[Bug symtab/27937] Performance regression of 8% due to assert in dwarf/read/attribute.h

["simark at simark dot ca" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=27937

Simon Marchi <simark at simark dot ca> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |simark at simark dot ca

--- Comment #1 from Simon Marchi <simark at simark dot ca> ---
I sometimes wonder if we should have a gdb_assert version that is only enabled
in development builds, that we could use in very hot paths like this one,
without affecting performance of production builds.  On one hand, it's not good
because problems could go unnoticed in production builds.  But if the
alternative is removing the assert because it is too costly, then having it
only in development builds is better than nothing.

Also, I was wondering if this would have any effect on your measurements:

diff --git a/gdbsupport/gdb_assert.h b/gdbsupport/gdb_assert.h
index 00553a786135..74dcd653cc4e 100644
--- a/gdbsupport/gdb_assert.h
+++ b/gdbsupport/gdb_assert.h
@@ -32,7 +32,7 @@
    replacing.  */

 #define gdb_assert(expr)                                                     
\
-  ((void) ((expr) ? 0 :                                                      
\
+  ((void) (__builtin_expect ((expr), 0) ? 0 :                                 
                     \
           (gdb_assert_fail (#expr, __FILE__, __LINE__, FUNCTION_NAME), 0)))

 /* This prints an "Assertion failed" message, asking the user if they

-- 
You are receiving this mail because:
You are on the CC list for the bug.