1 //===- ARMRegisterInfo.td - ARM Register defs -------------------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 //===----------------------------------------------------------------------===//
11 // Declarations that describe the ARM register file
12 //===----------------------------------------------------------------------===//
14 // Registers are identified with 4-bit ID numbers.
15 class ARMReg<bits<4> num, string n, list<Register> subregs = []> : Register<n> {
17 let Namespace = "ARM";
18 let SubRegs = subregs;
21 class ARMFReg<bits<6> num, string n> : Register<n> {
23 let Namespace = "ARM";
26 // Subregister indices.
27 let Namespace = "ARM" in {
28 // Note: Code depends on these having consecutive numbers.
29 def ssub_0 : SubRegIndex { let NumberHack = 1; }
30 def ssub_1 : SubRegIndex { let NumberHack = 2; }
31 def ssub_2 : SubRegIndex { let NumberHack = 3; }
32 def ssub_3 : SubRegIndex { let NumberHack = 4; }
34 def dsub_0 : SubRegIndex { let NumberHack = 5; }
35 def dsub_1 : SubRegIndex { let NumberHack = 6; }
36 def dsub_2 : SubRegIndex { let NumberHack = 7; }
37 def dsub_3 : SubRegIndex { let NumberHack = 8; }
38 def dsub_4 : SubRegIndex { let NumberHack = 9; }
39 def dsub_5 : SubRegIndex { let NumberHack = 10; }
40 def dsub_6 : SubRegIndex { let NumberHack = 11; }
41 def dsub_7 : SubRegIndex { let NumberHack = 12; }
43 def qsub_0 : SubRegIndex { let NumberHack = 13; }
44 def qsub_1 : SubRegIndex { let NumberHack = 14; }
45 def qsub_2 : SubRegIndex { let NumberHack = 15; }
46 def qsub_3 : SubRegIndex { let NumberHack = 16; }
48 def qqsub_0 : SubRegIndex { let NumberHack = 17; }
49 def qqsub_1 : SubRegIndex { let NumberHack = 18; }
53 def R0 : ARMReg< 0, "r0">, DwarfRegNum<[0]>;
54 def R1 : ARMReg< 1, "r1">, DwarfRegNum<[1]>;
55 def R2 : ARMReg< 2, "r2">, DwarfRegNum<[2]>;
56 def R3 : ARMReg< 3, "r3">, DwarfRegNum<[3]>;
57 def R4 : ARMReg< 4, "r4">, DwarfRegNum<[4]>;
58 def R5 : ARMReg< 5, "r5">, DwarfRegNum<[5]>;
59 def R6 : ARMReg< 6, "r6">, DwarfRegNum<[6]>;
60 def R7 : ARMReg< 7, "r7">, DwarfRegNum<[7]>;
61 def R8 : ARMReg< 8, "r8">, DwarfRegNum<[8]>;
62 def R9 : ARMReg< 9, "r9">, DwarfRegNum<[9]>;
63 def R10 : ARMReg<10, "r10">, DwarfRegNum<[10]>;
64 def R11 : ARMReg<11, "r11">, DwarfRegNum<[11]>;
65 def R12 : ARMReg<12, "r12">, DwarfRegNum<[12]>;
66 def SP : ARMReg<13, "sp">, DwarfRegNum<[13]>;
67 def LR : ARMReg<14, "lr">, DwarfRegNum<[14]>;
68 def PC : ARMReg<15, "pc">, DwarfRegNum<[15]>;
71 def S0 : ARMFReg< 0, "s0">; def S1 : ARMFReg< 1, "s1">;
72 def S2 : ARMFReg< 2, "s2">; def S3 : ARMFReg< 3, "s3">;
73 def S4 : ARMFReg< 4, "s4">; def S5 : ARMFReg< 5, "s5">;
74 def S6 : ARMFReg< 6, "s6">; def S7 : ARMFReg< 7, "s7">;
75 def S8 : ARMFReg< 8, "s8">; def S9 : ARMFReg< 9, "s9">;
76 def S10 : ARMFReg<10, "s10">; def S11 : ARMFReg<11, "s11">;
77 def S12 : ARMFReg<12, "s12">; def S13 : ARMFReg<13, "s13">;
78 def S14 : ARMFReg<14, "s14">; def S15 : ARMFReg<15, "s15">;
79 def S16 : ARMFReg<16, "s16">; def S17 : ARMFReg<17, "s17">;
80 def S18 : ARMFReg<18, "s18">; def S19 : ARMFReg<19, "s19">;
81 def S20 : ARMFReg<20, "s20">; def S21 : ARMFReg<21, "s21">;
82 def S22 : ARMFReg<22, "s22">; def S23 : ARMFReg<23, "s23">;
83 def S24 : ARMFReg<24, "s24">; def S25 : ARMFReg<25, "s25">;
84 def S26 : ARMFReg<26, "s26">; def S27 : ARMFReg<27, "s27">;
85 def S28 : ARMFReg<28, "s28">; def S29 : ARMFReg<29, "s29">;
86 def S30 : ARMFReg<30, "s30">; def S31 : ARMFReg<31, "s31">;
87 def SDummy : ARMFReg<63, "sINVALID">;
89 // Aliases of the F* registers used to hold 64-bit fp values (doubles)
90 def D0 : ARMReg< 0, "d0", [S0, S1]>;
91 def D1 : ARMReg< 1, "d1", [S2, S3]>;
92 def D2 : ARMReg< 2, "d2", [S4, S5]>;
93 def D3 : ARMReg< 3, "d3", [S6, S7]>;
94 def D4 : ARMReg< 4, "d4", [S8, S9]>;
95 def D5 : ARMReg< 5, "d5", [S10, S11]>;
96 def D6 : ARMReg< 6, "d6", [S12, S13]>;
97 def D7 : ARMReg< 7, "d7", [S14, S15]>;
98 def D8 : ARMReg< 8, "d8", [S16, S17]>;
99 def D9 : ARMReg< 9, "d9", [S18, S19]>;
100 def D10 : ARMReg<10, "d10", [S20, S21]>;
101 def D11 : ARMReg<11, "d11", [S22, S23]>;
102 def D12 : ARMReg<12, "d12", [S24, S25]>;
103 def D13 : ARMReg<13, "d13", [S26, S27]>;
104 def D14 : ARMReg<14, "d14", [S28, S29]>;
105 def D15 : ARMReg<15, "d15", [S30, S31]>;
107 // VFP3 defines 16 additional double registers
108 def D16 : ARMFReg<16, "d16">; def D17 : ARMFReg<17, "d17">;
109 def D18 : ARMFReg<18, "d18">; def D19 : ARMFReg<19, "d19">;
110 def D20 : ARMFReg<20, "d20">; def D21 : ARMFReg<21, "d21">;
111 def D22 : ARMFReg<22, "d22">; def D23 : ARMFReg<23, "d23">;
112 def D24 : ARMFReg<24, "d24">; def D25 : ARMFReg<25, "d25">;
113 def D26 : ARMFReg<26, "d26">; def D27 : ARMFReg<27, "d27">;
114 def D28 : ARMFReg<28, "d28">; def D29 : ARMFReg<29, "d29">;
115 def D30 : ARMFReg<30, "d30">; def D31 : ARMFReg<31, "d31">;
116 def DDummy : ARMFReg<31, "dINVALID">;
118 // Advanced SIMD (NEON) defines 16 quad-word aliases
119 def Q0 : ARMReg< 0, "q0", [D0, D1]>;
120 def Q1 : ARMReg< 1, "q1", [D2, D3]>;
121 def Q2 : ARMReg< 2, "q2", [D4, D5]>;
122 def Q3 : ARMReg< 3, "q3", [D6, D7]>;
123 def Q4 : ARMReg< 4, "q4", [D8, D9]>;
124 def Q5 : ARMReg< 5, "q5", [D10, D11]>;
125 def Q6 : ARMReg< 6, "q6", [D12, D13]>;
126 def Q7 : ARMReg< 7, "q7", [D14, D15]>;
127 def Q8 : ARMReg< 8, "q8", [D16, D17]>;
128 def Q9 : ARMReg< 9, "q9", [D18, D19]>;
129 def Q10 : ARMReg<10, "q10", [D20, D21]>;
130 def Q11 : ARMReg<11, "q11", [D22, D23]>;
131 def Q12 : ARMReg<12, "q12", [D24, D25]>;
132 def Q13 : ARMReg<13, "q13", [D26, D27]>;
133 def Q14 : ARMReg<14, "q14", [D28, D29]>;
134 def Q15 : ARMReg<15, "q15", [D30, D31]>;
135 def QDummy : ARMFReg<16, "qINVALID">;
137 // Pseudo 256-bit registers to represent pairs of Q registers. These should
138 // never be present in the emitted code.
139 // These are used for NEON load / store instructions, e.g. vld4, vst3.
140 // NOTE: It's possible to define more QQ registers since technical the
141 // starting D register number doesn't have to be multiple of 4. e.g.
142 // D1, D2, D3, D4 would be a legal quad. But that would make the sub-register
143 // stuffs very messy.
144 def QQ0 : ARMReg<0, "qq0", [Q0, Q1]>;
145 def QQ1 : ARMReg<1, "qq1", [Q2, Q3]>;
146 def QQ2 : ARMReg<2, "qq2", [Q4, Q5]>;
147 def QQ3 : ARMReg<3, "qq3", [Q6, Q7]>;
148 def QQ4 : ARMReg<4, "qq4", [Q8, Q9]>;
149 def QQ5 : ARMReg<5, "qq5", [Q10, Q11]>;
150 def QQ6 : ARMReg<6, "qq6", [Q12, Q13]>;
151 def QQ7 : ARMReg<7, "qq7", [Q14, Q15]>;
153 // Pseudo 512-bit registers to represent four consecutive Q registers.
154 def QQQQ0 : ARMReg<0, "qqqq0", [QQ0, QQ1]>;
155 def QQQQ1 : ARMReg<1, "qqqq1", [QQ2, QQ3]>;
156 def QQQQ2 : ARMReg<2, "qqqq2", [QQ4, QQ5]>;
157 def QQQQ3 : ARMReg<3, "qqqq3", [QQ6, QQ7]>;
159 // Current Program Status Register.
160 def CPSR : ARMReg<0, "cpsr">;
162 def FPSCR : ARMReg<1, "fpscr">;
166 // pc == Program Counter
167 // lr == Link Register
168 // sp == Stack Pointer
169 // r12 == ip (scratch)
170 // r7 == Frame Pointer (thumb-style backtraces)
171 // r9 == May be reserved as Thread Register
172 // r11 == Frame Pointer (arm-style backtraces)
173 // r10 == Stack Limit
175 def GPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6,
176 R7, R8, R9, R10, R11, R12,
178 let MethodProtos = [{
179 iterator allocation_order_begin(const MachineFunction &MF) const;
180 iterator allocation_order_end(const MachineFunction &MF) const;
182 let MethodBodies = [{
183 // FP is R11, R9 is available.
184 static const unsigned ARM_GPR_AO_1[] = {
185 ARM::R0, ARM::R1, ARM::R2, ARM::R3,
187 ARM::R4, ARM::R5, ARM::R6, ARM::R7,
188 ARM::R8, ARM::R9, ARM::R10,
190 // FP is R11, R9 is not available.
191 static const unsigned ARM_GPR_AO_2[] = {
192 ARM::R0, ARM::R1, ARM::R2, ARM::R3,
194 ARM::R4, ARM::R5, ARM::R6, ARM::R7,
197 // FP is R7, R9 is available as non-callee-saved register.
198 // This is used by Darwin.
199 static const unsigned ARM_GPR_AO_3[] = {
200 ARM::R0, ARM::R1, ARM::R2, ARM::R3,
201 ARM::R9, ARM::R12,ARM::LR,
202 ARM::R4, ARM::R5, ARM::R6,
203 ARM::R8, ARM::R10,ARM::R11,ARM::R7 };
204 // FP is R7, R9 is not available.
205 static const unsigned ARM_GPR_AO_4[] = {
206 ARM::R0, ARM::R1, ARM::R2, ARM::R3,
208 ARM::R4, ARM::R5, ARM::R6,
209 ARM::R8, ARM::R10,ARM::R11,
211 // FP is R7, R9 is available as callee-saved register.
212 // This is used by non-Darwin platform in Thumb mode.
213 static const unsigned ARM_GPR_AO_5[] = {
214 ARM::R0, ARM::R1, ARM::R2, ARM::R3,
216 ARM::R4, ARM::R5, ARM::R6,
217 ARM::R8, ARM::R9, ARM::R10,ARM::R11,ARM::R7 };
219 // For Thumb1 mode, we don't want to allocate hi regs at all, as we
220 // don't know how to spill them. If we make our prologue/epilogue code
221 // smarter at some point, we can go back to using the above allocation
222 // orders for the Thumb1 instructions that know how to use hi regs.
223 static const unsigned THUMB_GPR_AO[] = {
224 ARM::R0, ARM::R1, ARM::R2, ARM::R3,
225 ARM::R4, ARM::R5, ARM::R6, ARM::R7 };
228 GPRClass::allocation_order_begin(const MachineFunction &MF) const {
229 const TargetMachine &TM = MF.getTarget();
230 const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
231 if (Subtarget.isThumb1Only())
233 if (Subtarget.isTargetDarwin()) {
234 if (Subtarget.isR9Reserved())
239 if (Subtarget.isR9Reserved())
241 else if (Subtarget.isThumb())
249 GPRClass::allocation_order_end(const MachineFunction &MF) const {
250 const TargetMachine &TM = MF.getTarget();
251 const TargetRegisterInfo *RI = TM.getRegisterInfo();
252 const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
253 GPRClass::iterator I;
255 if (Subtarget.isThumb1Only()) {
256 I = THUMB_GPR_AO + (sizeof(THUMB_GPR_AO)/sizeof(unsigned));
257 // Mac OS X requires FP not to be clobbered for backtracing purpose.
258 return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
261 if (Subtarget.isTargetDarwin()) {
262 if (Subtarget.isR9Reserved())
263 I = ARM_GPR_AO_4 + (sizeof(ARM_GPR_AO_4)/sizeof(unsigned));
265 I = ARM_GPR_AO_3 + (sizeof(ARM_GPR_AO_3)/sizeof(unsigned));
267 if (Subtarget.isR9Reserved())
268 I = ARM_GPR_AO_2 + (sizeof(ARM_GPR_AO_2)/sizeof(unsigned));
269 else if (Subtarget.isThumb())
270 I = ARM_GPR_AO_5 + (sizeof(ARM_GPR_AO_5)/sizeof(unsigned));
272 I = ARM_GPR_AO_1 + (sizeof(ARM_GPR_AO_1)/sizeof(unsigned));
275 // Mac OS X requires FP not to be clobbered for backtracing purpose.
276 return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
281 // Thumb registers are R0-R7 normally. Some instructions can still use
282 // the general GPR register class above (MOV, e.g.)
283 def tGPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6, R7]> {
284 let MethodProtos = [{
285 iterator allocation_order_begin(const MachineFunction &MF) const;
286 iterator allocation_order_end(const MachineFunction &MF) const;
288 let MethodBodies = [{
289 static const unsigned THUMB_tGPR_AO[] = {
290 ARM::R0, ARM::R1, ARM::R2, ARM::R3,
291 ARM::R4, ARM::R5, ARM::R6, ARM::R7 };
293 // FP is R7, only low registers available.
295 tGPRClass::allocation_order_begin(const MachineFunction &MF) const {
296 return THUMB_tGPR_AO;
300 tGPRClass::allocation_order_end(const MachineFunction &MF) const {
301 const TargetMachine &TM = MF.getTarget();
302 const TargetRegisterInfo *RI = TM.getRegisterInfo();
303 const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
304 tGPRClass::iterator I =
305 THUMB_tGPR_AO + (sizeof(THUMB_tGPR_AO)/sizeof(unsigned));
306 // Mac OS X requires FP not to be clobbered for backtracing purpose.
307 return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
312 // Scalar single precision floating point register class..
313 def SPR : RegisterClass<"ARM", [f32], 32, [S0, S1, S2, S3, S4, S5, S6, S7, S8,
314 S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, S19, S20, S21, S22,
315 S23, S24, S25, S26, S27, S28, S29, S30, S31]>;
317 // Subset of SPR which can be used as a source of NEON scalars for 16-bit
319 def SPR_8 : RegisterClass<"ARM", [f32], 32,
320 [S0, S1, S2, S3, S4, S5, S6, S7,
321 S8, S9, S10, S11, S12, S13, S14, S15]>;
323 // Dummy f32 regclass to represent impossible subreg indices.
324 def SPR_INVALID : RegisterClass<"ARM", [f32], 32, [SDummy]> {
328 // Scalar double precision floating point / generic 64-bit vector register
330 // ARM requires only word alignment for double. It's more performant if it
331 // is double-word alignment though.
332 def DPR : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64,
333 [D0, D1, D2, D3, D4, D5, D6, D7,
334 D8, D9, D10, D11, D12, D13, D14, D15,
335 D16, D17, D18, D19, D20, D21, D22, D23,
336 D24, D25, D26, D27, D28, D29, D30, D31]> {
337 let MethodProtos = [{
338 iterator allocation_order_begin(const MachineFunction &MF) const;
339 iterator allocation_order_end(const MachineFunction &MF) const;
341 let MethodBodies = [{
343 static const unsigned ARM_DPR_VFP2[] = {
344 ARM::D0, ARM::D1, ARM::D2, ARM::D3,
345 ARM::D4, ARM::D5, ARM::D6, ARM::D7,
346 ARM::D8, ARM::D9, ARM::D10, ARM::D11,
347 ARM::D12, ARM::D13, ARM::D14, ARM::D15 };
349 static const unsigned ARM_DPR_VFP3[] = {
350 ARM::D0, ARM::D1, ARM::D2, ARM::D3,
351 ARM::D4, ARM::D5, ARM::D6, ARM::D7,
352 ARM::D8, ARM::D9, ARM::D10, ARM::D11,
353 ARM::D12, ARM::D13, ARM::D14, ARM::D15,
354 ARM::D16, ARM::D17, ARM::D18, ARM::D19,
355 ARM::D20, ARM::D21, ARM::D22, ARM::D23,
356 ARM::D24, ARM::D25, ARM::D26, ARM::D27,
357 ARM::D28, ARM::D29, ARM::D30, ARM::D31 };
359 DPRClass::allocation_order_begin(const MachineFunction &MF) const {
360 const TargetMachine &TM = MF.getTarget();
361 const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
362 if (Subtarget.hasVFP3())
368 DPRClass::allocation_order_end(const MachineFunction &MF) const {
369 const TargetMachine &TM = MF.getTarget();
370 const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
371 if (Subtarget.hasVFP3())
372 return ARM_DPR_VFP3 + (sizeof(ARM_DPR_VFP3)/sizeof(unsigned));
374 return ARM_DPR_VFP2 + (sizeof(ARM_DPR_VFP2)/sizeof(unsigned));
379 // Subset of DPR that are accessible with VFP2 (and so that also have
380 // 32-bit SPR subregs).
381 def DPR_VFP2 : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64,
382 [D0, D1, D2, D3, D4, D5, D6, D7,
383 D8, D9, D10, D11, D12, D13, D14, D15]> {
384 let SubRegClasses = [(SPR ssub_0, ssub_1)];
387 // Subset of DPR which can be used as a source of NEON scalars for 16-bit
389 def DPR_8 : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64,
390 [D0, D1, D2, D3, D4, D5, D6, D7]> {
391 let SubRegClasses = [(SPR_8 ssub_0, ssub_1)];
394 // Dummy 64-bit regclass to represent impossible subreg indices.
395 def DPR_INVALID : RegisterClass<"ARM",
396 [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64,
401 // Generic 128-bit vector register class.
402 def QPR : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], 128,
403 [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7,
404 Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15]> {
405 let SubRegClasses = [(DPR dsub_0, dsub_1)];
408 // Subset of QPR that have 32-bit SPR subregs.
409 def QPR_VFP2 : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
411 [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]> {
412 let SubRegClasses = [(SPR ssub_0, ssub_1, ssub_2, ssub_3),
413 (DPR_VFP2 dsub_0, dsub_1)];
416 // Subset of QPR that have DPR_8 and SPR_8 subregs.
417 def QPR_8 : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
420 let SubRegClasses = [(SPR_8 ssub_0, ssub_1, ssub_2, ssub_3),
421 (DPR_8 dsub_0, dsub_1)];
424 // Dummy 128-bit regclass to represent impossible subreg indices.
425 def QPR_INVALID : RegisterClass<"ARM",
426 [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], 128,
431 // Pseudo 256-bit vector register class to model pairs of Q registers
432 // (4 consecutive D registers).
433 def QQPR : RegisterClass<"ARM", [v4i64],
435 [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7]> {
436 let SubRegClasses = [(DPR dsub_0, dsub_1, dsub_2, dsub_3),
437 (QPR qsub_0, qsub_1)];
440 // Subset of QQPR that have 32-bit SPR subregs.
441 def QQPR_VFP2 : RegisterClass<"ARM", [v4i64],
443 [QQ0, QQ1, QQ2, QQ3]> {
444 let SubRegClasses = [(SPR ssub_0, ssub_1, ssub_2, ssub_3),
445 (DPR_VFP2 dsub_0, dsub_1, dsub_2, dsub_3),
446 (QPR_VFP2 qsub_0, qsub_1)];
450 // Pseudo 512-bit vector register class to model 4 consecutive Q registers
451 // (8 consecutive D registers).
452 def QQQQPR : RegisterClass<"ARM", [v8i64],
454 [QQQQ0, QQQQ1, QQQQ2, QQQQ3]> {
455 let SubRegClasses = [(DPR dsub_0, dsub_1, dsub_2, dsub_3,
456 dsub_4, dsub_5, dsub_6, dsub_7),
457 (QPR qsub_0, qsub_1, qsub_2, qsub_3)];
460 // Condition code registers.
461 def CCR : RegisterClass<"ARM", [i32], 32, [CPSR]>;
463 //===----------------------------------------------------------------------===//
464 // Subregister Set Definitions... now that we have all of the pieces, define the
465 // sub registers for each register.
468 // S sub-registers of D registers.
469 def : SubRegSet<1, [D0, D1, D2, D3, D4, D5, D6, D7,
470 D8, D9, D10, D11, D12, D13, D14, D15],
471 [S0, S2, S4, S6, S8, S10, S12, S14,
472 S16, S18, S20, S22, S24, S26, S28, S30]>;
473 def : SubRegSet<2, [D0, D1, D2, D3, D4, D5, D6, D7,
474 D8, D9, D10, D11, D12, D13, D14, D15],
475 [S1, S3, S5, S7, S9, S11, S13, S15,
476 S17, S19, S21, S23, S25, S27, S29, S31]>;
478 // S sub-registers of Q registers.
479 def : SubRegSet<1, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
480 [S0, S4, S8, S12, S16, S20, S24, S28]>;
481 def : SubRegSet<2, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
482 [S1, S5, S9, S13, S17, S21, S25, S29]>;
483 def : SubRegSet<3, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
484 [S2, S6, S10, S14, S18, S22, S26, S30]>;
485 def : SubRegSet<4, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
486 [S3, S7, S11, S15, S19, S23, S27, S31]>;
488 // D sub-registers of Q registers.
489 def : SubRegSet<5, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7,
490 Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15],
491 [D0, D2, D4, D6, D8, D10, D12, D14,
492 D16, D18, D20, D22, D24, D26, D28, D30]>;
493 def : SubRegSet<6, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7,
494 Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15],
495 [D1, D3, D5, D7, D9, D11, D13, D15,
496 D17, D19, D21, D23, D25, D27, D29, D31]>;
498 // S sub-registers of QQ registers. Note there are no sub-indices
499 // for referencing S4 - S7, S12 - S15, and S20 - S23. It doesn't
500 // look like we need them.
501 def : SubRegSet<1, [QQ0, QQ1, QQ2, QQ3],
503 def : SubRegSet<2, [QQ0, QQ1, QQ2, QQ3],
505 def : SubRegSet<3, [QQ0, QQ1, QQ2, QQ3],
506 [S2, S10, S18, S26]>;
507 def : SubRegSet<4, [QQ0, QQ1, QQ2, QQ3],
508 [S3, S11, S19, S27]>;
510 // D sub-registers of QQ registers.
511 def : SubRegSet<5, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
512 [D0, D4, D8, D12, D16, D20, D24, D28]>;
513 def : SubRegSet<6, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
514 [D1, D5, D9, D13, D17, D21, D25, D29]>;
515 def : SubRegSet<7, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
516 [D2, D6, D10, D14, D18, D22, D26, D30]>;
517 def : SubRegSet<8, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
518 [D3, D7, D11, D15, D19, D23, D27, D31]>;
520 // Q sub-registers of QQ registers.
521 def : SubRegSet<13, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
522 [Q0, Q2, Q4, Q6, Q8, Q10, Q12, Q14]>;
523 def : SubRegSet<14,[QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
524 [Q1, Q3, Q5, Q7, Q9, Q11, Q13, Q15]>;
527 // D sub-registers of QQQQ registers.
528 def : SubRegSet<5, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
530 def : SubRegSet<6, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
532 def : SubRegSet<7, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
533 [D2, D10, D18, D26]>;
534 def : SubRegSet<8, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
535 [D3, D11, D19, D27]>;
537 def : SubRegSet<9, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
538 [D4, D12, D20, D28]>;
539 def : SubRegSet<10, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
540 [D5, D13, D21, D29]>;
541 def : SubRegSet<11, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
542 [D6, D14, D22, D30]>;
543 def : SubRegSet<12, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
544 [D7, D15, D23, D31]>;
546 // Q sub-registers of QQQQQQQQ registers.
547 def : SubRegSet<13, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
549 def : SubRegSet<14, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
551 def : SubRegSet<15, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
553 def : SubRegSet<16, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
556 // QQ sub-registers of QQQQQQQQ registers.
557 def : SubRegSet<17, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
558 [QQ0, QQ2, QQ4, QQ6]>;
559 def : SubRegSet<18, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
560 [QQ1, QQ3, QQ5, QQ7]>;