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 ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd2|netbsd|nextstep3|cygwin32|mingw32)$/ ) {
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 ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|freebsd3|netbsd_elf)$/ ) {
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|netbsd_elf)$/) ? '#' : '/' ;
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|netbsd_elf/ ) {
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 || /^$TUS[@]?.*_done$TPOSTLBL[@]?$/o # PROF: _module_done
496 || /^$TUS[@]?_module_registered$TPOSTLBL[@]?$/o # PROF: _module_registered
499 $chkcat[$i] = 'data';
502 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
507 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
512 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_cc(s)?$TPOSTLBL[@]?$/o ) {
513 # all CC_ symbols go in the data section...
515 $chkcat[$i] = 'data';
518 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_(alt|dflt)$TPOSTLBL[@]?$/o ) {
520 $chkcat[$i] = 'misc';
523 #$chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
525 } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_vtbl$TPOSTLBL[@]?$/o ) {
527 $chkcat[$i] = 'vector';
532 # As a temporary solution for compiling "foreign export" declarations,
533 # we use global variables to pass arguments from C to STG land.
534 # These declarations live in the .hc file and not in the generated C
535 # stub file, so we let them pass through here.
536 } elsif ( /^[\t ]+\.comm[\t ]+__fexp_.*$/ ) {
538 $chkcat[$i] = 'data';
541 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
542 && /^(_uname|uname|stat|fstat):/ ) {
543 # for some utterly bizarre reason, this platform
544 # likes to drop little local C routines with these names
545 # into each and every .o file that #includes the
546 # relevant system .h file. Yuck. We just don't
547 # tolerate them in .hc files (which we are processing
548 # here). If you need to call one of these things from
549 # Haskell, make a call to your own C wrapper, then
550 # put that C wrapper (which calls one of these) in a
551 # plain .c file. WDP 95/12
553 $chkcat[$i] = 'toss';
556 } elsif ( /^$TUS[@]?[A-Za-z0-9_]/o
557 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
559 && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
560 || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
563 print "Funny global thing?: $_"
564 unless $KNOWN_FUNNY_THING{$thing}
565 || /^$TUS[@]?stg_.*$TPOSTLBL[@]?$/o # RTS internals
566 || /^$TUS[@]__fexp_.*$TPOSTLBL$/o # foreign export
567 || /^$TUS[@]?_reg.*$TPOSTLBL$/o # PROF: __reg<module>
568 || /^$TUS[@]?.*_btm$TPOSTLBL$/o # large bitmaps
569 || /^$TUS[@]?.*_closure_tbl$TPOSTLBL$/o; # closure tables
571 $chkcat[$i] = 'misc';
572 if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
573 { $chksymb[$i] = $thing; }
574 else { $chksymb[$i] = ''; };
576 } else { # simple line (duplicated at the top)
581 $numchks = $#chk + 1;
583 # the division into chunks is imperfect;
584 # we throw some things over the fence into the next
587 # also, there are things we would like to know
588 # about the whole module before we start spitting
591 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
593 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
595 # Alphas: NB: we start meddling at chunk 1, not chunk 0
596 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
597 # spits a ".quad 0" in after the v first ".rdata"; we
598 # detect this special case (tossing the ".quad 0")!
599 local($magic_rdata_seen) = 0;
601 # HPPAs, MIPSen: also start medding at chunk 1
603 if ($TargetPlatform =~ /^powerpc|^rs6000/) {
604 print OUTASM $T_HDR_toc; # yes, we have to put a .toc
605 # in the beginning of every file!
606 %tocequiv = (); # maps toc symbol number to toc symbol
607 %revtocequiv = (); # maps toc symbol to toc symbol number
608 for ($i = 1; $i < $numchks; $i++) {
609 $chk[$i] =~ s/\[RW\]//g;
610 $chk[$i] =~ s/\[DS\]//g;
611 $chk[$i] =~ s/^\.csect .*\[DS\]$//g;
613 if ( $chkcat[$i] eq 'toc' && $chk[$i] !~ /\.byte/ )
614 #ToDo: instead of all these changes, just fix mangle_powerpc_tailjump and delete/ignore these tocs?
615 { $chk[$i] =~ s/$T_MOVE_DIRVS//g;
616 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_fast\d+)/\t\.tc \1\[TC\],\.\2/;
617 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_entry)\n/\t\.tc \1\[TC\],\.\2\n/;
618 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(ret_\S+)/\t\.tc \1\[TC\],\.\2/;
619 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(alt_\S+)/\t\.tc \1\[TC\],\.\2/;
620 $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
622 $tocnumber = $chksymb[$i];
624 $tocsymb =~ s/^LC\.\.\d+:\n//;
625 $tocsymb =~ s/^\t\.tc \S+,(\S+)\n/\1/;
626 $tocequiv{$tocnumber} = $tocsymb;
628 } elsif ( $chkcat[$i] eq 'toc' && $chk[$i] =~ /\.byte/ ) {
629 $chkcat[$i] = 'literal';
634 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
635 $c = $chk[$i]; # convenience copy
637 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
639 # toss all prologue stuff; HPPA is pretty weird
641 $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/;
643 # be slightly paranoid to make sure there's
644 # nothing surprising in there
645 if ( $c =~ /--- BEGIN ---/ ) {
646 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
648 if ($TargetPlatform =~ /^i386-/) {
649 $p =~ s/^\tpushl \%edi\n//;
650 $p =~ s/^\tpushl \%esi\n//;
651 $p =~ s/^\tpushl \%ebx\n//;
652 $p =~ s/^\tsubl \$\d+,\%esp\n//;
653 $p =~ s/^\tmovl \$\d+,\%eax\n\tcall __alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
654 } elsif ($TargetPlatform =~ /^m68k-/) {
655 $p =~ s/^\tlink a6,#-?\d.*\n//;
656 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
657 # The above showed up in the asm code,
658 # so I added it here.
659 # I hope it's correct.
661 $p =~ s/^\tmovel d2,sp\@-\n//;
662 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
663 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
664 } elsif ($TargetPlatform =~ /^mips-/) {
665 # the .frame/.mask/.fmask that we use is the same
666 # as that produced by GCC for miniInterpret; this
667 # gives GDB some chance of figuring out what happened
668 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
669 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
670 $p =~ s/^\t\.(mask|fmask).*\n//g;
671 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
672 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
673 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
674 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
675 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
676 $p =~ s/__FRAME__/$FRAME/;
677 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000/) {
678 $p =~ s/^\tmflr 0\n//;
679 $p =~ s/^\tstm \d+,-\d+\(1\)\n//;
680 $p =~ s/^\tstw? 0,\d+\(1\)\n//g;
681 $p =~ s/^\tstw? 1,\d+\(1\)\n//g; #mc
682 $p =~ s/^\tlw?z 0,0\(1\)\n//g; #mc
683 $p =~ s/^\tstw?u 1,-\d+\(1\)\n//;
684 $p =~ s/^\tstw? \d+,-\d+\(1\)\n//g;
685 $p =~ s/^\tstfd \d+,-\d+\(1\)\n//g;
687 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
690 die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
691 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
693 # glue together what's left
695 $c =~ s/\n\t\n/\n/; # junk blank line
699 if ( $TargetPlatform =~ /^mips-/ ) {
700 # MIPS: first, this basic sequence may occur "--- END ---" or not
701 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
704 # toss all epilogue stuff; again, paranoidly
705 if ( $c =~ /--- END ---/ ) {
706 if (($r, $e) = split(/--- END ---/, $c)) {
707 if ($TargetPlatform =~ /^i386-/) {
709 $e =~ s/^\tpopl \%edi\n//;
710 $e =~ s/^\tpopl \%esi\n//;
711 $e =~ s/^\tpopl \%ecx\n//;
712 $e =~ s/^\taddl \$\d+,\%esp\n//;
713 } elsif ($TargetPlatform =~ /^m68k-/) {
714 $e =~ s/^\tunlk a6\n//;
716 } elsif ($TargetPlatform =~ /^mips-/) {
717 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
718 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
719 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
720 $e =~ s/^\tj\t\$31\n//;
721 } elsif ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
722 $e =~ s/^\taddi 1,1,\d+\n//;
723 $e =~ s/^\tcal 1,\d+\(1\)\n//;
724 $e =~ s/^\tlw?z? \d+,\d+\(1\)\n//;
725 $e =~ s/^\tmtlr 0\n//;
728 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
730 die "Epilogue junk?: $e\n" if $e =~ /^\t[^\.]/
731 && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
733 # glue together what's left
735 $c =~ s/\n\t\n/\n/; # junk blank line
739 # On SPARCs, we don't do --- BEGIN/END ---, we just
740 # toss the register-windowing save/restore/ret* instructions
742 if ( $TargetPlatform =~ /^sparc-/ ) {
743 $c =~ s/^\t(save.*|restore|ret|retl)\n//g;
744 # throw away PROLOGUE comments
745 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
748 # On Alphas, the prologue mangling is done a little later (below)
750 # toss all calls to __DISCARD__
751 $c =~ s/^\t(call|jbsr|jal)\s+$TUS[@]?__DISCARD__\n//go;
753 # MIPS: that may leave some gratuitous asm macros around
754 # (no harm done; but we get rid of them to be tidier)
755 $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/
756 if $TargetPlatform =~ /^mips-/;
758 # toss stack adjustment after DoSparks
759 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
760 if $TargetPlatform =~ /^m68k-/; # this looks old...
762 if ( $TargetPlatform =~ /^alpha-/ &&
763 ! $magic_rdata_seen &&
764 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
765 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
766 $magic_rdata_seen = 1;
769 # pick some end-things and move them to the next chunk
771 # pin a funny end-thing on (for easier matching):
772 $c .= 'FUNNY#END#THING';
774 while ( $c =~ /$TMOVEDIRVS[@]?FUNNY#END#THING/o ) { # [@]? is a silly hack to avoid having to use curlies for T_PRE_APP
775 # (this SEGVs perl4 on alphas, you see)
778 if ( $i < ($numchks - 1)
779 && ( $to_move =~ /$TCOPYDIRVS/
780 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
781 $chk[$i + 1] = $to_move . $chk[$i + 1];
782 # otherwise they're tossed
785 $c =~ s/$TMOVEDIRVS[@]?FUNNY#END#THING/FUNNY#END#THING/o; # [@]? is a hack (see above)
788 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
790 # toss all prologue stuff, except for loading gp, and the ..ng address
791 if (($p, $r) = split(/^\t\.prologue/, $c)) {
792 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
793 $c = $keep . "..ng:\n";
795 print STDERR "malformed code block ($ent)?\n"
798 $c .= "\t.frame \$30,0,\$26,0\n\t.prologue" . $r;
801 $c =~ s/FUNNY#END#THING//;
803 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
805 $chk[$i] = $c; # update w/ convenience copy
808 if ( $TargetPlatform =~ /^alpha-/ ) {
809 # print out the header stuff first
810 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
811 print OUTASM $chk[0];
813 } elsif ( $TargetPlatform =~ /^hppa/ ) {
814 print OUTASM $chk[0];
816 } elsif ( $TargetPlatform =~ /^mips-/ ) {
817 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
819 # get rid of horrible "<dollar>Revision: .*$" strings
820 local(@lines0) = split(/\n/, $chk[0]);
822 while ( $z <= $#lines0 ) {
823 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
826 while ( $z <= $#lines0 ) {
828 last if $lines0[$z] =~ /[,\t]0x0$/;
834 $chk[0] = join("\n", @lines0);
835 $chk[0] =~ s/\n\n+/\n/;
836 print OUTASM $chk[0];
839 # print out all the literal strings next
840 for ($i = 0; $i < $numchks; $i++) {
841 if ( $chkcat[$i] eq 'literal' ) {
842 print OUTASM $T_HDR_literal, $chk[$i];
843 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
845 $chkcat[$i] = 'DONE ALREADY';
849 # on the HPPA, print out all the bss next
850 if ( $TargetPlatform =~ /^hppa/ ) {
851 for ($i = 1; $i < $numchks; $i++) {
852 if ( $chkcat[$i] eq 'bss' ) {
853 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
854 print OUTASM $chk[$i];
856 $chkcat[$i] = 'DONE ALREADY';
861 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
862 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
864 next if $chkcat[$i] eq 'DONE ALREADY';
866 if ( $chkcat[$i] eq 'misc' ) {
867 if ($chk[$i] ne '') {
868 print OUTASM $T_HDR_misc;
869 if ($TargetPlatform =~ /^powerpc-|^rs6000/) {
870 $chksymb[$i] =~ s/://;
871 #new if ($chksymb[$i] =~ /ret.*upd/ || $KNOWN_FUNNY_THING{$chksymb[$i]}
872 #new || $chksymb[$i] =~ /^$.{T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o )
873 #new { print OUTASM "\t\.globl $chksymb[$i]\n"; }
874 # if ($chksymb[$i] ne '' && $chksymb[$i] !~ /ret_[a-z]/ && $chksymb[$i] !~ /djn_[a-z]/)
875 if ($chksymb[$i] ne '')
876 { print OUTASM "\t\.globl \.$chksymb[$i]\n"; };
877 if ($chk[$i] =~ /TOC\[tc0\], 0\n/)
878 { ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$i]); $printDS = 1;}
879 else { $r = $chk[$i]; $printDS = 0; };
880 $chk[$i] = &mangle_powerpc_tailjump($r);
882 &print_doctored($chk[$i], 0);
883 if ($TargetPlatform =~ /^powerpc-|^rs6000-/ && $printDS) {
884 #ok if ($chksymb[$i] !~ /\_regMain/) {
885 print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
886 print OUTASM "${p}TOC[tc0], 0\n";
891 } elsif ( $chkcat[$i] eq 'toss' ) {
892 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
894 } elsif ( $chkcat[$i] eq 'data' ) {
895 if ($chk[$i] ne '') {
896 print OUTASM $T_HDR_data;
897 print OUTASM $chk[$i];
900 } elsif ( $chkcat[$i] eq 'consist' ) {
901 if ( $chk[$i] =~ /$T_hsc_cc_PAT/o ) {
902 local($consist) = "$1.$2.$3";
904 $consist =~ s/\//./g;
906 $consist =~ s/[^A-Za-z0-9_.]/ZZ/g; # ToDo: properly?
908 # Using a cygnus-2.7-96q4 gcc build on hppas, the
909 # consistency chunk for ghc_cc_ID often (but not always!)
910 # gets lumped with a bunch of .IMPORT directives containing info on
911 # the code or data space nature of external symbols. We can't
912 # toss these, so once the consistency ID has been turned into
913 # a representable symbol, we substitute it for the symbol
914 # that the string was attached to in the first place (ghc_cc_ID.)
915 # (The original string is also substituted away.)
917 # This change may affect the code output on other platforms in
918 # adverse ways, hence we restrict this hack hppa targets only.
921 if ( $TargetPlatform =~ /^hppa/ ) {
922 $chk[$i] =~ s/^$TUS[@]?ghc.*c_ID$TPOSTLBL/$consist/o;
923 $chk[$i] =~ s/\t$T_hsc_cc_PAT/$T_HDR_misc/o;
924 $consist = $chk[$i]; #clumsily
926 print OUTASM $T_HDR_consist, "${consist}${T_POST_LBL}\n";
928 } elsif ( $TargetPlatform !~ /^(mips)-/ ) { # we just don't try in those case (ToDo)
929 # on mips: consistency string is just a v
930 # horrible bunch of .bytes,
931 # which I am too lazy to sort out (WDP 95/05)
933 print STDERR "Couldn't grok consistency: ", $chk[$i];
936 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
937 # we can just re-constitute this one...
938 # NB: we emit _three_ underscores no matter what,
939 # so ghc-split doesn't have to care.
940 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
942 } elsif ( $chkcat[$i] eq 'closure'
943 || $chkcat[$i] eq 'srt'
944 || $chkcat[$i] eq 'infotbl'
945 || $chkcat[$i] eq 'slow'
946 || $chkcat[$i] eq 'fast' ) { # do them in that order
947 $symb = $chksymb[$i];
950 if ( defined($closurechk{$symb}) ) {
951 print OUTASM $T_HDR_closure;
952 print OUTASM $chk[$closurechk{$symb}];
953 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
957 if ( defined($srtchk{$symb}) ) {
958 print OUTASM $T_HDR_srt;
959 print OUTASM $chk[$srtchk{$symb}];
960 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
964 if ( defined($infochk{$symb}) ) {
966 print OUTASM $T_HDR_info;
967 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
968 if ( !defined($slowchk{$symb}) && defined($fastchk{$symb}) ) {
969 $fastname = $chk[$fastchk{$symb}];
970 $fastname =~ s/([_A-Za-z]*_fast\d+):.*(.*\n)*/\1/;
971 $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long $fastname/;
973 $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
974 $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
975 print OUTASM $chk[$infochk{$symb}];
977 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
979 # entry code will be put here!
981 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
985 if ( defined($slowchk{$symb}) ) {
987 # teach it to drop through to the fast entry point:
988 $c = $chk[$slowchk{$symb}];
990 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
991 ($p, $r) = split(/TOC\[tc0\], 0\n/, $c);
992 if ($symb =~ /^[_A-Z]/)
994 print OUTASM "\t\.globl \.${chksymb[$i]}_entry\n";
995 print OUTASM "\.csect ${symb}_entry[DS]\n";
996 print OUTASM "${p}TOC[tc0], 0\n";
998 $r =~ s/\.csect \.text\[PR\]\n//; # todo: properly - andre
999 $c = &mangle_powerpc_tailjump($r);
1002 if ( defined($fastchk{$symb}) ) {
1003 if ( $TargetPlatform =~ /^alpha-/ ) {
1004 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1005 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1006 $c =~ s/^\s+ldil.*\n\s+ldo.*\n\s+bv.*\n(.*\n)?\s+\.EXIT/$1\t.EXIT/;
1007 } elsif ( $TargetPlatform =~ /^i386-/ ) {
1008 # Reg alloc depending, gcc generated code may jump to the fast entry point via
1009 # a number of registers.
1010 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edx\n\tjmp \*\%edx\n//;
1011 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%ecx\n\tjmp \*\%ecx\n//;
1012 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%eax\n\tjmp \*\%eax\n//;
1013 # The next two only apply if we're not stealing %esi or %edi.
1014 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%esi\n\tjmp \*\%esi\n// if ($StolenX86Regs < 3);
1015 $c =~ s/^\tmovl \$${T_US}${symb}_fast\d*,\%edi\n\tjmp \*\%edi\n// if ($StolenX86Regs < 4);
1016 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1017 $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/;
1018 } elsif ( $TargetPlatform =~ /^m68k-/ ) {
1019 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n\tnop\n//;
1020 $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n//;
1021 } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ ) {
1022 $c =~ s/^\tb \.${T_US}${symb}_fast\d+\n//;
1023 } elsif ( $TargetPlatform =~ /^sparc-/ ) {
1024 $c =~ s/^\tcall\s+${T_US}${symb}_fast\d+,.*\n\t\s*nop\n//;
1025 $c =~ s/^\tcall\s+${T_US}${symb}_fast\d+,.*\n(\t\s*[a-z].*\n)/$1/;
1027 print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n";
1031 if ( $TargetPlatform !~ /^(alpha-|hppa|mips-)/ ) {
1032 # On alphas, hppa: no very good way to look for "dangling"
1033 # references to fast-entry point.
1034 # (questionable re hppa and mips...)
1035 print STDERR "still has jump to fast entry point:\n$c"
1036 if $c =~ /$TUS[@]?$symb[@]?_fast/; # NB: paranoia
1039 print OUTASM $T_HDR_entry;
1041 &print_doctored($c, 1); # NB: the 1!!!
1043 $chkcat[$slowchk{$symb}] = 'DONE ALREADY';
1047 if ( defined($fastchk{$symb}) ) {
1048 $c = $chk[$fastchk{$symb}];
1049 if ( ! defined($slowchk{$symb})
1050 # ToDo: the || clause can go once we're no longer
1051 # concerned about producing exactly the same output as before
1052 #OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/
1054 print OUTASM $T_HDR_fast;
1056 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1057 local(@lbls) = split(/:/, $c);
1058 $fullname = $lbls[0];
1059 $fullname =~ s/$T_MOVE_DIRVS//g;
1060 if ( $fullname =~ /^[A-Z]/)
1061 { print OUTASM "\t\.globl \.${fullname}\n";
1063 # print OUTASM "\t\.lglobl \.${fullname}\n"; #todo: rm - andre
1065 $c =~ s/((.*\n)*)\t.long \S+, TOC\[tc0\], 0\n\.csect \.text\[PR\]\n((.*\n)*)/\1\3/;
1066 $c = &mangle_powerpc_tailjump($c);
1068 &print_doctored($c, 0);
1069 $chkcat[$fastchk{$symb}] = 'DONE ALREADY';
1072 } elsif ( $chkcat[$i] eq 'vector'
1073 || $chkcat[$i] eq 'direct' ) { # do them in that order
1074 $symb = $chksymb[$i];
1077 if ( defined($vectorchk{$symb}) ) {
1078 print OUTASM $T_HDR_vector;
1079 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1080 if ( $symb =~ /^[A-Z]/) {
1081 print OUTASM "\t\.globl \.${symb}_vtbl\n";
1082 print OUTASM "\t\.globl ${symb}_vtbl\n";
1084 $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
1085 print OUTASM ".${symb}_vtbl:\n";
1086 print OUTASM $chk[$vectorchk{$symb}];
1088 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1090 # direct return code will be put here!
1091 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1093 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1094 # Alphas: the commented nop is for the splitter, to ensure
1095 # that no module ends with a label as the very last
1096 # thing. (The linker will adjust the label to point
1097 # to the first code word of the next module linked in,
1098 # even if alignment constraints cause the label to move!)
1100 print OUTASM "\t# nop\n";
1103 } elsif ( $chkcat[$i] eq 'toc' ) {
1104 # silly optimisation to print tocs, since they come in groups...
1105 print OUTASM $T_HDR_toc;
1107 while ($chkcat[$j] eq 'toc')
1108 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1111 print OUTASM $chk[$j];
1113 $chkcat[$j] = 'DONE ALREADY';
1118 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1122 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1124 if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
1125 print OUTASM ".csect .text[PR]\n_section_.text:\n.csect .data[RW]\n\t.long _section_.text\n"
1129 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1130 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1135 sub mash_hppa_prologue { # OK, epilogue, too
1138 # toss all prologue stuff
1139 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1141 # Lie about our .CALLINFO
1142 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1149 # toss all epilogue stuff
1150 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1152 # Sorry; we moved the _info stuff to the code segment.
1153 s/_info,DATA/_info,CODE/g;
1160 sub print_doctored {
1161 local($_, $need_fallthru_patch) = @_;
1163 if ( $TargetPlatform !~ /^i386-/
1164 || ! /^\t[a-z]/ # no instructions in here, apparently
1165 || /^${T_US}_reg[A-Za-z0-9_]+${T_POST_LBL}/) {
1169 # OK, must do some x86 **HACKING**
1171 local($entry_patch) = '';
1172 local($exit_patch) = '';
1174 # gotta watch out for weird instructions that
1175 # invisibly smash various regs:
1176 # rep* %ecx used for counting
1177 # scas* %edi used for destination index
1178 # cmps* %e[sd]i used for indices
1179 # loop* %ecx used for counting
1184 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1186 # * GCC used an "STG reg" for its own purposes
1188 # * some secret uses of machine reg, requiring STG reg
1189 # to be saved/restored
1191 # The most dangerous "GCC uses" of an "STG reg" are when
1192 # the reg holds the target of a jmp -- it's tricky to
1193 # insert the patch-up code before we get to the target!
1194 # So here we change the jmps:
1196 # --------------------------------------------------------
1197 # it can happen that we have jumps of the form...
1198 # jmp *<something involving %esp>
1200 # jmp <something involving another naughty register...>
1202 # a reasonably-common case is:
1204 # movl $_blah,<bad-reg>
1208 # the short form may tickle perl bug:
1209 # s/^\tmovl \$${T_US}(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp $T_US$1/g;
1210 s/^\tmovl \$${T_US}(.*),\%eax\n\tjmp \*\%eax/\tjmp $T_US$1/g;
1211 s/^\tmovl \$${T_US}(.*),\%ebx\n\tjmp \*\%ebx/\tjmp $T_US$1/g;
1212 s/^\tmovl \$${T_US}(.*),\%ecx\n\tjmp \*\%ecx/\tjmp $T_US$1/g;
1213 s/^\tmovl \$${T_US}(.*),\%edx\n\tjmp \*\%edx/\tjmp $T_US$1/g;
1215 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1216 s/^\tmovl (.*),\%esi\n\tjmp \*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1217 s/^\tjmp \*(-?\d*)\((.*\%esi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
1218 s/^\tjmp \*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1219 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1220 if /(jmp|call) .*\%esi/;
1222 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1223 s/^\tmovl (.*),\%edi\n\tjmp \*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1224 s/^\tjmp \*(-?\d*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1225 s/^\tjmp \*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1226 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1227 if /(jmp|call) .*\%edi/;
1230 # OK, now we can decide what our patch-up code is going to
1233 # Offsets into register table - you'd better update these magic
1234 # numbers should you change its contents!
1235 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1236 local($OFFSET_Hp)=92;
1238 # Note funky ".=" stuff; we're *adding* to these _patch guys
1239 if ( $StolenX86Regs <= 2
1240 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1241 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1242 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1244 # nothing for call_{entry,exit} because %esi is callee-save
1246 if ( $StolenX86Regs <= 3
1247 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1248 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1249 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1251 # nothing for call_{entry,exit} because %edi is callee-save
1254 # --------------------------------------------------------
1255 # next, here we go with non-%esp patching!
1257 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1259 # Before calling GC we must set up the exit condition before the call
1260 # and entry condition when we come back
1262 # fix _all_ non-local jumps:
1264 s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1265 s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1267 s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
1269 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1270 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1272 if ($StolenX86Regs == 2 ) {
1273 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1274 if /^\t(jmp|call) .*\%e(si|di)/;
1275 } elsif ($StolenX86Regs == 3 ) {
1276 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1277 if /^\t(jmp|call) .*\%edi/;
1280 # --------------------------------------------------------
1281 # that's it -- print it
1283 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1287 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1288 print OUTASM $exit_patch;
1289 # ToDo: make it not print if there is a "jmp" at the end
1295 sub init_FUNNY_THINGS {
1296 %KNOWN_FUNNY_THING = (
1298 # "${T_US}stg_.*{T_POST_LBL}", 1,
1303 The following table reversal is used for both info tables and return
1304 vectors. In both cases, we remove the first entry from the table,
1305 reverse the table, put the label at the end, and paste some code
1306 (that which is normally referred to by the first entry in the table)
1307 right after the table itself. (The code pasting is done elsewhere.)
1311 local($symb, $tbl, $discard1) = @_;
1313 local($before) = '';
1315 local(@imports) = (); # hppa only
1318 local(@lines) = split(/\n/, $tbl);
1319 local($i, $j); #local ($i, $extra, $words_to_pad, $j);
1321 # see comment in mangleAsm as to why this silliness is needed.
1322 local($TDOTWORD) = ${T_DOT_WORD};
1323 local($TDOTGLOBAL) = ${T_DOT_GLOBAL};
1324 local($TUS) = ${T_US};
1325 local($TPOSTLBL) = ${T_POST_LBL};
1327 # Deal with the header...
1328 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?$TDOTWORD\s+/o; $i++) {
1329 $label .= $lines[$i] . "\n",
1330 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info$TPOSTLBL[@]?$/o
1331 || $lines[$i] =~ /$TDOTGLOBAL/o
1332 || $lines[$i] =~ /^$TUS[@]?\S+_vtbl$TPOSTLBL[@]?$/o;
1334 $before .= $lines[$i] . "\n"; # otherwise...
1337 # Grab the table data...
1338 if ( $TargetPlatform !~ /^hppa/ ) {
1339 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?$TDOTWORD\s+/o; $i++) {
1340 push(@words, $lines[$i]);
1342 } else { # hppa weirdness
1343 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+($TDOTWORD|\.IMPORT)/; $i++) {
1344 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1345 push(@imports, $lines[$i]);
1347 # We don't use HP's ``function pointers''
1348 # We just use labels in code space, like normal people
1349 $lines[$i] =~ s/P%//;
1350 push(@words, $lines[$i]);
1355 # now throw away the first word (SRT) iff it is empty.
1356 # The .zero business is for Linux/ELF.
1357 # The .skip business is for Sparc/Solaris/ELF.
1358 # The .blockz business is for HPPA.
1359 if ($discard1 && $words[0] =~ /^\t?($TDOTWORD\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1363 # Padding removed to reduce code size and improve performance on Pentiums.
1365 # for 486-cache-friendliness, we want our tables aligned
1366 # on 16-byte boundaries (.align 4). Let's pad:
1367 # $extra = ($#words + 1) % 4;
1368 # $words_to_pad = ($extra == 0) ? 0 : 4 - $extra;
1369 # for ($j = 0; $j < $words_to_pad; $j++) { push(@words, "\t${T_DOT_WORD} 0"); }
1371 for (; $i <= $#lines; $i++) {
1372 $after .= $lines[$i] . "\n";
1375 # Alphas:If we have anonymous text (not part of a procedure), the
1376 # linker may complain about missing exception information. Bleh.
1377 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1378 $before = "\t.ent $1\n" . $before;
1379 $after .= "\t.end $1\n";
1383 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1384 . join("\n", @words) . "\n"
1387 # print STDERR "before=$before\n";
1388 # print STDERR "label=$label\n";
1389 # print STDERR "words=",(reverse @words),"\n";
1390 # print STDERR "after=$after\n";
1396 The HP is a major nuisance. The threaded code mangler moved info
1397 tables from data space to code space, but unthreaded code in the RTS
1398 still has references to info tables in data space. Since the HP
1399 linker is very precise about where symbols live, we need to patch the
1400 references in the unthreaded RTS as well.
1403 sub mini_mangle_asm_hppa {
1404 local($in_asmf, $out_asmf) = @_;
1406 open(INASM, "< $in_asmf")
1407 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1408 open(OUTASM,"> $out_asmf")
1409 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1412 s/_info,DATA/_info,CODE/; # Move _info references to code space
1418 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1419 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1425 sub mini_mangle_asm_powerpc {
1426 local($in_asmf, $out_asmf) = @_;
1428 open(INASM, "< $in_asmf")
1429 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1430 open(OUTASM,"> $out_asmf")
1431 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1434 s/long _PRStart/long ._PRStart/;
1435 s/long _PRIn_/long ._PRIn_/;
1436 s/long _Dummy_(\S+)_entry/long ._Dummy_\1_entry/;
1437 s/long _PRMarking_MarkNextRoot\[DS\]/long ._PRMarking_MarkNextRoot/;
1438 s/long _PRMarking_MarkNextCAF\[DS\]/long ._PRMarking_MarkNextCAF/;
1439 s/long _PRMarking_MarkNextAStack\[DS\]/long ._PRMarking_MarkNextAStack/;
1440 s/long _PRMarking_MarkNextBStack\[DS\]/long ._PRMarking_MarkNextBStack/;
1441 s/\.tc EnterNodeCode\[TC]\,EnterNodeCode\[DS\]/\.tc EnterNodeCode\[TC]\,.EnterNodeCode/; # CONC
1442 s/\.tc CheckHeapCode\[TC]\,CheckHeapCode\[DS\]/\.tc CheckHeapCode\[TC]\,.CheckHeapCode/; # CONC
1447 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1448 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1451 sub mangle_powerpc_tailjump {
1453 local($maybe_more) = 1;
1454 while (($c =~ /\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n/) && $maybe_more)
1457 $lcsymb =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1458 # the checks for r1 and r2 are mostly paranoia...
1460 $r1 =~ s/(.*\n)*\tlw?z? (\d+),LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
1462 $r2 =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr (\d+)\n\tbctr\n(.*\n)*/\3/;
1465 $c =~ s/((.*\n)*)\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n((.*\n)*)/\1\tb $tocequiv{$lcsymb}\n\3/;
1471 # make "require"r happy...