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.
16 General note [chak]: Many regexps are very fragile because they rely on white
17 space being in the right place. This caused trouble with gcc 2.95 (at least
18 on Linux), where the use of white space in .s files generated by gcc suddenly
19 changed. To guarantee compatibility across different versions of gcc, make
20 sure (at least on i386-.*-linux) that regexps tolerate varying amounts of white
21 space between an assembler statement and its arguments as well as after a the
22 comma separating multiple arguments.
24 \emph{For the time being, I have corrected the regexps for i386-.*-linux. I
25 didn't touch all the regexps for other i386 platforms, as I don't have
26 a box to test these changes.}
31 The HP linker is very picky about symbols being in the appropriate
32 space (code vs. data). When we mangle the threaded code to put the
33 info tables just prior to the code, they wind up in code space
34 rather than data space. This means that references to *_info from
35 un-mangled parts of the RTS (e.g. unthreaded GC code) get
36 unresolved symbols. Solution: mini-mangler for .c files on HP. I
37 think this should really be triggered in the driver by a new -rts
38 option, so that user code doesn't get mangled inappropriately.
40 With reversed tables, jumps are to the _info label rather than to
41 the _entry label. The _info label is just an address in code
42 space, rather than an entry point with the descriptive blob we
43 talked about yesterday. As a result, you can't use the call-style
44 JMP_ macro. However, some JMP_ macros take _info labels as targets
45 and some take code entry points within the RTS. The latter won't
46 work with the goto-style JMP_ macro. Sigh. Solution: Use the goto
47 style JMP_ macro, and mangle some more assembly, changing all
48 "RP'literal" and "LP'literal" references to "R'literal" and
49 "L'literal," so that you get the real address of the code, rather
50 than the descriptive blob. Also change all ".word P%literal"
51 entries in info tables and vector tables to just ".word literal,"
52 for the same reason. Advantage: No more ridiculous call sequences.
55 %************************************************************************
57 \subsection{Top-level code}
59 %************************************************************************
62 $TargetPlatform = $TARGETPLATFORM;
64 ($Pgm = $0) =~ s|.*/||;
68 if ( $TargetPlatform =~ /^i386-/ ) {
72 $StolenX86Regs = $ARGV[2];
76 &mangle_asm($ifile,$ofile);
81 %************************************************************************
83 \subsection{Constants for various architectures}
85 %************************************************************************
88 sub init_TARGET_STUFF {
90 #--------------------------------------------------------#
91 if ( $TargetPlatform =~ /^alpha-.*-.*/ ) {
93 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
94 $T_US = ''; # _ if symbols have an underscore on the front
95 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
96 $T_CONST_LBL = '^\$L?C(\d+):$'; # regexp for what such a lbl looks like
99 $T_MOVE_DIRVS = '^(\s*(\$.*\.\.ng:|\.align\s+\d+|\.(globl|ent)\s+\S+|\#.*|\.(file|loc)\s+\S+\s+\S+|\.text|\.r?data)\n)';
100 $T_COPY_DIRVS = '^\s*(\$.*\.\.ng:|\#|\.(file|globl|ent|loc))';
102 $T_DOT_WORD = '\.(long|quad|byte|word)';
103 $T_DOT_GLOBAL = '^\t\.globl';
104 $T_HDR_literal = "\.rdata\n\t\.align 3\n";
105 $T_HDR_misc = "\.text\n\t\.align 3\n";
106 $T_HDR_data = "\.data\n\t\.align 3\n";
107 $T_HDR_rodata = "\.rdata\n\t\.align 3\n";
108 $T_HDR_closure = "\.data\n\t\.align 3\n";
109 $T_HDR_info = "\.text\n\t\.align 3\n";
110 $T_HDR_entry = "\.text\n\t\.align 3\n";
111 $T_HDR_vector = "\.text\n\t\.align 3\n";
113 #--------------------------------------------------------#
114 } elsif ( $TargetPlatform =~ /^hppa/ ) {
116 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
117 $T_US = ''; # _ if symbols have an underscore on the front
118 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
119 $T_CONST_LBL = '^L\$C(\d+)$'; # regexp for what such a lbl looks like
122 $T_MOVE_DIRVS = '^((\s+\.(IMPORT|EXPORT|PARAM).*|\s+\.align\s+\d+|\s+\.(SPACE|SUBSPA)\s+\S+|\s*)\n)';
123 $T_COPY_DIRVS = '^\s+\.(IMPORT|EXPORT)';
125 $T_DOT_WORD = '\.(blockz|word|half|byte)';
126 $T_DOT_GLOBAL = '^\s+\.EXPORT';
127 $T_HDR_literal = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
128 $T_HDR_misc = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
129 $T_HDR_data = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
130 $T_HDR_rodata = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
131 $T_HDR_closure = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
132 $T_HDR_info = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
133 $T_HDR_entry = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
134 $T_HDR_vector = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
136 #--------------------------------------------------------#
137 } elsif ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd2|nextstep3|cygwin32|mingw32)$/ ) {
138 # NeXT added but not tested. CaS
140 $T_STABBY = 1; # 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 $T_CONST_LBL = '^LC(\d+):$';
145 $T_X86_PRE_LLBL_PAT = 'L';
146 $T_X86_PRE_LLBL = 'L';
147 $T_X86_BADJMP = '^\tjmp [^L\*]';
149 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s.*|\.globl\s+\S+|\.text|\.data|\.stab[^n].*|\.type\s+.*|\.size\s+.*|\.lcomm.*)\n)';
150 $T_COPY_DIRVS = '\.(globl|stab|lcomm)';
151 $T_DOT_WORD = '\.(long|word|value|byte|space)';
152 $T_DOT_GLOBAL = '\.globl';
153 $T_HDR_literal = "\.text\n\t\.align 2\n";
154 $T_HDR_misc = "\.text\n\t\.align 2,0x90\n";
155 $T_HDR_data = "\.data\n\t\.align 2\n";
156 $T_HDR_rodata = "\.text\n\t\.align 2\n";
157 $T_HDR_closure = "\.data\n\t\.align 2\n";
158 $T_HDR_info = "\.text\n\t\.align 2\n"; # NB: requires padding
159 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
160 $T_HDR_vector = "\.text\n\t\.align 2\n"; # NB: requires padding
162 #--------------------------------------------------------#
163 } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|freebsd|netbsd|openbsd)$/ ) {
165 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
166 $T_US = ''; # _ if symbols have an underscore on the front
167 $T_PRE_APP = # regexp that says what comes before APP/NO_APP
168 ($TargetPlatform =~ /-(linux|freebsd|netbsd|openbsd)$/) ? '#' : '/' ;
169 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
171 $T_X86_PRE_LLBL_PAT = '\.L';
172 $T_X86_PRE_LLBL = '.L';
173 $T_X86_BADJMP = '^\tjmp\s+[^\.\*]';
175 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s.*|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.type\s+.*|\.size\s+\S+\s*,\s*\d+|\.ident.*|\.local.*)\n)';
176 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
178 $T_DOT_WORD = '\.(long|value|word|byte|zero)';
179 $T_DOT_GLOBAL = '\.globl';
180 $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11)
181 $T_HDR_misc = "\.text\n\t\.align 4\n";
182 $T_HDR_data = "\.data\n\t\.align 4\n";
183 $T_HDR_rodata = "\.section\t\.rodata\n\t\.align 4\n";
184 $T_HDR_closure = "\.data\n\t\.align 4\n";
185 $T_HDR_info = "\.text\n\t\.align 4\n";
186 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
187 $T_HDR_vector = "\.text\n\t\.align 4\n"; # NB: requires padding
189 #--------------------------------------------------------#
190 } elsif ( $TargetPlatform =~ /^ia64-.*-linux$/ ) {
192 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
193 $T_US = ''; # _ if symbols have an underscore on the front
195 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
198 $T_MOVE_DIRVS = '^(\s*\.(global|proc|pred\.safe_across_calls|text|data|section|subsection|align|size|type|ident)\s+.*\n)';
199 $T_COPY_DIRVS = '\.(global|proc)';
201 $T_DOT_WORD = '\.(long|value|byte|zero)';
202 $T_DOT_GLOBAL = '\.global';
203 $T_HDR_literal = "\.section\t\.rodata\n";
204 $T_HDR_misc = "\.text\n\t\.align 8\n";
205 $T_HDR_data = "\.data\n\t\.align 8\n";
206 $T_HDR_rodata = "\.section\t\.rodata\n\t\.align 8\n";
207 $T_HDR_closure = "\.data\n\t\.align 8\n";
208 $T_HDR_info = "\.text\n\t\.align 8\n";
209 $T_HDR_entry = "\.text\n\t\.align 16\n";
210 $T_HDR_vector = "\.text\n\t\.align 8\n";
212 #--------------------------------------------------------#
213 } elsif ( $TargetPlatform =~ /^x86_64-.*-(linux|openbsd)$/ ) {
215 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
216 $T_US = ''; # _ if symbols have an underscore on the front
218 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
221 $T_MOVE_DIRVS = '^(\s*\.(globl|text|data|section|align|size|type|ident|local)\s+.*\n)';
222 $T_COPY_DIRVS = '\.(globl|local)';
224 $T_DOT_WORD = '\.(quad|long|value|byte|zero)';
225 $T_DOT_GLOBAL = '\.global';
226 $T_HDR_literal = "\.section\t\.rodata\n";
227 $T_HDR_misc = "\.text\n\t\.align 8\n";
228 $T_HDR_data = "\.data\n\t\.align 8\n";
229 $T_HDR_rodata = "\.section\t\.rodata\n\t\.align 8\n";
230 $T_HDR_closure = "\.data\n\t\.align 8\n";
231 $T_HDR_info = "\.text\n\t\.align 8\n";
232 $T_HDR_entry = "\.text\n\t\.align 8\n";
233 $T_HDR_vector = "\.text\n\t\.align 8\n";
235 #--------------------------------------------------------#
236 } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
238 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
239 $T_US = '_'; # _ if symbols have an underscore on the front
240 $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
241 $T_CONST_LBL = '^LC(\d+):$';
244 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
245 $T_COPY_DIRVS = '\.(globl|proc|stab)';
247 $T_DOT_WORD = '\.long';
248 $T_DOT_GLOBAL = '\.globl';
249 $T_HDR_literal = "\.text\n\t\.even\n";
250 $T_HDR_misc = "\.text\n\t\.even\n";
251 $T_HDR_data = "\.data\n\t\.even\n";
252 $T_HDR_rodata = "\.text\n\t\.even\n";
253 $T_HDR_closure = "\.data\n\t\.even\n";
254 $T_HDR_info = "\.text\n\t\.even\n";
255 $T_HDR_entry = "\.text\n\t\.even\n";
256 $T_HDR_vector = "\.text\n\t\.even\n";
258 #--------------------------------------------------------#
259 } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
261 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
262 $T_US = ''; # _ if symbols have an underscore on the front
263 $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
264 $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
267 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
268 $T_COPY_DIRVS = '\.(globl|ent)';
270 $T_DOT_WORD = '\.word';
271 $T_DOT_GLOBAL = '^\t\.globl';
272 $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
273 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
274 $T_HDR_data = "\t\.data\n\t\.align 2\n";
275 $T_HDR_rodata = "\t\.rdata\n\t\.align 2\n";
276 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
277 $T_HDR_info = "\t\.text\n\t\.align 2\n";
278 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
279 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
281 #--------------------------------------------------------#
282 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ ) {
283 # Apple PowerPC Darwin/MacOS X.
284 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
285 $T_US = '_'; # _ if symbols have an underscore on the front
286 $T_PRE_APP = 'DOESNT APPLY'; # regexp that says what comes before APP/NO_APP
287 $T_CONST_LBL = '^\LC\d+:'; # regexp for what such a lbl looks like
290 $T_MOVE_DIRVS = '^(\s*(\.align \d+|\.text|\.data|\.const_data|\.cstring|\.non_lazy_symbol_pointer|\.const|\.static_const|\.literal4|\.literal8|\.static_data|\.globl \S+|\.section .*|\.lcomm.*)\n)';
291 $T_COPY_DIRVS = '\.(globl|lcomm)';
293 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
294 $T_DOT_GLOBAL = '\.globl';
295 $T_HDR_toc = "\.toc\n";
296 $T_HDR_literal = "\t\.const\n\t\.align 2\n";
297 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
298 $T_HDR_data = "\t\.data\n\t\.align 2\n";
299 $T_HDR_rodata = "\t\.const\n\t\.align 2\n";
300 $T_HDR_relrodata= "\t\.const_data\n\t\.align 2\n";
301 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
302 $T_HDR_info = "\t\.text\n\t\.align 2\n";
303 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
304 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
306 #--------------------------------------------------------#
307 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ ) {
309 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
310 $T_US = ''; # _ if symbols have an underscore on the front
311 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
312 $T_CONST_LBL = '^\.LC\d+:'; # regexp for what such a lbl looks like
315 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,\s*0x90)?|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.type\s+.*|\.size\s+\S+\s*,\s*\d+|\.ident.*|\.local.*)\n)';
316 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
318 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
319 $T_DOT_GLOBAL = '\.globl';
320 $T_HDR_toc = "\.toc\n";
321 $T_HDR_literal = "\t\.section\t.rodata\n\t\.align 2\n";
322 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
323 $T_HDR_data = "\t\.data\n\t\.align 2\n";
324 $T_HDR_rodata = "\t\.section\t.rodata\n\t\.align 2\n";
325 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
326 $T_HDR_info = "\t\.text\n\t\.align 2\n";
327 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
328 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
330 #--------------------------------------------------------#
331 } elsif ( $TargetPlatform =~ /^powerpc64-.*-linux/ ) {
333 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
334 $T_US = '\.'; # _ if symbols have an underscore on the front
335 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
336 $T_CONST_LBL = '^\.LC\d+:'; # regexp for what such a lbl looks like
339 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,\s*0x90)?|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.type\s+.*|\.size\s+\S+\s*,\s*\d+|\.ident.*|\.local.*)\n)';
340 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
342 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
343 $T_DOT_GLOBAL = '\.globl';
344 $T_HDR_toc = "\.toc\n";
345 $T_HDR_literal = "\t\.section\t\".toc\",\"aw\"\n";
346 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
347 $T_HDR_data = "\t\.data\n\t\.align 2\n";
348 $T_HDR_rodata = "\t\.section\t.rodata\n\t\.align 2\n";
349 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
350 $T_HDR_info = "\t\.text\n\t\.align 2\n";
351 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
352 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
354 #--------------------------------------------------------#
355 } elsif ( $TargetPlatform =~ /^sparc-.*-(solaris2|openbsd)/ ) {
357 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
358 $T_US = ''; # _ if symbols have an underscore on the front
359 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
360 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
363 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\s*\.section.*|\s+\.type.*|\s+\.size.*)\n)';
364 $T_COPY_DIRVS = '\.(global|proc|stab)';
366 $T_DOT_WORD = '\.(long|word|byte|half|skip|uahalf|uaword)';
367 $T_DOT_GLOBAL = '^\t\.global';
368 $T_HDR_literal = "\.text\n\t\.align 8\n";
369 $T_HDR_misc = "\.text\n\t\.align 4\n";
370 $T_HDR_data = "\.data\n\t\.align 8\n";
371 $T_HDR_rodata = "\.text\n\t\.align 4\n";
372 $T_HDR_closure = "\.data\n\t\.align 4\n";
373 $T_HDR_info = "\.text\n\t\.align 4\n";
374 $T_HDR_entry = "\.text\n\t\.align 4\n";
375 $T_HDR_vector = "\.text\n\t\.align 4\n";
377 #--------------------------------------------------------#
378 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
380 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
381 $T_US = '_'; # _ if symbols have an underscore on the front
382 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
383 $T_CONST_LBL = '^LC(\d+):$';
386 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
387 $T_COPY_DIRVS = '\.(global|proc|stab)';
389 $T_DOT_WORD = '\.word';
390 $T_DOT_GLOBAL = '^\t\.global';
391 $T_HDR_literal = "\.text\n\t\.align 8\n";
392 $T_HDR_misc = "\.text\n\t\.align 4\n";
393 $T_HDR_data = "\.data\n\t\.align 8\n";
394 $T_HDR_rodata = "\.text\n\t\.align 4\n";
395 $T_HDR_closure = "\.data\n\t\.align 4\n";
396 $T_HDR_info = "\.text\n\t\.align 4\n";
397 $T_HDR_entry = "\.text\n\t\.align 4\n";
398 $T_HDR_vector = "\.text\n\t\.align 4\n";
400 #--------------------------------------------------------#
401 } elsif ( $TargetPlatform =~ /^sparc-.*-linux/ ) {
402 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
403 $T_US = ''; # _ if symbols have an underscore on the front
404 $T_PRE_APP = '#'; # regexp that says what comes before APP/NO_APP
405 # Probably doesn't apply anyway
406 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
409 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.seg|\.stab.*|\s+?\.section.*|\s+\.type.*|\s+\.size.*)\n)';
410 $T_COPY_DIRVS = '\.(global|globl|proc|stab)';
412 $T_DOT_WORD = '\.(long|word|nword|xword|byte|half|short|skip|uahalf|uaword)';
413 $T_DOT_GLOBAL = '^\t\.global';
414 $T_HDR_literal = "\.text\n\t\.align 8\n";
415 $T_HDR_misc = "\.text\n\t\.align 4\n";
416 $T_HDR_data = "\.data\n\t\.align 8\n";
417 $T_HDR_rodata = "\.text\n\t\.align 4\n";
418 $T_HDR_closure = "\.data\n\t\.align 4\n";
419 $T_HDR_info = "\.text\n\t\.align 4\n";
420 $T_HDR_entry = "\.text\n\t\.align 4\n";
421 $T_HDR_vector = "\.text\n\t\.align 4\n";
423 #--------------------------------------------------------#
425 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
429 if($T_HDR_relrodata eq "") {
431 # relrodata defaults to rodata.
432 $T_HDR_relrodata = $T_HDR_rodata;
436 print STDERR "T_STABBY: $T_STABBY\n";
437 print STDERR "T_US: $T_US\n";
438 print STDERR "T_PRE_APP: $T_PRE_APP\n";
439 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
440 print STDERR "T_POST_LBL: $T_POST_LBL\n";
441 if ( $TargetPlatform =~ /^i386-/ ) {
442 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
443 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
444 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
446 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
447 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
448 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
449 print STDERR "T_HDR_literal: $T_HDR_literal\n";
450 print STDERR "T_HDR_misc: $T_HDR_misc\n";
451 print STDERR "T_HDR_data: $T_HDR_data\n";
452 print STDERR "T_HDR_rodata: $T_HDR_rodata\n";
453 print STDERR "T_HDR_closure: $T_HDR_closure\n";
454 print STDERR "T_HDR_info: $T_HDR_info\n";
455 print STDERR "T_HDR_entry: $T_HDR_entry\n";
456 print STDERR "T_HDR_vector: $T_HDR_vector\n";
462 %************************************************************************
464 \subsection{Mangle away}
466 %************************************************************************
470 local($in_asmf, $out_asmf) = @_;
472 # multi-line regexp matching:
477 &init_TARGET_STUFF();
478 &init_FUNNY_THINGS();
480 open(INASM, "< $in_asmf")
481 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
482 open(OUTASM,"> $out_asmf")
483 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
485 # read whole file, divide into "chunks":
486 # record some info about what we've found...
488 @chk = (); # contents of the chunk
489 $numchks = 0; # number of them
490 @chkcat = (); # what category of thing in each chunk
491 @chksymb = (); # what symbol(base) is defined in this chunk
492 %entrychk = (); # ditto, its entry code
493 %closurechk = (); # ditto, the (static) closure
494 %srtchk = (); # ditto, its SRT (for top-level things)
495 %infochk = (); # given a symbol base, say what chunk its info tbl is in
496 %vectorchk = (); # ditto, return vector table
497 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
499 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
502 tr/\r//d if $TargetPlatform =~ /-mingw32$/; # In case Perl doesn't convert line endings
503 next if $T_STABBY && /^\.stab.*${T_US}__stg_split_marker/o;
504 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
505 next if /^\t\.def.*endef$/;
506 next if /${T_PRE_APP}(NO_)?APP/o;
507 next if /^;/ && $TargetPlatform =~ /^hppa/;
509 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|ia64)-/;
511 if ( $TargetPlatform =~ /^mips-/
512 && /^\t\.(globl\S+\.text|comm\t)/ ) {
513 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
514 # Treat .comm variables as data. These show up in two (known) places:
516 # - the module_registered variable used in the __stginit fragment.
517 # even though these are declared static and initialised, gcc 3.3
518 # likes to make them .comm, presumably to save space in the
521 # - global variables used to pass arguments from C to STG in
522 # a foreign export. (is this still true? --SDM)
524 } elsif ( /^\t\.comm.*$/ ) {
526 $chkcat[$i] = 'data';
529 # Labels ending "_str": these are literal strings.
530 } elsif ( /^${T_US}([A-Za-z0-9_]+)_str${T_POST_LBL}$/ ) {
532 $chkcat[$i] = 'rodata';
534 } elsif ( $TargetPlatform =~ /-darwin/
535 && (/^\s*\.subsections_via_symbols/
536 ||/^\s*\.no_dead_strip.*/)) {
537 # Don't allow Apple's linker to do any dead-stripping of symbols
538 # in this file, because it will mess up info-tables in mangled
540 # The .no_dead_strip directives are actually put there by
541 # the gcc3 "used" attribute on entry points.
542 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
543 # duplicated from the bottom
547 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
548 # Alphas: Local labels not to be confused with new chunks
550 # NB: all the rest start with a non-space
552 } elsif ( $TargetPlatform =~ /^mips-/
553 && /^\d+:/ ) { # a funny-looking very-local label
556 } elsif ( /$T_CONST_LBL/o ) {
558 $chkcat[$i] = 'literal';
561 } elsif ( /^${T_US}__stg_split_marker(\d*)${T_POST_LBL}$/o ) {
563 $chkcat[$i] = 'splitmarker';
566 } elsif ( /^${T_US}([A-Za-z0-9_]+)_info${T_POST_LBL}$/o ) {
569 $chkcat[$i] = 'infotbl';
570 $chksymb[$i] = $symb;
572 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
574 $infochk{$symb} = $i;
576 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(entry|ret)${T_POST_LBL}$/o ) {
578 $chkcat[$i] = 'entry';
583 } elsif ( /^${T_US}([A-Za-z0-9_]+)_closure${T_POST_LBL}$/o ) {
585 $chkcat[$i] = 'closure';
588 $closurechk{$1} = $i;
590 } elsif ( /^${T_US}([A-Za-z0-9_]+)_srt${T_POST_LBL}$/o ) {
597 } elsif ( /^${T_US}([A-Za-z0-9_]+)_ct${T_POST_LBL}$/o ) {
599 $chkcat[$i] = 'data';
602 } elsif ( /^${T_US}(stg_ap_stack_entries|stg_stack_save_entries|stg_arg_bitmaps)${T_POST_LBL}$/o ) {
604 $chkcat[$i] = 'data';
607 } elsif ( /^(${T_US}__gnu_compiled_c|gcc2_compiled\.)${T_POST_LBL}/o ) {
610 } elsif ( /^${T_US}[A-Za-z0-9_]+\.\d+${T_POST_LBL}$/o
611 || /^${T_US}.*_CAT${T_POST_LBL}$/o # PROF: _entryname_CAT
612 || /^${T_US}.*_done${T_POST_LBL}$/o # PROF: _module_done
613 || /^${T_US}_module_registered${T_POST_LBL}$/o # PROF: _module_registered
616 $chkcat[$i] = 'data';
619 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
624 } elsif ( /^${T_US}([A-Za-z0-9_]+)_cc(s)?${T_POST_LBL}$/o ) {
625 # all CC_ symbols go in the data section...
627 $chkcat[$i] = 'data';
630 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(alt|dflt)${T_POST_LBL}$/o ) {
632 $chkcat[$i] = 'misc';
634 } elsif ( /^${T_US}([A-Za-z0-9_]+)_vtbl${T_POST_LBL}$/o ) {
636 $chkcat[$i] = 'vector';
641 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
642 && /^[A-Za-z0-9][A-Za-z0-9_]*:/ ) {
643 # Some Solaris system headers contain function definitions (as
644 # opposed to mere prototypes), which end up in the .hc file when
645 # a Haskell module foreign imports the corresponding system
646 # functions (most notably stat()). We put them into the text
647 # segment. Note that this currently does not extend to function
648 # names starting with an underscore.
651 $chkcat[$i] = 'misc';
654 } elsif ( /^${T_US}[A-Za-z0-9_]/o
655 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
657 && ( $TargetPlatform !~ /^powerpc64/ # we need to avoid local labels in this case
658 || ! /^\.L\d+:$/ ) ) {
665 /^${T_US}stg_.*${T_POST_LBL}$/o # RTS internals
666 || /^${T_US}__stg_.*${T_POST_LBL}$/o # more RTS internals
667 || /^${T_US}__fexp_.*${T_POST_LBL}$/o # foreign export
668 || /^${T_US}.*_slow${T_POST_LBL}$/o # slow entry
669 || /^${T_US}__stginit.*${T_POST_LBL}$/o # __stginit<module>
670 || /^${T_US}.*_btm${T_POST_LBL}$/o # large bitmaps
671 || /^${T_US}.*_fast${T_POST_LBL}$/o # primops
672 || /^_uname:/o # x86/Solaris2
675 $chkcat[$i] = 'misc';
677 /^${T_US}.*_srtd${T_POST_LBL}$/o # large bitmaps
678 || /^${T_US}.*_closure_tbl${T_POST_LBL}$/o # closure tables
681 $chkcat[$i] = 'relrodata';
684 print STDERR "Warning: retaining unknown function \`$thing' in output from C compiler\n";
685 $chkcat[$i] = 'unknown';
688 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && (
690 || /^\.section __TEXT,__picsymbol_stub1,.*/
691 || /^\.section __TEXT,__picsymbolstub1,.*/
693 || /^\.section __TEXT,__symbol_stub1,.*/
694 || /^\.section __TEXT,__symbolstub1,.*/
695 || /^\.lazy_symbol_pointer/
696 || /^\.non_lazy_symbol_pointer/ ))
699 $chkcat[$i] = 'dyld';
701 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.data/ && $chkcat[$i] eq 'dyld')
702 { # non_lazy_symbol_ptrs that point to local symbols
704 $chkcat[$i] = 'dyld';
706 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ && /^\.LCTOC1 = /o ) {
707 # PowerPC Linux's large-model PIC (-fPIC) generates a gobal offset
708 # table "by hand". Be sure to copy it over.
709 # Note that this label and all entries in the table should actually
710 # go into the .got2 section, but it isn't easy to distinguish them
711 # from other constant literals (.LC\d+), so we just put everything
714 $chkcat[$i] = 'literal';
715 $chksymb[$i] = 'LCTOC1';
716 } else { # simple line (duplicated at the top)
721 $numchks = $#chk + 1;
723 # open CHUNKS, ">/tmp/chunks1" or die "Cannot open /tmp/chunks1: $!\n";
724 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
727 # the division into chunks is imperfect;
728 # we throw some things over the fence into the next
731 # also, there are things we would like to know
732 # about the whole module before we start spitting
735 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
736 local($FIRST_TOSSABLE ) = ($TargetPlatform =~ /^(hppa|mips-)/) ? 1 : 0;
738 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
740 # Alphas: NB: we start meddling at chunk 1, not chunk 0
741 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
742 # spits a ".quad 0" in after the very first ".rdata"; we
743 # detect this special case (tossing the ".quad 0")!
744 local($magic_rdata_seen) = 0;
746 # HPPAs, MIPSen: also start medding at chunk 1
748 for ($i = $FIRST_TOSSABLE; $i < $numchks; $i++) {
749 $c = $chk[$i]; # convenience copy
751 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
753 # toss all prologue stuff; HPPA is pretty weird
755 $c = &hppa_mash_prologue($c) if $TargetPlatform =~ /^hppa-/;
757 # be slightly paranoid to make sure there's
758 # nothing surprising in there
759 if ( $c =~ /--- BEGIN ---/ ) {
760 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
762 # remove junk whitespace around the split point
766 if ($TargetPlatform =~ /^i386-/) {
767 $p =~ s/^\tpushl\s+\%edi\n//;
768 $p =~ s/^\tpushl\s+\%esi\n//;
769 $p =~ s/^\tpushl\s+\%ebx\n//;
770 $p =~ s/^\tmovl\s+\%esi,\s*\d*\(\%esp\)\n//;
771 $p =~ s/^\tmovl\s+\%edi,\s*\d*\(\%esp\)\n//;
772 $p =~ s/^\tsubl\s+\$\d+,\s*\%esp\n//;
773 $p =~ s/^\tmovl\s+\$\d+,\s*\%eax\n\tcall\s+__alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
775 # GCC 3.1 is in the habit of adding spurious writes to the
776 # stack in the prologue. Just to be on the safe side,
777 # chuck these over the fence into the main code.
778 while ($p =~ /^\tmovl\s+\$\d+,\s*\d*\(\%esp\)\n/) {
779 # print "Spurious instruction: $&";
784 } elsif ($TargetPlatform =~ /^x86_64-/) {
785 $p =~ s/^\tpushq\s+\%r(bx|bp|12|13|14)\n//g;
786 $p =~ s/^\tmovq\s+\%r(bx|bp|12|13|14),\s*\d*\(\%rsp\)\n//g;
787 $p =~ s/^\tsubq\s+\$\d+,\s*\%rsp\n//;
789 } elsif ($TargetPlatform =~ /^ia64-/) {
790 $p =~ s/^\t\.prologue .*\n//;
791 $p =~ s/^\t\.save ar\.pfs, r\d+\n\talloc r\d+ = ar\.pfs, 0, 3[12], \d+, 0\n//;
792 $p =~ s/^\t\.fframe \d+\n\tadds r12 = -\d+, r12\n//;
793 $p =~ s/^\t\.save rp, r\d+\n\tmov r\d+ = b0\n//;
794 $p =~ s/^\t\.(mii|mmi)\n//g; # bundling is no longer sensible
795 $p =~ s/^\t;;\n//g; # discard stops
796 $p =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
798 # GCC 3.3 saves r1 in the prologue, move this to the body
799 if ($p =~ /^\tmov r\d+ = r1\n/) {
803 } elsif ($TargetPlatform =~ /^m68k-/) {
804 $p =~ s/^\tlink a6,#-?\d.*\n//;
805 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
806 # The above showed up in the asm code,
807 # so I added it here.
808 # I hope it's correct.
810 $p =~ s/^\tmovel d2,sp\@-\n//;
811 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
812 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
813 } elsif ($TargetPlatform =~ /^mips-/) {
814 # the .frame/.mask/.fmask that we use is the same
815 # as that produced by GCC for miniInterpret; this
816 # gives GDB some chance of figuring out what happened
817 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
818 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
819 $p =~ s/^\t\.(mask|fmask).*\n//g;
820 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
821 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
822 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
823 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
824 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
825 $p =~ s/__FRAME__/$FRAME/;
826 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
828 $pcrel_label =~ s/(.|\n)*^(\"?L\d+\$pb\"?):\n(.|\n)*/$2/ or $pcrel_label = "";
830 $p =~ s/^\tmflr r0\n//;
831 $p =~ s/^\tbl saveFP # f\d+\n//;
832 $p =~ s/^\tbl saveFP ; save f\d+-f\d+\n//;
833 $p =~ s/^\"?L\d+\$pb\"?:\n//;
834 $p =~ s/^\tstmw r\d+,-\d+\(r1\)\n//;
835 $p =~ s/^\tstfd f\d+,-\d+\(r1\)\n//g;
836 $p =~ s/^\tstw r0,\d+\(r1\)\n//g;
837 $p =~ s/^\tstwu r1,-\d+\(r1\)\n//;
838 $p =~ s/^\tstw r\d+,-\d+\(r1\)\n//g;
839 $p =~ s/^\tbcl 20,31,L\d+\$pb\n//;
840 $p =~ s/^L\d+\$pb:\n//;
841 $p =~ s/^\tmflr r31\n//;
843 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
844 # under some circumstances, only when generating position dependent code.
845 # I have no idea why, and I don't think it is necessary, so let's toss it.
846 $p =~ s/^\tli r\d+,0\n//g;
847 $p =~ s/^\tstw r\d+,\d+\(r1\)\n//g;
848 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/) {
849 $p =~ s/^\tmflr 0\n//;
850 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
851 $p =~ s/^\tstfd \d+,\d+\(1\)\n//g;
852 $p =~ s/^\tstw r0,8\(1\)\n//;
853 $p =~ s/^\tstwu 1,-\d+\(1\)\n//;
854 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
856 # GCC's "large-model" PIC (-fPIC)
858 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
860 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
861 $p =~ s/^.LCF\d+:\n//;
862 $p =~ s/^\tmflr 30\n//;
863 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
864 $p =~ s/^\tadd 30,0,30\n//;
866 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
867 # under some circumstances, only when generating position dependent code.
868 # I have no idea why, and I don't think it is necessary, so let's toss it.
869 $p =~ s/^\tli \d+,0\n//g;
870 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
871 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
872 $p =~ s/^\tmr 31,1\n//;
873 $p =~ s/^\tmflr 0\n//;
874 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
875 $p =~ s/^\tstfd \d+,-?\d+\(1\)\n//g;
876 $p =~ s/^\tstd r0,8\(1\)\n//;
877 $p =~ s/^\tstdu 1,-\d+\(1\)\n//;
878 $p =~ s/^\tstd \d+,-?\d+\(1\)\n//g;
880 # GCC's "large-model" PIC (-fPIC)
882 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
884 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
885 $p =~ s/^.LCF\d+:\n//;
886 $p =~ s/^\tmflr 30\n//;
887 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
888 $p =~ s/^\tadd 30,0,30\n//;
890 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
891 # under some circumstances, only when generating position dependent code.
892 # I have no idea why, and I don't think it is necessary, so let's toss it.
893 $p =~ s/^\tli \d+,0\n//g;
894 $p =~ s/^\tstd \d+,\d+\(1\)\n//g;
896 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
899 # HWL HACK: dont die, just print a warning
900 #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/;
901 die "Prologue junk?: $p\n" if $p =~ /^\s+[^\s\.]/;
903 # For PIC, we want to keep part of the prologue
904 if ($TargetPlatform =~ /^powerpc-apple-.*/ && $pcrel_label ne "") {
905 # Darwin: load the current instruction pointer into register r31
906 $p .= "bcl 20,31,$pcrel_label\n";
907 $p .= "$pcrel_label:\n";
908 $p .= "\tmflr r31\n";
909 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/ && $pcrel_label ne "") {
910 # Linux: load the GOT pointer into register 30
911 $p .= "\tbcl 20,31,.LCF$pcrel_label\n";
912 $p .= ".LCF$pcrel_label:\n";
914 $p .= "\tlwz 0,.LCL$pcrel_label-.LCF$pcrel_label(30)\n";
915 $p .= "\tadd 30,0,30\n";
918 # glue together what's left
923 if ( $TargetPlatform =~ /^mips-/ ) {
924 # MIPS: first, this basic sequence may occur "--- END ---" or not
925 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
928 # toss all epilogue stuff; again, paranoidly
929 if ( $c =~ /--- END ---/ ) {
930 if (($r, $e) = split(/--- END ---/, $c)) {
931 if ($TargetPlatform =~ /^i386-/) {
933 $e =~ s/^\tpopl\s+\%edi\n//;
934 $e =~ s/^\tpopl\s+\%esi\n//;
935 $e =~ s/^\tpopl\s+\%edx\n//;
936 $e =~ s/^\tpopl\s+\%ecx\n//;
937 $e =~ s/^\taddl\s+\$\d+,\s*\%esp\n//;
938 $e =~ s/^\tsubl\s+\$-\d+,\s*\%esp\n//;
939 } elsif ($TargetPlatform =~ /^ia64-/) {
940 $e =~ s/^\tmov ar\.pfs = r\d+\n//;
941 $e =~ s/^\tmov b0 = r\d+\n//;
942 $e =~ s/^\t\.restore sp\n\tadds r12 = \d+, r12\n//;
943 $e =~ s/^\tbr\.ret\.sptk\.many b0\n//;
944 $e =~ s/^\t\.(mii|mmi|mib)\n//g; # bundling is no longer sensible
945 $e =~ s/^\t;;\n//g; # discard stops - stop at end of body is sufficient
946 $e =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
947 } elsif ($TargetPlatform =~ /^m68k-/) {
948 $e =~ s/^\tunlk a6\n//;
950 } elsif ($TargetPlatform =~ /^mips-/) {
951 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
952 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
953 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
954 $e =~ s/^\tj\t\$31\n//;
955 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
956 $e =~ s/^\taddi r1,r1,\d+\n//;
957 $e =~ s/^\tlwz r\d+,\d+\(r1\)\n//;
958 $e =~ s/^\tlmw r\d+,-\d+\(r1\)\n//;
959 $e =~ s/^\tmtlr r0\n//;
961 $e =~ s/^\tb restFP ;.*\n//;
962 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
963 $e =~ s/^\tmr 3,0\n//;
964 $e =~ s/^\taddi 1,1,\d+\n//;
965 $e =~ s/^\tld 0,16\(1\)\n//;
966 $e =~ s/^\tmtlr 0\n//;
968 # callee-save registers
969 $e =~ s/^\tld \d+,-?\d+\(1\)\n//g;
970 $e =~ s/^\tlfd \d+,-?\d+\(1\)\n//g;
972 # get rid of the debug junk along with the blr
973 $e =~ s/^\tblr\n\t.long .*\n\t.byte .*\n//;
975 # incase we missed it with the last one get the blr alone
978 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
981 print STDERR "WARNING: Epilogue junk?: $e\n" if $e =~ /^\t\s*[^\.\s\n]/;
983 # glue together what's left
985 $c =~ s/\n\t\n/\n/; # junk blank line
989 # On SPARCs, we don't do --- BEGIN/END ---, we just
990 # toss the register-windowing save/restore/ret* instructions
991 # directly unless they've been generated by function definitions in header
993 if ( $TargetPlatform =~ /^sparc-/ ) {
994 if ( ! ( $TargetPlatform =~ /solaris2$/ && $chkcat[$i] eq 'unknown' )) {
995 $c =~ s/^\t(save.*|restore.*|ret|retl)\n//g;
997 # throw away PROLOGUE comments
998 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
1001 # On Alphas, the prologue mangling is done a little later (below)
1003 # toss all calls to __DISCARD__
1004 $c =~ s/^\t(call|jbsr|jal)\s+${T_US}__DISCARD__\n//go;
1005 $c =~ s/^\tjsr\s+\$26\s*,\s*${T_US}__DISCARD__\n//go if $TargetPlatform =~ /^alpha-/;
1006 $c =~ s/^\tbl\s+L___DISCARD__\$stub\n//go if $TargetPlatform =~ /^powerpc-apple-.*/;
1007 $c =~ s/^\tbl\s+__DISCARD__(\@plt)?\n//go if $TargetPlatform =~ /^powerpc-.*-linux/;
1008 $c =~ s/^\tbl\s+\.__DISCARD__\n\s+nop\n//go if $TargetPlatform =~ /^powerpc64-.*-linux/;
1010 # IA64: mangle tailcalls into jumps here
1011 if ($TargetPlatform =~ /^ia64-/) {
1012 while ($c =~ s/^\tbr\.call\.sptk\.many b0 = (.*)\n(?:^\.L([0-9]*):\n)?(?:\t;;\n)?(?:\tmov r1 = r\d+\n)?(?:\t;;\n)?\t--- TAILCALL ---\n(?:\t;;\n\tbr \.L\d+\n)?/\tbr\.few $1\n/) {
1013 # Eek, the gcc optimiser is getting smarter... if we see a jump to the --- TAILCALL ---
1014 # marker then we reapply the substitution at the source sites
1015 $c =~ s/^\tbr \.L$2\n/\t--- TAILCALL ---\n/g if ($2);
1019 # MIPS: that may leave some gratuitous asm macros around
1020 # (no harm done; but we get rid of them to be tidier)
1021 $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/
1022 if $TargetPlatform =~ /^mips-/;
1024 # toss stack adjustment after DoSparks
1025 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
1026 if $TargetPlatform =~ /^m68k-/; # this looks old...
1028 if ( $TargetPlatform =~ /^alpha-/ &&
1029 ! $magic_rdata_seen &&
1030 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
1031 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
1032 $magic_rdata_seen = 1;
1035 # pick some end-things and move them to the next chunk
1037 # pin a funny end-thing on (for easier matching):
1038 $c .= 'FUNNY#END#THING';
1040 while ( $c =~ /${T_MOVE_DIRVS}FUNNY#END#THING/o ) {
1044 # on x86 we try not to copy any directives into a literal
1045 # chunk, rather we keep looking for the next real chunk. This
1046 # is because we get things like
1048 # .globl blah_closure
1054 if ( $TargetPlatform =~ /^(i386|sparc|powerpc)/ && $to_move =~ /${T_COPY_DIRVS}/ ) {
1056 while ( $j < $numchks && $chk[$j] =~ /$T_CONST_LBL/) {
1059 if ( $j < $numchks ) {
1060 $chk[$j] = $to_move . $chk[$j];
1064 elsif ( $i < ($numchks - 1)
1065 && ( $to_move =~ /${T_COPY_DIRVS}/
1066 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
1067 $chk[$i + 1] = $to_move . $chk[$i + 1];
1068 # otherwise they're tossed
1071 $c =~ s/${T_MOVE_DIRVS}FUNNY#END#THING/FUNNY#END#THING/o;
1074 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
1076 # toss all prologue stuff, except for loading gp, and the ..ng address
1077 unless ($c =~ /\.ent.*\n\$.*\.\.ng:/) {
1078 if (($p, $r) = split(/^\t\.prologue/, $c)) {
1079 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
1080 $keep =~ s/^\t\.frame.*\n/\t.frame \$30,0,\$26,0\n/;
1081 $keep =~ s/^\t\.(mask|fmask).*\n//g;
1082 $c = $keep . "..ng:\n";
1084 print STDERR "malformed code block ($ent)?\n"
1087 $c .= "\t.prologue" . $r;
1091 $c =~ s/FUNNY#END#THING//;
1093 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
1095 $chk[$i] = $c; # update w/ convenience copy
1098 # open CHUNKS, ">/tmp/chunks2" or die "Cannot open /tmp/chunks2: $!\n";
1099 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
1102 if ( $TargetPlatform =~ /^alpha-/ ) {
1103 # print out the header stuff first
1104 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
1105 print OUTASM $chk[0];
1107 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1108 print OUTASM $chk[0];
1110 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1111 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
1113 # get rid of horrible "<dollar>Revision: .*$" strings
1114 local(@lines0) = split(/\n/, $chk[0]);
1116 while ( $z <= $#lines0 ) {
1117 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
1120 while ( $z <= $#lines0 ) {
1122 last if $lines0[$z] =~ /[,\t]0x0$/;
1128 $chk[0] = join("\n", @lines0);
1129 $chk[0] =~ s/\n\n+/\n/;
1130 print OUTASM $chk[0];
1133 # print out all the literal strings next
1134 for ($i = 0; $i < $numchks; $i++) {
1135 if ( $chkcat[$i] eq 'literal' ) {
1136 print OUTASM $T_HDR_literal, $chk[$i];
1137 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
1139 $chkcat[$i] = 'DONE ALREADY';
1143 # on the HPPA, print out all the bss next
1144 if ( $TargetPlatform =~ /^hppa/ ) {
1145 for ($i = 1; $i < $numchks; $i++) {
1146 if ( $chkcat[$i] eq 'bss' ) {
1147 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
1148 print OUTASM $chk[$i];
1150 $chkcat[$i] = 'DONE ALREADY';
1155 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
1156 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
1158 next if $chkcat[$i] eq 'DONE ALREADY';
1160 if ( $chkcat[$i] eq 'misc' || $chkcat[$i] eq 'unknown' ) {
1161 if ($chk[$i] ne '') {
1162 print OUTASM $T_HDR_misc;
1163 &print_doctored($chk[$i], 0);
1166 } elsif ( $chkcat[$i] eq 'toss' ) {
1167 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
1169 } elsif ( $chkcat[$i] eq 'data' ) {
1170 if ($chk[$i] ne '') {
1171 print OUTASM $T_HDR_data;
1172 print OUTASM $chk[$i];
1175 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
1176 # we can just re-constitute this one...
1177 # NB: we emit _three_ underscores no matter what,
1178 # so ghc-split doesn't have to care.
1179 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
1181 } elsif ( $chkcat[$i] eq 'closure'
1182 || $chkcat[$i] eq 'srt'
1183 || $chkcat[$i] eq 'infotbl'
1184 || $chkcat[$i] eq 'entry') { # do them in that order
1185 $symb = $chksymb[$i];
1188 if ( defined($closurechk{$symb}) ) {
1189 print OUTASM $T_HDR_closure;
1190 print OUTASM $chk[$closurechk{$symb}];
1191 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
1195 if ( defined($srtchk{$symb}) ) {
1196 print OUTASM $T_HDR_relrodata;
1197 print OUTASM $chk[$srtchk{$symb}];
1198 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
1202 if ( defined($infochk{$symb}) ) {
1204 print OUTASM $T_HDR_info;
1205 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
1207 # entry code will be put here!
1209 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
1213 if ( defined($entrychk{$symb}) ) {
1215 $c = $chk[$entrychk{$symb}];
1217 # If this is an entry point with an info table,
1218 # eliminate the entry symbol and all directives involving it.
1219 if (defined($infochk{$symb}) && $TargetPlatform !~ /^ia64-/) {
1221 foreach $l (split(/\n/,$c)) {
1222 next if $l =~ /^.*$symb_(entry|ret)${T_POST_LBL}/;
1224 # If we have .type/.size direrctives involving foo_entry,
1225 # then make them refer to foo_info instead. The information
1226 # in these directives is used by the cachegrind annotator,
1227 # so it is worthwhile keeping.
1228 if ($l =~ /^\s*\.(type|size).*$symb_(entry|ret)/) {
1229 $l =~ s/$symb(_entry|_ret)/${symb}_info/g;
1233 next if $l =~ /^\s*\..*$symb.*\n?/;
1236 $c = join("\n",@o) . "\n";
1239 print OUTASM $T_HDR_entry;
1241 &print_doctored($c, 1); # NB: the 1!!!
1243 $chkcat[$entrychk{$symb}] = 'DONE ALREADY';
1246 } elsif ( $chkcat[$i] eq 'vector' ) {
1247 $symb = $chksymb[$i];
1250 if ( defined($vectorchk{$symb}) ) {
1251 print OUTASM $T_HDR_vector;
1252 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1254 # direct return code will be put here!
1255 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1257 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1258 # Alphas: the commented nop is for the splitter, to ensure
1259 # that no module ends with a label as the very last
1260 # thing. (The linker will adjust the label to point
1261 # to the first code word of the next module linked in,
1262 # even if alignment constraints cause the label to move!)
1264 print OUTASM "\t# nop\n";
1267 } elsif ( $chkcat[$i] eq 'rodata' ) {
1268 print OUTASM $T_HDR_rodata;
1269 print OUTASM $chk[$i];
1270 $chkcat[$i] = 'DONE ALREADY';
1271 } elsif ( $chkcat[$i] eq 'relrodata' ) {
1272 print OUTASM $T_HDR_relrodata;
1273 print OUTASM $chk[$i];
1274 $chkcat[$i] = 'DONE ALREADY';
1275 } elsif ( $chkcat[$i] eq 'toc' ) {
1276 # silly optimisation to print tocs, since they come in groups...
1277 print OUTASM $T_HDR_toc;
1279 while ($chkcat[$j] eq 'toc')
1280 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1283 print OUTASM $chk[$j];
1285 $chkcat[$j] = 'DONE ALREADY';
1289 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && $chkcat[$i] eq 'dyld' ) {
1290 # powerpc-apple: dynamic linker stubs
1291 if($chk[$i] !~ /\.indirect_symbol ___DISCARD__/)
1292 { # print them out unchanged, but remove the stubs for __DISCARD__
1293 print OUTASM $chk[$i];
1296 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1300 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1303 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1304 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1309 sub hppa_mash_prologue { # OK, epilogue, too
1312 # toss all prologue stuff
1313 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1315 # Lie about our .CALLINFO
1316 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1323 # toss all epilogue stuff
1324 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1326 # Sorry; we moved the _info stuff to the code segment.
1327 s/_info,DATA/_info,CODE/g;
1334 sub print_doctored {
1335 local($_, $need_fallthru_patch) = @_;
1337 if ( $TargetPlatform !~ /^i386-/
1338 || ! /^\t[a-z]/ # no instructions in here, apparently
1339 || /^${T_US}__stginit_[A-Za-z0-9_]+${T_POST_LBL}/) {
1343 # OK, must do some x86 **HACKING**
1345 local($entry_patch) = '';
1346 local($exit_patch) = '';
1348 # gotta watch out for weird instructions that
1349 # invisibly smash various regs:
1350 # rep* %ecx used for counting
1351 # scas* %edi used for destination index
1352 # cmps* %e[sd]i used for indices
1353 # loop* %ecx used for counting
1358 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1360 # * GCC used an "STG reg" for its own purposes
1362 # * some secret uses of machine reg, requiring STG reg
1363 # to be saved/restored
1365 # The most dangerous "GCC uses" of an "STG reg" are when
1366 # the reg holds the target of a jmp -- it's tricky to
1367 # insert the patch-up code before we get to the target!
1368 # So here we change the jmps:
1370 # --------------------------------------------------------
1371 # it can happen that we have jumps of the form...
1372 # jmp *<something involving %esp>
1374 # jmp <something involving another naughty register...>
1376 # a reasonably-common case is:
1378 # movl $_blah,<bad-reg>
1381 s/^\tmovl\s+\$${T_US}(.*),\s*(\%e[acd]x)\n\tjmp\s+\*\2/\tjmp $T_US$1/g;
1385 # movl -4(%ebx), %eax
1390 s/^\tmovl\s+(-?\d*\(\%e(bx|si)\)),\s*(\%e[acd]x)\n\tjmp\s+\*\3/\tjmp\t\*$1/g;
1392 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1393 s/^\tmovl\s+(.*),\s*\%esi\n\tjmp\s+\*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1394 s/^\tjmp\s+\*(.*\(.*\%esi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1395 s/^\tjmp\s+\*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1396 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1397 if /(jmp|call)\s+.*\%esi/;
1399 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1400 s/^\tmovl\s+(.*),\s*\%edi\n\tjmp\s+\*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1401 s/^\tjmp\s+\*(.*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1402 s/^\tjmp\s+\*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1403 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1404 if /(jmp|call)\s+.*\%edi/;
1407 # OK, now we can decide what our patch-up code is going to
1410 # Offsets into register table - you'd better update these magic
1411 # numbers should you change its contents!
1412 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1413 local($OFFSET_Hp)=88;
1415 # Note funky ".=" stuff; we're *adding* to these _patch guys
1416 if ( $StolenX86Regs <= 2
1417 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1418 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1419 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1421 # nothing for call_{entry,exit} because %esi is callee-save
1423 if ( $StolenX86Regs <= 3
1424 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1425 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1426 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1428 # nothing for call_{entry,exit} because %edi is callee-save
1431 # --------------------------------------------------------
1432 # next, here we go with non-%esp patching!
1434 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1436 # Before calling GC we must set up the exit condition before the call
1437 # and entry condition when we come back
1439 # fix _all_ non-local jumps:
1441 s/^\tjmp\s+\*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1442 s/^\tjmp\s+${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1444 s/^(\tjmp\s+.*\n)/$exit_patch$1/g; # here's the fix...
1446 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1447 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1449 if ($StolenX86Regs == 2 ) {
1450 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1451 if /^\t(jmp|call)\s+.*\%e(si|di)/;
1452 } elsif ($StolenX86Regs == 3 ) {
1453 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1454 if /^\t(jmp|call)\s+.*\%edi/;
1457 # --------------------------------------------------------
1458 # that's it -- print it
1460 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1464 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1465 print OUTASM $exit_patch;
1466 # ToDo: make it not print if there is a "jmp" at the end
1472 sub init_FUNNY_THINGS {
1473 %KNOWN_FUNNY_THING = (
1475 # "${T_US}stg_.*{T_POST_LBL}", 1,
1480 The following table reversal is used for both info tables and return
1481 vectors. In both cases, we remove the first entry from the table,
1482 reverse the table, put the label at the end, and paste some code
1483 (that which is normally referred to by the first entry in the table)
1484 right after the table itself. (The code pasting is done elsewhere.)
1488 local($symb, $tbl, $discard1) = @_;
1490 return ($tbl) if ($TargetPlatform =~ /^ia64-/);
1492 local($before) = '';
1494 local(@imports) = (); # hppa only
1497 local(@lines) = split(/\n/, $tbl);
1500 # Deal with the header...
1501 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1502 $label .= $lines[$i] . "\n",
1503 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info${T_POST_LBL}$/o
1504 || $lines[$i] =~ /${T_DOT_GLOBAL}/o
1505 || $lines[$i] =~ /^${T_US}\S+_vtbl${T_POST_LBL}$/o;
1507 $before .= $lines[$i] . "\n"; # otherwise...
1511 $infoname =~ s/(.|\n)*^([A-Za-z0-9_]+_info)${T_POST_LBL}$(.|\n)*/\2/;
1513 # Grab the table data...
1514 if ( $TargetPlatform !~ /^hppa/ ) {
1515 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1517 # Convert addresses of SRTs, slow entrypoints and large bitmaps
1518 # to offsets (relative to the info label),
1519 # in order to support position independent code.
1520 $line =~ s/$infoname/0/
1521 || $line =~ s/([A-Za-z0-9_]+_srtd)$/\1 - $infoname/
1522 || $line =~ s/([A-Za-z0-9_]+_srt(\+\d+)?)$/\1 - $infoname/
1523 || $line =~ s/([A-Za-z0-9_]+_slow)$/\1 - $infoname/
1524 || $line =~ s/([A-Za-z0-9_]+_btm)$/\1 - $infoname/
1525 || $line =~ s/([A-Za-z0-9_]+_alt)$/\1 - $infoname/
1526 || $line =~ s/([A-Za-z0-9_]+_dflt)$/\1 - $infoname/
1527 || $line =~ s/([A-Za-z0-9_]+_ret)$/\1 - $infoname/;
1528 push(@words, $line);
1530 } else { # hppa weirdness
1531 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+(${T_DOT_WORD}|\.IMPORT)/; $i++) {
1532 # FIXME: the RTS now expects offsets instead of addresses
1533 # for all labels in info tables.
1534 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1535 push(@imports, $lines[$i]);
1537 # We don't use HP's ``function pointers''
1538 # We just use labels in code space, like normal people
1539 $lines[$i] =~ s/P%//;
1540 push(@words, $lines[$i]);
1545 # Now throw away any initial zero word from the table. This is a hack
1546 # that lets us reduce the size of info tables when the SRT field is not
1547 # needed: see comments StgFunInfoTable in InfoTables.h.
1549 # The .zero business is for Linux/ELF.
1550 # The .skip business is for Sparc/Solaris/ELF.
1551 # The .blockz business is for HPPA.
1553 # if ($words[0] =~ /^\t?(${T_DOT_WORD}\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1558 for (; $i <= $#lines; $i++) {
1559 $after .= $lines[$i] . "\n";
1562 # Alphas: If we have anonymous text (not part of a procedure), the
1563 # linker may complain about missing exception information. Bleh.
1564 # To suppress this, we place a .ent/.end pair around the code.
1565 # At the same time, we have to be careful and not enclose any leading
1566 # .file/.loc directives.
1567 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1568 local ($ident) = $1;
1569 $before =~ s/^((\s*\.(file|loc)\s+[^\n]*\n)*)/$1\t.ent $ident\n/;
1570 $after .= "\t.end $ident\n";
1573 # Alphas: The heroic Simon Marlow found a bug in the Digital UNIX
1574 # assembler (!) wherein .quad constants inside .text sections are
1575 # first narrowed to 32 bits then sign-extended back to 64 bits.
1576 # This obviously screws up our 64-bit bitmaps, so we work around
1577 # the bug by replacing .quad with .align 3 + .long + .long [ccshan]
1578 if ( $TargetPlatform =~ /^alpha-/ ) {
1580 if (/^\s*\.quad\s+([-+0-9].*\S)\s*$/ && length $1 >= 10) {
1581 local ($number) = $1;
1582 if ($number =~ /^([-+])?(0x?)?([0-9]+)$/) {
1583 local ($sign, $base, $digits) = ($1, $2, $3);
1584 $base = (10, 8, 16)[length $base];
1585 local ($hi, $lo) = (0, 0);
1586 foreach $i (split(//, $digits)) {
1587 $j = $lo * $base + $i;
1588 $lo = $j % 4294967296;
1589 $hi = $hi * $base + ($j - $lo) / 4294967296;
1591 ($hi, $lo) = (4294967295 - $hi, 4294967296 - $lo)
1593 $_ = "\t.align 3\n\t.long $lo\n\t.long $hi\n";
1594 # printf STDERR "TURNING %s into 0x %08x %08x\n", $number, $hi, $lo;
1596 print STDERR "Cannot handle \".quad $number\" in info table\n";
1604 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1605 . join("\n", @words) . "\n"
1608 # print STDERR "before=$before\n";
1609 # print STDERR "label=$label\n";
1610 # print STDERR "words=",(reverse @words),"\n";
1611 # print STDERR "after=$after\n";
1617 The HP is a major nuisance. The threaded code mangler moved info
1618 tables from data space to code space, but unthreaded code in the RTS
1619 still has references to info tables in data space. Since the HP
1620 linker is very precise about where symbols live, we need to patch the
1621 references in the unthreaded RTS as well.
1624 sub mini_mangle_asm_hppa {
1625 local($in_asmf, $out_asmf) = @_;
1627 open(INASM, "< $in_asmf")
1628 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1629 open(OUTASM,"> $out_asmf")
1630 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1633 s/_info,DATA/_info,CODE/; # Move _info references to code space
1639 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1640 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1646 sub tidy_up_and_die {
1647 local($return_val, $msg) = @_;
1649 exit (($return_val == 0) ? 0 : 1);