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]>;
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:$XB, xo)))]>,
92 def HasVSX : Predicate<"PPCSubTarget->hasVSX()">;
93 def IsLittleEndian : Predicate<"PPCSubTarget->isLittleEndian()">;
94 def IsBigEndian : Predicate<"!PPCSubTarget->isLittleEndian()">;
96 let Predicates = [HasVSX] in {
97 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
98 let hasSideEffects = 0 in { // VSX instructions don't have side effects.
101 // Load indexed instructions
103 def LXSDX : XX1Form<31, 588,
104 (outs vsfrc:$XT), (ins memrr:$src),
105 "lxsdx $XT, $src", IIC_LdStLFD,
106 [(set f64:$XT, (load xoaddr:$src))]>;
108 def LXVD2X : XX1Form<31, 844,
109 (outs vsrc:$XT), (ins memrr:$src),
110 "lxvd2x $XT, $src", IIC_LdStLFD,
111 [(set v2f64:$XT, (int_ppc_vsx_lxvd2x xoaddr:$src))]>;
113 def LXVDSX : XX1Form<31, 332,
114 (outs vsrc:$XT), (ins memrr:$src),
115 "lxvdsx $XT, $src", IIC_LdStLFD, []>;
117 def LXVW4X : XX1Form<31, 780,
118 (outs vsrc:$XT), (ins memrr:$src),
119 "lxvw4x $XT, $src", IIC_LdStLFD,
120 [(set v4i32:$XT, (int_ppc_vsx_lxvw4x xoaddr:$src))]>;
123 // Store indexed instructions
124 let mayStore = 1 in {
125 def STXSDX : XX1Form<31, 716,
126 (outs), (ins vsfrc:$XT, memrr:$dst),
127 "stxsdx $XT, $dst", IIC_LdStSTFD,
128 [(store f64:$XT, xoaddr:$dst)]>;
130 def STXVD2X : XX1Form<31, 972,
131 (outs), (ins vsrc:$XT, memrr:$dst),
132 "stxvd2x $XT, $dst", IIC_LdStSTFD,
133 [(store v2f64:$XT, xoaddr:$dst)]>;
135 def STXVW4X : XX1Form<31, 908,
136 (outs), (ins vsrc:$XT, memrr:$dst),
137 "stxvw4x $XT, $dst", IIC_LdStSTFD,
138 [(store v4i32:$XT, xoaddr:$dst)]>;
142 // Add/Mul Instructions
143 let isCommutable = 1 in {
144 def XSADDDP : XX3Form<60, 32,
145 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
146 "xsadddp $XT, $XA, $XB", IIC_VecFP,
147 [(set f64:$XT, (fadd f64:$XA, f64:$XB))]>;
148 def XSMULDP : XX3Form<60, 48,
149 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
150 "xsmuldp $XT, $XA, $XB", IIC_VecFP,
151 [(set f64:$XT, (fmul f64:$XA, f64:$XB))]>;
153 def XVADDDP : XX3Form<60, 96,
154 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
155 "xvadddp $XT, $XA, $XB", IIC_VecFP,
156 [(set v2f64:$XT, (fadd v2f64:$XA, v2f64:$XB))]>;
158 def XVADDSP : XX3Form<60, 64,
159 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
160 "xvaddsp $XT, $XA, $XB", IIC_VecFP,
161 [(set v4f32:$XT, (fadd v4f32:$XA, v4f32:$XB))]>;
163 def XVMULDP : XX3Form<60, 112,
164 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
165 "xvmuldp $XT, $XA, $XB", IIC_VecFP,
166 [(set v2f64:$XT, (fmul v2f64:$XA, v2f64:$XB))]>;
168 def XVMULSP : XX3Form<60, 80,
169 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
170 "xvmulsp $XT, $XA, $XB", IIC_VecFP,
171 [(set v4f32:$XT, (fmul v4f32:$XA, v4f32:$XB))]>;
174 // Subtract Instructions
175 def XSSUBDP : XX3Form<60, 40,
176 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
177 "xssubdp $XT, $XA, $XB", IIC_VecFP,
178 [(set f64:$XT, (fsub f64:$XA, f64:$XB))]>;
180 def XVSUBDP : XX3Form<60, 104,
181 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
182 "xvsubdp $XT, $XA, $XB", IIC_VecFP,
183 [(set v2f64:$XT, (fsub v2f64:$XA, v2f64:$XB))]>;
184 def XVSUBSP : XX3Form<60, 72,
185 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
186 "xvsubsp $XT, $XA, $XB", IIC_VecFP,
187 [(set v4f32:$XT, (fsub v4f32:$XA, v4f32:$XB))]>;
190 let BaseName = "XSMADDADP" in {
191 let isCommutable = 1 in
192 def XSMADDADP : XX3Form<60, 33,
193 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
194 "xsmaddadp $XT, $XA, $XB", IIC_VecFP,
195 [(set f64:$XT, (fma f64:$XA, f64:$XB, f64:$XTi))]>,
196 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
198 let IsVSXFMAAlt = 1 in
199 def XSMADDMDP : XX3Form<60, 41,
200 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
201 "xsmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
202 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
206 let BaseName = "XSMSUBADP" in {
207 let isCommutable = 1 in
208 def XSMSUBADP : XX3Form<60, 49,
209 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
210 "xsmsubadp $XT, $XA, $XB", IIC_VecFP,
211 [(set f64:$XT, (fma f64:$XA, f64:$XB, (fneg f64:$XTi)))]>,
212 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
214 let IsVSXFMAAlt = 1 in
215 def XSMSUBMDP : XX3Form<60, 57,
216 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
217 "xsmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
218 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
222 let BaseName = "XSNMADDADP" in {
223 let isCommutable = 1 in
224 def XSNMADDADP : XX3Form<60, 161,
225 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
226 "xsnmaddadp $XT, $XA, $XB", IIC_VecFP,
227 [(set f64:$XT, (fneg (fma f64:$XA, f64:$XB, f64:$XTi)))]>,
228 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
230 let IsVSXFMAAlt = 1 in
231 def XSNMADDMDP : XX3Form<60, 169,
232 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
233 "xsnmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
234 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
238 let BaseName = "XSNMSUBADP" in {
239 let isCommutable = 1 in
240 def XSNMSUBADP : XX3Form<60, 177,
241 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
242 "xsnmsubadp $XT, $XA, $XB", IIC_VecFP,
243 [(set f64:$XT, (fneg (fma f64:$XA, f64:$XB, (fneg f64:$XTi))))]>,
244 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
246 let IsVSXFMAAlt = 1 in
247 def XSNMSUBMDP : XX3Form<60, 185,
248 (outs vsfrc:$XT), (ins vsfrc:$XTi, vsfrc:$XA, vsfrc:$XB),
249 "xsnmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
250 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
254 let BaseName = "XVMADDADP" in {
255 let isCommutable = 1 in
256 def XVMADDADP : XX3Form<60, 97,
257 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
258 "xvmaddadp $XT, $XA, $XB", IIC_VecFP,
259 [(set v2f64:$XT, (fma v2f64:$XA, v2f64:$XB, v2f64:$XTi))]>,
260 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
262 let IsVSXFMAAlt = 1 in
263 def XVMADDMDP : XX3Form<60, 105,
264 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
265 "xvmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
266 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
270 let BaseName = "XVMADDASP" in {
271 let isCommutable = 1 in
272 def XVMADDASP : XX3Form<60, 65,
273 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
274 "xvmaddasp $XT, $XA, $XB", IIC_VecFP,
275 [(set v4f32:$XT, (fma v4f32:$XA, v4f32:$XB, v4f32:$XTi))]>,
276 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
278 let IsVSXFMAAlt = 1 in
279 def XVMADDMSP : XX3Form<60, 73,
280 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
281 "xvmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
282 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
286 let BaseName = "XVMSUBADP" in {
287 let isCommutable = 1 in
288 def XVMSUBADP : XX3Form<60, 113,
289 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
290 "xvmsubadp $XT, $XA, $XB", IIC_VecFP,
291 [(set v2f64:$XT, (fma v2f64:$XA, v2f64:$XB, (fneg v2f64:$XTi)))]>,
292 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
294 let IsVSXFMAAlt = 1 in
295 def XVMSUBMDP : XX3Form<60, 121,
296 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
297 "xvmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
298 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
302 let BaseName = "XVMSUBASP" in {
303 let isCommutable = 1 in
304 def XVMSUBASP : XX3Form<60, 81,
305 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
306 "xvmsubasp $XT, $XA, $XB", IIC_VecFP,
307 [(set v4f32:$XT, (fma v4f32:$XA, v4f32:$XB, (fneg v4f32:$XTi)))]>,
308 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
310 let IsVSXFMAAlt = 1 in
311 def XVMSUBMSP : XX3Form<60, 89,
312 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
313 "xvmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
314 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
318 let BaseName = "XVNMADDADP" in {
319 let isCommutable = 1 in
320 def XVNMADDADP : XX3Form<60, 225,
321 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
322 "xvnmaddadp $XT, $XA, $XB", IIC_VecFP,
323 [(set v2f64:$XT, (fneg (fma v2f64:$XA, v2f64:$XB, v2f64:$XTi)))]>,
324 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
326 let IsVSXFMAAlt = 1 in
327 def XVNMADDMDP : XX3Form<60, 233,
328 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
329 "xvnmaddmdp $XT, $XA, $XB", IIC_VecFP, []>,
330 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
334 let BaseName = "XVNMADDASP" in {
335 let isCommutable = 1 in
336 def XVNMADDASP : XX3Form<60, 193,
337 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
338 "xvnmaddasp $XT, $XA, $XB", IIC_VecFP,
339 [(set v4f32:$XT, (fneg (fma v4f32:$XA, v4f32:$XB, v4f32:$XTi)))]>,
340 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
342 let IsVSXFMAAlt = 1 in
343 def XVNMADDMSP : XX3Form<60, 201,
344 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
345 "xvnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
346 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
350 let BaseName = "XVNMSUBADP" in {
351 let isCommutable = 1 in
352 def XVNMSUBADP : XX3Form<60, 241,
353 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
354 "xvnmsubadp $XT, $XA, $XB", IIC_VecFP,
355 [(set v2f64:$XT, (fneg (fma v2f64:$XA, v2f64:$XB, (fneg v2f64:$XTi))))]>,
356 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
358 let IsVSXFMAAlt = 1 in
359 def XVNMSUBMDP : XX3Form<60, 249,
360 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
361 "xvnmsubmdp $XT, $XA, $XB", IIC_VecFP, []>,
362 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
366 let BaseName = "XVNMSUBASP" in {
367 let isCommutable = 1 in
368 def XVNMSUBASP : XX3Form<60, 209,
369 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
370 "xvnmsubasp $XT, $XA, $XB", IIC_VecFP,
371 [(set v4f32:$XT, (fneg (fma v4f32:$XA, v4f32:$XB, (fneg v4f32:$XTi))))]>,
372 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
374 let IsVSXFMAAlt = 1 in
375 def XVNMSUBMSP : XX3Form<60, 217,
376 (outs vsrc:$XT), (ins vsrc:$XTi, vsrc:$XA, vsrc:$XB),
377 "xvnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
378 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
382 // Division Instructions
383 def XSDIVDP : XX3Form<60, 56,
384 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
385 "xsdivdp $XT, $XA, $XB", IIC_FPDivD,
386 [(set f64:$XT, (fdiv f64:$XA, f64:$XB))]>;
387 def XSSQRTDP : XX2Form<60, 75,
388 (outs vsfrc:$XT), (ins vsfrc:$XB),
389 "xssqrtdp $XT, $XB", IIC_FPSqrtD,
390 [(set f64:$XT, (fsqrt f64:$XB))]>;
392 def XSREDP : XX2Form<60, 90,
393 (outs vsfrc:$XT), (ins vsfrc:$XB),
394 "xsredp $XT, $XB", IIC_VecFP,
395 [(set f64:$XT, (PPCfre f64:$XB))]>;
396 def XSRSQRTEDP : XX2Form<60, 74,
397 (outs vsfrc:$XT), (ins vsfrc:$XB),
398 "xsrsqrtedp $XT, $XB", IIC_VecFP,
399 [(set f64:$XT, (PPCfrsqrte f64:$XB))]>;
401 def XSTDIVDP : XX3Form_1<60, 61,
402 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
403 "xstdivdp $crD, $XA, $XB", IIC_FPCompare, []>;
404 def XSTSQRTDP : XX2Form_1<60, 106,
405 (outs crrc:$crD), (ins vsfrc:$XB),
406 "xstsqrtdp $crD, $XB", IIC_FPCompare, []>;
408 def XVDIVDP : XX3Form<60, 120,
409 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
410 "xvdivdp $XT, $XA, $XB", IIC_FPDivD,
411 [(set v2f64:$XT, (fdiv v2f64:$XA, v2f64:$XB))]>;
412 def XVDIVSP : XX3Form<60, 88,
413 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
414 "xvdivsp $XT, $XA, $XB", IIC_FPDivS,
415 [(set v4f32:$XT, (fdiv v4f32:$XA, v4f32:$XB))]>;
417 def XVSQRTDP : XX2Form<60, 203,
418 (outs vsrc:$XT), (ins vsrc:$XB),
419 "xvsqrtdp $XT, $XB", IIC_FPSqrtD,
420 [(set v2f64:$XT, (fsqrt v2f64:$XB))]>;
421 def XVSQRTSP : XX2Form<60, 139,
422 (outs vsrc:$XT), (ins vsrc:$XB),
423 "xvsqrtsp $XT, $XB", IIC_FPSqrtS,
424 [(set v4f32:$XT, (fsqrt v4f32:$XB))]>;
426 def XVTDIVDP : XX3Form_1<60, 125,
427 (outs crrc:$crD), (ins vsrc:$XA, vsrc:$XB),
428 "xvtdivdp $crD, $XA, $XB", IIC_FPCompare, []>;
429 def XVTDIVSP : XX3Form_1<60, 93,
430 (outs crrc:$crD), (ins vsrc:$XA, vsrc:$XB),
431 "xvtdivsp $crD, $XA, $XB", IIC_FPCompare, []>;
433 def XVTSQRTDP : XX2Form_1<60, 234,
434 (outs crrc:$crD), (ins vsrc:$XB),
435 "xvtsqrtdp $crD, $XB", IIC_FPCompare, []>;
436 def XVTSQRTSP : XX2Form_1<60, 170,
437 (outs crrc:$crD), (ins vsrc:$XB),
438 "xvtsqrtsp $crD, $XB", IIC_FPCompare, []>;
440 def XVREDP : XX2Form<60, 218,
441 (outs vsrc:$XT), (ins vsrc:$XB),
442 "xvredp $XT, $XB", IIC_VecFP,
443 [(set v2f64:$XT, (PPCfre v2f64:$XB))]>;
444 def XVRESP : XX2Form<60, 154,
445 (outs vsrc:$XT), (ins vsrc:$XB),
446 "xvresp $XT, $XB", IIC_VecFP,
447 [(set v4f32:$XT, (PPCfre v4f32:$XB))]>;
449 def XVRSQRTEDP : XX2Form<60, 202,
450 (outs vsrc:$XT), (ins vsrc:$XB),
451 "xvrsqrtedp $XT, $XB", IIC_VecFP,
452 [(set v2f64:$XT, (PPCfrsqrte v2f64:$XB))]>;
453 def XVRSQRTESP : XX2Form<60, 138,
454 (outs vsrc:$XT), (ins vsrc:$XB),
455 "xvrsqrtesp $XT, $XB", IIC_VecFP,
456 [(set v4f32:$XT, (PPCfrsqrte v4f32:$XB))]>;
458 // Compare Instructions
459 def XSCMPODP : XX3Form_1<60, 43,
460 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
461 "xscmpodp $crD, $XA, $XB", IIC_FPCompare, []>;
462 def XSCMPUDP : XX3Form_1<60, 35,
463 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
464 "xscmpudp $crD, $XA, $XB", IIC_FPCompare, []>;
466 defm XVCMPEQDP : XX3Form_Rcr<60, 99,
467 "xvcmpeqdp", "$XT, $XA, $XB", IIC_VecFPCompare,
468 int_ppc_vsx_xvcmpeqdp, v2i64, v2f64>;
469 defm XVCMPEQSP : XX3Form_Rcr<60, 67,
470 "xvcmpeqsp", "$XT, $XA, $XB", IIC_VecFPCompare,
471 int_ppc_vsx_xvcmpeqsp, v4i32, v4f32>;
472 defm XVCMPGEDP : XX3Form_Rcr<60, 115,
473 "xvcmpgedp", "$XT, $XA, $XB", IIC_VecFPCompare,
474 int_ppc_vsx_xvcmpgedp, v2i64, v2f64>;
475 defm XVCMPGESP : XX3Form_Rcr<60, 83,
476 "xvcmpgesp", "$XT, $XA, $XB", IIC_VecFPCompare,
477 int_ppc_vsx_xvcmpgesp, v4i32, v4f32>;
478 defm XVCMPGTDP : XX3Form_Rcr<60, 107,
479 "xvcmpgtdp", "$XT, $XA, $XB", IIC_VecFPCompare,
480 int_ppc_vsx_xvcmpgtdp, v2i64, v2f64>;
481 defm XVCMPGTSP : XX3Form_Rcr<60, 75,
482 "xvcmpgtsp", "$XT, $XA, $XB", IIC_VecFPCompare,
483 int_ppc_vsx_xvcmpgtsp, v4i32, v4f32>;
486 def XSABSDP : XX2Form<60, 345,
487 (outs vsfrc:$XT), (ins vsfrc:$XB),
488 "xsabsdp $XT, $XB", IIC_VecFP,
489 [(set f64:$XT, (fabs f64:$XB))]>;
490 def XSNABSDP : XX2Form<60, 361,
491 (outs vsfrc:$XT), (ins vsfrc:$XB),
492 "xsnabsdp $XT, $XB", IIC_VecFP,
493 [(set f64:$XT, (fneg (fabs f64:$XB)))]>;
494 def XSNEGDP : XX2Form<60, 377,
495 (outs vsfrc:$XT), (ins vsfrc:$XB),
496 "xsnegdp $XT, $XB", IIC_VecFP,
497 [(set f64:$XT, (fneg f64:$XB))]>;
498 def XSCPSGNDP : XX3Form<60, 176,
499 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
500 "xscpsgndp $XT, $XA, $XB", IIC_VecFP,
501 [(set f64:$XT, (fcopysign f64:$XB, f64:$XA))]>;
503 def XVABSDP : XX2Form<60, 473,
504 (outs vsrc:$XT), (ins vsrc:$XB),
505 "xvabsdp $XT, $XB", IIC_VecFP,
506 [(set v2f64:$XT, (fabs v2f64:$XB))]>;
508 def XVABSSP : XX2Form<60, 409,
509 (outs vsrc:$XT), (ins vsrc:$XB),
510 "xvabssp $XT, $XB", IIC_VecFP,
511 [(set v4f32:$XT, (fabs v4f32:$XB))]>;
513 def XVCPSGNDP : XX3Form<60, 240,
514 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
515 "xvcpsgndp $XT, $XA, $XB", IIC_VecFP,
516 [(set v2f64:$XT, (fcopysign v2f64:$XB, v2f64:$XA))]>;
517 def XVCPSGNSP : XX3Form<60, 208,
518 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
519 "xvcpsgnsp $XT, $XA, $XB", IIC_VecFP,
520 [(set v4f32:$XT, (fcopysign v4f32:$XB, v4f32:$XA))]>;
522 def XVNABSDP : XX2Form<60, 489,
523 (outs vsrc:$XT), (ins vsrc:$XB),
524 "xvnabsdp $XT, $XB", IIC_VecFP,
525 [(set v2f64:$XT, (fneg (fabs v2f64:$XB)))]>;
526 def XVNABSSP : XX2Form<60, 425,
527 (outs vsrc:$XT), (ins vsrc:$XB),
528 "xvnabssp $XT, $XB", IIC_VecFP,
529 [(set v4f32:$XT, (fneg (fabs v4f32:$XB)))]>;
531 def XVNEGDP : XX2Form<60, 505,
532 (outs vsrc:$XT), (ins vsrc:$XB),
533 "xvnegdp $XT, $XB", IIC_VecFP,
534 [(set v2f64:$XT, (fneg v2f64:$XB))]>;
535 def XVNEGSP : XX2Form<60, 441,
536 (outs vsrc:$XT), (ins vsrc:$XB),
537 "xvnegsp $XT, $XB", IIC_VecFP,
538 [(set v4f32:$XT, (fneg v4f32:$XB))]>;
540 // Conversion Instructions
541 def XSCVDPSP : XX2Form<60, 265,
542 (outs vsfrc:$XT), (ins vsfrc:$XB),
543 "xscvdpsp $XT, $XB", IIC_VecFP, []>;
544 def XSCVDPSXDS : XX2Form<60, 344,
545 (outs vsfrc:$XT), (ins vsfrc:$XB),
546 "xscvdpsxds $XT, $XB", IIC_VecFP,
547 [(set f64:$XT, (PPCfctidz f64:$XB))]>;
548 def XSCVDPSXWS : XX2Form<60, 88,
549 (outs vsfrc:$XT), (ins vsfrc:$XB),
550 "xscvdpsxws $XT, $XB", IIC_VecFP,
551 [(set f64:$XT, (PPCfctiwz f64:$XB))]>;
552 def XSCVDPUXDS : XX2Form<60, 328,
553 (outs vsfrc:$XT), (ins vsfrc:$XB),
554 "xscvdpuxds $XT, $XB", IIC_VecFP,
555 [(set f64:$XT, (PPCfctiduz f64:$XB))]>;
556 def XSCVDPUXWS : XX2Form<60, 72,
557 (outs vsfrc:$XT), (ins vsfrc:$XB),
558 "xscvdpuxws $XT, $XB", IIC_VecFP,
559 [(set f64:$XT, (PPCfctiwuz f64:$XB))]>;
560 def XSCVSPDP : XX2Form<60, 329,
561 (outs vsfrc:$XT), (ins vsfrc:$XB),
562 "xscvspdp $XT, $XB", IIC_VecFP, []>;
563 def XSCVSXDDP : XX2Form<60, 376,
564 (outs vsfrc:$XT), (ins vsfrc:$XB),
565 "xscvsxddp $XT, $XB", IIC_VecFP,
566 [(set f64:$XT, (PPCfcfid f64:$XB))]>;
567 def XSCVUXDDP : XX2Form<60, 360,
568 (outs vsfrc:$XT), (ins vsfrc:$XB),
569 "xscvuxddp $XT, $XB", IIC_VecFP,
570 [(set f64:$XT, (PPCfcfidu f64:$XB))]>;
572 def XVCVDPSP : XX2Form<60, 393,
573 (outs vsrc:$XT), (ins vsrc:$XB),
574 "xvcvdpsp $XT, $XB", IIC_VecFP, []>;
575 def XVCVDPSXDS : XX2Form<60, 472,
576 (outs vsrc:$XT), (ins vsrc:$XB),
577 "xvcvdpsxds $XT, $XB", IIC_VecFP,
578 [(set v2i64:$XT, (fp_to_sint v2f64:$XB))]>;
579 def XVCVDPSXWS : XX2Form<60, 216,
580 (outs vsrc:$XT), (ins vsrc:$XB),
581 "xvcvdpsxws $XT, $XB", IIC_VecFP, []>;
582 def XVCVDPUXDS : XX2Form<60, 456,
583 (outs vsrc:$XT), (ins vsrc:$XB),
584 "xvcvdpuxds $XT, $XB", IIC_VecFP,
585 [(set v2i64:$XT, (fp_to_uint v2f64:$XB))]>;
586 def XVCVDPUXWS : XX2Form<60, 200,
587 (outs vsrc:$XT), (ins vsrc:$XB),
588 "xvcvdpuxws $XT, $XB", IIC_VecFP, []>;
590 def XVCVSPDP : XX2Form<60, 457,
591 (outs vsrc:$XT), (ins vsrc:$XB),
592 "xvcvspdp $XT, $XB", IIC_VecFP, []>;
593 def XVCVSPSXDS : XX2Form<60, 408,
594 (outs vsrc:$XT), (ins vsrc:$XB),
595 "xvcvspsxds $XT, $XB", IIC_VecFP, []>;
596 def XVCVSPSXWS : XX2Form<60, 152,
597 (outs vsrc:$XT), (ins vsrc:$XB),
598 "xvcvspsxws $XT, $XB", IIC_VecFP, []>;
599 def XVCVSPUXDS : XX2Form<60, 392,
600 (outs vsrc:$XT), (ins vsrc:$XB),
601 "xvcvspuxds $XT, $XB", IIC_VecFP, []>;
602 def XVCVSPUXWS : XX2Form<60, 136,
603 (outs vsrc:$XT), (ins vsrc:$XB),
604 "xvcvspuxws $XT, $XB", IIC_VecFP, []>;
605 def XVCVSXDDP : XX2Form<60, 504,
606 (outs vsrc:$XT), (ins vsrc:$XB),
607 "xvcvsxddp $XT, $XB", IIC_VecFP,
608 [(set v2f64:$XT, (sint_to_fp v2i64:$XB))]>;
609 def XVCVSXDSP : XX2Form<60, 440,
610 (outs vsrc:$XT), (ins vsrc:$XB),
611 "xvcvsxdsp $XT, $XB", IIC_VecFP, []>;
612 def XVCVSXWDP : XX2Form<60, 248,
613 (outs vsrc:$XT), (ins vsrc:$XB),
614 "xvcvsxwdp $XT, $XB", IIC_VecFP, []>;
615 def XVCVSXWSP : XX2Form<60, 184,
616 (outs vsrc:$XT), (ins vsrc:$XB),
617 "xvcvsxwsp $XT, $XB", IIC_VecFP,
618 [(set v4f32:$XT, (sint_to_fp v4i32:$XB))]>;
619 def XVCVUXDDP : XX2Form<60, 488,
620 (outs vsrc:$XT), (ins vsrc:$XB),
621 "xvcvuxddp $XT, $XB", IIC_VecFP,
622 [(set v2f64:$XT, (uint_to_fp v2i64:$XB))]>;
623 def XVCVUXDSP : XX2Form<60, 424,
624 (outs vsrc:$XT), (ins vsrc:$XB),
625 "xvcvuxdsp $XT, $XB", IIC_VecFP, []>;
626 def XVCVUXWDP : XX2Form<60, 232,
627 (outs vsrc:$XT), (ins vsrc:$XB),
628 "xvcvuxwdp $XT, $XB", IIC_VecFP, []>;
629 def XVCVUXWSP : XX2Form<60, 168,
630 (outs vsrc:$XT), (ins vsrc:$XB),
631 "xvcvuxwsp $XT, $XB", IIC_VecFP, []>;
633 // Rounding Instructions
634 def XSRDPI : XX2Form<60, 73,
635 (outs vsfrc:$XT), (ins vsfrc:$XB),
636 "xsrdpi $XT, $XB", IIC_VecFP,
637 [(set f64:$XT, (frnd f64:$XB))]>;
638 def XSRDPIC : XX2Form<60, 107,
639 (outs vsfrc:$XT), (ins vsfrc:$XB),
640 "xsrdpic $XT, $XB", IIC_VecFP,
641 [(set f64:$XT, (fnearbyint f64:$XB))]>;
642 def XSRDPIM : XX2Form<60, 121,
643 (outs vsfrc:$XT), (ins vsfrc:$XB),
644 "xsrdpim $XT, $XB", IIC_VecFP,
645 [(set f64:$XT, (ffloor f64:$XB))]>;
646 def XSRDPIP : XX2Form<60, 105,
647 (outs vsfrc:$XT), (ins vsfrc:$XB),
648 "xsrdpip $XT, $XB", IIC_VecFP,
649 [(set f64:$XT, (fceil f64:$XB))]>;
650 def XSRDPIZ : XX2Form<60, 89,
651 (outs vsfrc:$XT), (ins vsfrc:$XB),
652 "xsrdpiz $XT, $XB", IIC_VecFP,
653 [(set f64:$XT, (ftrunc f64:$XB))]>;
655 def XVRDPI : XX2Form<60, 201,
656 (outs vsrc:$XT), (ins vsrc:$XB),
657 "xvrdpi $XT, $XB", IIC_VecFP,
658 [(set v2f64:$XT, (frnd v2f64:$XB))]>;
659 def XVRDPIC : XX2Form<60, 235,
660 (outs vsrc:$XT), (ins vsrc:$XB),
661 "xvrdpic $XT, $XB", IIC_VecFP,
662 [(set v2f64:$XT, (fnearbyint v2f64:$XB))]>;
663 def XVRDPIM : XX2Form<60, 249,
664 (outs vsrc:$XT), (ins vsrc:$XB),
665 "xvrdpim $XT, $XB", IIC_VecFP,
666 [(set v2f64:$XT, (ffloor v2f64:$XB))]>;
667 def XVRDPIP : XX2Form<60, 233,
668 (outs vsrc:$XT), (ins vsrc:$XB),
669 "xvrdpip $XT, $XB", IIC_VecFP,
670 [(set v2f64:$XT, (fceil v2f64:$XB))]>;
671 def XVRDPIZ : XX2Form<60, 217,
672 (outs vsrc:$XT), (ins vsrc:$XB),
673 "xvrdpiz $XT, $XB", IIC_VecFP,
674 [(set v2f64:$XT, (ftrunc v2f64:$XB))]>;
676 def XVRSPI : XX2Form<60, 137,
677 (outs vsrc:$XT), (ins vsrc:$XB),
678 "xvrspi $XT, $XB", IIC_VecFP,
679 [(set v4f32:$XT, (frnd v4f32:$XB))]>;
680 def XVRSPIC : XX2Form<60, 171,
681 (outs vsrc:$XT), (ins vsrc:$XB),
682 "xvrspic $XT, $XB", IIC_VecFP,
683 [(set v4f32:$XT, (fnearbyint v4f32:$XB))]>;
684 def XVRSPIM : XX2Form<60, 185,
685 (outs vsrc:$XT), (ins vsrc:$XB),
686 "xvrspim $XT, $XB", IIC_VecFP,
687 [(set v4f32:$XT, (ffloor v4f32:$XB))]>;
688 def XVRSPIP : XX2Form<60, 169,
689 (outs vsrc:$XT), (ins vsrc:$XB),
690 "xvrspip $XT, $XB", IIC_VecFP,
691 [(set v4f32:$XT, (fceil v4f32:$XB))]>;
692 def XVRSPIZ : XX2Form<60, 153,
693 (outs vsrc:$XT), (ins vsrc:$XB),
694 "xvrspiz $XT, $XB", IIC_VecFP,
695 [(set v4f32:$XT, (ftrunc v4f32:$XB))]>;
697 // Max/Min Instructions
698 let isCommutable = 1 in {
699 def XSMAXDP : XX3Form<60, 160,
700 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
701 "xsmaxdp $XT, $XA, $XB", IIC_VecFP,
703 (int_ppc_vsx_xsmaxdp vsfrc:$XA, vsfrc:$XB))]>;
704 def XSMINDP : XX3Form<60, 168,
705 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
706 "xsmindp $XT, $XA, $XB", IIC_VecFP,
708 (int_ppc_vsx_xsmindp vsfrc:$XA, vsfrc:$XB))]>;
710 def XVMAXDP : XX3Form<60, 224,
711 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
712 "xvmaxdp $XT, $XA, $XB", IIC_VecFP,
714 (int_ppc_vsx_xvmaxdp vsrc:$XA, vsrc:$XB))]>;
715 def XVMINDP : XX3Form<60, 232,
716 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
717 "xvmindp $XT, $XA, $XB", IIC_VecFP,
719 (int_ppc_vsx_xvmindp vsrc:$XA, vsrc:$XB))]>;
721 def XVMAXSP : XX3Form<60, 192,
722 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
723 "xvmaxsp $XT, $XA, $XB", IIC_VecFP,
725 (int_ppc_vsx_xvmaxsp vsrc:$XA, vsrc:$XB))]>;
726 def XVMINSP : XX3Form<60, 200,
727 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
728 "xvminsp $XT, $XA, $XB", IIC_VecFP,
730 (int_ppc_vsx_xvminsp vsrc:$XA, vsrc:$XB))]>;
734 // Logical Instructions
735 let isCommutable = 1 in
736 def XXLAND : XX3Form<60, 130,
737 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
738 "xxland $XT, $XA, $XB", IIC_VecGeneral,
739 [(set v4i32:$XT, (and v4i32:$XA, v4i32:$XB))]>;
740 def XXLANDC : XX3Form<60, 138,
741 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
742 "xxlandc $XT, $XA, $XB", IIC_VecGeneral,
743 [(set v4i32:$XT, (and v4i32:$XA,
744 (vnot_ppc v4i32:$XB)))]>;
745 let isCommutable = 1 in {
746 def XXLNOR : XX3Form<60, 162,
747 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
748 "xxlnor $XT, $XA, $XB", IIC_VecGeneral,
749 [(set v4i32:$XT, (vnot_ppc (or v4i32:$XA,
751 def XXLOR : XX3Form<60, 146,
752 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
753 "xxlor $XT, $XA, $XB", IIC_VecGeneral,
754 [(set v4i32:$XT, (or v4i32:$XA, v4i32:$XB))]>;
755 let isCodeGenOnly = 1 in
756 def XXLORf: XX3Form<60, 146,
757 (outs vsfrc:$XT), (ins vsfrc:$XA, vsfrc:$XB),
758 "xxlor $XT, $XA, $XB", IIC_VecGeneral, []>;
759 def XXLXOR : XX3Form<60, 154,
760 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
761 "xxlxor $XT, $XA, $XB", IIC_VecGeneral,
762 [(set v4i32:$XT, (xor v4i32:$XA, v4i32:$XB))]>;
765 // Permutation Instructions
766 def XXMRGHW : XX3Form<60, 18,
767 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
768 "xxmrghw $XT, $XA, $XB", IIC_VecPerm, []>;
769 def XXMRGLW : XX3Form<60, 50,
770 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
771 "xxmrglw $XT, $XA, $XB", IIC_VecPerm, []>;
773 def XXPERMDI : XX3Form_2<60, 10,
774 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, u2imm:$DM),
775 "xxpermdi $XT, $XA, $XB, $DM", IIC_VecPerm, []>;
776 def XXSEL : XX4Form<60, 3,
777 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, vsrc:$XC),
778 "xxsel $XT, $XA, $XB, $XC", IIC_VecPerm, []>;
780 def XXSLDWI : XX3Form_2<60, 2,
781 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB, u2imm:$SHW),
782 "xxsldwi $XT, $XA, $XB, $SHW", IIC_VecPerm,
783 [(set v4i32:$XT, (PPCvecshl v4i32:$XA, v4i32:$XB,
784 imm32SExt16:$SHW))]>;
785 def XXSPLTW : XX2Form_2<60, 164,
786 (outs vsrc:$XT), (ins vsrc:$XB, u2imm:$UIM),
787 "xxspltw $XT, $XB, $UIM", IIC_VecPerm,
789 (PPCxxsplt v4i32:$XB, imm32SExt16:$UIM))]>;
792 // SELECT_CC_* - Used to implement the SELECT_CC DAG operation. Expanded after
793 // instruction selection into a branch sequence.
794 let usesCustomInserter = 1, // Expanded after instruction selection.
795 PPC970_Single = 1 in {
797 def SELECT_CC_VSRC: Pseudo<(outs vsrc:$dst),
798 (ins crrc:$cond, vsrc:$T, vsrc:$F, i32imm:$BROPC),
801 def SELECT_VSRC: Pseudo<(outs vsrc:$dst),
802 (ins crbitrc:$cond, vsrc:$T, vsrc:$F),
805 (select i1:$cond, v2f64:$T, v2f64:$F))]>;
806 def SELECT_CC_VSFRC: Pseudo<(outs f8rc:$dst),
807 (ins crrc:$cond, f8rc:$T, f8rc:$F,
808 i32imm:$BROPC), "#SELECT_CC_VSFRC",
810 def SELECT_VSFRC: Pseudo<(outs f8rc:$dst),
811 (ins crbitrc:$cond, f8rc:$T, f8rc:$F),
814 (select i1:$cond, f64:$T, f64:$F))]>;
815 def SELECT_CC_VSSRC: Pseudo<(outs f4rc:$dst),
816 (ins crrc:$cond, f4rc:$T, f4rc:$F,
817 i32imm:$BROPC), "#SELECT_CC_VSSRC",
819 def SELECT_VSSRC: Pseudo<(outs f4rc:$dst),
820 (ins crbitrc:$cond, f4rc:$T, f4rc:$F),
823 (select i1:$cond, f32:$T, f32:$F))]>;
824 } // usesCustomInserter
827 def : InstAlias<"xvmovdp $XT, $XB",
828 (XVCPSGNDP vsrc:$XT, vsrc:$XB, vsrc:$XB)>;
829 def : InstAlias<"xvmovsp $XT, $XB",
830 (XVCPSGNSP vsrc:$XT, vsrc:$XB, vsrc:$XB)>;
832 def : InstAlias<"xxspltd $XT, $XB, 0",
833 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 0)>;
834 def : InstAlias<"xxspltd $XT, $XB, 1",
835 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 3)>;
836 def : InstAlias<"xxmrghd $XT, $XA, $XB",
837 (XXPERMDI vsrc:$XT, vsrc:$XA, vsrc:$XB, 0)>;
838 def : InstAlias<"xxmrgld $XT, $XA, $XB",
839 (XXPERMDI vsrc:$XT, vsrc:$XA, vsrc:$XB, 3)>;
840 def : InstAlias<"xxswapd $XT, $XB",
841 (XXPERMDI vsrc:$XT, vsrc:$XB, vsrc:$XB, 2)>;
843 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
845 let Predicates = [IsBigEndian] in {
846 def : Pat<(v2f64 (scalar_to_vector f64:$A)),
847 (v2f64 (SUBREG_TO_REG (i64 1), $A, sub_64))>;
849 def : Pat<(f64 (extractelt v2f64:$S, 0)),
850 (f64 (EXTRACT_SUBREG $S, sub_64))>;
851 def : Pat<(f64 (extractelt v2f64:$S, 1)),
852 (f64 (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64))>;
855 let Predicates = [IsLittleEndian] in {
856 def : Pat<(v2f64 (scalar_to_vector f64:$A)),
857 (v2f64 (XXPERMDI (SUBREG_TO_REG (i64 1), $A, sub_64),
858 (SUBREG_TO_REG (i64 1), $A, sub_64), 0))>;
860 def : Pat<(f64 (extractelt v2f64:$S, 0)),
861 (f64 (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64))>;
862 def : Pat<(f64 (extractelt v2f64:$S, 1)),
863 (f64 (EXTRACT_SUBREG $S, sub_64))>;
866 // Additional fnmsub patterns: -a*c + b == -(a*c - b)
867 def : Pat<(fma (fneg f64:$A), f64:$C, f64:$B),
868 (XSNMSUBADP $B, $C, $A)>;
869 def : Pat<(fma f64:$A, (fneg f64:$C), f64:$B),
870 (XSNMSUBADP $B, $C, $A)>;
872 def : Pat<(fma (fneg v2f64:$A), v2f64:$C, v2f64:$B),
873 (XVNMSUBADP $B, $C, $A)>;
874 def : Pat<(fma v2f64:$A, (fneg v2f64:$C), v2f64:$B),
875 (XVNMSUBADP $B, $C, $A)>;
877 def : Pat<(fma (fneg v4f32:$A), v4f32:$C, v4f32:$B),
878 (XVNMSUBASP $B, $C, $A)>;
879 def : Pat<(fma v4f32:$A, (fneg v4f32:$C), v4f32:$B),
880 (XVNMSUBASP $B, $C, $A)>;
882 def : Pat<(v2f64 (bitconvert v4f32:$A)),
883 (COPY_TO_REGCLASS $A, VSRC)>;
884 def : Pat<(v2f64 (bitconvert v4i32:$A)),
885 (COPY_TO_REGCLASS $A, VSRC)>;
886 def : Pat<(v2f64 (bitconvert v8i16:$A)),
887 (COPY_TO_REGCLASS $A, VSRC)>;
888 def : Pat<(v2f64 (bitconvert v16i8:$A)),
889 (COPY_TO_REGCLASS $A, VSRC)>;
891 def : Pat<(v4f32 (bitconvert v2f64:$A)),
892 (COPY_TO_REGCLASS $A, VRRC)>;
893 def : Pat<(v4i32 (bitconvert v2f64:$A)),
894 (COPY_TO_REGCLASS $A, VRRC)>;
895 def : Pat<(v8i16 (bitconvert v2f64:$A)),
896 (COPY_TO_REGCLASS $A, VRRC)>;
897 def : Pat<(v16i8 (bitconvert v2f64:$A)),
898 (COPY_TO_REGCLASS $A, VRRC)>;
900 def : Pat<(v2i64 (bitconvert v4f32:$A)),
901 (COPY_TO_REGCLASS $A, VSRC)>;
902 def : Pat<(v2i64 (bitconvert v4i32:$A)),
903 (COPY_TO_REGCLASS $A, VSRC)>;
904 def : Pat<(v2i64 (bitconvert v8i16:$A)),
905 (COPY_TO_REGCLASS $A, VSRC)>;
906 def : Pat<(v2i64 (bitconvert v16i8:$A)),
907 (COPY_TO_REGCLASS $A, VSRC)>;
909 def : Pat<(v4f32 (bitconvert v2i64:$A)),
910 (COPY_TO_REGCLASS $A, VRRC)>;
911 def : Pat<(v4i32 (bitconvert v2i64:$A)),
912 (COPY_TO_REGCLASS $A, VRRC)>;
913 def : Pat<(v8i16 (bitconvert v2i64:$A)),
914 (COPY_TO_REGCLASS $A, VRRC)>;
915 def : Pat<(v16i8 (bitconvert v2i64:$A)),
916 (COPY_TO_REGCLASS $A, VRRC)>;
918 def : Pat<(v2f64 (bitconvert v2i64:$A)),
919 (COPY_TO_REGCLASS $A, VRRC)>;
920 def : Pat<(v2i64 (bitconvert v2f64:$A)),
921 (COPY_TO_REGCLASS $A, VRRC)>;
923 def : Pat<(v2f64 (bitconvert v1i128:$A)),
924 (COPY_TO_REGCLASS $A, VRRC)>;
925 def : Pat<(v1i128 (bitconvert v2f64:$A)),
926 (COPY_TO_REGCLASS $A, VRRC)>;
928 // sign extension patterns
929 // To extend "in place" from v2i32 to v2i64, we have input data like:
930 // | undef | i32 | undef | i32 |
931 // but xvcvsxwdp expects the input in big-Endian format:
932 // | i32 | undef | i32 | undef |
933 // so we need to shift everything to the left by one i32 (word) before
935 def : Pat<(sext_inreg v2i64:$C, v2i32),
936 (XVCVDPSXDS (XVCVSXWDP (XXSLDWI $C, $C, 1)))>;
937 def : Pat<(v2f64 (sint_to_fp (sext_inreg v2i64:$C, v2i32))),
938 (XVCVSXWDP (XXSLDWI $C, $C, 1))>;
940 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 0)),
941 (v2f64 (XVCVSXWDP (v2i64 (XXMRGHW $C, $C))))>;
942 def : Pat<(v2f64 (PPCsvec2fp v4i32:$C, 1)),
943 (v2f64 (XVCVSXWDP (v2i64 (XXMRGLW $C, $C))))>;
945 def : Pat<(v2f64 (PPCuvec2fp v4i32:$C, 0)),
946 (v2f64 (XVCVUXWDP (v2i64 (XXMRGHW $C, $C))))>;
947 def : Pat<(v2f64 (PPCuvec2fp v4i32:$C, 1)),
948 (v2f64 (XVCVUXWDP (v2i64 (XXMRGLW $C, $C))))>;
951 def : Pat<(v2f64 (load xoaddr:$src)), (LXVD2X xoaddr:$src)>;
952 def : Pat<(v2i64 (load xoaddr:$src)), (LXVD2X xoaddr:$src)>;
953 def : Pat<(v4i32 (load xoaddr:$src)), (LXVW4X xoaddr:$src)>;
954 def : Pat<(v2f64 (PPClxvd2x xoaddr:$src)), (LXVD2X xoaddr:$src)>;
957 def : Pat<(int_ppc_vsx_stxvd2x v2f64:$rS, xoaddr:$dst),
958 (STXVD2X $rS, xoaddr:$dst)>;
959 def : Pat<(store v2i64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
960 def : Pat<(int_ppc_vsx_stxvw4x v4i32:$rS, xoaddr:$dst),
961 (STXVW4X $rS, xoaddr:$dst)>;
962 def : Pat<(PPCstxvd2x v2f64:$rS, xoaddr:$dst), (STXVD2X $rS, xoaddr:$dst)>;
965 def : Pat<(v2f64 (PPCxxswapd v2f64:$src)), (XXPERMDI $src, $src, 2)>;
966 def : Pat<(v2i64 (PPCxxswapd v2i64:$src)), (XXPERMDI $src, $src, 2)>;
967 def : Pat<(v4f32 (PPCxxswapd v4f32:$src)), (XXPERMDI $src, $src, 2)>;
968 def : Pat<(v4i32 (PPCxxswapd v4i32:$src)), (XXPERMDI $src, $src, 2)>;
969 def : Pat<(v2f64 (PPCswapNoChain v2f64:$src)), (XXPERMDI $src, $src, 2)>;
972 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETLT)),
973 (SELECT_VSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
974 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETULT)),
975 (SELECT_VSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
976 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETLE)),
977 (SELECT_VSRC (CRORC $lhs, $rhs), $tval, $fval)>;
978 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETULE)),
979 (SELECT_VSRC (CRORC $rhs, $lhs), $tval, $fval)>;
980 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETEQ)),
981 (SELECT_VSRC (CREQV $lhs, $rhs), $tval, $fval)>;
982 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETGE)),
983 (SELECT_VSRC (CRORC $rhs, $lhs), $tval, $fval)>;
984 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETUGE)),
985 (SELECT_VSRC (CRORC $lhs, $rhs), $tval, $fval)>;
986 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETGT)),
987 (SELECT_VSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
988 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETUGT)),
989 (SELECT_VSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
990 def : Pat<(v2f64 (selectcc i1:$lhs, i1:$rhs, v2f64:$tval, v2f64:$fval, SETNE)),
991 (SELECT_VSRC (CRXOR $lhs, $rhs), $tval, $fval)>;
993 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETLT)),
994 (SELECT_VSFRC (CRANDC $lhs, $rhs), $tval, $fval)>;
995 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETULT)),
996 (SELECT_VSFRC (CRANDC $rhs, $lhs), $tval, $fval)>;
997 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETLE)),
998 (SELECT_VSFRC (CRORC $lhs, $rhs), $tval, $fval)>;
999 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETULE)),
1000 (SELECT_VSFRC (CRORC $rhs, $lhs), $tval, $fval)>;
1001 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETEQ)),
1002 (SELECT_VSFRC (CREQV $lhs, $rhs), $tval, $fval)>;
1003 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETGE)),
1004 (SELECT_VSFRC (CRORC $rhs, $lhs), $tval, $fval)>;
1005 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGE)),
1006 (SELECT_VSFRC (CRORC $lhs, $rhs), $tval, $fval)>;
1007 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETGT)),
1008 (SELECT_VSFRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1009 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETUGT)),
1010 (SELECT_VSFRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1011 def : Pat<(f64 (selectcc i1:$lhs, i1:$rhs, f64:$tval, f64:$fval, SETNE)),
1012 (SELECT_VSFRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1015 def : Pat<(int_ppc_vsx_xvdivsp v4f32:$A, v4f32:$B),
1017 def : Pat<(int_ppc_vsx_xvdivdp v2f64:$A, v2f64:$B),
1020 // Reciprocal estimate
1021 def : Pat<(int_ppc_vsx_xvresp v4f32:$A),
1023 def : Pat<(int_ppc_vsx_xvredp v2f64:$A),
1026 // Recip. square root estimate
1027 def : Pat<(int_ppc_vsx_xvrsqrtesp v4f32:$A),
1029 def : Pat<(int_ppc_vsx_xvrsqrtedp v2f64:$A),
1032 let Predicates = [IsLittleEndian] in {
1033 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1034 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1035 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1036 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1037 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1038 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1039 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1040 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1043 let Predicates = [IsBigEndian] in {
1044 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1045 (f64 (XSCVSXDDP (COPY_TO_REGCLASS $S, VSFRC)))>;
1046 def : Pat<(f64 (PPCfcfid (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1047 (f64 (XSCVSXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1048 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1049 (f64 (XSCVUXDDP (COPY_TO_REGCLASS $S, VSFRC)))>;
1050 def : Pat<(f64 (PPCfcfidu (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1051 (f64 (XSCVUXDDP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1054 } // AddedComplexity
1057 // The following VSX instructions were introduced in Power ISA 2.07
1058 /* FIXME: if the operands are v2i64, these patterns will not match.
1059 we should define new patterns or otherwise match the same patterns
1060 when the elements are larger than i32.
1062 def HasP8Vector : Predicate<"PPCSubTarget->hasP8Vector()">;
1063 def HasDirectMove : Predicate<"PPCSubTarget->hasDirectMove()">;
1064 let Predicates = [HasP8Vector] in {
1065 let AddedComplexity = 400 in { // Prefer VSX patterns over non-VSX patterns.
1066 let isCommutable = 1 in {
1067 def XXLEQV : XX3Form<60, 186,
1068 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1069 "xxleqv $XT, $XA, $XB", IIC_VecGeneral,
1070 [(set v4i32:$XT, (vnot_ppc (xor v4i32:$XA, v4i32:$XB)))]>;
1071 def XXLNAND : XX3Form<60, 178,
1072 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1073 "xxlnand $XT, $XA, $XB", IIC_VecGeneral,
1074 [(set v4i32:$XT, (vnot_ppc (and v4i32:$XA,
1078 def : Pat<(int_ppc_vsx_xxleqv v4i32:$A, v4i32:$B),
1081 def XXLORC : XX3Form<60, 170,
1082 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1083 "xxlorc $XT, $XA, $XB", IIC_VecGeneral,
1084 [(set v4i32:$XT, (or v4i32:$XA, (vnot_ppc v4i32:$XB)))]>;
1086 // VSX scalar loads introduced in ISA 2.07
1087 let mayLoad = 1 in {
1088 def LXSSPX : XX1Form<31, 524, (outs vssrc:$XT), (ins memrr:$src),
1089 "lxsspx $XT, $src", IIC_LdStLFD,
1090 [(set f32:$XT, (load xoaddr:$src))]>;
1091 def LXSIWAX : XX1Form<31, 76, (outs vsfrc:$XT), (ins memrr:$src),
1092 "lxsiwax $XT, $src", IIC_LdStLFD,
1093 [(set f64:$XT, (PPClfiwax xoaddr:$src))]>;
1094 def LXSIWZX : XX1Form<31, 12, (outs vsfrc:$XT), (ins memrr:$src),
1095 "lxsiwzx $XT, $src", IIC_LdStLFD,
1096 [(set f64:$XT, (PPClfiwzx xoaddr:$src))]>;
1099 // VSX scalar stores introduced in ISA 2.07
1100 let mayStore = 1 in {
1101 def STXSSPX : XX1Form<31, 652, (outs), (ins vssrc:$XT, memrr:$dst),
1102 "stxsspx $XT, $dst", IIC_LdStSTFD,
1103 [(store f32:$XT, xoaddr:$dst)]>;
1104 def STXSIWX : XX1Form<31, 140, (outs), (ins vsfrc:$XT, memrr:$dst),
1105 "stxsiwx $XT, $dst", IIC_LdStSTFD,
1106 [(PPCstfiwx f64:$XT, xoaddr:$dst)]>;
1109 def : Pat<(f64 (extloadf32 xoaddr:$src)),
1110 (COPY_TO_REGCLASS (LXSSPX xoaddr:$src), VSFRC)>;
1111 def : Pat<(f64 (fextend f32:$src)),
1112 (COPY_TO_REGCLASS $src, VSFRC)>;
1114 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLT)),
1115 (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1116 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULT)),
1117 (SELECT_VSSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1118 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETLE)),
1119 (SELECT_VSSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1120 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETULE)),
1121 (SELECT_VSSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1122 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETEQ)),
1123 (SELECT_VSSRC (CREQV $lhs, $rhs), $tval, $fval)>;
1124 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETGE)),
1125 (SELECT_VSSRC (CRORC $rhs, $lhs), $tval, $fval)>;
1126 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETUGE)),
1127 (SELECT_VSSRC (CRORC $lhs, $rhs), $tval, $fval)>;
1128 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETGT)),
1129 (SELECT_VSSRC (CRANDC $rhs, $lhs), $tval, $fval)>;
1130 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETUGT)),
1131 (SELECT_VSSRC (CRANDC $lhs, $rhs), $tval, $fval)>;
1132 def : Pat<(f32 (selectcc i1:$lhs, i1:$rhs, f32:$tval, f32:$fval, SETNE)),
1133 (SELECT_VSSRC (CRXOR $lhs, $rhs), $tval, $fval)>;
1135 // VSX Elementary Scalar FP arithmetic (SP)
1136 let isCommutable = 1 in {
1137 def XSADDSP : XX3Form<60, 0,
1138 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1139 "xsaddsp $XT, $XA, $XB", IIC_VecFP,
1140 [(set f32:$XT, (fadd f32:$XA, f32:$XB))]>;
1141 def XSMULSP : XX3Form<60, 16,
1142 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1143 "xsmulsp $XT, $XA, $XB", IIC_VecFP,
1144 [(set f32:$XT, (fmul f32:$XA, f32:$XB))]>;
1147 def XSDIVSP : XX3Form<60, 24,
1148 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1149 "xsdivsp $XT, $XA, $XB", IIC_FPDivS,
1150 [(set f32:$XT, (fdiv f32:$XA, f32:$XB))]>;
1151 def XSRESP : XX2Form<60, 26,
1152 (outs vssrc:$XT), (ins vssrc:$XB),
1153 "xsresp $XT, $XB", IIC_VecFP,
1154 [(set f32:$XT, (PPCfre f32:$XB))]>;
1155 def XSSQRTSP : XX2Form<60, 11,
1156 (outs vssrc:$XT), (ins vssrc:$XB),
1157 "xssqrtsp $XT, $XB", IIC_FPSqrtS,
1158 [(set f32:$XT, (fsqrt f32:$XB))]>;
1159 def XSRSQRTESP : XX2Form<60, 10,
1160 (outs vssrc:$XT), (ins vssrc:$XB),
1161 "xsrsqrtesp $XT, $XB", IIC_VecFP,
1162 [(set f32:$XT, (PPCfrsqrte f32:$XB))]>;
1163 def XSSUBSP : XX3Form<60, 8,
1164 (outs vssrc:$XT), (ins vssrc:$XA, vssrc:$XB),
1165 "xssubsp $XT, $XA, $XB", IIC_VecFP,
1166 [(set f32:$XT, (fsub f32:$XA, f32:$XB))]>;
1169 let BaseName = "XSMADDASP" in {
1170 let isCommutable = 1 in
1171 def XSMADDASP : XX3Form<60, 1,
1173 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1174 "xsmaddasp $XT, $XA, $XB", IIC_VecFP,
1175 [(set f32:$XT, (fma f32:$XA, f32:$XB, f32:$XTi))]>,
1176 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1178 let IsVSXFMAAlt = 1 in
1179 def XSMADDMSP : XX3Form<60, 9,
1181 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1182 "xsmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
1183 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1187 let BaseName = "XSMSUBASP" in {
1188 let isCommutable = 1 in
1189 def XSMSUBASP : XX3Form<60, 17,
1191 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1192 "xsmsubasp $XT, $XA, $XB", IIC_VecFP,
1193 [(set f32:$XT, (fma f32:$XA, f32:$XB,
1194 (fneg f32:$XTi)))]>,
1195 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1197 let IsVSXFMAAlt = 1 in
1198 def XSMSUBMSP : XX3Form<60, 25,
1200 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1201 "xsmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
1202 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1206 let BaseName = "XSNMADDASP" in {
1207 let isCommutable = 1 in
1208 def XSNMADDASP : XX3Form<60, 129,
1210 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1211 "xsnmaddasp $XT, $XA, $XB", IIC_VecFP,
1212 [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
1214 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1216 let IsVSXFMAAlt = 1 in
1217 def XSNMADDMSP : XX3Form<60, 137,
1219 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1220 "xsnmaddmsp $XT, $XA, $XB", IIC_VecFP, []>,
1221 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1225 let BaseName = "XSNMSUBASP" in {
1226 let isCommutable = 1 in
1227 def XSNMSUBASP : XX3Form<60, 145,
1229 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1230 "xsnmsubasp $XT, $XA, $XB", IIC_VecFP,
1231 [(set f32:$XT, (fneg (fma f32:$XA, f32:$XB,
1232 (fneg f32:$XTi))))]>,
1233 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1235 let IsVSXFMAAlt = 1 in
1236 def XSNMSUBMSP : XX3Form<60, 153,
1238 (ins vssrc:$XTi, vssrc:$XA, vssrc:$XB),
1239 "xsnmsubmsp $XT, $XA, $XB", IIC_VecFP, []>,
1240 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">,
1244 // Single Precision Conversions (FP <-> INT)
1245 def XSCVSXDSP : XX2Form<60, 312,
1246 (outs vssrc:$XT), (ins vsfrc:$XB),
1247 "xscvsxdsp $XT, $XB", IIC_VecFP,
1248 [(set f32:$XT, (PPCfcfids f64:$XB))]>;
1249 def XSCVUXDSP : XX2Form<60, 296,
1250 (outs vssrc:$XT), (ins vsfrc:$XB),
1251 "xscvuxdsp $XT, $XB", IIC_VecFP,
1252 [(set f32:$XT, (PPCfcfidus f64:$XB))]>;
1254 // Conversions between vector and scalar single precision
1255 def XSCVDPSPN : XX2Form<60, 267, (outs vsrc:$XT), (ins vssrc:$XB),
1256 "xscvdpspn $XT, $XB", IIC_VecFP, []>;
1257 def XSCVSPDPN : XX2Form<60, 331, (outs vssrc:$XT), (ins vsrc:$XB),
1258 "xscvspdpn $XT, $XB", IIC_VecFP, []>;
1260 let Predicates = [IsLittleEndian] in {
1261 def : Pat<(f32 (PPCfcfids (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1262 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1263 def : Pat<(f32 (PPCfcfids (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1264 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1265 def : Pat<(f32 (PPCfcfidus (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1266 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1267 def : Pat<(f32 (PPCfcfidus (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1268 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (f64 (COPY_TO_REGCLASS $S, VSRC)), VSFRC)))>;
1271 let Predicates = [IsBigEndian] in {
1272 def : Pat<(f32 (PPCfcfids (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1273 (f32 (XSCVSXDSP (COPY_TO_REGCLASS $S, VSFRC)))>;
1274 def : Pat<(f32 (PPCfcfids (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1275 (f32 (XSCVSXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1276 def : Pat<(f32 (PPCfcfidus (PPCmtvsra (i64 (vector_extract v2i64:$S, 0))))),
1277 (f32 (XSCVUXDSP (COPY_TO_REGCLASS $S, VSFRC)))>;
1278 def : Pat<(f32 (PPCfcfidus (PPCmtvsra (i64 (vector_extract v2i64:$S, 1))))),
1279 (f32 (XSCVUXDSP (COPY_TO_REGCLASS (XXPERMDI $S, $S, 2), VSFRC)))>;
1281 } // AddedComplexity = 400
1284 let Predicates = [HasDirectMove] in {
1285 // VSX direct move instructions
1286 def MFVSRD : XX1_RS6_RD5_XO<31, 51, (outs g8rc:$rA), (ins vsfrc:$XT),
1287 "mfvsrd $rA, $XT", IIC_VecGeneral,
1288 [(set i64:$rA, (PPCmfvsr f64:$XT))]>,
1289 Requires<[In64BitMode]>;
1290 def MFVSRWZ : XX1_RS6_RD5_XO<31, 115, (outs gprc:$rA), (ins vsfrc:$XT),
1291 "mfvsrwz $rA, $XT", IIC_VecGeneral,
1292 [(set i32:$rA, (PPCmfvsr f64:$XT))]>;
1293 def MTVSRD : XX1_RS6_RD5_XO<31, 179, (outs vsfrc:$XT), (ins g8rc:$rA),
1294 "mtvsrd $XT, $rA", IIC_VecGeneral,
1295 [(set f64:$XT, (PPCmtvsra i64:$rA))]>,
1296 Requires<[In64BitMode]>;
1297 def MTVSRWA : XX1_RS6_RD5_XO<31, 211, (outs vsfrc:$XT), (ins gprc:$rA),
1298 "mtvsrwa $XT, $rA", IIC_VecGeneral,
1299 [(set f64:$XT, (PPCmtvsra i32:$rA))]>;
1300 def MTVSRWZ : XX1_RS6_RD5_XO<31, 243, (outs vsfrc:$XT), (ins gprc:$rA),
1301 "mtvsrwz $XT, $rA", IIC_VecGeneral,
1302 [(set f64:$XT, (PPCmtvsrz i32:$rA))]>;
1305 let Predicates = [IsISA3_0, HasDirectMove] in {
1306 def MTVSRWS: XX1_RS6_RD5_XO<31, 403, (outs vsrc:$XT), (ins gprc:$rA),
1307 "mtvsrws $XT, $rA", IIC_VecGeneral,
1310 def MTVSRDD: XX1Form<31, 435, (outs vsrc:$XT), (ins g8rc:$rA, g8rc:$rB),
1311 "mtvsrdd $XT, $rA, $rB", IIC_VecGeneral,
1312 []>, Requires<[In64BitMode]>;
1314 def MFVSRLD: XX1_RS6_RD5_XO<31, 307, (outs g8rc:$rA), (ins vsrc:$XT),
1315 "mfvsrld $rA, $XT", IIC_VecGeneral,
1316 []>, Requires<[In64BitMode]>;
1318 } // IsISA3_0, HasDirectMove
1320 /* Direct moves of various widths from GPR's into VSR's. Each move lines
1321 the value up into element 0 (both BE and LE). Namely, entities smaller than
1322 a doubleword are shifted left and moved for BE. For LE, they're moved, then
1323 swapped to go into the least significant element of the VSR.
1329 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 56, 7));
1333 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 48, 15));
1337 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32), 32, 31));
1338 dag BE_DWORD_0 = (MTVSRD $A);
1340 dag LE_MTVSRW = (MTVSRD (INSERT_SUBREG (i64 (IMPLICIT_DEF)), $A, sub_32));
1341 dag LE_WORD_1 = (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1342 LE_MTVSRW, sub_64));
1343 dag LE_WORD_0 = (XXPERMDI LE_WORD_1, LE_WORD_1, 2);
1344 dag LE_DWORD_1 = (v2i64 (INSERT_SUBREG (v2i64 (IMPLICIT_DEF)),
1345 BE_DWORD_0, sub_64));
1346 dag LE_DWORD_0 = (XXPERMDI LE_DWORD_1, LE_DWORD_1, 2);
1349 /* Patterns for extracting elements out of vectors. Integer elements are
1350 extracted using direct move operations. Patterns for extracting elements
1351 whose indices are not available at compile time are also provided with
1352 various _VARIABLE_ patterns.
1353 The numbering for the DAG's is for LE, but when used on BE, the correct
1354 LE element can just be used (i.e. LE_BYTE_2 == BE_BYTE_13).
1356 def VectorExtractions {
1357 // Doubleword extraction
1361 (XXPERMDI (COPY_TO_REGCLASS $S, VSRC),
1362 (COPY_TO_REGCLASS $S, VSRC), 2), sub_64));
1363 dag LE_DWORD_1 = (MFVSRD
1365 (v2i64 (COPY_TO_REGCLASS $S, VSRC)), sub_64));
1368 dag LE_WORD_0 = (MFVSRWZ (EXTRACT_SUBREG (XXPERMDI $S, $S, 2), sub_64));
1369 dag LE_WORD_1 = (MFVSRWZ (EXTRACT_SUBREG (XXSLDWI $S, $S, 1), sub_64));
1370 dag LE_WORD_2 = (MFVSRWZ (EXTRACT_SUBREG
1371 (v2i64 (COPY_TO_REGCLASS $S, VSRC)), sub_64));
1372 dag LE_WORD_3 = (MFVSRWZ (EXTRACT_SUBREG (XXSLDWI $S, $S, 3), sub_64));
1374 // Halfword extraction
1375 dag LE_HALF_0 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 0, 48), sub_32));
1376 dag LE_HALF_1 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 48, 48), sub_32));
1377 dag LE_HALF_2 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 32, 48), sub_32));
1378 dag LE_HALF_3 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 16, 48), sub_32));
1379 dag LE_HALF_4 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 0, 48), sub_32));
1380 dag LE_HALF_5 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 48, 48), sub_32));
1381 dag LE_HALF_6 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 32, 48), sub_32));
1382 dag LE_HALF_7 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 16, 48), sub_32));
1385 dag LE_BYTE_0 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 0, 56), sub_32));
1386 dag LE_BYTE_1 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 56, 56), sub_32));
1387 dag LE_BYTE_2 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 48, 56), sub_32));
1388 dag LE_BYTE_3 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 40, 56), sub_32));
1389 dag LE_BYTE_4 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 32, 56), sub_32));
1390 dag LE_BYTE_5 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 24, 56), sub_32));
1391 dag LE_BYTE_6 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 16, 56), sub_32));
1392 dag LE_BYTE_7 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_0, 8, 56), sub_32));
1393 dag LE_BYTE_8 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 0, 56), sub_32));
1394 dag LE_BYTE_9 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 56, 56), sub_32));
1395 dag LE_BYTE_10 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 48, 56), sub_32));
1396 dag LE_BYTE_11 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 40, 56), sub_32));
1397 dag LE_BYTE_12 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 32, 56), sub_32));
1398 dag LE_BYTE_13 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 24, 56), sub_32));
1399 dag LE_BYTE_14 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 16, 56), sub_32));
1400 dag LE_BYTE_15 = (i32 (EXTRACT_SUBREG (RLDICL LE_DWORD_1, 8, 56), sub_32));
1402 /* Variable element number (BE and LE patterns must be specified separately)
1403 This is a rather involved process.
1405 Conceptually, this is how the move is accomplished:
1406 1. Identify which doubleword contains the element
1407 2. Shift in the VMX register so that the correct doubleword is correctly
1408 lined up for the MFVSRD
1409 3. Perform the move so that the element (along with some extra stuff)
1411 4. Right shift within the GPR so that the element is right-justified
1413 Of course, the index is an element number which has a different meaning
1414 on LE/BE so the patterns have to be specified separately.
1416 Note: The final result will be the element right-justified with high
1417 order bits being arbitrarily defined (namely, whatever was in the
1418 vector register to the left of the value originally).
1423 - For elements 0-7, we shift left by 8 bytes since they're on the right
1424 - For elements 8-15, we need not shift (shift left by zero bytes)
1425 This is accomplished by inverting the bits of the index and AND-ing
1426 with 0x8 (i.e. clearing all bits of the index and inverting bit 60).
1428 dag LE_VBYTE_PERM_VEC = (LVSL ZERO8, (ANDC8 (LI8 8), $Idx));
1431 // - Now that we set up the shift amount, we shift in the VMX register
1432 dag LE_VBYTE_PERMUTE = (VPERM $S, $S, LE_VBYTE_PERM_VEC);
1435 // - The doubleword containing our element is moved to a GPR
1436 dag LE_MV_VBYTE = (MFVSRD
1438 (v2i64 (COPY_TO_REGCLASS LE_VBYTE_PERMUTE, VSRC)),
1442 - Truncate the element number to the range 0-7 (8-15 are symmetrical
1443 and out of range values are truncated accordingly)
1444 - Multiply by 8 as we need to shift right by the number of bits, not bytes
1445 - Shift right in the GPR by the calculated value
1447 dag LE_VBYTE_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 7), $Idx), 3, 60),
1449 dag LE_VARIABLE_BYTE = (EXTRACT_SUBREG (SRD LE_MV_VBYTE, LE_VBYTE_SHIFT),
1452 /* LE variable halfword
1454 - For elements 0-3, we shift left by 8 since they're on the right
1455 - For elements 4-7, we need not shift (shift left by zero bytes)
1456 Similarly to the byte pattern, we invert the bits of the index, but we
1457 AND with 0x4 (i.e. clear all bits of the index and invert bit 61).
1458 Of course, the shift is still by 8 bytes, so we must multiply by 2.
1460 dag LE_VHALF_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDC8 (LI8 4), $Idx), 1, 62));
1463 // - Now that we set up the shift amount, we shift in the VMX register
1464 dag LE_VHALF_PERMUTE = (VPERM $S, $S, LE_VHALF_PERM_VEC);
1467 // - The doubleword containing our element is moved to a GPR
1468 dag LE_MV_VHALF = (MFVSRD
1470 (v2i64 (COPY_TO_REGCLASS LE_VHALF_PERMUTE, VSRC)),
1474 - Truncate the element number to the range 0-3 (4-7 are symmetrical
1475 and out of range values are truncated accordingly)
1476 - Multiply by 16 as we need to shift right by the number of bits
1477 - Shift right in the GPR by the calculated value
1479 dag LE_VHALF_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 3), $Idx), 4, 59),
1481 dag LE_VARIABLE_HALF = (EXTRACT_SUBREG (SRD LE_MV_VHALF, LE_VHALF_SHIFT),
1486 - For elements 0-1, we shift left by 8 since they're on the right
1487 - For elements 2-3, we need not shift
1489 dag LE_VWORD_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDC8 (LI8 2), $Idx), 2, 61));
1492 // - Now that we set up the shift amount, we shift in the VMX register
1493 dag LE_VWORD_PERMUTE = (VPERM $S, $S, LE_VWORD_PERM_VEC);
1496 // - The doubleword containing our element is moved to a GPR
1497 dag LE_MV_VWORD = (MFVSRD
1499 (v2i64 (COPY_TO_REGCLASS LE_VWORD_PERMUTE, VSRC)),
1503 - Truncate the element number to the range 0-1 (2-3 are symmetrical
1504 and out of range values are truncated accordingly)
1505 - Multiply by 32 as we need to shift right by the number of bits
1506 - Shift right in the GPR by the calculated value
1508 dag LE_VWORD_SHIFT = (EXTRACT_SUBREG (RLDICR (AND8 (LI8 1), $Idx), 5, 58),
1510 dag LE_VARIABLE_WORD = (EXTRACT_SUBREG (SRD LE_MV_VWORD, LE_VWORD_SHIFT),
1513 /* LE variable doubleword
1515 - For element 0, we shift left by 8 since it's on the right
1516 - For element 1, we need not shift
1518 dag LE_VDWORD_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDC8 (LI8 1), $Idx), 3, 60));
1521 // - Now that we set up the shift amount, we shift in the VMX register
1522 dag LE_VDWORD_PERMUTE = (VPERM $S, $S, LE_VDWORD_PERM_VEC);
1525 // - The doubleword containing our element is moved to a GPR
1526 // - Number 4. is not needed for the doubleword as the value is 64-bits
1527 dag LE_VARIABLE_DWORD =
1528 (MFVSRD (EXTRACT_SUBREG
1529 (v2i64 (COPY_TO_REGCLASS LE_VDWORD_PERMUTE, VSRC)),
1532 /* LE variable float
1533 - Shift the vector to line up the desired element to BE Word 0
1534 - Convert 32-bit float to a 64-bit single precision float
1536 dag LE_VFLOAT_PERM_VEC = (LVSL ZERO8, (RLDICR (XOR8 (LI8 3), $Idx), 2, 61));
1537 dag LE_VFLOAT_PERMUTE = (VPERM $S, $S, LE_VFLOAT_PERM_VEC);
1538 dag LE_VARIABLE_FLOAT = (XSCVSPDPN LE_VFLOAT_PERMUTE);
1540 /* LE variable double
1541 Same as the LE doubleword except there is no move.
1543 dag LE_VDOUBLE_PERMUTE = (VPERM (COPY_TO_REGCLASS $S, VRRC),
1544 (COPY_TO_REGCLASS $S, VRRC),
1545 LE_VDWORD_PERM_VEC);
1546 dag LE_VARIABLE_DOUBLE = (COPY_TO_REGCLASS LE_VDOUBLE_PERMUTE, VSRC);
1549 The algorithm here is the same as the LE variable byte except:
1550 - The shift in the VMX register is by 0/8 for opposite element numbers so
1551 we simply AND the element number with 0x8
1552 - The order of elements after the move to GPR is reversed, so we invert
1553 the bits of the index prior to truncating to the range 0-7
1555 dag BE_VBYTE_PERM_VEC = (LVSL ZERO8, (ANDIo8 $Idx, 8));
1556 dag BE_VBYTE_PERMUTE = (VPERM $S, $S, BE_VBYTE_PERM_VEC);
1557 dag BE_MV_VBYTE = (MFVSRD
1559 (v2i64 (COPY_TO_REGCLASS BE_VBYTE_PERMUTE, VSRC)),
1561 dag BE_VBYTE_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 7), $Idx), 3, 60),
1563 dag BE_VARIABLE_BYTE = (EXTRACT_SUBREG (SRD BE_MV_VBYTE, BE_VBYTE_SHIFT),
1566 /* BE variable halfword
1567 The algorithm here is the same as the LE variable halfword except:
1568 - The shift in the VMX register is by 0/8 for opposite element numbers so
1569 we simply AND the element number with 0x4 and multiply by 2
1570 - The order of elements after the move to GPR is reversed, so we invert
1571 the bits of the index prior to truncating to the range 0-3
1573 dag BE_VHALF_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDIo8 $Idx, 4), 1, 62));
1574 dag BE_VHALF_PERMUTE = (VPERM $S, $S, BE_VHALF_PERM_VEC);
1575 dag BE_MV_VHALF = (MFVSRD
1577 (v2i64 (COPY_TO_REGCLASS BE_VHALF_PERMUTE, VSRC)),
1579 dag BE_VHALF_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 3), $Idx), 4, 59),
1581 dag BE_VARIABLE_HALF = (EXTRACT_SUBREG (SRD BE_MV_VHALF, BE_VHALF_SHIFT),
1585 The algorithm is the same as the LE variable word except:
1586 - The shift in the VMX register happens for opposite element numbers
1587 - The order of elements after the move to GPR is reversed, so we invert
1588 the bits of the index prior to truncating to the range 0-1
1590 dag BE_VWORD_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDIo8 $Idx, 2), 2, 61));
1591 dag BE_VWORD_PERMUTE = (VPERM $S, $S, BE_VWORD_PERM_VEC);
1592 dag BE_MV_VWORD = (MFVSRD
1594 (v2i64 (COPY_TO_REGCLASS BE_VWORD_PERMUTE, VSRC)),
1596 dag BE_VWORD_SHIFT = (EXTRACT_SUBREG (RLDICR (ANDC8 (LI8 1), $Idx), 5, 58),
1598 dag BE_VARIABLE_WORD = (EXTRACT_SUBREG (SRD BE_MV_VWORD, BE_VWORD_SHIFT),
1601 /* BE variable doubleword
1602 Same as the LE doubleword except we shift in the VMX register for opposite
1605 dag BE_VDWORD_PERM_VEC = (LVSL ZERO8, (RLDICR (ANDIo8 $Idx, 1), 3, 60));
1606 dag BE_VDWORD_PERMUTE = (VPERM $S, $S, BE_VDWORD_PERM_VEC);
1607 dag BE_VARIABLE_DWORD =
1608 (MFVSRD (EXTRACT_SUBREG
1609 (v2i64 (COPY_TO_REGCLASS BE_VDWORD_PERMUTE, VSRC)),
1612 /* BE variable float
1613 - Shift the vector to line up the desired element to BE Word 0
1614 - Convert 32-bit float to a 64-bit single precision float
1616 dag BE_VFLOAT_PERM_VEC = (LVSL ZERO8, (RLDICR $Idx, 2, 61));
1617 dag BE_VFLOAT_PERMUTE = (VPERM $S, $S, BE_VFLOAT_PERM_VEC);
1618 dag BE_VARIABLE_FLOAT = (XSCVSPDPN BE_VFLOAT_PERMUTE);
1620 /* BE variable double
1621 Same as the BE doubleword except there is no move.
1623 dag BE_VDOUBLE_PERMUTE = (VPERM (COPY_TO_REGCLASS $S, VRRC),
1624 (COPY_TO_REGCLASS $S, VRRC),
1625 BE_VDWORD_PERM_VEC);
1626 dag BE_VARIABLE_DOUBLE = (COPY_TO_REGCLASS BE_VDOUBLE_PERMUTE, VSRC);
1629 // v4f32 scalar <-> vector conversions (BE)
1630 let Predicates = [IsBigEndian, HasP8Vector] in {
1631 def : Pat<(v4f32 (scalar_to_vector f32:$A)),
1632 (v4f32 (XSCVDPSPN $A))>;
1633 def : Pat<(f32 (vector_extract v4f32:$S, 0)),
1634 (f32 (XSCVSPDPN $S))>;
1635 def : Pat<(f32 (vector_extract v4f32:$S, 1)),
1636 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 1)))>;
1637 def : Pat<(f32 (vector_extract v4f32:$S, 2)),
1638 (f32 (XSCVSPDPN (XXPERMDI $S, $S, 2)))>;
1639 def : Pat<(f32 (vector_extract v4f32:$S, 3)),
1640 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 3)))>;
1641 def : Pat<(f32 (vector_extract v4f32:$S, i64:$Idx)),
1642 (f32 VectorExtractions.BE_VARIABLE_FLOAT)>;
1643 } // IsBigEndian, HasP8Vector
1645 // Variable index vector_extract for v2f64 does not require P8Vector
1646 let Predicates = [IsBigEndian, HasVSX] in
1647 def : Pat<(f64 (vector_extract v2f64:$S, i64:$Idx)),
1648 (f64 VectorExtractions.BE_VARIABLE_DOUBLE)>;
1650 let Predicates = [IsBigEndian, HasDirectMove] in {
1651 // v16i8 scalar <-> vector conversions (BE)
1652 def : Pat<(v16i8 (scalar_to_vector i32:$A)),
1653 (v16i8 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_BYTE_0, sub_64))>;
1654 def : Pat<(v8i16 (scalar_to_vector i32:$A)),
1655 (v8i16 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_HALF_0, sub_64))>;
1656 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
1657 (v4i32 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_WORD_0, sub_64))>;
1658 def : Pat<(v2i64 (scalar_to_vector i64:$A)),
1659 (v2i64 (SUBREG_TO_REG (i64 1), MovesToVSR.BE_DWORD_0, sub_64))>;
1660 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
1661 (i32 VectorExtractions.LE_BYTE_15)>;
1662 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
1663 (i32 VectorExtractions.LE_BYTE_14)>;
1664 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
1665 (i32 VectorExtractions.LE_BYTE_13)>;
1666 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
1667 (i32 VectorExtractions.LE_BYTE_12)>;
1668 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
1669 (i32 VectorExtractions.LE_BYTE_11)>;
1670 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
1671 (i32 VectorExtractions.LE_BYTE_10)>;
1672 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
1673 (i32 VectorExtractions.LE_BYTE_9)>;
1674 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
1675 (i32 VectorExtractions.LE_BYTE_8)>;
1676 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
1677 (i32 VectorExtractions.LE_BYTE_7)>;
1678 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
1679 (i32 VectorExtractions.LE_BYTE_6)>;
1680 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
1681 (i32 VectorExtractions.LE_BYTE_5)>;
1682 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
1683 (i32 VectorExtractions.LE_BYTE_4)>;
1684 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
1685 (i32 VectorExtractions.LE_BYTE_3)>;
1686 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
1687 (i32 VectorExtractions.LE_BYTE_2)>;
1688 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
1689 (i32 VectorExtractions.LE_BYTE_1)>;
1690 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
1691 (i32 VectorExtractions.LE_BYTE_0)>;
1692 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
1693 (i32 VectorExtractions.BE_VARIABLE_BYTE)>;
1695 // v8i16 scalar <-> vector conversions (BE)
1696 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
1697 (i32 VectorExtractions.LE_HALF_7)>;
1698 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
1699 (i32 VectorExtractions.LE_HALF_6)>;
1700 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
1701 (i32 VectorExtractions.LE_HALF_5)>;
1702 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
1703 (i32 VectorExtractions.LE_HALF_4)>;
1704 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
1705 (i32 VectorExtractions.LE_HALF_3)>;
1706 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
1707 (i32 VectorExtractions.LE_HALF_2)>;
1708 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
1709 (i32 VectorExtractions.LE_HALF_1)>;
1710 def : Pat<(i32 (vector_extract v8i16:$S, 7)),
1711 (i32 VectorExtractions.LE_HALF_0)>;
1712 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
1713 (i32 VectorExtractions.BE_VARIABLE_HALF)>;
1715 // v4i32 scalar <-> vector conversions (BE)
1716 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
1717 (i32 VectorExtractions.LE_WORD_3)>;
1718 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
1719 (i32 VectorExtractions.LE_WORD_2)>;
1720 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
1721 (i32 VectorExtractions.LE_WORD_1)>;
1722 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
1723 (i32 VectorExtractions.LE_WORD_0)>;
1724 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
1725 (i32 VectorExtractions.BE_VARIABLE_WORD)>;
1727 // v2i64 scalar <-> vector conversions (BE)
1728 def : Pat<(i64 (vector_extract v2i64:$S, 0)),
1729 (i64 VectorExtractions.LE_DWORD_1)>;
1730 def : Pat<(i64 (vector_extract v2i64:$S, 1)),
1731 (i64 VectorExtractions.LE_DWORD_0)>;
1732 def : Pat<(i64 (vector_extract v2i64:$S, i64:$Idx)),
1733 (i64 VectorExtractions.BE_VARIABLE_DWORD)>;
1734 } // IsBigEndian, HasDirectMove
1736 // v4f32 scalar <-> vector conversions (LE)
1737 let Predicates = [IsLittleEndian, HasP8Vector] in {
1738 def : Pat<(v4f32 (scalar_to_vector f32:$A)),
1739 (v4f32 (XXSLDWI (XSCVDPSPN $A), (XSCVDPSPN $A), 1))>;
1740 def : Pat<(f32 (vector_extract v4f32:$S, 0)),
1741 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 3)))>;
1742 def : Pat<(f32 (vector_extract v4f32:$S, 1)),
1743 (f32 (XSCVSPDPN (XXPERMDI $S, $S, 2)))>;
1744 def : Pat<(f32 (vector_extract v4f32:$S, 2)),
1745 (f32 (XSCVSPDPN (XXSLDWI $S, $S, 1)))>;
1746 def : Pat<(f32 (vector_extract v4f32:$S, 3)),
1747 (f32 (XSCVSPDPN $S))>;
1748 def : Pat<(f32 (vector_extract v4f32:$S, i64:$Idx)),
1749 (f32 VectorExtractions.LE_VARIABLE_FLOAT)>;
1750 } // IsLittleEndian, HasP8Vector
1752 // Variable index vector_extract for v2f64 does not require P8Vector
1753 let Predicates = [IsLittleEndian, HasVSX] in
1754 def : Pat<(f64 (vector_extract v2f64:$S, i64:$Idx)),
1755 (f64 VectorExtractions.LE_VARIABLE_DOUBLE)>;
1757 let Predicates = [IsLittleEndian, HasDirectMove] in {
1758 // v16i8 scalar <-> vector conversions (LE)
1759 def : Pat<(v16i8 (scalar_to_vector i32:$A)),
1760 (v16i8 (COPY_TO_REGCLASS MovesToVSR.LE_WORD_0, VSRC))>;
1761 def : Pat<(v8i16 (scalar_to_vector i32:$A)),
1762 (v8i16 (COPY_TO_REGCLASS MovesToVSR.LE_WORD_0, VSRC))>;
1763 def : Pat<(v4i32 (scalar_to_vector i32:$A)),
1764 (v4i32 MovesToVSR.LE_WORD_0)>;
1765 def : Pat<(v2i64 (scalar_to_vector i64:$A)),
1766 (v2i64 MovesToVSR.LE_DWORD_0)>;
1767 def : Pat<(i32 (vector_extract v16i8:$S, 0)),
1768 (i32 VectorExtractions.LE_BYTE_0)>;
1769 def : Pat<(i32 (vector_extract v16i8:$S, 1)),
1770 (i32 VectorExtractions.LE_BYTE_1)>;
1771 def : Pat<(i32 (vector_extract v16i8:$S, 2)),
1772 (i32 VectorExtractions.LE_BYTE_2)>;
1773 def : Pat<(i32 (vector_extract v16i8:$S, 3)),
1774 (i32 VectorExtractions.LE_BYTE_3)>;
1775 def : Pat<(i32 (vector_extract v16i8:$S, 4)),
1776 (i32 VectorExtractions.LE_BYTE_4)>;
1777 def : Pat<(i32 (vector_extract v16i8:$S, 5)),
1778 (i32 VectorExtractions.LE_BYTE_5)>;
1779 def : Pat<(i32 (vector_extract v16i8:$S, 6)),
1780 (i32 VectorExtractions.LE_BYTE_6)>;
1781 def : Pat<(i32 (vector_extract v16i8:$S, 7)),
1782 (i32 VectorExtractions.LE_BYTE_7)>;
1783 def : Pat<(i32 (vector_extract v16i8:$S, 8)),
1784 (i32 VectorExtractions.LE_BYTE_8)>;
1785 def : Pat<(i32 (vector_extract v16i8:$S, 9)),
1786 (i32 VectorExtractions.LE_BYTE_9)>;
1787 def : Pat<(i32 (vector_extract v16i8:$S, 10)),
1788 (i32 VectorExtractions.LE_BYTE_10)>;
1789 def : Pat<(i32 (vector_extract v16i8:$S, 11)),
1790 (i32 VectorExtractions.LE_BYTE_11)>;
1791 def : Pat<(i32 (vector_extract v16i8:$S, 12)),
1792 (i32 VectorExtractions.LE_BYTE_12)>;
1793 def : Pat<(i32 (vector_extract v16i8:$S, 13)),
1794 (i32 VectorExtractions.LE_BYTE_13)>;
1795 def : Pat<(i32 (vector_extract v16i8:$S, 14)),
1796 (i32 VectorExtractions.LE_BYTE_14)>;
1797 def : Pat<(i32 (vector_extract v16i8:$S, 15)),
1798 (i32 VectorExtractions.LE_BYTE_15)>;
1799 def : Pat<(i32 (vector_extract v16i8:$S, i64:$Idx)),
1800 (i32 VectorExtractions.LE_VARIABLE_BYTE)>;
1802 // v8i16 scalar <-> vector conversions (LE)
1803 def : Pat<(i32 (vector_extract v8i16:$S, 0)),
1804 (i32 VectorExtractions.LE_HALF_0)>;
1805 def : Pat<(i32 (vector_extract v8i16:$S, 1)),
1806 (i32 VectorExtractions.LE_HALF_1)>;
1807 def : Pat<(i32 (vector_extract v8i16:$S, 2)),
1808 (i32 VectorExtractions.LE_HALF_2)>;
1809 def : Pat<(i32 (vector_extract v8i16:$S, 3)),
1810 (i32 VectorExtractions.LE_HALF_3)>;
1811 def : Pat<(i32 (vector_extract v8i16:$S, 4)),
1812 (i32 VectorExtractions.LE_HALF_4)>;
1813 def : Pat<(i32 (vector_extract v8i16:$S, 5)),
1814 (i32 VectorExtractions.LE_HALF_5)>;
1815 def : Pat<(i32 (vector_extract v8i16:$S, 6)),
1816 (i32 VectorExtractions.LE_HALF_6)>;
1817 def : Pat<(i32 (vector_extract v8i16:$S, 7)),
1818 (i32 VectorExtractions.LE_HALF_7)>;
1819 def : Pat<(i32 (vector_extract v8i16:$S, i64:$Idx)),
1820 (i32 VectorExtractions.LE_VARIABLE_HALF)>;
1822 // v4i32 scalar <-> vector conversions (LE)
1823 def : Pat<(i32 (vector_extract v4i32:$S, 0)),
1824 (i32 VectorExtractions.LE_WORD_0)>;
1825 def : Pat<(i32 (vector_extract v4i32:$S, 1)),
1826 (i32 VectorExtractions.LE_WORD_1)>;
1827 def : Pat<(i32 (vector_extract v4i32:$S, 2)),
1828 (i32 VectorExtractions.LE_WORD_2)>;
1829 def : Pat<(i32 (vector_extract v4i32:$S, 3)),
1830 (i32 VectorExtractions.LE_WORD_3)>;
1831 def : Pat<(i32 (vector_extract v4i32:$S, i64:$Idx)),
1832 (i32 VectorExtractions.LE_VARIABLE_WORD)>;
1834 // v2i64 scalar <-> vector conversions (LE)
1835 def : Pat<(i64 (vector_extract v2i64:$S, 0)),
1836 (i64 VectorExtractions.LE_DWORD_0)>;
1837 def : Pat<(i64 (vector_extract v2i64:$S, 1)),
1838 (i64 VectorExtractions.LE_DWORD_1)>;
1839 def : Pat<(i64 (vector_extract v2i64:$S, i64:$Idx)),
1840 (i64 VectorExtractions.LE_VARIABLE_DWORD)>;
1841 } // IsLittleEndian, HasDirectMove
1843 let Predicates = [HasDirectMove, HasVSX] in {
1844 // bitconvert f32 -> i32
1845 // (convert to 32-bit fp single, shift right 1 word, move to GPR)
1846 def : Pat<(i32 (bitconvert f32:$S)),
1847 (i32 (MFVSRWZ (EXTRACT_SUBREG
1848 (XXSLDWI (XSCVDPSPN $S),(XSCVDPSPN $S), 3),
1850 // bitconvert i32 -> f32
1851 // (move to FPR, shift left 1 word, convert to 64-bit fp single)
1852 def : Pat<(f32 (bitconvert i32:$A)),
1854 (XXSLDWI MovesToVSR.LE_WORD_1, MovesToVSR.LE_WORD_1, 1)))>;
1856 // bitconvert f64 -> i64
1857 // (move to GPR, nothing else needed)
1858 def : Pat<(i64 (bitconvert f64:$S)),
1861 // bitconvert i64 -> f64
1862 // (move to FPR, nothing else needed)
1863 def : Pat<(f64 (bitconvert i64:$S)),
1868 dag F32_TO_BE_WORD1 = (v4f32 (XXSLDWI (XSCVDPSPN $B), (XSCVDPSPN $B), 3));
1869 dag I32_TO_BE_WORD1 = (COPY_TO_REGCLASS (MTVSRWZ $B), VSRC);
1872 // The following VSX instructions were introduced in Power ISA 3.0
1873 def HasP9Vector : Predicate<"PPCSubTarget->hasP9Vector()">;
1874 let AddedComplexity = 400, Predicates = [HasP9Vector] in {
1876 // [PO VRT XO VRB XO /]
1877 class X_VT5_XO5_VB5<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
1879 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vrrc:$vT), (ins vrrc:$vB),
1880 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
1882 // [PO VRT XO VRB XO RO], Round to Odd version of [PO VRT XO VRB XO /]
1883 class X_VT5_XO5_VB5_Ro<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
1885 : X_VT5_XO5_VB5<opcode, xo2, xo, opc, pattern>, isDOT;
1887 // [PO VRT XO VRB XO /], but the VRB is only used the left 64 bits (or less),
1888 // So we use different operand class for VRB
1889 class X_VT5_XO5_VB5_TyVB<bits<6> opcode, bits<5> xo2, bits<10> xo, string opc,
1890 RegisterOperand vbtype, list<dag> pattern>
1891 : X_RD5_XO5_RS5<opcode, xo2, xo, (outs vrrc:$vT), (ins vbtype:$vB),
1892 !strconcat(opc, " $vT, $vB"), IIC_VecFP, pattern>;
1894 // [PO T XO B XO BX /]
1895 class XX2_RT5_XO5_XB6<bits<6> opcode, bits<5> xo2, bits<9> xo, string opc,
1897 : XX2_RD5_XO5_RS6<opcode, xo2, xo, (outs g8rc:$rT), (ins vsfrc:$XB),
1898 !strconcat(opc, " $rT, $XB"), IIC_VecFP, pattern>;
1900 // [PO T XO B XO BX TX]
1901 class XX2_XT6_XO5_XB6<bits<6> opcode, bits<5> xo2, bits<9> xo, string opc,
1902 RegisterOperand vtype, list<dag> pattern>
1903 : XX2_RD6_XO5_RS6<opcode, xo2, xo, (outs vtype:$XT), (ins vtype:$XB),
1904 !strconcat(opc, " $XT, $XB"), IIC_VecFP, pattern>;
1906 // [PO T A B XO AX BX TX], src and dest register use different operand class
1907 class XX3_XT5_XA5_XB5<bits<6> opcode, bits<8> xo, string opc,
1908 RegisterOperand xty, RegisterOperand aty, RegisterOperand bty,
1909 InstrItinClass itin, list<dag> pattern>
1910 : XX3Form<opcode, xo, (outs xty:$XT), (ins aty:$XA, bty:$XB),
1911 !strconcat(opc, " $XT, $XA, $XB"), itin, pattern>;
1913 // [PO VRT VRA VRB XO /]
1914 class X_VT5_VA5_VB5<bits<6> opcode, bits<10> xo, string opc,
1916 : XForm_1<opcode, xo, (outs vrrc:$vT), (ins vrrc:$vA, vrrc:$vB),
1917 !strconcat(opc, " $vT, $vA, $vB"), IIC_VecFP, pattern>;
1919 // [PO VRT VRA VRB XO RO], Round to Odd version of [PO VRT VRA VRB XO /]
1920 class X_VT5_VA5_VB5_Ro<bits<6> opcode, bits<10> xo, string opc,
1922 : X_VT5_VA5_VB5<opcode, xo, opc, pattern>, isDOT;
1924 //===--------------------------------------------------------------------===//
1925 // Quad-Precision Scalar Move Instructions:
1928 def XSCPSGNQP : X_VT5_VA5_VB5<63, 100, "xscpsgnqp", []>;
1930 // Absolute/Negative-Absolute/Negate
1931 def XSABSQP : X_VT5_XO5_VB5<63, 0, 804, "xsabsqp" , []>;
1932 def XSNABSQP : X_VT5_XO5_VB5<63, 8, 804, "xsnabsqp", []>;
1933 def XSNEGQP : X_VT5_XO5_VB5<63, 16, 804, "xsnegqp" , []>;
1935 //===--------------------------------------------------------------------===//
1936 // Quad-Precision Scalar Floating-Point Arithmetic Instructions:
1938 // Add/Divide/Multiply/Subtract
1939 def XSADDQP : X_VT5_VA5_VB5 <63, 4, "xsaddqp" , []>;
1940 def XSADDQPO : X_VT5_VA5_VB5_Ro<63, 4, "xsaddqpo", []>;
1941 def XSDIVQP : X_VT5_VA5_VB5 <63, 548, "xsdivqp" , []>;
1942 def XSDIVQPO : X_VT5_VA5_VB5_Ro<63, 548, "xsdivqpo", []>;
1943 def XSMULQP : X_VT5_VA5_VB5 <63, 36, "xsmulqp" , []>;
1944 def XSMULQPO : X_VT5_VA5_VB5_Ro<63, 36, "xsmulqpo", []>;
1945 def XSSUBQP : X_VT5_VA5_VB5 <63, 516, "xssubqp" , []>;
1946 def XSSUBQPO : X_VT5_VA5_VB5_Ro<63, 516, "xssubqpo", []>;
1949 def XSSQRTQP : X_VT5_XO5_VB5 <63, 27, 804, "xssqrtqp" , []>;
1950 def XSSQRTQPO : X_VT5_XO5_VB5_Ro<63, 27, 804, "xssqrtqpo", []>;
1952 // (Negative) Multiply-{Add/Subtract}
1953 def XSMADDQP : X_VT5_VA5_VB5 <63, 388, "xsmaddqp" , []>;
1954 def XSMADDQPO : X_VT5_VA5_VB5_Ro<63, 388, "xsmaddqpo" , []>;
1955 def XSMSUBQP : X_VT5_VA5_VB5 <63, 420, "xsmsubqp" , []>;
1956 def XSMSUBQPO : X_VT5_VA5_VB5_Ro<63, 420, "xsmsubqpo" , []>;
1957 def XSNMADDQP : X_VT5_VA5_VB5 <63, 452, "xsnmaddqp" , []>;
1958 def XSNMADDQPO: X_VT5_VA5_VB5_Ro<63, 452, "xsnmaddqpo", []>;
1959 def XSNMSUBQP : X_VT5_VA5_VB5 <63, 484, "xsnmsubqp" , []>;
1960 def XSNMSUBQPO: X_VT5_VA5_VB5_Ro<63, 484, "xsnmsubqpo", []>;
1962 //===--------------------------------------------------------------------===//
1963 // Quad/Double-Precision Compare Instructions:
1965 // [PO BF // VRA VRB XO /]
1966 class X_BF3_VA5_VB5<bits<6> opcode, bits<10> xo, string opc,
1968 : XForm_17<opcode, xo, (outs crrc:$crD), (ins vrrc:$VA, vrrc:$VB),
1969 !strconcat(opc, " $crD, $VA, $VB"), IIC_FPCompare> {
1970 let Pattern = pattern;
1973 // QP Compare Ordered/Unordered
1974 def XSCMPOQP : X_BF3_VA5_VB5<63, 132, "xscmpoqp", []>;
1975 def XSCMPUQP : X_BF3_VA5_VB5<63, 644, "xscmpuqp", []>;
1977 // DP/QP Compare Exponents
1978 def XSCMPEXPDP : XX3Form_1<60, 59,
1979 (outs crrc:$crD), (ins vsfrc:$XA, vsfrc:$XB),
1980 "xscmpexpdp $crD, $XA, $XB", IIC_FPCompare, []>;
1981 def XSCMPEXPQP : X_BF3_VA5_VB5<63, 164, "xscmpexpqp", []>;
1983 // DP Compare ==, >=, >, !=
1984 // Use vsrc for XT, because the entire register of XT is set.
1985 // XT.dword[1] = 0x0000_0000_0000_0000
1986 def XSCMPEQDP : XX3_XT5_XA5_XB5<60, 3, "xscmpeqdp", vsrc, vsfrc, vsfrc,
1988 def XSCMPGEDP : XX3_XT5_XA5_XB5<60, 19, "xscmpgedp", vsrc, vsfrc, vsfrc,
1990 def XSCMPGTDP : XX3_XT5_XA5_XB5<60, 11, "xscmpgtdp", vsrc, vsfrc, vsfrc,
1992 def XSCMPNEDP : XX3_XT5_XA5_XB5<60, 27, "xscmpnedp", vsrc, vsfrc, vsfrc,
1994 // Vector Compare Not Equal
1995 def XVCMPNEDP : XX3Form_Rc<60, 123,
1996 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
1997 "xvcmpnedp $XT, $XA, $XB", IIC_VecFPCompare, []>;
1999 def XVCMPNEDPo : XX3Form_Rc<60, 123,
2000 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
2001 "xvcmpnedp. $XT, $XA, $XB", IIC_VecFPCompare, []>,
2003 def XVCMPNESP : XX3Form_Rc<60, 91,
2004 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
2005 "xvcmpnesp $XT, $XA, $XB", IIC_VecFPCompare, []>;
2007 def XVCMPNESPo : XX3Form_Rc<60, 91,
2008 (outs vsrc:$XT), (ins vsrc:$XA, vsrc:$XB),
2009 "xvcmpnesp. $XT, $XA, $XB", IIC_VecFPCompare, []>,
2012 //===--------------------------------------------------------------------===//
2013 // Quad-Precision Floating-Point Conversion Instructions:
2016 def XSCVDPQP : X_VT5_XO5_VB5_TyVB<63, 22, 836, "xscvdpqp", vsfrc, []>;
2018 // Round & Convert QP -> DP (dword[1] is set to zero)
2019 def XSCVQPDP : X_VT5_XO5_VB5 <63, 20, 836, "xscvqpdp" , []>;
2020 def XSCVQPDPO : X_VT5_XO5_VB5_Ro<63, 20, 836, "xscvqpdpo", []>;
2022 // Truncate & Convert QP -> (Un)Signed (D)Word (dword[1] is set to zero)
2023 def XSCVQPSDZ : X_VT5_XO5_VB5<63, 25, 836, "xscvqpsdz", []>;
2024 def XSCVQPSWZ : X_VT5_XO5_VB5<63, 9, 836, "xscvqpswz", []>;
2025 def XSCVQPUDZ : X_VT5_XO5_VB5<63, 17, 836, "xscvqpudz", []>;
2026 def XSCVQPUWZ : X_VT5_XO5_VB5<63, 1, 836, "xscvqpuwz", []>;
2028 // Convert (Un)Signed DWord -> QP
2029 def XSCVSDQP : X_VT5_XO5_VB5_TyVB<63, 10, 836, "xscvsdqp", vsfrc, []>;
2030 def XSCVUDQP : X_VT5_XO5_VB5_TyVB<63, 2, 836, "xscvudqp", vsfrc, []>;
2032 //===--------------------------------------------------------------------===//
2033 // Round to Floating-Point Integer Instructions
2035 // (Round &) Convert DP <-> HP
2036 // Note! xscvdphp's src and dest register both use the left 64 bits, so we use
2037 // vsfrc for src and dest register. xscvhpdp's src only use the left 16 bits,
2038 // but we still use vsfrc for it.
2039 def XSCVDPHP : XX2_XT6_XO5_XB6<60, 17, 347, "xscvdphp", vsfrc, []>;
2040 def XSCVHPDP : XX2_XT6_XO5_XB6<60, 16, 347, "xscvhpdp", vsfrc, []>;
2043 def XVCVHPSP : XX2_XT6_XO5_XB6<60, 24, 475, "xvcvhpsp", vsrc, []>;
2044 def XVCVSPHP : XX2_XT6_XO5_XB6<60, 25, 475, "xvcvsphp", vsrc, []>;
2046 class Z23_VT5_R1_VB5_RMC2_EX1<bits<6> opcode, bits<8> xo, bit ex, string opc,
2048 : Z23Form_1<opcode, xo,
2049 (outs vrrc:$vT), (ins u1imm:$r, vrrc:$vB, u2imm:$rmc),
2050 !strconcat(opc, " $r, $vT, $vB, $rmc"), IIC_VecFP, pattern> {
2054 // Round to Quad-Precision Integer [with Inexact]
2055 def XSRQPI : Z23_VT5_R1_VB5_RMC2_EX1<63, 5, 0, "xsrqpi" , []>;
2056 def XSRQPIX : Z23_VT5_R1_VB5_RMC2_EX1<63, 5, 1, "xsrqpix", []>;
2058 // Round Quad-Precision to Double-Extended Precision (fp80)
2059 def XSRQPXP : Z23_VT5_R1_VB5_RMC2_EX1<63, 37, 0, "xsrqpxp", []>;
2061 //===--------------------------------------------------------------------===//
2062 // Insert/Extract Instructions
2064 // Insert Exponent DP/QP
2065 // XT NOTE: XT.dword[1] = 0xUUUU_UUUU_UUUU_UUUU
2066 def XSIEXPDP : XX1Form <60, 918, (outs vsrc:$XT), (ins g8rc:$rA, g8rc:$rB),
2067 "xsiexpdp $XT, $rA, $rB", IIC_VecFP, []>;
2068 // vB NOTE: only vB.dword[0] is used, that's why we don't use
2069 // X_VT5_VA5_VB5 form
2070 def XSIEXPQP : XForm_18<63, 868, (outs vrrc:$vT), (ins vrrc:$vA, vsfrc:$vB),
2071 "xsiexpqp $vT, $vA, $vB", IIC_VecFP, []>;
2073 // Extract Exponent/Significand DP/QP
2074 def XSXEXPDP : XX2_RT5_XO5_XB6<60, 0, 347, "xsxexpdp", []>;
2075 def XSXSIGDP : XX2_RT5_XO5_XB6<60, 1, 347, "xsxsigdp", []>;
2076 def XSXEXPQP : X_VT5_XO5_VB5 <63, 2, 804, "xsxexpqp", []>;
2077 def XSXSIGQP : X_VT5_XO5_VB5 <63, 18, 804, "xsxsigqp", []>;
2079 // Vector Insert Word
2080 // XB NOTE: Only XB.dword[1] is used, but we use vsrc on XB.
2082 XX2_RD6_UIM5_RS6<60, 181, (outs vsrc:$XT),
2083 (ins vsrc:$XTi, vsrc:$XB, u4imm:$UIM),
2084 "xxinsertw $XT, $XB, $UIM", IIC_VecFP,
2085 [(set v4i32:$XT, (PPCxxinsert v4i32:$XTi, v4i32:$XB,
2086 imm32SExt16:$UIM))]>,
2087 RegConstraint<"$XTi = $XT">, NoEncode<"$XTi">;
2089 // Vector Extract Unsigned Word
2090 def XXEXTRACTUW : XX2_RD6_UIM5_RS6<60, 165,
2091 (outs vsfrc:$XT), (ins vsrc:$XB, u4imm:$UIMM),
2092 "xxextractuw $XT, $XB, $UIMM", IIC_VecFP, []>;
2094 // Vector Insert Exponent DP/SP
2095 def XVIEXPDP : XX3_XT5_XA5_XB5<60, 248, "xviexpdp", vsrc, vsrc, vsrc,
2097 def XVIEXPSP : XX3_XT5_XA5_XB5<60, 216, "xviexpsp", vsrc, vsrc, vsrc,
2100 // Vector Extract Exponent/Significand DP/SP
2101 def XVXEXPDP : XX2_XT6_XO5_XB6<60, 0, 475, "xvxexpdp", vsrc, []>;
2102 def XVXEXPSP : XX2_XT6_XO5_XB6<60, 8, 475, "xvxexpsp", vsrc, []>;
2103 def XVXSIGDP : XX2_XT6_XO5_XB6<60, 1, 475, "xvxsigdp", vsrc, []>;
2104 def XVXSIGSP : XX2_XT6_XO5_XB6<60, 9, 475, "xvxsigsp", vsrc, []>;
2106 //===--------------------------------------------------------------------===//
2108 // Test Data Class SP/DP/QP
2109 def XSTSTDCSP : XX2_BF3_DCMX7_RS6<60, 298,
2110 (outs crrc:$BF), (ins u7imm:$DCMX, vsfrc:$XB),
2111 "xststdcsp $BF, $XB, $DCMX", IIC_VecFP, []>;
2112 def XSTSTDCDP : XX2_BF3_DCMX7_RS6<60, 362,
2113 (outs crrc:$BF), (ins u7imm:$DCMX, vsfrc:$XB),
2114 "xststdcdp $BF, $XB, $DCMX", IIC_VecFP, []>;
2115 def XSTSTDCQP : X_BF3_DCMX7_RS5 <63, 708,
2116 (outs crrc:$BF), (ins u7imm:$DCMX, vrrc:$vB),
2117 "xststdcqp $BF, $vB, $DCMX", IIC_VecFP, []>;
2119 // Vector Test Data Class SP/DP
2120 def XVTSTDCSP : XX2_RD6_DCMX7_RS6<60, 13, 5,
2121 (outs vsrc:$XT), (ins u7imm:$DCMX, vsrc:$XB),
2122 "xvtstdcsp $XT, $XB, $DCMX", IIC_VecFP, []>;
2123 def XVTSTDCDP : XX2_RD6_DCMX7_RS6<60, 15, 5,
2124 (outs vsrc:$XT), (ins u7imm:$DCMX, vsrc:$XB),
2125 "xvtstdcdp $XT, $XB, $DCMX", IIC_VecFP, []>;
2127 //===--------------------------------------------------------------------===//
2129 // Maximum/Minimum Type-C/Type-J DP
2130 // XT.dword[1] = 0xUUUU_UUUU_UUUU_UUUU, so we use vsrc for XT
2131 def XSMAXCDP : XX3_XT5_XA5_XB5<60, 128, "xsmaxcdp", vsrc, vsfrc, vsfrc,
2133 def XSMAXJDP : XX3_XT5_XA5_XB5<60, 144, "xsmaxjdp", vsrc, vsfrc, vsfrc,
2135 def XSMINCDP : XX3_XT5_XA5_XB5<60, 136, "xsmincdp", vsrc, vsfrc, vsfrc,
2137 def XSMINJDP : XX3_XT5_XA5_XB5<60, 152, "xsminjdp", vsrc, vsfrc, vsfrc,
2140 //===--------------------------------------------------------------------===//
2142 // Vector Byte-Reverse H/W/D/Q Word
2143 def XXBRH : XX2_XT6_XO5_XB6<60, 7, 475, "xxbrh", vsrc, []>;
2144 def XXBRW : XX2_XT6_XO5_XB6<60, 15, 475, "xxbrw", vsrc, []>;
2145 def XXBRD : XX2_XT6_XO5_XB6<60, 23, 475, "xxbrd", vsrc, []>;
2146 def XXBRQ : XX2_XT6_XO5_XB6<60, 31, 475, "xxbrq", vsrc, []>;
2149 def XXPERM : XX3_XT5_XA5_XB5<60, 26, "xxperm" , vsrc, vsrc, vsrc,
2151 def XXPERMR : XX3_XT5_XA5_XB5<60, 58, "xxpermr", vsrc, vsrc, vsrc,
2154 // Vector Splat Immediate Byte
2155 def XXSPLTIB : X_RD6_IMM8<60, 360, (outs vsrc:$XT), (ins u8imm:$IMM8),
2156 "xxspltib $XT, $IMM8", IIC_VecPerm, []>;
2158 //===--------------------------------------------------------------------===//
2159 // Vector/Scalar Load/Store Instructions
2161 let mayLoad = 1 in {
2163 def LXV : DQ_RD6_RS5_DQ12<61, 1, (outs vsrc:$XT), (ins memrix16:$src),
2164 "lxv $XT, $src", IIC_LdStLFD, []>;
2166 def LXSD : DSForm_1<57, 2, (outs vrrc:$vD), (ins memrix:$src),
2167 "lxsd $vD, $src", IIC_LdStLFD, []>;
2168 // Load SP from src, convert it to DP, and place in dword[0]
2169 def LXSSP : DSForm_1<57, 3, (outs vrrc:$vD), (ins memrix:$src),
2170 "lxssp $vD, $src", IIC_LdStLFD, []>;
2172 // [PO T RA RB XO TX] almost equal to [PO S RA RB XO SX], but has different
2173 // "out" and "in" dag
2174 class X_XT6_RA5_RB5<bits<6> opcode, bits<10> xo, string opc,
2175 RegisterOperand vtype, list<dag> pattern>
2176 : XX1Form<opcode, xo, (outs vtype:$XT), (ins memrr:$src),
2177 !strconcat(opc, " $XT, $src"), IIC_LdStLFD, pattern>;
2179 // Load as Integer Byte/Halfword & Zero Indexed
2180 def LXSIBZX : X_XT6_RA5_RB5<31, 781, "lxsibzx", vsfrc, []>;
2181 def LXSIHZX : X_XT6_RA5_RB5<31, 813, "lxsihzx", vsfrc, []>;
2183 // Load Vector Halfword*8/Byte*16 Indexed
2184 def LXVH8X : X_XT6_RA5_RB5<31, 812, "lxvh8x" , vsrc, []>;
2185 def LXVB16X : X_XT6_RA5_RB5<31, 876, "lxvb16x", vsrc, []>;
2187 // Load Vector Indexed
2188 def LXVX : X_XT6_RA5_RB5<31, 268, "lxvx" , vsrc, []>;
2190 // Load Vector (Left-justified) with Length
2191 def LXVL : X_XT6_RA5_RB5<31, 269, "lxvl" , vsrc, []>;
2192 def LXVLL : X_XT6_RA5_RB5<31, 301, "lxvll" , vsrc, []>;
2194 // Load Vector Word & Splat Indexed
2195 def LXVWSX : X_XT6_RA5_RB5<31, 364, "lxvwsx" , vsrc, []>;
2198 let mayStore = 1 in {
2200 def STXV : DQ_RD6_RS5_DQ12<61, 5, (outs), (ins vsrc:$XT, memrix16:$dst),
2201 "stxv $XT, $dst", IIC_LdStSTFD, []>;
2203 def STXSD : DSForm_1<61, 2, (outs), (ins vrrc:$vS, memrix:$dst),
2204 "stxsd $vS, $dst", IIC_LdStSTFD, []>;
2205 // Convert DP of dword[0] to SP, and Store to dst
2206 def STXSSP : DSForm_1<61, 3, (outs), (ins vrrc:$vS, memrix:$dst),
2207 "stxssp $vS, $dst", IIC_LdStSTFD, []>;
2209 // [PO S RA RB XO SX]
2210 class X_XS6_RA5_RB5<bits<6> opcode, bits<10> xo, string opc,
2211 RegisterOperand vtype, list<dag> pattern>
2212 : XX1Form<opcode, xo, (outs), (ins vtype:$XT, memrr:$dst),
2213 !strconcat(opc, " $XT, $dst"), IIC_LdStSTFD, pattern>;
2215 // Store as Integer Byte/Halfword Indexed
2216 def STXSIBX : X_XS6_RA5_RB5<31, 909, "stxsibx" , vsfrc, []>;
2217 def STXSIHX : X_XS6_RA5_RB5<31, 941, "stxsihx" , vsfrc, []>;
2219 // Store Vector Halfword*8/Byte*16 Indexed
2220 def STXVH8X : X_XS6_RA5_RB5<31, 940, "stxvh8x" , vsrc, []>;
2221 def STXVB16X : X_XS6_RA5_RB5<31, 1004, "stxvb16x", vsrc, []>;
2223 // Store Vector Indexed
2224 def STXVX : X_XS6_RA5_RB5<31, 396, "stxvx" , vsrc, []>;
2226 // Store Vector (Left-justified) with Length
2227 def STXVL : X_XS6_RA5_RB5<31, 397, "stxvl" , vsrc, []>;
2228 def STXVLL : X_XS6_RA5_RB5<31, 429, "stxvll" , vsrc, []>;
2231 // Patterns for which instructions from ISA 3.0 are a better match
2232 let Predicates = [IsLittleEndian, HasP9Vector] in {
2233 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 0))))),
2234 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 12)))>;
2235 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 1))))),
2236 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 8)))>;
2237 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 2))))),
2238 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 4)))>;
2239 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 3))))),
2240 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 0)))>;
2241 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 0)),
2242 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 12))>;
2243 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 1)),
2244 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 8))>;
2245 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 2)),
2246 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 4))>;
2247 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 3)),
2248 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 0))>;
2249 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 0)),
2250 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 12))>;
2251 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 1)),
2252 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 8))>;
2253 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 2)),
2254 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 4))>;
2255 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 3)),
2256 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 0))>;
2257 } // IsLittleEndian, HasP9Vector
2259 let Predicates = [IsBigEndian, HasP9Vector] in {
2260 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 0))))),
2261 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 0)))>;
2262 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 1))))),
2263 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 4)))>;
2264 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 2))))),
2265 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 8)))>;
2266 def : Pat<(f32 (PPCfcfidus (PPCmtvsrz (i32 (extractelt v4i32:$A, 3))))),
2267 (f32 (XSCVUXDSP (XXEXTRACTUW $A, 12)))>;
2268 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 0)),
2269 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 0))>;
2270 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 1)),
2271 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 4))>;
2272 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 2)),
2273 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 8))>;
2274 def : Pat<(v4i32 (insertelt v4i32:$A, i32:$B, 3)),
2275 (v4i32 (XXINSERTW v4i32:$A, AlignValues.I32_TO_BE_WORD1, 12))>;
2276 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 0)),
2277 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 0))>;
2278 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 1)),
2279 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 4))>;
2280 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 2)),
2281 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 8))>;
2282 def : Pat<(v4f32 (insertelt v4f32:$A, f32:$B, 3)),
2283 (v4f32 (XXINSERTW v4f32:$A, AlignValues.F32_TO_BE_WORD1, 12))>;
2284 } // IsLittleEndian, HasP9Vector
2285 } // end HasP9Vector, AddedComplexity