* __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).