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 neverHasSideEffects = 1 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. R1 is the condition-code mask (all 1s)
36 // and R2 is the target address, which is always stored in %r14.
37 let isReturn = 1, isTerminator = 1, isBarrier = 1, hasCtrlDep = 1,
38 R1 = 15, R2 = 14, isCodeGenOnly = 1 in {
39 def RET : InstRR<0x07, (outs), (ins), "br\t%r14", [(z_retflag)]>;
42 // Unconditional branches. R1 is the condition-code mask (all 1s).
43 let isBranch = 1, isTerminator = 1, isBarrier = 1, R1 = 15 in {
44 let isIndirectBranch = 1 in
45 def BR : InstRR<0x07, (outs), (ins ADDR64:$R2),
46 "br\t$R2", [(brind ADDR64:$R2)]>;
48 // An assembler extended mnemonic for BRC.
49 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2), "j\t$I2",
52 // An assembler extended mnemonic for BRCL. (The extension is "G"
53 // rather than "L" because "JL" is "Jump if Less".)
54 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2), "jg\t$I2", []>;
57 // Conditional branches. It's easier for LLVM to handle these branches
58 // in their raw BRC/BRCL form, with the 4-bit condition-code mask being
59 // the first operand. It seems friendlier to use mnemonic forms like
60 // JE and JLH when writing out the assembly though.
62 // Using a custom inserter for BRC gives us a chance to convert the BRC
63 // and a preceding compare into a single compare-and-branch instruction.
64 // The inserter makes no change in cases where a separate branch really
66 multiclass CondBranches<Operand ccmask, string short, string long> {
67 let isBranch = 1, isTerminator = 1, Uses = [CC] in {
68 def "" : InstRI<0xA74, (outs), (ins ccmask:$R1, brtarget16:$I2), short, []>;
69 def L : InstRIL<0xC04, (outs), (ins ccmask:$R1, brtarget32:$I2), long, []>;
72 let isCodeGenOnly = 1, usesCustomInserter = 1 in
73 defm BRC : CondBranches<cond4, "j$R1\t$I2", "jg$R1\t$I2">;
74 defm AsmBRC : CondBranches<uimm8zx4, "brc\t$R1, $I2", "brcl\t$R1, $I2">;
76 def : Pat<(z_br_ccmask cond4:$cond, bb:$dst), (BRC cond4:$cond, bb:$dst)>;
78 // Fused compare-and-branch instructions. As for normal branches,
79 // we handle these instructions internally in their raw CRJ-like form,
80 // but use assembly macros like CRJE when writing them out.
82 // These instructions do not use or clobber the condition codes.
83 // We nevertheless pretend that they clobber CC, so that we can lower
84 // them to separate comparisons and BRCLs if the branch ends up being
86 multiclass CompareBranches<Operand ccmask, string pos1, string pos2> {
87 let isBranch = 1, isTerminator = 1, Defs = [CC] in {
88 def RJ : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2, ccmask:$M3,
90 "crj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
91 def GRJ : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2, ccmask:$M3,
93 "cgrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
94 def IJ : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2, ccmask:$M3,
96 "cij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
97 def GIJ : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2, ccmask:$M3,
99 "cgij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
102 let isCodeGenOnly = 1 in
103 defm C : CompareBranches<cond4, "$M3", "">;
104 defm AsmC : CompareBranches<uimm8zx4, "", "$M3, ">;
106 // Define AsmParser mnemonics for each general condition-code mask
107 // (integer or floating-point)
108 multiclass CondExtendedMnemonic<bits<4> ccmask, string name> {
110 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2),
111 "j"##name##"\t$I2", []>;
112 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2),
113 "jg"##name##"\t$I2", []>;
115 def LOCR : FixedCondUnaryRRF<"locr"##name, 0xB9F2, GR32, GR32, ccmask>;
116 def LOCGR : FixedCondUnaryRRF<"locgr"##name, 0xB9E2, GR64, GR64, ccmask>;
117 def LOC : FixedCondUnaryRSY<"loc"##name, 0xEBF2, GR32, ccmask, 4>;
118 def LOCG : FixedCondUnaryRSY<"locg"##name, 0xEBE2, GR64, ccmask, 8>;
119 def STOC : FixedCondStoreRSY<"stoc"##name, 0xEBF3, GR32, ccmask, 4>;
120 def STOCG : FixedCondStoreRSY<"stocg"##name, 0xEBE3, GR64, ccmask, 8>;
122 defm AsmO : CondExtendedMnemonic<1, "o">;
123 defm AsmH : CondExtendedMnemonic<2, "h">;
124 defm AsmNLE : CondExtendedMnemonic<3, "nle">;
125 defm AsmL : CondExtendedMnemonic<4, "l">;
126 defm AsmNHE : CondExtendedMnemonic<5, "nhe">;
127 defm AsmLH : CondExtendedMnemonic<6, "lh">;
128 defm AsmNE : CondExtendedMnemonic<7, "ne">;
129 defm AsmE : CondExtendedMnemonic<8, "e">;
130 defm AsmNLH : CondExtendedMnemonic<9, "nlh">;
131 defm AsmHE : CondExtendedMnemonic<10, "he">;
132 defm AsmNL : CondExtendedMnemonic<11, "nl">;
133 defm AsmLE : CondExtendedMnemonic<12, "le">;
134 defm AsmNH : CondExtendedMnemonic<13, "nh">;
135 defm AsmNO : CondExtendedMnemonic<14, "no">;
137 // Define AsmParser mnemonics for each integer condition-code mask.
138 // This is like the list above, except that condition 3 is not possible
139 // and that the low bit of the mask is therefore always 0. This means
140 // that each condition has two names. Conditions "o" and "no" are not used.
142 // We don't make one of the two names an alias of the other because
143 // we need the custom parsing routines to select the correct register class.
144 multiclass IntCondExtendedMnemonicA<bits<4> ccmask, string name> {
146 def CR : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2,
148 "crj"##name##"\t$R1, $R2, $RI4", []>;
149 def CGR : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2,
151 "cgrj"##name##"\t$R1, $R2, $RI4", []>;
152 def CI : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2,
154 "cij"##name##"\t$R1, $I2, $RI4", []>;
155 def CGI : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2,
157 "cgij"##name##"\t$R1, $I2, $RI4", []>;
160 multiclass IntCondExtendedMnemonic<bits<4> ccmask, string name1, string name2>
161 : IntCondExtendedMnemonicA<ccmask, name1> {
162 let isAsmParserOnly = 1 in
163 defm Alt : IntCondExtendedMnemonicA<ccmask, name2>;
165 defm AsmJH : IntCondExtendedMnemonic<2, "h", "nle">;
166 defm AsmJL : IntCondExtendedMnemonic<4, "l", "nhe">;
167 defm AsmJLH : IntCondExtendedMnemonic<6, "lh", "ne">;
168 defm AsmJE : IntCondExtendedMnemonic<8, "e", "nlh">;
169 defm AsmJHE : IntCondExtendedMnemonic<10, "he", "nl">;
170 defm AsmJLE : IntCondExtendedMnemonic<12, "le", "nh">;
172 //===----------------------------------------------------------------------===//
173 // Select instructions
174 //===----------------------------------------------------------------------===//
176 def Select32 : SelectWrapper<GR32>;
177 def Select64 : SelectWrapper<GR64>;
179 defm CondStore8_32 : CondStores<GR32, nonvolatile_truncstorei8,
180 nonvolatile_anyextloadi8, bdxaddr20only>;
181 defm CondStore16_32 : CondStores<GR32, nonvolatile_truncstorei16,
182 nonvolatile_anyextloadi16, bdxaddr20only>;
183 defm CondStore32_32 : CondStores<GR32, nonvolatile_store,
184 nonvolatile_load, bdxaddr20only>;
186 defm CondStore8 : CondStores<GR64, nonvolatile_truncstorei8,
187 nonvolatile_anyextloadi8, bdxaddr20only>;
188 defm CondStore16 : CondStores<GR64, nonvolatile_truncstorei16,
189 nonvolatile_anyextloadi16, bdxaddr20only>;
190 defm CondStore32 : CondStores<GR64, nonvolatile_truncstorei32,
191 nonvolatile_anyextloadi32, bdxaddr20only>;
192 defm CondStore64 : CondStores<GR64, nonvolatile_store,
193 nonvolatile_load, bdxaddr20only>;
195 //===----------------------------------------------------------------------===//
197 //===----------------------------------------------------------------------===//
199 // The definitions here are for the call-clobbered registers.
200 let isCall = 1, Defs = [R0D, R1D, R2D, R3D, R4D, R5D, R14D,
201 F0D, F1D, F2D, F3D, F4D, F5D, F6D, F7D],
202 R1 = 14, isCodeGenOnly = 1 in {
203 def BRAS : InstRI<0xA75, (outs), (ins pcrel16call:$I2, variable_ops),
204 "bras\t%r14, $I2", []>;
205 def BRASL : InstRIL<0xC05, (outs), (ins pcrel32call:$I2, variable_ops),
206 "brasl\t%r14, $I2", [(z_call pcrel32call:$I2)]>;
207 def BASR : InstRR<0x0D, (outs), (ins ADDR64:$R2, variable_ops),
208 "basr\t%r14, $R2", [(z_call ADDR64:$R2)]>;
211 // Define the general form of the call instructions for the asm parser.
212 // These instructions don't hard-code %r14 as the return address register.
213 def AsmBRAS : InstRI<0xA75, (outs), (ins GR64:$R1, brtarget16:$I2),
214 "bras\t$R1, $I2", []>;
215 def AsmBRASL : InstRIL<0xC05, (outs), (ins GR64:$R1, brtarget32:$I2),
216 "brasl\t$R1, $I2", []>;
217 def AsmBASR : InstRR<0x0D, (outs), (ins GR64:$R1, ADDR64:$R2),
218 "basr\t$R1, $R2", []>;
220 //===----------------------------------------------------------------------===//
222 //===----------------------------------------------------------------------===//
225 let neverHasSideEffects = 1 in {
226 def LR : UnaryRR <"l", 0x18, null_frag, GR32, GR32>;
227 def LGR : UnaryRRE<"lg", 0xB904, null_frag, GR64, GR64>;
230 // Move on condition.
231 let isCodeGenOnly = 1, Uses = [CC] in {
232 def LOCR : CondUnaryRRF<"loc", 0xB9F2, GR32, GR32>;
233 def LOCGR : CondUnaryRRF<"locg", 0xB9E2, GR64, GR64>;
236 def AsmLOCR : AsmCondUnaryRRF<"loc", 0xB9F2, GR32, GR32>;
237 def AsmLOCGR : AsmCondUnaryRRF<"locg", 0xB9E2, GR64, GR64>;
241 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
242 isReMaterializable = 1 in {
243 // 16-bit sign-extended immediates.
244 def LHI : UnaryRI<"lhi", 0xA78, bitconvert, GR32, imm32sx16>;
245 def LGHI : UnaryRI<"lghi", 0xA79, bitconvert, GR64, imm64sx16>;
247 // Other 16-bit immediates.
248 def LLILL : UnaryRI<"llill", 0xA5F, bitconvert, GR64, imm64ll16>;
249 def LLILH : UnaryRI<"llilh", 0xA5E, bitconvert, GR64, imm64lh16>;
250 def LLIHL : UnaryRI<"llihl", 0xA5D, bitconvert, GR64, imm64hl16>;
251 def LLIHH : UnaryRI<"llihh", 0xA5C, bitconvert, GR64, imm64hh16>;
253 // 32-bit immediates.
254 def LGFI : UnaryRIL<"lgfi", 0xC01, bitconvert, GR64, imm64sx32>;
255 def LLILF : UnaryRIL<"llilf", 0xC0F, bitconvert, GR64, imm64lf32>;
256 def LLIHF : UnaryRIL<"llihf", 0xC0E, bitconvert, GR64, imm64hf32>;
260 let canFoldAsLoad = 1, SimpleBDXLoad = 1 in {
261 defm L : UnaryRXPair<"l", 0x58, 0xE358, load, GR32, 4>;
262 def LG : UnaryRXY<"lg", 0xE304, load, GR64, 8>;
264 // These instructions are split after register allocation, so we don't
265 // want a custom inserter.
266 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
267 def L128 : Pseudo<(outs GR128:$dst), (ins bdxaddr20only128:$src),
268 [(set GR128:$dst, (load bdxaddr20only128:$src))]>;
271 let canFoldAsLoad = 1 in {
272 def LRL : UnaryRILPC<"lrl", 0xC4D, aligned_load, GR32>;
273 def LGRL : UnaryRILPC<"lgrl", 0xC48, aligned_load, GR64>;
276 // Load on condition.
277 let isCodeGenOnly = 1, Uses = [CC] in {
278 def LOC : CondUnaryRSY<"loc", 0xEBF2, GR32, 4>;
279 def LOCG : CondUnaryRSY<"locg", 0xEBE2, GR64, 8>;
282 def AsmLOC : AsmCondUnaryRSY<"loc", 0xEBF2, GR32, 4>;
283 def AsmLOCG : AsmCondUnaryRSY<"locg", 0xEBE2, GR64, 8>;
285 defm : CondLoad<LOC, GR32, nonvolatile_load>;
286 defm : CondLoad<LOCG, GR64, nonvolatile_load>;
289 let SimpleBDXStore = 1 in {
290 let isCodeGenOnly = 1 in
291 defm ST32 : StoreRXPair<"st", 0x50, 0xE350, store, GR32, 4>;
292 def STG : StoreRXY<"stg", 0xE324, store, GR64, 8>;
294 // These instructions are split after register allocation, so we don't
295 // want a custom inserter.
296 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
297 def ST128 : Pseudo<(outs), (ins GR128:$src, bdxaddr20only128:$dst),
298 [(store GR128:$src, bdxaddr20only128:$dst)]>;
301 let isCodeGenOnly = 1 in
302 def STRL32 : StoreRILPC<"strl", 0xC4F, aligned_store, GR32>;
303 def STGRL : StoreRILPC<"stgrl", 0xC4B, aligned_store, GR64>;
305 // Store on condition.
306 let isCodeGenOnly = 1, Uses = [CC] in {
307 def STOC32 : CondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
308 def STOC : CondStoreRSY<"stoc", 0xEBF3, GR64, 4>;
309 def STOCG : CondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
312 def AsmSTOC : AsmCondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
313 def AsmSTOCG : AsmCondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
316 // 8-bit immediate stores to 8-bit fields.
317 defm MVI : StoreSIPair<"mvi", 0x92, 0xEB52, truncstorei8, imm32zx8trunc>;
319 // 16-bit immediate stores to 16-, 32- or 64-bit fields.
320 def MVHHI : StoreSIL<"mvhhi", 0xE544, truncstorei16, imm32sx16trunc>;
321 def MVHI : StoreSIL<"mvhi", 0xE54C, store, imm32sx16>;
322 def MVGHI : StoreSIL<"mvghi", 0xE548, store, imm64sx16>;
324 // Memory-to-memory moves.
325 let mayLoad = 1, mayStore = 1 in
326 def MVC : InstSS<0xD2, (outs), (ins bdladdr12onlylen8:$BDL1,
328 "mvc\t$BDL1, $BD2", []>;
330 let mayLoad = 1, mayStore = 1, usesCustomInserter = 1 in
331 def MVCWrapper : Pseudo<(outs), (ins bdaddr12only:$dest, bdaddr12only:$src,
333 [(z_mvc bdaddr12only:$dest, bdaddr12only:$src,
334 imm32len8:$length)]>;
336 defm LoadStore8_32 : MVCLoadStore<anyextloadi8, truncstorei8, i32,
338 defm LoadStore16_32 : MVCLoadStore<anyextloadi16, truncstorei16, i32,
340 defm LoadStore32_32 : MVCLoadStore<load, store, i32, MVCWrapper, 4>;
342 defm LoadStore8 : MVCLoadStore<anyextloadi8, truncstorei8, i64,
344 defm LoadStore16 : MVCLoadStore<anyextloadi16, truncstorei16, i64,
346 defm LoadStore32 : MVCLoadStore<anyextloadi32, truncstorei32, i64,
348 defm LoadStore64 : MVCLoadStore<load, store, i64, MVCWrapper, 8>;
350 //===----------------------------------------------------------------------===//
352 //===----------------------------------------------------------------------===//
354 // 32-bit extensions from registers.
355 let neverHasSideEffects = 1 in {
356 def LBR : UnaryRRE<"lb", 0xB926, sext8, GR32, GR32>;
357 def LHR : UnaryRRE<"lh", 0xB927, sext16, GR32, GR32>;
360 // 64-bit extensions from registers.
361 let neverHasSideEffects = 1 in {
362 def LGBR : UnaryRRE<"lgb", 0xB906, sext8, GR64, GR64>;
363 def LGHR : UnaryRRE<"lgh", 0xB907, sext16, GR64, GR64>;
364 def LGFR : UnaryRRE<"lgf", 0xB914, sext32, GR64, GR32>;
367 // Match 32-to-64-bit sign extensions in which the source is already
368 // in a 64-bit register.
369 def : Pat<(sext_inreg GR64:$src, i32),
370 (LGFR (EXTRACT_SUBREG GR64:$src, subreg_32bit))>;
372 // 32-bit extensions from memory.
373 def LB : UnaryRXY<"lb", 0xE376, sextloadi8, GR32, 1>;
374 defm LH : UnaryRXPair<"lh", 0x48, 0xE378, sextloadi16, GR32, 2>;
375 def LHRL : UnaryRILPC<"lhrl", 0xC45, aligned_sextloadi16, GR32>;
377 // 64-bit extensions from memory.
378 def LGB : UnaryRXY<"lgb", 0xE377, sextloadi8, GR64, 1>;
379 def LGH : UnaryRXY<"lgh", 0xE315, sextloadi16, GR64, 2>;
380 def LGF : UnaryRXY<"lgf", 0xE314, sextloadi32, GR64, 4>;
381 def LGHRL : UnaryRILPC<"lghrl", 0xC44, aligned_sextloadi16, GR64>;
382 def LGFRL : UnaryRILPC<"lgfrl", 0xC4C, aligned_sextloadi32, GR64>;
384 // If the sign of a load-extend operation doesn't matter, use the signed ones.
385 // There's not really much to choose between the sign and zero extensions,
386 // but LH is more compact than LLH for small offsets.
387 def : Pat<(i32 (extloadi8 bdxaddr20only:$src)), (LB bdxaddr20only:$src)>;
388 def : Pat<(i32 (extloadi16 bdxaddr12pair:$src)), (LH bdxaddr12pair:$src)>;
389 def : Pat<(i32 (extloadi16 bdxaddr20pair:$src)), (LHY bdxaddr20pair:$src)>;
391 def : Pat<(i64 (extloadi8 bdxaddr20only:$src)), (LGB bdxaddr20only:$src)>;
392 def : Pat<(i64 (extloadi16 bdxaddr20only:$src)), (LGH bdxaddr20only:$src)>;
393 def : Pat<(i64 (extloadi32 bdxaddr20only:$src)), (LGF bdxaddr20only:$src)>;
395 // We want PC-relative addresses to be tried ahead of BD and BDX addresses.
396 // However, BDXs have two extra operands and are therefore 6 units more
398 let AddedComplexity = 7 in {
399 def : Pat<(i32 (extloadi16 pcrel32:$src)), (LHRL pcrel32:$src)>;
400 def : Pat<(i64 (extloadi16 pcrel32:$src)), (LGHRL pcrel32:$src)>;
403 //===----------------------------------------------------------------------===//
405 //===----------------------------------------------------------------------===//
407 // 32-bit extensions from registers.
408 let neverHasSideEffects = 1 in {
409 def LLCR : UnaryRRE<"llc", 0xB994, zext8, GR32, GR32>;
410 def LLHR : UnaryRRE<"llh", 0xB995, zext16, GR32, GR32>;
413 // 64-bit extensions from registers.
414 let neverHasSideEffects = 1 in {
415 def LLGCR : UnaryRRE<"llgc", 0xB984, zext8, GR64, GR64>;
416 def LLGHR : UnaryRRE<"llgh", 0xB985, zext16, GR64, GR64>;
417 def LLGFR : UnaryRRE<"llgf", 0xB916, zext32, GR64, GR32>;
420 // Match 32-to-64-bit zero extensions in which the source is already
421 // in a 64-bit register.
422 def : Pat<(and GR64:$src, 0xffffffff),
423 (LLGFR (EXTRACT_SUBREG GR64:$src, subreg_32bit))>;
425 // 32-bit extensions from memory.
426 def LLC : UnaryRXY<"llc", 0xE394, zextloadi8, GR32, 1>;
427 def LLH : UnaryRXY<"llh", 0xE395, zextloadi16, GR32, 2>;
428 def LLHRL : UnaryRILPC<"llhrl", 0xC42, aligned_zextloadi16, GR32>;
430 // 64-bit extensions from memory.
431 def LLGC : UnaryRXY<"llgc", 0xE390, zextloadi8, GR64, 1>;
432 def LLGH : UnaryRXY<"llgh", 0xE391, zextloadi16, GR64, 2>;
433 def LLGF : UnaryRXY<"llgf", 0xE316, zextloadi32, GR64, 4>;
434 def LLGHRL : UnaryRILPC<"llghrl", 0xC46, aligned_zextloadi16, GR64>;
435 def LLGFRL : UnaryRILPC<"llgfrl", 0xC4E, aligned_zextloadi32, GR64>;
437 //===----------------------------------------------------------------------===//
439 //===----------------------------------------------------------------------===//
441 // Truncations of 64-bit registers to 32-bit registers.
442 def : Pat<(i32 (trunc GR64:$src)),
443 (EXTRACT_SUBREG GR64:$src, subreg_32bit)>;
445 // Truncations of 32-bit registers to memory.
446 let isCodeGenOnly = 1 in {
447 defm STC32 : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR32, 1>;
448 defm STH32 : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR32, 2>;
449 def STHRL32 : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR32>;
452 // Truncations of 64-bit registers to memory.
453 defm STC : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR64, 1>;
454 defm STH : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR64, 2>;
455 def STHRL : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR64>;
456 defm ST : StoreRXPair<"st", 0x50, 0xE350, truncstorei32, GR64, 4>;
457 def STRL : StoreRILPC<"strl", 0xC4F, aligned_truncstorei32, GR64>;
459 //===----------------------------------------------------------------------===//
460 // Multi-register moves
461 //===----------------------------------------------------------------------===//
463 // Multi-register loads.
464 def LMG : LoadMultipleRSY<"lmg", 0xEB04, GR64>;
466 // Multi-register stores.
467 def STMG : StoreMultipleRSY<"stmg", 0xEB24, GR64>;
469 //===----------------------------------------------------------------------===//
471 //===----------------------------------------------------------------------===//
473 // Byte-swapping register moves.
474 let neverHasSideEffects = 1 in {
475 def LRVR : UnaryRRE<"lrv", 0xB91F, bswap, GR32, GR32>;
476 def LRVGR : UnaryRRE<"lrvg", 0xB90F, bswap, GR64, GR64>;
479 // Byte-swapping loads. Unlike normal loads, these instructions are
480 // allowed to access storage more than once.
481 def LRV : UnaryRXY<"lrv", 0xE31E, loadu<bswap, nonvolatile_load>, GR32, 4>;
482 def LRVG : UnaryRXY<"lrvg", 0xE30F, loadu<bswap, nonvolatile_load>, GR64, 8>;
484 // Likewise byte-swapping stores.
485 def STRV : StoreRXY<"strv", 0xE33E, storeu<bswap, nonvolatile_store>, GR32, 4>;
486 def STRVG : StoreRXY<"strvg", 0xE32F, storeu<bswap, nonvolatile_store>,
489 //===----------------------------------------------------------------------===//
490 // Load address instructions
491 //===----------------------------------------------------------------------===//
493 // Load BDX-style addresses.
494 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isReMaterializable = 1,
496 let DispSize = "12" in
497 def LA : InstRX<0x41, (outs GR64:$R1), (ins laaddr12pair:$XBD2),
499 [(set GR64:$R1, laaddr12pair:$XBD2)]>;
500 let DispSize = "20" in
501 def LAY : InstRXY<0xE371, (outs GR64:$R1), (ins laaddr20pair:$XBD2),
503 [(set GR64:$R1, laaddr20pair:$XBD2)]>;
506 // Load a PC-relative address. There's no version of this instruction
507 // with a 16-bit offset, so there's no relaxation.
508 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
509 isReMaterializable = 1 in {
510 def LARL : InstRIL<0xC00, (outs GR64:$R1), (ins pcrel32:$I2),
512 [(set GR64:$R1, pcrel32:$I2)]>;
515 //===----------------------------------------------------------------------===//
517 //===----------------------------------------------------------------------===//
520 def LCR : UnaryRR <"lc", 0x13, ineg, GR32, GR32>;
521 def LCGR : UnaryRRE<"lcg", 0xB903, ineg, GR64, GR64>;
522 def LCGFR : UnaryRRE<"lcgf", 0xB913, null_frag, GR64, GR32>;
524 defm : SXU<ineg, LCGFR>;
526 //===----------------------------------------------------------------------===//
528 //===----------------------------------------------------------------------===//
530 let isCodeGenOnly = 1 in
531 defm IC32 : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR32, zextloadi8, 1>;
532 defm IC : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR64, zextloadi8, 1>;
534 defm : InsertMem<"inserti8", IC32, GR32, zextloadi8, bdxaddr12pair>;
535 defm : InsertMem<"inserti8", IC32Y, GR32, zextloadi8, bdxaddr20pair>;
537 defm : InsertMem<"inserti8", IC, GR64, zextloadi8, bdxaddr12pair>;
538 defm : InsertMem<"inserti8", ICY, GR64, zextloadi8, bdxaddr20pair>;
540 // Insertions of a 16-bit immediate, leaving other bits unaffected.
541 // We don't have or_as_insert equivalents of these operations because
542 // OI is available instead.
543 let isCodeGenOnly = 1 in {
544 def IILL32 : BinaryRI<"iill", 0xA53, insertll, GR32, imm32ll16>;
545 def IILH32 : BinaryRI<"iilh", 0xA52, insertlh, GR32, imm32lh16>;
547 def IILL : BinaryRI<"iill", 0xA53, insertll, GR64, imm64ll16>;
548 def IILH : BinaryRI<"iilh", 0xA52, insertlh, GR64, imm64lh16>;
549 def IIHL : BinaryRI<"iihl", 0xA51, inserthl, GR64, imm64hl16>;
550 def IIHH : BinaryRI<"iihh", 0xA50, inserthh, GR64, imm64hh16>;
552 // ...likewise for 32-bit immediates. For GR32s this is a general
553 // full-width move. (We use IILF rather than something like LLILF
554 // for 32-bit moves because IILF leaves the upper 32 bits of the
556 let isCodeGenOnly = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
557 isReMaterializable = 1 in {
558 def IILF32 : UnaryRIL<"iilf", 0xC09, bitconvert, GR32, uimm32>;
560 def IILF : BinaryRIL<"iilf", 0xC09, insertlf, GR64, imm64lf32>;
561 def IIHF : BinaryRIL<"iihf", 0xC08, inserthf, GR64, imm64hf32>;
563 // An alternative model of inserthf, with the first operand being
564 // a zero-extended value.
565 def : Pat<(or (zext32 GR32:$src), imm64hf32:$imm),
566 (IIHF (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_32bit),
569 //===----------------------------------------------------------------------===//
571 //===----------------------------------------------------------------------===//
575 // Addition of a register.
576 let isCommutable = 1 in {
577 defm AR : BinaryRRAndK<"a", 0x1A, 0xB9F8, add, GR32, GR32>;
578 defm AGR : BinaryRREAndK<"ag", 0xB908, 0xB9E8, add, GR64, GR64>;
580 def AGFR : BinaryRRE<"agf", 0xB918, null_frag, GR64, GR32>;
582 // Addition of signed 16-bit immediates.
583 defm AHI : BinaryRIAndK<"ahi", 0xA7A, 0xECD8, add, GR32, imm32sx16>;
584 defm AGHI : BinaryRIAndK<"aghi", 0xA7B, 0xECD9, add, GR64, imm64sx16>;
586 // Addition of signed 32-bit immediates.
587 def AFI : BinaryRIL<"afi", 0xC29, add, GR32, simm32>;
588 def AGFI : BinaryRIL<"agfi", 0xC28, add, GR64, imm64sx32>;
590 // Addition of memory.
591 defm AH : BinaryRXPair<"ah", 0x4A, 0xE37A, add, GR32, sextloadi16, 2>;
592 defm A : BinaryRXPair<"a", 0x5A, 0xE35A, add, GR32, load, 4>;
593 def AGF : BinaryRXY<"agf", 0xE318, add, GR64, sextloadi32, 4>;
594 def AG : BinaryRXY<"ag", 0xE308, add, GR64, load, 8>;
596 // Addition to memory.
597 def ASI : BinarySIY<"asi", 0xEB6A, add, imm32sx8>;
598 def AGSI : BinarySIY<"agsi", 0xEB7A, add, imm64sx8>;
600 defm : SXB<add, GR64, AGFR>;
602 // Addition producing a carry.
604 // Addition of a register.
605 let isCommutable = 1 in {
606 defm ALR : BinaryRRAndK<"al", 0x1E, 0xB9FA, addc, GR32, GR32>;
607 defm ALGR : BinaryRREAndK<"alg", 0xB90A, 0xB9EA, addc, GR64, GR64>;
609 def ALGFR : BinaryRRE<"algf", 0xB91A, null_frag, GR64, GR32>;
611 // Addition of signed 16-bit immediates.
612 def ALHSIK : BinaryRIE<"alhsik", 0xECDA, addc, GR32, imm32sx16>,
613 Requires<[FeatureDistinctOps]>;
614 def ALGHSIK : BinaryRIE<"alghsik", 0xECDB, addc, GR64, imm64sx16>,
615 Requires<[FeatureDistinctOps]>;
617 // Addition of unsigned 32-bit immediates.
618 def ALFI : BinaryRIL<"alfi", 0xC2B, addc, GR32, uimm32>;
619 def ALGFI : BinaryRIL<"algfi", 0xC2A, addc, GR64, imm64zx32>;
621 // Addition of memory.
622 defm AL : BinaryRXPair<"al", 0x5E, 0xE35E, addc, GR32, load, 4>;
623 def ALGF : BinaryRXY<"algf", 0xE31A, addc, GR64, zextloadi32, 4>;
624 def ALG : BinaryRXY<"alg", 0xE30A, addc, GR64, load, 8>;
626 defm : ZXB<addc, GR64, ALGFR>;
628 // Addition producing and using a carry.
629 let Defs = [CC], Uses = [CC] in {
630 // Addition of a register.
631 def ALCR : BinaryRRE<"alc", 0xB998, adde, GR32, GR32>;
632 def ALCGR : BinaryRRE<"alcg", 0xB988, adde, GR64, GR64>;
634 // Addition of memory.
635 def ALC : BinaryRXY<"alc", 0xE398, adde, GR32, load, 4>;
636 def ALCG : BinaryRXY<"alcg", 0xE388, adde, GR64, load, 8>;
639 //===----------------------------------------------------------------------===//
641 //===----------------------------------------------------------------------===//
643 // Plain substraction. Although immediate forms exist, we use the
644 // add-immediate instruction instead.
646 // Subtraction of a register.
647 defm SR : BinaryRRAndK<"s", 0x1B, 0xB9F9, sub, GR32, GR32>;
648 def SGFR : BinaryRRE<"sgf", 0xB919, null_frag, GR64, GR32>;
649 defm SGR : BinaryRREAndK<"sg", 0xB909, 0xB9E9, sub, GR64, GR64>;
651 // Subtraction of memory.
652 defm SH : BinaryRXPair<"sh", 0x4B, 0xE37B, sub, GR32, sextloadi16, 2>;
653 defm S : BinaryRXPair<"s", 0x5B, 0xE35B, sub, GR32, load, 4>;
654 def SGF : BinaryRXY<"sgf", 0xE319, sub, GR64, sextloadi32, 4>;
655 def SG : BinaryRXY<"sg", 0xE309, sub, GR64, load, 8>;
657 defm : SXB<sub, GR64, SGFR>;
659 // Subtraction producing a carry.
661 // Subtraction of a register.
662 defm SLR : BinaryRRAndK<"sl", 0x1F, 0xB9FB, subc, GR32, GR32>;
663 def SLGFR : BinaryRRE<"slgf", 0xB91B, null_frag, GR64, GR32>;
664 defm SLGR : BinaryRREAndK<"slg", 0xB90B, 0xB9EB, subc, GR64, GR64>;
666 // Subtraction of unsigned 32-bit immediates. These don't match
667 // subc because we prefer addc for constants.
668 def SLFI : BinaryRIL<"slfi", 0xC25, null_frag, GR32, uimm32>;
669 def SLGFI : BinaryRIL<"slgfi", 0xC24, null_frag, GR64, imm64zx32>;
671 // Subtraction of memory.
672 defm SL : BinaryRXPair<"sl", 0x5F, 0xE35F, subc, GR32, load, 4>;
673 def SLGF : BinaryRXY<"slgf", 0xE31B, subc, GR64, zextloadi32, 4>;
674 def SLG : BinaryRXY<"slg", 0xE30B, subc, GR64, load, 8>;
676 defm : ZXB<subc, GR64, SLGFR>;
678 // Subtraction producing and using a carry.
679 let Defs = [CC], Uses = [CC] in {
680 // Subtraction of a register.
681 def SLBR : BinaryRRE<"slb", 0xB999, sube, GR32, GR32>;
682 def SLGBR : BinaryRRE<"slbg", 0xB989, sube, GR64, GR64>;
684 // Subtraction of memory.
685 def SLB : BinaryRXY<"slb", 0xE399, sube, GR32, load, 4>;
686 def SLBG : BinaryRXY<"slbg", 0xE389, sube, GR64, load, 8>;
689 //===----------------------------------------------------------------------===//
691 //===----------------------------------------------------------------------===//
694 // ANDs of a register.
695 let isCommutable = 1 in {
696 defm NR : BinaryRRAndK<"n", 0x14, 0xB9F4, and, GR32, GR32>;
697 defm NGR : BinaryRREAndK<"ng", 0xB980, 0xB9E4, and, GR64, GR64>;
700 // ANDs of a 16-bit immediate, leaving other bits unaffected.
701 let isCodeGenOnly = 1 in {
702 def NILL32 : BinaryRI<"nill", 0xA57, and, GR32, imm32ll16c>;
703 def NILH32 : BinaryRI<"nilh", 0xA56, and, GR32, imm32lh16c>;
705 def NILL : BinaryRI<"nill", 0xA57, and, GR64, imm64ll16c>;
706 def NILH : BinaryRI<"nilh", 0xA56, and, GR64, imm64lh16c>;
707 def NIHL : BinaryRI<"nihl", 0xA55, and, GR64, imm64hl16c>;
708 def NIHH : BinaryRI<"nihh", 0xA54, and, GR64, imm64hh16c>;
710 // ANDs of a 32-bit immediate, leaving other bits unaffected.
711 let isCodeGenOnly = 1 in
712 def NILF32 : BinaryRIL<"nilf", 0xC0B, and, GR32, uimm32>;
713 def NILF : BinaryRIL<"nilf", 0xC0B, and, GR64, imm64lf32c>;
714 def NIHF : BinaryRIL<"nihf", 0xC0A, and, GR64, imm64hf32c>;
717 defm N : BinaryRXPair<"n", 0x54, 0xE354, and, GR32, load, 4>;
718 def NG : BinaryRXY<"ng", 0xE380, and, GR64, load, 8>;
721 defm NI : BinarySIPair<"ni", 0x94, 0xEB54, null_frag, uimm8>;
723 defm : RMWIByte<and, bdaddr12pair, NI>;
724 defm : RMWIByte<and, bdaddr20pair, NIY>;
726 //===----------------------------------------------------------------------===//
728 //===----------------------------------------------------------------------===//
731 // ORs of a register.
732 let isCommutable = 1 in {
733 defm OR : BinaryRRAndK<"o", 0x16, 0xB9F6, or, GR32, GR32>;
734 defm OGR : BinaryRREAndK<"og", 0xB981, 0xB9E6, or, GR64, GR64>;
737 // ORs of a 16-bit immediate, leaving other bits unaffected.
738 let isCodeGenOnly = 1 in {
739 def OILL32 : BinaryRI<"oill", 0xA5B, or, GR32, imm32ll16>;
740 def OILH32 : BinaryRI<"oilh", 0xA5A, or, GR32, imm32lh16>;
742 def OILL : BinaryRI<"oill", 0xA5B, or, GR64, imm64ll16>;
743 def OILH : BinaryRI<"oilh", 0xA5A, or, GR64, imm64lh16>;
744 def OIHL : BinaryRI<"oihl", 0xA59, or, GR64, imm64hl16>;
745 def OIHH : BinaryRI<"oihh", 0xA58, or, GR64, imm64hh16>;
747 // ORs of a 32-bit immediate, leaving other bits unaffected.
748 let isCodeGenOnly = 1 in
749 def OILF32 : BinaryRIL<"oilf", 0xC0D, or, GR32, uimm32>;
750 def OILF : BinaryRIL<"oilf", 0xC0D, or, GR64, imm64lf32>;
751 def OIHF : BinaryRIL<"oihf", 0xC0C, or, GR64, imm64hf32>;
754 defm O : BinaryRXPair<"o", 0x56, 0xE356, or, GR32, load, 4>;
755 def OG : BinaryRXY<"og", 0xE381, or, GR64, load, 8>;
758 defm OI : BinarySIPair<"oi", 0x96, 0xEB56, null_frag, uimm8>;
760 defm : RMWIByte<or, bdaddr12pair, OI>;
761 defm : RMWIByte<or, bdaddr20pair, OIY>;
763 //===----------------------------------------------------------------------===//
765 //===----------------------------------------------------------------------===//
768 // XORs of a register.
769 let isCommutable = 1 in {
770 defm XR : BinaryRRAndK<"x", 0x17, 0xB9F7, xor, GR32, GR32>;
771 defm XGR : BinaryRREAndK<"xg", 0xB982, 0xB9E7, xor, GR64, GR64>;
774 // XORs of a 32-bit immediate, leaving other bits unaffected.
775 let isCodeGenOnly = 1 in
776 def XILF32 : BinaryRIL<"xilf", 0xC07, xor, GR32, uimm32>;
777 def XILF : BinaryRIL<"xilf", 0xC07, xor, GR64, imm64lf32>;
778 def XIHF : BinaryRIL<"xihf", 0xC06, xor, GR64, imm64hf32>;
781 defm X : BinaryRXPair<"x",0x57, 0xE357, xor, GR32, load, 4>;
782 def XG : BinaryRXY<"xg", 0xE382, xor, GR64, load, 8>;
785 defm XI : BinarySIPair<"xi", 0x97, 0xEB57, null_frag, uimm8>;
787 defm : RMWIByte<xor, bdaddr12pair, XI>;
788 defm : RMWIByte<xor, bdaddr20pair, XIY>;
790 //===----------------------------------------------------------------------===//
792 //===----------------------------------------------------------------------===//
794 // Multiplication of a register.
795 let isCommutable = 1 in {
796 def MSR : BinaryRRE<"ms", 0xB252, mul, GR32, GR32>;
797 def MSGR : BinaryRRE<"msg", 0xB90C, mul, GR64, GR64>;
799 def MSGFR : BinaryRRE<"msgf", 0xB91C, null_frag, GR64, GR32>;
800 defm : SXB<mul, GR64, MSGFR>;
802 // Multiplication of a signed 16-bit immediate.
803 def MHI : BinaryRI<"mhi", 0xA7C, mul, GR32, imm32sx16>;
804 def MGHI : BinaryRI<"mghi", 0xA7D, mul, GR64, imm64sx16>;
806 // Multiplication of a signed 32-bit immediate.
807 def MSFI : BinaryRIL<"msfi", 0xC21, mul, GR32, simm32>;
808 def MSGFI : BinaryRIL<"msgfi", 0xC20, mul, GR64, imm64sx32>;
810 // Multiplication of memory.
811 defm MH : BinaryRXPair<"mh", 0x4C, 0xE37C, mul, GR32, sextloadi16, 2>;
812 defm MS : BinaryRXPair<"ms", 0x71, 0xE351, mul, GR32, load, 4>;
813 def MSGF : BinaryRXY<"msgf", 0xE31C, mul, GR64, sextloadi32, 4>;
814 def MSG : BinaryRXY<"msg", 0xE30C, mul, GR64, load, 8>;
816 // Multiplication of a register, producing two results.
817 def MLGR : BinaryRRE<"mlg", 0xB986, z_umul_lohi64, GR128, GR64>;
819 // Multiplication of memory, producing two results.
820 def MLG : BinaryRXY<"mlg", 0xE386, z_umul_lohi64, GR128, load, 8>;
822 //===----------------------------------------------------------------------===//
823 // Division and remainder
824 //===----------------------------------------------------------------------===//
826 // Division and remainder, from registers.
827 def DSGFR : BinaryRRE<"dsgf", 0xB91D, z_sdivrem32, GR128, GR32>;
828 def DSGR : BinaryRRE<"dsg", 0xB90D, z_sdivrem64, GR128, GR64>;
829 def DLR : BinaryRRE<"dl", 0xB997, z_udivrem32, GR128, GR32>;
830 def DLGR : BinaryRRE<"dlg", 0xB987, z_udivrem64, GR128, GR64>;
832 // Division and remainder, from memory.
833 def DSGF : BinaryRXY<"dsgf", 0xE31D, z_sdivrem32, GR128, load, 4>;
834 def DSG : BinaryRXY<"dsg", 0xE30D, z_sdivrem64, GR128, load, 8>;
835 def DL : BinaryRXY<"dl", 0xE397, z_udivrem32, GR128, load, 4>;
836 def DLG : BinaryRXY<"dlg", 0xE387, z_udivrem64, GR128, load, 8>;
838 //===----------------------------------------------------------------------===//
840 //===----------------------------------------------------------------------===//
843 let neverHasSideEffects = 1 in {
844 defm SLL : ShiftRSAndK<"sll", 0x89, 0xEBDF, shl, GR32>;
845 def SLLG : ShiftRSY<"sllg", 0xEB0D, shl, GR64>;
848 // Logical shift right.
849 let neverHasSideEffects = 1 in {
850 defm SRL : ShiftRSAndK<"srl", 0x88, 0xEBDE, srl, GR32>;
851 def SRLG : ShiftRSY<"srlg", 0xEB0C, srl, GR64>;
854 // Arithmetic shift right.
856 defm SRA : ShiftRSAndK<"sra", 0x8A, 0xEBDC, sra, GR32>;
857 def SRAG : ShiftRSY<"srag", 0xEB0A, sra, GR64>;
861 let neverHasSideEffects = 1 in {
862 def RLL : ShiftRSY<"rll", 0xEB1D, rotl, GR32>;
863 def RLLG : ShiftRSY<"rllg", 0xEB1C, rotl, GR64>;
866 // Rotate second operand left and inserted selected bits into first operand.
867 // These can act like 32-bit operands provided that the constant start and
868 // end bits (operands 2 and 3) are in the range [32, 64)
870 let isCodeGenOnly = 1 in
871 def RISBG32 : RotateSelectRIEf<"risbg", 0xEC55, GR32, GR32>;
872 def RISBG : RotateSelectRIEf<"risbg", 0xEC55, GR64, GR64>;
875 // Rotate second operand left and perform a logical operation with selected
876 // bits of the first operand.
878 def RNSBG : RotateSelectRIEf<"rnsbg", 0xEC54, GR64, GR64>;
879 def ROSBG : RotateSelectRIEf<"rosbg", 0xEC56, GR64, GR64>;
880 def RXSBG : RotateSelectRIEf<"rxsbg", 0xEC57, GR64, GR64>;
883 //===----------------------------------------------------------------------===//
885 //===----------------------------------------------------------------------===//
887 // Signed comparisons.
889 // Comparison with a register.
890 def CR : CompareRR <"c", 0x19, z_cmp, GR32, GR32>;
891 def CGFR : CompareRRE<"cgf", 0xB930, null_frag, GR64, GR32>;
892 def CGR : CompareRRE<"cg", 0xB920, z_cmp, GR64, GR64>;
894 // Comparison with a signed 16-bit immediate.
895 def CHI : CompareRI<"chi", 0xA7E, z_cmp, GR32, imm32sx16>;
896 def CGHI : CompareRI<"cghi", 0xA7F, z_cmp, GR64, imm64sx16>;
898 // Comparison with a signed 32-bit immediate.
899 def CFI : CompareRIL<"cfi", 0xC2D, z_cmp, GR32, simm32>;
900 def CGFI : CompareRIL<"cgfi", 0xC2C, z_cmp, GR64, imm64sx32>;
902 // Comparison with memory.
903 defm CH : CompareRXPair<"ch", 0x49, 0xE379, z_cmp, GR32, sextloadi16, 2>;
904 defm C : CompareRXPair<"c", 0x59, 0xE359, z_cmp, GR32, load, 4>;
905 def CGH : CompareRXY<"cgh", 0xE334, z_cmp, GR64, sextloadi16, 2>;
906 def CGF : CompareRXY<"cgf", 0xE330, z_cmp, GR64, sextloadi32, 4>;
907 def CG : CompareRXY<"cg", 0xE320, z_cmp, GR64, load, 8>;
908 def CHRL : CompareRILPC<"chrl", 0xC65, z_cmp, GR32, aligned_sextloadi16>;
909 def CRL : CompareRILPC<"crl", 0xC6D, z_cmp, GR32, aligned_load>;
910 def CGHRL : CompareRILPC<"cghrl", 0xC64, z_cmp, GR64, aligned_sextloadi16>;
911 def CGFRL : CompareRILPC<"cgfrl", 0xC6C, z_cmp, GR64, aligned_sextloadi32>;
912 def CGRL : CompareRILPC<"cgrl", 0xC68, z_cmp, GR64, aligned_load>;
914 // Comparison between memory and a signed 16-bit immediate.
915 def CHHSI : CompareSIL<"chhsi", 0xE554, z_cmp, sextloadi16, imm32sx16>;
916 def CHSI : CompareSIL<"chsi", 0xE55C, z_cmp, load, imm32sx16>;
917 def CGHSI : CompareSIL<"cghsi", 0xE558, z_cmp, load, imm64sx16>;
919 defm : SXB<z_cmp, GR64, CGFR>;
921 // Unsigned comparisons.
923 // Comparison with a register.
924 def CLR : CompareRR <"cl", 0x15, z_ucmp, GR32, GR32>;
925 def CLGFR : CompareRRE<"clgf", 0xB931, null_frag, GR64, GR32>;
926 def CLGR : CompareRRE<"clg", 0xB921, z_ucmp, GR64, GR64>;
928 // Comparison with a signed 32-bit immediate.
929 def CLFI : CompareRIL<"clfi", 0xC2F, z_ucmp, GR32, uimm32>;
930 def CLGFI : CompareRIL<"clgfi", 0xC2E, z_ucmp, GR64, imm64zx32>;
932 // Comparison with memory.
933 defm CL : CompareRXPair<"cl", 0x55, 0xE355, z_ucmp, GR32, load, 4>;
934 def CLGF : CompareRXY<"clgf", 0xE331, z_ucmp, GR64, zextloadi32, 4>;
935 def CLG : CompareRXY<"clg", 0xE321, z_ucmp, GR64, load, 8>;
936 def CLHRL : CompareRILPC<"clhrl", 0xC67, z_ucmp, GR32,
937 aligned_zextloadi16>;
938 def CLRL : CompareRILPC<"clrl", 0xC6F, z_ucmp, GR32,
940 def CLGHRL : CompareRILPC<"clghrl", 0xC66, z_ucmp, GR64,
941 aligned_zextloadi16>;
942 def CLGFRL : CompareRILPC<"clgfrl", 0xC6E, z_ucmp, GR64,
943 aligned_zextloadi32>;
944 def CLGRL : CompareRILPC<"clgrl", 0xC6A, z_ucmp, GR64,
947 // Comparison between memory and an unsigned 8-bit immediate.
948 defm CLI : CompareSIPair<"cli", 0x95, 0xEB55, z_ucmp, zextloadi8, imm32zx8>;
950 // Comparison between memory and an unsigned 16-bit immediate.
951 def CLHHSI : CompareSIL<"clhhsi", 0xE555, z_ucmp, zextloadi16, imm32zx16>;
952 def CLFHSI : CompareSIL<"clfhsi", 0xE55D, z_ucmp, load, imm32zx16>;
953 def CLGHSI : CompareSIL<"clghsi", 0xE559, z_ucmp, load, imm64zx16>;
955 defm : ZXB<z_ucmp, GR64, CLGFR>;
957 //===----------------------------------------------------------------------===//
959 //===----------------------------------------------------------------------===//
961 def ATOMIC_SWAPW : AtomicLoadWBinaryReg<z_atomic_swapw>;
962 def ATOMIC_SWAP_32 : AtomicLoadBinaryReg32<atomic_swap_32>;
963 def ATOMIC_SWAP_64 : AtomicLoadBinaryReg64<atomic_swap_64>;
965 def ATOMIC_LOADW_AR : AtomicLoadWBinaryReg<z_atomic_loadw_add>;
966 def ATOMIC_LOADW_AFI : AtomicLoadWBinaryImm<z_atomic_loadw_add, simm32>;
967 def ATOMIC_LOAD_AR : AtomicLoadBinaryReg32<atomic_load_add_32>;
968 def ATOMIC_LOAD_AHI : AtomicLoadBinaryImm32<atomic_load_add_32, imm32sx16>;
969 def ATOMIC_LOAD_AFI : AtomicLoadBinaryImm32<atomic_load_add_32, simm32>;
970 def ATOMIC_LOAD_AGR : AtomicLoadBinaryReg64<atomic_load_add_64>;
971 def ATOMIC_LOAD_AGHI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx16>;
972 def ATOMIC_LOAD_AGFI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx32>;
974 def ATOMIC_LOADW_SR : AtomicLoadWBinaryReg<z_atomic_loadw_sub>;
975 def ATOMIC_LOAD_SR : AtomicLoadBinaryReg32<atomic_load_sub_32>;
976 def ATOMIC_LOAD_SGR : AtomicLoadBinaryReg64<atomic_load_sub_64>;
978 def ATOMIC_LOADW_NR : AtomicLoadWBinaryReg<z_atomic_loadw_and>;
979 def ATOMIC_LOADW_NILH : AtomicLoadWBinaryImm<z_atomic_loadw_and, imm32lh16c>;
980 def ATOMIC_LOAD_NR : AtomicLoadBinaryReg32<atomic_load_and_32>;
981 def ATOMIC_LOAD_NILL32 : AtomicLoadBinaryImm32<atomic_load_and_32, imm32ll16c>;
982 def ATOMIC_LOAD_NILH32 : AtomicLoadBinaryImm32<atomic_load_and_32, imm32lh16c>;
983 def ATOMIC_LOAD_NILF32 : AtomicLoadBinaryImm32<atomic_load_and_32, uimm32>;
984 def ATOMIC_LOAD_NGR : AtomicLoadBinaryReg64<atomic_load_and_64>;
985 def ATOMIC_LOAD_NILL : AtomicLoadBinaryImm64<atomic_load_and_64, imm64ll16c>;
986 def ATOMIC_LOAD_NILH : AtomicLoadBinaryImm64<atomic_load_and_64, imm64lh16c>;
987 def ATOMIC_LOAD_NIHL : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hl16c>;
988 def ATOMIC_LOAD_NIHH : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hh16c>;
989 def ATOMIC_LOAD_NILF : AtomicLoadBinaryImm64<atomic_load_and_64, imm64lf32c>;
990 def ATOMIC_LOAD_NIHF : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hf32c>;
992 def ATOMIC_LOADW_OR : AtomicLoadWBinaryReg<z_atomic_loadw_or>;
993 def ATOMIC_LOADW_OILH : AtomicLoadWBinaryImm<z_atomic_loadw_or, imm32lh16>;
994 def ATOMIC_LOAD_OR : AtomicLoadBinaryReg32<atomic_load_or_32>;
995 def ATOMIC_LOAD_OILL32 : AtomicLoadBinaryImm32<atomic_load_or_32, imm32ll16>;
996 def ATOMIC_LOAD_OILH32 : AtomicLoadBinaryImm32<atomic_load_or_32, imm32lh16>;
997 def ATOMIC_LOAD_OILF32 : AtomicLoadBinaryImm32<atomic_load_or_32, uimm32>;
998 def ATOMIC_LOAD_OGR : AtomicLoadBinaryReg64<atomic_load_or_64>;
999 def ATOMIC_LOAD_OILL : AtomicLoadBinaryImm64<atomic_load_or_64, imm64ll16>;
1000 def ATOMIC_LOAD_OILH : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lh16>;
1001 def ATOMIC_LOAD_OIHL : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hl16>;
1002 def ATOMIC_LOAD_OIHH : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hh16>;
1003 def ATOMIC_LOAD_OILF : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lf32>;
1004 def ATOMIC_LOAD_OIHF : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hf32>;
1006 def ATOMIC_LOADW_XR : AtomicLoadWBinaryReg<z_atomic_loadw_xor>;
1007 def ATOMIC_LOADW_XILF : AtomicLoadWBinaryImm<z_atomic_loadw_xor, uimm32>;
1008 def ATOMIC_LOAD_XR : AtomicLoadBinaryReg32<atomic_load_xor_32>;
1009 def ATOMIC_LOAD_XILF32 : AtomicLoadBinaryImm32<atomic_load_xor_32, uimm32>;
1010 def ATOMIC_LOAD_XGR : AtomicLoadBinaryReg64<atomic_load_xor_64>;
1011 def ATOMIC_LOAD_XILF : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64lf32>;
1012 def ATOMIC_LOAD_XIHF : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64hf32>;
1014 def ATOMIC_LOADW_NRi : AtomicLoadWBinaryReg<z_atomic_loadw_nand>;
1015 def ATOMIC_LOADW_NILHi : AtomicLoadWBinaryImm<z_atomic_loadw_nand,
1017 def ATOMIC_LOAD_NRi : AtomicLoadBinaryReg32<atomic_load_nand_32>;
1018 def ATOMIC_LOAD_NILL32i : AtomicLoadBinaryImm32<atomic_load_nand_32,
1020 def ATOMIC_LOAD_NILH32i : AtomicLoadBinaryImm32<atomic_load_nand_32,
1022 def ATOMIC_LOAD_NILF32i : AtomicLoadBinaryImm32<atomic_load_nand_32, uimm32>;
1023 def ATOMIC_LOAD_NGRi : AtomicLoadBinaryReg64<atomic_load_nand_64>;
1024 def ATOMIC_LOAD_NILLi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1026 def ATOMIC_LOAD_NILHi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1028 def ATOMIC_LOAD_NIHLi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1030 def ATOMIC_LOAD_NIHHi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1032 def ATOMIC_LOAD_NILFi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1034 def ATOMIC_LOAD_NIHFi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1037 def ATOMIC_LOADW_MIN : AtomicLoadWBinaryReg<z_atomic_loadw_min>;
1038 def ATOMIC_LOAD_MIN_32 : AtomicLoadBinaryReg32<atomic_load_min_32>;
1039 def ATOMIC_LOAD_MIN_64 : AtomicLoadBinaryReg64<atomic_load_min_64>;
1041 def ATOMIC_LOADW_MAX : AtomicLoadWBinaryReg<z_atomic_loadw_max>;
1042 def ATOMIC_LOAD_MAX_32 : AtomicLoadBinaryReg32<atomic_load_max_32>;
1043 def ATOMIC_LOAD_MAX_64 : AtomicLoadBinaryReg64<atomic_load_max_64>;
1045 def ATOMIC_LOADW_UMIN : AtomicLoadWBinaryReg<z_atomic_loadw_umin>;
1046 def ATOMIC_LOAD_UMIN_32 : AtomicLoadBinaryReg32<atomic_load_umin_32>;
1047 def ATOMIC_LOAD_UMIN_64 : AtomicLoadBinaryReg64<atomic_load_umin_64>;
1049 def ATOMIC_LOADW_UMAX : AtomicLoadWBinaryReg<z_atomic_loadw_umax>;
1050 def ATOMIC_LOAD_UMAX_32 : AtomicLoadBinaryReg32<atomic_load_umax_32>;
1051 def ATOMIC_LOAD_UMAX_64 : AtomicLoadBinaryReg64<atomic_load_umax_64>;
1053 def ATOMIC_CMP_SWAPW
1054 : Pseudo<(outs GR32:$dst), (ins bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1055 ADDR32:$bitshift, ADDR32:$negbitshift,
1058 (z_atomic_cmp_swapw bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1059 ADDR32:$bitshift, ADDR32:$negbitshift,
1060 uimm32:$bitsize))]> {
1064 let usesCustomInserter = 1;
1067 let Defs = [CC] in {
1068 defm CS : CmpSwapRSPair<"cs", 0xBA, 0xEB14, atomic_cmp_swap_32, GR32>;
1069 def CSG : CmpSwapRSY<"csg", 0xEB30, atomic_cmp_swap_64, GR64>;
1072 //===----------------------------------------------------------------------===//
1073 // Miscellaneous Instructions.
1074 //===----------------------------------------------------------------------===//
1076 // Read a 32-bit access register into a GR32. As with all GR32 operations,
1077 // the upper 32 bits of the enclosing GR64 remain unchanged, which is useful
1078 // when a 64-bit address is stored in a pair of access registers.
1079 def EAR : InstRRE<0xB24F, (outs GR32:$R1), (ins access_reg:$R2),
1081 [(set GR32:$R1, (z_extract_access access_reg:$R2))]>;
1083 // Find leftmost one, AKA count leading zeros. The instruction actually
1084 // returns a pair of GR64s, the first giving the number of leading zeros
1085 // and the second giving a copy of the source with the leftmost one bit
1086 // cleared. We only use the first result here.
1087 let Defs = [CC] in {
1088 def FLOGR : UnaryRRE<"flog", 0xB983, null_frag, GR128, GR64>;
1090 def : Pat<(ctlz GR64:$src),
1091 (EXTRACT_SUBREG (FLOGR GR64:$src), subreg_high)>;
1093 // Use subregs to populate the "don't care" bits in a 32-bit to 64-bit anyext.
1094 def : Pat<(i64 (anyext GR32:$src)),
1095 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_32bit)>;
1097 // There are no 32-bit equivalents of LLILL and LLILH, so use a full
1098 // 64-bit move followed by a subreg. This preserves the invariant that
1099 // all GR32 operations only modify the low 32 bits.
1100 def : Pat<(i32 imm32ll16:$src),
1101 (EXTRACT_SUBREG (LLILL (LL16 imm:$src)), subreg_32bit)>;
1102 def : Pat<(i32 imm32lh16:$src),
1103 (EXTRACT_SUBREG (LLILH (LH16 imm:$src)), subreg_32bit)>;
1105 // Extend GR32s and GR64s to GR128s.
1106 let usesCustomInserter = 1 in {
1107 def AEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1108 def ZEXT128_32 : Pseudo<(outs GR128:$dst), (ins GR32:$src), []>;
1109 def ZEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1112 //===----------------------------------------------------------------------===//
1114 //===----------------------------------------------------------------------===//
1116 // Use AL* for GR64 additions of unsigned 32-bit values.
1117 defm : ZXB<add, GR64, ALGFR>;
1118 def : Pat<(add GR64:$src1, imm64zx32:$src2),
1119 (ALGFI GR64:$src1, imm64zx32:$src2)>;
1120 def : Pat<(add GR64:$src1, (zextloadi32 bdxaddr20only:$addr)),
1121 (ALGF GR64:$src1, bdxaddr20only:$addr)>;
1123 // Use SL* for GR64 subtractions of unsigned 32-bit values.
1124 defm : ZXB<sub, GR64, SLGFR>;
1125 def : Pat<(add GR64:$src1, imm64zx32n:$src2),
1126 (SLGFI GR64:$src1, imm64zx32n:$src2)>;
1127 def : Pat<(sub GR64:$src1, (zextloadi32 bdxaddr20only:$addr)),
1128 (SLGF GR64:$src1, bdxaddr20only:$addr)>;
1130 // Optimize sign-extended 1/0 selects to -1/0 selects. This is important
1131 // for vector legalization.
1132 def : Pat<(sra (shl (i32 (z_select_ccmask 1, 0, imm:$cc)), (i32 31)), (i32 31)),
1133 (Select32 (LHI -1), (LHI 0), imm:$cc)>;
1134 def : Pat<(sra (shl (i64 (anyext (i32 (z_select_ccmask 1, 0, imm:$cc)))),
1137 (Select64 (LGHI -1), (LGHI 0), imm:$cc)>;