Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
1905 | serge | 1 | /* |
2 | optimize: get a grip on the different optimizations |
||
3 | |||
4 | copyright 2006-9 by the mpg123 project - free software under the terms of the LGPL 2.1 |
||
5 | see COPYING and AUTHORS files in distribution or http://mpg123.org |
||
6 | initially written by Thomas Orgis, inspired by 3DNow stuff in mpg123.[hc] |
||
7 | |||
8 | Currently, this file contains the struct and function to choose an optimization variant and works only when OPT_MULTI is in effect. |
||
9 | */ |
||
10 | |||
11 | #include "mpg123lib_intern.h" /* includes optimize.h */ |
||
12 | #include "debug.h" |
||
13 | |||
14 | /* Must match the enum dectype! */ |
||
15 | |||
16 | /* |
||
17 | It SUCKS having to define these names that way, but compile-time intialization of string arrays is a bitch. |
||
18 | GCC doesn't see constant stuff when it's wiggling in front of it! |
||
19 | Anyhow: Have a script for that: |
||
20 | names="generic generic_dither i386 i486 i586 i586_dither MMX 3DNow 3DNowExt AltiVec SSE x86-64" |
||
21 | for i in $names; do echo "##define dn_${i/-/_} \"$i\""; done |
||
22 | echo -n "static const char* decname[] = |
||
23 | { |
||
24 | \"auto\" |
||
25 | " |
||
26 | for i in $names; do echo -n ", dn_${i/-/_}"; done |
||
27 | echo " |
||
28 | , \"nodec\" |
||
29 | };" |
||
30 | */ |
||
31 | #define dn_generic "generic" |
||
32 | #define dn_generic_dither "generic_dither" |
||
33 | #define dn_i386 "i386" |
||
34 | #define dn_i486 "i486" |
||
35 | #define dn_i586 "i586" |
||
36 | #define dn_i586_dither "i586_dither" |
||
37 | #define dn_MMX "MMX" |
||
38 | #define dn_3DNow "3DNow" |
||
39 | #define dn_3DNowExt "3DNowExt" |
||
40 | #define dn_AltiVec "AltiVec" |
||
41 | #define dn_SSE "SSE" |
||
42 | #define dn_x86_64 "x86-64" |
||
43 | #define dn_ARM "ARM" |
||
44 | static const char* decname[] = |
||
45 | { |
||
46 | "auto" |
||
47 | , dn_generic, dn_generic_dither, dn_i386, dn_i486, dn_i586, dn_i586_dither, dn_MMX, dn_3DNow, dn_3DNowExt, dn_AltiVec, dn_SSE, dn_x86_64, dn_ARM |
||
48 | , "nodec" |
||
49 | }; |
||
50 | |||
51 | #if (defined OPT_X86) && (defined OPT_MULTI) |
||
52 | #include "getcpuflags.h" |
||
53 | struct cpuflags cpu_flags; |
||
54 | #else |
||
55 | /* Faking stuff for non-multi builds. The same code for synth function choice is used. |
||
56 | Just no runtime dependency of result... */ |
||
57 | char cpu_flags; |
||
58 | #define cpu_i586(s) 1 |
||
59 | #define cpu_fpu(s) 1 |
||
60 | #define cpu_mmx(s) 1 |
||
61 | #define cpu_3dnow(s) 1 |
||
62 | #define cpu_3dnowext(s) 1 |
||
63 | #define cpu_sse(s) 1 |
||
64 | #define cpu_sse2(s) 1 |
||
65 | #define cpu_sse3(s) 1 |
||
66 | #endif |
||
67 | |||
68 | /* Ugly macros to build conditional synth function array values. */ |
||
69 | |||
70 | #ifndef NO_8BIT |
||
71 | #define IF8(synth) synth, |
||
72 | #else |
||
73 | #define IF8(synth) |
||
74 | #endif |
||
75 | |||
76 | #ifndef NO_REAL |
||
77 | #define IFREAL(synth) synth, |
||
78 | #else |
||
79 | #define IFREAL(synth) |
||
80 | #endif |
||
81 | |||
82 | #ifndef NO_32BIT |
||
83 | #define IF32(synth) synth |
||
84 | #else |
||
85 | #define IF32(synth) |
||
86 | #endif |
||
87 | |||
88 | #ifndef NO_16BIT |
||
89 | # define OUT_SYNTHS(synth_16, synth_8, synth_real, synth_32) { synth_16, IF8(synth_8) IFREAL(synth_real) IF32(synth_32) } |
||
90 | #else |
||
91 | # define OUT_SYNTHS(synth_16, synth_8, synth_real, synth_32) { IF8(synth_8) IFREAL(synth_real) IF32(synth_32) } |
||
92 | #endif |
||
93 | |||
94 | const struct synth_s synth_base = |
||
95 | { |
||
96 | { /* plain */ |
||
97 | OUT_SYNTHS(synth_1to1, synth_1to1_8bit, synth_1to1_real, synth_1to1_s32) |
||
98 | # ifndef NO_DOWNSAMPLE |
||
99 | ,OUT_SYNTHS(synth_2to1, synth_2to1_8bit, synth_2to1_real, synth_2to1_s32) |
||
100 | ,OUT_SYNTHS(synth_4to1, synth_4to1_8bit, synth_4to1_real, synth_4to1_s32) |
||
101 | # endif |
||
102 | # ifndef NO_NTOM |
||
103 | ,OUT_SYNTHS(synth_ntom, synth_ntom_8bit, synth_ntom_real, synth_ntom_s32) |
||
104 | # endif |
||
105 | }, |
||
106 | { /* stereo, by default only wrappers over plain synth */ |
||
107 | OUT_SYNTHS(synth_stereo_wrap, synth_stereo_wrap, synth_stereo_wrap, synth_stereo_wrap) |
||
108 | # ifndef NO_DOWNSAMPLE |
||
109 | ,OUT_SYNTHS(synth_stereo_wrap, synth_stereo_wrap, synth_stereo_wrap, synth_stereo_wrap) |
||
110 | ,OUT_SYNTHS(synth_stereo_wrap, synth_stereo_wrap, synth_stereo_wrap, synth_stereo_wrap) |
||
111 | # endif |
||
112 | # ifndef NO_NTOM |
||
113 | ,OUT_SYNTHS(synth_stereo_wrap, synth_stereo_wrap, synth_stereo_wrap, synth_stereo_wrap) |
||
114 | # endif |
||
115 | }, |
||
116 | { /* mono2stereo */ |
||
117 | OUT_SYNTHS(synth_1to1_mono2stereo, synth_1to1_8bit_mono2stereo, synth_1to1_real_mono2stereo, synth_1to1_s32_mono2stereo) |
||
118 | # ifndef NO_DOWNSAMPLE |
||
119 | ,OUT_SYNTHS(synth_2to1_mono2stereo, synth_2to1_8bit_mono2stereo, synth_2to1_real_mono2stereo, synth_2to1_s32_mono2stereo) |
||
120 | ,OUT_SYNTHS(synth_4to1_mono2stereo, synth_4to1_8bit_mono2stereo, synth_4to1_real_mono2stereo, synth_4to1_s32_mono2stereo) |
||
121 | # endif |
||
122 | # ifndef NO_NTOM |
||
123 | ,OUT_SYNTHS(synth_ntom_mono2stereo, synth_ntom_8bit_mono2stereo, synth_ntom_real_mono2stereo, synth_ntom_s32_mono2stereo) |
||
124 | # endif |
||
125 | }, |
||
126 | { /* mono*/ |
||
127 | OUT_SYNTHS(synth_1to1_mono, synth_1to1_8bit_mono, synth_1to1_real_mono, synth_1to1_s32_mono) |
||
128 | # ifndef NO_DOWNSAMPLE |
||
129 | ,OUT_SYNTHS(synth_2to1_mono, synth_2to1_8bit_mono, synth_2to1_real_mono, synth_2to1_s32_mono) |
||
130 | ,OUT_SYNTHS(synth_4to1_mono, synth_4to1_8bit_mono, synth_4to1_real_mono, synth_4to1_s32_mono) |
||
131 | # endif |
||
132 | # ifndef NO_NTOM |
||
133 | ,OUT_SYNTHS(synth_ntom_mono, synth_ntom_8bit_mono, synth_ntom_real_mono, synth_ntom_s32_mono) |
||
134 | #endif |
||
135 | } |
||
136 | }; |
||
137 | |||
138 | #ifdef OPT_X86 |
||
139 | /* More plain synths for i386 */ |
||
140 | const func_synth plain_i386[r_limit][f_limit] = |
||
141 | { /* plain */ |
||
142 | OUT_SYNTHS(synth_1to1_i386, synth_1to1_8bit_i386, synth_1to1_real_i386, synth_1to1_s32_i386) |
||
143 | # ifndef NO_DOWNSAMPLE |
||
144 | ,OUT_SYNTHS(synth_2to1_i386, synth_2to1_8bit_i386, synth_2to1_real_i386, synth_2to1_s32_i386) |
||
145 | ,OUT_SYNTHS(synth_4to1_i386, synth_4to1_8bit_i386, synth_4to1_real_i386, synth_4to1_s32_i386) |
||
146 | # endif |
||
147 | # ifndef NO_NTOM |
||
148 | ,OUT_SYNTHS(synth_ntom, synth_ntom_8bit, synth_ntom_real, synth_ntom_s32) |
||
149 | # endif |
||
150 | }; |
||
151 | #endif |
||
152 | |||
153 | |||
154 | enum optdec defdec(void){ return defopt; } |
||
155 | |||
156 | enum optcla decclass(const enum optdec type) |
||
157 | { |
||
158 | return (type == mmx || type == sse || type == dreidnowext || type == x86_64 ) ? mmxsse : normal; |
||
159 | } |
||
160 | |||
161 | |||
162 | static int find_synth(func_synth synth, const func_synth synths[r_limit][f_limit]) |
||
163 | { |
||
164 | enum synth_resample ri; |
||
165 | enum synth_format fi; |
||
166 | for(ri=0; ri |
||
167 | for(fi=0; fi |
||
168 | if(synth == synths[ri][fi]) |
||
169 | return TRUE; |
||
170 | |||
171 | return FALSE; |
||
172 | } |
||
173 | |||
174 | /* Determine what kind of decoder is actually active |
||
175 | This depends on runtime choices which may cause fallback to i386 or generic code. */ |
||
176 | static int find_dectype(mpg123_handle *fr) |
||
177 | { |
||
178 | enum optdec type = nodec; |
||
179 | /* Direct and indirect usage, 1to1 stereo decoding. |
||
180 | Concentrating on the plain stereo synth should be fine, mono stuff is derived. */ |
||
181 | func_synth basic_synth = fr->synth; |
||
182 | #ifndef NO_8BIT |
||
183 | #ifndef NO_16BIT |
||
184 | if(basic_synth == synth_1to1_8bit_wrap) |
||
185 | basic_synth = fr->synths.plain[r_1to1][f_16]; /* That is what's really below the surface. */ |
||
186 | #endif |
||
187 | #endif |
||
188 | |||
189 | if(FALSE) ; /* Just to initialize the else if ladder. */ |
||
190 | #ifndef NO_16BIT |
||
191 | #ifdef OPT_3DNOWEXT |
||
192 | else if(basic_synth == synth_1to1_3dnowext) type = dreidnowext; |
||
193 | #endif |
||
194 | #ifdef OPT_SSE |
||
195 | else if(basic_synth == synth_1to1_sse) type = sse; |
||
196 | #endif |
||
197 | #ifdef OPT_3DNOW |
||
198 | else if(basic_synth == synth_1to1_3dnow) type = dreidnow; |
||
199 | #endif |
||
200 | #ifdef OPT_MMX |
||
201 | else if(basic_synth == synth_1to1_mmx) type = mmx; |
||
202 | #endif |
||
203 | #ifdef OPT_I586_DITHER |
||
204 | else if(basic_synth == synth_1to1_i586_dither) type = ifuenf_dither; |
||
205 | #endif |
||
206 | #ifdef OPT_I586 |
||
207 | else if(basic_synth == synth_1to1_i586) type = ifuenf; |
||
208 | #endif |
||
209 | #ifdef OPT_ALTIVEC |
||
210 | else if(basic_synth == synth_1to1_altivec) type = altivec; |
||
211 | #endif |
||
212 | #ifdef OPT_X86_64 |
||
213 | else if(basic_synth == synth_1to1_x86_64) type = x86_64; |
||
214 | #endif |
||
215 | #ifdef OPT_ARM |
||
216 | else if(basic_synth == synth_1to1_arm) type = arm; |
||
217 | #endif |
||
218 | #ifdef OPT_GENERIC_DITHER |
||
219 | else if(basic_synth == synth_1to1_dither) type = generic_dither; |
||
220 | #endif |
||
221 | #ifdef OPT_DITHER /* either i586 or generic! */ |
||
222 | #ifndef NO_DOWNSAMPLE |
||
223 | else if |
||
224 | ( |
||
225 | basic_synth == synth_2to1_dither |
||
226 | || basic_synth == synth_4to1_dither |
||
227 | ) type = generic_dither; |
||
228 | #endif |
||
229 | #endif |
||
230 | #endif /* 16bit */ |
||
231 | |||
232 | #ifndef NO_REAL |
||
233 | #ifdef OPT_SSE |
||
234 | else if(basic_synth == synth_1to1_real_sse) type = sse; |
||
235 | #endif |
||
236 | #ifdef OPT_X86_64 |
||
237 | else if(basic_synth == synth_1to1_real_x86_64) type = x86_64; |
||
238 | #endif |
||
239 | #ifdef OPT_ALTIVEC |
||
240 | else if(basic_synth == synth_1to1_real_altivec) type = altivec; |
||
241 | #endif |
||
242 | |||
243 | #endif /* real */ |
||
244 | |||
245 | #ifndef NO_32BIT |
||
246 | #ifdef OPT_SSE |
||
247 | else if(basic_synth == synth_1to1_s32_sse) type = sse; |
||
248 | #endif |
||
249 | #ifdef OPT_X86_64 |
||
250 | else if(basic_synth == synth_1to1_s32_x86_64) type = x86_64; |
||
251 | #endif |
||
252 | #ifdef OPT_ALTIVEC |
||
253 | else if(basic_synth == synth_1to1_s32_altivec) type = altivec; |
||
254 | #endif |
||
255 | #endif /* 32bit */ |
||
256 | |||
257 | #ifdef OPT_X86 |
||
258 | else if(find_synth(basic_synth, plain_i386)) |
||
259 | type = idrei; |
||
260 | #endif |
||
261 | |||
262 | else if(find_synth(basic_synth, synth_base.plain)) |
||
263 | type = generic; |
||
264 | |||
265 | |||
266 | |||
267 | #ifdef OPT_I486 |
||
268 | /* i486 is special ... the specific code is in use for 16bit 1to1 stereo |
||
269 | otherwise we have i386 active... but still, the distinction doesn't matter*/ |
||
270 | type = ivier; |
||
271 | #endif |
||
272 | |||
273 | if(type != nodec) |
||
274 | { |
||
275 | fr->cpu_opts.type = type; |
||
276 | fr->cpu_opts.class = decclass(type); |
||
277 | |||
278 | debug3("determined active decoder type %i (%s) of class %i", type, decname[type], fr->cpu_opts.class); |
||
279 | return MPG123_OK; |
||
280 | } |
||
281 | else |
||
282 | { |
||
283 | if(NOQUIET) error("Unable to determine active decoder type -- this is SERIOUS b0rkage!"); |
||
284 | |||
285 | fr->err = MPG123_BAD_DECODER_SETUP; |
||
286 | return MPG123_ERR; |
||
287 | } |
||
288 | } |
||
289 | |||
290 | /* set synth functions for current frame, optimizations handled by opt_* macros */ |
||
291 | int set_synth_functions(mpg123_handle *fr) |
||
292 | { |
||
293 | enum synth_resample resample = r_none; |
||
294 | enum synth_format basic_format = f_none; /* Default is always 16bit, or whatever. */ |
||
295 | |||
296 | /* Select the basic output format, different from 16bit: 8bit, real. */ |
||
297 | if(FALSE){} |
||
298 | #ifndef NO_16BIT |
||
299 | else if(fr->af.encoding & MPG123_ENC_16) |
||
300 | basic_format = f_16; |
||
301 | #endif |
||
302 | #ifndef NO_8BIT |
||
303 | else if(fr->af.encoding & MPG123_ENC_8) |
||
304 | basic_format = f_8; |
||
305 | #endif |
||
306 | #ifndef NO_REAL |
||
307 | else if(fr->af.encoding & MPG123_ENC_FLOAT) |
||
308 | basic_format = f_real; |
||
309 | #endif |
||
310 | #ifndef NO_32BIT |
||
311 | else if(fr->af.encoding & MPG123_ENC_32) |
||
312 | basic_format = f_32; |
||
313 | #endif |
||
314 | |||
315 | /* Make sure the chosen format is compiled into this lib. */ |
||
316 | if(basic_format == f_none) |
||
317 | { |
||
318 | if(NOQUIET) error("set_synth_functions: This output format is disabled in this build!"); |
||
319 | |||
320 | return -1; |
||
321 | } |
||
322 | |||
323 | /* Be explicit about downsampling variant. */ |
||
324 | switch(fr->down_sample) |
||
325 | { |
||
326 | case 0: resample = r_1to1; break; |
||
327 | #ifndef NO_DOWNSAMPLE |
||
328 | case 1: resample = r_2to1; break; |
||
329 | case 2: resample = r_4to1; break; |
||
330 | #endif |
||
331 | #ifndef NO_NTOM |
||
332 | case 3: resample = r_ntom; break; |
||
333 | #endif |
||
334 | } |
||
335 | |||
336 | if(resample == r_none) |
||
337 | { |
||
338 | if(NOQUIET) error("set_synth_functions: This resampling mode is not supported in this build!"); |
||
339 | |||
340 | return -1; |
||
341 | } |
||
342 | |||
343 | debug2("selecting synth: resample=%i format=%i", resample, basic_format); |
||
344 | /* Finally selecting the synth functions for stereo / mono. */ |
||
345 | fr->synth = fr->synths.plain[resample][basic_format]; |
||
346 | fr->synth_stereo = fr->synths.stereo[resample][basic_format]; |
||
347 | fr->synth_mono = fr->af.channels==2 |
||
348 | ? fr->synths.mono2stereo[resample][basic_format] /* Mono MPEG file decoded to stereo. */ |
||
349 | : fr->synths.mono[resample][basic_format]; /* Mono MPEG file decoded to mono. */ |
||
350 | |||
351 | if(find_dectype(fr) != MPG123_OK) /* Actually determine the currently active decoder breed. */ |
||
352 | { |
||
353 | fr->err = MPG123_BAD_DECODER_SETUP; |
||
354 | return MPG123_ERR; |
||
355 | } |
||
356 | |||
357 | if(frame_buffers(fr) != 0) |
||
358 | { |
||
359 | fr->err = MPG123_NO_BUFFERS; |
||
360 | if(NOQUIET) error("Failed to set up decoder buffers!"); |
||
361 | |||
362 | return MPG123_ERR; |
||
363 | } |
||
364 | |||
365 | #ifndef NO_8BIT |
||
366 | if(basic_format == f_8) |
||
367 | { |
||
368 | if(make_conv16to8_table(fr) != 0) |
||
369 | { |
||
370 | if(NOQUIET) error("Failed to set up conv16to8 table!"); |
||
371 | /* it's a bit more work to get proper error propagation up */ |
||
372 | return -1; |
||
373 | } |
||
374 | } |
||
375 | #endif |
||
376 | |||
377 | #ifdef OPT_MMXORSSE |
||
378 | /* Special treatment for MMX, SSE and 3DNowExt stuff. |
||
379 | The real-decoding SSE for x86-64 uses normal tables! */ |
||
380 | if(fr->cpu_opts.class == mmxsse |
||
381 | # ifndef NO_REAL |
||
382 | && basic_format != f_real |
||
383 | # endif |
||
384 | # ifndef NO_32BIT |
||
385 | && basic_format != f_32 |
||
386 | # endif |
||
387 | # ifdef ACCURATE_ROUNDING |
||
388 | && fr->cpu_opts.type != sse |
||
389 | && fr->cpu_opts.type != x86_64 |
||
390 | # endif |
||
391 | ) |
||
392 | { |
||
393 | #ifndef NO_LAYER3 |
||
394 | init_layer3_stuff(fr, init_layer3_gainpow2_mmx); |
||
395 | #endif |
||
396 | #ifndef NO_LAYER12 |
||
397 | init_layer12_stuff(fr, init_layer12_table_mmx); |
||
398 | #endif |
||
399 | fr->make_decode_tables = make_decode_tables_mmx; |
||
400 | } |
||
401 | else |
||
402 | #endif |
||
403 | { |
||
404 | #ifndef NO_LAYER3 |
||
405 | init_layer3_stuff(fr, init_layer3_gainpow2); |
||
406 | #endif |
||
407 | #ifndef NO_LAYER12 |
||
408 | init_layer12_stuff(fr, init_layer12_table); |
||
409 | #endif |
||
410 | fr->make_decode_tables = make_decode_tables; |
||
411 | } |
||
412 | |||
413 | /* We allocated the table buffers just now, so (re)create the tables. */ |
||
414 | fr->make_decode_tables(fr); |
||
415 | |||
416 | return 0; |
||
417 | } |
||
418 | |||
419 | int frame_cpu_opt(mpg123_handle *fr, const char* cpu) |
||
420 | { |
||
421 | const char* chosen = ""; /* the chosen decoder opt as string */ |
||
422 | enum optdec want_dec = nodec; |
||
423 | int done = 0; |
||
424 | int auto_choose = 0; |
||
425 | #ifdef OPT_DITHER |
||
426 | int dithered = FALSE; /* If some dithered decoder is chosen. */ |
||
427 | #endif |
||
428 | |||
429 | want_dec = dectype(cpu); |
||
430 | auto_choose = want_dec == autodec; |
||
431 | /* Fill whole array of synth functions with generic code first. */ |
||
432 | fr->synths = synth_base; |
||
433 | |||
434 | #ifndef OPT_MULTI |
||
435 | { |
||
436 | if(!auto_choose && want_dec != defopt) |
||
437 | { |
||
438 | if(NOQUIET) error2("you wanted decoder type %i, I only have %i", want_dec, defopt); |
||
439 | } |
||
440 | auto_choose = TRUE; /* There will be only one choice anyway. */ |
||
441 | } |
||
442 | #endif |
||
443 | |||
444 | fr->cpu_opts.type = nodec; |
||
445 | /* covers any i386+ cpu; they actually differ only in the synth_1to1 function, mostly... */ |
||
446 | #ifdef OPT_X86 |
||
447 | |||
448 | #ifndef NO_LAYER3 |
||
449 | #if (defined OPT_3DNOW || defined OPT_3DNOWEXT) |
||
450 | fr->cpu_opts.dct36 = dct36; |
||
451 | #endif |
||
452 | #endif |
||
453 | |||
454 | if(cpu_i586(cpu_flags)) |
||
455 | { |
||
456 | # ifdef OPT_MULTI |
||
457 | debug2("standard flags: 0x%08x\textended flags: 0x%08x", cpu_flags.std, cpu_flags.ext); |
||
458 | # endif |
||
459 | #ifdef OPT_SSE |
||
460 | if( !done && (auto_choose || want_dec == sse) |
||
461 | && cpu_sse(cpu_flags) && cpu_mmx(cpu_flags) ) |
||
462 | { |
||
463 | chosen = "SSE"; |
||
464 | fr->cpu_opts.type = sse; |
||
465 | # ifndef NO_16BIT |
||
466 | fr->synths.plain[r_1to1][f_16] = synth_1to1_sse; |
||
467 | # ifdef ACCURATE_ROUNDING |
||
468 | fr->synths.stereo[r_1to1][f_16] = synth_1to1_stereo_sse; |
||
469 | # endif |
||
470 | # endif |
||
471 | # ifndef NO_REAL |
||
472 | fr->synths.plain[r_1to1][f_real] = synth_1to1_real_sse; |
||
473 | fr->synths.stereo[r_1to1][f_real] = synth_1to1_real_stereo_sse; |
||
474 | # endif |
||
475 | # ifndef NO_32BIT |
||
476 | fr->synths.plain[r_1to1][f_32] = synth_1to1_s32_sse; |
||
477 | fr->synths.stereo[r_1to1][f_32] = synth_1to1_s32_stereo_sse; |
||
478 | # endif |
||
479 | done = 1; |
||
480 | } |
||
481 | #endif |
||
482 | # ifdef OPT_3DNOWEXT |
||
483 | if( !done && (auto_choose || want_dec == dreidnowext ) |
||
484 | && cpu_3dnow(cpu_flags) |
||
485 | && cpu_3dnowext(cpu_flags) |
||
486 | && cpu_mmx(cpu_flags) ) |
||
487 | { |
||
488 | chosen = "3DNowExt"; |
||
489 | fr->cpu_opts.type = dreidnowext; |
||
490 | # ifndef NO_LAYER3 |
||
491 | fr->cpu_opts.dct36 = dct36_3dnowext; |
||
492 | # endif |
||
493 | # ifndef NO_16BIT |
||
494 | fr->synths.plain[r_1to1][f_16] = synth_1to1_3dnowext; |
||
495 | # endif |
||
496 | done = 1; |
||
497 | } |
||
498 | #endif |
||
499 | #ifdef OPT_3DNOW |
||
500 | if( !done && (auto_choose || want_dec == dreidnow) |
||
501 | && cpu_3dnow(cpu_flags) && cpu_mmx(cpu_flags) ) |
||
502 | { |
||
503 | chosen = "3DNow"; |
||
504 | fr->cpu_opts.type = dreidnow; |
||
505 | # ifndef NO_LAYER3 |
||
506 | fr->cpu_opts.dct36 = dct36_3dnow; |
||
507 | # endif |
||
508 | # ifndef NO_16BIT |
||
509 | fr->synths.plain[r_1to1][f_16] = synth_1to1_3dnow; |
||
510 | # endif |
||
511 | done = 1; |
||
512 | } |
||
513 | #endif |
||
514 | #ifdef OPT_MMX |
||
515 | if( !done && (auto_choose || want_dec == mmx) |
||
516 | && cpu_mmx(cpu_flags) ) |
||
517 | { |
||
518 | chosen = "MMX"; |
||
519 | fr->cpu_opts.type = mmx; |
||
520 | # ifndef NO_16BIT |
||
521 | fr->synths.plain[r_1to1][f_16] = synth_1to1_mmx; |
||
522 | # endif |
||
523 | done = 1; |
||
524 | } |
||
525 | #endif |
||
526 | #ifdef OPT_I586 |
||
527 | if(!done && (auto_choose || want_dec == ifuenf)) |
||
528 | { |
||
529 | chosen = "i586/pentium"; |
||
530 | fr->cpu_opts.type = ifuenf; |
||
531 | # ifndef NO_16BIT |
||
532 | fr->synths.plain[r_1to1][f_16] = synth_1to1_i586; |
||
533 | # endif |
||
534 | done = 1; |
||
535 | } |
||
536 | #endif |
||
537 | #ifdef OPT_I586_DITHER |
||
538 | if(!done && (auto_choose || want_dec == ifuenf_dither)) |
||
539 | { |
||
540 | chosen = "dithered i586/pentium"; |
||
541 | fr->cpu_opts.type = ifuenf_dither; |
||
542 | dithered = TRUE; |
||
543 | # ifndef NO_16BIT |
||
544 | fr->synths.plain[r_1to1][f_16] = synth_1to1_i586_dither; |
||
545 | # ifndef NO_DOWNSAMPLE |
||
546 | fr->synths.plain[r_2to1][f_16] = synth_2to1_dither; |
||
547 | fr->synths.plain[r_4to1][f_16] = synth_4to1_dither; |
||
548 | # endif |
||
549 | # endif |
||
550 | done = 1; |
||
551 | } |
||
552 | #endif |
||
553 | } |
||
554 | #ifdef OPT_I486 |
||
555 | /* That won't cooperate in multi opt mode - forcing i486 in layer3.c |
||
556 | But still... here it is... maybe for real use in future. */ |
||
557 | if(!done && (auto_choose || want_dec == ivier)) |
||
558 | { |
||
559 | chosen = "i486"; |
||
560 | fr->cpu_opts.type = ivier; |
||
561 | done = 1; |
||
562 | } |
||
563 | #endif |
||
564 | #ifdef OPT_I386 |
||
565 | if(!done && (auto_choose || want_dec == idrei)) |
||
566 | { |
||
567 | chosen = "i386"; |
||
568 | fr->cpu_opts.type = idrei; |
||
569 | done = 1; |
||
570 | } |
||
571 | #endif |
||
572 | |||
573 | if(done) |
||
574 | { |
||
575 | /* |
||
576 | We have chosen some x86 decoder... fillup some i386 stuff. |
||
577 | There is an open question about using dithered synth_1to1 for 8bit wrappers. |
||
578 | For quality it won't make sense, but wrapped i586_dither wrapped may still be faster... |
||
579 | */ |
||
580 | enum synth_resample ri; |
||
581 | enum synth_format fi; |
||
582 | # ifndef NO_8BIT |
||
583 | # ifndef NO_16BIT /* possibility to use a 16->8 wrapper... */ |
||
584 | if(fr->synths.plain[r_1to1][f_16] != synth_base.plain[r_1to1][f_16]) |
||
585 | { |
||
586 | fr->synths.plain[r_1to1][f_8] = synth_1to1_8bit_wrap; |
||
587 | fr->synths.mono[r_1to1][f_8] = synth_1to1_8bit_wrap_mono; |
||
588 | fr->synths.mono2stereo[r_1to1][f_8] = synth_1to1_8bit_wrap_mono2stereo; |
||
589 | } |
||
590 | # endif |
||
591 | # endif |
||
592 | for(ri=0; ri |
||
593 | for(fi=0; fi |
||
594 | { |
||
595 | if(fr->synths.plain[ri][fi] == synth_base.plain[ri][fi]) |
||
596 | fr->synths.plain[ri][fi] = plain_i386[ri][fi]; |
||
597 | } |
||
598 | } |
||
599 | |||
600 | #endif /* OPT_X86 */ |
||
601 | |||
602 | #ifdef OPT_X86_64 |
||
603 | if(!done && (auto_choose || want_dec == x86_64)) |
||
604 | { |
||
605 | chosen = "x86-64 (SSE)"; |
||
606 | fr->cpu_opts.type = x86_64; |
||
607 | # ifndef NO_16BIT |
||
608 | fr->synths.plain[r_1to1][f_16] = synth_1to1_x86_64; |
||
609 | fr->synths.stereo[r_1to1][f_16] = synth_1to1_stereo_x86_64; |
||
610 | # endif |
||
611 | # ifndef NO_REAL |
||
612 | fr->synths.plain[r_1to1][f_real] = synth_1to1_real_x86_64; |
||
613 | fr->synths.stereo[r_1to1][f_real] = synth_1to1_real_stereo_x86_64; |
||
614 | # endif |
||
615 | # ifndef NO_32BIT |
||
616 | fr->synths.plain[r_1to1][f_32] = synth_1to1_s32_x86_64; |
||
617 | fr->synths.stereo[r_1to1][f_32] = synth_1to1_s32_stereo_x86_64; |
||
618 | # endif |
||
619 | done = 1; |
||
620 | } |
||
621 | #endif |
||
622 | |||
623 | #ifdef OPT_GENERIC_DITHER |
||
624 | if(!done && (auto_choose || want_dec == generic_dither)) |
||
625 | { |
||
626 | chosen = "dithered generic"; |
||
627 | fr->cpu_opts.type = generic_dither; |
||
628 | dithered = TRUE; |
||
629 | # ifndef NO_16BIT |
||
630 | fr->synths.plain[r_1to1][f_16] = synth_1to1_dither; |
||
631 | # ifndef NO_DOWNSAMPLE |
||
632 | fr->synths.plain[r_2to1][f_16] = synth_2to1_dither; |
||
633 | fr->synths.plain[r_4to1][f_16] = synth_4to1_dither; |
||
634 | # endif |
||
635 | # endif |
||
636 | done = 1; |
||
637 | } |
||
638 | #endif |
||
639 | |||
640 | # ifdef OPT_ALTIVEC |
||
641 | if(!done && (auto_choose || want_dec == altivec)) |
||
642 | { |
||
643 | chosen = "AltiVec"; |
||
644 | fr->cpu_opts.type = altivec; |
||
645 | # ifndef NO_16BIT |
||
646 | fr->synths.plain[r_1to1][f_16] = synth_1to1_altivec; |
||
647 | fr->synths.stereo[r_1to1][f_16] = synth_1to1_stereo_altivec; |
||
648 | # endif |
||
649 | # ifndef NO_REAL |
||
650 | fr->synths.plain[r_1to1][f_real] = synth_1to1_real_altivec; |
||
651 | fr->synths.stereo[r_1to1][f_real] = synth_1to1_real_stereo_altivec; |
||
652 | # endif |
||
653 | # ifndef NO_32BIT |
||
654 | fr->synths.plain[r_1to1][f_32] = synth_1to1_s32_altivec; |
||
655 | fr->synths.stereo[r_1to1][f_32] = synth_1to1_s32_stereo_altivec; |
||
656 | # endif |
||
657 | done = 1; |
||
658 | } |
||
659 | # endif |
||
660 | |||
661 | # ifdef OPT_ARM |
||
662 | if(!done && (auto_choose || want_dec == arm)) |
||
663 | { |
||
664 | chosen = "ARM"; |
||
665 | fr->cpu_opts.type = arm; |
||
666 | # ifndef NO_16BIT |
||
667 | fr->synths.plain[r_1to1][f_16] = synth_1to1_arm; |
||
668 | # endif |
||
669 | done = 1; |
||
670 | } |
||
671 | # endif |
||
672 | |||
673 | # ifdef OPT_GENERIC |
||
674 | if(!done && (auto_choose || want_dec == generic)) |
||
675 | { |
||
676 | chosen = "generic"; |
||
677 | fr->cpu_opts.type = generic; |
||
678 | done = 1; |
||
679 | } |
||
680 | # endif |
||
681 | |||
682 | fr->cpu_opts.class = decclass(fr->cpu_opts.type); |
||
683 | |||
684 | # ifndef NO_8BIT |
||
685 | # ifndef NO_16BIT /* possibility to use a 16->8 wrapper... */ |
||
686 | /* Last chance to use some optimized routine via generic wrappers (for 8bit). */ |
||
687 | if( fr->cpu_opts.type != ifuenf_dither |
||
688 | && fr->cpu_opts.type != generic_dither |
||
689 | && fr->synths.plain[r_1to1][f_16] != synth_base.plain[r_1to1][f_16] ) |
||
690 | { |
||
691 | fr->synths.plain[r_1to1][f_8] = synth_1to1_8bit_wrap; |
||
692 | fr->synths.mono[r_1to1][f_8] = synth_1to1_8bit_wrap_mono; |
||
693 | fr->synths.mono2stereo[r_1to1][f_8] = synth_1to1_8bit_wrap_mono2stereo; |
||
694 | } |
||
695 | # endif |
||
696 | # endif |
||
697 | |||
698 | #ifdef OPT_DITHER |
||
699 | if(done && dithered) |
||
700 | { |
||
701 | /* run-time dither noise table generation */ |
||
702 | if(!frame_dither_init(fr)) |
||
703 | { |
||
704 | if(NOQUIET) error("Dither noise setup failed!"); |
||
705 | return 0; |
||
706 | } |
||
707 | } |
||
708 | #endif |
||
709 | |||
710 | if(done) |
||
711 | { |
||
712 | if(VERBOSE) fprintf(stderr, "Decoder: %s\n", chosen); |
||
713 | return 1; |
||
714 | } |
||
715 | else |
||
716 | { |
||
717 | if(NOQUIET) error("Could not set optimization!"); |
||
718 | return 0; |
||
719 | } |
||
720 | } |
||
721 | |||
722 | enum optdec dectype(const char* decoder) |
||
723 | { |
||
724 | enum optdec dt; |
||
725 | if( (decoder == NULL) |
||
726 | || (decoder[0] == 0) ) |
||
727 | return autodec; |
||
728 | |||
729 | for(dt=autodec; dt |
||
730 | if(!strcasecmp(decoder, decname[dt])) return dt; |
||
731 | |||
732 | return nodec; /* If we found nothing... */ |
||
733 | } |
||
734 | |||
735 | #ifdef OPT_MULTI |
||
736 | |||
737 | /* same number of entries as full list, but empty at beginning */ |
||
738 | static const char *mpg123_supported_decoder_list[] = |
||
739 | { |
||
740 | #ifdef OPT_SSE |
||
741 | NULL, |
||
742 | #endif |
||
743 | #ifdef OPT_3DNOWEXT |
||
744 | NULL, |
||
745 | #endif |
||
746 | #ifdef OPT_3DNOW |
||
747 | NULL, |
||
748 | #endif |
||
749 | #ifdef OPT_MMX |
||
750 | NULL, |
||
751 | #endif |
||
752 | #ifdef OPT_I586 |
||
753 | NULL, |
||
754 | #endif |
||
755 | #ifdef OPT_I586_DITHER |
||
756 | NULL, |
||
757 | #endif |
||
758 | #ifdef OPT_I486 |
||
759 | NULL, |
||
760 | #endif |
||
761 | #ifdef OPT_I386 |
||
762 | NULL, |
||
763 | #endif |
||
764 | #ifdef OPT_ALTIVEC |
||
765 | NULL, |
||
766 | #endif |
||
767 | #ifdef OPT_X86_64 |
||
768 | NULL, |
||
769 | #endif |
||
770 | #ifdef OPT_ARM |
||
771 | NULL, |
||
772 | #endif |
||
773 | #ifdef OPT_GENERIC_FLOAT |
||
774 | NULL, |
||
775 | #endif |
||
776 | # ifdef OPT_GENERIC |
||
777 | NULL, |
||
778 | # endif |
||
779 | # ifdef OPT_GENERIC_DITHER |
||
780 | NULL, |
||
781 | # endif |
||
782 | NULL |
||
783 | }; |
||
784 | #endif |
||
785 | |||
786 | static const char *mpg123_decoder_list[] = |
||
787 | { |
||
788 | #ifdef OPT_SSE |
||
789 | dn_SSE, |
||
790 | #endif |
||
791 | #ifdef OPT_3DNOWEXT |
||
792 | dn_3DNowExt, |
||
793 | #endif |
||
794 | #ifdef OPT_3DNOW |
||
795 | dn_3DNow, |
||
796 | #endif |
||
797 | #ifdef OPT_MMX |
||
798 | dn_MMX, |
||
799 | #endif |
||
800 | #ifdef OPT_I586 |
||
801 | dn_i586, |
||
802 | #endif |
||
803 | #ifdef OPT_I586_DITHER |
||
804 | dn_i586_dither, |
||
805 | #endif |
||
806 | #ifdef OPT_I486 |
||
807 | dn_i486, |
||
808 | #endif |
||
809 | #ifdef OPT_I386 |
||
810 | dn_i386, |
||
811 | #endif |
||
812 | #ifdef OPT_ALTIVEC |
||
813 | dn_AltiVec, |
||
814 | #endif |
||
815 | #ifdef OPT_X86_64 |
||
816 | dn_x86_64, |
||
817 | #endif |
||
818 | #ifdef OPT_ARM |
||
819 | dn_ARM, |
||
820 | #endif |
||
821 | #ifdef OPT_GENERIC |
||
822 | dn_generic, |
||
823 | #endif |
||
824 | #ifdef OPT_GENERIC_DITHER |
||
825 | dn_generic_dither, |
||
826 | #endif |
||
827 | NULL |
||
828 | }; |
||
829 | |||
830 | void check_decoders(void ) |
||
831 | { |
||
832 | #ifndef OPT_MULTI |
||
833 | /* In non-multi mode, only the full list (one entry) is used. */ |
||
834 | return; |
||
835 | #else |
||
836 | const char **d = mpg123_supported_decoder_list; |
||
837 | #ifdef OPT_X86 |
||
838 | getcpuflags(&cpu_flags); |
||
839 | if(cpu_i586(cpu_flags)) |
||
840 | { |
||
841 | /* not yet: if(cpu_sse2(cpu_flags)) printf(" SSE2"); |
||
842 | if(cpu_sse3(cpu_flags)) printf(" SSE3"); */ |
||
843 | #ifdef OPT_SSE |
||
844 | if(cpu_sse(cpu_flags)) *(d++) = decname[sse]; |
||
845 | #endif |
||
846 | #ifdef OPT_3DNOWEXT |
||
847 | if(cpu_3dnowext(cpu_flags)) *(d++) = decname[dreidnowext]; |
||
848 | #endif |
||
849 | #ifdef OPT_3DNOW |
||
850 | if(cpu_3dnow(cpu_flags)) *(d++) = decname[dreidnow]; |
||
851 | #endif |
||
852 | #ifdef OPT_MMX |
||
853 | if(cpu_mmx(cpu_flags)) *(d++) = decname[mmx]; |
||
854 | #endif |
||
855 | #ifdef OPT_I586 |
||
856 | *(d++) = decname[ifuenf]; |
||
857 | #endif |
||
858 | #ifdef OPT_I586_DITHER |
||
859 | *(d++) = decname[ifuenf_dither]; |
||
860 | #endif |
||
861 | } |
||
862 | #endif |
||
863 | /* just assume that the i486 built is run on a i486 cpu... */ |
||
864 | #ifdef OPT_I486 |
||
865 | *(d++) = decname[ivier]; |
||
866 | #endif |
||
867 | #ifdef OPT_ALTIVEC |
||
868 | *(d++) = decname[altivec]; |
||
869 | #endif |
||
870 | /* every supported x86 can do i386, any cpu can do generic */ |
||
871 | #ifdef OPT_I386 |
||
872 | *(d++) = decname[idrei]; |
||
873 | #endif |
||
874 | #ifdef OPT_X86_64 |
||
875 | *(d++) = decname[x86_64]; |
||
876 | #endif |
||
877 | #ifdef OPT_ARM |
||
878 | *(d++) = decname[arm]; |
||
879 | #endif |
||
880 | #ifdef OPT_GENERIC |
||
881 | *(d++) = decname[generic]; |
||
882 | #endif |
||
883 | #ifdef OPT_GENERIC_DITHER |
||
884 | *(d++) = decname[generic_dither]; |
||
885 | #endif |
||
886 | #endif /* ndef OPT_MULTI */ |
||
887 | } |
||
888 | |||
889 | const char* attribute_align_arg mpg123_current_decoder(mpg123_handle *mh) |
||
890 | { |
||
891 | if(mh == NULL) return NULL; |
||
892 | |||
893 | return decname[mh->cpu_opts.type]; |
||
894 | } |
||
895 | |||
896 | const char attribute_align_arg **mpg123_decoders(){ return mpg123_decoder_list; } |
||
897 | const char attribute_align_arg **mpg123_supported_decoders() |
||
898 | { |
||
899 | #ifdef OPT_MULTI |
||
900 | return mpg123_supported_decoder_list; |
||
901 | #else |
||
902 | return mpg123_decoder_list; |
||
903 | #endif |
||
904 | } |