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_DO_GC = 'PerformGC_wrapper';
58 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
59 $T_CONST_LBL = '^\$C(\d+):$'; # regexp for what such a lbl looks like
62 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\#.*|\.(file|loc)\s+\S+\s+\S+|\.text|\.r?data)\n)';
63 $T_COPY_DIRVS = '^\s*(\#|\.(file|globl|ent|loc))';
65 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
66 $T_DOT_WORD = '\.quad';
67 $T_DOT_GLOBAL = '^\t\.globl';
68 $T_HDR_literal = "\.rdata\n\t\.align 3\n";
69 $T_HDR_misc = "\.text\n\t\.align 3\n";
70 $T_HDR_data = "\.data\n\t\.align 3\n";
71 $T_HDR_consist = "\.text\n";
72 $T_HDR_closure = "\.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_DO_GC = 'PerformGC_wrapper';
85 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
86 $T_CONST_LBL = '^L\$C(\d+)$'; # regexp for what such a lbl looks like
89 $T_MOVE_DIRVS = '^((\s+\.(IMPORT|EXPORT|PARAM).*|\s+\.align\s+\d+|\s+\.(SPACE|SUBSPA)\s+\S+|\s*)\n)';
90 $T_COPY_DIRVS = '^\s+\.(IMPORT|EXPORT)';
92 $T_hsc_cc_PAT = '\.STRING.*\)(hsc|cc) (.*)\\\\x09(.*)\\\\x00';
93 $T_DOT_WORD = '\.word';
94 $T_DOT_GLOBAL = '^\s+\.EXPORT';
95 $T_HDR_literal = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
96 $T_HDR_misc = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
97 $T_HDR_data = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
98 $T_HDR_consist = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
99 $T_HDR_closure = "\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|freebsd|nextstep3|cygwin32)/ ) {
108 # NeXT added but not tested. CaS
110 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
111 $T_US = '_'; # _ if symbols have an underscore on the front
112 $T_DO_GC = '_PerformGC_wrapper';
113 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
114 $T_CONST_LBL = '^LC(\d+):$';
116 $T_X86_PRE_LLBL_PAT = 'L';
117 $T_X86_PRE_LLBL = 'L';
118 $T_X86_BADJMP = '^\tjmp [^L\*]';
120 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+(,0x90)?|\.globl\s+\S+|\.text|\.data|\.stab[^n].*|\.type\s+.*|\.size\s+.*)\n)';
121 $T_COPY_DIRVS = '\.(globl|stab)';
122 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
123 $T_DOT_WORD = '\.long';
124 $T_DOT_GLOBAL = '\.globl';
125 $T_HDR_literal = "\.text\n\t\.align 2\n";
126 $T_HDR_misc = "\.text\n\t\.align 2,0x90\n";
127 $T_HDR_data = "\.data\n\t\.align 2\n";
128 $T_HDR_consist = "\.text\n";
129 $T_HDR_closure = "\.data\n\t\.align 2\n";
130 $T_HDR_info = "\.text\n\t\.align 2\n"; # NB: requires padding
131 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
132 $T_HDR_fast = "\.text\n\t\.align 2,0x90\n";
133 $T_HDR_vector = "\.text\n\t\.align 2\n"; # NB: requires padding
134 $T_HDR_direct = "\.text\n\t\.align 2,0x90\n";
136 #--------------------------------------------------------#
137 } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux)$/ ) {
139 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
140 $T_US = ''; # _ if symbols have an underscore on the front
141 $T_DO_GC = 'PerformGC_wrapper';
142 $T_PRE_APP = # regexp that says what comes before APP/NO_APP
143 ($TargetPlatform =~ /-linux$/) ? '#' : '/' ;
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*(\.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 $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
154 $T_DOT_WORD = '\.long';
155 $T_DOT_GLOBAL = '\.globl';
156 $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11)
157 $T_HDR_misc = "\.text\n\t\.align 16\n";
158 $T_HDR_data = "\.data\n\t\.align 4\n"; # ToDo: change align??
159 $T_HDR_consist = "\.text\n";
160 $T_HDR_closure = "\.data\n\t\.align 4\n"; # ToDo: change align?
161 $T_HDR_info = "\.text\n\t\.align 16\n"; # NB: requires padding
162 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
163 $T_HDR_fast = "\.text\n\t\.align 16\n";
164 $T_HDR_vector = "\.text\n\t\.align 16\n"; # NB: requires padding
165 $T_HDR_direct = "\.text\n\t\.align 16\n";
167 #--------------------------------------------------------#
168 } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
170 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
171 $T_US = '_'; # _ if symbols have an underscore on the front
172 $T_DO_GC = '_PerformGC_wrapper';
173 $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
174 $T_CONST_LBL = '^LC(\d+):$';
177 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
178 $T_COPY_DIRVS = '\.(globl|proc|stab)';
179 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
181 $T_DOT_WORD = '\.long';
182 $T_DOT_GLOBAL = '\.globl';
183 $T_HDR_literal = "\.text\n\t\.even\n";
184 $T_HDR_misc = "\.text\n\t\.even\n";
185 $T_HDR_data = "\.data\n\t\.even\n";
186 $T_HDR_consist = "\.text\n";
187 $T_HDR_closure = "\.data\n\t\.even\n";
188 $T_HDR_info = "\.text\n\t\.even\n";
189 $T_HDR_entry = "\.text\n\t\.even\n";
190 $T_HDR_fast = "\.text\n\t\.even\n";
191 $T_HDR_vector = "\.text\n\t\.even\n";
192 $T_HDR_direct = "\.text\n\t\.even\n";
194 #--------------------------------------------------------#
195 } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
197 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
198 $T_US = ''; # _ if symbols have an underscore on the front
199 $T_DO_GC = 'PerformGC_wrapper';
200 $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
201 $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
204 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
205 $T_COPY_DIRVS = '\.(globl|ent)';
207 $T_hsc_cc_PAT = 'I WAS TOO LAZY TO DO THIS BIT (WDP 95/05)';
208 $T_DOT_WORD = '\.word';
209 $T_DOT_GLOBAL = '^\t\.globl';
210 $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
211 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
212 $T_HDR_data = "\t\.data\n\t\.align 2\n";
213 $T_HDR_consist = 'TOO LAZY TO DO THIS TOO';
214 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
215 $T_HDR_info = "\t\.text\n\t\.align 2\n";
216 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
217 $T_HDR_fast = "\t\.text\n\t\.align 2\n";
218 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
219 $T_HDR_direct = "\t\.text\n\t\.align 2\n";
221 #--------------------------------------------------------#
222 } elsif ( $TargetPlatform =~ /^powerpc-.*|^rs6000-.*/ ) {
224 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
225 $T_US = ''; # _ if symbols have an underscore on the front
226 $T_DO_GC = '\.PerformGC_wrapper';
227 $T_PRE_APP = 'NOT APPLICABLE'; # regexp that says what comes before APP/NO_APP
228 $T_CONST_LBL = 'NOT APPLICABLE'; # regexp for what such a lbl looks like
231 $T_MOVE_DIRVS = '^(\s*(\.toc|\.align \d+|\.csect \S+|\t\.?l?globl \S+)\n)';
232 $T_COPY_DIRVS = '\.(l?globl)';
234 $T_hsc_cc_PAT = '\.byte.*\)(hsc|cc) (.*)"\n\t\.byte \d+\n\t\.byte "(.*)"\n\t\.byte \d+';
235 $T_DOT_WORD = '\.long';
236 $T_DOT_GLOBAL = '\.globl';
237 $T_HDR_toc = "\.toc\n";
238 $T_HDR_literal = "\.csect .data[RW]\n\t\.align 2\n"; #not RO!?
239 $T_HDR_misc = "# misc\n\.csect \.text[PR]\n\t\.align 2\n";
240 $T_HDR_data = "# data\n\.csect \.data[RW]\n\t\.align 2\n";
241 $T_HDR_consist = "# consist\n\.csect \.data[RW]\n\t\.align 2\n";
242 $T_HDR_closure = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
243 $T_HDR_info = "# info\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
244 $T_HDR_entry = "# entry\n\.csect \.text[PR]\n\t\.align 2\n";
245 $T_HDR_fast = "# fast\n\.csect \.text[PR]\n\t\.align 2\n";
246 $T_HDR_vector = "# vector\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
247 $T_HDR_direct = "# direct\n";
249 #--------------------------------------------------------#
250 } elsif ( $TargetPlatform =~ /^sparc-.*-solaris2/ ) {
252 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
253 $T_US = ''; # _ if symbols have an underscore on the front
254 $T_DO_GC = 'PerformGC_wrapper';
255 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
256 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
259 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\.section.*|\s+\.type.*|\s+\.size.*)\n)';
260 $T_COPY_DIRVS = '\.(global|proc|stab)';
262 $T_hsc_cc_PAT = '\.asciz.*\)(hsc|cc) (.*)\\\\t(.*)"';
263 $T_DOT_WORD = '\.word';
264 $T_DOT_GLOBAL = '^\t\.global';
265 $T_HDR_literal = "\.text\n\t\.align 8\n";
266 $T_HDR_misc = "\.text\n\t\.align 4\n";
267 $T_HDR_data = "\.data\n\t\.align 8\n";
268 $T_HDR_consist = "\.text\n";
269 $T_HDR_closure = "\.data\n\t\.align 4\n";
270 $T_HDR_info = "\.text\n\t\.align 4\n";
271 $T_HDR_entry = "\.text\n\t\.align 4\n";
272 $T_HDR_fast = "\.text\n\t\.align 4\n";
273 $T_HDR_vector = "\.text\n\t\.align 4\n";
274 $T_HDR_direct = "\.text\n\t\.align 4\n";
276 #--------------------------------------------------------#
277 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
279 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
280 $T_US = '_'; # _ if symbols have an underscore on the front
281 $T_DO_GC = '_PerformGC_wrapper';
282 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
283 $T_CONST_LBL = '^LC(\d+):$';
286 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
287 $T_COPY_DIRVS = '\.(global|proc|stab)';
288 $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
290 $T_DOT_WORD = '\.word';
291 $T_DOT_GLOBAL = '^\t\.global';
292 $T_HDR_literal = "\.text\n\t\.align 8\n";
293 $T_HDR_misc = "\.text\n\t\.align 4\n";
294 $T_HDR_data = "\.data\n\t\.align 8\n";
295 $T_HDR_consist = "\.text\n";
296 $T_HDR_closure = "\.data\n\t\.align 4\n";
297 $T_HDR_info = "\.text\n\t\.align 4\n";
298 $T_HDR_entry = "\.text\n\t\.align 4\n";
299 $T_HDR_fast = "\.text\n\t\.align 4\n";
300 $T_HDR_vector = "\.text\n\t\.align 4\n";
301 $T_HDR_direct = "\.text\n\t\.align 4\n";
303 #--------------------------------------------------------#
305 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
310 print STDERR "T_STABBY: $T_STABBY\n";
311 print STDERR "T_US: $T_US\n";
312 print STDERR "T_DO_GC: $T_DO_GC\n";
313 print STDERR "T_PRE_APP: $T_PRE_APP\n";
314 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
315 print STDERR "T_POST_LBL: $T_POST_LBL\n";
316 if ( $TargetPlatform =~ /^i386-/ ) {
317 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
318 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
319 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
321 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
322 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
323 print STDERR "T_hsc_cc_PAT: $T_hsc_cc_PAT\n";
324 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
325 print STDERR "T_HDR_literal: $T_HDR_literal\n";
326 print STDERR "T_HDR_misc: $T_HDR_misc\n";
327 print STDERR "T_HDR_data: $T_HDR_data\n";
328 print STDERR "T_HDR_consist: $T_HDR_consist\n";
329 print STDERR "T_HDR_closure: $T_HDR_closure\n";
330 print STDERR "T_HDR_info: $T_HDR_info\n";
331 print STDERR "T_HDR_entry: $T_HDR_entry\n";
332 print STDERR "T_HDR_fast: $T_HDR_fast\n";
333 print STDERR "T_HDR_vector: $T_HDR_vector\n";
334 print STDERR "T_HDR_direct: $T_HDR_direct\n";
340 %************************************************************************
342 \subsection{Mangle away}
344 %************************************************************************
348 local($in_asmf, $out_asmf) = @_;
350 # multi-line regexp matching:
355 &init_TARGET_STUFF();
356 &init_FUNNY_THINGS();
358 # perl4 on alphas SEGVs when give ${foo} substitutions in patterns.
359 # To avoid them we declare some locals that allows to avoid using curlies.
360 local($TUS) = ${T_US};
361 local($TPOSTLBL) = ${T_POST_LBL};
362 local($TMOVEDIRVS) = ${T_MOVE_DIRVS};
363 local($TPREAPP) = ${T_PRE_APP};
364 local($TCOPYDIRVS) = ${T_COPY_DIRVS};
365 local($TDOTWORD) = ${T_DOT_WORD};
367 open(INASM, "< $in_asmf")
368 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
369 open(OUTASM,"> $out_asmf")
370 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
372 # read whole file, divide into "chunks":
373 # record some info about what we've found...
375 @chk = (); # contents of the chunk
376 $numchks = 0; # number of them
377 @chkcat = (); # what category of thing in each chunk
378 @chksymb = (); # what symbol(base) is defined in this chunk
379 %slowchk = (); # ditto, its regular "slow" entry code
380 %fastchk = (); # ditto, fast entry code
381 %closurechk = (); # ditto, the (static) closure
382 %infochk = (); # given a symbol base, say what chunk its info tbl is in
383 %vectorchk = (); # ditto, return vector table
384 %directchk = (); # ditto, direct return code
385 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
387 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
390 next if $T_STABBY && /^\.stab.*$TUS[@]?__stg_split_marker/o;
391 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
392 next if /$TPREAPP(NO_)?APP/o;
393 next if /^;/ && $TargetPlatform =~ /^hppa/;
395 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|powerpc|rs6000)-/;
397 last if /^_section_\.text:$/ && $TargetPlatform =~ /^powerpc-|^rs6000-/;
399 if ( $TargetPlatform =~ /^mips-/
400 && /^\t\.(globl \S+ \.text|comm\t)/ ) {
401 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
403 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
404 # duplicated from the bottom
408 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
409 # Alphas: Local labels not to be confused with new chunks
412 # NB: all the rest start with a non-space
414 } elsif ( $TargetPlatform =~ /^mips-/
415 && /^\d+:/ ) { # a funny-looking very-local label
418 } elsif ( /$T_CONST_LBL/o ) {
420 $chkcat[$i] = 'literal';
423 } elsif ( /^$TUS[@]?__stg_split_marker(\d+)$TPOSTLBL[@]?$/o ) {
425 $chkcat[$i] = 'splitmarker';
428 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_info$TPOSTLBL[@]?$/o ) {
431 $chkcat[$i] = 'infotbl';
432 $chksymb[$i] = $symb;
434 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
436 $infochk{$symb} = $i;
438 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_entry$TPOSTLBL[@]?$/o ) {
440 $chkcat[$i] = 'slow';
445 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_fast\d+$TPOSTLBL[@]?$/o ) {
447 $chkcat[$i] = 'fast';
452 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_closure$TPOSTLBL[@]?$/o ) {
454 $chkcat[$i] = 'closure';
457 $closurechk{$1} = $i;
459 } elsif ( /^$TUS[@]?ghc.*c_ID$TPOSTLBL/o ) {
461 $chkcat[$i] = 'consist';
463 } elsif ( /^($TUS[@]?__gnu_compiled_c|gcc2_compiled\.)$TPOSTLBL/o ) {
466 } elsif ( /^$TUS[@]?ErrorIO_call_count$TPOSTLBL[@]?$/o # HACK!!!!
467 || /^$TUS[A-Za-z0-9_]+\.\d+$TPOSTLBL[@]?$/o
468 || /^$TUS[@]?.*_CAT$TPOSTLBL[@]?$/o # PROF: _entryname_CAT
469 || /^$TUS[@]?CC_.*_struct$TPOSTLBL[@]?$/o # PROF: _CC_ccident_struct
470 || /^$TUS[@]?.*_done$TPOSTLBL[@]?$/o # PROF: _module_done
471 || /^$TUS[@]?_module_registered$TPOSTLBL[@]?$/o # PROF: _module_registered
474 $chkcat[$i] = 'data';
477 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
482 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
487 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^CC_.*$/ ) {
488 # all CC_ symbols go in the data section...
490 $chkcat[$i] = 'data';
493 } elsif ( /^($TUS[@]?(ret_|djn_)[A-Za-z0-9_]+)/o ) {
495 $chkcat[$i] = 'misc';
498 $chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
500 } elsif ( /^$TUS[@]?vtbl_([A-Za-z0-9_]+)$TPOSTLBL[@]?$/o ) {
502 $chkcat[$i] = 'vector';
507 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)DirectReturn$TPOSTLBL[@]?$/o ) {
509 $chkcat[$i] = 'direct';
514 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_upd$TPOSTLBL[@]?$/o ) {
516 $chkcat[$i] = 'misc';
517 print STDERR "_upd!!!!! I guess this code is dead!!!!\n";
518 # I guess this is never entered, since all _upds are
519 # either vtbl_'s or ret_'s, caught above. - andre
522 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
523 && /^(_uname|uname|stat|fstat):/ ) {
524 # for some utterly bizarre reason, this platform
525 # likes to drop little local C routines with these names
526 # into each and every .o file that #includes the
527 # relevant system .h file. Yuck. We just don't
528 # tolerate them in .hc files (which we are processing
529 # here). If you need to call one of these things from
530 # Haskell, make a call to your own C wrapper, then
531 # put that C wrapper (which calls one of these) in a
532 # plain .c file. WDP 95/12
534 $chkcat[$i] = 'toss';
537 } elsif ( /^$TUS[@]?[A-Za-z0-9_]/o
538 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
540 && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
541 || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
544 print STDERR "Funny global thing?: $_"
545 unless $KNOWN_FUNNY_THING{$thing}
546 || /^$TUS[@]?_(PRIn|PRStart).*$TPOSTLBL[@]?$/o # pointer reversal GC routines
547 || /^$TUS[@]?CC_.*$TPOSTLBL$/o # PROF: _CC_ccident ([@]? is a silly hack (see above))
548 || /^$TUS[@]?_reg.*$TPOSTLBL$/o; # PROF: __reg<module>
550 $chkcat[$i] = 'misc';
551 if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
552 { $chksymb[$i] = $thing; }
553 else { $chksymb[$i] = ''; };
555 } else { # simple line (duplicated at the top)
560 $numchks = $#chk + 1;
562 # the division into chunks is imperfect;
563 # we throw some things over the fence into the next
566 # also, there are things we would like to know
567 # about the whole module before we start spitting
570 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
572 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
574 # Alphas: NB: we start meddling at chunk 1, not chunk 0
575 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
576 # spits a ".quad 0" in after the v first ".rdata"; we
577 # detect this special case (tossing the ".quad 0")!
578 local($magic_rdata_seen) = 0;
580 # HPPAs, MIPSen: also start medding at chunk 1
582 if ($TargetPlatform =~ /^powerpc|^rs6000/) {
583 print OUTASM $T_HDR_toc; # yes, we have to put a .toc
584 # in the beginning of every file!
585 %tocequiv = (); # maps toc symbol number to toc symbol
586 %revtocequiv = (); # maps toc symbol to toc symbol number
587 for ($i = 1; $i < $numchks; $i++) {
588 $chk[$i] =~ s/\[RW\]//g;
589 $chk[$i] =~ s/\[DS\]//g;
590 $chk[$i] =~ s/^\.csect .*\[DS\]$//g;
592 if ( $chkcat[$i] eq 'toc' && $chk[$i] !~ /\.byte/ )
593 #ToDo: instead of all these changes, just fix mangle_powerpc_tailjump and delete/ignore these tocs?
594 { $chk[$i] =~ s/$T_MOVE_DIRVS//g;
595 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_fast\d+)/\t\.tc \1\[TC\],\.\2/;
596 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_entry)\n/\t\.tc \1\[TC\],\.\2\n/;
597 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(ret_\S+)/\t\.tc \1\[TC\],\.\2/;
598 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(djn_\S+)/\t\.tc \1\[TC\],\.\2/;
599 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
600 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],ErrorIO_innards/\t\.tc \1\[TC\],\.ErrorIO_innards/;
601 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],startStgWorld/\t\.tc \1\[TC\],\.startStgWorld/;
602 $chk[$i] =~ s/\.tc UpdatePAP\[TC\],UpdatePAP/\.tc UpdatePAP\[TC\],\.UpdatePAP/;
603 $chk[$i] =~ s/\.tc _regMain\[TC\],_regMain/\.tc _regMain\[TC\],\._regMain/; #PROF
604 $chk[$i] =~ s/\.tc resumeThread\[TC\],resumeThread/\.tc resumeThread\[TC\],\.resumeThread/; #CONC
605 $chk[$i] =~ s/\.tc EnterNodeCode\[TC\],EnterNodeCode/\.tc EnterNodeCode\[TC\],\.EnterNodeCode/; #CONC
606 $chk[$i] =~ s/\.tc StackUnderflowEnterNode\[TC\],StackUnderflowEnterNode/\.tc StackUnderflowEnterNode\[TC\],\.StackUnderflowEnterNode/; #CONC
607 $chk[$i] =~ s/\.tc stopThreadDirectReturn\[TC\],stopThreadDirectReturn/\.tc stopThreadDirectReturn\[TC\],\.stopThreadDirectReturn/; #CONC
608 $chk[$i] =~ s/\.tc CommonUnderflow\[TC\],CommonUnderflow/\.tc CommonUnderflow\[TC\],\.CommonUnderflow/; #PAR
609 $chk[$i] =~ s/\.tc IndUpdRetDir\[TC\],IndUpdRetDir/\.tc IndUpdRetDir\[TC\],\.IndUpdRetDir/;
610 $chk[$i] =~ s/\t\.tc (_PRStart_\S+)\[TC\],_PRStart_\S+/\t\.tc \1\[TC\],\.\1/;
612 $tocnumber = $chksymb[$i];
614 $tocsymb =~ s/^LC\.\.\d+:\n//;
615 $tocsymb =~ s/^\t\.tc \S+,(\S+)\n/\1/;
616 $tocequiv{$tocnumber} = $tocsymb;
618 } elsif ( $chkcat[$i] eq 'toc' && $chk[$i] =~ /\.byte/ ) {
619 $chkcat[$i] = 'literal';
624 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
625 $c = $chk[$i]; # convenience copy
627 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
629 # toss all prologue stuff; HPPA is pretty weird
631 $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/;
633 # be slightly paranoid to make sure there's
634 # nothing surprising in there
635 if ( $c =~ /--- BEGIN ---/ ) {
636 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
638 if ($TargetPlatform =~ /^i386-/) {
639 $p =~ s/^\tpushl \%edi\n//;
640 $p =~ s/^\tpushl \%esi\n//;
641 $p =~ s/^\tsubl \$\d+,\%esp\n//;
642 $p =~ s/^\tmovl \$\d+,\%eax\n\tcall __alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
643 } elsif ($TargetPlatform =~ /^m68k-/) {
644 $p =~ s/^\tlink a6,#-?\d.*\n//;
645 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
646 # The above showed up in the asm code,
647 # so I added it here.
648 # I hope it's correct.
650 $p =~ s/^\tmovel d2,sp\@-\n//;
651 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
652 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
653 } elsif ($TargetPlatform =~ /^mips-/) {
654 # the .frame/.mask/.fmask that we use is the same
655 # as that produced by GCC for miniInterpret; this
656 # gives GDB some chance of figuring out what happened
657 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
658 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
659 $p =~ s/^\t\.(mask|fmask).*\n//g;
660 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
661 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
662 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
663 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
664 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
665 $p =~ s/__FRAME__/$FRAME/;
666 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000/) {
667 $p =~ s/^\tmflr 0\n//;
668 $p =~ s/^\tstm \d+,-\d+\(1\)\n//;
669 $p =~ s/^\tstw? 0,\d+\(1\)\n//g;
670 $p =~ s/^\tstw? 1,\d+\(1\)\n//g; #mc
671 $p =~ s/^\tlw?z 0,0\(1\)\n//g; #mc
672 $p =~ s/^\tstw?u 1,-\d+\(1\)\n//;
673 $p =~ s/^\tstw? \d+,-\d+\(1\)\n//g;
674 $p =~ s/^\tstfd \d+,-\d+\(1\)\n//g;
676 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
679 die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
680 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
682 # glue together what's left
684 $c =~ s/\n\t\n/\n/; # junk blank line
688 if ( $TargetPlatform =~ /^mips-/ ) {
689 # MIPS: first, this basic sequence may occur "--- END ---" or not
690 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
693 # toss all epilogue stuff; again, paranoidly
694 if ( $c =~ /--- END ---/ ) {
695 if (($r, $e) = split(/--- END ---/, $c)) {
696 if ($TargetPlatform =~ /^i386-/) {
698 $e =~ s/^\tpopl \%edi\n//;
699 $e =~ s/^\tpopl \%esi\n//;
700 $e =~ s/^\taddl \$\d+,\%esp\n//;
701 } elsif ($TargetPlatform =~ /^m68k-/) {
702 $e =~ s/^\tunlk a6\n//;
704 } elsif ($TargetPlatform =~ /^mips-/) {
705 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
706 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
707 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
708 $e =~ s/^\tj\t\$31\n//;
709 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
710 $e =~ s/^\taddi 1,1,\d+\n//;
711 $e =~ s/^\tcal 1,\d+\(1\)\n//;
712 $e =~ s/^\tlw?z? \d+,\d+\(1\)\n//;
713 $e =~ s/^\tmtlr 0\n//;
716 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
718 die "Epilogue junk?: $e\n" if $e =~ /^\t[^\.]/
719 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
721 # glue together what's left
723 $c =~ s/\n\t\n/\n/; # junk blank line
727 # On SPARCs, we don't do --- BEGIN/END ---, we just
728 # toss the register-windowing save/restore/ret* instructions
730 if ( $TargetPlatform =~ /^sparc-/ ) {
731 $c =~ s/^\t(save .*|restore|ret|retl)\n//g;
732 # throw away PROLOGUE comments
733 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
736 # On Alphas, the prologue mangling is done a little later (below)
738 # toss all calls to __DISCARD__
739 $c =~ s/^\t(call|jbsr|jal)\s+$TUS[@]?__DISCARD__\n//go;
741 # MIPS: that may leave some gratuitous asm macros around
742 # (no harm done; but we get rid of them to be tidier)
743 $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/
744 if $TargetPlatform =~ /^mips-/;
746 # toss stack adjustment after DoSparks
747 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
748 if $TargetPlatform =~ /^m68k-/; # this looks old...
750 if ( $TargetPlatform =~ /^alpha-/ &&
751 ! $magic_rdata_seen &&
752 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
753 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
754 $magic_rdata_seen = 1;
757 # pick some end-things and move them to the next chunk
759 # pin a funny end-thing on (for easier matching):
760 $c .= 'FUNNY#END#THING';
762 while ( $c =~ /$TMOVEDIRVS[@]?FUNNY#END#THING/o ) { # [@]? is a silly hack to avoid having to use curlies for T_PRE_APP
763 # (this SEGVs perl4 on alphas, you see)
766 if ( $i < ($numchks - 1)
767 && ( $to_move =~ /$TCOPYDIRVS/
768 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
769 $chk[$i + 1] = $to_move . $chk[$i + 1];
770 # otherwise they're tossed
773 $c =~ s/$TMOVEDIRVS[@]?FUNNY#END#THING/FUNNY#END#THING/o; # [@]? is a hack (see above)
776 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
778 # toss all prologue stuff, except for loading gp, and the ..ng address
779 if (($p, $r) = split(/^\t\.prologue/, $c)) {
780 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
781 $c = $keep . "..ng:\n";
783 print STDERR "malformed code block ($ent)?\n"
786 $c .= "\t.frame \$30,0,\$26,0\n\t.prologue" . $r;
789 $c =~ s/FUNNY#END#THING//;
791 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
793 $chk[$i] = $c; # update w/ convenience copy
796 if ( $TargetPlatform =~ /^alpha-/ ) {
797 # print out the header stuff first
798 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
799 print OUTASM $chk[0];
801 } elsif ( $TargetPlatform =~ /^hppa/ ) {
802 print OUTASM $chk[0];
804 } elsif ( $TargetPlatform =~ /^mips-/ ) {
805 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
807 # get rid of horrible "<dollar>Revision: .*$" strings
808 local(@lines0) = split(/\n/, $chk[0]);
810 while ( $z <= $#lines0 ) {
811 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
814 while ( $z <= $#lines0 ) {
816 last if $lines0[$z] =~ /[,\t]0x0$/;
822 $chk[0] = join("\n", @lines0);
823 $chk[0] =~ s/\n\n+/\n/;
824 print OUTASM $chk[0];
827 # print out all the literal strings next
828 for ($i = 0; $i < $numchks; $i++) {
829 if ( $chkcat[$i] eq 'literal' ) {
830 print OUTASM $T_HDR_literal, $chk[$i];
831 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
833 $chkcat[$i] = 'DONE ALREADY';
837 # on the HPPA, print out all the bss next
838 if ( $TargetPlatform =~ /^hppa/ ) {
839 for ($i = 1; $i < $numchks; $i++) {
840 if ( $chkcat[$i] eq 'bss' ) {
841 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
842 print OUTASM $chk[$i];
844 $chkcat[$i] = 'DONE ALREADY';
849 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
850 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
852 next if $chkcat[$i] eq 'DONE ALREADY';
854 if ( $chkcat[$i] eq 'misc' ) {
855 if ($chk[$i] ne '') {
856 print OUTASM $T_HDR_misc;
857 if ($TargetPlatform =~ /^powerpc-|^rs6000/) {
858 $chksymb[$i] =~ s/://;
859 #new if ($chksymb[$i] =~ /ret.*upd/ || $KNOWN_FUNNY_THING{$chksymb[$i]}
860 #new || $chksymb[$i] =~ /^$.{T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o )
861 #new { print OUTASM "\t\.globl $chksymb[$i]\n"; }
862 # if ($chksymb[$i] ne '' && $chksymb[$i] !~ /ret_[a-z]/ && $chksymb[$i] !~ /djn_[a-z]/)
863 if ($chksymb[$i] ne '')
864 { print OUTASM "\t\.globl \.$chksymb[$i]\n"; };
865 if ($chk[$i] =~ /TOC\[tc0\], 0\n/)
866 { ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$i]); $printDS = 1;}
867 else { $r = $chk[$i]; $printDS = 0; };
868 $chk[$i] = &mangle_powerpc_tailjump($r);
870 &print_doctored($chk[$i], 0);
871 if ($TargetPlatform =~ /^powerpc-|^rs6000-/ && $printDS) {
872 #ok if ($chksymb[$i] !~ /\_regMain/) {
873 print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
874 print OUTASM "${p}TOC[tc0], 0\n";
879 } elsif ( $chkcat[$i] eq 'toss' ) {
880 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
882 } elsif ( $chkcat[$i] eq 'data' ) {
883 if ($chk[$i] ne '') {
884 print OUTASM $T_HDR_data;
885 print OUTASM $chk[$i];
888 } elsif ( $chkcat[$i] eq 'consist' ) {
889 if ( $chk[$i] =~ /$T_hsc_cc_PAT/o ) {
890 local($consist) = "$1.$2.$3";
892 $consist =~ s/\//./g;
894 $consist =~ s/[^A-Za-z0-9_.]/ZZ/g; # ToDo: properly?
895 print OUTASM $T_HDR_consist, "${consist}${T_POST_LBL}\n";
897 } elsif ( $TargetPlatform !~ /^(mips)-/ ) { # we just don't try in those case (ToDo)
898 # on mips: consistency string is just a v
899 # horrible bunch of .bytes,
900 # which I am too lazy to sort out (WDP 95/05)
902 print STDERR "Couldn't grok consistency: ", $chk[$i];
905 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
906 # we can just re-constitute this one...
907 # NB: we emit _three_ underscores no matter what,
908 # so ghc-split doesn't have to care.
909 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
911 } elsif ( $chkcat[$i] eq 'closure'
912 || $chkcat[$i] eq 'infotbl'
913 || $chkcat[$i] eq 'slow'
914 || $chkcat[$i] eq 'fast' ) { # do them in that order
915 $symb = $chksymb[$i];
918 if ( defined($closurechk{$symb}) ) {
919 print OUTASM $T_HDR_closure;
920 print OUTASM $chk[$closurechk{$symb}];
921 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
925 if ( defined($infochk{$symb}) ) {
927 print OUTASM $T_HDR_info;
928 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
929 if ( !defined($slowchk{$symb}) && defined($fastchk{$symb}) ) {
930 $fastname = $chk[$fastchk{$symb}];
931 $fastname =~ s/([_A-Za-z]*_fast\d+):.*(.*\n)*/\1/;
932 $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long $fastname/;
934 $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
935 $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
936 $chk[$infochk{$symb}] =~ s/\.long (IndUpdRet\S+)/\.long \.\1/;
937 $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long \.StdErrorCode/;
938 $chk[$infochk{$symb}] =~ s/\.long UpdErr/\.long \.UpdErr/;
939 print OUTASM $chk[$infochk{$symb}];
941 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
943 # entry code will be put here!
946 if ( $chk[$infochk{$symb}] =~ /$TDOTWORD[@]?\s+([A-Za-z0-9_]+_entry)$/o
947 && $1 ne "${T_US}${symb}_entry" ) {
948 print STDERR "!!! entry point???\n",$chk[$infochk{$symb}];
951 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
955 if ( defined($slowchk{$symb}) ) {
957 # teach it to drop through to the fast entry point:
958 $c = $chk[$slowchk{$symb}];
960 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
961 ($p, $r) = split(/TOC\[tc0\], 0\n/, $c);
962 if ($symb =~ /^[_A-Z]/)
964 print OUTASM "\t\.globl \.${chksymb[$i]}_entry\n";
965 print OUTASM "\.csect ${symb}_entry[DS]\n";
966 print OUTASM "${p}TOC[tc0], 0\n";
968 $r =~ s/\.csect \.text\[PR\]\n//; # todo: properly - andre
969 $c = &mangle_powerpc_tailjump($r);
972 if ( defined($fastchk{$symb}) ) {
973 if ( $TargetPlatform =~ /^alpha-/ ) {
974 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
975 } elsif ( $TargetPlatform =~ /^hppa/ ) {
976 $c =~ s/^\s+ldil.*\n\s+ldo.*\n\s+bv.*\n(.*\n)?\s+\.EXIT/$1\t.EXIT/;
977 } elsif ( $TargetPlatform =~ /^i386-/ ) {
978 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d+,\%edx\n\tjmp \*\%edx\n//;
979 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d+,\%ecx\n\tjmp \*\%ecx\n//;
980 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d+,\%eax\n\tjmp \*\%eax\n//;
981 } elsif ( $TargetPlatform =~ /^mips-/ ) {
982 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
983 } elsif ( $TargetPlatform =~ /^m68k-/ ) {
984 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n\tnop\n//;
985 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n//;
986 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ ) {
987 $c =~ s/^\tb \.${T_US}${symb}_fast\d+\n//;
988 } elsif ( $TargetPlatform =~ /^sparc-/ ) {
989 $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n\tnop\n//;
990 $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n(\t[a-z].*\n)/$1/;
992 print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n";
996 if ( $TargetPlatform !~ /^(alpha-|hppa|mips-)/ ) {
997 # On alphas, hppa: no very good way to look for "dangling"
998 # references to fast-entry point.
999 # (questionable re hppa and mips...)
1000 print STDERR "still has jump to fast entry point:\n$c"
1001 if $c =~ /$TUS[@]?$symb[@]?_fast/; # NB: paranoia
1004 print OUTASM $T_HDR_entry;
1006 &print_doctored($c, 1); # NB: the 1!!!
1008 $chkcat[$slowchk{$symb}] = 'DONE ALREADY';
1012 if ( defined($fastchk{$symb}) ) {
1013 $c = $chk[$fastchk{$symb}];
1014 if ( ! defined($slowchk{$symb})
1015 # ToDo: the || clause can go once we're no longer
1016 # concerned about producing exactly the same output as before
1017 #OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/
1019 print OUTASM $T_HDR_fast;
1021 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1022 local(@lbls) = split(/:/, $c);
1023 $fullname = $lbls[0];
1024 $fullname =~ s/$T_MOVE_DIRVS//g;
1025 if ( $fullname =~ /^[A-Z]/)
1026 { print OUTASM "\t\.globl \.${fullname}\n";
1028 # print OUTASM "\t\.lglobl \.${fullname}\n"; #todo: rm - andre
1030 $c =~ s/((.*\n)*)\t.long \S+, TOC\[tc0\], 0\n\.csect \.text\[PR\]\n((.*\n)*)/\1\3/;
1031 $c = &mangle_powerpc_tailjump($c);
1033 &print_doctored($c, 0);
1034 $chkcat[$fastchk{$symb}] = 'DONE ALREADY';
1037 } elsif ( $chkcat[$i] eq 'vector'
1038 || $chkcat[$i] eq 'direct' ) { # do them in that order
1039 $symb = $chksymb[$i];
1042 if ( defined($vectorchk{$symb}) ) {
1043 print OUTASM $T_HDR_vector;
1044 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1045 if ( $symb =~ /^[A-Z]/) {
1046 print OUTASM "\t\.globl \.vtbl_${symb}\n";
1047 print OUTASM "\t\.globl vtbl_${symb}\n";
1049 $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
1050 print OUTASM ".vtbl_${symb}:\n";
1051 print OUTASM $chk[$vectorchk{$symb}];
1053 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1055 # direct return code will be put here!
1056 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1060 if ( defined($directchk{$symb}) ) {
1061 print OUTASM $T_HDR_direct;
1062 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1063 ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$directchk{$symb}]);
1064 &print_doctored($r, 0);
1065 print OUTASM "\.csect ${symb}DirectReturn[DS]\n";
1066 print OUTASM "${p}TOC[tc0], 0\n";
1068 &print_doctored($chk[$directchk{$symb}], 0);
1070 $chkcat[$directchk{$symb}] = 'DONE ALREADY';
1072 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1073 # Alphas: the commented nop is for the splitter, to ensure
1074 # that no module ends with a label as the very last
1075 # thing. (The linker will adjust the label to point
1076 # to the first code word of the next module linked in,
1077 # even if alignment constraints cause the label to move!)
1079 print OUTASM "\t# nop\n";
1082 } elsif ( $chkcat[$i] eq 'toc' ) {
1083 # silly optimisation to print tocs, since they come in groups...
1084 print OUTASM $T_HDR_toc;
1086 while ($chkcat[$j] eq 'toc')
1087 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1090 print OUTASM $chk[$j];
1092 $chkcat[$j] = 'DONE ALREADY';
1097 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1101 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1103 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1104 print OUTASM ".csect .text[PR]\n_section_.text:\n.csect .data[RW]\n\t.long _section_.text\n"
1108 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1109 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1114 sub mash_hppa_prologue { # OK, epilogue, too
1117 # toss all prologue stuff
1118 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1120 # Lie about our .CALLINFO
1121 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1128 # toss all epilogue stuff
1129 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1131 # Sorry; we moved the _info stuff to the code segment.
1132 s/_info,DATA/_info,CODE/g;
1139 sub print_doctored {
1140 local($_, $need_fallthru_patch) = @_;
1142 if ( $TargetPlatform !~ /^i386-/
1143 || ! /^\t[a-z]/ ) { # no instructions in here, apparently
1147 # OK, must do some x86 **HACKING**
1149 local($entry_patch) = '';
1150 local($exit_patch) = '';
1151 local($call_entry_patch)= '';
1152 local($call_exit_patch) = '';
1153 local($gc_call_entry_patch)= ''; # Patches before and after calls to Perform_GC_wrapper
1154 local($gc_call_exit_patch) = '';
1156 #OLD: # first, convert calls to *very magic form*: (ToDo: document
1160 # call _?PerformGC_wrapper
1164 # call _?PerformGC_wrapper
1166 # The reason we do this now is to remove the apparent use of
1167 # %esp, which would throw off the "what patch code do we need"
1170 # Special macros in ghc/includes/COptWraps.lh, used in
1171 # ghc/runtime/CallWrap_C.lc, are required for this to work!
1174 s/^\tpushl \$(\d+)\n\tcall ${T_DO_GC}\n\taddl \$4,\%esp\n/\tmovl \$$1,\%eax\n\tcall ${T_DO_GC}\n/go;
1175 s/^\tpushl \%eax\n\tcall ${T_DO_GC}\n\taddl \$4,\%esp\n/\tcall ${T_DO_GC}\n/go;
1176 s/^\tpushl \%edx\n\tcall ${T_DO_GC}\n\taddl \$4,\%esp\n/\tmovl \%edx,\%eax\n\tcall ${T_DO_GC}\n/go;
1178 #= if ( $StolenX86Regs <= 4 ) { # %ecx is ordinary reg
1179 #= s/^\tpushl \%ecx\n\tcall ${T_DO_GC}\n\taddl \$4,\%esp\n/\tmovl \%ecx,\%eax\n\tcall ${T_DO_GC}\n/go;
1182 # gotta watch out for weird instructions that
1183 # invisibly smash various regs:
1184 # rep* %ecx used for counting
1185 # scas* %edi used for destination index
1186 # cmps* %e[sd]i used for indices
1187 # loop* %ecx used for counting
1192 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1194 # * GCC used an "STG reg" for its own purposes
1196 # * some secret uses of machine reg, requiring STG reg
1197 # to be saved/restored
1199 # The most dangerous "GCC uses" of an "STG reg" are when
1200 # the reg holds the target of a jmp -- it's tricky to
1201 # insert the patch-up code before we get to the target!
1202 # So here we change the jmps:
1204 # --------------------------------------------------------
1205 # it can happen that we have jumps of the form...
1206 # jmp *<something involving %esp>
1208 # jmp <something involving another naughty register...>
1210 # a reasonably-common case is:
1212 # movl $_blah,<bad-reg>
1215 # which is easily fixed as:
1217 # sigh! try to hack around it...
1220 if ($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 $2,\%eax\n\tjmp \*$1\(\%eax\)\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/;
1234 #= if ($StolenX86Regs <= 4 ) { # spurious uses of ecx?
1235 #= s/^\tmovl (.*),\%ecx\n\tjmp \*%ecx\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1236 #= s/^\tjmp \*(-?\d*)\((.*\%ecx.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1237 #= s/^\tjmp \*\%ecx\n/\tmovl \%ecx,\%eax\n\tjmp \*\%eax\n/g;
1238 #= die "$Pgm: (mangler) still have jump involving \%ecx!\n$_"
1239 #= if /(jmp|call) .*\%ecx/;
1242 # OK, now we can decide what our patch-up code is going to
1245 # Note funky ".=" stuff; we're *adding* to these _patch guys
1246 if ( $StolenX86Regs <= 2
1247 && ( /32\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1248 $entry_patch .= "\tmovl \%esi,32(\%ebx)\n";
1249 $exit_patch .= "\tmovl 32(\%ebx),\%esi\n";
1251 $gc_call_entry_patch .= "\tmovl \%esi,32(\%ebx)\n";
1252 $gc_call_exit_patch .= "\tmovl 32(\%ebx),\%esi\n";
1254 # nothing for call_{entry,exit} because %esi is callee-save
1256 if ( $StolenX86Regs <= 3
1257 && ( /64\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # SpA (edi)
1258 $entry_patch .= "\tmovl \%edi,64(\%ebx)\n";
1259 $exit_patch .= "\tmovl 64(\%ebx),\%edi\n";
1261 $gc_call_entry_patch .= "\tmovl \%edi,64(\%ebx)\n";
1262 $gc_call_exit_patch .= "\tmovl 64(\%ebx),\%edi\n";
1264 # nothing for call_{entry,exit} because %edi is callee-save
1266 #= if ( $StolenX86Regs <= 4
1267 #= && ( /80\(\%ebx\)/ || /\%ecx/ || /^\t(rep|loop)/ ) ) { # Hp (ecx)
1268 #= $entry_patch .= "\tmovl \%ecx,80(\%ebx)\n";
1269 #= $exit_patch .= "\tmovl 80(\%ebx),\%ecx\n";
1271 #= $call_exit_patch .= "\tmovl \%ecx,108(\%ebx)\n";
1272 #= $call_entry_patch .= "\tmovl 108(\%ebx),\%ecx\n";
1273 #= # I have a really bad feeling about this if we ever
1274 #= # have a nested call...
1275 #= # NB: should just hide it somewhere in the C stack.
1277 # --------------------------------------------------------
1278 # next, here we go with non-%esp patching!
1280 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1282 # Actually, call_entry_patch and call_exit_patch never get set,
1283 # so let's nuke this one
1284 # s/^(\tcall .*\n(\taddl \$\d+,\%esp\n)?)/$call_exit_patch$1$call_entry_patch/g; # _all_ calls
1286 # Before calling GC we must set up the exit condition before the call
1287 # and entry condition when we come back
1288 s/^(\tcall ${T_DO_GC}\n(\taddl \$\d+,\%esp\n)?)/$gc_call_exit_patch$1$gc_call_entry_patch/g; # _all_ calls
1290 # fix _all_ non-local jumps:
1292 s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1293 s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1295 s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
1297 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1298 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1300 # fix post-PerformGC wrapper (re-)entries ???
1302 if ($StolenX86Regs == 2 ) {
1303 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1304 if /^\t(jmp|call) .*\%e(si|di)/;
1305 #= die "ARGH! Jump uses \%esi, \%edi, or \%ecx with -monly-2-regs:\n$_"
1306 #= if /^\t(jmp|call) .*\%e(si|di|cx)/;
1307 } elsif ($StolenX86Regs == 3 ) {
1308 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1309 if /^\t(jmp|call) .*\%edi/;
1310 #= die "ARGH! Jump uses \%edi or \%ecx with -monly-3-regs:\n$_"
1311 #= if /^\t(jmp|call) .*\%e(di|cx)/;
1312 #= } elsif ($StolenX86Regs == 4 ) {
1313 #= die "ARGH! Jump uses \%ecx with -monly-4-regs:\n$_"
1314 #= if /^\t(jmp|call) .*\%ecx/;
1317 # final peephole fixes
1319 s/^\tmovl \%eax,36\(\%ebx\)\n\tjmp \*36\(\%ebx\)\n/\tmovl \%eax,36\(\%ebx\)\n\tjmp \*\%eax\n/;
1320 # the short form may tickle perl bug:
1321 # s/^\tmovl \$${T_US}(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp $T_US$1/g;
1322 s/^\tmovl \$${T_US}(.*),\%eax\n\tjmp \*\%eax/\tjmp $T_US$1/g;
1323 s/^\tmovl \$${T_US}(.*),\%ebx\n\tjmp \*\%ebx/\tjmp $T_US$1/g;
1324 s/^\tmovl \$${T_US}(.*),\%ecx\n\tjmp \*\%ecx/\tjmp $T_US$1/g;
1325 s/^\tmovl \$${T_US}(.*),\%edx\n\tjmp \*\%edx/\tjmp $T_US$1/g;
1327 # Hacks to eliminate some reloads of Hp. Worth about 5% code size.
1328 # We could do much better than this, but at least it catches about
1329 # half of the unnecessary reloads.
1330 # Note that these will stop working if either:
1331 # (i) the offset of Hp from BaseReg changes from 80, or
1332 # (ii) the register assignment of BaseReg changes from %ebx
1334 s/^\tmovl 80\(\%ebx\),\%e.x\n\tmovl \$(.*),(-?[0-9]*)\(\%e.x\)\n\tmovl 80\(\%ebx\),\%e(.)x/\tmovl 80\(\%ebx\),\%e$3x\n\tmovl \$$1,$2\(\%e$3x\)/g;
1336 s/^\tmovl 80\(\%ebx\),\%e(.)x\n\tmovl (.*),\%e(.)x\n\tmovl \%e$3x,(-?[0-9]*\(\%e$1x\))\n\tmovl 80\(\%ebx\),\%e$1x/\tmovl 80\(\%ebx\),\%e$1x\n\tmovl $2,\%e$3x\n\tmovl \%e$3x,$4/g;
1338 s/^\tmovl 80\(\%ebx\),\%edx((\n\t(movl|addl) .*,((-?[0-9]*\(.*)|(\%e[abc]x)))+)\n\tmovl 80\(\%ebx\),\%edx/\tmovl 80\(\%ebx\),\%edx$1/g;
1339 s/^\tmovl 80\(\%ebx\),\%eax((\n\t(movl|addl) .*,((-?[0-9]*\(.*)|(\%e[bcd]x)))+)\n\tmovl 80\(\%ebx\),\%eax/\tmovl 80\(\%ebx\),\%eax$1/g;
1341 # --------------------------------------------------------
1342 # that's it -- print it
1344 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1348 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1349 print OUTASM $exit_patch;
1350 # ToDo: make it not print if there is a "jmp" at the end
1356 sub init_FUNNY_THINGS {
1357 %KNOWN_FUNNY_THING = (
1358 "${T_US}CheckHeapCode${T_POST_LBL}", 1,
1359 "${T_US}CommonUnderflow${T_POST_LBL}", 1,
1360 "${T_US}Continue${T_POST_LBL}", 1,
1361 "${T_US}EnterNodeCode${T_POST_LBL}", 1,
1362 "${T_US}ErrorIO_call_count${T_POST_LBL}", 1,
1363 "${T_US}ErrorIO_innards${T_POST_LBL}", 1,
1364 "${T_US}IndUpdRetDir${T_POST_LBL}", 1,
1365 "${T_US}IndUpdRetV0${T_POST_LBL}", 1,
1366 "${T_US}IndUpdRetV1${T_POST_LBL}", 1,
1367 "${T_US}IndUpdRetV2${T_POST_LBL}", 1,
1368 "${T_US}IndUpdRetV3${T_POST_LBL}", 1,
1369 "${T_US}IndUpdRetV4${T_POST_LBL}", 1,
1370 "${T_US}IndUpdRetV5${T_POST_LBL}", 1,
1371 "${T_US}IndUpdRetV6${T_POST_LBL}", 1,
1372 "${T_US}IndUpdRetV7${T_POST_LBL}", 1,
1373 "${T_US}PrimUnderflow${T_POST_LBL}", 1,
1374 "${T_US}StackUnderflowEnterNode${T_POST_LBL}", 1,
1375 "${T_US}StdErrorCode${T_POST_LBL}", 1,
1376 "${T_US}UnderflowVect0${T_POST_LBL}", 1,
1377 "${T_US}UnderflowVect1${T_POST_LBL}", 1,
1378 "${T_US}UnderflowVect2${T_POST_LBL}", 1,
1379 "${T_US}UnderflowVect3${T_POST_LBL}", 1,
1380 "${T_US}UnderflowVect4${T_POST_LBL}", 1,
1381 "${T_US}UnderflowVect5${T_POST_LBL}", 1,
1382 "${T_US}UnderflowVect6${T_POST_LBL}", 1,
1383 "${T_US}UnderflowVect7${T_POST_LBL}", 1,
1384 "${T_US}UpdErr${T_POST_LBL}", 1,
1385 "${T_US}UpdatePAP${T_POST_LBL}", 1,
1386 "${T_US}_Enter_Internal${T_POST_LBL}", 1,
1387 "${T_US}_PRMarking_MarkNextAStack${T_POST_LBL}", 1,
1388 "${T_US}_PRMarking_MarkNextBStack${T_POST_LBL}", 1,
1389 "${T_US}_PRMarking_MarkNextCAF${T_POST_LBL}", 1,
1390 "${T_US}_PRMarking_MarkNextGA${T_POST_LBL}", 1,
1391 "${T_US}_PRMarking_MarkNextRoot${T_POST_LBL}", 1,
1392 "${T_US}_PRMarking_MarkNextSpark${T_POST_LBL}", 1,
1393 "${T_US}_PRMarking_MarkNextEvent${T_POST_LBL}", 1,
1394 "${T_US}_PRMarking_MarkNextClosureInFetchBuffer${T_POST_LBL}", 1,
1395 "${T_US}_Scavenge_Forward_Ref${T_POST_LBL}", 1,
1396 "${T_US}__std_entry_error__${T_POST_LBL}", 1,
1397 "${T_US}_startMarkWorld${T_POST_LBL}", 1,
1398 "${T_US}resumeThread${T_POST_LBL}", 1,
1399 "${T_US}startCcRegisteringWorld${T_POST_LBL}", 1,
1400 "${T_US}startEnterFloat${T_POST_LBL}", 1,
1401 "${T_US}startEnterInt${T_POST_LBL}", 1,
1402 "${T_US}startPerformIO${T_POST_LBL}", 1,
1403 "${T_US}startStgWorld${T_POST_LBL}", 1,
1404 "${T_US}stopPerformIO${T_POST_LBL}", 1
1409 The following table reversal is used for both info tables and return
1410 vectors. In both cases, we remove the first entry from the table,
1411 reverse the table, put the label at the end, and paste some code
1412 (that which is normally referred to by the first entry in the table)
1413 right after the table itself. (The code pasting is done elsewhere.)
1417 local($symb, $tbl, $discard1) = @_;
1419 local($before) = '';
1421 local(@imports) = (); # hppa only
1424 local(@lines) = split(/\n/, $tbl);
1425 local($i, $j); #local ($i, $extra, $words_to_pad, $j);
1427 # see comment in mangleAsm as to why this silliness is needed.
1428 local($TDOTWORD) = ${T_DOT_WORD};
1429 local($TDOTGLOBAL) = ${T_DOT_GLOBAL};
1430 local($TUS) = ${T_US};
1431 local($TPOSTLBL) = ${T_POST_LBL};
1433 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t$TDOTWORD\s+/o; $i++) {
1434 $label .= $lines[$i] . "\n",
1435 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info$TPOSTLBL[@]?$/o
1436 || $lines[$i] =~ /$TDOTGLOBAL/o
1437 || $lines[$i] =~ /^$TUS[@]?vtbl_\S+$TPOSTLBL[@]?$/o;
1439 $before .= $lines[$i] . "\n"; # otherwise...
1442 if ( $TargetPlatform !~ /^hppa/ ) {
1443 for ( ; $i <= $#lines && $lines[$i] =~ /^\t$TDOTWORD\s+/o; $i++) {
1444 push(@words, $lines[$i]);
1446 } else { # hppa weirdness
1447 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+\.(word|IMPORT)/; $i++) {
1448 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1449 push(@imports, $lines[$i]);
1451 # We don't use HP's ``function pointers''
1452 # We just use labels in code space, like normal people
1453 $lines[$i] =~ s/P%//;
1454 push(@words, $lines[$i]);
1459 # now throw away the first word (entry code):
1460 shift(@words) if $discard1;
1462 # Padding removed to reduce code size and improve performance on Pentiums.
1464 # for 486-cache-friendliness, we want our tables aligned
1465 # on 16-byte boundaries (.align 4). Let's pad:
1466 # $extra = ($#words + 1) % 4;
1467 # $words_to_pad = ($extra == 0) ? 0 : 4 - $extra;
1468 # for ($j = 0; $j < $words_to_pad; $j++) { push(@words, "\t${T_DOT_WORD} 0"); }
1470 for (; $i <= $#lines; $i++) {
1471 $after .= $lines[$i] . "\n";
1474 # Alphas:If we have anonymous text (not part of a procedure), the
1475 # linker may complain about missing exception information. Bleh.
1476 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1477 $before = "\t.ent $1\n" . $before;
1478 $after .= "\t.end $1\n";
1482 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1483 . join("\n", (reverse @words)) . "\n"
1486 # print STDERR "before=$before\n";
1487 # print STDERR "label=$label\n";
1488 # print STDERR "words=",(reverse @words),"\n";
1489 # print STDERR "after=$after\n";
1496 sub mini_mangle_asm_i386 {
1497 local($in_asmf, $out_asmf) = @_;
1499 &init_TARGET_STUFF();
1501 # see mangleAsm comment
1502 local($TUS) = ${T_US};
1503 local($TPOSTLBL)=${T_POST_LBL};
1505 open(INASM, "< $in_asmf")
1506 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1507 open(OUTASM,"> $out_asmf")
1508 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1514 /^$TUS[@]?(PerformGC|StackOverflow|Yield|PerformReschedule)_wrapper$TPOSTLBL\n/o;
1515 print OUTASM "\tmovl \%esp, ${T_US}__temp_esp\n";
1516 print OUTASM "\tmovl \%eax, ${T_US}__temp_eax\n";
1520 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1521 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1525 The HP is a major nuisance. The threaded code mangler moved info
1526 tables from data space to code space, but unthreaded code in the RTS
1527 still has references to info tables in data space. Since the HP
1528 linker is very precise about where symbols live, we need to patch the
1529 references in the unthreaded RTS as well.
1532 sub mini_mangle_asm_hppa {
1533 local($in_asmf, $out_asmf) = @_;
1535 open(INASM, "< $in_asmf")
1536 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1537 open(OUTASM,"> $out_asmf")
1538 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1541 s/_info,DATA/_info,CODE/; # Move _info references to code space
1547 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1548 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1554 sub mini_mangle_asm_powerpc {
1555 local($in_asmf, $out_asmf) = @_;
1557 open(INASM, "< $in_asmf")
1558 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1559 open(OUTASM,"> $out_asmf")
1560 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1563 s/long _PRStart/long ._PRStart/;
1564 s/long _PRIn_/long ._PRIn_/;
1565 s/long _Dummy_(\S+)_entry/long ._Dummy_\1_entry/;
1566 s/long _PRMarking_MarkNextRoot\[DS\]/long ._PRMarking_MarkNextRoot/;
1567 s/long _PRMarking_MarkNextCAF\[DS\]/long ._PRMarking_MarkNextCAF/;
1568 s/long _PRMarking_MarkNextAStack\[DS\]/long ._PRMarking_MarkNextAStack/;
1569 s/long _PRMarking_MarkNextBStack\[DS\]/long ._PRMarking_MarkNextBStack/;
1570 s/\.tc EnterNodeCode\[TC]\,EnterNodeCode\[DS\]/\.tc EnterNodeCode\[TC]\,.EnterNodeCode/; # CONC
1571 s/\.tc CheckHeapCode\[TC]\,CheckHeapCode\[DS\]/\.tc CheckHeapCode\[TC]\,.CheckHeapCode/; # CONC
1576 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1577 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1580 sub mangle_powerpc_tailjump {
1582 local($maybe_more) = 1;
1583 while (($c =~ /\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n/) && $maybe_more)
1586 $lcsymb =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1587 # the checks for r1 and r2 are mostly paranoia...
1589 $r1 =~ s/(.*\n)*\tlw?z? (\d+),LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1591 $r2 =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr (\d+)\n\tbctr\n(.*\n)*/\3/;
1594 $c =~ s/((.*\n)*)\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n((.*\n)*)/\1\tb $tocequiv{$lcsymb}\n\3/;
1600 # make "require"r happy...