* java/net/natInetAddress.cc (lookup): On glibc2.0 systems, make
buffer larger to work around bug.
From Bryce McKinlay <bryce@albatross.co.nz>.
From-SVN: r26447
* alpha.c (aligned_memory_operand): Recognize the output of
LEGITIMIZE_RELOAD_ADDRESS. Examine reg_equiv_memory_loc in
the event of a pseudo.
(unaligned_memory_operand): Likewise. Don't otherwise accept
completely illegal addresses.
(normal_memory_operand): Likewise. Handle subregs of pseudos.
(get_aligned_mem): Revert previous change. Abort if we don't have a
mem. During reload, call find_replacement on all illegal memories.
(get_unaligned_address): Likewise.
* alpha.h (SECONDARY_INPUT_RELOAD_CLASS): Use !aligned_memory_operand
instead of unaligned_memory_operand.
* alpha.md: Revert extra argument to get_aligned_mem.
(reload_inqi): Use any_memory_operand in constraints. Abort if
we're not given some sort of mem.
(reload_inhi): Likewise.
(reload_outqi, reload_outhi): Likewise.
From-SVN: r26445
1999-04-14 12:37 -0400 Zack Weinberg <zack@rabi.columbia.edu>
* cpperror.c, cppexp.c, cpplib.c: Never call abort.
* cpphash.c: Only call abort when we detect corruption of the
malloc arena.
* cppmain.c: Don't define fatal or fancy_abort.
From-SVN: r26442
Jan Hubicka <hubicka@freesoft.cz>
* i386.c (x86_adjust_cost): Agi stall takes 1 cycle on Pentium, fst
requires value to be ready one extra cycle.
From-SVN: r26441
1999-04-14 Andrew Haley <aph@cygnus.com>
* java/lang/natDouble.java (doubleToLongBits): ensure that all
NaNs are always converted to the same long value.
* java/lang/natFloat.java (floatToIntBits): ditto, but for float
converted to int.
From-SVN: r26439
* i386.md (SImode logical compare): Avoid outputing non-pariable testw
and testl on Pentium.
(register and memory bit tests): Likewise.
(setcc, normal and reversed conditional branches): Use shorter
sequence for testing flags stored in EAX.
From-SVN: r26437
* i386.md (xorsi3): Do not output NOT instrctions on Pentium.
(xorqi3): Likewise.
(xorhi3): Likewise.
(notsi2): Likewise.
(notqi2): Likewise.
(nothi2): Likewise; do not output prefixed opcodes when possible.
From-SVN: r26434
* class.c (finish_struct_1): Look at the const-ness of the field's
type, not the TREE_READONLY-ness of the declaration.
* method.c (synthesize_method): Likewise.
* pt.c (tsubst_decl): Call c_apply_type_quals_to_decl when
creating new declarations.
From-SVN: r26420
* c-common.c (default_valid_lang_attribute): New fn.
(valid_lang_attribute): New callback ptr.
(decl_attributes): Call it. Move init_priority support into
C++ frontend.
From-SVN: r26419
* decl2.c (import_export_decl): Because vtables always reference
virtual functions, even if they are inlined, don't allow
-fno-implement-inlines to not emit them, instead, emit them with
the vtable.
* decl.c (start_function): Likewise.
Fixes vtable1.C
From-SVN: r26416
* config/c4x/c4x.md (storeqf_int, storeqf_int_clobber, loadqf_int,
loadqf_int_clobber): Add new patterns with corresponding splitters
to handle moves of floating point values into and out of intager
registers by using memory.
* config/c4x/c4x.c (c4x_check_legit_addr): Disallow PRE_INC for modes
other than QFmode and QImode.
(mixed_subreg_operand): New function.
(c4x_emit_move_sequence): If moving a floating point value into or
out of an integer register, use the new patterns storeqf_int_clobber
or loadqf_int_clobber.
(reg_imm_operand, *_reg_operand): Call reg_operand instead of
register_operand.
(reg_operand, src_operand): Disallow operand if it satisifes
mixed_subreg_operand.
* config/c4x/c4x.h (mixed_subreg_operand): Add prototype.
From-SVN: r26413
* pa.c: Use a register name, not a raw immediate in branch,
compare/clear, sub, subb, uaddcm and vshd instructions.
* pa.md: Likewise.
From-SVN: r26404
* pa.c: Remove space register specification in memory addresses,
except where it is actually needed.
* pa.md, pa.h, ee.asm, ee_fp.asm, lib2funcs.asm: Likewise.
From-SVN: r26398
