-//===- ARMRegisterInfo.td - ARM Register defs -------------------*- C++ -*-===//
+//===-- ARMRegisterInfo.td - ARM Register defs -------------*- tablegen -*-===//
//
// The LLVM Compiler Infrastructure
//
field bits<4> Num;
let Namespace = "ARM";
let SubRegs = subregs;
+ // All bits of ARM registers with sub-registers are covered by sub-registers.
+ let CoveredBySubRegs = 1;
}
-class ARMFReg<bits<5> num, string n> : Register<n> {
- field bits<5> Num;
+class ARMFReg<bits<6> num, string n> : Register<n> {
+ field bits<6> Num;
let Namespace = "ARM";
}
+// Subregister indices.
+let Namespace = "ARM" in {
+def qqsub_0 : SubRegIndex;
+def qqsub_1 : SubRegIndex;
+
+// Note: Code depends on these having consecutive numbers.
+def qsub_0 : SubRegIndex;
+def qsub_1 : SubRegIndex;
+def qsub_2 : SubRegIndex<[qqsub_1, qsub_0]>;
+def qsub_3 : SubRegIndex<[qqsub_1, qsub_1]>;
+
+def dsub_0 : SubRegIndex;
+def dsub_1 : SubRegIndex;
+def dsub_2 : SubRegIndex<[qsub_1, dsub_0]>;
+def dsub_3 : SubRegIndex<[qsub_1, dsub_1]>;
+def dsub_4 : SubRegIndex<[qsub_2, dsub_0]>;
+def dsub_5 : SubRegIndex<[qsub_2, dsub_1]>;
+def dsub_6 : SubRegIndex<[qsub_3, dsub_0]>;
+def dsub_7 : SubRegIndex<[qsub_3, dsub_1]>;
+
+def ssub_0 : SubRegIndex;
+def ssub_1 : SubRegIndex;
+def ssub_2 : SubRegIndex<[dsub_1, ssub_0]>;
+def ssub_3 : SubRegIndex<[dsub_1, ssub_1]>;
+// Let TableGen synthesize the remaining 12 ssub_* indices.
+// We don't need to name them.
+}
+
// Integer registers
def R0 : ARMReg< 0, "r0">, DwarfRegNum<[0]>;
def R1 : ARMReg< 1, "r1">, DwarfRegNum<[1]>;
def R5 : ARMReg< 5, "r5">, DwarfRegNum<[5]>;
def R6 : ARMReg< 6, "r6">, DwarfRegNum<[6]>;
def R7 : ARMReg< 7, "r7">, DwarfRegNum<[7]>;
+// These require 32-bit instructions.
+let CostPerUse = 1 in {
def R8 : ARMReg< 8, "r8">, DwarfRegNum<[8]>;
def R9 : ARMReg< 9, "r9">, DwarfRegNum<[9]>;
def R10 : ARMReg<10, "r10">, DwarfRegNum<[10]>;
def SP : ARMReg<13, "sp">, DwarfRegNum<[13]>;
def LR : ARMReg<14, "lr">, DwarfRegNum<[14]>;
def PC : ARMReg<15, "pc">, DwarfRegNum<[15]>;
+}
// Float registers
def S0 : ARMFReg< 0, "s0">; def S1 : ARMFReg< 1, "s1">;
def S30 : ARMFReg<30, "s30">; def S31 : ARMFReg<31, "s31">;
// Aliases of the F* registers used to hold 64-bit fp values (doubles)
-def D0 : ARMReg< 0, "d0", [S0, S1]>;
-def D1 : ARMReg< 1, "d1", [S2, S3]>;
-def D2 : ARMReg< 2, "d2", [S4, S5]>;
-def D3 : ARMReg< 3, "d3", [S6, S7]>;
-def D4 : ARMReg< 4, "d4", [S8, S9]>;
-def D5 : ARMReg< 5, "d5", [S10, S11]>;
-def D6 : ARMReg< 6, "d6", [S12, S13]>;
-def D7 : ARMReg< 7, "d7", [S14, S15]>;
-def D8 : ARMReg< 8, "d8", [S16, S17]>;
-def D9 : ARMReg< 9, "d9", [S18, S19]>;
-def D10 : ARMReg<10, "d10", [S20, S21]>;
-def D11 : ARMReg<11, "d11", [S22, S23]>;
-def D12 : ARMReg<12, "d12", [S24, S25]>;
-def D13 : ARMReg<13, "d13", [S26, S27]>;
-def D14 : ARMReg<14, "d14", [S28, S29]>;
-def D15 : ARMReg<15, "d15", [S30, S31]>;
+let SubRegIndices = [ssub_0, ssub_1] in {
+def D0 : ARMReg< 0, "d0", [S0, S1]>, DwarfRegNum<[256]>;
+def D1 : ARMReg< 1, "d1", [S2, S3]>, DwarfRegNum<[257]>;
+def D2 : ARMReg< 2, "d2", [S4, S5]>, DwarfRegNum<[258]>;
+def D3 : ARMReg< 3, "d3", [S6, S7]>, DwarfRegNum<[259]>;
+def D4 : ARMReg< 4, "d4", [S8, S9]>, DwarfRegNum<[260]>;
+def D5 : ARMReg< 5, "d5", [S10, S11]>, DwarfRegNum<[261]>;
+def D6 : ARMReg< 6, "d6", [S12, S13]>, DwarfRegNum<[262]>;
+def D7 : ARMReg< 7, "d7", [S14, S15]>, DwarfRegNum<[263]>;
+def D8 : ARMReg< 8, "d8", [S16, S17]>, DwarfRegNum<[264]>;
+def D9 : ARMReg< 9, "d9", [S18, S19]>, DwarfRegNum<[265]>;
+def D10 : ARMReg<10, "d10", [S20, S21]>, DwarfRegNum<[266]>;
+def D11 : ARMReg<11, "d11", [S22, S23]>, DwarfRegNum<[267]>;
+def D12 : ARMReg<12, "d12", [S24, S25]>, DwarfRegNum<[268]>;
+def D13 : ARMReg<13, "d13", [S26, S27]>, DwarfRegNum<[269]>;
+def D14 : ARMReg<14, "d14", [S28, S29]>, DwarfRegNum<[270]>;
+def D15 : ARMReg<15, "d15", [S30, S31]>, DwarfRegNum<[271]>;
+}
// VFP3 defines 16 additional double registers
-def D16 : ARMFReg<16, "d16">; def D17 : ARMFReg<17, "d17">;
-def D18 : ARMFReg<18, "d18">; def D19 : ARMFReg<19, "d19">;
-def D20 : ARMFReg<20, "d20">; def D21 : ARMFReg<21, "d21">;
-def D22 : ARMFReg<22, "d22">; def D23 : ARMFReg<23, "d23">;
-def D24 : ARMFReg<24, "d24">; def D25 : ARMFReg<25, "d25">;
-def D26 : ARMFReg<26, "d26">; def D27 : ARMFReg<27, "d27">;
-def D28 : ARMFReg<28, "d28">; def D29 : ARMFReg<29, "d29">;
-def D30 : ARMFReg<30, "d30">; def D31 : ARMFReg<31, "d31">;
+def D16 : ARMFReg<16, "d16">, DwarfRegNum<[272]>;
+def D17 : ARMFReg<17, "d17">, DwarfRegNum<[273]>;
+def D18 : ARMFReg<18, "d18">, DwarfRegNum<[274]>;
+def D19 : ARMFReg<19, "d19">, DwarfRegNum<[275]>;
+def D20 : ARMFReg<20, "d20">, DwarfRegNum<[276]>;
+def D21 : ARMFReg<21, "d21">, DwarfRegNum<[277]>;
+def D22 : ARMFReg<22, "d22">, DwarfRegNum<[278]>;
+def D23 : ARMFReg<23, "d23">, DwarfRegNum<[279]>;
+def D24 : ARMFReg<24, "d24">, DwarfRegNum<[280]>;
+def D25 : ARMFReg<25, "d25">, DwarfRegNum<[281]>;
+def D26 : ARMFReg<26, "d26">, DwarfRegNum<[282]>;
+def D27 : ARMFReg<27, "d27">, DwarfRegNum<[283]>;
+def D28 : ARMFReg<28, "d28">, DwarfRegNum<[284]>;
+def D29 : ARMFReg<29, "d29">, DwarfRegNum<[285]>;
+def D30 : ARMFReg<30, "d30">, DwarfRegNum<[286]>;
+def D31 : ARMFReg<31, "d31">, DwarfRegNum<[287]>;
// Advanced SIMD (NEON) defines 16 quad-word aliases
+let SubRegIndices = [dsub_0, dsub_1] in {
def Q0 : ARMReg< 0, "q0", [D0, D1]>;
-def Q1 : ARMReg< 1, "q1", [D2, D3]>;
+def Q1 : ARMReg< 1, "q1", [D2, D3]>;
def Q2 : ARMReg< 2, "q2", [D4, D5]>;
def Q3 : ARMReg< 3, "q3", [D6, D7]>;
def Q4 : ARMReg< 4, "q4", [D8, D9]>;
def Q5 : ARMReg< 5, "q5", [D10, D11]>;
def Q6 : ARMReg< 6, "q6", [D12, D13]>;
def Q7 : ARMReg< 7, "q7", [D14, D15]>;
+}
+let SubRegIndices = [dsub_0, dsub_1] in {
def Q8 : ARMReg< 8, "q8", [D16, D17]>;
def Q9 : ARMReg< 9, "q9", [D18, D19]>;
def Q10 : ARMReg<10, "q10", [D20, D21]>;
def Q13 : ARMReg<13, "q13", [D26, D27]>;
def Q14 : ARMReg<14, "q14", [D28, D29]>;
def Q15 : ARMReg<15, "q15", [D30, D31]>;
-
-// Aliases for superclasses of NEON quad registers.
-def Q2_0: ARMReg<0, "q2_0", [Q0, Q1]>;
-def Q2_1: ARMReg<1, "q2_1", [Q2, Q3]>;
-def Q2_2: ARMReg<2, "q2_2", [Q4, Q5]>;
-def Q2_3: ARMReg<3, "q2_3", [Q6, Q7]>;
-def Q2_4: ARMReg<4, "q2_4", [Q8, Q9]>;
-def Q2_5: ARMReg<5, "q2_5", [Q10, Q11]>;
-def Q2_6: ARMReg<6, "q2_6", [Q12, Q13]>;
-def Q2_7: ARMReg<7, "q2_7", [Q14, Q15]>;
-
-def Q4_0: ARMReg<0, "q4_0", [Q2_0, Q2_1]>;
-def Q4_1: ARMReg<1, "q4_1", [Q2_2, Q2_3]>;
-def Q4_2: ARMReg<2, "q4_2", [Q2_4, Q2_5]>;
-def Q4_3: ARMReg<3, "q4_3", [Q2_6, Q2_7]>;
+}
// Current Program Status Register.
-def CPSR : ARMReg<0, "cpsr">;
+// We model fpscr with two registers: FPSCR models the control bits and will be
+// reserved. FPSCR_NZCV models the flag bits and will be unreserved.
+def CPSR : ARMReg<0, "cpsr">;
+def APSR : ARMReg<1, "apsr">;
+def SPSR : ARMReg<2, "spsr">;
+def FPSCR : ARMReg<3, "fpscr">;
+def FPSCR_NZCV : ARMReg<3, "fpscr_nzcv"> {
+ let Aliases = [FPSCR];
+}
+def ITSTATE : ARMReg<4, "itstate">;
+
+// Special Registers - only available in privileged mode.
+def FPSID : ARMReg<0, "fpsid">;
+def FPEXC : ARMReg<8, "fpexc">;
// Register classes.
//
// r11 == Frame Pointer (arm-style backtraces)
// r10 == Stack Limit
//
-def GPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6,
- R7, R8, R9, R10, R12, R11,
- LR, SP, PC]> {
- let MethodProtos = [{
- iterator allocation_order_begin(const MachineFunction &MF) const;
- iterator allocation_order_end(const MachineFunction &MF) const;
+def GPR : RegisterClass<"ARM", [i32], 32, (add (sequence "R%u", 0, 12),
+ SP, LR, PC)> {
+ // Allocate LR as the first CSR since it is always saved anyway.
+ // For Thumb1 mode, we don't want to allocate hi regs at all, as we don't
+ // know how to spill them. If we make our prologue/epilogue code smarter at
+ // some point, we can go back to using the above allocation orders for the
+ // Thumb1 instructions that know how to use hi regs.
+ let AltOrders = [(add LR, GPR), (trunc GPR, 8)];
+ let AltOrderSelect = [{
+ return 1 + MF.getTarget().getSubtarget<ARMSubtarget>().isThumb1Only();
}];
- // FIXME: We are reserving r12 in case the PEI needs to use it to
- // generate large stack offset. Make it available once we have register
- // scavenging. Similarly r3 is reserved in Thumb mode for now.
- let MethodBodies = [{
- // FP is R11, R9 is available.
- static const unsigned ARM_GPR_AO_1[] = {
- ARM::R0, ARM::R1, ARM::R2, ARM::R3,
- ARM::R12,ARM::LR,
- ARM::R4, ARM::R5, ARM::R6, ARM::R7,
- ARM::R8, ARM::R9, ARM::R10,
- ARM::R11 };
- // FP is R11, R9 is not available.
- static const unsigned ARM_GPR_AO_2[] = {
- ARM::R0, ARM::R1, ARM::R2, ARM::R3,
- ARM::R12,ARM::LR,
- ARM::R4, ARM::R5, ARM::R6, ARM::R7,
- ARM::R8, ARM::R10,
- ARM::R11 };
- // FP is R7, R9 is available as non-callee-saved register.
- // This is used by Darwin.
- static const unsigned ARM_GPR_AO_3[] = {
- ARM::R0, ARM::R1, ARM::R2, ARM::R3,
- ARM::R9, ARM::R12,ARM::LR,
- ARM::R4, ARM::R5, ARM::R6,
- ARM::R8, ARM::R10,ARM::R11,ARM::R7 };
- // FP is R7, R9 is not available.
- static const unsigned ARM_GPR_AO_4[] = {
- ARM::R0, ARM::R1, ARM::R2, ARM::R3,
- ARM::R12,ARM::LR,
- ARM::R4, ARM::R5, ARM::R6,
- ARM::R8, ARM::R10,ARM::R11,
- ARM::R7 };
-
- GPRClass::iterator
- GPRClass::allocation_order_begin(const MachineFunction &MF) const {
- const TargetMachine &TM = MF.getTarget();
- const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
- if (Subtarget.isTargetDarwin()) {
- if (Subtarget.isR9Reserved())
- return ARM_GPR_AO_4;
- else
- return ARM_GPR_AO_3;
- } else {
- if (Subtarget.isR9Reserved())
- return ARM_GPR_AO_2;
- else
- return ARM_GPR_AO_1;
- }
- }
-
- GPRClass::iterator
- GPRClass::allocation_order_end(const MachineFunction &MF) const {
- const TargetMachine &TM = MF.getTarget();
- const TargetRegisterInfo *RI = TM.getRegisterInfo();
- const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
- GPRClass::iterator I;
-
- if (Subtarget.isTargetDarwin()) {
- if (Subtarget.isR9Reserved())
- I = ARM_GPR_AO_4 + (sizeof(ARM_GPR_AO_4)/sizeof(unsigned));
- else
- I = ARM_GPR_AO_3 + (sizeof(ARM_GPR_AO_3)/sizeof(unsigned));
- } else {
- if (Subtarget.isR9Reserved())
- I = ARM_GPR_AO_2 + (sizeof(ARM_GPR_AO_2)/sizeof(unsigned));
- else
- I = ARM_GPR_AO_1 + (sizeof(ARM_GPR_AO_1)/sizeof(unsigned));
- }
-
- // Mac OS X requires FP not to be clobbered for backtracing purpose.
- return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
- }
+}
+
+// GPRs without the PC. Some ARM instructions do not allow the PC in
+// certain operand slots, particularly as the destination. Primarily
+// useful for disassembly.
+def GPRnopc : RegisterClass<"ARM", [i32], 32, (sub GPR, PC)> {
+ let AltOrders = [(add LR, GPRnopc), (trunc GPRnopc, 8)];
+ let AltOrderSelect = [{
+ return 1 + MF.getTarget().getSubtarget<ARMSubtarget>().isThumb1Only();
+ }];
+}
+
+// GPRsp - Only the SP is legal. Used by Thumb1 instructions that want the
+// implied SP argument list.
+// FIXME: It would be better to not use this at all and refactor the
+// instructions to not have SP an an explicit argument. That makes
+// frame index resolution a bit trickier, though.
+def GPRsp : RegisterClass<"ARM", [i32], 32, (add SP)>;
+
+// restricted GPR register class. Many Thumb2 instructions allow the full
+// register range for operands, but have undefined behaviours when PC
+// or SP (R13 or R15) are used. The ARM ISA refers to these operands
+// via the BadReg() pseudo-code description.
+def rGPR : RegisterClass<"ARM", [i32], 32, (sub GPR, SP, PC)> {
+ let AltOrders = [(add LR, rGPR), (trunc rGPR, 8)];
+ let AltOrderSelect = [{
+ return 1 + MF.getTarget().getSubtarget<ARMSubtarget>().isThumb1Only();
}];
}
// Thumb registers are R0-R7 normally. Some instructions can still use
// the general GPR register class above (MOV, e.g.)
-def tGPR : RegisterClass<"ARM", [i32], 32, [R0, R1, R2, R3, R4, R5, R6, R7]> {
- let MethodProtos = [{
- iterator allocation_order_begin(const MachineFunction &MF) const;
- iterator allocation_order_end(const MachineFunction &MF) const;
- }];
- // FIXME: We are reserving r3 in Thumb mode in case the PEI needs to use it
- // to generate large stack offset. Make it available once we have register
- // scavenging.
- let MethodBodies = [{
- static const unsigned THUMB_tGPR_AO[] = {
- ARM::R0, ARM::R1, ARM::R2,
- ARM::R4, ARM::R5, ARM::R6, ARM::R7 };
-
- // FP is R7, only low registers available.
- tGPRClass::iterator
- tGPRClass::allocation_order_begin(const MachineFunction &MF) const {
- return THUMB_tGPR_AO;
- }
-
- tGPRClass::iterator
- tGPRClass::allocation_order_end(const MachineFunction &MF) const {
- const TargetMachine &TM = MF.getTarget();
- const TargetRegisterInfo *RI = TM.getRegisterInfo();
- const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
- tGPRClass::iterator I =
- THUMB_tGPR_AO + (sizeof(THUMB_tGPR_AO)/sizeof(unsigned));
- // Mac OS X requires FP not to be clobbered for backtracing purpose.
- return (Subtarget.isTargetDarwin() || RI->hasFP(MF)) ? I-1 : I;
- }
+def tGPR : RegisterClass<"ARM", [i32], 32, (trunc GPR, 8)>;
+
+// The high registers in thumb mode, R8-R15.
+def hGPR : RegisterClass<"ARM", [i32], 32, (sub GPR, tGPR)>;
+
+// For tail calls, we can't use callee-saved registers, as they are restored
+// to the saved value before the tail call, which would clobber a call address.
+// Note, getMinimalPhysRegClass(R0) returns tGPR because of the names of
+// this class and the preceding one(!) This is what we want.
+def tcGPR : RegisterClass<"ARM", [i32], 32, (add R0, R1, R2, R3, R9, R12)> {
+ let AltOrders = [(and tcGPR, tGPR)];
+ let AltOrderSelect = [{
+ return MF.getTarget().getSubtarget<ARMSubtarget>().isThumb1Only();
}];
}
+// Condition code registers.
+def CCR : RegisterClass<"ARM", [i32], 32, (add CPSR)> {
+ let CopyCost = -1; // Don't allow copying of status registers.
+ let isAllocatable = 0;
+}
+
// Scalar single precision floating point register class..
-def SPR : RegisterClass<"ARM", [f32], 32, [S0, S1, S2, S3, S4, S5, S6, S7, S8,
- S9, S10, S11, S12, S13, S14, S15, S16, S17, S18, S19, S20, S21, S22,
- S23, S24, S25, S26, S27, S28, S29, S30, S31]>;
+def SPR : RegisterClass<"ARM", [f32], 32, (sequence "S%u", 0, 31)>;
+
+// Subset of SPR which can be used as a source of NEON scalars for 16-bit
+// operations
+def SPR_8 : RegisterClass<"ARM", [f32], 32, (trunc SPR, 16)>;
// Scalar double precision floating point / generic 64-bit vector register
// class.
// ARM requires only word alignment for double. It's more performant if it
// is double-word alignment though.
def DPR : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64,
- [D0, D1, D2, D3, D4, D5, D6, D7,
- D8, D9, D10, D11, D12, D13, D14, D15]> {
- let SubRegClassList = [SPR, SPR];
- let MethodProtos = [{
- iterator allocation_order_begin(const MachineFunction &MF) const;
- iterator allocation_order_end(const MachineFunction &MF) const;
- }];
- let MethodBodies = [{
- // VFP2
- static const unsigned ARM_DPR_VFP2[] = {
- ARM::D0, ARM::D1, ARM::D2, ARM::D3,
- ARM::D4, ARM::D5, ARM::D6, ARM::D7,
- ARM::D8, ARM::D9, ARM::D10, ARM::D11,
- ARM::D12, ARM::D13, ARM::D14, ARM::D15 };
- // VFP3
- static const unsigned ARM_DPR_VFP3[] = {
- ARM::D0, ARM::D1, ARM::D2, ARM::D3,
- ARM::D4, ARM::D5, ARM::D6, ARM::D7,
- ARM::D8, ARM::D9, ARM::D10, ARM::D11,
- ARM::D12, ARM::D13, ARM::D14, ARM::D15,
- ARM::D16, ARM::D17, ARM::D18, ARM::D15,
- ARM::D20, ARM::D21, ARM::D22, ARM::D23,
- ARM::D24, ARM::D25, ARM::D26, ARM::D27,
- ARM::D28, ARM::D29, ARM::D30, ARM::D31 };
- DPRClass::iterator
- DPRClass::allocation_order_begin(const MachineFunction &MF) const {
- const TargetMachine &TM = MF.getTarget();
- const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
- if (Subtarget.hasVFP3())
- return ARM_DPR_VFP3;
- return ARM_DPR_VFP2;
- }
-
- DPRClass::iterator
- DPRClass::allocation_order_end(const MachineFunction &MF) const {
- const TargetMachine &TM = MF.getTarget();
- const ARMSubtarget &Subtarget = TM.getSubtarget<ARMSubtarget>();
- if (Subtarget.hasVFP3())
- return ARM_DPR_VFP3 + (sizeof(ARM_DPR_VFP3)/sizeof(unsigned));
- else
- return ARM_DPR_VFP2 + (sizeof(ARM_DPR_VFP2)/sizeof(unsigned));
- }
- }];
+ (sequence "D%u", 0, 31)> {
+ // Allocate non-VFP2 registers D16-D31 first.
+ let AltOrders = [(rotl DPR, 16)];
+ let AltOrderSelect = [{ return 1; }];
+}
+
+// Subset of DPR that are accessible with VFP2 (and so that also have
+// 32-bit SPR subregs).
+def DPR_VFP2 : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64,
+ (trunc DPR, 16)> {
+ let SubRegClasses = [(SPR ssub_0, ssub_1)];
+}
+
+// Subset of DPR which can be used as a source of NEON scalars for 16-bit
+// operations
+def DPR_8 : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64,
+ (trunc DPR, 8)> {
+ let SubRegClasses = [(SPR_8 ssub_0, ssub_1)];
}
// Generic 128-bit vector register class.
def QPR : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], 128,
- [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7,
- Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15]> {
- let SubRegClassList = [SPR, SPR, SPR, SPR, DPR, DPR];
+ (sequence "Q%u", 0, 15)> {
+ let SubRegClasses = [(DPR dsub_0, dsub_1)];
+ // Allocate non-VFP2 aliases Q8-Q15 first.
+ let AltOrders = [(rotl QPR, 8)];
+ let AltOrderSelect = [{ return 1; }];
}
-// Vector register class for NEON vector structures occupying either 3 or 4
-// DPR registers.
-def Q2PR: RegisterClass<"ARM", [v24i8, v12i16, v6i32, v3i64, v6f32,
- v32i8, v16i16, v8i32, v4i64, v8f32], 256,
- [Q2_0, Q2_1, Q2_2, Q2_3, Q2_4, Q2_5, Q2_6, Q2_7]> {
- let SubRegClassList = [SPR, SPR, SPR, SPR, SPR, SPR, SPR, SPR,
- DPR, DPR, DPR, DPR, QPR, QPR];
+// Subset of QPR that have 32-bit SPR subregs.
+def QPR_VFP2 : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
+ 128, (trunc QPR, 8)> {
+ let SubRegClasses = [(SPR ssub_0, ssub_1, ssub_2, ssub_3),
+ (DPR_VFP2 dsub_0, dsub_1)];
}
-// Vector register class for NEON vector structures occupying either 6 or 8
-// DPR registers.
-def Q4PR: RegisterClass<"ARM", [v48i8, v24i16, v12i32, v6i64, v12f32,
- v64i8, v32i16, v16i32, v8i64, v16f32], 512,
- [Q4_0, Q4_1, Q4_2, Q4_3]> {
- let SubRegClassList = [SPR, SPR, SPR, SPR, SPR, SPR, SPR, SPR,
- SPR, SPR, SPR, SPR, SPR, SPR, SPR, SPR,
- DPR, DPR, DPR, DPR, DPR, DPR, DPR, DPR,
- QPR, QPR, QPR, QPR, Q2PR, Q2PR];
+// Subset of QPR that have DPR_8 and SPR_8 subregs.
+def QPR_8 : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
+ 128, (trunc QPR, 4)> {
+ let SubRegClasses = [(SPR_8 ssub_0, ssub_1, ssub_2, ssub_3),
+ (DPR_8 dsub_0, dsub_1)];
}
-// Condition code registers.
-def CCR : RegisterClass<"ARM", [i32], 32, [CPSR]>;
+// Pseudo-registers representing odd-even pairs of D registers. The even-odd
+// pairs are already represented by the Q registers.
+// These are needed by NEON instructions requiring two consecutive D registers.
+// There is no D31_D0 register as that is always an UNPREDICTABLE encoding.
+def TuplesOE2D : RegisterTuples<[dsub_0, dsub_1],
+ [(decimate (shl DPR, 1), 2),
+ (decimate (shl DPR, 2), 2)]>;
+
+// Register class representing a pair of consecutive D registers.
+// Use the Q registers for the even-odd pairs.
+def DPair : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
+ 128, (interleave QPR, TuplesOE2D)> {
+ // Allocate starting at non-VFP2 registers D16-D31 first.
+ let AltOrders = [(rotl DPair, 16)];
+ let AltOrderSelect = [{ return 1; }];
+}
-//===----------------------------------------------------------------------===//
-// Subregister Set Definitions... now that we have all of the pieces, define the
-// sub registers for each register.
-//
+// Pseudo-registers representing 3 consecutive D registers.
+def Tuples3D : RegisterTuples<[dsub_0, dsub_1, dsub_2],
+ [(shl DPR, 0),
+ (shl DPR, 1),
+ (shl DPR, 2)]>;
+
+// 3 consecutive D registers.
+def DTriple : RegisterClass<"ARM", [untyped], 64, (add Tuples3D)> {
+ let Size = 192; // 3 x 64 bits, we have no predefined type of that size.
+}
+
+// Pseudo 256-bit registers to represent pairs of Q registers. These should
+// never be present in the emitted code.
+// These are used for NEON load / store instructions, e.g., vld4, vst3.
+def Tuples2Q : RegisterTuples<[qsub_0, qsub_1], [(shl QPR, 0), (shl QPR, 1)]>;
+
+// Pseudo 256-bit vector register class to model pairs of Q registers
+// (4 consecutive D registers).
+def QQPR : RegisterClass<"ARM", [v4i64], 256, (add Tuples2Q)> {
+ let SubRegClasses = [(DPR dsub_0, dsub_1, dsub_2, dsub_3),
+ (QPR qsub_0, qsub_1)];
+ // Allocate non-VFP2 aliases first.
+ let AltOrders = [(rotl QQPR, 8)];
+ let AltOrderSelect = [{ return 1; }];
+}
+
+// Tuples of 4 D regs that isn't also a pair of Q regs.
+def TuplesOE4D : RegisterTuples<[dsub_0, dsub_1, dsub_2, dsub_3],
+ [(decimate (shl DPR, 1), 2),
+ (decimate (shl DPR, 2), 2),
+ (decimate (shl DPR, 3), 2),
+ (decimate (shl DPR, 4), 2)]>;
+
+// 4 consecutive D registers.
+def DQuad : RegisterClass<"ARM", [v4i64], 256,
+ (interleave Tuples2Q, TuplesOE4D)>;
+
+// Pseudo 512-bit registers to represent four consecutive Q registers.
+def Tuples2QQ : RegisterTuples<[qqsub_0, qqsub_1],
+ [(shl QQPR, 0), (shl QQPR, 2)]>;
+
+// Pseudo 512-bit vector register class to model 4 consecutive Q registers
+// (8 consecutive D registers).
+def QQQQPR : RegisterClass<"ARM", [v8i64], 256, (add Tuples2QQ)> {
+ let SubRegClasses = [(DPR dsub_0, dsub_1, dsub_2, dsub_3,
+ dsub_4, dsub_5, dsub_6, dsub_7),
+ (QPR qsub_0, qsub_1, qsub_2, qsub_3)];
+ // Allocate non-VFP2 aliases first.
+ let AltOrders = [(rotl QQQQPR, 8)];
+ let AltOrderSelect = [{ return 1; }];
+}
+
+
+// Pseudo-registers representing 2-spaced consecutive D registers.
+def Tuples2DSpc : RegisterTuples<[dsub_0, dsub_2],
+ [(shl DPR, 0),
+ (shl DPR, 2)]>;
+
+// Spaced pairs of D registers.
+def DPairSpc : RegisterClass<"ARM", [v2i64], 64, (add Tuples2DSpc)>;
+
+def Tuples3DSpc : RegisterTuples<[dsub_0, dsub_2, dsub_4],
+ [(shl DPR, 0),
+ (shl DPR, 2),
+ (shl DPR, 4)]>;
+
+// Spaced triples of D registers.
+def DTripleSpc : RegisterClass<"ARM", [untyped], 64, (add Tuples3DSpc)> {
+ let Size = 192; // 3 x 64 bits, we have no predefined type of that size.
+}
+
+def Tuples4DSpc : RegisterTuples<[dsub_0, dsub_2, dsub_4, dsub_6],
+ [(shl DPR, 0),
+ (shl DPR, 2),
+ (shl DPR, 4),
+ (shl DPR, 6)]>;
-def arm_ssubreg_0 : PatLeaf<(i32 1)>;
-def arm_ssubreg_1 : PatLeaf<(i32 2)>;
-def arm_ssubreg_2 : PatLeaf<(i32 3)>;
-def arm_ssubreg_3 : PatLeaf<(i32 4)>;
-def arm_dsubreg_0 : PatLeaf<(i32 5)>;
-def arm_dsubreg_1 : PatLeaf<(i32 6)>;
-
-// S sub-registers of D registers.
-def : SubRegSet<1, [D0, D1, D2, D3, D4, D5, D6, D7,
- D8, D9, D10, D11, D12, D13, D14, D15],
- [S0, S2, S4, S6, S8, S10, S12, S14,
- S16, S18, S20, S22, S24, S26, S28, S30]>;
-def : SubRegSet<2, [D0, D1, D2, D3, D4, D5, D6, D7,
- D8, D9, D10, D11, D12, D13, D14, D15],
- [S1, S3, S5, S7, S9, S11, S13, S15,
- S17, S19, S21, S23, S25, S27, S29, S31]>;
-
-// S sub-registers of Q registers.
-def : SubRegSet<1, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
- [S0, S4, S8, S12, S16, S20, S24, S28]>;
-def : SubRegSet<2, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
- [S1, S5, S9, S13, S17, S21, S25, S29]>;
-def : SubRegSet<3, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
- [S2, S6, S10, S14, S18, S22, S26, S30]>;
-def : SubRegSet<4, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
- [S3, S7, S11, S15, S19, S23, S27, S31]>;
-
-// D sub-registers of Q registers.
-def : SubRegSet<5, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7,
- Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15],
- [D0, D2, D4, D6, D8, D10, D12, D14,
- D16, D18, D20, D22, D24, D26, D28, D30]>;
-def : SubRegSet<6, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7,
- Q8, Q9, Q10, Q11, Q12, Q13, Q14, Q15],
- [D1, D3, D5, D7, D9, D11, D13, D15,
- D17, D19, D21, D23, D25, D27, D29, D31]>;
-
-// S sub-registers of Q2 registers.
-def : SubRegSet<1, [Q2_0, Q2_1, Q2_2, Q2_3], [S0, S8, S16, S24]>;
-def : SubRegSet<2, [Q2_0, Q2_1, Q2_2, Q2_3], [S1, S9, S17, S25]>;
-def : SubRegSet<3, [Q2_0, Q2_1, Q2_2, Q2_3], [S2, S10, S18, S26]>;
-def : SubRegSet<4, [Q2_0, Q2_1, Q2_2, Q2_3], [S3, S11, S19, S27]>;
-def : SubRegSet<5, [Q2_0, Q2_1, Q2_2, Q2_3], [S4, S12, S20, S28]>;
-def : SubRegSet<6, [Q2_0, Q2_1, Q2_2, Q2_3], [S5, S13, S21, S29]>;
-def : SubRegSet<7, [Q2_0, Q2_1, Q2_2, Q2_3], [S6, S14, S22, S30]>;
-def : SubRegSet<8, [Q2_0, Q2_1, Q2_2, Q2_3], [S7, S15, S23, S31]>;
-
-// D sub-registers of Q2 registers.
-def : SubRegSet<9, [Q2_0, Q2_1, Q2_2, Q2_3, Q2_4, Q2_5, Q2_6, Q2_7],
- [D0, D4, D8, D12, D16, D20, D24, D28]>;
-def : SubRegSet<10, [Q2_0, Q2_1, Q2_2, Q2_3, Q2_4, Q2_5, Q2_6, Q2_7],
- [D1, D5, D9, D13, D17, D21, D25, D29]>;
-def : SubRegSet<11, [Q2_0, Q2_1, Q2_2, Q2_3, Q2_4, Q2_5, Q2_6, Q2_7],
- [D2, D6, D10, D14, D18, D22, D26, D30]>;
-def : SubRegSet<12, [Q2_0, Q2_1, Q2_2, Q2_3, Q2_4, Q2_5, Q2_6, Q2_7],
- [D3, D7, D11, D15, D19, D23, D27, D31]>;
-
-// Q sub-registers of Q2 registers.
-def : SubRegSet<13, [Q2_0, Q2_1, Q2_2, Q2_3, Q2_4, Q2_5, Q2_6, Q2_7],
- [Q0, Q2, Q4, Q6, Q8, Q10, Q12, Q14]>;
-def : SubRegSet<14, [Q2_0, Q2_1, Q2_2, Q2_3, Q2_4, Q2_5, Q2_6, Q2_7],
- [Q1, Q3, Q5, Q7, Q9, Q11, Q13, Q15]>;
-
-// S sub-registers of Q4 registers.
-def : SubRegSet<1, [Q4_0, Q4_1], [S0, S16]>;
-def : SubRegSet<2, [Q4_0, Q4_1], [S1, S17]>;
-def : SubRegSet<3, [Q4_0, Q4_1], [S2, S18]>;
-def : SubRegSet<4, [Q4_0, Q4_1], [S3, S19]>;
-def : SubRegSet<5, [Q4_0, Q4_1], [S4, S20]>;
-def : SubRegSet<6, [Q4_0, Q4_1], [S5, S21]>;
-def : SubRegSet<7, [Q4_0, Q4_1], [S6, S22]>;
-def : SubRegSet<8, [Q4_0, Q4_1], [S7, S23]>;
-def : SubRegSet<9, [Q4_0, Q4_1], [S8, S24]>;
-def : SubRegSet<10, [Q4_0, Q4_1], [S9, S25]>;
-def : SubRegSet<11, [Q4_0, Q4_1], [S10, S26]>;
-def : SubRegSet<12, [Q4_0, Q4_1], [S11, S27]>;
-def : SubRegSet<13, [Q4_0, Q4_1], [S12, S28]>;
-def : SubRegSet<14, [Q4_0, Q4_1], [S13, S29]>;
-def : SubRegSet<15, [Q4_0, Q4_1], [S14, S30]>;
-def : SubRegSet<16, [Q4_0, Q4_1], [S15, S31]>;
-
-// D sub-registers of Q4 registers.
-def : SubRegSet<17, [Q4_0, Q4_1, Q4_2, Q4_3], [D0, D8, D16, D24]>;
-def : SubRegSet<18, [Q4_0, Q4_1, Q4_2, Q4_3], [D1, D9, D17, D25]>;
-def : SubRegSet<19, [Q4_0, Q4_1, Q4_2, Q4_3], [D2, D10, D18, D26]>;
-def : SubRegSet<20, [Q4_0, Q4_1, Q4_2, Q4_3], [D3, D11, D19, D27]>;
-def : SubRegSet<21, [Q4_0, Q4_1, Q4_2, Q4_3], [D4, D12, D20, D28]>;
-def : SubRegSet<22, [Q4_0, Q4_1, Q4_2, Q4_3], [D5, D13, D21, D29]>;
-def : SubRegSet<23, [Q4_0, Q4_1, Q4_2, Q4_3], [D6, D14, D22, D30]>;
-def : SubRegSet<24, [Q4_0, Q4_1, Q4_2, Q4_3], [D7, D15, D23, D31]>;
-
-// Q sub-registers of Q4 registers.
-def : SubRegSet<25, [Q4_0, Q4_1, Q4_2, Q4_3], [Q0, Q4, Q8, Q12]>;
-def : SubRegSet<26, [Q4_0, Q4_1, Q4_2, Q4_3], [Q1, Q5, Q9, Q13]>;
-def : SubRegSet<27, [Q4_0, Q4_1, Q4_2, Q4_3], [Q2, Q6, Q10, Q14]>;
-def : SubRegSet<28, [Q4_0, Q4_1, Q4_2, Q4_3], [Q3, Q7, Q11, Q15]>;
-
-// Q2 sub-registers of Q4 registers.
-def : SubRegSet<29, [Q4_0, Q4_1, Q4_2, Q4_3], [Q2_0, Q2_2, Q2_4, Q2_6]>;
-def : SubRegSet<30, [Q4_0, Q4_1, Q4_2, Q4_3], [Q2_1, Q2_3, Q2_5, Q2_7]>;
+// Spaced quads of D registers.
+def DQuadSpc : RegisterClass<"ARM", [v4i64], 64, (add Tuples3DSpc)>;