* __builtin_assume warnings
@ 2020-08-12 9:38 Mattias Engdegård
2020-08-15 2:32 ` Paul Eggert
0 siblings, 1 reply; 8+ messages in thread
From: Mattias Engdegård @ 2020-08-12 9:38 UTC (permalink / raw)
To: Paul Eggert; +Cc: emacs-devel
Building on macOS with Apple clang version 11.0.0 (clang-1100.0.33.17) now gives a stream of warnings about:
In file included from ../../emacs/src/dispnew.c:27:
../../emacs/src/lisp.h:1812:12: warning: the argument to '__builtin_assume' has
side effects that will be discarded [-Wassume]
eassume (0 <= i && i < bool_vector_size (a));
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../../emacs/src/lisp.h:182:32: note: expanded from macro 'eassume'
# define eassume(cond) assume (cond)
^~~~
../../emacs/lib/verify.h:305:38: note: expanded from macro 'assume'
# define assume(R) __builtin_assume (R)
^
We could build with -Wno-assume, or alter the definition of 'assume' to
# define assume(R) _Pragma("clang diagnostic push") \
_Pragma("clang diagnostic ignored \"-Wassume\"") \
__builtin_assume (R) \
_Pragma("clang diagnostic pop")
but neither seems entirely satisfactory.
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: __builtin_assume warnings
2020-08-12 9:38 __builtin_assume warnings Mattias Engdegård
@ 2020-08-15 2:32 ` Paul Eggert
2020-08-16 8:23 ` Mattias Engdegård
0 siblings, 1 reply; 8+ messages in thread
From: Paul Eggert @ 2020-08-15 2:32 UTC (permalink / raw)
To: Mattias Engdegård; +Cc: emacs-devel
[-- Attachment #1: Type: text/plain, Size: 333 bytes --]
On 8/12/20 2:38 AM, Mattias Engdegård wrote:
> ../../emacs/src/lisp.h:1812:12: warning: the argument to '__builtin_assume' has
> side effects that will be discarded [-Wassume]
> eassume (0 <= i && i < bool_vector_size (a));
Thanks for mentioning that. I installed the attached, which should pacify clang
there.
[-- Attachment #2: 0001-Pacify-Apple-clang-11-__builtin_assume.patch --]
[-- Type: text/x-patch, Size: 1625 bytes --]
From 3296c0b2d1a61f099aa1d662b0a36e0b8f78ed28 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Fri, 14 Aug 2020 19:29:14 -0700
Subject: [PATCH] Pacify Apple clang 11 __builtin_assume
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Problem reported by Mattias Engdegård in:
https://lists.gnu.org/r/emacs-devel/2020-08/msg00300.html
* src/lisp.h (bool_vector_bitref, bool_vector_set):
Use eassert instead of eassume for bool_vector_size checks.
---
src/lisp.h | 11 ++++++-----
1 file changed, 6 insertions(+), 5 deletions(-)
diff --git a/src/lisp.h b/src/lisp.h
index 2962babb4f..eaf1c6ce6d 100644
--- a/src/lisp.h
+++ b/src/lisp.h
@@ -1809,7 +1809,8 @@ bool_vector_uchar_data (Lisp_Object a)
INLINE bool
bool_vector_bitref (Lisp_Object a, EMACS_INT i)
{
- eassume (0 <= i && i < bool_vector_size (a));
+ eassume (0 <= i);
+ eassert (i < bool_vector_size (a));
return !! (bool_vector_uchar_data (a)[i / BOOL_VECTOR_BITS_PER_CHAR]
& (1 << (i % BOOL_VECTOR_BITS_PER_CHAR)));
}
@@ -1825,11 +1826,11 @@ bool_vector_ref (Lisp_Object a, EMACS_INT i)
INLINE void
bool_vector_set (Lisp_Object a, EMACS_INT i, bool b)
{
- unsigned char *addr;
-
- eassume (0 <= i && i < bool_vector_size (a));
- addr = &bool_vector_uchar_data (a)[i / BOOL_VECTOR_BITS_PER_CHAR];
+ eassume (0 <= i);
+ eassert (i < bool_vector_size (a));
+ unsigned char *addr
+ = &bool_vector_uchar_data (a)[i / BOOL_VECTOR_BITS_PER_CHAR];
if (b)
*addr |= 1 << (i % BOOL_VECTOR_BITS_PER_CHAR);
else
--
2.17.1
^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: __builtin_assume warnings
2020-08-15 2:32 ` Paul Eggert
@ 2020-08-16 8:23 ` Mattias Engdegård
2020-08-17 14:05 ` Mattias Engdegård
0 siblings, 1 reply; 8+ messages in thread
From: Mattias Engdegård @ 2020-08-16 8:23 UTC (permalink / raw)
To: Paul Eggert; +Cc: emacs-devel
15 aug. 2020 kl. 04.32 skrev Paul Eggert <eggert@cs.ucla.edu>:
>
> On 8/12/20 2:38 AM, Mattias Engdegård wrote:
>> ../../emacs/src/lisp.h:1812:12: warning: the argument to '__builtin_assume' has
>> side effects that will be discarded [-Wassume]
>> eassume (0 <= i && i < bool_vector_size (a));
>
> Thanks for mentioning that. I installed the attached, which should pacify clang there.
Much better, thank you! (For the record, I believe it applies to any Clang version with that builtin and warning, not just the Apple 11 version.)
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: __builtin_assume warnings
2020-08-16 8:23 ` Mattias Engdegård
@ 2020-08-17 14:05 ` Mattias Engdegård
2020-08-17 22:11 ` Paul Eggert
0 siblings, 1 reply; 8+ messages in thread
From: Mattias Engdegård @ 2020-08-17 14:05 UTC (permalink / raw)
To: Paul Eggert; +Cc: emacs-devel
16 aug. 2020 kl. 10.23 skrev Mattias Engdegård <mattiase@acm.org>:
>
> 15 aug. 2020 kl. 04.32 skrev Paul Eggert <eggert@cs.ucla.edu>:
>>
>> On 8/12/20 2:38 AM, Mattias Engdegård wrote:
>>> ../../emacs/src/lisp.h:1812:12: warning: the argument to '__builtin_assume' has
>>> side effects that will be discarded [-Wassume]
>>> eassume (0 <= i && i < bool_vector_size (a));
>>
>> Thanks for mentioning that. I installed the attached, which should pacify clang there.
Actually, there are warnings in more places. Looks like Clang does not yet look inside inline functions for -Wassume.
frame.c:1686:12: warning: the argument to '__builtin_assume' has side effects
that will be discarded [-Wassume]
eassume (CONSP (Vframe_list));
^~~~~~~~~~~~~~~~~~~
./lisp.h:182:32: note: expanded from macro 'eassume'
# define eassume(cond) assume (cond)
^~~~
../lib/verify.h:305:38: note: expanded from macro 'assume'
# define assume(R) __builtin_assume (R)
^
frame.c:1710:12: warning: the argument to '__builtin_assume' has side effects
that will be discarded [-Wassume]
eassume (CONSP (Vframe_list));
^~~~~~~~~~~~~~~~~~~
./lisp.h:182:32: note: expanded from macro 'eassume'
# define eassume(cond) assume (cond)
^~~~
../lib/verify.h:305:38: note: expanded from macro 'assume'
# define assume(R) __builtin_assume (R)
^
frame.c:2020:16: warning: the argument to '__builtin_assume' has side effects
that will be discarded [-Wassume]
eassume (CONSP (Vframe_list));
^~~~~~~~~~~~~~~~~~~
./lisp.h:182:32: note: expanded from macro 'eassume'
# define eassume(cond) assume (cond)
^~~~
../lib/verify.h:305:38: note: expanded from macro 'assume'
# define assume(R) __builtin_assume (R)
^
character.c:814:16: warning: the argument to '__builtin_assume' has side effects
that will be discarded [-Wassume]
eassume (CHARACTERP (args[i]));
^~~~~~~~~~~~~~~~~~~~
./lisp.h:182:32: note: expanded from macro 'eassume'
# define eassume(cond) assume (cond)
^~~~
../lib/verify.h:305:38: note: expanded from macro 'assume'
# define assume(R) __builtin_assume (R)
^
alloc.c:1935:12: warning: the argument to '__builtin_assume' has side effects
that will be discarded [-Wassume]
eassume (STRING_MULTIBYTE (string));
^~~~~~~~~~~~~~~~~~~~~~~~~
./lisp.h:182:32: note: expanded from macro 'eassume'
# define eassume(cond) assume (cond)
^~~~
../lib/verify.h:305:38: note: expanded from macro 'assume'
# define assume(R) __builtin_assume (R)
^
data.c:3153:12: warning: the argument to '__builtin_assume' has side effects
that will be discarded [-Wassume]
eassume (FIXNUMP (value));
^~~~~~~~~~~~~~~
./lisp.h:182:32: note: expanded from macro 'eassume'
# define eassume(cond) assume (cond)
^~~~
../lib/verify.h:305:38: note: expanded from macro 'assume'
# define assume(R) __builtin_assume (R)
^
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: __builtin_assume warnings
2020-08-17 14:05 ` Mattias Engdegård
@ 2020-08-17 22:11 ` Paul Eggert
2020-08-18 7:11 ` Mattias Engdegård
0 siblings, 1 reply; 8+ messages in thread
From: Paul Eggert @ 2020-08-17 22:11 UTC (permalink / raw)
To: Mattias Engdegård; +Cc: emacs-devel
[-- Attachment #1: Type: text/plain, Size: 375 bytes --]
On 8/17/20 7:05 AM, Mattias Engdegård wrote:
> Looks like Clang does not yet look inside inline functions for -Wassume.
I guess that for that diagnostic, Clang assumes that every function call has a
side effect, which is pretty silly. I worked around the bug (and by an another
bogus-warning bug I ran into on Fedora 31's Clang 9.0.1) by installing the attached.
[-- Attachment #2: 0001-Update-from-Gnulib.patch --]
[-- Type: text/x-patch, Size: 2707 bytes --]
From 352b7dede0b3a28024a41a3da1c340859b110665 Mon Sep 17 00:00:00 2001
From: Paul Eggert <eggert@cs.ucla.edu>
Date: Mon, 17 Aug 2020 15:05:05 -0700
Subject: [PATCH] Update from Gnulib
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
This incorporates:
2020-08-17 verify: avoid __built_assume on Clang
2020-08-17 libc-config: avoid Clang’s __diagnose_if__
* lib/cdefs.h, lib/verify.h: Copy from Gnulib.
---
lib/cdefs.h | 2 +-
lib/verify.h | 18 +++++++-----------
2 files changed, 8 insertions(+), 12 deletions(-)
diff --git a/lib/cdefs.h b/lib/cdefs.h
index 0cc27806a1..b1870fd0a9 100644
--- a/lib/cdefs.h
+++ b/lib/cdefs.h
@@ -148,7 +148,7 @@ #define __bos0(ptr) __builtin_object_size (ptr, 0)
# define __warnattr(msg) __attribute__((__warning__ (msg)))
# define __errordecl(name, msg) \
extern void name (void) __attribute__((__error__ (msg)))
-#elif __glibc_clang_has_attribute (__diagnose_if__)
+#elif __glibc_clang_has_attribute (__diagnose_if__) && 0 /* fails on Fedora 31 with Clang 9. */
# define __warndecl(name, msg) \
extern void name (void) __attribute__((__diagnose_if__ (1, msg, "warning")))
# define __warnattr(msg) __attribute__((__diagnose_if__ (1, msg, "warning")))
diff --git a/lib/verify.h b/lib/verify.h
index 0ba8d57907..d485a0283a 100644
--- a/lib/verify.h
+++ b/lib/verify.h
@@ -246,13 +246,6 @@ #define _GL_VERIFY_TRUE(R, DIAGNOSTIC) \
/* @assert.h omit start@ */
-#if defined __has_builtin
-/* <https://clang.llvm.org/docs/LanguageExtensions.html#builtin-functions> */
-# define _GL_HAS_BUILTIN_ASSUME __has_builtin (__builtin_assume)
-#else
-# define _GL_HAS_BUILTIN_ASSUME 0
-#endif
-
#if 3 < __GNUC__ + (3 < __GNUC_MINOR__ + (4 <= __GNUC_PATCHLEVEL__))
# define _GL_HAS_BUILTIN_TRAP 1
#elif defined __has_builtin
@@ -312,11 +305,14 @@ #define verify_expr(R, E) \
Although assuming R can help a compiler generate better code or
diagnostics, performance can suffer if R uses hard-to-optimize
- features such as function calls not inlined by the compiler. */
+ features such as function calls not inlined by the compiler.
+
+ Avoid Clang’s __builtin_assume, as clang 9.0.1 -Wassume can
+ generate a bogus diagnostic "the argument to '__builtin_assume' has
+ side effects that will be discarded" even when the argument has no
+ side effects. */
-#if _GL_HAS_BUILTIN_ASSUME
-# define assume(R) __builtin_assume (R)
-#elif _GL_HAS_BUILTIN_UNREACHABLE
+#if _GL_HAS_BUILTIN_UNREACHABLE
# define assume(R) ((R) ? (void) 0 : __builtin_unreachable ())
#elif 1200 <= _MSC_VER
# define assume(R) __assume (R)
--
2.17.1
^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: __builtin_assume warnings
2020-08-17 22:11 ` Paul Eggert
@ 2020-08-18 7:11 ` Mattias Engdegård
2020-08-18 22:53 ` Paul Eggert
0 siblings, 1 reply; 8+ messages in thread
From: Mattias Engdegård @ 2020-08-18 7:11 UTC (permalink / raw)
To: Paul Eggert; +Cc: emacs-devel
18 aug. 2020 kl. 00.11 skrev Paul Eggert <eggert@cs.ucla.edu>:
> I guess that for that diagnostic, Clang assumes that every function call has a side effect, which is pretty silly. I worked around the bug (and by an another bogus-warning bug I ran into on Fedora 31's Clang 9.0.1) by installing the attached.
Thank you, but aren't we throwing the baby out with the bathwater? Wouldn't -Wno-assume, either locally (clang pragmas around __builtin_assume) or globally (configure?) be more effective?
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: __builtin_assume warnings
2020-08-18 7:11 ` Mattias Engdegård
@ 2020-08-18 22:53 ` Paul Eggert
2020-08-19 8:01 ` Mattias Engdegård
0 siblings, 1 reply; 8+ messages in thread
From: Paul Eggert @ 2020-08-18 22:53 UTC (permalink / raw)
To: Mattias Engdegård; +Cc: emacs-devel
On 8/18/20 12:11 AM, Mattias Engdegård wrote:
> Wouldn't -Wno-assume, either locally (clang pragmas around __builtin_assume) or globally (configure?) be more effective?
No, it's the other way around at least for me: having 'assume' use Clang's
__builtin_assume makes 'assume' slower. Without __builtin_assume, 'assume' falls
back on __builtin_unreachable, and Clang generates better code for
__builtin_unreachable than it does __builtin_assume. For the following code:
int x;
static int f (void) { return x; }
int g (void) { __builtin_assume (!f ()); return f (); }
int h (void) { if (f ()) __builtin_unreachable (); return f (); }
clang -O2 generates suboptimal machine code for g (the generated code loads from
'x') and better machine code for h (the generated code returns 0 without loading
from 'x'). This is clang version 9.0.1 (Fedora 9.0.1-2.fc31) on
x86_64-unknown-linux-gnu.
Perhaps someday the Clang folks will get their act together in this department,
but in the meantime __builtin_unreachable is a perfectly adequate substitute.
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: __builtin_assume warnings
2020-08-18 22:53 ` Paul Eggert
@ 2020-08-19 8:01 ` Mattias Engdegård
0 siblings, 0 replies; 8+ messages in thread
From: Mattias Engdegård @ 2020-08-19 8:01 UTC (permalink / raw)
To: Paul Eggert; +Cc: emacs-devel
19 aug. 2020 kl. 00.53 skrev Paul Eggert <eggert@cs.ucla.edu>:
> No, it's the other way around at least for me: having 'assume' use Clang's __builtin_assume makes 'assume' slower. Without __builtin_assume, 'assume' falls back on __builtin_unreachable, and Clang generates better code for __builtin_unreachable than it does __builtin_assume.
Right you are; as far as I can tell, __builtin_assume is strictly less useful. There does not seem to be any point in using it at this time. The only advantage over if(!x)__builtin_unreachable() appears to be that the argument isn't actually evaluated, but I'm not sure when that property would be useful.
Thank you for clearing that up.
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2020-08-19 8:01 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-08-12 9:38 __builtin_assume warnings Mattias Engdegård
2020-08-15 2:32 ` Paul Eggert
2020-08-16 8:23 ` Mattias Engdegård
2020-08-17 14:05 ` Mattias Engdegård
2020-08-17 22:11 ` Paul Eggert
2020-08-18 7:11 ` Mattias Engdegård
2020-08-18 22:53 ` Paul Eggert
2020-08-19 8:01 ` Mattias Engdegård
Code repositories for project(s) associated with this public inbox
https://git.savannah.gnu.org/cgit/emacs.git
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).