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{Top-level code}
47 %************************************************************************
50 $TargetPlatform = $TARGETPLATFORM;
52 ($Pgm = $0) =~ s|.*/||;
56 if ( $TargetPlatform =~ /^i386-/ ) {
57 $StolenX86Regs = $ARGV[2];
60 &mangle_asm($ifile,$ofile);
65 %************************************************************************
67 \subsection{Constants for various architectures}
69 %************************************************************************
72 sub init_TARGET_STUFF {
74 #--------------------------------------------------------#
75 if ( $TargetPlatform =~ /^alpha-.*-.*/ ) {
77 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
78 $T_US = ''; # _ if symbols have an underscore on the front
79 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
80 $T_CONST_LBL = '^\$C(\d+):$'; # regexp for what such a lbl looks like
83 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\#.*|\.(file|loc)\s+\S+\s+\S+|\.text|\.r?data)\n)';
84 $T_COPY_DIRVS = '^\s*(\#|\.(file|globl|ent|loc))';
86 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
87 $T_DOT_WORD = '\.quad';
88 $T_DOT_GLOBAL = '^\t\.globl';
89 $T_HDR_literal = "\.rdata\n\t\.align 3\n";
90 $T_HDR_misc = "\.text\n\t\.align 3\n";
91 $T_HDR_data = "\.data\n\t\.align 3\n";
92 $T_HDR_consist = "\.text\n";
93 $T_HDR_closure = "\.data\n\t\.align 3\n";
94 $T_HDR_srt = "\.text\n\t\.align 3\n";
95 $T_HDR_info = "\.text\n\t\.align 3\n";
96 $T_HDR_entry = "\.text\n\t\.align 3\n";
97 $T_HDR_fast = "\.text\n\t\.align 3\n";
98 $T_HDR_vector = "\.text\n\t\.align 3\n";
99 $T_HDR_direct = "\.text\n\t\.align 3\n";
101 #--------------------------------------------------------#
102 } elsif ( $TargetPlatform =~ /^hppa/ ) {
104 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
105 $T_US = ''; # _ if symbols have an underscore on the front
106 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
107 $T_CONST_LBL = '^L\$C(\d+)$'; # regexp for what such a lbl looks like
110 $T_MOVE_DIRVS = '^((\s+\.(IMPORT|EXPORT|PARAM).*|\s+\.align\s+\d+|\s+\.(SPACE|SUBSPA)\s+\S+|\s*)\n)';
111 $T_COPY_DIRVS = '^\s+\.(IMPORT|EXPORT)';
113 $T_hsc_cc_PAT = '\.STRING.*\)(hsc|cc) (.*)\\\\x09(.*)\\\\x00"';
114 $T_DOT_WORD = '\.(blockz|word|half|byte)';
115 $T_DOT_GLOBAL = '^\s+\.EXPORT';
116 $T_HDR_literal = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
117 $T_HDR_misc = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
118 $T_HDR_data = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
119 $T_HDR_consist = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
120 $T_HDR_closure = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
121 $T_HDR_srt = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
122 $T_HDR_info = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
123 $T_HDR_entry = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
124 $T_HDR_fast = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
125 $T_HDR_vector = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
126 $T_HDR_direct = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
128 #--------------------------------------------------------#
129 } elsif ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd2|netbsd|nextstep3|cygwin32|mingw32)$/ ) {
130 # NeXT added but not tested. CaS
132 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
133 $T_US = '_'; # _ if symbols have an underscore on the front
134 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
135 $T_CONST_LBL = '^LC(\d+):$';
137 $T_X86_PRE_LLBL_PAT = 'L';
138 $T_X86_PRE_LLBL = 'L';
139 $T_X86_BADJMP = '^\tjmp [^L\*]';
141 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,0x90)?|\.globl\s+\S+|\.text|\.data|\.stab[^n].*|\.type\s+.*|\.size\s+.*)\n)';
142 $T_COPY_DIRVS = '\.(globl|stab)';
143 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
144 $T_DOT_WORD = '\.(long|word|value|byte|space)';
145 $T_DOT_GLOBAL = '\.globl';
146 $T_HDR_literal = "\.text\n\t\.align 2\n";
147 $T_HDR_misc = "\.text\n\t\.align 2,0x90\n";
148 $T_HDR_data = "\.data\n\t\.align 2\n";
149 $T_HDR_consist = "\.text\n";
150 $T_HDR_closure = "\.data\n\t\.align 2\n";
151 $T_HDR_closure = "\.data\n\t\.align 2\n\t.long 0\n" if ( $TargetPlatform =~ /.*-mingw32$/ );
152 $T_HDR_srt = "\.text\n\t\.align 2\n";
153 $T_HDR_info = "\.text\n\t\.align 2\n"; # NB: requires padding
154 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
155 $T_HDR_fast = "\.text\n\t\.align 2,0x90\n";
156 $T_HDR_vector = "\.text\n\t\.align 2\n"; # NB: requires padding
157 $T_HDR_direct = "\.text\n\t\.align 2,0x90\n";
159 #--------------------------------------------------------#
160 } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|freebsd|netbsd_elf)$/ ) {
162 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
163 $T_US = ''; # _ if symbols have an underscore on the front
164 $T_PRE_APP = # regexp that says what comes before APP/NO_APP
165 ($TargetPlatform =~ /-(linux|freebsd|netbsd_elf)$/) ? '#' : '/' ;
166 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
168 $T_X86_PRE_LLBL_PAT = '\.L';
169 $T_X86_PRE_LLBL = '.L';
170 $T_X86_BADJMP = '^\tjmp [^\.\*]';
172 $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)';
173 $T_COPY_DIRVS = '\.(globl)';
175 if ( $TargetPlatform =~ /freebsd|netbsd_elf/ ) {
176 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
178 $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
181 $T_DOT_WORD = '\.(long|value|byte|zero)';
182 $T_DOT_GLOBAL = '\.globl';
183 $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11)
184 $T_HDR_misc = "\.text\n\t\.align 4\n";
185 $T_HDR_data = "\.data\n\t\.align 4\n"; # ToDo: change align??
186 $T_HDR_consist = "\.text\n";
187 $T_HDR_closure = "\.data\n\t\.align 4\n"; # ToDo: change align?
188 $T_HDR_srt = "\.text\n\t\.align 4\n"; # ToDo: change align?
189 $T_HDR_info = "\.text\n\t\.align 4\n"; # NB: requires padding
190 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
191 $T_HDR_fast = "\.text\n\t\.align 4\n";
192 $T_HDR_vector = "\.text\n\t\.align 4\n"; # NB: requires padding
193 $T_HDR_direct = "\.text\n\t\.align 4\n";
195 #--------------------------------------------------------#
196 } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
198 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
199 $T_US = '_'; # _ if symbols have an underscore on the front
200 $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
201 $T_CONST_LBL = '^LC(\d+):$';
204 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
205 $T_COPY_DIRVS = '\.(globl|proc|stab)';
206 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
208 $T_DOT_WORD = '\.long';
209 $T_DOT_GLOBAL = '\.globl';
210 $T_HDR_literal = "\.text\n\t\.even\n";
211 $T_HDR_misc = "\.text\n\t\.even\n";
212 $T_HDR_data = "\.data\n\t\.even\n";
213 $T_HDR_consist = "\.text\n";
214 $T_HDR_closure = "\.data\n\t\.even\n";
215 $T_HDR_srt = "\.text\n\t\.even\n";
216 $T_HDR_info = "\.text\n\t\.even\n";
217 $T_HDR_entry = "\.text\n\t\.even\n";
218 $T_HDR_fast = "\.text\n\t\.even\n";
219 $T_HDR_vector = "\.text\n\t\.even\n";
220 $T_HDR_direct = "\.text\n\t\.even\n";
222 #--------------------------------------------------------#
223 } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
225 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
226 $T_US = ''; # _ if symbols have an underscore on the front
227 $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
228 $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
231 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
232 $T_COPY_DIRVS = '\.(globl|ent)';
234 $T_hsc_cc_PAT = 'I WAS TOO LAZY TO DO THIS BIT (WDP 95/05)';
235 $T_DOT_WORD = '\.word';
236 $T_DOT_GLOBAL = '^\t\.globl';
237 $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
238 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
239 $T_HDR_data = "\t\.data\n\t\.align 2\n";
240 $T_HDR_consist = 'TOO LAZY TO DO THIS TOO';
241 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
242 $T_HDR_srt = "\t\.text\n\t\.align 2\n";
243 $T_HDR_info = "\t\.text\n\t\.align 2\n";
244 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
245 $T_HDR_fast = "\t\.text\n\t\.align 2\n";
246 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
247 $T_HDR_direct = "\t\.text\n\t\.align 2\n";
249 #--------------------------------------------------------#
250 } elsif ( $TargetPlatform =~ /^powerpc-.*|^rs6000-.*/ ) {
252 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
253 $T_US = ''; # _ if symbols have an underscore on the front
254 $T_PRE_APP = 'NOT APPLICABLE'; # regexp that says what comes before APP/NO_APP
255 $T_CONST_LBL = 'NOT APPLICABLE'; # regexp for what such a lbl looks like
258 $T_MOVE_DIRVS = '^(\s*(\.toc|\.align \d+|\.csect \S+|\t\.?l?globl \S+)\n)';
259 $T_COPY_DIRVS = '\.(l?globl)';
261 $T_hsc_cc_PAT = '\.byte.*\)(hsc|cc) (.*)"\n\t\.byte \d+\n\t\.byte "(.*)"\n\t\.byte \d+';
262 $T_DOT_WORD = '\.long';
263 $T_DOT_GLOBAL = '\.globl';
264 $T_HDR_toc = "\.toc\n";
265 $T_HDR_literal = "\.csect .data[RW]\n\t\.align 2\n"; #not RO!?
266 $T_HDR_misc = "# misc\n\.csect \.text[PR]\n\t\.align 2\n";
267 $T_HDR_data = "# data\n\.csect \.data[RW]\n\t\.align 2\n";
268 $T_HDR_consist = "# consist\n\.csect \.data[RW]\n\t\.align 2\n";
269 $T_HDR_closure = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
270 $T_HDR_srt = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
271 $T_HDR_info = "# info\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
272 $T_HDR_entry = "# entry\n\.csect \.text[PR]\n\t\.align 2\n";
273 $T_HDR_fast = "# fast\n\.csect \.text[PR]\n\t\.align 2\n";
274 $T_HDR_vector = "# vector\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
275 $T_HDR_direct = "# direct\n";
277 #--------------------------------------------------------#
278 } elsif ( $TargetPlatform =~ /^sparc-.*-solaris2/ ) {
280 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
281 $T_US = ''; # _ if symbols have an underscore on the front
282 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
283 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
286 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\.section.*|\s+\.type.*|\s+\.size.*)\n)';
287 $T_COPY_DIRVS = '\.(global|proc|stab)';
289 $T_hsc_cc_PAT = '\.asciz.*\)(hsc|cc) (.*)\\\\t(.*)"';
290 $T_DOT_WORD = '\.(word|byte|half|skip|uahalf|uaword)';
291 $T_DOT_GLOBAL = '^\t\.global';
292 $T_HDR_literal = "\.text\n\t\.align 8\n";
293 $T_HDR_misc = "\.text\n\t\.align 4\n";
294 $T_HDR_data = "\.data\n\t\.align 8\n";
295 $T_HDR_consist = "\.text\n";
296 $T_HDR_closure = "\.data\n\t\.align 4\n";
297 $T_HDR_srt = "\.data\n\t\.align 4\n";
298 $T_HDR_info = "\.text\n\t\.align 4\n";
299 $T_HDR_entry = "\.text\n\t\.align 4\n";
300 $T_HDR_fast = "\.text\n\t\.align 4\n";
301 $T_HDR_vector = "\.text\n\t\.align 4\n";
302 $T_HDR_direct = "\.text\n\t\.align 4\n";
304 #--------------------------------------------------------#
305 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
307 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
308 $T_US = '_'; # _ if symbols have an underscore on the front
309 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
310 $T_CONST_LBL = '^LC(\d+):$';
313 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
314 $T_COPY_DIRVS = '\.(global|proc|stab)';
315 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
317 $T_DOT_WORD = '\.word';
318 $T_DOT_GLOBAL = '^\t\.global';
319 $T_HDR_literal = "\.text\n\t\.align 8\n";
320 $T_HDR_misc = "\.text\n\t\.align 4\n";
321 $T_HDR_data = "\.data\n\t\.align 8\n";
322 $T_HDR_consist = "\.text\n";
323 $T_HDR_closure = "\.data\n\t\.align 4\n";
324 $T_HDR_srt = "\.data\n\t\.align 4\n";
325 $T_HDR_info = "\.text\n\t\.align 4\n";
326 $T_HDR_entry = "\.text\n\t\.align 4\n";
327 $T_HDR_fast = "\.text\n\t\.align 4\n";
328 $T_HDR_vector = "\.text\n\t\.align 4\n";
329 $T_HDR_direct = "\.text\n\t\.align 4\n";
331 #--------------------------------------------------------#
333 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
338 print STDERR "T_STABBY: $T_STABBY\n";
339 print STDERR "T_US: $T_US\n";
340 print STDERR "T_PRE_APP: $T_PRE_APP\n";
341 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
342 print STDERR "T_POST_LBL: $T_POST_LBL\n";
343 if ( $TargetPlatform =~ /^i386-/ ) {
344 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
345 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
346 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
348 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
349 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
350 print STDERR "T_hsc_cc_PAT: $T_hsc_cc_PAT\n";
351 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
352 print STDERR "T_HDR_literal: $T_HDR_literal\n";
353 print STDERR "T_HDR_misc: $T_HDR_misc\n";
354 print STDERR "T_HDR_data: $T_HDR_data\n";
355 print STDERR "T_HDR_consist: $T_HDR_consist\n";
356 print STDERR "T_HDR_closure: $T_HDR_closure\n";
357 print STDERR "T_HDR_info: $T_HDR_info\n";
358 print STDERR "T_HDR_entry: $T_HDR_entry\n";
359 print STDERR "T_HDR_fast: $T_HDR_fast\n";
360 print STDERR "T_HDR_vector: $T_HDR_vector\n";
361 print STDERR "T_HDR_direct: $T_HDR_direct\n";
367 %************************************************************************
369 \subsection{Mangle away}
371 %************************************************************************
375 local($in_asmf, $out_asmf) = @_;
377 # multi-line regexp matching:
382 &init_TARGET_STUFF();
383 &init_FUNNY_THINGS();
385 # perl4 on alphas SEGVs when give ${foo} substitutions in patterns.
386 # To avoid them we declare some locals that allows to avoid using curlies.
387 local($TUS) = ${T_US};
388 local($TPOSTLBL) = ${T_POST_LBL};
389 local($TMOVEDIRVS) = ${T_MOVE_DIRVS};
390 local($TPREAPP) = ${T_PRE_APP};
391 local($TCOPYDIRVS) = ${T_COPY_DIRVS};
392 local($TDOTWORD) = ${T_DOT_WORD};
394 open(INASM, "< $in_asmf")
395 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
396 open(OUTASM,"> $out_asmf")
397 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
399 # read whole file, divide into "chunks":
400 # record some info about what we've found...
402 @chk = (); # contents of the chunk
403 $numchks = 0; # number of them
404 @chkcat = (); # what category of thing in each chunk
405 @chksymb = (); # what symbol(base) is defined in this chunk
406 %slowchk = (); # ditto, its regular "slow" entry code
407 %fastchk = (); # ditto, fast entry code
408 %closurechk = (); # ditto, the (static) closure
409 %srtchk = (); # ditto, its SRT (for top-level things)
410 %infochk = (); # given a symbol base, say what chunk its info tbl is in
411 %vectorchk = (); # ditto, return vector table
412 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
414 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
417 next if $T_STABBY && /^\.stab.*$TUS[@]?__stg_split_marker/o;
418 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
419 next if /^\t\.def.*endef$/;
420 next if /$TPREAPP(NO_)?APP/o;
421 next if /^;/ && $TargetPlatform =~ /^hppa/;
423 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|powerpc|rs6000)-/;
425 last if /^_section_\.text:$/ && $TargetPlatform =~ /^powerpc-|^rs6000-/;
427 if ( $TargetPlatform =~ /^mips-/
428 && /^\t\.(globl \S+ \.text|comm\t)/ ) {
429 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
431 # As a temporary solution for compiling "foreign export" declarations,
432 # we use global variables to pass arguments from C to STG land.
433 # These declarations live in the .hc file and not in the generated C
434 # stub file, so we let them pass through here.
435 } elsif ( /^\t\.comm\t__fexp_.*$/ ) {
437 $chkcat[$i] = 'data';
440 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
441 # duplicated from the bottom
445 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
446 # Alphas: Local labels not to be confused with new chunks
449 # NB: all the rest start with a non-space
451 } elsif ( $TargetPlatform =~ /^mips-/
452 && /^\d+:/ ) { # a funny-looking very-local label
455 } elsif ( /$T_CONST_LBL/o ) {
457 $chkcat[$i] = 'literal';
460 } elsif ( /^$TUS[@]?__stg_split_marker(\d*)$TPOSTLBL[@]?$/o ) {
462 $chkcat[$i] = 'splitmarker';
465 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_info$TPOSTLBL[@]?$/o ) {
468 $chkcat[$i] = 'infotbl';
469 $chksymb[$i] = $symb;
471 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
473 $infochk{$symb} = $i;
475 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(entry|ret)$TPOSTLBL[@]?$/o ) {
477 $chkcat[$i] = 'slow';
482 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_fast\d*$TPOSTLBL[@]?$/o ) {
484 $chkcat[$i] = 'fast';
489 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_closure$TPOSTLBL[@]?$/o ) {
491 $chkcat[$i] = 'closure';
494 $closurechk{$1} = $i;
496 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_srt$TPOSTLBL[@]?$/o ) {
503 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_ct$TPOSTLBL[@]?$/o ) {
505 $chkcat[$i] = 'data';
508 } elsif ( /^$TUS[@]?ghc.*c_ID$TPOSTLBL/o ) {
510 $chkcat[$i] = 'consist';
512 } elsif ( /^($TUS[@]?__gnu_compiled_c|gcc2_compiled\.)$TPOSTLBL/o ) {
515 } elsif ( /^$TUS[A-Za-z0-9_]+\.\d+$TPOSTLBL[@]?$/o
516 || /^$TUS[@]?.*_CAT$TPOSTLBL[@]?$/o # PROF: _entryname_CAT
517 || /^$TUS[@]?.*_done$TPOSTLBL[@]?$/o # PROF: _module_done
518 || /^$TUS[@]?_module_registered$TPOSTLBL[@]?$/o # PROF: _module_registered
521 $chkcat[$i] = 'data';
524 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
529 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
534 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_cc(s)?$TPOSTLBL[@]?$/o ) {
535 # all CC_ symbols go in the data section...
537 $chkcat[$i] = 'data';
540 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(alt|dflt)$TPOSTLBL[@]?$/o ) {
542 $chkcat[$i] = 'misc';
545 #$chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
547 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_vtbl$TPOSTLBL[@]?$/o ) {
549 $chkcat[$i] = 'vector';
554 # As a temporary solution for compiling "foreign export" declarations,
555 # we use global variables to pass arguments from C to STG land.
556 # These declarations live in the .hc file and not in the generated C
557 # stub file, so we let them pass through here.
558 } elsif ( /^[\t ]+\.comm[\t ]+__fexp_.*$/ ) {
560 $chkcat[$i] = 'data';
563 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
564 && /^(_uname|uname|stat|fstat):/ ) {
565 # for some utterly bizarre reason, this platform
566 # likes to drop little local C routines with these names
567 # into each and every .o file that #includes the
568 # relevant system .h file. Yuck. We just don't
569 # tolerate them in .hc files (which we are processing
570 # here). If you need to call one of these things from
571 # Haskell, make a call to your own C wrapper, then
572 # put that C wrapper (which calls one of these) in a
573 # plain .c file. WDP 95/12
575 $chkcat[$i] = 'toss';
578 } elsif ( /^$TUS[@]?[A-Za-z0-9_]/o
579 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
581 && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
582 || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
585 print "Funny global thing?: $_"
586 unless $KNOWN_FUNNY_THING{$thing}
587 || /^$TUS[@]?stg_.*$TPOSTLBL[@]?$/o # RTS internals
588 || /^$TUS[@]__fexp_.*$TPOSTLBL$/o # foreign export
589 || /^$TUS[@]?__init.*$TPOSTLBL$/o # __init<module>
590 || /^$TUS[@]?.*_btm$TPOSTLBL$/o # large bitmaps
591 || /^$TUS[@]?.*_closure_tbl$TPOSTLBL$/o; # closure tables
593 $chkcat[$i] = 'misc';
594 if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
595 { $chksymb[$i] = $thing; }
596 else { $chksymb[$i] = ''; };
598 } else { # simple line (duplicated at the top)
603 $numchks = $#chk + 1;
605 # the division into chunks is imperfect;
606 # we throw some things over the fence into the next
609 # also, there are things we would like to know
610 # about the whole module before we start spitting
613 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
615 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
617 # Alphas: NB: we start meddling at chunk 1, not chunk 0
618 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
619 # spits a ".quad 0" in after the v first ".rdata"; we
620 # detect this special case (tossing the ".quad 0")!
621 local($magic_rdata_seen) = 0;
623 # HPPAs, MIPSen: also start medding at chunk 1
625 if ($TargetPlatform =~ /^powerpc|^rs6000/) {
626 print OUTASM $T_HDR_toc; # yes, we have to put a .toc
627 # in the beginning of every file!
628 %tocequiv = (); # maps toc symbol number to toc symbol
629 %revtocequiv = (); # maps toc symbol to toc symbol number
630 for ($i = 1; $i < $numchks; $i++) {
631 $chk[$i] =~ s/\[RW\]//g;
632 $chk[$i] =~ s/\[DS\]//g;
633 $chk[$i] =~ s/^\.csect .*\[DS\]$//g;
635 if ( $chkcat[$i] eq 'toc' && $chk[$i] !~ /\.byte/ )
636 #ToDo: instead of all these changes, just fix mangle_powerpc_tailjump and delete/ignore these tocs?
637 { $chk[$i] =~ s/$T_MOVE_DIRVS//g;
638 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_fast\d+)/\t\.tc \1\[TC\],\.\2/;
639 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_entry)\n/\t\.tc \1\[TC\],\.\2\n/;
640 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(ret_\S+)/\t\.tc \1\[TC\],\.\2/;
641 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(alt_\S+)/\t\.tc \1\[TC\],\.\2/;
642 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
644 $tocnumber = $chksymb[$i];
646 $tocsymb =~ s/^LC\.\.\d+:\n//;
647 $tocsymb =~ s/^\t\.tc \S+,(\S+)\n/\1/;
648 $tocequiv{$tocnumber} = $tocsymb;
650 } elsif ( $chkcat[$i] eq 'toc' && $chk[$i] =~ /\.byte/ ) {
651 $chkcat[$i] = 'literal';
656 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
657 $c = $chk[$i]; # convenience copy
659 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
661 # toss all prologue stuff; HPPA is pretty weird
663 $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/;
665 # be slightly paranoid to make sure there's
666 # nothing surprising in there
667 if ( $c =~ /--- BEGIN ---/ ) {
668 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
670 if ($TargetPlatform =~ /^i386-/) {
671 $p =~ s/^\tpushl \%edi\n//;
672 $p =~ s/^\tpushl \%esi\n//;
673 $p =~ s/^\tpushl \%ebx\n//;
674 $p =~ s/^\tsubl \$\d+,\%esp\n//;
675 $p =~ s/^\tmovl \$\d+,\%eax\n\tcall __alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
676 } elsif ($TargetPlatform =~ /^m68k-/) {
677 $p =~ s/^\tlink a6,#-?\d.*\n//;
678 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
679 # The above showed up in the asm code,
680 # so I added it here.
681 # I hope it's correct.
683 $p =~ s/^\tmovel d2,sp\@-\n//;
684 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
685 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
686 } elsif ($TargetPlatform =~ /^mips-/) {
687 # the .frame/.mask/.fmask that we use is the same
688 # as that produced by GCC for miniInterpret; this
689 # gives GDB some chance of figuring out what happened
690 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
691 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
692 $p =~ s/^\t\.(mask|fmask).*\n//g;
693 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
694 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
695 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
696 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
697 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
698 $p =~ s/__FRAME__/$FRAME/;
699 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000/) {
700 $p =~ s/^\tmflr 0\n//;
701 $p =~ s/^\tstm \d+,-\d+\(1\)\n//;
702 $p =~ s/^\tstw? 0,\d+\(1\)\n//g;
703 $p =~ s/^\tstw? 1,\d+\(1\)\n//g; #mc
704 $p =~ s/^\tlw?z 0,0\(1\)\n//g; #mc
705 $p =~ s/^\tstw?u 1,-\d+\(1\)\n//;
706 $p =~ s/^\tstw? \d+,-\d+\(1\)\n//g;
707 $p =~ s/^\tstfd \d+,-\d+\(1\)\n//g;
709 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
712 # HWL HACK: dont die, just print a warning
713 #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
714 # && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
715 die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
716 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
718 # glue together what's left
720 $c =~ s/\n\t\n/\n/; # junk blank line
724 if ( $TargetPlatform =~ /^mips-/ ) {
725 # MIPS: first, this basic sequence may occur "--- END ---" or not
726 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
729 # toss all epilogue stuff; again, paranoidly
730 if ( $c =~ /--- END ---/ ) {
731 if (($r, $e) = split(/--- END ---/, $c)) {
732 if ($TargetPlatform =~ /^i386-/) {
734 $e =~ s/^\tpopl \%edi\n//;
735 $e =~ s/^\tpopl \%esi\n//;
736 $e =~ s/^\tpopl \%ecx\n//;
737 $e =~ s/^\taddl \$\d+,\%esp\n//;
738 $e =~ s/^\tsubl \$-\d+,\%esp\n//;
739 } elsif ($TargetPlatform =~ /^m68k-/) {
740 $e =~ s/^\tunlk a6\n//;
742 } elsif ($TargetPlatform =~ /^mips-/) {
743 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
744 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
745 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
746 $e =~ s/^\tj\t\$31\n//;
747 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
748 $e =~ s/^\taddi 1,1,\d+\n//;
749 $e =~ s/^\tcal 1,\d+\(1\)\n//;
750 $e =~ s/^\tlw?z? \d+,\d+\(1\)\n//;
751 $e =~ s/^\tmtlr 0\n//;
754 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
756 # HWL HACK: dont die, just print a warning
757 #print stderr "HWL: this should die! Epilogue junk?: $e\n" if $e =~ /^\t[^\.]/
758 # && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
759 die "Epilogue junk?: $e\n" if $e =~ /^\t[^\.\n]/
760 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
762 # glue together what's left
764 $c =~ s/\n\t\n/\n/; # junk blank line
768 # On SPARCs, we don't do --- BEGIN/END ---, we just
769 # toss the register-windowing save/restore/ret* instructions
771 if ( $TargetPlatform =~ /^sparc-/ ) {
772 $c =~ s/^\t(save.*|restore|ret|retl)\n//g;
773 # throw away PROLOGUE comments
774 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
777 # On Alphas, the prologue mangling is done a little later (below)
779 # toss all calls to __DISCARD__
780 $c =~ s/^\t(call|jbsr|jal)\s+$TUS[@]?__DISCARD__\n//go;
782 # MIPS: that may leave some gratuitous asm macros around
783 # (no harm done; but we get rid of them to be tidier)
784 $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/
785 if $TargetPlatform =~ /^mips-/;
787 # toss stack adjustment after DoSparks
788 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
789 if $TargetPlatform =~ /^m68k-/; # this looks old...
791 if ( $TargetPlatform =~ /^alpha-/ &&
792 ! $magic_rdata_seen &&
793 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
794 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
795 $magic_rdata_seen = 1;
798 # pick some end-things and move them to the next chunk
800 # pin a funny end-thing on (for easier matching):
801 $c .= 'FUNNY#END#THING';
803 while ( $c =~ /$TMOVEDIRVS[@]?FUNNY#END#THING/o ) { # [@]? is a silly hack to avoid having to use curlies for T_PRE_APP
804 # (this SEGVs perl4 on alphas, you see)
808 # on x86 we try not to copy any directives into a literal
809 # chunk, rather we keep looking for the next real chunk. This
810 # is because we get things like
812 # .globl blah_closure
818 if ( $TargetPlatform =~ /^i386/ && $to_move =~ /$TCOPYDIRVS/ ) {
820 while ( $j < $numchks && $chk[$j] =~ /$T_CONST_LBL/) {
823 if ( $j < $numchks ) {
824 $chk[$j] = $to_move . $chk[$j];
828 elsif ( $i < ($numchks - 1)
829 && ( $to_move =~ /$TCOPYDIRVS/
830 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
831 $chk[$i + 1] = $to_move . $chk[$i + 1];
832 # otherwise they're tossed
835 $c =~ s/$TMOVEDIRVS[@]?FUNNY#END#THING/FUNNY#END#THING/o; # [@]? is a hack (see above)
838 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
840 # toss all prologue stuff, except for loading gp, and the ..ng address
841 if (($p, $r) = split(/^\t\.prologue/, $c)) {
842 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
843 $c = $keep . "..ng:\n";
845 print STDERR "malformed code block ($ent)?\n"
848 $c .= "\t.frame \$30,0,\$26,0\n\t.prologue" . $r;
851 $c =~ s/FUNNY#END#THING//;
853 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
855 $chk[$i] = $c; # update w/ convenience copy
858 if ( $TargetPlatform =~ /^alpha-/ ) {
859 # print out the header stuff first
860 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
861 print OUTASM $chk[0];
863 } elsif ( $TargetPlatform =~ /^hppa/ ) {
864 print OUTASM $chk[0];
866 } elsif ( $TargetPlatform =~ /^mips-/ ) {
867 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
869 # get rid of horrible "<dollar>Revision: .*$" strings
870 local(@lines0) = split(/\n/, $chk[0]);
872 while ( $z <= $#lines0 ) {
873 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
876 while ( $z <= $#lines0 ) {
878 last if $lines0[$z] =~ /[,\t]0x0$/;
884 $chk[0] = join("\n", @lines0);
885 $chk[0] =~ s/\n\n+/\n/;
886 print OUTASM $chk[0];
889 # print out all the literal strings next
890 for ($i = 0; $i < $numchks; $i++) {
891 if ( $chkcat[$i] eq 'literal' ) {
892 print OUTASM $T_HDR_literal, $chk[$i];
893 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
895 $chkcat[$i] = 'DONE ALREADY';
899 # on the HPPA, print out all the bss next
900 if ( $TargetPlatform =~ /^hppa/ ) {
901 for ($i = 1; $i < $numchks; $i++) {
902 if ( $chkcat[$i] eq 'bss' ) {
903 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
904 print OUTASM $chk[$i];
906 $chkcat[$i] = 'DONE ALREADY';
911 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
912 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
914 next if $chkcat[$i] eq 'DONE ALREADY';
916 if ( $chkcat[$i] eq 'misc' ) {
917 if ($chk[$i] ne '') {
918 print OUTASM $T_HDR_misc;
919 if ($TargetPlatform =~ /^powerpc-|^rs6000/) {
920 $chksymb[$i] =~ s/://;
921 #new if ($chksymb[$i] =~ /ret.*upd/ || $KNOWN_FUNNY_THING{$chksymb[$i]}
922 #new || $chksymb[$i] =~ /^$.{T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o )
923 #new { print OUTASM "\t\.globl $chksymb[$i]\n"; }
924 # if ($chksymb[$i] ne '' && $chksymb[$i] !~ /ret_[a-z]/ && $chksymb[$i] !~ /djn_[a-z]/)
925 if ($chksymb[$i] ne '')
926 { print OUTASM "\t\.globl \.$chksymb[$i]\n"; };
927 if ($chk[$i] =~ /TOC\[tc0\], 0\n/)
928 { ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$i]); $printDS = 1;}
929 else { $r = $chk[$i]; $printDS = 0; };
930 $chk[$i] = &mangle_powerpc_tailjump($r);
932 &print_doctored($chk[$i], 0);
933 if ($TargetPlatform =~ /^powerpc-|^rs6000-/ && $printDS) {
934 #ok if ($chksymb[$i] !~ /\__init_Main/) {
935 print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
936 print OUTASM "${p}TOC[tc0], 0\n";
941 } elsif ( $chkcat[$i] eq 'toss' ) {
942 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
944 } elsif ( $chkcat[$i] eq 'data' ) {
945 if ($chk[$i] ne '') {
946 print OUTASM $T_HDR_data;
947 print OUTASM $chk[$i];
950 } elsif ( $chkcat[$i] eq 'consist' ) {
951 if ( $chk[$i] =~ /$T_hsc_cc_PAT/o ) {
952 local($consist) = "$1.$2.$3";
954 $consist =~ s/\//./g;
956 $consist =~ s/[^A-Za-z0-9_.]/ZZ/g; # ToDo: properly?
958 # Using a cygnus-2.7-96q4 gcc build on hppas, the
959 # consistency chunk for ghc_cc_ID often (but not always!)
960 # gets lumped with a bunch of .IMPORT directives containing info on
961 # the code or data space nature of external symbols. We can't
962 # toss these, so once the consistency ID has been turned into
963 # a representable symbol, we substitute it for the symbol
964 # that the string was attached to in the first place (ghc_cc_ID.)
965 # (The original string is also substituted away.)
967 # This change may affect the code output on other platforms in
968 # adverse ways, hence we restrict this hack hppa targets only.
971 if ( $TargetPlatform =~ /^hppa/ ) {
972 $chk[$i] =~ s/^$TUS[@]?ghc.*c_ID$TPOSTLBL/$consist/o;
973 $chk[$i] =~ s/\t$T_hsc_cc_PAT/$T_HDR_misc/o;
974 $consist = $chk[$i]; #clumsily
976 print OUTASM $T_HDR_consist, "${consist}${T_POST_LBL}\n";
978 } elsif ( $TargetPlatform !~ /^(mips)-/ ) { # we just don't try in those case (ToDo)
979 # on mips: consistency string is just a v
980 # horrible bunch of .bytes,
981 # which I am too lazy to sort out (WDP 95/05)
983 print STDERR "Couldn't grok consistency: ", $chk[$i];
986 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
987 # we can just re-constitute this one...
988 # NB: we emit _three_ underscores no matter what,
989 # so ghc-split doesn't have to care.
990 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
992 } elsif ( $chkcat[$i] eq 'closure'
993 || $chkcat[$i] eq 'srt'
994 || $chkcat[$i] eq 'infotbl'
995 || $chkcat[$i] eq 'slow'
996 || $chkcat[$i] eq 'fast' ) { # do them in that order
997 $symb = $chksymb[$i];
1000 if ( defined($closurechk{$symb}) ) {
1001 print OUTASM $T_HDR_closure;
1002 print OUTASM $chk[$closurechk{$symb}];
1003 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
1007 if ( defined($srtchk{$symb}) ) {
1008 print OUTASM $T_HDR_srt;
1009 print OUTASM $chk[$srtchk{$symb}];
1010 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
1014 if ( defined($infochk{$symb}) ) {
1016 print OUTASM $T_HDR_info;
1017 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1018 if ( !defined($slowchk{$symb}) && defined($fastchk{$symb}) ) {
1019 $fastname = $chk[$fastchk{$symb}];
1020 $fastname =~ s/([_A-Za-z]*_fast\d+):.*(.*\n)*/\1/;
1021 $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long $fastname/;
1023 $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
1024 $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
1025 print OUTASM $chk[$infochk{$symb}];
1027 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
1029 # entry code will be put here!
1031 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
1035 if ( defined($slowchk{$symb}) ) {
1037 # teach it to drop through to the fast entry point:
1038 $c = $chk[$slowchk{$symb}];
1040 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1041 ($p, $r) = split(/TOC\[tc0\], 0\n/, $c);
1042 if ($symb =~ /^[_A-Z]/)
1044 print OUTASM "\t\.globl \.${chksymb[$i]}_entry\n";
1045 print OUTASM "\.csect ${symb}_entry[DS]\n";
1046 print OUTASM "${p}TOC[tc0], 0\n";
1048 $r =~ s/\.csect \.text\[PR\]\n//; # todo: properly - andre
1049 $c = &mangle_powerpc_tailjump($r);
1052 if ( defined($fastchk{$symb}) ) {
1053 if ( $TargetPlatform =~ /^alpha-/ ) {
1054 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1055 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1056 $c =~ s/^\s+ldil.*\n\s+ldo.*\n\s+bv.*\n(.*\n)?\s+\.EXIT/$1\t.EXIT/;
1057 } elsif ( $TargetPlatform =~ /^i386-/ ) {
1058 # Reg alloc depending, gcc generated code may jump to the fast entry point via
1059 # a number of registers.
1060 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edx\n\tjmp \*\%edx\n//;
1061 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%ecx\n\tjmp \*\%ecx\n//;
1062 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%eax\n\tjmp \*\%eax\n//;
1063 # The next two only apply if we're not stealing %esi or %edi.
1064 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%esi\n\tjmp \*\%esi\n// if ($StolenX86Regs < 3);
1065 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edi\n\tjmp \*\%edi\n// if ($StolenX86Regs < 4);
1066 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1067 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1068 } elsif ( $TargetPlatform =~ /^m68k-/ ) {
1069 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n\tnop\n//;
1070 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n//;
1071 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ ) {
1072 $c =~ s/^\tb \.${T_US}${symb}_fast\d+\n//;
1073 } elsif ( $TargetPlatform =~ /^sparc-/ ) {
1074 $c =~ s/^\tcall\s+${T_US}${symb}_fast\d+,.*\n\t\s*nop\n//;
1075 $c =~ s/^\tcall\s+${T_US}${symb}_fast\d+,.*\n(\t\s*[a-z].*\n)/$1/;
1077 print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n";
1081 if ( $TargetPlatform !~ /^(alpha-|hppa|mips-)/ ) {
1082 # On alphas, hppa: no very good way to look for "dangling"
1083 # references to fast-entry point.
1084 # (questionable re hppa and mips...)
1085 print STDERR "still has jump to fast entry point:\n$c"
1086 if $c =~ /$TUS[@]?$symb[@]?_fast/; # NB: paranoia
1089 print OUTASM $T_HDR_entry;
1091 &print_doctored($c, 1); # NB: the 1!!!
1093 $chkcat[$slowchk{$symb}] = 'DONE ALREADY';
1097 if ( defined($fastchk{$symb}) ) {
1098 $c = $chk[$fastchk{$symb}];
1099 if ( ! defined($slowchk{$symb})
1100 # ToDo: the || clause can go once we're no longer
1101 # concerned about producing exactly the same output as before
1102 #OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/
1104 print OUTASM $T_HDR_fast;
1106 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1107 local(@lbls) = split(/:/, $c);
1108 $fullname = $lbls[0];
1109 $fullname =~ s/$T_MOVE_DIRVS//g;
1110 if ( $fullname =~ /^[A-Z]/)
1111 { print OUTASM "\t\.globl \.${fullname}\n";
1113 # print OUTASM "\t\.lglobl \.${fullname}\n"; #todo: rm - andre
1115 $c =~ s/((.*\n)*)\t.long \S+, TOC\[tc0\], 0\n\.csect \.text\[PR\]\n((.*\n)*)/\1\3/;
1116 $c = &mangle_powerpc_tailjump($c);
1118 &print_doctored($c, 0);
1119 $chkcat[$fastchk{$symb}] = 'DONE ALREADY';
1122 } elsif ( $chkcat[$i] eq 'vector'
1123 || $chkcat[$i] eq 'direct' ) { # do them in that order
1124 $symb = $chksymb[$i];
1127 if ( defined($vectorchk{$symb}) ) {
1128 print OUTASM $T_HDR_vector;
1129 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1130 if ( $symb =~ /^[A-Z]/) {
1131 print OUTASM "\t\.globl \.${symb}_vtbl\n";
1132 print OUTASM "\t\.globl ${symb}_vtbl\n";
1134 $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
1135 print OUTASM ".${symb}_vtbl:\n";
1136 print OUTASM $chk[$vectorchk{$symb}];
1138 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1140 # direct return code will be put here!
1141 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1143 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1144 # Alphas: the commented nop is for the splitter, to ensure
1145 # that no module ends with a label as the very last
1146 # thing. (The linker will adjust the label to point
1147 # to the first code word of the next module linked in,
1148 # even if alignment constraints cause the label to move!)
1150 print OUTASM "\t# nop\n";
1153 } elsif ( $chkcat[$i] eq 'toc' ) {
1154 # silly optimisation to print tocs, since they come in groups...
1155 print OUTASM $T_HDR_toc;
1157 while ($chkcat[$j] eq 'toc')
1158 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1161 print OUTASM $chk[$j];
1163 $chkcat[$j] = 'DONE ALREADY';
1168 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1172 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1174 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1175 print OUTASM ".csect .text[PR]\n_section_.text:\n.csect .data[RW]\n\t.long _section_.text\n"
1179 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1180 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1185 sub mash_hppa_prologue { # OK, epilogue, too
1188 # toss all prologue stuff
1189 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1191 # Lie about our .CALLINFO
1192 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1199 # toss all epilogue stuff
1200 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1202 # Sorry; we moved the _info stuff to the code segment.
1203 s/_info,DATA/_info,CODE/g;
1210 sub print_doctored {
1211 local($_, $need_fallthru_patch) = @_;
1213 if ( $TargetPlatform !~ /^i386-/
1214 || ! /^\t[a-z]/ # no instructions in here, apparently
1215 || /^${T_US}__init_[A-Za-z0-9_]+${T_POST_LBL}/) {
1219 # OK, must do some x86 **HACKING**
1221 local($entry_patch) = '';
1222 local($exit_patch) = '';
1224 # gotta watch out for weird instructions that
1225 # invisibly smash various regs:
1226 # rep* %ecx used for counting
1227 # scas* %edi used for destination index
1228 # cmps* %e[sd]i used for indices
1229 # loop* %ecx used for counting
1234 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1236 # * GCC used an "STG reg" for its own purposes
1238 # * some secret uses of machine reg, requiring STG reg
1239 # to be saved/restored
1241 # The most dangerous "GCC uses" of an "STG reg" are when
1242 # the reg holds the target of a jmp -- it's tricky to
1243 # insert the patch-up code before we get to the target!
1244 # So here we change the jmps:
1246 # --------------------------------------------------------
1247 # it can happen that we have jumps of the form...
1248 # jmp *<something involving %esp>
1250 # jmp <something involving another naughty register...>
1252 # a reasonably-common case is:
1254 # movl $_blah,<bad-reg>
1258 # the short form may tickle perl bug:
1259 # s/^\tmovl \$${T_US}(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp $T_US$1/g;
1260 s/^\tmovl \$${T_US}(.*),\%eax\n\tjmp \*\%eax/\tjmp $T_US$1/g;
1261 s/^\tmovl \$${T_US}(.*),\%ebx\n\tjmp \*\%ebx/\tjmp $T_US$1/g;
1262 s/^\tmovl \$${T_US}(.*),\%ecx\n\tjmp \*\%ecx/\tjmp $T_US$1/g;
1263 s/^\tmovl \$${T_US}(.*),\%edx\n\tjmp \*\%edx/\tjmp $T_US$1/g;
1265 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1266 s/^\tmovl (.*),\%esi\n\tjmp \*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1267 s/^\tjmp \*(-?\d*)\((.*\%esi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1268 s/^\tjmp \*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1269 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1270 if /(jmp|call) .*\%esi/;
1272 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1273 s/^\tmovl (.*),\%edi\n\tjmp \*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1274 s/^\tjmp \*(-?\d*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1275 s/^\tjmp \*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1276 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1277 if /(jmp|call) .*\%edi/;
1280 # OK, now we can decide what our patch-up code is going to
1283 # Offsets into register table - you'd better update these magic
1284 # numbers should you change its contents!
1285 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1286 local($OFFSET_Hp)=92;
1288 # Note funky ".=" stuff; we're *adding* to these _patch guys
1289 if ( $StolenX86Regs <= 2
1290 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1291 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1292 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1294 # nothing for call_{entry,exit} because %esi is callee-save
1296 if ( $StolenX86Regs <= 3
1297 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1298 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1299 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1301 # nothing for call_{entry,exit} because %edi is callee-save
1304 # --------------------------------------------------------
1305 # next, here we go with non-%esp patching!
1307 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1309 # Before calling GC we must set up the exit condition before the call
1310 # and entry condition when we come back
1312 # fix _all_ non-local jumps:
1314 s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1315 s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1317 s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
1319 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1320 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1322 if ($StolenX86Regs == 2 ) {
1323 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1324 if /^\t(jmp|call) .*\%e(si|di)/;
1325 } elsif ($StolenX86Regs == 3 ) {
1326 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1327 if /^\t(jmp|call) .*\%edi/;
1330 # --------------------------------------------------------
1331 # that's it -- print it
1333 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1337 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1338 print OUTASM $exit_patch;
1339 # ToDo: make it not print if there is a "jmp" at the end
1345 sub init_FUNNY_THINGS {
1346 %KNOWN_FUNNY_THING = (
1348 # "${T_US}stg_.*{T_POST_LBL}", 1,
1353 The following table reversal is used for both info tables and return
1354 vectors. In both cases, we remove the first entry from the table,
1355 reverse the table, put the label at the end, and paste some code
1356 (that which is normally referred to by the first entry in the table)
1357 right after the table itself. (The code pasting is done elsewhere.)
1361 local($symb, $tbl, $discard1) = @_;
1363 local($before) = '';
1365 local(@imports) = (); # hppa only
1368 local(@lines) = split(/\n/, $tbl);
1369 local($i, $j); #local ($i, $extra, $words_to_pad, $j);
1371 # see comment in mangleAsm as to why this silliness is needed.
1372 local($TDOTWORD) = ${T_DOT_WORD};
1373 local($TDOTGLOBAL) = ${T_DOT_GLOBAL};
1374 local($TUS) = ${T_US};
1375 local($TPOSTLBL) = ${T_POST_LBL};
1377 # Deal with the header...
1378 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?$TDOTWORD\s+/o; $i++) {
1379 $label .= $lines[$i] . "\n",
1380 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info$TPOSTLBL[@]?$/o
1381 || $lines[$i] =~ /$TDOTGLOBAL/o
1382 || $lines[$i] =~ /^$TUS[@]?\S+_vtbl$TPOSTLBL[@]?$/o;
1384 $before .= $lines[$i] . "\n"; # otherwise...
1387 # Grab the table data...
1388 if ( $TargetPlatform !~ /^hppa/ ) {
1389 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?$TDOTWORD\s+/o; $i++) {
1390 push(@words, $lines[$i]);
1392 } else { # hppa weirdness
1393 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+($TDOTWORD|\.IMPORT)/; $i++) {
1394 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1395 push(@imports, $lines[$i]);
1397 # We don't use HP's ``function pointers''
1398 # We just use labels in code space, like normal people
1399 $lines[$i] =~ s/P%//;
1400 push(@words, $lines[$i]);
1405 # now throw away the first word (SRT) iff it is empty.
1406 # The .zero business is for Linux/ELF.
1407 # The .skip business is for Sparc/Solaris/ELF.
1408 # The .blockz business is for HPPA.
1409 if ($discard1 && $words[0] =~ /^\t?($TDOTWORD\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1413 # Padding removed to reduce code size and improve performance on Pentiums.
1415 # for 486-cache-friendliness, we want our tables aligned
1416 # on 16-byte boundaries (.align 4). Let's pad:
1417 # $extra = ($#words + 1) % 4;
1418 # $words_to_pad = ($extra == 0) ? 0 : 4 - $extra;
1419 # for ($j = 0; $j < $words_to_pad; $j++) { push(@words, "\t${T_DOT_WORD} 0"); }
1421 for (; $i <= $#lines; $i++) {
1422 $after .= $lines[$i] . "\n";
1425 # Alphas:If we have anonymous text (not part of a procedure), the
1426 # linker may complain about missing exception information. Bleh.
1427 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1428 $before = "\t.ent $1\n" . $before;
1429 $after .= "\t.end $1\n";
1433 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1434 . join("\n", @words) . "\n"
1437 # print STDERR "before=$before\n";
1438 # print STDERR "label=$label\n";
1439 # print STDERR "words=",(reverse @words),"\n";
1440 # print STDERR "after=$after\n";
1446 The HP is a major nuisance. The threaded code mangler moved info
1447 tables from data space to code space, but unthreaded code in the RTS
1448 still has references to info tables in data space. Since the HP
1449 linker is very precise about where symbols live, we need to patch the
1450 references in the unthreaded RTS as well.
1453 sub mini_mangle_asm_hppa {
1454 local($in_asmf, $out_asmf) = @_;
1456 open(INASM, "< $in_asmf")
1457 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1458 open(OUTASM,"> $out_asmf")
1459 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1462 s/_info,DATA/_info,CODE/; # Move _info references to code space
1468 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1469 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1475 sub mini_mangle_asm_powerpc {
1476 local($in_asmf, $out_asmf) = @_;
1478 open(INASM, "< $in_asmf")
1479 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1480 open(OUTASM,"> $out_asmf")
1481 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1484 s/long _PRStart/long ._PRStart/;
1485 s/long _PRIn_/long ._PRIn_/;
1486 s/long _Dummy_(\S+)_entry/long ._Dummy_\1_entry/;
1487 s/long _PRMarking_MarkNextRoot\[DS\]/long ._PRMarking_MarkNextRoot/;
1488 s/long _PRMarking_MarkNextCAF\[DS\]/long ._PRMarking_MarkNextCAF/;
1489 s/long _PRMarking_MarkNextAStack\[DS\]/long ._PRMarking_MarkNextAStack/;
1490 s/long _PRMarking_MarkNextBStack\[DS\]/long ._PRMarking_MarkNextBStack/;
1491 s/\.tc EnterNodeCode\[TC]\,EnterNodeCode\[DS\]/\.tc EnterNodeCode\[TC]\,.EnterNodeCode/; # CONC
1492 s/\.tc CheckHeapCode\[TC]\,CheckHeapCode\[DS\]/\.tc CheckHeapCode\[TC]\,.CheckHeapCode/; # CONC
1497 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1498 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1501 sub mangle_powerpc_tailjump {
1503 local($maybe_more) = 1;
1504 while (($c =~ /\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n/) && $maybe_more)
1507 $lcsymb =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1508 # the checks for r1 and r2 are mostly paranoia...
1510 $r1 =~ s/(.*\n)*\tlw?z? (\d+),LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1512 $r2 =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr (\d+)\n\tbctr\n(.*\n)*/\3/;
1515 $c =~ s/((.*\n)*)\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n((.*\n)*)/\1\tb $tocequiv{$lcsymb}\n\3/;
1523 sub tidy_up_and_die {
1524 local($return_val, $msg) = @_;
1526 exit (($return_val == 0) ? 0 : 1);