1 Thu Sep 10 14:48:59 1998 Martin von Löwis <loewis@informatik.hu-berlin.de>
3 * invoke.texi (C++ Dialect Options): Document -fhonor-std.
5 Thu Sep 10 01:38:05 1998 Jeffrey A Law (law@cygnus.com)
7 * reg-stack.c (straighten_stack): Do nothing if the virtual stack is
8 empty or has a single entry.
10 * toplev.c (rest_of_compilation): Open up the dump file for reg-stack
11 before calling reg_to_stack.
13 Thu Sep 10 00:03:34 1998 Richard Henderson <rth@cygnus.com>
15 * alpha.c (alphaev5_insn_pipe): Abort on default case.
16 (alphaev5_next_group): Swallow CLOBBERs and USEs.
18 * c-tree.h (warn_long_long): Declare it.
20 Wed Sep 9 23:31:36 1998 (Stephen L Moshier) <moshier@world.std.com>
22 * emit-rtl.c (gen_lowpart_common): Disable optimization of
23 initialized float-int union if the value is a NaN.
25 Wed Sep 9 23:00:48 1998 Nathan Sidwell <nathan@acm.org>
27 * c-lex.c (real_yylex): Don't warn about long long constants if
28 we're allowing long long
30 Wed Sep 9 21:58:41 1998 Bernd Schmidt <crux@pool.informatik.rwth-aachen.de>
32 * except.h (current_function_eh_stub_label): Declare.
33 (current_function_eh_old_stub_label): Declare.
34 * function.h (struct function): New members eh_stub_label and
36 * except.c (current_function_eh_stub_label): New variable.
37 (current_function_eh_old_stub_label): New variable.
38 (init_eh_for_function): Clear them.
39 (save_eh_status): Save them.
40 (restore_eh_status): Restore them.
41 (expand_builtin_eh_stub): Set current_function_eh_stub_label.
42 (expand_builtin_eh_stub_old): Set current_function_eh_old_stub_label.
43 * flow.c (find_basic_blocks_1): When handling a REG_LABEL note, don't
44 make an edge from the block that contains it to the block starting
45 with the label if this label is one of the eh stub labels.
46 If eh stub labels exist, show they are reachable from the last block
49 * reload1.c (reload): Break out several subroutines and make some
51 (calculate_needs_all_insns): New function, broken out of reload.
52 (calculate_needs): Likewise.
53 (find_reload_regs): Likewise.
54 (find_group): Likewise.
55 (find_tworeg_group): Likewise.
56 (something_needs_reloads): New global variable, formerly in reload.
57 (something_needs_elimination): Likewise.
58 (caller_save_spill_class): Likewise.
59 (caller_save_group_size): Likewise.
60 (max_needs): Likewise.
61 (group_size): Likewise.
62 (max_groups): Likewise.
63 (max_nongroups): Likewise.
64 (group_mode): Likewise.
65 (max_needs_insn): Likewise.
66 (max_groups_insn): Likewise.
67 (max_nongroups_insn): Likewise.
70 * print-rtl.c (print_rtx): For MEMs, print MEM_ALIAS_SET.
72 Wed Sep 9 13:14:41 1998 Richard Henderson <rth@cygnus.com>
74 * loop.c (load_mems): Copy rtx for output mem.
76 Wed Sep 9 15:16:58 1998 Gavin Romig-Koch <gavin@cygnus.com>
78 * mips/abi64.h (LONG_MAX_SPEC): Don't set LONG_MAX for
79 mips1 or mips2 either.
81 Wed Sep 9 12:31:35 1998 Jeffrey A Law (law@cygnus.com)
83 * pa.c (pa_reorg): New marking scheme for jumps inside switch
85 (pa_adjust_insn_length): Update to work with new marking scheme
86 for jumps inside switch tables.
87 * pa.md (switch_jump): Remove pattern.
88 (jump): Handle jumps inside jump tables.
90 * Makefile.in (profile.o): Depend on insn-config.h
92 Wed Sep 9 09:36:51 1998 Jim Wilson <wilson@cygnus.com>
94 * iris6.h (DWARF2_UNWIND_INFO): Undef.
96 Wed Sep 9 01:32:01 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
98 Add preliminary native sparcv9 Solaris support.
99 * configure.in: Recognize sparv9-*-solaris2*
101 * config.sub: Recognize sparcv9 just like sparc64.
102 * config/sparc/sol2-c1.asm config/sparc/sol2-ci.asm
103 config/sparc/sol2-cn.asm: Macroize so it can be shared between
104 32-bit and 64-bit Solaris systems.
105 * config/sparc/t-sol2: Assemble those with cpp.
106 * config/sparc/sparc.h (TARGET_CPU_sparcv9): New alias for v9.
107 (*TF*_LIBCALL): If ARCH64 use V9 names.
108 * config/sparc/{xm-sysv4-64,sol2-sld-64}.h: New files.
110 Wed Sep 9 01:07:30 1998 Jakub Jelinek <jj@sunsite.ms.mff.cuni.cz>
112 * config/sparc/sparc.h (TARGET_CM_MEDMID): Fix documentation.
113 (CASE_VECTOR_MODE): Set to SImode even if PTR64, when MEDLOW and
115 (ASM_OUTPUT_ADDR_{VEC,DIFF}_ELT): Check CASE_VECTOR_MODE not
117 * config/sparc/sparc.md (tablejump): Likewise, and sign extend op0
118 to Pmode if CASE_VECTOR_MODE is something else.
120 Wed Sep 9 00:10:31 1998 Jeffrey A Law (law@cygnus.com)
122 * prefix.c (update_path): Correctly handle cases where PATH is
123 a substring of the builtin prefix, but specifies a different
126 Tue Sep 8 23:46:04 1998 Hans-Peter Nilsson <hp@axis.se>
128 * expr.c: Corrected comment about what MOVE_RATIO does.
129 * config/alpha/alpha.h: Ditto.
130 * config/1750a/1750a.h: Ditto.
131 * config/clipper/clipper.h: Ditto.
132 * config/i386/i386.h: Ditto.
134 Tue Sep 8 22:56:12 1998 Jeffrey A Law (law@cygnus.com)
136 * configure.in (m68k-next-nextstep3*): Use collect2.
137 Similarly for x86 NeXT configurations.
138 * configure: Rebuilt.
140 Tue Sep 8 01:38:57 1998 Nathan Sidwell <nathan@acm.org>
142 * configure.in: Don't assume srcdir is .../gcc
143 * configure: Rebuilt.
145 Sat Sep 5 16:34:34 EDT 1998 John Wehle (john@feith.com)
147 * global.c: Update comments.
148 (global_alloc): Assign allocation-numbers
149 even for registers allocated by local_alloc in case
150 they are later spilled and retry_global_alloc is called.
151 (mark_reg_store, mark_reg_clobber,
152 mark_reg_conflicts, mark_reg_death): Always record a
153 conflict with a pseudo register even if it has been
154 assigned to a hard register.
155 (dump_conflicts): Don't list pseudo registers already assigned to
156 a hard register as needing to be allocated, but do list their
158 * local-alloc.c: Update comment.
160 Mon Sep 7 23:38:01 1998 Jeffrey A Law (law@cygnus.com)
162 * configure.in: Check for bogus GCC_EXEC_PREFIX and LIBRARY_PATH.
163 * configure: Rebuilt.
165 Mon Sep 7 22:41:46 1998 Michael Meissner <meissner@cygnus.com>
167 * rs6000.c (rs6000_override_options): Fix name for ec603e, to add
169 * t-ppccomm (MULTILIB_MATCHES_FLOAT): Add support for -mcpu=xxx
170 for all targets that set -msoft-float.
172 Mon Sep 7 23:30:07 1998 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
174 * toplev.c (print_switch_values): Make static to match prototype.
176 Mon Sep 7 19:13:59 1998 Jeffrey A Law (law@cygnus.com)
178 * configure.in: If we are unable to find the "gnatbind" program,
179 then do not configure the ada subdir.
180 * configure: Rebuilt.
182 Sun Sep 6 14:03:58 PDT 1998 Jeff Law (law@cygnus.com)
184 * version.c: Bump for snapshot.
186 Sun Sep 6 13:28:07 PDT 1998 Jeff Law (law@cygnus.com)
188 * version.c: Bump for snapshot.
190 Sun Sep 6 08:54:14 1998 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
192 * Makefile.in (toplev.o): Depend on $(EXPR_H).
193 (insn-extract.o, insn-attrtab.o): Depend on toplev.h.
195 * gansidecl.h: Define ATTRIBUTE_NORETURN.
197 * genattrtab.c: Have insn-attrtab.c include toplev.h.
199 * genextract.c: Have insn-extract.c include toplev.h.
201 * rtl.h: Don't prototype `fatal_insn_not_found' and `fatal_insn'.
203 * toplev.c: Include expr.h.
204 (really_sorry, fancy_abort): Remove prototypes.
205 (set_target_switch): Add argument in prototype.
206 (vfatal): Mark prototype with ATTRIBUTE_NORETURN.
207 (v_really_sorry): Likewise.
208 (print_version, print_single_switch, print_switch_values): Make
209 static and add prototype arguments.
210 (decl_printable_name): Add prototype arguments.
211 (lang_expand_expr_t): New typedef.
212 (lang_expand_expr): Declare as a lang_expand_expr_t.
213 (incomplete_decl_finalize_hook): Add prototype argument.
214 (decl_name): Mark variable `verbosity' with ATTRIBUTE_UNUSED.
215 (botch): Likewise for variable `s'.
216 (rest_of_type_compilation): Mark variables `type' and `toplev'
217 with ATTRIBUTE_UNUSED if none of DBX_DEBUGGING_INFO,
218 XCOFF_DEBUGGING_INFO or SDB_DEBUGGING_INFO are defined.
219 (display_help): Make variable `i' an `unsigned long'.
220 (main): Remove unused parameter `envp'.
221 Cast assignment to `lang_expand_expr' to a `lang_expand_expr_t'.
222 Cast -1 when comparing it with a `size_t'.
224 * toplev.h (fatal, fatal_io_error, pfatal_with_name): Mark
225 prototype with ATTRIBUTE_NORETURN.
226 (fatal_insn_not_found, fatal_insn, really_sorry,
227 push_float_handler, pop_float_handler): Add prototypes.
228 (fancy_abort): Mark prototype with ATTRIBUTE_NORETURN.
229 (do_abort, botch): Add prototypes.
231 Sat Sep 6 12:05:18 1998 John Carr <jfc@mit.edu>
233 * final.c (final): If a label is reached only from a single jump,
234 call NOTICE_UPDATE_CC on the jump and its predecessor before
235 emitting the insn after the label.
237 * i386.h: Add AMD K6 support.
238 Change TARGET_* macros to use table lookup.
239 (INITIALIZE_TRAMPOLINE): Improve trampoline code.
240 (ADJUST_COST): Change definition to call function in i386.c.
241 (ISSUE_RATE): Define as 2 for anything newer than an 80486.
242 * i386.c: Add AMD K6 support.
243 Add constants for feature tests used by TARGET_* macros.
244 (split_di): If before reload, call gen_lowpart and gen_highpart.
245 (x86_adjust_cost): New function.
246 (put_jump_code): New function.
247 (print_operand): New codes 'D' and 'd'.
248 * i386.md: New insn types. New insn attribute "memory".
249 Redefine scheduling parameters to use new types and add AMD K6
250 support. Explicitly set type of most insns.
251 (move insns): K6 prefers movl $0,reg to xorl reg,reg. Pentium
252 Pro and K6 prefer movl $1,reg to incl reg.
253 (adddi3, subdi3): Set cc_status.
254 (DImode shift patterns): Change label counters from HOST_WIDE_INT
255 to int; x86 can't have more than 2^31 DImode shifts per file.
256 (setcc): Combine all setcc patterns. Allow writing memory.
257 Combine all jump patterns using match_operator.
258 (*bzero): Name pattern. Emit mutliple stos instructions when that
259 is faster than rep stos.
260 (xordi3, anddi3, iordi3): Simplify DImode logical patterns and
263 Sun Sep 6 11:17:20 1998 Dave Love <d.love@dl.ac.uk>
265 * config/m68k/x-next (BOOT_LDFLAGS): Define suitably for f771
268 Sat Sep 5 22:05:25 1998 Richard Henderson <rth@cygnus.com>
270 * alpha.c (alpha_ra_ever_killed): Inspect the topmost sequence,
271 not whatever we're generating now.
273 * alpha.c (set_frame_related_p, FRP): New.
274 (alpha_expand_prologue): Mark frame related insns.
275 (alpha_expand_epilogue): Likewise, but with a null FRP.
276 * alpha.h (INCOMING_RETURN_ADDR_RTX): New.
277 * alpha.md (exception_receiver): New.
278 * alpha/crtbegin.asm (.eh_frame): New beginning.
279 (__do_frame_setup, __do_frame_takedown): New.
280 * alpha/crtend.asm (.eh_frame): New ending.
281 * alpha/elf.h (DWARF2_DEBUGGING_INFO): Define.
282 (ASM_SPEC): Don't emit both dwarf2 and mdebug.
283 (ASM_FILE_START): Don't emit .file for dwarf2.
285 * rtl.h (enum reg_note): Add REG_FRAME_RELATED_EXPR.
286 * rtl.c (reg_note_name): Likewise.
287 * rtl.texi (REG_NOTES): Likewise.
288 * dwarf2out.c (dwarf2out_frame_debug): Use it. Recognize a store
291 Sat Sep 5 14:47:17 1998 Richard Henderson <rth@cygnus.com>
293 * i386.h (PREFERRED_RELOAD_CLASS): Standard fp constants load to TOS.
294 * i386.md (movsf, movdf, movxf): Validate memory address returned
295 from force_const_mem. Kill useless REG_EQUAL setting code.
297 Sat Sep 5 14:23:31 1998 Torbjorn Granlund <tege@matematik.su.se>
299 * m68k.md (zero_extendsidi2): Fix typo.
301 Sat Sep 5 13:40:24 1998 Krister Walfridsson <cato@df.lth.se>
303 * configure.in: Removed references to the removed file
304 * config/xm-netbsd.h. Use ${cpu_type}/xm-netbsd.h for
305 * arm*-*-netbsd* and ns32k-*-netbsd*.
306 * config/i386/xm-netbsd.h: Removed unnecessary file.
307 * config/m68k/xm-netbsd.h: Likewise.
308 * config/sparc/xm-netbsd.h: Likewise.
309 * config/mips/xm-netbsd.h: Likewise.
311 Sat Aug 29 13:32:58 1998 Mumit Khan <khan@xraylith.wisc.edu>
313 * i386/cygwin32.h (BIGGEST_ALIGNMENT): Define.
314 (PCC_BITFIELD_TYPE_MATTERS): Define to be 0.
316 * i386/cygwin32.h (ASM_OUTPUT_SECTION_NAME): Don't check for
317 for exact section attributions.
319 * i386/mingw32.h (CPP_PREDEFINES): Add __MSVCRT__ for msvc
321 * i386/crtdll.h (CPP_PREDEFINES): Define.
323 Sat Sep 5 03:23:05 1998 Jeffrey A Law (law@cygnus.com)
325 * m68k.md (5200 movqi): Do not allow byte sized memory references
327 * m68k.c (output_move_qimode): Do not use byte sized operations on
330 * Makefile.in (pexecute.o): Use pexecute.c from libiberty. Provide
331 explicit rules for building. Similarly for alloca, vfprintf,
332 choose-temp and mkstemp, getopt, getopt1, and obstack.
333 (INCLUDES): Add $(srcdir)/../include.
334 * pexecute.c, alloca.c, vfprintf.c, choose-temp.c, mkstemp.c: Delete.
335 * getopt.h, getopt.c getopt1.c, obstack.c, obstack.h: Likewise.
337 Fri Sep 4 11:57:50 1998 Tom Tromey <tromey@cygnus.com>
339 * gcc.c (do_spec_1): [case 'o'] Account for
340 lang_specific_extra_outfiles.
341 (main): Correctly clear all slots in outfiles for
342 lang_specific_extra_outfiles. Set input_file_number before
343 calling lang_specific_pre_link.
345 Fri Sep 4 10:37:07 1998 Jim Wilson <wilson@cygnus.com>
347 * loop.c (load_mems): Fix JUMP_LABEL field after for_each_rtx call.
349 Fri Sep 4 02:01:05 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
351 * config/sparc/sparc.c (output_double_int): In all V9 symbolic
353 (sparc_output_deferred_case_vectors): If no work to do, return.
354 Fix thinko in Sept 1 change.
356 1998-09-03 SL Baur <steve@altair.xemacs.org>
358 * Makefile.in: add semicolon in BISON definition for portability.
360 Thu Sep 3 13:34:41 1998 Toon Moene <toon@moene.indiv.nluug.nl>
362 * config/nextstep.c (handle_pragma): Correct name of third
365 Tue Sep 1 11:30:33 1998 Nick Clifton <nickc@cygnus.com>
367 * config/m32r/m32r.md: Change (reg:CC 17) to (reg:SI 17).
368 * config/m32r/m32r.h: Make register 17 be fixed.
369 * config/m32r/m32r.c: Use SImode for cc operations.
371 Thu Sep 3 18:17:34 1998 Benjamin Kosnik <bkoz@cygnus.com>
373 * invoke.texi (Warning Options): Add -Wnon-template-friend
376 Thu Sep 3 18:16:16 1998 Michael Meissner <meissner@cygnus.com>
378 * rs6000.c (rs6000_override_options): Add -mcpu={401,e603e}.
380 Thu Sep 3 18:05:16 1998 David Edelsohn <edelsohn@mhpcc.edu>
382 * rs6000.md (movsf): Disable explicit secondary-reload-like
383 functionality if TARGET_POWERPC64.
384 (movdf): Remove TARGET_POWERPC64 explicit secondary-reload-like
388 Thu Sep 3 11:41:40 1998 Robert Lipe <robertl@dgii.com>
390 * fixinc.sco: Borrow code to wrap 'bool' typedefs from tinfo.h
391 and term.h from fixinc.wrap.
393 Thu Sep 3 09:47:31 1998 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
395 * aclocal.m4 (GCC_HEADER_STRING): New macro to detect if it is
396 safe to include both string.h and strings.h together.
397 (GCC_NEED_DECLARATION): Test STRING_WITH_STRINGS when deciding
398 which headers to search for function declarations. Continue to
399 prefer string.h over strings.h when both are not acceptable.
401 * acconfig.h (STRING_WITH_STRINGS): Add stub.
403 * configure.in: Call GCC_HEADER_STRING.
405 * system.h: Test STRING_WITH_STRINGS when deciding which headers
406 to include. Continue to prefer string.h over strings.h when both
409 Wed Sep 2 23:56:29 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
411 * config/sparc/sparc.c (output_double_int): If V9 and MEDLOW, do
412 not assume top 32-bits of symbolic addresses are zero if
415 Thu Sep 3 00:23:21 1998 Richard Henderson <rth@cygnus.com>
417 * ginclude/va-alpha.h: Protect entire second portion of the
418 file against double inclusion.
420 Thu Sep 3 00:37:55 1998 Ovidiu Predescu <ovidiu@aracnet.com>
422 Added support for the Boehm's garbage collector.
423 * configure.in: Handle --enable-objc-gc.
424 * configure: Rebuilt.
425 * Makefile.in (CHECK_TARGETS): Add check-objc.
426 (check-objc): New rule.
427 * objc/Make-lang.in: Build a different Objective-C library that
428 runs with the Boehm's collector.
429 * objc/encoding.c (objc_round_acc_size_for_types): New function.
430 * objc/encoding.c: Correctly compute the size of compound types in
431 the presence of bitfields. Skip the variable name of the type if
432 any. Added support for long long.
433 * objc/encoding.h (_C_GCINVISIBLE): New specifier.
434 (_F_GCINVISIBLE): New mask.
435 * objc/gc.c: New file. Compute the type memory mask associated with
436 a class based on the runtime information.
437 * objc/misc.c: Added the hooks that use the Boehm's collector
438 allocation functions.
439 * objc/objc-act.c (build_class_template): Generate a new class
440 member (gc_object_type) to hold the class' type memory mask.
441 (build_shared_structure_initializer): Initialize the new member to
443 (encode_complete_bitfield): New function. Generate the new
445 (encode_field_decl): Generate the new encoding only for the GNU
447 * objc/objc-api.h (_C_LNG_LNG, _C_ULNG_LNG): New specifiers for the
449 (class_get_gc_object_type): New function to mark a pointer instance
450 variable as a weak pointer.
451 * objc/objc-features.texi: New file.
452 * objc/objc.h (gc_object_type): New class member.
453 * objc/objects.c (class_create_instance): Create a typed memory
454 object when compiled with Boehm's collector support.
455 * objc/sendmsg.c (__objc_init_install_dtable): Call
456 __objc_send_initialize instead of setting the initialize flag.
457 (__objc_send_initialize): Call __objc_generate_gc_type_description
458 to generate the class type memory mask. Rewrite the code that
459 sends the +initialize so that it is called only once (bug report
460 and fix from Ronald Pijnacker <Ronald.Pijnacker@best.ms.philips.com>).
461 * testsuite/objc: New testsuite for Objective-C type encoding.
462 * testsuite/lib/objc-torture.exp: New file.
463 * testsuite/lib/objc.exp: New file.
465 Wed Sep 2 14:47:36 1998 Jim Wilson <wilson@cygnus.com>
467 * jump.c (jump_optimize): In if/then/else transformations, add
468 another call to modified_between_p for the jump insn.
470 Wed Sep 2 14:16:49 1998 Jeffrey A Law (law@cygnus.com)
472 * fix-header.c (symlink): Treat like readlink.
474 Wed Sep 2 19:30:06 1998 J"orn Rennecke <amylaar@cygnus.co.uk>
476 * dwarfout.c (fundamental_type_code): Encode 32 bit floats/doubles
479 Wed Sep 2 10:06:07 1998 Nick Clifton <nickc@cygnus.com>
481 * config/nextstep.h: Update HANDLE_PRAGMA macro.
482 * config/h8300/h8300.h: Update HANDLE_PRAGMA macro.
483 * config/i960/i960.h: Update HANDLE_PRAGMA macro.
485 * config/nextstep.c (handle_pragma): Take three arguments, as per
486 the new HANDLE_PRAGMA macro specification.
487 * config/h8300/h8300.c (handle_pragma): Take three arguments, as
488 per the new HANDLE_PRAGMA macro specification.
489 * config/i960/i960.c (process_pragma): Take three arguments, as
490 per the new HANDLE_PRAGMA macro specification.
492 Wed Sep 2 09:25:29 1998 Nick Clifton <nickc@cygnus.com>
494 * c-lex.c (check_newline): Call HANDLE_PRAGMA before
495 HANDLE_SYSV_PRAGMA if both are defined. Generate warning messages
496 if unknown pragmas are encountered.
497 (handle_sysv_pragma): Interpret return code from
498 handle_pragma_token (). Return success/failure indication rather
499 than next unprocessed character.
500 (pragma_getc): New function: retrieves characters from the
501 input stream. Defined when HANDLE_PRAGMA is enabled.
502 (pragma_ungetc): New function: replaces characters back into the
503 input stream. Defined when HANDLE_PRAGMA is enabled.
505 * c-pragma.c (handle_pragma_token): Return success/failure status
508 * c-pragma.h: Change prototype of handle_pragma_token().
510 * varasm.c: (handle_pragma_weak): Only create this function if
511 HANDLE_PRAGMA_WEAK is defined.
513 * c-common,c (decl_attributes): If defined call the expression
514 contained within the INSERT_ATTRIBUTES macro before adding
515 attributes to a decl.
517 * tm.texi (HANDLE_PRAGMA): Document the new verion of
518 HANDLE_PRAGMA, which takes three arguments.
519 (INSERT_ATTRIBUTES): Document this new macro.
521 * LANGUAGES: Document the new version of HANDLE_PRAGMA and the
522 new INSERT_ATTRIBUTES macro.
524 Wed Sep 2 02:03:23 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
526 * config/sparc/sparc.md (movdf): Only generate special RTL for
528 (move_label_di): Remove
529 (movdi_pic_label_ref, movdi_high_pic_label_ref,
530 movdi_lo_sum_pic_label_ref): New patterns for 64-bit label
532 * config/sparc/sparc.h (ASM_OUTPUT_ADDR_VEC_ELT,
533 ASM_OUTPUT_ADDR_DIFF_ELT): Don't do anything special for MEDLOW,
534 output an .xword for all 64-bit cases.
536 Tue Sep 1 15:55:17 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
538 * config/sparc/sparc.c (finalize_pic): Don't output arbitrary
539 alignment, use FUNCTION_BOUNDARY instead.
540 (sparc_output_deferred_case_vectors): Likewise.
542 Mon Aug 31 17:25:41 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
544 * config/sparc/sparc.md (movsf_const_intreg): Kill warning.
545 (movtf_insn_sp64, movtf_no_e_insn_sp64): Reorder alternatives.
547 Mon Aug 31 13:57:55 1998 Richard Henderson <rth@cygnus.com>
549 * alpha/va_list.h: New file.
550 * alpha/x-alpha (EXTRA_HEADERS): New. Add va_list.h.
552 Mon Aug 31 14:55:02 1998 Jeffrey A Law (law@cygnus.com)
554 * NEWS: Add SCO Openserver and Unixware 7 notes.
558 Mon Aug 31 15:42:18 1998 Dave Brolley <brolley@cygnus.com>
560 * varasm.c (compare_constant_1): Handle RANGE_EXPR.
561 (record_constant_1): Handle RANGE_EXPR.
563 Mon Aug 31 10:54:03 1998 Richard Henderson <rth@cygnus.com>
565 * print-rtl.c (print_rtx): NOTE_INSN_LIVE has an rtx not a bitmap.
566 * haifa-sched.c (sched_analyze): Handle NOTE_INSN_RANGE_START
567 and NOTE_INSN_RANGE_END specially.
568 (reemit_notes): Likewise.
570 Mon Aug 31 10:18:52 1998 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
572 * sparc.c (TMASK, UMASK): Use `(unsigned)1' not `1U'.
573 (ultrasparc_sched_init): Remove unneeded &.
575 Mon Aug 31 10:47:16 1998 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
577 * config/m68k/m68k.h (TARGET_SWITCHES): Don't remove MASK_68040
578 for m68020-60, to prevent the use of fintrz.
580 Sun Aug 30 22:17:20 1998 Mark Mitchell <mark@markmitchell.com>
582 * configure.in: If the native compiler is GCC use $(WARN_CFLAGS)
584 * Makefile.in: Likewise.
585 * configure: Regenerated.
587 Sun Aug 30 22:15:41 1998 H.J. Lu (hjl@gnu.org)
589 * configure.in (gxx_include_dir): Changed to
590 '${prefix}/include/g++'-${libstdcxx_interface}.
591 * configure: Rebuilt.
593 Sun Aug 30 20:19:43 1998 Hans-Peter Nilsson <hp@axis.se>
595 * expr.c (expand_expr): Change ">" to ">=" making MOVE_RATIO use
597 * tm.texi (Costs): Say MOVE_RATIO is number of mem-mem move
598 *sequences* *below* which scalar moves will be used.
600 Sun Aug 30 17:18:43 1998 Jeffrey A Law (law@cygnus.com)
602 * collect2.c (mktemp): Delete unused declaration.
604 * config/xm-netbsd.h: Remove unnecessary file.
605 * config/*/xm-netbsd.h: Do not include the generic xm-netbsd.h
606 file anymore, it is not needed.
608 Sun Aug 30 16:05:45 1998 Mark Mitchell <mark@markmitchell.com>
610 * convert.c (convert_to_integer): Issue an error on conversions to
613 Sun Aug 30 16:47:20 1998 Martin von Lvwis <loewis@informatik.hu-berlin.de>
615 * Makefile.in: Add lang_tree_files and gencheck.h.
616 * configure.in: Generate them.
617 * gencheck.c: Include gencheck.h.
619 Sat Aug 29 21:38:24 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
621 * config/sparc/sparc.md (pic_lo_sum_di, pic_sethi_di): Rename to
622 movdi_lo_sum_pic and movdi_high_pic and make visible.
623 * config/sparc/sparc.c (legitimize_pic_address): For -fPIC,
624 emit these when Pmode is not SImode.
625 * config/sparc/linux64.h (SPARC_DEFAULT_CMODEL): Make CM_MEDLOW.
627 Sat Aug 29 14:59:32 1998 Mumit Khan <khan@xraylith.wisc.edu>
629 * i386/cygwin32.h (ASM_OUTPUT_SECTION_NAME): Don't emit
630 .linkonce directive after the first time.
632 Sat Aug 29 12:39:56 1998 Jeffrey A Law (law@cygnus.com)
634 * m68k.md (beq0_di): Generate correct (and more efficient) code when
635 the clobbered operand overlaps with an input.
636 (bne0_di): Similarly.
638 * Makefile.in (INSTALL): Remove "--no-header" argument.
640 * NEWS: Various updates.
642 Fri Aug 28 19:00:44 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
644 * config/sparc/sparc.c (arith_operand, const64_operand,
645 const64_high_operand, arith_double_4096_operand): Mark mode as
647 (create_simple_focus_bits): Remove unused arg highest_bit_set, all
649 (sparc_emit_set_const64): Remove unused variable i.
650 (sparc_splitdi_legitimate): Likewise for addr_part.
651 (ultra_code_from_mask): Likewise for mask.
652 (ultra_cmove_results_ready_p): Fixup entry modulo calc. and
653 reverse return values so it matches usage and comments.
654 (ultra_flush_pipeline): Likewise.
655 (ultra_fpmode_conflict_exists): Likewise, remove unused variable
656 this_type, and allow loads and stores of differing FP modes as
657 they do not create a conflict.
658 (ultra_find_type): Initialize fpmode to SFmode, fix
659 parenthesization thinkos in large conditional.
660 (ultrasparc_sched_init): Mark dump and sched_verbose as unused.
661 Init free_slot_mask after ultra_cur_hist is reset, not before.
662 (ultrasparc_rescan_pipeline_state): Remove unused variable ucode.
663 (ultrasparc_sched_reorder): Don't bzero current pipeline state,
664 use ultra_flush_pipeline instead, then re-init group pointer.
665 Fix statement with no effect. If no progress made in, and no
666 instructions scheduled at all, advance to new pipeline cycle else
667 we get into an endless loop.
668 (ultrasparc_adjust_cost): Remove previous arg.
669 * config/sparc/sparc.h (ADJUST_COST): Update to reflect that.
671 Fri Aug 28 13:52:35 1998 Jim Wilson <wilson@cygnus.com>
673 * sparc.md (DImode, DFmode, TFmode splits): Delete self_reference
674 code. Use reg_overlap_mentioned_p to detect when source and
676 (negtf2_notv9+1): Use DFmode instead of SFmode in last two operands.
678 1998-08-28 Brendan Kehoe <brendan@cygnus.com>
680 * loop.c (check_dbra_loop): Pass COMPARISON_VALUE, not
681 COMPARISON_VAL, into invariant_p.
683 Fri Aug 28 15:13:25 1998 J"orn Rennecke <amylaar@cygnus.co.uk>
685 * regmove.c (regclass_compatible_p): New function.
686 (regmove_optimize): Use it.
688 Use NREGS parameter insted of calling max_reg_num.
690 (fixup_match_1): Don't use code = MINUS when later tieing with
691 a hard register is likely.
693 Fri Aug 28 14:54:07 1998 J"orn Rennecke <amylaar@cygnus.co.uk>
695 * loop.c (check_dbra_loop): Fix calculation of FINAL_VALUE when
696 COMPARISON_VAL was normalized.
698 Thu Aug 27 20:10:46 1998 Jeffrey A Law (law@cygnus.com)
700 * loop.c (check_dbra_loop): The loop ending comparison value
701 must be an invariant or we can not reverse the loop.
703 * loop.c (scan_loop): Count down from max_reg_num - 1 to
704 FIRST_PSEUDO_REGISTER to avoid calling max_reg_num each iteration
706 (load_mems_and_recount_loop_regs_set): Likewise.
708 * i386.c (print_operand): Remove obsolete 'c' docs.
710 Wed Aug 26 17:13:37 1998 Tom Tromey <tromey@cygnus.com>
712 * gthr.h: Document __GTHREAD_MUTEX_INIT_FUNCTION.
713 * frame.c (init_object_mutex): New function.
714 (init_object_mutex_once): Likewise.
716 (__register_frame_info): Likewise.
717 (__register_frame_info_table): Likewise.
718 (__deregister_frame_info): Likewise.
720 Thu Aug 27 15:14:18 1998 Jeffrey A Law (law@cygnus.com)
722 * haifa-sched.c (sched_analyze_insn): Fix thinko in last change.
724 Thu Aug 27 16:34:51 1998 J"orn Rennecke <amylaar@cygnus.co.uk>
726 * loop.c (check_dbra_loop): Enable code for reversal
727 of some loops without a known constant loop end.
729 Wed Aug 26 18:38:15 1998 Richard Henderson <rth@cygnus.com>
731 * haifa-sched.c (last_clock_var): New.
732 (schedule_block): Initialize it.
733 (schedule_insn): Use it to fill insn modes with issue information.
735 * alpha.c (alpha_handle_trap_shadows): Remove do-nothing exit.
736 Tag trapb and next insn with TImode.
737 (alphaev5_insn_pipe, alphaev5_next_group, alphaev5_align_insns): New.
738 (alpha_reorg): Add conditional for alpha_handle_trap_shadows.
739 Invoke alphaev5_align_insns as appropriate.
740 * alpha.h (LABEL_ALIGN_AFTER_BARRIER): Was ALIGN_LABEL_AFTER_BARRIER.
741 (MD_SCHED_VARIABLE_ISSUE): New.
742 * alpha.md (attr type): Add multi.
743 (define_asm_attributes): New.
744 (prologue_stack_probe_loop, builtin_setjmp_receiver): Set type multi.
745 (arg_home): Likewise.
746 (fnop, unop, realign): New.
748 Wed Aug 26 15:55:41 1998 Jim Wilson <wilson@cygnus.com>
750 * iris5.h (PREFERRED_DEBUGGING_TYPE): Undef.
751 * iris5gas.h (PREFERRED_DEBUGGING_TYPE): Define.
753 * configure.in (powerpc-ibm-aix4.[12]*): Change from 4.[12].*.
754 (rs6000-ibm-aix4.[12]*): Likewise.
755 * configure: Regnerate.
757 Wed Aug 26 09:30:59 1998 Nick Clifton <nickc@cygnus.com>
759 * config/arm/thumb.c (thumb_exit): Do not move a4 into lr if it
760 already contains the return address.
762 Wed Aug 26 12:57:09 1998 Jeffrey A Law (law@cygnus.com)
764 * calls.c (expand_call): Use bitfield instructions to extract/deposit
765 word sized hunks when loading unaligned args into registers.
767 * haifa-sched.c (sched_analyze_insn): Only create scheduling
768 barriers for LOOP, EH and SETJMP notes on the loop_notes list.
770 * mn10300.h (RTX_COSTS): Handle UDIV and UMOD too.
772 Wed Aug 26 16:35:37 1998 J"orn Rennecke <amylaar@cygnus.co.uk>
774 * loop.c (check_dbra_loop): Add some code that would allow reversal
775 of some loops without a known constant loop end if it were enabled.
777 Wed Aug 26 11:08:44 1998 Gavin Romig-Koch <gavin@cygnus.com>
779 * mips.md (lshrsi3_internal2+2): Fix type-o.
781 Wed Aug 26 10:53:03 1998 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
783 * system.h: Include stdarg.h/varargs.h, make sure they are ordered
784 correctly with regards to stdio.h.
786 * calls.c: Remove stdarg.h/varargs.h.
789 * combine.c: Likewise.
790 * cpperror.c: Likewise.
791 * cpplib.c: Likewise.
792 * cpplib.h: Likewise.
793 * doprint.c: Likewise.
794 * emit-rtl.c: Likewise.
796 * fix-header.c: Likewise.
798 * genattr.c: Likewise.
799 * genattrtab.c: Likewise.
800 * gencodes.c: Likewise.
801 * genconfig.c: Likewise.
802 * genemit.c: Likewise.
803 * genextract.c: Likewise.
804 * genflags.c: Likewise.
805 * genopinit.c: Likewise.
806 * genoutput.c: Likewise.
807 * genpeep.c: Likewise.
808 * genrecog.c: Likewise.
809 * mips-tfile.c: Likewise.
810 * prefix.c: Likewise.
811 * protoize.c: Likewise.
812 * regmove.c: Likewise.
813 * toplev.c: Likewise.
816 Wed Aug 26 05:09:27 1998 Jakub Jelinek <jj@sunsite.ms.mff.cuni.cz>
818 * config/sparc/sparc.c (sparc_override_options): If not
819 TARGET_FPU, turn off TARGET_VIS.
820 * config/sparc/sparc.h (TARGET_SWITCHES): Add no-vis.
821 (LEGITIMATE_CONSTANT_P): Allow SF/DF mode zero when TARGET_VIS.
822 * config/sparc/sparc.md (movsi_insn): Use fzeros not fzero.
823 (movdi_insn_sp64): Add VIS fzero alternative.
824 (clear_sf, clear_df): New VIS patterns.
825 (movsf, movdf expanders): Allow fp_zero_operand flat out when
827 (one_cmpldi2_sp64): Provide new fnot1 VIS alternative.
829 Tue Aug 25 10:57:41 1998 Mark Mitchell <mark@markmitchell.com>
831 * loop.c (n_times_set, n_times_used, may_not_optimize,
832 reg_single_usage): Convert to varrays. All uses changed.
833 (insert_loop_mem): Return a value.
834 (scan_loop): Tweak AVOID_CC_MODE_COPIES code.
835 (load_mems_and_recount_loop_regs_set): Likewise. Grow the arrays, if
838 Tue Aug 25 23:57:12 1998 Jeffrey A Law (law@cygnus.com)
841 * configure.in: Do not set thread_file to "irix" since no such
844 * sparc.md (float abs/neg splits): Check reload_completed before
845 calling alter_subreg.
847 Tue Aug 25 19:17:59 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
849 * config/sparc/sparc.c (sparc_absnegfloat_split_legitimate): New
851 * config/sparc/sparc.h: Declare it.
852 * config/sparc/sparc.md (float abs/neg splits): Use it.
853 (all other splits): Handle SUBREGs properly where necessary.
854 (unnamed (1<<x)-1 V8PLUS pattern): Disable for now.
856 Tue Aug 25 19:48:46 1998 Jeffrey A Law (law@cygnus.com)
858 * reorg.c (fill_simple_delay_slots): Do not abort if we encounter
859 an insn on the unfilled_slots_list that has no delay slots.
860 (fill_eager_delay_slots): Similarly.
862 Tue Aug 25 13:35:20 1998 Nick Clifton <nickc@cygnus.com>
864 * config/v850/v850.c (movsi_source_operand): Treat CONSTANT_P_RTX
865 as an ordinary operand.
867 Tue Aug 25 12:54:57 1998 Jason Merrill <jason@yorick.cygnus.com>
869 * tree.c (valid_machine_attribute): Don't apply attributes to both
872 Tue Aug 25 12:23:20 PDT 1998 Richard Henderson <rth@cygnus.com>
874 * reload.c (operands_match_p): Handle rtvecs.
876 * i386.c (legitimate_pic_address_disp_p): New.
877 (legitimate_address_p): Use it.
878 (legitimize_pic_address): Use unspecs to represent @GOT and @GOTOFF.
879 Handle constant pool symbols just like statics.
880 (emit_pic_move): Use Pmode not SImode for clarity.
881 (output_pic_addr_const) [SYMBOL_REF]: Remove @GOT and @GOTOFF hacks.
882 [UNSPEC]: New, handling what we killed above.
883 [PLUS]: Detect and abort on invalid symbol arithmetic.
884 * i386.h (CONSTANT_ADDRESS_P): Remove HIGH.
886 Tue Aug 25 12:02:23 1998 Mark Mitchell <mark@markmitchell.com>
888 * alias.c: Include output.h.
889 (DIFFERENT_ALIAS_SETS_P): Don't treat alias sets as
890 different if we're in a varargs function.
891 * Makefile.in (alias.o): Depend on output.h
893 Tue Aug 25 19:20:12 1998 J"orn Rennecke <amylaar@cygnus.co.uk>
895 * sh.h (GIV_SORT_CRITERION): Delete.
897 Tue Aug 25 13:19:46 1998 Dave Brolley <brolley@cygnus.com>
899 * regclass.c (regclass): Use xmalloc/free instead of alloca.
900 * stupid.c (stupid_life_analysis): Ditto.
901 * reload1.c (reload): Ditto.
903 Tue Aug 25 05:48:18 1998 Jakub Jelinek <jj@sunsite.ms.mff.cuni.cz>
905 * config/sparc/sparc.c (arith_4096_operand, arith_add_operand,
906 arith_double_4096_operand, arith_double_add_operand): New
908 * config/sparc/sparc.h (PREDICATE_CODES): Add them, declare them.
909 * config/sparc/sparc.md (adddi3, addsi3, subdi3, subsi3): Use
910 them to transform add/sub 4096 into add/sub -4096.
912 Mon Aug 24 23:31:03 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
914 * loop.c (scan_loop): Allocate some slop to handle pseudos
915 generated by move_movables.
916 (load_mems_and_recount_loop_regs_set): Honor AVOID_CC_MODE_COPIES
919 Mon Aug 24 19:45:40 1998 Jim Wilson <wilson@cygnus.com>
921 * tree.def (DECL_RESULT): Correct documentation.
923 Tue Aug 25 01:15:27 1998 J"orn Rennecke <amylaar@cygnus.co.uk>
925 * reload1.c (reload_reg_free_before_p): New argument EQUIV; Changed
926 all callers. Abort for RELOAD_FOR_INSN. RELOAD_FOR_OUTADDR_ADDR:
927 conflicts will all RELOAD_FOR_OUTPUT reloads.
929 * reload1.c (reload_cse_regs_1): When deleting a no-op move that
930 loads the function result, substitute with a USE.
932 Mon Aug 24 15:20:19 1998 David Edelsohn <edelsohn@mhpcc.edu>
934 * rs6000.h (GO_IF_LEGITIMATE_ADDRESS): Use TARGET_POWERPC64
935 when testing LEGITIMATE_INDEXED_ADDRESS_P DFmode and DImode.
936 (LEGITIMIZE_ADDRESS): Use TARGET_POWERPC64 for INDEXED fixup.
937 * rs6000.c (print_operand, case 'L'): Add UNITS_PER_WORD, not 4.
938 (print_operand, cases 'O' and 'T'): Fix typos in lossage strings.
939 * rs6000.md (fix_truncdfsi2_store): Remove %w from non-CONST_INT
941 (movdf_softfloat32, movdf_hardfloat64, movdf_softfloat64): Change
942 'o' to 'm' for GPR variant constraints.
944 Mon Aug 24 10:25:46 1998 Jeffrey A Law (law@cygnus.com)
946 * loop.c (scan_loop): Honor AVOID_CC_MODE_COPIES.
948 * h8300.h (STRIP_NAME_ENCODING): Fix typo.
950 * sparc.md (TFmode splits): Use reg_overlap_mentioned_p to detect
951 when the source and destination overlap.
953 * stmt.c (emit_case_nodes): Change rtx_function to rtx_fn to avoid
954 clash with global type.
956 Mon Aug 24 00:53:53 1998 Jason Merrill <jason@yorick.cygnus.com>
958 * fixinc.irix: Add curses.h handling from fixinc.wrap.
960 * c-common.c (combine_strings): Also set TREE_READONLY.
961 Change warn_write_strings to flag_const_strings.
962 * c-decl.c, c-tree.h: Likewise.
964 Sun Aug 23 18:39:11 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
966 * config/sparc/sparc.c (sparc_emit_set_const32): If outputting a
967 CONST_INT, not a symbolic reference, don't use a HIGH/LO_SUM
968 sequence, use SET/IOR instead so CSE can see it.
969 * config/sparc/sparc.md (movhi_const64_special,
970 movsi_const64_special): New patterns necessitated by that change.
971 (movhi_high): Remove.
972 (movhi_lo_sum): Change to match an IOR.
973 (movdf_insn_sp32): Test TARGET_V9 not TARGET_ARCH64.
974 (movdf_insn_v9only): New pattern for when V9 but not ARCH64.
975 (movdf_insn_sp64): Test both TARGET_V9 and TARGET_ARCH64.
976 (movdf splits): Allow when not V9 or when not ARCH64 and integer
977 registers are involved.
978 (snesi_zero_extend split): Remove reload_completed test.
979 (unnamed plus and minus zero_extend sidi splits): Add it.
981 Sun Aug 23 11:56:08 1998 Mark Mitchell <mark@markmitchell.com>
983 * extend.texi: Remove description of extension to explicit
984 instantiation that is now endorsed by standard C++.
986 Sun Aug 23 09:39:09 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
988 * config/arc/arc.c (arc_initialize_pic): Remove.
989 * config/arc/arc.h (INITIALIZE_PIC): Similarly, this routine does
990 nothing on any platform and is invoked by no-one, it does not even
991 appear in the documentation.
992 * config/sparc/sparc.h (INITIALIZE_PIC): Likewise.
993 * config/sparc/sparc.c (initialize_pic): Likewise.
994 (find_addr_reg): Remove this as well, no longer referenced after
997 Sun Aug 23 00:17:14 1998 Jeffrey A Law (law@cygnus.com)
999 * recog.c (validate_replace_rtx_group): New function.
1000 * recog.h (validate_replace_rtx_group): Declare it.
1001 * regmove.c (optimize_reg_copy_3): If any substitution fails, then undo
1002 the entire group of substitutions.
1004 Sat Aug 22 23:31:00 1998 Klaus-Georg Adams (Klaus-Georg.Adams@chemie.uni-karlsruhe.de)
1006 * loop.c (load_mems): Fix initializers.
1008 Fri Aug 21 23:07:46 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1010 * config/sparc/sparc.md (TFmode splits): Handle destination
1011 registers being referenced in the address correctly.
1013 * expmed.c (make_tree) [CONST_INT]: Sign extend even if
1014 TREE_UNSIGNED, when bitsize of type's mode is larger than
1015 HOST_BITS_PER_WIDE_INT.
1017 Fri Aug 21 19:31:31 1998 Alexandre Petit-Bianco <apbianco@cygnus.com>
1019 * tree.def (LABELED_BLOCK_EXPR, EXIT_BLOCK_EXPR): New tree nodes.
1020 * tree.h (LABELED_BLOCK_LABEL, LABELED_BLOCK_BODY,
1021 EXIT_BLOCK_LABELED_BLOCK, EXIT_BLOCK_RETURN, LOOP_EXPR_BODY): New
1023 * expr.c (expand_expr): Handle LABELED_BLOCK_EXPR and
1026 Thu Aug 20 19:43:44 1998 Jeffrey A Law (law@cygnus.com)
1028 * h8300.c (h8300_encode_label): Use '&' for tiny data items.
1029 * h8300.h (TINY_DATA_NAME_P): Likewise.
1030 (STRIP_NAME_ENCODING): Handle '&'.
1032 * mn10200.h (REG_OK_FOR_INDEX_P): Do not check the mode of the
1033 register (it could be accessed via an outer SUBREG).
1034 (REG_OK_FOR_BASE_P): Likewwise.
1035 (GO_IF_LEGITIMATE_ADDRESS): Consistently use REGNO_OK_FOR_BASE_P.
1037 * remove.c (optimize_reg_copy_3): Abort instead of silently generating
1040 * jump.c (rtx_renumbered_equal_p): Do not consider PLUS commutative.
1042 Thu Aug 20 17:35:20 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1044 * config/sparc/sparc.md (movtf_insn_sp32): All memory operands
1045 must be offsettable so the splits can be made.
1047 Thu Aug 20 13:56:53 1998 Michael Meissner <meissner@cygnus.com>
1049 * config/i386/winnt.c: Include system.h, not stdio.h to get
1050 sys/param.h pulled in before rtl.h in case the system defines MIN
1053 Thu Aug 20 13:44:20 1998 David Edelsohn <edelsohn@mhpcc.edu>
1055 * rs6000.md (movqi, movhi): Add CONSTANT_P_RTX.
1057 Thu Aug 20 13:15:11 1998 Dave Brolley <brolley@cygnus.com>
1059 * stor-layout.c (layout_type): Compute TYPE_SIZE_UNIT correctly for
1061 * cpplib.c (cpp_define): Handle macros with parameters.
1063 Wed Aug 19 21:33:19 1998 David Edelsohn <edelsohn@mhpcc.edu>
1065 * rs6000.c (rs6000_output_load_toc_table): Use ld for 64-bit.
1066 (output_toc): Use single TOC slot or llong minimal-toc for DFmode
1067 and DImode 64-bit. Use llong for minimal-toc SFmode and
1068 SYMBOL_REF / LABEL_REF 64-bit.
1069 (output_function_profiler): Use llong for profiler label and ld to
1070 load 64-bit label address.
1072 Wed Aug 19 17:52:27 1998 Nick Clifton (nickc@cygnus.com)
1074 * config/arm/thumb.md (extendqisi2_insn): Cope with REG +
1077 Wed Aug 19 14:13:31 PDT 1998 Jeff Law (law@cygnus.com)
1079 * version.c: Bump for snapshot.
1081 Wed Aug 19 13:10:30 PDT 1998 Jeff Law (law@cygnus.com)
1083 * version.c: Bump for snapshot.
1085 Wed Aug 19 13:06:47 1998 Jason Merrill <jason@yorick.cygnus.com>
1087 * collect2.c (extract_init_priority): Use atoi instead of strtoul.
1089 Wed Aug 19 13:51:35 1998 Hans-Peter Nilsson <hp@axis.se>
1091 * tm.texi (Misc): Fix typo "teh".
1093 * tm.texi (PIC): Fix typo "PPIC".
1095 * tm.texi (Caller Saves): Say that DEFAULT_CALLER_SAVES has no
1096 effect when -O2 and higher.
1097 * invoke.texi (Optimize Options): Likewise for -fcaller-saves.
1099 1998-08-19 Michael Hayes <michaelh@ongaonga.chch.cri.nz>
1101 * regclass.c: Changed register set documentation to be consistent
1104 * final.c (final_start_function) Removed redundant test for
1107 Wed Aug 19 13:28:41 1998 Mark Mitchell <mark@markmitchell.com>
1109 * rtl.h (rtx_function): New type.
1110 (for_each_rtx): New function.
1111 * rtlanal.c (for_each_rtx): Define it.
1113 * recog.c (change_t): New type.
1114 (change_objects, change_old_codes, change_locs, change_olds):
1116 (changes): New variable.
1117 (validate_change): Dynamically allocate room for more changes, if
1118 necessary. Uses changes array instead of change_objects, etc.
1119 (apply_change_group): Use changes array instead of
1120 change_objects, etc.
1122 * loop.c (loop_mem_info): New type.
1123 (loop_mems): New variable.
1124 (loop_mems_idx): Likewise.
1125 (looop_mems_allocated): Likewise.
1126 (scan_loop): Remove nregs parameter.
1127 (next_insn_in_loop): New function.
1128 (load_mems_and_recount_loop_regs_set): Likewise.
1129 (load_mems): Likewise.
1130 (insert_loop_mem): Likewise.
1131 (replace_loop_mem): Likewise.
1132 (replace_label): Likewise.
1133 (INSN_IN_RANGE_P): New macro.
1134 (loop_optimize): Don't pass max_reg_num() to scan_loop.
1135 (scan_loop): Remove nregs parameter, compute it after any new
1136 registers are created by load_mems. Use INSN_IN_RANGE_P and
1137 next_insn_in_loop rather than expanding them inline. Call
1138 load_mems to load memory into pseudos, if appropriate.
1139 (prescan_loop): Figure out whether or not there are jumps from the
1140 loop to targets other than the label immediately following the
1141 loop. Call insert_loop_mem to notice all the MEMs used in the
1142 loop, if it could be safe to pull MEMs into REGs for the duration
1144 (strength_reduce): Use next_insn_in_loop. Tweak comments.
1146 Wed Aug 19 08:29:44 1998 Richard Earnshaw (rearnsha@arm.com)
1148 * arm.c (arm_override_options): Remove lie about ignoring PIC flag.
1150 Wed Aug 19 07:08:15 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1152 * config/sparc/sparc.c (finalize_pic): Check for the correct
1153 nonlocal_goto_receiver UNSPEC number.
1154 * config/sparc/sparc.md (nonlocal_goto_receiver): Add comment
1155 making note of this dependency existing in sparc.c
1156 (negtf2_notv9 split): Give NEG SFmode.
1157 (negsf2): Fix insn output string.
1159 Tue Aug 18 12:40:27 1998 Richard Henderson <rth@cygnus.com>
1161 * c-common.c (decl_attributes): Issue an error if the argument
1162 to alias is not a string.
1164 Tue Aug 18 10:33:30 1998 Jeffrey A Law (law@cygnus.com)
1166 * haifa-sched.c (sched_analyze): Put all JUMP_INSNs on the last
1167 pending memory flush list.
1169 * combine.c (can_combine_p): Allow combining insns with REG_RETVAL
1171 (try_combine): Allow combining insns with REG_LIBCALL notes.
1173 * expr.c (emit_block_move): Do not call memcpy as a libcall
1174 instead build up a CALL_EXPR and call it like any other
1176 (clear_storage): Similarly for memset.
1178 * regmove.c (fixup_match_2): Do not call reg_overlap_mentioned_p
1181 * Makefile.in (cplus-dem.o): Provide explicit rules for building
1184 * regmove.c (optimize_reg_copy_1): Update REG_N_CALLS_CROSSED
1185 and REG_LIVE_LENGTH as successful substitutions are made.
1187 Tue Aug 18 07:15:27 1998 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
1189 * config/sparc/sparc.c (ultra_find_type): Add empty semicolon
1190 statement after end of loop label.
1192 Tue Aug 18 07:13:27 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1194 * config/sparc/sparc.c (ultra_types_avail): New variable.
1195 (ultra_build_types_avail): New function to record mask of insn
1196 types in ready list at this cycle.
1197 (ultrasparc_sched_reorder): Call it.
1198 (ultra_find_type): Use it to quicken the search. Also simplif
1199 dependency check, don't use rtx_equal_p because we know exactly
1200 what we are looking for.
1202 Tue Aug 18 03:20:53 1998 Richard Earnshaw (rearnsha@arm.com)
1204 * arm.h (SECONDARY_INPUT_RELOAD_CLASS): Return NO_REGS if compiling
1205 for architecture v4.
1207 Mon Aug 17 21:26:38 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1209 * config/sparc/sparc.md (sltu, sgeu): Don't FAIL, call
1211 (movsf_const_intreg, movsf_const_high, movsf_const_lo,
1212 movdf_const_intreg and helper splits): New patterns to move float
1213 constants into integer registers.
1214 (negtf2, negdf2, abstf2, absdf2): Rework using new patterns and
1217 Mon Aug 17 11:46:19 1998 Jeffrey A Law (law@cygnus.com)
1220 * tree.c (build_index_type): Copy TYPE_SIZE_UNIT from sizetype
1222 * c-decl.c (finish_enum): Copy TYPE_SIZ_UNIT from enumtype to tem.
1224 * rs6000.c (secondary_reload_class): For TARGET_ELF, indicate that
1225 a BASE_REGS register is needed as an intermediate when copying
1226 a symbolic value into any register class other than BASE_REGS.
1228 * expr.c (move_by_pieces): No longer static. Remove prototype.
1229 * rtl.h (move_by_pieces): Add extern prototype.
1230 * mips.c (expand_block_move): Handle aligned straight line copy by
1231 calling move_by_pieces.
1233 * expr.c (expand_expr): Allow assignments from TImode PARM_DECLs
1236 Mon Aug 17 10:28:52 1998 Mark Mitchell <mark@markmitchell.com>
1238 * stmt.c (expand_end_loop): Tidy. Allow unconditional
1239 jumps out of the loop to be treated as part of the exit test.
1241 Mon Aug 17 10:06:11 1998 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
1242 Jeff Law <law@cygnus.com>
1244 * Makefile.in (cplus-dep.o): Use cplus-dem.c from libiberty.
1245 * cplus-dem.c: Delete.
1247 * Makefile.in (fold-const.o): depend on $(RTL_H).
1249 * fold-const.c: Include rtl.h to get the prototype for
1250 `set_identifier_local_value'.
1252 * loop.c (express_from_1): Remove unused variable `tmp'.
1253 (combine_givs): cast the first argument of bzero to char *.
1255 * toplev.c (display_help): Remove unused variable `looking_for_start'.
1257 * c-decl.c (init_decl_processing): Remove unneeded &.
1259 * alpha.h (alpha_initialize_trampoline): Provide prototype.
1261 * except.c (set_exception_lang_code, set_exception_version_code):
1262 Change parameter from `short' to `int' to avoid using a gcc
1265 * except.h (set_exception_lang_code, set_exception_version_code):
1266 Likewise for prototypes.
1268 * flow.c (count_reg_references): Remove unused variables `regno'
1271 * gcse.c (hash_scan_insn): Declare parameter `in_libcall_block'.
1273 * prefix.c (translate_name): Cast the result of `alloca'.
1275 * varray.h (VARRAY_FREE): Reimplement as a `do-while(0)' statement.
1277 Mon Aug 17 09:23:42 1998 Andreas Schwab <schwab@issan.informatik.uni-dortmund.de>
1279 * config/m68k/m68k.c: Include "system.h" instead of <stdio.h>.
1281 (valid_dbcc_comparison_p): Mark mode argument as unused.
1282 (symbolic_operand): Likewise.
1283 (legitimize_pic_address): Likewise.
1284 (const_uint32_operand): Likewise.
1285 (const_sint32_operand): Likewise.
1286 * sched.c [!INSN_SCHEDULING]: Define only dummy function
1287 schedule_insns and comment out rest of file.
1289 * m68k.c (output_move_simode_const): Use subl to move a zero into an
1291 (output_move_[hq]imode): Likewise.
1293 Mon Aug 17 09:15:47 1998 Jeffrey A Law (law@cygnus.com)
1295 * toplev.c (main): Enable -fstrict-aliasing for -O2 and above.
1296 * invoke.texi: Corresponding changes.
1298 Mon Aug 17 02:03:55 1998 Richard Henderson <rth@cygnus.com>
1300 * regclass.c (allocate_reg_info): Respect MIN when clearing data.
1302 Sun Aug 16 17:37:06 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1304 * config/sparc/sparc.c (ultra_code_from_mask,
1305 ultra_cmove_results_ready_p, ultra_fpmode_conflict_exists,
1306 ultra_find_type, ultra_schedule_insn, ultra_flush_pipeline,
1307 ultrasparc_sched_init, ultrasparc_variable_issue,
1308 ultra_rescan_pipeline_state, ultrasparc_sched_reorder): New
1309 functions to describe UltraSPARC pipeline exactly to Haifa.
1310 (ultrasparc_adjust_cost): Indicate IMUL type insns have zero cost,
1311 as there is nothing the scheduler can do about it. Indicate that
1312 REG_DEP_OUTPUT's collide. Fixup formatting.
1313 * config/sparc/sparc.h (RTX_COSTS): Fixup integer multiply and
1314 divide costs on Ultra for DImode.
1315 (MD_SCHED_INIT, MD_SCHED_REORDER, MD_SCHED_VARIABLE_ISSUE):
1317 * config/sparc/sparc.md (ieu_unnamed function unit): Rename to
1318 ieuN and add call_no_delay_slot to type list.
1319 (cti function unit): New unit for branches on UltraSPARC.
1320 (subx/addx insns): Set type to misc.
1321 (sidi zero/sign extension insns on arch64): Set type to shift.
1322 (sign_extendhidi2_insn): Set type to sload.
1324 Sun Aug 16 13:52:00 1998 David Edelsohn <edelsohn@mhpcc.edu>
1326 * rs6000.c (rs6000_stack_info): Use if == 0 for sizes.
1327 (output_epilog): Use if != 0 for offset.
1328 (rs6000_fatal_bad_address): Prepare for Intl.
1329 * rs6000.h (rs6000_fatal_bad_address): Declare.
1330 * rs6000.md (movsfcc, movdfcc): Use else if.
1331 (elf_high): Use {liu|lis}.
1332 (elf_low): Use {cal|la}. Remove %a template from old mnemonics.
1333 (movsi): Use rs6000_fatal_bad_address.
1335 Sun Aug 16 01:53:21 1998 Richard Henderson <rth@cygnus.com>
1337 * reload.c (find_equiv_reg): Reject equivalences separated
1338 by a volatile instruction.
1340 Sun Aug 16 00:21:44 1998 Franz Sirl <Franz.Sirl-kernel@lauterbach.com>
1342 * rs6000/linux.h (CPP_OS_DEFAULT_SPEC): Define.
1344 Sat Aug 15 20:51:35 1998 Richard Henderson <rth@cygnus.com>
1346 * alpha.md (movsicc): Fix mode mismatch.
1348 Sat Aug 15 20:22:33 1998 H.J. Lu (hjl@gnu.org)
1350 * config/alpha/alpha.h (ASM_OUTPUT_MI_THUNK): Handle aggregated
1352 * config/alpha/win-nt.h (ASM_OUTPUT_MI_THUNK): Likewise.
1354 Sat Aug 15 08:39:49 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1356 * config/sparc/sparc.md (movsi_lo_sum_pic_label_reg): Remove
1357 write-only modifier from operand 1 constraint.
1359 Sat Aug 15 06:28:19 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1361 * config/sparc/sparc.c (sparc_emit_set_const64_quick1): If
1362 emitting a XOR of -1 at the end, emit a NOT instead for combine's
1364 (sparc_emit_set_const64): Likewise, also when computing trailing
1365 bits do not negate low_bits and make fast_int an int.
1367 Fri Aug 14 21:07:03 1998 Jeffrey A Law (law@cygnus.com)
1369 * loop.c (add_label_notes): Do not ignore references to labels
1370 before dispatch tables. Mirrors Apr 8 change to mark_jump_label.
1371 * gcse.c (add_label_notes): Similarly.
1373 * pa.h (ASM_OUTPUT_MI_THUNK): Strip name encoding.
1375 * m68k.md (adddi_dilshr32): One of the operands must be a register.
1376 (adddi_dishl32): Similarly.
1378 Fri Aug 14 14:12:59 1998 Jason Merrill <jason@yorick.cygnus.com>
1380 * i386.h (MODES_TIEABLE_P): Reorganize to shut up warnings.
1381 * alias.c (memrefs_conflict_p): Add braces to shut up warnings.
1382 * cse.c (cse_basic_block): Add parens to shut up warnings.
1384 Fri Aug 14 12:58:21 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1386 * config/sparc/sparc.c (sparc_emit_set_const64_quick2,
1387 sparc_emit_set_const64_longway, const64_is_2insns,
1388 create_simple_focus_bits, sparc_emit_set_const64): Fix more bugs
1389 in 64-bit constant formation.
1390 * config/sparc/sparc.md (snesi_zero_extend split): Generate
1391 rtl for addx not subx.
1392 (define_insn movdi_const64_special): Make available even when
1393 HOST_BITS_PER_WIDE_INT is not 64.
1394 (movdi_lo_sum_sp64_cint, movdi_high_sp64_cint): Remove.
1395 (losum_di_medlow, sethm, setlo): Make op2 symbolic_operand.
1396 (cmp_siqi_trunc_set, cmp_diqi_trunc_set): Encapsulate both
1397 instances of operand 1 inside a QI subreg.
1398 (xordi3_sp64_dbl): Remove '%' constraint for op1.
1399 (one_cmpldi2_sp64): Fix output string.
1400 (one_cmplsi2_not_liveg0): Rewrite to remove unneeded extra
1402 (unnamed arch64 ashift DI): Truncate shift count if greater than
1405 Fri Aug 14 21:52:53 1998 J"orn Rennecke <amylaar@cygnus.co.uk>
1407 * expr.c (store_expr): Don't optimize away load-store pair
1408 when either source or destination have a side effect.
1410 Fri Aug 14 16:50:10 1998 John Carr <jfc@mit.edu>
1412 * genrecog.c (add_to_sequence): Fatal error if the modes of the
1413 operands of SET are incompatible.
1415 * alpha.md: Fix max and min patterns so modes of SET operands match.
1417 Fri Aug 14 12:22:55 1998 Ian Lance Taylor <ian@cygnus.com>
1419 * configure.in: Avoid [[ by using test and changequote.
1420 * configure: Rebuild.
1422 Fri Aug 14 01:22:31 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1424 * rtl.def (CONSTANT_P_RTX): Fix typo in string name.
1426 * config/sparc/sparc.md (seqdi_special_trunc, snedi_special_trunc,
1427 seqsi_special_extend, snesi_special_extend, snesi_zero_extend and
1428 split, snedi_zero_trunc and split, seqsi_zero_extend and split,
1429 seqdi_zero_trunc and split, pic_lo_sum_di, pic_sethi_di,
1430 movdi_cc_sp64_trunc, movdi_cc_reg_sp64_trunc, addx_extend_sp32 and
1431 split, addx_extend_sp64, subx_extend_sp64, subx_extend and split):
1432 Fix mismatching modes in SET operands.
1433 (conditional move patterns): Fix formatting.
1434 (unnamed subx arch64 pattern): Remove duplicate insn.
1436 Fri Aug 14 00:34:34 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1438 * config/sparc/sparc.c (const64_operand, const64_high_operand):
1439 Get it right when HOST_BITS_PER_WIDE_INT is not 64.
1440 (input_operand): Fixup test for what we accept for constant
1442 (sparc_emit_set_const32, sparc_emit_set_symbolic_const64): Give
1444 (safe_constDI): Remove.
1445 (sparc_emit_set_safe_HIGH64, gen_safe_SET64, gen_safe_OR64,
1446 gen_safe_XOR64): New functions.
1447 (sparc_emit_set_const64_quick1, sparc_emit_set_const64_quick2,
1448 sparc_emit_set_const64_longway, sparc_emit_set_const64): Use
1450 * config/sparc/sparc.md (define_insn xordi3_sp64_dbl): Only make
1451 available when HOST_BITS_PER_WIDE_INT is not 64.
1452 (define_insn movdi_sp64_dbl, movdi_const64_special): Likewise and
1453 move before movdi_insn_sp64 pattern.
1454 (define_insn movdi_lo_sum_sp64_dbl, movdi_high_sp64_dbl): Remove.
1455 (define_insn sethi_di_medlow, seth44, setm44, sethh): Use
1456 symbolic_operand as predicate for second operand.
1457 (DImode minus split on arch32, negsi2 expander, one_cmplsi2
1458 expander): Give set VOIDmode.
1460 Fri Aug 14 01:45:06 1998 Mumit Khan <khan@xraylith.wisc.edu>
1462 * i386/cygwin32 (DEFAULT_PCC_STRUCT_RETURN): Define.
1464 Fri Aug 14 01:40:21 1998 Geoffrey Keating <geoffk@ozemail.com.au>
1466 * rs6000/linux.h (LINK_SPEC): Pass -G args to the linker.
1468 Fri Aug 14 01:23:23 1998 Richard Earnshaw (rearnsha@arm.com)
1470 * arm/netbsd.h (TARGET_DEFAULT): Default includes software floating
1472 (CPP_FLOAT_DEFAULT_SPEC): Re-define accordingly.
1474 Fri Aug 14 01:19:08 1998 Robert Lipe <robertl@dgii.com>
1476 * install.texi: Various SCO OpenServer tweaks.
1478 Thu Aug 13 20:14:40 1998 Jim Wilson <wilson@cygnus.com>
1480 * reload1.c (eliminate_regs_in_insn): Handle another case when
1481 eliminating the frame pointer to the hard frame pointer. Add
1482 missing ep->to_rtx check to one existing case.
1484 * mips/mips.md (movhi_internal2+2): Fix typo mem:SI -> mem:HI.
1486 Thu Aug 13 17:08:11 1998 Jason Merrill <jason@yorick.cygnus.com>
1488 * tree.h: De-conditionalize init_priority code.
1490 * mips.h (NM_FLAGS): Change from -Bp to -Bn.
1491 * collect2.c (NM_FLAGS): Change from -p to -n.
1493 * configure.in: Turn on collect2 for mipstx39-elf.
1494 Handle use_collect2=no properly.
1496 * c-common.c: De-conditionalize init_priority code.
1497 * collect2.c (extract_init_priority, sort_ids): New fns.
1498 (main): Call sort_ids.
1499 Move sequence_number to file scope.
1501 * configure.in: Handle --enable-init-priority.
1502 * c-common.c (attrs): Add A_INIT_PRIORITY.
1503 (init_attributes, decl_attributes): Likewise.
1504 * tree.h (DEFAULT_INIT_PRIORITY, MAX_INIT_PRIORITY): New macros.
1505 * tree.c (get_file_function_name_long): Split out...
1506 (get_file_function_name): ...from here.
1508 Thu Aug 13 16:09:53 1998 Martin von Loewis <loewis@informatik.hu-berlin.de>
1510 * expr.c (safe_from_p): Change code to ERROR_MARK only when not
1513 Thu Aug 13 15:24:48 1998 Jason Merrill <jason@yorick.cygnus.com>
1515 * toplev.c (display_help): Add braces to shut up warnings.
1516 * tree.c (simple_cst_equal): Likewise.
1518 * fold-const.c (non_lvalue): Don't deal with null pointer
1520 (fold, case COMPOUND_EXPR): Wrap a constant 0 in a NOP_EXPR.
1522 * c-typeck.c (initializer_constant_valid_p): Allow conversion of 0
1523 of any size to a pointer.
1525 Thu Aug 13 12:53:13 1998 Jim Wilson <wilson@cygnus.com>
1527 * i386/winnt.c (i386_pe_asm_file_end): Check TREE_SYMBOL_REFERENCED.
1529 Wed Aug 12 17:25:18 1998 Jeffrey A Law (law@cygnus.com)
1531 * mn10300.c (REG_SAVE_BYTES): Only reserve space for registers
1532 which will be saved.
1533 * mn10300.md (prologue insn): Only save registers which need saving.
1534 (epilogue insn): Similarly.
1536 * mn10300.c, mn10300.h, mn10300.md: Remove "global zero register"
1539 Wed Aug 12 12:39:16 1998 Gavin Romig-Koch <gavin@cygnus.com>
1541 * mips/mips.h (ENCODE_SECTION_INFO): Set SYMBOL_REF_FLAG for
1542 VAR_DECL's in gp addressable sections.
1544 Tue Aug 11 23:02:31 1998 John Carr <jfc@mit.edu>
1546 * sparc.c: Change return <exp> to <exp>; return; in functions
1548 * sparc.md: Add empty semicolon statement after final label in
1551 Tue Aug 11 22:42:01 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1553 * config/sparc/sparc.md (define_insn addx_extend): Rename to
1554 addx_extend_sp64, only allow when TARGET_ARCH64.
1555 (define_insn addx_extend_sp32 and split): Version that works when
1557 (define_insn subx_extend): Likewise.
1558 (define_split adddi3 and subdi3 with zero extension): Fixup and
1559 correct bugs when not TARGET_ARCH64.
1561 Tue Aug 11 16:04:34 1998 John Carr <jfc@mit.edu>
1563 * except.c (set_exception_lang_code, set_exception_version_code):
1564 Use prototype-style definition if __STDC__, to match declaration
1567 * genemit.c: Change FAIL and DONE macros not to use loops.
1569 Tue Aug 11 12:27:03 1998 Jim Wilson <wilson@cygnus.com>
1571 * dwarf2out.c (ASM_OUTPUT_DWARF_ADDR_CONST): Use
1572 ASM_OUTPUT_DWARF2_ADDR_CONST if defined.
1574 * mips/mips.md (reload_outsi): Use M16_REG_P when TARGET_MIPS16.
1576 Tue Aug 11 18:12:53 1998 Dave Love <d.love@dl.ac.uk>
1578 * README.g77: Update from Craig.
1580 Tue Aug 11 04:46:01 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1582 * config/sparc/sparc.c (sparc_emit_set_const32): INTVAL is of
1584 (safe_constDI sparc_emit_set_const64_quick1,
1585 sparc_emit_set_const64_quick2, sparc_emit_set_const64_longway,
1586 analyze_64bit_constant, const64_is_2insns,
1587 create_simple_focus_bits): Fix some bugs when compiled on real
1589 (function_arg_record_value_3, function_arg_record_value_2,
1590 function_arg_record_value): Add fully prototyped forward decls.
1591 * config/sparc/sparc.md (define_insn cmpsi_insn_sp32): Rename back
1592 to cmpsi_insn and use on both 64 and 32 bit targets.
1593 (define_insn cmpsi_insn_sp64): Remove.
1594 (define_expand zero_extendsidi2): Allow for 32-bit target too.
1595 (define_insn zero_extendsidi2_insn): Rename to
1596 zero_extendsidi2_insn_sp64.
1597 (define_insn zero_extendsidi2_insn_sp32): New pattern and
1598 assosciated forced split for it.
1600 * config/sparc/sparc.c (const64_operand, const64_high_operand):
1602 * config/sparc/sparc.h: Declare them.
1603 (PREDICATE_CODES): Add them.
1604 * config/sparc/sparc.md (movdi_lo_sum_sp64_dbl,
1605 movdi_high_sp64_dbl, xordi3_sp64_dbl): Use them.
1607 Mon Aug 10 22:57:24 1998 John Carr <jfc@mit.edu>
1609 * config/sparc/sparc.md (define_insn jump): Output ba,pt not b,pt
1610 in v9 case as the latter makes the Solaris assembler crash.
1612 Mon Aug 10 22:39:09 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1614 * config/sparc/sparc.c (input_operand): Do not accept a LO_SUM MEM
1615 for TFmode when !v9. We require offsettable memory addresses.
1616 * config/sparc/sparc.h (ALTER_HARD_SUBREG): Handle TFmode to
1617 DFmode register number conversions.
1618 * config/sparc/sparc.md (define_split DFmode moves): If register
1619 is a SUBREG do alter_subreg on it before using.
1620 (define_expand movtf): Fixup comment about alignment on v9.
1621 (define_split TFmode moves): Don't use gen_{high,low}part, create
1622 explicit SUBREGs instead.
1624 Mon Aug 10 19:02:55 1998 John Carr <jfc@mit.edu>
1626 * Makefile.in (mbchar.o): Depend on mbchar.c.
1628 Mon Aug 10 04:28:13 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
1629 Richard Henderson <rth@cygnus.com>
1631 Rewrite Sparc backend for better code generation and
1632 improved sparc64 support.
1633 * config/sparc/sp64-elf.h: Set JUMP_TABLES_IN_TEXT_SECTION to
1635 * config/sparc/sysv4.h: Likewise.
1636 * config/sparc/sparc.c (v8plus_regcmp_p, sparc_operand,
1637 move_operand, v8plus_regcmp_op, emit_move_sequence,
1638 singlemove_string, doublemove_string, mem_aligned_8,
1639 output_move_double, output_move_quad, output_fp_move_double,
1640 move_quad_direction, output_fp_move_quad, output_scc_insn):
1642 (small_int_or_double): New predicate.
1643 (gen_compare_reg): Remove TARGET_V8PLUS cmpdi_v8plus emission.
1644 (legitimize_pic_address): Emit movsi_{high,lo_sum}_pic instead of
1645 old pic_{sethi,lo_sum}_si patterns.
1646 (mem_min_alignment): New generic function to replace
1647 mem_aligned_8, which uses REGNO_POINTER_ALIGN information when
1648 available and can test for arbitrary alignments. All callers
1650 (save_regs, restore_regs, build_big_number,
1651 output_function_prologue, output_cbranch, output_return,
1652 sparc_flat_save_restore, sparc_flat_output_function_prologue,
1653 sparc_flat_output_function_epilogue): Prettify
1655 (output_function_epilogue): Likewise and add code to output
1656 deferred case vectors.
1657 (output_v9branch): Likewise, add new arg INSN and use it to tack
1658 on branch prediction settings. All callers changed.
1659 (print_operand): Likewise and output %l44 for LO_SUMs when
1661 (sparc_splitdi_legitimate): New function to make sure DImode
1662 splits can be run properly when !arch64.
1663 (sparc_initialize_trampoline, sparc64_initialize_trampoline):
1664 Reformat example code in comments.
1665 (set_extends): Remove UNSPEC/v8plus_clear_high case.
1666 (sparc_addr_diff_list, sparc_addr_list): New statics to keep track
1667 of deferred case vectors we need to output.
1668 (sparc_defer_case_vector): Record a case vector.
1669 (sparc_output_addr_vec, sparc_output_addr_diff_vec,
1670 sparc_output_deferred_case_vectors): New functions to output them.
1671 (sparc_emit_set_const32): New function to form 32-bit constants in
1672 registers when that requires more than one instruction.
1673 (safe_constDI, sparc_emit_set_const64_quick1,
1674 sparc_emit_set_const64_quick2, sparc_emit_set_const64_longway,
1675 analyze_64bit_constant, const64_is_2insns,
1676 create_simple_focus_bits, sparc_emit_set_const64): New functions
1677 which do the same for 64-bit constants when arch64.
1678 (sparc_emit_set_symbolic_const64): New function to emit address
1679 loading for all code models on v9.
1680 * config/sparc/sparc.h (CONDITIONAL_REGISTER_USAGE): Do not make
1681 %g1 fixed when arch64, unfix %g0 when TARGET_LIVE_G0.
1682 (ALTER_HARD_SUBREG): Fix thinko, return REGNO + 1 not 1.
1683 (SECONDARY_INPUT_RELOAD_CLASS, SECONDARY_OUTPUT_RELOAD_CLASS): Fix
1684 inaccuracies in comments, add symbolic and text_segment operands
1685 when TARGET_CM_MEDANY and TARGET_CM_EMBMEDANY respectively. Use
1686 GENERAL_REGS in these cases as a temp REG is needed to load these
1687 addresses into a register properly.
1688 (EXTRA_CONSTRAINT): Document more accurately, remove Q case as it
1690 (GO_IF_LEGITIMATE_ADDRESS): Allow TFmode for LO_SUM on v9 since fp
1691 quads are guarenteed to have 16-byte alignment.
1692 (LEGITIMIZE_ADDRESS): For SYMBOL_REF, CONST, and LABEL_REF use
1693 copy_to_suggested_reg instead of explicit LO_SUM and HIGH.
1694 (ASM_OUTPUT_ADDR_VEC, ASM_OUTPUT_ADDR_DIFF_VEC): New macros for
1695 deferred case vector implementation.
1696 (ASM_OUTPUT_ADDR_VEC_ELT): Use fputc to output newline.
1697 (ASM_OUTPUT_ADDR_DIFF_ELT): Parenthesize LABEL in macro calls.
1698 Generate "internal label - label" instead of "label - 1b".
1699 (PRINT_OPERAND_ADDRESS): For LO_SUM use %l44 on TARGET_CM_MEDMID.
1700 (PREDICATE_CODES): Remove sparc_operand, move_operand,
1701 v8plus_regcmp_op. Add small_int_or_double, input_operand, and
1703 (doublemove_string, output_block_move, output_fp_move_double,
1704 output_fp_move_quad, output_move_double, output_move_quad,
1705 output_scc_insn, singlemove_string, mem_aligned_8, move_operand,
1706 sparc_operand, v8plus_regcmp_op, v8plus_regcmp_p): Remove externs.
1707 (sparc_emit_set_const32, sparc_emit_set_const64,
1708 sparc_emit_set_symbolic_const64, input_operand, zero_operand,
1709 mem_min_alignment, small_int_or_double): Add externs.
1710 * config/sparc/sparc.md: Document the many uses of UNSPEC and
1711 UNSPEC_VOLATILE in this backend.
1712 (define_function_unit ieu): Rename to ieu_unnamed. Add move and
1713 unary to types which execute in it.
1714 (define_function_unit ieu_shift): Rename to ieu0.
1715 (define_function_unit ieu1): New, executes compare, call, and
1716 uncond_branch type insns.
1717 (define_function_units for type fdivs, fdivd, fsqrt): These
1718 execute in the fpu multiply unit not the adder on UltraSparc.
1719 (define_expand cmpdi): Disallow TARGET_V8PLUS.
1720 (define_insn cmpsi_insn): Rename to cmpsi_insn_sp32.
1721 (define_insn cmpsi_insn_sp64): New, same as sp32 variant except it
1722 allows the arith_double_operand predicate and rHI constraint when
1724 (define_insn cmpdi_sp64, cmpsf_fpe, cmpdf_fpe, cmptf_fpe,
1725 cmpsf_fp, cmpdf_fp, cmptf_fp, sltu_insn, neg_sltu_insn,
1726 neg_sltu_minux_x, neg_sltu_plus_x, sgeu_insn, neg_sgeu_insn,
1727 sltu_plus_x, sltu_plus_x, sltu_plus_x_plus_y, x_minus_sltu,
1728 sgeu_plus_x, x_minus_sgeu, movqi_cc_sp64, movhi_cc_sp64,
1729 movsi_cc_sp64, movdi_cc_sp64, movsf_cc_sp64, movdf_cc_sp64,
1730 movtf_cc_sp64, movqi_cc_reg_sp64, movhi_cc_reg_sp64,
1731 movsi_cc_reg_sp64, movdi_cc_reg_sp64, movsf_cc_reg_sp64,
1732 movdf_cc_reg_sp64, movtf_cc_reg_sp64, zero_extendhisi2_insn,
1733 cmp_siqi_trunc, cmp_siqi_trunc_set, sign_extendhisi2_insn,
1734 sign_extendqihi2_insn, sign_extendqisi2_insn,
1735 sign_extendqidi2_insn, sign_extendhidi2_insn,
1736 extendsfdf2, extendsftf2, extenddftf2, truncdfsf2, trunctfsf2,
1737 trunctfdf2, floatsisf2, floatsidf2, floatsitf2, floatdisf2,
1738 floatdidf2, floatditf2, fix_truncsfsi2, fix_truncdfsi2,
1739 fix_trunctfsi2, fix_truncsfdi2, fix_truncdfdi2, fix_trunctfdi2,
1740 adddi3_sp64, addsi3, cmp_ccx_plus, cmp_cc_plus_set, subdi_sp64,
1741 subsi3, cmp_minus_ccx, cmp_minus_ccx_set, mulsi3, muldi3,
1742 muldi3_v8plus, cmp_mul_set, mulsidi3, mulsidi3_v8plus,
1743 const_mulsidi3_v8plus, mulsidi3_sp32, const_mulsidi3,
1744 smulsi3_highpart_v8plus, unnamed subreg mult,
1745 const_smulsi3_highpart_v8plus, smulsi3_highpart_sp32,
1746 const_smulsi3_highpart, umulsidi3_v8plus, umulsidi3_sp32,
1747 const_umulsidi3, const_umulsidi3_v8plus, umulsi3_highpart_v8plus,
1748 const_umulsi3_highpart_v8plus, umulsi3_highpart_sp32,
1749 const_umulsi3_highpart, divsi3, divdi3, cmp_sdiv_cc_set, udivsi3,
1750 udivdi3, cmp_udiv_cc_set, smacsi, smacdi, umacdi, anddi3_sp64,
1751 andsi3, and_not_di_sp64, and_not_si, iordi3_sp64, iorsi3,
1752 or_not_di_sp64, or_not_si, xordi3_sp64, xorsi3, xor_not_di_sp64,
1753 xor_not_si, cmp_cc_arith_op, cmp_ccx_arith_op,
1754 cmp_cc_arith_op_set, cmp_ccx_arith_op_set, cmp_ccx_xor_not,
1755 cmp_cc_xor_not_set, cmp_ccx_xor_not_set, cmp_cc_arith_op_not,
1756 cmp_ccx_arith_op_not, cmp_cc_arith_op_not_set,
1757 cmp_ccx_arith_op_not_set, negdi2_sp64, cmp_cc_neg, cmp_ccx_neg,
1758 cmp_cc_set_neg, cmp_ccx_set_neg, one_cmpldi2_sp64, cmp_cc_not,
1759 cmp_ccx_not, cmp_cc_set_not, cmp_ccx_set_not, addtf3, adddf3,
1760 addsf3, subtf3, subdf3, subsf3, multf3, muldf3, mulsf3,
1761 muldf3_extend, multf3_extend, divtf3, divdf3, divsf3, negtf2,
1762 negdf2, negsf2, abstf2, absdf2, abssf2, sqrttf2, sqrtdf2, sqrtsf2,
1763 ashlsi3, ashldi3, unnamed DI ashift, cmp_cc_ashift_1,
1764 cmp_cc_set_ashift_1, ashrsi3, ashrdi3, unnamed DI ashiftrt,
1765 ashrdi3_v8plus, lshrsi3, lshrdi3, unnamed DI lshiftrt,
1766 lshrdi3_v8plus, tablejump_sp32, tablejump_sp64, call_address_sp32,
1767 call_symbolic_sp32, call_address_sp64, call_symbolic_sp64,
1768 call_address_struct_value_sp32, call_symbolic_struct_value_sp32,
1769 call_address_untyped_struct_value_sp32,
1770 call_symbolic_untyped_struct_value_sp32, call_value_address_sp32,
1771 call_value_symbolic_sp32, call_value_address_sp64,
1772 call_value_symbolic_sp64, branch_sp32, branch_sp64,
1773 flush_register_windows, goto_handler_and_restore,
1774 goto_handler_and_restore_v9, goto_handler_and_restore_v9_sp64,
1775 flush, all ldd/std peepholes, return_qi, return_hi, return_si,
1776 return_addsi, return_di, return_adddi, return_sf, all call+jump
1777 peepholes, trap, unnamed trap insns): Prettify output strings.
1778 (define_insn anddi3_sp32, and_not_di_sp32, iordi3_sp32,
1779 or_not_di_sp32, xordi3_sp32, xor_not_di_sp32, one_cmpldi2):
1780 Likewise and force + implement splits for integer cases.
1781 (define_insn return_sf_no_fpu): Likewise and allow to match when
1782 no-fpu because of our subreg SFmode splits.
1783 (define_insn zero_extendqihi2, zero_extendqisi2_insn,
1784 zero_extendqidi2_insn, zero_extendhidi2_insn,
1785 zero_extendsidi2_insn, sign_extendsidi2_insn): Likewise and use
1786 input_operand for second operand.
1787 (cmp_minus_cc, cmp_minus_cc_set): Likewise and use
1788 reg_or_0_operand for operand 2 so new splits can use it.
1789 (cmp_zero_extendqisi2, cmp_zero_extendqisi2_set, cmp_cc_plus,
1790 cmp_cc_xor_not): Likewise and don't forget to check TARGET_LIVE_G0
1792 (cmp_zero_extract, cmp_zero_extract_sp64): Likewise and allow
1793 CONST_DOUBLEs for operand 2.
1794 (define_insn move_label_di): Likewise and label distance
1795 optimization because it no longer works with new deferred case
1796 vector scheme. To be revisited.
1797 (define_insn x_minus_y_minus_sltu, x_minus_sltu_plus_y): Likewise
1798 and allow reg_or_0_operand and J constraint for second operand.
1799 (define_insn jump): Set branch predict taken on V9.
1800 (define_insn tablejump): Emit LABEL_REF + PLUS memory address for
1801 new deferred case vector scheme.
1802 (define_insn pic_tablejump_32, pic_tablejump_64): Remove.
1803 (define_insn negdi2_sp32): Force + implement splits.
1804 (define_insn negsi2, one_cmplsi2): Rename to negsi2_not_liveg0 and
1805 one_cmplsi2_not_liveg0 respectively, and create expander of original
1806 names which emit special rtl for TARGET_LIVE_G0.
1807 (define_insn cmpdi_v8plus, scc_si, scc_di): Remove.
1808 (define_insn seq, sne, slt, sge, sle, sltu, sgeu): Don't do
1809 gen_compare_reg, FAIL instead.
1810 (define_insn sgtu, sleu): Likewise and check gen_s*() return
1811 values when trying to reverse condition codes, if they FAIL then
1813 (define_insn snesi_zero, neg_snesi_zero, snesi_zero_extend,
1814 snedi_zero, neg_snedi_zero, snedi_zero_trunc, seqsi_zero,
1815 neg_seqsi_zero, seqsi_zero_extend, seqdi_zero, neg_seqdi_zero,
1816 seqdi_zero_trunc, x_plus_i_ne_0, x_minus_i_ne_0, x_plus_i_eq_0,
1817 x_minus_i_eq_0): Add new splits to perform these multi-insn cases,
1818 set output string to # to indicate they are mandatory splits.
1819 (define_insn pic_lo_sum_si, pic_sethi_si, pic_lo_sum_di,
1820 pic_sethi_di, move_pic_label_si): Remove.
1821 (define_insn movsi_low_sum, movsi_high, movsi_lo_sum_pic,
1822 movsi_high_pic, movsi_pic_label_reg): New patterns to take their
1824 (define_expand movsi_pic_label_ref, define_insn
1825 movsi_high_pic_label_ref, movsi_lo_sum_pic_label_ref): New
1826 expander and insns to handle PIC label references and deferred
1828 (define_insn get_pc_via_rdpc): Comment out as it is no longer
1830 (define_expand movqi, movhi, movsi, movdi, movsf, movdf, movtf):
1831 Rewrite to not use emit_move_sequence, make use of new constant
1832 formation code, and new splits for all multi-insn cases.
1833 (define_insn movqi_insn): Remove sethi case, it can never happen.
1834 Use reg_or_zero_operand instead of const0_rtx explicit test,
1835 use input_operand instead of move_operand for source, and use
1836 general_operand now for dest.
1837 (define_insn movhi_insn): Similar but leave sethi case.
1838 (define_insn lo_sum_qi, store_qi, store_hi): Remove.
1839 (define_insn sethi_hi lo_sum_hi): Rename to movhi_high and
1840 movhi_lo_sum respectively, prettify output string.
1841 (define_insn movsi_zero_liveg0): New pattern to put zero into a
1842 register when needed on TARGET_LIVE_G0.
1843 (define_insn movsi_insn): Use general_operand and input_operand
1844 for dest and src respectively. Simplify applicability test.
1845 Prettify output strings, and add clr alternative for J
1847 (define_insn movdi_sp32_v9, movdi_sp32, define_splits for
1848 deprecated std and reg-reg DI moves): Remove and...
1849 (define_insn movdi_insn_sp32, movdi_insn_sp64): Replace with new
1850 implementation which uses forced splits for all non-single insn
1852 (define_split DI move cases on !arch64): New splits to handle all
1853 situations of 64-bit double register DImode on 32bit, and
1854 unaligned registers and memory addresses for all subtargets.
1855 (define_insn movsf_const_insn, movdf_const_insn, store_sf):
1857 (define_insn movsf_insn, movsf_no_f_insn): Use general_operand and
1858 input_operand for dest and src respectively, prettify output
1860 (define_insn movdf_insn, movdf_no_e_insn, store_df,
1861 movtf_const_insn, movtf_insn, movtf_no_e_insn, store_tf): Remove
1863 (define_insn movdf_insn_sp32, movdf_no_e_insn_sp32,
1864 movdf_insn_sp64, movdf_no_e_insn_sp64, movtf_insn,
1865 movtf_no_e_insn_sp32, movtf_insn_hq_sp64, movtf_insn_sp64,
1866 movtf_no_e_insn_sp64) Replace with new
1867 implementation which uses forced splits for all non-single insn
1869 (define_split DF move cases): New splits in similar vein to DI
1871 (define_insn sethi_di_medlow, sethi_di_medium_pic,
1872 sethi_di_embmedany_data, sethi_di_embmedany_text, sethi_di_sp64,
1873 movdi_sp64_insn): Remove old v9 code model and constant loading
1875 (define_insn pic_lo_sum_di, pic_sethi_di,
1876 sethi_di_medlow_embmedany_pic, sethi_di_medlow, losum_di_medlow,
1877 seth44, setm44, setl44, sethh, setlm, sethm, setlo,
1878 embmedany_sethi, embmedany_losum, embmedany_brsum,
1879 embmedany_textuhi, embmedany_texthi, embmedany_textulo,
1880 embmedany_textlo, movdi_lo_sum_sp64_cint, movdi_lo_sum_sp64_dbl,
1881 movdi_high_sp64_cint, movdi_high_sp64_dbl): Replace with new
1882 scheme, using unspecs, secondary reloads, and one to one sparc
1883 insn to rtl insn mapping for better scheduling and code gen.
1884 (define_expand reload_indi, reload_outdi): Reload helpers for
1885 MEDANY and EMBMEDANY symbol address loading cases which require a
1887 (define_expand movsicc): Remove v8plus_regcmp cases.
1888 (define_insn movdi_cc_sp64_trunc, movdi_cc_reg_sp64_trunc,
1889 cmp_zero_extendqidi2, cmp_zero_extendqidi2_set, cmp_qidi_trunc,
1890 cmp_diqi_trunc_set): New patterns used by some of the new scc
1892 (define_insn xordi3_sp64_dbl): New pattern used for constant
1893 formation when crossing from 32-bit targets.
1894 (define_insn movsi_cc_reg_v8plus, v8plus_clear_high, and helper
1896 (define_insn addx, subx): Make visible and prettify.
1897 (define_insn adddi3_insn_sp32): Likewise and force split.
1898 (define_insn addx_extend, subx_extend, unnamed): New patterns for
1899 64bit scc split usage.
1900 (define_insn unnamed plusDI zero_extend, unnamed minusDI
1901 zero_extend, subdi3): Force and implement splits.
1903 * final.c (final_scan_insn): Don't output labels if target
1904 specifies ASM_OUTPUT_ADDR_{DIFF}_VEC. Do these macro operations
1907 * reorg.c (dbr_schedule): When taking on BR_PRED notes at the end,
1908 don't forget to walk inside SEQUENCESs too as these are what the
1909 delay slot scheduler will create.
1911 Mon Aug 10 01:21:01 1998 Richard Henderson <rth@cygnus.com>
1913 * alpha.md (extxl+1,+2): New patterns to work around
1916 Sat Aug 8 19:20:22 1998 Gary Thomas (gdt@linuxppc.org)
1918 * rs6000.c (rs6000_allocate_stack_space) Fix typo which
1919 caused bad assembly code to be generated.
1921 Sat Aug 8 18:53:28 1998 Jeffrey A Law (law@cygnus.com)
1923 * netbsd.h: Fix typo.
1925 Mon Aug 3 00:06:42 1998 Robert Lipe <robertl@dgii.com>
1927 * config.sub: Fix typo.
1929 Sun Aug 2 22:39:08 1998 Hans-Peter Nilsson <hp@axis.se>
1931 * invoke.texi (Environment Variables): Typo: Change "ascpects"
1933 (Running Protoize): Typo: Change "ther" into "other".
1935 Sun Aug 2 00:42:50 1998 Jeffrey A Law (law@cygnus.com)
1937 * i386/netbsd.h: Undo previous change to DWARF2_UNWIND_INFO.
1938 * m68k/netbsd.h: Likewise.
1939 * ns32k/netbsd.h: Likewise.
1940 * sparc/netbsd.h: Likewise.
1942 Sat Aug 1 17:59:30 1998 Richard Henderson <rth@cygnus.com>
1944 * ginclude/va-alpha.h (va_list): Use a typedef, not a define.
1945 * ginclude/va-clipper.h (va_list): Likewise.
1947 Fri Jul 31 20:22:02 1998 Michael Meissner <meissner@cygnus.com>
1949 * rs6000.c (rs6000_override_options): If big endian and -Os, use
1950 load/store multiple instructions unless user overrides.
1952 Fri Jul 31 17:08:59 1998 Jeffrey A Law (law@cygnus.com)
1954 * ns32k/netbsd.h: Fix typo.
1956 Fri Jul 31 10:23:55 1998 Doug Evans <devans@canuck.cygnus.com>
1958 * m32r/m32r.h (ASM_OUTPUT_SOURCE_LINE): Always output line number
1959 labels with .debugsym if no parallel insns.
1961 Thu Jul 30 19:15:53 1998 Richard Henderson <rth@cygnus.com>
1963 * alpha.md (fp cmp): Replicate patterns for ALPHA_TP_INSN.
1964 (fcmov): Remove ALPHA_TP_INSN patterns -- fcmov doesn't trap.
1966 Thu Jul 30 19:50:15 1998 David Edelsohn <edelsohn@mhpcc.edu>
1968 * rs6000/x-aix43 (AR_FOR_TARGET_FLAGS): Delete.
1969 (AR_FOR_TARGET): Define.
1971 Thu Jul 30 12:29:12 1998 Mark Mitchell <mark@markmitchell.com>
1973 * dyn-string.h: New file.
1974 * dyn-string.c: Likewise.
1975 * Makefile.in (OBJS): Add dyn-string.o.
1976 (dwarf2out.o): Add dyn-string.h dependency.
1977 (dyn-string.o): List dependencies.
1978 * dwarf2out.c: Include dyn-string.h.
1979 (ASM_NAME_TO_STRING): Use dyn_string_append, rather than strcpy.
1980 (addr_const_to_string): Take a dyn_string_t, not a char* as a
1981 prototype. Use dyn_string_append rather than strcat, throughout.
1982 (addr_to_string): Use dyn_string_t.
1984 Thu Jul 30 13:08:07 1998 Ken Raeburn <raeburn@cygnus.com>
1986 Function entry/exit profiling instrumentation:
1987 * expr.h (profile_function_entry_libfunc,
1988 profile_function_exit_libfunc): Declare new variables.
1989 * optabs.c: Define them here.
1990 (init_optabs): Initialize them.
1991 * tree.h (struct tree_decl): New flag
1992 no_instrument_function_entry_exit.
1993 (DECL_NO_INSTRUMENT_FUNCTION_ENTRY_EXIT): New accessor macro.
1994 * c-decl.c (duplicate_decls): Merge it.
1995 * c-common.c (enum attrs): New value A_NO_INSTRUMENT_FUNCTION.
1996 (init_attributes): Use it for "no_instrument_function".
1997 (decl_attributes): Handle it, for functions that have not yet been
1998 compiled. Set decl flag.
1999 * flags.h (flag_instrument_function_entry_exit): Declare new
2001 * toplev.c (flag_instrument_function_entry_exit): Define it here.
2002 (f_options): New option "instrument-functions".
2003 * function.h (struct function): New field instrument_entry_exit.
2004 * function.c (current_function_instrument_entry_exit): New
2006 (push_function_context_to, pop_function_context_from): Save and
2008 (expand_function_start): Set current_ variable, maybe emit return
2009 label and entry profile call.
2010 (expand_function_end): Maybe emit exit profile call.
2012 Thu Jul 30 00:58:34 1998 Jeffrey A Law (law@cygnus.com)
2014 * i386.md (movqi): When optimizing a load of (const_int 1) into a
2015 NON_QI_REG_P, pretend the register is SImode.
2017 Wed Jul 29 23:49:23 1998 Todd Vierling <tv@netbsd.org>
2019 * configure.in: Use xm-netbsd.h as the NetBSD xm file (not xm-siglist).
2020 Accept arm32 as arm, m68k4k as m68k, mipsle as mips-dec, and any
2021 manufacturer id for ns32k.
2022 * configure: Regenerated.
2023 * config/netbsd.h: When using ASM_WEAKEN_LABEL, make it global too.
2024 * config/t-netbsd: Don't compile libgcc1-test as the fns are in libc.
2025 * config/i386/netbsd.h: Undefine DWARF2_UNWIND_INFO, not define as 0.
2026 * config/m68k/netbsd.h: Same.
2027 * config/ns32k/netbsd.h: Same.
2028 * config/sparc/netbsd.h: Same.
2030 Wed Jul 29 22:39:21 1998 Jeffrey A Law (law@cygnus.com)
2032 * unroll.c (unroll_loop): Do not abort for an UNROLL_MODULO
2033 or UNROLL_COMPLETELY loop that starts with a jump to its
2036 Wed Jul 29 22:18:14 1998 David Edelsohn <edelsohn@mhpcc.edu>
2038 * rs6000/rs6000.md (absdi2 define_split): Swap operands of MINUS.
2039 * rs6000/rs6000.c (mask64_operand): Use HOST_BITS_PER_WIDE_INT.
2040 (print_operand, case 'B'): Don't fall through.
2041 (print_operand, case 'S'): Correct mask begin/end computation.
2042 Use HOST_BITS_PER_WIDE_INT.
2043 * rs6000/rs6000.h (CPP_PREDEFINES): Define _LONG_LONG.
2044 (CONDITIONAL_REGISTER_USAGE): GPR13 fixed if TARGET_64BIT.
2045 * rs6000/aix41.h (CPP_PREDEFINES): Same.
2046 * rs6000/aix43.h (CPP_PREDEFINES): Same.
2048 Wed Jul 29 11:47:10 1998 Nick Clifton <nickc@cygnus.com>
2050 * config/arm/thumb.md (extendqisi2_insn): Remove earlyclobber
2051 constraint from second alternative.
2053 Tue Jul 28 23:29:04 1998 Jason Merrill <jason@yorick.cygnus.com>
2055 * configure.in: Fix --without/--disable cases for local-prefix,
2056 gxx-include-dir and checking.
2058 Tue Jul 28 22:01:23 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
2060 * configure.in (enable_haifa): Set by default for sparc64 too.
2063 Tue Jul 28 23:29:04 1998 Jason Merrill <jason@yorick.cygnus.com>
2065 * i386/cygwin32.h (VALID_MACHINE_TYPE_ATTRIBUTE): New macro.
2066 * i386/winnt.c (associated_type): New fn.
2067 (i386_pe_valid_type_attribute_p): New fn.
2068 (i386_pe_check_vtable_importexport): Remove.
2069 (i386_pe_dllexport_p): Use associated_type.
2070 (i386_pe_dllimport_p): Likewise.
2072 From Antonio M. O. Neto <anmendes@cruzeironet.com.br>:
2073 * i386.c (i386_valid_type_attribute_p): Also accept
2074 attributes for METHOD_TYPEs.
2076 Tue Jul 28 23:17:39 1998 Peter Gerwinski <peter@gerwinski.de>
2078 * tree.c (build_range_type): Copy TYPE_SIZE_UNIT.
2080 Tue Jul 28 22:31:12 1998 Craig Burley <burley@gnu.org>
2082 * gcc.c: Fix commentary describing %g, %u, %U, and %O.
2084 * gcc.c (do_spec_1): Fix handling of %g%O and %U%O to prevent
2085 them from generating a new base name for each occurence of
2088 1998-07-28 Vladimir N. Makarov <vmakarov@cygnus.com>
2090 * cse.c (cse_insn): Enable subsitution inside libcall only for REG,
2092 * rtlanal.c (replace_rtx): Prohibit replaces in CONST_DOUBLE.
2096 * cplus-dem.c (type_kind_t): New type.
2097 (demangle_template_value_parm): Add type_kind_t parameter. Rely
2098 on this paramter, rather than demangling the type again.
2099 (demangle_integral_value): Pass tk_integral.
2100 (demangle_template_: Pass the value returned from do_type.
2101 (do_type): Return a type_kind_t. Pass tk_integral to
2102 demangle_template_value_parm for array bounds.
2103 (demangle_fund_type): Likewise.
2105 Mon Jul 27 00:54:41 1998 Jason Merrill <jason@yorick.cygnus.com>
2107 * tree.c (simple_cst_equal, case CONSTRUCTOR): OK if the elts are
2110 Mon Jul 27 22:18:36 1998 Jeffrey A Law (law@cygnus.com)
2112 * pa.c (move_operand): Accept CONSTANT_P_RTX.
2114 Mon Jul 27 17:18:52 1998 Dave Brolley <brolley@cygnus.com>
2116 * stor-layout.c (layout_type): Handle arrays of bits, for Chill.
2118 * expr.c (get_inner_reference): Handle zero-based, unsigned, array
2121 Mon Jul 27 14:51:33 1998 Jeffrey A Law (law@cygnus.com)
2123 * mn10300.h (DEBUGGER_AUTO_OFFSET): Define.
2124 (DEBUGGER_ARG_OFFSET): Likewise.
2126 * mn10300.md (movsf): Remove last change. Not needed.
2128 Mon Jul 27 14:22:36 1998 Dave Brolley <brolley@cygnus.com>
2130 * c-lex.c (yylex): Fix boundary conditions in character literal and
2131 string literal loops.
2133 Mon Jul 27 11:43:54 1998 Stan Cox <scox@cygnus.com>
2135 * longlong.h (count_leading_zeros): Sparclite scan instruction was
2136 being invoked incorrectly.
2138 * i386.c (ix86_prologue): Added SUBTARGET_PROLOGUE invocation.
2139 * i386/cygwin32.h (STARTFILE_SPEC, LIB_SPEC, SUBTARGET_PROLOGUE):
2141 * i386/win32.h: New file. Hybrid mingw32.h/cygwin32.h configuration.
2142 * configure.in: Added i[34567]86-*-win32
2143 * config.sub: Ditto.
2144 * configure: Rebuilt.
2146 Sun Jul 26 01:11:12 1998 H.J. Lu (hjl@gnu.org)
2148 * i386.h (CONST_DOUBLE_OK_FOR_LETTER_P): Return 0 when eliminating
2149 the frame pointer and compiling PIC code and reload has not completed.
2151 * i386.c (output_to_reg): Add code to emulate non-popping DImode
2154 Sun Jul 26 01:01:32 1998 Jeffrey A Law (law@cygnus.com)
2156 * regmove.c (regmove_optimize): Fix typo initializing regmove_bb_head.
2158 Sat Jul 25 23:29:23 1998 Gerald Pfeifer <pfeifer@dbai.tuwien.ac.at>
2160 * Makefile.in (install-info): Only try to update the info
2161 directory file if it exists in the first place.
2163 Fri Jul 24 18:58:37 1998 Klaus Espenlaub <kespenla@student.informatik.uni-ulm.de>
2165 * rs6000.h (ASM_OUTPUT_CONSTRUCTOR, ASM_OUTPUT_DESTRUCTOR): Delete.
2167 Fri Jul 24 14:20:26 1998 Jeffrey A Law (law@cygnus.com)
2169 * mn10300.md (movqi, movhi, movsi, movsf): Correctly handle
2170 CONST_DOUBLE source.
2172 Fri Jul 24 11:17:04 1998 Nick Clifton <nickc@cygnus.com>
2174 * config/arm/thumb.c (thumb_print_operand): Decode %_ in asm
2175 strings as the insertion of USER_LABEL_PREFIX.
2176 * config/arm/thumb.h (PRINT_OPERAND_PUNCT_VALID_P): Accept _ as a
2178 * config/arm/thumb.md: Use %_ as a prefix to gcc library function
2181 Thu Jul 23 18:53:20 1998 Jim Wilson <wilson@cygnus.com>
2183 * dbxout.c (dbxout_range_type): Only call dbxout_type_index for
2184 already defined type.
2186 Thu Jul 23 13:49:41 1998 Jeffrey A Law (law@cygnus.com)
2188 * expr.c (check_max_integer_computation_mode): Allow conversions
2189 of constant integers to MAX_INTEGER_COMPUTATION_MODE.
2190 (expand_expr): Likewise.
2192 Thu Jul 23 11:12:06 1998 Alexandre Petit-Bianco <apbianco@cygnus.com>
2194 * expr.c (expand_expr): Expand RETURN_EXPR.
2196 Thu Jul 23 11:00:29 1998 Jim Wilson <wilson@cygnus.com>
2198 * dwarf2out.c (dwarf2out_finish): Call stripattributes on TEXT_SECTION.
2200 Wed Jul 22 19:10:00 1998 Catherine Moore <clm@cygnus.com>
2202 * dwarf2out.c (output_aranges): Call stripattributes
2203 for TEXT_SECTION references.
2204 (output_line_info): Likewise.
2206 Wed Jul 22 14:08:54 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
2208 * profile.c (branch_prob): Call allocate_reg_info after outputting
2209 profile rtl in instrument_arcs.
2211 Wed Jul 22 12:47:49 1998 Jim Wilson <wilson@cygnus.com>
2213 * fixinc.irix (math.h): Install wrapper instead of copying.
2215 Wed Jul 22 12:37:14 1998 Alexandre Petit-Bianco <apbianco@cygnus.com>
2217 * tree.def (EXPR_WITH_FILE_LOCATION): Defined as an 'e' expression
2218 so WFL are expanded correctly when contained in a COMPOUND_EXPR.
2219 * tree.h (EXPR_WFL_EMIT_LINE_NOTE): Change macro not to use
2220 lang_flag_0. Added documentation in the flag table.
2222 Tue Jul 21 23:28:35 1998 Klaus Kaempf <kkaempf@rmi.de>
2224 * cccp.c (do_include): Fix vax c style include handling.
2226 Tue Jul 21 13:28:19 1998 Jason Merrill <jason@yorick.cygnus.com>
2228 * cplus-dem.c (do_type): Use demangle_template_value_parm for arrays.
2230 Sun Jul 12 01:27:05 1998 Jason Merrill <jason@yorick.cygnus.com>
2232 * fold-const.c (non_lvalue): Don't deal with null pointer
2234 (fold, case COMPOUND_EXPR): Wrap a constant 0 in a NOP_EXPR.
2236 Tue Jul 21 15:49:31 1998 David Edelsohn <edelsohn@mhpcc.edu>
2238 * rs6000.h (PREDICATE_CODES): Add CONSTANT_P_RTX.
2239 * rs6000.md (movsi, movdi): Add CONSTANT_P_RTX.
2240 * rs6000.c (short_cint_operand): Add CONSTANT_P_RTX.
2241 (u_short_cint_operand): Same.
2242 (reg_or_cint_operand): Same.
2243 (logical_operand): Same.
2244 (input_operand): Same.
2245 (reg_or_short_operand): Use u_short_cint_operand.
2247 Tue Jul 21 08:56:42 1998 Richard Henderson <rth@cygnus.com>
2249 * alpha.md (fix_truncdfsi2, fix_truncsfsi2): Remove the define_expands,
2250 but keep the insns and splits. Adjust so when the ultimate destination
2251 is memory, use cvtql.
2253 Tue Jul 21 08:55:09 1998 Richard Henderson <rth@cygnus.com>
2255 * flow.c (regno_uninitialized): Fixed regs are never uninitialized.
2257 Tue Jul 21 00:31:01 1998 Jeffrey A Law (law@cygnus.com)
2259 * gcc.c (do_spec): Call "error" not "warning".
2261 * configure.in: Fix minor problems with gas feature detection code.
2262 * configure: Rebuilt.
2264 * gcc.c (do_spec): Issue a warning for '%[]' usage.
2267 * gcc.c: Delete %[spec] support.
2268 (do_spec_1, case '('): Likewise.
2269 (do_spec_1, case '['): Call error.
2271 Mon Jul 20 22:34:17 1998 Richard Henderson <rth@cygnus.com>
2273 * alpha.h (CPP_SPEC): Tidy. Hook to cpp_cpu and cpp_subtarget.
2274 (CPP_SUBTARGET_SPEC): Default to empty string.
2275 (CPP_AM_*, CPP_IM_*, CPP_CPU_*, CPP_CPU_SPEC): New.
2276 (EXTRA_SPECS, SUBTARGET_EXTRA_SPECS): New.
2277 * alpha/elf.h (LD_SPEC): Use %(elf_dynamic_linker).
2278 * alpha/linux-elf.h (SUBTARGET_EXTRA_SPECS): New.
2280 * alpha/linux.h (CPP_PREDEFINES): Tidy.
2281 * alpha/netbsd-elf.h (SUBTARGET_EXTRA_SPECS): New.
2282 * alpha/netbsd.h (CPP_PREDEFINES): Tidy.
2283 * alpha/osf.h (CPP_PREDEFINES): Remove bits subsumed by CPP_CPU_SPEC.
2284 * alpha/win-nt.h (CPP_PREDEFINES): Likewise.
2285 * alpha/vsf.h (CPP_PREDEFINES): Likewise.
2286 (CPP_SUBTARGET_SPEC): New. Do this instead of overriding CPP_SPEC.
2287 * alpha/vxworks.h: Likewise.
2289 Mon Jul 20 22:51:57 1998 Ken Raeburn <raeburn@cygnus.com>
2291 * mips.md (reload_outsi): Added missing REGNO call.
2292 (smulsi3_highpart, umulsi3_highpart): Provide prototype for
2294 (mul_acc_di, mul_acc_64bit_di): Don't use match_op_dup, use
2295 another match_operator and compare the codes.
2297 * mips.h (MASK_DEBUG_E, MASK_DEBUG_I): Set to zero.
2299 * MIPS multiply pattern fixes:
2300 * mips.h (enum reg_class, REG_CLASS_NAMES, REG_CLASS_CONTENTS):
2301 Add union classes for HI, LO, or HILO plus general registers.
2302 (GENERATE_MADD): Deleted.
2303 * mips.md (mulsi3_mult3): Don't disparage output-LO alternative.
2304 Add TARGET_MAD to condition.
2305 (mulsi3): Test HAVE_mulsi3_mult3, not specific flags.
2306 (mul_acc_si): Expand GENERATE_MADD here; it's the only use. Use
2307 "*d" for accumulator, to give preference to LO initially but not
2310 Mon Jul 20 16:16:38 1998 Dave Brolley <brolley@cygnus.com>
2312 * configure.in (enable_c_mbchar): New configure option.
2313 (extra_cpp_objs): Always available now.
2315 * cexp.y (mbchar.h): #include it.
2316 (yylex): Handle Multibyte characters in character literals.
2318 * cccp.c (mbchar.h): #include it.
2319 (main): Set character set based on LANG environment variable.
2320 (rescan): Handle multibyte characters in comments.
2321 (skip_if_group): See above.
2322 (validate_else): See above.
2323 (skip_to_end_of_comment): See above.
2324 (macarg1): See above.
2325 (discard_comments): See above.
2326 (rescan): Handle multibyte characters in string and character literals.
2327 (collect_expansion): See above.
2328 (skip_quoted_string): See above.
2329 (macroexpand): See above.
2330 (macarg1): See above.
2331 (discard_comments): See above.
2332 (change_newlines): See above.
2334 * c-lex.c (mbchar.h): #include it.
2335 (GET_ENVIRONMENT): New macro.
2336 (init_lex): Set character set based on LANG environment variable.
2337 (yylex): Handle multibyte characters in character literals.
2338 (yylex): Handle multibyte characters in string literals.
2340 * Makefile.in (mbchar.o): New target.
2341 (cccp$(exeext)): @extra_cpp_objs@ is always available.
2342 (cppmain$(exeext)): @extra_cpp_objs@ is always available.
2344 * mbchar.[ch]: New files for multibyte character handling.
2346 Mon Jul 20 01:11:11 1998 David S. Miller <davem@pierdol.cobaltmicro.com>
2348 * jump.c (jump_optimize): When simplifying noop moves and
2349 PUSH_ROUNDING, fix thinko so we use same criterion for identifying
2350 the PUSHes to rewrite in second loop as we did in the first.
2352 Sun Jul 19 08:23:53 1998 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
2354 * cplus-dem.c (demangle_nested_args): Make function definition
2355 static to match the prototype.
2357 Fri Jul 17 14:58:44 1998 Richard Henderson <rth@cygnus.com>
2359 * alloca.c: Respect USE_C_ALLOCA.
2360 * gencheck.c (xmalloc): Ignore __GNUC__ for definition.
2361 * gengenrtl.c (xmalloc): Likewise.
2363 Fri Jul 17 14:18:14 1998 Richard Henderson <rth@cygnus.com>
2365 * loop.h (struct induction): Add no_const_addval.
2366 * loop.c (the_movables, reg_address_cost): New variables.
2367 (init_loop): Init reg_address_cost.
2368 (loop_optimize): Call end_alias_analysis.
2369 (scan_loop): Init the_movables.
2370 (record_giv): Init induction->no_const_addval.
2371 (basic_induction_var) [PLUS]: Use rtx_equal_p instead of ==.
2372 [REG]: Rearrange loop search test to catch more cases.
2373 (general_induction_var): Return success not benefit; take an extra
2374 argument for that. Change all callers.
2375 (simplify_giv_expr) [PLUS]: Always combine invariants. Use sge_plus.
2376 [MULT]: Use rtx_equal_p instead of ==. Combine simple invariants.
2377 [default]: Search the_movables for additional combinations.
2378 (sge_plus_constant, sge_plus): New functions.
2379 (express_from_1): New function.
2380 (express_from): Always define. Rewrite using express_from_1.
2381 (combine_givs_p): Handle more cases. Ignore address cost.
2382 (cmp_combine_givs_stats): New function.
2383 (combine_givs_used_once, combine_givs_benefit_from): New functions.
2384 (combine_givs): Rewrite to do best-fit combination.
2386 * fold-const.c (operand_equal_p): Handle RTL_EXPR.
2387 (fold): Do a complete (A*C)+(B*C) association check.
2389 Fri Jul 17 11:21:55 1998 Jim Wilson <wilson@cygnus.com>
2391 * function.c (fixup_var_refs_insns): Handle CLOBBER of a CONCAT.
2393 Fri Jul 17 11:48:55 1998 Jeffrey A Law (law@cygnus.com)
2395 * mn10300.c (MODES_TIEABLE_P): Fix typo.
2397 Fri Jul 17 03:26:12 1998 Rihcard Earnshaw (rearnsha@arm.com)
2399 * tree.c (valid_machine_attribute): Only create a new type variant if
2400 there is a decl to use it.
2402 Thu Jul 16 14:48:04 1998 Nick Clifton <nickc@cygnus.com>
2404 * gcc.c (do_spec_1): Cope with %g/%u/%U options which do not have
2407 Fri Jul 17 03:24:40 1998 Hans-Peter Nilsson <hp@axis.se>
2409 * extend.texi (Explicit Reg Vars): Typo: change "may deleted" into "may
2412 Thu Jul 16 14:48:47 1998 Jeffrey A Law (law@cygnus.com)
2414 * mn10300.c (count_tst_insns): New arg oreg_countp. Callers changed.
2415 Simplify tests for clearing an address register.
2416 (expand_prologue): Corresponding changes.
2418 * mn10300.md (movXX patterns): Make sure the destination is an
2419 ADDRESS_REG when substituting "zero_areg" for (const_int 0).
2420 (logical patterns): Split into expanders + patterns
2421 (zero and sign extension patterns): Similarly.
2422 (shift patterns): Similarly.
2424 Thu Jul 16 01:17:44 1998 Richard Henderson <rth@cygnus.com>
2426 * loop.c (emit_iv_add_mult): Scan the entire insn list generated
2427 for the sequence, recording base values.
2429 Wed Jul 15 10:49:55 1998 Richard Henderson <rth@cygnus.com>
2431 * i386.h (CPP_CPU_SPEC): Remove -Asystem(unix).
2433 Tue Jul 14 14:15:30 1998 Nick Clifton <nickc@cygnus.com>
2435 * gcc.c: Remove ANSI-C ism from --help code.
2437 * toplev.c: Support --help with USE_CPPLIB.
2439 Tue Jul 14 14:46:08 1998 Jeffrey A Law (law@cygnus.com)
2441 * configure.in: Rework gas feature code to work with symlink based
2444 * extend.texi: Clarify some issues related to local variables
2445 assigned to explicit registers.
2447 * mn10300.md (mulsi): Turn into expander + pattern.
2449 * mn10300.md (movsi, movsf, movdi, movdf): Remove "x" from I -> a
2452 Tue Jul 14 07:41:59 1998 Richard Earnshaw (rearnsha@arm.com)
2454 * arm/tcoff.h (USER_LABEL_PREFIX): Make it empty to match coff.h.
2456 Tue Jul 14 03:02:44 1998 Jeffrey A Law (law@cygnus.com)
2458 * version.c: Bump again to distinguish mainline tree from the
2461 See ChangeLog.0 for earlier changes.
2464 add-log-time-format: current-time-string