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$I2", []>;
57 // Conditional branches. It's easier for LLVM to handle these branches
58 // in their raw BRC/BRCL form, with the 4-bit condition-code mask being
59 // the first operand. It seems friendlier to use mnemonic forms like
60 // JE and JLH when writing out the assembly though.
61 let isBranch = 1, isTerminator = 1, Uses = [CC] in {
62 let isCodeGenOnly = 1, CCMaskFirst = 1 in {
63 def BRC : InstRI<0xA74, (outs), (ins cond4:$valid, cond4:$R1,
64 brtarget16:$I2), "j$R1\t$I2",
65 [(z_br_ccmask cond4:$valid, cond4:$R1, bb:$I2)]>;
66 def BRCL : InstRIL<0xC04, (outs), (ins cond4:$valid, cond4:$R1,
67 brtarget32:$I2), "jg$R1\t$I2", []>;
69 def AsmBRC : InstRI<0xA74, (outs), (ins uimm8zx4:$R1, brtarget16:$I2),
71 def AsmBRCL : InstRIL<0xC04, (outs), (ins uimm8zx4:$R1, brtarget32:$I2),
72 "brcl\t$R1, $I2", []>;
75 // Fused compare-and-branch instructions. As for normal branches,
76 // we handle these instructions internally in their raw CRJ-like form,
77 // but use assembly macros like CRJE when writing them out.
79 // These instructions do not use or clobber the condition codes.
80 // We nevertheless pretend that they clobber CC, so that we can lower
81 // them to separate comparisons and BRCLs if the branch ends up being
83 multiclass CompareBranches<Operand ccmask, string pos1, string pos2> {
84 let isBranch = 1, isTerminator = 1, Defs = [CC] in {
85 def RJ : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2, ccmask:$M3,
87 "crj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
88 def GRJ : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2, ccmask:$M3,
90 "cgrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
91 def IJ : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2, ccmask:$M3,
93 "cij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
94 def GIJ : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2, ccmask:$M3,
96 "cgij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
99 let isCodeGenOnly = 1 in
100 defm C : CompareBranches<cond4, "$M3", "">;
101 defm AsmC : CompareBranches<uimm8zx4, "", "$M3, ">;
103 // Define AsmParser mnemonics for each general condition-code mask
104 // (integer or floating-point)
105 multiclass CondExtendedMnemonic<bits<4> ccmask, string name> {
107 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2),
108 "j"##name##"\t$I2", []>;
109 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2),
110 "jg"##name##"\t$I2", []>;
112 def LOCR : FixedCondUnaryRRF<"locr"##name, 0xB9F2, GR32, GR32, ccmask>;
113 def LOCGR : FixedCondUnaryRRF<"locgr"##name, 0xB9E2, GR64, GR64, ccmask>;
114 def LOC : FixedCondUnaryRSY<"loc"##name, 0xEBF2, GR32, ccmask, 4>;
115 def LOCG : FixedCondUnaryRSY<"locg"##name, 0xEBE2, GR64, ccmask, 8>;
116 def STOC : FixedCondStoreRSY<"stoc"##name, 0xEBF3, GR32, ccmask, 4>;
117 def STOCG : FixedCondStoreRSY<"stocg"##name, 0xEBE3, GR64, ccmask, 8>;
119 defm AsmO : CondExtendedMnemonic<1, "o">;
120 defm AsmH : CondExtendedMnemonic<2, "h">;
121 defm AsmNLE : CondExtendedMnemonic<3, "nle">;
122 defm AsmL : CondExtendedMnemonic<4, "l">;
123 defm AsmNHE : CondExtendedMnemonic<5, "nhe">;
124 defm AsmLH : CondExtendedMnemonic<6, "lh">;
125 defm AsmNE : CondExtendedMnemonic<7, "ne">;
126 defm AsmE : CondExtendedMnemonic<8, "e">;
127 defm AsmNLH : CondExtendedMnemonic<9, "nlh">;
128 defm AsmHE : CondExtendedMnemonic<10, "he">;
129 defm AsmNL : CondExtendedMnemonic<11, "nl">;
130 defm AsmLE : CondExtendedMnemonic<12, "le">;
131 defm AsmNH : CondExtendedMnemonic<13, "nh">;
132 defm AsmNO : CondExtendedMnemonic<14, "no">;
134 // Define AsmParser mnemonics for each integer condition-code mask.
135 // This is like the list above, except that condition 3 is not possible
136 // and that the low bit of the mask is therefore always 0. This means
137 // that each condition has two names. Conditions "o" and "no" are not used.
139 // We don't make one of the two names an alias of the other because
140 // we need the custom parsing routines to select the correct register class.
141 multiclass IntCondExtendedMnemonicA<bits<4> ccmask, string name> {
143 def CR : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2,
145 "crj"##name##"\t$R1, $R2, $RI4", []>;
146 def CGR : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2,
148 "cgrj"##name##"\t$R1, $R2, $RI4", []>;
149 def CI : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2,
151 "cij"##name##"\t$R1, $I2, $RI4", []>;
152 def CGI : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2,
154 "cgij"##name##"\t$R1, $I2, $RI4", []>;
157 multiclass IntCondExtendedMnemonic<bits<4> ccmask, string name1, string name2>
158 : IntCondExtendedMnemonicA<ccmask, name1> {
159 let isAsmParserOnly = 1 in
160 defm Alt : IntCondExtendedMnemonicA<ccmask, name2>;
162 defm AsmJH : IntCondExtendedMnemonic<2, "h", "nle">;
163 defm AsmJL : IntCondExtendedMnemonic<4, "l", "nhe">;
164 defm AsmJLH : IntCondExtendedMnemonic<6, "lh", "ne">;
165 defm AsmJE : IntCondExtendedMnemonic<8, "e", "nlh">;
166 defm AsmJHE : IntCondExtendedMnemonic<10, "he", "nl">;
167 defm AsmJLE : IntCondExtendedMnemonic<12, "le", "nh">;
169 // Decrement a register and branch if it is nonzero. These don't clobber CC,
170 // but we might need to split long branches into sequences that do.
172 def BRCT : BranchUnaryRI<"brct", 0xA76, GR32>;
173 def BRCTG : BranchUnaryRI<"brctg", 0xA77, GR64>;
176 //===----------------------------------------------------------------------===//
177 // Select instructions
178 //===----------------------------------------------------------------------===//
180 def Select32 : SelectWrapper<GR32>;
181 def Select64 : SelectWrapper<GR64>;
183 defm CondStore8_32 : CondStores<GR32, nonvolatile_truncstorei8,
184 nonvolatile_anyextloadi8, bdxaddr20only>;
185 defm CondStore16_32 : CondStores<GR32, nonvolatile_truncstorei16,
186 nonvolatile_anyextloadi16, bdxaddr20only>;
187 defm CondStore32_32 : CondStores<GR32, nonvolatile_store,
188 nonvolatile_load, bdxaddr20only>;
190 defm CondStore8 : CondStores<GR64, nonvolatile_truncstorei8,
191 nonvolatile_anyextloadi8, bdxaddr20only>;
192 defm CondStore16 : CondStores<GR64, nonvolatile_truncstorei16,
193 nonvolatile_anyextloadi16, bdxaddr20only>;
194 defm CondStore32 : CondStores<GR64, nonvolatile_truncstorei32,
195 nonvolatile_anyextloadi32, bdxaddr20only>;
196 defm CondStore64 : CondStores<GR64, nonvolatile_store,
197 nonvolatile_load, bdxaddr20only>;
199 //===----------------------------------------------------------------------===//
201 //===----------------------------------------------------------------------===//
203 // The definitions here are for the call-clobbered registers.
204 let isCall = 1, Defs = [R0D, R1D, R2D, R3D, R4D, R5D, R14D,
205 F0D, F1D, F2D, F3D, F4D, F5D, F6D, F7D, CC],
206 R1 = 14, isCodeGenOnly = 1 in {
207 def BRAS : InstRI<0xA75, (outs), (ins pcrel16call:$I2, variable_ops),
208 "bras\t%r14, $I2", []>;
209 def BRASL : InstRIL<0xC05, (outs), (ins pcrel32call:$I2, variable_ops),
210 "brasl\t%r14, $I2", [(z_call pcrel32call:$I2)]>;
211 def BASR : InstRR<0x0D, (outs), (ins ADDR64:$R2, variable_ops),
212 "basr\t%r14, $R2", [(z_call ADDR64:$R2)]>;
215 // Sibling calls. Indirect sibling calls must be via R1, since R2 upwards
216 // are argument registers and since branching to R0 is a no-op.
217 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1,
218 isCodeGenOnly = 1, R1 = 15 in {
219 def CallJG : InstRIL<0xC04, (outs), (ins pcrel32call:$I2),
220 "jg\t$I2", [(z_sibcall pcrel32call:$I2)]>;
221 let R2 = 1, Uses = [R1D] in
222 def CallBR : InstRR<0x07, (outs), (ins), "br\t%r1", [(z_sibcall R1D)]>;
225 // Define the general form of the call instructions for the asm parser.
226 // These instructions don't hard-code %r14 as the return address register.
227 def AsmBRAS : InstRI<0xA75, (outs), (ins GR64:$R1, brtarget16:$I2),
228 "bras\t$R1, $I2", []>;
229 def AsmBRASL : InstRIL<0xC05, (outs), (ins GR64:$R1, brtarget32:$I2),
230 "brasl\t$R1, $I2", []>;
231 def AsmBASR : InstRR<0x0D, (outs), (ins GR64:$R1, ADDR64:$R2),
232 "basr\t$R1, $R2", []>;
234 //===----------------------------------------------------------------------===//
236 //===----------------------------------------------------------------------===//
239 let neverHasSideEffects = 1 in {
240 def LR : UnaryRR <"l", 0x18, null_frag, GR32, GR32>;
241 def LGR : UnaryRRE<"lg", 0xB904, null_frag, GR64, GR64>;
243 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
244 def LTR : UnaryRR <"lt", 0x12, null_frag, GR32, GR32>;
245 def LTGR : UnaryRRE<"ltg", 0xB902, null_frag, GR64, GR64>;
248 // Move on condition.
249 let isCodeGenOnly = 1, Uses = [CC] in {
250 def LOCR : CondUnaryRRF<"loc", 0xB9F2, GR32, GR32>;
251 def LOCGR : CondUnaryRRF<"locg", 0xB9E2, GR64, GR64>;
254 def AsmLOCR : AsmCondUnaryRRF<"loc", 0xB9F2, GR32, GR32>;
255 def AsmLOCGR : AsmCondUnaryRRF<"locg", 0xB9E2, GR64, GR64>;
259 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
260 isReMaterializable = 1 in {
261 // 16-bit sign-extended immediates.
262 def LHI : UnaryRI<"lhi", 0xA78, bitconvert, GR32, imm32sx16>;
263 def LGHI : UnaryRI<"lghi", 0xA79, bitconvert, GR64, imm64sx16>;
265 // Other 16-bit immediates.
266 def LLILL : UnaryRI<"llill", 0xA5F, bitconvert, GR64, imm64ll16>;
267 def LLILH : UnaryRI<"llilh", 0xA5E, bitconvert, GR64, imm64lh16>;
268 def LLIHL : UnaryRI<"llihl", 0xA5D, bitconvert, GR64, imm64hl16>;
269 def LLIHH : UnaryRI<"llihh", 0xA5C, bitconvert, GR64, imm64hh16>;
271 // 32-bit immediates.
272 def LGFI : UnaryRIL<"lgfi", 0xC01, bitconvert, GR64, imm64sx32>;
273 def LLILF : UnaryRIL<"llilf", 0xC0F, bitconvert, GR64, imm64lf32>;
274 def LLIHF : UnaryRIL<"llihf", 0xC0E, bitconvert, GR64, imm64hf32>;
278 let canFoldAsLoad = 1, SimpleBDXLoad = 1 in {
279 defm L : UnaryRXPair<"l", 0x58, 0xE358, load, GR32, 4>;
280 def LG : UnaryRXY<"lg", 0xE304, load, GR64, 8>;
282 // These instructions are split after register allocation, so we don't
283 // want a custom inserter.
284 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
285 def L128 : Pseudo<(outs GR128:$dst), (ins bdxaddr20only128:$src),
286 [(set GR128:$dst, (load bdxaddr20only128:$src))]>;
289 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
290 def LT : UnaryRXY<"lt", 0xE312, load, GR32, 4>;
291 def LTG : UnaryRXY<"ltg", 0xE302, load, GR64, 8>;
294 let canFoldAsLoad = 1 in {
295 def LRL : UnaryRILPC<"lrl", 0xC4D, aligned_load, GR32>;
296 def LGRL : UnaryRILPC<"lgrl", 0xC48, aligned_load, GR64>;
299 // Load on condition.
300 let isCodeGenOnly = 1, Uses = [CC] in {
301 def LOC : CondUnaryRSY<"loc", 0xEBF2, nonvolatile_load, GR32, 4>;
302 def LOCG : CondUnaryRSY<"locg", 0xEBE2, nonvolatile_load, GR64, 8>;
305 def AsmLOC : AsmCondUnaryRSY<"loc", 0xEBF2, GR32, 4>;
306 def AsmLOCG : AsmCondUnaryRSY<"locg", 0xEBE2, GR64, 8>;
310 let SimpleBDXStore = 1 in {
311 let isCodeGenOnly = 1 in
312 defm ST32 : StoreRXPair<"st", 0x50, 0xE350, store, GR32, 4>;
313 def STG : StoreRXY<"stg", 0xE324, store, GR64, 8>;
315 // These instructions are split after register allocation, so we don't
316 // want a custom inserter.
317 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
318 def ST128 : Pseudo<(outs), (ins GR128:$src, bdxaddr20only128:$dst),
319 [(store GR128:$src, bdxaddr20only128:$dst)]>;
322 let isCodeGenOnly = 1 in
323 def STRL32 : StoreRILPC<"strl", 0xC4F, aligned_store, GR32>;
324 def STGRL : StoreRILPC<"stgrl", 0xC4B, aligned_store, GR64>;
326 // Store on condition.
327 let isCodeGenOnly = 1, Uses = [CC] in {
328 def STOC32 : CondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
329 def STOC : CondStoreRSY<"stoc", 0xEBF3, GR64, 4>;
330 def STOCG : CondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
333 def AsmSTOC : AsmCondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
334 def AsmSTOCG : AsmCondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
337 // 8-bit immediate stores to 8-bit fields.
338 defm MVI : StoreSIPair<"mvi", 0x92, 0xEB52, truncstorei8, imm32zx8trunc>;
340 // 16-bit immediate stores to 16-, 32- or 64-bit fields.
341 def MVHHI : StoreSIL<"mvhhi", 0xE544, truncstorei16, imm32sx16trunc>;
342 def MVHI : StoreSIL<"mvhi", 0xE54C, store, imm32sx16>;
343 def MVGHI : StoreSIL<"mvghi", 0xE548, store, imm64sx16>;
345 // Memory-to-memory moves.
346 let mayLoad = 1, mayStore = 1 in
347 defm MVC : MemorySS<"mvc", 0xD2, z_mvc, z_mvc_loop>;
350 let mayLoad = 1, mayStore = 1, Defs = [CC], Uses = [R0W] in
351 defm MVST : StringRRE<"mvst", 0xB255, z_stpcpy>;
353 defm LoadStore8_32 : MVCLoadStore<anyextloadi8, truncstorei8, i32,
355 defm LoadStore16_32 : MVCLoadStore<anyextloadi16, truncstorei16, i32,
357 defm LoadStore32_32 : MVCLoadStore<load, store, i32, MVCSequence, 4>;
359 defm LoadStore8 : MVCLoadStore<anyextloadi8, truncstorei8, i64,
361 defm LoadStore16 : MVCLoadStore<anyextloadi16, truncstorei16, i64,
363 defm LoadStore32 : MVCLoadStore<anyextloadi32, truncstorei32, i64,
365 defm LoadStore64 : MVCLoadStore<load, store, i64, MVCSequence, 8>;
367 //===----------------------------------------------------------------------===//
369 //===----------------------------------------------------------------------===//
371 // 32-bit extensions from registers.
372 let neverHasSideEffects = 1 in {
373 def LBR : UnaryRRE<"lb", 0xB926, sext8, GR32, GR32>;
374 def LHR : UnaryRRE<"lh", 0xB927, sext16, GR32, GR32>;
377 // 64-bit extensions from registers.
378 let neverHasSideEffects = 1 in {
379 def LGBR : UnaryRRE<"lgb", 0xB906, sext8, GR64, GR64>;
380 def LGHR : UnaryRRE<"lgh", 0xB907, sext16, GR64, GR64>;
381 def LGFR : UnaryRRE<"lgf", 0xB914, sext32, GR64, GR32>;
383 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in
384 def LTGFR : UnaryRRE<"ltgf", 0xB912, null_frag, GR64, GR64>;
386 // Match 32-to-64-bit sign extensions in which the source is already
387 // in a 64-bit register.
388 def : Pat<(sext_inreg GR64:$src, i32),
389 (LGFR (EXTRACT_SUBREG GR64:$src, subreg_32bit))>;
391 // 32-bit extensions from memory.
392 def LB : UnaryRXY<"lb", 0xE376, sextloadi8, GR32, 1>;
393 defm LH : UnaryRXPair<"lh", 0x48, 0xE378, sextloadi16, GR32, 2>;
394 def LHRL : UnaryRILPC<"lhrl", 0xC45, aligned_sextloadi16, GR32>;
396 // 64-bit extensions from memory.
397 def LGB : UnaryRXY<"lgb", 0xE377, sextloadi8, GR64, 1>;
398 def LGH : UnaryRXY<"lgh", 0xE315, sextloadi16, GR64, 2>;
399 def LGF : UnaryRXY<"lgf", 0xE314, sextloadi32, GR64, 4>;
400 def LGHRL : UnaryRILPC<"lghrl", 0xC44, aligned_sextloadi16, GR64>;
401 def LGFRL : UnaryRILPC<"lgfrl", 0xC4C, aligned_sextloadi32, GR64>;
402 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in
403 def LTGF : UnaryRXY<"ltgf", 0xE332, sextloadi32, GR64, 4>;
405 // If the sign of a load-extend operation doesn't matter, use the signed ones.
406 // There's not really much to choose between the sign and zero extensions,
407 // but LH is more compact than LLH for small offsets.
408 def : Pat<(i32 (extloadi8 bdxaddr20only:$src)), (LB bdxaddr20only:$src)>;
409 def : Pat<(i32 (extloadi16 bdxaddr12pair:$src)), (LH bdxaddr12pair:$src)>;
410 def : Pat<(i32 (extloadi16 bdxaddr20pair:$src)), (LHY bdxaddr20pair:$src)>;
412 def : Pat<(i64 (extloadi8 bdxaddr20only:$src)), (LGB bdxaddr20only:$src)>;
413 def : Pat<(i64 (extloadi16 bdxaddr20only:$src)), (LGH bdxaddr20only:$src)>;
414 def : Pat<(i64 (extloadi32 bdxaddr20only:$src)), (LGF bdxaddr20only:$src)>;
416 // We want PC-relative addresses to be tried ahead of BD and BDX addresses.
417 // However, BDXs have two extra operands and are therefore 6 units more
419 let AddedComplexity = 7 in {
420 def : Pat<(i32 (extloadi16 pcrel32:$src)), (LHRL pcrel32:$src)>;
421 def : Pat<(i64 (extloadi16 pcrel32:$src)), (LGHRL pcrel32:$src)>;
424 //===----------------------------------------------------------------------===//
426 //===----------------------------------------------------------------------===//
428 // 32-bit extensions from registers.
429 let neverHasSideEffects = 1 in {
430 def LLCR : UnaryRRE<"llc", 0xB994, zext8, GR32, GR32>;
431 def LLHR : UnaryRRE<"llh", 0xB995, zext16, GR32, GR32>;
434 // 64-bit extensions from registers.
435 let neverHasSideEffects = 1 in {
436 def LLGCR : UnaryRRE<"llgc", 0xB984, zext8, GR64, GR64>;
437 def LLGHR : UnaryRRE<"llgh", 0xB985, zext16, GR64, GR64>;
438 def LLGFR : UnaryRRE<"llgf", 0xB916, zext32, GR64, GR32>;
441 // Match 32-to-64-bit zero extensions in which the source is already
442 // in a 64-bit register.
443 def : Pat<(and GR64:$src, 0xffffffff),
444 (LLGFR (EXTRACT_SUBREG GR64:$src, subreg_32bit))>;
446 // 32-bit extensions from memory.
447 def LLC : UnaryRXY<"llc", 0xE394, zextloadi8, GR32, 1>;
448 def LLH : UnaryRXY<"llh", 0xE395, zextloadi16, GR32, 2>;
449 def LLHRL : UnaryRILPC<"llhrl", 0xC42, aligned_zextloadi16, GR32>;
451 // 64-bit extensions from memory.
452 def LLGC : UnaryRXY<"llgc", 0xE390, zextloadi8, GR64, 1>;
453 def LLGH : UnaryRXY<"llgh", 0xE391, zextloadi16, GR64, 2>;
454 def LLGF : UnaryRXY<"llgf", 0xE316, zextloadi32, GR64, 4>;
455 def LLGHRL : UnaryRILPC<"llghrl", 0xC46, aligned_zextloadi16, GR64>;
456 def LLGFRL : UnaryRILPC<"llgfrl", 0xC4E, aligned_zextloadi32, GR64>;
458 //===----------------------------------------------------------------------===//
460 //===----------------------------------------------------------------------===//
462 // Truncations of 64-bit registers to 32-bit registers.
463 def : Pat<(i32 (trunc GR64:$src)),
464 (EXTRACT_SUBREG GR64:$src, subreg_32bit)>;
466 // Truncations of 32-bit registers to memory.
467 let isCodeGenOnly = 1 in {
468 defm STC32 : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR32, 1>;
469 defm STH32 : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR32, 2>;
470 def STHRL32 : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR32>;
473 // Truncations of 64-bit registers to memory.
474 defm STC : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR64, 1>;
475 defm STH : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR64, 2>;
476 def STHRL : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR64>;
477 defm ST : StoreRXPair<"st", 0x50, 0xE350, truncstorei32, GR64, 4>;
478 def STRL : StoreRILPC<"strl", 0xC4F, aligned_truncstorei32, GR64>;
480 //===----------------------------------------------------------------------===//
481 // Multi-register moves
482 //===----------------------------------------------------------------------===//
484 // Multi-register loads.
485 def LMG : LoadMultipleRSY<"lmg", 0xEB04, GR64>;
487 // Multi-register stores.
488 def STMG : StoreMultipleRSY<"stmg", 0xEB24, GR64>;
490 //===----------------------------------------------------------------------===//
492 //===----------------------------------------------------------------------===//
494 // Byte-swapping register moves.
495 let neverHasSideEffects = 1 in {
496 def LRVR : UnaryRRE<"lrv", 0xB91F, bswap, GR32, GR32>;
497 def LRVGR : UnaryRRE<"lrvg", 0xB90F, bswap, GR64, GR64>;
500 // Byte-swapping loads. Unlike normal loads, these instructions are
501 // allowed to access storage more than once.
502 def LRV : UnaryRXY<"lrv", 0xE31E, loadu<bswap, nonvolatile_load>, GR32, 4>;
503 def LRVG : UnaryRXY<"lrvg", 0xE30F, loadu<bswap, nonvolatile_load>, GR64, 8>;
505 // Likewise byte-swapping stores.
506 def STRV : StoreRXY<"strv", 0xE33E, storeu<bswap, nonvolatile_store>, GR32, 4>;
507 def STRVG : StoreRXY<"strvg", 0xE32F, storeu<bswap, nonvolatile_store>,
510 //===----------------------------------------------------------------------===//
511 // Load address instructions
512 //===----------------------------------------------------------------------===//
514 // Load BDX-style addresses.
515 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isReMaterializable = 1,
517 let DispSize = "12" in
518 def LA : InstRX<0x41, (outs GR64:$R1), (ins laaddr12pair:$XBD2),
520 [(set GR64:$R1, laaddr12pair:$XBD2)]>;
521 let DispSize = "20" in
522 def LAY : InstRXY<0xE371, (outs GR64:$R1), (ins laaddr20pair:$XBD2),
524 [(set GR64:$R1, laaddr20pair:$XBD2)]>;
527 // Load a PC-relative address. There's no version of this instruction
528 // with a 16-bit offset, so there's no relaxation.
529 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
530 isReMaterializable = 1 in {
531 def LARL : InstRIL<0xC00, (outs GR64:$R1), (ins pcrel32:$I2),
533 [(set GR64:$R1, pcrel32:$I2)]>;
536 //===----------------------------------------------------------------------===//
537 // Absolute and Negation
538 //===----------------------------------------------------------------------===//
541 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
542 def LPR : UnaryRR <"lp", 0x10, z_iabs32, GR32, GR32>;
543 def LPGR : UnaryRRE<"lpg", 0xB900, z_iabs64, GR64, GR64>;
545 let CCValues = 0xE, CompareZeroCCMask = 0xE in
546 def LPGFR : UnaryRRE<"lpgf", 0xB910, null_frag, GR64, GR32>;
548 defm : SXU<z_iabs64, LPGFR>;
551 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
552 def LNR : UnaryRR <"ln", 0x11, z_inegabs32, GR32, GR32>;
553 def LNGR : UnaryRRE<"lng", 0xB901, z_inegabs64, GR64, GR64>;
555 let CCValues = 0xE, CompareZeroCCMask = 0xE in
556 def LNGFR : UnaryRRE<"lngf", 0xB911, null_frag, GR64, GR32>;
558 defm : SXU<z_inegabs64, LNGFR>;
561 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
562 def LCR : UnaryRR <"lc", 0x13, ineg, GR32, GR32>;
563 def LCGR : UnaryRRE<"lcg", 0xB903, ineg, GR64, GR64>;
565 let CCValues = 0xE, CompareZeroCCMask = 0xE in
566 def LCGFR : UnaryRRE<"lcgf", 0xB913, null_frag, GR64, GR32>;
568 defm : SXU<ineg, LCGFR>;
570 //===----------------------------------------------------------------------===//
572 //===----------------------------------------------------------------------===//
574 let isCodeGenOnly = 1 in
575 defm IC32 : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR32, zextloadi8, 1>;
576 defm IC : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR64, zextloadi8, 1>;
578 defm : InsertMem<"inserti8", IC32, GR32, zextloadi8, bdxaddr12pair>;
579 defm : InsertMem<"inserti8", IC32Y, GR32, zextloadi8, bdxaddr20pair>;
581 defm : InsertMem<"inserti8", IC, GR64, zextloadi8, bdxaddr12pair>;
582 defm : InsertMem<"inserti8", ICY, GR64, zextloadi8, bdxaddr20pair>;
584 // Insertions of a 16-bit immediate, leaving other bits unaffected.
585 // We don't have or_as_insert equivalents of these operations because
586 // OI is available instead.
587 let isCodeGenOnly = 1 in {
588 def IILL32 : BinaryRI<"iill", 0xA53, insertll, GR32, imm32ll16>;
589 def IILH32 : BinaryRI<"iilh", 0xA52, insertlh, GR32, imm32lh16>;
591 def IILL : BinaryRI<"iill", 0xA53, insertll, GR64, imm64ll16>;
592 def IILH : BinaryRI<"iilh", 0xA52, insertlh, GR64, imm64lh16>;
593 def IIHL : BinaryRI<"iihl", 0xA51, inserthl, GR64, imm64hl16>;
594 def IIHH : BinaryRI<"iihh", 0xA50, inserthh, GR64, imm64hh16>;
596 // ...likewise for 32-bit immediates. For GR32s this is a general
597 // full-width move. (We use IILF rather than something like LLILF
598 // for 32-bit moves because IILF leaves the upper 32 bits of the
600 let isCodeGenOnly = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
601 isReMaterializable = 1 in {
602 def IILF32 : UnaryRIL<"iilf", 0xC09, bitconvert, GR32, uimm32>;
604 def IILF : BinaryRIL<"iilf", 0xC09, insertlf, GR64, imm64lf32>;
605 def IIHF : BinaryRIL<"iihf", 0xC08, inserthf, GR64, imm64hf32>;
607 // An alternative model of inserthf, with the first operand being
608 // a zero-extended value.
609 def : Pat<(or (zext32 GR32:$src), imm64hf32:$imm),
610 (IIHF (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_32bit),
613 //===----------------------------------------------------------------------===//
615 //===----------------------------------------------------------------------===//
618 let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in {
619 // Addition of a register.
620 let isCommutable = 1 in {
621 defm AR : BinaryRRAndK<"a", 0x1A, 0xB9F8, add, GR32, GR32>;
622 defm AGR : BinaryRREAndK<"ag", 0xB908, 0xB9E8, add, GR64, GR64>;
624 def AGFR : BinaryRRE<"agf", 0xB918, null_frag, GR64, GR32>;
626 // Addition of signed 16-bit immediates.
627 defm AHI : BinaryRIAndK<"ahi", 0xA7A, 0xECD8, add, GR32, imm32sx16>;
628 defm AGHI : BinaryRIAndK<"aghi", 0xA7B, 0xECD9, add, GR64, imm64sx16>;
630 // Addition of signed 32-bit immediates.
631 def AFI : BinaryRIL<"afi", 0xC29, add, GR32, simm32>;
632 def AGFI : BinaryRIL<"agfi", 0xC28, add, GR64, imm64sx32>;
634 // Addition of memory.
635 defm AH : BinaryRXPair<"ah", 0x4A, 0xE37A, add, GR32, sextloadi16, 2>;
636 defm A : BinaryRXPair<"a", 0x5A, 0xE35A, add, GR32, load, 4>;
637 def AGF : BinaryRXY<"agf", 0xE318, add, GR64, sextloadi32, 4>;
638 def AG : BinaryRXY<"ag", 0xE308, add, GR64, load, 8>;
640 // Addition to memory.
641 def ASI : BinarySIY<"asi", 0xEB6A, add, imm32sx8>;
642 def AGSI : BinarySIY<"agsi", 0xEB7A, add, imm64sx8>;
644 defm : SXB<add, GR64, AGFR>;
646 // Addition producing a carry.
648 // Addition of a register.
649 let isCommutable = 1 in {
650 defm ALR : BinaryRRAndK<"al", 0x1E, 0xB9FA, addc, GR32, GR32>;
651 defm ALGR : BinaryRREAndK<"alg", 0xB90A, 0xB9EA, addc, GR64, GR64>;
653 def ALGFR : BinaryRRE<"algf", 0xB91A, null_frag, GR64, GR32>;
655 // Addition of signed 16-bit immediates.
656 def ALHSIK : BinaryRIE<"alhsik", 0xECDA, addc, GR32, imm32sx16>,
657 Requires<[FeatureDistinctOps]>;
658 def ALGHSIK : BinaryRIE<"alghsik", 0xECDB, addc, GR64, imm64sx16>,
659 Requires<[FeatureDistinctOps]>;
661 // Addition of unsigned 32-bit immediates.
662 def ALFI : BinaryRIL<"alfi", 0xC2B, addc, GR32, uimm32>;
663 def ALGFI : BinaryRIL<"algfi", 0xC2A, addc, GR64, imm64zx32>;
665 // Addition of memory.
666 defm AL : BinaryRXPair<"al", 0x5E, 0xE35E, addc, GR32, load, 4>;
667 def ALGF : BinaryRXY<"algf", 0xE31A, addc, GR64, zextloadi32, 4>;
668 def ALG : BinaryRXY<"alg", 0xE30A, addc, GR64, load, 8>;
670 defm : ZXB<addc, GR64, ALGFR>;
672 // Addition producing and using a carry.
673 let Defs = [CC], Uses = [CC] in {
674 // Addition of a register.
675 def ALCR : BinaryRRE<"alc", 0xB998, adde, GR32, GR32>;
676 def ALCGR : BinaryRRE<"alcg", 0xB988, adde, GR64, GR64>;
678 // Addition of memory.
679 def ALC : BinaryRXY<"alc", 0xE398, adde, GR32, load, 4>;
680 def ALCG : BinaryRXY<"alcg", 0xE388, adde, GR64, load, 8>;
683 //===----------------------------------------------------------------------===//
685 //===----------------------------------------------------------------------===//
687 // Plain substraction. Although immediate forms exist, we use the
688 // add-immediate instruction instead.
689 let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in {
690 // Subtraction of a register.
691 defm SR : BinaryRRAndK<"s", 0x1B, 0xB9F9, sub, GR32, GR32>;
692 def SGFR : BinaryRRE<"sgf", 0xB919, null_frag, GR64, GR32>;
693 defm SGR : BinaryRREAndK<"sg", 0xB909, 0xB9E9, sub, GR64, GR64>;
695 // Subtraction of memory.
696 defm SH : BinaryRXPair<"sh", 0x4B, 0xE37B, sub, GR32, sextloadi16, 2>;
697 defm S : BinaryRXPair<"s", 0x5B, 0xE35B, sub, GR32, load, 4>;
698 def SGF : BinaryRXY<"sgf", 0xE319, sub, GR64, sextloadi32, 4>;
699 def SG : BinaryRXY<"sg", 0xE309, sub, GR64, load, 8>;
701 defm : SXB<sub, GR64, SGFR>;
703 // Subtraction producing a carry.
705 // Subtraction of a register.
706 defm SLR : BinaryRRAndK<"sl", 0x1F, 0xB9FB, subc, GR32, GR32>;
707 def SLGFR : BinaryRRE<"slgf", 0xB91B, null_frag, GR64, GR32>;
708 defm SLGR : BinaryRREAndK<"slg", 0xB90B, 0xB9EB, subc, GR64, GR64>;
710 // Subtraction of unsigned 32-bit immediates. These don't match
711 // subc because we prefer addc for constants.
712 def SLFI : BinaryRIL<"slfi", 0xC25, null_frag, GR32, uimm32>;
713 def SLGFI : BinaryRIL<"slgfi", 0xC24, null_frag, GR64, imm64zx32>;
715 // Subtraction of memory.
716 defm SL : BinaryRXPair<"sl", 0x5F, 0xE35F, subc, GR32, load, 4>;
717 def SLGF : BinaryRXY<"slgf", 0xE31B, subc, GR64, zextloadi32, 4>;
718 def SLG : BinaryRXY<"slg", 0xE30B, subc, GR64, load, 8>;
720 defm : ZXB<subc, GR64, SLGFR>;
722 // Subtraction producing and using a carry.
723 let Defs = [CC], Uses = [CC] in {
724 // Subtraction of a register.
725 def SLBR : BinaryRRE<"slb", 0xB999, sube, GR32, GR32>;
726 def SLGBR : BinaryRRE<"slbg", 0xB989, sube, GR64, GR64>;
728 // Subtraction of memory.
729 def SLB : BinaryRXY<"slb", 0xE399, sube, GR32, load, 4>;
730 def SLBG : BinaryRXY<"slbg", 0xE389, sube, GR64, load, 8>;
733 //===----------------------------------------------------------------------===//
735 //===----------------------------------------------------------------------===//
738 // ANDs of a register.
739 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
740 defm NR : BinaryRRAndK<"n", 0x14, 0xB9F4, and, GR32, GR32>;
741 defm NGR : BinaryRREAndK<"ng", 0xB980, 0xB9E4, and, GR64, GR64>;
744 let isConvertibleToThreeAddress = 1 in {
745 // ANDs of a 16-bit immediate, leaving other bits unaffected.
746 // The CC result only reflects the 16-bit field, not the full register.
747 let isCodeGenOnly = 1 in {
748 def NILL32 : BinaryRI<"nill", 0xA57, and, GR32, imm32ll16c>;
749 def NILH32 : BinaryRI<"nilh", 0xA56, and, GR32, imm32lh16c>;
751 def NILL : BinaryRI<"nill", 0xA57, and, GR64, imm64ll16c>;
752 def NILH : BinaryRI<"nilh", 0xA56, and, GR64, imm64lh16c>;
753 def NIHL : BinaryRI<"nihl", 0xA55, and, GR64, imm64hl16c>;
754 def NIHH : BinaryRI<"nihh", 0xA54, and, GR64, imm64hh16c>;
756 // ANDs of a 32-bit immediate, leaving other bits unaffected.
757 // The CC result only reflects the 32-bit field, which means we can
758 // use it as a zero indicator for i32 operations but not otherwise.
759 let isCodeGenOnly = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in
760 def NILF32 : BinaryRIL<"nilf", 0xC0B, and, GR32, uimm32>;
761 def NILF : BinaryRIL<"nilf", 0xC0B, and, GR64, imm64lf32c>;
762 def NIHF : BinaryRIL<"nihf", 0xC0A, and, GR64, imm64hf32c>;
766 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
767 defm N : BinaryRXPair<"n", 0x54, 0xE354, and, GR32, load, 4>;
768 def NG : BinaryRXY<"ng", 0xE380, and, GR64, load, 8>;
772 defm NI : BinarySIPair<"ni", 0x94, 0xEB54, null_frag, uimm8>;
774 defm : RMWIByte<and, bdaddr12pair, NI>;
775 defm : RMWIByte<and, bdaddr20pair, NIY>;
777 //===----------------------------------------------------------------------===//
779 //===----------------------------------------------------------------------===//
782 // ORs of a register.
783 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
784 defm OR : BinaryRRAndK<"o", 0x16, 0xB9F6, or, GR32, GR32>;
785 defm OGR : BinaryRREAndK<"og", 0xB981, 0xB9E6, or, GR64, GR64>;
788 // ORs of a 16-bit immediate, leaving other bits unaffected.
789 // The CC result only reflects the 16-bit field, not the full register.
790 let isCodeGenOnly = 1 in {
791 def OILL32 : BinaryRI<"oill", 0xA5B, or, GR32, imm32ll16>;
792 def OILH32 : BinaryRI<"oilh", 0xA5A, or, GR32, imm32lh16>;
794 def OILL : BinaryRI<"oill", 0xA5B, or, GR64, imm64ll16>;
795 def OILH : BinaryRI<"oilh", 0xA5A, or, GR64, imm64lh16>;
796 def OIHL : BinaryRI<"oihl", 0xA59, or, GR64, imm64hl16>;
797 def OIHH : BinaryRI<"oihh", 0xA58, or, GR64, imm64hh16>;
799 // ORs of a 32-bit immediate, leaving other bits unaffected.
800 // The CC result only reflects the 32-bit field, which means we can
801 // use it as a zero indicator for i32 operations but not otherwise.
802 let isCodeGenOnly = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in
803 def OILF32 : BinaryRIL<"oilf", 0xC0D, or, GR32, uimm32>;
804 def OILF : BinaryRIL<"oilf", 0xC0D, or, GR64, imm64lf32>;
805 def OIHF : BinaryRIL<"oihf", 0xC0C, or, GR64, imm64hf32>;
808 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
809 defm O : BinaryRXPair<"o", 0x56, 0xE356, or, GR32, load, 4>;
810 def OG : BinaryRXY<"og", 0xE381, or, GR64, load, 8>;
814 defm OI : BinarySIPair<"oi", 0x96, 0xEB56, null_frag, uimm8>;
816 defm : RMWIByte<or, bdaddr12pair, OI>;
817 defm : RMWIByte<or, bdaddr20pair, OIY>;
819 //===----------------------------------------------------------------------===//
821 //===----------------------------------------------------------------------===//
824 // XORs of a register.
825 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
826 defm XR : BinaryRRAndK<"x", 0x17, 0xB9F7, xor, GR32, GR32>;
827 defm XGR : BinaryRREAndK<"xg", 0xB982, 0xB9E7, xor, GR64, GR64>;
830 // XORs of a 32-bit immediate, leaving other bits unaffected.
831 // The CC result only reflects the 32-bit field, which means we can
832 // use it as a zero indicator for i32 operations but not otherwise.
833 let isCodeGenOnly = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in
834 def XILF32 : BinaryRIL<"xilf", 0xC07, xor, GR32, uimm32>;
835 def XILF : BinaryRIL<"xilf", 0xC07, xor, GR64, imm64lf32>;
836 def XIHF : BinaryRIL<"xihf", 0xC06, xor, GR64, imm64hf32>;
839 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
840 defm X : BinaryRXPair<"x",0x57, 0xE357, xor, GR32, load, 4>;
841 def XG : BinaryRXY<"xg", 0xE382, xor, GR64, load, 8>;
845 defm XI : BinarySIPair<"xi", 0x97, 0xEB57, null_frag, uimm8>;
847 defm : RMWIByte<xor, bdaddr12pair, XI>;
848 defm : RMWIByte<xor, bdaddr20pair, XIY>;
850 //===----------------------------------------------------------------------===//
852 //===----------------------------------------------------------------------===//
854 // Multiplication of a register.
855 let isCommutable = 1 in {
856 def MSR : BinaryRRE<"ms", 0xB252, mul, GR32, GR32>;
857 def MSGR : BinaryRRE<"msg", 0xB90C, mul, GR64, GR64>;
859 def MSGFR : BinaryRRE<"msgf", 0xB91C, null_frag, GR64, GR32>;
860 defm : SXB<mul, GR64, MSGFR>;
862 // Multiplication of a signed 16-bit immediate.
863 def MHI : BinaryRI<"mhi", 0xA7C, mul, GR32, imm32sx16>;
864 def MGHI : BinaryRI<"mghi", 0xA7D, mul, GR64, imm64sx16>;
866 // Multiplication of a signed 32-bit immediate.
867 def MSFI : BinaryRIL<"msfi", 0xC21, mul, GR32, simm32>;
868 def MSGFI : BinaryRIL<"msgfi", 0xC20, mul, GR64, imm64sx32>;
870 // Multiplication of memory.
871 defm MH : BinaryRXPair<"mh", 0x4C, 0xE37C, mul, GR32, sextloadi16, 2>;
872 defm MS : BinaryRXPair<"ms", 0x71, 0xE351, mul, GR32, load, 4>;
873 def MSGF : BinaryRXY<"msgf", 0xE31C, mul, GR64, sextloadi32, 4>;
874 def MSG : BinaryRXY<"msg", 0xE30C, mul, GR64, load, 8>;
876 // Multiplication of a register, producing two results.
877 def MLGR : BinaryRRE<"mlg", 0xB986, z_umul_lohi64, GR128, GR64>;
879 // Multiplication of memory, producing two results.
880 def MLG : BinaryRXY<"mlg", 0xE386, z_umul_lohi64, GR128, load, 8>;
882 //===----------------------------------------------------------------------===//
883 // Division and remainder
884 //===----------------------------------------------------------------------===//
886 // Division and remainder, from registers.
887 def DSGFR : BinaryRRE<"dsgf", 0xB91D, z_sdivrem32, GR128, GR32>;
888 def DSGR : BinaryRRE<"dsg", 0xB90D, z_sdivrem64, GR128, GR64>;
889 def DLR : BinaryRRE<"dl", 0xB997, z_udivrem32, GR128, GR32>;
890 def DLGR : BinaryRRE<"dlg", 0xB987, z_udivrem64, GR128, GR64>;
892 // Division and remainder, from memory.
893 def DSGF : BinaryRXY<"dsgf", 0xE31D, z_sdivrem32, GR128, load, 4>;
894 def DSG : BinaryRXY<"dsg", 0xE30D, z_sdivrem64, GR128, load, 8>;
895 def DL : BinaryRXY<"dl", 0xE397, z_udivrem32, GR128, load, 4>;
896 def DLG : BinaryRXY<"dlg", 0xE387, z_udivrem64, GR128, load, 8>;
898 //===----------------------------------------------------------------------===//
900 //===----------------------------------------------------------------------===//
903 let neverHasSideEffects = 1 in {
904 defm SLL : ShiftRSAndK<"sll", 0x89, 0xEBDF, shl, GR32>;
905 def SLLG : ShiftRSY<"sllg", 0xEB0D, shl, GR64>;
908 // Logical shift right.
909 let neverHasSideEffects = 1 in {
910 defm SRL : ShiftRSAndK<"srl", 0x88, 0xEBDE, srl, GR32>;
911 def SRLG : ShiftRSY<"srlg", 0xEB0C, srl, GR64>;
914 // Arithmetic shift right.
915 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
916 defm SRA : ShiftRSAndK<"sra", 0x8A, 0xEBDC, sra, GR32>;
917 def SRAG : ShiftRSY<"srag", 0xEB0A, sra, GR64>;
921 let neverHasSideEffects = 1 in {
922 def RLL : ShiftRSY<"rll", 0xEB1D, rotl, GR32>;
923 def RLLG : ShiftRSY<"rllg", 0xEB1C, rotl, GR64>;
926 // Rotate second operand left and inserted selected bits into first operand.
927 // These can act like 32-bit operands provided that the constant start and
928 // end bits (operands 2 and 3) are in the range [32, 64).
930 let isCodeGenOnly = 1 in
931 def RISBG32 : RotateSelectRIEf<"risbg", 0xEC55, GR32, GR32>;
932 let CCValues = 0xE, CompareZeroCCMask = 0xE in
933 def RISBG : RotateSelectRIEf<"risbg", 0xEC55, GR64, GR64>;
936 // Forms of RISBG that only affect one word of the destination register.
937 // They do not set CC.
938 let isCodeGenOnly = 1 in
939 def RISBLG32 : RotateSelectRIEf<"risblg", 0xEC51, GR32, GR32>,
940 Requires<[FeatureHighWord]>;
941 def RISBHG : RotateSelectRIEf<"risbhg", 0xEC5D, GR64, GR64>,
942 Requires<[FeatureHighWord]>;
943 def RISBLG : RotateSelectRIEf<"risblg", 0xEC51, GR64, GR64>,
944 Requires<[FeatureHighWord]>;
946 // Rotate second operand left and perform a logical operation with selected
947 // bits of the first operand. The CC result only describes the selected bits,
948 // so isn't useful for a full comparison against zero.
950 def RNSBG : RotateSelectRIEf<"rnsbg", 0xEC54, GR64, GR64>;
951 def ROSBG : RotateSelectRIEf<"rosbg", 0xEC56, GR64, GR64>;
952 def RXSBG : RotateSelectRIEf<"rxsbg", 0xEC57, GR64, GR64>;
955 //===----------------------------------------------------------------------===//
957 //===----------------------------------------------------------------------===//
959 // Signed comparisons.
960 let Defs = [CC], CCValues = 0xE in {
961 // Comparison with a register.
962 def CR : CompareRR <"c", 0x19, z_cmp, GR32, GR32>;
963 def CGFR : CompareRRE<"cgf", 0xB930, null_frag, GR64, GR32>;
964 def CGR : CompareRRE<"cg", 0xB920, z_cmp, GR64, GR64>;
966 // Comparison with a signed 16-bit immediate.
967 def CHI : CompareRI<"chi", 0xA7E, z_cmp, GR32, imm32sx16>;
968 def CGHI : CompareRI<"cghi", 0xA7F, z_cmp, GR64, imm64sx16>;
970 // Comparison with a signed 32-bit immediate.
971 def CFI : CompareRIL<"cfi", 0xC2D, z_cmp, GR32, simm32>;
972 def CGFI : CompareRIL<"cgfi", 0xC2C, z_cmp, GR64, imm64sx32>;
974 // Comparison with memory.
975 defm CH : CompareRXPair<"ch", 0x49, 0xE379, z_cmp, GR32, sextloadi16, 2>;
976 defm C : CompareRXPair<"c", 0x59, 0xE359, z_cmp, GR32, load, 4>;
977 def CGH : CompareRXY<"cgh", 0xE334, z_cmp, GR64, sextloadi16, 2>;
978 def CGF : CompareRXY<"cgf", 0xE330, z_cmp, GR64, sextloadi32, 4>;
979 def CG : CompareRXY<"cg", 0xE320, z_cmp, GR64, load, 8>;
980 def CHRL : CompareRILPC<"chrl", 0xC65, z_cmp, GR32, aligned_sextloadi16>;
981 def CRL : CompareRILPC<"crl", 0xC6D, z_cmp, GR32, aligned_load>;
982 def CGHRL : CompareRILPC<"cghrl", 0xC64, z_cmp, GR64, aligned_sextloadi16>;
983 def CGFRL : CompareRILPC<"cgfrl", 0xC6C, z_cmp, GR64, aligned_sextloadi32>;
984 def CGRL : CompareRILPC<"cgrl", 0xC68, z_cmp, GR64, aligned_load>;
986 // Comparison between memory and a signed 16-bit immediate.
987 def CHHSI : CompareSIL<"chhsi", 0xE554, z_cmp, sextloadi16, imm32sx16>;
988 def CHSI : CompareSIL<"chsi", 0xE55C, z_cmp, load, imm32sx16>;
989 def CGHSI : CompareSIL<"cghsi", 0xE558, z_cmp, load, imm64sx16>;
991 defm : SXB<z_cmp, GR64, CGFR>;
993 // Unsigned comparisons.
994 let Defs = [CC], CCValues = 0xE, IsLogical = 1 in {
995 // Comparison with a register.
996 def CLR : CompareRR <"cl", 0x15, z_ucmp, GR32, GR32>;
997 def CLGFR : CompareRRE<"clgf", 0xB931, null_frag, GR64, GR32>;
998 def CLGR : CompareRRE<"clg", 0xB921, z_ucmp, GR64, GR64>;
1000 // Comparison with a signed 32-bit immediate.
1001 def CLFI : CompareRIL<"clfi", 0xC2F, z_ucmp, GR32, uimm32>;
1002 def CLGFI : CompareRIL<"clgfi", 0xC2E, z_ucmp, GR64, imm64zx32>;
1004 // Comparison with memory.
1005 defm CL : CompareRXPair<"cl", 0x55, 0xE355, z_ucmp, GR32, load, 4>;
1006 def CLGF : CompareRXY<"clgf", 0xE331, z_ucmp, GR64, zextloadi32, 4>;
1007 def CLG : CompareRXY<"clg", 0xE321, z_ucmp, GR64, load, 8>;
1008 def CLHRL : CompareRILPC<"clhrl", 0xC67, z_ucmp, GR32,
1009 aligned_zextloadi16>;
1010 def CLRL : CompareRILPC<"clrl", 0xC6F, z_ucmp, GR32,
1012 def CLGHRL : CompareRILPC<"clghrl", 0xC66, z_ucmp, GR64,
1013 aligned_zextloadi16>;
1014 def CLGFRL : CompareRILPC<"clgfrl", 0xC6E, z_ucmp, GR64,
1015 aligned_zextloadi32>;
1016 def CLGRL : CompareRILPC<"clgrl", 0xC6A, z_ucmp, GR64,
1019 // Comparison between memory and an unsigned 8-bit immediate.
1020 defm CLI : CompareSIPair<"cli", 0x95, 0xEB55, z_ucmp, zextloadi8, imm32zx8>;
1022 // Comparison between memory and an unsigned 16-bit immediate.
1023 def CLHHSI : CompareSIL<"clhhsi", 0xE555, z_ucmp, zextloadi16, imm32zx16>;
1024 def CLFHSI : CompareSIL<"clfhsi", 0xE55D, z_ucmp, load, imm32zx16>;
1025 def CLGHSI : CompareSIL<"clghsi", 0xE559, z_ucmp, load, imm64zx16>;
1027 defm : ZXB<z_ucmp, GR64, CLGFR>;
1029 // Memory-to-memory comparison.
1030 let mayLoad = 1, Defs = [CC] in
1031 defm CLC : MemorySS<"clc", 0xD5, z_clc, z_clc_loop>;
1033 // String comparison.
1034 let mayLoad = 1, Defs = [CC], Uses = [R0W] in
1035 defm CLST : StringRRE<"clst", 0xB25D, z_strcmp>;
1038 let Defs = [CC] in {
1039 let isCodeGenOnly = 1 in {
1040 def TMLL32 : CompareRI<"tmll", 0xA71, z_tm, GR32, imm32ll16>;
1041 def TMLH32 : CompareRI<"tmlh", 0xA70, z_tm, GR32, imm32lh16>;
1044 def TMLL : CompareRI<"tmll", 0xA71, z_tm, GR64, imm64ll16>;
1045 def TMLH : CompareRI<"tmlh", 0xA70, z_tm, GR64, imm64lh16>;
1046 def TMHL : CompareRI<"tmhl", 0xA73, z_tm, GR64, imm64hl16>;
1047 def TMHH : CompareRI<"tmhh", 0xA72, z_tm, GR64, imm64hh16>;
1050 //===----------------------------------------------------------------------===//
1052 //===----------------------------------------------------------------------===//
1054 def PFD : PrefetchRXY<"pfd", 0xE336, z_prefetch>;
1055 def PFDRL : PrefetchRILPC<"pfdrl", 0xC62, z_prefetch>;
1057 //===----------------------------------------------------------------------===//
1058 // Atomic operations
1059 //===----------------------------------------------------------------------===//
1061 def ATOMIC_SWAPW : AtomicLoadWBinaryReg<z_atomic_swapw>;
1062 def ATOMIC_SWAP_32 : AtomicLoadBinaryReg32<atomic_swap_32>;
1063 def ATOMIC_SWAP_64 : AtomicLoadBinaryReg64<atomic_swap_64>;
1065 def ATOMIC_LOADW_AR : AtomicLoadWBinaryReg<z_atomic_loadw_add>;
1066 def ATOMIC_LOADW_AFI : AtomicLoadWBinaryImm<z_atomic_loadw_add, simm32>;
1067 def ATOMIC_LOAD_AR : AtomicLoadBinaryReg32<atomic_load_add_32>;
1068 def ATOMIC_LOAD_AHI : AtomicLoadBinaryImm32<atomic_load_add_32, imm32sx16>;
1069 def ATOMIC_LOAD_AFI : AtomicLoadBinaryImm32<atomic_load_add_32, simm32>;
1070 def ATOMIC_LOAD_AGR : AtomicLoadBinaryReg64<atomic_load_add_64>;
1071 def ATOMIC_LOAD_AGHI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx16>;
1072 def ATOMIC_LOAD_AGFI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx32>;
1074 def ATOMIC_LOADW_SR : AtomicLoadWBinaryReg<z_atomic_loadw_sub>;
1075 def ATOMIC_LOAD_SR : AtomicLoadBinaryReg32<atomic_load_sub_32>;
1076 def ATOMIC_LOAD_SGR : AtomicLoadBinaryReg64<atomic_load_sub_64>;
1078 def ATOMIC_LOADW_NR : AtomicLoadWBinaryReg<z_atomic_loadw_and>;
1079 def ATOMIC_LOADW_NILH : AtomicLoadWBinaryImm<z_atomic_loadw_and, imm32lh16c>;
1080 def ATOMIC_LOAD_NR : AtomicLoadBinaryReg32<atomic_load_and_32>;
1081 def ATOMIC_LOAD_NILL32 : AtomicLoadBinaryImm32<atomic_load_and_32, imm32ll16c>;
1082 def ATOMIC_LOAD_NILH32 : AtomicLoadBinaryImm32<atomic_load_and_32, imm32lh16c>;
1083 def ATOMIC_LOAD_NILF32 : AtomicLoadBinaryImm32<atomic_load_and_32, uimm32>;
1084 def ATOMIC_LOAD_NGR : AtomicLoadBinaryReg64<atomic_load_and_64>;
1085 def ATOMIC_LOAD_NILL : AtomicLoadBinaryImm64<atomic_load_and_64, imm64ll16c>;
1086 def ATOMIC_LOAD_NILH : AtomicLoadBinaryImm64<atomic_load_and_64, imm64lh16c>;
1087 def ATOMIC_LOAD_NIHL : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hl16c>;
1088 def ATOMIC_LOAD_NIHH : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hh16c>;
1089 def ATOMIC_LOAD_NILF : AtomicLoadBinaryImm64<atomic_load_and_64, imm64lf32c>;
1090 def ATOMIC_LOAD_NIHF : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hf32c>;
1092 def ATOMIC_LOADW_OR : AtomicLoadWBinaryReg<z_atomic_loadw_or>;
1093 def ATOMIC_LOADW_OILH : AtomicLoadWBinaryImm<z_atomic_loadw_or, imm32lh16>;
1094 def ATOMIC_LOAD_OR : AtomicLoadBinaryReg32<atomic_load_or_32>;
1095 def ATOMIC_LOAD_OILL32 : AtomicLoadBinaryImm32<atomic_load_or_32, imm32ll16>;
1096 def ATOMIC_LOAD_OILH32 : AtomicLoadBinaryImm32<atomic_load_or_32, imm32lh16>;
1097 def ATOMIC_LOAD_OILF32 : AtomicLoadBinaryImm32<atomic_load_or_32, uimm32>;
1098 def ATOMIC_LOAD_OGR : AtomicLoadBinaryReg64<atomic_load_or_64>;
1099 def ATOMIC_LOAD_OILL : AtomicLoadBinaryImm64<atomic_load_or_64, imm64ll16>;
1100 def ATOMIC_LOAD_OILH : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lh16>;
1101 def ATOMIC_LOAD_OIHL : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hl16>;
1102 def ATOMIC_LOAD_OIHH : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hh16>;
1103 def ATOMIC_LOAD_OILF : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lf32>;
1104 def ATOMIC_LOAD_OIHF : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hf32>;
1106 def ATOMIC_LOADW_XR : AtomicLoadWBinaryReg<z_atomic_loadw_xor>;
1107 def ATOMIC_LOADW_XILF : AtomicLoadWBinaryImm<z_atomic_loadw_xor, uimm32>;
1108 def ATOMIC_LOAD_XR : AtomicLoadBinaryReg32<atomic_load_xor_32>;
1109 def ATOMIC_LOAD_XILF32 : AtomicLoadBinaryImm32<atomic_load_xor_32, uimm32>;
1110 def ATOMIC_LOAD_XGR : AtomicLoadBinaryReg64<atomic_load_xor_64>;
1111 def ATOMIC_LOAD_XILF : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64lf32>;
1112 def ATOMIC_LOAD_XIHF : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64hf32>;
1114 def ATOMIC_LOADW_NRi : AtomicLoadWBinaryReg<z_atomic_loadw_nand>;
1115 def ATOMIC_LOADW_NILHi : AtomicLoadWBinaryImm<z_atomic_loadw_nand,
1117 def ATOMIC_LOAD_NRi : AtomicLoadBinaryReg32<atomic_load_nand_32>;
1118 def ATOMIC_LOAD_NILL32i : AtomicLoadBinaryImm32<atomic_load_nand_32,
1120 def ATOMIC_LOAD_NILH32i : AtomicLoadBinaryImm32<atomic_load_nand_32,
1122 def ATOMIC_LOAD_NILF32i : AtomicLoadBinaryImm32<atomic_load_nand_32, uimm32>;
1123 def ATOMIC_LOAD_NGRi : AtomicLoadBinaryReg64<atomic_load_nand_64>;
1124 def ATOMIC_LOAD_NILLi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1126 def ATOMIC_LOAD_NILHi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1128 def ATOMIC_LOAD_NIHLi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1130 def ATOMIC_LOAD_NIHHi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1132 def ATOMIC_LOAD_NILFi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1134 def ATOMIC_LOAD_NIHFi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1137 def ATOMIC_LOADW_MIN : AtomicLoadWBinaryReg<z_atomic_loadw_min>;
1138 def ATOMIC_LOAD_MIN_32 : AtomicLoadBinaryReg32<atomic_load_min_32>;
1139 def ATOMIC_LOAD_MIN_64 : AtomicLoadBinaryReg64<atomic_load_min_64>;
1141 def ATOMIC_LOADW_MAX : AtomicLoadWBinaryReg<z_atomic_loadw_max>;
1142 def ATOMIC_LOAD_MAX_32 : AtomicLoadBinaryReg32<atomic_load_max_32>;
1143 def ATOMIC_LOAD_MAX_64 : AtomicLoadBinaryReg64<atomic_load_max_64>;
1145 def ATOMIC_LOADW_UMIN : AtomicLoadWBinaryReg<z_atomic_loadw_umin>;
1146 def ATOMIC_LOAD_UMIN_32 : AtomicLoadBinaryReg32<atomic_load_umin_32>;
1147 def ATOMIC_LOAD_UMIN_64 : AtomicLoadBinaryReg64<atomic_load_umin_64>;
1149 def ATOMIC_LOADW_UMAX : AtomicLoadWBinaryReg<z_atomic_loadw_umax>;
1150 def ATOMIC_LOAD_UMAX_32 : AtomicLoadBinaryReg32<atomic_load_umax_32>;
1151 def ATOMIC_LOAD_UMAX_64 : AtomicLoadBinaryReg64<atomic_load_umax_64>;
1153 def ATOMIC_CMP_SWAPW
1154 : Pseudo<(outs GR32:$dst), (ins bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1155 ADDR32:$bitshift, ADDR32:$negbitshift,
1158 (z_atomic_cmp_swapw bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1159 ADDR32:$bitshift, ADDR32:$negbitshift,
1160 uimm32:$bitsize))]> {
1164 let usesCustomInserter = 1;
1167 let Defs = [CC] in {
1168 defm CS : CmpSwapRSPair<"cs", 0xBA, 0xEB14, atomic_cmp_swap_32, GR32>;
1169 def CSG : CmpSwapRSY<"csg", 0xEB30, atomic_cmp_swap_64, GR64>;
1172 //===----------------------------------------------------------------------===//
1173 // Miscellaneous Instructions.
1174 //===----------------------------------------------------------------------===//
1176 // Extract CC into bits 29 and 28 of a register.
1178 def IPM : InherentRRE<"ipm", 0xB222, GR32, (z_ipm)>;
1180 // Read a 32-bit access register into a GR32. As with all GR32 operations,
1181 // the upper 32 bits of the enclosing GR64 remain unchanged, which is useful
1182 // when a 64-bit address is stored in a pair of access registers.
1183 def EAR : InstRRE<0xB24F, (outs GR32:$R1), (ins access_reg:$R2),
1185 [(set GR32:$R1, (z_extract_access access_reg:$R2))]>;
1187 // Find leftmost one, AKA count leading zeros. The instruction actually
1188 // returns a pair of GR64s, the first giving the number of leading zeros
1189 // and the second giving a copy of the source with the leftmost one bit
1190 // cleared. We only use the first result here.
1191 let Defs = [CC] in {
1192 def FLOGR : UnaryRRE<"flog", 0xB983, null_frag, GR128, GR64>;
1194 def : Pat<(ctlz GR64:$src),
1195 (EXTRACT_SUBREG (FLOGR GR64:$src), subreg_high)>;
1197 // Use subregs to populate the "don't care" bits in a 32-bit to 64-bit anyext.
1198 def : Pat<(i64 (anyext GR32:$src)),
1199 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_32bit)>;
1201 // There are no 32-bit equivalents of LLILL and LLILH, so use a full
1202 // 64-bit move followed by a subreg. This preserves the invariant that
1203 // all GR32 operations only modify the low 32 bits.
1204 def : Pat<(i32 imm32ll16:$src),
1205 (EXTRACT_SUBREG (LLILL (LL16 imm:$src)), subreg_32bit)>;
1206 def : Pat<(i32 imm32lh16:$src),
1207 (EXTRACT_SUBREG (LLILH (LH16 imm:$src)), subreg_32bit)>;
1209 // Extend GR32s and GR64s to GR128s.
1210 let usesCustomInserter = 1 in {
1211 def AEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1212 def ZEXT128_32 : Pseudo<(outs GR128:$dst), (ins GR32:$src), []>;
1213 def ZEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1216 // Search a block of memory for a character.
1217 let mayLoad = 1, Defs = [CC], Uses = [R0W] in
1218 defm SRST : StringRRE<"srst", 0xb25e, z_search_string>;
1220 //===----------------------------------------------------------------------===//
1222 //===----------------------------------------------------------------------===//
1224 // Use AL* for GR64 additions of unsigned 32-bit values.
1225 defm : ZXB<add, GR64, ALGFR>;
1226 def : Pat<(add GR64:$src1, imm64zx32:$src2),
1227 (ALGFI GR64:$src1, imm64zx32:$src2)>;
1228 def : Pat<(add GR64:$src1, (zextloadi32 bdxaddr20only:$addr)),
1229 (ALGF GR64:$src1, bdxaddr20only:$addr)>;
1231 // Use SL* for GR64 subtractions of unsigned 32-bit values.
1232 defm : ZXB<sub, GR64, SLGFR>;
1233 def : Pat<(add GR64:$src1, imm64zx32n:$src2),
1234 (SLGFI GR64:$src1, imm64zx32n:$src2)>;
1235 def : Pat<(sub GR64:$src1, (zextloadi32 bdxaddr20only:$addr)),
1236 (SLGF GR64:$src1, bdxaddr20only:$addr)>;
1238 // Optimize sign-extended 1/0 selects to -1/0 selects. This is important
1239 // for vector legalization.
1240 def : Pat<(sra (shl (i32 (z_select_ccmask 1, 0, uimm8zx4:$valid, uimm8zx4:$cc)),
1243 (Select32 (LHI -1), (LHI 0), uimm8zx4:$valid, uimm8zx4:$cc)>;
1244 def : Pat<(sra (shl (i64 (anyext (i32 (z_select_ccmask 1, 0, uimm8zx4:$valid,
1248 (Select64 (LGHI -1), (LGHI 0), uimm8zx4:$valid, uimm8zx4:$cc)>;