diff options
Diffstat (limited to 'gnu/usr.bin/gcc/f/ChangeLog')
| -rw-r--r-- | gnu/usr.bin/gcc/f/ChangeLog | 667 |
1 files changed, 667 insertions, 0 deletions
diff --git a/gnu/usr.bin/gcc/f/ChangeLog b/gnu/usr.bin/gcc/f/ChangeLog index 031133da65e..6e483dcadef 100644 --- a/gnu/usr.bin/gcc/f/ChangeLog +++ b/gnu/usr.bin/gcc/f/ChangeLog @@ -1,3 +1,374 @@ +Fri Feb 28 13:16:50 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * Version 0.5.20 released. + +Fri Feb 28 01:45:25 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * Make-lang.in (RUNTIMESTAGESTUFF, LIBU77STAGESTUFF): + Move some files incorrectly in the former to the latter, + and add another file or two to the latter. + + New meanings for (KIND=n), and new denotations in the + little language describing intrinsics: + * com.c (ffecom_init_0): Assign new meanings. + * intdoc.c: Document new meanings. + Support the new denotations. + * intrin.c: Employ new meanings, mapping them to internal + values (which are the same as they ever were for now). + Support the new denotations. + * intrin.def: Switch DEFIMP table to the new denotations. + + * intrin.c (ffeintrin_check_): Fix bug that was leaving + LOC() and %LOC() returning INTEGER*4 on systems where + it should return INTEGER*8. + + * type.c: Canonicalize function definitions, for etags + and such. + +Wed Feb 26 20:43:03 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * com.c (ffecom_init_0): Choose INTEGER(KIND=n) types, + where n is 2, 3, and 4, according to the new docs + instead of according to the old C correspondences + (which seem less useful at this point). + + * equiv.c (ffeequiv_destroy_): New function. + (ffeequiv_layout_local_): Use this new function + whenever the laying out of a local equivalence chain + is aborted for any reason. + Otherwise ensure that symbols no longer reference + the stale ffeequiv entries that result when they + are killed off in this procedure. + Also, the rooted symbol is one that has storage, + it really is irrelevant whether it has an equiv entry + at this point (though the code to remove the equiv + entry was put in at the end, just in case). + (ffeequiv_kill): When doing internal checks, make + sure the victim isn't named by any symbols it points + to. Not as complete a check as looking through the + entire symbol table (which does matter, since some + code in equiv.c used to remove symbols from the lists + for an ffeequiv victim but not remove that victim as the + symbol's equiv info), but this check did find some + real bugs in the code (that were fixed). + +Mon Feb 24 16:42:13 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * com.c (ffecom_expr_intrinsic_): Fix a couple of + warnings about uninitialized variables. + * intrin.c (ffeintrin_check_): Ditto, but there were + a couple of _real_ uninitialized-variable _bugs_ here! + (ffeintrin_fulfill_specific): Ditto, no real bug here. + +Sun Feb 23 15:01:20 1997 Craig Burley <burley@gnu.ai.mit.edu> + + Clean up diagnostics (especially about intrinsics): + * bad.def (FFEBAD_UNIMPL_STMT): Remove. + (FFEBAD_INTRINSIC_*, FFEBAD_NEED_INTRINSIC): Clean these + up so they're friendlier. + (FFEBAD_INTRINSIC_CMPAMBIG): New. + * intrin.c (ffeintrin_fulfill_generic, + ffeintrin_fulfill_specific, ffeintrin_is_intrinsic): + Always choose + generic or specific name text (which is for doc purposes + anyway) over implementation name text (which is for + internal use). + * intrin.def: Use more descriptive name texts for generics + and specifics in cases where the names themselves are not + enough (e.g. IDATE, which has two forms). + + Fix some intrinsic mappings: + * intrin.def (FFEINTRIN_specIDINT, FFEINTRIN_specAND, + FFEINTRIN_specDFLOAT, FFEINTRIN_specDREAL, FFEINTRIN_specOR, + FFEINTRIN_specXOR): Now have their own implementations, + instead of borrowing from others. + (FFEINTRIN_specAJMAX0, FFEINTRIN_specAJMIN0, FFEINTRIN_specBJTEST, + FFEINTRIN_specDFLOTJ, FFEINTRIN_specFLOATJ, FFEINTRIN_specJIABS, + FFEINTRIN_specJIAND, FFEINTRIN_specJIBCLR, FFEINTRIN_specJIBITS, + FFEINTRIN_specJIBSET, FFEINTRIN_specJIDIM, FFEINTRIN_specJIDINT, + FFEINTRIN_specJIDNNT, FFEINTRIN_specJIEOR, FFEINTRIN_specJIFIX, + FFEINTRIN_specJINT, FFEINTRIN_specJIOR, FFEINTRIN_specJISHFT, + FFEINTRIN_specJISHFTC, FFEINTRIN_specJISIN, FFEINTRIN_specJMAX0, + FFEINTRIN_specJMAX1, FFEINTRIN_specJMIN0, FFEINTRIN_specJMIN1, + FFEINTRIN_specJMOD, FFEINTRIN_specJNINT, FFEINTRIN_specJNOT,): + Turn these implementations off, since it's not clear + just what types they expect in the context of portable Fortran. + (DFLOAT): Now in FVZ family, since f2c supports them + + Support intrinsic inquiry functions (BIT_SIZE, LEN): + * intrin.c: Allow `i' in <arg_extra>. + * intrin.def (FFEINTRIN_impBIT_SIZE, FFEINTRIN_impLEN): + Mark args with `i'. + +Sat Feb 22 13:34:09 1997 Craig Burley <burley@gnu.ai.mit.edu> + + Only warn, don't error, for reference to unimplemented + intrinsic: + * bad.def (FFEBAD_INTRINSIC_UNIMPLW): Warning version + of _UNIMPL. + * intrin.c (ffeintrin_is_intrinsic): Use new warning + version of _UNIMPL (FFEBAD_INTRINSIC_UNIMPLW). + + Complain about REAL(Z) and AIMAG(Z) (Z is DOUBLE COMPLEX): + * bad.def (FFEBAD_INTRINSIC_CMPAMBIG): New diagnostic. + * expr.c: Needed #include "intrin.h" anyway. + (ffeexpr_token_intrincheck_): New function handles delayed + diagnostic for "REAL(REAL(expr)" if next token isn't ")". + (ffeexpr_token_arguments_): Do most of the actual checking here. + * intrin.h, intrin.c (ffeintrin_fulfill_specific): New + argument, check_intrin, to tell caller that intrin is REAL(Z) + or AIMAG(Z). All callers updated, mostly to pass NULL in + for this. + (ffeintrin_check_): Also has new arg check_intrin for same + purpose. All callers updated the same way. + * intrin.def (FFEINTRIN_impAIMAG): Change return type + from "R0" to "RC", to accommodate f2c (and perhaps other + non-F90 F77 compilers). + * top.h, top.c: New option -fugly-complex. + + New GNU intrinsics REALPART, IMAGPART, and COMPLEX: + * com.c (ffecom_expr_intrinsic_): Implement impCOMPLEX + and impREALPART here. (specIMAGPART => specAIMAG.) + * intrin.def: Add the intrinsics here. + + Rename implementations of VXTIDATE and VXTTIME to IDATEVXT + and TIMEVXT, so they sort more consistently: + * com.c (ffecom_expr_intrinsic_): + * intrin.def: + + Delete intrinsic group `dcp', add `gnu', etc.: + * intrin.c (ffeintrin_state_family): FFEINTRIN_familyGNU + replaces FFEINTRIN_familyDCP, and gets state from `gnu' + group. + Get rid of FFEINTRIN_familyF2Z, nobody needs it. + Move FFEINTRIN_specDCMPLX from DCP family to FVZ family, + as f2c has it. + Move FFEINTRIN_specDFLOAT from F2C family to FVZ family. + (FFEINTRIN_specZABS, FFEINTRIN_specZCOS, FFEINTRIN_specZEXP, + FFEINTRIN_specZLOG, FFEINTRIN_specZSIN, FFEINTRIN_specZSQRT): + Move these from F2Z family to F2C family. + * intrin.h (FFEINTRIN_familyF2Z, FFEINTRIN_familyDCP): Remove. + (FFEINTRIN_familyGNU): Add. + * top.h, top.c: Replace `dcp' with `gnu'. + + * com.c (ffecom_expr_intrinsic_): Clean up by collecting + simple conversions into one nice, conceptual place. + Fix up some intrinsic subroutines (MVBITS, KILL, UMASK) to + properly push and pop call temps, to avoid wasting temp + registers. + + * g77.c (doit): Toon says variables should be defined + before being referenced. Spoilsport. + + * intrin.c (ffeintrin_check_): Now Dave's worried about + warnings about uninitialized variables. Okay, so for + basic return values 'g' and 's', they _were_ + uninitialized -- is determinism really _that_ useful? + + * intrin.def (FFEINTRIN_impFGETC): Fix STATUS argument + so that it is INTENT(OUT) instead of INTENT(IN). + +1997-02-21 Dave Love <d.love@dl.ac.uk> + + * intrin.def, com.c: Support Sun-type `short' and `long' + intrinsics. Perhaps should also do Microcruft-style `int2'. + +Thu Feb 20 15:16:53 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * com.c (ffecom_expr_intrinsic_): Clean up indentation. + Support SECONDSUBR intrinsic implementation. + Rename SECOND to SECONDFUNC for direct support via library. + + * g77.c: Fix to return proper status value to shell, + by obtaining it from processes it spawns. + + * intdoc.c: Fix minor typo. + + * intrin.def: Turn SECOND into generic that maps into + function and subroutine forms. + + * intrin.def: Make FLOAT and SNGL into specific intrinsics. + + * intrin.def, intrin.h: Change the way DEFGEN and DEFSPEC + macros work, to save on verbage. + +Mon Feb 17 02:08:04 1997 Craig Burley <burley@gnu.ai.mit.edu> + + New subsystem to automatically generate documentation + on intrinsics: + * Make-lang.in ($(srcdir)/f/g77.info, + $(srcdir)/f/g77.dvi): Move g77 doc rules around. + Add to g77 doc rules the new subsystem. + (f77.mostlyclean, f77.maintainer-clean): Also clean up + after new doc subsystem. + * intdoc.c, intdoc.h: New doc subsystem code. + * intrin.h [FFEINTRIN_DOC]: When 1, don't pull in + stuff not needed by doc subsystem. + + Improve on intrinsics mechanism to both be more + self-documenting and to catch more user errors: + * intrin.c (ffeintrin_check_): Recognize new arg-len + and arg-rank information, and check it. + Move goto and signal indicators to the basic type. + Permit reference to arbitrary argument number, not + just first argument (for BESJN and BESYN). + (ffeintrin_init_0): Check and accept new notations. + * intrin.c, intrin.def: Value in COL now identifies + arguments starting with number 0 being the first. + + Some minor intrinsics cleanups (resulting from doc work): + * com.c (ffecom_expr_intrinsic_): Implement FLUSH + directly once again, handle its optional argument, + so it need not be a generic (awkward to handle in docs). + * intrin.def (BESJ0, BESJ1, BESJN, BESY0, BESY1, BESYN, + CHDIR, CHMOD, CTIME, DBESJ0, DBESJ1, DBESJN, DBESY0, + DBESY1, DBESYN, DDIM, ETIME, FGETC, FNUM, FPUTC, FSTAT, + GERROR, GETCWD, GETGID, GETLOG, GETPID, GETUID, GMTIME, + HOSTNM, IDATE, IERRNO, IIDINT, IRAND, ISATTY, ITIME, JIDINT, + LNBLNK, LSTAT, LTIME, MCLOCK, PERROR, SRAND, SYMLNK, TTYNAM, + UMASK): Change capitalization of initcaps (official) name + to be consistent with Burley's somewhat arbitrary rules. + (BESJN, BESYN): These have return arguments of same type + as their _second_ argument. + (FLUSH): Now a specific, not generic, intrinsic, with one + optional argument. + (FLUSH1): Eliminated. + Add arg-len and arg-rank info to several intrinsics. + (ITIME): Change argument type from REAL to INTEGER. + +Tue Feb 11 14:04:42 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * Make-lang.in (f771): Invocation of Makefile now done + with $(srcdir)=gcc to go along with $(VPATH)=gcc. + ($(srcdir)/f/runtime/configure, + $(srcdir)/f/runtime/libU77/configure): Break these out + so spurious triggers of this rule don't happen (as when + configure.in is more recent than libU77/configure). + (f77.rebuilt): Distinguish source versus build files, + so this target can be invoked from build directory and + still work. + * Makefile.in: This now expects $(srcdir) to be the gcc + source directory, not gcc/f, to agree with $(VPATH). + Accordingly, $(INCLUDES) has been fixed, various cruft + removed, the removal of f771 has been fixed to remove + the _real_ f771 (not the one in gcc's parent directory), + and so on. + + * lex.c: Part of ffelex_finish_statement_() now done + by new function ffelex_prepare_eos_(), so that, in one + popular case, the EOS can be prepared while the pointer + is at the end of the non-continued line instead of the + end of the line that marks no continuation. This improves + the appearance of diagnostics substantially. + +Mon Feb 10 12:44:06 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * Make-lang.in: runtime Makefile's, and include/f2c.h, + also depend on f/runtime/configure and f/runtime/libU77/configure. + + Fix various libU77 routines: + * com-rt.def (FFECOM_gfrtCTIME, FFECOM_gfrtMCLOCK, + FFECOM_gfrtTIME): These now use INTEGER*8 for time values, + for compatibility with systems like Alpha. + (FFECOM_gfrtSYSTEM_CLOCK, FFECOM_gfrtTTYNAM): Delete incorrect + trailing underscore in routine names. + * intrin.c, intrin.def: Support INTEGER*8 return values and + arguments ('4'). Change FFEINTRIN_impCTIME, FFEINTRIN_impMCLOCK, + and FFEINTRIN_impTIME accordingly. + (ffeintrin_is_intrinsic): Don't give caller a clue about + form of intrinsic -- shouldn't be needed at this point. + + Cope with generic intrinsics that are subroutines and functions: + * com.c (ffecom_finish_symbol_transform_, ffecom_expr_transform_): + Don't transform an intrinsic that is not known to be a subroutine + or a function. (Maybe someday have to avoid transforming + any intrinsic with an undecided or unknown implementation.) + * expr.c (ffeexpr_declare_unadorned_, + ffeexpr_declare_parenthesized_): Ok to invoke generic + intrinsic that has at least one subroutine form as a + subroutine. + Ok to pass intrinsic as actual arg if it has a known specific + intrinsic form that is valid as actual arg. + (ffeexpr_declare_parenthesized_): An unknown kind of + intrinsic has a paren_type chosen based on context. + (ffeexpr_token_arguments_): Build funcref/subrref based + on context, not on kind of procedure being called. + * intrin.h, intrin.c (ffeintrin_is_intrinsic): Undo changes of + Tue Feb 4 23:12:04 1997 by me, change all callers to leave + intrinsics as FFEINFO_kindNONE at this point. (Some callers + also had unused variables deleted as a result.) + + Enable all intrinsic groups (especially f90 and vxt): + * target.h (FFETARGET_defaultSTATE_DCP, FFETARGET_defaultSTATE_F2C, + FFETARGET_defaultSTATE_F90, FFETARGET_defaultSTATE_MIL, + FFETARGET_defaultSTATE_UNIX, FFETARGET_defaultSTATE_VXT): + Delete these macros, let top.c set them directly. + * top.c (ffeintrinsic_state_dcp_, ffe_intrinsic_state_f2c_, + ffe_intrinsic_state_f90_, ffe_intrinsic_state_mil_, + ffe_intrinsic_state_unix_, ffe_intrinsic_state_vxt_): + Enable all these directly. + +Sat Feb 8 03:21:50 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * g77.c: Incorporate recent changes to ../gcc.c. + For version magic (e.g. `g77 -v'), instead of compiling + /dev/null, write, compile, run, and then delete a small + program that prints the version numbers of the three + components of libf2c (libF77, libI77, and libU77), + so we get this info with bug reports. + Also, this change reduces the chances of accidentally + linking to an old (complex-alias-problem) libf2c. + Fix `-L' so the argument is expected in `-Larg'. + + * com.h (FFECOM_f2cLONGINT): For INTEGER*8 support in f2c.h, + dynamically determine proper type here, instead of + assuming `long long int' is correct. + +Tue Feb 4 23:12:04 1997 Craig Burley <burley@gnu.ai.mit.edu> + + Add libU77 library from Dave Love <d.love@dl.ac.uk>: + * Make-lang.in (f77-runtime): Depend on new Makefile. + (f/runtime/libU77/Makefile): New rule. + Also configure libU77. + ($(srcdir)/f/runtime/configure: Use Makefile.in, + so configuration doesn't have to have happened. + (f77.mostlyclean, f77.clean, f77.distclean, + f77.maintainer-clean): Some fixups here, but more work + needed. + (RUNTIMESTAGESTUFF): Add libU77's config.status. + (LIBU77STAGESTUFF, f77.stage1, f77.stage2, f77.stage3, + f77.stage4): New macro, appropriate uses added. + * com-rt.def: Add libU77 procedures. + * com.c (ffecom_f2c_ptr_to_integer_type_node, + ffecom_f2c_ptr_to_real_type_node): New type nodes. + (FFECOM_rttypeCHARACTER_): New type of run-time function. + (ffecom_char_args_): Handle CHARACTER*n intrinsics + where n != 1 here, instead of in ffecom_expr_intrinsic_. + (ffecom_expr_intrinsic_): New code to handle new + intrinsics. + In particular, change how FFEINTRIN_impFLUSH is handled. + (ffecom_make_gfrt_): Handle new type of run-time function. + (ffecom_init_0): Initialize new type nodes. + * config-lang.in: New libU77 directory. + * intrin.h, intrin.c (ffeintrin_is_intrinsic): Handle + potential generic for subroutine _and_ function + specifics via two new arguments. All callers changed. + Properly ignore deleted/disabled intrinsics in resolving + generics. + (ffeintrin_check_, ffeintrin_init_0): Handle CHARACTER intrinsics of (*) + length. + * intrin.def: Permission granted by FSF to place this in + public domain, which will allow it to serve as source + for both g77 program and its documentation. + Add libU77 intrinsics. + (FLUSH): Now a generic, not specific, intrinsic. + (DEFIMP): Now support return modifier for CHARACTER intrinsics. + + * com-rt.def (FFECOM_gfrtDIM, FFECOM_gfrtERF, + FFECOM_gfrtERFC, FFECOM_gfrtEXP, FFECOM_gfrtSIGN, + FFECOM_gfrtSIN, FFECOM_gfrtSINH, FFECOM_gfrtTAN, + FFECOM_gfrtTANH, FFECOM_gfrtPOW_RI): Change "&r" to "&f". + Sat Feb 1 12:15:09 1997 Craig Burley <burley@gnu.ai.mit.edu> * Version 0.5.19.1 released. @@ -12,6 +383,224 @@ Sat Feb 1 12:15:09 1997 Craig Burley <burley@gnu.ai.mit.edu> FFECOM_gfrtDIV_CC, FFECOM_gfrtDIV_ZZ: These all require result to _not_ overlap one or more inputs. +Sat Feb 1 00:25:55 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * com.c (ffecom_init_0): Do internal checks only if + -fset-g77-defaults not specified. + + Fix %LOC(), LOC() to return sufficiently wide type: + * com.h, com.c (ffecom_pointer_kind_, ffecom_label_kind_, + ffecom_pointer_kind(), ffecom_label_kind()): New globals + and accessor macros hold kind for integer pointers on target + machine. + (ffecom_init_0): Determine narrowest INTEGER type that + can hold a pointer (usually INTEGER*4 or INTEGER*8), + store it in ffecom_pointer_kind_, etc. + * expr.c (ffeexpr_cb_end_loc_): Use right type for %LOC(). + * intrin.c (ffeintrin_check_, ffeintrin_init_0): Support + new 'p' kind for type of intrinsic. + * intrin.def (FFEINTRIN_impLOC): Returns "Ip" instead of "I1", + so LOC() type is correct for target machine. + + Support -fugly-assign: + * lang-options.h, top.h, top.c (ffe_decode_option): + Accept -fugly-assign and -fno-ugly-assign. + * com.c (ffecom_expr_): Handle -fugly-assign. + * expr.c (ffeexpr_finished_): Check right type for ASSIGN + contexts. + +Fri Jan 31 14:30:00 1997 Craig Burley <burley@gnu.ai.mit.edu> + + Remove last vestiges of -fvxt-not-f90: + * stb.c (ffestb_R10012_, ffestb_R10014_, ffestb_V0201_): + top.c, top.h: + +Fri Jan 31 02:13:54 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * top.c (ffe_decode_option): Warn if -fugly is specified, + it'll go away soon. + + * symbol.h: No need to #include "bad.h". + + Reorganize features from -fvxt-not-f90 to -fvxt: + * lang-options.h, top.h, top.c: + Accept -fvxt and -fno-vxt, but not -fvxt-not-f90 or -ff90-not-vxt. + Warn if the latter two are used. + * expr.c (ffeexpr_nil_rhs_): Double-quote means octal constant. + (ffeexpr_token_rhs_): Double-quote means octal constant. + * target.h (FFETARGET_defaultIS_VXT_NOT_90): Delete macro + definition, no longer needed. + + Make some -ff90 features the default: + * data.c (ffedata_value): DATA implies SAVE. + * src.h (ffesrc_is_name_noninit): Underscores always okay. + + Fix up some more #error directives by quoting their text: + * bld.c (ffebld_constant_is_zero): + * target.h: + +Sat Jan 18 18:22:09 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * g77.c (lookup_option, main): Recognize `-Xlinker', + `-Wl,', `-l', `-L', `--library-directory', `-o', + `--output'. + (lookup_option): Don't depend on SWITCH_TAKES_ARG + being correct, it might or might not have `-x' in + it depending on host. + Return NULL argument if it would be an empty string. + (main): If no input files (by gcc.c's definition) + but `-o' or `--output' specified, produce diagnostic + to avoid overwriting output via gcc. + Recognize C++ `+e' options. + Treat -L as another non-magical option (like -B). + Don't append_arg `-x' twice. + +Fri Jan 10 23:36:00 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * top.c [BUILT_FOR_270] (ffe_decode_option): Make + -fargument-noalias-global the default. + +Fri Jan 10 07:42:27 1997 Craig Burley <burley@gnu.ai.mit.edu> + + Enable inlining of previously-compiled program units: + * com.c (ffecom_do_entry_, ffecom_start_progunit_): + Register new public function in ffeglobal database. + (ffecom_sym_transform_): Any GLOBAL or potentially GLOBAL + symbol should be looked up in ffeglobal database and + that tree node used, if found. That way, gcc knows + the references are to those earlier definitions, so it + can emit shorter branches/calls, inline, etc. + (ffecom_transform_common_): Minor change for clarity. + * expr.c (ffeexpr_sym-lhs_call_, ffeexpr_sym_lhs_extfunc_, + ffeexpr_sym_rhs_actualarg_, ffeexpr_paren_rhs_let_, + ffeexpr_token_funsubstr_): Globalize symbol as needed. + * global.c (ffeglobal_promoted): New function to look up + existing local symbol in ffeglobal database. + * global.h: Declare new function. + * name.h (ffename_token): New macro, plus alphabetize. + * stc.c (ffestc_R1207_item): Globalize EXTERNAL symbol. + * stu.c (ffestu_sym_end_transition, ffestu_sym_exec_transition): + Globalize symbol as needed. + * symbol.h, symbol.c (ffesymbol_globalize): New function. + +Thu Jan 9 14:20:00 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * ste.c (ffeste_R809): Produce a diagnostic for SELECT CASE + on CHARACTER type, instead of crashing. + +Thu Jan 9 00:52:45 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * stc.c (ffestc_order_entry_, ffestc_order_format_, + ffestc_R1226): Allow ENTRY and FORMAT before IMPLICIT + NONE, by having them transition only to state 1 instead + of state 2 (which is disallowed by IMPLICIT NONE). + +Mon Jan 6 22:44:53 1997 Craig Burley <burley@gnu.ai.mit.edu> + + Fix AXP bug found by Rick Niles (961201-1.f): + * com.c (ffecom_init_0): Undo my 1996-05-14 change, as + it is incorrect and prevented easily finding this bug. + * target.h [__alpha__] (ffetargetReal1, ffetargetReal2): + Use int instead of long. + (ffetarget_cvt_r1_to_rv_, ffetarget_cvt_rv_to_r1_, + ffetarget_cvt_r2_to_rv_, ffetarget_cvt_rv_to_r2_): + New functions that intercede for callers of + REAL_VALUE_(TO|UNTO)_TARGET_(SINGLE|DOUBLE). + All callers changed, and damaging casts to (long *) removed. + +Sun Jan 5 03:26:11 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * Make-lang.in (g77, g77-cross): Depend on both g77.c and + zzz.c, in $(srcdir)/f/. + + Better design for -fugly-assumed: + * stc.c (ffestc_R501_item, ffestc_R524_item, + ffestc_R547_item_object): Pass new is_ugly_assumed flag. + * stt.c, stt.h (ffestt_dimlist_as_expr, + ffestt_dimlist_type): New is_ugly_assumed flag now + controls whether "1" is treated as "*". + Don't treat "2-1" or other collapsed constants as "*". + +Sat Jan 4 15:26:22 1997 Craig Burley <burley@gnu.ai.mit.edu> + + * stb.c (ffestb_R10012_): Don't confirm on FORMAT(A,) + or even FORMAT(A,,B), as R1229 only warns about the + former currently, and this seems reasonable. + + Improvements to diagnostics: + * sta.c (ffesta_second_): Don't add any ffestb parsers + unless they're specifically called for. + Set up ffesta_tokens[0] before calling ffestc_exec_transition, + else stale info might get used. + (ffesta_save_): Do a better job picking which parser to run + after running all parsers with no confirmed possibles. + (FFESTA_maxPOSSIBLES_): Decrease from 100 now that so few + possibles are ever on the list at a given time. + (struct _ffesta_possible): Add named attribute. + (ffesta_add_possible_exec_, ffesta_add_possible_nonexec_): + Make these into macros that call a single function that now + sets the named attribute. + (ffesta_add_possible_unnamed_exec_, + ffeseta_add_possible_unnamed_nonexec_): New macros. + (ffesta_second_): Designate unnamed possibles as + appropriate. + * stb.c (ffestb_R1229, ffestb_R12291_): Use more general + diagnostic, so things like "POINTER (FOO, BAR)" are + diagnosed as unrecognized statements, not invalid statement + functions. + * stb.h, stb.c (ffestb_unimplemented): Remove function. + +1996-12-30 Dave Love <d.love@dl.ac.uk> + + * com.c: #include libU77/config.h + (ffecom_f2c_ptr_to_integer_type_node, + ffecom_f2c_ptr_to_integer_type_node): New variables. + (ffecom_init_0): Use them. + (ffecom_expr_intrinsic_): Many news cases for libU77 intrinsics. + + * com-rt.def: New definitions for libU77. + * intrin.def: Likewise. Also correct ftell arg spec. + + * Makefile.in (f/runtime/libU77/config.h): New target for com.c + dependency. + * Make-lang.in (f771): Depend on f/runtime/Makefile for the above. + +Sat Dec 28 12:28:29 1996 Craig Burley <burley@gnu.ai.mit.edu> + + * stt.c (ffestt_dimlist_type): Treat ([...,]1) in dimlist + as ([...,]*) if -fugly-assumed, so assumed-size array + detected early enough. + +Thu Dec 19 14:01:57 1996 Craig Burley <burley@gnu.ai.mit.edu> + + * target.h (FFETARGET_REAL_VALUE_FROM_INT_): Conditionalize + definition on BUILT_FOR_280, not BUILT_WITH_280, since + the name of the macro was (properly) changed since 0.5.19. + + Fix warnings/errors resulting from ffetargetOffset becoming + `long long int' instead of `unsigned long' as of 0.5.19, + while ffebitCount remains `unsigned long': + * bld.c (ffebld_constantarray_dump): Avoid warnings by + using loop var of appropriate type, and using casts. + * com.c (ffecom_expr_): Use right type for loop var. + (ffecom_sym_transform_, ffecom_transform_equiv_): + Cast to right type in assertions. + * data.c (ffedata_gather_, ffedata_value_): Cast to right + type in assertions and comparisons. + +Wed Dec 18 12:07:11 1996 Craig Burley <burley@gnu.ai.mit.edu> + + Patch from Alexandre Oliva <oliva@dcc.unicamp.br>: + * Makefile.in (all.indirect): Don't pass -bbigtoc option + to GNU ld. + + Cope with new versions of gcc: + * com.h (BUILT_FOR_280): New macro. + * com.c (ffecom_ptr_to_expr): Conditionalize test of + OFFSET_REF. + (ffecom_build_complex_constant_): Conditionalize calling + sequence for build_complex. + Sat Dec 7 07:15:17 1996 Craig Burley <burley@gnu.ai.mit.edu> * Version 0.5.19 released. @@ -200,6 +789,11 @@ Thu Oct 31 20:56:28 1996 Craig Burley <burley@gnu.ai.mit.edu> if pre-2.7.0 gcc is involved (since our patch for that doesn't add support for tooning). +Sat Oct 26 05:56:51 1996 Craig Burley <burley@gnu.ai.mit.edu> + + * bad.def (FFEBAD_TYPELESS_TOO_LARGE): Remove this + unused and redundant diagnostic. + Sat Oct 26 00:45:42 1996 Craig Burley <burley@gnu.ai.mit.edu> * target.c (ffetarget_integerhex): Fix dumb bug. @@ -246,6 +840,12 @@ Sun Oct 13 22:15:23 1996 Craig Burley <burley@gnu.ai.mit.edu> * top.c (ffe_decode_option): Don't set back-end flags that are nonexistent prior to gcc 2.7.0. +Sun Oct 13 12:48:45 1996 Craig Burley <burley@gnu.ai.mit.edu> + + * com.c (convert): Don't convert emulated complex expr to + real (via REALPART_EXPR) if the target type is (emulated) + complex. + Wed Oct 2 21:57:12 1996 Craig Burley <burley@gnu.ai.mit.edu> * com.c (ffecom_debug_kludge_): Set DECL_IN_SYSTEM_HEADER so @@ -299,6 +899,62 @@ Thu Sep 26 00:18:40 1996 Craig Burley <burley@gnu.ai.mit.edu> Depend on stmp-hdrs, not stmp-int-hdrs, since libF77 needs float.h. +Sat Jun 22 18:17:11 1996 Craig Burley <burley@gnu.ai.mit.edu> + + * com.c (ffecom_tree_divide_): Fix RECORD_TYPE case to + look at type of first field, properly, to determine + whether to call c_div or z_div. + +Tue Jun 4 04:27:18 1996 Craig Burley <burley@gnu.ai.mit.edu> + + * com.c (ffecom_build_complex_constant_): Explicitly specify + TREE_PURPOSE. + (ffecom_expr_): Fix thinko. + (ffecom_2): For COMPLEX_EXPR, explicitly specify TREE_PURPOSE. + +Mon May 27 16:23:43 1996 Craig Burley <burley@gnu.ai.mit.edu> + + Changes to optionally avoid gcc's back-end complex support: + * com.c (ffecom_stabilize_aggregate_): New function. + (ffecom_convert_to_complex_): New function. + (ffecom_make_complex_type_): New function. + (ffecom_build_complex_constant_): New function. + (ffecom_expr_): For opCONVERT of non-COMPLEX to COMPLEX, + don't bother explicitly converting to the subtype first, + because gcc does that anyway, and more code would have + to be added to find the subtype for the emulated-complex + case. + (ffecom_f2c_make_type_): Use ffecom_make_complex_type_ + instead of make_node etc. to make a complex type. + (ffecom_1, ffecom_2): Translate operations on COMPLEX operands + to appropriate operations when emulating complex. + (ffecom_constantunion): Use ffecom_build_complex_constant_ + instead of build_complex to build a complex constant. + (ffecom_init_0): Change point at which types are laid out + for improved consistency. + Use ffecom_make_complex_type_ instead of make_node etc. + to make a complex type. + Always calculate storage sizes from TYPE_SIZE, never TYPE_PRECISION. + (convert): Use e, not expr, since we've copied into that anyway. + For RECORD_TYPE cases, do emulated-complex conversions. + (ffecom_f2c_set_lio_code_): Always calculate storage sizes + from TYPE_SIZE, never TYPE_PRECISION. + (ffecom_tree_divide_): Allow RECORD_TYPE to also be handled + by run-time library. + (ffecom_expr_intrinsic_): Handle possible RECORD_TYPE as argument + to AIMAG intrinsic. + + * top.h, top.c, lang-options.h: Support new -f(no-)emulate-complex option. + + * com.c (ffecom_sym_transform_): Clarify and fix typos in comments. + +Mon May 20 02:06:27 1996 Craig Burley <burley@gnu.ai.mit.edu> + + * target.h: Use new REAL_VALUE_UNTO_TARGET_* macros instead + of REAL_VALUE_FROM_TARGET_DOUBLE and _SINGLE. + Explicitly use long instead of HOST_WIDE_INT for emulation + of ffetargetReal1 and ffetargetReal2. + 1996-05-20 Dave Love <d.love@dl.ac.uk> * config-lang.in: @@ -317,6 +973,17 @@ Thu Sep 26 00:18:40 1996 Craig Burley <burley@gnu.ai.mit.edu> * g77.c (DEFAULT_SWITCH_TAKES_ARG): Define a la gcc.c in the current GCC snapshot. +Tue May 14 00:24:07 1996 Craig Burley <burley@gnu.ai.mit.edu> + + Changes for DEC Alpha AXP support: + * com.c (ffecom_init_0): REAL_ARITHMETIC means internal + REAL/DOUBLE PRECISION might well have a different size + than the compiled type, so don't crash if this is the + case. + * target.h: Use `int' for ffetargetInteger1, + ffetargetLogical1, and magical tests. Set _f format + strings accordingly. + Tue Apr 16 14:08:28 1996 Craig Burley <burley@gnu.ai.mit.edu> * top.c (ffe_decode_option): -Wall no longer implies |