vm: added 1-bit variants of lsr/lsl opcodes

This commit is contained in:
Irmen de Jong
2022-04-11 00:25:00 +02:00
parent 890b1c2d52
commit ec7501782d
6 changed files with 150 additions and 67 deletions

View File

@@ -133,7 +133,6 @@ internal class BuiltinFuncGen(private val codeGen: CodeGen, private val exprGen:
}
private fun funcPoke(call: PtBuiltinFunctionCall, resultRegister: Int): VmCodeChunk {
// should just be a memory write
val addressReg = codeGen.vmRegisters.nextFree()
val valueReg = codeGen.vmRegisters.nextFree()
val code = VmCodeChunk()
@@ -152,7 +151,6 @@ internal class BuiltinFuncGen(private val codeGen: CodeGen, private val exprGen:
}
private fun funcPeek(call: PtBuiltinFunctionCall, resultRegister: Int): VmCodeChunk {
// should just be a memory read
val code = VmCodeChunk()
val addressReg = codeGen.vmRegisters.nextFree()
code += exprGen.translateExpression(call.args.single(), addressReg)
@@ -187,6 +185,7 @@ internal class BuiltinFuncGen(private val codeGen: CodeGen, private val exprGen:
}
private fun funcLsb(call: PtBuiltinFunctionCall, resultRegister: Int): VmCodeChunk {
// TODO optimized code gen
val code = VmCodeChunk()
code += exprGen.translateExpression(call.args.single(), resultRegister)
// note: if a word result is needed, the upper byte is cleared by the typecast that follows. No need to do it here.
@@ -194,9 +193,10 @@ internal class BuiltinFuncGen(private val codeGen: CodeGen, private val exprGen:
}
private fun funcMsb(call: PtBuiltinFunctionCall, resultRegister: Int): VmCodeChunk {
// TODO optimized code gen
val code = VmCodeChunk()
code += exprGen.translateExpression(call.args.single(), resultRegister)
code += VmCodeInstruction(Opcode.SWAP, VmDataType.BYTE, reg1 = resultRegister)
code += VmCodeInstruction(Opcode.MSIG, VmDataType.BYTE, reg1 = resultRegister, reg2=resultRegister)
// note: if a word result is needed, the upper byte is cleared by the typecast that follows. No need to do it here.
return code
}
@@ -248,7 +248,6 @@ internal class BuiltinFuncGen(private val codeGen: CodeGen, private val exprGen:
}
private fun funcRolRor2(opcode: Opcode, call: PtBuiltinFunctionCall, resultRegister: Int): VmCodeChunk {
// bit rotate left without carry, in-place
val vmDt = codeGen.vmType(call.args[0].type)
val code = VmCodeChunk()
code += exprGen.translateExpression(call.args[0], resultRegister)

View File

@@ -372,23 +372,29 @@ class CodeGen(internal val program: PtProgram,
private val powersOfTwo = (0..16).map { 2.0.pow(it.toDouble()).toInt() }
private fun multiplyByConst(dt: VmDataType, reg: Int, factor: UInt): VmCodeChunk {
internal fun multiplyByConst(dt: VmDataType, reg: Int, factor: Int): VmCodeChunk {
require(factor>=0)
val code = VmCodeChunk()
val pow2 = powersOfTwo.indexOf(factor.toInt())
if(pow2>=1) {
// just shift bits
code += VmCodeInstruction(Opcode.LSL, dt, reg1=reg, reg2=reg, reg3=pow2)
if(factor==1)
return code
val pow2 = powersOfTwo.indexOf(factor)
if(pow2==1) {
// just shift 1 bit
code += VmCodeInstruction(Opcode.LSL, dt, reg1=reg)
}
else if(pow2>=1) {
// just shift multiple bits
val pow2reg = vmRegisters.nextFree()
code += VmCodeInstruction(Opcode.LOAD, dt, reg1=pow2reg, value=pow2)
code += VmCodeInstruction(Opcode.LSLM, dt, reg1=reg, reg2=reg, reg3=pow2reg)
} else {
when(factor) {
0u -> {
code += VmCodeInstruction(Opcode.LOAD, dt, reg1=reg, value=0)
}
1u -> { /* do nothing */ }
else -> {
val factorReg = vmRegisters.nextFree()
code += VmCodeInstruction(Opcode.LOAD, dt, reg1=factorReg, value=factor.toInt())
code += VmCodeInstruction(Opcode.MUL, dt, reg1=reg, reg2=reg, reg3=factorReg)
}
if (factor == 0) {
code += VmCodeInstruction(Opcode.LOAD, dt, reg1=reg, value=0)
}
else {
val factorReg = vmRegisters.nextFree()
code += VmCodeInstruction(Opcode.LOAD, dt, reg1=factorReg, value= factor)
code += VmCodeInstruction(Opcode.MUL, dt, reg1=reg, reg2=reg, reg3=factorReg)
}
}
return code

View File

@@ -145,11 +145,8 @@ internal class ExpressionGen(private val codeGen: CodeGen) {
code += VmCodeInstruction(Opcode.LOADM, vmDt, reg1=resultRegister, value=arrayLocation+memOffset)
} else {
code += translateExpression(arrayIx.index, idxReg)
if(eltSize>1) {
val factorReg = codeGen.vmRegisters.nextFree()
code += VmCodeInstruction(Opcode.LOAD, VmDataType.BYTE, reg1=factorReg, value=eltSize)
code += VmCodeInstruction(Opcode.MUL, VmDataType.BYTE, reg1=idxReg, reg2=idxReg, reg3=factorReg)
}
if(eltSize>1)
code += codeGen.multiplyByConst(VmDataType.BYTE, idxReg, eltSize)
code += VmCodeInstruction(Opcode.LOADX, vmDt, reg1=resultRegister, reg2=idxReg, value = arrayLocation)
}
return code
@@ -267,6 +264,7 @@ internal class ExpressionGen(private val codeGen: CodeGen) {
val rightResultReg = codeGen.vmRegisters.nextFree()
// TODO: optimized codegen when left or right operand is known 0 or 1 or whatever. But only if this would result in a different opcode such as ADD 1 -> INC, MUL 1 -> NOP
// actually optimizing the code should not be done here but in a tailored code optimizer step.
// multiplyByConst()
val leftCode = translateExpression(binExpr.left, leftResultReg)
val rightCode = translateExpression(binExpr.right, rightResultReg)
code += leftCode
@@ -299,10 +297,10 @@ internal class ExpressionGen(private val codeGen: CodeGen) {
code += VmCodeInstruction(Opcode.XOR, vmDt, reg1=resultRegister, reg2=leftResultReg, reg3=rightResultReg)
}
"<<" -> {
code += VmCodeInstruction(Opcode.LSL, vmDt, reg1=resultRegister, reg2=leftResultReg, reg3=rightResultReg)
code += VmCodeInstruction(Opcode.LSLM, vmDt, reg1=resultRegister, reg2=leftResultReg, reg3=rightResultReg)
}
">>" -> {
val opc = if(signed) Opcode.ASR else Opcode.LSR
val opc = if(signed) Opcode.ASRM else Opcode.LSRM
code += VmCodeInstruction(opc, vmDt, reg1=resultRegister, reg2=leftResultReg, reg3=rightResultReg)
}
"==" -> {