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 (br %r14).
36 let isReturn = 1, isTerminator = 1, isBarrier = 1, hasCtrlDep = 1 in
37 def Return : Alias<2, (outs), (ins), [(z_retflag)]>;
39 // Unconditional branches. R1 is the condition-code mask (all 1s).
40 let isBranch = 1, isTerminator = 1, isBarrier = 1, R1 = 15 in {
41 let isIndirectBranch = 1 in
42 def BR : InstRR<0x07, (outs), (ins ADDR64:$R2),
43 "br\t$R2", [(brind ADDR64:$R2)]>;
45 // An assembler extended mnemonic for BRC.
46 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2), "j\t$I2",
49 // An assembler extended mnemonic for BRCL. (The extension is "G"
50 // rather than "L" because "JL" is "Jump if Less".)
51 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2), "jg\t$I2", []>;
54 // Conditional branches. It's easier for LLVM to handle these branches
55 // in their raw BRC/BRCL form, with the 4-bit condition-code mask being
56 // the first operand. It seems friendlier to use mnemonic forms like
57 // JE and JLH when writing out the assembly though.
58 let isBranch = 1, isTerminator = 1, Uses = [CC] in {
59 let isCodeGenOnly = 1, CCMaskFirst = 1 in {
60 def BRC : InstRI<0xA74, (outs), (ins cond4:$valid, cond4:$R1,
61 brtarget16:$I2), "j$R1\t$I2",
62 [(z_br_ccmask cond4:$valid, cond4:$R1, bb:$I2)]>;
63 def BRCL : InstRIL<0xC04, (outs), (ins cond4:$valid, cond4:$R1,
64 brtarget32:$I2), "jg$R1\t$I2", []>;
66 def AsmBRC : InstRI<0xA74, (outs), (ins uimm8zx4:$R1, brtarget16:$I2),
68 def AsmBRCL : InstRIL<0xC04, (outs), (ins uimm8zx4:$R1, brtarget32:$I2),
69 "brcl\t$R1, $I2", []>;
72 // Fused compare-and-branch instructions. As for normal branches,
73 // we handle these instructions internally in their raw CRJ-like form,
74 // but use assembly macros like CRJE when writing them out.
76 // These instructions do not use or clobber the condition codes.
77 // We nevertheless pretend that they clobber CC, so that we can lower
78 // them to separate comparisons and BRCLs if the branch ends up being
80 multiclass CompareBranches<Operand ccmask, string pos1, string pos2> {
81 let isBranch = 1, isTerminator = 1, Defs = [CC] in {
82 def RJ : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2, ccmask:$M3,
84 "crj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
85 def GRJ : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2, ccmask:$M3,
87 "cgrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
88 def IJ : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2, ccmask:$M3,
90 "cij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
91 def GIJ : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2, ccmask:$M3,
93 "cgij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
94 def LRJ : InstRIEb<0xEC77, (outs), (ins GR32:$R1, GR32:$R2, ccmask:$M3,
96 "clrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
97 def LGRJ : InstRIEb<0xEC65, (outs), (ins GR64:$R1, GR64:$R2, ccmask:$M3,
99 "clgrj"##pos1##"\t$R1, $R2, "##pos2##"$RI4", []>;
100 def LIJ : InstRIEc<0xEC7F, (outs), (ins GR32:$R1, imm32zx8:$I2, ccmask:$M3,
102 "clij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
103 def LGIJ : InstRIEc<0xEC7D, (outs), (ins GR64:$R1, imm64zx8:$I2, ccmask:$M3,
105 "clgij"##pos1##"\t$R1, $I2, "##pos2##"$RI4", []>;
108 let isCodeGenOnly = 1 in
109 defm C : CompareBranches<cond4, "$M3", "">;
110 defm AsmC : CompareBranches<uimm8zx4, "", "$M3, ">;
112 // Define AsmParser mnemonics for each general condition-code mask
113 // (integer or floating-point)
114 multiclass CondExtendedMnemonic<bits<4> ccmask, string name> {
116 def J : InstRI<0xA74, (outs), (ins brtarget16:$I2),
117 "j"##name##"\t$I2", []>;
118 def JG : InstRIL<0xC04, (outs), (ins brtarget32:$I2),
119 "jg"##name##"\t$I2", []>;
121 def LOCR : FixedCondUnaryRRF<"locr"##name, 0xB9F2, GR32, GR32, ccmask>;
122 def LOCGR : FixedCondUnaryRRF<"locgr"##name, 0xB9E2, GR64, GR64, ccmask>;
123 def LOC : FixedCondUnaryRSY<"loc"##name, 0xEBF2, GR32, ccmask, 4>;
124 def LOCG : FixedCondUnaryRSY<"locg"##name, 0xEBE2, GR64, ccmask, 8>;
125 def STOC : FixedCondStoreRSY<"stoc"##name, 0xEBF3, GR32, ccmask, 4>;
126 def STOCG : FixedCondStoreRSY<"stocg"##name, 0xEBE3, GR64, ccmask, 8>;
128 defm AsmO : CondExtendedMnemonic<1, "o">;
129 defm AsmH : CondExtendedMnemonic<2, "h">;
130 defm AsmNLE : CondExtendedMnemonic<3, "nle">;
131 defm AsmL : CondExtendedMnemonic<4, "l">;
132 defm AsmNHE : CondExtendedMnemonic<5, "nhe">;
133 defm AsmLH : CondExtendedMnemonic<6, "lh">;
134 defm AsmNE : CondExtendedMnemonic<7, "ne">;
135 defm AsmE : CondExtendedMnemonic<8, "e">;
136 defm AsmNLH : CondExtendedMnemonic<9, "nlh">;
137 defm AsmHE : CondExtendedMnemonic<10, "he">;
138 defm AsmNL : CondExtendedMnemonic<11, "nl">;
139 defm AsmLE : CondExtendedMnemonic<12, "le">;
140 defm AsmNH : CondExtendedMnemonic<13, "nh">;
141 defm AsmNO : CondExtendedMnemonic<14, "no">;
143 // Define AsmParser mnemonics for each integer condition-code mask.
144 // This is like the list above, except that condition 3 is not possible
145 // and that the low bit of the mask is therefore always 0. This means
146 // that each condition has two names. Conditions "o" and "no" are not used.
148 // We don't make one of the two names an alias of the other because
149 // we need the custom parsing routines to select the correct register class.
150 multiclass IntCondExtendedMnemonicA<bits<4> ccmask, string name> {
152 def CR : InstRIEb<0xEC76, (outs), (ins GR32:$R1, GR32:$R2,
154 "crj"##name##"\t$R1, $R2, $RI4", []>;
155 def CGR : InstRIEb<0xEC64, (outs), (ins GR64:$R1, GR64:$R2,
157 "cgrj"##name##"\t$R1, $R2, $RI4", []>;
158 def CI : InstRIEc<0xEC7E, (outs), (ins GR32:$R1, imm32sx8:$I2,
160 "cij"##name##"\t$R1, $I2, $RI4", []>;
161 def CGI : InstRIEc<0xEC7C, (outs), (ins GR64:$R1, imm64sx8:$I2,
163 "cgij"##name##"\t$R1, $I2, $RI4", []>;
164 def CLR : InstRIEb<0xEC77, (outs), (ins GR32:$R1, GR32:$R2,
166 "clrj"##name##"\t$R1, $R2, $RI4", []>;
167 def CLGR : InstRIEb<0xEC65, (outs), (ins GR64:$R1, GR64:$R2,
169 "clgrj"##name##"\t$R1, $R2, $RI4", []>;
170 def CLI : InstRIEc<0xEC7F, (outs), (ins GR32:$R1, imm32zx8:$I2,
172 "clij"##name##"\t$R1, $I2, $RI4", []>;
173 def CLGI : InstRIEc<0xEC7D, (outs), (ins GR64:$R1, imm64zx8:$I2,
175 "clgij"##name##"\t$R1, $I2, $RI4", []>;
178 multiclass IntCondExtendedMnemonic<bits<4> ccmask, string name1, string name2>
179 : IntCondExtendedMnemonicA<ccmask, name1> {
180 let isAsmParserOnly = 1 in
181 defm Alt : IntCondExtendedMnemonicA<ccmask, name2>;
183 defm AsmJH : IntCondExtendedMnemonic<2, "h", "nle">;
184 defm AsmJL : IntCondExtendedMnemonic<4, "l", "nhe">;
185 defm AsmJLH : IntCondExtendedMnemonic<6, "lh", "ne">;
186 defm AsmJE : IntCondExtendedMnemonic<8, "e", "nlh">;
187 defm AsmJHE : IntCondExtendedMnemonic<10, "he", "nl">;
188 defm AsmJLE : IntCondExtendedMnemonic<12, "le", "nh">;
190 // Decrement a register and branch if it is nonzero. These don't clobber CC,
191 // but we might need to split long branches into sequences that do.
193 def BRCT : BranchUnaryRI<"brct", 0xA76, GR32>;
194 def BRCTG : BranchUnaryRI<"brctg", 0xA77, GR64>;
197 //===----------------------------------------------------------------------===//
198 // Select instructions
199 //===----------------------------------------------------------------------===//
201 def Select32 : SelectWrapper<GR32>;
202 def Select64 : SelectWrapper<GR64>;
204 defm CondStore8 : CondStores<GR32, nonvolatile_truncstorei8,
205 nonvolatile_anyextloadi8, bdxaddr20only>;
206 defm CondStore16 : CondStores<GR32, nonvolatile_truncstorei16,
207 nonvolatile_anyextloadi16, bdxaddr20only>;
208 defm CondStore32 : CondStores<GR32, nonvolatile_store,
209 nonvolatile_load, bdxaddr20only>;
211 defm : CondStores64<CondStore8, CondStore8Inv, nonvolatile_truncstorei8,
212 nonvolatile_anyextloadi8, bdxaddr20only>;
213 defm : CondStores64<CondStore16, CondStore16Inv, nonvolatile_truncstorei16,
214 nonvolatile_anyextloadi16, bdxaddr20only>;
215 defm : CondStores64<CondStore32, CondStore32Inv, nonvolatile_truncstorei32,
216 nonvolatile_anyextloadi32, bdxaddr20only>;
217 defm CondStore64 : CondStores<GR64, nonvolatile_store,
218 nonvolatile_load, bdxaddr20only>;
220 //===----------------------------------------------------------------------===//
222 //===----------------------------------------------------------------------===//
224 // The definitions here are for the call-clobbered registers.
225 let isCall = 1, Defs = [R0D, R1D, R2D, R3D, R4D, R5D, R14D,
226 F0D, F1D, F2D, F3D, F4D, F5D, F6D, F7D, CC],
227 R1 = 14, isCodeGenOnly = 1 in {
228 def BRAS : InstRI<0xA75, (outs), (ins pcrel16call:$I2, variable_ops),
229 "bras\t%r14, $I2", []>;
230 def BRASL : InstRIL<0xC05, (outs), (ins pcrel32call:$I2, variable_ops),
231 "brasl\t%r14, $I2", [(z_call pcrel32call:$I2)]>;
232 def BASR : InstRR<0x0D, (outs), (ins ADDR64:$R2, variable_ops),
233 "basr\t%r14, $R2", [(z_call ADDR64:$R2)]>;
236 // Sibling calls. Indirect sibling calls must be via R1, since R2 upwards
237 // are argument registers and since branching to R0 is a no-op.
238 let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1,
239 isCodeGenOnly = 1, R1 = 15 in {
240 def CallJG : InstRIL<0xC04, (outs), (ins pcrel32call:$I2),
241 "jg\t$I2", [(z_sibcall pcrel32call:$I2)]>;
242 let R2 = 1, Uses = [R1D] in
243 def CallBR : InstRR<0x07, (outs), (ins), "br\t%r1", [(z_sibcall R1D)]>;
246 // Define the general form of the call instructions for the asm parser.
247 // These instructions don't hard-code %r14 as the return address register.
248 def AsmBRAS : InstRI<0xA75, (outs), (ins GR64:$R1, brtarget16:$I2),
249 "bras\t$R1, $I2", []>;
250 def AsmBRASL : InstRIL<0xC05, (outs), (ins GR64:$R1, brtarget32:$I2),
251 "brasl\t$R1, $I2", []>;
252 def AsmBASR : InstRR<0x0D, (outs), (ins GR64:$R1, ADDR64:$R2),
253 "basr\t$R1, $R2", []>;
255 //===----------------------------------------------------------------------===//
257 //===----------------------------------------------------------------------===//
260 let neverHasSideEffects = 1 in {
261 def LR : UnaryRR <"l", 0x18, null_frag, GR32, GR32>;
262 def LGR : UnaryRRE<"lg", 0xB904, null_frag, GR64, GR64>;
264 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
265 def LTR : UnaryRR <"lt", 0x12, null_frag, GR32, GR32>;
266 def LTGR : UnaryRRE<"ltg", 0xB902, null_frag, GR64, GR64>;
269 // Move on condition.
270 let isCodeGenOnly = 1, Uses = [CC] in {
271 def LOCR : CondUnaryRRF<"loc", 0xB9F2, GR32, GR32>;
272 def LOCGR : CondUnaryRRF<"locg", 0xB9E2, GR64, GR64>;
275 def AsmLOCR : AsmCondUnaryRRF<"loc", 0xB9F2, GR32, GR32>;
276 def AsmLOCGR : AsmCondUnaryRRF<"locg", 0xB9E2, GR64, GR64>;
280 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
281 isReMaterializable = 1 in {
282 // 16-bit sign-extended immediates.
283 def LHI : UnaryRI<"lhi", 0xA78, bitconvert, GR32, imm32sx16>;
284 def LGHI : UnaryRI<"lghi", 0xA79, bitconvert, GR64, imm64sx16>;
286 // Other 16-bit immediates.
287 def LLILL : UnaryRI<"llill", 0xA5F, bitconvert, GR64, imm64ll16>;
288 def LLILH : UnaryRI<"llilh", 0xA5E, bitconvert, GR64, imm64lh16>;
289 def LLIHL : UnaryRI<"llihl", 0xA5D, bitconvert, GR64, imm64hl16>;
290 def LLIHH : UnaryRI<"llihh", 0xA5C, bitconvert, GR64, imm64hh16>;
292 // 32-bit immediates.
293 def LGFI : UnaryRIL<"lgfi", 0xC01, bitconvert, GR64, imm64sx32>;
294 def LLILF : UnaryRIL<"llilf", 0xC0F, bitconvert, GR64, imm64lf32>;
295 def LLIHF : UnaryRIL<"llihf", 0xC0E, bitconvert, GR64, imm64hf32>;
299 let canFoldAsLoad = 1, SimpleBDXLoad = 1 in {
300 defm L : UnaryRXPair<"l", 0x58, 0xE358, load, GR32, 4>;
301 def LG : UnaryRXY<"lg", 0xE304, load, GR64, 8>;
303 // These instructions are split after register allocation, so we don't
304 // want a custom inserter.
305 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
306 def L128 : Pseudo<(outs GR128:$dst), (ins bdxaddr20only128:$src),
307 [(set GR128:$dst, (load bdxaddr20only128:$src))]>;
310 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
311 def LT : UnaryRXY<"lt", 0xE312, load, GR32, 4>;
312 def LTG : UnaryRXY<"ltg", 0xE302, load, GR64, 8>;
315 let canFoldAsLoad = 1 in {
316 def LRL : UnaryRILPC<"lrl", 0xC4D, aligned_load, GR32>;
317 def LGRL : UnaryRILPC<"lgrl", 0xC48, aligned_load, GR64>;
320 // Load on condition.
321 let isCodeGenOnly = 1, Uses = [CC] in {
322 def LOC : CondUnaryRSY<"loc", 0xEBF2, nonvolatile_load, GR32, 4>;
323 def LOCG : CondUnaryRSY<"locg", 0xEBE2, nonvolatile_load, GR64, 8>;
326 def AsmLOC : AsmCondUnaryRSY<"loc", 0xEBF2, GR32, 4>;
327 def AsmLOCG : AsmCondUnaryRSY<"locg", 0xEBE2, GR64, 8>;
331 let SimpleBDXStore = 1 in {
332 defm ST : StoreRXPair<"st", 0x50, 0xE350, store, GR32, 4>;
333 def STG : StoreRXY<"stg", 0xE324, store, GR64, 8>;
335 // These instructions are split after register allocation, so we don't
336 // want a custom inserter.
337 let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
338 def ST128 : Pseudo<(outs), (ins GR128:$src, bdxaddr20only128:$dst),
339 [(store GR128:$src, bdxaddr20only128:$dst)]>;
342 def STRL : StoreRILPC<"strl", 0xC4F, aligned_store, GR32>;
343 def STGRL : StoreRILPC<"stgrl", 0xC4B, aligned_store, GR64>;
345 // Store on condition.
346 let isCodeGenOnly = 1, Uses = [CC] in {
347 def STOC : CondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
348 def STOCG : CondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
351 def AsmSTOC : AsmCondStoreRSY<"stoc", 0xEBF3, GR32, 4>;
352 def AsmSTOCG : AsmCondStoreRSY<"stocg", 0xEBE3, GR64, 8>;
355 // 8-bit immediate stores to 8-bit fields.
356 defm MVI : StoreSIPair<"mvi", 0x92, 0xEB52, truncstorei8, imm32zx8trunc>;
358 // 16-bit immediate stores to 16-, 32- or 64-bit fields.
359 def MVHHI : StoreSIL<"mvhhi", 0xE544, truncstorei16, imm32sx16trunc>;
360 def MVHI : StoreSIL<"mvhi", 0xE54C, store, imm32sx16>;
361 def MVGHI : StoreSIL<"mvghi", 0xE548, store, imm64sx16>;
363 // Memory-to-memory moves.
364 let mayLoad = 1, mayStore = 1 in
365 defm MVC : MemorySS<"mvc", 0xD2, z_mvc, z_mvc_loop>;
368 let mayLoad = 1, mayStore = 1, Defs = [CC], Uses = [R0W] in
369 defm MVST : StringRRE<"mvst", 0xB255, z_stpcpy>;
371 //===----------------------------------------------------------------------===//
373 //===----------------------------------------------------------------------===//
375 // Note that putting these before zero extensions mean that we will prefer
376 // them for anyextload*. There's not really much to choose between the two
377 // either way, but signed-extending loads have a short LH and a long LHY,
378 // while zero-extending loads have only the long LLH.
380 //===----------------------------------------------------------------------===//
382 // 32-bit extensions from registers.
383 let neverHasSideEffects = 1 in {
384 def LBR : UnaryRRE<"lb", 0xB926, sext8, GR32, GR32>;
385 def LHR : UnaryRRE<"lh", 0xB927, sext16, GR32, GR32>;
388 // 64-bit extensions from registers.
389 let neverHasSideEffects = 1 in {
390 def LGBR : UnaryRRE<"lgb", 0xB906, sext8, GR64, GR64>;
391 def LGHR : UnaryRRE<"lgh", 0xB907, sext16, GR64, GR64>;
392 def LGFR : UnaryRRE<"lgf", 0xB914, sext32, GR64, GR32>;
394 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in
395 def LTGFR : UnaryRRE<"ltgf", 0xB912, null_frag, GR64, GR64>;
397 // Match 32-to-64-bit sign extensions in which the source is already
398 // in a 64-bit register.
399 def : Pat<(sext_inreg GR64:$src, i32),
400 (LGFR (EXTRACT_SUBREG GR64:$src, subreg_32bit))>;
402 // 32-bit extensions from memory.
403 def LB : UnaryRXY<"lb", 0xE376, asextloadi8, GR32, 1>;
404 defm LH : UnaryRXPair<"lh", 0x48, 0xE378, asextloadi16, GR32, 2>;
405 def LHRL : UnaryRILPC<"lhrl", 0xC45, aligned_asextloadi16, GR32>;
407 // 64-bit extensions from memory.
408 def LGB : UnaryRXY<"lgb", 0xE377, asextloadi8, GR64, 1>;
409 def LGH : UnaryRXY<"lgh", 0xE315, asextloadi16, GR64, 2>;
410 def LGF : UnaryRXY<"lgf", 0xE314, asextloadi32, GR64, 4>;
411 def LGHRL : UnaryRILPC<"lghrl", 0xC44, aligned_asextloadi16, GR64>;
412 def LGFRL : UnaryRILPC<"lgfrl", 0xC4C, aligned_asextloadi32, GR64>;
413 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in
414 def LTGF : UnaryRXY<"ltgf", 0xE332, asextloadi32, GR64, 4>;
416 //===----------------------------------------------------------------------===//
418 //===----------------------------------------------------------------------===//
420 // 32-bit extensions from registers.
421 let neverHasSideEffects = 1 in {
422 def LLCR : UnaryRRE<"llc", 0xB994, zext8, GR32, GR32>;
423 def LLHR : UnaryRRE<"llh", 0xB995, zext16, GR32, GR32>;
426 // 64-bit extensions from registers.
427 let neverHasSideEffects = 1 in {
428 def LLGCR : UnaryRRE<"llgc", 0xB984, zext8, GR64, GR64>;
429 def LLGHR : UnaryRRE<"llgh", 0xB985, zext16, GR64, GR64>;
430 def LLGFR : UnaryRRE<"llgf", 0xB916, zext32, GR64, GR32>;
433 // Match 32-to-64-bit zero extensions in which the source is already
434 // in a 64-bit register.
435 def : Pat<(and GR64:$src, 0xffffffff),
436 (LLGFR (EXTRACT_SUBREG GR64:$src, subreg_32bit))>;
438 // 32-bit extensions from memory.
439 def LLC : UnaryRXY<"llc", 0xE394, azextloadi8, GR32, 1>;
440 def LLH : UnaryRXY<"llh", 0xE395, azextloadi16, GR32, 2>;
441 def LLHRL : UnaryRILPC<"llhrl", 0xC42, aligned_azextloadi16, GR32>;
443 // 64-bit extensions from memory.
444 def LLGC : UnaryRXY<"llgc", 0xE390, azextloadi8, GR64, 1>;
445 def LLGH : UnaryRXY<"llgh", 0xE391, azextloadi16, GR64, 2>;
446 def LLGF : UnaryRXY<"llgf", 0xE316, azextloadi32, GR64, 4>;
447 def LLGHRL : UnaryRILPC<"llghrl", 0xC46, aligned_azextloadi16, GR64>;
448 def LLGFRL : UnaryRILPC<"llgfrl", 0xC4E, aligned_azextloadi32, GR64>;
450 //===----------------------------------------------------------------------===//
452 //===----------------------------------------------------------------------===//
454 // Truncations of 64-bit registers to 32-bit registers.
455 def : Pat<(i32 (trunc GR64:$src)),
456 (EXTRACT_SUBREG GR64:$src, subreg_32bit)>;
458 // Truncations of 32-bit registers to memory.
459 defm STC : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR32, 1>;
460 defm STH : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR32, 2>;
461 def STHRL : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR32>;
463 // Truncations of 64-bit registers to memory.
464 defm : StoreGR64Pair<STC, STCY, truncstorei8>;
465 defm : StoreGR64Pair<STH, STHY, truncstorei16>;
466 def : StoreGR64PC<STHRL, aligned_truncstorei16>;
467 defm : StoreGR64Pair<ST, STY, truncstorei32>;
468 def : StoreGR64PC<STRL, aligned_truncstorei32>;
470 //===----------------------------------------------------------------------===//
471 // Multi-register moves
472 //===----------------------------------------------------------------------===//
474 // Multi-register loads.
475 def LMG : LoadMultipleRSY<"lmg", 0xEB04, GR64>;
477 // Multi-register stores.
478 def STMG : StoreMultipleRSY<"stmg", 0xEB24, GR64>;
480 //===----------------------------------------------------------------------===//
482 //===----------------------------------------------------------------------===//
484 // Byte-swapping register moves.
485 let neverHasSideEffects = 1 in {
486 def LRVR : UnaryRRE<"lrv", 0xB91F, bswap, GR32, GR32>;
487 def LRVGR : UnaryRRE<"lrvg", 0xB90F, bswap, GR64, GR64>;
490 // Byte-swapping loads. Unlike normal loads, these instructions are
491 // allowed to access storage more than once.
492 def LRV : UnaryRXY<"lrv", 0xE31E, loadu<bswap, nonvolatile_load>, GR32, 4>;
493 def LRVG : UnaryRXY<"lrvg", 0xE30F, loadu<bswap, nonvolatile_load>, GR64, 8>;
495 // Likewise byte-swapping stores.
496 def STRV : StoreRXY<"strv", 0xE33E, storeu<bswap, nonvolatile_store>, GR32, 4>;
497 def STRVG : StoreRXY<"strvg", 0xE32F, storeu<bswap, nonvolatile_store>,
500 //===----------------------------------------------------------------------===//
501 // Load address instructions
502 //===----------------------------------------------------------------------===//
504 // Load BDX-style addresses.
505 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isReMaterializable = 1,
507 let DispSize = "12" in
508 def LA : InstRX<0x41, (outs GR64:$R1), (ins laaddr12pair:$XBD2),
510 [(set GR64:$R1, laaddr12pair:$XBD2)]>;
511 let DispSize = "20" in
512 def LAY : InstRXY<0xE371, (outs GR64:$R1), (ins laaddr20pair:$XBD2),
514 [(set GR64:$R1, laaddr20pair:$XBD2)]>;
517 // Load a PC-relative address. There's no version of this instruction
518 // with a 16-bit offset, so there's no relaxation.
519 let neverHasSideEffects = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
520 isReMaterializable = 1 in {
521 def LARL : InstRIL<0xC00, (outs GR64:$R1), (ins pcrel32:$I2),
523 [(set GR64:$R1, pcrel32:$I2)]>;
526 //===----------------------------------------------------------------------===//
527 // Absolute and Negation
528 //===----------------------------------------------------------------------===//
531 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
532 def LPR : UnaryRR <"lp", 0x10, z_iabs32, GR32, GR32>;
533 def LPGR : UnaryRRE<"lpg", 0xB900, z_iabs64, GR64, GR64>;
535 let CCValues = 0xE, CompareZeroCCMask = 0xE in
536 def LPGFR : UnaryRRE<"lpgf", 0xB910, null_frag, GR64, GR32>;
538 defm : SXU<z_iabs64, LPGFR>;
541 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
542 def LNR : UnaryRR <"ln", 0x11, z_inegabs32, GR32, GR32>;
543 def LNGR : UnaryRRE<"lng", 0xB901, z_inegabs64, GR64, GR64>;
545 let CCValues = 0xE, CompareZeroCCMask = 0xE in
546 def LNGFR : UnaryRRE<"lngf", 0xB911, null_frag, GR64, GR32>;
548 defm : SXU<z_inegabs64, LNGFR>;
551 let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
552 def LCR : UnaryRR <"lc", 0x13, ineg, GR32, GR32>;
553 def LCGR : UnaryRRE<"lcg", 0xB903, ineg, GR64, GR64>;
555 let CCValues = 0xE, CompareZeroCCMask = 0xE in
556 def LCGFR : UnaryRRE<"lcgf", 0xB913, null_frag, GR64, GR32>;
558 defm : SXU<ineg, LCGFR>;
560 //===----------------------------------------------------------------------===//
562 //===----------------------------------------------------------------------===//
564 let isCodeGenOnly = 1 in
565 defm IC32 : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR32, azextloadi8, 1>;
566 defm IC : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR64, azextloadi8, 1>;
568 defm : InsertMem<"inserti8", IC32, GR32, azextloadi8, bdxaddr12pair>;
569 defm : InsertMem<"inserti8", IC32Y, GR32, azextloadi8, bdxaddr20pair>;
571 defm : InsertMem<"inserti8", IC, GR64, azextloadi8, bdxaddr12pair>;
572 defm : InsertMem<"inserti8", ICY, GR64, azextloadi8, bdxaddr20pair>;
574 // Insertions of a 16-bit immediate, leaving other bits unaffected.
575 // We don't have or_as_insert equivalents of these operations because
576 // OI is available instead.
577 let isCodeGenOnly = 1 in {
578 def IILL32 : BinaryRI<"iill", 0xA53, insertll, GR32, imm32ll16>;
579 def IILH32 : BinaryRI<"iilh", 0xA52, insertlh, GR32, imm32lh16>;
581 def IILL : BinaryRI<"iill", 0xA53, insertll, GR64, imm64ll16>;
582 def IILH : BinaryRI<"iilh", 0xA52, insertlh, GR64, imm64lh16>;
583 def IIHL : BinaryRI<"iihl", 0xA51, inserthl, GR64, imm64hl16>;
584 def IIHH : BinaryRI<"iihh", 0xA50, inserthh, GR64, imm64hh16>;
586 // ...likewise for 32-bit immediates. For GR32s this is a general
587 // full-width move. (We use IILF rather than something like LLILF
588 // for 32-bit moves because IILF leaves the upper 32 bits of the
590 let isCodeGenOnly = 1, isAsCheapAsAMove = 1, isMoveImm = 1,
591 isReMaterializable = 1 in {
592 def IILF32 : UnaryRIL<"iilf", 0xC09, bitconvert, GR32, uimm32>;
594 def IILF : BinaryRIL<"iilf", 0xC09, insertlf, GR64, imm64lf32>;
595 def IIHF : BinaryRIL<"iihf", 0xC08, inserthf, GR64, imm64hf32>;
597 // An alternative model of inserthf, with the first operand being
598 // a zero-extended value.
599 def : Pat<(or (zext32 GR32:$src), imm64hf32:$imm),
600 (IIHF (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_32bit),
603 //===----------------------------------------------------------------------===//
605 //===----------------------------------------------------------------------===//
608 let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in {
609 // Addition of a register.
610 let isCommutable = 1 in {
611 defm AR : BinaryRRAndK<"a", 0x1A, 0xB9F8, add, GR32, GR32>;
612 defm AGR : BinaryRREAndK<"ag", 0xB908, 0xB9E8, add, GR64, GR64>;
614 def AGFR : BinaryRRE<"agf", 0xB918, null_frag, GR64, GR32>;
616 // Addition of signed 16-bit immediates.
617 defm AHI : BinaryRIAndK<"ahi", 0xA7A, 0xECD8, add, GR32, imm32sx16>;
618 defm AGHI : BinaryRIAndK<"aghi", 0xA7B, 0xECD9, add, GR64, imm64sx16>;
620 // Addition of signed 32-bit immediates.
621 def AFI : BinaryRIL<"afi", 0xC29, add, GR32, simm32>;
622 def AGFI : BinaryRIL<"agfi", 0xC28, add, GR64, imm64sx32>;
624 // Addition of memory.
625 defm AH : BinaryRXPair<"ah", 0x4A, 0xE37A, add, GR32, asextloadi16, 2>;
626 defm A : BinaryRXPair<"a", 0x5A, 0xE35A, add, GR32, load, 4>;
627 def AGF : BinaryRXY<"agf", 0xE318, add, GR64, asextloadi32, 4>;
628 def AG : BinaryRXY<"ag", 0xE308, add, GR64, load, 8>;
630 // Addition to memory.
631 def ASI : BinarySIY<"asi", 0xEB6A, add, imm32sx8>;
632 def AGSI : BinarySIY<"agsi", 0xEB7A, add, imm64sx8>;
634 defm : SXB<add, GR64, AGFR>;
636 // Addition producing a carry.
638 // Addition of a register.
639 let isCommutable = 1 in {
640 defm ALR : BinaryRRAndK<"al", 0x1E, 0xB9FA, addc, GR32, GR32>;
641 defm ALGR : BinaryRREAndK<"alg", 0xB90A, 0xB9EA, addc, GR64, GR64>;
643 def ALGFR : BinaryRRE<"algf", 0xB91A, null_frag, GR64, GR32>;
645 // Addition of signed 16-bit immediates.
646 def ALHSIK : BinaryRIE<"alhsik", 0xECDA, addc, GR32, imm32sx16>,
647 Requires<[FeatureDistinctOps]>;
648 def ALGHSIK : BinaryRIE<"alghsik", 0xECDB, addc, GR64, imm64sx16>,
649 Requires<[FeatureDistinctOps]>;
651 // Addition of unsigned 32-bit immediates.
652 def ALFI : BinaryRIL<"alfi", 0xC2B, addc, GR32, uimm32>;
653 def ALGFI : BinaryRIL<"algfi", 0xC2A, addc, GR64, imm64zx32>;
655 // Addition of memory.
656 defm AL : BinaryRXPair<"al", 0x5E, 0xE35E, addc, GR32, load, 4>;
657 def ALGF : BinaryRXY<"algf", 0xE31A, addc, GR64, azextloadi32, 4>;
658 def ALG : BinaryRXY<"alg", 0xE30A, addc, GR64, load, 8>;
660 defm : ZXB<addc, GR64, ALGFR>;
662 // Addition producing and using a carry.
663 let Defs = [CC], Uses = [CC] in {
664 // Addition of a register.
665 def ALCR : BinaryRRE<"alc", 0xB998, adde, GR32, GR32>;
666 def ALCGR : BinaryRRE<"alcg", 0xB988, adde, GR64, GR64>;
668 // Addition of memory.
669 def ALC : BinaryRXY<"alc", 0xE398, adde, GR32, load, 4>;
670 def ALCG : BinaryRXY<"alcg", 0xE388, adde, GR64, load, 8>;
673 //===----------------------------------------------------------------------===//
675 //===----------------------------------------------------------------------===//
677 // Plain substraction. Although immediate forms exist, we use the
678 // add-immediate instruction instead.
679 let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in {
680 // Subtraction of a register.
681 defm SR : BinaryRRAndK<"s", 0x1B, 0xB9F9, sub, GR32, GR32>;
682 def SGFR : BinaryRRE<"sgf", 0xB919, null_frag, GR64, GR32>;
683 defm SGR : BinaryRREAndK<"sg", 0xB909, 0xB9E9, sub, GR64, GR64>;
685 // Subtraction of memory.
686 defm SH : BinaryRXPair<"sh", 0x4B, 0xE37B, sub, GR32, asextloadi16, 2>;
687 defm S : BinaryRXPair<"s", 0x5B, 0xE35B, sub, GR32, load, 4>;
688 def SGF : BinaryRXY<"sgf", 0xE319, sub, GR64, asextloadi32, 4>;
689 def SG : BinaryRXY<"sg", 0xE309, sub, GR64, load, 8>;
691 defm : SXB<sub, GR64, SGFR>;
693 // Subtraction producing a carry.
695 // Subtraction of a register.
696 defm SLR : BinaryRRAndK<"sl", 0x1F, 0xB9FB, subc, GR32, GR32>;
697 def SLGFR : BinaryRRE<"slgf", 0xB91B, null_frag, GR64, GR32>;
698 defm SLGR : BinaryRREAndK<"slg", 0xB90B, 0xB9EB, subc, GR64, GR64>;
700 // Subtraction of unsigned 32-bit immediates. These don't match
701 // subc because we prefer addc for constants.
702 def SLFI : BinaryRIL<"slfi", 0xC25, null_frag, GR32, uimm32>;
703 def SLGFI : BinaryRIL<"slgfi", 0xC24, null_frag, GR64, imm64zx32>;
705 // Subtraction of memory.
706 defm SL : BinaryRXPair<"sl", 0x5F, 0xE35F, subc, GR32, load, 4>;
707 def SLGF : BinaryRXY<"slgf", 0xE31B, subc, GR64, azextloadi32, 4>;
708 def SLG : BinaryRXY<"slg", 0xE30B, subc, GR64, load, 8>;
710 defm : ZXB<subc, GR64, SLGFR>;
712 // Subtraction producing and using a carry.
713 let Defs = [CC], Uses = [CC] in {
714 // Subtraction of a register.
715 def SLBR : BinaryRRE<"slb", 0xB999, sube, GR32, GR32>;
716 def SLGBR : BinaryRRE<"slbg", 0xB989, sube, GR64, GR64>;
718 // Subtraction of memory.
719 def SLB : BinaryRXY<"slb", 0xE399, sube, GR32, load, 4>;
720 def SLBG : BinaryRXY<"slbg", 0xE389, sube, GR64, load, 8>;
723 //===----------------------------------------------------------------------===//
725 //===----------------------------------------------------------------------===//
728 // ANDs of a register.
729 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
730 defm NR : BinaryRRAndK<"n", 0x14, 0xB9F4, and, GR32, GR32>;
731 defm NGR : BinaryRREAndK<"ng", 0xB980, 0xB9E4, and, GR64, GR64>;
734 let isConvertibleToThreeAddress = 1 in {
735 // ANDs of a 16-bit immediate, leaving other bits unaffected.
736 // The CC result only reflects the 16-bit field, not the full register.
737 let isCodeGenOnly = 1 in {
738 def NILL32 : BinaryRI<"nill", 0xA57, and, GR32, imm32ll16c>;
739 def NILH32 : BinaryRI<"nilh", 0xA56, and, GR32, imm32lh16c>;
741 def NILL : BinaryRI<"nill", 0xA57, and, GR64, imm64ll16c>;
742 def NILH : BinaryRI<"nilh", 0xA56, and, GR64, imm64lh16c>;
743 def NIHL : BinaryRI<"nihl", 0xA55, and, GR64, imm64hl16c>;
744 def NIHH : BinaryRI<"nihh", 0xA54, and, GR64, imm64hh16c>;
746 // ANDs of a 32-bit immediate, leaving other bits unaffected.
747 // The CC result only reflects the 32-bit field, which means we can
748 // use it as a zero indicator for i32 operations but not otherwise.
749 let isCodeGenOnly = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in
750 def NILF32 : BinaryRIL<"nilf", 0xC0B, and, GR32, uimm32>;
751 def NILF : BinaryRIL<"nilf", 0xC0B, and, GR64, imm64lf32c>;
752 def NIHF : BinaryRIL<"nihf", 0xC0A, and, GR64, imm64hf32c>;
756 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
757 defm N : BinaryRXPair<"n", 0x54, 0xE354, and, GR32, load, 4>;
758 def NG : BinaryRXY<"ng", 0xE380, and, GR64, load, 8>;
762 defm NI : BinarySIPair<"ni", 0x94, 0xEB54, null_frag, uimm8>;
765 let mayLoad = 1, mayStore = 1 in
766 defm NC : MemorySS<"nc", 0xD4, z_nc, z_nc_loop>;
768 defm : RMWIByte<and, bdaddr12pair, NI>;
769 defm : RMWIByte<and, bdaddr20pair, NIY>;
771 //===----------------------------------------------------------------------===//
773 //===----------------------------------------------------------------------===//
776 // ORs of a register.
777 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
778 defm OR : BinaryRRAndK<"o", 0x16, 0xB9F6, or, GR32, GR32>;
779 defm OGR : BinaryRREAndK<"og", 0xB981, 0xB9E6, or, GR64, GR64>;
782 // ORs of a 16-bit immediate, leaving other bits unaffected.
783 // The CC result only reflects the 16-bit field, not the full register.
784 let isCodeGenOnly = 1 in {
785 def OILL32 : BinaryRI<"oill", 0xA5B, or, GR32, imm32ll16>;
786 def OILH32 : BinaryRI<"oilh", 0xA5A, or, GR32, imm32lh16>;
788 def OILL : BinaryRI<"oill", 0xA5B, or, GR64, imm64ll16>;
789 def OILH : BinaryRI<"oilh", 0xA5A, or, GR64, imm64lh16>;
790 def OIHL : BinaryRI<"oihl", 0xA59, or, GR64, imm64hl16>;
791 def OIHH : BinaryRI<"oihh", 0xA58, or, GR64, imm64hh16>;
793 // ORs of a 32-bit immediate, leaving other bits unaffected.
794 // The CC result only reflects the 32-bit field, which means we can
795 // use it as a zero indicator for i32 operations but not otherwise.
796 let isCodeGenOnly = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in
797 def OILF32 : BinaryRIL<"oilf", 0xC0D, or, GR32, uimm32>;
798 def OILF : BinaryRIL<"oilf", 0xC0D, or, GR64, imm64lf32>;
799 def OIHF : BinaryRIL<"oihf", 0xC0C, or, GR64, imm64hf32>;
802 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
803 defm O : BinaryRXPair<"o", 0x56, 0xE356, or, GR32, load, 4>;
804 def OG : BinaryRXY<"og", 0xE381, or, GR64, load, 8>;
808 defm OI : BinarySIPair<"oi", 0x96, 0xEB56, null_frag, uimm8>;
811 let mayLoad = 1, mayStore = 1 in
812 defm OC : MemorySS<"oc", 0xD6, z_oc, z_oc_loop>;
814 defm : RMWIByte<or, bdaddr12pair, OI>;
815 defm : RMWIByte<or, bdaddr20pair, OIY>;
817 //===----------------------------------------------------------------------===//
819 //===----------------------------------------------------------------------===//
822 // XORs of a register.
823 let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
824 defm XR : BinaryRRAndK<"x", 0x17, 0xB9F7, xor, GR32, GR32>;
825 defm XGR : BinaryRREAndK<"xg", 0xB982, 0xB9E7, xor, GR64, GR64>;
828 // XORs of a 32-bit immediate, leaving other bits unaffected.
829 // The CC result only reflects the 32-bit field, which means we can
830 // use it as a zero indicator for i32 operations but not otherwise.
831 let isCodeGenOnly = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in
832 def XILF32 : BinaryRIL<"xilf", 0xC07, xor, GR32, uimm32>;
833 def XILF : BinaryRIL<"xilf", 0xC07, xor, GR64, imm64lf32>;
834 def XIHF : BinaryRIL<"xihf", 0xC06, xor, GR64, imm64hf32>;
837 let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
838 defm X : BinaryRXPair<"x",0x57, 0xE357, xor, GR32, load, 4>;
839 def XG : BinaryRXY<"xg", 0xE382, xor, GR64, load, 8>;
843 defm XI : BinarySIPair<"xi", 0x97, 0xEB57, null_frag, uimm8>;
846 let mayLoad = 1, mayStore = 1 in
847 defm XC : MemorySS<"xc", 0xD7, z_xc, z_xc_loop>;
849 defm : RMWIByte<xor, bdaddr12pair, XI>;
850 defm : RMWIByte<xor, bdaddr20pair, XIY>;
852 //===----------------------------------------------------------------------===//
854 //===----------------------------------------------------------------------===//
856 // Multiplication of a register.
857 let isCommutable = 1 in {
858 def MSR : BinaryRRE<"ms", 0xB252, mul, GR32, GR32>;
859 def MSGR : BinaryRRE<"msg", 0xB90C, mul, GR64, GR64>;
861 def MSGFR : BinaryRRE<"msgf", 0xB91C, null_frag, GR64, GR32>;
862 defm : SXB<mul, GR64, MSGFR>;
864 // Multiplication of a signed 16-bit immediate.
865 def MHI : BinaryRI<"mhi", 0xA7C, mul, GR32, imm32sx16>;
866 def MGHI : BinaryRI<"mghi", 0xA7D, mul, GR64, imm64sx16>;
868 // Multiplication of a signed 32-bit immediate.
869 def MSFI : BinaryRIL<"msfi", 0xC21, mul, GR32, simm32>;
870 def MSGFI : BinaryRIL<"msgfi", 0xC20, mul, GR64, imm64sx32>;
872 // Multiplication of memory.
873 defm MH : BinaryRXPair<"mh", 0x4C, 0xE37C, mul, GR32, asextloadi16, 2>;
874 defm MS : BinaryRXPair<"ms", 0x71, 0xE351, mul, GR32, load, 4>;
875 def MSGF : BinaryRXY<"msgf", 0xE31C, mul, GR64, asextloadi32, 4>;
876 def MSG : BinaryRXY<"msg", 0xE30C, mul, GR64, load, 8>;
878 // Multiplication of a register, producing two results.
879 def MLGR : BinaryRRE<"mlg", 0xB986, z_umul_lohi64, GR128, GR64>;
881 // Multiplication of memory, producing two results.
882 def MLG : BinaryRXY<"mlg", 0xE386, z_umul_lohi64, GR128, load, 8>;
884 //===----------------------------------------------------------------------===//
885 // Division and remainder
886 //===----------------------------------------------------------------------===//
888 // Division and remainder, from registers.
889 def DSGFR : BinaryRRE<"dsgf", 0xB91D, z_sdivrem32, GR128, GR32>;
890 def DSGR : BinaryRRE<"dsg", 0xB90D, z_sdivrem64, GR128, GR64>;
891 def DLR : BinaryRRE<"dl", 0xB997, z_udivrem32, GR128, GR32>;
892 def DLGR : BinaryRRE<"dlg", 0xB987, z_udivrem64, GR128, GR64>;
894 // Division and remainder, from memory.
895 def DSGF : BinaryRXY<"dsgf", 0xE31D, z_sdivrem32, GR128, load, 4>;
896 def DSG : BinaryRXY<"dsg", 0xE30D, z_sdivrem64, GR128, load, 8>;
897 def DL : BinaryRXY<"dl", 0xE397, z_udivrem32, GR128, load, 4>;
898 def DLG : BinaryRXY<"dlg", 0xE387, z_udivrem64, GR128, load, 8>;
900 //===----------------------------------------------------------------------===//
902 //===----------------------------------------------------------------------===//
905 let neverHasSideEffects = 1 in {
906 defm SLL : ShiftRSAndK<"sll", 0x89, 0xEBDF, shl, GR32>;
907 def SLLG : ShiftRSY<"sllg", 0xEB0D, shl, GR64>;
910 // Logical shift right.
911 let neverHasSideEffects = 1 in {
912 defm SRL : ShiftRSAndK<"srl", 0x88, 0xEBDE, srl, GR32>;
913 def SRLG : ShiftRSY<"srlg", 0xEB0C, srl, GR64>;
916 // Arithmetic shift right.
917 let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
918 defm SRA : ShiftRSAndK<"sra", 0x8A, 0xEBDC, sra, GR32>;
919 def SRAG : ShiftRSY<"srag", 0xEB0A, sra, GR64>;
923 let neverHasSideEffects = 1 in {
924 def RLL : ShiftRSY<"rll", 0xEB1D, rotl, GR32>;
925 def RLLG : ShiftRSY<"rllg", 0xEB1C, rotl, GR64>;
928 // Rotate second operand left and inserted selected bits into first operand.
929 // These can act like 32-bit operands provided that the constant start and
930 // end bits (operands 2 and 3) are in the range [32, 64).
932 let isCodeGenOnly = 1 in
933 def RISBG32 : RotateSelectRIEf<"risbg", 0xEC55, GR32, GR32>;
934 let CCValues = 0xE, CompareZeroCCMask = 0xE in
935 def RISBG : RotateSelectRIEf<"risbg", 0xEC55, GR64, GR64>;
938 // Forms of RISBG that only affect one word of the destination register.
939 // They do not set CC.
940 let isCodeGenOnly = 1 in
941 def RISBLG32 : RotateSelectRIEf<"risblg", 0xEC51, GR32, GR32>,
942 Requires<[FeatureHighWord]>;
943 def RISBHG : RotateSelectRIEf<"risbhg", 0xEC5D, GR64, GR64>,
944 Requires<[FeatureHighWord]>;
945 def RISBLG : RotateSelectRIEf<"risblg", 0xEC51, GR64, GR64>,
946 Requires<[FeatureHighWord]>;
948 // Rotate second operand left and perform a logical operation with selected
949 // bits of the first operand. The CC result only describes the selected bits,
950 // so isn't useful for a full comparison against zero.
952 def RNSBG : RotateSelectRIEf<"rnsbg", 0xEC54, GR64, GR64>;
953 def ROSBG : RotateSelectRIEf<"rosbg", 0xEC56, GR64, GR64>;
954 def RXSBG : RotateSelectRIEf<"rxsbg", 0xEC57, GR64, GR64>;
957 //===----------------------------------------------------------------------===//
959 //===----------------------------------------------------------------------===//
961 // Signed comparisons. We put these before the unsigned comparisons because
962 // some of the signed forms have COMPARE AND BRANCH equivalents whereas none
963 // of the unsigned forms do.
964 let Defs = [CC], CCValues = 0xE in {
965 // Comparison with a register.
966 def CR : CompareRR <"c", 0x19, z_scmp, GR32, GR32>;
967 def CGFR : CompareRRE<"cgf", 0xB930, null_frag, GR64, GR32>;
968 def CGR : CompareRRE<"cg", 0xB920, z_scmp, GR64, GR64>;
970 // Comparison with a signed 16-bit immediate.
971 def CHI : CompareRI<"chi", 0xA7E, z_scmp, GR32, imm32sx16>;
972 def CGHI : CompareRI<"cghi", 0xA7F, z_scmp, GR64, imm64sx16>;
974 // Comparison with a signed 32-bit immediate.
975 def CFI : CompareRIL<"cfi", 0xC2D, z_scmp, GR32, simm32>;
976 def CGFI : CompareRIL<"cgfi", 0xC2C, z_scmp, GR64, imm64sx32>;
978 // Comparison with memory.
979 defm CH : CompareRXPair<"ch", 0x49, 0xE379, z_scmp, GR32, asextloadi16, 2>;
980 defm C : CompareRXPair<"c", 0x59, 0xE359, z_scmp, GR32, load, 4>;
981 def CGH : CompareRXY<"cgh", 0xE334, z_scmp, GR64, asextloadi16, 2>;
982 def CGF : CompareRXY<"cgf", 0xE330, z_scmp, GR64, asextloadi32, 4>;
983 def CG : CompareRXY<"cg", 0xE320, z_scmp, GR64, load, 8>;
984 def CHRL : CompareRILPC<"chrl", 0xC65, z_scmp, GR32, aligned_asextloadi16>;
985 def CRL : CompareRILPC<"crl", 0xC6D, z_scmp, GR32, aligned_load>;
986 def CGHRL : CompareRILPC<"cghrl", 0xC64, z_scmp, GR64, aligned_asextloadi16>;
987 def CGFRL : CompareRILPC<"cgfrl", 0xC6C, z_scmp, GR64, aligned_asextloadi32>;
988 def CGRL : CompareRILPC<"cgrl", 0xC68, z_scmp, GR64, aligned_load>;
990 // Comparison between memory and a signed 16-bit immediate.
991 def CHHSI : CompareSIL<"chhsi", 0xE554, z_scmp, asextloadi16, imm32sx16>;
992 def CHSI : CompareSIL<"chsi", 0xE55C, z_scmp, load, imm32sx16>;
993 def CGHSI : CompareSIL<"cghsi", 0xE558, z_scmp, load, imm64sx16>;
995 defm : SXB<z_scmp, GR64, CGFR>;
997 // Unsigned comparisons.
998 let Defs = [CC], CCValues = 0xE, IsLogical = 1 in {
999 // Comparison with a register.
1000 def CLR : CompareRR <"cl", 0x15, z_ucmp, GR32, GR32>;
1001 def CLGFR : CompareRRE<"clgf", 0xB931, null_frag, GR64, GR32>;
1002 def CLGR : CompareRRE<"clg", 0xB921, z_ucmp, GR64, GR64>;
1004 // Comparison with a signed 32-bit immediate.
1005 def CLFI : CompareRIL<"clfi", 0xC2F, z_ucmp, GR32, uimm32>;
1006 def CLGFI : CompareRIL<"clgfi", 0xC2E, z_ucmp, GR64, imm64zx32>;
1008 // Comparison with memory.
1009 defm CL : CompareRXPair<"cl", 0x55, 0xE355, z_ucmp, GR32, load, 4>;
1010 def CLGF : CompareRXY<"clgf", 0xE331, z_ucmp, GR64, azextloadi32, 4>;
1011 def CLG : CompareRXY<"clg", 0xE321, z_ucmp, GR64, load, 8>;
1012 def CLHRL : CompareRILPC<"clhrl", 0xC67, z_ucmp, GR32,
1013 aligned_azextloadi16>;
1014 def CLRL : CompareRILPC<"clrl", 0xC6F, z_ucmp, GR32,
1016 def CLGHRL : CompareRILPC<"clghrl", 0xC66, z_ucmp, GR64,
1017 aligned_azextloadi16>;
1018 def CLGFRL : CompareRILPC<"clgfrl", 0xC6E, z_ucmp, GR64,
1019 aligned_azextloadi32>;
1020 def CLGRL : CompareRILPC<"clgrl", 0xC6A, z_ucmp, GR64,
1023 // Comparison between memory and an unsigned 8-bit immediate.
1024 defm CLI : CompareSIPair<"cli", 0x95, 0xEB55, z_ucmp, azextloadi8, imm32zx8>;
1026 // Comparison between memory and an unsigned 16-bit immediate.
1027 def CLHHSI : CompareSIL<"clhhsi", 0xE555, z_ucmp, azextloadi16, imm32zx16>;
1028 def CLFHSI : CompareSIL<"clfhsi", 0xE55D, z_ucmp, load, imm32zx16>;
1029 def CLGHSI : CompareSIL<"clghsi", 0xE559, z_ucmp, load, imm64zx16>;
1031 defm : ZXB<z_ucmp, GR64, CLGFR>;
1033 // Memory-to-memory comparison.
1034 let mayLoad = 1, Defs = [CC] in
1035 defm CLC : MemorySS<"clc", 0xD5, z_clc, z_clc_loop>;
1037 // String comparison.
1038 let mayLoad = 1, Defs = [CC], Uses = [R0W] in
1039 defm CLST : StringRRE<"clst", 0xB25D, z_strcmp>;
1042 let Defs = [CC] in {
1043 let isCodeGenOnly = 1 in {
1044 def TMLL32 : CompareRI<"tmll", 0xA71, z_tm_reg, GR32, imm32ll16>;
1045 def TMLH32 : CompareRI<"tmlh", 0xA70, z_tm_reg, GR32, imm32lh16>;
1048 def TMLL : CompareRI<"tmll", 0xA71, z_tm_reg, GR64, imm64ll16>;
1049 def TMLH : CompareRI<"tmlh", 0xA70, z_tm_reg, GR64, imm64lh16>;
1050 def TMHL : CompareRI<"tmhl", 0xA73, z_tm_reg, GR64, imm64hl16>;
1051 def TMHH : CompareRI<"tmhh", 0xA72, z_tm_reg, GR64, imm64hh16>;
1053 defm TM : CompareSIPair<"tm", 0x91, 0xEB51, z_tm_mem, anyextloadi8, imm32zx8>;
1056 //===----------------------------------------------------------------------===//
1058 //===----------------------------------------------------------------------===//
1060 def PFD : PrefetchRXY<"pfd", 0xE336, z_prefetch>;
1061 def PFDRL : PrefetchRILPC<"pfdrl", 0xC62, z_prefetch>;
1063 //===----------------------------------------------------------------------===//
1064 // Atomic operations
1065 //===----------------------------------------------------------------------===//
1067 def ATOMIC_SWAPW : AtomicLoadWBinaryReg<z_atomic_swapw>;
1068 def ATOMIC_SWAP_32 : AtomicLoadBinaryReg32<atomic_swap_32>;
1069 def ATOMIC_SWAP_64 : AtomicLoadBinaryReg64<atomic_swap_64>;
1071 def ATOMIC_LOADW_AR : AtomicLoadWBinaryReg<z_atomic_loadw_add>;
1072 def ATOMIC_LOADW_AFI : AtomicLoadWBinaryImm<z_atomic_loadw_add, simm32>;
1073 def ATOMIC_LOAD_AR : AtomicLoadBinaryReg32<atomic_load_add_32>;
1074 def ATOMIC_LOAD_AHI : AtomicLoadBinaryImm32<atomic_load_add_32, imm32sx16>;
1075 def ATOMIC_LOAD_AFI : AtomicLoadBinaryImm32<atomic_load_add_32, simm32>;
1076 def ATOMIC_LOAD_AGR : AtomicLoadBinaryReg64<atomic_load_add_64>;
1077 def ATOMIC_LOAD_AGHI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx16>;
1078 def ATOMIC_LOAD_AGFI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx32>;
1080 def ATOMIC_LOADW_SR : AtomicLoadWBinaryReg<z_atomic_loadw_sub>;
1081 def ATOMIC_LOAD_SR : AtomicLoadBinaryReg32<atomic_load_sub_32>;
1082 def ATOMIC_LOAD_SGR : AtomicLoadBinaryReg64<atomic_load_sub_64>;
1084 def ATOMIC_LOADW_NR : AtomicLoadWBinaryReg<z_atomic_loadw_and>;
1085 def ATOMIC_LOADW_NILH : AtomicLoadWBinaryImm<z_atomic_loadw_and, imm32lh16c>;
1086 def ATOMIC_LOAD_NR : AtomicLoadBinaryReg32<atomic_load_and_32>;
1087 def ATOMIC_LOAD_NILL32 : AtomicLoadBinaryImm32<atomic_load_and_32, imm32ll16c>;
1088 def ATOMIC_LOAD_NILH32 : AtomicLoadBinaryImm32<atomic_load_and_32, imm32lh16c>;
1089 def ATOMIC_LOAD_NILF32 : AtomicLoadBinaryImm32<atomic_load_and_32, uimm32>;
1090 def ATOMIC_LOAD_NGR : AtomicLoadBinaryReg64<atomic_load_and_64>;
1091 def ATOMIC_LOAD_NILL : AtomicLoadBinaryImm64<atomic_load_and_64, imm64ll16c>;
1092 def ATOMIC_LOAD_NILH : AtomicLoadBinaryImm64<atomic_load_and_64, imm64lh16c>;
1093 def ATOMIC_LOAD_NIHL : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hl16c>;
1094 def ATOMIC_LOAD_NIHH : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hh16c>;
1095 def ATOMIC_LOAD_NILF : AtomicLoadBinaryImm64<atomic_load_and_64, imm64lf32c>;
1096 def ATOMIC_LOAD_NIHF : AtomicLoadBinaryImm64<atomic_load_and_64, imm64hf32c>;
1098 def ATOMIC_LOADW_OR : AtomicLoadWBinaryReg<z_atomic_loadw_or>;
1099 def ATOMIC_LOADW_OILH : AtomicLoadWBinaryImm<z_atomic_loadw_or, imm32lh16>;
1100 def ATOMIC_LOAD_OR : AtomicLoadBinaryReg32<atomic_load_or_32>;
1101 def ATOMIC_LOAD_OILL32 : AtomicLoadBinaryImm32<atomic_load_or_32, imm32ll16>;
1102 def ATOMIC_LOAD_OILH32 : AtomicLoadBinaryImm32<atomic_load_or_32, imm32lh16>;
1103 def ATOMIC_LOAD_OILF32 : AtomicLoadBinaryImm32<atomic_load_or_32, uimm32>;
1104 def ATOMIC_LOAD_OGR : AtomicLoadBinaryReg64<atomic_load_or_64>;
1105 def ATOMIC_LOAD_OILL : AtomicLoadBinaryImm64<atomic_load_or_64, imm64ll16>;
1106 def ATOMIC_LOAD_OILH : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lh16>;
1107 def ATOMIC_LOAD_OIHL : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hl16>;
1108 def ATOMIC_LOAD_OIHH : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hh16>;
1109 def ATOMIC_LOAD_OILF : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lf32>;
1110 def ATOMIC_LOAD_OIHF : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hf32>;
1112 def ATOMIC_LOADW_XR : AtomicLoadWBinaryReg<z_atomic_loadw_xor>;
1113 def ATOMIC_LOADW_XILF : AtomicLoadWBinaryImm<z_atomic_loadw_xor, uimm32>;
1114 def ATOMIC_LOAD_XR : AtomicLoadBinaryReg32<atomic_load_xor_32>;
1115 def ATOMIC_LOAD_XILF32 : AtomicLoadBinaryImm32<atomic_load_xor_32, uimm32>;
1116 def ATOMIC_LOAD_XGR : AtomicLoadBinaryReg64<atomic_load_xor_64>;
1117 def ATOMIC_LOAD_XILF : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64lf32>;
1118 def ATOMIC_LOAD_XIHF : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64hf32>;
1120 def ATOMIC_LOADW_NRi : AtomicLoadWBinaryReg<z_atomic_loadw_nand>;
1121 def ATOMIC_LOADW_NILHi : AtomicLoadWBinaryImm<z_atomic_loadw_nand,
1123 def ATOMIC_LOAD_NRi : AtomicLoadBinaryReg32<atomic_load_nand_32>;
1124 def ATOMIC_LOAD_NILL32i : AtomicLoadBinaryImm32<atomic_load_nand_32,
1126 def ATOMIC_LOAD_NILH32i : AtomicLoadBinaryImm32<atomic_load_nand_32,
1128 def ATOMIC_LOAD_NILF32i : AtomicLoadBinaryImm32<atomic_load_nand_32, uimm32>;
1129 def ATOMIC_LOAD_NGRi : AtomicLoadBinaryReg64<atomic_load_nand_64>;
1130 def ATOMIC_LOAD_NILLi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1132 def ATOMIC_LOAD_NILHi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1134 def ATOMIC_LOAD_NIHLi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1136 def ATOMIC_LOAD_NIHHi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1138 def ATOMIC_LOAD_NILFi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1140 def ATOMIC_LOAD_NIHFi : AtomicLoadBinaryImm64<atomic_load_nand_64,
1143 def ATOMIC_LOADW_MIN : AtomicLoadWBinaryReg<z_atomic_loadw_min>;
1144 def ATOMIC_LOAD_MIN_32 : AtomicLoadBinaryReg32<atomic_load_min_32>;
1145 def ATOMIC_LOAD_MIN_64 : AtomicLoadBinaryReg64<atomic_load_min_64>;
1147 def ATOMIC_LOADW_MAX : AtomicLoadWBinaryReg<z_atomic_loadw_max>;
1148 def ATOMIC_LOAD_MAX_32 : AtomicLoadBinaryReg32<atomic_load_max_32>;
1149 def ATOMIC_LOAD_MAX_64 : AtomicLoadBinaryReg64<atomic_load_max_64>;
1151 def ATOMIC_LOADW_UMIN : AtomicLoadWBinaryReg<z_atomic_loadw_umin>;
1152 def ATOMIC_LOAD_UMIN_32 : AtomicLoadBinaryReg32<atomic_load_umin_32>;
1153 def ATOMIC_LOAD_UMIN_64 : AtomicLoadBinaryReg64<atomic_load_umin_64>;
1155 def ATOMIC_LOADW_UMAX : AtomicLoadWBinaryReg<z_atomic_loadw_umax>;
1156 def ATOMIC_LOAD_UMAX_32 : AtomicLoadBinaryReg32<atomic_load_umax_32>;
1157 def ATOMIC_LOAD_UMAX_64 : AtomicLoadBinaryReg64<atomic_load_umax_64>;
1159 def ATOMIC_CMP_SWAPW
1160 : Pseudo<(outs GR32:$dst), (ins bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1161 ADDR32:$bitshift, ADDR32:$negbitshift,
1164 (z_atomic_cmp_swapw bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1165 ADDR32:$bitshift, ADDR32:$negbitshift,
1166 uimm32:$bitsize))]> {
1170 let usesCustomInserter = 1;
1173 let Defs = [CC] in {
1174 defm CS : CmpSwapRSPair<"cs", 0xBA, 0xEB14, atomic_cmp_swap_32, GR32>;
1175 def CSG : CmpSwapRSY<"csg", 0xEB30, atomic_cmp_swap_64, GR64>;
1178 //===----------------------------------------------------------------------===//
1179 // Miscellaneous Instructions.
1180 //===----------------------------------------------------------------------===//
1182 // Extract CC into bits 29 and 28 of a register.
1184 def IPM : InherentRRE<"ipm", 0xB222, GR32, (z_ipm)>;
1186 // Read a 32-bit access register into a GR32. As with all GR32 operations,
1187 // the upper 32 bits of the enclosing GR64 remain unchanged, which is useful
1188 // when a 64-bit address is stored in a pair of access registers.
1189 def EAR : InstRRE<0xB24F, (outs GR32:$R1), (ins access_reg:$R2),
1191 [(set GR32:$R1, (z_extract_access access_reg:$R2))]>;
1193 // Find leftmost one, AKA count leading zeros. The instruction actually
1194 // returns a pair of GR64s, the first giving the number of leading zeros
1195 // and the second giving a copy of the source with the leftmost one bit
1196 // cleared. We only use the first result here.
1197 let Defs = [CC] in {
1198 def FLOGR : UnaryRRE<"flog", 0xB983, null_frag, GR128, GR64>;
1200 def : Pat<(ctlz GR64:$src),
1201 (EXTRACT_SUBREG (FLOGR GR64:$src), subreg_high)>;
1203 // Use subregs to populate the "don't care" bits in a 32-bit to 64-bit anyext.
1204 def : Pat<(i64 (anyext GR32:$src)),
1205 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_32bit)>;
1207 // Extend GR32s and GR64s to GR128s.
1208 let usesCustomInserter = 1 in {
1209 def AEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1210 def ZEXT128_32 : Pseudo<(outs GR128:$dst), (ins GR32:$src), []>;
1211 def ZEXT128_64 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
1214 // Search a block of memory for a character.
1215 let mayLoad = 1, Defs = [CC], Uses = [R0W] in
1216 defm SRST : StringRRE<"srst", 0xb25e, z_search_string>;
1218 //===----------------------------------------------------------------------===//
1220 //===----------------------------------------------------------------------===//
1222 // Use AL* for GR64 additions of unsigned 32-bit values.
1223 defm : ZXB<add, GR64, ALGFR>;
1224 def : Pat<(add GR64:$src1, imm64zx32:$src2),
1225 (ALGFI GR64:$src1, imm64zx32:$src2)>;
1226 def : Pat<(add GR64:$src1, (azextloadi32 bdxaddr20only:$addr)),
1227 (ALGF GR64:$src1, bdxaddr20only:$addr)>;
1229 // Use SL* for GR64 subtractions of unsigned 32-bit values.
1230 defm : ZXB<sub, GR64, SLGFR>;
1231 def : Pat<(add GR64:$src1, imm64zx32n:$src2),
1232 (SLGFI GR64:$src1, imm64zx32n:$src2)>;
1233 def : Pat<(sub GR64:$src1, (azextloadi32 bdxaddr20only:$addr)),
1234 (SLGF GR64:$src1, bdxaddr20only:$addr)>;
1236 // Optimize sign-extended 1/0 selects to -1/0 selects. This is important
1237 // for vector legalization.
1238 def : Pat<(sra (shl (i32 (z_select_ccmask 1, 0, uimm8zx4:$valid, uimm8zx4:$cc)),
1241 (Select32 (LHI -1), (LHI 0), uimm8zx4:$valid, uimm8zx4:$cc)>;
1242 def : Pat<(sra (shl (i64 (anyext (i32 (z_select_ccmask 1, 0, uimm8zx4:$valid,
1246 (Select64 (LGHI -1), (LGHI 0), uimm8zx4:$valid, uimm8zx4:$cc)>;
1248 // Peepholes for turning scalar operations into block operations.
1249 defm : BlockLoadStore<anyextloadi8, i32, MVCSequence, NCSequence, OCSequence,
1251 defm : BlockLoadStore<anyextloadi16, i32, MVCSequence, NCSequence, OCSequence,
1253 defm : BlockLoadStore<load, i32, MVCSequence, NCSequence, OCSequence,
1255 defm : BlockLoadStore<anyextloadi8, i64, MVCSequence, NCSequence,
1256 OCSequence, XCSequence, 1>;
1257 defm : BlockLoadStore<anyextloadi16, i64, MVCSequence, NCSequence, OCSequence,
1259 defm : BlockLoadStore<anyextloadi32, i64, MVCSequence, NCSequence, OCSequence,
1261 defm : BlockLoadStore<load, i64, MVCSequence, NCSequence, OCSequence,