diff options
author | Maurus Cuelenaere <mcuelenaere@gmail.com> | 2009-07-07 13:36:34 +0000 |
---|---|---|
committer | Maurus Cuelenaere <mcuelenaere@gmail.com> | 2009-07-07 13:36:34 +0000 |
commit | 8d4d4610b6814c0c4a4abe9523b4b6b2b80ac9a4 (patch) | |
tree | 5823f6ea576834eb0b7fd9f1fc78a85997d0e224 /apps/fixedpoint.c | |
parent | 616905f965ed626c398624ec2ffa988f69899c32 (diff) | |
download | rockbox-8d4d4610b6814c0c4a4abe9523b4b6b2b80ac9a4.tar.gz rockbox-8d4d4610b6814c0c4a4abe9523b4b6b2b80ac9a4.zip |
* FS#10411 - Fixed point math code is bloated by Jeffrey Goode
* Set svn:keywords properties
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@21701 a1c6a512-1295-4272-9138-f99709370657
Diffstat (limited to 'apps/fixedpoint.c')
-rw-r--r-- | apps/fixedpoint.c | 131 |
1 files changed, 41 insertions, 90 deletions
diff --git a/apps/fixedpoint.c b/apps/fixedpoint.c index 917f624258..f9903f301f 100644 --- a/apps/fixedpoint.c +++ b/apps/fixedpoint.c | |||
@@ -5,7 +5,7 @@ | |||
5 | * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < < | 5 | * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < < |
6 | * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \ | 6 | * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \ |
7 | * \/ \/ \/ \/ \/ | 7 | * \/ \/ \/ \/ \/ |
8 | * $Id: fixedpoint.c -1 $ | 8 | * $Id$ |
9 | * | 9 | * |
10 | * Copyright (C) 2006 Jens Arnold | 10 | * Copyright (C) 2006 Jens Arnold |
11 | * | 11 | * |
@@ -261,24 +261,18 @@ long fp16_log(int x) { | |||
261 | 261 | ||
262 | #if (!defined(PLUGIN) && !defined(CODEC)) | 262 | #if (!defined(PLUGIN) && !defined(CODEC)) |
263 | /** MODIFIED FROM replaygain.c */ | 263 | /** MODIFIED FROM replaygain.c */ |
264 | /* These math routines have 64-bit internal precision to avoid overflows. | ||
265 | * Arguments and return values are 32-bit (long) precision. | ||
266 | */ | ||
267 | 264 | ||
268 | #define FP_MUL64(x, y) (((x) * (y)) >> (fracbits)) | 265 | #define FP_MUL_FRAC(x, y) fp_mul(x, y, fracbits) |
269 | #define FP_DIV64(x, y) (((x) << (fracbits)) / (y)) | 266 | #define FP_DIV_FRAC(x, y) fp_div(x, y, fracbits) |
270 | |||
271 | static long long fp_exp10(long long x, unsigned int fracbits); | ||
272 | /* static long long fp_log10(long long n, unsigned int fracbits); */ | ||
273 | 267 | ||
274 | /* constants in fixed point format, 28 fractional bits */ | 268 | /* constants in fixed point format, 28 fractional bits */ |
275 | #define FP28_LN2 (186065279LL) /* ln(2) */ | 269 | #define FP28_LN2 (186065279L) /* ln(2) */ |
276 | #define FP28_LN2_INV (387270501LL) /* 1/ln(2) */ | 270 | #define FP28_LN2_INV (387270501L) /* 1/ln(2) */ |
277 | #define FP28_EXP_ZERO (44739243LL) /* 1/6 */ | 271 | #define FP28_EXP_ZERO (44739243L) /* 1/6 */ |
278 | #define FP28_EXP_ONE (-745654LL) /* -1/360 */ | 272 | #define FP28_EXP_ONE (-745654L) /* -1/360 */ |
279 | #define FP28_EXP_TWO (12428LL) /* 1/21600 */ | 273 | #define FP28_EXP_TWO (12428L) /* 1/21600 */ |
280 | #define FP28_LN10 (618095479LL) /* ln(10) */ | 274 | #define FP28_LN10 (618095479L) /* ln(10) */ |
281 | #define FP28_LOG10OF2 (80807124LL) /* log10(2) */ | 275 | #define FP28_LOG10OF2 (80807124L) /* log10(2) */ |
282 | 276 | ||
283 | #define TOL_BITS 2 /* log calculation tolerance */ | 277 | #define TOL_BITS 2 /* log calculation tolerance */ |
284 | 278 | ||
@@ -290,24 +284,24 @@ static long long fp_exp10(long long x, unsigned int fracbits); | |||
290 | /** FIXED POINT EXP10 | 284 | /** FIXED POINT EXP10 |
291 | * Return 10^x as FP integer. Argument is FP integer. | 285 | * Return 10^x as FP integer. Argument is FP integer. |
292 | */ | 286 | */ |
293 | static long long fp_exp10(long long x, unsigned int fracbits) | 287 | static long fp_exp10(long x, unsigned int fracbits) |
294 | { | 288 | { |
295 | long long k; | 289 | long k; |
296 | long long z; | 290 | long z; |
297 | long long R; | 291 | long R; |
298 | long long xp; | 292 | long xp; |
299 | 293 | ||
300 | /* scale constants */ | 294 | /* scale constants */ |
301 | const long long fp_one = (1 << fracbits); | 295 | const long fp_one = (1 << fracbits); |
302 | const long long fp_half = (1 << (fracbits - 1)); | 296 | const long fp_half = (1 << (fracbits - 1)); |
303 | const long long fp_two = (2 << fracbits); | 297 | const long fp_two = (2 << fracbits); |
304 | const long long fp_mask = (fp_one - 1); | 298 | const long fp_mask = (fp_one - 1); |
305 | const long long fp_ln2_inv = (FP28_LN2_INV >> (28 - fracbits)); | 299 | const long fp_ln2_inv = (FP28_LN2_INV >> (28 - fracbits)); |
306 | const long long fp_ln2 = (FP28_LN2 >> (28 - fracbits)); | 300 | const long fp_ln2 = (FP28_LN2 >> (28 - fracbits)); |
307 | const long long fp_ln10 = (FP28_LN10 >> (28 - fracbits)); | 301 | const long fp_ln10 = (FP28_LN10 >> (28 - fracbits)); |
308 | const long long fp_exp_zero = (FP28_EXP_ZERO >> (28 - fracbits)); | 302 | const long fp_exp_zero = (FP28_EXP_ZERO >> (28 - fracbits)); |
309 | const long long fp_exp_one = (FP28_EXP_ONE >> (28 - fracbits)); | 303 | const long fp_exp_one = (FP28_EXP_ONE >> (28 - fracbits)); |
310 | const long long fp_exp_two = (FP28_EXP_TWO >> (28 - fracbits)); | 304 | const long fp_exp_two = (FP28_EXP_TWO >> (28 - fracbits)); |
311 | 305 | ||
312 | /* exp(0) = 1 */ | 306 | /* exp(0) = 1 */ |
313 | if (x == 0) | 307 | if (x == 0) |
@@ -316,21 +310,21 @@ static long long fp_exp10(long long x, unsigned int fracbits) | |||
316 | } | 310 | } |
317 | 311 | ||
318 | /* convert from base 10 to base e */ | 312 | /* convert from base 10 to base e */ |
319 | x = FP_MUL64(x, fp_ln10); | 313 | x = FP_MUL_FRAC(x, fp_ln10); |
320 | 314 | ||
321 | /* calculate exp(x) */ | 315 | /* calculate exp(x) */ |
322 | k = (FP_MUL64(abs(x), fp_ln2_inv) + fp_half) & ~fp_mask; | 316 | k = (FP_MUL_FRAC(abs(x), fp_ln2_inv) + fp_half) & ~fp_mask; |
323 | 317 | ||
324 | if (x < 0) | 318 | if (x < 0) |
325 | { | 319 | { |
326 | k = -k; | 320 | k = -k; |
327 | } | 321 | } |
328 | 322 | ||
329 | x -= FP_MUL64(k, fp_ln2); | 323 | x -= FP_MUL_FRAC(k, fp_ln2); |
330 | z = FP_MUL64(x, x); | 324 | z = FP_MUL_FRAC(x, x); |
331 | R = fp_two + FP_MUL64(z, fp_exp_zero + FP_MUL64(z, fp_exp_one | 325 | R = fp_two + FP_MUL_FRAC(z, fp_exp_zero + FP_MUL_FRAC(z, fp_exp_one |
332 | + FP_MUL64(z, fp_exp_two))); | 326 | + FP_MUL_FRAC(z, fp_exp_two))); |
333 | xp = fp_one + FP_DIV64(FP_MUL64(fp_two, x), R - x); | 327 | xp = fp_one + FP_DIV_FRAC(FP_MUL_FRAC(fp_two, x), R - x); |
334 | 328 | ||
335 | if (k < 0) | 329 | if (k < 0) |
336 | { | 330 | { |
@@ -341,7 +335,7 @@ static long long fp_exp10(long long x, unsigned int fracbits) | |||
341 | k = fp_one << (k >> fracbits); | 335 | k = fp_one << (k >> fracbits); |
342 | } | 336 | } |
343 | 337 | ||
344 | return FP_MUL64(k, xp); | 338 | return FP_MUL_FRAC(k, xp); |
345 | } | 339 | } |
346 | 340 | ||
347 | 341 | ||
@@ -349,13 +343,13 @@ static long long fp_exp10(long long x, unsigned int fracbits) | |||
349 | /** FIXED POINT LOG10 | 343 | /** FIXED POINT LOG10 |
350 | * Return log10(x) as FP integer. Argument is FP integer. | 344 | * Return log10(x) as FP integer. Argument is FP integer. |
351 | */ | 345 | */ |
352 | static long long fp_log10(long long n, unsigned int fracbits) | 346 | static long fp_log10(long n, unsigned int fracbits) |
353 | { | 347 | { |
354 | /* Calculate log2 of argument */ | 348 | /* Calculate log2 of argument */ |
355 | 349 | ||
356 | long long log2, frac; | 350 | long log2, frac; |
357 | const long long fp_one = (1 << fracbits); | 351 | const long fp_one = (1 << fracbits); |
358 | const long long fp_two = (2 << fracbits); | 352 | const long fp_two = (2 << fracbits); |
359 | const long tolerance = (1 << ((fracbits / 2) + 2)); | 353 | const long tolerance = (1 << ((fracbits / 2) + 2)); |
360 | 354 | ||
361 | if (n <=0) return FP_NEGINF; | 355 | if (n <=0) return FP_NEGINF; |
@@ -378,7 +372,7 @@ static long long fp_log10(long long n, unsigned int fracbits) | |||
378 | while (frac > tolerance) | 372 | while (frac > tolerance) |
379 | { | 373 | { |
380 | frac >>= 1; | 374 | frac >>= 1; |
381 | n = FP_MUL64(n, n); | 375 | n = FP_MUL_FRAC(n, n); |
382 | if (n >= fp_two) | 376 | if (n >= fp_two) |
383 | { | 377 | { |
384 | n >>= 1; | 378 | n >>= 1; |
@@ -387,31 +381,15 @@ static long long fp_log10(long long n, unsigned int fracbits) | |||
387 | } | 381 | } |
388 | 382 | ||
389 | /* convert log2 to log10 */ | 383 | /* convert log2 to log10 */ |
390 | return FP_MUL64(log2, (FP28_LOG10OF2 >> (28 - fracbits))); | 384 | return FP_MUL_FRAC(log2, (FP28_LOG10OF2 >> (28 - fracbits))); |
391 | } | 385 | } |
392 | 386 | ||
393 | 387 | ||
394 | /** CONVERT FACTOR TO DECIBELS */ | 388 | /** CONVERT FACTOR TO DECIBELS */ |
395 | long fp_decibels(unsigned long factor, unsigned int fracbits) | 389 | long fp_decibels(unsigned long factor, unsigned int fracbits) |
396 | { | 390 | { |
397 | long long decibels; | ||
398 | long long f = (long long)factor; | ||
399 | bool neg; | ||
400 | |||
401 | /* keep factor in signed long range */ | ||
402 | if (f >= (1LL << 31)) | ||
403 | f = (1LL << 31) - 1; | ||
404 | |||
405 | /* decibels = 20 * log10(factor) */ | 391 | /* decibels = 20 * log10(factor) */ |
406 | decibels = FP_MUL64((20LL << fracbits), fp_log10(f, fracbits)); | 392 | return FP_MUL_FRAC((20L << fracbits), fp_log10(factor, fracbits)); |
407 | |||
408 | /* keep result in signed long range */ | ||
409 | if ((neg = (decibels < 0))) | ||
410 | decibels = -decibels; | ||
411 | if (decibels >= (1LL << 31)) | ||
412 | return neg ? FP_NEGINF : FP_INF; | ||
413 | |||
414 | return neg ? (long)-decibels : (long)decibels; | ||
415 | } | 393 | } |
416 | #endif /* unused code */ | 394 | #endif /* unused code */ |
417 | 395 | ||
@@ -419,34 +397,7 @@ long fp_decibels(unsigned long factor, unsigned int fracbits) | |||
419 | /** CONVERT DECIBELS TO FACTOR */ | 397 | /** CONVERT DECIBELS TO FACTOR */ |
420 | long fp_factor(long decibels, unsigned int fracbits) | 398 | long fp_factor(long decibels, unsigned int fracbits) |
421 | { | 399 | { |
422 | bool neg; | ||
423 | long long factor; | ||
424 | long long db = (long long)decibels; | ||
425 | |||
426 | /* if decibels is 0, factor is 1 */ | ||
427 | if (db == 0) | ||
428 | return (1L << fracbits); | ||
429 | |||
430 | /* calculate for positive decibels only */ | ||
431 | if ((neg = (db < 0))) | ||
432 | db = -db; | ||
433 | |||
434 | /* factor = 10 ^ (decibels / 20) */ | 400 | /* factor = 10 ^ (decibels / 20) */ |
435 | factor = fp_exp10(FP_DIV64(db, (20LL << fracbits)), fracbits); | 401 | return fp_exp10(FP_DIV_FRAC(decibels, (20L << fracbits)), fracbits); |
436 | |||
437 | /* keep result in signed long range, return 0 if very small */ | ||
438 | if (factor >= (1LL << 31)) | ||
439 | { | ||
440 | if (neg) | ||
441 | return 0; | ||
442 | else | ||
443 | return FP_INF; | ||
444 | } | ||
445 | |||
446 | /* if negative argument, factor is 1 / result */ | ||
447 | if (neg) | ||
448 | factor = FP_DIV64((1LL << fracbits), factor); | ||
449 | |||
450 | return (long)factor; | ||
451 | } | 402 | } |
452 | #endif /* !PLUGIN and !CODEC */ | 403 | #endif /* !PLUGIN and !CODEC */ |