1 %************************************************************************
3 \section[Driver-asm-fiddling]{Fiddling with assembler files}
5 %************************************************************************
10 Utterly stomp out C functions' prologues and epilogues; i.e., the
11 stuff to do with the C stack.
13 Any other required tidying up.
19 The HP linker is very picky about symbols being in the appropriate
20 space (code vs. data). When we mangle the threaded code to put the
21 info tables just prior to the code, they wind up in code space
22 rather than data space. This means that references to *_info from
23 un-mangled parts of the RTS (e.g. unthreaded GC code) get
24 unresolved symbols. Solution: mini-mangler for .c files on HP. I
25 think this should really be triggered in the driver by a new -rts
26 option, so that user code doesn't get mangled inappropriately.
28 With reversed tables, jumps are to the _info label rather than to
29 the _entry label. The _info label is just an address in code
30 space, rather than an entry point with the descriptive blob we
31 talked about yesterday. As a result, you can't use the call-style
32 JMP_ macro. However, some JMP_ macros take _info labels as targets
33 and some take code entry points within the RTS. The latter won't
34 work with the goto-style JMP_ macro. Sigh. Solution: Use the goto
35 style JMP_ macro, and mangle some more assembly, changing all
36 "RP'literal" and "LP'literal" references to "R'literal" and
37 "L'literal," so that you get the real address of the code, rather
38 than the descriptive blob. Also change all ".word P%literal"
39 entries in info tables and vector tables to just ".word literal,"
40 for the same reason. Advantage: No more ridiculous call sequences.
43 %************************************************************************
45 \subsection{Constants for various architectures}
47 %************************************************************************
50 sub init_TARGET_STUFF {
52 #--------------------------------------------------------#
53 if ( $TargetPlatform =~ /^alpha-.*-.*/ ) {
55 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
56 $T_US = ''; # _ if symbols have an underscore on the front
57 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
58 $T_CONST_LBL = '^\$C(\d+):$'; # regexp for what such a lbl looks like
61 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\#.*|\.(file|loc)\s+\S+\s+\S+|\.text|\.r?data)\n)';
62 $T_COPY_DIRVS = '^\s*(\#|\.(file|globl|ent|loc))';
64 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
65 $T_DOT_WORD = '\.quad';
66 $T_DOT_GLOBAL = '^\t\.globl';
67 $T_HDR_literal = "\.rdata\n\t\.align 3\n";
68 $T_HDR_misc = "\.text\n\t\.align 3\n";
69 $T_HDR_data = "\.data\n\t\.align 3\n";
70 $T_HDR_consist = "\.text\n";
71 $T_HDR_closure = "\.data\n\t\.align 3\n";
72 $T_HDR_srt = "\.data\n\t\.align 3\n";
73 $T_HDR_info = "\.text\n\t\.align 3\n";
74 $T_HDR_entry = "\.text\n\t\.align 3\n";
75 $T_HDR_fast = "\.text\n\t\.align 3\n";
76 $T_HDR_vector = "\.text\n\t\.align 3\n";
77 $T_HDR_direct = "\.text\n\t\.align 3\n";
79 #--------------------------------------------------------#
80 } elsif ( $TargetPlatform =~ /^hppa/ ) {
82 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
83 $T_US = ''; # _ if symbols have an underscore on the front
84 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
85 $T_CONST_LBL = '^L\$C(\d+)$'; # regexp for what such a lbl looks like
88 $T_MOVE_DIRVS = '^((\s+\.(IMPORT|EXPORT|PARAM).*|\s+\.align\s+\d+|\s+\.(SPACE|SUBSPA)\s+\S+|\s*)\n)';
89 $T_COPY_DIRVS = '^\s+\.(IMPORT|EXPORT)';
91 $T_hsc_cc_PAT = '\.STRING.*\)(hsc|cc) (.*)\\\\x09(.*)\\\\x00"';
92 $T_DOT_WORD = '\.word';
93 $T_DOT_GLOBAL = '^\s+\.EXPORT';
94 $T_HDR_literal = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
95 $T_HDR_misc = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
96 $T_HDR_data = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
97 $T_HDR_consist = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
98 $T_HDR_closure = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
99 $T_HDR_srt = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
100 $T_HDR_info = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
101 $T_HDR_entry = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
102 $T_HDR_fast = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
103 $T_HDR_vector = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
104 $T_HDR_direct = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
106 #--------------------------------------------------------#
107 } elsif ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd2|nextstep3|cygwin32|mingw32)$/ ) {
108 # NeXT added but not tested. CaS
110 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
111 $T_US = '_'; # _ if symbols have an underscore on the front
112 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
113 $T_CONST_LBL = '^LC(\d+):$';
115 $T_X86_PRE_LLBL_PAT = 'L';
116 $T_X86_PRE_LLBL = 'L';
117 $T_X86_BADJMP = '^\tjmp [^L\*]';
119 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,0x90)?|\.globl\s+\S+|\.text|\.data|\.stab[^n].*|\.type\s+.*|\.size\s+.*)\n)';
120 $T_COPY_DIRVS = '\.(globl|stab)';
121 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
122 $T_DOT_WORD = '\.(long|word|value|byte|space)';
123 $T_DOT_GLOBAL = '\.globl';
124 $T_HDR_literal = "\.text\n\t\.align 2\n";
125 $T_HDR_misc = "\.text\n\t\.align 2,0x90\n";
126 $T_HDR_data = "\.data\n\t\.align 2\n";
127 $T_HDR_consist = "\.text\n";
128 $T_HDR_closure = "\.data\n\t\.align 2\n";
129 $T_HDR_closure = "\.data\n\t\.align 2\n\t.long 0\n" if ( $TargetPlatform =~ /.*-mingw32$/ );
130 $T_HDR_srt = "\.data\n\t\.align 2\n";
131 $T_HDR_info = "\.text\n\t\.align 2\n"; # NB: requires padding
132 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
133 $T_HDR_fast = "\.text\n\t\.align 2,0x90\n";
134 $T_HDR_vector = "\.text\n\t\.align 2\n"; # NB: requires padding
135 $T_HDR_direct = "\.text\n\t\.align 2,0x90\n";
137 #--------------------------------------------------------#
138 } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|freebsd3)$/ ) {
140 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
141 $T_US = ''; # _ if symbols have an underscore on the front
142 $T_PRE_APP = # regexp that says what comes before APP/NO_APP
143 ($TargetPlatform =~ /-(linux|freebsd3)$/) ? '#' : '/' ;
144 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
146 $T_X86_PRE_LLBL_PAT = '\.L';
147 $T_X86_PRE_LLBL = '.L';
148 $T_X86_BADJMP = '^\tjmp [^\.\*]';
150 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,0x90)?|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.type\s+.*|\.Lfe.*\n\t\.size\s+.*|\.size\s+.*|\.ident.*)\n)';
151 $T_COPY_DIRVS = '\.(globl)';
153 if ( $TargetPlatform =~ /freebsd3/ ) {
154 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
156 $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
159 $T_DOT_WORD = '\.(long|value|byte|zero)';
160 $T_DOT_GLOBAL = '\.globl';
161 $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11)
162 $T_HDR_misc = "\.text\n\t\.align 4\n";
163 $T_HDR_data = "\.data\n\t\.align 4\n"; # ToDo: change align??
164 $T_HDR_consist = "\.text\n";
165 $T_HDR_closure = "\.data\n\t\.align 4\n"; # ToDo: change align?
166 $T_HDR_srt = "\.data\n\t\.align 4\n"; # ToDo: change align?
167 $T_HDR_info = "\.text\n\t\.align 16\n"; # NB: requires padding
168 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
169 $T_HDR_fast = "\.text\n\t\.align 16\n";
170 $T_HDR_vector = "\.text\n\t\.align 16\n"; # NB: requires padding
171 $T_HDR_direct = "\.text\n\t\.align 16\n";
173 #--------------------------------------------------------#
174 } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
176 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
177 $T_US = '_'; # _ if symbols have an underscore on the front
178 $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
179 $T_CONST_LBL = '^LC(\d+):$';
182 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
183 $T_COPY_DIRVS = '\.(globl|proc|stab)';
184 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
186 $T_DOT_WORD = '\.long';
187 $T_DOT_GLOBAL = '\.globl';
188 $T_HDR_literal = "\.text\n\t\.even\n";
189 $T_HDR_misc = "\.text\n\t\.even\n";
190 $T_HDR_data = "\.data\n\t\.even\n";
191 $T_HDR_consist = "\.text\n";
192 $T_HDR_closure = "\.data\n\t\.even\n";
193 $T_HDR_srt = "\.data\n\t\.even\n";
194 $T_HDR_info = "\.text\n\t\.even\n";
195 $T_HDR_entry = "\.text\n\t\.even\n";
196 $T_HDR_fast = "\.text\n\t\.even\n";
197 $T_HDR_vector = "\.text\n\t\.even\n";
198 $T_HDR_direct = "\.text\n\t\.even\n";
200 #--------------------------------------------------------#
201 } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
203 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
204 $T_US = ''; # _ if symbols have an underscore on the front
205 $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
206 $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
209 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
210 $T_COPY_DIRVS = '\.(globl|ent)';
212 $T_hsc_cc_PAT = 'I WAS TOO LAZY TO DO THIS BIT (WDP 95/05)';
213 $T_DOT_WORD = '\.word';
214 $T_DOT_GLOBAL = '^\t\.globl';
215 $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
216 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
217 $T_HDR_data = "\t\.data\n\t\.align 2\n";
218 $T_HDR_consist = 'TOO LAZY TO DO THIS TOO';
219 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
220 $T_HDR_srt = "\t\.data\n\t\.align 2\n";
221 $T_HDR_info = "\t\.text\n\t\.align 2\n";
222 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
223 $T_HDR_fast = "\t\.text\n\t\.align 2\n";
224 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
225 $T_HDR_direct = "\t\.text\n\t\.align 2\n";
227 #--------------------------------------------------------#
228 } elsif ( $TargetPlatform =~ /^powerpc-.*|^rs6000-.*/ ) {
230 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
231 $T_US = ''; # _ if symbols have an underscore on the front
232 $T_PRE_APP = 'NOT APPLICABLE'; # regexp that says what comes before APP/NO_APP
233 $T_CONST_LBL = 'NOT APPLICABLE'; # regexp for what such a lbl looks like
236 $T_MOVE_DIRVS = '^(\s*(\.toc|\.align \d+|\.csect \S+|\t\.?l?globl \S+)\n)';
237 $T_COPY_DIRVS = '\.(l?globl)';
239 $T_hsc_cc_PAT = '\.byte.*\)(hsc|cc) (.*)"\n\t\.byte \d+\n\t\.byte "(.*)"\n\t\.byte \d+';
240 $T_DOT_WORD = '\.long';
241 $T_DOT_GLOBAL = '\.globl';
242 $T_HDR_toc = "\.toc\n";
243 $T_HDR_literal = "\.csect .data[RW]\n\t\.align 2\n"; #not RO!?
244 $T_HDR_misc = "# misc\n\.csect \.text[PR]\n\t\.align 2\n";
245 $T_HDR_data = "# data\n\.csect \.data[RW]\n\t\.align 2\n";
246 $T_HDR_consist = "# consist\n\.csect \.data[RW]\n\t\.align 2\n";
247 $T_HDR_closure = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
248 $T_HDR_srt = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
249 $T_HDR_info = "# info\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
250 $T_HDR_entry = "# entry\n\.csect \.text[PR]\n\t\.align 2\n";
251 $T_HDR_fast = "# fast\n\.csect \.text[PR]\n\t\.align 2\n";
252 $T_HDR_vector = "# vector\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
253 $T_HDR_direct = "# direct\n";
255 #--------------------------------------------------------#
256 } elsif ( $TargetPlatform =~ /^sparc-.*-solaris2/ ) {
258 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
259 $T_US = ''; # _ if symbols have an underscore on the front
260 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
261 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
264 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\.section.*|\s+\.type.*|\s+\.size.*)\n)';
265 $T_COPY_DIRVS = '\.(global|proc|stab)';
267 $T_hsc_cc_PAT = '\.asciz.*\)(hsc|cc) (.*)\\\\t(.*)"';
268 $T_DOT_WORD = '\.(word|byte|half|skip)';
269 $T_DOT_GLOBAL = '^\t\.global';
270 $T_HDR_literal = "\.text\n\t\.align 8\n";
271 $T_HDR_misc = "\.text\n\t\.align 4\n";
272 $T_HDR_data = "\.data\n\t\.align 8\n";
273 $T_HDR_consist = "\.text\n";
274 $T_HDR_closure = "\.data\n\t\.align 4\n";
275 $T_HDR_srt = "\.data\n\t\.align 4\n";
276 $T_HDR_info = "\.text\n\t\.align 4\n";
277 $T_HDR_entry = "\.text\n\t\.align 4\n";
278 $T_HDR_fast = "\.text\n\t\.align 4\n";
279 $T_HDR_vector = "\.text\n\t\.align 4\n";
280 $T_HDR_direct = "\.text\n\t\.align 4\n";
282 #--------------------------------------------------------#
283 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
285 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
286 $T_US = '_'; # _ if symbols have an underscore on the front
287 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
288 $T_CONST_LBL = '^LC(\d+):$';
291 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
292 $T_COPY_DIRVS = '\.(global|proc|stab)';
293 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
295 $T_DOT_WORD = '\.word';
296 $T_DOT_GLOBAL = '^\t\.global';
297 $T_HDR_literal = "\.text\n\t\.align 8\n";
298 $T_HDR_misc = "\.text\n\t\.align 4\n";
299 $T_HDR_data = "\.data\n\t\.align 8\n";
300 $T_HDR_consist = "\.text\n";
301 $T_HDR_closure = "\.data\n\t\.align 4\n";
302 $T_HDR_srt = "\.data\n\t\.align 4\n";
303 $T_HDR_info = "\.text\n\t\.align 4\n";
304 $T_HDR_entry = "\.text\n\t\.align 4\n";
305 $T_HDR_fast = "\.text\n\t\.align 4\n";
306 $T_HDR_vector = "\.text\n\t\.align 4\n";
307 $T_HDR_direct = "\.text\n\t\.align 4\n";
309 #--------------------------------------------------------#
311 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
316 print STDERR "T_STABBY: $T_STABBY\n";
317 print STDERR "T_US: $T_US\n";
318 print STDERR "T_PRE_APP: $T_PRE_APP\n";
319 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
320 print STDERR "T_POST_LBL: $T_POST_LBL\n";
321 if ( $TargetPlatform =~ /^i386-/ ) {
322 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
323 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
324 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
326 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
327 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
328 print STDERR "T_hsc_cc_PAT: $T_hsc_cc_PAT\n";
329 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
330 print STDERR "T_HDR_literal: $T_HDR_literal\n";
331 print STDERR "T_HDR_misc: $T_HDR_misc\n";
332 print STDERR "T_HDR_data: $T_HDR_data\n";
333 print STDERR "T_HDR_consist: $T_HDR_consist\n";
334 print STDERR "T_HDR_closure: $T_HDR_closure\n";
335 print STDERR "T_HDR_info: $T_HDR_info\n";
336 print STDERR "T_HDR_entry: $T_HDR_entry\n";
337 print STDERR "T_HDR_fast: $T_HDR_fast\n";
338 print STDERR "T_HDR_vector: $T_HDR_vector\n";
339 print STDERR "T_HDR_direct: $T_HDR_direct\n";
345 %************************************************************************
347 \subsection{Mangle away}
349 %************************************************************************
353 local($in_asmf, $out_asmf) = @_;
355 # multi-line regexp matching:
360 &init_TARGET_STUFF();
361 &init_FUNNY_THINGS();
363 # perl4 on alphas SEGVs when give ${foo} substitutions in patterns.
364 # To avoid them we declare some locals that allows to avoid using curlies.
365 local($TUS) = ${T_US};
366 local($TPOSTLBL) = ${T_POST_LBL};
367 local($TMOVEDIRVS) = ${T_MOVE_DIRVS};
368 local($TPREAPP) = ${T_PRE_APP};
369 local($TCOPYDIRVS) = ${T_COPY_DIRVS};
370 local($TDOTWORD) = ${T_DOT_WORD};
372 open(INASM, "< $in_asmf")
373 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
374 open(OUTASM,"> $out_asmf")
375 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
377 # read whole file, divide into "chunks":
378 # record some info about what we've found...
380 @chk = (); # contents of the chunk
381 $numchks = 0; # number of them
382 @chkcat = (); # what category of thing in each chunk
383 @chksymb = (); # what symbol(base) is defined in this chunk
384 %slowchk = (); # ditto, its regular "slow" entry code
385 %fastchk = (); # ditto, fast entry code
386 %closurechk = (); # ditto, the (static) closure
387 %srtchk = (); # ditto, its SRT (for top-level things)
388 %infochk = (); # given a symbol base, say what chunk its info tbl is in
389 %vectorchk = (); # ditto, return vector table
390 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
392 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
395 next if $T_STABBY && /^\.stab.*$TUS[@]?__stg_split_marker/o;
396 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
397 next if /^\t\.def.*endef$/;
398 next if /$TPREAPP(NO_)?APP/o;
399 next if /^;/ && $TargetPlatform =~ /^hppa/;
401 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|powerpc|rs6000)-/;
403 last if /^_section_\.text:$/ && $TargetPlatform =~ /^powerpc-|^rs6000-/;
405 if ( $TargetPlatform =~ /^mips-/
406 && /^\t\.(globl \S+ \.text|comm\t)/ ) {
407 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
409 # As a temporary solution for compiling "foreign export" declarations,
410 # we use global variables to pass arguments from C to STG land.
411 # These declarations live in the .hc file and not in the generated C
412 # stub file, so we let them pass through here.
413 } elsif ( /^\t\.comm\t__fexp_.*$/ ) {
415 $chkcat[$i] = 'data';
418 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
419 # duplicated from the bottom
423 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
424 # Alphas: Local labels not to be confused with new chunks
427 # NB: all the rest start with a non-space
429 } elsif ( $TargetPlatform =~ /^mips-/
430 && /^\d+:/ ) { # a funny-looking very-local label
433 } elsif ( /$T_CONST_LBL/o ) {
435 $chkcat[$i] = 'literal';
438 } elsif ( /^$TUS[@]?__stg_split_marker(\d+)$TPOSTLBL[@]?$/o ) {
440 $chkcat[$i] = 'splitmarker';
443 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_info$TPOSTLBL[@]?$/o ) {
446 $chkcat[$i] = 'infotbl';
447 $chksymb[$i] = $symb;
449 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
451 $infochk{$symb} = $i;
453 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(entry|ret)$TPOSTLBL[@]?$/o ) {
455 $chkcat[$i] = 'slow';
460 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_fast\d*$TPOSTLBL[@]?$/o ) {
462 $chkcat[$i] = 'fast';
467 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_closure$TPOSTLBL[@]?$/o ) {
469 $chkcat[$i] = 'closure';
472 $closurechk{$1} = $i;
474 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_srt$TPOSTLBL[@]?$/o ) {
481 } elsif ( /^$TUS[@]?ghc.*c_ID$TPOSTLBL/o ) {
483 $chkcat[$i] = 'consist';
485 } elsif ( /^($TUS[@]?__gnu_compiled_c|gcc2_compiled\.)$TPOSTLBL/o ) {
488 } elsif ( /^$TUS[A-Za-z0-9_]+\.\d+$TPOSTLBL[@]?$/o
489 || /^$TUS[@]?.*_CAT$TPOSTLBL[@]?$/o # PROF: _entryname_CAT
490 || /^$TUS[@]?CC_.*_struct$TPOSTLBL[@]?$/o # PROF: _CC_ccident_struct
491 || /^$TUS[@]?.*_done$TPOSTLBL[@]?$/o # PROF: _module_done
492 || /^$TUS[@]?_module_registered$TPOSTLBL[@]?$/o # PROF: _module_registered
495 $chkcat[$i] = 'data';
498 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
503 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
508 } elsif ( /^$TUS[@]?CC(S)?_.*$/ ) {
509 # all CC_ symbols go in the data section...
511 $chkcat[$i] = 'data';
514 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(alt|dflt)$TPOSTLBL[@]?$/o ) {
516 $chkcat[$i] = 'misc';
519 #$chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
521 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_vtbl$TPOSTLBL[@]?$/o ) {
523 $chkcat[$i] = 'vector';
528 # As a temporary solution for compiling "foreign export" declarations,
529 # we use global variables to pass arguments from C to STG land.
530 # These declarations live in the .hc file and not in the generated C
531 # stub file, so we let them pass through here.
532 } elsif ( /^[\t ]+\.comm[\t ]+__fexp_.*$/ ) {
534 $chkcat[$i] = 'data';
537 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
538 && /^(_uname|uname|stat|fstat):/ ) {
539 # for some utterly bizarre reason, this platform
540 # likes to drop little local C routines with these names
541 # into each and every .o file that #includes the
542 # relevant system .h file. Yuck. We just don't
543 # tolerate them in .hc files (which we are processing
544 # here). If you need to call one of these things from
545 # Haskell, make a call to your own C wrapper, then
546 # put that C wrapper (which calls one of these) in a
547 # plain .c file. WDP 95/12
549 $chkcat[$i] = 'toss';
552 } elsif ( /^$TUS[@]?[A-Za-z0-9_]/o
553 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
555 && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
556 || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
559 print "Funny global thing?: $_"
560 unless $KNOWN_FUNNY_THING{$thing}
561 || /^$TUS[@]?stg_.*$TPOSTLBL[@]?$/o # RTS internals
562 || /^$TUS[@]__fexp_.*$TPOSTLBL$/o # foreign export
563 || /^$TUS[@]?_reg.*$TPOSTLBL$/o # PROF: __reg<module>
564 || /^$TUS[@]?.*_btm$TPOSTLBL$/o # large bitmaps
565 || /^$TUS[@]?.*_closure_tbl$TPOSTLBL$/o; # closure tables
567 $chkcat[$i] = 'misc';
568 if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
569 { $chksymb[$i] = $thing; }
570 else { $chksymb[$i] = ''; };
572 } else { # simple line (duplicated at the top)
577 $numchks = $#chk + 1;
579 # the division into chunks is imperfect;
580 # we throw some things over the fence into the next
583 # also, there are things we would like to know
584 # about the whole module before we start spitting
587 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
589 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
591 # Alphas: NB: we start meddling at chunk 1, not chunk 0
592 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
593 # spits a ".quad 0" in after the v first ".rdata"; we
594 # detect this special case (tossing the ".quad 0")!
595 local($magic_rdata_seen) = 0;
597 # HPPAs, MIPSen: also start medding at chunk 1
599 if ($TargetPlatform =~ /^powerpc|^rs6000/) {
600 print OUTASM $T_HDR_toc; # yes, we have to put a .toc
601 # in the beginning of every file!
602 %tocequiv = (); # maps toc symbol number to toc symbol
603 %revtocequiv = (); # maps toc symbol to toc symbol number
604 for ($i = 1; $i < $numchks; $i++) {
605 $chk[$i] =~ s/\[RW\]//g;
606 $chk[$i] =~ s/\[DS\]//g;
607 $chk[$i] =~ s/^\.csect .*\[DS\]$//g;
609 if ( $chkcat[$i] eq 'toc' && $chk[$i] !~ /\.byte/ )
610 #ToDo: instead of all these changes, just fix mangle_powerpc_tailjump and delete/ignore these tocs?
611 { $chk[$i] =~ s/$T_MOVE_DIRVS//g;
612 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_fast\d+)/\t\.tc \1\[TC\],\.\2/;
613 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_entry)\n/\t\.tc \1\[TC\],\.\2\n/;
614 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(ret_\S+)/\t\.tc \1\[TC\],\.\2/;
615 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(alt_\S+)/\t\.tc \1\[TC\],\.\2/;
616 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
618 $tocnumber = $chksymb[$i];
620 $tocsymb =~ s/^LC\.\.\d+:\n//;
621 $tocsymb =~ s/^\t\.tc \S+,(\S+)\n/\1/;
622 $tocequiv{$tocnumber} = $tocsymb;
624 } elsif ( $chkcat[$i] eq 'toc' && $chk[$i] =~ /\.byte/ ) {
625 $chkcat[$i] = 'literal';
630 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
631 $c = $chk[$i]; # convenience copy
633 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
635 # toss all prologue stuff; HPPA is pretty weird
637 $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/;
639 # be slightly paranoid to make sure there's
640 # nothing surprising in there
641 if ( $c =~ /--- BEGIN ---/ ) {
642 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
644 if ($TargetPlatform =~ /^i386-/) {
645 $p =~ s/^\tpushl \%edi\n//;
646 $p =~ s/^\tpushl \%esi\n//;
647 $p =~ s/^\tsubl \$\d+,\%esp\n//;
648 $p =~ s/^\tmovl \$\d+,\%eax\n\tcall __alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
649 } elsif ($TargetPlatform =~ /^m68k-/) {
650 $p =~ s/^\tlink a6,#-?\d.*\n//;
651 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
652 # The above showed up in the asm code,
653 # so I added it here.
654 # I hope it's correct.
656 $p =~ s/^\tmovel d2,sp\@-\n//;
657 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
658 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
659 } elsif ($TargetPlatform =~ /^mips-/) {
660 # the .frame/.mask/.fmask that we use is the same
661 # as that produced by GCC for miniInterpret; this
662 # gives GDB some chance of figuring out what happened
663 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
664 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
665 $p =~ s/^\t\.(mask|fmask).*\n//g;
666 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
667 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
668 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
669 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
670 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
671 $p =~ s/__FRAME__/$FRAME/;
672 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000/) {
673 $p =~ s/^\tmflr 0\n//;
674 $p =~ s/^\tstm \d+,-\d+\(1\)\n//;
675 $p =~ s/^\tstw? 0,\d+\(1\)\n//g;
676 $p =~ s/^\tstw? 1,\d+\(1\)\n//g; #mc
677 $p =~ s/^\tlw?z 0,0\(1\)\n//g; #mc
678 $p =~ s/^\tstw?u 1,-\d+\(1\)\n//;
679 $p =~ s/^\tstw? \d+,-\d+\(1\)\n//g;
680 $p =~ s/^\tstfd \d+,-\d+\(1\)\n//g;
682 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
685 die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
686 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
688 # glue together what's left
690 $c =~ s/\n\t\n/\n/; # junk blank line
694 if ( $TargetPlatform =~ /^mips-/ ) {
695 # MIPS: first, this basic sequence may occur "--- END ---" or not
696 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
699 # toss all epilogue stuff; again, paranoidly
700 if ( $c =~ /--- END ---/ ) {
701 if (($r, $e) = split(/--- END ---/, $c)) {
702 if ($TargetPlatform =~ /^i386-/) {
704 $e =~ s/^\tpopl \%edi\n//;
705 $e =~ s/^\tpopl \%esi\n//;
706 $e =~ s/^\taddl \$\d+,\%esp\n//;
707 } elsif ($TargetPlatform =~ /^m68k-/) {
708 $e =~ s/^\tunlk a6\n//;
710 } elsif ($TargetPlatform =~ /^mips-/) {
711 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
712 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
713 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
714 $e =~ s/^\tj\t\$31\n//;
715 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
716 $e =~ s/^\taddi 1,1,\d+\n//;
717 $e =~ s/^\tcal 1,\d+\(1\)\n//;
718 $e =~ s/^\tlw?z? \d+,\d+\(1\)\n//;
719 $e =~ s/^\tmtlr 0\n//;
722 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
724 die "Epilogue junk?: $e\n" if $e =~ /^\t[^\.]/
725 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
727 # glue together what's left
729 $c =~ s/\n\t\n/\n/; # junk blank line
733 # On SPARCs, we don't do --- BEGIN/END ---, we just
734 # toss the register-windowing save/restore/ret* instructions
736 if ( $TargetPlatform =~ /^sparc-/ ) {
737 $c =~ s/^\t(save .*|restore|ret|retl)\n//g;
738 # throw away PROLOGUE comments
739 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
742 # On Alphas, the prologue mangling is done a little later (below)
744 # toss all calls to __DISCARD__
745 $c =~ s/^\t(call|jbsr|jal)\s+$TUS[@]?__DISCARD__\n//go;
747 # MIPS: that may leave some gratuitous asm macros around
748 # (no harm done; but we get rid of them to be tidier)
749 $c =~ s/^\t\.set\tnoreorder\n\t\.set\tnomacro\n\taddu\t(\S+)\n\t\.set\tmacro\n\t\.set\treorder\n/\taddu\t$1\n/
750 if $TargetPlatform =~ /^mips-/;
752 # toss stack adjustment after DoSparks
753 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
754 if $TargetPlatform =~ /^m68k-/; # this looks old...
756 if ( $TargetPlatform =~ /^alpha-/ &&
757 ! $magic_rdata_seen &&
758 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
759 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
760 $magic_rdata_seen = 1;
763 # pick some end-things and move them to the next chunk
765 # pin a funny end-thing on (for easier matching):
766 $c .= 'FUNNY#END#THING';
768 while ( $c =~ /$TMOVEDIRVS[@]?FUNNY#END#THING/o ) { # [@]? is a silly hack to avoid having to use curlies for T_PRE_APP
769 # (this SEGVs perl4 on alphas, you see)
772 if ( $i < ($numchks - 1)
773 && ( $to_move =~ /$TCOPYDIRVS/
774 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
775 $chk[$i + 1] = $to_move . $chk[$i + 1];
776 # otherwise they're tossed
779 $c =~ s/$TMOVEDIRVS[@]?FUNNY#END#THING/FUNNY#END#THING/o; # [@]? is a hack (see above)
782 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
784 # toss all prologue stuff, except for loading gp, and the ..ng address
785 if (($p, $r) = split(/^\t\.prologue/, $c)) {
786 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
787 $c = $keep . "..ng:\n";
789 print STDERR "malformed code block ($ent)?\n"
792 $c .= "\t.frame \$30,0,\$26,0\n\t.prologue" . $r;
795 $c =~ s/FUNNY#END#THING//;
797 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
799 $chk[$i] = $c; # update w/ convenience copy
802 if ( $TargetPlatform =~ /^alpha-/ ) {
803 # print out the header stuff first
804 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
805 print OUTASM $chk[0];
807 } elsif ( $TargetPlatform =~ /^hppa/ ) {
808 print OUTASM $chk[0];
810 } elsif ( $TargetPlatform =~ /^mips-/ ) {
811 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
813 # get rid of horrible "<dollar>Revision: .*$" strings
814 local(@lines0) = split(/\n/, $chk[0]);
816 while ( $z <= $#lines0 ) {
817 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
820 while ( $z <= $#lines0 ) {
822 last if $lines0[$z] =~ /[,\t]0x0$/;
828 $chk[0] = join("\n", @lines0);
829 $chk[0] =~ s/\n\n+/\n/;
830 print OUTASM $chk[0];
833 # print out all the literal strings next
834 for ($i = 0; $i < $numchks; $i++) {
835 if ( $chkcat[$i] eq 'literal' ) {
836 print OUTASM $T_HDR_literal, $chk[$i];
837 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
839 $chkcat[$i] = 'DONE ALREADY';
843 # on the HPPA, print out all the bss next
844 if ( $TargetPlatform =~ /^hppa/ ) {
845 for ($i = 1; $i < $numchks; $i++) {
846 if ( $chkcat[$i] eq 'bss' ) {
847 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
848 print OUTASM $chk[$i];
850 $chkcat[$i] = 'DONE ALREADY';
855 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
856 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
858 next if $chkcat[$i] eq 'DONE ALREADY';
860 if ( $chkcat[$i] eq 'misc' ) {
861 if ($chk[$i] ne '') {
862 print OUTASM $T_HDR_misc;
863 if ($TargetPlatform =~ /^powerpc-|^rs6000/) {
864 $chksymb[$i] =~ s/://;
865 #new if ($chksymb[$i] =~ /ret.*upd/ || $KNOWN_FUNNY_THING{$chksymb[$i]}
866 #new || $chksymb[$i] =~ /^$.{T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o )
867 #new { print OUTASM "\t\.globl $chksymb[$i]\n"; }
868 # if ($chksymb[$i] ne '' && $chksymb[$i] !~ /ret_[a-z]/ && $chksymb[$i] !~ /djn_[a-z]/)
869 if ($chksymb[$i] ne '')
870 { print OUTASM "\t\.globl \.$chksymb[$i]\n"; };
871 if ($chk[$i] =~ /TOC\[tc0\], 0\n/)
872 { ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$i]); $printDS = 1;}
873 else { $r = $chk[$i]; $printDS = 0; };
874 $chk[$i] = &mangle_powerpc_tailjump($r);
876 &print_doctored($chk[$i], 0);
877 if ($TargetPlatform =~ /^powerpc-|^rs6000-/ && $printDS) {
878 #ok if ($chksymb[$i] !~ /\_regMain/) {
879 print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
880 print OUTASM "${p}TOC[tc0], 0\n";
885 } elsif ( $chkcat[$i] eq 'toss' ) {
886 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
888 } elsif ( $chkcat[$i] eq 'data' ) {
889 if ($chk[$i] ne '') {
890 print OUTASM $T_HDR_data;
891 print OUTASM $chk[$i];
894 } elsif ( $chkcat[$i] eq 'consist' ) {
895 if ( $chk[$i] =~ /$T_hsc_cc_PAT/o ) {
896 local($consist) = "$1.$2.$3";
898 $consist =~ s/\//./g;
900 $consist =~ s/[^A-Za-z0-9_.]/ZZ/g; # ToDo: properly?
902 # Using a cygnus-2.7-96q4 gcc build on hppas, the
903 # consistency chunk for ghc_cc_ID often (but not always!)
904 # gets lumped with a bunch of .IMPORT directives containing info on
905 # the code or data space nature of external symbols. We can't
906 # toss these, so once the consistency ID has been turned into
907 # a representable symbol, we substitute it for the symbol
908 # that the string was attached to in the first place (ghc_cc_ID.)
909 # (The original string is also substituted away.)
911 # This change may affect the code output on other platforms in
912 # adverse ways, hence we restrict this hack hppa targets only.
915 if ( $TargetPlatform =~ /^hppa/ ) {
916 $chk[$i] =~ s/^$TUS[@]?ghc.*c_ID$TPOSTLBL/$consist/o;
917 $chk[$i] =~ s/\t$T_hsc_cc_PAT/$T_HDR_misc/o;
918 $consist = $chk[$i]; #clumsily
920 print OUTASM $T_HDR_consist, "${consist}${T_POST_LBL}\n";
922 } elsif ( $TargetPlatform !~ /^(mips)-/ ) { # we just don't try in those case (ToDo)
923 # on mips: consistency string is just a v
924 # horrible bunch of .bytes,
925 # which I am too lazy to sort out (WDP 95/05)
927 print STDERR "Couldn't grok consistency: ", $chk[$i];
930 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
931 # we can just re-constitute this one...
932 # NB: we emit _three_ underscores no matter what,
933 # so ghc-split doesn't have to care.
934 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
936 } elsif ( $chkcat[$i] eq 'closure'
937 || $chkcat[$i] eq 'srt'
938 || $chkcat[$i] eq 'infotbl'
939 || $chkcat[$i] eq 'slow'
940 || $chkcat[$i] eq 'fast' ) { # do them in that order
941 $symb = $chksymb[$i];
944 if ( defined($closurechk{$symb}) ) {
945 print OUTASM $T_HDR_closure;
946 print OUTASM $chk[$closurechk{$symb}];
947 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
951 if ( defined($srtchk{$symb}) ) {
952 print OUTASM $T_HDR_srt;
953 print OUTASM $chk[$srtchk{$symb}];
954 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
958 if ( defined($infochk{$symb}) ) {
960 print OUTASM $T_HDR_info;
961 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
962 if ( !defined($slowchk{$symb}) && defined($fastchk{$symb}) ) {
963 $fastname = $chk[$fastchk{$symb}];
964 $fastname =~ s/([_A-Za-z]*_fast\d+):.*(.*\n)*/\1/;
965 $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long $fastname/;
967 $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
968 $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
969 print OUTASM $chk[$infochk{$symb}];
971 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
973 # entry code will be put here!
975 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
979 if ( defined($slowchk{$symb}) ) {
981 # teach it to drop through to the fast entry point:
982 $c = $chk[$slowchk{$symb}];
984 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
985 ($p, $r) = split(/TOC\[tc0\], 0\n/, $c);
986 if ($symb =~ /^[_A-Z]/)
988 print OUTASM "\t\.globl \.${chksymb[$i]}_entry\n";
989 print OUTASM "\.csect ${symb}_entry[DS]\n";
990 print OUTASM "${p}TOC[tc0], 0\n";
992 $r =~ s/\.csect \.text\[PR\]\n//; # todo: properly - andre
993 $c = &mangle_powerpc_tailjump($r);
996 if ( defined($fastchk{$symb}) ) {
997 if ( $TargetPlatform =~ /^alpha-/ ) {
998 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
999 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1000 $c =~ s/^\s+ldil.*\n\s+ldo.*\n\s+bv.*\n(.*\n)?\s+\.EXIT/$1\t.EXIT/;
1001 } elsif ( $TargetPlatform =~ /^i386-/ ) {
1002 # Reg alloc depending, gcc generated code may jump to the fast entry point via
1003 # a number of registers.
1004 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edx\n\tjmp \*\%edx\n//;
1005 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%ecx\n\tjmp \*\%ecx\n//;
1006 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%eax\n\tjmp \*\%eax\n//;
1007 # The next two only apply if we're not stealing %esi or %edi.
1008 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%esi\n\tjmp \*\%esi\n// if ($StolenX86Regs < 3);
1009 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edi\n\tjmp \*\%edi\n// if ($StolenX86Regs < 4);
1010 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1011 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1012 } elsif ( $TargetPlatform =~ /^m68k-/ ) {
1013 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n\tnop\n//;
1014 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n//;
1015 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ ) {
1016 $c =~ s/^\tb \.${T_US}${symb}_fast\d+\n//;
1017 } elsif ( $TargetPlatform =~ /^sparc-/ ) {
1018 $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n\tnop\n//;
1019 $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n(\t[a-z].*\n)/$1/;
1021 print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n";
1025 if ( $TargetPlatform !~ /^(alpha-|hppa|mips-)/ ) {
1026 # On alphas, hppa: no very good way to look for "dangling"
1027 # references to fast-entry point.
1028 # (questionable re hppa and mips...)
1029 print STDERR "still has jump to fast entry point:\n$c"
1030 if $c =~ /$TUS[@]?$symb[@]?_fast/; # NB: paranoia
1033 print OUTASM $T_HDR_entry;
1035 &print_doctored($c, 1); # NB: the 1!!!
1037 $chkcat[$slowchk{$symb}] = 'DONE ALREADY';
1041 if ( defined($fastchk{$symb}) ) {
1042 $c = $chk[$fastchk{$symb}];
1043 if ( ! defined($slowchk{$symb})
1044 # ToDo: the || clause can go once we're no longer
1045 # concerned about producing exactly the same output as before
1046 #OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/
1048 print OUTASM $T_HDR_fast;
1050 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1051 local(@lbls) = split(/:/, $c);
1052 $fullname = $lbls[0];
1053 $fullname =~ s/$T_MOVE_DIRVS//g;
1054 if ( $fullname =~ /^[A-Z]/)
1055 { print OUTASM "\t\.globl \.${fullname}\n";
1057 # print OUTASM "\t\.lglobl \.${fullname}\n"; #todo: rm - andre
1059 $c =~ s/((.*\n)*)\t.long \S+, TOC\[tc0\], 0\n\.csect \.text\[PR\]\n((.*\n)*)/\1\3/;
1060 $c = &mangle_powerpc_tailjump($c);
1062 &print_doctored($c, 0);
1063 $chkcat[$fastchk{$symb}] = 'DONE ALREADY';
1066 } elsif ( $chkcat[$i] eq 'vector'
1067 || $chkcat[$i] eq 'direct' ) { # do them in that order
1068 $symb = $chksymb[$i];
1071 if ( defined($vectorchk{$symb}) ) {
1072 print OUTASM $T_HDR_vector;
1073 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1074 if ( $symb =~ /^[A-Z]/) {
1075 print OUTASM "\t\.globl \.${symb}_vtbl\n";
1076 print OUTASM "\t\.globl ${symb}_vtbl\n";
1078 $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
1079 print OUTASM ".${symb}_vtbl:\n";
1080 print OUTASM $chk[$vectorchk{$symb}];
1082 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1084 # direct return code will be put here!
1085 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1087 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1088 # Alphas: the commented nop is for the splitter, to ensure
1089 # that no module ends with a label as the very last
1090 # thing. (The linker will adjust the label to point
1091 # to the first code word of the next module linked in,
1092 # even if alignment constraints cause the label to move!)
1094 print OUTASM "\t# nop\n";
1097 } elsif ( $chkcat[$i] eq 'toc' ) {
1098 # silly optimisation to print tocs, since they come in groups...
1099 print OUTASM $T_HDR_toc;
1101 while ($chkcat[$j] eq 'toc')
1102 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1105 print OUTASM $chk[$j];
1107 $chkcat[$j] = 'DONE ALREADY';
1112 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1116 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1118 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1119 print OUTASM ".csect .text[PR]\n_section_.text:\n.csect .data[RW]\n\t.long _section_.text\n"
1123 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1124 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1129 sub mash_hppa_prologue { # OK, epilogue, too
1132 # toss all prologue stuff
1133 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1135 # Lie about our .CALLINFO
1136 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1143 # toss all epilogue stuff
1144 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1146 # Sorry; we moved the _info stuff to the code segment.
1147 s/_info,DATA/_info,CODE/g;
1154 sub print_doctored {
1155 local($_, $need_fallthru_patch) = @_;
1157 if ( $TargetPlatform !~ /^i386-/
1158 || ! /^\t[a-z]/ # no instructions in here, apparently
1159 || /^${T_US}_reg[A-Za-z0-9_]+:/) {
1163 # OK, must do some x86 **HACKING**
1165 local($entry_patch) = '';
1166 local($exit_patch) = '';
1168 # gotta watch out for weird instructions that
1169 # invisibly smash various regs:
1170 # rep* %ecx used for counting
1171 # scas* %edi used for destination index
1172 # cmps* %e[sd]i used for indices
1173 # loop* %ecx used for counting
1178 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1180 # * GCC used an "STG reg" for its own purposes
1182 # * some secret uses of machine reg, requiring STG reg
1183 # to be saved/restored
1185 # The most dangerous "GCC uses" of an "STG reg" are when
1186 # the reg holds the target of a jmp -- it's tricky to
1187 # insert the patch-up code before we get to the target!
1188 # So here we change the jmps:
1190 # --------------------------------------------------------
1191 # it can happen that we have jumps of the form...
1192 # jmp *<something involving %esp>
1194 # jmp <something involving another naughty register...>
1196 # a reasonably-common case is:
1198 # movl $_blah,<bad-reg>
1202 # the short form may tickle perl bug:
1203 # s/^\tmovl \$${T_US}(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp $T_US$1/g;
1204 s/^\tmovl \$${T_US}(.*),\%eax\n\tjmp \*\%eax/\tjmp $T_US$1/g;
1205 s/^\tmovl \$${T_US}(.*),\%ebx\n\tjmp \*\%ebx/\tjmp $T_US$1/g;
1206 s/^\tmovl \$${T_US}(.*),\%ecx\n\tjmp \*\%ecx/\tjmp $T_US$1/g;
1207 s/^\tmovl \$${T_US}(.*),\%edx\n\tjmp \*\%edx/\tjmp $T_US$1/g;
1209 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1210 s/^\tmovl (.*),\%esi\n\tjmp \*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1211 s/^\tjmp \*(-?\d*)\((.*\%esi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1212 s/^\tjmp \*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1213 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1214 if /(jmp|call) .*\%esi/;
1216 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1217 s/^\tmovl (.*),\%edi\n\tjmp \*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1218 s/^\tjmp \*(-?\d*)\((.*\%edi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1219 s/^\tjmp \*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1220 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1221 if /(jmp|call) .*\%edi/;
1224 # OK, now we can decide what our patch-up code is going to
1227 # Offsets into register table - you'd better update these magic
1228 # numbers should you change its contents!
1229 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1230 local($OFFSET_Hp)=92;
1232 # Note funky ".=" stuff; we're *adding* to these _patch guys
1233 if ( $StolenX86Regs <= 2
1234 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1235 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1236 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1238 # nothing for call_{entry,exit} because %esi is callee-save
1240 if ( $StolenX86Regs <= 3
1241 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1242 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1243 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1245 # nothing for call_{entry,exit} because %edi is callee-save
1248 # --------------------------------------------------------
1249 # next, here we go with non-%esp patching!
1251 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1253 # Before calling GC we must set up the exit condition before the call
1254 # and entry condition when we come back
1256 # fix _all_ non-local jumps:
1258 s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1259 s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1261 s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
1263 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1264 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1266 if ($StolenX86Regs == 2 ) {
1267 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1268 if /^\t(jmp|call) .*\%e(si|di)/;
1269 } elsif ($StolenX86Regs == 3 ) {
1270 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1271 if /^\t(jmp|call) .*\%edi/;
1274 # --------------------------------------------------------
1275 # that's it -- print it
1277 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1281 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1282 print OUTASM $exit_patch;
1283 # ToDo: make it not print if there is a "jmp" at the end
1289 sub init_FUNNY_THINGS {
1290 %KNOWN_FUNNY_THING = (
1292 # "${T_US}stg_.*{T_POST_LBL}", 1,
1297 The following table reversal is used for both info tables and return
1298 vectors. In both cases, we remove the first entry from the table,
1299 reverse the table, put the label at the end, and paste some code
1300 (that which is normally referred to by the first entry in the table)
1301 right after the table itself. (The code pasting is done elsewhere.)
1305 local($symb, $tbl, $discard1) = @_;
1307 local($before) = '';
1309 local(@imports) = (); # hppa only
1312 local(@lines) = split(/\n/, $tbl);
1313 local($i, $j); #local ($i, $extra, $words_to_pad, $j);
1315 # see comment in mangleAsm as to why this silliness is needed.
1316 local($TDOTWORD) = ${T_DOT_WORD};
1317 local($TDOTGLOBAL) = ${T_DOT_GLOBAL};
1318 local($TUS) = ${T_US};
1319 local($TPOSTLBL) = ${T_POST_LBL};
1321 # Deal with the header...
1322 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?$TDOTWORD\s+/o; $i++) {
1323 $label .= $lines[$i] . "\n",
1324 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info$TPOSTLBL[@]?$/o
1325 || $lines[$i] =~ /$TDOTGLOBAL/o
1326 || $lines[$i] =~ /^$TUS[@]?\S+_vtbl$TPOSTLBL[@]?$/o;
1328 $before .= $lines[$i] . "\n"; # otherwise...
1331 # Grab the table data...
1332 if ( $TargetPlatform !~ /^hppa/ ) {
1333 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?$TDOTWORD\s+/o; $i++) {
1334 push(@words, $lines[$i]);
1336 } else { # hppa weirdness
1337 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+\.(word|IMPORT)/; $i++) {
1338 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1339 push(@imports, $lines[$i]);
1341 # We don't use HP's ``function pointers''
1342 # We just use labels in code space, like normal people
1343 $lines[$i] =~ s/P%//;
1344 push(@words, $lines[$i]);
1349 # now throw away the first word (SRT) iff it is empty.
1350 # The .zero business is for Linux/ELF.
1351 # The .skip business is for Sparc/Solaris/ELF.
1352 if ($discard1 && $words[0] =~ /^\t?($TDOTWORD\s+0|\.zero\s+4|\.skip\s+4)/) {
1356 # Padding removed to reduce code size and improve performance on Pentiums.
1358 # for 486-cache-friendliness, we want our tables aligned
1359 # on 16-byte boundaries (.align 4). Let's pad:
1360 # $extra = ($#words + 1) % 4;
1361 # $words_to_pad = ($extra == 0) ? 0 : 4 - $extra;
1362 # for ($j = 0; $j < $words_to_pad; $j++) { push(@words, "\t${T_DOT_WORD} 0"); }
1364 for (; $i <= $#lines; $i++) {
1365 $after .= $lines[$i] . "\n";
1368 # Alphas:If we have anonymous text (not part of a procedure), the
1369 # linker may complain about missing exception information. Bleh.
1370 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1371 $before = "\t.ent $1\n" . $before;
1372 $after .= "\t.end $1\n";
1376 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1377 . join("\n", @words) . "\n"
1380 # print STDERR "before=$before\n";
1381 # print STDERR "label=$label\n";
1382 # print STDERR "words=",(reverse @words),"\n";
1383 # print STDERR "after=$after\n";
1389 The HP is a major nuisance. The threaded code mangler moved info
1390 tables from data space to code space, but unthreaded code in the RTS
1391 still has references to info tables in data space. Since the HP
1392 linker is very precise about where symbols live, we need to patch the
1393 references in the unthreaded RTS as well.
1396 sub mini_mangle_asm_hppa {
1397 local($in_asmf, $out_asmf) = @_;
1399 open(INASM, "< $in_asmf")
1400 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1401 open(OUTASM,"> $out_asmf")
1402 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1405 s/_info,DATA/_info,CODE/; # Move _info references to code space
1411 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1412 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1418 sub mini_mangle_asm_powerpc {
1419 local($in_asmf, $out_asmf) = @_;
1421 open(INASM, "< $in_asmf")
1422 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1423 open(OUTASM,"> $out_asmf")
1424 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1427 s/long _PRStart/long ._PRStart/;
1428 s/long _PRIn_/long ._PRIn_/;
1429 s/long _Dummy_(\S+)_entry/long ._Dummy_\1_entry/;
1430 s/long _PRMarking_MarkNextRoot\[DS\]/long ._PRMarking_MarkNextRoot/;
1431 s/long _PRMarking_MarkNextCAF\[DS\]/long ._PRMarking_MarkNextCAF/;
1432 s/long _PRMarking_MarkNextAStack\[DS\]/long ._PRMarking_MarkNextAStack/;
1433 s/long _PRMarking_MarkNextBStack\[DS\]/long ._PRMarking_MarkNextBStack/;
1434 s/\.tc EnterNodeCode\[TC]\,EnterNodeCode\[DS\]/\.tc EnterNodeCode\[TC]\,.EnterNodeCode/; # CONC
1435 s/\.tc CheckHeapCode\[TC]\,CheckHeapCode\[DS\]/\.tc CheckHeapCode\[TC]\,.CheckHeapCode/; # CONC
1440 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1441 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1444 sub mangle_powerpc_tailjump {
1446 local($maybe_more) = 1;
1447 while (($c =~ /\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n/) && $maybe_more)
1450 $lcsymb =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1451 # the checks for r1 and r2 are mostly paranoia...
1453 $r1 =~ s/(.*\n)*\tlw?z? (\d+),LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1455 $r2 =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr (\d+)\n\tbctr\n(.*\n)*/\3/;
1458 $c =~ s/((.*\n)*)\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n((.*\n)*)/\1\tb $tocequiv{$lcsymb}\n\3/;
1464 # make "require"r happy...