def visit_Assert(self, node): # pylint: disable=invalid-name, no-self-use
"""Visit Assert
Convert BinOp asserts into unittest's assert equals
"""
args = [node.test]
assertion = 'assertTrue'
if isinstance(node.test, ast.Compare) and len(node.test.ops) == 1:
args = [node.test.left, node.test.comparators[0]]
assertion = {
ast.Eq: 'assertEqual',
ast.NotEq: 'assertNotEqual',
ast.Lt: 'assertLess',
ast.LtE: 'assertLessEqual',
ast.Gt: 'assertGreater',
ast.GtE: 'assertGreaterEqual',
ast.Is: 'assertIs',
ast.IsNot: 'assertIsNot',
ast.In: 'assertIn',
ast.NotIn: 'assertNotIn',
}[node.test.ops[0].__class__]
if node.msg:
args.append(node.msg)
return ast.copy_location(ast.Expr(ast.copy_location(call(
ast.copy_location(ast.Attribute(
ast.Name('self', ast.Load()),
assertion, ast.Load()
), node),
args
), node)), node)
def visit_Name(self, old_node):
node = nodes.Name(old_node.id, old_node.ctx)
ast.copy_location(node, old_node)
# Set some defaults
node.cf_maybe_null = True
node.cf_is_null = False
node.allow_null = False
node.name = node.id
if isinstance(node.ctx, ast.Param):
var = self.symtab[node.name]
var.is_arg = True
self.flow.mark_assignment(node, None, var, assignment=None)
elif isinstance(node.ctx, ast.Load):
var = self.symtab.lookup(node.name)
if var:
# Local variable
self.flow.mark_reference(node, var)
# Set position of assignment of this definition
if isinstance(node.ctx, (ast.Param, ast.Store)):
var = self.symtab[node.name]
if var.lineno == -1:
var.lineno = getattr(node, "lineno", 0)
var.col_offset = getattr(node, "col_offset", 0)
return node
def visit_BinOp(self, node):
node = self.generic_visit(node)
if isinstance(node.left, ast.Num) and isinstance(node.right, ast.Num):
value = eval(compile(ast.copy_location(ast.Expression(body=node), node), '', 'eval'))
return ast.copy_location(ast.Num(n=value), node)
else:
return node
def test_load_const(self):
consts = [None,
True, False,
124,
2.0,
3j,
"unicode",
b'bytes',
(1, 2, 3)]
code = '\n'.join(['x={!r}'.format(const) for const in consts])
code += '\nx = ...'
consts.extend((Ellipsis, None))
tree = ast.parse(code)
self.assertEqual(self.get_load_const(tree),
consts)
# Replace expression nodes with constants
for assign, const in zip(tree.body, consts):
assert isinstance(assign, ast.Assign), ast.dump(assign)
new_node = ast.Constant(value=const)
ast.copy_location(new_node, assign.value)
assign.value = new_node
self.assertEqual(self.get_load_const(tree),
consts)
def prepare(self):
for i in self.children:
i.prepare()
# Compile the keywords.
keyword_values = { }
keyword_keys = [ ]
keyword_exprs = [ ]
for k, expr in self.keyword:
node = py_compile(expr, 'eval', ast_node=True)
if is_constant(node):
keyword_values[k] = py_eval_bytecode(compile_expr(node))
else:
keyword_keys.append(ast.Str(s=k))
keyword_exprs.append(node)
if keyword_values:
self.keyword_values = keyword_values
else:
self.keyword_values = None
if keyword_keys:
node = ast.Dict(keys=keyword_keys, values=keyword_exprs)
ast.copy_location(node, keyword_exprs[0])
self.keyword_exprs = compile_expr(node)
else:
self.keyword_exprs = None
def prepare(self):
SLBlock.prepare(self)
# Prepare the positional arguments.
exprs = [ ]
values = [ ]
has_exprs = False
has_values = False
for a in self.positional:
node = py_compile(a, 'eval', ast_node=True)
if is_constant(node):
values.append(py_eval_bytecode(compile_expr(node)))
exprs.append(ast.Num(n=0))
has_values = True
else:
values.append(use_expression)
exprs.append(node)
has_exprs = True
if has_values:
self.positional_values = values
else:
self.positional_values = None
if has_exprs:
t = ast.Tuple(elts=exprs, ctx=ast.Load())
ast.copy_location(t, exprs[0])
self.positional_exprs = compile_expr(t)
else:
self.positional_exprs = None
def visit_FunctionDef(self, node):
if node.name.startswith('test'):
statements = [stmt for stmt in node.body
if isinstance(stmt, ast.Assign) or
isinstance(stmt, ast.Expr)
]
self.statements += statements
self.tracking[node.name] = None
body = node.body
new_stmts = []
for _node in node.body:
new_stmts.append(createNode(node.name, _node))
new_node_body = []
for i in xrange(0, len(new_stmts)):
new_node = new_stmts[i]
old_node = body[i]
if isinstance(new_node, list):
for _new_node in new_node:
ast.copy_location(_new_node, old_node)
new_node_body.append(_new_node)
new_node_body.append(old_node)
elif isinstance(new_node, ast.TryExcept):
ast.copy_location(new_node, old_node)
new_node_body.append(new_node)
new_node_body.append(old_node)
else:
new_node_body.append(new_node)
node.body = new_node_body
ast.fix_missing_locations(node)
return node
ast.fix_missing_locations(node)
return node
def __visit_FunctionDef(self, node):
new_node = ast.FunctionDef(args=self.visit_arguments(node.args),
body=self._visit_list(node.body),
decorator_list=self._visit_list(node.decorator_list),
name=node.name)
ast.copy_location(new_node, node)
return new_node
def visit_BinOp(self, node: ast.BinOp):
node = self.generic_visit(node)
if self._is_numeric_pow(node):
left, right = node.left, node.right
degree = ( right.n if isinstance(right, ast.Num)
else -right.operand.n if isinstance(right.op, ast.USub)
else right.operand.n )
degree = int(degree)
if abs(degree) == 0:
node = ast.copy_location(ast.Num(n = 1), node)
elif abs(degree) == 1:
node = node.left
elif 2 <= abs(degree) <= self.MAX_DEGREE:
for _ in range(1, abs(degree)):
new_node = ast.BinOp\
( left = left
, op = ast.Mult()
, right = copy(node.left)
)
left = new_node = ast.copy_location(new_node, node)
node = new_node
else:
return node
if degree < 0:
new_node = ast.BinOp\
( left = ast.Num(n = 1)
, op = ast.Div()
, right = node
)
node = ast.copy_location(new_node, node)
return node
def visit_Name(self, node):
if self.randomize():
if node.id == 'forall':
return ast.copy_location(_ast.Name(id='exists'), node)
elif node.id == 'exists':
return ast.copy_location(_ast.Name(id='forall'), node)
return node
def test_load_const(self):
consts = [None,
True, False,
124,
2.0,
3j,
"unicode",
b'bytes',
(1, 2, 3)]
code = '\n'.join(map(repr, consts))
code += '\n...'
code_consts = [const for const in consts
if (not isinstance(const, (str, int, float, complex))
or isinstance(const, bool))]
code_consts.append(Ellipsis)
# the compiler adds a final "LOAD_CONST None"
code_consts.append(None)
tree = ast.parse(code)
self.assertEqual(self.get_load_const(tree), code_consts)
# Replace expression nodes with constants
for expr_node, const in zip(tree.body, consts):
assert isinstance(expr_node, ast.Expr)
new_node = ast.Constant(value=const)
ast.copy_location(new_node, expr_node.value)
expr_node.value = new_node
self.assertEqual(self.get_load_const(tree), code_consts)
def visit_If(self, node):
node = self.generic_visit(node)
if (node.orelse
and len(node.orelse) == 1
and isinstance(node.orelse[0], ast.Pass)
):
node.orelse = []
if (len(node.body) == 1
and isinstance(node.body[0], ast.Pass)
):
if node.orelse:
node_test = ast.UnaryOp(op=ast.Not(), operand=node.test)
if (len(node.orelse) == 1
and isinstance(node.orelse[0], ast.If)
):
node_test = ast.BoolOp\
( op = ast.And()
, values = [node_test, node.orelse[0].test]
)
node.test = ast.copy_location(node_test, node.orelse[0].test)
node.body = node.orelse[0].body
node.orelse = node.orelse[0].orelse
else:
node.test = ast.copy_location(node_test, node.test)
node.body = node.orelse
node.orelse = []
else:
node = None
return node
def visit_If(self, node):
exit_block = self.flow.exit_block(label='exit_if', pos=node)
# Condition
cond_block = self.flow.nextblock(self.flow.block, label='if_cond',
is_expr=True, pos=node.test)
node.test = self.visit(node.test)
# Body
if_block = self.flow.nextblock(label='if_body', pos=node.body[0])
self.visitlist(node.body)
if self.flow.block:
self.flow.block.add_child(exit_block)
# Else clause
if node.orelse:
else_block = self.flow.nextblock(cond_block,
label='else_body',
pos=node.orelse[0])
self.visitlist(node.orelse)
if self.flow.block:
self.flow.block.add_child(exit_block)
else:
cond_block.add_child(exit_block)
else_block = None
new_node = nodes.build_if(cond_block=cond_block, test=node.test,
if_block=if_block, body=node.body,
else_block=else_block, orelse=node.orelse,
exit_block=exit_block)
ast.copy_location(new_node, node)
return self.exit_block(exit_block, new_node)
def visit_Name(self, node):
new_node = ast.Name(
self._visit(node.id),
self._visit(node.ctx),
)
ast.copy_location(new_node, node)
return new_node
def test_ast_2(self):
self.ns['FOOBAR'] = ''
expr = ast.Expression()
expr.body = ast.parse('FOO+BAR').body[0].value
ast.copy_location(expr, expr.body)
self.ns['FOOBAR'] = Expression(expr)
self.assertEqual(self.ns['FOOBAR'].get(), 'foobar')
def visit_Assign(self, node):
if isinstance(node.value, ast.Subscript) and isinstance(node.value.value, ast.Call):
subscr = node.value
call = subscr.value
if len(node.targets) > 1:
error.error('Cannot use multiple assignment in array declaration.', node)
variable_name = node.targets[0].id
value_type = call.func.id
declaration_args = call.args
# Get the indices being accessed.
shape = slice_node_to_tuple_of_numbers(subscr.slice)
new_assigns = []
for indices in itertools.product(*[range(n) for n in shape]):
index_name = flattened_array_name(variable_name, indices)
new_index_name_node = ast.copy_location(ast.Name(index_name, ast.Store()), node)
new_value_type_node = ast.copy_location(ast.Name(value_type, ast.Load()), node)
new_declaration_args = [copy.deepcopy(arg) for arg in declaration_args]
new_call_node = ast.copy_location(ast.Call(new_value_type_node, new_declaration_args, [], None, None), node)
new_assign = ast.Assign([new_index_name_node], new_call_node)
new_assign = ast.copy_location(new_assign, node)
new_assigns.append(new_assign)
return new_assigns
else:
return node
def CALL(self, tree):
is_demandload = is_demandload_regex = False
if isinstance(tree.func, _ast.Attribute):
if not isinstance(tree.func.value, _ast.Name):
# this means it's a multilevel lookup;
# pkgcore.ebuild.ebuild_src.some_func
# ignore it; it *could* miss a direct
# snakeoil.demandload.demandload, but
# I don't care, bad form of access imo.
return self.handleChildren(tree)
src = self.scope.get(tree.func.value.id)
if getattr(src, 'is_demandload_module', False):
if tree.func.attr == 'demandload':
is_demandload = True
elif tree.func.attr == 'demand_compile_regexp':
is_demandload_regex = True
elif hasattr(tree.func, 'id'):
is_demandload = getattr(self.scope.get(tree.func.id), 'is_demandload_func', False)
is_demandload_regex = getattr(self.scope.get(tree.func.id), 'is_demandload_regex', False)
if is_demandload_regex:
# should do validation here.
if len(tree.args) < 3:
self.report(BadDemandloadRegexCall, tree.lineno)
elif tree.args[1].__class__.__name__.upper() not in ("STR", "UNICODE"):
self.report(BadDemandloadRegexCall, tree.lineno, "name argument isn't string nor unicode")
elif tree.args[2].__class__.__name__.upper() not in ("STR", "UNICODE"):
self.report(BadDemandloadRegexCall, tree.lineno, "regex argument isn't string nor unicode")
else:
code = "%s = re.compile(%r)\n" % (tree.args[1].s, tree.args[2].s)
fakenode = _ast.copy_location(compile(code, self.filename, "exec", _ast.PyCF_ONLY_AST).body[0],
tree)
self.addBinding(tree.lineno, _checker.Assignment(tree.args[1].s, fakenode))
if is_demandload:
if len(tree.args) < 2:
self.report(BadDemandloadCall, tree.lineno)
return self.handleChildren(tree)
for chunk in tree.args[1:]:
chunk_cls = chunk.__class__.__name__
if chunk_cls.upper() not in ('STR', 'UNICODE'):
self.report(BadDemandloadCall, chunk.lineno,
"invoked with non string/unicode arg: %r" % (chunk_cls,))
continue
s = chunk.s
try:
targets = list(parse_demandload([s]))
except ValueError, ve:
self.report(BadDemandloadCall, chunk.lineno, ve)
continue
for src, asname in targets:
fakenode = _ast.copy_location(compile("import %s as %s\n" % (src, asname),
self.filename, "exec", _ast.PyCF_ONLY_AST).body[0],
chunk)
self.addBinding(chunk.lineno,
DemandloadImportation(asname, fakenode))
def search_def_methods(self, class_node):
for node in class_node.body:
if isinstance(node, ast.FunctionDef):
if self.is_test_method(node.name):
newnode = ast.arguments(args=[], vararg=None, kwarg=None, defaults=[])
ast.copy_location(newnode, node.args)
node.args = newnode
self.methods_to_run.append(node)
def visit_With(self, node):
new_node = ast.With(
self._visit(node.items[0].context_expr),
self._visit(node.items[0].optional_vars),
self._visit(node.body)
)
ast.copy_location(new_node, node)
return new_node
def visit_Expr(self, node):
""" When capturing a call to include, we must grab it here, so we can replace the whole Expr(Call('include')).
"""
if type(node.value) is Call:
call = node.value
if type(call.func) is Name and call.func.id == 'include':
if len(call.args) < 1:
raise FormatError("include requires at least a filename as an argument.")
root = None
# if the original call to include had an additional argument
# use that argument as the root
# print('call',ast.dump(call))
if len(call.args) > 1:
root = call.args[1].s
# or if there was a root= kwarg provided, use that
elif len(call.keywords) > 0:
for k in call.keywords:
if k.arg == "root":
root = k.value.s
if root is None:
# if we didn't get one from the call to include
# look for one that was given as an argument to the template() call
root = self.root
if type(root) is str:
root = root.split(os.path.sep)
# the first argument to include() is the filename
template_name = call.args[0].s
# get the ast tree that comes from this included file
check, fundef = include_ast(template_name, root)
# each include produces the code to execute, plus some code to check for freshness
# this code absolutely must run first, because we can't restart the generator once it has already yielded
self.preamble.append(check)
if fundef is None:
raise FormatError("include_ast returned None")
# return a copy of the the cached ast tree, because it will be further modified to fit with the including template
fundef = copy.deepcopy(fundef)
_yieldall(fundef.body)
for expr in fundef.body:
self.visit(expr)
return fundef.body
elif type(node.value) is Yield:
y = node.value
if type(y.value) == Str:
if self.stripWhitespace:
s = strip_whitespace(y.value.s)
if len(s) == 0:
return None # dont even compile in the Expr(Yield) if it was only yielding white space
else:
y.value.s = s
elif type(y.value) == Call:
call = y.value
if type(call.func) is Name:
if self.seenFuncs.get(call.func.id, False) is not False: # was defined locally
# replace the Call with one to ''.join(Call)
y.value = _call(Attribute(value=Str(s=''), attr='join', ctx=Load()), [y.value])
ast.copy_location(y.value, node)
self.generic_visit(node)
return node
def visit_Str(self, node):
if "html" == self._get_template_type():
n = ast.Name(id='_q_htmltext', ctx=ast.Load())
ast.copy_location(n, node)
n = ast.Call(func=n, args=[node], keywords=[], starargs=None,
kwargs=None)
return ast.copy_location(n, node)
else:
return node
def visit_BinOp(self, node):
self.generic_visit(node)
if isinstance(node.op, Pow):
fnode = Name("pow", Load())
copy_location(fnode, node)
cnode = Call(fnode, [node.left, node.right], [], None, None)
copy_location(cnode, node)
return cnode
return node
def visit_TryFinally(self, node):
new_node = gast.Try(
self._visit(node.body),
[], # handlers
[], # orelse
self._visit(node.finalbody)
)
ast.copy_location(new_node, node)
return new_node
def visit_TryExcept(self, node):
new_node = gast.Try(
self._visit(node.body),
self._visit(node.handlers),
self._visit(node.orelse),
[] # finalbody
)
ast.copy_location(new_node, node)
return new_node
def visit_Return(self, node):
assign = ast.Assign(targets = [ast.Name(id = 'y' , ctx = ast.Store())], value = ast.Num(8))
ast.increment_lineno(node, 1)
ast.copy_location(assign, node)
ast.fix_missing_locations(assign)
#assign.col_offset = 8
# lists = list(ast.iter_child_nodes(assign))
# print lists
return assign
def visit_FunctionDef(self, node):
new_node = ast.FunctionDef(
self._visit(node.name),
self._visit(node.args),
self._visit(node.body),
self._visit(node.decorator_list),
)
ast.copy_location(new_node, node)
return new_node
def _seconds_to_mu(ref_period, node):
divided = ast.copy_location(
ast.BinOp(left=node,
op=ast.Div(),
right=value_to_ast(ref_period)),
node)
return ast.copy_location(
ast.Call(func=ast.Name("round64", ast.Load()),
args=[divided], keywords=[]),
divided)
def visit_Assert(self, assert_node):
if assert_node.msg is not None:
return assert_node
statements = AssertionChildVisitor().visit(assert_node.test)
for s in statements:
ast.copy_location(s, assert_node)
ast.fix_missing_locations(s)
return statements
def visit_ClassDef(self, node):
new_node = ast.ClassDef(
self._visit(node.name),
self._visit(node.bases),
self._visit(node.body),
self._visit(node.decorator_list),
)
ast.copy_location(new_node, node)
return new_node
def visit_With(self, node):
new_node = gast.With(
[gast.withitem(
self._visit(node.context_expr),
self._visit(node.optional_vars)
)],
self._visit(node.body)
)
ast.copy_location(new_node, node)
return new_node
def visit(self, node):
new_node = super().visit(node)
if new_node is not node:
return ast.copy_location(new_node, node)
return node
def visit_Num(self, node: ast.Num):
newname = f'__uu{self.id}'
self.gvars[newname] = node.n
self.id += 1
return ast.copy_location(ast.Name(id=newname, ctx=ast.Load()), node)
def _aug_assign(self, target, oper, value):
# transform a += 1 to a = a + 1, then we can use assign and eval
new_value = ast.BinOp(left=target, op=oper, right=value)
ast.copy_location(new_value, target)
self._assign(target, new_value)
def ast_call(node):
"""Visit and transform ast call node"""
distributed_mode = util_global.get_value("distributed_mode", "")
is_not_strategy = distributed_mode in ("horovod", "")
is_not_horovod = distributed_mode in ("tf_strategy", "")
convert_loss_scale_api(node)
if _call_name_match(node.func, "set_experimental_options"):
log_msg(
getattr(node, 'lineno', 'None'),
'change set_experimental_options(*) to set_experimental_options(experimental_options)'
)
node.args = [ast.Name(id='experimental_options', ctx=ast.Load())]
node.keywords = []
util_global.set_value('need_conver', True)
if isinstance(node.func,
ast.Name) and node.func.id == 'check_available_gpus':
log_msg(getattr(node, 'lineno', 'None'),
"change check_available_gpus() to ['/device:CPU:0']")
util_global.set_value('need_conver', True)
return ast.List(elts=[ast.Str(s="/device:CPU:0")], ctx=ast.Load())
if ((isinstance(node.func, ast.Name) and node.func.id == 'GraphOptions')
or (isinstance(node.func, ast.Attribute)
and node.func.attr == 'GraphOptions')):
log_success_report(getattr(node, 'lineno', 'None'), 'GraphOptions()')
src = copy.deepcopy(node)
node.func = ast.Name(id='npu_graph_options', ctx=ast.Load())
node.args = []
node.keywords = []
node.keywords.append(ast.keyword(arg='graph_options', value=src))
util_global.set_value('need_conver', True)
return node
if (isinstance(node.func, ast.Name) and node.func.id == 'OptimizerOptions') or \
(isinstance(node.func, ast.Attribute) and node.func.attr == 'OptimizerOptions'):
log_success_report(getattr(node, 'lineno', 'None'),
'OptimizerOptions()')
src = copy.deepcopy(node)
node.func = ast.Name(id='npu_optimizer_options', ctx=ast.Load())
node.args = []
node.keywords = []
node.keywords.append(ast.keyword(arg='optimizer_options', value=src))
util_global.set_value('need_conver', True)
return node
if _call_name_match(node.func, "Session"):
return convert_origin_func_to_npu(node, tf_func_map["tf.Session"],
"tf.Session", ["config"])
if _call_name_match(node.func, "InteractiveSession"):
return convert_origin_func_to_npu(node,
tf_func_map["tf.InteractiveSession"],
"tf.InteractiveSession", ["config"])
if isinstance(node.func, ast.Attribute
) and node.func.attr == "BroadcastGlobalVariablesHook":
if isinstance(node.func.value,
ast.Name) and node.func.value.id == "hvd":
if is_not_horovod:
log_strategy_distributed_mode_error(node)
return node
log_msg(
getattr(node, "lineno", "None"),
'change hvd.BroadcastGlobalVariablesHook to NPUBroadcastGlobalVariablesHook'
)
node = pasta.parse(
"NPUBroadcastGlobalVariablesHook(0, int(os.getenv('RANK_ID', '0')))"
)
util_global.set_value('need_conver', True)
return node
if isinstance(node.func, ast.Attribute
) and node.func.attr == "BroadcastGlobalVariablesCallback":
if isinstance(node.func.value,
ast.Attribute) and node.func.value.attr == "callbacks":
if is_not_horovod:
log_strategy_distributed_mode_error(node)
return node
log_msg(
getattr(node, "lineno", "None"),
'change hvd.callbacks.BroadcastGlobalVariablesCallback to NPUBroadcastGlobalVariablesCallback'
)
node = pasta.parse(
"NPUBroadcastGlobalVariablesCallback(root_rank=0)")
util_global.set_value('need_conver', True)
return node
if isinstance(node.func,
ast.Attribute) and node.func.attr == "DistributedOptimizer":
if isinstance(node.func.value,
ast.Name) and node.func.value.id == "hvd":
if is_not_horovod:
log_strategy_distributed_mode_error(node)
return node
return convert_hvd_distributed_api(node)
if isinstance(node.func, ast.Attribute) and node.func.attr == 'shard':
log_success_report(getattr(node, "lineno", "None"), 'shard')
node.args = [
pasta.parse("int(os.getenv('RANK_SIZE', '1'))"),
pasta.parse("int(os.getenv('RANK_ID', '0'))")
]
node.keywords.clear()
util_global.set_value('need_conver', True)
return node
if isinstance(node.func, ast.Attribute) and node.func.attr == 'dropout':
if isinstance(node.func.value,
ast.Attribute) and node.func.value.attr == 'nn':
for index, _ in enumerate(node.args):
if index == 2:
return node
for keyword in node.keywords:
if keyword.arg == "noise_shape":
return node
log_success_report(getattr(node, "lineno", "None"), 'dropout')
node.func = ast.Attribute(value=ast.Name(id='npu_ops',
ctx=ast.Load()),
attr='dropout',
ctx=ast.Load())
keywords_new = []
for keyword in node.keywords:
if keyword.arg != 'rate':
keywords_new.append(keyword)
else:
keywords_new.append(
ast.keyword(arg='keep_prob',
value=ast.BinOp(left=ast.Num(n=1),
op=ast.Sub(),
right=keyword.value)))
node.keywords = keywords_new
util_global.set_value('need_conver', True)
return node
if isinstance(node.func, ast.Attribute) and \
((node.func.attr == 'map_and_batch') or
(node.func.attr == 'batch' and (not isinstance(node.func.value, ast.Attribute) or (
isinstance(node.func.value, ast.Attribute) and node.func.value.attr != 'train')))):
exist = False
for keyword in node.keywords:
if keyword.arg == 'drop_remainder':
exist = True
if ((isinstance(keyword.value, ast.NameConstant)
and not keyword.value.value)
or (not isinstance(keyword.value, ast.NameConstant))):
log_success_report(getattr(node, "lineno", "None"),
node.func.attr)
keyword.value = pasta.parse('True')
util_global.set_value('need_conver', True)
if not exist:
log_success_report(getattr(node, "lineno", "None"), node.func.attr)
keyword = ast.keyword(arg='drop_remainder',
value=pasta.parse('True'))
node.keywords.insert(0, keyword)
util_global.set_value('need_conver', True)
return node
if (isinstance(node.func, ast.Attribute)
and isinstance(node.func.value, ast.Name)
and node.func.value.id == 'tf' and node.func.attr == 'device'):
log_success_report(getattr(node, "lineno", "None"), node.func.attr)
node.args = [ast.Str(s='/cpu:0')]
util_global.set_value('need_conver', True)
return node
if isinstance(node.func, ast.Attribute) and \
(node.func.attr == "get_distribution_strategy" or
node.func.attr == "MirroredStrategy" or
node.func.attr == "MultiWorkerMirroredStrategy"):
if is_not_strategy:
log_hvd_distributed_mode_error(node)
return node
log_success_report(getattr(node, "lineno", "None"), node.func.attr)
new_func = ast.Attribute(value=ast.Name(id="npu_strategy",
ctx=ast.Load()),
attr="NPUStrategy",
ctx=ast.Load())
ast.copy_location(new_func, node.func)
node.func = new_func
node.keywords = []
node.args = []
util_global.set_value('need_conver', True)
return node
if (isinstance(node.func, ast.Attribute) and (node.func.attr == 'RunConfig')) and \
(_call_name_match(node.func.value, 'estimator') or _call_name_match(node.func.value, 'tpu')):
if node.keywords.count("train_distribute") or node.keywords.count(
"eval_distribute"):
if is_not_strategy:
log_hvd_distributed_mode_error(node)
save_summary_steps = None
for keyword in node.keywords:
if keyword.arg == 'save_summary_steps':
save_summary_steps = keyword
break
if len(node.args) < 3 and not save_summary_steps:
log_msg(getattr(node, 'lineno'),
'RunConfig() add save_summary_steps=0')
util_global.set_value('need_conver', True)
node.keywords.append(
ast.keyword(arg='save_summary_steps', value=pasta.parse('0')))
return node
if isinstance(node.func, ast.Attribute) and (node.func.attr == 'TPUEstimator') and \
((isinstance(node.func.value, ast.Attribute) and (node.func.value.attr == 'tpu')) or
(isinstance(node.func.value, ast.Name) and (node.func.value.id == 'tpu'))):
add_eval_on_tpu = True
add_use_tpu = True
add_export_to_tpu = True
for keyword in node.keywords:
if (keyword.arg == 'eval_on_tpu') or (
keyword.arg == 'use_tpu') or (keyword.arg
== 'export_to_tpu'):
if (not isinstance(keyword.value, ast.NameConstant)) or \
(isinstance(keyword.value, ast.NameConstant) and (keyword.value.value)):
log_success_report(getattr(node, 'lineno', 'None'),
'TPUEstimator(' + keyword.arg + '=*)')
keyword.value = pasta.parse('False')
util_global.set_value('need_conver', True)
if add_eval_on_tpu and (keyword.arg == 'eval_on_tpu'):
add_eval_on_tpu = False
if add_use_tpu and (keyword.arg == 'use_tpu'):
add_use_tpu = False
if add_export_to_tpu and (keyword.arg == 'export_to_tpu'):
add_export_to_tpu = False
if add_eval_on_tpu:
log_success_report(getattr(node, 'lineno', 'None'),
'TPUEstimator(eval_on_tpu=*)')
node.keywords.append(
ast.keyword(arg='eval_on_tpu', value=pasta.parse('False')))
util_global.set_value('need_conver', True)
if add_use_tpu:
log_success_report(getattr(node, 'lineno', 'None'),
'TPUEstimator(use_tpu=*)')
node.keywords.append(
ast.keyword(arg='use_tpu', value=pasta.parse('False')))
util_global.set_value('need_conver', True)
if add_export_to_tpu:
log_success_report(getattr(node, 'lineno', 'None'),
'TPUEstimator(export_to_tpu=*)')
node.keywords.append(
ast.keyword(arg='export_to_tpu', value=pasta.parse('False')))
util_global.set_value('need_conver', True)
if isinstance(node.func,
ast.Attribute) and (node.func.attr
== 'VirtualDeviceConfiguration'):
log_success_report(getattr(node, 'lineno', 'None'),
'VirtualDeviceConfiguration')
util_global.set_value('need_conver', True)
memory_limit = None
for keyword in node.keywords:
if keyword.arg == 'memory_limit':
memory_limit = keyword
break
if memory_limit:
memory_limit.value = ast.NameConstant(value=None)
else:
node.keywords.append(
ast.keyword(arg='memory_limit',
value=ast.NameConstant(value=None)))
return node
if isinstance(node.func,
ast.Attribute) and (node.func.attr
== 'set_soft_device_placement'):
log_success_report(getattr(node, 'lineno', 'None'),
'set_soft_device_placement')
util_global.set_value('need_conver', True)
node.args = []
node.keywords = [
ast.keyword(arg='enabled', value=ast.NameConstant(value=True))
]
return node
if isinstance(node.func, ast.Attribute) and (node.func.attr
== 'set_memory_growth'):
log_success_report(getattr(node, 'lineno', 'None'),
'set_memory_growth')
util_global.set_value('need_conver', True)
node = ast.NameConstant(value=None)
return node
if isinstance(node.func,
ast.Attribute) and (node.func.attr
== 'set_virtual_device_configuration'):
log_success_report(getattr(node, 'lineno', 'None'),
'set_virtual_device_configuration')
util_global.set_value('need_conver', True)
node = ast.NameConstant(value=None)
return node
if isinstance(node.func, ast.Attribute) and (node.func.attr
== 'jit_scope'):
if isinstance(node.func.value, ast.Attribute) and (node.func.value.attr
== 'experimental'):
if isinstance(node.func.value.value,
ast.Attribute) and (node.func.value.value.attr
== 'xla'):
log_success_report(getattr(node, 'lineno', 'None'),
'*.xla.experimental.jit_scope')
util_global.set_value('need_conver', True)
compile_ops = None
for keyword in node.keywords:
if keyword.arg == 'compile_ops':
compile_ops = keyword
break
if compile_ops:
compile_ops.value = pasta.parse('False')
else:
node.keywords.append(
ast.keyword(arg='compile_ops',
value=pasta.parse('False')))
return node
for estimator in util_global.get_value('Estimators', []):
if (isinstance(node.func, ast.Attribute) and (node.func.attr == estimator)) \
or (isinstance(node.func, ast.Name) and (node.func.id == estimator)):
log_msg(
getattr(node, 'lineno'),
"".join([estimator, '() add config=npu_run_config_init()']))
config = None
for keyword in node.keywords:
if keyword.arg == 'config':
config = keyword
break
if config:
new_value = ast.Call(func=ast.Name(id='npu_run_config_init',
ctx=ast.Load()),
args=[],
keywords=[
ast.keyword(arg='run_config',
value=config.value)
])
ast.copy_location(new_value, config.value)
config.value = new_value
else:
node.keywords.append(
ast.keyword(arg='config',
value=pasta.parse('npu_run_config_init()')))
util_global.set_value('need_conver', True)
return node
if isinstance(node.func, ast.Attribute) and (node.func.attr
== 'clear_session'):
log_msg(getattr(node, 'lineno'),
"change keras.clear_session() to npu_clear_session()")
node = ast.Call(func=ast.Name(id='npu_clear_session', ctx=ast.Load()),
args=[],
keywords=[])
util_global.set_value('need_conver', True)
if _call_name_match(node.func, "MonitoredTrainingSession"):
return convert_origin_func_to_npu(
node, tf_func_map["tf.train.MonitoredTrainingSession"],
"MonitoredTrainingSession", ["config", "hooks"])
if isinstance(node.func,
ast.Attribute) and node.func.attr == "managed_session":
return convert_origin_func_to_npu(
node, tf_func_map["tf.train.Supervisor.managed_session"],
"managed_session", ["config"], True)
if distributed_mode == "tf_strategy": # this cond should be placed at the end of the Call function.
return convert_distributed_strategy_apis(node)
return node
def visit_Assign(self, node):
assert (len(node.targets) == 1)
self.generic_visit(node)
decorated = isinstance(node.value, ast.Call) and isinstance(
node.value.func, ast.Attribute) and isinstance(node.value.func.value, ast.Name) \
and node.value.func.value.id == 'ti'
is_static_assign = False
if decorated:
attr = node.value.func
if attr.attr == 'static':
is_static_assign = True
else:
pass # eg. x = ti.cast(xx) will reach here, but they're not decorators, so no raising errors here
if is_static_assign:
return node
if isinstance(node.targets[0], ast.Tuple):
targets = node.targets[0].elts
# Create
stmts = []
holder = self.parse_stmt('__tmp_tuple = ti.expr_init_list(0, '
f'{len(targets)})')
holder.value.args[0] = node.value
stmts.append(holder)
def tuple_indexed(i):
indexing = self.parse_stmt('__tmp_tuple[0]')
indexing.value.slice.value = self.parse_expr("{}".format(i))
return indexing.value
for i, target in enumerate(targets):
is_local = isinstance(target, ast.Name)
if is_local and self.is_creation(target.id):
var_name = target.id
target.ctx = ast.Store()
# Create
init = ast.Attribute(value=ast.Name(id='ti',
ctx=ast.Load()),
attr='expr_init',
ctx=ast.Load())
rhs = ast.Call(
func=init,
args=[tuple_indexed(i)],
keywords=[],
)
self.create_variable(var_name)
stmts.append(ast.Assign(targets=[target], value=rhs))
else:
# Assign
target.ctx = ast.Load()
func = ast.Attribute(value=target,
attr='assign',
ctx=ast.Load())
call = ast.Call(func=func,
args=[tuple_indexed(i)],
keywords=[])
stmts.append(ast.Expr(value=call))
for stmt in stmts:
ast.copy_location(stmt, node)
stmts.append(self.parse_stmt('del __tmp_tuple'))
return self.make_single_statement(stmts)
else:
is_local = isinstance(node.targets[0], ast.Name)
if is_local and self.is_creation(node.targets[0].id):
var_name = node.targets[0].id
# Create
init = ast.Attribute(value=ast.Name(id='ti', ctx=ast.Load()),
attr='expr_init',
ctx=ast.Load())
rhs = ast.Call(
func=init,
args=[node.value],
keywords=[],
)
self.create_variable(var_name)
return ast.copy_location(
ast.Assign(targets=node.targets, value=rhs), node)
else:
# Assign
node.targets[0].ctx = ast.Load()
func = ast.Attribute(value=node.targets[0],
attr='assign',
ctx=ast.Load())
call = ast.Call(func=func, args=[node.value], keywords=[])
return ast.copy_location(ast.Expr(value=call), node)
def visit_For(self, node):
if node.orelse:
raise TaichiSyntaxError(
"'else' clause for 'for' not supported in Taichi kernels")
decorated = isinstance(node.iter, ast.Call) and isinstance(
node.iter.func, ast.Attribute) and isinstance(node.iter.func.value, ast.Name) \
and node.iter.func.value.id == 'ti'
is_ndrange_for = False
is_static_for = False
is_grouped = False
if decorated:
attr = node.iter.func
if attr.attr == 'static':
is_static_for = True
elif attr.attr == 'grouped':
is_grouped = True
elif attr.attr == 'ndrange':
is_ndrange_for = True
else:
raise Exception('Not supported')
is_range_for = isinstance(node.iter, ast.Call) and isinstance(
node.iter.func, ast.Name) and node.iter.func.id == 'range'
ast.fix_missing_locations(node)
if not is_ndrange_for:
self.generic_visit(node, ['body'])
if is_ndrange_for:
template = '''
if ti.static(1):
__ndrange = 0
for __ndrange_I in range(0):
__I = __ndrange_I
'''
t = ast.parse(template).body[0]
t.body[0].value = node.iter
t.body[1].iter.args[0] = self.parse_expr(
'__ndrange.acc_dimensions[0]')
targets = node.target
if isinstance(targets, ast.Tuple):
targets = [name.id for name in targets.elts]
else:
targets = [targets.id]
loop_body = t.body[1].body
for i in range(len(targets)):
if i + 1 < len(targets):
stmt = '__{} = __I // __ndrange.acc_dimensions[{}]'.format(
targets[i], i + 1)
else:
stmt = '__{} = __I'.format(targets[i])
loop_body.append(self.parse_stmt(stmt))
stmt = '{} = __{} + __ndrange.bounds[{}][0]'.format(
targets[i], targets[i], i)
loop_body.append(self.parse_stmt(stmt))
if i + 1 < len(targets):
stmt = '__I = __I - __{} * __ndrange.acc_dimensions[{}]'.format(
targets[i], i + 1)
loop_body.append(self.parse_stmt(stmt))
loop_body += node.body
node = ast.copy_location(t, node)
return self.visit(node) # further translate as a range for
elif is_static_for:
return node
elif is_range_for:
loop_var = node.target.id
self.check_loop_var(loop_var)
template = '''
if 1:
{} = ti.Expr(ti.core.make_id_expr(''))
___begin = ti.Expr(0)
___end = ti.Expr(0)
___begin = ti.cast(___begin, ti.i32)
___end = ti.cast(___end, ti.i32)
ti.core.begin_frontend_range_for({}.ptr, ___begin.ptr, ___end.ptr)
ti.core.end_frontend_range_for()
'''.format(loop_var, loop_var)
t = ast.parse(template).body[0]
assert len(node.iter.args) in [1, 2]
if len(node.iter.args) == 2:
bgn = node.iter.args[0]
end = node.iter.args[1]
else:
bgn = self.make_constant(value=0)
end = node.iter.args[0]
t.body[1].value.args[0] = bgn
t.body[2].value.args[0] = end
t.body = t.body[:6] + node.body + t.body[6:]
t.body.append(self.parse_stmt('del {}'.format(loop_var)))
return ast.copy_location(t, node)
else: # Struct for
if isinstance(node.target, ast.Name):
elts = [node.target]
else:
elts = node.target.elts
for loop_var in elts:
self.check_loop_var(loop_var.id)
var_decl = ''.join(
' {} = ti.Expr(ti.core.make_id_expr(""))\n'.format(ind.id)
for ind in elts)
vars = ', '.join(ind.id for ind in elts)
if is_grouped:
template = '''
if 1:
___loop_var = 0
{} = ti.make_var_vector(size=___loop_var.loop_range().dim())
___expr_group = ti.make_expr_group({})
ti.core.begin_frontend_struct_for(___expr_group, ___loop_var.loop_range().ptr)
ti.core.end_frontend_range_for()
'''.format(vars, vars)
t = ast.parse(template).body[0]
cut = 4
t.body[0].value = node.iter
t.body = t.body[:cut] + node.body + t.body[cut:]
else:
template = '''
if 1:
{}
___loop_var = 0
___expr_group = ti.make_expr_group({})
ti.core.begin_frontend_struct_for(___expr_group, ___loop_var.loop_range().ptr)
ti.core.end_frontend_range_for()
'''.format(var_decl, vars)
t = ast.parse(template).body[0]
cut = len(elts) + 3
t.body[cut - 3].value = node.iter
t.body = t.body[:cut] + node.body + t.body[cut:]
for loop_var in reversed(elts):
t.body.append(self.parse_stmt('del {}'.format(loop_var.id)))
return ast.copy_location(t, node)
def visit(self, node):
method = 'visit_' + node.__class__.__name__
visitor = getattr(self, method, self.generic_visit)
return ast.copy_location(visitor(node), self.__benchmark)
def _unellipsify(self, node, slices, subscript_node):
"""
Given an array node `node`, process all AST slices and create the
final type:
- process newaxes (None or numpy.newaxis)
- replace Ellipsis with a bunch of ast.Slice objects
- process integer indices
- append any missing slices in trailing dimensions
"""
type = node.variable.type
if not type.is_array:
assert type.is_object
return minitypes.object_, node
if (len(slices) == 1 and self._is_constant_index(slices[0])
and slices[0].value.pyval is Ellipsis):
# A[...]
return type, node
result = []
seen_ellipsis = False
# Filter out newaxes
newaxes = [newaxis for newaxis in slices if self._is_newaxis(newaxis)]
n_indices = len(slices) - len(newaxes)
full_slice = ast.Slice(lower=None, upper=None, step=None)
full_slice.variable = Variable(numba_types.SliceType())
ast.copy_location(full_slice, slices[0])
# process ellipses and count integer indices
indices_seen = 0
for slice_node in slices[::-1]:
slice_type = slice_node.variable.type
if slice_type.is_ellipsis:
if seen_ellipsis:
result.append(full_slice)
else:
nslices = type.ndim - n_indices + 1
result.extend([full_slice] * nslices)
seen_ellipsis = True
elif (slice_type.is_slice or slice_type.is_int
or self._is_newaxis(slice_node)):
indices_seen += slice_type.is_int
result.append(slice_node)
else:
# TODO: Coerce all object operands to integer indices?
# TODO: (This will break indexing with the Ellipsis object or
# TODO: with slice objects that we couldn't infer)
return minitypes.object_, nodes.CoercionNode(
node, minitypes.object_)
# append any missing slices (e.g. a2d[:]
result_length = len(result) - len(newaxes)
if result_length < type.ndim:
nslices = type.ndim - result_length
result.extend([full_slice] * nslices)
result.reverse()
subscript_node.slice = ast.ExtSlice(result)
ast.copy_location(subscript_node.slice, slices[0])
# create the final array type and set it in value.variable
result_dtype = node.variable.type.dtype
result_ndim = node.variable.type.ndim + len(newaxes) - indices_seen
if result_ndim > 0:
result_type = result_dtype[(slice(None), ) * result_ndim]
elif result_ndim == 0:
result_type = result_dtype
else:
result_type = minitypes.object_
return result_type, node
def visit_FunctionDef(self, orig_ast: ast.FunctionDef) -> ast.AST:
cascades_body = []
# Compute selected features.
for node in selected_feature_nodes:
cascades_body += node.get_ast()
# Predict with approximate model.
approximate_model_ast = create_model_ast(
function_name=predict_proba_function.__name__,
output_name="__willump_approximate_preds",
input_names=selected_feature_names,
model_param="__willump_approximate_model")
cascades_body += approximate_model_ast
# Get indices that can't be approximated.
unapproximated_indices_ast = create_function_ast(
function_name="__willump_get_unapproximated_indices",
output_name="__willump_unapproximated_indices",
input_names=[
"__willump_approximate_preds",
"__willump_cascade_threshold"
])
cascades_body += unapproximated_indices_ast
# Only compute remaining features for unapproximated indices.
shortened_inputs = set()
for node in remaining_feature_nodes:
for input_name in node.input_names:
if input_name not in shortened_inputs:
shorten_ast = create_function_ast(
function_name=
"__willump_select_unapproximated_rows",
output_name=input_name,
input_names=[
input_name, "__willump_unapproximated_indices"
])
cascades_body += shorten_ast
shortened_inputs.add(input_name)
cascades_body += node.get_ast()
# Shorten the selected features.
for name in selected_feature_names:
shorten_ast = create_function_ast(
function_name="__willump_select_unapproximated_rows",
output_name=name,
input_names=[name, "__willump_unapproximated_indices"])
cascades_body += shorten_ast
# Predict with full model.
full_model_ast = create_model_ast(
function_name=predict_function.__name__,
output_name="__willump_full_preds",
input_names=model_node.input_names,
model_param="__willump_full_model")
cascades_body += full_model_ast
# Return combined predictions.
combine_predictions_ast = create_function_ast(
function_name="__willump_combine_predictions",
output_name="__willump_final_predictions",
input_names=[
"__willump_approximate_preds", "__willump_full_preds",
"__willump_cascade_threshold"
])
cascades_body += combine_predictions_ast
return_ast = ast.parse("return __willump_final_predictions",
"exec").body
cascades_body += return_ast
# Finalize AST.
new_ast = copy.deepcopy(orig_ast)
new_ast.body = cascades_body
# No recursion allowed!
new_ast.decorator_list = []
return ast.copy_location(new_ast, orig_ast)
def visit_Return(self, node, visit_count):
"""Visit a ``Return`` node."""
if not self.ordinal or visit_count in self.ordinal:
return ast.copy_location(self.inject(node), node)
self.generic_visit(node)
return node
def visit_With(self, node):
new_node = ast.With(self._visit(node.items[0].context_expr),
self._visit(node.items[0].optional_vars),
self._visit(node.body))
ast.copy_location(new_node, node)
return new_node
def _copy_location(newnode, node):
return ast.fix_missing_locations(ast.copy_location(newnode, node))
def visitName(self, node: ast.Name):
if node.id == "__debug__":
return copy_location(Constant(not self.optimize), node)
return self.generic_visit(node)
def passer(_: ast3.NodeTransformer, node: cls) -> None:
return ast.copy_location(ast.Pass(), node)
def visit_Assign(self, node):
assert (len(node.targets) == 1)
self.generic_visit(node)
if isinstance(node.targets[0], ast.Tuple):
targets = node.targets[0].elts
# Create
stmts = []
holder = self.parse_stmt('__tmp_tuple = 0')
holder.value = node.value
stmts.append(holder)
def tuple_indexed(i):
indexing = self.parse_stmt('__tmp_tuple[0]')
indexing.value.slice.value = self.parse_expr("{}".format(i))
return indexing.value
for i, target in enumerate(targets):
is_local = isinstance(target, ast.Name)
if is_local and self.is_creation(target.id):
var_name = target.id
target.ctx = ast.Store()
# Create
init = ast.Attribute(value=ast.Name(id='ti',
ctx=ast.Load()),
attr='expr_init',
ctx=ast.Load())
rhs = ast.Call(
func=init,
args=[tuple_indexed(i)],
keywords=[],
)
self.create_variable(var_name)
stmts.append(ast.Assign(targets=[target], value=rhs))
else:
# Assign
target.ctx = ast.Load()
func = ast.Attribute(value=target,
attr='assign',
ctx=ast.Load())
call = ast.Call(func=func,
args=[tuple_indexed(i)],
keywords=[])
stmts.append(ast.Expr(value=call))
for stmt in stmts:
ast.copy_location(stmt, node)
stmts.append(self.parse_stmt('del __tmp_tuple'))
return self.make_single_statement(stmts)
else:
is_local = isinstance(node.targets[0], ast.Name)
if is_local and self.is_creation(node.targets[0].id):
var_name = node.targets[0].id
# Create
init = ast.Attribute(value=ast.Name(id='ti', ctx=ast.Load()),
attr='expr_init',
ctx=ast.Load())
rhs = ast.Call(
func=init,
args=[node.value],
keywords=[],
)
self.create_variable(var_name)
return ast.copy_location(
ast.Assign(targets=node.targets, value=rhs), node)
else:
# Assign
node.targets[0].ctx = ast.Load()
func = ast.Attribute(value=node.targets[0],
attr='assign',
ctx=ast.Load())
call = ast.Call(func=func, args=[node.value], keywords=[])
return ast.copy_location(ast.Expr(value=call), node)
def fix_location(new, old):
ast.copy_location(new, old)
ast.fix_missing_locations(new)
return new
def visit_Assert(self, assert_: ast.Assert) -> List[ast.stmt]:
"""Return the AST statements to replace the ast.Assert instance.
This rewrites the test of an assertion to provide
intermediate values and replace it with an if statement which
raises an assertion error with a detailed explanation in case
the expression is false.
"""
if isinstance(assert_.test, ast.Tuple) and len(assert_.test.elts) >= 1:
from _pytest.warning_types import PytestAssertRewriteWarning
import warnings
# TODO: This assert should not be needed.
assert self.module_path is not None
warnings.warn_explicit(
PytestAssertRewriteWarning(
"assertion is always true, perhaps remove parentheses?"
),
category=None,
filename=self.module_path,
lineno=assert_.lineno,
)
self.statements: List[ast.stmt] = []
self.variables: List[str] = []
self.variable_counter = itertools.count()
if self.enable_assertion_pass_hook:
self.format_variables: List[str] = []
self.stack: List[Dict[str, ast.expr]] = []
self.expl_stmts: List[ast.stmt] = []
self.push_format_context()
# Rewrite assert into a bunch of statements.
top_condition, explanation = self.visit(assert_.test)
negation = ast.UnaryOp(ast.Not(), top_condition)
if self.enable_assertion_pass_hook: # Experimental pytest_assertion_pass hook
msg = self.pop_format_context(ast.Str(explanation))
# Failed
if assert_.msg:
assertmsg = self.helper("_format_assertmsg", assert_.msg)
gluestr = "\n>assert "
else:
assertmsg = ast.Str("")
gluestr = "assert "
err_explanation = ast.BinOp(ast.Str(gluestr), ast.Add(), msg)
err_msg = ast.BinOp(assertmsg, ast.Add(), err_explanation)
err_name = ast.Name("AssertionError", ast.Load())
fmt = self.helper("_format_explanation", err_msg)
exc = ast.Call(err_name, [fmt], [])
raise_ = ast.Raise(exc, None)
statements_fail = []
statements_fail.extend(self.expl_stmts)
statements_fail.append(raise_)
# Passed
fmt_pass = self.helper("_format_explanation", msg)
orig = _get_assertion_exprs(self.source)[assert_.lineno]
hook_call_pass = ast.Expr(
self.helper(
"_call_assertion_pass",
ast.Num(assert_.lineno),
ast.Str(orig),
fmt_pass,
)
)
# If any hooks implement assert_pass hook
hook_impl_test = ast.If(
self.helper("_check_if_assertion_pass_impl"),
self.expl_stmts + [hook_call_pass],
[],
)
statements_pass = [hook_impl_test]
# Test for assertion condition
main_test = ast.If(negation, statements_fail, statements_pass)
self.statements.append(main_test)
if self.format_variables:
variables = [
ast.Name(name, ast.Store()) for name in self.format_variables
]
clear_format = ast.Assign(variables, ast.NameConstant(None))
self.statements.append(clear_format)
else: # Original assertion rewriting
# Create failure message.
body = self.expl_stmts
self.statements.append(ast.If(negation, body, []))
if assert_.msg:
assertmsg = self.helper("_format_assertmsg", assert_.msg)
explanation = "\n>assert " + explanation
else:
assertmsg = ast.Str("")
explanation = "assert " + explanation
template = ast.BinOp(assertmsg, ast.Add(), ast.Str(explanation))
msg = self.pop_format_context(template)
fmt = self.helper("_format_explanation", msg)
err_name = ast.Name("AssertionError", ast.Load())
exc = ast.Call(err_name, [fmt], [])
raise_ = ast.Raise(exc, None)
body.append(raise_)
# Clear temporary variables by setting them to None.
if self.variables:
variables = [ast.Name(name, ast.Store()) for name in self.variables]
clear = ast.Assign(variables, ast.NameConstant(None))
self.statements.append(clear)
# Fix locations (line numbers/column offsets).
for stmt in self.statements:
for node in traverse_node(stmt):
ast.copy_location(node, assert_)
return self.statements
def dyn(n):
return ast.copy_location(ast.Name(id='Any', ctx=ast.Load()), n)
def convert_distributed_strategy_apis(node):
"""Convert distributed strategy API"""
if isinstance(node.func, ast.Attribute) and isinstance(
node.func.value, ast.Attribute):
if ("Optimizer" in node.func.attr
and node.func.attr != "ScipyOptimizerInterface"
and node.func.attr != "MixedPrecisionLossScaleOptimizer"):
log_msg(getattr(node, "lineno", "None"),
"add npu distribute optimizer to tensorflow optimizer")
new_node = ast.Call(func=ast.Name(
id="npu_distributed_optimizer_wrapper", ctx=ast.Load()),
args=[node],
keywords=[])
ast.copy_location(new_node, node)
util_global.set_value('need_conver', True)
return new_node
if isinstance(
node.func, ast.Name
) and "Optimizer" in node.func.id and node.func.id != "NPULossScaleOptimizer":
log_msg(getattr(node, "lineno", "None"),
"add npu distribute optimizer to tensorflow optimizer")
new_node = ast.Call(func=ast.Name(
id="npu_distributed_optimizer_wrapper", ctx=ast.Load()),
args=[node],
keywords=[])
ast.copy_location(new_node, node)
util_global.set_value('need_conver', True)
return new_node
if _call_name_match(node.func, "TrainSpec"):
return convert_origin_func_to_npu(
node, tf_func_map["tf.estimator.TrainSpec"], "TrainSpec",
["hooks"])
if _call_name_match(node.func, "EvalSpec"):
return convert_origin_func_to_npu(node,
tf_func_map["tf.estimator.EvalSpec"],
"EvalSpec", ["hooks"])
if isinstance(node.func, ast.Attribute) and node.func.attr == "train":
if isinstance(node.func.value,
ast.Attribute) and node.func.value.attr == "learning":
return node
return convert_origin_func_to_npu(
node, tf_func_map["tf.estimator.Estimator.train"],
"Estimator.train", ["hooks"], True)
if isinstance(node.func, ast.Attribute) and (node.func.attr == 'compile'):
if isinstance(node.func.value,
ast.Name) and node.func.value.id == "re":
return node
return convert_origin_func_to_npu(
node, tf_func_map["tf.keras.Model.compile"], "Model.compile",
["optimizer"], True)
if isinstance(node.func, ast.Attribute) and node.func.attr == "fit":
return convert_origin_func_to_npu(node,
tf_func_map["tf.keras.Model.fit"],
"Model.fit", ["callbacks"], True)
if isinstance(node.func,
ast.Attribute) and node.func.attr == "fit_generator":
return convert_origin_func_to_npu(
node, tf_func_map["tf.keras.Model.fit_generator"],
"Model.fit_generator", ["callbacks"], True)
if isinstance(node.func, ast.Attribute) and node.func.attr == "gradients" and \
isinstance(node.func.value, ast.Name) and node.func.value.id == "tf":
return convert_tf_gradient_distributed(node)
return node
def visit_Subscript(self, node: ast.Subscript):
if rname(node) in self.keywords:
return ast.copy_location(node.value, node)
return self.generic_visit(node)
def visit_BinOp(self, node):
self.generic_visit(node)
self.binop_count += 1
# Check if this is the node we want to alter. We can accomplish this by
# keeping track of a counter, which we increment every time encounter
# a BinOp. Since the traversal through the AST is deterministic using the visitor
# pattern (IT IS NOT DETERMINISTIC IF YOU USE ast.walk), we can identify AST nodes
# uniquely by the value of the counter
if (self.binop_count == self.count_of_node_to_mutate):
# We make sure to use deepcopy so that we preserve all extra
# information we don't explicitly modify
new_node = copy.deepcopy(node)
ast.copy_location(new_node, node)
# figure out a way to randomize what operator it transforms to based on the current operator
if isinstance(node.op, ast.Add):
#randomly generate a number which will associate to a certain type of transformation
num = random.randint(0, 2)
print('random number', num)
if num == 0:
new_node.op = ast.Mult()
if num == 1:
new_node.op = ast.Sub()
if num == 2:
new_node.op = ast.Div()
if isinstance(node.op, ast.Mult):
num = random.randint(0, 3)
if num == 0:
new_node.op = ast.Div()
if num == 1:
new_node.op = ast.Add()
if num == 2:
new_node.op = ast.FloorDiv()
if num == 3:
new_node.op = ast.Sub()
if isinstance(node.op, ast.Div):
num = random.randint(0, 2)
if num == 0:
new_node.op = ast.FloorDiv()
if num == 1:
new_node.op = ast.Mult()
if num == 2:
new_node.op = ast.Add()
if isinstance(node.op, ast.Sub):
num = random.randint(0, 2)
if num == 0:
new_node.op = ast.Add()
if num == 1:
new_node.op = ast.Mult()
if num == 2:
new_node.op = ast.Div()
if isinstance(node.op, ast.FloorDiv):
new_node.op = ast.Div()
print('I AM CREATING A NEW NODE HERE', self.binop_count)
return new_node
else:
# If we're not looking at an add node we want to change, don't modify
# this node whatsoever
return node
def visit_Str(self, node):
return ast.copy_location(ast.Str(s=node.s.lower()), node)
def visit_For(self, node):
if node.orelse:
raise TaichiSyntaxError(
"'else' clause for 'for' not supported in Taichi kernels")
self.generic_visit(node, ['body'])
decorated = isinstance(node.iter, ast.Call) and isinstance(
node.iter.func, ast.Attribute)
is_static_for = False
is_grouped = False
if decorated:
attr = node.iter.func
if attr.attr == 'static':
is_static_for = True
elif attr.attr == 'grouped':
is_grouped = True
is_range_for = isinstance(node.iter, ast.Call) and isinstance(
node.iter.func, ast.Name) and node.iter.func.id == 'range'
if is_static_for:
return node
elif is_range_for:
loop_var = node.target.id
self.check_loop_var(loop_var)
template = '''
if 1:
{} = ti.Expr(ti.core.make_id_expr(''))
___begin = ti.Expr(0)
___end = ti.Expr(0)
___begin = ti.cast(___begin, ti.i32)
___end = ti.cast(___end, ti.i32)
ti.core.begin_frontend_range_for({}.ptr, ___begin.ptr, ___end.ptr)
ti.core.end_frontend_range_for()
'''.format(loop_var, loop_var)
t = ast.parse(template).body[0]
assert len(node.iter.args) in [1, 2]
if len(node.iter.args) == 2:
bgn = node.iter.args[0]
end = node.iter.args[1]
else:
bgn = self.make_constant(value=0)
end = node.iter.args[0]
t.body[1].value.args[0] = bgn
t.body[2].value.args[0] = end
t.body = t.body[:6] + node.body + t.body[6:]
t.body.append(self.parse_stmt('del {}'.format(loop_var)))
return ast.copy_location(t, node)
else: # Struct for
if isinstance(node.target, ast.Name):
elts = [node.target]
else:
elts = node.target.elts
for loop_var in elts:
self.check_loop_var(loop_var.id)
var_decl = ''.join(
' {} = ti.Expr(ti.core.make_id_expr(""))\n'.format(ind.id)
for ind in elts)
vars = ', '.join(ind.id for ind in elts)
if is_grouped:
template = '''
if 1:
___loop_var = 0
{} = ti.make_var_vector(size=___loop_var.loop_range().dim())
___expr_group = ti.make_expr_group({})
ti.core.begin_frontend_struct_for(___expr_group, ___loop_var.loop_range().ptr)
ti.core.end_frontend_range_for()
'''.format(vars, vars)
t = ast.parse(template).body[0]
cut = 4
t.body[0].value = node.iter
t.body = t.body[:cut] + node.body + t.body[cut:]
else:
template = '''
if 1:
{}
___loop_var = 0
___expr_group = ti.make_expr_group({})
ti.core.begin_frontend_struct_for(___expr_group, ___loop_var.loop_range().ptr)
ti.core.end_frontend_range_for()
'''.format(var_decl, vars)
t = ast.parse(template).body[0]
cut = len(elts) + 3
t.body[cut - 3].value = node.iter
t.body = t.body[:cut] + node.body + t.body[cut:]
for loop_var in reversed(elts):
t.body.append(self.parse_stmt('del {}'.format(loop_var.id)))
return ast.copy_location(t, node)
def visit_Assign(self, node):
target = rname(node.targets[-1])
if target not in self.memlets:
return self.generic_visit(node)
memlet, nc, wcr, dtype = self.memlets[target]
value = self.visit(node.value)
if not isinstance(node.targets[-1], ast.Subscript):
# Dynamic accesses or streams -> every access counts
try:
if memlet and memlet.data and (memlet.dynamic or isinstance(
self.sdfg.arrays[memlet.data], data.Stream)):
if wcr is not None:
newnode = ast.Name(
id=self.codegen.write_and_resolve_expr(
self.sdfg,
memlet,
nc,
target,
cppunparse.cppunparse(value,
expr_semicolon=False),
dtype=dtype))
node.value = ast.copy_location(newnode, node.value)
return node
elif isinstance(self.sdfg.arrays[memlet.data], data.Stream):
newnode = ast.Name(id="%s.push(%s);" % (
memlet.data,
cppunparse.cppunparse(value, expr_semicolon=False),
))
else:
var_type, ctypedef = self.codegen._dispatcher.defined_vars.get(
memlet.data)
if var_type == DefinedType.Scalar:
newnode = ast.Name(id="%s = %s;" % (
memlet.data,
cppunparse.cppunparse(value,
expr_semicolon=False),
))
else:
newnode = ast.Name(id="%s = %s;" % (
cpp_array_expr(self.sdfg, memlet),
cppunparse.cppunparse(value,
expr_semicolon=False),
))
return self._replace_assignment(newnode, node)
except TypeError: # cannot determine truth value of Relational
pass
return self.generic_visit(node)
subscript = self._subscript_expr(node.targets[-1].slice, target)
if wcr is not None:
newnode = ast.Name(id=self.codegen.write_and_resolve_expr(
self.sdfg,
memlet,
nc,
target,
cppunparse.cppunparse(value, expr_semicolon=False),
indices=sym2cpp(subscript),
dtype=dtype) + ';')
else:
newnode = ast.Name(
id="%s[%s] = %s;" %
(target, sym2cpp(subscript),
cppunparse.cppunparse(value, expr_semicolon=False)))
return self._replace_assignment(newnode, node)
def visit_Attribute(self, node):
return ast.copy_location(
ast.Attribute(value=self.visit(node.value),
attr=node.attr.lower()), node)
def apply(self, state: SDFGState, sdfg: SDFG):
input: nodes.AccessNode = self.input
tasklet: nodes.Tasklet = self.tasklet
output: nodes.AccessNode = self.output
# If state fission is necessary to keep semantics, do it first
if (self.expr_index == 0 and state.in_degree(input) > 0
and state.out_degree(output) == 0):
newstate = sdfg.add_state_after(state)
newstate.add_node(tasklet)
new_input, new_output = None, None
# Keep old edges for after we remove tasklet from the original state
in_edges = list(state.in_edges(tasklet))
out_edges = list(state.out_edges(tasklet))
for e in in_edges:
r = newstate.add_read(e.src.data)
newstate.add_edge(r, e.src_conn, e.dst, e.dst_conn, e.data)
if e.src is input:
new_input = r
for e in out_edges:
w = newstate.add_write(e.dst.data)
newstate.add_edge(e.src, e.src_conn, w, e.dst_conn, e.data)
if e.dst is output:
new_output = w
# Remove tasklet and resulting isolated nodes
state.remove_node(tasklet)
for e in in_edges:
if state.degree(e.src) == 0:
state.remove_node(e.src)
for e in out_edges:
if state.degree(e.dst) == 0:
state.remove_node(e.dst)
# Reset state and nodes for rest of transformation
input = new_input
output = new_output
state = newstate
# End of state fission
if self.expr_index == 0:
inedges = state.edges_between(input, tasklet)
outedge = state.edges_between(tasklet, output)[0]
else:
me = self.map_entry
mx = self.map_exit
inedges = state.edges_between(me, tasklet)
outedge = state.edges_between(tasklet, mx)[0]
# Get relevant output connector
outconn = outedge.src_conn
ops = '[%s]' % ''.join(
re.escape(o) for o in AugAssignToWCR._EXPRESSIONS)
# Change tasklet code
if tasklet.language is dtypes.Language.Python:
# Match a single assignment with a binary operation as RHS
ast_node: ast.Assign = tasklet.code.code[0]
lhs: ast.Name = ast_node.targets[0]
rhs: ast.BinOp = ast_node.value
op = AugAssignToWCR._PYOP_MAP[type(rhs.op)]
inconns = list(edge.dst_conn for edge in inedges)
for n in (rhs.left, rhs.right):
if isinstance(n, ast.Name) and n.id in inconns:
inedge = inedges[inconns.index(n.id)]
else:
new_rhs = n
new_node = ast.copy_location(
ast.Assign(targets=[lhs], value=new_rhs), ast_node)
tasklet.code.code = [new_node]
elif tasklet.language is dtypes.Language.CPP:
cstr = tasklet.code.as_string.strip()
for edge in inedges:
inconn = edge.dst_conn
match = re.match(
r'^\s*%s\s*=\s*%s\s*(%s)(.*);$' %
(re.escape(outconn), re.escape(inconn), ops), cstr)
if match is None:
# match = re.match(
# r'^\s*%s\s*=\s*(.*)\s*(%s)\s*%s;$' %
# (re.escape(outconn), ops, re.escape(inconn)), cstr)
# if match is None:
continue
# op = match.group(2)
# expr = match.group(1)
else:
op = match.group(1)
expr = match.group(2)
if edge.data.subset != outedge.data.subset:
continue
# Map asymmetric WCRs to symmetric ones if possible
if op in AugAssignToWCR._EXPR_MAP:
op, newexpr = AugAssignToWCR._EXPR_MAP[op]
expr = newexpr.format(expr=expr)
tasklet.code.code = '%s = %s;' % (outconn, expr)
inedge = edge
break
else:
raise NotImplementedError
# Change output edge
outedge.data.wcr = f'lambda a,b: a {op} b'
if self.expr_index == 0:
# Remove input node and connector
state.remove_edge_and_connectors(inedge)
if state.degree(input) == 0:
state.remove_node(input)
else:
# Remove input edge and dst connector, but not necessarily src
state.remove_memlet_path(inedge)
# If outedge leads to non-transient, and this is a nested SDFG,
# propagate outwards
sd = sdfg
while (not sd.arrays[outedge.data.data].transient
and sd.parent_nsdfg_node is not None):
nsdfg = sd.parent_nsdfg_node
nstate = sd.parent
sd = sd.parent_sdfg
outedge = next(
iter(nstate.out_edges_by_connector(nsdfg, outedge.data.data)))
for outedge in nstate.memlet_path(outedge):
outedge.data.wcr = f'lambda a,b: a {op} b'
def visit_Name(self, node):
return ast.copy_location(ast.Name(id=node.id.lower()), node)
def replace(self, node, new_node):
copy_location(new_node, node)
NodeVisitor.generic_visit(self, new_node)
return new_node
def visit_arg(self, node):
new_node = ast.Name(node.arg, ast.Param())
ast.copy_location(new_node, node)
return new_node
def visit_Raise(self, node):
if node.cause:
raise error.NumbaError(node, "Cause to 'raise' not supported")
newnode = Raise(type=node.exc, inst=None, tback=None)
return ast.copy_location(newnode, node)
