-//===- ARMRegisterInfo.td - ARM Register defs --------------*- tablegen -*-===//
+//===-- ARMRegisterInfo.td - ARM Register defs -------------*- tablegen -*-===//
//
// The LLVM Compiler Infrastructure
//
//===----------------------------------------------------------------------===//
// Registers are identified with 4-bit ID numbers.
-class ARMReg<bits<4> num, string n, list<Register> subregs = []> : Register<n> {
- field bits<4> Num;
+class ARMReg<bits<16> Enc, string n, list<Register> subregs = []> : Register<n> {
+ let HWEncoding = Enc;
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<6> num, string n> : Register<n> {
- field bits<6> Num;
+class ARMFReg<bits<16> Enc, string n> : Register<n> {
+ let HWEncoding = Enc;
let Namespace = "ARM";
}
// Subregister indices.
let Namespace = "ARM" in {
+def qqsub_0 : SubRegIndex<256>;
+def qqsub_1 : SubRegIndex<256, 256>;
+
// Note: Code depends on these having consecutive numbers.
-def ssub_0 : SubRegIndex;
-def ssub_1 : SubRegIndex;
-def ssub_2 : SubRegIndex; // In a Q reg.
-def ssub_3 : SubRegIndex;
-
-def dsub_0 : SubRegIndex;
-def dsub_1 : SubRegIndex;
-def dsub_2 : SubRegIndex;
-def dsub_3 : SubRegIndex;
-def dsub_4 : SubRegIndex;
-def dsub_5 : SubRegIndex;
-def dsub_6 : SubRegIndex;
-def dsub_7 : SubRegIndex;
-
-def qsub_0 : SubRegIndex;
-def qsub_1 : SubRegIndex;
-def qsub_2 : SubRegIndex;
-def qsub_3 : SubRegIndex;
-
-def qqsub_0 : SubRegIndex;
-def qqsub_1 : SubRegIndex;
+def qsub_0 : SubRegIndex<128>;
+def qsub_1 : SubRegIndex<128, 128>;
+def qsub_2 : ComposedSubRegIndex<qqsub_1, qsub_0>;
+def qsub_3 : ComposedSubRegIndex<qqsub_1, qsub_1>;
+
+def dsub_0 : SubRegIndex<64>;
+def dsub_1 : SubRegIndex<64, 64>;
+def dsub_2 : ComposedSubRegIndex<qsub_1, dsub_0>;
+def dsub_3 : ComposedSubRegIndex<qsub_1, dsub_1>;
+def dsub_4 : ComposedSubRegIndex<qsub_2, dsub_0>;
+def dsub_5 : ComposedSubRegIndex<qsub_2, dsub_1>;
+def dsub_6 : ComposedSubRegIndex<qsub_3, dsub_0>;
+def dsub_7 : ComposedSubRegIndex<qsub_3, dsub_1>;
+
+def ssub_0 : SubRegIndex<32>;
+def ssub_1 : SubRegIndex<32, 32>;
+def ssub_2 : ComposedSubRegIndex<dsub_1, ssub_0>;
+def ssub_3 : ComposedSubRegIndex<dsub_1, ssub_1>;
+
+def gsub_0 : SubRegIndex<32>;
+def gsub_1 : SubRegIndex<32, 32>;
+// Let TableGen synthesize the remaining 12 ssub_* indices.
+// We don't need to name them.
}
// Integer registers
}
// VFP3 defines 16 additional double registers
-def D16 : ARMFReg<16, "d16">, DwarfRegNum<[272]>;
+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 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 D31 : ARMFReg<31, "d31">, DwarfRegNum<[287]>;
// Advanced SIMD (NEON) defines 16 quad-word aliases
-let SubRegIndices = [dsub_0, dsub_1],
- CompositeIndices = [(ssub_2 dsub_1, ssub_0),
- (ssub_3 dsub_1, ssub_1)] in {
+let SubRegIndices = [dsub_0, dsub_1] in {
def Q0 : ARMReg< 0, "q0", [D0, D1]>;
def Q1 : ARMReg< 1, "q1", [D2, D3]>;
def Q2 : ARMReg< 2, "q2", [D4, D5]>;
}
// Current Program Status Register.
-def CPSR : ARMReg<0, "cpsr">;
-def APSR : ARMReg<1, "apsr">;
-def SPSR : ARMReg<2, "spsr">;
-def FPSCR : ARMReg<3, "fpscr">;
-def ITSTATE : ARMReg<4, "itstate">;
+// 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. APSR_NZCV
+// models the APSR when it's accessed by some special instructions. In such cases
+// it has the same encoding as PC.
+def CPSR : ARMReg<0, "cpsr">;
+def APSR : ARMReg<1, "apsr">;
+def APSR_NZCV : ARMReg<15, "apsr_nzcv">;
+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">;
+def FPSID : ARMReg<0, "fpsid">;
+def MVFR2 : ARMReg<5, "mvfr2">;
+def MVFR1 : ARMReg<6, "mvfr1">;
+def MVFR0 : ARMReg<7, "mvfr0">;
+def FPEXC : ARMReg<8, "fpexc">;
+def FPINST : ARMReg<9, "fpinst">;
+def FPINST2 : ARMReg<10, "fpinst2">;
// Register classes.
//
// Thumb1 instructions that know how to use hi regs.
let AltOrders = [(add LR, GPR), (trunc GPR, 8)];
let AltOrderSelect = [{
- return 1 + MF.get
Target().getSubtarget<ARMSubtarget>().isThumb1Only();
+ return 1 + MF.getSubtarget<ARMSubtarget>().isThumb1Only();
}];
}
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();
+ return 1 + MF.getSubtarget<ARMSubtarget>().isThumb1Only();
+ }];
+}
+
+// GPRs without the PC but with APSR. Some instructions allow accessing the
+// APSR, while actually encoding PC in the register field. This is useful
+// for assembly and disassembly only.
+def GPRwithAPSR : RegisterClass<"ARM", [i32], 32, (add (sub GPR, PC), APSR_NZCV)> {
+ let AltOrders = [(add LR, GPRnopc), (trunc GPRnopc, 8)];
+ let AltOrderSelect = [{
+ return 1 + MF.getSubtarget<ARMSubtarget>().isThumb1Only();
}];
}
def rGPR : RegisterClass<"ARM", [i32], 32, (sub GPR, SP, PC)> {
let AltOrders = [(add LR, rGPR), (trunc rGPR, 8)];
let AltOrderSelect = [{
- return 1 + MF.get
Target().getSubtarget<ARMSubtarget>().isThumb1Only();
+ return 1 + MF.getSubtarget<ARMSubtarget>().isThumb1Only();
}];
}
// 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)> {
+def tcGPR : RegisterClass<"ARM", [i32], 32, (add R0, R1, R2, R3, R12)> {
let AltOrders = [(and tcGPR, tGPR)];
let AltOrderSelect = [{
- return MF.get
Target().getSubtarget<ARMSubtarget>().isThumb1Only();
+ return MF.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, (sequence "S%u", 0, 31)>;
+// FIXME: Allocation order changed to s0, s2, ... or s0, s4, ... as a quick hack
+// to avoid partial-write dependencies on D or Q (depending on platform)
+// registers (S registers are renamed as portions of D/Q registers).
+def SPR : RegisterClass<"ARM", [f32], 32, (sequence "S%u", 0, 31)> {
+ let AltOrders = [(add (decimate SPR, 2), SPR),
+ (add (decimate SPR, 4),
+ (decimate SPR, 2),
+ (decimate (rotl SPR, 1), 4),
+ (decimate (rotl SPR, 1), 2))];
+ let AltOrderSelect = [{
+ return 1 + MF.getSubtarget<ARMSubtarget>().useStride4VFPs(MF);
+ }];
+}
// 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)>;
+def SPR_8 : RegisterClass<"ARM", [f32], 32, (sequence "S%u", 0, 15)>;
// Scalar double precision floating point / generic 64-bit vector register
// class.
// is double-word alignment though.
def DPR : RegisterClass<"ARM", [f64, v8i8, v4i16, v2i32, v1i64, v2f32], 64,
(sequence "D%u", 0, 31)> {
- // Allocate non-VFP2 registers D16-D31 first.
- let AltOrders = [(rotl DPR, 16)];
- let AltOrderSelect = [{ return 1; }];
+ // Allocate non-VFP2 registers D16-D31 first, and prefer even registers on
+ // Darwin platforms.
+ let AltOrders = [(rotl DPR, 16),
+ (add (decimate (rotl DPR, 16), 2), (rotl DPR, 16))];
+ let AltOrderSelect = [{
+ return 1 + MF.getSubtarget<ARMSubtarget>().useStride4VFPs(MF);
+ }];
}
// 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)];
-}
+ (trunc DPR, 16)>;
// 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)];
-}
+ (trunc DPR, 8)>;
// Generic 128-bit vector register class.
def QPR : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64], 128,
(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; }];
// 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)];
-}
+ 128, (trunc QPR, 8)>;
// 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)];
+ 128, (trunc QPR, 4)>;
+
+// 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.
+ // Prefer even-odd pairs as they are easier to copy.
+ let AltOrders = [(add (rotl QPR, 8), (rotl DPair, 16))];
+ let AltOrderSelect = [{ return 1; }];
+}
+
+// Pseudo-registers representing even-odd pairs of GPRs from R1 to R13/SP.
+// These are needed by instructions (e.g. ldrexd/strexd) requiring even-odd GPRs.
+def Tuples2R : RegisterTuples<[gsub_0, gsub_1],
+ [(add R0, R2, R4, R6, R8, R10, R12),
+ (add R1, R3, R5, R7, R9, R11, SP)]>;
+
+// Register class representing a pair of even-odd GPRs.
+def GPRPair : RegisterClass<"ARM", [untyped], 64, (add Tuples2R)> {
+ let Size = 64; // 2 x 32 bits, we have no predefined type of that size.
+}
+
+// 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.
-// NOTE: It's possible to define more QQ registers since technically the
-// starting D register number doesn't have to be multiple of 4, e.g.,
-// D1, D2, D3, D4 would be a legal quad, but that would make the subregister
-// stuff very messy.
-def Tuples2Q : RegisterTuples<[qsub_0, qsub_1],
- [(decimate QPR, 2),
- (decimate (shl QPR, 1), 2)]> {
- let CompositeIndices = [(dsub_2 qsub_1, dsub_0), (dsub_3 qsub_1, dsub_1)];
-}
+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, 4)];
+ 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],
- [(decimate QQPR, 2),
- (decimate (shl QQPR, 1), 2)]> {
- let CompositeIndices = [(qsub_2 qqsub_1, qsub_0), (qsub_3 qqsub_1, qsub_1),
- (dsub_4 qqsub_1, dsub_0), (dsub_5 qqsub_1, dsub_1),
- (dsub_6 qqsub_1, dsub_2), (dsub_7 qqsub_1, dsub_3)];
-}
+ [(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, 2)];
+ let AltOrders = [(rotl QQQQPR, 8)];
let AltOrderSelect = [{ return 1; }];
}
-// Condition code registers.
-def CCR : RegisterClass<"ARM", [i32], 32, (add CPSR)> {
- let CopyCost = -1; // Don't allow copying of status registers.
- let isAllocatable = 0;
+
+// 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)]>;
+
+// Spaced quads of D registers.
+def DQuadSpc : RegisterClass<"ARM", [v4i64], 64, (add Tuples3DSpc)>;
projects / oota-llvm.git / blobdiff