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+.*|\.type\s+.*|\.Lfe.*\n\t\.size\s+.*|\.size\s+.*|\.ident.*)\n)';
151 $T_COPY_DIRVS = '\.(globl)';
153 if ( $TargetPlatform =~ /freebsd3/ ) {
154 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
156 $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
159 $T_DOT_WORD = '\.(long|value|byte|zero)';
160 $T_DOT_GLOBAL = '\.globl';
161 $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11)
162 $T_HDR_misc = "\.text\n\t\.align 4\n";
163 $T_HDR_data = "\.data\n\t\.align 4\n"; # ToDo: change align??
164 $T_HDR_consist = "\.text\n";
165 $T_HDR_closure = "\.data\n\t\.align 4\n"; # ToDo: change align?
166 $T_HDR_srt = "\.data\n\t\.align 4\n"; # ToDo: change align?
167 $T_HDR_info = "\.text\n\t\.align 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 ( /^$TUS[@]?([A-Za-z0-9_]+)_ct$TPOSTLBL[@]?$/o ) {
483 $chkcat[$i] = 'data';
486 } elsif ( /^$TUS[@]?ghc.*c_ID$TPOSTLBL/o ) {
488 $chkcat[$i] = 'consist';
490 } elsif ( /^($TUS[@]?__gnu_compiled_c|gcc2_compiled\.)$TPOSTLBL/o ) {
493 } elsif ( /^$TUS[A-Za-z0-9_]+\.\d+$TPOSTLBL[@]?$/o
494 || /^$TUS[@]?.*_CAT$TPOSTLBL[@]?$/o # PROF: _entryname_CAT
495 || /^$TUS[@]?CC_.*_struct$TPOSTLBL[@]?$/o # PROF: _CC_ccident_struct
496 || /^$TUS[@]?.*_done$TPOSTLBL[@]?$/o # PROF: _module_done
497 || /^$TUS[@]?_module_registered$TPOSTLBL[@]?$/o # PROF: _module_registered
500 $chkcat[$i] = 'data';
503 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
508 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
513 } elsif ( /^$TUS[@]?CC(S)?_.*$/ ) {
514 # all CC_ symbols go in the data section...
516 $chkcat[$i] = 'data';
519 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(alt|dflt)$TPOSTLBL[@]?$/o ) {
521 $chkcat[$i] = 'misc';
524 #$chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
526 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_vtbl$TPOSTLBL[@]?$/o ) {
528 $chkcat[$i] = 'vector';
533 # As a temporary solution for compiling "foreign export" declarations,
534 # we use global variables to pass arguments from C to STG land.
535 # These declarations live in the .hc file and not in the generated C
536 # stub file, so we let them pass through here.
537 } elsif ( /^[\t ]+\.comm[\t ]+__fexp_.*$/ ) {
539 $chkcat[$i] = 'data';
542 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
543 && /^(_uname|uname|stat|fstat):/ ) {
544 # for some utterly bizarre reason, this platform
545 # likes to drop little local C routines with these names
546 # into each and every .o file that #includes the
547 # relevant system .h file. Yuck. We just don't
548 # tolerate them in .hc files (which we are processing
549 # here). If you need to call one of these things from
550 # Haskell, make a call to your own C wrapper, then
551 # put that C wrapper (which calls one of these) in a
552 # plain .c file. WDP 95/12
554 $chkcat[$i] = 'toss';
557 } elsif ( /^$TUS[@]?[A-Za-z0-9_]/o
558 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
560 && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
561 || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
564 print "Funny global thing?: $_"
565 unless $KNOWN_FUNNY_THING{$thing}
566 || /^$TUS[@]?stg_.*$TPOSTLBL[@]?$/o # RTS internals
567 || /^$TUS[@]__fexp_.*$TPOSTLBL$/o # foreign export
568 || /^$TUS[@]?_reg.*$TPOSTLBL$/o # PROF: __reg<module>
569 || /^$TUS[@]?.*_btm$TPOSTLBL$/o # large bitmaps
570 || /^$TUS[@]?.*_closure_tbl$TPOSTLBL$/o; # closure tables
572 $chkcat[$i] = 'misc';
573 if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
574 { $chksymb[$i] = $thing; }
575 else { $chksymb[$i] = ''; };
577 } else { # simple line (duplicated at the top)
582 $numchks = $#chk + 1;
584 # the division into chunks is imperfect;
585 # we throw some things over the fence into the next
588 # also, there are things we would like to know
589 # about the whole module before we start spitting
592 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
594 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
596 # Alphas: NB: we start meddling at chunk 1, not chunk 0
597 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
598 # spits a ".quad 0" in after the v first ".rdata"; we
599 # detect this special case (tossing the ".quad 0")!
600 local($magic_rdata_seen) = 0;
602 # HPPAs, MIPSen: also start medding at chunk 1
604 if ($TargetPlatform =~ /^powerpc|^rs6000/) {
605 print OUTASM $T_HDR_toc; # yes, we have to put a .toc
606 # in the beginning of every file!
607 %tocequiv = (); # maps toc symbol number to toc symbol
608 %revtocequiv = (); # maps toc symbol to toc symbol number
609 for ($i = 1; $i < $numchks; $i++) {
610 $chk[$i] =~ s/\[RW\]//g;
611 $chk[$i] =~ s/\[DS\]//g;
612 $chk[$i] =~ s/^\.csect .*\[DS\]$//g;
614 if ( $chkcat[$i] eq 'toc' && $chk[$i] !~ /\.byte/ )
615 #ToDo: instead of all these changes, just fix mangle_powerpc_tailjump and delete/ignore these tocs?
616 { $chk[$i] =~ s/$T_MOVE_DIRVS//g;
617 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_fast\d+)/\t\.tc \1\[TC\],\.\2/;
618 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_entry)\n/\t\.tc \1\[TC\],\.\2\n/;
619 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(ret_\S+)/\t\.tc \1\[TC\],\.\2/;
620 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(alt_\S+)/\t\.tc \1\[TC\],\.\2/;
621 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
623 $tocnumber = $chksymb[$i];
625 $tocsymb =~ s/^LC\.\.\d+:\n//;
626 $tocsymb =~ s/^\t\.tc \S+,(\S+)\n/\1/;
627 $tocequiv{$tocnumber} = $tocsymb;
629 } elsif ( $chkcat[$i] eq 'toc' && $chk[$i] =~ /\.byte/ ) {
630 $chkcat[$i] = 'literal';
635 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
636 $c = $chk[$i]; # convenience copy
638 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
640 # toss all prologue stuff; HPPA is pretty weird
642 $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/;
644 # be slightly paranoid to make sure there's
645 # nothing surprising in there
646 if ( $c =~ /--- BEGIN ---/ ) {
647 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
649 if ($TargetPlatform =~ /^i386-/) {
650 $p =~ s/^\tpushl \%edi\n//;
651 $p =~ s/^\tpushl \%esi\n//;
652 $p =~ s/^\tpushl \%ebx\n//;
653 $p =~ s/^\tsubl \$\d+,\%esp\n//;
654 $p =~ s/^\tmovl \$\d+,\%eax\n\tcall __alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
655 } elsif ($TargetPlatform =~ /^m68k-/) {
656 $p =~ s/^\tlink a6,#-?\d.*\n//;
657 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
658 # The above showed up in the asm code,
659 # so I added it here.
660 # I hope it's correct.
662 $p =~ s/^\tmovel d2,sp\@-\n//;
663 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
664 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
665 } elsif ($TargetPlatform =~ /^mips-/) {
666 # the .frame/.mask/.fmask that we use is the same
667 # as that produced by GCC for miniInterpret; this
668 # gives GDB some chance of figuring out what happened
669 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
670 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
671 $p =~ s/^\t\.(mask|fmask).*\n//g;
672 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
673 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
674 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
675 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
676 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
677 $p =~ s/__FRAME__/$FRAME/;
678 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000/) {
679 $p =~ s/^\tmflr 0\n//;
680 $p =~ s/^\tstm \d+,-\d+\(1\)\n//;
681 $p =~ s/^\tstw? 0,\d+\(1\)\n//g;
682 $p =~ s/^\tstw? 1,\d+\(1\)\n//g; #mc
683 $p =~ s/^\tlw?z 0,0\(1\)\n//g; #mc
684 $p =~ s/^\tstw?u 1,-\d+\(1\)\n//;
685 $p =~ s/^\tstw? \d+,-\d+\(1\)\n//g;
686 $p =~ s/^\tstfd \d+,-\d+\(1\)\n//g;
688 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
691 die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
692 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
694 # glue together what's left
696 $c =~ s/\n\t\n/\n/; # junk blank line
700 if ( $TargetPlatform =~ /^mips-/ ) {
701 # MIPS: first, this basic sequence may occur "--- END ---" or not
702 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
705 # toss all epilogue stuff; again, paranoidly
706 if ( $c =~ /--- END ---/ ) {
707 if (($r, $e) = split(/--- END ---/, $c)) {
708 if ($TargetPlatform =~ /^i386-/) {
710 $e =~ s/^\tpopl \%edi\n//;
711 $e =~ s/^\tpopl \%esi\n//;
712 $e =~ s/^\tpopl \%ecx\n//;
713 $e =~ s/^\taddl \$\d+,\%esp\n//;
714 } elsif ($TargetPlatform =~ /^m68k-/) {
715 $e =~ s/^\tunlk a6\n//;
717 } elsif ($TargetPlatform =~ /^mips-/) {
718 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
719 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
720 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
721 $e =~ s/^\tj\t\$31\n//;
722 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
723 $e =~ s/^\taddi 1,1,\d+\n//;
724 $e =~ s/^\tcal 1,\d+\(1\)\n//;
725 $e =~ s/^\tlw?z? \d+,\d+\(1\)\n//;
726 $e =~ s/^\tmtlr 0\n//;
729 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
731 die "Epilogue junk?: $e\n" if $e =~ /^\t[^\.]/
732 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
734 # glue together what's left
736 $c =~ s/\n\t\n/\n/; # junk blank line
740 # On SPARCs, we don't do --- BEGIN/END ---, we just
741 # toss the register-windowing save/restore/ret* instructions
743 if ( $TargetPlatform =~ /^sparc-/ ) {
744 $c =~ s/^\t(save.*|restore|ret|retl)\n//g;
745 # throw away PROLOGUE comments
746 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
749 # On Alphas, the prologue mangling is done a little later (below)
751 # toss all calls to __DISCARD__
752 $c =~ s/^\t(call|jbsr|jal)\s+$TUS[@]?__DISCARD__\n//go;
754 # MIPS: that may leave some gratuitous asm macros around
755 # (no harm done; but we get rid of them to be tidier)
756 $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/
757 if $TargetPlatform =~ /^mips-/;
759 # toss stack adjustment after DoSparks
760 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
761 if $TargetPlatform =~ /^m68k-/; # this looks old...
763 if ( $TargetPlatform =~ /^alpha-/ &&
764 ! $magic_rdata_seen &&
765 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
766 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
767 $magic_rdata_seen = 1;
770 # pick some end-things and move them to the next chunk
772 # pin a funny end-thing on (for easier matching):
773 $c .= 'FUNNY#END#THING';
775 while ( $c =~ /$TMOVEDIRVS[@]?FUNNY#END#THING/o ) { # [@]? is a silly hack to avoid having to use curlies for T_PRE_APP
776 # (this SEGVs perl4 on alphas, you see)
779 if ( $i < ($numchks - 1)
780 && ( $to_move =~ /$TCOPYDIRVS/
781 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
782 $chk[$i + 1] = $to_move . $chk[$i + 1];
783 # otherwise they're tossed
786 $c =~ s/$TMOVEDIRVS[@]?FUNNY#END#THING/FUNNY#END#THING/o; # [@]? is a hack (see above)
789 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
791 # toss all prologue stuff, except for loading gp, and the ..ng address
792 if (($p, $r) = split(/^\t\.prologue/, $c)) {
793 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
794 $c = $keep . "..ng:\n";
796 print STDERR "malformed code block ($ent)?\n"
799 $c .= "\t.frame \$30,0,\$26,0\n\t.prologue" . $r;
802 $c =~ s/FUNNY#END#THING//;
804 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
806 $chk[$i] = $c; # update w/ convenience copy
809 if ( $TargetPlatform =~ /^alpha-/ ) {
810 # print out the header stuff first
811 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
812 print OUTASM $chk[0];
814 } elsif ( $TargetPlatform =~ /^hppa/ ) {
815 print OUTASM $chk[0];
817 } elsif ( $TargetPlatform =~ /^mips-/ ) {
818 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
820 # get rid of horrible "<dollar>Revision: .*$" strings
821 local(@lines0) = split(/\n/, $chk[0]);
823 while ( $z <= $#lines0 ) {
824 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
827 while ( $z <= $#lines0 ) {
829 last if $lines0[$z] =~ /[,\t]0x0$/;
835 $chk[0] = join("\n", @lines0);
836 $chk[0] =~ s/\n\n+/\n/;
837 print OUTASM $chk[0];
840 # print out all the literal strings next
841 for ($i = 0; $i < $numchks; $i++) {
842 if ( $chkcat[$i] eq 'literal' ) {
843 print OUTASM $T_HDR_literal, $chk[$i];
844 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
846 $chkcat[$i] = 'DONE ALREADY';
850 # on the HPPA, print out all the bss next
851 if ( $TargetPlatform =~ /^hppa/ ) {
852 for ($i = 1; $i < $numchks; $i++) {
853 if ( $chkcat[$i] eq 'bss' ) {
854 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
855 print OUTASM $chk[$i];
857 $chkcat[$i] = 'DONE ALREADY';
862 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
863 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
865 next if $chkcat[$i] eq 'DONE ALREADY';
867 if ( $chkcat[$i] eq 'misc' ) {
868 if ($chk[$i] ne '') {
869 print OUTASM $T_HDR_misc;
870 if ($TargetPlatform =~ /^powerpc-|^rs6000/) {
871 $chksymb[$i] =~ s/://;
872 #new if ($chksymb[$i] =~ /ret.*upd/ || $KNOWN_FUNNY_THING{$chksymb[$i]}
873 #new || $chksymb[$i] =~ /^$.{T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o )
874 #new { print OUTASM "\t\.globl $chksymb[$i]\n"; }
875 # if ($chksymb[$i] ne '' && $chksymb[$i] !~ /ret_[a-z]/ && $chksymb[$i] !~ /djn_[a-z]/)
876 if ($chksymb[$i] ne '')
877 { print OUTASM "\t\.globl \.$chksymb[$i]\n"; };
878 if ($chk[$i] =~ /TOC\[tc0\], 0\n/)
879 { ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$i]); $printDS = 1;}
880 else { $r = $chk[$i]; $printDS = 0; };
881 $chk[$i] = &mangle_powerpc_tailjump($r);
883 &print_doctored($chk[$i], 0);
884 if ($TargetPlatform =~ /^powerpc-|^rs6000-/ && $printDS) {
885 #ok if ($chksymb[$i] !~ /\_regMain/) {
886 print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
887 print OUTASM "${p}TOC[tc0], 0\n";
892 } elsif ( $chkcat[$i] eq 'toss' ) {
893 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
895 } elsif ( $chkcat[$i] eq 'data' ) {
896 if ($chk[$i] ne '') {
897 print OUTASM $T_HDR_data;
898 print OUTASM $chk[$i];
901 } elsif ( $chkcat[$i] eq 'consist' ) {
902 if ( $chk[$i] =~ /$T_hsc_cc_PAT/o ) {
903 local($consist) = "$1.$2.$3";
905 $consist =~ s/\//./g;
907 $consist =~ s/[^A-Za-z0-9_.]/ZZ/g; # ToDo: properly?
909 # Using a cygnus-2.7-96q4 gcc build on hppas, the
910 # consistency chunk for ghc_cc_ID often (but not always!)
911 # gets lumped with a bunch of .IMPORT directives containing info on
912 # the code or data space nature of external symbols. We can't
913 # toss these, so once the consistency ID has been turned into
914 # a representable symbol, we substitute it for the symbol
915 # that the string was attached to in the first place (ghc_cc_ID.)
916 # (The original string is also substituted away.)
918 # This change may affect the code output on other platforms in
919 # adverse ways, hence we restrict this hack hppa targets only.
922 if ( $TargetPlatform =~ /^hppa/ ) {
923 $chk[$i] =~ s/^$TUS[@]?ghc.*c_ID$TPOSTLBL/$consist/o;
924 $chk[$i] =~ s/\t$T_hsc_cc_PAT/$T_HDR_misc/o;
925 $consist = $chk[$i]; #clumsily
927 print OUTASM $T_HDR_consist, "${consist}${T_POST_LBL}\n";
929 } elsif ( $TargetPlatform !~ /^(mips)-/ ) { # we just don't try in those case (ToDo)
930 # on mips: consistency string is just a v
931 # horrible bunch of .bytes,
932 # which I am too lazy to sort out (WDP 95/05)
934 print STDERR "Couldn't grok consistency: ", $chk[$i];
937 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
938 # we can just re-constitute this one...
939 # NB: we emit _three_ underscores no matter what,
940 # so ghc-split doesn't have to care.
941 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
943 } elsif ( $chkcat[$i] eq 'closure'
944 || $chkcat[$i] eq 'srt'
945 || $chkcat[$i] eq 'infotbl'
946 || $chkcat[$i] eq 'slow'
947 || $chkcat[$i] eq 'fast' ) { # do them in that order
948 $symb = $chksymb[$i];
951 if ( defined($closurechk{$symb}) ) {
952 print OUTASM $T_HDR_closure;
953 print OUTASM $chk[$closurechk{$symb}];
954 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
958 if ( defined($srtchk{$symb}) ) {
959 print OUTASM $T_HDR_srt;
960 print OUTASM $chk[$srtchk{$symb}];
961 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
965 if ( defined($infochk{$symb}) ) {
967 print OUTASM $T_HDR_info;
968 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
969 if ( !defined($slowchk{$symb}) && defined($fastchk{$symb}) ) {
970 $fastname = $chk[$fastchk{$symb}];
971 $fastname =~ s/([_A-Za-z]*_fast\d+):.*(.*\n)*/\1/;
972 $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long $fastname/;
974 $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
975 $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
976 print OUTASM $chk[$infochk{$symb}];
978 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
980 # entry code will be put here!
982 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
986 if ( defined($slowchk{$symb}) ) {
988 # teach it to drop through to the fast entry point:
989 $c = $chk[$slowchk{$symb}];
991 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
992 ($p, $r) = split(/TOC\[tc0\], 0\n/, $c);
993 if ($symb =~ /^[_A-Z]/)
995 print OUTASM "\t\.globl \.${chksymb[$i]}_entry\n";
996 print OUTASM "\.csect ${symb}_entry[DS]\n";
997 print OUTASM "${p}TOC[tc0], 0\n";
999 $r =~ s/\.csect \.text\[PR\]\n//; # todo: properly - andre
1000 $c = &mangle_powerpc_tailjump($r);
1003 if ( defined($fastchk{$symb}) ) {
1004 if ( $TargetPlatform =~ /^alpha-/ ) {
1005 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1006 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1007 $c =~ s/^\s+ldil.*\n\s+ldo.*\n\s+bv.*\n(.*\n)?\s+\.EXIT/$1\t.EXIT/;
1008 } elsif ( $TargetPlatform =~ /^i386-/ ) {
1009 # Reg alloc depending, gcc generated code may jump to the fast entry point via
1010 # a number of registers.
1011 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edx\n\tjmp \*\%edx\n//;
1012 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%ecx\n\tjmp \*\%ecx\n//;
1013 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%eax\n\tjmp \*\%eax\n//;
1014 # The next two only apply if we're not stealing %esi or %edi.
1015 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%esi\n\tjmp \*\%esi\n// if ($StolenX86Regs < 3);
1016 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edi\n\tjmp \*\%edi\n// if ($StolenX86Regs < 4);
1017 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1018 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1019 } elsif ( $TargetPlatform =~ /^m68k-/ ) {
1020 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n\tnop\n//;
1021 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n//;
1022 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ ) {
1023 $c =~ s/^\tb \.${T_US}${symb}_fast\d+\n//;
1024 } elsif ( $TargetPlatform =~ /^sparc-/ ) {
1025 $c =~ s/^\tcall\s+${T_US}${symb}_fast\d+,.*\n\t\s*nop\n//;
1026 $c =~ s/^\tcall\s+${T_US}${symb}_fast\d+,.*\n(\t\s*[a-z].*\n)/$1/;
1028 print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n";
1032 if ( $TargetPlatform !~ /^(alpha-|hppa|mips-)/ ) {
1033 # On alphas, hppa: no very good way to look for "dangling"
1034 # references to fast-entry point.
1035 # (questionable re hppa and mips...)
1036 print STDERR "still has jump to fast entry point:\n$c"
1037 if $c =~ /$TUS[@]?$symb[@]?_fast/; # NB: paranoia
1040 print OUTASM $T_HDR_entry;
1042 &print_doctored($c, 1); # NB: the 1!!!
1044 $chkcat[$slowchk{$symb}] = 'DONE ALREADY';
1048 if ( defined($fastchk{$symb}) ) {
1049 $c = $chk[$fastchk{$symb}];
1050 if ( ! defined($slowchk{$symb})
1051 # ToDo: the || clause can go once we're no longer
1052 # concerned about producing exactly the same output as before
1053 #OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/
1055 print OUTASM $T_HDR_fast;
1057 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1058 local(@lbls) = split(/:/, $c);
1059 $fullname = $lbls[0];
1060 $fullname =~ s/$T_MOVE_DIRVS//g;
1061 if ( $fullname =~ /^[A-Z]/)
1062 { print OUTASM "\t\.globl \.${fullname}\n";
1064 # print OUTASM "\t\.lglobl \.${fullname}\n"; #todo: rm - andre
1066 $c =~ s/((.*\n)*)\t.long \S+, TOC\[tc0\], 0\n\.csect \.text\[PR\]\n((.*\n)*)/\1\3/;
1067 $c = &mangle_powerpc_tailjump($c);
1069 &print_doctored($c, 0);
1070 $chkcat[$fastchk{$symb}] = 'DONE ALREADY';
1073 } elsif ( $chkcat[$i] eq 'vector'
1074 || $chkcat[$i] eq 'direct' ) { # do them in that order
1075 $symb = $chksymb[$i];
1078 if ( defined($vectorchk{$symb}) ) {
1079 print OUTASM $T_HDR_vector;
1080 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1081 if ( $symb =~ /^[A-Z]/) {
1082 print OUTASM "\t\.globl \.${symb}_vtbl\n";
1083 print OUTASM "\t\.globl ${symb}_vtbl\n";
1085 $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
1086 print OUTASM ".${symb}_vtbl:\n";
1087 print OUTASM $chk[$vectorchk{$symb}];
1089 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1091 # direct return code will be put here!
1092 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1094 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1095 # Alphas: the commented nop is for the splitter, to ensure
1096 # that no module ends with a label as the very last
1097 # thing. (The linker will adjust the label to point
1098 # to the first code word of the next module linked in,
1099 # even if alignment constraints cause the label to move!)
1101 print OUTASM "\t# nop\n";
1104 } elsif ( $chkcat[$i] eq 'toc' ) {
1105 # silly optimisation to print tocs, since they come in groups...
1106 print OUTASM $T_HDR_toc;
1108 while ($chkcat[$j] eq 'toc')
1109 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1112 print OUTASM $chk[$j];
1114 $chkcat[$j] = 'DONE ALREADY';
1119 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1123 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1125 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1126 print OUTASM ".csect .text[PR]\n_section_.text:\n.csect .data[RW]\n\t.long _section_.text\n"
1130 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1131 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1136 sub mash_hppa_prologue { # OK, epilogue, too
1139 # toss all prologue stuff
1140 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1142 # Lie about our .CALLINFO
1143 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1150 # toss all epilogue stuff
1151 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1153 # Sorry; we moved the _info stuff to the code segment.
1154 s/_info,DATA/_info,CODE/g;
1161 sub print_doctored {
1162 local($_, $need_fallthru_patch) = @_;
1164 if ( $TargetPlatform !~ /^i386-/
1165 || ! /^\t[a-z]/ # no instructions in here, apparently
1166 || /^${T_US}_reg[A-Za-z0-9_]+${T_POST_LBL}/) {
1170 # OK, must do some x86 **HACKING**
1172 local($entry_patch) = '';
1173 local($exit_patch) = '';
1175 # gotta watch out for weird instructions that
1176 # invisibly smash various regs:
1177 # rep* %ecx used for counting
1178 # scas* %edi used for destination index
1179 # cmps* %e[sd]i used for indices
1180 # loop* %ecx used for counting
1185 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1187 # * GCC used an "STG reg" for its own purposes
1189 # * some secret uses of machine reg, requiring STG reg
1190 # to be saved/restored
1192 # The most dangerous "GCC uses" of an "STG reg" are when
1193 # the reg holds the target of a jmp -- it's tricky to
1194 # insert the patch-up code before we get to the target!
1195 # So here we change the jmps:
1197 # --------------------------------------------------------
1198 # it can happen that we have jumps of the form...
1199 # jmp *<something involving %esp>
1201 # jmp <something involving another naughty register...>
1203 # a reasonably-common case is:
1205 # movl $_blah,<bad-reg>
1209 # the short form may tickle perl bug:
1210 # s/^\tmovl \$${T_US}(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp $T_US$1/g;
1211 s/^\tmovl \$${T_US}(.*),\%eax\n\tjmp \*\%eax/\tjmp $T_US$1/g;
1212 s/^\tmovl \$${T_US}(.*),\%ebx\n\tjmp \*\%ebx/\tjmp $T_US$1/g;
1213 s/^\tmovl \$${T_US}(.*),\%ecx\n\tjmp \*\%ecx/\tjmp $T_US$1/g;
1214 s/^\tmovl \$${T_US}(.*),\%edx\n\tjmp \*\%edx/\tjmp $T_US$1/g;
1216 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1217 s/^\tmovl (.*),\%esi\n\tjmp \*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1218 s/^\tjmp \*(-?\d*)\((.*\%esi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1219 s/^\tjmp \*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1220 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1221 if /(jmp|call) .*\%esi/;
1223 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1224 s/^\tmovl (.*),\%edi\n\tjmp \*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1225 s/^\tjmp \*(-?\d*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1226 s/^\tjmp \*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1227 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1228 if /(jmp|call) .*\%edi/;
1231 # OK, now we can decide what our patch-up code is going to
1234 # Offsets into register table - you'd better update these magic
1235 # numbers should you change its contents!
1236 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1237 local($OFFSET_Hp)=92;
1239 # Note funky ".=" stuff; we're *adding* to these _patch guys
1240 if ( $StolenX86Regs <= 2
1241 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1242 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1243 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1245 # nothing for call_{entry,exit} because %esi is callee-save
1247 if ( $StolenX86Regs <= 3
1248 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1249 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1250 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1252 # nothing for call_{entry,exit} because %edi is callee-save
1255 # --------------------------------------------------------
1256 # next, here we go with non-%esp patching!
1258 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1260 # Before calling GC we must set up the exit condition before the call
1261 # and entry condition when we come back
1263 # fix _all_ non-local jumps:
1265 s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1266 s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1268 s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
1270 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1271 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1273 if ($StolenX86Regs == 2 ) {
1274 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1275 if /^\t(jmp|call) .*\%e(si|di)/;
1276 } elsif ($StolenX86Regs == 3 ) {
1277 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1278 if /^\t(jmp|call) .*\%edi/;
1281 # --------------------------------------------------------
1282 # that's it -- print it
1284 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1288 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1289 print OUTASM $exit_patch;
1290 # ToDo: make it not print if there is a "jmp" at the end
1296 sub init_FUNNY_THINGS {
1297 %KNOWN_FUNNY_THING = (
1299 # "${T_US}stg_.*{T_POST_LBL}", 1,
1304 The following table reversal is used for both info tables and return
1305 vectors. In both cases, we remove the first entry from the table,
1306 reverse the table, put the label at the end, and paste some code
1307 (that which is normally referred to by the first entry in the table)
1308 right after the table itself. (The code pasting is done elsewhere.)
1312 local($symb, $tbl, $discard1) = @_;
1314 local($before) = '';
1316 local(@imports) = (); # hppa only
1319 local(@lines) = split(/\n/, $tbl);
1320 local($i, $j); #local ($i, $extra, $words_to_pad, $j);
1322 # see comment in mangleAsm as to why this silliness is needed.
1323 local($TDOTWORD) = ${T_DOT_WORD};
1324 local($TDOTGLOBAL) = ${T_DOT_GLOBAL};
1325 local($TUS) = ${T_US};
1326 local($TPOSTLBL) = ${T_POST_LBL};
1328 # Deal with the header...
1329 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?$TDOTWORD\s+/o; $i++) {
1330 $label .= $lines[$i] . "\n",
1331 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info$TPOSTLBL[@]?$/o
1332 || $lines[$i] =~ /$TDOTGLOBAL/o
1333 || $lines[$i] =~ /^$TUS[@]?\S+_vtbl$TPOSTLBL[@]?$/o;
1335 $before .= $lines[$i] . "\n"; # otherwise...
1338 # Grab the table data...
1339 if ( $TargetPlatform !~ /^hppa/ ) {
1340 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?$TDOTWORD\s+/o; $i++) {
1341 push(@words, $lines[$i]);
1343 } else { # hppa weirdness
1344 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+($TDOTWORD|\.IMPORT)/; $i++) {
1345 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1346 push(@imports, $lines[$i]);
1348 # We don't use HP's ``function pointers''
1349 # We just use labels in code space, like normal people
1350 $lines[$i] =~ s/P%//;
1351 push(@words, $lines[$i]);
1356 # now throw away the first word (SRT) iff it is empty.
1357 # The .zero business is for Linux/ELF.
1358 # The .skip business is for Sparc/Solaris/ELF.
1359 # The .blockz business is for HPPA.
1360 if ($discard1 && $words[0] =~ /^\t?($TDOTWORD\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1364 # Padding removed to reduce code size and improve performance on Pentiums.
1366 # for 486-cache-friendliness, we want our tables aligned
1367 # on 16-byte boundaries (.align 4). Let's pad:
1368 # $extra = ($#words + 1) % 4;
1369 # $words_to_pad = ($extra == 0) ? 0 : 4 - $extra;
1370 # for ($j = 0; $j < $words_to_pad; $j++) { push(@words, "\t${T_DOT_WORD} 0"); }
1372 for (; $i <= $#lines; $i++) {
1373 $after .= $lines[$i] . "\n";
1376 # Alphas:If we have anonymous text (not part of a procedure), the
1377 # linker may complain about missing exception information. Bleh.
1378 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1379 $before = "\t.ent $1\n" . $before;
1380 $after .= "\t.end $1\n";
1384 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1385 . join("\n", @words) . "\n"
1388 # print STDERR "before=$before\n";
1389 # print STDERR "label=$label\n";
1390 # print STDERR "words=",(reverse @words),"\n";
1391 # print STDERR "after=$after\n";
1397 The HP is a major nuisance. The threaded code mangler moved info
1398 tables from data space to code space, but unthreaded code in the RTS
1399 still has references to info tables in data space. Since the HP
1400 linker is very precise about where symbols live, we need to patch the
1401 references in the unthreaded RTS as well.
1404 sub mini_mangle_asm_hppa {
1405 local($in_asmf, $out_asmf) = @_;
1407 open(INASM, "< $in_asmf")
1408 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1409 open(OUTASM,"> $out_asmf")
1410 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1413 s/_info,DATA/_info,CODE/; # Move _info references to code space
1419 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1420 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1426 sub mini_mangle_asm_powerpc {
1427 local($in_asmf, $out_asmf) = @_;
1429 open(INASM, "< $in_asmf")
1430 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1431 open(OUTASM,"> $out_asmf")
1432 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1435 s/long _PRStart/long ._PRStart/;
1436 s/long _PRIn_/long ._PRIn_/;
1437 s/long _Dummy_(\S+)_entry/long ._Dummy_\1_entry/;
1438 s/long _PRMarking_MarkNextRoot\[DS\]/long ._PRMarking_MarkNextRoot/;
1439 s/long _PRMarking_MarkNextCAF\[DS\]/long ._PRMarking_MarkNextCAF/;
1440 s/long _PRMarking_MarkNextAStack\[DS\]/long ._PRMarking_MarkNextAStack/;
1441 s/long _PRMarking_MarkNextBStack\[DS\]/long ._PRMarking_MarkNextBStack/;
1442 s/\.tc EnterNodeCode\[TC]\,EnterNodeCode\[DS\]/\.tc EnterNodeCode\[TC]\,.EnterNodeCode/; # CONC
1443 s/\.tc CheckHeapCode\[TC]\,CheckHeapCode\[DS\]/\.tc CheckHeapCode\[TC]\,.CheckHeapCode/; # CONC
1448 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1449 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1452 sub mangle_powerpc_tailjump {
1454 local($maybe_more) = 1;
1455 while (($c =~ /\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n/) && $maybe_more)
1458 $lcsymb =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1459 # the checks for r1 and r2 are mostly paranoia...
1461 $r1 =~ s/(.*\n)*\tlw?z? (\d+),LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1463 $r2 =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr (\d+)\n\tbctr\n(.*\n)*/\3/;
1466 $c =~ s/((.*\n)*)\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n((.*\n)*)/\1\tb $tocequiv{$lcsymb}\n\3/;
1472 # make "require"r happy...