Re: [PATCH] Fix ODR violations in code using
On 05/07/19 19:44 +0100, Jonathan Wakely wrote:
On 05/07/19 20:23 +0200, Daniel Krügler wrote:
Am Fr., 5. Juli 2019 um 18:13 Uhr schrieb Jonathan Wakely :
[..]
I decided against the simplification in the second patch, and
committed the attached one which is closer to the first patch I sent
(preserving the __atomic_add and __exchange_and_add functions even
when they just call the built-ins).
Tested x86_64-linux, powerpc64-linux, powerpc-aix. Committed to trunk.
Unrelated to the actual patch, I noticed some explicit "throw()" forms
used as exception specifications - shouldn't these be replaced by
either explicit "noexcept" or at least by a library macro that expands
to one or the other?
Yes, they should be _GLIBCXX_NOTHROW.
(I'm sorry, if such unrelated questions are
considered as inappropriate for this list).
Entirely appropriate, thanks!
Here's a patch to fix that and the dumb mistake I made in
__atomic_add_dispatch. I'll commit after testing finishes.
commit 1542776ebab8848109d125d05a548fca5efb513a
Author: Jonathan Wakely
Date: Sat Jul 6 21:24:51 2019 +0100
Fix recent regression in __atomic_add_dispatch
* include/ext/atomicity.h (__exchange_and_add, __atomic_add): Replace
throw() with _GLIBCXX_NOTHROW.
(__atomic_add_dispatch): Return after performing atomic increment.
diff --git a/libstdc++-v3/include/ext/atomicity.h b/libstdc++-v3/include/ext/atomicity.h
index 73225b3de20..333c8843e14 100644
--- a/libstdc++-v3/include/ext/atomicity.h
+++ b/libstdc++-v3/include/ext/atomicity.h
@@ -55,10 +55,10 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
{ __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL); }
#else
_Atomic_word
- __exchange_and_add(volatile _Atomic_word*, int) throw ();
+ __exchange_and_add(volatile _Atomic_word*, int) _GLIBCXX_NOTHROW;
void
- __atomic_add(volatile _Atomic_word*, int) throw ();
+ __atomic_add(volatile _Atomic_word*, int) _GLIBCXX_NOTHROW;
#endif
inline _Atomic_word
@@ -92,7 +92,10 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
{
#ifdef __GTHREADS
if (__gthread_active_p())
- __atomic_add(__mem, __val);
+ {
+ __atomic_add(__mem, __val);
+ return;
+ }
#endif
__atomic_add_single(__mem, __val);
}
Re: [PATCH] Fix ODR violations in code using
On 05/07/19 20:23 +0200, Daniel Krügler wrote: Am Fr., 5. Juli 2019 um 18:13 Uhr schrieb Jonathan Wakely : [..] I decided against the simplification in the second patch, and committed the attached one which is closer to the first patch I sent (preserving the __atomic_add and __exchange_and_add functions even when they just call the built-ins). Tested x86_64-linux, powerpc64-linux, powerpc-aix. Committed to trunk. Unrelated to the actual patch, I noticed some explicit "throw()" forms used as exception specifications - shouldn't these be replaced by either explicit "noexcept" or at least by a library macro that expands to one or the other? Yes, they should be _GLIBCXX_NOTHROW. (I'm sorry, if such unrelated questions are considered as inappropriate for this list). Entirely appropriate, thanks!
Re: [PATCH] Fix ODR violations in code using
Am Fr., 5. Juli 2019 um 18:13 Uhr schrieb Jonathan Wakely : > [..] > I decided against the simplification in the second patch, and > committed the attached one which is closer to the first patch I sent > (preserving the __atomic_add and __exchange_and_add functions even > when they just call the built-ins). > > Tested x86_64-linux, powerpc64-linux, powerpc-aix. Committed to trunk. Unrelated to the actual patch, I noticed some explicit "throw()" forms used as exception specifications - shouldn't these be replaced by either explicit "noexcept" or at least by a library macro that expands to one or the other? (I'm sorry, if such unrelated questions are considered as inappropriate for this list). - Daniel
Re: [PATCH] Fix ODR violations in code using
On 21/06/19 18:13 +0100, Jonathan Wakely wrote:
On 21/06/19 18:08 +0100, Jonathan Wakely wrote:
On 21/06/19 13:01 -0400, Nathan Sidwell wrote:
On 6/21/19 12:01 PM, Jonathan Wakely wrote:
Nathan noticed that the 'static inline' functions in
cause ODR violations when used from inline functions or templates (see
[basic.def.odr] p12 bullet (12.2)). His modules branch now diagnoses
those violations, so we need a fix.
Looking at the history (r114044 by Paolo) I believe the idea was indeed
to allow different definitions to be used in different TUs. I think
__attribute__((__always_inline__)) is a better match for that than
'static inline', and doesn't violate the ODR (at least, not if all TUs
have the same values for __GTHREADS and _GLIBCXX_ATOMIC_BUILTINS).
These functions still violate this rule in [dcl.inline]:
C++17: "If a function with external linkage is declared inline in one
translation unit, it shall be declared inline in all translation units
in which it appears; no diagnostic is required."
C++2a WP: "If a function or variable with external or module linkage
is declared inline in one translation unit, there shall be a reachable
inline declaration in all translation units in which it is declared;
no diagnostic is required."
But that doesn't seem to be diagnosable by today's implementations.
Does this change seem reasonable?
yes, thanks!
Actually, I think I prefer this version. This produces identical code
to the always_inline version, but without the indirection to
additional inline functions, i.e. just inline the relevant code into
the _dispatch functions.
Tests are still running but no failures so far, as expected.
Oops, I spoke too soon:
FAIL: tr1/2_general_utilities/shared_ptr/thread/default_weaktoshared.cc (test
for excess errors)
UNRESOLVED: tr1/2_general_utilities/shared_ptr/thread/default_weaktoshared.cc
compilation failed to produce executable
FAIL: tr1/2_general_utilities/shared_ptr/thread/mutex_weaktoshared.cc (test for
excess errors)
UNRESOLVED: tr1/2_general_utilities/shared_ptr/thread/mutex_weaktoshared.cc
compilation failed to produce executable
Those tests explicitly use the __atomic_add function that the patch
removes. But those would be easy to adjust.
I decided against the simplification in the second patch, and
committed the attached one which is closer to the first patch I sent
(preserving the __atomic_add and __exchange_and_add functions even
when they just call the built-ins).
Tested x86_64-linux, powerpc64-linux, powerpc-aix. Committed to trunk.
commit 0d26b6506f20dcb1c956338baba0c26f623e25f2
Author: Jonathan Wakely
Date: Fri Jul 5 15:15:46 2019 +0100
Fix ODR violations in code using
Because the inline versions of __exchange_and_add and __atomic_add are
also marked static, they cannot be used from templates or other inline
functions without ODR violations. This change gives them external
linkage, but adds the always_inline attribute.
* include/ext/atomicity.h [_GLIBCXX_ATOMIC_BUILTINS] (__atomic_add)
(__exchange_and_add): Replace static specifier with always_inline
attribute.
(__exchange_and_add_single, __atomic_add_single): Likewise.
(__exchange_and_add_dispatch, __atomic_add_dispatch): Likewise. Also
combine !__gthread_active_p() and !__GTHREADS branches.
diff --git a/libstdc++-v3/include/ext/atomicity.h b/libstdc++-v3/include/ext/atomicity.h
index 0783a58e01a..73225b3de20 100644
--- a/libstdc++-v3/include/ext/atomicity.h
+++ b/libstdc++-v3/include/ext/atomicity.h
@@ -44,24 +44,25 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
// __exchange_and_add_dispatch
// __atomic_add_dispatch
#ifdef _GLIBCXX_ATOMIC_BUILTINS
- static inline _Atomic_word
+ inline _Atomic_word
+ __attribute__((__always_inline__))
__exchange_and_add(volatile _Atomic_word* __mem, int __val)
{ return __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL); }
- static inline void
+ inline void
+ __attribute__((__always_inline__))
__atomic_add(volatile _Atomic_word* __mem, int __val)
{ __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL); }
#else
_Atomic_word
- __attribute__ ((__unused__))
__exchange_and_add(volatile _Atomic_word*, int) throw ();
void
- __attribute__ ((__unused__))
__atomic_add(volatile _Atomic_word*, int) throw ();
#endif
- static inline _Atomic_word
+ inline _Atomic_word
+ __attribute__((__always_inline__))
__exchange_and_add_single(_Atomic_word* __mem, int __val)
{
_Atomic_word __result = *__mem;
@@ -69,36 +70,31 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
return __result;
}
- static inline void
+ inline void
+ __attribute__((__always_inline__))
__atomic_add_single(_Atomic_word* __mem, int __val)
{ *__mem += __val; }
- static inline _Atomic_word
- __attribute__ ((__unused__))
+ inline _Atomic_word
+ __attribute__ ((__always_inline__))
__exchange_and_add_dispatch(_Atomic_word* __mem, int __val)
Re: [PATCH] Fix ODR violations in code using
On 21/06/19 18:08 +0100, Jonathan Wakely wrote: On 21/06/19 13:01 -0400, Nathan Sidwell wrote: On 6/21/19 12:01 PM, Jonathan Wakely wrote: Nathan noticed that the 'static inline' functions in cause ODR violations when used from inline functions or templates (see [basic.def.odr] p12 bullet (12.2)). His modules branch now diagnoses those violations, so we need a fix. Looking at the history (r114044 by Paolo) I believe the idea was indeed to allow different definitions to be used in different TUs. I think __attribute__((__always_inline__)) is a better match for that than 'static inline', and doesn't violate the ODR (at least, not if all TUs have the same values for __GTHREADS and _GLIBCXX_ATOMIC_BUILTINS). These functions still violate this rule in [dcl.inline]: C++17: "If a function with external linkage is declared inline in one translation unit, it shall be declared inline in all translation units in which it appears; no diagnostic is required." C++2a WP: "If a function or variable with external or module linkage is declared inline in one translation unit, there shall be a reachable inline declaration in all translation units in which it is declared; no diagnostic is required." But that doesn't seem to be diagnosable by today's implementations. Does this change seem reasonable? yes, thanks! Actually, I think I prefer this version. This produces identical code to the always_inline version, but without the indirection to additional inline functions, i.e. just inline the relevant code into the _dispatch functions. Tests are still running but no failures so far, as expected. Oops, I spoke too soon: FAIL: tr1/2_general_utilities/shared_ptr/thread/default_weaktoshared.cc (test for excess errors) UNRESOLVED: tr1/2_general_utilities/shared_ptr/thread/default_weaktoshared.cc compilation failed to produce executable FAIL: tr1/2_general_utilities/shared_ptr/thread/mutex_weaktoshared.cc (test for excess errors) UNRESOLVED: tr1/2_general_utilities/shared_ptr/thread/mutex_weaktoshared.cc compilation failed to produce executable Those tests explicitly use the __atomic_add function that the patch removes. But those would be easy to adjust.
Re: [PATCH] Fix ODR violations in code using
On 21/06/19 13:01 -0400, Nathan Sidwell wrote:
On 6/21/19 12:01 PM, Jonathan Wakely wrote:
Nathan noticed that the 'static inline' functions in
cause ODR violations when used from inline functions or templates (see
[basic.def.odr] p12 bullet (12.2)). His modules branch now diagnoses
those violations, so we need a fix.
Looking at the history (r114044 by Paolo) I believe the idea was indeed
to allow different definitions to be used in different TUs. I think
__attribute__((__always_inline__)) is a better match for that than
'static inline', and doesn't violate the ODR (at least, not if all TUs
have the same values for __GTHREADS and _GLIBCXX_ATOMIC_BUILTINS).
These functions still violate this rule in [dcl.inline]:
C++17: "If a function with external linkage is declared inline in one
translation unit, it shall be declared inline in all translation units
in which it appears; no diagnostic is required."
C++2a WP: "If a function or variable with external or module linkage
is declared inline in one translation unit, there shall be a reachable
inline declaration in all translation units in which it is declared;
no diagnostic is required."
But that doesn't seem to be diagnosable by today's implementations.
Does this change seem reasonable?
yes, thanks!
Actually, I think I prefer this version. This produces identical code
to the always_inline version, but without the indirection to
additional inline functions, i.e. just inline the relevant code into
the _dispatch functions.
Tests are still running but no failures so far, as expected.
commit 56040dad3b55a98f6e5090d04cb49945c40df39c
Author: Jonathan Wakely
Date: Fri Jun 21 17:52:23 2019 +0100
Simplify
diff --git a/libstdc++-v3/include/ext/atomicity.h b/libstdc++-v3/include/ext/atomicity.h
index 0783a58e01a..30453c92315 100644
--- a/libstdc++-v3/include/ext/atomicity.h
+++ b/libstdc++-v3/include/ext/atomicity.h
@@ -43,62 +43,49 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
// To abstract locking primitives across all thread policies, use:
// __exchange_and_add_dispatch
// __atomic_add_dispatch
-#ifdef _GLIBCXX_ATOMIC_BUILTINS
- static inline _Atomic_word
- __exchange_and_add(volatile _Atomic_word* __mem, int __val)
- { return __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL); }
-
- static inline void
- __atomic_add(volatile _Atomic_word* __mem, int __val)
- { __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL); }
-#else
+#ifndef _GLIBCXX_ATOMIC_BUILTINS
_Atomic_word
- __attribute__ ((__unused__))
__exchange_and_add(volatile _Atomic_word*, int) throw ();
void
- __attribute__ ((__unused__))
__atomic_add(volatile _Atomic_word*, int) throw ();
#endif
- static inline _Atomic_word
- __exchange_and_add_single(_Atomic_word* __mem, int __val)
+ inline _Atomic_word
+ __attribute__ ((__always_inline__))
+ __exchange_and_add_dispatch(_Atomic_word* __mem, int __val)
{
+#ifdef __GTHREADS
+if (__gthread_active_p())
+ {
+#ifdef _GLIBCXX_ATOMIC_BUILTINS
+ return __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL);
+#else
+ return __exchange_and_add(__mem, __val);
+#endif
+ }
+#endif
_Atomic_word __result = *__mem;
*__mem += __val;
return __result;
}
- static inline void
- __atomic_add_single(_Atomic_word* __mem, int __val)
- { *__mem += __val; }
-
- static inline _Atomic_word
- __attribute__ ((__unused__))
- __exchange_and_add_dispatch(_Atomic_word* __mem, int __val)
- {
-#ifdef __GTHREADS
-if (__gthread_active_p())
- return __exchange_and_add(__mem, __val);
-else
- return __exchange_and_add_single(__mem, __val);
-#else
-return __exchange_and_add_single(__mem, __val);
-#endif
- }
-
- static inline void
- __attribute__ ((__unused__))
+ inline void
+ __attribute__ ((__always_inline__))
__atomic_add_dispatch(_Atomic_word* __mem, int __val)
{
#ifdef __GTHREADS
if (__gthread_active_p())
- __atomic_add(__mem, __val);
-else
- __atomic_add_single(__mem, __val);
+ {
+#ifdef _GLIBCXX_ATOMIC_BUILTINS
+ __atomic_fetch_add(__mem, __val, __ATOMIC_ACQ_REL);
#else
-__atomic_add_single(__mem, __val);
+ __atomic_add(__mem, __val);
#endif
+ }
+else
+#endif
+*__mem += __val;
}
_GLIBCXX_END_NAMESPACE_VERSION
diff --git a/libstdc++-v3/src/c++98/mt_allocator.cc b/libstdc++-v3/src/c++98/mt_allocator.cc
index f842c6f9cfd..9b26f3dc993 100644
--- a/libstdc++-v3/src/c++98/mt_allocator.cc
+++ b/libstdc++-v3/src/c++98/mt_allocator.cc
@@ -302,7 +302,8 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
if (__reclaimed > 1024)
{
__bin._M_used[__thread_id] -= __reclaimed;
- __atomic_add(&__reclaimed_base[__thread_id], -__reclaimed);
+ __atomic_fetch_add(&__reclaimed_base[__thread_id], -__reclaimed,
+ __ATOMIC_ACQ_REL);
}
if (__remove >= __net_used)
@@ -332,7 +333,8 @@ _GLIBCXX_BEGIN_NAMESPACE_VERSION
if (__block->_M_thread_id == __thread_id)
--__bin._M_used[__thread_id];
else
-
Re: [PATCH] Fix ODR violations in code using
On 6/21/19 12:01 PM, Jonathan Wakely wrote: Nathan noticed that the 'static inline' functions in cause ODR violations when used from inline functions or templates (see [basic.def.odr] p12 bullet (12.2)). His modules branch now diagnoses those violations, so we need a fix. Looking at the history (r114044 by Paolo) I believe the idea was indeed to allow different definitions to be used in different TUs. I think __attribute__((__always_inline__)) is a better match for that than 'static inline', and doesn't violate the ODR (at least, not if all TUs have the same values for __GTHREADS and _GLIBCXX_ATOMIC_BUILTINS). These functions still violate this rule in [dcl.inline]: C++17: "If a function with external linkage is declared inline in one translation unit, it shall be declared inline in all translation units in which it appears; no diagnostic is required." C++2a WP: "If a function or variable with external or module linkage is declared inline in one translation unit, there shall be a reachable inline declaration in all translation units in which it is declared; no diagnostic is required." But that doesn't seem to be diagnosable by today's implementations. Does this change seem reasonable? yes, thanks! nathan -- Nathan Sidwell
[PATCH] Fix ODR violations in code using
Nathan noticed that the 'static inline' functions in cause ODR violations when used from inline functions or templates (see [basic.def.odr] p12 bullet (12.2)). His modules branch now diagnoses those violations, so we need a fix. Looking at the history (r114044 by Paolo) I believe the idea was indeed to allow different definitions to be used in different TUs. I think __attribute__((__always_inline__)) is a better match for that than 'static inline', and doesn't violate the ODR (at least, not if all TUs have the same values for __GTHREADS and _GLIBCXX_ATOMIC_BUILTINS). These functions still violate this rule in [dcl.inline]: C++17: "If a function with external linkage is declared inline in one translation unit, it shall be declared inline in all translation units in which it appears; no diagnostic is required." C++2a WP: "If a function or variable with external or module linkage is declared inline in one translation unit, there shall be a reachable inline declaration in all translation units in which it is declared; no diagnostic is required." But that doesn't seem to be diagnosable by today's implementations. Does this change seem reasonable? It does change the size of a few files, e.g. for x86_64 some are bigger and some are smaller, overall slightly smaller: text data bss dec hex filename 54008 35257601680 src/c++11/cow-locale_init.o 54028 35257621682 src/c++11/cow-locale_init.o [always_inline] 22772 2056 0 2482860fc src/c++11/cow-shim_facets.o 22710 2056 0 2476660be src/c++11/cow-shim_facets.o [always_inline] 49458 2184 0 51642c9ba src/c++11/cow-sstream-inst.o 49482 2184 0 51666c9d2 src/c++11/cow-sstream-inst.o [always_inline] 74618 074691d2d src/c++11/cow-stdexcept.o 74648 074721d30 src/c++11/cow-stdexcept.o [always_inline] 167538 32 167934199 src/c++11/cow-string-inst.o 167568 32 16796419c src/c++11/cow-string-inst.o [always_inline] 170098 32 170494299 src/c++11/cow-wstring-inst.o 170088 32 170484298 src/c++11/cow-wstring-inst.o [always_inline] 20342 2072 0 22414578e src/c++11/cxx11-shim_facets.o 20276 2072 0 22348574c src/c++11/cxx11-shim_facets.o [always_inline] 2929 240 83177 c69 src/c++11/future.o 2945 240 83193 c79 src/c++11/future.o [always_inline] 6789 120 069091afd src/c++11/ios-inst.o 6750 120 068701ad6 src/c++11/ios-inst.o [always_inline] 2364 57 82429 97d src/c++11/ios.o 2362 57 82427 97b src/c++11/ios.o [always_inline] 94493 2712 192 97397 17c75 src/c++11/locale-inst.o 94498 2712 192 97402 17c7a src/c++11/locale-inst.o [always_inline] 93094 2712 192 95998 176fe src/c++11/wlocale-inst.o 93113 2712 192 96017 17711 src/c++11/wlocale-inst.o [always_inline] 20189 208 0 203974fad src/c++17/cow-fs_dir.o 20118 208 0 203264f66 src/c++17/cow-fs_dir.o [always_inline] 550778 0 55085d72d src/c++17/cow-fs_ops.o 551038 0 55111d747 src/c++17/cow-fs_ops.o [always_inline] 40200 304 0 405049e38 src/c++17/cow-fs_path.o 40204 304 0 405089e3c src/c++17/cow-fs_path.o [always_inline] 18672 208 0 1888049c0 src/c++17/fs_dir.o 18628 208 0 188364994 src/c++17/fs_dir.o [always_inline] 60604 160 0 60764ed5c src/c++17/fs_ops.o 60606 160 0 60766ed5e src/c++17/fs_ops.o [always_inline] 39568 304 0 398729bc0 src/c++17/fs_path.o 39561 304 0 398659bb9 src/c++17/fs_path.o [always_inline] 36538 03661 e4d src/c++98/ios_init.o 36568 03664 e50 src/c++98/ios_init.o [always_inline] 6878 184 8471461bea src/c++98/locale.o 6896 184 8471641bfc src/c++98/locale.o [always_inline] 6693 7685368 12829321d src/c++98/locale_init.o 6710 7685368 12846322e src/c++98/locale_init.o [always_inline] 72228 072301c3e src/c++98/localename.o 72238 072311c3f src/c++98/localename.o [always_inline] 4036 144 20843881124 src/c++98/pool_allocator.o 4044 144 2084396112c src/c++98/pool_allocator.o [always_inline] 1955 584 02539 9eb src/c++98/stdexcept.o 1957 584 02541 9ed src/c++98/stdexcept.o [always_inline] 25199 208 0 25407633f src/filesystem/cow-dir.o 25248 208 0 254566370 src/filesystem/cow-dir.o [always_inline] 711348 0 71142 115e6
