Bernd Schmidt [Tue, 12 Jan 1999 00:28:21 +0000 (17:28 -0700)]
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
I will remember to check in ChangeLog files.
Bernd Schmidt [Wed, 6 Jan 1999 23:43:14 +0000 (16:43 -0700)]
combine.c (num_sign_bit_copies): In NEG...
* combine.c (num_sign_bit_copies): In NEG, MULT, DIV and MOD cases,
when a test can't be performed due to limited width of
HOST_BITS_PER_WIDE_INT, use the more conservative approximation.
Fix UDIV case for cases where the first operand has the highest bit
set.
And fix a couple copyrights.
expr.c (store_expr): If the lhs is a memory location pointed to be a postincremented (or...
�
* expr.c (store_expr): If the lhs is a memory location pointed
to be a postincremented (or postdecremented) pointer, always
force the rhs to be evaluated into a pseudo.
Jeffrey A Law [Wed, 30 Dec 1998 23:39:55 +0000 (23:39 +0000)]
cse.c (invalidate_skipped_block): Call invalidate_from_clobbers for each insn in the skipped block.
* cse.c (invalidate_skipped_block): Call invalidate_from_clobbers
for each insn in the skipped block.
* i386.md (doubleword shifts): Avoid namespace pollution.
Mike Stump [Mon, 23 Nov 1998 20:37:20 +0000 (15:37 -0500)]
libgcc2.c (top_elt): Remove top_elt, it isn't thread safe.
* libgcc2.c (top_elt): Remove top_elt, it isn't thread safe.
The strategy we now use is to pre allocate the top_elt along
with the EH context so that each thread has its own top_elt.
This is necessary as the dynmanic cleanup chain is used on the
top element of the stack and each thread MUST have its own.
(new_eh_context): Likewise.
(__sjthrow): Likewise.
Andrew MacLeod [Mon, 23 Nov 1998 08:52:34 +0000 (08:52 +0000)]
cplus-dem.c (demangle_prefix): Use the last "__" in the mangled name when looking for the signature.
Mon Nov 23 09:49:49 1998 Andrew MacLeod <amacleod@cygnus.com>
*cplus-dem.c (demangle_prefix): Use the last "__"
in the mangled name when looking for the signature. This allows
template names to begin with "__".
Jeffrey A Law [Sun, 1 Nov 1998 10:13:01 +0000 (10:13 +0000)]
linux.h (CPP_PREDEFINES): Bring back -Di386 for the last time.
* i386/linux.h (CPP_PREDEFINES): Bring back -Di386 for the last time.
* From Christian Gafton:
* i386/linux.h (CPP_PREDEFINES): Add -D__i386__.
* sparc/linux.h (CPP_PREDEFINES): Add -D__sparc__.
* sparc/linux64.h (CPP_PREDEFIENS): Add -D__sparc__.
Jeffrey A Law [Fri, 30 Oct 1998 23:41:35 +0000 (23:41 +0000)]
jump.c (jump_optimize): Initialize mappings from INSN_UID to EH region if...
* jump.c (jump_optimize): Initialize mappings from INSN_UID to
EH region if exceptions are enabled and we're performing cross
jump optimizations.
(find_cross_jump): Exit loop if the insns are in different EH regions.
David Edelsohn [Wed, 28 Oct 1998 19:18:09 +0000 (19:18 +0000)]
collect2.c (aix64_flag): New variable.
* collect2.c (aix64_flag): New variable.
(main, case 'b'): Parse it.
(GCC_CHECK_HDR): object magic number must match mode.
(scan_prog_file): Only check for shared object if valid header.
Print debugging if header/mode mismatch.
* README.RS6000: Update.
Jason Merrill [Tue, 27 Oct 1998 00:05:30 +0000 (00:05 +0000)]
expr.c (pending_chain): Move up.
* expr.c (pending_chain): Move up.
(save_expr_status): Do save pending_chain.
(restore_expr_status): And restore it.
* function.h (struct function): Add pending_chain.
Jason Merrill [Sun, 25 Oct 1998 17:35:47 +0000 (12:35 -0500)]
parse.y (condition): Convert VAR_DECL from reference to indirect reference.
* parse.y (condition): Convert VAR_DECL from reference to indirect
reference.
* decl2.c (validate_nonmember_using_decl): Fix using-directives of
std if std is ignored.
* method.c (build_decl_overload_real): Clear
numeric_output_need_bar after __.
* decl.c (cp_finish_decl): Make statics in extern inlines and
templates common, if possible and the target doesn't support weak
symbols.
* decl2.c (merge_functions): Remove duplicates.
* typeck.c (build_conditional_expr): Only fold if ifexp is an
INTEGER_CST.
Jason Merrill [Sun, 25 Oct 1998 16:52:48 +0000 (16:52 +0000)]
tinfo2.cc (fast_compare): Remove.
* tinfo2.cc (fast_compare): Remove.
(before): Just use strcmp.
* tinfo.cc (operator==): Just use strcmp.
* inc/typeinfo: Add #pragma interface.
(operator!=): Just call operator==.
* tinfo.cc: Add #pragma implementation.
(operator==): Move from inc/typeinfo and tinfo2.cc.
Jeffrey A Law [Fri, 23 Oct 1998 21:41:12 +0000 (21:41 +0000)]
m68k.md (5200 movqi): Do not allow byte sized memory references using address regs.
* m68k.md (5200 movqi): Do not allow byte sized memory references
using address regs.
* m68k.c (output_move_qimode): Do not use byte sized operations on
address registers.
Geoff Keating [Fri, 2 Oct 1998 00:26:21 +0000 (00:26 +0000)]
gcse.c: New definition NEVER_SET for reg_first_set...
* gcse.c: New definition NEVER_SET for reg_first_set, reg_last_set,
mem_first_set, mem_last_set; because 0 can be a CUID.
(oprs_unchanged_p): Use new definition.
(record_last_reg_set_info): Likewise.
(record_last_mem_set_info): Likewise.
(compute_hash_table): Likewise.
arm.c (add_constant): New parameter address_only, change caller.
* arm.c (add_constant): New parameter address_only, change caller.
Set it non-zero if taking the address of an item in the pool.
(arm_reorg): Handle cases where we need the address of an item in
the pool.
* arm.c (bad_signed_byte_operand): Check both arms of a sum in
a memory address.
* arm.md (splits for *extendqihi_insn and *extendqisi_insn): Handle
memory addresses that are not in standard canonical form.
Dave Love [Wed, 30 Sep 1998 02:59:33 +0000 (02:59 +0000)]
open.c: Back out part of last Netlib update affecting scratch files which clashed with...
1998-09-28 Dave Love <d.love@dl.ac.uk>
* libI77/open.c: Back out part of last Netlib update affecting
scratch files which clashed with the g77 variations and broke
implicit endfile on rewind.