1 //===-- PPCInstrAltivec.td - The PowerPC Altivec Extension -*- tablegen -*-===//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 // This file describes the Altivec extension to the PowerPC instruction set.
11 //===----------------------------------------------------------------------===//
13 // *********************************** NOTE ***********************************
14 // ** For POWER8 Little Endian, the VSX swap optimization relies on knowing **
15 // ** which VMX and VSX instructions are lane-sensitive and which are not. **
16 // ** A lane-sensitive instruction relies, implicitly or explicitly, on **
17 // ** whether lanes are numbered from left to right. An instruction like **
18 // ** VADDFP is not lane-sensitive, because each lane of the result vector **
19 // ** relies only on the corresponding lane of the source vectors. However, **
20 // ** an instruction like VMULESB is lane-sensitive, because "even" and **
21 // ** "odd" lanes are different for big-endian and little-endian numbering. **
23 // ** When adding new VMX and VSX instructions, please consider whether they **
24 // ** are lane-sensitive. If so, they must be added to a switch statement **
25 // ** in PPCVSXSwapRemoval::gatherVectorInstructions(). **
26 // ****************************************************************************
29 //===----------------------------------------------------------------------===//
30 // Altivec transformation functions and pattern fragments.
33 // Since we canonicalize buildvectors to v16i8, all vnots "-1" operands will be
35 def vnot_ppc : PatFrag<(ops node:$in),
36 (xor node:$in, (bitconvert (v16i8 immAllOnesV)))>;
38 def vpkuhum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
39 (vector_shuffle node:$lhs, node:$rhs), [{
40 return PPC::isVPKUHUMShuffleMask(cast<ShuffleVectorSDNode>(N), 0, *CurDAG);
42 def vpkuwum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
43 (vector_shuffle node:$lhs, node:$rhs), [{
44 return PPC::isVPKUWUMShuffleMask(cast<ShuffleVectorSDNode>(N), 0, *CurDAG);
46 def vpkudum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
47 (vector_shuffle node:$lhs, node:$rhs), [{
48 return PPC::isVPKUDUMShuffleMask(cast<ShuffleVectorSDNode>(N), 0, *CurDAG);
50 def vpkuhum_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
51 (vector_shuffle node:$lhs, node:$rhs), [{
52 return PPC::isVPKUHUMShuffleMask(cast<ShuffleVectorSDNode>(N), 1, *CurDAG);
54 def vpkuwum_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
55 (vector_shuffle node:$lhs, node:$rhs), [{
56 return PPC::isVPKUWUMShuffleMask(cast<ShuffleVectorSDNode>(N), 1, *CurDAG);
58 def vpkudum_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
59 (vector_shuffle node:$lhs, node:$rhs), [{
60 return PPC::isVPKUDUMShuffleMask(cast<ShuffleVectorSDNode>(N), 1, *CurDAG);
63 // These fragments are provided for little-endian, where the inputs must be
64 // swapped for correct semantics.
65 def vpkuhum_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
66 (vector_shuffle node:$lhs, node:$rhs), [{
67 return PPC::isVPKUHUMShuffleMask(cast<ShuffleVectorSDNode>(N), 2, *CurDAG);
69 def vpkuwum_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
70 (vector_shuffle node:$lhs, node:$rhs), [{
71 return PPC::isVPKUWUMShuffleMask(cast<ShuffleVectorSDNode>(N), 2, *CurDAG);
73 def vpkudum_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
74 (vector_shuffle node:$lhs, node:$rhs), [{
75 return PPC::isVPKUDUMShuffleMask(cast<ShuffleVectorSDNode>(N), 2, *CurDAG);
78 def vmrglb_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
79 (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
80 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 1, 0, *CurDAG);
82 def vmrglh_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
83 (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
84 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 2, 0, *CurDAG);
86 def vmrglw_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
87 (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
88 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 4, 0, *CurDAG);
90 def vmrghb_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
91 (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
92 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 1, 0, *CurDAG);
94 def vmrghh_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
95 (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
96 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 2, 0, *CurDAG);
98 def vmrghw_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
99 (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
100 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 4, 0, *CurDAG);
104 def vmrglb_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
105 (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
106 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 1, 1, *CurDAG);
108 def vmrglh_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
109 (vector_shuffle node:$lhs, node:$rhs), [{
110 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 2, 1, *CurDAG);
112 def vmrglw_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
113 (vector_shuffle node:$lhs, node:$rhs), [{
114 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 4, 1, *CurDAG);
116 def vmrghb_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
117 (vector_shuffle node:$lhs, node:$rhs), [{
118 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 1, 1, *CurDAG);
120 def vmrghh_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
121 (vector_shuffle node:$lhs, node:$rhs), [{
122 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 2, 1, *CurDAG);
124 def vmrghw_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
125 (vector_shuffle node:$lhs, node:$rhs), [{
126 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 4, 1, *CurDAG);
130 // These fragments are provided for little-endian, where the inputs must be
131 // swapped for correct semantics.
132 def vmrglb_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
133 (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
134 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 1, 2, *CurDAG);
136 def vmrglh_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
137 (vector_shuffle node:$lhs, node:$rhs), [{
138 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 2, 2, *CurDAG);
140 def vmrglw_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
141 (vector_shuffle node:$lhs, node:$rhs), [{
142 return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 4, 2, *CurDAG);
144 def vmrghb_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
145 (vector_shuffle node:$lhs, node:$rhs), [{
146 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 1, 2, *CurDAG);
148 def vmrghh_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
149 (vector_shuffle node:$lhs, node:$rhs), [{
150 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 2, 2, *CurDAG);
152 def vmrghw_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
153 (vector_shuffle node:$lhs, node:$rhs), [{
154 return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 4, 2, *CurDAG);
158 def vmrgew_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
159 (vector_shuffle node:$lhs, node:$rhs), [{
160 return PPC::isVMRGEOShuffleMask(cast<ShuffleVectorSDNode>(N), true, 0, *CurDAG);
162 def vmrgow_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
163 (vector_shuffle node:$lhs, node:$rhs), [{
164 return PPC::isVMRGEOShuffleMask(cast<ShuffleVectorSDNode>(N), false, 0, *CurDAG);
166 def vmrgew_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
167 (vector_shuffle node:$lhs, node:$rhs), [{
168 return PPC::isVMRGEOShuffleMask(cast<ShuffleVectorSDNode>(N), true, 1, *CurDAG);
170 def vmrgow_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
171 (vector_shuffle node:$lhs, node:$rhs), [{
172 return PPC::isVMRGEOShuffleMask(cast<ShuffleVectorSDNode>(N), false, 1, *CurDAG);
174 def vmrgew_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
175 (vector_shuffle node:$lhs, node:$rhs), [{
176 return PPC::isVMRGEOShuffleMask(cast<ShuffleVectorSDNode>(N), true, 2, *CurDAG);
178 def vmrgow_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
179 (vector_shuffle node:$lhs, node:$rhs), [{
180 return PPC::isVMRGEOShuffleMask(cast<ShuffleVectorSDNode>(N), false, 2, *CurDAG);
185 def VSLDOI_get_imm : SDNodeXForm<vector_shuffle, [{
186 return getI32Imm(PPC::isVSLDOIShuffleMask(N, 0, *CurDAG), SDLoc(N));
188 def vsldoi_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
189 (vector_shuffle node:$lhs, node:$rhs), [{
190 return PPC::isVSLDOIShuffleMask(N, 0, *CurDAG) != -1;
194 /// VSLDOI_unary* - These are used to match vsldoi(X,X), which is turned into
195 /// vector_shuffle(X,undef,mask) by the dag combiner.
196 def VSLDOI_unary_get_imm : SDNodeXForm<vector_shuffle, [{
197 return getI32Imm(PPC::isVSLDOIShuffleMask(N, 1, *CurDAG), SDLoc(N));
199 def vsldoi_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
200 (vector_shuffle node:$lhs, node:$rhs), [{
201 return PPC::isVSLDOIShuffleMask(N, 1, *CurDAG) != -1;
202 }], VSLDOI_unary_get_imm>;
205 /// VSLDOI_swapped* - These fragments are provided for little-endian, where
206 /// the inputs must be swapped for correct semantics.
207 def VSLDOI_swapped_get_imm : SDNodeXForm<vector_shuffle, [{
208 return getI32Imm(PPC::isVSLDOIShuffleMask(N, 2, *CurDAG), SDLoc(N));
210 def vsldoi_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
211 (vector_shuffle node:$lhs, node:$rhs), [{
212 return PPC::isVSLDOIShuffleMask(N, 2, *CurDAG) != -1;
216 // VSPLT*_get_imm xform function: convert vector_shuffle mask to VSPLT* imm.
217 def VSPLTB_get_imm : SDNodeXForm<vector_shuffle, [{
218 return getI32Imm(PPC::getSplatIdxForPPCMnemonics(N, 1, *CurDAG), SDLoc(N));
220 def vspltb_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
221 (vector_shuffle node:$lhs, node:$rhs), [{
222 return PPC::isSplatShuffleMask(cast<ShuffleVectorSDNode>(N), 1);
224 def VSPLTH_get_imm : SDNodeXForm<vector_shuffle, [{
225 return getI32Imm(PPC::getSplatIdxForPPCMnemonics(N, 2, *CurDAG), SDLoc(N));
227 def vsplth_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
228 (vector_shuffle node:$lhs, node:$rhs), [{
229 return PPC::isSplatShuffleMask(cast<ShuffleVectorSDNode>(N), 2);
231 def VSPLTW_get_imm : SDNodeXForm<vector_shuffle, [{
232 return getI32Imm(PPC::getSplatIdxForPPCMnemonics(N, 4, *CurDAG), SDLoc(N));
234 def vspltw_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
235 (vector_shuffle node:$lhs, node:$rhs), [{
236 return PPC::isSplatShuffleMask(cast<ShuffleVectorSDNode>(N), 4);
240 // VSPLTISB_get_imm xform function: convert build_vector to VSPLTISB imm.
241 def VSPLTISB_get_imm : SDNodeXForm<build_vector, [{
242 return PPC::get_VSPLTI_elt(N, 1, *CurDAG);
244 def vecspltisb : PatLeaf<(build_vector), [{
245 return PPC::get_VSPLTI_elt(N, 1, *CurDAG).getNode() != nullptr;
246 }], VSPLTISB_get_imm>;
248 // VSPLTISH_get_imm xform function: convert build_vector to VSPLTISH imm.
249 def VSPLTISH_get_imm : SDNodeXForm<build_vector, [{
250 return PPC::get_VSPLTI_elt(N, 2, *CurDAG);
252 def vecspltish : PatLeaf<(build_vector), [{
253 return PPC::get_VSPLTI_elt(N, 2, *CurDAG).getNode() != nullptr;
254 }], VSPLTISH_get_imm>;
256 // VSPLTISW_get_imm xform function: convert build_vector to VSPLTISW imm.
257 def VSPLTISW_get_imm : SDNodeXForm<build_vector, [{
258 return PPC::get_VSPLTI_elt(N, 4, *CurDAG);
260 def vecspltisw : PatLeaf<(build_vector), [{
261 return PPC::get_VSPLTI_elt(N, 4, *CurDAG).getNode() != nullptr;
262 }], VSPLTISW_get_imm>;
264 //===----------------------------------------------------------------------===//
265 // Helpers for defining instructions that directly correspond to intrinsics.
267 // VA1a_Int_Ty - A VAForm_1a intrinsic definition of specific type.
268 class VA1a_Int_Ty<bits<6> xo, string opc, Intrinsic IntID, ValueType Ty>
269 : VAForm_1a<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
270 !strconcat(opc, " $vD, $vA, $vB, $vC"), IIC_VecFP,
271 [(set Ty:$vD, (IntID Ty:$vA, Ty:$vB, Ty:$vC))]>;
273 // VA1a_Int_Ty2 - A VAForm_1a intrinsic definition where the type of the
274 // inputs doesn't match the type of the output.
275 class VA1a_Int_Ty2<bits<6> xo, string opc, Intrinsic IntID, ValueType OutTy,
277 : VAForm_1a<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
278 !strconcat(opc, " $vD, $vA, $vB, $vC"), IIC_VecFP,
279 [(set OutTy:$vD, (IntID InTy:$vA, InTy:$vB, InTy:$vC))]>;
281 // VA1a_Int_Ty3 - A VAForm_1a intrinsic definition where there are two
282 // input types and an output type.
283 class VA1a_Int_Ty3<bits<6> xo, string opc, Intrinsic IntID, ValueType OutTy,
284 ValueType In1Ty, ValueType In2Ty>
285 : VAForm_1a<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
286 !strconcat(opc, " $vD, $vA, $vB, $vC"), IIC_VecFP,
288 (IntID In1Ty:$vA, In1Ty:$vB, In2Ty:$vC))]>;
290 // VX1_Int_Ty - A VXForm_1 intrinsic definition of specific type.
291 class VX1_Int_Ty<bits<11> xo, string opc, Intrinsic IntID, ValueType Ty>
292 : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
293 !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP,
294 [(set Ty:$vD, (IntID Ty:$vA, Ty:$vB))]>;
296 // VX1_Int_Ty2 - A VXForm_1 intrinsic definition where the type of the
297 // inputs doesn't match the type of the output.
298 class VX1_Int_Ty2<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
300 : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
301 !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP,
302 [(set OutTy:$vD, (IntID InTy:$vA, InTy:$vB))]>;
304 // VX1_Int_Ty3 - A VXForm_1 intrinsic definition where there are two
305 // input types and an output type.
306 class VX1_Int_Ty3<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
307 ValueType In1Ty, ValueType In2Ty>
308 : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
309 !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP,
310 [(set OutTy:$vD, (IntID In1Ty:$vA, In2Ty:$vB))]>;
312 // VX2_Int_SP - A VXForm_2 intrinsic definition of vector single-precision type.
313 class VX2_Int_SP<bits<11> xo, string opc, Intrinsic IntID>
314 : VXForm_2<xo, (outs vrrc:$vD), (ins vrrc:$vB),
315 !strconcat(opc, " $vD, $vB"), IIC_VecFP,
316 [(set v4f32:$vD, (IntID v4f32:$vB))]>;
318 // VX2_Int_Ty2 - A VXForm_2 intrinsic definition where the type of the
319 // inputs doesn't match the type of the output.
320 class VX2_Int_Ty2<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
322 : VXForm_2<xo, (outs vrrc:$vD), (ins vrrc:$vB),
323 !strconcat(opc, " $vD, $vB"), IIC_VecFP,
324 [(set OutTy:$vD, (IntID InTy:$vB))]>;
326 class VXBX_Int_Ty<bits<11> xo, string opc, Intrinsic IntID, ValueType Ty>
327 : VXForm_BX<xo, (outs vrrc:$vD), (ins vrrc:$vA),
328 !strconcat(opc, " $vD, $vA"), IIC_VecFP,
329 [(set Ty:$vD, (IntID Ty:$vA))]>;
331 class VXCR_Int_Ty<bits<11> xo, string opc, Intrinsic IntID, ValueType Ty>
332 : VXForm_CR<xo, (outs vrrc:$vD), (ins vrrc:$vA, u1imm:$ST, u4imm:$SIX),
333 !strconcat(opc, " $vD, $vA, $ST, $SIX"), IIC_VecFP,
334 [(set Ty:$vD, (IntID Ty:$vA, timm:$ST, timm:$SIX))]>;
336 //===----------------------------------------------------------------------===//
337 // Instruction Definitions.
339 def HasAltivec : Predicate<"PPCSubTarget->hasAltivec()">;
340 let Predicates = [HasAltivec] in {
342 def DSS : DSS_Form<0, 822, (outs), (ins u5imm:$STRM),
343 "dss $STRM", IIC_LdStLoad /*FIXME*/, [(int_ppc_altivec_dss imm:$STRM)]>,
344 Deprecated<DeprecatedDST> {
349 def DSSALL : DSS_Form<1, 822, (outs), (ins),
350 "dssall", IIC_LdStLoad /*FIXME*/, [(int_ppc_altivec_dssall)]>,
351 Deprecated<DeprecatedDST> {
357 def DST : DSS_Form<0, 342, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB),
358 "dst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
359 [(int_ppc_altivec_dst i32:$rA, i32:$rB, imm:$STRM)]>,
360 Deprecated<DeprecatedDST>;
362 def DSTT : DSS_Form<1, 342, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB),
363 "dstt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
364 [(int_ppc_altivec_dstt i32:$rA, i32:$rB, imm:$STRM)]>,
365 Deprecated<DeprecatedDST>;
367 def DSTST : DSS_Form<0, 374, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB),
368 "dstst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
369 [(int_ppc_altivec_dstst i32:$rA, i32:$rB, imm:$STRM)]>,
370 Deprecated<DeprecatedDST>;
372 def DSTSTT : DSS_Form<1, 374, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB),
373 "dststt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
374 [(int_ppc_altivec_dststt i32:$rA, i32:$rB, imm:$STRM)]>,
375 Deprecated<DeprecatedDST>;
377 let isCodeGenOnly = 1 in {
378 // The very same instructions as above, but formally matching 64bit registers.
379 def DST64 : DSS_Form<0, 342, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB),
380 "dst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
381 [(int_ppc_altivec_dst i64:$rA, i32:$rB, imm:$STRM)]>,
382 Deprecated<DeprecatedDST>;
384 def DSTT64 : DSS_Form<1, 342, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB),
385 "dstt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
386 [(int_ppc_altivec_dstt i64:$rA, i32:$rB, imm:$STRM)]>,
387 Deprecated<DeprecatedDST>;
389 def DSTST64 : DSS_Form<0, 374, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB),
390 "dstst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
391 [(int_ppc_altivec_dstst i64:$rA, i32:$rB,
393 Deprecated<DeprecatedDST>;
395 def DSTSTT64 : DSS_Form<1, 374, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB),
396 "dststt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
397 [(int_ppc_altivec_dststt i64:$rA, i32:$rB,
399 Deprecated<DeprecatedDST>;
402 def MFVSCR : VXForm_4<1540, (outs vrrc:$vD), (ins),
403 "mfvscr $vD", IIC_LdStStore,
404 [(set v8i16:$vD, (int_ppc_altivec_mfvscr))]>;
405 def MTVSCR : VXForm_5<1604, (outs), (ins vrrc:$vB),
406 "mtvscr $vB", IIC_LdStLoad,
407 [(int_ppc_altivec_mtvscr v4i32:$vB)]>;
409 let PPC970_Unit = 2, mayLoad = 1, mayStore = 0 in { // Loads.
410 def LVEBX: XForm_1_memOp<31, 7, (outs vrrc:$vD), (ins memrr:$src),
411 "lvebx $vD, $src", IIC_LdStLoad,
412 [(set v16i8:$vD, (int_ppc_altivec_lvebx xoaddr:$src))]>;
413 def LVEHX: XForm_1_memOp<31, 39, (outs vrrc:$vD), (ins memrr:$src),
414 "lvehx $vD, $src", IIC_LdStLoad,
415 [(set v8i16:$vD, (int_ppc_altivec_lvehx xoaddr:$src))]>;
416 def LVEWX: XForm_1_memOp<31, 71, (outs vrrc:$vD), (ins memrr:$src),
417 "lvewx $vD, $src", IIC_LdStLoad,
418 [(set v4i32:$vD, (int_ppc_altivec_lvewx xoaddr:$src))]>;
419 def LVX : XForm_1_memOp<31, 103, (outs vrrc:$vD), (ins memrr:$src),
420 "lvx $vD, $src", IIC_LdStLoad,
421 [(set v4i32:$vD, (int_ppc_altivec_lvx xoaddr:$src))]>;
422 def LVXL : XForm_1_memOp<31, 359, (outs vrrc:$vD), (ins memrr:$src),
423 "lvxl $vD, $src", IIC_LdStLoad,
424 [(set v4i32:$vD, (int_ppc_altivec_lvxl xoaddr:$src))]>;
427 def LVSL : XForm_1_memOp<31, 6, (outs vrrc:$vD), (ins memrr:$src),
428 "lvsl $vD, $src", IIC_LdStLoad,
429 [(set v16i8:$vD, (int_ppc_altivec_lvsl xoaddr:$src))]>,
431 def LVSR : XForm_1_memOp<31, 38, (outs vrrc:$vD), (ins memrr:$src),
432 "lvsr $vD, $src", IIC_LdStLoad,
433 [(set v16i8:$vD, (int_ppc_altivec_lvsr xoaddr:$src))]>,
436 let PPC970_Unit = 2, mayStore = 1, mayLoad = 0 in { // Stores.
437 def STVEBX: XForm_8_memOp<31, 135, (outs), (ins vrrc:$rS, memrr:$dst),
438 "stvebx $rS, $dst", IIC_LdStStore,
439 [(int_ppc_altivec_stvebx v16i8:$rS, xoaddr:$dst)]>;
440 def STVEHX: XForm_8_memOp<31, 167, (outs), (ins vrrc:$rS, memrr:$dst),
441 "stvehx $rS, $dst", IIC_LdStStore,
442 [(int_ppc_altivec_stvehx v8i16:$rS, xoaddr:$dst)]>;
443 def STVEWX: XForm_8_memOp<31, 199, (outs), (ins vrrc:$rS, memrr:$dst),
444 "stvewx $rS, $dst", IIC_LdStStore,
445 [(int_ppc_altivec_stvewx v4i32:$rS, xoaddr:$dst)]>;
446 def STVX : XForm_8_memOp<31, 231, (outs), (ins vrrc:$rS, memrr:$dst),
447 "stvx $rS, $dst", IIC_LdStStore,
448 [(int_ppc_altivec_stvx v4i32:$rS, xoaddr:$dst)]>;
449 def STVXL : XForm_8_memOp<31, 487, (outs), (ins vrrc:$rS, memrr:$dst),
450 "stvxl $rS, $dst", IIC_LdStStore,
451 [(int_ppc_altivec_stvxl v4i32:$rS, xoaddr:$dst)]>;
454 let PPC970_Unit = 5 in { // VALU Operations.
455 // VA-Form instructions. 3-input AltiVec ops.
456 let isCommutable = 1 in {
457 def VMADDFP : VAForm_1<46, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vC, vrrc:$vB),
458 "vmaddfp $vD, $vA, $vC, $vB", IIC_VecFP,
460 (fma v4f32:$vA, v4f32:$vC, v4f32:$vB))]>;
462 // FIXME: The fma+fneg pattern won't match because fneg is not legal.
463 def VNMSUBFP: VAForm_1<47, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vC, vrrc:$vB),
464 "vnmsubfp $vD, $vA, $vC, $vB", IIC_VecFP,
465 [(set v4f32:$vD, (fneg (fma v4f32:$vA, v4f32:$vC,
466 (fneg v4f32:$vB))))]>;
468 def VMHADDSHS : VA1a_Int_Ty<32, "vmhaddshs", int_ppc_altivec_vmhaddshs, v8i16>;
469 def VMHRADDSHS : VA1a_Int_Ty<33, "vmhraddshs", int_ppc_altivec_vmhraddshs,
471 def VMLADDUHM : VA1a_Int_Ty<34, "vmladduhm", int_ppc_altivec_vmladduhm, v8i16>;
474 def VPERM : VA1a_Int_Ty3<43, "vperm", int_ppc_altivec_vperm,
475 v4i32, v4i32, v16i8>;
476 def VSEL : VA1a_Int_Ty<42, "vsel", int_ppc_altivec_vsel, v4i32>;
479 def VSLDOI : VAForm_2<44, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, u4imm:$SH),
480 "vsldoi $vD, $vA, $vB, $SH", IIC_VecFP,
482 (PPCvecshl v16i8:$vA, v16i8:$vB, imm32SExt16:$SH))]>;
484 // VX-Form instructions. AltiVec arithmetic ops.
485 let isCommutable = 1 in {
486 def VADDFP : VXForm_1<10, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
487 "vaddfp $vD, $vA, $vB", IIC_VecFP,
488 [(set v4f32:$vD, (fadd v4f32:$vA, v4f32:$vB))]>;
490 def VADDUBM : VXForm_1<0, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
491 "vaddubm $vD, $vA, $vB", IIC_VecGeneral,
492 [(set v16i8:$vD, (add v16i8:$vA, v16i8:$vB))]>;
493 def VADDUHM : VXForm_1<64, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
494 "vadduhm $vD, $vA, $vB", IIC_VecGeneral,
495 [(set v8i16:$vD, (add v8i16:$vA, v8i16:$vB))]>;
496 def VADDUWM : VXForm_1<128, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
497 "vadduwm $vD, $vA, $vB", IIC_VecGeneral,
498 [(set v4i32:$vD, (add v4i32:$vA, v4i32:$vB))]>;
500 def VADDCUW : VX1_Int_Ty<384, "vaddcuw", int_ppc_altivec_vaddcuw, v4i32>;
501 def VADDSBS : VX1_Int_Ty<768, "vaddsbs", int_ppc_altivec_vaddsbs, v16i8>;
502 def VADDSHS : VX1_Int_Ty<832, "vaddshs", int_ppc_altivec_vaddshs, v8i16>;
503 def VADDSWS : VX1_Int_Ty<896, "vaddsws", int_ppc_altivec_vaddsws, v4i32>;
504 def VADDUBS : VX1_Int_Ty<512, "vaddubs", int_ppc_altivec_vaddubs, v16i8>;
505 def VADDUHS : VX1_Int_Ty<576, "vadduhs", int_ppc_altivec_vadduhs, v8i16>;
506 def VADDUWS : VX1_Int_Ty<640, "vadduws", int_ppc_altivec_vadduws, v4i32>;
509 let isCommutable = 1 in
510 def VAND : VXForm_1<1028, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
511 "vand $vD, $vA, $vB", IIC_VecFP,
512 [(set v4i32:$vD, (and v4i32:$vA, v4i32:$vB))]>;
513 def VANDC : VXForm_1<1092, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
514 "vandc $vD, $vA, $vB", IIC_VecFP,
515 [(set v4i32:$vD, (and v4i32:$vA,
516 (vnot_ppc v4i32:$vB)))]>;
518 def VCFSX : VXForm_1<842, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
519 "vcfsx $vD, $vB, $UIMM", IIC_VecFP,
521 (int_ppc_altivec_vcfsx v4i32:$vB, imm:$UIMM))]>;
522 def VCFUX : VXForm_1<778, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
523 "vcfux $vD, $vB, $UIMM", IIC_VecFP,
525 (int_ppc_altivec_vcfux v4i32:$vB, imm:$UIMM))]>;
526 def VCTSXS : VXForm_1<970, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
527 "vctsxs $vD, $vB, $UIMM", IIC_VecFP,
529 (int_ppc_altivec_vctsxs v4f32:$vB, imm:$UIMM))]>;
530 def VCTUXS : VXForm_1<906, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
531 "vctuxs $vD, $vB, $UIMM", IIC_VecFP,
533 (int_ppc_altivec_vctuxs v4f32:$vB, imm:$UIMM))]>;
535 // Defines with the UIM field set to 0 for floating-point
536 // to integer (fp_to_sint/fp_to_uint) conversions and integer
537 // to floating-point (sint_to_fp/uint_to_fp) conversions.
538 let isCodeGenOnly = 1, VA = 0 in {
539 def VCFSX_0 : VXForm_1<842, (outs vrrc:$vD), (ins vrrc:$vB),
540 "vcfsx $vD, $vB, 0", IIC_VecFP,
542 (int_ppc_altivec_vcfsx v4i32:$vB, 0))]>;
543 def VCTUXS_0 : VXForm_1<906, (outs vrrc:$vD), (ins vrrc:$vB),
544 "vctuxs $vD, $vB, 0", IIC_VecFP,
546 (int_ppc_altivec_vctuxs v4f32:$vB, 0))]>;
547 def VCFUX_0 : VXForm_1<778, (outs vrrc:$vD), (ins vrrc:$vB),
548 "vcfux $vD, $vB, 0", IIC_VecFP,
550 (int_ppc_altivec_vcfux v4i32:$vB, 0))]>;
551 def VCTSXS_0 : VXForm_1<970, (outs vrrc:$vD), (ins vrrc:$vB),
552 "vctsxs $vD, $vB, 0", IIC_VecFP,
554 (int_ppc_altivec_vctsxs v4f32:$vB, 0))]>;
556 def VEXPTEFP : VX2_Int_SP<394, "vexptefp", int_ppc_altivec_vexptefp>;
557 def VLOGEFP : VX2_Int_SP<458, "vlogefp", int_ppc_altivec_vlogefp>;
559 let isCommutable = 1 in {
560 def VAVGSB : VX1_Int_Ty<1282, "vavgsb", int_ppc_altivec_vavgsb, v16i8>;
561 def VAVGSH : VX1_Int_Ty<1346, "vavgsh", int_ppc_altivec_vavgsh, v8i16>;
562 def VAVGSW : VX1_Int_Ty<1410, "vavgsw", int_ppc_altivec_vavgsw, v4i32>;
563 def VAVGUB : VX1_Int_Ty<1026, "vavgub", int_ppc_altivec_vavgub, v16i8>;
564 def VAVGUH : VX1_Int_Ty<1090, "vavguh", int_ppc_altivec_vavguh, v8i16>;
565 def VAVGUW : VX1_Int_Ty<1154, "vavguw", int_ppc_altivec_vavguw, v4i32>;
567 def VMAXFP : VX1_Int_Ty<1034, "vmaxfp", int_ppc_altivec_vmaxfp, v4f32>;
568 def VMAXSB : VX1_Int_Ty< 258, "vmaxsb", int_ppc_altivec_vmaxsb, v16i8>;
569 def VMAXSH : VX1_Int_Ty< 322, "vmaxsh", int_ppc_altivec_vmaxsh, v8i16>;
570 def VMAXSW : VX1_Int_Ty< 386, "vmaxsw", int_ppc_altivec_vmaxsw, v4i32>;
571 def VMAXUB : VX1_Int_Ty< 2, "vmaxub", int_ppc_altivec_vmaxub, v16i8>;
572 def VMAXUH : VX1_Int_Ty< 66, "vmaxuh", int_ppc_altivec_vmaxuh, v8i16>;
573 def VMAXUW : VX1_Int_Ty< 130, "vmaxuw", int_ppc_altivec_vmaxuw, v4i32>;
574 def VMINFP : VX1_Int_Ty<1098, "vminfp", int_ppc_altivec_vminfp, v4f32>;
575 def VMINSB : VX1_Int_Ty< 770, "vminsb", int_ppc_altivec_vminsb, v16i8>;
576 def VMINSH : VX1_Int_Ty< 834, "vminsh", int_ppc_altivec_vminsh, v8i16>;
577 def VMINSW : VX1_Int_Ty< 898, "vminsw", int_ppc_altivec_vminsw, v4i32>;
578 def VMINUB : VX1_Int_Ty< 514, "vminub", int_ppc_altivec_vminub, v16i8>;
579 def VMINUH : VX1_Int_Ty< 578, "vminuh", int_ppc_altivec_vminuh, v8i16>;
580 def VMINUW : VX1_Int_Ty< 642, "vminuw", int_ppc_altivec_vminuw, v4i32>;
583 def VMRGHB : VXForm_1< 12, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
584 "vmrghb $vD, $vA, $vB", IIC_VecFP,
585 [(set v16i8:$vD, (vmrghb_shuffle v16i8:$vA, v16i8:$vB))]>;
586 def VMRGHH : VXForm_1< 76, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
587 "vmrghh $vD, $vA, $vB", IIC_VecFP,
588 [(set v16i8:$vD, (vmrghh_shuffle v16i8:$vA, v16i8:$vB))]>;
589 def VMRGHW : VXForm_1<140, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
590 "vmrghw $vD, $vA, $vB", IIC_VecFP,
591 [(set v16i8:$vD, (vmrghw_shuffle v16i8:$vA, v16i8:$vB))]>;
592 def VMRGLB : VXForm_1<268, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
593 "vmrglb $vD, $vA, $vB", IIC_VecFP,
594 [(set v16i8:$vD, (vmrglb_shuffle v16i8:$vA, v16i8:$vB))]>;
595 def VMRGLH : VXForm_1<332, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
596 "vmrglh $vD, $vA, $vB", IIC_VecFP,
597 [(set v16i8:$vD, (vmrglh_shuffle v16i8:$vA, v16i8:$vB))]>;
598 def VMRGLW : VXForm_1<396, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
599 "vmrglw $vD, $vA, $vB", IIC_VecFP,
600 [(set v16i8:$vD, (vmrglw_shuffle v16i8:$vA, v16i8:$vB))]>;
602 def VMSUMMBM : VA1a_Int_Ty3<37, "vmsummbm", int_ppc_altivec_vmsummbm,
603 v4i32, v16i8, v4i32>;
604 def VMSUMSHM : VA1a_Int_Ty3<40, "vmsumshm", int_ppc_altivec_vmsumshm,
605 v4i32, v8i16, v4i32>;
606 def VMSUMSHS : VA1a_Int_Ty3<41, "vmsumshs", int_ppc_altivec_vmsumshs,
607 v4i32, v8i16, v4i32>;
608 def VMSUMUBM : VA1a_Int_Ty3<36, "vmsumubm", int_ppc_altivec_vmsumubm,
609 v4i32, v16i8, v4i32>;
610 def VMSUMUHM : VA1a_Int_Ty3<38, "vmsumuhm", int_ppc_altivec_vmsumuhm,
611 v4i32, v8i16, v4i32>;
612 def VMSUMUHS : VA1a_Int_Ty3<39, "vmsumuhs", int_ppc_altivec_vmsumuhs,
613 v4i32, v8i16, v4i32>;
615 let isCommutable = 1 in {
616 def VMULESB : VX1_Int_Ty2<776, "vmulesb", int_ppc_altivec_vmulesb,
618 def VMULESH : VX1_Int_Ty2<840, "vmulesh", int_ppc_altivec_vmulesh,
620 def VMULEUB : VX1_Int_Ty2<520, "vmuleub", int_ppc_altivec_vmuleub,
622 def VMULEUH : VX1_Int_Ty2<584, "vmuleuh", int_ppc_altivec_vmuleuh,
624 def VMULOSB : VX1_Int_Ty2<264, "vmulosb", int_ppc_altivec_vmulosb,
626 def VMULOSH : VX1_Int_Ty2<328, "vmulosh", int_ppc_altivec_vmulosh,
628 def VMULOUB : VX1_Int_Ty2< 8, "vmuloub", int_ppc_altivec_vmuloub,
630 def VMULOUH : VX1_Int_Ty2< 72, "vmulouh", int_ppc_altivec_vmulouh,
634 def VREFP : VX2_Int_SP<266, "vrefp", int_ppc_altivec_vrefp>;
635 def VRFIM : VX2_Int_SP<714, "vrfim", int_ppc_altivec_vrfim>;
636 def VRFIN : VX2_Int_SP<522, "vrfin", int_ppc_altivec_vrfin>;
637 def VRFIP : VX2_Int_SP<650, "vrfip", int_ppc_altivec_vrfip>;
638 def VRFIZ : VX2_Int_SP<586, "vrfiz", int_ppc_altivec_vrfiz>;
639 def VRSQRTEFP : VX2_Int_SP<330, "vrsqrtefp", int_ppc_altivec_vrsqrtefp>;
641 def VSUBCUW : VX1_Int_Ty<1408, "vsubcuw", int_ppc_altivec_vsubcuw, v4i32>;
643 def VSUBFP : VXForm_1<74, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
644 "vsubfp $vD, $vA, $vB", IIC_VecGeneral,
645 [(set v4f32:$vD, (fsub v4f32:$vA, v4f32:$vB))]>;
646 def VSUBUBM : VXForm_1<1024, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
647 "vsububm $vD, $vA, $vB", IIC_VecGeneral,
648 [(set v16i8:$vD, (sub v16i8:$vA, v16i8:$vB))]>;
649 def VSUBUHM : VXForm_1<1088, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
650 "vsubuhm $vD, $vA, $vB", IIC_VecGeneral,
651 [(set v8i16:$vD, (sub v8i16:$vA, v8i16:$vB))]>;
652 def VSUBUWM : VXForm_1<1152, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
653 "vsubuwm $vD, $vA, $vB", IIC_VecGeneral,
654 [(set v4i32:$vD, (sub v4i32:$vA, v4i32:$vB))]>;
656 def VSUBSBS : VX1_Int_Ty<1792, "vsubsbs" , int_ppc_altivec_vsubsbs, v16i8>;
657 def VSUBSHS : VX1_Int_Ty<1856, "vsubshs" , int_ppc_altivec_vsubshs, v8i16>;
658 def VSUBSWS : VX1_Int_Ty<1920, "vsubsws" , int_ppc_altivec_vsubsws, v4i32>;
659 def VSUBUBS : VX1_Int_Ty<1536, "vsububs" , int_ppc_altivec_vsububs, v16i8>;
660 def VSUBUHS : VX1_Int_Ty<1600, "vsubuhs" , int_ppc_altivec_vsubuhs, v8i16>;
661 def VSUBUWS : VX1_Int_Ty<1664, "vsubuws" , int_ppc_altivec_vsubuws, v4i32>;
663 def VSUMSWS : VX1_Int_Ty<1928, "vsumsws" , int_ppc_altivec_vsumsws, v4i32>;
664 def VSUM2SWS: VX1_Int_Ty<1672, "vsum2sws", int_ppc_altivec_vsum2sws, v4i32>;
666 def VSUM4SBS: VX1_Int_Ty3<1800, "vsum4sbs", int_ppc_altivec_vsum4sbs,
667 v4i32, v16i8, v4i32>;
668 def VSUM4SHS: VX1_Int_Ty3<1608, "vsum4shs", int_ppc_altivec_vsum4shs,
669 v4i32, v8i16, v4i32>;
670 def VSUM4UBS: VX1_Int_Ty3<1544, "vsum4ubs", int_ppc_altivec_vsum4ubs,
671 v4i32, v16i8, v4i32>;
673 def VNOR : VXForm_1<1284, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
674 "vnor $vD, $vA, $vB", IIC_VecFP,
675 [(set v4i32:$vD, (vnot_ppc (or v4i32:$vA,
677 let isCommutable = 1 in {
678 def VOR : VXForm_1<1156, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
679 "vor $vD, $vA, $vB", IIC_VecFP,
680 [(set v4i32:$vD, (or v4i32:$vA, v4i32:$vB))]>;
681 def VXOR : VXForm_1<1220, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
682 "vxor $vD, $vA, $vB", IIC_VecFP,
683 [(set v4i32:$vD, (xor v4i32:$vA, v4i32:$vB))]>;
686 def VRLB : VX1_Int_Ty< 4, "vrlb", int_ppc_altivec_vrlb, v16i8>;
687 def VRLH : VX1_Int_Ty< 68, "vrlh", int_ppc_altivec_vrlh, v8i16>;
688 def VRLW : VX1_Int_Ty< 132, "vrlw", int_ppc_altivec_vrlw, v4i32>;
690 def VSL : VX1_Int_Ty< 452, "vsl" , int_ppc_altivec_vsl, v4i32 >;
691 def VSLO : VX1_Int_Ty<1036, "vslo", int_ppc_altivec_vslo, v4i32>;
693 def VSLB : VX1_Int_Ty< 260, "vslb", int_ppc_altivec_vslb, v16i8>;
694 def VSLH : VX1_Int_Ty< 324, "vslh", int_ppc_altivec_vslh, v8i16>;
695 def VSLW : VX1_Int_Ty< 388, "vslw", int_ppc_altivec_vslw, v4i32>;
697 def VSPLTB : VXForm_1<524, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
698 "vspltb $vD, $vB, $UIMM", IIC_VecPerm,
700 (vspltb_shuffle:$UIMM v16i8:$vB, (undef)))]>;
701 def VSPLTH : VXForm_1<588, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
702 "vsplth $vD, $vB, $UIMM", IIC_VecPerm,
704 (vsplth_shuffle:$UIMM v16i8:$vB, (undef)))]>;
705 def VSPLTW : VXForm_1<652, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
706 "vspltw $vD, $vB, $UIMM", IIC_VecPerm,
708 (vspltw_shuffle:$UIMM v16i8:$vB, (undef)))]>;
709 let isCodeGenOnly = 1 in {
710 def VSPLTBs : VXForm_1<524, (outs vrrc:$vD), (ins u5imm:$UIMM, vfrc:$vB),
711 "vspltb $vD, $vB, $UIMM", IIC_VecPerm, []>;
712 def VSPLTHs : VXForm_1<588, (outs vrrc:$vD), (ins u5imm:$UIMM, vfrc:$vB),
713 "vsplth $vD, $vB, $UIMM", IIC_VecPerm, []>;
716 def VSR : VX1_Int_Ty< 708, "vsr" , int_ppc_altivec_vsr, v4i32>;
717 def VSRO : VX1_Int_Ty<1100, "vsro" , int_ppc_altivec_vsro, v4i32>;
719 def VSRAB : VX1_Int_Ty< 772, "vsrab", int_ppc_altivec_vsrab, v16i8>;
720 def VSRAH : VX1_Int_Ty< 836, "vsrah", int_ppc_altivec_vsrah, v8i16>;
721 def VSRAW : VX1_Int_Ty< 900, "vsraw", int_ppc_altivec_vsraw, v4i32>;
722 def VSRB : VX1_Int_Ty< 516, "vsrb" , int_ppc_altivec_vsrb , v16i8>;
723 def VSRH : VX1_Int_Ty< 580, "vsrh" , int_ppc_altivec_vsrh , v8i16>;
724 def VSRW : VX1_Int_Ty< 644, "vsrw" , int_ppc_altivec_vsrw , v4i32>;
727 def VSPLTISB : VXForm_3<780, (outs vrrc:$vD), (ins s5imm:$SIMM),
728 "vspltisb $vD, $SIMM", IIC_VecPerm,
729 [(set v16i8:$vD, (v16i8 vecspltisb:$SIMM))]>;
730 def VSPLTISH : VXForm_3<844, (outs vrrc:$vD), (ins s5imm:$SIMM),
731 "vspltish $vD, $SIMM", IIC_VecPerm,
732 [(set v8i16:$vD, (v8i16 vecspltish:$SIMM))]>;
733 def VSPLTISW : VXForm_3<908, (outs vrrc:$vD), (ins s5imm:$SIMM),
734 "vspltisw $vD, $SIMM", IIC_VecPerm,
735 [(set v4i32:$vD, (v4i32 vecspltisw:$SIMM))]>;
738 def VPKPX : VX1_Int_Ty2<782, "vpkpx", int_ppc_altivec_vpkpx,
740 def VPKSHSS : VX1_Int_Ty2<398, "vpkshss", int_ppc_altivec_vpkshss,
742 def VPKSHUS : VX1_Int_Ty2<270, "vpkshus", int_ppc_altivec_vpkshus,
744 def VPKSWSS : VX1_Int_Ty2<462, "vpkswss", int_ppc_altivec_vpkswss,
746 def VPKSWUS : VX1_Int_Ty2<334, "vpkswus", int_ppc_altivec_vpkswus,
748 def VPKUHUM : VXForm_1<14, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
749 "vpkuhum $vD, $vA, $vB", IIC_VecFP,
751 (vpkuhum_shuffle v16i8:$vA, v16i8:$vB))]>;
752 def VPKUHUS : VX1_Int_Ty2<142, "vpkuhus", int_ppc_altivec_vpkuhus,
754 def VPKUWUM : VXForm_1<78, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
755 "vpkuwum $vD, $vA, $vB", IIC_VecFP,
757 (vpkuwum_shuffle v16i8:$vA, v16i8:$vB))]>;
758 def VPKUWUS : VX1_Int_Ty2<206, "vpkuwus", int_ppc_altivec_vpkuwus,
762 def VUPKHPX : VX2_Int_Ty2<846, "vupkhpx", int_ppc_altivec_vupkhpx,
764 def VUPKHSB : VX2_Int_Ty2<526, "vupkhsb", int_ppc_altivec_vupkhsb,
766 def VUPKHSH : VX2_Int_Ty2<590, "vupkhsh", int_ppc_altivec_vupkhsh,
768 def VUPKLPX : VX2_Int_Ty2<974, "vupklpx", int_ppc_altivec_vupklpx,
770 def VUPKLSB : VX2_Int_Ty2<654, "vupklsb", int_ppc_altivec_vupklsb,
772 def VUPKLSH : VX2_Int_Ty2<718, "vupklsh", int_ppc_altivec_vupklsh,
776 // Altivec Comparisons.
778 class VCMP<bits<10> xo, string asmstr, ValueType Ty>
779 : VXRForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), asmstr,
781 [(set Ty:$vD, (Ty (PPCvcmp Ty:$vA, Ty:$vB, xo)))]>;
782 class VCMPo<bits<10> xo, string asmstr, ValueType Ty>
783 : VXRForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), asmstr,
785 [(set Ty:$vD, (Ty (PPCvcmp_o Ty:$vA, Ty:$vB, xo)))]> {
790 // f32 element comparisons.0
791 def VCMPBFP : VCMP <966, "vcmpbfp $vD, $vA, $vB" , v4f32>;
792 def VCMPBFPo : VCMPo<966, "vcmpbfp. $vD, $vA, $vB" , v4f32>;
793 def VCMPEQFP : VCMP <198, "vcmpeqfp $vD, $vA, $vB" , v4f32>;
794 def VCMPEQFPo : VCMPo<198, "vcmpeqfp. $vD, $vA, $vB", v4f32>;
795 def VCMPGEFP : VCMP <454, "vcmpgefp $vD, $vA, $vB" , v4f32>;
796 def VCMPGEFPo : VCMPo<454, "vcmpgefp. $vD, $vA, $vB", v4f32>;
797 def VCMPGTFP : VCMP <710, "vcmpgtfp $vD, $vA, $vB" , v4f32>;
798 def VCMPGTFPo : VCMPo<710, "vcmpgtfp. $vD, $vA, $vB", v4f32>;
800 // i8 element comparisons.
801 def VCMPEQUB : VCMP < 6, "vcmpequb $vD, $vA, $vB" , v16i8>;
802 def VCMPEQUBo : VCMPo< 6, "vcmpequb. $vD, $vA, $vB", v16i8>;
803 def VCMPGTSB : VCMP <774, "vcmpgtsb $vD, $vA, $vB" , v16i8>;
804 def VCMPGTSBo : VCMPo<774, "vcmpgtsb. $vD, $vA, $vB", v16i8>;
805 def VCMPGTUB : VCMP <518, "vcmpgtub $vD, $vA, $vB" , v16i8>;
806 def VCMPGTUBo : VCMPo<518, "vcmpgtub. $vD, $vA, $vB", v16i8>;
808 // i16 element comparisons.
809 def VCMPEQUH : VCMP < 70, "vcmpequh $vD, $vA, $vB" , v8i16>;
810 def VCMPEQUHo : VCMPo< 70, "vcmpequh. $vD, $vA, $vB", v8i16>;
811 def VCMPGTSH : VCMP <838, "vcmpgtsh $vD, $vA, $vB" , v8i16>;
812 def VCMPGTSHo : VCMPo<838, "vcmpgtsh. $vD, $vA, $vB", v8i16>;
813 def VCMPGTUH : VCMP <582, "vcmpgtuh $vD, $vA, $vB" , v8i16>;
814 def VCMPGTUHo : VCMPo<582, "vcmpgtuh. $vD, $vA, $vB", v8i16>;
816 // i32 element comparisons.
817 def VCMPEQUW : VCMP <134, "vcmpequw $vD, $vA, $vB" , v4i32>;
818 def VCMPEQUWo : VCMPo<134, "vcmpequw. $vD, $vA, $vB", v4i32>;
819 def VCMPGTSW : VCMP <902, "vcmpgtsw $vD, $vA, $vB" , v4i32>;
820 def VCMPGTSWo : VCMPo<902, "vcmpgtsw. $vD, $vA, $vB", v4i32>;
821 def VCMPGTUW : VCMP <646, "vcmpgtuw $vD, $vA, $vB" , v4i32>;
822 def VCMPGTUWo : VCMPo<646, "vcmpgtuw. $vD, $vA, $vB", v4i32>;
824 let isCodeGenOnly = 1, isMoveImm = 1, isAsCheapAsAMove = 1,
825 isReMaterializable = 1 in {
827 def V_SET0B : VXForm_setzero<1220, (outs vrrc:$vD), (ins),
828 "vxor $vD, $vD, $vD", IIC_VecFP,
829 [(set v16i8:$vD, (v16i8 immAllZerosV))]>;
830 def V_SET0H : VXForm_setzero<1220, (outs vrrc:$vD), (ins),
831 "vxor $vD, $vD, $vD", IIC_VecFP,
832 [(set v8i16:$vD, (v8i16 immAllZerosV))]>;
833 def V_SET0 : VXForm_setzero<1220, (outs vrrc:$vD), (ins),
834 "vxor $vD, $vD, $vD", IIC_VecFP,
835 [(set v4i32:$vD, (v4i32 immAllZerosV))]>;
838 def V_SETALLONESB : VXForm_3<908, (outs vrrc:$vD), (ins),
839 "vspltisw $vD, -1", IIC_VecFP,
840 [(set v16i8:$vD, (v16i8 immAllOnesV))]>;
841 def V_SETALLONESH : VXForm_3<908, (outs vrrc:$vD), (ins),
842 "vspltisw $vD, -1", IIC_VecFP,
843 [(set v8i16:$vD, (v8i16 immAllOnesV))]>;
844 def V_SETALLONES : VXForm_3<908, (outs vrrc:$vD), (ins),
845 "vspltisw $vD, -1", IIC_VecFP,
846 [(set v4i32:$vD, (v4i32 immAllOnesV))]>;
849 } // VALU Operations.
851 //===----------------------------------------------------------------------===//
852 // Additional Altivec Patterns
855 // Extended mnemonics
856 def : InstAlias<"vmr $vD, $vA", (VOR vrrc:$vD, vrrc:$vA, vrrc:$vA)>;
857 def : InstAlias<"vnot $vD, $vA", (VNOR vrrc:$vD, vrrc:$vA, vrrc:$vA)>;
860 def : Pat<(v4i32 (load xoaddr:$src)), (LVX xoaddr:$src)>;
863 def : Pat<(store v4i32:$rS, xoaddr:$dst),
864 (STVX $rS, xoaddr:$dst)>;
867 def : Pat<(v16i8 (bitconvert (v8i16 VRRC:$src))), (v16i8 VRRC:$src)>;
868 def : Pat<(v16i8 (bitconvert (v4i32 VRRC:$src))), (v16i8 VRRC:$src)>;
869 def : Pat<(v16i8 (bitconvert (v4f32 VRRC:$src))), (v16i8 VRRC:$src)>;
870 def : Pat<(v16i8 (bitconvert (v2i64 VRRC:$src))), (v16i8 VRRC:$src)>;
871 def : Pat<(v16i8 (bitconvert (v1i128 VRRC:$src))), (v16i8 VRRC:$src)>;
873 def : Pat<(v8i16 (bitconvert (v16i8 VRRC:$src))), (v8i16 VRRC:$src)>;
874 def : Pat<(v8i16 (bitconvert (v4i32 VRRC:$src))), (v8i16 VRRC:$src)>;
875 def : Pat<(v8i16 (bitconvert (v4f32 VRRC:$src))), (v8i16 VRRC:$src)>;
876 def : Pat<(v8i16 (bitconvert (v2i64 VRRC:$src))), (v8i16 VRRC:$src)>;
877 def : Pat<(v8i16 (bitconvert (v1i128 VRRC:$src))), (v8i16 VRRC:$src)>;
879 def : Pat<(v4i32 (bitconvert (v16i8 VRRC:$src))), (v4i32 VRRC:$src)>;
880 def : Pat<(v4i32 (bitconvert (v8i16 VRRC:$src))), (v4i32 VRRC:$src)>;
881 def : Pat<(v4i32 (bitconvert (v4f32 VRRC:$src))), (v4i32 VRRC:$src)>;
882 def : Pat<(v4i32 (bitconvert (v2i64 VRRC:$src))), (v4i32 VRRC:$src)>;
883 def : Pat<(v4i32 (bitconvert (v1i128 VRRC:$src))), (v4i32 VRRC:$src)>;
885 def : Pat<(v4f32 (bitconvert (v16i8 VRRC:$src))), (v4f32 VRRC:$src)>;
886 def : Pat<(v4f32 (bitconvert (v8i16 VRRC:$src))), (v4f32 VRRC:$src)>;
887 def : Pat<(v4f32 (bitconvert (v4i32 VRRC:$src))), (v4f32 VRRC:$src)>;
888 def : Pat<(v4f32 (bitconvert (v2i64 VRRC:$src))), (v4f32 VRRC:$src)>;
889 def : Pat<(v4f32 (bitconvert (v1i128 VRRC:$src))), (v4f32 VRRC:$src)>;
891 def : Pat<(v2i64 (bitconvert (v16i8 VRRC:$src))), (v2i64 VRRC:$src)>;
892 def : Pat<(v2i64 (bitconvert (v8i16 VRRC:$src))), (v2i64 VRRC:$src)>;
893 def : Pat<(v2i64 (bitconvert (v4i32 VRRC:$src))), (v2i64 VRRC:$src)>;
894 def : Pat<(v2i64 (bitconvert (v4f32 VRRC:$src))), (v2i64 VRRC:$src)>;
895 def : Pat<(v2i64 (bitconvert (v1i128 VRRC:$src))), (v2i64 VRRC:$src)>;
897 def : Pat<(v1i128 (bitconvert (v16i8 VRRC:$src))), (v1i128 VRRC:$src)>;
898 def : Pat<(v1i128 (bitconvert (v8i16 VRRC:$src))), (v1i128 VRRC:$src)>;
899 def : Pat<(v1i128 (bitconvert (v4i32 VRRC:$src))), (v1i128 VRRC:$src)>;
900 def : Pat<(v1i128 (bitconvert (v4f32 VRRC:$src))), (v1i128 VRRC:$src)>;
901 def : Pat<(v1i128 (bitconvert (v2i64 VRRC:$src))), (v1i128 VRRC:$src)>;
904 def : Pat<(v16i8 (umax v16i8:$src1, v16i8:$src2)),
905 (v16i8 (VMAXUB $src1, $src2))>;
906 def : Pat<(v16i8 (smax v16i8:$src1, v16i8:$src2)),
907 (v16i8 (VMAXSB $src1, $src2))>;
908 def : Pat<(v8i16 (umax v8i16:$src1, v8i16:$src2)),
909 (v8i16 (VMAXUH $src1, $src2))>;
910 def : Pat<(v8i16 (smax v8i16:$src1, v8i16:$src2)),
911 (v8i16 (VMAXSH $src1, $src2))>;
912 def : Pat<(v4i32 (umax v4i32:$src1, v4i32:$src2)),
913 (v4i32 (VMAXUW $src1, $src2))>;
914 def : Pat<(v4i32 (smax v4i32:$src1, v4i32:$src2)),
915 (v4i32 (VMAXSW $src1, $src2))>;
916 def : Pat<(v16i8 (umin v16i8:$src1, v16i8:$src2)),
917 (v16i8 (VMINUB $src1, $src2))>;
918 def : Pat<(v16i8 (smin v16i8:$src1, v16i8:$src2)),
919 (v16i8 (VMINSB $src1, $src2))>;
920 def : Pat<(v8i16 (umin v8i16:$src1, v8i16:$src2)),
921 (v8i16 (VMINUH $src1, $src2))>;
922 def : Pat<(v8i16 (smin v8i16:$src1, v8i16:$src2)),
923 (v8i16 (VMINSH $src1, $src2))>;
924 def : Pat<(v4i32 (umin v4i32:$src1, v4i32:$src2)),
925 (v4i32 (VMINUW $src1, $src2))>;
926 def : Pat<(v4i32 (smin v4i32:$src1, v4i32:$src2)),
927 (v4i32 (VMINSW $src1, $src2))>;
931 // Match vsldoi(x,x), vpkuwum(x,x), vpkuhum(x,x)
932 def:Pat<(vsldoi_unary_shuffle:$in v16i8:$vA, undef),
933 (VSLDOI $vA, $vA, (VSLDOI_unary_get_imm $in))>;
934 def:Pat<(vpkuwum_unary_shuffle v16i8:$vA, undef),
936 def:Pat<(vpkuhum_unary_shuffle v16i8:$vA, undef),
938 def:Pat<(vsldoi_shuffle:$SH v16i8:$vA, v16i8:$vB),
939 (VSLDOI v16i8:$vA, v16i8:$vB, (VSLDOI_get_imm $SH))>;
942 // Match vsldoi(y,x), vpkuwum(y,x), vpkuhum(y,x), i.e., swapped operands.
943 // These fragments are matched for little-endian, where the inputs must
944 // be swapped for correct semantics.
945 def:Pat<(vsldoi_swapped_shuffle:$in v16i8:$vA, v16i8:$vB),
946 (VSLDOI $vB, $vA, (VSLDOI_swapped_get_imm $in))>;
947 def:Pat<(vpkuwum_swapped_shuffle v16i8:$vA, v16i8:$vB),
949 def:Pat<(vpkuhum_swapped_shuffle v16i8:$vA, v16i8:$vB),
953 def:Pat<(vmrglb_unary_shuffle v16i8:$vA, undef),
955 def:Pat<(vmrglh_unary_shuffle v16i8:$vA, undef),
957 def:Pat<(vmrglw_unary_shuffle v16i8:$vA, undef),
959 def:Pat<(vmrghb_unary_shuffle v16i8:$vA, undef),
961 def:Pat<(vmrghh_unary_shuffle v16i8:$vA, undef),
963 def:Pat<(vmrghw_unary_shuffle v16i8:$vA, undef),
966 // Match vmrg*(y,x), i.e., swapped operands. These fragments
967 // are matched for little-endian, where the inputs must be
968 // swapped for correct semantics.
969 def:Pat<(vmrglb_swapped_shuffle v16i8:$vA, v16i8:$vB),
971 def:Pat<(vmrglh_swapped_shuffle v16i8:$vA, v16i8:$vB),
973 def:Pat<(vmrglw_swapped_shuffle v16i8:$vA, v16i8:$vB),
975 def:Pat<(vmrghb_swapped_shuffle v16i8:$vA, v16i8:$vB),
977 def:Pat<(vmrghh_swapped_shuffle v16i8:$vA, v16i8:$vB),
979 def:Pat<(vmrghw_swapped_shuffle v16i8:$vA, v16i8:$vB),
982 // Logical Operations
983 def : Pat<(vnot_ppc v4i32:$vA), (VNOR $vA, $vA)>;
985 def : Pat<(vnot_ppc (or v4i32:$A, v4i32:$B)),
987 def : Pat<(and v4i32:$A, (vnot_ppc v4i32:$B)),
990 def : Pat<(fmul v4f32:$vA, v4f32:$vB),
992 (v4i32 (VSLW (v4i32 (V_SETALLONES)), (v4i32 (V_SETALLONES)))))>;
994 // Fused multiply add and multiply sub for packed float. These are represented
995 // separately from the real instructions above, for operations that must have
996 // the additional precision, such as Newton-Rhapson (used by divide, sqrt)
997 def : Pat<(PPCvmaddfp v4f32:$A, v4f32:$B, v4f32:$C),
998 (VMADDFP $A, $B, $C)>;
999 def : Pat<(PPCvnmsubfp v4f32:$A, v4f32:$B, v4f32:$C),
1000 (VNMSUBFP $A, $B, $C)>;
1002 def : Pat<(int_ppc_altivec_vmaddfp v4f32:$A, v4f32:$B, v4f32:$C),
1003 (VMADDFP $A, $B, $C)>;
1004 def : Pat<(int_ppc_altivec_vnmsubfp v4f32:$A, v4f32:$B, v4f32:$C),
1005 (VNMSUBFP $A, $B, $C)>;
1007 def : Pat<(PPCvperm v16i8:$vA, v16i8:$vB, v16i8:$vC),
1008 (VPERM $vA, $vB, $vC)>;
1010 def : Pat<(PPCfre v4f32:$A), (VREFP $A)>;
1011 def : Pat<(PPCfrsqrte v4f32:$A), (VRSQRTEFP $A)>;
1014 def : Pat<(v16i8 (shl v16i8:$vA, v16i8:$vB)),
1015 (v16i8 (VSLB $vA, $vB))>;
1016 def : Pat<(v8i16 (shl v8i16:$vA, v8i16:$vB)),
1017 (v8i16 (VSLH $vA, $vB))>;
1018 def : Pat<(v4i32 (shl v4i32:$vA, v4i32:$vB)),
1019 (v4i32 (VSLW $vA, $vB))>;
1020 def : Pat<(v1i128 (shl v1i128:$vA, v1i128:$vB)),
1021 (v1i128 (VSL (v16i8 (VSLO $vA, $vB)), (v16i8 (VSPLTB 15, $vB))))>;
1022 def : Pat<(v16i8 (PPCshl v16i8:$vA, v16i8:$vB)),
1023 (v16i8 (VSLB $vA, $vB))>;
1024 def : Pat<(v8i16 (PPCshl v8i16:$vA, v8i16:$vB)),
1025 (v8i16 (VSLH $vA, $vB))>;
1026 def : Pat<(v4i32 (PPCshl v4i32:$vA, v4i32:$vB)),
1027 (v4i32 (VSLW $vA, $vB))>;
1028 def : Pat<(v1i128 (PPCshl v1i128:$vA, v1i128:$vB)),
1029 (v1i128 (VSL (v16i8 (VSLO $vA, $vB)), (v16i8 (VSPLTB 15, $vB))))>;
1031 def : Pat<(v16i8 (srl v16i8:$vA, v16i8:$vB)),
1032 (v16i8 (VSRB $vA, $vB))>;
1033 def : Pat<(v8i16 (srl v8i16:$vA, v8i16:$vB)),
1034 (v8i16 (VSRH $vA, $vB))>;
1035 def : Pat<(v4i32 (srl v4i32:$vA, v4i32:$vB)),
1036 (v4i32 (VSRW $vA, $vB))>;
1037 def : Pat<(v1i128 (srl v1i128:$vA, v1i128:$vB)),
1038 (v1i128 (VSR (v16i8 (VSRO $vA, $vB)), (v16i8 (VSPLTB 15, $vB))))>;
1039 def : Pat<(v16i8 (PPCsrl v16i8:$vA, v16i8:$vB)),
1040 (v16i8 (VSRB $vA, $vB))>;
1041 def : Pat<(v8i16 (PPCsrl v8i16:$vA, v8i16:$vB)),
1042 (v8i16 (VSRH $vA, $vB))>;
1043 def : Pat<(v4i32 (PPCsrl v4i32:$vA, v4i32:$vB)),
1044 (v4i32 (VSRW $vA, $vB))>;
1045 def : Pat<(v1i128 (PPCsrl v1i128:$vA, v1i128:$vB)),
1046 (v1i128 (VSR (v16i8 (VSRO $vA, $vB)), (v16i8 (VSPLTB 15, $vB))))>;
1048 def : Pat<(v16i8 (sra v16i8:$vA, v16i8:$vB)),
1049 (v16i8 (VSRAB $vA, $vB))>;
1050 def : Pat<(v8i16 (sra v8i16:$vA, v8i16:$vB)),
1051 (v8i16 (VSRAH $vA, $vB))>;
1052 def : Pat<(v4i32 (sra v4i32:$vA, v4i32:$vB)),
1053 (v4i32 (VSRAW $vA, $vB))>;
1054 def : Pat<(v16i8 (PPCsra v16i8:$vA, v16i8:$vB)),
1055 (v16i8 (VSRAB $vA, $vB))>;
1056 def : Pat<(v8i16 (PPCsra v8i16:$vA, v8i16:$vB)),
1057 (v8i16 (VSRAH $vA, $vB))>;
1058 def : Pat<(v4i32 (PPCsra v4i32:$vA, v4i32:$vB)),
1059 (v4i32 (VSRAW $vA, $vB))>;
1061 // Float to integer and integer to float conversions
1062 def : Pat<(v4i32 (fp_to_sint v4f32:$vA)),
1064 def : Pat<(v4i32 (fp_to_uint v4f32:$vA)),
1066 def : Pat<(v4f32 (sint_to_fp v4i32:$vA)),
1068 def : Pat<(v4f32 (uint_to_fp v4i32:$vA)),
1071 // Floating-point rounding
1072 def : Pat<(v4f32 (ffloor v4f32:$vA)),
1074 def : Pat<(v4f32 (fceil v4f32:$vA)),
1076 def : Pat<(v4f32 (ftrunc v4f32:$vA)),
1078 def : Pat<(v4f32 (fnearbyint v4f32:$vA)),
1082 def : Pat<(v16i8 (vselect v16i8:$vA, v16i8:$vB, v16i8:$vC)),
1083 (VSEL $vC, $vB, $vA)>;
1084 def : Pat<(v8i16 (vselect v8i16:$vA, v8i16:$vB, v8i16:$vC)),
1085 (VSEL $vC, $vB, $vA)>;
1086 def : Pat<(v4i32 (vselect v4i32:$vA, v4i32:$vB, v4i32:$vC)),
1087 (VSEL $vC, $vB, $vA)>;
1088 def : Pat<(v2i64 (vselect v2i64:$vA, v2i64:$vB, v2i64:$vC)),
1089 (VSEL $vC, $vB, $vA)>;
1090 def : Pat<(v4f32 (vselect v4i32:$vA, v4f32:$vB, v4f32:$vC)),
1091 (VSEL $vC, $vB, $vA)>;
1092 def : Pat<(v2f64 (vselect v2i64:$vA, v2f64:$vB, v2f64:$vC)),
1093 (VSEL $vC, $vB, $vA)>;
1097 def HasP8Altivec : Predicate<"PPCSubTarget->hasP8Altivec()">;
1098 def HasP8Crypto : Predicate<"PPCSubTarget->hasP8Crypto()">;
1099 let Predicates = [HasP8Altivec] in {
1101 let isCommutable = 1 in {
1102 def VMULESW : VX1_Int_Ty2<904, "vmulesw", int_ppc_altivec_vmulesw,
1104 def VMULEUW : VX1_Int_Ty2<648, "vmuleuw", int_ppc_altivec_vmuleuw,
1106 def VMULOSW : VX1_Int_Ty2<392, "vmulosw", int_ppc_altivec_vmulosw,
1108 def VMULOUW : VX1_Int_Ty2<136, "vmulouw", int_ppc_altivec_vmulouw,
1110 def VMULUWM : VXForm_1<137, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1111 "vmuluwm $vD, $vA, $vB", IIC_VecGeneral,
1112 [(set v4i32:$vD, (mul v4i32:$vA, v4i32:$vB))]>;
1113 def VMAXSD : VX1_Int_Ty<450, "vmaxsd", int_ppc_altivec_vmaxsd, v2i64>;
1114 def VMAXUD : VX1_Int_Ty<194, "vmaxud", int_ppc_altivec_vmaxud, v2i64>;
1115 def VMINSD : VX1_Int_Ty<962, "vminsd", int_ppc_altivec_vminsd, v2i64>;
1116 def VMINUD : VX1_Int_Ty<706, "vminud", int_ppc_altivec_vminud, v2i64>;
1120 def VMRGEW : VXForm_1<1932, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1121 "vmrgew $vD, $vA, $vB", IIC_VecFP,
1123 (v16i8 (vmrgew_shuffle v16i8:$vA, v16i8:$vB)))]>;
1124 def VMRGOW : VXForm_1<1676, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1125 "vmrgow $vD, $vA, $vB", IIC_VecFP,
1127 (v16i8 (vmrgow_shuffle v16i8:$vA, v16i8:$vB)))]>;
1129 // Match vmrgew(x,x) and vmrgow(x,x)
1130 def:Pat<(vmrgew_unary_shuffle v16i8:$vA, undef),
1132 def:Pat<(vmrgow_unary_shuffle v16i8:$vA, undef),
1135 // Match vmrgew(y,x) and vmrgow(y,x), i.e., swapped operands. These fragments
1136 // are matched for little-endian, where the inputs must be swapped for correct
1138 def:Pat<(vmrgew_swapped_shuffle v16i8:$vA, v16i8:$vB),
1140 def:Pat<(vmrgow_swapped_shuffle v16i8:$vA, v16i8:$vB),
1145 def VRLD : VX1_Int_Ty<196, "vrld", int_ppc_altivec_vrld, v2i64>;
1146 def VSLD : VXForm_1<1476, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1147 "vsld $vD, $vA, $vB", IIC_VecGeneral, []>;
1148 def VSRD : VXForm_1<1732, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1149 "vsrd $vD, $vA, $vB", IIC_VecGeneral, []>;
1150 def VSRAD : VXForm_1<964, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1151 "vsrad $vD, $vA, $vB", IIC_VecGeneral, []>;
1153 def : Pat<(v2i64 (shl v2i64:$vA, v2i64:$vB)),
1154 (v2i64 (VSLD $vA, $vB))>;
1155 def : Pat<(v2i64 (PPCshl v2i64:$vA, v2i64:$vB)),
1156 (v2i64 (VSLD $vA, $vB))>;
1157 def : Pat<(v2i64 (srl v2i64:$vA, v2i64:$vB)),
1158 (v2i64 (VSRD $vA, $vB))>;
1159 def : Pat<(v2i64 (PPCsrl v2i64:$vA, v2i64:$vB)),
1160 (v2i64 (VSRD $vA, $vB))>;
1161 def : Pat<(v2i64 (sra v2i64:$vA, v2i64:$vB)),
1162 (v2i64 (VSRAD $vA, $vB))>;
1163 def : Pat<(v2i64 (PPCsra v2i64:$vA, v2i64:$vB)),
1164 (v2i64 (VSRAD $vA, $vB))>;
1166 // Vector Integer Arithmetic Instructions
1167 let isCommutable = 1 in {
1168 def VADDUDM : VXForm_1<192, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1169 "vaddudm $vD, $vA, $vB", IIC_VecGeneral,
1170 [(set v2i64:$vD, (add v2i64:$vA, v2i64:$vB))]>;
1171 def VADDUQM : VXForm_1<256, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1172 "vadduqm $vD, $vA, $vB", IIC_VecGeneral,
1173 [(set v1i128:$vD, (add v1i128:$vA, v1i128:$vB))]>;
1176 // Vector Quadword Add
1177 def VADDEUQM : VA1a_Int_Ty<60, "vaddeuqm", int_ppc_altivec_vaddeuqm, v1i128>;
1178 def VADDCUQ : VX1_Int_Ty<320, "vaddcuq", int_ppc_altivec_vaddcuq, v1i128>;
1179 def VADDECUQ : VA1a_Int_Ty<61, "vaddecuq", int_ppc_altivec_vaddecuq, v1i128>;
1181 // Vector Doubleword Subtract
1182 def VSUBUDM : VXForm_1<1216, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1183 "vsubudm $vD, $vA, $vB", IIC_VecGeneral,
1184 [(set v2i64:$vD, (sub v2i64:$vA, v2i64:$vB))]>;
1186 // Vector Quadword Subtract
1187 def VSUBUQM : VXForm_1<1280, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1188 "vsubuqm $vD, $vA, $vB", IIC_VecGeneral,
1189 [(set v1i128:$vD, (sub v1i128:$vA, v1i128:$vB))]>;
1190 def VSUBEUQM : VA1a_Int_Ty<62, "vsubeuqm", int_ppc_altivec_vsubeuqm, v1i128>;
1191 def VSUBCUQ : VX1_Int_Ty<1344, "vsubcuq", int_ppc_altivec_vsubcuq, v1i128>;
1192 def VSUBECUQ : VA1a_Int_Ty<63, "vsubecuq", int_ppc_altivec_vsubecuq, v1i128>;
1194 // Count Leading Zeros
1195 def VCLZB : VXForm_2<1794, (outs vrrc:$vD), (ins vrrc:$vB),
1196 "vclzb $vD, $vB", IIC_VecGeneral,
1197 [(set v16i8:$vD, (ctlz v16i8:$vB))]>;
1198 def VCLZH : VXForm_2<1858, (outs vrrc:$vD), (ins vrrc:$vB),
1199 "vclzh $vD, $vB", IIC_VecGeneral,
1200 [(set v8i16:$vD, (ctlz v8i16:$vB))]>;
1201 def VCLZW : VXForm_2<1922, (outs vrrc:$vD), (ins vrrc:$vB),
1202 "vclzw $vD, $vB", IIC_VecGeneral,
1203 [(set v4i32:$vD, (ctlz v4i32:$vB))]>;
1204 def VCLZD : VXForm_2<1986, (outs vrrc:$vD), (ins vrrc:$vB),
1205 "vclzd $vD, $vB", IIC_VecGeneral,
1206 [(set v2i64:$vD, (ctlz v2i64:$vB))]>;
1209 def VPOPCNTB : VXForm_2<1795, (outs vrrc:$vD), (ins vrrc:$vB),
1210 "vpopcntb $vD, $vB", IIC_VecGeneral,
1211 [(set v16i8:$vD, (ctpop v16i8:$vB))]>;
1212 def VPOPCNTH : VXForm_2<1859, (outs vrrc:$vD), (ins vrrc:$vB),
1213 "vpopcnth $vD, $vB", IIC_VecGeneral,
1214 [(set v8i16:$vD, (ctpop v8i16:$vB))]>;
1215 def VPOPCNTW : VXForm_2<1923, (outs vrrc:$vD), (ins vrrc:$vB),
1216 "vpopcntw $vD, $vB", IIC_VecGeneral,
1217 [(set v4i32:$vD, (ctpop v4i32:$vB))]>;
1218 def VPOPCNTD : VXForm_2<1987, (outs vrrc:$vD), (ins vrrc:$vB),
1219 "vpopcntd $vD, $vB", IIC_VecGeneral,
1220 [(set v2i64:$vD, (ctpop v2i64:$vB))]>;
1222 let isCommutable = 1 in {
1223 // FIXME: Use AddedComplexity > 400 to ensure these patterns match before the
1224 // VSX equivalents. We need to fix this up at some point. Two possible
1225 // solutions for this problem:
1226 // 1. Disable Altivec patterns that compete with VSX patterns using the
1227 // !HasVSX predicate. This essentially favours VSX over Altivec, in
1228 // hopes of reducing register pressure (larger register set using VSX
1229 // instructions than VMX instructions)
1230 // 2. Employ a more disciplined use of AddedComplexity, which would provide
1231 // more fine-grained control than option 1. This would be beneficial
1232 // if we find situations where Altivec is really preferred over VSX.
1233 def VEQV : VXForm_1<1668, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1234 "veqv $vD, $vA, $vB", IIC_VecGeneral,
1235 [(set v4i32:$vD, (vnot_ppc (xor v4i32:$vA, v4i32:$vB)))]>;
1236 def VNAND : VXForm_1<1412, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1237 "vnand $vD, $vA, $vB", IIC_VecGeneral,
1238 [(set v4i32:$vD, (vnot_ppc (and v4i32:$vA, v4i32:$vB)))]>;
1241 def VORC : VXForm_1<1348, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1242 "vorc $vD, $vA, $vB", IIC_VecGeneral,
1243 [(set v4i32:$vD, (or v4i32:$vA,
1244 (vnot_ppc v4i32:$vB)))]>;
1246 // i64 element comparisons.
1247 def VCMPEQUD : VCMP <199, "vcmpequd $vD, $vA, $vB" , v2i64>;
1248 def VCMPEQUDo : VCMPo<199, "vcmpequd. $vD, $vA, $vB", v2i64>;
1249 def VCMPGTSD : VCMP <967, "vcmpgtsd $vD, $vA, $vB" , v2i64>;
1250 def VCMPGTSDo : VCMPo<967, "vcmpgtsd. $vD, $vA, $vB", v2i64>;
1251 def VCMPGTUD : VCMP <711, "vcmpgtud $vD, $vA, $vB" , v2i64>;
1252 def VCMPGTUDo : VCMPo<711, "vcmpgtud. $vD, $vA, $vB", v2i64>;
1254 // The cryptography instructions that do not require Category:Vector.Crypto
1255 def VPMSUMB : VX1_Int_Ty<1032, "vpmsumb",
1256 int_ppc_altivec_crypto_vpmsumb, v16i8>;
1257 def VPMSUMH : VX1_Int_Ty<1096, "vpmsumh",
1258 int_ppc_altivec_crypto_vpmsumh, v8i16>;
1259 def VPMSUMW : VX1_Int_Ty<1160, "vpmsumw",
1260 int_ppc_altivec_crypto_vpmsumw, v4i32>;
1261 def VPMSUMD : VX1_Int_Ty<1224, "vpmsumd",
1262 int_ppc_altivec_crypto_vpmsumd, v2i64>;
1263 def VPERMXOR : VA1a_Int_Ty<45, "vpermxor",
1264 int_ppc_altivec_crypto_vpermxor, v16i8>;
1266 // Vector doubleword integer pack and unpack.
1267 def VPKSDSS : VX1_Int_Ty2<1486, "vpksdss", int_ppc_altivec_vpksdss,
1269 def VPKSDUS : VX1_Int_Ty2<1358, "vpksdus", int_ppc_altivec_vpksdus,
1271 def VPKUDUM : VXForm_1<1102, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1272 "vpkudum $vD, $vA, $vB", IIC_VecFP,
1274 (vpkudum_shuffle v16i8:$vA, v16i8:$vB))]>;
1275 def VPKUDUS : VX1_Int_Ty2<1230, "vpkudus", int_ppc_altivec_vpkudus,
1277 def VUPKHSW : VX2_Int_Ty2<1614, "vupkhsw", int_ppc_altivec_vupkhsw,
1279 def VUPKLSW : VX2_Int_Ty2<1742, "vupklsw", int_ppc_altivec_vupklsw,
1282 // Shuffle patterns for unary and swapped (LE) vector pack modulo.
1283 def:Pat<(vpkudum_unary_shuffle v16i8:$vA, undef),
1284 (VPKUDUM $vA, $vA)>;
1285 def:Pat<(vpkudum_swapped_shuffle v16i8:$vA, v16i8:$vB),
1286 (VPKUDUM $vB, $vA)>;
1288 def VGBBD : VX2_Int_Ty2<1292, "vgbbd", int_ppc_altivec_vgbbd, v16i8, v16i8>;
1289 def VBPERMQ : VX1_Int_Ty2<1356, "vbpermq", int_ppc_altivec_vbpermq,
1291 } // end HasP8Altivec
1293 // Crypto instructions (from builtins)
1294 let Predicates = [HasP8Crypto] in {
1295 def VSHASIGMAW : VXCR_Int_Ty<1666, "vshasigmaw",
1296 int_ppc_altivec_crypto_vshasigmaw, v4i32>;
1297 def VSHASIGMAD : VXCR_Int_Ty<1730, "vshasigmad",
1298 int_ppc_altivec_crypto_vshasigmad, v2i64>;
1299 def VCIPHER : VX1_Int_Ty<1288, "vcipher", int_ppc_altivec_crypto_vcipher,
1301 def VCIPHERLAST : VX1_Int_Ty<1289, "vcipherlast",
1302 int_ppc_altivec_crypto_vcipherlast, v2i64>;
1303 def VNCIPHER : VX1_Int_Ty<1352, "vncipher",
1304 int_ppc_altivec_crypto_vncipher, v2i64>;
1305 def VNCIPHERLAST : VX1_Int_Ty<1353, "vncipherlast",
1306 int_ppc_altivec_crypto_vncipherlast, v2i64>;
1307 def VSBOX : VXBX_Int_Ty<1480, "vsbox", int_ppc_altivec_crypto_vsbox, v2i64>;
1310 // The following altivec instructions were introduced in Power ISA 3.0
1311 def HasP9Altivec : Predicate<"PPCSubTarget->hasP9Altivec()">;
1312 let Predicates = [HasP9Altivec] in {
1314 // i8 element comparisons.
1315 def VCMPNEB : VCMP < 7, "vcmpneb $vD, $vA, $vB" , v16i8>;
1316 def VCMPNEBo : VCMPo < 7, "vcmpneb. $vD, $vA, $vB" , v16i8>;
1317 def VCMPNEZB : VCMP <263, "vcmpnezb $vD, $vA, $vB" , v16i8>;
1318 def VCMPNEZBo : VCMPo<263, "vcmpnezb. $vD, $vA, $vB", v16i8>;
1320 // i16 element comparisons.
1321 def VCMPNEH : VCMP < 71, "vcmpneh $vD, $vA, $vB" , v8i16>;
1322 def VCMPNEHo : VCMPo< 71, "vcmpneh. $vD, $vA, $vB" , v8i16>;
1323 def VCMPNEZH : VCMP <327, "vcmpnezh $vD, $vA, $vB" , v8i16>;
1324 def VCMPNEZHo : VCMPo<327, "vcmpnezh. $vD, $vA, $vB", v8i16>;
1326 // i32 element comparisons.
1327 def VCMPNEW : VCMP <135, "vcmpnew $vD, $vA, $vB" , v4i32>;
1328 def VCMPNEWo : VCMPo<135, "vcmpnew. $vD, $vA, $vB" , v4i32>;
1329 def VCMPNEZW : VCMP <391, "vcmpnezw $vD, $vA, $vB" , v4i32>;
1330 def VCMPNEZWo : VCMPo<391, "vcmpnezw. $vD, $vA, $vB", v4i32>;
1332 // VX-Form: [PO VRT / UIM VRB XO].
1333 // We use VXForm_1 to implement it, that is, we use "VRA" (5 bit) to represent
1334 // "/ UIM" (1 + 4 bit)
1335 class VX1_VT5_UIM5_VB5<bits<11> xo, string opc, list<dag> pattern>
1336 : VXForm_1<xo, (outs vrrc:$vD), (ins u4imm:$UIMM, vrrc:$vB),
1337 !strconcat(opc, " $vD, $vB, $UIMM"), IIC_VecGeneral, pattern>;
1339 class VX1_RT5_RA5_VB5<bits<11> xo, string opc, list<dag> pattern>
1340 : VXForm_1<xo, (outs g8rc:$rD), (ins g8rc:$rA, vrrc:$vB),
1341 !strconcat(opc, " $rD, $rA, $vB"), IIC_VecGeneral, pattern>;
1343 // Vector Extract Unsigned
1344 def VEXTRACTUB : VX1_VT5_UIM5_VB5<525, "vextractub", []>;
1345 def VEXTRACTUH : VX1_VT5_UIM5_VB5<589, "vextractuh", []>;
1346 def VEXTRACTUW : VX1_VT5_UIM5_VB5<653, "vextractuw", []>;
1347 def VEXTRACTD : VX1_VT5_UIM5_VB5<717, "vextractd" , []>;
1349 // Vector Extract Unsigned Byte/Halfword/Word Left/Right-Indexed
1350 def VEXTUBLX : VX1_RT5_RA5_VB5<1549, "vextublx", []>;
1351 def VEXTUBRX : VX1_RT5_RA5_VB5<1805, "vextubrx", []>;
1352 def VEXTUHLX : VX1_RT5_RA5_VB5<1613, "vextuhlx", []>;
1353 def VEXTUHRX : VX1_RT5_RA5_VB5<1869, "vextuhrx", []>;
1354 def VEXTUWLX : VX1_RT5_RA5_VB5<1677, "vextuwlx", []>;
1355 def VEXTUWRX : VX1_RT5_RA5_VB5<1933, "vextuwrx", []>;
1357 // Vector Insert Element Instructions
1358 def VINSERTB : VXForm_1<781, (outs vrrc:$vD),
1359 (ins vrrc:$vDi, u4imm:$UIM, vrrc:$vB),
1360 "vinsertb $vD, $vB, $UIM", IIC_VecGeneral,
1361 [(set v16i8:$vD, (PPCvecinsert v16i8:$vDi, v16i8:$vB,
1362 imm32SExt16:$UIM))]>,
1363 RegConstraint<"$vDi = $vD">, NoEncode<"$vDi">;
1364 def VINSERTH : VXForm_1<845, (outs vrrc:$vD),
1365 (ins vrrc:$vDi, u4imm:$UIM, vrrc:$vB),
1366 "vinserth $vD, $vB, $UIM", IIC_VecGeneral,
1367 [(set v8i16:$vD, (PPCvecinsert v8i16:$vDi, v8i16:$vB,
1368 imm32SExt16:$UIM))]>,
1369 RegConstraint<"$vDi = $vD">, NoEncode<"$vDi">;
1370 def VINSERTW : VX1_VT5_UIM5_VB5<909, "vinsertw", []>;
1371 def VINSERTD : VX1_VT5_UIM5_VB5<973, "vinsertd", []>;
1373 class VX_VT5_EO5_VB5<bits<11> xo, bits<5> eo, string opc, list<dag> pattern>
1374 : VXForm_RD5_XO5_RS5<xo, eo, (outs vrrc:$vD), (ins vrrc:$vB),
1375 !strconcat(opc, " $vD, $vB"), IIC_VecGeneral, pattern>;
1376 class VX_VT5_EO5_VB5s<bits<11> xo, bits<5> eo, string opc, list<dag> pattern>
1377 : VXForm_RD5_XO5_RS5<xo, eo, (outs vfrc:$vD), (ins vfrc:$vB),
1378 !strconcat(opc, " $vD, $vB"), IIC_VecGeneral, pattern>;
1380 // Vector Count Leading/Trailing Zero LSB. Result is placed into GPR[rD]
1381 def VCLZLSBB : VXForm_RD5_XO5_RS5<1538, 0, (outs gprc:$rD), (ins vrrc:$vB),
1382 "vclzlsbb $rD, $vB", IIC_VecGeneral,
1383 [(set i32:$rD, (int_ppc_altivec_vclzlsbb
1385 def VCTZLSBB : VXForm_RD5_XO5_RS5<1538, 1, (outs gprc:$rD), (ins vrrc:$vB),
1386 "vctzlsbb $rD, $vB", IIC_VecGeneral,
1387 [(set i32:$rD, (int_ppc_altivec_vctzlsbb
1389 // Vector Count Trailing Zeros
1390 def VCTZB : VX_VT5_EO5_VB5<1538, 28, "vctzb",
1391 [(set v16i8:$vD, (cttz v16i8:$vB))]>;
1392 def VCTZH : VX_VT5_EO5_VB5<1538, 29, "vctzh",
1393 [(set v8i16:$vD, (cttz v8i16:$vB))]>;
1394 def VCTZW : VX_VT5_EO5_VB5<1538, 30, "vctzw",
1395 [(set v4i32:$vD, (cttz v4i32:$vB))]>;
1396 def VCTZD : VX_VT5_EO5_VB5<1538, 31, "vctzd",
1397 [(set v2i64:$vD, (cttz v2i64:$vB))]>;
1399 // Vector Extend Sign
1400 def VEXTSB2W : VX_VT5_EO5_VB5<1538, 16, "vextsb2w", []>;
1401 def VEXTSH2W : VX_VT5_EO5_VB5<1538, 17, "vextsh2w", []>;
1402 def VEXTSB2D : VX_VT5_EO5_VB5<1538, 24, "vextsb2d", []>;
1403 def VEXTSH2D : VX_VT5_EO5_VB5<1538, 25, "vextsh2d", []>;
1404 def VEXTSW2D : VX_VT5_EO5_VB5<1538, 26, "vextsw2d", []>;
1405 let isCodeGenOnly = 1 in {
1406 def VEXTSB2Ws : VX_VT5_EO5_VB5s<1538, 16, "vextsb2w", []>;
1407 def VEXTSH2Ws : VX_VT5_EO5_VB5s<1538, 17, "vextsh2w", []>;
1408 def VEXTSB2Ds : VX_VT5_EO5_VB5s<1538, 24, "vextsb2d", []>;
1409 def VEXTSH2Ds : VX_VT5_EO5_VB5s<1538, 25, "vextsh2d", []>;
1410 def VEXTSW2Ds : VX_VT5_EO5_VB5s<1538, 26, "vextsw2d", []>;
1413 // Vector Integer Negate
1414 def VNEGW : VX_VT5_EO5_VB5<1538, 6, "vnegw",
1416 (sub (v4i32 immAllZerosV), v4i32:$vB))]>;
1418 def VNEGD : VX_VT5_EO5_VB5<1538, 7, "vnegd",
1420 (sub (v2i64 (bitconvert (v4i32 immAllZerosV))),
1423 // Vector Parity Byte
1424 def VPRTYBW : VX_VT5_EO5_VB5<1538, 8, "vprtybw", [(set v4i32:$vD,
1425 (int_ppc_altivec_vprtybw v4i32:$vB))]>;
1426 def VPRTYBD : VX_VT5_EO5_VB5<1538, 9, "vprtybd", [(set v2i64:$vD,
1427 (int_ppc_altivec_vprtybd v2i64:$vB))]>;
1428 def VPRTYBQ : VX_VT5_EO5_VB5<1538, 10, "vprtybq", [(set v1i128:$vD,
1429 (int_ppc_altivec_vprtybq v1i128:$vB))]>;
1431 // Vector (Bit) Permute (Right-indexed)
1432 def VBPERMD : VXForm_1<1484, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1433 "vbpermd $vD, $vA, $vB", IIC_VecFP, []>;
1434 def VPERMR : VAForm_1a<59, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
1435 "vpermr $vD, $vA, $vB, $vC", IIC_VecFP, []>;
1437 class VX1_VT5_VA5_VB5<bits<11> xo, string opc, list<dag> pattern>
1438 : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1439 !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP, pattern>;
1441 // Vector Rotate Left Mask/Mask-Insert
1442 def VRLWNM : VX1_VT5_VA5_VB5<389, "vrlwnm",
1444 (int_ppc_altivec_vrlwnm v4i32:$vA,
1446 def VRLWMI : VXForm_1<133, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vDi),
1447 "vrlwmi $vD, $vA, $vB", IIC_VecFP,
1449 (int_ppc_altivec_vrlwmi v4i32:$vA, v4i32:$vB,
1451 RegConstraint<"$vDi = $vD">, NoEncode<"$vDi">;
1452 def VRLDNM : VX1_VT5_VA5_VB5<453, "vrldnm",
1454 (int_ppc_altivec_vrldnm v2i64:$vA,
1456 def VRLDMI : VXForm_1<197, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vDi),
1457 "vrldmi $vD, $vA, $vB", IIC_VecFP,
1459 (int_ppc_altivec_vrldmi v2i64:$vA, v2i64:$vB,
1461 RegConstraint<"$vDi = $vD">, NoEncode<"$vDi">;
1463 // Vector Shift Left/Right
1464 def VSLV : VX1_VT5_VA5_VB5<1860, "vslv",
1465 [(set v16i8 : $vD, (int_ppc_altivec_vslv v16i8 : $vA, v16i8 : $vB))]>;
1466 def VSRV : VX1_VT5_VA5_VB5<1796, "vsrv",
1467 [(set v16i8 : $vD, (int_ppc_altivec_vsrv v16i8 : $vA, v16i8 : $vB))]>;
1469 // Vector Multiply-by-10 (& Write Carry) Unsigned Quadword
1470 def VMUL10UQ : VXForm_BX<513, (outs vrrc:$vD), (ins vrrc:$vA),
1471 "vmul10uq $vD, $vA", IIC_VecFP, []>;
1472 def VMUL10CUQ : VXForm_BX< 1, (outs vrrc:$vD), (ins vrrc:$vA),
1473 "vmul10cuq $vD, $vA", IIC_VecFP, []>;
1475 // Vector Multiply-by-10 Extended (& Write Carry) Unsigned Quadword
1476 def VMUL10EUQ : VX1_VT5_VA5_VB5<577, "vmul10euq" , []>;
1477 def VMUL10ECUQ : VX1_VT5_VA5_VB5< 65, "vmul10ecuq", []>;
1479 // Decimal Integer Format Conversion Instructions
1481 // [PO VRT EO VRB 1 PS XO], "_o" means CR6 is set.
1482 class VX_VT5_EO5_VB5_PS1_XO9_o<bits<5> eo, bits<9> xo, string opc,
1484 : VX_RD5_EO5_RS5_PS1_XO9<eo, xo, (outs vrrc:$vD), (ins vrrc:$vB, u1imm:$PS),
1485 !strconcat(opc, " $vD, $vB, $PS"), IIC_VecFP, pattern> {
1489 // [PO VRT EO VRB 1 / XO]
1490 class VX_VT5_EO5_VB5_XO9_o<bits<5> eo, bits<9> xo, string opc,
1492 : VX_RD5_EO5_RS5_PS1_XO9<eo, xo, (outs vrrc:$vD), (ins vrrc:$vB),
1493 !strconcat(opc, " $vD, $vB"), IIC_VecFP, pattern> {
1498 // Decimal Convert From/to National/Zoned/Signed-QWord
1499 def BCDCFNo : VX_VT5_EO5_VB5_PS1_XO9_o<7, 385, "bcdcfn." , []>;
1500 def BCDCFZo : VX_VT5_EO5_VB5_PS1_XO9_o<6, 385, "bcdcfz." , []>;
1501 def BCDCTNo : VX_VT5_EO5_VB5_XO9_o <5, 385, "bcdctn." , []>;
1502 def BCDCTZo : VX_VT5_EO5_VB5_PS1_XO9_o<4, 385, "bcdctz." , []>;
1503 def BCDCFSQo : VX_VT5_EO5_VB5_PS1_XO9_o<2, 385, "bcdcfsq.", []>;
1504 def BCDCTSQo : VX_VT5_EO5_VB5_XO9_o <0, 385, "bcdctsq.", []>;
1506 // Decimal Copy-Sign/Set-Sign
1508 def BCDCPSGNo : VX1_VT5_VA5_VB5<833, "bcdcpsgn.", []>;
1510 def BCDSETSGNo : VX_VT5_EO5_VB5_PS1_XO9_o<31, 385, "bcdsetsgn.", []>;
1512 // [PO VRT VRA VRB 1 PS XO], "_o" means CR6 is set.
1513 class VX_VT5_VA5_VB5_PS1_XO9_o<bits<9> xo, string opc, list<dag> pattern>
1514 : VX_RD5_RSp5_PS1_XO9<xo,
1515 (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, u1imm:$PS),
1516 !strconcat(opc, " $vD, $vA, $vB, $PS"), IIC_VecFP, pattern> {
1520 // [PO VRT VRA VRB 1 / XO]
1521 class VX_VT5_VA5_VB5_XO9_o<bits<9> xo, string opc, list<dag> pattern>
1522 : VX_RD5_RSp5_PS1_XO9<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1523 !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP, pattern> {
1528 // Decimal Shift/Unsigned-Shift/Shift-and-Round
1529 def BCDSo : VX_VT5_VA5_VB5_PS1_XO9_o<193, "bcds." , []>;
1530 def BCDUSo : VX_VT5_VA5_VB5_XO9_o <129, "bcdus.", []>;
1531 def BCDSRo : VX_VT5_VA5_VB5_PS1_XO9_o<449, "bcdsr.", []>;
1533 // Decimal (Unsigned) Truncate
1534 def BCDTRUNCo : VX_VT5_VA5_VB5_PS1_XO9_o<257, "bcdtrunc." , []>;
1535 def BCDUTRUNCo : VX_VT5_VA5_VB5_XO9_o <321, "bcdutrunc.", []>;
1537 // Absolute Difference
1538 def VABSDUB : VXForm_1<1027, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1539 "vabsdub $vD, $vA, $vB", IIC_VecGeneral,
1540 [(set v16i8:$vD, (int_ppc_altivec_vabsdub v16i8:$vA, v16i8:$vB))]>;
1541 def VABSDUH : VXForm_1<1091, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1542 "vabsduh $vD, $vA, $vB", IIC_VecGeneral,
1543 [(set v8i16:$vD, (int_ppc_altivec_vabsduh v8i16:$vA, v8i16:$vB))]>;
1544 def VABSDUW : VXForm_1<1155, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1545 "vabsduw $vD, $vA, $vB", IIC_VecGeneral,
1546 [(set v4i32:$vD, (int_ppc_altivec_vabsduw v4i32:$vA, v4i32:$vB))]>;
1548 } // end HasP9Altivec