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+\d+(,0x90)?|\.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+\d+(,\s*0x90)?|\.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.*|\.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';
535 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
536 # duplicated from the bottom
540 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
541 # Alphas: Local labels not to be confused with new chunks
543 } elsif ( $TargetPlatform =~ /-darwin/
544 && /^\t\.subsections_via_symbols/) {
545 # Don't allow Apple's linker to do any dead-stripping of symbols
546 # in this file, because it will mess up info-tables in mangled
549 # NB: all the rest start with a non-space
551 } elsif ( $TargetPlatform =~ /^mips-/
552 && /^\d+:/ ) { # a funny-looking very-local label
555 } elsif ( /$T_CONST_LBL/o ) {
557 $chkcat[$i] = 'literal';
560 } elsif ( /^${T_US}__stg_split_marker(\d*)${T_POST_LBL}$/o ) {
562 $chkcat[$i] = 'splitmarker';
565 } elsif ( /^${T_US}([A-Za-z0-9_]+)_info${T_POST_LBL}$/o ) {
568 $chkcat[$i] = 'infotbl';
569 $chksymb[$i] = $symb;
571 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
573 $infochk{$symb} = $i;
575 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(entry|ret)${T_POST_LBL}$/o ) {
577 $chkcat[$i] = 'entry';
582 } elsif ( /^${T_US}([A-Za-z0-9_]+)_closure${T_POST_LBL}$/o ) {
584 $chkcat[$i] = 'closure';
587 $closurechk{$1} = $i;
589 } elsif ( /^${T_US}([A-Za-z0-9_]+)_srt${T_POST_LBL}$/o ) {
596 } elsif ( /^${T_US}([A-Za-z0-9_]+)_ct${T_POST_LBL}$/o ) {
598 $chkcat[$i] = 'data';
601 } elsif ( /^${T_US}(stg_ap_stack_entries|stg_stack_save_entries|stg_arg_bitmaps)${T_POST_LBL}$/o ) {
603 $chkcat[$i] = 'data';
606 } elsif ( /^(${T_US}__gnu_compiled_c|gcc2_compiled\.)${T_POST_LBL}/o ) {
609 } elsif ( /^${T_US}[A-Za-z0-9_]+\.\d+${T_POST_LBL}$/o
610 || /^${T_US}.*_CAT${T_POST_LBL}$/o # PROF: _entryname_CAT
611 || /^${T_US}.*_done${T_POST_LBL}$/o # PROF: _module_done
612 || /^${T_US}_module_registered${T_POST_LBL}$/o # PROF: _module_registered
615 $chkcat[$i] = 'data';
618 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
623 } elsif ( /^${T_US}([A-Za-z0-9_]+)_cc(s)?${T_POST_LBL}$/o ) {
624 # all CC_ symbols go in the data section...
626 $chkcat[$i] = 'data';
629 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(alt|dflt)${T_POST_LBL}$/o ) {
631 $chkcat[$i] = 'misc';
633 } elsif ( /^${T_US}([A-Za-z0-9_]+)_vtbl${T_POST_LBL}$/o ) {
635 $chkcat[$i] = 'vector';
640 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
641 && /^[A-Za-z0-9][A-Za-z0-9_]*:/ ) {
642 # Some Solaris system headers contain function definitions (as
643 # opposed to mere prototypes), which end up in the .hc file when
644 # a Haskell module foreign imports the corresponding system
645 # functions (most notably stat()). We put them into the text
646 # segment. Note that this currently does not extend to function
647 # names starting with an underscore.
650 $chkcat[$i] = 'misc';
653 } elsif ( /^${T_US}[A-Za-z0-9_]/o
654 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
656 && ( $TargetPlatform !~ /^powerpc64/ # we need to avoid local labels in this case
657 || ! /^\.L\d+:$/ ) ) {
664 /^${T_US}stg_.*${T_POST_LBL}$/o # RTS internals
665 || /^${T_US}__stg_.*${T_POST_LBL}$/o # more RTS internals
666 || /^${T_US}__fexp_.*${T_POST_LBL}$/o # foreign export
667 || /^${T_US}.*_slow${T_POST_LBL}$/o # slow entry
668 || /^${T_US}__stginit.*${T_POST_LBL}$/o # __stginit<module>
669 || /^${T_US}.*_btm${T_POST_LBL}$/o # large bitmaps
670 || /^${T_US}.*_fast${T_POST_LBL}$/o # primops
671 || /^_uname:/o # x86/Solaris2
674 $chkcat[$i] = 'misc';
676 /^${T_US}.*_srtd${T_POST_LBL}$/o # large bitmaps
677 || /^${T_US}.*_closure_tbl${T_POST_LBL}$/o # closure tables
680 $chkcat[$i] = 'relrodata';
683 print STDERR "Warning: retaining unknown function \`$thing' in output from C compiler\n";
684 $chkcat[$i] = 'unknown';
687 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && (
689 || /^\.section __TEXT,__picsymbol_stub1,.*/
690 || /^\.section __TEXT,__picsymbolstub1,.*/
692 || /^\.section __TEXT,__symbol_stub1,.*/
693 || /^\.section __TEXT,__symbolstub1,.*/
694 || /^\.lazy_symbol_pointer/
695 || /^\.non_lazy_symbol_pointer/ ))
698 $chkcat[$i] = 'dyld';
700 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.data/ && $chkcat[$i] eq 'dyld')
701 { # non_lazy_symbol_ptrs that point to local symbols
703 $chkcat[$i] = 'dyld';
705 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ && /^\.LCTOC1 = /o ) {
706 # PowerPC Linux's large-model PIC (-fPIC) generates a gobal offset
707 # table "by hand". Be sure to copy it over.
708 # Note that this label and all entries in the table should actually
709 # go into the .got2 section, but it isn't easy to distinguish them
710 # from other constant literals (.LC\d+), so we just put everything
713 $chkcat[$i] = 'literal';
714 $chksymb[$i] = 'LCTOC1';
715 } else { # simple line (duplicated at the top)
720 $numchks = $#chk + 1;
722 # open CHUNKS, ">/tmp/chunks1" or die "Cannot open /tmp/chunks1: $!\n";
723 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
726 # the division into chunks is imperfect;
727 # we throw some things over the fence into the next
730 # also, there are things we would like to know
731 # about the whole module before we start spitting
734 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
735 local($FIRST_TOSSABLE ) = ($TargetPlatform =~ /^(hppa|mips-)/) ? 1 : 0;
737 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
739 # Alphas: NB: we start meddling at chunk 1, not chunk 0
740 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
741 # spits a ".quad 0" in after the very first ".rdata"; we
742 # detect this special case (tossing the ".quad 0")!
743 local($magic_rdata_seen) = 0;
745 # HPPAs, MIPSen: also start medding at chunk 1
747 for ($i = $FIRST_TOSSABLE; $i < $numchks; $i++) {
748 $c = $chk[$i]; # convenience copy
750 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
752 # toss all prologue stuff; HPPA is pretty weird
754 $c = &hppa_mash_prologue($c) if $TargetPlatform =~ /^hppa-/;
756 # be slightly paranoid to make sure there's
757 # nothing surprising in there
758 if ( $c =~ /--- BEGIN ---/ ) {
759 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
761 # remove junk whitespace around the split point
765 if ($TargetPlatform =~ /^i386-/) {
766 $p =~ s/^\tpushl\s+\%edi\n//;
767 $p =~ s/^\tpushl\s+\%esi\n//;
768 $p =~ s/^\tpushl\s+\%ebx\n//;
769 $p =~ s/^\tmovl\s+\%esi,\s*\d*\(\%esp\)\n//;
770 $p =~ s/^\tmovl\s+\%edi,\s*\d*\(\%esp\)\n//;
771 $p =~ s/^\tsubl\s+\$\d+,\s*\%esp\n//;
772 $p =~ s/^\tmovl\s+\$\d+,\s*\%eax\n\tcall\s+__alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
774 # GCC 3.1 is in the habit of adding spurious writes to the
775 # stack in the prologue. Just to be on the safe side,
776 # chuck these over the fence into the main code.
777 while ($p =~ /^\tmovl\s+\$\d+,\s*\d*\(\%esp\)\n/) {
778 # print "Spurious instruction: $&";
783 } elsif ($TargetPlatform =~ /^x86_64-/) {
784 $p =~ s/^\tpushq\s+\%r(bx|bp|12|13|14)\n//g;
785 $p =~ s/^\tmovq\s+\%r(bx|bp|12|13|14),\s*\d*\(\%rsp\)\n//g;
786 $p =~ s/^\tsubq\s+\$\d+,\s*\%rsp\n//;
788 } elsif ($TargetPlatform =~ /^ia64-/) {
789 $p =~ s/^\t\.prologue .*\n//;
790 $p =~ s/^\t\.save ar\.pfs, r\d+\n\talloc r\d+ = ar\.pfs, 0, 3[12], \d+, 0\n//;
791 $p =~ s/^\t\.fframe \d+\n\tadds r12 = -\d+, r12\n//;
792 $p =~ s/^\t\.save rp, r\d+\n\tmov r\d+ = b0\n//;
793 $p =~ s/^\t\.(mii|mmi)\n//g; # bundling is no longer sensible
794 $p =~ s/^\t;;\n//g; # discard stops
795 $p =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
797 # GCC 3.3 saves r1 in the prologue, move this to the body
798 if ($p =~ /^\tmov r\d+ = r1\n/) {
802 } elsif ($TargetPlatform =~ /^m68k-/) {
803 $p =~ s/^\tlink a6,#-?\d.*\n//;
804 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
805 # The above showed up in the asm code,
806 # so I added it here.
807 # I hope it's correct.
809 $p =~ s/^\tmovel d2,sp\@-\n//;
810 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
811 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
812 } elsif ($TargetPlatform =~ /^mips-/) {
813 # the .frame/.mask/.fmask that we use is the same
814 # as that produced by GCC for miniInterpret; this
815 # gives GDB some chance of figuring out what happened
816 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
817 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
818 $p =~ s/^\t\.(mask|fmask).*\n//g;
819 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
820 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
821 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
822 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
823 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
824 $p =~ s/__FRAME__/$FRAME/;
825 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
827 $pcrel_label =~ s/(.|\n)*^(\"?L\d+\$pb\"?):\n(.|\n)*/$2/ or $pcrel_label = "";
829 $p =~ s/^\tmflr r0\n//;
830 $p =~ s/^\tbl saveFP # f\d+\n//;
831 $p =~ s/^\tbl saveFP ; save f\d+-f\d+\n//;
832 $p =~ s/^\"?L\d+\$pb\"?:\n//;
833 $p =~ s/^\tstmw r\d+,-\d+\(r1\)\n//;
834 $p =~ s/^\tstfd f\d+,-\d+\(r1\)\n//g;
835 $p =~ s/^\tstw r0,\d+\(r1\)\n//g;
836 $p =~ s/^\tstwu r1,-\d+\(r1\)\n//;
837 $p =~ s/^\tstw r\d+,-\d+\(r1\)\n//g;
838 $p =~ s/^\tbcl 20,31,L\d+\$pb\n//;
839 $p =~ s/^L\d+\$pb:\n//;
840 $p =~ s/^\tmflr r31\n//;
842 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
843 # under some circumstances, only when generating position dependent code.
844 # I have no idea why, and I don't think it is necessary, so let's toss it.
845 $p =~ s/^\tli r\d+,0\n//g;
846 $p =~ s/^\tstw r\d+,\d+\(r1\)\n//g;
847 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/) {
848 $p =~ s/^\tmflr 0\n//;
849 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
850 $p =~ s/^\tstfd \d+,\d+\(1\)\n//g;
851 $p =~ s/^\tstw r0,8\(1\)\n//;
852 $p =~ s/^\tstwu 1,-\d+\(1\)\n//;
853 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
855 # GCC's "large-model" PIC (-fPIC)
857 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
859 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
860 $p =~ s/^.LCF\d+:\n//;
861 $p =~ s/^\tmflr 30\n//;
862 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
863 $p =~ s/^\tadd 30,0,30\n//;
865 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
866 # under some circumstances, only when generating position dependent code.
867 # I have no idea why, and I don't think it is necessary, so let's toss it.
868 $p =~ s/^\tli \d+,0\n//g;
869 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
870 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
871 $p =~ s/^\tmr 31,1\n//;
872 $p =~ s/^\tmflr 0\n//;
873 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
874 $p =~ s/^\tstfd \d+,-?\d+\(1\)\n//g;
875 $p =~ s/^\tstd r0,8\(1\)\n//;
876 $p =~ s/^\tstdu 1,-\d+\(1\)\n//;
877 $p =~ s/^\tstd \d+,-?\d+\(1\)\n//g;
879 # GCC's "large-model" PIC (-fPIC)
881 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
883 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
884 $p =~ s/^.LCF\d+:\n//;
885 $p =~ s/^\tmflr 30\n//;
886 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
887 $p =~ s/^\tadd 30,0,30\n//;
889 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
890 # under some circumstances, only when generating position dependent code.
891 # I have no idea why, and I don't think it is necessary, so let's toss it.
892 $p =~ s/^\tli \d+,0\n//g;
893 $p =~ s/^\tstd \d+,\d+\(1\)\n//g;
895 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
898 # HWL HACK: dont die, just print a warning
899 #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/;
900 die "Prologue junk?: $p\n" if $p =~ /^\s+[^\s\.]/;
902 # For PIC, we want to keep part of the prologue
903 if ($TargetPlatform =~ /^powerpc-apple-.*/ && $pcrel_label ne "") {
904 # Darwin: load the current instruction pointer into register r31
905 $p .= "bcl 20,31,$pcrel_label\n";
906 $p .= "$pcrel_label:\n";
907 $p .= "\tmflr r31\n";
908 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/ && $pcrel_label ne "") {
909 # Linux: load the GOT pointer into register 30
910 $p .= "\tbcl 20,31,.LCF$pcrel_label\n";
911 $p .= ".LCF$pcrel_label:\n";
913 $p .= "\tlwz 0,.LCL$pcrel_label-.LCF$pcrel_label(30)\n";
914 $p .= "\tadd 30,0,30\n";
917 # glue together what's left
922 if ( $TargetPlatform =~ /^mips-/ ) {
923 # MIPS: first, this basic sequence may occur "--- END ---" or not
924 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
927 # toss all epilogue stuff; again, paranoidly
928 if ( $c =~ /--- END ---/ ) {
929 if (($r, $e) = split(/--- END ---/, $c)) {
930 if ($TargetPlatform =~ /^i386-/) {
932 $e =~ s/^\tpopl\s+\%edi\n//;
933 $e =~ s/^\tpopl\s+\%esi\n//;
934 $e =~ s/^\tpopl\s+\%edx\n//;
935 $e =~ s/^\tpopl\s+\%ecx\n//;
936 $e =~ s/^\taddl\s+\$\d+,\s*\%esp\n//;
937 $e =~ s/^\tsubl\s+\$-\d+,\s*\%esp\n//;
938 } elsif ($TargetPlatform =~ /^ia64-/) {
939 $e =~ s/^\tmov ar\.pfs = r\d+\n//;
940 $e =~ s/^\tmov b0 = r\d+\n//;
941 $e =~ s/^\t\.restore sp\n\tadds r12 = \d+, r12\n//;
942 $e =~ s/^\tbr\.ret\.sptk\.many b0\n//;
943 $e =~ s/^\t\.(mii|mmi|mib)\n//g; # bundling is no longer sensible
944 $e =~ s/^\t;;\n//g; # discard stops - stop at end of body is sufficient
945 $e =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
946 } elsif ($TargetPlatform =~ /^m68k-/) {
947 $e =~ s/^\tunlk a6\n//;
949 } elsif ($TargetPlatform =~ /^mips-/) {
950 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
951 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
952 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
953 $e =~ s/^\tj\t\$31\n//;
954 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
955 $e =~ s/^\taddi r1,r1,\d+\n//;
956 $e =~ s/^\tlwz r\d+,\d+\(r1\)\n//;
957 $e =~ s/^\tlmw r\d+,-\d+\(r1\)\n//;
958 $e =~ s/^\tmtlr r0\n//;
960 $e =~ s/^\tb restFP ;.*\n//;
961 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
962 $e =~ s/^\tmr 3,0\n//;
963 $e =~ s/^\taddi 1,1,\d+\n//;
964 $e =~ s/^\tld 0,16\(1\)\n//;
965 $e =~ s/^\tmtlr 0\n//;
967 # callee-save registers
968 $e =~ s/^\tld \d+,-?\d+\(1\)\n//g;
969 $e =~ s/^\tlfd \d+,-?\d+\(1\)\n//g;
971 # get rid of the debug junk along with the blr
972 $e =~ s/^\tblr\n\t.long .*\n\t.byte .*\n//;
974 # incase we missed it with the last one get the blr alone
977 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
980 print STDERR "WARNING: Epilogue junk?: $e\n" if $e =~ /^\t\s*[^\.\s\n]/;
982 # glue together what's left
984 $c =~ s/\n\t\n/\n/; # junk blank line
988 # On SPARCs, we don't do --- BEGIN/END ---, we just
989 # toss the register-windowing save/restore/ret* instructions
990 # directly unless they've been generated by function definitions in header
992 if ( $TargetPlatform =~ /^sparc-/ ) {
993 if ( ! ( $TargetPlatform =~ /solaris2$/ && $chkcat[$i] eq 'unknown' )) {
994 $c =~ s/^\t(save.*|restore.*|ret|retl)\n//g;
996 # throw away PROLOGUE comments
997 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
1000 # On Alphas, the prologue mangling is done a little later (below)
1002 # toss all calls to __DISCARD__
1003 $c =~ s/^\t(call|jbsr|jal)\s+${T_US}__DISCARD__\n//go;
1004 $c =~ s/^\tjsr\s+\$26\s*,\s*${T_US}__DISCARD__\n//go if $TargetPlatform =~ /^alpha-/;
1005 $c =~ s/^\tbl\s+L___DISCARD__\$stub\n//go if $TargetPlatform =~ /^powerpc-apple-.*/;
1006 $c =~ s/^\tbl\s+__DISCARD__(\@plt)?\n//go if $TargetPlatform =~ /^powerpc-.*-linux/;
1007 $c =~ s/^\tbl\s+\.__DISCARD__\n\s+nop\n//go if $TargetPlatform =~ /^powerpc64-.*-linux/;
1009 # IA64: mangle tailcalls into jumps here
1010 if ($TargetPlatform =~ /^ia64-/) {
1011 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/) {
1012 # Eek, the gcc optimiser is getting smarter... if we see a jump to the --- TAILCALL ---
1013 # marker then we reapply the substitution at the source sites
1014 $c =~ s/^\tbr \.L$2\n/\t--- TAILCALL ---\n/g if ($2);
1018 # MIPS: that may leave some gratuitous asm macros around
1019 # (no harm done; but we get rid of them to be tidier)
1020 $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/
1021 if $TargetPlatform =~ /^mips-/;
1023 # toss stack adjustment after DoSparks
1024 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
1025 if $TargetPlatform =~ /^m68k-/; # this looks old...
1027 if ( $TargetPlatform =~ /^alpha-/ &&
1028 ! $magic_rdata_seen &&
1029 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
1030 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
1031 $magic_rdata_seen = 1;
1034 # pick some end-things and move them to the next chunk
1036 # pin a funny end-thing on (for easier matching):
1037 $c .= 'FUNNY#END#THING';
1039 while ( $c =~ /${T_MOVE_DIRVS}FUNNY#END#THING/o ) {
1043 # on x86 we try not to copy any directives into a literal
1044 # chunk, rather we keep looking for the next real chunk. This
1045 # is because we get things like
1047 # .globl blah_closure
1053 if ( $TargetPlatform =~ /^(i386|sparc|powerpc)/ && $to_move =~ /${T_COPY_DIRVS}/ ) {
1055 while ( $j < $numchks && $chk[$j] =~ /$T_CONST_LBL/) {
1058 if ( $j < $numchks ) {
1059 $chk[$j] = $to_move . $chk[$j];
1063 elsif ( $i < ($numchks - 1)
1064 && ( $to_move =~ /${T_COPY_DIRVS}/
1065 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
1066 $chk[$i + 1] = $to_move . $chk[$i + 1];
1067 # otherwise they're tossed
1070 $c =~ s/${T_MOVE_DIRVS}FUNNY#END#THING/FUNNY#END#THING/o;
1073 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
1075 # toss all prologue stuff, except for loading gp, and the ..ng address
1076 unless ($c =~ /\.ent.*\n\$.*\.\.ng:/) {
1077 if (($p, $r) = split(/^\t\.prologue/, $c)) {
1078 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
1079 $keep =~ s/^\t\.frame.*\n/\t.frame \$30,0,\$26,0\n/;
1080 $keep =~ s/^\t\.(mask|fmask).*\n//g;
1081 $c = $keep . "..ng:\n";
1083 print STDERR "malformed code block ($ent)?\n"
1086 $c .= "\t.prologue" . $r;
1090 $c =~ s/FUNNY#END#THING//;
1092 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
1094 $chk[$i] = $c; # update w/ convenience copy
1097 # open CHUNKS, ">/tmp/chunks2" or die "Cannot open /tmp/chunks2: $!\n";
1098 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
1101 if ( $TargetPlatform =~ /^alpha-/ ) {
1102 # print out the header stuff first
1103 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
1104 print OUTASM $chk[0];
1106 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1107 print OUTASM $chk[0];
1109 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1110 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
1112 # get rid of horrible "<dollar>Revision: .*$" strings
1113 local(@lines0) = split(/\n/, $chk[0]);
1115 while ( $z <= $#lines0 ) {
1116 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
1119 while ( $z <= $#lines0 ) {
1121 last if $lines0[$z] =~ /[,\t]0x0$/;
1127 $chk[0] = join("\n", @lines0);
1128 $chk[0] =~ s/\n\n+/\n/;
1129 print OUTASM $chk[0];
1132 # print out all the literal strings next
1133 for ($i = 0; $i < $numchks; $i++) {
1134 if ( $chkcat[$i] eq 'literal' ) {
1135 print OUTASM $T_HDR_literal, $chk[$i];
1136 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
1138 $chkcat[$i] = 'DONE ALREADY';
1142 # on the HPPA, print out all the bss next
1143 if ( $TargetPlatform =~ /^hppa/ ) {
1144 for ($i = 1; $i < $numchks; $i++) {
1145 if ( $chkcat[$i] eq 'bss' ) {
1146 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
1147 print OUTASM $chk[$i];
1149 $chkcat[$i] = 'DONE ALREADY';
1154 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
1155 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
1157 next if $chkcat[$i] eq 'DONE ALREADY';
1159 if ( $chkcat[$i] eq 'misc' || $chkcat[$i] eq 'unknown' ) {
1160 if ($chk[$i] ne '') {
1161 print OUTASM $T_HDR_misc;
1162 &print_doctored($chk[$i], 0);
1165 } elsif ( $chkcat[$i] eq 'toss' ) {
1166 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
1168 } elsif ( $chkcat[$i] eq 'data' ) {
1169 if ($chk[$i] ne '') {
1170 print OUTASM $T_HDR_data;
1171 print OUTASM $chk[$i];
1174 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
1175 # we can just re-constitute this one...
1176 # NB: we emit _three_ underscores no matter what,
1177 # so ghc-split doesn't have to care.
1178 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
1180 } elsif ( $chkcat[$i] eq 'closure'
1181 || $chkcat[$i] eq 'srt'
1182 || $chkcat[$i] eq 'infotbl'
1183 || $chkcat[$i] eq 'entry') { # do them in that order
1184 $symb = $chksymb[$i];
1187 if ( defined($closurechk{$symb}) ) {
1188 print OUTASM $T_HDR_closure;
1189 print OUTASM $chk[$closurechk{$symb}];
1190 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
1194 if ( defined($srtchk{$symb}) ) {
1195 print OUTASM $T_HDR_relrodata;
1196 print OUTASM $chk[$srtchk{$symb}];
1197 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
1201 if ( defined($infochk{$symb}) ) {
1203 print OUTASM $T_HDR_info;
1204 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
1206 # entry code will be put here!
1208 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
1212 if ( defined($entrychk{$symb}) ) {
1214 $c = $chk[$entrychk{$symb}];
1216 # If this is an entry point with an info table,
1217 # eliminate the entry symbol and all directives involving it.
1218 if (defined($infochk{$symb}) && $TargetPlatform !~ /^ia64-/) {
1219 $c =~ s/^.*$symb_(entry|ret)${T_POST_LBL}\n//;
1220 $c =~ s/^\s*\..*$symb.*\n//g;
1223 print OUTASM $T_HDR_entry;
1225 &print_doctored($c, 1); # NB: the 1!!!
1227 $chkcat[$entrychk{$symb}] = 'DONE ALREADY';
1230 } elsif ( $chkcat[$i] eq 'vector' ) {
1231 $symb = $chksymb[$i];
1234 if ( defined($vectorchk{$symb}) ) {
1235 print OUTASM $T_HDR_vector;
1236 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1238 # direct return code will be put here!
1239 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1241 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1242 # Alphas: the commented nop is for the splitter, to ensure
1243 # that no module ends with a label as the very last
1244 # thing. (The linker will adjust the label to point
1245 # to the first code word of the next module linked in,
1246 # even if alignment constraints cause the label to move!)
1248 print OUTASM "\t# nop\n";
1251 } elsif ( $chkcat[$i] eq 'rodata' ) {
1252 print OUTASM $T_HDR_rodata;
1253 print OUTASM $chk[$i];
1254 $chkcat[$i] = 'DONE ALREADY';
1255 } elsif ( $chkcat[$i] eq 'relrodata' ) {
1256 print OUTASM $T_HDR_relrodata;
1257 print OUTASM $chk[$i];
1258 $chkcat[$i] = 'DONE ALREADY';
1259 } elsif ( $chkcat[$i] eq 'toc' ) {
1260 # silly optimisation to print tocs, since they come in groups...
1261 print OUTASM $T_HDR_toc;
1263 while ($chkcat[$j] eq 'toc')
1264 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1267 print OUTASM $chk[$j];
1269 $chkcat[$j] = 'DONE ALREADY';
1273 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && $chkcat[$i] eq 'dyld' ) {
1274 # powerpc-apple: dynamic linker stubs
1275 if($chk[$i] !~ /\.indirect_symbol ___DISCARD__/)
1276 { # print them out unchanged, but remove the stubs for __DISCARD__
1277 print OUTASM $chk[$i];
1280 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1284 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1287 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1288 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1293 sub hppa_mash_prologue { # OK, epilogue, too
1296 # toss all prologue stuff
1297 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1299 # Lie about our .CALLINFO
1300 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1307 # toss all epilogue stuff
1308 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1310 # Sorry; we moved the _info stuff to the code segment.
1311 s/_info,DATA/_info,CODE/g;
1318 sub print_doctored {
1319 local($_, $need_fallthru_patch) = @_;
1321 if ( $TargetPlatform !~ /^i386-/
1322 || ! /^\t[a-z]/ # no instructions in here, apparently
1323 || /^${T_US}__stginit_[A-Za-z0-9_]+${T_POST_LBL}/) {
1327 # OK, must do some x86 **HACKING**
1329 local($entry_patch) = '';
1330 local($exit_patch) = '';
1332 # gotta watch out for weird instructions that
1333 # invisibly smash various regs:
1334 # rep* %ecx used for counting
1335 # scas* %edi used for destination index
1336 # cmps* %e[sd]i used for indices
1337 # loop* %ecx used for counting
1342 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1344 # * GCC used an "STG reg" for its own purposes
1346 # * some secret uses of machine reg, requiring STG reg
1347 # to be saved/restored
1349 # The most dangerous "GCC uses" of an "STG reg" are when
1350 # the reg holds the target of a jmp -- it's tricky to
1351 # insert the patch-up code before we get to the target!
1352 # So here we change the jmps:
1354 # --------------------------------------------------------
1355 # it can happen that we have jumps of the form...
1356 # jmp *<something involving %esp>
1358 # jmp <something involving another naughty register...>
1360 # a reasonably-common case is:
1362 # movl $_blah,<bad-reg>
1366 s/^\tmovl\s+\$${T_US}(.*),\s*(\%e[abcd]x)\n\tjmp\s+\*\2/\tjmp $T_US$1/g;
1368 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1369 s/^\tmovl\s+(.*),\s*\%esi\n\tjmp\s+\*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1370 s/^\tjmp\s+\*(.*\(.*\%esi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1371 s/^\tjmp\s+\*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1372 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1373 if /(jmp|call)\s+.*\%esi/;
1375 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1376 s/^\tmovl\s+(.*),\s*\%edi\n\tjmp\s+\*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1377 s/^\tjmp\s+\*(.*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1378 s/^\tjmp\s+\*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1379 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1380 if /(jmp|call)\s+.*\%edi/;
1383 # OK, now we can decide what our patch-up code is going to
1386 # Offsets into register table - you'd better update these magic
1387 # numbers should you change its contents!
1388 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1389 local($OFFSET_Hp)=88;
1391 # Note funky ".=" stuff; we're *adding* to these _patch guys
1392 if ( $StolenX86Regs <= 2
1393 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1394 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1395 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1397 # nothing for call_{entry,exit} because %esi is callee-save
1399 if ( $StolenX86Regs <= 3
1400 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1401 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1402 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1404 # nothing for call_{entry,exit} because %edi is callee-save
1407 # --------------------------------------------------------
1408 # next, here we go with non-%esp patching!
1410 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1412 # Before calling GC we must set up the exit condition before the call
1413 # and entry condition when we come back
1415 # fix _all_ non-local jumps:
1417 s/^\tjmp\s+\*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1418 s/^\tjmp\s+${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1420 s/^(\tjmp\s+.*\n)/$exit_patch$1/g; # here's the fix...
1422 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1423 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1425 if ($StolenX86Regs == 2 ) {
1426 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1427 if /^\t(jmp|call)\s+.*\%e(si|di)/;
1428 } elsif ($StolenX86Regs == 3 ) {
1429 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1430 if /^\t(jmp|call)\s+.*\%edi/;
1433 # --------------------------------------------------------
1434 # that's it -- print it
1436 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1440 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1441 print OUTASM $exit_patch;
1442 # ToDo: make it not print if there is a "jmp" at the end
1448 sub init_FUNNY_THINGS {
1449 %KNOWN_FUNNY_THING = (
1451 # "${T_US}stg_.*{T_POST_LBL}", 1,
1456 The following table reversal is used for both info tables and return
1457 vectors. In both cases, we remove the first entry from the table,
1458 reverse the table, put the label at the end, and paste some code
1459 (that which is normally referred to by the first entry in the table)
1460 right after the table itself. (The code pasting is done elsewhere.)
1464 local($symb, $tbl, $discard1) = @_;
1466 return ($tbl) if ($TargetPlatform =~ /^ia64-/);
1468 local($before) = '';
1470 local(@imports) = (); # hppa only
1473 local(@lines) = split(/\n/, $tbl);
1476 # Deal with the header...
1477 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1478 $label .= $lines[$i] . "\n",
1479 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info${T_POST_LBL}$/o
1480 || $lines[$i] =~ /${T_DOT_GLOBAL}/o
1481 || $lines[$i] =~ /^${T_US}\S+_vtbl${T_POST_LBL}$/o;
1483 $before .= $lines[$i] . "\n"; # otherwise...
1487 $infoname =~ s/(.|\n)*^([A-Za-z0-9_]+_info)${T_POST_LBL}$(.|\n)*/\2/;
1489 # Grab the table data...
1490 if ( $TargetPlatform !~ /^hppa/ ) {
1491 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1493 # Convert addresses of SRTs, slow entrypoints and large bitmaps
1494 # to offsets (relative to the info label),
1495 # in order to support position independent code.
1496 $line =~ s/$infoname/0/
1497 || $line =~ s/([A-Za-z0-9_]+_srtd)$/\1 - $infoname/
1498 || $line =~ s/([A-Za-z0-9_]+_srt(\+\d+)?)$/\1 - $infoname/
1499 || $line =~ s/([A-Za-z0-9_]+_slow)$/\1 - $infoname/
1500 || $line =~ s/([A-Za-z0-9_]+_btm)$/\1 - $infoname/
1501 || $line =~ s/([A-Za-z0-9_]+_alt)$/\1 - $infoname/
1502 || $line =~ s/([A-Za-z0-9_]+_dflt)$/\1 - $infoname/
1503 || $line =~ s/([A-Za-z0-9_]+_ret)$/\1 - $infoname/;
1504 push(@words, $line);
1506 } else { # hppa weirdness
1507 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+(${T_DOT_WORD}|\.IMPORT)/; $i++) {
1508 # FIXME: the RTS now expects offsets instead of addresses
1509 # for all labels in info tables.
1510 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1511 push(@imports, $lines[$i]);
1513 # We don't use HP's ``function pointers''
1514 # We just use labels in code space, like normal people
1515 $lines[$i] =~ s/P%//;
1516 push(@words, $lines[$i]);
1521 # Now throw away any initial zero word from the table. This is a hack
1522 # that lets us reduce the size of info tables when the SRT field is not
1523 # needed: see comments StgFunInfoTable in InfoTables.h.
1525 # The .zero business is for Linux/ELF.
1526 # The .skip business is for Sparc/Solaris/ELF.
1527 # The .blockz business is for HPPA.
1529 # if ($words[0] =~ /^\t?(${T_DOT_WORD}\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1534 for (; $i <= $#lines; $i++) {
1535 $after .= $lines[$i] . "\n";
1538 # Alphas: If we have anonymous text (not part of a procedure), the
1539 # linker may complain about missing exception information. Bleh.
1540 # To suppress this, we place a .ent/.end pair around the code.
1541 # At the same time, we have to be careful and not enclose any leading
1542 # .file/.loc directives.
1543 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1544 local ($ident) = $1;
1545 $before =~ s/^((\s*\.(file|loc)\s+[^\n]*\n)*)/$1\t.ent $ident\n/;
1546 $after .= "\t.end $ident\n";
1549 # Alphas: The heroic Simon Marlow found a bug in the Digital UNIX
1550 # assembler (!) wherein .quad constants inside .text sections are
1551 # first narrowed to 32 bits then sign-extended back to 64 bits.
1552 # This obviously screws up our 64-bit bitmaps, so we work around
1553 # the bug by replacing .quad with .align 3 + .long + .long [ccshan]
1554 if ( $TargetPlatform =~ /^alpha-/ ) {
1556 if (/^\s*\.quad\s+([-+0-9].*\S)\s*$/ && length $1 >= 10) {
1557 local ($number) = $1;
1558 if ($number =~ /^([-+])?(0x?)?([0-9]+)$/) {
1559 local ($sign, $base, $digits) = ($1, $2, $3);
1560 $base = (10, 8, 16)[length $base];
1561 local ($hi, $lo) = (0, 0);
1562 foreach $i (split(//, $digits)) {
1563 $j = $lo * $base + $i;
1564 $lo = $j % 4294967296;
1565 $hi = $hi * $base + ($j - $lo) / 4294967296;
1567 ($hi, $lo) = (4294967295 - $hi, 4294967296 - $lo)
1569 $_ = "\t.align 3\n\t.long $lo\n\t.long $hi\n";
1570 # printf STDERR "TURNING %s into 0x %08x %08x\n", $number, $hi, $lo;
1572 print STDERR "Cannot handle \".quad $number\" in info table\n";
1580 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1581 . join("\n", @words) . "\n"
1584 # print STDERR "before=$before\n";
1585 # print STDERR "label=$label\n";
1586 # print STDERR "words=",(reverse @words),"\n";
1587 # print STDERR "after=$after\n";
1593 The HP is a major nuisance. The threaded code mangler moved info
1594 tables from data space to code space, but unthreaded code in the RTS
1595 still has references to info tables in data space. Since the HP
1596 linker is very precise about where symbols live, we need to patch the
1597 references in the unthreaded RTS as well.
1600 sub mini_mangle_asm_hppa {
1601 local($in_asmf, $out_asmf) = @_;
1603 open(INASM, "< $in_asmf")
1604 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1605 open(OUTASM,"> $out_asmf")
1606 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1609 s/_info,DATA/_info,CODE/; # Move _info references to code space
1615 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1616 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1622 sub tidy_up_and_die {
1623 local($return_val, $msg) = @_;
1625 exit (($return_val == 0) ? 0 : 1);