riscv/rocket-chip
1
0
Fork 0
Code Releases Activity

Shave a few gate delays off IBuf control logic

Browse Source 
It takes a while for the pipeline to compute the stall signal, so avoid
using it until the last logic levels in the clock cycle.
This commit is contained in:
Andrew Waterman 2017-08-04 00:39:10 -07:00
parent 82ff81e40d
commit ecc2ee366c
1 changed files with 20 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

39
src/main/scala/rocket/IBuf.scala
View File

@ -43,24 +43,24 @@ class IBuf(implicit p: Parameters) extends CoreModule {
val nIC = Mux(io.imem.bits.btb.valid && io.imem.bits.btb.bits.taken, io.imem.bits.btb.bits.bridx +& 1, UInt(fetchWidth)) - pcWordBits
val nICReady = nReady - nBufValid
val nValid = Mux(io.imem.valid, nIC, UInt(0)) + nBufValid
io.imem.ready := nReady >= nBufValid && (nICReady >= nIC || n >= nIC - nICReady)
io.imem.ready := io.inst(0).ready && nReady >= nBufValid && (nICReady >= nIC || n >= nIC - nICReady)
if (n > 0) {
nBufValid := Mux(nReady >= nBufValid, UInt(0), nBufValid - nReady)
if (n > 1) when (nReady > 0 && nReady < nBufValid) {
val shiftedBuf = shiftInsnRight(buf.data(n*coreInstBits-1, coreInstBits), (nReady-1)(log2Ceil(n-1)-1,0))
buf.data := Cat(buf.data(n*coreInstBits-1, (n-1)*coreInstBits), shiftedBuf((n-1)*coreInstBits-1, 0))
buf.pc := buf.pc & ~pcWordMask | (buf.pc + (nReady << log2Ceil(coreInstBytes))) & pcWordMask
ibufBTBResp.bridx := ibufBTBResp.bridx - nReady
}
when (io.imem.valid && nReady >= nBufValid && nICReady < nIC && n >= nIC - nICReady) {
val shamt = pcWordBits + nICReady
nBufValid := nIC - nICReady
buf := io.imem.bits
buf.data := shiftInsnRight(io.imem.bits.data, shamt)(n*coreInstBits-1,0)
buf.pc := io.imem.bits.pc & ~pcWordMask | (io.imem.bits.pc + (nICReady << log2Ceil(coreInstBytes))) & pcWordMask
ibufBTBHit := io.imem.bits.btb.valid
when (io.imem.bits.btb.valid) {
when (io.inst(0).ready) {
nBufValid := Mux(nReady >= nBufValid, UInt(0), nBufValid - nReady)
if (n > 1) when (nReady > 0 && nReady < nBufValid) {
val shiftedBuf = shiftInsnRight(buf.data(n*coreInstBits-1, coreInstBits), (nReady-1)(log2Ceil(n-1)-1,0))
buf.data := Cat(buf.data(n*coreInstBits-1, (n-1)*coreInstBits), shiftedBuf((n-1)*coreInstBits-1, 0))
buf.pc := buf.pc & ~pcWordMask | (buf.pc + (nReady << log2Ceil(coreInstBytes))) & pcWordMask
ibufBTBResp.bridx := ibufBTBResp.bridx - nReady
}
when (io.imem.valid && nReady >= nBufValid && nICReady < nIC && n >= nIC - nICReady) {
val shamt = pcWordBits + nICReady
nBufValid := nIC - nICReady
buf := io.imem.bits
buf.data := shiftInsnRight(io.imem.bits.data, shamt)(n*coreInstBits-1,0)
buf.pc := io.imem.bits.pc & ~pcWordMask | (io.imem.bits.pc + (nICReady << log2Ceil(coreInstBytes))) & pcWordMask
ibufBTBHit := io.imem.bits.btb.valid
ibufBTBResp := io.imem.bits.btb.bits
ibufBTBResp.bridx := io.imem.bits.btb.bits.bridx + nICReady
}
@ -97,18 +97,19 @@ class IBuf(implicit p: Parameters) extends CoreModule {
if (usingCompressed) {
val replay = ic_replay(j) || (!exp.io.rvc && (btbHitMask(j) || ic_replay(j+1)))
io.inst(i).valid := valid(j) && (exp.io.rvc || valid(j+1) || xcpt(j+1).asUInt.orR || replay)
val full_insn = exp.io.rvc || valid(j+1) || xcpt(j+1).asUInt.orR || replay
io.inst(i).valid := valid(j) && full_insn
io.inst(i).bits.xcpt0 := xcpt(j)
io.inst(i).bits.xcpt1 := Mux(exp.io.rvc, 0.U, xcpt(j+1).asUInt).asTypeOf(new FrontendExceptions)
io.inst(i).bits.replay := replay
io.inst(i).bits.btb_hit := btbHitMask(j) || (!exp.io.rvc && btbHitMask(j+1))
io.inst(i).bits.rvc := exp.io.rvc
when (io.inst(i).fire()) { nReady := Mux(exp.io.rvc, j+1, j+2) }
when (full_insn && (i == 0 || io.inst(i).ready)) { nReady := Mux(exp.io.rvc, j+1, j+2) }
expand(i+1, Mux(exp.io.rvc, j+1, j+2), Mux(exp.io.rvc, curInst >> 16, curInst >> 32))
} else {
when (io.inst(i).ready) { nReady := i+1 }
when (i == 0 || io.inst(i).ready) { nReady := i+1 }
io.inst(i).valid := valid(i)
io.inst(i).bits.xcpt0 := xcpt(i)
io.inst(i).bits.xcpt1 := 0.U.asTypeOf(new FrontendExceptions)