Misc optimizations, a couple of bugfixes.
Merged doc/ChangeLog.
Ralph Campbell's MIPS support.
And any other changes I've been putting in, save whitespace/comment adjustments.
* config/obj-coffbfd.c (fill_section): Don't set vaddr here.
(write_object_file): Set it here instead, so that fixup_segment
can see the correct value.
as soon as I check in as.h.
(s_lcomm): Likewise with bss_section.
(s_fill): Use memset, not bzero.
whitespace/comment changes.
deleted some "#ifdef DONTDEF" sections relating to GDB symtabs.
* app.c, config/tc-*.c: Don't include read.h, since it is already
included by as.h.
* These are based on patches from Minh Tran-le
<mtranle@paris.intellicorp.com>.
* configure.in (i[34]86-ibm-aix*): Accept i486 for host. Use
obj_format coffbfd and gas_target i386coff for target.
(i[34]86-*-isc*): New host (uses sysv).
* config/i386aix.mt: Removed (no longer used).
* config/mh-i386aix (RANLIB): Use true rather than /bin/true.
(MINUS_G): Removed.
(LDFLAGS): Added, defined as -shlib.
* config/te-i386aix.h (REVERSE_SORT_RELOCS): Undefine.
* config/te-sco386.h (LOCAL_LABEL): Don't define.
(DOT_LABEL_PREFIX): Define.
* expr.c (operand): If DOT_LABEL_PREFIX, use .L0\001 as a label
name rather than L0\001.
* read.c (s_lcomm): Make a frag in SEG_BSS rather than using
local_bss_counter.
* symbols.c, symbols.h (local_bss_counter): Removed.
* write.c (write_object_file): bss no longer uses
local_bss_counter. Pass correct data and bss size to
VMS_write_object_file.
* config/obj-vms.c (VMS_write_object_file): Accept bss size as
argument, rather than using local_bss_counter.
* config/tc-m88k.c (s_bss): Don't use local_bss_counter.
* config/tc-sparc.c (s_reserve): Don't use local_bss_counter.
* config/obj-coffbfd.c (had_lineno, had_reloc): Removed.
(size_section): Restored sanity check.
(do_relocs_for): Base section address on s_paddr rather than
computing it. Adjust a29k R_IHIHALF special case to account for
section paddr (used to require paddr to be zero). If there are no
reclos, set s_relptr to 0. Set relocation size in object_headers.
(fill_section): Always set s_vaddr here, removing
ZERO_BASED_SEGMENTS case. Force s_scnptr for bss to 0. Don't set
NOLOAD for i386 .bss, because it confuses the SVR3 native linker.
Set STYP_INFO for .comment.
(coff_header_append): Use object headers and H_{SET,GET}_* macros.
Make aouthdr writing depend on OBJ_COFF_OMIT_OPTIONAL_HEADER.
(crawl_symbols): Handle 8 character section name correctly. Use
H_{SET,GET}_* macros.
(do_linenos_for): Set lineno size in object_headers.
(write_object_file): Use H_{SET,GET}_* macros. Don't bother to
set s_vaddr here. If string_byte_count remains 4, set it back to
0, and only write strings out if there are some. Call
fill_section before do_relocs_for and do_linenos_for.
(obj_coff_section): Handle optional quoted second argument giving
section characteristics.
(obj_coff_bss): Added to handle .bss.
(obj_coff_ident): Added to handle .ident (puts string in .comment
section).
(obj_coff_lcomm): Put common symbols in .bss, not .data.
(fixup_mdeps): Change to segment. Call frag_wane after
md_convert_frag.
(fixup_segment): Explicitly check S_IS_COMMON before making 386
adjustment (already happened only for common symbols, but this is
clearer).
* config/obj-coffbfd.h (OBJ_COFF_OMIT_OPTIONAL_HEADER): Define.
* config/tc-i386.c (s_bss): Don't use if I386COFF.
(md_pseudo_table): Ignore .optim and .noopt.
(tc_coff_sizemachdep): New function.
* config/tc-i386.h (REVERSE_SORT_RELOCS): Undef, for SVR3
compatibility.
(LOCAL_LABEL): Removed definition.
(DOT_LABEL_PREFIX): Defined.
* listing.c: Call xmalloc, not malloc; don't declare malloc.
* Changes to support SCO 3.2v4:
* read.c (s_align_bytes, s_align_ptwo): If not SEG_DATA or
SEG_BSS, fill with NOP_OPCODE.
* config/i386coff.mt: Add opcode/i386.h to TARG_CPU_DEPENDENTS.
* config/obj-coffbfd.c (do_relocs_for): Increment addr even if not
using ZERO_BASED_SEGMENTS.
(fill_section): If ZERO_BASED_SEGMENTS, set segment addresses, but
never set segment address for SEG_E2 (.bss) and don't write out
SEG_E2 contents. Set .init and .fini sections to STYP_TEXT.
(obj_coff_endef): Don't merge labels or symbols awaiting forward
definitions, and don't merge tags with non-tags. Check for .bf
rather than just checking whether the second character is b and
the third character is f.
(obj_coff_val): gcc can generate values which we don't handle
correctly; discard information for now, since it only affects the
debugging information.
(tag_find_or_name): Don't insert tags in the symbol table.
(yank_symbols): Don't merge labels.
(write_object_file): Don't define SUB_SEGMENT_ALIGN if it is
already defined. Fill subsegments with NOP_OPCODE, not 0. Don't
set segment address if ZERO_BASED_SEGMENTS.
(obj_coff_section): Accept and ignore a trailing quoted string, as
used in AT&T i386 syntax.
(fixup_segment): Take segment as argument. On the i386, adjust PC
relative addends by the segment vaddr.
* tc-i386.h: Define SUB_SEGMENT_ALIGN.
* tc-a29k.h: Define ZERO_BASED_SEGMENTS.
* tc-i386.c: (i386_operand): If I386COFF, accept any segment type.
Deleted some unused code.
Fixed some declarations to use PARAMS macro.
Fixed up configure.in for new targets.
Some whitespace/comment fixes.
Merged config/ChangeLog.
Removed some unneeded files.
obj-coff.c (obj_coff_endef): Use as_warn, not fprintf.
tc-m68k.c (md_assemble): 68000+68881 is okay -- could be emulating.
p3 into devo.
* tc-m68k.c (m68k_reg_parse): If REGISTER_PREFIX isn't defined,
still accept (but don't require) OPTIONAL_REGISTER_PREFIX before
the register name.
* tc-m68k.h (OPTIONAL_REGISTER_PREFIX): Define to be "%", if not
M68KCOFF.
support when using /opt/SUNWspro/SC2.0/acc.
Tue Nov 10 09:50:25 1992 Ian Lance Taylor (ian@cygnus.com)
* obj-coffbfd.c (fill_section): set STYP_NOLOAD bit for .bss
section.
* atof-ieee.c, atof-ns32k.c, tc-*.c: made EXP_CHARS, FLT_CHARS,
comment_chars, line_comment_chars and line_seperator_chars
consistently const, and always initialized them. Included read.h.
* Support for i386-sysv.
obj-coffbfd.c (do_relocs_for, write_object_file): set segment
addresses to reasonable sizes. New define ZERO_BASED_SEGMENTS can
be used to set them all to zero as was done before.
(fill_section): segment addresses now set in write_object_file.
(fill_section): Don't set STYP_NOLOAD for .bss section.
(fixup_segment): 386 uses strange common symbol format.
tc-i386.c (tc_coff_fix2rtype): use R_DIR32, not R_RELLONG, for
compatibility with SVR3.2 linker.
* obj-coff.c (c_dot_file_symbol, obj_coff_ln, obj_coff_line):
support for C source listings. (obj_coff_endef): look in the right
part of the symbol for the symbol name
* tc-i960.h, tc-i960.c: avoid the ANSI preprocessor addition
#elif, since it is not supported by old compilers.
ho-rs6000.h, tc-m68k.c: the native RS/6000 compiler miscompiles a
couple of expressions in tc-m68k.c.
* tc-m68k.h: if M68KCOFF, define DOT_LABEL_PREFIX (to require
local labels to start with a .) and set REGISTER_PREFIX to %.
tc-m68k.c (m68k_reg_parse): accept REGISTER_PREFIX if defined.
* obj-coffbfd.c (do_relocs_for,fill_section): now allocate all
sections starting from zero, rather than making them consecutive.
This makes subsequent reloc calculations easier, esp if the
object format doesn't understand addends. (obj_coff_lcomm): (maybe
temporarily) allocate lcomm in .data rather than in .bss. It
seems that some tools can't cope with a non-zero sized bss
before linkage.
* Added WARN_SIGNED_OVERFLOW_WORD define to give an error if any
.word is < -32768 or > 32767. The -J flag causes the error to be
ignored. This is to catch over-sized switches generated by gcc on
systems which don't support the broken .word hack.
as.c (main): permit -J if WARN_SIGNED_OVERFLOW_WORD.
write.c (fixup_segment): check for signed .word overflow if
WARN_SIGNED_OVERFLOW_WORD.
* input-scrub.c (input_scrub_push): call input_file_begin, not
input_scrub_begin.
messages.c (as_perror): print ": " between the passed in error and
the strerror, like perror does.
* tc-m68k.c (m68k_ip, m68k_ip_op, get_num, try_moto_index): merge
Motorola and MIT syntax; gas can now assemble either type of
file.
tc-m68kmote.c, tc-m68kmote.h: removed now superfluous files.
From Steve Chamberlain:
m68kcoff.mt: for m68k COFF.
obj-coffbfd.c: (fixup_mdeps) added
(size_section) removed bad sanity check
(fill_section) added rs_machine_dependent case
(write_object_file) call fixup_mdeps
(fixup_segment) set fx_subsy to 0.
obj-coffbfd.h: define WORKING_DOT_WORD (too hard to support) and
handle m68k.
tc-m68k.c, config/tc-m68k.h: added m68k COFF support and Motorala
pseudo ops.
* Makefile.in: use gas_target instead of modifying target_cpu.
From Steve Chamberlain:
Makefile.in: Handle m68*-*-coff*.
read.c, read.h: add mult argument to s_space
* obj-bout.h, obj-bout.c (obj_header_append, obj_symbol_to_chars),
tc-i960.c (md_ri_to_chars): Always output bout object file in
little endian byte order (used to use endianness of host).
data, so that "bc" is available to 68040 cache instructions. Added
"tt0", "tt1", and 68ec030 variants.
(md_assemble): Complain if 68000 (only) and 68881 are specified.
(enum _register): Added TT0, TT1.
(m68k_ip, cases '3' and 't'): Handle new operand type codes. Pass
line number correctly in "internal error" messages. Don't print
architecture-mismatch message for operand errors.
From Colin Smith (colin@wrs.com):
* config/tc-m68k.c (m68k_ip, case '_'): Use addword twice rather than
install_operand.
data, so that "bc" is available to 68040 cache instructions. Added
"tt0", "tt1", and 68ec030 variants.
(md_assemble): Complain if 68000 (only) and 68881 are specified.
(enum _register): Added TT0, TT1.
(m68k_ip, cases '3' and 't'): Handle new operand type codes. Pass
line number correctly in "internal error" messages. Don't print
architecture-mismatch message for operand errors.
From Colin Smith (colin@wrs.com):
* config/tc-m68k.c (m68k_ip, case '_'): Use addword twice rather than
install_operand.
* tc-m68k.c: use TARGET_CPU to choose default cpu type.
* te-generic.h: default to LOCAL_LABELS_DOLLAR and LOCAL_LABELS_FB
so that we can assemble hand-written libgcc code.
* messages.c (as_warn): Use fputs, not fprintf, with a buffer that
has already been formatted (but may still contain %-characters).
(as_bad): Likewise.
Wed Aug 19 11:20:59 1992 Ian Lance Taylor (ian@cygnus.com)
* tc-m68k.c, tc-m68kmote.c: the cas2 instruction is supposed to be
written with indirection on the last two operands, which can be
either data or address registers. Added a new operand type 'r'
which accepts either register type. Added '(' to notend stuff in
tc-m68kmote.c to accept (a0):(a2) in cas2 instruction.
Tue Aug 18 14:16:38 1992 Ian Lance Taylor (ian@cygnus.com)
* Makefile.in: always create installation directories. Removed
MINUS_G, set CFLAGS to default to -g, added FLAGS_TO_PASS, passed
FLAGS_TO_PASS to recursive makes.
assembling for an older CPU than a 68020.
This bug was found at Adobe because GAS did not diagnose incorrect
68020 code generated by GCC when compiling for 68010. We had to debug it
in the target system as failing instructions.
processor/opcode mismatch, so reword the error message.
(md_assemble): If no CPU has been set (even if FPU/PMMU
characteristics have been), default to 68020. Don't need extra
quotes around error string.
* configure.in (mips host): Accept "ultrix" with version number.
* expr.c (floating_constant): Separate "=-" to avoid confusing
ancient or broken compilers.
* expr.c (floating_constant): Separate "=-" to avoid confusing
ancient or broken compilers.
* config/tc-m68k.c (m68k_ip): Mismatch error could also indicate
processor/opcode mismatch, so reword the error message.
(md_assemble): If no CPU has been set (even if FPU/PMMU
characteristics have been), default to 68020. Don't need extra
quotes around error string.
(2) structural changes triggered thereby
(3) verify m4 and makeinfo run correctly on all versions of the doc
represented in the Makefile and m4 config definition files
as.texinfo: disabled .lsym
exploit makeinfo node defaulting
removed some conditional hair around nodes
fixed some bad conditionals in a29k config
a29k.m4, i960.m4, sparc.m4, vax.m4: supply missing switch definitions
(looks like these hadn't been tested for a while)
as-all.texinfo: mechanically derived from above
* read.c (potable): align should be nbytes, not ptwo!
* write.c (write_object_file): extra glue for new bss attributes
(relax_segment): SEG_BSS is ok now
* config/tc-m68k.c (m68k_ip_op): can now parse more @( modes
* app.c (process_escape): new function to handle escapes the right
way, (do_scrub_next_char): use new function
* cond.c (s_ifdef): do ifdef/ifndef right
* read.c (s_fill): make the , expressions optional like the doc
says
* config/tc-h8300.[ch]: better warnings
as data0_frchainP
* write.c (write_object_file): various changes to handle data in
the BSS segment in much the same was as stuff in the DATA segment.
* subsegs.c (subseg_change): allow and handle a change into SEG_BSS.
* tc-m68k.c (m68kip): Fix typo so that only arch's >=68020 do
pcrel data stuff. (md_estimate_size_before_relax): when relaxing a
68010 bxx into a bra+6 jmpxx, put the bytes of the jmp opcode into
the right place. (s_bss): Don't put .bss stuff into SEG_DATA, put
it into SEG_BSS
* expr.c(expr): allow SEG_REGISTER in expressions.
* read.c(pseudo_set): register expressions can be the source of a
set.
* subsegs.c (subseg_new): Now -R forces all changes to SEG_DATA to
goto SEG_TEXT (if a.out)
* write.c (write_object_file): If a.out don't use the old way for
-R.
* config/obj-a.out (s_sect): complain if the user tries to use a
subsegment with a value which might interfere with out -R hackery.
* config/tc-m68k.c (m68k_reg_parse): lookup names in symbol table
rather than use ugly if tree. (init_regtable): insert register
names into symbol table.
* expr.c: remove limitation that local_labels_dollar or
local_labels_fb must be < 10.
* symbols.c: remove local_labels_dollar, replace with a function
interface for a sparse array. All users adjusted.
checked in earlier. These are Makefiles relevant to the progressive
release currently in progress.
* added exec_prefix
* default bindir now exec_prefix/bin
* default libdir now exec_prefix/lib
* default mandir now prefix/man
* default infodir now prefix/info
* added oldincludedir, default NULL
* remove all traces of the old -subdirs and -namesubdir support,
including code, for loops, and all subdir/unsubdir stuff.
* remove old style staging targets {de-}stage[1234], comparison,
bootstrapN, from all but the bootstrappable tools.
* added check targets.
* removed any "Last Mod" lines.
* forced copyrights to '92 and shifted some from Cygnus to FSF.
* added tooldir and program_prefix where needed.
* updated texinfo and man pages, removed all references to multiple
hosts or targets.
* expr.c: LOCAL_LABELS_FB had been changed to lower case - so
local labels didn't work.
* listing.c (list_symbol_table): don't core dump when there's no
symbol there.
* write.c, write.h: call fix_new with the right number of args on
the H8.
* config/tc-h8300.[ch] : fix bugs reported by HMSI, and make
errors nicer
* as.c: fix typo recently introduced.
* as.h : Don't include aout/reloc.h - it's not right for COFF!
* expr.c: Much rewriting, to accomodate MRI syntax for
expressions. Also easier to read now.
* listing.c: Put back defuns
* read.c: modified to accept MRI syntax, put back listing pseudo
ops so that an assembler built with NO_LISTING ignores list ops
rather than pukes.
* write.c, write.h: fixs - only keep a reloc type in a fix if the target
machine is a SPARC or a 29K.
* config/obj-aout.c: added s_sect pseudo op
* config/obj-coffbfd.c: lints, set the filehdr flags right and
fill in the timestamp.
* config/obj-coffbfd.h: Since we don't include aout/reloc.h
anymore, define all the relocs which the tc-<x> bit will use so we
can translate from them to the coff types.
* config/tc-a29k.c: reloc_type isn't ane enum any more
* config/tc-m68k.c: Added NO_RELOC definition.
Now compiles for sparc aout, 68k aout (MRI and MIT syntax),
29k coff.
So far works as replacement for sparc and 68k /bin/as.
Jim Wilson
Ken Raeburn
K. Richard Pixley
Ian Lance Taylor
David D. Zuhn
Steve Chamberlain
Roland Pesch
Mark Eichin
John Gilmore
Brendan Kehoe
Sean Eric Fagan
Per Bothner
David Henkel-Wallace
Michael Tiemann