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 ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd2|nextstep3|cygwin32)$/ ) {
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_srt = "\.data\n\t\.align 2\n";
130 $T_HDR_info = "\.text\n\t\.align 2\n"; # NB: requires padding
131 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
132 $T_HDR_fast = "\.text\n\t\.align 2,0x90\n";
133 $T_HDR_vector = "\.text\n\t\.align 2\n"; # NB: requires padding
134 $T_HDR_direct = "\.text\n\t\.align 2,0x90\n";
136 #--------------------------------------------------------#
137 } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|freebsd3)$/ ) {
139 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
140 $T_US = ''; # _ if symbols have an underscore on the front
141 $T_PRE_APP = # regexp that says what comes before APP/NO_APP
142 ($TargetPlatform =~ /-(linux|freebsd3)$/) ? '#' : '/' ;
143 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
145 $T_X86_PRE_LLBL_PAT = '\.L';
146 $T_X86_PRE_LLBL = '.L';
147 $T_X86_BADJMP = '^\tjmp [^\.\*]';
149 $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)';
150 $T_COPY_DIRVS = '\.(globl)';
152 if ( $TargetPlatform =~ /freebsd3/ ) {
153 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
155 $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
158 $T_DOT_WORD = '\.(long|value|byte|zero)';
159 $T_DOT_GLOBAL = '\.globl';
160 $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11)
161 $T_HDR_misc = "\.text\n\t\.align 4\n";
162 $T_HDR_data = "\.data\n\t\.align 4\n"; # ToDo: change align??
163 $T_HDR_consist = "\.text\n";
164 $T_HDR_closure = "\.data\n\t\.align 4\n"; # ToDo: change align?
165 $T_HDR_srt = "\.data\n\t\.align 4\n"; # ToDo: change align?
166 $T_HDR_info = "\.text\n\t\.align 16\n"; # NB: requires padding
167 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
168 $T_HDR_fast = "\.text\n\t\.align 16\n";
169 $T_HDR_vector = "\.text\n\t\.align 16\n"; # NB: requires padding
170 $T_HDR_direct = "\.text\n\t\.align 16\n";
172 #--------------------------------------------------------#
173 } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
175 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
176 $T_US = '_'; # _ if symbols have an underscore on the front
177 $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
178 $T_CONST_LBL = '^LC(\d+):$';
181 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
182 $T_COPY_DIRVS = '\.(globl|proc|stab)';
183 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
185 $T_DOT_WORD = '\.long';
186 $T_DOT_GLOBAL = '\.globl';
187 $T_HDR_literal = "\.text\n\t\.even\n";
188 $T_HDR_misc = "\.text\n\t\.even\n";
189 $T_HDR_data = "\.data\n\t\.even\n";
190 $T_HDR_consist = "\.text\n";
191 $T_HDR_closure = "\.data\n\t\.even\n";
192 $T_HDR_srt = "\.data\n\t\.even\n";
193 $T_HDR_info = "\.text\n\t\.even\n";
194 $T_HDR_entry = "\.text\n\t\.even\n";
195 $T_HDR_fast = "\.text\n\t\.even\n";
196 $T_HDR_vector = "\.text\n\t\.even\n";
197 $T_HDR_direct = "\.text\n\t\.even\n";
199 #--------------------------------------------------------#
200 } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
202 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
203 $T_US = ''; # _ if symbols have an underscore on the front
204 $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
205 $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
208 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
209 $T_COPY_DIRVS = '\.(globl|ent)';
211 $T_hsc_cc_PAT = 'I WAS TOO LAZY TO DO THIS BIT (WDP 95/05)';
212 $T_DOT_WORD = '\.word';
213 $T_DOT_GLOBAL = '^\t\.globl';
214 $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
215 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
216 $T_HDR_data = "\t\.data\n\t\.align 2\n";
217 $T_HDR_consist = 'TOO LAZY TO DO THIS TOO';
218 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
219 $T_HDR_srt = "\t\.data\n\t\.align 2\n";
220 $T_HDR_info = "\t\.text\n\t\.align 2\n";
221 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
222 $T_HDR_fast = "\t\.text\n\t\.align 2\n";
223 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
224 $T_HDR_direct = "\t\.text\n\t\.align 2\n";
226 #--------------------------------------------------------#
227 } elsif ( $TargetPlatform =~ /^powerpc-.*|^rs6000-.*/ ) {
229 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
230 $T_US = ''; # _ if symbols have an underscore on the front
231 $T_PRE_APP = 'NOT APPLICABLE'; # regexp that says what comes before APP/NO_APP
232 $T_CONST_LBL = 'NOT APPLICABLE'; # regexp for what such a lbl looks like
235 $T_MOVE_DIRVS = '^(\s*(\.toc|\.align \d+|\.csect \S+|\t\.?l?globl \S+)\n)';
236 $T_COPY_DIRVS = '\.(l?globl)';
238 $T_hsc_cc_PAT = '\.byte.*\)(hsc|cc) (.*)"\n\t\.byte \d+\n\t\.byte "(.*)"\n\t\.byte \d+';
239 $T_DOT_WORD = '\.long';
240 $T_DOT_GLOBAL = '\.globl';
241 $T_HDR_toc = "\.toc\n";
242 $T_HDR_literal = "\.csect .data[RW]\n\t\.align 2\n"; #not RO!?
243 $T_HDR_misc = "# misc\n\.csect \.text[PR]\n\t\.align 2\n";
244 $T_HDR_data = "# data\n\.csect \.data[RW]\n\t\.align 2\n";
245 $T_HDR_consist = "# consist\n\.csect \.data[RW]\n\t\.align 2\n";
246 $T_HDR_closure = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
247 $T_HDR_srt = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
248 $T_HDR_info = "# info\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
249 $T_HDR_entry = "# entry\n\.csect \.text[PR]\n\t\.align 2\n";
250 $T_HDR_fast = "# fast\n\.csect \.text[PR]\n\t\.align 2\n";
251 $T_HDR_vector = "# vector\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
252 $T_HDR_direct = "# direct\n";
254 #--------------------------------------------------------#
255 } elsif ( $TargetPlatform =~ /^sparc-.*-solaris2/ ) {
257 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
258 $T_US = ''; # _ if symbols have an underscore on the front
259 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
260 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
263 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\.section.*|\s+\.type.*|\s+\.size.*)\n)';
264 $T_COPY_DIRVS = '\.(global|proc|stab)';
266 $T_hsc_cc_PAT = '\.asciz.*\)(hsc|cc) (.*)\\\\t(.*)"';
267 $T_DOT_WORD = '\.(word|byte|half|skip)';
268 $T_DOT_GLOBAL = '^\t\.global';
269 $T_HDR_literal = "\.text\n\t\.align 8\n";
270 $T_HDR_misc = "\.text\n\t\.align 4\n";
271 $T_HDR_data = "\.data\n\t\.align 8\n";
272 $T_HDR_consist = "\.text\n";
273 $T_HDR_closure = "\.data\n\t\.align 4\n";
274 $T_HDR_srt = "\.data\n\t\.align 4\n";
275 $T_HDR_info = "\.text\n\t\.align 4\n";
276 $T_HDR_entry = "\.text\n\t\.align 4\n";
277 $T_HDR_fast = "\.text\n\t\.align 4\n";
278 $T_HDR_vector = "\.text\n\t\.align 4\n";
279 $T_HDR_direct = "\.text\n\t\.align 4\n";
281 #--------------------------------------------------------#
282 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
284 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
285 $T_US = '_'; # _ if symbols have an underscore on the front
286 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
287 $T_CONST_LBL = '^LC(\d+):$';
290 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
291 $T_COPY_DIRVS = '\.(global|proc|stab)';
292 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
294 $T_DOT_WORD = '\.word';
295 $T_DOT_GLOBAL = '^\t\.global';
296 $T_HDR_literal = "\.text\n\t\.align 8\n";
297 $T_HDR_misc = "\.text\n\t\.align 4\n";
298 $T_HDR_data = "\.data\n\t\.align 8\n";
299 $T_HDR_consist = "\.text\n";
300 $T_HDR_closure = "\.data\n\t\.align 4\n";
301 $T_HDR_srt = "\.data\n\t\.align 4\n";
302 $T_HDR_info = "\.text\n\t\.align 4\n";
303 $T_HDR_entry = "\.text\n\t\.align 4\n";
304 $T_HDR_fast = "\.text\n\t\.align 4\n";
305 $T_HDR_vector = "\.text\n\t\.align 4\n";
306 $T_HDR_direct = "\.text\n\t\.align 4\n";
308 #--------------------------------------------------------#
310 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
315 print STDERR "T_STABBY: $T_STABBY\n";
316 print STDERR "T_US: $T_US\n";
317 print STDERR "T_PRE_APP: $T_PRE_APP\n";
318 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
319 print STDERR "T_POST_LBL: $T_POST_LBL\n";
320 if ( $TargetPlatform =~ /^i386-/ ) {
321 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
322 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
323 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
325 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
326 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
327 print STDERR "T_hsc_cc_PAT: $T_hsc_cc_PAT\n";
328 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
329 print STDERR "T_HDR_literal: $T_HDR_literal\n";
330 print STDERR "T_HDR_misc: $T_HDR_misc\n";
331 print STDERR "T_HDR_data: $T_HDR_data\n";
332 print STDERR "T_HDR_consist: $T_HDR_consist\n";
333 print STDERR "T_HDR_closure: $T_HDR_closure\n";
334 print STDERR "T_HDR_info: $T_HDR_info\n";
335 print STDERR "T_HDR_entry: $T_HDR_entry\n";
336 print STDERR "T_HDR_fast: $T_HDR_fast\n";
337 print STDERR "T_HDR_vector: $T_HDR_vector\n";
338 print STDERR "T_HDR_direct: $T_HDR_direct\n";
344 %************************************************************************
346 \subsection{Mangle away}
348 %************************************************************************
352 local($in_asmf, $out_asmf) = @_;
354 # multi-line regexp matching:
359 &init_TARGET_STUFF();
360 &init_FUNNY_THINGS();
362 # perl4 on alphas SEGVs when give ${foo} substitutions in patterns.
363 # To avoid them we declare some locals that allows to avoid using curlies.
364 local($TUS) = ${T_US};
365 local($TPOSTLBL) = ${T_POST_LBL};
366 local($TMOVEDIRVS) = ${T_MOVE_DIRVS};
367 local($TPREAPP) = ${T_PRE_APP};
368 local($TCOPYDIRVS) = ${T_COPY_DIRVS};
369 local($TDOTWORD) = ${T_DOT_WORD};
371 open(INASM, "< $in_asmf")
372 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
373 open(OUTASM,"> $out_asmf")
374 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
376 # read whole file, divide into "chunks":
377 # record some info about what we've found...
379 @chk = (); # contents of the chunk
380 $numchks = 0; # number of them
381 @chkcat = (); # what category of thing in each chunk
382 @chksymb = (); # what symbol(base) is defined in this chunk
383 %slowchk = (); # ditto, its regular "slow" entry code
384 %fastchk = (); # ditto, fast entry code
385 %closurechk = (); # ditto, the (static) closure
386 %srtchk = (); # ditto, its SRT (for top-level things)
387 %infochk = (); # given a symbol base, say what chunk its info tbl is in
388 %vectorchk = (); # ditto, return vector table
389 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
391 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
394 next if $T_STABBY && /^\.stab.*$TUS[@]?__stg_split_marker/o;
395 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
396 next if /$TPREAPP(NO_)?APP/o;
397 next if /^;/ && $TargetPlatform =~ /^hppa/;
399 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|powerpc|rs6000)-/;
401 last if /^_section_\.text:$/ && $TargetPlatform =~ /^powerpc-|^rs6000-/;
403 if ( $TargetPlatform =~ /^mips-/
404 && /^\t\.(globl \S+ \.text|comm\t)/ ) {
405 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
407 # As a temporary solution for compiling "foreign export" declarations,
408 # we use global variables to pass arguments from C to STG land.
409 # These declarations live in the .hc file and not in the generated C
410 # stub file, so we let them pass through here.
411 } elsif ( /^\t\.comm\t__fexp_.*$/ ) {
413 $chkcat[$i] = 'data';
416 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
417 # duplicated from the bottom
421 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
422 # Alphas: Local labels not to be confused with new chunks
425 # NB: all the rest start with a non-space
427 } elsif ( $TargetPlatform =~ /^mips-/
428 && /^\d+:/ ) { # a funny-looking very-local label
431 } elsif ( /$T_CONST_LBL/o ) {
433 $chkcat[$i] = 'literal';
436 } elsif ( /^$TUS[@]?__stg_split_marker(\d+)$TPOSTLBL[@]?$/o ) {
438 $chkcat[$i] = 'splitmarker';
441 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_info$TPOSTLBL[@]?$/o ) {
444 $chkcat[$i] = 'infotbl';
445 $chksymb[$i] = $symb;
447 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
449 $infochk{$symb} = $i;
451 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(entry|ret)$TPOSTLBL[@]?$/o ) {
453 $chkcat[$i] = 'slow';
458 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_fast\d*$TPOSTLBL[@]?$/o ) {
460 $chkcat[$i] = 'fast';
465 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_closure$TPOSTLBL[@]?$/o ) {
467 $chkcat[$i] = 'closure';
470 $closurechk{$1} = $i;
472 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_srt$TPOSTLBL[@]?$/o ) {
479 } elsif ( /^$TUS[@]?ghc.*c_ID$TPOSTLBL/o ) {
481 $chkcat[$i] = 'consist';
483 } elsif ( /^($TUS[@]?__gnu_compiled_c|gcc2_compiled\.)$TPOSTLBL/o ) {
486 } elsif ( /^$TUS[A-Za-z0-9_]+\.\d+$TPOSTLBL[@]?$/o
487 || /^$TUS[@]?.*_CAT$TPOSTLBL[@]?$/o # PROF: _entryname_CAT
488 || /^$TUS[@]?CC_.*_struct$TPOSTLBL[@]?$/o # PROF: _CC_ccident_struct
489 || /^$TUS[@]?.*_done$TPOSTLBL[@]?$/o # PROF: _module_done
490 || /^$TUS[@]?_module_registered$TPOSTLBL[@]?$/o # PROF: _module_registered
493 $chkcat[$i] = 'data';
496 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
501 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
506 } elsif ( /^$TUS[@]?CC(S)?_.*$/ ) {
507 # all CC_ symbols go in the data section...
509 $chkcat[$i] = 'data';
512 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(alt|dflt)$TPOSTLBL[@]?$/o ) {
514 $chkcat[$i] = 'misc';
517 #$chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
519 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_vtbl$TPOSTLBL[@]?$/o ) {
521 $chkcat[$i] = 'vector';
526 # As a temporary solution for compiling "foreign export" declarations,
527 # we use global variables to pass arguments from C to STG land.
528 # These declarations live in the .hc file and not in the generated C
529 # stub file, so we let them pass through here.
530 } elsif ( /^[\t ]+\.comm[\t ]+__fexp_.*$/ ) {
532 $chkcat[$i] = 'data';
535 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
536 && /^(_uname|uname|stat|fstat):/ ) {
537 # for some utterly bizarre reason, this platform
538 # likes to drop little local C routines with these names
539 # into each and every .o file that #includes the
540 # relevant system .h file. Yuck. We just don't
541 # tolerate them in .hc files (which we are processing
542 # here). If you need to call one of these things from
543 # Haskell, make a call to your own C wrapper, then
544 # put that C wrapper (which calls one of these) in a
545 # plain .c file. WDP 95/12
547 $chkcat[$i] = 'toss';
550 } elsif ( /^$TUS[@]?[A-Za-z0-9_]/o
551 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
553 && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
554 || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
557 print "Funny global thing?: $_"
558 unless $KNOWN_FUNNY_THING{$thing}
559 || /^$TUS[@]?stg_.*$TPOSTLBL[@]?$/o # RTS internals
560 || /^$TUS[@]__fexp_.*$TPOSTLBL$/o # foreign export
561 || /^$TUS[@]?_reg.*$TPOSTLBL$/o # PROF: __reg<module>
562 || /^$TUS[@]?.*_btm$TPOSTLBL$/o # large bitmaps
563 || /^$TUS[@]?.*_closure_tbl$TPOSTLBL$/o; # closure tables
565 $chkcat[$i] = 'misc';
566 if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
567 { $chksymb[$i] = $thing; }
568 else { $chksymb[$i] = ''; };
570 } else { # simple line (duplicated at the top)
575 $numchks = $#chk + 1;
577 # the division into chunks is imperfect;
578 # we throw some things over the fence into the next
581 # also, there are things we would like to know
582 # about the whole module before we start spitting
585 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
587 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
589 # Alphas: NB: we start meddling at chunk 1, not chunk 0
590 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
591 # spits a ".quad 0" in after the v first ".rdata"; we
592 # detect this special case (tossing the ".quad 0")!
593 local($magic_rdata_seen) = 0;
595 # HPPAs, MIPSen: also start medding at chunk 1
597 if ($TargetPlatform =~ /^powerpc|^rs6000/) {
598 print OUTASM $T_HDR_toc; # yes, we have to put a .toc
599 # in the beginning of every file!
600 %tocequiv = (); # maps toc symbol number to toc symbol
601 %revtocequiv = (); # maps toc symbol to toc symbol number
602 for ($i = 1; $i < $numchks; $i++) {
603 $chk[$i] =~ s/\[RW\]//g;
604 $chk[$i] =~ s/\[DS\]//g;
605 $chk[$i] =~ s/^\.csect .*\[DS\]$//g;
607 if ( $chkcat[$i] eq 'toc' && $chk[$i] !~ /\.byte/ )
608 #ToDo: instead of all these changes, just fix mangle_powerpc_tailjump and delete/ignore these tocs?
609 { $chk[$i] =~ s/$T_MOVE_DIRVS//g;
610 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_fast\d+)/\t\.tc \1\[TC\],\.\2/;
611 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_entry)\n/\t\.tc \1\[TC\],\.\2\n/;
612 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(ret_\S+)/\t\.tc \1\[TC\],\.\2/;
613 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(alt_\S+)/\t\.tc \1\[TC\],\.\2/;
614 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
616 $tocnumber = $chksymb[$i];
618 $tocsymb =~ s/^LC\.\.\d+:\n//;
619 $tocsymb =~ s/^\t\.tc \S+,(\S+)\n/\1/;
620 $tocequiv{$tocnumber} = $tocsymb;
622 } elsif ( $chkcat[$i] eq 'toc' && $chk[$i] =~ /\.byte/ ) {
623 $chkcat[$i] = 'literal';
628 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
629 $c = $chk[$i]; # convenience copy
631 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
633 # toss all prologue stuff; HPPA is pretty weird
635 $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/;
637 # be slightly paranoid to make sure there's
638 # nothing surprising in there
639 if ( $c =~ /--- BEGIN ---/ ) {
640 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
642 if ($TargetPlatform =~ /^i386-/) {
643 $p =~ s/^\tpushl \%edi\n//;
644 $p =~ s/^\tpushl \%esi\n//;
645 $p =~ s/^\tsubl \$\d+,\%esp\n//;
646 $p =~ s/^\tmovl \$\d+,\%eax\n\tcall __alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
647 } elsif ($TargetPlatform =~ /^m68k-/) {
648 $p =~ s/^\tlink a6,#-?\d.*\n//;
649 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
650 # The above showed up in the asm code,
651 # so I added it here.
652 # I hope it's correct.
654 $p =~ s/^\tmovel d2,sp\@-\n//;
655 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
656 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
657 } elsif ($TargetPlatform =~ /^mips-/) {
658 # the .frame/.mask/.fmask that we use is the same
659 # as that produced by GCC for miniInterpret; this
660 # gives GDB some chance of figuring out what happened
661 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
662 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
663 $p =~ s/^\t\.(mask|fmask).*\n//g;
664 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
665 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
666 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
667 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
668 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
669 $p =~ s/__FRAME__/$FRAME/;
670 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000/) {
671 $p =~ s/^\tmflr 0\n//;
672 $p =~ s/^\tstm \d+,-\d+\(1\)\n//;
673 $p =~ s/^\tstw? 0,\d+\(1\)\n//g;
674 $p =~ s/^\tstw? 1,\d+\(1\)\n//g; #mc
675 $p =~ s/^\tlw?z 0,0\(1\)\n//g; #mc
676 $p =~ s/^\tstw?u 1,-\d+\(1\)\n//;
677 $p =~ s/^\tstw? \d+,-\d+\(1\)\n//g;
678 $p =~ s/^\tstfd \d+,-\d+\(1\)\n//g;
680 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
683 die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
684 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
686 # glue together what's left
688 $c =~ s/\n\t\n/\n/; # junk blank line
692 if ( $TargetPlatform =~ /^mips-/ ) {
693 # MIPS: first, this basic sequence may occur "--- END ---" or not
694 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
697 # toss all epilogue stuff; again, paranoidly
698 if ( $c =~ /--- END ---/ ) {
699 if (($r, $e) = split(/--- END ---/, $c)) {
700 if ($TargetPlatform =~ /^i386-/) {
702 $e =~ s/^\tpopl \%edi\n//;
703 $e =~ s/^\tpopl \%esi\n//;
704 $e =~ s/^\taddl \$\d+,\%esp\n//;
705 } elsif ($TargetPlatform =~ /^m68k-/) {
706 $e =~ s/^\tunlk a6\n//;
708 } elsif ($TargetPlatform =~ /^mips-/) {
709 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
710 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
711 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
712 $e =~ s/^\tj\t\$31\n//;
713 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
714 $e =~ s/^\taddi 1,1,\d+\n//;
715 $e =~ s/^\tcal 1,\d+\(1\)\n//;
716 $e =~ s/^\tlw?z? \d+,\d+\(1\)\n//;
717 $e =~ s/^\tmtlr 0\n//;
720 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
722 die "Epilogue junk?: $e\n" if $e =~ /^\t[^\.]/
723 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
725 # glue together what's left
727 $c =~ s/\n\t\n/\n/; # junk blank line
731 # On SPARCs, we don't do --- BEGIN/END ---, we just
732 # toss the register-windowing save/restore/ret* instructions
734 if ( $TargetPlatform =~ /^sparc-/ ) {
735 $c =~ s/^\t(save .*|restore|ret|retl)\n//g;
736 # throw away PROLOGUE comments
737 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
740 # On Alphas, the prologue mangling is done a little later (below)
742 # toss all calls to __DISCARD__
743 $c =~ s/^\t(call|jbsr|jal)\s+$TUS[@]?__DISCARD__\n//go;
745 # MIPS: that may leave some gratuitous asm macros around
746 # (no harm done; but we get rid of them to be tidier)
747 $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/
748 if $TargetPlatform =~ /^mips-/;
750 # toss stack adjustment after DoSparks
751 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
752 if $TargetPlatform =~ /^m68k-/; # this looks old...
754 if ( $TargetPlatform =~ /^alpha-/ &&
755 ! $magic_rdata_seen &&
756 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
757 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
758 $magic_rdata_seen = 1;
761 # pick some end-things and move them to the next chunk
763 # pin a funny end-thing on (for easier matching):
764 $c .= 'FUNNY#END#THING';
766 while ( $c =~ /$TMOVEDIRVS[@]?FUNNY#END#THING/o ) { # [@]? is a silly hack to avoid having to use curlies for T_PRE_APP
767 # (this SEGVs perl4 on alphas, you see)
770 if ( $i < ($numchks - 1)
771 && ( $to_move =~ /$TCOPYDIRVS/
772 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
773 $chk[$i + 1] = $to_move . $chk[$i + 1];
774 # otherwise they're tossed
777 $c =~ s/$TMOVEDIRVS[@]?FUNNY#END#THING/FUNNY#END#THING/o; # [@]? is a hack (see above)
780 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
782 # toss all prologue stuff, except for loading gp, and the ..ng address
783 if (($p, $r) = split(/^\t\.prologue/, $c)) {
784 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
785 $c = $keep . "..ng:\n";
787 print STDERR "malformed code block ($ent)?\n"
790 $c .= "\t.frame \$30,0,\$26,0\n\t.prologue" . $r;
793 $c =~ s/FUNNY#END#THING//;
795 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
797 $chk[$i] = $c; # update w/ convenience copy
800 if ( $TargetPlatform =~ /^alpha-/ ) {
801 # print out the header stuff first
802 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
803 print OUTASM $chk[0];
805 } elsif ( $TargetPlatform =~ /^hppa/ ) {
806 print OUTASM $chk[0];
808 } elsif ( $TargetPlatform =~ /^mips-/ ) {
809 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
811 # get rid of horrible "<dollar>Revision: .*$" strings
812 local(@lines0) = split(/\n/, $chk[0]);
814 while ( $z <= $#lines0 ) {
815 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
818 while ( $z <= $#lines0 ) {
820 last if $lines0[$z] =~ /[,\t]0x0$/;
826 $chk[0] = join("\n", @lines0);
827 $chk[0] =~ s/\n\n+/\n/;
828 print OUTASM $chk[0];
831 # print out all the literal strings next
832 for ($i = 0; $i < $numchks; $i++) {
833 if ( $chkcat[$i] eq 'literal' ) {
834 print OUTASM $T_HDR_literal, $chk[$i];
835 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
837 $chkcat[$i] = 'DONE ALREADY';
841 # on the HPPA, print out all the bss next
842 if ( $TargetPlatform =~ /^hppa/ ) {
843 for ($i = 1; $i < $numchks; $i++) {
844 if ( $chkcat[$i] eq 'bss' ) {
845 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
846 print OUTASM $chk[$i];
848 $chkcat[$i] = 'DONE ALREADY';
853 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
854 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
856 next if $chkcat[$i] eq 'DONE ALREADY';
858 if ( $chkcat[$i] eq 'misc' ) {
859 if ($chk[$i] ne '') {
860 print OUTASM $T_HDR_misc;
861 if ($TargetPlatform =~ /^powerpc-|^rs6000/) {
862 $chksymb[$i] =~ s/://;
863 #new if ($chksymb[$i] =~ /ret.*upd/ || $KNOWN_FUNNY_THING{$chksymb[$i]}
864 #new || $chksymb[$i] =~ /^$.{T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o )
865 #new { print OUTASM "\t\.globl $chksymb[$i]\n"; }
866 # if ($chksymb[$i] ne '' && $chksymb[$i] !~ /ret_[a-z]/ && $chksymb[$i] !~ /djn_[a-z]/)
867 if ($chksymb[$i] ne '')
868 { print OUTASM "\t\.globl \.$chksymb[$i]\n"; };
869 if ($chk[$i] =~ /TOC\[tc0\], 0\n/)
870 { ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$i]); $printDS = 1;}
871 else { $r = $chk[$i]; $printDS = 0; };
872 $chk[$i] = &mangle_powerpc_tailjump($r);
874 &print_doctored($chk[$i], 0);
875 if ($TargetPlatform =~ /^powerpc-|^rs6000-/ && $printDS) {
876 #ok if ($chksymb[$i] !~ /\_regMain/) {
877 print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
878 print OUTASM "${p}TOC[tc0], 0\n";
883 } elsif ( $chkcat[$i] eq 'toss' ) {
884 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
886 } elsif ( $chkcat[$i] eq 'data' ) {
887 if ($chk[$i] ne '') {
888 print OUTASM $T_HDR_data;
889 print OUTASM $chk[$i];
892 } elsif ( $chkcat[$i] eq 'consist' ) {
893 if ( $chk[$i] =~ /$T_hsc_cc_PAT/o ) {
894 local($consist) = "$1.$2.$3";
896 $consist =~ s/\//./g;
898 $consist =~ s/[^A-Za-z0-9_.]/ZZ/g; # ToDo: properly?
900 # Using a cygnus-2.7-96q4 gcc build on hppas, the
901 # consistency chunk for ghc_cc_ID often (but not always!)
902 # gets lumped with a bunch of .IMPORT directives containing info on
903 # the code or data space nature of external symbols. We can't
904 # toss these, so once the consistency ID has been turned into
905 # a representable symbol, we substitute it for the symbol
906 # that the string was attached to in the first place (ghc_cc_ID.)
907 # (The original string is also substituted away.)
909 # This change may affect the code output on other platforms in
910 # adverse ways, hence we restrict this hack hppa targets only.
913 if ( $TargetPlatform =~ /^hppa/ ) {
914 $chk[$i] =~ s/^$TUS[@]?ghc.*c_ID$TPOSTLBL/$consist/o;
915 $chk[$i] =~ s/\t$T_hsc_cc_PAT/$T_HDR_misc/o;
916 $consist = $chk[$i]; #clumsily
918 print OUTASM $T_HDR_consist, "${consist}${T_POST_LBL}\n";
920 } elsif ( $TargetPlatform !~ /^(mips)-/ ) { # we just don't try in those case (ToDo)
921 # on mips: consistency string is just a v
922 # horrible bunch of .bytes,
923 # which I am too lazy to sort out (WDP 95/05)
925 print STDERR "Couldn't grok consistency: ", $chk[$i];
928 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
929 # we can just re-constitute this one...
930 # NB: we emit _three_ underscores no matter what,
931 # so ghc-split doesn't have to care.
932 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
934 } elsif ( $chkcat[$i] eq 'closure'
935 || $chkcat[$i] eq 'srt'
936 || $chkcat[$i] eq 'infotbl'
937 || $chkcat[$i] eq 'slow'
938 || $chkcat[$i] eq 'fast' ) { # do them in that order
939 $symb = $chksymb[$i];
942 if ( defined($closurechk{$symb}) ) {
943 print OUTASM $T_HDR_closure;
944 print OUTASM $chk[$closurechk{$symb}];
945 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
949 if ( defined($srtchk{$symb}) ) {
950 print OUTASM $T_HDR_srt;
951 print OUTASM $chk[$srtchk{$symb}];
952 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
956 if ( defined($infochk{$symb}) ) {
958 print OUTASM $T_HDR_info;
959 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
960 if ( !defined($slowchk{$symb}) && defined($fastchk{$symb}) ) {
961 $fastname = $chk[$fastchk{$symb}];
962 $fastname =~ s/([_A-Za-z]*_fast\d+):.*(.*\n)*/\1/;
963 $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long $fastname/;
965 $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
966 $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
967 print OUTASM $chk[$infochk{$symb}];
969 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
971 # entry code will be put here!
973 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
977 if ( defined($slowchk{$symb}) ) {
979 # teach it to drop through to the fast entry point:
980 $c = $chk[$slowchk{$symb}];
982 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
983 ($p, $r) = split(/TOC\[tc0\], 0\n/, $c);
984 if ($symb =~ /^[_A-Z]/)
986 print OUTASM "\t\.globl \.${chksymb[$i]}_entry\n";
987 print OUTASM "\.csect ${symb}_entry[DS]\n";
988 print OUTASM "${p}TOC[tc0], 0\n";
990 $r =~ s/\.csect \.text\[PR\]\n//; # todo: properly - andre
991 $c = &mangle_powerpc_tailjump($r);
994 if ( defined($fastchk{$symb}) ) {
995 if ( $TargetPlatform =~ /^alpha-/ ) {
996 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
997 } elsif ( $TargetPlatform =~ /^hppa/ ) {
998 $c =~ s/^\s+ldil.*\n\s+ldo.*\n\s+bv.*\n(.*\n)?\s+\.EXIT/$1\t.EXIT/;
999 } elsif ( $TargetPlatform =~ /^i386-/ ) {
1000 # Reg alloc depending, gcc generated code may jump to the fast entry point via
1001 # a number of registers.
1002 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edx\n\tjmp \*\%edx\n//;
1003 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%ecx\n\tjmp \*\%ecx\n//;
1004 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%eax\n\tjmp \*\%eax\n//;
1005 # The next two only apply if we're not stealing %esi or %edi.
1006 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%esi\n\tjmp \*\%esi\n// if ($StolenX86Regs < 3);
1007 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edi\n\tjmp \*\%edi\n// if ($StolenX86Regs < 4);
1008 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1009 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1010 } elsif ( $TargetPlatform =~ /^m68k-/ ) {
1011 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n\tnop\n//;
1012 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n//;
1013 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ ) {
1014 $c =~ s/^\tb \.${T_US}${symb}_fast\d+\n//;
1015 } elsif ( $TargetPlatform =~ /^sparc-/ ) {
1016 $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n\tnop\n//;
1017 $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n(\t[a-z].*\n)/$1/;
1019 print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n";
1023 if ( $TargetPlatform !~ /^(alpha-|hppa|mips-)/ ) {
1024 # On alphas, hppa: no very good way to look for "dangling"
1025 # references to fast-entry point.
1026 # (questionable re hppa and mips...)
1027 print STDERR "still has jump to fast entry point:\n$c"
1028 if $c =~ /$TUS[@]?$symb[@]?_fast/; # NB: paranoia
1031 print OUTASM $T_HDR_entry;
1033 &print_doctored($c, 1); # NB: the 1!!!
1035 $chkcat[$slowchk{$symb}] = 'DONE ALREADY';
1039 if ( defined($fastchk{$symb}) ) {
1040 $c = $chk[$fastchk{$symb}];
1041 if ( ! defined($slowchk{$symb})
1042 # ToDo: the || clause can go once we're no longer
1043 # concerned about producing exactly the same output as before
1044 #OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/
1046 print OUTASM $T_HDR_fast;
1048 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1049 local(@lbls) = split(/:/, $c);
1050 $fullname = $lbls[0];
1051 $fullname =~ s/$T_MOVE_DIRVS//g;
1052 if ( $fullname =~ /^[A-Z]/)
1053 { print OUTASM "\t\.globl \.${fullname}\n";
1055 # print OUTASM "\t\.lglobl \.${fullname}\n"; #todo: rm - andre
1057 $c =~ s/((.*\n)*)\t.long \S+, TOC\[tc0\], 0\n\.csect \.text\[PR\]\n((.*\n)*)/\1\3/;
1058 $c = &mangle_powerpc_tailjump($c);
1060 &print_doctored($c, 0);
1061 $chkcat[$fastchk{$symb}] = 'DONE ALREADY';
1064 } elsif ( $chkcat[$i] eq 'vector'
1065 || $chkcat[$i] eq 'direct' ) { # do them in that order
1066 $symb = $chksymb[$i];
1069 if ( defined($vectorchk{$symb}) ) {
1070 print OUTASM $T_HDR_vector;
1071 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1072 if ( $symb =~ /^[A-Z]/) {
1073 print OUTASM "\t\.globl \.${symb}_vtbl\n";
1074 print OUTASM "\t\.globl ${symb}_vtbl\n";
1076 $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
1077 print OUTASM ".${symb}_vtbl:\n";
1078 print OUTASM $chk[$vectorchk{$symb}];
1080 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1082 # direct return code will be put here!
1083 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1085 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1086 # Alphas: the commented nop is for the splitter, to ensure
1087 # that no module ends with a label as the very last
1088 # thing. (The linker will adjust the label to point
1089 # to the first code word of the next module linked in,
1090 # even if alignment constraints cause the label to move!)
1092 print OUTASM "\t# nop\n";
1095 } elsif ( $chkcat[$i] eq 'toc' ) {
1096 # silly optimisation to print tocs, since they come in groups...
1097 print OUTASM $T_HDR_toc;
1099 while ($chkcat[$j] eq 'toc')
1100 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1103 print OUTASM $chk[$j];
1105 $chkcat[$j] = 'DONE ALREADY';
1110 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1114 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1116 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1117 print OUTASM ".csect .text[PR]\n_section_.text:\n.csect .data[RW]\n\t.long _section_.text\n"
1121 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1122 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1127 sub mash_hppa_prologue { # OK, epilogue, too
1130 # toss all prologue stuff
1131 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1133 # Lie about our .CALLINFO
1134 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1141 # toss all epilogue stuff
1142 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1144 # Sorry; we moved the _info stuff to the code segment.
1145 s/_info,DATA/_info,CODE/g;
1152 sub print_doctored {
1153 local($_, $need_fallthru_patch) = @_;
1155 if ( $TargetPlatform !~ /^i386-/
1156 || ! /^\t[a-z]/ ) { # no instructions in here, apparently
1160 # OK, must do some x86 **HACKING**
1162 local($entry_patch) = '';
1163 local($exit_patch) = '';
1165 # gotta watch out for weird instructions that
1166 # invisibly smash various regs:
1167 # rep* %ecx used for counting
1168 # scas* %edi used for destination index
1169 # cmps* %e[sd]i used for indices
1170 # loop* %ecx used for counting
1175 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1177 # * GCC used an "STG reg" for its own purposes
1179 # * some secret uses of machine reg, requiring STG reg
1180 # to be saved/restored
1182 # The most dangerous "GCC uses" of an "STG reg" are when
1183 # the reg holds the target of a jmp -- it's tricky to
1184 # insert the patch-up code before we get to the target!
1185 # So here we change the jmps:
1187 # --------------------------------------------------------
1188 # it can happen that we have jumps of the form...
1189 # jmp *<something involving %esp>
1191 # jmp <something involving another naughty register...>
1193 # a reasonably-common case is:
1195 # movl $_blah,<bad-reg>
1199 # the short form may tickle perl bug:
1200 # s/^\tmovl \$${T_US}(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp $T_US$1/g;
1201 s/^\tmovl \$${T_US}(.*),\%eax\n\tjmp \*\%eax/\tjmp $T_US$1/g;
1202 s/^\tmovl \$${T_US}(.*),\%ebx\n\tjmp \*\%ebx/\tjmp $T_US$1/g;
1203 s/^\tmovl \$${T_US}(.*),\%ecx\n\tjmp \*\%ecx/\tjmp $T_US$1/g;
1204 s/^\tmovl \$${T_US}(.*),\%edx\n\tjmp \*\%edx/\tjmp $T_US$1/g;
1206 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1207 s/^\tmovl (.*),\%esi\n\tjmp \*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1208 s/^\tjmp \*(-?\d*)\((.*\%esi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1209 s/^\tjmp \*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1210 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1211 if /(jmp|call) .*\%esi/;
1213 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1214 s/^\tmovl (.*),\%edi\n\tjmp \*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1215 s/^\tjmp \*(-?\d*)\((.*\%edi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1216 s/^\tjmp \*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1217 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1218 if /(jmp|call) .*\%edi/;
1221 # OK, now we can decide what our patch-up code is going to
1224 # Offsets into register table - you'd better update these magic
1225 # numbers should you change its contents!
1226 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1227 local($OFFSET_Hp)=92;
1229 # Note funky ".=" stuff; we're *adding* to these _patch guys
1230 if ( $StolenX86Regs <= 2
1231 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1232 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1233 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1235 # nothing for call_{entry,exit} because %esi is callee-save
1237 if ( $StolenX86Regs <= 3
1238 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1239 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1240 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1242 # nothing for call_{entry,exit} because %edi is callee-save
1245 # --------------------------------------------------------
1246 # next, here we go with non-%esp patching!
1248 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1250 # Before calling GC we must set up the exit condition before the call
1251 # and entry condition when we come back
1253 # fix _all_ non-local jumps:
1255 s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1256 s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1258 s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
1260 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1261 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1263 if ($StolenX86Regs == 2 ) {
1264 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1265 if /^\t(jmp|call) .*\%e(si|di)/;
1266 } elsif ($StolenX86Regs == 3 ) {
1267 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1268 if /^\t(jmp|call) .*\%edi/;
1271 # --------------------------------------------------------
1272 # that's it -- print it
1274 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1278 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1279 print OUTASM $exit_patch;
1280 # ToDo: make it not print if there is a "jmp" at the end
1286 sub init_FUNNY_THINGS {
1287 %KNOWN_FUNNY_THING = (
1289 # "${T_US}stg_.*{T_POST_LBL}", 1,
1294 The following table reversal is used for both info tables and return
1295 vectors. In both cases, we remove the first entry from the table,
1296 reverse the table, put the label at the end, and paste some code
1297 (that which is normally referred to by the first entry in the table)
1298 right after the table itself. (The code pasting is done elsewhere.)
1302 local($symb, $tbl, $discard1) = @_;
1304 local($before) = '';
1306 local(@imports) = (); # hppa only
1309 local(@lines) = split(/\n/, $tbl);
1310 local($i, $j); #local ($i, $extra, $words_to_pad, $j);
1312 # see comment in mangleAsm as to why this silliness is needed.
1313 local($TDOTWORD) = ${T_DOT_WORD};
1314 local($TDOTGLOBAL) = ${T_DOT_GLOBAL};
1315 local($TUS) = ${T_US};
1316 local($TPOSTLBL) = ${T_POST_LBL};
1318 # Deal with the header...
1319 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?$TDOTWORD\s+/o; $i++) {
1320 $label .= $lines[$i] . "\n",
1321 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info$TPOSTLBL[@]?$/o
1322 || $lines[$i] =~ /$TDOTGLOBAL/o
1323 || $lines[$i] =~ /^$TUS[@]?\S+_vtbl$TPOSTLBL[@]?$/o;
1325 $before .= $lines[$i] . "\n"; # otherwise...
1328 # Grab the table data...
1329 if ( $TargetPlatform !~ /^hppa/ ) {
1330 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?$TDOTWORD\s+/o; $i++) {
1331 push(@words, $lines[$i]);
1333 } else { # hppa weirdness
1334 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+\.(word|IMPORT)/; $i++) {
1335 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1336 push(@imports, $lines[$i]);
1338 # We don't use HP's ``function pointers''
1339 # We just use labels in code space, like normal people
1340 $lines[$i] =~ s/P%//;
1341 push(@words, $lines[$i]);
1346 # now throw away the first word (SRT) iff it is empty.
1347 # The .zero business is for Linux/ELF.
1348 # The .skip business is for Sparc/Solaris/ELF.
1349 if ($discard1 && $words[0] =~ /^\t?($TDOTWORD\s+0|\.zero\s+4|\.skip\s+4)/) {
1353 # Padding removed to reduce code size and improve performance on Pentiums.
1355 # for 486-cache-friendliness, we want our tables aligned
1356 # on 16-byte boundaries (.align 4). Let's pad:
1357 # $extra = ($#words + 1) % 4;
1358 # $words_to_pad = ($extra == 0) ? 0 : 4 - $extra;
1359 # for ($j = 0; $j < $words_to_pad; $j++) { push(@words, "\t${T_DOT_WORD} 0"); }
1361 for (; $i <= $#lines; $i++) {
1362 $after .= $lines[$i] . "\n";
1365 # Alphas:If we have anonymous text (not part of a procedure), the
1366 # linker may complain about missing exception information. Bleh.
1367 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1368 $before = "\t.ent $1\n" . $before;
1369 $after .= "\t.end $1\n";
1373 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1374 . join("\n", @words) . "\n"
1377 # print STDERR "before=$before\n";
1378 # print STDERR "label=$label\n";
1379 # print STDERR "words=",(reverse @words),"\n";
1380 # print STDERR "after=$after\n";
1386 The HP is a major nuisance. The threaded code mangler moved info
1387 tables from data space to code space, but unthreaded code in the RTS
1388 still has references to info tables in data space. Since the HP
1389 linker is very precise about where symbols live, we need to patch the
1390 references in the unthreaded RTS as well.
1393 sub mini_mangle_asm_hppa {
1394 local($in_asmf, $out_asmf) = @_;
1396 open(INASM, "< $in_asmf")
1397 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1398 open(OUTASM,"> $out_asmf")
1399 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1402 s/_info,DATA/_info,CODE/; # Move _info references to code space
1408 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1409 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1415 sub mini_mangle_asm_powerpc {
1416 local($in_asmf, $out_asmf) = @_;
1418 open(INASM, "< $in_asmf")
1419 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1420 open(OUTASM,"> $out_asmf")
1421 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1424 s/long _PRStart/long ._PRStart/;
1425 s/long _PRIn_/long ._PRIn_/;
1426 s/long _Dummy_(\S+)_entry/long ._Dummy_\1_entry/;
1427 s/long _PRMarking_MarkNextRoot\[DS\]/long ._PRMarking_MarkNextRoot/;
1428 s/long _PRMarking_MarkNextCAF\[DS\]/long ._PRMarking_MarkNextCAF/;
1429 s/long _PRMarking_MarkNextAStack\[DS\]/long ._PRMarking_MarkNextAStack/;
1430 s/long _PRMarking_MarkNextBStack\[DS\]/long ._PRMarking_MarkNextBStack/;
1431 s/\.tc EnterNodeCode\[TC]\,EnterNodeCode\[DS\]/\.tc EnterNodeCode\[TC]\,.EnterNodeCode/; # CONC
1432 s/\.tc CheckHeapCode\[TC]\,CheckHeapCode\[DS\]/\.tc CheckHeapCode\[TC]\,.CheckHeapCode/; # CONC
1437 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1438 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1441 sub mangle_powerpc_tailjump {
1443 local($maybe_more) = 1;
1444 while (($c =~ /\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n/) && $maybe_more)
1447 $lcsymb =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1448 # the checks for r1 and r2 are mostly paranoia...
1450 $r1 =~ s/(.*\n)*\tlw?z? (\d+),LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1452 $r2 =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr (\d+)\n\tbctr\n(.*\n)*/\3/;
1455 $c =~ s/((.*\n)*)\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n((.*\n)*)/\1\tb $tocequiv{$lcsymb}\n\3/;
1461 # make "require"r happy...