ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/OpenCL/OpenCL.xs
(Generate patch)

Comparing OpenCL/OpenCL.xs (file contents):
Revision 1.27 by root, Thu Apr 19 13:18:31 2012 UTC vs.
Revision 1.73 by root, Sat May 5 13:30:07 2012 UTC

1#include "EXTERN.h" 1#include "EXTERN.h"
2#include "perl.h" 2#include "perl.h"
3#include "XSUB.h" 3#include "XSUB.h"
4
5#include "ecb.h"//D
6
7#define X_STACKSIZE sizeof (void *) * 512 * 1024 // 2-4mb should be enough, really
8#include "xthread.h"
9#include "schmorp.h"
10
11#ifdef I_DLFCN
12 #include <dlfcn.h>
13#endif
14
15// how stupid is that, the 1.2 header files define CL_VERSION_1_1,
16// but then fail to define the api functions unless you ALSO define
17// this. This breaks 100% of the opencl 1.1 apps, for what reason?
18// after all, the functions are deprecated, not removed.
19// in addition, you cannot test for this in any future-proof way.
20// each time a new opencl version comes out, you need to make a new
21// release.
22#define CL_USE_DEPRECATED_OPENCL_1_2_APIS /* just guessing, you stupid idiots */
23
24#ifndef PREFER_1_1
25 #define PREFER_1_1 1
26#endif
27
28#if PREFER_1_1
29 #define CL_USE_DEPRECATED_OPENCL_1_1_APIS
30#endif
4 31
5#ifdef __APPLE__ 32#ifdef __APPLE__
6 #include <OpenCL/opencl.h> 33 #include <OpenCL/opencl.h>
7#else 34#else
8 #include <CL/opencl.h> 35 #include <CL/opencl.h>
9#endif 36#endif
10 37
38#ifndef CL_VERSION_1_2
39 #undef PREFER_1_1
40 #define PREFER_1_1 1
41#endif
42
11typedef cl_platform_id OpenCL__Platform; 43typedef cl_platform_id OpenCL__Platform;
12typedef cl_device_id OpenCL__Device; 44typedef cl_device_id OpenCL__Device;
45typedef cl_device_id OpenCL__SubDevice;
13typedef cl_context OpenCL__Context; 46typedef cl_context OpenCL__Context;
14typedef cl_command_queue OpenCL__Queue; 47typedef cl_command_queue OpenCL__Queue;
15typedef cl_mem OpenCL__Memory; 48typedef cl_mem OpenCL__Memory;
16typedef cl_mem OpenCL__Buffer; 49typedef cl_mem OpenCL__Buffer;
17typedef cl_mem OpenCL__BufferObj; 50typedef cl_mem OpenCL__BufferObj;
18typedef cl_mem OpenCL__Image; 51typedef cl_mem OpenCL__Image;
19typedef cl_mem OpenCL__Image2D;
20typedef cl_mem OpenCL__Image3D;
21typedef cl_mem OpenCL__Memory_ornull; 52typedef cl_mem OpenCL__Memory_ornull;
22typedef cl_mem OpenCL__Buffer_ornull; 53typedef cl_mem OpenCL__Buffer_ornull;
23typedef cl_mem OpenCL__Image_ornull; 54typedef cl_mem OpenCL__Image_ornull;
24typedef cl_mem OpenCL__Image2D_ornull;
25typedef cl_mem OpenCL__Image3D_ornull;
26typedef cl_sampler OpenCL__Sampler; 55typedef cl_sampler OpenCL__Sampler;
27typedef cl_program OpenCL__Program; 56typedef cl_program OpenCL__Program;
28typedef cl_kernel OpenCL__Kernel; 57typedef cl_kernel OpenCL__Kernel;
29typedef cl_event OpenCL__Event; 58typedef cl_event OpenCL__Event;
30typedef cl_event OpenCL__UserEvent; 59typedef cl_event OpenCL__UserEvent;
31 60
32typedef SV *FUTURE; 61typedef struct mapped * OpenCL__Mapped;
62
63static HV
64 *stash_platform,
65 *stash_device,
66 *stash_subdevice,
67 *stash_context,
68 *stash_queue,
69 *stash_program,
70 *stash_kernel,
71 *stash_sampler,
72 *stash_event,
73 *stash_userevent,
74 *stash_memory,
75 *stash_buffer,
76 *stash_bufferobj,
77 *stash_image,
78 *stash_image1d,
79 *stash_image1darray,
80 *stash_image1dbuffer,
81 *stash_image2d,
82 *stash_image2darray,
83 *stash_image3d,
84 *stash_mapped,
85 *stash_mappedbuffer,
86 *stash_mappedimage;
33 87
34/*****************************************************************************/ 88/*****************************************************************************/
35 89
90// name must include a leading underscore
91// all of this horrors would be unneceesary if somebody wrote a proper OpenGL module
92// for perl. doh.
93static void *
94glsym (const char *name)
95{
96 void *fun = 0;
97
98 #if defined I_DLFCN && defined RTLD_DEFAULT
99 fun = dlsym (RTLD_DEFAULT, name + 1);
100 if (!fun) fun = dlsym (RTLD_DEFAULT, name);
101
102 if (!fun)
103 {
104 static void *libgl;
105 static const char *glso[] = {
106 "libGL.so.1",
107 "libGL.so.3",
108 "libGL.so.4.0",
109 "libGL.so",
110 "/usr/lib/libGL.so",
111 "/usr/X11R6/lib/libGL.1.dylib"
112 };
113 int i;
114
115 for (i = 0; !libgl && i < sizeof (glso) / sizeof (glso [0]); ++i)
116 {
117 libgl = dlopen (glso [i], RTLD_LAZY);
118 if (libgl)
119 break;
120 }
121
122 if (libgl)
123 {
124 fun = dlsym (libgl, name + 1);
125 if (!fun) fun = dlsym (libgl, name);
126 }
127 }
128 #endif
129
130 return fun;
131}
132
133/*****************************************************************************/
134
36/* up to two temporary buffers */ 135/* up to two temporary buffers */
37static void * 136static void * ecb_noinline
38tmpbuf (size_t size) 137tmpbuf (size_t size)
39{ 138{
40 enum { buffers = 3 }; 139 enum { buffers = 4 };
41 static int idx; 140 static int idx;
42 static void *buf [buffers]; 141 static void *buf [buffers];
43 static size_t len [buffers]; 142 static size_t len [buffers];
44 143
45 idx = ++idx % buffers; 144 idx = (idx + 1) % buffers;
46 145
47 if (len [idx] < size) 146 if (len [idx] < size)
48 { 147 {
49 free (buf [idx]); 148 free (buf [idx]);
50 len [idx] = ((size + 31) & ~4095) + 4096 - 32; 149 len [idx] = ((size + 31) & ~4095) + 4096 - 32;
51 buf [idx] = malloc (len [idx]); 150 buf [idx] = malloc (len [idx]);
52 } 151 }
53 152
54 return buf [idx]; 153 return buf [idx];
154}
155
156static const char * ecb_noinline
157cv_get_name (CV *cv)
158{
159 static char fullname [256];
160
161 GV *gv = CvGV (cv); // gv better be nonzero
162
163 HV *stash = GvSTASH (gv);
164 const char *hvname = HvNAME_get (stash); // stash also better be nonzero
165 const char *gvname = GvNAME (gv);
166
167 snprintf (fullname, sizeof (fullname), "%s::%s", hvname, gvname);
168 return fullname;
55} 169}
56 170
57/*****************************************************************************/ 171/*****************************************************************************/
58 172
59typedef struct 173typedef struct
118 if (res) \ 232 if (res) \
119 FAIL (name); 233 FAIL (name);
120 234
121/*****************************************************************************/ 235/*****************************************************************************/
122 236
123#define NEW_MORTAL_OBJ(class,ptr) sv_setref_pv (sv_newmortal (), class, ptr)
124#define XPUSH_NEW_OBJ(class,ptr) XPUSHs (NEW_MORTAL_OBJ (class, ptr))
125
126static void * 237static SV *
238new_clobj (HV *stash, IV id)
239{
240 return sv_2mortal (sv_bless (newRV_noinc (newSViv (id)), stash));
241}
242
243#define PUSH_CLOBJ(stash,id) PUSHs (new_clobj ((stash), (IV)(id)))
244#define XPUSH_CLOBJ(stash,id) XPUSHs (new_clobj ((stash), (IV)(id)))
245
246/* cl objects are either \$iv, or [$iv, ...] */
247/* they can be upgraded at runtime to the array form */
248static void * ecb_noinline
127SvPTROBJ (const char *func, const char *svname, SV *sv, const char *pkg) 249SvCLOBJ (CV *cv, const char *svname, SV *sv, const char *pkg)
128{ 250{
251 // sv_derived_from is quite slow :(
129 if (SvROK (sv) && sv_derived_from (sv, pkg)) 252 if (SvROK (sv) && sv_derived_from (sv, pkg))
130 return (void *)SvIV (SvRV (sv)); 253 return (void *)SvIV (SvRV (sv));
131 254
132 croak ("%s: %s is not of type %s", func, svname, pkg); 255 croak ("%s: %s is not of type %s", cv_get_name (cv), svname, pkg);
256}
257
258// the "no-inherit" version of the above
259static void * ecb_noinline
260SvCLOBJ_ni (CV *cv, const char *svname, SV *sv, HV *stash)
261{
262 if (SvROK (sv) && SvSTASH (SvRV (sv)) == stash)
263 return (void *)SvIV (SvRV (sv));
264
265 croak ("%s: %s is not of type %s", cv_get_name (cv), svname, HvNAME (stash));
133} 266}
134 267
135/*****************************************************************************/ 268/*****************************************************************************/
136 269
137static cl_context_properties * 270static cl_context_properties * ecb_noinline
138SvCONTEXTPROPERTIES (const char *func, const char *svname, SV *sv, cl_context_properties *extra, int extracount) 271SvCONTEXTPROPERTIES (CV *cv, const char *svname, SV *sv, cl_context_properties *extra, int extracount)
139{ 272{
140 if (!sv || !SvOK (sv)) 273 if (!sv || !SvOK (sv))
141 if (extra) 274 if (extra)
142 sv = sv_2mortal (newRV_noinc ((SV *)newAV ())); // slow, but rarely used hopefully 275 sv = sv_2mortal (newRV_noinc ((SV *)newAV ())); // slow, but rarely used hopefully
143 else 276 else
144 return 0; 277 return 0;
145 278
146 if (SvROK (sv) && SvTYPE (SvRV (sv)) == SVt_PVAV) 279 if (SvROK (sv) && SvTYPE (SvRV (sv)) == SVt_PVAV)
147 { 280 {
148 AV *av = (AV *)SvRV (sv); 281 AV *av = (AV *)SvRV (sv);
149 int i, len = av_len (av); 282 int i, len = av_len (av) + 1;
150 cl_context_properties *p = tmpbuf (sizeof (cl_context_properties) * (len + extracount + 1)); 283 cl_context_properties *p = tmpbuf (sizeof (cl_context_properties) * (len + extracount + 1));
151 cl_context_properties *l = p; 284 cl_context_properties *l = p;
152 285
153 if (len & 1) 286 if (len & 1)
154 croak ("%s: %s is not a property list (must be even number of elements)", func, svname); 287 croak ("%s: %s is not a property list (must contain an even number of elements)", cv_get_name (cv), svname);
155 288
156 while (extracount--) 289 while (extracount--)
157 *l++ = *extra++; 290 *l++ = *extra++;
158 291
159 for (i = 0; i < len; i += 2) 292 for (i = 0; i < len; i += 2)
160 { 293 {
161 cl_context_properties t = SvIV (*av_fetch (av, i, 0)); 294 cl_context_properties t = SvIV (*av_fetch (av, i , 0));
162 cl_context_properties v; 295 SV *p_sv = *av_fetch (av, i + 1, 0);
296 cl_context_properties v = SvIV (p_sv); // code below can override
163 297
164 switch (t) 298 switch (t)
165 { 299 {
300 case CL_CONTEXT_PLATFORM:
301 if (SvROK (p_sv))
302 v = (cl_context_properties)SvCLOBJ (cv, svname, p_sv, "OpenCL::Platform");
303 break;
304
305 case CL_GLX_DISPLAY_KHR:
306 if (!SvOK (p_sv))
307 {
308 void *func = glsym ("_glXGetCurrentDisplay");
309 if (func)
310 v = (cl_context_properties)((void *(*)(void))func)();
311 }
312 break;
313
314 case CL_GL_CONTEXT_KHR:
315 if (!SvOK (p_sv))
316 {
317 void *func = glsym ("_glXGetCurrentContext");
318 if (func)
319 v = (cl_context_properties)((void *(*)(void))func)();
320 }
321 break;
322
166 default: 323 default:
167 /* unknown property, treat as int */ 324 /* unknown property, treat as int */
168 v = SvIV (*av_fetch (av, i, 0));
169 break; 325 break;
170 } 326 }
171 327
172 *l++ = t; 328 *l++ = t;
173 *l++ = v; 329 *l++ = v;
176 *l = 0; 332 *l = 0;
177 333
178 return p; 334 return p;
179 } 335 }
180 336
181 croak ("%s: %s is not a property list (either undef or [type => value, ...])", func, svname); 337 croak ("%s: %s is not a property list (either undef or [type => value, ...])", cv_get_name (cv), svname);
338}
339
340// parse an array of CLOBJ into a void ** array in C - works only for CLOBJs whose representation
341// is a pointer (and only on well-behaved systems).
342static void * ecb_noinline
343object_list (CV *cv, int or_undef, const char *argname, SV *arg, const char *klass, cl_uint *rcount)
344{
345 if (!SvROK (arg) || SvTYPE (SvRV (arg)) != SVt_PVAV)
346 croak ("%s: '%s' parameter must be %sa reference to an array of %s objects",
347 cv_get_name (cv), argname, or_undef ? "undef or " : "", klass);
348
349 AV *av = (AV *)SvRV (arg);
350 void **list = 0;
351 cl_uint count = av_len (av) + 1;
352
353 if (count)
354 {
355 list = tmpbuf (sizeof (*list) * count);
356 int i;
357 for (i = 0; i < count; ++i)
358 list [i] = SvCLOBJ (cv, argname, *av_fetch (av, i, 1), klass);
359 }
360
361 if (!count && !or_undef)
362 croak ("%s: '%s' must contain at least one %s object",
363 cv_get_name (cv), argname, klass);
364
365 *rcount = count;
366 return (void *)list;
182} 367}
183 368
184/*****************************************************************************/ 369/*****************************************************************************/
370/* callback stuff */
371
372/* default context callback, log to stderr */
373static void CL_CALLBACK
374context_default_notify (const char *msg, const void *info, size_t cb, void *data)
375{
376 fprintf (stderr, "OpenCL Context Notify: %s\n", msg);
377}
378
379typedef struct
380{
381 int free_cb;
382 void (*push)(void *data1, void *data2, void *data3);
383} eq_vtbl;
384
385typedef struct eq_item
386{
387 struct eq_item *next;
388 eq_vtbl *vtbl;
389 SV *cb;
390 void *data1, *data2, *data3;
391} eq_item;
392
393static void (*eq_signal_func)(void *signal_arg, int value);
394static void *eq_signal_arg;
395static xmutex_t eq_lock = X_MUTEX_INIT;
396static eq_item *eq_head, *eq_tail;
397
398static void ecb_noinline
399eq_enq (eq_vtbl *vtbl, SV *cb, void *data1, void *data2, void *data3)
400{
401 eq_item *item = malloc (sizeof (eq_item));
402
403 item->next = 0;
404 item->vtbl = vtbl;
405 item->cb = cb;
406 item->data1 = data1;
407 item->data2 = data2;
408 item->data3 = data3;
409
410 X_LOCK (eq_lock);
411
412 *(eq_head ? &eq_tail->next : &eq_head) = item;
413 eq_tail = item;
414
415 X_UNLOCK (eq_lock);
416
417 eq_signal_func (eq_signal_arg, 0);
418}
419
420static eq_item *
421eq_dec (void)
422{
423 eq_item *res;
424
425 X_LOCK (eq_lock);
426
427 res = eq_head;
428 if (res)
429 eq_head = res->next;
430
431 X_UNLOCK (eq_lock);
432
433 return res;
434}
435
436static void
437eq_poll (void)
438{
439 eq_item *item;
440
441 while ((item = eq_dec ()))
442 {
443 ENTER;
444 SAVETMPS;
445
446 dSP;
447 PUSHMARK (SP);
448 EXTEND (SP, 2);
449
450 if (item->vtbl->free_cb)
451 sv_2mortal (item->cb);
452
453 PUTBACK;
454 item->vtbl->push (item->data1, item->data2, item->data3);
455
456 SV *cb = item->cb;
457 free (item);
458
459 call_sv (cb, G_DISCARD | G_VOID);
460
461 FREETMPS;
462 LEAVE;
463 }
464}
465
466static void
467eq_poll_interrupt (pTHX_ void *c_arg, int value)
468{
469 eq_poll ();
470}
471
472/*****************************************************************************/
473/* context notify */
474
475static void ecb_noinline
476eq_context_push (void *data1, void *data2, void *data3)
477{
478 dSP;
479 PUSHs (sv_2mortal (newSVpv (data1, 0)));
480 PUSHs (sv_2mortal (newSVpvn (data2, (STRLEN)data3)));
481 PUTBACK;
482
483 free (data1);
484 free (data2);
485}
486
487static eq_vtbl eq_context_vtbl = { 0, eq_context_push };
488
489static void CL_CALLBACK
490eq_context_notify (const char *msg, const void *pvt, size_t cb, void *user_data)
491{
492 void *pvt_copy = malloc (cb);
493 memcpy (pvt_copy, pvt, cb);
494 eq_enq (&eq_context_vtbl, user_data, strdup (msg), pvt_copy, (void *)cb);
495}
496
497#define CONTEXT_NOTIFY_CALLBACK \
498 void (CL_CALLBACK *pfn_notify)(const char *, const void *, size_t, void *) = context_default_notify; \
499 void *user_data = 0; \
500 \
501 if (SvOK (notify)) \
502 { \
503 pfn_notify = eq_context_notify; \
504 user_data = s_get_cv (notify); \
505 }
506
507static SV * ecb_noinline
508new_clobj_context (cl_context ctx, void *user_data)
509{
510 SV *sv = new_clobj (stash_context, (IV)ctx);
511
512 if (user_data)
513 sv_magicext (SvRV (sv), user_data, PERL_MAGIC_ext, 0, 0, 0);
514
515 return sv;
516}
517
518#define XPUSH_CLOBJ_CONTEXT XPUSHs (new_clobj_context (ctx, user_data));
519
520/*****************************************************************************/
521/* build/compile/link notify */
522
523static void
524eq_program_push (void *data1, void *data2, void *data3)
525{
526 dSP;
527 PUSH_CLOBJ (stash_program, data1);
528 PUTBACK;
529}
530
531static eq_vtbl eq_program_vtbl = { 1, eq_program_push };
532
533static void CL_CALLBACK
534eq_program_notify (cl_program program, void *user_data)
535{
536 clRetainProgram (program);
537
538 eq_enq (&eq_program_vtbl, user_data, (void *)program, 0, 0);
539}
540
541typedef void (CL_CALLBACK *program_callback)(cl_program program, void *user_data);
542
543static program_callback ecb_noinline
544make_program_callback (SV *notify, void **ruser_data)
545{
546 if (SvOK (notify))
547 {
548 *ruser_data = SvREFCNT_inc (s_get_cv (notify));
549 return eq_program_notify;
550 }
551 else
552 {
553 *ruser_data = 0;
554 return 0;
555 }
556}
557
558struct build_args
559{
560 cl_program program;
561 char *options;
562 void *user_data;
563 cl_uint num_devices;
564};
565
566X_THREAD_PROC (build_program_thread)
567{
568 struct build_args *arg = thr_arg;
569
570 clBuildProgram (arg->program, arg->num_devices, arg->num_devices ? (void *)(arg + 1) : 0, arg->options, 0, 0);
571
572 if (arg->user_data)
573 eq_program_notify (arg->program, arg->user_data);
574 else
575 clReleaseProgram (arg->program);
576
577 free (arg->options);
578 free (arg);
579
580 return 0;
581}
582
583static void
584build_program_async (cl_program program, cl_uint num_devices, const cl_device_id *device_list, const char *options, void *user_data)
585{
586 struct build_args *arg = malloc (sizeof (struct build_args) + sizeof (*device_list) * num_devices);
587
588 arg->program = program;
589 arg->options = strdup (options);
590 arg->user_data = user_data;
591 arg->num_devices = num_devices;
592 memcpy (arg + 1, device_list, sizeof (*device_list) * num_devices);
593
594 xthread_t id;
595 thread_create (&id, build_program_thread, arg);
596}
597
598/*****************************************************************************/
599/* event objects */
600
601static void
602eq_event_push (void *data1, void *data2, void *data3)
603{
604 dSP;
605 PUSH_CLOBJ (stash_event, data1);
606 PUSHs (sv_2mortal (newSViv ((IV)data2)));
607 PUTBACK;
608}
609
610static eq_vtbl eq_event_vtbl = { 1, eq_event_push };
611
612static void CL_CALLBACK
613eq_event_notify (cl_event event, cl_int event_command_exec_status, void *user_data)
614{
615 clRetainEvent (event);
616 eq_enq (&eq_event_vtbl, user_data, (void *)event, (void *)(IV)event_command_exec_status, 0);
617}
618
619/*****************************************************************************/
620/* utilities for XS code */
185 621
186static size_t 622static size_t
187img_row_pitch (cl_mem img) 623img_row_pitch (cl_mem img)
188{ 624{
189 size_t res; 625 size_t res;
190 clGetImageInfo (img, CL_IMAGE_ROW_PITCH, sizeof (res), &res, 0); 626 clGetImageInfo (img, CL_IMAGE_ROW_PITCH, sizeof (res), &res, 0);
191 return res; 627 return res;
192} 628}
193 629
194static cl_event * 630static cl_event * ecb_noinline
195event_list (SV **items, int count) 631event_list (CV *cv, SV **items, cl_uint *rcount, cl_event extra)
196{ 632{
197 if (!count) 633 cl_uint count = *rcount;
634
635 if (count > 0x7fffffffU) // yeah, it's a hack - the caller might have underflowed
636 *rcount = count = 0;
637
638 if (!count && !extra)
198 return 0; 639 return 0;
199 640
200 cl_event *list = tmpbuf (sizeof (cl_event) * count); 641 cl_event *list = tmpbuf (sizeof (cl_event) * (count + 1));
642 int i = 0;
201 643
202 while (count--) 644 while (count--)
645 if (SvOK (items [count]))
203 list [count] = SvPTROBJ ("clEnqueue", "wait_events", items [count], "OpenCL::Event"); 646 list [i++] = SvCLOBJ (cv, "wait_events", items [count], "OpenCL::Event");
204 647
648 if (extra)
649 list [i++] = extra;
650
651 *rcount = i;
652
205 return list; 653 return i ? list : 0;
206} 654}
207 655
208#define EVENT_LIST(items,count) \ 656#define EVENT_LIST(skip) \
209 cl_uint event_list_count = (count); \ 657 cl_uint event_list_count = items - (skip); \
210 cl_event *event_list_ptr = event_list (&ST (items), event_list_count) 658 cl_event *event_list_ptr = event_list (cv, &ST (skip), &event_list_count, 0)
211 659
212#define INFO(class) \ 660#define INFO(class) \
213{ \ 661{ \
214 size_t size; \ 662 size_t size; \
215 NEED_SUCCESS (Get ## class ## Info, (self, name, 0, 0, &size)); \ 663 NEED_SUCCESS (Get ## class ## Info, (self, name, 0, 0, &size)); \
219 SvCUR_set (sv, size); \ 667 SvCUR_set (sv, size); \
220 NEED_SUCCESS (Get ## class ## Info, (self, name, size, SvPVX (sv), 0)); \ 668 NEED_SUCCESS (Get ## class ## Info, (self, name, size, SvPVX (sv), 0)); \
221 XPUSHs (sv); \ 669 XPUSHs (sv); \
222} 670}
223 671
672/*****************************************************************************/
673/* mapped_xxx */
674
675static OpenCL__Mapped
676SvMAPPED (SV *sv)
677{
678 // no typechecking atm., keep your fingers crossed
679 return (OpenCL__Mapped)SvMAGIC (SvRV (sv))->mg_ptr;
680}
681
682struct mapped
683{
684 cl_command_queue queue;
685 cl_mem memobj;
686 void *ptr;
687 size_t cb;
688 cl_event event;
689 size_t row_pitch;
690 size_t slice_pitch;
691};
692
693static SV *
694mapped_new (HV *stash, cl_command_queue queue, cl_mem memobj, cl_map_flags flags, void *ptr, size_t cb, cl_event ev, size_t row_pitch, size_t slice_pitch)
695{
696 SV *data = newSV (0);
697 SvUPGRADE (data, SVt_PVMG);
698
699 OpenCL__Mapped mapped;
700 New (0, mapped, 1, struct mapped);
701
702 clRetainCommandQueue (queue);
703
704 mapped->queue = queue;
705 mapped->memobj = memobj;
706 mapped->ptr = ptr;
707 mapped->cb = cb;
708 mapped->event = ev;
709 mapped->row_pitch = row_pitch;
710 mapped->slice_pitch = slice_pitch;
711
712 sv_magicext (data, 0, PERL_MAGIC_ext, 0, (char *)mapped, 0);
713
714 if (SvLEN (data))
715 Safefree (data);
716
717 SvPVX (data) = (char *)ptr;
718 SvCUR_set (data, cb);
719 SvLEN_set (data, 0);
720 SvPOK_only (data);
721
722 SV *obj = sv_2mortal (sv_bless (newRV_noinc (data), stash));
723
724 if (!(flags & CL_MAP_WRITE))
725 SvREADONLY_on (data);
726
727 return obj;
728}
729
730static void
731mapped_detach (SV *sv, OpenCL__Mapped mapped)
732{
733 SV *data = SvRV (sv);
734
735 // the next check checks both before AND after detach, where SvPVX should STILL be 0
736 if (SvPVX (data) != (char *)mapped->ptr)
737 warn ("FATAL: OpenCL memory mapped scalar changed location, detected");
738 else
739 {
740 SvREADONLY_off (data);
741 SvCUR_set (data, 0);
742 SvPVX (data) = 0;
743 SvOK_off (data);
744 }
745
746 mapped->ptr = 0;
747}
748
749static void
750mapped_unmap (CV *cv, SV *self, OpenCL__Mapped mapped, cl_command_queue queue, SV **wait_list, cl_uint event_list_count)
751{
752 cl_event *event_list_ptr = event_list (cv, wait_list, &event_list_count, mapped->event);
753 cl_event ev;
754
755 NEED_SUCCESS (EnqueueUnmapMemObject, (queue, mapped->memobj, mapped->ptr, event_list_count, event_list_ptr, &ev));
756
757 clReleaseEvent (mapped->event);
758 mapped->event = ev;
759
760 mapped_detach (self, mapped);
761}
762
763/*****************************************************************************/
764
224MODULE = OpenCL PACKAGE = OpenCL 765MODULE = OpenCL PACKAGE = OpenCL
225 766
226PROTOTYPES: ENABLE 767PROTOTYPES: ENABLE
768
769void
770poll ()
771 CODE:
772 eq_poll ();
773
774void
775_eq_initialise (IV func, IV arg)
776 CODE:
777 eq_signal_func = (void (*)(void *, int))func;
778 eq_signal_arg = (void*)arg;
227 779
228BOOT: 780BOOT:
229{ 781{
230 HV *stash = gv_stashpv ("OpenCL", 1); 782 HV *stash = gv_stashpv ("OpenCL", 1);
231 static const ivstr *civ, const_iv[] = { 783 static const ivstr *civ, const_iv[] = {
238 { sizeof (cl_long ), "SIZEOF_LONG" }, 790 { sizeof (cl_long ), "SIZEOF_LONG" },
239 { sizeof (cl_ulong ), "SIZEOF_ULONG" }, 791 { sizeof (cl_ulong ), "SIZEOF_ULONG" },
240 { sizeof (cl_half ), "SIZEOF_HALF" }, 792 { sizeof (cl_half ), "SIZEOF_HALF" },
241 { sizeof (cl_float ), "SIZEOF_FLOAT" }, 793 { sizeof (cl_float ), "SIZEOF_FLOAT" },
242 { sizeof (cl_double), "SIZEOF_DOUBLE" }, 794 { sizeof (cl_double), "SIZEOF_DOUBLE" },
795 { PREFER_1_1 , "PREFER_1_1" },
243#include "constiv.h" 796#include "constiv.h"
244 }; 797 };
798
245 for (civ = const_iv + sizeof (const_iv) / sizeof (const_iv [0]); civ > const_iv; civ--) 799 for (civ = const_iv + sizeof (const_iv) / sizeof (const_iv [0]); civ > const_iv; civ--)
246 newCONSTSUB (stash, (char *)civ[-1].name, newSViv (civ[-1].iv)); 800 newCONSTSUB (stash, (char *)civ[-1].name, newSViv (civ[-1].iv));
801
802 stash_platform = gv_stashpv ("OpenCL::Platform", GV_ADD);
803 stash_device = gv_stashpv ("OpenCL::Device", GV_ADD);
804 stash_subdevice = gv_stashpv ("OpenCL::SubDevice", GV_ADD);
805 stash_context = gv_stashpv ("OpenCL::Context", GV_ADD);
806 stash_queue = gv_stashpv ("OpenCL::Queue", GV_ADD);
807 stash_program = gv_stashpv ("OpenCL::Program", GV_ADD);
808 stash_kernel = gv_stashpv ("OpenCL::Kernel", GV_ADD);
809 stash_sampler = gv_stashpv ("OpenCL::Sampler", GV_ADD);
810 stash_event = gv_stashpv ("OpenCL::Event", GV_ADD);
811 stash_userevent = gv_stashpv ("OpenCL::UserEvent", GV_ADD);
812 stash_memory = gv_stashpv ("OpenCL::Memory", GV_ADD);
813 stash_buffer = gv_stashpv ("OpenCL::Buffer", GV_ADD);
814 stash_bufferobj = gv_stashpv ("OpenCL::BufferObj", GV_ADD);
815 stash_image = gv_stashpv ("OpenCL::Image", GV_ADD);
816 stash_image1d = gv_stashpv ("OpenCL::Image1D", GV_ADD);
817 stash_image1darray = gv_stashpv ("OpenCL::Image1DArray", GV_ADD);
818 stash_image1dbuffer = gv_stashpv ("OpenCL::Image1DBuffer", GV_ADD);
819 stash_image2d = gv_stashpv ("OpenCL::Image2D", GV_ADD);
820 stash_image2darray = gv_stashpv ("OpenCL::Image2DArray", GV_ADD);
821 stash_image3d = gv_stashpv ("OpenCL::Image3D", GV_ADD);
822 stash_mapped = gv_stashpv ("OpenCL::Mapped", GV_ADD);
823 stash_mappedbuffer = gv_stashpv ("OpenCL::MappedBuffer", GV_ADD);
824 stash_mappedimage = gv_stashpv ("OpenCL::MappedImage", GV_ADD);
825
826 sv_setiv (perl_get_sv ("OpenCL::POLL_FUNC", TRUE), (IV)eq_poll_interrupt);
247} 827}
248 828
249cl_int 829cl_int
250errno () 830errno ()
251 CODE: 831 CODE:
252 errno = res; 832 RETVAL = res;
833 OUTPUT:
834 RETVAL
253 835
254const char * 836const char *
255err2str (cl_int err) 837err2str (cl_int err = res)
256 838
257const char * 839const char *
258enum2str (cl_uint value) 840enum2str (cl_uint value)
259 841
260void 842void
268 list = tmpbuf (sizeof (*list) * count); 850 list = tmpbuf (sizeof (*list) * count);
269 NEED_SUCCESS (GetPlatformIDs, (count, list, 0)); 851 NEED_SUCCESS (GetPlatformIDs, (count, list, 0));
270 852
271 EXTEND (SP, count); 853 EXTEND (SP, count);
272 for (i = 0; i < count; ++i) 854 for (i = 0; i < count; ++i)
273 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Platform", list [i])); 855 PUSH_CLOBJ (stash_platform, list [i]);
274 856
275void 857void
276context_from_type (cl_context_properties *properties = 0, cl_device_type type = CL_DEVICE_TYPE_DEFAULT, FUTURE notify = 0) 858context_from_type (cl_context_properties *properties = 0, cl_device_type type = CL_DEVICE_TYPE_DEFAULT, SV *notify = &PL_sv_undef)
277 PPCODE: 859 PPCODE:
860 CONTEXT_NOTIFY_CALLBACK;
278 NEED_SUCCESS_ARG (cl_context ctx, CreateContextFromType, (properties, type, 0, 0, &res)); 861 NEED_SUCCESS_ARG (cl_context ctx, CreateContextFromType, (properties, type, pfn_notify, user_data, &res));
279 XPUSH_NEW_OBJ ("OpenCL::Context", ctx); 862 XPUSH_CLOBJ_CONTEXT;
280 863
281void 864void
282context (cl_context_properties *properties = 0, FUTURE devices, FUTURE notify = 0) 865context (cl_context_properties *properties, SV *devices, SV *notify = &PL_sv_undef)
283 PPCODE: 866 PPCODE:
284 /* der Gipfel der Kunst */ 867 cl_uint device_count;
868 cl_device_id *device_list = object_list (cv, 0, "devices", devices, "OpenCL::Device", &device_count);
869
870 CONTEXT_NOTIFY_CALLBACK;
871 NEED_SUCCESS_ARG (cl_context ctx, CreateContext, (properties, device_count, device_list, pfn_notify, user_data, &res));
872 XPUSH_CLOBJ_CONTEXT;
285 873
286void 874void
287wait_for_events (...) 875wait_for_events (...)
288 CODE: 876 CODE:
289 EVENT_LIST (0, items); 877 EVENT_LIST (0);
290 NEED_SUCCESS (WaitForEvents, (event_list_count, event_list_ptr)); 878 NEED_SUCCESS (WaitForEvents, (event_list_count, event_list_ptr));
291 879
292PROTOTYPES: DISABLE 880PROTOTYPES: DISABLE
293 881
294MODULE = OpenCL PACKAGE = OpenCL::Platform 882MODULE = OpenCL PACKAGE = OpenCL::Platform
295 883
296void 884void
297info (OpenCL::Platform self, cl_platform_info name) 885info (OpenCL::Platform self, cl_platform_info name)
298 PPCODE: 886 PPCODE:
299 INFO (Platform) 887 INFO (Platform)
888
889void
890unload_compiler (OpenCL::Platform self)
891 CODE:
892#if CL_VERSION_1_2
893 clUnloadPlatformCompiler (self);
894#endif
300 895
301#BEGIN:platform 896#BEGIN:platform
302 897
303void 898void
304profile (OpenCL::Platform self) 899profile (OpenCL::Platform self)
330 list = tmpbuf (sizeof (*list) * count); 925 list = tmpbuf (sizeof (*list) * count);
331 NEED_SUCCESS (GetDeviceIDs, (self, type, count, list, 0)); 926 NEED_SUCCESS (GetDeviceIDs, (self, type, count, list, 0));
332 927
333 EXTEND (SP, count); 928 EXTEND (SP, count);
334 for (i = 0; i < count; ++i) 929 for (i = 0; i < count; ++i)
335 PUSHs (sv_setref_pv (sv_newmortal (), "OpenCL::Device", list [i])); 930 PUSH_CLOBJ (stash_device, list [i]);
336 931
337void 932void
338context (OpenCL::Platform self, cl_context_properties *properties = 0, SV *devices, FUTURE notify = 0) 933context (OpenCL::Platform self, SV *properties, SV *devices, SV *notify = &PL_sv_undef)
339 PPCODE: 934 PPCODE:
935 cl_context_properties extra[] = { CL_CONTEXT_PLATFORM, (cl_context_properties)self };
936 cl_context_properties *props = SvCONTEXTPROPERTIES (cv, "properties", properties, extra, 2);
937
938 cl_uint device_count;
939 cl_device_id *device_list = object_list (cv, 0, "devices", devices, "OpenCL::Device", &device_count);
940
941 CONTEXT_NOTIFY_CALLBACK;
942 NEED_SUCCESS_ARG (cl_context ctx, CreateContext, (props, device_count, device_list, pfn_notify, user_data, &res));
943 XPUSH_CLOBJ_CONTEXT;
944
945void
946context_from_type (OpenCL::Platform self, SV *properties = 0, cl_device_type type = CL_DEVICE_TYPE_DEFAULT, SV *notify = &PL_sv_undef)
947 PPCODE:
948 cl_context_properties extra[] = { CL_CONTEXT_PLATFORM, (cl_context_properties)self };
949 cl_context_properties *props = SvCONTEXTPROPERTIES (cv, "properties", properties, extra, 2);
950
951 CONTEXT_NOTIFY_CALLBACK;
952 NEED_SUCCESS_ARG (cl_context ctx, CreateContextFromType, (props, type, pfn_notify, user_data, &res));
953 XPUSH_CLOBJ_CONTEXT;
954
955MODULE = OpenCL PACKAGE = OpenCL::Device
956
957void
958info (OpenCL::Device self, cl_device_info name)
959 PPCODE:
960 INFO (Device)
961
962#if CL_VERSION_1_2
963
964void
965sub_devices (OpenCL::Device self, SV *properties)
966 PPCODE:
340 if (!SvROK (devices) || SvTYPE (SvRV (devices)) != SVt_PVAV) 967 if (!SvROK (properties) || SvTYPE (SvRV (properties)) != SVt_PVAV)
341 croak ("OpenCL::Platform::context argument 'device' must be an arrayref with device objects, in call"); 968 croak ("OpenCL::Device::sub_devices: properties must be specified as reference to an array of property-value pairs");
342 969
343 AV *av = (AV *)SvRV (devices); 970 properties = SvRV (properties);
344 cl_uint num_devices = av_len (av) + 1; 971
345 cl_device_id *device_list = tmpbuf (sizeof (cl_device_id) * num_devices); 972 cl_uint count = av_len ((AV *)properties) + 1;
973 cl_device_partition_property *props = tmpbuf (sizeof (*props) * count + 1);
974
346 int i; 975 int i;
976 for (i = 0; i < count; ++i)
977 props [i] = (cl_device_partition_property)SvIV (*av_fetch ((AV *)properties, i, 0));
347 978
348 for (i = num_devices; i--; ) 979 props [count] = 0;
349 device_list [i] = SvPTROBJ ("clCreateContext", "devices", *av_fetch (av, i, 0), "OpenCL::Device");
350 980
351 NEED_SUCCESS_ARG (cl_context ctx, CreateContext, (properties, num_devices, device_list, 0, 0, &res)); 981 cl_uint num_devices;
352 XPUSH_NEW_OBJ ("OpenCL::Context", ctx); 982 NEED_SUCCESS (CreateSubDevices, (self, props, 0, 0, &num_devices));
983 cl_device_id *list = tmpbuf (sizeof (*list) * num_devices);
984 NEED_SUCCESS (CreateSubDevices, (self, props, num_devices, list, 0));
353 985
354void 986 EXTEND (SP, num_devices);
355context_from_type (OpenCL::Platform self, SV *properties = 0, cl_device_type type = CL_DEVICE_TYPE_DEFAULT, FUTURE notify = 0) 987 for (i = 0; i < count; ++i)
356 PPCODE: 988 PUSH_CLOBJ (stash_subdevice, list [i]);
357 cl_context_properties extra[] = { CL_CONTEXT_PLATFORM, (cl_context_properties)self };
358 cl_context_properties *props = SvCONTEXTPROPERTIES ("OpenCL::Platform::context_from_type", "properties", properties, extra, 2);
359 NEED_SUCCESS_ARG (cl_context ctx, CreateContextFromType, (props, type, 0, 0, &res));
360 XPUSH_NEW_OBJ ("OpenCL::Context", ctx);
361 989
362MODULE = OpenCL PACKAGE = OpenCL::Device 990#endif
363
364void
365info (OpenCL::Device self, cl_device_info name)
366 PPCODE:
367 INFO (Device)
368 991
369#BEGIN:device 992#BEGIN:device
370 993
371void 994void
372type (OpenCL::Device self) 995type (OpenCL::Device self)
373 PPCODE: 996 PPCODE:
374 cl_device_type value [1]; 997 cl_device_type value [1];
375 NEED_SUCCESS (GetDeviceInfo, (self, CL_DEVICE_TYPE, sizeof (value), value, 0)); 998 NEED_SUCCESS (GetDeviceInfo, (self, CL_DEVICE_TYPE, sizeof (value), value, 0));
376 EXTEND (SP, 1); 999 EXTEND (SP, 1);
377 const int i = 0; 1000 const int i = 0;
378 PUSHs (sv_2mortal (newSViv (value [i]))); 1001 PUSHs (sv_2mortal (newSVuv (value [i])));
379 1002
380void 1003void
381vendor_id (OpenCL::Device self) 1004vendor_id (OpenCL::Device self)
382 ALIAS: 1005 ALIAS:
383 vendor_id = CL_DEVICE_VENDOR_ID 1006 vendor_id = CL_DEVICE_VENDOR_ID
404 native_vector_width_int = CL_DEVICE_NATIVE_VECTOR_WIDTH_INT 1027 native_vector_width_int = CL_DEVICE_NATIVE_VECTOR_WIDTH_INT
405 native_vector_width_long = CL_DEVICE_NATIVE_VECTOR_WIDTH_LONG 1028 native_vector_width_long = CL_DEVICE_NATIVE_VECTOR_WIDTH_LONG
406 native_vector_width_float = CL_DEVICE_NATIVE_VECTOR_WIDTH_FLOAT 1029 native_vector_width_float = CL_DEVICE_NATIVE_VECTOR_WIDTH_FLOAT
407 native_vector_width_double = CL_DEVICE_NATIVE_VECTOR_WIDTH_DOUBLE 1030 native_vector_width_double = CL_DEVICE_NATIVE_VECTOR_WIDTH_DOUBLE
408 native_vector_width_half = CL_DEVICE_NATIVE_VECTOR_WIDTH_HALF 1031 native_vector_width_half = CL_DEVICE_NATIVE_VECTOR_WIDTH_HALF
409 reference_count_ext = CL_DEVICE_REFERENCE_COUNT_EXT 1032 reference_count_ext = CL_DEVICE_REFERENCE_COUNT_EXT
410 PPCODE: 1033 PPCODE:
411 cl_uint value [1]; 1034 cl_uint value [1];
412 NEED_SUCCESS (GetDeviceInfo, (self, ix, sizeof (value), value, 0)); 1035 NEED_SUCCESS (GetDeviceInfo, (self, ix, sizeof (value), value, 0));
413 EXTEND (SP, 1); 1036 EXTEND (SP, 1);
414 const int i = 0; 1037 const int i = 0;
528 PPCODE: 1151 PPCODE:
529 cl_command_queue_properties value [1]; 1152 cl_command_queue_properties value [1];
530 NEED_SUCCESS (GetDeviceInfo, (self, CL_DEVICE_QUEUE_PROPERTIES, sizeof (value), value, 0)); 1153 NEED_SUCCESS (GetDeviceInfo, (self, CL_DEVICE_QUEUE_PROPERTIES, sizeof (value), value, 0));
531 EXTEND (SP, 1); 1154 EXTEND (SP, 1);
532 const int i = 0; 1155 const int i = 0;
533 PUSHs (sv_2mortal (newSViv (value [i]))); 1156 PUSHs (sv_2mortal (newSVuv (value [i])));
534 1157
535void 1158void
536platform (OpenCL::Device self) 1159platform (OpenCL::Device self)
537 PPCODE: 1160 PPCODE:
538 cl_platform_id value [1]; 1161 cl_platform_id value [1];
539 NEED_SUCCESS (GetDeviceInfo, (self, CL_DEVICE_PLATFORM, sizeof (value), value, 0)); 1162 NEED_SUCCESS (GetDeviceInfo, (self, CL_DEVICE_PLATFORM, sizeof (value), value, 0));
540 EXTEND (SP, 1); 1163 EXTEND (SP, 1);
541 const int i = 0; 1164 const int i = 0;
542 { 1165 {
543 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Platform", value [i])); 1166 PUSH_CLOBJ (stash_platform, value [i]);
544 } 1167 }
545 1168
546void 1169void
547name (OpenCL::Device self) 1170name (OpenCL::Device self)
548 ALIAS: 1171 ALIAS:
567 cl_device_id value [1]; 1190 cl_device_id value [1];
568 NEED_SUCCESS (GetDeviceInfo, (self, CL_DEVICE_PARENT_DEVICE_EXT, sizeof (value), value, 0)); 1191 NEED_SUCCESS (GetDeviceInfo, (self, CL_DEVICE_PARENT_DEVICE_EXT, sizeof (value), value, 0));
569 EXTEND (SP, 1); 1192 EXTEND (SP, 1);
570 const int i = 0; 1193 const int i = 0;
571 { 1194 {
572 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Device", value [i])); 1195 PUSH_CLOBJ (stash_device, value [i]);
573 } 1196 }
574 1197
575void 1198void
576partition_types_ext (OpenCL::Device self) 1199partition_types_ext (OpenCL::Device self)
577 ALIAS: 1200 ALIAS:
588 for (i = 0; i < n; ++i) 1211 for (i = 0; i < n; ++i)
589 PUSHs (sv_2mortal (newSVuv (value [i]))); 1212 PUSHs (sv_2mortal (newSVuv (value [i])));
590 1213
591#END:device 1214#END:device
592 1215
1216MODULE = OpenCL PACKAGE = OpenCL::SubDevice
1217
1218#if CL_VERSION_1_2
1219
1220void
1221DESTROY (OpenCL::SubDevice self)
1222 CODE:
1223 clReleaseDevice (self);
1224
1225#endif
1226
593MODULE = OpenCL PACKAGE = OpenCL::Context 1227MODULE = OpenCL PACKAGE = OpenCL::Context
594 1228
595void 1229void
596DESTROY (OpenCL::Context context) 1230DESTROY (OpenCL::Context self)
597 CODE: 1231 CODE:
598 clReleaseContext (context); 1232 clReleaseContext (self);
599 1233
600void 1234void
601info (OpenCL::Context self, cl_context_info name) 1235info (OpenCL::Context self, cl_context_info name)
602 PPCODE: 1236 PPCODE:
603 INFO (Context) 1237 INFO (Context)
604 1238
605void 1239void
606queue (OpenCL::Context self, OpenCL::Device device, cl_command_queue_properties properties = 0) 1240queue (OpenCL::Context self, OpenCL::Device device, cl_command_queue_properties properties = 0)
607 PPCODE: 1241 PPCODE:
608 NEED_SUCCESS_ARG (cl_command_queue queue, CreateCommandQueue, (self, device, properties, &res)); 1242 NEED_SUCCESS_ARG (cl_command_queue queue, CreateCommandQueue, (self, device, properties, &res));
609 XPUSH_NEW_OBJ ("OpenCL::Queue", queue); 1243 XPUSH_CLOBJ (stash_queue, queue);
610 1244
611void 1245void
612user_event (OpenCL::Context self) 1246user_event (OpenCL::Context self)
613 PPCODE: 1247 PPCODE:
614 NEED_SUCCESS_ARG (cl_event ev, CreateUserEvent, (self, &res)); 1248 NEED_SUCCESS_ARG (cl_event ev, CreateUserEvent, (self, &res));
615 XPUSH_NEW_OBJ ("OpenCL::UserEvent", ev); 1249 XPUSH_CLOBJ (stash_userevent, ev);
616 1250
617void 1251void
618buffer (OpenCL::Context self, cl_mem_flags flags, size_t len) 1252buffer (OpenCL::Context self, cl_mem_flags flags, size_t len)
619 PPCODE: 1253 PPCODE:
620 if (flags & (CL_MEM_USE_HOST_PTR | CL_MEM_COPY_HOST_PTR)) 1254 if (flags & (CL_MEM_USE_HOST_PTR | CL_MEM_COPY_HOST_PTR))
621 croak ("OpenCL::Context::buffer: cannot use/copy host ptr when no data is given, use $context->buffer_sv instead?"); 1255 croak ("OpenCL::Context::buffer: cannot use/copy host ptr when no data is given, use $context->buffer_sv instead?");
622 1256
623 NEED_SUCCESS_ARG (cl_mem mem, CreateBuffer, (self, flags, len, 0, &res)); 1257 NEED_SUCCESS_ARG (cl_mem mem, CreateBuffer, (self, flags, len, 0, &res));
624 XPUSH_NEW_OBJ ("OpenCL::BufferObj", mem); 1258 XPUSH_CLOBJ (stash_bufferobj, mem);
625 1259
626void 1260void
627buffer_sv (OpenCL::Context self, cl_mem_flags flags, SV *data) 1261buffer_sv (OpenCL::Context self, cl_mem_flags flags, SV *data)
628 PPCODE: 1262 PPCODE:
629 STRLEN len; 1263 STRLEN len;
630 char *ptr = SvOK (data) ? SvPVbyte (data, len) : 0; 1264 char *ptr = SvOK (data) ? SvPVbyte (data, len) : 0;
631 if (!(flags & (CL_MEM_USE_HOST_PTR | CL_MEM_COPY_HOST_PTR))) 1265 if (!(flags & (CL_MEM_USE_HOST_PTR | CL_MEM_COPY_HOST_PTR)))
632 croak ("OpenCL::Context::buffer_sv: you have to specify use or copy host ptr when buffer data is given, use $context->buffer instead?"); 1266 croak ("OpenCL::Context::buffer_sv: you have to specify use or copy host ptr when buffer data is given, use $context->buffer instead?");
633 NEED_SUCCESS_ARG (cl_mem mem, CreateBuffer, (self, flags, len, ptr, &res)); 1267 NEED_SUCCESS_ARG (cl_mem mem, CreateBuffer, (self, flags, len, ptr, &res));
634 XPUSH_NEW_OBJ ("OpenCL::BufferObj", mem); 1268 XPUSH_CLOBJ (stash_bufferobj, mem);
635 1269
1270#if CL_VERSION_1_2
1271
636void 1272void
637image2d (OpenCL::Context self, cl_mem_flags flags, cl_channel_order channel_order, cl_channel_type channel_type, size_t width, size_t height, size_t row_pitch = 0, SV *data = &PL_sv_undef) 1273image (OpenCL::Context self, cl_mem_flags flags, cl_channel_order channel_order, cl_channel_type channel_type, cl_mem_object_type type, size_t width, size_t height, size_t depth = 0, size_t array_size = 0, size_t row_pitch = 0, size_t slice_pitch = 0, cl_uint num_mip_level = 0, cl_uint num_samples = 0, SV *data = &PL_sv_undef)
638 PPCODE: 1274 PPCODE:
639 STRLEN len; 1275 STRLEN len;
640 char *ptr = SvOK (data) ? SvPVbyte (data, len) : 0; 1276 char *ptr = SvOK (data) ? SvPVbyte (data, len) : 0;
641 const cl_image_format format = { channel_order, channel_type }; 1277 const cl_image_format format = { channel_order, channel_type };
1278 const cl_image_desc desc = {
1279 type,
1280 width, height, depth,
1281 array_size, row_pitch, slice_pitch,
1282 num_mip_level, num_samples,
1283 type == CL_MEM_OBJECT_IMAGE1D_BUFFER ? (cl_mem)SvCLOBJ (cv, "data", data, "OpenCL::Buffer") : 0
1284 };
642 NEED_SUCCESS_ARG (cl_mem mem, CreateImage2D, (self, flags, &format, width, height, row_pitch, ptr, &res)); 1285 NEED_SUCCESS_ARG (cl_mem mem, CreateImage, (self, flags, &format, &desc, ptr, &res));
643 XPUSH_NEW_OBJ ("OpenCL::Image2D", mem); 1286 HV *stash = stash_image;
1287 switch (type)
1288 {
1289 case CL_MEM_OBJECT_IMAGE1D_BUFFER: stash = stash_image1dbuffer; break;
1290 case CL_MEM_OBJECT_IMAGE1D: stash = stash_image1d; break;
1291 case CL_MEM_OBJECT_IMAGE1D_ARRAY: stash = stash_image2darray; break;
1292 case CL_MEM_OBJECT_IMAGE2D: stash = stash_image2d; break;
1293 case CL_MEM_OBJECT_IMAGE2D_ARRAY: stash = stash_image2darray; break;
1294 case CL_MEM_OBJECT_IMAGE3D: stash = stash_image3d; break;
1295 }
1296 XPUSH_CLOBJ (stash, mem);
644 1297
1298#endif
1299
645void 1300void
646image3d (OpenCL::Context self, cl_mem_flags flags, cl_channel_order channel_order, cl_channel_type channel_type, size_t width, size_t height, size_t depth, size_t row_pitch = 0, size_t slice_pitch = 0, SV *data = &PL_sv_undef) 1301image2d (OpenCL::Context self, cl_mem_flags flags, cl_channel_order channel_order, cl_channel_type channel_type, size_t width, size_t height, size_t row_pitch = 0, SV *data = &PL_sv_undef)
647 PPCODE: 1302 PPCODE:
648 STRLEN len; 1303 STRLEN len;
649 char *ptr = SvOK (data) ? SvPVbyte (data, len) : 0; 1304 char *ptr = SvOK (data) ? SvPVbyte (data, len) : 0;
650 const cl_image_format format = { channel_order, channel_type }; 1305 const cl_image_format format = { channel_order, channel_type };
1306#if PREFER_1_1
1307 NEED_SUCCESS_ARG (cl_mem mem, CreateImage2D, (self, flags, &format, width, height, row_pitch, ptr, &res));
1308#else
1309 const cl_image_desc desc = { CL_MEM_OBJECT_IMAGE2D, width, height, 0, 0, row_pitch, 0, 0, 0, 0 };
1310 NEED_SUCCESS_ARG (cl_mem mem, CreateImage, (self, flags, &format, &desc, ptr, &res));
1311#endif
1312 XPUSH_CLOBJ (stash_image2d, mem);
1313
1314void
1315image3d (OpenCL::Context self, cl_mem_flags flags, cl_channel_order channel_order, cl_channel_type channel_type, size_t width, size_t height, size_t depth, size_t row_pitch = 0, size_t slice_pitch = 0, SV *data = &PL_sv_undef)
1316 PPCODE:
1317 STRLEN len;
1318 char *ptr = SvOK (data) ? SvPVbyte (data, len) : 0;
1319 const cl_image_format format = { channel_order, channel_type };
1320#if PREFER_1_1
651 NEED_SUCCESS_ARG (cl_mem mem, CreateImage3D, (self, flags, &format, width, height, depth, row_pitch, slice_pitch, ptr, &res)); 1321 NEED_SUCCESS_ARG (cl_mem mem, CreateImage3D, (self, flags, &format, width, height, depth, row_pitch, slice_pitch, ptr, &res));
652 XPUSH_NEW_OBJ ("OpenCL::Image3D", mem); 1322#else
1323 const cl_image_desc desc = { CL_MEM_OBJECT_IMAGE3D, width, height, depth, 0, row_pitch, slice_pitch, 0, 0, 0 };
1324 NEED_SUCCESS_ARG (cl_mem mem, CreateImage, (self, flags, &format, &desc, ptr, &res));
1325#endif
1326 XPUSH_CLOBJ (stash_image3d, mem);
653 1327
654#if cl_apple_gl_sharing || cl_khr_gl_sharing 1328#if cl_apple_gl_sharing || cl_khr_gl_sharing
655 1329
656void 1330void
657gl_buffer (OpenCL::Context self, cl_mem_flags flags, cl_GLuint bufobj) 1331gl_buffer (OpenCL::Context self, cl_mem_flags flags, cl_GLuint bufobj)
658 PPCODE: 1332 PPCODE:
659 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLBuffer, (self, flags, bufobj, &res)); 1333 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLBuffer, (self, flags, bufobj, &res));
660 XPUSH_NEW_OBJ ("OpenCL::BufferObj", mem); 1334 XPUSH_CLOBJ (stash_bufferobj, mem);
1335
1336void
1337gl_renderbuffer (OpenCL::Context self, cl_mem_flags flags, cl_GLuint renderbuffer)
1338 PPCODE:
1339 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLRenderbuffer, (self, flags, renderbuffer, &res));
1340 XPUSH_CLOBJ (stash_image2d, mem);
1341
1342#if CL_VERSION_1_2
1343
1344void
1345gl_texture (OpenCL::Context self, cl_mem_flags flags, cl_GLenum target, cl_GLint miplevel, cl_GLuint texture)
1346 ALIAS:
1347 PPCODE:
1348 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLTexture, (self, flags, target, miplevel, texture, &res));
1349 cl_gl_object_type type;
1350 NEED_SUCCESS (GetGLObjectInfo, (mem, &type, 0)); // TODO: use target instead?
1351 HV *stash = stash_memory;
1352 switch (type)
1353 {
1354 case CL_GL_OBJECT_TEXTURE_BUFFER: stash = stash_image1dbuffer; break;
1355 case CL_GL_OBJECT_TEXTURE1D: stash = stash_image1d; break;
1356 case CL_GL_OBJECT_TEXTURE1D_ARRAY: stash = stash_image2darray; break;
1357 case CL_GL_OBJECT_TEXTURE2D: stash = stash_image2d; break;
1358 case CL_GL_OBJECT_TEXTURE2D_ARRAY: stash = stash_image2darray; break;
1359 case CL_GL_OBJECT_TEXTURE3D: stash = stash_image3d; break;
1360 }
1361 XPUSH_CLOBJ (stash, mem);
1362
1363#endif
661 1364
662void 1365void
663gl_texture2d (OpenCL::Context self, cl_mem_flags flags, cl_GLenum target, cl_GLint miplevel, cl_GLuint texture) 1366gl_texture2d (OpenCL::Context self, cl_mem_flags flags, cl_GLenum target, cl_GLint miplevel, cl_GLuint texture)
664 PPCODE: 1367 PPCODE:
1368#if PREFER_1_1
665 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLTexture2D, (self, flags, target, miplevel, texture, &res)); 1369 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLTexture2D, (self, flags, target, miplevel, texture, &res));
666 XPUSH_NEW_OBJ ("OpenCL::Image2D", mem); 1370#else
1371 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLTexture , (self, flags, target, miplevel, texture, &res));
1372#endif
1373 XPUSH_CLOBJ (stash_image2d, mem);
667 1374
668void 1375void
669gl_texture3d (OpenCL::Context self, cl_mem_flags flags, cl_GLenum target, cl_GLint miplevel, cl_GLuint texture) 1376gl_texture3d (OpenCL::Context self, cl_mem_flags flags, cl_GLenum target, cl_GLint miplevel, cl_GLuint texture)
670 PPCODE: 1377 PPCODE:
1378#if PREFER_1_1
671 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLTexture3D, (self, flags, target, miplevel, texture, &res)); 1379 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLTexture3D, (self, flags, target, miplevel, texture, &res));
672 XPUSH_NEW_OBJ ("OpenCL::Image3D", mem); 1380#else
673 1381 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLTexture , (self, flags, target, miplevel, texture, &res));
674void 1382#endif
675gl_renderbuffer (OpenCL::Context self, cl_mem_flags flags, cl_GLuint renderbuffer) 1383 XPUSH_CLOBJ (stash_image3d, mem);
676 PPCODE:
677 NEED_SUCCESS_ARG (cl_mem mem, CreateFromGLRenderbuffer, (self, flags, renderbuffer, &res));
678 XPUSH_NEW_OBJ ("OpenCL::Image2D", mem);
679 1384
680#endif 1385#endif
681 1386
682void 1387void
683supported_image_formats (OpenCL::Context self, cl_mem_flags flags, cl_mem_object_type image_type) 1388supported_image_formats (OpenCL::Context self, cl_mem_flags flags, cl_mem_object_type image_type)
703 1408
704void 1409void
705sampler (OpenCL::Context self, cl_bool normalized_coords, cl_addressing_mode addressing_mode, cl_filter_mode filter_mode) 1410sampler (OpenCL::Context self, cl_bool normalized_coords, cl_addressing_mode addressing_mode, cl_filter_mode filter_mode)
706 PPCODE: 1411 PPCODE:
707 NEED_SUCCESS_ARG (cl_sampler sampler, CreateSampler, (self, normalized_coords, addressing_mode, filter_mode, &res)); 1412 NEED_SUCCESS_ARG (cl_sampler sampler, CreateSampler, (self, normalized_coords, addressing_mode, filter_mode, &res));
708 XPUSH_NEW_OBJ ("OpenCL::Sampler", sampler); 1413 XPUSH_CLOBJ (stash_sampler, sampler);
709 1414
710void 1415void
711program_with_source (OpenCL::Context self, SV *program) 1416program_with_source (OpenCL::Context self, SV *program)
712 PPCODE: 1417 PPCODE:
713 STRLEN len; 1418 STRLEN len;
714 size_t len2; 1419 size_t len2;
715 const char *ptr = SvPVbyte (program, len); 1420 const char *ptr = SvPVbyte (program, len);
716 1421
717 len2 = len; 1422 len2 = len;
718 NEED_SUCCESS_ARG (cl_program prog, CreateProgramWithSource, (self, 1, &ptr, &len2, &res)); 1423 NEED_SUCCESS_ARG (cl_program prog, CreateProgramWithSource, (self, 1, &ptr, &len2, &res));
719 XPUSH_NEW_OBJ ("OpenCL::Program", prog); 1424 XPUSH_CLOBJ (stash_program, prog);
1425
1426void
1427program_with_binary (OpenCL::Context self, SV *devices, SV *binaries)
1428 PPCODE:
1429 cl_uint device_count;
1430 cl_device_id *device_list = object_list (cv, 0, "devices", devices, "OpenCL::Device", &device_count);
1431
1432 if (!SvROK (binaries) || SvTYPE (SvRV (binaries)) != SVt_PVAV)
1433 croak ("OpenCL::Context::program_with_binary: binaries must be specified as reference to an array of strings");
1434
1435 binaries = SvRV (binaries);
1436
1437 if (device_count != av_len ((AV *)binaries) + 1)
1438 croak ("OpenCL::Context::program_with_binary: differing numbers of devices and binaries are not allowed");
1439
1440 size_t *length_list = tmpbuf (sizeof (*length_list) * device_count);
1441 const unsigned char **binary_list = tmpbuf (sizeof (*binary_list) * device_count);
1442 cl_int *status_list = tmpbuf (sizeof (*status_list) * device_count);
1443
1444 int i;
1445 for (i = 0; i < device_count; ++i)
1446 {
1447 STRLEN len;
1448 binary_list [i] = (const unsigned char *)SvPVbyte (*av_fetch ((AV *)binaries, i, 0), len);
1449 length_list [i] = len;
1450 }
1451
1452 NEED_SUCCESS_ARG (cl_program prog, CreateProgramWithBinary, (self, device_count, device_list,
1453 length_list, binary_list,
1454 GIMME_V == G_ARRAY ? status_list : 0, &res));
1455
1456 EXTEND (SP, 2);
1457 PUSH_CLOBJ (stash_program, prog);
1458
1459 if (GIMME_V == G_ARRAY)
1460 {
1461 AV *av = newAV ();
1462 PUSHs (sv_2mortal (newRV_noinc ((SV *)av)));
1463
1464 for (i = device_count; i--; )
1465 av_store (av, i, newSViv (status_list [i]));
1466 }
1467
1468#if CL_VERSION_1_2
1469
1470void
1471program_with_built_in_kernels (OpenCL::Context self, SV *devices, SV *kernel_names)
1472 PPCODE:
1473 cl_uint device_count;
1474 cl_device_id *device_list = object_list (cv, 0, "devices", devices, "OpenCL::Device", &device_count);
1475
1476 NEED_SUCCESS_ARG (cl_program prog, CreateProgramWithBuiltInKernels, (self, device_count, device_list, SvPVbyte_nolen (kernel_names), &res));
1477
1478 XPUSH_CLOBJ (stash_program, prog);
1479
1480void
1481link_program (OpenCL::Context self, SV *devices, SV *options, SV *programs, SV *notify = &PL_sv_undef)
1482 CODE:
1483 cl_uint device_count = 0;
1484 cl_device_id *device_list = 0;
1485
1486 if (SvOK (devices))
1487 device_list = object_list (cv, 1, "devices", devices, "OpenCL::Device", &device_count);
1488
1489 cl_uint program_count;
1490 cl_program *program_list = object_list (cv, 0, "programs", programs, "OpenCL::Program", &program_count);
1491
1492 void *user_data;
1493 program_callback pfn_notify = make_program_callback (notify, &user_data);
1494
1495 NEED_SUCCESS_ARG (cl_program prog, LinkProgram, (self, device_count, device_list, SvPVbyte_nolen (options),
1496 program_count, program_list, pfn_notify, user_data, &res));
1497
1498 XPUSH_CLOBJ (stash_program, prog);
1499
1500#endif
720 1501
721#BEGIN:context 1502#BEGIN:context
722 1503
723void 1504void
724reference_count (OpenCL::Context self) 1505reference_count (OpenCL::Context self)
741 NEED_SUCCESS (GetContextInfo, (self, CL_CONTEXT_DEVICES, size, value, 0)); 1522 NEED_SUCCESS (GetContextInfo, (self, CL_CONTEXT_DEVICES, size, value, 0));
742 int i, n = size / sizeof (*value); 1523 int i, n = size / sizeof (*value);
743 EXTEND (SP, n); 1524 EXTEND (SP, n);
744 for (i = 0; i < n; ++i) 1525 for (i = 0; i < n; ++i)
745 { 1526 {
746 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Device", value [i])); 1527 PUSH_CLOBJ (stash_device, value [i]);
747 } 1528 }
748 1529
749void 1530void
750properties (OpenCL::Context self) 1531properties (OpenCL::Context self)
751 PPCODE: 1532 PPCODE:
766DESTROY (OpenCL::Queue self) 1547DESTROY (OpenCL::Queue self)
767 CODE: 1548 CODE:
768 clReleaseCommandQueue (self); 1549 clReleaseCommandQueue (self);
769 1550
770void 1551void
771enqueue_read_buffer (OpenCL::Queue self, OpenCL::Buffer mem, cl_bool blocking, size_t offset, size_t len, SV *data, ...) 1552read_buffer (OpenCL::Queue self, OpenCL::Buffer mem, cl_bool blocking, size_t offset, size_t len, SV *data, ...)
1553 ALIAS:
1554 enqueue_read_buffer = 0
772 PPCODE: 1555 PPCODE:
773 cl_event ev = 0;
774 EVENT_LIST (6, items - 6); 1556 EVENT_LIST (6);
775 1557
776 SvUPGRADE (data, SVt_PV); 1558 SvUPGRADE (data, SVt_PV);
777 SvGROW (data, len); 1559 SvGROW (data, len);
778 SvPOK_only (data); 1560 SvPOK_only (data);
779 SvCUR_set (data, len); 1561 SvCUR_set (data, len);
1562
1563 cl_event ev = 0;
780 NEED_SUCCESS (EnqueueReadBuffer, (self, mem, blocking, offset, len, SvPVX (data), event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1564 NEED_SUCCESS (EnqueueReadBuffer, (self, mem, blocking, offset, len, SvPVX (data), event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
781 1565
782 if (ev) 1566 if (ev)
783 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1567 XPUSH_CLOBJ (stash_event, ev);
784 1568
785void 1569void
786enqueue_write_buffer (OpenCL::Queue self, OpenCL::Buffer mem, cl_bool blocking, size_t offset, SV *data, ...) 1570write_buffer (OpenCL::Queue self, OpenCL::Buffer mem, cl_bool blocking, size_t offset, SV *data, ...)
1571 ALIAS:
1572 enqueue_write_buffer = 0
787 PPCODE: 1573 PPCODE:
788 cl_event ev = 0; 1574 EVENT_LIST (5);
1575
789 STRLEN len; 1576 STRLEN len;
790 char *ptr = SvPVbyte (data, len); 1577 char *ptr = SvPVbyte (data, len);
791 EVENT_LIST (5, items - 5);
792 1578
1579 cl_event ev = 0;
793 NEED_SUCCESS (EnqueueReadBuffer, (self, mem, blocking, offset, len, ptr, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1580 NEED_SUCCESS (EnqueueWriteBuffer, (self, mem, blocking, offset, len, ptr, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
794 1581
795 if (ev) 1582 if (ev)
796 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1583 XPUSH_CLOBJ (stash_event, ev);
797 1584
1585#if CL_VERSION_1_2
1586
798void 1587void
799enqueue_copy_buffer (OpenCL::Queue self, OpenCL::Buffer src, OpenCL::Buffer dst, size_t src_offset, size_t dst_offset, size_t len, ...) 1588fill_buffer (OpenCL::Queue self, OpenCL::Buffer mem, SV *data, size_t offset, size_t size, ...)
1589 ALIAS:
1590 enqueue_fill_buffer = 0
800 PPCODE: 1591 PPCODE:
1592 EVENT_LIST (5);
1593
1594 STRLEN len;
1595 char *ptr = SvPVbyte (data, len);
1596
801 cl_event ev = 0; 1597 cl_event ev = 0;
1598 NEED_SUCCESS (EnqueueFillBuffer, (self, mem, ptr, len, offset, size, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
1599
1600 if (ev)
1601 XPUSH_CLOBJ (stash_event, ev);
1602
1603void
1604fill_image (OpenCL::Queue self, OpenCL::Image img, NV r, NV g, NV b, NV a, size_t x, size_t y, size_t z, size_t width, size_t height, size_t depth, ...)
1605 ALIAS:
1606 enqueue_fill_image = 0
1607 PPCODE:
1608 EVENT_LIST (12);
1609
1610 const size_t origin [3] = { x, y, z };
1611 const size_t region [3] = { width, height, depth };
1612
1613 const cl_float c_f [4] = { r, g, b, a };
1614 const cl_uint c_u [4] = { r, g, b, a };
1615 const cl_int c_s [4] = { r, g, b, a };
1616 const void *c_fus [3] = { &c_f, &c_u, &c_s };
1617 static const char fus [] = { 0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 1, 1, 1, 0, 0 };
1618 cl_image_format format;
1619 NEED_SUCCESS (GetImageInfo, (img, CL_IMAGE_FORMAT, sizeof (format), &format, 0));
1620 assert (sizeof (fus) == CL_FLOAT + 1 - CL_SNORM_INT8);
1621 if (format.image_channel_data_type < CL_SNORM_INT8 || CL_FLOAT < format.image_channel_data_type)
1622 croak ("enqueue_fill_image: image has unsupported channel type, only opencl 1.2 channel types supported.");
1623
1624 cl_event ev = 0;
1625 NEED_SUCCESS (EnqueueFillImage, (self, img, c_fus [fus [format.image_channel_data_type - CL_SNORM_INT8]],
1626 origin, region, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
1627
1628 if (ev)
1629 XPUSH_CLOBJ (stash_event, ev);
1630
1631#endif
1632
1633void
1634copy_buffer (OpenCL::Queue self, OpenCL::Buffer src, OpenCL::Buffer dst, size_t src_offset, size_t dst_offset, size_t len, ...)
1635 ALIAS:
1636 enqueue_copy_buffer = 0
1637 PPCODE:
802 EVENT_LIST (6, items - 6); 1638 EVENT_LIST (6);
803 1639
1640 cl_event ev = 0;
804 NEED_SUCCESS (EnqueueCopyBuffer, (self, src, dst, src_offset, dst_offset, len, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1641 NEED_SUCCESS (EnqueueCopyBuffer, (self, src, dst, src_offset, dst_offset, len, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
805 1642
806 if (ev) 1643 if (ev)
807 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1644 XPUSH_CLOBJ (stash_event, ev);
808 1645
809void 1646void
810enqueue_read_buffer_rect (OpenCL::Queue self, OpenCL::Memory buf, cl_bool blocking, size_t buf_x, size_t buf_y, size_t buf_z, size_t host_x, size_t host_y, size_t host_z, size_t width, size_t height, size_t depth, size_t buf_row_pitch, size_t buf_slice_pitch, size_t host_row_pitch, size_t host_slice_pitch, SV *data, ...) 1647read_buffer_rect (OpenCL::Queue self, OpenCL::Memory buf, cl_bool blocking, size_t buf_x, size_t buf_y, size_t buf_z, size_t host_x, size_t host_y, size_t host_z, size_t width, size_t height, size_t depth, size_t buf_row_pitch, size_t buf_slice_pitch, size_t host_row_pitch, size_t host_slice_pitch, SV *data, ...)
1648 ALIAS:
1649 enqueue_read_buffer_rect = 0
811 PPCODE: 1650 PPCODE:
812 cl_event ev = 0; 1651 EVENT_LIST (17);
1652
813 const size_t buf_origin [3] = { buf_x , buf_y , buf_z }; 1653 const size_t buf_origin [3] = { buf_x , buf_y , buf_z };
814 const size_t host_origin[3] = { host_x, host_y, host_z }; 1654 const size_t host_origin[3] = { host_x, host_y, host_z };
815 const size_t region[3] = { width, height, depth }; 1655 const size_t region[3] = { width, height, depth };
816 EVENT_LIST (17, items - 17);
817 1656
818 if (!buf_row_pitch) 1657 if (!buf_row_pitch)
819 buf_row_pitch = region [0]; 1658 buf_row_pitch = region [0];
820 1659
821 if (!buf_slice_pitch) 1660 if (!buf_slice_pitch)
831 1670
832 SvUPGRADE (data, SVt_PV); 1671 SvUPGRADE (data, SVt_PV);
833 SvGROW (data, len); 1672 SvGROW (data, len);
834 SvPOK_only (data); 1673 SvPOK_only (data);
835 SvCUR_set (data, len); 1674 SvCUR_set (data, len);
1675
1676 cl_event ev = 0;
836 NEED_SUCCESS (EnqueueReadBufferRect, (self, buf, blocking, buf_origin, host_origin, region, buf_row_pitch, buf_slice_pitch, host_row_pitch, host_slice_pitch, SvPVX (data), event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1677 NEED_SUCCESS (EnqueueReadBufferRect, (self, buf, blocking, buf_origin, host_origin, region, buf_row_pitch, buf_slice_pitch, host_row_pitch, host_slice_pitch, SvPVX (data), event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
837 1678
838 if (ev) 1679 if (ev)
839 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1680 XPUSH_CLOBJ (stash_event, ev);
840 1681
841void 1682void
842enqueue_write_buffer_rect (OpenCL::Queue self, OpenCL::Memory buf, cl_bool blocking, size_t buf_x, size_t buf_y, size_t buf_z, size_t host_x, size_t host_y, size_t host_z, size_t width, size_t height, size_t depth, size_t buf_row_pitch, size_t buf_slice_pitch, size_t host_row_pitch, size_t host_slice_pitch, SV *data, ...) 1683write_buffer_rect (OpenCL::Queue self, OpenCL::Memory buf, cl_bool blocking, size_t buf_x, size_t buf_y, size_t buf_z, size_t host_x, size_t host_y, size_t host_z, size_t width, size_t height, size_t depth, size_t buf_row_pitch, size_t buf_slice_pitch, size_t host_row_pitch, size_t host_slice_pitch, SV *data, ...)
1684 ALIAS:
1685 enqueue_write_buffer_rect = 0
843 PPCODE: 1686 PPCODE:
844 cl_event ev = 0; 1687 EVENT_LIST (17);
1688
845 const size_t buf_origin [3] = { buf_x , buf_y , buf_z }; 1689 const size_t buf_origin [3] = { buf_x , buf_y , buf_z };
846 const size_t host_origin[3] = { host_x, host_y, host_z }; 1690 const size_t host_origin[3] = { host_x, host_y, host_z };
847 const size_t region[3] = { width, height, depth }; 1691 const size_t region[3] = { width, height, depth };
848 STRLEN len; 1692 STRLEN len;
849 char *ptr = SvPVbyte (data, len); 1693 char *ptr = SvPVbyte (data, len);
850 EVENT_LIST (17, items - 17);
851 1694
852 if (!buf_row_pitch) 1695 if (!buf_row_pitch)
853 buf_row_pitch = region [0]; 1696 buf_row_pitch = region [0];
854 1697
855 if (!buf_slice_pitch) 1698 if (!buf_slice_pitch)
864 size_t min_len = host_row_pitch * host_slice_pitch * region [2]; 1707 size_t min_len = host_row_pitch * host_slice_pitch * region [2];
865 1708
866 if (len < min_len) 1709 if (len < min_len)
867 croak ("clEnqueueWriteImage: data string is shorter than what would be transferred"); 1710 croak ("clEnqueueWriteImage: data string is shorter than what would be transferred");
868 1711
1712 cl_event ev = 0;
869 NEED_SUCCESS (EnqueueWriteBufferRect, (self, buf, blocking, buf_origin, host_origin, region, buf_row_pitch, buf_slice_pitch, host_row_pitch, host_slice_pitch, SvPVX (data), event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1713 NEED_SUCCESS (EnqueueWriteBufferRect, (self, buf, blocking, buf_origin, host_origin, region, buf_row_pitch, buf_slice_pitch, host_row_pitch, host_slice_pitch, ptr, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
870 1714
871 if (ev) 1715 if (ev)
872 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1716 XPUSH_CLOBJ (stash_event, ev);
873 1717
874void 1718void
875enqueue_copy_buffer_rect (OpenCL::Queue self, OpenCL::Buffer src, OpenCL::Buffer dst, size_t src_x, size_t src_y, size_t src_z, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, size_t src_row_pitch, size_t src_slice_pitch, size_t dst_row_pitch, size_t dst_slice_pitch, ...) 1719copy_buffer_rect (OpenCL::Queue self, OpenCL::Buffer src, OpenCL::Buffer dst, size_t src_x, size_t src_y, size_t src_z, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, size_t src_row_pitch, size_t src_slice_pitch, size_t dst_row_pitch, size_t dst_slice_pitch, ...)
1720 ALIAS:
1721 enqueue_copy_buffer_rect = 0
876 PPCODE: 1722 PPCODE:
877 cl_event ev = 0; 1723 EVENT_LIST (16);
1724
878 const size_t src_origin[3] = { src_x, src_y, src_z }; 1725 const size_t src_origin[3] = { src_x, src_y, src_z };
879 const size_t dst_origin[3] = { dst_x, dst_y, dst_z }; 1726 const size_t dst_origin[3] = { dst_x, dst_y, dst_z };
880 const size_t region[3] = { width, height, depth }; 1727 const size_t region[3] = { width, height, depth };
881 EVENT_LIST (16, items - 16);
882 1728
1729 cl_event ev = 0;
883 NEED_SUCCESS (EnqueueCopyBufferRect, (self, src, dst, src_origin, dst_origin, region, src_row_pitch, src_slice_pitch, dst_row_pitch, dst_slice_pitch, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1730 NEED_SUCCESS (EnqueueCopyBufferRect, (self, src, dst, src_origin, dst_origin, region, src_row_pitch, src_slice_pitch, dst_row_pitch, dst_slice_pitch, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
884 1731
885 if (ev) 1732 if (ev)
886 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1733 XPUSH_CLOBJ (stash_event, ev);
887 1734
888void 1735void
889enqueue_read_image (OpenCL::Queue self, OpenCL::Image src, cl_bool blocking, size_t src_x, size_t src_y, size_t src_z, size_t width, size_t height, size_t depth, size_t row_pitch, size_t slice_pitch, SV *data, ...) 1736read_image (OpenCL::Queue self, OpenCL::Image src, cl_bool blocking, size_t src_x, size_t src_y, size_t src_z, size_t width, size_t height, size_t depth, size_t row_pitch, size_t slice_pitch, SV *data, ...)
1737 ALIAS:
1738 enqueue_read_image = 0
890 PPCODE: 1739 PPCODE:
891 cl_event ev = 0; 1740 EVENT_LIST (12);
1741
892 const size_t src_origin[3] = { src_x, src_y, src_z }; 1742 const size_t src_origin[3] = { src_x, src_y, src_z };
893 const size_t region[3] = { width, height, depth }; 1743 const size_t region[3] = { width, height, depth };
894 EVENT_LIST (12, items - 12);
895 1744
896 if (!row_pitch) 1745 if (!row_pitch)
897 row_pitch = img_row_pitch (src); 1746 row_pitch = img_row_pitch (src);
898 1747
899 if (depth > 1 && !slice_pitch) 1748 if (depth > 1 && !slice_pitch)
903 1752
904 SvUPGRADE (data, SVt_PV); 1753 SvUPGRADE (data, SVt_PV);
905 SvGROW (data, len); 1754 SvGROW (data, len);
906 SvPOK_only (data); 1755 SvPOK_only (data);
907 SvCUR_set (data, len); 1756 SvCUR_set (data, len);
1757
1758 cl_event ev = 0;
908 NEED_SUCCESS (EnqueueReadImage, (self, src, blocking, src_origin, region, row_pitch, slice_pitch, SvPVX (data), event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1759 NEED_SUCCESS (EnqueueReadImage, (self, src, blocking, src_origin, region, row_pitch, slice_pitch, SvPVX (data), event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
909 1760
910 if (ev) 1761 if (ev)
911 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1762 XPUSH_CLOBJ (stash_event, ev);
912 1763
913void 1764void
914enqueue_write_image (OpenCL::Queue self, OpenCL::Image dst, cl_bool blocking, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, size_t row_pitch, size_t slice_pitch, SV *data, ...) 1765write_image (OpenCL::Queue self, OpenCL::Image dst, cl_bool blocking, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, size_t row_pitch, size_t slice_pitch, SV *data, ...)
1766 ALIAS:
1767 enqueue_write_image = 0
915 PPCODE: 1768 PPCODE:
916 cl_event ev = 0; 1769 EVENT_LIST (12);
1770
917 const size_t dst_origin[3] = { dst_x, dst_y, dst_z }; 1771 const size_t dst_origin[3] = { dst_x, dst_y, dst_z };
918 const size_t region[3] = { width, height, depth }; 1772 const size_t region[3] = { width, height, depth };
919 STRLEN len; 1773 STRLEN len;
920 char *ptr = SvPVbyte (data, len); 1774 char *ptr = SvPVbyte (data, len);
921 EVENT_LIST (12, items - 12);
922 1775
923 if (!row_pitch) 1776 if (!row_pitch)
924 row_pitch = img_row_pitch (dst); 1777 row_pitch = img_row_pitch (dst);
925 1778
926 if (depth > 1 && !slice_pitch) 1779 if (depth > 1 && !slice_pitch)
929 size_t min_len = slice_pitch ? slice_pitch * depth : row_pitch * height; 1782 size_t min_len = slice_pitch ? slice_pitch * depth : row_pitch * height;
930 1783
931 if (len < min_len) 1784 if (len < min_len)
932 croak ("clEnqueueWriteImage: data string is shorter than what would be transferred"); 1785 croak ("clEnqueueWriteImage: data string is shorter than what would be transferred");
933 1786
1787 cl_event ev = 0;
934 NEED_SUCCESS (EnqueueWriteImage, (self, dst, blocking, dst_origin, region, row_pitch, slice_pitch, SvPVX (data), event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1788 NEED_SUCCESS (EnqueueWriteImage, (self, dst, blocking, dst_origin, region, row_pitch, slice_pitch, ptr, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
935 1789
936 if (ev) 1790 if (ev)
937 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1791 XPUSH_CLOBJ (stash_event, ev);
938 1792
939void 1793void
940enqueue_copy_image (OpenCL::Queue self, OpenCL::Image src, OpenCL::Image dst, size_t src_x, size_t src_y, size_t src_z, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, ...) 1794copy_image (OpenCL::Queue self, OpenCL::Image src, OpenCL::Image dst, size_t src_x, size_t src_y, size_t src_z, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, ...)
1795 ALIAS:
1796 enqueue_copy_image = 0
941 PPCODE: 1797 PPCODE:
942 cl_event ev = 0; 1798 EVENT_LIST (12);
1799
943 const size_t src_origin[3] = { src_x, src_y, src_z }; 1800 const size_t src_origin[3] = { src_x, src_y, src_z };
944 const size_t dst_origin[3] = { dst_x, dst_y, dst_z }; 1801 const size_t dst_origin[3] = { dst_x, dst_y, dst_z };
945 const size_t region[3] = { width, height, depth }; 1802 const size_t region[3] = { width, height, depth };
946 EVENT_LIST (12, items - 12);
947 1803
1804 cl_event ev = 0;
948 NEED_SUCCESS (EnqueueCopyImage, (self, src, dst, src_origin, dst_origin, region, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1805 NEED_SUCCESS (EnqueueCopyImage, (self, src, dst, src_origin, dst_origin, region, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
949 1806
950 if (ev) 1807 if (ev)
951 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1808 XPUSH_CLOBJ (stash_event, ev);
952 1809
953void 1810void
954enqueue_copy_image_to_buffer (OpenCL::Queue self, OpenCL::Image src, OpenCL::Buffer dst, size_t src_x, size_t src_y, size_t src_z, size_t width, size_t height, size_t depth, size_t dst_offset, ...) 1811copy_image_to_buffer (OpenCL::Queue self, OpenCL::Image src, OpenCL::Buffer dst, size_t src_x, size_t src_y, size_t src_z, size_t width, size_t height, size_t depth, size_t dst_offset, ...)
1812 ALIAS:
1813 enqueue_copy_image_to_buffer = 0
955 PPCODE: 1814 PPCODE:
1815 EVENT_LIST (10);
1816
1817 const size_t src_origin[3] = { src_x, src_y, src_z };
1818 const size_t region [3] = { width, height, depth };
1819
956 cl_event ev = 0; 1820 cl_event ev = 0;
1821 NEED_SUCCESS (EnqueueCopyImageToBuffer, (self, src, dst, src_origin, region, dst_offset, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
1822
1823 if (ev)
1824 XPUSH_CLOBJ (stash_event, ev);
1825
1826void
1827copy_buffer_to_image (OpenCL::Queue self, OpenCL::Buffer src, OpenCL::Image dst, size_t src_offset, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, ...)
1828 ALIAS:
1829 enqueue_copy_buffer_to_image = 0
1830 PPCODE:
1831 EVENT_LIST (10);
1832
957 const size_t src_origin[3] = { src_x, src_y, src_z }; 1833 const size_t dst_origin[3] = { dst_x, dst_y, dst_z };
1834 const size_t region [3] = { width, height, depth };
1835
1836 cl_event ev = 0;
1837 NEED_SUCCESS (EnqueueCopyBufferToImage, (self, src, dst, src_offset, dst_origin, region, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
1838
1839 if (ev)
1840 XPUSH_CLOBJ (stash_event, ev);
1841
1842void
1843map_buffer (OpenCL::Queue self, OpenCL::Buffer buf, cl_bool blocking = 1, cl_map_flags map_flags = CL_MAP_READ | CL_MAP_WRITE, size_t offset = 0, SV *cb_ = &PL_sv_undef, ...)
1844 ALIAS:
1845 enqueue_map_buffer = 0
1846 PPCODE:
1847 EVENT_LIST (6);
1848
1849 size_t cb = SvIV (cb_);
1850
1851 if (!SvOK (cb_))
1852 {
1853 NEED_SUCCESS (GetMemObjectInfo, (buf, CL_MEM_SIZE, sizeof (cb), &cb, 0));
1854 cb -= offset;
1855 }
1856
1857 cl_event ev;
1858 NEED_SUCCESS_ARG (void *ptr, EnqueueMapBuffer, (self, buf, blocking, map_flags, offset, cb, event_list_count, event_list_ptr, &ev, &res));
1859 XPUSHs (mapped_new (stash_mappedbuffer, self, buf, map_flags, ptr, cb, ev, 0, 0));
1860
1861void
1862map_image (OpenCL::Queue self, OpenCL::Image img, cl_bool blocking = 1, cl_map_flags map_flags = CL_MAP_READ | CL_MAP_WRITE, size_t x = 0, size_t y = 0, size_t z = 0, SV *width_ = &PL_sv_undef, SV *height_ = &PL_sv_undef, SV *depth_ = &PL_sv_undef, ...)
1863 ALIAS:
1864 enqueue_map_image = 0
1865 PPCODE:
1866 size_t width = SvIV (width_);
1867 if (!SvOK (width_))
1868 {
1869 NEED_SUCCESS (GetImageInfo, (img, CL_IMAGE_WIDTH, sizeof (width), &width, 0));
1870 width -= x;
1871 }
1872
1873 size_t height = SvIV (width_);
1874 if (!SvOK (height_))
1875 {
1876 NEED_SUCCESS (GetImageInfo, (img, CL_IMAGE_HEIGHT, sizeof (height), &height, 0));
1877 height -= y;
1878 }
1879
1880 size_t depth = SvIV (width_);
1881 if (!SvOK (depth_))
1882 {
1883 NEED_SUCCESS (GetImageInfo, (img, CL_IMAGE_DEPTH, sizeof (depth), &depth, 0));
1884 depth -= z;
1885
1886 // stupid opencl returns 0 for depth, but requires 1 for 2d images
1887 if (!depth)
1888 depth = 1;
1889 }
1890
1891 const size_t origin[3] = { x, y, z };
958 const size_t region[3] = { width, height, depth }; 1892 const size_t region[3] = { width, height, depth };
1893 size_t row_pitch, slice_pitch;
959 EVENT_LIST (10, items - 10); 1894 EVENT_LIST (10);
960 1895
1896 cl_event ev;
1897 NEED_SUCCESS_ARG (void *ptr, EnqueueMapImage, (self, img, blocking, map_flags, origin, region, &row_pitch, &slice_pitch, event_list_count, event_list_ptr, &ev, &res));
1898
1899 size_t cb = slice_pitch ? slice_pitch * region [2]
1900 : row_pitch ? row_pitch * region [1]
1901 : region [0];
1902
1903 XPUSHs (mapped_new (stash_mappedimage, self, img, map_flags, ptr, cb, ev, row_pitch, slice_pitch));
1904
1905void
1906unmap (OpenCL::Queue self, OpenCL::Mapped mapped, ...)
1907 PPCODE:
1908 mapped_unmap (cv, ST (1), mapped, self, &ST (2), items - 2);
1909 if (GIMME_V != G_VOID)
1910 {
1911 clRetainEvent (mapped->event);
1912 XPUSH_CLOBJ (stash_event, mapped->event);
1913 }
1914
1915void
1916task (OpenCL::Queue self, OpenCL::Kernel kernel, ...)
1917 ALIAS:
1918 enqueue_task = 0
1919 PPCODE:
1920 EVENT_LIST (2);
1921
1922 cl_event ev = 0;
961 NEED_SUCCESS (EnqueueCopyImageToBuffer, (self, src, dst, src_origin, region, dst_offset, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1923 NEED_SUCCESS (EnqueueTask, (self, kernel, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
962 1924
963 if (ev) 1925 if (ev)
964 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1926 XPUSH_CLOBJ (stash_event, ev);
965 1927
966void 1928void
967enqueue_copy_buffer_to_image (OpenCL::Queue self, OpenCL::Buffer src, OpenCL::Image dst, size_t src_offset, size_t dst_x, size_t dst_y, size_t dst_z, size_t width, size_t height, size_t depth, ...)
968 PPCODE:
969 cl_event ev = 0;
970 const size_t dst_origin[3] = { dst_x, dst_y, dst_z };
971 const size_t region[3] = { width, height, depth };
972 EVENT_LIST (10, items - 10);
973
974 NEED_SUCCESS (EnqueueCopyBufferToImage, (self, src, dst, src_offset, dst_origin, region, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
975
976 if (ev)
977 XPUSH_NEW_OBJ ("OpenCL::Event", ev);
978
979void
980enqueue_task (OpenCL::Queue self, OpenCL::Kernel kernel, ...)
981 PPCODE:
982 cl_event ev = 0;
983 EVENT_LIST (2, items - 2);
984
985 NEED_SUCCESS (EnqueueTask, (self, kernel, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
986
987 if (ev)
988 XPUSH_NEW_OBJ ("OpenCL::Event", ev);
989
990void
991enqueue_nd_range_kernel (OpenCL::Queue self, OpenCL::Kernel kernel, SV *global_work_offset, SV *global_work_size, SV *local_work_size = &PL_sv_undef, ...) 1929nd_range_kernel (OpenCL::Queue self, OpenCL::Kernel kernel, SV *global_work_offset, SV *global_work_size, SV *local_work_size = &PL_sv_undef, ...)
1930 ALIAS:
1931 enqueue_nd_range_kernel = 0
992 PPCODE: 1932 PPCODE:
993 cl_event ev = 0; 1933 EVENT_LIST (5);
1934
994 size_t *gwo = 0, *gws, *lws = 0; 1935 size_t *gwo = 0, *gws, *lws = 0;
995 int gws_len; 1936 int gws_len;
996 size_t *lists; 1937 size_t *lists;
997 int i; 1938 int i;
998 EVENT_LIST (5, items - 5);
999 1939
1000 if (!SvROK (global_work_size) || SvTYPE (SvRV (global_work_size)) != SVt_PVAV) 1940 if (!SvROK (global_work_size) || SvTYPE (SvRV (global_work_size)) != SVt_PVAV)
1001 croak ("clEnqueueNDRangeKernel: global_work_size must be an array reference"); 1941 croak ("clEnqueueNDRangeKernel: global_work_size must be an array reference");
1002 1942
1003 gws_len = AvFILLp (SvRV (global_work_size)) + 1; 1943 gws_len = AvFILLp (SvRV (global_work_size)) + 1;
1004 1944
1005 lists = tmpbuf (sizeof (size_t) * 3 * gws_len); 1945 lists = tmpbuf (sizeof (size_t) * 3 * gws_len);
1006 1946
1007 gws = lists + gws_len * 0; 1947 gws = lists + gws_len * 0;
1008 for (i = 0; i < gws_len; ++i) 1948 for (i = 0; i < gws_len; ++i)
1949 {
1009 gws [i] = SvIV (AvARRAY (SvRV (global_work_size))[i]); 1950 gws [i] = SvIV (AvARRAY (SvRV (global_work_size))[i]);
1951 // at least nvidia crashes for 0-sized work group sizes, work around
1952 if (!gws [i])
1953 croak ("clEnqueueNDRangeKernel: global_work_size[%d] is zero, must be non-zero", i);
1954 }
1010 1955
1011 if (SvOK (global_work_offset)) 1956 if (SvOK (global_work_offset))
1012 { 1957 {
1013 if (!SvROK (global_work_offset) || SvTYPE (SvRV (global_work_offset)) != SVt_PVAV) 1958 if (!SvROK (global_work_offset) || SvTYPE (SvRV (global_work_offset)) != SVt_PVAV)
1014 croak ("clEnqueueNDRangeKernel: global_work_offset must be undef or an array reference"); 1959 croak ("clEnqueueNDRangeKernel: global_work_offset must be undef or an array reference");
1021 gwo [i] = SvIV (AvARRAY (SvRV (global_work_offset))[i]); 1966 gwo [i] = SvIV (AvARRAY (SvRV (global_work_offset))[i]);
1022 } 1967 }
1023 1968
1024 if (SvOK (local_work_size)) 1969 if (SvOK (local_work_size))
1025 { 1970 {
1026 if (SvOK (local_work_size) && !SvROK (local_work_size) || SvTYPE (SvRV (local_work_size)) != SVt_PVAV) 1971 if ((SvOK (local_work_size) && !SvROK (local_work_size)) || SvTYPE (SvRV (local_work_size)) != SVt_PVAV)
1027 croak ("clEnqueueNDRangeKernel: global_work_size must be undef or an array reference"); 1972 croak ("clEnqueueNDRangeKernel: local_work_size must be undef or an array reference");
1028 1973
1029 if (AvFILLp (SvRV (local_work_size)) + 1 != gws_len) 1974 if (AvFILLp (SvRV (local_work_size)) + 1 != gws_len)
1030 croak ("clEnqueueNDRangeKernel: local_work_local must be undef or an array of same size as global_work_size"); 1975 croak ("clEnqueueNDRangeKernel: local_work_local must be undef or an array of same size as global_work_size");
1031 1976
1032 lws = lists + gws_len * 2; 1977 lws = lists + gws_len * 2;
1033 for (i = 0; i < gws_len; ++i) 1978 for (i = 0; i < gws_len; ++i)
1979 {
1034 lws [i] = SvIV (AvARRAY (SvRV (local_work_size))[i]); 1980 lws [i] = SvIV (AvARRAY (SvRV (local_work_size))[i]);
1981 // at least nvidia crashes for 0-sized work group sizes, work around
1982 if (!lws [i])
1983 croak ("clEnqueueNDRangeKernel: local_work_size[%d] is zero, must be non-zero", i);
1984 }
1035 } 1985 }
1036 1986
1987 cl_event ev = 0;
1037 NEED_SUCCESS (EnqueueNDRangeKernel, (self, kernel, gws_len, gwo, gws, lws, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 1988 NEED_SUCCESS (EnqueueNDRangeKernel, (self, kernel, gws_len, gwo, gws, lws, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
1038 1989
1039 if (ev) 1990 if (ev)
1040 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 1991 XPUSH_CLOBJ (stash_event, ev);
1992
1993#if CL_VERSION_1_2
1994
1995void
1996migrate_mem_objects (OpenCL::Queue self, SV *objects, cl_mem_migration_flags flags, ...)
1997 ALIAS:
1998 enqueue_migrate_mem_objects = 0
1999 PPCODE:
2000 EVENT_LIST (3);
2001
2002 cl_uint obj_count;
2003 cl_mem *obj_list = object_list (cv, 0, "objects", objects, "OpenCL::Memory", &obj_count);
2004
2005 cl_event ev = 0;
2006 NEED_SUCCESS (EnqueueMigrateMemObjects, (self, obj_count, obj_list, flags, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
2007
2008 if (ev)
2009 XPUSH_CLOBJ (stash_event, ev);
2010
2011#endif
1041 2012
1042#if cl_apple_gl_sharing || cl_khr_gl_sharing 2013#if cl_apple_gl_sharing || cl_khr_gl_sharing
1043 2014
1044void 2015void
1045enqueue_acquire_gl_objects (OpenCL::Queue self, SV *objects, ...) 2016acquire_gl_objects (OpenCL::Queue self, SV *objects, ...)
1046 ALIAS: 2017 ALIAS:
2018 release_gl_objects = 1
2019 enqueue_acquire_gl_objects = 0
1047 enqueue_release_gl_objects = 1 2020 enqueue_release_gl_objects = 1
1048 CODE: 2021 PPCODE:
1049 if (!SvROK (objects) || SvTYPE (SvRV (objects)) != SVt_PVAV) 2022 EVENT_LIST (2);
1050 croak ("OpenCL::Queue::enqueue_acquire/release_gl_objects argument 'objects' must be an arrayref with memory objects, in call"); 2023
2024 cl_uint obj_count;
2025 cl_mem *obj_list = object_list (cv, 0, "objects", objects, "OpenCL::Memory", &obj_count);
1051 2026
1052 cl_event ev = 0; 2027 cl_event ev = 0;
1053 EVENT_LIST (2, items - 2);
1054 AV *av = (AV *)SvRV (objects);
1055 cl_uint num_objects = av_len (av) + 1;
1056 cl_mem *object_list = tmpbuf (sizeof (cl_mem) * num_objects);
1057 int i;
1058
1059 for (i = num_objects; i--; )
1060 object_list [i] = SvPTROBJ ("OpenCL::Queue::enqueue_acquire/release_gl_objects", "objects", *av_fetch (av, i, 0), "OpenCL::Memory");
1061 2028
1062 if (ix) 2029 if (ix)
1063 NEED_SUCCESS (EnqueueReleaseGLObjects, (self, num_objects, object_list, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 2030 NEED_SUCCESS (EnqueueReleaseGLObjects, (self, obj_count, obj_list, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
1064 else 2031 else
1065 NEED_SUCCESS (EnqueueAcquireGLObjects, (self, num_objects, object_list, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0)); 2032 NEED_SUCCESS (EnqueueAcquireGLObjects, (self, obj_count, obj_list, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
1066 2033
1067 if (ev) 2034 if (ev)
1068 XPUSH_NEW_OBJ ("OpenCL::Event", ev); 2035 XPUSH_CLOBJ (stash_event, ev);
1069 2036
1070#endif 2037#endif
1071 2038
1072void 2039void
1073enqueue_marker (OpenCL::Queue self)
1074 PPCODE:
1075 cl_event ev;
1076 NEED_SUCCESS (EnqueueMarker, (self, &ev));
1077 XPUSH_NEW_OBJ ("OpenCL::Event", ev);
1078
1079void
1080enqueue_wait_for_events (OpenCL::Queue self, ...) 2040wait_for_events (OpenCL::Queue self, ...)
2041 ALIAS:
2042 enqueue_wait_for_events = 0
1081 CODE: 2043 CODE:
1082 EVENT_LIST (1, items - 1); 2044 EVENT_LIST (1);
2045#if PREFER_1_1
1083 NEED_SUCCESS (EnqueueWaitForEvents, (self, event_list_count, event_list_ptr)); 2046 NEED_SUCCESS (EnqueueWaitForEvents, (self, event_list_count, event_list_ptr));
2047#else
2048 NEED_SUCCESS (EnqueueBarrierWithWaitList, (self, event_list_count, event_list_ptr, 0));
2049#endif
1084 2050
1085void 2051void
1086enqueue_barrier (OpenCL::Queue self) 2052marker (OpenCL::Queue self, ...)
2053 ALIAS:
2054 enqueue_marker = 0
1087 CODE: 2055 PPCODE:
2056 EVENT_LIST (1);
2057 cl_event ev = 0;
2058#if PREFER_1_1
2059 if (!event_list_count)
2060 NEED_SUCCESS (EnqueueMarker, (self, GIMME_V != G_VOID ? &ev : 0));
2061 else
2062#if CL_VERSION_1_2
2063 NEED_SUCCESS (EnqueueMarkerWithWaitList, (self, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
2064#else
2065 {
2066 NEED_SUCCESS (EnqueueWaitForEvents, (self, event_list_count, event_list_ptr)); // also a barrier
2067 NEED_SUCCESS (EnqueueMarker, (self, GIMME_V != G_VOID ? &ev : 0));
2068 }
2069#endif
2070#else
2071 NEED_SUCCESS (EnqueueMarkerWithWaitList, (self, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
2072#endif
2073 if (ev)
2074 XPUSH_CLOBJ (stash_event, ev);
2075
2076void
2077barrier (OpenCL::Queue self, ...)
2078 ALIAS:
2079 enqueue_barrier = 0
2080 PPCODE:
2081 EVENT_LIST (1);
2082 cl_event ev = 0;
2083#if PREFER_1_1
2084 if (!event_list_count && GIMME_V == G_VOID)
1088 NEED_SUCCESS (EnqueueBarrier, (self)); 2085 NEED_SUCCESS (EnqueueBarrier, (self));
2086 else
2087#if CL_VERSION_1_2
2088 NEED_SUCCESS (EnqueueBarrierWithWaitList, (self, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
2089#else
2090 {
2091 if (event_list_count)
2092 NEED_SUCCESS (EnqueueWaitForEvents, (self, event_list_count, event_list_ptr));
2093
2094 if (GIMME_V != G_VOID)
2095 NEED_SUCCESS (EnqueueMarker, (self, &ev));
2096 }
2097#endif
2098#else
2099 NEED_SUCCESS (EnqueueBarrierWithWaitList, (self, event_list_count, event_list_ptr, GIMME_V != G_VOID ? &ev : 0));
2100#endif
2101 if (ev)
2102 XPUSH_CLOBJ (stash_event, ev);
1089 2103
1090void 2104void
1091flush (OpenCL::Queue self) 2105flush (OpenCL::Queue self)
1092 CODE: 2106 CODE:
1093 NEED_SUCCESS (Flush, (self)); 2107 NEED_SUCCESS (Flush, (self));
1111 NEED_SUCCESS (GetCommandQueueInfo, (self, CL_QUEUE_CONTEXT, sizeof (value), value, 0)); 2125 NEED_SUCCESS (GetCommandQueueInfo, (self, CL_QUEUE_CONTEXT, sizeof (value), value, 0));
1112 EXTEND (SP, 1); 2126 EXTEND (SP, 1);
1113 const int i = 0; 2127 const int i = 0;
1114 { 2128 {
1115 NEED_SUCCESS (RetainContext, (value [i])); 2129 NEED_SUCCESS (RetainContext, (value [i]));
1116 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Context", value [i])); 2130 PUSH_CLOBJ (stash_context, value [i]);
1117 } 2131 }
1118 2132
1119void 2133void
1120device (OpenCL::Queue self) 2134device (OpenCL::Queue self)
1121 PPCODE: 2135 PPCODE:
1122 cl_device_id value [1]; 2136 cl_device_id value [1];
1123 NEED_SUCCESS (GetCommandQueueInfo, (self, CL_QUEUE_DEVICE, sizeof (value), value, 0)); 2137 NEED_SUCCESS (GetCommandQueueInfo, (self, CL_QUEUE_DEVICE, sizeof (value), value, 0));
1124 EXTEND (SP, 1); 2138 EXTEND (SP, 1);
1125 const int i = 0; 2139 const int i = 0;
1126 { 2140 {
1127 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Device", value [i])); 2141 PUSH_CLOBJ (stash_device, value [i]);
1128 } 2142 }
1129 2143
1130void 2144void
1131reference_count (OpenCL::Queue self) 2145reference_count (OpenCL::Queue self)
1132 PPCODE: 2146 PPCODE:
1141 PPCODE: 2155 PPCODE:
1142 cl_command_queue_properties value [1]; 2156 cl_command_queue_properties value [1];
1143 NEED_SUCCESS (GetCommandQueueInfo, (self, CL_QUEUE_PROPERTIES, sizeof (value), value, 0)); 2157 NEED_SUCCESS (GetCommandQueueInfo, (self, CL_QUEUE_PROPERTIES, sizeof (value), value, 0));
1144 EXTEND (SP, 1); 2158 EXTEND (SP, 1);
1145 const int i = 0; 2159 const int i = 0;
1146 PUSHs (sv_2mortal (newSViv (value [i]))); 2160 PUSHs (sv_2mortal (newSVuv (value [i])));
1147 2161
1148#END:command_queue 2162#END:command_queue
1149 2163
1150MODULE = OpenCL PACKAGE = OpenCL::Memory 2164MODULE = OpenCL PACKAGE = OpenCL::Memory
1151 2165
1166 PPCODE: 2180 PPCODE:
1167 cl_mem_object_type value [1]; 2181 cl_mem_object_type value [1];
1168 NEED_SUCCESS (GetMemObjectInfo, (self, CL_MEM_TYPE, sizeof (value), value, 0)); 2182 NEED_SUCCESS (GetMemObjectInfo, (self, CL_MEM_TYPE, sizeof (value), value, 0));
1169 EXTEND (SP, 1); 2183 EXTEND (SP, 1);
1170 const int i = 0; 2184 const int i = 0;
1171 PUSHs (sv_2mortal (newSViv (value [i]))); 2185 PUSHs (sv_2mortal (newSVuv (value [i])));
1172 2186
1173void 2187void
1174flags (OpenCL::Memory self) 2188flags (OpenCL::Memory self)
1175 PPCODE: 2189 PPCODE:
1176 cl_mem_flags value [1]; 2190 cl_mem_flags value [1];
1177 NEED_SUCCESS (GetMemObjectInfo, (self, CL_MEM_FLAGS, sizeof (value), value, 0)); 2191 NEED_SUCCESS (GetMemObjectInfo, (self, CL_MEM_FLAGS, sizeof (value), value, 0));
1178 EXTEND (SP, 1); 2192 EXTEND (SP, 1);
1179 const int i = 0; 2193 const int i = 0;
1180 PUSHs (sv_2mortal (newSViv (value [i]))); 2194 PUSHs (sv_2mortal (newSVuv (value [i])));
1181 2195
1182void 2196void
1183size (OpenCL::Memory self) 2197size (OpenCL::Memory self)
1184 ALIAS: 2198 ALIAS:
1185 size = CL_MEM_SIZE 2199 size = CL_MEM_SIZE
1219 NEED_SUCCESS (GetMemObjectInfo, (self, CL_MEM_CONTEXT, sizeof (value), value, 0)); 2233 NEED_SUCCESS (GetMemObjectInfo, (self, CL_MEM_CONTEXT, sizeof (value), value, 0));
1220 EXTEND (SP, 1); 2234 EXTEND (SP, 1);
1221 const int i = 0; 2235 const int i = 0;
1222 { 2236 {
1223 NEED_SUCCESS (RetainContext, (value [i])); 2237 NEED_SUCCESS (RetainContext, (value [i]));
1224 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Context", value [i])); 2238 PUSH_CLOBJ (stash_context, value [i]);
1225 } 2239 }
1226 2240
1227void 2241void
1228associated_memobject (OpenCL::Memory self) 2242associated_memobject (OpenCL::Memory self)
1229 PPCODE: 2243 PPCODE:
1231 NEED_SUCCESS (GetMemObjectInfo, (self, CL_MEM_ASSOCIATED_MEMOBJECT, sizeof (value), value, 0)); 2245 NEED_SUCCESS (GetMemObjectInfo, (self, CL_MEM_ASSOCIATED_MEMOBJECT, sizeof (value), value, 0));
1232 EXTEND (SP, 1); 2246 EXTEND (SP, 1);
1233 const int i = 0; 2247 const int i = 0;
1234 { 2248 {
1235 NEED_SUCCESS (RetainMemObject, (value [i])); 2249 NEED_SUCCESS (RetainMemObject, (value [i]));
1236 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Memory", value [i])); 2250 PUSH_CLOBJ (stash_memory, value [i]);
1237 } 2251 }
1238 2252
1239#END:mem 2253#END:mem
1240 2254
1241#if cl_apple_gl_sharing || cl_khr_gl_sharing 2255#if cl_apple_gl_sharing || cl_khr_gl_sharing
1243void 2257void
1244gl_object_info (OpenCL::Memory self) 2258gl_object_info (OpenCL::Memory self)
1245 PPCODE: 2259 PPCODE:
1246 cl_gl_object_type type; 2260 cl_gl_object_type type;
1247 cl_GLuint name; 2261 cl_GLuint name;
1248 NEED_SUCCESS (clGetGLObjectInfo, (self, &type, &name)); 2262 NEED_SUCCESS (GetGLObjectInfo, (self, &type, &name));
1249 EXTEND (SP, 2); 2263 EXTEND (SP, 2);
1250 PUSHs (sv_2mortal (newSVuv (type))); 2264 PUSHs (sv_2mortal (newSVuv (type)));
1251 PUSHs (sv_2mortal (newSVuv (name))); 2265 PUSHs (sv_2mortal (newSVuv (name)));
1252 2266
1253#endif 2267#endif
1261 croak ("clCreateSubBuffer: cannot use/copy/alloc host ptr, doesn't make sense, check your flags!"); 2275 croak ("clCreateSubBuffer: cannot use/copy/alloc host ptr, doesn't make sense, check your flags!");
1262 2276
1263 cl_buffer_region crdata = { origin, size }; 2277 cl_buffer_region crdata = { origin, size };
1264 2278
1265 NEED_SUCCESS_ARG (cl_mem mem, CreateSubBuffer, (self, flags, CL_BUFFER_CREATE_TYPE_REGION, &crdata, &res)); 2279 NEED_SUCCESS_ARG (cl_mem mem, CreateSubBuffer, (self, flags, CL_BUFFER_CREATE_TYPE_REGION, &crdata, &res));
1266 XPUSH_NEW_OBJ ("OpenCL::Buffer", mem); 2280 XPUSH_CLOBJ (stash_buffer, mem);
1267 2281
1268MODULE = OpenCL PACKAGE = OpenCL::Image 2282MODULE = OpenCL PACKAGE = OpenCL::Image
1269 2283
1270void 2284void
1271image_info (OpenCL::Image self, cl_image_info name) 2285image_info (OpenCL::Image self, cl_image_info name)
1272 PPCODE: 2286 PPCODE:
1273 INFO (Image) 2287 INFO (Image)
2288
2289void
2290format (OpenCL::Image self)
2291 PPCODE:
2292 cl_image_format format;
2293 NEED_SUCCESS (GetImageInfo, (self, CL_IMAGE_FORMAT, sizeof (format), &format, 0));
2294 EXTEND (SP, 2);
2295 PUSHs (sv_2mortal (newSVuv (format.image_channel_order)));
2296 PUSHs (sv_2mortal (newSVuv (format.image_channel_data_type)));
1274 2297
1275#BEGIN:image 2298#BEGIN:image
1276 2299
1277void 2300void
1278element_size (OpenCL::Image self) 2301element_size (OpenCL::Image self)
1298 2321
1299void 2322void
1300target (OpenCL::Image self) 2323target (OpenCL::Image self)
1301 PPCODE: 2324 PPCODE:
1302 cl_GLenum value [1]; 2325 cl_GLenum value [1];
1303 NEED_SUCCESS (GetGlTextureInfo, (self, CL_GL_TEXTURE_TARGET, sizeof (value), value, 0)); 2326 NEED_SUCCESS (GetGLTextureInfo, (self, CL_GL_TEXTURE_TARGET, sizeof (value), value, 0));
1304 EXTEND (SP, 1); 2327 EXTEND (SP, 1);
1305 const int i = 0; 2328 const int i = 0;
1306 PUSHs (sv_2mortal (newSVuv (value [i]))); 2329 PUSHs (sv_2mortal (newSVuv (value [i])));
1307 2330
1308void 2331void
1309gl_mipmap_level (OpenCL::Image self) 2332gl_mipmap_level (OpenCL::Image self)
1310 PPCODE: 2333 PPCODE:
1311 cl_GLint value [1]; 2334 cl_GLint value [1];
1312 NEED_SUCCESS (GetGlTextureInfo, (self, CL_GL_MIPMAP_LEVEL, sizeof (value), value, 0)); 2335 NEED_SUCCESS (GetGLTextureInfo, (self, CL_GL_MIPMAP_LEVEL, sizeof (value), value, 0));
1313 EXTEND (SP, 1); 2336 EXTEND (SP, 1);
1314 const int i = 0; 2337 const int i = 0;
1315 PUSHs (sv_2mortal (newSViv (value [i]))); 2338 PUSHs (sv_2mortal (newSViv (value [i])));
1316 2339
1317#END:gl_texture 2340#END:gl_texture
1348 NEED_SUCCESS (GetSamplerInfo, (self, CL_SAMPLER_CONTEXT, sizeof (value), value, 0)); 2371 NEED_SUCCESS (GetSamplerInfo, (self, CL_SAMPLER_CONTEXT, sizeof (value), value, 0));
1349 EXTEND (SP, 1); 2372 EXTEND (SP, 1);
1350 const int i = 0; 2373 const int i = 0;
1351 { 2374 {
1352 NEED_SUCCESS (RetainContext, (value [i])); 2375 NEED_SUCCESS (RetainContext, (value [i]));
1353 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Context", value [i])); 2376 PUSH_CLOBJ (stash_context, value [i]);
1354 } 2377 }
1355 2378
1356void 2379void
1357normalized_coords (OpenCL::Sampler self) 2380normalized_coords (OpenCL::Sampler self)
1358 PPCODE: 2381 PPCODE:
1359 cl_addressing_mode value [1]; 2382 cl_addressing_mode value [1];
1360 NEED_SUCCESS (GetSamplerInfo, (self, CL_SAMPLER_NORMALIZED_COORDS, sizeof (value), value, 0)); 2383 NEED_SUCCESS (GetSamplerInfo, (self, CL_SAMPLER_NORMALIZED_COORDS, sizeof (value), value, 0));
1361 EXTEND (SP, 1); 2384 EXTEND (SP, 1);
1362 const int i = 0; 2385 const int i = 0;
1363 PUSHs (sv_2mortal (newSViv (value [i]))); 2386 PUSHs (sv_2mortal (newSVuv (value [i])));
1364 2387
1365void 2388void
1366addressing_mode (OpenCL::Sampler self) 2389addressing_mode (OpenCL::Sampler self)
1367 PPCODE: 2390 PPCODE:
1368 cl_filter_mode value [1]; 2391 cl_filter_mode value [1];
1369 NEED_SUCCESS (GetSamplerInfo, (self, CL_SAMPLER_ADDRESSING_MODE, sizeof (value), value, 0)); 2392 NEED_SUCCESS (GetSamplerInfo, (self, CL_SAMPLER_ADDRESSING_MODE, sizeof (value), value, 0));
1370 EXTEND (SP, 1); 2393 EXTEND (SP, 1);
1371 const int i = 0; 2394 const int i = 0;
1372 PUSHs (sv_2mortal (newSViv (value [i]))); 2395 PUSHs (sv_2mortal (newSVuv (value [i])));
1373 2396
1374void 2397void
1375filter_mode (OpenCL::Sampler self) 2398filter_mode (OpenCL::Sampler self)
1376 PPCODE: 2399 PPCODE:
1377 cl_bool value [1]; 2400 cl_bool value [1];
1388DESTROY (OpenCL::Program self) 2411DESTROY (OpenCL::Program self)
1389 CODE: 2412 CODE:
1390 clReleaseProgram (self); 2413 clReleaseProgram (self);
1391 2414
1392void 2415void
1393build (OpenCL::Program self, OpenCL::Device device, SV *options = &PL_sv_undef) 2416build (OpenCL::Program self, SV *devices = &PL_sv_undef, SV *options = &PL_sv_undef, SV *notify = &PL_sv_undef)
2417 ALIAS:
2418 build_async = 1
1394 CODE: 2419 CODE:
2420 cl_uint device_count = 0;
2421 cl_device_id *device_list = 0;
2422
2423 if (SvOK (devices))
2424 device_list = object_list (cv, 1, "devices", devices, "OpenCL::Device", &device_count);
2425
2426 void *user_data;
2427 program_callback pfn_notify = make_program_callback (notify, &user_data);
2428
2429 if (ix)
2430 build_program_async (self, device_count, device_list, SvPVbyte_nolen (options), user_data);
2431 else
2432 NEED_SUCCESS (BuildProgram, (self, device_count, device_list, SvPVbyte_nolen (options), pfn_notify, user_data));
2433
2434#if CL_VERSION_1_2
2435
2436void
2437compile (OpenCL::Program self, SV *devices, SV *options = &PL_sv_undef, SV *headers = &PL_sv_undef, SV *notify = &PL_sv_undef)
2438 CODE:
2439 cl_uint device_count = 0;
2440 cl_device_id *device_list = 0;
2441
2442 if (SvOK (devices))
2443 device_list = object_list (cv, 1, "devices", devices, "OpenCL::Device", &device_count);
2444
2445 cl_uint header_count = 0;
2446 cl_program *header_list = 0;
2447 const char **header_name = 0;
2448
2449 if (SvOK (headers))
2450 {
2451 if (!SvROK (devices) || SvTYPE (SvRV (devices)) != SVt_PVHV)
2452 croak ("clCompileProgram: headers must be undef or a hashref of name => OpenCL::Program pairs");
2453
2454 HV *hv = (HV *)SvRV (devices);
2455
2456 header_count = hv_iterinit (hv);
2457 header_list = tmpbuf (sizeof (*header_list) * header_count);
2458 header_name = tmpbuf (sizeof (*header_name) * header_count);
2459
2460 HE *he;
2461 int i = 0;
2462 while (he = hv_iternext (hv))
2463 {
2464 header_name [i] = SvPVbyte_nolen (HeSVKEY_force (he));
2465 header_list [i] = SvCLOBJ (cv, "headers", HeVAL (he), "OpenCL::Program");
2466 ++i;
2467 }
2468 }
2469
2470 void *user_data;
2471 program_callback pfn_notify = make_program_callback (notify, &user_data);
2472
1395 NEED_SUCCESS (BuildProgram, (self, 1, &device, SvPVbyte_nolen (options), 0, 0)); 2473 NEED_SUCCESS (CompileProgram, (self, device_count, device_list, SvPVbyte_nolen (options),
2474 header_count, header_list, header_name, pfn_notify, user_data));
2475
2476#endif
1396 2477
1397void 2478void
1398build_info (OpenCL::Program self, OpenCL::Device device, cl_program_build_info name) 2479build_info (OpenCL::Program self, OpenCL::Device device, cl_program_build_info name)
1399 PPCODE: 2480 PPCODE:
1400 size_t size; 2481 size_t size;
1429 NEED_SUCCESS (GetProgramBuildInfo, (self, device, ix, size, value, 0)); 2510 NEED_SUCCESS (GetProgramBuildInfo, (self, device, ix, size, value, 0));
1430 EXTEND (SP, 1); 2511 EXTEND (SP, 1);
1431 const int i = 0; 2512 const int i = 0;
1432 PUSHs (sv_2mortal (newSVpv (value, 0))); 2513 PUSHs (sv_2mortal (newSVpv (value, 0)));
1433 2514
2515void
2516binary_type (OpenCL::Program self, OpenCL::Device device)
2517 PPCODE:
2518 cl_program_binary_type value [1];
2519 NEED_SUCCESS (GetProgramBuildInfo, (self, device, CL_PROGRAM_BINARY_TYPE, sizeof (value), value, 0));
2520 EXTEND (SP, 1);
2521 const int i = 0;
2522 PUSHs (sv_2mortal (newSVuv (UV)value [i])));
2523
1434#END:program_build 2524#END:program_build
1435 2525
1436void 2526void
1437kernel (OpenCL::Program program, SV *function) 2527kernel (OpenCL::Program program, SV *function)
1438 PPCODE: 2528 PPCODE:
1439 NEED_SUCCESS_ARG (cl_kernel kernel, CreateKernel, (program, SvPVbyte_nolen (function), &res)); 2529 NEED_SUCCESS_ARG (cl_kernel kernel, CreateKernel, (program, SvPVbyte_nolen (function), &res));
1440 XPUSH_NEW_OBJ ("OpenCL::Kernel", kernel); 2530 XPUSH_CLOBJ (stash_kernel, kernel);
2531
2532void
2533kernels_in_program (OpenCL::Program program)
2534 PPCODE:
2535 cl_uint num_kernels;
2536 NEED_SUCCESS (CreateKernelsInProgram, (program, 0, 0, &num_kernels));
2537 cl_kernel *kernels = tmpbuf (sizeof (cl_kernel) * num_kernels);
2538 NEED_SUCCESS (CreateKernelsInProgram, (program, num_kernels, kernels, 0));
2539
2540 int i;
2541 EXTEND (SP, num_kernels);
2542 for (i = 0; i < num_kernels; ++i)
2543 PUSH_CLOBJ (stash_kernel, kernels [i]);
1441 2544
1442void 2545void
1443info (OpenCL::Program self, cl_program_info name) 2546info (OpenCL::Program self, cl_program_info name)
1444 PPCODE: 2547 PPCODE:
1445 INFO (Program) 2548 INFO (Program)
1463 { 2566 {
1464 SV *sv = sv_2mortal (newSV (sizes [i])); 2567 SV *sv = sv_2mortal (newSV (sizes [i]));
1465 SvUPGRADE (sv, SVt_PV); 2568 SvUPGRADE (sv, SVt_PV);
1466 SvPOK_only (sv); 2569 SvPOK_only (sv);
1467 SvCUR_set (sv, sizes [i]); 2570 SvCUR_set (sv, sizes [i]);
1468 ptrs [i] = SvPVX (sv); 2571 ptrs [i] = (void *)SvPVX (sv);
1469 PUSHs (sv); 2572 PUSHs (sv);
1470 } 2573 }
1471 2574
1472 NEED_SUCCESS (GetProgramInfo, (self, CL_PROGRAM_BINARIES , sizeof (*ptrs ) * n, ptrs , &size)); 2575 NEED_SUCCESS (GetProgramInfo, (self, CL_PROGRAM_BINARIES , sizeof (*ptrs ) * n, ptrs , &size));
1473 if (size != sizeof (*ptrs) * n) XSRETURN_EMPTY; 2576 if (size != sizeof (*ptrs) * n) XSRETURN_EMPTY;
1493 NEED_SUCCESS (GetProgramInfo, (self, CL_PROGRAM_CONTEXT, sizeof (value), value, 0)); 2596 NEED_SUCCESS (GetProgramInfo, (self, CL_PROGRAM_CONTEXT, sizeof (value), value, 0));
1494 EXTEND (SP, 1); 2597 EXTEND (SP, 1);
1495 const int i = 0; 2598 const int i = 0;
1496 { 2599 {
1497 NEED_SUCCESS (RetainContext, (value [i])); 2600 NEED_SUCCESS (RetainContext, (value [i]));
1498 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Context", value [i])); 2601 PUSH_CLOBJ (stash_context, value [i]);
1499 } 2602 }
1500 2603
1501void 2604void
1502devices (OpenCL::Program self) 2605devices (OpenCL::Program self)
1503 PPCODE: 2606 PPCODE:
1507 NEED_SUCCESS (GetProgramInfo, (self, CL_PROGRAM_DEVICES, size, value, 0)); 2610 NEED_SUCCESS (GetProgramInfo, (self, CL_PROGRAM_DEVICES, size, value, 0));
1508 int i, n = size / sizeof (*value); 2611 int i, n = size / sizeof (*value);
1509 EXTEND (SP, n); 2612 EXTEND (SP, n);
1510 for (i = 0; i < n; ++i) 2613 for (i = 0; i < n; ++i)
1511 { 2614 {
1512 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Device", value [i])); 2615 PUSH_CLOBJ (stash_device, value [i]);
1513 } 2616 }
1514 2617
1515void 2618void
1516source (OpenCL::Program self) 2619source (OpenCL::Program self)
1517 PPCODE: 2620 PPCODE:
1543DESTROY (OpenCL::Kernel self) 2646DESTROY (OpenCL::Kernel self)
1544 CODE: 2647 CODE:
1545 clReleaseKernel (self); 2648 clReleaseKernel (self);
1546 2649
1547void 2650void
2651setf (OpenCL::Kernel self, const char *format, ...)
2652 CODE:
2653 int i;
2654 for (i = 2; ; ++i)
2655 {
2656 while (*format == ' ')
2657 ++format;
2658
2659 char type = *format++;
2660
2661 if (!type)
2662 break;
2663
2664 if (i >= items)
2665 croak ("OpenCL::Kernel::setf format string too long (not enough arguments)");
2666
2667 SV *sv = ST (i);
2668
2669 union
2670 {
2671 cl_char cc; cl_uchar cC; cl_short cs; cl_ushort cS;
2672 cl_int ci; cl_uint cI; cl_long cl; cl_ulong cL;
2673 cl_half ch; cl_float cf; cl_double cd;
2674 cl_mem cm;
2675 cl_sampler ca;
2676 size_t cz;
2677 cl_event ce;
2678 } arg;
2679 size_t size;
2680 int nullarg = 0;
2681
2682 switch (type)
2683 {
2684 case 'c': arg.cc = SvIV (sv); size = sizeof (arg.cc); break;
2685 case 'C': arg.cC = SvUV (sv); size = sizeof (arg.cC); break;
2686 case 's': arg.cs = SvIV (sv); size = sizeof (arg.cs); break;
2687 case 'S': arg.cS = SvUV (sv); size = sizeof (arg.cS); break;
2688 case 'i': arg.ci = SvIV (sv); size = sizeof (arg.ci); break;
2689 case 'I': arg.cI = SvUV (sv); size = sizeof (arg.cI); break;
2690 case 'l': arg.cl = SvIV (sv); size = sizeof (arg.cl); break;
2691 case 'L': arg.cL = SvUV (sv); size = sizeof (arg.cL); break;
2692
2693 case 'h': arg.ch = SvUV (sv); size = sizeof (arg.ch); break;
2694 case 'f': arg.cf = SvNV (sv); size = sizeof (arg.cf); break;
2695 case 'd': arg.cd = SvNV (sv); size = sizeof (arg.cd); break;
2696
2697 case 'z': nullarg = 1; size = SvIV (sv); break;
2698
2699 case 'm': nullarg = !SvOK (sv); arg.cm = SvCLOBJ (cv, "m", sv, "OpenCL::Memory" ); size = sizeof (arg.cm); break;
2700 case 'a': nullarg = !SvOK (sv); arg.ca = SvCLOBJ (cv, "a", sv, "OpenCL::Sampler"); size = sizeof (arg.ca); break;
2701 case 'e': nullarg = !SvOK (sv); arg.ca = SvCLOBJ (cv, "e", sv, "OpenCL::Event" ); size = sizeof (arg.ce); break;
2702
2703 default:
2704 croak ("OpenCL::Kernel::setf format character '%c' not supported", type);
2705 }
2706
2707 res = clSetKernelArg (self, i - 2, size, nullarg ? 0 : &arg);
2708 if (res)
2709 croak ("OpenCL::Kernel::setf kernel parameter '%c' (#%d): %s", type, i - 2, err2str (res));
2710 }
2711
2712 if (i != items)
2713 croak ("OpenCL::Kernel::setf format string too short (too many arguments)");
2714
2715void
1548set_char (OpenCL::Kernel self, cl_uint idx, cl_char value) 2716set_char (OpenCL::Kernel self, cl_uint idx, cl_char value)
1549 CODE: 2717 CODE:
1550 clSetKernelArg (self, idx, sizeof (value), &value); 2718 clSetKernelArg (self, idx, sizeof (value), &value);
1551 2719
1552void 2720void
1600 clSetKernelArg (self, idx, sizeof (value), &value); 2768 clSetKernelArg (self, idx, sizeof (value), &value);
1601 2769
1602void 2770void
1603set_memory (OpenCL::Kernel self, cl_uint idx, OpenCL::Memory_ornull value) 2771set_memory (OpenCL::Kernel self, cl_uint idx, OpenCL::Memory_ornull value)
1604 CODE: 2772 CODE:
2773 clSetKernelArg (self, idx, sizeof (value), value ? &value : 0);
2774
2775void
2776set_buffer (OpenCL::Kernel self, cl_uint idx, OpenCL::Buffer_ornull value)
2777 CODE:
2778 clSetKernelArg (self, idx, sizeof (value), value ? &value : 0);
2779
2780void
2781set_image (OpenCL::Kernel self, cl_uint idx, OpenCL::Image_ornull value)
2782 ALIAS:
2783 set_image2d = 0
2784 set_image3d = 0
2785 CODE:
2786 clSetKernelArg (self, idx, sizeof (value), value ? &value : 0);
2787
2788void
2789set_sampler (OpenCL::Kernel self, cl_uint idx, OpenCL::Sampler value)
2790 CODE:
1605 clSetKernelArg (self, idx, sizeof (value), &value); 2791 clSetKernelArg (self, idx, sizeof (value), &value);
1606 2792
1607void 2793void
1608set_buffer (OpenCL::Kernel self, cl_uint idx, OpenCL::Buffer_ornull value) 2794set_local (OpenCL::Kernel self, cl_uint idx, size_t size)
1609 CODE: 2795 CODE:
1610 clSetKernelArg (self, idx, sizeof (value), &value); 2796 clSetKernelArg (self, idx, size, 0);
1611
1612void
1613set_image2d (OpenCL::Kernel self, cl_uint idx, OpenCL::Image2D_ornull value)
1614 CODE:
1615 clSetKernelArg (self, idx, sizeof (value), &value);
1616
1617void
1618set_image3d (OpenCL::Kernel self, cl_uint idx, OpenCL::Image3D_ornull value)
1619 CODE:
1620 clSetKernelArg (self, idx, sizeof (value), &value);
1621
1622void
1623set_sampler (OpenCL::Kernel self, cl_uint idx, OpenCL::Sampler value)
1624 CODE:
1625 clSetKernelArg (self, idx, sizeof (value), &value);
1626 2797
1627void 2798void
1628set_event (OpenCL::Kernel self, cl_uint idx, OpenCL::Event value) 2799set_event (OpenCL::Kernel self, cl_uint idx, OpenCL::Event value)
1629 CODE: 2800 CODE:
1630 clSetKernelArg (self, idx, sizeof (value), &value); 2801 clSetKernelArg (self, idx, sizeof (value), &value);
1666 NEED_SUCCESS (GetKernelInfo, (self, CL_KERNEL_CONTEXT, sizeof (value), value, 0)); 2837 NEED_SUCCESS (GetKernelInfo, (self, CL_KERNEL_CONTEXT, sizeof (value), value, 0));
1667 EXTEND (SP, 1); 2838 EXTEND (SP, 1);
1668 const int i = 0; 2839 const int i = 0;
1669 { 2840 {
1670 NEED_SUCCESS (RetainContext, (value [i])); 2841 NEED_SUCCESS (RetainContext, (value [i]));
1671 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Context", value [i])); 2842 PUSH_CLOBJ (stash_context, value [i]);
1672 } 2843 }
1673 2844
1674void 2845void
1675program (OpenCL::Kernel self) 2846program (OpenCL::Kernel self)
1676 PPCODE: 2847 PPCODE:
1678 NEED_SUCCESS (GetKernelInfo, (self, CL_KERNEL_PROGRAM, sizeof (value), value, 0)); 2849 NEED_SUCCESS (GetKernelInfo, (self, CL_KERNEL_PROGRAM, sizeof (value), value, 0));
1679 EXTEND (SP, 1); 2850 EXTEND (SP, 1);
1680 const int i = 0; 2851 const int i = 0;
1681 { 2852 {
1682 NEED_SUCCESS (RetainProgram, (value [i])); 2853 NEED_SUCCESS (RetainProgram, (value [i]));
1683 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Program", value [i])); 2854 PUSH_CLOBJ (stash_program, value [i]);
1684 } 2855 }
1685 2856
1686#END:kernel 2857#END:kernel
1687 2858
1688void 2859void
1735 const int i = 0; 2906 const int i = 0;
1736 PUSHs (sv_2mortal (newSVuv (value [i]))); 2907 PUSHs (sv_2mortal (newSVuv (value [i])));
1737 2908
1738#END:kernel_work_group 2909#END:kernel_work_group
1739 2910
2911#if CL_VERSION_1_2
2912
2913void
2914arg_info (OpenCL::Kernel self, cl_uint idx, cl_kernel_arg_info name)
2915 PPCODE:
2916 size_t size;
2917 NEED_SUCCESS (GetKernelArgInfo, (self, idx, name, 0, 0, &size));
2918 SV *sv = sv_2mortal (newSV (size));
2919 SvUPGRADE (sv, SVt_PV);
2920 SvPOK_only (sv);
2921 SvCUR_set (sv, size);
2922 NEED_SUCCESS (GetKernelArgInfo, (self, idx, name, size, SvPVX (sv), 0));
2923 XPUSHs (sv);
2924
2925#BEGIN:kernel_arg
2926
2927void
2928arg_address_qualifier (OpenCL::Kernel self, cl_uint idx)
2929 PPCODE:
2930 cl_kernel_arg_address_qualifier value [1];
2931 NEED_SUCCESS (GetKernelArgInfo, (self, idx, CL_KERNEL_ARG_ADDRESS_QUALIFIER, sizeof (value), value, 0));
2932 EXTEND (SP, 1);
2933 const int i = 0;
2934 PUSHs (sv_2mortal (newSVuv (value [i])));
2935
2936void
2937arg_access_qualifier (OpenCL::Kernel self, cl_uint idx)
2938 PPCODE:
2939 cl_kernel_arg_access_qualifier value [1];
2940 NEED_SUCCESS (GetKernelArgInfo, (self, idx, CL_KERNEL_ARG_ACCESS_QUALIFIER, sizeof (value), value, 0));
2941 EXTEND (SP, 1);
2942 const int i = 0;
2943 PUSHs (sv_2mortal (newSVuv (value [i])));
2944
2945void
2946arg_type_name (OpenCL::Kernel self, cl_uint idx)
2947 ALIAS:
2948 arg_type_name = CL_KERNEL_ARG_TYPE_NAME
2949 arg_name = CL_KERNEL_ARG_NAME
2950 PPCODE:
2951 size_t size;
2952 NEED_SUCCESS (GetKernelArgInfo, (self, idx, ix, 0, 0, &size));
2953 char *value = tmpbuf (size);
2954 NEED_SUCCESS (GetKernelArgInfo, (self, idx, ix, size, value, 0));
2955 EXTEND (SP, 1);
2956 const int i = 0;
2957 PUSHs (sv_2mortal (newSVpv (value, 0)));
2958
2959void
2960arg_type_qualifier (OpenCL::Kernel self, cl_uint idx)
2961 PPCODE:
2962 cl_kernel_arg_type_qualifier value [1];
2963 NEED_SUCCESS (GetKernelArgInfo, (self, idx, CL_KERNEL_ARG_TYPE_QUALIFIER, sizeof (value), value, 0));
2964 EXTEND (SP, 1);
2965 const int i = 0;
2966 PUSHs (sv_2mortal (newSVuv (value [i])));
2967
2968#END:kernel_arg
2969
2970#endif
2971
1740MODULE = OpenCL PACKAGE = OpenCL::Event 2972MODULE = OpenCL PACKAGE = OpenCL::Event
1741 2973
1742void 2974void
1743DESTROY (OpenCL::Event self) 2975DESTROY (OpenCL::Event self)
1744 CODE: 2976 CODE:
1746 2978
1747void 2979void
1748wait (OpenCL::Event self) 2980wait (OpenCL::Event self)
1749 CODE: 2981 CODE:
1750 clWaitForEvents (1, &self); 2982 clWaitForEvents (1, &self);
2983
2984void
2985cb (OpenCL::Event self, cl_int command_exec_callback_type, SV *cb)
2986 CODE:
2987 clSetEventCallback (self, command_exec_callback_type, eq_event_notify, SvREFCNT_inc (s_get_cv (cb)));
1751 2988
1752void 2989void
1753info (OpenCL::Event self, cl_event_info name) 2990info (OpenCL::Event self, cl_event_info name)
1754 PPCODE: 2991 PPCODE:
1755 INFO (Event) 2992 INFO (Event)
1763 NEED_SUCCESS (GetEventInfo, (self, CL_EVENT_COMMAND_QUEUE, sizeof (value), value, 0)); 3000 NEED_SUCCESS (GetEventInfo, (self, CL_EVENT_COMMAND_QUEUE, sizeof (value), value, 0));
1764 EXTEND (SP, 1); 3001 EXTEND (SP, 1);
1765 const int i = 0; 3002 const int i = 0;
1766 { 3003 {
1767 NEED_SUCCESS (RetainCommandQueue, (value [i])); 3004 NEED_SUCCESS (RetainCommandQueue, (value [i]));
1768 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Queue", value [i])); 3005 PUSH_CLOBJ (stash_queue, value [i]);
1769 } 3006 }
1770 3007
1771void 3008void
1772command_type (OpenCL::Event self) 3009command_type (OpenCL::Event self)
1773 PPCODE: 3010 PPCODE:
1796 NEED_SUCCESS (GetEventInfo, (self, CL_EVENT_CONTEXT, sizeof (value), value, 0)); 3033 NEED_SUCCESS (GetEventInfo, (self, CL_EVENT_CONTEXT, sizeof (value), value, 0));
1797 EXTEND (SP, 1); 3034 EXTEND (SP, 1);
1798 const int i = 0; 3035 const int i = 0;
1799 { 3036 {
1800 NEED_SUCCESS (RetainContext, (value [i])); 3037 NEED_SUCCESS (RetainContext, (value [i]));
1801 PUSHs (NEW_MORTAL_OBJ ("OpenCL::Context", value [i])); 3038 PUSH_CLOBJ (stash_context, value [i]);
1802 } 3039 }
1803 3040
1804#END:event 3041#END:event
1805 3042
1806void 3043void
1831void 3068void
1832set_status (OpenCL::UserEvent self, cl_int execution_status) 3069set_status (OpenCL::UserEvent self, cl_int execution_status)
1833 CODE: 3070 CODE:
1834 clSetUserEventStatus (self, execution_status); 3071 clSetUserEventStatus (self, execution_status);
1835 3072
3073MODULE = OpenCL PACKAGE = OpenCL::Mapped
3074
3075void
3076DESTROY (SV *self)
3077 CODE:
3078 OpenCL__Mapped mapped = SvMAPPED (self);
3079
3080 clEnqueueUnmapMemObject (mapped->queue, mapped->memobj, mapped->ptr, 1, &mapped->event, 0);
3081 mapped_detach (self, mapped);
3082
3083 clReleaseCommandQueue (mapped->queue);
3084 clReleaseEvent (mapped->event);
3085 Safefree (mapped);
3086
3087void
3088unmap (OpenCL::Mapped self, ...)
3089 CODE:
3090 mapped_unmap (cv, ST (0), self, self->queue, &ST (1), items - 1);
3091
3092bool
3093mapped (OpenCL::Mapped self)
3094 CODE:
3095 RETVAL = !!self->ptr;
3096 OUTPUT:
3097 RETVAL
3098
3099void
3100wait (OpenCL::Mapped self)
3101 PPCODE:
3102 if (self->event)
3103 NEED_SUCCESS (WaitForEvents, (1, &self->event));
3104
3105void
3106event (OpenCL::Mapped self)
3107 PPCODE:
3108 if (!self->event)
3109 XSRETURN_UNDEF;
3110
3111 clRetainEvent (self->event);
3112 XPUSH_CLOBJ (stash_event, self->event);
3113
3114size_t
3115size (OpenCL::Mapped self)
3116 CODE:
3117 RETVAL = self->cb;
3118 OUTPUT:
3119 RETVAL
3120
3121IV
3122ptr (OpenCL::Mapped self)
3123 CODE:
3124 RETVAL = PTR2IV (self->ptr);
3125 OUTPUT:
3126 RETVAL
3127
3128void
3129set (OpenCL::Mapped self, size_t offset, SV *data)
3130 CODE:
3131 STRLEN len;
3132 const char *ptr = SvPVbyte (data, len);
3133
3134 if (offset + len > self->cb)
3135 croak ("OpenCL::Mapped::set out of bound condition detected");
3136
3137 memcpy (offset + (char *)self->ptr, ptr, len);
3138
3139MODULE = OpenCL PACKAGE = OpenCL::MappedBuffer
3140
3141MODULE = OpenCL PACKAGE = OpenCL::MappedImage
3142
3143IV
3144row_pitch (OpenCL::Mapped self)
3145 ALIAS:
3146 slice_pitch = 1
3147 CODE:
3148 RETVAL = ix ? self->slice_pitch : self->row_pitch;
3149 OUTPUT:
3150 RETVAL
3151
3152
3153

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines