Author: Wim Lavrijsen <[email protected]>
Branch: cppyy-packaging
Changeset: r92721:63eea9e3fded
Date: 2017-10-06 14:24 -0700
http://bitbucket.org/pypy/pypy/changeset/63eea9e3fded/
Log: merge default into branch
diff --git a/.hgtags b/.hgtags
--- a/.hgtags
+++ b/.hgtags
@@ -40,3 +40,7 @@
2875f328eae2216a87f3d6f335092832eb031f56 release-pypy3.5-v5.7.1
c925e73810367cd960a32592dd7f728f436c125c release-pypy2.7-v5.8.0
a37ecfe5f142bc971a86d17305cc5d1d70abec64 release-pypy3.5-v5.8.0
+03d614975835870da65ff0481e1edad68ebbcb8d release-pypy2.7-v5.9.0
+d72f9800a42b46a8056951b1da2426d2c2d8d502 release-pypy3.5-v5.9.0
+03d614975835870da65ff0481e1edad68ebbcb8d release-pypy2.7-v5.9.0
+84a2f3e6a7f88f2fe698e473998755b3bd1a12e2 release-pypy2.7-v5.9.0
diff --git a/pypy/module/_cppyy/test/conftest.py
b/pypy/module/_cppyy/test/conftest.py
--- a/pypy/module/_cppyy/test/conftest.py
+++ b/pypy/module/_cppyy/test/conftest.py
@@ -4,7 +4,7 @@
def pytest_runtest_setup(item):
if py.path.local.sysfind('genreflex') is None:
import pypy.module._cppyy.capi.loadable_capi as lcapi
- if 'dummy' in lcapi.reflection_library:
+ if 'dummy' in lcapi.backend_library:
# run only tests that are covered by the dummy backend and tests
# that do not rely on reflex
import os
@@ -33,7 +33,7 @@
import pypy.module._cppyy.capi.loadable_capi as lcapi
try:
import ctypes
- ctypes.CDLL(lcapi.reflection_library)
+ ctypes.CDLL(lcapi.backend_library)
except Exception as e:
if config.option.runappdirect:
return # "can't run dummy tests in -A"
@@ -71,4 +71,4 @@
return
raise
- lcapi.reflection_library = str(soname)
+ lcapi.backend_library = str(soname)
diff --git a/rpython/rlib/rsiphash.py b/rpython/rlib/rsiphash.py
--- a/rpython/rlib/rsiphash.py
+++ b/rpython/rlib/rsiphash.py
@@ -11,6 +11,7 @@
from rpython.rlib.objectmodel import not_rpython, always_inline
from rpython.rlib.objectmodel import we_are_translated, dont_inline
from rpython.rlib.objectmodel import keepalive_until_here
+from rpython.rlib.objectmodel import specialize
from rpython.rlib import rgc, jit, rposix
from rpython.rlib.rarithmetic import r_uint64, r_uint32, r_uint
from rpython.rlib.rawstorage import misaligned_is_fine
@@ -26,9 +27,11 @@
return x
def _le32toh(x):
return x
+ BIG_ENDIAN = False
else:
_le64toh = rarithmetic.byteswap
_le32toh = rarithmetic.byteswap
+ BIG_ENDIAN = True
def _decode64(s):
return (r_uint64(ord(s[0])) |
@@ -49,6 +52,7 @@
"""'s' is a string of length 16"""
seed.k0l = _decode64(s)
seed.k1l = _decode64(s[8:16])
+ _update_prebuilt_hashes()
random_ctx = rurandom.init_urandom()
@@ -151,23 +155,20 @@
else:
# NOTE: a latin-1 unicode string must have the same hash as the
# corresponding byte string. If the unicode is all within
- # 0-255, then we need to allocate a byte buffer and copy the
- # latin-1 encoding in it manually. Note also that we give a
+ # 0-255, then we call _siphash24() with a special argument that
+ # will make it load only one byte for every unicode char.
+ # Note also that we give a
# different hash result than CPython on ucs4 platforms, for
# unicode strings where CPython uses 2 bytes per character.
+ addr = rstr._get_raw_buf_unicode(rstr.UNICODE, ll_s, 0)
+ SZ = rffi.sizeof(rstr.UNICODE.chars.OF)
for i in range(length):
if ord(ll_s.chars[i]) > 0xFF:
- addr = rstr._get_raw_buf_unicode(rstr.UNICODE, ll_s, 0)
- length *= rffi.sizeof(rstr.UNICODE.chars.OF)
+ length *= SZ
break
else:
- p = lltype.malloc(rffi.CCHARP.TO, length, flavor='raw')
- i = 0
- while i < length:
- p[i] = chr(ord(ll_s.chars[i]))
- i += 1
- x = _siphash24(llmemory.cast_ptr_to_adr(p), length)
- lltype.free(p, flavor='raw')
+ x = _siphash24(addr, length, SZ)
+ keepalive_until_here(ll_s)
return intmask(x)
x = _siphash24(addr, length)
keepalive_until_here(ll_s)
@@ -175,16 +176,22 @@
@contextmanager
-def choosen_seed(new_k0, new_k1, test_misaligned_path=False):
+def choosen_seed(new_k0, new_k1, test_misaligned_path=False,
+ test_prebuilt=False):
"""For tests."""
- global misaligned_is_fine
- old = seed.k0l, seed.k1l, misaligned_is_fine
+ global misaligned_is_fine, seed
+ old = seed, misaligned_is_fine
+ seed = Seed()
seed.k0l = r_uint64(new_k0)
seed.k1l = r_uint64(new_k1)
+ if test_prebuilt:
+ _update_prebuilt_hashes()
+ else:
+ seed.bound_prebuilt_size = 0
if test_misaligned_path:
misaligned_is_fine = False
yield
- seed.k0l, seed.k1l, misaligned_is_fine = old
+ seed, misaligned_is_fine = old
magic0 = r_uint64(0x736f6d6570736575)
magic1 = r_uint64(0x646f72616e646f6d)
@@ -215,10 +222,22 @@
@rgc.no_collect
-def _siphash24(addr_in, size):
[email protected](2)
+def _siphash24(addr_in, size, SZ=1):
"""Takes an address pointer and a size. Returns the hash as a r_uint64,
which can then be casted to the expected type."""
+ if BIG_ENDIAN:
+ index = SZ - 1
+ else:
+ index = 0
+ if size < seed.bound_prebuilt_size:
+ if size <= 0:
+ return seed.hash_empty
+ else:
+ t = rarithmetic.intmask(llop.raw_load(rffi.UCHAR, addr_in, index))
+ return seed.hash_single[t]
+
k0 = seed.k0l
k1 = seed.k1l
b = r_uint64(size) << 56
@@ -227,10 +246,10 @@
v2 = k0 ^ magic2
v3 = k1 ^ magic3
- direct = (misaligned_is_fine or
+ direct = (SZ == 1) and (misaligned_is_fine or
(rffi.cast(lltype.Signed, addr_in) & 7) == 0)
- index = 0
if direct:
+ assert SZ == 1
while size >= 8:
mi = llop.raw_load(rffi.ULONGLONG, addr_in, index)
mi = _le64toh(mi)
@@ -242,30 +261,30 @@
else:
while size >= 8:
mi = (
- r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index)) |
- r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 1)) << 8 |
- r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 2)) << 16 |
- r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 3)) << 24 |
- r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 4)) << 32 |
- r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 5)) << 40 |
- r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 6)) << 48 |
- r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 7)) << 56
+ r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index)) |
+ r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 1*SZ)) << 8 |
+ r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 2*SZ)) << 16
|
+ r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 3*SZ)) << 24
|
+ r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 4*SZ)) << 32
|
+ r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 5*SZ)) << 40
|
+ r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 6*SZ)) << 48
|
+ r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 7*SZ)) << 56
)
size -= 8
- index += 8
+ index += 8*SZ
v3 ^= mi
v0, v1, v2, v3 = _double_round(v0, v1, v2, v3)
v0 ^= mi
t = r_uint64(0)
if size == 7:
- t = r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 6)) << 48
+ t = r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 6*SZ)) << 48
size = 6
if size == 6:
- t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 5)) << 40
+ t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 5*SZ)) << 40
size = 5
if size == 5:
- t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 4)) << 32
+ t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 4*SZ)) << 32
size = 4
if size == 4:
if direct:
@@ -273,13 +292,13 @@
t |= r_uint64(v)
size = 0
else:
- t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 3)) << 24
+ t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 3*SZ))<<
24
size = 3
if size == 3:
- t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 2)) << 16
+ t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 2*SZ)) << 16
size = 2
if size == 2:
- t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 1)) << 8
+ t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index + 1*SZ)) << 8
size = 1
if size == 1:
t |= r_uint64(llop.raw_load(rffi.UCHAR, addr_in, index))
@@ -306,3 +325,17 @@
"""
with rffi.scoped_nonmovingbuffer(s) as p:
return _siphash24(llmemory.cast_ptr_to_adr(p), len(s))
+
+
+# Prebuilt hashes are precomputed here
+def _update_prebuilt_hashes():
+ seed.bound_prebuilt_size = 0
+ with lltype.scoped_alloc(rffi.CCHARP.TO, 1) as p:
+ addr = llmemory.cast_ptr_to_adr(p)
+ seed.hash_single = [r_uint64(0)] * 256
+ for i in range(256):
+ p[0] = chr(i)
+ seed.hash_single[i] = _siphash24(addr, 1)
+ seed.hash_empty = _siphash24(addr, 0)
+ seed.bound_prebuilt_size = 2
+_update_prebuilt_hashes()
diff --git a/rpython/rlib/test/test_rsiphash.py
b/rpython/rlib/test/test_rsiphash.py
--- a/rpython/rlib/test/test_rsiphash.py
+++ b/rpython/rlib/test/test_rsiphash.py
@@ -1,8 +1,10 @@
import os
from rpython.rlib.rsiphash import siphash24, _siphash24, choosen_seed
from rpython.rlib.rsiphash import initialize_from_env, enable_siphash24
+from rpython.rlib.rsiphash import ll_hash_string_siphash24
from rpython.rlib.objectmodel import compute_hash
from rpython.rlib.rarithmetic import intmask
+from rpython.rtyper.annlowlevel import llstr, llunicode
from rpython.rtyper.lltypesystem import llmemory, rffi
from rpython.translator.c.test.test_genc import compile
@@ -38,14 +40,30 @@
test_misaligned_path=True):
x = siphash24(s)
y = _siphash24(llmemory.cast_ptr_to_adr(rffi.ptradd(q, 1)), len(s))
+ z = ll_hash_string_siphash24(llstr(s))
rffi.free_charp(q)
assert x == y
+ assert z == intmask(x)
return x
def test_siphash24():
for expected, string in CASES:
assert check(string) == expected
+def check_latin1(s, expected, test_prebuilt=False):
+ with choosen_seed(0x8a9f065a358479f4, 0x11cb1e9ee7f40e1f,
+ test_misaligned_path=True, test_prebuilt=test_prebuilt):
+ z = ll_hash_string_siphash24(llunicode(s))
+ assert z == intmask(expected)
+
+def test_siphash24_latin1_unicode():
+ for expected, string in CASES:
+ check_latin1(string.decode('latin1'), expected)
+
+def test_siphash24_latin1_unicode_prebuilt():
+ for expected, string in CASES:
+ check_latin1(string.decode('latin1'), expected, test_prebuilt=True)
+
def test_fix_seed():
old_val = os.environ.get('PYTHONHASHSEED', None)
try:
_______________________________________________
pypy-commit mailing list
[email protected]
https://mail.python.org/mailman/listinfo/pypy-commit
