555 lines
20 KiB
Scala
555 lines
20 KiB
Scala
// See LICENSE for license details.
|
|
|
|
package uncore
|
|
import Chisel._
|
|
|
|
class BigMem[T <: Data](n: Int, preLatency: Int, postLatency: Int, leaf: Mem[UInt], noMask: Boolean = false)(gen: => T) extends Module
|
|
{
|
|
class Inputs extends Bundle {
|
|
val addr = UInt(INPUT, log2Up(n))
|
|
val rw = Bool(INPUT)
|
|
val wdata = gen.asInput
|
|
val wmask = gen.asInput
|
|
override def clone = new Inputs().asInstanceOf[this.type]
|
|
}
|
|
val io = new Bundle {
|
|
val in = Valid(new Inputs).flip
|
|
val rdata = gen.asOutput
|
|
}
|
|
val data = gen
|
|
val colMux = if (2*data.getWidth <= leaf.data.getWidth && n > leaf.n) 1 << math.floor(math.log(leaf.data.getWidth/data.getWidth)/math.log(2)).toInt else 1
|
|
val nWide = if (data.getWidth > leaf.data.getWidth) 1+(data.getWidth-1)/leaf.data.getWidth else 1
|
|
val nDeep = if (n > colMux*leaf.n) 1+(n-1)/(colMux*leaf.n) else 1
|
|
if (nDeep > 1 || colMux > 1)
|
|
require(isPow2(n) && isPow2(leaf.n))
|
|
|
|
val rdataDeep = Vec.fill(nDeep){Bits()}
|
|
val rdataSel = Vec.fill(nDeep){Bool()}
|
|
for (i <- 0 until nDeep) {
|
|
val in = Pipe(io.in.valid && (if (nDeep == 1) Bool(true) else UInt(i) === io.in.bits.addr(log2Up(n)-1, log2Up(n/nDeep))), io.in.bits, preLatency)
|
|
val idx = in.bits.addr(log2Up(n/nDeep/colMux)-1, 0)
|
|
val wdata = in.bits.wdata.toBits
|
|
val wmask = in.bits.wmask.toBits
|
|
val ren = in.valid && !in.bits.rw
|
|
val reg_ren = Reg(next=ren)
|
|
val rdata = Vec.fill(nWide){Bits()}
|
|
|
|
val r = Pipe(ren, in.bits.addr, postLatency)
|
|
|
|
for (j <- 0 until nWide) {
|
|
val mem = leaf.clone
|
|
var dout: Bits = null
|
|
val ridx = if (postLatency > 0) Reg(Bits()) else null
|
|
|
|
var wmask0 = Fill(colMux, wmask(math.min(wmask.getWidth, leaf.data.getWidth*(j+1))-1, leaf.data.getWidth*j))
|
|
if (colMux > 1)
|
|
wmask0 = wmask0 & FillInterleaved(gen.getWidth, UIntToOH(in.bits.addr(log2Up(n/nDeep)-1, log2Up(n/nDeep/colMux)), log2Up(colMux)))
|
|
val wdata0 = Fill(colMux, wdata(math.min(wdata.getWidth, leaf.data.getWidth*(j+1))-1, leaf.data.getWidth*j))
|
|
when (in.valid) {
|
|
when (in.bits.rw) {
|
|
if (noMask)
|
|
mem.write(idx, wdata0)
|
|
else
|
|
mem.write(idx, wdata0, wmask0)
|
|
}
|
|
.otherwise { if (postLatency > 0) ridx := idx }
|
|
}
|
|
|
|
if (postLatency == 0) {
|
|
dout = mem(idx)
|
|
} else if (postLatency == 1) {
|
|
dout = mem(ridx)
|
|
} else
|
|
dout = Pipe(reg_ren, mem(ridx), postLatency-1).bits
|
|
|
|
rdata(j) := dout
|
|
}
|
|
val rdataWide = rdata.reduceLeft((x, y) => Cat(y, x))
|
|
|
|
var colMuxOut = rdataWide
|
|
if (colMux > 1) {
|
|
val colMuxIn = Vec((0 until colMux).map(k => rdataWide(gen.getWidth*(k+1)-1, gen.getWidth*k)))
|
|
colMuxOut = colMuxIn(r.bits(log2Up(n/nDeep)-1, log2Up(n/nDeep/colMux)))
|
|
}
|
|
|
|
rdataDeep(i) := colMuxOut
|
|
rdataSel(i) := r.valid
|
|
}
|
|
|
|
io.rdata := Mux1H(rdataSel, rdataDeep)
|
|
}
|
|
|
|
class LLCDataReq(ways: Int) extends MemReqCmd
|
|
{
|
|
val way = UInt(width = log2Up(ways))
|
|
val isWriteback = Bool()
|
|
override def clone = new LLCDataReq(ways).asInstanceOf[this.type]
|
|
}
|
|
|
|
class LLCTagReq(ways: Int) extends HasMemAddr
|
|
{
|
|
val way = UInt(width = log2Up(ways))
|
|
override def clone = new LLCTagReq(ways).asInstanceOf[this.type]
|
|
}
|
|
|
|
class LLCMSHRFile(sets: Int, ways: Int, outstanding: Int, refill_cycles: Int) extends Module
|
|
{
|
|
val io = new Bundle {
|
|
val cpu = Decoupled(new MemReqCmd).flip
|
|
val repl_way = UInt(INPUT, log2Up(ways))
|
|
val repl_dirty = Bool(INPUT)
|
|
val repl_tag = UInt(INPUT, params(MIFAddrBits) - log2Up(sets))
|
|
val data = Decoupled(new LLCDataReq(ways))
|
|
val tag = Decoupled(new LLCTagReq(ways))
|
|
val mem = new MemPipeIO
|
|
val mem_resp_set = UInt(OUTPUT, log2Up(sets))
|
|
val mem_resp_way = UInt(OUTPUT, log2Up(ways))
|
|
}
|
|
|
|
class MSHR extends Bundle {
|
|
val addr = UInt(width = params(MIFAddrBits))
|
|
val way = UInt(width = log2Up(ways))
|
|
val tag = io.cpu.bits.tag.clone
|
|
val refilled = Bool()
|
|
val refillCount = UInt(width = log2Up(refill_cycles))
|
|
val requested = Bool()
|
|
val old_dirty = Bool()
|
|
val old_tag = UInt(width = params(MIFAddrBits) - log2Up(sets))
|
|
val wb_busy = Bool()
|
|
|
|
override def clone = new MSHR().asInstanceOf[this.type]
|
|
}
|
|
|
|
val valid = Vec.fill(outstanding){Reg(init=Bool(false))}
|
|
val validBits = valid.toBits
|
|
val freeId = PriorityEncoder(~validBits)
|
|
val mshr = Vec.fill(outstanding){Reg(new MSHR)}
|
|
when (io.cpu.valid && io.cpu.ready) {
|
|
valid(freeId) := Bool(true)
|
|
mshr(freeId).addr := io.cpu.bits.addr
|
|
mshr(freeId).tag := io.cpu.bits.tag
|
|
mshr(freeId).way := io.repl_way
|
|
mshr(freeId).old_dirty := io.repl_dirty
|
|
mshr(freeId).old_tag := io.repl_tag
|
|
mshr(freeId).wb_busy := Bool(false)
|
|
mshr(freeId).requested := Bool(false)
|
|
mshr(freeId).refillCount := UInt(0)
|
|
mshr(freeId).refilled := Bool(false)
|
|
}
|
|
|
|
val requests = Cat(Bits(0), (outstanding-1 to 0 by -1).map(i => valid(i) && !mshr(i).old_dirty && !mshr(i).wb_busy && !mshr(i).requested):_*)
|
|
val request = requests.orR && io.data.ready // allow in-flight hits to drain
|
|
val requestId = PriorityEncoder(requests)
|
|
when (io.mem.req_cmd.valid && io.mem.req_cmd.ready) { mshr(requestId).requested := Bool(true) }
|
|
|
|
val refillId = io.mem.resp.bits.tag(log2Up(outstanding)-1, 0)
|
|
val refillCount = mshr(refillId).refillCount
|
|
when (io.mem.resp.valid) {
|
|
mshr(refillId).refillCount := refillCount + UInt(1)
|
|
when (refillCount === UInt(refill_cycles-1)) { mshr(refillId).refilled := Bool(true) }
|
|
}
|
|
|
|
val replays = Cat(Bits(0), (outstanding-1 to 0 by -1).map(i => valid(i) && mshr(i).refilled):_*)
|
|
val replay = replays.orR
|
|
val replayId = PriorityEncoder(replays)
|
|
when (replay && io.data.ready && io.tag.ready) { valid(replayId) := Bool(false) }
|
|
|
|
val writebacks = Cat(Bits(0), (outstanding-1 to 0 by -1).map(i => valid(i) && mshr(i).old_dirty):_*)
|
|
val writeback = writebacks.orR
|
|
val writebackId = PriorityEncoder(writebacks)
|
|
when (writeback && io.data.ready && !replay) {
|
|
mshr(writebackId).old_dirty := Bool(false)
|
|
mshr(writebackId).wb_busy := Bool(true)
|
|
}
|
|
mshr.foreach(m => when (m.wb_busy && io.data.ready) { m.wb_busy := Bool(false) })
|
|
|
|
val conflicts = Cat(Bits(0), (0 until outstanding).map(i => valid(i) && io.cpu.bits.addr(log2Up(sets)-1, 0) === mshr(i).addr(log2Up(sets)-1, 0)):_*)
|
|
io.cpu.ready := !conflicts.orR && !validBits.andR
|
|
|
|
io.data.valid := writeback
|
|
io.data.bits.rw := Bool(false)
|
|
io.data.bits.tag := mshr(replayId).tag
|
|
io.data.bits.isWriteback := Bool(true)
|
|
io.data.bits.addr := Cat(mshr(writebackId).old_tag, mshr(writebackId).addr(log2Up(sets)-1, 0)).toUInt
|
|
io.data.bits.way := mshr(writebackId).way
|
|
when (replay) {
|
|
io.data.valid := io.tag.ready
|
|
io.data.bits.isWriteback := Bool(false)
|
|
io.data.bits.addr := mshr(replayId).addr
|
|
io.data.bits.way := mshr(replayId).way
|
|
}
|
|
io.tag.valid := replay && io.data.ready
|
|
io.tag.bits.addr := io.data.bits.addr
|
|
io.tag.bits.way := io.data.bits.way
|
|
|
|
io.mem.req_cmd.valid := request
|
|
io.mem.req_cmd.bits.rw := Bool(false)
|
|
io.mem.req_cmd.bits.addr := mshr(requestId).addr
|
|
io.mem.req_cmd.bits.tag := requestId
|
|
io.mem_resp_set := mshr(refillId).addr
|
|
io.mem_resp_way := mshr(refillId).way
|
|
}
|
|
|
|
class LLCWriteback(requestors: Int, refill_cycles: Int) extends Module
|
|
{
|
|
val io = new Bundle {
|
|
val req = Vec.fill(requestors){Decoupled(UInt(width = params(MIFAddrBits))).flip }
|
|
val data = Vec.fill(requestors){Decoupled(new MemData).flip }
|
|
val mem = new MemPipeIO
|
|
}
|
|
|
|
val valid = Reg(init=Bool(false))
|
|
val who = Reg(UInt())
|
|
val addr = Reg(UInt())
|
|
val cmd_sent = Reg(Bool())
|
|
val data_sent = Reg(Bool())
|
|
val count = Reg(init=UInt(0, log2Up(refill_cycles)))
|
|
|
|
var anyReq = Bool(false)
|
|
for (i <- 0 until requestors) {
|
|
io.req(i).ready := !valid && !anyReq
|
|
io.data(i).ready := valid && who === UInt(i) && io.mem.req_data.ready
|
|
anyReq = anyReq || io.req(i).valid
|
|
}
|
|
|
|
val nextWho = PriorityEncoder(io.req.map(_.valid))
|
|
when (!valid && io.req.map(_.valid).reduceLeft(_||_)) {
|
|
valid := Bool(true)
|
|
cmd_sent := Bool(false)
|
|
data_sent := Bool(false)
|
|
who := nextWho
|
|
addr := io.req(nextWho).bits
|
|
}
|
|
|
|
when (io.mem.req_data.valid && io.mem.req_data.ready) {
|
|
count := count + UInt(1)
|
|
when (count === UInt(refill_cycles-1)) {
|
|
data_sent := Bool(true)
|
|
when (cmd_sent) { valid := Bool(false) }
|
|
}
|
|
}
|
|
when (io.mem.req_cmd.valid && io.mem.req_cmd.ready) { cmd_sent := Bool(true) }
|
|
when (valid && cmd_sent && data_sent) { valid := Bool(false) }
|
|
|
|
io.mem.req_cmd.valid := valid && !cmd_sent
|
|
io.mem.req_cmd.bits.addr := addr
|
|
io.mem.req_cmd.bits.rw := Bool(true)
|
|
|
|
io.mem.req_data.valid := valid && !data_sent && io.data(who).valid
|
|
io.mem.req_data.bits := io.data(who).bits
|
|
}
|
|
|
|
class LLCData(latency: Int, sets: Int, ways: Int, refill_cycles: Int, leaf: Mem[UInt]) extends Module
|
|
{
|
|
val io = new Bundle {
|
|
val req = Decoupled(new LLCDataReq(ways)).flip
|
|
val req_data = Decoupled(new MemData).flip
|
|
val writeback = Decoupled(UInt(width = params(MIFAddrBits)))
|
|
val writeback_data = Decoupled(new MemData)
|
|
val resp = Decoupled(new MemResp)
|
|
val mem_resp = Valid(new MemResp).flip
|
|
val mem_resp_set = UInt(INPUT, log2Up(sets))
|
|
val mem_resp_way = UInt(INPUT, log2Up(ways))
|
|
}
|
|
|
|
val data = Module(new BigMem(sets*ways*refill_cycles, 1, latency-1, leaf, true)(Bits(width = params(MIFDataBits))))
|
|
class QEntry extends MemResp {
|
|
val isWriteback = Bool()
|
|
override def clone = new QEntry().asInstanceOf[this.type]
|
|
}
|
|
val q = Module(new Queue(new QEntry, latency+2))
|
|
val qReady = q.io.count <= UInt(q.entries-latency-1)
|
|
val valid = Reg(init=Bool(false))
|
|
val req = Reg(io.req.bits.clone)
|
|
val count = Reg(init=UInt(0, log2Up(refill_cycles)))
|
|
val refillCount = Reg(init=UInt(0, log2Up(refill_cycles)))
|
|
|
|
when (data.io.in.valid && !io.mem_resp.valid) {
|
|
count := count + UInt(1)
|
|
when (valid && count === UInt(refill_cycles-1)) { valid := Bool(false) }
|
|
}
|
|
when (io.req.valid && io.req.ready) { valid := Bool(true); req := io.req.bits }
|
|
when (io.mem_resp.valid) { refillCount := refillCount + UInt(1) }
|
|
|
|
data.io.in.valid := io.req.valid && io.req.ready && Mux(io.req.bits.rw, io.req_data.valid, qReady)
|
|
data.io.in.bits.addr := Cat(io.req.bits.way, io.req.bits.addr(log2Up(sets)-1, 0), count).toUInt
|
|
data.io.in.bits.rw := io.req.bits.rw
|
|
data.io.in.bits.wdata := io.req_data.bits.data
|
|
when (valid) {
|
|
data.io.in.valid := Mux(req.rw, io.req_data.valid, qReady)
|
|
data.io.in.bits.addr := Cat(req.way, req.addr(log2Up(sets)-1, 0), count).toUInt
|
|
data.io.in.bits.rw := req.rw
|
|
}
|
|
when (io.mem_resp.valid) {
|
|
data.io.in.valid := Bool(true)
|
|
data.io.in.bits.addr := Cat(io.mem_resp_way, io.mem_resp_set, refillCount).toUInt
|
|
data.io.in.bits.rw := Bool(true)
|
|
data.io.in.bits.wdata := io.mem_resp.bits.data
|
|
}
|
|
|
|
val tagPipe = Pipe(data.io.in.valid && !data.io.in.bits.rw, Mux(valid, req.tag, io.req.bits.tag), latency)
|
|
q.io.enq.valid := tagPipe.valid
|
|
q.io.enq.bits.tag := tagPipe.bits
|
|
q.io.enq.bits.isWriteback := Pipe(Mux(valid, req.isWriteback, io.req.bits.isWriteback), Bool(false), latency).valid
|
|
q.io.enq.bits.data := data.io.rdata
|
|
|
|
io.req.ready := !valid && Mux(io.req.bits.isWriteback, io.writeback.ready, Bool(true))
|
|
io.req_data.ready := !io.mem_resp.valid && Mux(valid, req.rw, io.req.valid && io.req.bits.rw)
|
|
|
|
io.writeback.valid := io.req.valid && io.req.ready && io.req.bits.isWriteback
|
|
io.writeback.bits := io.req.bits.addr
|
|
|
|
q.io.deq.ready := Mux(q.io.deq.bits.isWriteback, io.writeback_data.ready, io.resp.ready)
|
|
io.resp.valid := q.io.deq.valid && !q.io.deq.bits.isWriteback
|
|
io.resp.bits := q.io.deq.bits
|
|
io.writeback_data.valid := q.io.deq.valid && q.io.deq.bits.isWriteback
|
|
io.writeback_data.bits := q.io.deq.bits
|
|
}
|
|
|
|
class MemReqArb(n: Int, refill_cycles: Int) extends Module
|
|
{
|
|
val io = new Bundle {
|
|
val cpu = Vec.fill(n){new MemIO().flip}
|
|
val mem = new MemIO
|
|
}
|
|
|
|
val lock = Reg(init=Bool(false))
|
|
val locker = Reg(UInt())
|
|
|
|
val arb = Module(new RRArbiter(new MemReqCmd, n))
|
|
val respWho = io.mem.resp.bits.tag(log2Up(n)-1,0)
|
|
val respTag = io.mem.resp.bits.tag >> UInt(log2Up(n))
|
|
for (i <- 0 until n) {
|
|
val me = UInt(i, log2Up(n))
|
|
arb.io.in(i).valid := io.cpu(i).req_cmd.valid
|
|
arb.io.in(i).bits := io.cpu(i).req_cmd.bits
|
|
arb.io.in(i).bits.tag := Cat(io.cpu(i).req_cmd.bits.tag, me)
|
|
io.cpu(i).req_cmd.ready := arb.io.in(i).ready
|
|
io.cpu(i).req_data.ready := Bool(false)
|
|
|
|
val getLock = io.cpu(i).req_cmd.fire() && io.cpu(i).req_cmd.bits.rw && !lock
|
|
val haveLock = lock && locker === me
|
|
when (getLock) {
|
|
lock := Bool(true)
|
|
locker := UInt(i)
|
|
}
|
|
when (getLock || haveLock) {
|
|
io.cpu(i).req_data.ready := io.mem.req_data.ready
|
|
io.mem.req_data.valid := Bool(true)
|
|
io.mem.req_data.bits := io.cpu(i).req_data.bits
|
|
}
|
|
|
|
io.cpu(i).resp.valid := io.mem.resp.valid && respWho === me
|
|
io.cpu(i).resp.bits := io.mem.resp.bits
|
|
io.cpu(i).resp.bits.tag := respTag
|
|
}
|
|
io.mem.resp.ready := io.cpu(respWho).resp.ready
|
|
|
|
val unlock = Counter(io.mem.req_data.fire(), refill_cycles)._2
|
|
when (unlock) { lock := Bool(false) }
|
|
}
|
|
|
|
abstract class DRAMSideLLCLike extends Module {
|
|
val io = new Bundle {
|
|
val cpu = new MemIO().flip
|
|
val mem = new MemPipeIO
|
|
}
|
|
}
|
|
|
|
// DRAMSideLLC has a known bug now. DO NOT USE. We are working on a brand new
|
|
// L2$. Stay stuned.
|
|
//
|
|
// Bug description:
|
|
// There's a race condition between the writeback unit (the module which
|
|
// sends the data out to the backside interface) and the data unit (the module
|
|
// which writes the data into the SRAM in the L2$). The "hit status"
|
|
// is saved in a register, however, is updated again when there's
|
|
// a transaction coming from the core without waiting until the writeback unit
|
|
// has sent out all its data to the outer memory system. This is why the
|
|
// problem manifests at a higher probability with the slow backup memory port.
|
|
|
|
class DRAMSideLLC_HasKnownBug(sets: Int, ways: Int, outstanding: Int, refill_cycles: Int, tagLeaf: Mem[UInt], dataLeaf: Mem[UInt]) extends DRAMSideLLCLike
|
|
{
|
|
val tagWidth = params(MIFAddrBits) - log2Up(sets)
|
|
val metaWidth = tagWidth + 2 // valid + dirty
|
|
|
|
val memCmdArb = Module(new Arbiter(new MemReqCmd, 2))
|
|
val dataArb = Module(new Arbiter(new LLCDataReq(ways), 2))
|
|
val mshr = Module(new LLCMSHRFile(sets, ways, outstanding, refill_cycles))
|
|
val tags = Module(new BigMem(sets, 0, 1, tagLeaf)(Bits(width = metaWidth*ways)))
|
|
val data = Module(new LLCData(4, sets, ways, refill_cycles, dataLeaf))
|
|
val writeback = Module(new LLCWriteback(2, refill_cycles))
|
|
|
|
val initCount = Reg(init=UInt(0, log2Up(sets+1)))
|
|
val initialize = !initCount(log2Up(sets))
|
|
when (initialize) { initCount := initCount + UInt(1) }
|
|
|
|
val replay_s2 = Reg(init=Bool(false))
|
|
val s2_valid = Reg(init=Bool(false))
|
|
val s2 = Reg(new MemReqCmd)
|
|
val s3_rdy = Bool()
|
|
val replay_s2_rdy = Bool()
|
|
|
|
val s1_valid = Reg(next = io.cpu.req_cmd.fire() || replay_s2 && replay_s2_rdy, init = Bool(false))
|
|
val s1 = Reg(new MemReqCmd)
|
|
when (io.cpu.req_cmd.fire()) { s1 := io.cpu.req_cmd.bits }
|
|
when (replay_s2 && replay_s2_rdy) { s1 := s2 }
|
|
|
|
s2_valid := s1_valid
|
|
replay_s2 := s2_valid && !s3_rdy || replay_s2 && !replay_s2_rdy
|
|
val s2_tags = Vec.fill(ways){Reg(Bits(width = metaWidth))}
|
|
when (s1_valid) {
|
|
s2 := s1
|
|
for (i <- 0 until ways)
|
|
s2_tags(i) := tags.io.rdata(metaWidth*(i+1)-1, metaWidth*i)
|
|
}
|
|
val s2_hits = s2_tags.map(t => t(tagWidth) && s2.addr(s2.addr.getWidth-1, s2.addr.getWidth-tagWidth) === t(tagWidth-1, 0))
|
|
val s2_hit_way = OHToUInt(s2_hits)
|
|
val s2_hit = s2_hits.reduceLeft(_||_)
|
|
val s2_hit_dirty = s2_tags(s2_hit_way)(tagWidth+1)
|
|
val repl_way = LFSR16(s2_valid)(log2Up(ways)-1, 0)
|
|
val repl_tag = s2_tags(repl_way).toUInt
|
|
val setDirty = s2_valid && s2.rw && s2_hit && !s2_hit_dirty
|
|
|
|
val tag_we = initialize || setDirty || mshr.io.tag.fire()
|
|
val tag_waddr = Mux(initialize, initCount, Mux(setDirty, s2.addr, mshr.io.tag.bits.addr))
|
|
val tag_wdata = Cat(setDirty, !initialize, Mux(setDirty, s2.addr, mshr.io.tag.bits.addr)(mshr.io.tag.bits.addr.getWidth-1, mshr.io.tag.bits.addr.getWidth-tagWidth))
|
|
val tag_wmask = Mux(initialize, SInt(-1, ways), UIntToOH(Mux(setDirty, s2_hit_way, mshr.io.tag.bits.way)))
|
|
tags.io.in.valid := io.cpu.req_cmd.fire() || replay_s2 && replay_s2_rdy || tag_we
|
|
tags.io.in.bits.addr := Mux(tag_we, tag_waddr, Mux(replay_s2, s2.addr, io.cpu.req_cmd.bits.addr)(log2Up(sets)-1,0))
|
|
tags.io.in.bits.rw := tag_we
|
|
tags.io.in.bits.wdata := Fill(ways, tag_wdata)
|
|
tags.io.in.bits.wmask := FillInterleaved(metaWidth, tag_wmask)
|
|
|
|
mshr.io.cpu.valid := s2_valid && !s2_hit && !s2.rw
|
|
mshr.io.cpu.bits := s2
|
|
mshr.io.repl_way := repl_way
|
|
mshr.io.repl_dirty := repl_tag(tagWidth+1, tagWidth).andR
|
|
mshr.io.repl_tag := repl_tag
|
|
mshr.io.mem.resp := io.mem.resp
|
|
mshr.io.tag.ready := !s1_valid && !s2_valid
|
|
|
|
data.io.req <> dataArb.io.out
|
|
data.io.mem_resp := io.mem.resp
|
|
data.io.mem_resp_set := mshr.io.mem_resp_set
|
|
data.io.mem_resp_way := mshr.io.mem_resp_way
|
|
data.io.req_data.bits := io.cpu.req_data.bits
|
|
data.io.req_data.valid := io.cpu.req_data.valid
|
|
|
|
writeback.io.req(0) <> data.io.writeback
|
|
writeback.io.data(0) <> data.io.writeback_data
|
|
writeback.io.req(1).valid := s2_valid && !s2_hit && s2.rw && mshr.io.cpu.ready
|
|
writeback.io.req(1).bits := s2.addr
|
|
writeback.io.data(1).valid := io.cpu.req_data.valid
|
|
writeback.io.data(1).bits := io.cpu.req_data.bits
|
|
|
|
memCmdArb.io.in(0) <> mshr.io.mem.req_cmd
|
|
memCmdArb.io.in(1) <> writeback.io.mem.req_cmd
|
|
|
|
dataArb.io.in(0) <> mshr.io.data
|
|
dataArb.io.in(1).valid := s2_valid && s2_hit && mshr.io.cpu.ready
|
|
dataArb.io.in(1).bits := s2
|
|
dataArb.io.in(1).bits.way := s2_hit_way
|
|
dataArb.io.in(1).bits.isWriteback := Bool(false)
|
|
|
|
s3_rdy := mshr.io.cpu.ready && Mux(s2_hit, dataArb.io.in(1).ready, !s2.rw || writeback.io.req(1).ready)
|
|
replay_s2_rdy := s3_rdy && !tag_we
|
|
|
|
io.cpu.resp <> data.io.resp
|
|
io.cpu.req_data.ready := writeback.io.data(1).ready || data.io.req_data.ready
|
|
io.mem.req_cmd <> memCmdArb.io.out
|
|
io.mem.req_data <> writeback.io.mem.req_data
|
|
io.cpu.req_cmd.ready := !(s1_valid || s2_valid || replay_s2 || tag_we ||
|
|
io.cpu.req_cmd.bits.rw && io.cpu.req_data.ready)
|
|
}
|
|
|
|
class HellaFlowQueue[T <: Data](val entries: Int)(data: => T) extends Module
|
|
{
|
|
val io = new QueueIO(data, entries)
|
|
require(isPow2(entries) && entries > 1)
|
|
|
|
val do_flow = Bool()
|
|
val do_enq = io.enq.fire() && !do_flow
|
|
val do_deq = io.deq.fire() && !do_flow
|
|
|
|
val maybe_full = Reg(init=Bool(false))
|
|
val enq_ptr = Counter(do_enq, entries)._1
|
|
val deq_ptr = Counter(do_deq, entries)._1
|
|
when (do_enq != do_deq) { maybe_full := do_enq }
|
|
|
|
val ptr_match = enq_ptr === deq_ptr
|
|
val empty = ptr_match && !maybe_full
|
|
val full = ptr_match && maybe_full
|
|
val atLeastTwo = full || enq_ptr - deq_ptr >= UInt(2)
|
|
do_flow := empty && io.deq.ready
|
|
|
|
val ram = Mem(data, entries, seqRead = true)
|
|
val ram_addr = Reg(Bits())
|
|
val ram_out_valid = Reg(Bool())
|
|
ram_out_valid := Bool(false)
|
|
when (do_enq) { ram(enq_ptr) := io.enq.bits }
|
|
when (io.deq.ready && (atLeastTwo || !io.deq.valid && !empty)) {
|
|
ram_out_valid := Bool(true)
|
|
ram_addr := Mux(io.deq.valid, deq_ptr + UInt(1), deq_ptr)
|
|
}
|
|
|
|
io.deq.valid := Mux(empty, io.enq.valid, ram_out_valid)
|
|
io.enq.ready := !full
|
|
io.deq.bits := Mux(empty, io.enq.bits, ram(ram_addr))
|
|
}
|
|
|
|
class HellaQueue[T <: Data](val entries: Int)(data: => T) extends Module
|
|
{
|
|
val io = new QueueIO(data, entries)
|
|
|
|
val fq = Module(new HellaFlowQueue(entries)(data))
|
|
io.enq <> fq.io.enq
|
|
io.deq <> Queue(fq.io.deq, 1, pipe = true)
|
|
}
|
|
|
|
object HellaQueue
|
|
{
|
|
def apply[T <: Data](enq: DecoupledIO[T], entries: Int) = {
|
|
val q = Module((new HellaQueue(entries)) { enq.bits.clone })
|
|
q.io.enq.valid := enq.valid // not using <> so that override is allowed
|
|
q.io.enq.bits := enq.bits
|
|
enq.ready := q.io.enq.ready
|
|
q.io.deq
|
|
}
|
|
}
|
|
|
|
class DRAMSideLLCNull(numRequests: Int, refillCycles: Int) extends DRAMSideLLCLike {
|
|
val numEntries = numRequests * refillCycles
|
|
val size = log2Down(numEntries) + 1
|
|
|
|
val inc = Bool()
|
|
val dec = Bool()
|
|
val count = Reg(init=UInt(numEntries, size))
|
|
val watermark = count >= UInt(refillCycles)
|
|
|
|
when (inc && !dec) {
|
|
count := count + UInt(1)
|
|
}
|
|
when (!inc && dec) {
|
|
count := count - UInt(refillCycles)
|
|
}
|
|
when (inc && dec) {
|
|
count := count - UInt(refillCycles-1)
|
|
}
|
|
|
|
val cmdq_mask = io.cpu.req_cmd.bits.rw || watermark
|
|
|
|
io.mem.req_cmd.valid := io.cpu.req_cmd.valid && cmdq_mask
|
|
io.cpu.req_cmd.ready := io.mem.req_cmd.ready && cmdq_mask
|
|
io.mem.req_cmd.bits := io.cpu.req_cmd.bits
|
|
|
|
io.mem.req_data <> io.cpu.req_data
|
|
|
|
val resp_dataq = Module((new HellaQueue(numEntries)) { new MemResp })
|
|
resp_dataq.io.enq <> io.mem.resp
|
|
io.cpu.resp <> resp_dataq.io.deq
|
|
|
|
inc := resp_dataq.io.deq.fire()
|
|
dec := io.mem.req_cmd.fire() && !io.mem.req_cmd.bits.rw
|
|
}
|