mirror of
https://github.com/irmen/prog8.git
synced 2024-11-29 17:50:35 +00:00
268 lines
13 KiB
Python
268 lines
13 KiB
Python
"""
|
|
Programming Language for 6502/6510 microprocessors, codename 'Sick'
|
|
This is the code generator for variable declarations and initialization.
|
|
|
|
Written by Irmen de Jong (irmen@razorvine.net) - license: GNU GPL 3.0
|
|
"""
|
|
|
|
from collections import defaultdict
|
|
from typing import Dict, List, Callable, Any, no_type_check
|
|
from ..plyparse import Block, VarType, VarDef, LiteralValue
|
|
from ..datatypes import DataType, STRING_DATATYPES
|
|
from . import to_hex, to_mflpt5, CodeError
|
|
|
|
|
|
def generate_block_init(out: Callable, block: Block) -> None:
|
|
# generate the block initializer
|
|
# @todo add a block initializer subroutine that can contain custom reset/init code? (static initializer)
|
|
# @todo will be called at program start automatically, so there's no risk of forgetting to call it manually
|
|
|
|
def _memset(varname: str, value: int, size: int) -> None:
|
|
if size > 6:
|
|
out("\vlda #<" + varname)
|
|
out("\vsta il65_lib.SCRATCH_ZPWORD1")
|
|
out("\vlda #>" + varname)
|
|
out("\vsta il65_lib.SCRATCH_ZPWORD1+1")
|
|
out("\vlda #" + to_hex(value))
|
|
out("\vldx #<" + to_hex(size))
|
|
out("\vldy #>" + to_hex(size))
|
|
out("\vjsr il65_lib.memset")
|
|
else:
|
|
out("\vlda #" + to_hex(value))
|
|
for i in range(size):
|
|
out("\vsta {:s}+{:d}".format(varname, i))
|
|
|
|
def _memsetw(varname: str, value: int, size: int) -> None:
|
|
if size > 4:
|
|
out("\vlda #<" + varname)
|
|
out("\vsta il65_lib.SCRATCH_ZPWORD1")
|
|
out("\vlda #>" + varname)
|
|
out("\vsta il65_lib.SCRATCH_ZPWORD1+1")
|
|
out("\vlda #<" + to_hex(size))
|
|
out("\vsta il65_lib.SCRATCH_ZPWORD2")
|
|
out("\vlda #>" + to_hex(size))
|
|
out("\vsta il65_lib.SCRATCH_ZPWORD2+1")
|
|
out("\vlda #<" + to_hex(value))
|
|
out("\vldx #>" + to_hex(value))
|
|
out("\vjsr il65_lib.memsetw")
|
|
else:
|
|
out("\vlda #<" + to_hex(value))
|
|
out("\vldy #>" + to_hex(value))
|
|
for i in range(size):
|
|
out("\vsta {:s}+{:d}".format(varname, i * 2))
|
|
out("\vsty {:s}+{:d}".format(varname, i * 2 + 1))
|
|
|
|
out("_il65_init_block\v; (re)set vars to initial values")
|
|
float_inits = {}
|
|
prev_value_a, prev_value_x = None, None
|
|
vars_by_datatype = defaultdict(list) # type: Dict[DataType, List[VarDef]]
|
|
for vardef in block.all_nodes(VarDef):
|
|
if vardef.vartype == VarType.VAR: # type: ignore
|
|
vars_by_datatype[vardef.datatype].append(vardef) # type: ignore
|
|
for bytevar in sorted(vars_by_datatype[DataType.BYTE], key=lambda vd: vd.value):
|
|
assert isinstance(bytevar.value, LiteralValue) and type(bytevar.value.value) is int
|
|
if bytevar.value.value != prev_value_a:
|
|
out("\vlda #${:02x}".format(bytevar.value.value))
|
|
prev_value_a = bytevar.value.value
|
|
out("\vsta {:s}".format(bytevar.name))
|
|
for wordvar in sorted(vars_by_datatype[DataType.WORD], key=lambda vd: vd.value):
|
|
assert isinstance(wordvar.value, LiteralValue) and type(wordvar.value.value) is int
|
|
v_hi, v_lo = divmod(wordvar.value.value, 256)
|
|
if v_hi != prev_value_a:
|
|
out("\vlda #${:02x}".format(v_hi))
|
|
prev_value_a = v_hi
|
|
if v_lo != prev_value_x:
|
|
out("\vldx #${:02x}".format(v_lo))
|
|
prev_value_x = v_lo
|
|
out("\vsta {:s}".format(wordvar.name))
|
|
out("\vstx {:s}+1".format(wordvar.name))
|
|
for floatvar in vars_by_datatype[DataType.FLOAT]:
|
|
assert isinstance(floatvar.value, LiteralValue) and type(floatvar.value.value) in (int, float)
|
|
fpbytes = to_mflpt5(floatvar.value.value) # type: ignore
|
|
float_inits[floatvar.name] = (floatvar.name, fpbytes, floatvar.value)
|
|
for arrayvar in vars_by_datatype[DataType.BYTEARRAY]:
|
|
assert isinstance(arrayvar.value, LiteralValue) and type(arrayvar.value.value) is int
|
|
_memset(arrayvar.name, arrayvar.value.value, arrayvar.size[0])
|
|
for arrayvar in vars_by_datatype[DataType.WORDARRAY]:
|
|
assert isinstance(arrayvar.value, LiteralValue) and type(arrayvar.value.value) is int
|
|
_memsetw(arrayvar.name, arrayvar.value.value, arrayvar.size[0])
|
|
for arrayvar in vars_by_datatype[DataType.MATRIX]:
|
|
assert isinstance(arrayvar.value, LiteralValue) and type(arrayvar.value.value) is int
|
|
_memset(arrayvar.name, arrayvar.value.value, arrayvar.size[0] * arrayvar.size[1])
|
|
if float_inits:
|
|
out("\vldx #4")
|
|
out("-")
|
|
for varname, (vname, b, fv) in sorted(float_inits.items()):
|
|
out("\vlda _init_float_{:s},x".format(varname))
|
|
out("\vsta {:s},x".format(vname))
|
|
out("\vdex")
|
|
out("\vbpl -")
|
|
out("\vrts\n")
|
|
for varname, (vname, fpbytes, fpvalue) in sorted(float_inits.items()):
|
|
assert isinstance(fpvalue, LiteralValue)
|
|
out("_init_float_{:s}\t\t.byte ${:02x}, ${:02x}, ${:02x}, ${:02x}, ${:02x}\t; {}".format(varname, *fpbytes, fpvalue.value))
|
|
all_string_vars = []
|
|
for svtype in STRING_DATATYPES:
|
|
all_string_vars.extend(vars_by_datatype[svtype])
|
|
for strvar in all_string_vars:
|
|
# string vars are considered to be a constant, and are statically initialized.
|
|
_generate_string_var(out, strvar)
|
|
out("")
|
|
|
|
|
|
def generate_block_vars(out: Callable, block: Block, zeropage: bool=False) -> None:
|
|
# Generate the block variable storage.
|
|
# The memory bytes of the allocated variables is set to zero (so it compresses very well),
|
|
# their actual starting values are set by the block init code.
|
|
vars_by_vartype = defaultdict(list) # type: Dict[VarType, List[VarDef]]
|
|
for vardef in block.all_nodes(VarDef):
|
|
vars_by_vartype[vardef.vartype].append(vardef) # type: ignore
|
|
out("; constants")
|
|
for vardef in vars_by_vartype.get(VarType.CONST, []):
|
|
if vardef.datatype == DataType.FLOAT:
|
|
out("\v{:s} = {}".format(vardef.name, _numeric_value_str(vardef.value)))
|
|
elif vardef.datatype in (DataType.BYTE, DataType.WORD):
|
|
assert isinstance(vardef.value.value, int) # type: ignore
|
|
out("\v{:s} = {:s}".format(vardef.name, _numeric_value_str(vardef.value, True)))
|
|
elif vardef.datatype.isstring():
|
|
# a const string is just a string variable in the generated assembly
|
|
_generate_string_var(out, vardef)
|
|
else:
|
|
raise CodeError("invalid const type", vardef)
|
|
# @todo float constants that are used in expressions
|
|
out("; memory mapped variables")
|
|
for vardef in vars_by_vartype.get(VarType.MEMORY, []):
|
|
# create a definition for variables at a specific place in memory (memory-mapped)
|
|
assert isinstance(vardef.value.value, int) # type: ignore
|
|
if vardef.datatype.isnumeric():
|
|
assert vardef.size == [1]
|
|
out("\v{:s} = {:s}\t; {:s}".format(vardef.name, to_hex(vardef.value.value), vardef.datatype.name.lower())) # type: ignore
|
|
elif vardef.datatype == DataType.BYTEARRAY:
|
|
assert len(vardef.size) == 1
|
|
out("\v{:s} = {:s}\t; array of {:d} bytes".format(vardef.name, to_hex(vardef.value.value), vardef.size[0])) # type: ignore
|
|
elif vardef.datatype == DataType.WORDARRAY:
|
|
assert len(vardef.size) == 1
|
|
out("\v{:s} = {:s}\t; array of {:d} words".format(vardef.name, to_hex(vardef.value.value), vardef.size[0])) # type: ignore
|
|
elif vardef.datatype == DataType.MATRIX:
|
|
assert len(vardef.size) in (2, 3)
|
|
if len(vardef.size) == 2:
|
|
comment = "matrix of {:d} by {:d} = {:d} bytes".format(vardef.size[0], vardef.size[1], vardef.size[0]*vardef.size[1])
|
|
elif len(vardef.size) == 3:
|
|
comment = "matrix of {:d} by {:d}, interleave {:d}".format(vardef.size[0], vardef.size[1], vardef.size[2])
|
|
else:
|
|
raise CodeError("matrix size must be 2 or 3 numbers")
|
|
out("\v{:s} = {:s}\t; {:s}".format(vardef.name, to_hex(vardef.value.value), comment)) # type: ignore
|
|
else:
|
|
raise CodeError("invalid var type")
|
|
out("; normal variables - initial values will be set by init code")
|
|
if zeropage:
|
|
# zeropage uses the zp_address we've allocated, instead of allocating memory here
|
|
for vardef in vars_by_vartype.get(VarType.VAR, []):
|
|
assert vardef.zp_address is not None
|
|
if vardef.datatype.isstring():
|
|
raise CodeError("cannot put strings in the zeropage", vardef.sourceref)
|
|
if vardef.datatype.isarray():
|
|
size_str = "size " + str(vardef.size)
|
|
else:
|
|
size_str = ""
|
|
out("\v{:s} = {:s}\t; {:s} {:s}".format(vardef.name, to_hex(vardef.zp_address), vardef.datatype.name.lower(), size_str))
|
|
else:
|
|
# create definitions for the variables that takes up empty space and will be initialized at startup
|
|
string_vars = []
|
|
for vardef in vars_by_vartype.get(VarType.VAR, []):
|
|
if vardef.datatype.isnumeric():
|
|
assert vardef.size == [1]
|
|
if vardef.datatype == DataType.BYTE:
|
|
out("{:s}\v.byte ?".format(vardef.name))
|
|
elif vardef.datatype == DataType.WORD:
|
|
out("{:s}\v.word ?".format(vardef.name))
|
|
elif vardef.datatype == DataType.FLOAT:
|
|
out("{:s}\v.fill 5\t\t; float".format(vardef.name))
|
|
else:
|
|
raise CodeError("weird datatype")
|
|
elif vardef.datatype in (DataType.BYTEARRAY, DataType.WORDARRAY):
|
|
assert len(vardef.size) == 1
|
|
if vardef.datatype == DataType.BYTEARRAY:
|
|
out("{:s}\v.fill {:d}\t\t; bytearray".format(vardef.name, vardef.size[0]))
|
|
elif vardef.datatype == DataType.WORDARRAY:
|
|
out("{:s}\v.fill {:d}*2\t\t; wordarray".format(vardef.name, vardef.size[0]))
|
|
else:
|
|
raise CodeError("invalid datatype", vardef.datatype)
|
|
elif vardef.datatype == DataType.MATRIX:
|
|
assert len(vardef.size) == 2
|
|
out("{:s}\v.fill {:d}\t\t; matrix {:d}*{:d} bytes"
|
|
.format(vardef.name, vardef.size[0] * vardef.size[1], vardef.size[0], vardef.size[1]))
|
|
elif vardef.datatype.isstring():
|
|
string_vars.append(vardef)
|
|
else:
|
|
raise CodeError("unknown variable type " + str(vardef.datatype))
|
|
# string vars are considered to be a constant, and are not re-initialized.
|
|
out("")
|
|
|
|
|
|
@no_type_check
|
|
def _generate_string_var(out: Callable, vardef: VarDef) -> None:
|
|
if vardef.datatype == DataType.STRING:
|
|
# 0-terminated string
|
|
out("{:s}\n\v.null {:s}".format(vardef.name, _format_string(str(vardef.value.value))))
|
|
elif vardef.datatype == DataType.STRING_P:
|
|
# pascal string
|
|
out("{:s}\n\v.ptext {:s}".format(vardef.name, _format_string(str(vardef.value.value))))
|
|
elif vardef.datatype == DataType.STRING_S:
|
|
# 0-terminated string in screencode encoding
|
|
out(".enc 'screen'")
|
|
out("{:s}\n\v.null {:s}".format(vardef.name, _format_string(str(vardef.value.value), True)))
|
|
out(".enc 'none'")
|
|
elif vardef.datatype == DataType.STRING_PS:
|
|
# 0-terminated pascal string in screencode encoding
|
|
out(".enc 'screen'")
|
|
out("{:s}n\v.ptext {:s}".format(vardef.name, _format_string(str(vardef.value.value), True)))
|
|
out(".enc 'none'")
|
|
|
|
|
|
def _format_string(value: str, screencodes: bool = False) -> str:
|
|
if len(value) == 1 and screencodes:
|
|
if value[0].isprintable() and ord(value[0]) < 128:
|
|
return "'{:s}'".format(value[0])
|
|
else:
|
|
return str(ord(value[0]))
|
|
result = '"'
|
|
for char in value:
|
|
if char in "{}":
|
|
result += '", {:d}, "'.format(ord(char))
|
|
elif char.isprintable() and ord(char) < 128:
|
|
result += char
|
|
else:
|
|
if screencodes:
|
|
result += '", {:d}, "'.format(ord(char))
|
|
else:
|
|
if char == '\f':
|
|
result += "{clear}"
|
|
elif char == '\b':
|
|
result += "{delete}"
|
|
elif char == '\n':
|
|
result += "{cr}"
|
|
elif char == '\r':
|
|
result += "{down}"
|
|
elif char == '\t':
|
|
result += "{tab}"
|
|
else:
|
|
result += '", {:d}, "'.format(ord(char))
|
|
return result + '"'
|
|
|
|
|
|
def _numeric_value_str(value: Any, as_hex: bool=False) -> str:
|
|
if isinstance(value, LiteralValue):
|
|
value = value.value
|
|
if isinstance(value, bool):
|
|
return "1" if value else "0"
|
|
if type(value) is int:
|
|
if as_hex:
|
|
return to_hex(value)
|
|
return str(value)
|
|
if isinstance(value, (int, float)):
|
|
if as_hex:
|
|
raise TypeError("cannot output float as hex")
|
|
return str(value)
|
|
raise TypeError("no numeric representation possible", value)
|