yystopf
/
riscv-lab
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

修改bpu

This commit is contained in:
Liphen 2023-11-20 14:44:23 +08:00
parent 79b8387218
commit bb3942d119
4 changed files with 140 additions and 178 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
chisel/playground/src/Core.scala
View File

@ -67,9 +67,6 @@
// bpu.decoder.pht_index := decoderUnit.bpu.pht_index // bpu.decoder.pht_index := decoderUnit.bpu.pht_index
// decoderUnit.bpu.update_pht_index := bpu.decoder.update_pht_index // decoderUnit.bpu.update_pht_index := bpu.decoder.update_pht_index
// bpu.execute <> executeUnit.bpu // bpu.execute <> executeUnit.bpu
// if (config.branchPredictor == "pesudo") {
// bpu.regfile.get <> regfile.bpu.get
// }
// decoderUnit.bpu.branch_inst := bpu.decoder.branch_inst // decoderUnit.bpu.branch_inst := bpu.decoder.branch_inst
// decoderUnit.bpu.pred_branch := bpu.decoder.pred_branch // decoderUnit.bpu.pred_branch := bpu.decoder.pred_branch
// decoderUnit.bpu.branch_target := bpu.decoder.branch_target // decoderUnit.bpu.branch_target := bpu.decoder.branch_target

2
chisel/playground/src/CpuConfig.scala
View File

@ -19,7 +19,7 @@ case class CpuConfig(
val instFifoDepth: Int = 8, // 指令缓存深度 val instFifoDepth: Int = 8, // 指令缓存深度
val mulClockNum: Int = 2, // 乘法器的时钟周期数 val mulClockNum: Int = 2, // 乘法器的时钟周期数
val divClockNum: Int = 8, // 除法器的时钟周期数 val divClockNum: Int = 8, // 除法器的时钟周期数
val branchPredictor: String = "adaptive" // adaptive, pesudo, global val branchPredictor: String = "adaptive" // adaptive, global
) )
case class BranchPredictorConfig( case class BranchPredictorConfig(

310
chisel/playground/src/pipeline/fetch/BranchPredictorUnit.scala
View File

@ -1,199 +1,163 @@
// package cpu.pipeline.fetch package cpu.pipeline.fetch
// import chisel3._ import chisel3._
// import chisel3.util._ import chisel3.util._
// import cpu.defines.Const._ import cpu.defines.Const._
// import cpu._ import cpu._
// import cpu.pipeline.decoder.Src12Read import cpu.pipeline.decoder.Src12Read
import cpu.defines.ALUOpType
import cpu.defines.FuOpType
// class ExecuteUnitBranchPredictor extends Bundle { class ExecuteUnitBranchPredictor extends Bundle {
// val bpuConfig = new BranchPredictorConfig() val bpuConfig = new BranchPredictorConfig()
// val pc = Output(UInt(PC_WID.W)) val pc = Output(UInt(PC_WID.W))
// val update_pht_index = Output(UInt(bpuConfig.phtDepth.W)) val update_pht_index = Output(UInt(bpuConfig.phtDepth.W))
// val branch_inst = Output(Bool()) val branch_inst = Output(Bool())
// val branch = Output(Bool()) val branch = Output(Bool())
// } }
// class BranchPredictorIO(implicit config: CpuConfig) extends Bundle { class BranchPredictorIO(implicit config: CpuConfig) extends Bundle {
// val bpuConfig = new BranchPredictorConfig() val bpuConfig = new BranchPredictorConfig()
// val decoder = new Bundle { val decoder = new Bundle {
// val inst = Input(UInt(INST_WID.W)) val inst = Input(UInt(INST_WID.W))
// val op = Input(UInt(OP_WID.W)) val op = Input(FuOpType())
// val ena = Input(Bool()) val ena = Input(Bool())
// val pc = Input(UInt(PC_WID.W)) val pc = Input(UInt(PC_WID.W))
// val pc_plus4 = Input(UInt(PC_WID.W)) val pc_plus4 = Input(UInt(PC_WID.W))
// val pht_index = Input(UInt(bpuConfig.phtDepth.W)) val pht_index = Input(UInt(bpuConfig.phtDepth.W))
// val rs1 = Input(UInt(REG_ADDR_WID.W)) val rs1 = Input(UInt(REG_ADDR_WID.W))
// val rs2 = Input(UInt(REG_ADDR_WID.W)) val rs2 = Input(UInt(REG_ADDR_WID.W))
// val branch_inst = Output(Bool()) val branch_inst = Output(Bool())
// val pred_branch = Output(Bool()) val pred_branch = Output(Bool())
// val branch_target = Output(UInt(PC_WID.W)) val branch_target = Output(UInt(PC_WID.W))
// val update_pht_index = Output(UInt(bpuConfig.phtDepth.W)) val update_pht_index = Output(UInt(bpuConfig.phtDepth.W))
// } }
// val instBuffer = new Bundle { val instBuffer = new Bundle {
// val pc = Input(Vec(config.instFetchNum, UInt(PC_WID.W))) val pc = Input(Vec(config.instFetchNum, UInt(PC_WID.W)))
// val pht_index = Output(Vec(config.instFetchNum, UInt(bpuConfig.phtDepth.W))) val pht_index = Output(Vec(config.instFetchNum, UInt(bpuConfig.phtDepth.W)))
// } }
// val execute = Flipped(new ExecuteUnitBranchPredictor()) val execute = Flipped(new ExecuteUnitBranchPredictor())
}
// val regfile = if (config.branchPredictor == "pesudo") Some(new Src12Read()) else None class BranchPredictorUnit(implicit config: CpuConfig) extends Module {
// } val io = IO(new BranchPredictorIO())
// class BranchPredictorUnit(implicit config: CpuConfig) extends Module { if (config.branchPredictor == "adaptive") {
// val io = IO(new BranchPredictorIO()) val adaptive_predictor = Module(new AdaptiveTwoLevelPredictor())
io <> adaptive_predictor.io
}
// if (config.branchPredictor == "adaptive") { if (config.branchPredictor == "global") {
// val adaptive_predictor = Module(new AdaptiveTwoLevelPredictor()) val global_predictor = Module(new GlobalBranchPredictor())
// io <> adaptive_predictor.io io <> global_predictor.io
// } }
}
// if (config.branchPredictor == "pesudo") { class GlobalBranchPredictor(
// val pesudo_predictor = Module(new PesudoBranchPredictor()) GHR_DEPTH: Int = 4, // 可以记录的历史记录个数
// io <> pesudo_predictor.io PC_HASH_WID: Int = 4, // 取得PC的宽度
// } PHT_DEPTH: Int = 6, // 可以记录的历史个数
BHT_DEPTH: Int = 4 // 取得PC的宽度
)(
implicit
config: CpuConfig)
extends Module {
val io = IO(new BranchPredictorIO())
// if (config.branchPredictor == "global") { val strongly_not_taken :: weakly_not_taken :: weakly_taken :: strongly_taken :: Nil = Enum(4)
// val global_predictor = Module(new GlobalBranchPredictor())
// io <> global_predictor.io
// }
// }
// class PesudoBranchPredictor(implicit config: CpuConfig) extends Module { io.decoder.branch_inst := ALUOpType.isBru(io.decoder.op) && ALUOpType.isBranch(io.decoder.op)
// val io = IO(new BranchPredictorIO()) io.decoder.branch_target := io.decoder.pc_plus4 + Cat(
// io.decoder.branch_inst := VecInit(EXE_BEQ, EXE_BNE, EXE_BGTZ, EXE_BLEZ, EXE_BGEZ, EXE_BGEZAL, EXE_BLTZ, EXE_BLTZAL) Fill(14, io.decoder.inst(15)),
// .contains(io.decoder.op) io.decoder.inst(15, 0),
// io.decoder.branch_target := io.decoder.pc_plus4 + Cat( 0.U(2.W)
// Fill(14, io.decoder.inst(15)), )
// io.decoder.inst(15, 0), // 局部预测模式
// 0.U(2.W)
// )
// io.regfile.get.src1.raddr := io.decoder.rs1 val bht = RegInit(VecInit(Seq.fill(1 << BHT_DEPTH)(0.U(PHT_DEPTH.W))))
// io.regfile.get.src2.raddr := io.decoder.rs2 val pht = RegInit(VecInit(Seq.fill(1 << PHT_DEPTH)(strongly_taken)))
// val (src1, src2) = (io.regfile.get.src1.rdata, io.regfile.get.src2.rdata) val bht_index = io.decoder.pc(1 + BHT_DEPTH, 2)
// val pred_branch = MuxLookup(io.decoder.op, false.B)( val pht_index = bht(bht_index)
// Seq(
// EXE_BEQ -> (src1 === src2),
// EXE_BNE -> (src1 =/= src2),
// EXE_BGTZ -> (!src1(31) && (src1 =/= 0.U)),
// EXE_BLEZ -> (src1(31) || src1 === 0.U),
// EXE_BGEZ -> (!src1(31)),
// EXE_BGEZAL -> (!src1(31)),
// EXE_BLTZ -> (src1(31)),
// EXE_BLTZAL -> (src1(31))
// )
// )
// io.decoder.pred_branch := io.decoder.ena && io.decoder.branch_inst && pred_branch io.decoder.pred_branch :=
// } io.decoder.ena && io.decoder.branch_inst && (pht(pht_index) === weakly_taken || pht(pht_index) === strongly_taken)
val update_bht_index = io.execute.pc(1 + BHT_DEPTH, 2)
val update_pht_index = bht(update_bht_index)
// class GlobalBranchPredictor( when(io.execute.branch_inst) {
// GHR_DEPTH: Int = 4, // 可以记录的历史记录个数 bht(update_bht_index) := Cat(bht(update_bht_index)(PHT_DEPTH - 2, 0), io.execute.branch)
// PC_HASH_WID: Int = 4, // 取得PC的宽度 switch(pht(update_pht_index)) {
// PHT_DEPTH: Int = 6, // 可以记录的历史个数 is(strongly_not_taken) {
// BHT_DEPTH: Int = 4 // 取得PC的宽度 pht(update_pht_index) := Mux(io.execute.branch, weakly_not_taken, strongly_not_taken)
// )( }
// implicit is(weakly_not_taken) {
// config: CpuConfig) pht(update_pht_index) := Mux(io.execute.branch, weakly_taken, strongly_not_taken)
// extends Module { }
// val io = IO(new BranchPredictorIO()) is(weakly_taken) {
pht(update_pht_index) := Mux(io.execute.branch, strongly_taken, weakly_not_taken)
}
is(strongly_taken) {
pht(update_pht_index) := Mux(io.execute.branch, strongly_taken, weakly_taken)
}
}
}
// val strongly_not_taken :: weakly_not_taken :: weakly_taken :: strongly_taken :: Nil = Enum(4) }
// io.decoder.branch_inst := VecInit(EXE_BEQ, EXE_BNE, EXE_BGTZ, EXE_BLEZ, EXE_BGEZ, EXE_BGEZAL, EXE_BLTZ, EXE_BLTZAL) class AdaptiveTwoLevelPredictor(
// .contains(io.decoder.op) )(
// io.decoder.branch_target := io.decoder.pc_plus4 + Cat( implicit
// Fill(14, io.decoder.inst(15)), config: CpuConfig)
// io.decoder.inst(15, 0), extends Module {
// 0.U(2.W) val bpuConfig = new BranchPredictorConfig()
// ) val PHT_DEPTH = bpuConfig.phtDepth
// // 局部预测模式 val BHT_DEPTH = bpuConfig.bhtDepth
val io = IO(new BranchPredictorIO())
// val bht = RegInit(VecInit(Seq.fill(1 << BHT_DEPTH)(0.U(PHT_DEPTH.W)))) val strongly_not_taken :: weakly_not_taken :: weakly_taken :: strongly_taken :: Nil = Enum(4)
// val pht = RegInit(VecInit(Seq.fill(1 << PHT_DEPTH)(strongly_taken)))
// val bht_index = io.decoder.pc(1 + BHT_DEPTH, 2)
// val pht_index = bht(bht_index)
// io.decoder.pred_branch := io.decoder.branch_inst := ALUOpType.isBru(io.decoder.op) && ALUOpType.isBranch(io.decoder.op)
// io.decoder.ena && io.decoder.branch_inst && (pht(pht_index) === weakly_taken || pht(pht_index) === strongly_taken) io.decoder.branch_target := io.decoder.pc_plus4 + Cat(
// val update_bht_index = io.execute.pc(1 + BHT_DEPTH, 2) Fill(14, io.decoder.inst(15)),
// val update_pht_index = bht(update_bht_index) io.decoder.inst(15, 0),
0.U(2.W)
)
// when(io.execute.branch_inst) { val bht = RegInit(VecInit(Seq.fill(1 << BHT_DEPTH)(0.U(PHT_DEPTH.W))))
// bht(update_bht_index) := Cat(bht(update_bht_index)(PHT_DEPTH - 2, 0), io.execute.branch) val pht = RegInit(VecInit(Seq.fill(1 << PHT_DEPTH)(strongly_taken)))
// switch(pht(update_pht_index)) { val pht_index = io.decoder.pht_index
// is(strongly_not_taken) {
// pht(update_pht_index) := Mux(io.execute.branch, weakly_not_taken, strongly_not_taken)
// }
// is(weakly_not_taken) {
// pht(update_pht_index) := Mux(io.execute.branch, weakly_taken, strongly_not_taken)
// }
// is(weakly_taken) {
// pht(update_pht_index) := Mux(io.execute.branch, strongly_taken, weakly_not_taken)
// }
// is(strongly_taken) {
// pht(update_pht_index) := Mux(io.execute.branch, strongly_taken, weakly_taken)
// }
// }
// }
// } for (i <- 0 until config.instFetchNum) {
io.instBuffer.pht_index(i) := bht(io.instBuffer.pc(i)(1 + BHT_DEPTH, 2))
}
// class AdaptiveTwoLevelPredictor( io.decoder.pred_branch :=
// )( io.decoder.ena && io.decoder.branch_inst && (pht(pht_index) === weakly_taken || pht(pht_index) === strongly_taken)
// implicit io.decoder.update_pht_index := bht(io.decoder.pc(1 + BHT_DEPTH, 2))
// config: CpuConfig)
// extends Module {
// val bpuConfig = new BranchPredictorConfig()
// val PHT_DEPTH = bpuConfig.phtDepth
// val BHT_DEPTH = bpuConfig.bhtDepth
// val io = IO(new BranchPredictorIO())
// val strongly_not_taken :: weakly_not_taken :: weakly_taken :: strongly_taken :: Nil = Enum(4) val update_bht_index = io.execute.pc(1 + BHT_DEPTH, 2)
val update_pht_index = io.execute.update_pht_index
// io.decoder.branch_inst := when(io.execute.branch_inst) {
// VecInit(EXE_BEQ, EXE_BNE, EXE_BGTZ, EXE_BLEZ, EXE_BGEZ, EXE_BGEZAL, EXE_BLTZ, EXE_BLTZAL).contains(io.decoder.op) bht(update_bht_index) := Cat(bht(update_bht_index)(PHT_DEPTH - 2, 0), io.execute.branch)
// io.decoder.branch_target := io.decoder.pc_plus4 + Cat( switch(pht(update_pht_index)) {
// Fill(14, io.decoder.inst(15)), is(strongly_not_taken) {
// io.decoder.inst(15, 0), pht(update_pht_index) := Mux(io.execute.branch, weakly_not_taken, strongly_not_taken)
// 0.U(2.W) }
// ) is(weakly_not_taken) {
pht(update_pht_index) := Mux(io.execute.branch, weakly_taken, strongly_not_taken)
}
is(weakly_taken) {
pht(update_pht_index) := Mux(io.execute.branch, strongly_taken, weakly_not_taken)
}
is(strongly_taken) {
pht(update_pht_index) := Mux(io.execute.branch, strongly_taken, weakly_taken)
}
}
}
// val bht = RegInit(VecInit(Seq.fill(1 << BHT_DEPTH)(0.U(PHT_DEPTH.W)))) }
// val pht = RegInit(VecInit(Seq.fill(1 << PHT_DEPTH)(strongly_taken)))
// val pht_index = io.decoder.pht_index
// for (i <- 0 until config.instFetchNum) {
// io.instBuffer.pht_index(i) := bht(io.instBuffer.pc(i)(1 + BHT_DEPTH, 2))
// }
// io.decoder.pred_branch :=
// io.decoder.ena && io.decoder.branch_inst && (pht(pht_index) === weakly_taken || pht(pht_index) === strongly_taken)
// io.decoder.update_pht_index := bht(io.decoder.pc(1 + BHT_DEPTH, 2))
// val update_bht_index = io.execute.pc(1 + BHT_DEPTH, 2)
// val update_pht_index = io.execute.update_pht_index
// when(io.execute.branch_inst) {
// bht(update_bht_index) := Cat(bht(update_bht_index)(PHT_DEPTH - 2, 0), io.execute.branch)
// switch(pht(update_pht_index)) {
// is(strongly_not_taken) {
// pht(update_pht_index) := Mux(io.execute.branch, weakly_not_taken, strongly_not_taken)
// }
// is(weakly_not_taken) {
// pht(update_pht_index) := Mux(io.execute.branch, weakly_taken, strongly_not_taken)
// }
// is(weakly_taken) {
// pht(update_pht_index) := Mux(io.execute.branch, strongly_taken, weakly_not_taken)
// }
// is(strongly_taken) {
// pht(update_pht_index) := Mux(io.execute.branch, strongly_taken, weakly_taken)
// }
// }
// }
// }

3
chisel/playground/test/src/TestMain.scala
View File

@ -4,10 +4,11 @@ import cache.Cache
import cpu.pipeline.decoder.Decoder import cpu.pipeline.decoder.Decoder
import cpu.pipeline.decoder.DecoderUnit import cpu.pipeline.decoder.DecoderUnit
import cache.ICache import cache.ICache
import cpu.pipeline.fetch.BranchPredictorUnit
object TestMain extends App { object TestMain extends App {
implicit val config = new CpuConfig() implicit val config = new CpuConfig()
def top = new Cache() def top = new BranchPredictorUnit()
val useMFC = false // use MLIR-based firrtl compiler val useMFC = false // use MLIR-based firrtl compiler
val generator = Seq(chisel3.stage.ChiselGeneratorAnnotation(() => top)) val generator = Seq(chisel3.stage.ChiselGeneratorAnnotation(() => top))
if (useMFC) { if (useMFC) {