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 STOC : FixedCondStoreRSY<"stoc"##name, 0xEBF3, GR32, ccmask, 4>;
116 def STOCG : FixedCondStoreRSY<"stocg"##name, 0xEBE3, GR64, ccmask, 8>;
118 defm AsmO : CondExtendedMnemonic<1, "o">;
119 defm AsmH : CondExtendedMnemonic<2, "h">;
120 defm AsmNLE : CondExtendedMnemonic<3, "nle">;
121 defm AsmL : CondExtendedMnemonic<4, "l">;
122 defm AsmNHE : CondExtendedMnemonic<5, "nhe">;
123 defm AsmLH : CondExtendedMnemonic<6, "lh">;
124 defm AsmNE : CondExtendedMnemonic<7, "ne">;
125 defm AsmE : CondExtendedMnemonic<8, "e">;
126 defm AsmNLH : CondExtendedMnemonic<9, "nlh">;
127 defm AsmHE : CondExtendedMnemonic<10, "he">;
128 defm AsmNL : CondExtendedMnemonic<11, "nl">;
129 defm AsmLE : CondExtendedMnemonic<12, "le">;
130 defm AsmNH : CondExtendedMnemonic<13, "nh">;
131 defm AsmNO : CondExtendedMnemonic<14, "no">;
133 // Define AsmParser mnemonics for each integer condition-code mask.
134 // This is like the list above, except that condition 3 is not possible
135 // and that the low bit of the mask is therefore always 0. This means
136 // that each condition has two names. Conditions "o" and "no" are not used.
138 // We don't make one of the two names an alias of the other because
139 // we need the custom parsing routines to select the correct register class.
140 multiclass IntCondExtendedMnemonicA<bits<4> ccmask, string name> {
142 def CR : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2,
144 "crj"##name##"\t$R1, $R2, $RI4", []>;
145 def CGR : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2,
147 "cgrj"##name##"\t$R1, $R2, $RI4", []>;
148 def CI : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2,
150 "cij"##name##"\t$R1, $I2, $RI4", []>;
151 def CGI : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2,
153 "cgij"##name##"\t$R1, $I2, $RI4", []>;
156 multiclass IntCondExtendedMnemonic<bits<4> ccmask, string name1, string name2>
157 : IntCondExtendedMnemonicA<ccmask, name1> {
158 let isAsmParserOnly = 1 in
159 defm Alt : IntCondExtendedMnemonicA<ccmask, name2>;
161 defm AsmJH : IntCondExtendedMnemonic<2, "h", "nle">;
162 defm AsmJL : IntCondExtendedMnemonic<4, "l", "nhe">;
163 defm AsmJLH : IntCondExtendedMnemonic<6, "lh", "ne">;
164 defm AsmJE : IntCondExtendedMnemonic<8, "e", "nlh">;
165 defm AsmJHE : IntCondExtendedMnemonic<10, "he", "nl">;
166 defm AsmJLE : IntCondExtendedMnemonic<12, "le", "nh">;
168 //===----------------------------------------------------------------------===//
169 // Select instructions
170 //===----------------------------------------------------------------------===//
172 def Select32 : SelectWrapper<GR32>;
173 def Select64 : SelectWrapper<GR64>;
175 defm CondStore8_32 : CondStores<GR32, nonvolatile_truncstorei8,
176 nonvolatile_anyextloadi8, bdxaddr20only>;
177 defm CondStore16_32 : CondStores<GR32, nonvolatile_truncstorei16,
178 nonvolatile_anyextloadi16, bdxaddr20only>;
179 defm CondStore32_32 : CondStores<GR32, nonvolatile_store,
180 nonvolatile_load, bdxaddr20only>;
182 defm CondStore8 : CondStores<GR64, nonvolatile_truncstorei8,
183 nonvolatile_anyextloadi8, bdxaddr20only>;
184 defm CondStore16 : CondStores<GR64, nonvolatile_truncstorei16,
185 nonvolatile_anyextloadi16, bdxaddr20only>;
186 defm CondStore32 : CondStores<GR64, nonvolatile_truncstorei32,
187 nonvolatile_anyextloadi32, bdxaddr20only>;
188 defm CondStore64 : CondStores<GR64, nonvolatile_store,
189 nonvolatile_load, bdxaddr20only>;
191 //===----------------------------------------------------------------------===//
193 //===----------------------------------------------------------------------===//
195 // The definitions here are for the call-clobbered registers.
196 let isCall = 1, Defs = [R0D, R1D, R2D, R3D, R4D, R5D, R14D,
197 F0D, F1D, F2D, F3D, F4D, F5D, F6D, F7D],
198 R1 = 14, isCodeGenOnly = 1 in {
199 def BRAS : InstRI<0xA75, (outs), (ins pcrel16call:$I2, variable_ops),
200 "bras\t%r14, $I2", []>;
201 def BRASL : InstRIL<0xC05, (outs), (ins pcrel32call:$I2, variable_ops),
202 "brasl\t%r14, $I2", [(z_call pcrel32call:$I2)]>;
203 def BASR : InstRR<0x0D, (outs), (ins ADDR64:$R2, variable_ops),
204 "basr\t%r14, $R2", [(z_call ADDR64:$R2)]>;
207 // Define the general form of the call instructions for the asm parser.
208 // These instructions don't hard-code %r14 as the return address register.
209 def AsmBRAS : InstRI<0xA75, (outs), (ins GR64:$R1, brtarget16:$I2),
210 "bras\t$R1, $I2", []>;
211 def AsmBRASL : InstRIL<0xC05, (outs), (ins GR64:$R1, brtarget32:$I2),
212 "brasl\t$R1, $I2", []>;
213 def AsmBASR : InstRR<0x0D, (outs), (ins GR64:$R1, ADDR64:$R2),
214 "basr\t$R1, $R2", []>;
216 //===----------------------------------------------------------------------===//
218 //===----------------------------------------------------------------------===//
221 let neverHasSideEffects = 1 in {
222 def LR : UnaryRR <"l", 0x18, null_frag, GR32, GR32>;
223 def LGR : UnaryRRE<"lg", 0xB904, null_frag, GR64, GR64>;
227 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
228 isReMaterializable = 1 in {
229 // 16-bit sign-extended immediates.
230 def LHI : UnaryRI<"lhi", 0xA78, bitconvert, GR32, imm32sx16>;
231 def LGHI : UnaryRI<"lghi", 0xA79, bitconvert, GR64, imm64sx16>;
233 // Other 16-bit immediates.
234 def LLILL : UnaryRI<"llill", 0xA5F, bitconvert, GR64, imm64ll16>;
235 def LLILH : UnaryRI<"llilh", 0xA5E, bitconvert, GR64, imm64lh16>;
236 def LLIHL : UnaryRI<"llihl", 0xA5D, bitconvert, GR64, imm64hl16>;
237 def LLIHH : UnaryRI<"llihh", 0xA5C, bitconvert, GR64, imm64hh16>;
239 // 32-bit immediates.
240 def LGFI : UnaryRIL<"lgfi", 0xC01, bitconvert, GR64, imm64sx32>;
241 def LLILF : UnaryRIL<"llilf", 0xC0F, bitconvert, GR64, imm64lf32>;
242 def LLIHF : UnaryRIL<"llihf", 0xC0E, bitconvert, GR64, imm64hf32>;
246 let canFoldAsLoad = 1, SimpleBDXLoad = 1 in {
247 defm L : UnaryRXPair<"l", 0x58, 0xE358, load, GR32, 4>;
248 def LG : UnaryRXY<"lg", 0xE304, load, GR64, 8>;
250 // These instructions are split after register allocation, so we don't
251 // want a custom inserter.
252 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
253 def L128 : Pseudo<(outs GR128:$dst), (ins bdxaddr20only128:$src),
254 [(set GR128:$dst, (load bdxaddr20only128:$src))]>;
257 let canFoldAsLoad = 1 in {
258 def LRL : UnaryRILPC<"lrl", 0xC4D, aligned_load, GR32>;
259 def LGRL : UnaryRILPC<"lgrl", 0xC48, aligned_load, GR64>;
263 let SimpleBDXStore = 1 in {
264 let isCodeGenOnly = 1 in
265 defm ST32 : StoreRXPair<"st", 0x50, 0xE350, store, GR32, 4>;
266 def STG : StoreRXY<"stg", 0xE324, store, GR64, 8>;
268 // These instructions are split after register allocation, so we don't
269 // want a custom inserter.
270 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
271 def ST128 : Pseudo<(outs), (ins GR128:$src, bdxaddr20only128:$dst),
272 [(store GR128:$src, bdxaddr20only128:$dst)]>;
275 let isCodeGenOnly = 1 in
276 def STRL32 : StoreRILPC<"strl", 0xC4F, aligned_store, GR32>;
277 def STGRL : StoreRILPC<"stgrl", 0xC4B, aligned_store, GR64>;
279 // Store on condition.
280 let isCodeGenOnly = 1, Uses = [CC] in {
281 def STOC32 : CondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
282 def STOC : CondStoreRSY<"stoc", 0xEBF3, GR64, 4>;
283 def STOCG : CondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
286 def AsmSTOC : AsmCondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
287 def AsmSTOCG : AsmCondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
290 // 8-bit immediate stores to 8-bit fields.
291 defm MVI : StoreSIPair<"mvi", 0x92, 0xEB52, truncstorei8, imm32zx8trunc>;
293 // 16-bit immediate stores to 16-, 32- or 64-bit fields.
294 def MVHHI : StoreSIL<"mvhhi", 0xE544, truncstorei16, imm32sx16trunc>;
295 def MVHI : StoreSIL<"mvhi", 0xE54C, store, imm32sx16>;
296 def MVGHI : StoreSIL<"mvghi", 0xE548, store, imm64sx16>;
298 // Memory-to-memory moves.
299 let mayLoad = 1, mayStore = 1 in
300 def MVC : InstSS<0xD2, (outs), (ins bdladdr12onlylen8:$BDL1,
302 "mvc\t$BDL1, $BD2", []>;
304 let mayLoad = 1, mayStore = 1, usesCustomInserter = 1 in
305 def MVCWrapper : Pseudo<(outs), (ins bdaddr12only:$dest, bdaddr12only:$src,
307 [(z_mvc bdaddr12only:$dest, bdaddr12only:$src,
308 imm32len8:$length)]>;
310 defm LoadStore8_32 : MVCLoadStore<anyextloadi8, truncstorei8, i32,
312 defm LoadStore16_32 : MVCLoadStore<anyextloadi16, truncstorei16, i32,
314 defm LoadStore32_32 : MVCLoadStore<load, store, i32, MVCWrapper, 4>;
316 defm LoadStore8 : MVCLoadStore<anyextloadi8, truncstorei8, i64,
318 defm LoadStore16 : MVCLoadStore<anyextloadi16, truncstorei16, i64,
320 defm LoadStore32 : MVCLoadStore<anyextloadi32, truncstorei32, i64,
322 defm LoadStore64 : MVCLoadStore<load, store, i64, MVCWrapper, 8>;
324 //===----------------------------------------------------------------------===//
326 //===----------------------------------------------------------------------===//
328 // 32-bit extensions from registers.
329 let neverHasSideEffects = 1 in {
330 def LBR : UnaryRRE<"lb", 0xB926, sext8, GR32, GR32>;
331 def LHR : UnaryRRE<"lh", 0xB927, sext16, GR32, GR32>;
334 // 64-bit extensions from registers.
335 let neverHasSideEffects = 1 in {
336 def LGBR : UnaryRRE<"lgb", 0xB906, sext8, GR64, GR64>;
337 def LGHR : UnaryRRE<"lgh", 0xB907, sext16, GR64, GR64>;
338 def LGFR : UnaryRRE<"lgf", 0xB914, sext32, GR64, GR32>;
341 // Match 32-to-64-bit sign extensions in which the source is already
342 // in a 64-bit register.
343 def : Pat<(sext_inreg GR64:$src, i32),
344 (LGFR (EXTRACT_SUBREG GR64:$src, subreg_32bit))>;
346 // 32-bit extensions from memory.
347 def LB : UnaryRXY<"lb", 0xE376, sextloadi8, GR32, 1>;
348 defm LH : UnaryRXPair<"lh", 0x48, 0xE378, sextloadi16, GR32, 2>;
349 def LHRL : UnaryRILPC<"lhrl", 0xC45, aligned_sextloadi16, GR32>;
351 // 64-bit extensions from memory.
352 def LGB : UnaryRXY<"lgb", 0xE377, sextloadi8, GR64, 1>;
353 def LGH : UnaryRXY<"lgh", 0xE315, sextloadi16, GR64, 2>;
354 def LGF : UnaryRXY<"lgf", 0xE314, sextloadi32, GR64, 4>;
355 def LGHRL : UnaryRILPC<"lghrl", 0xC44, aligned_sextloadi16, GR64>;
356 def LGFRL : UnaryRILPC<"lgfrl", 0xC4C, aligned_sextloadi32, GR64>;
358 // If the sign of a load-extend operation doesn't matter, use the signed ones.
359 // There's not really much to choose between the sign and zero extensions,
360 // but LH is more compact than LLH for small offsets.
361 def : Pat<(i32 (extloadi8 bdxaddr20only:$src)), (LB bdxaddr20only:$src)>;
362 def : Pat<(i32 (extloadi16 bdxaddr12pair:$src)), (LH bdxaddr12pair:$src)>;
363 def : Pat<(i32 (extloadi16 bdxaddr20pair:$src)), (LHY bdxaddr20pair:$src)>;
365 def : Pat<(i64 (extloadi8 bdxaddr20only:$src)), (LGB bdxaddr20only:$src)>;
366 def : Pat<(i64 (extloadi16 bdxaddr20only:$src)), (LGH bdxaddr20only:$src)>;
367 def : Pat<(i64 (extloadi32 bdxaddr20only:$src)), (LGF bdxaddr20only:$src)>;
369 // We want PC-relative addresses to be tried ahead of BD and BDX addresses.
370 // However, BDXs have two extra operands and are therefore 6 units more
372 let AddedComplexity = 7 in {
373 def : Pat<(i32 (extloadi16 pcrel32:$src)), (LHRL pcrel32:$src)>;
374 def : Pat<(i64 (extloadi16 pcrel32:$src)), (LGHRL pcrel32:$src)>;
377 //===----------------------------------------------------------------------===//
379 //===----------------------------------------------------------------------===//
381 // 32-bit extensions from registers.
382 let neverHasSideEffects = 1 in {
383 def LLCR : UnaryRRE<"llc", 0xB994, zext8, GR32, GR32>;
384 def LLHR : UnaryRRE<"llh", 0xB995, zext16, GR32, GR32>;
387 // 64-bit extensions from registers.
388 let neverHasSideEffects = 1 in {
389 def LLGCR : UnaryRRE<"llgc", 0xB984, zext8, GR64, GR64>;
390 def LLGHR : UnaryRRE<"llgh", 0xB985, zext16, GR64, GR64>;
391 def LLGFR : UnaryRRE<"llgf", 0xB916, zext32, GR64, GR32>;
394 // Match 32-to-64-bit zero extensions in which the source is already
395 // in a 64-bit register.
396 def : Pat<(and GR64:$src, 0xffffffff),
397 (LLGFR (EXTRACT_SUBREG GR64:$src, subreg_32bit))>;
399 // 32-bit extensions from memory.
400 def LLC : UnaryRXY<"llc", 0xE394, zextloadi8, GR32, 1>;
401 def LLH : UnaryRXY<"llh", 0xE395, zextloadi16, GR32, 2>;
402 def LLHRL : UnaryRILPC<"llhrl", 0xC42, aligned_zextloadi16, GR32>;
404 // 64-bit extensions from memory.
405 def LLGC : UnaryRXY<"llgc", 0xE390, zextloadi8, GR64, 1>;
406 def LLGH : UnaryRXY<"llgh", 0xE391, zextloadi16, GR64, 2>;
407 def LLGF : UnaryRXY<"llgf", 0xE316, zextloadi32, GR64, 4>;
408 def LLGHRL : UnaryRILPC<"llghrl", 0xC46, aligned_zextloadi16, GR64>;
409 def LLGFRL : UnaryRILPC<"llgfrl", 0xC4E, aligned_zextloadi32, GR64>;
411 //===----------------------------------------------------------------------===//
413 //===----------------------------------------------------------------------===//
415 // Truncations of 64-bit registers to 32-bit registers.
416 def : Pat<(i32 (trunc GR64:$src)),
417 (EXTRACT_SUBREG GR64:$src, subreg_32bit)>;
419 // Truncations of 32-bit registers to memory.
420 let isCodeGenOnly = 1 in {
421 defm STC32 : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR32, 1>;
422 defm STH32 : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR32, 2>;
423 def STHRL32 : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR32>;
426 // Truncations of 64-bit registers to memory.
427 defm STC : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR64, 1>;
428 defm STH : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR64, 2>;
429 def STHRL : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR64>;
430 defm ST : StoreRXPair<"st", 0x50, 0xE350, truncstorei32, GR64, 4>;
431 def STRL : StoreRILPC<"strl", 0xC4F, aligned_truncstorei32, GR64>;
433 //===----------------------------------------------------------------------===//
434 // Multi-register moves
435 //===----------------------------------------------------------------------===//
437 // Multi-register loads.
438 def LMG : LoadMultipleRSY<"lmg", 0xEB04, GR64>;
440 // Multi-register stores.
441 def STMG : StoreMultipleRSY<"stmg", 0xEB24, GR64>;
443 //===----------------------------------------------------------------------===//
445 //===----------------------------------------------------------------------===//
447 // Byte-swapping register moves.
448 let neverHasSideEffects = 1 in {
449 def LRVR : UnaryRRE<"lrv", 0xB91F, bswap, GR32, GR32>;
450 def LRVGR : UnaryRRE<"lrvg", 0xB90F, bswap, GR64, GR64>;
453 // Byte-swapping loads. Unlike normal loads, these instructions are
454 // allowed to access storage more than once.
455 def LRV : UnaryRXY<"lrv", 0xE31E, loadu<bswap, nonvolatile_load>, GR32, 4>;
456 def LRVG : UnaryRXY<"lrvg", 0xE30F, loadu<bswap, nonvolatile_load>, GR64, 8>;
458 // Likewise byte-swapping stores.
459 def STRV : StoreRXY<"strv", 0xE33E, storeu<bswap, nonvolatile_store>, GR32, 4>;
460 def STRVG : StoreRXY<"strvg", 0xE32F, storeu<bswap, nonvolatile_store>,
463 //===----------------------------------------------------------------------===//
464 // Load address instructions
465 //===----------------------------------------------------------------------===//
467 // Load BDX-style addresses.
468 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isReMaterializable = 1,
470 let DispSize = "12" in
471 def LA : InstRX<0x41, (outs GR64:$R1), (ins laaddr12pair:$XBD2),
473 [(set GR64:$R1, laaddr12pair:$XBD2)]>;
474 let DispSize = "20" in
475 def LAY : InstRXY<0xE371, (outs GR64:$R1), (ins laaddr20pair:$XBD2),
477 [(set GR64:$R1, laaddr20pair:$XBD2)]>;
480 // Load a PC-relative address. There's no version of this instruction
481 // with a 16-bit offset, so there's no relaxation.
482 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
483 isReMaterializable = 1 in {
484 def LARL : InstRIL<0xC00, (outs GR64:$R1), (ins pcrel32:$I2),
486 [(set GR64:$R1, pcrel32:$I2)]>;
489 //===----------------------------------------------------------------------===//
491 //===----------------------------------------------------------------------===//
494 def LCR : UnaryRR <"lc", 0x13, ineg, GR32, GR32>;
495 def LCGR : UnaryRRE<"lcg", 0xB903, ineg, GR64, GR64>;
496 def LCGFR : UnaryRRE<"lcgf", 0xB913, null_frag, GR64, GR32>;
498 defm : SXU<ineg, LCGFR>;
500 //===----------------------------------------------------------------------===//
502 //===----------------------------------------------------------------------===//
504 let isCodeGenOnly = 1 in
505 defm IC32 : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR32, zextloadi8, 1>;
506 defm IC : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR64, zextloadi8, 1>;
508 defm : InsertMem<"inserti8", IC32, GR32, zextloadi8, bdxaddr12pair>;
509 defm : InsertMem<"inserti8", IC32Y, GR32, zextloadi8, bdxaddr20pair>;
511 defm : InsertMem<"inserti8", IC, GR64, zextloadi8, bdxaddr12pair>;
512 defm : InsertMem<"inserti8", ICY, GR64, zextloadi8, bdxaddr20pair>;
514 // Insertions of a 16-bit immediate, leaving other bits unaffected.
515 // We don't have or_as_insert equivalents of these operations because
516 // OI is available instead.
517 let isCodeGenOnly = 1 in {
518 def IILL32 : BinaryRI<"iill", 0xA53, insertll, GR32, imm32ll16>;
519 def IILH32 : BinaryRI<"iilh", 0xA52, insertlh, GR32, imm32lh16>;
521 def IILL : BinaryRI<"iill", 0xA53, insertll, GR64, imm64ll16>;
522 def IILH : BinaryRI<"iilh", 0xA52, insertlh, GR64, imm64lh16>;
523 def IIHL : BinaryRI<"iihl", 0xA51, inserthl, GR64, imm64hl16>;
524 def IIHH : BinaryRI<"iihh", 0xA50, inserthh, GR64, imm64hh16>;
526 // ...likewise for 32-bit immediates. For GR32s this is a general
527 // full-width move. (We use IILF rather than something like LLILF
528 // for 32-bit moves because IILF leaves the upper 32 bits of the
530 let isCodeGenOnly = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
531 isReMaterializable = 1 in {
532 def IILF32 : UnaryRIL<"iilf", 0xC09, bitconvert, GR32, uimm32>;
534 def IILF : BinaryRIL<"iilf", 0xC09, insertlf, GR64, imm64lf32>;
535 def IIHF : BinaryRIL<"iihf", 0xC08, inserthf, GR64, imm64hf32>;
537 // An alternative model of inserthf, with the first operand being
538 // a zero-extended value.
539 def : Pat<(or (zext32 GR32:$src), imm64hf32:$imm),
540 (IIHF (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_32bit),
543 //===----------------------------------------------------------------------===//
545 //===----------------------------------------------------------------------===//
549 // Addition of a register.
550 let isCommutable = 1 in {
551 defm AR : BinaryRRAndK<"a", 0x1A, 0xB9F8, add, GR32, GR32>;
552 defm AGR : BinaryRREAndK<"ag", 0xB908, 0xB9E8, add, GR64, GR64>;
554 def AGFR : BinaryRRE<"agf", 0xB918, null_frag, GR64, GR32>;
556 // Addition of signed 16-bit immediates.
557 defm AHI : BinaryRIAndK<"ahi", 0xA7A, 0xECD8, add, GR32, imm32sx16>;
558 defm AGHI : BinaryRIAndK<"aghi", 0xA7B, 0xECD9, add, GR64, imm64sx16>;
560 // Addition of signed 32-bit immediates.
561 def AFI : BinaryRIL<"afi", 0xC29, add, GR32, simm32>;
562 def AGFI : BinaryRIL<"agfi", 0xC28, add, GR64, imm64sx32>;
564 // Addition of memory.
565 defm AH : BinaryRXPair<"ah", 0x4A, 0xE37A, add, GR32, sextloadi16, 2>;
566 defm A : BinaryRXPair<"a", 0x5A, 0xE35A, add, GR32, load, 4>;
567 def AGF : BinaryRXY<"agf", 0xE318, add, GR64, sextloadi32, 4>;
568 def AG : BinaryRXY<"ag", 0xE308, add, GR64, load, 8>;
570 // Addition to memory.
571 def ASI : BinarySIY<"asi", 0xEB6A, add, imm32sx8>;
572 def AGSI : BinarySIY<"agsi", 0xEB7A, add, imm64sx8>;
574 defm : SXB<add, GR64, AGFR>;
576 // Addition producing a carry.
578 // Addition of a register.
579 let isCommutable = 1 in {
580 defm ALR : BinaryRRAndK<"al", 0x1E, 0xB9FA, addc, GR32, GR32>;
581 defm ALGR : BinaryRREAndK<"alg", 0xB90A, 0xB9EA, addc, GR64, GR64>;
583 def ALGFR : BinaryRRE<"algf", 0xB91A, null_frag, GR64, GR32>;
585 // Addition of signed 16-bit immediates.
586 def ALHSIK : BinaryRIE<"alhsik", 0xECDA, addc, GR32, imm32sx16>,
587 Requires<[FeatureDistinctOps]>;
588 def ALGHSIK : BinaryRIE<"alghsik", 0xECDB, addc, GR64, imm64sx16>,
589 Requires<[FeatureDistinctOps]>;
591 // Addition of unsigned 32-bit immediates.
592 def ALFI : BinaryRIL<"alfi", 0xC2B, addc, GR32, uimm32>;
593 def ALGFI : BinaryRIL<"algfi", 0xC2A, addc, GR64, imm64zx32>;
595 // Addition of memory.
596 defm AL : BinaryRXPair<"al", 0x5E, 0xE35E, addc, GR32, load, 4>;
597 def ALGF : BinaryRXY<"algf", 0xE31A, addc, GR64, zextloadi32, 4>;
598 def ALG : BinaryRXY<"alg", 0xE30A, addc, GR64, load, 8>;
600 defm : ZXB<addc, GR64, ALGFR>;
602 // Addition producing and using a carry.
603 let Defs = [CC], Uses = [CC] in {
604 // Addition of a register.
605 def ALCR : BinaryRRE<"alc", 0xB998, adde, GR32, GR32>;
606 def ALCGR : BinaryRRE<"alcg", 0xB988, adde, GR64, GR64>;
608 // Addition of memory.
609 def ALC : BinaryRXY<"alc", 0xE398, adde, GR32, load, 4>;
610 def ALCG : BinaryRXY<"alcg", 0xE388, adde, GR64, load, 8>;
613 //===----------------------------------------------------------------------===//
615 //===----------------------------------------------------------------------===//
617 // Plain substraction. Although immediate forms exist, we use the
618 // add-immediate instruction instead.
620 // Subtraction of a register.
621 defm SR : BinaryRRAndK<"s", 0x1B, 0xB9F9, sub, GR32, GR32>;
622 def SGFR : BinaryRRE<"sgf", 0xB919, null_frag, GR64, GR32>;
623 defm SGR : BinaryRREAndK<"sg", 0xB909, 0xB9E9, sub, GR64, GR64>;
625 // Subtraction of memory.
626 defm SH : BinaryRXPair<"sh", 0x4B, 0xE37B, sub, GR32, sextloadi16, 2>;
627 defm S : BinaryRXPair<"s", 0x5B, 0xE35B, sub, GR32, load, 4>;
628 def SGF : BinaryRXY<"sgf", 0xE319, sub, GR64, sextloadi32, 4>;
629 def SG : BinaryRXY<"sg", 0xE309, sub, GR64, load, 8>;
631 defm : SXB<sub, GR64, SGFR>;
633 // Subtraction producing a carry.
635 // Subtraction of a register.
636 defm SLR : BinaryRRAndK<"sl", 0x1F, 0xB9FB, subc, GR32, GR32>;
637 def SLGFR : BinaryRRE<"slgf", 0xB91B, null_frag, GR64, GR32>;
638 defm SLGR : BinaryRREAndK<"slg", 0xB90B, 0xB9EB, subc, GR64, GR64>;
640 // Subtraction of unsigned 32-bit immediates. These don't match
641 // subc because we prefer addc for constants.
642 def SLFI : BinaryRIL<"slfi", 0xC25, null_frag, GR32, uimm32>;
643 def SLGFI : BinaryRIL<"slgfi", 0xC24, null_frag, GR64, imm64zx32>;
645 // Subtraction of memory.
646 defm SL : BinaryRXPair<"sl", 0x5F, 0xE35F, subc, GR32, load, 4>;
647 def SLGF : BinaryRXY<"slgf", 0xE31B, subc, GR64, zextloadi32, 4>;
648 def SLG : BinaryRXY<"slg", 0xE30B, subc, GR64, load, 8>;
650 defm : ZXB<subc, GR64, SLGFR>;
652 // Subtraction producing and using a carry.
653 let Defs = [CC], Uses = [CC] in {
654 // Subtraction of a register.
655 def SLBR : BinaryRRE<"slb", 0xB999, sube, GR32, GR32>;
656 def SLGBR : BinaryRRE<"slbg", 0xB989, sube, GR64, GR64>;
658 // Subtraction of memory.
659 def SLB : BinaryRXY<"slb", 0xE399, sube, GR32, load, 4>;
660 def SLBG : BinaryRXY<"slbg", 0xE389, sube, GR64, load, 8>;
663 //===----------------------------------------------------------------------===//
665 //===----------------------------------------------------------------------===//
668 // ANDs of a register.
669 let isCommutable = 1 in {
670 defm NR : BinaryRRAndK<"n", 0x14, 0xB9F4, and, GR32, GR32>;
671 defm NGR : BinaryRREAndK<"ng", 0xB980, 0xB9E4, and, GR64, GR64>;
674 // ANDs of a 16-bit immediate, leaving other bits unaffected.
675 let isCodeGenOnly = 1 in {
676 def NILL32 : BinaryRI<"nill", 0xA57, and, GR32, imm32ll16c>;
677 def NILH32 : BinaryRI<"nilh", 0xA56, and, GR32, imm32lh16c>;
679 def NILL : BinaryRI<"nill", 0xA57, and, GR64, imm64ll16c>;
680 def NILH : BinaryRI<"nilh", 0xA56, and, GR64, imm64lh16c>;
681 def NIHL : BinaryRI<"nihl", 0xA55, and, GR64, imm64hl16c>;
682 def NIHH : BinaryRI<"nihh", 0xA54, and, GR64, imm64hh16c>;
684 // ANDs of a 32-bit immediate, leaving other bits unaffected.
685 let isCodeGenOnly = 1 in
686 def NILF32 : BinaryRIL<"nilf", 0xC0B, and, GR32, uimm32>;
687 def NILF : BinaryRIL<"nilf", 0xC0B, and, GR64, imm64lf32c>;
688 def NIHF : BinaryRIL<"nihf", 0xC0A, and, GR64, imm64hf32c>;
691 defm N : BinaryRXPair<"n", 0x54, 0xE354, and, GR32, load, 4>;
692 def NG : BinaryRXY<"ng", 0xE380, and, GR64, load, 8>;
695 defm NI : BinarySIPair<"ni", 0x94, 0xEB54, null_frag, uimm8>;
697 defm : RMWIByte<and, bdaddr12pair, NI>;
698 defm : RMWIByte<and, bdaddr20pair, NIY>;
700 //===----------------------------------------------------------------------===//
702 //===----------------------------------------------------------------------===//
705 // ORs of a register.
706 let isCommutable = 1 in {
707 defm OR : BinaryRRAndK<"o", 0x16, 0xB9F6, or, GR32, GR32>;
708 defm OGR : BinaryRREAndK<"og", 0xB981, 0xB9E6, or, GR64, GR64>;
711 // ORs of a 16-bit immediate, leaving other bits unaffected.
712 let isCodeGenOnly = 1 in {
713 def OILL32 : BinaryRI<"oill", 0xA5B, or, GR32, imm32ll16>;
714 def OILH32 : BinaryRI<"oilh", 0xA5A, or, GR32, imm32lh16>;
716 def OILL : BinaryRI<"oill", 0xA5B, or, GR64, imm64ll16>;
717 def OILH : BinaryRI<"oilh", 0xA5A, or, GR64, imm64lh16>;
718 def OIHL : BinaryRI<"oihl", 0xA59, or, GR64, imm64hl16>;
719 def OIHH : BinaryRI<"oihh", 0xA58, or, GR64, imm64hh16>;
721 // ORs of a 32-bit immediate, leaving other bits unaffected.
722 let isCodeGenOnly = 1 in
723 def OILF32 : BinaryRIL<"oilf", 0xC0D, or, GR32, uimm32>;
724 def OILF : BinaryRIL<"oilf", 0xC0D, or, GR64, imm64lf32>;
725 def OIHF : BinaryRIL<"oihf", 0xC0C, or, GR64, imm64hf32>;
728 defm O : BinaryRXPair<"o", 0x56, 0xE356, or, GR32, load, 4>;
729 def OG : BinaryRXY<"og", 0xE381, or, GR64, load, 8>;
732 defm OI : BinarySIPair<"oi", 0x96, 0xEB56, null_frag, uimm8>;
734 defm : RMWIByte<or, bdaddr12pair, OI>;
735 defm : RMWIByte<or, bdaddr20pair, OIY>;
737 //===----------------------------------------------------------------------===//
739 //===----------------------------------------------------------------------===//
742 // XORs of a register.
743 let isCommutable = 1 in {
744 defm XR : BinaryRRAndK<"x", 0x17, 0xB9F7, xor, GR32, GR32>;
745 defm XGR : BinaryRREAndK<"xg", 0xB982, 0xB9E7, xor, GR64, GR64>;
748 // XORs of a 32-bit immediate, leaving other bits unaffected.
749 let isCodeGenOnly = 1 in
750 def XILF32 : BinaryRIL<"xilf", 0xC07, xor, GR32, uimm32>;
751 def XILF : BinaryRIL<"xilf", 0xC07, xor, GR64, imm64lf32>;
752 def XIHF : BinaryRIL<"xihf", 0xC06, xor, GR64, imm64hf32>;
755 defm X : BinaryRXPair<"x",0x57, 0xE357, xor, GR32, load, 4>;
756 def XG : BinaryRXY<"xg", 0xE382, xor, GR64, load, 8>;
759 defm XI : BinarySIPair<"xi", 0x97, 0xEB57, null_frag, uimm8>;
761 defm : RMWIByte<xor, bdaddr12pair, XI>;
762 defm : RMWIByte<xor, bdaddr20pair, XIY>;
764 //===----------------------------------------------------------------------===//
766 //===----------------------------------------------------------------------===//
768 // Multiplication of a register.
769 let isCommutable = 1 in {
770 def MSR : BinaryRRE<"ms", 0xB252, mul, GR32, GR32>;
771 def MSGR : BinaryRRE<"msg", 0xB90C, mul, GR64, GR64>;
773 def MSGFR : BinaryRRE<"msgf", 0xB91C, null_frag, GR64, GR32>;
774 defm : SXB<mul, GR64, MSGFR>;
776 // Multiplication of a signed 16-bit immediate.
777 def MHI : BinaryRI<"mhi", 0xA7C, mul, GR32, imm32sx16>;
778 def MGHI : BinaryRI<"mghi", 0xA7D, mul, GR64, imm64sx16>;
780 // Multiplication of a signed 32-bit immediate.
781 def MSFI : BinaryRIL<"msfi", 0xC21, mul, GR32, simm32>;
782 def MSGFI : BinaryRIL<"msgfi", 0xC20, mul, GR64, imm64sx32>;
784 // Multiplication of memory.
785 defm MH : BinaryRXPair<"mh", 0x4C, 0xE37C, mul, GR32, sextloadi16, 2>;
786 defm MS : BinaryRXPair<"ms", 0x71, 0xE351, mul, GR32, load, 4>;
787 def MSGF : BinaryRXY<"msgf", 0xE31C, mul, GR64, sextloadi32, 4>;
788 def MSG : BinaryRXY<"msg", 0xE30C, mul, GR64, load, 8>;
790 // Multiplication of a register, producing two results.
791 def MLGR : BinaryRRE<"mlg", 0xB986, z_umul_lohi64, GR128, GR64>;
793 // Multiplication of memory, producing two results.
794 def MLG : BinaryRXY<"mlg", 0xE386, z_umul_lohi64, GR128, load, 8>;
796 //===----------------------------------------------------------------------===//
797 // Division and remainder
798 //===----------------------------------------------------------------------===//
800 // Division and remainder, from registers.
801 def DSGFR : BinaryRRE<"dsgf", 0xB91D, z_sdivrem32, GR128, GR32>;
802 def DSGR : BinaryRRE<"dsg", 0xB90D, z_sdivrem64, GR128, GR64>;
803 def DLR : BinaryRRE<"dl", 0xB997, z_udivrem32, GR128, GR32>;
804 def DLGR : BinaryRRE<"dlg", 0xB987, z_udivrem64, GR128, GR64>;
806 // Division and remainder, from memory.
807 def DSGF : BinaryRXY<"dsgf", 0xE31D, z_sdivrem32, GR128, load, 4>;
808 def DSG : BinaryRXY<"dsg", 0xE30D, z_sdivrem64, GR128, load, 8>;
809 def DL : BinaryRXY<"dl", 0xE397, z_udivrem32, GR128, load, 4>;
810 def DLG : BinaryRXY<"dlg", 0xE387, z_udivrem64, GR128, load, 8>;
812 //===----------------------------------------------------------------------===//
814 //===----------------------------------------------------------------------===//
817 let neverHasSideEffects = 1 in {
818 defm SLL : ShiftRSAndK<"sll", 0x89, 0xEBDF, shl, GR32>;
819 def SLLG : ShiftRSY<"sllg", 0xEB0D, shl, GR64>;
822 // Logical shift right.
823 let neverHasSideEffects = 1 in {
824 defm SRL : ShiftRSAndK<"srl", 0x88, 0xEBDE, srl, GR32>;
825 def SRLG : ShiftRSY<"srlg", 0xEB0C, srl, GR64>;
828 // Arithmetic shift right.
830 defm SRA : ShiftRSAndK<"sra", 0x8A, 0xEBDC, sra, GR32>;
831 def SRAG : ShiftRSY<"srag", 0xEB0A, sra, GR64>;
835 let neverHasSideEffects = 1 in {
836 def RLL : ShiftRSY<"rll", 0xEB1D, rotl, GR32>;
837 def RLLG : ShiftRSY<"rllg", 0xEB1C, rotl, GR64>;
840 // Rotate second operand left and inserted selected bits into first operand.
841 // These can act like 32-bit operands provided that the constant start and
842 // end bits (operands 2 and 3) are in the range [32, 64)
844 let isCodeGenOnly = 1 in
845 def RISBG32 : RotateSelectRIEf<"risbg", 0xEC55, GR32, GR32>;
846 def RISBG : RotateSelectRIEf<"risbg", 0xEC55, GR64, GR64>;
849 // Rotate second operand left and perform a logical operation with selected
850 // bits of the first operand.
852 def RNSBG : RotateSelectRIEf<"rnsbg", 0xEC54, GR64, GR64>;
853 def ROSBG : RotateSelectRIEf<"rosbg", 0xEC56, GR64, GR64>;
854 def RXSBG : RotateSelectRIEf<"rxsbg", 0xEC57, GR64, GR64>;
857 //===----------------------------------------------------------------------===//
859 //===----------------------------------------------------------------------===//
861 // Signed comparisons.
863 // Comparison with a register.
864 def CR : CompareRR <"c", 0x19, z_cmp, GR32, GR32>;
865 def CGFR : CompareRRE<"cgf", 0xB930, null_frag, GR64, GR32>;
866 def CGR : CompareRRE<"cg", 0xB920, z_cmp, GR64, GR64>;
868 // Comparison with a signed 16-bit immediate.
869 def CHI : CompareRI<"chi", 0xA7E, z_cmp, GR32, imm32sx16>;
870 def CGHI : CompareRI<"cghi", 0xA7F, z_cmp, GR64, imm64sx16>;
872 // Comparison with a signed 32-bit immediate.
873 def CFI : CompareRIL<"cfi", 0xC2D, z_cmp, GR32, simm32>;
874 def CGFI : CompareRIL<"cgfi", 0xC2C, z_cmp, GR64, imm64sx32>;
876 // Comparison with memory.
877 defm CH : CompareRXPair<"ch", 0x49, 0xE379, z_cmp, GR32, sextloadi16, 2>;
878 defm C : CompareRXPair<"c", 0x59, 0xE359, z_cmp, GR32, load, 4>;
879 def CGH : CompareRXY<"cgh", 0xE334, z_cmp, GR64, sextloadi16, 2>;
880 def CGF : CompareRXY<"cgf", 0xE330, z_cmp, GR64, sextloadi32, 4>;
881 def CG : CompareRXY<"cg", 0xE320, z_cmp, GR64, load, 8>;
882 def CHRL : CompareRILPC<"chrl", 0xC65, z_cmp, GR32, aligned_sextloadi16>;
883 def CRL : CompareRILPC<"crl", 0xC6D, z_cmp, GR32, aligned_load>;
884 def CGHRL : CompareRILPC<"cghrl", 0xC64, z_cmp, GR64, aligned_sextloadi16>;
885 def CGFRL : CompareRILPC<"cgfrl", 0xC6C, z_cmp, GR64, aligned_sextloadi32>;
886 def CGRL : CompareRILPC<"cgrl", 0xC68, z_cmp, GR64, aligned_load>;
888 // Comparison between memory and a signed 16-bit immediate.
889 def CHHSI : CompareSIL<"chhsi", 0xE554, z_cmp, sextloadi16, imm32sx16>;
890 def CHSI : CompareSIL<"chsi", 0xE55C, z_cmp, load, imm32sx16>;
891 def CGHSI : CompareSIL<"cghsi", 0xE558, z_cmp, load, imm64sx16>;
893 defm : SXB<z_cmp, GR64, CGFR>;
895 // Unsigned comparisons.
897 // Comparison with a register.
898 def CLR : CompareRR <"cl", 0x15, z_ucmp, GR32, GR32>;
899 def CLGFR : CompareRRE<"clgf", 0xB931, null_frag, GR64, GR32>;
900 def CLGR : CompareRRE<"clg", 0xB921, z_ucmp, GR64, GR64>;
902 // Comparison with a signed 32-bit immediate.
903 def CLFI : CompareRIL<"clfi", 0xC2F, z_ucmp, GR32, uimm32>;
904 def CLGFI : CompareRIL<"clgfi", 0xC2E, z_ucmp, GR64, imm64zx32>;
906 // Comparison with memory.
907 defm CL : CompareRXPair<"cl", 0x55, 0xE355, z_ucmp, GR32, load, 4>;
908 def CLGF : CompareRXY<"clgf", 0xE331, z_ucmp, GR64, zextloadi32, 4>;
909 def CLG : CompareRXY<"clg", 0xE321, z_ucmp, GR64, load, 8>;
910 def CLHRL : CompareRILPC<"clhrl", 0xC67, z_ucmp, GR32,
911 aligned_zextloadi16>;
912 def CLRL : CompareRILPC<"clrl", 0xC6F, z_ucmp, GR32,
914 def CLGHRL : CompareRILPC<"clghrl", 0xC66, z_ucmp, GR64,
915 aligned_zextloadi16>;
916 def CLGFRL : CompareRILPC<"clgfrl", 0xC6E, z_ucmp, GR64,
917 aligned_zextloadi32>;
918 def CLGRL : CompareRILPC<"clgrl", 0xC6A, z_ucmp, GR64,
921 // Comparison between memory and an unsigned 8-bit immediate.
922 defm CLI : CompareSIPair<"cli", 0x95, 0xEB55, z_ucmp, zextloadi8, imm32zx8>;
924 // Comparison between memory and an unsigned 16-bit immediate.
925 def CLHHSI : CompareSIL<"clhhsi", 0xE555, z_ucmp, zextloadi16, imm32zx16>;
926 def CLFHSI : CompareSIL<"clfhsi", 0xE55D, z_ucmp, load, imm32zx16>;
927 def CLGHSI : CompareSIL<"clghsi", 0xE559, z_ucmp, load, imm64zx16>;
929 defm : ZXB<z_ucmp, GR64, CLGFR>;
931 //===----------------------------------------------------------------------===//
933 //===----------------------------------------------------------------------===//
935 def ATOMIC_SWAPW : AtomicLoadWBinaryReg<z_atomic_swapw>;
936 def ATOMIC_SWAP_32 : AtomicLoadBinaryReg32<atomic_swap_32>;
937 def ATOMIC_SWAP_64 : AtomicLoadBinaryReg64<atomic_swap_64>;
939 def ATOMIC_LOADW_AR : AtomicLoadWBinaryReg<z_atomic_loadw_add>;
940 def ATOMIC_LOADW_AFI : AtomicLoadWBinaryImm<z_atomic_loadw_add, simm32>;
941 def ATOMIC_LOAD_AR : AtomicLoadBinaryReg32<atomic_load_add_32>;
942 def ATOMIC_LOAD_AHI : AtomicLoadBinaryImm32<atomic_load_add_32, imm32sx16>;
943 def ATOMIC_LOAD_AFI : AtomicLoadBinaryImm32<atomic_load_add_32, simm32>;
944 def ATOMIC_LOAD_AGR : AtomicLoadBinaryReg64<atomic_load_add_64>;
945 def ATOMIC_LOAD_AGHI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx16>;
946 def ATOMIC_LOAD_AGFI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx32>;
948 def ATOMIC_LOADW_SR : AtomicLoadWBinaryReg<z_atomic_loadw_sub>;
949 def ATOMIC_LOAD_SR : AtomicLoadBinaryReg32<atomic_load_sub_32>;
950 def ATOMIC_LOAD_SGR : AtomicLoadBinaryReg64<atomic_load_sub_64>;
952 def ATOMIC_LOADW_NR : AtomicLoadWBinaryReg<z_atomic_loadw_and>;
953 def ATOMIC_LOADW_NILH : AtomicLoadWBinaryImm<z_atomic_loadw_and, imm32lh16c>;
954 def ATOMIC_LOAD_NR : AtomicLoadBinaryReg32<atomic_load_and_32>;
955 def ATOMIC_LOAD_NILL32 : AtomicLoadBinaryImm32<atomic_load_and_32, imm32ll16c>;
956 def ATOMIC_LOAD_NILH32 : AtomicLoadBinaryImm32<atomic_load_and_32, imm32lh16c>;
957 def ATOMIC_LOAD_NILF32 : AtomicLoadBinaryImm32<atomic_load_and_32, uimm32>;
958 def ATOMIC_LOAD_NGR : AtomicLoadBinaryReg64<atomic_load_and_64>;
959 def ATOMIC_LOAD_NILL : AtomicLoadBinaryImm64<atomic_load_and_64, imm64ll16c>;
960 def ATOMIC_LOAD_NILH : AtomicLoadBinaryImm64<atomic_load_and_64, imm64lh16c>;
961 def ATOMIC_LOAD_NIHL : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hl16c>;
962 def ATOMIC_LOAD_NIHH : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hh16c>;
963 def ATOMIC_LOAD_NILF : AtomicLoadBinaryImm64<atomic_load_and_64, imm64lf32c>;
964 def ATOMIC_LOAD_NIHF : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hf32c>;
966 def ATOMIC_LOADW_OR : AtomicLoadWBinaryReg<z_atomic_loadw_or>;
967 def ATOMIC_LOADW_OILH : AtomicLoadWBinaryImm<z_atomic_loadw_or, imm32lh16>;
968 def ATOMIC_LOAD_OR : AtomicLoadBinaryReg32<atomic_load_or_32>;
969 def ATOMIC_LOAD_OILL32 : AtomicLoadBinaryImm32<atomic_load_or_32, imm32ll16>;
970 def ATOMIC_LOAD_OILH32 : AtomicLoadBinaryImm32<atomic_load_or_32, imm32lh16>;
971 def ATOMIC_LOAD_OILF32 : AtomicLoadBinaryImm32<atomic_load_or_32, uimm32>;
972 def ATOMIC_LOAD_OGR : AtomicLoadBinaryReg64<atomic_load_or_64>;
973 def ATOMIC_LOAD_OILL : AtomicLoadBinaryImm64<atomic_load_or_64, imm64ll16>;
974 def ATOMIC_LOAD_OILH : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lh16>;
975 def ATOMIC_LOAD_OIHL : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hl16>;
976 def ATOMIC_LOAD_OIHH : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hh16>;
977 def ATOMIC_LOAD_OILF : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lf32>;
978 def ATOMIC_LOAD_OIHF : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hf32>;
980 def ATOMIC_LOADW_XR : AtomicLoadWBinaryReg<z_atomic_loadw_xor>;
981 def ATOMIC_LOADW_XILF : AtomicLoadWBinaryImm<z_atomic_loadw_xor, uimm32>;
982 def ATOMIC_LOAD_XR : AtomicLoadBinaryReg32<atomic_load_xor_32>;
983 def ATOMIC_LOAD_XILF32 : AtomicLoadBinaryImm32<atomic_load_xor_32, uimm32>;
984 def ATOMIC_LOAD_XGR : AtomicLoadBinaryReg64<atomic_load_xor_64>;
985 def ATOMIC_LOAD_XILF : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64lf32>;
986 def ATOMIC_LOAD_XIHF : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64hf32>;
988 def ATOMIC_LOADW_NRi : AtomicLoadWBinaryReg<z_atomic_loadw_nand>;
989 def ATOMIC_LOADW_NILHi : AtomicLoadWBinaryImm<z_atomic_loadw_nand,
991 def ATOMIC_LOAD_NRi : AtomicLoadBinaryReg32<atomic_load_nand_32>;
992 def ATOMIC_LOAD_NILL32i : AtomicLoadBinaryImm32<atomic_load_nand_32,
994 def ATOMIC_LOAD_NILH32i : AtomicLoadBinaryImm32<atomic_load_nand_32,
996 def ATOMIC_LOAD_NILF32i : AtomicLoadBinaryImm32<atomic_load_nand_32, uimm32>;
997 def ATOMIC_LOAD_NGRi : AtomicLoadBinaryReg64<atomic_load_nand_64>;
998 def ATOMIC_LOAD_NILLi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1000 def ATOMIC_LOAD_NILHi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1002 def ATOMIC_LOAD_NIHLi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1004 def ATOMIC_LOAD_NIHHi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1006 def ATOMIC_LOAD_NILFi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1008 def ATOMIC_LOAD_NIHFi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1011 def ATOMIC_LOADW_MIN : AtomicLoadWBinaryReg<z_atomic_loadw_min>;
1012 def ATOMIC_LOAD_MIN_32 : AtomicLoadBinaryReg32<atomic_load_min_32>;
1013 def ATOMIC_LOAD_MIN_64 : AtomicLoadBinaryReg64<atomic_load_min_64>;
1015 def ATOMIC_LOADW_MAX : AtomicLoadWBinaryReg<z_atomic_loadw_max>;
1016 def ATOMIC_LOAD_MAX_32 : AtomicLoadBinaryReg32<atomic_load_max_32>;
1017 def ATOMIC_LOAD_MAX_64 : AtomicLoadBinaryReg64<atomic_load_max_64>;
1019 def ATOMIC_LOADW_UMIN : AtomicLoadWBinaryReg<z_atomic_loadw_umin>;
1020 def ATOMIC_LOAD_UMIN_32 : AtomicLoadBinaryReg32<atomic_load_umin_32>;
1021 def ATOMIC_LOAD_UMIN_64 : AtomicLoadBinaryReg64<atomic_load_umin_64>;
1023 def ATOMIC_LOADW_UMAX : AtomicLoadWBinaryReg<z_atomic_loadw_umax>;
1024 def ATOMIC_LOAD_UMAX_32 : AtomicLoadBinaryReg32<atomic_load_umax_32>;
1025 def ATOMIC_LOAD_UMAX_64 : AtomicLoadBinaryReg64<atomic_load_umax_64>;
1027 def ATOMIC_CMP_SWAPW
1028 : Pseudo<(outs GR32:$dst), (ins bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1029 ADDR32:$bitshift, ADDR32:$negbitshift,
1032 (z_atomic_cmp_swapw bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1033 ADDR32:$bitshift, ADDR32:$negbitshift,
1034 uimm32:$bitsize))]> {
1038 let usesCustomInserter = 1;
1041 let Defs = [CC] in {
1042 defm CS : CmpSwapRSPair<"cs", 0xBA, 0xEB14, atomic_cmp_swap_32, GR32>;
1043 def CSG : CmpSwapRSY<"csg", 0xEB30, atomic_cmp_swap_64, GR64>;
1046 //===----------------------------------------------------------------------===//
1047 // Miscellaneous Instructions.
1048 //===----------------------------------------------------------------------===//
1050 // Read a 32-bit access register into a GR32. As with all GR32 operations,
1051 // the upper 32 bits of the enclosing GR64 remain unchanged, which is useful
1052 // when a 64-bit address is stored in a pair of access registers.
1053 def EAR : InstRRE<0xB24F, (outs GR32:$R1), (ins access_reg:$R2),
1055 [(set GR32:$R1, (z_extract_access access_reg:$R2))]>;
1057 // Find leftmost one, AKA count leading zeros. The instruction actually
1058 // returns a pair of GR64s, the first giving the number of leading zeros
1059 // and the second giving a copy of the source with the leftmost one bit
1060 // cleared. We only use the first result here.
1061 let Defs = [CC] in {
1062 def FLOGR : UnaryRRE<"flog", 0xB983, null_frag, GR128, GR64>;
1064 def : Pat<(ctlz GR64:$src),
1065 (EXTRACT_SUBREG (FLOGR GR64:$src), subreg_high)>;
1067 // Use subregs to populate the "don't care" bits in a 32-bit to 64-bit anyext.
1068 def : Pat<(i64 (anyext GR32:$src)),
1069 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_32bit)>;
1071 // There are no 32-bit equivalents of LLILL and LLILH, so use a full
1072 // 64-bit move followed by a subreg. This preserves the invariant that
1073 // all GR32 operations only modify the low 32 bits.
1074 def : Pat<(i32 imm32ll16:$src),
1075 (EXTRACT_SUBREG (LLILL (LL16 imm:$src)), subreg_32bit)>;
1076 def : Pat<(i32 imm32lh16:$src),
1077 (EXTRACT_SUBREG (LLILH (LH16 imm:$src)), subreg_32bit)>;
1079 // Extend GR32s and GR64s to GR128s.
1080 let usesCustomInserter = 1 in {
1081 def AEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1082 def ZEXT128_32 : Pseudo<(outs GR128:$dst), (ins GR32:$src), []>;
1083 def ZEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1086 //===----------------------------------------------------------------------===//
1088 //===----------------------------------------------------------------------===//
1090 // Use AL* for GR64 additions of unsigned 32-bit values.
1091 defm : ZXB<add, GR64, ALGFR>;
1092 def : Pat<(add GR64:$src1, imm64zx32:$src2),
1093 (ALGFI GR64:$src1, imm64zx32:$src2)>;
1094 def : Pat<(add GR64:$src1, (zextloadi32 bdxaddr20only:$addr)),
1095 (ALGF GR64:$src1, bdxaddr20only:$addr)>;
1097 // Use SL* for GR64 subtractions of unsigned 32-bit values.
1098 defm : ZXB<sub, GR64, SLGFR>;
1099 def : Pat<(add GR64:$src1, imm64zx32n:$src2),
1100 (SLGFI GR64:$src1, imm64zx32n:$src2)>;
1101 def : Pat<(sub GR64:$src1, (zextloadi32 bdxaddr20only:$addr)),
1102 (SLGF GR64:$src1, bdxaddr20only:$addr)>;
1104 // Optimize sign-extended 1/0 selects to -1/0 selects. This is important
1105 // for vector legalization.
1106 def : Pat<(sra (shl (i32 (z_select_ccmask 1, 0, imm:$cc)), (i32 31)), (i32 31)),
1107 (Select32 (LHI -1), (LHI 0), imm:$cc)>;
1108 def : Pat<(sra (shl (i64 (anyext (i32 (z_select_ccmask 1, 0, imm:$cc)))),
1111 (Select64 (LGHI -1), (LGHI 0), imm:$cc)>;