"""Flow Graph Simplification 'Syntactic-ish' simplifications on a flow graph. simplify_graph() applies all simplifications defined in this file. """ import py from pypy.objspace.flow.model import SpaceOperation from pypy.objspace.flow.model import Variable, Constant, Block, Link from pypy.objspace.flow.model import c_last_exception from pypy.objspace.flow.model import checkgraph, traverse, mkentrymap from pypy.rpython.lltypesystem import lloperation, lltype from pypy.rpython.ootypesystem import ootype def get_funcobj(func): """ Return an object which is supposed to have attributes such as graph and _callable """ if hasattr(func, '_obj'): return func._obj # lltypesystem else: return func # ootypesystem def get_functype(TYPE): if isinstance(TYPE, lltype.Ptr): return TYPE.TO elif isinstance(TYPE, (ootype.StaticMethod, ootype.ForwardReference)): return TYPE assert False def get_graph(arg, translator): from pypy.translator.translator import graphof if isinstance(arg, Variable): return None f = arg.value from pypy.rpython.lltypesystem import lltype from pypy.rpython.ootypesystem import ootype if not isinstance(f, lltype._ptr) and not isinstance(f, ootype._callable): return None funcobj = get_funcobj(f) try: callable = funcobj._callable except (AttributeError, KeyError, AssertionError): return None try: return funcobj.graph except AttributeError: return None try: callable = funcobj._callable return graphof(translator, callable) except (AttributeError, KeyError, AssertionError): return None def replace_exitswitch_by_constant(block, const): assert isinstance(const, Constant) assert const != c_last_exception newexits = [link for link in block.exits if link.exitcase == const.value] if len(newexits) == 0: newexits = [link for link in block.exits if link.exitcase == 'default'] assert len(newexits) == 1 newexits[0].exitcase = None if hasattr(newexits[0], 'llexitcase'): newexits[0].llexitcase = None block.exitswitch = None block.recloseblock(*newexits) return newexits # ____________________________________________________________ def eliminate_empty_blocks(graph): """Eliminate basic blocks that do not contain any operations. When this happens, we need to replace the preceeding link with the following link. Arguments of the links should be updated.""" def visit(link): if isinstance(link, Link): while not link.target.operations: block1 = link.target if block1.exitswitch is not None: break if not block1.exits: break exit = block1.exits[0] assert block1 is not exit.target, ( "the graph contains an empty infinite loop") outputargs = [] for v in exit.args: if isinstance(v, Variable): # this variable is valid in the context of block1 # but it must come from 'link' i = block1.inputargs.index(v) v = link.args[i] outputargs.append(v) link.args = outputargs link.target = exit.target # the while loop above will simplify recursively the new link traverse(visit, graph) def transform_ovfcheck(graph): """The special function calls ovfcheck and ovfcheck_lshift need to be translated into primitive operations. ovfcheck is called directly after an operation that should be turned into an overflow-checked version. It is considered a syntax error if the resulting _ovf is not defined in objspace/flow/objspace.py. ovfcheck_lshift is special because there is no preceding operation. Instead, it will be replaced by an OP_LSHIFT_OVF operation. """ from pypy.rlib.rarithmetic import ovfcheck, ovfcheck_lshift from pypy.objspace.flow.objspace import implicit_exceptions covf = Constant(ovfcheck) covfls = Constant(ovfcheck_lshift) def check_syntax(opname): exlis = implicit_exceptions.get("%s_ovf" % (opname,), []) if OverflowError not in exlis: raise Exception("ovfcheck in %s: Operation %s has no" " overflow variant" % (graph.name, opname)) for block in graph.iterblocks(): for i in range(len(block.operations)-1, -1, -1): op = block.operations[i] if op.opname != 'simple_call': continue if op.args[0] == covf: if i == 0: # hard case: ovfcheck() on an operation that occurs # in the previous block, like 'floordiv'. The generic # exception handling around the ovfcheck() is enough # to cover all cases; kill the one around the previous op. entrymap = mkentrymap(graph) links = entrymap[block] assert len(links) == 1 prevblock = links[0].prevblock assert prevblock.exits[0].target is block prevblock.exitswitch = None prevblock.exits = (links[0],) join_blocks(graph) # merge the two blocks together transform_ovfcheck(graph) # ...and try again return op1 = block.operations[i-1] check_syntax(op1.opname) op1.opname += '_ovf' del block.operations[i] block.renamevariables({op.result: op1.result}) elif op.args[0] == covfls: op.opname = 'lshift_ovf' del op.args[0] def simplify_exceptions(graph): """The exception handling caused by non-implicit exceptions starts with an exitswitch on Exception, followed by a lengthy chain of is_/issubtype tests. We collapse them all into the block's single list of exits. """ clastexc = c_last_exception renaming = {} def rename(v): return renaming.get(v, v) def visit(block): if not (isinstance(block, Block) and block.exitswitch == clastexc and block.exits[-1].exitcase is Exception): return covered = [link.exitcase for link in block.exits[1:-1]] seen = [] preserve = list(block.exits[:-1]) exc = block.exits[-1] last_exception = exc.last_exception last_exc_value = exc.last_exc_value query = exc.target switches = [] # collect the targets while len(query.exits) == 2: newrenaming = {} for lprev, ltarg in zip(exc.args, query.inputargs): newrenaming[ltarg] = rename(lprev) op = query.operations[0] if not (op.opname in ("is_", "issubtype") and newrenaming.get(op.args[0]) == last_exception): break renaming.update(newrenaming) case = query.operations[0].args[-1].value assert issubclass(case, py.builtin.BaseException) lno, lyes = query.exits assert lno.exitcase == False and lyes.exitcase == True if case not in seen: is_covered = False for cov in covered: if issubclass(case, cov): is_covered = True break if not is_covered: switches.append( (case, lyes) ) seen.append(case) exc = lno query = exc.target if Exception not in seen: switches.append( (Exception, exc) ) # construct the block's new exits exits = [] for case, oldlink in switches: link = oldlink.copy(rename) assert case is not None link.last_exception = last_exception link.last_exc_value = last_exc_value # make the above two variables unique renaming2 = {} def rename2(v): return renaming2.get(v, v) for v in link.getextravars(): renaming2[v] = Variable(v) link = link.copy(rename2) link.exitcase = case link.prevblock = block exits.append(link) block.recloseblock(*(preserve + exits)) traverse(visit, graph) def transform_xxxitem(graph): # xxx setitem too for block in graph.iterblocks(): if block.operations and block.exitswitch == c_last_exception: last_op = block.operations[-1] if last_op.opname == 'getitem': postfx = [] for exit in block.exits: if exit.exitcase is IndexError: postfx.append('idx') elif exit.exitcase is KeyError: postfx.append('key') if postfx: last_op.opname = last_op.opname + '_' + '_'.join(postfx) def remove_dead_exceptions(graph): """Exceptions can be removed if they are unreachable""" clastexc = c_last_exception def issubclassofmember(cls, seq): for member in seq: if member and issubclass(cls, member): return True return False def visit(block): if not (isinstance(block, Block) and block.exitswitch == clastexc): return exits = [] seen = [] for link in block.exits: case = link.exitcase # check whether exceptions are shadowed if issubclassofmember(case, seen): continue # see if the previous case can be merged while len(exits) > 1: prev = exits[-1] if not (issubclass(prev.exitcase, link.exitcase) and prev.target is link.target and prev.args == link.args): break exits.pop() exits.append(link) seen.append(case) block.recloseblock(*exits) traverse(visit, graph) def join_blocks(graph): """Links can be deleted if they are the single exit of a block and the single entry point of the next block. When this happens, we can append all the operations of the following block to the preceeding block (but renaming variables with the appropriate arguments.) """ entrymap = mkentrymap(graph) block = graph.startblock seen = {block: True} stack = list(block.exits) while stack: link = stack.pop() if (link.prevblock.exitswitch is None and len(entrymap[link.target]) == 1 and link.target.exits): # stop at the returnblock assert len(link.prevblock.exits) == 1 renaming = {} for vprev, vtarg in zip(link.args, link.target.inputargs): renaming[vtarg] = vprev def rename(v): return renaming.get(v, v) def rename_op(op): args = [rename(a) for a in op.args] op = SpaceOperation(op.opname, args, rename(op.result), op.offset) # special case... if op.opname == 'indirect_call': if isinstance(op.args[0], Constant): assert isinstance(op.args[-1], Constant) del op.args[-1] op.opname = 'direct_call' return op for op in link.target.operations: link.prevblock.operations.append(rename_op(op)) exits = [] for exit in link.target.exits: newexit = exit.copy(rename) exits.append(newexit) newexitswitch = rename(link.target.exitswitch) link.prevblock.exitswitch = newexitswitch link.prevblock.recloseblock(*exits) if isinstance(newexitswitch, Constant) and newexitswitch != c_last_exception: exits = replace_exitswitch_by_constant(link.prevblock, newexitswitch) stack.extend(exits) else: if link.target not in seen: stack.extend(link.target.exits) seen[link.target] = True def remove_assertion_errors(graph): """Remove branches that go directly to raising an AssertionError, assuming that AssertionError shouldn't occur at run-time. Note that this is how implicit exceptions are removed (see _implicit_ in flowcontext.py). """ def visit(block): if isinstance(block, Block): for i in range(len(block.exits)-1, -1, -1): exit = block.exits[i] if not (exit.target is graph.exceptblock and exit.args[0] == Constant(AssertionError)): continue # can we remove this exit without breaking the graph? if len(block.exits) < 2: break if block.exitswitch == c_last_exception: if exit.exitcase is None: break if len(block.exits) == 2: # removing the last non-exceptional exit block.exitswitch = None exit.exitcase = None # remove this exit lst = list(block.exits) del lst[i] block.recloseblock(*lst) traverse(visit, graph) # _____________________________________________________________________ # decide whether a function has side effects def op_has_side_effects(op): return lloperation.LL_OPERATIONS[op.opname].sideeffects def has_no_side_effects(translator, graph, seen=None): #is the graph specialized? if no we can't say anything #don't cache the result though if translator.rtyper is None: return False else: if graph.startblock not in translator.rtyper.already_seen: return False if seen is None: seen = {} elif graph in seen: return True newseen = seen.copy() newseen[graph] = True for block in graph.iterblocks(): if block is graph.exceptblock: return False # graphs explicitly raising have side-effects for op in block.operations: if rec_op_has_side_effects(translator, op, newseen): return False return True def rec_op_has_side_effects(translator, op, seen=None): if op.opname == "direct_call": g = get_graph(op.args[0], translator) if g is None: return True if not has_no_side_effects(translator, g, seen): return True elif op.opname == "indirect_call": graphs = op.args[-1].value if graphs is None: return True for g in graphs: if not has_no_side_effects(translator, g, seen): return True else: return op_has_side_effects(op) # ___________________________________________________________________________ # remove operations if their result is not used and they have no side effects def transform_dead_op_vars(graph, translator=None): """Remove dead operations and variables that are passed over a link but not used in the target block. Input is a graph.""" return transform_dead_op_vars_in_blocks(list(graph.iterblocks()), translator) # the set of operations that can safely be removed # (they have no side effects, at least in R-Python) CanRemove = {} for _op in ''' newtuple newlist newdict is_true is_ id type issubtype repr str len hash getattr getitem pos neg nonzero abs hex oct ord invert add sub mul truediv floordiv div mod divmod pow lshift rshift and_ or_ xor int float long lt le eq ne gt ge cmp coerce contains iter get'''.split(): CanRemove[_op] = True from pypy.rpython.lltypesystem.lloperation import enum_ops_without_sideeffects for _op in enum_ops_without_sideeffects(): CanRemove[_op] = True del _op CanRemoveBuiltins = { isinstance: True, hasattr: True, } def transform_dead_op_vars_in_blocks(blocks, translator=None): """Remove dead operations and variables that are passed over a link but not used in the target block. Input is a set of blocks""" read_vars = {} # set of variables really used variable_flow = {} # map {Var: list-of-Vars-it-depends-on} def canremove(op, block): if op.opname not in CanRemove: return False if block.exitswitch != c_last_exception: return True # cannot remove the exc-raising operation return op is not block.operations[-1] # compute variable_flow and an initial read_vars for block in blocks: # figure out which variables are ever read for op in block.operations: if not canremove(op, block): # mark the inputs as really needed for arg in op.args: read_vars[arg] = True else: # if CanRemove, only mark dependencies of the result # on the input variables deps = variable_flow.setdefault(op.result, []) deps.extend(op.args) if isinstance(block.exitswitch, Variable): read_vars[block.exitswitch] = True if block.exits: for link in block.exits: if link.target not in blocks: for arg, targetarg in zip(link.args, link.target.inputargs): read_vars[arg] = True read_vars[targetarg] = True else: for arg, targetarg in zip(link.args, link.target.inputargs): deps = variable_flow.setdefault(targetarg, []) deps.append(arg) else: # return and except blocks implicitely use their input variable(s) for arg in block.inputargs: read_vars[arg] = True # an input block's inputargs should not be modified, even if some # of the function's input arguments are not actually used if block.isstartblock: for arg in block.inputargs: read_vars[arg] = True # flow read_vars backwards so that any variable on which a read_vars # depends is also included in read_vars def flow_read_var_backward(pending): pending = list(pending) for var in pending: for prevvar in variable_flow.get(var, []): if prevvar not in read_vars: read_vars[prevvar] = True pending.append(prevvar) flow_read_var_backward(read_vars) for block in blocks: # look for removable operations whose result is never used for i in range(len(block.operations)-1, -1, -1): op = block.operations[i] if op.result not in read_vars: if canremove(op, block): del block.operations[i] elif op.opname == 'simple_call': # XXX we want to have a more effective and safe # way to check if this operation has side effects # ... if op.args and isinstance(op.args[0], Constant): func = op.args[0].value try: if func in CanRemoveBuiltins: del block.operations[i] except TypeError: # func is not hashable pass elif op.opname == 'direct_call': if translator is not None: graph = get_graph(op.args[0], translator) if (graph is not None and has_no_side_effects(translator, graph) and (block.exitswitch != c_last_exception or i != len(block.operations)- 1)): del block.operations[i] # look for output variables never used # warning: this must be completely done *before* we attempt to # remove the corresponding variables from block.inputargs! # Otherwise the link.args get out of sync with the # link.target.inputargs. for link in block.exits: assert len(link.args) == len(link.target.inputargs) for i in range(len(link.args)-1, -1, -1): if link.target.inputargs[i] not in read_vars: del link.args[i] # the above assert would fail here for block in blocks: # look for input variables never used # The corresponding link.args have already been all removed above for i in range(len(block.inputargs)-1, -1, -1): if block.inputargs[i] not in read_vars: del block.inputargs[i] def remove_identical_vars(graph): """When the same variable is passed multiple times into the next block, pass it only once. This enables further optimizations by the annotator, which otherwise doesn't realize that tests performed on one of the copies of the variable also affect the other.""" # This algorithm is based on DataFlowFamilyBuilder, used as a # "phi node remover" (in the SSA sense). 'variable_families' is a # UnionFind object that groups variables by families; variables from the # same family can be identified, and if two input arguments of a block # end up in the same family, then we really remove one of them in favor # of the other. # # The idea is to identify as much variables as possible by trying # iteratively two kinds of phi node removal: # # * "vertical", by identifying variables from different blocks, when # we see that a value just flows unmodified into the next block without # needing any merge (this is what backendopt.ssa.SSI_to_SSA() would do # as well); # # * "horizontal", by identifying two input variables of the same block, # when these two variables' phi nodes have the same argument -- i.e. # when for all possible incoming paths they would get twice the same # value (this is really the purpose of remove_identical_vars()). # from pypy.translator.backendopt.ssa import DataFlowFamilyBuilder builder = DataFlowFamilyBuilder(graph) variable_families = builder.get_variable_families() # vertical removal while True: if not builder.merge_identical_phi_nodes(): # horizontal removal break if not builder.complete(): # vertical removal break for block, links in mkentrymap(graph).items(): if block is graph.startblock: continue renaming = {} family2blockvar = {} kills = [] for i, v in enumerate(block.inputargs): v1 = variable_families.find_rep(v) if v1 in family2blockvar: # already seen -- this variable can be shared with the # previous one renaming[v] = family2blockvar[v1] kills.append(i) else: family2blockvar[v1] = v if renaming: block.renamevariables(renaming) # remove the now-duplicate input variables kills.reverse() # starting from the end for i in kills: del block.inputargs[i] for link in links: del link.args[i] def coalesce_is_true(graph): """coalesce paths that go through an is_true and a directly successive is_true both on the same value, transforming the link into the second is_true from the first to directly jump to the correct target out of the second.""" candidates = [] def has_is_true_exitpath(block): tgts = [] start_op = block.operations[-1] cond_v = start_op.args[0] if block.exitswitch == start_op.result: for exit in block.exits: tgt = exit.target if tgt == block: continue rrenaming = dict(zip(tgt.inputargs,exit.args)) if len(tgt.operations) == 1 and tgt.operations[0].opname == 'is_true': tgt_op = tgt.operations[0] if tgt.exitswitch == tgt_op.result and rrenaming.get(tgt_op.args[0]) == cond_v: tgts.append((exit.exitcase, tgt)) return tgts def visit(block): if isinstance(block, Block) and block.operations and block.operations[-1].opname == 'is_true': tgts = has_is_true_exitpath(block) if tgts: candidates.append((block, tgts)) traverse(visit, graph) while candidates: cand, tgts = candidates.pop() newexits = list(cand.exits) for case, tgt in tgts: exit = cand.exits[case] rrenaming = dict(zip(tgt.inputargs,exit.args)) rrenaming[tgt.operations[0].result] = cand.operations[-1].result def rename(v): return rrenaming.get(v,v) newlink = tgt.exits[case].copy(rename) newexits[case] = newlink cand.recloseblock(*newexits) newtgts = has_is_true_exitpath(cand) if newtgts: candidates.append((cand, newtgts)) # ____________________________________________________________ def detect_list_comprehension(graph): """Look for the pattern: Replace it with marker operations: v0 = newlist() v2 = newlist() v1 = hint(v0, iterable, {'maxlength'}) loop start loop start ... ... exactly one append per loop v1.append(..) and nothing else done with v2 ... ... loop end v2 = hint(v1, {'fence'}) """ # NB. this assumes RPythonicity: we can only iterate over something # that has a len(), and this len() cannot change as long as we are # using the iterator. from pypy.translator.backendopt.ssa import DataFlowFamilyBuilder builder = DataFlowFamilyBuilder(graph) variable_families = builder.get_variable_families() c_append = Constant('append') newlist_v = {} iter_v = {} append_v = [] loopnextblocks = [] # collect relevant operations based on the family of their result for block in graph.iterblocks(): if (len(block.operations) == 1 and block.operations[0].opname == 'next' and block.exitswitch == c_last_exception and len(block.exits) >= 2): cases = [link.exitcase for link in block.exits] if None in cases and StopIteration in cases: # it's a straightforward loop start block loopnextblocks.append((block, block.operations[0].args[0])) continue for op in block.operations: if op.opname == 'newlist' and not op.args: vlist = variable_families.find_rep(op.result) newlist_v[vlist] = block if op.opname == 'iter': viter = variable_families.find_rep(op.result) iter_v[viter] = block loops = [] for block, viter in loopnextblocks: viterfamily = variable_families.find_rep(viter) if viterfamily in iter_v: # we have a next(viter) operation where viter comes from a # single known iter() operation. Check that the iter() # operation is in the block just before. iterblock = iter_v[viterfamily] if (len(iterblock.exits) == 1 and iterblock.exitswitch is None and iterblock.exits[0].target is block): # yes - simple case. loops.append((block, iterblock, viterfamily)) if not newlist_v or not loops: return # XXX works with Python >= 2.4 only: find calls to append encoded as # getattr/simple_call pairs, as produced by the LIST_APPEND bytecode. for block in graph.iterblocks(): for i in range(len(block.operations)-1): op = block.operations[i] if op.opname == 'getattr' and op.args[1] == c_append: vlist = variable_families.find_rep(op.args[0]) if vlist in newlist_v: op2 = block.operations[i+1] if (op2.opname == 'simple_call' and len(op2.args) == 2 and op2.args[0] is op.result): append_v.append((op.args[0], op.result, block)) if not append_v: return detector = ListComprehensionDetector(graph, loops, newlist_v, variable_families) graphmutated = False for location in append_v: if graphmutated: # new variables introduced, must restart the whole process return detect_list_comprehension(graph) try: detector.run(*location) except DetectorFailed: pass else: graphmutated = True class DetectorFailed(Exception): pass class ListComprehensionDetector(object): def __init__(self, graph, loops, newlist_v, variable_families): self.graph = graph self.loops = loops self.newlist_v = newlist_v self.variable_families = variable_families self.reachable_cache = {} def enum_blocks_with_vlist_from(self, fromblock, avoid): found = {avoid: True} pending = [fromblock] while pending: block = pending.pop() if block in found: continue if not self.vlist_alive(block): continue yield block found[block] = True for exit in block.exits: pending.append(exit.target) def enum_reachable_blocks(self, fromblock, stop_at, stay_within=None): if fromblock is stop_at: return found = {stop_at: True} pending = [fromblock] while pending: block = pending.pop() if block in found: continue found[block] = True for exit in block.exits: if stay_within is None or exit.target in stay_within: yield exit.target pending.append(exit.target) def reachable_within(self, fromblock, toblock, avoid, stay_within): if toblock is avoid: return False for block in self.enum_reachable_blocks(fromblock, avoid, stay_within): if block is toblock: return True return False def reachable(self, fromblock, toblock, avoid): if toblock is avoid: return False try: return self.reachable_cache[fromblock, toblock, avoid] except KeyError: pass future = [fromblock] for block in self.enum_reachable_blocks(fromblock, avoid): self.reachable_cache[fromblock, block, avoid] = True if block is toblock: return True future.append(block) # 'toblock' is unreachable from 'fromblock', so it is also # unreachable from any of the 'future' blocks for block in future: self.reachable_cache[block, toblock, avoid] = False return False def vlist_alive(self, block): # check if 'block' is in the "cone" of blocks where # the vlistfamily lives try: return self.vlistcone[block] except KeyError: result = bool(self.contains_vlist(block.inputargs)) self.vlistcone[block] = result return result def vlist_escapes(self, block): # check if the vlist "escapes" to uncontrolled places in that block try: return self.escapes[block] except KeyError: for op in block.operations: if op.result is self.vmeth: continue # the single getattr(vlist, 'append') is ok if op.opname == 'getitem': continue # why not allow getitem(vlist, index) too if self.contains_vlist(op.args): result = True break else: result = False self.escapes[block] = result return result def contains_vlist(self, args): for arg in args: if self.variable_families.find_rep(arg) is self.vlistfamily: return arg else: return None def remove_vlist(self, args): removed = 0 for i in range(len(args)-1, -1, -1): arg = self.variable_families.find_rep(args[i]) if arg is self.vlistfamily: del args[i] removed += 1 assert removed == 1 def run(self, vlist, vmeth, appendblock): # first check that the 'append' method object doesn't escape for op in appendblock.operations: if op.opname == 'simple_call' and op.args[0] is vmeth: pass elif vmeth in op.args: raise DetectorFailed # used in another operation for link in appendblock.exits: if vmeth in link.args: raise DetectorFailed # escapes to a next block self.vmeth = vmeth self.vlistfamily = self.variable_families.find_rep(vlist) newlistblock = self.newlist_v[self.vlistfamily] self.vlistcone = {newlistblock: True} self.escapes = {self.graph.returnblock: True, self.graph.exceptblock: True} # in which loop are we? for loopnextblock, iterblock, viterfamily in self.loops: # check that the vlist is alive across the loop head block, # which ensures that we have a whole loop where the vlist # doesn't change if not self.vlist_alive(loopnextblock): continue # no - unrelated loop # check that we cannot go from 'newlist' to 'append' without # going through the 'iter' of our loop (and the following 'next'). # This ensures that the lifetime of vlist is cleanly divided in # "before" and "after" the loop... if self.reachable(newlistblock, appendblock, avoid=iterblock): continue # ... with the possible exception of links from the loop # body jumping back to the loop prologue, between 'newlist' and # 'iter', which we must forbid too: if self.reachable(loopnextblock, iterblock, avoid=newlistblock): continue # there must not be a larger number of calls to 'append' than # the number of elements that 'next' returns, so we must ensure # that we cannot go from 'append' to 'append' again without # passing 'next'... if self.reachable(appendblock, appendblock, avoid=loopnextblock): continue # ... and when the iterator is exhausted, we should no longer # reach 'append' at all. stopblocks = [link.target for link in loopnextblock.exits if link.exitcase is not None] accepted = True for stopblock1 in stopblocks: if self.reachable(stopblock1, appendblock, avoid=newlistblock): accepted = False if not accepted: continue # now explicitly find the "loop body" blocks: they are the ones # from which we can reach 'append' without going through 'iter'. # (XXX inefficient) loopbody = {} for block in self.graph.iterblocks(): if (self.vlist_alive(block) and self.reachable(block, appendblock, iterblock)): loopbody[block] = True # if the 'append' is actually after a 'break' or on a path that # can only end up in a 'break', then it won't be recorded as part # of the loop body at all. This is a strange case where we have # basically proved that the list will be of length 1... too # uncommon to worry about, I suspect if appendblock not in loopbody: continue # This candidate loop is acceptable if the list is not escaping # too early, i.e. in the loop header or in the loop body. loopheader = list(self.enum_blocks_with_vlist_from(newlistblock, avoid=loopnextblock)) assert loopheader[0] is newlistblock escapes = False for block in loopheader + loopbody.keys(): assert self.vlist_alive(block) if self.vlist_escapes(block): escapes = True break if not escapes: break # accept this loop! else: raise DetectorFailed # no suitable loop # Found a suitable loop, let's patch the graph: assert iterblock not in loopbody assert loopnextblock in loopbody for stopblock1 in stopblocks: assert stopblock1 not in loopbody # at StopIteration, the new list is exactly of the same length as # the one we iterate over if it's not possible to skip the appendblock # in the body: exactlength = not self.reachable_within(loopnextblock, loopnextblock, avoid = appendblock, stay_within = loopbody) # - add a hint(vlist, iterable, {'maxlength'}) in the iterblock, # where we can compute the known maximum length link = iterblock.exits[0] vlist = self.contains_vlist(link.args) assert vlist for op in iterblock.operations: res = self.variable_families.find_rep(op.result) if res is viterfamily: break else: raise AssertionError("lost 'iter' operation") vlength = Variable('maxlength') vlist2 = Variable(vlist) chint = Constant({'maxlength': True}) iterblock.operations += [ SpaceOperation('hint', [vlist, op.args[0], chint], vlist2)] link.args = list(link.args) for i in range(len(link.args)): if link.args[i] is vlist: link.args[i] = vlist2 # - wherever the list exits the loop body, add a 'hint({fence})' from pypy.translator.unsimplify import insert_empty_block for block in loopbody: for link in block.exits: if link.target not in loopbody: vlist = self.contains_vlist(link.args) if vlist is None: continue # list not passed along this link anyway hints = {'fence': True} if (exactlength and block is loopnextblock and link.target in stopblocks): hints['exactlength'] = True chints = Constant(hints) newblock = insert_empty_block(None, link) index = link.args.index(vlist) vlist2 = newblock.inputargs[index] vlist3 = Variable(vlist2) newblock.inputargs[index] = vlist3 newblock.operations.append( SpaceOperation('hint', [vlist3, chints], vlist2)) # done! # ____ all passes & simplify_graph all_passes = [ eliminate_empty_blocks, remove_assertion_errors, join_blocks, coalesce_is_true, transform_dead_op_vars, remove_identical_vars, transform_ovfcheck, simplify_exceptions, transform_xxxitem, remove_dead_exceptions, ] def simplify_graph(graph, passes=True): # can take a list of passes to apply, True meaning all """inplace-apply all the existing optimisations to the graph.""" if passes is True: passes = all_passes checkgraph(graph) for pass_ in passes: pass_(graph) checkgraph(graph) def cleanup_graph(graph): checkgraph(graph) eliminate_empty_blocks(graph) join_blocks(graph) remove_identical_vars(graph) checkgraph(graph)