| Age | Commit message (Collapse) | Author |
|---|
|
derefs in at least one case and I do not have time to debug this before the
release.
ok deraadt@
|
|
compile current libm. ok martynas@
|
|
can use these functions in C++.
ok millert@. also looked over by espie@, fgsch@.
|
|
|
|
|
|
|
|
uses -1, 0 and 1 which is one value too much for boolean.
Ok krw@, millert@
|
|
|
|
as has been done for the __sentinel__ attribute, since all toolchains support
it.
|
|
|
|
ldscripts). Correct the xd and xdc ldscripts (pie and pie + combreloc)
on i386 and add two new variants for pie + combreloc + Z and pie + Z on
all arches. ok drahn@
|
|
unbreaks libm on gcc2
ok miod@ (who created almost the same diff)
|
|
the OpenBSD-specific Mdocdate expansion. This is useful to avoid
conflicts that arise between the server's expansion of Mdocdate and
the upstream code when maintaining forked OpenBSD derived code on an
OpenBSD CVS server.
"I must grudgingly agree that this is neccessary" deraadt@
|
|
with several ports on alpha and does not produce a significant slowdown.
Suggestion to patch gcc directly from millert@
Diff and ok from miod@
ok pvalchev@, martin@, deraadt@
|
|
spotted by ajacoutot@, ok ajacoutot@, millert@
|
|
Crank max queue children in bulk config
Use the C flag in DAEMON_OPTIONS instead of nocanonify in bulk config
|
|
last update.
|
|
ok miod@
|
|
|
|
as the demangler provides new interfaces.
|
|
- pexecute implementation replaced with the older gcc 2.95 implementation,
as gcc 2.95's collect2 relies upon an implementation behaviour which has
been lost in further libiberty updates.
- keep the old hash table interfaces, which are used by gcc 2.95.
- keep the OpenBSD replacement for the md5 interface.
- keep the OpenBSD strl{cat,cpy} and snprinft usage.
Bump libiberty minor version due to the addition of a few new symbols.
|
|
identifies itself as the gcc 3 version...)
|
|
(longer an) RCS file in CVSROOT, both -- GNU cvs and OpenCVS -- segfaulted.
ok joris
|
|
anymore.
|
|
okay millert@
|
|
|
|
"looks okay" espie@
|
|
warnings and/or unexpected operation when such statements are multi-insn
macros, such as li/dli. Backported from gcc 3.3.6.
|
|
``[3.4 Regression] Incorrect optimisation of loop termination condition''
This is a complete backport of this commit:
2003-12-04 J"orn Rennecke <joern.rennecke@superh.com>
PR optimization/13260
* sh-protos.h (sh_expand_t_scc): Declare.
* sh.h (PREDICATE_CODES): Add cmpsi_operand.
* sh.c (cmpsi_operand, sh_expand_t_scc): New functions.
* sh.md (cmpsi): Use cmpsi_operand. If T_REG is compared to
something that is not a CONST_INT, copy it into a pseudo register.
(subc): Fix description of new T value.
(slt, sgt, sge, sgtu): Don't clobber T after rtl generation is over.
(sltu, sleu, sgeu): Likewise.
(seq, sne): Likewise. Use sh_expand_t_scc.
* PR target/15886
``[3.3/3.4/4.0 Regression] SH: Miscompilation with -O2 -fPIC''
This is a complete backport of this commit:
2004-06-11 J"orn Rennecke <joern.rennecke@superh.com>
* sh.h (ALLOCATE_INITIAL_VALUE): Use return_address_pointer_rtx.
* PR target/21255
``%R and %S are not safe to use from asms''
This is a complete backport of this commit:
2005-08-31 J"orn Rennecke <joern.rennecke@st.com>
PR target/21255
* sh.c (print_operand, %R and %S): Add handling of floating point
registers, memory, constants and invalid operands.
* Non PR
- CAN_ELIMINATE() is modified to prevent fp -> sp elimination when
a frame pointer needs to be preserved. I have serious reasons to
think this will help propolice operation.
|
|
``[3.3 only] set_fpscr isn't PIC even when __PIC__ defined; bad for
prelinking, causes glibc-3.2.3 check-textrel test to fail''
Correctly address fpscr_values[] in set_fpscr() when compiled PIC.
This is a partial backport of this commit:
2003-08-11 Kaz Kojima <kkojima@gcc.gnu.org>
* config/sh/lib1funcs.asm (__udivdi3): Add .type and .size
information in SHmedia case too.
(__divdi3, __umoddi3, __moddi3, __init_trampoline, __ic_invalidate):
Likewise.
(__set_fpscr): Use an access via GOT for PIC case.
|
|
|
|
|
|
GNU developers wear grey. GNU developers use autoconf and friends to configure
their software in the best possible way for the platform it is intended to run
on.
This frees us from the tedious job of, say, editing files hidden deep inside
the sources to, say, tell the code that 64-bit platforms (you know, those
which always define __LPA64__ and _LP64 among other things) have 64-bit
integer types (one such platform even defines the completely unexpected symbol
__mips64__.)
GNU developers work much harder than we do, because they're so frightfully
clever. I'm really awfully glad I'm a BSD developer.
|
|
#ifdefs. Found by ragge@
|
|
install.
Prompted and tested by drahn@
|
|
|
|
on some (not in-tree) configurations.
ok espie@ kettenis@
|
|
The sh architecture required some different linkage than most platforms,
This fixes those other requirements.
|
|
each cmsg_len (ie. msg_controllen = sum of CMSG_ALIGN(cmsg_len). This
works now that kernel fd passing has been fixed to accept a bit of
sloppiness because of this ABI repair.
lots of discussion with kettenis
|
|
sizeof(buffer) is the right size (alignments can mess it up). Code is
apparently not used, but it is better if all examples show the correct
idiom
ok hshoexer
|
|
of the writable sections. This is useful for garbage collectors such as
boehm-gc to locate these sections in a uniform way.
input and okay drahn@, miod@
|
|
okay otto@
|
|
an unknown flag specified in CVSROOT/modules.
found by tobias@, ok millert@
|
|
okay millert@
|
|
From Matthew Dillon@DragonFlyBSD
ok millert@ dlg@
|
|
default logic works better.
|
|
|
|
okay miod@
|
|
|
|
-1.
ok henning@ beck@ ray@
|
