… | |
… | |
4 | |
4 | |
5 | =head1 SYNOPSIS |
5 | =head1 SYNOPSIS |
6 | |
6 | |
7 | use Faster; |
7 | use Faster; |
8 | |
8 | |
|
|
9 | perl -MFaster ... |
|
|
10 | |
9 | =head1 DESCRIPTION |
11 | =head1 DESCRIPTION |
10 | |
12 | |
|
|
13 | This module implements a very simple-minded "JIT" (or actually AIT, ahead |
|
|
14 | of time compiler). It works by more or less translating every function it |
|
|
15 | sees into a C program, compiling it and then replacing the function by the |
|
|
16 | compiled code. |
|
|
17 | |
|
|
18 | As a result, startup times are immense, as every function might lead to a |
|
|
19 | full-blown compilation. |
|
|
20 | |
|
|
21 | The speed improvements are also not great, you can expect 20% or so on |
|
|
22 | average, for code that runs very often. The reason for this is that data |
|
|
23 | handling is mostly being done by the same old code, it just gets called |
|
|
24 | a bit faster. Regexes and string operations won't get faster. Airhtmetic |
|
|
25 | doresn't become any faster. Just the operands and other stuff is put on |
|
|
26 | the stack faster, and the opcodes themselves have a bit less overhead. |
|
|
27 | |
|
|
28 | Faster is in the early stages of development. Due to its design its |
|
|
29 | relatively safe to use (it will either work or simply slowdown the program |
|
|
30 | immensely, but rarely cause bugs). |
|
|
31 | |
|
|
32 | More intelligent algorithms (loop optimisation, type inference) could |
|
|
33 | improve that easily, but requires a much more elaborate presentation and |
|
|
34 | optimiser than what is in place. There are no plans to improve Faster in |
|
|
35 | this way, yet, but it would provide a reasonably good place to start. |
|
|
36 | |
|
|
37 | Usage is very easy, just C<use Faster> and every function called from then |
|
|
38 | on will be compiled. |
|
|
39 | |
|
|
40 | Right now, Faster can leave lots of F<*.c> and F<*.so> files in your |
|
|
41 | F<$FASTER_CACHEDIR> (by default F<$HOME/.perl-faster-cache>), and it will |
|
|
42 | even create those temporary files in an insecure manner, so watch out. |
|
|
43 | |
11 | =over 4 |
44 | =over 4 |
12 | |
45 | |
13 | =cut |
46 | =cut |
14 | |
47 | |
15 | package Faster; |
48 | package Faster; |
|
|
49 | |
|
|
50 | no warnings; |
16 | |
51 | |
17 | use strict; |
52 | use strict; |
18 | use Config; |
53 | use Config; |
19 | use B (); |
54 | use B (); |
|
|
55 | use DynaLoader (); |
20 | use Digest::MD5 (); |
56 | use Digest::MD5 (); |
21 | use DynaLoader (); |
57 | use Storable (); |
|
|
58 | use Fcntl (); |
22 | |
59 | |
23 | BEGIN { |
60 | BEGIN { |
24 | our $VERSION = '0.01'; |
61 | our $VERSION = '0.01'; |
25 | |
62 | |
26 | require XSLoader; |
63 | require XSLoader; |
27 | XSLoader::load __PACKAGE__, $VERSION; |
64 | XSLoader::load __PACKAGE__, $VERSION; |
28 | } |
65 | } |
|
|
66 | |
|
|
67 | my $CACHEDIR = |
|
|
68 | $ENV{FASTER_CACHE} |
|
|
69 | || (exists $ENV{HOME} && "$ENV{HOME}/.perl-faster-cache") |
|
|
70 | || do { |
|
|
71 | require File::Temp; |
|
|
72 | File::Temp::tempdir (CLEANUP => 1) |
|
|
73 | }; |
29 | |
74 | |
30 | my $COMPILE = "$Config{cc} -c -I$Config{archlibexp}/CORE $Config{optimize} $Config{ccflags} $Config{cccdlflags}"; |
75 | my $COMPILE = "$Config{cc} -c -I$Config{archlibexp}/CORE $Config{optimize} $Config{ccflags} $Config{cccdlflags}"; |
31 | my $LINK = "$Config{ld} $Config{ldflags} $Config{lddlflags} $Config{ccdlflags}"; |
76 | my $LINK = "$Config{ld} $Config{ldflags} $Config{lddlflags} $Config{ccdlflags}"; |
32 | my $LIBS = "$Config{libs}"; |
77 | my $LIBS = "$Config{libs}"; |
33 | my $_o = $Config{_o}; |
78 | my $_o = $Config{_o}; |
… | |
… | |
35 | |
80 | |
36 | # we don't need no steenking PIC on x86 |
81 | # we don't need no steenking PIC on x86 |
37 | $COMPILE =~ s/-f(?:PIC|pic)//g |
82 | $COMPILE =~ s/-f(?:PIC|pic)//g |
38 | if $Config{archname} =~ /^(i[3456]86)-/; |
83 | if $Config{archname} =~ /^(i[3456]86)-/; |
39 | |
84 | |
40 | my $opt_assert = 1; |
85 | my $opt_assert = $ENV{FASTER_DEBUG} > 1; |
|
|
86 | my $verbose = $ENV{FASTER_VERBOSE}+0; |
|
|
87 | |
|
|
88 | warn "Faster: CACHEDIR is $CACHEDIR\n" if $verbose > 2; |
41 | |
89 | |
42 | our $source; |
90 | our $source; |
43 | |
91 | |
44 | my @ops; |
92 | our @ops; |
45 | my $op; |
93 | our $insn; |
|
|
94 | our $op; |
46 | my $op_name; |
95 | our $op_name; |
47 | my @loop; |
96 | our @op_loop; |
|
|
97 | our %op_regcomp; |
48 | |
98 | |
49 | my %flag; |
99 | # ops that cause immediate return to the interpreter |
|
|
100 | my %f_unsafe = map +($_ => undef), qw( |
|
|
101 | leavesub leavesublv return |
|
|
102 | goto last redo next |
|
|
103 | eval flip leaveeval entertry |
|
|
104 | formline grepstart mapstart |
|
|
105 | substcont entereval require |
|
|
106 | ); |
50 | |
107 | |
51 | # complex flag steting is no longer required, rewrite this ugly code |
108 | # ops with known stack extend behaviour |
52 | for (split /\n/, <<EOF) { |
109 | # the values given are maximum values |
53 | leavesub unsafe |
110 | my %extend = ( |
54 | leavesublv unsafe |
111 | pushmark => 0, |
55 | return unsafe |
112 | nextstate => 0, # might reduce the stack |
56 | flip unsafe |
113 | unstack => 0, |
57 | goto unsafe |
114 | enter => 0, |
58 | last unsafe |
|
|
59 | redo unsafe |
|
|
60 | next unsafe |
|
|
61 | eval unsafe |
|
|
62 | leaveeval unsafe |
|
|
63 | entertry unsafe |
|
|
64 | formline unsafe |
|
|
65 | grepstart unsafe |
|
|
66 | mapstart unsafe |
|
|
67 | substcont unsafe |
|
|
68 | entereval unsafe noasync todo |
|
|
69 | require unsafe |
|
|
70 | |
115 | |
71 | mapstart noasync |
116 | stringify => 0, |
72 | grepstart noasync |
117 | not => 0, |
73 | match noasync |
118 | and => 0, |
|
|
119 | or => 0, |
|
|
120 | gvsv => 0, |
|
|
121 | rv2gv => 0, |
|
|
122 | preinc => 0, |
|
|
123 | predec => 0, |
|
|
124 | postinc => 0, |
|
|
125 | postdec => 0, |
|
|
126 | aelem => 0, |
|
|
127 | helem => 0, |
|
|
128 | qr => 1, #??? |
|
|
129 | pushre => 1, |
|
|
130 | gv => 1, |
|
|
131 | aelemfast => 1, |
|
|
132 | aelem => 0, |
|
|
133 | padsv => 1, |
|
|
134 | const => 1, |
|
|
135 | pop => 1, |
|
|
136 | shift => 1, |
|
|
137 | eq => -1, |
|
|
138 | ne => -1, |
|
|
139 | gt => -1, |
|
|
140 | lt => -1, |
|
|
141 | ge => -1, |
|
|
142 | lt => -1, |
|
|
143 | cond_expr => -1, |
|
|
144 | add => -1, |
|
|
145 | subtract => -1, |
|
|
146 | multiply => -1, |
|
|
147 | divide => -1, |
|
|
148 | aassign => 0, |
|
|
149 | sassign => -2, |
|
|
150 | method => 0, |
|
|
151 | method_named => 1, |
|
|
152 | ); |
74 | |
153 | |
75 | last noasync |
154 | # ops that do not need an ASYNC_CHECK |
76 | next noasync |
155 | my %f_noasync = map +($_ => undef), qw( |
77 | redo noasync |
156 | mapstart grepstart match entereval |
78 | seq noasync |
157 | enteriter entersub leaveloop |
79 | pushmark noasync extend=0 |
|
|
80 | padsv noasync extend=1 |
|
|
81 | padav noasync extend=1 |
|
|
82 | padhv noasync extend=1 |
|
|
83 | padany noasync extend=1 |
|
|
84 | entersub noasync |
|
|
85 | aassign noasync |
|
|
86 | sassign noasync |
|
|
87 | rv2av noasync |
|
|
88 | rv2cv noasync |
|
|
89 | rv2gv noasync |
|
|
90 | rv2hv noasync |
|
|
91 | refgen noasync |
|
|
92 | nextstate noasync |
|
|
93 | gv noasync |
|
|
94 | gvsv noasync |
|
|
95 | add noasync |
|
|
96 | subtract noasync |
|
|
97 | multiply noasync |
|
|
98 | divide noasync |
|
|
99 | complement noasync |
|
|
100 | cond_expr noasync |
|
|
101 | and noasync |
|
|
102 | or noasync |
|
|
103 | not noasync |
|
|
104 | defined noasync |
|
|
105 | method_named noasync |
|
|
106 | preinc noasync |
|
|
107 | postinc noasync |
|
|
108 | predec noasync |
|
|
109 | postdec noasync |
|
|
110 | stub noasync |
|
|
111 | unstack noasync |
|
|
112 | leaveloop noasync |
|
|
113 | aelem noasync |
|
|
114 | aelemfast noasync |
|
|
115 | helem noasync |
|
|
116 | pushre noasync |
|
|
117 | subst noasync |
|
|
118 | const noasync extend=1 |
|
|
119 | list noasync |
|
|
120 | join noasync |
|
|
121 | split noasync |
|
|
122 | concat noasync |
|
|
123 | push noasync |
|
|
124 | pop noasync |
|
|
125 | shift noasync |
|
|
126 | unshift noasync |
|
|
127 | length noasync |
|
|
128 | substr noasync |
|
|
129 | stringify noasync |
|
|
130 | eq noasync |
|
|
131 | ne noasync |
|
|
132 | gt noasync |
|
|
133 | lt noasync |
|
|
134 | ge noasync |
|
|
135 | le noasync |
|
|
136 | enteriter noasync |
|
|
137 | ord noasync |
|
|
138 | |
158 | |
139 | iter async |
159 | pushmark nextstate caller |
140 | EOF |
|
|
141 | my (undef, $op, @flags) = split /\s+/; |
|
|
142 | |
160 | |
143 | undef $flag{$_}{$op} |
161 | const stub unstack |
144 | for ("known", @flags); |
162 | last next redo goto seq |
145 | } |
163 | padsv padav padhv padany |
|
|
164 | aassign sassign orassign |
|
|
165 | rv2av rv2cv rv2gv rv2hv refgen |
|
|
166 | gv gvsv |
|
|
167 | add subtract multiply divide |
|
|
168 | complement cond_expr and or not |
|
|
169 | bit_and bit_or bit_xor |
|
|
170 | defined |
|
|
171 | method method_named bless |
|
|
172 | preinc postinc predec postdec |
|
|
173 | aelem aelemfast helem delete exists |
|
|
174 | pushre subst list lslice join split concat |
|
|
175 | length substr stringify ord |
|
|
176 | push pop shift unshift |
|
|
177 | eq ne gt lt ge le |
|
|
178 | regcomp regcreset regcmaybe |
|
|
179 | ); |
146 | |
180 | |
147 | my %callop = ( |
181 | my %callop = ( |
148 | entersub => "(PL_ppaddr [OP_ENTERSUB]) (aTHX)", |
182 | entersub => "(PL_op->op_ppaddr) (aTHX)", |
149 | mapstart => "Perl_pp_grepstart (aTHX)", |
183 | mapstart => "Perl_pp_grepstart (aTHX)", |
150 | ); |
184 | ); |
151 | |
185 | |
152 | sub callop { |
186 | sub callop { |
153 | $callop{$op_name} || "Perl_pp_$op_name (aTHX)" |
187 | $callop{$op_name} || "Perl_pp_$op_name (aTHX)" |
… | |
… | |
161 | sub out_callop { |
195 | sub out_callop { |
162 | assert "nextop == (OP *)$$op"; |
196 | assert "nextop == (OP *)$$op"; |
163 | $source .= " PL_op = nextop; nextop = " . (callop $op) . ";\n"; |
197 | $source .= " PL_op = nextop; nextop = " . (callop $op) . ";\n"; |
164 | } |
198 | } |
165 | |
199 | |
|
|
200 | sub out_cond_jump { |
|
|
201 | $source .= " if (nextop == (OP *)${$_[0]}L) goto op_${$_[0]};\n"; |
|
|
202 | } |
|
|
203 | |
166 | sub out_jump_next { |
204 | sub out_jump_next { |
|
|
205 | out_cond_jump $op_regcomp{$$op} |
|
|
206 | if $op_regcomp{$$op}; |
|
|
207 | |
167 | assert "nextop == (OP *)${$op->next}"; |
208 | assert "nextop == (OP *)${$op->next}"; |
168 | $source .= " goto op_${$op->next};\n"; |
209 | $source .= " goto op_${$op->next};\n"; |
169 | } |
210 | } |
170 | |
211 | |
171 | sub out_next { |
212 | sub out_next { |
… | |
… | |
175 | } |
216 | } |
176 | |
217 | |
177 | sub out_linear { |
218 | sub out_linear { |
178 | out_callop; |
219 | out_callop; |
179 | out_jump_next; |
220 | out_jump_next; |
180 | } |
|
|
181 | |
|
|
182 | sub out_cond_jump { |
|
|
183 | $source .= " if (nextop == (OP *)${$_[0]}L) goto op_${$_[0]};\n"; |
|
|
184 | } |
221 | } |
185 | |
222 | |
186 | sub op_entersub { |
223 | sub op_entersub { |
187 | out_callop; |
224 | out_callop; |
188 | $source .= " RUNOPS_TILL ((OP *)${$op->next}L);\n"; |
225 | $source .= " RUNOPS_TILL ((OP *)${$op->next}L);\n"; |
… | |
… | |
198 | |
235 | |
199 | out_next; |
236 | out_next; |
200 | } |
237 | } |
201 | |
238 | |
202 | sub op_pushmark { |
239 | sub op_pushmark { |
203 | $source .= " PUSHMARK (PL_stack_sp);\n"; |
240 | $source .= " faster_PUSHMARK (PL_stack_sp);\n"; |
204 | |
241 | |
205 | out_next; |
242 | out_next; |
206 | } |
243 | } |
207 | |
244 | |
208 | if ($Config{useithreads} ne "define") { |
245 | if ($Config{useithreads} ne "define") { |
209 | # disable optimisations on ithreads |
246 | # disable optimisations on ithreads |
210 | |
247 | |
211 | *op_const = sub { |
248 | *op_const = sub { |
212 | $source .= " { dSP; XPUSHs ((SV *)${$op->sv}L); PUTBACK; }\n"; |
249 | $source .= " { dSP; PUSHs ((SV *)${$op->sv}L); PUTBACK; }\n"; |
|
|
250 | |
|
|
251 | $ops[0]{follows_const}++ if @ops;#d# |
213 | |
252 | |
214 | out_next; |
253 | out_next; |
215 | }; |
254 | }; |
216 | |
255 | |
217 | *op_gv = \&op_const; |
256 | *op_gv = \&op_const; |
… | |
… | |
237 | if (!($op->flags & B::OPf_MOD)) { |
276 | if (!($op->flags & B::OPf_MOD)) { |
238 | $source .= " if (SvGMAGICAL (sv)) sv = sv_mortalcopy (sv);\n"; |
277 | $source .= " if (SvGMAGICAL (sv)) sv = sv_mortalcopy (sv);\n"; |
239 | } |
278 | } |
240 | |
279 | |
241 | $source .= " dSP;\n"; |
280 | $source .= " dSP;\n"; |
242 | $source .= " XPUSHs (sv);\n"; |
281 | $source .= " PUSHs (sv);\n"; |
243 | $source .= " PUTBACK;\n"; |
282 | $source .= " PUTBACK;\n"; |
244 | $source .= " }\n"; |
283 | $source .= " }\n"; |
245 | |
284 | |
246 | out_next; |
285 | out_next; |
247 | }; |
286 | }; |
248 | |
287 | |
249 | *op_gvsv = sub { |
288 | *op_gvsv = sub { |
250 | $source .= " {\n"; |
289 | $source .= " {\n"; |
251 | $source .= " dSP;\n"; |
290 | $source .= " dSP;\n"; |
252 | $source .= " EXTEND (SP, 1);\n"; |
|
|
253 | |
291 | |
254 | if ($op->private & B::OPpLVAL_INTRO) { |
292 | if ($op->private & B::OPpLVAL_INTRO) { |
255 | $source .= " PUSHs (save_scalar ((GV *)${$op->sv}L));\n"; |
293 | $source .= " PUSHs (save_scalar ((GV *)${$op->sv}L));\n"; |
256 | } else { |
294 | } else { |
257 | $source .= " PUSHs (GvSV ((GV *)${$op->sv}L));\n"; |
295 | $source .= " PUSHs (GvSV ((GV *)${$op->sv}L));\n"; |
… | |
… | |
317 | out_next; |
355 | out_next; |
318 | } |
356 | } |
319 | |
357 | |
320 | sub op_padsv { |
358 | sub op_padsv { |
321 | my $flags = $op->flags; |
359 | my $flags = $op->flags; |
322 | my $target = $op->targ; |
360 | my $padofs = "(PADOFFSET)" . $op->targ; |
323 | |
361 | |
324 | $source .= <<EOF; |
362 | $source .= <<EOF; |
325 | { |
363 | { |
326 | dSP; |
364 | dSP; |
327 | XPUSHs (PAD_SV ((PADOFFSET)$target)); |
365 | SV *sv = PAD_SVl ($padofs); |
|
|
366 | EOF |
|
|
367 | |
|
|
368 | if (($flags & B::OPf_MOD) && ($op->private & B::OPpLVAL_INTRO)) { |
|
|
369 | $source .= " SAVECLEARSV (PAD_SVl ($padofs));\n"; |
|
|
370 | $ops[0]{follows_padsv_lval_intro}++ if @ops;#d# |
|
|
371 | } |
|
|
372 | |
|
|
373 | $source .= <<EOF; |
|
|
374 | PUSHs (sv); |
328 | PUTBACK; |
375 | PUTBACK; |
329 | EOF |
376 | EOF |
330 | if ($op->flags & B::OPf_MOD) { |
377 | |
331 | if ($op->private & B::OPpLVAL_INTRO) { |
378 | if (($flags & B::OPf_MOD) && ($op->private & B::OPpDEREF)) { |
332 | $source .= " SAVECLEARSV (PAD_SVl ((PADOFFSET)$target));\n"; |
379 | $source .= " if (!SvROK (sv)) vivify_ref (sv, " . $op->private . " & OPpDEREF);\n"; |
333 | } elsif ($op->private & B::OPpDEREF) { |
|
|
334 | my $deref = $op->private & B::OPpDEREF; |
|
|
335 | $source .= " Perl_vivify_ref (PAD_SVl ((PADOFFSET)$target), $deref);\n"; |
|
|
336 | } |
|
|
337 | } |
380 | } |
|
|
381 | $source .= " }\n"; |
|
|
382 | |
|
|
383 | out_next; |
|
|
384 | } |
|
|
385 | |
|
|
386 | sub op_sassign { |
|
|
387 | $source .= <<EOF; |
|
|
388 | { |
|
|
389 | dSP; |
|
|
390 | dPOPTOPssrl; |
|
|
391 | EOF |
|
|
392 | $source .= " SV *temp = left; left = right; right = temp;\n" |
|
|
393 | if $op->private & B::OPpASSIGN_BACKWARDS; |
|
|
394 | |
|
|
395 | if ($insn->{follows_padsv_lval_intro} && !($op->private & B::OPpASSIGN_BACKWARDS)) { |
|
|
396 | # simple assignment - the target exists, but is basically undef |
|
|
397 | $source .= " SvSetSV (right, left);\n"; |
|
|
398 | } else { |
|
|
399 | $source .= " SvSetMagicSV (right, left);\n"; |
|
|
400 | } |
|
|
401 | |
338 | $source .= <<EOF; |
402 | $source .= <<EOF; |
|
|
403 | SETs (right); |
|
|
404 | PUTBACK; |
339 | } |
405 | } |
340 | EOF |
406 | EOF |
341 | |
407 | |
342 | out_next; |
408 | out_next; |
343 | } |
409 | } |
344 | |
410 | |
345 | # pattern const+ (or general push1) |
411 | # pattern const+ (or general push1) |
346 | # pattern pushmark return(?) |
|
|
347 | # pattern pushmark gv rv2av pushmark padsv+o.ä. aassign |
412 | # pattern pushmark gv rv2av pushmark padsv+o.ä. aassign |
348 | |
413 | |
349 | # pattern const method_named |
|
|
350 | sub op_method_named { |
414 | sub op_method_named { |
|
|
415 | if ($insn->{follows_const}) { |
351 | $source .= <<EOF; |
416 | $source .= <<EOF; |
|
|
417 | { |
|
|
418 | dSP; |
|
|
419 | static SV *last_cv; |
|
|
420 | static U32 last_sub_generation; |
|
|
421 | |
|
|
422 | /* simple "polymorphic" inline cache */ |
|
|
423 | if (PL_sub_generation == last_sub_generation) |
|
|
424 | { |
|
|
425 | PUSHs (last_cv); |
|
|
426 | PUTBACK; |
|
|
427 | } |
|
|
428 | else |
|
|
429 | { |
|
|
430 | PL_op = nextop; nextop = Perl_pp_method_named (aTHX); |
|
|
431 | |
|
|
432 | SPAGAIN; |
|
|
433 | last_sub_generation = PL_sub_generation; |
|
|
434 | last_cv = TOPs; |
|
|
435 | } |
|
|
436 | } |
|
|
437 | EOF |
|
|
438 | } else { |
|
|
439 | $source .= <<EOF; |
352 | { |
440 | { |
353 | static HV *last_stash; |
441 | static HV *last_stash; |
354 | static SV *last_cv; |
442 | static SV *last_cv; |
355 | static U32 last_sub_generation; |
443 | static U32 last_sub_generation; |
356 | |
444 | |
… | |
… | |
363 | |
451 | |
364 | /* simple "polymorphic" inline cache */ |
452 | /* simple "polymorphic" inline cache */ |
365 | if (stash == last_stash |
453 | if (stash == last_stash |
366 | && PL_sub_generation == last_sub_generation) |
454 | && PL_sub_generation == last_sub_generation) |
367 | { |
455 | { |
368 | XPUSHs (last_cv); |
456 | PUSHs (last_cv); |
369 | PUTBACK; |
457 | PUTBACK; |
370 | } |
458 | } |
371 | else |
459 | else |
372 | { |
460 | { |
373 | PL_op = nextop; nextop = Perl_pp_method_named (aTHX); |
461 | PL_op = nextop; nextop = Perl_pp_method_named (aTHX); |
… | |
… | |
383 | /* error case usually */ |
471 | /* error case usually */ |
384 | PL_op = nextop; nextop = Perl_pp_method_named (aTHX); |
472 | PL_op = nextop; nextop = Perl_pp_method_named (aTHX); |
385 | } |
473 | } |
386 | } |
474 | } |
387 | EOF |
475 | EOF |
|
|
476 | } |
388 | |
477 | |
389 | out_next; |
478 | out_next; |
390 | } |
479 | } |
391 | |
480 | |
392 | sub op_grepstart { |
481 | sub op_grepstart { |
… | |
… | |
409 | my ($idx) = @_; |
498 | my ($idx) = @_; |
410 | |
499 | |
411 | out_callop; |
500 | out_callop; |
412 | |
501 | |
413 | out_cond_jump $_->[$idx] |
502 | out_cond_jump $_->[$idx] |
414 | for reverse @loop; |
503 | for reverse @op_loop; |
415 | |
504 | |
416 | $source .= " return nextop;\n"; |
505 | $source .= " return nextop;\n"; |
417 | } |
506 | } |
418 | |
507 | |
419 | sub xop_next { |
508 | sub xop_next { |
… | |
… | |
429 | } |
518 | } |
430 | |
519 | |
431 | sub cv2c { |
520 | sub cv2c { |
432 | my ($cv) = @_; |
521 | my ($cv) = @_; |
433 | |
522 | |
434 | @loop = (); |
523 | local @ops; |
|
|
524 | local @op_loop; |
|
|
525 | local %op_regcomp; |
435 | |
526 | |
436 | my %opsseen; |
527 | my %opsseen; |
437 | my @todo = $cv->START; |
528 | my @todo = $cv->START; |
|
|
529 | my %op_target; |
|
|
530 | my $numpushmark; |
438 | |
531 | |
439 | while (my $op = shift @todo) { |
532 | while (my $op = shift @todo) { |
440 | for (; $$op; $op = $op->next) { |
533 | for (; $$op; $op = $op->next) { |
441 | last if $opsseen{$$op}++; |
534 | last if $opsseen{$$op}++; |
442 | push @ops, $op; |
|
|
443 | |
535 | |
444 | my $name = $op->name; |
536 | my $name = $op->name; |
445 | my $class = B::class $op; |
537 | my $class = B::class $op; |
446 | |
538 | |
|
|
539 | my $insn = { op => $op }; |
|
|
540 | |
|
|
541 | push @ops, $insn; |
|
|
542 | |
|
|
543 | if (exists $extend{$name}) { |
|
|
544 | my $extend = $extend{$name}; |
|
|
545 | $extend = $extend->($op) if ref $extend; |
|
|
546 | $insn->{extend} = $extend if defined $extend; |
|
|
547 | } |
|
|
548 | |
|
|
549 | push @todo, $op->next; |
|
|
550 | |
447 | if ($class eq "LOGOP") { |
551 | if ($class eq "LOGOP") { |
448 | unshift @todo, $op->other; # unshift vs. push saves jumps |
552 | push @todo, $op->other; |
|
|
553 | $op_target{${$op->other}}++; |
|
|
554 | |
|
|
555 | # regcomp/o patches ops at runtime, lets expect that |
|
|
556 | if ($name eq "regcomp" && $op->other->pmflags & B::PMf_KEEP) { |
|
|
557 | $op_target{${$op->first}}++; |
|
|
558 | $op_regcomp{${$op->first}} = $op->next; |
|
|
559 | } |
|
|
560 | |
449 | } elsif ($class eq "PMOP") { |
561 | } elsif ($class eq "PMOP") { |
|
|
562 | if (${$op->pmreplstart}) { |
450 | unshift @todo, $op->pmreplstart if ${$op->pmreplstart}; |
563 | unshift @todo, $op->pmreplstart; |
|
|
564 | $op_target{${$op->pmreplstart}}++; |
|
|
565 | } |
|
|
566 | |
451 | } elsif ($class eq "LOOP") { |
567 | } elsif ($class eq "LOOP") { |
452 | push @loop, [$op->nextop, $op->lastop->next, $op->redoop->next]; |
568 | my @targ = ($op->nextop, $op->lastop->next, $op->redoop->next); |
|
|
569 | |
|
|
570 | push @op_loop, \@targ; |
|
|
571 | push @todo, @targ; |
|
|
572 | |
|
|
573 | $op_target{$$_}++ for @targ; |
|
|
574 | |
|
|
575 | } elsif ($class eq "COP") { |
|
|
576 | $insn->{bblock}++ if defined $op->label; |
|
|
577 | |
|
|
578 | } else { |
|
|
579 | if ($name eq "pushmark") { |
|
|
580 | $numpushmark++; |
|
|
581 | } |
453 | } |
582 | } |
454 | } |
583 | } |
455 | } |
584 | } |
|
|
585 | |
|
|
586 | $_->{bblock}++ for grep $op_target{${$_->{op}}}, @ops; |
456 | |
587 | |
457 | local $source = <<EOF; |
588 | local $source = <<EOF; |
458 | OP *%%%FUNC%%% (pTHX) |
589 | OP *%%%FUNC%%% (pTHX) |
459 | { |
590 | { |
460 | register OP *nextop = (OP *)${$ops[0]}L; |
591 | register OP *nextop = (OP *)${$ops[0]->{op}}L; |
461 | EOF |
592 | EOF |
|
|
593 | |
|
|
594 | $source .= " faster_PUSHMARK_PREALLOC ($numpushmark);\n" |
|
|
595 | if $numpushmark; |
462 | |
596 | |
463 | while (@ops) { |
597 | while (@ops) { |
464 | $op = shift @ops; |
598 | $insn = shift @ops; |
|
|
599 | |
|
|
600 | $op = $insn->{op}; |
465 | $op_name = $op->name; |
601 | $op_name = $op->name; |
466 | |
602 | |
|
|
603 | my $class = B::class $op; |
|
|
604 | |
|
|
605 | $source .= "\n/* start basic block */\n" if exists $insn->{bblock};#d# |
467 | $source .= "op_$$op: /* $op_name */\n"; |
606 | $source .= "op_$$op: /* $op_name */\n"; |
468 | #$source .= "fprintf (stderr, \"$$op in op $op_name\\n\");\n";#d# |
607 | #$source .= "fprintf (stderr, \"$$op in op $op_name\\n\");\n";#d# |
469 | #$source .= "{ dSP; sv_dump (TOPs); }\n";#d# |
608 | #$source .= "{ dSP; sv_dump (TOPs); }\n";#d# |
470 | |
609 | |
471 | $source .= " PERL_ASYNC_CHECK ();\n" |
610 | $source .= " PERL_ASYNC_CHECK ();\n" |
472 | unless exists $flag{noasync}{$op_name}; |
611 | unless exists $f_noasync{$op_name}; |
473 | |
612 | |
474 | if (my $can = __PACKAGE__->can ("op_$op_name")) { |
613 | if (my $can = __PACKAGE__->can ("op_$op_name")) { |
475 | # handcrafted replacement |
614 | # handcrafted replacement |
|
|
615 | |
|
|
616 | if ($insn->{extend} > 0) { |
|
|
617 | # coalesce EXTENDs |
|
|
618 | # TODO: properly take negative preceeding and following EXTENDs into account |
|
|
619 | for my $i (@ops) { |
|
|
620 | last if exists $i->{bblock}; |
|
|
621 | last unless exists $i->{extend}; |
|
|
622 | my $extend = delete $i->{extend}; |
|
|
623 | $insn->{extend} += $extend if $extend > 0; |
|
|
624 | } |
|
|
625 | |
|
|
626 | $source .= " { dSP; EXTEND (SP, $insn->{extend}); PUTBACK; }\n" |
|
|
627 | if $insn->{extend} > 0; |
|
|
628 | } |
|
|
629 | |
476 | $can->($op); |
630 | $can->($op); |
477 | |
631 | |
478 | } elsif (exists $flag{unsafe}{$op_name}) { |
632 | } elsif (exists $f_unsafe{$op_name}) { |
479 | # unsafe, return to interpreter |
633 | # unsafe, return to interpreter |
480 | assert "nextop == (OP *)$$op"; |
634 | assert "nextop == (OP *)$$op"; |
481 | $source .= " return nextop;\n"; |
635 | $source .= " return nextop;\n"; |
482 | |
636 | |
483 | } elsif ("LOGOP" eq B::class $op) { |
637 | } elsif ("LOGOP" eq $class) { |
484 | # logical operation with optionaö branch |
638 | # logical operation with optional branch |
485 | out_callop; |
639 | out_callop; |
486 | out_cond_jump $op->other; |
640 | out_cond_jump $op->other; |
487 | out_jump_next; |
641 | out_jump_next; |
488 | |
642 | |
489 | } elsif ("PMOP" eq B::class $op) { |
643 | } elsif ("PMOP" eq $class) { |
490 | # regex-thingy |
644 | # regex-thingy |
491 | out_callop; |
645 | out_callop; |
492 | out_cond_jump $op->pmreplroot if ${$op->pmreplroot}; |
646 | out_cond_jump $op->pmreplroot if $op_name ne "pushre" && ${$op->pmreplroot}; |
493 | out_jump_next; |
647 | out_jump_next; |
494 | |
648 | |
495 | } else { |
649 | } else { |
496 | # normal operator, linear execution |
650 | # normal operator, linear execution |
497 | out_linear; |
651 | out_linear; |
… | |
… | |
508 | #warn $source; |
662 | #warn $source; |
509 | |
663 | |
510 | $source |
664 | $source |
511 | } |
665 | } |
512 | |
666 | |
|
|
667 | my $uid = "aaaaaaa0"; |
|
|
668 | my %so; |
|
|
669 | |
513 | sub source2ptr { |
670 | sub func2ptr { |
514 | my ($source) = @_; |
671 | my (@func) = @_; |
515 | |
672 | |
516 | my $md5 = Digest::MD5::md5_hex $source; |
673 | #LOCK |
517 | $source =~ s/%%%FUNC%%%/Faster_$md5/; |
674 | mkdir $CACHEDIR, 0777; |
|
|
675 | sysopen my $meta_fh, "$CACHEDIR/meta", &Fcntl::O_RDWR | &Fcntl::O_CREAT, 0666 |
|
|
676 | or die "$$CACHEDIR/meta: $!"; |
|
|
677 | binmode $meta_fh, ":raw:perlio"; |
|
|
678 | fcntl_lock fileno $meta_fh |
|
|
679 | or die "$CACHEDIR/meta: $!"; |
518 | |
680 | |
519 | my $stem = "/tmp/$md5"; |
681 | my $meta = eval { Storable::fd_retrieve $meta_fh } || { version => 1 }; |
520 | |
682 | |
521 | unless (-e "$stem$_so") { |
683 | for my $f (@func) { |
|
|
684 | $f->{func} = "F" . Digest::MD5::md5_hex ($f->{source}); |
|
|
685 | $f->{so} = $meta->{$f->{func}}; |
|
|
686 | } |
|
|
687 | |
|
|
688 | if (grep !$_->{so}, @func) { |
|
|
689 | my $stem; |
|
|
690 | |
|
|
691 | do { |
|
|
692 | $stem = "$CACHEDIR/$$-" . $uid++; |
|
|
693 | } while -e "$stem$_so"; |
|
|
694 | |
522 | open FILE, ">:raw", "$stem.c"; |
695 | open my $fh, ">:raw", "$stem.c"; |
523 | print FILE <<EOF; |
696 | print $fh <<EOF; |
524 | #define PERL_NO_GET_CONTEXT |
697 | #define PERL_NO_GET_CONTEXT |
|
|
698 | #define PERL_CORE |
525 | |
699 | |
526 | #include <assert.h> |
700 | #include <assert.h> |
527 | |
701 | |
528 | #include "EXTERN.h" |
702 | #include "EXTERN.h" |
529 | #include "perl.h" |
703 | #include "perl.h" |
530 | #include "XSUB.h" |
704 | #include "XSUB.h" |
|
|
705 | |
|
|
706 | #if 1 |
|
|
707 | # define faster_PUSHMARK_PREALLOC(count) while (PL_markstack_ptr + (count) >= PL_markstack_max) markstack_grow () |
|
|
708 | # define faster_PUSHMARK(p) *++PL_markstack_ptr = (p) - PL_stack_base |
|
|
709 | #else |
|
|
710 | # define faster_PUSHMARK_PREALLOC(count) 1 |
|
|
711 | # define faster_PUSHMARK(p) PUSHMARK(p) |
|
|
712 | #endif |
531 | |
713 | |
532 | #define RUNOPS_TILL(op) \\ |
714 | #define RUNOPS_TILL(op) \\ |
533 | while (nextop != (op)) \\ |
715 | while (nextop != (op)) \\ |
534 | { \\ |
716 | { \\ |
535 | PERL_ASYNC_CHECK (); \\ |
717 | PERL_ASYNC_CHECK (); \\ |
536 | PL_op = nextop; nextop = (PL_op->op_ppaddr)(aTHX); \\ |
718 | PL_op = nextop; nextop = (PL_op->op_ppaddr)(aTHX); \\ |
537 | } |
719 | } |
538 | |
720 | |
539 | EOF |
721 | EOF |
|
|
722 | for my $f (grep !$_->{so}, @func) { |
|
|
723 | next if $f->{so} = $meta->{$f->{func}}; # some cv's alias others |
|
|
724 | |
|
|
725 | warn "compiling $f->{name} to $stem$_so:$f->{func}\n" if $verbose > 1; |
|
|
726 | my $source = $f->{source}; |
|
|
727 | $source =~ s/%%%FUNC%%%/$f->{func}/g; |
540 | print FILE $source; |
728 | print $fh $source; |
|
|
729 | $meta->{$f->{func}} = $f->{so} = $stem; |
|
|
730 | } |
|
|
731 | |
541 | close FILE; |
732 | close $fh; |
542 | system "$COMPILE -o $stem$_o $stem.c"; |
733 | system "$COMPILE -o $stem$_o $stem.c"; |
|
|
734 | unlink "$stem.c" unless $ENV{FASTER_DEBUG} > 0; |
543 | system "$LINK -o $stem$_so $stem$_o $LIBS"; |
735 | system "$LINK -o $stem$_so $stem$_o $LIBS"; |
|
|
736 | unlink "$stem$_o"; |
544 | } |
737 | } |
545 | |
738 | |
546 | # warn $source; |
739 | for my $f (@func) { |
|
|
740 | my $stem = $f->{so}; |
|
|
741 | |
547 | my $so = DynaLoader::dl_load_file "$stem$_so" |
742 | my $so = ($so{$stem} ||= DynaLoader::dl_load_file "$stem$_so") |
548 | or die "$stem$_so: $!"; |
743 | or die "$stem$_so: $!"; |
549 | |
744 | |
550 | DynaLoader::dl_find_symbol $so, "Faster_$md5" |
745 | #unlink "$stem$_so"; |
551 | or die "Faster_$md5: $!" |
746 | |
|
|
747 | $f->{ptr} = DynaLoader::dl_find_symbol $so, $f->{func} |
|
|
748 | or die "$f->{func} not found in $stem$_so: $!"; |
|
|
749 | } |
|
|
750 | |
|
|
751 | seek $meta_fh, 0, 0 or die "$CACHEDIR/meta: $!"; |
|
|
752 | Storable::nstore_fd $meta, $meta_fh; |
|
|
753 | truncate $meta_fh, tell $meta_fh; |
|
|
754 | |
|
|
755 | # UNLOCK (by closing $meta_fh) |
552 | } |
756 | } |
|
|
757 | |
|
|
758 | my %ignore; |
553 | |
759 | |
554 | sub entersub { |
760 | sub entersub { |
555 | my ($cv) = @_; |
761 | my ($cv) = @_; |
556 | |
762 | |
557 | # always compile the whole stash |
763 | my $pkg = $cv->STASH->NAME; |
558 | # my @stash = $cv->STASH->ARRAY; |
764 | |
559 | # warn join ":", @stash; |
765 | return if $ignore{$pkg}; |
560 | # exit; |
766 | |
|
|
767 | warn "optimising ", $cv->STASH->NAME, "\n" |
|
|
768 | if $verbose; |
561 | |
769 | |
562 | eval { |
770 | eval { |
|
|
771 | my @func; |
|
|
772 | |
|
|
773 | push @func, { |
|
|
774 | cv => $cv, |
|
|
775 | name => "<>", |
563 | my $source = cv2c $cv; |
776 | source => cv2c $cv, |
|
|
777 | }; |
564 | |
778 | |
565 | my $ptr = source2ptr $source; |
779 | # always compile the whole stash |
|
|
780 | my %stash = $cv->STASH->ARRAY; |
|
|
781 | while (my ($k, $v) = each %stash) { |
|
|
782 | $v->isa (B::GV::) |
|
|
783 | or next; |
566 | |
784 | |
|
|
785 | my $cv = $v->CV; |
|
|
786 | |
|
|
787 | if ($cv->isa (B::CV::) |
|
|
788 | && ${$cv->START} |
|
|
789 | && $cv->START->name ne "null") { |
|
|
790 | |
|
|
791 | push @func, { |
|
|
792 | cv => $cv, |
|
|
793 | name => $k, |
|
|
794 | source => cv2c $cv, |
|
|
795 | }; |
|
|
796 | } |
|
|
797 | } |
|
|
798 | |
|
|
799 | func2ptr @func; |
|
|
800 | |
|
|
801 | for my $f (@func) { |
567 | patch_cv $cv, $ptr; |
802 | patch_cv $f->{cv}, $f->{ptr}; |
|
|
803 | } |
568 | }; |
804 | }; |
569 | |
805 | |
570 | warn $@ if $@; |
806 | if ($@) { |
|
|
807 | $ignore{$pkg}++; |
|
|
808 | warn $@; |
|
|
809 | } |
571 | } |
810 | } |
572 | |
811 | |
573 | hook_entersub; |
812 | hook_entersub; |
574 | |
813 | |
575 | 1; |
814 | 1; |
|
|
815 | |
|
|
816 | =back |
|
|
817 | |
|
|
818 | =head1 ENVIRONMENT VARIABLES |
|
|
819 | |
|
|
820 | The following environment variables influence the behaviour of Faster: |
|
|
821 | |
|
|
822 | =over 4 |
|
|
823 | |
|
|
824 | =item FASTER_VERBOSE |
|
|
825 | |
|
|
826 | Faster will output more informational messages when set to values higher |
|
|
827 | than C<0>. Currently, C<1> outputs which packages are being compiled, C<3> |
|
|
828 | outputs the cache directory and C<10> outputs information on which perl |
|
|
829 | function is compiled into which shared object. |
|
|
830 | |
|
|
831 | =item FASTER_DEBUG |
|
|
832 | |
|
|
833 | Add debugging code when set to values higher than C<0>. Currently, this |
|
|
834 | adds 1-3 C<assert>'s per perl op (FASTER_DEBUG > 1), to ensure that opcode |
|
|
835 | order and C execution order are compatible. |
|
|
836 | |
|
|
837 | =item FASTER_CACHE |
|
|
838 | |
|
|
839 | Set a persistent cache directory that caches compiled code fragments. The |
|
|
840 | default is C<$HOME/.perl-faster-cache> if C<HOME> is set and a temporary |
|
|
841 | directory otherwise. |
|
|
842 | |
|
|
843 | This directory will always grow in size, so you might need to erase it |
|
|
844 | from time to time. |
576 | |
845 | |
577 | =back |
846 | =back |
578 | |
847 | |
579 | =head1 BUGS/LIMITATIONS |
848 | =head1 BUGS/LIMITATIONS |
580 | |
849 | |
581 | Perl will check much less often for asynchronous signals in |
850 | Perl will check much less often for asynchronous signals in |
582 | Faster-compiled code. It tries to check on every function call, loop |
851 | Faster-compiled code. It tries to check on every function call, loop |
583 | iteration and every I/O operator, though. |
852 | iteration and every I/O operator, though. |
584 | |
853 | |
585 | The following things will disable Faster. If you manage to enable them at |
854 | The following things will disable Faster. If you manage to enable them at |
586 | runtime, bad things will happen. |
855 | runtime, bad things will happen. Enabling them at startup will be fine, |
|
|
856 | though. |
587 | |
857 | |
588 | enabled tainting |
858 | enabled tainting |
589 | enabled debugging |
859 | enabled debugging |
590 | |
860 | |
591 | This will dramatically reduce Faster's performance: |
861 | Thread-enabled builds of perl will dramatically reduce Faster's |
|
|
862 | performance, but you don't care about speed if you enable threads anyway. |
592 | |
863 | |
593 | threads (but you don't care about speed if you use threads anyway) |
|
|
594 | |
|
|
595 | These constructs will force the use of the interpreter as soon as they are |
864 | These constructs will force the use of the interpreter for the currently |
596 | being executed, for the rest of the currently executed: |
865 | executed function as soon as they are being encountered during execution. |
597 | |
866 | |
598 | .., ... (flipflop operators) |
|
|
599 | goto |
867 | goto |
600 | next, redo (but not well-behaved last's) |
868 | next, redo (but not well-behaved last's) |
601 | eval |
869 | eval |
602 | require |
870 | require |
603 | any use of formats |
871 | any use of formats |
|
|
872 | .., ... (flipflop operators) |
604 | |
873 | |
605 | =head1 AUTHOR |
874 | =head1 AUTHOR |
606 | |
875 | |
607 | Marc Lehmann <schmorp@schmorp.de> |
876 | Marc Lehmann <schmorp@schmorp.de> |
608 | http://home.schmorp.de/ |
877 | http://home.schmorp.de/ |