1 %************************************************************************
3 \section[Driver-asm-fiddling]{Fiddling with assembler files}
5 %************************************************************************
10 Utterly stomp out C functions' prologues and epilogues; i.e., the
11 stuff to do with the C stack.
13 Any other required tidying up.
16 General note [chak]: Many regexps are very fragile because they rely on white
17 space being in the right place. This caused trouble with gcc 2.95 (at least
18 on Linux), where the use of white space in .s files generated by gcc suddenly
19 changed. To guarantee compatibility across different versions of gcc, make
20 sure (at least on i386-.*-linux) that regexps tolerate varying amounts of white
21 space between an assembler statement and its arguments as well as after a the
22 comma separating multiple arguments.
24 \emph{For the time being, I have corrected the regexps for i386-.*-linux. I
25 didn't touch all the regexps for other i386 platforms, as I don't have
26 a box to test these changes.}
31 The HP linker is very picky about symbols being in the appropriate
32 space (code vs. data). When we mangle the threaded code to put the
33 info tables just prior to the code, they wind up in code space
34 rather than data space. This means that references to *_info from
35 un-mangled parts of the RTS (e.g. unthreaded GC code) get
36 unresolved symbols. Solution: mini-mangler for .c files on HP. I
37 think this should really be triggered in the driver by a new -rts
38 option, so that user code doesn't get mangled inappropriately.
40 With reversed tables, jumps are to the _info label rather than to
41 the _entry label. The _info label is just an address in code
42 space, rather than an entry point with the descriptive blob we
43 talked about yesterday. As a result, you can't use the call-style
44 JMP_ macro. However, some JMP_ macros take _info labels as targets
45 and some take code entry points within the RTS. The latter won't
46 work with the goto-style JMP_ macro. Sigh. Solution: Use the goto
47 style JMP_ macro, and mangle some more assembly, changing all
48 "RP'literal" and "LP'literal" references to "R'literal" and
49 "L'literal," so that you get the real address of the code, rather
50 than the descriptive blob. Also change all ".word P%literal"
51 entries in info tables and vector tables to just ".word literal,"
52 for the same reason. Advantage: No more ridiculous call sequences.
55 %************************************************************************
57 \subsection{Top-level code}
59 %************************************************************************
62 $TargetPlatform = $TARGETPLATFORM;
64 ($Pgm = $0) =~ s|.*/||;
68 if ( $TargetPlatform =~ /^i386-/ ) {
72 $StolenX86Regs = $ARGV[2];
76 &mangle_asm($ifile,$ofile);
81 %************************************************************************
83 \subsection{Constants for various architectures}
85 %************************************************************************
88 sub init_TARGET_STUFF {
90 #--------------------------------------------------------#
91 if ( $TargetPlatform =~ /^alpha-.*-.*/ ) {
93 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
94 $T_US = ''; # _ if symbols have an underscore on the front
95 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
96 $T_CONST_LBL = '^\$L?C(\d+):$'; # regexp for what such a lbl looks like
99 $T_MOVE_DIRVS = '^(\s*(\$.*\.\.ng:|\.align\s+\d+|\.(globl|ent)\s+\S+|\#.*|\.(file|loc)\s+\S+\s+\S+|\.text|\.r?data)\n)';
100 $T_COPY_DIRVS = '^\s*(\$.*\.\.ng:|\#|\.(file|globl|ent|loc))';
102 $T_DOT_WORD = '\.(long|quad|byte|word)';
103 $T_DOT_GLOBAL = '^\t\.globl';
104 $T_HDR_literal = "\.rdata\n\t\.align 3\n";
105 $T_HDR_misc = "\.text\n\t\.align 3\n";
106 $T_HDR_data = "\.data\n\t\.align 3\n";
107 $T_HDR_rodata = "\.rdata\n\t\.align 3\n";
108 $T_HDR_closure = "\.data\n\t\.align 3\n";
109 $T_HDR_info = "\.text\n\t\.align 3\n";
110 $T_HDR_entry = "\.text\n\t\.align 3\n";
111 $T_HDR_vector = "\.text\n\t\.align 3\n";
113 #--------------------------------------------------------#
114 } elsif ( $TargetPlatform =~ /^hppa/ ) {
116 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
117 $T_US = ''; # _ if symbols have an underscore on the front
118 $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
119 $T_CONST_LBL = '^L\$C(\d+)$'; # regexp for what such a lbl looks like
122 $T_MOVE_DIRVS = '^((\s+\.(IMPORT|EXPORT|PARAM).*|\s+\.align\s+\d+|\s+\.(SPACE|SUBSPA)\s+\S+|\s*)\n)';
123 $T_COPY_DIRVS = '^\s+\.(IMPORT|EXPORT)';
125 $T_DOT_WORD = '\.(blockz|word|half|byte)';
126 $T_DOT_GLOBAL = '^\s+\.EXPORT';
127 $T_HDR_literal = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
128 $T_HDR_misc = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
129 $T_HDR_data = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
130 $T_HDR_rodata = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
131 $T_HDR_closure = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
132 $T_HDR_info = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
133 $T_HDR_entry = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
134 $T_HDR_vector = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
136 #--------------------------------------------------------#
137 } elsif ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd2|nextstep3|cygwin32|mingw32)$/ ) {
138 # NeXT added but not tested. CaS
140 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
141 $T_US = '_'; # _ if symbols have an underscore on the front
142 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
143 $T_CONST_LBL = '^LC(\d+):$';
145 $T_X86_PRE_LLBL_PAT = 'L';
146 $T_X86_PRE_LLBL = 'L';
147 $T_X86_BADJMP = '^\tjmp [^L\*]';
149 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s.*|\.globl\s+\S+|\.text|\.data|\.stab[^n].*|\.type\s+.*|\.size\s+.*|\.lcomm.*)\n)';
150 $T_COPY_DIRVS = '\.(globl|stab|lcomm)';
151 $T_DOT_WORD = '\.(long|word|value|byte|space)';
152 $T_DOT_GLOBAL = '\.globl';
153 $T_HDR_literal = "\.text\n\t\.align 2\n";
154 $T_HDR_misc = "\.text\n\t\.align 2,0x90\n";
155 $T_HDR_data = "\.data\n\t\.align 2\n";
156 $T_HDR_rodata = "\.text\n\t\.align 2\n";
157 $T_HDR_closure = "\.data\n\t\.align 2\n";
158 $T_HDR_info = "\.text\n\t\.align 2\n"; # NB: requires padding
159 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
160 $T_HDR_vector = "\.text\n\t\.align 2\n"; # NB: requires padding
162 #--------------------------------------------------------#
163 } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|freebsd|netbsd|openbsd)$/ ) {
165 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
166 $T_US = ''; # _ if symbols have an underscore on the front
167 $T_PRE_APP = # regexp that says what comes before APP/NO_APP
168 ($TargetPlatform =~ /-(linux|freebsd|netbsd|openbsd)$/) ? '#' : '/' ;
169 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
171 $T_X86_PRE_LLBL_PAT = '\.L';
172 $T_X86_PRE_LLBL = '.L';
173 $T_X86_BADJMP = '^\tjmp\s+[^\.\*]';
175 $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s.*|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.type\s+.*|\.size\s+\S+\s*,\s*\d+|\.ident.*|\.local.*)\n)';
176 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
178 $T_DOT_WORD = '\.(long|value|word|byte|zero)';
179 $T_DOT_GLOBAL = '\.globl';
180 $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11)
181 $T_HDR_misc = "\.text\n\t\.align 4\n";
182 $T_HDR_data = "\.data\n\t\.align 4\n";
183 $T_HDR_rodata = "\.section\t\.rodata\n\t\.align 4\n";
184 $T_HDR_closure = "\.data\n\t\.align 4\n";
185 $T_HDR_info = "\.text\n\t\.align 4\n";
186 $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?)
187 $T_HDR_vector = "\.text\n\t\.align 4\n"; # NB: requires padding
189 #--------------------------------------------------------#
190 } elsif ( $TargetPlatform =~ /^ia64-.*-linux$/ ) {
192 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
193 $T_US = ''; # _ if symbols have an underscore on the front
195 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
198 $T_MOVE_DIRVS = '^(\s*\.(global|proc|pred\.safe_across_calls|text|data|section|subsection|align|size|type|ident)\s+.*\n)';
199 $T_COPY_DIRVS = '\.(global|proc)';
201 $T_DOT_WORD = '\.(long|value|byte|zero)';
202 $T_DOT_GLOBAL = '\.global';
203 $T_HDR_literal = "\.section\t\.rodata\n";
204 $T_HDR_misc = "\.text\n\t\.align 8\n";
205 $T_HDR_data = "\.data\n\t\.align 8\n";
206 $T_HDR_rodata = "\.section\t\.rodata\n\t\.align 8\n";
207 $T_HDR_closure = "\.data\n\t\.align 8\n";
208 $T_HDR_info = "\.text\n\t\.align 8\n";
209 $T_HDR_entry = "\.text\n\t\.align 16\n";
210 $T_HDR_vector = "\.text\n\t\.align 8\n";
212 #--------------------------------------------------------#
213 } elsif ( $TargetPlatform =~ /^x86_64-.*-(linux|openbsd)$/ ) {
215 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
216 $T_US = ''; # _ if symbols have an underscore on the front
218 $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
221 $T_MOVE_DIRVS = '^(\s*\.(globl|text|data|section|align|size|type|ident|local)\s+.*\n)';
222 $T_COPY_DIRVS = '\.(globl|local)';
224 $T_DOT_WORD = '\.(quad|long|value|byte|zero)';
225 $T_DOT_GLOBAL = '\.global';
226 $T_HDR_literal = "\.section\t\.rodata\n";
227 $T_HDR_misc = "\.text\n\t\.align 8\n";
228 $T_HDR_data = "\.data\n\t\.align 8\n";
229 $T_HDR_rodata = "\.section\t\.rodata\n\t\.align 8\n";
231 # the assembler on x86_64/Linux refuses to generate code for
233 # where x is in the text section and y in the rodata section.
234 # It works if y is in the text section, though. This is probably
235 # going to cause difficulties for PIC, I imagine.
236 $T_HDR_relrodata= "\.text\n\t\.align 8\n";
238 $T_HDR_closure = "\.data\n\t\.align 8\n";
239 $T_HDR_info = "\.text\n\t\.align 8\n";
240 $T_HDR_entry = "\.text\n\t\.align 8\n";
241 $T_HDR_vector = "\.text\n\t\.align 8\n";
243 #--------------------------------------------------------#
244 } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
246 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
247 $T_US = '_'; # _ if symbols have an underscore on the front
248 $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
249 $T_CONST_LBL = '^LC(\d+):$';
252 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
253 $T_COPY_DIRVS = '\.(globl|proc|stab)';
255 $T_DOT_WORD = '\.long';
256 $T_DOT_GLOBAL = '\.globl';
257 $T_HDR_literal = "\.text\n\t\.even\n";
258 $T_HDR_misc = "\.text\n\t\.even\n";
259 $T_HDR_data = "\.data\n\t\.even\n";
260 $T_HDR_rodata = "\.text\n\t\.even\n";
261 $T_HDR_closure = "\.data\n\t\.even\n";
262 $T_HDR_info = "\.text\n\t\.even\n";
263 $T_HDR_entry = "\.text\n\t\.even\n";
264 $T_HDR_vector = "\.text\n\t\.even\n";
266 #--------------------------------------------------------#
267 } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
269 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
270 $T_US = ''; # _ if symbols have an underscore on the front
271 $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
272 $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
275 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
276 $T_COPY_DIRVS = '\.(globl|ent)';
278 $T_DOT_WORD = '\.word';
279 $T_DOT_GLOBAL = '^\t\.globl';
280 $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
281 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
282 $T_HDR_data = "\t\.data\n\t\.align 2\n";
283 $T_HDR_rodata = "\t\.rdata\n\t\.align 2\n";
284 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
285 $T_HDR_info = "\t\.text\n\t\.align 2\n";
286 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
287 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
289 #--------------------------------------------------------#
290 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ ) {
291 # Apple PowerPC Darwin/MacOS X.
292 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
293 $T_US = '_'; # _ if symbols have an underscore on the front
294 $T_PRE_APP = 'DOESNT APPLY'; # regexp that says what comes before APP/NO_APP
295 $T_CONST_LBL = '^\LC\d+:'; # regexp for what such a lbl looks like
298 $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)';
299 $T_COPY_DIRVS = '\.(globl|lcomm)';
301 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
302 $T_DOT_GLOBAL = '\.globl';
303 $T_HDR_toc = "\.toc\n";
304 $T_HDR_literal = "\t\.const\n\t\.align 2\n";
305 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
306 $T_HDR_data = "\t\.data\n\t\.align 2\n";
307 $T_HDR_rodata = "\t\.const\n\t\.align 2\n";
308 $T_HDR_relrodata= "\t\.const_data\n\t\.align 2\n";
309 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
310 $T_HDR_info = "\t\.text\n\t\.align 2\n";
311 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
312 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
314 #--------------------------------------------------------#
315 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ ) {
317 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
318 $T_US = ''; # _ if symbols have an underscore on the front
319 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
320 $T_CONST_LBL = '^\.LC\d+:'; # regexp for what such a lbl looks like
323 $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)';
324 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
326 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
327 $T_DOT_GLOBAL = '\.globl';
328 $T_HDR_toc = "\.toc\n";
329 $T_HDR_literal = "\t\.section\t.rodata\n\t\.align 2\n";
330 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
331 $T_HDR_data = "\t\.data\n\t\.align 2\n";
332 $T_HDR_rodata = "\t\.section\t.rodata\n\t\.align 2\n";
333 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
334 $T_HDR_info = "\t\.text\n\t\.align 2\n";
335 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
336 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
338 #--------------------------------------------------------#
339 } elsif ( $TargetPlatform =~ /^powerpc64-.*-linux/ ) {
341 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
342 $T_US = '\.'; # _ if symbols have an underscore on the front
343 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
344 $T_CONST_LBL = '^\.LC\d+:'; # regexp for what such a lbl looks like
347 $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)';
348 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
350 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
351 $T_DOT_GLOBAL = '\.globl';
352 $T_HDR_toc = "\.toc\n";
353 $T_HDR_literal = "\t\.section\t\".toc\",\"aw\"\n";
354 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
355 $T_HDR_data = "\t\.data\n\t\.align 2\n";
356 $T_HDR_rodata = "\t\.section\t.rodata\n\t\.align 2\n";
357 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
358 $T_HDR_info = "\t\.text\n\t\.align 2\n";
359 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
360 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
362 #--------------------------------------------------------#
363 } elsif ( $TargetPlatform =~ /^sparc-.*-(solaris2|openbsd)/ ) {
365 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
366 $T_US = ''; # _ if symbols have an underscore on the front
367 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
368 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
371 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\s*\.section.*|\s+\.type.*|\s+\.size.*)\n)';
372 $T_COPY_DIRVS = '\.(global|proc|stab)';
374 $T_DOT_WORD = '\.(long|word|byte|half|skip|uahalf|uaword)';
375 $T_DOT_GLOBAL = '^\t\.global';
376 $T_HDR_literal = "\.text\n\t\.align 8\n";
377 $T_HDR_misc = "\.text\n\t\.align 4\n";
378 $T_HDR_data = "\.data\n\t\.align 8\n";
379 $T_HDR_rodata = "\.text\n\t\.align 4\n";
380 $T_HDR_closure = "\.data\n\t\.align 4\n";
381 $T_HDR_info = "\.text\n\t\.align 4\n";
382 $T_HDR_entry = "\.text\n\t\.align 4\n";
383 $T_HDR_vector = "\.text\n\t\.align 4\n";
385 #--------------------------------------------------------#
386 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
388 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
389 $T_US = '_'; # _ if symbols have an underscore on the front
390 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
391 $T_CONST_LBL = '^LC(\d+):$';
394 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
395 $T_COPY_DIRVS = '\.(global|proc|stab)';
397 $T_DOT_WORD = '\.word';
398 $T_DOT_GLOBAL = '^\t\.global';
399 $T_HDR_literal = "\.text\n\t\.align 8\n";
400 $T_HDR_misc = "\.text\n\t\.align 4\n";
401 $T_HDR_data = "\.data\n\t\.align 8\n";
402 $T_HDR_rodata = "\.text\n\t\.align 4\n";
403 $T_HDR_closure = "\.data\n\t\.align 4\n";
404 $T_HDR_info = "\.text\n\t\.align 4\n";
405 $T_HDR_entry = "\.text\n\t\.align 4\n";
406 $T_HDR_vector = "\.text\n\t\.align 4\n";
408 #--------------------------------------------------------#
409 } elsif ( $TargetPlatform =~ /^sparc-.*-linux/ ) {
410 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
411 $T_US = ''; # _ if symbols have an underscore on the front
412 $T_PRE_APP = '#'; # regexp that says what comes before APP/NO_APP
413 # Probably doesn't apply anyway
414 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
417 $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)';
418 $T_COPY_DIRVS = '\.(global|globl|proc|stab)';
420 $T_DOT_WORD = '\.(long|word|nword|xword|byte|half|short|skip|uahalf|uaword)';
421 $T_DOT_GLOBAL = '^\t\.global';
422 $T_HDR_literal = "\.text\n\t\.align 8\n";
423 $T_HDR_misc = "\.text\n\t\.align 4\n";
424 $T_HDR_data = "\.data\n\t\.align 8\n";
425 $T_HDR_rodata = "\.text\n\t\.align 4\n";
426 $T_HDR_closure = "\.data\n\t\.align 4\n";
427 $T_HDR_info = "\.text\n\t\.align 4\n";
428 $T_HDR_entry = "\.text\n\t\.align 4\n";
429 $T_HDR_vector = "\.text\n\t\.align 4\n";
431 #--------------------------------------------------------#
433 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
437 if($T_HDR_relrodata eq "") {
439 # relrodata defaults to rodata.
440 $T_HDR_relrodata = $T_HDR_rodata;
444 print STDERR "T_STABBY: $T_STABBY\n";
445 print STDERR "T_US: $T_US\n";
446 print STDERR "T_PRE_APP: $T_PRE_APP\n";
447 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
448 print STDERR "T_POST_LBL: $T_POST_LBL\n";
449 if ( $TargetPlatform =~ /^i386-/ ) {
450 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
451 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
452 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
454 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
455 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
456 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
457 print STDERR "T_HDR_literal: $T_HDR_literal\n";
458 print STDERR "T_HDR_misc: $T_HDR_misc\n";
459 print STDERR "T_HDR_data: $T_HDR_data\n";
460 print STDERR "T_HDR_rodata: $T_HDR_rodata\n";
461 print STDERR "T_HDR_closure: $T_HDR_closure\n";
462 print STDERR "T_HDR_info: $T_HDR_info\n";
463 print STDERR "T_HDR_entry: $T_HDR_entry\n";
464 print STDERR "T_HDR_vector: $T_HDR_vector\n";
470 %************************************************************************
472 \subsection{Mangle away}
474 %************************************************************************
478 local($in_asmf, $out_asmf) = @_;
480 # multi-line regexp matching:
485 &init_TARGET_STUFF();
486 &init_FUNNY_THINGS();
488 open(INASM, "< $in_asmf")
489 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
490 open(OUTASM,"> $out_asmf")
491 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
493 # read whole file, divide into "chunks":
494 # record some info about what we've found...
496 @chk = (); # contents of the chunk
497 $numchks = 0; # number of them
498 @chkcat = (); # what category of thing in each chunk
499 @chksymb = (); # what symbol(base) is defined in this chunk
500 %entrychk = (); # ditto, its entry code
501 %closurechk = (); # ditto, the (static) closure
502 %srtchk = (); # ditto, its SRT (for top-level things)
503 %infochk = (); # given a symbol base, say what chunk its info tbl is in
504 %vectorchk = (); # ditto, return vector table
505 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
507 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
510 tr/\r//d if $TargetPlatform =~ /-mingw32$/; # In case Perl doesn't convert line endings
511 next if $T_STABBY && /^\.stab.*${T_US}__stg_split_marker/o;
512 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
513 next if /^\t\.def.*endef$/;
514 next if /${T_PRE_APP}(NO_)?APP/o;
515 next if /^;/ && $TargetPlatform =~ /^hppa/;
517 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|ia64)-/;
519 if ( $TargetPlatform =~ /^mips-/
520 && /^\t\.(globl\S+\.text|comm\t)/ ) {
521 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
522 # Treat .comm variables as data. These show up in two (known) places:
524 # - the module_registered variable used in the __stginit fragment.
525 # even though these are declared static and initialised, gcc 3.3
526 # likes to make them .comm, presumably to save space in the
529 # - global variables used to pass arguments from C to STG in
530 # a foreign export. (is this still true? --SDM)
532 } elsif ( /^\t\.comm.*$/ ) {
534 $chkcat[$i] = 'data';
537 # Labels ending "_str": these are literal strings.
538 } elsif ( /^${T_US}([A-Za-z0-9_]+)_str${T_POST_LBL}$/ ) {
540 $chkcat[$i] = 'rodata';
542 } elsif ( $TargetPlatform =~ /-darwin/
543 && (/^\s*\.subsections_via_symbols/
544 ||/^\s*\.no_dead_strip.*/)) {
545 # Don't allow Apple's linker to do any dead-stripping of symbols
546 # in this file, because it will mess up info-tables in mangled
548 # The .no_dead_strip directives are actually put there by
549 # the gcc3 "used" attribute on entry points.
550 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
551 # duplicated from the bottom
555 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
556 # Alphas: Local labels not to be confused with new chunks
558 # NB: all the rest start with a non-space
560 } elsif ( $TargetPlatform =~ /^mips-/
561 && /^\d+:/ ) { # a funny-looking very-local label
564 } elsif ( /$T_CONST_LBL/o ) {
566 $chkcat[$i] = 'literal';
569 } elsif ( /^${T_US}__stg_split_marker(\d*)${T_POST_LBL}$/o ) {
571 $chkcat[$i] = 'splitmarker';
574 } elsif ( /^${T_US}([A-Za-z0-9_]+)_info${T_POST_LBL}$/o ) {
577 $chkcat[$i] = 'infotbl';
578 $chksymb[$i] = $symb;
580 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
582 $infochk{$symb} = $i;
584 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(entry|ret)${T_POST_LBL}$/o ) {
586 $chkcat[$i] = 'entry';
591 } elsif ( /^${T_US}([A-Za-z0-9_]+)_closure${T_POST_LBL}$/o ) {
593 $chkcat[$i] = 'closure';
596 $closurechk{$1} = $i;
598 } elsif ( /^${T_US}([A-Za-z0-9_]+)_srt${T_POST_LBL}$/o ) {
605 } elsif ( /^${T_US}([A-Za-z0-9_]+)_ct${T_POST_LBL}$/o ) {
607 $chkcat[$i] = 'data';
610 } elsif ( /^${T_US}(stg_ap_stack_entries|stg_stack_save_entries|stg_arg_bitmaps)${T_POST_LBL}$/o ) {
612 $chkcat[$i] = 'data';
615 } elsif ( /^(${T_US}__gnu_compiled_c|gcc2_compiled\.)${T_POST_LBL}/o ) {
618 } elsif ( /^${T_US}[A-Za-z0-9_]+\.\d+${T_POST_LBL}$/o
619 || /^${T_US}.*_CAT${T_POST_LBL}$/o # PROF: _entryname_CAT
620 || /^${T_US}.*_done${T_POST_LBL}$/o # PROF: _module_done
621 || /^${T_US}_module_registered${T_POST_LBL}$/o # PROF: _module_registered
624 $chkcat[$i] = 'data';
627 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
632 } elsif ( /^${T_US}([A-Za-z0-9_]+)_cc(s)?${T_POST_LBL}$/o ) {
633 # all CC_ symbols go in the data section...
635 $chkcat[$i] = 'data';
638 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(alt|dflt)${T_POST_LBL}$/o ) {
640 $chkcat[$i] = 'misc';
642 } elsif ( /^${T_US}([A-Za-z0-9_]+)_vtbl${T_POST_LBL}$/o ) {
644 $chkcat[$i] = 'vector';
649 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
650 && /^[A-Za-z0-9][A-Za-z0-9_]*:/ ) {
651 # Some Solaris system headers contain function definitions (as
652 # opposed to mere prototypes), which end up in the .hc file when
653 # a Haskell module foreign imports the corresponding system
654 # functions (most notably stat()). We put them into the text
655 # segment. Note that this currently does not extend to function
656 # names starting with an underscore.
659 $chkcat[$i] = 'misc';
662 } elsif ( /^${T_US}[A-Za-z0-9_]/o
663 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
665 && ( $TargetPlatform !~ /^powerpc64/ # we need to avoid local labels in this case
666 || ! /^\.L\d+:$/ ) ) {
673 /^${T_US}stg_.*${T_POST_LBL}$/o # RTS internals
674 || /^${T_US}__stg_.*${T_POST_LBL}$/o # more RTS internals
675 || /^${T_US}__fexp_.*${T_POST_LBL}$/o # foreign export
676 || /^${T_US}.*_slow${T_POST_LBL}$/o # slow entry
677 || /^${T_US}__stginit.*${T_POST_LBL}$/o # __stginit<module>
678 || /^${T_US}.*_btm${T_POST_LBL}$/o # large bitmaps
679 || /^${T_US}.*_fast${T_POST_LBL}$/o # primops
680 || /^_uname:/o # x86/Solaris2
683 $chkcat[$i] = 'misc';
685 /^${T_US}.*_srtd${T_POST_LBL}$/o # large bitmaps
686 || /^${T_US}.*_closure_tbl${T_POST_LBL}$/o # closure tables
689 $chkcat[$i] = 'relrodata';
692 print STDERR "Warning: retaining unknown function \`$thing' in output from C compiler\n";
693 $chkcat[$i] = 'unknown';
696 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && (
698 || /^\.section __TEXT,__picsymbol_stub1,.*/
699 || /^\.section __TEXT,__picsymbolstub1,.*/
701 || /^\.section __TEXT,__symbol_stub1,.*/
702 || /^\.section __TEXT,__symbolstub1,.*/
703 || /^\.lazy_symbol_pointer/
704 || /^\.non_lazy_symbol_pointer/ ))
707 $chkcat[$i] = 'dyld';
709 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.data/ && $chkcat[$i] eq 'dyld')
710 { # non_lazy_symbol_ptrs that point to local symbols
712 $chkcat[$i] = 'dyld';
714 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ && /^\.LCTOC1 = /o ) {
715 # PowerPC Linux's large-model PIC (-fPIC) generates a gobal offset
716 # table "by hand". Be sure to copy it over.
717 # Note that this label and all entries in the table should actually
718 # go into the .got2 section, but it isn't easy to distinguish them
719 # from other constant literals (.LC\d+), so we just put everything
722 $chkcat[$i] = 'literal';
723 $chksymb[$i] = 'LCTOC1';
724 } else { # simple line (duplicated at the top)
729 $numchks = $#chk + 1;
731 # open CHUNKS, ">/tmp/chunks1" or die "Cannot open /tmp/chunks1: $!\n";
732 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
735 # the division into chunks is imperfect;
736 # we throw some things over the fence into the next
739 # also, there are things we would like to know
740 # about the whole module before we start spitting
743 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
744 local($FIRST_TOSSABLE ) = ($TargetPlatform =~ /^(hppa|mips-)/) ? 1 : 0;
746 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
748 # Alphas: NB: we start meddling at chunk 1, not chunk 0
749 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
750 # spits a ".quad 0" in after the very first ".rdata"; we
751 # detect this special case (tossing the ".quad 0")!
752 local($magic_rdata_seen) = 0;
754 # HPPAs, MIPSen: also start medding at chunk 1
756 for ($i = $FIRST_TOSSABLE; $i < $numchks; $i++) {
757 $c = $chk[$i]; # convenience copy
759 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
761 # toss all prologue stuff; HPPA is pretty weird
763 $c = &hppa_mash_prologue($c) if $TargetPlatform =~ /^hppa-/;
765 # be slightly paranoid to make sure there's
766 # nothing surprising in there
767 if ( $c =~ /--- BEGIN ---/ ) {
768 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
770 # remove junk whitespace around the split point
774 if ($TargetPlatform =~ /^i386-/) {
775 $p =~ s/^\tpushl\s+\%edi\n//;
776 $p =~ s/^\tpushl\s+\%esi\n//;
777 $p =~ s/^\tpushl\s+\%ebx\n//;
778 $p =~ s/^\tmovl\s+\%esi,\s*\d*\(\%esp\)\n//;
779 $p =~ s/^\tmovl\s+\%edi,\s*\d*\(\%esp\)\n//;
780 $p =~ s/^\tsubl\s+\$\d+,\s*\%esp\n//;
781 $p =~ s/^\tmovl\s+\$\d+,\s*\%eax\n\tcall\s+__alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
783 # GCC 3.1 is in the habit of adding spurious writes to the
784 # stack in the prologue. Just to be on the safe side,
785 # chuck these over the fence into the main code.
786 while ($p =~ /^\tmovl\s+\$\d+,\s*\d*\(\%esp\)\n/) {
787 # print "Spurious instruction: $&";
792 } elsif ($TargetPlatform =~ /^x86_64-/) {
793 $p =~ s/^\tpushq\s+\%r(bx|bp|12|13|14)\n//g;
794 $p =~ s/^\tmovq\s+\%r(bx|bp|12|13|14),\s*\d*\(\%rsp\)\n//g;
795 $p =~ s/^\tsubq\s+\$\d+,\s*\%rsp\n//;
797 } elsif ($TargetPlatform =~ /^ia64-/) {
798 $p =~ s/^\t\.prologue .*\n//;
799 $p =~ s/^\t\.save ar\.pfs, r\d+\n\talloc r\d+ = ar\.pfs, 0, 3[12], \d+, 0\n//;
800 $p =~ s/^\t\.fframe \d+\n\tadds r12 = -\d+, r12\n//;
801 $p =~ s/^\t\.save rp, r\d+\n\tmov r\d+ = b0\n//;
802 $p =~ s/^\t\.(mii|mmi)\n//g; # bundling is no longer sensible
803 $p =~ s/^\t;;\n//g; # discard stops
804 $p =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
806 # GCC 3.3 saves r1 in the prologue, move this to the body
807 if ($p =~ /^\tmov r\d+ = r1\n/) {
811 } elsif ($TargetPlatform =~ /^m68k-/) {
812 $p =~ s/^\tlink a6,#-?\d.*\n//;
813 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
814 # The above showed up in the asm code,
815 # so I added it here.
816 # I hope it's correct.
818 $p =~ s/^\tmovel d2,sp\@-\n//;
819 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
820 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
821 } elsif ($TargetPlatform =~ /^mips-/) {
822 # the .frame/.mask/.fmask that we use is the same
823 # as that produced by GCC for miniInterpret; this
824 # gives GDB some chance of figuring out what happened
825 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
826 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
827 $p =~ s/^\t\.(mask|fmask).*\n//g;
828 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
829 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
830 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
831 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
832 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
833 $p =~ s/__FRAME__/$FRAME/;
834 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
836 $pcrel_label =~ s/(.|\n)*^(\"?L\d+\$pb\"?):\n(.|\n)*/$2/ or $pcrel_label = "";
838 $p =~ s/^\tmflr r0\n//;
839 $p =~ s/^\tbl saveFP # f\d+\n//;
840 $p =~ s/^\tbl saveFP ; save f\d+-f\d+\n//;
841 $p =~ s/^\"?L\d+\$pb\"?:\n//;
842 $p =~ s/^\tstmw r\d+,-\d+\(r1\)\n//;
843 $p =~ s/^\tstfd f\d+,-\d+\(r1\)\n//g;
844 $p =~ s/^\tstw r0,\d+\(r1\)\n//g;
845 $p =~ s/^\tstwu r1,-\d+\(r1\)\n//;
846 $p =~ s/^\tstw r\d+,-\d+\(r1\)\n//g;
847 $p =~ s/^\tbcl 20,31,L\d+\$pb\n//;
848 $p =~ s/^L\d+\$pb:\n//;
849 $p =~ s/^\tmflr r31\n//;
851 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
852 # under some circumstances, only when generating position dependent code.
853 # I have no idea why, and I don't think it is necessary, so let's toss it.
854 $p =~ s/^\tli r\d+,0\n//g;
855 $p =~ s/^\tstw r\d+,\d+\(r1\)\n//g;
856 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/) {
857 $p =~ s/^\tmflr 0\n//;
858 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
859 $p =~ s/^\tstfd \d+,\d+\(1\)\n//g;
860 $p =~ s/^\tstw r0,8\(1\)\n//;
861 $p =~ s/^\tstwu 1,-\d+\(1\)\n//;
862 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
864 # GCC's "large-model" PIC (-fPIC)
866 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
868 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
869 $p =~ s/^.LCF\d+:\n//;
870 $p =~ s/^\tmflr 30\n//;
871 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
872 $p =~ s/^\tadd 30,0,30\n//;
874 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
875 # under some circumstances, only when generating position dependent code.
876 # I have no idea why, and I don't think it is necessary, so let's toss it.
877 $p =~ s/^\tli \d+,0\n//g;
878 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
879 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
880 $p =~ s/^\tmr 31,1\n//;
881 $p =~ s/^\tmflr 0\n//;
882 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
883 $p =~ s/^\tstfd \d+,-?\d+\(1\)\n//g;
884 $p =~ s/^\tstd r0,8\(1\)\n//;
885 $p =~ s/^\tstdu 1,-\d+\(1\)\n//;
886 $p =~ s/^\tstd \d+,-?\d+\(1\)\n//g;
888 # GCC's "large-model" PIC (-fPIC)
890 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
892 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
893 $p =~ s/^.LCF\d+:\n//;
894 $p =~ s/^\tmflr 30\n//;
895 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
896 $p =~ s/^\tadd 30,0,30\n//;
898 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
899 # under some circumstances, only when generating position dependent code.
900 # I have no idea why, and I don't think it is necessary, so let's toss it.
901 $p =~ s/^\tli \d+,0\n//g;
902 $p =~ s/^\tstd \d+,\d+\(1\)\n//g;
904 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
907 # HWL HACK: dont die, just print a warning
908 #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/;
909 die "Prologue junk?: $p\n" if $p =~ /^\s+[^\s\.]/;
911 # For PIC, we want to keep part of the prologue
912 if ($TargetPlatform =~ /^powerpc-apple-.*/ && $pcrel_label ne "") {
913 # Darwin: load the current instruction pointer into register r31
914 $p .= "bcl 20,31,$pcrel_label\n";
915 $p .= "$pcrel_label:\n";
916 $p .= "\tmflr r31\n";
917 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/ && $pcrel_label ne "") {
918 # Linux: load the GOT pointer into register 30
919 $p .= "\tbcl 20,31,.LCF$pcrel_label\n";
920 $p .= ".LCF$pcrel_label:\n";
922 $p .= "\tlwz 0,.LCL$pcrel_label-.LCF$pcrel_label(30)\n";
923 $p .= "\tadd 30,0,30\n";
926 # glue together what's left
931 if ( $TargetPlatform =~ /^mips-/ ) {
932 # MIPS: first, this basic sequence may occur "--- END ---" or not
933 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
936 # toss all epilogue stuff; again, paranoidly
937 if ( $c =~ /--- END ---/ ) {
938 if (($r, $e) = split(/--- END ---/, $c)) {
939 if ($TargetPlatform =~ /^i386-/) {
941 $e =~ s/^\tpopl\s+\%edi\n//;
942 $e =~ s/^\tpopl\s+\%esi\n//;
943 $e =~ s/^\tpopl\s+\%edx\n//;
944 $e =~ s/^\tpopl\s+\%ecx\n//;
945 $e =~ s/^\taddl\s+\$\d+,\s*\%esp\n//;
946 $e =~ s/^\tsubl\s+\$-\d+,\s*\%esp\n//;
947 } elsif ($TargetPlatform =~ /^ia64-/) {
948 $e =~ s/^\tmov ar\.pfs = r\d+\n//;
949 $e =~ s/^\tmov b0 = r\d+\n//;
950 $e =~ s/^\t\.restore sp\n\tadds r12 = \d+, r12\n//;
951 $e =~ s/^\tbr\.ret\.sptk\.many b0\n//;
952 $e =~ s/^\t\.(mii|mmi|mib)\n//g; # bundling is no longer sensible
953 $e =~ s/^\t;;\n//g; # discard stops - stop at end of body is sufficient
954 $e =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
955 } elsif ($TargetPlatform =~ /^m68k-/) {
956 $e =~ s/^\tunlk a6\n//;
958 } elsif ($TargetPlatform =~ /^mips-/) {
959 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
960 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
961 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
962 $e =~ s/^\tj\t\$31\n//;
963 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
964 $e =~ s/^\taddi r1,r1,\d+\n//;
965 $e =~ s/^\tlwz r\d+,\d+\(r1\)\n//;
966 $e =~ s/^\tlmw r\d+,-\d+\(r1\)\n//;
967 $e =~ s/^\tmtlr r0\n//;
969 $e =~ s/^\tb restFP ;.*\n//;
970 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
971 $e =~ s/^\tmr 3,0\n//;
972 $e =~ s/^\taddi 1,1,\d+\n//;
973 $e =~ s/^\tld 0,16\(1\)\n//;
974 $e =~ s/^\tmtlr 0\n//;
976 # callee-save registers
977 $e =~ s/^\tld \d+,-?\d+\(1\)\n//g;
978 $e =~ s/^\tlfd \d+,-?\d+\(1\)\n//g;
980 # get rid of the debug junk along with the blr
981 $e =~ s/^\tblr\n\t.long .*\n\t.byte .*\n//;
983 # incase we missed it with the last one get the blr alone
986 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
989 print STDERR "WARNING: Epilogue junk?: $e\n" if $e =~ /^\t\s*[^\.\s\n]/;
991 # glue together what's left
993 $c =~ s/\n\t\n/\n/; # junk blank line
997 # On SPARCs, we don't do --- BEGIN/END ---, we just
998 # toss the register-windowing save/restore/ret* instructions
999 # directly unless they've been generated by function definitions in header
1001 if ( $TargetPlatform =~ /^sparc-/ ) {
1002 if ( ! ( $TargetPlatform =~ /solaris2$/ && $chkcat[$i] eq 'unknown' )) {
1003 $c =~ s/^\t(save.*|restore.*|ret|retl)\n//g;
1005 # throw away PROLOGUE comments
1006 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
1009 # On Alphas, the prologue mangling is done a little later (below)
1011 # toss all calls to __DISCARD__
1012 $c =~ s/^\t(call|jbsr|jal)\s+${T_US}__DISCARD__\n//go;
1013 $c =~ s/^\tjsr\s+\$26\s*,\s*${T_US}__DISCARD__\n//go if $TargetPlatform =~ /^alpha-/;
1014 $c =~ s/^\tbl\s+L___DISCARD__\$stub\n//go if $TargetPlatform =~ /^powerpc-apple-.*/;
1015 $c =~ s/^\tbl\s+__DISCARD__(\@plt)?\n//go if $TargetPlatform =~ /^powerpc-.*-linux/;
1016 $c =~ s/^\tbl\s+\.__DISCARD__\n\s+nop\n//go if $TargetPlatform =~ /^powerpc64-.*-linux/;
1018 # IA64: mangle tailcalls into jumps here
1019 if ($TargetPlatform =~ /^ia64-/) {
1020 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/) {
1021 # Eek, the gcc optimiser is getting smarter... if we see a jump to the --- TAILCALL ---
1022 # marker then we reapply the substitution at the source sites
1023 $c =~ s/^\tbr \.L$2\n/\t--- TAILCALL ---\n/g if ($2);
1027 # MIPS: that may leave some gratuitous asm macros around
1028 # (no harm done; but we get rid of them to be tidier)
1029 $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/
1030 if $TargetPlatform =~ /^mips-/;
1032 # toss stack adjustment after DoSparks
1033 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
1034 if $TargetPlatform =~ /^m68k-/; # this looks old...
1036 if ( $TargetPlatform =~ /^alpha-/ &&
1037 ! $magic_rdata_seen &&
1038 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
1039 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
1040 $magic_rdata_seen = 1;
1043 # pick some end-things and move them to the next chunk
1045 # pin a funny end-thing on (for easier matching):
1046 $c .= 'FUNNY#END#THING';
1048 while ( $c =~ /${T_MOVE_DIRVS}FUNNY#END#THING/o ) {
1052 # on x86 we try not to copy any directives into a literal
1053 # chunk, rather we keep looking for the next real chunk. This
1054 # is because we get things like
1056 # .globl blah_closure
1062 if ( $TargetPlatform =~ /^(i386|sparc|powerpc)/ && $to_move =~ /${T_COPY_DIRVS}/ ) {
1064 while ( $j < $numchks && $chk[$j] =~ /$T_CONST_LBL/) {
1067 if ( $j < $numchks ) {
1068 $chk[$j] = $to_move . $chk[$j];
1072 elsif ( $i < ($numchks - 1)
1073 && ( $to_move =~ /${T_COPY_DIRVS}/
1074 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
1075 $chk[$i + 1] = $to_move . $chk[$i + 1];
1076 # otherwise they're tossed
1079 $c =~ s/${T_MOVE_DIRVS}FUNNY#END#THING/FUNNY#END#THING/o;
1082 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
1084 # toss all prologue stuff, except for loading gp, and the ..ng address
1085 unless ($c =~ /\.ent.*\n\$.*\.\.ng:/) {
1086 if (($p, $r) = split(/^\t\.prologue/, $c)) {
1087 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
1088 $keep =~ s/^\t\.frame.*\n/\t.frame \$30,0,\$26,0\n/;
1089 $keep =~ s/^\t\.(mask|fmask).*\n//g;
1090 $c = $keep . "..ng:\n";
1092 print STDERR "malformed code block ($ent)?\n"
1095 $c .= "\t.prologue" . $r;
1099 $c =~ s/FUNNY#END#THING//;
1101 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
1103 $chk[$i] = $c; # update w/ convenience copy
1106 # open CHUNKS, ">/tmp/chunks2" or die "Cannot open /tmp/chunks2: $!\n";
1107 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
1110 if ( $TargetPlatform =~ /^alpha-/ ) {
1111 # print out the header stuff first
1112 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
1113 print OUTASM $chk[0];
1115 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1116 print OUTASM $chk[0];
1118 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1119 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
1121 # get rid of horrible "<dollar>Revision: .*$" strings
1122 local(@lines0) = split(/\n/, $chk[0]);
1124 while ( $z <= $#lines0 ) {
1125 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
1128 while ( $z <= $#lines0 ) {
1130 last if $lines0[$z] =~ /[,\t]0x0$/;
1136 $chk[0] = join("\n", @lines0);
1137 $chk[0] =~ s/\n\n+/\n/;
1138 print OUTASM $chk[0];
1141 # print out all the literal strings next
1142 for ($i = 0; $i < $numchks; $i++) {
1143 if ( $chkcat[$i] eq 'literal' ) {
1144 print OUTASM $T_HDR_literal, $chk[$i];
1145 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
1147 $chkcat[$i] = 'DONE ALREADY';
1151 # on the HPPA, print out all the bss next
1152 if ( $TargetPlatform =~ /^hppa/ ) {
1153 for ($i = 1; $i < $numchks; $i++) {
1154 if ( $chkcat[$i] eq 'bss' ) {
1155 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
1156 print OUTASM $chk[$i];
1158 $chkcat[$i] = 'DONE ALREADY';
1163 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
1164 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
1166 next if $chkcat[$i] eq 'DONE ALREADY';
1168 if ( $chkcat[$i] eq 'misc' || $chkcat[$i] eq 'unknown' ) {
1169 if ($chk[$i] ne '') {
1170 print OUTASM $T_HDR_misc;
1171 &print_doctored($chk[$i], 0);
1174 } elsif ( $chkcat[$i] eq 'toss' ) {
1175 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
1177 } elsif ( $chkcat[$i] eq 'data' ) {
1178 if ($chk[$i] ne '') {
1179 print OUTASM $T_HDR_data;
1180 print OUTASM $chk[$i];
1183 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
1184 # we can just re-constitute this one...
1185 # NB: we emit _three_ underscores no matter what,
1186 # so ghc-split doesn't have to care.
1187 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
1189 } elsif ( $chkcat[$i] eq 'closure'
1190 || $chkcat[$i] eq 'srt'
1191 || $chkcat[$i] eq 'infotbl'
1192 || $chkcat[$i] eq 'entry') { # do them in that order
1193 $symb = $chksymb[$i];
1196 if ( defined($closurechk{$symb}) ) {
1197 print OUTASM $T_HDR_closure;
1198 print OUTASM $chk[$closurechk{$symb}];
1199 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
1203 if ( defined($srtchk{$symb}) ) {
1204 print OUTASM $T_HDR_relrodata;
1205 print OUTASM $chk[$srtchk{$symb}];
1206 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
1210 if ( defined($infochk{$symb}) ) {
1212 print OUTASM $T_HDR_info;
1213 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
1215 # entry code will be put here!
1217 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
1221 if ( defined($entrychk{$symb}) ) {
1223 $c = $chk[$entrychk{$symb}];
1225 # If this is an entry point with an info table,
1226 # eliminate the entry symbol and all directives involving it.
1227 if (defined($infochk{$symb}) && $TargetPlatform !~ /^ia64-/) {
1229 foreach $l (split(/\n/,$c)) {
1230 next if $l =~ /^.*$symb_(entry|ret)${T_POST_LBL}/;
1232 # If we have .type/.size direrctives involving foo_entry,
1233 # then make them refer to foo_info instead. The information
1234 # in these directives is used by the cachegrind annotator,
1235 # so it is worthwhile keeping.
1236 if ($l =~ /^\s*\.(type|size).*$symb_(entry|ret)/) {
1237 $l =~ s/$symb(_entry|_ret)/${symb}_info/g;
1241 next if $l =~ /^\s*\..*$symb.*\n?/;
1244 $c = join("\n",@o) . "\n";
1247 print OUTASM $T_HDR_entry;
1249 &print_doctored($c, 1); # NB: the 1!!!
1251 $chkcat[$entrychk{$symb}] = 'DONE ALREADY';
1254 } elsif ( $chkcat[$i] eq 'vector' ) {
1255 $symb = $chksymb[$i];
1258 if ( defined($vectorchk{$symb}) ) {
1259 print OUTASM $T_HDR_vector;
1260 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1262 # direct return code will be put here!
1263 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1265 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1266 # Alphas: the commented nop is for the splitter, to ensure
1267 # that no module ends with a label as the very last
1268 # thing. (The linker will adjust the label to point
1269 # to the first code word of the next module linked in,
1270 # even if alignment constraints cause the label to move!)
1272 print OUTASM "\t# nop\n";
1275 } elsif ( $chkcat[$i] eq 'rodata' ) {
1276 print OUTASM $T_HDR_rodata;
1277 print OUTASM $chk[$i];
1278 $chkcat[$i] = 'DONE ALREADY';
1279 } elsif ( $chkcat[$i] eq 'relrodata' ) {
1280 print OUTASM $T_HDR_relrodata;
1281 print OUTASM $chk[$i];
1282 $chkcat[$i] = 'DONE ALREADY';
1283 } elsif ( $chkcat[$i] eq 'toc' ) {
1284 # silly optimisation to print tocs, since they come in groups...
1285 print OUTASM $T_HDR_toc;
1287 while ($chkcat[$j] eq 'toc')
1288 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1291 print OUTASM $chk[$j];
1293 $chkcat[$j] = 'DONE ALREADY';
1297 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && $chkcat[$i] eq 'dyld' ) {
1298 # powerpc-apple: dynamic linker stubs
1299 if($chk[$i] !~ /\.indirect_symbol ___DISCARD__/)
1300 { # print them out unchanged, but remove the stubs for __DISCARD__
1301 print OUTASM $chk[$i];
1304 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1308 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1311 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1312 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1317 sub hppa_mash_prologue { # OK, epilogue, too
1320 # toss all prologue stuff
1321 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1323 # Lie about our .CALLINFO
1324 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1331 # toss all epilogue stuff
1332 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1334 # Sorry; we moved the _info stuff to the code segment.
1335 s/_info,DATA/_info,CODE/g;
1342 sub print_doctored {
1343 local($_, $need_fallthru_patch) = @_;
1345 if ( $TargetPlatform !~ /^i386-/
1346 || ! /^\t[a-z]/ # no instructions in here, apparently
1347 || /^${T_US}__stginit_[A-Za-z0-9_]+${T_POST_LBL}/) {
1351 # OK, must do some x86 **HACKING**
1353 local($entry_patch) = '';
1354 local($exit_patch) = '';
1356 # gotta watch out for weird instructions that
1357 # invisibly smash various regs:
1358 # rep* %ecx used for counting
1359 # scas* %edi used for destination index
1360 # cmps* %e[sd]i used for indices
1361 # loop* %ecx used for counting
1366 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1368 # * GCC used an "STG reg" for its own purposes
1370 # * some secret uses of machine reg, requiring STG reg
1371 # to be saved/restored
1373 # The most dangerous "GCC uses" of an "STG reg" are when
1374 # the reg holds the target of a jmp -- it's tricky to
1375 # insert the patch-up code before we get to the target!
1376 # So here we change the jmps:
1378 # --------------------------------------------------------
1379 # it can happen that we have jumps of the form...
1380 # jmp *<something involving %esp>
1382 # jmp <something involving another naughty register...>
1384 # a reasonably-common case is:
1386 # movl $_blah,<bad-reg>
1389 s/^\tmovl\s+\$${T_US}(.*),\s*(\%e[acd]x)\n\tjmp\s+\*\2/\tjmp $T_US$1/g;
1393 # movl -4(%ebx), %eax
1398 s/^\tmovl\s+(-?\d*\(\%e(bx|si)\)),\s*(\%e[acd]x)\n\tjmp\s+\*\3/\tjmp\t\*$1/g;
1400 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1401 s/^\tmovl\s+(.*),\s*\%esi\n\tjmp\s+\*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1402 s/^\tjmp\s+\*(.*\(.*\%esi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1403 s/^\tjmp\s+\*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1404 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1405 if /(jmp|call)\s+.*\%esi/;
1407 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1408 s/^\tmovl\s+(.*),\s*\%edi\n\tjmp\s+\*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1409 s/^\tjmp\s+\*(.*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1410 s/^\tjmp\s+\*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1411 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1412 if /(jmp|call)\s+.*\%edi/;
1415 # OK, now we can decide what our patch-up code is going to
1418 # Offsets into register table - you'd better update these magic
1419 # numbers should you change its contents!
1420 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1421 local($OFFSET_Hp)=88;
1423 # Note funky ".=" stuff; we're *adding* to these _patch guys
1424 if ( $StolenX86Regs <= 2
1425 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1426 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1427 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1429 # nothing for call_{entry,exit} because %esi is callee-save
1431 if ( $StolenX86Regs <= 3
1432 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1433 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1434 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1436 # nothing for call_{entry,exit} because %edi is callee-save
1439 # --------------------------------------------------------
1440 # next, here we go with non-%esp patching!
1442 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1444 # Before calling GC we must set up the exit condition before the call
1445 # and entry condition when we come back
1447 # fix _all_ non-local jumps:
1449 s/^\tjmp\s+\*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1450 s/^\tjmp\s+${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1452 s/^(\tjmp\s+.*\n)/$exit_patch$1/g; # here's the fix...
1454 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1455 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1457 if ($StolenX86Regs == 2 ) {
1458 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1459 if /^\t(jmp|call)\s+.*\%e(si|di)/;
1460 } elsif ($StolenX86Regs == 3 ) {
1461 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1462 if /^\t(jmp|call)\s+.*\%edi/;
1465 # --------------------------------------------------------
1466 # that's it -- print it
1468 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1472 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1473 print OUTASM $exit_patch;
1474 # ToDo: make it not print if there is a "jmp" at the end
1480 sub init_FUNNY_THINGS {
1481 %KNOWN_FUNNY_THING = (
1483 # "${T_US}stg_.*{T_POST_LBL}", 1,
1488 The following table reversal is used for both info tables and return
1489 vectors. In both cases, we remove the first entry from the table,
1490 reverse the table, put the label at the end, and paste some code
1491 (that which is normally referred to by the first entry in the table)
1492 right after the table itself. (The code pasting is done elsewhere.)
1496 local($symb, $tbl, $discard1) = @_;
1498 return ($tbl) if ($TargetPlatform =~ /^ia64-/);
1500 local($before) = '';
1502 local(@imports) = (); # hppa only
1505 local(@lines) = split(/\n/, $tbl);
1508 # Deal with the header...
1509 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1510 $label .= $lines[$i] . "\n",
1511 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info${T_POST_LBL}$/o
1512 || $lines[$i] =~ /${T_DOT_GLOBAL}/o
1513 || $lines[$i] =~ /^${T_US}\S+_vtbl${T_POST_LBL}$/o;
1515 $before .= $lines[$i] . "\n"; # otherwise...
1519 $infoname =~ s/(.|\n)*^([A-Za-z0-9_]+_info)${T_POST_LBL}$(.|\n)*/\2/;
1521 # Grab the table data...
1522 if ( $TargetPlatform !~ /^hppa/ ) {
1523 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1525 # Convert addresses of SRTs, slow entrypoints and large bitmaps
1526 # to offsets (relative to the info label),
1527 # in order to support position independent code.
1528 $line =~ s/$infoname/0/
1529 || $line =~ s/([A-Za-z0-9_]+_srtd)$/\1 - $infoname/
1530 || $line =~ s/([A-Za-z0-9_]+_srt(\+\d+)?)$/\1 - $infoname/
1531 || $line =~ s/([A-Za-z0-9_]+_slow)$/\1 - $infoname/
1532 || $line =~ s/([A-Za-z0-9_]+_btm)$/\1 - $infoname/
1533 || $line =~ s/([A-Za-z0-9_]+_alt)$/\1 - $infoname/
1534 || $line =~ s/([A-Za-z0-9_]+_dflt)$/\1 - $infoname/
1535 || $line =~ s/([A-Za-z0-9_]+_ret)$/\1 - $infoname/;
1536 push(@words, $line);
1538 } else { # hppa weirdness
1539 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+(${T_DOT_WORD}|\.IMPORT)/; $i++) {
1540 # FIXME: the RTS now expects offsets instead of addresses
1541 # for all labels in info tables.
1542 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1543 push(@imports, $lines[$i]);
1545 # We don't use HP's ``function pointers''
1546 # We just use labels in code space, like normal people
1547 $lines[$i] =~ s/P%//;
1548 push(@words, $lines[$i]);
1553 # Now throw away any initial zero word from the table. This is a hack
1554 # that lets us reduce the size of info tables when the SRT field is not
1555 # needed: see comments StgFunInfoTable in InfoTables.h.
1557 # The .zero business is for Linux/ELF.
1558 # The .skip business is for Sparc/Solaris/ELF.
1559 # The .blockz business is for HPPA.
1561 # if ($words[0] =~ /^\t?(${T_DOT_WORD}\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1566 for (; $i <= $#lines; $i++) {
1567 $after .= $lines[$i] . "\n";
1570 # Alphas: If we have anonymous text (not part of a procedure), the
1571 # linker may complain about missing exception information. Bleh.
1572 # To suppress this, we place a .ent/.end pair around the code.
1573 # At the same time, we have to be careful and not enclose any leading
1574 # .file/.loc directives.
1575 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1576 local ($ident) = $1;
1577 $before =~ s/^((\s*\.(file|loc)\s+[^\n]*\n)*)/$1\t.ent $ident\n/;
1578 $after .= "\t.end $ident\n";
1581 # Alphas: The heroic Simon Marlow found a bug in the Digital UNIX
1582 # assembler (!) wherein .quad constants inside .text sections are
1583 # first narrowed to 32 bits then sign-extended back to 64 bits.
1584 # This obviously screws up our 64-bit bitmaps, so we work around
1585 # the bug by replacing .quad with .align 3 + .long + .long [ccshan]
1586 if ( $TargetPlatform =~ /^alpha-/ ) {
1588 if (/^\s*\.quad\s+([-+0-9].*\S)\s*$/ && length $1 >= 10) {
1589 local ($number) = $1;
1590 if ($number =~ /^([-+])?(0x?)?([0-9]+)$/) {
1591 local ($sign, $base, $digits) = ($1, $2, $3);
1592 $base = (10, 8, 16)[length $base];
1593 local ($hi, $lo) = (0, 0);
1594 foreach $i (split(//, $digits)) {
1595 $j = $lo * $base + $i;
1596 $lo = $j % 4294967296;
1597 $hi = $hi * $base + ($j - $lo) / 4294967296;
1599 ($hi, $lo) = (4294967295 - $hi, 4294967296 - $lo)
1601 $_ = "\t.align 3\n\t.long $lo\n\t.long $hi\n";
1602 # printf STDERR "TURNING %s into 0x %08x %08x\n", $number, $hi, $lo;
1604 print STDERR "Cannot handle \".quad $number\" in info table\n";
1612 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1613 . join("\n", @words) . "\n"
1616 # print STDERR "before=$before\n";
1617 # print STDERR "label=$label\n";
1618 # print STDERR "words=",(reverse @words),"\n";
1619 # print STDERR "after=$after\n";
1625 The HP is a major nuisance. The threaded code mangler moved info
1626 tables from data space to code space, but unthreaded code in the RTS
1627 still has references to info tables in data space. Since the HP
1628 linker is very precise about where symbols live, we need to patch the
1629 references in the unthreaded RTS as well.
1632 sub mini_mangle_asm_hppa {
1633 local($in_asmf, $out_asmf) = @_;
1635 open(INASM, "< $in_asmf")
1636 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1637 open(OUTASM,"> $out_asmf")
1638 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1641 s/_info,DATA/_info,CODE/; # Move _info references to code space
1647 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1648 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1654 sub tidy_up_and_die {
1655 local($return_val, $msg) = @_;
1657 exit (($return_val == 0) ? 0 : 1);