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 = '\.(blockz|word|half|byte)';
93 $T_DOT_GLOBAL = '^\s+\.EXPORT';
94 $T_HDR_literal = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
95 $T_HDR_misc = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
96 $T_HDR_data = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
97 $T_HDR_consist = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
98 $T_HDR_closure = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
99 $T_HDR_srt = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
100 $T_HDR_info = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
101 $T_HDR_entry = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
102 $T_HDR_fast = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
103 $T_HDR_vector = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
104 $T_HDR_direct = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
106 #--------------------------------------------------------#
107 } elsif ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd2|nextstep3|cygwin32|mingw32)$/ ) {
108 # NeXT added but not tested. CaS
110 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
111 $T_US = '_'; # _ if symbols have an underscore on the front
112 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
113 $T_CONST_LBL = '^LC(\d+):$';
115 $T_X86_PRE_LLBL_PAT = 'L';
116 $T_X86_PRE_LLBL = 'L';
117 $T_X86_BADJMP = '^\tjmp [^L\*]';
119 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,0x90)?|\.globl\s+\S+|\.text|\.data|\.stab[^n].*|\.type\s+.*|\.size\s+.*)\n)';
120 $T_COPY_DIRVS = '\.(globl|stab)';
121 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
122 $T_DOT_WORD = '\.(long|word|value|byte|space)';
123 $T_DOT_GLOBAL = '\.globl';
124 $T_HDR_literal = "\.text\n\t\.align 2\n";
125 $T_HDR_misc = "\.text\n\t\.align 2,0x90\n";
126 $T_HDR_data = "\.data\n\t\.align 2\n";
127 $T_HDR_consist = "\.text\n";
128 $T_HDR_closure = "\.data\n\t\.align 2\n";
129 $T_HDR_closure = "\.data\n\t\.align 2\n\t.long 0\n" if ( $TargetPlatform =~ /.*-mingw32$/ );
130 $T_HDR_srt = "\.data\n\t\.align 2\n";
131 $T_HDR_info = "\.text\n\t\.align 2\n"; # NB: requires padding
132 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
133 $T_HDR_fast = "\.text\n\t\.align 2,0x90\n";
134 $T_HDR_vector = "\.text\n\t\.align 2\n"; # NB: requires padding
135 $T_HDR_direct = "\.text\n\t\.align 2,0x90\n";
137 #--------------------------------------------------------#
138 } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|freebsd3)$/ ) {
140 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
141 $T_US = ''; # _ if symbols have an underscore on the front
142 $T_PRE_APP = # regexp that says what comes before APP/NO_APP
143 ($TargetPlatform =~ /-(linux|freebsd3)$/) ? '#' : '/' ;
144 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
146 $T_X86_PRE_LLBL_PAT = '\.L';
147 $T_X86_PRE_LLBL = '.L';
148 $T_X86_BADJMP = '^\tjmp [^\.\*]';
150 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,0x90)?|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.size\s+.*|\.type\s+.*|\.Lfe.*\n\t\.size\s+.*|\.ident.*)\n)';
151 $T_COPY_DIRVS = '\.(globl)';
153 if ( $TargetPlatform =~ /freebsd3/ ) {
154 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
156 $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
159 $T_DOT_WORD = '\.(long|value|byte|zero)';
160 $T_DOT_GLOBAL = '\.globl';
161 $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11)
162 $T_HDR_misc = "\.text\n\t\.align 4\n";
163 $T_HDR_data = "\.data\n\t\.align 4\n"; # ToDo: change align??
164 $T_HDR_consist = "\.text\n";
165 $T_HDR_closure = "\.data\n\t\.align 4\n"; # ToDo: change align?
166 $T_HDR_srt = "\.data\n\t\.align 4\n"; # ToDo: change align?
167 $T_HDR_info = "\.text\n\t\.align 4\n"; # NB: requires padding
168 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
169 $T_HDR_fast = "\.text\n\t\.align 4\n";
170 $T_HDR_vector = "\.text\n\t\.align 4\n"; # NB: requires padding
171 $T_HDR_direct = "\.text\n\t\.align 4\n";
173 #--------------------------------------------------------#
174 } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
176 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
177 $T_US = '_'; # _ if symbols have an underscore on the front
178 $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
179 $T_CONST_LBL = '^LC(\d+):$';
182 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
183 $T_COPY_DIRVS = '\.(globl|proc|stab)';
184 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
186 $T_DOT_WORD = '\.long';
187 $T_DOT_GLOBAL = '\.globl';
188 $T_HDR_literal = "\.text\n\t\.even\n";
189 $T_HDR_misc = "\.text\n\t\.even\n";
190 $T_HDR_data = "\.data\n\t\.even\n";
191 $T_HDR_consist = "\.text\n";
192 $T_HDR_closure = "\.data\n\t\.even\n";
193 $T_HDR_srt = "\.data\n\t\.even\n";
194 $T_HDR_info = "\.text\n\t\.even\n";
195 $T_HDR_entry = "\.text\n\t\.even\n";
196 $T_HDR_fast = "\.text\n\t\.even\n";
197 $T_HDR_vector = "\.text\n\t\.even\n";
198 $T_HDR_direct = "\.text\n\t\.even\n";
200 #--------------------------------------------------------#
201 } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
203 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
204 $T_US = ''; # _ if symbols have an underscore on the front
205 $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
206 $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
209 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
210 $T_COPY_DIRVS = '\.(globl|ent)';
212 $T_hsc_cc_PAT = 'I WAS TOO LAZY TO DO THIS BIT (WDP 95/05)';
213 $T_DOT_WORD = '\.word';
214 $T_DOT_GLOBAL = '^\t\.globl';
215 $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
216 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
217 $T_HDR_data = "\t\.data\n\t\.align 2\n";
218 $T_HDR_consist = 'TOO LAZY TO DO THIS TOO';
219 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
220 $T_HDR_srt = "\t\.data\n\t\.align 2\n";
221 $T_HDR_info = "\t\.text\n\t\.align 2\n";
222 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
223 $T_HDR_fast = "\t\.text\n\t\.align 2\n";
224 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
225 $T_HDR_direct = "\t\.text\n\t\.align 2\n";
227 #--------------------------------------------------------#
228 } elsif ( $TargetPlatform =~ /^powerpc-.*|^rs6000-.*/ ) {
230 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
231 $T_US = ''; # _ if symbols have an underscore on the front
232 $T_PRE_APP = 'NOT APPLICABLE'; # regexp that says what comes before APP/NO_APP
233 $T_CONST_LBL = 'NOT APPLICABLE'; # regexp for what such a lbl looks like
236 $T_MOVE_DIRVS = '^(\s*(\.toc|\.align \d+|\.csect \S+|\t\.?l?globl \S+)\n)';
237 $T_COPY_DIRVS = '\.(l?globl)';
239 $T_hsc_cc_PAT = '\.byte.*\)(hsc|cc) (.*)"\n\t\.byte \d+\n\t\.byte "(.*)"\n\t\.byte \d+';
240 $T_DOT_WORD = '\.long';
241 $T_DOT_GLOBAL = '\.globl';
242 $T_HDR_toc = "\.toc\n";
243 $T_HDR_literal = "\.csect .data[RW]\n\t\.align 2\n"; #not RO!?
244 $T_HDR_misc = "# misc\n\.csect \.text[PR]\n\t\.align 2\n";
245 $T_HDR_data = "# data\n\.csect \.data[RW]\n\t\.align 2\n";
246 $T_HDR_consist = "# consist\n\.csect \.data[RW]\n\t\.align 2\n";
247 $T_HDR_closure = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
248 $T_HDR_srt = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
249 $T_HDR_info = "# info\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
250 $T_HDR_entry = "# entry\n\.csect \.text[PR]\n\t\.align 2\n";
251 $T_HDR_fast = "# fast\n\.csect \.text[PR]\n\t\.align 2\n";
252 $T_HDR_vector = "# vector\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
253 $T_HDR_direct = "# direct\n";
255 #--------------------------------------------------------#
256 } elsif ( $TargetPlatform =~ /^sparc-.*-solaris2/ ) {
258 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
259 $T_US = ''; # _ if symbols have an underscore on the front
260 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
261 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
264 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\.section.*|\s+\.type.*|\s+\.size.*)\n)';
265 $T_COPY_DIRVS = '\.(global|proc|stab)';
267 $T_hsc_cc_PAT = '\.asciz.*\)(hsc|cc) (.*)\\\\t(.*)"';
268 $T_DOT_WORD = '\.(word|byte|half|skip|uahalf|uaword)';
269 $T_DOT_GLOBAL = '^\t\.global';
270 $T_HDR_literal = "\.text\n\t\.align 8\n";
271 $T_HDR_misc = "\.text\n\t\.align 4\n";
272 $T_HDR_data = "\.data\n\t\.align 8\n";
273 $T_HDR_consist = "\.text\n";
274 $T_HDR_closure = "\.data\n\t\.align 4\n";
275 $T_HDR_srt = "\.data\n\t\.align 4\n";
276 $T_HDR_info = "\.text\n\t\.align 4\n";
277 $T_HDR_entry = "\.text\n\t\.align 4\n";
278 $T_HDR_fast = "\.text\n\t\.align 4\n";
279 $T_HDR_vector = "\.text\n\t\.align 4\n";
280 $T_HDR_direct = "\.text\n\t\.align 4\n";
282 #--------------------------------------------------------#
283 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
285 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
286 $T_US = '_'; # _ if symbols have an underscore on the front
287 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
288 $T_CONST_LBL = '^LC(\d+):$';
291 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
292 $T_COPY_DIRVS = '\.(global|proc|stab)';
293 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
295 $T_DOT_WORD = '\.word';
296 $T_DOT_GLOBAL = '^\t\.global';
297 $T_HDR_literal = "\.text\n\t\.align 8\n";
298 $T_HDR_misc = "\.text\n\t\.align 4\n";
299 $T_HDR_data = "\.data\n\t\.align 8\n";
300 $T_HDR_consist = "\.text\n";
301 $T_HDR_closure = "\.data\n\t\.align 4\n";
302 $T_HDR_srt = "\.data\n\t\.align 4\n";
303 $T_HDR_info = "\.text\n\t\.align 4\n";
304 $T_HDR_entry = "\.text\n\t\.align 4\n";
305 $T_HDR_fast = "\.text\n\t\.align 4\n";
306 $T_HDR_vector = "\.text\n\t\.align 4\n";
307 $T_HDR_direct = "\.text\n\t\.align 4\n";
309 #--------------------------------------------------------#
311 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
316 print STDERR "T_STABBY: $T_STABBY\n";
317 print STDERR "T_US: $T_US\n";
318 print STDERR "T_PRE_APP: $T_PRE_APP\n";
319 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
320 print STDERR "T_POST_LBL: $T_POST_LBL\n";
321 if ( $TargetPlatform =~ /^i386-/ ) {
322 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
323 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
324 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
326 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
327 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
328 print STDERR "T_hsc_cc_PAT: $T_hsc_cc_PAT\n";
329 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
330 print STDERR "T_HDR_literal: $T_HDR_literal\n";
331 print STDERR "T_HDR_misc: $T_HDR_misc\n";
332 print STDERR "T_HDR_data: $T_HDR_data\n";
333 print STDERR "T_HDR_consist: $T_HDR_consist\n";
334 print STDERR "T_HDR_closure: $T_HDR_closure\n";
335 print STDERR "T_HDR_info: $T_HDR_info\n";
336 print STDERR "T_HDR_entry: $T_HDR_entry\n";
337 print STDERR "T_HDR_fast: $T_HDR_fast\n";
338 print STDERR "T_HDR_vector: $T_HDR_vector\n";
339 print STDERR "T_HDR_direct: $T_HDR_direct\n";
345 %************************************************************************
347 \subsection{Mangle away}
349 %************************************************************************
353 local($in_asmf, $out_asmf) = @_;
355 # multi-line regexp matching:
360 &init_TARGET_STUFF();
361 &init_FUNNY_THINGS();
363 # perl4 on alphas SEGVs when give ${foo} substitutions in patterns.
364 # To avoid them we declare some locals that allows to avoid using curlies.
365 local($TUS) = ${T_US};
366 local($TPOSTLBL) = ${T_POST_LBL};
367 local($TMOVEDIRVS) = ${T_MOVE_DIRVS};
368 local($TPREAPP) = ${T_PRE_APP};
369 local($TCOPYDIRVS) = ${T_COPY_DIRVS};
370 local($TDOTWORD) = ${T_DOT_WORD};
372 open(INASM, "< $in_asmf")
373 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
374 open(OUTASM,"> $out_asmf")
375 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
377 # read whole file, divide into "chunks":
378 # record some info about what we've found...
380 @chk = (); # contents of the chunk
381 $numchks = 0; # number of them
382 @chkcat = (); # what category of thing in each chunk
383 @chksymb = (); # what symbol(base) is defined in this chunk
384 %slowchk = (); # ditto, its regular "slow" entry code
385 %fastchk = (); # ditto, fast entry code
386 %closurechk = (); # ditto, the (static) closure
387 %srtchk = (); # ditto, its SRT (for top-level things)
388 %infochk = (); # given a symbol base, say what chunk its info tbl is in
389 %vectorchk = (); # ditto, return vector table
390 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
392 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
395 next if $T_STABBY && /^\.stab.*$TUS[@]?__stg_split_marker/o;
396 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
397 next if /^\t\.def.*endef$/;
398 next if /$TPREAPP(NO_)?APP/o;
399 next if /^;/ && $TargetPlatform =~ /^hppa/;
401 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|powerpc|rs6000)-/;
403 last if /^_section_\.text:$/ && $TargetPlatform =~ /^powerpc-|^rs6000-/;
405 if ( $TargetPlatform =~ /^mips-/
406 && /^\t\.(globl \S+ \.text|comm\t)/ ) {
407 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
409 # As a temporary solution for compiling "foreign export" declarations,
410 # we use global variables to pass arguments from C to STG land.
411 # These declarations live in the .hc file and not in the generated C
412 # stub file, so we let them pass through here.
413 } elsif ( /^\t\.comm\t__fexp_.*$/ ) {
415 $chkcat[$i] = 'data';
418 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
419 # duplicated from the bottom
423 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
424 # Alphas: Local labels not to be confused with new chunks
427 # NB: all the rest start with a non-space
429 } elsif ( $TargetPlatform =~ /^mips-/
430 && /^\d+:/ ) { # a funny-looking very-local label
433 } elsif ( /$T_CONST_LBL/o ) {
435 $chkcat[$i] = 'literal';
438 } elsif ( /^$TUS[@]?__stg_split_marker(\d+)$TPOSTLBL[@]?$/o ) {
440 $chkcat[$i] = 'splitmarker';
443 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_info$TPOSTLBL[@]?$/o ) {
446 $chkcat[$i] = 'infotbl';
447 $chksymb[$i] = $symb;
449 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
451 $infochk{$symb} = $i;
453 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(entry|ret)$TPOSTLBL[@]?$/o ) {
455 $chkcat[$i] = 'slow';
460 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_fast\d*$TPOSTLBL[@]?$/o ) {
462 $chkcat[$i] = 'fast';
467 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_closure$TPOSTLBL[@]?$/o ) {
469 $chkcat[$i] = 'closure';
472 $closurechk{$1} = $i;
474 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_srt$TPOSTLBL[@]?$/o ) {
481 } elsif ( /^$TUS[@]?ghc.*c_ID$TPOSTLBL/o ) {
483 $chkcat[$i] = 'consist';
485 } elsif ( /^($TUS[@]?__gnu_compiled_c|gcc2_compiled\.)$TPOSTLBL/o ) {
488 } elsif ( /^$TUS[A-Za-z0-9_]+\.\d+$TPOSTLBL[@]?$/o
489 || /^$TUS[@]?.*_CAT$TPOSTLBL[@]?$/o # PROF: _entryname_CAT
490 || /^$TUS[@]?CC_.*_struct$TPOSTLBL[@]?$/o # PROF: _CC_ccident_struct
491 || /^$TUS[@]?.*_done$TPOSTLBL[@]?$/o # PROF: _module_done
492 || /^$TUS[@]?_module_registered$TPOSTLBL[@]?$/o # PROF: _module_registered
495 $chkcat[$i] = 'data';
498 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
503 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
508 } elsif ( /^$TUS[@]?CC(S)?_.*$/ ) {
509 # all CC_ symbols go in the data section...
511 $chkcat[$i] = 'data';
514 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(alt|dflt)$TPOSTLBL[@]?$/o ) {
516 $chkcat[$i] = 'misc';
519 #$chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
521 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_vtbl$TPOSTLBL[@]?$/o ) {
523 $chkcat[$i] = 'vector';
528 # As a temporary solution for compiling "foreign export" declarations,
529 # we use global variables to pass arguments from C to STG land.
530 # These declarations live in the .hc file and not in the generated C
531 # stub file, so we let them pass through here.
532 } elsif ( /^[\t ]+\.comm[\t ]+__fexp_.*$/ ) {
534 $chkcat[$i] = 'data';
537 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
538 && /^(_uname|uname|stat|fstat):/ ) {
539 # for some utterly bizarre reason, this platform
540 # likes to drop little local C routines with these names
541 # into each and every .o file that #includes the
542 # relevant system .h file. Yuck. We just don't
543 # tolerate them in .hc files (which we are processing
544 # here). If you need to call one of these things from
545 # Haskell, make a call to your own C wrapper, then
546 # put that C wrapper (which calls one of these) in a
547 # plain .c file. WDP 95/12
549 $chkcat[$i] = 'toss';
552 } elsif ( /^$TUS[@]?[A-Za-z0-9_]/o
553 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
555 && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
556 || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
559 print "Funny global thing?: $_"
560 unless $KNOWN_FUNNY_THING{$thing}
561 || /^$TUS[@]?stg_.*$TPOSTLBL[@]?$/o # RTS internals
562 || /^$TUS[@]__fexp_.*$TPOSTLBL$/o # foreign export
563 || /^$TUS[@]?_reg.*$TPOSTLBL$/o # PROF: __reg<module>
564 || /^$TUS[@]?.*_btm$TPOSTLBL$/o # large bitmaps
565 || /^$TUS[@]?.*_closure_tbl$TPOSTLBL$/o; # closure tables
567 $chkcat[$i] = 'misc';
568 if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
569 { $chksymb[$i] = $thing; }
570 else { $chksymb[$i] = ''; };
572 } else { # simple line (duplicated at the top)
577 $numchks = $#chk + 1;
579 # the division into chunks is imperfect;
580 # we throw some things over the fence into the next
583 # also, there are things we would like to know
584 # about the whole module before we start spitting
587 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
589 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
591 # Alphas: NB: we start meddling at chunk 1, not chunk 0
592 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
593 # spits a ".quad 0" in after the v first ".rdata"; we
594 # detect this special case (tossing the ".quad 0")!
595 local($magic_rdata_seen) = 0;
597 # HPPAs, MIPSen: also start medding at chunk 1
599 if ($TargetPlatform =~ /^powerpc|^rs6000/) {
600 print OUTASM $T_HDR_toc; # yes, we have to put a .toc
601 # in the beginning of every file!
602 %tocequiv = (); # maps toc symbol number to toc symbol
603 %revtocequiv = (); # maps toc symbol to toc symbol number
604 for ($i = 1; $i < $numchks; $i++) {
605 $chk[$i] =~ s/\[RW\]//g;
606 $chk[$i] =~ s/\[DS\]//g;
607 $chk[$i] =~ s/^\.csect .*\[DS\]$//g;
609 if ( $chkcat[$i] eq 'toc' && $chk[$i] !~ /\.byte/ )
610 #ToDo: instead of all these changes, just fix mangle_powerpc_tailjump and delete/ignore these tocs?
611 { $chk[$i] =~ s/$T_MOVE_DIRVS//g;
612 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_fast\d+)/\t\.tc \1\[TC\],\.\2/;
613 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_entry)\n/\t\.tc \1\[TC\],\.\2\n/;
614 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(ret_\S+)/\t\.tc \1\[TC\],\.\2/;
615 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(alt_\S+)/\t\.tc \1\[TC\],\.\2/;
616 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
618 $tocnumber = $chksymb[$i];
620 $tocsymb =~ s/^LC\.\.\d+:\n//;
621 $tocsymb =~ s/^\t\.tc \S+,(\S+)\n/\1/;
622 $tocequiv{$tocnumber} = $tocsymb;
624 } elsif ( $chkcat[$i] eq 'toc' && $chk[$i] =~ /\.byte/ ) {
625 $chkcat[$i] = 'literal';
630 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
631 $c = $chk[$i]; # convenience copy
633 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
635 # toss all prologue stuff; HPPA is pretty weird
637 $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/;
639 # be slightly paranoid to make sure there's
640 # nothing surprising in there
641 if ( $c =~ /--- BEGIN ---/ ) {
642 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
644 if ($TargetPlatform =~ /^i386-/) {
645 $p =~ s/^\tpushl \%edi\n//;
646 $p =~ s/^\tpushl \%esi\n//;
647 $p =~ s/^\tpushl \%ebx\n//;
648 $p =~ s/^\tsubl \$\d+,\%esp\n//;
649 $p =~ s/^\tmovl \$\d+,\%eax\n\tcall __alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
650 } elsif ($TargetPlatform =~ /^m68k-/) {
651 $p =~ s/^\tlink a6,#-?\d.*\n//;
652 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
653 # The above showed up in the asm code,
654 # so I added it here.
655 # I hope it's correct.
657 $p =~ s/^\tmovel d2,sp\@-\n//;
658 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
659 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
660 } elsif ($TargetPlatform =~ /^mips-/) {
661 # the .frame/.mask/.fmask that we use is the same
662 # as that produced by GCC for miniInterpret; this
663 # gives GDB some chance of figuring out what happened
664 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
665 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
666 $p =~ s/^\t\.(mask|fmask).*\n//g;
667 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
668 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
669 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
670 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
671 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
672 $p =~ s/__FRAME__/$FRAME/;
673 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000/) {
674 $p =~ s/^\tmflr 0\n//;
675 $p =~ s/^\tstm \d+,-\d+\(1\)\n//;
676 $p =~ s/^\tstw? 0,\d+\(1\)\n//g;
677 $p =~ s/^\tstw? 1,\d+\(1\)\n//g; #mc
678 $p =~ s/^\tlw?z 0,0\(1\)\n//g; #mc
679 $p =~ s/^\tstw?u 1,-\d+\(1\)\n//;
680 $p =~ s/^\tstw? \d+,-\d+\(1\)\n//g;
681 $p =~ s/^\tstfd \d+,-\d+\(1\)\n//g;
683 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
686 die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
687 && $TargetPlatform !~ /^(powerpc-|i386-)/; #ToDo: remove test
689 # glue together what's left
691 $c =~ s/\n\t\n/\n/; # junk blank line
695 if ( $TargetPlatform =~ /^mips-/ ) {
696 # MIPS: first, this basic sequence may occur "--- END ---" or not
697 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
700 # toss all epilogue stuff; again, paranoidly
701 if ( $c =~ /--- END ---/ ) {
702 if (($r, $e) = split(/--- END ---/, $c)) {
703 if ($TargetPlatform =~ /^i386-/) {
705 $e =~ s/^\tpopl \%edi\n//;
706 $e =~ s/^\tpopl \%esi\n//;
707 $e =~ s/^\tpopl \%ebx\n//;
708 $e =~ s/^\tpopl \%ecx\n//;
709 $e =~ s/^\taddl \$\d+,\%esp\n//;
710 } elsif ($TargetPlatform =~ /^m68k-/) {
711 $e =~ s/^\tunlk a6\n//;
713 } elsif ($TargetPlatform =~ /^mips-/) {
714 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
715 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
716 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
717 $e =~ s/^\tj\t\$31\n//;
718 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
719 $e =~ s/^\taddi 1,1,\d+\n//;
720 $e =~ s/^\tcal 1,\d+\(1\)\n//;
721 $e =~ s/^\tlw?z? \d+,\d+\(1\)\n//;
722 $e =~ s/^\tmtlr 0\n//;
725 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
727 die "Epilogue junk?: $e\n" if $e =~ /^\t[^\.]/
728 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
730 # glue together what's left
732 $c =~ s/\n\t\n/\n/; # junk blank line
736 # On SPARCs, we don't do --- BEGIN/END ---, we just
737 # toss the register-windowing save/restore/ret* instructions
739 if ( $TargetPlatform =~ /^sparc-/ ) {
740 $c =~ s/^\t(save .*|restore|ret|retl)\n//g;
741 # throw away PROLOGUE comments
742 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
745 # On Alphas, the prologue mangling is done a little later (below)
747 # toss all calls to __DISCARD__
748 $c =~ s/^\t(call|jbsr|jal)\s+$TUS[@]?__DISCARD__(\@PLT)?\n//go;
750 # MIPS: that may leave some gratuitous asm macros around
751 # (no harm done; but we get rid of them to be tidier)
752 $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/
753 if $TargetPlatform =~ /^mips-/;
755 # toss stack adjustment after DoSparks
756 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
757 if $TargetPlatform =~ /^m68k-/; # this looks old...
759 if ( $TargetPlatform =~ /^alpha-/ &&
760 ! $magic_rdata_seen &&
761 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
762 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
763 $magic_rdata_seen = 1;
766 # pick some end-things and move them to the next chunk
768 # pin a funny end-thing on (for easier matching):
769 $c .= 'FUNNY#END#THING';
771 while ( $c =~ /$TMOVEDIRVS[@]?FUNNY#END#THING/o ) { # [@]? is a silly hack to avoid having to use curlies for T_PRE_APP
772 # (this SEGVs perl4 on alphas, you see)
775 if ( $i < ($numchks - 1)
776 && ( $to_move =~ /$TCOPYDIRVS/
777 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
778 $chk[$i + 1] = $to_move . $chk[$i + 1];
779 # otherwise they're tossed
782 $c =~ s/$TMOVEDIRVS[@]?FUNNY#END#THING/FUNNY#END#THING/o; # [@]? is a hack (see above)
785 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
787 # toss all prologue stuff, except for loading gp, and the ..ng address
788 if (($p, $r) = split(/^\t\.prologue/, $c)) {
789 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
790 $c = $keep . "..ng:\n";
792 print STDERR "malformed code block ($ent)?\n"
795 $c .= "\t.frame \$30,0,\$26,0\n\t.prologue" . $r;
798 $c =~ s/FUNNY#END#THING//;
800 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
802 $chk[$i] = $c; # update w/ convenience copy
805 if ( $TargetPlatform =~ /^alpha-/ ) {
806 # print out the header stuff first
807 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
808 print OUTASM $chk[0];
810 } elsif ( $TargetPlatform =~ /^hppa/ ) {
811 print OUTASM $chk[0];
813 } elsif ( $TargetPlatform =~ /^mips-/ ) {
814 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
816 # get rid of horrible "<dollar>Revision: .*$" strings
817 local(@lines0) = split(/\n/, $chk[0]);
819 while ( $z <= $#lines0 ) {
820 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
823 while ( $z <= $#lines0 ) {
825 last if $lines0[$z] =~ /[,\t]0x0$/;
831 $chk[0] = join("\n", @lines0);
832 $chk[0] =~ s/\n\n+/\n/;
833 print OUTASM $chk[0];
836 # print out all the literal strings next
837 for ($i = 0; $i < $numchks; $i++) {
838 if ( $chkcat[$i] eq 'literal' ) {
839 print OUTASM $T_HDR_literal, $chk[$i];
840 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
842 $chkcat[$i] = 'DONE ALREADY';
846 # on the HPPA, print out all the bss next
847 if ( $TargetPlatform =~ /^hppa/ ) {
848 for ($i = 1; $i < $numchks; $i++) {
849 if ( $chkcat[$i] eq 'bss' ) {
850 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
851 print OUTASM $chk[$i];
853 $chkcat[$i] = 'DONE ALREADY';
858 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
859 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
861 next if $chkcat[$i] eq 'DONE ALREADY';
863 if ( $chkcat[$i] eq 'misc' ) {
864 if ($chk[$i] ne '') {
865 print OUTASM $T_HDR_misc;
866 if ($TargetPlatform =~ /^powerpc-|^rs6000/) {
867 $chksymb[$i] =~ s/://;
868 #new if ($chksymb[$i] =~ /ret.*upd/ || $KNOWN_FUNNY_THING{$chksymb[$i]}
869 #new || $chksymb[$i] =~ /^$.{T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o )
870 #new { print OUTASM "\t\.globl $chksymb[$i]\n"; }
871 # if ($chksymb[$i] ne '' && $chksymb[$i] !~ /ret_[a-z]/ && $chksymb[$i] !~ /djn_[a-z]/)
872 if ($chksymb[$i] ne '')
873 { print OUTASM "\t\.globl \.$chksymb[$i]\n"; };
874 if ($chk[$i] =~ /TOC\[tc0\], 0\n/)
875 { ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$i]); $printDS = 1;}
876 else { $r = $chk[$i]; $printDS = 0; };
877 $chk[$i] = &mangle_powerpc_tailjump($r);
879 &print_doctored($chk[$i], 0);
880 if ($TargetPlatform =~ /^powerpc-|^rs6000-/ && $printDS) {
881 #ok if ($chksymb[$i] !~ /\_regMain/) {
882 print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
883 print OUTASM "${p}TOC[tc0], 0\n";
888 } elsif ( $chkcat[$i] eq 'toss' ) {
889 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
891 } elsif ( $chkcat[$i] eq 'data' ) {
892 if ($chk[$i] ne '') {
893 print OUTASM $T_HDR_data;
894 print OUTASM $chk[$i];
897 } elsif ( $chkcat[$i] eq 'consist' ) {
898 if ( $chk[$i] =~ /$T_hsc_cc_PAT/o ) {
899 local($consist) = "$1.$2.$3";
901 $consist =~ s/\//./g;
903 $consist =~ s/[^A-Za-z0-9_.]/ZZ/g; # ToDo: properly?
905 # Using a cygnus-2.7-96q4 gcc build on hppas, the
906 # consistency chunk for ghc_cc_ID often (but not always!)
907 # gets lumped with a bunch of .IMPORT directives containing info on
908 # the code or data space nature of external symbols. We can't
909 # toss these, so once the consistency ID has been turned into
910 # a representable symbol, we substitute it for the symbol
911 # that the string was attached to in the first place (ghc_cc_ID.)
912 # (The original string is also substituted away.)
914 # This change may affect the code output on other platforms in
915 # adverse ways, hence we restrict this hack hppa targets only.
918 if ( $TargetPlatform =~ /^hppa/ ) {
919 $chk[$i] =~ s/^$TUS[@]?ghc.*c_ID$TPOSTLBL/$consist/o;
920 $chk[$i] =~ s/\t$T_hsc_cc_PAT/$T_HDR_misc/o;
921 $consist = $chk[$i]; #clumsily
923 print OUTASM $T_HDR_consist, "${consist}${T_POST_LBL}\n";
925 } elsif ( $TargetPlatform !~ /^(mips)-/ ) { # we just don't try in those case (ToDo)
926 # on mips: consistency string is just a v
927 # horrible bunch of .bytes,
928 # which I am too lazy to sort out (WDP 95/05)
930 print STDERR "Couldn't grok consistency: ", $chk[$i];
933 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
934 # we can just re-constitute this one...
935 # NB: we emit _three_ underscores no matter what,
936 # so ghc-split doesn't have to care.
937 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
939 } elsif ( $chkcat[$i] eq 'closure'
940 || $chkcat[$i] eq 'srt'
941 || $chkcat[$i] eq 'infotbl'
942 || $chkcat[$i] eq 'slow'
943 || $chkcat[$i] eq 'fast' ) { # do them in that order
944 $symb = $chksymb[$i];
947 if ( defined($closurechk{$symb}) ) {
948 print OUTASM $T_HDR_closure;
949 print OUTASM $chk[$closurechk{$symb}];
950 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
954 if ( defined($srtchk{$symb}) ) {
955 print OUTASM $T_HDR_srt;
956 print OUTASM $chk[$srtchk{$symb}];
957 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
961 if ( defined($infochk{$symb}) ) {
963 print OUTASM $T_HDR_info;
964 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
965 if ( !defined($slowchk{$symb}) && defined($fastchk{$symb}) ) {
966 $fastname = $chk[$fastchk{$symb}];
967 $fastname =~ s/([_A-Za-z]*_fast\d+):.*(.*\n)*/\1/;
968 $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long $fastname/;
970 $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
971 $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
972 print OUTASM $chk[$infochk{$symb}];
974 # for shared libs --SDM
975 print OUTASM $chk[$infochk{$symb}];
976 # print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
978 # entry code will be put here!
980 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
984 if ( defined($slowchk{$symb}) ) {
986 # teach it to drop through to the fast entry point:
987 $c = $chk[$slowchk{$symb}];
989 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
990 ($p, $r) = split(/TOC\[tc0\], 0\n/, $c);
991 if ($symb =~ /^[_A-Z]/)
993 print OUTASM "\t\.globl \.${chksymb[$i]}_entry\n";
994 print OUTASM "\.csect ${symb}_entry[DS]\n";
995 print OUTASM "${p}TOC[tc0], 0\n";
997 $r =~ s/\.csect \.text\[PR\]\n//; # todo: properly - andre
998 $c = &mangle_powerpc_tailjump($r);
1001 if ( defined($fastchk{$symb}) ) {
1002 if ( $TargetPlatform =~ /^alpha-/ ) {
1003 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1004 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1005 $c =~ s/^\s+ldil.*\n\s+ldo.*\n\s+bv.*\n(.*\n)?\s+\.EXIT/$1\t.EXIT/;
1006 } elsif ( $TargetPlatform =~ /^i386-/ ) {
1007 # Reg alloc depending, gcc generated code may jump to the fast entry point via
1008 # a number of registers.
1009 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edx\n\tjmp \*\%edx\n//;
1010 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%ecx\n\tjmp \*\%ecx\n//;
1011 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%eax\n\tjmp \*\%eax\n//;
1012 # The next two only apply if we're not stealing %esi or %edi.
1013 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%esi\n\tjmp \*\%esi\n// if ($StolenX86Regs < 3);
1014 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edi\n\tjmp \*\%edi\n// if ($StolenX86Regs < 4);
1016 $c =~ s/^\tleal ${T_US}${symb}_fast\d*\@GOTOFF\(\%ebx\),\%eax\n\tjmp \*\%eax\n//;
1017 $c =~ s/^\tjmp \*${T_US}${symb}_fast\d*\@GOT\(\%ebx\)\n//;
1018 $c =~ s/^\.LPR\d+\:\n\tmovl \(\%esp\),\%ebx\n\tret\n//;
1019 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1020 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1021 } elsif ( $TargetPlatform =~ /^m68k-/ ) {
1022 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n\tnop\n//;
1023 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n//;
1024 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ ) {
1025 $c =~ s/^\tb \.${T_US}${symb}_fast\d+\n//;
1026 } elsif ( $TargetPlatform =~ /^sparc-/ ) {
1027 $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n\tnop\n//;
1028 $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n(\t[a-z].*\n)/$1/;
1030 print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n";
1034 if ( $TargetPlatform !~ /^(alpha-|hppa|mips-)/ ) {
1035 # On alphas, hppa: no very good way to look for "dangling"
1036 # references to fast-entry point.
1037 # (questionable re hppa and mips...)
1038 print STDERR "still has jump to fast entry point:\n$c"
1039 if $c =~ /$TUS[@]?$symb[@]?_fast/; # NB: paranoia
1042 print OUTASM $T_HDR_entry;
1044 &print_doctored($c, 1); # NB: the 1!!!
1046 $chkcat[$slowchk{$symb}] = 'DONE ALREADY';
1050 if ( defined($fastchk{$symb}) ) {
1051 $c = $chk[$fastchk{$symb}];
1052 if ( ! defined($slowchk{$symb})
1053 # ToDo: the || clause can go once we're no longer
1054 # concerned about producing exactly the same output as before
1055 #OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/
1057 print OUTASM $T_HDR_fast;
1059 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1060 local(@lbls) = split(/:/, $c);
1061 $fullname = $lbls[0];
1062 $fullname =~ s/$T_MOVE_DIRVS//g;
1063 if ( $fullname =~ /^[A-Z]/)
1064 { print OUTASM "\t\.globl \.${fullname}\n";
1066 # print OUTASM "\t\.lglobl \.${fullname}\n"; #todo: rm - andre
1068 $c =~ s/((.*\n)*)\t.long \S+, TOC\[tc0\], 0\n\.csect \.text\[PR\]\n((.*\n)*)/\1\3/;
1069 $c = &mangle_powerpc_tailjump($c);
1071 &print_doctored($c, 0);
1072 $chkcat[$fastchk{$symb}] = 'DONE ALREADY';
1075 } elsif ( $chkcat[$i] eq 'vector'
1076 || $chkcat[$i] eq 'direct' ) { # do them in that order
1077 $symb = $chksymb[$i];
1080 if ( defined($vectorchk{$symb}) ) {
1081 print OUTASM $T_HDR_vector;
1082 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1083 if ( $symb =~ /^[A-Z]/) {
1084 print OUTASM "\t\.globl \.${symb}_vtbl\n";
1085 print OUTASM "\t\.globl ${symb}_vtbl\n";
1087 $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
1088 print OUTASM ".${symb}_vtbl:\n";
1089 print OUTASM $chk[$vectorchk{$symb}];
1091 # for shared libs --SDM print OUTASM $chk[$vectorchk{$symb}];
1092 print OUTASM $chk[$vectorchk{$symb}];
1093 # print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1095 # direct return code will be put here!
1096 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1098 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1099 # Alphas: the commented nop is for the splitter, to ensure
1100 # that no module ends with a label as the very last
1101 # thing. (The linker will adjust the label to point
1102 # to the first code word of the next module linked in,
1103 # even if alignment constraints cause the label to move!)
1105 print OUTASM "\t# nop\n";
1108 } elsif ( $chkcat[$i] eq 'toc' ) {
1109 # silly optimisation to print tocs, since they come in groups...
1110 print OUTASM $T_HDR_toc;
1112 while ($chkcat[$j] eq 'toc')
1113 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1116 print OUTASM $chk[$j];
1118 $chkcat[$j] = 'DONE ALREADY';
1123 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1127 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1129 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1130 print OUTASM ".csect .text[PR]\n_section_.text:\n.csect .data[RW]\n\t.long _section_.text\n"
1134 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1135 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1140 sub mash_hppa_prologue { # OK, epilogue, too
1143 # toss all prologue stuff
1144 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1146 # Lie about our .CALLINFO
1147 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1154 # toss all epilogue stuff
1155 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1157 # Sorry; we moved the _info stuff to the code segment.
1158 s/_info,DATA/_info,CODE/g;
1165 sub print_doctored {
1166 local($_, $need_fallthru_patch) = @_;
1168 if ( $TargetPlatform !~ /^i386-/
1169 || ! /^\t[a-z]/ # no instructions in here, apparently
1170 || /^${T_US}_reg[A-Za-z0-9_]+${T_POST_LBL}/) {
1174 # OK, must do some x86 **HACKING**
1176 local($entry_patch) = '';
1177 local($exit_patch) = '';
1179 # gotta watch out for weird instructions that
1180 # invisibly smash various regs:
1181 # rep* %ecx used for counting
1182 # scas* %edi used for destination index
1183 # cmps* %e[sd]i used for indices
1184 # loop* %ecx used for counting
1189 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1191 # * GCC used an "STG reg" for its own purposes
1193 # * some secret uses of machine reg, requiring STG reg
1194 # to be saved/restored
1196 # The most dangerous "GCC uses" of an "STG reg" are when
1197 # the reg holds the target of a jmp -- it's tricky to
1198 # insert the patch-up code before we get to the target!
1199 # So here we change the jmps:
1201 # --------------------------------------------------------
1202 # it can happen that we have jumps of the form...
1203 # jmp *<something involving %esp>
1205 # jmp <something involving another naughty register...>
1207 # a reasonably-common case is:
1209 # movl $_blah,<bad-reg>
1213 # the short form may tickle perl bug:
1214 # s/^\tmovl \$${T_US}(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp $T_US$1/g;
1215 s/^\tmovl \$${T_US}(.*),\%eax\n\tjmp \*\%eax/\tjmp $T_US$1/g;
1216 s/^\tmovl \$${T_US}(.*),\%ebx\n\tjmp \*\%ebx/\tjmp $T_US$1/g;
1217 s/^\tmovl \$${T_US}(.*),\%ecx\n\tjmp \*\%ecx/\tjmp $T_US$1/g;
1218 s/^\tmovl \$${T_US}(.*),\%edx\n\tjmp \*\%edx/\tjmp $T_US$1/g;
1220 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1221 s/^\tmovl (.*),\%esi\n\tjmp \*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1222 s/^\tjmp \*(-?\d*)\((.*\%esi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1223 s/^\tjmp \*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1224 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1225 if /(jmp|call) .*\%esi/;
1227 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1228 s/^\tmovl (.*),\%edi\n\tjmp \*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1229 s/^\tjmp \*(-?\d*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1230 s/^\tjmp \*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1231 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1232 if /(jmp|call) .*\%edi/;
1235 # OK, now we can decide what our patch-up code is going to
1238 # Offsets into register table - you'd better update these magic
1239 # numbers should you change its contents!
1240 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1241 local($OFFSET_Hp)=92;
1243 # Note funky ".=" stuff; we're *adding* to these _patch guys
1244 if ( $StolenX86Regs <= 2
1245 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1246 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1247 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1249 # nothing for call_{entry,exit} because %esi is callee-save
1251 if ( $StolenX86Regs <= 3
1252 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1253 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1254 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1256 # nothing for call_{entry,exit} because %edi is callee-save
1259 # --------------------------------------------------------
1260 # next, here we go with non-%esp patching!
1262 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1264 # Before calling GC we must set up the exit condition before the call
1265 # and entry condition when we come back
1267 # fix _all_ non-local jumps:
1269 s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1270 s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1272 s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
1274 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1275 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1277 if ($StolenX86Regs == 2 ) {
1278 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1279 if /^\t(jmp|call) .*\%e(si|di)/;
1280 } elsif ($StolenX86Regs == 3 ) {
1281 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1282 if /^\t(jmp|call) .*\%edi/;
1285 # --------------------------------------------------------
1286 # that's it -- print it
1288 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1292 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1293 print OUTASM $exit_patch;
1294 # ToDo: make it not print if there is a "jmp" at the end
1300 sub init_FUNNY_THINGS {
1301 %KNOWN_FUNNY_THING = (
1303 # "${T_US}stg_.*{T_POST_LBL}", 1,
1308 The following table reversal is used for both info tables and return
1309 vectors. In both cases, we remove the first entry from the table,
1310 reverse the table, put the label at the end, and paste some code
1311 (that which is normally referred to by the first entry in the table)
1312 right after the table itself. (The code pasting is done elsewhere.)
1316 local($symb, $tbl, $discard1) = @_;
1318 local($before) = '';
1320 local(@imports) = (); # hppa only
1323 local(@lines) = split(/\n/, $tbl);
1324 local($i, $j); #local ($i, $extra, $words_to_pad, $j);
1326 # see comment in mangleAsm as to why this silliness is needed.
1327 local($TDOTWORD) = ${T_DOT_WORD};
1328 local($TDOTGLOBAL) = ${T_DOT_GLOBAL};
1329 local($TUS) = ${T_US};
1330 local($TPOSTLBL) = ${T_POST_LBL};
1332 # Deal with the header...
1333 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?$TDOTWORD\s+/o; $i++) {
1334 $label .= $lines[$i] . "\n",
1335 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info$TPOSTLBL[@]?$/o
1336 || $lines[$i] =~ /$TDOTGLOBAL/o
1337 || $lines[$i] =~ /^$TUS[@]?\S+_vtbl$TPOSTLBL[@]?$/o;
1339 $before .= $lines[$i] . "\n"; # otherwise...
1342 # Grab the table data...
1343 if ( $TargetPlatform !~ /^hppa/ ) {
1344 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?$TDOTWORD\s+/o; $i++) {
1345 push(@words, $lines[$i]);
1347 } else { # hppa weirdness
1348 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+($TDOTWORD|\.IMPORT)/; $i++) {
1349 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1350 push(@imports, $lines[$i]);
1352 # We don't use HP's ``function pointers''
1353 # We just use labels in code space, like normal people
1354 $lines[$i] =~ s/P%//;
1355 push(@words, $lines[$i]);
1360 # now throw away the first word (SRT) iff it is empty.
1361 # The .zero business is for Linux/ELF.
1362 # The .skip business is for Sparc/Solaris/ELF.
1363 # The .blockz business is for HPPA.
1364 if ($discard1 && $words[0] =~ /^\t?($TDOTWORD\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1368 # Padding removed to reduce code size and improve performance on Pentiums.
1370 # for 486-cache-friendliness, we want our tables aligned
1371 # on 16-byte boundaries (.align 4). Let's pad:
1372 # $extra = ($#words + 1) % 4;
1373 # $words_to_pad = ($extra == 0) ? 0 : 4 - $extra;
1374 # for ($j = 0; $j < $words_to_pad; $j++) { push(@words, "\t${T_DOT_WORD} 0"); }
1376 for (; $i <= $#lines; $i++) {
1377 $after .= $lines[$i] . "\n";
1380 # Alphas:If we have anonymous text (not part of a procedure), the
1381 # linker may complain about missing exception information. Bleh.
1382 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1383 $before = "\t.ent $1\n" . $before;
1384 $after .= "\t.end $1\n";
1388 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1389 . join("\n", @words) . "\n"
1392 # print STDERR "before=$before\n";
1393 # print STDERR "label=$label\n";
1394 # print STDERR "words=",(reverse @words),"\n";
1395 # print STDERR "after=$after\n";
1401 The HP is a major nuisance. The threaded code mangler moved info
1402 tables from data space to code space, but unthreaded code in the RTS
1403 still has references to info tables in data space. Since the HP
1404 linker is very precise about where symbols live, we need to patch the
1405 references in the unthreaded RTS as well.
1408 sub mini_mangle_asm_hppa {
1409 local($in_asmf, $out_asmf) = @_;
1411 open(INASM, "< $in_asmf")
1412 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1413 open(OUTASM,"> $out_asmf")
1414 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1417 s/_info,DATA/_info,CODE/; # Move _info references to code space
1423 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1424 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1430 sub mini_mangle_asm_powerpc {
1431 local($in_asmf, $out_asmf) = @_;
1433 open(INASM, "< $in_asmf")
1434 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1435 open(OUTASM,"> $out_asmf")
1436 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1439 s/long _PRStart/long ._PRStart/;
1440 s/long _PRIn_/long ._PRIn_/;
1441 s/long _Dummy_(\S+)_entry/long ._Dummy_\1_entry/;
1442 s/long _PRMarking_MarkNextRoot\[DS\]/long ._PRMarking_MarkNextRoot/;
1443 s/long _PRMarking_MarkNextCAF\[DS\]/long ._PRMarking_MarkNextCAF/;
1444 s/long _PRMarking_MarkNextAStack\[DS\]/long ._PRMarking_MarkNextAStack/;
1445 s/long _PRMarking_MarkNextBStack\[DS\]/long ._PRMarking_MarkNextBStack/;
1446 s/\.tc EnterNodeCode\[TC]\,EnterNodeCode\[DS\]/\.tc EnterNodeCode\[TC]\,.EnterNodeCode/; # CONC
1447 s/\.tc CheckHeapCode\[TC]\,CheckHeapCode\[DS\]/\.tc CheckHeapCode\[TC]\,.CheckHeapCode/; # CONC
1452 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1453 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1456 sub mangle_powerpc_tailjump {
1458 local($maybe_more) = 1;
1459 while (($c =~ /\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n/) && $maybe_more)
1462 $lcsymb =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1463 # the checks for r1 and r2 are mostly paranoia...
1465 $r1 =~ s/(.*\n)*\tlw?z? (\d+),LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1467 $r2 =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr (\d+)\n\tbctr\n(.*\n)*/\3/;
1470 $c =~ s/((.*\n)*)\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n((.*\n)*)/\1\tb $tocequiv{$lcsymb}\n\3/;
1476 # make "require"r happy...