xref: /freebsd/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCInstrAltivec.td (revision a134ebd6e63f658f2d3d04ac0c60d23bcaa86dd7)
1//===-- PPCInstrAltivec.td - The PowerPC Altivec Extension -*- tablegen -*-===//
2//
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
6//
7//===----------------------------------------------------------------------===//
8//
9// This file describes the Altivec extension to the PowerPC instruction set.
10//
11//===----------------------------------------------------------------------===//
12
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.  **
22// **                                                                        **
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// ****************************************************************************
27
28
29//===----------------------------------------------------------------------===//
30// Altivec transformation functions and pattern fragments.
31//
32
33// Since we canonicalize buildvectors to v16i8, all vnots "-1" operands will be
34// of that type.
35def vnot_ppc : PatFrag<(ops node:$in),
36                       (xor node:$in, (bitconvert (v16i8 immAllOnesV)))>;
37
38def vpkuhum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
39                              (vector_shuffle node:$lhs, node:$rhs), [{
40  return PPC::isVPKUHUMShuffleMask(cast<ShuffleVectorSDNode>(N), 0, *CurDAG);
41}]>;
42def vpkuwum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
43                              (vector_shuffle node:$lhs, node:$rhs), [{
44  return PPC::isVPKUWUMShuffleMask(cast<ShuffleVectorSDNode>(N), 0, *CurDAG);
45}]>;
46def vpkudum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
47                              (vector_shuffle node:$lhs, node:$rhs), [{
48  return PPC::isVPKUDUMShuffleMask(cast<ShuffleVectorSDNode>(N), 0, *CurDAG);
49}]>;
50def 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);
53}]>;
54def 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);
57}]>;
58def 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);
61}]>;
62
63// These fragments are provided for little-endian, where the inputs must be
64// swapped for correct semantics.
65def 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);
68}]>;
69def 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);
72}]>;
73def 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);
76}]>;
77
78def 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);
81}]>;
82def 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);
85}]>;
86def 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);
89}]>;
90def 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);
93}]>;
94def 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);
97}]>;
98def 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);
101}]>;
102
103
104def 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);
107}]>;
108def 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);
111}]>;
112def 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);
115}]>;
116def 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);
119}]>;
120def 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);
123}]>;
124def 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);
127}]>;
128
129
130// These fragments are provided for little-endian, where the inputs must be
131// swapped for correct semantics.
132def 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);
135}]>;
136def 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);
139}]>;
140def 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);
143}]>;
144def 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);
147}]>;
148def 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);
151}]>;
152def 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);
155}]>;
156
157
158def 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);
161}]>;
162def 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);
165}]>;
166def 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);
169}]>;
170def 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);
173}]>;
174def 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);
177}]>;
178def 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);
181}]>;
182
183
184
185def VSLDOI_get_imm : SDNodeXForm<vector_shuffle, [{
186  return getI32Imm(PPC::isVSLDOIShuffleMask(N, 0, *CurDAG), SDLoc(N));
187}]>;
188def vsldoi_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
189                             (vector_shuffle node:$lhs, node:$rhs), [{
190  return PPC::isVSLDOIShuffleMask(N, 0, *CurDAG) != -1;
191}], VSLDOI_get_imm>;
192
193
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.
196def VSLDOI_unary_get_imm : SDNodeXForm<vector_shuffle, [{
197  return getI32Imm(PPC::isVSLDOIShuffleMask(N, 1, *CurDAG), SDLoc(N));
198}]>;
199def 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>;
203
204
205/// VSLDOI_swapped* - These fragments are provided for little-endian, where
206/// the inputs must be swapped for correct semantics.
207def VSLDOI_swapped_get_imm : SDNodeXForm<vector_shuffle, [{
208  return getI32Imm(PPC::isVSLDOIShuffleMask(N, 2, *CurDAG), SDLoc(N));
209}]>;
210def vsldoi_swapped_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
211                                     (vector_shuffle node:$lhs, node:$rhs), [{
212  return PPC::isVSLDOIShuffleMask(N, 2, *CurDAG) != -1;
213}], VSLDOI_get_imm>;
214
215
216// VSPLT*_get_imm xform function: convert vector_shuffle mask to VSPLT* imm.
217def VSPLTB_get_imm : SDNodeXForm<vector_shuffle, [{
218  return getI32Imm(PPC::getSplatIdxForPPCMnemonics(N, 1, *CurDAG), SDLoc(N));
219}]>;
220def vspltb_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
221                             (vector_shuffle node:$lhs, node:$rhs), [{
222  return PPC::isSplatShuffleMask(cast<ShuffleVectorSDNode>(N), 1);
223}], VSPLTB_get_imm>;
224def VSPLTH_get_imm : SDNodeXForm<vector_shuffle, [{
225  return getI32Imm(PPC::getSplatIdxForPPCMnemonics(N, 2, *CurDAG), SDLoc(N));
226}]>;
227def vsplth_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
228                             (vector_shuffle node:$lhs, node:$rhs), [{
229  return PPC::isSplatShuffleMask(cast<ShuffleVectorSDNode>(N), 2);
230}], VSPLTH_get_imm>;
231def VSPLTW_get_imm : SDNodeXForm<vector_shuffle, [{
232  return getI32Imm(PPC::getSplatIdxForPPCMnemonics(N, 4, *CurDAG), SDLoc(N));
233}]>;
234def vspltw_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
235                             (vector_shuffle node:$lhs, node:$rhs), [{
236  return PPC::isSplatShuffleMask(cast<ShuffleVectorSDNode>(N), 4);
237}], VSPLTW_get_imm>;
238
239
240// VSPLTISB_get_imm xform function: convert build_vector to VSPLTISB imm.
241def VSPLTISB_get_imm : SDNodeXForm<build_vector, [{
242  return PPC::get_VSPLTI_elt(N, 1, *CurDAG);
243}]>;
244def vecspltisb : PatLeaf<(build_vector), [{
245  return PPC::get_VSPLTI_elt(N, 1, *CurDAG).getNode() != nullptr;
246}], VSPLTISB_get_imm>;
247
248// VSPLTISH_get_imm xform function: convert build_vector to VSPLTISH imm.
249def VSPLTISH_get_imm : SDNodeXForm<build_vector, [{
250  return PPC::get_VSPLTI_elt(N, 2, *CurDAG);
251}]>;
252def vecspltish : PatLeaf<(build_vector), [{
253  return PPC::get_VSPLTI_elt(N, 2, *CurDAG).getNode() != nullptr;
254}], VSPLTISH_get_imm>;
255
256// VSPLTISW_get_imm xform function: convert build_vector to VSPLTISW imm.
257def VSPLTISW_get_imm : SDNodeXForm<build_vector, [{
258  return PPC::get_VSPLTI_elt(N, 4, *CurDAG);
259}]>;
260def vecspltisw : PatLeaf<(build_vector), [{
261  return PPC::get_VSPLTI_elt(N, 4, *CurDAG).getNode() != nullptr;
262}], VSPLTISW_get_imm>;
263
264def immEQOneV : PatLeaf<(build_vector), [{
265  if (ConstantSDNode *C = cast<BuildVectorSDNode>(N)->getConstantSplatNode())
266    return C->isOne();
267  return false;
268}]>;
269//===----------------------------------------------------------------------===//
270// Helpers for defining instructions that directly correspond to intrinsics.
271
272// VA1a_Int_Ty - A VAForm_1a intrinsic definition of specific type.
273class VA1a_Int_Ty<bits<6> xo, string opc, Intrinsic IntID, ValueType Ty>
274  : VAForm_1a<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
275              !strconcat(opc, " $vD, $vA, $vB, $vC"), IIC_VecFP,
276                       [(set Ty:$vD, (IntID Ty:$vA, Ty:$vB, Ty:$vC))]>;
277
278// VA1a_Int_Ty2 - A VAForm_1a intrinsic definition where the type of the
279// inputs doesn't match the type of the output.
280class VA1a_Int_Ty2<bits<6> xo, string opc, Intrinsic IntID, ValueType OutTy,
281                   ValueType InTy>
282  : VAForm_1a<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
283              !strconcat(opc, " $vD, $vA, $vB, $vC"), IIC_VecFP,
284                       [(set OutTy:$vD, (IntID InTy:$vA, InTy:$vB, InTy:$vC))]>;
285
286// VA1a_Int_Ty3 - A VAForm_1a intrinsic definition where there are two
287// input types and an output type.
288class VA1a_Int_Ty3<bits<6> xo, string opc, Intrinsic IntID, ValueType OutTy,
289                   ValueType In1Ty, ValueType In2Ty>
290  : VAForm_1a<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
291              !strconcat(opc, " $vD, $vA, $vB, $vC"), IIC_VecFP,
292                       [(set OutTy:$vD,
293                         (IntID In1Ty:$vA, In1Ty:$vB, In2Ty:$vC))]>;
294
295// VX1_Int_Ty - A VXForm_1 intrinsic definition of specific type.
296class VX1_Int_Ty<bits<11> xo, string opc, Intrinsic IntID, ValueType Ty>
297  : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
298             !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP,
299             [(set Ty:$vD, (IntID Ty:$vA, Ty:$vB))]>;
300
301// VX1_Int_Ty2 - A VXForm_1 intrinsic definition where the type of the
302// inputs doesn't match the type of the output.
303class VX1_Int_Ty2<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
304                  ValueType InTy>
305  : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
306             !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP,
307             [(set OutTy:$vD, (IntID InTy:$vA, InTy:$vB))]>;
308
309// VX1_Int_Ty3 - A VXForm_1 intrinsic definition where there are two
310// input types and an output type.
311class VX1_Int_Ty3<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
312                  ValueType In1Ty, ValueType In2Ty>
313  : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
314             !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP,
315             [(set OutTy:$vD, (IntID In1Ty:$vA, In2Ty:$vB))]>;
316
317// VX2_Int_SP - A VXForm_2 intrinsic definition of vector single-precision type.
318class VX2_Int_SP<bits<11> xo, string opc, Intrinsic IntID>
319  : VXForm_2<xo, (outs vrrc:$vD), (ins vrrc:$vB),
320             !strconcat(opc, " $vD, $vB"), IIC_VecFP,
321             [(set v4f32:$vD, (IntID v4f32:$vB))]>;
322
323// VX2_Int_Ty2 - A VXForm_2 intrinsic definition where the type of the
324// inputs doesn't match the type of the output.
325class VX2_Int_Ty2<bits<11> xo, string opc, Intrinsic IntID, ValueType OutTy,
326                  ValueType InTy>
327  : VXForm_2<xo, (outs vrrc:$vD), (ins vrrc:$vB),
328             !strconcat(opc, " $vD, $vB"), IIC_VecFP,
329             [(set OutTy:$vD, (IntID InTy:$vB))]>;
330
331class VXBX_Int_Ty<bits<11> xo, string opc, Intrinsic IntID, ValueType Ty>
332  : VXForm_BX<xo, (outs vrrc:$vD), (ins vrrc:$vA),
333             !strconcat(opc, " $vD, $vA"), IIC_VecFP,
334             [(set Ty:$vD, (IntID Ty:$vA))]>;
335
336class VXCR_Int_Ty<bits<11> xo, string opc, Intrinsic IntID, ValueType Ty>
337  : VXForm_CR<xo, (outs vrrc:$vD), (ins vrrc:$vA, u1imm:$ST, u4imm:$SIX),
338              !strconcat(opc, " $vD, $vA, $ST, $SIX"), IIC_VecFP,
339              [(set Ty:$vD, (IntID Ty:$vA, timm:$ST, timm:$SIX))]>;
340
341//===----------------------------------------------------------------------===//
342// Instruction Definitions.
343
344def HasAltivec : Predicate<"PPCSubTarget->hasAltivec()">;
345let Predicates = [HasAltivec] in {
346
347def DSS      : DSS_Form<0, 822, (outs), (ins u5imm:$STRM),
348                        "dss $STRM", IIC_LdStLoad /*FIXME*/, [(int_ppc_altivec_dss imm:$STRM)]>,
349                        Deprecated<DeprecatedDST> {
350  let A = 0;
351  let B = 0;
352}
353
354def DSSALL   : DSS_Form<1, 822, (outs), (ins),
355                        "dssall", IIC_LdStLoad /*FIXME*/, [(int_ppc_altivec_dssall)]>,
356                        Deprecated<DeprecatedDST> {
357  let STRM = 0;
358  let A = 0;
359  let B = 0;
360}
361
362def DST      : DSS_Form<0, 342, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB),
363                        "dst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
364                        [(int_ppc_altivec_dst i32:$rA, i32:$rB, imm:$STRM)]>,
365                        Deprecated<DeprecatedDST>;
366
367def DSTT     : DSS_Form<1, 342, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB),
368                        "dstt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
369                        [(int_ppc_altivec_dstt i32:$rA, i32:$rB, imm:$STRM)]>,
370                        Deprecated<DeprecatedDST>;
371
372def DSTST    : DSS_Form<0, 374, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB),
373                        "dstst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
374                        [(int_ppc_altivec_dstst i32:$rA, i32:$rB, imm:$STRM)]>,
375                        Deprecated<DeprecatedDST>;
376
377def DSTSTT   : DSS_Form<1, 374, (outs), (ins u5imm:$STRM, gprc:$rA, gprc:$rB),
378                        "dststt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
379                        [(int_ppc_altivec_dststt i32:$rA, i32:$rB, imm:$STRM)]>,
380                        Deprecated<DeprecatedDST>;
381
382let isCodeGenOnly = 1 in {
383  // The very same instructions as above, but formally matching 64bit registers.
384  def DST64    : DSS_Form<0, 342, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB),
385                          "dst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
386                          [(int_ppc_altivec_dst i64:$rA, i32:$rB, imm:$STRM)]>,
387                          Deprecated<DeprecatedDST>;
388
389  def DSTT64   : DSS_Form<1, 342, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB),
390                          "dstt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
391                          [(int_ppc_altivec_dstt i64:$rA, i32:$rB, imm:$STRM)]>,
392                          Deprecated<DeprecatedDST>;
393
394  def DSTST64  : DSS_Form<0, 374, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB),
395                          "dstst $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
396                          [(int_ppc_altivec_dstst i64:$rA, i32:$rB,
397                                                  imm:$STRM)]>,
398                          Deprecated<DeprecatedDST>;
399
400  def DSTSTT64 : DSS_Form<1, 374, (outs), (ins u5imm:$STRM, g8rc:$rA, gprc:$rB),
401                          "dststt $rA, $rB, $STRM", IIC_LdStLoad /*FIXME*/,
402                          [(int_ppc_altivec_dststt i64:$rA, i32:$rB,
403                                                   imm:$STRM)]>,
404                          Deprecated<DeprecatedDST>;
405}
406
407def MFVSCR : VXForm_4<1540, (outs vrrc:$vD), (ins),
408                      "mfvscr $vD", IIC_LdStStore,
409                      [(set v8i16:$vD, (int_ppc_altivec_mfvscr))]>;
410def MTVSCR : VXForm_5<1604, (outs), (ins vrrc:$vB),
411                      "mtvscr $vB", IIC_LdStLoad,
412                      [(int_ppc_altivec_mtvscr v4i32:$vB)]>;
413
414let PPC970_Unit = 2, mayLoad = 1, mayStore = 0 in {  // Loads.
415def LVEBX: XForm_1_memOp<31,   7, (outs vrrc:$vD), (ins memrr:$src),
416                   "lvebx $vD, $src", IIC_LdStLoad,
417                   [(set v16i8:$vD, (int_ppc_altivec_lvebx xoaddr:$src))]>;
418def LVEHX: XForm_1_memOp<31,  39, (outs vrrc:$vD), (ins memrr:$src),
419                   "lvehx $vD, $src", IIC_LdStLoad,
420                   [(set v8i16:$vD, (int_ppc_altivec_lvehx xoaddr:$src))]>;
421def LVEWX: XForm_1_memOp<31,  71, (outs vrrc:$vD), (ins memrr:$src),
422                   "lvewx $vD, $src", IIC_LdStLoad,
423                   [(set v4i32:$vD, (int_ppc_altivec_lvewx xoaddr:$src))]>;
424def LVX  : XForm_1_memOp<31, 103, (outs vrrc:$vD), (ins memrr:$src),
425                   "lvx $vD, $src", IIC_LdStLoad,
426                   [(set v4i32:$vD, (int_ppc_altivec_lvx xoaddr:$src))]>;
427def LVXL : XForm_1_memOp<31, 359, (outs vrrc:$vD), (ins memrr:$src),
428                   "lvxl $vD, $src", IIC_LdStLoad,
429                   [(set v4i32:$vD, (int_ppc_altivec_lvxl xoaddr:$src))]>;
430}
431
432def LVSL : XForm_1_memOp<31,   6, (outs vrrc:$vD), (ins memrr:$src),
433                   "lvsl $vD, $src", IIC_LdStLoad,
434                   [(set v16i8:$vD, (int_ppc_altivec_lvsl xoaddr:$src))]>,
435                   PPC970_Unit_LSU;
436def LVSR : XForm_1_memOp<31,  38, (outs vrrc:$vD), (ins memrr:$src),
437                   "lvsr $vD, $src", IIC_LdStLoad,
438                   [(set v16i8:$vD, (int_ppc_altivec_lvsr xoaddr:$src))]>,
439                   PPC970_Unit_LSU;
440
441let PPC970_Unit = 2, mayStore = 1, mayLoad = 0 in {   // Stores.
442def STVEBX: XForm_8_memOp<31, 135, (outs), (ins vrrc:$rS, memrr:$dst),
443                   "stvebx $rS, $dst", IIC_LdStStore,
444                   [(int_ppc_altivec_stvebx v16i8:$rS, xoaddr:$dst)]>;
445def STVEHX: XForm_8_memOp<31, 167, (outs), (ins vrrc:$rS, memrr:$dst),
446                   "stvehx $rS, $dst", IIC_LdStStore,
447                   [(int_ppc_altivec_stvehx v8i16:$rS, xoaddr:$dst)]>;
448def STVEWX: XForm_8_memOp<31, 199, (outs), (ins vrrc:$rS, memrr:$dst),
449                   "stvewx $rS, $dst", IIC_LdStStore,
450                   [(int_ppc_altivec_stvewx v4i32:$rS, xoaddr:$dst)]>;
451def STVX  : XForm_8_memOp<31, 231, (outs), (ins vrrc:$rS, memrr:$dst),
452                   "stvx $rS, $dst", IIC_LdStStore,
453                   [(int_ppc_altivec_stvx v4i32:$rS, xoaddr:$dst)]>;
454def STVXL : XForm_8_memOp<31, 487, (outs), (ins vrrc:$rS, memrr:$dst),
455                   "stvxl $rS, $dst", IIC_LdStStore,
456                   [(int_ppc_altivec_stvxl v4i32:$rS, xoaddr:$dst)]>;
457}
458
459let PPC970_Unit = 5 in {  // VALU Operations.
460// VA-Form instructions.  3-input AltiVec ops.
461let isCommutable = 1 in {
462def VMADDFP : VAForm_1<46, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vC, vrrc:$vB),
463                       "vmaddfp $vD, $vA, $vC, $vB", IIC_VecFP,
464                       [(set v4f32:$vD,
465                        (fma v4f32:$vA, v4f32:$vC, v4f32:$vB))]>;
466
467// FIXME: The fma+fneg pattern won't match because fneg is not legal.
468def VNMSUBFP: VAForm_1<47, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vC, vrrc:$vB),
469                       "vnmsubfp $vD, $vA, $vC, $vB", IIC_VecFP,
470                       [(set v4f32:$vD, (fneg (fma v4f32:$vA, v4f32:$vC,
471                                                  (fneg v4f32:$vB))))]>;
472
473def VMHADDSHS  : VA1a_Int_Ty<32, "vmhaddshs", int_ppc_altivec_vmhaddshs, v8i16>;
474def VMHRADDSHS : VA1a_Int_Ty<33, "vmhraddshs", int_ppc_altivec_vmhraddshs,
475                             v8i16>;
476def VMLADDUHM  : VA1a_Int_Ty<34, "vmladduhm", int_ppc_altivec_vmladduhm, v8i16>;
477} // isCommutable
478
479def VPERM      : VA1a_Int_Ty3<43, "vperm", int_ppc_altivec_vperm,
480                              v4i32, v4i32, v16i8>;
481def VSEL       : VA1a_Int_Ty<42, "vsel",  int_ppc_altivec_vsel, v4i32>;
482
483// Shuffles.
484def VSLDOI  : VAForm_2<44, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, u4imm:$SH),
485                       "vsldoi $vD, $vA, $vB, $SH", IIC_VecFP,
486                       [(set v16i8:$vD,
487                         (PPCvecshl v16i8:$vA, v16i8:$vB, imm32SExt16:$SH))]>;
488
489// VX-Form instructions.  AltiVec arithmetic ops.
490let isCommutable = 1 in {
491def VADDFP : VXForm_1<10, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
492                      "vaddfp $vD, $vA, $vB", IIC_VecFP,
493                      [(set v4f32:$vD, (fadd v4f32:$vA, v4f32:$vB))]>;
494
495def VADDUBM : VXForm_1<0, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
496                      "vaddubm $vD, $vA, $vB", IIC_VecGeneral,
497                      [(set v16i8:$vD, (add v16i8:$vA, v16i8:$vB))]>;
498def VADDUHM : VXForm_1<64, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
499                      "vadduhm $vD, $vA, $vB", IIC_VecGeneral,
500                      [(set v8i16:$vD, (add v8i16:$vA, v8i16:$vB))]>;
501def VADDUWM : VXForm_1<128, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
502                      "vadduwm $vD, $vA, $vB", IIC_VecGeneral,
503                      [(set v4i32:$vD, (add v4i32:$vA, v4i32:$vB))]>;
504
505def VADDCUW : VX1_Int_Ty<384, "vaddcuw", int_ppc_altivec_vaddcuw, v4i32>;
506def VADDSBS : VX1_Int_Ty<768, "vaddsbs", int_ppc_altivec_vaddsbs, v16i8>;
507def VADDSHS : VX1_Int_Ty<832, "vaddshs", int_ppc_altivec_vaddshs, v8i16>;
508def VADDSWS : VX1_Int_Ty<896, "vaddsws", int_ppc_altivec_vaddsws, v4i32>;
509def VADDUBS : VX1_Int_Ty<512, "vaddubs", int_ppc_altivec_vaddubs, v16i8>;
510def VADDUHS : VX1_Int_Ty<576, "vadduhs", int_ppc_altivec_vadduhs, v8i16>;
511def VADDUWS : VX1_Int_Ty<640, "vadduws", int_ppc_altivec_vadduws, v4i32>;
512} // isCommutable
513
514let isCommutable = 1 in
515def VAND : VXForm_1<1028, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
516                    "vand $vD, $vA, $vB", IIC_VecFP,
517                    [(set v4i32:$vD, (and v4i32:$vA, v4i32:$vB))]>;
518def VANDC : VXForm_1<1092, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
519                     "vandc $vD, $vA, $vB", IIC_VecFP,
520                     [(set v4i32:$vD, (and v4i32:$vA,
521                                           (vnot_ppc v4i32:$vB)))]>;
522
523def VCFSX  : VXForm_1<842, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
524                      "vcfsx $vD, $vB, $UIMM", IIC_VecFP,
525                      [(set v4f32:$vD,
526                             (int_ppc_altivec_vcfsx v4i32:$vB, timm:$UIMM))]>;
527def VCFUX  : VXForm_1<778, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
528                      "vcfux $vD, $vB, $UIMM", IIC_VecFP,
529                      [(set v4f32:$vD,
530                             (int_ppc_altivec_vcfux v4i32:$vB, timm:$UIMM))]>;
531def VCTSXS : VXForm_1<970, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
532                      "vctsxs $vD, $vB, $UIMM", IIC_VecFP,
533                      [(set v4i32:$vD,
534                             (int_ppc_altivec_vctsxs v4f32:$vB, timm:$UIMM))]>;
535def VCTUXS : VXForm_1<906, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
536                      "vctuxs $vD, $vB, $UIMM", IIC_VecFP,
537                      [(set v4i32:$vD,
538                             (int_ppc_altivec_vctuxs v4f32:$vB, timm:$UIMM))]>;
539
540// Defines with the UIM field set to 0 for floating-point
541// to integer (fp_to_sint/fp_to_uint) conversions and integer
542// to floating-point (sint_to_fp/uint_to_fp) conversions.
543let isCodeGenOnly = 1, VA = 0 in {
544def VCFSX_0 : VXForm_1<842, (outs vrrc:$vD), (ins vrrc:$vB),
545                       "vcfsx $vD, $vB, 0", IIC_VecFP,
546                       [(set v4f32:$vD,
547                             (int_ppc_altivec_vcfsx v4i32:$vB, 0))]>;
548def VCTUXS_0 : VXForm_1<906, (outs vrrc:$vD), (ins vrrc:$vB),
549                        "vctuxs $vD, $vB, 0", IIC_VecFP,
550                        [(set v4i32:$vD,
551                               (int_ppc_altivec_vctuxs v4f32:$vB, 0))]>;
552def VCFUX_0 : VXForm_1<778, (outs vrrc:$vD), (ins vrrc:$vB),
553                       "vcfux $vD, $vB, 0", IIC_VecFP,
554                       [(set v4f32:$vD,
555                               (int_ppc_altivec_vcfux v4i32:$vB, 0))]>;
556def VCTSXS_0 : VXForm_1<970, (outs vrrc:$vD), (ins vrrc:$vB),
557                      "vctsxs $vD, $vB, 0", IIC_VecFP,
558                      [(set v4i32:$vD,
559                             (int_ppc_altivec_vctsxs v4f32:$vB, 0))]>;
560}
561def VEXPTEFP : VX2_Int_SP<394, "vexptefp", int_ppc_altivec_vexptefp>;
562def VLOGEFP  : VX2_Int_SP<458, "vlogefp",  int_ppc_altivec_vlogefp>;
563
564let isCommutable = 1 in {
565def VAVGSB : VX1_Int_Ty<1282, "vavgsb", int_ppc_altivec_vavgsb, v16i8>;
566def VAVGSH : VX1_Int_Ty<1346, "vavgsh", int_ppc_altivec_vavgsh, v8i16>;
567def VAVGSW : VX1_Int_Ty<1410, "vavgsw", int_ppc_altivec_vavgsw, v4i32>;
568def VAVGUB : VX1_Int_Ty<1026, "vavgub", int_ppc_altivec_vavgub, v16i8>;
569def VAVGUH : VX1_Int_Ty<1090, "vavguh", int_ppc_altivec_vavguh, v8i16>;
570def VAVGUW : VX1_Int_Ty<1154, "vavguw", int_ppc_altivec_vavguw, v4i32>;
571
572def VMAXFP : VX1_Int_Ty<1034, "vmaxfp", int_ppc_altivec_vmaxfp, v4f32>;
573def VMAXSB : VX1_Int_Ty< 258, "vmaxsb", int_ppc_altivec_vmaxsb, v16i8>;
574def VMAXSH : VX1_Int_Ty< 322, "vmaxsh", int_ppc_altivec_vmaxsh, v8i16>;
575def VMAXSW : VX1_Int_Ty< 386, "vmaxsw", int_ppc_altivec_vmaxsw, v4i32>;
576def VMAXUB : VX1_Int_Ty<   2, "vmaxub", int_ppc_altivec_vmaxub, v16i8>;
577def VMAXUH : VX1_Int_Ty<  66, "vmaxuh", int_ppc_altivec_vmaxuh, v8i16>;
578def VMAXUW : VX1_Int_Ty< 130, "vmaxuw", int_ppc_altivec_vmaxuw, v4i32>;
579def VMINFP : VX1_Int_Ty<1098, "vminfp", int_ppc_altivec_vminfp, v4f32>;
580def VMINSB : VX1_Int_Ty< 770, "vminsb", int_ppc_altivec_vminsb, v16i8>;
581def VMINSH : VX1_Int_Ty< 834, "vminsh", int_ppc_altivec_vminsh, v8i16>;
582def VMINSW : VX1_Int_Ty< 898, "vminsw", int_ppc_altivec_vminsw, v4i32>;
583def VMINUB : VX1_Int_Ty< 514, "vminub", int_ppc_altivec_vminub, v16i8>;
584def VMINUH : VX1_Int_Ty< 578, "vminuh", int_ppc_altivec_vminuh, v8i16>;
585def VMINUW : VX1_Int_Ty< 642, "vminuw", int_ppc_altivec_vminuw, v4i32>;
586} // isCommutable
587
588def VMRGHB : VXForm_1< 12, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
589                      "vmrghb $vD, $vA, $vB", IIC_VecFP,
590                      [(set v16i8:$vD, (vmrghb_shuffle v16i8:$vA, v16i8:$vB))]>;
591def VMRGHH : VXForm_1< 76, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
592                      "vmrghh $vD, $vA, $vB", IIC_VecFP,
593                      [(set v16i8:$vD, (vmrghh_shuffle v16i8:$vA, v16i8:$vB))]>;
594def VMRGHW : VXForm_1<140, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
595                      "vmrghw $vD, $vA, $vB", IIC_VecFP,
596                      [(set v16i8:$vD, (vmrghw_shuffle v16i8:$vA, v16i8:$vB))]>;
597def VMRGLB : VXForm_1<268, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
598                      "vmrglb $vD, $vA, $vB", IIC_VecFP,
599                      [(set v16i8:$vD, (vmrglb_shuffle v16i8:$vA, v16i8:$vB))]>;
600def VMRGLH : VXForm_1<332, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
601                      "vmrglh $vD, $vA, $vB", IIC_VecFP,
602                      [(set v16i8:$vD, (vmrglh_shuffle v16i8:$vA, v16i8:$vB))]>;
603def VMRGLW : VXForm_1<396, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
604                      "vmrglw $vD, $vA, $vB", IIC_VecFP,
605                      [(set v16i8:$vD, (vmrglw_shuffle v16i8:$vA, v16i8:$vB))]>;
606
607def VMSUMMBM : VA1a_Int_Ty3<37, "vmsummbm", int_ppc_altivec_vmsummbm,
608                            v4i32, v16i8, v4i32>;
609def VMSUMSHM : VA1a_Int_Ty3<40, "vmsumshm", int_ppc_altivec_vmsumshm,
610                            v4i32, v8i16, v4i32>;
611def VMSUMSHS : VA1a_Int_Ty3<41, "vmsumshs", int_ppc_altivec_vmsumshs,
612                            v4i32, v8i16, v4i32>;
613def VMSUMUBM : VA1a_Int_Ty3<36, "vmsumubm", int_ppc_altivec_vmsumubm,
614                            v4i32, v16i8, v4i32>;
615def VMSUMUHM : VA1a_Int_Ty3<38, "vmsumuhm", int_ppc_altivec_vmsumuhm,
616                            v4i32, v8i16, v4i32>;
617def VMSUMUHS : VA1a_Int_Ty3<39, "vmsumuhs", int_ppc_altivec_vmsumuhs,
618                            v4i32, v8i16, v4i32>;
619
620let isCommutable = 1 in {
621def VMULESB : VX1_Int_Ty2<776, "vmulesb", int_ppc_altivec_vmulesb,
622                          v8i16, v16i8>;
623def VMULESH : VX1_Int_Ty2<840, "vmulesh", int_ppc_altivec_vmulesh,
624                          v4i32, v8i16>;
625def VMULEUB : VX1_Int_Ty2<520, "vmuleub", int_ppc_altivec_vmuleub,
626                          v8i16, v16i8>;
627def VMULEUH : VX1_Int_Ty2<584, "vmuleuh", int_ppc_altivec_vmuleuh,
628                          v4i32, v8i16>;
629def VMULOSB : VX1_Int_Ty2<264, "vmulosb", int_ppc_altivec_vmulosb,
630                          v8i16, v16i8>;
631def VMULOSH : VX1_Int_Ty2<328, "vmulosh", int_ppc_altivec_vmulosh,
632                          v4i32, v8i16>;
633def VMULOUB : VX1_Int_Ty2<  8, "vmuloub", int_ppc_altivec_vmuloub,
634                          v8i16, v16i8>;
635def VMULOUH : VX1_Int_Ty2< 72, "vmulouh", int_ppc_altivec_vmulouh,
636                          v4i32, v8i16>;
637} // isCommutable
638
639def VREFP     : VX2_Int_SP<266, "vrefp",     int_ppc_altivec_vrefp>;
640def VRFIM     : VX2_Int_SP<714, "vrfim",     int_ppc_altivec_vrfim>;
641def VRFIN     : VX2_Int_SP<522, "vrfin",     int_ppc_altivec_vrfin>;
642def VRFIP     : VX2_Int_SP<650, "vrfip",     int_ppc_altivec_vrfip>;
643def VRFIZ     : VX2_Int_SP<586, "vrfiz",     int_ppc_altivec_vrfiz>;
644def VRSQRTEFP : VX2_Int_SP<330, "vrsqrtefp", int_ppc_altivec_vrsqrtefp>;
645
646def VSUBCUW : VX1_Int_Ty<1408, "vsubcuw", int_ppc_altivec_vsubcuw, v4i32>;
647
648def VSUBFP  : VXForm_1<74, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
649                      "vsubfp $vD, $vA, $vB", IIC_VecGeneral,
650                      [(set v4f32:$vD, (fsub v4f32:$vA, v4f32:$vB))]>;
651def VSUBUBM : VXForm_1<1024, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
652                      "vsububm $vD, $vA, $vB", IIC_VecGeneral,
653                      [(set v16i8:$vD, (sub v16i8:$vA, v16i8:$vB))]>;
654def VSUBUHM : VXForm_1<1088, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
655                      "vsubuhm $vD, $vA, $vB", IIC_VecGeneral,
656                      [(set v8i16:$vD, (sub v8i16:$vA, v8i16:$vB))]>;
657def VSUBUWM : VXForm_1<1152, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
658                      "vsubuwm $vD, $vA, $vB", IIC_VecGeneral,
659                      [(set v4i32:$vD, (sub v4i32:$vA, v4i32:$vB))]>;
660
661def VSUBSBS : VX1_Int_Ty<1792, "vsubsbs" , int_ppc_altivec_vsubsbs, v16i8>;
662def VSUBSHS : VX1_Int_Ty<1856, "vsubshs" , int_ppc_altivec_vsubshs, v8i16>;
663def VSUBSWS : VX1_Int_Ty<1920, "vsubsws" , int_ppc_altivec_vsubsws, v4i32>;
664def VSUBUBS : VX1_Int_Ty<1536, "vsububs" , int_ppc_altivec_vsububs, v16i8>;
665def VSUBUHS : VX1_Int_Ty<1600, "vsubuhs" , int_ppc_altivec_vsubuhs, v8i16>;
666def VSUBUWS : VX1_Int_Ty<1664, "vsubuws" , int_ppc_altivec_vsubuws, v4i32>;
667
668def VSUMSWS : VX1_Int_Ty<1928, "vsumsws" , int_ppc_altivec_vsumsws, v4i32>;
669def VSUM2SWS: VX1_Int_Ty<1672, "vsum2sws", int_ppc_altivec_vsum2sws, v4i32>;
670
671def VSUM4SBS: VX1_Int_Ty3<1800, "vsum4sbs", int_ppc_altivec_vsum4sbs,
672                          v4i32, v16i8, v4i32>;
673def VSUM4SHS: VX1_Int_Ty3<1608, "vsum4shs", int_ppc_altivec_vsum4shs,
674                          v4i32, v8i16, v4i32>;
675def VSUM4UBS: VX1_Int_Ty3<1544, "vsum4ubs", int_ppc_altivec_vsum4ubs,
676                          v4i32, v16i8, v4i32>;
677
678def VNOR : VXForm_1<1284, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
679                    "vnor $vD, $vA, $vB", IIC_VecFP,
680                    [(set v4i32:$vD, (vnot_ppc (or v4i32:$vA,
681                                                   v4i32:$vB)))]>;
682let isCommutable = 1 in {
683def VOR : VXForm_1<1156, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
684                      "vor $vD, $vA, $vB", IIC_VecFP,
685                      [(set v4i32:$vD, (or v4i32:$vA, v4i32:$vB))]>;
686def VXOR : VXForm_1<1220, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
687                      "vxor $vD, $vA, $vB", IIC_VecFP,
688                      [(set v4i32:$vD, (xor v4i32:$vA, v4i32:$vB))]>;
689} // isCommutable
690
691def VRLB   : VX1_Int_Ty<   4, "vrlb", int_ppc_altivec_vrlb, v16i8>;
692def VRLH   : VX1_Int_Ty<  68, "vrlh", int_ppc_altivec_vrlh, v8i16>;
693def VRLW   : VX1_Int_Ty< 132, "vrlw", int_ppc_altivec_vrlw, v4i32>;
694
695def VSL    : VX1_Int_Ty< 452, "vsl" , int_ppc_altivec_vsl,  v4i32 >;
696def VSLO   : VX1_Int_Ty<1036, "vslo", int_ppc_altivec_vslo, v4i32>;
697
698def VSLB   : VX1_Int_Ty< 260, "vslb", int_ppc_altivec_vslb, v16i8>;
699def VSLH   : VX1_Int_Ty< 324, "vslh", int_ppc_altivec_vslh, v8i16>;
700def VSLW   : VX1_Int_Ty< 388, "vslw", int_ppc_altivec_vslw, v4i32>;
701
702def VSPLTB : VXForm_1<524, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
703                      "vspltb $vD, $vB, $UIMM", IIC_VecPerm,
704                      [(set v16i8:$vD,
705                        (vspltb_shuffle:$UIMM v16i8:$vB, (undef)))]>;
706def VSPLTH : VXForm_1<588, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
707                      "vsplth $vD, $vB, $UIMM", IIC_VecPerm,
708                      [(set v16i8:$vD,
709                        (vsplth_shuffle:$UIMM v16i8:$vB, (undef)))]>;
710def VSPLTW : VXForm_1<652, (outs vrrc:$vD), (ins u5imm:$UIMM, vrrc:$vB),
711                      "vspltw $vD, $vB, $UIMM", IIC_VecPerm,
712                      [(set v16i8:$vD,
713                        (vspltw_shuffle:$UIMM v16i8:$vB, (undef)))]>;
714let isCodeGenOnly = 1, hasSideEffects = 0 in {
715  def VSPLTBs : VXForm_1<524, (outs vrrc:$vD), (ins u5imm:$UIMM, vfrc:$vB),
716                         "vspltb $vD, $vB, $UIMM", IIC_VecPerm, []>;
717  def VSPLTHs : VXForm_1<588, (outs vrrc:$vD), (ins u5imm:$UIMM, vfrc:$vB),
718                         "vsplth $vD, $vB, $UIMM", IIC_VecPerm, []>;
719}
720
721def VSR    : VX1_Int_Ty< 708, "vsr"  , int_ppc_altivec_vsr,  v4i32>;
722def VSRO   : VX1_Int_Ty<1100, "vsro" , int_ppc_altivec_vsro, v4i32>;
723
724def VSRAB  : VX1_Int_Ty< 772, "vsrab", int_ppc_altivec_vsrab, v16i8>;
725def VSRAH  : VX1_Int_Ty< 836, "vsrah", int_ppc_altivec_vsrah, v8i16>;
726def VSRAW  : VX1_Int_Ty< 900, "vsraw", int_ppc_altivec_vsraw, v4i32>;
727def VSRB   : VX1_Int_Ty< 516, "vsrb" , int_ppc_altivec_vsrb , v16i8>;
728def VSRH   : VX1_Int_Ty< 580, "vsrh" , int_ppc_altivec_vsrh , v8i16>;
729def VSRW   : VX1_Int_Ty< 644, "vsrw" , int_ppc_altivec_vsrw , v4i32>;
730
731
732def VSPLTISB : VXForm_3<780, (outs vrrc:$vD), (ins s5imm:$SIMM),
733                       "vspltisb $vD, $SIMM", IIC_VecPerm,
734                       [(set v16i8:$vD, (v16i8 vecspltisb:$SIMM))]>;
735def VSPLTISH : VXForm_3<844, (outs vrrc:$vD), (ins s5imm:$SIMM),
736                       "vspltish $vD, $SIMM", IIC_VecPerm,
737                       [(set v8i16:$vD, (v8i16 vecspltish:$SIMM))]>;
738def VSPLTISW : VXForm_3<908, (outs vrrc:$vD), (ins s5imm:$SIMM),
739                       "vspltisw $vD, $SIMM", IIC_VecPerm,
740                       [(set v4i32:$vD, (v4i32 vecspltisw:$SIMM))]>;
741
742// Vector Pack.
743def VPKPX   : VX1_Int_Ty2<782, "vpkpx", int_ppc_altivec_vpkpx,
744                          v8i16, v4i32>;
745def VPKSHSS : VX1_Int_Ty2<398, "vpkshss", int_ppc_altivec_vpkshss,
746                          v16i8, v8i16>;
747def VPKSHUS : VX1_Int_Ty2<270, "vpkshus", int_ppc_altivec_vpkshus,
748                          v16i8, v8i16>;
749def VPKSWSS : VX1_Int_Ty2<462, "vpkswss", int_ppc_altivec_vpkswss,
750                          v8i16, v4i32>;
751def VPKSWUS : VX1_Int_Ty2<334, "vpkswus", int_ppc_altivec_vpkswus,
752                          v8i16, v4i32>;
753def VPKUHUM : VXForm_1<14, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
754                       "vpkuhum $vD, $vA, $vB", IIC_VecFP,
755                       [(set v16i8:$vD,
756                         (vpkuhum_shuffle v16i8:$vA, v16i8:$vB))]>;
757def VPKUHUS : VX1_Int_Ty2<142, "vpkuhus", int_ppc_altivec_vpkuhus,
758                          v16i8, v8i16>;
759def VPKUWUM : VXForm_1<78, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
760                       "vpkuwum $vD, $vA, $vB", IIC_VecFP,
761                       [(set v16i8:$vD,
762                         (vpkuwum_shuffle v16i8:$vA, v16i8:$vB))]>;
763def VPKUWUS : VX1_Int_Ty2<206, "vpkuwus", int_ppc_altivec_vpkuwus,
764                          v8i16, v4i32>;
765
766// Vector Unpack.
767def VUPKHPX : VX2_Int_Ty2<846, "vupkhpx", int_ppc_altivec_vupkhpx,
768                          v4i32, v8i16>;
769def VUPKHSB : VX2_Int_Ty2<526, "vupkhsb", int_ppc_altivec_vupkhsb,
770                          v8i16, v16i8>;
771def VUPKHSH : VX2_Int_Ty2<590, "vupkhsh", int_ppc_altivec_vupkhsh,
772                          v4i32, v8i16>;
773def VUPKLPX : VX2_Int_Ty2<974, "vupklpx", int_ppc_altivec_vupklpx,
774                          v4i32, v8i16>;
775def VUPKLSB : VX2_Int_Ty2<654, "vupklsb", int_ppc_altivec_vupklsb,
776                          v8i16, v16i8>;
777def VUPKLSH : VX2_Int_Ty2<718, "vupklsh", int_ppc_altivec_vupklsh,
778                          v4i32, v8i16>;
779
780
781// Altivec Comparisons.
782
783class VCMP<bits<10> xo, string asmstr, ValueType Ty>
784  : VXRForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), asmstr,
785              IIC_VecFPCompare,
786              [(set Ty:$vD, (Ty (PPCvcmp Ty:$vA, Ty:$vB, xo)))]>;
787class VCMPo<bits<10> xo, string asmstr, ValueType Ty>
788  : VXRForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB), asmstr,
789              IIC_VecFPCompare,
790              [(set Ty:$vD, (Ty (PPCvcmp_o Ty:$vA, Ty:$vB, xo)))]> {
791  let Defs = [CR6];
792  let RC = 1;
793}
794
795// f32 element comparisons.0
796def VCMPBFP   : VCMP <966, "vcmpbfp $vD, $vA, $vB"  , v4f32>;
797def VCMPBFP_rec  : VCMPo<966, "vcmpbfp. $vD, $vA, $vB" , v4f32>;
798def VCMPEQFP  : VCMP <198, "vcmpeqfp $vD, $vA, $vB" , v4f32>;
799def VCMPEQFP_rec : VCMPo<198, "vcmpeqfp. $vD, $vA, $vB", v4f32>;
800def VCMPGEFP  : VCMP <454, "vcmpgefp $vD, $vA, $vB" , v4f32>;
801def VCMPGEFP_rec : VCMPo<454, "vcmpgefp. $vD, $vA, $vB", v4f32>;
802def VCMPGTFP  : VCMP <710, "vcmpgtfp $vD, $vA, $vB" , v4f32>;
803def VCMPGTFP_rec : VCMPo<710, "vcmpgtfp. $vD, $vA, $vB", v4f32>;
804
805// i8 element comparisons.
806def VCMPEQUB  : VCMP <  6, "vcmpequb $vD, $vA, $vB" , v16i8>;
807def VCMPEQUB_rec : VCMPo<  6, "vcmpequb. $vD, $vA, $vB", v16i8>;
808def VCMPGTSB  : VCMP <774, "vcmpgtsb $vD, $vA, $vB" , v16i8>;
809def VCMPGTSB_rec : VCMPo<774, "vcmpgtsb. $vD, $vA, $vB", v16i8>;
810def VCMPGTUB  : VCMP <518, "vcmpgtub $vD, $vA, $vB" , v16i8>;
811def VCMPGTUB_rec : VCMPo<518, "vcmpgtub. $vD, $vA, $vB", v16i8>;
812
813// i16 element comparisons.
814def VCMPEQUH  : VCMP < 70, "vcmpequh $vD, $vA, $vB" , v8i16>;
815def VCMPEQUH_rec : VCMPo< 70, "vcmpequh. $vD, $vA, $vB", v8i16>;
816def VCMPGTSH  : VCMP <838, "vcmpgtsh $vD, $vA, $vB" , v8i16>;
817def VCMPGTSH_rec : VCMPo<838, "vcmpgtsh. $vD, $vA, $vB", v8i16>;
818def VCMPGTUH  : VCMP <582, "vcmpgtuh $vD, $vA, $vB" , v8i16>;
819def VCMPGTUH_rec : VCMPo<582, "vcmpgtuh. $vD, $vA, $vB", v8i16>;
820
821// i32 element comparisons.
822def VCMPEQUW  : VCMP <134, "vcmpequw $vD, $vA, $vB" , v4i32>;
823def VCMPEQUW_rec : VCMPo<134, "vcmpequw. $vD, $vA, $vB", v4i32>;
824def VCMPGTSW  : VCMP <902, "vcmpgtsw $vD, $vA, $vB" , v4i32>;
825def VCMPGTSW_rec : VCMPo<902, "vcmpgtsw. $vD, $vA, $vB", v4i32>;
826def VCMPGTUW  : VCMP <646, "vcmpgtuw $vD, $vA, $vB" , v4i32>;
827def VCMPGTUW_rec : VCMPo<646, "vcmpgtuw. $vD, $vA, $vB", v4i32>;
828
829let isCodeGenOnly = 1, isMoveImm = 1, isAsCheapAsAMove = 1,
830    isReMaterializable = 1 in {
831
832def V_SET0B : VXForm_setzero<1220, (outs vrrc:$vD), (ins),
833                      "vxor $vD, $vD, $vD", IIC_VecFP,
834                      [(set v16i8:$vD, (v16i8 immAllZerosV))]>;
835def V_SET0H : VXForm_setzero<1220, (outs vrrc:$vD), (ins),
836                      "vxor $vD, $vD, $vD", IIC_VecFP,
837                      [(set v8i16:$vD, (v8i16 immAllZerosV))]>;
838def V_SET0  : VXForm_setzero<1220, (outs vrrc:$vD), (ins),
839                      "vxor $vD, $vD, $vD", IIC_VecFP,
840                      [(set v4i32:$vD, (v4i32 immAllZerosV))]>;
841
842let IMM=-1 in {
843def V_SETALLONESB : VXForm_3<908, (outs vrrc:$vD), (ins),
844                      "vspltisw $vD, -1", IIC_VecFP,
845                      [(set v16i8:$vD, (v16i8 immAllOnesV))]>;
846def V_SETALLONESH : VXForm_3<908, (outs vrrc:$vD), (ins),
847                      "vspltisw $vD, -1", IIC_VecFP,
848                      [(set v8i16:$vD, (v8i16 immAllOnesV))]>;
849def V_SETALLONES  : VXForm_3<908, (outs vrrc:$vD), (ins),
850                      "vspltisw $vD, -1", IIC_VecFP,
851                      [(set v4i32:$vD, (v4i32 immAllOnesV))]>;
852}
853}
854} // VALU Operations.
855
856//===----------------------------------------------------------------------===//
857// Additional Altivec Patterns
858//
859
860// Extended mnemonics
861def : InstAlias<"vmr $vD, $vA", (VOR vrrc:$vD, vrrc:$vA, vrrc:$vA)>;
862def : InstAlias<"vnot $vD, $vA", (VNOR vrrc:$vD, vrrc:$vA, vrrc:$vA)>;
863
864// Rotates.
865def : Pat<(v16i8 (rotl v16i8:$vA, v16i8:$vB)),
866          (v16i8 (VRLB v16i8:$vA, v16i8:$vB))>;
867def : Pat<(v8i16 (rotl v8i16:$vA, v8i16:$vB)),
868          (v8i16 (VRLH v8i16:$vA, v8i16:$vB))>;
869def : Pat<(v4i32 (rotl v4i32:$vA, v4i32:$vB)),
870          (v4i32 (VRLW v4i32:$vA, v4i32:$vB))>;
871
872// Loads.
873def : Pat<(v4i32 (load xoaddr:$src)), (LVX xoaddr:$src)>;
874
875// Stores.
876def : Pat<(store v4i32:$rS, xoaddr:$dst),
877          (STVX $rS, xoaddr:$dst)>;
878
879// Bit conversions.
880def : Pat<(v16i8 (bitconvert (v8i16 VRRC:$src))), (v16i8 VRRC:$src)>;
881def : Pat<(v16i8 (bitconvert (v4i32 VRRC:$src))), (v16i8 VRRC:$src)>;
882def : Pat<(v16i8 (bitconvert (v4f32 VRRC:$src))), (v16i8 VRRC:$src)>;
883def : Pat<(v16i8 (bitconvert (v2i64 VRRC:$src))), (v16i8 VRRC:$src)>;
884def : Pat<(v16i8 (bitconvert (v1i128 VRRC:$src))), (v16i8 VRRC:$src)>;
885
886def : Pat<(v8i16 (bitconvert (v16i8 VRRC:$src))), (v8i16 VRRC:$src)>;
887def : Pat<(v8i16 (bitconvert (v4i32 VRRC:$src))), (v8i16 VRRC:$src)>;
888def : Pat<(v8i16 (bitconvert (v4f32 VRRC:$src))), (v8i16 VRRC:$src)>;
889def : Pat<(v8i16 (bitconvert (v2i64 VRRC:$src))), (v8i16 VRRC:$src)>;
890def : Pat<(v8i16 (bitconvert (v1i128 VRRC:$src))), (v8i16 VRRC:$src)>;
891
892def : Pat<(v4i32 (bitconvert (v16i8 VRRC:$src))), (v4i32 VRRC:$src)>;
893def : Pat<(v4i32 (bitconvert (v8i16 VRRC:$src))), (v4i32 VRRC:$src)>;
894def : Pat<(v4i32 (bitconvert (v4f32 VRRC:$src))), (v4i32 VRRC:$src)>;
895def : Pat<(v4i32 (bitconvert (v2i64 VRRC:$src))), (v4i32 VRRC:$src)>;
896def : Pat<(v4i32 (bitconvert (v1i128 VRRC:$src))), (v4i32 VRRC:$src)>;
897
898def : Pat<(v4f32 (bitconvert (v16i8 VRRC:$src))), (v4f32 VRRC:$src)>;
899def : Pat<(v4f32 (bitconvert (v8i16 VRRC:$src))), (v4f32 VRRC:$src)>;
900def : Pat<(v4f32 (bitconvert (v4i32 VRRC:$src))), (v4f32 VRRC:$src)>;
901def : Pat<(v4f32 (bitconvert (v2i64 VRRC:$src))), (v4f32 VRRC:$src)>;
902def : Pat<(v4f32 (bitconvert (v1i128 VRRC:$src))), (v4f32 VRRC:$src)>;
903
904def : Pat<(v2i64 (bitconvert (v16i8 VRRC:$src))), (v2i64 VRRC:$src)>;
905def : Pat<(v2i64 (bitconvert (v8i16 VRRC:$src))), (v2i64 VRRC:$src)>;
906def : Pat<(v2i64 (bitconvert (v4i32 VRRC:$src))), (v2i64 VRRC:$src)>;
907def : Pat<(v2i64 (bitconvert (v4f32 VRRC:$src))), (v2i64 VRRC:$src)>;
908def : Pat<(v2i64 (bitconvert (v1i128 VRRC:$src))), (v2i64 VRRC:$src)>;
909
910def : Pat<(v1i128 (bitconvert (v16i8 VRRC:$src))), (v1i128 VRRC:$src)>;
911def : Pat<(v1i128 (bitconvert (v8i16 VRRC:$src))), (v1i128 VRRC:$src)>;
912def : Pat<(v1i128 (bitconvert (v4i32 VRRC:$src))), (v1i128 VRRC:$src)>;
913def : Pat<(v1i128 (bitconvert (v4f32 VRRC:$src))), (v1i128 VRRC:$src)>;
914def : Pat<(v1i128 (bitconvert (v2i64 VRRC:$src))), (v1i128 VRRC:$src)>;
915
916// Max/Min
917def : Pat<(v16i8 (umax v16i8:$src1, v16i8:$src2)),
918          (v16i8 (VMAXUB $src1, $src2))>;
919def : Pat<(v16i8 (smax v16i8:$src1, v16i8:$src2)),
920          (v16i8 (VMAXSB $src1, $src2))>;
921def : Pat<(v8i16 (umax v8i16:$src1, v8i16:$src2)),
922          (v8i16 (VMAXUH $src1, $src2))>;
923def : Pat<(v8i16 (smax v8i16:$src1, v8i16:$src2)),
924          (v8i16 (VMAXSH $src1, $src2))>;
925def : Pat<(v4i32 (umax v4i32:$src1, v4i32:$src2)),
926          (v4i32 (VMAXUW $src1, $src2))>;
927def : Pat<(v4i32 (smax v4i32:$src1, v4i32:$src2)),
928          (v4i32 (VMAXSW $src1, $src2))>;
929def : Pat<(v16i8 (umin v16i8:$src1, v16i8:$src2)),
930          (v16i8 (VMINUB $src1, $src2))>;
931def : Pat<(v16i8 (smin v16i8:$src1, v16i8:$src2)),
932          (v16i8 (VMINSB $src1, $src2))>;
933def : Pat<(v8i16 (umin v8i16:$src1, v8i16:$src2)),
934          (v8i16 (VMINUH $src1, $src2))>;
935def : Pat<(v8i16 (smin v8i16:$src1, v8i16:$src2)),
936          (v8i16 (VMINSH $src1, $src2))>;
937def : Pat<(v4i32 (umin v4i32:$src1, v4i32:$src2)),
938          (v4i32 (VMINUW $src1, $src2))>;
939def : Pat<(v4i32 (smin v4i32:$src1, v4i32:$src2)),
940          (v4i32 (VMINSW $src1, $src2))>;
941
942// Shuffles.
943
944// Match vsldoi(x,x), vpkuwum(x,x), vpkuhum(x,x)
945def:Pat<(vsldoi_unary_shuffle:$in v16i8:$vA, undef),
946        (VSLDOI $vA, $vA, (VSLDOI_unary_get_imm $in))>;
947def:Pat<(vpkuwum_unary_shuffle v16i8:$vA, undef),
948        (VPKUWUM $vA, $vA)>;
949def:Pat<(vpkuhum_unary_shuffle v16i8:$vA, undef),
950        (VPKUHUM $vA, $vA)>;
951def:Pat<(vsldoi_shuffle:$SH v16i8:$vA, v16i8:$vB),
952        (VSLDOI v16i8:$vA, v16i8:$vB, (VSLDOI_get_imm $SH))>;
953
954
955// Match vsldoi(y,x), vpkuwum(y,x), vpkuhum(y,x), i.e., swapped operands.
956// These fragments are matched for little-endian, where the inputs must
957// be swapped for correct semantics.
958def:Pat<(vsldoi_swapped_shuffle:$in v16i8:$vA, v16i8:$vB),
959        (VSLDOI $vB, $vA, (VSLDOI_swapped_get_imm $in))>;
960def:Pat<(vpkuwum_swapped_shuffle v16i8:$vA, v16i8:$vB),
961        (VPKUWUM $vB, $vA)>;
962def:Pat<(vpkuhum_swapped_shuffle v16i8:$vA, v16i8:$vB),
963        (VPKUHUM $vB, $vA)>;
964
965// Match vmrg*(x,x)
966def:Pat<(vmrglb_unary_shuffle v16i8:$vA, undef),
967        (VMRGLB $vA, $vA)>;
968def:Pat<(vmrglh_unary_shuffle v16i8:$vA, undef),
969        (VMRGLH $vA, $vA)>;
970def:Pat<(vmrglw_unary_shuffle v16i8:$vA, undef),
971        (VMRGLW $vA, $vA)>;
972def:Pat<(vmrghb_unary_shuffle v16i8:$vA, undef),
973        (VMRGHB $vA, $vA)>;
974def:Pat<(vmrghh_unary_shuffle v16i8:$vA, undef),
975        (VMRGHH $vA, $vA)>;
976def:Pat<(vmrghw_unary_shuffle v16i8:$vA, undef),
977        (VMRGHW $vA, $vA)>;
978
979// Match vmrg*(y,x), i.e., swapped operands.  These fragments
980// are matched for little-endian, where the inputs must be
981// swapped for correct semantics.
982def:Pat<(vmrglb_swapped_shuffle v16i8:$vA, v16i8:$vB),
983        (VMRGLB $vB, $vA)>;
984def:Pat<(vmrglh_swapped_shuffle v16i8:$vA, v16i8:$vB),
985        (VMRGLH $vB, $vA)>;
986def:Pat<(vmrglw_swapped_shuffle v16i8:$vA, v16i8:$vB),
987        (VMRGLW $vB, $vA)>;
988def:Pat<(vmrghb_swapped_shuffle v16i8:$vA, v16i8:$vB),
989        (VMRGHB $vB, $vA)>;
990def:Pat<(vmrghh_swapped_shuffle v16i8:$vA, v16i8:$vB),
991        (VMRGHH $vB, $vA)>;
992def:Pat<(vmrghw_swapped_shuffle v16i8:$vA, v16i8:$vB),
993        (VMRGHW $vB, $vA)>;
994
995// Logical Operations
996def : Pat<(vnot_ppc v4i32:$vA), (VNOR $vA, $vA)>;
997
998def : Pat<(vnot_ppc (or v4i32:$A, v4i32:$B)),
999          (VNOR $A, $B)>;
1000def : Pat<(and v4i32:$A, (vnot_ppc v4i32:$B)),
1001          (VANDC $A, $B)>;
1002
1003def : Pat<(fmul v4f32:$vA, v4f32:$vB),
1004          (VMADDFP $vA, $vB,
1005             (v4i32 (VSLW (v4i32 (V_SETALLONES)), (v4i32 (V_SETALLONES)))))>;
1006
1007// Fused multiply add and multiply sub for packed float.  These are represented
1008// separately from the real instructions above, for operations that must have
1009// the additional precision, such as Newton-Rhapson (used by divide, sqrt)
1010def : Pat<(PPCvmaddfp v4f32:$A, v4f32:$B, v4f32:$C),
1011          (VMADDFP $A, $B, $C)>;
1012def : Pat<(PPCvnmsubfp v4f32:$A, v4f32:$B, v4f32:$C),
1013          (VNMSUBFP $A, $B, $C)>;
1014
1015def : Pat<(int_ppc_altivec_vmaddfp v4f32:$A, v4f32:$B, v4f32:$C),
1016          (VMADDFP $A, $B, $C)>;
1017def : Pat<(int_ppc_altivec_vnmsubfp v4f32:$A, v4f32:$B, v4f32:$C),
1018          (VNMSUBFP $A, $B, $C)>;
1019
1020def : Pat<(PPCvperm v16i8:$vA, v16i8:$vB, v16i8:$vC),
1021          (VPERM $vA, $vB, $vC)>;
1022
1023def : Pat<(PPCfre v4f32:$A), (VREFP $A)>;
1024def : Pat<(PPCfrsqrte v4f32:$A), (VRSQRTEFP $A)>;
1025
1026// Vector shifts
1027def : Pat<(v16i8 (shl v16i8:$vA, v16i8:$vB)),
1028          (v16i8 (VSLB $vA, $vB))>;
1029def : Pat<(v8i16 (shl v8i16:$vA, v8i16:$vB)),
1030          (v8i16 (VSLH $vA, $vB))>;
1031def : Pat<(v4i32 (shl v4i32:$vA, v4i32:$vB)),
1032          (v4i32 (VSLW $vA, $vB))>;
1033def : Pat<(v1i128 (shl v1i128:$vA, v1i128:$vB)),
1034          (v1i128 (VSL (v16i8 (VSLO $vA, $vB)), (v16i8 (VSPLTB 15, $vB))))>;
1035def : Pat<(v16i8 (PPCshl v16i8:$vA, v16i8:$vB)),
1036          (v16i8 (VSLB $vA, $vB))>;
1037def : Pat<(v8i16 (PPCshl v8i16:$vA, v8i16:$vB)),
1038          (v8i16 (VSLH $vA, $vB))>;
1039def : Pat<(v4i32 (PPCshl v4i32:$vA, v4i32:$vB)),
1040          (v4i32 (VSLW $vA, $vB))>;
1041def : Pat<(v1i128 (PPCshl v1i128:$vA, v1i128:$vB)),
1042          (v1i128 (VSL (v16i8 (VSLO $vA, $vB)), (v16i8 (VSPLTB 15, $vB))))>;
1043
1044def : Pat<(v16i8 (srl v16i8:$vA, v16i8:$vB)),
1045          (v16i8 (VSRB $vA, $vB))>;
1046def : Pat<(v8i16 (srl v8i16:$vA, v8i16:$vB)),
1047          (v8i16 (VSRH $vA, $vB))>;
1048def : Pat<(v4i32 (srl v4i32:$vA, v4i32:$vB)),
1049          (v4i32 (VSRW $vA, $vB))>;
1050def : Pat<(v1i128 (srl v1i128:$vA, v1i128:$vB)),
1051          (v1i128 (VSR (v16i8 (VSRO $vA, $vB)), (v16i8 (VSPLTB 15, $vB))))>;
1052def : Pat<(v16i8 (PPCsrl v16i8:$vA, v16i8:$vB)),
1053          (v16i8 (VSRB $vA, $vB))>;
1054def : Pat<(v8i16 (PPCsrl v8i16:$vA, v8i16:$vB)),
1055          (v8i16 (VSRH $vA, $vB))>;
1056def : Pat<(v4i32 (PPCsrl v4i32:$vA, v4i32:$vB)),
1057          (v4i32 (VSRW $vA, $vB))>;
1058def : Pat<(v1i128 (PPCsrl v1i128:$vA, v1i128:$vB)),
1059          (v1i128 (VSR (v16i8 (VSRO $vA, $vB)), (v16i8 (VSPLTB 15, $vB))))>;
1060
1061def : Pat<(v16i8 (sra v16i8:$vA, v16i8:$vB)),
1062          (v16i8 (VSRAB $vA, $vB))>;
1063def : Pat<(v8i16 (sra v8i16:$vA, v8i16:$vB)),
1064          (v8i16 (VSRAH $vA, $vB))>;
1065def : Pat<(v4i32 (sra v4i32:$vA, v4i32:$vB)),
1066          (v4i32 (VSRAW $vA, $vB))>;
1067def : Pat<(v16i8 (PPCsra v16i8:$vA, v16i8:$vB)),
1068          (v16i8 (VSRAB $vA, $vB))>;
1069def : Pat<(v8i16 (PPCsra v8i16:$vA, v8i16:$vB)),
1070          (v8i16 (VSRAH $vA, $vB))>;
1071def : Pat<(v4i32 (PPCsra v4i32:$vA, v4i32:$vB)),
1072          (v4i32 (VSRAW $vA, $vB))>;
1073
1074// Float to integer and integer to float conversions
1075def : Pat<(v4i32 (fp_to_sint v4f32:$vA)),
1076           (VCTSXS_0 $vA)>;
1077def : Pat<(v4i32 (fp_to_uint v4f32:$vA)),
1078           (VCTUXS_0 $vA)>;
1079def : Pat<(v4f32 (sint_to_fp v4i32:$vA)),
1080           (VCFSX_0 $vA)>;
1081def : Pat<(v4f32 (uint_to_fp v4i32:$vA)),
1082           (VCFUX_0 $vA)>;
1083
1084// Floating-point rounding
1085def : Pat<(v4f32 (ffloor v4f32:$vA)),
1086          (VRFIM $vA)>;
1087def : Pat<(v4f32 (fceil v4f32:$vA)),
1088          (VRFIP $vA)>;
1089def : Pat<(v4f32 (ftrunc v4f32:$vA)),
1090          (VRFIZ $vA)>;
1091def : Pat<(v4f32 (fnearbyint v4f32:$vA)),
1092          (VRFIN $vA)>;
1093
1094// Vector selection
1095def : Pat<(v16i8 (vselect v16i8:$vA, v16i8:$vB, v16i8:$vC)),
1096          (VSEL $vC, $vB, $vA)>;
1097def : Pat<(v8i16 (vselect v8i16:$vA, v8i16:$vB, v8i16:$vC)),
1098          (VSEL $vC, $vB, $vA)>;
1099def : Pat<(v4i32 (vselect v4i32:$vA, v4i32:$vB, v4i32:$vC)),
1100          (VSEL $vC, $vB, $vA)>;
1101def : Pat<(v2i64 (vselect v2i64:$vA, v2i64:$vB, v2i64:$vC)),
1102          (VSEL $vC, $vB, $vA)>;
1103def : Pat<(v4f32 (vselect v4i32:$vA, v4f32:$vB, v4f32:$vC)),
1104          (VSEL $vC, $vB, $vA)>;
1105def : Pat<(v2f64 (vselect v2i64:$vA, v2f64:$vB, v2f64:$vC)),
1106          (VSEL $vC, $vB, $vA)>;
1107
1108// Vector Integer Average Instructions
1109def : Pat<(v4i32 (sra (sub v4i32:$vA, (vnot_ppc v4i32:$vB)),
1110          (v4i32 (immEQOneV)))), (v4i32 (VAVGSW $vA, $vB))>;
1111def : Pat<(v8i16 (sra (sub v8i16:$vA, (v8i16 (bitconvert(vnot_ppc v4i32:$vB)))),
1112          (v8i16 (immEQOneV)))), (v8i16 (VAVGSH $vA, $vB))>;
1113def : Pat<(v16i8 (sra (sub v16i8:$vA, (v16i8 (bitconvert(vnot_ppc v4i32:$vB)))),
1114          (v16i8 (immEQOneV)))), (v16i8 (VAVGSB $vA, $vB))>;
1115def : Pat<(v4i32 (srl (sub v4i32:$vA, (vnot_ppc v4i32:$vB)),
1116          (v4i32 (immEQOneV)))), (v4i32 (VAVGUW $vA, $vB))>;
1117def : Pat<(v8i16 (srl (sub v8i16:$vA, (v8i16 (bitconvert(vnot_ppc v4i32:$vB)))),
1118          (v8i16 (immEQOneV)))), (v8i16 (VAVGUH $vA, $vB))>;
1119def : Pat<(v16i8 (srl (sub v16i8:$vA, (v16i8 (bitconvert(vnot_ppc v4i32:$vB)))),
1120          (v16i8 (immEQOneV)))), (v16i8 (VAVGUB $vA, $vB))>;
1121
1122} // end HasAltivec
1123
1124def HasP8Altivec : Predicate<"PPCSubTarget->hasP8Altivec()">;
1125def HasP8Crypto : Predicate<"PPCSubTarget->hasP8Crypto()">;
1126let Predicates = [HasP8Altivec] in {
1127
1128let isCommutable = 1 in {
1129def VMULESW : VX1_Int_Ty2<904, "vmulesw", int_ppc_altivec_vmulesw,
1130                          v2i64, v4i32>;
1131def VMULEUW : VX1_Int_Ty2<648, "vmuleuw", int_ppc_altivec_vmuleuw,
1132                          v2i64, v4i32>;
1133def VMULOSW : VX1_Int_Ty2<392, "vmulosw", int_ppc_altivec_vmulosw,
1134                          v2i64, v4i32>;
1135def VMULOUW : VX1_Int_Ty2<136, "vmulouw", int_ppc_altivec_vmulouw,
1136                          v2i64, v4i32>;
1137def VMULUWM : VXForm_1<137, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1138                       "vmuluwm $vD, $vA, $vB", IIC_VecGeneral,
1139                       [(set v4i32:$vD, (mul v4i32:$vA, v4i32:$vB))]>;
1140def VMAXSD : VX1_Int_Ty<450, "vmaxsd", int_ppc_altivec_vmaxsd, v2i64>;
1141def VMAXUD : VX1_Int_Ty<194, "vmaxud", int_ppc_altivec_vmaxud, v2i64>;
1142def VMINSD : VX1_Int_Ty<962, "vminsd", int_ppc_altivec_vminsd, v2i64>;
1143def VMINUD : VX1_Int_Ty<706, "vminud", int_ppc_altivec_vminud, v2i64>;
1144} // isCommutable
1145
1146// Vector merge
1147def VMRGEW : VXForm_1<1932, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1148                      "vmrgew $vD, $vA, $vB", IIC_VecFP,
1149                      [(set v16i8:$vD,
1150                            (v16i8 (vmrgew_shuffle v16i8:$vA, v16i8:$vB)))]>;
1151def VMRGOW : VXForm_1<1676, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1152                      "vmrgow $vD, $vA, $vB", IIC_VecFP,
1153                      [(set v16i8:$vD,
1154                            (v16i8 (vmrgow_shuffle v16i8:$vA, v16i8:$vB)))]>;
1155
1156// Match vmrgew(x,x) and vmrgow(x,x)
1157def:Pat<(vmrgew_unary_shuffle v16i8:$vA, undef),
1158        (VMRGEW $vA, $vA)>;
1159def:Pat<(vmrgow_unary_shuffle v16i8:$vA, undef),
1160        (VMRGOW $vA, $vA)>;
1161
1162// Match vmrgew(y,x) and vmrgow(y,x), i.e., swapped operands.  These fragments
1163// are matched for little-endian, where the inputs must be swapped for correct
1164// semantics.w
1165def:Pat<(vmrgew_swapped_shuffle v16i8:$vA, v16i8:$vB),
1166        (VMRGEW $vB, $vA)>;
1167def:Pat<(vmrgow_swapped_shuffle v16i8:$vA, v16i8:$vB),
1168        (VMRGOW $vB, $vA)>;
1169
1170// Vector rotates.
1171def VRLD : VX1_Int_Ty<196, "vrld", int_ppc_altivec_vrld, v2i64>;
1172
1173def : Pat<(v2i64 (rotl v2i64:$vA, v2i64:$vB)),
1174          (v2i64 (VRLD v2i64:$vA, v2i64:$vB))>;
1175
1176// Vector shifts
1177def VSLD : VXForm_1<1476, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1178                    "vsld $vD, $vA, $vB", IIC_VecGeneral, []>;
1179def VSRD : VXForm_1<1732, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1180                   "vsrd $vD, $vA, $vB", IIC_VecGeneral, []>;
1181def VSRAD : VXForm_1<964, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1182                    "vsrad $vD, $vA, $vB", IIC_VecGeneral, []>;
1183
1184def : Pat<(v2i64 (shl v2i64:$vA, v2i64:$vB)),
1185          (v2i64 (VSLD $vA, $vB))>;
1186def : Pat<(v2i64 (PPCshl v2i64:$vA, v2i64:$vB)),
1187          (v2i64 (VSLD $vA, $vB))>;
1188def : Pat<(v2i64 (srl v2i64:$vA, v2i64:$vB)),
1189          (v2i64 (VSRD $vA, $vB))>;
1190def : Pat<(v2i64 (PPCsrl v2i64:$vA, v2i64:$vB)),
1191          (v2i64 (VSRD $vA, $vB))>;
1192def : Pat<(v2i64 (sra v2i64:$vA, v2i64:$vB)),
1193          (v2i64 (VSRAD $vA, $vB))>;
1194def : Pat<(v2i64 (PPCsra v2i64:$vA, v2i64:$vB)),
1195          (v2i64 (VSRAD $vA, $vB))>;
1196
1197// Vector Integer Arithmetic Instructions
1198let isCommutable = 1 in {
1199def VADDUDM : VXForm_1<192, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1200                       "vaddudm $vD, $vA, $vB", IIC_VecGeneral,
1201                       [(set v2i64:$vD, (add v2i64:$vA, v2i64:$vB))]>;
1202def VADDUQM : VXForm_1<256, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1203                       "vadduqm $vD, $vA, $vB", IIC_VecGeneral,
1204                       [(set v1i128:$vD, (add v1i128:$vA, v1i128:$vB))]>;
1205} // isCommutable
1206
1207// Vector Quadword Add
1208def VADDEUQM : VA1a_Int_Ty<60, "vaddeuqm", int_ppc_altivec_vaddeuqm, v1i128>;
1209def VADDCUQ  : VX1_Int_Ty<320, "vaddcuq", int_ppc_altivec_vaddcuq, v1i128>;
1210def VADDECUQ : VA1a_Int_Ty<61, "vaddecuq", int_ppc_altivec_vaddecuq, v1i128>;
1211
1212// Vector Doubleword Subtract
1213def VSUBUDM : VXForm_1<1216, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1214                       "vsubudm $vD, $vA, $vB", IIC_VecGeneral,
1215                       [(set v2i64:$vD, (sub v2i64:$vA, v2i64:$vB))]>;
1216
1217// Vector Quadword Subtract
1218def VSUBUQM : VXForm_1<1280, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1219                       "vsubuqm $vD, $vA, $vB", IIC_VecGeneral,
1220                       [(set v1i128:$vD, (sub v1i128:$vA, v1i128:$vB))]>;
1221def VSUBEUQM : VA1a_Int_Ty<62, "vsubeuqm", int_ppc_altivec_vsubeuqm, v1i128>;
1222def VSUBCUQ  : VX1_Int_Ty<1344, "vsubcuq", int_ppc_altivec_vsubcuq, v1i128>;
1223def VSUBECUQ : VA1a_Int_Ty<63, "vsubecuq", int_ppc_altivec_vsubecuq, v1i128>;
1224
1225// Count Leading Zeros
1226def VCLZB : VXForm_2<1794, (outs vrrc:$vD), (ins vrrc:$vB),
1227                     "vclzb $vD, $vB", IIC_VecGeneral,
1228                     [(set v16i8:$vD, (ctlz v16i8:$vB))]>;
1229def VCLZH : VXForm_2<1858, (outs vrrc:$vD), (ins vrrc:$vB),
1230                     "vclzh $vD, $vB", IIC_VecGeneral,
1231                     [(set v8i16:$vD, (ctlz v8i16:$vB))]>;
1232def VCLZW : VXForm_2<1922, (outs vrrc:$vD), (ins vrrc:$vB),
1233                     "vclzw $vD, $vB", IIC_VecGeneral,
1234                     [(set v4i32:$vD, (ctlz v4i32:$vB))]>;
1235def VCLZD : VXForm_2<1986, (outs vrrc:$vD), (ins vrrc:$vB),
1236                     "vclzd $vD, $vB", IIC_VecGeneral,
1237                     [(set v2i64:$vD, (ctlz v2i64:$vB))]>;
1238
1239// Population Count
1240def VPOPCNTB : VXForm_2<1795, (outs vrrc:$vD), (ins vrrc:$vB),
1241                        "vpopcntb $vD, $vB", IIC_VecGeneral,
1242                        [(set v16i8:$vD, (ctpop v16i8:$vB))]>;
1243def VPOPCNTH : VXForm_2<1859, (outs vrrc:$vD), (ins vrrc:$vB),
1244                        "vpopcnth $vD, $vB", IIC_VecGeneral,
1245                        [(set v8i16:$vD, (ctpop v8i16:$vB))]>;
1246def VPOPCNTW : VXForm_2<1923, (outs vrrc:$vD), (ins vrrc:$vB),
1247                        "vpopcntw $vD, $vB", IIC_VecGeneral,
1248                        [(set v4i32:$vD, (ctpop v4i32:$vB))]>;
1249def VPOPCNTD : VXForm_2<1987, (outs vrrc:$vD), (ins vrrc:$vB),
1250                        "vpopcntd $vD, $vB", IIC_VecGeneral,
1251                        [(set v2i64:$vD, (ctpop v2i64:$vB))]>;
1252
1253let isCommutable = 1 in {
1254// FIXME: Use AddedComplexity > 400 to ensure these patterns match before the
1255//        VSX equivalents. We need to fix this up at some point. Two possible
1256//        solutions for this problem:
1257//        1. Disable Altivec patterns that compete with VSX patterns using the
1258//           !HasVSX predicate. This essentially favours VSX over Altivec, in
1259//           hopes of reducing register pressure (larger register set using VSX
1260//           instructions than VMX instructions)
1261//        2. Employ a more disciplined use of AddedComplexity, which would provide
1262//           more fine-grained control than option 1. This would be beneficial
1263//           if we find situations where Altivec is really preferred over VSX.
1264def VEQV  : VXForm_1<1668, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1265                     "veqv $vD, $vA, $vB", IIC_VecGeneral,
1266                     [(set v4i32:$vD, (vnot_ppc (xor v4i32:$vA, v4i32:$vB)))]>;
1267def VNAND : VXForm_1<1412, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1268                     "vnand $vD, $vA, $vB", IIC_VecGeneral,
1269                     [(set v4i32:$vD, (vnot_ppc (and v4i32:$vA, v4i32:$vB)))]>;
1270} // isCommutable
1271
1272def VORC : VXForm_1<1348, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1273                      "vorc $vD, $vA, $vB", IIC_VecGeneral,
1274                      [(set v4i32:$vD, (or v4i32:$vA,
1275                                           (vnot_ppc v4i32:$vB)))]>;
1276
1277// i64 element comparisons.
1278def VCMPEQUD  : VCMP <199, "vcmpequd $vD, $vA, $vB" , v2i64>;
1279def VCMPEQUD_rec : VCMPo<199, "vcmpequd. $vD, $vA, $vB", v2i64>;
1280def VCMPGTSD  : VCMP <967, "vcmpgtsd $vD, $vA, $vB" , v2i64>;
1281def VCMPGTSD_rec : VCMPo<967, "vcmpgtsd. $vD, $vA, $vB", v2i64>;
1282def VCMPGTUD  : VCMP <711, "vcmpgtud $vD, $vA, $vB" , v2i64>;
1283def VCMPGTUD_rec : VCMPo<711, "vcmpgtud. $vD, $vA, $vB", v2i64>;
1284
1285// The cryptography instructions that do not require Category:Vector.Crypto
1286def VPMSUMB : VX1_Int_Ty<1032, "vpmsumb",
1287                         int_ppc_altivec_crypto_vpmsumb, v16i8>;
1288def VPMSUMH : VX1_Int_Ty<1096, "vpmsumh",
1289                         int_ppc_altivec_crypto_vpmsumh, v8i16>;
1290def VPMSUMW : VX1_Int_Ty<1160, "vpmsumw",
1291                         int_ppc_altivec_crypto_vpmsumw, v4i32>;
1292def VPMSUMD : VX1_Int_Ty<1224, "vpmsumd",
1293                         int_ppc_altivec_crypto_vpmsumd, v2i64>;
1294def VPERMXOR : VA1a_Int_Ty<45, "vpermxor",
1295                         int_ppc_altivec_crypto_vpermxor, v16i8>;
1296
1297// Vector doubleword integer pack and unpack.
1298def VPKSDSS : VX1_Int_Ty2<1486, "vpksdss", int_ppc_altivec_vpksdss,
1299                          v4i32, v2i64>;
1300def VPKSDUS : VX1_Int_Ty2<1358, "vpksdus", int_ppc_altivec_vpksdus,
1301                          v4i32, v2i64>;
1302def VPKUDUM : VXForm_1<1102, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1303                       "vpkudum $vD, $vA, $vB", IIC_VecFP,
1304                       [(set v16i8:$vD,
1305                         (vpkudum_shuffle v16i8:$vA, v16i8:$vB))]>;
1306def VPKUDUS : VX1_Int_Ty2<1230, "vpkudus", int_ppc_altivec_vpkudus,
1307                          v4i32, v2i64>;
1308def VUPKHSW : VX2_Int_Ty2<1614, "vupkhsw", int_ppc_altivec_vupkhsw,
1309                          v2i64, v4i32>;
1310def VUPKLSW : VX2_Int_Ty2<1742, "vupklsw", int_ppc_altivec_vupklsw,
1311                          v2i64, v4i32>;
1312
1313// Shuffle patterns for unary and swapped (LE) vector pack modulo.
1314def:Pat<(vpkudum_unary_shuffle v16i8:$vA, undef),
1315        (VPKUDUM $vA, $vA)>;
1316def:Pat<(vpkudum_swapped_shuffle v16i8:$vA, v16i8:$vB),
1317        (VPKUDUM $vB, $vA)>;
1318
1319def VGBBD : VX2_Int_Ty2<1292, "vgbbd", int_ppc_altivec_vgbbd, v16i8, v16i8>;
1320def VBPERMQ : VX1_Int_Ty2<1356, "vbpermq", int_ppc_altivec_vbpermq,
1321                          v2i64, v16i8>;
1322} // end HasP8Altivec
1323
1324// Crypto instructions (from builtins)
1325let Predicates = [HasP8Crypto] in {
1326def VSHASIGMAW : VXCR_Int_Ty<1666, "vshasigmaw",
1327                              int_ppc_altivec_crypto_vshasigmaw, v4i32>;
1328def VSHASIGMAD : VXCR_Int_Ty<1730, "vshasigmad",
1329                              int_ppc_altivec_crypto_vshasigmad, v2i64>;
1330def VCIPHER : VX1_Int_Ty<1288, "vcipher", int_ppc_altivec_crypto_vcipher,
1331                         v2i64>;
1332def VCIPHERLAST : VX1_Int_Ty<1289, "vcipherlast",
1333                              int_ppc_altivec_crypto_vcipherlast, v2i64>;
1334def VNCIPHER : VX1_Int_Ty<1352, "vncipher",
1335                          int_ppc_altivec_crypto_vncipher, v2i64>;
1336def VNCIPHERLAST : VX1_Int_Ty<1353, "vncipherlast",
1337                              int_ppc_altivec_crypto_vncipherlast, v2i64>;
1338def VSBOX : VXBX_Int_Ty<1480, "vsbox", int_ppc_altivec_crypto_vsbox, v2i64>;
1339} // HasP8Crypto
1340
1341// The following altivec instructions were introduced in Power ISA 3.0
1342def HasP9Altivec : Predicate<"PPCSubTarget->hasP9Altivec()">;
1343let Predicates = [HasP9Altivec] in {
1344
1345// Vector Multiply-Sum
1346def VMSUMUDM : VA1a_Int_Ty3<35, "vmsumudm", int_ppc_altivec_vmsumudm,
1347                            v1i128, v2i64, v1i128>;
1348
1349// i8 element comparisons.
1350def VCMPNEB   : VCMP   <  7, "vcmpneb $vD, $vA, $vB"  , v16i8>;
1351def VCMPNEB_rec  : VCMPo  <  7, "vcmpneb. $vD, $vA, $vB" , v16i8>;
1352def VCMPNEZB  : VCMP <263, "vcmpnezb $vD, $vA, $vB" , v16i8>;
1353def VCMPNEZB_rec : VCMPo<263, "vcmpnezb. $vD, $vA, $vB", v16i8>;
1354
1355// i16 element comparisons.
1356def VCMPNEH   : VCMP < 71, "vcmpneh $vD, $vA, $vB"  , v8i16>;
1357def VCMPNEH_rec  : VCMPo< 71, "vcmpneh. $vD, $vA, $vB" , v8i16>;
1358def VCMPNEZH  : VCMP <327, "vcmpnezh $vD, $vA, $vB" , v8i16>;
1359def VCMPNEZH_rec : VCMPo<327, "vcmpnezh. $vD, $vA, $vB", v8i16>;
1360
1361// i32 element comparisons.
1362def VCMPNEW   : VCMP <135, "vcmpnew $vD, $vA, $vB"  , v4i32>;
1363def VCMPNEW_rec  : VCMPo<135, "vcmpnew. $vD, $vA, $vB" , v4i32>;
1364def VCMPNEZW  : VCMP <391, "vcmpnezw $vD, $vA, $vB" , v4i32>;
1365def VCMPNEZW_rec : VCMPo<391, "vcmpnezw. $vD, $vA, $vB", v4i32>;
1366
1367// VX-Form: [PO VRT / UIM VRB XO].
1368// We use VXForm_1 to implement it, that is, we use "VRA" (5 bit) to represent
1369// "/ UIM" (1 + 4 bit)
1370class VX1_VT5_UIM5_VB5<bits<11> xo, string opc, list<dag> pattern>
1371  : VXForm_1<xo, (outs vrrc:$vD), (ins u4imm:$UIMM, vrrc:$vB),
1372             !strconcat(opc, " $vD, $vB, $UIMM"), IIC_VecGeneral, pattern>;
1373
1374class VX1_RT5_RA5_VB5<bits<11> xo, string opc, list<dag> pattern>
1375  : VXForm_1<xo, (outs g8rc:$rD), (ins g8rc:$rA, vrrc:$vB),
1376             !strconcat(opc, " $rD, $rA, $vB"), IIC_VecGeneral, pattern>;
1377
1378// Vector Extract Unsigned
1379def VEXTRACTUB : VX1_VT5_UIM5_VB5<525, "vextractub", []>;
1380def VEXTRACTUH : VX1_VT5_UIM5_VB5<589, "vextractuh", []>;
1381def VEXTRACTUW : VX1_VT5_UIM5_VB5<653, "vextractuw", []>;
1382def VEXTRACTD  : VX1_VT5_UIM5_VB5<717, "vextractd" , []>;
1383
1384// Vector Extract Unsigned Byte/Halfword/Word Left/Right-Indexed
1385let hasSideEffects = 0 in {
1386def VEXTUBLX : VX1_RT5_RA5_VB5<1549, "vextublx", []>;
1387def VEXTUBRX : VX1_RT5_RA5_VB5<1805, "vextubrx", []>;
1388def VEXTUHLX : VX1_RT5_RA5_VB5<1613, "vextuhlx", []>;
1389def VEXTUHRX : VX1_RT5_RA5_VB5<1869, "vextuhrx", []>;
1390def VEXTUWLX : VX1_RT5_RA5_VB5<1677, "vextuwlx", []>;
1391def VEXTUWRX : VX1_RT5_RA5_VB5<1933, "vextuwrx", []>;
1392}
1393
1394// Vector Insert Element Instructions
1395def VINSERTB : VXForm_1<781, (outs vrrc:$vD),
1396                        (ins vrrc:$vDi, u4imm:$UIM, vrrc:$vB),
1397                        "vinsertb $vD, $vB, $UIM", IIC_VecGeneral,
1398                        [(set v16i8:$vD, (PPCvecinsert v16i8:$vDi, v16i8:$vB,
1399                                                      imm32SExt16:$UIM))]>,
1400                        RegConstraint<"$vDi = $vD">, NoEncode<"$vDi">;
1401def VINSERTH : VXForm_1<845, (outs vrrc:$vD),
1402                        (ins vrrc:$vDi, u4imm:$UIM, vrrc:$vB),
1403                        "vinserth $vD, $vB, $UIM", IIC_VecGeneral,
1404                        [(set v8i16:$vD, (PPCvecinsert v8i16:$vDi, v8i16:$vB,
1405                                                      imm32SExt16:$UIM))]>,
1406                        RegConstraint<"$vDi = $vD">, NoEncode<"$vDi">;
1407def VINSERTW : VX1_VT5_UIM5_VB5<909, "vinsertw", []>;
1408def VINSERTD : VX1_VT5_UIM5_VB5<973, "vinsertd", []>;
1409
1410class VX_VT5_EO5_VB5<bits<11> xo, bits<5> eo, string opc, list<dag> pattern>
1411  : VXForm_RD5_XO5_RS5<xo, eo, (outs vrrc:$vD), (ins vrrc:$vB),
1412                       !strconcat(opc, " $vD, $vB"), IIC_VecGeneral, pattern>;
1413class VX_VT5_EO5_VB5s<bits<11> xo, bits<5> eo, string opc, list<dag> pattern>
1414  : VXForm_RD5_XO5_RS5<xo, eo, (outs vfrc:$vD), (ins vfrc:$vB),
1415                       !strconcat(opc, " $vD, $vB"), IIC_VecGeneral, pattern>;
1416
1417// Vector Count Leading/Trailing Zero LSB. Result is placed into GPR[rD]
1418def VCLZLSBB : VXForm_RD5_XO5_RS5<1538, 0, (outs gprc:$rD), (ins vrrc:$vB),
1419                                  "vclzlsbb $rD, $vB", IIC_VecGeneral,
1420                                  [(set i32:$rD, (int_ppc_altivec_vclzlsbb
1421                                     v16i8:$vB))]>;
1422def VCTZLSBB : VXForm_RD5_XO5_RS5<1538, 1, (outs gprc:$rD), (ins vrrc:$vB),
1423                                  "vctzlsbb $rD, $vB", IIC_VecGeneral,
1424                                  [(set i32:$rD, (int_ppc_altivec_vctzlsbb
1425                                     v16i8:$vB))]>;
1426// Vector Count Trailing Zeros
1427def VCTZB : VX_VT5_EO5_VB5<1538, 28, "vctzb",
1428                           [(set v16i8:$vD, (cttz v16i8:$vB))]>;
1429def VCTZH : VX_VT5_EO5_VB5<1538, 29, "vctzh",
1430                           [(set v8i16:$vD, (cttz v8i16:$vB))]>;
1431def VCTZW : VX_VT5_EO5_VB5<1538, 30, "vctzw",
1432                           [(set v4i32:$vD, (cttz v4i32:$vB))]>;
1433def VCTZD : VX_VT5_EO5_VB5<1538, 31, "vctzd",
1434                           [(set v2i64:$vD, (cttz v2i64:$vB))]>;
1435
1436// Vector Extend Sign
1437def VEXTSB2W : VX_VT5_EO5_VB5<1538, 16, "vextsb2w", []>;
1438def VEXTSH2W : VX_VT5_EO5_VB5<1538, 17, "vextsh2w", []>;
1439def VEXTSB2D : VX_VT5_EO5_VB5<1538, 24, "vextsb2d", []>;
1440def VEXTSH2D : VX_VT5_EO5_VB5<1538, 25, "vextsh2d", []>;
1441def VEXTSW2D : VX_VT5_EO5_VB5<1538, 26, "vextsw2d", []>;
1442let isCodeGenOnly = 1 in {
1443  def VEXTSB2Ws : VX_VT5_EO5_VB5s<1538, 16, "vextsb2w", []>;
1444  def VEXTSH2Ws : VX_VT5_EO5_VB5s<1538, 17, "vextsh2w", []>;
1445  def VEXTSB2Ds : VX_VT5_EO5_VB5s<1538, 24, "vextsb2d", []>;
1446  def VEXTSH2Ds : VX_VT5_EO5_VB5s<1538, 25, "vextsh2d", []>;
1447  def VEXTSW2Ds : VX_VT5_EO5_VB5s<1538, 26, "vextsw2d", []>;
1448}
1449
1450def : Pat<(v4i32 (sext_inreg v4i32:$VRB, v4i8)), (v4i32 (VEXTSB2W $VRB))>;
1451def : Pat<(v4i32 (sext_inreg v4i32:$VRB, v4i16)), (v4i32 (VEXTSH2W $VRB))>;
1452def : Pat<(v2i64 (sext_inreg v2i64:$VRB, v2i8)), (v2i64 (VEXTSB2D $VRB))>;
1453def : Pat<(v2i64 (sext_inreg v2i64:$VRB, v2i16)), (v2i64 (VEXTSH2D $VRB))>;
1454def : Pat<(v2i64 (sext_inreg v2i64:$VRB, v2i32)), (v2i64 (VEXTSW2D $VRB))>;
1455
1456// Vector Integer Negate
1457def VNEGW : VX_VT5_EO5_VB5<1538, 6, "vnegw",
1458                           [(set v4i32:$vD,
1459                            (sub (v4i32 immAllZerosV), v4i32:$vB))]>;
1460
1461def VNEGD : VX_VT5_EO5_VB5<1538, 7, "vnegd",
1462                           [(set v2i64:$vD,
1463                            (sub (v2i64 (bitconvert (v4i32 immAllZerosV))),
1464                                  v2i64:$vB))]>;
1465
1466// Vector Parity Byte
1467def VPRTYBW : VX_VT5_EO5_VB5<1538, 8, "vprtybw", [(set v4i32:$vD,
1468                            (int_ppc_altivec_vprtybw v4i32:$vB))]>;
1469def VPRTYBD : VX_VT5_EO5_VB5<1538,  9, "vprtybd", [(set v2i64:$vD,
1470                            (int_ppc_altivec_vprtybd v2i64:$vB))]>;
1471def VPRTYBQ : VX_VT5_EO5_VB5<1538, 10, "vprtybq", [(set v1i128:$vD,
1472                            (int_ppc_altivec_vprtybq v1i128:$vB))]>;
1473
1474// Vector (Bit) Permute (Right-indexed)
1475def VBPERMD : VXForm_1<1484, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1476                       "vbpermd $vD, $vA, $vB", IIC_VecFP, []>;
1477def VPERMR : VAForm_1a<59, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vC),
1478                       "vpermr $vD, $vA, $vB, $vC", IIC_VecFP, []>;
1479
1480class VX1_VT5_VA5_VB5<bits<11> xo, string opc, list<dag> pattern>
1481  : VXForm_1<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1482             !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP, pattern>;
1483
1484// Vector Rotate Left Mask/Mask-Insert
1485def VRLWNM : VX1_VT5_VA5_VB5<389, "vrlwnm",
1486                             [(set v4i32:$vD,
1487                                 (int_ppc_altivec_vrlwnm v4i32:$vA,
1488                                                         v4i32:$vB))]>;
1489def VRLWMI : VXForm_1<133, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vDi),
1490                      "vrlwmi $vD, $vA, $vB", IIC_VecFP,
1491                      [(set v4i32:$vD,
1492                         (int_ppc_altivec_vrlwmi v4i32:$vA, v4i32:$vB,
1493                                                 v4i32:$vDi))]>,
1494                      RegConstraint<"$vDi = $vD">, NoEncode<"$vDi">;
1495def VRLDNM : VX1_VT5_VA5_VB5<453, "vrldnm",
1496                             [(set v2i64:$vD,
1497                                 (int_ppc_altivec_vrldnm v2i64:$vA,
1498                                                         v2i64:$vB))]>;
1499def VRLDMI : VXForm_1<197, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, vrrc:$vDi),
1500                      "vrldmi $vD, $vA, $vB", IIC_VecFP,
1501                      [(set v2i64:$vD,
1502                         (int_ppc_altivec_vrldmi v2i64:$vA, v2i64:$vB,
1503                                                 v2i64:$vDi))]>,
1504                      RegConstraint<"$vDi = $vD">, NoEncode<"$vDi">;
1505
1506// Vector Shift Left/Right
1507def VSLV : VX1_VT5_VA5_VB5<1860, "vslv",
1508                           [(set v16i8 : $vD, (int_ppc_altivec_vslv v16i8 : $vA, v16i8 : $vB))]>;
1509def VSRV : VX1_VT5_VA5_VB5<1796, "vsrv",
1510                           [(set v16i8 : $vD, (int_ppc_altivec_vsrv v16i8 : $vA, v16i8 : $vB))]>;
1511
1512// Vector Multiply-by-10 (& Write Carry) Unsigned Quadword
1513def VMUL10UQ   : VXForm_BX<513, (outs vrrc:$vD), (ins vrrc:$vA),
1514                           "vmul10uq $vD, $vA", IIC_VecFP, []>;
1515def VMUL10CUQ  : VXForm_BX<  1, (outs vrrc:$vD), (ins vrrc:$vA),
1516                           "vmul10cuq $vD, $vA", IIC_VecFP, []>;
1517
1518// Vector Multiply-by-10 Extended (& Write Carry) Unsigned Quadword
1519def VMUL10EUQ  : VX1_VT5_VA5_VB5<577, "vmul10euq" , []>;
1520def VMUL10ECUQ : VX1_VT5_VA5_VB5< 65, "vmul10ecuq", []>;
1521
1522// Decimal Integer Format Conversion Instructions
1523
1524// [PO VRT EO VRB 1 PS XO], "_o" means CR6 is set.
1525class VX_VT5_EO5_VB5_PS1_XO9_o<bits<5> eo, bits<9> xo, string opc,
1526                               list<dag> pattern>
1527  : VX_RD5_EO5_RS5_PS1_XO9<eo, xo, (outs vrrc:$vD), (ins vrrc:$vB, u1imm:$PS),
1528                        !strconcat(opc, " $vD, $vB, $PS"), IIC_VecFP, pattern> {
1529  let Defs = [CR6];
1530}
1531
1532// [PO VRT EO VRB 1 / XO]
1533class VX_VT5_EO5_VB5_XO9_o<bits<5> eo, bits<9> xo, string opc,
1534                           list<dag> pattern>
1535  : VX_RD5_EO5_RS5_PS1_XO9<eo, xo, (outs vrrc:$vD), (ins vrrc:$vB),
1536                           !strconcat(opc, " $vD, $vB"), IIC_VecFP, pattern> {
1537  let Defs = [CR6];
1538  let PS = 0;
1539}
1540
1541// Decimal Convert From/to National/Zoned/Signed-QWord
1542def BCDCFN_rec  : VX_VT5_EO5_VB5_PS1_XO9_o<7, 385, "bcdcfn." , []>;
1543def BCDCFZ_rec  : VX_VT5_EO5_VB5_PS1_XO9_o<6, 385, "bcdcfz." , []>;
1544def BCDCTN_rec  : VX_VT5_EO5_VB5_XO9_o    <5, 385, "bcdctn." , []>;
1545def BCDCTZ_rec  : VX_VT5_EO5_VB5_PS1_XO9_o<4, 385, "bcdctz." , []>;
1546def BCDCFSQ_rec : VX_VT5_EO5_VB5_PS1_XO9_o<2, 385, "bcdcfsq.", []>;
1547def BCDCTSQ_rec : VX_VT5_EO5_VB5_XO9_o    <0, 385, "bcdctsq.", []>;
1548
1549// Decimal Copy-Sign/Set-Sign
1550let Defs = [CR6] in
1551def BCDCPSGN_rec : VX1_VT5_VA5_VB5<833, "bcdcpsgn.", []>;
1552
1553def BCDSETSGN_rec : VX_VT5_EO5_VB5_PS1_XO9_o<31, 385, "bcdsetsgn.", []>;
1554
1555// [PO VRT VRA VRB 1 PS XO], "_o" means CR6 is set.
1556class VX_VT5_VA5_VB5_PS1_XO9_o<bits<9> xo, string opc, list<dag> pattern>
1557  : VX_RD5_RSp5_PS1_XO9<xo,
1558                   (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB, u1imm:$PS),
1559                   !strconcat(opc, " $vD, $vA, $vB, $PS"), IIC_VecFP, pattern> {
1560  let Defs = [CR6];
1561}
1562
1563// [PO VRT VRA VRB 1 / XO]
1564class VX_VT5_VA5_VB5_XO9_o<bits<9> xo, string opc, list<dag> pattern>
1565  : VX_RD5_RSp5_PS1_XO9<xo, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1566                        !strconcat(opc, " $vD, $vA, $vB"), IIC_VecFP, pattern> {
1567  let Defs = [CR6];
1568  let PS = 0;
1569}
1570
1571// Decimal Shift/Unsigned-Shift/Shift-and-Round
1572def BCDS_rec :  VX_VT5_VA5_VB5_PS1_XO9_o<193, "bcds." , []>;
1573def BCDUS_rec : VX_VT5_VA5_VB5_XO9_o    <129, "bcdus.", []>;
1574def BCDSR_rec : VX_VT5_VA5_VB5_PS1_XO9_o<449, "bcdsr.", []>;
1575
1576// Decimal (Unsigned) Truncate
1577def BCDTRUNC_rec :  VX_VT5_VA5_VB5_PS1_XO9_o<257, "bcdtrunc." , []>;
1578def BCDUTRUNC_rec : VX_VT5_VA5_VB5_XO9_o    <321, "bcdutrunc.", []>;
1579
1580// Absolute Difference
1581def VABSDUB : VXForm_1<1027, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1582                       "vabsdub $vD, $vA, $vB", IIC_VecGeneral,
1583                       [(set v16i8:$vD, (int_ppc_altivec_vabsdub v16i8:$vA, v16i8:$vB))]>;
1584def VABSDUH : VXForm_1<1091, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1585                       "vabsduh $vD, $vA, $vB", IIC_VecGeneral,
1586                       [(set v8i16:$vD, (int_ppc_altivec_vabsduh v8i16:$vA, v8i16:$vB))]>;
1587def VABSDUW : VXForm_1<1155, (outs vrrc:$vD), (ins vrrc:$vA, vrrc:$vB),
1588                       "vabsduw $vD, $vA, $vB", IIC_VecGeneral,
1589                       [(set v4i32:$vD, (int_ppc_altivec_vabsduw v4i32:$vA, v4i32:$vB))]>;
1590
1591} // end HasP9Altivec
1592