def parseLoop(nodes: List[Element], loc: Location) -> OP.Loop:
if len(nodes) < 3:
raise ParseError('incorrect number of args passed to op_par_loop', loc)
# Parse loop kernel and set
kernel = parseIdentifier(nodes[0])
set_ = parseIdentifier(nodes[1])
loop_args = []
# Parse loop args
for raw_arg in nodes[2:]:
name = parseIdentifier(raw_arg)
arg_loc = parseLocation(raw_arg)
args = raw_arg.findall('name/subscripts/subscript')
if name == 'op_arg_dat':
loop_args.append(parseArgDat(args, arg_loc))
elif name == 'op_opt_arg_dat':
loop_args.append(parseOptArgDat(args, arg_loc))
elif name == 'op_arg_gbl':
loop_args.append(parseArgGbl(args, arg_loc))
elif name == 'op_opt_arg_gbl':
loop_args.append(parseOptArgGbl(args, arg_loc))
else:
raise ParseError(f'invalid loop argument {name}')
return OP.Loop(kernel, set_, loc, loop_args)
def parseIntLit(node: Element, signed: bool = True) -> int:
# Parse location
loc = parseLocation(node)
# Assume the literal is not negated
negation = False
# Check if the node is wrapped in a valid unary negation
if signed and node.find('operation'):
node = node.find('operation')
if node.attrib['type'] == 'unary' and node.find('operator') and node.find('operator').attrib['operator'] == '-':
negation = True
node = node.find('operand')
# Descend to child literal node
node = node.find('literal')
# Verify and typecheck the literal node
if not node or node.attrib['type'] != 'int':
if not signed:
raise ParseError('expected unsigned integer literal', loc)
else:
raise ParseError('expected integer literal', loc)
# Extract the value
value = int(node.attrib['value'])
return -value if negation else value
def parseLoop(nodes: List[Cursor], loc: Location) -> OP.Loop:
if len(nodes) < 3:
raise ParseError('incorrect number of args passed to op_par_loop')
# Parse loop kernel and set
kernel = parseIdentifier(nodes[0])
_ = parseStringLit(nodes[1])
set_ = parseIdentifier(nodes[2])
loop_args = []
# Parse loop args
for node in nodes[3:]:
node = descend(descend(node))
name = node.spelling
arg_loc = parseLocation(node)
args = list(node.get_children())[1:]
if name == 'op_arg_dat':
loop_args.append(parseArgDat(args, arg_loc))
elif name == 'op_opt_arg_dat':
loop_args.append(parseOptArgDat(args, arg_loc))
elif name == 'op_arg_gbl':
loop_args.append(parseArgGbl(args, arg_loc))
elif name == 'op_opt_arg_gbl':
loop_args.append(parseOptArgGbl(args, arg_loc))
else:
raise ParseError(f'invalid loop argument {name}',
parseLocation(node))
return OP.Loop(kernel, set_, loc, loop_args)
def parseIdentifier(node: Cursor, regex: str = None) -> str:
# TODO: Check this
while node.kind == CursorKind.CSTYLE_CAST_EXPR:
node = list(node.get_children())[1]
# Descend to child node
if node.kind == CursorKind.UNEXPOSED_EXPR:
node = descend(node)
# Descend to child node
if node.kind == CursorKind.UNARY_OPERATOR and next(
node.get_tokens()).spelling in ('&', '*'):
node = descend(node)
# Check for null
if node.kind == CursorKind.GNU_NULL_EXPR:
return ''
# Validate the node
if node.kind != CursorKind.DECL_REF_EXPR:
raise ParseError('expected identifier')
value = node.spelling
# Apply conditional regex constraint
if regex and not re.match(regex, value):
raise ParseError(f'expected identifier matching {regex}')
return value
def parseKernel(self, path: Path, name: str) -> Kernel:
# Invoke Clang parser on kernel source
translation_unit = Index.create().parse(path)
# Collect root-level nodes
nodes = translation_unit.cursor.get_children()
# Search for kernel function
node = safeFind(
nodes,
lambda n: n.kind == CursorKind.FUNCTION_DECL and n.spelling == name)
if not node:
raise ParseError(f'failed to locate kernel function {name}',
parseLocation(node))
# Collect parameter types
params = []
for n in node.get_children():
if n.kind == CursorKind.PARM_DECL:
type = n.type.get_pointee() or n.type
type = re.sub(r'\s*const\s*', '', type.spelling)
param = (n.spelling, type)
params.append(param)
return Kernel(name, path, translation_unit.cursor, params)
def parseKernel(self, path: Path, name: str) -> Kernel:
# Parse AST
ast = parse(path)
# Search for kernel function
nodes = ast.findall('file/subroutine')
node = safeFind(nodes, lambda n: n.attrib['name'] == name)
if not node:
raise ParseError(f'failed to locate kernel function {name}')
# Parse parameter identifiers
param_identifiers = [ n.attrib['name'] for n in node.findall('header/arguments/argument') ]
params = [ ('', '') ] * len(param_identifiers)
# TODO: Cleanup
for decl in node.findall('body/specification/declaration'):
if decl.attrib and decl.attrib['type'] == 'variable':
type = decl.find('type')
for variable in decl.findall('variables/variable'):
if variable.attrib:
identifier = variable.attrib['name']
if identifier in param_identifiers:
index = param_identifiers.index(identifier)
params[index] = (identifier, parseType(type))
return Kernel(name, path, ast, params)
def parseSet(nodes: List[Element], loc: Location) -> OP.Set:
if len(nodes) != 3:
raise ParseError('incorrect number of nodes passed to op_decl_set', loc)
_ = parseIdentifier(nodes[0])
ptr = parseIdentifier(nodes[1])
debug = parseStringLit(nodes[2])
return OP.Set(ptr)
def parseConst(nodes: List[Element], loc: Location) -> OP.Const:
if len(nodes) != 3:
raise ParseError('incorrect number of args passed to op_decl_const', loc)
ptr = parseIdentifier(nodes[0])
dim = parseIntLit(nodes[1], signed=False)
debug = parseStringLit(nodes[2])
return OP.Const(ptr, dim, debug, loc)
def parseIdentifier(node: Element, regex: str = None) -> str:
# Parse location
loc = parseLocation(node)
# Descend to child node
node = node.find('name')
# Validate the node
if not node or not node.attrib['id']:
raise ParseError('expected identifier', loc)
value = node.attrib['id']
# Apply conditional regex constraint
if regex and not re.match(regex, value):
raise ParseError(f'expected identifier matching {regex}', loc)
return value
def parseStringLit(node: Element, regex: str = None) -> str:
# Parse location
loc = parseLocation(node)
# Descend to child literal node
node = node.find('literal')
# Validate the node
if not node or node.attrib['type'] != 'char':
raise ParseError('expected string literal', loc)
# Extract value from string delimeters
value = node.attrib['value'][1:-1]
# Apply conditional regex constraint
if regex and not re.match(regex, value):
raise ParseError(f'expected string literal matching {regex}', loc)
return value
def parse(path: Path) -> Element:
try:
# Track the current file for parse errors
global _current_file
_current_file = str(path)
# Invoke OFP on the source
return fp.parse(path, raise_on_error=True)
except CalledProcessError as error:
raise ParseError(error.output)
def parseMap(nodes: List[Cursor], ptr: str, loc: Location) -> OP.Map:
if len(nodes) != 5:
raise ParseError('incorrect number of args passed to op_decl_map', loc)
from_set = parseIdentifier(nodes[0])
to_set = parseIdentifier(nodes[1])
dim = parseIntLit(nodes[2], signed=False)
_ = parseIdentifier(nodes[3])
debug = parseStringLit(nodes[4])
return OP.Map(from_set, to_set, dim, ptr, loc)
def parseData(nodes: List[Cursor], ptr: str, loc: Location) -> OP.Data:
if len(nodes) != 5:
raise ParseError('incorrect number of args passed to op_decl_dat', loc)
set_ = parseIdentifier(nodes[0])
dim = parseIntLit(nodes[1], signed=False)
typ = parseStringLit(nodes[2])
_ = parseIdentifier(nodes[3])
debug = parseStringLit(nodes[4])
return OP.Data(set_, dim, typ, ptr, loc)
def parseStringLit(node: Cursor, regex: str = None) -> str:
# Validate the node
if node.kind != CursorKind.UNEXPOSED_EXPR:
raise ParseError('expected string literal')
# Descend to child node
node = descend(node)
# Validate the node
if node.kind != CursorKind.STRING_LITERAL:
raise ParseError('expected string literal')
# Extract value from string delimeters
value = node.spelling[1:-1]
# Apply conditional regex constraint
if regex and not re.match(regex, value):
raise ParseError(f'expected string literal matching {regex}')
return value
def parseArgGbl(nodes: List[Element], loc: Location) -> OP.Arg:
if len(nodes) != 4:
raise ParseError('incorrect number of args passed to op_arg_gbl', loc)
access_regex = enumRegex(OP.GBL_ACCESS_TYPES)
var = parseIdentifier(nodes[0])
dim = parseIntLit(nodes[1], signed=False)
typ = normaliseType(parseStringLit(nodes[2]))
acc = parseIdentifier(nodes[3], regex=access_regex)
return OP.Arg(var, dim, typ, acc, loc)
def parseData(nodes: List[Element], loc: Location) -> OP.Data:
if len(nodes) != 6:
raise ParseError('incorrect number of args passed to op_decl_dat', loc)
set_ = parseIdentifier(nodes[0])
dim = parseIntLit(nodes[1], signed=False)
typ = normaliseType(parseStringLit(nodes[2]))
_ = parseIdentifier(nodes[3])
ptr = parseIdentifier(nodes[4])
debug = parseStringLit(nodes[5])
return OP.Data(set_, dim, typ, ptr, loc)
def parseIntLit(node: Cursor, signed: bool = True) -> int:
# Assume the literal is not negated
negation = False
# Check if the node is wrapped in a valid unary negation
if signed and node.kind == CursorKind.UNARY_OPERATOR and next(
node.get_tokens()).spelling == '-':
negation = True
node = descend(node)
# Validate the node
if node.kind != CursorKind.INTEGER_LITERAL:
if not signed:
raise ParseError('expected unsigned integer literal')
else:
raise ParseError('expected integer literal')
# Extract the value
value = int(next(node.get_tokens()).spelling)
return -value if negation else value
def parseOptArgGbl(nodes: List[Element], loc: Location) -> OP.Arg:
if len(nodes) != 5:
ParseError('incorrect number of args passed to op_opt_arg_gbl', loc)
# Parse opt argument
opt = parseIdentifier(nodes[0])
# Parse standard argGbl arguments
dat = parseArgGbl(nodes[1:], loc)
# Return augmented dat
dat.opt = opt
return dat
def parseArgGbl(nodes: List[Cursor], loc: Location) -> OP.Arg:
if len(nodes) != 4:
raise ParseError('incorrect number of args passed to op_arg_gbl', loc)
access_regex = enumRegex(OP.GBL_ACCESS_TYPES)
var = parseIdentifier(nodes[0])
dim = parseIntLit(nodes[1], signed=False)
typ = parseStringLit(nodes[2])
acc = macro_instances[(nodes[3].location.line,
nodes[3].location.column)] # TODO: Cleanup
return OP.Arg(var, dim, typ, acc, loc)
def parseArgDat(nodes: List[Cursor], loc: Location) -> OP.Arg:
if len(nodes) != 6:
raise ParseError('incorrect number of args passed to op_arg_dat', loc)
access_regex = enumRegex(OP.DAT_ACCESS_TYPES)
var = parseIdentifier(nodes[0])
idx = parseIntLit(nodes[1], signed=True)
map_ = parseIdentifier(nodes[2]) or OP.ID
dim = parseIntLit(nodes[3], signed=False)
typ = parseStringLit(nodes[4])
acc = macro_instances[(nodes[5].location.line,
nodes[5].location.column)] # TODO: Cleanup
return OP.Arg(var, dim, typ, acc, loc, map_, idx)
def parseProgram(self, path: Path, include_dirs: Set[Path]) -> Program:
# Locate OP2 install
op2_install = os.getenv('OP2_INSTALL_PATH')
if not op2_install:
exit('Fatal: OP2_INSTALL_PATH not set')
# Add OP2 includes
op2_include = Path(op2_install).joinpath('c/include')
include_dirs.add(op2_include)
# Form Clang args
args = [f'-I{dir}' for dir in include_dirs]
# Invoke Clang parser on the program source
translation_unit = Index.create().parse(
path,
args=args,
options=TranslationUnit.PARSE_DETAILED_PROCESSING_RECORD)
# Throw the parse error first parse error caught in the diagnostics
error = next(iter(translation_unit.diagnostics), None)
if error:
raise ParseError(error.spelling, parseLocation(error))
# Initialise search stack
program = Program(path)
stack = []
for child in translation_unit.cursor.get_children():
# Ignore macro definitions and cursors outside of the program file
if child.kind != CursorKind.MACRO_DEFINITION and child.location.file.name == translation_unit.spelling:
# Collect the locations and identifiers of macro instances
if child.kind == CursorKind.MACRO_INSTANTIATION:
macro_instances[(child.location.line,
child.location.column)] = child.displayname
# Populate with the top-level cursors from the target file
else:
stack.append(child)
ptr: str
# DFS
while stack:
# Manage the stack
node = stack.pop()
stack.extend(node.get_children())
# TODO: This is a bit of a hack. Cleanup
if node.kind == CursorKind.VAR_DECL:
if safeFind(node.get_children(),
lambda n: n.kind == CursorKind.CALL_EXPR):
ptr = node.spelling
# Focus on function calls
if node.kind == CursorKind.CALL_EXPR:
name = node.spelling
args = list(node.get_children())[1:]
loc = parseLocation(node)
if name == 'op_init':
program.recordInit(loc)
elif name == 'op_decl_set':
program.sets.append(parseSet(args, ptr, loc))
elif name == 'op_decl_map':
program.maps.append(parseMap(args, ptr, loc))
elif name == 'op_decl_dat':
program.datas.append(parseData(args, ptr, loc))
elif name == 'op_decl_const':
program.consts.append(parseConst(args, loc))
elif name == 'op_par_loop':
program.loops.append(parseLoop(args, loc))
elif name == 'op_exit':
program.recordExit()
return program