Hardware
/
VexRiscv
mirror of https://github.com/SpinalHDL/VexRiscv.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fpu improve FMax and add asyncronus regfile support

This commit is contained in:
Dolu1990 2021-03-16 14:45:59 +01:00
parent 0d628b4706
commit 02c572b6f1
2 changed files with 60 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

89
src/main/scala/vexriscv/ip/fpu/FpuCore.scala
View File

@ -278,36 +278,38 @@ case class FpuCore( portCount : Int, p : FpuParameter) extends Component{
val s0 = cmdArbiter.output.pipelined() //TODO may need to remove m2s for store latency val s0 = cmdArbiter.output.pipelined() //TODO may need to remove m2s for store latency
val s1 = s0.m2sPipe() val s1 = s0.m2sPipe()
val output = s1.swapPayload(RfReadOutput()) val output = s1.swapPayload(RfReadOutput())
val rs1Entry = rf.ram.readSync(s0.source @@ s0.rs1,enable = !output.isStall) val rs = if(p.asyncRegFile){
val rs2Entry = rf.ram.readSync(s0.source @@ s0.rs2,enable = !output.isStall) List(s1.rs1, s1.rs2, s1.rs3).map(a => rf.ram.readAsync(s1.source @@ a))
val rs3Entry = rf.ram.readSync(s0.source @@ s0.rs3,enable = !output.isStall) } else {
List(s0.rs1, s0.rs2, s0.rs3).map(a => rf.ram.readSync(s0.source @@ a, enable = !output.isStall))
}
output.source := s1.source output.source := s1.source
output.opcode := s1.opcode output.opcode := s1.opcode
output.arg := s1.arg output.arg := s1.arg
output.roundMode := s1.roundMode output.roundMode := s1.roundMode
output.rd := s1.rd output.rd := s1.rd
output.rs1 := rs1Entry.value output.rs1 := rs(0).value
output.rs2 := rs2Entry.value output.rs2 := rs(1).value
output.rs3 := rs3Entry.value output.rs3 := rs(2).value
if(p.withDouble){ if(p.withDouble){
output.rs1Boxed := rs1Entry.boxed output.rs1Boxed := rs(0).boxed
output.rs2Boxed := rs2Entry.boxed output.rs2Boxed := rs(1).boxed
output.format := s1.format output.format := s1.format
val store = s1.opcode === FpuOpcode.STORE ||s1.opcode === FpuOpcode.FMV_X_W val store = s1.opcode === FpuOpcode.STORE ||s1.opcode === FpuOpcode.FMV_X_W
val sgnjBypass = s1.opcode === FpuOpcode.SGNJ && s1.format === FpuFormat.DOUBLE val sgnjBypass = s1.opcode === FpuOpcode.SGNJ && s1.format === FpuFormat.DOUBLE
when(!sgnjBypass) { when(!sgnjBypass) {
when(store) { //Pass through when(store) { //Pass through
output.format := rs1Entry.boxed ? FpuFormat.FLOAT | FpuFormat.DOUBLE output.format := rs(0).boxed ? FpuFormat.FLOAT | FpuFormat.DOUBLE
} elsewhen (s1.format === FpuFormat.FLOAT =/= rs1Entry.boxed) { } elsewhen (s1.format === FpuFormat.FLOAT =/= rs(0).boxed) {
output.rs1.setNanQuiet output.rs1.setNanQuiet
output.rs1.sign := False output.rs1.sign := False
} }
} }
when(s1.format === FpuFormat.FLOAT =/= rs2Entry.boxed) { when(s1.format === FpuFormat.FLOAT =/= rs(1).boxed) {
output.rs2.setNanQuiet output.rs2.setNanQuiet
output.rs2.sign := False output.rs2.sign := False
} }
when(s1.format === FpuFormat.FLOAT =/= rs3Entry.boxed) { when(s1.format === FpuFormat.FLOAT =/= rs(2).boxed) {
output.rs3.setNanQuiet output.rs3.setNanQuiet
} }
} }
@ -1003,23 +1005,26 @@ case class FpuCore( portCount : Int, p : FpuParameter) extends Component{
output.NV := NV output.NV := NV
output.DZ := False output.DZ := False
decode.mulToAdd.valid := input.valid && input.add val mulToAdd = Stream(AddInput())
decode.mulToAdd.source := input.source decode.mulToAdd << mulToAdd.stage()
decode.mulToAdd.rs1.mantissa := norm.output.mantissa >> 1 //FMA Precision lost
decode.mulToAdd.rs1.exponent := norm.output.exponent mulToAdd.valid := input.valid && input.add
decode.mulToAdd.rs1.sign := norm.output.sign mulToAdd.source := input.source
decode.mulToAdd.rs1.special := norm.output.special mulToAdd.rs1.mantissa := norm.output.mantissa >> 1 //FMA Precision lost
decode.mulToAdd.rs2 := input.rs3 mulToAdd.rs1.exponent := norm.output.exponent
decode.mulToAdd.rd := input.rd mulToAdd.rs1.sign := norm.output.sign
decode.mulToAdd.roundMode := input.roundMode mulToAdd.rs1.special := norm.output.special
decode.mulToAdd.needCommit := False mulToAdd.rs2 := input.rs3
if (p.withDouble) decode.mulToAdd.format := input.format mulToAdd.rd := input.rd
mulToAdd.roundMode := input.roundMode
mulToAdd.needCommit := False
if (p.withDouble) mulToAdd.format := input.format
when(NV){ when(NV){
decode.mulToAdd.rs1.mantissa.msb := False mulToAdd.rs1.mantissa.msb := False
} }
input.ready := (input.add ? decode.mulToAdd.ready | output.ready) || input.divSqrt input.ready := (input.add ? mulToAdd.ready | output.ready) || input.divSqrt
} }
} }
@ -1348,6 +1353,27 @@ case class FpuCore( portCount : Int, p : FpuParameter) extends Component{
val add = p.withAdd generate new Area{ val add = p.withAdd generate new Area{
class PreShifterOutput extends AddInput{
val absRs1Bigger = Bool()
val rs1ExponentBigger = Bool()
}
val preShifter = new Area{
val input = decode.add.combStage()
val output = input.swapPayload(new PreShifterOutput)
val exp21 = input.rs2.exponent -^ input.rs1.exponent
val rs1ExponentBigger = (exp21.msb || input.rs2.isZero) && !input.rs1.isZero
val rs1ExponentEqual = input.rs1.exponent === input.rs2.exponent
val rs1MantissaBigger = input.rs1.mantissa > input.rs2.mantissa
val absRs1Bigger = ((rs1ExponentBigger || rs1ExponentEqual && rs1MantissaBigger) && !input.rs1.isZero || input.rs1.isInfinity) && !input.rs2.isInfinity
output.payload.assignSomeByName(input.payload)
output.absRs1Bigger := absRs1Bigger
output.rs1ExponentBigger := rs1ExponentBigger
}
class ShifterOutput extends AddInput{ class ShifterOutput extends AddInput{
val xSign, ySign = Bool() val xSign, ySign = Bool()
val xMantissa, yMantissa = UInt(p.internalMantissaSize+3 bits) val xMantissa, yMantissa = UInt(p.internalMantissaSize+3 bits)
@ -1357,19 +1383,22 @@ case class FpuCore( portCount : Int, p : FpuParameter) extends Component{
} }
val shifter = new Area { val shifter = new Area {
val input = decode.add.stage() val input = preShifter.output.stage()
val output = input.swapPayload(new ShifterOutput) val output = input.swapPayload(new ShifterOutput)
output.payload.assignSomeByName(input.payload) output.payload.assignSomeByName(input.payload)
val exp21 = input.rs2.exponent -^ input.rs1.exponent val exp21 = input.rs2.exponent -^ input.rs1.exponent
val rs1ExponentBigger = (exp21.msb || input.rs2.isZero) && !input.rs1.isZero // val rs1ExponentBigger = (exp21.msb || input.rs2.isZero) && !input.rs1.isZero
val rs1ExponentEqual = input.rs1.exponent === input.rs2.exponent // val rs1ExponentEqual = input.rs1.exponent === input.rs2.exponent
val rs1MantissaBigger = input.rs1.mantissa > input.rs2.mantissa // val rs1MantissaBigger = input.rs1.mantissa > input.rs2.mantissa
val absRs1Bigger = ((rs1ExponentBigger || rs1ExponentEqual && rs1MantissaBigger) && !input.rs1.isZero || input.rs1.isInfinity) && !input.rs2.isInfinity // val absRs1Bigger = ((rs1ExponentBigger || rs1ExponentEqual && rs1MantissaBigger) && !input.rs1.isZero || input.rs1.isInfinity) && !input.rs2.isInfinity
val shiftBy = exp21.asSInt.abs//rs1ExponentBigger ? (0-exp21) | exp21 val shiftBy = exp21.asSInt.abs//rs1ExponentBigger ? (0-exp21) | exp21
val shiftOverflow = (shiftBy >= p.internalMantissaSize+3) val shiftOverflow = (shiftBy >= p.internalMantissaSize+3)
val passThrough = shiftOverflow || (input.rs1.isZero) || (input.rs2.isZero) val passThrough = shiftOverflow || (input.rs1.isZero) || (input.rs2.isZero)
def absRs1Bigger = input.absRs1Bigger
def rs1ExponentBigger = input.rs1ExponentBigger
//Note that rs1ExponentBigger can be replaced by absRs1Bigger bellow to avoid xsigned two complement in math block at expense of combinatorial path //Note that rs1ExponentBigger can be replaced by absRs1Bigger bellow to avoid xsigned two complement in math block at expense of combinatorial path
val xySign = absRs1Bigger ? input.rs1.sign | input.rs2.sign val xySign = absRs1Bigger ? input.rs1.sign | input.rs2.sign
output.xSign := xySign ^ (rs1ExponentBigger ? input.rs1.sign | input.rs2.sign) output.xSign := xySign ^ (rs1ExponentBigger ? input.rs1.sign | input.rs2.sign)

1
src/main/scala/vexriscv/ip/fpu/Interface.scala
View File

@ -116,6 +116,7 @@ object FpuRoundModeInstr extends SpinalEnum(){
case class FpuParameter( withDouble : Boolean, case class FpuParameter( withDouble : Boolean,
asyncRegFile : Boolean = false,
mulWidthA : Int = 18, mulWidthA : Int = 18,
mulWidthB : Int = 18, mulWidthB : Int = 18,
sim : Boolean = false, sim : Boolean = false,