Re: Forward declarations and variable alignment weirdness
> > PR45112. See also http://gcc.gnu.org/ml/gcc-patches/2010-08/msg00283.html > > Thanks a lot for info. > If I got the discussion right, it's ok for the definition to use more > strict alignment than the declaration. Not quite. Any mismatch is user error, and may cause nasal daemons. Whether this actually causes failures depends on the context, and you can construct code that fails either way. In practice giving the definition greater alignment than the declaration will probably lead to sub-optimal code rather than active borkage. Paul
RE: CLEAR_INSN_CACHE
That does the trick. Thanks, steve -Original Message- From: Ian Lance Taylor [mailto:i...@google.com] Sent: 21 August 2010 19:50 To: Kilbane, Stephen Cc: gcc@gcc.gnu.org Subject: Re: CLEAR_INSN_CACHE "Kilbane, Stephen" writes: > I'm trying to add support to 4.3 for cache flushing when setting up a > trampoline. > Reading around, it looks like I should be defining CLEAR_INSN_CACHE: > > To clear the instruction cache when a trampoline is initialized, define the > following macro. >CLEAR_INSN_CACHE (beg, end) > > but I can't see anywhere that this macro gets used during trampoline > creation. I can see > that it provides an implementation for __clear_cache() in libgcc, but nothing > that would > invoke that routine, either, short of an explicit call in the source file > being compiled. > > Am I missing something? No. You do need to arrange to flush the cache yourself, typically by generating the appropriate instructions in the TARGET_TRAMPOLINE_INIT target hook. On targets which need to clear the cache, that hook will often emit a call to __clear_cache, as in emit_library_call (gen_rtx_SYMBOL_REF (Pmode, "__clear_cache"), LCT_NORMAL, VOIDmode, 2, a_tramp, Pmode, plus_constant (a_tramp, TRAMPOLINE_SIZE), Pmode); where a_tramp is the address of the trampoline. The CLEAR_INSN_CACHE macro will then provide the body of the __clear_cache function. Of course, if clearing the cache is reasonably simple, the instructions can be emitted inline by TARGET_TRAMPOLINE_INIT; e.g., that's what the MIPS target does. Ian
[testsuite] How to conditionally skip a single dejagnu directive
Hello,
I have a runnable (fortran) test on which I want to check that a
conditional part is optimized away.
Here are the corresponding tree-dump directives :
! { dg-final { scan-tree-dump-times "_gfortran_abort" 32 "original" } }
! { dg-final { scan-tree-dump-times "_gfortran_abort" 30 "optimized" } }
This works, but as the fortran run test is a torture one (it tries
several optimization options), there is a single failure for the -O0
case (because there is no optimization).
I would like to skip the dg directive if the options match "-O0" but on
the tree dump only.
There is dg-skip-if, but it skips the whole test, and I don't want to
skip unrelated dg directives. Especially I don't want to skip the
testcase execution.
So, is there something existing to skip a single dump scan based on the
compilation options ?
Thanks in advance
Mikael
build: are there situations where 'ar' does not work?
Hello GCC developers, Libtool recently added a configure test for @file support in ar (and other tools) in order to avoid the slower partial building or static libraries on some systems (and have an escape route on systems where the archiver doesn't allow adding to an archive). Now, to avoid issues when GCC uses newer Libtool: are there situations where running $AR does not work? Cross compile? In-tree binutils? Thanks, Ralf
gcc-4.3-20100822 is now available
Snapshot gcc-4.3-20100822 is now available on ftp://gcc.gnu.org/pub/gcc/snapshots/4.3-20100822/ and on various mirrors, see http://gcc.gnu.org/mirrors.html for details. This snapshot has been generated from the GCC 4.3 SVN branch with the following options: svn://gcc.gnu.org/svn/gcc/branches/gcc-4_3-branch revision 163461 You'll find: gcc-4.3-20100822.tar.bz2 Complete GCC (includes all of below) MD5=810d4a5d9b1d73e56efa9dc37ad48aa8 SHA1=5b586b77ae0361962a82615021cd0ce451cdc46b gcc-core-4.3-20100822.tar.bz2C front end and core compiler MD5=8eae4f42d291630c8734ce5e4a61918c SHA1=fda3debfc449ad96dc3e30dfd6fd3c9db382af59 gcc-ada-4.3-20100822.tar.bz2 Ada front end and runtime MD5=21afba18c2eda1523e4afc3075c70bf1 SHA1=1e86bf6616833a57272c8d8af843e528c782d060 gcc-fortran-4.3-20100822.tar.bz2 Fortran front end and runtime MD5=af31c43011978a3163f5a77c15829c34 SHA1=ec28811ae73eafad5e2854818e9252db54408a62 gcc-g++-4.3-20100822.tar.bz2 C++ front end and runtime MD5=19491c72aa7d574e1b9c8cdd83077494 SHA1=04a6b7be26b9db3ba5a7c760e261dffba9f6c753 gcc-java-4.3-20100822.tar.bz2Java front end and runtime MD5=96b2930e7f86adb75d850d416dde02ad SHA1=4e6d112f35b12035820972987242e72f4ac5b1cc gcc-objc-4.3-20100822.tar.bz2Objective-C front end and runtime MD5=61c7fd3c7203907c8514cc0b1114813b SHA1=a35b8ad866e851a4432817dccf8833424a9a5482 gcc-testsuite-4.3-20100822.tar.bz2 The GCC testsuite MD5=3af791b71b3739f4522abfe99977503a SHA1=52b38d7be027ec5b6db96e857d64df3d7974 Diffs from 4.3-20100815 are available in the diffs/ subdirectory. When a particular snapshot is ready for public consumption the LATEST-4.3 link is updated and a message is sent to the gcc list. Please do not use a snapshot before it has been announced that way.
�y�o�k� o�H� Be�ge_mmjh
ТанÑÑÑÑÐ°Ñ cÑÑkа Ð´Ð¾Ð»Ð±Ð¸Ñ Ð¼ÐµÐ¶Ð´Ñ Ð½Ð¾Ð³ banana.com.mk/wp-content/uploads/2010/08/data3.html
Re: build: are there situations where 'ar' does not work?
Ralf Wildenhues writes: > Libtool recently added a configure test for @file support in ar > (and other tools) in order to avoid the slower partial building > or static libraries on some systems (and have an escape route on > systems where the archiver doesn't allow adding to an archive). > > Now, to avoid issues when GCC uses newer Libtool: are there situations > where running $AR does not work? Cross compile? In-tree binutils? When building a target library (libgcc, libstdc++, etc.) $AR should always work. Ian
