSTDIVDP : XX3Form_1<60, 61,
427 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
428 "xstdivdp $crD, $XA, $XB", IIC_FPCompare, []>;
429 def XSTSQRTDP : XX2Form_1<60, 106,
430 (outs crrc:$crD), (ins vsfrc:$XB),
431 "xstsqrtdp $crD, $XB", IIC_FPCompare, []>;
433 def XVDIVDP : XX3Form<60, 120,
434 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
435 "xvdivdp $XT, $XA, $XB", IIC_FPDivD,
436 [(set v2f64:$XT, (fdiv v2f64:$XA, v2f64:$XB))]>;
437 def XVDIVSP : XX3Form<60, 88,
438 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
439 "xvdivsp $XT, $XA, $XB", IIC_FPDivS,
440 [(set v4f32:$XT, (fdiv v4f32:$XA, v4f32:$XB))]>;
442 def XVSQRTDP : XX2Form<60, 203,
443 (outs vsrc:$XT), (ins vsrc:$XB),
444 "xvsqrtdp $XT, $XB", IIC_FPSqrtD,
445 [(set v2f64:$XT, (fsqrt v2f64:$XB))]>;
446 def XVSQRTSP : XX2Form<60, 139,
447 (outs vsrc:$XT), (ins vsrc:$XB),
448 "xvsqrtsp $XT, $XB", IIC_FPSqrtS,
449 [(set v4f32:$XT, (fsqrt v4f32:$XB))]>;
451 def XVTDIVDP : XX3Form_1<60, 125,
452 (outs crrc:$crD), (ins vsrc:$XA, vsrc:$XB),
453 "xvtdivdp $crD, $XA, $XB", IIC_FPCompare, []>;
454 def XVTDIVSP : XX3Form_1<60, 93,
455 (outs crrc:$crD), (ins vsrc:$XA, vsrc:$XB),
456 "xvtdivsp $crD, $XA, $XB", IIC_FPCompare, []>;
458 def XVTSQRTDP : XX2Form_1<60, 234,
459 (outs crrc:$crD), (ins vsrc:$XB),
460 "xvtsqrtdp $crD, $XB", IIC_FPCompare, []>;
461 def XVTSQRTSP : XX2Form_1<60, 170,
462 (outs crrc:$crD), (ins vsrc:$XB),
463 "xvtsqrtsp $crD, $XB", IIC_FPCompare, []>;
465 def XVREDP : XX2Form<60, 218,
466 (outs vsrc:$XT), (ins vsrc:$XB),
467 "xvredp $XT, $XB", IIC_VecFP,
468 [(set v2f64:$XT, (PPCfre v2f64:$XB))]>;
469 def XVRESP : XX2Form<60, 154,
470 (outs vsrc:$XT), (ins vsrc:$XB),
471 "xvresp $XT, $XB", IIC_VecFP,
472 [(set v4f32:$XT, (PPCfre v4f32:$XB))]>;
474 def XVRSQRTEDP : XX2Form<60, 202,
475 (outs vsrc:$XT), (ins vsrc:$XB),
476 "xvrsqrtedp $XT, $XB", IIC_VecFP,
477 [(set v2f64:$XT, (PPCfrsqrte v2f64:$XB))]>;
478 def XVRSQRTESP : XX2Form<60, 138,
479 (outs vsrc:$XT), (ins vsrc:$XB),
480 "xvrsqrtesp $XT, $XB", IIC_VecFP,
481 [(set v4f32:$XT, (PPCfrsqrte v4f32:$XB))]>;
483 // Compare Instructions
484 def XSCMPODP : XX3Form_1<60, 43,
485 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
486 "xscmpodp $crD, $XA, $XB", IIC_FPCompare, []>;
487 def XSCMPUDP : XX3Form_1<60, 35,
488 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
489 "xscmpudp $crD, $XA, $XB", IIC_FPCompare, []>;
491 defm XVCMPEQDP : XX3Form_Rcr<60, 99,
492 "xvcmpeqdp", "$XT, $XA, $XB", IIC_VecFPCompare,
493 int_ppc_vsx_xvcmpeqdp, v2i64, v2f64>;
494 defm XVCMPEQSP : XX3Form_Rcr<60, 67,
495 "xvcmpeqsp", "$XT, $XA, $XB", IIC_VecFPCompare,
496 int_ppc_vsx_xvcmpeqsp, v4i32, v4f32>;
497 defm XVCMPGEDP : XX3Form_Rcr<60, 115,
498 "xvcmpgedp", "$XT, $XA, $XB", IIC_VecFPCompare,
499 int_ppc_vsx_xvcmpgedp, v2i64, v2f64>;
500 defm XVCMPGESP : XX3Form_Rcr<60, 83,
501 "xvcmpgesp", "$XT, $XA, $XB", IIC_VecFPCompare,
502 int_ppc_vsx_xvcmpgesp, v4i32, v4f32>;
503 defm XVCMPGTDP : XX3Form_Rcr<60, 107,
504 "xvcmpgtdp", "$XT, $XA, $XB", IIC_VecFPCompare,
505 int_ppc_vsx_xvcmpgtdp, v2i64, v2f64>;
506 defm XVCMPGTSP : XX3Form_Rcr<60, 75,
507 "xvcmpgtsp", "$XT, $XA, $XB", IIC_VecFPCompare,
508 int_ppc_vsx_xvcmpgtsp, v4i32, v4f32>;
511 def XSABSDP : XX2Form<60, 345,
512 (outs vsfrc:$XT), (ins vsfrc:$XB),
513 "xsabsdp $XT, $XB", IIC_VecFP,
514 [(set f64:$XT, (fabs f64:$XB))]>;
515 def XSNABSDP : XX2Form<60, 361,
516 (outs vsfrc:$XT), (ins vsfrc:$XB),
517 "xsnabsdp $XT, $XB", IIC_VecFP,
518 [(set f64:$XT, (fneg (fabs f64:$XB)))]>;
519 def XSNEGDP : XX2Form<60, 377,
520 (outs vsfrc:$XT), (ins vsfrc:$XB),
521 "xsnegdp $XT, $XB", IIC_VecFP,
522 [(set f64:$XT, (fneg f64:$XB))]>;
523 def XSCPSGNDP : XX3Form<60, 176,
524 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
525 "xscpsgndp $XT, $XA, $XB", IIC_VecFP,
526 [(set f64:$XT, (fcopysign f64:$XB, f64:$XA))]>;
528 def XVABSDP : XX2Form<60, 473,
529 (outs vsrc:$XT), (ins vsrc:$XB),
530 "xvabsdp $XT, $XB", IIC_VecFP,
531 [(set v2f64:$XT, (fabs v2f64:$XB))]>;
533 def XVABSSP : XX2Form<60, 409,
534 (outs vsrc:$XT), (ins vsrc:$XB),
535 "xvabssp $XT, $XB", IIC_VecFP,
536 [(set v4f32:$XT, (fabs v4f32:$XB))]>;
538 def XVCPSGNDP : XX3Form<60, 240,
539 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
540 "xvcpsgndp $XT, $XA, $XB", IIC_VecFP,
541 [(set v2f64:$XT, (fcopysign v2f64:$XB, v2f64:$XA))]>;
542 def XVCPSGNSP : XX3Form<60, 208,
543 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
544 "xvcpsgnsp $XT, $XA, $XB", IIC_VecFP,
545 [(set v4f32:$XT, (fcopysign v4f32:$XB, v4f32:$XA))]>;
547 def XVNABSDP : XX2Form<60, 489,
548 (outs vsrc:$XT), (ins vsrc:$XB),
549 "xvnabsdp $XT, $XB", IIC_VecFP,
550 [(set v2f64:$XT, (fneg (fabs v2f64:$XB)))]>;
551 def XVNABSSP : XX2Form<60, 425,
552 (outs vsrc:$XT), (ins vsrc:$XB),
553 "xvnabssp $XT, $XB", IIC_VecFP,
554 [(set v4f32:$XT, (fneg (fabs v4f32:$XB)))]>;
556 def XVNEGDP : XX2Form<60, 505,
557 (outs vsrc:$XT), (ins vsrc:$XB),
558 "xvnegdp $XT, $XB", IIC_VecFP,
559 [(set v2f64:$XT, (fneg v2f64:$XB))]>;
560 def XVNEGSP : XX2Form<60, 441,
561 (outs vsrc:$XT), (ins vsrc:$XB),
562 "xvnegsp $XT, $XB", IIC_VecFP,
563 [(set v4f32:$XT, (fneg v4f32:$XB))]>;
565 // Conversion Instructions
566 def XSCVDPSP : XX2Form<60, 265,
567 (outs vsfrc:$XT), (ins vsfrc:$XB),
568 "xscvdpsp $XT, $XB", IIC_VecFP, []>;
569 def XSCVDPSXDS : XX2Form<60, 344,
570 (outs vsfrc:$XT), (ins vsfrc:$XB),
571 "xscvdpsxds $XT, $XB", IIC_VecFP,
572 [(set f64:$XT, (PPCfctidz f64:$XB))]>;
573 let isCodeGenOnly = 1 in
574 def XSCVDPSXDSs : XX2Form<60, 344,
575 (outs vssrc:$XT), (ins vssrc:$XB),
576 "xscvdpsxds $XT, $XB", IIC_VecFP,
577 [(set f32:$XT, (PPCfctidz f32:$XB))]>;
578 def XSCVDPSXWS : XX2Form<60, 88,
579 (outs vsfrc:$XT), (ins vsfrc:$XB),
580 "xscvdpsxws $XT, $XB", IIC_VecFP,
581 [(set f64:$XT, (PPCfctiwz f64:$XB))]>;
582 let isCodeGenOnly = 1 in
583 def XSCVDPSXWSs : XX2Form<60, 88,
584 (outs vssrc:$XT), (ins vssrc:$XB),
585 "xscvdpsxws $XT, $XB", IIC_VecFP,
586 [(set f32:$XT, (PPCfctiwz f32:$XB))]>;
587 def XSCVDPUXDS : XX2Form<60, 328,
588 (outs vsfrc:$XT), (ins vsfrc:$XB),
589 "xscvdpuxds $XT, $XB", IIC_VecFP,
590 [(set f64:$XT, (PPCfctiduz f64:$XB))]>;
591 let isCodeGenOnly = 1 in
592 def XSCVDPUXDSs : XX2Form<60, 328,
593 (outs vssrc:$XT), (ins vssrc:$XB),
594 "xscvdpuxds $XT, $XB", IIC_VecFP,
595 [(set f32:$XT, (PPCfctiduz f32:$XB))]>;
596 def XSCVDPUXWS : XX2Form<60, 72,
597 (outs vsfrc:$XT), (ins vsfrc:$XB),
598 "xscvdpuxws $XT, $XB", IIC_VecFP,
599 [(set f64:$XT, (PPCfctiwuz f64:$XB))]>;
600 let isCodeGenOnly = 1 in
601 def XSCVDPUXWSs : XX2Form<60, 72,
602 (outs vssrc:$XT), (ins vssrc:$XB),
603 "xscvdpuxws $XT, $XB", IIC_VecFP,
604 [(set f32:$XT, (PPCfctiwuz f32:$XB))]>;
605 def XSCVSPDP : XX2Form<60, 329,
606 (outs vsfrc:$XT), (ins vsfrc:$XB),
607 "xscvspdp $XT, $XB", IIC_VecFP, []>;
608 def XSCVSXDDP : XX2Form<60, 376,
609 (outs vsfrc:$XT), (ins vsfrc:$XB),
610 "xscvsxddp $XT, $XB", IIC_VecFP,
611 [(set f64:$XT, (PPCfcfid f64:$XB))]>;
612 def XSCVUXDDP : XX2Form<60, 360,
613 (outs vsfrc:$XT), (ins vsfrc:$XB),
614 "xscvuxddp $XT, $XB", IIC_VecFP,
615 [(set f64:$XT, (PPCfcfidu f64:$XB))]>;
617 def XVCVDPSP : XX2Form<60, 393,
618 (outs vsrc:$XT), (ins vsrc:$XB),
619 "xvcvdpsp $XT, $XB", IIC_VecFP,
620 [(set v4f32:$XT, (int_ppc_vsx_xvcvdpsp v2f64:$XB))]>;
621 def XVCVDPSXDS : XX2Form<60, 472,
622 (outs vsrc:$XT), (ins vsrc:$XB),
623 "xvcvdpsxds $XT, $XB", IIC_VecFP,
624 [(set v2i64:$XT, (fp_to_sint v2f64:$XB))]>;
625 def XVCVDPSXWS : XX2Form<60, 216,
626 (outs vsrc:$XT), (ins vsrc:$XB),
627 "xvcvdpsxws $XT, $XB", IIC_VecFP,
628 [(set v4i32:$XT, (int_ppc_vsx_xvcvdpsxws v2f64:$XB))]>;
629 def XVCVDPUXDS : XX2Form<60, 456,
630 (outs vsrc:$XT), (ins vsrc:$XB),
631 "xvcvdpuxds $XT, $XB", IIC_VecFP,
632 [(set v2i64:$XT, (fp_to_uint v2f64:$XB))]>;
633 def XVCVDPUXWS : XX2Form<60, 200,
634 (outs vsrc:$XT), (ins vsrc:$XB),
635 "xvcvdpuxws $XT, $XB", IIC_VecFP,
636 [(set v4i32:$XT, (int_ppc_vsx_xvcvdpuxws v2f64:$XB))]>;
638 def XVCVSPDP : XX2Form<60, 457,
639 (outs vsrc:$XT), (ins vsrc:$XB),
640 "xvcvspdp $XT, $XB", IIC_VecFP,
641 [(set v2f64:$XT, (int_ppc_vsx_xvcvspdp v4f32:$XB))]>;
642 def XVCVSPSXDS : XX2Form<60, 408,
643 (outs vsrc:$XT), (ins vsrc:$XB),
644 "xvcvspsxds $XT, $XB", IIC_VecFP, []>;
645 def XVCVSPSXWS : XX2Form<60, 152,
646 (outs vsrc:$XT), (ins vsrc:$XB),
647 "xvcvspsxws $XT, $XB", IIC_VecFP,
648 [(set v4i32:$XT, (fp_to_sint v4f32:$XB))]>;
649 def XVCVSPUXDS : XX2Form<60, 392,
650 (outs vsrc:$XT), (ins vsrc:$XB),
651 "xvcvspuxds $XT, $XB", IIC_VecFP, []>;
652 def XVCVSPUXWS : XX2Form<60, 136,
653 (outs vsrc:$XT), (ins vsrc:$XB),
654 "xvcvspuxws $XT, $XB", IIC_VecFP,
655 [(set v4i32:$XT, (fp_to_uint v4f32:$XB))]>;
656 def XVCVSXDDP : XX2Form<60, 504,
657 (outs vsrc:$XT), (ins vsrc:$XB),
658 "xvcvsxddp $XT, $XB", IIC_VecFP,
659 [(set v2f64:$XT, (sint_to_fp v2i64:$XB))]>;
660 def XVCVSXDSP : XX2Form<60, 440,
661 (outs vsrc:$XT), (ins vsrc:$XB),
662 "xvcvsxdsp $XT, $XB", IIC_VecFP,
663 [(set v4f32:$XT, (int_ppc_vsx_xvcvsxdsp v2i64:$XB))]>;
664 def XVCVSXWDP : XX2Form<60, 248,
665 (outs vsrc:$XT), (ins vsrc:$XB),
666 "xvcvsxwdp $XT, $XB", IIC_VecFP,
667 [(set v2f64:$XT, (int_ppc_vsx_xvcvsxwdp v4i32:$XB))]>;
668 def XVCVSXWSP : XX2Form<60, 184,
669 (outs vsrc:$XT), (ins vsrc:$XB),
670 "xvcvsxwsp $XT, $XB", IIC_VecFP,
671 [(set v4f32:$XT, (sint_to_fp v4i32:$XB))]>;
672 def XVCVUXDDP : XX2Form<60, 488,
673 (outs vsrc:$XT), (ins vsrc:$XB),
674 "xvcvuxddp $XT, $XB", IIC_VecFP,
675 [(set v2f64:$XT, (uint_to_fp v2i64:$XB))]>;
676 def XVCVUXDSP : XX2Form<60, 424,
677 (outs vsrc:$XT), (ins vsrc:$XB),
678 "xvcvuxdsp $XT, $XB", IIC_VecFP,
679 [(set v4f32:$XT, (int_ppc_vsx_xvcvuxdsp v2i64:$XB))]>;
680 def XVCVUXWDP : XX2Form<60, 232,
681 (outs vsrc:$XT), (ins vsrc:$XB),
682 "xvcvuxwdp $XT, $XB", IIC_VecFP,
683 [(set v2f64:$XT, (int_ppc_vsx_xvcvuxwdp v4i32:$XB))]>;
684 def XVCVUXWSP : XX2Form<60, 168,
685 (outs vsrc:$XT), (ins vsrc:$XB),
686 "xvcvuxwsp $XT, $XB", IIC_VecFP,
687 [(set v4f32:$XT, (uint_to_fp v4i32:$XB))]>;
689 // Rounding Instructions
690 def XSRDPI : XX2Form<60, 73,
691 (outs vsfrc:$XT), (ins vsfrc:$XB),
692 "xsrdpi $XT, $XB", IIC_VecFP,
693 [(set f64:$XT, (fround f64:$XB))]>;
694 def XSRDPIC : XX2Form<60, 107,
695 (outs vsfrc:$XT), (ins vsfrc:$XB),
696 "xsrdpic $XT, $XB", IIC_VecFP,
697 [(set f64:$XT, (fnearbyint f64:$XB))]>;
698 def XSRDPIM : XX2Form<60, 121,
699 (outs vsfrc:$XT), (ins vsfrc:$XB),
700 "xsrdpim $XT, $XB", IIC_VecFP,
701 [(set f64:$XT, (ffloor f64:$XB))]>;
702 def XSRDPIP : XX2Form<60, 105,
703 (outs vsfrc:$XT), (ins vsfrc:$XB),
704 "xsrdpip $XT, $XB", IIC_VecFP,
705 [(set f64:$XT, (fceil f64:$XB))]>;
706 def XSRDPIZ : XX2Form<60, 89,
707 (outs vsfrc:$XT), (ins vsfrc:$XB),
708 "xsrdpiz $XT, $XB", IIC_VecFP,
709 [(set f64:$XT, (ftrunc f64:$XB))]>;
711 def XVRDPI : XX2Form<60, 201,
712 (outs vsrc:$XT), (ins vsrc:$XB),
713 "xvrdpi $XT, $XB", IIC_VecFP,
714 [(set v2f64:$XT, (fround v2f64:$XB))]>;
715 def XVRDPIC : XX2Form<60, 235,
716 (outs vsrc:$XT), (ins vsrc:$XB),
717 "xvrdpic $XT, $XB", IIC_VecFP,
718 [(set v2f64:$XT, (fnearbyint v2f64:$XB))]>;
719 def XVRDPIM : XX2Form<60, 249,
720 (outs vsrc:$XT), (ins vsrc:$XB),
721 "xvrdpim $XT, $XB", IIC_VecFP,
722 [(set v2f64:$XT, (ffloor v2f64:$XB))]>;
723 def XVRDPIP : XX2Form<60, 233,
724 (outs vsrc:$XT), (ins vsrc:$XB),
725 "xvrdpip $XT, $XB", IIC_VecFP,
726 [(set v2f64:$XT, (fceil v2f64:$XB))]>;
727 def XVRDPIZ : XX2Form<60, 217,
728 (outs vsrc:$XT), (ins vsrc:$XB),
729 "xvrdpiz $XT, $XB", IIC_VecFP,
730 [(set v2f64:$XT, (ftrunc v2f64:$XB))]>;
732 def XVRSPI : XX2Form<60, 137,
733 (outs vsrc:$XT), (ins vsrc:$XB),
734 "xvrspi $XT, $XB", IIC_VecFP,
735 [(set v4f32:$XT, (fround v4f32:$XB))]>;
736 def XVRSPIC : XX2Form<60, 171,
737 (outs vsrc:$XT), (ins vsrc:$XB),
738 "xvrspic $XT, $XB", IIC_VecFP,
739 [(set v4f32:$XT, (fnearbyint v4f32:$XB))]>;
740 def XVRSPIM : XX2Form<60, 185,
741 (outs vsrc:$XT), (ins vsrc:$XB),
742 "xvrspim $XT, $XB", IIC_VecFP,
743 [(set v4f32:$XT, (ffloor v4f32:$XB))]>;
744 def XVRSPIP : XX2Form<60, 169,
745 (outs vsrc:$XT), (ins vsrc:$XB),
746 "xvrspip $XT, $XB", IIC_VecFP,
747 [(set v4f32:$XT, (fceil v4f32:$XB))]>;
748 def XVRSPIZ : XX2Form<60, 153,
749 (outs vsrc:$XT), (ins vsrc:$XB),
750 "xvrspiz $XT, $XB", IIC_VecFP,
751 [(set v4f32:$XT, (ftrunc v4f32:$XB))]>;
753 // Max/Min Instructions
754 let isCommutable = 1 in {
755 def XSMAXDP : XX3Form<60, 160,
756 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
757 "xsmaxdp $XT, $XA, $XB", IIC_VecFP,
759 (int_ppc_vsx_xsmaxdp vsfrc:$XA, vsfrc:$XB))]>;
760 def XSMINDP : XX3Form<60, 168,
761 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
762 "xsmindp $XT, $XA, $XB", IIC_VecFP,
764 (int_ppc_vsx_xsmindp vsfrc:$XA, vsfrc:$XB))]>;
766 def XVMAXDP : XX3Form<60, 224,
767 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
768 "xvmaxdp $XT, $XA, $XB", IIC_VecFP,
770 (int_ppc_vsx_xvmaxdp vsrc:$XA, vsrc:$XB))]>;
771 def XVMINDP : XX3Form<60, 232,
772 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
773 "xvmindp $XT, $XA, $XB", IIC_VecFP,
775 (int_ppc_vsx_xvmindp vsrc:$XA, vsrc:$XB))]>;
777 def XVMAXSP : XX3Form<60, 192,
778 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
779 "xvmaxsp $XT, $XA, $XB", IIC_VecFP,
781 (int_ppc_vsx_xvmaxsp vsrc:$XA, vsrc:$XB))]>;
782 def XVMINSP : XX3Form<60, 200,
783 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
784 "xvminsp $XT, $XA, $XB", IIC_VecFP,
786 (int_ppc_vsx_xvminsp vsrc:$XA, vsrc:$XB))]>;
790 // Logical Instructions
791 let isCommutable = 1 in
792 def XXLAND : XX3Form<60, 130,
793 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
794 "xxland $XT, $XA, $XB", IIC_VecGeneral,
795 [(set v4i32:$XT, (and v4i32:$XA, v4i32:$XB))]>;
796 def XXLANDC : XX3Form<60, 138,
797 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
798 "xxlandc $XT, $XA, $XB", IIC_VecGeneral,
799 [(set v4i32:$XT, (and v4i32:$XA,
800 (vnot_ppc v4i32:$XB)))]>;
801 let isCommutable = 1 in {
802 def XXLNOR : XX3Form<60, 162,
803 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
804 "xxlnor $XT, $XA, $XB", IIC_VecGeneral,
805 [(set v4i32:$XT, (vnot_ppc (or v4i32:$XA,
807 def XXLOR : XX3Form<60, 146,
808 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
809 "xxlor $XT, $XA, $XB", IIC_VecGeneral,
810 [(set v4i32:$XT, (or v4i32:$XA, v4i32:$XB))]>;
811 let isCodeGenOnly = 1 in
812 def XXLORf: XX3Form<60, 146,
813 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
814 "xxlor $XT, $XA, $XB", IIC_VecGeneral, []>;
815 def XXLXOR : XX3Form<60, 154,
816 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
817 "xxlxor $XT, $XA, $XB", IIC_VecGeneral,
818 [(set v4i32:$XT, (xor v4i32:$XA, v4i32:$XB))]>;
820 let isCodeGenOnly = 1 in
821 def XXLXORz : XX3Form_Zero<60, 154, (outs vsrc:$XT), (ins),
822 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
823 [(set v4i32:$XT, (v4i32 immAllZerosV))]>;
825 let isCodeGenOnly = 1 in {
826 def XXLXORdpz : XX3Form_SetZero<60, 154,
827 (outs vsfrc:$XT), (ins),
828 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
829 [(set f64:$XT, (fpimm0))]>;
830 def XXLXORspz : XX3Form_SetZero<60, 154,
831 (outs vssrc:$XT), (ins),
832 "xxlxor $XT, $XT, $XT", IIC_VecGeneral,
833 [(set f32:$XT, (fpimm0))]>;
836 // Permutation Instructions
837 def XXMRGHW : XX3Form<60, 18,
838 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
839 "xxmrghw $XT, $XA, $XB", IIC_VecPerm, []>;
840 def XXMRGLW : XX3Form<60, 50,
841 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
842 "xxmrglw $XT, $XA, $XB", IIC_VecPerm, []>;
844 def XXPERMDI : XX3Form_2<60, 10,
845 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, u2imm:$DM),
846 "xxpermdi $XT, $XA, $XB, $DM", IIC_VecPerm, []>;
847 let isCodeGenOnly = 1 in
848 def XXPERMDIs : XX3Form_2s<60, 10, (outs vsrc:$XT), (ins vsfrc:$XA, u2imm:$DM),
849 "xxpermdi $XT, $XA, $XA, $DM", IIC_VecPerm, []>;
850 def XXSEL : XX4Form<60, 3,
851 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, vsrc:$XC),
852 "xxsel $XT, $XA, $XB, $XC", IIC_VecPerm, []>;
854 def XXSLDWI : XX3Form_2<60, 2,
855 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, u2imm:$SHW),
856 "xxsldwi $XT, $XA, $XB, $SHW", IIC_VecPerm,
857 [(set v4i32:$XT, (PPCvecshl v4i32:$XA, v4i32:$XB,
858 imm32SExt16:$SHW))]>;
859 def XXSPLTW : XX2Form_2<60, 164,
860 (outs vsrc:$XT), (ins vsrc:$XB, u2imm:$UIM),
861 "xxspltw $XT, $XB, $UIM", IIC_VecPerm,
863 (PPCxxsplt v4i32:$XB, imm32SExt16:$UIM))]>;
864 let isCodeGenOnly = 1 in
865 def XXSPLTWs : XX2Form_2<60, 164,
866 (outs vsrc:$XT), (ins vfrc:$XB, u2imm:$UIM),
867 "xxspltw $XT, $XB, $UIM", IIC_VecPerm, []>;
871 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation. Expanded after
872 // instruction selection into a branch sequence.
873 let usesCustomInserter = 1, // Expanded after instruction selection.
874 PPC970_Single = 1 in {
876 def SELECT_CC_VSRC: Pseudo<(outs vsrc:$dst),
877 (ins crrc:$cond, vsrc:$T, vsrc:$F, i32imm:$BROPC),
880 def SELECT_VSRC: Pseudo<(outs vsrc:$dst),
881 (ins crbitrc:$cond, vsrc:$T, vsrc:$F),
884 (select i1:$cond, v2f64:$T, v2f64:$F))]>;
885 def SELECT_CC_VSFRC: Pseudo<(outs f8rc:$dst),
886 (ins crrc:$cond, f8rc:$T, f8rc:$F,
887 i32imm:$BROPC), "#SELECT_CC_VSFRC",
889 def SELECT_VSFRC: Pseudo<(outs f8rc:$dst),
890 (ins crbitrc:$cond, f8rc:$T, f8rc:$F),
893 (select i1:$cond, f64:$T, f64:$F))]>;
894 def SELECT_CC_VSSRC: Pseudo<(outs f4rc:$dst),
895 (ins crrc:$cond, f4rc:$T, f4rc:$F,
896 i32imm:$BROPC), "#SELECT_CC_VSSRC",
898 def SELECT_VSSRC: Pseudo<(outs f4rc:$dst),
899 (ins crbitrc:$cond, f4rc:$T, f4rc:$F),
902 (select i1:$cond, f32:$T, f32:$F))]>;
903 } // usesCustomInserter
906 def : InstAlias<"xvmovdp $XT, $XB",
907 (XVCPSGNDP vsrc:$XT, vsrc:$XB, vsrc:$XB)>;
908 def : InstAlias<"xvmovsp $XT, $XB",
909 (XVCPSGNSP vsrc:$XT, vsrc:$XB, vsrc:$XB)>;
911 def : InstAlias<"xxspltd $XT, $XB, 0",
912 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 0)>;
913 def : InstAlias<"xxspltd $XT, $XB, 1",
914 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 3)>;
915 def : InstAlias<"xxmrghd $XT, $XA, $XB",
916 (XXPERMDI vsrc:$XT, vsrc:$XA, vsrc:$XB, 0)>;
917 def : InstAlias<"xxmrgld $XT, $XA, $XB",
918 (XXPERMDI vsrc:$XT, vsrc:$XA, vsrc:$XB, 3)>;
919 def : InstAlias<"xxswapd $XT, $XB",
920 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 2)>;
921 def : InstAlias<"xxspltd $XT, $XB, 0",
922 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 0)>;
923 def : InstAlias<"xxspltd $XT, $XB, 1",
924 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 3)>;
925 def : InstAlias<"xxswapd $XT, $XB",
926 (XXPERMDIs vsrc:$XT, vsfrc:$XB, 2)>;
928 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
930 def : Pat<(v4i32 (vnot_ppc v4i32:$A)),
931 (v4i32 (XXLNOR $A, $A))>;
932 let Predicates = [IsBigEndian] in {
933 def : Pat<(v2f64 (scalar_to_vector f64:$A)),
934 (v2f64 (SUBREG_TO_REG (i64 1), $A, sub_64))>;
936 def : Pat<(f64 (extractelt v2f64:$S, 0)),
937 (f64 (EXTRACT_SUBREG $S, sub_64))>;
938 def : Pat<(f64 (extractelt v2f64:$S, 1)),
939 (f64 (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64))>;
942 let Predicates = [IsLittleEndian] in {
943 def : Pat<(v2f64 (scalar_to_vector f64:$A)),
944 (v2f64 (XXPERMDI (SUBREG_TO_REG (i64 1), $A, sub_64),
945 (SUBREG_TO_REG (i64 1), $A, sub_64), 0))>;
947 def : Pat<(f64 (extractelt v2f64:$S, 0)),
948 (f64 (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64))>;
949 def : Pat<(f64 (extractelt v2f64:$S, 1)),
950 (f64 (EXTRACT_SUBREG $S, sub_64))>;
953 // Additional fnmsub patterns: -a*c + b == -(a*c - b)
954 def : Pat<(fma (fneg f64:$A), f64:$C, f64:$B),
955 (XSNMSUBADP $B, $C, $A)>;
956 def : Pat<(fma f64:$A, (fneg f64:$C), f64:$B),
957 (XSNMSUBADP $B, $C, $A)>;
959 def : Pat<(fma (fneg v2f64:$A), v2f64:$C, v2f64:$B),
960 (XVNMSUBADP $B, $C, $A)>;
961 def : Pat<(fma v2f64:$A, (fneg v2f64:$C), v2f64:$B),
962 (XVNMSUBADP $B, $C, $A)>;
964 def : Pat<(fma (fneg v4f32:$A), v4f32:$C, v4f32:$B),
965 (XVNMSUBASP $B, $C, $A)>;
966 def : Pat<(fma v4f32:$A, (fneg v4f32:$C), v4f32:$B),
967 (XVNMSUBASP $B, $C, $A)>;
969 def : Pat<(v2f64 (bitconvert v4f32:$A)),
970 (COPY_TO_REGCLASS $A, VSRC)>;
971 def : Pat<(v2f64 (bitconvert v4i32:$A)),
972 (COPY_TO_REGCLASS $A, VSRC)>;
973 def : Pat<(v2f64 (bitconvert v8i16:$A)),
974 (COPY_TO_REGCLASS $A, VSRC)>;
975 def : Pat<(v2f64 (bitconvert v16i8:$A)),
976 (COPY_TO_REGCLASS $A, VSRC)>;
978 def : Pat<(v4f32 (bitconvert v2f64:$A)),
979 (COPY_TO_REGCLASS $A, VRRC)>;
980 def : Pat<(v4i32 (bitconvert v2f64:$A)),
981 (COPY_TO_REGCLASS $A, VRRC)>;
982 def : Pat<(v8i16 (bitconvert v2f64:$A)),
983 (COPY_TO_REGCLASS $A, VRRC)>;
984 def : Pat<(v16i8 (bitconvert v2f64:$A)),
985 (COPY_TO_REGCLASS $A, VRRC)>;
987 def : Pat<(v2i64 (bitconvert v4f32:$A)),
988 (COPY_TO_REGCLASS $A, VSRC)>;
989 def : Pat<(v2i64 (bitconvert v4i32:$A)),
990 (COPY_TO_REGCLASS $A, VSRC)>;
991 def : Pat<(v2i64 (bitconvert v8i16:$A)),
992 (COPY_TO_REGCLASS $A, VSRC)>;
993 def : Pat<(v2i64 (bitconvert v16i8:$A)),
994 (COPY_TO_REGCLASS $A, VSRC)>;
996 def : Pat<(v4f32 (bitconvert v2i64:$A)),
997 (COPY_TO_REGCLASS $A, VRRC)>;
998 def : Pat<(v4i32 (bitconvert v2i64:$A)),
999 (COPY_TO_REGCLASS $A, VRRC)>;
1000 def : Pat<(v8i16 (bitconvert v2i64:$A)),
1001 (COPY_TO_REGCLASS $A, VRRC)>;
1002 def : Pat<(v16i8 (bitconvert v2i64:$A)),
1003 (COPY_TO_REGCLASS $A, VRRC)>;
1005 def : Pat<(v2f64 (bitconvert v2i64:$A)),
1006 (COPY_TO_REGCLASS $A, VRRC)>;
1007 def : Pat<(v2i64 (bitconvert v2f64:$A)),
1008 (COPY_TO_REGCLASS $A, VRRC)>;
1010 def : Pat<(v2f64 (bitconvert v1i128:$A)),
1011 (COPY_TO_REGCLASS $A, VRRC)>;
1012 def : Pat<(v1i128 (bitconvert v2f64:$A)),
1013 (COPY_TO_REGCLASS $A, VRRC)>;
1015 // sign extension patterns
1016 // To extend "in place" from v2i32 to v2i64, we have input data like:
1017 // | undef | i32 | undef | i32 |
1018 // but xvcvsxwdp expects the input in big-Endian format:
1019 // | i32 | undef | i32 | undef |
1020 // so we need to shift everything to the left by one i32 (word) before
1022 def : Pat<(sext_inreg v2i64:$C, v2i32),
1023 (XVCVDPSXDS (XVCVSXWDP (XXSLDWI $C, $C, 1)))>;
1024 def : Pat<(v2f64 (sint_to_fp (sext_inreg v2i64:$C, v2i32))),
1025 (XVCVSXWDP (XXSLDWI $C, $C, 1))>;
1027 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 0)),
1028 (v2f64 (XVCVSXWDP (v2i64 (XXMRGHW $C, $C))))>;
1029 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 1)),
1030 (v2f64 (XVCVSXWDP (v2i64 (XXMRGLW $C, $C))))>;
1032 def : Pat<(v2f64 (PPCuvec2fp v4i32:$C, 0)),
1033 (v2f64 (XVCVUXWDP (v2i64 (XXMRGHW $C, $C))))>;
1034 def : Pat<(v2f64 (PPCuvec2fp v4i32:$C, 1)),
1035 (v2f64 (XVCVUXWDP (v2i64 (XXMRGLW $C, $C))))>;
1038 let Predicates = [HasVSX, HasOnlySwappingMemOps] in {
1039 def : Pat<(v2f64 (PPClxvd2x xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1042 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, xoaddr:$dst),
1043 (STXVD2X $rS, xoaddr:$dst)>;
1044 def : Pat<(int_ppc_vsx_stxvd2x_be v2f64:$rS, xoaddr:$dst),
1045 (STXVD2X $rS, xoaddr:$dst)>;
1046 def : Pat<(int_ppc_vsx_stxvw4x_be v4i32:$rS, xoaddr:$dst),
1047 (STXVW4X $rS, xoaddr:$dst)>;
1048 def : Pat<(PPCstxvd2x v2f64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1050 let Predicates = [IsBigEndian, HasVSX, HasOnlySwappingMemOps] in {
1051 def : Pat<(v2f64 (load xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1052 def : Pat<(v2i64 (load xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1053 def : Pat<(v4i32 (load xoaddr:$src)), (LXVW4X xoaddr:$src)>;
1054 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x xoaddr:$src)), (LXVW4X xoaddr:$src)>;
1055 def : Pat<(store v2f64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1056 def : Pat<(store v2i64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
1057 def : Pat<(store v4i32:$XT, xoaddr:$dst), (STXVW4X $XT, xoaddr:$dst)>;
1058 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, xoaddr:$dst),
1059 (STXVW4X $rS, xoaddr:$dst)>;
1063 def : Pat<(v2f64 (PPCxxswapd v2f64:$src)), (XXPERMDI $src, $src, 2)>;
1064 def : Pat<(v2i64 (PPCxxswapd v2i64:$src)), (XXPERMDI $src, $src, 2)>;
1065 def : Pat<(v4f32 (PPCxxswapd v4f32:$src)), (XXPERMDI $src, $src, 2)>;
1066 def : Pat<(v4i32 (PPCxxswapd v4i32:$src)), (XXPERMDI $src, $src, 2)>;
1067 def : Pat<(v2f64 (PPCswapNoChain v2f64:$src)), (XXPERMDI $src, $src, 2)>;
1069 // PPCvecshl XT, XA, XA, 2 can be selected to both XXSLDWI XT,XA,XA,2 and
1070 // XXSWAPD XT,XA (i.e. XXPERMDI XT,XA,XA,2), the later one is more profitable.
1071 def : Pat<(v4i32 (PPCvecshl v4i32:$src, v4i32:$src, 2)), (XXPERMDI $src, $src, 2)>;
1074 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETLT)),
1075 (SELECT_VSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1076 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETULT)),
1077 (SELECT_VSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1078 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETLE)),
1079 (SELECT_VSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1080 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETULE)),
1081 (SELECT_VSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1082 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETEQ)),
1083 (SELECT_VSRC (CREQV $lhs, $rhs), $tval, $fval)>;
1084 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETGE)),
1085 (SELECT_VSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1086 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETUGE)),
1087 (SELECT_VSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1088 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETGT)),
1089 (SELECT_VSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1090 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETUGT)),
1091 (SELECT_VSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1092 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETNE)),
1093 (SELECT_VSRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1095 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETLT)),
1096 (SELECT_VSFRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1097 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETULT)),
1098 (SELECT_VSFRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1099 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETLE)),
1100 (SELECT_VSFRC (CRORC $lhs, $rhs), $tval, $fval)>;
1101 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETULE)),
1102 (SELECT_VSFRC (CRORC $rhs, $lhs), $tval, $fval)>;
1103 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETEQ)),
1104 (SELECT_VSFRC (CREQV $lhs, $rhs), $tval, $fval)>;
1105 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETGE)),
1106 (SELECT_VSFRC (CRORC $rhs, $lhs), $tval, $fval)>;
1107 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGE)),
1108 (SELECT_VSFRC (CRORC $lhs, $rhs), $tval, $fval)>;
1109 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETGT)),
1110 (SELECT_VSFRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1111 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGT)),
1112 (SELECT_VSFRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1113 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETNE)),
1114 (SELECT_VSFRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1117 def : Pat<(int_ppc_vsx_xvdivsp v4f32:$A, v4f32:$B),
1119 def : Pat<(int_ppc_vsx_xvdivdp v2f64:$A, v2f64:$B),
1122 // Reciprocal estimate
1123 def : Pat<(int_ppc_vsx_xvresp v4f32:$A),
1125 def : Pat<(int_ppc_vsx_xvredp v2f64:$A),
1128 // Recip. square root estimate
1129 def : Pat<(int_ppc_vsx_xvrsqrtesp v4f32:$A),
1131 def : Pat<(int_ppc_vsx_xvrsqrtedp v2f64:$A),
1134 let Predicates = [IsLittleEndian] in {
1135 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1136 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1137 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1138 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1139 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1140 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1141 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1142 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1145 let Predicates = [IsBigEndian] in {
1146 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1147 (f64 (XSCVSXDDP (COPY_TO_REGCLASS $S, VSFRC)))>;
1148 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1149 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1150 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1151 (f64 (XSCVUXDDP (COPY_TO_REGCLASS $S, VSFRC)))>;
1152 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1153 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1156 } // AddedComplexity
1160 dag Li8 = (i32 (extloadi8 xoaddr:$src));
1161 dag ZELi8 = (i32 (zextloadi8 xoaddr:$src));
1162 dag ZELi8i64 = (i64 (zextloadi8 xoaddr:$src));
1163 dag SELi8 = (i32 (sext_inreg (extloadi8 xoaddr:$src), i8));
1164 dag SELi8i64 = (i64 (sext_inreg (extloadi8 xoaddr:$src), i8));
1166 dag Li16 = (i32 (extloadi16 xoaddr:$src));
1167 dag ZELi16 = (i32 (zextloadi16 xoaddr:$src));
1168 dag ZELi16i64 = (i64 (zextloadi16 xoaddr:$src));
1169 dag SELi16 = (i32 (sextloadi16 xoaddr:$src));
1170 dag SELi16i64 = (i64 (sextloadi16 xoaddr:$src));
1172 dag Li32 = (i32 (load xoaddr:$src));
1175 // The following VSX instructions were introduced in Power ISA 2.07
1176 /* FIXME: if the operands are v2i64, these patterns will not match.
1177 we should define new patterns or otherwise match the same patterns
1178 when the elements are larger than i32.
1180 def HasP8Vector : Predicate<"PPCSubTarget->hasP8Vector()">;
1181 def HasDirectMove : Predicate<"PPCSubTarget->hasDirectMove()">;
1182 let Predicates = [HasP8Vector] in {
1183 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
1184 let isCommutable = 1, UseVSXReg = 1 in {
1185 def XXLEQV : XX3Form<60, 186,
1186 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1187 "xxleqv $XT, $XA, $XB", IIC_VecGeneral,
1188 [(set v4i32:$XT, (vnot_ppc (xor v4i32:$XA, v4i32:$XB)))]>;
1189 def XXLNAND : XX3Form<60, 178,
1190 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1191 "xxlnand $XT, $XA, $XB", IIC_VecGeneral,
1192 [(set v4i32:$XT, (vnot_ppc (and v4i32:$XA,
1194 } // isCommutable, UseVSXReg
1196 def : Pat<(int_ppc_vsx_xxleqv v4i32:$A, v4i32:$B),
1199 let UseVSXReg = 1 in {
1200 def XXLORC : XX3Form<60, 170,
1201 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1202 "xxlorc $XT, $XA, $XB", IIC_VecGeneral,
1203 [(set v4i32:$XT, (or v4i32:$XA, (vnot_ppc v4i32:$XB)))]>;
1205 // VSX scalar loads introduced in ISA 2.07
1206 let mayLoad = 1, mayStore = 0 in {
1208 def LXSSPX : XX1Form<31, 524, (outs vssrc:$XT), (ins memrr:$src),
1209 "lxsspx $XT, $src", IIC_LdStLFD,
1210 [(set f32:$XT, (load xoaddr:$src))]>;
1211 def LXSIWAX : XX1Form<31, 76, (outs vsfrc:$XT), (ins memrr:$src),
1212 "lxsiwax $XT, $src", IIC_LdStLFD,
1213 [(set f64:$XT, (PPClfiwax xoaddr:$src))]>;
1214 def LXSIWZX : XX1Form<31, 12, (outs vsfrc:$XT), (ins memrr:$src),
1215 "lxsiwzx $XT, $src", IIC_LdStLFD,
1216 [(set f64:$XT, (PPClfiwzx xoaddr:$src))]>;
1219 // VSX scalar stores introduced in ISA 2.07
1220 let mayStore = 1, mayLoad = 0 in {
1222 def STXSSPX : XX1Form<31, 652, (outs), (ins vssrc:$XT, memrr:$dst),
1223 "stxsspx $XT, $dst", IIC_LdStSTFD,
1224 [(store f32:$XT, xoaddr:$dst)]>;
1225 def STXSIWX : XX1Form<31, 140, (outs), (ins vsfrc:$XT, memrr:$dst),
1226 "stxsiwx $XT, $dst", IIC_LdStSTFD,
1227 [(PPCstfiwx f64:$XT, xoaddr:$dst)]>;
1231 def : Pat<(f64 (extloadf32 xoaddr:$src)),
1232 (COPY_TO_REGCLASS (LXSSPX xoaddr:$src), VSFRC)>;
1233 def : Pat<(f32 (fpround (extloadf32 xoaddr:$src))),
1234 (f32 (LXSSPX xoaddr:$src))>;
1235 def : Pat<(f64 (fpextend f32:$src)),
1236 (COPY_TO_REGCLASS $src, VSFRC)>;
1238 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLT)),
1239 (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1240 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULT)),
1241 (SELECT_VSSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1242 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLE)),
1243 (SELECT_VSSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1244 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULE)),
1245 (SELECT_VSSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1246 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETEQ)),
1247 (SELECT_VSSRC (CREQV $lhs, $rhs), $tval, $fval)>;
1248 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETGE)),
1249 (SELECT_VSSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1250 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETUGE)),
1251 (SELECT_VSSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1252 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETGT)),
1253 (SELECT_VSSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1254 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETUGT)),
1255 (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1256 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETNE)),
1257 (SELECT_VSSRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1259 let UseVSXReg = 1 in {
1260 // VSX Elementary Scalar FP arithmetic (SP)
1261 let isCommutable = 1 in {
1262 def XSADDSP : XX3Form<60, 0,
1263 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1264 "xsaddsp $XT, $XA, $XB", IIC_VecFP,
1265 [(set f32:$XT, (fadd f32:$XA, f32:$XB))]>;
1266 def XSMULSP : XX3Form<60, 16,
1267 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1268 "xsmulsp $XT, $XA, $XB", IIC_VecFP,
1269 [(set f32:$XT, (fmul f32:$XA, f32:$XB))]>;
1272 def XSDIVSP : XX3Form<60, 24,
1273 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1274 "xsdivsp $XT, $XA, $XB", IIC_FPDivS,
1275 [(set f32:$XT, (fdiv f32:$XA, f32:$XB))]>;
1276 def XSRESP : XX2Form<60, 26,
1277 (outs vssrc:$XT), (ins vssrc:$XB),
1278 "xsresp $XT, $XB", IIC_VecFP,
1279 [(set f32:$XT, (PPCfre f32:$XB))]>;
1280 def XSSQRTSP : XX2Form<60, 11,
1281 (outs vssrc:$XT), (ins vssrc:$XB),
1282 "xssqrtsp $XT, $XB", IIC_FPSqrtS,
1283 [(set f32:$XT, (fsqrt f32:$XB))]>;
1284 def XSRSQRTESP : XX2Form<60, 10,
1285 (outs vssrc:$XT), (ins vssrc:$XB),
1286 "xsrsqrtesp $XT, $XB", IIC_VecFP,
1287 [(set f32:$XT, (PPCfrsqrte f32:$XB))]>;
1288 def XSSUBSP : XX3Form<60, 8,
1289 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1290 "xssubsp $XT, $XA, $XB", IIC_VecFP,
1291 [(set f32:$XT, (fsub f32:$XA, f32:$XB))]>;
1294 let BaseName = "XSMADDASP" in {
1295 let isCommutable = 1 in
1296 def XSMADDASP : XX3Form<60, 1,
1298 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1299 "xsmaddasp $XT, $XA, $XB", IIC_VecFP,
1300 [(set f32:$XT, (fma f32:$XA, f32:$XB, f32:$XTi))]>,
1301 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1303 let IsVSXFMAAlt = 1 in
1304 def XSMADDMSP : XX3Form<60, 9,
1306 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1307 "xsmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
1308 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1312 let BaseName = "XSMSUBASP" in {
1313 let isCommutable = 1 in
1314 def XSMSUBASP : XX3Form<60, 17,
1316 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1317 "xsmsubasp $XT, $XA, $XB", IIC_VecFP,
1318 [(set f32:$XT, (fma f32:$XA, f32:$XB,
1319 (fneg f32:$XTi)))]>,
1320 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1322 let IsVSXFMAAlt = 1 in
1323 def XSMSUBMSP : XX3Form<60, 25,
1325 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1326 "xsmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
1327 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1331 let BaseName = "XSNMADDASP" in {
1332 let isCommutable = 1 in
1333 def XSNMADDASP : XX3Form<60, 129,
1335 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1336 "xsnmaddasp $XT, $XA, $XB", IIC_VecFP,
1337 [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
1339 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1341 let IsVSXFMAAlt = 1 in
1342 def XSNMADDMSP : XX3Form<60, 137,
1344 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1345 "xsnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
1346 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1350 let BaseName = "XSNMSUBASP" in {
1351 let isCommutable = 1 in
1352 def XSNMSUBASP : XX3Form<60, 145,
1354 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1355 "xsnmsubasp $XT, $XA, $XB", IIC_VecFP,
1356 [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
1357 (fneg f32:$XTi))))]>,
1358 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1360 let IsVSXFMAAlt = 1 in
1361 def XSNMSUBMSP : XX3Form<60, 153,
1363 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1364 "xsnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
1365 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1369 // Single Precision Conversions (FP <-> INT)
1370 def XSCVSXDSP : XX2Form<60, 312,
1371 (outs vssrc:$XT), (ins vsfrc:$XB),
1372 "xscvsxdsp $XT, $XB", IIC_VecFP,
1373 [(set f32:$XT, (PPCfcfids f64:$XB))]>;
1374 def XSCVUXDSP : XX2Form<60, 296,
1375 (outs vssrc:$XT), (ins vsfrc:$XB),
1376 "xscvuxdsp $XT, $XB", IIC_VecFP,
1377 [(set f32:$XT, (PPCfcfidus f64:$XB))]>;
1379 // Conversions between vector and scalar single precision
1380 def XSCVDPSPN : XX2Form<60, 267, (outs vsrc:$XT), (ins vssrc:$XB),
1381 "xscvdpspn $XT, $XB", IIC_VecFP, []>;
1382 def XSCVSPDPN : XX2Form<60, 331, (outs vssrc:$XT), (ins vsrc:$XB),
1383 "xscvspdpn $XT, $XB", IIC_VecFP, []>;
1386 let Predicates = [IsLittleEndian] in {
1387 def : Pat<(f32 (PPCfcfids (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1388 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1389 def : Pat<(f32 (PPCfcfids (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1390 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1391 def : Pat<(f32 (PPCfcfidus (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1392 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1393 def : Pat<(f32 (PPCfcfidus (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1394 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1397 let Predicates = [IsBigEndian] in {
1398 def : Pat<(f32 (PPCfcfids (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1399 (f32 (XSCVSXDSP (COPY_TO_REGCLASS $S, VSFRC)))>;
1400 def : Pat<(f32 (PPCfcfids (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1401 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1402 def : Pat<(f32 (PPCfcfidus (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1403 (f32 (XSCVUXDSP (COPY_TO_REGCLASS $S, VSFRC)))>;
1404 def : Pat<(f32 (PPCfcfidus (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1405 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1407 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.Li32)),
1408 (v4i32 (XXSPLTWs (LXSIWAX xoaddr:$src), 1))>;
1409 } // AddedComplexity = 400
1412 let UseVSXReg = 1, AddedComplexity = 400 in {
1413 let Predicates = [HasDirectMove] in {
1414 // VSX direct move instructions
1415 def MFVSRD : XX1_RS6_RD5_XO<31, 51, (outs g8rc:$rA), (ins vsfrc:$XT),
1416 "mfvsrd $rA, $XT", IIC_VecGeneral,
1417 [(set i64:$rA, (PPCmfvsr f64:$XT))]>,
1418 Requires<[In64BitMode]>;
1419 let isCodeGenOnly = 1 in
1420 def MFVRD : XX1_RS6_RD5_XO<31, 51, (outs g8rc:$rA), (ins vrrc:$XT),
1421 "mfvsrd $rA, $XT", IIC_VecGeneral,
1423 Requires<[In64BitMode]>;
1424 def MFVSRWZ : XX1_RS6_RD5_XO<31, 115, (outs gprc:$rA), (ins vsfrc:$XT),
1425 "mfvsrwz $rA, $XT", IIC_VecGeneral,
1426 [(set i32:$rA, (PPCmfvsr f64:$XT))]>;
1427 def MTVSRD : XX1_RS6_RD5_XO<31, 179, (outs vsfrc:$XT), (ins g8rc:$rA),
1428 "mtvsrd $XT, $rA", IIC_VecGeneral,
1429 [(set f64:$XT, (PPCmtvsra i64:$rA))]>,
1430 Requires<[In64BitMode]>;
1431 def MTVSRWA : XX1_RS6_RD5_XO<31, 211, (outs vsfrc:$XT), (ins gprc:$rA),
1432 "mtvsrwa $XT, $rA", IIC_VecGeneral,
1433 [(set f64:$XT, (PPCmtvsra i32:$rA))]>;
1434 def MTVSRWZ : XX1_RS6_RD5_XO<31, 243, (outs vsfrc:$XT), (ins gprc:$rA),
1435 "mtvsrwz $XT, $rA", IIC_VecGeneral,
1436 [(set f64:$XT, (PPCmtvsrz i32:$rA))]>;
1439 let Predicates = [IsISA3_0, HasDirectMove] in {
1440 def MTVSRWS: XX1_RS6_RD5_XO<31, 403, (outs vsrc:$XT), (ins gprc:$rA),
1441 "mtvsrws $XT, $rA", IIC_VecGeneral, []>;
1443 def MTVSRDD: XX1Form<31, 435, (outs vsrc:$XT), (ins g8rc_nox0:$rA, g8rc:$rB),
1444 "mtvsrdd $XT, $rA, $rB", IIC_VecGeneral,
1445 []>, Requires<[In64BitMode]>;
1447 def MFVSRLD: XX1_RS6_RD5_XO<31, 307, (outs g8rc:$rA), (ins vsrc:$XT),
1448 "mfvsrld $rA, $XT", IIC_VecGeneral,
1449 []>, Requires<[In64BitMode]>;
1451 } // IsISA3_0, HasDirectMove
1454 // We want to parse this from asm, but we don't want to emit this as it would
1455 // be emitted with a VSX reg. So leave Emit = 0 here.
1456 def : InstAlias<"mfvrd $rA, $XT",
1457 (MFVRD g8rc:$rA, vrrc:$XT), 0>;
1458 def : InstAlias<"mffprd $rA, $src",
1459 (MFVSRD g8rc:$rA, f8rc:$src)>;
1461 /* Direct moves of various widths from GPR's into VSR's. Each move lines
1462 the value up into element 0 (both BE and LE). Namely, entities smaller than
1463 a doubleword are shifted left and moved for BE. For LE, they're moved, then
1464 swapped to go into the least significant element of the VSR.
1470 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 56, 7));
1474 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 48, 15));
1478 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 32, 31));
1479 dag BE_DWORD_0 = (MTVSRD $A);
1481 dag LE_MTVSRW = (MTVSRD (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32));
1482 dag LE_WORD_1 = (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1483 LE_MTVSRW, sub_64));
1484 dag LE_WORD_0 = (XXPERMDI LE_WORD_1, LE_WORD_1, 2);
1485 dag LE_DWORD_1 = (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1486 BE_DWORD_0, sub_64));
1487 dag LE_DWORD_0 = (XXPERMDI LE_DWORD_1, LE_DWORD_1, 2);
1490 /* Patterns for extracting elements out of vectors. Integer elements are
1491 extracted using direct move operations. Patterns for extracting elements
1492 whose indices are not available at compile time are also provided with
1493 various _VARIABLE_ patterns.
1494 The numbering for the DAG's is for LE, but when used on BE, the correct
1495 LE element can just be used (i.e. LE_BYTE_2 == BE_BYTE_13).
1497 def VectorExtractions {
1498 // Doubleword extraction
1502 (XXPERMDI (COPY_TO_REGCLASS $S, VSRC),
1503 (COPY_TO_REGCLASS $S, VSRC), 2), sub_64));
1504 dag LE_DWORD_1 = (MFVSRD
1506 (v2i64 (COPY_TO_REGCLASS $S, VSRC)), sub_64));
1509 dag LE_WORD_0 = (MFVSRWZ (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64));
1510 dag LE_WORD_1 = (MFVSRWZ (EXTRACT_SUBREG (XXSLDWI $S, $S, 1), sub_64));
1511 dag LE_WORD_2 = (MFVSRWZ (EXTRACT_SUBREG
1512 (v2i64 (COPY_TO_REGCLASS $S, VSRC)), sub_64));
1513 dag LE_WORD_3 = (MFVSRWZ (EXTRACT_SUBREG (XXSLDWI $S, $S, 3), sub_64));
1515 // Halfword extraction
1516 dag LE_HALF_0 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 0, 48), sub_32));
1517 dag LE_HALF_1 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 48, 48), sub_32));
1518 dag LE_HALF_2 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 32, 48), sub_32));
1519 dag LE_HALF_3 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 16, 48), sub_32));
1520 dag LE_HALF_4 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 0, 48), sub_32));
1521 dag LE_HALF_5 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 48, 48), sub_32));
1522 dag LE_HALF_6 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 32, 48), sub_32));
1523 dag LE_HALF_7 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 16, 48), sub_32));
1526 dag LE_BYTE_0 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 0, 56), sub_32));
1527 dag LE_BYTE_1 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 56, 56), sub_32));
1528 dag LE_BYTE_2 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 48, 56), sub_32));
1529 dag LE_BYTE_3 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 40, 56), sub_32));
1530 dag LE_BYTE_4 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 32, 56), sub_32));
1531 dag LE_BYTE_5 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 24, 56), sub_32));
1532 dag LE_BYTE_6 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 16, 56), sub_32));
1533 dag LE_BYTE_7 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 8, 56), sub_32));
1534 dag LE_BYTE_8 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 0, 56), sub_32));
1535 dag LE_BYTE_9 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 56, 56), sub_32));
1536 dag LE_BYTE_10 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 48, 56), sub_32));
1537 dag LE_BYTE_11 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 40, 56), sub_32));
1538 dag LE_BYTE_12 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 32, 56), sub_32));
1539 dag LE_BYTE_13 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 24, 56), sub_32));
1540 dag LE_BYTE_14 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 16, 56), sub_32));
1541 dag LE_BYTE_15 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 8, 56), sub_32));
1543 /* Variable element number (BE and LE patterns must be specified separately)
1544 This is a rather involved process.
1546 Conceptually, this is how the move is accomplished:
1547 1. Identify which doubleword contains the element
1548 2. Shift in the VMX register so that the correct doubleword is correctly
1549 lined up for the MFVSRD
1550 3. Perform the move so that the element (along with some extra stuff)
1552 4. Right shift within the GPR so that the element is right-justified
1554 Of course, the index is an element number which has a different meaning
1555 on LE/BE so the patterns have to be specified separately.
1557 Note: The final result will be the element right-justified with high
1558 order bits being arbitrarily defined (namely, whatever was in the
1559 vector register to the left of the value originally).
1564 - For elements 0-7, we shift left by 8 bytes since they're on the right
1565 - For elements 8-15, we need not shift (shift left by zero bytes)
1566 This is accomplished by inverting the bits of the index and AND-ing
1567 with 0x8 (i.e. clearing all bits of the index and inverting bit 60).
1569 dag LE_VBYTE_PERM_VEC = (LVSL ZERO8, (ANDC8 (LI8 8), $Idx));
1572 // - Now that we set up the shift amount, we shift in the VMX register
1573 dag LE_VBYTE_PERMUTE = (VPERM $S, $S, LE_VBYTE_PERM_VEC);
1576 // - The doubleword containing our element is moved to a GPR
1577 dag LE_MV_VBYTE = (MFVSRD
1579 (v2i64 (COPY_TO_REGCLASS LE_VBYTE_PERMUTE, VSRC)),
1583 - Truncate the element number to the range 0-7 (8-15 are symmetrical
1584 and out of range values are truncated accordingly)
1585 - Multiply by 8 as we need to shift right by the number of bits, not bytes
1586 - Shift right in the GPR by the calculated value
1588 dag LE_VBYTE_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 7), $Idx), 3, 60),
1590 dag LE_VARIABLE_BYTE = (EXTRACT_SUBREG (SRD LE_MV_VBYTE, LE_VBYTE_SHIFT),
1593 /* LE variable halfword
1595 - For elements 0-3, we shift left by 8 since they're on the right
1596 - For elements 4-7, we need not shift (shift left by zero bytes)
1597 Similarly to the byte pattern, we invert the bits of the index, but we
1598 AND with 0x4 (i.e. clear all bits of the index and invert bit 61).
1599 Of course, the shift is still by 8 bytes, so we must multiply by 2.
1601 dag LE_VHALF_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDC8 (LI8 4), $Idx), 1, 62));
1604 // - Now that we set up the shift amount, we shift in the VMX register
1605 dag LE_VHALF_PERMUTE = (VPERM $S, $S, LE_VHALF_PERM_VEC);
1608 // - The doubleword containing our element is moved to a GPR
1609 dag LE_MV_VHALF = (MFVSRD
1611 (v2i64 (COPY_TO_REGCLASS LE_VHALF_PERMUTE, VSRC)),
1615 - Truncate the element number to the range 0-3 (4-7 are symmetrical
1616 and out of range values are truncated accordingly)
1617 - Multiply by 16 as we need to shift right by the number of bits
1618 - Shift right in the GPR by the calculated value
1620 dag LE_VHALF_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 3), $Idx), 4, 59),
1622 dag LE_VARIABLE_HALF = (EXTRACT_SUBREG (SRD LE_MV_VHALF, LE_VHALF_SHIFT),
1627 - For elements 0-1, we shift left by 8 since they're on the right
1628 - For elements 2-3, we need not shift
1630 dag LE_VWORD_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDC8 (LI8 2), $Idx), 2, 61));
1633 // - Now that we set up the shift amount, we shift in the VMX register
1634 dag LE_VWORD_PERMUTE = (VPERM $S, $S, LE_VWORD_PERM_VEC);
1637 // - The doubleword containing our element is moved to a GPR
1638 dag LE_MV_VWORD = (MFVSRD
1640 (v2i64 (COPY_TO_REGCLASS LE_VWORD_PERMUTE, VSRC)),
1644 - Truncate the element number to the range 0-1 (2-3 are symmetrical
1645 and out of range values are truncated accordingly)
1646 - Multiply by 32 as we need to shift right by the number of bits
1647 - Shift right in the GPR by the calculated value
1649 dag LE_VWORD_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 1), $Idx), 5, 58),
1651 dag LE_VARIABLE_WORD = (EXTRACT_SUBREG (SRD LE_MV_VWORD, LE_VWORD_SHIFT),
1654 /* LE variable doubleword
1656 - For element 0, we shift left by 8 since it's on the right
1657 - For element 1, we need not shift
1659 dag LE_VDWORD_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDC8 (LI8 1), $Idx), 3, 60));
1662 // - Now that we set up the shift amount, we shift in the VMX register
1663 dag LE_VDWORD_PERMUTE = (VPERM $S, $S, LE_VDWORD_PERM_VEC);
1666 // - The doubleword containing our element is moved to a GPR
1667 // - Number 4. is not needed for the doubleword as the value is 64-bits
1668 dag LE_VARIABLE_DWORD =
1669 (MFVSRD (EXTRACT_SUBREG
1670 (v2i64 (COPY_TO_REGCLASS LE_VDWORD_PERMUTE, VSRC)),
1673 /* LE variable float
1674 - Shift the vector to line up the desired element to BE Word 0
1675 - Convert 32-bit float to a 64-bit single precision float
1677 dag LE_VFLOAT_PERM_VEC = (LVSL ZERO8, (RLDICR (XOR8 (LI8 3), $Idx), 2, 61));
1678 dag LE_VFLOAT_PERMUTE = (VPERM $S, $S, LE_VFLOAT_PERM_VEC);
1679 dag LE_VARIABLE_FLOAT = (XSCVSPDPN LE_VFLOAT_PERMUTE);
1681 /* LE variable double
1682 Same as the LE doubleword except there is no move.
1684 dag LE_VDOUBLE_PERMUTE = (VPERM (COPY_TO_REGCLASS $S, VRRC),
1685 (COPY_TO_REGCLASS $S, VRRC),
1686 LE_VDWORD_PERM_VEC);
1687 dag LE_VARIABLE_DOUBLE = (COPY_TO_REGCLASS LE_VDOUBLE_PERMUTE, VSRC);
1690 The algorithm here is the same as the LE variable byte except:
1691 - The shift in the VMX register is by 0/8 for opposite element numbers so
1692 we simply AND the element number with 0x8
1693 - The order of elements after the move to GPR is reversed, so we invert
1694 the bits of the index prior to truncating to the range 0-7
1696 dag BE_VBYTE_PERM_VEC = (LVSL ZERO8, (ANDIo8 $Idx, 8));
1697 dag BE_VBYTE_PERMUTE = (VPERM $S, $S, BE_VBYTE_PERM_VEC);
1698 dag BE_MV_VBYTE = (MFVSRD
1700 (v2i64 (COPY_TO_REGCLASS BE_VBYTE_PERMUTE, VSRC)),
1702 dag BE_VBYTE_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 7), $Idx), 3, 60),
1704 dag BE_VARIABLE_BYTE = (EXTRACT_SUBREG (SRD BE_MV_VBYTE, BE_VBYTE_SHIFT),
1707 /* BE variable halfword
1708 The algorithm here is the same as the LE variable halfword except:
1709 - The shift in the VMX register is by 0/8 for opposite element numbers so
1710 we simply AND the element number with 0x4 and multiply by 2
1711 - The order of elements after the move to GPR is reversed, so we invert
1712 the bits of the index prior to truncating to the range 0-3
1714 dag BE_VHALF_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDIo8 $Idx, 4), 1, 62));
1715 dag BE_VHALF_PERMUTE = (VPERM $S, $S, BE_VHALF_PERM_VEC);
1716 dag BE_MV_VHALF = (MFVSRD
1718 (v2i64 (COPY_TO_REGCLASS BE_VHALF_PERMUTE, VSRC)),
1720 dag BE_VHALF_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 3), $Idx), 4, 59),
1722 dag BE_VARIABLE_HALF = (EXTRACT_SUBREG (SRD BE_MV_VHALF, BE_VHALF_SHIFT),
1726 The algorithm is the same as the LE variable word except:
1727 - The shift in the VMX register happens for opposite element numbers
1728 - The order of elements after the move to GPR is reversed, so we invert
1729 the bits of the index prior to truncating to the range 0-1
1731 dag BE_VWORD_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDIo8 $Idx, 2), 2, 61));
1732 dag BE_VWORD_PERMUTE = (VPERM $S, $S, BE_VWORD_PERM_VEC);
1733 dag BE_MV_VWORD = (MFVSRD
1735 (v2i64 (COPY_TO_REGCLASS BE_VWORD_PERMUTE, VSRC)),
1737 dag BE_VWORD_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 1), $Idx), 5, 58),
1739 dag BE_VARIABLE_WORD = (EXTRACT_SUBREG (SRD BE_MV_VWORD, BE_VWORD_SHIFT),
1742 /* BE variable doubleword
1743 Same as the LE doubleword except we shift in the VMX register for opposite
1746 dag BE_VDWORD_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDIo8 $Idx, 1), 3, 60));
1747 dag BE_VDWORD_PERMUTE = (VPERM $S, $S, BE_VDWORD_PERM_VEC);
1748 dag BE_VARIABLE_DWORD =
1749 (MFVSRD (EXTRACT_SUBREG
1750 (v2i64 (COPY_TO_REGCLASS BE_VDWORD_PERMUTE, VSRC)),
1753 /* BE variable float
1754 - Shift the vector to line up the desired element to BE Word 0
1755 - Convert 32-bit float to a 64-bit single precision float
1757 dag BE_VFLOAT_PERM_VEC = (LVSL ZERO8, (RLDICR $Idx, 2, 61));
1758 dag BE_VFLOAT_PERMUTE = (VPERM $S, $S, BE_VFLOAT_PERM_VEC);
1759 dag BE_VARIABLE_FLOAT = (XSCVSPDPN BE_VFLOAT_PERMUTE);
1761 /* BE variable double
1762 Same as the BE doubleword except there is no move.
1764 dag BE_VDOUBLE_PERMUTE = (VPERM (COPY_TO_REGCLASS $S, VRRC),
1765 (COPY_TO_REGCLASS $S, VRRC),
1766 BE_VDWORD_PERM_VEC);
1767 dag BE_VARIABLE_DOUBLE = (COPY_TO_REGCLASS BE_VDOUBLE_PERMUTE, VSRC);
1770 let AddedComplexity = 400 in {
1771 // v4f32 scalar <-> vector conversions (BE)
1772 let Predicates = [IsBigEndian, HasP8Vector] in {
1773 def : Pat<(v4f32 (scalar_to_vector f32:$A)),
1774 (v4f32 (XSCVDPSPN $A))>;
1775 def : Pat<(f32 (vector_extract v4f32:$S, 0)),
1776 (f32 (XSCVSPDPN $S))>;
1777 def : Pat<(f32 (vector_extract v4f32:$S, 1)),
1778 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 1)))>;
1779 def : Pat<(f32 (vector_extract v4f32:$S, 2)),
1780 (f32 (XSCVSPDPN (XXPERMDI $S, $S, 2)))>;
1781 def : Pat<(f32 (vector_extract v4f32:$S, 3)),
1782 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 3)))>;
1783 def : Pat<(f32 (vector_extract v4f32:$S, i64:$Idx)),
1784 (f32 VectorExtractions.BE_VARIABLE_FLOAT)>;
1785 } // IsBigEndian, HasP8Vector
1787 // Variable index vector_extract for v2f64 does not require P8Vector
1788 let Predicates = [IsBigEndian, HasVSX] in
1789 def : Pat<(f64 (vector_extract v2f64:$S, i64:$Idx)),
1790 (f64 VectorExtractions.BE_VARIABLE_DOUBLE)>;
1792 let Predicates = [IsBigEndian, HasDirectMove] in {
1793 // v16i8 scalar <-> vector conversions (BE)
1794 def : Pat<(v16i8 (scalar_to_vector i32:$A)),
1795 (v16i8 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_BYTE_0, sub_64))>;
1796 def : Pat<(v8i16 (scalar_to_vector i32:$A)),
1797 (v8i16 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_HALF_0, sub_64))>;
1798 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
1799 (v4i32 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_WORD_0, sub_64))>;
1800 def : Pat<(v2i64 (scalar_to_vector i64:$A)),
1801 (v2i64 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_DWORD_0, sub_64))>;
1802 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
1803 (i32 VectorExtractions.LE_BYTE_15)>;
1804 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
1805 (i32 VectorExtractions.LE_BYTE_14)>;
1806 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
1807 (i32 VectorExtractions.LE_BYTE_13)>;
1808 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
1809 (i32 VectorExtractions.LE_BYTE_12)>;
1810 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
1811 (i32 VectorExtractions.LE_BYTE_11)>;
1812 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
1813 (i32 VectorExtractions.LE_BYTE_10)>;
1814 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
1815 (i32 VectorExtractions.LE_BYTE_9)>;
1816 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
1817 (i32 VectorExtractions.LE_BYTE_8)>;
1818 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
1819 (i32 VectorExtractions.LE_BYTE_7)>;
1820 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
1821 (i32 VectorExtractions.LE_BYTE_6)>;
1822 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
1823 (i32 VectorExtractions.LE_BYTE_5)>;
1824 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
1825 (i32 VectorExtractions.LE_BYTE_4)>;
1826 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
1827 (i32 VectorExtractions.LE_BYTE_3)>;
1828 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
1829 (i32 VectorExtractions.LE_BYTE_2)>;
1830 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
1831 (i32 VectorExtractions.LE_BYTE_1)>;
1832 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
1833 (i32 VectorExtractions.LE_BYTE_0)>;
1834 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
1835 (i32 VectorExtractions.BE_VARIABLE_BYTE)>;
1837 // v8i16 scalar <-> vector conversions (BE)
1838 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
1839 (i32 VectorExtractions.LE_HALF_7)>;
1840 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
1841 (i32 VectorExtractions.LE_HALF_6)>;
1842 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
1843 (i32 VectorExtractions.LE_HALF_5)>;
1844 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
1845 (i32 VectorExtractions.LE_HALF_4)>;
1846 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
1847 (i32 VectorExtractions.LE_HALF_3)>;
1848 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
1849 (i32 VectorExtractions.LE_HALF_2)>;
1850 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
1851 (i32 VectorExtractions.LE_HALF_1)>;
1852 def : Pat<(i32 (vector_extract v8i16:$S, 7)),
1853 (i32 VectorExtractions.LE_HALF_0)>;
1854 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
1855 (i32 VectorExtractions.BE_VARIABLE_HALF)>;
1857 // v4i32 scalar <-> vector conversions (BE)
1858 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
1859 (i32 VectorExtractions.LE_WORD_3)>;
1860 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
1861 (i32 VectorExtractions.LE_WORD_2)>;
1862 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
1863 (i32 VectorExtractions.LE_WORD_1)>;
1864 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
1865 (i32 VectorExtractions.LE_WORD_0)>;
1866 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
1867 (i32 VectorExtractions.BE_VARIABLE_WORD)>;
1869 // v2i64 scalar <-> vector conversions (BE)
1870 def : Pat<(i64 (vector_extract v2i64:$S, 0)),
1871 (i64 VectorExtractions.LE_DWORD_1)>;
1872 def : Pat<(i64 (vector_extract v2i64:$S, 1)),
1873 (i64 VectorExtractions.LE_DWORD_0)>;
1874 def : Pat<(i64 (vector_extract v2i64:$S, i64:$Idx)),
1875 (i64 VectorExtractions.BE_VARIABLE_DWORD)>;
1876 } // IsBigEndian, HasDirectMove
1878 // v4f32 scalar <-> vector conversions (LE)
1879 let Predicates = [IsLittleEndian, HasP8Vector] in {
1880 def : Pat<(v4f32 (scalar_to_vector f32:$A)),
1881 (v4f32 (XXSLDWI (XSCVDPSPN $A), (XSCVDPSPN $A), 1))>;
1882 def : Pat<(f32 (vector_extract v4f32:$S, 0)),
1883 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 3)))>;
1884 def : Pat<(f32 (vector_extract v4f32:$S, 1)),
1885 (f32 (XSCVSPDPN (XXPERMDI $S, $S, 2)))>;
1886 def : Pat<(f32 (vector_extract v4f32:$S, 2)),
1887 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 1)))>;
1888 def : Pat<(f32 (vector_extract v4f32:$S, 3)),
1889 (f32 (XSCVSPDPN $S))>;
1890 def : Pat<(f32 (vector_extract v4f32:$S, i64:$Idx)),
1891 (f32 VectorExtractions.LE_VARIABLE_FLOAT)>;
1892 } // IsLittleEndian, HasP8Vector
1894 // Variable index vector_extract for v2f64 does not require P8Vector
1895 let Predicates = [IsLittleEndian, HasVSX] in
1896 def : Pat<(f64 (vector_extract v2f64:$S, i64:$Idx)),
1897 (f64 VectorExtractions.LE_VARIABLE_DOUBLE)>;
1899 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x_be xoaddr:$src)), (LXVW4X xoaddr:$src)>;
1900 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x_be xoaddr:$src)), (LXVD2X xoaddr:$src)>;
1902 let Predicates = [IsLittleEndian, HasDirectMove] in {
1903 // v16i8 scalar <-> vector conversions (LE)
1904 def : Pat<(v16i8 (scalar_to_vector i32:$A)),
1905 (v16i8 (COPY_TO_REGCLASS MovesToVSR.LE_WORD_0, VSRC))>;
1906 def : Pat<(v8i16 (scalar_to_vector i32:$A)),
1907 (v8i16 (COPY_TO_REGCLASS MovesToVSR.LE_WORD_0, VSRC))>;
1908 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
1909 (v4i32 MovesToVSR.LE_WORD_0)>;
1910 def : Pat<(v2i64 (scalar_to_vector i64:$A)),
1911 (v2i64 MovesToVSR.LE_DWORD_0)>;
1912 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
1913 (i32 VectorExtractions.LE_BYTE_0)>;
1914 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
1915 (i32 VectorExtractions.LE_BYTE_1)>;
1916 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
1917 (i32 VectorExtractions.LE_BYTE_2)>;
1918 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
1919 (i32 VectorExtractions.LE_BYTE_3)>;
1920 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
1921 (i32 VectorExtractions.LE_BYTE_4)>;
1922 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
1923 (i32 VectorExtractions.LE_BYTE_5)>;
1924 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
1925 (i32 VectorExtractions.LE_BYTE_6)>;
1926 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
1927 (i32 VectorExtractions.LE_BYTE_7)>;
1928 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
1929 (i32 VectorExtractions.LE_BYTE_8)>;
1930 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
1931 (i32 VectorExtractions.LE_BYTE_9)>;
1932 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
1933 (i32 VectorExtractions.LE_BYTE_10)>;
1934 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
1935 (i32 VectorExtractions.LE_BYTE_11)>;
1936 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
1937 (i32 VectorExtractions.LE_BYTE_12)>;
1938 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
1939 (i32 VectorExtractions.LE_BYTE_13)>;
1940 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
1941 (i32 VectorExtractions.LE_BYTE_14)>;
1942 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
1943 (i32 VectorExtractions.LE_BYTE_15)>;
1944 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
1945 (i32 VectorExtractions.LE_VARIABLE_BYTE)>;
1947 // v8i16 scalar <-> vector conversions (LE)
1948 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
1949 (i32 VectorExtractions.LE_HALF_0)>;
1950 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
1951 (i32 VectorExtractions.LE_HALF_1)>;
1952 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
1953 (i32 VectorExtractions.LE_HALF_2)>;
1954 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
1955 (i32 VectorExtractions.LE_HALF_3)>;
1956 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
1957 (i32 VectorExtractions.LE_HALF_4)>;
1958 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
1959 (i32 VectorExtractions.LE_HALF_5)>;
1960 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
1961 (i32 VectorExtractions.LE_HALF_6)>;
1962 def : Pat<(i32 (vector_extract v8i16:$S, 7)),
1963 (i32 VectorExtractions.LE_HALF_7)>;
1964 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
1965 (i32 VectorExtractions.LE_VARIABLE_HALF)>;
1967 // v4i32 scalar <-> vector conversions (LE)
1968 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
1969 (i32 VectorExtractions.LE_WORD_0)>;
1970 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
1971 (i32 VectorExtractions.LE_WORD_1)>;
1972 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
1973 (i32 VectorExtractions.LE_WORD_2)>;
1974 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
1975 (i32 VectorExtractions.LE_WORD_3)>;
1976 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
1977 (i32 VectorExtractions.LE_VARIABLE_WORD)>;
1979 // v2i64 scalar <-> vector conversions (LE)
1980 def : Pat<(i64 (vector_extract v2i64:$S, 0)),
1981 (i64 VectorExtractions.LE_DWORD_0)>;
1982 def : Pat<(i64 (vector_extract v2i64:$S, 1)),
1983 (i64 VectorExtractions.LE_DWORD_1)>;
1984 def : Pat<(i64 (vector_extract v2i64:$S, i64:$Idx)),
1985 (i64 VectorExtractions.LE_VARIABLE_DWORD)>;
1986 } // IsLittleEndian, HasDirectMove
1988 let Predicates = [HasDirectMove, HasVSX] in {
1989 // bitconvert f32 -> i32
1990 // (convert to 32-bit fp single, shift right 1 word, move to GPR)
1991 def : Pat<(i32 (bitconvert f32:$S)),
1992 (i32 (MFVSRWZ (EXTRACT_SUBREG
1993 (XXSLDWI (XSCVDPSPN $S),(XSCVDPSPN $S), 3),
1995 // bitconvert i32 -> f32
1996 // (move to FPR, shift left 1 word, convert to 64-bit fp single)
1997 def : Pat<(f32 (bitconvert i32:$A)),
1999 (XXSLDWI MovesToVSR.LE_WORD_1, MovesToVSR.LE_WORD_1, 1)))>;
2001 // bitconvert f64 -> i64
2002 // (move to GPR, nothing else needed)
2003 def : Pat<(i64 (bitconvert f64:$S)),
2006 // bitconvert i64 -> f64
2007 // (move to FPR, nothing else needed)
2008 def : Pat<(f64 (bitconvert i64:$S)),
2012 // Materialize a zero-vector of long long
2013 def : Pat<(v2i64 immAllZerosV),
2018 dag F32_TO_BE_WORD1 = (v4f32 (XXSLDWI (XSCVDPSPN $B), (XSCVDPSPN $B), 3));
2019 dag I32_TO_BE_WORD1 = (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC);
2022 // The following VSX instructions were introduced in Power ISA 3.0
2023 def HasP9Vector : Predicate<"PPCSubTarget->hasP9Vector()">;
2024 let AddedComplexity = 400, Predicates = [HasP9Vector] in {
2026 // [PO VRT XO VRB XO /]
2027 class X_VT5_XO5_VB5<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2029 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vrrc:$vT), (ins vrrc:$vB),
2030 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
2032 // [PO VRT XO VRB XO RO], Round to Odd version of [PO VRT XO VRB XO /]
2033 class X_VT5_XO5_VB5_Ro<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2035 : X_VT5_XO5_VB5<opcode, xo2, xo, opc, pattern>, isDOT;
2037 // [PO VRT XO VRB XO /], but the VRB is only used the left 64 bits (or less),
2038 // So we use different operand class for VRB
2039 class X_VT5_XO5_VB5_TyVB<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
2040 RegisterOperand vbtype, list<dag> pattern>
2041 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vrrc:$vT), (ins vbtype:$vB),
2042 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
2044 let UseVSXReg = 1 in {
2045 // [PO T XO B XO BX /]
2046 class XX2_RT5_XO5_XB6<bits<6> opcode, bits<5> xo2, bits<9> xo, string opc,
2048 : XX2_RD5_XO5_RS6<opcode, xo2, xo, (outs g8rc:$rT), (ins vsfrc:$XB),
2049 !strconcat(opc, " $rT, $XB"), IIC_VecFP, pattern>;
2051 // [PO T XO B XO BX TX]
2052 class XX2_XT6_XO5_XB6<bits<6> opcode, bits<5> xo2, bits<9> xo, string opc,
2053 RegisterOperand vtype, list<dag> pattern>
2054 : XX2_RD6_XO5_RS6<opcode, xo2, xo, (outs vtype:$XT), (ins vtype:$XB),
2055 !strconcat(opc, " $XT, $XB"), IIC_VecFP, pattern>;
2057 // [PO T A B XO AX BX TX], src and dest register use different operand class
2058 class XX3_XT5_XA5_XB5<bits<6> opcode, bits<8> xo, string opc,
2059 RegisterOperand xty, RegisterOperand aty, RegisterOperand bty,
2060 InstrItinClass itin, list<dag> pattern>
2061 : XX3Form<opcode, xo, (outs xty:$XT), (ins aty:$XA, bty:$XB),
2062 !strconcat(opc, " $XT, $XA, $XB"), itin, pattern>;
2065 // [PO VRT VRA VRB XO /]
2066 class X_VT5_VA5_VB5<bits<6> opcode, bits<10> xo, string opc,
2068 : XForm_1<opcode, xo, (outs vrrc:$vT), (ins vrrc:$vA, vrrc:$vB),
2069 !strconcat(opc, " $vT, $vA, $vB"), IIC_VecFP, pattern>;
2071 // [PO VRT VRA VRB XO RO], Round to Odd version of [PO VRT VRA VRB XO /]
2072 class X_VT5_VA5_VB5_Ro<bits<6> opcode, bits<10> xo, string opc,
2074 : X_VT5_VA5_VB5<opcode, xo, opc, pattern>, isDOT;
2076 //===--------------------------------------------------------------------===//
2077 // Quad-Precision Scalar Move Instructions:
2080 def XSCPSGNQP : X_VT5_VA5_VB5<63, 100, "xscpsgnqp", []>;
2082 // Absolute/Negative-Absolute/Negate
2083 def XSABSQP : X_VT5_XO5_VB5<63, 0, 804, "xsabsqp" , []>;
2084 def XSNABSQP : X_VT5_XO5_VB5<63, 8, 804, "xsnabsqp", []>;
2085 def XSNEGQP : X_VT5_XO5_VB5<63, 16, 804, "xsnegqp" , []>;
2087 //===--------------------------------------------------------------------===//
2088 // Quad-Precision Scalar Floating-Point Arithmetic Instructions:
2090 // Add/Divide/Multiply/Subtract
2091 def XSADDQP : X_VT5_VA5_VB5 <63, 4, "xsaddqp" , []>;
2092 def XSADDQPO : X_VT5_VA5_VB5_Ro<63, 4, "xsaddqpo", []>;
2093 def XSDIVQP : X_VT5_VA5_VB5 <63, 548, "xsdivqp" , []>;
2094 def XSDIVQPO : X_VT5_VA5_VB5_Ro<63, 548, "xsdivqpo", []>;
2095 def XSMULQP : X_VT5_VA5_VB5 <63, 36, "xsmulqp" , []>;
2096 def XSMULQPO : X_VT5_VA5_VB5_Ro<63, 36, "xsmulqpo", []>;
2097 def XSSUBQP : X_VT5_VA5_VB5 <63, 516, "xssubqp" , []>;
2098 def XSSUBQPO : X_VT5_VA5_VB5_Ro<63, 516, "xssubqpo", []>;
2101 def XSSQRTQP : X_VT5_XO5_VB5 <63, 27, 804, "xssqrtqp" , []>;
2102 def XSSQRTQPO : X_VT5_XO5_VB5_Ro<63, 27, 804, "xssqrtqpo", []>;
2104 // (Negative) Multiply-{Add/Subtract}
2105 def XSMADDQP : X_VT5_VA5_VB5 <63, 388, "xsmaddqp" , []>;
2106 def XSMADDQPO : X_VT5_VA5_VB5_Ro<63, 388, "xsmaddqpo" , []>;
2107 def XSMSUBQP : X_VT5_VA5_VB5 <63, 420, "xsmsubqp" , []>;
2108 def XSMSUBQPO : X_VT5_VA5_VB5_Ro<63, 420, "xsmsubqpo" , []>;
2109 def XSNMADDQP : X_VT5_VA5_VB5 <63, 452, "xsnmaddqp" , []>;
2110 def XSNMADDQPO: X_VT5_VA5_VB5_Ro<63, 452, "xsnmaddqpo", []>;
2111 def XSNMSUBQP : X_VT5_VA5_VB5 <63, 484, "xsnmsubqp" , []>;
2112 def XSNMSUBQPO: X_VT5_VA5_VB5_Ro<63, 484, "xsnmsubqpo", []>;
2114 //===--------------------------------------------------------------------===//
2115 // Quad/Double-Precision Compare Instructions:
2117 // [PO BF // VRA VRB XO /]
2118 class X_BF3_VA5_VB5<bits<6> opcode, bits<10> xo, string opc,
2120 : XForm_17<opcode, xo, (outs crrc:$crD), (ins vrrc:$VA, vrrc:$VB),
2121 !strconcat(opc, " $crD, $VA, $VB"), IIC_FPCompare> {
2122 let Pattern = pattern;
2125 // QP Compare Ordered/Unordered
2126 def XSCMPOQP : X_BF3_VA5_VB5<63, 132, "xscmpoqp", []>;
2127 def XSCMPUQP : X_BF3_VA5_VB5<63, 644, "xscmpuqp", []>;
2129 // DP/QP Compare Exponents
2130 def XSCMPEXPDP : XX3Form_1<60, 59,
2131 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
2132 "xscmpexpdp $crD, $XA, $XB", IIC_FPCompare, []>,
2134 def XSCMPEXPQP : X_BF3_VA5_VB5<63, 164, "xscmpexpqp", []>;
2136 // DP Compare ==, >=, >, !=
2137 // Use vsrc for XT, because the entire register of XT is set.
2138 // XT.dword[1] = 0x0000_0000_0000_0000
2139 def XSCMPEQDP : XX3_XT5_XA5_XB5<60, 3, "xscmpeqdp", vsrc, vsfrc, vsfrc,
2141 def XSCMPGEDP : XX3_XT5_XA5_XB5<60, 19, "xscmpgedp", vsrc, vsfrc, vsfrc,
2143 def XSCMPGTDP : XX3_XT5_XA5_XB5<60, 11, "xscmpgtdp", vsrc, vsfrc, vsfrc,
2145 def XSCMPNEDP : XX3_XT5_XA5_XB5<60, 27, "xscmpnedp", vsrc, vsfrc, vsfrc,
2147 let UseVSXReg = 1 in {
2148 // Vector Compare Not Equal
2149 def XVCMPNEDP : XX3Form_Rc<60, 123,
2150 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
2151 "xvcmpnedp $XT, $XA, $XB", IIC_VecFPCompare, []>;
2153 def XVCMPNEDPo : XX3Form_Rc<60, 123,
2154 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
2155 "xvcmpnedp. $XT, $XA, $XB", IIC_VecFPCompare, []>,
2157 def XVCMPNESP : XX3Form_Rc<60, 91,
2158 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
2159 "xvcmpnesp $XT, $XA, $XB", IIC_VecFPCompare, []>;
2161 def XVCMPNESPo : XX3Form_Rc<60, 91,
2162 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
2163 "xvcmpnesp. $XT, $XA, $XB", IIC_VecFPCompare, []>,
2167 //===--------------------------------------------------------------------===//
2168 // Quad-Precision Floating-Point Conversion Instructions:
2171 def XSCVDPQP : X_VT5_XO5_VB5_TyVB<63, 22, 836, "xscvdpqp", vfrc, []>;
2173 // Round & Convert QP -> DP (dword[1] is set to zero)
2174 def XSCVQPDP : X_VT5_XO5_VB5 <63, 20, 836, "xscvqpdp" , []>;
2175 def XSCVQPDPO : X_VT5_XO5_VB5_Ro<63, 20, 836, "xscvqpdpo", []>;
2177 // Truncate & Convert QP -> (Un)Signed (D)Word (dword[1] is set to zero)
2178 def XSCVQPSDZ : X_VT5_XO5_VB5<63, 25, 836, "xscvqpsdz", []>;
2179 def XSCVQPSWZ : X_VT5_XO5_VB5<63, 9, 836, "xscvqpswz", []>;
2180 def XSCVQPUDZ : X_VT5_XO5_VB5<63, 17, 836, "xscvqpudz", []>;
2181 def XSCVQPUWZ : X_VT5_XO5_VB5<63, 1, 836, "xscvqpuwz", []>;
2183 // Convert (Un)Signed DWord -> QP
2184 def XSCVSDQP : X_VT5_XO5_VB5_TyVB<63, 10, 836, "xscvsdqp", vfrc, []>;
2185 def XSCVUDQP : X_VT5_XO5_VB5_TyVB<63, 2, 836, "xscvudqp", vfrc, []>;
2187 let UseVSXReg = 1 in {
2188 //===--------------------------------------------------------------------===//
2189 // Round to Floating-Point Integer Instructions
2191 // (Round &) Convert DP <-> HP
2192 // Note! xscvdphp's src and dest register both use the left 64 bits, so we use
2193 // vsfrc for src and dest register. xscvhpdp's src only use the left 16 bits,
2194 // but we still use vsfrc for it.
2195 def XSCVDPHP : XX2_XT6_XO5_XB6<60, 17, 347, "xscvdphp", vsfrc, []>;
2196 def XSCVHPDP : XX2_XT6_XO5_XB6<60, 16, 347, "xscvhpdp", vsfrc, []>;
2199 def XVCVHPSP : XX2_XT6_XO5_XB6<60, 24, 475, "xvcvhpsp", vsrc, []>;
2200 def XVCVSPHP : XX2_XT6_XO5_XB6<60, 25, 475, "xvcvsphp", vsrc,
2202 (int_ppc_vsx_xvcvsphp v4f32:$XB))]>;
2206 // Pattern for matching Vector HP -> Vector SP intrinsic. Defined as a
2207 // separate pattern so that it can convert the input register class from
2208 // VRRC(v8i16) to VSRC.
2209 def : Pat<(v4f32 (int_ppc_vsx_xvcvhpsp v8i16:$A)),
2210 (v4f32 (XVCVHPSP (COPY_TO_REGCLASS $A, VSRC)))>;
2212 class Z23_VT5_R1_VB5_RMC2_EX1<bits<6> opcode, bits<8> xo, bit ex, string opc,
2214 : Z23Form_1<opcode, xo,
2215 (outs vrrc:$vT), (ins u1imm:$r, vrrc:$vB, u2imm:$rmc),
2216 !strconcat(opc, " $r, $vT, $vB, $rmc"), IIC_VecFP, pattern> {
2220 // Round to Quad-Precision Integer [with Inexact]
2221 def XSRQPI : Z23_VT5_R1_VB5_RMC2_EX1<63, 5, 0, "xsrqpi" , []>;
2222 def XSRQPIX : Z23_VT5_R1_VB5_RMC2_EX1<63, 5, 1, "xsrqpix", []>;
2224 // Round Quad-Precision to Double-Extended Precision (fp80)
2225 def XSRQPXP : Z23_VT5_R1_VB5_RMC2_EX1<63, 37, 0, "xsrqpxp", []>;
2227 //===--------------------------------------------------------------------===//
2228 // Insert/Extract Instructions
2230 // Insert Exponent DP/QP
2231 // XT NOTE: XT.dword[1] = 0xUUUU_UUUU_UUUU_UUUU
2232 def XSIEXPDP : XX1Form <60, 918, (outs vsrc:$XT), (ins g8rc:$rA, g8rc:$rB),
2233 "xsiexpdp $XT, $rA, $rB", IIC_VecFP, []>, UseVSXReg;
2234 // vB NOTE: only vB.dword[0] is used, that's why we don't use
2235 // X_VT5_VA5_VB5 form
2236 def XSIEXPQP : XForm_18<63, 868, (outs vrrc:$vT), (ins vrrc:$vA, vsfrc:$vB),
2237 "xsiexpqp $vT, $vA, $vB", IIC_VecFP, []>;
2239 // Extract Exponent/Significand DP/QP
2240 def XSXEXPDP : XX2_RT5_XO5_XB6<60, 0, 347, "xsxexpdp", []>;
2241 def XSXSIGDP : XX2_RT5_XO5_XB6<60, 1, 347, "xsxsigdp", []>;
2243 def XSXEXPQP : X_VT5_XO5_VB5 <63, 2, 804, "xsxexpqp", []>;
2244 def XSXSIGQP : X_VT5_XO5_VB5 <63, 18, 804, "xsxsigqp", []>;
2246 // Vector Insert Word
2247 let UseVSXReg = 1 in {
2248 // XB NOTE: Only XB.dword[1] is used, but we use vsrc on XB.
2250 XX2_RD6_UIM5_RS6<60, 181, (outs vsrc:$XT),
2251 (ins vsrc:$XTi, vsrc:$XB, u4imm:$UIM),
2252 "xxinsertw $XT, $XB, $UIM", IIC_VecFP,
2253 [(set v4i32:$XT, (PPCxxinsert v4i32:$XTi, v4i32:$XB,
2254 imm32SExt16:$UIM))]>,
2255 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">;
2257 // Vector Extract Unsigned Word
2258 def XXEXTRACTUW : XX2_RD6_UIM5_RS6<60, 165,
2259 (outs vsfrc:$XT), (ins vsrc:$XB, u4imm:$UIMM),
2260 "xxextractuw $XT, $XB, $UIMM", IIC_VecFP, []>;
2263 // Vector Insert Exponent DP/SP
2264 def XVIEXPDP : XX3_XT5_XA5_XB5<60, 248, "xviexpdp", vsrc, vsrc, vsrc,
2265 IIC_VecFP, [(set v2f64: $XT,(int_ppc_vsx_xviexpdp v2i64:$XA, v2i64:$XB))]>;
2266 def XVIEXPSP : XX3_XT5_XA5_XB5<60, 216, "xviexpsp", vsrc, vsrc, vsrc,
2267 IIC_VecFP, [(set v4f32: $XT,(int_ppc_vsx_xviexpsp v4i32:$XA, v4i32:$XB))]>;
2269 // Vector Extract Exponent/Significand DP/SP
2270 def XVXEXPDP : XX2_XT6_XO5_XB6<60, 0, 475, "xvxexpdp", vsrc,
2272 (int_ppc_vsx_xvxexpdp v2f64:$XB))]>;
2273 def XVXEXPSP : XX2_XT6_XO5_XB6<60, 8, 475, "xvxexpsp", vsrc,
2275 (int_ppc_vsx_xvxexpsp v4f32:$XB))]>;
2276 def XVXSIGDP : XX2_XT6_XO5_XB6<60, 1, 475, "xvxsigdp", vsrc,
2278 (int_ppc_vsx_xvxsigdp v2f64:$XB))]>;
2279 def XVXSIGSP : XX2_XT6_XO5_XB6<60, 9, 475, "xvxsigsp", vsrc,
2281 (int_ppc_vsx_xvxsigsp v4f32:$XB))]>;
2283 let AddedComplexity = 400, Predicates = [HasP9Vector] in {
2284 // Extra patterns expanding to vector Extract Word/Insert Word
2285 def : Pat<(v4i32 (int_ppc_vsx_xxinsertw v4i32:$A, v2i64:$B, imm:$IMM)),
2286 (v4i32 (XXINSERTW $A, $B, imm:$IMM))>;
2287 def : Pat<(v2i64 (int_ppc_vsx_xxextractuw v2i64:$A, imm:$IMM)),
2288 (v2i64 (COPY_TO_REGCLASS (XXEXTRACTUW $A, imm:$IMM), VSRC))>;
2289 } // AddedComplexity = 400, HasP9Vector
2291 //===--------------------------------------------------------------------===//
2293 // Test Data Class SP/DP/QP
2294 let UseVSXReg = 1 in {
2295 def XSTSTDCSP : XX2_BF3_DCMX7_RS6<60, 298,
2296 (outs crrc:$BF), (ins u7imm:$DCMX, vsfrc:$XB),
2297 "xststdcsp $BF, $XB, $DCMX", IIC_VecFP, []>;
2298 def XSTSTDCDP : XX2_BF3_DCMX7_RS6<60, 362,
2299 (outs crrc:$BF), (ins u7imm:$DCMX, vsfrc:$XB),
2300 "xststdcdp $BF, $XB, $DCMX", IIC_VecFP, []>;
2302 def XSTSTDCQP : X_BF3_DCMX7_RS5 <63, 708,
2303 (outs crrc:$BF), (ins u7imm:$DCMX, vrrc:$vB),
2304 "xststdcqp $BF, $vB, $DCMX", IIC_VecFP, []>;
2306 // Vector Test Data Class SP/DP
2307 let UseVSXReg = 1 in {
2308 def XVTSTDCSP : XX2_RD6_DCMX7_RS6<60, 13, 5,
2309 (outs vsrc:$XT), (ins u7imm:$DCMX, vsrc:$XB),
2310 "xvtstdcsp $XT, $XB, $DCMX", IIC_VecFP,
2312 (int_ppc_vsx_xvtstdcsp v4f32:$XB, imm:$DCMX))]>;
2313 def XVTSTDCDP : XX2_RD6_DCMX7_RS6<60, 15, 5,
2314 (outs vsrc:$XT), (ins u7imm:$DCMX, vsrc:$XB),
2315 "xvtstdcdp $XT, $XB, $DCMX", IIC_VecFP,
2317 (int_ppc_vsx_xvtstdcdp v2f64:$XB, imm:$DCMX))]>;
2320 //===--------------------------------------------------------------------===//
2322 // Maximum/Minimum Type-C/Type-J DP
2323 // XT.dword[1] = 0xUUUU_UUUU_UUUU_UUUU, so we use vsrc for XT
2324 def XSMAXCDP : XX3_XT5_XA5_XB5<60, 128, "xsmaxcdp", vsrc, vsfrc, vsfrc,
2326 def XSMAXJDP : XX3_XT5_XA5_XB5<60, 144, "xsmaxjdp", vsrc, vsfrc, vsfrc,
2328 def XSMINCDP : XX3_XT5_XA5_XB5<60, 136, "xsmincdp", vsrc, vsfrc, vsfrc,
2330 def XSMINJDP : XX3_XT5_XA5_XB5<60, 152, "xsminjdp", vsrc, vsfrc, vsfrc,
2333 //===--------------------------------------------------------------------===//
2335 // Vector Byte-Reverse H/W/D/Q Word
2336 def XXBRH : XX2_XT6_XO5_XB6<60, 7, 475, "xxbrh", vsrc, []>;
2337 def XXBRW : XX2_XT6_XO5_XB6<60, 15, 475, "xxbrw", vsrc, []>;
2338 def XXBRD : XX2_XT6_XO5_XB6<60, 23, 475, "xxbrd", vsrc, []>;
2339 def XXBRQ : XX2_XT6_XO5_XB6<60, 31, 475, "xxbrq", vsrc, []>;
2342 def XXPERM : XX3_XT5_XA5_XB5<60, 26, "xxperm" , vsrc, vsrc, vsrc,
2344 def XXPERMR : XX3_XT5_XA5_XB5<60, 58, "xxpermr", vsrc, vsrc, vsrc,
2347 // Vector Splat Immediate Byte
2348 def XXSPLTIB : X_RD6_IMM8<60, 360, (outs vsrc:$XT), (ins u8imm:$IMM8),
2349 "xxspltib $XT, $IMM8", IIC_VecPerm, []>, UseVSXReg;
2351 //===--------------------------------------------------------------------===//
2352 // Vector/Scalar Load/Store Instructions
2354 // When adding new D-Form loads/stores, be sure to update the ImmToIdxMap in
2355 // PPCRegisterInfo::PPCRegisterInfo and maybe save yourself some debugging.
2356 let mayLoad = 1, mayStore = 0 in {
2358 def LXV : DQ_RD6_RS5_DQ12<61, 1, (outs vsrc:$XT), (ins memrix16:$src),
2359 "lxv $XT, $src", IIC_LdStLFD, []>, UseVSXReg;
2361 def LXSD : DSForm_1<57, 2, (outs vfrc:$vD), (ins memrix:$src),
2362 "lxsd $vD, $src", IIC_LdStLFD, []>;
2363 // Load SP from src, convert it to DP, and place in dword[0]
2364 def LXSSP : DSForm_1<57, 3, (outs vfrc:$vD), (ins memrix:$src),
2365 "lxssp $vD, $src", IIC_LdStLFD, []>;
2367 // [PO T RA RB XO TX] almost equal to [PO S RA RB XO SX], but has different
2368 // "out" and "in" dag
2369 class X_XT6_RA5_RB5<bits<6> opcode, bits<10> xo, string opc,
2370 RegisterOperand vtype, list<dag> pattern>
2371 : XX1Form<opcode, xo, (outs vtype:$XT), (ins memrr:$src),
2372 !strconcat(opc, " $XT, $src"), IIC_LdStLFD, pattern>, UseVSXReg;
2374 // Load as Integer Byte/Halfword & Zero Indexed
2375 def LXSIBZX : X_XT6_RA5_RB5<31, 781, "lxsibzx", vsfrc,
2376 [(set f64:$XT, (PPClxsizx xoaddr:$src, 1))]>;
2377 def LXSIHZX : X_XT6_RA5_RB5<31, 813, "lxsihzx", vsfrc,
2378 [(set f64:$XT, (PPClxsizx xoaddr:$src, 2))]>;
2380 // Load Vector Halfword*8/Byte*16 Indexed
2381 def LXVH8X : X_XT6_RA5_RB5<31, 812, "lxvh8x" , vsrc, []>;
2382 def LXVB16X : X_XT6_RA5_RB5<31, 876, "lxvb16x", vsrc, []>;
2384 // Load Vector Indexed
2385 def LXVX : X_XT6_RA5_RB5<31, 268, "lxvx" , vsrc,
2386 [(set v2f64:$XT, (load xaddr:$src))]>;
2387 // Load Vector (Left-justified) with Length
2388 def LXVL : XX1Form<31, 269, (outs vsrc:$XT), (ins memr:$src, g8rc:$rB),
2389 "lxvl $XT, $src, $rB", IIC_LdStLoad,
2390 [(set v4i32:$XT, (int_ppc_vsx_lxvl addr:$src, i64:$rB))]>,
2392 def LXVLL : XX1Form<31,301, (outs vsrc:$XT), (ins memr:$src, g8rc:$rB),
2393 "lxvll $XT, $src, $rB", IIC_LdStLoad,
2394 [(set v4i32:$XT, (int_ppc_vsx_lxvll addr:$src, i64:$rB))]>,
2397 // Load Vector Word & Splat Indexed
2398 def LXVWSX : X_XT6_RA5_RB5<31, 364, "lxvwsx" , vsrc, []>;
2401 // When adding new D-Form loads/stores, be sure to update the ImmToIdxMap in
2402 // PPCRegisterInfo::PPCRegisterInfo and maybe save yourself some debugging.
2403 let mayStore = 1, mayLoad = 0 in {
2405 def STXV : DQ_RD6_RS5_DQ12<61, 5, (outs), (ins vsrc:$XT, memrix16:$dst),
2406 "stxv $XT, $dst", IIC_LdStSTFD, []>, UseVSXReg;
2408 def STXSD : DSForm_1<61, 2, (outs), (ins vfrc:$vS, memrix:$dst),
2409 "stxsd $vS, $dst", IIC_LdStSTFD, []>;
2410 // Convert DP of dword[0] to SP, and Store to dst
2411 def STXSSP : DSForm_1<61, 3, (outs), (ins vfrc:$vS, memrix:$dst),
2412 "stxssp $vS, $dst", IIC_LdStSTFD, []>;
2414 // [PO S RA RB XO SX]
2415 class X_XS6_RA5_RB5<bits<6> opcode, bits<10> xo, string opc,
2416 RegisterOperand vtype, list<dag> pattern>
2417 : XX1Form<opcode, xo, (outs), (ins vtype:$XT, memrr:$dst),
2418 !strconcat(opc, " $XT, $dst"), IIC_LdStSTFD, pattern>, UseVSXReg;
2420 // Store as Integer Byte/Halfword Indexed
2421 def STXSIBX : X_XS6_RA5_RB5<31, 909, "stxsibx" , vsfrc,
2422 [(PPCstxsix f64:$XT, xoaddr:$dst, 1)]>;
2423 def STXSIHX : X_XS6_RA5_RB5<31, 941, "stxsihx" , vsfrc,
2424 [(PPCstxsix f64:$XT, xoaddr:$dst, 2)]>;
2425 let isCodeGenOnly = 1 in {
2426 def STXSIBXv : X_XS6_RA5_RB5<31, 909, "stxsibx" , vrrc, []>;
2427 def STXSIHXv : X_XS6_RA5_RB5<31, 941, "stxsihx" , vrrc, []>;
2430 // Store Vector Halfword*8/Byte*16 Indexed
2431 def STXVH8X : X_XS6_RA5_RB5<31, 940, "stxvh8x" , vsrc, []>;
2432 def STXVB16X : X_XS6_RA5_RB5<31, 1004, "stxvb16x", vsrc, []>;
2434 // Store Vector Indexed
2435 def STXVX : X_XS6_RA5_RB5<31, 396, "stxvx" , vsrc,
2436 [(store v2f64:$XT, xaddr:$dst)]>;
2438 // Store Vector (Left-justified) with Length
2439 def STXVL : XX1Form<31, 397, (outs), (ins vsrc:$XT, memr:$dst, g8rc:$rB),
2440 "stxvl $XT, $dst, $rB", IIC_LdStLoad,
2441 [(int_ppc_vsx_stxvl v4i32:$XT, addr:$dst, i64:$rB)]>,
2443 def STXVLL : XX1Form<31, 429, (outs), (ins vsrc:$XT, memr:$dst, g8rc:$rB),
2444 "stxvll $XT, $dst, $rB", IIC_LdStLoad,
2445 [(int_ppc_vsx_stxvll v4i32:$XT, addr:$dst, i64:$rB)]>,
2449 // Patterns for which instructions from ISA 3.0 are a better match
2450 let Predicates = [IsLittleEndian, HasP9Vector] in {
2451 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 0))))),
2452 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 12)))>;
2453 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 1))))),
2454 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 8)))>;
2455 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 2))))),
2456 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 4)))>;
2457 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 3))))),
2458 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 0)))>;
2459 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 0)),
2460 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 12))>;
2461 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 1)),
2462 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 8))>;
2463 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 2)),
2464 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 4))>;
2465 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 3)),
2466 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 0))>;
2467 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 0)),
2468 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 12))>;
2469 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 1)),
2470 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 8))>;
2471 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 2)),
2472 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 4))>;
2473 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 3)),
2474 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 0))>;
2475 } // IsLittleEndian, HasP9Vector
2477 let Predicates = [IsBigEndian, HasP9Vector] in {
2478 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 0))))),
2479 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 0)))>;
2480 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 1))))),
2481 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 4)))>;
2482 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 2))))),
2483 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 8)))>;
2484 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 3))))),
2485 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 12)))>;
2486 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 0)),
2487 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 0))>;
2488 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 1)),
2489 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 4))>;
2490 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 2)),
2491 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 8))>;
2492 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 3)),
2493 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 12))>;
2494 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 0)),
2495 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 0))>;
2496 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 1)),
2497 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 4))>;
2498 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 2)),
2499 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 8))>;
2500 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 3)),
2501 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 12))>;
2502 } // IsLittleEndian, HasP9Vector
2504 // D-Form Load/Store
2505 def : Pat<(v4i32 (load iaddr:$src)), (LXV memrix16:$src)>;
2506 def : Pat<(v4f32 (load iaddr:$src)), (LXV memrix16:$src)>;
2507 def : Pat<(v2i64 (load iaddr:$src)), (LXV memrix16:$src)>;
2508 def : Pat<(v2f64 (load iaddr:$src)), (LXV memrix16:$src)>;
2509 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x iaddr:$src)), (LXV memrix16:$src)>;
2510 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x iaddr:$src)), (LXV memrix16:$src)>;
2512 def : Pat<(store v4f32:$rS, iaddr:$dst), (STXV $rS, memrix16:$dst)>;
2513 def : Pat<(store v4i32:$rS, iaddr:$dst), (STXV $rS, memrix16:$dst)>;
2514 def : Pat<(store v2f64:$rS, iaddr:$dst), (STXV $rS, memrix16:$dst)>;
2515 def : Pat<(store v2i64:$rS, iaddr:$dst), (STXV $rS, memrix16:$dst)>;
2516 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, iaddr:$dst),
2517 (STXV $rS, memrix16:$dst)>;
2518 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, iaddr:$dst),
2519 (STXV $rS, memrix16:$dst)>;
2522 def : Pat<(v2f64 (load xaddr:$src)), (LXVX xaddr:$src)>;
2523 def : Pat<(v2i64 (load xaddr:$src)), (LXVX xaddr:$src)>;
2524 def : Pat<(v4f32 (load xaddr:$src)), (LXVX xaddr:$src)>;
2525 def : Pat<(v4i32 (load xaddr:$src)), (LXVX xaddr:$src)>;
2526 def : Pat<(v4i32 (int_ppc_vsx_lxvw4x xaddr:$src)), (LXVX xaddr:$src)>;
2527 def : Pat<(v2f64 (int_ppc_vsx_lxvd2x xaddr:$src)), (LXVX xaddr:$src)>;
2528 def : Pat<(store v2f64:$rS, xaddr:$dst), (STXVX $rS, xaddr:$dst)>;
2529 def : Pat<(store v2i64:$rS, xaddr:$dst), (STXVX $rS, xaddr:$dst)>;
2530 def : Pat<(store v4f32:$rS, xaddr:$dst), (STXVX $rS, xaddr:$dst)>;
2531 def : Pat<(store v4i32:$rS, xaddr:$dst), (STXVX $rS, xaddr:$dst)>;
2532 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, xaddr:$dst),
2533 (STXVX $rS, xaddr:$dst)>;
2534 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, xaddr:$dst),
2535 (STXVX $rS, xaddr:$dst)>;
2536 def : Pat<(v4i32 (scalar_to_vector (i32 (load xoaddr:$src)))),
2537 (v4i32 (LXVWSX xoaddr:$src))>;
2538 def : Pat<(v4f32 (scalar_to_vector (f32 (load xoaddr:$src)))),
2539 (v4f32 (LXVWSX xoaddr:$src))>;
2540 def : Pat<(v4f32 (scalar_to_vector (f32 (fpround (extloadf32 xoaddr:$src))))),
2541 (v4f32 (LXVWSX xoaddr:$src))>;
2543 // Build vectors from i8 loads
2544 def : Pat<(v16i8 (scalar_to_vector ScalarLoads.Li8)),
2545 (v16i8 (VSPLTBs 7, (LXSIBZX xoaddr:$src)))>;
2546 def : Pat<(v8i16 (scalar_to_vector ScalarLoads.ZELi8)),
2547 (v8i16 (VSPLTHs 3, (LXSIBZX xoaddr:$src)))>;
2548 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.ZELi8)),
2549 (v4i32 (XXSPLTWs (LXSIBZX xoaddr:$src), 1))>;
2550 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.ZELi8i64)),
2551 (v2i64 (XXPERMDIs (LXSIBZX xoaddr:$src), 0))>;
2552 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.SELi8)),
2553 (v4i32 (XXSPLTWs (VEXTSB2Ws (LXSIBZX xoaddr:$src)), 1))>;
2554 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.SELi8i64)),
2555 (v2i64 (XXPERMDIs (VEXTSB2Ds (LXSIBZX xoaddr:$src)), 0))>;
2557 // Build vectors from i16 loads
2558 def : Pat<(v8i16 (scalar_to_vector ScalarLoads.Li16)),
2559 (v8i16 (VSPLTHs 3, (LXSIHZX xoaddr:$src)))>;
2560 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.ZELi16)),
2561 (v4i32 (XXSPLTWs (LXSIHZX xoaddr:$src), 1))>;
2562 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.ZELi16i64)),
2563 (v2i64 (XXPERMDIs (LXSIHZX xoaddr:$src), 0))>;
2564 def : Pat<(v4i32 (scalar_to_vector ScalarLoads.SELi16)),
2565 (v4i32 (XXSPLTWs (VEXTSH2Ws (LXSIHZX xoaddr:$src)), 1))>;
2566 def : Pat<(v2i64 (scalar_to_vector ScalarLoads.SELi16i64)),
2567 (v2i64 (XXPERMDIs (VEXTSH2Ds (LXSIHZX xoaddr:$src)), 0))>;
2569 let Predicates = [IsBigEndian, HasP9Vector] in {
2570 // Scalar stores of i8
2571 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 0)), xoaddr:$dst),
2572 (STXSIBXv (VSLDOI $S, $S, 9), xoaddr:$dst)>;
2573 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 1)), xoaddr:$dst),
2574 (STXSIBXv (VSLDOI $S, $S, 10), xoaddr:$dst)>;
2575 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 2)), xoaddr:$dst),
2576 (STXSIBXv (VSLDOI $S, $S, 11), xoaddr:$dst)>;
2577 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 3)), xoaddr:$dst),
2578 (STXSIBXv (VSLDOI $S, $S, 12), xoaddr:$dst)>;
2579 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 4)), xoaddr:$dst),
2580 (STXSIBXv (VSLDOI $S, $S, 13), xoaddr:$dst)>;
2581 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 5)), xoaddr:$dst),
2582 (STXSIBXv (VSLDOI $S, $S, 14), xoaddr:$dst)>;
2583 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 6)), xoaddr:$dst),
2584 (STXSIBXv (VSLDOI $S, $S, 15), xoaddr:$dst)>;
2585 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 7)), xoaddr:$dst),
2586 (STXSIBXv $S, xoaddr:$dst)>;
2587 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 8)), xoaddr:$dst),
2588 (STXSIBXv (VSLDOI $S, $S, 1), xoaddr:$dst)>;
2589 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 9)), xoaddr:$dst),
2590 (STXSIBXv (VSLDOI $S, $S, 2), xoaddr:$dst)>;
2591 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 10)), xoaddr:$dst),
2592 (STXSIBXv (VSLDOI $S, $S, 3), xoaddr:$dst)>;
2593 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 11)), xoaddr:$dst),
2594 (STXSIBXv (VSLDOI $S, $S, 4), xoaddr:$dst)>;
2595 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 12)), xoaddr:$dst),
2596 (STXSIBXv (VSLDOI $S, $S, 5), xoaddr:$dst)>;
2597 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 13)), xoaddr:$dst),
2598 (STXSIBXv (VSLDOI $S, $S, 6), xoaddr:$dst)>;
2599 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 14)), xoaddr:$dst),
2600 (STXSIBXv (VSLDOI $S, $S, 7), xoaddr:$dst)>;
2601 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 15)), xoaddr:$dst),
2602 (STXSIBXv (VSLDOI $S, $S, 8), xoaddr:$dst)>;
2604 // Scalar stores of i16
2605 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 0)), xoaddr:$dst),
2606 (STXSIHXv (VSLDOI $S, $S, 10), xoaddr:$dst)>;
2607 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 1)), xoaddr:$dst),
2608 (STXSIHXv (VSLDOI $S, $S, 12), xoaddr:$dst)>;
2609 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 2)), xoaddr:$dst),
2610 (STXSIHXv (VSLDOI $S, $S, 14), xoaddr:$dst)>;
2611 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 3)), xoaddr:$dst),
2612 (STXSIHXv $S, xoaddr:$dst)>;
2613 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 4)), xoaddr:$dst),
2614 (STXSIHXv (VSLDOI $S, $S, 2), xoaddr:$dst)>;
2615 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 5)), xoaddr:$dst),
2616 (STXSIHXv (VSLDOI $S, $S, 4), xoaddr:$dst)>;
2617 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 6)), xoaddr:$dst),
2618 (STXSIHXv (VSLDOI $S, $S, 6), xoaddr:$dst)>;
2619 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 7)), xoaddr:$dst),
2620 (STXSIHXv (VSLDOI $S, $S, 8), xoaddr:$dst)>;
2621 } // IsBigEndian, HasP9Vector
2623 let Predicates = [IsLittleEndian, HasP9Vector] in {
2624 // Scalar stores of i8
2625 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 0)), xoaddr:$dst),
2626 (STXSIBXv (VSLDOI $S, $S, 8), xoaddr:$dst)>;
2627 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 1)), xoaddr:$dst),
2628 (STXSIBXv (VSLDOI $S, $S, 7), xoaddr:$dst)>;
2629 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 2)), xoaddr:$dst),
2630 (STXSIBXv (VSLDOI $S, $S, 6), xoaddr:$dst)>;
2631 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 3)), xoaddr:$dst),
2632 (STXSIBXv (VSLDOI $S, $S, 5), xoaddr:$dst)>;
2633 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 4)), xoaddr:$dst),
2634 (STXSIBXv (VSLDOI $S, $S, 4), xoaddr:$dst)>;
2635 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 5)), xoaddr:$dst),
2636 (STXSIBXv (VSLDOI $S, $S, 3), xoaddr:$dst)>;
2637 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 6)), xoaddr:$dst),
2638 (STXSIBXv (VSLDOI $S, $S, 2), xoaddr:$dst)>;
2639 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 7)), xoaddr:$dst),
2640 (STXSIBXv (VSLDOI $S, $S, 1), xoaddr:$dst)>;
2641 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 8)), xoaddr:$dst),
2642 (STXSIBXv $S, xoaddr:$dst)>;
2643 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 9)), xoaddr:$dst),
2644 (STXSIBXv (VSLDOI $S, $S, 15), xoaddr:$dst)>;
2645 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 10)), xoaddr:$dst),
2646 (STXSIBXv (VSLDOI $S, $S, 14), xoaddr:$dst)>;
2647 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 11)), xoaddr:$dst),
2648 (STXSIBXv (VSLDOI $S, $S, 13), xoaddr:$dst)>;
2649 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 12)), xoaddr:$dst),
2650 (STXSIBXv (VSLDOI $S, $S, 12), xoaddr:$dst)>;
2651 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 13)), xoaddr:$dst),
2652 (STXSIBXv (VSLDOI $S, $S, 11), xoaddr:$dst)>;
2653 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 14)), xoaddr:$dst),
2654 (STXSIBXv (VSLDOI $S, $S, 10), xoaddr:$dst)>;
2655 def : Pat<(truncstorei8 (i32 (vector_extract v16i8:$S, 15)), xoaddr:$dst),
2656 (STXSIBXv (VSLDOI $S, $S, 9), xoaddr:$dst)>;
2658 // Scalar stores of i16
2659 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 0)), xoaddr:$dst),
2660 (STXSIHXv (VSLDOI $S, $S, 8), xoaddr:$dst)>;
2661 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 1)), xoaddr:$dst),
2662 (STXSIHXv (VSLDOI $S, $S, 6), xoaddr:$dst)>;
2663 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 2)), xoaddr:$dst),
2664 (STXSIHXv (VSLDOI $S, $S, 4), xoaddr:$dst)>;
2665 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 3)), xoaddr:$dst),
2666 (STXSIHXv (VSLDOI $S, $S, 2), xoaddr:$dst)>;
2667 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 4)), xoaddr:$dst),
2668 (STXSIHXv $S, xoaddr:$dst)>;
2669 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 5)), xoaddr:$dst),
2670 (STXSIHXv (VSLDOI $S, $S, 14), xoaddr:$dst)>;
2671 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 6)), xoaddr:$dst),
2672 (STXSIHXv (VSLDOI $S, $S, 12), xoaddr:$dst)>;
2673 def : Pat<(truncstorei16 (i32 (vector_extract v8i16:$S, 7)), xoaddr:$dst),
2674 (STXSIHXv (VSLDOI $S, $S, 10), xoaddr:$dst)>;
2675 } // IsLittleEndian, HasP9Vector
2678 // Vector sign extensions
2679 def : Pat<(f64 (PPCVexts f64:$A, 1)),
2680 (f64 (COPY_TO_REGCLASS (VEXTSB2Ds $A), VSFRC))>;
2681 def : Pat<(f64 (PPCVexts f64:$A, 2)),
2682 (f64 (COPY_TO_REGCLASS (VEXTSH2Ds $A), VSFRC))>;
2684 let isPseudo = 1 in {
2685 def DFLOADf32 : Pseudo<(outs vssrc:$XT), (ins memrix:$src),
2687 [(set f32:$XT, (load iaddr:$src))]>;
2688 def DFLOADf64 : Pseudo<(outs vsfrc:$XT), (ins memrix:$src),
2690 [(set f64:$XT, (load iaddr:$src))]>;
2691 def DFSTOREf32 : Pseudo<(outs), (ins vssrc:$XT, memrix:$dst),
2693 [(store f32:$XT, iaddr:$dst)]>;
2694 def DFSTOREf64 : Pseudo<(outs), (ins vsfrc:$XT, memrix:$dst),
2696 [(store f64:$XT, iaddr:$dst)]>;
2698 def : Pat<(f64 (extloadf32 iaddr:$src)),
2699 (COPY_TO_REGCLASS (DFLOADf32 iaddr:$src), VSFRC)>;
2700 def : Pat<(f32 (fpround (extloadf32 iaddr:$src))),
2701 (f32 (DFLOADf32 iaddr:$src))>;
2702 } // end HasP9Vector, AddedComplexity
2704 // Integer extend helper dags 32 -> 64
2706 dag A = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32);
2707 dag B = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $B, sub_32);
2708 dag C = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $C, sub_32);
2709 dag D = (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $D, sub_32);
2713 dag A0 = (f32 (fpround (f64 (extractelt v2f64:$A, 0))));
2714 dag A1 = (f32 (fpround (f64 (extractelt v2f64:$A, 1))));
2715 dag B0 = (f32 (fpround (f64 (extractelt v2f64:$B, 0))));
2716 dag B1 = (f32 (fpround (f64 (extractelt v2f64:$B, 1))));
2719 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (extloadf32 xoaddr:$A)))));
2722 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (extloadf32 xoaddr:$A)))));
2725 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (extloadf32 xoaddr:$A)))));
2727 def FltToLongLoadP9 {
2728 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (extloadf32 iaddr:$A)))));
2730 def FltToULongLoad {
2731 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (extloadf32 xoaddr:$A)))));
2733 def FltToULongLoadP9 {
2734 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (extloadf32 iaddr:$A)))));
2737 dag A = (i64 (PPCmfvsr (PPCfctidz (fpextend f32:$A))));
2740 dag A = (i64 (PPCmfvsr (PPCfctiduz (fpextend f32:$A))));
2743 dag A = (i32 (PPCmfvsr (f64 (PPCfctiwz f64:$A))));
2746 dag A = (i32 (PPCmfvsr (f64 (PPCfctiwuz f64:$A))));
2749 dag A = (i64 (PPCmfvsr (f64 (PPCfctidz f64:$A))));
2752 dag A = (i64 (PPCmfvsr (f64 (PPCfctiduz f64:$A))));
2755 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (load xoaddr:$A)))));
2757 def DblToIntLoadP9 {
2758 dag A = (i32 (PPCmfvsr (PPCfctiwz (f64 (load iaddr:$A)))));
2761 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (load xoaddr:$A)))));
2763 def DblToUIntLoadP9 {
2764 dag A = (i32 (PPCmfvsr (PPCfctiwuz (f64 (load iaddr:$A)))));
2767 dag A = (i64 (PPCmfvsr (PPCfctidz (f64 (load xoaddr:$A)))));
2769 def DblToULongLoad {
2770 dag A = (i64 (PPCmfvsr (PPCfctiduz (f64 (load xoaddr:$A)))));
2773 // FP merge dags (for f32 -> v4f32)
2775 dag AC = (XVCVDPSP (XXPERMDI (COPY_TO_REGCLASS $A, VSRC),
2776 (COPY_TO_REGCLASS $C, VSRC), 0));
2777 dag BD = (XVCVDPSP (XXPERMDI (COPY_TO_REGCLASS $B, VSRC),
2778 (COPY_TO_REGCLASS $D, VSRC), 0));
2779 dag ABhToFlt = (XVCVDPSP (XXPERMDI $A, $B, 0));
2780 dag ABlToFlt = (XVCVDPSP (XXPERMDI $A, $B, 3));
2781 dag BAhToFlt = (XVCVDPSP (XXPERMDI $B, $A, 0));
2782 dag BAlToFlt = (XVCVDPSP (XXPERMDI $B, $A, 3));
2785 // Patterns for BUILD_VECTOR nodes.
2786 def NoP9Vector : Predicate<"!PPCSubTarget->hasP9Vector()">;
2787 let AddedComplexity = 400 in {
2789 let Predicates = [HasVSX] in {
2790 // Build vectors of floating point converted to i32.
2791 def : Pat<(v4i32 (build_vector DblToInt.A, DblToInt.A,
2792 DblToInt.A, DblToInt.A)),
2793 (v4i32 (XXSPLTW (COPY_TO_REGCLASS (XSCVDPSXWS $A), VSRC), 1))>;
2794 def : Pat<(v4i32 (build_vector DblToUInt.A, DblToUInt.A,
2795 DblToUInt.A, DblToUInt.A)),
2796 (v4i32 (XXSPLTW (COPY_TO_REGCLASS (XSCVDPUXWS $A), VSRC), 1))>;
2797 def : Pat<(v2i64 (build_vector DblToLong.A, DblToLong.A)),
2798 (v2i64 (XXPERMDI (COPY_TO_REGCLASS (XSCVDPSXDS $A), VSRC),
2799 (COPY_TO_REGCLASS (XSCVDPSXDS $A), VSRC), 0))>;
2800 def : Pat<(v2i64 (build_vector DblToULong.A, DblToULong.A)),
2801 (v2i64 (XXPERMDI (COPY_TO_REGCLASS (XSCVDPUXDS $A), VSRC),
2802 (COPY_TO_REGCLASS (XSCVDPUXDS $A), VSRC), 0))>;
2803 def : Pat<(v4i32 (scalar_to_vector FltToIntLoad.A)),
2804 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
2805 (XSCVDPSXWSs (LXSSPX xoaddr:$A)), VSRC), 1))>;
2806 def : Pat<(v4i32 (scalar_to_vector FltToUIntLoad.A)),
2807 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
2808 (XSCVDPUXWSs (LXSSPX xoaddr:$A)), VSRC), 1))>;
2809 def : Pat<(v4f32 (build_vector f32:$A, f32:$A, f32:$A, f32:$A)),
2810 (v4f32 (XXSPLTW (v4f32 (XSCVDPSPN $A)), 0))>;
2812 // Build vectors of floating point converted to i64.
2813 def : Pat<(v2i64 (build_vector FltToLong.A, FltToLong.A)),
2815 (COPY_TO_REGCLASS (XSCVDPSXDSs $A), VSFRC), 0))>;
2816 def : Pat<(v2i64 (build_vector FltToULong.A, FltToULong.A)),
2818 (COPY_TO_REGCLASS (XSCVDPUXDSs $A), VSFRC), 0))>;
2819 def : Pat<(v2i64 (scalar_to_vector DblToLongLoad.A)),
2820 (v2i64 (XVCVDPSXDS (LXVDSX xoaddr:$A)))>;
2821 def : Pat<(v2i64 (scalar_to_vector DblToULongLoad.A)),
2822 (v2i64 (XVCVDPUXDS (LXVDSX xoaddr:$A)))>;
2825 let Predicates = [HasVSX, NoP9Vector] in {
2826 // Load-and-splat with fp-to-int conversion (using X-Form VSX loads).
2827 def : Pat<(v4i32 (scalar_to_vector DblToIntLoad.A)),
2828 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
2829 (XSCVDPSXWS (LXSDX xoaddr:$A)), VSRC), 1))>;
2830 def : Pat<(v4i32 (scalar_to_vector DblToUIntLoad.A)),
2831 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
2832 (XSCVDPUXWS (LXSDX xoaddr:$A)), VSRC), 1))>;
2833 def : Pat<(v2i64 (scalar_to_vector FltToLongLoad.A)),
2834 (v2i64 (XXPERMDIs (XSCVDPSXDS (COPY_TO_REGCLASS
2835 (LXSSPX xoaddr:$A), VSFRC)), 0))>;
2836 def : Pat<(v2i64 (scalar_to_vector FltToULongLoad.A)),
2837 (v2i64 (XXPERMDIs (XSCVDPUXDS (COPY_TO_REGCLASS
2838 (LXSSPX xoaddr:$A), VSFRC)), 0))>;
2841 // Big endian, available on all targets with VSX
2842 let Predicates = [IsBigEndian, HasVSX] in {
2843 def : Pat<(v2f64 (build_vector f64:$A, f64:$B)),
2845 (COPY_TO_REGCLASS $A, VSRC),
2846 (COPY_TO_REGCLASS $B, VSRC), 0))>;
2848 def : Pat<(v4f32 (build_vector f32:$A, f32:$B, f32:$C, f32:$D)),
2849 (VMRGEW MrgFP.AC, MrgFP.BD)>;
2850 def : Pat<(v4f32 (build_vector DblToFlt.A0, DblToFlt.A1,
2851 DblToFlt.B0, DblToFlt.B1)),
2852 (v4f32 (VMRGEW MrgFP.ABhToFlt, MrgFP.ABlToFlt))>;
2855 let Predicates = [IsLittleEndian, HasVSX] in {
2856 // Little endian, available on all targets with VSX
2857 def : Pat<(v2f64 (build_vector f64:$A, f64:$B)),
2859 (COPY_TO_REGCLASS $B, VSRC),
2860 (COPY_TO_REGCLASS $A, VSRC), 0))>;
2862 def : Pat<(v4f32 (build_vector f32:$D, f32:$C, f32:$B, f32:$A)),
2863 (VMRGEW MrgFP.AC, MrgFP.BD)>;
2864 def : Pat<(v4f32 (build_vector DblToFlt.A0, DblToFlt.A1,
2865 DblToFlt.B0, DblToFlt.B1)),
2866 (v4f32 (VMRGEW MrgFP.BAhToFlt, MrgFP.BAlToFlt))>;
2869 let Predicates = [HasDirectMove] in {
2870 // Endianness-neutral constant splat on P8 and newer targets. The reason
2871 // for this pattern is that on targets with direct moves, we don't expand
2872 // BUILD_VECTOR nodes for v4i32.
2873 def : Pat<(v4i32 (build_vector immSExt5NonZero:$A, immSExt5NonZero:$A,
2874 immSExt5NonZero:$A, immSExt5NonZero:$A)),
2875 (v4i32 (VSPLTISW imm:$A))>;
2878 let Predicates = [IsBigEndian, HasDirectMove, NoP9Vector] in {
2879 // Big endian integer vectors using direct moves.
2880 def : Pat<(v2i64 (build_vector i64:$A, i64:$B)),
2882 (COPY_TO_REGCLASS (MTVSRD $A), VSRC),
2883 (COPY_TO_REGCLASS (MTVSRD $B), VSRC), 0))>;
2884 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
2885 (VMRGOW (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC),
2886 (COPY_TO_REGCLASS (MTVSRWZ $C), VSRC), 0),
2887 (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC),
2888 (COPY_TO_REGCLASS (MTVSRWZ $D), VSRC), 0))>;
2889 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
2890 (XXSPLTW (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 1)>;
2893 let Predicates = [IsLittleEndian, HasDirectMove, NoP9Vector] in {
2894 // Little endian integer vectors using direct moves.
2895 def : Pat<(v2i64 (build_vector i64:$A, i64:$B)),
2897 (COPY_TO_REGCLASS (MTVSRD $B), VSRC),
2898 (COPY_TO_REGCLASS (MTVSRD $A), VSRC), 0))>;
2899 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
2900 (VMRGOW (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $D), VSRC),
2901 (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC), 0),
2902 (XXPERMDI (COPY_TO_REGCLASS (MTVSRWZ $C), VSRC),
2903 (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 0))>;
2904 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
2905 (XXSPLTW (COPY_TO_REGCLASS (MTVSRWZ $A), VSRC), 1)>;
2908 let Predicates = [HasP9Vector] in {
2909 // Endianness-neutral patterns for const splats with ISA 3.0 instructions.
2910 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
2911 (v4i32 (MTVSRWS $A))>;
2912 def : Pat<(v4i32 (build_vector i32:$A, i32:$A, i32:$A, i32:$A)),
2913 (v4i32 (MTVSRWS $A))>;
2914 def : Pat<(v16i8 (build_vector immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
2915 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
2916 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
2917 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
2918 immAnyExt8:$A, immAnyExt8:$A, immAnyExt8:$A,
2920 (v16i8 (COPY_TO_REGCLASS (XXSPLTIB imm:$A), VSRC))>;
2921 def : Pat<(v16i8 immAllOnesV),
2922 (v16i8 (COPY_TO_REGCLASS (XXSPLTIB 255), VSRC))>;
2923 def : Pat<(v8i16 immAllOnesV),
2924 (v8i16 (COPY_TO_REGCLASS (XXSPLTIB 255), VSRC))>;
2925 def : Pat<(v4i32 immAllOnesV),
2926 (v4i32 (XXSPLTIB 255))>;
2927 def : Pat<(v2i64 immAllOnesV),
2928 (v2i64 (XXSPLTIB 255))>;
2929 def : Pat<(v4i32 (scalar_to_vector FltToIntLoad.A)),
2930 (v4i32 (XVCVSPSXWS (LXVWSX xoaddr:$A)))>;
2931 def : Pat<(v4i32 (scalar_to_vector FltToUIntLoad.A)),
2932 (v4i32 (XVCVSPUXWS (LXVWSX xoaddr:$A)))>;
2933 def : Pat<(v4i32 (scalar_to_vector DblToIntLoadP9.A)),
2934 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
2935 (XSCVDPSXWS (DFLOADf64 iaddr:$A)), VSRC), 1))>;
2936 def : Pat<(v4i32 (scalar_to_vector DblToUIntLoadP9.A)),
2937 (v4i32 (XXSPLTW (COPY_TO_REGCLASS
2938 (XSCVDPUXWS (DFLOADf64 iaddr:$A)), VSRC), 1))>;
2939 def : Pat<(v2i64 (scalar_to_vector FltToLongLoadP9.A)),
2940 (v2i64 (XXPERMDIs (XSCVDPSXDS (COPY_TO_REGCLASS
2941 (DFLOADf32 iaddr:$A),
2943 def : Pat<(v2i64 (scalar_to_vector FltToULongLoadP9.A)),
2944 (v2i64 (XXPERMDIs (XSCVDPUXDS (COPY_TO_REGCLASS
2945 (DFLOADf32 iaddr:$A),
2949 let Predicates = [IsISA3_0, HasDirectMove, IsBigEndian] in {
2950 def : Pat<(i64 (extractelt v2i64:$A, 1)),
2951 (i64 (MFVSRLD $A))>;
2952 // Better way to build integer vectors if we have MTVSRDD. Big endian.
2953 def : Pat<(v2i64 (build_vector i64:$rB, i64:$rA)),
2954 (v2i64 (MTVSRDD $rB, $rA))>;
2955 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
2956 (VMRGOW (COPY_TO_REGCLASS (MTVSRDD AnyExts.A, AnyExts.C), VSRC),
2957 (COPY_TO_REGCLASS (MTVSRDD AnyExts.B, AnyExts.D), VSRC))>;
2960 let Predicates = [IsISA3_0, HasDirectMove, IsLittleEndian] in {
2961 def : Pat<(i64 (extractelt v2i64:$A, 0)),
2962 (i64 (MFVSRLD $A))>;
2963 // Better way to build integer vectors if we have MTVSRDD. Little endian.
2964 def : Pat<(v2i64 (build_vector i64:$rA, i64:$rB)),
2965 (v2i64 (MTVSRDD $rB, $rA))>;
2966 def : Pat<(v4i32 (build_vector i32:$A, i32:$B, i32:$C, i32:$D)),
2967 (VMRGOW (COPY_TO_REGCLASS (MTVSRDD AnyExts.D, AnyExts.B), VSRC),
2968 (COPY_TO_REGCLASS (MTVSRDD AnyExts.C, AnyExts.A), VSRC))>;