-
Notifications
You must be signed in to change notification settings - Fork 2
/
ocl.h
572 lines (408 loc) · 17.9 KB
/
ocl.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
#ifndef OCL_H_INCLUDED
#define OCL_H_INCLUDED
#include <cstdio>
#include <CL/cl.h>
#include <iostream>
#include <sfml/graphics.hpp>
#include "clstate.h"
#include <windows.h>
#include <boost/compute/source.hpp>
#include <boost/compute/system.hpp>
#include <boost/compute/algorithm/iota.hpp>
#include <boost/compute/interop/opengl.hpp>
#include <unordered_map>
#include "logging.hpp"
#include <thread>
namespace compute = boost::compute;
extern std::thread build_thread;
#if 0
///to manage the difference between cl_mem classes
///and raw values
///we can pass cl_mems straight to opencl
///but we have to pass raw values by value
struct auto_argument
{
void* ptr = nullptr;
int is_naturally_pointer = 1;
auto_argument(void* pptr, int is_ptr)
{
if(is_ptr)
{
ptr = pptr;
return;
}
else
{
void** nptr = new void*;
///copy pointer value
*nptr = pptr;
ptr = (void*)(&nptr);
}
}
};
#endif
extern std::map<std::string,void*> registered_automatic_argument_map;
extern std::vector<automatic_argument_identifiers> parsed_automatic_arguments;
#include <string>
#include <sstream>
#include <vector>
bool supports_extension(const std::string& ext_name);
struct driver_blacklist_info
{
std::string friendly_name;
std::string driver_name;
std::string vendor;
int is_blacklisted = 0;
std::string get_blacklist_string()
{
return vendor + " OpenCL Driver version: " + driver_name + " Friendly driver name: " + friendly_name;
}
};
///thanks AMD
extern std::map<std::string, driver_blacklist_info> driver_blacklist;
///can't be used if we're already crashing before doing driver stuff!
bool is_driver_blacklisted(const std::string& version);
void print_blacklist_error_info();
///blatantly nicked from nvidia
cl_int oclGetPlatformID(cl_platform_id* clSelectedPlatformID);
char* file_contents(const char *filename, int *length);
///left in headers in case of future type safety
inline
void register_automatic(void* buf, const std::string& name)
{
registered_automatic_argument_map[name] = buf;
}
inline
void* get_automatic(const std::string& name, bool& success)
{
auto it = registered_automatic_argument_map.find(name);
success = false;
if(it == registered_automatic_argument_map.end())
{
return nullptr;
}
void* ptr = it->second;
if(ptr == nullptr)
return nullptr;
success = true;
return ptr;
}
extern void program_ensure_built();
static kernel load_kernel(const compute::program &p, const std::string& name)
{
program_ensure_built();
kernel k;
k.kernel = compute::kernel(p, name);
k.name = name;
k.loaded = true;
for(int i=0; i<parsed_automatic_arguments.size(); i++)
{
if(parsed_automatic_arguments[i].kernel_name == name)
{
k.automatic_arguments = parsed_automatic_arguments[i];
lg::log("Found an autoargument pack of size ", k.automatic_arguments.args.size(), " in kernel ", k.name);
break;
}
}
size_t ret = 128;
clGetKernelWorkGroupInfo(k.kernel.get(), cl::device.id(), CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &ret, NULL);
k.work_size = ret;
return k;
}
extern std::unordered_map<std::string, std::map<int, const void*>> kernel_map;
///I hate parsing things, but sometimes you've gotta do what you've gotta do
///because I aint parsing this for real
///although actualy, we could just look for kernel and do proper parsing, with attributes
///but lets take the easy and well thought out route for the moment
std::vector<automatic_argument_identifiers> parse_automatic_arguments(char* text);
bool use_3d_texture_array();
inline
std::vector<char> get_device_info(cl_device_id id, cl_device_info inf)
{
size_t size_ret;
clGetDeviceInfo(id, inf, 0, nullptr, &size_ret);
if(size_ret <= 1)
return std::vector<char>(1);
char* ret = new char[size_ret]();
clGetDeviceInfo(id, inf, size_ret, ret, nullptr);
std::vector<char> rstr;
for(int i=0; i<size_ret; i++)
{
rstr.push_back(ret[i]);
}
delete [] ret;
return rstr;
}
inline void build(const std::string& file, int w, int h, int lres, bool only_3d, const std::string& extra_build_commands)
{
int src_size=0;
char *source;
source = file_contents(file.c_str(), &src_size);
parsed_automatic_arguments = parse_automatic_arguments(source);
lg::log("Loaded file");
compute::program program = compute::program::create_with_source(source, cl::context);
free(source);
lg::log("Created program");
std::ostringstream convert;
convert << w;
std::string wstr = convert.str();
std::ostringstream converth;
converth << h;
std::string hstr = converth.str();
std::ostringstream convertlres;
convertlres << lres;
std::string lresstr = convertlres.str();
std::string pure_3d;
if(only_3d)
pure_3d = " -D fONLY_3D";
///does not compile properly without (breaks texture filtering), investigate this at some point
std::string buildoptions = "-cl-fast-relaxed-math -cl-no-signed-zeros -cl-single-precision-constant -cl-denorms-are-zero -D SCREENWIDTH=" + wstr + " -D SCREENHEIGHT=" + hstr + " -D LIGHTBUFFERDIM=" + lresstr + pure_3d;// + " -D BECKY_HACK=" + sbecky;
if(use_3d_texture_array())
{
buildoptions = buildoptions + " -D supports_3d_writes";
}
buildoptions = buildoptions + " -D SHADOWBIAS=50";
buildoptions = buildoptions + " " + extra_build_commands;
#ifdef BECKY_HACK
buildoptions = buildoptions + std::string(" -D BECKY_HACK=1");
#endif
try
{
program.build(buildoptions.c_str());
}
catch(...)
{
lg::log(program.build_log());
exit(1232345);
}
lg::log("Built program");
cl::program = program;
///make this more automatic
///calling load kernel in this thread recurses rather a lot, tis a silly place
/*cl::kernel1 = load_kernel(program, "kernel1");
cl::kernel2 = load_kernel(program, "kernel2");
cl::kernel3 = load_kernel(program, "kernel3");
cl::prearrange = load_kernel(program, "prearrange");
cl::prearrange_light = load_kernel(program, "prearrange_light");
cl::kernel1_light = load_kernel(program, "kernel1_light");
cl::clear_screen_buffer = load_kernel(program, "clear_screen_buffer");*/
#ifdef OCULUS
cl::kernel1_oculus = load_kernel(program, "kernel1_oculus");
cl::kernel2_oculus= load_kernel(program, "kernel2_oculus");
cl::kernel3_oculus = load_kernel(program, "kernel3_oculus");
cl::prearrange_oculus = load_kernel(program, "prearrange_oculus");
#endif
//cl::cloth_simulate = load_kernel(program, "cloth_simulate");
/*if(!only_3d)
{
cl::tile_clear = load_kernel(program, "tile_clear");
cl::point_cloud_depth = load_kernel(program, "point_cloud_depth_pass");
cl::point_cloud_recover = load_kernel(program, "point_cloud_recovery_pass");
cl::space_dust = load_kernel(program, "space_dust");
cl::space_dust_no_tile = load_kernel(program, "space_dust_no_tiling");
cl::draw_ui = load_kernel(program, "draw_ui");
cl::draw_hologram = load_kernel(program, "draw_hologram");
cl::blit_with_id = load_kernel(program, "blit_with_id");
cl::blit_clear = load_kernel(program, "blit_clear");
cl::clear_id_buf = load_kernel(program, "clear_id_buf");
cl::clear_screen_dbuf = load_kernel(program, "clear_screen_dbuf");
cl::draw_voxel_octree = load_kernel(program, "draw_voxel_octree");
cl::create_distortion_offset = load_kernel(program, "create_distortion_offset");
cl::draw_fancy_projectile = load_kernel(program, "draw_fancy_projectile");
cl::reproject_depth = load_kernel(program, "reproject_depth");
cl::reproject_screen = load_kernel(program, "reproject_screen");
cl::space_nebulae = load_kernel(program, "space_nebulae");
cl::edge_smoothing = load_kernel(program, "edge_smoothing");
cl::shadowmap_smoothing_x = load_kernel(program, "shadowmap_smoothing_x");
cl::shadowmap_smoothing_y = load_kernel(program, "shadowmap_smoothing_y");
cl::raytrace = load_kernel(program, "raytrace");
cl::render_voxels = load_kernel(program, "render_voxels");
cl::render_voxels_tex = load_kernel(program, "render_voxels_tex");
cl::render_voxel_cube = load_kernel(program, "render_voxel_cube");
cl::diffuse_unstable = load_kernel(program, "diffuse_unstable");
cl::diffuse_unstable_tex = load_kernel(program, "diffuse_unstable_tex");
cl::advect = load_kernel(program, "advect");
cl::advect_tex = load_kernel(program, "advect_tex");
cl::post_upscale = load_kernel(program, "post_upscale");
cl::warp_oculus = load_kernel(program, "warp_oculus");
cl::goo_diffuse = load_kernel(program, "goo_diffuse");
cl::goo_advect = load_kernel(program, "goo_advect");
cl::fluid_amount = load_kernel(program, "fluid_amount");
cl::update_boundary = load_kernel(program, "update_boundary");
cl::fluid_initialise_mem = load_kernel(program, "fluid_initialise_mem");
cl::fluid_initialise_mem_3d = load_kernel(program, "fluid_initialise_mem_3d");
cl::fluid_timestep = load_kernel(program, "fluid_timestep");
cl::fluid_timestep_3d = load_kernel(program, "fluid_timestep_3d");
cl::displace_fluid = load_kernel(program, "displace_fluid");
cl::process_skins = load_kernel(program, "process_skins");
cl::draw_hermite_skin = load_kernel(program, "draw_hermite_skin");
}*/
kernel_map.clear();
cl::kernels.clear();
lg::log("Loaded obscene numbers of kernels");
}
inline void oclstuff(const std::string& file, int w, int h, int lres, bool only_3d, const std::string& extra_build_commands)
{
///need to initialise context and the like
///cant use boost::compute as it does not support opengl context sharing on windows
cl_int error = 0; // Used to handle error codes
cl_platform_id platform;
// Platform
error = oclGetPlatformID(&platform);
if(error != CL_SUCCESS)
{
lg::log("Error getting platform id: ", error);
print_blacklist_error_info();
exit(error);
}
else
{
lg::log("Got platform IDs");
}
cl_uint num;
cl_device_id device[100];
// Device
error = clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, device, &num);
lg::log("Found ", num, " devices");
if(error != CL_SUCCESS)
{
lg::log("Error getting device ids: ", error);
print_blacklist_error_info();
exit(error);
}
else
{
lg::log("Got device ids");
}
///this is essentially black magic
cl_context_properties props[] =
{
CL_GL_CONTEXT_KHR, (cl_context_properties)wglGetCurrentContext(),
CL_WGL_HDC_KHR, (cl_context_properties)wglGetCurrentDC(),
CL_CONTEXT_PLATFORM, (cl_context_properties)platform,
0
};
///I think the context is invalid
///because all the resources were created under the other context
///maybe create a new reload function which skips all this
// Context
cl_context context = clCreateContext(props, 1, &device[0], NULL, NULL, &error);
if(error != CL_SUCCESS)
{
lg::log("Error creating context: ", error);
print_blacklist_error_info();
exit(error);
}
else
{
lg::log("Created context");
}
///you know... I'm not sure we *do* want to retain the context here...!
cl::context = compute::context(context, true);
lg::log("Bound context");
cl::device = compute::device(device[0], true);
lg::log("Bound device");
std::vector<char> size_device = get_device_info(cl::device.get(), CL_DEVICE_GLOBAL_MEM_SIZE);
if(size_device.size() == sizeof(cl_ulong))
lg::log("Mem size of device ", *(cl_ulong*)&size_device[0] / 1024 / 1024);
std::vector<char> opencl_version = get_device_info(cl::device.get(), CL_DEVICE_OPENCL_C_VERSION);
lg::log("OpenCL version ", &opencl_version[0]);
std::vector<char> device_version = get_device_info(cl::device.get(), CL_DEVICE_VERSION);
lg::log("Device OpenCL version ", &device_version[0]);
std::vector<char> driver_version = get_device_info(cl::device.get(), CL_DRIVER_VERSION);
lg::log("Driver OpenCL version ", &driver_version[0]);
std::vector<char> interop_sync = get_device_info(cl::device.get(), CL_DEVICE_PREFERRED_INTEROP_USER_SYNC);
if(interop_sync.size() == sizeof(cl_bool))
lg::log("Does device prefer user interop sync ", *(cl_bool*)&interop_sync[0]);
std::vector<char> unified_memory = get_device_info(cl::device.get(), CL_DEVICE_HOST_UNIFIED_MEMORY);
if(unified_memory.size() == sizeof(cl_bool))
lg::log("Does device have unified memory ", *(cl_bool*)&unified_memory[0]);
std::vector<char> device_name = get_device_info(cl::device.get(), CL_DEVICE_NAME);
lg::log("Device name ", &device_name[0]);
std::vector<char> device_vendor = get_device_info(cl::device.get(), CL_DEVICE_VENDOR);
lg::log("Device vendor ", &device_vendor[0]);
std::vector<char> image_width = get_device_info(cl::device.get(), CL_DEVICE_IMAGE3D_MAX_WIDTH);
if(image_width.size() == sizeof(size_t))
lg::log("Max image width ", *(size_t*)&image_width[0]);
std::vector<char> image_height = get_device_info(cl::device.get(), CL_DEVICE_IMAGE3D_MAX_HEIGHT);
if(image_height.size() == sizeof(size_t))
lg::log("Max image height ", *(size_t*)&image_height[0]);
std::vector<char> image_depth = get_device_info(cl::device.get(), CL_DEVICE_IMAGE3D_MAX_DEPTH);
if(image_depth.size() == sizeof(size_t))
lg::log("Max image depth ", *(size_t*)&image_depth[0]);
std::vector<char> supports_images = get_device_info(cl::device.get(), CL_DEVICE_IMAGE_SUPPORT);
if(supports_images.size() == sizeof(cl_bool))
lg::log("Supports images ", *(cl_bool*)&supports_images[0]);
std::vector<char> max_allocation_size = get_device_info(cl::device.get(), CL_DEVICE_MAX_MEM_ALLOC_SIZE);
if(max_allocation_size.size() == sizeof(cl_ulong))
lg::log("Max allocation size ", *(cl_ulong*)&max_allocation_size[0] / 1024 / 1024);
std::vector<char> max_parameter_size = get_device_info(cl::device.get(), CL_DEVICE_MAX_PARAMETER_SIZE);
if(max_parameter_size.size() == sizeof(size_t))
lg::log("Max parameter size ", *(size_t*)&max_parameter_size[0], " bytes");
std::vector<char> max_work_group_size = get_device_info(cl::device.get(), CL_DEVICE_MAX_WORK_GROUP_SIZE);
if(max_work_group_size.size() == sizeof(size_t))
lg::log("Max work group size ", *(size_t*)&max_work_group_size[0]);
std::vector<char> max_work_item_dims = get_device_info(cl::device.get(), CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS);
cl_uint nargs = 0;
if(max_work_item_dims.size() == sizeof(cl_uint))
{
nargs = *(cl_uint*)&max_work_item_dims[0];
lg::log("Max work item dims ", nargs);
}
std::vector<char> max_item_sizes = get_device_info(cl::device.get(), CL_DEVICE_MAX_WORK_ITEM_SIZES);
for(int i=0; i<nargs; i++)
{
int c = i * sizeof(size_t);
size_t n = *(size_t*)&max_item_sizes[c];
lg::log("Item max dim i ", i, " ", n);
}
if(!supports_extension("cl_khr_gl_sharing"))
{
lg::log("This device and driver combination does not support OpenCL/OpenGL sharing and will almost certainly not work");
lg::log("Try updating your drivers, or if the latest drivers don't support it, complain to whoever provides them");
}
bool supports_efficient_clgl_interop = supports_extension("cl_khr_gl_event");
lg::log("Does this device support efficient cl/gl interop with cl_khr_gl_event: ", supports_efficient_clgl_interop);
if(!supports_efficient_clgl_interop)
{
lg::log("Warning: Device does not support fast implicit synchronisation, update your drivers or complain to your gpu vendor");
}
#ifdef PROFILING
cl::cqueue = compute::command_queue(cl::context, cl::device, CL_QUEUE_PROFILING_ENABLE);
cl::cqueue2 = compute::command_queue(cl::context, cl::device, CL_QUEUE_PROFILING_ENABLE);
cl::cqueue_ooo = compute::command_queue(cl::context, cl::device, CL_QUEUE_PROFILING_ENABLE | CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE);
#else
cl::cqueue = compute::command_queue(cl::context, cl::device);
cl::cqueue2 = compute::command_queue(cl::context, cl::device); //CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE
cl::cqueue_ooo = compute::command_queue(cl::context, cl::device, CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE);
#endif
lg::log("Created command queue");
cl::any_built_requested = true;
build_thread = std::thread(build, file, w, h, lres, only_3d, extra_build_commands);
//build_thread.join();
//build(file, w, h, lres, only_3d, extra_build_commands);
}
inline
void reset_program_built()
{
if(cl::any_built_requested)
program_ensure_built();
cl::program_built = false;
}
inline
void program_ensure_built()
{
if(cl::program_built)
return;
lg::log("Trying to join thread");
build_thread.join();
cl::program_built = true;
lg::log("Program thread joined");
}
template<typename T>
compute::buffer make_single_element()
{
T def = T();
return compute::buffer(cl::context, sizeof(T), CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR, &def);
}
#endif // OCL_H_INCLUDED