let Namespace = "ARM";
}
+// Subregister indices.
+let Namespace = "ARM" in {
+// Note: Code depends on these having consecutive numbers.
+def ssub_0 : SubRegIndex { let NumberHack = 1; }
+def ssub_1 : SubRegIndex { let NumberHack = 2; }
+def ssub_2 : SubRegIndex { let NumberHack = 3; }
+def ssub_3 : SubRegIndex { let NumberHack = 4; }
+
+def dsub_0 : SubRegIndex { let NumberHack = 5; }
+def dsub_1 : SubRegIndex { let NumberHack = 6; }
+def dsub_2 : SubRegIndex { let NumberHack = 7; }
+def dsub_3 : SubRegIndex { let NumberHack = 8; }
+def dsub_4 : SubRegIndex { let NumberHack = 9; }
+def dsub_5 : SubRegIndex { let NumberHack = 10; }
+def dsub_6 : SubRegIndex { let NumberHack = 11; }
+def dsub_7 : SubRegIndex { let NumberHack = 12; }
+
+def qsub_0 : SubRegIndex { let NumberHack = 13; }
+def qsub_1 : SubRegIndex { let NumberHack = 14; }
+def qsub_2 : SubRegIndex { let NumberHack = 15; }
+def qsub_3 : SubRegIndex { let NumberHack = 16; }
+
+def qqsub_0 : SubRegIndex { let NumberHack = 17; }
+def qqsub_1 : SubRegIndex { let NumberHack = 18; }
+}
+
// Integer registers
def R0 : ARMReg< 0, "r0">, DwarfRegNum<[0]>;
def R1 : ARMReg< 1, "r1">, DwarfRegNum<[1]>;
def S26 : ARMFReg<26, "s26">; def S27 : ARMFReg<27, "s27">;
def S28 : ARMFReg<28, "s28">; def S29 : ARMFReg<29, "s29">;
def S30 : ARMFReg<30, "s30">; def S31 : ARMFReg<31, "s31">;
-def SDummy : ARMFReg<63, "sINVALID">;
// Aliases of the F* registers used to hold 64-bit fp values (doubles)
def D0 : ARMReg< 0, "d0", [S0, S1]>;
def Q14 : ARMReg<14, "q14", [D28, D29]>;
def Q15 : ARMReg<15, "q15", [D30, D31]>;
+// 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 technical 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 sub-register
+// stuffs very messy.
+def QQ0 : ARMReg<0, "qq0", [Q0, Q1]>;
+def QQ1 : ARMReg<1, "qq1", [Q2, Q3]>;
+def QQ2 : ARMReg<2, "qq2", [Q4, Q5]>;
+def QQ3 : ARMReg<3, "qq3", [Q6, Q7]>;
+def QQ4 : ARMReg<4, "qq4", [Q8, Q9]>;
+def QQ5 : ARMReg<5, "qq5", [Q10, Q11]>;
+def QQ6 : ARMReg<6, "qq6", [Q12, Q13]>;
+def QQ7 : ARMReg<7, "qq7", [Q14, Q15]>;
+
+// Pseudo 512-bit registers to represent four consecutive Q registers.
+def QQQQ0 : ARMReg<0, "qqqq0", [QQ0, QQ1]>;
+def QQQQ1 : ARMReg<1, "qqqq1", [QQ2, QQ3]>;
+def QQQQ2 : ARMReg<2, "qqqq2", [QQ4, QQ5]>;
+def QQQQ3 : ARMReg<3, "qqqq3", [QQ6, QQ7]>;
+
// Current Program Status Register.
def CPSR : ARMReg<0, "cpsr">;
[S0, S1, S2, S3, S4, S5, S6, S7,
S8, S9, S10, S11, S12, S13, S14, S15]>;
-// Dummy f32 regclass to represent impossible subreg indices.
-def SPR_INVALID : RegisterClass<"ARM", [f32], 32, [SDummy]> {
- let CopyCost = -1;
-}
-
// Scalar double precision floating point / generic 64-bit vector register
// class.
// ARM requires only word alignment for double. It's more performant if it
D8, D9, D10, D11, D12, D13, D14, D15,
D16, D17, D18, D19, D20, D21, D22, D23,
D24, D25, D26, D27, D28, D29, D30, D31]> {
- let SubRegClassList = [SPR_INVALID, SPR_INVALID];
let MethodProtos = [{
iterator allocation_order_begin(const MachineFunction &MF) const;
iterator allocation_order_end(const MachineFunction &MF) const;
def DPR_VFP2 : 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 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,
[D0, D1, D2, D3, D4, D5, D6, D7]> {
- let SubRegClassList = [SPR_8, SPR_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_INVALID, SPR_INVALID, SPR_INVALID, SPR_INVALID,
- DPR, DPR];
+ let SubRegClasses = [(DPR dsub_0, dsub_1)];
}
// Subset of QPR that have 32-bit SPR subregs.
def QPR_VFP2 : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
128,
[Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7]> {
- let SubRegClassList = [SPR, SPR, SPR, SPR, DPR_VFP2, DPR_VFP2];
+ let SubRegClasses = [(SPR ssub_0, ssub_1, ssub_2, ssub_3),
+ (DPR_VFP2 dsub_0, dsub_1)];
}
// Subset of QPR that have DPR_8 and SPR_8 subregs.
def QPR_8 : RegisterClass<"ARM", [v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
128,
[Q0, Q1, Q2, Q3]> {
- let SubRegClassList = [SPR_8, SPR_8, SPR_8, SPR_8, DPR_8, DPR_8];
+ let SubRegClasses = [(SPR_8 ssub_0, ssub_1, ssub_2, ssub_3),
+ (DPR_8 dsub_0, dsub_1)];
+}
+
+// Pseudo 256-bit vector register class to model pairs of Q registers
+// (4 consecutive D registers).
+def QQPR : RegisterClass<"ARM", [v4i64],
+ 256,
+ [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7]> {
+ let SubRegClasses = [(DPR dsub_0, dsub_1, dsub_2, dsub_3),
+ (QPR qsub_0, qsub_1)];
+}
+
+// Subset of QQPR that have 32-bit SPR subregs.
+def QQPR_VFP2 : RegisterClass<"ARM", [v4i64],
+ 256,
+ [QQ0, QQ1, QQ2, QQ3]> {
+ let SubRegClasses = [(SPR ssub_0, ssub_1, ssub_2, ssub_3),
+ (DPR_VFP2 dsub_0, dsub_1, dsub_2, dsub_3),
+ (QPR_VFP2 qsub_0, qsub_1)];
+
+}
+
+// Pseudo 512-bit vector register class to model 4 consecutive Q registers
+// (8 consecutive D registers).
+def QQQQPR : RegisterClass<"ARM", [v8i64],
+ 256,
+ [QQQQ0, QQQQ1, QQQQ2, QQQQ3]> {
+ 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)];
}
// Condition code registers.
// sub registers for each register.
//
-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]>;
+def : SubRegSet<ssub_0, [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<ssub_1, [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]>;
+def : SubRegSet<ssub_0, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
+ [S0, S4, S8, S12, S16, S20, S24, S28]>;
+def : SubRegSet<ssub_1, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
+ [S1, S5, S9, S13, S17, S21, S25, S29]>;
+def : SubRegSet<ssub_2, [Q0, Q1, Q2, Q3, Q4, Q5, Q6, Q7],
+ [S2, S6, S10, S14, S18, S22, S26, S30]>;
+def : SubRegSet<ssub_3, [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]>;
+def : SubRegSet<dsub_0, [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<dsub_1, [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 QQ registers. Note there are no sub-indices
+// for referencing S4 - S7, S12 - S15, and S20 - S23. It doesn't
+// look like we need them.
+def : SubRegSet<ssub_0, [QQ0, QQ1, QQ2, QQ3],
+ [S0, S8, S16, S24]>;
+def : SubRegSet<ssub_1, [QQ0, QQ1, QQ2, QQ3],
+ [S1, S9, S17, S25]>;
+def : SubRegSet<ssub_2, [QQ0, QQ1, QQ2, QQ3],
+ [S2, S10, S18, S26]>;
+def : SubRegSet<ssub_3, [QQ0, QQ1, QQ2, QQ3],
+ [S3, S11, S19, S27]>;
+
+// D sub-registers of QQ registers.
+def : SubRegSet<dsub_0, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
+ [D0, D4, D8, D12, D16, D20, D24, D28]>;
+def : SubRegSet<dsub_1, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
+ [D1, D5, D9, D13, D17, D21, D25, D29]>;
+def : SubRegSet<dsub_2, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
+ [D2, D6, D10, D14, D18, D22, D26, D30]>;
+def : SubRegSet<dsub_3, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
+ [D3, D7, D11, D15, D19, D23, D27, D31]>;
+
+// Q sub-registers of QQ registers.
+def : SubRegSet<qsub_0, [QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
+ [Q0, Q2, Q4, Q6, Q8, Q10, Q12, Q14]>;
+def : SubRegSet<qsub_1,[QQ0, QQ1, QQ2, QQ3, QQ4, QQ5, QQ6, QQ7],
+ [Q1, Q3, Q5, Q7, Q9, Q11, Q13, Q15]>;
+
+
+// D sub-registers of QQQQ registers.
+def : SubRegSet<dsub_0, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [D0, D8, D16, D24]>;
+def : SubRegSet<dsub_1, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [D1, D9, D17, D25]>;
+def : SubRegSet<dsub_2, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [D2, D10, D18, D26]>;
+def : SubRegSet<dsub_3, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [D3, D11, D19, D27]>;
+
+def : SubRegSet<dsub_4, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [D4, D12, D20, D28]>;
+def : SubRegSet<dsub_5, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [D5, D13, D21, D29]>;
+def : SubRegSet<dsub_6, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [D6, D14, D22, D30]>;
+def : SubRegSet<dsub_7, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [D7, D15, D23, D31]>;
+
+// Q sub-registers of QQQQ registers.
+def : SubRegSet<qsub_0, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [Q0, Q4, Q8, Q12]>;
+def : SubRegSet<qsub_1, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [Q1, Q5, Q9, Q13]>;
+def : SubRegSet<qsub_2, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [Q2, Q6, Q10, Q14]>;
+def : SubRegSet<qsub_3, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [Q3, Q7, Q11, Q15]>;
+
+// QQ sub-registers of QQQQ registers.
+def : SubRegSet<qqsub_0, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [QQ0, QQ2, QQ4, QQ6]>;
+def : SubRegSet<qqsub_1, [QQQQ0, QQQQ1, QQQQ2, QQQQ3],
+ [QQ1, QQ3, QQ5, QQ7]>;
+
projects / oota-llvm.git / blobdiff