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 ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|gnu|freebsd|netbsd|openbsd)$/ ) {
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|gnu|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';
227 $T_HDR_literal16 = "\.section\t\.rodata.cst16\n\t.align 16\n";
228 $T_HDR_literal = "\.section\t\.rodata\n";
230 $T_HDR_misc = "\.text\n\t\.align 8\n";
231 $T_HDR_data = "\.data\n\t\.align 8\n";
232 $T_HDR_rodata = "\.section\t\.rodata\n\t\.align 8\n";
234 # the assembler on x86_64/Linux refuses to generate code for
236 # where x is in the text section and y in the rodata section.
237 # It works if y is in the text section, though. This is probably
238 # going to cause difficulties for PIC, I imagine.
239 $T_HDR_relrodata= "\.text\n\t\.align 8\n";
241 $T_HDR_closure = "\.data\n\t\.align 8\n";
242 $T_HDR_info = "\.text\n\t\.align 8\n";
243 $T_HDR_entry = "\.text\n\t\.align 8\n";
244 $T_HDR_vector = "\.text\n\t\.align 8\n";
246 #--------------------------------------------------------#
247 } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
249 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
250 $T_US = '_'; # _ if symbols have an underscore on the front
251 $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
252 $T_CONST_LBL = '^LC(\d+):$';
255 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
256 $T_COPY_DIRVS = '\.(globl|proc|stab)';
258 $T_DOT_WORD = '\.long';
259 $T_DOT_GLOBAL = '\.globl';
260 $T_HDR_literal = "\.text\n\t\.even\n";
261 $T_HDR_misc = "\.text\n\t\.even\n";
262 $T_HDR_data = "\.data\n\t\.even\n";
263 $T_HDR_rodata = "\.text\n\t\.even\n";
264 $T_HDR_closure = "\.data\n\t\.even\n";
265 $T_HDR_info = "\.text\n\t\.even\n";
266 $T_HDR_entry = "\.text\n\t\.even\n";
267 $T_HDR_vector = "\.text\n\t\.even\n";
269 #--------------------------------------------------------#
270 } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
272 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
273 $T_US = ''; # _ if symbols have an underscore on the front
274 $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
275 $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
278 $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
279 $T_COPY_DIRVS = '\.(globl|ent)';
281 $T_DOT_WORD = '\.word';
282 $T_DOT_GLOBAL = '^\t\.globl';
283 $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
284 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
285 $T_HDR_data = "\t\.data\n\t\.align 2\n";
286 $T_HDR_rodata = "\t\.rdata\n\t\.align 2\n";
287 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
288 $T_HDR_info = "\t\.text\n\t\.align 2\n";
289 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
290 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
292 #--------------------------------------------------------#
293 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ ) {
294 # Apple PowerPC Darwin/MacOS X.
295 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
296 $T_US = '_'; # _ if symbols have an underscore on the front
297 $T_PRE_APP = 'DOESNT APPLY'; # regexp that says what comes before APP/NO_APP
298 $T_CONST_LBL = '^\LC\d+:'; # regexp for what such a lbl looks like
301 $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)';
302 $T_COPY_DIRVS = '\.(globl|lcomm)';
304 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
305 $T_DOT_GLOBAL = '\.globl';
306 $T_HDR_toc = "\.toc\n";
307 $T_HDR_literal = "\t\.const\n\t\.align 2\n";
308 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
309 $T_HDR_data = "\t\.data\n\t\.align 2\n";
310 $T_HDR_rodata = "\t\.const\n\t\.align 2\n";
311 $T_HDR_relrodata= "\t\.const_data\n\t\.align 2\n";
312 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
313 $T_HDR_info = "\t\.text\n\t\.align 2\n";
314 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
315 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
317 #--------------------------------------------------------#
318 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ ) {
320 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
321 $T_US = ''; # _ if symbols have an underscore on the front
322 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
323 $T_CONST_LBL = '^\.LC\d+:'; # regexp for what such a lbl looks like
326 $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)';
327 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
329 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
330 $T_DOT_GLOBAL = '\.globl';
331 $T_HDR_toc = "\.toc\n";
332 $T_HDR_literal = "\t\.section\t.rodata\n\t\.align 2\n";
333 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
334 $T_HDR_data = "\t\.data\n\t\.align 2\n";
335 $T_HDR_rodata = "\t\.section\t.rodata\n\t\.align 2\n";
336 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
337 $T_HDR_info = "\t\.text\n\t\.align 2\n";
338 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
339 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
341 #--------------------------------------------------------#
342 } elsif ( $TargetPlatform =~ /^powerpc64-.*-linux/ ) {
344 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
345 $T_US = '\.'; # _ if symbols have an underscore on the front
346 $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
347 $T_CONST_LBL = '^\.LC\d+:'; # regexp for what such a lbl looks like
350 $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)';
351 $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)';
353 $T_DOT_WORD = '\.(long|short|byte|fill|space)';
354 $T_DOT_GLOBAL = '\.globl';
355 $T_HDR_toc = "\.toc\n";
356 $T_HDR_literal = "\t\.section\t\".toc\",\"aw\"\n";
357 $T_HDR_misc = "\t\.text\n\t\.align 2\n";
358 $T_HDR_data = "\t\.data\n\t\.align 2\n";
359 $T_HDR_rodata = "\t\.section\t.rodata\n\t\.align 2\n";
360 $T_HDR_closure = "\t\.data\n\t\.align 2\n";
361 $T_HDR_info = "\t\.text\n\t\.align 2\n";
362 $T_HDR_entry = "\t\.text\n\t\.align 2\n";
363 $T_HDR_vector = "\t\.text\n\t\.align 2\n";
365 #--------------------------------------------------------#
366 } elsif ( $TargetPlatform =~ /^sparc-.*-(solaris2|openbsd)/ ) {
368 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
369 $T_US = ''; # _ if symbols have an underscore on the front
370 $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
371 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
374 $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)';
375 $T_COPY_DIRVS = '\.(global|proc|stab)';
377 $T_DOT_WORD = '\.(long|word|byte|half|skip|uahalf|uaword)';
378 $T_DOT_GLOBAL = '^\t\.global';
379 $T_HDR_literal = "\.text\n\t\.align 8\n";
380 $T_HDR_misc = "\.text\n\t\.align 4\n";
381 $T_HDR_data = "\.data\n\t\.align 8\n";
382 $T_HDR_rodata = "\.text\n\t\.align 4\n";
383 $T_HDR_closure = "\.data\n\t\.align 4\n";
384 $T_HDR_info = "\.text\n\t\.align 4\n";
385 $T_HDR_entry = "\.text\n\t\.align 4\n";
386 $T_HDR_vector = "\.text\n\t\.align 4\n";
388 #--------------------------------------------------------#
389 } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
391 $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
392 $T_US = '_'; # _ if symbols have an underscore on the front
393 $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
394 $T_CONST_LBL = '^LC(\d+):$';
397 $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
398 $T_COPY_DIRVS = '\.(global|proc|stab)';
400 $T_DOT_WORD = '\.word';
401 $T_DOT_GLOBAL = '^\t\.global';
402 $T_HDR_literal = "\.text\n\t\.align 8\n";
403 $T_HDR_misc = "\.text\n\t\.align 4\n";
404 $T_HDR_data = "\.data\n\t\.align 8\n";
405 $T_HDR_rodata = "\.text\n\t\.align 4\n";
406 $T_HDR_closure = "\.data\n\t\.align 4\n";
407 $T_HDR_info = "\.text\n\t\.align 4\n";
408 $T_HDR_entry = "\.text\n\t\.align 4\n";
409 $T_HDR_vector = "\.text\n\t\.align 4\n";
411 #--------------------------------------------------------#
412 } elsif ( $TargetPlatform =~ /^sparc-.*-linux/ ) {
413 $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
414 $T_US = ''; # _ if symbols have an underscore on the front
415 $T_PRE_APP = '#'; # regexp that says what comes before APP/NO_APP
416 # Probably doesn't apply anyway
417 $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
420 $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)';
421 $T_COPY_DIRVS = '\.(global|globl|proc|stab)';
423 $T_DOT_WORD = '\.(long|word|nword|xword|byte|half|short|skip|uahalf|uaword)';
424 $T_DOT_GLOBAL = '^\t\.global';
425 $T_HDR_literal = "\.text\n\t\.align 8\n";
426 $T_HDR_misc = "\.text\n\t\.align 4\n";
427 $T_HDR_data = "\.data\n\t\.align 8\n";
428 $T_HDR_rodata = "\.text\n\t\.align 4\n";
429 $T_HDR_closure = "\.data\n\t\.align 4\n";
430 $T_HDR_info = "\.text\n\t\.align 4\n";
431 $T_HDR_entry = "\.text\n\t\.align 4\n";
432 $T_HDR_vector = "\.text\n\t\.align 4\n";
434 #--------------------------------------------------------#
436 print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
440 if($T_HDR_relrodata eq "") {
442 # relrodata defaults to rodata.
443 $T_HDR_relrodata = $T_HDR_rodata;
447 print STDERR "T_STABBY: $T_STABBY\n";
448 print STDERR "T_US: $T_US\n";
449 print STDERR "T_PRE_APP: $T_PRE_APP\n";
450 print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
451 print STDERR "T_POST_LBL: $T_POST_LBL\n";
452 if ( $TargetPlatform =~ /^i386-/ ) {
453 print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
454 print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
455 print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
457 print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
458 print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
459 print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
460 print STDERR "T_HDR_literal: $T_HDR_literal\n";
461 print STDERR "T_HDR_misc: $T_HDR_misc\n";
462 print STDERR "T_HDR_data: $T_HDR_data\n";
463 print STDERR "T_HDR_rodata: $T_HDR_rodata\n";
464 print STDERR "T_HDR_closure: $T_HDR_closure\n";
465 print STDERR "T_HDR_info: $T_HDR_info\n";
466 print STDERR "T_HDR_entry: $T_HDR_entry\n";
467 print STDERR "T_HDR_vector: $T_HDR_vector\n";
473 %************************************************************************
475 \subsection{Mangle away}
477 %************************************************************************
481 local($in_asmf, $out_asmf) = @_;
483 # multi-line regexp matching:
488 &init_TARGET_STUFF();
489 &init_FUNNY_THINGS();
491 open(INASM, "< $in_asmf")
492 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
493 open(OUTASM,"> $out_asmf")
494 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
496 # read whole file, divide into "chunks":
497 # record some info about what we've found...
499 @chk = (); # contents of the chunk
500 $numchks = 0; # number of them
501 @chkcat = (); # what category of thing in each chunk
502 @chksymb = (); # what symbol(base) is defined in this chunk
503 %entrychk = (); # ditto, its entry code
504 %closurechk = (); # ditto, the (static) closure
505 %srtchk = (); # ditto, its SRT (for top-level things)
506 %infochk = (); # given a symbol base, say what chunk its info tbl is in
507 %vectorchk = (); # ditto, return vector table
508 $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
510 $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
513 tr/\r//d if $TargetPlatform =~ /-mingw32$/; # In case Perl doesn't convert line endings
514 next if $T_STABBY && /^\.stab.*${T_US}__stg_split_marker/o;
515 next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
516 next if /^\t\.def.*endef$/;
517 next if /${T_PRE_APP}(NO_)?APP/o;
518 next if /^;/ && $TargetPlatform =~ /^hppa/;
520 next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|ia64)-/;
522 if ( $TargetPlatform =~ /^mips-/
523 && /^\t\.(globl\S+\.text|comm\t)/ ) {
524 $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
525 # Treat .comm variables as data. These show up in two (known) places:
527 # - the module_registered variable used in the __stginit fragment.
528 # even though these are declared static and initialised, gcc 3.3
529 # likes to make them .comm, presumably to save space in the
532 # - global variables used to pass arguments from C to STG in
533 # a foreign export. (is this still true? --SDM)
535 } elsif ( /^\t\.comm.*$/ ) {
537 $chkcat[$i] = 'data';
540 # Labels ending "_str": these are literal strings.
541 } elsif ( /^${T_US}([A-Za-z0-9_]+)_str${T_POST_LBL}$/ ) {
543 $chkcat[$i] = 'rodata';
545 } elsif ( $TargetPlatform =~ /-darwin/
546 && (/^\s*\.subsections_via_symbols/
547 ||/^\s*\.no_dead_strip.*/)) {
548 # Don't allow Apple's linker to do any dead-stripping of symbols
549 # in this file, because it will mess up info-tables in mangled
551 # The .no_dead_strip directives are actually put there by
552 # the gcc3 "used" attribute on entry points.
554 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && (
555 /^\s*\.picsymbol_stub/
556 || /^\s*\.section __TEXT,__picsymbol_stub1,.*/
557 || /^\s*\.section __TEXT,__picsymbolstub1,.*/
558 || /^\s*\.symbol_stub/
559 || /^\s*\.section __TEXT,__symbol_stub1,.*/
560 || /^\s*\.section __TEXT,__symbolstub1,.*/
561 || /^\s*\.lazy_symbol_pointer/
562 || /^\s*\.non_lazy_symbol_pointer/ ))
565 $chkcat[$i] = 'dyld';
568 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && $chkcat[$i] eq 'dyld' && /^\s*\.data/)
569 { # non_lazy_symbol_ptrs that point to local symbols
571 $chkcat[$i] = 'dyld';
574 } elsif ( /^\s+/ ) { # most common case first -- a simple line!
575 # duplicated from the bottom
579 } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
580 # Alphas: Local labels not to be confused with new chunks
582 # NB: all the rest start with a non-space
584 } elsif ( $TargetPlatform =~ /^mips-/
585 && /^\d+:/ ) { # a funny-looking very-local label
588 } elsif ( /$T_CONST_LBL/o ) {
590 $chkcat[$i] = 'literal';
593 } elsif ( /^${T_US}__stg_split_marker(\d*)${T_POST_LBL}$/o ) {
595 $chkcat[$i] = 'splitmarker';
598 } elsif ( /^${T_US}([A-Za-z0-9_]+)_info${T_POST_LBL}$/o ) {
601 $chkcat[$i] = 'infotbl';
602 $chksymb[$i] = $symb;
604 die "Info table already? $symb; $i\n" if defined($infochk{$symb});
606 $infochk{$symb} = $i;
608 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(entry|ret)${T_POST_LBL}$/o ) {
610 $chkcat[$i] = 'entry';
615 } elsif ( /^${T_US}([A-Za-z0-9_]+)_closure${T_POST_LBL}$/o ) {
617 $chkcat[$i] = 'closure';
620 $closurechk{$1} = $i;
622 } elsif ( /^${T_US}([A-Za-z0-9_]+)_srt${T_POST_LBL}$/o ) {
629 } elsif ( /^${T_US}([A-Za-z0-9_]+)_ct${T_POST_LBL}$/o ) {
631 $chkcat[$i] = 'data';
634 } elsif ( /^${T_US}(stg_ap_stack_entries|stg_stack_save_entries|stg_arg_bitmaps)${T_POST_LBL}$/o ) {
636 $chkcat[$i] = 'data';
639 } elsif ( /^(${T_US}__gnu_compiled_c|gcc2_compiled\.)${T_POST_LBL}/o ) {
642 } elsif ( /^${T_US}[A-Za-z0-9_]+\.\d+${T_POST_LBL}$/o
643 || /^${T_US}.*_CAT${T_POST_LBL}$/o # PROF: _entryname_CAT
644 || /^${T_US}.*_done${T_POST_LBL}$/o # PROF: _module_done
645 || /^${T_US}_module_registered${T_POST_LBL}$/o # PROF: _module_registered
648 $chkcat[$i] = 'data';
651 } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
656 } elsif ( /^${T_US}([A-Za-z0-9_]+)_cc(s)?${T_POST_LBL}$/o ) {
657 # all CC_ symbols go in the data section...
659 $chkcat[$i] = 'data';
662 } elsif ( /^${T_US}([A-Za-z0-9_]+)_(alt|dflt)${T_POST_LBL}$/o ) {
664 $chkcat[$i] = 'misc';
666 } elsif ( /^${T_US}([A-Za-z0-9_]+)_vtbl${T_POST_LBL}$/o ) {
668 $chkcat[$i] = 'vector';
673 } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
674 && /^[A-Za-z0-9][A-Za-z0-9_]*:/ ) {
675 # Some Solaris system headers contain function definitions (as
676 # opposed to mere prototypes), which end up in the .hc file when
677 # a Haskell module foreign imports the corresponding system
678 # functions (most notably stat()). We put them into the text
679 # segment. Note that this currently does not extend to function
680 # names starting with an underscore.
683 $chkcat[$i] = 'misc';
686 } elsif ( /^${T_US}[A-Za-z0-9_]/o
687 && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
689 && ( $TargetPlatform !~ /^powerpc64/ # we need to avoid local labels in this case
690 || ! /^\.L\d+:$/ ) ) {
697 /^${T_US}stg_.*${T_POST_LBL}$/o # RTS internals
698 || /^${T_US}__stg_.*${T_POST_LBL}$/o # more RTS internals
699 || /^${T_US}__fexp_.*${T_POST_LBL}$/o # foreign export
700 || /^${T_US}.*_slow${T_POST_LBL}$/o # slow entry
701 || /^${T_US}__stginit.*${T_POST_LBL}$/o # __stginit<module>
702 || /^${T_US}.*_btm${T_POST_LBL}$/o # large bitmaps
703 || /^${T_US}.*_fast${T_POST_LBL}$/o # primops
704 || /^_uname:/o # x86/Solaris2
707 $chkcat[$i] = 'misc';
709 /^${T_US}.*_srtd${T_POST_LBL}$/o # large bitmaps
710 || /^${T_US}.*_closure_tbl${T_POST_LBL}$/o # closure tables
713 $chkcat[$i] = 'relrodata';
716 print STDERR "Warning: retaining unknown function \`$thing' in output from C compiler\n";
717 $chkcat[$i] = 'unknown';
720 } elsif ( $TargetPlatform =~ /^powerpc-.*-linux/ && /^\.LCTOC1 = /o ) {
721 # PowerPC Linux's large-model PIC (-fPIC) generates a gobal offset
722 # table "by hand". Be sure to copy it over.
723 # Note that this label and all entries in the table should actually
724 # go into the .got2 section, but it isn't easy to distinguish them
725 # from other constant literals (.LC\d+), so we just put everything
728 $chkcat[$i] = 'literal';
729 $chksymb[$i] = 'LCTOC1';
730 } else { # simple line (duplicated at the top)
735 $numchks = $#chk + 1;
737 # open CHUNKS, ">/tmp/chunks1" or die "Cannot open /tmp/chunks1: $!\n";
738 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
741 # the division into chunks is imperfect;
742 # we throw some things over the fence into the next
745 # also, there are things we would like to know
746 # about the whole module before we start spitting
749 local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
750 local($FIRST_TOSSABLE ) = ($TargetPlatform =~ /^(hppa|mips-)/) ? 1 : 0;
752 # print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
754 # Alphas: NB: we start meddling at chunk 1, not chunk 0
755 # The first ".rdata" is quite magical; as of GCC 2.7.x, it
756 # spits a ".quad 0" in after the very first ".rdata"; we
757 # detect this special case (tossing the ".quad 0")!
758 local($magic_rdata_seen) = 0;
760 # HPPAs, MIPSen: also start medding at chunk 1
762 for ($i = $FIRST_TOSSABLE; $i < $numchks; $i++) {
763 $c = $chk[$i]; # convenience copy
765 # print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
767 # toss all prologue stuff; HPPA is pretty weird
769 $c = &hppa_mash_prologue($c) if $TargetPlatform =~ /^hppa-/;
771 # be slightly paranoid to make sure there's
772 # nothing surprising in there
773 if ( $c =~ /--- BEGIN ---/ ) {
774 if (($p, $r) = split(/--- BEGIN ---/, $c)) {
776 # remove junk whitespace around the split point
780 if ($TargetPlatform =~ /^i386-/) {
781 if ($p =~ /^\tsubl\s+\$(\d+),\s*\%esp\n/) {
783 die "Error: reserved stack space exceeded!\n Possible workarounds: compile with -fasm, or try another version of gcc.\n"
787 $p =~ s/^\tpushl\s+\%edi\n//;
788 $p =~ s/^\tpushl\s+\%esi\n//;
789 $p =~ s/^\tpushl\s+\%ebx\n//;
790 $p =~ s/^\tmovl\s+\%esi,\s*\d*\(\%esp\)\n//;
791 $p =~ s/^\tmovl\s+\%edi,\s*\d*\(\%esp\)\n//;
792 $p =~ s/^\tsubl\s+\$\d+,\s*\%esp\n//;
793 $p =~ s/^\tmovl\s+\$\d+,\s*\%eax\n\tcall\s+__alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
795 # GCC 3.1 is in the habit of adding spurious writes to the
796 # stack in the prologue. Just to be on the safe side,
797 # chuck these over the fence into the main code.
798 while ($p =~ /^\tmovl\s+\$\d+,\s*\d*\(\%esp\)\n/) {
799 # print "Spurious instruction: $&";
804 } elsif ($TargetPlatform =~ /^x86_64-/) {
805 $p =~ s/^\tpushq\s+\%r(bx|bp|12|13|14)\n//g;
806 $p =~ s/^\tmovq\s+\%r(bx|bp|12|13|14),\s*\d*\(\%rsp\)\n//g;
807 $p =~ s/^\tsubq\s+\$\d+,\s*\%rsp\n//;
809 } elsif ($TargetPlatform =~ /^ia64-/) {
810 $p =~ s/^\t\.prologue .*\n//;
811 $p =~ s/^\t\.save ar\.pfs, r\d+\n\talloc r\d+ = ar\.pfs, 0, 3[12], \d+, 0\n//;
812 $p =~ s/^\t\.fframe \d+\n\tadds r12 = -\d+, r12\n//;
813 $p =~ s/^\t\.save rp, r\d+\n\tmov r\d+ = b0\n//;
814 $p =~ s/^\t\.(mii|mmi)\n//g; # bundling is no longer sensible
815 $p =~ s/^\t;;\n//g; # discard stops
816 $p =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
818 # GCC 3.3 saves r1 in the prologue, move this to the body
819 if ($p =~ /^\tmov r\d+ = r1\n/) {
823 } elsif ($TargetPlatform =~ /^m68k-/) {
824 $p =~ s/^\tlink a6,#-?\d.*\n//;
825 $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
826 # The above showed up in the asm code,
827 # so I added it here.
828 # I hope it's correct.
830 $p =~ s/^\tmovel d2,sp\@-\n//;
831 $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
832 $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
833 } elsif ($TargetPlatform =~ /^mips-/) {
834 # the .frame/.mask/.fmask that we use is the same
835 # as that produced by GCC for miniInterpret; this
836 # gives GDB some chance of figuring out what happened
837 $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
838 $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
839 $p =~ s/^\t\.(mask|fmask).*\n//g;
840 $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
841 $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
842 $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
843 $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
844 $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
845 $p =~ s/__FRAME__/$FRAME/;
846 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
848 $pcrel_label =~ s/(.|\n)*^(\"?L\d+\$pb\"?):\n(.|\n)*/$2/ or $pcrel_label = "";
850 $p =~ s/^\tmflr r0\n//;
851 $p =~ s/^\tbl saveFP # f\d+\n//;
852 $p =~ s/^\tbl saveFP ; save f\d+-f\d+\n//;
853 $p =~ s/^\"?L\d+\$pb\"?:\n//;
854 $p =~ s/^\tstmw r\d+,-\d+\(r1\)\n//;
855 $p =~ s/^\tstfd f\d+,-\d+\(r1\)\n//g;
856 $p =~ s/^\tstw r0,\d+\(r1\)\n//g;
857 $p =~ s/^\tstwu r1,-\d+\(r1\)\n//;
858 $p =~ s/^\tstw r\d+,-\d+\(r1\)\n//g;
859 $p =~ s/^\tbcl 20,31,L\d+\$pb\n//;
860 $p =~ s/^L\d+\$pb:\n//;
861 $p =~ s/^\tmflr r31\n//;
863 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
864 # under some circumstances, only when generating position dependent code.
865 # I have no idea why, and I don't think it is necessary, so let's toss it.
866 $p =~ s/^\tli r\d+,0\n//g;
867 $p =~ s/^\tstw r\d+,\d+\(r1\)\n//g;
868 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/) {
869 $p =~ s/^\tmflr 0\n//;
870 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
871 $p =~ s/^\tstfd \d+,\d+\(1\)\n//g;
872 $p =~ s/^\tstw r0,8\(1\)\n//;
873 $p =~ s/^\tstwu 1,-\d+\(1\)\n//;
874 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
876 # GCC's "large-model" PIC (-fPIC)
878 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
880 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
881 $p =~ s/^.LCF\d+:\n//;
882 $p =~ s/^\tmflr 30\n//;
883 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
884 $p =~ s/^\tadd 30,0,30\n//;
886 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
887 # under some circumstances, only when generating position dependent code.
888 # I have no idea why, and I don't think it is necessary, so let's toss it.
889 $p =~ s/^\tli \d+,0\n//g;
890 $p =~ s/^\tstw \d+,\d+\(1\)\n//g;
891 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
892 $p =~ s/^\tmr 31,1\n//;
893 $p =~ s/^\tmflr 0\n//;
894 $p =~ s/^\tstmw \d+,\d+\(1\)\n//;
895 $p =~ s/^\tstfd \d+,-?\d+\(1\)\n//g;
896 $p =~ s/^\tstd r0,8\(1\)\n//;
897 $p =~ s/^\tstdu 1,-\d+\(1\)\n//;
898 $p =~ s/^\tstd \d+,-?\d+\(1\)\n//g;
900 # GCC's "large-model" PIC (-fPIC)
902 $pcrel_label =~ s/(.|\n)*^.LCF(\d+):\n(.|\n)*/$2/ or $pcrel_label = "";
904 $p =~ s/^\tbcl 20,31,.LCF\d+\n//;
905 $p =~ s/^.LCF\d+:\n//;
906 $p =~ s/^\tmflr 30\n//;
907 $p =~ s/^\tlwz 0,\.LCL\d+-\.LCF\d+\(30\)\n//;
908 $p =~ s/^\tadd 30,0,30\n//;
910 # This is bad: GCC 3 seems to zero-fill some local variables in the prologue
911 # under some circumstances, only when generating position dependent code.
912 # I have no idea why, and I don't think it is necessary, so let's toss it.
913 $p =~ s/^\tli \d+,0\n//g;
914 $p =~ s/^\tstd \d+,\d+\(1\)\n//g;
916 print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
919 # HWL HACK: dont die, just print a warning
920 #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/;
921 die "Prologue junk?: $p\n" if $p =~ /^\s+[^\s\.]/;
923 # For PIC, we want to keep part of the prologue
924 if ($TargetPlatform =~ /^powerpc-apple-.*/ && $pcrel_label ne "") {
925 # Darwin: load the current instruction pointer into register r31
926 $p .= "bcl 20,31,$pcrel_label\n";
927 $p .= "$pcrel_label:\n";
928 $p .= "\tmflr r31\n";
929 } elsif ($TargetPlatform =~ /^powerpc-.*-linux/ && $pcrel_label ne "") {
930 # Linux: load the GOT pointer into register 30
931 $p .= "\tbcl 20,31,.LCF$pcrel_label\n";
932 $p .= ".LCF$pcrel_label:\n";
934 $p .= "\tlwz 0,.LCL$pcrel_label-.LCF$pcrel_label(30)\n";
935 $p .= "\tadd 30,0,30\n";
938 # glue together what's left
943 if ( $TargetPlatform =~ /^mips-/ ) {
944 # MIPS: first, this basic sequence may occur "--- END ---" or not
945 $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
948 # toss all epilogue stuff; again, paranoidly
949 if ( $c =~ /--- END ---/ ) {
950 if (($r, $e) = split(/--- END ---/, $c)) {
951 if ($TargetPlatform =~ /^i386-/) {
953 $e =~ s/^\tpopl\s+\%edi\n//;
954 $e =~ s/^\tpopl\s+\%esi\n//;
955 $e =~ s/^\tpopl\s+\%edx\n//;
956 $e =~ s/^\tpopl\s+\%ecx\n//;
957 $e =~ s/^\taddl\s+\$\d+,\s*\%esp\n//;
958 $e =~ s/^\tsubl\s+\$-\d+,\s*\%esp\n//;
959 } elsif ($TargetPlatform =~ /^ia64-/) {
960 $e =~ s/^\tmov ar\.pfs = r\d+\n//;
961 $e =~ s/^\tmov b0 = r\d+\n//;
962 $e =~ s/^\t\.restore sp\n\tadds r12 = \d+, r12\n//;
963 $e =~ s/^\tbr\.ret\.sptk\.many b0\n//;
964 $e =~ s/^\t\.(mii|mmi|mib)\n//g; # bundling is no longer sensible
965 $e =~ s/^\t;;\n//g; # discard stops - stop at end of body is sufficient
966 $e =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments
967 } elsif ($TargetPlatform =~ /^m68k-/) {
968 $e =~ s/^\tunlk a6\n//;
970 } elsif ($TargetPlatform =~ /^mips-/) {
971 $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
972 $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
973 $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
974 $e =~ s/^\tj\t\$31\n//;
975 } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) {
976 $e =~ s/^\taddi r1,r1,\d+\n//;
977 $e =~ s/^\tlwz r\d+,\d+\(r1\)\n//;
978 $e =~ s/^\tlmw r\d+,-\d+\(r1\)\n//;
979 $e =~ s/^\tmtlr r0\n//;
981 $e =~ s/^\tb restFP ;.*\n//;
982 } elsif ($TargetPlatform =~ /^powerpc64-.*-linux/) {
983 $e =~ s/^\tmr 3,0\n//;
984 $e =~ s/^\taddi 1,1,\d+\n//;
985 $e =~ s/^\tld 0,16\(1\)\n//;
986 $e =~ s/^\tmtlr 0\n//;
988 # callee-save registers
989 $e =~ s/^\tld \d+,-?\d+\(1\)\n//g;
990 $e =~ s/^\tlfd \d+,-?\d+\(1\)\n//g;
992 # get rid of the debug junk along with the blr
993 $e =~ s/^\tblr\n\t.long .*\n\t.byte .*\n//;
995 # incase we missed it with the last one get the blr alone
998 print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
1001 print STDERR "WARNING: Epilogue junk?: $e\n" if $e =~ /^\t\s*[^\.\s\n]/;
1003 # glue together what's left
1005 $c =~ s/\n\t\n/\n/; # junk blank line
1009 # On SPARCs, we don't do --- BEGIN/END ---, we just
1010 # toss the register-windowing save/restore/ret* instructions
1011 # directly unless they've been generated by function definitions in header
1013 if ( $TargetPlatform =~ /^sparc-/ ) {
1014 if ( ! ( $TargetPlatform =~ /solaris2$/ && $chkcat[$i] eq 'unknown' )) {
1015 $c =~ s/^\t(save.*|restore.*|ret|retl)\n//g;
1017 # throw away PROLOGUE comments
1018 $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
1021 # On Alphas, the prologue mangling is done a little later (below)
1023 # toss all calls to __DISCARD__
1024 $c =~ s/^\t(call|jbsr|jal)\s+${T_US}__DISCARD__\n//go;
1025 $c =~ s/^\tjsr\s+\$26\s*,\s*${T_US}__DISCARD__\n//go if $TargetPlatform =~ /^alpha-/;
1026 $c =~ s/^\tbl\s+L___DISCARD__\$stub\n//go if $TargetPlatform =~ /^powerpc-apple-.*/;
1027 $c =~ s/^\tbl\s+__DISCARD__(\@plt)?\n//go if $TargetPlatform =~ /^powerpc-.*-linux/;
1028 $c =~ s/^\tbl\s+\.__DISCARD__\n\s+nop\n//go if $TargetPlatform =~ /^powerpc64-.*-linux/;
1030 # IA64: mangle tailcalls into jumps here
1031 if ($TargetPlatform =~ /^ia64-/) {
1032 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/) {
1033 # Eek, the gcc optimiser is getting smarter... if we see a jump to the --- TAILCALL ---
1034 # marker then we reapply the substitution at the source sites
1035 $c =~ s/^\tbr \.L$2\n/\t--- TAILCALL ---\n/g if ($2);
1039 # MIPS: that may leave some gratuitous asm macros around
1040 # (no harm done; but we get rid of them to be tidier)
1041 $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/
1042 if $TargetPlatform =~ /^mips-/;
1044 # toss stack adjustment after DoSparks
1045 $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
1046 if $TargetPlatform =~ /^m68k-/; # this looks old...
1048 if ( $TargetPlatform =~ /^alpha-/ &&
1049 ! $magic_rdata_seen &&
1050 $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
1051 $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
1052 $magic_rdata_seen = 1;
1055 # pick some end-things and move them to the next chunk
1057 # pin a funny end-thing on (for easier matching):
1058 $c .= 'FUNNY#END#THING';
1060 while ( $c =~ /${T_MOVE_DIRVS}FUNNY#END#THING/o ) {
1064 # on x86 we try not to copy any directives into a literal
1065 # chunk, rather we keep looking for the next real chunk. This
1066 # is because we get things like
1068 # .globl blah_closure
1074 if ( $TargetPlatform =~ /^(i386|sparc|powerpc)/ && $to_move =~ /${T_COPY_DIRVS}/ ) {
1076 while ( $j < $numchks && $chk[$j] =~ /$T_CONST_LBL/) {
1079 if ( $j < $numchks ) {
1080 $chk[$j] = $to_move . $chk[$j];
1084 elsif ( $i < ($numchks - 1)
1085 && ( $to_move =~ /${T_COPY_DIRVS}/
1086 || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
1087 $chk[$i + 1] = $to_move . $chk[$i + 1];
1088 # otherwise they're tossed
1091 $c =~ s/${T_MOVE_DIRVS}FUNNY#END#THING/FUNNY#END#THING/o;
1094 if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) {
1096 # toss all prologue stuff, except for loading gp, and the ..ng address
1097 unless ($c =~ /\.ent.*\n\$.*\.\.ng:/) {
1098 if (($p, $r) = split(/^\t\.prologue/, $c)) {
1099 if (($keep, $junk) = split(/\.\.ng:/, $p)) {
1100 $keep =~ s/^\t\.frame.*\n/\t.frame \$30,0,\$26,0\n/;
1101 $keep =~ s/^\t\.(mask|fmask).*\n//g;
1102 $c = $keep . "..ng:\n";
1104 print STDERR "malformed code block ($ent)?\n"
1107 $c .= "\t.prologue" . $r;
1111 $c =~ s/FUNNY#END#THING//;
1113 # print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
1115 $chk[$i] = $c; # update w/ convenience copy
1118 # open CHUNKS, ">/tmp/chunks2" or die "Cannot open /tmp/chunks2: $!\n";
1119 # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
1122 if ( $TargetPlatform =~ /^alpha-/ ) {
1123 # print out the header stuff first
1124 $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
1125 print OUTASM $chk[0];
1127 } elsif ( $TargetPlatform =~ /^hppa/ ) {
1128 print OUTASM $chk[0];
1130 } elsif ( $TargetPlatform =~ /^mips-/ ) {
1131 $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
1133 # get rid of horrible "<dollar>Revision: .*$" strings
1134 local(@lines0) = split(/\n/, $chk[0]);
1136 while ( $z <= $#lines0 ) {
1137 if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
1140 while ( $z <= $#lines0 ) {
1142 last if $lines0[$z] =~ /[,\t]0x0$/;
1148 $chk[0] = join("\n", @lines0);
1149 $chk[0] =~ s/\n\n+/\n/;
1150 print OUTASM $chk[0];
1153 # print out all the literal strings next
1154 for ($i = 0; $i < $numchks; $i++) {
1155 if ( $chkcat[$i] eq 'literal' ) {
1157 # HACK: try to detect 16-byte constants and align them
1158 # on a 16-byte boundary. x86_64 sometimes needs 128-bit
1159 # aligned constants.
1160 if ( $TargetPlatform =~ /^x86_64/ ) {
1162 if ($z =~ /(\.long.*\n.*\.long.*\n.*\.long.*\n.*\.long|\.quad.*\n.*\.quad)/) {
1163 print OUTASM $T_HDR_literal16;
1165 print OUTASM $T_HDR_literal;
1168 print OUTASM $T_HDR_literal;
1171 print OUTASM $chk[$i];
1172 print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
1174 $chkcat[$i] = 'DONE ALREADY';
1178 # on the HPPA, print out all the bss next
1179 if ( $TargetPlatform =~ /^hppa/ ) {
1180 for ($i = 1; $i < $numchks; $i++) {
1181 if ( $chkcat[$i] eq 'bss' ) {
1182 print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
1183 print OUTASM $chk[$i];
1185 $chkcat[$i] = 'DONE ALREADY';
1190 for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
1191 # print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
1193 next if $chkcat[$i] eq 'DONE ALREADY';
1195 if ( $chkcat[$i] eq 'misc' || $chkcat[$i] eq 'unknown' ) {
1196 if ($chk[$i] ne '') {
1197 print OUTASM $T_HDR_misc;
1198 &print_doctored($chk[$i], 0);
1201 } elsif ( $chkcat[$i] eq 'toss' ) {
1202 print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
1204 } elsif ( $chkcat[$i] eq 'data' ) {
1205 if ($chk[$i] ne '') {
1206 print OUTASM $T_HDR_data;
1207 print OUTASM $chk[$i];
1210 } elsif ( $chkcat[$i] eq 'splitmarker' ) {
1211 # we can just re-constitute this one...
1212 # NB: we emit _three_ underscores no matter what,
1213 # so ghc-split doesn't have to care.
1214 print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
1216 } elsif ( $chkcat[$i] eq 'closure'
1217 || $chkcat[$i] eq 'srt'
1218 || $chkcat[$i] eq 'infotbl'
1219 || $chkcat[$i] eq 'entry') { # do them in that order
1220 $symb = $chksymb[$i];
1223 if ( defined($closurechk{$symb}) ) {
1224 print OUTASM $T_HDR_closure;
1225 print OUTASM $chk[$closurechk{$symb}];
1226 $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
1230 if ( defined($srtchk{$symb}) ) {
1231 print OUTASM $T_HDR_relrodata;
1232 print OUTASM $chk[$srtchk{$symb}];
1233 $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
1237 if ( defined($infochk{$symb}) ) {
1239 print OUTASM $T_HDR_info;
1240 print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
1242 # entry code will be put here!
1244 $chkcat[$infochk{$symb}] = 'DONE ALREADY';
1248 if ( defined($entrychk{$symb}) ) {
1250 $c = $chk[$entrychk{$symb}];
1252 # If this is an entry point with an info table,
1253 # eliminate the entry symbol and all directives involving it.
1254 if (defined($infochk{$symb}) && $TargetPlatform !~ /^ia64-/) {
1256 foreach $l (split(/\n/,$c)) {
1257 next if $l =~ /^.*$symb_(entry|ret)${T_POST_LBL}/;
1259 # If we have .type/.size direrctives involving foo_entry,
1260 # then make them refer to foo_info instead. The information
1261 # in these directives is used by the cachegrind annotator,
1262 # so it is worthwhile keeping.
1263 if ($l =~ /^\s*\.(type|size).*$symb_(entry|ret)/) {
1264 $l =~ s/$symb(_entry|_ret)/${symb}_info/g;
1268 next if $l =~ /^\s*\..*$symb.*\n?/;
1271 $c = join("\n",@o) . "\n";
1274 print OUTASM $T_HDR_entry;
1276 &print_doctored($c, 1); # NB: the 1!!!
1278 $chkcat[$entrychk{$symb}] = 'DONE ALREADY';
1281 } elsif ( $chkcat[$i] eq 'vector' ) {
1282 $symb = $chksymb[$i];
1285 if ( defined($vectorchk{$symb}) ) {
1286 print OUTASM $T_HDR_vector;
1287 print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
1289 # direct return code will be put here!
1290 $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
1292 } elsif ( $TargetPlatform =~ /^alpha-/ ) {
1293 # Alphas: the commented nop is for the splitter, to ensure
1294 # that no module ends with a label as the very last
1295 # thing. (The linker will adjust the label to point
1296 # to the first code word of the next module linked in,
1297 # even if alignment constraints cause the label to move!)
1299 print OUTASM "\t# nop\n";
1302 } elsif ( $chkcat[$i] eq 'rodata' ) {
1303 print OUTASM $T_HDR_rodata;
1304 print OUTASM $chk[$i];
1305 $chkcat[$i] = 'DONE ALREADY';
1306 } elsif ( $chkcat[$i] eq 'relrodata' ) {
1307 print OUTASM $T_HDR_relrodata;
1308 print OUTASM $chk[$i];
1309 $chkcat[$i] = 'DONE ALREADY';
1310 } elsif ( $chkcat[$i] eq 'toc' ) {
1311 # silly optimisation to print tocs, since they come in groups...
1312 print OUTASM $T_HDR_toc;
1314 while ($chkcat[$j] eq 'toc')
1315 { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
1318 print OUTASM $chk[$j];
1320 $chkcat[$j] = 'DONE ALREADY';
1324 } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && $chkcat[$i] eq 'dyld' ) {
1325 # powerpc-apple: dynamic linker stubs
1326 if($chk[$i] !~ /\.indirect_symbol ___DISCARD__/)
1327 { # print them out unchanged, but remove the stubs for __DISCARD__
1328 print OUTASM $chk[$i];
1331 &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
1335 print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
1338 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1339 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1344 sub hppa_mash_prologue { # OK, epilogue, too
1347 # toss all prologue stuff
1348 s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
1350 # Lie about our .CALLINFO
1351 s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
1358 # toss all epilogue stuff
1359 s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
1361 # Sorry; we moved the _info stuff to the code segment.
1362 s/_info,DATA/_info,CODE/g;
1369 sub print_doctored {
1370 local($_, $need_fallthru_patch) = @_;
1372 if ( $TargetPlatform !~ /^i386-/
1373 || ! /^\t[a-z]/ # no instructions in here, apparently
1374 || /^${T_US}__stginit_[A-Za-z0-9_]+${T_POST_LBL}/) {
1378 # OK, must do some x86 **HACKING**
1380 local($entry_patch) = '';
1381 local($exit_patch) = '';
1383 # gotta watch out for weird instructions that
1384 # invisibly smash various regs:
1385 # rep* %ecx used for counting
1386 # scas* %edi used for destination index
1387 # cmps* %e[sd]i used for indices
1388 # loop* %ecx used for counting
1393 # * use of STG reg [ nn(%ebx) ] where no machine reg avail
1395 # * GCC used an "STG reg" for its own purposes
1397 # * some secret uses of machine reg, requiring STG reg
1398 # to be saved/restored
1400 # The most dangerous "GCC uses" of an "STG reg" are when
1401 # the reg holds the target of a jmp -- it's tricky to
1402 # insert the patch-up code before we get to the target!
1403 # So here we change the jmps:
1405 # --------------------------------------------------------
1406 # it can happen that we have jumps of the form...
1407 # jmp *<something involving %esp>
1409 # jmp <something involving another naughty register...>
1411 # a reasonably-common case is:
1413 # movl $_blah,<bad-reg>
1416 s/^\tmovl\s+\$${T_US}(.*),\s*(\%e[acd]x)\n\tjmp\s+\*\2/\tjmp $T_US$1/g;
1420 # movl -4(%ebx), %eax
1425 s/^\tmovl\s+(-?\d*\(\%e(bx|si)\)),\s*(\%e[acd]x)\n\tjmp\s+\*\3/\tjmp\t\*$1/g;
1427 if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
1428 s/^\tmovl\s+(.*),\s*\%esi\n\tjmp\s+\*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1429 s/^\tjmp\s+\*(.*\(.*\%esi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1430 s/^\tjmp\s+\*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
1431 die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
1432 if /(jmp|call)\s+.*\%esi/;
1434 if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
1435 s/^\tmovl\s+(.*),\s*\%edi\n\tjmp\s+\*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1436 s/^\tjmp\s+\*(.*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
1437 s/^\tjmp\s+\*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
1438 die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
1439 if /(jmp|call)\s+.*\%edi/;
1442 # OK, now we can decide what our patch-up code is going to
1445 # Offsets into register table - you'd better update these magic
1446 # numbers should you change its contents!
1447 # local($OFFSET_R1)=0; No offset for R1 in new RTS.
1448 local($OFFSET_Hp)=88;
1450 # Note funky ".=" stuff; we're *adding* to these _patch guys
1451 if ( $StolenX86Regs <= 2
1452 && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
1453 $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
1454 $exit_patch .= "\tmovl (\%ebx),\%esi\n";
1456 # nothing for call_{entry,exit} because %esi is callee-save
1458 if ( $StolenX86Regs <= 3
1459 && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
1460 $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
1461 $exit_patch .= "\tmovl ${OFFSET_Hp}(\%ebx),\%edi\n";
1463 # nothing for call_{entry,exit} because %edi is callee-save
1466 # --------------------------------------------------------
1467 # next, here we go with non-%esp patching!
1469 s/^(\t[a-z])/$entry_patch$1/; # before first instruction
1471 # Before calling GC we must set up the exit condition before the call
1472 # and entry condition when we come back
1474 # fix _all_ non-local jumps:
1476 s/^\tjmp\s+\*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
1477 s/^\tjmp\s+${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
1479 s/^(\tjmp\s+.*\n)/$exit_patch$1/g; # here's the fix...
1481 s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
1482 s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
1484 if ($StolenX86Regs == 2 ) {
1485 die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
1486 if /^\t(jmp|call)\s+.*\%e(si|di)/;
1487 } elsif ($StolenX86Regs == 3 ) {
1488 die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
1489 if /^\t(jmp|call)\s+.*\%edi/;
1492 # --------------------------------------------------------
1493 # that's it -- print it
1495 #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
1499 if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
1500 print OUTASM $exit_patch;
1501 # ToDo: make it not print if there is a "jmp" at the end
1507 sub init_FUNNY_THINGS {
1508 %KNOWN_FUNNY_THING = (
1510 # "${T_US}stg_.*{T_POST_LBL}", 1,
1515 The following table reversal is used for both info tables and return
1516 vectors. In both cases, we remove the first entry from the table,
1517 reverse the table, put the label at the end, and paste some code
1518 (that which is normally referred to by the first entry in the table)
1519 right after the table itself. (The code pasting is done elsewhere.)
1523 local($symb, $tbl, $discard1) = @_;
1525 return ($tbl) if ($TargetPlatform =~ /^ia64-/);
1527 local($before) = '';
1529 local(@imports) = (); # hppa only
1532 local(@lines) = split(/\n/, $tbl);
1535 # Deal with the header...
1536 for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1537 $label .= $lines[$i] . "\n",
1538 next if $lines[$i] =~ /^[A-Za-z0-9_]+_info${T_POST_LBL}$/o
1539 || $lines[$i] =~ /${T_DOT_GLOBAL}/o
1540 || $lines[$i] =~ /^${T_US}\S+_vtbl${T_POST_LBL}$/o;
1542 $before .= $lines[$i] . "\n"; # otherwise...
1546 $infoname =~ s/(.|\n)*^([A-Za-z0-9_]+_info)${T_POST_LBL}$(.|\n)*/\2/;
1548 # Grab the table data...
1549 if ( $TargetPlatform !~ /^hppa/ ) {
1550 for ( ; $i <= $#lines && $lines[$i] =~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
1552 # Convert addresses of SRTs, slow entrypoints and large bitmaps
1553 # to offsets (relative to the info label),
1554 # in order to support position independent code.
1555 $line =~ s/$infoname/0/
1556 || $line =~ s/([A-Za-z0-9_]+_srtd)$/\1 - $infoname/
1557 || $line =~ s/([A-Za-z0-9_]+_srt(\+\d+)?)$/\1 - $infoname/
1558 || $line =~ s/([A-Za-z0-9_]+_slow)$/\1 - $infoname/
1559 || $line =~ s/([A-Za-z0-9_]+_btm)$/\1 - $infoname/
1560 || $line =~ s/([A-Za-z0-9_]+_alt)$/\1 - $infoname/
1561 || $line =~ s/([A-Za-z0-9_]+_dflt)$/\1 - $infoname/
1562 || $line =~ s/([A-Za-z0-9_]+_ret)$/\1 - $infoname/;
1563 push(@words, $line);
1565 } else { # hppa weirdness
1566 for ( ; $i <= $#lines && $lines[$i] =~ /^\s+(${T_DOT_WORD}|\.IMPORT)/; $i++) {
1567 # FIXME: the RTS now expects offsets instead of addresses
1568 # for all labels in info tables.
1569 if ($lines[$i] =~ /^\s+\.IMPORT/) {
1570 push(@imports, $lines[$i]);
1572 # We don't use HP's ``function pointers''
1573 # We just use labels in code space, like normal people
1574 $lines[$i] =~ s/P%//;
1575 push(@words, $lines[$i]);
1580 # Now throw away any initial zero word from the table. This is a hack
1581 # that lets us reduce the size of info tables when the SRT field is not
1582 # needed: see comments StgFunInfoTable in InfoTables.h.
1584 # The .zero business is for Linux/ELF.
1585 # The .skip business is for Sparc/Solaris/ELF.
1586 # The .blockz business is for HPPA.
1588 # if ($words[0] =~ /^\t?(${T_DOT_WORD}\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) {
1593 for (; $i <= $#lines; $i++) {
1594 $after .= $lines[$i] . "\n";
1597 # Alphas: If we have anonymous text (not part of a procedure), the
1598 # linker may complain about missing exception information. Bleh.
1599 # To suppress this, we place a .ent/.end pair around the code.
1600 # At the same time, we have to be careful and not enclose any leading
1601 # .file/.loc directives.
1602 if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
1603 local ($ident) = $1;
1604 $before =~ s/^((\s*\.(file|loc)\s+[^\n]*\n)*)/$1\t.ent $ident\n/;
1605 $after .= "\t.end $ident\n";
1608 # Alphas: The heroic Simon Marlow found a bug in the Digital UNIX
1609 # assembler (!) wherein .quad constants inside .text sections are
1610 # first narrowed to 32 bits then sign-extended back to 64 bits.
1611 # This obviously screws up our 64-bit bitmaps, so we work around
1612 # the bug by replacing .quad with .align 3 + .long + .long [ccshan]
1613 if ( $TargetPlatform =~ /^alpha-/ ) {
1615 if (/^\s*\.quad\s+([-+0-9].*\S)\s*$/ && length $1 >= 10) {
1616 local ($number) = $1;
1617 if ($number =~ /^([-+])?(0x?)?([0-9]+)$/) {
1618 local ($sign, $base, $digits) = ($1, $2, $3);
1619 $base = (10, 8, 16)[length $base];
1620 local ($hi, $lo) = (0, 0);
1621 foreach $i (split(//, $digits)) {
1622 $j = $lo * $base + $i;
1623 $lo = $j % 4294967296;
1624 $hi = $hi * $base + ($j - $lo) / 4294967296;
1626 ($hi, $lo) = (4294967295 - $hi, 4294967296 - $lo)
1628 $_ = "\t.align 3\n\t.long $lo\n\t.long $hi\n";
1629 # printf STDERR "TURNING %s into 0x %08x %08x\n", $number, $hi, $lo;
1631 print STDERR "Cannot handle \".quad $number\" in info table\n";
1639 . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
1640 . join("\n", @words) . "\n"
1643 # print STDERR "before=$before\n";
1644 # print STDERR "label=$label\n";
1645 # print STDERR "words=",(reverse @words),"\n";
1646 # print STDERR "after=$after\n";
1652 The HP is a major nuisance. The threaded code mangler moved info
1653 tables from data space to code space, but unthreaded code in the RTS
1654 still has references to info tables in data space. Since the HP
1655 linker is very precise about where symbols live, we need to patch the
1656 references in the unthreaded RTS as well.
1659 sub mini_mangle_asm_hppa {
1660 local($in_asmf, $out_asmf) = @_;
1662 open(INASM, "< $in_asmf")
1663 || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
1664 open(OUTASM,"> $out_asmf")
1665 || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
1668 s/_info,DATA/_info,CODE/; # Move _info references to code space
1674 close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
1675 close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
1681 sub tidy_up_and_die {
1682 local($return_val, $msg) = @_;
1684 exit (($return_val == 0) ? 0 : 1);