1 |
/*************************************************************************** |
2 |
* * |
3 |
* LinuxSampler - modular, streaming capable sampler * |
4 |
* * |
5 |
* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck * |
6 |
* Copyright (C) 2005, 2006 Christian Schoenebeck * |
7 |
* * |
8 |
* This program is free software; you can redistribute it and/or modify * |
9 |
* it under the terms of the GNU General Public License as published by * |
10 |
* the Free Software Foundation; either version 2 of the License, or * |
11 |
* (at your option) any later version. * |
12 |
* * |
13 |
* This program is distributed in the hope that it will be useful, * |
14 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of * |
15 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
16 |
* GNU General Public License for more details. * |
17 |
* * |
18 |
* You should have received a copy of the GNU General Public License * |
19 |
* along with this program; if not, write to the Free Software * |
20 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, * |
21 |
* MA 02111-1307 USA * |
22 |
***************************************************************************/ |
23 |
|
24 |
#ifndef RINGBUFFER_H |
25 |
#define RINGBUFFER_H |
26 |
|
27 |
#define DEFAULT_WRAP_ELEMENTS 0 |
28 |
|
29 |
#include <string.h> |
30 |
|
31 |
#include "atomic.h" |
32 |
|
33 |
/** @brief Real-time safe and type safe RingBuffer implementation. |
34 |
* |
35 |
* This constant size buffer can be used to send data from exactly one |
36 |
* sender / writing thread to exactly one receiver / reading thread. It is |
37 |
* real-time safe due to the fact that data is only allocated when this |
38 |
* RingBuffer is created and no system level mechanisms are used for |
39 |
* ensuring thread safety of this class. |
40 |
* |
41 |
* <b>Important:</b> There are two distinct behaviors of this RingBuffer |
42 |
* which has to be given as template argument @c T_DEEP_COPY, which is a |
43 |
* boolean flag: |
44 |
* |
45 |
* - @c true: The RingBuffer will copy elements of type @c T by using type |
46 |
* @c T's assignment operator. This behavior is mandatory for all data |
47 |
* structures (classes) which additionally allocate memory on the heap. |
48 |
* Type @c T's needs to have an assignment operator implementation though, |
49 |
* otherwise this will cause a compilation error. This behavior is more |
50 |
* safe, but usually slower (except for very small buffer sizes, where it |
51 |
* might be even faster). |
52 |
* - @c false: The RingBuffer will copy elements of type @c T by flatly |
53 |
* copying their structural data ( i.e. with @c memcpy() ) in one piece. |
54 |
* This will only work if class @c T (and all of its subelements) does not |
55 |
* allocate any additional data on the heap by itself. So use this option |
56 |
* with great care, because otherwise it will result in very ugly behavior |
57 |
* and crashes! For larger buffer sizes, this behavior will most probably |
58 |
* be faster. |
59 |
*/ |
60 |
template<class T, bool T_DEEP_COPY> |
61 |
class RingBuffer |
62 |
{ |
63 |
public: |
64 |
RingBuffer (int sz, int wrap_elements = DEFAULT_WRAP_ELEMENTS) { |
65 |
int power_of_two; |
66 |
|
67 |
this->wrap_elements = wrap_elements; |
68 |
|
69 |
for (power_of_two = 1; |
70 |
1<<power_of_two < sz; |
71 |
power_of_two++); |
72 |
|
73 |
size = 1<<power_of_two; |
74 |
size_mask = size; |
75 |
size_mask -= 1; |
76 |
atomic_set(&write_ptr, 0); |
77 |
atomic_set(&read_ptr, 0); |
78 |
buf = new T[size + wrap_elements]; |
79 |
}; |
80 |
|
81 |
virtual ~RingBuffer() { |
82 |
delete [] buf; |
83 |
} |
84 |
|
85 |
/** |
86 |
* Sets all remaining write space elements to zero. The write pointer |
87 |
* will currently not be incremented after, but that might change in |
88 |
* future. |
89 |
* |
90 |
* @e Caution: for @c T_DEEP_COPY=true you might probably @e NOT want |
91 |
* to call this method at all, at least not in case type @c T allocates |
92 |
* any additional data on the heap by itself. |
93 |
*/ |
94 |
inline void fill_write_space_with_null() { |
95 |
int w = atomic_read(&write_ptr), |
96 |
r = atomic_read(&read_ptr); |
97 |
memset(get_write_ptr(), 0, sizeof(T)*write_space_to_end()); |
98 |
if (r && w >= r) { |
99 |
memset(get_buffer_begin(), 0, sizeof(T)*(r - 1)); |
100 |
} |
101 |
|
102 |
// set the wrap space elements to null |
103 |
if (wrap_elements) memset(&buf[size], 0, sizeof(T)*wrap_elements); |
104 |
} |
105 |
|
106 |
__inline int read (T *dest, int cnt); |
107 |
__inline int write (T *src, int cnt); |
108 |
|
109 |
inline int push(T* src) { return write(src,1); } |
110 |
inline int pop(T* dst) { return read(dst,1); } |
111 |
|
112 |
__inline T *get_buffer_begin(); |
113 |
|
114 |
__inline T *get_read_ptr(void) { |
115 |
return(&buf[atomic_read(&read_ptr)]); |
116 |
} |
117 |
|
118 |
/** |
119 |
* Returns a pointer to the element from the current read position, |
120 |
* advanced by \a offset elements. |
121 |
*/ |
122 |
/*inline T* get_read_ptr(int offset) { |
123 |
int r = atomic_read(&read_ptr); |
124 |
r += offset; |
125 |
r &= size_mask; |
126 |
return &buf[r]; |
127 |
}*/ |
128 |
|
129 |
__inline T *get_write_ptr(); |
130 |
__inline void increment_read_ptr(int cnt) { |
131 |
atomic_set(&read_ptr , (atomic_read(&read_ptr) + cnt) & size_mask); |
132 |
} |
133 |
__inline void set_read_ptr(int val) { |
134 |
atomic_set(&read_ptr , val); |
135 |
} |
136 |
|
137 |
__inline void increment_write_ptr(int cnt) { |
138 |
atomic_set(&write_ptr, (atomic_read(&write_ptr) + cnt) & size_mask); |
139 |
} |
140 |
|
141 |
/* this function increments the write_ptr by cnt, if the buffer wraps then |
142 |
subtract size from the write_ptr value so that it stays within 0<write_ptr<size |
143 |
use this function to increment the write ptr after you filled the buffer |
144 |
with a number of elements given by write_space_to_end_with_wrap(). |
145 |
This ensures that the data that is written to the buffer fills up all |
146 |
the wrap space that resides past the regular buffer. The wrap_space is needed for |
147 |
interpolation. So that the audio thread sees the ringbuffer like a linear space |
148 |
which allows us to use faster routines. |
149 |
When the buffer wraps the wrap part is memcpy()ied to the beginning of the buffer |
150 |
and the write ptr incremented accordingly. |
151 |
*/ |
152 |
__inline void increment_write_ptr_with_wrap(int cnt) { |
153 |
int w=atomic_read(&write_ptr); |
154 |
w += cnt; |
155 |
if(w >= size) { |
156 |
w -= size; |
157 |
copy(&buf[0], &buf[size], w); |
158 |
//printf("DEBUG !!!! increment_write_ptr_with_wrap: buffer wrapped, elements wrapped = %d (wrap_elements %d)\n",w,wrap_elements); |
159 |
} |
160 |
atomic_set(&write_ptr, w); |
161 |
} |
162 |
|
163 |
/* this function returns the available write space in the buffer |
164 |
when the read_ptr > write_ptr it returns the space inbetween, otherwise |
165 |
when the read_ptr < write_ptr it returns the space between write_ptr and |
166 |
the buffer end, including the wrap_space. |
167 |
There is an exception to it. When read_ptr <= wrap_elements. In that |
168 |
case we return the write space to buffer end (-1) without the wrap_elements, |
169 |
this is needed because a subsequent increment_write_ptr which copies the |
170 |
data that resides into the wrap space to the beginning of the buffer and increments |
171 |
the write_ptr would cause the write_ptr overstepping the read_ptr which would be an error. |
172 |
So basically the if(r<=wrap_elements) we return the buffer space to end - 1 which |
173 |
ensures that at the next call there will be one element free to write before the buffer wraps |
174 |
and usually (except in EOF situations) the next write_space_to_end_with_wrap() will return |
175 |
1 + wrap_elements which ensures that the wrap part gets fully filled with data |
176 |
*/ |
177 |
__inline int write_space_to_end_with_wrap() { |
178 |
int w, r; |
179 |
|
180 |
w = atomic_read(&write_ptr); |
181 |
r = atomic_read(&read_ptr); |
182 |
//printf("write_space_to_end: w=%d r=%d\n",w,r); |
183 |
if(r > w) { |
184 |
//printf("DEBUG: write_space_to_end_with_wrap: r>w r=%d w=%d val=%d\n",r,w,r - w - 1); |
185 |
return(r - w - 1); |
186 |
} |
187 |
if(r <= wrap_elements) { |
188 |
//printf("DEBUG: write_space_to_end_with_wrap: ATTENTION r <= wrap_elements: r=%d w=%d val=%d\n",r,w,size - w -1); |
189 |
return(size - w -1); |
190 |
} |
191 |
if(r) { |
192 |
//printf("DEBUG: write_space_to_end_with_wrap: r=%d w=%d val=%d\n",r,w,size - w + wrap_elements); |
193 |
return(size - w + wrap_elements); |
194 |
} |
195 |
//printf("DEBUG: write_space_to_end_with_wrap: r=0 w=%d val=%d\n",w,size - w - 1 + wrap_elements); |
196 |
return(size - w - 1 + wrap_elements); |
197 |
} |
198 |
|
199 |
/* this function adjusts the number of elements to write into the ringbuffer |
200 |
in a way that the size boundary is avoided and that the wrap space always gets |
201 |
entirely filled. |
202 |
cnt contains the write_space_to_end_with_wrap() amount while |
203 |
capped_cnt contains a capped amount of samples to read. |
204 |
normally capped_cnt == cnt but in some cases eg when the disk thread needs |
205 |
to refill tracks with smaller blocks because lots of streams require immediate |
206 |
refill because lots of notes were started simultaneously. |
207 |
In that case we set for example capped_cnt to a fixed amount (< cnt, eg 64k), |
208 |
which helps to reduce the buffer refill latencies that occur between streams. |
209 |
the first if() checks if the current write_ptr + capped_cnt resides within |
210 |
the wrap area but is < size+wrap_elements. in that case we cannot return |
211 |
capped_cnt because it would lead to a write_ptr wrapping and only a partial fill |
212 |
of wrap space which would lead to errors. So we simply return cnt which ensures |
213 |
that the the entire wrap space will get filled correctly. |
214 |
In all other cases (which are not problematic because no write_ptr wrapping |
215 |
occurs) we simply return capped_cnt. |
216 |
*/ |
217 |
__inline int adjust_write_space_to_avoid_boundary(int cnt, int capped_cnt) { |
218 |
int w; |
219 |
w = atomic_read(&write_ptr); |
220 |
if((w+capped_cnt) >= size && (w+capped_cnt) < (size+wrap_elements)) { |
221 |
//printf("adjust_write_space_to_avoid_boundary returning cnt = %d\n",cnt); |
222 |
return(cnt); |
223 |
} |
224 |
//printf("adjust_write_space_to_avoid_boundary returning capped_cnt = %d\n",capped_cnt); |
225 |
return(capped_cnt); |
226 |
} |
227 |
|
228 |
__inline int write_space_to_end() { |
229 |
int w, r; |
230 |
|
231 |
w = atomic_read(&write_ptr); |
232 |
r = atomic_read(&read_ptr); |
233 |
//printf("write_space_to_end: w=%d r=%d\n",w,r); |
234 |
if(r > w) return(r - w - 1); |
235 |
if(r) return(size - w); |
236 |
return(size - w - 1); |
237 |
} |
238 |
|
239 |
__inline int read_space_to_end() { |
240 |
int w, r; |
241 |
|
242 |
w = atomic_read(&write_ptr); |
243 |
r = atomic_read(&read_ptr); |
244 |
if(w >= r) return(w - r); |
245 |
return(size - r); |
246 |
} |
247 |
__inline void init() { |
248 |
atomic_set(&write_ptr, 0); |
249 |
atomic_set(&read_ptr, 0); |
250 |
// wrap=0; |
251 |
} |
252 |
|
253 |
int write_space () { |
254 |
int w, r; |
255 |
|
256 |
w = atomic_read(&write_ptr); |
257 |
r = atomic_read(&read_ptr); |
258 |
|
259 |
if (w > r) { |
260 |
return ((r - w + size) & size_mask) - 1; |
261 |
} else if (w < r) { |
262 |
return (r - w) - 1; |
263 |
} else { |
264 |
return size - 1; |
265 |
} |
266 |
} |
267 |
|
268 |
int read_space () { |
269 |
int w, r; |
270 |
|
271 |
w = atomic_read(&write_ptr); |
272 |
r = atomic_read(&read_ptr); |
273 |
|
274 |
if (w >= r) { |
275 |
return w - r; |
276 |
} else { |
277 |
return (w - r + size) & size_mask; |
278 |
} |
279 |
} |
280 |
|
281 |
int size; |
282 |
int wrap_elements; |
283 |
|
284 |
/** |
285 |
* Independent, random access reading from a RingBuffer. This class |
286 |
* allows to read from a RingBuffer without being forced to free read |
287 |
* data while reading / positioning. |
288 |
*/ |
289 |
template<class T1, bool T1_DEEP_COPY> |
290 |
class _NonVolatileReader { |
291 |
public: |
292 |
int read_space() { |
293 |
int r = read_ptr; |
294 |
int w = atomic_read(&pBuf->write_ptr); |
295 |
return (w >= r) ? w - r : (w - r + pBuf->size) & pBuf->size_mask; |
296 |
} |
297 |
|
298 |
/** |
299 |
* Prefix decrement operator, for reducing NonVolatileReader's |
300 |
* read position by one. |
301 |
*/ |
302 |
inline void operator--() { |
303 |
if (read_ptr == atomic_read(&pBuf->read_ptr)) return; //TODO: or should we react oh this case (e.g. force segfault), as this is a very odd case? |
304 |
read_ptr = (read_ptr-1) & pBuf->size_mask; |
305 |
} |
306 |
|
307 |
/** |
308 |
* Postfix decrement operator, for reducing NonVolatileReader's |
309 |
* read position by one. |
310 |
*/ |
311 |
inline void operator--(int) { |
312 |
--*this; |
313 |
} |
314 |
|
315 |
/** |
316 |
* Returns pointer to the RingBuffer data of current |
317 |
* NonVolatileReader's read position and increments |
318 |
* NonVolatileReader's read position by one. |
319 |
* |
320 |
* @returns pointer to element of current read position |
321 |
*/ |
322 |
T* pop() { |
323 |
if (!read_space()) return NULL; |
324 |
T* pData = &pBuf->buf[read_ptr]; |
325 |
read_ptr++; |
326 |
read_ptr &= pBuf->size_mask; |
327 |
return pData; |
328 |
} |
329 |
|
330 |
/** |
331 |
* Reads one element from the NonVolatileReader's current read |
332 |
* position and copies it to the variable pointed by \a dst and |
333 |
* finally increments the NonVolatileReader's read position by |
334 |
* one. |
335 |
* |
336 |
* @param dst - where the element is copied to |
337 |
* @returns 1 on success, 0 otherwise |
338 |
*/ |
339 |
int pop(T* dst) { return read(dst,1); } |
340 |
|
341 |
/** |
342 |
* Reads \a cnt elements from the NonVolatileReader's current |
343 |
* read position and copies it to the buffer pointed by \a dest |
344 |
* and finally increments the NonVolatileReader's read position |
345 |
* by the number of read elements. |
346 |
* |
347 |
* @param dest - destination buffer |
348 |
* @param cnt - number of elements to read |
349 |
* @returns number of read elements |
350 |
*/ |
351 |
int read(T* dest, int cnt) { |
352 |
int free_cnt; |
353 |
int cnt2; |
354 |
int to_read; |
355 |
int n1, n2; |
356 |
int priv_read_ptr; |
357 |
|
358 |
priv_read_ptr = read_ptr; |
359 |
|
360 |
if ((free_cnt = read_space()) == 0) return 0; |
361 |
|
362 |
to_read = cnt > free_cnt ? free_cnt : cnt; |
363 |
|
364 |
cnt2 = priv_read_ptr + to_read; |
365 |
|
366 |
if (cnt2 > pBuf->size) { |
367 |
n1 = pBuf->size - priv_read_ptr; |
368 |
n2 = cnt2 & pBuf->size_mask; |
369 |
} else { |
370 |
n1 = to_read; |
371 |
n2 = 0; |
372 |
} |
373 |
|
374 |
copy(dest, &pBuf->buf[priv_read_ptr], n1); |
375 |
priv_read_ptr = (priv_read_ptr + n1) & pBuf->size_mask; |
376 |
|
377 |
if (n2) { |
378 |
copy(dest+n1, pBuf->buf, n2); |
379 |
priv_read_ptr = n2; |
380 |
} |
381 |
|
382 |
this->read_ptr = priv_read_ptr; |
383 |
return to_read; |
384 |
} |
385 |
|
386 |
/** |
387 |
* Finally when the read data is not needed anymore, this method |
388 |
* should be called to free the data in the RingBuffer up to the |
389 |
* current read position of this NonVolatileReader. |
390 |
* |
391 |
* @see RingBuffer::increment_read_ptr() |
392 |
*/ |
393 |
void free() { |
394 |
atomic_set(&pBuf->read_ptr, read_ptr); |
395 |
} |
396 |
|
397 |
protected: |
398 |
_NonVolatileReader(RingBuffer<T1,T1_DEEP_COPY>* pBuf) { |
399 |
this->pBuf = pBuf; |
400 |
this->read_ptr = atomic_read(&pBuf->read_ptr); |
401 |
} |
402 |
|
403 |
RingBuffer<T1,T1_DEEP_COPY>* pBuf; |
404 |
int read_ptr; |
405 |
|
406 |
friend class RingBuffer<T1,T1_DEEP_COPY>; |
407 |
}; |
408 |
|
409 |
typedef _NonVolatileReader<T,T_DEEP_COPY> NonVolatileReader; |
410 |
|
411 |
NonVolatileReader get_non_volatile_reader() { return NonVolatileReader(this); } |
412 |
|
413 |
protected: |
414 |
T *buf; |
415 |
atomic_t write_ptr; |
416 |
atomic_t read_ptr; |
417 |
int size_mask; |
418 |
|
419 |
/** |
420 |
* Copies \a n amount of elements from the buffer given by |
421 |
* \a pSrc to the buffer given by \a pDst. |
422 |
*/ |
423 |
inline static void copy(T* pDst, T* pSrc, int n); |
424 |
|
425 |
friend class _NonVolatileReader<T,T_DEEP_COPY>; |
426 |
}; |
427 |
|
428 |
template<class T, bool T_DEEP_COPY> |
429 |
T* RingBuffer<T,T_DEEP_COPY>::get_write_ptr (void) { |
430 |
return(&buf[atomic_read(&write_ptr)]); |
431 |
} |
432 |
|
433 |
template<class T, bool T_DEEP_COPY> |
434 |
T* RingBuffer<T,T_DEEP_COPY>::get_buffer_begin (void) { |
435 |
return(buf); |
436 |
} |
437 |
|
438 |
|
439 |
|
440 |
template<class T, bool T_DEEP_COPY> |
441 |
int RingBuffer<T,T_DEEP_COPY>::read(T* dest, int cnt) |
442 |
{ |
443 |
int free_cnt; |
444 |
int cnt2; |
445 |
int to_read; |
446 |
int n1, n2; |
447 |
int priv_read_ptr; |
448 |
|
449 |
priv_read_ptr=atomic_read(&read_ptr); |
450 |
|
451 |
if ((free_cnt = read_space ()) == 0) { |
452 |
return 0; |
453 |
} |
454 |
|
455 |
to_read = cnt > free_cnt ? free_cnt : cnt; |
456 |
|
457 |
cnt2 = priv_read_ptr + to_read; |
458 |
|
459 |
if (cnt2 > size) { |
460 |
n1 = size - priv_read_ptr; |
461 |
n2 = cnt2 & size_mask; |
462 |
} else { |
463 |
n1 = to_read; |
464 |
n2 = 0; |
465 |
} |
466 |
|
467 |
copy(dest, &buf[priv_read_ptr], n1); |
468 |
priv_read_ptr = (priv_read_ptr + n1) & size_mask; |
469 |
|
470 |
if (n2) { |
471 |
copy(dest+n1, buf, n2); |
472 |
priv_read_ptr = n2; |
473 |
} |
474 |
|
475 |
atomic_set(&read_ptr, priv_read_ptr); |
476 |
return to_read; |
477 |
} |
478 |
|
479 |
template<class T, bool T_DEEP_COPY> |
480 |
int RingBuffer<T,T_DEEP_COPY>::write(T* src, int cnt) |
481 |
{ |
482 |
int free_cnt; |
483 |
int cnt2; |
484 |
int to_write; |
485 |
int n1, n2; |
486 |
int priv_write_ptr; |
487 |
|
488 |
priv_write_ptr=atomic_read(&write_ptr); |
489 |
|
490 |
if ((free_cnt = write_space ()) == 0) { |
491 |
return 0; |
492 |
} |
493 |
|
494 |
to_write = cnt > free_cnt ? free_cnt : cnt; |
495 |
|
496 |
cnt2 = priv_write_ptr + to_write; |
497 |
|
498 |
if (cnt2 > size) { |
499 |
n1 = size - priv_write_ptr; |
500 |
n2 = cnt2 & size_mask; |
501 |
} else { |
502 |
n1 = to_write; |
503 |
n2 = 0; |
504 |
} |
505 |
|
506 |
copy(&buf[priv_write_ptr], src, n1); |
507 |
priv_write_ptr = (priv_write_ptr + n1) & size_mask; |
508 |
|
509 |
if (n2) { |
510 |
copy(buf, src+n1, n2); |
511 |
priv_write_ptr = n2; |
512 |
} |
513 |
atomic_set(&write_ptr, priv_write_ptr); |
514 |
return to_write; |
515 |
} |
516 |
|
517 |
template<class T, bool T_DEEP_COPY> |
518 |
void RingBuffer<T,T_DEEP_COPY>::copy(T* pDst, T* pSrc, int n) { |
519 |
if (T_DEEP_COPY) { // deep copy - won't work for data structures without assignment operator implementation |
520 |
for (int i = 0; i < n; i++) pDst[i] = pSrc[i]; |
521 |
} else { // flat copy - won't work for complex data structures ! |
522 |
memcpy(pDst, pSrc, n * sizeof(T)); |
523 |
} |
524 |
} |
525 |
|
526 |
#endif /* RINGBUFFER_H */ |