diff options
| -rw-r--r-- | lib/Target/ARM/ARMInstrInfo.td | 45 | ||||
| -rw-r--r-- | lib/Target/ARM/ARMSchedule.td | 7 | ||||
| -rw-r--r-- | lib/Target/ARM/ARMScheduleA9.td | 19 | ||||
| -rw-r--r-- | lib/Target/ARM/ARMScheduleSwift.td | 23 |
4 files changed, 70 insertions, 24 deletions
diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td index 9409f35..e0190a6 100644 --- a/lib/Target/ARM/ARMInstrInfo.td +++ b/lib/Target/ARM/ARMInstrInfo.td @@ -1010,7 +1010,8 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc, let isReMaterializable = 1 in { def ri : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), DPFrm, iii, opc, "\t$Rd, $Rn, $imm", - [(set GPR:$Rd, (opnode GPR:$Rn, so_imm:$imm))]> { + [(set GPR:$Rd, (opnode GPR:$Rn, so_imm:$imm))]>, + Sched<[WriteALU, ReadAdvanceALU]> { bits<4> Rd; bits<4> Rn; bits<12> imm; @@ -1022,7 +1023,8 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc, } def rr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), DPFrm, iir, opc, "\t$Rd, $Rn, $Rm", - [(set GPR:$Rd, (opnode GPR:$Rn, GPR:$Rm))]> { + [(set GPR:$Rd, (opnode GPR:$Rn, GPR:$Rm))]>, + Sched<[WriteALU, ReadAdvanceALU, ReadAdvanceALU]> { bits<4> Rd; bits<4> Rn; bits<4> Rm; @@ -1037,7 +1039,8 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc, def rsi : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_reg_imm:$shift), DPSoRegImmFrm, iis, opc, "\t$Rd, $Rn, $shift", - [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_imm:$shift))]> { + [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_imm:$shift))]>, + Sched<[WriteALUsi, ReadAdvanceALU]> { bits<4> Rd; bits<4> Rn; bits<12> shift; @@ -1052,7 +1055,8 @@ multiclass AsI1_bin_irs<bits<4> opcod, string opc, def rsr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_reg_reg:$shift), DPSoRegRegFrm, iis, opc, "\t$Rd, $Rn, $shift", - [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_reg:$shift))]> { + [(set GPR:$Rd, (opnode GPR:$Rn, so_reg_reg:$shift))]>, + Sched<[WriteALUsr, ReadAdvanceALUsr]> { bits<4> Rd; bits<4> Rn; bits<12> shift; @@ -1079,7 +1083,8 @@ multiclass AsI1_rbin_irs<bits<4> opcod, string opc, let isReMaterializable = 1 in { def ri : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm), DPFrm, iii, opc, "\t$Rd, $Rn, $imm", - [(set GPR:$Rd, (opnode so_imm:$imm, GPR:$Rn))]> { + [(set GPR:$Rd, (opnode so_imm:$imm, GPR:$Rn))]>, + Sched<[WriteALU, ReadAdvanceALU]> { bits<4> Rd; bits<4> Rn; bits<12> imm; @@ -1091,7 +1096,8 @@ multiclass AsI1_rbin_irs<bits<4> opcod, string opc, } def rr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm), DPFrm, iir, opc, "\t$Rd, $Rn, $Rm", - [/* pattern left blank */]> { + [/* pattern left blank */]>, + Sched<[WriteALU, ReadAdvanceALU, ReadAdvanceALU]> { bits<4> Rd; bits<4> Rn; bits<4> Rm; @@ -1105,7 +1111,8 @@ multiclass AsI1_rbin_irs<bits<4> opcod, string opc, def rsi : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_reg_imm:$shift), DPSoRegImmFrm, iis, opc, "\t$Rd, $Rn, $shift", - [(set GPR:$Rd, (opnode so_reg_imm:$shift, GPR:$Rn))]> { + [(set GPR:$Rd, (opnode so_reg_imm:$shift, GPR:$Rn))]>, + Sched<[WriteALUsi, ReadAdvanceALU]> { bits<4> Rd; bits<4> Rn; bits<12> shift; @@ -1120,7 +1127,8 @@ multiclass AsI1_rbin_irs<bits<4> opcod, string opc, def rsr : AsI1<opcod, (outs GPR:$Rd), (ins GPR:$Rn, so_reg_reg:$shift), DPSoRegRegFrm, iis, opc, "\t$Rd, $Rn, $shift", - [(set GPR:$Rd, (opnode so_reg_reg:$shift, GPR:$Rn))]> { + [(set GPR:$Rd, (opnode so_reg_reg:$shift, GPR:$Rn))]>, + Sched<[WriteALUsr, ReadAdvanceALUsr]> { bits<4> Rd; bits<4> Rn; bits<12> shift; @@ -1145,24 +1153,28 @@ multiclass AsI1_bin_s_irs<InstrItinClass iii, InstrItinClass iir, bit Commutable = 0> { def ri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm, pred:$p), 4, iii, - [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, so_imm:$imm))]>; + [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, so_imm:$imm))]>, + Sched<[WriteALU, ReadAdvanceALU]>; def rr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, GPR:$Rm, pred:$p), 4, iir, - [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, GPR:$Rm))]> { + [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, GPR:$Rm))]>, + Sched<[WriteALU, ReadAdvanceALU, ReadAdvanceALU]> { let isCommutable = Commutable; } def rsi : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg_imm:$shift, pred:$p), 4, iis, [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, - so_reg_imm:$shift))]>; + so_reg_imm:$shift))]>, + Sched<[WriteALUsi, ReadAdvanceALU]>; def rsr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg_reg:$shift, pred:$p), 4, iis, [(set GPR:$Rd, CPSR, (opnode GPR:$Rn, - so_reg_reg:$shift))]>; + so_reg_reg:$shift))]>, + Sched<[WriteALUSsr, ReadAdvanceALUsr]>; } } @@ -1174,19 +1186,22 @@ multiclass AsI1_rbin_s_is<InstrItinClass iii, InstrItinClass iir, bit Commutable = 0> { def ri : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_imm:$imm, pred:$p), 4, iii, - [(set GPR:$Rd, CPSR, (opnode so_imm:$imm, GPR:$Rn))]>; + [(set GPR:$Rd, CPSR, (opnode so_imm:$imm, GPR:$Rn))]>, + Sched<[WriteALU, ReadAdvanceALU]>; def rsi : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg_imm:$shift, pred:$p), 4, iis, [(set GPR:$Rd, CPSR, (opnode so_reg_imm:$shift, - GPR:$Rn))]>; + GPR:$Rn))]>, + Sched<[WriteALUsi, ReadAdvanceALU]>; def rsr : ARMPseudoInst<(outs GPR:$Rd), (ins GPR:$Rn, so_reg_reg:$shift, pred:$p), 4, iis, [(set GPR:$Rd, CPSR, (opnode so_reg_reg:$shift, - GPR:$Rn))]>; + GPR:$Rn))]>, + Sched<[WriteALUSsr, ReadAdvanceALUsr]>; } } diff --git a/lib/Target/ARM/ARMSchedule.td b/lib/Target/ARM/ARMSchedule.td index ff1ff2f..1bca554 100644 --- a/lib/Target/ARM/ARMSchedule.td +++ b/lib/Target/ARM/ARMSchedule.td @@ -64,6 +64,13 @@ def WriteALUsr : SchedWrite; // Shift by register. def WriteALUSsr : SchedWrite; // Shift by register (flag setting). def ReadAdvanceALUsr : SchedRead; // Some operands are read later. +// Define TII for use in SchedVariant Predicates. +def : PredicateProlog<[{ + const ARMBaseInstrInfo *TII = + static_cast<const ARMBaseInstrInfo*>(SchedModel->getInstrInfo()); + (void)TII; +}]>; + //===----------------------------------------------------------------------===// // Instruction Itinerary classes used for ARM // diff --git a/lib/Target/ARM/ARMScheduleA9.td b/lib/Target/ARM/ARMScheduleA9.td index 4191931..cec20b1 100644 --- a/lib/Target/ARM/ARMScheduleA9.td +++ b/lib/Target/ARM/ARMScheduleA9.td @@ -1898,6 +1898,8 @@ def CortexA9Model : SchedMachineModel { //===----------------------------------------------------------------------===// // Define each kind of processor resource and number available. +let SchedModel = CortexA9Model in { + def A9UnitALU : ProcResource<2>; def A9UnitMul : ProcResource<1> { let Super = A9UnitALU; } def A9UnitAGU : ProcResource<1>; @@ -2003,13 +2005,6 @@ foreach NumCycles = 2-8 in { def A9WriteCycle#NumCycles : WriteSequence<[A9WriteCycle1], NumCycles>; } // foreach NumCycles -// Define TII for use in SchedVariant Predicates. -def : PredicateProlog<[{ - const ARMBaseInstrInfo *TII = - static_cast<const ARMBaseInstrInfo*>(SchedModel->getInstrInfo()); - (void)TII; -}]>; - // Define address generation sequences and predicates for 8 flavors of LDMs. foreach NumAddr = 1-8 in { @@ -2279,7 +2274,6 @@ def A9Read4 : SchedReadAdvance<3>; // This table follows the ARM Cortex-A9 Technical Reference Manuals, // mostly in order. -let SchedModel = CortexA9Model in { def :ItinRW<[A9WriteI], [IIC_iMOVi,IIC_iMOVr,IIC_iMOVsi, IIC_iMVNi,IIC_iMVNsi, @@ -2486,4 +2480,13 @@ def :ItinRW<[A9WriteV9, A9Read3, A9Read2], [IIC_VMACD, IIC_VFMACD]>; def :ItinRW<[A9WriteV10, A9Read3, A9Read2], [IIC_VMACQ, IIC_VFMACQ]>; def :ItinRW<[A9WriteV9, A9Read2, A9Read2], [IIC_VRECSD]>; def :ItinRW<[A9WriteV10, A9Read2, A9Read2], [IIC_VRECSQ]>; + +// New (incomplete) model mappings that don't rely on itinerary mappings. +def : SchedAlias<WriteALU, A9WriteA>; +def : SchedAlias<WriteALUsi, A9WriteAsi>; +def : SchedAlias<WriteALUsr, A9WriteAsr>; +def : SchedAlias<WriteALUSsr, A9WriteAsr>; +def : SchedAlias<ReadAdvanceALU, A9ReadA>; +def : SchedAlias<ReadAdvanceALUsr, A9ReadA>; + } // SchedModel = CortexA9Model diff --git a/lib/Target/ARM/ARMScheduleSwift.td b/lib/Target/ARM/ARMScheduleSwift.td index e9bc3e0..becf41b 100644 --- a/lib/Target/ARM/ARMScheduleSwift.td +++ b/lib/Target/ARM/ARMScheduleSwift.td @@ -1078,8 +1078,29 @@ def SwiftModel : SchedMachineModel { let IssueWidth = 3; // 3 micro-ops are dispatched per cycle. let MinLatency = 0; // Data dependencies are allowed within dispatch groups. let LoadLatency = 3; + let MispredictPenalty = 14; // A branch direction mispredict. let Itineraries = SwiftItineraries; } -// TODO: Add Swift processor and scheduler resources. +// Swift resource mapping. +let SchedModel = SwiftModel in { + // Processor resources. + def SwiftUnitP01 : ProcResource<2>; // ALU unit. + def SwiftUnitP0 : ProcResource<1> { let Super = SwiftUnitP01; } // Mul unit. + def SwiftUnitP1 : ProcResource<1> { let Super = SwiftUnitP01; } // Br unit. + def SwiftUnitP2 : ProcResource<1>; // LS unit. + def SwiftUnitDiv : ProcResource<1>; + + // 4.2.4 Arithmetic and Logical. + // ADC,ADD,NEG,RSB,RSC,SBC,SUB,ADR + // AND,BIC, EOR,ORN,ORR + // CLZ,RBIT,REV,REV16,REVSH,PKH + // Single cycle. + def : WriteRes<WriteALU, [SwiftUnitP01]>; + def : WriteRes<WriteALUsi, [SwiftUnitP01]>; + def : WriteRes<WriteALUsr, [SwiftUnitP01]>; + def : WriteRes<WriteALUSsr, [SwiftUnitP01]>; + def : ReadAdvance<ReadAdvanceALU, 0>; + def : ReadAdvance<ReadAdvanceALUsr, 2>; +} |