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_data\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_data\n\t\.align 2\n";
300 $T_HDR_closure = "\t\.const_data\n\t\.align 2\n";
301 $T_HDR_info = "\t\.text\n\t\.align 2\n";
302 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
303 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
305 #--------------------------------------------------------#
306 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ ) {
308 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
309 $T_US = ''; # _ if symbols have an underscore on the front
310 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
311 $T_CONST_LBL = '^\.LC\d+:'; # regexp for what such a lbl looks like
314 $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)';
315 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
317 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
318 $T_DOT_GLOBAL = '\.globl';
319 $T_HDR_toc = "\.toc\n";
320 $T_HDR_literal = "\t\.section\t.rodata\n\t\.align 2\n";
321 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
322 $T_HDR_data = "\t\.data\n\t\.align 2\n";
323 $T_HDR_rodata = "\t\.section\t.rodata\n\t\.align 2\n";
324 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
325 $T_HDR_info = "\t\.text\n\t\.align 2\n";
326 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
327 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
329 #--------------------------------------------------------#
330 } elsif ( $TargetPlatform =~ /^powerpc64-.*-linux/ ) {
332 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
333 $T_US = '\.'; # _ if symbols have an underscore on the front
334 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
335 $T_CONST_LBL = '^\.LC\d+:'; # regexp for what such a lbl looks like
338 $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)';
339 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
341 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
342 $T_DOT_GLOBAL = '\.globl';
343 $T_HDR_toc = "\.toc\n";
344 $T_HDR_literal = "\t\.section\t\".toc\",\"aw\"\n";
345 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
346 $T_HDR_data = "\t\.data\n\t\.align 2\n";
347 $T_HDR_rodata = "\t\.section\t.rodata\n\t\.align 2\n";
348 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
349 $T_HDR_info = "\t\.text\n\t\.align 2\n";
350 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
351 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
353 #--------------------------------------------------------#
354 } elsif ( $TargetPlatform =~ /^sparc-.*-(solaris2|openbsd)/ ) {
356 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
357 $T_US = ''; # _ if symbols have an underscore on the front
358 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
359 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
362 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\.section.*|\s+\.type.*|\s+\.size.*)\n)';
363 $T_COPY_DIRVS = '\.(global|proc|stab)';
365 $T_DOT_WORD = '\.(long|word|byte|half|skip|uahalf|uaword)';
366 $T_DOT_GLOBAL = '^\t\.global';
367 $T_HDR_literal = "\.text\n\t\.align 8\n";
368 $T_HDR_misc = "\.text\n\t\.align 4\n";
369 $T_HDR_data = "\.data\n\t\.align 8\n";
370 $T_HDR_rodata = "\.text\n\t\.align 4\n";
371 $T_HDR_closure = "\.data\n\t\.align 4\n";
372 $T_HDR_info = "\.text\n\t\.align 4\n";
373 $T_HDR_entry = "\.text\n\t\.align 4\n";
374 $T_HDR_vector = "\.text\n\t\.align 4\n";
376 #--------------------------------------------------------#
377 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
379 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
380 $T_US = '_'; # _ if symbols have an underscore on the front
381 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
382 $T_CONST_LBL = '^LC(\d+):$';
385 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
386 $T_COPY_DIRVS = '\.(global|proc|stab)';
388 $T_DOT_WORD = '\.word';
389 $T_DOT_GLOBAL = '^\t\.global';
390 $T_HDR_literal = "\.text\n\t\.align 8\n";
391 $T_HDR_misc = "\.text\n\t\.align 4\n";
392 $T_HDR_data = "\.data\n\t\.align 8\n";
393 $T_HDR_rodata = "\.text\n\t\.align 4\n";
394 $T_HDR_closure = "\.data\n\t\.align 4\n";
395 $T_HDR_info = "\.text\n\t\.align 4\n";
396 $T_HDR_entry = "\.text\n\t\.align 4\n";
397 $T_HDR_vector = "\.text\n\t\.align 4\n";
399 #--------------------------------------------------------#
400 } elsif ( $TargetPlatform =~ /^sparc-.*-linux/ ) {
401 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
402 $T_US = ''; # _ if symbols have an underscore on the front
403 $T_PRE_APP = '#'; # regexp that says what comes before APP/NO_APP
404 # Probably doesn't apply anyway
405 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
408 $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)';
409 $T_COPY_DIRVS = '\.(global|globl|proc|stab)';
411 $T_DOT_WORD = '\.(long|word|nword|xword|byte|half|short|skip|uahalf|uaword)';
412 $T_DOT_GLOBAL = '^\t\.global';
413 $T_HDR_literal = "\.text\n\t\.align 8\n";
414 $T_HDR_misc = "\.text\n\t\.align 4\n";
415 $T_HDR_data = "\.data\n\t\.align 8\n";
416 $T_HDR_rodata = "\.text\n\t\.align 4\n";
417 $T_HDR_closure = "\.data\n\t\.align 4\n";
418 $T_HDR_info = "\.text\n\t\.align 4\n";
419 $T_HDR_entry = "\.text\n\t\.align 4\n";
420 $T_HDR_vector = "\.text\n\t\.align 4\n";
422 #--------------------------------------------------------#
424 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
429 print STDERR "T_STABBY: $T_STABBY\n";
430 print STDERR "T_US: $T_US\n";
431 print STDERR "T_PRE_APP: $T_PRE_APP\n";
432 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
433 print STDERR "T_POST_LBL: $T_POST_LBL\n";
434 if ( $TargetPlatform =~ /^i386-/ ) {
435 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
436 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
437 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
439 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
440 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
441 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
442 print STDERR "T_HDR_literal: $T_HDR_literal\n";
443 print STDERR "T_HDR_misc: $T_HDR_misc\n";
444 print STDERR "T_HDR_data: $T_HDR_data\n";
445 print STDERR "T_HDR_rodata: $T_HDR_rodata\n";
446 print STDERR "T_HDR_closure: $T_HDR_closure\n";
447 print STDERR "T_HDR_info: $T_HDR_info\n";
448 print STDERR "T_HDR_entry: $T_HDR_entry\n";
449 print STDERR "T_HDR_vector: $T_HDR_vector\n";
455 %************************************************************************
457 \subsection{Mangle away}
459 %************************************************************************
463 local($in_asmf, $out_asmf) = @_;
465 # multi-line regexp matching:
470 &init_TARGET_STUFF();
471 &init_FUNNY_THINGS();
473 open(INASM, "< $in_asmf")
474 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
475 open(OUTASM,"> $out_asmf")
476 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
478 # read whole file, divide into "chunks":
479 # record some info about what we've found...
481 @chk = (); # contents of the chunk
482 $numchks = 0; # number of them
483 @chkcat = (); # what category of thing in each chunk
484 @chksymb = (); # what symbol(base) is defined in this chunk
485 %entrychk = (); # ditto, its entry code
486 %closurechk = (); # ditto, the (static) closure
487 %srtchk = (); # ditto, its SRT (for top-level things)
488 %infochk = (); # given a symbol base, say what chunk its info tbl is in
489 %vectorchk = (); # ditto, return vector table
490 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
492 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
495 tr/\r//d if $TargetPlatform =~ /-mingw32$/; # In case Perl doesn't convert line endings
496 next if $T_STABBY && /^\.stab.*${T_US}__stg_split_marker/o;
497 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
498 next if /^\t\.def.*endef$/;
499 next if /${T_PRE_APP}(NO_)?APP/o;
500 next if /^;/ && $TargetPlatform =~ /^hppa/;
502 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|ia64)-/;
504 if ( $TargetPlatform =~ /^mips-/
505 && /^\t\.(globl\S+\.text|comm\t)/ ) {
506 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
507 # Treat .comm variables as data. These show up in two (known) places:
509 # - the module_registered variable used in the __stginit fragment.
510 # even though these are declared static and initialised, gcc 3.3
511 # likes to make them .comm, presumably to save space in the
514 # - global variables used to pass arguments from C to STG in
515 # a foreign export. (is this still true? --SDM)
517 } elsif ( /^\t\.comm.*$/ ) {
519 $chkcat[$i] = 'data';
522 # Labels ending "_str": these are literal strings.
523 } elsif ( /^${T_US}([A-Za-z0-9_]+)_str${T_POST_LBL}$/ ) {
525 $chkcat[$i] = 'rodata';
528 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
529 # duplicated from the bottom
533 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
534 # Alphas: Local labels not to be confused with new chunks
536 } elsif ( $TargetPlatform =~ /-darwin/
537 && /^\t\.subsections_via_symbols/) {
538 # Don't allow Apple's linker to do any dead-stripping of symbols
539 # in this file, because it will mess up info-tables in mangled
542 # NB: all the rest start with a non-space
544 } elsif ( $TargetPlatform =~ /^mips-/
545 && /^\d+:/ ) { # a funny-looking very-local label
548 } elsif ( /$T_CONST_LBL/o ) {
550 $chkcat[$i] = 'literal';
553 } elsif ( /^${T_US}__stg_split_marker(\d*)${T_POST_LBL}$/o ) {
555 $chkcat[$i] = 'splitmarker';
558 } elsif ( /^${T_US}([A-Za-z0-9_]+)_info${T_POST_LBL}$/o ) {
561 $chkcat[$i] = 'infotbl';
562 $chksymb[$i] = $symb;
564 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
566 $infochk{$symb} = $i;
568 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(entry|ret)${T_POST_LBL}$/o ) {
570 $chkcat[$i] = 'entry';
575 } elsif ( /^${T_US}([A-Za-z0-9_]+)_closure${T_POST_LBL}$/o ) {
577 $chkcat[$i] = 'closure';
580 $closurechk{$1} = $i;
582 } elsif ( /^${T_US}([A-Za-z0-9_]+)_srt${T_POST_LBL}$/o ) {
589 } elsif ( /^${T_US}([A-Za-z0-9_]+)_ct${T_POST_LBL}$/o ) {
591 $chkcat[$i] = 'data';
594 } elsif ( /^${T_US}(stg_ap_stack_entries|stg_stack_save_entries|stg_arg_bitmaps)${T_POST_LBL}$/o ) {
596 $chkcat[$i] = 'data';
599 } elsif ( /^(${T_US}__gnu_compiled_c|gcc2_compiled\.)${T_POST_LBL}/o ) {
602 } elsif ( /^${T_US}[A-Za-z0-9_]+\.\d+${T_POST_LBL}$/o
603 || /^${T_US}.*_CAT${T_POST_LBL}$/o # PROF: _entryname_CAT
604 || /^${T_US}.*_done${T_POST_LBL}$/o # PROF: _module_done
605 || /^${T_US}_module_registered${T_POST_LBL}$/o # PROF: _module_registered
608 $chkcat[$i] = 'data';
611 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
616 } elsif ( /^${T_US}([A-Za-z0-9_]+)_cc(s)?${T_POST_LBL}$/o ) {
617 # all CC_ symbols go in the data section...
619 $chkcat[$i] = 'data';
622 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(alt|dflt)${T_POST_LBL}$/o ) {
624 $chkcat[$i] = 'misc';
626 } elsif ( /^${T_US}([A-Za-z0-9_]+)_vtbl${T_POST_LBL}$/o ) {
628 $chkcat[$i] = 'vector';
633 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
634 && /^[A-Za-z0-9][A-Za-z0-9_]*:/ ) {
635 # Some Solaris system headers contain function definitions (as
636 # opposed to mere prototypes), which end up in the .hc file when
637 # a Haskell module foreign imports the corresponding system
638 # functions (most notably stat()). We put them into the text
639 # segment. Note that this currently does not extend to function
640 # names starting with an underscore.
643 $chkcat[$i] = 'misc';
646 } elsif ( /^${T_US}[A-Za-z0-9_]/o
647 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
649 && ( $TargetPlatform !~ /^powerpc64/ # we need to avoid local labels in this case
650 || ! /^\.L\d+:$/ ) ) {
657 /^${T_US}stg_.*${T_POST_LBL}$/o # RTS internals
658 || /^${T_US}__stg_.*${T_POST_LBL}$/o # more RTS internals
659 || /^${T_US}__fexp_.*${T_POST_LBL}$/o # foreign export
660 || /^${T_US}.*_slow${T_POST_LBL}$/o # slow entry
661 || /^${T_US}__stginit.*${T_POST_LBL}$/o # __stginit<module>
662 || /^${T_US}.*_btm${T_POST_LBL}$/o # large bitmaps
663 || /^${T_US}.*_fast${T_POST_LBL}$/o # primops
664 || /^_uname:/o # x86/Solaris2
667 $chkcat[$i] = 'misc';
669 /^${T_US}.*_srtd${T_POST_LBL}$/o # large bitmaps
670 || /^${T_US}.*_closure_tbl${T_POST_LBL}$/o # closure tables
673 $chkcat[$i] = 'rodata';
676 print STDERR "Warning: retaining unknown function \`$thing' in output from C compiler\n";
677 $chkcat[$i] = 'unknown';
680 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && (
682 || /^\.section __TEXT,__picsymbol_stub1,.*/
683 || /^\.section __TEXT,__picsymbolstub1,.*/
685 || /^\.section __TEXT,__symbol_stub1,.*/
686 || /^\.section __TEXT,__symbolstub1,.*/
687 || /^\.lazy_symbol_pointer/
688 || /^\.non_lazy_symbol_pointer/ ))
691 $chkcat[$i] = 'dyld';
693 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.data/ && $chkcat[$i] eq 'dyld')
694 { # non_lazy_symbol_ptrs that point to local symbols
696 $chkcat[$i] = 'dyld';
698 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ && /^\.LCTOC1 = /o ) {
699 # PowerPC Linux's large-model PIC (-fPIC) generates a gobal offset
700 # table "by hand". Be sure to copy it over.
701 # Note that this label and all entries in the table should actually
702 # go into the .got2 section, but it isn't easy to distinguish them
703 # from other constant literals (.LC\d+), so we just put everything
706 $chkcat[$i] = 'literal';
707 $chksymb[$i] = 'LCTOC1';
708 } else { # simple line (duplicated at the top)
713 $numchks = $#chk + 1;
715 # open CHUNKS, ">/tmp/chunks1" or die "Cannot open /tmp/chunks1: $!\n";
716 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
719 # the division into chunks is imperfect;
720 # we throw some things over the fence into the next
723 # also, there are things we would like to know
724 # about the whole module before we start spitting
727 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
728 local($FIRST_TOSSABLE ) = ($TargetPlatform =~ /^(hppa|mips-)/) ? 1 : 0;
730 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
732 # Alphas: NB: we start meddling at chunk 1, not chunk 0
733 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
734 # spits a ".quad 0" in after the very first ".rdata"; we
735 # detect this special case (tossing the ".quad 0")!
736 local($magic_rdata_seen) = 0;
738 # HPPAs, MIPSen: also start medding at chunk 1
740 for ($i = $FIRST_TOSSABLE; $i < $numchks; $i++) {
741 $c = $chk[$i]; # convenience copy
743 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
745 # toss all prologue stuff; HPPA is pretty weird
747 $c = &hppa_mash_prologue($c) if $TargetPlatform =~ /^hppa-/;
749 # be slightly paranoid to make sure there's
750 # nothing surprising in there
751 if ( $c =~ /--- BEGIN ---/ ) {
752 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
754 # remove junk whitespace around the split point
758 if ($TargetPlatform =~ /^i386-/) {
759 $p =~ s/^\tpushl\s+\%edi\n//;
760 $p =~ s/^\tpushl\s+\%esi\n//;
761 $p =~ s/^\tpushl\s+\%ebx\n//;
762 $p =~ s/^\tmovl\s+\%esi,\s*\d*\(\%esp\)\n//;
763 $p =~ s/^\tmovl\s+\%edi,\s*\d*\(\%esp\)\n//;
764 $p =~ s/^\tsubl\s+\$\d+,\s*\%esp\n//;
765 $p =~ s/^\tmovl\s+\$\d+,\s*\%eax\n\tcall\s+__alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
767 # GCC 3.1 is in the habit of adding spurious writes to the
768 # stack in the prologue. Just to be on the safe side,
769 # chuck these over the fence into the main code.
770 while ($p =~ /^\tmovl\s+\$\d+,\s*\d*\(\%esp\)\n/) {
771 # print "Spurious instruction: $&";
776 } elsif ($TargetPlatform =~ /^x86_64-/) {
777 $p =~ s/^\tpushq\s+\%r(bx|bp|12|13|14)\n//g;
778 $p =~ s/^\tmovq\s+\%r(bx|bp|12|13|14),\s*\d*\(\%rsp\)\n//g;
779 $p =~ s/^\tsubq\s+\$\d+,\s*\%rsp\n//;
781 } elsif ($TargetPlatform =~ /^ia64-/) {
782 $p =~ s/^\t\.prologue .*\n//;
783 $p =~ s/^\t\.save ar\.pfs, r\d+\n\talloc r\d+ = ar\.pfs, 0, 3[12], \d+, 0\n//;
784 $p =~ s/^\t\.fframe \d+\n\tadds r12 = -\d+, r12\n//;
785 $p =~ s/^\t\.save rp, r\d+\n\tmov r\d+ = b0\n//;
786 $p =~ s/^\t\.(mii|mmi)\n//g; # bundling is no longer sensible
787 $p =~ s/^\t;;\n//g; # discard stops
788 $p =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
790 # GCC 3.3 saves r1 in the prologue, move this to the body
791 if ($p =~ /^\tmov r\d+ = r1\n/) {
795 } elsif ($TargetPlatform =~ /^m68k-/) {
796 $p =~ s/^\tlink a6,#-?\d.*\n//;
797 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
798 # The above showed up in the asm code,
799 # so I added it here.
800 # I hope it's correct.
802 $p =~ s/^\tmovel d2,sp\@-\n//;
803 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
804 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
805 } elsif ($TargetPlatform =~ /^mips-/) {
806 # the .frame/.mask/.fmask that we use is the same
807 # as that produced by GCC for miniInterpret; this
808 # gives GDB some chance of figuring out what happened
809 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
810 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
811 $p =~ s/^\t\.(mask|fmask).*\n//g;
812 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
813 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
814 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
815 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
816 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
817 $p =~ s/__FRAME__/$FRAME/;
818 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
820 $pcrel_label =~ s/(.|\n)*^(\"?L\d+\$pb\"?):\n(.|\n)*/$2/ or $pcrel_label = "";
822 $p =~ s/^\tmflr r0\n//;
823 $p =~ s/^\tbl saveFP # f\d+\n//;
824 $p =~ s/^\tbl saveFP ; save f\d+-f\d+\n//;
825 $p =~ s/^\"?L\d+\$pb\"?:\n//;
826 $p =~ s/^\tstmw r\d+,-\d+\(r1\)\n//;
827 $p =~ s/^\tstfd f\d+,-\d+\(r1\)\n//g;
828 $p =~ s/^\tstw r0,\d+\(r1\)\n//g;
829 $p =~ s/^\tstwu r1,-\d+\(r1\)\n//;
830 $p =~ s/^\tstw r\d+,-\d+\(r1\)\n//g;
831 $p =~ s/^\tbcl 20,31,L\d+\$pb\n//;
832 $p =~ s/^L\d+\$pb:\n//;
833 $p =~ s/^\tmflr r31\n//;
835 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
836 # under some circumstances, only when generating position dependent code.
837 # I have no idea why, and I don't think it is necessary, so let's toss it.
838 $p =~ s/^\tli r\d+,0\n//g;
839 $p =~ s/^\tstw r\d+,\d+\(r1\)\n//g;
840 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/) {
841 $p =~ s/^\tmflr 0\n//;
842 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
843 $p =~ s/^\tstfd \d+,\d+\(1\)\n//g;
844 $p =~ s/^\tstw r0,8\(1\)\n//;
845 $p =~ s/^\tstwu 1,-\d+\(1\)\n//;
846 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
848 # GCC's "large-model" PIC (-fPIC)
850 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
852 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
853 $p =~ s/^.LCF\d+:\n//;
854 $p =~ s/^\tmflr 30\n//;
855 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
856 $p =~ s/^\tadd 30,0,30\n//;
858 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
859 # under some circumstances, only when generating position dependent code.
860 # I have no idea why, and I don't think it is necessary, so let's toss it.
861 $p =~ s/^\tli \d+,0\n//g;
862 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
863 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
864 $p =~ s/^\tmr 31,1\n//;
865 $p =~ s/^\tmflr 0\n//;
866 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
867 $p =~ s/^\tstfd \d+,-?\d+\(1\)\n//g;
868 $p =~ s/^\tstd r0,8\(1\)\n//;
869 $p =~ s/^\tstdu 1,-\d+\(1\)\n//;
870 $p =~ s/^\tstd \d+,-?\d+\(1\)\n//g;
872 # GCC's "large-model" PIC (-fPIC)
874 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
876 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
877 $p =~ s/^.LCF\d+:\n//;
878 $p =~ s/^\tmflr 30\n//;
879 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
880 $p =~ s/^\tadd 30,0,30\n//;
882 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
883 # under some circumstances, only when generating position dependent code.
884 # I have no idea why, and I don't think it is necessary, so let's toss it.
885 $p =~ s/^\tli \d+,0\n//g;
886 $p =~ s/^\tstd \d+,\d+\(1\)\n//g;
888 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
891 # HWL HACK: dont die, just print a warning
892 #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/;
893 die "Prologue junk?: $p\n" if $p =~ /^\s+[^\s\.]/;
895 # For PIC, we want to keep part of the prologue
896 if ($TargetPlatform =~ /^powerpc-apple-.*/ && $pcrel_label ne "") {
897 # Darwin: load the current instruction pointer into register r31
898 $p .= "bcl 20,31,$pcrel_label\n";
899 $p .= "$pcrel_label:\n";
900 $p .= "\tmflr r31\n";
901 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/ && $pcrel_label ne "") {
902 # Linux: load the GOT pointer into register 30
903 $p .= "\tbcl 20,31,.LCF$pcrel_label\n";
904 $p .= ".LCF$pcrel_label:\n";
906 $p .= "\tlwz 0,.LCL$pcrel_label-.LCF$pcrel_label(30)\n";
907 $p .= "\tadd 30,0,30\n";
910 # glue together what's left
915 if ( $TargetPlatform =~ /^mips-/ ) {
916 # MIPS: first, this basic sequence may occur "--- END ---" or not
917 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
920 # toss all epilogue stuff; again, paranoidly
921 if ( $c =~ /--- END ---/ ) {
922 if (($r, $e) = split(/--- END ---/, $c)) {
923 if ($TargetPlatform =~ /^i386-/) {
925 $e =~ s/^\tpopl\s+\%edi\n//;
926 $e =~ s/^\tpopl\s+\%esi\n//;
927 $e =~ s/^\tpopl\s+\%edx\n//;
928 $e =~ s/^\tpopl\s+\%ecx\n//;
929 $e =~ s/^\taddl\s+\$\d+,\s*\%esp\n//;
930 $e =~ s/^\tsubl\s+\$-\d+,\s*\%esp\n//;
931 } elsif ($TargetPlatform =~ /^ia64-/) {
932 $e =~ s/^\tmov ar\.pfs = r\d+\n//;
933 $e =~ s/^\tmov b0 = r\d+\n//;
934 $e =~ s/^\t\.restore sp\n\tadds r12 = \d+, r12\n//;
935 $e =~ s/^\tbr\.ret\.sptk\.many b0\n//;
936 $e =~ s/^\t\.(mii|mmi|mib)\n//g; # bundling is no longer sensible
937 $e =~ s/^\t;;\n//g; # discard stops - stop at end of body is sufficient
938 $e =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
939 } elsif ($TargetPlatform =~ /^m68k-/) {
940 $e =~ s/^\tunlk a6\n//;
942 } elsif ($TargetPlatform =~ /^mips-/) {
943 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
944 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
945 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
946 $e =~ s/^\tj\t\$31\n//;
947 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
948 $e =~ s/^\taddi r1,r1,\d+\n//;
949 $e =~ s/^\tlwz r\d+,\d+\(r1\)\n//;
950 $e =~ s/^\tlmw r\d+,-\d+\(r1\)\n//;
951 $e =~ s/^\tmtlr r0\n//;
953 $e =~ s/^\tb restFP ;.*\n//;
954 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
955 $e =~ s/^\tmr 3,0\n//;
956 $e =~ s/^\taddi 1,1,\d+\n//;
957 $e =~ s/^\tld 0,16\(1\)\n//;
958 $e =~ s/^\tmtlr 0\n//;
960 # callee-save registers
961 $e =~ s/^\tld \d+,-?\d+\(1\)\n//g;
962 $e =~ s/^\tlfd \d+,-?\d+\(1\)\n//g;
964 # get rid of the debug junk along with the blr
965 $e =~ s/^\tblr\n\t.long .*\n\t.byte .*\n//;
967 # incase we missed it with the last one get the blr alone
970 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
973 print STDERR "WARNING: Epilogue junk?: $e\n" if $e =~ /^\t\s*[^\.\s\n]/;
975 # glue together what's left
977 $c =~ s/\n\t\n/\n/; # junk blank line
981 # On SPARCs, we don't do --- BEGIN/END ---, we just
982 # toss the register-windowing save/restore/ret* instructions
983 # directly unless they've been generated by function definitions in header
985 if ( $TargetPlatform =~ /^sparc-/ ) {
986 if ( ! ( $TargetPlatform =~ /solaris2$/ && $chkcat[$i] eq 'unknown' )) {
987 $c =~ s/^\t(save.*|restore.*|ret|retl)\n//g;
989 # throw away PROLOGUE comments
990 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
993 # On Alphas, the prologue mangling is done a little later (below)
995 # toss all calls to __DISCARD__
996 $c =~ s/^\t(call|jbsr|jal)\s+${T_US}__DISCARD__\n//go;
997 $c =~ s/^\tjsr\s+\$26\s*,\s*${T_US}__DISCARD__\n//go if $TargetPlatform =~ /^alpha-/;
998 $c =~ s/^\tbl\s+L___DISCARD__\$stub\n//go if $TargetPlatform =~ /^powerpc-apple-.*/;
999 $c =~ s/^\tbl\s+__DISCARD__(\@plt)?\n//go if $TargetPlatform =~ /^powerpc-.*-linux/;
1000 $c =~ s/^\tbl\s+\.__DISCARD__\n\s+nop\n//go if $TargetPlatform =~ /^powerpc64-.*-linux/;
1002 # IA64: mangle tailcalls into jumps here
1003 if ($TargetPlatform =~ /^ia64-/) {
1004 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/) {
1005 # Eek, the gcc optimiser is getting smarter... if we see a jump to the --- TAILCALL ---
1006 # marker then we reapply the substitution at the source sites
1007 $c =~ s/^\tbr \.L$2\n/\t--- TAILCALL ---\n/g if ($2);
1011 # MIPS: that may leave some gratuitous asm macros around
1012 # (no harm done; but we get rid of them to be tidier)
1013 $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/
1014 if $TargetPlatform =~ /^mips-/;
1016 # toss stack adjustment after DoSparks
1017 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
1018 if $TargetPlatform =~ /^m68k-/; # this looks old...
1020 if ( $TargetPlatform =~ /^alpha-/ &&
1021 ! $magic_rdata_seen &&
1022 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
1023 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
1024 $magic_rdata_seen = 1;
1027 # pick some end-things and move them to the next chunk
1029 # pin a funny end-thing on (for easier matching):
1030 $c .= 'FUNNY#END#THING';
1032 while ( $c =~ /${T_MOVE_DIRVS}FUNNY#END#THING/o ) {
1036 # on x86 we try not to copy any directives into a literal
1037 # chunk, rather we keep looking for the next real chunk. This
1038 # is because we get things like
1040 # .globl blah_closure
1046 if ( $TargetPlatform =~ /^(i386|sparc|powerpc)/ && $to_move =~ /${T_COPY_DIRVS}/ ) {
1048 while ( $j < $numchks && $chk[$j] =~ /$T_CONST_LBL/) {
1051 if ( $j < $numchks ) {
1052 $chk[$j] = $to_move . $chk[$j];
1056 elsif ( $i < ($numchks - 1)
1057 && ( $to_move =~ /${T_COPY_DIRVS}/
1058 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
1059 $chk[$i + 1] = $to_move . $chk[$i + 1];
1060 # otherwise they're tossed
1063 $c =~ s/${T_MOVE_DIRVS}FUNNY#END#THING/FUNNY#END#THING/o;
1066 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
1068 # toss all prologue stuff, except for loading gp, and the ..ng address
1069 unless ($c =~ /\.ent.*\n\$.*\.\.ng:/) {
1070 if (($p, $r) = split(/^\t\.prologue/, $c)) {
1071 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
1072 $keep =~ s/^\t\.frame.*\n/\t.frame \$30,0,\$26,0\n/;
1073 $keep =~ s/^\t\.(mask|fmask).*\n//g;
1074 $c = $keep . "..ng:\n";
1076 print STDERR "malformed code block ($ent)?\n"
1079 $c .= "\t.prologue" . $r;
1083 $c =~ s/FUNNY#END#THING//;
1085 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
1087 $chk[$i] = $c; # update w/ convenience copy
1090 # open CHUNKS, ">/tmp/chunks2" or die "Cannot open /tmp/chunks2: $!\n";
1091 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
1094 if ( $TargetPlatform =~ /^alpha-/ ) {
1095 # print out the header stuff first
1096 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
1097 print OUTASM $chk[0];
1099 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1100 print OUTASM $chk[0];
1102 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1103 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
1105 # get rid of horrible "<dollar>Revision: .*$" strings
1106 local(@lines0) = split(/\n/, $chk[0]);
1108 while ( $z <= $#lines0 ) {
1109 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
1112 while ( $z <= $#lines0 ) {
1114 last if $lines0[$z] =~ /[,\t]0x0$/;
1120 $chk[0] = join("\n", @lines0);
1121 $chk[0] =~ s/\n\n+/\n/;
1122 print OUTASM $chk[0];
1125 # print out all the literal strings next
1126 for ($i = 0; $i < $numchks; $i++) {
1127 if ( $chkcat[$i] eq 'literal' ) {
1128 print OUTASM $T_HDR_literal, $chk[$i];
1129 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
1131 $chkcat[$i] = 'DONE ALREADY';
1135 # on the HPPA, print out all the bss next
1136 if ( $TargetPlatform =~ /^hppa/ ) {
1137 for ($i = 1; $i < $numchks; $i++) {
1138 if ( $chkcat[$i] eq 'bss' ) {
1139 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
1140 print OUTASM $chk[$i];
1142 $chkcat[$i] = 'DONE ALREADY';
1147 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
1148 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
1150 next if $chkcat[$i] eq 'DONE ALREADY';
1152 if ( $chkcat[$i] eq 'misc' || $chkcat[$i] eq 'unknown' ) {
1153 if ($chk[$i] ne '') {
1154 print OUTASM $T_HDR_misc;
1155 &print_doctored($chk[$i], 0);
1158 } elsif ( $chkcat[$i] eq 'toss' ) {
1159 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
1161 } elsif ( $chkcat[$i] eq 'data' ) {
1162 if ($chk[$i] ne '') {
1163 print OUTASM $T_HDR_data;
1164 print OUTASM $chk[$i];
1167 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
1168 # we can just re-constitute this one...
1169 # NB: we emit _three_ underscores no matter what,
1170 # so ghc-split doesn't have to care.
1171 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
1173 } elsif ( $chkcat[$i] eq 'closure'
1174 || $chkcat[$i] eq 'srt'
1175 || $chkcat[$i] eq 'infotbl'
1176 || $chkcat[$i] eq 'entry') { # do them in that order
1177 $symb = $chksymb[$i];
1180 if ( defined($closurechk{$symb}) ) {
1181 print OUTASM $T_HDR_closure;
1182 print OUTASM $chk[$closurechk{$symb}];
1183 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
1187 if ( defined($srtchk{$symb}) ) {
1188 print OUTASM $T_HDR_rodata;
1189 print OUTASM $chk[$srtchk{$symb}];
1190 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
1194 if ( defined($infochk{$symb}) ) {
1196 print OUTASM $T_HDR_info;
1197 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
1199 # entry code will be put here!
1201 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
1205 if ( defined($entrychk{$symb}) ) {
1207 $c = $chk[$entrychk{$symb}];
1209 # If this is an entry point with an info table,
1210 # eliminate the entry symbol and all directives involving it.
1211 if (defined($infochk{$symb}) && $TargetPlatform !~ /^ia64-/) {
1212 $c =~ s/^.*$symb_(entry|ret)${T_POST_LBL}\n//;
1213 $c =~ s/^\s*\..*$symb.*\n//g;
1216 print OUTASM $T_HDR_entry;
1218 &print_doctored($c, 1); # NB: the 1!!!
1220 $chkcat[$entrychk{$symb}] = 'DONE ALREADY';
1223 } elsif ( $chkcat[$i] eq 'vector' ) {
1224 $symb = $chksymb[$i];
1227 if ( defined($vectorchk{$symb}) ) {
1228 print OUTASM $T_HDR_vector;
1229 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1231 # direct return code will be put here!
1232 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1234 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1235 # Alphas: the commented nop is for the splitter, to ensure
1236 # that no module ends with a label as the very last
1237 # thing. (The linker will adjust the label to point
1238 # to the first code word of the next module linked in,
1239 # even if alignment constraints cause the label to move!)
1241 print OUTASM "\t# nop\n";
1244 } elsif ( $chkcat[$i] eq 'rodata' ) {
1245 print OUTASM $T_HDR_rodata;
1246 print OUTASM $chk[$i];
1247 $chkcat[$i] = 'DONE ALREADY';
1249 } elsif ( $chkcat[$i] eq 'toc' ) {
1250 # silly optimisation to print tocs, since they come in groups...
1251 print OUTASM $T_HDR_toc;
1253 while ($chkcat[$j] eq 'toc')
1254 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1257 print OUTASM $chk[$j];
1259 $chkcat[$j] = 'DONE ALREADY';
1263 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && $chkcat[$i] eq 'dyld' ) {
1264 # powerpc-apple: dynamic linker stubs
1265 if($chk[$i] !~ /\.indirect_symbol ___DISCARD__/)
1266 { # print them out unchanged, but remove the stubs for __DISCARD__
1267 print OUTASM $chk[$i];
1270 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1274 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1277 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1278 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1283 sub hppa_mash_prologue { # OK, epilogue, too
1286 # toss all prologue stuff
1287 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1289 # Lie about our .CALLINFO
1290 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1297 # toss all epilogue stuff
1298 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1300 # Sorry; we moved the _info stuff to the code segment.
1301 s/_info,DATA/_info,CODE/g;
1308 sub print_doctored {
1309 local($_, $need_fallthru_patch) = @_;
1311 if ( $TargetPlatform !~ /^i386-/
1312 || ! /^\t[a-z]/ # no instructions in here, apparently
1313 || /^${T_US}__stginit_[A-Za-z0-9_]+${T_POST_LBL}/) {
1317 # OK, must do some x86 **HACKING**
1319 local($entry_patch) = '';
1320 local($exit_patch) = '';
1322 # gotta watch out for weird instructions that
1323 # invisibly smash various regs:
1324 # rep* %ecx used for counting
1325 # scas* %edi used for destination index
1326 # cmps* %e[sd]i used for indices
1327 # loop* %ecx used for counting
1332 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1334 # * GCC used an "STG reg" for its own purposes
1336 # * some secret uses of machine reg, requiring STG reg
1337 # to be saved/restored
1339 # The most dangerous "GCC uses" of an "STG reg" are when
1340 # the reg holds the target of a jmp -- it's tricky to
1341 # insert the patch-up code before we get to the target!
1342 # So here we change the jmps:
1344 # --------------------------------------------------------
1345 # it can happen that we have jumps of the form...
1346 # jmp *<something involving %esp>
1348 # jmp <something involving another naughty register...>
1350 # a reasonably-common case is:
1352 # movl $_blah,<bad-reg>
1356 s/^\tmovl\s+\$${T_US}(.*),\s*(\%e[abcd]x)\n\tjmp\s+\*\2/\tjmp $T_US$1/g;
1358 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1359 s/^\tmovl\s+(.*),\s*\%esi\n\tjmp\s+\*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1360 s/^\tjmp\s+\*(.*\(.*\%esi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1361 s/^\tjmp\s+\*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1362 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1363 if /(jmp|call)\s+.*\%esi/;
1365 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1366 s/^\tmovl\s+(.*),\s*\%edi\n\tjmp\s+\*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1367 s/^\tjmp\s+\*(.*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1368 s/^\tjmp\s+\*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1369 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1370 if /(jmp|call)\s+.*\%edi/;
1373 # OK, now we can decide what our patch-up code is going to
1376 # Offsets into register table - you'd better update these magic
1377 # numbers should you change its contents!
1378 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1379 local($OFFSET_Hp)=88;
1381 # Note funky ".=" stuff; we're *adding* to these _patch guys
1382 if ( $StolenX86Regs <= 2
1383 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1384 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1385 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1387 # nothing for call_{entry,exit} because %esi is callee-save
1389 if ( $StolenX86Regs <= 3
1390 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1391 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1392 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1394 # nothing for call_{entry,exit} because %edi is callee-save
1397 # --------------------------------------------------------
1398 # next, here we go with non-%esp patching!
1400 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1402 # Before calling GC we must set up the exit condition before the call
1403 # and entry condition when we come back
1405 # fix _all_ non-local jumps:
1407 s/^\tjmp\s+\*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1408 s/^\tjmp\s+${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1410 s/^(\tjmp\s+.*\n)/$exit_patch$1/g; # here's the fix...
1412 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1413 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1415 if ($StolenX86Regs == 2 ) {
1416 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1417 if /^\t(jmp|call)\s+.*\%e(si|di)/;
1418 } elsif ($StolenX86Regs == 3 ) {
1419 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1420 if /^\t(jmp|call)\s+.*\%edi/;
1423 # --------------------------------------------------------
1424 # that's it -- print it
1426 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1430 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1431 print OUTASM $exit_patch;
1432 # ToDo: make it not print if there is a "jmp" at the end
1438 sub init_FUNNY_THINGS {
1439 %KNOWN_FUNNY_THING = (
1441 # "${T_US}stg_.*{T_POST_LBL}", 1,
1446 The following table reversal is used for both info tables and return
1447 vectors. In both cases, we remove the first entry from the table,
1448 reverse the table, put the label at the end, and paste some code
1449 (that which is normally referred to by the first entry in the table)
1450 right after the table itself. (The code pasting is done elsewhere.)
1454 local($symb, $tbl, $discard1) = @_;
1456 return ($tbl) if ($TargetPlatform =~ /^ia64-/);
1458 local($before) = '';
1460 local(@imports) = (); # hppa only
1463 local(@lines) = split(/\n/, $tbl);
1466 # Deal with the header...
1467 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1468 $label .= $lines[$i] . "\n",
1469 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info${T_POST_LBL}$/o
1470 || $lines[$i] =~ /${T_DOT_GLOBAL}/o
1471 || $lines[$i] =~ /^${T_US}\S+_vtbl${T_POST_LBL}$/o;
1473 $before .= $lines[$i] . "\n"; # otherwise...
1477 $infoname =~ s/(.|\n)*^([A-Za-z0-9_]+_info)${T_POST_LBL}$(.|\n)*/\2/;
1479 # Grab the table data...
1480 if ( $TargetPlatform !~ /^hppa/ ) {
1481 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1483 # Convert addresses of SRTs, slow entrypoints and large bitmaps
1484 # to offsets (relative to the info label),
1485 # in order to support position independent code.
1486 $line =~ s/$infoname/0/
1487 || $line =~ s/([A-Za-z0-9_]+_srtd)$/\1 - $infoname/
1488 || $line =~ s/([A-Za-z0-9_]+_srt(\+\d+)?)$/\1 - $infoname/
1489 || $line =~ s/([A-Za-z0-9_]+_slow)$/\1 - $infoname/
1490 || $line =~ s/([A-Za-z0-9_]+_btm)$/\1 - $infoname/
1491 || $line =~ s/([A-Za-z0-9_]+_alt)$/\1 - $infoname/
1492 || $line =~ s/([A-Za-z0-9_]+_dflt)$/\1 - $infoname/
1493 || $line =~ s/([A-Za-z0-9_]+_ret)$/\1 - $infoname/;
1494 push(@words, $line);
1496 } else { # hppa weirdness
1497 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+(${T_DOT_WORD}|\.IMPORT)/; $i++) {
1498 # FIXME: the RTS now expects offsets instead of addresses
1499 # for all labels in info tables.
1500 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1501 push(@imports, $lines[$i]);
1503 # We don't use HP's ``function pointers''
1504 # We just use labels in code space, like normal people
1505 $lines[$i] =~ s/P%//;
1506 push(@words, $lines[$i]);
1511 # Now throw away any initial zero word from the table. This is a hack
1512 # that lets us reduce the size of info tables when the SRT field is not
1513 # needed: see comments StgFunInfoTable in InfoTables.h.
1515 # The .zero business is for Linux/ELF.
1516 # The .skip business is for Sparc/Solaris/ELF.
1517 # The .blockz business is for HPPA.
1519 # if ($words[0] =~ /^\t?(${T_DOT_WORD}\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1524 for (; $i <= $#lines; $i++) {
1525 $after .= $lines[$i] . "\n";
1528 # Alphas: If we have anonymous text (not part of a procedure), the
1529 # linker may complain about missing exception information. Bleh.
1530 # To suppress this, we place a .ent/.end pair around the code.
1531 # At the same time, we have to be careful and not enclose any leading
1532 # .file/.loc directives.
1533 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1534 local ($ident) = $1;
1535 $before =~ s/^((\s*\.(file|loc)\s+[^\n]*\n)*)/$1\t.ent $ident\n/;
1536 $after .= "\t.end $ident\n";
1539 # Alphas: The heroic Simon Marlow found a bug in the Digital UNIX
1540 # assembler (!) wherein .quad constants inside .text sections are
1541 # first narrowed to 32 bits then sign-extended back to 64 bits.
1542 # This obviously screws up our 64-bit bitmaps, so we work around
1543 # the bug by replacing .quad with .align 3 + .long + .long [ccshan]
1544 if ( $TargetPlatform =~ /^alpha-/ ) {
1546 if (/^\s*\.quad\s+([-+0-9].*\S)\s*$/ && length $1 >= 10) {
1547 local ($number) = $1;
1548 if ($number =~ /^([-+])?(0x?)?([0-9]+)$/) {
1549 local ($sign, $base, $digits) = ($1, $2, $3);
1550 $base = (10, 8, 16)[length $base];
1551 local ($hi, $lo) = (0, 0);
1552 foreach $i (split(//, $digits)) {
1553 $j = $lo * $base + $i;
1554 $lo = $j % 4294967296;
1555 $hi = $hi * $base + ($j - $lo) / 4294967296;
1557 ($hi, $lo) = (4294967295 - $hi, 4294967296 - $lo)
1559 $_ = "\t.align 3\n\t.long $lo\n\t.long $hi\n";
1560 # printf STDERR "TURNING %s into 0x %08x %08x\n", $number, $hi, $lo;
1562 print STDERR "Cannot handle \".quad $number\" in info table\n";
1570 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1571 . join("\n", @words) . "\n"
1574 # print STDERR "before=$before\n";
1575 # print STDERR "label=$label\n";
1576 # print STDERR "words=",(reverse @words),"\n";
1577 # print STDERR "after=$after\n";
1583 The HP is a major nuisance. The threaded code mangler moved info
1584 tables from data space to code space, but unthreaded code in the RTS
1585 still has references to info tables in data space. Since the HP
1586 linker is very precise about where symbols live, we need to patch the
1587 references in the unthreaded RTS as well.
1590 sub mini_mangle_asm_hppa {
1591 local($in_asmf, $out_asmf) = @_;
1593 open(INASM, "< $in_asmf")
1594 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1595 open(OUTASM,"> $out_asmf")
1596 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1599 s/_info,DATA/_info,CODE/; # Move _info references to code space
1605 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1606 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1612 sub tidy_up_and_die {
1613 local($return_val, $msg) = @_;
1615 exit (($return_val == 0) ? 0 : 1);