1 //===-- SystemZInstrInfo.td - General SystemZ instructions ----*- tblgen-*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 //===----------------------------------------------------------------------===//
12 //===----------------------------------------------------------------------===//
14 def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i64imm:$amt),
15 [(callseq_start timm:$amt)]>;
16 def ADJCALLSTACKUP : Pseudo<(outs), (ins i64imm:$amt1, i64imm:$amt2),
17 [(callseq_end timm:$amt1, timm:$amt2)]>;
19 let hasSideEffects = 0 in {
20 // Takes as input the value of the stack pointer after a dynamic allocation
21 // has been made. Sets the output to the address of the dynamically-
22 // allocated area itself, skipping the outgoing arguments.
24 // This expands to an LA or LAY instruction. We restrict the offset
25 // to the range of LA and keep the LAY range in reserve for when
26 // the size of the outgoing arguments is added.
27 def ADJDYNALLOC : Pseudo<(outs GR64:$dst), (ins dynalloc12only:$src),
28 [(set GR64:$dst, dynalloc12only:$src)]>;
31 //===----------------------------------------------------------------------===//
32 // Control flow instructions
33 //===----------------------------------------------------------------------===//
35 // A return instruction (br %r14).
36 let isReturn = 1, isTerminator = 1, isBarrier = 1, hasCtrlDep = 1 in
37 def Return : Alias<2, (outs), (ins), [(z_retflag)]>;
39 // Unconditional branches. R1 is the condition-code mask (all 1s).
40 let isBranch = 1, isTerminator = 1, isBarrier = 1, R1 = 15 in {
41 let isIndirectBranch = 1 in
42 def BR : InstRR<0x07, (outs), (ins ADDR64:$R2),
43 "br\t$R2", [(brind ADDR64:$R2)]>;
45 // An assembler extended mnemonic for BRC.
46 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2), "j\t$I2",
49 // An assembler extended mnemonic for BRCL. (The extension is "G"
50 // rather than "L" because "JL" is "Jump if Less".)
51 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2), "jg\t$I2", []>;
54 // Conditional branches. It's easier for LLVM to handle these branches
55 // in their raw BRC/BRCL form, with the 4-bit condition-code mask being
56 // the first operand. It seems friendlier to use mnemonic forms like
57 // JE and JLH when writing out the assembly though.
58 let isBranch = 1, isTerminator = 1, Uses = [CC] in {
59 let isCodeGenOnly = 1, CCMaskFirst = 1 in {
60 def BRC : InstRI<0xA74, (outs), (ins cond4:$valid, cond4:$R1,
61 brtarget16:$I2), "j$R1\t$I2",
62 [(z_br_ccmask cond4:$valid, cond4:$R1, bb:$I2)]>;
63 def BRCL : InstRIL<0xC04, (outs), (ins cond4:$valid, cond4:$R1,
64 brtarget32:$I2), "jg$R1\t$I2", []>;
66 def AsmBRC : InstRI<0xA74, (outs), (ins imm32zx4:$R1, brtarget16:$I2),
68 def AsmBRCL : InstRIL<0xC04, (outs), (ins imm32zx4:$R1, brtarget32:$I2),
69 "brcl\t$R1, $I2", []>;
70 def AsmBCR : InstRR<0x07, (outs), (ins imm32zx4:$R1, GR64:$R2),
74 // Fused compare-and-branch instructions. As for normal branches,
75 // we handle these instructions internally in their raw CRJ-like form,
76 // but use assembly macros like CRJE when writing them out.
78 // These instructions do not use or clobber the condition codes.
79 // We nevertheless pretend that they clobber CC, so that we can lower
80 // them to separate comparisons and BRCLs if the branch ends up being
82 multiclass CompareBranches<Operand ccmask, string pos1, string pos2> {
83 let isBranch = 1, isTerminator = 1, Defs = [CC] in {
84 def RJ : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2, ccmask:$M3,
86 "crj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
87 def GRJ : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2, ccmask:$M3,
89 "cgrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
90 def IJ : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2, ccmask:$M3,
92 "cij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
93 def GIJ : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2, ccmask:$M3,
95 "cgij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
96 def LRJ : InstRIEb<0xEC77, (outs), (ins GR32:$R1, GR32:$R2, ccmask:$M3,
98 "clrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
99 def LGRJ : InstRIEb<0xEC65, (outs), (ins GR64:$R1, GR64:$R2, ccmask:$M3,
101 "clgrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
102 def LIJ : InstRIEc<0xEC7F, (outs), (ins GR32:$R1, imm32zx8:$I2, ccmask:$M3,
104 "clij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
105 def LGIJ : InstRIEc<0xEC7D, (outs), (ins GR64:$R1, imm64zx8:$I2, ccmask:$M3,
107 "clgij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
110 let isCodeGenOnly = 1 in
111 defm C : CompareBranches<cond4, "$M3", "">;
112 defm AsmC : CompareBranches<imm32zx4, "", "$M3, ">;
114 // Define AsmParser mnemonics for each general condition-code mask
115 // (integer or floating-point)
116 multiclass CondExtendedMnemonic<bits<4> ccmask, string name> {
118 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2),
119 "j"##name##"\t$I2", []>;
120 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2),
121 "jg"##name##"\t$I2", []>;
122 def BR : InstRR<0x07, (outs), (ins ADDR64:$R2), "b"##name##"r\t$R2", []>;
124 def LOCR : FixedCondUnaryRRF<"locr"##name, 0xB9F2, GR32, GR32, ccmask>;
125 def LOCGR : FixedCondUnaryRRF<"locgr"##name, 0xB9E2, GR64, GR64, ccmask>;
126 def LOC : FixedCondUnaryRSY<"loc"##name, 0xEBF2, GR32, ccmask, 4>;
127 def LOCG : FixedCondUnaryRSY<"locg"##name, 0xEBE2, GR64, ccmask, 8>;
128 def STOC : FixedCondStoreRSY<"stoc"##name, 0xEBF3, GR32, ccmask, 4>;
129 def STOCG : FixedCondStoreRSY<"stocg"##name, 0xEBE3, GR64, ccmask, 8>;
131 defm AsmO : CondExtendedMnemonic<1, "o">;
132 defm AsmH : CondExtendedMnemonic<2, "h">;
133 defm AsmNLE : CondExtendedMnemonic<3, "nle">;
134 defm AsmL : CondExtendedMnemonic<4, "l">;
135 defm AsmNHE : CondExtendedMnemonic<5, "nhe">;
136 defm AsmLH : CondExtendedMnemonic<6, "lh">;
137 defm AsmNE : CondExtendedMnemonic<7, "ne">;
138 defm AsmE : CondExtendedMnemonic<8, "e">;
139 defm AsmNLH : CondExtendedMnemonic<9, "nlh">;
140 defm AsmHE : CondExtendedMnemonic<10, "he">;
141 defm AsmNL : CondExtendedMnemonic<11, "nl">;
142 defm AsmLE : CondExtendedMnemonic<12, "le">;
143 defm AsmNH : CondExtendedMnemonic<13, "nh">;
144 defm AsmNO : CondExtendedMnemonic<14, "no">;
146 // Define AsmParser mnemonics for each integer condition-code mask.
147 // This is like the list above, except that condition 3 is not possible
148 // and that the low bit of the mask is therefore always 0. This means
149 // that each condition has two names. Conditions "o" and "no" are not used.
151 // We don't make one of the two names an alias of the other because
152 // we need the custom parsing routines to select the correct register class.
153 multiclass IntCondExtendedMnemonicA<bits<4> ccmask, string name> {
155 def CR : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2,
157 "crj"##name##"\t$R1, $R2, $RI4", []>;
158 def CGR : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2,
160 "cgrj"##name##"\t$R1, $R2, $RI4", []>;
161 def CI : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2,
163 "cij"##name##"\t$R1, $I2, $RI4", []>;
164 def CGI : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2,
166 "cgij"##name##"\t$R1, $I2, $RI4", []>;
167 def CLR : InstRIEb<0xEC77, (outs), (ins GR32:$R1, GR32:$R2,
169 "clrj"##name##"\t$R1, $R2, $RI4", []>;
170 def CLGR : InstRIEb<0xEC65, (outs), (ins GR64:$R1, GR64:$R2,
172 "clgrj"##name##"\t$R1, $R2, $RI4", []>;
173 def CLI : InstRIEc<0xEC7F, (outs), (ins GR32:$R1, imm32zx8:$I2,
175 "clij"##name##"\t$R1, $I2, $RI4", []>;
176 def CLGI : InstRIEc<0xEC7D, (outs), (ins GR64:$R1, imm64zx8:$I2,
178 "clgij"##name##"\t$R1, $I2, $RI4", []>;
181 multiclass IntCondExtendedMnemonic<bits<4> ccmask, string name1, string name2>
182 : IntCondExtendedMnemonicA<ccmask, name1> {
183 let isAsmParserOnly = 1 in
184 defm Alt : IntCondExtendedMnemonicA<ccmask, name2>;
186 defm AsmJH : IntCondExtendedMnemonic<2, "h", "nle">;
187 defm AsmJL : IntCondExtendedMnemonic<4, "l", "nhe">;
188 defm AsmJLH : IntCondExtendedMnemonic<6, "lh", "ne">;
189 defm AsmJE : IntCondExtendedMnemonic<8, "e", "nlh">;
190 defm AsmJHE : IntCondExtendedMnemonic<10, "he", "nl">;
191 defm AsmJLE : IntCondExtendedMnemonic<12, "le", "nh">;
193 // Decrement a register and branch if it is nonzero. These don't clobber CC,
194 // but we might need to split long branches into sequences that do.
196 def BRCT : BranchUnaryRI<"brct", 0xA76, GR32>;
197 def BRCTG : BranchUnaryRI<"brctg", 0xA77, GR64>;
200 //===----------------------------------------------------------------------===//
201 // Select instructions
202 //===----------------------------------------------------------------------===//
204 def Select32Mux : SelectWrapper<GRX32>, Requires<[FeatureHighWord]>;
205 def Select32 : SelectWrapper<GR32>;
206 def Select64 : SelectWrapper<GR64>;
208 // We don't define 32-bit Mux stores because the low-only STOC should
209 // always be used if possible.
210 defm CondStore8Mux : CondStores<GRX32, nonvolatile_truncstorei8,
211 nonvolatile_anyextloadi8, bdxaddr20only>,
212 Requires<[FeatureHighWord]>;
213 defm CondStore16Mux : CondStores<GRX32, nonvolatile_truncstorei16,
214 nonvolatile_anyextloadi16, bdxaddr20only>,
215 Requires<[FeatureHighWord]>;
216 defm CondStore8 : CondStores<GR32, nonvolatile_truncstorei8,
217 nonvolatile_anyextloadi8, bdxaddr20only>;
218 defm CondStore16 : CondStores<GR32, nonvolatile_truncstorei16,
219 nonvolatile_anyextloadi16, bdxaddr20only>;
220 defm CondStore32 : CondStores<GR32, nonvolatile_store,
221 nonvolatile_load, bdxaddr20only>;
223 defm : CondStores64<CondStore8, CondStore8Inv, nonvolatile_truncstorei8,
224 nonvolatile_anyextloadi8, bdxaddr20only>;
225 defm : CondStores64<CondStore16, CondStore16Inv, nonvolatile_truncstorei16,
226 nonvolatile_anyextloadi16, bdxaddr20only>;
227 defm : CondStores64<CondStore32, CondStore32Inv, nonvolatile_truncstorei32,
228 nonvolatile_anyextloadi32, bdxaddr20only>;
229 defm CondStore64 : CondStores<GR64, nonvolatile_store,
230 nonvolatile_load, bdxaddr20only>;
232 //===----------------------------------------------------------------------===//
234 //===----------------------------------------------------------------------===//
236 let isCall = 1, Defs = [R14D, CC] in {
237 def CallBRASL : Alias<6, (outs), (ins pcrel32:$I2, variable_ops),
238 [(z_call pcrel32:$I2)]>;
239 def CallBASR : Alias<2, (outs), (ins ADDR64:$R2, variable_ops),
240 [(z_call ADDR64:$R2)]>;
243 // Sibling calls. Indirect sibling calls must be via R1, since R2 upwards
244 // are argument registers and since branching to R0 is a no-op.
245 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in {
246 def CallJG : Alias<6, (outs), (ins pcrel32:$I2),
247 [(z_sibcall pcrel32:$I2)]>;
249 def CallBR : Alias<2, (outs), (ins), [(z_sibcall R1D)]>;
252 // Define the general form of the call instructions for the asm parser.
253 // These instructions don't hard-code %r14 as the return address register.
254 // Allow an optional TLS marker symbol to generate TLS call relocations.
255 def BRAS : InstRI<0xA75, (outs), (ins GR64:$R1, brtarget16tls:$I2),
256 "bras\t$R1, $I2", []>;
257 def BRASL : InstRIL<0xC05, (outs), (ins GR64:$R1, brtarget32tls:$I2),
258 "brasl\t$R1, $I2", []>;
259 def BASR : InstRR<0x0D, (outs), (ins GR64:$R1, ADDR64:$R2),
260 "basr\t$R1, $R2", []>;
262 //===----------------------------------------------------------------------===//
264 //===----------------------------------------------------------------------===//
267 let hasSideEffects = 0 in {
268 // Expands to LR, RISBHG or RISBLG, depending on the choice of registers.
269 def LRMux : UnaryRRPseudo<"l", null_frag, GRX32, GRX32>,
270 Requires<[FeatureHighWord]>;
271 def LR : UnaryRR <"l", 0x18, null_frag, GR32, GR32>;
272 def LGR : UnaryRRE<"lg", 0xB904, null_frag, GR64, GR64>;
274 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
275 def LTR : UnaryRR <"lt", 0x12, null_frag, GR32, GR32>;
276 def LTGR : UnaryRRE<"ltg", 0xB902, null_frag, GR64, GR64>;
279 // Move on condition.
280 let isCodeGenOnly = 1, Uses = [CC] in {
281 def LOCR : CondUnaryRRF<"loc", 0xB9F2, GR32, GR32>;
282 def LOCGR : CondUnaryRRF<"locg", 0xB9E2, GR64, GR64>;
285 def AsmLOCR : AsmCondUnaryRRF<"loc", 0xB9F2, GR32, GR32>;
286 def AsmLOCGR : AsmCondUnaryRRF<"locg", 0xB9E2, GR64, GR64>;
290 let hasSideEffects = 0, isAsCheapAsAMove = 1, isMoveImm = 1,
291 isReMaterializable = 1 in {
292 // 16-bit sign-extended immediates. LHIMux expands to LHI or IIHF,
293 // deopending on the choice of register.
294 def LHIMux : UnaryRIPseudo<bitconvert, GRX32, imm32sx16>,
295 Requires<[FeatureHighWord]>;
296 def LHI : UnaryRI<"lhi", 0xA78, bitconvert, GR32, imm32sx16>;
297 def LGHI : UnaryRI<"lghi", 0xA79, bitconvert, GR64, imm64sx16>;
299 // Other 16-bit immediates.
300 def LLILL : UnaryRI<"llill", 0xA5F, bitconvert, GR64, imm64ll16>;
301 def LLILH : UnaryRI<"llilh", 0xA5E, bitconvert, GR64, imm64lh16>;
302 def LLIHL : UnaryRI<"llihl", 0xA5D, bitconvert, GR64, imm64hl16>;
303 def LLIHH : UnaryRI<"llihh", 0xA5C, bitconvert, GR64, imm64hh16>;
305 // 32-bit immediates.
306 def LGFI : UnaryRIL<"lgfi", 0xC01, bitconvert, GR64, imm64sx32>;
307 def LLILF : UnaryRIL<"llilf", 0xC0F, bitconvert, GR64, imm64lf32>;
308 def LLIHF : UnaryRIL<"llihf", 0xC0E, bitconvert, GR64, imm64hf32>;
312 let canFoldAsLoad = 1, SimpleBDXLoad = 1 in {
313 // Expands to L, LY or LFH, depending on the choice of register.
314 def LMux : UnaryRXYPseudo<"l", load, GRX32, 4>,
315 Requires<[FeatureHighWord]>;
316 defm L : UnaryRXPair<"l", 0x58, 0xE358, load, GR32, 4>;
317 def LFH : UnaryRXY<"lfh", 0xE3CA, load, GRH32, 4>,
318 Requires<[FeatureHighWord]>;
319 def LG : UnaryRXY<"lg", 0xE304, load, GR64, 8>;
321 // These instructions are split after register allocation, so we don't
322 // want a custom inserter.
323 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
324 def L128 : Pseudo<(outs GR128:$dst), (ins bdxaddr20only128:$src),
325 [(set GR128:$dst, (load bdxaddr20only128:$src))]>;
328 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
329 def LT : UnaryRXY<"lt", 0xE312, load, GR32, 4>;
330 def LTG : UnaryRXY<"ltg", 0xE302, load, GR64, 8>;
333 let canFoldAsLoad = 1 in {
334 def LRL : UnaryRILPC<"lrl", 0xC4D, aligned_load, GR32>;
335 def LGRL : UnaryRILPC<"lgrl", 0xC48, aligned_load, GR64>;
338 // Load on condition.
339 let isCodeGenOnly = 1, Uses = [CC] in {
340 def LOC : CondUnaryRSY<"loc", 0xEBF2, nonvolatile_load, GR32, 4>;
341 def LOCG : CondUnaryRSY<"locg", 0xEBE2, nonvolatile_load, GR64, 8>;
344 def AsmLOC : AsmCondUnaryRSY<"loc", 0xEBF2, GR32, 4>;
345 def AsmLOCG : AsmCondUnaryRSY<"locg", 0xEBE2, GR64, 8>;
349 let SimpleBDXStore = 1 in {
350 // Expands to ST, STY or STFH, depending on the choice of register.
351 def STMux : StoreRXYPseudo<store, GRX32, 4>,
352 Requires<[FeatureHighWord]>;
353 defm ST : StoreRXPair<"st", 0x50, 0xE350, store, GR32, 4>;
354 def STFH : StoreRXY<"stfh", 0xE3CB, store, GRH32, 4>,
355 Requires<[FeatureHighWord]>;
356 def STG : StoreRXY<"stg", 0xE324, store, GR64, 8>;
358 // These instructions are split after register allocation, so we don't
359 // want a custom inserter.
360 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
361 def ST128 : Pseudo<(outs), (ins GR128:$src, bdxaddr20only128:$dst),
362 [(store GR128:$src, bdxaddr20only128:$dst)]>;
365 def STRL : StoreRILPC<"strl", 0xC4F, aligned_store, GR32>;
366 def STGRL : StoreRILPC<"stgrl", 0xC4B, aligned_store, GR64>;
368 // Store on condition.
369 let isCodeGenOnly = 1, Uses = [CC] in {
370 def STOC : CondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
371 def STOCG : CondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
374 def AsmSTOC : AsmCondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
375 def AsmSTOCG : AsmCondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
378 // 8-bit immediate stores to 8-bit fields.
379 defm MVI : StoreSIPair<"mvi", 0x92, 0xEB52, truncstorei8, imm32zx8trunc>;
381 // 16-bit immediate stores to 16-, 32- or 64-bit fields.
382 def MVHHI : StoreSIL<"mvhhi", 0xE544, truncstorei16, imm32sx16trunc>;
383 def MVHI : StoreSIL<"mvhi", 0xE54C, store, imm32sx16>;
384 def MVGHI : StoreSIL<"mvghi", 0xE548, store, imm64sx16>;
386 // Memory-to-memory moves.
387 let mayLoad = 1, mayStore = 1 in
388 defm MVC : MemorySS<"mvc", 0xD2, z_mvc, z_mvc_loop>;
391 let mayLoad = 1, mayStore = 1, Defs = [CC], Uses = [R0L] in
392 defm MVST : StringRRE<"mvst", 0xB255, z_stpcpy>;
394 //===----------------------------------------------------------------------===//
396 //===----------------------------------------------------------------------===//
398 // Note that putting these before zero extensions mean that we will prefer
399 // them for anyextload*. There's not really much to choose between the two
400 // either way, but signed-extending loads have a short LH and a long LHY,
401 // while zero-extending loads have only the long LLH.
403 //===----------------------------------------------------------------------===//
405 // 32-bit extensions from registers.
406 let hasSideEffects = 0 in {
407 def LBR : UnaryRRE<"lb", 0xB926, sext8, GR32, GR32>;
408 def LHR : UnaryRRE<"lh", 0xB927, sext16, GR32, GR32>;
411 // 64-bit extensions from registers.
412 let hasSideEffects = 0 in {
413 def LGBR : UnaryRRE<"lgb", 0xB906, sext8, GR64, GR64>;
414 def LGHR : UnaryRRE<"lgh", 0xB907, sext16, GR64, GR64>;
415 def LGFR : UnaryRRE<"lgf", 0xB914, sext32, GR64, GR32>;
417 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in
418 def LTGFR : UnaryRRE<"ltgf", 0xB912, null_frag, GR64, GR64>;
420 // Match 32-to-64-bit sign extensions in which the source is already
421 // in a 64-bit register.
422 def : Pat<(sext_inreg GR64:$src, i32),
423 (LGFR (EXTRACT_SUBREG GR64:$src, subreg_l32))>;
425 // 32-bit extensions from 8-bit memory. LBMux expands to LB or LBH,
426 // depending on the choice of register.
427 def LBMux : UnaryRXYPseudo<"lb", asextloadi8, GRX32, 1>,
428 Requires<[FeatureHighWord]>;
429 def LB : UnaryRXY<"lb", 0xE376, asextloadi8, GR32, 1>;
430 def LBH : UnaryRXY<"lbh", 0xE3C0, asextloadi8, GRH32, 1>,
431 Requires<[FeatureHighWord]>;
433 // 32-bit extensions from 16-bit memory. LHMux expands to LH or LHH,
434 // depending on the choice of register.
435 def LHMux : UnaryRXYPseudo<"lh", asextloadi16, GRX32, 2>,
436 Requires<[FeatureHighWord]>;
437 defm LH : UnaryRXPair<"lh", 0x48, 0xE378, asextloadi16, GR32, 2>;
438 def LHH : UnaryRXY<"lhh", 0xE3C4, asextloadi16, GRH32, 2>,
439 Requires<[FeatureHighWord]>;
440 def LHRL : UnaryRILPC<"lhrl", 0xC45, aligned_asextloadi16, GR32>;
442 // 64-bit extensions from memory.
443 def LGB : UnaryRXY<"lgb", 0xE377, asextloadi8, GR64, 1>;
444 def LGH : UnaryRXY<"lgh", 0xE315, asextloadi16, GR64, 2>;
445 def LGF : UnaryRXY<"lgf", 0xE314, asextloadi32, GR64, 4>;
446 def LGHRL : UnaryRILPC<"lghrl", 0xC44, aligned_asextloadi16, GR64>;
447 def LGFRL : UnaryRILPC<"lgfrl", 0xC4C, aligned_asextloadi32, GR64>;
448 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in
449 def LTGF : UnaryRXY<"ltgf", 0xE332, asextloadi32, GR64, 4>;
451 //===----------------------------------------------------------------------===//
453 //===----------------------------------------------------------------------===//
455 // 32-bit extensions from registers.
456 let hasSideEffects = 0 in {
457 // Expands to LLCR or RISB[LH]G, depending on the choice of registers.
458 def LLCRMux : UnaryRRPseudo<"llc", zext8, GRX32, GRX32>,
459 Requires<[FeatureHighWord]>;
460 def LLCR : UnaryRRE<"llc", 0xB994, zext8, GR32, GR32>;
461 // Expands to LLHR or RISB[LH]G, depending on the choice of registers.
462 def LLHRMux : UnaryRRPseudo<"llh", zext16, GRX32, GRX32>,
463 Requires<[FeatureHighWord]>;
464 def LLHR : UnaryRRE<"llh", 0xB995, zext16, GR32, GR32>;
467 // 64-bit extensions from registers.
468 let hasSideEffects = 0 in {
469 def LLGCR : UnaryRRE<"llgc", 0xB984, zext8, GR64, GR64>;
470 def LLGHR : UnaryRRE<"llgh", 0xB985, zext16, GR64, GR64>;
471 def LLGFR : UnaryRRE<"llgf", 0xB916, zext32, GR64, GR32>;
474 // Match 32-to-64-bit zero extensions in which the source is already
475 // in a 64-bit register.
476 def : Pat<(and GR64:$src, 0xffffffff),
477 (LLGFR (EXTRACT_SUBREG GR64:$src, subreg_l32))>;
479 // 32-bit extensions from 8-bit memory. LLCMux expands to LLC or LLCH,
480 // depending on the choice of register.
481 def LLCMux : UnaryRXYPseudo<"llc", azextloadi8, GRX32, 1>,
482 Requires<[FeatureHighWord]>;
483 def LLC : UnaryRXY<"llc", 0xE394, azextloadi8, GR32, 1>;
484 def LLCH : UnaryRXY<"llch", 0xE3C2, azextloadi8, GR32, 1>,
485 Requires<[FeatureHighWord]>;
487 // 32-bit extensions from 16-bit memory. LLHMux expands to LLH or LLHH,
488 // depending on the choice of register.
489 def LLHMux : UnaryRXYPseudo<"llh", azextloadi16, GRX32, 2>,
490 Requires<[FeatureHighWord]>;
491 def LLH : UnaryRXY<"llh", 0xE395, azextloadi16, GR32, 2>;
492 def LLHH : UnaryRXY<"llhh", 0xE3C6, azextloadi16, GR32, 2>,
493 Requires<[FeatureHighWord]>;
494 def LLHRL : UnaryRILPC<"llhrl", 0xC42, aligned_azextloadi16, GR32>;
496 // 64-bit extensions from memory.
497 def LLGC : UnaryRXY<"llgc", 0xE390, azextloadi8, GR64, 1>;
498 def LLGH : UnaryRXY<"llgh", 0xE391, azextloadi16, GR64, 2>;
499 def LLGF : UnaryRXY<"llgf", 0xE316, azextloadi32, GR64, 4>;
500 def LLGHRL : UnaryRILPC<"llghrl", 0xC46, aligned_azextloadi16, GR64>;
501 def LLGFRL : UnaryRILPC<"llgfrl", 0xC4E, aligned_azextloadi32, GR64>;
503 //===----------------------------------------------------------------------===//
505 //===----------------------------------------------------------------------===//
507 // Truncations of 64-bit registers to 32-bit registers.
508 def : Pat<(i32 (trunc GR64:$src)),
509 (EXTRACT_SUBREG GR64:$src, subreg_l32)>;
511 // Truncations of 32-bit registers to 8-bit memory. STCMux expands to
512 // STC, STCY or STCH, depending on the choice of register.
513 def STCMux : StoreRXYPseudo<truncstorei8, GRX32, 1>,
514 Requires<[FeatureHighWord]>;
515 defm STC : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR32, 1>;
516 def STCH : StoreRXY<"stch", 0xE3C3, truncstorei8, GRH32, 1>,
517 Requires<[FeatureHighWord]>;
519 // Truncations of 32-bit registers to 16-bit memory. STHMux expands to
520 // STH, STHY or STHH, depending on the choice of register.
521 def STHMux : StoreRXYPseudo<truncstorei16, GRX32, 1>,
522 Requires<[FeatureHighWord]>;
523 defm STH : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR32, 2>;
524 def STHH : StoreRXY<"sthh", 0xE3C7, truncstorei16, GRH32, 2>,
525 Requires<[FeatureHighWord]>;
526 def STHRL : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR32>;
528 // Truncations of 64-bit registers to memory.
529 defm : StoreGR64Pair<STC, STCY, truncstorei8>;
530 defm : StoreGR64Pair<STH, STHY, truncstorei16>;
531 def : StoreGR64PC<STHRL, aligned_truncstorei16>;
532 defm : StoreGR64Pair<ST, STY, truncstorei32>;
533 def : StoreGR64PC<STRL, aligned_truncstorei32>;
535 //===----------------------------------------------------------------------===//
536 // Multi-register moves
537 //===----------------------------------------------------------------------===//
539 // Multi-register loads.
540 def LMG : LoadMultipleRSY<"lmg", 0xEB04, GR64>;
542 // Multi-register stores.
543 def STMG : StoreMultipleRSY<"stmg", 0xEB24, GR64>;
545 //===----------------------------------------------------------------------===//
547 //===----------------------------------------------------------------------===//
549 // Byte-swapping register moves.
550 let hasSideEffects = 0 in {
551 def LRVR : UnaryRRE<"lrv", 0xB91F, bswap, GR32, GR32>;
552 def LRVGR : UnaryRRE<"lrvg", 0xB90F, bswap, GR64, GR64>;
555 // Byte-swapping loads. Unlike normal loads, these instructions are
556 // allowed to access storage more than once.
557 def LRV : UnaryRXY<"lrv", 0xE31E, loadu<bswap, nonvolatile_load>, GR32, 4>;
558 def LRVG : UnaryRXY<"lrvg", 0xE30F, loadu<bswap, nonvolatile_load>, GR64, 8>;
560 // Likewise byte-swapping stores.
561 def STRV : StoreRXY<"strv", 0xE33E, storeu<bswap, nonvolatile_store>, GR32, 4>;
562 def STRVG : StoreRXY<"strvg", 0xE32F, storeu<bswap, nonvolatile_store>,
565 //===----------------------------------------------------------------------===//
566 // Load address instructions
567 //===----------------------------------------------------------------------===//
569 // Load BDX-style addresses.
570 let hasSideEffects = 0, isAsCheapAsAMove = 1, isReMaterializable = 1,
572 let DispSize = "12" in
573 def LA : InstRX<0x41, (outs GR64:$R1), (ins laaddr12pair:$XBD2),
575 [(set GR64:$R1, laaddr12pair:$XBD2)]>;
576 let DispSize = "20" in
577 def LAY : InstRXY<0xE371, (outs GR64:$R1), (ins laaddr20pair:$XBD2),
579 [(set GR64:$R1, laaddr20pair:$XBD2)]>;
582 // Load a PC-relative address. There's no version of this instruction
583 // with a 16-bit offset, so there's no relaxation.
584 let hasSideEffects = 0, isAsCheapAsAMove = 1, isMoveImm = 1,
585 isReMaterializable = 1 in {
586 def LARL : InstRIL<0xC00, (outs GR64:$R1), (ins pcrel32:$I2),
588 [(set GR64:$R1, pcrel32:$I2)]>;
591 //===----------------------------------------------------------------------===//
592 // Absolute and Negation
593 //===----------------------------------------------------------------------===//
596 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
597 def LPR : UnaryRR <"lp", 0x10, z_iabs, GR32, GR32>;
598 def LPGR : UnaryRRE<"lpg", 0xB900, z_iabs, GR64, GR64>;
600 let CCValues = 0xE, CompareZeroCCMask = 0xE in
601 def LPGFR : UnaryRRE<"lpgf", 0xB910, null_frag, GR64, GR32>;
603 def : Pat<(z_iabs32 GR32:$src), (LPR GR32:$src)>;
604 def : Pat<(z_iabs64 GR64:$src), (LPGR GR64:$src)>;
605 defm : SXU<z_iabs, LPGFR>;
606 defm : SXU<z_iabs64, LPGFR>;
609 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
610 def LNR : UnaryRR <"ln", 0x11, z_inegabs, GR32, GR32>;
611 def LNGR : UnaryRRE<"lng", 0xB901, z_inegabs, GR64, GR64>;
613 let CCValues = 0xE, CompareZeroCCMask = 0xE in
614 def LNGFR : UnaryRRE<"lngf", 0xB911, null_frag, GR64, GR32>;
616 def : Pat<(z_inegabs32 GR32:$src), (LNR GR32:$src)>;
617 def : Pat<(z_inegabs64 GR64:$src), (LNGR GR64:$src)>;
618 defm : SXU<z_inegabs, LNGFR>;
619 defm : SXU<z_inegabs64, LNGFR>;
622 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
623 def LCR : UnaryRR <"lc", 0x13, ineg, GR32, GR32>;
624 def LCGR : UnaryRRE<"lcg", 0xB903, ineg, GR64, GR64>;
626 let CCValues = 0xE, CompareZeroCCMask = 0xE in
627 def LCGFR : UnaryRRE<"lcgf", 0xB913, null_frag, GR64, GR32>;
629 defm : SXU<ineg, LCGFR>;
631 //===----------------------------------------------------------------------===//
633 //===----------------------------------------------------------------------===//
635 let isCodeGenOnly = 1 in
636 defm IC32 : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR32, azextloadi8, 1>;
637 defm IC : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR64, azextloadi8, 1>;
639 defm : InsertMem<"inserti8", IC32, GR32, azextloadi8, bdxaddr12pair>;
640 defm : InsertMem<"inserti8", IC32Y, GR32, azextloadi8, bdxaddr20pair>;
642 defm : InsertMem<"inserti8", IC, GR64, azextloadi8, bdxaddr12pair>;
643 defm : InsertMem<"inserti8", ICY, GR64, azextloadi8, bdxaddr20pair>;
645 // Insertions of a 16-bit immediate, leaving other bits unaffected.
646 // We don't have or_as_insert equivalents of these operations because
647 // OI is available instead.
649 // IIxMux expands to II[LH]x, depending on the choice of register.
650 def IILMux : BinaryRIPseudo<insertll, GRX32, imm32ll16>,
651 Requires<[FeatureHighWord]>;
652 def IIHMux : BinaryRIPseudo<insertlh, GRX32, imm32lh16>,
653 Requires<[FeatureHighWord]>;
654 def IILL : BinaryRI<"iill", 0xA53, insertll, GR32, imm32ll16>;
655 def IILH : BinaryRI<"iilh", 0xA52, insertlh, GR32, imm32lh16>;
656 def IIHL : BinaryRI<"iihl", 0xA51, insertll, GRH32, imm32ll16>;
657 def IIHH : BinaryRI<"iihh", 0xA50, insertlh, GRH32, imm32lh16>;
658 def IILL64 : BinaryAliasRI<insertll, GR64, imm64ll16>;
659 def IILH64 : BinaryAliasRI<insertlh, GR64, imm64lh16>;
660 def IIHL64 : BinaryAliasRI<inserthl, GR64, imm64hl16>;
661 def IIHH64 : BinaryAliasRI<inserthh, GR64, imm64hh16>;
663 // ...likewise for 32-bit immediates. For GR32s this is a general
664 // full-width move. (We use IILF rather than something like LLILF
665 // for 32-bit moves because IILF leaves the upper 32 bits of the
667 let isAsCheapAsAMove = 1, isMoveImm = 1, isReMaterializable = 1 in {
668 def IIFMux : UnaryRIPseudo<bitconvert, GRX32, uimm32>,
669 Requires<[FeatureHighWord]>;
670 def IILF : UnaryRIL<"iilf", 0xC09, bitconvert, GR32, uimm32>;
671 def IIHF : UnaryRIL<"iihf", 0xC08, bitconvert, GRH32, uimm32>;
673 def IILF64 : BinaryAliasRIL<insertlf, GR64, imm64lf32>;
674 def IIHF64 : BinaryAliasRIL<inserthf, GR64, imm64hf32>;
676 // An alternative model of inserthf, with the first operand being
677 // a zero-extended value.
678 def : Pat<(or (zext32 GR32:$src), imm64hf32:$imm),
679 (IIHF64 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_l32),
682 //===----------------------------------------------------------------------===//
684 //===----------------------------------------------------------------------===//
687 let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in {
688 // Addition of a register.
689 let isCommutable = 1 in {
690 defm AR : BinaryRRAndK<"a", 0x1A, 0xB9F8, add, GR32, GR32>;
691 defm AGR : BinaryRREAndK<"ag", 0xB908, 0xB9E8, add, GR64, GR64>;
693 def AGFR : BinaryRRE<"agf", 0xB918, null_frag, GR64, GR32>;
695 // Addition of signed 16-bit immediates.
696 defm AHIMux : BinaryRIAndKPseudo<"ahimux", add, GRX32, imm32sx16>;
697 defm AHI : BinaryRIAndK<"ahi", 0xA7A, 0xECD8, add, GR32, imm32sx16>;
698 defm AGHI : BinaryRIAndK<"aghi", 0xA7B, 0xECD9, add, GR64, imm64sx16>;
700 // Addition of signed 32-bit immediates.
701 def AFIMux : BinaryRIPseudo<add, GRX32, simm32>,
702 Requires<[FeatureHighWord]>;
703 def AFI : BinaryRIL<"afi", 0xC29, add, GR32, simm32>;
704 def AIH : BinaryRIL<"aih", 0xCC8, add, GRH32, simm32>,
705 Requires<[FeatureHighWord]>;
706 def AGFI : BinaryRIL<"agfi", 0xC28, add, GR64, imm64sx32>;
708 // Addition of memory.
709 defm AH : BinaryRXPair<"ah", 0x4A, 0xE37A, add, GR32, asextloadi16, 2>;
710 defm A : BinaryRXPair<"a", 0x5A, 0xE35A, add, GR32, load, 4>;
711 def AGF : BinaryRXY<"agf", 0xE318, add, GR64, asextloadi32, 4>;
712 def AG : BinaryRXY<"ag", 0xE308, add, GR64, load, 8>;
714 // Addition to memory.
715 def ASI : BinarySIY<"asi", 0xEB6A, add, imm32sx8>;
716 def AGSI : BinarySIY<"agsi", 0xEB7A, add, imm64sx8>;
718 defm : SXB<add, GR64, AGFR>;
720 // Addition producing a carry.
722 // Addition of a register.
723 let isCommutable = 1 in {
724 defm ALR : BinaryRRAndK<"al", 0x1E, 0xB9FA, addc, GR32, GR32>;
725 defm ALGR : BinaryRREAndK<"alg", 0xB90A, 0xB9EA, addc, GR64, GR64>;
727 def ALGFR : BinaryRRE<"algf", 0xB91A, null_frag, GR64, GR32>;
729 // Addition of signed 16-bit immediates.
730 def ALHSIK : BinaryRIE<"alhsik", 0xECDA, addc, GR32, imm32sx16>,
731 Requires<[FeatureDistinctOps]>;
732 def ALGHSIK : BinaryRIE<"alghsik", 0xECDB, addc, GR64, imm64sx16>,
733 Requires<[FeatureDistinctOps]>;
735 // Addition of unsigned 32-bit immediates.
736 def ALFI : BinaryRIL<"alfi", 0xC2B, addc, GR32, uimm32>;
737 def ALGFI : BinaryRIL<"algfi", 0xC2A, addc, GR64, imm64zx32>;
739 // Addition of memory.
740 defm AL : BinaryRXPair<"al", 0x5E, 0xE35E, addc, GR32, load, 4>;
741 def ALGF : BinaryRXY<"algf", 0xE31A, addc, GR64, azextloadi32, 4>;
742 def ALG : BinaryRXY<"alg", 0xE30A, addc, GR64, load, 8>;
744 defm : ZXB<addc, GR64, ALGFR>;
746 // Addition producing and using a carry.
747 let Defs = [CC], Uses = [CC] in {
748 // Addition of a register.
749 def ALCR : BinaryRRE<"alc", 0xB998, adde, GR32, GR32>;
750 def ALCGR : BinaryRRE<"alcg", 0xB988, adde, GR64, GR64>;
752 // Addition of memory.
753 def ALC : BinaryRXY<"alc", 0xE398, adde, GR32, load, 4>;
754 def ALCG : BinaryRXY<"alcg", 0xE388, adde, GR64, load, 8>;
757 //===----------------------------------------------------------------------===//
759 //===----------------------------------------------------------------------===//
761 // Plain subtraction. Although immediate forms exist, we use the
762 // add-immediate instruction instead.
763 let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in {
764 // Subtraction of a register.
765 defm SR : BinaryRRAndK<"s", 0x1B, 0xB9F9, sub, GR32, GR32>;
766 def SGFR : BinaryRRE<"sgf", 0xB919, null_frag, GR64, GR32>;
767 defm SGR : BinaryRREAndK<"sg", 0xB909, 0xB9E9, sub, GR64, GR64>;
769 // Subtraction of memory.
770 defm SH : BinaryRXPair<"sh", 0x4B, 0xE37B, sub, GR32, asextloadi16, 2>;
771 defm S : BinaryRXPair<"s", 0x5B, 0xE35B, sub, GR32, load, 4>;
772 def SGF : BinaryRXY<"sgf", 0xE319, sub, GR64, asextloadi32, 4>;
773 def SG : BinaryRXY<"sg", 0xE309, sub, GR64, load, 8>;
775 defm : SXB<sub, GR64, SGFR>;
777 // Subtraction producing a carry.
779 // Subtraction of a register.
780 defm SLR : BinaryRRAndK<"sl", 0x1F, 0xB9FB, subc, GR32, GR32>;
781 def SLGFR : BinaryRRE<"slgf", 0xB91B, null_frag, GR64, GR32>;
782 defm SLGR : BinaryRREAndK<"slg", 0xB90B, 0xB9EB, subc, GR64, GR64>;
784 // Subtraction of unsigned 32-bit immediates. These don't match
785 // subc because we prefer addc for constants.
786 def SLFI : BinaryRIL<"slfi", 0xC25, null_frag, GR32, uimm32>;
787 def SLGFI : BinaryRIL<"slgfi", 0xC24, null_frag, GR64, imm64zx32>;
789 // Subtraction of memory.
790 defm SL : BinaryRXPair<"sl", 0x5F, 0xE35F, subc, GR32, load, 4>;
791 def SLGF : BinaryRXY<"slgf", 0xE31B, subc, GR64, azextloadi32, 4>;
792 def SLG : BinaryRXY<"slg", 0xE30B, subc, GR64, load, 8>;
794 defm : ZXB<subc, GR64, SLGFR>;
796 // Subtraction producing and using a carry.
797 let Defs = [CC], Uses = [CC] in {
798 // Subtraction of a register.
799 def SLBR : BinaryRRE<"slb", 0xB999, sube, GR32, GR32>;
800 def SLGBR : BinaryRRE<"slbg", 0xB989, sube, GR64, GR64>;
802 // Subtraction of memory.
803 def SLB : BinaryRXY<"slb", 0xE399, sube, GR32, load, 4>;
804 def SLBG : BinaryRXY<"slbg", 0xE389, sube, GR64, load, 8>;
807 //===----------------------------------------------------------------------===//
809 //===----------------------------------------------------------------------===//
812 // ANDs of a register.
813 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
814 defm NR : BinaryRRAndK<"n", 0x14, 0xB9F4, and, GR32, GR32>;
815 defm NGR : BinaryRREAndK<"ng", 0xB980, 0xB9E4, and, GR64, GR64>;
818 let isConvertibleToThreeAddress = 1 in {
819 // ANDs of a 16-bit immediate, leaving other bits unaffected.
820 // The CC result only reflects the 16-bit field, not the full register.
822 // NIxMux expands to NI[LH]x, depending on the choice of register.
823 def NILMux : BinaryRIPseudo<and, GRX32, imm32ll16c>,
824 Requires<[FeatureHighWord]>;
825 def NIHMux : BinaryRIPseudo<and, GRX32, imm32lh16c>,
826 Requires<[FeatureHighWord]>;
827 def NILL : BinaryRI<"nill", 0xA57, and, GR32, imm32ll16c>;
828 def NILH : BinaryRI<"nilh", 0xA56, and, GR32, imm32lh16c>;
829 def NIHL : BinaryRI<"nihl", 0xA55, and, GRH32, imm32ll16c>;
830 def NIHH : BinaryRI<"nihh", 0xA54, and, GRH32, imm32lh16c>;
831 def NILL64 : BinaryAliasRI<and, GR64, imm64ll16c>;
832 def NILH64 : BinaryAliasRI<and, GR64, imm64lh16c>;
833 def NIHL64 : BinaryAliasRI<and, GR64, imm64hl16c>;
834 def NIHH64 : BinaryAliasRI<and, GR64, imm64hh16c>;
836 // ANDs of a 32-bit immediate, leaving other bits unaffected.
837 // The CC result only reflects the 32-bit field, which means we can
838 // use it as a zero indicator for i32 operations but not otherwise.
839 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
840 // Expands to NILF or NIHF, depending on the choice of register.
841 def NIFMux : BinaryRIPseudo<and, GRX32, uimm32>,
842 Requires<[FeatureHighWord]>;
843 def NILF : BinaryRIL<"nilf", 0xC0B, and, GR32, uimm32>;
844 def NIHF : BinaryRIL<"nihf", 0xC0A, and, GRH32, uimm32>;
846 def NILF64 : BinaryAliasRIL<and, GR64, imm64lf32c>;
847 def NIHF64 : BinaryAliasRIL<and, GR64, imm64hf32c>;
851 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
852 defm N : BinaryRXPair<"n", 0x54, 0xE354, and, GR32, load, 4>;
853 def NG : BinaryRXY<"ng", 0xE380, and, GR64, load, 8>;
857 defm NI : BinarySIPair<"ni", 0x94, 0xEB54, null_frag, imm32zx8>;
860 let mayLoad = 1, mayStore = 1 in
861 defm NC : MemorySS<"nc", 0xD4, z_nc, z_nc_loop>;
863 defm : RMWIByte<and, bdaddr12pair, NI>;
864 defm : RMWIByte<and, bdaddr20pair, NIY>;
866 //===----------------------------------------------------------------------===//
868 //===----------------------------------------------------------------------===//
871 // ORs of a register.
872 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
873 defm OR : BinaryRRAndK<"o", 0x16, 0xB9F6, or, GR32, GR32>;
874 defm OGR : BinaryRREAndK<"og", 0xB981, 0xB9E6, or, GR64, GR64>;
877 // ORs of a 16-bit immediate, leaving other bits unaffected.
878 // The CC result only reflects the 16-bit field, not the full register.
880 // OIxMux expands to OI[LH]x, depending on the choice of register.
881 def OILMux : BinaryRIPseudo<or, GRX32, imm32ll16>,
882 Requires<[FeatureHighWord]>;
883 def OIHMux : BinaryRIPseudo<or, GRX32, imm32lh16>,
884 Requires<[FeatureHighWord]>;
885 def OILL : BinaryRI<"oill", 0xA5B, or, GR32, imm32ll16>;
886 def OILH : BinaryRI<"oilh", 0xA5A, or, GR32, imm32lh16>;
887 def OIHL : BinaryRI<"oihl", 0xA59, or, GRH32, imm32ll16>;
888 def OIHH : BinaryRI<"oihh", 0xA58, or, GRH32, imm32lh16>;
889 def OILL64 : BinaryAliasRI<or, GR64, imm64ll16>;
890 def OILH64 : BinaryAliasRI<or, GR64, imm64lh16>;
891 def OIHL64 : BinaryAliasRI<or, GR64, imm64hl16>;
892 def OIHH64 : BinaryAliasRI<or, GR64, imm64hh16>;
894 // ORs of a 32-bit immediate, leaving other bits unaffected.
895 // The CC result only reflects the 32-bit field, which means we can
896 // use it as a zero indicator for i32 operations but not otherwise.
897 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
898 // Expands to OILF or OIHF, depending on the choice of register.
899 def OIFMux : BinaryRIPseudo<or, GRX32, uimm32>,
900 Requires<[FeatureHighWord]>;
901 def OILF : BinaryRIL<"oilf", 0xC0D, or, GR32, uimm32>;
902 def OIHF : BinaryRIL<"oihf", 0xC0C, or, GRH32, uimm32>;
904 def OILF64 : BinaryAliasRIL<or, GR64, imm64lf32>;
905 def OIHF64 : BinaryAliasRIL<or, GR64, imm64hf32>;
908 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
909 defm O : BinaryRXPair<"o", 0x56, 0xE356, or, GR32, load, 4>;
910 def OG : BinaryRXY<"og", 0xE381, or, GR64, load, 8>;
914 defm OI : BinarySIPair<"oi", 0x96, 0xEB56, null_frag, imm32zx8>;
917 let mayLoad = 1, mayStore = 1 in
918 defm OC : MemorySS<"oc", 0xD6, z_oc, z_oc_loop>;
920 defm : RMWIByte<or, bdaddr12pair, OI>;
921 defm : RMWIByte<or, bdaddr20pair, OIY>;
923 //===----------------------------------------------------------------------===//
925 //===----------------------------------------------------------------------===//
928 // XORs of a register.
929 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
930 defm XR : BinaryRRAndK<"x", 0x17, 0xB9F7, xor, GR32, GR32>;
931 defm XGR : BinaryRREAndK<"xg", 0xB982, 0xB9E7, xor, GR64, GR64>;
934 // XORs of a 32-bit immediate, leaving other bits unaffected.
935 // The CC result only reflects the 32-bit field, which means we can
936 // use it as a zero indicator for i32 operations but not otherwise.
937 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
938 // Expands to XILF or XIHF, depending on the choice of register.
939 def XIFMux : BinaryRIPseudo<xor, GRX32, uimm32>,
940 Requires<[FeatureHighWord]>;
941 def XILF : BinaryRIL<"xilf", 0xC07, xor, GR32, uimm32>;
942 def XIHF : BinaryRIL<"xihf", 0xC06, xor, GRH32, uimm32>;
944 def XILF64 : BinaryAliasRIL<xor, GR64, imm64lf32>;
945 def XIHF64 : BinaryAliasRIL<xor, GR64, imm64hf32>;
948 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
949 defm X : BinaryRXPair<"x",0x57, 0xE357, xor, GR32, load, 4>;
950 def XG : BinaryRXY<"xg", 0xE382, xor, GR64, load, 8>;
954 defm XI : BinarySIPair<"xi", 0x97, 0xEB57, null_frag, imm32zx8>;
957 let mayLoad = 1, mayStore = 1 in
958 defm XC : MemorySS<"xc", 0xD7, z_xc, z_xc_loop>;
960 defm : RMWIByte<xor, bdaddr12pair, XI>;
961 defm : RMWIByte<xor, bdaddr20pair, XIY>;
963 //===----------------------------------------------------------------------===//
965 //===----------------------------------------------------------------------===//
967 // Multiplication of a register.
968 let isCommutable = 1 in {
969 def MSR : BinaryRRE<"ms", 0xB252, mul, GR32, GR32>;
970 def MSGR : BinaryRRE<"msg", 0xB90C, mul, GR64, GR64>;
972 def MSGFR : BinaryRRE<"msgf", 0xB91C, null_frag, GR64, GR32>;
973 defm : SXB<mul, GR64, MSGFR>;
975 // Multiplication of a signed 16-bit immediate.
976 def MHI : BinaryRI<"mhi", 0xA7C, mul, GR32, imm32sx16>;
977 def MGHI : BinaryRI<"mghi", 0xA7D, mul, GR64, imm64sx16>;
979 // Multiplication of a signed 32-bit immediate.
980 def MSFI : BinaryRIL<"msfi", 0xC21, mul, GR32, simm32>;
981 def MSGFI : BinaryRIL<"msgfi", 0xC20, mul, GR64, imm64sx32>;
983 // Multiplication of memory.
984 defm MH : BinaryRXPair<"mh", 0x4C, 0xE37C, mul, GR32, asextloadi16, 2>;
985 defm MS : BinaryRXPair<"ms", 0x71, 0xE351, mul, GR32, load, 4>;
986 def MSGF : BinaryRXY<"msgf", 0xE31C, mul, GR64, asextloadi32, 4>;
987 def MSG : BinaryRXY<"msg", 0xE30C, mul, GR64, load, 8>;
989 // Multiplication of a register, producing two results.
990 def MLGR : BinaryRRE<"mlg", 0xB986, z_umul_lohi64, GR128, GR64>;
992 // Multiplication of memory, producing two results.
993 def MLG : BinaryRXY<"mlg", 0xE386, z_umul_lohi64, GR128, load, 8>;
995 //===----------------------------------------------------------------------===//
996 // Division and remainder
997 //===----------------------------------------------------------------------===//
999 // Division and remainder, from registers.
1000 def DSGFR : BinaryRRE<"dsgf", 0xB91D, z_sdivrem32, GR128, GR32>;
1001 def DSGR : BinaryRRE<"dsg", 0xB90D, z_sdivrem64, GR128, GR64>;
1002 def DLR : BinaryRRE<"dl", 0xB997, z_udivrem32, GR128, GR32>;
1003 def DLGR : BinaryRRE<"dlg", 0xB987, z_udivrem64, GR128, GR64>;
1005 // Division and remainder, from memory.
1006 def DSGF : BinaryRXY<"dsgf", 0xE31D, z_sdivrem32, GR128, load, 4>;
1007 def DSG : BinaryRXY<"dsg", 0xE30D, z_sdivrem64, GR128, load, 8>;
1008 def DL : BinaryRXY<"dl", 0xE397, z_udivrem32, GR128, load, 4>;
1009 def DLG : BinaryRXY<"dlg", 0xE387, z_udivrem64, GR128, load, 8>;
1011 //===----------------------------------------------------------------------===//
1013 //===----------------------------------------------------------------------===//
1016 let hasSideEffects = 0 in {
1017 defm SLL : BinaryRSAndK<"sll", 0x89, 0xEBDF, shl, GR32>;
1018 def SLLG : BinaryRSY<"sllg", 0xEB0D, shl, GR64>;
1021 // Logical shift right.
1022 let hasSideEffects = 0 in {
1023 defm SRL : BinaryRSAndK<"srl", 0x88, 0xEBDE, srl, GR32>;
1024 def SRLG : BinaryRSY<"srlg", 0xEB0C, srl, GR64>;
1027 // Arithmetic shift right.
1028 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
1029 defm SRA : BinaryRSAndK<"sra", 0x8A, 0xEBDC, sra, GR32>;
1030 def SRAG : BinaryRSY<"srag", 0xEB0A, sra, GR64>;
1034 let hasSideEffects = 0 in {
1035 def RLL : BinaryRSY<"rll", 0xEB1D, rotl, GR32>;
1036 def RLLG : BinaryRSY<"rllg", 0xEB1C, rotl, GR64>;
1039 // Rotate second operand left and inserted selected bits into first operand.
1040 // These can act like 32-bit operands provided that the constant start and
1041 // end bits (operands 2 and 3) are in the range [32, 64).
1042 let Defs = [CC] in {
1043 let isCodeGenOnly = 1 in
1044 def RISBG32 : RotateSelectRIEf<"risbg", 0xEC55, GR32, GR32>;
1045 let CCValues = 0xE, CompareZeroCCMask = 0xE in
1046 def RISBG : RotateSelectRIEf<"risbg", 0xEC55, GR64, GR64>;
1049 // Forms of RISBG that only affect one word of the destination register.
1050 // They do not set CC.
1051 let Predicates = [FeatureHighWord] in {
1052 def RISBMux : RotateSelectRIEfPseudo<GRX32, GRX32>;
1053 def RISBLL : RotateSelectAliasRIEf<GR32, GR32>;
1054 def RISBLH : RotateSelectAliasRIEf<GR32, GRH32>;
1055 def RISBHL : RotateSelectAliasRIEf<GRH32, GR32>;
1056 def RISBHH : RotateSelectAliasRIEf<GRH32, GRH32>;
1057 def RISBLG : RotateSelectRIEf<"risblg", 0xEC51, GR32, GR64>;
1058 def RISBHG : RotateSelectRIEf<"risbhg", 0xEC5D, GRH32, GR64>;
1061 // Rotate second operand left and perform a logical operation with selected
1062 // bits of the first operand. The CC result only describes the selected bits,
1063 // so isn't useful for a full comparison against zero.
1064 let Defs = [CC] in {
1065 def RNSBG : RotateSelectRIEf<"rnsbg", 0xEC54, GR64, GR64>;
1066 def ROSBG : RotateSelectRIEf<"rosbg", 0xEC56, GR64, GR64>;
1067 def RXSBG : RotateSelectRIEf<"rxsbg", 0xEC57, GR64, GR64>;
1070 //===----------------------------------------------------------------------===//
1072 //===----------------------------------------------------------------------===//
1074 // Signed comparisons. We put these before the unsigned comparisons because
1075 // some of the signed forms have COMPARE AND BRANCH equivalents whereas none
1076 // of the unsigned forms do.
1077 let Defs = [CC], CCValues = 0xE in {
1078 // Comparison with a register.
1079 def CR : CompareRR <"c", 0x19, z_scmp, GR32, GR32>;
1080 def CGFR : CompareRRE<"cgf", 0xB930, null_frag, GR64, GR32>;
1081 def CGR : CompareRRE<"cg", 0xB920, z_scmp, GR64, GR64>;
1083 // Comparison with a signed 16-bit immediate.
1084 def CHI : CompareRI<"chi", 0xA7E, z_scmp, GR32, imm32sx16>;
1085 def CGHI : CompareRI<"cghi", 0xA7F, z_scmp, GR64, imm64sx16>;
1087 // Comparison with a signed 32-bit immediate. CFIMux expands to CFI or CIH,
1088 // depending on the choice of register.
1089 def CFIMux : CompareRIPseudo<z_scmp, GRX32, simm32>,
1090 Requires<[FeatureHighWord]>;
1091 def CFI : CompareRIL<"cfi", 0xC2D, z_scmp, GR32, simm32>;
1092 def CIH : CompareRIL<"cih", 0xCCD, z_scmp, GRH32, simm32>,
1093 Requires<[FeatureHighWord]>;
1094 def CGFI : CompareRIL<"cgfi", 0xC2C, z_scmp, GR64, imm64sx32>;
1096 // Comparison with memory.
1097 defm CH : CompareRXPair<"ch", 0x49, 0xE379, z_scmp, GR32, asextloadi16, 2>;
1098 def CMux : CompareRXYPseudo<z_scmp, GRX32, load, 4>,
1099 Requires<[FeatureHighWord]>;
1100 defm C : CompareRXPair<"c", 0x59, 0xE359, z_scmp, GR32, load, 4>;
1101 def CHF : CompareRXY<"chf", 0xE3CD, z_scmp, GRH32, load, 4>,
1102 Requires<[FeatureHighWord]>;
1103 def CGH : CompareRXY<"cgh", 0xE334, z_scmp, GR64, asextloadi16, 2>;
1104 def CGF : CompareRXY<"cgf", 0xE330, z_scmp, GR64, asextloadi32, 4>;
1105 def CG : CompareRXY<"cg", 0xE320, z_scmp, GR64, load, 8>;
1106 def CHRL : CompareRILPC<"chrl", 0xC65, z_scmp, GR32, aligned_asextloadi16>;
1107 def CRL : CompareRILPC<"crl", 0xC6D, z_scmp, GR32, aligned_load>;
1108 def CGHRL : CompareRILPC<"cghrl", 0xC64, z_scmp, GR64, aligned_asextloadi16>;
1109 def CGFRL : CompareRILPC<"cgfrl", 0xC6C, z_scmp, GR64, aligned_asextloadi32>;
1110 def CGRL : CompareRILPC<"cgrl", 0xC68, z_scmp, GR64, aligned_load>;
1112 // Comparison between memory and a signed 16-bit immediate.
1113 def CHHSI : CompareSIL<"chhsi", 0xE554, z_scmp, asextloadi16, imm32sx16>;
1114 def CHSI : CompareSIL<"chsi", 0xE55C, z_scmp, load, imm32sx16>;
1115 def CGHSI : CompareSIL<"cghsi", 0xE558, z_scmp, load, imm64sx16>;
1117 defm : SXB<z_scmp, GR64, CGFR>;
1119 // Unsigned comparisons.
1120 let Defs = [CC], CCValues = 0xE, IsLogical = 1 in {
1121 // Comparison with a register.
1122 def CLR : CompareRR <"cl", 0x15, z_ucmp, GR32, GR32>;
1123 def CLGFR : CompareRRE<"clgf", 0xB931, null_frag, GR64, GR32>;
1124 def CLGR : CompareRRE<"clg", 0xB921, z_ucmp, GR64, GR64>;
1126 // Comparison with an unsigned 32-bit immediate. CLFIMux expands to CLFI
1127 // or CLIH, depending on the choice of register.
1128 def CLFIMux : CompareRIPseudo<z_ucmp, GRX32, uimm32>,
1129 Requires<[FeatureHighWord]>;
1130 def CLFI : CompareRIL<"clfi", 0xC2F, z_ucmp, GR32, uimm32>;
1131 def CLIH : CompareRIL<"clih", 0xCCF, z_ucmp, GR32, uimm32>,
1132 Requires<[FeatureHighWord]>;
1133 def CLGFI : CompareRIL<"clgfi", 0xC2E, z_ucmp, GR64, imm64zx32>;
1135 // Comparison with memory.
1136 def CLMux : CompareRXYPseudo<z_ucmp, GRX32, load, 4>,
1137 Requires<[FeatureHighWord]>;
1138 defm CL : CompareRXPair<"cl", 0x55, 0xE355, z_ucmp, GR32, load, 4>;
1139 def CLHF : CompareRXY<"clhf", 0xE3CF, z_ucmp, GRH32, load, 4>,
1140 Requires<[FeatureHighWord]>;
1141 def CLGF : CompareRXY<"clgf", 0xE331, z_ucmp, GR64, azextloadi32, 4>;
1142 def CLG : CompareRXY<"clg", 0xE321, z_ucmp, GR64, load, 8>;
1143 def CLHRL : CompareRILPC<"clhrl", 0xC67, z_ucmp, GR32,
1144 aligned_azextloadi16>;
1145 def CLRL : CompareRILPC<"clrl", 0xC6F, z_ucmp, GR32,
1147 def CLGHRL : CompareRILPC<"clghrl", 0xC66, z_ucmp, GR64,
1148 aligned_azextloadi16>;
1149 def CLGFRL : CompareRILPC<"clgfrl", 0xC6E, z_ucmp, GR64,
1150 aligned_azextloadi32>;
1151 def CLGRL : CompareRILPC<"clgrl", 0xC6A, z_ucmp, GR64,
1154 // Comparison between memory and an unsigned 8-bit immediate.
1155 defm CLI : CompareSIPair<"cli", 0x95, 0xEB55, z_ucmp, azextloadi8, imm32zx8>;
1157 // Comparison between memory and an unsigned 16-bit immediate.
1158 def CLHHSI : CompareSIL<"clhhsi", 0xE555, z_ucmp, azextloadi16, imm32zx16>;
1159 def CLFHSI : CompareSIL<"clfhsi", 0xE55D, z_ucmp, load, imm32zx16>;
1160 def CLGHSI : CompareSIL<"clghsi", 0xE559, z_ucmp, load, imm64zx16>;
1162 defm : ZXB<z_ucmp, GR64, CLGFR>;
1164 // Memory-to-memory comparison.
1165 let mayLoad = 1, Defs = [CC] in
1166 defm CLC : MemorySS<"clc", 0xD5, z_clc, z_clc_loop>;
1168 // String comparison.
1169 let mayLoad = 1, Defs = [CC], Uses = [R0L] in
1170 defm CLST : StringRRE<"clst", 0xB25D, z_strcmp>;
1173 let Defs = [CC] in {
1174 // TMxMux expands to TM[LH]x, depending on the choice of register.
1175 def TMLMux : CompareRIPseudo<z_tm_reg, GRX32, imm32ll16>,
1176 Requires<[FeatureHighWord]>;
1177 def TMHMux : CompareRIPseudo<z_tm_reg, GRX32, imm32lh16>,
1178 Requires<[FeatureHighWord]>;
1179 def TMLL : CompareRI<"tmll", 0xA71, z_tm_reg, GR32, imm32ll16>;
1180 def TMLH : CompareRI<"tmlh", 0xA70, z_tm_reg, GR32, imm32lh16>;
1181 def TMHL : CompareRI<"tmhl", 0xA73, z_tm_reg, GRH32, imm32ll16>;
1182 def TMHH : CompareRI<"tmhh", 0xA72, z_tm_reg, GRH32, imm32lh16>;
1184 def TMLL64 : CompareAliasRI<z_tm_reg, GR64, imm64ll16>;
1185 def TMLH64 : CompareAliasRI<z_tm_reg, GR64, imm64lh16>;
1186 def TMHL64 : CompareAliasRI<z_tm_reg, GR64, imm64hl16>;
1187 def TMHH64 : CompareAliasRI<z_tm_reg, GR64, imm64hh16>;
1189 defm TM : CompareSIPair<"tm", 0x91, 0xEB51, z_tm_mem, anyextloadi8, imm32zx8>;
1192 //===----------------------------------------------------------------------===//
1194 //===----------------------------------------------------------------------===//
1196 def PFD : PrefetchRXY<"pfd", 0xE336, z_prefetch>;
1197 def PFDRL : PrefetchRILPC<"pfdrl", 0xC62, z_prefetch>;
1199 //===----------------------------------------------------------------------===//
1200 // Atomic operations
1201 //===----------------------------------------------------------------------===//
1203 def Serialize : Alias<2, (outs), (ins), [(z_serialize)]>;
1205 let Predicates = [FeatureInterlockedAccess1], Defs = [CC] in {
1206 def LAA : LoadAndOpRSY<"laa", 0xEBF8, atomic_load_add_32, GR32>;
1207 def LAAG : LoadAndOpRSY<"laag", 0xEBE8, atomic_load_add_64, GR64>;
1208 def LAAL : LoadAndOpRSY<"laal", 0xEBFA, null_frag, GR32>;
1209 def LAALG : LoadAndOpRSY<"laalg", 0xEBEA, null_frag, GR64>;
1210 def LAN : LoadAndOpRSY<"lan", 0xEBF4, atomic_load_and_32, GR32>;
1211 def LANG : LoadAndOpRSY<"lang", 0xEBE4, atomic_load_and_64, GR64>;
1212 def LAO : LoadAndOpRSY<"lao", 0xEBF6, atomic_load_or_32, GR32>;
1213 def LAOG : LoadAndOpRSY<"laog", 0xEBE6, atomic_load_or_64, GR64>;
1214 def LAX : LoadAndOpRSY<"lax", 0xEBF7, atomic_load_xor_32, GR32>;
1215 def LAXG : LoadAndOpRSY<"laxg", 0xEBE7, atomic_load_xor_64, GR64>;
1218 def ATOMIC_SWAPW : AtomicLoadWBinaryReg<z_atomic_swapw>;
1219 def ATOMIC_SWAP_32 : AtomicLoadBinaryReg32<atomic_swap_32>;
1220 def ATOMIC_SWAP_64 : AtomicLoadBinaryReg64<atomic_swap_64>;
1222 def ATOMIC_LOADW_AR : AtomicLoadWBinaryReg<z_atomic_loadw_add>;
1223 def ATOMIC_LOADW_AFI : AtomicLoadWBinaryImm<z_atomic_loadw_add, simm32>;
1224 let Predicates = [FeatureNoInterlockedAccess1] in {
1225 def ATOMIC_LOAD_AR : AtomicLoadBinaryReg32<atomic_load_add_32>;
1226 def ATOMIC_LOAD_AHI : AtomicLoadBinaryImm32<atomic_load_add_32, imm32sx16>;
1227 def ATOMIC_LOAD_AFI : AtomicLoadBinaryImm32<atomic_load_add_32, simm32>;
1228 def ATOMIC_LOAD_AGR : AtomicLoadBinaryReg64<atomic_load_add_64>;
1229 def ATOMIC_LOAD_AGHI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx16>;
1230 def ATOMIC_LOAD_AGFI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx32>;
1233 def ATOMIC_LOADW_SR : AtomicLoadWBinaryReg<z_atomic_loadw_sub>;
1234 def ATOMIC_LOAD_SR : AtomicLoadBinaryReg32<atomic_load_sub_32>;
1235 def ATOMIC_LOAD_SGR : AtomicLoadBinaryReg64<atomic_load_sub_64>;
1237 def ATOMIC_LOADW_NR : AtomicLoadWBinaryReg<z_atomic_loadw_and>;
1238 def ATOMIC_LOADW_NILH : AtomicLoadWBinaryImm<z_atomic_loadw_and, imm32lh16c>;
1239 let Predicates = [FeatureNoInterlockedAccess1] in {
1240 def ATOMIC_LOAD_NR : AtomicLoadBinaryReg32<atomic_load_and_32>;
1241 def ATOMIC_LOAD_NILL : AtomicLoadBinaryImm32<atomic_load_and_32,
1243 def ATOMIC_LOAD_NILH : AtomicLoadBinaryImm32<atomic_load_and_32,
1245 def ATOMIC_LOAD_NILF : AtomicLoadBinaryImm32<atomic_load_and_32, uimm32>;
1246 def ATOMIC_LOAD_NGR : AtomicLoadBinaryReg64<atomic_load_and_64>;
1247 def ATOMIC_LOAD_NILL64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1249 def ATOMIC_LOAD_NILH64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1251 def ATOMIC_LOAD_NIHL64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1253 def ATOMIC_LOAD_NIHH64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1255 def ATOMIC_LOAD_NILF64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1257 def ATOMIC_LOAD_NIHF64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1261 def ATOMIC_LOADW_OR : AtomicLoadWBinaryReg<z_atomic_loadw_or>;
1262 def ATOMIC_LOADW_OILH : AtomicLoadWBinaryImm<z_atomic_loadw_or, imm32lh16>;
1263 let Predicates = [FeatureNoInterlockedAccess1] in {
1264 def ATOMIC_LOAD_OR : AtomicLoadBinaryReg32<atomic_load_or_32>;
1265 def ATOMIC_LOAD_OILL : AtomicLoadBinaryImm32<atomic_load_or_32, imm32ll16>;
1266 def ATOMIC_LOAD_OILH : AtomicLoadBinaryImm32<atomic_load_or_32, imm32lh16>;
1267 def ATOMIC_LOAD_OILF : AtomicLoadBinaryImm32<atomic_load_or_32, uimm32>;
1268 def ATOMIC_LOAD_OGR : AtomicLoadBinaryReg64<atomic_load_or_64>;
1269 def ATOMIC_LOAD_OILL64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64ll16>;
1270 def ATOMIC_LOAD_OILH64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lh16>;
1271 def ATOMIC_LOAD_OIHL64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hl16>;
1272 def ATOMIC_LOAD_OIHH64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hh16>;
1273 def ATOMIC_LOAD_OILF64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lf32>;
1274 def ATOMIC_LOAD_OIHF64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hf32>;
1277 def ATOMIC_LOADW_XR : AtomicLoadWBinaryReg<z_atomic_loadw_xor>;
1278 def ATOMIC_LOADW_XILF : AtomicLoadWBinaryImm<z_atomic_loadw_xor, uimm32>;
1279 let Predicates = [FeatureNoInterlockedAccess1] in {
1280 def ATOMIC_LOAD_XR : AtomicLoadBinaryReg32<atomic_load_xor_32>;
1281 def ATOMIC_LOAD_XILF : AtomicLoadBinaryImm32<atomic_load_xor_32, uimm32>;
1282 def ATOMIC_LOAD_XGR : AtomicLoadBinaryReg64<atomic_load_xor_64>;
1283 def ATOMIC_LOAD_XILF64 : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64lf32>;
1284 def ATOMIC_LOAD_XIHF64 : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64hf32>;
1287 def ATOMIC_LOADW_NRi : AtomicLoadWBinaryReg<z_atomic_loadw_nand>;
1288 def ATOMIC_LOADW_NILHi : AtomicLoadWBinaryImm<z_atomic_loadw_nand,
1290 def ATOMIC_LOAD_NRi : AtomicLoadBinaryReg32<atomic_load_nand_32>;
1291 def ATOMIC_LOAD_NILLi : AtomicLoadBinaryImm32<atomic_load_nand_32,
1293 def ATOMIC_LOAD_NILHi : AtomicLoadBinaryImm32<atomic_load_nand_32,
1295 def ATOMIC_LOAD_NILFi : AtomicLoadBinaryImm32<atomic_load_nand_32, uimm32>;
1296 def ATOMIC_LOAD_NGRi : AtomicLoadBinaryReg64<atomic_load_nand_64>;
1297 def ATOMIC_LOAD_NILL64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1299 def ATOMIC_LOAD_NILH64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1301 def ATOMIC_LOAD_NIHL64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1303 def ATOMIC_LOAD_NIHH64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1305 def ATOMIC_LOAD_NILF64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1307 def ATOMIC_LOAD_NIHF64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1310 def ATOMIC_LOADW_MIN : AtomicLoadWBinaryReg<z_atomic_loadw_min>;
1311 def ATOMIC_LOAD_MIN_32 : AtomicLoadBinaryReg32<atomic_load_min_32>;
1312 def ATOMIC_LOAD_MIN_64 : AtomicLoadBinaryReg64<atomic_load_min_64>;
1314 def ATOMIC_LOADW_MAX : AtomicLoadWBinaryReg<z_atomic_loadw_max>;
1315 def ATOMIC_LOAD_MAX_32 : AtomicLoadBinaryReg32<atomic_load_max_32>;
1316 def ATOMIC_LOAD_MAX_64 : AtomicLoadBinaryReg64<atomic_load_max_64>;
1318 def ATOMIC_LOADW_UMIN : AtomicLoadWBinaryReg<z_atomic_loadw_umin>;
1319 def ATOMIC_LOAD_UMIN_32 : AtomicLoadBinaryReg32<atomic_load_umin_32>;
1320 def ATOMIC_LOAD_UMIN_64 : AtomicLoadBinaryReg64<atomic_load_umin_64>;
1322 def ATOMIC_LOADW_UMAX : AtomicLoadWBinaryReg<z_atomic_loadw_umax>;
1323 def ATOMIC_LOAD_UMAX_32 : AtomicLoadBinaryReg32<atomic_load_umax_32>;
1324 def ATOMIC_LOAD_UMAX_64 : AtomicLoadBinaryReg64<atomic_load_umax_64>;
1326 def ATOMIC_CMP_SWAPW
1327 : Pseudo<(outs GR32:$dst), (ins bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1328 ADDR32:$bitshift, ADDR32:$negbitshift,
1331 (z_atomic_cmp_swapw bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1332 ADDR32:$bitshift, ADDR32:$negbitshift,
1333 uimm32:$bitsize))]> {
1337 let usesCustomInserter = 1;
1340 let Defs = [CC] in {
1341 defm CS : CmpSwapRSPair<"cs", 0xBA, 0xEB14, atomic_cmp_swap_32, GR32>;
1342 def CSG : CmpSwapRSY<"csg", 0xEB30, atomic_cmp_swap_64, GR64>;
1345 //===----------------------------------------------------------------------===//
1346 // Miscellaneous Instructions.
1347 //===----------------------------------------------------------------------===//
1349 // Extract CC into bits 29 and 28 of a register.
1351 def IPM : InherentRRE<"ipm", 0xB222, GR32, (z_ipm)>;
1353 // Read a 32-bit access register into a GR32. As with all GR32 operations,
1354 // the upper 32 bits of the enclosing GR64 remain unchanged, which is useful
1355 // when a 64-bit address is stored in a pair of access registers.
1356 def EAR : InstRRE<0xB24F, (outs GR32:$R1), (ins access_reg:$R2),
1358 [(set GR32:$R1, (z_extract_access access_reg:$R2))]>;
1360 // Find leftmost one, AKA count leading zeros. The instruction actually
1361 // returns a pair of GR64s, the first giving the number of leading zeros
1362 // and the second giving a copy of the source with the leftmost one bit
1363 // cleared. We only use the first result here.
1364 let Defs = [CC] in {
1365 def FLOGR : UnaryRRE<"flog", 0xB983, null_frag, GR128, GR64>;
1367 def : Pat<(ctlz GR64:$src),
1368 (EXTRACT_SUBREG (FLOGR GR64:$src), subreg_h64)>;
1370 // Use subregs to populate the "don't care" bits in a 32-bit to 64-bit anyext.
1371 def : Pat<(i64 (anyext GR32:$src)),
1372 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_l32)>;
1374 // Extend GR32s and GR64s to GR128s.
1375 let usesCustomInserter = 1 in {
1376 def AEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1377 def ZEXT128_32 : Pseudo<(outs GR128:$dst), (ins GR32:$src), []>;
1378 def ZEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1381 // Search a block of memory for a character.
1382 let mayLoad = 1, Defs = [CC], Uses = [R0L] in
1383 defm SRST : StringRRE<"srst", 0xb25e, z_search_string>;
1385 //===----------------------------------------------------------------------===//
1387 //===----------------------------------------------------------------------===//
1389 // Use AL* for GR64 additions of unsigned 32-bit values.
1390 defm : ZXB<add, GR64, ALGFR>;
1391 def : Pat<(add GR64:$src1, imm64zx32:$src2),
1392 (ALGFI GR64:$src1, imm64zx32:$src2)>;
1393 def : Pat<(add GR64:$src1, (azextloadi32 bdxaddr20only:$addr)),
1394 (ALGF GR64:$src1, bdxaddr20only:$addr)>;
1396 // Use SL* for GR64 subtractions of unsigned 32-bit values.
1397 defm : ZXB<sub, GR64, SLGFR>;
1398 def : Pat<(add GR64:$src1, imm64zx32n:$src2),
1399 (SLGFI GR64:$src1, imm64zx32n:$src2)>;
1400 def : Pat<(sub GR64:$src1, (azextloadi32 bdxaddr20only:$addr)),
1401 (SLGF GR64:$src1, bdxaddr20only:$addr)>;
1403 // Optimize sign-extended 1/0 selects to -1/0 selects. This is important
1404 // for vector legalization.
1405 def : Pat<(sra (shl (i32 (z_select_ccmask 1, 0, imm32zx4:$valid, imm32zx4:$cc)),
1408 (Select32 (LHI -1), (LHI 0), imm32zx4:$valid, imm32zx4:$cc)>;
1409 def : Pat<(sra (shl (i64 (anyext (i32 (z_select_ccmask 1, 0, imm32zx4:$valid,
1413 (Select64 (LGHI -1), (LGHI 0), imm32zx4:$valid, imm32zx4:$cc)>;
1415 // Peepholes for turning scalar operations into block operations.
1416 defm : BlockLoadStore<anyextloadi8, i32, MVCSequence, NCSequence, OCSequence,
1418 defm : BlockLoadStore<anyextloadi16, i32, MVCSequence, NCSequence, OCSequence,
1420 defm : BlockLoadStore<load, i32, MVCSequence, NCSequence, OCSequence,
1422 defm : BlockLoadStore<anyextloadi8, i64, MVCSequence, NCSequence,
1423 OCSequence, XCSequence, 1>;
1424 defm : BlockLoadStore<anyextloadi16, i64, MVCSequence, NCSequence, OCSequence,
1426 defm : BlockLoadStore<anyextloadi32, i64, MVCSequence, NCSequence, OCSequence,
1428 defm : BlockLoadStore<load, i64, MVCSequence, NCSequence, OCSequence,