HexagonScheduleV4.td [plain text]
//=-HexagonScheduleV4.td - HexagonV4 Scheduling Definitions --*- tablegen -*-=//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
// There are four SLOTS (four parallel pipelines) in Hexagon V4 machine.
// This file describes that machine information.
//
// |===========|==================================================|
// | PIPELINE | Instruction Classes |
// |===========|==================================================|
// | SLOT0 | LD ST ALU32 MEMOP NV SYSTEM |
// |-----------|--------------------------------------------------|
// | SLOT1 | LD ST ALU32 |
// |-----------|--------------------------------------------------|
// | SLOT2 | XTYPE ALU32 J JR |
// |-----------|--------------------------------------------------|
// | SLOT3 | XTYPE ALU32 J CR |
// |===========|==================================================|
// Functional Units.
def SLOT0 : FuncUnit;
def SLOT1 : FuncUnit;
def SLOT2 : FuncUnit;
def SLOT3 : FuncUnit;
// Endloop is a pseudo instruction that is encoded with 2 bits in a packet
// rather than taking an execution slot. This special unit is needed
// to schedule an ENDLOOP with 4 other instructions.
def SLOT_ENDLOOP: FuncUnit;
// Itinerary classes.
def PSEUDO : InstrItinClass;
def PSEUDOM : InstrItinClass;
// ALU64/M/S Instruction classes of V2 are collectively knownn as XTYPE in V4.
def DUPLEX : InstrItinClass;
def PREFIX : InstrItinClass;
def COMPOUND : InstrItinClass;
def ALU32_2op_tc_1_SLOT0123 : InstrItinClass;
def ALU32_2op_tc_2early_SLOT0123 : InstrItinClass;
def ALU32_3op_tc_2early_SLOT0123 : InstrItinClass;
def ALU32_3op_tc_1_SLOT0123 : InstrItinClass;
def ALU32_3op_tc_2_SLOT0123 : InstrItinClass;
def ALU32_ADDI_tc_1_SLOT0123 : InstrItinClass;
def ALU64_tc_1_SLOT23 : InstrItinClass;
def ALU64_tc_1or2_SLOT23 : InstrItinClass;
def ALU64_tc_2_SLOT23 : InstrItinClass;
def ALU64_tc_2early_SLOT23 : InstrItinClass;
def ALU64_tc_3x_SLOT23 : InstrItinClass;
def CR_tc_2_SLOT3 : InstrItinClass;
def CR_tc_2early_SLOT23 : InstrItinClass;
def CR_tc_2early_SLOT3 : InstrItinClass;
def CR_tc_3x_SLOT23 : InstrItinClass;
def CR_tc_3x_SLOT3 : InstrItinClass;
def J_tc_2early_SLOT23 : InstrItinClass;
def J_tc_2early_SLOT2 : InstrItinClass;
def LD_tc_ld_SLOT01 : InstrItinClass;
def LD_tc_ld_SLOT0 : InstrItinClass;
def LD_tc_3or4stall_SLOT0 : InstrItinClass;
def M_tc_1_SLOT23 : InstrItinClass;
def M_tc_1or2_SLOT23 : InstrItinClass;
def M_tc_2_SLOT23 : InstrItinClass;
def M_tc_3_SLOT23 : InstrItinClass;
def M_tc_3x_SLOT23 : InstrItinClass;
def M_tc_3or4x_SLOT23 : InstrItinClass;
def ST_tc_st_SLOT01 : InstrItinClass;
def ST_tc_st_SLOT0 : InstrItinClass;
def ST_tc_ld_SLOT0 : InstrItinClass;
def ST_tc_3stall_SLOT0 : InstrItinClass;
def S_2op_tc_1_SLOT23 : InstrItinClass;
def S_2op_tc_2_SLOT23 : InstrItinClass;
def S_2op_tc_2early_SLOT23 : InstrItinClass;
def S_2op_tc_3or4x_SLOT23 : InstrItinClass;
def S_3op_tc_1_SLOT23 : InstrItinClass;
def S_3op_tc_1or2_SLOT23 : InstrItinClass;
def S_3op_tc_2_SLOT23 : InstrItinClass;
def S_3op_tc_2early_SLOT23 : InstrItinClass;
def S_3op_tc_3_SLOT23 : InstrItinClass;
def S_3op_tc_3x_SLOT23 : InstrItinClass;
def NCJ_tc_3or4stall_SLOT0 : InstrItinClass;
def V2LDST_tc_ld_SLOT01 : InstrItinClass;
def V2LDST_tc_st_SLOT0 : InstrItinClass;
def V2LDST_tc_st_SLOT01 : InstrItinClass;
def V4LDST_tc_ld_SLOT01 : InstrItinClass;
def V4LDST_tc_st_SLOT0 : InstrItinClass;
def V4LDST_tc_st_SLOT01 : InstrItinClass;
def J_tc_2early_SLOT0123 : InstrItinClass;
def EXTENDER_tc_1_SLOT0123 : InstrItinClass;
def HexagonItinerariesV4 :
ProcessorItineraries<[SLOT0, SLOT1, SLOT2, SLOT3, SLOT_ENDLOOP], [], [
// ALU32
InstrItinData<ALU32_2op_tc_1_SLOT0123 ,
[InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
InstrItinData<ALU32_2op_tc_2early_SLOT0123,
[InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
InstrItinData<ALU32_3op_tc_1_SLOT0123 ,
[InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
InstrItinData<ALU32_3op_tc_2early_SLOT0123,
[InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
InstrItinData<ALU32_3op_tc_2_SLOT0123 ,
[InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
InstrItinData<ALU32_ADDI_tc_1_SLOT0123 ,
[InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
// ALU64
InstrItinData<ALU64_tc_1_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<ALU64_tc_1or2_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<ALU64_tc_2_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<ALU64_tc_2early_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<ALU64_tc_3x_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
// CR -> System
InstrItinData<CR_tc_2_SLOT3 , [InstrStage<1, [SLOT3]>]>,
InstrItinData<CR_tc_2early_SLOT3 , [InstrStage<1, [SLOT3]>]>,
InstrItinData<CR_tc_3x_SLOT3 , [InstrStage<1, [SLOT3]>]>,
// Jump (conditional/unconditional/return etc)
// CR
InstrItinData<CR_tc_2early_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<CR_tc_3x_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
// J
InstrItinData<J_tc_2early_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
// JR
InstrItinData<J_tc_2early_SLOT2 , [InstrStage<1, [SLOT2]>]>,
//Load
InstrItinData<LD_tc_ld_SLOT01 , [InstrStage<1, [SLOT0, SLOT1]>]>,
InstrItinData<LD_tc_ld_SLOT0 , [InstrStage<1, [SLOT0]>]>,
InstrItinData<LD_tc_3or4stall_SLOT0 , [InstrStage<1, [SLOT0]>]>,
// M
InstrItinData<M_tc_1_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<M_tc_1or2_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<M_tc_2_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<M_tc_3_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<M_tc_3x_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<M_tc_3or4x_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
// Store
// ST
InstrItinData<ST_tc_st_SLOT01 , [InstrStage<1, [SLOT0, SLOT1]>]>,
// ST0
InstrItinData<ST_tc_st_SLOT0 , [InstrStage<1, [SLOT0]>]>,
InstrItinData<ST_tc_ld_SLOT0 , [InstrStage<1, [SLOT0]>]>,
// S
InstrItinData<S_2op_tc_1_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_2op_tc_2_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_2op_tc_2early_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_2op_tc_3or4x_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_3op_tc_1_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_3op_tc_1or2_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_3op_tc_2early_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_3op_tc_2_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_3op_tc_3_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<S_3op_tc_3x_SLOT23 , [InstrStage<1, [SLOT2, SLOT3]>]>,
// SYS
InstrItinData<ST_tc_3stall_SLOT0 , [InstrStage<1, [SLOT0]>]>,
// New Value Compare Jump
InstrItinData<NCJ_tc_3or4stall_SLOT0 , [InstrStage<1, [SLOT0]>]>,
// Mem ops - MEM_V4
InstrItinData<V2LDST_tc_st_SLOT0 , [InstrStage<1, [SLOT0]>]>,
InstrItinData<V2LDST_tc_ld_SLOT01 , [InstrStage<1, [SLOT0, SLOT1]>]>,
InstrItinData<V2LDST_tc_st_SLOT01 , [InstrStage<1, [SLOT0, SLOT1]>]>,
InstrItinData<V4LDST_tc_st_SLOT0 , [InstrStage<1, [SLOT0]>]>,
InstrItinData<V4LDST_tc_ld_SLOT01 , [InstrStage<1, [SLOT0, SLOT1]>]>,
InstrItinData<V4LDST_tc_st_SLOT01 , [InstrStage<1, [SLOT0, SLOT1]>]>,
InstrItinData<DUPLEX , [InstrStage<1, [SLOT0]>]>,
// ENDLOOP
InstrItinData<J_tc_2early_SLOT0123 , [InstrStage<1, [SLOT_ENDLOOP]>]>,
// Extender/PREFIX
InstrItinData<EXTENDER_tc_1_SLOT0123,
[InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
InstrItinData<COMPOUND , [InstrStage<1, [SLOT2, SLOT3]>]>,
InstrItinData<PSEUDO , [InstrStage<1, [SLOT0, SLOT1, SLOT2, SLOT3]>]>,
InstrItinData<PSEUDOM, [InstrStage<1, [SLOT2, SLOT3], 0>,
InstrStage<1, [SLOT2, SLOT3]>]>
]>;
def HexagonModelV4 : SchedMachineModel {
// Max issue per cycle == bundle width.
let IssueWidth = 4;
let Itineraries = HexagonItinerariesV4;
let LoadLatency = 1;
}
//===----------------------------------------------------------------------===//
// Hexagon V4 Resource Definitions -
//===----------------------------------------------------------------------===//