1//===-- SystemZInstrInfo.td - General SystemZ instructions ----*- tblgen-*-===//
2//
3//                     The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9
10//===----------------------------------------------------------------------===//
11// Stack allocation
12//===----------------------------------------------------------------------===//
13
14let hasNoSchedulingInfo = 1 in {
15  def ADJCALLSTACKDOWN : Pseudo<(outs), (ins i64imm:$amt1, i64imm:$amt2),
16                                [(callseq_start timm:$amt1, timm:$amt2)]>;
17  def ADJCALLSTACKUP   : Pseudo<(outs), (ins i64imm:$amt1, i64imm:$amt2),
18                                [(callseq_end timm:$amt1, timm:$amt2)]>;
19}
20
21let hasSideEffects = 0 in {
22  // Takes as input the value of the stack pointer after a dynamic allocation
23  // has been made.  Sets the output to the address of the dynamically-
24  // allocated area itself, skipping the outgoing arguments.
25  //
26  // This expands to an LA or LAY instruction.  We restrict the offset
27  // to the range of LA and keep the LAY range in reserve for when
28  // the size of the outgoing arguments is added.
29  def ADJDYNALLOC : Pseudo<(outs GR64:$dst), (ins dynalloc12only:$src),
30                           [(set GR64:$dst, dynalloc12only:$src)]>;
31}
32
33//===----------------------------------------------------------------------===//
34// Branch instructions
35//===----------------------------------------------------------------------===//
36
37// Conditional branches.
38let isBranch = 1, isTerminator = 1, Uses = [CC] in {
39  // It's easier for LLVM to handle these branches in their raw BRC/BRCL form
40  // with the condition-code mask being the first operand.  It seems friendlier
41  // to use mnemonic forms like JE and JLH when writing out the assembly though.
42  let isCodeGenOnly = 1 in {
43    // An assembler extended mnemonic for BRC.
44    def BRC  : CondBranchRI <"j#",  0xA74, z_br_ccmask>;
45    // An assembler extended mnemonic for BRCL.  (The extension is "G"
46    // rather than "L" because "JL" is "Jump if Less".)
47    def BRCL : CondBranchRIL<"jg#", 0xC04>;
48    let isIndirectBranch = 1 in {
49      def BC  : CondBranchRX<"b#",  0x47>;
50      def BCR : CondBranchRR<"b#r", 0x07>;
51    }
52  }
53
54  // Allow using the raw forms directly from the assembler (and occasional
55  // special code generation needs) as well.
56  def BRCAsm  : AsmCondBranchRI <"brc",  0xA74>;
57  def BRCLAsm : AsmCondBranchRIL<"brcl", 0xC04>;
58  let isIndirectBranch = 1 in {
59    def BCAsm  : AsmCondBranchRX<"bc",  0x47>;
60    def BCRAsm : AsmCondBranchRR<"bcr", 0x07>;
61  }
62
63  // Define AsmParser extended mnemonics for each general condition-code mask
64  // (integer or floating-point)
65  foreach V = [ "E", "NE", "H", "NH", "L", "NL", "HE", "NHE", "LE", "NLE",
66                "Z", "NZ", "P", "NP", "M", "NM", "LH", "NLH", "O", "NO" ] in {
67    def JAsm#V  : FixedCondBranchRI <CV<V>, "j#",  0xA74>;
68    def JGAsm#V : FixedCondBranchRIL<CV<V>, "jg#", 0xC04>;
69    let isIndirectBranch = 1 in {
70      def BAsm#V  : FixedCondBranchRX <CV<V>, "b#",  0x47>;
71      def BRAsm#V : FixedCondBranchRR <CV<V>, "b#r", 0x07>;
72    }
73  }
74}
75
76// Unconditional branches.  These are in fact simply variants of the
77// conditional branches with the condition mask set to "always".
78let isBranch = 1, isTerminator = 1, isBarrier = 1 in {
79  def J  : FixedCondBranchRI <CondAlways, "j",  0xA74, br>;
80  def JG : FixedCondBranchRIL<CondAlways, "jg", 0xC04>;
81  let isIndirectBranch = 1 in {
82    def B  : FixedCondBranchRX<CondAlways, "b",  0x47>;
83    def BR : FixedCondBranchRR<CondAlways, "br", 0x07, brind>;
84  }
85}
86
87// NOPs.  These are again variants of the conditional branches,
88// with the condition mask set to "never".
89def NOP  : InstAlias<"nop\t$XBD", (BCAsm 0, bdxaddr12only:$XBD), 0>;
90def NOPR : InstAlias<"nopr\t$R", (BCRAsm 0, GR64:$R), 0>;
91
92// Fused compare-and-branch instructions.
93//
94// These instructions do not use or clobber the condition codes.
95// We nevertheless pretend that the relative compare-and-branch
96// instructions clobber CC, so that we can lower them to separate
97// comparisons and BRCLs if the branch ends up being out of range.
98let isBranch = 1, isTerminator = 1 in {
99  // As for normal branches, we handle these instructions internally in
100  // their raw CRJ-like form, but use assembly macros like CRJE when writing
101  // them out.  Using the *Pair multiclasses, we also create the raw forms.
102  let Defs = [CC] in {
103    defm CRJ   : CmpBranchRIEbPair<"crj",   0xEC76, GR32>;
104    defm CGRJ  : CmpBranchRIEbPair<"cgrj",  0xEC64, GR64>;
105    defm CIJ   : CmpBranchRIEcPair<"cij",   0xEC7E, GR32, imm32sx8>;
106    defm CGIJ  : CmpBranchRIEcPair<"cgij",  0xEC7C, GR64, imm64sx8>;
107    defm CLRJ  : CmpBranchRIEbPair<"clrj",  0xEC77, GR32>;
108    defm CLGRJ : CmpBranchRIEbPair<"clgrj", 0xEC65, GR64>;
109    defm CLIJ  : CmpBranchRIEcPair<"clij",  0xEC7F, GR32, imm32zx8>;
110    defm CLGIJ : CmpBranchRIEcPair<"clgij", 0xEC7D, GR64, imm64zx8>;
111  }
112  let isIndirectBranch = 1 in {
113    defm CRB   : CmpBranchRRSPair<"crb",   0xECF6, GR32>;
114    defm CGRB  : CmpBranchRRSPair<"cgrb",  0xECE4, GR64>;
115    defm CIB   : CmpBranchRISPair<"cib",   0xECFE, GR32, imm32sx8>;
116    defm CGIB  : CmpBranchRISPair<"cgib",  0xECFC, GR64, imm64sx8>;
117    defm CLRB  : CmpBranchRRSPair<"clrb",  0xECF7, GR32>;
118    defm CLGRB : CmpBranchRRSPair<"clgrb", 0xECE5, GR64>;
119    defm CLIB  : CmpBranchRISPair<"clib",  0xECFF, GR32, imm32zx8>;
120    defm CLGIB : CmpBranchRISPair<"clgib", 0xECFD, GR64, imm64zx8>;
121  }
122
123  // Define AsmParser mnemonics for each integer condition-code mask.
124  foreach V = [ "E", "H", "L", "HE", "LE", "LH",
125                "NE", "NH", "NL", "NHE", "NLE", "NLH" ] in {
126    let Defs = [CC] in {
127      def CRJAsm#V   : FixedCmpBranchRIEb<ICV<V>, "crj",   0xEC76, GR32>;
128      def CGRJAsm#V  : FixedCmpBranchRIEb<ICV<V>, "cgrj",  0xEC64, GR64>;
129      def CIJAsm#V   : FixedCmpBranchRIEc<ICV<V>, "cij",   0xEC7E, GR32,
130                                          imm32sx8>;
131      def CGIJAsm#V  : FixedCmpBranchRIEc<ICV<V>, "cgij",  0xEC7C, GR64,
132                                          imm64sx8>;
133      def CLRJAsm#V  : FixedCmpBranchRIEb<ICV<V>, "clrj",  0xEC77, GR32>;
134      def CLGRJAsm#V : FixedCmpBranchRIEb<ICV<V>, "clgrj", 0xEC65, GR64>;
135      def CLIJAsm#V  : FixedCmpBranchRIEc<ICV<V>, "clij",  0xEC7F, GR32,
136                                          imm32zx8>;
137      def CLGIJAsm#V : FixedCmpBranchRIEc<ICV<V>, "clgij", 0xEC7D, GR64,
138                                          imm64zx8>;
139    }
140    let isIndirectBranch = 1 in {
141      def CRBAsm#V   : FixedCmpBranchRRS<ICV<V>, "crb",   0xECF6, GR32>;
142      def CGRBAsm#V  : FixedCmpBranchRRS<ICV<V>, "cgrb",  0xECE4, GR64>;
143      def CIBAsm#V   : FixedCmpBranchRIS<ICV<V>, "cib",   0xECFE, GR32,
144                                         imm32sx8>;
145      def CGIBAsm#V  : FixedCmpBranchRIS<ICV<V>, "cgib",  0xECFC, GR64,
146                                         imm64sx8>;
147      def CLRBAsm#V  : FixedCmpBranchRRS<ICV<V>, "clrb",  0xECF7, GR32>;
148      def CLGRBAsm#V : FixedCmpBranchRRS<ICV<V>, "clgrb", 0xECE5, GR64>;
149      def CLIBAsm#V  : FixedCmpBranchRIS<ICV<V>, "clib",  0xECFF, GR32,
150                                         imm32zx8>;
151      def CLGIBAsm#V : FixedCmpBranchRIS<ICV<V>, "clgib", 0xECFD, GR64,
152                                         imm64zx8>;
153    }
154  }
155}
156
157// Decrement a register and branch if it is nonzero.  These don't clobber CC,
158// but we might need to split long relative branches into sequences that do.
159let isBranch = 1, isTerminator = 1 in {
160  let Defs = [CC] in {
161    def BRCT  : BranchUnaryRI<"brct",  0xA76, GR32>;
162    def BRCTG : BranchUnaryRI<"brctg", 0xA77, GR64>;
163  }
164  // This doesn't need to clobber CC since we never need to split it.
165  def BRCTH : BranchUnaryRIL<"brcth", 0xCC6, GRH32>,
166              Requires<[FeatureHighWord]>;
167
168  def BCT   : BranchUnaryRX<"bct",  0x46,GR32>;
169  def BCTR  : BranchUnaryRR<"bctr", 0x06, GR32>;
170  def BCTG  : BranchUnaryRXY<"bctg",  0xE346, GR64>;
171  def BCTGR : BranchUnaryRRE<"bctgr", 0xB946, GR64>;
172}
173
174let isBranch = 1, isTerminator = 1 in {
175  let Defs = [CC] in {
176    def BRXH  : BranchBinaryRSI<"brxh",  0x84, GR32>;
177    def BRXLE : BranchBinaryRSI<"brxle", 0x85, GR32>;
178    def BRXHG : BranchBinaryRIEe<"brxhg", 0xEC44, GR64>;
179    def BRXLG : BranchBinaryRIEe<"brxlg", 0xEC45, GR64>;
180  }
181  def BXH   : BranchBinaryRS<"bxh",  0x86, GR32>;
182  def BXLE  : BranchBinaryRS<"bxle", 0x87, GR32>;
183  def BXHG  : BranchBinaryRSY<"bxhg",  0xEB44, GR64>;
184  def BXLEG : BranchBinaryRSY<"bxleg", 0xEB45, GR64>;
185}
186
187//===----------------------------------------------------------------------===//
188// Trap instructions
189//===----------------------------------------------------------------------===//
190
191// Unconditional trap.
192let hasCtrlDep = 1 in
193  def Trap : Alias<4, (outs), (ins), [(trap)]>;
194
195// Conditional trap.
196let hasCtrlDep = 1, Uses = [CC] in
197  def CondTrap : Alias<4, (outs), (ins cond4:$valid, cond4:$R1), []>;
198
199// Fused compare-and-trap instructions.
200let hasCtrlDep = 1 in {
201  // These patterns work the same way as for compare-and-branch.
202  defm CRT   : CmpBranchRRFcPair<"crt",   0xB972, GR32>;
203  defm CGRT  : CmpBranchRRFcPair<"cgrt",  0xB960, GR64>;
204  defm CLRT  : CmpBranchRRFcPair<"clrt",  0xB973, GR32>;
205  defm CLGRT : CmpBranchRRFcPair<"clgrt", 0xB961, GR64>;
206  defm CIT   : CmpBranchRIEaPair<"cit",   0xEC72, GR32, imm32sx16>;
207  defm CGIT  : CmpBranchRIEaPair<"cgit",  0xEC70, GR64, imm64sx16>;
208  defm CLFIT : CmpBranchRIEaPair<"clfit", 0xEC73, GR32, imm32zx16>;
209  defm CLGIT : CmpBranchRIEaPair<"clgit", 0xEC71, GR64, imm64zx16>;
210  let Predicates = [FeatureMiscellaneousExtensions] in {
211    defm CLT  : CmpBranchRSYbPair<"clt",  0xEB23, GR32>;
212    defm CLGT : CmpBranchRSYbPair<"clgt", 0xEB2B, GR64>;
213  }
214
215  foreach V = [ "E", "H", "L", "HE", "LE", "LH",
216                "NE", "NH", "NL", "NHE", "NLE", "NLH" ] in {
217    def CRTAsm#V   : FixedCmpBranchRRFc<ICV<V>, "crt",   0xB972, GR32>;
218    def CGRTAsm#V  : FixedCmpBranchRRFc<ICV<V>, "cgrt",  0xB960, GR64>;
219    def CLRTAsm#V  : FixedCmpBranchRRFc<ICV<V>, "clrt",  0xB973, GR32>;
220    def CLGRTAsm#V : FixedCmpBranchRRFc<ICV<V>, "clgrt", 0xB961, GR64>;
221    def CITAsm#V   : FixedCmpBranchRIEa<ICV<V>, "cit",   0xEC72, GR32,
222                                         imm32sx16>;
223    def CGITAsm#V  : FixedCmpBranchRIEa<ICV<V>, "cgit",  0xEC70, GR64,
224                                         imm64sx16>;
225    def CLFITAsm#V : FixedCmpBranchRIEa<ICV<V>, "clfit", 0xEC73, GR32,
226                                         imm32zx16>;
227    def CLGITAsm#V : FixedCmpBranchRIEa<ICV<V>, "clgit", 0xEC71, GR64,
228                                         imm64zx16>;
229    let Predicates = [FeatureMiscellaneousExtensions] in {
230      def CLTAsm#V  : FixedCmpBranchRSYb<ICV<V>, "clt",  0xEB23, GR32>;
231      def CLGTAsm#V : FixedCmpBranchRSYb<ICV<V>, "clgt", 0xEB2B, GR64>;
232    }
233  }
234}
235
236//===----------------------------------------------------------------------===//
237// Call and return instructions
238//===----------------------------------------------------------------------===//
239
240// Define the general form of the call instructions for the asm parser.
241// These instructions don't hard-code %r14 as the return address register.
242let isCall = 1, Defs = [CC] in {
243  def BRAS  : CallRI <"bras", 0xA75>;
244  def BRASL : CallRIL<"brasl", 0xC05>;
245  def BAS   : CallRX <"bas", 0x4D>;
246  def BASR  : CallRR <"basr", 0x0D>;
247}
248
249// Regular calls.
250let isCall = 1, Defs = [R14D, CC] in {
251  def CallBRASL : Alias<6, (outs), (ins pcrel32:$I2, variable_ops),
252                        [(z_call pcrel32:$I2)]>;
253  def CallBASR  : Alias<2, (outs), (ins ADDR64:$R2, variable_ops),
254                        [(z_call ADDR64:$R2)]>;
255}
256
257// TLS calls.  These will be lowered into a call to __tls_get_offset,
258// with an extra relocation specifying the TLS symbol.
259let isCall = 1, Defs = [R14D, CC] in {
260  def TLS_GDCALL : Alias<6, (outs), (ins tlssym:$I2, variable_ops),
261                         [(z_tls_gdcall tglobaltlsaddr:$I2)]>;
262  def TLS_LDCALL : Alias<6, (outs), (ins tlssym:$I2, variable_ops),
263                         [(z_tls_ldcall tglobaltlsaddr:$I2)]>;
264}
265
266// Sibling calls.  Indirect sibling calls must be via R1, since R2 upwards
267// are argument registers and since branching to R0 is a no-op.
268let isCall = 1, isTerminator = 1, isReturn = 1, isBarrier = 1 in {
269  def CallJG : Alias<6, (outs), (ins pcrel32:$I2),
270                     [(z_sibcall pcrel32:$I2)]>;
271  let Uses = [R1D] in
272    def CallBR : Alias<2, (outs), (ins), [(z_sibcall R1D)]>;
273}
274
275// Conditional sibling calls.
276let CCMaskFirst = 1, isCall = 1, isTerminator = 1, isReturn = 1 in {
277  def CallBRCL : Alias<6, (outs), (ins cond4:$valid, cond4:$R1,
278                                   pcrel32:$I2), []>;
279  let Uses = [R1D] in
280    def CallBCR : Alias<2, (outs), (ins cond4:$valid, cond4:$R1), []>;
281}
282
283// Fused compare and conditional sibling calls.
284let isCall = 1, isTerminator = 1, isReturn = 1, Uses = [R1D] in {
285  def CRBCall : Alias<6, (outs), (ins GR32:$R1, GR32:$R2, cond4:$M3), []>;
286  def CGRBCall : Alias<6, (outs), (ins GR64:$R1, GR64:$R2, cond4:$M3), []>;
287  def CIBCall : Alias<6, (outs), (ins GR32:$R1, imm32sx8:$I2, cond4:$M3), []>;
288  def CGIBCall : Alias<6, (outs), (ins GR64:$R1, imm64sx8:$I2, cond4:$M3), []>;
289  def CLRBCall : Alias<6, (outs), (ins GR32:$R1, GR32:$R2, cond4:$M3), []>;
290  def CLGRBCall : Alias<6, (outs), (ins GR64:$R1, GR64:$R2, cond4:$M3), []>;
291  def CLIBCall : Alias<6, (outs), (ins GR32:$R1, imm32zx8:$I2, cond4:$M3), []>;
292  def CLGIBCall : Alias<6, (outs), (ins GR64:$R1, imm64zx8:$I2, cond4:$M3), []>;
293}
294
295// A return instruction (br %r14).
296let isReturn = 1, isTerminator = 1, isBarrier = 1, hasCtrlDep = 1 in
297  def Return : Alias<2, (outs), (ins), [(z_retflag)]>;
298
299// A conditional return instruction (bcr <cond>, %r14).
300let isReturn = 1, isTerminator = 1, hasCtrlDep = 1, CCMaskFirst = 1, Uses = [CC] in
301  def CondReturn : Alias<2, (outs), (ins cond4:$valid, cond4:$R1), []>;
302
303// Fused compare and conditional returns.
304let isReturn = 1, isTerminator = 1, hasCtrlDep = 1 in {
305  def CRBReturn : Alias<6, (outs), (ins GR32:$R1, GR32:$R2, cond4:$M3), []>;
306  def CGRBReturn : Alias<6, (outs), (ins GR64:$R1, GR64:$R2, cond4:$M3), []>;
307  def CIBReturn : Alias<6, (outs), (ins GR32:$R1, imm32sx8:$I2, cond4:$M3), []>;
308  def CGIBReturn : Alias<6, (outs), (ins GR64:$R1, imm64sx8:$I2, cond4:$M3), []>;
309  def CLRBReturn : Alias<6, (outs), (ins GR32:$R1, GR32:$R2, cond4:$M3), []>;
310  def CLGRBReturn : Alias<6, (outs), (ins GR64:$R1, GR64:$R2, cond4:$M3), []>;
311  def CLIBReturn : Alias<6, (outs), (ins GR32:$R1, imm32zx8:$I2, cond4:$M3), []>;
312  def CLGIBReturn : Alias<6, (outs), (ins GR64:$R1, imm64zx8:$I2, cond4:$M3), []>;
313}
314
315//===----------------------------------------------------------------------===//
316// Select instructions
317//===----------------------------------------------------------------------===//
318
319def Select32Mux : SelectWrapper<GRX32>, Requires<[FeatureHighWord]>;
320def Select32    : SelectWrapper<GR32>;
321def Select64    : SelectWrapper<GR64>;
322
323// We don't define 32-bit Mux stores if we don't have STOCFH, because the
324// low-only STOC should then always be used if possible.
325defm CondStore8Mux  : CondStores<GRX32, nonvolatile_truncstorei8,
326                                 nonvolatile_anyextloadi8, bdxaddr20only>,
327                      Requires<[FeatureHighWord]>;
328defm CondStore16Mux : CondStores<GRX32, nonvolatile_truncstorei16,
329                                 nonvolatile_anyextloadi16, bdxaddr20only>,
330                      Requires<[FeatureHighWord]>;
331defm CondStore32Mux : CondStores<GRX32, nonvolatile_store,
332                                 nonvolatile_load, bdxaddr20only>,
333                      Requires<[FeatureLoadStoreOnCond2]>;
334defm CondStore8     : CondStores<GR32, nonvolatile_truncstorei8,
335                                 nonvolatile_anyextloadi8, bdxaddr20only>;
336defm CondStore16    : CondStores<GR32, nonvolatile_truncstorei16,
337                                 nonvolatile_anyextloadi16, bdxaddr20only>;
338defm CondStore32    : CondStores<GR32, nonvolatile_store,
339                                 nonvolatile_load, bdxaddr20only>;
340
341defm : CondStores64<CondStore8, CondStore8Inv, nonvolatile_truncstorei8,
342                    nonvolatile_anyextloadi8, bdxaddr20only>;
343defm : CondStores64<CondStore16, CondStore16Inv, nonvolatile_truncstorei16,
344                    nonvolatile_anyextloadi16, bdxaddr20only>;
345defm : CondStores64<CondStore32, CondStore32Inv, nonvolatile_truncstorei32,
346                    nonvolatile_anyextloadi32, bdxaddr20only>;
347defm CondStore64 : CondStores<GR64, nonvolatile_store,
348                              nonvolatile_load, bdxaddr20only>;
349
350//===----------------------------------------------------------------------===//
351// Move instructions
352//===----------------------------------------------------------------------===//
353
354// Register moves.
355let hasSideEffects = 0 in {
356  // Expands to LR, RISBHG or RISBLG, depending on the choice of registers.
357  def LRMux : UnaryRRPseudo<"lr", null_frag, GRX32, GRX32>,
358              Requires<[FeatureHighWord]>;
359  def LR  : UnaryRR <"lr",  0x18,   null_frag, GR32, GR32>;
360  def LGR : UnaryRRE<"lgr", 0xB904, null_frag, GR64, GR64>;
361}
362let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
363  def LTR  : UnaryRR <"ltr",  0x12,   null_frag, GR32, GR32>;
364  def LTGR : UnaryRRE<"ltgr", 0xB902, null_frag, GR64, GR64>;
365}
366
367// Immediate moves.
368let hasSideEffects = 0, isAsCheapAsAMove = 1, isMoveImm = 1,
369    isReMaterializable = 1 in {
370  // 16-bit sign-extended immediates.  LHIMux expands to LHI or IIHF,
371  // deopending on the choice of register.
372  def LHIMux : UnaryRIPseudo<bitconvert, GRX32, imm32sx16>,
373               Requires<[FeatureHighWord]>;
374  def LHI  : UnaryRI<"lhi",  0xA78, bitconvert, GR32, imm32sx16>;
375  def LGHI : UnaryRI<"lghi", 0xA79, bitconvert, GR64, imm64sx16>;
376
377  // Other 16-bit immediates.
378  def LLILL : UnaryRI<"llill", 0xA5F, bitconvert, GR64, imm64ll16>;
379  def LLILH : UnaryRI<"llilh", 0xA5E, bitconvert, GR64, imm64lh16>;
380  def LLIHL : UnaryRI<"llihl", 0xA5D, bitconvert, GR64, imm64hl16>;
381  def LLIHH : UnaryRI<"llihh", 0xA5C, bitconvert, GR64, imm64hh16>;
382
383  // 32-bit immediates.
384  def LGFI  : UnaryRIL<"lgfi",  0xC01, bitconvert, GR64, imm64sx32>;
385  def LLILF : UnaryRIL<"llilf", 0xC0F, bitconvert, GR64, imm64lf32>;
386  def LLIHF : UnaryRIL<"llihf", 0xC0E, bitconvert, GR64, imm64hf32>;
387}
388
389// Register loads.
390let canFoldAsLoad = 1, SimpleBDXLoad = 1 in {
391  // Expands to L, LY or LFH, depending on the choice of register.
392  def LMux : UnaryRXYPseudo<"l", load, GRX32, 4>,
393             Requires<[FeatureHighWord]>;
394  defm L : UnaryRXPair<"l", 0x58, 0xE358, load, GR32, 4>;
395  def LFH : UnaryRXY<"lfh", 0xE3CA, load, GRH32, 4>,
396            Requires<[FeatureHighWord]>;
397  def LG : UnaryRXY<"lg", 0xE304, load, GR64, 8>;
398
399  // These instructions are split after register allocation, so we don't
400  // want a custom inserter.
401  let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
402    def L128 : Pseudo<(outs GR128:$dst), (ins bdxaddr20only128:$src),
403                      [(set GR128:$dst, (load bdxaddr20only128:$src))]>;
404  }
405}
406let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
407  def LT  : UnaryRXY<"lt",  0xE312, load, GR32, 4>;
408  def LTG : UnaryRXY<"ltg", 0xE302, load, GR64, 8>;
409}
410
411let canFoldAsLoad = 1 in {
412  def LRL  : UnaryRILPC<"lrl",  0xC4D, aligned_load, GR32>;
413  def LGRL : UnaryRILPC<"lgrl", 0xC48, aligned_load, GR64>;
414}
415
416// Load and zero rightmost byte.
417let Predicates = [FeatureLoadAndZeroRightmostByte] in {
418  def LZRF : UnaryRXY<"lzrf", 0xE33B, null_frag, GR32, 4>;
419  def LZRG : UnaryRXY<"lzrg", 0xE32A, null_frag, GR64, 8>;
420  def : Pat<(and (i32 (load bdxaddr20only:$src)), 0xffffff00),
421            (LZRF bdxaddr20only:$src)>;
422  def : Pat<(and (i64 (load bdxaddr20only:$src)), 0xffffffffffffff00),
423            (LZRG bdxaddr20only:$src)>;
424}
425
426// Load and trap.
427let Predicates = [FeatureLoadAndTrap] in {
428  def LAT   : UnaryRXY<"lat",   0xE39F, null_frag, GR32, 4>;
429  def LFHAT : UnaryRXY<"lfhat", 0xE3C8, null_frag, GRH32, 4>;
430  def LGAT  : UnaryRXY<"lgat",  0xE385, null_frag, GR64, 8>;
431}
432
433// Register stores.
434let SimpleBDXStore = 1 in {
435  // Expands to ST, STY or STFH, depending on the choice of register.
436  def STMux : StoreRXYPseudo<store, GRX32, 4>,
437              Requires<[FeatureHighWord]>;
438  defm ST : StoreRXPair<"st", 0x50, 0xE350, store, GR32, 4>;
439  def STFH : StoreRXY<"stfh", 0xE3CB, store, GRH32, 4>,
440             Requires<[FeatureHighWord]>;
441  def STG : StoreRXY<"stg", 0xE324, store, GR64, 8>;
442
443  // These instructions are split after register allocation, so we don't
444  // want a custom inserter.
445  let Has20BitOffset = 1, HasIndex = 1, Is128Bit = 1 in {
446    def ST128 : Pseudo<(outs), (ins GR128:$src, bdxaddr20only128:$dst),
447                       [(store GR128:$src, bdxaddr20only128:$dst)]>;
448  }
449}
450def STRL  : StoreRILPC<"strl", 0xC4F, aligned_store, GR32>;
451def STGRL : StoreRILPC<"stgrl", 0xC4B, aligned_store, GR64>;
452
453// 8-bit immediate stores to 8-bit fields.
454defm MVI : StoreSIPair<"mvi", 0x92, 0xEB52, truncstorei8, imm32zx8trunc>;
455
456// 16-bit immediate stores to 16-, 32- or 64-bit fields.
457def MVHHI : StoreSIL<"mvhhi", 0xE544, truncstorei16, imm32sx16trunc>;
458def MVHI  : StoreSIL<"mvhi",  0xE54C, store,         imm32sx16>;
459def MVGHI : StoreSIL<"mvghi", 0xE548, store,         imm64sx16>;
460
461// Memory-to-memory moves.
462let mayLoad = 1, mayStore = 1 in
463  defm MVC : MemorySS<"mvc", 0xD2, z_mvc, z_mvc_loop>;
464let mayLoad = 1, mayStore = 1, Defs = [CC] in {
465  def MVCL  : SideEffectBinaryMemMemRR<"mvcl", 0x0E, GR128, GR128>;
466  def MVCLE : SideEffectTernaryMemMemRS<"mvcle", 0xA8, GR128, GR128>;
467  def MVCLU : SideEffectTernaryMemMemRSY<"mvclu", 0xEB8E, GR128, GR128>;
468}
469
470// String moves.
471let mayLoad = 1, mayStore = 1, Defs = [CC] in
472  defm MVST : StringRRE<"mvst", 0xB255, z_stpcpy>;
473
474//===----------------------------------------------------------------------===//
475// Conditional move instructions
476//===----------------------------------------------------------------------===//
477
478let Predicates = [FeatureLoadStoreOnCond2], Uses = [CC] in {
479  // Load immediate on condition.  Matched via DAG pattern and created
480  // by the PeepholeOptimizer via FoldImmediate.
481  let hasSideEffects = 0 in {
482    // Expands to LOCHI or LOCHHI, depending on the choice of register.
483    def LOCHIMux : CondBinaryRIEPseudo<GRX32, imm32sx16>;
484    defm LOCHHI  : CondBinaryRIEPair<"lochhi", 0xEC4E, GRH32, imm32sx16>;
485    defm LOCHI   : CondBinaryRIEPair<"lochi",  0xEC42, GR32, imm32sx16>;
486    defm LOCGHI  : CondBinaryRIEPair<"locghi", 0xEC46, GR64, imm64sx16>;
487  }
488
489  // Move register on condition.  Expanded from Select* pseudos and
490  // created by early if-conversion.
491  let hasSideEffects = 0, isCommutable = 1 in {
492    // Expands to LOCR or LOCFHR or a branch-and-move sequence,
493    // depending on the choice of registers.
494    def LOCRMux : CondBinaryRRFPseudo<GRX32, GRX32>;
495    defm LOCFHR : CondBinaryRRFPair<"locfhr", 0xB9E0, GRH32, GRH32>;
496  }
497
498  // Load on condition.  Matched via DAG pattern.
499  // Expands to LOC or LOCFH, depending on the choice of register.
500  def LOCMux : CondUnaryRSYPseudo<nonvolatile_load, GRX32, 4>;
501  defm LOCFH : CondUnaryRSYPair<"locfh", 0xEBE0, nonvolatile_load, GRH32, 4>;
502
503  // Store on condition.  Expanded from CondStore* pseudos.
504  // Expands to STOC or STOCFH, depending on the choice of register.
505  def STOCMux : CondStoreRSYPseudo<GRX32, 4>;
506  defm STOCFH : CondStoreRSYPair<"stocfh", 0xEBE1, GRH32, 4>;
507
508  // Define AsmParser extended mnemonics for each general condition-code mask.
509  foreach V = [ "E", "NE", "H", "NH", "L", "NL", "HE", "NHE", "LE", "NLE",
510                "Z", "NZ", "P", "NP", "M", "NM", "LH", "NLH", "O", "NO" ] in {
511    def LOCHIAsm#V  : FixedCondBinaryRIE<CV<V>, "lochi",  0xEC42, GR32,
512                                         imm32sx16>;
513    def LOCGHIAsm#V : FixedCondBinaryRIE<CV<V>, "locghi", 0xEC46, GR64,
514                                         imm64sx16>;
515    def LOCHHIAsm#V : FixedCondBinaryRIE<CV<V>, "lochhi", 0xEC4E, GRH32,
516                                         imm32sx16>;
517    def LOCFHRAsm#V : FixedCondBinaryRRF<CV<V>, "locfhr", 0xB9E0, GRH32, GRH32>;
518    def LOCFHAsm#V  : FixedCondUnaryRSY<CV<V>, "locfh",  0xEBE0, GRH32, 4>;
519    def STOCFHAsm#V : FixedCondStoreRSY<CV<V>, "stocfh", 0xEBE1, GRH32, 4>;
520  }
521}
522
523let Predicates = [FeatureLoadStoreOnCond], Uses = [CC] in {
524  // Move register on condition.  Expanded from Select* pseudos and
525  // created by early if-conversion.
526  let hasSideEffects = 0, isCommutable = 1 in {
527    defm LOCR  : CondBinaryRRFPair<"locr",  0xB9F2, GR32, GR32>;
528    defm LOCGR : CondBinaryRRFPair<"locgr", 0xB9E2, GR64, GR64>;
529  }
530
531  // Load on condition.  Matched via DAG pattern.
532  defm LOC  : CondUnaryRSYPair<"loc",  0xEBF2, nonvolatile_load, GR32, 4>;
533  defm LOCG : CondUnaryRSYPair<"locg", 0xEBE2, nonvolatile_load, GR64, 8>;
534
535  // Store on condition.  Expanded from CondStore* pseudos.
536  defm STOC  : CondStoreRSYPair<"stoc",  0xEBF3, GR32, 4>;
537  defm STOCG : CondStoreRSYPair<"stocg", 0xEBE3, GR64, 8>;
538
539  // Define AsmParser extended mnemonics for each general condition-code mask.
540  foreach V = [ "E", "NE", "H", "NH", "L", "NL", "HE", "NHE", "LE", "NLE",
541                "Z", "NZ", "P", "NP", "M", "NM", "LH", "NLH", "O", "NO" ] in {
542    def LOCRAsm#V   : FixedCondBinaryRRF<CV<V>, "locr",  0xB9F2, GR32, GR32>;
543    def LOCGRAsm#V  : FixedCondBinaryRRF<CV<V>, "locgr", 0xB9E2, GR64, GR64>;
544    def LOCAsm#V    : FixedCondUnaryRSY<CV<V>, "loc",   0xEBF2, GR32, 4>;
545    def LOCGAsm#V   : FixedCondUnaryRSY<CV<V>, "locg",  0xEBE2, GR64, 8>;
546    def STOCAsm#V   : FixedCondStoreRSY<CV<V>, "stoc",  0xEBF3, GR32, 4>;
547    def STOCGAsm#V  : FixedCondStoreRSY<CV<V>, "stocg", 0xEBE3, GR64, 8>;
548  }
549}
550//===----------------------------------------------------------------------===//
551// Sign extensions
552//===----------------------------------------------------------------------===//
553//
554// Note that putting these before zero extensions mean that we will prefer
555// them for anyextload*.  There's not really much to choose between the two
556// either way, but signed-extending loads have a short LH and a long LHY,
557// while zero-extending loads have only the long LLH.
558//
559//===----------------------------------------------------------------------===//
560
561// 32-bit extensions from registers.
562let hasSideEffects = 0 in {
563  def LBR : UnaryRRE<"lbr", 0xB926, sext8,  GR32, GR32>;
564  def LHR : UnaryRRE<"lhr", 0xB927, sext16, GR32, GR32>;
565}
566
567// 64-bit extensions from registers.
568let hasSideEffects = 0 in {
569  def LGBR : UnaryRRE<"lgbr", 0xB906, sext8,  GR64, GR64>;
570  def LGHR : UnaryRRE<"lghr", 0xB907, sext16, GR64, GR64>;
571  def LGFR : UnaryRRE<"lgfr", 0xB914, sext32, GR64, GR32>;
572}
573let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in
574  def LTGFR : UnaryRRE<"ltgfr", 0xB912, null_frag, GR64, GR32>;
575
576// Match 32-to-64-bit sign extensions in which the source is already
577// in a 64-bit register.
578def : Pat<(sext_inreg GR64:$src, i32),
579          (LGFR (EXTRACT_SUBREG GR64:$src, subreg_l32))>;
580
581// 32-bit extensions from 8-bit memory.  LBMux expands to LB or LBH,
582// depending on the choice of register.
583def LBMux : UnaryRXYPseudo<"lb", asextloadi8, GRX32, 1>,
584            Requires<[FeatureHighWord]>;
585def LB  : UnaryRXY<"lb", 0xE376, asextloadi8, GR32, 1>;
586def LBH : UnaryRXY<"lbh", 0xE3C0, asextloadi8, GRH32, 1>,
587          Requires<[FeatureHighWord]>;
588
589// 32-bit extensions from 16-bit memory.  LHMux expands to LH or LHH,
590// depending on the choice of register.
591def LHMux : UnaryRXYPseudo<"lh", asextloadi16, GRX32, 2>,
592            Requires<[FeatureHighWord]>;
593defm LH   : UnaryRXPair<"lh", 0x48, 0xE378, asextloadi16, GR32, 2>;
594def  LHH  : UnaryRXY<"lhh", 0xE3C4, asextloadi16, GRH32, 2>,
595            Requires<[FeatureHighWord]>;
596def  LHRL : UnaryRILPC<"lhrl", 0xC45, aligned_asextloadi16, GR32>;
597
598// 64-bit extensions from memory.
599def LGB   : UnaryRXY<"lgb", 0xE377, asextloadi8,  GR64, 1>;
600def LGH   : UnaryRXY<"lgh", 0xE315, asextloadi16, GR64, 2>;
601def LGF   : UnaryRXY<"lgf", 0xE314, asextloadi32, GR64, 4>;
602def LGHRL : UnaryRILPC<"lghrl", 0xC44, aligned_asextloadi16, GR64>;
603def LGFRL : UnaryRILPC<"lgfrl", 0xC4C, aligned_asextloadi32, GR64>;
604let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in
605  def LTGF : UnaryRXY<"ltgf", 0xE332, asextloadi32, GR64, 4>;
606
607//===----------------------------------------------------------------------===//
608// Zero extensions
609//===----------------------------------------------------------------------===//
610
611// 32-bit extensions from registers.
612let hasSideEffects = 0 in {
613  // Expands to LLCR or RISB[LH]G, depending on the choice of registers.
614  def LLCRMux : UnaryRRPseudo<"llcr", zext8, GRX32, GRX32>,
615                Requires<[FeatureHighWord]>;
616  def LLCR    : UnaryRRE<"llcr", 0xB994, zext8,  GR32, GR32>;
617  // Expands to LLHR or RISB[LH]G, depending on the choice of registers.
618  def LLHRMux : UnaryRRPseudo<"llhr", zext16, GRX32, GRX32>,
619                Requires<[FeatureHighWord]>;
620  def LLHR    : UnaryRRE<"llhr", 0xB995, zext16, GR32, GR32>;
621}
622
623// 64-bit extensions from registers.
624let hasSideEffects = 0 in {
625  def LLGCR : UnaryRRE<"llgcr", 0xB984, zext8,  GR64, GR64>;
626  def LLGHR : UnaryRRE<"llghr", 0xB985, zext16, GR64, GR64>;
627  def LLGFR : UnaryRRE<"llgfr", 0xB916, zext32, GR64, GR32>;
628}
629
630// Match 32-to-64-bit zero extensions in which the source is already
631// in a 64-bit register.
632def : Pat<(and GR64:$src, 0xffffffff),
633          (LLGFR (EXTRACT_SUBREG GR64:$src, subreg_l32))>;
634
635// 32-bit extensions from 8-bit memory.  LLCMux expands to LLC or LLCH,
636// depending on the choice of register.
637def LLCMux : UnaryRXYPseudo<"llc", azextloadi8, GRX32, 1>,
638             Requires<[FeatureHighWord]>;
639def LLC  : UnaryRXY<"llc", 0xE394, azextloadi8, GR32, 1>;
640def LLCH : UnaryRXY<"llch", 0xE3C2, azextloadi8, GRH32, 1>,
641           Requires<[FeatureHighWord]>;
642
643// 32-bit extensions from 16-bit memory.  LLHMux expands to LLH or LLHH,
644// depending on the choice of register.
645def LLHMux : UnaryRXYPseudo<"llh", azextloadi16, GRX32, 2>,
646             Requires<[FeatureHighWord]>;
647def LLH   : UnaryRXY<"llh", 0xE395, azextloadi16, GR32, 2>;
648def LLHH  : UnaryRXY<"llhh", 0xE3C6, azextloadi16, GRH32, 2>,
649            Requires<[FeatureHighWord]>;
650def LLHRL : UnaryRILPC<"llhrl", 0xC42, aligned_azextloadi16, GR32>;
651
652// 64-bit extensions from memory.
653def LLGC   : UnaryRXY<"llgc", 0xE390, azextloadi8,  GR64, 1>;
654def LLGH   : UnaryRXY<"llgh", 0xE391, azextloadi16, GR64, 2>;
655def LLGF   : UnaryRXY<"llgf", 0xE316, azextloadi32, GR64, 4>;
656def LLGHRL : UnaryRILPC<"llghrl", 0xC46, aligned_azextloadi16, GR64>;
657def LLGFRL : UnaryRILPC<"llgfrl", 0xC4E, aligned_azextloadi32, GR64>;
658
659// 31-to-64-bit zero extensions.
660def LLGTR : UnaryRRE<"llgtr", 0xB917, null_frag, GR64, GR64>;
661def LLGT  : UnaryRXY<"llgt",  0xE317, null_frag, GR64, 4>;
662def : Pat<(and GR64:$src, 0x7fffffff),
663          (LLGTR GR64:$src)>;
664def : Pat<(and (i64 (azextloadi32 bdxaddr20only:$src)), 0x7fffffff),
665          (LLGT bdxaddr20only:$src)>;
666
667// Load and zero rightmost byte.
668let Predicates = [FeatureLoadAndZeroRightmostByte] in {
669  def LLZRGF : UnaryRXY<"llzrgf", 0xE33A, null_frag, GR64, 4>;
670  def : Pat<(and (i64 (azextloadi32 bdxaddr20only:$src)), 0xffffff00),
671            (LLZRGF bdxaddr20only:$src)>;
672}
673
674// Load and trap.
675let Predicates = [FeatureLoadAndTrap] in {
676  def LLGFAT : UnaryRXY<"llgfat", 0xE39D, null_frag, GR64, 4>;
677  def LLGTAT : UnaryRXY<"llgtat", 0xE39C, null_frag, GR64, 4>;
678}
679
680// Extend GR64s to GR128s.
681let usesCustomInserter = 1 in
682  def ZEXT128 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
683
684//===----------------------------------------------------------------------===//
685// "Any" extensions
686//===----------------------------------------------------------------------===//
687
688// Use subregs to populate the "don't care" bits in a 32-bit to 64-bit anyext.
689def : Pat<(i64 (anyext GR32:$src)),
690          (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_l32)>;
691
692// Extend GR64s to GR128s.
693let usesCustomInserter = 1 in
694  def AEXT128 : Pseudo<(outs GR128:$dst), (ins GR64:$src), []>;
695
696//===----------------------------------------------------------------------===//
697// Truncations
698//===----------------------------------------------------------------------===//
699
700// Truncations of 64-bit registers to 32-bit registers.
701def : Pat<(i32 (trunc GR64:$src)),
702          (EXTRACT_SUBREG GR64:$src, subreg_l32)>;
703
704// Truncations of 32-bit registers to 8-bit memory.  STCMux expands to
705// STC, STCY or STCH, depending on the choice of register.
706def STCMux : StoreRXYPseudo<truncstorei8, GRX32, 1>,
707             Requires<[FeatureHighWord]>;
708defm STC : StoreRXPair<"stc", 0x42, 0xE372, truncstorei8, GR32, 1>;
709def STCH : StoreRXY<"stch", 0xE3C3, truncstorei8, GRH32, 1>,
710           Requires<[FeatureHighWord]>;
711
712// Truncations of 32-bit registers to 16-bit memory.  STHMux expands to
713// STH, STHY or STHH, depending on the choice of register.
714def STHMux : StoreRXYPseudo<truncstorei16, GRX32, 1>,
715             Requires<[FeatureHighWord]>;
716defm STH : StoreRXPair<"sth", 0x40, 0xE370, truncstorei16, GR32, 2>;
717def STHH : StoreRXY<"sthh", 0xE3C7, truncstorei16, GRH32, 2>,
718           Requires<[FeatureHighWord]>;
719def STHRL : StoreRILPC<"sthrl", 0xC47, aligned_truncstorei16, GR32>;
720
721// Truncations of 64-bit registers to memory.
722defm : StoreGR64Pair<STC, STCY, truncstorei8>;
723defm : StoreGR64Pair<STH, STHY, truncstorei16>;
724def  : StoreGR64PC<STHRL, aligned_truncstorei16>;
725defm : StoreGR64Pair<ST, STY, truncstorei32>;
726def  : StoreGR64PC<STRL, aligned_truncstorei32>;
727
728// Store characters under mask -- not (yet) used for codegen.
729defm STCM : StoreBinaryRSPair<"stcm", 0xBE, 0xEB2D, GR32, 0>;
730def STCMH : StoreBinaryRSY<"stcmh", 0xEB2C, GRH32, 0>;
731
732//===----------------------------------------------------------------------===//
733// Multi-register moves
734//===----------------------------------------------------------------------===//
735
736// Multi-register loads.
737defm LM : LoadMultipleRSPair<"lm", 0x98, 0xEB98, GR32>;
738def LMG : LoadMultipleRSY<"lmg", 0xEB04, GR64>;
739def LMH : LoadMultipleRSY<"lmh", 0xEB96, GRH32>;
740def LMD : LoadMultipleSSe<"lmd", 0xEF, GR64>;
741
742// Multi-register stores.
743defm STM : StoreMultipleRSPair<"stm", 0x90, 0xEB90, GR32>;
744def STMG : StoreMultipleRSY<"stmg", 0xEB24, GR64>;
745def STMH : StoreMultipleRSY<"stmh", 0xEB26, GRH32>;
746
747//===----------------------------------------------------------------------===//
748// Byte swaps
749//===----------------------------------------------------------------------===//
750
751// Byte-swapping register moves.
752let hasSideEffects = 0 in {
753  def LRVR  : UnaryRRE<"lrvr",  0xB91F, bswap, GR32, GR32>;
754  def LRVGR : UnaryRRE<"lrvgr", 0xB90F, bswap, GR64, GR64>;
755}
756
757// Byte-swapping loads.  Unlike normal loads, these instructions are
758// allowed to access storage more than once.
759def LRVH : UnaryRXY<"lrvh", 0xE31F, z_lrvh, GR32, 2>;
760def LRV  : UnaryRXY<"lrv",  0xE31E, z_lrv,  GR32, 4>;
761def LRVG : UnaryRXY<"lrvg", 0xE30F, z_lrvg, GR64, 8>;
762
763// Likewise byte-swapping stores.
764def STRVH : StoreRXY<"strvh", 0xE33F, z_strvh, GR32, 2>;
765def STRV  : StoreRXY<"strv",  0xE33E, z_strv,  GR32, 4>;
766def STRVG : StoreRXY<"strvg", 0xE32F, z_strvg, GR64, 8>;
767
768// Byte-swapping memory-to-memory moves.
769let mayLoad = 1, mayStore = 1 in
770  def MVCIN : SideEffectBinarySSa<"mvcin", 0xE8>;
771
772//===----------------------------------------------------------------------===//
773// Load address instructions
774//===----------------------------------------------------------------------===//
775
776// Load BDX-style addresses.
777let hasSideEffects = 0, isAsCheapAsAMove = 1, isReMaterializable = 1 in
778  defm LA : LoadAddressRXPair<"la", 0x41, 0xE371, bitconvert>;
779
780// Load a PC-relative address.  There's no version of this instruction
781// with a 16-bit offset, so there's no relaxation.
782let hasSideEffects = 0, isAsCheapAsAMove = 1, isMoveImm = 1,
783    isReMaterializable = 1 in
784  def LARL : LoadAddressRIL<"larl", 0xC00, bitconvert>;
785
786// Load the Global Offset Table address.  This will be lowered into a
787//     larl $R1, _GLOBAL_OFFSET_TABLE_
788// instruction.
789def GOT : Alias<6, (outs GR64:$R1), (ins),
790                [(set GR64:$R1, (global_offset_table))]>;
791
792//===----------------------------------------------------------------------===//
793// Absolute and Negation
794//===----------------------------------------------------------------------===//
795
796let Defs = [CC] in {
797  let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
798    def LPR  : UnaryRR <"lpr",  0x10,   z_iabs, GR32, GR32>;
799    def LPGR : UnaryRRE<"lpgr", 0xB900, z_iabs, GR64, GR64>;
800  }
801  let CCValues = 0xE, CompareZeroCCMask = 0xE in
802    def LPGFR : UnaryRRE<"lpgfr", 0xB910, null_frag, GR64, GR32>;
803}
804def : Pat<(z_iabs32 GR32:$src), (LPR  GR32:$src)>;
805def : Pat<(z_iabs64 GR64:$src), (LPGR GR64:$src)>;
806defm : SXU<z_iabs,   LPGFR>;
807defm : SXU<z_iabs64, LPGFR>;
808
809let Defs = [CC] in {
810  let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
811    def LNR  : UnaryRR <"lnr",  0x11,   z_inegabs, GR32, GR32>;
812    def LNGR : UnaryRRE<"lngr", 0xB901, z_inegabs, GR64, GR64>;
813  }
814  let CCValues = 0xE, CompareZeroCCMask = 0xE in
815    def LNGFR : UnaryRRE<"lngfr", 0xB911, null_frag, GR64, GR32>;
816}
817def : Pat<(z_inegabs32 GR32:$src), (LNR  GR32:$src)>;
818def : Pat<(z_inegabs64 GR64:$src), (LNGR GR64:$src)>;
819defm : SXU<z_inegabs,   LNGFR>;
820defm : SXU<z_inegabs64, LNGFR>;
821
822let Defs = [CC] in {
823  let CCValues = 0xF, CompareZeroCCMask = 0x8 in {
824    def LCR  : UnaryRR <"lcr",  0x13,   ineg, GR32, GR32>;
825    def LCGR : UnaryRRE<"lcgr", 0xB903, ineg, GR64, GR64>;
826  }
827  let CCValues = 0xE, CompareZeroCCMask = 0xE in
828    def LCGFR : UnaryRRE<"lcgfr", 0xB913, null_frag, GR64, GR32>;
829}
830defm : SXU<ineg, LCGFR>;
831
832//===----------------------------------------------------------------------===//
833// Insertion
834//===----------------------------------------------------------------------===//
835
836let isCodeGenOnly = 1 in
837  defm IC32 : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR32, azextloadi8, 1>;
838defm IC : BinaryRXPair<"ic", 0x43, 0xE373, inserti8, GR64, azextloadi8, 1>;
839
840defm : InsertMem<"inserti8", IC32,  GR32, azextloadi8, bdxaddr12pair>;
841defm : InsertMem<"inserti8", IC32Y, GR32, azextloadi8, bdxaddr20pair>;
842
843defm : InsertMem<"inserti8", IC,  GR64, azextloadi8, bdxaddr12pair>;
844defm : InsertMem<"inserti8", ICY, GR64, azextloadi8, bdxaddr20pair>;
845
846// Insert characters under mask -- not (yet) used for codegen.
847let Defs = [CC] in {
848  defm ICM : TernaryRSPair<"icm", 0xBF, 0xEB81, GR32, 0>;
849  def ICMH : TernaryRSY<"icmh", 0xEB80, GRH32, 0>;
850}
851
852// Insertions of a 16-bit immediate, leaving other bits unaffected.
853// We don't have or_as_insert equivalents of these operations because
854// OI is available instead.
855//
856// IIxMux expands to II[LH]x, depending on the choice of register.
857def IILMux : BinaryRIPseudo<insertll, GRX32, imm32ll16>,
858             Requires<[FeatureHighWord]>;
859def IIHMux : BinaryRIPseudo<insertlh, GRX32, imm32lh16>,
860             Requires<[FeatureHighWord]>;
861def IILL : BinaryRI<"iill", 0xA53, insertll, GR32, imm32ll16>;
862def IILH : BinaryRI<"iilh", 0xA52, insertlh, GR32, imm32lh16>;
863def IIHL : BinaryRI<"iihl", 0xA51, insertll, GRH32, imm32ll16>;
864def IIHH : BinaryRI<"iihh", 0xA50, insertlh, GRH32, imm32lh16>;
865def IILL64 : BinaryAliasRI<insertll, GR64, imm64ll16>;
866def IILH64 : BinaryAliasRI<insertlh, GR64, imm64lh16>;
867def IIHL64 : BinaryAliasRI<inserthl, GR64, imm64hl16>;
868def IIHH64 : BinaryAliasRI<inserthh, GR64, imm64hh16>;
869
870// ...likewise for 32-bit immediates.  For GR32s this is a general
871// full-width move.  (We use IILF rather than something like LLILF
872// for 32-bit moves because IILF leaves the upper 32 bits of the
873// GR64 unchanged.)
874let isAsCheapAsAMove = 1, isMoveImm = 1, isReMaterializable = 1 in {
875  def IIFMux : UnaryRIPseudo<bitconvert, GRX32, uimm32>,
876               Requires<[FeatureHighWord]>;
877  def IILF : UnaryRIL<"iilf", 0xC09, bitconvert, GR32, uimm32>;
878  def IIHF : UnaryRIL<"iihf", 0xC08, bitconvert, GRH32, uimm32>;
879}
880def IILF64 : BinaryAliasRIL<insertlf, GR64, imm64lf32>;
881def IIHF64 : BinaryAliasRIL<inserthf, GR64, imm64hf32>;
882
883// An alternative model of inserthf, with the first operand being
884// a zero-extended value.
885def : Pat<(or (zext32 GR32:$src), imm64hf32:$imm),
886          (IIHF64 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_l32),
887                  imm64hf32:$imm)>;
888
889//===----------------------------------------------------------------------===//
890// Addition
891//===----------------------------------------------------------------------===//
892
893// Plain addition.
894let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in {
895  // Addition of a register.
896  let isCommutable = 1 in {
897    defm AR : BinaryRRAndK<"ar", 0x1A, 0xB9F8, add, GR32, GR32>;
898    defm AGR : BinaryRREAndK<"agr", 0xB908, 0xB9E8, add, GR64, GR64>;
899  }
900  def AGFR : BinaryRRE<"agfr", 0xB918, null_frag, GR64, GR32>;
901
902  // Addition to a high register.
903  def AHHHR : BinaryRRFa<"ahhhr", 0xB9C8, null_frag, GRH32, GRH32, GRH32>,
904              Requires<[FeatureHighWord]>;
905  def AHHLR : BinaryRRFa<"ahhlr", 0xB9D8, null_frag, GRH32, GRH32, GR32>,
906              Requires<[FeatureHighWord]>;
907
908  // Addition of signed 16-bit immediates.
909  defm AHIMux : BinaryRIAndKPseudo<"ahimux", add, GRX32, imm32sx16>;
910  defm AHI  : BinaryRIAndK<"ahi",  0xA7A, 0xECD8, add, GR32, imm32sx16>;
911  defm AGHI : BinaryRIAndK<"aghi", 0xA7B, 0xECD9, add, GR64, imm64sx16>;
912
913  // Addition of signed 32-bit immediates.
914  def AFIMux : BinaryRIPseudo<add, GRX32, simm32>,
915               Requires<[FeatureHighWord]>;
916  def AFI  : BinaryRIL<"afi",  0xC29, add, GR32, simm32>;
917  def AIH  : BinaryRIL<"aih",  0xCC8, add, GRH32, simm32>,
918             Requires<[FeatureHighWord]>;
919  def AGFI : BinaryRIL<"agfi", 0xC28, add, GR64, imm64sx32>;
920
921  // Addition of memory.
922  defm AH  : BinaryRXPair<"ah", 0x4A, 0xE37A, add, GR32, asextloadi16, 2>;
923  defm A   : BinaryRXPair<"a",  0x5A, 0xE35A, add, GR32, load, 4>;
924  def  AGF : BinaryRXY<"agf", 0xE318, add, GR64, asextloadi32, 4>;
925  def  AG  : BinaryRXY<"ag",  0xE308, add, GR64, load, 8>;
926
927  // Addition to memory.
928  def ASI  : BinarySIY<"asi",  0xEB6A, add, imm32sx8>;
929  def AGSI : BinarySIY<"agsi", 0xEB7A, add, imm64sx8>;
930}
931defm : SXB<add, GR64, AGFR>;
932
933// Addition producing a carry.
934let Defs = [CC] in {
935  // Addition of a register.
936  let isCommutable = 1 in {
937    defm ALR : BinaryRRAndK<"alr", 0x1E, 0xB9FA, addc, GR32, GR32>;
938    defm ALGR : BinaryRREAndK<"algr", 0xB90A, 0xB9EA, addc, GR64, GR64>;
939  }
940  def ALGFR : BinaryRRE<"algfr", 0xB91A, null_frag, GR64, GR32>;
941
942  // Addition to a high register.
943  def ALHHHR : BinaryRRFa<"alhhhr", 0xB9CA, null_frag, GRH32, GRH32, GRH32>,
944               Requires<[FeatureHighWord]>;
945  def ALHHLR : BinaryRRFa<"alhhlr", 0xB9DA, null_frag, GRH32, GRH32, GR32>,
946               Requires<[FeatureHighWord]>;
947
948  // Addition of signed 16-bit immediates.
949  def ALHSIK  : BinaryRIE<"alhsik",  0xECDA, addc, GR32, imm32sx16>,
950                Requires<[FeatureDistinctOps]>;
951  def ALGHSIK : BinaryRIE<"alghsik", 0xECDB, addc, GR64, imm64sx16>,
952                Requires<[FeatureDistinctOps]>;
953
954  // Addition of unsigned 32-bit immediates.
955  def ALFI  : BinaryRIL<"alfi",  0xC2B, addc, GR32, uimm32>;
956  def ALGFI : BinaryRIL<"algfi", 0xC2A, addc, GR64, imm64zx32>;
957
958  // Addition of signed 32-bit immediates.
959  def ALSIH : BinaryRIL<"alsih", 0xCCA, null_frag, GRH32, simm32>,
960              Requires<[FeatureHighWord]>;
961
962  // Addition of memory.
963  defm AL   : BinaryRXPair<"al", 0x5E, 0xE35E, addc, GR32, load, 4>;
964  def  ALGF : BinaryRXY<"algf", 0xE31A, addc, GR64, azextloadi32, 4>;
965  def  ALG  : BinaryRXY<"alg",  0xE30A, addc, GR64, load, 8>;
966
967  // Addition to memory.
968  def ALSI  : BinarySIY<"alsi",  0xEB6E, null_frag, imm32sx8>;
969  def ALGSI : BinarySIY<"algsi", 0xEB7E, null_frag, imm64sx8>;
970}
971defm : ZXB<addc, GR64, ALGFR>;
972
973// Addition producing and using a carry.
974let Defs = [CC], Uses = [CC] in {
975  // Addition of a register.
976  def ALCR  : BinaryRRE<"alcr",  0xB998, adde, GR32, GR32>;
977  def ALCGR : BinaryRRE<"alcgr", 0xB988, adde, GR64, GR64>;
978
979  // Addition of memory.
980  def ALC  : BinaryRXY<"alc",  0xE398, adde, GR32, load, 4>;
981  def ALCG : BinaryRXY<"alcg", 0xE388, adde, GR64, load, 8>;
982}
983
984// Addition that does not modify the condition code.
985def ALSIHN : BinaryRIL<"alsihn", 0xCCB, null_frag, GRH32, simm32>,
986             Requires<[FeatureHighWord]>;
987
988//===----------------------------------------------------------------------===//
989// Subtraction
990//===----------------------------------------------------------------------===//
991
992// Plain subtraction.  Although immediate forms exist, we use the
993// add-immediate instruction instead.
994let Defs = [CC], CCValues = 0xF, CompareZeroCCMask = 0x8 in {
995  // Subtraction of a register.
996  defm SR : BinaryRRAndK<"sr", 0x1B, 0xB9F9, sub, GR32, GR32>;
997  def SGFR : BinaryRRE<"sgfr", 0xB919, null_frag, GR64, GR32>;
998  defm SGR : BinaryRREAndK<"sgr", 0xB909, 0xB9E9, sub, GR64, GR64>;
999
1000  // Subtraction from a high register.
1001  def SHHHR : BinaryRRFa<"shhhr", 0xB9C9, null_frag, GRH32, GRH32, GRH32>,
1002              Requires<[FeatureHighWord]>;
1003  def SHHLR : BinaryRRFa<"shhlr", 0xB9D9, null_frag, GRH32, GRH32, GR32>,
1004              Requires<[FeatureHighWord]>;
1005
1006  // Subtraction of memory.
1007  defm SH  : BinaryRXPair<"sh", 0x4B, 0xE37B, sub, GR32, asextloadi16, 2>;
1008  defm S   : BinaryRXPair<"s", 0x5B, 0xE35B, sub, GR32, load, 4>;
1009  def  SGF : BinaryRXY<"sgf", 0xE319, sub, GR64, asextloadi32, 4>;
1010  def  SG  : BinaryRXY<"sg",  0xE309, sub, GR64, load, 8>;
1011}
1012defm : SXB<sub, GR64, SGFR>;
1013
1014// Subtraction producing a carry.
1015let Defs = [CC] in {
1016  // Subtraction of a register.
1017  defm SLR : BinaryRRAndK<"slr", 0x1F, 0xB9FB, subc, GR32, GR32>;
1018  def SLGFR : BinaryRRE<"slgfr", 0xB91B, null_frag, GR64, GR32>;
1019  defm SLGR : BinaryRREAndK<"slgr", 0xB90B, 0xB9EB, subc, GR64, GR64>;
1020
1021  // Subtraction from a high register.
1022  def SLHHHR : BinaryRRFa<"slhhhr", 0xB9CB, null_frag, GRH32, GRH32, GRH32>,
1023               Requires<[FeatureHighWord]>;
1024  def SLHHLR : BinaryRRFa<"slhhlr", 0xB9DB, null_frag, GRH32, GRH32, GR32>,
1025               Requires<[FeatureHighWord]>;
1026
1027  // Subtraction of unsigned 32-bit immediates.  These don't match
1028  // subc because we prefer addc for constants.
1029  def SLFI  : BinaryRIL<"slfi",  0xC25, null_frag, GR32, uimm32>;
1030  def SLGFI : BinaryRIL<"slgfi", 0xC24, null_frag, GR64, imm64zx32>;
1031
1032  // Subtraction of memory.
1033  defm SL   : BinaryRXPair<"sl", 0x5F, 0xE35F, subc, GR32, load, 4>;
1034  def  SLGF : BinaryRXY<"slgf", 0xE31B, subc, GR64, azextloadi32, 4>;
1035  def  SLG  : BinaryRXY<"slg",  0xE30B, subc, GR64, load, 8>;
1036}
1037defm : ZXB<subc, GR64, SLGFR>;
1038
1039// Subtraction producing and using a carry.
1040let Defs = [CC], Uses = [CC] in {
1041  // Subtraction of a register.
1042  def SLBR  : BinaryRRE<"slbr",  0xB999, sube, GR32, GR32>;
1043  def SLBGR : BinaryRRE<"slbgr", 0xB989, sube, GR64, GR64>;
1044
1045  // Subtraction of memory.
1046  def SLB  : BinaryRXY<"slb",  0xE399, sube, GR32, load, 4>;
1047  def SLBG : BinaryRXY<"slbg", 0xE389, sube, GR64, load, 8>;
1048}
1049
1050//===----------------------------------------------------------------------===//
1051// AND
1052//===----------------------------------------------------------------------===//
1053
1054let Defs = [CC] in {
1055  // ANDs of a register.
1056  let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1057    defm NR : BinaryRRAndK<"nr", 0x14, 0xB9F4, and, GR32, GR32>;
1058    defm NGR : BinaryRREAndK<"ngr", 0xB980, 0xB9E4, and, GR64, GR64>;
1059  }
1060
1061  let isConvertibleToThreeAddress = 1 in {
1062    // ANDs of a 16-bit immediate, leaving other bits unaffected.
1063    // The CC result only reflects the 16-bit field, not the full register.
1064    //
1065    // NIxMux expands to NI[LH]x, depending on the choice of register.
1066    def NILMux : BinaryRIPseudo<and, GRX32, imm32ll16c>,
1067                 Requires<[FeatureHighWord]>;
1068    def NIHMux : BinaryRIPseudo<and, GRX32, imm32lh16c>,
1069                 Requires<[FeatureHighWord]>;
1070    def NILL : BinaryRI<"nill", 0xA57, and, GR32, imm32ll16c>;
1071    def NILH : BinaryRI<"nilh", 0xA56, and, GR32, imm32lh16c>;
1072    def NIHL : BinaryRI<"nihl", 0xA55, and, GRH32, imm32ll16c>;
1073    def NIHH : BinaryRI<"nihh", 0xA54, and, GRH32, imm32lh16c>;
1074    def NILL64 : BinaryAliasRI<and, GR64, imm64ll16c>;
1075    def NILH64 : BinaryAliasRI<and, GR64, imm64lh16c>;
1076    def NIHL64 : BinaryAliasRI<and, GR64, imm64hl16c>;
1077    def NIHH64 : BinaryAliasRI<and, GR64, imm64hh16c>;
1078
1079    // ANDs of a 32-bit immediate, leaving other bits unaffected.
1080    // The CC result only reflects the 32-bit field, which means we can
1081    // use it as a zero indicator for i32 operations but not otherwise.
1082    let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1083      // Expands to NILF or NIHF, depending on the choice of register.
1084      def NIFMux : BinaryRIPseudo<and, GRX32, uimm32>,
1085                   Requires<[FeatureHighWord]>;
1086      def NILF : BinaryRIL<"nilf", 0xC0B, and, GR32, uimm32>;
1087      def NIHF : BinaryRIL<"nihf", 0xC0A, and, GRH32, uimm32>;
1088    }
1089    def NILF64 : BinaryAliasRIL<and, GR64, imm64lf32c>;
1090    def NIHF64 : BinaryAliasRIL<and, GR64, imm64hf32c>;
1091  }
1092
1093  // ANDs of memory.
1094  let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1095    defm N  : BinaryRXPair<"n", 0x54, 0xE354, and, GR32, load, 4>;
1096    def  NG : BinaryRXY<"ng", 0xE380, and, GR64, load, 8>;
1097  }
1098
1099  // AND to memory
1100  defm NI : BinarySIPair<"ni", 0x94, 0xEB54, null_frag, imm32zx8>;
1101
1102  // Block AND.
1103  let mayLoad = 1, mayStore = 1 in
1104    defm NC : MemorySS<"nc", 0xD4, z_nc, z_nc_loop>;
1105}
1106defm : RMWIByte<and, bdaddr12pair, NI>;
1107defm : RMWIByte<and, bdaddr20pair, NIY>;
1108
1109//===----------------------------------------------------------------------===//
1110// OR
1111//===----------------------------------------------------------------------===//
1112
1113let Defs = [CC] in {
1114  // ORs of a register.
1115  let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1116    defm OR : BinaryRRAndK<"or", 0x16, 0xB9F6, or, GR32, GR32>;
1117    defm OGR : BinaryRREAndK<"ogr", 0xB981, 0xB9E6, or, GR64, GR64>;
1118  }
1119
1120  // ORs of a 16-bit immediate, leaving other bits unaffected.
1121  // The CC result only reflects the 16-bit field, not the full register.
1122  //
1123  // OIxMux expands to OI[LH]x, depending on the choice of register.
1124  def OILMux : BinaryRIPseudo<or, GRX32, imm32ll16>,
1125               Requires<[FeatureHighWord]>;
1126  def OIHMux : BinaryRIPseudo<or, GRX32, imm32lh16>,
1127               Requires<[FeatureHighWord]>;
1128  def OILL : BinaryRI<"oill", 0xA5B, or, GR32, imm32ll16>;
1129  def OILH : BinaryRI<"oilh", 0xA5A, or, GR32, imm32lh16>;
1130  def OIHL : BinaryRI<"oihl", 0xA59, or, GRH32, imm32ll16>;
1131  def OIHH : BinaryRI<"oihh", 0xA58, or, GRH32, imm32lh16>;
1132  def OILL64 : BinaryAliasRI<or, GR64, imm64ll16>;
1133  def OILH64 : BinaryAliasRI<or, GR64, imm64lh16>;
1134  def OIHL64 : BinaryAliasRI<or, GR64, imm64hl16>;
1135  def OIHH64 : BinaryAliasRI<or, GR64, imm64hh16>;
1136
1137  // ORs of a 32-bit immediate, leaving other bits unaffected.
1138  // The CC result only reflects the 32-bit field, which means we can
1139  // use it as a zero indicator for i32 operations but not otherwise.
1140  let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1141    // Expands to OILF or OIHF, depending on the choice of register.
1142    def OIFMux : BinaryRIPseudo<or, GRX32, uimm32>,
1143                 Requires<[FeatureHighWord]>;
1144    def OILF : BinaryRIL<"oilf", 0xC0D, or, GR32, uimm32>;
1145    def OIHF : BinaryRIL<"oihf", 0xC0C, or, GRH32, uimm32>;
1146  }
1147  def OILF64 : BinaryAliasRIL<or, GR64, imm64lf32>;
1148  def OIHF64 : BinaryAliasRIL<or, GR64, imm64hf32>;
1149
1150  // ORs of memory.
1151  let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1152    defm O  : BinaryRXPair<"o", 0x56, 0xE356, or, GR32, load, 4>;
1153    def  OG : BinaryRXY<"og", 0xE381, or, GR64, load, 8>;
1154  }
1155
1156  // OR to memory
1157  defm OI : BinarySIPair<"oi", 0x96, 0xEB56, null_frag, imm32zx8>;
1158
1159  // Block OR.
1160  let mayLoad = 1, mayStore = 1 in
1161    defm OC : MemorySS<"oc", 0xD6, z_oc, z_oc_loop>;
1162}
1163defm : RMWIByte<or, bdaddr12pair, OI>;
1164defm : RMWIByte<or, bdaddr20pair, OIY>;
1165
1166//===----------------------------------------------------------------------===//
1167// XOR
1168//===----------------------------------------------------------------------===//
1169
1170let Defs = [CC] in {
1171  // XORs of a register.
1172  let isCommutable = 1, CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1173    defm XR : BinaryRRAndK<"xr", 0x17, 0xB9F7, xor, GR32, GR32>;
1174    defm XGR : BinaryRREAndK<"xgr", 0xB982, 0xB9E7, xor, GR64, GR64>;
1175  }
1176
1177  // XORs of a 32-bit immediate, leaving other bits unaffected.
1178  // The CC result only reflects the 32-bit field, which means we can
1179  // use it as a zero indicator for i32 operations but not otherwise.
1180  let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1181    // Expands to XILF or XIHF, depending on the choice of register.
1182    def XIFMux : BinaryRIPseudo<xor, GRX32, uimm32>,
1183                 Requires<[FeatureHighWord]>;
1184    def XILF : BinaryRIL<"xilf", 0xC07, xor, GR32, uimm32>;
1185    def XIHF : BinaryRIL<"xihf", 0xC06, xor, GRH32, uimm32>;
1186  }
1187  def XILF64 : BinaryAliasRIL<xor, GR64, imm64lf32>;
1188  def XIHF64 : BinaryAliasRIL<xor, GR64, imm64hf32>;
1189
1190  // XORs of memory.
1191  let CCValues = 0xC, CompareZeroCCMask = 0x8 in {
1192    defm X  : BinaryRXPair<"x",0x57, 0xE357, xor, GR32, load, 4>;
1193    def  XG : BinaryRXY<"xg", 0xE382, xor, GR64, load, 8>;
1194  }
1195
1196  // XOR to memory
1197  defm XI : BinarySIPair<"xi", 0x97, 0xEB57, null_frag, imm32zx8>;
1198
1199  // Block XOR.
1200  let mayLoad = 1, mayStore = 1 in
1201    defm XC : MemorySS<"xc", 0xD7, z_xc, z_xc_loop>;
1202}
1203defm : RMWIByte<xor, bdaddr12pair, XI>;
1204defm : RMWIByte<xor, bdaddr20pair, XIY>;
1205
1206//===----------------------------------------------------------------------===//
1207// Multiplication
1208//===----------------------------------------------------------------------===//
1209
1210// Multiplication of a register.
1211let isCommutable = 1 in {
1212  def MSR  : BinaryRRE<"msr",  0xB252, mul, GR32, GR32>;
1213  def MSGR : BinaryRRE<"msgr", 0xB90C, mul, GR64, GR64>;
1214}
1215def MSGFR : BinaryRRE<"msgfr", 0xB91C, null_frag, GR64, GR32>;
1216defm : SXB<mul, GR64, MSGFR>;
1217
1218// Multiplication of a signed 16-bit immediate.
1219def MHI  : BinaryRI<"mhi",  0xA7C, mul, GR32, imm32sx16>;
1220def MGHI : BinaryRI<"mghi", 0xA7D, mul, GR64, imm64sx16>;
1221
1222// Multiplication of a signed 32-bit immediate.
1223def MSFI  : BinaryRIL<"msfi",  0xC21, mul, GR32, simm32>;
1224def MSGFI : BinaryRIL<"msgfi", 0xC20, mul, GR64, imm64sx32>;
1225
1226// Multiplication of memory.
1227defm MH   : BinaryRXPair<"mh", 0x4C, 0xE37C, mul, GR32, asextloadi16, 2>;
1228defm MS   : BinaryRXPair<"ms", 0x71, 0xE351, mul, GR32, load, 4>;
1229def  MSGF : BinaryRXY<"msgf", 0xE31C, mul, GR64, asextloadi32, 4>;
1230def  MSG  : BinaryRXY<"msg",  0xE30C, mul, GR64, load, 8>;
1231
1232// Multiplication of a register, producing two results.
1233def MR   : BinaryRR <"mr",   0x1C,   null_frag, GR128, GR32>;
1234def MLR  : BinaryRRE<"mlr",  0xB996, null_frag, GR128, GR32>;
1235def MLGR : BinaryRRE<"mlgr", 0xB986, null_frag, GR128, GR64>;
1236def : Pat<(z_umul_lohi GR64:$src1, GR64:$src2),
1237          (MLGR (AEXT128 GR64:$src1), GR64:$src2)>;
1238
1239// Multiplication of memory, producing two results.
1240def M   : BinaryRX <"m",   0x5C,   null_frag, GR128, load, 4>;
1241def MFY : BinaryRXY<"mfy", 0xE35C, null_frag, GR128, load, 4>;
1242def ML  : BinaryRXY<"ml",  0xE396, null_frag, GR128, load, 4>;
1243def MLG : BinaryRXY<"mlg", 0xE386, null_frag, GR128, load, 8>;
1244def : Pat<(z_umul_lohi GR64:$src1, (i64 (load bdxaddr20only:$src2))),
1245          (MLG (AEXT128 GR64:$src1), bdxaddr20only:$src2)>;
1246
1247//===----------------------------------------------------------------------===//
1248// Division and remainder
1249//===----------------------------------------------------------------------===//
1250
1251let hasSideEffects = 1 in {  // Do not speculatively execute.
1252  // Division and remainder, from registers.
1253  def DR    : BinaryRR <"dr",    0x1D,   null_frag, GR128, GR32>;
1254  def DSGFR : BinaryRRE<"dsgfr", 0xB91D, null_frag, GR128, GR32>;
1255  def DSGR  : BinaryRRE<"dsgr",  0xB90D, null_frag, GR128, GR64>;
1256  def DLR   : BinaryRRE<"dlr",   0xB997, null_frag, GR128, GR32>;
1257  def DLGR  : BinaryRRE<"dlgr",  0xB987, null_frag, GR128, GR64>;
1258
1259  // Division and remainder, from memory.
1260  def D    : BinaryRX <"d",    0x5D,   null_frag, GR128, load, 4>;
1261  def DSGF : BinaryRXY<"dsgf", 0xE31D, null_frag, GR128, load, 4>;
1262  def DSG  : BinaryRXY<"dsg",  0xE30D, null_frag, GR128, load, 8>;
1263  def DL   : BinaryRXY<"dl",   0xE397, null_frag, GR128, load, 4>;
1264  def DLG  : BinaryRXY<"dlg",  0xE387, null_frag, GR128, load, 8>;
1265}
1266def : Pat<(z_sdivrem GR64:$src1, GR32:$src2),
1267          (DSGFR (AEXT128 GR64:$src1), GR32:$src2)>;
1268def : Pat<(z_sdivrem GR64:$src1, (i32 (load bdxaddr20only:$src2))),
1269          (DSGF (AEXT128 GR64:$src1), bdxaddr20only:$src2)>;
1270def : Pat<(z_sdivrem GR64:$src1, GR64:$src2),
1271          (DSGR (AEXT128 GR64:$src1), GR64:$src2)>;
1272def : Pat<(z_sdivrem GR64:$src1, (i64 (load bdxaddr20only:$src2))),
1273          (DSG (AEXT128 GR64:$src1), bdxaddr20only:$src2)>;
1274
1275def : Pat<(z_udivrem GR32:$src1, GR32:$src2),
1276          (DLR (ZEXT128 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src1,
1277                                       subreg_l32)), GR32:$src2)>;
1278def : Pat<(z_udivrem GR32:$src1, (i32 (load bdxaddr20only:$src2))),
1279          (DL (ZEXT128 (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src1,
1280                                      subreg_l32)), bdxaddr20only:$src2)>;
1281def : Pat<(z_udivrem GR64:$src1, GR64:$src2),
1282          (DLGR (ZEXT128 GR64:$src1), GR64:$src2)>;
1283def : Pat<(z_udivrem GR64:$src1, (i64 (load bdxaddr20only:$src2))),
1284          (DLG (ZEXT128 GR64:$src1), bdxaddr20only:$src2)>;
1285
1286//===----------------------------------------------------------------------===//
1287// Shifts
1288//===----------------------------------------------------------------------===//
1289
1290// Logical shift left.
1291let hasSideEffects = 0 in {
1292  defm SLL : BinaryRSAndK<"sll", 0x89, 0xEBDF, shl, GR32>;
1293  def SLLG : BinaryRSY<"sllg", 0xEB0D, shl, GR64>;
1294  def SLDL : BinaryRS<"sldl", 0x8D, null_frag, GR128>;
1295}
1296
1297// Arithmetic shift left.
1298let Defs = [CC] in {
1299  defm SLA : BinaryRSAndK<"sla", 0x8B, 0xEBDD, null_frag, GR32>;
1300  def SLAG : BinaryRSY<"slag", 0xEB0B, null_frag, GR64>;
1301  def SLDA : BinaryRS<"slda", 0x8F, null_frag, GR128>;
1302}
1303
1304// Logical shift right.
1305let hasSideEffects = 0 in {
1306  defm SRL : BinaryRSAndK<"srl", 0x88, 0xEBDE, srl, GR32>;
1307  def SRLG : BinaryRSY<"srlg", 0xEB0C, srl, GR64>;
1308  def SRDL : BinaryRS<"srdl", 0x8C, null_frag, GR128>;
1309}
1310
1311// Arithmetic shift right.
1312let Defs = [CC], CCValues = 0xE, CompareZeroCCMask = 0xE in {
1313  defm SRA : BinaryRSAndK<"sra", 0x8A, 0xEBDC, sra, GR32>;
1314  def SRAG : BinaryRSY<"srag", 0xEB0A, sra, GR64>;
1315  def SRDA : BinaryRS<"srda", 0x8E, null_frag, GR128>;
1316}
1317
1318// Rotate left.
1319let hasSideEffects = 0 in {
1320  def RLL  : BinaryRSY<"rll",  0xEB1D, rotl, GR32>;
1321  def RLLG : BinaryRSY<"rllg", 0xEB1C, rotl, GR64>;
1322}
1323
1324// Rotate second operand left and inserted selected bits into first operand.
1325// These can act like 32-bit operands provided that the constant start and
1326// end bits (operands 2 and 3) are in the range [32, 64).
1327let Defs = [CC] in {
1328  let isCodeGenOnly = 1 in
1329    def RISBG32 : RotateSelectRIEf<"risbg", 0xEC55, GR32, GR32>;
1330  let CCValues = 0xE, CompareZeroCCMask = 0xE in
1331    def RISBG : RotateSelectRIEf<"risbg", 0xEC55, GR64, GR64>;
1332}
1333
1334// On zEC12 we have a variant of RISBG that does not set CC.
1335let Predicates = [FeatureMiscellaneousExtensions] in
1336  def RISBGN : RotateSelectRIEf<"risbgn", 0xEC59, GR64, GR64>;
1337
1338// Forms of RISBG that only affect one word of the destination register.
1339// They do not set CC.
1340let Predicates = [FeatureHighWord] in {
1341  def RISBMux : RotateSelectRIEfPseudo<GRX32, GRX32>;
1342  def RISBLL  : RotateSelectAliasRIEf<GR32,  GR32>;
1343  def RISBLH  : RotateSelectAliasRIEf<GR32,  GRH32>;
1344  def RISBHL  : RotateSelectAliasRIEf<GRH32, GR32>;
1345  def RISBHH  : RotateSelectAliasRIEf<GRH32, GRH32>;
1346  def RISBLG  : RotateSelectRIEf<"risblg", 0xEC51, GR32, GR64>;
1347  def RISBHG  : RotateSelectRIEf<"risbhg", 0xEC5D, GRH32, GR64>;
1348}
1349
1350// Rotate second operand left and perform a logical operation with selected
1351// bits of the first operand.  The CC result only describes the selected bits,
1352// so isn't useful for a full comparison against zero.
1353let Defs = [CC] in {
1354  def RNSBG : RotateSelectRIEf<"rnsbg", 0xEC54, GR64, GR64>;
1355  def ROSBG : RotateSelectRIEf<"rosbg", 0xEC56, GR64, GR64>;
1356  def RXSBG : RotateSelectRIEf<"rxsbg", 0xEC57, GR64, GR64>;
1357}
1358
1359//===----------------------------------------------------------------------===//
1360// Comparison
1361//===----------------------------------------------------------------------===//
1362
1363// Signed comparisons.  We put these before the unsigned comparisons because
1364// some of the signed forms have COMPARE AND BRANCH equivalents whereas none
1365// of the unsigned forms do.
1366let Defs = [CC], CCValues = 0xE in {
1367  // Comparison with a register.
1368  def CR   : CompareRR <"cr",   0x19,   z_scmp,    GR32, GR32>;
1369  def CGFR : CompareRRE<"cgfr", 0xB930, null_frag, GR64, GR32>;
1370  def CGR  : CompareRRE<"cgr",  0xB920, z_scmp,    GR64, GR64>;
1371
1372  // Comparison with a high register.
1373  def CHHR : CompareRRE<"chhr", 0xB9CD, null_frag, GRH32, GRH32>,
1374             Requires<[FeatureHighWord]>;
1375  def CHLR : CompareRRE<"chlr", 0xB9DD, null_frag, GRH32, GR32>,
1376             Requires<[FeatureHighWord]>;
1377
1378  // Comparison with a signed 16-bit immediate.  CHIMux expands to CHI or CIH,
1379  // depending on the choice of register.
1380  def CHIMux : CompareRIPseudo<z_scmp, GRX32, imm32sx16>,
1381               Requires<[FeatureHighWord]>;
1382  def CHI  : CompareRI<"chi",  0xA7E, z_scmp, GR32, imm32sx16>;
1383  def CGHI : CompareRI<"cghi", 0xA7F, z_scmp, GR64, imm64sx16>;
1384
1385  // Comparison with a signed 32-bit immediate.  CFIMux expands to CFI or CIH,
1386  // depending on the choice of register.
1387  def CFIMux : CompareRIPseudo<z_scmp, GRX32, simm32>,
1388               Requires<[FeatureHighWord]>;
1389  def CFI  : CompareRIL<"cfi",  0xC2D, z_scmp, GR32, simm32>;
1390  def CIH  : CompareRIL<"cih",  0xCCD, z_scmp, GRH32, simm32>,
1391             Requires<[FeatureHighWord]>;
1392  def CGFI : CompareRIL<"cgfi", 0xC2C, z_scmp, GR64, imm64sx32>;
1393
1394  // Comparison with memory.
1395  defm CH    : CompareRXPair<"ch", 0x49, 0xE379, z_scmp, GR32, asextloadi16, 2>;
1396  def  CMux  : CompareRXYPseudo<z_scmp, GRX32, load, 4>,
1397               Requires<[FeatureHighWord]>;
1398  defm C     : CompareRXPair<"c",  0x59, 0xE359, z_scmp, GR32, load, 4>;
1399  def  CHF   : CompareRXY<"chf", 0xE3CD, z_scmp, GRH32, load, 4>,
1400               Requires<[FeatureHighWord]>;
1401  def  CGH   : CompareRXY<"cgh", 0xE334, z_scmp, GR64, asextloadi16, 2>;
1402  def  CGF   : CompareRXY<"cgf", 0xE330, z_scmp, GR64, asextloadi32, 4>;
1403  def  CG    : CompareRXY<"cg",  0xE320, z_scmp, GR64, load, 8>;
1404  def  CHRL  : CompareRILPC<"chrl",  0xC65, z_scmp, GR32, aligned_asextloadi16>;
1405  def  CRL   : CompareRILPC<"crl",   0xC6D, z_scmp, GR32, aligned_load>;
1406  def  CGHRL : CompareRILPC<"cghrl", 0xC64, z_scmp, GR64, aligned_asextloadi16>;
1407  def  CGFRL : CompareRILPC<"cgfrl", 0xC6C, z_scmp, GR64, aligned_asextloadi32>;
1408  def  CGRL  : CompareRILPC<"cgrl",  0xC68, z_scmp, GR64, aligned_load>;
1409
1410  // Comparison between memory and a signed 16-bit immediate.
1411  def CHHSI : CompareSIL<"chhsi", 0xE554, z_scmp, asextloadi16, imm32sx16>;
1412  def CHSI  : CompareSIL<"chsi",  0xE55C, z_scmp, load, imm32sx16>;
1413  def CGHSI : CompareSIL<"cghsi", 0xE558, z_scmp, load, imm64sx16>;
1414}
1415defm : SXB<z_scmp, GR64, CGFR>;
1416
1417// Unsigned comparisons.
1418let Defs = [CC], CCValues = 0xE, IsLogical = 1 in {
1419  // Comparison with a register.
1420  def CLR   : CompareRR <"clr",   0x15,   z_ucmp,    GR32, GR32>;
1421  def CLGFR : CompareRRE<"clgfr", 0xB931, null_frag, GR64, GR32>;
1422  def CLGR  : CompareRRE<"clgr",  0xB921, z_ucmp,    GR64, GR64>;
1423
1424  // Comparison with a high register.
1425  def CLHHR : CompareRRE<"clhhr", 0xB9CF, null_frag, GRH32, GRH32>,
1426              Requires<[FeatureHighWord]>;
1427  def CLHLR : CompareRRE<"clhlr", 0xB9DF, null_frag, GRH32, GR32>,
1428              Requires<[FeatureHighWord]>;
1429
1430  // Comparison with an unsigned 32-bit immediate.  CLFIMux expands to CLFI
1431  // or CLIH, depending on the choice of register.
1432  def CLFIMux : CompareRIPseudo<z_ucmp, GRX32, uimm32>,
1433                Requires<[FeatureHighWord]>;
1434  def CLFI  : CompareRIL<"clfi",  0xC2F, z_ucmp, GR32, uimm32>;
1435  def CLIH  : CompareRIL<"clih",  0xCCF, z_ucmp, GRH32, uimm32>,
1436              Requires<[FeatureHighWord]>;
1437  def CLGFI : CompareRIL<"clgfi", 0xC2E, z_ucmp, GR64, imm64zx32>;
1438
1439  // Comparison with memory.
1440  def  CLMux  : CompareRXYPseudo<z_ucmp, GRX32, load, 4>,
1441                Requires<[FeatureHighWord]>;
1442  defm CL     : CompareRXPair<"cl", 0x55, 0xE355, z_ucmp, GR32, load, 4>;
1443  def  CLHF   : CompareRXY<"clhf", 0xE3CF, z_ucmp, GRH32, load, 4>,
1444                Requires<[FeatureHighWord]>;
1445  def  CLGF   : CompareRXY<"clgf", 0xE331, z_ucmp, GR64, azextloadi32, 4>;
1446  def  CLG    : CompareRXY<"clg",  0xE321, z_ucmp, GR64, load, 8>;
1447  def  CLHRL  : CompareRILPC<"clhrl",  0xC67, z_ucmp, GR32,
1448                             aligned_azextloadi16>;
1449  def  CLRL   : CompareRILPC<"clrl",   0xC6F, z_ucmp, GR32,
1450                             aligned_load>;
1451  def  CLGHRL : CompareRILPC<"clghrl", 0xC66, z_ucmp, GR64,
1452                             aligned_azextloadi16>;
1453  def  CLGFRL : CompareRILPC<"clgfrl", 0xC6E, z_ucmp, GR64,
1454                             aligned_azextloadi32>;
1455  def  CLGRL  : CompareRILPC<"clgrl",  0xC6A, z_ucmp, GR64,
1456                             aligned_load>;
1457
1458  // Comparison between memory and an unsigned 8-bit immediate.
1459  defm CLI : CompareSIPair<"cli", 0x95, 0xEB55, z_ucmp, azextloadi8, imm32zx8>;
1460
1461  // Comparison between memory and an unsigned 16-bit immediate.
1462  def CLHHSI : CompareSIL<"clhhsi", 0xE555, z_ucmp, azextloadi16, imm32zx16>;
1463  def CLFHSI : CompareSIL<"clfhsi", 0xE55D, z_ucmp, load, imm32zx16>;
1464  def CLGHSI : CompareSIL<"clghsi", 0xE559, z_ucmp, load, imm64zx16>;
1465}
1466defm : ZXB<z_ucmp, GR64, CLGFR>;
1467
1468// Memory-to-memory comparison.
1469let mayLoad = 1, Defs = [CC] in {
1470  defm CLC : MemorySS<"clc", 0xD5, z_clc, z_clc_loop>;
1471  def CLCL  : SideEffectBinaryMemMemRR<"clcl", 0x0F, GR128, GR128>;
1472  def CLCLE : SideEffectTernaryMemMemRS<"clcle", 0xA9, GR128, GR128>;
1473  def CLCLU : SideEffectTernaryMemMemRSY<"clclu", 0xEB8F, GR128, GR128>;
1474}
1475
1476// String comparison.
1477let mayLoad = 1, Defs = [CC] in
1478  defm CLST : StringRRE<"clst", 0xB25D, z_strcmp>;
1479
1480// Test under mask.
1481let Defs = [CC] in {
1482  // TMxMux expands to TM[LH]x, depending on the choice of register.
1483  def TMLMux : CompareRIPseudo<z_tm_reg, GRX32, imm32ll16>,
1484               Requires<[FeatureHighWord]>;
1485  def TMHMux : CompareRIPseudo<z_tm_reg, GRX32, imm32lh16>,
1486               Requires<[FeatureHighWord]>;
1487  def TMLL : CompareRI<"tmll", 0xA71, z_tm_reg, GR32, imm32ll16>;
1488  def TMLH : CompareRI<"tmlh", 0xA70, z_tm_reg, GR32, imm32lh16>;
1489  def TMHL : CompareRI<"tmhl", 0xA73, z_tm_reg, GRH32, imm32ll16>;
1490  def TMHH : CompareRI<"tmhh", 0xA72, z_tm_reg, GRH32, imm32lh16>;
1491
1492  def TMLL64 : CompareAliasRI<z_tm_reg, GR64, imm64ll16>;
1493  def TMLH64 : CompareAliasRI<z_tm_reg, GR64, imm64lh16>;
1494  def TMHL64 : CompareAliasRI<z_tm_reg, GR64, imm64hl16>;
1495  def TMHH64 : CompareAliasRI<z_tm_reg, GR64, imm64hh16>;
1496
1497  defm TM : CompareSIPair<"tm", 0x91, 0xEB51, z_tm_mem, anyextloadi8, imm32zx8>;
1498}
1499
1500def TML : InstAlias<"tml\t$R, $I", (TMLL GR32:$R, imm32ll16:$I), 0>;
1501def TMH : InstAlias<"tmh\t$R, $I", (TMLH GR32:$R, imm32lh16:$I), 0>;
1502
1503// Compare logical characters under mask -- not (yet) used for codegen.
1504let Defs = [CC] in {
1505  defm CLM : CompareRSPair<"clm", 0xBD, 0xEB21, GR32, 0>;
1506  def CLMH : CompareRSY<"clmh", 0xEB20, GRH32, 0>;
1507}
1508
1509//===----------------------------------------------------------------------===//
1510// Prefetch and execution hint
1511//===----------------------------------------------------------------------===//
1512
1513def PFD : PrefetchRXY<"pfd", 0xE336, z_prefetch>;
1514def PFDRL : PrefetchRILPC<"pfdrl", 0xC62, z_prefetch>;
1515
1516let Predicates = [FeatureExecutionHint] in {
1517  // Branch Prediction Preload
1518  def BPP : BranchPreloadSMI<"bpp", 0xC7>;
1519  def BPRP : BranchPreloadMII<"bprp", 0xC5>;
1520
1521  // Next Instruction Access Intent
1522  def NIAI : SideEffectBinaryIE<"niai", 0xB2FA, imm32zx4, imm32zx4>;
1523}
1524
1525//===----------------------------------------------------------------------===//
1526// Atomic operations
1527//===----------------------------------------------------------------------===//
1528
1529// A serialization instruction that acts as a barrier for all memory
1530// accesses, which expands to "bcr 14, 0".
1531let hasSideEffects = 1 in
1532def Serialize : Alias<2, (outs), (ins), []>;
1533
1534// A pseudo instruction that serves as a compiler barrier.
1535let hasSideEffects = 1, hasNoSchedulingInfo = 1 in
1536def MemBarrier : Pseudo<(outs), (ins), [(z_membarrier)]>;
1537
1538let Predicates = [FeatureInterlockedAccess1], Defs = [CC] in {
1539  def LAA   : LoadAndOpRSY<"laa",   0xEBF8, atomic_load_add_32, GR32>;
1540  def LAAG  : LoadAndOpRSY<"laag",  0xEBE8, atomic_load_add_64, GR64>;
1541  def LAAL  : LoadAndOpRSY<"laal",  0xEBFA, null_frag, GR32>;
1542  def LAALG : LoadAndOpRSY<"laalg", 0xEBEA, null_frag, GR64>;
1543  def LAN   : LoadAndOpRSY<"lan",   0xEBF4, atomic_load_and_32, GR32>;
1544  def LANG  : LoadAndOpRSY<"lang",  0xEBE4, atomic_load_and_64, GR64>;
1545  def LAO   : LoadAndOpRSY<"lao",   0xEBF6, atomic_load_or_32, GR32>;
1546  def LAOG  : LoadAndOpRSY<"laog",  0xEBE6, atomic_load_or_64, GR64>;
1547  def LAX   : LoadAndOpRSY<"lax",   0xEBF7, atomic_load_xor_32, GR32>;
1548  def LAXG  : LoadAndOpRSY<"laxg",  0xEBE7, atomic_load_xor_64, GR64>;
1549}
1550
1551def ATOMIC_SWAPW   : AtomicLoadWBinaryReg<z_atomic_swapw>;
1552def ATOMIC_SWAP_32 : AtomicLoadBinaryReg32<atomic_swap_32>;
1553def ATOMIC_SWAP_64 : AtomicLoadBinaryReg64<atomic_swap_64>;
1554
1555def ATOMIC_LOADW_AR  : AtomicLoadWBinaryReg<z_atomic_loadw_add>;
1556def ATOMIC_LOADW_AFI : AtomicLoadWBinaryImm<z_atomic_loadw_add, simm32>;
1557let Predicates = [FeatureNoInterlockedAccess1] in {
1558  def ATOMIC_LOAD_AR   : AtomicLoadBinaryReg32<atomic_load_add_32>;
1559  def ATOMIC_LOAD_AHI  : AtomicLoadBinaryImm32<atomic_load_add_32, imm32sx16>;
1560  def ATOMIC_LOAD_AFI  : AtomicLoadBinaryImm32<atomic_load_add_32, simm32>;
1561  def ATOMIC_LOAD_AGR  : AtomicLoadBinaryReg64<atomic_load_add_64>;
1562  def ATOMIC_LOAD_AGHI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx16>;
1563  def ATOMIC_LOAD_AGFI : AtomicLoadBinaryImm64<atomic_load_add_64, imm64sx32>;
1564}
1565
1566def ATOMIC_LOADW_SR : AtomicLoadWBinaryReg<z_atomic_loadw_sub>;
1567def ATOMIC_LOAD_SR  : AtomicLoadBinaryReg32<atomic_load_sub_32>;
1568def ATOMIC_LOAD_SGR : AtomicLoadBinaryReg64<atomic_load_sub_64>;
1569
1570def ATOMIC_LOADW_NR   : AtomicLoadWBinaryReg<z_atomic_loadw_and>;
1571def ATOMIC_LOADW_NILH : AtomicLoadWBinaryImm<z_atomic_loadw_and, imm32lh16c>;
1572let Predicates = [FeatureNoInterlockedAccess1] in {
1573  def ATOMIC_LOAD_NR     : AtomicLoadBinaryReg32<atomic_load_and_32>;
1574  def ATOMIC_LOAD_NILL   : AtomicLoadBinaryImm32<atomic_load_and_32,
1575                                                 imm32ll16c>;
1576  def ATOMIC_LOAD_NILH   : AtomicLoadBinaryImm32<atomic_load_and_32,
1577                                                 imm32lh16c>;
1578  def ATOMIC_LOAD_NILF   : AtomicLoadBinaryImm32<atomic_load_and_32, uimm32>;
1579  def ATOMIC_LOAD_NGR    : AtomicLoadBinaryReg64<atomic_load_and_64>;
1580  def ATOMIC_LOAD_NILL64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1581                                                 imm64ll16c>;
1582  def ATOMIC_LOAD_NILH64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1583                                                 imm64lh16c>;
1584  def ATOMIC_LOAD_NIHL64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1585                                                 imm64hl16c>;
1586  def ATOMIC_LOAD_NIHH64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1587                                                 imm64hh16c>;
1588  def ATOMIC_LOAD_NILF64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1589                                                 imm64lf32c>;
1590  def ATOMIC_LOAD_NIHF64 : AtomicLoadBinaryImm64<atomic_load_and_64,
1591                                                 imm64hf32c>;
1592}
1593
1594def ATOMIC_LOADW_OR     : AtomicLoadWBinaryReg<z_atomic_loadw_or>;
1595def ATOMIC_LOADW_OILH   : AtomicLoadWBinaryImm<z_atomic_loadw_or, imm32lh16>;
1596let Predicates = [FeatureNoInterlockedAccess1] in {
1597  def ATOMIC_LOAD_OR     : AtomicLoadBinaryReg32<atomic_load_or_32>;
1598  def ATOMIC_LOAD_OILL   : AtomicLoadBinaryImm32<atomic_load_or_32, imm32ll16>;
1599  def ATOMIC_LOAD_OILH   : AtomicLoadBinaryImm32<atomic_load_or_32, imm32lh16>;
1600  def ATOMIC_LOAD_OILF   : AtomicLoadBinaryImm32<atomic_load_or_32, uimm32>;
1601  def ATOMIC_LOAD_OGR    : AtomicLoadBinaryReg64<atomic_load_or_64>;
1602  def ATOMIC_LOAD_OILL64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64ll16>;
1603  def ATOMIC_LOAD_OILH64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lh16>;
1604  def ATOMIC_LOAD_OIHL64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hl16>;
1605  def ATOMIC_LOAD_OIHH64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hh16>;
1606  def ATOMIC_LOAD_OILF64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64lf32>;
1607  def ATOMIC_LOAD_OIHF64 : AtomicLoadBinaryImm64<atomic_load_or_64, imm64hf32>;
1608}
1609
1610def ATOMIC_LOADW_XR     : AtomicLoadWBinaryReg<z_atomic_loadw_xor>;
1611def ATOMIC_LOADW_XILF   : AtomicLoadWBinaryImm<z_atomic_loadw_xor, uimm32>;
1612let Predicates = [FeatureNoInterlockedAccess1] in {
1613  def ATOMIC_LOAD_XR     : AtomicLoadBinaryReg32<atomic_load_xor_32>;
1614  def ATOMIC_LOAD_XILF   : AtomicLoadBinaryImm32<atomic_load_xor_32, uimm32>;
1615  def ATOMIC_LOAD_XGR    : AtomicLoadBinaryReg64<atomic_load_xor_64>;
1616  def ATOMIC_LOAD_XILF64 : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64lf32>;
1617  def ATOMIC_LOAD_XIHF64 : AtomicLoadBinaryImm64<atomic_load_xor_64, imm64hf32>;
1618}
1619
1620def ATOMIC_LOADW_NRi    : AtomicLoadWBinaryReg<z_atomic_loadw_nand>;
1621def ATOMIC_LOADW_NILHi  : AtomicLoadWBinaryImm<z_atomic_loadw_nand,
1622                                               imm32lh16c>;
1623def ATOMIC_LOAD_NRi     : AtomicLoadBinaryReg32<atomic_load_nand_32>;
1624def ATOMIC_LOAD_NILLi   : AtomicLoadBinaryImm32<atomic_load_nand_32,
1625                                                imm32ll16c>;
1626def ATOMIC_LOAD_NILHi   : AtomicLoadBinaryImm32<atomic_load_nand_32,
1627                                                imm32lh16c>;
1628def ATOMIC_LOAD_NILFi   : AtomicLoadBinaryImm32<atomic_load_nand_32, uimm32>;
1629def ATOMIC_LOAD_NGRi    : AtomicLoadBinaryReg64<atomic_load_nand_64>;
1630def ATOMIC_LOAD_NILL64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1631                                                imm64ll16c>;
1632def ATOMIC_LOAD_NILH64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1633                                                imm64lh16c>;
1634def ATOMIC_LOAD_NIHL64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1635                                                imm64hl16c>;
1636def ATOMIC_LOAD_NIHH64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1637                                                imm64hh16c>;
1638def ATOMIC_LOAD_NILF64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1639                                                imm64lf32c>;
1640def ATOMIC_LOAD_NIHF64i : AtomicLoadBinaryImm64<atomic_load_nand_64,
1641                                                imm64hf32c>;
1642
1643def ATOMIC_LOADW_MIN    : AtomicLoadWBinaryReg<z_atomic_loadw_min>;
1644def ATOMIC_LOAD_MIN_32  : AtomicLoadBinaryReg32<atomic_load_min_32>;
1645def ATOMIC_LOAD_MIN_64  : AtomicLoadBinaryReg64<atomic_load_min_64>;
1646
1647def ATOMIC_LOADW_MAX    : AtomicLoadWBinaryReg<z_atomic_loadw_max>;
1648def ATOMIC_LOAD_MAX_32  : AtomicLoadBinaryReg32<atomic_load_max_32>;
1649def ATOMIC_LOAD_MAX_64  : AtomicLoadBinaryReg64<atomic_load_max_64>;
1650
1651def ATOMIC_LOADW_UMIN   : AtomicLoadWBinaryReg<z_atomic_loadw_umin>;
1652def ATOMIC_LOAD_UMIN_32 : AtomicLoadBinaryReg32<atomic_load_umin_32>;
1653def ATOMIC_LOAD_UMIN_64 : AtomicLoadBinaryReg64<atomic_load_umin_64>;
1654
1655def ATOMIC_LOADW_UMAX   : AtomicLoadWBinaryReg<z_atomic_loadw_umax>;
1656def ATOMIC_LOAD_UMAX_32 : AtomicLoadBinaryReg32<atomic_load_umax_32>;
1657def ATOMIC_LOAD_UMAX_64 : AtomicLoadBinaryReg64<atomic_load_umax_64>;
1658
1659def ATOMIC_CMP_SWAPW
1660  : Pseudo<(outs GR32:$dst), (ins bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1661                                  ADDR32:$bitshift, ADDR32:$negbitshift,
1662                                  uimm32:$bitsize),
1663           [(set GR32:$dst,
1664                 (z_atomic_cmp_swapw bdaddr20only:$addr, GR32:$cmp, GR32:$swap,
1665                                     ADDR32:$bitshift, ADDR32:$negbitshift,
1666                                     uimm32:$bitsize))]> {
1667  let Defs = [CC];
1668  let mayLoad = 1;
1669  let mayStore = 1;
1670  let usesCustomInserter = 1;
1671  let hasNoSchedulingInfo = 1;
1672}
1673
1674// Test and set.
1675let mayLoad = 1, Defs = [CC] in
1676  def TS : StoreInherentS<"ts", 0x9300, null_frag, 1>;
1677
1678// Compare and swap.
1679let Defs = [CC] in {
1680  defm CS  : CmpSwapRSPair<"cs", 0xBA, 0xEB14, atomic_cmp_swap_32, GR32>;
1681  def  CSG : CmpSwapRSY<"csg", 0xEB30, atomic_cmp_swap_64, GR64>;
1682}
1683
1684// Compare double and swap.
1685let Defs = [CC] in {
1686  defm CDS  : CmpSwapRSPair<"cds", 0xBB, 0xEB31, null_frag, GR128>;
1687  def  CDSG : CmpSwapRSY<"cdsg", 0xEB3E, null_frag, GR128>;
1688}
1689
1690// Compare and swap and store.
1691let Uses = [R0L, R1D], Defs = [CC], mayStore = 1, mayLoad = 1 in
1692  def CSST : SideEffectTernarySSF<"csst", 0xC82, GR64>;
1693
1694// Perform locked operation.
1695let Uses = [R0L, R1D], Defs = [CC], mayStore = 1, mayLoad =1 in
1696  def PLO : SideEffectQuaternarySSe<"plo", 0xEE, GR64>;
1697
1698// Load/store pair from/to quadword.
1699def LPQ  : UnaryRXY<"lpq", 0xE38F, null_frag, GR128, 16>;
1700def STPQ : StoreRXY<"stpq", 0xE38E, null_frag, GR128, 16>;
1701
1702// Load pair disjoint.
1703let Predicates = [FeatureInterlockedAccess1], Defs = [CC] in {
1704  def LPD  : BinarySSF<"lpd", 0xC84, GR128>;
1705  def LPDG : BinarySSF<"lpdg", 0xC85, GR128>;
1706}
1707
1708//===----------------------------------------------------------------------===//
1709// Translate and convert
1710//===----------------------------------------------------------------------===//
1711
1712let mayLoad = 1, mayStore = 1 in
1713  def TR : SideEffectBinarySSa<"tr", 0xDC>;
1714
1715let mayLoad = 1, Defs = [CC, R0L, R1D] in {
1716  def TRT  : SideEffectBinarySSa<"trt", 0xDD>;
1717  def TRTR : SideEffectBinarySSa<"trtr", 0xD0>;
1718}
1719
1720let mayLoad = 1, mayStore = 1, Uses = [R0L] in
1721  def TRE : SideEffectBinaryMemMemRRE<"tre", 0xB2A5, GR128, GR64>;
1722
1723let mayLoad = 1, Uses = [R1D], Defs = [CC] in {
1724  defm TRTE  : BinaryMemRRFcOpt<"trte",  0xB9BF, GR128, GR64>;
1725  defm TRTRE : BinaryMemRRFcOpt<"trtre", 0xB9BD, GR128, GR64>;
1726}
1727
1728let mayLoad = 1, mayStore = 1, Uses = [R0L, R1D], Defs = [CC] in {
1729  defm TROO : SideEffectTernaryMemMemRRFcOpt<"troo", 0xB993, GR128, GR64>;
1730  defm TROT : SideEffectTernaryMemMemRRFcOpt<"trot", 0xB992, GR128, GR64>;
1731  defm TRTO : SideEffectTernaryMemMemRRFcOpt<"trto", 0xB991, GR128, GR64>;
1732  defm TRTT : SideEffectTernaryMemMemRRFcOpt<"trtt", 0xB990, GR128, GR64>;
1733}
1734
1735let mayLoad = 1, mayStore = 1, Defs = [CC] in {
1736  defm CU12 : SideEffectTernaryMemMemRRFcOpt<"cu12", 0xB2A7, GR128, GR128>;
1737  defm CU14 : SideEffectTernaryMemMemRRFcOpt<"cu14", 0xB9B0, GR128, GR128>;
1738  defm CU21 : SideEffectTernaryMemMemRRFcOpt<"cu21", 0xB2A6, GR128, GR128>;
1739  defm CU24 : SideEffectTernaryMemMemRRFcOpt<"cu24", 0xB9B1, GR128, GR128>;
1740  def  CU41 : SideEffectBinaryMemMemRRE<"cu41", 0xB9B2, GR128, GR128>;
1741  def  CU42 : SideEffectBinaryMemMemRRE<"cu42", 0xB9B3, GR128, GR128>;
1742
1743  let isAsmParserOnly = 1 in {
1744    defm CUUTF : SideEffectTernaryMemMemRRFcOpt<"cuutf", 0xB2A6, GR128, GR128>;
1745    defm CUTFU : SideEffectTernaryMemMemRRFcOpt<"cutfu", 0xB2A7, GR128, GR128>;
1746  }
1747}
1748
1749//===----------------------------------------------------------------------===//
1750// Message-security assist
1751//===----------------------------------------------------------------------===//
1752
1753let mayLoad = 1, mayStore = 1, Uses = [R0L, R1D], Defs = [CC] in {
1754  def KM  : SideEffectBinaryMemMemRRE<"km",  0xB92E, GR128, GR128>;
1755  def KMC : SideEffectBinaryMemMemRRE<"kmc", 0xB92F, GR128, GR128>;
1756
1757  def KIMD : SideEffectBinaryMemRRE<"kimd", 0xB93E, GR64, GR128>;
1758  def KLMD : SideEffectBinaryMemRRE<"klmd", 0xB93F, GR64, GR128>;
1759  def KMAC : SideEffectBinaryMemRRE<"kmac", 0xB91E, GR64, GR128>;
1760
1761  let Predicates = [FeatureMessageSecurityAssist4] in {
1762    def KMF   : SideEffectBinaryMemMemRRE<"kmf", 0xB92A, GR128, GR128>;
1763    def KMO   : SideEffectBinaryMemMemRRE<"kmo", 0xB92B, GR128, GR128>;
1764    def KMCTR : SideEffectTernaryMemMemMemRRFb<"kmctr", 0xB92D,
1765                                               GR128, GR128, GR128>;
1766    def PCC   : SideEffectInherentRRE<"pcc", 0xB92C>;
1767  }
1768  let Predicates = [FeatureMessageSecurityAssist5] in
1769    def PPNO  : SideEffectBinaryMemMemRRE<"ppno", 0xB93C, GR128, GR128>;
1770}
1771
1772//===----------------------------------------------------------------------===//
1773// Decimal arithmetic
1774//===----------------------------------------------------------------------===//
1775
1776defm CVB  : BinaryRXPair<"cvb",0x4F, 0xE306, null_frag, GR32, load, 4>;
1777def  CVBG : BinaryRXY<"cvbg", 0xE30E, null_frag, GR64, load, 8>;
1778
1779defm CVD  : StoreRXPair<"cvd", 0x4E, 0xE326, null_frag, GR32, 4>;
1780def  CVDG : StoreRXY<"cvdg", 0xE32E, null_frag, GR64, 8>;
1781
1782let mayLoad = 1, mayStore = 1 in {
1783  def MVN : SideEffectBinarySSa<"mvn", 0xD1>;
1784  def MVZ : SideEffectBinarySSa<"mvz", 0xD3>;
1785  def MVO : SideEffectBinarySSb<"mvo", 0xF1>;
1786
1787  def PACK : SideEffectBinarySSb<"pack", 0xF2>;
1788  def PKA  : SideEffectBinarySSf<"pka", 0xE9>;
1789  def PKU  : SideEffectBinarySSf<"pku", 0xE1>;
1790  def UNPK : SideEffectBinarySSb<"unpk", 0xF3>;
1791  let Defs = [CC] in {
1792    def UNPKA : SideEffectBinarySSa<"unpka", 0xEA>;
1793    def UNPKU : SideEffectBinarySSa<"unpku", 0xE2>;
1794  }
1795}
1796
1797let mayLoad = 1, mayStore = 1 in {
1798  let Defs = [CC] in {
1799    def AP : SideEffectBinarySSb<"ap", 0xFA>;
1800    def SP : SideEffectBinarySSb<"sp", 0xFB>;
1801    def ZAP : SideEffectBinarySSb<"zap", 0xF8>;
1802    def SRP : SideEffectTernarySSc<"srp", 0xF0>;
1803  }
1804  def MP : SideEffectBinarySSb<"mp", 0xFC>;
1805  def DP : SideEffectBinarySSb<"dp", 0xFD>;
1806  let Defs = [CC] in {
1807    def ED : SideEffectBinarySSa<"ed", 0xDE>;
1808    def EDMK : SideEffectBinarySSa<"edmk", 0xDF>;
1809  }
1810}
1811
1812let Defs = [CC] in {
1813  def CP : CompareSSb<"cp", 0xF9>;
1814  def TP : TestRSL<"tp", 0xEBC0>;
1815}
1816
1817//===----------------------------------------------------------------------===//
1818// Access registers
1819//===----------------------------------------------------------------------===//
1820
1821// Read a 32-bit access register into a GR32.  As with all GR32 operations,
1822// the upper 32 bits of the enclosing GR64 remain unchanged, which is useful
1823// when a 64-bit address is stored in a pair of access registers.
1824def EAR : UnaryRRE<"ear", 0xB24F, null_frag, GR32, AR32>;
1825
1826// Set access register.
1827def SAR : UnaryRRE<"sar", 0xB24E, null_frag, AR32, GR32>;
1828
1829// Copy access register.
1830def CPYA : UnaryRRE<"cpya", 0xB24D, null_frag, AR32, AR32>;
1831
1832// Load address extended.
1833defm LAE : LoadAddressRXPair<"lae", 0x51, 0xE375, null_frag>;
1834
1835// Load access multiple.
1836defm LAM : LoadMultipleRSPair<"lam", 0x9A, 0xEB9A, AR32>;
1837
1838// Load access multiple.
1839defm STAM : StoreMultipleRSPair<"stam", 0x9B, 0xEB9B, AR32>;
1840
1841//===----------------------------------------------------------------------===//
1842// Program mask and addressing mode
1843//===----------------------------------------------------------------------===//
1844
1845// Extract CC and program mask into a register.  CC ends up in bits 29 and 28.
1846let Uses = [CC] in
1847  def IPM : InherentRRE<"ipm", 0xB222, GR32, z_ipm>;
1848
1849// Set CC and program mask from a register.
1850let hasSideEffects = 1, Defs = [CC] in
1851  def SPM : SideEffectUnaryRR<"spm", 0x04, GR32>;
1852
1853// Branch and link - like BAS, but also extracts CC and program mask.
1854let isCall = 1, Uses = [CC], Defs = [CC] in {
1855  def BAL  : CallRX<"bal", 0x45>;
1856  def BALR : CallRR<"balr", 0x05>;
1857}
1858
1859// Test addressing mode.
1860let Defs = [CC] in
1861  def TAM : SideEffectInherentE<"tam", 0x010B>;
1862
1863// Set addressing mode.
1864let hasSideEffects = 1 in {
1865  def SAM24 : SideEffectInherentE<"sam24", 0x010C>;
1866  def SAM31 : SideEffectInherentE<"sam31", 0x010D>;
1867  def SAM64 : SideEffectInherentE<"sam64", 0x010E>;
1868}
1869
1870// Branch and set mode.  Not really a call, but also sets an output register.
1871let isBranch = 1, isTerminator = 1, isBarrier = 1 in
1872  def BSM : CallRR<"bsm", 0x0B>;
1873
1874// Branch and save and set mode.
1875let isCall = 1, Defs = [CC] in
1876  def BASSM : CallRR<"bassm", 0x0C>;
1877
1878//===----------------------------------------------------------------------===//
1879// Transactional execution
1880//===----------------------------------------------------------------------===//
1881
1882let hasSideEffects = 1, Predicates = [FeatureTransactionalExecution] in {
1883  // Transaction Begin
1884  let mayStore = 1, usesCustomInserter = 1, Defs = [CC] in {
1885    def TBEGIN : SideEffectBinarySIL<"tbegin", 0xE560, z_tbegin, imm32zx16>;
1886    def TBEGIN_nofloat : SideEffectBinarySILPseudo<z_tbegin_nofloat, imm32zx16>;
1887
1888    def TBEGINC : SideEffectBinarySIL<"tbeginc", 0xE561,
1889                                      int_s390_tbeginc, imm32zx16>;
1890  }
1891
1892  // Transaction End
1893  let Defs = [CC] in
1894    def TEND : SideEffectInherentS<"tend", 0xB2F8, z_tend>;
1895
1896  // Transaction Abort
1897  let isTerminator = 1, isBarrier = 1 in
1898    def TABORT : SideEffectAddressS<"tabort", 0xB2FC, int_s390_tabort>;
1899
1900  // Nontransactional Store
1901  def NTSTG : StoreRXY<"ntstg", 0xE325, int_s390_ntstg, GR64, 8>;
1902
1903  // Extract Transaction Nesting Depth
1904  def ETND : InherentRRE<"etnd", 0xB2EC, GR32, int_s390_etnd>;
1905}
1906
1907//===----------------------------------------------------------------------===//
1908// Processor assist
1909//===----------------------------------------------------------------------===//
1910
1911let Predicates = [FeatureProcessorAssist] in {
1912  let hasSideEffects = 1 in
1913    def PPA : SideEffectTernaryRRFc<"ppa", 0xB2E8, GR64, GR64, imm32zx4>;
1914  def : Pat<(int_s390_ppa_txassist GR32:$src),
1915            (PPA (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$src, subreg_l32),
1916                 0, 1)>;
1917}
1918
1919//===----------------------------------------------------------------------===//
1920// Miscellaneous Instructions.
1921//===----------------------------------------------------------------------===//
1922
1923// Find leftmost one, AKA count leading zeros.  The instruction actually
1924// returns a pair of GR64s, the first giving the number of leading zeros
1925// and the second giving a copy of the source with the leftmost one bit
1926// cleared.  We only use the first result here.
1927let Defs = [CC] in
1928  def FLOGR : UnaryRRE<"flogr", 0xB983, null_frag, GR128, GR64>;
1929def : Pat<(ctlz GR64:$src),
1930          (EXTRACT_SUBREG (FLOGR GR64:$src), subreg_h64)>;
1931
1932// Population count.  Counts bits set per byte.
1933let Predicates = [FeaturePopulationCount], Defs = [CC] in
1934  def POPCNT : UnaryRRE<"popcnt", 0xB9E1, z_popcnt, GR64, GR64>;
1935
1936// Search a block of memory for a character.
1937let mayLoad = 1, Defs = [CC] in
1938  defm SRST : StringRRE<"srst", 0xB25E, z_search_string>;
1939let mayLoad = 1, Defs = [CC], Uses = [R0L] in
1940  def SRSTU : SideEffectBinaryMemMemRRE<"srstu", 0xB9BE, GR64, GR64>;
1941
1942// Compare until substring equal.
1943let mayLoad = 1, Defs = [CC], Uses = [R0L, R1L] in
1944  def CUSE : SideEffectBinaryMemMemRRE<"cuse", 0xB257, GR128, GR128>;
1945
1946// Compare and form codeword.
1947let mayLoad = 1, Defs = [CC, R1D, R2D, R3D], Uses = [R1D, R2D, R3D] in
1948  def CFC : SideEffectAddressS<"cfc", 0xB21A, null_frag>;
1949
1950// Update tree.
1951let mayLoad = 1, mayStore = 1, Defs = [CC, R0D, R1D, R2D, R3D, R5D],
1952    Uses = [R0D, R1D, R2D, R3D, R4D, R5D] in
1953  def UPT : SideEffectInherentE<"upt", 0x0102>;
1954
1955// Checksum.
1956let mayLoad = 1, Defs = [CC] in
1957  def CKSM : SideEffectBinaryMemMemRRE<"cksm", 0xB241, GR64, GR128>;
1958
1959// Compression call.
1960let mayLoad = 1, mayStore = 1, Defs = [CC, R1D], Uses = [R0L, R1D] in
1961  def CMPSC : SideEffectBinaryMemMemRRE<"cmpsc", 0xB263, GR128, GR128>;
1962
1963// Execute.
1964let hasSideEffects = 1 in {
1965  def EX   : SideEffectBinaryRX<"ex", 0x44, GR64>;
1966  def EXRL : SideEffectBinaryRILPC<"exrl", 0xC60, GR64>;
1967}
1968
1969//===----------------------------------------------------------------------===//
1970// .insn directive instructions
1971//===----------------------------------------------------------------------===//
1972
1973let isCodeGenOnly = 1 in {
1974  def InsnE   : DirectiveInsnE<(outs), (ins imm64zx16:$enc), ".insn e,$enc", []>;
1975  def InsnRI  : DirectiveInsnRI<(outs), (ins imm64zx32:$enc, AnyReg:$R1,
1976                                             imm32sx16:$I2),
1977                                ".insn ri,$enc,$R1,$I2", []>;
1978  def InsnRIE : DirectiveInsnRIE<(outs), (ins imm64zx48:$enc, AnyReg:$R1,
1979                                              AnyReg:$R3, brtarget16:$I2),
1980                                 ".insn rie,$enc,$R1,$R3,$I2", []>;
1981  def InsnRIL : DirectiveInsnRIL<(outs), (ins imm64zx48:$enc, AnyReg:$R1,
1982                                              brtarget32:$I2),
1983                                 ".insn ril,$enc,$R1,$I2", []>;
1984  def InsnRILU : DirectiveInsnRIL<(outs), (ins imm64zx48:$enc, AnyReg:$R1,
1985                                               uimm32:$I2),
1986                                  ".insn rilu,$enc,$R1,$I2", []>;
1987  def InsnRIS : DirectiveInsnRIS<(outs),
1988                                 (ins imm64zx48:$enc, AnyReg:$R1,
1989                                      imm32sx8:$I2, imm32zx4:$M3,
1990                                      bdaddr12only:$BD4),
1991                                 ".insn ris,$enc,$R1,$I2,$M3,$BD4", []>;
1992  def InsnRR : DirectiveInsnRR<(outs),
1993                               (ins imm64zx16:$enc, AnyReg:$R1, AnyReg:$R2),
1994                               ".insn rr,$enc,$R1,$R2", []>;
1995  def InsnRRE : DirectiveInsnRRE<(outs), (ins imm64zx32:$enc,
1996                                              AnyReg:$R1, AnyReg:$R2),
1997                                 ".insn rre,$enc,$R1,$R2", []>;
1998  def InsnRRF : DirectiveInsnRRF<(outs),
1999                                 (ins imm64zx32:$enc, AnyReg:$R1, AnyReg:$R2,
2000                                      AnyReg:$R3, imm32zx4:$M4),
2001                                 ".insn rrf,$enc,$R1,$R2,$R3,$M4", []>;
2002  def InsnRRS : DirectiveInsnRRS<(outs),
2003                                 (ins imm64zx48:$enc, AnyReg:$R1,
2004                                      AnyReg:$R2, imm32zx4:$M3,
2005                                      bdaddr12only:$BD4),
2006                                 ".insn rrs,$enc,$R1,$R2,$M3,$BD4", []>;
2007  def InsnRS  : DirectiveInsnRS<(outs),
2008                                (ins imm64zx32:$enc, AnyReg:$R1,
2009                                     AnyReg:$R3, bdaddr12only:$BD2),
2010                                ".insn rs,$enc,$R1,$R3,$BD2", []>;
2011  def InsnRSE : DirectiveInsnRSE<(outs),
2012                                 (ins imm64zx48:$enc, AnyReg:$R1,
2013                                      AnyReg:$R3, bdaddr12only:$BD2),
2014                                 ".insn rse,$enc,$R1,$R3,$BD2", []>;
2015  def InsnRSI : DirectiveInsnRSI<(outs),
2016                                 (ins imm64zx48:$enc, AnyReg:$R1,
2017                                      AnyReg:$R3, brtarget16:$RI2),
2018                                 ".insn rsi,$enc,$R1,$R3,$RI2", []>;
2019  def InsnRSY : DirectiveInsnRSY<(outs),
2020                                 (ins imm64zx48:$enc, AnyReg:$R1,
2021                                      AnyReg:$R3, bdaddr20only:$BD2),
2022                                 ".insn rsy,$enc,$R1,$R3,$BD2", []>;
2023  def InsnRX  : DirectiveInsnRX<(outs), (ins imm64zx32:$enc, AnyReg:$R1,
2024                                             bdxaddr12only:$XBD2),
2025                                ".insn rx,$enc,$R1,$XBD2", []>;
2026  def InsnRXE : DirectiveInsnRXE<(outs), (ins imm64zx48:$enc, AnyReg:$R1,
2027                                              bdxaddr12only:$XBD2),
2028                                 ".insn rxe,$enc,$R1,$XBD2", []>;
2029  def InsnRXF : DirectiveInsnRXF<(outs),
2030                                 (ins imm64zx48:$enc, AnyReg:$R1,
2031                                      AnyReg:$R3, bdxaddr12only:$XBD2),
2032                                 ".insn rxf,$enc,$R1,$R3,$XBD2", []>;
2033  def InsnRXY : DirectiveInsnRXY<(outs), (ins imm64zx48:$enc, AnyReg:$R1,
2034                                              bdxaddr20only:$XBD2),
2035                                 ".insn rxy,$enc,$R1,$XBD2", []>;
2036  def InsnS : DirectiveInsnS<(outs),
2037                             (ins imm64zx32:$enc, bdaddr12only:$BD2),
2038                             ".insn s,$enc,$BD2", []>;
2039  def InsnSI : DirectiveInsnSI<(outs),
2040                               (ins imm64zx32:$enc, bdaddr12only:$BD1,
2041                                    imm32sx8:$I2),
2042                               ".insn si,$enc,$BD1,$I2", []>;
2043  def InsnSIY : DirectiveInsnSIY<(outs),
2044                                 (ins imm64zx48:$enc,
2045                                      bdaddr20only:$BD1, imm32zx8:$I2),
2046                                 ".insn siy,$enc,$BD1,$I2", []>;
2047  def InsnSIL : DirectiveInsnSIL<(outs),
2048                                 (ins imm64zx48:$enc, bdaddr12only:$BD1,
2049                                      imm32zx16:$I2),
2050                                 ".insn sil,$enc,$BD1,$I2", []>;
2051  def InsnSS : DirectiveInsnSS<(outs),
2052                               (ins imm64zx48:$enc, bdraddr12only:$RBD1,
2053                                    bdaddr12only:$BD2, AnyReg:$R3),
2054                               ".insn ss,$enc,$RBD1,$BD2,$R3", []>;
2055  def InsnSSE : DirectiveInsnSSE<(outs),
2056                                 (ins imm64zx48:$enc,
2057                                      bdaddr12only:$BD1,bdaddr12only:$BD2),
2058                                 ".insn sse,$enc,$BD1,$BD2", []>;
2059  def InsnSSF : DirectiveInsnSSF<(outs),
2060                                 (ins imm64zx48:$enc, bdaddr12only:$BD1,
2061                                      bdaddr12only:$BD2, AnyReg:$R3),
2062                                 ".insn ssf,$enc,$BD1,$BD2,$R3", []>;
2063}
2064
2065//===----------------------------------------------------------------------===//
2066// Peepholes.
2067//===----------------------------------------------------------------------===//
2068
2069// Use AL* for GR64 additions of unsigned 32-bit values.
2070defm : ZXB<add, GR64, ALGFR>;
2071def  : Pat<(add GR64:$src1, imm64zx32:$src2),
2072           (ALGFI GR64:$src1, imm64zx32:$src2)>;
2073def  : Pat<(add GR64:$src1, (azextloadi32 bdxaddr20only:$addr)),
2074           (ALGF GR64:$src1, bdxaddr20only:$addr)>;
2075
2076// Use SL* for GR64 subtractions of unsigned 32-bit values.
2077defm : ZXB<sub, GR64, SLGFR>;
2078def  : Pat<(add GR64:$src1, imm64zx32n:$src2),
2079           (SLGFI GR64:$src1, imm64zx32n:$src2)>;
2080def  : Pat<(sub GR64:$src1, (azextloadi32 bdxaddr20only:$addr)),
2081           (SLGF GR64:$src1, bdxaddr20only:$addr)>;
2082
2083// Optimize sign-extended 1/0 selects to -1/0 selects.  This is important
2084// for vector legalization.
2085def : Pat<(sra (shl (i32 (z_select_ccmask 1, 0, imm32zx4:$valid, imm32zx4:$cc)),
2086                         (i32 31)),
2087                    (i32 31)),
2088          (Select32 (LHI -1), (LHI 0), imm32zx4:$valid, imm32zx4:$cc)>;
2089def : Pat<(sra (shl (i64 (anyext (i32 (z_select_ccmask 1, 0, imm32zx4:$valid,
2090                                                       imm32zx4:$cc)))),
2091                    (i32 63)),
2092               (i32 63)),
2093          (Select64 (LGHI -1), (LGHI 0), imm32zx4:$valid, imm32zx4:$cc)>;
2094
2095// Avoid generating 2 XOR instructions. (xor (and x, y), y) is
2096// equivalent to (and (xor x, -1), y)
2097def : Pat<(and (xor GR64:$x, (i64 -1)), GR64:$y),
2098                          (XGR GR64:$y, (NGR GR64:$y, GR64:$x))>;
2099
2100// Shift/rotate instructions only use the last 6 bits of the second operand
2101// register, so we can safely use NILL (16 fewer bits than NILF) to only AND the
2102// last 16 bits.
2103// Complexity is added so that we match this before we match NILF on the AND
2104// operation alone.
2105let AddedComplexity = 4 in {
2106  def : Pat<(shl GR32:$val, (and GR32:$shift, uimm32:$imm)),
2107            (SLL GR32:$val, (NILL GR32:$shift, uimm32:$imm), 0)>;
2108
2109  def : Pat<(sra GR32:$val, (and GR32:$shift, uimm32:$imm)),
2110            (SRA GR32:$val, (NILL GR32:$shift, uimm32:$imm), 0)>;
2111
2112  def : Pat<(srl GR32:$val, (and GR32:$shift, uimm32:$imm)),
2113            (SRL GR32:$val, (NILL GR32:$shift, uimm32:$imm), 0)>;
2114
2115  def : Pat<(shl GR64:$val, (and GR32:$shift, uimm32:$imm)),
2116            (SLLG GR64:$val, (NILL GR32:$shift, uimm32:$imm), 0)>;
2117
2118  def : Pat<(sra GR64:$val, (and GR32:$shift, uimm32:$imm)),
2119            (SRAG GR64:$val, (NILL GR32:$shift, uimm32:$imm), 0)>;
2120
2121  def : Pat<(srl GR64:$val, (and GR32:$shift, uimm32:$imm)),
2122            (SRLG GR64:$val, (NILL GR32:$shift, uimm32:$imm), 0)>;
2123
2124  def : Pat<(rotl GR32:$val, (and GR32:$shift, uimm32:$imm)),
2125            (RLL GR32:$val, (NILL GR32:$shift, uimm32:$imm), 0)>;
2126
2127  def : Pat<(rotl GR64:$val, (and GR32:$shift, uimm32:$imm)),
2128            (RLLG GR64:$val, (NILL GR32:$shift, uimm32:$imm), 0)>;
2129}
2130
2131// Peepholes for turning scalar operations into block operations.
2132defm : BlockLoadStore<anyextloadi8, i32, MVCSequence, NCSequence, OCSequence,
2133                      XCSequence, 1>;
2134defm : BlockLoadStore<anyextloadi16, i32, MVCSequence, NCSequence, OCSequence,
2135                      XCSequence, 2>;
2136defm : BlockLoadStore<load, i32, MVCSequence, NCSequence, OCSequence,
2137                      XCSequence, 4>;
2138defm : BlockLoadStore<anyextloadi8, i64, MVCSequence, NCSequence,
2139                      OCSequence, XCSequence, 1>;
2140defm : BlockLoadStore<anyextloadi16, i64, MVCSequence, NCSequence, OCSequence,
2141                      XCSequence, 2>;
2142defm : BlockLoadStore<anyextloadi32, i64, MVCSequence, NCSequence, OCSequence,
2143                      XCSequence, 4>;
2144defm : BlockLoadStore<load, i64, MVCSequence, NCSequence, OCSequence,
2145                      XCSequence, 8>;
2146