1 //===- PPCInstrAltivec.td - The PowerPC Altivec Extension --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by Chris Lattner and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the Altivec extension to the PowerPC instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // Altivec transformation functions and pattern fragments.
18 // VSPLT*_get_imm xform function: convert vector_shuffle mask to VSPLT* imm.
19 def VSPLTB_get_imm : SDNodeXForm<build_vector, [{
20 return getI32Imm(PPC::getVSPLTImmediate(N, 1));
22 def VSPLTB_shuffle_mask : PatLeaf<(build_vector), [{
23 return PPC::isSplatShuffleMask(N, 1);
25 def VSPLTH_get_imm : SDNodeXForm<build_vector, [{
26 return getI32Imm(PPC::getVSPLTImmediate(N, 2));
28 def VSPLTH_shuffle_mask : PatLeaf<(build_vector), [{
29 return PPC::isSplatShuffleMask(N, 2);
31 def VSPLTW_get_imm : SDNodeXForm<build_vector, [{
32 return getI32Imm(PPC::getVSPLTImmediate(N, 4));
34 def VSPLTW_shuffle_mask : PatLeaf<(build_vector), [{
35 return PPC::isSplatShuffleMask(N, 4);
39 // VSPLTISB_get_imm xform function: convert build_vector to VSPLTISB imm.
40 def VSPLTISB_get_imm : SDNodeXForm<build_vector, [{
42 PPC::isVecSplatImm(N, 1, &Val);
43 return getI32Imm(Val);
45 def vecspltisb : PatLeaf<(build_vector), [{
46 return PPC::isVecSplatImm(N, 1);
47 }], VSPLTISB_get_imm>;
49 // VSPLTISH_get_imm xform function: convert build_vector to VSPLTISH imm.
50 def VSPLTISH_get_imm : SDNodeXForm<build_vector, [{
52 PPC::isVecSplatImm(N, 2, &Val);
53 return getI32Imm(Val);
55 def vecspltish : PatLeaf<(build_vector), [{
56 return PPC::isVecSplatImm(N, 2);
57 }], VSPLTISH_get_imm>;
59 // VSPLTISW_get_imm xform function: convert build_vector to VSPLTISW imm.
60 def VSPLTISW_get_imm : SDNodeXForm<build_vector, [{
62 PPC::isVecSplatImm(N, 4, &Val);
63 return getI32Imm(Val);
65 def vecspltisw : PatLeaf<(build_vector), [{
66 return PPC::isVecSplatImm(N, 4);
67 }], VSPLTISW_get_imm>;
69 //===----------------------------------------------------------------------===//
70 // Helpers for defining instructions that directly correspond to intrinsics.
72 // VA1a_Int - A VAForm_1a intrinsic definition.
73 class VA1a_Int<bits<6> xo, string opc, Intrinsic IntID>
74 : VAForm_1a<xo, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB, VRRC:$vC),
75 !strconcat(opc, " $vD, $vA, $vB, $vC"), VecFP,
76 [(set VRRC:$vD, (IntID VRRC:$vA, VRRC:$vB, VRRC:$vC))]>;
78 // VX1_Int - A VXForm_1 intrinsic definition.
79 class VX1_Int<bits<11> xo, string opc, Intrinsic IntID>
80 : VXForm_1<xo, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
81 !strconcat(opc, " $vD, $vA, $vB"), VecFP,
82 [(set VRRC:$vD, (IntID VRRC:$vA, VRRC:$vB))]>;
84 // VX2_Int - A VXForm_2 intrinsic definition.
85 class VX2_Int<bits<11> xo, string opc, Intrinsic IntID>
86 : VXForm_2<xo, (ops VRRC:$vD, VRRC:$vB),
87 !strconcat(opc, " $vD, $vB"), VecFP,
88 [(set VRRC:$vD, (IntID VRRC:$vB))]>;
90 //===----------------------------------------------------------------------===//
91 // Instruction Definitions.
93 def IMPLICIT_DEF_VRRC : Pseudo<(ops VRRC:$rD), "; $rD = IMPLICIT_DEF_VRRC",
94 [(set VRRC:$rD, (v4f32 (undef)))]>;
96 let noResults = 1 in {
97 def DSS : DSS_Form<822, (ops u5imm:$A, u5imm:$STRM,u5imm:$ZERO1,u5imm:$ZERO2),
98 "dss $STRM, $A", LdStGeneral /*FIXME*/, []>;
99 def DST : DSS_Form<342, (ops u5imm:$T, u5imm:$STRM, GPRC:$rA, GPRC:$rB),
100 "dst $rA, $rB, $STRM, $T", LdStGeneral /*FIXME*/, []>;
101 def DSTST : DSS_Form<374, (ops u5imm:$T, u5imm:$STRM, GPRC:$rA, GPRC:$rB),
102 "dstst $rA, $rB, $STRM, $T", LdStGeneral /*FIXME*/, []>;
105 def MFVSCR : VXForm_4<1540, (ops VRRC:$vD),
106 "mfvcr $vD", LdStGeneral,
107 [(set VRRC:$vD, (int_ppc_altivec_mfvscr))]>;
108 def MTVSCR : VXForm_5<1604, (ops VRRC:$vB),
109 "mtvcr $vB", LdStGeneral,
110 [(int_ppc_altivec_mtvscr VRRC:$vB)]>;
112 let isLoad = 1, PPC970_Unit = 2 in { // Loads.
113 def LVEBX: XForm_1<31, 7, (ops VRRC:$vD, memrr:$src),
114 "lvebx $vD, $src", LdStGeneral,
115 [(set VRRC:$vD, (int_ppc_altivec_lvebx xoaddr:$src))]>;
116 def LVEHX: XForm_1<31, 39, (ops VRRC:$vD, memrr:$src),
117 "lvehx $vD, $src", LdStGeneral,
118 [(set VRRC:$vD, (int_ppc_altivec_lvehx xoaddr:$src))]>;
119 def LVEWX: XForm_1<31, 71, (ops VRRC:$vD, memrr:$src),
120 "lvewx $vD, $src", LdStGeneral,
121 [(set VRRC:$vD, (int_ppc_altivec_lvewx xoaddr:$src))]>;
122 def LVX : XForm_1<31, 103, (ops VRRC:$vD, memrr:$src),
123 "lvx $vD, $src", LdStGeneral,
124 [(set VRRC:$vD, (int_ppc_altivec_lvx xoaddr:$src))]>;
125 def LVXL : XForm_1<31, 359, (ops VRRC:$vD, memrr:$src),
126 "lvxl $vD, $src", LdStGeneral,
127 [(set VRRC:$vD, (int_ppc_altivec_lvxl xoaddr:$src))]>;
130 def LVSL : XForm_1<31, 6, (ops VRRC:$vD, memrr:$src),
131 "lvsl $vD, $src", LdStGeneral,
132 [(set VRRC:$vD, (int_ppc_altivec_lvsl xoaddr:$src))]>,
134 def LVSR : XForm_1<31, 38, (ops VRRC:$vD, memrr:$src),
135 "lvsr $vD, $src", LdStGeneral,
136 [(set VRRC:$vD, (int_ppc_altivec_lvsr xoaddr:$src))]>,
139 let isStore = 1, noResults = 1, PPC970_Unit = 2 in { // Stores.
140 def STVEBX: XForm_8<31, 135, (ops VRRC:$rS, memrr:$dst),
141 "stvebx $rS, $dst", LdStGeneral,
142 [(int_ppc_altivec_stvebx VRRC:$rS, xoaddr:$dst)]>;
143 def STVEHX: XForm_8<31, 167, (ops VRRC:$rS, memrr:$dst),
144 "stvehx $rS, $dst", LdStGeneral,
145 [(int_ppc_altivec_stvehx VRRC:$rS, xoaddr:$dst)]>;
146 def STVEWX: XForm_8<31, 199, (ops VRRC:$rS, memrr:$dst),
147 "stvewx $rS, $dst", LdStGeneral,
148 [(int_ppc_altivec_stvewx VRRC:$rS, xoaddr:$dst)]>;
149 def STVX : XForm_8<31, 231, (ops VRRC:$rS, memrr:$dst),
150 "stvx $rS, $dst", LdStGeneral,
151 [(int_ppc_altivec_stvx VRRC:$rS, xoaddr:$dst)]>;
152 def STVXL : XForm_8<31, 487, (ops VRRC:$rS, memrr:$dst),
153 "stvxl $rS, $dst", LdStGeneral,
154 [(int_ppc_altivec_stvxl VRRC:$rS, xoaddr:$dst)]>;
157 let PPC970_Unit = 5 in { // VALU Operations.
158 // VA-Form instructions. 3-input AltiVec ops.
159 def VMADDFP : VAForm_1<46, (ops VRRC:$vD, VRRC:$vA, VRRC:$vC, VRRC:$vB),
160 "vmaddfp $vD, $vA, $vC, $vB", VecFP,
161 [(set VRRC:$vD, (fadd (fmul VRRC:$vA, VRRC:$vC),
163 Requires<[FPContractions]>;
164 def VNMSUBFP: VAForm_1<47, (ops VRRC:$vD, VRRC:$vA, VRRC:$vC, VRRC:$vB),
165 "vnmsubfp $vD, $vA, $vC, $vB", VecFP,
166 [(set VRRC:$vD, (fneg (fsub (fmul VRRC:$vA, VRRC:$vC),
168 Requires<[FPContractions]>;
170 def VMHADDSHS : VA1a_Int<32, "vmhaddshs", int_ppc_altivec_vmhaddshs>;
171 def VMHRADDSHS : VA1a_Int<33, "vmhraddshs", int_ppc_altivec_vmhraddshs>;
172 def VMLADDUHM : VA1a_Int<34, "vmladduhm", int_ppc_altivec_vmladduhm>;
173 def VPERM : VA1a_Int<43, "vperm", int_ppc_altivec_vperm>;
174 def VSEL : VA1a_Int<42, "vsel", int_ppc_altivec_vsel>;
176 def VSLDOI : VAForm_2<44, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB, u5imm:$SH),
177 "vsldoi $vD, $vA, $vB, $SH", VecFP,
179 (int_ppc_altivec_vsldoi VRRC:$vA, VRRC:$vB,
182 // VX-Form instructions. AltiVec arithmetic ops.
183 def VADDFP : VXForm_1<10, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
184 "vaddfp $vD, $vA, $vB", VecFP,
185 [(set VRRC:$vD, (fadd VRRC:$vA, VRRC:$vB))]>;
187 def VADDUBM : VXForm_1<0, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
188 "vaddubm $vD, $vA, $vB", VecGeneral,
189 [(set VRRC:$vD, (add (v16i8 VRRC:$vA), VRRC:$vB))]>;
190 def VADDUHM : VXForm_1<64, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
191 "vadduhm $vD, $vA, $vB", VecGeneral,
192 [(set VRRC:$vD, (add (v8i16 VRRC:$vA), VRRC:$vB))]>;
193 def VADDUWM : VXForm_1<128, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
194 "vadduwm $vD, $vA, $vB", VecGeneral,
195 [(set VRRC:$vD, (add (v4i32 VRRC:$vA), VRRC:$vB))]>;
197 def VADDCUW : VX1_Int<384, "vaddcuw", int_ppc_altivec_vaddcuw>;
198 def VADDSBS : VX1_Int<768, "vaddsbs", int_ppc_altivec_vaddsbs>;
199 def VADDSHS : VX1_Int<832, "vaddshs", int_ppc_altivec_vaddshs>;
200 def VADDSWS : VX1_Int<896, "vaddsws", int_ppc_altivec_vaddsws>;
201 def VADDUBS : VX1_Int<512, "vaddubs", int_ppc_altivec_vaddubs>;
202 def VADDUHS : VX1_Int<576, "vadduhs", int_ppc_altivec_vadduhs>;
203 def VADDUWS : VX1_Int<640, "vadduws", int_ppc_altivec_vadduws>;
206 def VAND : VXForm_1<1028, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
207 "vand $vD, $vA, $vB", VecFP,
208 [(set VRRC:$vD, (and (v4i32 VRRC:$vA), VRRC:$vB))]>;
209 def VANDC : VXForm_1<1092, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
210 "vandc $vD, $vA, $vB", VecFP,
211 [(set VRRC:$vD, (and (v4i32 VRRC:$vA), (vnot VRRC:$vB)))]>;
213 def VCFSX : VXForm_1<842, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
214 "vcfsx $vD, $vB, $UIMM", VecFP,
216 (int_ppc_altivec_vcfsx VRRC:$vB, imm:$UIMM))]>;
217 def VCFUX : VXForm_1<778, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
218 "vcfux $vD, $vB, $UIMM", VecFP,
220 (int_ppc_altivec_vcfux VRRC:$vB, imm:$UIMM))]>;
221 def VCTSXS : VXForm_1<970, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
222 "vctsxs $vD, $vB, $UIMM", VecFP,
224 (int_ppc_altivec_vctsxs VRRC:$vB, imm:$UIMM))]>;
225 def VCTUXS : VXForm_1<906, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
226 "vctuxs $vD, $vB, $UIMM", VecFP,
228 (int_ppc_altivec_vctuxs VRRC:$vB, imm:$UIMM))]>;
229 def VEXPTEFP : VX2_Int<394, "vexptefp", int_ppc_altivec_vexptefp>;
230 def VLOGEFP : VX2_Int<458, "vlogefp", int_ppc_altivec_vlogefp>;
232 def VAVGSB : VX1_Int<1282, "vavgsb", int_ppc_altivec_vavgsb>;
233 def VAVGSH : VX1_Int<1346, "vavgsh", int_ppc_altivec_vavgsh>;
234 def VAVGSW : VX1_Int<1410, "vavgsw", int_ppc_altivec_vavgsw>;
235 def VAVGUB : VX1_Int<1026, "vavgub", int_ppc_altivec_vavgub>;
236 def VAVGUH : VX1_Int<1090, "vavguh", int_ppc_altivec_vavguh>;
237 def VAVGUW : VX1_Int<1154, "vavguw", int_ppc_altivec_vavguw>;
239 def VMAXFP : VX1_Int<1034, "vmaxfp", int_ppc_altivec_vmaxfp>;
240 def VMAXSB : VX1_Int< 258, "vmaxsb", int_ppc_altivec_vmaxsb>;
241 def VMAXSH : VX1_Int< 322, "vmaxsh", int_ppc_altivec_vmaxsh>;
242 def VMAXSW : VX1_Int< 386, "vmaxsw", int_ppc_altivec_vmaxsw>;
243 def VMAXUB : VX1_Int< 2, "vmaxub", int_ppc_altivec_vmaxub>;
244 def VMAXUH : VX1_Int< 66, "vmaxuh", int_ppc_altivec_vmaxuh>;
245 def VMAXUW : VX1_Int< 130, "vmaxuw", int_ppc_altivec_vmaxuw>;
246 def VMINFP : VX1_Int<1098, "vminfp", int_ppc_altivec_vminfp>;
247 def VMINSB : VX1_Int< 770, "vminsb", int_ppc_altivec_vminsb>;
248 def VMINSH : VX1_Int< 834, "vminsh", int_ppc_altivec_vminsh>;
249 def VMINSW : VX1_Int< 896, "vminsw", int_ppc_altivec_vminsw>;
250 def VMINUB : VX1_Int< 514, "vminub", int_ppc_altivec_vminub>;
251 def VMINUH : VX1_Int< 578, "vminuh", int_ppc_altivec_vminuh>;
252 def VMINUW : VX1_Int< 642, "vminuw", int_ppc_altivec_vminuw>;
254 def VMRGHB : VX1_Int<12 , "vmrghb", int_ppc_altivec_vmrghb>;
255 def VMRGHH : VX1_Int<76 , "vmrghh", int_ppc_altivec_vmrghh>;
256 def VMRGHW : VX1_Int<140, "vmrghw", int_ppc_altivec_vmrghw>;
257 def VMRGLB : VX1_Int<268, "vmrglb", int_ppc_altivec_vmrglb>;
258 def VMRGLH : VX1_Int<332, "vmrglh", int_ppc_altivec_vmrglh>;
259 def VMRGLW : VX1_Int<396, "vmrglw", int_ppc_altivec_vmrglw>;
261 def VMSUMMBM : VA1a_Int<37, "vmsummbm", int_ppc_altivec_vmsummbm>;
262 def VMSUMSHM : VA1a_Int<40, "vmsumshm", int_ppc_altivec_vmsumshm>;
263 def VMSUMSHS : VA1a_Int<41, "vmsumshs", int_ppc_altivec_vmsumshs>;
264 def VMSUMUBM : VA1a_Int<36, "vmsumubm", int_ppc_altivec_vmsumubm>;
265 def VMSUMUHM : VA1a_Int<38, "vmsumuhm", int_ppc_altivec_vmsumuhm>;
266 def VMSUMUHS : VA1a_Int<39, "vmsumuhs", int_ppc_altivec_vmsumuhs>;
268 def VMULESB : VX1_Int<776, "vmulesb", int_ppc_altivec_vmulesb>;
269 def VMULESH : VX1_Int<840, "vmulesh", int_ppc_altivec_vmulesh>;
270 def VMULEUB : VX1_Int<520, "vmuleub", int_ppc_altivec_vmuleub>;
271 def VMULEUH : VX1_Int<584, "vmuleuh", int_ppc_altivec_vmuleuh>;
272 def VMULOSB : VX1_Int<264, "vmulosb", int_ppc_altivec_vmulosb>;
273 def VMULOSH : VX1_Int<328, "vmulosh", int_ppc_altivec_vmulosh>;
274 def VMULOUB : VX1_Int< 8, "vmuloub", int_ppc_altivec_vmuloub>;
275 def VMULOUH : VX1_Int< 72, "vmulouh", int_ppc_altivec_vmulouh>;
277 def VREFP : VX2_Int<266, "vrefp", int_ppc_altivec_vrefp>;
278 def VRFIM : VX2_Int<714, "vrfim", int_ppc_altivec_vrfim>;
279 def VRFIN : VX2_Int<522, "vrfin", int_ppc_altivec_vrfin>;
280 def VRFIP : VX2_Int<650, "vrfip", int_ppc_altivec_vrfip>;
281 def VRFIZ : VX2_Int<586, "vrfiz", int_ppc_altivec_vrfiz>;
282 def VRSQRTEFP : VX2_Int<330, "vrsqrtefp", int_ppc_altivec_vrsqrtefp>;
284 def VSUBCUW : VX1_Int<74, "vsubcuw", int_ppc_altivec_vsubcuw>;
286 def VSUBFP : VXForm_1<74, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
287 "vsubfp $vD, $vA, $vB", VecGeneral,
288 [(set VRRC:$vD, (fsub VRRC:$vA, VRRC:$vB))]>;
289 def VSUBUBM : VXForm_1<1024, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
290 "vsububm $vD, $vA, $vB", VecGeneral,
291 [(set VRRC:$vD, (sub (v16i8 VRRC:$vA), VRRC:$vB))]>;
292 def VSUBUHM : VXForm_1<1088, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
293 "vsubuhm $vD, $vA, $vB", VecGeneral,
294 [(set VRRC:$vD, (sub (v8i16 VRRC:$vA), VRRC:$vB))]>;
295 def VSUBUWM : VXForm_1<1152, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
296 "vsubuwm $vD, $vA, $vB", VecGeneral,
297 [(set VRRC:$vD, (sub (v4i32 VRRC:$vA), VRRC:$vB))]>;
299 def VSUBSBS : VX1_Int<1792, "vsubsbs" , int_ppc_altivec_vsubsbs>;
300 def VSUBSHS : VX1_Int<1856, "vsubshs" , int_ppc_altivec_vsubshs>;
301 def VSUBSWS : VX1_Int<1920, "vsubsws" , int_ppc_altivec_vsubsws>;
302 def VSUBUBS : VX1_Int<1536, "vsububs" , int_ppc_altivec_vsububs>;
303 def VSUBUHS : VX1_Int<1600, "vsubuhs" , int_ppc_altivec_vsubuhs>;
304 def VSUBUWS : VX1_Int<1664, "vsubuws" , int_ppc_altivec_vsubuws>;
305 def VSUMSWS : VX1_Int<1928, "vsumsws" , int_ppc_altivec_vsumsws>;
306 def VSUM2SWS: VX1_Int<1672, "vsum2sws", int_ppc_altivec_vsum2sws>;
307 def VSUM4SBS: VX1_Int<1672, "vsum4sbs", int_ppc_altivec_vsum4sbs>;
308 def VSUM4SHS: VX1_Int<1608, "vsum4shs", int_ppc_altivec_vsum4shs>;
309 def VSUM4UBS: VX1_Int<1544, "vsum4ubs", int_ppc_altivec_vsum4ubs>;
311 def VNOR : VXForm_1<1284, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
312 "vnor $vD, $vA, $vB", VecFP,
313 [(set VRRC:$vD, (vnot (or (v4i32 VRRC:$vA), VRRC:$vB)))]>;
314 def VOR : VXForm_1<1156, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
315 "vor $vD, $vA, $vB", VecFP,
316 [(set VRRC:$vD, (or (v4i32 VRRC:$vA), VRRC:$vB))]>;
317 def VXOR : VXForm_1<1220, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
318 "vxor $vD, $vA, $vB", VecFP,
319 [(set VRRC:$vD, (xor (v4i32 VRRC:$vA), VRRC:$vB))]>;
321 def VRLB : VX1_Int< 4, "vrlb", int_ppc_altivec_vrlb>;
322 def VRLH : VX1_Int< 68, "vrlh", int_ppc_altivec_vrlh>;
323 def VRLW : VX1_Int< 132, "vrlw", int_ppc_altivec_vrlw>;
325 def VSL : VX1_Int< 452, "vsl" , int_ppc_altivec_vsl >;
326 def VSLO : VX1_Int<1036, "vslo", int_ppc_altivec_vslo>;
327 def VSLB : VX1_Int< 260, "vslb", int_ppc_altivec_vslb>;
328 def VSLH : VX1_Int< 324, "vslh", int_ppc_altivec_vslh>;
329 def VSLW : VX1_Int< 388, "vslw", int_ppc_altivec_vslw>;
331 def VSPLTB : VXForm_1<524, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
332 "vspltb $vD, $vB, $UIMM", VecPerm,
333 [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vB), (undef),
334 VSPLTB_shuffle_mask:$UIMM))]>;
335 def VSPLTH : VXForm_1<588, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
336 "vsplth $vD, $vB, $UIMM", VecPerm,
337 [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vB), (undef),
338 VSPLTH_shuffle_mask:$UIMM))]>;
339 def VSPLTW : VXForm_1<652, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
340 "vspltw $vD, $vB, $UIMM", VecPerm,
341 [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vB), (undef),
342 VSPLTW_shuffle_mask:$UIMM))]>;
344 def VSR : VX1_Int< 708, "vsr" , int_ppc_altivec_vsr>;
345 def VSRO : VX1_Int<1100, "vsro" , int_ppc_altivec_vsro>;
346 def VSRAB : VX1_Int< 772, "vsrab", int_ppc_altivec_vsrab>;
347 def VSRAH : VX1_Int< 836, "vsrah", int_ppc_altivec_vsrah>;
348 def VSRAW : VX1_Int< 900, "vsraw", int_ppc_altivec_vsraw>;
349 def VSRB : VX1_Int< 516, "vsrb" , int_ppc_altivec_vsrb>;
350 def VSRH : VX1_Int< 580, "vsrh" , int_ppc_altivec_vsrh>;
351 def VSRW : VX1_Int< 644, "vsrw" , int_ppc_altivec_vsrw>;
354 def VSPLTISB : VXForm_3<780, (ops VRRC:$vD, s5imm:$SIMM),
355 "vspltisb $vD, $SIMM", VecPerm,
356 [(set VRRC:$vD, (v4f32 vecspltisb:$SIMM))]>;
357 def VSPLTISH : VXForm_3<844, (ops VRRC:$vD, s5imm:$SIMM),
358 "vspltish $vD, $SIMM", VecPerm,
359 [(set VRRC:$vD, (v4f32 vecspltish:$SIMM))]>;
360 def VSPLTISW : VXForm_3<908, (ops VRRC:$vD, s5imm:$SIMM),
361 "vspltisw $vD, $SIMM", VecPerm,
362 [(set VRRC:$vD, (v4f32 vecspltisw:$SIMM))]>;
365 def VPKPX : VX1_Int<782, "vpkpx", int_ppc_altivec_vpkpx>;
366 def VPKSHSS : VX1_Int<398, "vpkshss", int_ppc_altivec_vpkshss>;
367 def VPKSHUS : VX1_Int<270, "vpkshus", int_ppc_altivec_vpkshus>;
368 def VPKSWSS : VX1_Int<462, "vpkswss", int_ppc_altivec_vpkswss>;
369 def VPKSWUS : VX1_Int<334, "vpkswus", int_ppc_altivec_vpkswus>;
370 def VPKUHUM : VXForm_1<14, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
371 "vpkuhum $vD, $vA, $vB", VecFP,
373 def VPKUHUS : VX1_Int<142, "vpkuhus", int_ppc_altivec_vpkuhus>;
374 def VPKUWUM : VXForm_1<78, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
375 "vpkuwum $vD, $vA, $vB", VecFP,
377 def VPKUWUS : VX1_Int<206, "vpkuwus", int_ppc_altivec_vpkuwus>;
380 def VUPKHPX : VX2_Int<846, "vupkhpx", int_ppc_altivec_vupkhpx>;
381 def VUPKHSB : VX2_Int<526, "vupkhsb", int_ppc_altivec_vupkhsb>;
382 def VUPKHSH : VX2_Int<590, "vupkhsh", int_ppc_altivec_vupkhsh>;
383 def VUPKLPX : VX2_Int<974, "vupklpx", int_ppc_altivec_vupklpx>;
384 def VUPKLSB : VX2_Int<654, "vupklsb", int_ppc_altivec_vupklsb>;
385 def VUPKLSH : VX2_Int<718, "vupklsh", int_ppc_altivec_vupklsh>;
388 // Altivec Comparisons.
390 class VCMP<bits<10> xo, string asmstr, ValueType Ty>
391 : VXRForm_1<xo, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB), asmstr, VecFPCompare,
392 [(set VRRC:$vD, (Ty (PPCvcmp VRRC:$vA, VRRC:$vB, xo)))]>;
393 class VCMPo<bits<10> xo, string asmstr, ValueType Ty>
394 : VXRForm_1<xo, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB), asmstr, VecFPCompare,
395 [(set VRRC:$vD, (Ty (PPCvcmp_o VRRC:$vA, VRRC:$vB, xo)))]> {
400 // f32 element comparisons.0
401 def VCMPBFP : VCMP <966, "vcmpbfp $vD, $vA, $vB" , v4f32>;
402 def VCMPBFPo : VCMPo<966, "vcmpbfp. $vD, $vA, $vB" , v4f32>;
403 def VCMPEQFP : VCMP <198, "vcmpeqfp $vD, $vA, $vB" , v4f32>;
404 def VCMPEQFPo : VCMPo<198, "vcmpeqfp. $vD, $vA, $vB", v4f32>;
405 def VCMPGEFP : VCMP <454, "vcmpgefp $vD, $vA, $vB" , v4f32>;
406 def VCMPGEFPo : VCMPo<454, "vcmpgefp. $vD, $vA, $vB", v4f32>;
407 def VCMPGTFP : VCMP <710, "vcmpgtfp $vD, $vA, $vB" , v4f32>;
408 def VCMPGTFPo : VCMPo<710, "vcmpgtfp. $vD, $vA, $vB", v4f32>;
410 // i8 element comparisons.
411 def VCMPEQUB : VCMP < 6, "vcmpequb $vD, $vA, $vB" , v16i8>;
412 def VCMPEQUBo : VCMPo< 6, "vcmpequb. $vD, $vA, $vB", v16i8>;
413 def VCMPGTSB : VCMP <774, "vcmpgtsb $vD, $vA, $vB" , v16i8>;
414 def VCMPGTSBo : VCMPo<774, "vcmpgtsb. $vD, $vA, $vB", v16i8>;
415 def VCMPGTUB : VCMP <518, "vcmpgtub $vD, $vA, $vB" , v16i8>;
416 def VCMPGTUBo : VCMPo<518, "vcmpgtub. $vD, $vA, $vB", v16i8>;
418 // i16 element comparisons.
419 def VCMPEQUH : VCMP < 70, "vcmpequh $vD, $vA, $vB" , v8i16>;
420 def VCMPEQUHo : VCMPo< 70, "vcmpequh. $vD, $vA, $vB", v8i16>;
421 def VCMPGTSH : VCMP <838, "vcmpgtsh $vD, $vA, $vB" , v8i16>;
422 def VCMPGTSHo : VCMPo<838, "vcmpgtsh. $vD, $vA, $vB", v8i16>;
423 def VCMPGTUH : VCMP <582, "vcmpgtuh $vD, $vA, $vB" , v8i16>;
424 def VCMPGTUHo : VCMPo<582, "vcmpgtuh. $vD, $vA, $vB", v8i16>;
426 // i32 element comparisons.
427 def VCMPEQUW : VCMP <134, "vcmpequw $vD, $vA, $vB" , v4i32>;
428 def VCMPEQUWo : VCMPo<134, "vcmpequw. $vD, $vA, $vB", v4i32>;
429 def VCMPGTSW : VCMP <902, "vcmpgtsw $vD, $vA, $vB" , v4i32>;
430 def VCMPGTSWo : VCMPo<902, "vcmpgtsw. $vD, $vA, $vB", v4i32>;
431 def VCMPGTUW : VCMP <646, "vcmpgtuw $vD, $vA, $vB" , v4i32>;
432 def VCMPGTUWo : VCMPo<646, "vcmpgtuw. $vD, $vA, $vB", v4i32>;
434 def V_SET0 : VXForm_setzero<1220, (ops VRRC:$vD),
435 "vxor $vD, $vD, $vD", VecFP,
436 [(set VRRC:$vD, (v4f32 immAllZerosV))]>;
439 //===----------------------------------------------------------------------===//
440 // Additional Altivec Patterns
444 def : Pat<(int_ppc_altivec_dss imm:$STRM), (DSS 0, imm:$STRM, 0, 0)>;
445 def : Pat<(int_ppc_altivec_dssall), (DSS 1, 0, 0, 0)>;
446 def : Pat<(int_ppc_altivec_dst GPRC:$rA, GPRC:$rB, imm:$STRM),
447 (DST 0, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
448 def : Pat<(int_ppc_altivec_dstt GPRC:$rA, GPRC:$rB, imm:$STRM),
449 (DST 1, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
450 def : Pat<(int_ppc_altivec_dstst GPRC:$rA, GPRC:$rB, imm:$STRM),
451 (DSTST 0, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
452 def : Pat<(int_ppc_altivec_dststt GPRC:$rA, GPRC:$rB, imm:$STRM),
453 (DSTST 1, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
456 def : Pat<(v16i8 (undef)), (v16i8 (IMPLICIT_DEF_VRRC))>;
457 def : Pat<(v8i16 (undef)), (v8i16 (IMPLICIT_DEF_VRRC))>;
458 def : Pat<(v4i32 (undef)), (v4i32 (IMPLICIT_DEF_VRRC))>;
459 def : Pat<(v16i8 immAllZerosV), (v16i8 (V_SET0))>;
460 def : Pat<(v8i16 immAllZerosV), (v8i16 (V_SET0))>;
461 def : Pat<(v4i32 immAllZerosV), (v4i32 (V_SET0))>;
464 def : Pat<(v16i8 (load xoaddr:$src)), (v16i8 (LVX xoaddr:$src))>;
465 def : Pat<(v8i16 (load xoaddr:$src)), (v8i16 (LVX xoaddr:$src))>;
466 def : Pat<(v4i32 (load xoaddr:$src)), (v4i32 (LVX xoaddr:$src))>;
467 def : Pat<(v4f32 (load xoaddr:$src)), (v4f32 (LVX xoaddr:$src))>;
470 def : Pat<(store (v16i8 VRRC:$rS), xoaddr:$dst),
471 (STVX (v16i8 VRRC:$rS), xoaddr:$dst)>;
472 def : Pat<(store (v8i16 VRRC:$rS), xoaddr:$dst),
473 (STVX (v8i16 VRRC:$rS), xoaddr:$dst)>;
474 def : Pat<(store (v4i32 VRRC:$rS), xoaddr:$dst),
475 (STVX (v4i32 VRRC:$rS), xoaddr:$dst)>;
476 def : Pat<(store (v4f32 VRRC:$rS), xoaddr:$dst),
477 (STVX (v4f32 VRRC:$rS), xoaddr:$dst)>;
480 def : Pat<(v16i8 (bitconvert (v8i16 VRRC:$src))), (v16i8 VRRC:$src)>;
481 def : Pat<(v16i8 (bitconvert (v4i32 VRRC:$src))), (v16i8 VRRC:$src)>;
482 def : Pat<(v16i8 (bitconvert (v4f32 VRRC:$src))), (v16i8 VRRC:$src)>;
484 def : Pat<(v8i16 (bitconvert (v16i8 VRRC:$src))), (v8i16 VRRC:$src)>;
485 def : Pat<(v8i16 (bitconvert (v4i32 VRRC:$src))), (v8i16 VRRC:$src)>;
486 def : Pat<(v8i16 (bitconvert (v4f32 VRRC:$src))), (v8i16 VRRC:$src)>;
488 def : Pat<(v4i32 (bitconvert (v16i8 VRRC:$src))), (v4i32 VRRC:$src)>;
489 def : Pat<(v4i32 (bitconvert (v8i16 VRRC:$src))), (v4i32 VRRC:$src)>;
490 def : Pat<(v4i32 (bitconvert (v4f32 VRRC:$src))), (v4i32 VRRC:$src)>;
492 def : Pat<(v4f32 (bitconvert (v16i8 VRRC:$src))), (v4f32 VRRC:$src)>;
493 def : Pat<(v4f32 (bitconvert (v8i16 VRRC:$src))), (v4f32 VRRC:$src)>;
494 def : Pat<(v4f32 (bitconvert (v4i32 VRRC:$src))), (v4f32 VRRC:$src)>;
496 // Immediate vector formation with vsplti*.
497 def : Pat<(v16i8 vecspltisb:$invec), (v16i8 (VSPLTISB vecspltisb:$invec))>;
498 def : Pat<(v16i8 vecspltish:$invec), (v16i8 (VSPLTISH vecspltish:$invec))>;
499 def : Pat<(v16i8 vecspltisw:$invec), (v16i8 (VSPLTISW vecspltisw:$invec))>;
501 def : Pat<(v8i16 vecspltisb:$invec), (v8i16 (VSPLTISB vecspltisb:$invec))>;
502 def : Pat<(v8i16 vecspltish:$invec), (v8i16 (VSPLTISH vecspltish:$invec))>;
503 def : Pat<(v8i16 vecspltisw:$invec), (v8i16 (VSPLTISW vecspltisw:$invec))>;
505 def : Pat<(v4i32 vecspltisb:$invec), (v4i32 (VSPLTISB vecspltisb:$invec))>;
506 def : Pat<(v4i32 vecspltish:$invec), (v4i32 (VSPLTISH vecspltish:$invec))>;
507 def : Pat<(v4i32 vecspltisw:$invec), (v4i32 (VSPLTISW vecspltisw:$invec))>;
509 // Logical Operations
510 def : Pat<(v16i8 (vnot VRRC:$vA)), (v16i8 (VNOR VRRC:$vA, VRRC:$vA))>;
511 def : Pat<(v8i16 (vnot VRRC:$vA)), (v8i16 (VNOR VRRC:$vA, VRRC:$vA))>;
512 def : Pat<(v4i32 (vnot VRRC:$vA)), (v4i32 (VNOR VRRC:$vA, VRRC:$vA))>;
514 def : Pat<(v16i8 (and VRRC:$A, VRRC:$B)), (v16i8 (VAND VRRC:$A, VRRC:$B))>;
515 def : Pat<(v8i16 (and VRRC:$A, VRRC:$B)), (v8i16 (VAND VRRC:$A, VRRC:$B))>;
516 def : Pat<(v16i8 (or VRRC:$A, VRRC:$B)), (v16i8 (VOR VRRC:$A, VRRC:$B))>;
517 def : Pat<(v8i16 (or VRRC:$A, VRRC:$B)), (v8i16 (VOR VRRC:$A, VRRC:$B))>;
518 def : Pat<(v16i8 (xor VRRC:$A, VRRC:$B)), (v16i8 (VXOR VRRC:$A, VRRC:$B))>;
519 def : Pat<(v8i16 (xor VRRC:$A, VRRC:$B)), (v8i16 (VXOR VRRC:$A, VRRC:$B))>;
520 def : Pat<(v16i8 (vnot (or VRRC:$A, VRRC:$B))),(v16i8 (VNOR VRRC:$A, VRRC:$B))>;
521 def : Pat<(v8i16 (vnot (or VRRC:$A, VRRC:$B))),(v8i16 (VNOR VRRC:$A, VRRC:$B))>;
522 def : Pat<(v16i8 (and VRRC:$A, (vnot VRRC:$B))),
523 (v16i8 (VANDC VRRC:$A, VRRC:$B))>;
524 def : Pat<(v8i16 (and VRRC:$A, (vnot VRRC:$B))),
525 (v8i16 (VANDC VRRC:$A, VRRC:$B))>;
527 def : Pat<(fmul VRRC:$vA, VRRC:$vB),
528 (VMADDFP VRRC:$vA, VRRC:$vB, (V_SET0))>;
530 // Fused multiply add and multiply sub for packed float. These are represented
531 // separately from the real instructions above, for operations that must have
532 // the additional precision, such as Newton-Rhapson (used by divide, sqrt)
533 def : Pat<(PPCvmaddfp VRRC:$A, VRRC:$B, VRRC:$C),
534 (VMADDFP VRRC:$A, VRRC:$B, VRRC:$C)>;
535 def : Pat<(PPCvnmsubfp VRRC:$A, VRRC:$B, VRRC:$C),
536 (VNMSUBFP VRRC:$A, VRRC:$B, VRRC:$C)>;
538 def : Pat<(int_ppc_altivec_vmaddfp VRRC:$A, VRRC:$B, VRRC:$C),
539 (VMADDFP VRRC:$A, VRRC:$B, VRRC:$C)>;
540 def : Pat<(int_ppc_altivec_vnmsubfp VRRC:$A, VRRC:$B, VRRC:$C),
541 (VNMSUBFP VRRC:$A, VRRC:$B, VRRC:$C)>;
543 def : Pat<(PPCvperm (v16i8 VRRC:$vA), VRRC:$vB, VRRC:$vC),
544 (v16i8 (VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC))>;