[pypy-commit] pypy jit-short_from_state: hg merge default
hakanardo
noreply at buildbot.pypy.org
Fri Aug 5 14:42:49 CEST 2011
Author: Hakan Ardo <hakan at debian.org>
Branch: jit-short_from_state
Changeset: r46287:314b7370dbf0
Date: 2011-08-05 14:13 +0200
http://bitbucket.org/pypy/pypy/changeset/314b7370dbf0/
Log: hg merge default
diff --git a/.hgtags b/.hgtags
--- a/.hgtags
+++ b/.hgtags
@@ -1,1 +1,2 @@
b590cf6de4190623aad9aa698694c22e614d67b9 release-1.5
+b48df0bf4e75b81d98f19ce89d4a7dc3e1dab5e5 benchmarked
diff --git a/LICENSE b/LICENSE
--- a/LICENSE
+++ b/LICENSE
@@ -37,22 +37,22 @@
Armin Rigo
Maciej Fijalkowski
Carl Friedrich Bolz
+ Antonio Cuni
Amaury Forgeot d'Arc
- Antonio Cuni
Samuele Pedroni
Michael Hudson
Holger Krekel
+ Benjamin Peterson
Christian Tismer
- Benjamin Peterson
+ Hakan Ardo
+ Alex Gaynor
Eric van Riet Paap
- Anders Chrigström
- Håkan Ardö
+ Anders Chrigstrom
+ David Schneider
Richard Emslie
Dan Villiom Podlaski Christiansen
Alexander Schremmer
- Alex Gaynor
- David Schneider
- Aurelién Campeas
+ Aurelien Campeas
Anders Lehmann
Camillo Bruni
Niklaus Haldimann
@@ -63,16 +63,17 @@
Bartosz Skowron
Jakub Gustak
Guido Wesdorp
+ Daniel Roberts
Adrien Di Mascio
Laura Creighton
Ludovic Aubry
Niko Matsakis
- Daniel Roberts
Jason Creighton
- Jacob Hallén
+ Jacob Hallen
Alex Martelli
Anders Hammarquist
Jan de Mooij
+ Wim Lavrijsen
Stephan Diehl
Michael Foord
Stefan Schwarzer
@@ -83,9 +84,13 @@
Alexandre Fayolle
Marius Gedminas
Simon Burton
+ Justin Peel
Jean-Paul Calderone
John Witulski
+ Lukas Diekmann
+ holger krekel
Wim Lavrijsen
+ Dario Bertini
Andreas Stührk
Jean-Philippe St. Pierre
Guido van Rossum
@@ -97,15 +102,16 @@
Georg Brandl
Gerald Klix
Wanja Saatkamp
+ Ronny Pfannschmidt
Boris Feigin
Oscar Nierstrasz
- Dario Bertini
David Malcolm
Eugene Oden
Henry Mason
+ Sven Hager
Lukas Renggli
+ Ilya Osadchiy
Guenter Jantzen
- Ronny Pfannschmidt
Bert Freudenberg
Amit Regmi
Ben Young
@@ -122,8 +128,8 @@
Jared Grubb
Karl Bartel
Gabriel Lavoie
+ Victor Stinner
Brian Dorsey
- Victor Stinner
Stuart Williams
Toby Watson
Antoine Pitrou
@@ -134,19 +140,23 @@
Jonathan David Riehl
Elmo Mäntynen
Anders Qvist
- Beatrice Düring
+ Beatrice During
Alexander Sedov
+ Timo Paulssen
+ Corbin Simpson
Vincent Legoll
+ Romain Guillebert
Alan McIntyre
- Romain Guillebert
Alex Perry
Jens-Uwe Mager
+ Simon Cross
Dan Stromberg
- Lukas Diekmann
+ Guillebert Romain
Carl Meyer
Pieter Zieschang
Alejandro J. Cura
Sylvain Thenault
+ Christoph Gerum
Travis Francis Athougies
Henrik Vendelbo
Lutz Paelike
@@ -157,6 +167,7 @@
Miguel de Val Borro
Ignas Mikalajunas
Artur Lisiecki
+ Philip Jenvey
Joshua Gilbert
Godefroid Chappelle
Yusei Tahara
@@ -165,27 +176,31 @@
Gustavo Niemeyer
William Leslie
Akira Li
- Kristján Valur Jónsson
+ Kristjan Valur Jonsson
Bobby Impollonia
+ Michael Hudson-Doyle
Andrew Thompson
Anders Sigfridsson
+ Floris Bruynooghe
Jacek Generowicz
Dan Colish
- Sven Hager
Zooko Wilcox-O Hearn
+ Dan Villiom Podlaski Christiansen
Anders Hammarquist
+ Chris Lambacher
Dinu Gherman
Dan Colish
+ Brett Cannon
Daniel Neuhäuser
Michael Chermside
Konrad Delong
Anna Ravencroft
Greg Price
Armin Ronacher
+ Christian Muirhead
Jim Baker
- Philip Jenvey
Rodrigo Araújo
- Brett Cannon
+ Romain Guillebert
Heinrich-Heine University, Germany
Open End AB (formerly AB Strakt), Sweden
diff --git a/lib-python/conftest.py b/lib-python/conftest.py
--- a/lib-python/conftest.py
+++ b/lib-python/conftest.py
@@ -154,18 +154,18 @@
RegrTest('test_cmd.py'),
RegrTest('test_cmd_line_script.py'),
RegrTest('test_codeccallbacks.py', core=True),
- RegrTest('test_codecencodings_cn.py'),
- RegrTest('test_codecencodings_hk.py'),
- RegrTest('test_codecencodings_jp.py'),
- RegrTest('test_codecencodings_kr.py'),
- RegrTest('test_codecencodings_tw.py'),
+ RegrTest('test_codecencodings_cn.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecencodings_hk.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecencodings_jp.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecencodings_kr.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecencodings_tw.py', usemodules='_multibytecodec'),
- RegrTest('test_codecmaps_cn.py'),
- RegrTest('test_codecmaps_hk.py'),
- RegrTest('test_codecmaps_jp.py'),
- RegrTest('test_codecmaps_kr.py'),
- RegrTest('test_codecmaps_tw.py'),
- RegrTest('test_codecs.py', core=True),
+ RegrTest('test_codecmaps_cn.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecmaps_hk.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecmaps_jp.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecmaps_kr.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecmaps_tw.py', usemodules='_multibytecodec'),
+ RegrTest('test_codecs.py', core=True, usemodules='_multibytecodec'),
RegrTest('test_codeop.py', core=True),
RegrTest('test_coercion.py', core=True),
RegrTest('test_collections.py'),
@@ -314,7 +314,7 @@
RegrTest('test_mmap.py'),
RegrTest('test_module.py', core=True),
RegrTest('test_modulefinder.py'),
- RegrTest('test_multibytecodec.py'),
+ RegrTest('test_multibytecodec.py', usemodules='_multibytecodec'),
RegrTest('test_multibytecodec_support.py', skip="not a test"),
RegrTest('test_multifile.py'),
RegrTest('test_multiprocessing.py', skip='FIXME leaves subprocesses'),
diff --git a/lib-python/modified-2.7/ctypes/__init__.py b/lib-python/modified-2.7/ctypes/__init__.py
--- a/lib-python/modified-2.7/ctypes/__init__.py
+++ b/lib-python/modified-2.7/ctypes/__init__.py
@@ -489,9 +489,12 @@
_flags_ = _FUNCFLAG_CDECL | _FUNCFLAG_PYTHONAPI
return CFunctionType
-_cast = PYFUNCTYPE(py_object, c_void_p, py_object, py_object)(_cast_addr)
def cast(obj, typ):
- return _cast(obj, obj, typ)
+ try:
+ c_void_p.from_param(obj)
+ except TypeError, e:
+ raise ArgumentError(str(e))
+ return _cast_addr(obj, obj, typ)
_string_at = PYFUNCTYPE(py_object, c_void_p, c_int)(_string_at_addr)
def string_at(ptr, size=-1):
diff --git a/lib-python/modified-2.7/test/test_multibytecodec.py b/lib-python/modified-2.7/test/test_multibytecodec.py
--- a/lib-python/modified-2.7/test/test_multibytecodec.py
+++ b/lib-python/modified-2.7/test/test_multibytecodec.py
@@ -148,7 +148,8 @@
class Test_StreamReader(unittest.TestCase):
def test_bug1728403(self):
try:
- open(TESTFN, 'w').write('\xa1')
+ with open(TESTFN, 'w') as f:
+ f.write('\xa1')
f = codecs.open(TESTFN, encoding='cp949')
self.assertRaises(UnicodeDecodeError, f.read, 2)
finally:
diff --git a/lib_pypy/_ctypes/function.py b/lib_pypy/_ctypes/function.py
--- a/lib_pypy/_ctypes/function.py
+++ b/lib_pypy/_ctypes/function.py
@@ -78,8 +78,6 @@
_com_iid = None
_is_fastpath = False
- __restype_set = False
-
def _getargtypes(self):
return self._argtypes_
@@ -93,13 +91,15 @@
raise TypeError(
"item %d in _argtypes_ has no from_param method" % (
i + 1,))
- #
- if all([hasattr(argtype, '_ffiargshape') for argtype in argtypes]):
- fastpath_cls = make_fastpath_subclass(self.__class__)
- fastpath_cls.enable_fastpath_maybe(self)
self._argtypes_ = list(argtypes)
+ self._check_argtypes_for_fastpath()
argtypes = property(_getargtypes, _setargtypes)
+ def _check_argtypes_for_fastpath(self):
+ if all([hasattr(argtype, '_ffiargshape') for argtype in self._argtypes_]):
+ fastpath_cls = make_fastpath_subclass(self.__class__)
+ fastpath_cls.enable_fastpath_maybe(self)
+
def _getparamflags(self):
return self._paramflags
@@ -149,7 +149,6 @@
return self._restype_
def _setrestype(self, restype):
- self.__restype_set = True
self._ptr = None
if restype is int:
from ctypes import c_int
@@ -219,6 +218,7 @@
import ctypes
restype = ctypes.c_int
self._ptr = self._getfuncptr_fromaddress(self._argtypes_, restype)
+ self._check_argtypes_for_fastpath()
return
@@ -296,13 +296,12 @@
"This function takes %d argument%s (%s given)"
% (len(self._argtypes_), plural, len(args)))
- # check that arguments are convertible
- ## XXX Not as long as ctypes.cast is a callback function with
- ## py_object arguments...
- ## self._convert_args(self._argtypes_, args, {})
-
try:
- res = self.callable(*args)
+ newargs = self._convert_args_for_callback(argtypes, args)
+ except (UnicodeError, TypeError, ValueError), e:
+ raise ArgumentError(str(e))
+ try:
+ res = self.callable(*newargs)
except:
exc_info = sys.exc_info()
traceback.print_tb(exc_info[2], file=sys.stderr)
@@ -316,10 +315,6 @@
warnings.warn('C function without declared arguments called',
RuntimeWarning, stacklevel=2)
argtypes = []
-
- if not self.__restype_set:
- warnings.warn('C function without declared return type called',
- RuntimeWarning, stacklevel=2)
if self._com_index:
from ctypes import cast, c_void_p, POINTER
@@ -366,7 +361,10 @@
if self._flags_ & _rawffi.FUNCFLAG_USE_LASTERROR:
set_last_error(_rawffi.get_last_error())
#
- return self._build_result(self._restype_, result, newargs)
+ try:
+ return self._build_result(self._restype_, result, newargs)
+ finally:
+ funcptr.free_temp_buffers()
def _do_errcheck(self, result, args):
# The 'errcheck' protocol
@@ -466,6 +464,18 @@
return cobj, cobj._to_ffi_param(), type(cobj)
+ def _convert_args_for_callback(self, argtypes, args):
+ assert len(argtypes) == len(args)
+ newargs = []
+ for argtype, arg in zip(argtypes, args):
+ param = argtype.from_param(arg)
+ if argtype._type_ == 'P': # special-case for c_void_p
+ param = param._get_buffer_value()
+ elif self._is_primitive(argtype):
+ param = param.value
+ newargs.append(param)
+ return newargs
+
def _convert_args(self, argtypes, args, kwargs, marker=object()):
newargs = []
outargs = []
@@ -556,6 +566,9 @@
newargtypes.append(newargtype)
return keepalives, newargs, newargtypes, outargs
+ @staticmethod
+ def _is_primitive(argtype):
+ return argtype.__bases__[0] is _SimpleCData
def _wrap_result(self, restype, result):
"""
@@ -564,7 +577,7 @@
"""
# hack for performance: if restype is a "simple" primitive type, don't
# allocate the buffer because it's going to be thrown away immediately
- if restype.__bases__[0] is _SimpleCData and not restype._is_pointer_like():
+ if self._is_primitive(restype) and not restype._is_pointer_like():
return result
#
shape = restype._ffishape
@@ -680,7 +693,7 @@
try:
result = self._call_funcptr(funcptr, *args)
result = self._do_errcheck(result, args)
- except (TypeError, ArgumentError): # XXX, should be FFITypeError
+ except (TypeError, ArgumentError, UnicodeDecodeError):
assert self._slowpath_allowed
return CFuncPtr.__call__(self, *args)
return result
diff --git a/lib_pypy/_ctypes/primitive.py b/lib_pypy/_ctypes/primitive.py
--- a/lib_pypy/_ctypes/primitive.py
+++ b/lib_pypy/_ctypes/primitive.py
@@ -10,6 +10,8 @@
from _ctypes.builtin import ConvMode
from _ctypes.array import Array
from _ctypes.pointer import _Pointer, as_ffi_pointer
+#from _ctypes.function import CFuncPtr # this import is moved at the bottom
+ # because else it's circular
class NULL(object):
pass
@@ -86,7 +88,7 @@
return res
if isinstance(value, Array):
return value
- if isinstance(value, _Pointer):
+ if isinstance(value, (_Pointer, CFuncPtr)):
return cls.from_address(value._buffer.buffer)
if isinstance(value, (int, long)):
return cls(value)
@@ -338,3 +340,5 @@
def __nonzero__(self):
return self._buffer[0] not in (0, '\x00')
+
+from _ctypes.function import CFuncPtr
diff --git a/lib_pypy/_ctypes/structure.py b/lib_pypy/_ctypes/structure.py
--- a/lib_pypy/_ctypes/structure.py
+++ b/lib_pypy/_ctypes/structure.py
@@ -34,16 +34,18 @@
for i, field in enumerate(all_fields):
name = field[0]
value = field[1]
+ is_bitfield = (len(field) == 3)
fields[name] = Field(name,
self._ffistruct.fieldoffset(name),
self._ffistruct.fieldsize(name),
- value, i)
+ value, i, is_bitfield)
if anonymous_fields:
resnames = []
for i, field in enumerate(all_fields):
name = field[0]
value = field[1]
+ is_bitfield = (len(field) == 3)
startpos = self._ffistruct.fieldoffset(name)
if name in anonymous_fields:
for subname in value._names:
@@ -52,7 +54,7 @@
subvalue = value._fieldtypes[subname].ctype
fields[subname] = Field(subname,
relpos, subvalue._sizeofinstances(),
- subvalue, i)
+ subvalue, i, is_bitfield)
else:
resnames.append(name)
names = resnames
@@ -60,8 +62,8 @@
self._fieldtypes = fields
class Field(object):
- def __init__(self, name, offset, size, ctype, num):
- for k in ('name', 'offset', 'size', 'ctype', 'num'):
+ def __init__(self, name, offset, size, ctype, num, is_bitfield):
+ for k in ('name', 'offset', 'size', 'ctype', 'num', 'is_bitfield'):
self.__dict__[k] = locals()[k]
def __setattr__(self, name, value):
@@ -225,7 +227,7 @@
field = self._fieldtypes[name]
except KeyError:
return _CData.__getattribute__(self, name)
- if field.size >> 16:
+ if field.is_bitfield:
# bitfield member, use direct access
return self._buffer.__getattr__(name)
else:
diff --git a/pypy/annotation/builtin.py b/pypy/annotation/builtin.py
--- a/pypy/annotation/builtin.py
+++ b/pypy/annotation/builtin.py
@@ -416,7 +416,8 @@
from pypy.annotation.model import SomePtr
from pypy.rpython.lltypesystem import lltype
-def malloc(s_T, s_n=None, s_flavor=None, s_zero=None, s_track_allocation=None):
+def malloc(s_T, s_n=None, s_flavor=None, s_zero=None, s_track_allocation=None,
+ s_add_memory_pressure=None):
assert (s_n is None or s_n.knowntype == int
or issubclass(s_n.knowntype, pypy.rlib.rarithmetic.base_int))
assert s_T.is_constant()
@@ -432,6 +433,8 @@
else:
assert s_flavor.is_constant()
assert s_track_allocation is None or s_track_allocation.is_constant()
+ assert (s_add_memory_pressure is None or
+ s_add_memory_pressure.is_constant())
# not sure how to call malloc() for the example 'p' in the
# presence of s_extraargs
r = SomePtr(lltype.Ptr(s_T.const))
diff --git a/pypy/config/translationoption.py b/pypy/config/translationoption.py
--- a/pypy/config/translationoption.py
+++ b/pypy/config/translationoption.py
@@ -13,6 +13,10 @@
DEFL_LOW_INLINE_THRESHOLD = DEFL_INLINE_THRESHOLD / 2.0
DEFL_GC = "minimark"
+if sys.platform.startswith("linux"):
+ DEFL_ROOTFINDER_WITHJIT = "asmgcc"
+else:
+ DEFL_ROOTFINDER_WITHJIT = "shadowstack"
IS_64_BITS = sys.maxint > 2147483647
@@ -109,7 +113,7 @@
BoolOption("jit", "generate a JIT",
default=False,
suggests=[("translation.gc", DEFL_GC),
- ("translation.gcrootfinder", "asmgcc"),
+ ("translation.gcrootfinder", DEFL_ROOTFINDER_WITHJIT),
("translation.list_comprehension_operations", True)]),
ChoiceOption("jit_backend", "choose the backend for the JIT",
["auto", "x86", "x86-without-sse2", "llvm"],
diff --git a/pypy/doc/contributor.rst b/pypy/doc/contributor.rst
--- a/pypy/doc/contributor.rst
+++ b/pypy/doc/contributor.rst
@@ -9,22 +9,22 @@
Armin Rigo
Maciej Fijalkowski
Carl Friedrich Bolz
+ Antonio Cuni
Amaury Forgeot d'Arc
- Antonio Cuni
Samuele Pedroni
Michael Hudson
Holger Krekel
+ Benjamin Peterson
Christian Tismer
- Benjamin Peterson
+ Hakan Ardo
+ Alex Gaynor
Eric van Riet Paap
- Anders Chrigström
- Håkan Ardö
+ Anders Chrigstrom
+ David Schneider
Richard Emslie
Dan Villiom Podlaski Christiansen
Alexander Schremmer
- Alex Gaynor
- David Schneider
- Aurelién Campeas
+ Aurelien Campeas
Anders Lehmann
Camillo Bruni
Niklaus Haldimann
@@ -35,16 +35,17 @@
Bartosz Skowron
Jakub Gustak
Guido Wesdorp
+ Daniel Roberts
Adrien Di Mascio
Laura Creighton
Ludovic Aubry
Niko Matsakis
- Daniel Roberts
Jason Creighton
- Jacob Hallén
+ Jacob Hallen
Alex Martelli
Anders Hammarquist
Jan de Mooij
+ Wim Lavrijsen
Stephan Diehl
Michael Foord
Stefan Schwarzer
@@ -55,9 +56,13 @@
Alexandre Fayolle
Marius Gedminas
Simon Burton
+ Justin Peel
Jean-Paul Calderone
John Witulski
+ Lukas Diekmann
+ holger krekel
Wim Lavrijsen
+ Dario Bertini
Andreas Stührk
Jean-Philippe St. Pierre
Guido van Rossum
@@ -69,15 +74,16 @@
Georg Brandl
Gerald Klix
Wanja Saatkamp
+ Ronny Pfannschmidt
Boris Feigin
Oscar Nierstrasz
- Dario Bertini
David Malcolm
Eugene Oden
Henry Mason
+ Sven Hager
Lukas Renggli
+ Ilya Osadchiy
Guenter Jantzen
- Ronny Pfannschmidt
Bert Freudenberg
Amit Regmi
Ben Young
@@ -94,8 +100,8 @@
Jared Grubb
Karl Bartel
Gabriel Lavoie
+ Victor Stinner
Brian Dorsey
- Victor Stinner
Stuart Williams
Toby Watson
Antoine Pitrou
@@ -106,19 +112,23 @@
Jonathan David Riehl
Elmo Mäntynen
Anders Qvist
- Beatrice Düring
+ Beatrice During
Alexander Sedov
+ Timo Paulssen
+ Corbin Simpson
Vincent Legoll
+ Romain Guillebert
Alan McIntyre
- Romain Guillebert
Alex Perry
Jens-Uwe Mager
+ Simon Cross
Dan Stromberg
- Lukas Diekmann
+ Guillebert Romain
Carl Meyer
Pieter Zieschang
Alejandro J. Cura
Sylvain Thenault
+ Christoph Gerum
Travis Francis Athougies
Henrik Vendelbo
Lutz Paelike
@@ -129,6 +139,7 @@
Miguel de Val Borro
Ignas Mikalajunas
Artur Lisiecki
+ Philip Jenvey
Joshua Gilbert
Godefroid Chappelle
Yusei Tahara
@@ -137,24 +148,29 @@
Gustavo Niemeyer
William Leslie
Akira Li
- Kristján Valur Jónsson
+ Kristjan Valur Jonsson
Bobby Impollonia
+ Michael Hudson-Doyle
Andrew Thompson
Anders Sigfridsson
+ Floris Bruynooghe
Jacek Generowicz
Dan Colish
- Sven Hager
Zooko Wilcox-O Hearn
+ Dan Villiom Podlaski Christiansen
Anders Hammarquist
+ Chris Lambacher
Dinu Gherman
Dan Colish
+ Brett Cannon
Daniel Neuhäuser
Michael Chermside
Konrad Delong
Anna Ravencroft
Greg Price
Armin Ronacher
+ Christian Muirhead
Jim Baker
- Philip Jenvey
Rodrigo Araújo
+ Romain Guillebert
diff --git a/pypy/doc/how-to-release.rst b/pypy/doc/how-to-release.rst
--- a/pypy/doc/how-to-release.rst
+++ b/pypy/doc/how-to-release.rst
@@ -21,8 +21,8 @@
Release Steps
----------------
-* at code freeze make a release branch under
- http://codepeak.net/svn/pypy/release/x.y(.z). IMPORTANT: bump the
+* at code freeze make a release branch using release-x.x.x in mercurial
+ IMPORTANT: bump the
pypy version number in module/sys/version.py and in
module/cpyext/include/patchlevel.h, notice that the branch
will capture the revision number of this change for the release;
@@ -48,12 +48,6 @@
the release announcement should contain a direct link to the download page
* update pypy.org (under extradoc/pypy.org), rebuild and commit
-* update http://codespeak.net/pypy/trunk:
- code0> + chmod -R yourname:users /www/codespeak.net/htdocs/pypy/trunk
- local> cd ..../pypy/doc && py.test
- local> cd ..../pypy
- local> rsync -az doc codespeak.net:/www/codespeak.net/htdocs/pypy/trunk/pypy/
-
* post announcement on morepypy.blogspot.com
* send announcements to pypy-dev, python-list,
python-announce, python-dev ...
diff --git a/pypy/doc/release-1.6.0.rst b/pypy/doc/release-1.6.0.rst
new file mode 100644
--- /dev/null
+++ b/pypy/doc/release-1.6.0.rst
@@ -0,0 +1,85 @@
+===========================
+PyPy 1.6 - faster than ever
+===========================
+
+We're pleased to announce the 1.6 release of PyPy. This release brings a lot
+of bugfixes and performance improvements over 1.5, and improves support for
+Windows 32bit and OS X 64bit. This version fully implements Python 2.7.1 and
+has beta level support for loading CPython C extensions. You can download it
+here:
+
+ http://pypy.org/download.html
+
+What is PyPy?
+=============
+
+PyPy is a very compliant Python interpreter, almost a drop-in replacement for
+CPython 2.7.1. It's fast (`pypy 1.5 and cpython 2.6.2`_ performance comparison)
+due to its integrated tracing JIT compiler. XXX: compare to 2.7.1
+
+This release supports x86 machines running Linux 32/64 or Mac OS X. Windows 32
+is beta (it roughly works but a lot of small issues have not been fixed so
+far). Windows 64 is not yet supported.
+
+The main topics of this release are speed and stability: on average, PyPy 1.6
+is between 20% and 30% faster than PyPy 1.5, and overall it's 4.3 times faster
+than CPython when running our set of benchmarks.
+
+The speed improvements have been made possible by optimizing many of the
+layers which compose PyPy. In particular, we improved: the Garbage Collector,
+the JIT warmup time, the optimizations performed by the JIT, the quality of
+the generated machine code and the implementation of our Python interpreter.
+
+
+Highlights
+==========
+
+* Numerous performance improvements, overall giving considerable speedups:
+
+ - better GC behavior when dealing with very large objects and arrays
+
+ - `fast ctypes`_: now calls to ctypes functions are seen and optimized
+ by the JIT, and they are up to 60 times faster than PyPy 1.5 and 10 times
+ faster than CPython
+
+ - improved generators(1): simple generators now are inlined into the caller
+ loop, making performance up to 3.5 times faster than PyPy 1.5.
+
+ - improved generators(2): thanks to other optimizations, even generators
+ that are not inlined are between 10% and 20% faster than PyPy 1.5.
+
+ - faster warmup time for the JIT
+
+ - JIT support for single floats (e.g., for ``array('f')``)
+
+ - optimized dictionaries: the internal representation of dictionaries is now
+ dynamically selected depending on the type of stored objects, resulting in
+ faster code and smaller memory footprint. For example, dictionaries whose
+ keys are all strings, or all integers.
+
+* JitViewer: this is the first official release which includes the JitViewer,
+ a web-based tool which helps you to see which parts of your Python code have
+ been compiled by the JIT, down until the assembler. XXX: publish a public
+ demo?
+
+- The CPython extension module API has been improved and now supports many
+ more extensions. For information on which one are supported, please refer to
+ our `compatibility wiki`_.
+
+* Multibyte encoding support: this was of of the last areas in which we were
+ still behind CPython, but now we fully support them. (XXX: is that true?)
+
+* Preliminary support for NumPy: this release includes a preview of a very
+ fast NumPy module integrated with the PyPy JIT. Unfortunately, this does
+ not mean that you can expect to take an existing NumPy program and run it on
+ PyPy, because the module is still unfinished and supports only some of the
+ numpy API. However, what works is blazingly fast :-)
+
+* Bugfixes: since the 1.5 release we fixed 53 bugs in our `bug tracker`_, not
+ counting the numerous bugs that were found and reported through other
+ channels than the bug tracker.
+
+Cheers,
+
+Carl Friedrich Bolz, Laura Creighton, Antonio Cuni, Maciej Fijalkowski,
+Amaury Forgeot d'Arc, Alex Gaynor, Armin Rigo and the PyPy team
diff --git a/pypy/interpreter/baseobjspace.py b/pypy/interpreter/baseobjspace.py
--- a/pypy/interpreter/baseobjspace.py
+++ b/pypy/interpreter/baseobjspace.py
@@ -111,6 +111,9 @@
def setslotvalue(self, index, w_val):
raise NotImplementedError
+ def delslotvalue(self, index):
+ raise NotImplementedError
+
def descr_call_mismatch(self, space, opname, RequiredClass, args):
if RequiredClass is None:
classname = '?'
diff --git a/pypy/interpreter/gateway.py b/pypy/interpreter/gateway.py
--- a/pypy/interpreter/gateway.py
+++ b/pypy/interpreter/gateway.py
@@ -64,7 +64,7 @@
self.visit_self(el[1], *args)
else:
self.visit_function(el, *args)
- else:
+ elif isinstance(el, type):
for typ in self.bases_order:
if issubclass(el, typ):
visit = getattr(self, "visit__%s" % (typ.__name__,))
@@ -73,6 +73,8 @@
else:
raise Exception("%s: no match for unwrap_spec element %s" % (
self.__class__.__name__, el))
+ else:
+ raise Exception("unable to dispatch, %s, perhaps your parameter should have started with w_?" % el)
def apply_over(self, unwrap_spec, *extra):
dispatch = self.dispatch
diff --git a/pypy/interpreter/typedef.py b/pypy/interpreter/typedef.py
--- a/pypy/interpreter/typedef.py
+++ b/pypy/interpreter/typedef.py
@@ -258,6 +258,11 @@
self.slots_w = [None] * nslots
def setslotvalue(self, index, w_value):
self.slots_w[index] = w_value
+ def delslotvalue(self, index):
+ if self.slots_w[index] is None:
+ return False
+ self.slots_w[index] = None
+ return True
def getslotvalue(self, index):
return self.slots_w[index]
add(Proto)
@@ -530,11 +535,10 @@
"""member.__delete__(obj)
Delete the value of the slot 'member' from the given 'obj'."""
self.typecheck(space, w_obj)
- w_oldresult = w_obj.getslotvalue(self.index)
- if w_oldresult is None:
+ success = w_obj.delslotvalue(self.index)
+ if not success:
raise OperationError(space.w_AttributeError,
space.wrap(self.name)) # XXX better message
- w_obj.setslotvalue(self.index, None)
Member.typedef = TypeDef(
"member_descriptor",
diff --git a/pypy/jit/backend/llgraph/llimpl.py b/pypy/jit/backend/llgraph/llimpl.py
--- a/pypy/jit/backend/llgraph/llimpl.py
+++ b/pypy/jit/backend/llgraph/llimpl.py
@@ -1071,6 +1071,8 @@
return heaptracker.adr2int(llmemory.cast_ptr_to_adr(x))
if TP == llmemory.Address:
return heaptracker.adr2int(x)
+ if TP is lltype.SingleFloat:
+ return longlong.singlefloat2int(x)
return lltype.cast_primitive(lltype.Signed, x)
def cast_from_int(TYPE, x):
@@ -1086,6 +1088,9 @@
x = llmemory.cast_int_to_adr(x)
assert lltype.typeOf(x) == llmemory.Address
return x
+ elif TYPE is lltype.SingleFloat:
+ assert lltype.typeOf(x) is lltype.Signed
+ return longlong.int2singlefloat(x)
else:
if lltype.typeOf(x) == llmemory.Address:
x = heaptracker.adr2int(x)
@@ -1140,6 +1145,7 @@
del _future_values[:]
def set_future_value_int(index, value):
+ assert lltype.typeOf(value) is lltype.Signed
set_future_value_ref(index, value)
def set_future_value_float(index, value):
@@ -1488,6 +1494,7 @@
'i': lltype.Signed,
'f': lltype.Float,
'L': lltype.SignedLongLong,
+ 'S': lltype.SingleFloat,
'v': lltype.Void,
}
diff --git a/pypy/jit/backend/llgraph/runner.py b/pypy/jit/backend/llgraph/runner.py
--- a/pypy/jit/backend/llgraph/runner.py
+++ b/pypy/jit/backend/llgraph/runner.py
@@ -91,6 +91,7 @@
class BaseCPU(model.AbstractCPU):
supports_floats = True
supports_longlong = llimpl.IS_32_BIT
+ supports_singlefloats = True
def __init__(self, rtyper, stats=None, opts=None,
translate_support_code=False,
@@ -327,12 +328,16 @@
def calldescrof_dynamic(self, ffi_args, ffi_result, extrainfo=None):
from pypy.jit.backend.llsupport.ffisupport import get_ffi_type_kind
+ from pypy.jit.backend.llsupport.ffisupport import UnsupportedKind
arg_types = []
- for arg in ffi_args:
- kind = get_ffi_type_kind(arg)
- if kind != history.VOID:
- arg_types.append(kind)
- reskind = get_ffi_type_kind(ffi_result)
+ try:
+ for arg in ffi_args:
+ kind = get_ffi_type_kind(self, arg)
+ if kind != history.VOID:
+ arg_types.append(kind)
+ reskind = get_ffi_type_kind(self, ffi_result)
+ except UnsupportedKind:
+ return None
return self.getdescr(0, reskind, extrainfo=extrainfo,
arg_types=''.join(arg_types))
diff --git a/pypy/jit/backend/llgraph/test/test_llgraph.py b/pypy/jit/backend/llgraph/test/test_llgraph.py
--- a/pypy/jit/backend/llgraph/test/test_llgraph.py
+++ b/pypy/jit/backend/llgraph/test/test_llgraph.py
@@ -19,6 +19,9 @@
def setup_method(self, _):
self.cpu = self.cpu_type(None)
+ def test_memoryerror(self):
+ py.test.skip("does not make much sense on the llgraph backend")
+
def test_cast_adr_to_int_and_back():
X = lltype.Struct('X', ('foo', lltype.Signed))
diff --git a/pypy/jit/backend/llsupport/descr.py b/pypy/jit/backend/llsupport/descr.py
--- a/pypy/jit/backend/llsupport/descr.py
+++ b/pypy/jit/backend/llsupport/descr.py
@@ -303,6 +303,8 @@
c = 'f'
elif c == 'f' and longlong.supports_longlong:
return 'longlong.getrealfloat(%s)' % (process('L'),)
+ elif c == 'S':
+ return 'longlong.int2singlefloat(%s)' % (process('i'),)
arg = 'args_%s[%d]' % (c, seen[c])
seen[c] += 1
return arg
@@ -318,6 +320,8 @@
return lltype.Void
elif arg == 'L':
return lltype.SignedLongLong
+ elif arg == 'S':
+ return lltype.SingleFloat
else:
raise AssertionError(arg)
@@ -334,6 +338,8 @@
result = 'rffi.cast(lltype.SignedLongLong, res)'
elif self.get_return_type() == history.VOID:
result = 'None'
+ elif self.get_return_type() == 'S':
+ result = 'longlong.singlefloat2int(res)'
else:
assert 0
source = py.code.Source("""
@@ -344,14 +350,15 @@
""" % locals())
ARGS = [TYPE(arg) for arg in self.arg_classes]
FUNC = lltype.FuncType(ARGS, RESULT)
- d = locals().copy()
- d.update(globals())
+ d = globals().copy()
+ d.update(locals())
exec source.compile() in d
self.call_stub = d['call_stub']
def verify_types(self, args_i, args_r, args_f, return_type):
assert self._return_type in return_type
- assert self.arg_classes.count('i') == len(args_i or ())
+ assert (self.arg_classes.count('i') +
+ self.arg_classes.count('S')) == len(args_i or ())
assert self.arg_classes.count('r') == len(args_r or ())
assert (self.arg_classes.count('f') +
self.arg_classes.count('L')) == len(args_f or ())
@@ -428,23 +435,39 @@
def get_result_size(self, translate_support_code):
return 0
+_SingleFloatCallDescr = None # built lazily
+
def getCallDescrClass(RESULT):
if RESULT is lltype.Void:
return VoidCallDescr
if RESULT is lltype.Float:
return FloatCallDescr
+ if RESULT is lltype.SingleFloat:
+ global _SingleFloatCallDescr
+ if _SingleFloatCallDescr is None:
+ assert rffi.sizeof(rffi.UINT) == rffi.sizeof(RESULT)
+ class SingleFloatCallDescr(getCallDescrClass(rffi.UINT)):
+ _clsname = 'SingleFloatCallDescr'
+ _return_type = 'S'
+ _SingleFloatCallDescr = SingleFloatCallDescr
+ return _SingleFloatCallDescr
if is_longlong(RESULT):
return LongLongCallDescr
return getDescrClass(RESULT, BaseIntCallDescr, GcPtrCallDescr,
NonGcPtrCallDescr, 'Call', 'get_result_size',
Ellipsis, # <= floatattrname should not be used here
'_is_result_signed')
+getCallDescrClass._annspecialcase_ = 'specialize:memo'
def get_call_descr(gccache, ARGS, RESULT, extrainfo=None):
arg_classes = []
for ARG in ARGS:
kind = getkind(ARG)
- if kind == 'int': arg_classes.append('i')
+ if kind == 'int':
+ if ARG is lltype.SingleFloat:
+ arg_classes.append('S')
+ else:
+ arg_classes.append('i')
elif kind == 'ref': arg_classes.append('r')
elif kind == 'float':
if is_longlong(ARG):
@@ -476,6 +499,9 @@
return GcPtrDescr
else:
return NonGcPtrDescr
+ if TYPE is lltype.SingleFloat:
+ assert rffi.sizeof(rffi.UINT) == rffi.sizeof(TYPE)
+ TYPE = rffi.UINT
try:
return _cache[nameprefix, TYPE]
except KeyError:
diff --git a/pypy/jit/backend/llsupport/ffisupport.py b/pypy/jit/backend/llsupport/ffisupport.py
--- a/pypy/jit/backend/llsupport/ffisupport.py
+++ b/pypy/jit/backend/llsupport/ffisupport.py
@@ -1,19 +1,21 @@
from pypy.rlib.rarithmetic import intmask
from pypy.jit.metainterp import history
-from pypy.jit.backend.llsupport.descr import DynamicIntCallDescr, NonGcPtrCallDescr,\
- FloatCallDescr, VoidCallDescr
+from pypy.rpython.lltypesystem import rffi
+from pypy.jit.backend.llsupport.descr import (
+ DynamicIntCallDescr, NonGcPtrCallDescr, FloatCallDescr, VoidCallDescr,
+ LongLongCallDescr, getCallDescrClass)
class UnsupportedKind(Exception):
pass
-def get_call_descr_dynamic(ffi_args, ffi_result, extrainfo=None):
+def get_call_descr_dynamic(cpu, ffi_args, ffi_result, extrainfo=None):
"""Get a call descr: the types of result and args are represented by
rlib.libffi.types.*"""
try:
- reskind = get_ffi_type_kind(ffi_result)
- argkinds = [get_ffi_type_kind(arg) for arg in ffi_args]
+ reskind = get_ffi_type_kind(cpu, ffi_result)
+ argkinds = [get_ffi_type_kind(cpu, arg) for arg in ffi_args]
except UnsupportedKind:
- return None # ??
+ return None
arg_classes = ''.join(argkinds)
if reskind == history.INT:
size = intmask(ffi_result.c_size)
@@ -25,17 +27,26 @@
return FloatCallDescr(arg_classes, extrainfo)
elif reskind == history.VOID:
return VoidCallDescr(arg_classes, extrainfo)
+ elif reskind == 'L':
+ return LongLongCallDescr(arg_classes, extrainfo)
+ elif reskind == 'S':
+ SingleFloatCallDescr = getCallDescrClass(rffi.FLOAT)
+ return SingleFloatCallDescr(arg_classes, extrainfo)
assert False
-def get_ffi_type_kind(ffi_type):
+def get_ffi_type_kind(cpu, ffi_type):
from pypy.rlib.libffi import types
kind = types.getkind(ffi_type)
if kind == 'i' or kind == 'u':
return history.INT
- elif kind == 'f':
+ elif cpu.supports_floats and kind == 'f':
return history.FLOAT
elif kind == 'v':
return history.VOID
+ elif cpu.supports_longlong and (kind == 'I' or kind == 'U'): # longlong
+ return 'L'
+ elif cpu.supports_singlefloats and kind == 's': # singlefloat
+ return 'S'
raise UnsupportedKind("Unsupported kind '%s'" % kind)
def is_ffi_type_signed(ffi_type):
diff --git a/pypy/jit/backend/llsupport/gc.py b/pypy/jit/backend/llsupport/gc.py
--- a/pypy/jit/backend/llsupport/gc.py
+++ b/pypy/jit/backend/llsupport/gc.py
@@ -544,18 +544,19 @@
assert self.GCClass.inline_simple_malloc
assert self.GCClass.inline_simple_malloc_varsize
- # make a malloc function, with three arguments
+ # make a malloc function, with two arguments
def malloc_basic(size, tid):
type_id = llop.extract_ushort(llgroup.HALFWORD, tid)
has_finalizer = bool(tid & (1<<llgroup.HALFSHIFT))
check_typeid(type_id)
- try:
- res = llop1.do_malloc_fixedsize_clear(llmemory.GCREF,
- type_id, size, True,
- has_finalizer, False)
- except MemoryError:
- fatalerror("out of memory (from JITted code)")
- res = lltype.nullptr(llmemory.GCREF.TO)
+ res = llop1.do_malloc_fixedsize_clear(llmemory.GCREF,
+ type_id, size, True,
+ has_finalizer, False)
+ # In case the operation above failed, we are returning NULL
+ # from this function to assembler. There is also an RPython
+ # exception set, typically MemoryError; but it's easier and
+ # faster to check for the NULL return value, as done by
+ # translator/exceptiontransform.py.
#llop.debug_print(lltype.Void, "\tmalloc_basic", size, type_id,
# "-->", res)
return res
@@ -571,14 +572,10 @@
def malloc_array(itemsize, tid, num_elem):
type_id = llop.extract_ushort(llgroup.HALFWORD, tid)
check_typeid(type_id)
- try:
- return llop1.do_malloc_varsize_clear(
- llmemory.GCREF,
- type_id, num_elem, self.array_basesize, itemsize,
- self.array_length_ofs, True)
- except MemoryError:
- fatalerror("out of memory (from JITted code)")
- return lltype.nullptr(llmemory.GCREF.TO)
+ return llop1.do_malloc_varsize_clear(
+ llmemory.GCREF,
+ type_id, num_elem, self.array_basesize, itemsize,
+ self.array_length_ofs, True)
self.malloc_array = malloc_array
self.GC_MALLOC_ARRAY = lltype.Ptr(lltype.FuncType(
[lltype.Signed] * 3, llmemory.GCREF))
@@ -591,23 +588,15 @@
unicode_type_id = self.layoutbuilder.get_type_id(rstr.UNICODE)
#
def malloc_str(length):
- try:
- return llop1.do_malloc_varsize_clear(
- llmemory.GCREF,
- str_type_id, length, str_basesize, str_itemsize,
- str_ofs_length, True)
- except MemoryError:
- fatalerror("out of memory (from JITted code)")
- return lltype.nullptr(llmemory.GCREF.TO)
+ return llop1.do_malloc_varsize_clear(
+ llmemory.GCREF,
+ str_type_id, length, str_basesize, str_itemsize,
+ str_ofs_length, True)
def malloc_unicode(length):
- try:
- return llop1.do_malloc_varsize_clear(
- llmemory.GCREF,
- unicode_type_id, length, unicode_basesize,unicode_itemsize,
- unicode_ofs_length, True)
- except MemoryError:
- fatalerror("out of memory (from JITted code)")
- return lltype.nullptr(llmemory.GCREF.TO)
+ return llop1.do_malloc_varsize_clear(
+ llmemory.GCREF,
+ unicode_type_id, length, unicode_basesize,unicode_itemsize,
+ unicode_ofs_length, True)
self.malloc_str = malloc_str
self.malloc_unicode = malloc_unicode
self.GC_MALLOC_STR_UNICODE = lltype.Ptr(lltype.FuncType(
@@ -628,16 +617,12 @@
if self.DEBUG:
random_usage_of_xmm_registers()
assert size >= self.minimal_size_in_nursery
- try:
- # NB. although we call do_malloc_fixedsize_clear() here,
- # it's a bit of a hack because we set tid to 0 and may
- # also use it to allocate varsized objects. The tid
- # and possibly the length are both set afterward.
- gcref = llop1.do_malloc_fixedsize_clear(llmemory.GCREF,
- 0, size, True, False, False)
- except MemoryError:
- fatalerror("out of memory (from JITted code)")
- return 0
+ # NB. although we call do_malloc_fixedsize_clear() here,
+ # it's a bit of a hack because we set tid to 0 and may
+ # also use it to allocate varsized objects. The tid
+ # and possibly the length are both set afterward.
+ gcref = llop1.do_malloc_fixedsize_clear(llmemory.GCREF,
+ 0, size, True, False, False)
return rffi.cast(lltype.Signed, gcref)
self.malloc_slowpath = malloc_slowpath
self.MALLOC_SLOWPATH = lltype.FuncType([lltype.Signed], lltype.Signed)
diff --git a/pypy/jit/backend/llsupport/llmodel.py b/pypy/jit/backend/llsupport/llmodel.py
--- a/pypy/jit/backend/llsupport/llmodel.py
+++ b/pypy/jit/backend/llsupport/llmodel.py
@@ -259,7 +259,7 @@
def calldescrof_dynamic(self, ffi_args, ffi_result, extrainfo=None):
from pypy.jit.backend.llsupport import ffisupport
- return ffisupport.get_call_descr_dynamic(ffi_args, ffi_result,
+ return ffisupport.get_call_descr_dynamic(self, ffi_args, ffi_result,
extrainfo)
def get_overflow_error(self):
@@ -499,7 +499,7 @@
def bh_call_i(self, func, calldescr, args_i, args_r, args_f):
assert isinstance(calldescr, BaseIntCallDescr)
if not we_are_translated():
- calldescr.verify_types(args_i, args_r, args_f, history.INT)
+ calldescr.verify_types(args_i, args_r, args_f, history.INT + 'S')
return calldescr.call_stub(func, args_i, args_r, args_f)
def bh_call_r(self, func, calldescr, args_i, args_r, args_f):
diff --git a/pypy/jit/backend/llsupport/test/test_descr.py b/pypy/jit/backend/llsupport/test/test_descr.py
--- a/pypy/jit/backend/llsupport/test/test_descr.py
+++ b/pypy/jit/backend/llsupport/test/test_descr.py
@@ -52,7 +52,8 @@
S = lltype.GcStruct('S', ('x', lltype.Char),
('y', lltype.Ptr(T)),
('z', lltype.Ptr(U)),
- ('f', lltype.Float))
+ ('f', lltype.Float),
+ ('s', lltype.SingleFloat))
assert getFieldDescrClass(lltype.Ptr(T)) is GcPtrFieldDescr
assert getFieldDescrClass(lltype.Ptr(U)) is NonGcPtrFieldDescr
cls = getFieldDescrClass(lltype.Char)
@@ -61,6 +62,10 @@
clsf = getFieldDescrClass(lltype.Float)
assert clsf != cls
assert clsf == getFieldDescrClass(lltype.Float)
+ clss = getFieldDescrClass(lltype.SingleFloat)
+ assert clss not in (cls, clsf)
+ assert clss == getFieldDescrClass(lltype.SingleFloat)
+ assert clss == getFieldDescrClass(rffi.UINT) # for now
#
c0 = GcCache(False)
c1 = GcCache(True)
@@ -72,14 +77,17 @@
descr_y = get_field_descr(c2, S, 'y')
descr_z = get_field_descr(c2, S, 'z')
descr_f = get_field_descr(c2, S, 'f')
+ descr_s = get_field_descr(c2, S, 's')
assert descr_x.__class__ is cls
assert descr_y.__class__ is GcPtrFieldDescr
assert descr_z.__class__ is NonGcPtrFieldDescr
assert descr_f.__class__ is clsf
+ assert descr_s.__class__ is clss
assert descr_x.name == 'S.x'
assert descr_y.name == 'S.y'
assert descr_z.name == 'S.z'
assert descr_f.name == 'S.f'
+ assert descr_s.name == 'S.s'
if not tsc:
assert descr_x.offset < descr_y.offset < descr_z.offset
assert descr_x.sort_key() < descr_y.sort_key() < descr_z.sort_key()
@@ -87,23 +95,29 @@
assert descr_y.get_field_size(False) == rffi.sizeof(lltype.Ptr(T))
assert descr_z.get_field_size(False) == rffi.sizeof(lltype.Ptr(U))
assert descr_f.get_field_size(False) == rffi.sizeof(lltype.Float)
+ assert descr_s.get_field_size(False) == rffi.sizeof(
+ lltype.SingleFloat)
else:
assert isinstance(descr_x.offset, Symbolic)
assert isinstance(descr_y.offset, Symbolic)
assert isinstance(descr_z.offset, Symbolic)
assert isinstance(descr_f.offset, Symbolic)
+ assert isinstance(descr_s.offset, Symbolic)
assert isinstance(descr_x.get_field_size(True), Symbolic)
assert isinstance(descr_y.get_field_size(True), Symbolic)
assert isinstance(descr_z.get_field_size(True), Symbolic)
assert isinstance(descr_f.get_field_size(True), Symbolic)
+ assert isinstance(descr_s.get_field_size(True), Symbolic)
assert not descr_x.is_pointer_field()
assert descr_y.is_pointer_field()
assert not descr_z.is_pointer_field()
assert not descr_f.is_pointer_field()
+ assert not descr_s.is_pointer_field()
assert not descr_x.is_float_field()
assert not descr_y.is_float_field()
assert not descr_z.is_float_field()
assert descr_f.is_float_field()
+ assert not descr_s.is_float_field()
def test_get_field_descr_sign():
@@ -135,6 +149,7 @@
A2 = lltype.GcArray(lltype.Ptr(T))
A3 = lltype.GcArray(lltype.Ptr(U))
A4 = lltype.GcArray(lltype.Float)
+ A5 = lltype.GcArray(lltype.SingleFloat)
assert getArrayDescrClass(A2) is GcPtrArrayDescr
assert getArrayDescrClass(A3) is NonGcPtrArrayDescr
cls = getArrayDescrClass(A1)
@@ -143,25 +158,32 @@
clsf = getArrayDescrClass(A4)
assert clsf != cls
assert clsf == getArrayDescrClass(lltype.GcArray(lltype.Float))
+ clss = getArrayDescrClass(A5)
+ assert clss not in (clsf, cls)
+ assert clss == getArrayDescrClass(lltype.GcArray(rffi.UINT))
#
c0 = GcCache(False)
descr1 = get_array_descr(c0, A1)
descr2 = get_array_descr(c0, A2)
descr3 = get_array_descr(c0, A3)
descr4 = get_array_descr(c0, A4)
+ descr5 = get_array_descr(c0, A5)
assert descr1.__class__ is cls
assert descr2.__class__ is GcPtrArrayDescr
assert descr3.__class__ is NonGcPtrArrayDescr
assert descr4.__class__ is clsf
+ assert descr5.__class__ is clss
assert descr1 == get_array_descr(c0, lltype.GcArray(lltype.Char))
assert not descr1.is_array_of_pointers()
assert descr2.is_array_of_pointers()
assert not descr3.is_array_of_pointers()
assert not descr4.is_array_of_pointers()
+ assert not descr5.is_array_of_pointers()
assert not descr1.is_array_of_floats()
assert not descr2.is_array_of_floats()
assert not descr3.is_array_of_floats()
assert descr4.is_array_of_floats()
+ assert not descr5.is_array_of_floats()
#
def get_alignment(code):
# Retrieve default alignment for the compiler/platform
@@ -170,27 +192,33 @@
assert descr2.get_base_size(False) == get_alignment('p')
assert descr3.get_base_size(False) == get_alignment('p')
assert descr4.get_base_size(False) == get_alignment('d')
+ assert descr5.get_base_size(False) == get_alignment('f')
assert descr1.get_ofs_length(False) == 0
assert descr2.get_ofs_length(False) == 0
assert descr3.get_ofs_length(False) == 0
assert descr4.get_ofs_length(False) == 0
+ assert descr5.get_ofs_length(False) == 0
assert descr1.get_item_size(False) == rffi.sizeof(lltype.Char)
assert descr2.get_item_size(False) == rffi.sizeof(lltype.Ptr(T))
assert descr3.get_item_size(False) == rffi.sizeof(lltype.Ptr(U))
assert descr4.get_item_size(False) == rffi.sizeof(lltype.Float)
+ assert descr5.get_item_size(False) == rffi.sizeof(lltype.SingleFloat)
#
assert isinstance(descr1.get_base_size(True), Symbolic)
assert isinstance(descr2.get_base_size(True), Symbolic)
assert isinstance(descr3.get_base_size(True), Symbolic)
assert isinstance(descr4.get_base_size(True), Symbolic)
+ assert isinstance(descr5.get_base_size(True), Symbolic)
assert isinstance(descr1.get_ofs_length(True), Symbolic)
assert isinstance(descr2.get_ofs_length(True), Symbolic)
assert isinstance(descr3.get_ofs_length(True), Symbolic)
assert isinstance(descr4.get_ofs_length(True), Symbolic)
+ assert isinstance(descr5.get_ofs_length(True), Symbolic)
assert isinstance(descr1.get_item_size(True), Symbolic)
assert isinstance(descr2.get_item_size(True), Symbolic)
assert isinstance(descr3.get_item_size(True), Symbolic)
assert isinstance(descr4.get_item_size(True), Symbolic)
+ assert isinstance(descr5.get_item_size(True), Symbolic)
CA = rffi.CArray(lltype.Signed)
descr = get_array_descr(c0, CA)
assert not descr.is_array_of_floats()
@@ -210,6 +238,11 @@
assert descr.is_array_of_floats()
assert descr.get_base_size(False) == 0
assert descr.get_ofs_length(False) == -1
+ CA = rffi.CArray(rffi.FLOAT)
+ descr = get_array_descr(c0, CA)
+ assert not descr.is_array_of_floats()
+ assert descr.get_base_size(False) == 0
+ assert descr.get_ofs_length(False) == -1
def test_get_array_descr_sign():
@@ -257,6 +290,11 @@
assert descr4.get_result_size(False) == rffi.sizeof(lltype.Float)
assert descr4.get_return_type() == history.FLOAT
assert descr4.arg_classes == "ff"
+ #
+ descr5 = get_call_descr(c0, [lltype.SingleFloat], lltype.SingleFloat)
+ assert descr5.get_result_size(False) == rffi.sizeof(lltype.SingleFloat)
+ assert descr5.get_return_type() == "S"
+ assert descr5.arg_classes == "S"
def test_get_call_descr_not_translated_longlong():
if sys.maxint > 2147483647:
@@ -286,6 +324,11 @@
assert isinstance(descr4.get_result_size(True), Symbolic)
assert descr4.get_return_type() == history.FLOAT
assert descr4.arg_classes == "ff"
+ #
+ descr5 = get_call_descr(c1, [lltype.SingleFloat], lltype.SingleFloat)
+ assert isinstance(descr5.get_result_size(True), Symbolic)
+ assert descr5.get_return_type() == "S"
+ assert descr5.arg_classes == "S"
def test_call_descr_extra_info():
c1 = GcCache(True)
@@ -345,8 +388,11 @@
#
descr4f = get_call_descr(c0, [lltype.Char, lltype.Ptr(S)], lltype.Float)
assert 'FloatCallDescr' in descr4f.repr_of_descr()
+ #
+ descr5f = get_call_descr(c0, [lltype.Char], lltype.SingleFloat)
+ assert 'SingleFloatCallDescr' in descr5f.repr_of_descr()
-def test_call_stubs():
+def test_call_stubs_1():
c0 = GcCache(False)
ARGS = [lltype.Char, lltype.Signed]
RES = lltype.Char
@@ -360,6 +406,8 @@
res = call_stub(rffi.cast(lltype.Signed, fnptr), [1, 2], None, None)
assert res == ord('c')
+def test_call_stubs_2():
+ c0 = GcCache(False)
ARRAY = lltype.GcArray(lltype.Signed)
ARGS = [lltype.Float, lltype.Ptr(ARRAY)]
RES = lltype.Float
@@ -375,3 +423,27 @@
res = descr2.call_stub(rffi.cast(lltype.Signed, fnptr),
[], [opaquea], [longlong.getfloatstorage(3.5)])
assert longlong.getrealfloat(res) == 4.5
+
+def test_call_stubs_single_float():
+ from pypy.rlib.longlong2float import uint2singlefloat, singlefloat2uint
+ from pypy.rlib.rarithmetic import r_singlefloat, intmask
+ #
+ c0 = GcCache(False)
+ ARGS = [lltype.SingleFloat, lltype.SingleFloat, lltype.SingleFloat]
+ RES = lltype.SingleFloat
+
+ def f(a, b, c):
+ a = float(a)
+ b = float(b)
+ c = float(c)
+ x = a - (b / c)
+ return r_singlefloat(x)
+
+ fnptr = llhelper(lltype.Ptr(lltype.FuncType(ARGS, RES)), f)
+ descr2 = get_call_descr(c0, ARGS, RES)
+ a = intmask(singlefloat2uint(r_singlefloat(-10.0)))
+ b = intmask(singlefloat2uint(r_singlefloat(3.0)))
+ c = intmask(singlefloat2uint(r_singlefloat(2.0)))
+ res = descr2.call_stub(rffi.cast(lltype.Signed, fnptr),
+ [a, b, c], [], [])
+ assert float(uint2singlefloat(rffi.r_uint(res))) == -11.5
diff --git a/pypy/jit/backend/llsupport/test/test_ffisupport.py b/pypy/jit/backend/llsupport/test/test_ffisupport.py
--- a/pypy/jit/backend/llsupport/test/test_ffisupport.py
+++ b/pypy/jit/backend/llsupport/test/test_ffisupport.py
@@ -1,24 +1,52 @@
from pypy.rlib.libffi import types
-from pypy.jit.backend.llsupport.ffisupport import get_call_descr_dynamic, \
- VoidCallDescr, DynamicIntCallDescr
-
+from pypy.jit.codewriter.longlong import is_64_bit
+from pypy.jit.backend.llsupport.ffisupport import *
+
+
+class FakeCPU:
+ def __init__(self, supports_floats=False, supports_longlong=False,
+ supports_singlefloats=False):
+ self.supports_floats = supports_floats
+ self.supports_longlong = supports_longlong
+ self.supports_singlefloats = supports_singlefloats
+
+
def test_call_descr_dynamic():
+ args = [types.sint, types.pointer]
+ descr = get_call_descr_dynamic(FakeCPU(), args, types.sint)
+ assert isinstance(descr, DynamicIntCallDescr)
+ assert descr.arg_classes == 'ii'
args = [types.sint, types.double, types.pointer]
- descr = get_call_descr_dynamic(args, types.void)
+ descr = get_call_descr_dynamic(FakeCPU(), args, types.void)
+ assert descr is None # missing floats
+ descr = get_call_descr_dynamic(FakeCPU(supports_floats=True),
+ args, types.void)
assert isinstance(descr, VoidCallDescr)
assert descr.arg_classes == 'ifi'
- descr = get_call_descr_dynamic([], types.sint8)
+ descr = get_call_descr_dynamic(FakeCPU(), [], types.sint8)
assert isinstance(descr, DynamicIntCallDescr)
assert descr.get_result_size(False) == 1
assert descr.is_result_signed() == True
- descr = get_call_descr_dynamic([], types.uint8)
+ descr = get_call_descr_dynamic(FakeCPU(), [], types.uint8)
assert isinstance(descr, DynamicIntCallDescr)
assert descr.get_result_size(False) == 1
assert descr.is_result_signed() == False
- descr = get_call_descr_dynamic([], types.float)
- assert descr is None # single floats are not supported so far
-
+ if not is_64_bit:
+ descr = get_call_descr_dynamic(FakeCPU(), [], types.slonglong)
+ assert descr is None # missing longlongs
+ descr = get_call_descr_dynamic(FakeCPU(supports_longlong=True),
+ [], types.slonglong)
+ assert isinstance(descr, LongLongCallDescr)
+ else:
+ assert types.slonglong is types.slong
+
+ descr = get_call_descr_dynamic(FakeCPU(), [], types.float)
+ assert descr is None # missing singlefloats
+ descr = get_call_descr_dynamic(FakeCPU(supports_singlefloats=True),
+ [], types.float)
+ SingleFloatCallDescr = getCallDescrClass(rffi.FLOAT)
+ assert isinstance(descr, SingleFloatCallDescr)
diff --git a/pypy/jit/backend/model.py b/pypy/jit/backend/model.py
--- a/pypy/jit/backend/model.py
+++ b/pypy/jit/backend/model.py
@@ -8,12 +8,13 @@
# ^^^ This is only useful on 32-bit platforms. If True,
# longlongs are supported by the JIT, but stored as doubles.
# Boxes and Consts are BoxFloats and ConstFloats.
+ supports_singlefloats = False
done_with_this_frame_void_v = -1
done_with_this_frame_int_v = -1
done_with_this_frame_ref_v = -1
done_with_this_frame_float_v = -1
- exit_frame_with_exception_v = -1
+ propagate_exception_v = -1
total_compiled_loops = 0
total_compiled_bridges = 0
total_freed_loops = 0
diff --git a/pypy/jit/backend/test/calling_convention_test.py b/pypy/jit/backend/test/calling_convention_test.py
--- a/pypy/jit/backend/test/calling_convention_test.py
+++ b/pypy/jit/backend/test/calling_convention_test.py
@@ -290,3 +290,58 @@
assert abs(x - expected_result) < 0.0001
finally:
del self.cpu.done_with_this_frame_float_v
+
+ def test_call_with_singlefloats(self):
+ cpu = self.cpu
+ if not cpu.supports_floats or not cpu.supports_singlefloats:
+ py.test.skip('requires floats and singlefloats')
+
+ import random
+ from pypy.rlib.libffi import types
+ from pypy.rlib.rarithmetic import r_singlefloat
+
+ def func(*args):
+ res = 0.0
+ for i, x in enumerate(args):
+ res += (i + 1.1) * float(x)
+ return res
+
+ F = lltype.Float
+ S = lltype.SingleFloat
+ I = lltype.Signed
+ floats = [random.random() - 0.5 for i in range(8)]
+ singlefloats = [r_singlefloat(random.random() - 0.5) for i in range(8)]
+ ints = [random.randrange(-99, 99) for i in range(8)]
+ for repeat in range(100):
+ args = []
+ argvalues = []
+ argslist = []
+ local_floats = list(floats)
+ local_singlefloats = list(singlefloats)
+ local_ints = list(ints)
+ for i in range(8):
+ case = random.randrange(0, 3)
+ if case == 0:
+ args.append(F)
+ arg = local_floats.pop()
+ argslist.append(boxfloat(arg))
+ elif case == 1:
+ args.append(S)
+ arg = local_singlefloats.pop()
+ argslist.append(BoxInt(longlong.singlefloat2int(arg)))
+ else:
+ args.append(I)
+ arg = local_ints.pop()
+ argslist.append(BoxInt(arg))
+ argvalues.append(arg)
+ FUNC = self.FuncType(args, F)
+ FPTR = self.Ptr(FUNC)
+ func_ptr = llhelper(FPTR, func)
+ calldescr = cpu.calldescrof(FUNC, FUNC.ARGS, FUNC.RESULT)
+ funcbox = self.get_funcbox(cpu, func_ptr)
+
+ res = self.execute_operation(rop.CALL,
+ [funcbox] + argslist,
+ 'float', descr=calldescr)
+ expected = func(*argvalues)
+ assert abs(res.getfloat() - expected) < 0.0001
diff --git a/pypy/jit/backend/test/runner_test.py b/pypy/jit/backend/test/runner_test.py
--- a/pypy/jit/backend/test/runner_test.py
+++ b/pypy/jit/backend/test/runner_test.py
@@ -2734,6 +2734,65 @@
'float', descr=calldescr)
assert res.getfloatstorage() == expected
+ def test_singlefloat_result_of_call_direct(self):
+ if not self.cpu.supports_singlefloats:
+ py.test.skip("singlefloat test")
+ from pypy.translator.tool.cbuild import ExternalCompilationInfo
+ from pypy.rlib.rarithmetic import r_singlefloat
+ eci = ExternalCompilationInfo(
+ separate_module_sources=["""
+ float fn_test_result_of_call(float x)
+ {
+ return x / 2.0f;
+ }
+ """],
+ export_symbols=['fn_test_result_of_call'])
+ f = rffi.llexternal('fn_test_result_of_call', [lltype.SingleFloat],
+ lltype.SingleFloat,
+ compilation_info=eci, _nowrapper=True)
+ value = r_singlefloat(-42.5)
+ expected = r_singlefloat(-21.25)
+ assert f(value) == expected
+ #
+ FUNC = self.FuncType([lltype.SingleFloat], lltype.SingleFloat)
+ FPTR = self.Ptr(FUNC)
+ calldescr = self.cpu.calldescrof(FUNC, FUNC.ARGS, FUNC.RESULT)
+ ivalue = longlong.singlefloat2int(value)
+ iexpected = longlong.singlefloat2int(expected)
+ x = self.cpu.bh_call_i(self.get_funcbox(self.cpu, f).value,
+ calldescr, [ivalue], None, None)
+ assert x == iexpected
+
+ def test_singlefloat_result_of_call_compiled(self):
+ if not self.cpu.supports_singlefloats:
+ py.test.skip("test of singlefloat result")
+ from pypy.translator.tool.cbuild import ExternalCompilationInfo
+ from pypy.rlib.rarithmetic import r_singlefloat
+ eci = ExternalCompilationInfo(
+ separate_module_sources=["""
+ float fn_test_result_of_call(float x)
+ {
+ return x / 2.0f;
+ }
+ """],
+ export_symbols=['fn_test_result_of_call'])
+ f = rffi.llexternal('fn_test_result_of_call', [lltype.SingleFloat],
+ lltype.SingleFloat,
+ compilation_info=eci, _nowrapper=True)
+ value = r_singlefloat(-42.5)
+ expected = r_singlefloat(-21.25)
+ assert f(value) == expected
+ #
+ FUNC = self.FuncType([lltype.SingleFloat], lltype.SingleFloat)
+ FPTR = self.Ptr(FUNC)
+ calldescr = self.cpu.calldescrof(FUNC, FUNC.ARGS, FUNC.RESULT)
+ funcbox = self.get_funcbox(self.cpu, f)
+ ivalue = longlong.singlefloat2int(value)
+ iexpected = longlong.singlefloat2int(expected)
+ res = self.execute_operation(rop.CALL, [funcbox, BoxInt(ivalue)],
+ 'int', descr=calldescr)
+ assert res.value == iexpected
+
def test_free_loop_and_bridges(self):
from pypy.jit.backend.llsupport.llmodel import AbstractLLCPU
if not isinstance(self.cpu, AbstractLLCPU):
@@ -2748,6 +2807,26 @@
assert mem2 < mem1
assert mem2 == mem0
+ def test_memoryerror(self):
+ excdescr = BasicFailDescr(666)
+ self.cpu.propagate_exception_v = self.cpu.get_fail_descr_number(
+ excdescr)
+ self.cpu.setup_once() # xxx redo it, because we added
+ # propagate_exception_v
+ i0 = BoxInt()
+ p0 = BoxPtr()
+ operations = [
+ ResOperation(rop.NEWUNICODE, [i0], p0),
+ ResOperation(rop.FINISH, [p0], None, descr=BasicFailDescr(1))
+ ]
+ inputargs = [i0]
+ looptoken = LoopToken()
+ self.cpu.compile_loop(inputargs, operations, looptoken)
+ # overflowing value:
+ self.cpu.set_future_value_int(0, sys.maxint // 4 + 1)
+ fail = self.cpu.execute_token(looptoken)
+ assert fail.identifier == excdescr.identifier
+
class OOtypeBackendTest(BaseBackendTest):
diff --git a/pypy/jit/backend/x86/arch.py b/pypy/jit/backend/x86/arch.py
--- a/pypy/jit/backend/x86/arch.py
+++ b/pypy/jit/backend/x86/arch.py
@@ -27,3 +27,6 @@
# which are used in the malloc itself. They are:
# ecx, ebx, esi, edi [32 and 64 bits]
# r8, r9, r10, r12, r13, r14, r15 [64 bits only]
+#
+# Note that with asmgcc, the locations corresponding to callee-save registers
+# are never used.
diff --git a/pypy/jit/backend/x86/assembler.py b/pypy/jit/backend/x86/assembler.py
--- a/pypy/jit/backend/x86/assembler.py
+++ b/pypy/jit/backend/x86/assembler.py
@@ -56,7 +56,9 @@
self.exc = exc
self.is_guard_not_invalidated = is_guard_not_invalidated
-DEBUG_COUNTER = lltype.Struct('DEBUG_COUNTER', ('i', lltype.Signed))
+DEBUG_COUNTER = lltype.Struct('DEBUG_COUNTER', ('i', lltype.Signed),
+ ('bridge', lltype.Signed), # 0 or 1
+ ('number', lltype.Signed))
class Assembler386(object):
_regalloc = None
@@ -89,6 +91,7 @@
self._current_depths_cache = (0, 0)
self.datablockwrapper = None
self.stack_check_slowpath = 0
+ self.propagate_exception_path = 0
self.teardown()
def leave_jitted_hook(self):
@@ -125,6 +128,7 @@
self._build_failure_recovery(True, withfloats=True)
support.ensure_sse2_floats()
self._build_float_constants()
+ self._build_propagate_exception_path()
if gc_ll_descr.get_malloc_slowpath_addr is not None:
self._build_malloc_slowpath()
self._build_stack_check_slowpath()
@@ -138,6 +142,9 @@
assert self.memcpy_addr != 0, "setup_once() not called?"
self.current_clt = looptoken.compiled_loop_token
self.pending_guard_tokens = []
+ if WORD == 8:
+ self.pending_memoryerror_trampoline_from = []
+ self.error_trampoline_64 = 0
self.mc = codebuf.MachineCodeBlockWrapper()
#assert self.datablockwrapper is None --- but obscure case
# possible, e.g. getting MemoryError and continuing
@@ -147,6 +154,8 @@
def teardown(self):
self.pending_guard_tokens = None
+ if WORD == 8:
+ self.pending_memoryerror_trampoline_from = None
self.mc = None
self.looppos = -1
self.currently_compiling_loop = None
@@ -155,9 +164,12 @@
def finish_once(self):
if self._debug:
debug_start('jit-backend-counts')
- for i in range(len(self.loop_run_counters)):
- struct = self.loop_run_counters[i]
- debug_print(str(i) + ':' + str(struct.i))
+ for struct in self.loop_run_counters:
+ if struct.bridge:
+ prefix = 'bridge '
+ else:
+ prefix = 'loop '
+ debug_print(prefix + str(struct.number) + ':' + str(struct.i))
debug_stop('jit-backend-counts')
def _build_float_constants(self):
@@ -181,6 +193,7 @@
# instructions in assembler, with a mark_gc_roots in between.
# With shadowstack, this is not needed, so we produce a single helper.
gcrootmap = self.cpu.gc_ll_descr.gcrootmap
+ shadow_stack = (gcrootmap is not None and gcrootmap.is_shadow_stack)
#
# ---------- first helper for the slow path of malloc ----------
mc = codebuf.MachineCodeBlockWrapper()
@@ -190,10 +203,19 @@
mc.SUB_rr(edx.value, eax.value) # compute the size we want
addr = self.cpu.gc_ll_descr.get_malloc_slowpath_addr()
#
- if gcrootmap is not None and gcrootmap.is_shadow_stack:
+ # The registers to save in the copy area: with shadowstack, most
+ # registers need to be saved. With asmgcc, the callee-saved registers
+ # don't need to.
+ save_in_copy_area = gpr_reg_mgr_cls.REGLOC_TO_COPY_AREA_OFS.items()
+ if not shadow_stack:
+ save_in_copy_area = [(reg, ofs) for (reg, ofs) in save_in_copy_area
+ if reg not in gpr_reg_mgr_cls.REGLOC_TO_GCROOTMAP_REG_INDEX]
+ #
+ for reg, ofs in save_in_copy_area:
+ mc.MOV_br(ofs, reg.value)
+ #
+ if shadow_stack:
# ---- shadowstack ----
- for reg, ofs in gpr_reg_mgr_cls.REGLOC_TO_COPY_AREA_OFS.items():
- mc.MOV_br(ofs, reg.value)
mc.SUB_ri(esp.value, 16 - WORD) # stack alignment of 16 bytes
if IS_X86_32:
mc.MOV_sr(0, edx.value) # push argument
@@ -201,15 +223,13 @@
mc.MOV_rr(edi.value, edx.value)
mc.CALL(imm(addr))
mc.ADD_ri(esp.value, 16 - WORD)
- for reg, ofs in gpr_reg_mgr_cls.REGLOC_TO_COPY_AREA_OFS.items():
- mc.MOV_rb(reg.value, ofs)
else:
# ---- asmgcc ----
if IS_X86_32:
mc.MOV_sr(WORD, edx.value) # save it as the new argument
elif IS_X86_64:
- # rdi can be clobbered: its content was forced to the stack
- # by _fastpath_malloc(), like all other save_around_call_regs.
+ # rdi can be clobbered: its content was saved in the
+ # copy area of the stack
mc.MOV_rr(edi.value, edx.value)
mc.JMP(imm(addr)) # tail call to the real malloc
rawstart = mc.materialize(self.cpu.asmmemmgr, [])
@@ -217,18 +237,54 @@
# ---------- second helper for the slow path of malloc ----------
mc = codebuf.MachineCodeBlockWrapper()
#
+ for reg, ofs in save_in_copy_area:
+ mc.MOV_rb(reg.value, ofs)
+ assert reg is not eax and reg is not edx
+ #
if self.cpu.supports_floats: # restore the XMM registers
for i in range(self.cpu.NUM_REGS):# from where they were saved
mc.MOVSD_xs(i, (WORD*2)+8*i)
+ #
+ # Note: we check this after the code above, just because the code
+ # above is more than 127 bytes on 64-bits...
+ mc.TEST_rr(eax.value, eax.value)
+ mc.J_il8(rx86.Conditions['Z'], 0) # patched later
+ jz_location = mc.get_relative_pos()
+ #
nursery_free_adr = self.cpu.gc_ll_descr.get_nursery_free_addr()
mc.MOV(edx, heap(nursery_free_adr)) # load this in EDX
mc.RET()
+ #
+ # If the slowpath malloc failed, we raise a MemoryError that
+ # always interrupts the current loop, as a "good enough"
+ # approximation. Also note that we didn't RET from this helper;
+ # but the code we jump to will actually restore the stack
+ # position based on EBP, which will get us out of here for free.
+ offset = mc.get_relative_pos() - jz_location
+ assert 0 < offset <= 127
+ mc.overwrite(jz_location-1, chr(offset))
+ mc.JMP(imm(self.propagate_exception_path))
+ #
rawstart = mc.materialize(self.cpu.asmmemmgr, [])
self.malloc_slowpath2 = rawstart
+ def _build_propagate_exception_path(self):
+ if self.cpu.propagate_exception_v < 0:
+ return # not supported (for tests, or non-translated)
+ #
+ self.mc = codebuf.MachineCodeBlockWrapper()
+ # call on_leave_jitted_save_exc()
+ addr = self.cpu.get_on_leave_jitted_int(save_exception=True)
+ self.mc.CALL(imm(addr))
+ self.mc.MOV_ri(eax.value, self.cpu.propagate_exception_v)
+ self._call_footer()
+ rawstart = self.mc.materialize(self.cpu.asmmemmgr, [])
+ self.propagate_exception_path = rawstart
+ self.mc = None
+
def _build_stack_check_slowpath(self):
_, _, slowpathaddr = self.cpu.insert_stack_check()
- if slowpathaddr == 0 or self.cpu.exit_frame_with_exception_v < 0:
+ if slowpathaddr == 0 or self.cpu.propagate_exception_v < 0:
return # no stack check (for tests, or non-translated)
#
# make a "function" that is called immediately at the start of
@@ -284,19 +340,11 @@
offset = mc.get_relative_pos() - jnz_location
assert 0 < offset <= 127
mc.overwrite(jnz_location-1, chr(offset))
- # clear the exception from the global position
- mc.MOV(eax, heap(self.cpu.pos_exc_value()))
- mc.MOV(heap(self.cpu.pos_exception()), imm0)
- mc.MOV(heap(self.cpu.pos_exc_value()), imm0)
- # save the current exception instance into fail_boxes_ptr[0]
- adr = self.fail_boxes_ptr.get_addr_for_num(0)
- mc.MOV(heap(adr), eax)
- # call the helper function to set the GC flag on the fail_boxes_ptr
- # array (note that there is no exception any more here)
- addr = self.cpu.get_on_leave_jitted_int(save_exception=False)
+ # call on_leave_jitted_save_exc()
+ addr = self.cpu.get_on_leave_jitted_int(save_exception=True)
mc.CALL(imm(addr))
#
- mc.MOV_ri(eax.value, self.cpu.exit_frame_with_exception_v)
+ mc.MOV_ri(eax.value, self.cpu.propagate_exception_v)
#
# footer -- note the ADD, which skips the return address of this
# function, and will instead return to the caller's caller. Note
@@ -392,7 +440,7 @@
self.setup(looptoken)
self.currently_compiling_loop = looptoken
if log:
- self._register_counter()
+ self._register_counter(False, looptoken.number)
operations = self._inject_debugging_code(looptoken, operations)
regalloc = RegAlloc(self, self.cpu.translate_support_code)
@@ -461,7 +509,7 @@
self.setup(original_loop_token)
if log:
- self._register_counter()
+ self._register_counter(True, descr_number)
operations = self._inject_debugging_code(faildescr, operations)
arglocs = self.rebuild_faillocs_from_descr(failure_recovery)
@@ -508,6 +556,8 @@
# at the end of self.mc.
for tok in self.pending_guard_tokens:
tok.pos_recovery_stub = self.generate_quick_failure(tok)
+ if WORD == 8 and len(self.pending_memoryerror_trampoline_from) > 0:
+ self.error_trampoline_64 = self.generate_propagate_error_64()
def patch_pending_failure_recoveries(self, rawstart):
# after we wrote the assembler to raw memory, set up
@@ -544,6 +594,12 @@
# less, we would run into the issue that overwriting the
# 5 bytes here might get a few nonsense bytes at the
# return address of the following CALL.
+ if WORD == 8:
+ for pos_after_jz in self.pending_memoryerror_trampoline_from:
+ assert self.error_trampoline_64 != 0 # only if non-empty
+ mc = codebuf.MachineCodeBlockWrapper()
+ mc.writeimm32(self.error_trampoline_64 - pos_after_jz)
+ mc.copy_to_raw_memory(rawstart + pos_after_jz - 4)
def get_asmmemmgr_blocks(self, looptoken):
clt = looptoken.compiled_loop_token
@@ -558,7 +614,7 @@
return self.mc.materialize(self.cpu.asmmemmgr, allblocks,
self.cpu.gc_ll_descr.gcrootmap)
- def _register_counter(self):
+ def _register_counter(self, bridge, number):
if self._debug:
# YYY very minor leak -- we need the counters to stay alive
# forever, just because we want to report them at the end
@@ -566,6 +622,8 @@
struct = lltype.malloc(DEBUG_COUNTER, flavor='raw',
track_allocation=False)
struct.i = 0
+ struct.bridge = int(bridge)
+ struct.number = number
self.loop_run_counters.append(struct)
def _find_failure_recovery_bytecode(self, faildescr):
@@ -1056,9 +1114,10 @@
self.implement_guard(guard_token, checkfalsecond)
return genop_cmp_guard_float
- def _emit_call(self, force_index, x, arglocs, start=0, tmp=eax):
+ def _emit_call(self, force_index, x, arglocs, start=0, tmp=eax,
+ argtypes=None):
if IS_X86_64:
- return self._emit_call_64(force_index, x, arglocs, start)
+ return self._emit_call_64(force_index, x, arglocs, start, argtypes)
p = 0
n = len(arglocs)
@@ -1086,12 +1145,13 @@
self.mc.CALL(x)
self.mark_gc_roots(force_index)
- def _emit_call_64(self, force_index, x, arglocs, start):
+ def _emit_call_64(self, force_index, x, arglocs, start, argtypes):
src_locs = []
dst_locs = []
xmm_src_locs = []
xmm_dst_locs = []
pass_on_stack = []
+ singlefloats = None
# In reverse order for use with pop()
unused_gpr = [r9, r8, ecx, edx, esi, edi]
@@ -1111,6 +1171,11 @@
xmm_dst_locs.append(unused_xmm.pop())
else:
pass_on_stack.append(loc)
+ elif (argtypes is not None and argtypes[i-start] == 'S' and
+ len(unused_xmm) > 0):
+ # Singlefloat argument
+ if singlefloats is None: singlefloats = []
+ singlefloats.append((loc, unused_xmm.pop()))
else:
if len(unused_gpr) > 0:
src_locs.append(loc)
@@ -1138,9 +1203,15 @@
else:
self.mc.MOV_sr(i*WORD, loc.value)
- # Handle register arguments
+ # Handle register arguments: first remap the xmm arguments
+ remap_frame_layout(self, xmm_src_locs, xmm_dst_locs,
+ X86_64_XMM_SCRATCH_REG)
+ # Load the singlefloat arguments from main regs or stack to xmm regs
+ if singlefloats is not None:
+ for src, dst in singlefloats:
+ self.mc.MOVD(dst, src)
+ # Finally remap the arguments in the main regs
remap_frame_layout(self, src_locs, dst_locs, X86_64_SCRATCH_REG)
- remap_frame_layout(self, xmm_src_locs, xmm_dst_locs, X86_64_XMM_SCRATCH_REG)
self._regalloc.reserve_param(len(pass_on_stack))
self.mc.CALL(x)
@@ -1255,6 +1326,20 @@
def genop_cast_int_to_float(self, op, arglocs, resloc):
self.mc.CVTSI2SD(resloc, arglocs[0])
+ def genop_cast_float_to_singlefloat(self, op, arglocs, resloc):
+ loc0, loctmp = arglocs
+ self.mc.CVTSD2SS(loctmp, loc0)
+ assert isinstance(resloc, RegLoc)
+ assert isinstance(loctmp, RegLoc)
+ self.mc.MOVD_rx(resloc.value, loctmp.value)
+
+ def genop_cast_singlefloat_to_float(self, op, arglocs, resloc):
+ loc0, = arglocs
+ assert isinstance(resloc, RegLoc)
+ assert isinstance(loc0, RegLoc)
+ self.mc.MOVD_xr(resloc.value, loc0.value)
+ self.mc.CVTSS2SD_xx(resloc.value, resloc.value)
+
def genop_guard_int_is_true(self, op, guard_op, guard_token, arglocs, resloc):
guard_opnum = guard_op.getopnum()
self.mc.CMP(arglocs[0], imm0)
@@ -1376,7 +1461,7 @@
assert isinstance(loc_vtable, ImmedLoc)
arglocs = arglocs[:-1]
self.call(self.malloc_func_addr, arglocs, eax)
- # xxx ignore NULL returns for now
+ self.propagate_memoryerror_if_eax_is_null()
self.set_vtable(eax, loc_vtable)
def set_vtable(self, loc, loc_vtable):
@@ -1395,18 +1480,35 @@
def genop_new(self, op, arglocs, result_loc):
assert result_loc is eax
self.call(self.malloc_func_addr, arglocs, eax)
+ self.propagate_memoryerror_if_eax_is_null()
def genop_new_array(self, op, arglocs, result_loc):
assert result_loc is eax
self.call(self.malloc_array_func_addr, arglocs, eax)
+ self.propagate_memoryerror_if_eax_is_null()
def genop_newstr(self, op, arglocs, result_loc):
assert result_loc is eax
self.call(self.malloc_str_func_addr, arglocs, eax)
+ self.propagate_memoryerror_if_eax_is_null()
def genop_newunicode(self, op, arglocs, result_loc):
assert result_loc is eax
self.call(self.malloc_unicode_func_addr, arglocs, eax)
+ self.propagate_memoryerror_if_eax_is_null()
+
+ def propagate_memoryerror_if_eax_is_null(self):
+ # if self.propagate_exception_path == 0 (tests), this may jump to 0
+ # and segfaults. too bad. the alternative is to continue anyway
+ # with eax==0, but that will segfault too.
+ self.mc.TEST_rr(eax.value, eax.value)
+ if WORD == 4:
+ self.mc.J_il(rx86.Conditions['Z'], self.propagate_exception_path)
+ self.mc.add_pending_relocation()
+ elif WORD == 8:
+ self.mc.J_il(rx86.Conditions['Z'], 0)
+ pos = self.mc.get_relative_pos()
+ self.pending_memoryerror_trampoline_from.append(pos)
# ----------
@@ -1678,6 +1780,12 @@
return GuardToken(faildescr, failargs, fail_locs, exc,
is_guard_not_invalidated)
+ def generate_propagate_error_64(self):
+ assert WORD == 8
+ startpos = self.mc.get_relative_pos()
+ self.mc.JMP(imm(self.propagate_exception_path))
+ return startpos
+
def generate_quick_failure(self, guardtok):
"""Generate the initial code for handling a failure. We try to
keep it as compact as possible.
@@ -2013,7 +2121,8 @@
else:
tmp = eax
- self._emit_call(force_index, x, arglocs, 3, tmp=tmp)
+ self._emit_call(force_index, x, arglocs, 3, tmp=tmp,
+ argtypes=op.getdescr().get_arg_types())
if IS_X86_32 and isinstance(resloc, StackLoc) and resloc.width == 8:
# a float or a long long return
@@ -2025,7 +2134,19 @@
# and this way is simpler also because the result loc
# can just be always a stack location
else:
- self.mc.FSTP_b(resloc.value) # float return
+ self.mc.FSTPL_b(resloc.value) # float return
+ elif op.getdescr().get_return_type() == 'S':
+ # singlefloat return
+ assert resloc is eax
+ if IS_X86_32:
+ # must convert ST(0) to a 32-bit singlefloat and load it into EAX
+ # mess mess mess
+ self.mc.SUB_ri(esp.value, 4)
+ self.mc.FSTPS_s(0)
+ self.mc.POP_r(eax.value)
+ elif IS_X86_64:
+ # must copy from the lower 32 bits of XMM0 into eax
+ self.mc.MOVD_rx(eax.value, xmm0.value)
elif size == WORD:
assert resloc is eax or resloc is xmm0 # a full word
elif size == 0:
@@ -2183,7 +2304,7 @@
self._emit_call(fail_index, imm(asm_helper_adr), [eax, arglocs[1]], 0,
tmp=ecx)
if IS_X86_32 and isinstance(result_loc, StackLoc) and result_loc.type == FLOAT:
- self.mc.FSTP_b(result_loc.value)
+ self.mc.FSTPL_b(result_loc.value)
#else: result_loc is already either eax or None, checked below
self.mc.JMP_l8(0) # jump to done, patched later
jmp_location = self.mc.get_relative_pos()
@@ -2424,8 +2545,7 @@
# there are two helpers to call only with asmgcc
slowpath_addr1 = self.malloc_slowpath1
self.mc.CALL(imm(slowpath_addr1))
- self.mark_gc_roots(self.write_new_force_index(),
- use_copy_area=shadow_stack)
+ self.mark_gc_roots(self.write_new_force_index(), use_copy_area=True)
slowpath_addr2 = self.malloc_slowpath2
self.mc.CALL(imm(slowpath_addr2))
diff --git a/pypy/jit/backend/x86/codebuf.py b/pypy/jit/backend/x86/codebuf.py
--- a/pypy/jit/backend/x86/codebuf.py
+++ b/pypy/jit/backend/x86/codebuf.py
@@ -25,8 +25,11 @@
self.init_block_builder()
# a list of relative positions; for each position p, the bytes
# at [p-4:p] encode an absolute address that will need to be
- # made relative.
- self.relocations = []
+ # made relative. Only works on 32-bit!
+ if WORD == 4:
+ self.relocations = []
+ else:
+ self.relocations = None
#
# ResOperation --> offset in the assembly.
# ops_offset[None] represents the beginning of the code after the last op
@@ -42,9 +45,10 @@
def copy_to_raw_memory(self, addr):
self._copy_to_raw_memory(addr)
- for reloc in self.relocations:
- p = addr + reloc
- adr = rffi.cast(rffi.LONGP, p - WORD)
- adr[0] = intmask(adr[0] - p)
+ if self.relocations is not None:
+ for reloc in self.relocations:
+ p = addr + reloc
+ adr = rffi.cast(rffi.LONGP, p - WORD)
+ adr[0] = intmask(adr[0] - p)
valgrind.discard_translations(addr, self.get_relative_pos())
self._dump(addr, "jit-backend-dump", backend_name)
diff --git a/pypy/jit/backend/x86/regalloc.py b/pypy/jit/backend/x86/regalloc.py
--- a/pypy/jit/backend/x86/regalloc.py
+++ b/pypy/jit/backend/x86/regalloc.py
@@ -705,6 +705,17 @@
self.Perform(op, [loc0], loc1)
self.rm.possibly_free_var(op.getarg(0))
+ def consider_cast_float_to_singlefloat(self, op):
+ loc0 = self.xrm.make_sure_var_in_reg(op.getarg(0))
+ loc1 = self.rm.force_allocate_reg(op.result)
+ self.xrm.possibly_free_var(op.getarg(0))
+ tmpxvar = TempBox()
+ loctmp = self.xrm.force_allocate_reg(tmpxvar) # may be equal to loc0
+ self.xrm.possibly_free_var(tmpxvar)
+ self.Perform(op, [loc0, loctmp], loc1)
+
+ consider_cast_singlefloat_to_float = consider_cast_int_to_float
+
def _consider_llong_binop_xx(self, op):
# must force both arguments into xmm registers, because we don't
# know if they will be suitably aligned. Exception: if the second
@@ -921,27 +932,13 @@
def _do_fastpath_malloc(self, op, size, tid):
gc_ll_descr = self.assembler.cpu.gc_ll_descr
self.rm.force_allocate_reg(op.result, selected_reg=eax)
-
- if gc_ll_descr.gcrootmap and gc_ll_descr.gcrootmap.is_shadow_stack:
- # ---- shadowstack ----
- # We need edx as a temporary, but otherwise don't save any more
- # register. See comments in _build_malloc_slowpath().
- tmp_box = TempBox()
- self.rm.force_allocate_reg(tmp_box, selected_reg=edx)
- self.rm.possibly_free_var(tmp_box)
- else:
- # ---- asmgcc ----
- # We need to force-allocate each of save_around_call_regs now.
- # The alternative would be to save and restore them around the
- # actual call to malloc(), in the rare case where we need to do
- # it; however, mark_gc_roots() would need to be adapted to know
- # where the variables end up being saved. Messy.
- for reg in self.rm.save_around_call_regs:
- if reg is not eax:
- tmp_box = TempBox()
- self.rm.force_allocate_reg(tmp_box, selected_reg=reg)
- self.rm.possibly_free_var(tmp_box)
-
+ #
+ # We need edx as a temporary, but otherwise don't save any more
+ # register. See comments in _build_malloc_slowpath().
+ tmp_box = TempBox()
+ self.rm.force_allocate_reg(tmp_box, selected_reg=edx)
+ self.rm.possibly_free_var(tmp_box)
+ #
self.assembler.malloc_cond(
gc_ll_descr.get_nursery_free_addr(),
gc_ll_descr.get_nursery_top_addr(),
@@ -1337,14 +1334,26 @@
if reg is eax:
continue # ok to ignore this one
if (isinstance(v, BoxPtr) and self.rm.stays_alive(v)):
- if use_copy_area:
- assert reg in self.rm.REGLOC_TO_COPY_AREA_OFS
- area_offset = self.rm.REGLOC_TO_COPY_AREA_OFS[reg]
- gcrootmap.add_frame_offset(shape, area_offset)
- else:
- assert reg in self.rm.REGLOC_TO_GCROOTMAP_REG_INDEX
- gcrootmap.add_callee_save_reg(
- shape, self.rm.REGLOC_TO_GCROOTMAP_REG_INDEX[reg])
+ #
+ # The register 'reg' is alive across this call.
+ gcrootmap = self.assembler.cpu.gc_ll_descr.gcrootmap
+ if gcrootmap is None or not gcrootmap.is_shadow_stack:
+ #
+ # Asmgcc: if reg is a callee-save register, we can
+ # explicitly mark it as containing a BoxPtr.
+ if reg in self.rm.REGLOC_TO_GCROOTMAP_REG_INDEX:
+ gcrootmap.add_callee_save_reg(
+ shape, self.rm.REGLOC_TO_GCROOTMAP_REG_INDEX[reg])
+ continue
+ #
+ # Else, 'use_copy_area' must be True (otherwise this BoxPtr
+ # should not be in a register). The copy area contains the
+ # real value of the register.
+ assert use_copy_area
+ assert reg in self.rm.REGLOC_TO_COPY_AREA_OFS
+ area_offset = self.rm.REGLOC_TO_COPY_AREA_OFS[reg]
+ gcrootmap.add_frame_offset(shape, area_offset)
+ #
return gcrootmap.compress_callshape(shape,
self.assembler.datablockwrapper)
diff --git a/pypy/jit/backend/x86/regloc.py b/pypy/jit/backend/x86/regloc.py
--- a/pypy/jit/backend/x86/regloc.py
+++ b/pypy/jit/backend/x86/regloc.py
@@ -521,6 +521,8 @@
UCOMISD = _binaryop('UCOMISD')
CVTSI2SD = _binaryop('CVTSI2SD')
CVTTSD2SI = _binaryop('CVTTSD2SI')
+ CVTSD2SS = _binaryop('CVTSD2SS')
+ CVTSS2SD = _binaryop('CVTSS2SD')
SQRTSD = _binaryop('SQRTSD')
@@ -534,6 +536,8 @@
PXOR = _binaryop('PXOR')
PCMPEQD = _binaryop('PCMPEQD')
+ MOVD = _binaryop('MOVD')
+
CALL = _relative_unaryop('CALL')
JMP = _relative_unaryop('JMP')
diff --git a/pypy/jit/backend/x86/runner.py b/pypy/jit/backend/x86/runner.py
--- a/pypy/jit/backend/x86/runner.py
+++ b/pypy/jit/backend/x86/runner.py
@@ -19,6 +19,7 @@
class AbstractX86CPU(AbstractLLCPU):
debug = True
supports_floats = True
+ supports_singlefloats = True
BOOTSTRAP_TP = lltype.FuncType([], lltype.Signed)
dont_keepalive_stuff = False # for tests
diff --git a/pypy/jit/backend/x86/rx86.py b/pypy/jit/backend/x86/rx86.py
--- a/pypy/jit/backend/x86/rx86.py
+++ b/pypy/jit/backend/x86/rx86.py
@@ -573,7 +573,8 @@
BTS_jr = insn(rex_w, '\x0F\xAB', register(2,8), abs_, immediate(1))
# x87 instructions
- FSTP_b = insn('\xDD', orbyte(3<<3), stack_bp(1))
+ FSTPL_b = insn('\xDD', orbyte(3<<3), stack_bp(1)) # rffi.DOUBLE ('as' wants L??)
+ FSTPS_s = insn('\xD9', orbyte(3<<3), stack_sp(1)) # lltype.SingleFloat
# ------------------------------ Random mess -----------------------
RDTSC = insn('\x0F\x31')
@@ -590,8 +591,18 @@
CVTTSD2SI_rx = xmminsn('\xF2', rex_w, '\x0F\x2C', register(1, 8), register(2), '\xC0')
CVTTSD2SI_rb = xmminsn('\xF2', rex_w, '\x0F\x2C', register(1, 8), stack_bp(2))
- MOVD_rx = xmminsn('\x66', rex_w, '\x0F\x7E', register(2, 8), register(1), '\xC0')
- MOVD_xr = xmminsn('\x66', rex_w, '\x0F\x6E', register(1, 8), register(2), '\xC0')
+ CVTSD2SS_xx = xmminsn('\xF2', rex_nw, '\x0F\x5A',
+ register(1, 8), register(2), '\xC0')
+ CVTSD2SS_xb = xmminsn('\xF2', rex_nw, '\x0F\x5A',
+ register(1, 8), stack_bp(2))
+ CVTSS2SD_xx = xmminsn('\xF3', rex_nw, '\x0F\x5A',
+ register(1, 8), register(2), '\xC0')
+ CVTSS2SD_xb = xmminsn('\xF3', rex_nw, '\x0F\x5A',
+ register(1, 8), stack_bp(2))
+
+ MOVD_rx = xmminsn('\x66', rex_nw, '\x0F\x7E', register(2, 8), register(1), '\xC0')
+ MOVD_xr = xmminsn('\x66', rex_nw, '\x0F\x6E', register(1, 8), register(2), '\xC0')
+ MOVD_xb = xmminsn('\x66', rex_nw, '\x0F\x6E', register(1, 8), stack_bp(2))
PSRAD_xi = xmminsn('\x66', rex_nw, '\x0F\x72', register(1), '\xE0', immediate(2, 'b'))
diff --git a/pypy/jit/backend/x86/test/test_regloc.py b/pypy/jit/backend/x86/test/test_regloc.py
--- a/pypy/jit/backend/x86/test/test_regloc.py
+++ b/pypy/jit/backend/x86/test/test_regloc.py
@@ -62,7 +62,7 @@
assert mc.relocations == [5]
expected = "\xE8" + struct.pack('<i', target - (rawstart + 5))
elif IS_X86_64:
- assert mc.relocations == []
+ assert mc.relocations is None
if 0 <= target <= 0xffffffff:
assert length == 9
expected = (
diff --git a/pypy/jit/backend/x86/test/test_runner.py b/pypy/jit/backend/x86/test/test_runner.py
--- a/pypy/jit/backend/x86/test/test_runner.py
+++ b/pypy/jit/backend/x86/test/test_runner.py
@@ -463,7 +463,7 @@
self.cpu.finish_once()
finally:
debug._log = None
- assert ('jit-backend-counts', [('debug_print', '0:10')]) in dlog
+ assert ('jit-backend-counts', [('debug_print', 'loop -1:10')]) in dlog
def test_debugger_checksum(self):
loop = """
diff --git a/pypy/jit/backend/x86/test/test_rx86_32_auto_encoding.py b/pypy/jit/backend/x86/test/test_rx86_32_auto_encoding.py
--- a/pypy/jit/backend/x86/test/test_rx86_32_auto_encoding.py
+++ b/pypy/jit/backend/x86/test/test_rx86_32_auto_encoding.py
@@ -36,6 +36,14 @@
def hexdump(s):
return ' '.join(["%02X" % ord(c) for c in s])
+def reduce_to_32bit(s):
+ if s[:2] != '%r':
+ return s
+ if s[2:].isdigit():
+ return s + 'd'
+ else:
+ return '%e' + s[2:]
+
# ____________________________________________________________
COUNT1 = 15
@@ -180,12 +188,14 @@
## for m, extra in args:
## if m in (i386.MODRM, i386.MODRM8) or all:
## suffix = suffixes[sizes[m]] + suffix
- if argmodes and not self.is_xmm_insn:
+ if (argmodes and not self.is_xmm_insn
+ and not instrname.startswith('FSTP')):
suffix = suffixes[self.WORD]
# Special case: On 64-bit CPUs, rx86 assumes 64-bit integer
# operands when converting to/from floating point, so we need to
# indicate that with a suffix
- if (self.WORD == 8) and instrname.startswith('CVT'):
+ if (self.WORD == 8) and (instrname.startswith('CVT') and
+ 'SI' in instrname):
suffix = suffixes[self.WORD]
if instr_suffix is not None:
@@ -218,10 +228,10 @@
and ops[1].startswith('%r')):
# movq $xxx, %rax => movl $xxx, %eax
suffix = 'l'
- if ops[1][2:].isdigit():
- ops[1] += 'd'
- else:
- ops[1] = '%e' + ops[1][2:]
+ ops[1] = reduce_to_32bit(ops[1])
+ if instrname.lower() == 'movd':
+ ops[0] = reduce_to_32bit(ops[0])
+ ops[1] = reduce_to_32bit(ops[1])
#
op = '\t%s%s %s%s' % (instrname.lower(), suffix,
', '.join(ops), following)
diff --git a/pypy/jit/codewriter/assembler.py b/pypy/jit/codewriter/assembler.py
--- a/pypy/jit/codewriter/assembler.py
+++ b/pypy/jit/codewriter/assembler.py
@@ -76,6 +76,8 @@
TYPE = llmemory.Address
if TYPE == llmemory.Address:
value = heaptracker.adr2int(value)
+ if TYPE is lltype.SingleFloat:
+ value = longlong.singlefloat2int(value)
if not isinstance(value, (llmemory.AddressAsInt,
ComputedIntSymbolic)):
value = lltype.cast_primitive(lltype.Signed, value)
diff --git a/pypy/jit/codewriter/call.py b/pypy/jit/codewriter/call.py
--- a/pypy/jit/codewriter/call.py
+++ b/pypy/jit/codewriter/call.py
@@ -228,8 +228,10 @@
elif loopinvariant:
extraeffect = EffectInfo.EF_LOOPINVARIANT
elif elidable:
- # XXX check what to do about exceptions (also MemoryError?)
- extraeffect = EffectInfo.EF_ELIDABLE
+ if self._canraise(op):
+ extraeffect = EffectInfo.EF_ELIDABLE_CAN_RAISE
+ else:
+ extraeffect = EffectInfo.EF_ELIDABLE_CANNOT_RAISE
elif self._canraise(op):
extraeffect = EffectInfo.EF_CAN_RAISE
else:
@@ -263,7 +265,7 @@
def calldescr_canraise(self, calldescr):
effectinfo = calldescr.get_extra_info()
return (effectinfo is None or
- effectinfo.extraeffect >= EffectInfo.EF_CAN_RAISE)
+ effectinfo.extraeffect > EffectInfo.EF_CANNOT_RAISE)
def jitdriver_sd_from_portal_graph(self, graph):
for jd in self.jitdrivers_sd:
diff --git a/pypy/jit/codewriter/effectinfo.py b/pypy/jit/codewriter/effectinfo.py
--- a/pypy/jit/codewriter/effectinfo.py
+++ b/pypy/jit/codewriter/effectinfo.py
@@ -9,10 +9,11 @@
_cache = {}
# the 'extraeffect' field is one of the following values:
- EF_ELIDABLE = 0 #elidable function (and cannot raise)
+ EF_ELIDABLE_CANNOT_RAISE = 0 #elidable function (and cannot raise)
EF_LOOPINVARIANT = 1 #special: call it only once per loop
EF_CANNOT_RAISE = 2 #a function which cannot raise
- EF_CAN_RAISE = 3 #normal function (can raise)
+ EF_ELIDABLE_CAN_RAISE = 3 #elidable function (but can raise)
+ EF_CAN_RAISE = 4 #normal function (can raise)
EF_FORCES_VIRTUAL_OR_VIRTUALIZABLE = 5 #can raise and force virtualizables
# the 'oopspecindex' field is one of the following values:
@@ -94,7 +95,8 @@
result.readonly_descrs_fields = readonly_descrs_fields
result.readonly_descrs_arrays = readonly_descrs_arrays
if extraeffect == EffectInfo.EF_LOOPINVARIANT or \
- extraeffect == EffectInfo.EF_ELIDABLE:
+ extraeffect == EffectInfo.EF_ELIDABLE_CANNOT_RAISE or \
+ extraeffect == EffectInfo.EF_ELIDABLE_CAN_RAISE:
result.write_descrs_fields = []
result.write_descrs_arrays = []
else:
diff --git a/pypy/jit/codewriter/jitcode.py b/pypy/jit/codewriter/jitcode.py
--- a/pypy/jit/codewriter/jitcode.py
+++ b/pypy/jit/codewriter/jitcode.py
@@ -1,7 +1,6 @@
from pypy.jit.metainterp.history import AbstractDescr
from pypy.jit.codewriter import heaptracker
from pypy.rlib.objectmodel import we_are_translated
-from pypy.rpython.lltypesystem import llmemory
class JitCode(AbstractDescr):
@@ -102,7 +101,7 @@
def _clone_if_mutable(self):
raise NotImplementedError
-
+
class MissingLiveness(Exception):
pass
diff --git a/pypy/jit/codewriter/jtransform.py b/pypy/jit/codewriter/jtransform.py
--- a/pypy/jit/codewriter/jtransform.py
+++ b/pypy/jit/codewriter/jtransform.py
@@ -1,18 +1,16 @@
-import py, sys
-from pypy.rpython.lltypesystem import lltype, llmemory, rstr, rclass
-from pypy.rpython import rlist
-from pypy.jit.metainterp.history import getkind
-from pypy.objspace.flow.model import SpaceOperation, Variable, Constant
-from pypy.objspace.flow.model import Block, Link, c_last_exception
-from pypy.jit.codewriter.flatten import ListOfKind, IndirectCallTargets
+import py
from pypy.jit.codewriter import support, heaptracker, longlong
from pypy.jit.codewriter.effectinfo import EffectInfo
+from pypy.jit.codewriter.flatten import ListOfKind, IndirectCallTargets
from pypy.jit.codewriter.policy import log
+from pypy.jit.metainterp import quasiimmut
+from pypy.jit.metainterp.history import getkind
from pypy.jit.metainterp.typesystem import deref, arrayItem
-from pypy.jit.metainterp import quasiimmut
-from pypy.rpython.rclass import IR_QUASIIMMUTABLE, IR_QUASIIMMUTABLE_ARRAY
+from pypy.objspace.flow.model import SpaceOperation, Variable, Constant, c_last_exception
from pypy.rlib import objectmodel
from pypy.rlib.jit import _we_are_jitted
+from pypy.rpython.lltypesystem import lltype, llmemory, rstr, rclass, rffi
+from pypy.rpython.rclass import IR_QUASIIMMUTABLE, IR_QUASIIMMUTABLE_ARRAY
from pypy.translator.simplify import get_funcobj
from pypy.translator.unsimplify import varoftype
@@ -200,7 +198,6 @@
self.vable_array_vars[op.result]= self.vable_array_vars[op.args[0]]
rewrite_op_cast_pointer = rewrite_op_same_as
- rewrite_op_cast_opaque_ptr = rewrite_op_same_as # rlib.rerased
def rewrite_op_cast_bool_to_int(self, op): pass
def rewrite_op_cast_bool_to_uint(self, op): pass
def rewrite_op_cast_char_to_int(self, op): pass
@@ -787,7 +784,6 @@
op2.result = op.result
return op2
elif toll:
- from pypy.rpython.lltypesystem import rffi
size, unsigned = rffi.size_and_sign(op.args[0].concretetype)
if unsigned:
INTERMEDIATE = lltype.Unsigned
@@ -809,21 +805,27 @@
return self.force_cast_without_longlong(op.args[0], op.result)
def force_cast_without_longlong(self, v_arg, v_result):
- from pypy.rpython.lltypesystem.rffi import size_and_sign, sizeof, FLOAT
- from pypy.rlib.rarithmetic import intmask
- #
- if (v_result.concretetype in (FLOAT, lltype.Float) or
- v_arg.concretetype in (FLOAT, lltype.Float)):
- assert (v_result.concretetype == lltype.Float and
- v_arg.concretetype == lltype.Float), "xxx unsupported cast"
+ if v_result.concretetype == v_arg.concretetype:
return
- #
- size2, unsigned2 = size_and_sign(v_result.concretetype)
- assert size2 <= sizeof(lltype.Signed)
- if size2 == sizeof(lltype.Signed):
+ if v_arg.concretetype == rffi.FLOAT:
+ assert v_result.concretetype == lltype.Float, "cast %s -> %s" % (
+ v_arg.concretetype, v_result.concretetype)
+ return SpaceOperation('cast_singlefloat_to_float', [v_arg],
+ v_result)
+ if v_result.concretetype == rffi.FLOAT:
+ assert v_arg.concretetype == lltype.Float, "cast %s -> %s" % (
+ v_arg.concretetype, v_result.concretetype)
+ return SpaceOperation('cast_float_to_singlefloat', [v_arg],
+ v_result)
+ return self.force_cast_without_singlefloat(v_arg, v_result)
+
+ def force_cast_without_singlefloat(self, v_arg, v_result):
+ size2, unsigned2 = rffi.size_and_sign(v_result.concretetype)
+ assert size2 <= rffi.sizeof(lltype.Signed)
+ if size2 == rffi.sizeof(lltype.Signed):
return # the target type is LONG or ULONG
- size1, unsigned1 = size_and_sign(v_arg.concretetype)
- assert size1 <= sizeof(lltype.Signed)
+ size1, unsigned1 = rffi.size_and_sign(v_arg.concretetype)
+ assert size1 <= rffi.sizeof(lltype.Signed)
#
def bounds(size, unsigned):
if unsigned:
@@ -852,7 +854,6 @@
return result
def rewrite_op_direct_ptradd(self, op):
- from pypy.rpython.lltypesystem import rffi
# xxx otherwise, not implemented:
assert op.args[0].concretetype == rffi.CCHARP
#
@@ -905,7 +906,7 @@
op1 = self.prepare_builtin_call(op, "llong_%s", args)
op2 = self._handle_oopspec_call(op1, args,
EffectInfo.OS_LLONG_%s,
- EffectInfo.EF_ELIDABLE)
+ EffectInfo.EF_ELIDABLE_CANNOT_RAISE)
if %r == "TO_INT":
assert op2.result.concretetype == lltype.Signed
return op2
@@ -1366,15 +1367,15 @@
otherindex += EffectInfo._OS_offset_uni
self._register_extra_helper(otherindex, othername,
argtypes, resulttype,
- EffectInfo.EF_ELIDABLE)
+ EffectInfo.EF_ELIDABLE_CANNOT_RAISE)
#
return self._handle_oopspec_call(op, args, dict[oopspec_name],
- EffectInfo.EF_ELIDABLE)
+ EffectInfo.EF_ELIDABLE_CANNOT_RAISE)
def _handle_str2unicode_call(self, op, oopspec_name, args):
- # ll_str2unicode is not EF_ELIDABLE, because it can raise
- # UnicodeDecodeError...
- return self._handle_oopspec_call(op, args, EffectInfo.OS_STR2UNICODE)
+ # ll_str2unicode can raise UnicodeDecodeError
+ return self._handle_oopspec_call(op, args, EffectInfo.OS_STR2UNICODE,
+ EffectInfo.EF_ELIDABLE_CAN_RAISE)
# ----------
# VirtualRefs.
@@ -1412,13 +1413,13 @@
assert vinfo is not None
self.vable_flags[op.args[0]] = op.args[2].value
return []
-
+
# ---------
# ll_math.sqrt_nonneg()
-
+
def _handle_math_sqrt_call(self, op, oopspec_name, args):
return self._handle_oopspec_call(op, args, EffectInfo.OS_MATH_SQRT,
- EffectInfo.EF_ELIDABLE)
+ EffectInfo.EF_ELIDABLE_CANNOT_RAISE)
def rewrite_op_jit_force_quasi_immutable(self, op):
v_inst, c_fieldname = op.args
diff --git a/pypy/jit/codewriter/longlong.py b/pypy/jit/codewriter/longlong.py
--- a/pypy/jit/codewriter/longlong.py
+++ b/pypy/jit/codewriter/longlong.py
@@ -7,7 +7,8 @@
"""
import sys
-from pypy.rpython.lltypesystem import lltype
+from pypy.rpython.lltypesystem import lltype, rffi
+from pypy.rlib import rarithmetic, longlong2float
if sys.maxint > 2147483647:
@@ -31,8 +32,6 @@
# ---------- 32-bit platform ----------
# the type FloatStorage is r_longlong, and conversion is needed
- from pypy.rlib import rarithmetic, longlong2float
-
is_64_bit = False
supports_longlong = True
r_float_storage = rarithmetic.r_longlong
@@ -41,9 +40,19 @@
getfloatstorage = longlong2float.float2longlong
getrealfloat = longlong2float.longlong2float
gethash = lambda xll: rarithmetic.intmask(xll - (xll >> 32))
- is_longlong = lambda TYPE: (TYPE == lltype.SignedLongLong or
- TYPE == lltype.UnsignedLongLong)
+ is_longlong = lambda TYPE: (TYPE is lltype.SignedLongLong or
+ TYPE is lltype.UnsignedLongLong)
# -------------------------------------
ZEROF = getfloatstorage(0.0)
+
+# ____________________________________________________________
+
+def int2singlefloat(x):
+ x = rffi.r_uint(x)
+ return longlong2float.uint2singlefloat(x)
+
+def singlefloat2int(x):
+ x = longlong2float.singlefloat2uint(x)
+ return rffi.cast(lltype.Signed, x)
diff --git a/pypy/jit/codewriter/policy.py b/pypy/jit/codewriter/policy.py
--- a/pypy/jit/codewriter/policy.py
+++ b/pypy/jit/codewriter/policy.py
@@ -1,9 +1,7 @@
-from pypy.translator.simplify import get_funcobj
from pypy.jit.metainterp import history
-from pypy.rpython.lltypesystem import lltype, rclass
from pypy.tool.udir import udir
-import py, sys
+import py
from pypy.tool.ansi_print import ansi_log
log = py.log.Producer('jitcodewriter')
py.log.setconsumer('jitcodewriter', ansi_log)
@@ -14,6 +12,7 @@
self.unsafe_loopy_graphs = set()
self.supports_floats = False
self.supports_longlong = False
+ self.supports_singlefloats = False
def set_supports_floats(self, flag):
self.supports_floats = flag
@@ -21,6 +20,9 @@
def set_supports_longlong(self, flag):
self.supports_longlong = flag
+ def set_supports_singlefloats(self, flag):
+ self.supports_singlefloats = flag
+
def dump_unsafe_loops(self):
f = udir.join("unsafe-loops.txt").open('w')
strs = [str(graph) for graph in self.unsafe_loopy_graphs]
@@ -60,8 +62,9 @@
func, '_jit_unroll_safe_', False)
unsupported = contains_unsupported_variable_type(graph,
- self.supports_floats,
- self.supports_longlong)
+ self.supports_floats,
+ self.supports_longlong,
+ self.supports_singlefloats)
res = see_function and not unsupported
if res and contains_loop:
self.unsafe_loopy_graphs.add(graph)
@@ -82,17 +85,24 @@
return res
def contains_unsupported_variable_type(graph, supports_floats,
- supports_longlong):
+ supports_longlong,
+ supports_singlefloats):
getkind = history.getkind
try:
for block in graph.iterblocks():
for v in block.inputargs:
- getkind(v.concretetype, supports_floats, supports_longlong)
+ getkind(v.concretetype, supports_floats,
+ supports_longlong,
+ supports_singlefloats)
for op in block.operations:
for v in op.args:
- getkind(v.concretetype, supports_floats, supports_longlong)
+ getkind(v.concretetype, supports_floats,
+ supports_longlong,
+ supports_singlefloats)
v = op.result
- getkind(v.concretetype, supports_floats, supports_longlong)
+ getkind(v.concretetype, supports_floats,
+ supports_longlong,
+ supports_singlefloats)
except NotImplementedError, e:
log.WARNING('%s, ignoring graph' % (e,))
log.WARNING(' %s' % (graph,))
diff --git a/pypy/jit/codewriter/regalloc.py b/pypy/jit/codewriter/regalloc.py
--- a/pypy/jit/codewriter/regalloc.py
+++ b/pypy/jit/codewriter/regalloc.py
@@ -1,129 +1,8 @@
-import sys
-from pypy.objspace.flow.model import Variable
-from pypy.tool.algo.color import DependencyGraph
-from pypy.tool.algo.unionfind import UnionFind
+from pypy.tool.algo import regalloc
from pypy.jit.metainterp.history import getkind
from pypy.jit.codewriter.flatten import ListOfKind
+
def perform_register_allocation(graph, kind):
- """Perform register allocation for the Variables of the given 'kind'
- in the 'graph'."""
- regalloc = RegAllocator(graph, kind)
- regalloc.make_dependencies()
- regalloc.coalesce_variables()
- regalloc.find_node_coloring()
- return regalloc
-
-
-class RegAllocator(object):
- DEBUG_REGALLOC = False
-
- def __init__(self, graph, kind):
- self.graph = graph
- self.kind = kind
-
- def make_dependencies(self):
- dg = DependencyGraph()
- for block in self.graph.iterblocks():
- # Compute die_at = {Variable: index_of_operation_with_last_usage}
- die_at = dict.fromkeys(block.inputargs, 0)
- for i, op in enumerate(block.operations):
- for v in op.args:
- if isinstance(v, Variable):
- die_at[v] = i
- elif isinstance(v, ListOfKind):
- for v1 in v:
- if isinstance(v1, Variable):
- die_at[v1] = i
- if op.result is not None:
- die_at[op.result] = i + 1
- if isinstance(block.exitswitch, tuple):
- for x in block.exitswitch:
- die_at.pop(x, None)
- else:
- die_at.pop(block.exitswitch, None)
- for link in block.exits:
- for v in link.args:
- die_at.pop(v, None)
- die_at = [(value, key) for (key, value) in die_at.items()]
- die_at.sort()
- die_at.append((sys.maxint,))
- # Done. XXX the code above this line runs 3 times
- # (for kind in KINDS) to produce the same result...
- livevars = [v for v in block.inputargs
- if getkind(v.concretetype) == self.kind]
- # Add the variables of this block to the dependency graph
- for i, v in enumerate(livevars):
- dg.add_node(v)
- for j in range(i):
- dg.add_edge(livevars[j], v)
- livevars = set(livevars)
- die_index = 0
- for i, op in enumerate(block.operations):
- while die_at[die_index][0] == i:
- try:
- livevars.remove(die_at[die_index][1])
- except KeyError:
- pass
- die_index += 1
- if (op.result is not None and
- getkind(op.result.concretetype) == self.kind):
- dg.add_node(op.result)
- for v in livevars:
- if getkind(v.concretetype) == self.kind:
- dg.add_edge(v, op.result)
- livevars.add(op.result)
- self._depgraph = dg
-
- def coalesce_variables(self):
- self._unionfind = UnionFind()
- pendingblocks = list(self.graph.iterblocks())
- while pendingblocks:
- block = pendingblocks.pop()
- # Aggressively try to coalesce each source variable with its
- # target. We start from the end of the graph instead of
- # from the beginning. This is a bit arbitrary, but the idea
- # is that the end of the graph runs typically more often
- # than the start, given that we resume execution from the
- # middle during blackholing.
- for link in block.exits:
- if link.last_exception is not None:
- self._depgraph.add_node(link.last_exception)
- if link.last_exc_value is not None:
- self._depgraph.add_node(link.last_exc_value)
- for i, v in enumerate(link.args):
- self._try_coalesce(v, link.target.inputargs[i])
-
- def _try_coalesce(self, v, w):
- if isinstance(v, Variable) and getkind(v.concretetype) == self.kind:
- assert getkind(w.concretetype) == self.kind
- dg = self._depgraph
- uf = self._unionfind
- v0 = uf.find_rep(v)
- w0 = uf.find_rep(w)
- if v0 is not w0 and v0 not in dg.neighbours[w0]:
- _, rep, _ = uf.union(v0, w0)
- assert uf.find_rep(v0) is uf.find_rep(w0) is rep
- if rep is v0:
- dg.coalesce(w0, v0)
- else:
- assert rep is w0
- dg.coalesce(v0, w0)
-
- def find_node_coloring(self):
- self._coloring = self._depgraph.find_node_coloring()
- if self.DEBUG_REGALLOC:
- for block in self.graph.iterblocks():
- print block
- for v in block.getvariables():
- print '\t', v, '\t', self.getcolor(v)
-
- def getcolor(self, v):
- return self._coloring[self._unionfind.find_rep(v)]
-
- def swapcolors(self, col1, col2):
- for key, value in self._coloring.items():
- if value == col1:
- self._coloring[key] = col2
- elif value == col2:
- self._coloring[key] = col1
+ checkkind = lambda v: getkind(v.concretetype) == kind
+ return regalloc.perform_register_allocation(graph, checkkind, ListOfKind)
diff --git a/pypy/jit/codewriter/support.py b/pypy/jit/codewriter/support.py
--- a/pypy/jit/codewriter/support.py
+++ b/pypy/jit/codewriter/support.py
@@ -20,6 +20,7 @@
from pypy.rpython.annlowlevel import MixLevelHelperAnnotator
from pypy.jit.metainterp.typesystem import deref
from pypy.rlib import rgc
+from pypy.rlib.jit import elidable
from pypy.rlib.rarithmetic import r_longlong, r_ulonglong, r_uint, intmask
def getargtypes(annotator, values):
@@ -167,9 +168,14 @@
_ll_5_list_ll_arraycopy = rgc.ll_arraycopy
+ at elidable
def _ll_1_gc_identityhash(x):
return lltype.identityhash(x)
+# the following function should not be "@elidable": I can think of
+# a corner case in which id(const) is constant-folded, and then 'const'
+# disappears and is collected too early (possibly causing another object
+# with the same id() to appear).
def _ll_1_gc_id(ptr):
return llop.gc_id(lltype.Signed, ptr)
@@ -420,10 +426,6 @@
_ll_1_dict_values.need_result_type = True
_ll_1_dict_items .need_result_type = True
- def _ll_1_newdictiter(ITER, d):
- return ll_rdict.ll_dictiter(lltype.Ptr(ITER), d)
- _ll_1_newdictiter.need_result_type = True
-
_dictnext_keys = staticmethod(ll_rdict.ll_dictnext_group['keys'])
_dictnext_values = staticmethod(ll_rdict.ll_dictnext_group['values'])
_dictnext_items = staticmethod(ll_rdict.ll_dictnext_group['items'])
@@ -574,10 +576,6 @@
_ll_1_dict_values.need_result_type = True
_ll_1_dict_items .need_result_type = True
- def _ll_1_newdictiter(ITER, d):
- return oo_rdict.ll_dictiter(ITER, d)
- _ll_1_newdictiter.need_result_type = True
-
_dictnext_keys = staticmethod(oo_rdict.ll_dictnext_group['keys'])
_dictnext_values = staticmethod(oo_rdict.ll_dictnext_group['values'])
_dictnext_items = staticmethod(oo_rdict.ll_dictnext_group['items'])
diff --git a/pypy/jit/codewriter/test/test_jtransform.py b/pypy/jit/codewriter/test/test_jtransform.py
--- a/pypy/jit/codewriter/test/test_jtransform.py
+++ b/pypy/jit/codewriter/test/test_jtransform.py
@@ -120,9 +120,9 @@
assert argtypes[0] == [v.concretetype for v in op.args[1:]]
assert argtypes[1] == op.result.concretetype
if oopspecindex == EI.OS_STR2UNICODE:
- assert extraeffect == None # not pure, can raise!
+ assert extraeffect == EI.EF_ELIDABLE_CAN_RAISE
else:
- assert extraeffect == EI.EF_ELIDABLE
+ assert extraeffect == EI.EF_ELIDABLE_CANNOT_RAISE
return 'calldescr-%d' % oopspecindex
def calldescr_canraise(self, calldescr):
return False
@@ -769,7 +769,7 @@
def get_vinfo(self, v):
return None
def could_be_green_field(self, S1, name1):
- assert S1 is S
+ assert S1 == S
assert name1 == 'x'
return True
S = lltype.GcStruct('S', ('x', lltype.Char),
diff --git a/pypy/jit/codewriter/test/test_longlong.py b/pypy/jit/codewriter/test/test_longlong.py
--- a/pypy/jit/codewriter/test/test_longlong.py
+++ b/pypy/jit/codewriter/test/test_longlong.py
@@ -230,3 +230,18 @@
assert list(op1.args[3]) == []
assert list(op1.args[4]) == vlist
assert op1.result == v_result
+
+
+##def test_singlefloat_constants():
+## v_x = varoftype(TYPE)
+## vlist = [v_x, const(rffi.cast(TYPE, 7))]
+## v_result = varoftype(TYPE)
+## op = SpaceOperation('llong_add', vlist, v_result)
+## tr = Transformer(FakeCPU(), FakeBuiltinCallControl())
+## op1 = tr.rewrite_operation(op)
+## #
+## assert op1.opname == 'residual_call_irf_f'
+## assert list(op1.args[2]) == []
+## assert list(op1.args[3]) == []
+## assert list(op1.args[4]) == vlist
+## assert op1.result == v_result
diff --git a/pypy/jit/codewriter/test/test_policy.py b/pypy/jit/codewriter/test/test_policy.py
--- a/pypy/jit/codewriter/test/test_policy.py
+++ b/pypy/jit/codewriter/test/test_policy.py
@@ -12,24 +12,30 @@
graph = support.getgraph(f, [5])
for sf in [False, True]:
for sll in [False, True]:
- assert not contains_unsupported_variable_type(graph, sf, sll)
+ for ssf in [False, True]:
+ assert not contains_unsupported_variable_type(graph, sf,
+ sll, ssf)
#
graph = support.getgraph(f, [5.5])
for sf in [False, True]:
for sll in [False, True]:
- res = contains_unsupported_variable_type(graph, sf, sll)
- assert res is not sf
+ for ssf in [False, True]:
+ res = contains_unsupported_variable_type(graph, sf, sll, ssf)
+ assert res is not sf
#
graph = support.getgraph(f, [r_singlefloat(5.5)])
for sf in [False, True]:
for sll in [False, True]:
- assert contains_unsupported_variable_type(graph, sf, sll)
+ for ssf in [False, True]:
+ res = contains_unsupported_variable_type(graph, sf, sll, ssf)
+ assert res == (not ssf)
#
graph = support.getgraph(f, [r_longlong(5)])
for sf in [False, True]:
for sll in [False, True]:
- res = contains_unsupported_variable_type(graph, sf, sll)
- assert res == (sys.maxint == 2147483647 and not sll)
+ for ssf in [False, True]:
+ res = contains_unsupported_variable_type(graph, sf, sll, ssf)
+ assert res == (sys.maxint == 2147483647 and not sll)
def test_regular_function():
diff --git a/pypy/jit/metainterp/blackhole.py b/pypy/jit/metainterp/blackhole.py
--- a/pypy/jit/metainterp/blackhole.py
+++ b/pypy/jit/metainterp/blackhole.py
@@ -500,6 +500,9 @@
@arguments("r", returns="i")
def bhimpl_ptr_nonzero(a):
return bool(a)
+ @arguments("r", returns="r")
+ def bhimpl_cast_opaque_ptr(a):
+ return a
@arguments("i", returns="i")
def bhimpl_int_copy(a):
@@ -623,6 +626,19 @@
x = float(a)
return longlong.getfloatstorage(x)
+ @arguments("f", returns="i")
+ def bhimpl_cast_float_to_singlefloat(a):
+ from pypy.rlib.rarithmetic import r_singlefloat
+ a = longlong.getrealfloat(a)
+ a = r_singlefloat(a)
+ return longlong.singlefloat2int(a)
+
+ @arguments("i", returns="f")
+ def bhimpl_cast_singlefloat_to_float(a):
+ a = longlong.int2singlefloat(a)
+ a = float(a)
+ return longlong.getfloatstorage(a)
+
# ----------
# control flow operations
diff --git a/pypy/jit/metainterp/compile.py b/pypy/jit/metainterp/compile.py
--- a/pypy/jit/metainterp/compile.py
+++ b/pypy/jit/metainterp/compile.py
@@ -673,10 +673,9 @@
def handle_fail(self, metainterp_sd, jitdriver_sd):
cpu = metainterp_sd.cpu
exception = cpu.grab_exc_value()
+ assert exception, "PropagateExceptionDescr: no exception??"
raise metainterp_sd.ExitFrameWithExceptionRef(cpu, exception)
-propagate_exception_descr = PropagateExceptionDescr()
-
def compile_tmp_callback(cpu, jitdriver_sd, greenboxes, redboxes,
memory_manager=None):
"""Make a LoopToken that corresponds to assembler code that just
@@ -710,7 +709,7 @@
finishargs = []
#
jd = jitdriver_sd
- faildescr = propagate_exception_descr
+ faildescr = PropagateExceptionDescr()
operations = [
ResOperation(rop.CALL, callargs, result, descr=jd.portal_calldescr),
ResOperation(rop.GUARD_NO_EXCEPTION, [], None, descr=faildescr),
diff --git a/pypy/jit/metainterp/executor.py b/pypy/jit/metainterp/executor.py
--- a/pypy/jit/metainterp/executor.py
+++ b/pypy/jit/metainterp/executor.py
@@ -50,7 +50,7 @@
func = argboxes[0].getint()
# do the call using the correct function from the cpu
rettype = descr.get_return_type()
- if rettype == INT:
+ if rettype == INT or rettype == 'S': # *S*ingle float
try:
result = cpu.bh_call_i(func, descr, args_i, args_r, args_f)
except Exception, e:
@@ -64,7 +64,7 @@
metainterp.execute_raised(e)
result = NULL
return BoxPtr(result)
- if rettype == FLOAT or rettype == 'L':
+ if rettype == FLOAT or rettype == 'L': # *L*ong long
try:
result = cpu.bh_call_f(func, descr, args_i, args_r, args_f)
except Exception, e:
diff --git a/pypy/jit/metainterp/history.py b/pypy/jit/metainterp/history.py
--- a/pypy/jit/metainterp/history.py
+++ b/pypy/jit/metainterp/history.py
@@ -20,12 +20,16 @@
FAILARGS_LIMIT = 1000
-def getkind(TYPE, supports_floats=True, supports_longlong=True):
+def getkind(TYPE, supports_floats=True,
+ supports_longlong=True,
+ supports_singlefloats=True):
if TYPE is lltype.Void:
return "void"
elif isinstance(TYPE, lltype.Primitive):
if TYPE is lltype.Float and supports_floats:
return 'float'
+ if TYPE is lltype.SingleFloat and supports_singlefloats:
+ return 'int' # singlefloats are stored in an int
if TYPE in (lltype.Float, lltype.SingleFloat):
raise NotImplementedError("type %s not supported" % TYPE)
# XXX fix this for oo...
@@ -145,6 +149,7 @@
""" Implement in call descr.
Must return INT, REF, FLOAT, or 'v' for void.
On 32-bit (hack) it can also be 'L' for longlongs.
+ Additionally it can be 'S' for singlefloats.
"""
raise NotImplementedError
diff --git a/pypy/jit/metainterp/optimizeopt/__init__.py b/pypy/jit/metainterp/optimizeopt/__init__.py
--- a/pypy/jit/metainterp/optimizeopt/__init__.py
+++ b/pypy/jit/metainterp/optimizeopt/__init__.py
@@ -51,7 +51,7 @@
def optimize_loop_1(metainterp_sd, loop, enable_opts,
- inline_short_preamble=True, retraced=False):
+ inline_short_preamble=True, retraced=False, bridge=False):
"""Optimize loop.operations to remove internal overheadish operations.
"""
@@ -60,7 +60,7 @@
if unroll:
optimize_unroll(metainterp_sd, loop, optimizations)
else:
- optimizer = Optimizer(metainterp_sd, loop, optimizations)
+ optimizer = Optimizer(metainterp_sd, loop, optimizations, bridge)
optimizer.propagate_all_forward()
def optimize_bridge_1(metainterp_sd, bridge, enable_opts,
@@ -72,7 +72,7 @@
except KeyError:
pass
optimize_loop_1(metainterp_sd, bridge, enable_opts,
- inline_short_preamble, retraced)
+ inline_short_preamble, retraced, bridge=True)
if __name__ == '__main__':
print ALL_OPTS_NAMES
diff --git a/pypy/jit/metainterp/optimizeopt/fficall.py b/pypy/jit/metainterp/optimizeopt/fficall.py
--- a/pypy/jit/metainterp/optimizeopt/fficall.py
+++ b/pypy/jit/metainterp/optimizeopt/fficall.py
@@ -1,12 +1,11 @@
from pypy.rpython.annlowlevel import cast_base_ptr_to_instance
from pypy.rlib.objectmodel import we_are_translated
from pypy.rlib.libffi import Func
-from pypy.rlib.debug import debug_start, debug_stop, debug_print, have_debug_prints
+from pypy.rlib.debug import debug_print
from pypy.jit.codewriter.effectinfo import EffectInfo
from pypy.jit.metainterp.resoperation import rop, ResOperation
from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
from pypy.jit.metainterp.optimizeopt.optimizer import Optimization
-from pypy.jit.backend.llsupport.ffisupport import UnsupportedKind
class FuncInfo(object):
@@ -20,11 +19,8 @@
self.funcval = funcval
self.opargs = []
argtypes, restype = self._get_signature(funcval)
- try:
- self.descr = cpu.calldescrof_dynamic(argtypes, restype)
- except UnsupportedKind:
- # e.g., I or U for long longs
- self.descr = None
+ self.descr = cpu.calldescrof_dynamic(argtypes, restype)
+ # ^^^ may be None if unsupported
self.prepare_op = prepare_op
self.delayed_ops = []
@@ -48,7 +44,7 @@
inst_argtypes is actually a low-level array, but we can use it
directly since the only thing we do with it is to read its items
"""
-
+
llfunc = funcval.box.getref_base()
if we_are_translated():
func = cast_base_ptr_to_instance(Func, llfunc)
@@ -78,14 +74,6 @@
else:
self.logops = None
- def propagate_begin_forward(self):
- debug_start('jit-log-ffiopt')
- Optimization.propagate_begin_forward(self)
-
- def propagate_end_forward(self):
- debug_stop('jit-log-ffiopt')
- Optimization.propagate_end_forward(self)
-
def new(self):
return OptFfiCall()
@@ -188,7 +176,8 @@
def do_call(self, op):
funcval = self._get_funcval(op)
funcinfo = self.funcinfo
- if not funcinfo or funcinfo.funcval is not funcval:
+ if (not funcinfo or funcinfo.funcval is not funcval or
+ funcinfo.descr is None):
return [op] # cannot optimize
funcsymval = self.getvalue(op.getarg(2))
arglist = [funcsymval.force_box()]
diff --git a/pypy/jit/metainterp/optimizeopt/heap.py b/pypy/jit/metainterp/optimizeopt/heap.py
--- a/pypy/jit/metainterp/optimizeopt/heap.py
+++ b/pypy/jit/metainterp/optimizeopt/heap.py
@@ -1,10 +1,11 @@
import os
-from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
-from pypy.jit.metainterp.resoperation import rop, ResOperation
-from pypy.rlib.objectmodel import we_are_translated
+
from pypy.jit.metainterp.jitexc import JitException
from pypy.jit.metainterp.optimizeopt.optimizer import Optimization, MODE_ARRAY
from pypy.jit.metainterp.history import ConstInt, Const
+from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
+from pypy.jit.metainterp.resoperation import rop
+from pypy.rlib.objectmodel import we_are_translated
class CachedField(object):
@@ -284,13 +285,14 @@
return
cf.force_lazy_setfield(self, can_cache)
- def force_lazy_setarrayitem(self, arraydescr, can_cache=True):
+ def force_lazy_setarrayitem(self, arraydescr, indexvalue=None, can_cache=True):
try:
submap = self.cached_arrayitems[arraydescr]
except KeyError:
return
- for cf in submap.values():
- cf.force_lazy_setfield(self, can_cache)
+ for idx, cf in submap.iteritems():
+ if indexvalue is None or indexvalue.intbound.contains(idx):
+ cf.force_lazy_setfield(self, can_cache)
def fixup_guard_situation(self):
# hackish: reverse the order of the last two operations if it makes
@@ -403,7 +405,7 @@
return
else:
# variable index, so make sure the lazy setarrayitems are done
- self.force_lazy_setarrayitem(op.getdescr())
+ self.force_lazy_setarrayitem(op.getdescr(), indexvalue=indexvalue)
# default case: produce the operation
arrayvalue.ensure_nonnull()
self.emit_operation(op)
@@ -429,7 +431,7 @@
cf.do_setfield(self, op)
else:
# variable index, so make sure the lazy setarrayitems are done
- self.force_lazy_setarrayitem(op.getdescr(), can_cache=False)
+ self.force_lazy_setarrayitem(op.getdescr(), indexvalue=indexvalue, can_cache=False)
# and then emit the operation
self.emit_operation(op)
diff --git a/pypy/jit/metainterp/optimizeopt/intbounds.py b/pypy/jit/metainterp/optimizeopt/intbounds.py
--- a/pypy/jit/metainterp/optimizeopt/intbounds.py
+++ b/pypy/jit/metainterp/optimizeopt/intbounds.py
@@ -1,10 +1,11 @@
from pypy.jit.metainterp.optimizeopt.optimizer import Optimization, CONST_1, CONST_0, \
MODE_ARRAY, MODE_STR, MODE_UNICODE
+from pypy.jit.metainterp.history import ConstInt
+from pypy.jit.metainterp.optimizeopt.intutils import (IntBound, IntLowerBound,
+ IntUpperBound)
from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
-from pypy.jit.metainterp.optimizeopt.intutils import (IntBound, IntUnbounded,
- IntLowerBound, IntUpperBound)
-from pypy.jit.metainterp.history import Const, ConstInt
-from pypy.jit.metainterp.resoperation import rop, ResOperation
+from pypy.jit.metainterp.resoperation import rop
+
class OptIntBounds(Optimization):
"""Keeps track of the bounds placed on integers by guards and remove
@@ -129,6 +130,17 @@
r = self.getvalue(op.result)
r.intbound.intersect(v1.intbound.div_bound(v2.intbound))
+ def optimize_INT_MOD(self, op):
+ self.emit_operation(op)
+ v2 = self.getvalue(op.getarg(1))
+ if v2.is_constant():
+ val = v2.box.getint()
+ r = self.getvalue(op.result)
+ if val < 0:
+ val = -val
+ r.intbound.make_gt(IntBound(-val, -val))
+ r.intbound.make_lt(IntBound(val, val))
+
def optimize_INT_LSHIFT(self, op):
v1 = self.getvalue(op.getarg(0))
v2 = self.getvalue(op.getarg(1))
diff --git a/pypy/jit/metainterp/optimizeopt/optimizer.py b/pypy/jit/metainterp/optimizeopt/optimizer.py
--- a/pypy/jit/metainterp/optimizeopt/optimizer.py
+++ b/pypy/jit/metainterp/optimizeopt/optimizer.py
@@ -1,19 +1,13 @@
-from pypy.jit.metainterp.history import Box, BoxInt, LoopToken, BoxFloat,\
- ConstFloat
-from pypy.jit.metainterp.history import Const, ConstInt, ConstPtr, ConstObj, REF
-from pypy.jit.metainterp.resoperation import rop, ResOperation
-from pypy.jit.metainterp import jitprof
+from pypy.jit.metainterp import jitprof, resume, compile
from pypy.jit.metainterp.executor import execute_nonspec
-from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method, sort_descrs
-from pypy.jit.metainterp.optimizeopt.util import descrlist_dict, args_dict
-from pypy.jit.metainterp.optimize import InvalidLoop
-from pypy.jit.metainterp import resume, compile
-from pypy.jit.metainterp.typesystem import llhelper, oohelper
-from pypy.rpython.lltypesystem import lltype
-from pypy.jit.metainterp.history import AbstractDescr, make_hashable_int
+from pypy.jit.metainterp.history import BoxInt, BoxFloat, Const, ConstInt, REF
from pypy.jit.metainterp.optimizeopt.intutils import IntBound, IntUnbounded, \
ImmutableIntUnbounded, \
IntLowerBound, MININT, MAXINT
+from pypy.jit.metainterp.optimizeopt.util import (make_dispatcher_method,
+ args_dict)
+from pypy.jit.metainterp.resoperation import rop, ResOperation
+from pypy.jit.metainterp.typesystem import llhelper, oohelper
from pypy.tool.pairtype import extendabletype
from pypy.rlib.debug import debug_start, debug_stop, debug_print
@@ -337,10 +331,11 @@
class Optimizer(Optimization):
- def __init__(self, metainterp_sd, loop, optimizations=None):
+ def __init__(self, metainterp_sd, loop, optimizations=None, bridge=False):
self.metainterp_sd = metainterp_sd
self.cpu = metainterp_sd.cpu
self.loop = loop
+ self.bridge = bridge
self.values = {}
self.interned_refs = self.cpu.ts.new_ref_dict()
self.resumedata_memo = resume.ResumeDataLoopMemo(metainterp_sd)
@@ -352,6 +347,7 @@
self.posponedop = None
self.exception_might_have_happened = False
self.quasi_immutable_deps = None
+ self.opaque_pointers = {}
self.newoperations = []
self.emitting_dissabled = False
if loop is not None:
@@ -543,9 +539,7 @@
return CVAL_ZERO
def propagate_all_forward(self):
- self.exception_might_have_happened = True
- # ^^^ at least at the start of bridges. For loops, we could set
- # it to False, but we probably don't care
+ self.exception_might_have_happened = self.bridge
self.newoperations = []
self.first_optimization.propagate_begin_forward()
self.i = 0
@@ -692,6 +686,11 @@
def optimize_DEBUG_MERGE_POINT(self, op):
self.emit_operation(op)
+ def optimize_CAST_OPAQUE_PTR(self, op):
+ value = self.getvalue(op.getarg(0))
+ self.opaque_pointers[value] = True
+ self.make_equal_to(op.result, value)
+
def optimize_GETARRAYITEM_GC_PURE(self, op):
indexvalue = self.getvalue(op.getarg(1))
if indexvalue.is_constant():
diff --git a/pypy/jit/metainterp/optimizeopt/rewrite.py b/pypy/jit/metainterp/optimizeopt/rewrite.py
--- a/pypy/jit/metainterp/optimizeopt/rewrite.py
+++ b/pypy/jit/metainterp/optimizeopt/rewrite.py
@@ -1,10 +1,11 @@
+from pypy.jit.codewriter.effectinfo import EffectInfo
+from pypy.jit.metainterp.history import ConstInt, make_hashable_int
+from pypy.jit.metainterp.optimize import InvalidLoop
+from pypy.jit.metainterp.optimizeopt.intutils import IntBound
from pypy.jit.metainterp.optimizeopt.optimizer import *
-from pypy.jit.metainterp.resoperation import opboolinvers, opboolreflex
-from pypy.jit.metainterp.history import ConstInt
from pypy.jit.metainterp.optimizeopt.util import _findall, make_dispatcher_method
-from pypy.jit.metainterp.resoperation import rop, ResOperation
-from pypy.jit.codewriter.effectinfo import EffectInfo
-from pypy.jit.metainterp.optimizeopt.intutils import IntBound
+from pypy.jit.metainterp.resoperation import (opboolinvers, opboolreflex, rop,
+ ResOperation)
from pypy.rlib.rarithmetic import highest_bit
diff --git a/pypy/jit/metainterp/optimizeopt/simplify.py b/pypy/jit/metainterp/optimizeopt/simplify.py
--- a/pypy/jit/metainterp/optimizeopt/simplify.py
+++ b/pypy/jit/metainterp/optimizeopt/simplify.py
@@ -1,7 +1,7 @@
-
-from pypy.jit.metainterp.resoperation import ResOperation, rop
from pypy.jit.metainterp.optimizeopt.optimizer import Optimization
from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
+from pypy.jit.metainterp.resoperation import ResOperation, rop
+
class OptSimplify(Optimization):
def optimize_CALL_PURE(self, op):
@@ -25,6 +25,8 @@
# but it's a bit hard to implement robustly if heap.py is also run
pass
+ optimize_CAST_OPAQUE_PTR = optimize_VIRTUAL_REF
+
dispatch_opt = make_dispatcher_method(OptSimplify, 'optimize_',
default=OptSimplify.emit_operation)
diff --git a/pypy/jit/metainterp/optimizeopt/test/test_optimizebasic.py b/pypy/jit/metainterp/optimizeopt/test/test_optimizebasic.py
--- a/pypy/jit/metainterp/optimizeopt/test/test_optimizebasic.py
+++ b/pypy/jit/metainterp/optimizeopt/test/test_optimizebasic.py
@@ -693,7 +693,6 @@
"""
expected = """
[i]
- guard_no_exception() []
i1 = int_add(i, 3)
i2 = call(i1, descr=nonwritedescr)
guard_no_exception() [i1, i2]
@@ -4532,7 +4531,7 @@
escape(i1)
jump(p0, i0)
"""
- self.optimize_loop(ops, expected)
+ self.optimize_strunicode_loop(ops, expected)
def test_int_is_true_bounds(self):
ops = """
@@ -4551,7 +4550,7 @@
guard_true(i1) []
jump(p0)
"""
- self.optimize_loop(ops, expected)
+ self.optimize_strunicode_loop(ops, expected)
def test_strslice_subtraction_folds(self):
ops = """
@@ -4586,6 +4585,132 @@
"""
self.optimize_loop(ops, expected)
+ def test_null_char_str(self):
+ ops = """
+ [p0]
+ p1 = newstr(4)
+ setfield_gc(p0, p1, descr=valuedescr)
+ jump(p0)
+ """
+ # It used to be the case that this would have a series of
+ # strsetitem(p1, idx, 0), which was silly because memory is 0 filled
+ # when allocated.
+ expected = """
+ [p0]
+ p1 = newstr(4)
+ setfield_gc(p0, p1, descr=valuedescr)
+ jump(p0)
+ """
+ self.optimize_strunicode_loop(ops, expected)
+
+ def test_newstr_strlen(self):
+ ops = """
+ [i0]
+ p0 = newstr(i0)
+ escape(p0)
+ i1 = strlen(p0)
+ i2 = int_add(i1, 1)
+ jump(i2)
+ """
+ expected = """
+ [i0]
+ p0 = newstr(i0)
+ escape(p0)
+ i1 = int_add(i0, 1)
+ jump(i1)
+ """
+ self.optimize_strunicode_loop(ops, expected)
+
+ def test_intmod_bounds(self):
+ ops = """
+ [i0, i1]
+ i2 = int_mod(i0, 12)
+ i3 = int_gt(i2, 12)
+ guard_false(i3) []
+ i4 = int_lt(i2, -12)
+ guard_false(i4) []
+ i5 = int_mod(i1, -12)
+ i6 = int_lt(i5, -12)
+ guard_false(i6) []
+ i7 = int_gt(i5, 12)
+ guard_false(i7) []
+ jump(i2, i5)
+ """
+ expected = """
+ [i0, i1]
+ i2 = int_mod(i0, 12)
+ i5 = int_mod(i1, -12)
+ jump(i2, i5)
+ """
+ self.optimize_loop(ops, expected)
+
+ # This the sequence of resoperations that is generated for a Python
+ # app-level int % int. When the modulus is constant and when i0
+ # is known non-negative it should be optimized to a single int_mod.
+ ops = """
+ [i0]
+ i5 = int_ge(i0, 0)
+ guard_true(i5) []
+ i1 = int_mod(i0, 42)
+ i2 = int_rshift(i1, 63)
+ i3 = int_and(42, i2)
+ i4 = int_add(i1, i3)
+ finish(i4)
+ """
+ expected = """
+ [i0]
+ i5 = int_ge(i0, 0)
+ guard_true(i5) []
+ i1 = int_mod(i0, 42)
+ finish(i1)
+ """
+ py.test.skip("in-progress")
+ self.optimize_loop(ops, expected)
+
+ # Also, 'n % power-of-two' can be turned into int_and(),
+ # but that's a bit harder to detect here because it turns into
+ # several operations, and of course it is wrong to just turn
+ # int_mod(i0, 16) into int_and(i0, 15).
+ ops = """
+ [i0]
+ i1 = int_mod(i0, 16)
+ i2 = int_rshift(i1, 63)
+ i3 = int_and(16, i2)
+ i4 = int_add(i1, i3)
+ finish(i4)
+ """
+ expected = """
+ [i0]
+ i4 = int_and(i0, 15)
+ finish(i4)
+ """
+ py.test.skip("harder")
+ self.optimize_loop(ops, expected)
+
+ def test_bounded_lazy_setfield(self):
+ ops = """
+ [p0, i0]
+ i1 = int_gt(i0, 2)
+ guard_true(i1) []
+ setarrayitem_gc(p0, 0, 3)
+ setarrayitem_gc(p0, 2, 4)
+ setarrayitem_gc(p0, i0, 15)
+ i2 = getarrayitem_gc(p0, 2)
+ jump(p0, i2)
+ """
+ # Remove the getarrayitem_gc, because we know that p[i0] does not alias
+ # p0[2]
+ expected = """
+ [p0, i0]
+ i1 = int_gt(i0, 2)
+ guard_true(i1) []
+ setarrayitem_gc(p0, i0, 15)
+ setarrayitem_gc(p0, 0, 3)
+ setarrayitem_gc(p0, 2, 4)
+ jump(p0, 4)
+ """
+ self.optimize_loop(ops, expected)
+
class TestLLtype(BaseTestOptimizeBasic, LLtypeMixin):
pass
diff --git a/pypy/jit/metainterp/optimizeopt/test/test_optimizeopt.py b/pypy/jit/metainterp/optimizeopt/test/test_optimizeopt.py
--- a/pypy/jit/metainterp/optimizeopt/test/test_optimizeopt.py
+++ b/pypy/jit/metainterp/optimizeopt/test/test_optimizeopt.py
@@ -917,12 +917,10 @@
i3 = call(i2, descr=nonwritedescr)
jump(i1) # the exception is considered lost when we loop back
"""
- # note that 'guard_no_exception' at the very start must be kept
- # around: bridges may start with one. (In case of loops we could
- # remove it, but we probably don't care.)
+ # note that 'guard_no_exception' at the very start is kept around
+ # for bridges, but not for loops
preamble = """
[i]
- guard_no_exception() []
i1 = int_add(i, 3)
i2 = call(i1, descr=nonwritedescr)
guard_no_exception() [i1, i2]
@@ -3247,6 +3245,38 @@
'''
self.optimize_loop(ops, expected, preamble, call_pure_results)
+ def test_call_pure_constant_folding_exc(self):
+ # CALL_PURE may be followed by GUARD_NO_EXCEPTION
+ arg_consts = [ConstInt(i) for i in (123456, 4, 5, 6)]
+ call_pure_results = {tuple(arg_consts): ConstInt(42)}
+ ops = '''
+ [i0, i1, i2]
+ escape(i1)
+ escape(i2)
+ i3 = call_pure(123456, 4, 5, 6, descr=plaincalldescr)
+ guard_no_exception() []
+ i4 = call_pure(123456, 4, i0, 6, descr=plaincalldescr)
+ guard_no_exception() []
+ jump(i0, i3, i4)
+ '''
+ preamble = '''
+ [i0, i1, i2]
+ escape(i1)
+ escape(i2)
+ i4 = call(123456, 4, i0, 6, descr=plaincalldescr)
+ guard_no_exception() []
+ jump(i0, i4)
+ '''
+ expected = '''
+ [i0, i2]
+ escape(42)
+ escape(i2)
+ i4 = call(123456, 4, i0, 6, descr=plaincalldescr)
+ guard_no_exception() []
+ jump(i0, i4)
+ '''
+ self.optimize_loop(ops, expected, preamble, call_pure_results)
+
# ----------
def test_vref_nonvirtual_nonescape(self):
diff --git a/pypy/jit/metainterp/optimizeopt/unroll.py b/pypy/jit/metainterp/optimizeopt/unroll.py
--- a/pypy/jit/metainterp/optimizeopt/unroll.py
+++ b/pypy/jit/metainterp/optimizeopt/unroll.py
@@ -1,15 +1,14 @@
+from pypy.jit.codewriter.effectinfo import EffectInfo
+from pypy.jit.metainterp.optimizeopt.virtualstate import VirtualStateAdder, ShortBoxes
+from pypy.jit.metainterp.compile import ResumeGuardDescr
+from pypy.jit.metainterp.history import TreeLoop, LoopToken
+from pypy.jit.metainterp.jitexc import JitException
+from pypy.jit.metainterp.optimize import InvalidLoop, RetraceLoop
from pypy.jit.metainterp.optimizeopt.optimizer import *
-from pypy.jit.metainterp.optimizeopt.virtualstate import VirtualStateAdder, ShortBoxes
+from pypy.jit.metainterp.optimizeopt.generalize import KillHugeIntBounds
from pypy.jit.metainterp.resoperation import rop, ResOperation
-from pypy.jit.metainterp.compile import ResumeGuardDescr
from pypy.jit.metainterp.resume import Snapshot
-from pypy.jit.metainterp.history import TreeLoop, LoopToken
-from pypy.rlib.debug import debug_start, debug_stop, debug_print
-from pypy.jit.metainterp.optimize import InvalidLoop, RetraceLoop
-from pypy.jit.metainterp.jitexc import JitException
-from pypy.jit.metainterp.history import make_hashable_int
-from pypy.jit.codewriter.effectinfo import EffectInfo
-from pypy.jit.metainterp.optimizeopt.generalize import KillHugeIntBounds
+from pypy.rlib.debug import debug_print
# Assumptions
# ===========
@@ -23,7 +22,7 @@
# are also recreated to allow virtuals not supported to be forced.
#
# First of all, the optimizations are not allowed to introduce new
-# boxes. It is the unoptimized version of the trace that is inlined to
+# boxes. It is the unoptimized version of the trace that is inlined to
# form the second iteration of the loop. Otherwise the
# state of the virtuals would not be updated correctly. Whenever some
# box from the first iteration is reused in the second iteration, it
@@ -58,7 +57,7 @@
# be absorbed into the virtual p2 and never seen by the heap
# optimizer. At the end of the loop both p2 and p3 are virtuals, but
# the loop needs p2 to be a pointer to be able to call itself. So it
-# is forced producing the operations
+# is forced producing the operations
#
# p2 = new_with_vtable(ConstClass(node_vtable))
# setfield_gc(p2, i2, descr=nextdescr)
@@ -69,7 +68,7 @@
# the trace were optimized under the wrong assumption that the
# setfield_gc was store sinked which could lead to errors. In this
# case what would happen is that it would be inserted once more in
-# front of the guard.
+# front of the guard.
@@ -116,7 +115,7 @@
def inline_descr_inplace(self, descr):
if isinstance(descr, ResumeGuardDescr):
descr.rd_snapshot = self.inline_snapshot(descr.rd_snapshot)
-
+
def inline_arg(self, arg):
if arg is None:
return None
@@ -132,11 +131,12 @@
self.snapshot_map[snapshot] = new_snapshot
return new_snapshot
+
class UnrollOptimizer(Optimization):
"""Unroll the loop into two iterations. The first one will
become the preamble or entry bridge (don't think there is a
distinction anymore)"""
-
+
def __init__(self, metainterp_sd, loop, optimizations):
self.optimizer = Optimizer(metainterp_sd, loop, optimizations)
self.cloned_operations = []
@@ -329,7 +329,7 @@
for op in short_loop.operations:
if op.result:
op.result.forget_value()
-
+
def inline(self, inputargs, loop_operations, loop_args, short_inputargs, virtual_state):
inliner = self.inliner
@@ -546,7 +546,7 @@
"at preamble position: ", preamble_i,
"loop position: ", loop_i)
return None
-
+
if self.sameop(newop, loop_ops[loop_i]) \
and loop_i < len(loop_ops):
try:
@@ -567,7 +567,7 @@
"loop position: ", loop_i)
return None
short_preamble.append(op)
-
+
state.update(op)
preamble_i += 1
@@ -577,7 +577,7 @@
"at position", loop_i)
return None
-
+
jumpargs = []
for i in range(len(loop.inputargs)):
try:
@@ -605,7 +605,7 @@
return None
if op.result:
seen[op.result] = True
-
+
return short_preamble
class ExeState(object):
@@ -615,13 +615,16 @@
self.unsafe_getitem = {}
self.unsafe_getarrayitem = {}
self.unsafe_getarrayitem_indexes = {}
-
+
# Make sure it is safe to move the instrucions in short_preamble
# to the top making short_preamble followed by loop equvivalent
# to preamble
def safe_to_move(self, op):
opnum = op.getopnum()
descr = op.getdescr()
+ for box in op.getarglist():
+ if self.optimizer.getvalue(box) in self.optimizer.opaque_pointers:
+ return False
if op.is_always_pure() or op.is_foldable_guard():
return True
elif opnum == rop.JUMP:
@@ -652,15 +655,17 @@
elif opnum == rop.CALL:
effectinfo = descr.get_extra_info()
if effectinfo is not None:
- if effectinfo.extraeffect == EffectInfo.EF_LOOPINVARIANT or \
- effectinfo.extraeffect == EffectInfo.EF_ELIDABLE:
+ ef = effectinfo.extraeffect
+ if ef == EffectInfo.EF_LOOPINVARIANT or \
+ ef == EffectInfo.EF_ELIDABLE_CANNOT_RAISE or \
+ ef == EffectInfo.EF_ELIDABLE_CAN_RAISE:
return True
return False
-
+
def update(self, op):
if (op.has_no_side_effect() or
op.is_ovf() or
- op.is_guard()):
+ op.is_guard()):
return
opnum = op.getopnum()
descr = op.getdescr()
@@ -673,7 +678,7 @@
if (opnum == rop.SETARRAYITEM_GC or
opnum == rop.SETARRAYITEM_RAW):
index = op.getarg(1)
- if isinstance(index, Const):
+ if isinstance(index, Const):
d = self.unsafe_getarrayitem_indexes.get(descr, None)
if d is None:
d = self.unsafe_getarrayitem_indexes[descr] = {}
@@ -699,7 +704,7 @@
def __init__(self):
self.map = {}
-
+
def link_ops(self, preambleop, loopop):
pargs = preambleop.getarglist()
largs = loopop.getarglist()
@@ -713,7 +718,7 @@
if not loopop.result:
raise ImpossibleLink
self.link_boxes(preambleop.result, loopop.result)
-
+
def link_boxes(self, pbox, lbox):
if lbox in self.map:
@@ -736,11 +741,11 @@
def new(self):
return OptInlineShortPreamble(self.retraced)
-
+
def reconstruct_for_next_iteration(self, short_boxes, surviving_boxes,
optimizer, valuemap):
return OptInlineShortPreamble(self.retraced)
-
+
def propagate_forward(self, op):
if op.getopnum() == rop.JUMP:
loop_token = op.getdescr()
@@ -773,7 +778,7 @@
sh.virtual_state.generate_guards(virtual_state,
args, cpu,
extra_guards)
-
+
ok = True
debugmsg = 'Guarded to match '
except InvalidLoop:
@@ -821,6 +826,7 @@
else:
debug_print("Retracing (%d of %d)" % (retraced_count,
limit))
+
raise RetraceLoop
else:
if not loop_token.failed_states:
@@ -828,3 +834,7 @@
else:
loop_token.failed_states.append(virtual_state)
self.emit_operation(op)
+
+
+
+
diff --git a/pypy/jit/metainterp/optimizeopt/virtualize.py b/pypy/jit/metainterp/optimizeopt/virtualize.py
--- a/pypy/jit/metainterp/optimizeopt/virtualize.py
+++ b/pypy/jit/metainterp/optimizeopt/virtualize.py
@@ -1,12 +1,12 @@
+from pypy.jit.codewriter.heaptracker import vtable2descr
+from pypy.jit.metainterp.executor import execute
from pypy.jit.metainterp.history import Const, ConstInt, BoxInt
+from pypy.jit.metainterp.optimizeopt import optimizer
+from pypy.jit.metainterp.optimizeopt.util import (make_dispatcher_method,
+ descrlist_dict, sort_descrs)
from pypy.jit.metainterp.resoperation import rop, ResOperation
-from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
-from pypy.jit.metainterp.optimizeopt.util import descrlist_dict, sort_descrs
from pypy.rlib.objectmodel import we_are_translated
-from pypy.jit.metainterp.optimizeopt import optimizer
from pypy.jit.metainterp.optimizeopt.optimizer import OptValue
-from pypy.jit.metainterp.executor import execute
-from pypy.jit.codewriter.heaptracker import vtable2descr
class AbstractVirtualValue(optimizer.OptValue):
diff --git a/pypy/jit/metainterp/optimizeopt/vstring.py b/pypy/jit/metainterp/optimizeopt/vstring.py
--- a/pypy/jit/metainterp/optimizeopt/vstring.py
+++ b/pypy/jit/metainterp/optimizeopt/vstring.py
@@ -1,18 +1,14 @@
-from pypy.rpython.lltypesystem import lltype, rstr, llmemory
+from pypy.jit.codewriter.effectinfo import EffectInfo
+from pypy.jit.metainterp.history import (BoxInt, Const, ConstInt, ConstPtr,
+ get_const_ptr_for_string, get_const_ptr_for_unicode)
+from pypy.jit.metainterp.optimizeopt import optimizer, virtualize
+from pypy.jit.metainterp.optimizeopt.optimizer import CONST_0, CONST_1, llhelper
+from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
+from pypy.jit.metainterp.resoperation import rop, ResOperation
+from pypy.rlib.objectmodel import specialize, we_are_translated
+from pypy.rlib.unroll import unrolling_iterable
from pypy.rpython import annlowlevel
-from pypy.jit.metainterp.history import Box, BoxInt, BoxPtr
-from pypy.jit.metainterp.history import Const, ConstInt, ConstPtr
-from pypy.jit.metainterp.history import get_const_ptr_for_string
-from pypy.jit.metainterp.history import get_const_ptr_for_unicode
-from pypy.jit.metainterp.resoperation import rop, ResOperation
-from pypy.jit.metainterp.optimizeopt import optimizer, virtualize
-from pypy.jit.metainterp.optimizeopt.optimizer import CONST_0, CONST_1
-from pypy.jit.metainterp.optimizeopt.optimizer import llhelper
-from pypy.jit.metainterp.optimizeopt.util import make_dispatcher_method
-from pypy.jit.codewriter.effectinfo import EffectInfo
-from pypy.jit.codewriter import heaptracker
-from pypy.rlib.unroll import unrolling_iterable
-from pypy.rlib.objectmodel import specialize, we_are_translated
+from pypy.rpython.lltypesystem import lltype, rstr
class StrOrUnicode(object):
@@ -147,10 +143,11 @@
def string_copy_parts(self, optimizer, targetbox, offsetbox, mode):
for i in range(len(self._chars)):
charbox = self._chars[i].force_box()
- optimizer.emit_operation(ResOperation(mode.STRSETITEM, [targetbox,
- offsetbox,
- charbox],
- None))
+ if not (isinstance(charbox, Const) and charbox.same_constant(CONST_0)):
+ optimizer.emit_operation(ResOperation(mode.STRSETITEM, [targetbox,
+ offsetbox,
+ charbox],
+ None))
offsetbox = _int_add(optimizer, offsetbox, CONST_1)
return offsetbox
@@ -405,6 +402,7 @@
else:
self.getvalue(op.result).ensure_nonnull()
self.emit_operation(op)
+ self.pure(mode.STRLEN, [op.result], op.getarg(0))
def optimize_STRSETITEM(self, op):
value = self.getvalue(op.getarg(0))
diff --git a/pypy/jit/metainterp/pyjitpl.py b/pypy/jit/metainterp/pyjitpl.py
--- a/pypy/jit/metainterp/pyjitpl.py
+++ b/pypy/jit/metainterp/pyjitpl.py
@@ -215,6 +215,7 @@
for _opimpl in ['int_is_true', 'int_is_zero', 'int_neg', 'int_invert',
'cast_float_to_int', 'cast_int_to_float',
+ 'cast_float_to_singlefloat', 'cast_singlefloat_to_float',
'float_neg', 'float_abs',
]:
exec py.code.Source('''
@@ -232,6 +233,10 @@
return self.execute(rop.PTR_EQ, box, history.CONST_NULL)
@arguments("box")
+ def opimpl_cast_opaque_ptr(self, box):
+ return self.execute(rop.CAST_OPAQUE_PTR, box)
+
+ @arguments("box")
def _opimpl_any_return(self, box):
self.metainterp.finishframe(box)
@@ -1199,7 +1204,7 @@
return self.metainterp.execute_and_record(opnum, descr, *argboxes)
@specialize.arg(1)
- def execute_varargs(self, opnum, argboxes, descr, exc):
+ def execute_varargs(self, opnum, argboxes, descr, exc, pure):
self.metainterp.clear_exception()
resbox = self.metainterp.execute_and_record_varargs(opnum, argboxes,
descr=descr)
@@ -1207,6 +1212,9 @@
self.make_result_of_lastop(resbox)
# ^^^ this is done before handle_possible_exception() because we
# need the box to show up in get_list_of_active_boxes()
+ if pure and self.metainterp.last_exc_value_box is None:
+ resbox = self.metainterp.record_result_of_call_pure(resbox)
+ exc = exc and not isinstance(resbox, Const)
if exc:
self.metainterp.handle_possible_exception()
else:
@@ -1224,7 +1232,7 @@
src_i = src_r = src_f = 0
i = 1
for kind in descr.get_arg_types():
- if kind == history.INT:
+ if kind == history.INT or kind == 'S': # single float
while True:
box = argboxes[src_i]
src_i += 1
@@ -1269,16 +1277,14 @@
return resbox
else:
effect = effectinfo.extraeffect
- if effect == effectinfo.EF_CANNOT_RAISE:
- return self.execute_varargs(rop.CALL, allboxes, descr, False)
- elif effect == effectinfo.EF_ELIDABLE:
- return self.metainterp.record_result_of_call_pure(
- self.execute_varargs(rop.CALL, allboxes, descr, False))
- elif effect == effectinfo.EF_LOOPINVARIANT:
+ if effect == effectinfo.EF_LOOPINVARIANT:
return self.execute_varargs(rop.CALL_LOOPINVARIANT, allboxes,
- descr, False)
- else:
- return self.execute_varargs(rop.CALL, allboxes, descr, True)
+ descr, False, False)
+ exc = (effect != effectinfo.EF_CANNOT_RAISE and
+ effect != effectinfo.EF_ELIDABLE_CANNOT_RAISE)
+ pure = (effect == effectinfo.EF_ELIDABLE_CAN_RAISE or
+ effect == effectinfo.EF_ELIDABLE_CANNOT_RAISE)
+ return self.execute_varargs(rop.CALL, allboxes, descr, exc, pure)
def do_residual_or_indirect_call(self, funcbox, calldescr, argboxes):
"""The 'residual_call' operation is emitted in two cases:
@@ -1377,9 +1383,9 @@
num = self.cpu.get_fail_descr_number(tokens[0].finishdescr)
setattr(self.cpu, 'done_with_this_frame_%s_v' % name, num)
#
- tokens = self.loop_tokens_exit_frame_with_exception_ref
- num = self.cpu.get_fail_descr_number(tokens[0].finishdescr)
- self.cpu.exit_frame_with_exception_v = num
+ exc_descr = compile.PropagateExceptionDescr()
+ num = self.cpu.get_fail_descr_number(exc_descr)
+ self.cpu.propagate_exception_v = num
#
self.globaldata = MetaInterpGlobalData(self)
@@ -1686,8 +1692,12 @@
return
if opnum == rop.CALL:
effectinfo = descr.get_extra_info()
- if effectinfo.extraeffect == effectinfo.EF_ELIDABLE:
- return
+ if effectinfo is not None:
+ ef = effectinfo.extraeffect
+ if ef == effectinfo.EF_LOOPINVARIANT or \
+ ef == effectinfo.EF_ELIDABLE_CANNOT_RAISE or \
+ ef == effectinfo.EF_ELIDABLE_CAN_RAISE:
+ return
if self.heap_cache:
self.heap_cache.clear()
if self.heap_array_cache:
@@ -2375,6 +2385,7 @@
tobox = newbox
if change:
self.heap_cache[descr] = frombox, tobox
+ # XXX what about self.heap_array_cache?
def find_biggest_function(self):
start_stack = []
diff --git a/pypy/jit/metainterp/resoperation.py b/pypy/jit/metainterp/resoperation.py
--- a/pypy/jit/metainterp/resoperation.py
+++ b/pypy/jit/metainterp/resoperation.py
@@ -408,6 +408,8 @@
'FLOAT_ABS/1',
'CAST_FLOAT_TO_INT/1',
'CAST_INT_TO_FLOAT/1',
+ 'CAST_FLOAT_TO_SINGLEFLOAT/1',
+ 'CAST_SINGLEFLOAT_TO_FLOAT/1',
#
'INT_LT/2b',
'INT_LE/2b',
@@ -435,6 +437,7 @@
#
'PTR_EQ/2b',
'PTR_NE/2b',
+ 'CAST_OPAQUE_PTR/1b',
#
'ARRAYLEN_GC/1d',
'STRLEN/1',
diff --git a/pypy/jit/metainterp/test/support.py b/pypy/jit/metainterp/test/support.py
--- a/pypy/jit/metainterp/test/support.py
+++ b/pypy/jit/metainterp/test/support.py
@@ -277,3 +277,15 @@
NODE._add_fields({'value': ootype.Signed,
'next': NODE})
return NODE
+
+# ____________________________________________________________
+
+class _Foo:
+ pass
+
+def noConst(x):
+ """Helper function for tests, returning 'x' as a BoxInt/BoxPtr
+ even if it is a ConstInt/ConstPtr."""
+ f1 = _Foo(); f2 = _Foo()
+ f1.x = x; f2.x = 0
+ return f1.x
diff --git a/pypy/jit/metainterp/test/test_ajit.py b/pypy/jit/metainterp/test/test_ajit.py
--- a/pypy/jit/metainterp/test/test_ajit.py
+++ b/pypy/jit/metainterp/test/test_ajit.py
@@ -14,7 +14,7 @@
from pypy.rpython.lltypesystem import lltype, llmemory, rffi
from pypy.rpython.ootypesystem import ootype
from pypy.jit.metainterp.optimizeopt import ALL_OPTS_DICT
-from pypy.jit.metainterp.test.support import LLJitMixin, OOJitMixin
+from pypy.jit.metainterp.test.support import LLJitMixin, OOJitMixin, noConst
class BasicTests:
@@ -441,6 +441,58 @@
# the CALL_PURE is constant-folded away by optimizeopt.py
self.check_loops(int_sub=1, call=0, call_pure=0, getfield_gc=0)
+ def test_elidable_raising(self):
+ myjitdriver = JitDriver(greens = ['m'], reds = ['n'])
+ @elidable
+ def externfn(x):
+ if x <= 0:
+ raise ValueError
+ return x - 1
+ def f(n, m):
+ while n > 0:
+ myjitdriver.can_enter_jit(n=n, m=m)
+ myjitdriver.jit_merge_point(n=n, m=m)
+ try:
+ n -= externfn(m)
+ except ValueError:
+ n -= 1
+ return n
+ res = self.meta_interp(f, [22, 6])
+ assert res == -3
+ # the CALL_PURE is constant-folded away during tracing
+ self.check_loops(int_sub=1, call=0, call_pure=0)
+ #
+ res = self.meta_interp(f, [22, -5])
+ assert res == 0
+ # raises: becomes CALL and is not constant-folded away
+ self.check_loops(int_sub=1, call=1, call_pure=0)
+
+ def test_elidable_raising_2(self):
+ myjitdriver = JitDriver(greens = ['m'], reds = ['n'])
+ @elidable
+ def externfn(x):
+ if x <= 0:
+ raise ValueError
+ return x - 1
+ def f(n, m):
+ while n > 0:
+ myjitdriver.can_enter_jit(n=n, m=m)
+ myjitdriver.jit_merge_point(n=n, m=m)
+ try:
+ n -= externfn(noConst(m))
+ except ValueError:
+ n -= 1
+ return n
+ res = self.meta_interp(f, [22, 6])
+ assert res == -3
+ # the CALL_PURE is constant-folded away by optimizeopt.py
+ self.check_loops(int_sub=1, call=0, call_pure=0)
+ #
+ res = self.meta_interp(f, [22, -5])
+ assert res == 0
+ # raises: becomes CALL and is not constant-folded away
+ self.check_loops(int_sub=1, call=1, call_pure=0)
+
def test_constant_across_mp(self):
myjitdriver = JitDriver(greens = [], reds = ['n'])
class X(object):
diff --git a/pypy/jit/metainterp/test/test_dict.py b/pypy/jit/metainterp/test/test_dict.py
--- a/pypy/jit/metainterp/test/test_dict.py
+++ b/pypy/jit/metainterp/test/test_dict.py
@@ -157,7 +157,7 @@
# the same arguments are not folded, because we have conflicting
# definitions of pure, once strhash can be appropriately folded
# this should be decreased to seven.
- self.check_loops({"call": 8, "guard_false": 1, "guard_no_exception": 5,
+ self.check_loops({"call": 8, "guard_false": 1, "guard_no_exception": 6,
"guard_true": 1, "int_and": 1, "int_gt": 1,
"int_is_true": 1, "int_sub": 1, "jump": 1,
"new_with_vtable": 1, "setfield_gc": 1})
diff --git a/pypy/jit/metainterp/test/test_fficall.py b/pypy/jit/metainterp/test/test_fficall.py
--- a/pypy/jit/metainterp/test/test_fficall.py
+++ b/pypy/jit/metainterp/test/test_fficall.py
@@ -3,7 +3,7 @@
from pypy.rlib.rarithmetic import r_singlefloat, r_longlong, r_ulonglong
from pypy.rlib.jit import JitDriver, promote, dont_look_inside
from pypy.rlib.unroll import unrolling_iterable
-from pypy.rlib.libffi import ArgChain, longlong2float, float2longlong
+from pypy.rlib.libffi import ArgChain
from pypy.rlib.libffi import IS_32_BIT
from pypy.rlib.test.test_libffi import TestLibffiCall as _TestLibffiCall
from pypy.rpython.lltypesystem import lltype, rffi
@@ -12,10 +12,11 @@
from pypy.jit.metainterp.test.support import LLJitMixin
class TestFfiCall(LLJitMixin, _TestLibffiCall):
+ supports_all = False # supports_{floats,longlong,singlefloats}
# ===> ../../../rlib/test/test_libffi.py
- def call(self, funcspec, args, RESULT, init_result=0, is_struct=False):
+ def call(self, funcspec, args, RESULT, is_struct=False, jitif=[]):
"""
Call the function specified by funcspec in a loop, and let the jit to
see and optimize it.
@@ -24,14 +25,7 @@
lib, name, argtypes, restype = funcspec
method_and_args = []
for argval in args:
- if type(argval) is r_singlefloat:
- method_name = 'arg_singlefloat'
- argval = float(argval)
- elif IS_32_BIT and type(argval) in [r_longlong, r_ulonglong]:
- method_name = 'arg_longlong'
- argval = rffi.cast(rffi.LONGLONG, argval)
- argval = longlong2float(argval)
- elif isinstance(argval, tuple):
+ if isinstance(argval, tuple):
method_name, argval = argval
else:
method_name = 'arg'
@@ -39,10 +33,20 @@
method_and_args = unrolling_iterable(method_and_args)
#
reds = ['n', 'res', 'func']
- if (RESULT in [rffi.FLOAT, rffi.DOUBLE] or
+ if (RESULT is rffi.DOUBLE or
IS_32_BIT and RESULT in [rffi.LONGLONG, rffi.ULONGLONG]):
- reds = ['n', 'func', 'res'] # floats must be *after* refs
+ reds = ['n', 'func', 'res'] # 'double' floats must be *after* refs
driver = JitDriver(reds=reds, greens=[])
+ init_result = rffi.cast(RESULT, 0)
+ #
+ def g(func):
+ # a different function, which is marked as "dont_look_inside"
+ # in case it uses an unsupported argument
+ argchain = ArgChain()
+ # this loop is unrolled
+ for method_name, argval in method_and_args:
+ getattr(argchain, method_name)(argval)
+ return func.call(argchain, RESULT, is_struct=is_struct)
#
def f(n):
func = lib.getpointer(name, argtypes, restype)
@@ -50,18 +54,44 @@
while n < 10:
driver.jit_merge_point(n=n, res=res, func=func)
promote(func)
- argchain = ArgChain()
- # this loop is unrolled
- for method_name, argval in method_and_args:
- getattr(argchain, method_name)(argval)
- res = func.call(argchain, RESULT, is_struct=is_struct)
+ res = g(func)
n += 1
return res
#
- res = self.meta_interp(f, [0], backendopt=True)
+ res = self.meta_interp(f, [0], backendopt=True,
+ supports_floats = self.supports_all,
+ supports_longlong = self.supports_all,
+ supports_singlefloats = self.supports_all)
+ d = {'floats': self.supports_all,
+ 'longlong': self.supports_all or not IS_32_BIT,
+ 'singlefloats': self.supports_all,
+ 'byval': False}
+ supported = all(d[check] for check in jitif)
+ if supported:
+ self.check_loops(
+ call_release_gil=1, # a CALL_RELEASE_GIL, and no other CALLs
+ call=0,
+ call_may_force=0,
+ guard_no_exception=1,
+ guard_not_forced=1,
+ int_add=1,
+ int_lt=1,
+ guard_true=1,
+ jump=1)
+ else:
+ self.check_loops(
+ call_release_gil=0, # no CALL_RELEASE_GIL
+ int_add=1,
+ int_lt=1,
+ guard_true=1,
+ jump=1)
return res
def test_byval_result(self):
_TestLibffiCall.test_byval_result(self)
test_byval_result.__doc__ = _TestLibffiCall.test_byval_result.__doc__
test_byval_result.dont_track_allocations = True
+
+
+class TestFfiCallSupportAll(TestFfiCall):
+ supports_all = True # supports_{floats,longlong,singlefloats}
diff --git a/pypy/jit/metainterp/test/test_float.py b/pypy/jit/metainterp/test/test_float.py
--- a/pypy/jit/metainterp/test/test_float.py
+++ b/pypy/jit/metainterp/test/test_float.py
@@ -36,6 +36,15 @@
res = self.interp_operations(f, [x])
assert res == -x
+ def test_singlefloat(self):
+ from pypy.rlib.rarithmetic import r_singlefloat
+ def f(a):
+ a = float(r_singlefloat(a))
+ a *= 4.25
+ return float(r_singlefloat(a))
+ res = self.interp_operations(f, [-2.0])
+ assert res == -8.5
+
class TestOOtype(FloatTests, OOJitMixin):
pass
diff --git a/pypy/jit/metainterp/test/test_loop.py b/pypy/jit/metainterp/test/test_loop.py
--- a/pypy/jit/metainterp/test/test_loop.py
+++ b/pypy/jit/metainterp/test/test_loop.py
@@ -800,6 +800,37 @@
res = self.meta_interp(f, [200])
+
+ def test_unerased_pointers_in_short_preamble(self):
+ from pypy.rlib.rerased import new_erasing_pair
+ from pypy.rpython.lltypesystem import lltype
+ class A(object):
+ def __init__(self, val):
+ self.val = val
+ erase_A, unerase_A = new_erasing_pair('A')
+ erase_TP, unerase_TP = new_erasing_pair('TP')
+ TP = lltype.GcArray(lltype.Signed)
+ myjitdriver = JitDriver(greens = [], reds = ['n', 'm', 'i', 'j', 'sa', 'p'])
+ def f(n, m, j):
+ i = sa = 0
+ p = erase_A(A(7))
+ while i < n:
+ myjitdriver.jit_merge_point(n=n, m=m, i=i, j=j, sa=sa, p=p)
+ if i < m:
+ sa += unerase_A(p).val
+ elif i == m:
+ a = lltype.malloc(TP, 5)
+ a[0] = 42
+ p = erase_TP(a)
+ else:
+ sa += unerase_TP(p)[0]
+ sa += A(i).val
+ assert n>0 and m>0
+ i += j
+ return sa
+ res = self.meta_interp(f, [20, 10, 1])
+ assert res == f(20, 10, 1)
+
class TestOOtype(LoopTest, OOJitMixin):
pass
diff --git a/pypy/jit/metainterp/test/test_string.py b/pypy/jit/metainterp/test/test_string.py
--- a/pypy/jit/metainterp/test/test_string.py
+++ b/pypy/jit/metainterp/test/test_string.py
@@ -515,3 +515,22 @@
self.check_loops(call=3, # str(), _str(), escape()
newunicode=1, unicodegetitem=0,
unicodesetitem=1, copyunicodecontent=1)
+
+ def test_str2unicode_fold(self):
+ _str = self._str
+ jitdriver = JitDriver(greens = ['g'], reds = ['m'])
+ @dont_look_inside
+ def escape(x):
+ print str(x)
+ def f(g, m):
+ g = str(g)
+ while m >= 0:
+ jitdriver.can_enter_jit(g=g, m=m)
+ jitdriver.jit_merge_point(g=g, m=m)
+ escape(_str(g))
+ m -= 1
+ return 42
+ self.meta_interp(f, [6, 7])
+ self.check_loops(call_pure=0, call=1,
+ newunicode=0, unicodegetitem=0,
+ unicodesetitem=0, copyunicodecontent=0)
diff --git a/pypy/jit/metainterp/test/test_warmspot.py b/pypy/jit/metainterp/test/test_warmspot.py
--- a/pypy/jit/metainterp/test/test_warmspot.py
+++ b/pypy/jit/metainterp/test/test_warmspot.py
@@ -303,6 +303,7 @@
class FakeCPU(object):
supports_floats = False
supports_longlong = False
+ supports_singlefloats = False
ts = llhelper
translate_support_code = False
stats = "stats"
diff --git a/pypy/jit/metainterp/test/test_warmstate.py b/pypy/jit/metainterp/test/test_warmstate.py
--- a/pypy/jit/metainterp/test/test_warmstate.py
+++ b/pypy/jit/metainterp/test/test_warmstate.py
@@ -1,13 +1,14 @@
from pypy.rpython.test.test_llinterp import interpret
-from pypy.rpython.lltypesystem import lltype, llmemory, rstr
+from pypy.rpython.lltypesystem import lltype, llmemory, rstr, rffi
from pypy.rpython.ootypesystem import ootype
from pypy.rpython.annlowlevel import llhelper
-from pypy.jit.metainterp.warmstate import wrap, unwrap
+from pypy.jit.metainterp.warmstate import wrap, unwrap, specialize_value
from pypy.jit.metainterp.warmstate import equal_whatever, hash_whatever
from pypy.jit.metainterp.warmstate import WarmEnterState, JitCell
from pypy.jit.metainterp.history import BoxInt, BoxFloat, BoxPtr
from pypy.jit.metainterp.history import ConstInt, ConstFloat, ConstPtr
from pypy.jit.codewriter import longlong
+from pypy.rlib.rarithmetic import r_singlefloat
def boxfloat(x):
return BoxFloat(longlong.getfloatstorage(x))
@@ -40,6 +41,28 @@
assert _is(wrap(None, 42, in_const_box=True), ConstInt(42))
assert _is(wrap(None, 42.5, in_const_box=True), constfloat(42.5))
assert _is(wrap(None, p, in_const_box=True), ConstPtr(po))
+ if longlong.supports_longlong:
+ import sys
+ from pypy.rlib.rarithmetic import r_longlong, r_ulonglong
+ value = r_longlong(-sys.maxint*17)
+ assert _is(wrap(None, value), BoxFloat(value))
+ assert _is(wrap(None, value, in_const_box=True), ConstFloat(value))
+ value_unsigned = r_ulonglong(-sys.maxint*17)
+ assert _is(wrap(None, value_unsigned), BoxFloat(value))
+ sfval = r_singlefloat(42.5)
+ ival = longlong.singlefloat2int(sfval)
+ assert _is(wrap(None, sfval), BoxInt(ival))
+ assert _is(wrap(None, sfval, in_const_box=True), ConstInt(ival))
+
+def test_specialize_value():
+ assert specialize_value(lltype.Char, 0x41) == '\x41'
+ if longlong.supports_longlong:
+ import sys
+ value = longlong.r_float_storage(sys.maxint*17)
+ assert specialize_value(lltype.SignedLongLong, value) == sys.maxint*17
+ sfval = r_singlefloat(42.5)
+ ival = longlong.singlefloat2int(sfval)
+ assert specialize_value(rffi.FLOAT, ival) == sfval
def test_hash_equal_whatever_lltype():
s1 = rstr.mallocstr(2)
@@ -186,6 +209,7 @@
_get_printable_location_ptr = None
_confirm_enter_jit_ptr = None
_can_never_inline_ptr = None
+ _should_unroll_one_iteration_ptr = None
class FakeCell:
dont_trace_here = False
state = WarmEnterState(FakeWarmRunnerDesc(), FakeJitDriverSD())
@@ -214,6 +238,7 @@
_confirm_enter_jit_ptr = None
_can_never_inline_ptr = None
_get_jitcell_at_ptr = None
+ _should_unroll_one_iteration_ptr = None
state = WarmEnterState(FakeWarmRunnerDesc(), FakeJitDriverSD())
state.make_jitdriver_callbacks()
res = state.get_location_str([ConstInt(5), constfloat(42.5)])
@@ -238,6 +263,7 @@
_confirm_enter_jit_ptr = llhelper(ENTER_JIT, confirm_enter_jit)
_can_never_inline_ptr = None
_get_jitcell_at_ptr = None
+ _should_unroll_one_iteration_ptr = None
state = WarmEnterState(FakeWarmRunnerDesc(), FakeJitDriverSD())
state.make_jitdriver_callbacks()
@@ -262,6 +288,7 @@
_confirm_enter_jit_ptr = None
_can_never_inline_ptr = llhelper(CAN_NEVER_INLINE, can_never_inline)
_get_jitcell_at_ptr = None
+ _should_unroll_one_iteration_ptr = None
state = WarmEnterState(FakeWarmRunnerDesc(), FakeJitDriverSD())
state.make_jitdriver_callbacks()
diff --git a/pypy/jit/metainterp/warmspot.py b/pypy/jit/metainterp/warmspot.py
--- a/pypy/jit/metainterp/warmspot.py
+++ b/pypy/jit/metainterp/warmspot.py
@@ -173,6 +173,7 @@
policy = JitPolicy()
policy.set_supports_floats(self.cpu.supports_floats)
policy.set_supports_longlong(self.cpu.supports_longlong)
+ policy.set_supports_singlefloats(self.cpu.supports_singlefloats)
graphs = self.codewriter.find_all_graphs(policy)
policy.dump_unsafe_loops()
self.check_access_directly_sanity(graphs)
@@ -283,7 +284,9 @@
auto_inline_graphs(self.translator, graphs, 0.01)
def build_cpu(self, CPUClass, translate_support_code=False,
- no_stats=False, **kwds):
+ no_stats=False, supports_floats=True,
+ supports_longlong=True, supports_singlefloats=True,
+ **kwds):
assert CPUClass is not None
self.opt = history.Options(**kwds)
if no_stats:
@@ -295,6 +298,9 @@
self.annhelper = MixLevelHelperAnnotator(self.translator.rtyper)
cpu = CPUClass(self.translator.rtyper, self.stats, self.opt,
translate_support_code, gcdescr=self.gcdescr)
+ if not supports_floats: cpu.supports_floats = False
+ if not supports_longlong: cpu.supports_longlong = False
+ if not supports_singlefloats: cpu.supports_singlefloats = False
self.cpu = cpu
def build_meta_interp(self, ProfilerClass):
@@ -409,6 +415,7 @@
jd.warmstate = state
def crash_in_jit(e):
+ tb = not we_are_translated() and sys.exc_info()[2]
try:
raise e
except JitException:
@@ -422,8 +429,8 @@
print "~~~ Crash in JIT!"
print '~~~ %s: %s' % (e.__class__, e)
if sys.stdout == sys.__stdout__:
- import pdb; pdb.post_mortem(sys.exc_info()[2])
- raise
+ import pdb; pdb.post_mortem(tb)
+ raise e.__class__, e, tb
fatalerror('~~~ Crash in JIT! %s' % (e,), traceback=True)
crash_in_jit._dont_inline_ = True
diff --git a/pypy/jit/metainterp/warmstate.py b/pypy/jit/metainterp/warmstate.py
--- a/pypy/jit/metainterp/warmstate.py
+++ b/pypy/jit/metainterp/warmstate.py
@@ -25,9 +25,13 @@
if isinstance(TYPE, lltype.Ptr) and TYPE.TO._gckind == 'raw':
# non-gc pointer
return rffi.cast(TYPE, x)
+ elif TYPE is lltype.SingleFloat:
+ return longlong.int2singlefloat(x)
else:
return lltype.cast_primitive(TYPE, x)
elif INPUT is longlong.FLOATSTORAGE:
+ if longlong.is_longlong(TYPE):
+ return rffi.cast(TYPE, x)
assert TYPE is lltype.Float
return longlong.getrealfloat(x)
else:
@@ -84,8 +88,12 @@
return history.ConstObj(value)
else:
return history.BoxObj(value)
- elif isinstance(value, float):
- value = longlong.getfloatstorage(value)
+ elif (isinstance(value, float) or
+ longlong.is_longlong(lltype.typeOf(value))):
+ if isinstance(value, float):
+ value = longlong.getfloatstorage(value)
+ else:
+ value = rffi.cast(lltype.SignedLongLong, value)
if in_const_box:
return history.ConstFloat(value)
else:
@@ -93,6 +101,8 @@
elif isinstance(value, str) or isinstance(value, unicode):
assert len(value) == 1 # must be a character
value = ord(value)
+ elif lltype.typeOf(value) is lltype.SingleFloat:
+ value = longlong.singlefloat2int(value)
else:
value = intmask(value)
if in_const_box:
diff --git a/pypy/jit/tool/oparser.py b/pypy/jit/tool/oparser.py
--- a/pypy/jit/tool/oparser.py
+++ b/pypy/jit/tool/oparser.py
@@ -53,7 +53,7 @@
class OpParser(object):
use_mock_model = False
-
+
def __init__(self, input, cpu, namespace, type_system, boxkinds,
invent_fail_descr=default_fail_descr,
nonstrict=False):
@@ -187,7 +187,7 @@
poss_descr = allargs[-1].strip()
if poss_descr.startswith('descr='):
descr = self.get_descr(poss_descr[len('descr='):])
- allargs = allargs[:-1]
+ allargs = allargs[:-1]
for arg in allargs:
arg = arg.strip()
try:
@@ -240,7 +240,7 @@
fail_args = None
if opnum == rop.FINISH:
if descr is None and self.invent_fail_descr:
- descr = self.invent_fail_descr(self.model)
+ descr = self.invent_fail_descr(self.model, fail_args)
elif opnum == rop.JUMP:
if descr is None and self.invent_fail_descr:
descr = self.looptoken
diff --git a/pypy/module/_ffi/interp_ffi.py b/pypy/module/_ffi/interp_ffi.py
--- a/pypy/module/_ffi/interp_ffi.py
+++ b/pypy/module/_ffi/interp_ffi.py
@@ -11,6 +11,7 @@
from pypy.rlib import libffi
from pypy.rlib.rdynload import DLOpenError
from pypy.rlib.rarithmetic import intmask, r_uint
+from pypy.rlib.objectmodel import we_are_translated
class W_FFIType(Wrappable):
@@ -74,6 +75,13 @@
def is_struct(self):
return libffi.types.is_struct(self.ffitype)
+ def is_char_p(self):
+ return self is app_types.char_p
+
+ def is_unichar_p(self):
+ return self is app_types.unichar_p
+
+
W_FFIType.typedef = TypeDef(
'FFIType',
__repr__ = interp2app(W_FFIType.repr),
@@ -115,7 +123,12 @@
## 'Z' : ffi_type_pointer,
]
- return dict([(t.name, t) for t in types])
+ d = dict([(t.name, t) for t in types])
+ w_char = d['char']
+ w_unichar = d['unichar']
+ d['char_p'] = W_FFIType('char_p', libffi.types.pointer, w_pointer_to = w_char)
+ d['unichar_p'] = W_FFIType('unichar_p', libffi.types.pointer, w_pointer_to = w_unichar)
+ return d
class app_types:
pass
@@ -125,9 +138,14 @@
try:
return descr_new_pointer.cache[w_pointer_to]
except KeyError:
- w_pointer_to = space.interp_w(W_FFIType, w_pointer_to)
- name = '(pointer to %s)' % w_pointer_to.name
- w_result = W_FFIType(name, libffi.types.pointer, w_pointer_to = w_pointer_to)
+ if w_pointer_to is app_types.char:
+ w_result = app_types.char_p
+ elif w_pointer_to is app_types.unichar:
+ w_result = app_types.unichar_p
+ else:
+ w_pointer_to = space.interp_w(W_FFIType, w_pointer_to)
+ name = '(pointer to %s)' % w_pointer_to.name
+ w_result = W_FFIType(name, libffi.types.pointer, w_pointer_to = w_pointer_to)
descr_new_pointer.cache[w_pointer_to] = w_result
return w_result
descr_new_pointer.cache = {}
@@ -164,6 +182,7 @@
self.func = func
self.argtypes_w = argtypes_w
self.w_restype = w_restype
+ self.to_free = []
@jit.unroll_safe
def build_argchain(self, space, args_w):
@@ -188,6 +207,9 @@
self.arg_longlong(space, argchain, w_arg)
elif w_argtype.is_signed():
argchain.arg(unwrap_truncate_int(rffi.LONG, space, w_arg))
+ elif self.add_char_p_maybe(space, argchain, w_arg, w_argtype):
+ # the argument is added to the argchain direcly by the method above
+ pass
elif w_argtype.is_pointer():
w_arg = self.convert_pointer_arg_maybe(space, w_arg, w_argtype)
argchain.arg(intmask(space.uint_w(w_arg)))
@@ -200,9 +222,9 @@
w_arg = space.ord(w_arg)
argchain.arg(space.int_w(w_arg))
elif w_argtype.is_double():
- argchain.arg(space.float_w(w_arg))
+ self.arg_float(space, argchain, w_arg)
elif w_argtype.is_singlefloat():
- argchain.arg_singlefloat(space.float_w(w_arg))
+ self.arg_singlefloat(space, argchain, w_arg)
elif w_argtype.is_struct():
# arg_raw directly takes value to put inside ll_args
w_arg = space.interp_w(W_StructureInstance, w_arg)
@@ -212,6 +234,29 @@
assert False, "Argument shape '%s' not supported" % w_argtype
return argchain
+ def add_char_p_maybe(self, space, argchain, w_arg, w_argtype):
+ """
+ Automatic conversion from string to char_p. The allocated buffer will
+ be automatically freed after the call.
+ """
+ w_type = jit.promote(space.type(w_arg))
+ if w_argtype.is_char_p() and w_type is space.w_str:
+ strval = space.str_w(w_arg)
+ buf = rffi.str2charp(strval)
+ self.to_free.append(rffi.cast(rffi.VOIDP, buf))
+ addr = rffi.cast(rffi.ULONG, buf)
+ argchain.arg(addr)
+ return True
+ elif w_argtype.is_unichar_p() and (w_type is space.w_str or
+ w_type is space.w_unicode):
+ unicodeval = space.unicode_w(w_arg)
+ buf = rffi.unicode2wcharp(unicodeval)
+ self.to_free.append(rffi.cast(rffi.VOIDP, buf))
+ addr = rffi.cast(rffi.ULONG, buf)
+ argchain.arg(addr)
+ return True
+ return False
+
def convert_pointer_arg_maybe(self, space, w_arg, w_argtype):
"""
Try to convert the argument by calling _as_ffi_pointer_()
@@ -222,26 +267,47 @@
else:
return w_arg
- @jit.dont_look_inside
+ def arg_float(self, space, argchain, w_arg):
+ # a separate function, which can be seen by the jit or not,
+ # depending on whether floats are supported
+ argchain.arg(space.float_w(w_arg))
+
def arg_longlong(self, space, argchain, w_arg):
+ # a separate function, which can be seen by the jit or not,
+ # depending on whether longlongs are supported
bigarg = space.bigint_w(w_arg)
ullval = bigarg.ulonglongmask()
llval = rffi.cast(rffi.LONGLONG, ullval)
- # this is a hack: we store the 64 bits of the long long into the
- # 64 bits of a float (i.e., a C double)
- floatval = libffi.longlong2float(llval)
- argchain.arg_longlong(floatval)
+ argchain.arg(llval)
+
+ def arg_singlefloat(self, space, argchain, w_arg):
+ # a separate function, which can be seen by the jit or not,
+ # depending on whether singlefloats are supported
+ from pypy.rlib.rarithmetic import r_singlefloat
+ fval = space.float_w(w_arg)
+ sfval = r_singlefloat(fval)
+ argchain.arg(sfval)
def call(self, space, args_w):
self = jit.promote(self)
argchain = self.build_argchain(space, args_w)
+ return self._do_call(space, argchain)
+
+ def free_temp_buffers(self, space):
+ for buf in self.to_free:
+ if not we_are_translated():
+ buf[0] = '\00' # invalidate the buffer, so that
+ # test_keepalive_temp_buffer can fail
+ lltype.free(buf, flavor='raw')
+ self.to_free = []
+
+ def _do_call(self, space, argchain):
w_restype = self.w_restype
if w_restype.is_longlong():
# note that we must check for longlong first, because either
# is_signed or is_unsigned returns true anyway
assert libffi.IS_32_BIT
- reskind = libffi.types.getkind(self.func.restype) # XXX: remove the kind
- return self._call_longlong(space, argchain, reskind)
+ return self._call_longlong(space, argchain)
elif w_restype.is_signed():
return self._call_int(space, argchain)
elif w_restype.is_unsigned() or w_restype.is_pointer():
@@ -253,12 +319,9 @@
intres = self.func.call(argchain, rffi.WCHAR_T)
return space.wrap(unichr(intres))
elif w_restype.is_double():
- floatres = self.func.call(argchain, rffi.DOUBLE)
- return space.wrap(floatres)
+ return self._call_float(space, argchain)
elif w_restype.is_singlefloat():
- # the result is a float, but widened to be inside a double
- floatres = self.func.call(argchain, rffi.FLOAT)
- return space.wrap(floatres)
+ return self._call_singlefloat(space, argchain)
elif w_restype.is_struct():
w_datashape = w_restype.w_datashape
assert isinstance(w_datashape, W_Structure)
@@ -327,19 +390,32 @@
space.wrap('Unsupported restype'))
return space.wrap(intres)
- @jit.dont_look_inside
- def _call_longlong(self, space, argchain, reskind):
- # this is a hack: we store the 64 bits of the long long into the 64
- # bits of a float (i.e., a C double)
- floatres = self.func.call(argchain, rffi.LONGLONG)
- llres = libffi.float2longlong(floatres)
- if reskind == 'I':
+ def _call_float(self, space, argchain):
+ # a separate function, which can be seen by the jit or not,
+ # depending on whether floats are supported
+ floatres = self.func.call(argchain, rffi.DOUBLE)
+ return space.wrap(floatres)
+
+ def _call_longlong(self, space, argchain):
+ # a separate function, which can be seen by the jit or not,
+ # depending on whether longlongs are supported
+ restype = self.func.restype
+ call = self.func.call
+ if restype is libffi.types.slonglong:
+ llres = call(argchain, rffi.LONGLONG)
return space.wrap(llres)
- elif reskind == 'U':
- ullres = rffi.cast(rffi.ULONGLONG, llres)
+ elif restype is libffi.types.ulonglong:
+ ullres = call(argchain, rffi.ULONGLONG)
return space.wrap(ullres)
else:
- assert False
+ raise OperationError(space.w_ValueError,
+ space.wrap('Unsupported longlong restype'))
+
+ def _call_singlefloat(self, space, argchain):
+ # a separate function, which can be seen by the jit or not,
+ # depending on whether singlefloats are supported
+ sfres = self.func.call(argchain, rffi.FLOAT)
+ return space.wrap(float(sfres))
def getaddr(self, space):
"""
@@ -372,6 +448,7 @@
'_ffi.FuncPtr',
__call__ = interp2app(W_FuncPtr.call),
getaddr = interp2app(W_FuncPtr.getaddr),
+ free_temp_buffers = interp2app(W_FuncPtr.free_temp_buffers),
fromaddr = interp2app(descr_fromaddr, as_classmethod=True)
)
diff --git a/pypy/module/_ffi/test/test__ffi.py b/pypy/module/_ffi/test/test__ffi.py
--- a/pypy/module/_ffi/test/test__ffi.py
+++ b/pypy/module/_ffi/test/test__ffi.py
@@ -188,6 +188,75 @@
assert get_dummy() == 123
set_val_to_ptr(ptr2, 0)
+ def test_convert_strings_to_char_p(self):
+ """
+ long mystrlen(char* s)
+ {
+ long len = 0;
+ while(*s++)
+ len++;
+ return len;
+ }
+ """
+ from _ffi import CDLL, types
+ import _rawffi
+ libfoo = CDLL(self.libfoo_name)
+ mystrlen = libfoo.getfunc('mystrlen', [types.char_p], types.slong)
+ #
+ # first, try automatic conversion from a string
+ assert mystrlen('foobar') == 6
+ # then, try to pass an explicit pointer
+ CharArray = _rawffi.Array('c')
+ mystr = CharArray(7, 'foobar')
+ assert mystrlen(mystr.buffer) == 6
+ mystr.free()
+ mystrlen.free_temp_buffers()
+
+ def test_convert_unicode_to_unichar_p(self):
+ """
+ #include <wchar.h>
+ long mystrlen_u(wchar_t* s)
+ {
+ long len = 0;
+ while(*s++)
+ len++;
+ return len;
+ }
+ """
+ from _ffi import CDLL, types
+ import _rawffi
+ libfoo = CDLL(self.libfoo_name)
+ mystrlen = libfoo.getfunc('mystrlen_u', [types.unichar_p], types.slong)
+ #
+ # first, try automatic conversion from strings and unicode
+ assert mystrlen('foobar') == 6
+ assert mystrlen(u'foobar') == 6
+ assert mystrlen(u'ab\u2070') == 3
+ # then, try to pass an explicit pointer
+ UniCharArray = _rawffi.Array('u')
+ mystr = UniCharArray(7, u'foobar')
+ assert mystrlen(mystr.buffer) == 6
+ mystr.free()
+ mystrlen.free_temp_buffers()
+
+ def test_keepalive_temp_buffer(self):
+ """
+ char* do_nothing(char* s)
+ {
+ return s;
+ }
+ """
+ from _ffi import CDLL, types
+ import _rawffi
+ libfoo = CDLL(self.libfoo_name)
+ do_nothing = libfoo.getfunc('do_nothing', [types.char_p], types.char_p)
+ CharArray = _rawffi.Array('c')
+ #
+ ptr = do_nothing('foobar')
+ array = CharArray.fromaddress(ptr, 7)
+ assert list(array) == list('foobar\00')
+ do_nothing.free_temp_buffers()
+
def test_typed_pointer(self):
from _ffi import types
intptr = types.Pointer(types.sint) # create a typed pointer to sint
@@ -204,6 +273,13 @@
assert x is y
assert x is not z
+ def test_char_p_cached(self):
+ from _ffi import types
+ x = types.Pointer(types.char)
+ assert x is types.char_p
+ x = types.Pointer(types.unichar)
+ assert x is types.unichar_p
+
def test_typed_pointer_args(self):
"""
extern int dummy; // defined in test_void_result
diff --git a/pypy/module/_multibytecodec/__init__.py b/pypy/module/_multibytecodec/__init__.py
--- a/pypy/module/_multibytecodec/__init__.py
+++ b/pypy/module/_multibytecodec/__init__.py
@@ -7,13 +7,14 @@
# for compatibility this name is obscured, and should be called
# via the _codecs_*.py modules written in lib_pypy.
'__getcodec': 'interp_multibytecodec.getcodec',
+
+ 'MultibyteIncrementalDecoder':
+ 'interp_incremental.MultibyteIncrementalDecoder',
+ 'MultibyteIncrementalEncoder':
+ 'interp_incremental.MultibyteIncrementalEncoder',
}
appleveldefs = {
- 'MultibyteIncrementalEncoder':
- 'app_multibytecodec.MultibyteIncrementalEncoder',
- 'MultibyteIncrementalDecoder':
- 'app_multibytecodec.MultibyteIncrementalDecoder',
'MultibyteStreamReader':
'app_multibytecodec.MultibyteStreamReader',
'MultibyteStreamWriter':
diff --git a/pypy/module/_multibytecodec/app_multibytecodec.py b/pypy/module/_multibytecodec/app_multibytecodec.py
--- a/pypy/module/_multibytecodec/app_multibytecodec.py
+++ b/pypy/module/_multibytecodec/app_multibytecodec.py
@@ -1,34 +1,47 @@
# NOT_RPYTHON
#
-# These classes are not supported so far.
-#
-# My theory is that they are not widely used on CPython either, because
-# I found two bugs just by looking at their .c source: they always call
-# encreset() after a piece of data, even though I think it's wrong ---
-# it should be called only once at the end; and mbiencoder_reset() calls
-# decreset() instead of encreset().
-#
+# The interface here may be a little bit on the lightweight side.
-class MultibyteIncrementalEncoder(object):
- def __init__(self, *args, **kwds):
- raise LookupError(
- "MultibyteIncrementalEncoder not implemented; "
- "see pypy/module/_multibytecodec/app_multibytecodec.py")
+from _multibytecodec import MultibyteIncrementalDecoder
+from _multibytecodec import MultibyteIncrementalEncoder
-class MultibyteIncrementalDecoder(object):
- def __init__(self, *args, **kwds):
- raise LookupError(
- "MultibyteIncrementalDecoder not implemented; "
- "see pypy/module/_multibytecodec/app_multibytecodec.py")
-class MultibyteStreamReader(object):
- def __init__(self, *args, **kwds):
- raise LookupError(
- "MultibyteStreamReader not implemented; "
- "see pypy/module/_multibytecodec/app_multibytecodec.py")
+class MultibyteStreamReader(MultibyteIncrementalDecoder):
+ def __new__(cls, stream, errors=None):
+ self = MultibyteIncrementalDecoder.__new__(cls, errors)
+ self.stream = stream
+ return self
-class MultibyteStreamWriter(object):
- def __init__(self, *args, **kwds):
- raise LookupError(
- "MultibyteStreamWriter not implemented; "
- "see pypy/module/_multibytecodec/app_multibytecodec.py")
+ def __read(self, read, size):
+ if size is None or size < 0:
+ return MultibyteIncrementalDecoder.decode(self, read(), True)
+ while True:
+ data = read(size)
+ final = not data
+ output = MultibyteIncrementalDecoder.decode(self, data, final)
+ if output or final:
+ return output
+ size = 1 # read 1 more byte and retry
+
+ def read(self, size=None):
+ return self.__read(self.stream.read, size)
+
+ def readline(self, size=None):
+ return self.__read(self.stream.readline, size)
+
+ def readlines(self, sizehint=None):
+ return self.__read(self.stream.read, sizehint).splitlines(True)
+
+
+class MultibyteStreamWriter(MultibyteIncrementalEncoder):
+ def __new__(cls, stream, errors=None):
+ self = MultibyteIncrementalEncoder.__new__(cls, errors)
+ self.stream = stream
+ return self
+
+ def write(self, data):
+ self.stream.write(MultibyteIncrementalEncoder.encode(self, data))
+
+ def writelines(self, lines):
+ for data in lines:
+ self.write(data)
diff --git a/pypy/module/_multibytecodec/c_codecs.py b/pypy/module/_multibytecodec/c_codecs.py
--- a/pypy/module/_multibytecodec/c_codecs.py
+++ b/pypy/module/_multibytecodec/c_codecs.py
@@ -52,11 +52,13 @@
includes = ['src/cjkcodecs/multibytecodec.h'],
include_dirs = [str(srcdir)],
export_symbols = [
+ "pypy_cjk_dec_new",
"pypy_cjk_dec_init", "pypy_cjk_dec_free", "pypy_cjk_dec_chunk",
"pypy_cjk_dec_outbuf", "pypy_cjk_dec_outlen",
"pypy_cjk_dec_inbuf_remaining", "pypy_cjk_dec_inbuf_consumed",
"pypy_cjk_dec_replace_on_error",
+ "pypy_cjk_enc_new",
"pypy_cjk_enc_init", "pypy_cjk_enc_free", "pypy_cjk_enc_chunk",
"pypy_cjk_enc_reset", "pypy_cjk_enc_outbuf", "pypy_cjk_enc_outlen",
"pypy_cjk_enc_inbuf_remaining", "pypy_cjk_enc_inbuf_consumed",
@@ -92,9 +94,11 @@
# Decoding
DECODEBUF_P = rffi.COpaquePtr('struct pypy_cjk_dec_s', compilation_info=eci)
+pypy_cjk_dec_new = llexternal('pypy_cjk_dec_new',
+ [MULTIBYTECODEC_P], DECODEBUF_P)
pypy_cjk_dec_init = llexternal('pypy_cjk_dec_init',
- [MULTIBYTECODEC_P, rffi.CCHARP, rffi.SSIZE_T],
- DECODEBUF_P)
+ [DECODEBUF_P, rffi.CCHARP, rffi.SSIZE_T],
+ rffi.SSIZE_T)
pypy_cjk_dec_free = llexternal('pypy_cjk_dec_free', [DECODEBUF_P],
lltype.Void)
pypy_cjk_dec_chunk = llexternal('pypy_cjk_dec_chunk', [DECODEBUF_P],
@@ -113,25 +117,30 @@
rffi.SSIZE_T)
def decode(codec, stringdata, errors="strict", errorcb=None, namecb=None):
+ decodebuf = pypy_cjk_dec_new(codec)
+ if not decodebuf:
+ raise MemoryError
+ try:
+ return decodeex(decodebuf, stringdata, errors, errorcb, namecb)
+ finally:
+ pypy_cjk_dec_free(decodebuf)
+
+def decodeex(decodebuf, stringdata, errors="strict", errorcb=None, namecb=None,
+ ignore_error=0):
inleft = len(stringdata)
inbuf = rffi.get_nonmovingbuffer(stringdata)
try:
- decodebuf = pypy_cjk_dec_init(codec, inbuf, inleft)
- if not decodebuf:
+ if pypy_cjk_dec_init(decodebuf, inbuf, inleft) < 0:
raise MemoryError
- try:
- while True:
- r = pypy_cjk_dec_chunk(decodebuf)
- if r == 0:
- break
- multibytecodec_decerror(decodebuf, r, errors,
- errorcb, namecb, stringdata)
- src = pypy_cjk_dec_outbuf(decodebuf)
- length = pypy_cjk_dec_outlen(decodebuf)
- return rffi.wcharpsize2unicode(src, length)
- #
- finally:
- pypy_cjk_dec_free(decodebuf)
+ while True:
+ r = pypy_cjk_dec_chunk(decodebuf)
+ if r == 0 or r == ignore_error:
+ break
+ multibytecodec_decerror(decodebuf, r, errors,
+ errorcb, namecb, stringdata)
+ src = pypy_cjk_dec_outbuf(decodebuf)
+ length = pypy_cjk_dec_outlen(decodebuf)
+ return rffi.wcharpsize2unicode(src, length)
#
finally:
rffi.free_nonmovingbuffer(stringdata, inbuf)
@@ -174,13 +183,15 @@
# ____________________________________________________________
# Encoding
ENCODEBUF_P = rffi.COpaquePtr('struct pypy_cjk_enc_s', compilation_info=eci)
+pypy_cjk_enc_new = llexternal('pypy_cjk_enc_new',
+ [MULTIBYTECODEC_P], ENCODEBUF_P)
pypy_cjk_enc_init = llexternal('pypy_cjk_enc_init',
- [MULTIBYTECODEC_P, rffi.CWCHARP, rffi.SSIZE_T],
- ENCODEBUF_P)
+ [ENCODEBUF_P, rffi.CWCHARP, rffi.SSIZE_T],
+ rffi.SSIZE_T)
pypy_cjk_enc_free = llexternal('pypy_cjk_enc_free', [ENCODEBUF_P],
lltype.Void)
-pypy_cjk_enc_chunk = llexternal('pypy_cjk_enc_chunk', [ENCODEBUF_P],
- rffi.SSIZE_T)
+pypy_cjk_enc_chunk = llexternal('pypy_cjk_enc_chunk',
+ [ENCODEBUF_P, rffi.SSIZE_T], rffi.SSIZE_T)
pypy_cjk_enc_reset = llexternal('pypy_cjk_enc_reset', [ENCODEBUF_P],
rffi.SSIZE_T)
pypy_cjk_enc_outbuf = llexternal('pypy_cjk_enc_outbuf', [ENCODEBUF_P],
@@ -195,39 +206,52 @@
[ENCODEBUF_P, rffi.CCHARP,
rffi.SSIZE_T, rffi.SSIZE_T],
rffi.SSIZE_T)
+pypy_cjk_enc_getcodec = llexternal('pypy_cjk_enc_getcodec',
+ [ENCODEBUF_P], MULTIBYTECODEC_P)
+MBENC_FLUSH = 1
+MBENC_RESET = 2
def encode(codec, unicodedata, errors="strict", errorcb=None, namecb=None):
+ encodebuf = pypy_cjk_enc_new(codec)
+ if not encodebuf:
+ raise MemoryError
+ try:
+ return encodeex(encodebuf, unicodedata, errors, errorcb, namecb)
+ finally:
+ pypy_cjk_enc_free(encodebuf)
+
+def encodeex(encodebuf, unicodedata, errors="strict", errorcb=None,
+ namecb=None, ignore_error=0):
inleft = len(unicodedata)
inbuf = rffi.get_nonmoving_unicodebuffer(unicodedata)
try:
- encodebuf = pypy_cjk_enc_init(codec, inbuf, inleft)
- if not encodebuf:
+ if pypy_cjk_enc_init(encodebuf, inbuf, inleft) < 0:
raise MemoryError
- try:
- while True:
- r = pypy_cjk_enc_chunk(encodebuf)
- if r == 0:
- break
- multibytecodec_encerror(encodebuf, r, errors,
- codec, errorcb, namecb, unicodedata)
- while True:
- r = pypy_cjk_enc_reset(encodebuf)
- if r == 0:
- break
- multibytecodec_encerror(encodebuf, r, errors,
- codec, errorcb, namecb, unicodedata)
- src = pypy_cjk_enc_outbuf(encodebuf)
- length = pypy_cjk_enc_outlen(encodebuf)
- return rffi.charpsize2str(src, length)
- #
- finally:
- pypy_cjk_enc_free(encodebuf)
+ if ignore_error == 0:
+ flags = MBENC_FLUSH | MBENC_RESET
+ else:
+ flags = MBENC_RESET
+ while True:
+ r = pypy_cjk_enc_chunk(encodebuf, flags)
+ if r == 0 or r == ignore_error:
+ break
+ multibytecodec_encerror(encodebuf, r, errors,
+ errorcb, namecb, unicodedata)
+ while True:
+ r = pypy_cjk_enc_reset(encodebuf)
+ if r == 0:
+ break
+ multibytecodec_encerror(encodebuf, r, errors,
+ errorcb, namecb, unicodedata)
+ src = pypy_cjk_enc_outbuf(encodebuf)
+ length = pypy_cjk_enc_outlen(encodebuf)
+ return rffi.charpsize2str(src, length)
#
finally:
rffi.free_nonmoving_unicodebuffer(unicodedata, inbuf)
def multibytecodec_encerror(encodebuf, e, errors,
- codec, errorcb, namecb, unicodedata):
+ errorcb, namecb, unicodedata):
if e > 0:
reason = "illegal multibyte sequence"
esize = e
@@ -248,6 +272,7 @@
elif errors == "ignore":
replace = ""
elif errors == "replace":
+ codec = pypy_cjk_enc_getcodec(encodebuf)
try:
replace = encode(codec, u"?")
except EncodeDecodeError:
diff --git a/pypy/module/_multibytecodec/interp_incremental.py b/pypy/module/_multibytecodec/interp_incremental.py
new file mode 100644
--- /dev/null
+++ b/pypy/module/_multibytecodec/interp_incremental.py
@@ -0,0 +1,141 @@
+from pypy.rpython.lltypesystem import lltype
+from pypy.module._multibytecodec import c_codecs
+from pypy.module._multibytecodec.interp_multibytecodec import (
+ MultibyteCodec, wrap_unicodedecodeerror, wrap_runtimeerror,
+ wrap_unicodeencodeerror)
+from pypy.interpreter.baseobjspace import Wrappable
+from pypy.interpreter.gateway import interp2app, unwrap_spec
+from pypy.interpreter.typedef import TypeDef, GetSetProperty
+from pypy.module._codecs.interp_codecs import CodecState
+
+
+class MultibyteIncrementalBase(Wrappable):
+
+ def __init__(self, space, errors):
+ if errors is None:
+ errors = 'strict'
+ self.space = space
+ self.errors = errors
+ w_codec = space.getattr(space.wrap(self), space.wrap("codec"))
+ codec = space.interp_w(MultibyteCodec, w_codec)
+ self.codec = codec.codec
+ self.name = codec.name
+ self._initialize()
+
+ def __del__(self):
+ self._free()
+
+ def reset_w(self):
+ self._free()
+ self._initialize()
+
+ def fget_errors(self, space):
+ return space.wrap(self.errors)
+
+ def fset_errors(self, space, w_errors):
+ self.errors = space.str_w(w_errors)
+
+
+class MultibyteIncrementalDecoder(MultibyteIncrementalBase):
+
+ def _initialize(self):
+ self.decodebuf = c_codecs.pypy_cjk_dec_new(self.codec)
+ self.pending = ""
+
+ def _free(self):
+ self.pending = None
+ if self.decodebuf:
+ c_codecs.pypy_cjk_dec_free(self.decodebuf)
+ self.decodebuf = lltype.nullptr(c_codecs.DECODEBUF_P.TO)
+
+ @unwrap_spec(object=str, final=bool)
+ def decode_w(self, object, final=False):
+ space = self.space
+ state = space.fromcache(CodecState)
+ if len(self.pending) > 0:
+ object = self.pending + object
+ try:
+ output = c_codecs.decodeex(self.decodebuf, object, self.errors,
+ state.decode_error_handler, self.name,
+ get_ignore_error(final))
+ except c_codecs.EncodeDecodeError, e:
+ raise wrap_unicodedecodeerror(space, e, object, self.name)
+ except RuntimeError:
+ raise wrap_runtimeerror(space)
+ pos = c_codecs.pypy_cjk_dec_inbuf_consumed(self.decodebuf)
+ assert 0 <= pos <= len(object)
+ self.pending = object[pos:]
+ return space.wrap(output)
+
+
+ at unwrap_spec(errors="str_or_None")
+def mbidecoder_new(space, w_subtype, errors=None):
+ r = space.allocate_instance(MultibyteIncrementalDecoder, w_subtype)
+ r.__init__(space, errors)
+ return space.wrap(r)
+
+MultibyteIncrementalDecoder.typedef = TypeDef(
+ 'MultibyteIncrementalDecoder',
+ __module__ = '_multibytecodec',
+ __new__ = interp2app(mbidecoder_new),
+ decode = interp2app(MultibyteIncrementalDecoder.decode_w),
+ reset = interp2app(MultibyteIncrementalDecoder.reset_w),
+ errors = GetSetProperty(MultibyteIncrementalDecoder.fget_errors,
+ MultibyteIncrementalDecoder.fset_errors),
+ )
+
+
+class MultibyteIncrementalEncoder(MultibyteIncrementalBase):
+
+ def _initialize(self):
+ self.encodebuf = c_codecs.pypy_cjk_enc_new(self.codec)
+ self.pending = u""
+
+ def _free(self):
+ self.pending = None
+ if self.encodebuf:
+ c_codecs.pypy_cjk_enc_free(self.encodebuf)
+ self.encodebuf = lltype.nullptr(c_codecs.ENCODEBUF_P.TO)
+
+ @unwrap_spec(object=unicode, final=bool)
+ def encode_w(self, object, final=False):
+ space = self.space
+ state = space.fromcache(CodecState)
+ if len(self.pending) > 0:
+ object = self.pending + object
+ try:
+ output = c_codecs.encodeex(self.encodebuf, object, self.errors,
+ state.encode_error_handler, self.name,
+ get_ignore_error(final))
+ except c_codecs.EncodeDecodeError, e:
+ raise wrap_unicodeencodeerror(space, e, object, self.name)
+ except RuntimeError:
+ raise wrap_runtimeerror(space)
+ pos = c_codecs.pypy_cjk_enc_inbuf_consumed(self.encodebuf)
+ assert 0 <= pos <= len(object)
+ self.pending = object[pos:]
+ return space.wrap(output)
+
+
+ at unwrap_spec(errors="str_or_None")
+def mbiencoder_new(space, w_subtype, errors=None):
+ r = space.allocate_instance(MultibyteIncrementalEncoder, w_subtype)
+ r.__init__(space, errors)
+ return space.wrap(r)
+
+MultibyteIncrementalEncoder.typedef = TypeDef(
+ 'MultibyteIncrementalEncoder',
+ __module__ = '_multibytecodec',
+ __new__ = interp2app(mbiencoder_new),
+ encode = interp2app(MultibyteIncrementalEncoder.encode_w),
+ reset = interp2app(MultibyteIncrementalEncoder.reset_w),
+ errors = GetSetProperty(MultibyteIncrementalEncoder.fget_errors,
+ MultibyteIncrementalEncoder.fset_errors),
+ )
+
+
+def get_ignore_error(final):
+ if final:
+ return 0 # don't ignore any error
+ else:
+ return c_codecs.MBERR_TOOFEW
diff --git a/pypy/module/_multibytecodec/interp_multibytecodec.py b/pypy/module/_multibytecodec/interp_multibytecodec.py
--- a/pypy/module/_multibytecodec/interp_multibytecodec.py
+++ b/pypy/module/_multibytecodec/interp_multibytecodec.py
@@ -22,17 +22,9 @@
output = c_codecs.decode(self.codec, input, errors,
state.decode_error_handler, self.name)
except c_codecs.EncodeDecodeError, e:
- raise OperationError(
- space.w_UnicodeDecodeError,
- space.newtuple([
- space.wrap(self.name),
- space.wrap(input),
- space.wrap(e.start),
- space.wrap(e.end),
- space.wrap(e.reason)]))
+ raise wrap_unicodedecodeerror(space, e, input, self.name)
except RuntimeError:
- raise OperationError(space.w_RuntimeError,
- space.wrap("internal codec error"))
+ raise wrap_runtimeerror(space)
return space.newtuple([space.wrap(output),
space.wrap(len(input))])
@@ -46,17 +38,9 @@
output = c_codecs.encode(self.codec, input, errors,
state.encode_error_handler, self.name)
except c_codecs.EncodeDecodeError, e:
- raise OperationError(
- space.w_UnicodeEncodeError,
- space.newtuple([
- space.wrap(self.name),
- space.wrap(input),
- space.wrap(e.start),
- space.wrap(e.end),
- space.wrap(e.reason)]))
+ raise wrap_unicodeencodeerror(space, e, input, self.name)
except RuntimeError:
- raise OperationError(space.w_RuntimeError,
- space.wrap("internal codec error"))
+ raise wrap_runtimeerror(space)
return space.newtuple([space.wrap(output),
space.wrap(len(input))])
@@ -78,3 +62,28 @@
raise OperationError(space.w_LookupError,
space.wrap("no such codec is supported."))
return space.wrap(MultibyteCodec(name, codec))
+
+
+def wrap_unicodedecodeerror(space, e, input, name):
+ return OperationError(
+ space.w_UnicodeDecodeError,
+ space.newtuple([
+ space.wrap(name),
+ space.wrap(input),
+ space.wrap(e.start),
+ space.wrap(e.end),
+ space.wrap(e.reason)]))
+
+def wrap_unicodeencodeerror(space, e, input, name):
+ raise OperationError(
+ space.w_UnicodeEncodeError,
+ space.newtuple([
+ space.wrap(name),
+ space.wrap(input),
+ space.wrap(e.start),
+ space.wrap(e.end),
+ space.wrap(e.reason)]))
+
+def wrap_runtimeerror(space):
+ raise OperationError(space.w_RuntimeError,
+ space.wrap("internal codec error"))
diff --git a/pypy/module/_multibytecodec/test/test_app_incremental.py b/pypy/module/_multibytecodec/test/test_app_incremental.py
new file mode 100644
--- /dev/null
+++ b/pypy/module/_multibytecodec/test/test_app_incremental.py
@@ -0,0 +1,163 @@
+from pypy.conftest import gettestobjspace
+
+
+class AppTestClasses:
+ def setup_class(cls):
+ cls.space = gettestobjspace(usemodules=['_multibytecodec'])
+ cls.w_IncrementalHzDecoder = cls.space.appexec([], """():
+ import _codecs_cn
+ from _multibytecodec import MultibyteIncrementalDecoder
+
+ class IncrementalHzDecoder(MultibyteIncrementalDecoder):
+ codec = _codecs_cn.getcodec('hz')
+
+ return IncrementalHzDecoder
+ """)
+ cls.w_IncrementalHzEncoder = cls.space.appexec([], """():
+ import _codecs_cn
+ from _multibytecodec import MultibyteIncrementalEncoder
+
+ class IncrementalHzEncoder(MultibyteIncrementalEncoder):
+ codec = _codecs_cn.getcodec('hz')
+
+ return IncrementalHzEncoder
+ """)
+ cls.w_IncrementalBig5hkscsEncoder = cls.space.appexec([], """():
+ import _codecs_cn
+ from _multibytecodec import MultibyteIncrementalEncoder
+
+ class IncrementalBig5hkscsEncoder(MultibyteIncrementalEncoder):
+ codec = _codecs_cn.getcodec('big5hkscs')
+
+ return IncrementalBig5hkscsEncoder
+ """)
+
+ def test_decode_hz(self):
+ d = self.IncrementalHzDecoder()
+ r = d.decode("~{abcd~}")
+ assert r == u'\u5f95\u6c85'
+ r = d.decode("~{efgh~}")
+ assert r == u'\u5f50\u73b7'
+ for c, output in zip("!~{abcd~}xyz~{efgh",
+ [u'!', # !
+ u'', # ~
+ u'', # {
+ u'', # a
+ u'\u5f95', # b
+ u'', # c
+ u'\u6c85', # d
+ u'', # ~
+ u'', # }
+ u'x', # x
+ u'y', # y
+ u'z', # z
+ u'', # ~
+ u'', # {
+ u'', # e
+ u'\u5f50', # f
+ u'', # g
+ u'\u73b7', # h
+ ]):
+ r = d.decode(c)
+ assert r == output
+
+ def test_decode_hz_final(self):
+ d = self.IncrementalHzDecoder()
+ r = d.decode("~{", True)
+ assert r == u''
+ raises(UnicodeDecodeError, d.decode, "~", True)
+ raises(UnicodeDecodeError, d.decode, "~{a", True)
+
+ def test_decode_hz_reset(self):
+ d = self.IncrementalHzDecoder()
+ r = d.decode("ab")
+ assert r == u'ab'
+ r = d.decode("~{")
+ assert r == u''
+ r = d.decode("ab")
+ assert r == u'\u5f95'
+ r = d.decode("ab")
+ assert r == u'\u5f95'
+ d.reset()
+ r = d.decode("ab")
+ assert r == u'ab'
+
+ def test_decode_hz_error(self):
+ d = self.IncrementalHzDecoder()
+ raises(UnicodeDecodeError, d.decode, "~{abc", True)
+ d = self.IncrementalHzDecoder("ignore")
+ r = d.decode("~{abc", True)
+ assert r == u'\u5f95'
+ d = self.IncrementalHzDecoder()
+ d.errors = "replace"
+ r = d.decode("~{abc", True)
+ assert r == u'\u5f95\ufffd'
+
+ def test_decode_hz_buffer_grow(self):
+ d = self.IncrementalHzDecoder()
+ for i in range(13):
+ r = d.decode("a" * (2**i))
+ assert r == u"a" * (2**i)
+
+ def test_encode_hz(self):
+ e = self.IncrementalHzEncoder()
+ r = e.encode("abcd")
+ assert r == 'abcd'
+ r = e.encode(u"\u5f95\u6c85")
+ assert r == '~{abcd~}'
+ r = e.encode(u"\u5f50")
+ assert r == '~{ef~}'
+ r = e.encode(u"\u73b7")
+ assert r == '~{gh~}'
+
+ def test_encode_hz_final(self):
+ e = self.IncrementalHzEncoder()
+ r = e.encode(u"xyz\u5f95\u6c85", True)
+ assert r == 'xyz~{abcd~}'
+ # This is a bit hard to test, because the only way I can see that
+ # encoders can return MBERR_TOOFEW is with surrogates, which only
+ # occur with 2-byte unicode characters... We will just have to
+ # trust that the logic works, because it is exactly the same one
+ # as in the decode case :-/
+
+ def test_encode_hz_reset(self):
+ # Same issue as with test_encode_hz_final
+ e = self.IncrementalHzEncoder()
+ r = e.encode(u"xyz\u5f95\u6c85", True)
+ assert r == 'xyz~{abcd~}'
+ e.reset()
+ r = e.encode(u"xyz\u5f95\u6c85")
+ assert r == 'xyz~{abcd~}'
+
+ def test_encode_hz_error(self):
+ e = self.IncrementalHzEncoder()
+ raises(UnicodeEncodeError, e.encode, u"\u4321", True)
+ e = self.IncrementalHzEncoder("ignore")
+ r = e.encode(u"xy\u4321z", True)
+ assert r == 'xyz'
+ e = self.IncrementalHzEncoder()
+ e.errors = "replace"
+ r = e.encode(u"xy\u4321z", True)
+ assert r == 'xy?z'
+
+ def test_encode_hz_buffer_grow(self):
+ e = self.IncrementalHzEncoder()
+ for i in range(13):
+ r = e.encode(u"a" * (2**i))
+ assert r == "a" * (2**i)
+
+ def test_encode_big5hkscs(self):
+ #e = self.IncrementalBig5hkscsEncoder()
+ #r = e.encode(u'\xca', True)
+ #assert r == '\x88f'
+ #r = e.encode(u'\xca', True)
+ #assert r == '\x88f'
+ #raises(UnicodeEncodeError, e.encode, u'\u0304', True)
+ #
+ e = self.IncrementalBig5hkscsEncoder()
+ r = e.encode(u'\xca')
+ assert r == ''
+ r = e.encode(u'\xca')
+ assert r == '\x88f'
+ r = e.encode(u'\u0304')
+ assert r == '\x88b'
diff --git a/pypy/module/_multibytecodec/test/test_app_stream.py b/pypy/module/_multibytecodec/test/test_app_stream.py
new file mode 100644
--- /dev/null
+++ b/pypy/module/_multibytecodec/test/test_app_stream.py
@@ -0,0 +1,93 @@
+from pypy.conftest import gettestobjspace
+
+
+class AppTestStreams:
+ def setup_class(cls):
+ cls.space = gettestobjspace(usemodules=['_multibytecodec'])
+ cls.w_HzStreamReader = cls.space.appexec([], """():
+ import _codecs_cn
+ from _multibytecodec import MultibyteStreamReader
+
+ class HzStreamReader(MultibyteStreamReader):
+ codec = _codecs_cn.getcodec('hz')
+
+ return HzStreamReader
+ """)
+ cls.w_HzStreamWriter = cls.space.appexec([], """():
+ import _codecs_cn
+ from _multibytecodec import MultibyteStreamWriter
+
+ class HzStreamWriter(MultibyteStreamWriter):
+ codec = _codecs_cn.getcodec('hz')
+
+ return HzStreamWriter
+ """)
+ cls.w_ShiftJisx0213StreamWriter = cls.space.appexec([], """():
+ import _codecs_jp
+ from _multibytecodec import MultibyteStreamWriter
+
+ class ShiftJisx0213StreamWriter(MultibyteStreamWriter):
+ codec = _codecs_jp.getcodec('shift_jisx0213')
+
+ return ShiftJisx0213StreamWriter
+ """)
+
+ def test_reader(self):
+ class FakeFile:
+ def __init__(self, data):
+ self.data = data
+ self.pos = 0
+ def read(self, size):
+ res = self.data[self.pos : self.pos + size]
+ self.pos += size
+ return res
+ #
+ r = self.HzStreamReader(FakeFile("!~{abcd~}xyz~{efgh"))
+ for expected in u'!\u5f95\u6c85xyz\u5f50\u73b7':
+ c = r.read(1)
+ assert c == expected
+ c = r.read(1)
+ assert c == ''
+
+ def test_reader_replace(self):
+ class FakeFile:
+ def __init__(self, data):
+ self.data = data
+ def read(self):
+ return self.data
+ #
+ r = self.HzStreamReader(FakeFile("!~{a"), "replace")
+ c = r.read()
+ assert c == u'!\ufffd'
+ #
+ r = self.HzStreamReader(FakeFile("!~{a"))
+ r.errors = "replace"
+ assert r.errors == "replace"
+ c = r.read()
+ assert c == u'!\ufffd'
+
+ def test_writer(self):
+ class FakeFile:
+ def __init__(self):
+ self.output = []
+ def write(self, data):
+ self.output.append(data)
+ #
+ w = self.HzStreamWriter(FakeFile())
+ for input in u'!\u5f95\u6c85xyz\u5f50\u73b7':
+ w.write(input)
+ assert w.stream.output == ['!', '~{ab~}', '~{cd~}', 'x', 'y', 'z',
+ '~{ef~}', '~{gh~}']
+
+ def test_no_flush(self):
+ class FakeFile:
+ def __init__(self):
+ self.output = []
+ def write(self, data):
+ self.output.append(data)
+ #
+ w = self.ShiftJisx0213StreamWriter(FakeFile())
+ w.write(u'\u30ce')
+ w.write(u'\u304b')
+ w.write(u'\u309a')
+ assert w.stream.output == ['\x83m', '', '\x82\xf5']
diff --git a/pypy/module/_multibytecodec/test/test_c_codecs.py b/pypy/module/_multibytecodec/test/test_c_codecs.py
--- a/pypy/module/_multibytecodec/test/test_c_codecs.py
+++ b/pypy/module/_multibytecodec/test/test_c_codecs.py
@@ -2,6 +2,7 @@
from pypy.module._multibytecodec.c_codecs import getcodec, codecs
from pypy.module._multibytecodec.c_codecs import decode, encode
from pypy.module._multibytecodec.c_codecs import EncodeDecodeError
+from pypy.module._multibytecodec import c_codecs
def test_codecs_existence():
@@ -22,6 +23,52 @@
c = getcodec("hz")
u = decode(c, "~{abc}")
assert u == u'\u5f95\u6cef'
+ u = decode(c, "~{")
+ assert u == u''
+
+def test_decodeex_hz():
+ c = getcodec("hz")
+ decodebuf = c_codecs.pypy_cjk_dec_new(c)
+ u = c_codecs.decodeex(decodebuf, "~{abcd~}")
+ assert u == u'\u5f95\u6c85'
+ u = c_codecs.decodeex(decodebuf, "~{efgh~}")
+ assert u == u'\u5f50\u73b7'
+ u = c_codecs.decodeex(decodebuf, "!~{abcd~}xyz~{efgh")
+ assert u == u'!\u5f95\u6c85xyz\u5f50\u73b7'
+ c_codecs.pypy_cjk_dec_free(decodebuf)
+
+def test_decodeex_hz_incomplete():
+ c = getcodec("hz")
+ decodebuf = c_codecs.pypy_cjk_dec_new(c)
+ buf = ''
+ for c, output in zip("!~{abcd~}xyz~{efgh",
+ [u'!', # !
+ u'', # ~
+ u'', # {
+ u'', # a
+ u'\u5f95', # b
+ u'', # c
+ u'\u6c85', # d
+ u'', # ~
+ u'', # }
+ u'x', # x
+ u'y', # y
+ u'z', # z
+ u'', # ~
+ u'', # {
+ u'', # e
+ u'\u5f50', # f
+ u'', # g
+ u'\u73b7', # h
+ ]):
+ buf += c
+ u = c_codecs.decodeex(decodebuf, buf,
+ ignore_error = c_codecs.MBERR_TOOFEW)
+ assert u == output
+ incompletepos = c_codecs.pypy_cjk_dec_inbuf_consumed(decodebuf)
+ buf = buf[incompletepos:]
+ assert buf == ''
+ c_codecs.pypy_cjk_dec_free(decodebuf)
def test_decode_hz_error():
# error
diff --git a/pypy/module/_rawffi/interp_rawffi.py b/pypy/module/_rawffi/interp_rawffi.py
--- a/pypy/module/_rawffi/interp_rawffi.py
+++ b/pypy/module/_rawffi/interp_rawffi.py
@@ -268,7 +268,7 @@
self.ll_buffer = rffi.cast(rffi.VOIDP, address)
else:
self.ll_buffer = lltype.malloc(rffi.VOIDP.TO, size, flavor='raw',
- zero=True)
+ zero=True, add_memory_pressure=True)
if tracker.DO_TRACING:
ll_buf = rffi.cast(lltype.Signed, self.ll_buffer)
tracker.trace_allocation(ll_buf, self)
diff --git a/pypy/module/array/interp_array.py b/pypy/module/array/interp_array.py
--- a/pypy/module/array/interp_array.py
+++ b/pypy/module/array/interp_array.py
@@ -14,8 +14,6 @@
from pypy.rpython.lltypesystem import lltype, rffi
-memcpy = rffi.llexternal("memcpy", [rffi.VOIDP, rffi.VOIDP, rffi.SIZE_T], lltype.Void)
-
@unwrap_spec(typecode=str)
def w_array(space, w_cls, typecode, __args__):
if len(__args__.arguments_w) > 1:
@@ -228,7 +226,8 @@
some += size >> 3
self.allocated = size + some
new_buffer = lltype.malloc(mytype.arraytype,
- self.allocated, flavor='raw')
+ self.allocated, flavor='raw',
+ add_memory_pressure=True)
for i in range(min(size, self.len)):
new_buffer[i] = self.buffer[i]
else:
@@ -617,7 +616,7 @@
def array_copy__Array(space, self):
w_a = mytype.w_class(self.space)
w_a.setlen(self.len)
- memcpy(
+ rffi.c_memcpy(
rffi.cast(rffi.VOIDP, w_a.buffer),
rffi.cast(rffi.VOIDP, self.buffer),
self.len * mytype.bytes
diff --git a/pypy/module/cpyext/include/patchlevel.h b/pypy/module/cpyext/include/patchlevel.h
--- a/pypy/module/cpyext/include/patchlevel.h
+++ b/pypy/module/cpyext/include/patchlevel.h
@@ -31,8 +31,9 @@
/* PyPy version as a string */
#define PYPY_VERSION "1.6.0"
-/* Subversion Revision number of this file (not of the repository) */
-#define PY_PATCHLEVEL_REVISION "$Revision: 77872 $"
+/* Subversion Revision number of this file (not of the repository).
+ * Empty since Mercurial migration. */
+#define PY_PATCHLEVEL_REVISION ""
/* Version as a single 4-byte hex number, e.g. 0x010502B2 == 1.5.2b2.
Use this for numeric comparisons, e.g. #if PY_VERSION_HEX >= ... */
diff --git a/pypy/module/itertools/interp_itertools.py b/pypy/module/itertools/interp_itertools.py
--- a/pypy/module/itertools/interp_itertools.py
+++ b/pypy/module/itertools/interp_itertools.py
@@ -339,16 +339,21 @@
start = 0
else:
start = space.int_w(w_startstop)
+ if start < 0:
+ raise OperationError(space.w_ValueError, space.wrap(
+ "Indicies for islice() must be non-negative integers."))
w_stop = args_w[0]
else:
raise OperationError(space.w_TypeError, space.wrap("islice() takes at most 4 arguments (" + str(num_args) + " given)"))
if space.is_w(w_stop, space.w_None):
stop = -1
- stoppable = False
else:
stop = space.int_w(w_stop)
- stoppable = True
+ if stop < 0:
+ raise OperationError(space.w_ValueError, space.wrap(
+ "Stop argument must be a non-negative integer or None."))
+ stop = max(start, stop) # for obscure CPython compatibility
if num_args == 2:
w_step = args_w[1]
@@ -356,38 +361,37 @@
step = 1
else:
step = space.int_w(w_step)
+ if step < 1:
+ raise OperationError(space.w_ValueError, space.wrap(
+ "Step must be one or lager for islice()."))
else:
step = 1
- if start < 0:
- raise OperationError(space.w_ValueError, space.wrap("Indicies for islice() must be non-negative integers."))
- if stoppable and stop < 0:
- raise OperationError(space.w_ValueError, space.wrap("Stop argument must be a non-negative integer or None."))
- if step < 1:
- raise OperationError(space.w_ValueError, space.wrap("Step must be one or lager for islice()."))
-
+ self.ignore = step - 1
self.start = start
self.stop = stop
- self.step = step
def iter_w(self):
return self.space.wrap(self)
def next_w(self):
if self.start >= 0: # first call only
- consume = self.start + 1
+ ignore = self.start
self.start = -1
else: # all following calls
- consume = self.step
- if consume > 1:
- self._ignore_items(consume-1)
- if self.stop >= 0:
- if self.stop < consume:
+ ignore = self.ignore
+ stop = self.stop
+ if stop >= 0:
+ if stop <= ignore:
self.stop = 0 # reset the state so that a following next_w()
- self.step = 1 # has no effect any more
+ # has no effect any more
+ if stop > 0:
+ self._ignore_items(stop)
raise OperationError(self.space.w_StopIteration,
self.space.w_None)
- self.stop -= consume
+ self.stop = stop - (ignore + 1)
+ if ignore > 0:
+ self._ignore_items(ignore)
return self.space.next(self.iterable)
def _ignore_items(self, num):
diff --git a/pypy/module/itertools/test/test_itertools.py b/pypy/module/itertools/test/test_itertools.py
--- a/pypy/module/itertools/test/test_itertools.py
+++ b/pypy/module/itertools/test/test_itertools.py
@@ -266,6 +266,13 @@
raises(StopIteration, islc.next) # drops the 6th and raise
assert it.next() == "j"
+ it = iter("abcdefghij")
+ islc = itertools.islice(it, 3, 4, 3)
+ assert islc.next() == "d" # drops 0, 1, 2, returns item #3
+ assert it.next() == "e"
+ raises(StopIteration, islc.next) # item #4 is 'stop', so just raise
+ assert it.next() == "f"
+
def test_islice_overflow(self):
import itertools
import sys
diff --git a/pypy/module/micronumpy/__init__.py b/pypy/module/micronumpy/__init__.py
--- a/pypy/module/micronumpy/__init__.py
+++ b/pypy/module/micronumpy/__init__.py
@@ -31,6 +31,9 @@
'sin': 'interp_ufuncs.sin',
'cos': 'interp_ufuncs.cos',
'tan': 'interp_ufuncs.tan',
+ 'arcsin': 'interp_ufuncs.arcsin',
+ 'arccos': 'interp_ufuncs.arccos',
+ 'arctan': 'interp_ufuncs.arctan',
}
appleveldefs = {
diff --git a/pypy/module/micronumpy/interp_numarray.py b/pypy/module/micronumpy/interp_numarray.py
--- a/pypy/module/micronumpy/interp_numarray.py
+++ b/pypy/module/micronumpy/interp_numarray.py
@@ -187,17 +187,17 @@
def _getnums(self, comma):
if self.find_size() > 1000:
nums = [
- float2string(self.getitem(index))
+ float2string(self.eval(index))
for index in range(3)
]
nums.append("..." + "," * comma)
nums.extend([
- float2string(self.getitem(index))
+ float2string(self.eval(index))
for index in range(self.find_size() - 3, self.find_size())
])
else:
nums = [
- float2string(self.getitem(index))
+ float2string(self.eval(index))
for index in range(self.find_size())
]
return nums
@@ -229,7 +229,7 @@
start, stop, step, slice_length = space.decode_index4(w_idx, self.find_size())
if step == 0:
# Single index
- return space.wrap(self.get_concrete().getitem(start))
+ return space.wrap(self.get_concrete().eval(start))
else:
# Slice
res = SingleDimSlice(start, stop, step, slice_length, self, self.signature.transition(SingleDimSlice.static_signature))
@@ -416,14 +416,12 @@
# in fact, ViewArray never gets "concrete" as it never stores data.
# This implementation is needed for BaseArray getitem/setitem to work,
# can be refactored.
+ self.parent.get_concrete()
return self
def eval(self, i):
return self.parent.eval(self.calc_index(i))
- def getitem(self, item):
- return self.parent.getitem(self.calc_index(item))
-
@unwrap_spec(item=int, value=float)
def setitem(self, item, value):
return self.parent.setitem(self.calc_index(item), value)
@@ -479,7 +477,8 @@
BaseArray.__init__(self)
self.size = size
self.storage = lltype.malloc(TP, size, zero=True,
- flavor='raw', track_allocation=False)
+ flavor='raw', track_allocation=False,
+ add_memory_pressure=True)
# XXX find out why test_zjit explodes with trackign of allocations
def get_concrete(self):
@@ -497,9 +496,6 @@
def descr_len(self, space):
return space.wrap(self.size)
- def getitem(self, item):
- return self.storage[item]
-
def setitem(self, item, value):
self.invalidated()
self.storage[item] = value
@@ -511,7 +507,7 @@
self._sliceloop2(start, stop, step, arr, self)
def __del__(self):
- lltype.free(self.storage, flavor='raw')
+ lltype.free(self.storage, flavor='raw', track_allocation=False)
def new_numarray(space, w_size_or_iterable):
l = space.listview(w_size_or_iterable)
diff --git a/pypy/module/micronumpy/interp_ufuncs.py b/pypy/module/micronumpy/interp_ufuncs.py
--- a/pypy/module/micronumpy/interp_ufuncs.py
+++ b/pypy/module/micronumpy/interp_ufuncs.py
@@ -120,3 +120,20 @@
@ufunc2
def mod(lvalue, rvalue):
return math.fmod(lvalue, rvalue)
+
+
+ at ufunc
+def arcsin(value):
+ if value < -1.0 or value > 1.0:
+ return rfloat.NAN
+ return math.asin(value)
+
+ at ufunc
+def arccos(value):
+ if value < -1.0 or value > 1.0:
+ return rfloat.NAN
+ return math.acos(value)
+
+ at ufunc
+def arctan(value):
+ return math.atan(value)
diff --git a/pypy/module/micronumpy/test/test_numarray.py b/pypy/module/micronumpy/test/test_numarray.py
--- a/pypy/module/micronumpy/test/test_numarray.py
+++ b/pypy/module/micronumpy/test/test_numarray.py
@@ -70,6 +70,7 @@
from numpy import array, zeros
a = array(range(5))
assert str(a) == "[0.0 1.0 2.0 3.0 4.0]"
+ assert str((2*a)[:]) == "[0.0 2.0 4.0 6.0 8.0]"
a = zeros(1001)
assert str(a) == "[0.0 0.0 0.0 ..., 0.0 0.0 0.0]"
diff --git a/pypy/module/micronumpy/test/test_ufuncs.py b/pypy/module/micronumpy/test/test_ufuncs.py
--- a/pypy/module/micronumpy/test/test_ufuncs.py
+++ b/pypy/module/micronumpy/test/test_ufuncs.py
@@ -205,3 +205,47 @@
b = tan(a)
for i in range(len(a)):
assert b[i] == math.tan(a[i])
+
+
+ def test_arcsin(self):
+ import math
+ from numpy import array, arcsin
+
+ a = array([-1, -0.5, -0.33, 0, 0.33, 0.5, 1])
+ b = arcsin(a)
+ for i in range(len(a)):
+ assert b[i] == math.asin(a[i])
+
+ a = array([-10, -1.5, -1.01, 1.01, 1.5, 10, float('nan'), float('inf'), float('-inf')])
+ b = arcsin(a)
+ for f in b:
+ assert math.isnan(f)
+
+ def test_arccos(self):
+ import math
+ from numpy import array, arccos
+
+ a = array([-1, -0.5, -0.33, 0, 0.33, 0.5, 1])
+ b = arccos(a)
+ for i in range(len(a)):
+ assert b[i] == math.acos(a[i])
+
+
+ a = array([-10, -1.5, -1.01, 1.01, 1.5, 10, float('nan'), float('inf'), float('-inf')])
+ b = arccos(a)
+ for f in b:
+ assert math.isnan(f)
+
+ def test_arctan(self):
+ import math
+ from numpy import array, arctan
+
+ a = array([-3, -2, -1, 0, 1, 2, 3, float('inf'), float('-inf')])
+ b = arctan(a)
+ for i in range(len(a)):
+ assert b[i] == math.atan(a[i])
+
+ a = array([float('nan')])
+ b = arctan(a)
+ assert math.isnan(b[0])
+
diff --git a/pypy/module/pypyjit/test_pypy_c/test_array.py b/pypy/module/pypyjit/test_pypy_c/test_array.py
--- a/pypy/module/pypyjit/test_pypy_c/test_array.py
+++ b/pypy/module/pypyjit/test_pypy_c/test_array.py
@@ -1,4 +1,4 @@
-import py
+import py, sys
from pypy.module.pypyjit.test_pypy_c.test_00_model import BaseTestPyPyC
class TestArray(BaseTestPyPyC):
@@ -88,6 +88,73 @@
jump(p0, p1, p2, p3, p4, p5, p6, i28, i15, p9, i10, i11, descr=<Loop0>)
""")
+ def test_array_of_doubles(self):
+ def main():
+ from array import array
+ img = array('d', [21.5]*1000)
+ i = 0
+ while i < 1000:
+ img[i] += 20.5
+ assert img[i] == 42.0
+ i += 1
+ return 123
+ #
+ log = self.run(main, [])
+ assert log.result == 123
+ loop, = log.loops_by_filename(self.filepath)
+ assert loop.match("""
+ i10 = int_lt(i6, 1000)
+ guard_true(i10, descr=...)
+ i11 = int_lt(i6, i7)
+ guard_true(i11, descr=...)
+ f13 = getarrayitem_raw(i8, i6, descr=<FloatArrayNoLengthDescr>)
+ f15 = float_add(f13, 20.500000)
+ setarrayitem_raw(i8, i6, f15, descr=<FloatArrayNoLengthDescr>)
+ f16 = getarrayitem_raw(i8, i6, descr=<FloatArrayNoLengthDescr>)
+ i18 = float_eq(f16, 42.000000)
+ guard_true(i18, descr=...)
+ i20 = int_add(i6, 1)
+ --TICK--
+ jump(..., descr=<Loop0>)
+ """)
+
+ def test_array_of_floats(self):
+ def main():
+ from array import array
+ img = array('f', [21.5]*1000)
+ i = 0
+ while i < 1000:
+ img[i] += 20.5
+ assert img[i] == 42.0
+ i += 1
+ return 321
+ #
+ log = self.run(main, [])
+ assert log.result == 321
+ loop, = log.loops_by_filename(self.filepath)
+ if sys.maxint == 2147483647:
+ arraydescr = 'UnsignedArrayNoLengthDescr'
+ else:
+ arraydescr = 'UINTArrayNoLengthDescr'
+ assert loop.match("""
+ i10 = int_lt(i6, 1000)
+ guard_true(i10, descr=...)
+ i11 = int_lt(i6, i7)
+ guard_true(i11, descr=...)
+ i13 = getarrayitem_raw(i8, i6, descr=<%s>)
+ f14 = cast_singlefloat_to_float(i13)
+ f16 = float_add(f14, 20.500000)
+ i17 = cast_float_to_singlefloat(f16)
+ setarrayitem_raw(i8, i6,i17, descr=<%s>)
+ i18 = getarrayitem_raw(i8, i6, descr=<%s>)
+ f19 = cast_singlefloat_to_float(i18)
+ i21 = float_eq(f19, 42.000000)
+ guard_true(i21, descr=...)
+ i23 = int_add(i6, 1)
+ --TICK--
+ jump(..., descr=<Loop0>)
+ """ % (arraydescr, arraydescr, arraydescr))
+
def test_zeropadded(self):
def main():
diff --git a/pypy/module/pypyjit/test_pypy_c/test_misc.py b/pypy/module/pypyjit/test_pypy_c/test_misc.py
--- a/pypy/module/pypyjit/test_pypy_c/test_misc.py
+++ b/pypy/module/pypyjit/test_pypy_c/test_misc.py
@@ -63,6 +63,7 @@
i7 = int_gt(i4, 1)
guard_true(i7, descr=...)
p9 = call(ConstClass(fromint), i4, descr=...)
+ guard_no_exception(descr=...)
p11 = call(ConstClass(rbigint.mul), p5, p9, descr=...)
guard_no_exception(descr=...)
i13 = int_sub(i4, 1)
diff --git a/pypy/module/pypyjit/test_pypy_c/test_string.py b/pypy/module/pypyjit/test_pypy_c/test_string.py
--- a/pypy/module/pypyjit/test_pypy_c/test_string.py
+++ b/pypy/module/pypyjit/test_pypy_c/test_string.py
@@ -90,12 +90,12 @@
i46 = call(ConstClass(ll_startswith__rpy_stringPtr_rpy_stringPtr), p28, ConstPtr(ptr45), descr=<BoolCallDescr>)
guard_false(i46, descr=...)
p51 = new_with_vtable(21136408)
- setfield_gc(p51, p28, descr=<GcPtrFieldDescr .*NumberStringParser.inst_literal .*>)
- setfield_gc(p51, ConstPtr(ptr51), descr=<GcPtrFieldDescr pypy.objspace.std.strutil.NumberStringParser.inst_fname .*>)
- setfield_gc(p51, 1, descr=<SignedFieldDescr .*NumberStringParser.inst_sign .*>)
- setfield_gc(p51, 16, descr=<SignedFieldDescr .*NumberStringParser.inst_base .*>)
- setfield_gc(p51, p28, descr=<GcPtrFieldDescr .*NumberStringParser.inst_s .*>)
- setfield_gc(p51, i29, descr=<SignedFieldDescr .*NumberStringParser.inst_n .*>)
+ setfield_gc(p51, _, descr=...) # 6 setfields, but the order is dict-order-dependent
+ setfield_gc(p51, _, descr=...)
+ setfield_gc(p51, _, descr=...)
+ setfield_gc(p51, _, descr=...)
+ setfield_gc(p51, _, descr=...)
+ setfield_gc(p51, _, descr=...)
p55 = call(ConstClass(parse_digit_string), p51, descr=<GcPtrCallDescr>)
guard_no_exception(descr=...)
i57 = call(ConstClass(rbigint.toint), p55, descr=<SignedCallDescr>)
diff --git a/pypy/module/rctime/interp_time.py b/pypy/module/rctime/interp_time.py
--- a/pypy/module/rctime/interp_time.py
+++ b/pypy/module/rctime/interp_time.py
@@ -207,13 +207,13 @@
t = (((c_time(lltype.nullptr(rffi.TIME_TP.TO))) / YEAR) * YEAR)
# we cannot have reference to stack variable, put it on the heap
t_ref = lltype.malloc(rffi.TIME_TP.TO, 1, flavor='raw')
- t_ref[0] = t
+ t_ref[0] = rffi.cast(rffi.TIME_T, t)
p = c_localtime(t_ref)
janzone = -p.c_tm_gmtoff
tm_zone = rffi.charp2str(p.c_tm_zone)
janname = [" ", tm_zone][bool(tm_zone)]
tt = t + YEAR / 2
- t_ref[0] = tt
+ t_ref[0] = rffi.cast(rffi.TIME_T, tt)
p = c_localtime(t_ref)
lltype.free(t_ref, flavor='raw')
tm_zone = rffi.charp2str(p.c_tm_zone)
@@ -292,11 +292,14 @@
else:
seconds = space.float_w(w_seconds)
try:
- ovfcheck_float_to_int(seconds)
+ seconds = ovfcheck_float_to_int(seconds)
+ t = rffi.r_time_t(seconds)
+ if rffi.cast(lltype.Signed, t) != seconds:
+ raise OverflowError
except OverflowError:
raise OperationError(space.w_ValueError,
space.wrap("time argument too large"))
- return rffi.r_time_t(seconds)
+ return t
def _tm_to_tuple(space, t):
time_tuple = [
@@ -317,7 +320,7 @@
def _gettmarg(space, w_tup, allowNone=True):
if allowNone and space.is_w(w_tup, space.w_None):
# default to the current local time
- tt = rffi.r_time_t(pytime.time())
+ tt = rffi.r_time_t(int(pytime.time()))
t_ref = lltype.malloc(rffi.TIME_TP.TO, 1, flavor='raw')
t_ref[0] = tt
pbuf = c_localtime(t_ref)
diff --git a/pypy/module/test_lib_pypy/ctypes_tests/test_callbacks.py b/pypy/module/test_lib_pypy/ctypes_tests/test_callbacks.py
--- a/pypy/module/test_lib_pypy/ctypes_tests/test_callbacks.py
+++ b/pypy/module/test_lib_pypy/ctypes_tests/test_callbacks.py
@@ -14,14 +14,27 @@
return args[-1]
def check_type(self, typ, arg):
+ unwrapped_types = {
+ c_float: (float,),
+ c_double: (float,),
+ c_char: (str,),
+ c_char_p: (str,),
+ c_uint: (int, long),
+ c_ulong: (int, long),
+ }
+
PROTO = self.functype.im_func(typ, typ)
- result = PROTO(self.callback)(arg)
+ cfunc = PROTO(self.callback)
+ result = cfunc(arg)
if typ == c_float:
assert abs(result - arg) < 0.000001
else:
assert self.got_args == (arg,)
assert result == arg
+ result2 = cfunc(typ(arg))
+ assert type(result2) in unwrapped_types.get(typ, (int, long))
+
PROTO = self.functype.im_func(typ, c_byte, typ)
result = PROTO(self.callback)(-3, arg)
if typ == c_float:
@@ -222,3 +235,20 @@
out, err = capsys.readouterr()
assert (out, err) == ("", "")
+
+ def test_callback_pyobject(self):
+ def callback(obj):
+ return obj
+
+ FUNC = CFUNCTYPE(py_object, py_object)
+ cfunc = FUNC(callback)
+ param = c_int(42)
+ assert cfunc(param) is param
+
+ def test_raise_argumenterror(self):
+ def callback(x):
+ pass
+ FUNC = CFUNCTYPE(None, c_void_p)
+ cfunc = FUNC(callback)
+ param = c_uint(42)
+ py.test.raises(ArgumentError, "cfunc(param)")
diff --git a/pypy/module/test_lib_pypy/ctypes_tests/test_cast.py b/pypy/module/test_lib_pypy/ctypes_tests/test_cast.py
--- a/pypy/module/test_lib_pypy/ctypes_tests/test_cast.py
+++ b/pypy/module/test_lib_pypy/ctypes_tests/test_cast.py
@@ -90,3 +90,8 @@
assert sqrt._objects is my_sqrt._objects # on CPython too
my_sqrt._objects.clear()
my_sqrt._objects.update(saved_objects)
+
+ def test_cast_argumenterror(self):
+ param = c_uint(42)
+ py.test.raises(ArgumentError, "cast(param, c_void_p)")
+
diff --git a/pypy/module/test_lib_pypy/ctypes_tests/test_fastpath.py b/pypy/module/test_lib_pypy/ctypes_tests/test_fastpath.py
--- a/pypy/module/test_lib_pypy/ctypes_tests/test_fastpath.py
+++ b/pypy/module/test_lib_pypy/ctypes_tests/test_fastpath.py
@@ -1,4 +1,4 @@
-from ctypes import CDLL, POINTER, pointer, c_byte, c_int, c_char_p
+from ctypes import CDLL, POINTER, pointer, c_byte, c_int, c_char_p, CFUNCTYPE, c_void_p, c_size_t
import sys
import py
from support import BaseCTypesTestChecker
@@ -46,6 +46,24 @@
tf_b.argtypes = (c_byte,)
assert tf_b(-126) == -42
+ def test_from_cfunctype(self):
+ from _ctypes import _memmove_addr
+ functype = CFUNCTYPE(c_void_p, c_void_p, c_void_p, c_size_t)
+ my_memmove = functype(_memmove_addr)
+ assert my_memmove._is_fastpath
+
+ def test_undeclared_restype(self):
+ # make sure we get a fresh function
+ try:
+ del dll.tf_i
+ except AttributeError:
+ pass
+ tf_i = dll.tf_i
+ assert not tf_i._is_fastpath
+ tf_i.argtypes = (c_int,)
+ assert tf_i._is_fastpath
+ assert tf_i(12) == 4
+
def test_pointer_args(self):
f = dll._testfunc_p_p
f.restype = POINTER(c_int)
@@ -63,13 +81,10 @@
result = f(mystr, ord("b"))
assert result == "bcd"
- @py.test.mark.xfail
def test_strings(self):
f = dll.my_strchr
f.argtypes = [c_char_p, c_int]
f.restype = c_char_p
- # python strings need to be converted to c_char_p, but this is
- # supported only in the slow path so far
result = f("abcd", ord("b"))
assert result == "bcd"
diff --git a/pypy/module/test_lib_pypy/ctypes_tests/test_functions.py b/pypy/module/test_lib_pypy/ctypes_tests/test_functions.py
--- a/pypy/module/test_lib_pypy/ctypes_tests/test_functions.py
+++ b/pypy/module/test_lib_pypy/ctypes_tests/test_functions.py
@@ -488,11 +488,9 @@
warnings.simplefilter("always")
with warnings.catch_warnings(record=True) as w:
dll.get_an_integer()
- assert len(w) == 2
+ assert len(w) == 1
assert issubclass(w[0].category, RuntimeWarning)
- assert issubclass(w[1].category, RuntimeWarning)
assert "C function without declared arguments called" in str(w[0].message)
- assert "C function without declared return type called" in str(w[1].message)
def test_errcheck(self):
py.test.skip('fixme')
diff --git a/pypy/module/test_lib_pypy/ctypes_tests/test_structures.py b/pypy/module/test_lib_pypy/ctypes_tests/test_structures.py
--- a/pypy/module/test_lib_pypy/ctypes_tests/test_structures.py
+++ b/pypy/module/test_lib_pypy/ctypes_tests/test_structures.py
@@ -424,6 +424,15 @@
sys.settrace(oldtrace)
events = None
+ def test_large_fields(self):
+ # make sure that large fields are not "confused" with bitfields
+ # (because the bitfields use the higher bits of the "size" attribute)
+ Array = c_long * 8192
+ class X(Structure):
+ _fields_ = [('items', Array)]
+ obj = X()
+ assert isinstance(obj.items, Array)
+
class TestPointerMember(BaseCTypesTestChecker):
def test_1(self):
diff --git a/pypy/objspace/descroperation.py b/pypy/objspace/descroperation.py
--- a/pypy/objspace/descroperation.py
+++ b/pypy/objspace/descroperation.py
@@ -35,6 +35,13 @@
return w_hash
object_hash._annspecialcase_ = 'specialize:memo'
+def type_eq(space):
+ "Utility that returns the app-level descriptor type.__eq__."
+ w_src, w_eq = space.lookup_in_type_where(space.w_type,
+ '__eq__')
+ return w_eq
+type_eq._annspecialcase_ = 'specialize:memo'
+
def raiseattrerror(space, w_obj, name, w_descr=None):
w_type = space.type(w_obj)
typename = w_type.getname(space)
diff --git a/pypy/objspace/flow/operation.py b/pypy/objspace/flow/operation.py
--- a/pypy/objspace/flow/operation.py
+++ b/pypy/objspace/flow/operation.py
@@ -359,10 +359,10 @@
# All arguments are constants: call the operator now
try:
result = op(*args)
- except:
- etype, evalue, etb = sys.exc_info()
- msg = "generated by a constant operation: %s%r" % (
- name, tuple(args))
+ except Exception, e:
+ etype = e.__class__
+ msg = "generated by a constant operation: %s" % (
+ name)
raise OperationThatShouldNotBePropagatedError(
self.wrap(etype), self.wrap(msg))
else:
diff --git a/pypy/objspace/std/bytearrayobject.py b/pypy/objspace/std/bytearrayobject.py
--- a/pypy/objspace/std/bytearrayobject.py
+++ b/pypy/objspace/std/bytearrayobject.py
@@ -282,8 +282,8 @@
return space.wrap(''.join(w_bytearray.data))
def _convert_idx_params(space, w_self, w_start, w_stop):
- start = slicetype._Eval_SliceIndex(space, w_start)
- stop = slicetype._Eval_SliceIndex(space, w_stop)
+ start = slicetype.eval_slice_index(space, w_start)
+ stop = slicetype.eval_slice_index(space, w_stop)
length = len(w_self.data)
if start < 0:
start += length
diff --git a/pypy/objspace/std/mapdict.py b/pypy/objspace/std/mapdict.py
--- a/pypy/objspace/std/mapdict.py
+++ b/pypy/objspace/std/mapdict.py
@@ -421,6 +421,14 @@
key = ("slot", SLOTS_STARTING_FROM + index)
self._get_mapdict_map().write(self, key, w_value)
+ def delslotvalue(self, index):
+ key = ("slot", SLOTS_STARTING_FROM + index)
+ new_obj = self._get_mapdict_map().delete(self, key)
+ if new_obj is None:
+ return False
+ self._become(new_obj)
+ return True
+
# used by _weakref implemenation
def getweakref(self):
diff --git a/pypy/objspace/std/sliceobject.py b/pypy/objspace/std/sliceobject.py
--- a/pypy/objspace/std/sliceobject.py
+++ b/pypy/objspace/std/sliceobject.py
@@ -4,7 +4,7 @@
from pypy.interpreter import gateway
from pypy.objspace.std.model import registerimplementation, W_Object
from pypy.objspace.std.register_all import register_all
-from pypy.objspace.std.slicetype import _Eval_SliceIndex
+from pypy.objspace.std.slicetype import eval_slice_index
class W_SliceObject(W_Object):
from pypy.objspace.std.slicetype import slice_typedef as typedef
@@ -25,7 +25,7 @@
if space.is_w(w_slice.w_step, space.w_None):
step = 1
else:
- step = _Eval_SliceIndex(space, w_slice.w_step)
+ step = eval_slice_index(space, w_slice.w_step)
if step == 0:
raise OperationError(space.w_ValueError,
space.wrap("slice step cannot be zero"))
@@ -35,7 +35,7 @@
else:
start = 0
else:
- start = _Eval_SliceIndex(space, w_slice.w_start)
+ start = eval_slice_index(space, w_slice.w_start)
if start < 0:
start += length
if start < 0:
@@ -54,7 +54,7 @@
else:
stop = length
else:
- stop = _Eval_SliceIndex(space, w_slice.w_stop)
+ stop = eval_slice_index(space, w_slice.w_stop)
if stop < 0:
stop += length
if stop < 0:
diff --git a/pypy/objspace/std/slicetype.py b/pypy/objspace/std/slicetype.py
--- a/pypy/objspace/std/slicetype.py
+++ b/pypy/objspace/std/slicetype.py
@@ -14,7 +14,7 @@
' normal slices.')
# utility functions
-def _Eval_SliceIndex(space, w_int):
+def eval_slice_index(space, w_int):
try:
return space.getindex_w(w_int, None) # clamp if long integer too large
except OperationError, err:
@@ -25,7 +25,7 @@
"None or have an __index__ method"))
def adapt_lower_bound(space, size, w_index):
- index = _Eval_SliceIndex(space, w_index)
+ index = eval_slice_index(space, w_index)
if index < 0:
index = index + size
if index < 0:
@@ -34,7 +34,7 @@
return index
def adapt_bound(space, size, w_index):
- index = _Eval_SliceIndex(space, w_index)
+ index = eval_slice_index(space, w_index)
if index < 0:
index = index + size
if index < 0:
diff --git a/pypy/objspace/std/stringobject.py b/pypy/objspace/std/stringobject.py
--- a/pypy/objspace/std/stringobject.py
+++ b/pypy/objspace/std/stringobject.py
@@ -913,7 +913,7 @@
def repr__String(space, w_str):
s = w_str._value
- buf = StringBuilder(50)
+ buf = StringBuilder(len(s) + 2)
quote = "'"
if quote in s and '"' not in s:
diff --git a/pypy/objspace/std/test/test_identitydict.py b/pypy/objspace/std/test/test_identitydict.py
--- a/pypy/objspace/std/test/test_identitydict.py
+++ b/pypy/objspace/std/test/test_identitydict.py
@@ -32,10 +32,20 @@
def __hash__(self):
return 0
+ class TypeSubclass(type):
+ pass
+
+ class TypeSubclassCustomCmp(type):
+ def __cmp__(self, other):
+ return 0
+
assert self.compares_by_identity(Plain)
assert not self.compares_by_identity(CustomEq)
assert not self.compares_by_identity(CustomCmp)
assert not self.compares_by_identity(CustomHash)
+ assert self.compares_by_identity(type)
+ assert self.compares_by_identity(TypeSubclass)
+ assert not self.compares_by_identity(TypeSubclassCustomCmp)
def test_modify_class(self):
class X(object):
diff --git a/pypy/objspace/std/test/test_mapdict.py b/pypy/objspace/std/test/test_mapdict.py
--- a/pypy/objspace/std/test/test_mapdict.py
+++ b/pypy/objspace/std/test/test_mapdict.py
@@ -210,6 +210,12 @@
assert obj2.storage == [501, 601, 701, 51, 61, 71]
assert obj.map is obj2.map
+ assert obj2.getslotvalue(b) == 601
+ assert obj2.delslotvalue(b)
+ assert obj2.getslotvalue(b) is None
+ assert obj2.storage == [501, 701, 51, 61, 71]
+ assert not obj2.delslotvalue(b)
+
def test_slots_no_dict():
cls = Class(hasdict=False)
@@ -631,6 +637,14 @@
a.__dict__ = {}
a.__dict__ = {}
+ def test_delete_slot(self):
+ class A(object):
+ __slots__ = ['x']
+
+ a = A()
+ a.x = 42
+ del a.x
+ raises(AttributeError, "a.x")
class AppTestWithMapDictAndCounters(object):
def setup_class(cls):
diff --git a/pypy/objspace/std/tupleobject.py b/pypy/objspace/std/tupleobject.py
--- a/pypy/objspace/std/tupleobject.py
+++ b/pypy/objspace/std/tupleobject.py
@@ -154,7 +154,7 @@
x = 0x345678
z = len(wrappeditems)
for w_item in wrappeditems:
- y = space.int_w(space.hash(w_item))
+ y = space.hash_w(w_item)
x = (x ^ y) * mult
z -= 1
mult += 82520 + z + z
@@ -172,8 +172,8 @@
return space.wrap(count)
def tuple_index__Tuple_ANY_ANY_ANY(space, w_tuple, w_obj, w_start, w_stop):
- start = slicetype._Eval_SliceIndex(space, w_start)
- stop = slicetype._Eval_SliceIndex(space, w_stop)
+ start = slicetype.eval_slice_index(space, w_start)
+ stop = slicetype.eval_slice_index(space, w_stop)
length = len(w_tuple.wrappeditems)
if start < 0:
start += length
diff --git a/pypy/objspace/std/typeobject.py b/pypy/objspace/std/typeobject.py
--- a/pypy/objspace/std/typeobject.py
+++ b/pypy/objspace/std/typeobject.py
@@ -173,8 +173,6 @@
# ^^^ conservative default, fixed during real usage
if space.config.objspace.std.withidentitydict:
- did_compare_by_identity = (
- w_self.compares_by_identity_status == COMPARES_BY_IDENTITY)
if (key is None or key == '__eq__' or
key == '__cmp__' or key == '__hash__'):
w_self.compares_by_identity_status = UNKNOWN
@@ -229,7 +227,7 @@
return w_self.getattribute_if_not_from_object() is None
def compares_by_identity(w_self):
- from pypy.objspace.descroperation import object_hash
+ from pypy.objspace.descroperation import object_hash, type_eq
if not w_self.space.config.objspace.std.withidentitydict:
return False # conservative
#
@@ -238,7 +236,9 @@
return w_self.compares_by_identity_status == COMPARES_BY_IDENTITY
#
default_hash = object_hash(w_self.space)
- overrides_eq_cmp_or_hash = (w_self.lookup('__eq__') or
+ my_eq = w_self.lookup('__eq__')
+ overrides_eq = (my_eq and my_eq is not type_eq(w_self.space))
+ overrides_eq_cmp_or_hash = (overrides_eq or
w_self.lookup('__cmp__') or
w_self.lookup('__hash__') is not default_hash)
if overrides_eq_cmp_or_hash:
diff --git a/pypy/rlib/jit.py b/pypy/rlib/jit.py
--- a/pypy/rlib/jit.py
+++ b/pypy/rlib/jit.py
@@ -18,7 +18,8 @@
Most importantly it doesn't mean that an elidable function has no observable
side effect, but those side effects are idempotent (ie caching).
- For now, such a function should never raise an exception.
+ If a particular call to this function ends up raising an exception, then it
+ is handled like a normal function call (this decorator is ignored).
"""
func._elidable_function_ = True
return func
diff --git a/pypy/rlib/libffi.py b/pypy/rlib/libffi.py
--- a/pypy/rlib/libffi.py
+++ b/pypy/rlib/libffi.py
@@ -2,14 +2,13 @@
from pypy.rpython.lltypesystem import rffi, lltype
from pypy.rlib.objectmodel import specialize, enforceargs, we_are_translated
-from pypy.rlib.rarithmetic import intmask, r_uint, r_singlefloat
+from pypy.rlib.rarithmetic import intmask, r_uint, r_singlefloat, r_longlong
from pypy.rlib import jit
from pypy.rlib import clibffi
from pypy.rlib.clibffi import get_libc_name, FUNCFLAG_CDECL, AbstractFuncPtr, \
push_arg_as_ffiptr, c_ffi_call, FFI_TYPE_STRUCT
from pypy.rlib.rdynload import dlopen, dlclose, dlsym, dlsym_byordinal
from pypy.rlib.rdynload import DLLHANDLE
-from pypy.rlib.longlong2float import longlong2float, float2longlong
class types(object):
"""
@@ -122,9 +121,10 @@
elif TYPE is rffi.DOUBLE:
cls = FloatArg
elif TYPE is rffi.LONGLONG or TYPE is rffi.ULONGLONG:
- raise TypeError, 'r_(u)longlong not supported by arg(), use arg_(u)longlong()'
+ cls = LongLongArg
+ val = rffi.cast(rffi.LONGLONG, val)
elif TYPE is rffi.FLOAT:
- raise TypeError, 'r_singlefloat not supported by arg(), use arg_singlefloat()'
+ cls = SingleFloatArg
else:
raise TypeError, 'Unsupported argument type: %s' % TYPE
self._append(cls(val))
@@ -133,25 +133,6 @@
def arg_raw(self, val):
self._append(RawArg(val))
- def arg_longlong(self, val):
- """
- Note: this is a hack. So far, the JIT does not support long longs, so
- you must pass it as if it were a python Float (rffi.DOUBLE). You can
- use the convenience functions longlong2float and float2longlong to do
- the conversions. Note that if you use long longs, the call won't
- be jitted at all.
- """
- assert IS_32_BIT # use a normal integer on 64-bit platforms
- self._append(LongLongArg(val))
-
- def arg_singlefloat(self, val):
- """
- Note: you must pass a python Float (rffi.DOUBLE), not a r_singlefloat
- (else the jit complains). Note that if you use single floats, the
- call won't be jitted at all.
- """
- self._append(SingleFloatArg(val))
-
def _append(self, arg):
if self.first is None:
self.first = self.last = arg
@@ -196,25 +177,25 @@
func._push_raw(self.ptrval, ll_args, i)
class SingleFloatArg(AbstractArg):
- """ An argument representing a C float (but holding a C double)
+ """ An argument representing a C float
"""
- def __init__(self, floatval):
- self.floatval = floatval
+ def __init__(self, singlefloatval):
+ self.singlefloatval = singlefloatval
def push(self, func, ll_args, i):
- func._push_single_float(self.floatval, ll_args, i)
+ func._push_singlefloat(self.singlefloatval, ll_args, i)
class LongLongArg(AbstractArg):
- """ An argument representing a C long long (but holding a C double)
+ """ An argument representing a C long long
"""
- def __init__(self, floatval):
- self.floatval = floatval
+ def __init__(self, longlongval):
+ self.longlongval = longlongval
def push(self, func, ll_args, i):
- func._push_longlong(self.floatval, ll_args, i)
+ func._push_longlong(self.longlongval, ll_args, i)
# ======================================================================
@@ -274,15 +255,10 @@
elif RESULT is rffi.DOUBLE:
return self._do_call_float(self.funcsym, ll_args)
elif RESULT is rffi.FLOAT:
- # XXX: even if RESULT is FLOAT, we still return a DOUBLE, else the
- # jit complains. Note that the jit is disabled in this case
- return self._do_call_single_float(self.funcsym, ll_args)
+ return self._do_call_singlefloat(self.funcsym, ll_args)
elif RESULT is rffi.LONGLONG or RESULT is rffi.ULONGLONG:
- # XXX: even if RESULT is LONGLONG, we still return a DOUBLE, else the
- # jit complains. Note that the jit is disabled in this case
- # (it's not a typo, we really return a DOUBLE)
assert IS_32_BIT
- return self._do_call_longlong(self.funcsym, ll_args)
+ res = self._do_call_longlong(self.funcsym, ll_args)
elif RESULT is lltype.Void:
return self._do_call_void(self.funcsym, ll_args)
else:
@@ -320,16 +296,15 @@
def _push_float(self, value, ll_args, i):
self._push_arg(value, ll_args, i)
- @jit.dont_look_inside
- def _push_single_float(self, value, ll_args, i):
- self._push_arg(r_singlefloat(value), ll_args, i)
+ @jit.oopspec('libffi_push_singlefloat(self, value, ll_args, i)')
+ @enforceargs(None, r_singlefloat, None, int) # fix the annotation for tests
+ def _push_singlefloat(self, value, ll_args, i):
+ self._push_arg(value, ll_args, i)
- @jit.dont_look_inside
- def _push_longlong(self, floatval, ll_args, i):
- """
- Takes a longlong represented as a python Float. It's a hack for the
- jit, else we could not see the whole libffi module at all"""
- self._push_arg(float2longlong(floatval), ll_args, i)
+ @jit.oopspec('libffi_push_longlong(self, value, ll_args, i)')
+ @enforceargs(None, r_longlong, None, int) # fix the annotation for tests
+ def _push_longlong(self, value, ll_args, i):
+ self._push_arg(value, ll_args, i)
@jit.oopspec('libffi_call_int(self, funcsym, ll_args)')
def _do_call_int(self, funcsym, ll_args):
@@ -339,20 +314,18 @@
def _do_call_float(self, funcsym, ll_args):
return self._do_call(funcsym, ll_args, rffi.DOUBLE)
- @jit.dont_look_inside
- def _do_call_single_float(self, funcsym, ll_args):
- single_res = self._do_call(funcsym, ll_args, rffi.FLOAT)
- return float(single_res)
+ @jit.oopspec('libffi_call_singlefloat(self, funcsym, ll_args)')
+ def _do_call_singlefloat(self, funcsym, ll_args):
+ return self._do_call(funcsym, ll_args, rffi.FLOAT)
@jit.dont_look_inside
def _do_call_raw(self, funcsym, ll_args):
# same as _do_call_int, but marked as jit.dont_look_inside
return self._do_call(funcsym, ll_args, rffi.LONG)
- @jit.dont_look_inside
+ @jit.oopspec('libffi_call_longlong(self, funcsym, ll_args)')
def _do_call_longlong(self, funcsym, ll_args):
- llres = self._do_call(funcsym, ll_args, rffi.LONGLONG)
- return longlong2float(llres)
+ return self._do_call(funcsym, ll_args, rffi.LONGLONG)
@jit.oopspec('libffi_call_void(self, funcsym, ll_args)')
def _do_call_void(self, funcsym, ll_args):
diff --git a/pypy/rlib/longlong2float.py b/pypy/rlib/longlong2float.py
--- a/pypy/rlib/longlong2float.py
+++ b/pypy/rlib/longlong2float.py
@@ -11,6 +11,8 @@
# -------- implement longlong2float and float2longlong --------
DOUBLE_ARRAY_PTR = lltype.Ptr(lltype.Array(rffi.DOUBLE))
LONGLONG_ARRAY_PTR = lltype.Ptr(lltype.Array(rffi.LONGLONG))
+UINT_ARRAY_PTR = lltype.Ptr(lltype.Array(rffi.UINT))
+FLOAT_ARRAY_PTR = lltype.Ptr(lltype.Array(rffi.FLOAT))
# these definitions are used only in tests, when not translated
def longlong2float_emulator(llval):
@@ -29,6 +31,22 @@
lltype.free(d_array, flavor='raw')
return llval
+def uint2singlefloat_emulator(ival):
+ f_array = lltype.malloc(FLOAT_ARRAY_PTR.TO, 1, flavor='raw')
+ i_array = rffi.cast(UINT_ARRAY_PTR, f_array)
+ i_array[0] = ival
+ singlefloatval = f_array[0]
+ lltype.free(f_array, flavor='raw')
+ return singlefloatval
+
+def singlefloat2uint_emulator(singlefloatval):
+ f_array = lltype.malloc(FLOAT_ARRAY_PTR.TO, 1, flavor='raw')
+ i_array = rffi.cast(UINT_ARRAY_PTR, f_array)
+ f_array[0] = singlefloatval
+ ival = i_array[0]
+ lltype.free(f_array, flavor='raw')
+ return ival
+
from pypy.translator.tool.cbuild import ExternalCompilationInfo
eci = ExternalCompilationInfo(includes=['string.h', 'assert.h'],
post_include_bits=["""
@@ -44,6 +62,18 @@
memcpy(&ll, &x, 8);
return ll;
}
+static float pypy__uint2singlefloat(unsigned int x) {
+ float ff;
+ assert(sizeof(float) == 4 && sizeof(unsigned int) == 4);
+ memcpy(&ff, &x, 4);
+ return ff;
+}
+static unsigned int pypy__singlefloat2uint(float x) {
+ unsigned int ii;
+ assert(sizeof(float) == 4 && sizeof(unsigned int) == 4);
+ memcpy(&ii, &x, 4);
+ return ii;
+}
"""])
longlong2float = rffi.llexternal(
@@ -55,3 +85,13 @@
"pypy__float2longlong", [rffi.DOUBLE], rffi.LONGLONG,
_callable=float2longlong_emulator, compilation_info=eci,
_nowrapper=True, elidable_function=True)
+
+uint2singlefloat = rffi.llexternal(
+ "pypy__uint2singlefloat", [rffi.UINT], rffi.FLOAT,
+ _callable=uint2singlefloat_emulator, compilation_info=eci,
+ _nowrapper=True, elidable_function=True)
+
+singlefloat2uint = rffi.llexternal(
+ "pypy__singlefloat2uint", [rffi.FLOAT], rffi.UINT,
+ _callable=singlefloat2uint_emulator, compilation_info=eci,
+ _nowrapper=True, elidable_function=True)
diff --git a/pypy/rlib/rarithmetic.py b/pypy/rlib/rarithmetic.py
--- a/pypy/rlib/rarithmetic.py
+++ b/pypy/rlib/rarithmetic.py
@@ -71,9 +71,8 @@
return int(n)
def longlongmask(n):
- if isinstance(n, int):
- n = long(n)
- assert isinstance(n, long)
+ assert isinstance(n, (int, long))
+ n = long(n)
n &= LONGLONG_MASK
if n >= LONGLONG_TEST:
n -= 2*LONGLONG_TEST
diff --git a/pypy/rlib/rstring.py b/pypy/rlib/rstring.py
--- a/pypy/rlib/rstring.py
+++ b/pypy/rlib/rstring.py
@@ -1,8 +1,8 @@
""" String builder interface and string functions
"""
-from pypy.annotation.model import SomeObject, SomeString, s_None,\
- SomeChar, SomeInteger, SomeUnicodeCodePoint, SomeUnicodeString
+from pypy.annotation.model import (SomeObject, SomeString, s_None, SomeChar,
+ SomeInteger, SomeUnicodeCodePoint, SomeUnicodeString, SomePtr)
from pypy.rpython.extregistry import ExtRegistryEntry
@@ -65,6 +65,12 @@
assert isinstance(c, self.tp)
self.l.append(c * times)
+ def append_charpsize(self, s, size):
+ l = []
+ for i in xrange(size):
+ l.append(s[i])
+ self.l.append(self.tp("").join(l))
+
def build(self):
return self.tp("").join(self.l)
@@ -100,6 +106,11 @@
assert isinstance(s_times, SomeInteger)
return s_None
+ def method_append_charpsize(self, s_ptr, s_size):
+ assert isinstance(s_ptr, SomePtr)
+ assert isinstance(s_size, SomeInteger)
+ return s_None
+
def method_getlength(self):
return SomeInteger(nonneg=True)
@@ -127,6 +138,11 @@
assert isinstance(s_times, SomeInteger)
return s_None
+ def method_append_charpsize(self, s_ptr, s_size):
+ assert isinstance(s_ptr, SomePtr)
+ assert isinstance(s_size, SomeInteger)
+ return s_None
+
def method_getlength(self):
return SomeInteger(nonneg=True)
diff --git a/pypy/rlib/rzlib.py b/pypy/rlib/rzlib.py
--- a/pypy/rlib/rzlib.py
+++ b/pypy/rlib/rzlib.py
@@ -1,8 +1,11 @@
import sys
+
+from pypy.rlib.rstring import StringBuilder
from pypy.rpython.lltypesystem import rffi, lltype
from pypy.rpython.tool import rffi_platform
+from pypy.translator.platform import platform as compiler, CompilationError
from pypy.translator.tool.cbuild import ExternalCompilationInfo
-from pypy.translator.platform import platform as compiler, CompilationError
+
if compiler.name == "msvc":
libname = 'zlib'
@@ -337,23 +340,18 @@
"""Common code for compress() and decompress().
"""
# Prepare the input buffer for the stream
- inbuf = lltype.malloc(rffi.CCHARP.TO, len(data), flavor='raw')
- try:
+ with lltype.scoped_alloc(rffi.CCHARP.TO, len(data)) as inbuf:
for i in xrange(len(data)):
inbuf[i] = data[i]
stream.c_next_in = rffi.cast(Bytefp, inbuf)
rffi.setintfield(stream, 'c_avail_in', len(data))
# Prepare the output buffer
- outbuf = lltype.malloc(rffi.CCHARP.TO, OUTPUT_BUFFER_SIZE,
- flavor='raw')
- try:
- # Strategy: we call deflate() to get as much output data as
- # fits in the buffer, then accumulate all output into a list
- # of characters 'result'. We don't need to gradually
- # increase the output buffer size because there is no
- # quadratic factor.
- result = []
+ with lltype.scoped_alloc(rffi.CCHARP.TO, OUTPUT_BUFFER_SIZE) as outbuf:
+ # Strategy: we call deflate() to get as much output data as fits in
+ # the buffer, then accumulate all output into a StringBuffer
+ # 'result'.
+ result = StringBuilder()
while True:
stream.c_next_out = rffi.cast(Bytefp, outbuf)
@@ -369,8 +367,7 @@
if err == Z_OK or err == Z_STREAM_END:
# accumulate data into 'result'
avail_out = rffi.cast(lltype.Signed, stream.c_avail_out)
- for i in xrange(bufsize - avail_out):
- result.append(outbuf[i])
+ result.append_charpsize(outbuf, bufsize - avail_out)
# if the output buffer is full, there might be more data
# so we need to try again. Otherwise, we're done.
if avail_out > 0:
@@ -393,14 +390,9 @@
# fallback case: report this error
raise RZlibError.fromstream(stream, err, while_doing)
- finally:
- lltype.free(outbuf, flavor='raw')
- finally:
- lltype.free(inbuf, flavor='raw')
-
# When decompressing, if the compressed stream of data was truncated,
# then the zlib simply returns Z_OK and waits for more. If it is
# complete it returns Z_STREAM_END.
- return (''.join(result),
+ return (result.build(),
err,
rffi.cast(lltype.Signed, stream.c_avail_in))
diff --git a/pypy/rlib/streamio.py b/pypy/rlib/streamio.py
--- a/pypy/rlib/streamio.py
+++ b/pypy/rlib/streamio.py
@@ -894,13 +894,10 @@
self.buf.append(data)
self.buflen += datalen
elif buflen:
- i = self.bufsize - buflen
- assert i >= 0
- self.buf.append(data[:i])
+ self.buf.append(data)
self.do_write(''.join(self.buf))
self.buf = []
self.buflen = 0
- self.write(data[i:])
else:
self.do_write(data)
diff --git a/pypy/rlib/test/test_libffi.py b/pypy/rlib/test/test_libffi.py
--- a/pypy/rlib/test/test_libffi.py
+++ b/pypy/rlib/test/test_libffi.py
@@ -5,7 +5,7 @@
from pypy.rlib.rarithmetic import r_singlefloat, r_longlong, r_ulonglong
from pypy.rlib.test.test_clibffi import BaseFfiTest, get_libm_name, make_struct_ffitype_e
from pypy.rlib.libffi import CDLL, Func, get_libc_name, ArgChain, types
-from pypy.rlib.libffi import longlong2float, float2longlong, IS_32_BIT
+from pypy.rlib.libffi import IS_32_BIT
class TestLibffiMisc(BaseFfiTest):
@@ -52,19 +52,6 @@
del lib
assert not ALLOCATED
- def test_longlong_as_float(self):
- from pypy.translator.c.test.test_genc import compile
- maxint64 = r_longlong(9223372036854775807)
- def fn(x):
- d = longlong2float(x)
- ll = float2longlong(d)
- return ll
- assert fn(maxint64) == maxint64
- #
- fn2 = compile(fn, [r_longlong])
- res = fn2(maxint64)
- assert res == maxint64
-
class TestLibffiCall(BaseFfiTest):
"""
Test various kind of calls through libffi.
@@ -111,7 +98,7 @@
def get_libfoo(self):
return self.CDLL(self.libfoo_name)
- def call(self, funcspec, args, RESULT, init_result=0, is_struct=False):
+ def call(self, funcspec, args, RESULT, is_struct=False, jitif=[]):
"""
Call the specified function after constructing and ArgChain with the
arguments in ``args``.
@@ -128,14 +115,7 @@
func = lib.getpointer(name, argtypes, restype)
chain = ArgChain()
for arg in args:
- if isinstance(arg, r_singlefloat):
- chain.arg_singlefloat(float(arg))
- elif IS_32_BIT and isinstance(arg, r_longlong):
- chain.arg_longlong(longlong2float(arg))
- elif IS_32_BIT and isinstance(arg, r_ulonglong):
- arg = rffi.cast(rffi.LONGLONG, arg)
- chain.arg_longlong(longlong2float(arg))
- elif isinstance(arg, tuple):
+ if isinstance(arg, tuple):
methname, arg = arg
meth = getattr(chain, methname)
meth(arg)
@@ -143,13 +123,19 @@
chain.arg(arg)
return func.call(chain, RESULT, is_struct=is_struct)
- def check_loops(self, *args, **kwds):
+ # ------------------------------------------------------------------------
+
+ def test_very_simple(self):
"""
- Ignored here, but does something in the JIT tests
+ int diff_xy(int x, long y)
+ {
+ return x - y;
+ }
"""
- pass
-
- # ------------------------------------------------------------------------
+ libfoo = self.get_libfoo()
+ func = (libfoo, 'diff_xy', [types.sint, types.slong], types.sint)
+ res = self.call(func, [50, 8], lltype.Signed)
+ assert res == 42
def test_simple(self):
"""
@@ -160,23 +146,14 @@
"""
libfoo = self.get_libfoo()
func = (libfoo, 'sum_xy', [types.sint, types.double], types.sint)
- res = self.call(func, [38, 4.2], rffi.LONG)
+ res = self.call(func, [38, 4.2], lltype.Signed, jitif=["floats"])
assert res == 42
- self.check_loops({
- 'call_release_gil': 1,
- 'guard_no_exception': 1,
- 'guard_not_forced': 1,
- 'int_add': 1,
- 'int_lt': 1,
- 'guard_true': 1,
- 'jump': 1})
def test_float_result(self):
libm = self.get_libm()
func = (libm, 'pow', [types.double, types.double], types.double)
- res = self.call(func, [2.0, 3.0], rffi.DOUBLE, init_result=0.0)
+ res = self.call(func, [2.0, 3.0], rffi.DOUBLE, jitif=["floats"])
assert res == 8.0
- self.check_loops(call_release_gil=1, guard_no_exception=1, guard_not_forced=1)
def test_cast_result(self):
"""
@@ -189,7 +166,6 @@
func = (libfoo, 'cast_to_uchar_and_ovf', [types.sint], types.uchar)
res = self.call(func, [0], rffi.UCHAR)
assert res == 200
- self.check_loops(call_release_gil=1, guard_no_exception=1, guard_not_forced=1)
def test_cast_argument(self):
"""
@@ -271,8 +247,7 @@
libfoo = self.get_libfoo()
func = (libfoo, 'get_pointer_to_b', [], types.pointer)
LONGP = lltype.Ptr(rffi.CArray(rffi.LONG))
- null = lltype.nullptr(LONGP.TO)
- res = self.call(func, [], LONGP, init_result=null)
+ res = self.call(func, [], LONGP)
assert res[0] == 20
def test_void_result(self):
@@ -287,7 +262,7 @@
#
initval = self.call(get_dummy, [], rffi.LONG)
#
- res = self.call(set_dummy, [initval+1], lltype.Void, init_result=None)
+ res = self.call(set_dummy, [initval+1], lltype.Void)
assert res is None
#
res = self.call(get_dummy, [], rffi.LONG)
@@ -305,9 +280,9 @@
func = (libfoo, 'sum_xy_float', [types.float, types.float], types.float)
x = r_singlefloat(12.34)
y = r_singlefloat(56.78)
- res = self.call(func, [x, y], rffi.FLOAT, init_result=0.0)
+ res = self.call(func, [x, y], rffi.FLOAT, jitif=["singlefloats"])
expected = c_float(c_float(12.34).value + c_float(56.78).value).value
- assert res == expected
+ assert float(res) == expected
def test_slonglong_args(self):
"""
@@ -325,16 +300,10 @@
if IS_32_BIT:
x = r_longlong(maxint32+1)
y = r_longlong(maxint32+2)
- zero = longlong2float(r_longlong(0))
else:
x = maxint32+1
y = maxint32+2
- zero = 0
- res = self.call(func, [x, y], rffi.LONGLONG, init_result=zero)
- if IS_32_BIT:
- # obscure, on 32bit it's really a long long, so it returns a
- # DOUBLE because of the JIT hack
- res = float2longlong(res)
+ res = self.call(func, [x, y], rffi.LONGLONG, jitif=["longlong"])
expected = maxint32*2 + 3
assert res == expected
@@ -354,12 +323,7 @@
types.ulonglong)
x = r_ulonglong(maxint64+1)
y = r_ulonglong(2)
- res = self.call(func, [x, y], rffi.ULONGLONG, init_result=0)
- if IS_32_BIT:
- # obscure, on 32bit it's really a long long, so it returns a
- # DOUBLE because of the JIT hack
- res = float2longlong(res)
- res = rffi.cast(rffi.ULONGLONG, res)
+ res = self.call(func, [x, y], rffi.ULONGLONG, jitif=["longlong"])
expected = maxint64 + 3
assert res == expected
@@ -406,7 +370,8 @@
buf[0] = 30
buf[1] = 12
adr = rffi.cast(rffi.VOIDP, buf)
- res = self.call(sum_point, [('arg_raw', adr)], rffi.LONG, init_result=0)
+ res = self.call(sum_point, [('arg_raw', adr)], rffi.LONG,
+ jitif=["byval"])
assert res == 42
# check that we still have the ownership on the buffer
assert buf[0] == 30
@@ -431,8 +396,8 @@
make_point = (libfoo, 'make_point', [types.slong, types.slong], ffi_point)
#
PTR = lltype.Ptr(rffi.CArray(rffi.LONG))
- p = self.call(make_point, [12, 34], PTR, init_result=lltype.nullptr(PTR.TO),
- is_struct=True)
+ p = self.call(make_point, [12, 34], PTR, is_struct=True,
+ jitif=["byval"])
assert p[0] == 12
assert p[1] == 34
lltype.free(p, flavor='raw')
diff --git a/pypy/rlib/test/test_longlong2float.py b/pypy/rlib/test/test_longlong2float.py
--- a/pypy/rlib/test/test_longlong2float.py
+++ b/pypy/rlib/test/test_longlong2float.py
@@ -1,5 +1,7 @@
from pypy.translator.c.test.test_genc import compile
from pypy.rlib.longlong2float import longlong2float, float2longlong
+from pypy.rlib.longlong2float import uint2singlefloat, singlefloat2uint
+from pypy.rlib.rarithmetic import r_singlefloat
def fn(f1):
@@ -28,3 +30,23 @@
for x in enum_floats():
res = fn2(x)
assert repr(res) == repr(x)
+
+# ____________________________________________________________
+
+def fnsingle(f1):
+ sf1 = r_singlefloat(f1)
+ ii = singlefloat2uint(sf1)
+ sf2 = uint2singlefloat(ii)
+ f2 = float(sf2)
+ return f2
+
+def test_int_as_singlefloat():
+ for x in enum_floats():
+ res = fnsingle(x)
+ assert repr(res) == repr(float(r_singlefloat(x)))
+
+def test_compiled_single():
+ fn2 = compile(fnsingle, [float])
+ for x in enum_floats():
+ res = fn2(x)
+ assert repr(res) == repr(float(r_singlefloat(x)))
diff --git a/pypy/rpython/lltypesystem/ll2ctypes.py b/pypy/rpython/lltypesystem/ll2ctypes.py
--- a/pypy/rpython/lltypesystem/ll2ctypes.py
+++ b/pypy/rpython/lltypesystem/ll2ctypes.py
@@ -27,7 +27,11 @@
from pypy.rpython import raddress
from pypy.translator.platform import platform
from array import array
-from thread import _local as tlsobject
+try:
+ from thread import _local as tlsobject
+except ImportError:
+ class tlsobject(object):
+ pass
# ____________________________________________________________
@@ -688,6 +692,8 @@
res = ctypes.cast(res, ctypes.c_void_p).value
if res is None:
return 0
+ if T.TO.RESULT == lltype.SingleFloat:
+ res = res.value # baaaah, cannot return a c_float()
return res
def callback(*cargs):
diff --git a/pypy/rpython/lltypesystem/lltype.py b/pypy/rpython/lltypesystem/lltype.py
--- a/pypy/rpython/lltypesystem/lltype.py
+++ b/pypy/rpython/lltypesystem/lltype.py
@@ -1,7 +1,7 @@
import py
from pypy.rlib.rarithmetic import (r_int, r_uint, intmask, r_singlefloat,
r_ulonglong, r_longlong, r_longfloat,
- base_int, normalizedinttype)
+ base_int, normalizedinttype, longlongmask)
from pypy.rlib.objectmodel import Symbolic
from pypy.tool.uid import Hashable
from pypy.tool.identity_dict import identity_dict
@@ -654,6 +654,9 @@
_numbertypes = {int: Number("Signed", int, intmask)}
_numbertypes[r_int] = _numbertypes[int]
+if r_longlong is not r_int:
+ _numbertypes[r_longlong] = Number("SignedLongLong", r_longlong,
+ longlongmask)
def build_number(name, type):
try:
@@ -1146,7 +1149,7 @@
try:
return self._lookup_adtmeth(field_name)
except AttributeError:
- raise AttributeError("%r instance has no field %r" % (self._T,
+ raise AttributeError("%r instance has no field %r" % (self._T._name,
field_name))
def __setattr__(self, field_name, val):
@@ -1936,7 +1939,7 @@
def malloc(T, n=None, flavor='gc', immortal=False, zero=False,
- track_allocation=True):
+ track_allocation=True, add_memory_pressure=False):
assert flavor in ('gc', 'raw')
if zero or immortal:
initialization = 'example'
diff --git a/pypy/rpython/lltypesystem/rbuilder.py b/pypy/rpython/lltypesystem/rbuilder.py
--- a/pypy/rpython/lltypesystem/rbuilder.py
+++ b/pypy/rpython/lltypesystem/rbuilder.py
@@ -1,13 +1,13 @@
-
+from pypy.rlib import rgc
+from pypy.rlib.objectmodel import enforceargs
+from pypy.rlib.rarithmetic import ovfcheck
+from pypy.rpython.annlowlevel import llstr
+from pypy.rpython.rptr import PtrRepr
+from pypy.rpython.lltypesystem import lltype, rstr
+from pypy.rpython.lltypesystem.lltype import staticAdtMethod
+from pypy.rpython.lltypesystem.rstr import (STR, UNICODE, char_repr,
+ string_repr, unichar_repr, unicode_repr)
from pypy.rpython.rbuilder import AbstractStringBuilderRepr
-from pypy.rpython.lltypesystem import lltype, rstr
-from pypy.rpython.lltypesystem.rstr import STR, UNICODE, char_repr,\
- string_repr, unichar_repr, unicode_repr
-from pypy.rpython.annlowlevel import llstr
-from pypy.rlib import rgc
-from pypy.rlib.rarithmetic import ovfcheck
-from pypy.rlib.objectmodel import enforceargs
-from pypy.rpython.lltypesystem.lltype import staticAdtMethod
from pypy.tool.sourcetools import func_with_new_name
# Think about heuristics below, maybe we can come up with something
@@ -73,7 +73,7 @@
ll_builder.grow(ll_builder, lgt)
ll_str.copy_contents(ll_str, ll_builder.buf, 0, used, lgt)
ll_builder.used = needed
-
+
@staticmethod
def ll_append_char(ll_builder, char):
if ll_builder.used == ll_builder.allocated:
@@ -102,6 +102,16 @@
ll_builder.used = used
@staticmethod
+ def ll_append_charpsize(ll_builder, charp, size):
+ used = ll_builder.used
+ if used + size > ll_builder.allocated:
+ ll_builder.grow(ll_builder, size)
+ for i in xrange(size):
+ ll_builder.buf.chars[used] = charp[i]
+ used += 1
+ ll_builder.used = used
+
+ @staticmethod
def ll_getlength(ll_builder):
return ll_builder.used
@@ -119,6 +129,9 @@
mallocfn = staticmethod(rstr.mallocstr)
string_repr = string_repr
char_repr = char_repr
+ raw_ptr_repr = PtrRepr(
+ lltype.Ptr(lltype.Array(lltype.Char, hints={'nolength': True}))
+ )
class UnicodeBuilderRepr(BaseStringBuilderRepr):
lowleveltype = lltype.Ptr(UNICODEBUILDER)
@@ -126,6 +139,9 @@
mallocfn = staticmethod(rstr.mallocunicode)
string_repr = unicode_repr
char_repr = unichar_repr
+ raw_ptr_repr = PtrRepr(
+ lltype.Ptr(lltype.Array(lltype.UniChar, hints={'nolength': True}))
+ )
unicodebuilder_repr = UnicodeBuilderRepr()
stringbuilder_repr = StringBuilderRepr()
diff --git a/pypy/rpython/lltypesystem/rdict.py b/pypy/rpython/lltypesystem/rdict.py
--- a/pypy/rpython/lltypesystem/rdict.py
+++ b/pypy/rpython/lltypesystem/rdict.py
@@ -29,7 +29,7 @@
# DICTVALUE value;
# int f_hash; # (optional) key hash, if hard to recompute
# }
-#
+#
# struct dicttable {
# int num_items;
# int num_pristine_entries; # never used entries
@@ -50,12 +50,12 @@
self.custom_eq_hash = custom_eq_hash is not None
if not isinstance(key_repr, rmodel.Repr): # not computed yet, done by setup()
assert callable(key_repr)
- self._key_repr_computer = key_repr
+ self._key_repr_computer = key_repr
else:
self.external_key_repr, self.key_repr = self.pickkeyrepr(key_repr)
if not isinstance(value_repr, rmodel.Repr): # not computed yet, done by setup()
assert callable(value_repr)
- self._value_repr_computer = value_repr
+ self._value_repr_computer = value_repr
else:
self.external_value_repr, self.value_repr = self.pickrepr(value_repr)
self.dictkey = dictkey
@@ -176,7 +176,7 @@
self.DICTENTRYARRAY = lltype.GcArray(self.DICTENTRY,
adtmeths=entrymeths)
fields = [ ("num_items", lltype.Signed),
- ("num_pristine_entries", lltype.Signed),
+ ("num_pristine_entries", lltype.Signed),
("entries", lltype.Ptr(self.DICTENTRYARRAY)) ]
if self.custom_eq_hash:
self.r_rdict_eqfn, self.r_rdict_hashfn = self._custom_eq_hash_repr()
@@ -211,7 +211,7 @@
def convert_const(self, dictobj):
from pypy.rpython.lltypesystem import llmemory
# get object from bound dict methods
- #dictobj = getattr(dictobj, '__self__', dictobj)
+ #dictobj = getattr(dictobj, '__self__', dictobj)
if dictobj is None:
return lltype.nullptr(self.DICT)
if not isinstance(dictobj, (dict, objectmodel.r_dict)):
@@ -222,7 +222,7 @@
except KeyError:
self.setup()
l_dict = ll_newdict_size(self.DICT, len(dictobj))
- self.dict_cache[key] = l_dict
+ self.dict_cache[key] = l_dict
r_key = self.key_repr
if r_key.lowleveltype == llmemory.Address:
raise TypeError("No prebuilt dicts of address keys")
@@ -274,7 +274,7 @@
hop.exception_cannot_occur()
v_res = hop.gendirectcall(ll_setdefault, v_dict, v_key, v_default)
return self.recast_value(hop.llops, v_res)
-
+
def rtype_method_copy(self, hop):
v_dict, = hop.inputargs(self)
hop.exception_cannot_occur()
@@ -325,7 +325,7 @@
hop.exception_is_here()
return hop.gendirectcall(ll_popitem, cTUPLE, v_dict)
-class __extend__(pairtype(DictRepr, rmodel.Repr)):
+class __extend__(pairtype(DictRepr, rmodel.Repr)):
def rtype_getitem((r_dict, r_key), hop):
v_dict, v_key = hop.inputargs(r_dict, r_dict.key_repr)
@@ -338,7 +338,7 @@
def rtype_delitem((r_dict, r_key), hop):
v_dict, v_key = hop.inputargs(r_dict, r_dict.key_repr)
if not r_dict.custom_eq_hash:
- hop.has_implicit_exception(KeyError) # record that we know about it
+ hop.has_implicit_exception(KeyError) # record that we know about it
hop.exception_is_here()
return hop.gendirectcall(ll_dict_delitem, v_dict, v_key)
@@ -354,11 +354,11 @@
v_dict, v_key = hop.inputargs(r_dict, r_dict.key_repr)
hop.exception_is_here()
return hop.gendirectcall(ll_contains, v_dict, v_key)
-
+
class __extend__(pairtype(DictRepr, DictRepr)):
def convert_from_to((r_dict1, r_dict2), v, llops):
# check that we don't convert from Dicts with
- # different key/value types
+ # different key/value types
if r_dict1.dictkey is None or r_dict2.dictkey is None:
return NotImplemented
if r_dict1.dictkey is not r_dict2.dictkey:
@@ -430,7 +430,7 @@
return hlinvoke(DICT.r_rdict_eqfn, d.fnkeyeq, key1, key2)
def ll_dict_len(d):
- return d.num_items
+ return d.num_items
def ll_dict_is_true(d):
# check if a dict is True, allowing for None
@@ -491,8 +491,8 @@
if i & HIGHEST_BIT:
raise KeyError
_ll_dict_del(d, i)
-ll_dict_delitem.oopspec = 'dict.delitem(d, key)'
+ at jit.dont_look_inside
def _ll_dict_del(d, i):
d.entries.mark_deleted(i)
d.num_items -= 1
@@ -510,7 +510,7 @@
def ll_dict_resize(d):
old_entries = d.entries
- old_size = len(old_entries)
+ old_size = len(old_entries)
# make a 'new_size' estimate and shrink it if there are many
# deleted entry markers
new_size = old_size * 2
@@ -538,7 +538,7 @@
direct_compare = not hasattr(ENTRIES, 'no_direct_compare')
mask = len(entries) - 1
i = hash & mask
- # do the first try before any looping
+ # do the first try before any looping
if entries.valid(i):
checkingkey = entries[i].key
if direct_compare and checkingkey == key:
@@ -562,8 +562,8 @@
# In the loop, a deleted entry (everused and not valid) is by far
# (factor of 100s) the least likely outcome, so test for that last.
- perturb = r_uint(hash)
- while 1:
+ perturb = r_uint(hash)
+ while 1:
# compute the next index using unsigned arithmetic
i = r_uint(i)
i = (i << 2) + i + perturb + 1
@@ -591,7 +591,7 @@
if found:
return i # found the entry
elif freeslot == -1:
- freeslot = i
+ freeslot = i
perturb >>= PERTURB_SHIFT
def ll_dict_lookup_clean(d, hash):
@@ -601,7 +601,7 @@
entries = d.entries
mask = len(entries) - 1
i = hash & mask
- perturb = r_uint(hash)
+ perturb = r_uint(hash)
while entries.everused(i):
i = r_uint(i)
i = (i << 2) + i + perturb + 1
@@ -687,7 +687,6 @@
iter.dict = d
iter.index = 0
return iter
-ll_dictiter.oopspec = 'newdictiter(d)'
def _make_ll_dictnext(kind):
# make three versions of the following function: keys, values, items
diff --git a/pypy/rpython/lltypesystem/rffi.py b/pypy/rpython/lltypesystem/rffi.py
--- a/pypy/rpython/lltypesystem/rffi.py
+++ b/pypy/rpython/lltypesystem/rffi.py
@@ -789,8 +789,7 @@
# char* and size -> str (which can contain null bytes)
def charpsize2str(cp, size):
b = builder_class(size)
- for i in xrange(size):
- b.append(cp[i])
+ b.append_charpsize(cp, size)
return b.build()
charpsize2str._annenforceargs_ = [None, int]
@@ -1062,3 +1061,11 @@
keep_unicodebuffer_alive_until_here(self.raw, self.gc_buf)
def str(self, length):
return unicode_from_buffer(self.raw, self.gc_buf, self.size, length)
+
+# You would have to have a *huge* amount of data for this to block long enough
+# to be worth it to release the GIL.
+c_memcpy = llexternal("memcpy",
+ [VOIDP, VOIDP, SIZE_T],
+ lltype.Void,
+ threadsafe=False
+)
\ No newline at end of file
diff --git a/pypy/rpython/lltypesystem/rlist.py b/pypy/rpython/lltypesystem/rlist.py
--- a/pypy/rpython/lltypesystem/rlist.py
+++ b/pypy/rpython/lltypesystem/rlist.py
@@ -14,7 +14,6 @@
from pypy.rpython.lltypesystem import rstr
from pypy.rpython import robject
from pypy.rlib.debug import ll_assert
-from pypy.rlib.rarithmetic import ovfcheck
from pypy.rpython.lltypesystem import rffi
from pypy.rpython.lltypesystem.lloperation import llop
from pypy.rlib import rgc
@@ -200,12 +199,11 @@
else:
some = 6
some += newsize >> 3
- try:
- new_allocated = ovfcheck(newsize + some)
- except OverflowError:
- raise MemoryError
+ new_allocated = newsize + some
# new_allocated is a bit more than newsize, enough to ensure an amortized
- # linear complexity for e.g. repeated usage of l.append().
+ # linear complexity for e.g. repeated usage of l.append(). In case
+ # it overflows sys.maxint, it is guaranteed negative, and the following
+ # malloc() will fail.
items = l.items
newitems = malloc(typeOf(l).TO.items.TO, new_allocated)
before_len = l.length
diff --git a/pypy/rpython/lltypesystem/rstr.py b/pypy/rpython/lltypesystem/rstr.py
--- a/pypy/rpython/lltypesystem/rstr.py
+++ b/pypy/rpython/lltypesystem/rstr.py
@@ -345,6 +345,8 @@
def ll_strconcat(s1, s2):
len1 = len(s1.chars)
len2 = len(s2.chars)
+ # a single '+' like this is allowed to overflow: it gets
+ # a negative result, and the gc will complain
newstr = s1.malloc(len1 + len2)
s1.copy_contents(s1, newstr, 0, 0, len1)
s1.copy_contents(s2, newstr, 0, len1, len2)
@@ -412,9 +414,18 @@
itemslen = 0
i = 0
while i < num_items:
- itemslen += len(items[i].chars)
+ try:
+ itemslen = ovfcheck(itemslen + len(items[i].chars))
+ except OverflowError:
+ raise MemoryError
i += 1
- result = s.malloc(itemslen + s_len * (num_items - 1))
+ try:
+ seplen = ovfcheck(s_len * (num_items - 1))
+ except OverflowError:
+ raise MemoryError
+ # a single '+' at the end is allowed to overflow: it gets
+ # a negative result, and the gc will complain
+ result = s.malloc(itemslen + seplen)
res_index = len(items[0].chars)
s.copy_contents(items[0], result, 0, 0, res_index)
i = 1
@@ -688,7 +699,10 @@
itemslen = 0
i = 0
while i < num_items:
- itemslen += len(items[i].chars)
+ try:
+ itemslen = ovfcheck(itemslen + len(items[i].chars))
+ except OverflowError:
+ raise MemoryError
i += 1
if typeOf(items).TO.OF.TO == STR:
malloc = mallocstr
diff --git a/pypy/rpython/memory/gc/minimark.py b/pypy/rpython/memory/gc/minimark.py
--- a/pypy/rpython/memory/gc/minimark.py
+++ b/pypy/rpython/memory/gc/minimark.py
@@ -49,6 +49,7 @@
from pypy.rpython.lltypesystem.llmemory import raw_malloc_usage
from pypy.rpython.memory.gc.base import GCBase, MovingGCBase
from pypy.rpython.memory.gc import minimarkpage, env
+from pypy.rpython.memory.support import mangle_hash
from pypy.rlib.rarithmetic import ovfcheck, LONG_BIT, intmask, r_uint
from pypy.rlib.rarithmetic import LONG_BIT_SHIFT
from pypy.rlib.debug import ll_assert, debug_print, debug_start, debug_stop
@@ -389,6 +390,11 @@
# initialize the threshold
self.min_heap_size = max(self.min_heap_size, self.nursery_size *
self.major_collection_threshold)
+ # the following two values are usually equal, but during raw mallocs
+ # of arrays, next_major_collection_threshold is decremented to make
+ # the next major collection arrive earlier.
+ # See translator/c/test/test_newgc, test_nongc_attached_to_gc
+ self.next_major_collection_initial = self.min_heap_size
self.next_major_collection_threshold = self.min_heap_size
self.set_major_threshold_from(0.0)
debug_stop("gc-set-nursery-size")
@@ -396,7 +402,7 @@
def set_major_threshold_from(self, threshold, reserving_size=0):
# Set the next_major_collection_threshold.
- threshold_max = (self.next_major_collection_threshold *
+ threshold_max = (self.next_major_collection_initial *
self.growth_rate_max)
if threshold > threshold_max:
threshold = threshold_max
@@ -411,6 +417,7 @@
else:
bounded = False
#
+ self.next_major_collection_initial = threshold
self.next_major_collection_threshold = threshold
return bounded
@@ -510,17 +517,19 @@
# constant-folded because self.nonlarge_max, size and itemsize
# are all constants (the arguments are constant due to
# inlining).
- if not raw_malloc_usage(itemsize):
- too_many_items = raw_malloc_usage(nonvarsize) > self.nonlarge_max
+ maxsize = self.nonlarge_max - raw_malloc_usage(nonvarsize)
+ if maxsize < 0:
+ toobig = r_uint(0) # the nonvarsize alone is too big
+ elif raw_malloc_usage(itemsize):
+ toobig = r_uint(maxsize // raw_malloc_usage(itemsize)) + 1
else:
- maxlength = self.nonlarge_max - raw_malloc_usage(nonvarsize)
- maxlength = maxlength // raw_malloc_usage(itemsize)
- too_many_items = length > maxlength
+ toobig = r_uint(sys.maxint) + 1
- if too_many_items:
+ if r_uint(length) >= r_uint(toobig):
#
# If the total size of the object would be larger than
- # 'nonlarge_max', then allocate it externally.
+ # 'nonlarge_max', then allocate it externally. We also
+ # go there if 'length' is actually negative.
obj = self.external_malloc(typeid, length)
#
else:
@@ -603,13 +612,18 @@
# this includes the case of fixed-size objects, for which we
# should not even ask for the varsize_item_sizes().
totalsize = nonvarsize
- else:
+ elif length > 0:
+ # var-sized allocation with at least one item
itemsize = self.varsize_item_sizes(typeid)
try:
varsize = ovfcheck(itemsize * length)
totalsize = ovfcheck(nonvarsize + varsize)
except OverflowError:
raise MemoryError
+ else:
+ # negative length! This likely comes from an overflow
+ # earlier. We will just raise MemoryError here.
+ raise MemoryError
#
# If somebody calls this function a lot, we must eventually
# force a full collection.
@@ -717,9 +731,18 @@
def set_max_heap_size(self, size):
self.max_heap_size = float(size)
if self.max_heap_size > 0.0:
+ if self.max_heap_size < self.next_major_collection_initial:
+ self.next_major_collection_initial = self.max_heap_size
if self.max_heap_size < self.next_major_collection_threshold:
self.next_major_collection_threshold = self.max_heap_size
+ def raw_malloc_memory_pressure(self, sizehint):
+ self.next_major_collection_threshold -= sizehint
+ if self.next_major_collection_threshold < 0:
+ # cannot trigger a full collection now, but we can ensure
+ # that one will occur very soon
+ self.nursery_free = self.nursery_top
+
def can_malloc_nonmovable(self):
return True
@@ -1599,7 +1622,7 @@
# Max heap size: gives an upper bound on the threshold. If we
# already have at least this much allocated, raise MemoryError.
if bounded and (float(self.get_total_memory_used()) + reserving_size >=
- self.next_major_collection_threshold):
+ self.next_major_collection_initial):
#
# First raise MemoryError, giving the program a chance to
# quit cleanly. It might still allocate in the nursery,
@@ -1732,7 +1755,7 @@
# ----------
# id() and identityhash() support
- def id_or_identityhash(self, gcobj, special_case_prebuilt):
+ def id_or_identityhash(self, gcobj, is_hash):
"""Implement the common logic of id() and identityhash()
of an object, given as a GCREF.
"""
@@ -1775,7 +1798,7 @@
# The answer is the address of the shadow.
obj = shadow
#
- elif special_case_prebuilt:
+ elif is_hash:
if self.header(obj).tid & GCFLAG_HAS_SHADOW:
#
# For identityhash(), we need a special case for some
@@ -1784,10 +1807,14 @@
# after the object. But we cannot use it for id()
# because the stored value might clash with a real one.
size = self.get_size(obj)
- return (obj + size).signed[0]
+ i = (obj + size).signed[0]
+ # Important: the returned value is not mangle_hash()ed!
+ return i
#
- return llmemory.cast_adr_to_int(obj)
-
+ i = llmemory.cast_adr_to_int(obj)
+ if is_hash:
+ i = mangle_hash(i)
+ return i
def id(self, gcobj):
return self.id_or_identityhash(gcobj, False)
diff --git a/pypy/rpython/memory/gc/test/test_minimark.py b/pypy/rpython/memory/gc/test/test_minimark.py
--- a/pypy/rpython/memory/gc/test/test_minimark.py
+++ b/pypy/rpython/memory/gc/test/test_minimark.py
@@ -34,6 +34,7 @@
growth_rate_max=1.5)
gc.min_heap_size = 100.0
gc.max_heap_size = 300.0
+ gc.next_major_collection_initial = 0.0
gc.next_major_collection_threshold = 0.0
# first, we don't grow past min_heap_size
for i in range(5):
diff --git a/pypy/rpython/memory/gctransform/framework.py b/pypy/rpython/memory/gctransform/framework.py
--- a/pypy/rpython/memory/gctransform/framework.py
+++ b/pypy/rpython/memory/gctransform/framework.py
@@ -386,6 +386,18 @@
else:
self.malloc_varsize_nonmovable_ptr = None
+ if getattr(GCClass, 'raw_malloc_memory_pressure', False):
+ def raw_malloc_memory_pressure(length, itemsize):
+ totalmem = length * itemsize
+ if totalmem > 0:
+ gcdata.gc.raw_malloc_memory_pressure(totalmem)
+ #else: probably an overflow -- the following rawmalloc
+ # will fail then
+ self.raw_malloc_memory_pressure_ptr = getfn(
+ raw_malloc_memory_pressure,
+ [annmodel.SomeInteger(), annmodel.SomeInteger()],
+ annmodel.s_None, minimal_transform = False)
+
self.identityhash_ptr = getfn(GCClass.identityhash.im_func,
[s_gc, s_gcref],
annmodel.SomeInteger(),
@@ -525,7 +537,8 @@
self.c_vtinfo_skip_offset = rmodel.inputconst(lltype.typeOf(sko), sko)
def build_root_walker(self):
- return ShadowStackRootWalker(self)
+ from pypy.rpython.memory.gctransform import shadowstack
+ return shadowstack.ShadowStackRootWalker(self)
def consider_constant(self, TYPE, value):
self.layoutbuilder.consider_constant(TYPE, value, self.gcdata.gc)
@@ -932,10 +945,10 @@
def gct_gc_identityhash(self, hop):
livevars = self.push_roots(hop)
[v_ptr] = hop.spaceop.args
- v_adr = hop.genop("cast_ptr_to_adr", [v_ptr],
- resulttype=llmemory.Address)
+ v_ptr = hop.genop("cast_opaque_ptr", [v_ptr],
+ resulttype=llmemory.GCREF)
hop.genop("direct_call",
- [self.identityhash_ptr, self.c_const_gc, v_adr],
+ [self.identityhash_ptr, self.c_const_gc, v_ptr],
resultvar=hop.spaceop.result)
self.pop_roots(hop, livevars)
@@ -1323,217 +1336,3 @@
def need_thread_support(self, gctransformer, getfn):
raise Exception("%s does not support threads" % (
self.__class__.__name__,))
-
-
-class ShadowStackRootWalker(BaseRootWalker):
- need_root_stack = True
- collect_stacks_from_other_threads = None
-
- def __init__(self, gctransformer):
- BaseRootWalker.__init__(self, gctransformer)
- self.rootstacksize = sizeofaddr * gctransformer.root_stack_depth
- # NB. 'self' is frozen, but we can use self.gcdata to store state
- gcdata = self.gcdata
-
- def incr_stack(n):
- top = gcdata.root_stack_top
- gcdata.root_stack_top = top + n*sizeofaddr
- return top
- self.incr_stack = incr_stack
-
- def decr_stack(n):
- top = gcdata.root_stack_top - n*sizeofaddr
- gcdata.root_stack_top = top
- return top
- self.decr_stack = decr_stack
-
- self.rootstackhook = gctransformer.root_stack_jit_hook
- if self.rootstackhook is None:
- def collect_stack_root(callback, gc, addr):
- if gc.points_to_valid_gc_object(addr):
- callback(gc, addr)
- return sizeofaddr
- self.rootstackhook = collect_stack_root
-
- def push_stack(self, addr):
- top = self.incr_stack(1)
- top.address[0] = addr
-
- def pop_stack(self):
- top = self.decr_stack(1)
- return top.address[0]
-
- def allocate_stack(self):
- return llmemory.raw_malloc(self.rootstacksize)
-
- def setup_root_walker(self):
- stackbase = self.allocate_stack()
- ll_assert(bool(stackbase), "could not allocate root stack")
- self.gcdata.root_stack_top = stackbase
- self.gcdata.root_stack_base = stackbase
- BaseRootWalker.setup_root_walker(self)
-
- def walk_stack_roots(self, collect_stack_root):
- gcdata = self.gcdata
- gc = self.gc
- rootstackhook = self.rootstackhook
- addr = gcdata.root_stack_base
- end = gcdata.root_stack_top
- while addr != end:
- addr += rootstackhook(collect_stack_root, gc, addr)
- if self.collect_stacks_from_other_threads is not None:
- self.collect_stacks_from_other_threads(collect_stack_root)
-
- def need_thread_support(self, gctransformer, getfn):
- from pypy.module.thread import ll_thread # xxx fish
- from pypy.rpython.memory.support import AddressDict
- from pypy.rpython.memory.support import copy_without_null_values
- gcdata = self.gcdata
- # the interfacing between the threads and the GC is done via
- # three completely ad-hoc operations at the moment:
- # gc_thread_prepare, gc_thread_run, gc_thread_die.
- # See docstrings below.
-
- def get_aid():
- """Return the thread identifier, cast to an (opaque) address."""
- return llmemory.cast_int_to_adr(ll_thread.get_ident())
-
- def thread_setup():
- """Called once when the program starts."""
- aid = get_aid()
- gcdata.main_thread = aid
- gcdata.active_thread = aid
- gcdata.thread_stacks = AddressDict() # {aid: root_stack_top}
- gcdata._fresh_rootstack = llmemory.NULL
- gcdata.dead_threads_count = 0
-
- def thread_prepare():
- """Called just before thread.start_new_thread(). This
- allocates a new shadow stack to be used by the future
- thread. If memory runs out, this raises a MemoryError
- (which can be handled by the caller instead of just getting
- ignored if it was raised in the newly starting thread).
- """
- if not gcdata._fresh_rootstack:
- gcdata._fresh_rootstack = self.allocate_stack()
- if not gcdata._fresh_rootstack:
- raise MemoryError
-
- def thread_run():
- """Called whenever the current thread (re-)acquired the GIL.
- This should ensure that the shadow stack installed in
- gcdata.root_stack_top/root_stack_base is the one corresponding
- to the current thread.
- """
- aid = get_aid()
- if gcdata.active_thread != aid:
- switch_shadow_stacks(aid)
-
- def thread_die():
- """Called just before the final GIL release done by a dying
- thread. After a thread_die(), no more gc operation should
- occur in this thread.
- """
- aid = get_aid()
- if aid == gcdata.main_thread:
- return # ignore calls to thread_die() in the main thread
- # (which can occur after a fork()).
- gcdata.thread_stacks.setitem(aid, llmemory.NULL)
- old = gcdata.root_stack_base
- if gcdata._fresh_rootstack == llmemory.NULL:
- gcdata._fresh_rootstack = old
- else:
- llmemory.raw_free(old)
- install_new_stack(gcdata.main_thread)
- # from time to time, rehash the dictionary to remove
- # old NULL entries
- gcdata.dead_threads_count += 1
- if (gcdata.dead_threads_count & 511) == 0:
- copy = copy_without_null_values(gcdata.thread_stacks)
- gcdata.thread_stacks.delete()
- gcdata.thread_stacks = copy
-
- def switch_shadow_stacks(new_aid):
- save_away_current_stack()
- install_new_stack(new_aid)
- switch_shadow_stacks._dont_inline_ = True
-
- def save_away_current_stack():
- old_aid = gcdata.active_thread
- # save root_stack_base on the top of the stack
- self.push_stack(gcdata.root_stack_base)
- # store root_stack_top into the dictionary
- gcdata.thread_stacks.setitem(old_aid, gcdata.root_stack_top)
-
- def install_new_stack(new_aid):
- # look for the new stack top
- top = gcdata.thread_stacks.get(new_aid, llmemory.NULL)
- if top == llmemory.NULL:
- # first time we see this thread. It is an error if no
- # fresh new stack is waiting.
- base = gcdata._fresh_rootstack
- gcdata._fresh_rootstack = llmemory.NULL
- ll_assert(base != llmemory.NULL, "missing gc_thread_prepare")
- gcdata.root_stack_top = base
- gcdata.root_stack_base = base
- else:
- # restore the root_stack_base from the top of the stack
- gcdata.root_stack_top = top
- gcdata.root_stack_base = self.pop_stack()
- # done
- gcdata.active_thread = new_aid
-
- def collect_stack(aid, stacktop, callback):
- if stacktop != llmemory.NULL and aid != gcdata.active_thread:
- # collect all valid stacks from the dict (the entry
- # corresponding to the current thread is not valid)
- gc = self.gc
- rootstackhook = self.rootstackhook
- end = stacktop - sizeofaddr
- addr = end.address[0]
- while addr != end:
- addr += rootstackhook(callback, gc, addr)
-
- def collect_more_stacks(callback):
- ll_assert(get_aid() == gcdata.active_thread,
- "collect_more_stacks(): invalid active_thread")
- gcdata.thread_stacks.foreach(collect_stack, callback)
-
- def _free_if_not_current(aid, stacktop, _):
- if stacktop != llmemory.NULL and aid != gcdata.active_thread:
- end = stacktop - sizeofaddr
- base = end.address[0]
- llmemory.raw_free(base)
-
- def thread_after_fork(result_of_fork, opaqueaddr):
- # we don't need a thread_before_fork in this case, so
- # opaqueaddr == NULL. This is called after fork().
- if result_of_fork == 0:
- # We are in the child process. Assumes that only the
- # current thread survived, so frees the shadow stacks
- # of all the other ones.
- gcdata.thread_stacks.foreach(_free_if_not_current, None)
- # Clears the dict (including the current thread, which
- # was an invalid entry anyway and will be recreated by
- # the next call to save_away_current_stack()).
- gcdata.thread_stacks.clear()
- # Finally, reset the stored thread IDs, in case it
- # changed because of fork(). Also change the main
- # thread to the current one (because there is not any
- # other left).
- aid = get_aid()
- gcdata.main_thread = aid
- gcdata.active_thread = aid
-
- self.thread_setup = thread_setup
- self.thread_prepare_ptr = getfn(thread_prepare, [], annmodel.s_None)
- self.thread_run_ptr = getfn(thread_run, [], annmodel.s_None,
- inline=True)
- # no thread_start_ptr here
- self.thread_die_ptr = getfn(thread_die, [], annmodel.s_None)
- # no thread_before_fork_ptr here
- self.thread_after_fork_ptr = getfn(thread_after_fork,
- [annmodel.SomeInteger(),
- annmodel.SomeAddress()],
- annmodel.s_None)
- self.collect_stacks_from_other_threads = collect_more_stacks
diff --git a/pypy/rpython/memory/gctransform/shadowstack.py b/pypy/rpython/memory/gctransform/shadowstack.py
new file mode 100644
--- /dev/null
+++ b/pypy/rpython/memory/gctransform/shadowstack.py
@@ -0,0 +1,219 @@
+from pypy.rpython.memory.gctransform.framework import BaseRootWalker
+from pypy.rpython.memory.gctransform.framework import sizeofaddr
+from pypy.rlib.debug import ll_assert
+from pypy.rpython.lltypesystem import llmemory
+from pypy.annotation import model as annmodel
+
+
+class ShadowStackRootWalker(BaseRootWalker):
+ need_root_stack = True
+ collect_stacks_from_other_threads = None
+
+ def __init__(self, gctransformer):
+ BaseRootWalker.__init__(self, gctransformer)
+ self.rootstacksize = sizeofaddr * gctransformer.root_stack_depth
+ # NB. 'self' is frozen, but we can use self.gcdata to store state
+ gcdata = self.gcdata
+
+ def incr_stack(n):
+ top = gcdata.root_stack_top
+ gcdata.root_stack_top = top + n*sizeofaddr
+ return top
+ self.incr_stack = incr_stack
+
+ def decr_stack(n):
+ top = gcdata.root_stack_top - n*sizeofaddr
+ gcdata.root_stack_top = top
+ return top
+ self.decr_stack = decr_stack
+
+ self.rootstackhook = gctransformer.root_stack_jit_hook
+ if self.rootstackhook is None:
+ def collect_stack_root(callback, gc, addr):
+ if gc.points_to_valid_gc_object(addr):
+ callback(gc, addr)
+ return sizeofaddr
+ self.rootstackhook = collect_stack_root
+
+ def push_stack(self, addr):
+ top = self.incr_stack(1)
+ top.address[0] = addr
+
+ def pop_stack(self):
+ top = self.decr_stack(1)
+ return top.address[0]
+
+ def allocate_stack(self):
+ return llmemory.raw_malloc(self.rootstacksize)
+
+ def setup_root_walker(self):
+ stackbase = self.allocate_stack()
+ ll_assert(bool(stackbase), "could not allocate root stack")
+ self.gcdata.root_stack_top = stackbase
+ self.gcdata.root_stack_base = stackbase
+ BaseRootWalker.setup_root_walker(self)
+
+ def walk_stack_roots(self, collect_stack_root):
+ gcdata = self.gcdata
+ gc = self.gc
+ rootstackhook = self.rootstackhook
+ addr = gcdata.root_stack_base
+ end = gcdata.root_stack_top
+ while addr != end:
+ addr += rootstackhook(collect_stack_root, gc, addr)
+ if self.collect_stacks_from_other_threads is not None:
+ self.collect_stacks_from_other_threads(collect_stack_root)
+
+ def need_thread_support(self, gctransformer, getfn):
+ from pypy.module.thread import ll_thread # xxx fish
+ from pypy.rpython.memory.support import AddressDict
+ from pypy.rpython.memory.support import copy_without_null_values
+ gcdata = self.gcdata
+ # the interfacing between the threads and the GC is done via
+ # three completely ad-hoc operations at the moment:
+ # gc_thread_prepare, gc_thread_run, gc_thread_die.
+ # See docstrings below.
+
+ def get_aid():
+ """Return the thread identifier, cast to an (opaque) address."""
+ return llmemory.cast_int_to_adr(ll_thread.get_ident())
+
+ def thread_setup():
+ """Called once when the program starts."""
+ aid = get_aid()
+ gcdata.main_thread = aid
+ gcdata.active_thread = aid
+ gcdata.thread_stacks = AddressDict() # {aid: root_stack_top}
+ gcdata._fresh_rootstack = llmemory.NULL
+ gcdata.dead_threads_count = 0
+
+ def thread_prepare():
+ """Called just before thread.start_new_thread(). This
+ allocates a new shadow stack to be used by the future
+ thread. If memory runs out, this raises a MemoryError
+ (which can be handled by the caller instead of just getting
+ ignored if it was raised in the newly starting thread).
+ """
+ if not gcdata._fresh_rootstack:
+ gcdata._fresh_rootstack = self.allocate_stack()
+ if not gcdata._fresh_rootstack:
+ raise MemoryError
+
+ def thread_run():
+ """Called whenever the current thread (re-)acquired the GIL.
+ This should ensure that the shadow stack installed in
+ gcdata.root_stack_top/root_stack_base is the one corresponding
+ to the current thread.
+ """
+ aid = get_aid()
+ if gcdata.active_thread != aid:
+ switch_shadow_stacks(aid)
+
+ def thread_die():
+ """Called just before the final GIL release done by a dying
+ thread. After a thread_die(), no more gc operation should
+ occur in this thread.
+ """
+ aid = get_aid()
+ if aid == gcdata.main_thread:
+ return # ignore calls to thread_die() in the main thread
+ # (which can occur after a fork()).
+ gcdata.thread_stacks.setitem(aid, llmemory.NULL)
+ old = gcdata.root_stack_base
+ if gcdata._fresh_rootstack == llmemory.NULL:
+ gcdata._fresh_rootstack = old
+ else:
+ llmemory.raw_free(old)
+ install_new_stack(gcdata.main_thread)
+ # from time to time, rehash the dictionary to remove
+ # old NULL entries
+ gcdata.dead_threads_count += 1
+ if (gcdata.dead_threads_count & 511) == 0:
+ copy = copy_without_null_values(gcdata.thread_stacks)
+ gcdata.thread_stacks.delete()
+ gcdata.thread_stacks = copy
+
+ def switch_shadow_stacks(new_aid):
+ save_away_current_stack()
+ install_new_stack(new_aid)
+ switch_shadow_stacks._dont_inline_ = True
+
+ def save_away_current_stack():
+ old_aid = gcdata.active_thread
+ # save root_stack_base on the top of the stack
+ self.push_stack(gcdata.root_stack_base)
+ # store root_stack_top into the dictionary
+ gcdata.thread_stacks.setitem(old_aid, gcdata.root_stack_top)
+
+ def install_new_stack(new_aid):
+ # look for the new stack top
+ top = gcdata.thread_stacks.get(new_aid, llmemory.NULL)
+ if top == llmemory.NULL:
+ # first time we see this thread. It is an error if no
+ # fresh new stack is waiting.
+ base = gcdata._fresh_rootstack
+ gcdata._fresh_rootstack = llmemory.NULL
+ ll_assert(base != llmemory.NULL, "missing gc_thread_prepare")
+ gcdata.root_stack_top = base
+ gcdata.root_stack_base = base
+ else:
+ # restore the root_stack_base from the top of the stack
+ gcdata.root_stack_top = top
+ gcdata.root_stack_base = self.pop_stack()
+ # done
+ gcdata.active_thread = new_aid
+
+ def collect_stack(aid, stacktop, callback):
+ if stacktop != llmemory.NULL and aid != gcdata.active_thread:
+ # collect all valid stacks from the dict (the entry
+ # corresponding to the current thread is not valid)
+ gc = self.gc
+ rootstackhook = self.rootstackhook
+ end = stacktop - sizeofaddr
+ addr = end.address[0]
+ while addr != end:
+ addr += rootstackhook(callback, gc, addr)
+
+ def collect_more_stacks(callback):
+ ll_assert(get_aid() == gcdata.active_thread,
+ "collect_more_stacks(): invalid active_thread")
+ gcdata.thread_stacks.foreach(collect_stack, callback)
+
+ def _free_if_not_current(aid, stacktop, _):
+ if stacktop != llmemory.NULL and aid != gcdata.active_thread:
+ end = stacktop - sizeofaddr
+ base = end.address[0]
+ llmemory.raw_free(base)
+
+ def thread_after_fork(result_of_fork, opaqueaddr):
+ # we don't need a thread_before_fork in this case, so
+ # opaqueaddr == NULL. This is called after fork().
+ if result_of_fork == 0:
+ # We are in the child process. Assumes that only the
+ # current thread survived, so frees the shadow stacks
+ # of all the other ones.
+ gcdata.thread_stacks.foreach(_free_if_not_current, None)
+ # Clears the dict (including the current thread, which
+ # was an invalid entry anyway and will be recreated by
+ # the next call to save_away_current_stack()).
+ gcdata.thread_stacks.clear()
+ # Finally, reset the stored thread IDs, in case it
+ # changed because of fork(). Also change the main
+ # thread to the current one (because there is not any
+ # other left).
+ aid = get_aid()
+ gcdata.main_thread = aid
+ gcdata.active_thread = aid
+
+ self.thread_setup = thread_setup
+ self.thread_prepare_ptr = getfn(thread_prepare, [], annmodel.s_None)
+ self.thread_run_ptr = getfn(thread_run, [], annmodel.s_None,
+ inline=True)
+ # no thread_start_ptr here
+ self.thread_die_ptr = getfn(thread_die, [], annmodel.s_None)
+ # no thread_before_fork_ptr here
+ self.thread_after_fork_ptr = getfn(thread_after_fork,
+ [annmodel.SomeInteger(),
+ annmodel.SomeAddress()],
+ annmodel.s_None)
+ self.collect_stacks_from_other_threads = collect_more_stacks
diff --git a/pypy/rpython/memory/gctransform/transform.py b/pypy/rpython/memory/gctransform/transform.py
--- a/pypy/rpython/memory/gctransform/transform.py
+++ b/pypy/rpython/memory/gctransform/transform.py
@@ -590,6 +590,11 @@
def gct_fv_raw_malloc_varsize(self, hop, flags, TYPE, v_length, c_const_size, c_item_size,
c_offset_to_length):
+ if flags.get('add_memory_pressure', False):
+ if hasattr(self, 'raw_malloc_memory_pressure_ptr'):
+ hop.genop("direct_call",
+ [self.raw_malloc_memory_pressure_ptr,
+ v_length, c_item_size])
if c_offset_to_length is None:
if flags.get('zero'):
fnptr = self.raw_malloc_varsize_no_length_zero_ptr
diff --git a/pypy/rpython/memory/lldict.py b/pypy/rpython/memory/lldict.py
--- a/pypy/rpython/memory/lldict.py
+++ b/pypy/rpython/memory/lldict.py
@@ -1,6 +1,7 @@
from pypy.rpython.lltypesystem import lltype, llmemory
from pypy.rpython.lltypesystem import rdict
from pypy.rlib.objectmodel import we_are_translated
+from pypy.rpython.memory.support import mangle_hash
# This is a low-level AddressDict, reusing a lot of the logic from rdict.py.
# xxx this is very dependent on the details of rdict.py
@@ -40,7 +41,8 @@
lltype.free(entries, flavor="raw")
if not we_are_translated(): count_alloc(-1)
-_hash = llmemory.cast_adr_to_int
+def _hash(adr):
+ return mangle_hash(llmemory.cast_adr_to_int(adr))
def dict_keyhash(d, key):
return _hash(key)
diff --git a/pypy/rpython/memory/support.py b/pypy/rpython/memory/support.py
--- a/pypy/rpython/memory/support.py
+++ b/pypy/rpython/memory/support.py
@@ -4,6 +4,15 @@
from pypy.rlib.debug import ll_assert
from pypy.tool.identity_dict import identity_dict
+
+def mangle_hash(i):
+ # To hash pointers in dictionaries. Assumes that i shows some
+ # alignment (to 4, 8, maybe 16 bytes), so we use the following
+ # formula to avoid the trailing bits being always 0.
+ return i ^ (i >> 4)
+
+# ____________________________________________________________
+
DEFAULT_CHUNK_SIZE = 1019
diff --git a/pypy/rpython/rbuilder.py b/pypy/rpython/rbuilder.py
--- a/pypy/rpython/rbuilder.py
+++ b/pypy/rpython/rbuilder.py
@@ -36,6 +36,11 @@
hop.exception_cannot_occur()
return hop.gendirectcall(self.ll_append_multiple_char, *vlist)
+ def rtype_method_append_charpsize(self, hop):
+ vlist = hop.inputargs(self, self.raw_ptr_repr, lltype.Signed)
+ hop.exception_cannot_occur()
+ return hop.gendirectcall(self.ll_append_charpsize, *vlist)
+
def rtype_method_getlength(self, hop):
vlist = hop.inputargs(self)
hop.exception_cannot_occur()
diff --git a/pypy/rpython/rbuiltin.py b/pypy/rpython/rbuiltin.py
--- a/pypy/rpython/rbuiltin.py
+++ b/pypy/rpython/rbuiltin.py
@@ -345,14 +345,17 @@
BUILTIN_TYPER[object.__init__] = rtype_object__init__
# annotation of low-level types
-def rtype_malloc(hop, i_flavor=None, i_zero=None, i_track_allocation=None):
+def rtype_malloc(hop, i_flavor=None, i_zero=None, i_track_allocation=None,
+ i_add_memory_pressure=None):
assert hop.args_s[0].is_constant()
vlist = [hop.inputarg(lltype.Void, arg=0)]
opname = 'malloc'
- v_flavor, v_zero, v_track_allocation = parse_kwds(hop,
+ v_flavor, v_zero, v_track_allocation, v_add_memory_pressure = parse_kwds(
+ hop,
(i_flavor, lltype.Void),
(i_zero, None),
- (i_track_allocation, None))
+ (i_track_allocation, None),
+ (i_add_memory_pressure, None))
flags = {'flavor': 'gc'}
if v_flavor is not None:
@@ -361,8 +364,11 @@
flags['zero'] = v_zero.value
if i_track_allocation is not None:
flags['track_allocation'] = v_track_allocation.value
+ if i_add_memory_pressure is not None:
+ flags['add_memory_pressure'] = v_add_memory_pressure.value
vlist.append(hop.inputconst(lltype.Void, flags))
-
+
+ assert 1 <= hop.nb_args <= 2
if hop.nb_args == 2:
vlist.append(hop.inputarg(lltype.Signed, arg=1))
opname += '_varsize'
diff --git a/pypy/rpython/rptr.py b/pypy/rpython/rptr.py
--- a/pypy/rpython/rptr.py
+++ b/pypy/rpython/rptr.py
@@ -22,7 +22,7 @@
class __extend__(annmodel.SomeInteriorPtr):
def rtyper_makerepr(self, rtyper):
return InteriorPtrRepr(self.ll_ptrtype)
-
+
class PtrRepr(Repr):
@@ -91,7 +91,7 @@
vlist = hop.inputargs(*hop.args_r)
nexpected = len(self.lowleveltype.TO.ARGS)
nactual = len(vlist)-1
- if nactual != nexpected:
+ if nactual != nexpected:
raise TyperError("argcount mismatch: expected %d got %d" %
(nexpected, nactual))
if isinstance(vlist[0], flowmodel.Constant):
@@ -111,7 +111,12 @@
hop.swap_fst_snd_args()
hop.r_s_popfirstarg()
return self.rtype_simple_call(hop)
-
+
+class __extend__(pairtype(PtrRepr, PtrRepr)):
+ def convert_from_to((r_ptr1, r_ptr2), v, llop):
+ assert r_ptr1.lowleveltype == r_ptr2.lowleveltype
+ return v
+
class __extend__(pairtype(PtrRepr, IntegerRepr)):
@@ -205,7 +210,7 @@
self.lowleveltype = adtmeth.ll_ptrtype
self.ll_ptrtype = adtmeth.ll_ptrtype
self.lowleveltype = rtyper.getrepr(annmodel.lltype_to_annotation(adtmeth.ll_ptrtype)).lowleveltype
-
+
def rtype_simple_call(self, hop):
hop2 = hop.copy()
func = self.func
@@ -242,7 +247,7 @@
if numitemoffsets > 0:
self.lowleveltype = lltype.Ptr(self.parentptrtype._interior_ptr_type_with_index(self.resulttype.TO))
else:
- self.lowleveltype = self.parentptrtype
+ self.lowleveltype = self.parentptrtype
def getinteriorfieldargs(self, hop, v_self):
vlist = []
@@ -305,7 +310,7 @@
class __extend__(pairtype(InteriorPtrRepr, IntegerRepr)):
- def rtype_getitem((r_ptr, r_item), hop):
+ def rtype_getitem((r_ptr, r_item), hop):
ARRAY = r_ptr.resulttype.TO
ITEM_TYPE = ARRAY.OF
if isinstance(ITEM_TYPE, lltype.ContainerType):
@@ -325,7 +330,7 @@
vlist = r_ptr.getinteriorfieldargs(hop, v_self) + [v_index]
return hop.genop('getinteriorfield', vlist,
resulttype=ITEM_TYPE)
-
+
def rtype_setitem((r_ptr, r_index), hop):
ARRAY = r_ptr.resulttype.TO
ITEM_TYPE = ARRAY.OF
@@ -333,11 +338,11 @@
v_self, v_index, v_value = hop.inputargs(r_ptr, lltype.Signed, hop.args_r[2])
vlist = r_ptr.getinteriorfieldargs(hop, v_self) + [v_index, v_value]
hop.genop('setinteriorfield', vlist)
-
+
class __extend__(pairtype(InteriorPtrRepr, LLADTMethRepr)):
def convert_from_to((r_from, r_to), v, llops):
if r_from.lowleveltype == r_to.lowleveltype:
return v
return NotImplemented
-
+
diff --git a/pypy/rpython/test/test_rbuilder.py b/pypy/rpython/test/test_rbuilder.py
--- a/pypy/rpython/test/test_rbuilder.py
+++ b/pypy/rpython/test/test_rbuilder.py
@@ -1,8 +1,10 @@
import py
+
+from pypy.rlib.rstring import StringBuilder, UnicodeBuilder
+from pypy.rpython.annlowlevel import llstr, hlstr
+from pypy.rpython.lltypesystem import rffi
+from pypy.rpython.lltypesystem.rbuilder import *
from pypy.rpython.test.tool import BaseRtypingTest, LLRtypeMixin, OORtypeMixin
-from pypy.rpython.lltypesystem.rbuilder import *
-from pypy.rpython.annlowlevel import llstr, hlstr
-from pypy.rlib.rstring import StringBuilder, UnicodeBuilder
class TestStringBuilderDirect(object):
@@ -73,6 +75,15 @@
res = self.interpret(func, [])
assert res == 4
+ def test_append_charpsize(self):
+ def func(l):
+ s = StringBuilder()
+ with rffi.scoped_str2charp("hello world") as x:
+ s.append_charpsize(x, l)
+ return s.build()
+ res = self.ll_to_string(self.interpret(func, [5]))
+ assert res == "hello"
+
class TestLLtype(BaseTestStringBuilder, LLRtypeMixin):
pass
@@ -81,3 +92,5 @@
py.test.skip("getlength(): not implemented on ootype")
def test_unicode_getlength(self):
py.test.skip("getlength(): not implemented on ootype")
+ def test_append_charpsize(self):
+ py.test.skip("append_charpsize(): not implemented on ootype")
\ No newline at end of file
diff --git a/pypy/jit/codewriter/regalloc.py b/pypy/tool/algo/regalloc.py
copy from pypy/jit/codewriter/regalloc.py
copy to pypy/tool/algo/regalloc.py
--- a/pypy/jit/codewriter/regalloc.py
+++ b/pypy/tool/algo/regalloc.py
@@ -2,13 +2,11 @@
from pypy.objspace.flow.model import Variable
from pypy.tool.algo.color import DependencyGraph
from pypy.tool.algo.unionfind import UnionFind
-from pypy.jit.metainterp.history import getkind
-from pypy.jit.codewriter.flatten import ListOfKind
-def perform_register_allocation(graph, kind):
+def perform_register_allocation(graph, consider_var, ListOfKind=()):
"""Perform register allocation for the Variables of the given 'kind'
in the 'graph'."""
- regalloc = RegAllocator(graph, kind)
+ regalloc = RegAllocator(graph, consider_var, ListOfKind)
regalloc.make_dependencies()
regalloc.coalesce_variables()
regalloc.find_node_coloring()
@@ -18,9 +16,10 @@
class RegAllocator(object):
DEBUG_REGALLOC = False
- def __init__(self, graph, kind):
+ def __init__(self, graph, consider_var, ListOfKind):
self.graph = graph
- self.kind = kind
+ self.consider_var = consider_var
+ self.ListOfKind = ListOfKind
def make_dependencies(self):
dg = DependencyGraph()
@@ -31,7 +30,7 @@
for v in op.args:
if isinstance(v, Variable):
die_at[v] = i
- elif isinstance(v, ListOfKind):
+ elif isinstance(v, self.ListOfKind):
for v1 in v:
if isinstance(v1, Variable):
die_at[v1] = i
@@ -51,7 +50,7 @@
# Done. XXX the code above this line runs 3 times
# (for kind in KINDS) to produce the same result...
livevars = [v for v in block.inputargs
- if getkind(v.concretetype) == self.kind]
+ if self.consider_var(v)]
# Add the variables of this block to the dependency graph
for i, v in enumerate(livevars):
dg.add_node(v)
@@ -67,10 +66,10 @@
pass
die_index += 1
if (op.result is not None and
- getkind(op.result.concretetype) == self.kind):
+ self.consider_var(op.result)):
dg.add_node(op.result)
for v in livevars:
- if getkind(v.concretetype) == self.kind:
+ if self.consider_var(v):
dg.add_edge(v, op.result)
livevars.add(op.result)
self._depgraph = dg
@@ -95,8 +94,8 @@
self._try_coalesce(v, link.target.inputargs[i])
def _try_coalesce(self, v, w):
- if isinstance(v, Variable) and getkind(v.concretetype) == self.kind:
- assert getkind(w.concretetype) == self.kind
+ if isinstance(v, Variable) and self.consider_var(v):
+ assert self.consider_var(w)
dg = self._depgraph
uf = self._unionfind
v0 = uf.find_rep(v)
diff --git a/pypy/tool/gdb_pypy.py b/pypy/tool/gdb_pypy.py
new file mode 100644
--- /dev/null
+++ b/pypy/tool/gdb_pypy.py
@@ -0,0 +1,125 @@
+"""
+Some convenience macros for gdb. If you have pypy in your path, you can simply do:
+
+(gdb) python import pypy.tool.gdb_pypy
+
+Or, alternatively:
+
+(gdb) python execfile('/path/to/gdb_pypy.py')
+"""
+
+from __future__ import with_statement
+
+import sys
+import os.path
+
+try:
+ # when running inside gdb
+ from gdb import Command
+except ImportError:
+ # whenn running outside gdb: mock class for testing
+ class Command(object):
+ def __init__(self, name, command_class):
+ pass
+
+
+def find_field_with_suffix(val, suffix):
+ """
+ Return ``val[field]``, where ``field`` is the only one whose name ends
+ with ``suffix``. If there is no such field, or more than one, raise KeyError.
+ """
+ names = []
+ for field in val.type.fields():
+ if field.name.endswith(suffix):
+ names.append(field.name)
+ #
+ if len(names) == 1:
+ return val[names[0]]
+ elif len(names) == 0:
+ raise KeyError, "cannot find field *%s" % suffix
+ else:
+ raise KeyError, "too many matching fields: %s" % ', '.join(names)
+
+def lookup(val, suffix):
+ """
+ Lookup a field which ends with ``suffix`` following the rpython struct
+ inheritance hierarchy (i.e., looking both at ``val`` and
+ ``val['*_super']``, recursively.
+ """
+ try:
+ return find_field_with_suffix(val, suffix)
+ except KeyError:
+ baseobj = find_field_with_suffix(val, '_super')
+ return lookup(baseobj, suffix)
+
+
+class RPyType(Command):
+ """
+ Prints the RPython type of the expression (remember to dereference it!)
+ It assumes to find ``typeids.txt`` in the current directory.
+ E.g.:
+
+ (gdb) rpy_type *l_v123
+ GcStruct pypy.foo.Bar { super, inst_xxx, inst_yyy }
+ """
+
+ prog2typeids = {}
+
+ def __init__(self, gdb=None):
+ # dependency injection, for tests
+ if gdb is None:
+ import gdb
+ self.gdb = gdb
+ Command.__init__(self, "rpy_type", self.gdb.COMMAND_NONE)
+
+ def invoke(self, arg, from_tty):
+ # some magic code to automatically reload the python file while developing
+ ## from pypy.tool import gdb_pypy
+ ## reload(gdb_pypy)
+ ## gdb_pypy.RPyType.prog2typeids = self.prog2typeids # persist the cache
+ ## self.__class__ = gdb_pypy.RPyType
+ print self.do_invoke(arg, from_tty)
+
+ def do_invoke(self, arg, from_tty):
+ obj = self.gdb.parse_and_eval(arg)
+ hdr = lookup(obj, '_gcheader')
+ tid = hdr['h_tid']
+ offset = tid & 0xFFFFFFFF # 64bit only
+ offset = int(offset) # convert from gdb.Value to python int
+ typeids = self.get_typeids()
+ if offset in typeids:
+ return typeids[offset]
+ else:
+ return 'Cannot find the type with offset %d' % offset
+
+ def get_typeids(self):
+ progspace = self.gdb.current_progspace()
+ try:
+ return self.prog2typeids[progspace]
+ except KeyError:
+ typeids = self.load_typeids(progspace)
+ self.prog2typeids[progspace] = typeids
+ return typeids
+
+ def load_typeids(self, progspace):
+ """
+ Returns a mapping offset --> description
+ """
+ exename = progspace.filename
+ root = os.path.dirname(exename)
+ typeids_txt = os.path.join(root, 'typeids.txt')
+ print 'loading', typeids_txt
+ typeids = {}
+ with open(typeids_txt) as f:
+ for line in f:
+ member, descr = map(str.strip, line.split(None, 1))
+ expr = "((char*)(&pypy_g_typeinfo.%s)) - (char*)&pypy_g_typeinfo" % member
+ offset = int(self.gdb.parse_and_eval(expr))
+ typeids[offset] = descr
+ return typeids
+
+try:
+ import gdb
+ RPyType() # side effects
+except ImportError:
+ pass
diff --git a/pypy/tool/jitlogparser/parser.py b/pypy/tool/jitlogparser/parser.py
--- a/pypy/tool/jitlogparser/parser.py
+++ b/pypy/tool/jitlogparser/parser.py
@@ -330,6 +330,8 @@
if op.is_guard() and bridges.get('loop-' + str(op.guard_no), None):
res.append(op)
i = 0
+ if hasattr(op.bridge, 'force_asm'):
+ op.bridge.force_asm()
ops = op.bridge.operations
else:
res.append(op)
diff --git a/pypy/tool/logparser.py b/pypy/tool/logparser.py
--- a/pypy/tool/logparser.py
+++ b/pypy/tool/logparser.py
@@ -4,7 +4,8 @@
python logparser.py <action> <logfilename> <output> <options...>
Actions:
- draw-time draw a timeline image of the log (format PNG by default)
+ draw-time draw a timeline image of the log (format PNG by default)
+ print-summary print a summary of the log
"""
import autopath
import sys, re
@@ -385,6 +386,23 @@
else:
image.save(output)
+def print_summary(log, out):
+ totaltimes = gettotaltimes(log)
+ if out == '-':
+ outfile = sys.stdout
+ else:
+ outfile = open(out, "w")
+ l = totaltimes.items()
+ l.sort(cmp=lambda a, b: cmp(b[1], a[1]))
+ total = sum([b for a, b in l])
+ for a, b in l:
+ if a is None:
+ a = 'interpret'
+ s = " " * (50 - len(a))
+ print >>outfile, a, s, str(b*100/total) + "%"
+ if out != '-':
+ outfile.close()
+
# ____________________________________________________________
@@ -393,6 +411,7 @@
'mainwidth=', 'mainheight=',
'summarywidth=', 'summarybarheight=',
]),
+ 'print-summary': (print_summary, []),
}
if __name__ == '__main__':
diff --git a/pypy/tool/release/force-builds.py b/pypy/tool/release/force-builds.py
--- a/pypy/tool/release/force-builds.py
+++ b/pypy/tool/release/force-builds.py
@@ -22,7 +22,7 @@
# 'own-macosx-x86-32',
# 'pypy-c-app-level-linux-x86-32',
# 'pypy-c-app-level-linux-x86-64',
- 'pypy-c-stackless-app-level-linux-x86-32',
+# 'pypy-c-stackless-app-level-linux-x86-32',
'pypy-c-app-level-win-x86-32',
'pypy-c-jit-linux-x86-32',
'pypy-c-jit-linux-x86-64',
diff --git a/pypy/tool/test/test_gdb_pypy.py b/pypy/tool/test/test_gdb_pypy.py
new file mode 100644
--- /dev/null
+++ b/pypy/tool/test/test_gdb_pypy.py
@@ -0,0 +1,105 @@
+import py
+from pypy.tool import gdb_pypy
+
+class FakeGdb(object):
+
+ COMMAND_NONE = -1
+
+ def __init__(self, exprs, progspace=None):
+ self.exprs = exprs
+ self.progspace = progspace
+
+ def parse_and_eval(self, expr):
+ return self.exprs[expr]
+
+ def current_progspace(self):
+ return self.progspace
+
+
+class Mock(object):
+ def __init__(self, **attrs):
+ self.__dict__.update(attrs)
+
+class Field(Mock):
+ pass
+
+class Struct(object):
+ def __init__(self, fieldnames):
+ self._fields = [Field(name=name) for name in fieldnames]
+
+ def fields(self):
+ return self._fields[:]
+
+class Value(dict):
+ def __init__(self, *args, **kwds):
+ dict.__init__(self, *args, **kwds)
+ self.type = Struct(self.keys())
+ for key, val in self.iteritems():
+ if isinstance(val, dict):
+ self[key] = Value(val)
+
+def test_mock_objects():
+ d = {'a': 1,
+ 'b': 2,
+ 'super': {
+ 'c': 3,
+ }
+ }
+ val = Value(d)
+ assert val['a'] == 1
+ assert val['b'] == 2
+ assert isinstance(val['super'], Value)
+ assert val['super']['c'] == 3
+ fields = val.type.fields()
+ names = [f.name for f in fields]
+ assert sorted(names) == ['a', 'b', 'super']
+
+def test_find_field_with_suffix():
+ obj = Value(x_foo = 1,
+ y_bar = 2,
+ z_foobar = 3)
+ assert gdb_pypy.find_field_with_suffix(obj, 'foo') == 1
+ assert gdb_pypy.find_field_with_suffix(obj, 'foobar') == 3
+ py.test.raises(KeyError, "gdb_pypy.find_field_with_suffix(obj, 'bar')")
+ py.test.raises(KeyError, "gdb_pypy.find_field_with_suffix(obj, 'xxx')")
+
+def test_lookup():
+ d = {'r_super': {
+ '_gcheader': {
+ 'h_tid': 123,
+ }
+ },
+ 'r_foo': 42,
+ }
+ obj = Value(d)
+ assert gdb_pypy.lookup(obj, 'foo') == 42
+ hdr = gdb_pypy.lookup(obj, 'gcheader')
+ assert hdr['h_tid'] == 123
+
+def test_RPyType(tmpdir):
+ exe = tmpdir.join('pypy-c')
+ typeids = tmpdir.join('typeids.txt')
+ typeids.write("""
+member0 GcStruct xxx {}
+member1 GcStruct yyy {}
+member2 GcStruct zzz {}
+""".strip())
+ #
+ progspace = Mock(filename=str(exe))
+ d = {'r_super': {
+ '_gcheader': {
+ 'h_tid': 123,
+ }
+ },
+ 'r_foo': 42,
+ }
+ myvar = Value(d)
+ exprs = {
+ '*myvar': myvar,
+ '((char*)(&pypy_g_typeinfo.member0)) - (char*)&pypy_g_typeinfo': 0,
+ '((char*)(&pypy_g_typeinfo.member1)) - (char*)&pypy_g_typeinfo': 123,
+ '((char*)(&pypy_g_typeinfo.member2)) - (char*)&pypy_g_typeinfo': 456,
+ }
+ gdb = FakeGdb(exprs, progspace)
+ cmd = gdb_pypy.RPyType(gdb)
+ assert cmd.do_invoke('*myvar', True) == 'GcStruct yyy {}'
diff --git a/pypy/translator/c/src/cjkcodecs/multibytecodec.c b/pypy/translator/c/src/cjkcodecs/multibytecodec.c
--- a/pypy/translator/c/src/cjkcodecs/multibytecodec.c
+++ b/pypy/translator/c/src/cjkcodecs/multibytecodec.c
@@ -3,31 +3,38 @@
#include "src/cjkcodecs/multibytecodec.h"
-struct pypy_cjk_dec_s *pypy_cjk_dec_init(const MultibyteCodec *codec,
- char *inbuf, Py_ssize_t inlen)
+struct pypy_cjk_dec_s *pypy_cjk_dec_new(const MultibyteCodec *codec)
{
struct pypy_cjk_dec_s *d = malloc(sizeof(struct pypy_cjk_dec_s));
if (!d)
return NULL;
if (codec->decinit != NULL && codec->decinit(&d->state, codec->config) != 0)
- goto errorexit;
+ {
+ free(d);
+ return NULL;
+ }
+ d->codec = codec;
+ d->outbuf_start = NULL;
+ return d;
+}
- d->codec = codec;
+Py_ssize_t pypy_cjk_dec_init(struct pypy_cjk_dec_s *d,
+ char *inbuf, Py_ssize_t inlen)
+{
d->inbuf_start = inbuf;
d->inbuf = inbuf;
d->inbuf_end = inbuf + inlen;
- d->outbuf_start = (inlen <= (PY_SSIZE_T_MAX / sizeof(Py_UNICODE)) ?
- malloc(inlen * sizeof(Py_UNICODE)) :
- NULL);
- if (!d->outbuf_start)
- goto errorexit;
+ if (d->outbuf_start == NULL)
+ {
+ d->outbuf_start = (inlen <= (PY_SSIZE_T_MAX / sizeof(Py_UNICODE)) ?
+ malloc(inlen * sizeof(Py_UNICODE)) :
+ NULL);
+ if (d->outbuf_start == NULL)
+ return -1;
+ d->outbuf_end = d->outbuf_start + inlen;
+ }
d->outbuf = d->outbuf_start;
- d->outbuf_end = d->outbuf_start + inlen;
- return d;
-
- errorexit:
- free(d);
- return NULL;
+ return 0;
}
void pypy_cjk_dec_free(struct pypy_cjk_dec_s *d)
@@ -112,34 +119,40 @@
/************************************************************/
-struct pypy_cjk_enc_s *pypy_cjk_enc_init(const MultibyteCodec *codec,
- Py_UNICODE *inbuf, Py_ssize_t inlen)
+struct pypy_cjk_enc_s *pypy_cjk_enc_new(const MultibyteCodec *codec)
{
- Py_ssize_t outlen;
struct pypy_cjk_enc_s *d = malloc(sizeof(struct pypy_cjk_enc_s));
if (!d)
return NULL;
if (codec->encinit != NULL && codec->encinit(&d->state, codec->config) != 0)
- goto errorexit;
+ {
+ free(d);
+ return NULL;
+ }
+ d->codec = codec;
+ d->outbuf_start = NULL;
+ return d;
+}
- d->codec = codec;
+Py_ssize_t pypy_cjk_enc_init(struct pypy_cjk_enc_s *d,
+ Py_UNICODE *inbuf, Py_ssize_t inlen)
+{
+ Py_ssize_t outlen;
d->inbuf_start = inbuf;
d->inbuf = inbuf;
d->inbuf_end = inbuf + inlen;
-
- if (inlen > (PY_SSIZE_T_MAX - 16) / 2)
- goto errorexit;
- outlen = inlen * 2 + 16;
- d->outbuf_start = malloc(outlen);
- if (!d->outbuf_start)
- goto errorexit;
+ if (d->outbuf_start == NULL)
+ {
+ if (inlen > (PY_SSIZE_T_MAX - 16) / 2)
+ return -1;
+ outlen = inlen * 2 + 16;
+ d->outbuf_start = malloc(outlen);
+ if (d->outbuf_start == NULL)
+ return -1;
+ d->outbuf_end = d->outbuf_start + outlen;
+ }
d->outbuf = d->outbuf_start;
- d->outbuf_end = d->outbuf_start + outlen;
- return d;
-
- errorexit:
- free(d);
- return NULL;
+ return 0;
}
void pypy_cjk_enc_free(struct pypy_cjk_enc_s *d)
@@ -167,11 +180,8 @@
return 0;
}
-#define MBENC_RESET MBENC_MAX<<1
-
-Py_ssize_t pypy_cjk_enc_chunk(struct pypy_cjk_enc_s *d)
+Py_ssize_t pypy_cjk_enc_chunk(struct pypy_cjk_enc_s *d, Py_ssize_t flags)
{
- int flags = MBENC_FLUSH | MBENC_RESET; /* XXX always, for now */
while (1)
{
Py_ssize_t r;
@@ -242,3 +252,8 @@
d->inbuf = d->inbuf_start + in_offset;
return 0;
}
+
+const MultibyteCodec *pypy_cjk_enc_getcodec(struct pypy_cjk_enc_s *d)
+{
+ return d->codec;
+}
diff --git a/pypy/translator/c/src/cjkcodecs/multibytecodec.h b/pypy/translator/c/src/cjkcodecs/multibytecodec.h
--- a/pypy/translator/c/src/cjkcodecs/multibytecodec.h
+++ b/pypy/translator/c/src/cjkcodecs/multibytecodec.h
@@ -94,8 +94,9 @@
Py_UNICODE *outbuf_start, *outbuf, *outbuf_end;
};
-struct pypy_cjk_dec_s *pypy_cjk_dec_init(const MultibyteCodec *codec,
- char *inbuf, Py_ssize_t inlen);
+struct pypy_cjk_dec_s *pypy_cjk_dec_new(const MultibyteCodec *codec);
+Py_ssize_t pypy_cjk_dec_init(struct pypy_cjk_dec_s *d,
+ char *inbuf, Py_ssize_t inlen);
void pypy_cjk_dec_free(struct pypy_cjk_dec_s *);
Py_ssize_t pypy_cjk_dec_chunk(struct pypy_cjk_dec_s *);
Py_UNICODE *pypy_cjk_dec_outbuf(struct pypy_cjk_dec_s *);
@@ -112,10 +113,11 @@
unsigned char *outbuf_start, *outbuf, *outbuf_end;
};
-struct pypy_cjk_enc_s *pypy_cjk_enc_init(const MultibyteCodec *codec,
- Py_UNICODE *inbuf, Py_ssize_t inlen);
+struct pypy_cjk_enc_s *pypy_cjk_enc_new(const MultibyteCodec *codec);
+Py_ssize_t pypy_cjk_enc_init(struct pypy_cjk_enc_s *d,
+ Py_UNICODE *inbuf, Py_ssize_t inlen);
void pypy_cjk_enc_free(struct pypy_cjk_enc_s *);
-Py_ssize_t pypy_cjk_enc_chunk(struct pypy_cjk_enc_s *);
+Py_ssize_t pypy_cjk_enc_chunk(struct pypy_cjk_enc_s *, Py_ssize_t);
Py_ssize_t pypy_cjk_enc_reset(struct pypy_cjk_enc_s *);
char *pypy_cjk_enc_outbuf(struct pypy_cjk_enc_s *);
Py_ssize_t pypy_cjk_enc_outlen(struct pypy_cjk_enc_s *);
@@ -123,6 +125,7 @@
Py_ssize_t pypy_cjk_enc_inbuf_consumed(struct pypy_cjk_enc_s* d);
Py_ssize_t pypy_cjk_enc_replace_on_error(struct pypy_cjk_enc_s* d,
char *, Py_ssize_t, Py_ssize_t);
+const MultibyteCodec *pypy_cjk_enc_getcodec(struct pypy_cjk_enc_s *);
/* list of codecs defined in the .c files */
diff --git a/pypy/translator/c/test/test_newgc.py b/pypy/translator/c/test/test_newgc.py
--- a/pypy/translator/c/test/test_newgc.py
+++ b/pypy/translator/c/test/test_newgc.py
@@ -1390,6 +1390,35 @@
def test_gc_heap_stats(self):
py.test.skip("not implemented")
+ def define_nongc_attached_to_gc(cls):
+ from pypy.rpython.lltypesystem import rffi
+ ARRAY = rffi.CArray(rffi.INT)
+ class A:
+ def __init__(self, n):
+ self.buf = lltype.malloc(ARRAY, n, flavor='raw',
+ add_memory_pressure=True)
+ def __del__(self):
+ lltype.free(self.buf, flavor='raw')
+ A(6)
+ def f():
+ # allocate a total of ~77GB, but if the automatic gc'ing works,
+ # it should never need more than a few MBs at once
+ am1 = am2 = am3 = None
+ res = 0
+ for i in range(1, 100001):
+ if am3 is not None:
+ res += rffi.cast(lltype.Signed, am3.buf[0])
+ am3 = am2
+ am2 = am1
+ am1 = A(i * 4)
+ am1.buf[0] = rffi.cast(rffi.INT, i-50000)
+ return res
+ return f
+
+ def test_nongc_attached_to_gc(self):
+ res = self.run("nongc_attached_to_gc")
+ assert res == -99997
+
# ____________________________________________________________________
class TaggedPointersTest(object):
diff --git a/pypy/translator/jvm/src/pypy/PyPy.java b/pypy/translator/jvm/src/pypy/PyPy.java
--- a/pypy/translator/jvm/src/pypy/PyPy.java
+++ b/pypy/translator/jvm/src/pypy/PyPy.java
@@ -1100,9 +1100,9 @@
if (Double.isNaN(x))
return interlink.recordFloatSigned(x, 0);
- // Infinity: Python throws exception
+ // Infinity: Python returns (inf, 0)
if (Double.isInfinite(x))
- interlink.throwOverflowError();
+ return interlink.recordFloatSigned(x, 0);
// Extract the various parts of the format:
final long e=11, f=52; // number of bits in IEEE format
More information about the pypy-commit
mailing list