1
/*******************************************************************************
2
* Copyright (c) 2008-2016 The Khronos Group Inc.
4
* Permission is hereby granted, free of charge, to any person obtaining a
5
* copy of this software and/or associated documentation files (the
6
* "Materials"), to deal in the Materials without restriction, including
7
* without limitation the rights to use, copy, modify, merge, publish,
8
* distribute, sublicense, and/or sell copies of the Materials, and to
9
* permit persons to whom the Materials are furnished to do so, subject to
10
* the following conditions:
12
* The above copyright notice and this permission notice shall be included
13
* in all copies or substantial portions of the Materials.
15
* MODIFICATIONS TO THIS FILE MAY MEAN IT NO LONGER ACCURATELY REFLECTS
16
* KHRONOS STANDARDS. THE UNMODIFIED, NORMATIVE VERSIONS OF KHRONOS
17
* SPECIFICATIONS AND HEADER INFORMATION ARE LOCATED AT
18
* https://www.khronos.org/registry/
20
* THE MATERIALS ARE PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
21
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
22
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
23
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
24
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
25
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
26
* MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
27
******************************************************************************/
31
* \brief C++ bindings for OpenCL 1.0 (rev 48), OpenCL 1.1 (rev 33),
32
* OpenCL 1.2 (rev 15), OpenCL 2.0 (rev 29), OpenCL 2.1 (rev 17),
33
* and OpenCL 2.2 (V2.2-11).
34
* \author Lee Howes and Bruce Merry
36
* Derived from the OpenCL 1.x C++ bindings written by
37
* Benedict R. Gaster, Laurent Morichetti and Lee Howes
38
* With additions and fixes from:
39
* Brian Cole, March 3rd 2010 and April 2012
40
* Matt Gruenke, April 2012.
41
* Bruce Merry, February 2013.
42
* Tom Deakin and Simon McIntosh-Smith, July 2013
47
* Optional extension support
49
* cl_ext_device_fission
50
* #define CL_HPP_USE_CL_DEVICE_FISSION
51
* cl_khr_d3d10_sharing
52
* #define CL_HPP_USE_DX_INTEROP
54
* #define CL_HPP_USE_CL_SUB_GROUPS_KHR
55
* cl_khr_image2d_from_buffer
56
* #define CL_HPP_USE_CL_IMAGE2D_FROM_BUFFER_KHR
58
* Doxygen documentation for this header is available here:
60
* http://khronosgroup.github.io/OpenCL-CLHPP/
62
* The latest version of this header can be found on the GitHub releases page:
64
* https://github.com/KhronosGroup/OpenCL-CLHPP/releases
66
* Bugs and patches can be submitted to the GitHub repository:
68
* https://github.com/KhronosGroup/OpenCL-CLHPP
72
* \section intro Introduction
73
* For many large applications C++ is the language of choice and so it seems
74
* reasonable to define C++ bindings for OpenCL.
76
* The interface is contained with a single C++ header file \em cl2.hpp and all
77
* definitions are contained within the namespace \em cl. There is no additional
78
* requirement to include \em cl.h and to use either the C++ or original C
79
* bindings; it is enough to simply include \em cl2.hpp.
81
* The bindings themselves are lightweight and correspond closely to the
82
* underlying C API. Using the C++ bindings introduces no additional execution
85
* There are numerous compatibility, portability and memory management
86
* fixes in the new header as well as additional OpenCL 2.0 features.
87
* As a result the header is not directly backward compatible and for this
88
* reason we release it as cl2.hpp rather than a new version of cl.hpp.
91
* \section compatibility Compatibility
92
* Due to the evolution of the underlying OpenCL API the 2.0 C++ bindings
93
* include an updated approach to defining supported feature versions
94
* and the range of valid underlying OpenCL runtime versions supported.
96
* The combination of preprocessor macros CL_HPP_TARGET_OPENCL_VERSION and
97
* CL_HPP_MINIMUM_OPENCL_VERSION control this range. These are three digit
98
* decimal values representing OpenCL runime versions. The default for
99
* the target is 200, representing OpenCL 2.0 and the minimum is also
100
* defined as 200. These settings would use 2.0 API calls only.
101
* If backward compatibility with a 1.2 runtime is required, the minimum
102
* version may be set to 120.
104
* Note that this is a compile-time setting, and so affects linking against
105
* a particular SDK version rather than the versioning of the loaded runtime.
107
* The earlier versions of the header included basic vector and string
108
* classes based loosely on STL versions. These were difficult to
109
* maintain and very rarely used. For the 2.0 header we now assume
110
* the presence of the standard library unless requested otherwise.
111
* We use std::array, std::vector, std::shared_ptr and std::string
112
* throughout to safely manage memory and reduce the chance of a
113
* recurrance of earlier memory management bugs.
115
* These classes are used through typedefs in the cl namespace:
116
* cl::array, cl::vector, cl::pointer and cl::string.
117
* In addition cl::allocate_pointer forwards to std::allocate_shared
119
* In all cases these standard library classes can be replaced with
120
* custom interface-compatible versions using the CL_HPP_NO_STD_ARRAY,
121
* CL_HPP_NO_STD_VECTOR, CL_HPP_NO_STD_UNIQUE_PTR and
122
* CL_HPP_NO_STD_STRING macros.
124
* The OpenCL 1.x versions of the C++ bindings included a size_t wrapper
125
* class to interface with kernel enqueue. This caused unpleasant interactions
126
* with the standard size_t declaration and led to namespacing bugs.
127
* In the 2.0 version we have replaced this with a std::array-based interface.
128
* However, the old behaviour can be regained for backward compatibility
129
* using the CL_HPP_ENABLE_SIZE_T_COMPATIBILITY macro.
131
* Finally, the program construction interface used a clumsy vector-of-pairs
132
* design in the earlier versions. We have replaced that with a cleaner
133
* vector-of-vectors and vector-of-strings design. However, for backward
134
* compatibility old behaviour can be regained with the
135
* CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY macro.
137
* In OpenCL 2.0 OpenCL C is not entirely backward compatibility with
138
* earlier versions. As a result a flag must be passed to the OpenCL C
139
* compiled to request OpenCL 2.0 compilation of kernels with 1.2 as
140
* the default in the absence of the flag.
141
* In some cases the C++ bindings automatically compile code for ease.
142
* For those cases the compilation defaults to OpenCL C 2.0.
143
* If this is not wanted, the CL_HPP_CL_1_2_DEFAULT_BUILD macro may
144
* be specified to assume 1.2 compilation.
145
* If more fine-grained decisions on a per-kernel bases are required
146
* then explicit build operations that take the flag should be used.
149
* \section parameterization Parameters
150
* This header may be parameterized by a set of preprocessor macros.
152
* - CL_HPP_TARGET_OPENCL_VERSION
154
* Defines the target OpenCL runtime version to build the header
155
* against. Defaults to 200, representing OpenCL 2.0.
157
* - CL_HPP_NO_STD_STRING
159
* Do not use the standard library string class. cl::string is not
160
* defined and may be defined by the user before cl2.hpp is
163
* - CL_HPP_NO_STD_VECTOR
165
* Do not use the standard library vector class. cl::vector is not
166
* defined and may be defined by the user before cl2.hpp is
169
* - CL_HPP_NO_STD_ARRAY
171
* Do not use the standard library array class. cl::array is not
172
* defined and may be defined by the user before cl2.hpp is
175
* - CL_HPP_NO_STD_UNIQUE_PTR
177
* Do not use the standard library unique_ptr class. cl::pointer and
178
* the cl::allocate_pointer functions are not defined and may be
179
* defined by the user before cl2.hpp is included.
181
* - CL_HPP_ENABLE_DEVICE_FISSION
183
* Enables device fission for OpenCL 1.2 platforms.
185
* - CL_HPP_ENABLE_EXCEPTIONS
187
* Enable exceptions for use in the C++ bindings header. This is the
188
* preferred error handling mechanism but is not required.
190
* - CL_HPP_ENABLE_SIZE_T_COMPATIBILITY
192
* Backward compatibility option to support cl.hpp-style size_t
193
* class. Replaces the updated std::array derived version and
194
* removal of size_t from the namespace. Note that in this case the
195
* new size_t class is placed in the cl::compatibility namespace and
196
* thus requires an additional using declaration for direct backward
199
* - CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY
201
* Enable older vector of pairs interface for construction of
204
* - CL_HPP_CL_1_2_DEFAULT_BUILD
206
* Default to OpenCL C 1.2 compilation rather than OpenCL C 2.0
207
* applies to use of cl::Program construction and other program
210
* - CL_HPP_USE_CL_SUB_GROUPS_KHR
212
* Enable the cl_khr_subgroups extension.
214
* - CL_HPP_USE_IL_KHR
216
* Enable the cl_khr_il_program extension.
219
* \section example Example
221
* The following example shows a general use case for the C++
222
* bindings, including support for the optional exception feature and
223
* also the supplied vector and string classes, see following sections for
224
* decriptions of these features.
227
#define CL_HPP_ENABLE_EXCEPTIONS
228
#define CL_HPP_TARGET_OPENCL_VERSION 200
230
#include <CL/cl2.hpp>
236
const int numElements = 32;
240
// Filter for a 2.0 platform and set it as the default
241
std::vector<cl::Platform> platforms;
242
cl::Platform::get(&platforms);
244
for (auto &p : platforms) {
245
std::string platver = p.getInfo<CL_PLATFORM_VERSION>();
246
if (platver.find("OpenCL 2.") != std::string::npos) {
251
std::cout << "No OpenCL 2.0 platform found.";
255
cl::Platform newP = cl::Platform::setDefault(plat);
257
std::cout << "Error setting default platform.";
261
// Use C++11 raw string literals for kernel source code
262
std::string kernel1{R"CLC(
264
kernel void updateGlobal()
269
std::string kernel2{R"CLC(
270
typedef struct { global int *bar; } Foo;
271
kernel void vectorAdd(global const Foo* aNum, global const int *inputA, global const int *inputB,
272
global int *output, int val, write_only pipe int outPipe, queue_t childQueue)
274
output[get_global_id(0)] = inputA[get_global_id(0)] + inputB[get_global_id(0)] + val + *(aNum->bar);
275
write_pipe(outPipe, &val);
276
queue_t default_queue = get_default_queue();
277
ndrange_t ndrange = ndrange_1D(get_global_size(0)/2, get_global_size(0)/2);
279
// Have a child kernel write into third quarter of output
280
enqueue_kernel(default_queue, CLK_ENQUEUE_FLAGS_WAIT_KERNEL, ndrange,
282
output[get_global_size(0)*2 + get_global_id(0)] =
283
inputA[get_global_size(0)*2 + get_global_id(0)] + inputB[get_global_size(0)*2 + get_global_id(0)] + globalA;
286
// Have a child kernel write into last quarter of output
287
enqueue_kernel(childQueue, CLK_ENQUEUE_FLAGS_WAIT_KERNEL, ndrange,
289
output[get_global_size(0)*3 + get_global_id(0)] =
290
inputA[get_global_size(0)*3 + get_global_id(0)] + inputB[get_global_size(0)*3 + get_global_id(0)] + globalA + 2;
295
// New simpler string interface style
296
std::vector<std::string> programStrings {kernel1, kernel2};
298
cl::Program vectorAddProgram(programStrings);
300
vectorAddProgram.build("-cl-std=CL2.0");
303
// Print build info for all devices
304
cl_int buildErr = CL_SUCCESS;
305
auto buildInfo = vectorAddProgram.getBuildInfo<CL_PROGRAM_BUILD_LOG>(&buildErr);
306
for (auto &pair : buildInfo) {
307
std::cerr << pair.second << std::endl << std::endl;
313
typedef struct { int *bar; } Foo;
315
// Get and run kernel that initializes the program-scope global
316
// A test for kernels that take no arguments
317
auto program2Kernel =
318
cl::KernelFunctor<>(vectorAddProgram, "updateGlobal");
326
auto anSVMInt = cl::allocate_svm<int, cl::SVMTraitCoarse<>>();
328
cl::SVMAllocator<Foo, cl::SVMTraitCoarse<cl::SVMTraitReadOnly<>>> svmAllocReadOnly;
329
auto fooPointer = cl::allocate_pointer<Foo>(svmAllocReadOnly);
330
fooPointer->bar = anSVMInt.get();
331
cl::SVMAllocator<int, cl::SVMTraitCoarse<>> svmAlloc;
332
std::vector<int, cl::SVMAllocator<int, cl::SVMTraitCoarse<>>> inputA(numElements, 1, svmAlloc);
333
cl::coarse_svm_vector<int> inputB(numElements, 2, svmAlloc);
338
// Traditional cl_mem allocations
339
std::vector<int> output(numElements, 0xdeadbeef);
340
cl::Buffer outputBuffer(begin(output), end(output), false);
341
cl::Pipe aPipe(sizeof(cl_int), numElements / 2);
343
// Default command queue, also passed in as a parameter
344
cl::DeviceCommandQueue defaultDeviceQueue = cl::DeviceCommandQueue::makeDefault(
345
cl::Context::getDefault(), cl::Device::getDefault());
347
auto vectorAddKernel =
349
decltype(fooPointer)&,
351
cl::coarse_svm_vector<int>&,
355
cl::DeviceCommandQueue
356
>(vectorAddProgram, "vectorAdd");
358
// Ensure that the additional SVM pointer is available to the kernel
359
// This one was not passed as a parameter
360
vectorAddKernel.setSVMPointers(anSVMInt);
362
// Hand control of coarse allocations to runtime
363
cl::enqueueUnmapSVM(anSVMInt);
364
cl::enqueueUnmapSVM(fooPointer);
365
cl::unmapSVM(inputB);
366
cl::unmapSVM(output2);
371
cl::NDRange(numElements/2),
372
cl::NDRange(numElements/2)),
383
cl::copy(outputBuffer, begin(output), end(output));
384
// Grab the SVM output vector using a map
387
cl::Device d = cl::Device::getDefault();
389
std::cout << "Output:\n";
390
for (int i = 1; i < numElements; ++i) {
391
std::cout << "\t" << output[i] << "\n";
404
/* Handle deprecated preprocessor definitions. In each case, we only check for
405
* the old name if the new name is not defined, so that user code can define
406
* both and hence work with either version of the bindings.
408
#if !defined(CL_HPP_USE_DX_INTEROP) && defined(USE_DX_INTEROP)
409
# pragma message("cl2.hpp: USE_DX_INTEROP is deprecated. Define CL_HPP_USE_DX_INTEROP instead")
410
# define CL_HPP_USE_DX_INTEROP
412
#if !defined(CL_HPP_USE_CL_DEVICE_FISSION) && defined(USE_CL_DEVICE_FISSION)
413
# pragma message("cl2.hpp: USE_CL_DEVICE_FISSION is deprecated. Define CL_HPP_USE_CL_DEVICE_FISSION instead")
414
# define CL_HPP_USE_CL_DEVICE_FISSION
416
#if !defined(CL_HPP_ENABLE_EXCEPTIONS) && defined(__CL_ENABLE_EXCEPTIONS)
417
# pragma message("cl2.hpp: __CL_ENABLE_EXCEPTIONS is deprecated. Define CL_HPP_ENABLE_EXCEPTIONS instead")
418
# define CL_HPP_ENABLE_EXCEPTIONS
420
#if !defined(CL_HPP_NO_STD_VECTOR) && defined(__NO_STD_VECTOR)
421
# pragma message("cl2.hpp: __NO_STD_VECTOR is deprecated. Define CL_HPP_NO_STD_VECTOR instead")
422
# define CL_HPP_NO_STD_VECTOR
424
#if !defined(CL_HPP_NO_STD_STRING) && defined(__NO_STD_STRING)
425
# pragma message("cl2.hpp: __NO_STD_STRING is deprecated. Define CL_HPP_NO_STD_STRING instead")
426
# define CL_HPP_NO_STD_STRING
428
#if defined(VECTOR_CLASS)
429
# pragma message("cl2.hpp: VECTOR_CLASS is deprecated. Alias cl::vector instead")
431
#if defined(STRING_CLASS)
432
# pragma message("cl2.hpp: STRING_CLASS is deprecated. Alias cl::string instead.")
434
#if !defined(CL_HPP_USER_OVERRIDE_ERROR_STRINGS) && defined(__CL_USER_OVERRIDE_ERROR_STRINGS)
435
# pragma message("cl2.hpp: __CL_USER_OVERRIDE_ERROR_STRINGS is deprecated. Define CL_HPP_USER_OVERRIDE_ERROR_STRINGS instead")
436
# define CL_HPP_USER_OVERRIDE_ERROR_STRINGS
439
/* Warn about features that are no longer supported
441
#if defined(__USE_DEV_VECTOR)
442
# pragma message("cl2.hpp: __USE_DEV_VECTOR is no longer supported. Expect compilation errors")
444
#if defined(__USE_DEV_STRING)
445
# pragma message("cl2.hpp: __USE_DEV_STRING is no longer supported. Expect compilation errors")
448
/* Detect which version to target */
449
#if !defined(CL_HPP_TARGET_OPENCL_VERSION)
450
# pragma message("cl2.hpp: CL_HPP_TARGET_OPENCL_VERSION is not defined. It will default to 220 (OpenCL 2.2)")
451
# define CL_HPP_TARGET_OPENCL_VERSION 220
453
#if CL_HPP_TARGET_OPENCL_VERSION != 100 && \
454
CL_HPP_TARGET_OPENCL_VERSION != 110 && \
455
CL_HPP_TARGET_OPENCL_VERSION != 120 && \
456
CL_HPP_TARGET_OPENCL_VERSION != 200 && \
457
CL_HPP_TARGET_OPENCL_VERSION != 210 && \
458
CL_HPP_TARGET_OPENCL_VERSION != 220
459
# pragma message("cl2.hpp: CL_HPP_TARGET_OPENCL_VERSION is not a valid value (100, 110, 120, 200, 210 or 220). It will be set to 220")
460
# undef CL_HPP_TARGET_OPENCL_VERSION
461
# define CL_HPP_TARGET_OPENCL_VERSION 220
464
/* Forward target OpenCL version to C headers if necessary */
465
#if defined(CL_TARGET_OPENCL_VERSION)
466
/* Warn if prior definition of CL_TARGET_OPENCL_VERSION is lower than
467
* requested C++ bindings version */
468
#if CL_TARGET_OPENCL_VERSION < CL_HPP_TARGET_OPENCL_VERSION
469
# pragma message("CL_TARGET_OPENCL_VERSION is already defined as is lower than CL_HPP_TARGET_OPENCL_VERSION")
472
# define CL_TARGET_OPENCL_VERSION CL_HPP_TARGET_OPENCL_VERSION
475
#if !defined(CL_HPP_MINIMUM_OPENCL_VERSION)
476
# define CL_HPP_MINIMUM_OPENCL_VERSION 200
478
#if CL_HPP_MINIMUM_OPENCL_VERSION != 100 && \
479
CL_HPP_MINIMUM_OPENCL_VERSION != 110 && \
480
CL_HPP_MINIMUM_OPENCL_VERSION != 120 && \
481
CL_HPP_MINIMUM_OPENCL_VERSION != 200 && \
482
CL_HPP_MINIMUM_OPENCL_VERSION != 210 && \
483
CL_HPP_MINIMUM_OPENCL_VERSION != 220
484
# pragma message("cl2.hpp: CL_HPP_MINIMUM_OPENCL_VERSION is not a valid value (100, 110, 120, 200, 210 or 220). It will be set to 100")
485
# undef CL_HPP_MINIMUM_OPENCL_VERSION
486
# define CL_HPP_MINIMUM_OPENCL_VERSION 100
488
#if CL_HPP_MINIMUM_OPENCL_VERSION > CL_HPP_TARGET_OPENCL_VERSION
489
# error "CL_HPP_MINIMUM_OPENCL_VERSION must not be greater than CL_HPP_TARGET_OPENCL_VERSION"
492
#if CL_HPP_MINIMUM_OPENCL_VERSION <= 100 && !defined(CL_USE_DEPRECATED_OPENCL_1_0_APIS)
493
# define CL_USE_DEPRECATED_OPENCL_1_0_APIS
495
#if CL_HPP_MINIMUM_OPENCL_VERSION <= 110 && !defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
496
# define CL_USE_DEPRECATED_OPENCL_1_1_APIS
498
#if CL_HPP_MINIMUM_OPENCL_VERSION <= 120 && !defined(CL_USE_DEPRECATED_OPENCL_1_2_APIS)
499
# define CL_USE_DEPRECATED_OPENCL_1_2_APIS
501
#if CL_HPP_MINIMUM_OPENCL_VERSION <= 200 && !defined(CL_USE_DEPRECATED_OPENCL_2_0_APIS)
502
# define CL_USE_DEPRECATED_OPENCL_2_0_APIS
504
#if CL_HPP_MINIMUM_OPENCL_VERSION <= 210 && !defined(CL_USE_DEPRECATED_OPENCL_2_1_APIS)
505
# define CL_USE_DEPRECATED_OPENCL_2_1_APIS
507
#if CL_HPP_MINIMUM_OPENCL_VERSION <= 220 && !defined(CL_USE_DEPRECATED_OPENCL_2_2_APIS)
508
# define CL_USE_DEPRECATED_OPENCL_2_2_APIS
515
#if defined(CL_HPP_USE_DX_INTEROP)
516
#include <CL/cl_d3d10.h>
517
#include <CL/cl_dx9_media_sharing.h>
521
#if defined(_MSC_VER)
525
// Check for a valid C++ version
527
// Need to do both tests here because for some reason __cplusplus is not
528
// updated in visual studio
529
#if (!defined(_MSC_VER) && __cplusplus < 201103L) || (defined(_MSC_VER) && _MSC_VER < 1700)
530
#error Visual studio 2013 or another C++11-supporting compiler required
534
#if defined(CL_HPP_USE_CL_DEVICE_FISSION) || defined(CL_HPP_USE_CL_SUB_GROUPS_KHR)
535
#include <CL/cl_ext.h>
538
#if defined(__APPLE__) || defined(__MACOSX)
539
#include <OpenCL/opencl.h>
541
#include <CL/opencl.h>
544
#if (__cplusplus >= 201103L)
545
#define CL_HPP_NOEXCEPT_ noexcept
547
#define CL_HPP_NOEXCEPT_
550
#if defined(_MSC_VER)
551
# define CL_HPP_DEFINE_STATIC_MEMBER_ __declspec(selectany)
552
#elif defined(__MINGW32__)
553
# define CL_HPP_DEFINE_STATIC_MEMBER_ __attribute__((selectany))
555
# define CL_HPP_DEFINE_STATIC_MEMBER_ __attribute__((weak))
558
// Define deprecated prefixes and suffixes to ensure compilation
559
// in case they are not pre-defined
560
#if !defined(CL_EXT_PREFIX__VERSION_1_1_DEPRECATED)
561
#define CL_EXT_PREFIX__VERSION_1_1_DEPRECATED
562
#endif // #if !defined(CL_EXT_PREFIX__VERSION_1_1_DEPRECATED)
563
#if !defined(CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED)
564
#define CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED
565
#endif // #if !defined(CL_EXT_PREFIX__VERSION_1_1_DEPRECATED)
567
#if !defined(CL_EXT_PREFIX__VERSION_1_2_DEPRECATED)
568
#define CL_EXT_PREFIX__VERSION_1_2_DEPRECATED
569
#endif // #if !defined(CL_EXT_PREFIX__VERSION_1_2_DEPRECATED)
570
#if !defined(CL_EXT_SUFFIX__VERSION_1_2_DEPRECATED)
571
#define CL_EXT_SUFFIX__VERSION_1_2_DEPRECATED
572
#endif // #if !defined(CL_EXT_PREFIX__VERSION_1_2_DEPRECATED)
574
#if !defined(CL_CALLBACK)
583
#include <functional>
586
// Define a size_type to represent a correctly resolved size_t
587
#if defined(CL_HPP_ENABLE_SIZE_T_COMPATIBILITY)
589
using size_type = ::size_t;
591
#else // #if defined(CL_HPP_ENABLE_SIZE_T_COMPATIBILITY)
593
using size_type = size_t;
595
#endif // #if defined(CL_HPP_ENABLE_SIZE_T_COMPATIBILITY)
598
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
600
#endif // #if defined(CL_HPP_ENABLE_EXCEPTIONS)
602
#if !defined(CL_HPP_NO_STD_VECTOR)
605
template < class T, class Alloc = std::allocator<T> >
606
using vector = std::vector<T, Alloc>;
608
#endif // #if !defined(CL_HPP_NO_STD_VECTOR)
610
#if !defined(CL_HPP_NO_STD_STRING)
613
using string = std::string;
615
#endif // #if !defined(CL_HPP_NO_STD_STRING)
617
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
619
#if !defined(CL_HPP_NO_STD_UNIQUE_PTR)
622
// Replace unique_ptr and allocate_pointer for internal use
623
// to allow user to replace them
624
template<class T, class D>
625
using pointer = std::unique_ptr<T, D>;
628
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
629
#if !defined(CL_HPP_NO_STD_ARRAY)
632
template < class T, size_type N >
633
using array = std::array<T, N>;
635
#endif // #if !defined(CL_HPP_NO_STD_ARRAY)
637
// Define size_type appropriately to allow backward-compatibility
638
// use of the old size_t interface class
639
#if defined(CL_HPP_ENABLE_SIZE_T_COMPATIBILITY)
641
namespace compatibility {
642
/*! \brief class used to interface between C++ and
643
* OpenCL C calls that require arrays of size_t values, whose
644
* size is known statically.
653
//! \brief Initialize size_t to all 0s
656
for (int i = 0; i < N; ++i) {
661
size_t(const array<size_type, N> &rhs)
663
for (int i = 0; i < N; ++i) {
668
size_type& operator[](int index)
673
const size_type& operator[](int index) const
678
//! \brief Conversion operator to T*.
679
operator size_type* () { return data_; }
681
//! \brief Conversion operator to const T*.
682
operator const size_type* () const { return data_; }
684
operator array<size_type, N>() const
686
array<size_type, N> ret;
688
for (int i = 0; i < N; ++i) {
694
} // namespace compatibility
697
using size_t = compatibility::size_t<N>;
699
#endif // #if defined(CL_HPP_ENABLE_SIZE_T_COMPATIBILITY)
701
// Helper alias to avoid confusing the macros
704
using size_t_array = array<size_type, 3>;
705
} // namespace detail
711
* \brief The OpenCL C++ bindings are defined within this namespace.
717
#define CL_HPP_INIT_CL_EXT_FCN_PTR_(name) \
719
pfn_##name = (PFN_##name) \
720
clGetExtensionFunctionAddress(#name); \
725
#define CL_HPP_INIT_CL_EXT_FCN_PTR_PLATFORM_(platform, name) \
727
pfn_##name = (PFN_##name) \
728
clGetExtensionFunctionAddressForPlatform(platform, #name); \
737
class DeviceCommandQueue;
742
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
743
/*! \brief Exception class
745
* This may be thrown by API functions when CL_HPP_ENABLE_EXCEPTIONS is defined.
747
class Error : public std::exception
751
const char * errStr_;
753
/*! \brief Create a new CL error exception for a given error code
754
* and corresponding message.
756
* \param err error code value.
758
* \param errStr a descriptive string that must remain in scope until
759
* handling of the exception has concluded. If set, it
760
* will be returned by what().
762
Error(cl_int err, const char * errStr = NULL) : err_(err), errStr_(errStr)
767
/*! \brief Get error string associated with exception
769
* \return A memory pointer to the error message string.
771
virtual const char * what() const throw ()
773
if (errStr_ == NULL) {
781
/*! \brief Get error code associated with exception
783
* \return The error code.
785
cl_int err(void) const { return err_; }
787
#define CL_HPP_ERR_STR_(x) #x
789
#define CL_HPP_ERR_STR_(x) NULL
790
#endif // CL_HPP_ENABLE_EXCEPTIONS
795
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
796
static inline cl_int errHandler (
798
const char * errStr = NULL)
800
if (err != CL_SUCCESS) {
801
throw Error(err, errStr);
806
static inline cl_int errHandler (cl_int err, const char * errStr = NULL)
808
(void) errStr; // suppress unused variable warning
811
#endif // CL_HPP_ENABLE_EXCEPTIONS
816
//! \cond DOXYGEN_DETAIL
817
#if !defined(CL_HPP_USER_OVERRIDE_ERROR_STRINGS)
818
#define __GET_DEVICE_INFO_ERR CL_HPP_ERR_STR_(clGetDeviceInfo)
819
#define __GET_PLATFORM_INFO_ERR CL_HPP_ERR_STR_(clGetPlatformInfo)
820
#define __GET_DEVICE_IDS_ERR CL_HPP_ERR_STR_(clGetDeviceIDs)
821
#define __GET_PLATFORM_IDS_ERR CL_HPP_ERR_STR_(clGetPlatformIDs)
822
#define __GET_CONTEXT_INFO_ERR CL_HPP_ERR_STR_(clGetContextInfo)
823
#define __GET_EVENT_INFO_ERR CL_HPP_ERR_STR_(clGetEventInfo)
824
#define __GET_EVENT_PROFILE_INFO_ERR CL_HPP_ERR_STR_(clGetEventProfileInfo)
825
#define __GET_MEM_OBJECT_INFO_ERR CL_HPP_ERR_STR_(clGetMemObjectInfo)
826
#define __GET_IMAGE_INFO_ERR CL_HPP_ERR_STR_(clGetImageInfo)
827
#define __GET_SAMPLER_INFO_ERR CL_HPP_ERR_STR_(clGetSamplerInfo)
828
#define __GET_KERNEL_INFO_ERR CL_HPP_ERR_STR_(clGetKernelInfo)
829
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
830
#define __GET_KERNEL_ARG_INFO_ERR CL_HPP_ERR_STR_(clGetKernelArgInfo)
831
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
832
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
833
#define __GET_KERNEL_SUB_GROUP_INFO_ERR CL_HPP_ERR_STR_(clGetKernelSubGroupInfo)
834
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
835
#define __GET_KERNEL_WORK_GROUP_INFO_ERR CL_HPP_ERR_STR_(clGetKernelWorkGroupInfo)
836
#define __GET_PROGRAM_INFO_ERR CL_HPP_ERR_STR_(clGetProgramInfo)
837
#define __GET_PROGRAM_BUILD_INFO_ERR CL_HPP_ERR_STR_(clGetProgramBuildInfo)
838
#define __GET_COMMAND_QUEUE_INFO_ERR CL_HPP_ERR_STR_(clGetCommandQueueInfo)
840
#define __CREATE_CONTEXT_ERR CL_HPP_ERR_STR_(clCreateContext)
841
#define __CREATE_CONTEXT_FROM_TYPE_ERR CL_HPP_ERR_STR_(clCreateContextFromType)
842
#define __GET_SUPPORTED_IMAGE_FORMATS_ERR CL_HPP_ERR_STR_(clGetSupportedImageFormats)
844
#define __CREATE_BUFFER_ERR CL_HPP_ERR_STR_(clCreateBuffer)
845
#define __COPY_ERR CL_HPP_ERR_STR_(cl::copy)
846
#define __CREATE_SUBBUFFER_ERR CL_HPP_ERR_STR_(clCreateSubBuffer)
847
#define __CREATE_GL_BUFFER_ERR CL_HPP_ERR_STR_(clCreateFromGLBuffer)
848
#define __CREATE_GL_RENDER_BUFFER_ERR CL_HPP_ERR_STR_(clCreateFromGLBuffer)
849
#define __GET_GL_OBJECT_INFO_ERR CL_HPP_ERR_STR_(clGetGLObjectInfo)
850
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
851
#define __CREATE_IMAGE_ERR CL_HPP_ERR_STR_(clCreateImage)
852
#define __CREATE_GL_TEXTURE_ERR CL_HPP_ERR_STR_(clCreateFromGLTexture)
853
#define __IMAGE_DIMENSION_ERR CL_HPP_ERR_STR_(Incorrect image dimensions)
854
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
855
#define __SET_MEM_OBJECT_DESTRUCTOR_CALLBACK_ERR CL_HPP_ERR_STR_(clSetMemObjectDestructorCallback)
857
#define __CREATE_USER_EVENT_ERR CL_HPP_ERR_STR_(clCreateUserEvent)
858
#define __SET_USER_EVENT_STATUS_ERR CL_HPP_ERR_STR_(clSetUserEventStatus)
859
#define __SET_EVENT_CALLBACK_ERR CL_HPP_ERR_STR_(clSetEventCallback)
860
#define __WAIT_FOR_EVENTS_ERR CL_HPP_ERR_STR_(clWaitForEvents)
862
#define __CREATE_KERNEL_ERR CL_HPP_ERR_STR_(clCreateKernel)
863
#define __SET_KERNEL_ARGS_ERR CL_HPP_ERR_STR_(clSetKernelArg)
864
#define __CREATE_PROGRAM_WITH_SOURCE_ERR CL_HPP_ERR_STR_(clCreateProgramWithSource)
865
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
866
#define __CREATE_PROGRAM_WITH_IL_ERR CL_HPP_ERR_STR_(clCreateProgramWithIL)
867
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
868
#define __CREATE_PROGRAM_WITH_BINARY_ERR CL_HPP_ERR_STR_(clCreateProgramWithBinary)
869
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
870
#define __CREATE_PROGRAM_WITH_IL_ERR CL_HPP_ERR_STR_(clCreateProgramWithIL)
871
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 210
872
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
873
#define __CREATE_PROGRAM_WITH_BUILT_IN_KERNELS_ERR CL_HPP_ERR_STR_(clCreateProgramWithBuiltInKernels)
874
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
875
#define __BUILD_PROGRAM_ERR CL_HPP_ERR_STR_(clBuildProgram)
876
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
877
#define __COMPILE_PROGRAM_ERR CL_HPP_ERR_STR_(clCompileProgram)
878
#define __LINK_PROGRAM_ERR CL_HPP_ERR_STR_(clLinkProgram)
879
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
880
#define __CREATE_KERNELS_IN_PROGRAM_ERR CL_HPP_ERR_STR_(clCreateKernelsInProgram)
882
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
883
#define __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR CL_HPP_ERR_STR_(clCreateCommandQueueWithProperties)
884
#define __CREATE_SAMPLER_WITH_PROPERTIES_ERR CL_HPP_ERR_STR_(clCreateSamplerWithProperties)
885
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
886
#define __SET_COMMAND_QUEUE_PROPERTY_ERR CL_HPP_ERR_STR_(clSetCommandQueueProperty)
887
#define __ENQUEUE_READ_BUFFER_ERR CL_HPP_ERR_STR_(clEnqueueReadBuffer)
888
#define __ENQUEUE_READ_BUFFER_RECT_ERR CL_HPP_ERR_STR_(clEnqueueReadBufferRect)
889
#define __ENQUEUE_WRITE_BUFFER_ERR CL_HPP_ERR_STR_(clEnqueueWriteBuffer)
890
#define __ENQUEUE_WRITE_BUFFER_RECT_ERR CL_HPP_ERR_STR_(clEnqueueWriteBufferRect)
891
#define __ENQEUE_COPY_BUFFER_ERR CL_HPP_ERR_STR_(clEnqueueCopyBuffer)
892
#define __ENQEUE_COPY_BUFFER_RECT_ERR CL_HPP_ERR_STR_(clEnqueueCopyBufferRect)
893
#define __ENQUEUE_FILL_BUFFER_ERR CL_HPP_ERR_STR_(clEnqueueFillBuffer)
894
#define __ENQUEUE_READ_IMAGE_ERR CL_HPP_ERR_STR_(clEnqueueReadImage)
895
#define __ENQUEUE_WRITE_IMAGE_ERR CL_HPP_ERR_STR_(clEnqueueWriteImage)
896
#define __ENQUEUE_COPY_IMAGE_ERR CL_HPP_ERR_STR_(clEnqueueCopyImage)
897
#define __ENQUEUE_FILL_IMAGE_ERR CL_HPP_ERR_STR_(clEnqueueFillImage)
898
#define __ENQUEUE_COPY_IMAGE_TO_BUFFER_ERR CL_HPP_ERR_STR_(clEnqueueCopyImageToBuffer)
899
#define __ENQUEUE_COPY_BUFFER_TO_IMAGE_ERR CL_HPP_ERR_STR_(clEnqueueCopyBufferToImage)
900
#define __ENQUEUE_MAP_BUFFER_ERR CL_HPP_ERR_STR_(clEnqueueMapBuffer)
901
#define __ENQUEUE_MAP_IMAGE_ERR CL_HPP_ERR_STR_(clEnqueueMapImage)
902
#define __ENQUEUE_UNMAP_MEM_OBJECT_ERR CL_HPP_ERR_STR_(clEnqueueUnMapMemObject)
903
#define __ENQUEUE_NDRANGE_KERNEL_ERR CL_HPP_ERR_STR_(clEnqueueNDRangeKernel)
904
#define __ENQUEUE_NATIVE_KERNEL CL_HPP_ERR_STR_(clEnqueueNativeKernel)
905
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
906
#define __ENQUEUE_MIGRATE_MEM_OBJECTS_ERR CL_HPP_ERR_STR_(clEnqueueMigrateMemObjects)
907
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
908
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
909
#define __ENQUEUE_MIGRATE_SVM_ERR CL_HPP_ERR_STR_(clEnqueueSVMMigrateMem)
910
#define __SET_DEFAULT_DEVICE_COMMAND_QUEUE_ERR CL_HPP_ERR_STR_(clSetDefaultDeviceCommandQueue)
911
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 210
914
#define __ENQUEUE_ACQUIRE_GL_ERR CL_HPP_ERR_STR_(clEnqueueAcquireGLObjects)
915
#define __ENQUEUE_RELEASE_GL_ERR CL_HPP_ERR_STR_(clEnqueueReleaseGLObjects)
917
#define __CREATE_PIPE_ERR CL_HPP_ERR_STR_(clCreatePipe)
918
#define __GET_PIPE_INFO_ERR CL_HPP_ERR_STR_(clGetPipeInfo)
921
#define __RETAIN_ERR CL_HPP_ERR_STR_(Retain Object)
922
#define __RELEASE_ERR CL_HPP_ERR_STR_(Release Object)
923
#define __FLUSH_ERR CL_HPP_ERR_STR_(clFlush)
924
#define __FINISH_ERR CL_HPP_ERR_STR_(clFinish)
925
#define __VECTOR_CAPACITY_ERR CL_HPP_ERR_STR_(Vector capacity error)
927
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
928
#define __GET_HOST_TIMER_ERR CL_HPP_ERR_STR_(clGetHostTimer)
929
#define __GET_DEVICE_AND_HOST_TIMER_ERR CL_HPP_ERR_STR_(clGetDeviceAndHostTimer)
931
#if CL_HPP_TARGET_OPENCL_VERSION >= 220
932
#define __SET_PROGRAM_RELEASE_CALLBACK_ERR CL_HPP_ERR_STR_(clSetProgramReleaseCallback)
933
#define __SET_PROGRAM_SPECIALIZATION_CONSTANT_ERR CL_HPP_ERR_STR_(clSetProgramSpecializationConstant)
938
* CL 1.2 version that uses device fission.
940
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
941
#define __CREATE_SUB_DEVICES_ERR CL_HPP_ERR_STR_(clCreateSubDevices)
943
#define __CREATE_SUB_DEVICES_ERR CL_HPP_ERR_STR_(clCreateSubDevicesEXT)
944
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
947
* Deprecated APIs for 1.2
949
#if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
950
#define __ENQUEUE_MARKER_ERR CL_HPP_ERR_STR_(clEnqueueMarker)
951
#define __ENQUEUE_WAIT_FOR_EVENTS_ERR CL_HPP_ERR_STR_(clEnqueueWaitForEvents)
952
#define __ENQUEUE_BARRIER_ERR CL_HPP_ERR_STR_(clEnqueueBarrier)
953
#define __UNLOAD_COMPILER_ERR CL_HPP_ERR_STR_(clUnloadCompiler)
954
#define __CREATE_GL_TEXTURE_2D_ERR CL_HPP_ERR_STR_(clCreateFromGLTexture2D)
955
#define __CREATE_GL_TEXTURE_3D_ERR CL_HPP_ERR_STR_(clCreateFromGLTexture3D)
956
#define __CREATE_IMAGE2D_ERR CL_HPP_ERR_STR_(clCreateImage2D)
957
#define __CREATE_IMAGE3D_ERR CL_HPP_ERR_STR_(clCreateImage3D)
958
#endif // #if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
961
* Deprecated APIs for 2.0
963
#if defined(CL_USE_DEPRECATED_OPENCL_1_2_APIS)
964
#define __CREATE_COMMAND_QUEUE_ERR CL_HPP_ERR_STR_(clCreateCommandQueue)
965
#define __ENQUEUE_TASK_ERR CL_HPP_ERR_STR_(clEnqueueTask)
966
#define __CREATE_SAMPLER_ERR CL_HPP_ERR_STR_(clCreateSampler)
967
#endif // #if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
970
* CL 1.2 marker and barrier commands
972
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
973
#define __ENQUEUE_MARKER_WAIT_LIST_ERR CL_HPP_ERR_STR_(clEnqueueMarkerWithWaitList)
974
#define __ENQUEUE_BARRIER_WAIT_LIST_ERR CL_HPP_ERR_STR_(clEnqueueBarrierWithWaitList)
975
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
977
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
978
#define __CLONE_KERNEL_ERR CL_HPP_ERR_STR_(clCloneKernel)
979
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 210
981
#endif // CL_HPP_USER_OVERRIDE_ERROR_STRINGS
987
// Generic getInfoHelper. The final parameter is used to guide overload
988
// resolution: the actual parameter passed is an int, which makes this
989
// a worse conversion sequence than a specialization that declares the
990
// parameter as an int.
991
template<typename Functor, typename T>
992
inline cl_int getInfoHelper(Functor f, cl_uint name, T* param, long)
994
return f(name, sizeof(T), param, NULL);
997
// Specialized for getInfo<CL_PROGRAM_BINARIES>
998
// Assumes that the output vector was correctly resized on the way in
999
template <typename Func>
1000
inline cl_int getInfoHelper(Func f, cl_uint name, vector<vector<unsigned char>>* param, int)
1002
if (name != CL_PROGRAM_BINARIES) {
1003
return CL_INVALID_VALUE;
1006
// Create array of pointers, calculate total size and pass pointer array in
1007
size_type numBinaries = param->size();
1008
vector<unsigned char*> binariesPointers(numBinaries);
1010
for (size_type i = 0; i < numBinaries; ++i)
1012
binariesPointers[i] = (*param)[i].data();
1015
cl_int err = f(name, numBinaries * sizeof(unsigned char*), binariesPointers.data(), NULL);
1017
if (err != CL_SUCCESS) {
1026
// Specialized getInfoHelper for vector params
1027
template <typename Func, typename T>
1028
inline cl_int getInfoHelper(Func f, cl_uint name, vector<T>* param, long)
1031
cl_int err = f(name, 0, NULL, &required);
1032
if (err != CL_SUCCESS) {
1035
const size_type elements = required / sizeof(T);
1037
// Temporary to avoid changing param on an error
1038
vector<T> localData(elements);
1039
err = f(name, required, localData.data(), NULL);
1040
if (err != CL_SUCCESS) {
1044
*param = std::move(localData);
1050
/* Specialization for reference-counted types. This depends on the
1051
* existence of Wrapper<T>::cl_type, and none of the other types having the
1052
* cl_type member. Note that simplify specifying the parameter as Wrapper<T>
1053
* does not work, because when using a derived type (e.g. Context) the generic
1054
* template will provide a better match.
1056
template <typename Func, typename T>
1057
inline cl_int getInfoHelper(
1058
Func f, cl_uint name, vector<T>* param, int, typename T::cl_type = 0)
1061
cl_int err = f(name, 0, NULL, &required);
1062
if (err != CL_SUCCESS) {
1066
const size_type elements = required / sizeof(typename T::cl_type);
1068
vector<typename T::cl_type> value(elements);
1069
err = f(name, required, value.data(), NULL);
1070
if (err != CL_SUCCESS) {
1075
// Assign to convert CL type to T for each element
1076
param->resize(elements);
1078
// Assign to param, constructing with retain behaviour
1079
// to correctly capture each underlying CL object
1080
for (size_type i = 0; i < elements; i++) {
1081
(*param)[i] = T(value[i], true);
1087
// Specialized GetInfoHelper for string params
1088
template <typename Func>
1089
inline cl_int getInfoHelper(Func f, cl_uint name, string* param, long)
1092
cl_int err = f(name, 0, NULL, &required);
1093
if (err != CL_SUCCESS) {
1097
// std::string has a constant data member
1098
// a char vector does not
1100
vector<char> value(required);
1101
err = f(name, required, value.data(), NULL);
1102
if (err != CL_SUCCESS) {
1106
param->assign(begin(value), prev(end(value)));
1115
// Specialized GetInfoHelper for clsize_t params
1116
template <typename Func, size_type N>
1117
inline cl_int getInfoHelper(Func f, cl_uint name, array<size_type, N>* param, long)
1120
cl_int err = f(name, 0, NULL, &required);
1121
if (err != CL_SUCCESS) {
1125
size_type elements = required / sizeof(size_type);
1126
vector<size_type> value(elements, 0);
1128
err = f(name, required, value.data(), NULL);
1129
if (err != CL_SUCCESS) {
1133
// Bound the copy with N to prevent overruns
1134
// if passed N > than the amount copied
1138
for (size_type i = 0; i < elements; ++i) {
1139
(*param)[i] = value[i];
1145
template<typename T> struct ReferenceHandler;
1147
/* Specialization for reference-counted types. This depends on the
1148
* existence of Wrapper<T>::cl_type, and none of the other types having the
1149
* cl_type member. Note that simplify specifying the parameter as Wrapper<T>
1150
* does not work, because when using a derived type (e.g. Context) the generic
1151
* template will provide a better match.
1153
template<typename Func, typename T>
1154
inline cl_int getInfoHelper(Func f, cl_uint name, T* param, int, typename T::cl_type = 0)
1156
typename T::cl_type value;
1157
cl_int err = f(name, sizeof(value), &value, NULL);
1158
if (err != CL_SUCCESS) {
1164
err = param->retain();
1165
if (err != CL_SUCCESS) {
1172
#define CL_HPP_PARAM_NAME_INFO_1_0_(F) \
1173
F(cl_platform_info, CL_PLATFORM_PROFILE, string) \
1174
F(cl_platform_info, CL_PLATFORM_VERSION, string) \
1175
F(cl_platform_info, CL_PLATFORM_NAME, string) \
1176
F(cl_platform_info, CL_PLATFORM_VENDOR, string) \
1177
F(cl_platform_info, CL_PLATFORM_EXTENSIONS, string) \
1179
F(cl_device_info, CL_DEVICE_TYPE, cl_device_type) \
1180
F(cl_device_info, CL_DEVICE_VENDOR_ID, cl_uint) \
1181
F(cl_device_info, CL_DEVICE_MAX_COMPUTE_UNITS, cl_uint) \
1182
F(cl_device_info, CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, cl_uint) \
1183
F(cl_device_info, CL_DEVICE_MAX_WORK_GROUP_SIZE, size_type) \
1184
F(cl_device_info, CL_DEVICE_MAX_WORK_ITEM_SIZES, cl::vector<size_type>) \
1185
F(cl_device_info, CL_DEVICE_PREFERRED_VECTOR_WIDTH_CHAR, cl_uint) \
1186
F(cl_device_info, CL_DEVICE_PREFERRED_VECTOR_WIDTH_SHORT, cl_uint) \
1187
F(cl_device_info, CL_DEVICE_PREFERRED_VECTOR_WIDTH_INT, cl_uint) \
1188
F(cl_device_info, CL_DEVICE_PREFERRED_VECTOR_WIDTH_LONG, cl_uint) \
1189
F(cl_device_info, CL_DEVICE_PREFERRED_VECTOR_WIDTH_FLOAT, cl_uint) \
1190
F(cl_device_info, CL_DEVICE_PREFERRED_VECTOR_WIDTH_DOUBLE, cl_uint) \
1191
F(cl_device_info, CL_DEVICE_MAX_CLOCK_FREQUENCY, cl_uint) \
1192
F(cl_device_info, CL_DEVICE_ADDRESS_BITS, cl_uint) \
1193
F(cl_device_info, CL_DEVICE_MAX_READ_IMAGE_ARGS, cl_uint) \
1194
F(cl_device_info, CL_DEVICE_MAX_WRITE_IMAGE_ARGS, cl_uint) \
1195
F(cl_device_info, CL_DEVICE_MAX_MEM_ALLOC_SIZE, cl_ulong) \
1196
F(cl_device_info, CL_DEVICE_IMAGE2D_MAX_WIDTH, size_type) \
1197
F(cl_device_info, CL_DEVICE_IMAGE2D_MAX_HEIGHT, size_type) \
1198
F(cl_device_info, CL_DEVICE_IMAGE3D_MAX_WIDTH, size_type) \
1199
F(cl_device_info, CL_DEVICE_IMAGE3D_MAX_HEIGHT, size_type) \
1200
F(cl_device_info, CL_DEVICE_IMAGE3D_MAX_DEPTH, size_type) \
1201
F(cl_device_info, CL_DEVICE_IMAGE_SUPPORT, cl_bool) \
1202
F(cl_device_info, CL_DEVICE_MAX_PARAMETER_SIZE, size_type) \
1203
F(cl_device_info, CL_DEVICE_MAX_SAMPLERS, cl_uint) \
1204
F(cl_device_info, CL_DEVICE_MEM_BASE_ADDR_ALIGN, cl_uint) \
1205
F(cl_device_info, CL_DEVICE_MIN_DATA_TYPE_ALIGN_SIZE, cl_uint) \
1206
F(cl_device_info, CL_DEVICE_SINGLE_FP_CONFIG, cl_device_fp_config) \
1207
F(cl_device_info, CL_DEVICE_DOUBLE_FP_CONFIG, cl_device_fp_config) \
1208
F(cl_device_info, CL_DEVICE_HALF_FP_CONFIG, cl_device_fp_config) \
1209
F(cl_device_info, CL_DEVICE_GLOBAL_MEM_CACHE_TYPE, cl_device_mem_cache_type) \
1210
F(cl_device_info, CL_DEVICE_GLOBAL_MEM_CACHELINE_SIZE, cl_uint)\
1211
F(cl_device_info, CL_DEVICE_GLOBAL_MEM_CACHE_SIZE, cl_ulong) \
1212
F(cl_device_info, CL_DEVICE_GLOBAL_MEM_SIZE, cl_ulong) \
1213
F(cl_device_info, CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE, cl_ulong) \
1214
F(cl_device_info, CL_DEVICE_MAX_CONSTANT_ARGS, cl_uint) \
1215
F(cl_device_info, CL_DEVICE_LOCAL_MEM_TYPE, cl_device_local_mem_type) \
1216
F(cl_device_info, CL_DEVICE_LOCAL_MEM_SIZE, cl_ulong) \
1217
F(cl_device_info, CL_DEVICE_ERROR_CORRECTION_SUPPORT, cl_bool) \
1218
F(cl_device_info, CL_DEVICE_PROFILING_TIMER_RESOLUTION, size_type) \
1219
F(cl_device_info, CL_DEVICE_ENDIAN_LITTLE, cl_bool) \
1220
F(cl_device_info, CL_DEVICE_AVAILABLE, cl_bool) \
1221
F(cl_device_info, CL_DEVICE_COMPILER_AVAILABLE, cl_bool) \
1222
F(cl_device_info, CL_DEVICE_EXECUTION_CAPABILITIES, cl_device_exec_capabilities) \
1223
F(cl_device_info, CL_DEVICE_PLATFORM, cl_platform_id) \
1224
F(cl_device_info, CL_DEVICE_NAME, string) \
1225
F(cl_device_info, CL_DEVICE_VENDOR, string) \
1226
F(cl_device_info, CL_DRIVER_VERSION, string) \
1227
F(cl_device_info, CL_DEVICE_PROFILE, string) \
1228
F(cl_device_info, CL_DEVICE_VERSION, string) \
1229
F(cl_device_info, CL_DEVICE_EXTENSIONS, string) \
1231
F(cl_context_info, CL_CONTEXT_REFERENCE_COUNT, cl_uint) \
1232
F(cl_context_info, CL_CONTEXT_DEVICES, cl::vector<Device>) \
1233
F(cl_context_info, CL_CONTEXT_PROPERTIES, cl::vector<cl_context_properties>) \
1235
F(cl_event_info, CL_EVENT_COMMAND_QUEUE, cl::CommandQueue) \
1236
F(cl_event_info, CL_EVENT_COMMAND_TYPE, cl_command_type) \
1237
F(cl_event_info, CL_EVENT_REFERENCE_COUNT, cl_uint) \
1238
F(cl_event_info, CL_EVENT_COMMAND_EXECUTION_STATUS, cl_int) \
1240
F(cl_profiling_info, CL_PROFILING_COMMAND_QUEUED, cl_ulong) \
1241
F(cl_profiling_info, CL_PROFILING_COMMAND_SUBMIT, cl_ulong) \
1242
F(cl_profiling_info, CL_PROFILING_COMMAND_START, cl_ulong) \
1243
F(cl_profiling_info, CL_PROFILING_COMMAND_END, cl_ulong) \
1245
F(cl_mem_info, CL_MEM_TYPE, cl_mem_object_type) \
1246
F(cl_mem_info, CL_MEM_FLAGS, cl_mem_flags) \
1247
F(cl_mem_info, CL_MEM_SIZE, size_type) \
1248
F(cl_mem_info, CL_MEM_HOST_PTR, void*) \
1249
F(cl_mem_info, CL_MEM_MAP_COUNT, cl_uint) \
1250
F(cl_mem_info, CL_MEM_REFERENCE_COUNT, cl_uint) \
1251
F(cl_mem_info, CL_MEM_CONTEXT, cl::Context) \
1253
F(cl_image_info, CL_IMAGE_FORMAT, cl_image_format) \
1254
F(cl_image_info, CL_IMAGE_ELEMENT_SIZE, size_type) \
1255
F(cl_image_info, CL_IMAGE_ROW_PITCH, size_type) \
1256
F(cl_image_info, CL_IMAGE_SLICE_PITCH, size_type) \
1257
F(cl_image_info, CL_IMAGE_WIDTH, size_type) \
1258
F(cl_image_info, CL_IMAGE_HEIGHT, size_type) \
1259
F(cl_image_info, CL_IMAGE_DEPTH, size_type) \
1261
F(cl_sampler_info, CL_SAMPLER_REFERENCE_COUNT, cl_uint) \
1262
F(cl_sampler_info, CL_SAMPLER_CONTEXT, cl::Context) \
1263
F(cl_sampler_info, CL_SAMPLER_NORMALIZED_COORDS, cl_bool) \
1264
F(cl_sampler_info, CL_SAMPLER_ADDRESSING_MODE, cl_addressing_mode) \
1265
F(cl_sampler_info, CL_SAMPLER_FILTER_MODE, cl_filter_mode) \
1267
F(cl_program_info, CL_PROGRAM_REFERENCE_COUNT, cl_uint) \
1268
F(cl_program_info, CL_PROGRAM_CONTEXT, cl::Context) \
1269
F(cl_program_info, CL_PROGRAM_NUM_DEVICES, cl_uint) \
1270
F(cl_program_info, CL_PROGRAM_DEVICES, cl::vector<Device>) \
1271
F(cl_program_info, CL_PROGRAM_SOURCE, string) \
1272
F(cl_program_info, CL_PROGRAM_BINARY_SIZES, cl::vector<size_type>) \
1273
F(cl_program_info, CL_PROGRAM_BINARIES, cl::vector<cl::vector<unsigned char>>) \
1275
F(cl_program_build_info, CL_PROGRAM_BUILD_STATUS, cl_build_status) \
1276
F(cl_program_build_info, CL_PROGRAM_BUILD_OPTIONS, string) \
1277
F(cl_program_build_info, CL_PROGRAM_BUILD_LOG, string) \
1279
F(cl_kernel_info, CL_KERNEL_FUNCTION_NAME, string) \
1280
F(cl_kernel_info, CL_KERNEL_NUM_ARGS, cl_uint) \
1281
F(cl_kernel_info, CL_KERNEL_REFERENCE_COUNT, cl_uint) \
1282
F(cl_kernel_info, CL_KERNEL_CONTEXT, cl::Context) \
1283
F(cl_kernel_info, CL_KERNEL_PROGRAM, cl::Program) \
1285
F(cl_kernel_work_group_info, CL_KERNEL_WORK_GROUP_SIZE, size_type) \
1286
F(cl_kernel_work_group_info, CL_KERNEL_COMPILE_WORK_GROUP_SIZE, cl::detail::size_t_array) \
1287
F(cl_kernel_work_group_info, CL_KERNEL_LOCAL_MEM_SIZE, cl_ulong) \
1289
F(cl_command_queue_info, CL_QUEUE_CONTEXT, cl::Context) \
1290
F(cl_command_queue_info, CL_QUEUE_DEVICE, cl::Device) \
1291
F(cl_command_queue_info, CL_QUEUE_REFERENCE_COUNT, cl_uint) \
1292
F(cl_command_queue_info, CL_QUEUE_PROPERTIES, cl_command_queue_properties)
1295
#define CL_HPP_PARAM_NAME_INFO_1_1_(F) \
1296
F(cl_context_info, CL_CONTEXT_NUM_DEVICES, cl_uint)\
1297
F(cl_device_info, CL_DEVICE_PREFERRED_VECTOR_WIDTH_HALF, cl_uint) \
1298
F(cl_device_info, CL_DEVICE_NATIVE_VECTOR_WIDTH_CHAR, cl_uint) \
1299
F(cl_device_info, CL_DEVICE_NATIVE_VECTOR_WIDTH_SHORT, cl_uint) \
1300
F(cl_device_info, CL_DEVICE_NATIVE_VECTOR_WIDTH_INT, cl_uint) \
1301
F(cl_device_info, CL_DEVICE_NATIVE_VECTOR_WIDTH_LONG, cl_uint) \
1302
F(cl_device_info, CL_DEVICE_NATIVE_VECTOR_WIDTH_FLOAT, cl_uint) \
1303
F(cl_device_info, CL_DEVICE_NATIVE_VECTOR_WIDTH_DOUBLE, cl_uint) \
1304
F(cl_device_info, CL_DEVICE_NATIVE_VECTOR_WIDTH_HALF, cl_uint) \
1305
F(cl_device_info, CL_DEVICE_OPENCL_C_VERSION, string) \
1307
F(cl_mem_info, CL_MEM_ASSOCIATED_MEMOBJECT, cl::Memory) \
1308
F(cl_mem_info, CL_MEM_OFFSET, size_type) \
1310
F(cl_kernel_work_group_info, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, size_type) \
1311
F(cl_kernel_work_group_info, CL_KERNEL_PRIVATE_MEM_SIZE, cl_ulong) \
1313
F(cl_event_info, CL_EVENT_CONTEXT, cl::Context)
1315
#define CL_HPP_PARAM_NAME_INFO_1_2_(F) \
1316
F(cl_program_info, CL_PROGRAM_NUM_KERNELS, size_type) \
1317
F(cl_program_info, CL_PROGRAM_KERNEL_NAMES, string) \
1319
F(cl_program_build_info, CL_PROGRAM_BINARY_TYPE, cl_program_binary_type) \
1321
F(cl_kernel_info, CL_KERNEL_ATTRIBUTES, string) \
1323
F(cl_kernel_arg_info, CL_KERNEL_ARG_ADDRESS_QUALIFIER, cl_kernel_arg_address_qualifier) \
1324
F(cl_kernel_arg_info, CL_KERNEL_ARG_ACCESS_QUALIFIER, cl_kernel_arg_access_qualifier) \
1325
F(cl_kernel_arg_info, CL_KERNEL_ARG_TYPE_NAME, string) \
1326
F(cl_kernel_arg_info, CL_KERNEL_ARG_NAME, string) \
1327
F(cl_kernel_arg_info, CL_KERNEL_ARG_TYPE_QUALIFIER, cl_kernel_arg_type_qualifier) \
1329
F(cl_device_info, CL_DEVICE_PARENT_DEVICE, cl::Device) \
1330
F(cl_device_info, CL_DEVICE_PARTITION_PROPERTIES, cl::vector<cl_device_partition_property>) \
1331
F(cl_device_info, CL_DEVICE_PARTITION_TYPE, cl::vector<cl_device_partition_property>) \
1332
F(cl_device_info, CL_DEVICE_REFERENCE_COUNT, cl_uint) \
1333
F(cl_device_info, CL_DEVICE_PREFERRED_INTEROP_USER_SYNC, size_type) \
1334
F(cl_device_info, CL_DEVICE_PARTITION_AFFINITY_DOMAIN, cl_device_affinity_domain) \
1335
F(cl_device_info, CL_DEVICE_BUILT_IN_KERNELS, string) \
1337
F(cl_image_info, CL_IMAGE_ARRAY_SIZE, size_type) \
1338
F(cl_image_info, CL_IMAGE_NUM_MIP_LEVELS, cl_uint) \
1339
F(cl_image_info, CL_IMAGE_NUM_SAMPLES, cl_uint)
1341
#define CL_HPP_PARAM_NAME_INFO_2_0_(F) \
1342
F(cl_device_info, CL_DEVICE_QUEUE_ON_HOST_PROPERTIES, cl_command_queue_properties) \
1343
F(cl_device_info, CL_DEVICE_QUEUE_ON_DEVICE_PROPERTIES, cl_command_queue_properties) \
1344
F(cl_device_info, CL_DEVICE_QUEUE_ON_DEVICE_PREFERRED_SIZE, cl_uint) \
1345
F(cl_device_info, CL_DEVICE_QUEUE_ON_DEVICE_MAX_SIZE, cl_uint) \
1346
F(cl_device_info, CL_DEVICE_MAX_ON_DEVICE_QUEUES, cl_uint) \
1347
F(cl_device_info, CL_DEVICE_MAX_ON_DEVICE_EVENTS, cl_uint) \
1348
F(cl_device_info, CL_DEVICE_MAX_PIPE_ARGS, cl_uint) \
1349
F(cl_device_info, CL_DEVICE_PIPE_MAX_ACTIVE_RESERVATIONS, cl_uint) \
1350
F(cl_device_info, CL_DEVICE_PIPE_MAX_PACKET_SIZE, cl_uint) \
1351
F(cl_device_info, CL_DEVICE_SVM_CAPABILITIES, cl_device_svm_capabilities) \
1352
F(cl_device_info, CL_DEVICE_PREFERRED_PLATFORM_ATOMIC_ALIGNMENT, cl_uint) \
1353
F(cl_device_info, CL_DEVICE_PREFERRED_GLOBAL_ATOMIC_ALIGNMENT, cl_uint) \
1354
F(cl_device_info, CL_DEVICE_PREFERRED_LOCAL_ATOMIC_ALIGNMENT, cl_uint) \
1355
F(cl_command_queue_info, CL_QUEUE_SIZE, cl_uint) \
1356
F(cl_mem_info, CL_MEM_USES_SVM_POINTER, cl_bool) \
1357
F(cl_program_build_info, CL_PROGRAM_BUILD_GLOBAL_VARIABLE_TOTAL_SIZE, size_type) \
1358
F(cl_pipe_info, CL_PIPE_PACKET_SIZE, cl_uint) \
1359
F(cl_pipe_info, CL_PIPE_MAX_PACKETS, cl_uint)
1361
#define CL_HPP_PARAM_NAME_INFO_SUBGROUP_KHR_(F) \
1362
F(cl_kernel_sub_group_info, CL_KERNEL_MAX_SUB_GROUP_SIZE_FOR_NDRANGE_KHR, size_type) \
1363
F(cl_kernel_sub_group_info, CL_KERNEL_SUB_GROUP_COUNT_FOR_NDRANGE_KHR, size_type)
1365
#define CL_HPP_PARAM_NAME_INFO_IL_KHR_(F) \
1366
F(cl_device_info, CL_DEVICE_IL_VERSION_KHR, string) \
1367
F(cl_program_info, CL_PROGRAM_IL_KHR, cl::vector<unsigned char>)
1369
#define CL_HPP_PARAM_NAME_INFO_2_1_(F) \
1370
F(cl_platform_info, CL_PLATFORM_HOST_TIMER_RESOLUTION, size_type) \
1371
F(cl_program_info, CL_PROGRAM_IL, cl::vector<unsigned char>) \
1372
F(cl_kernel_info, CL_KERNEL_MAX_NUM_SUB_GROUPS, size_type) \
1373
F(cl_kernel_info, CL_KERNEL_COMPILE_NUM_SUB_GROUPS, size_type) \
1374
F(cl_device_info, CL_DEVICE_MAX_NUM_SUB_GROUPS, cl_uint) \
1375
F(cl_device_info, CL_DEVICE_IL_VERSION, string) \
1376
F(cl_device_info, CL_DEVICE_SUB_GROUP_INDEPENDENT_FORWARD_PROGRESS, cl_bool) \
1377
F(cl_command_queue_info, CL_QUEUE_DEVICE_DEFAULT, cl::DeviceCommandQueue) \
1378
F(cl_kernel_sub_group_info, CL_KERNEL_MAX_SUB_GROUP_SIZE_FOR_NDRANGE, size_type) \
1379
F(cl_kernel_sub_group_info, CL_KERNEL_SUB_GROUP_COUNT_FOR_NDRANGE, size_type) \
1380
F(cl_kernel_sub_group_info, CL_KERNEL_LOCAL_SIZE_FOR_SUB_GROUP_COUNT, cl::detail::size_t_array)
1382
#define CL_HPP_PARAM_NAME_INFO_2_2_(F) \
1383
F(cl_program_info, CL_PROGRAM_SCOPE_GLOBAL_CTORS_PRESENT, cl_bool) \
1384
F(cl_program_info, CL_PROGRAM_SCOPE_GLOBAL_DTORS_PRESENT, cl_bool)
1386
#define CL_HPP_PARAM_NAME_DEVICE_FISSION_(F) \
1387
F(cl_device_info, CL_DEVICE_PARENT_DEVICE_EXT, cl_device_id) \
1388
F(cl_device_info, CL_DEVICE_PARTITION_TYPES_EXT, cl::vector<cl_device_partition_property_ext>) \
1389
F(cl_device_info, CL_DEVICE_AFFINITY_DOMAINS_EXT, cl::vector<cl_device_partition_property_ext>) \
1390
F(cl_device_info, CL_DEVICE_REFERENCE_COUNT_EXT , cl_uint) \
1391
F(cl_device_info, CL_DEVICE_PARTITION_STYLE_EXT, cl::vector<cl_device_partition_property_ext>)
1393
template <typename enum_type, cl_int Name>
1394
struct param_traits {};
1396
#define CL_HPP_DECLARE_PARAM_TRAITS_(token, param_name, T) \
1399
struct param_traits<detail:: token,param_name> \
1401
enum { value = param_name }; \
1402
typedef T param_type; \
1405
CL_HPP_PARAM_NAME_INFO_1_0_(CL_HPP_DECLARE_PARAM_TRAITS_)
1406
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
1407
CL_HPP_PARAM_NAME_INFO_1_1_(CL_HPP_DECLARE_PARAM_TRAITS_)
1408
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 110
1409
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
1410
CL_HPP_PARAM_NAME_INFO_1_2_(CL_HPP_DECLARE_PARAM_TRAITS_)
1411
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
1412
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
1413
CL_HPP_PARAM_NAME_INFO_2_0_(CL_HPP_DECLARE_PARAM_TRAITS_)
1414
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
1415
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
1416
CL_HPP_PARAM_NAME_INFO_2_1_(CL_HPP_DECLARE_PARAM_TRAITS_)
1417
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 210
1418
#if CL_HPP_TARGET_OPENCL_VERSION >= 220
1419
CL_HPP_PARAM_NAME_INFO_2_2_(CL_HPP_DECLARE_PARAM_TRAITS_)
1420
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 220
1422
#if defined(CL_HPP_USE_CL_SUB_GROUPS_KHR) && CL_HPP_TARGET_OPENCL_VERSION < 210
1423
CL_HPP_PARAM_NAME_INFO_SUBGROUP_KHR_(CL_HPP_DECLARE_PARAM_TRAITS_)
1424
#endif // #if defined(CL_HPP_USE_CL_SUB_GROUPS_KHR) && CL_HPP_TARGET_OPENCL_VERSION < 210
1426
#if defined(CL_HPP_USE_IL_KHR)
1427
CL_HPP_PARAM_NAME_INFO_IL_KHR_(CL_HPP_DECLARE_PARAM_TRAITS_)
1428
#endif // #if defined(CL_HPP_USE_IL_KHR)
1431
// Flags deprecated in OpenCL 2.0
1432
#define CL_HPP_PARAM_NAME_INFO_1_0_DEPRECATED_IN_2_0_(F) \
1433
F(cl_device_info, CL_DEVICE_QUEUE_PROPERTIES, cl_command_queue_properties)
1435
#define CL_HPP_PARAM_NAME_INFO_1_1_DEPRECATED_IN_2_0_(F) \
1436
F(cl_device_info, CL_DEVICE_HOST_UNIFIED_MEMORY, cl_bool)
1438
#define CL_HPP_PARAM_NAME_INFO_1_2_DEPRECATED_IN_2_0_(F) \
1439
F(cl_image_info, CL_IMAGE_BUFFER, cl::Buffer)
1441
// Include deprecated query flags based on versions
1442
// Only include deprecated 1.0 flags if 2.0 not active as there is an enum clash
1443
#if CL_HPP_TARGET_OPENCL_VERSION > 100 && CL_HPP_MINIMUM_OPENCL_VERSION < 200 && CL_HPP_TARGET_OPENCL_VERSION < 200
1444
CL_HPP_PARAM_NAME_INFO_1_0_DEPRECATED_IN_2_0_(CL_HPP_DECLARE_PARAM_TRAITS_)
1445
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 110
1446
#if CL_HPP_TARGET_OPENCL_VERSION > 110 && CL_HPP_MINIMUM_OPENCL_VERSION < 200
1447
CL_HPP_PARAM_NAME_INFO_1_1_DEPRECATED_IN_2_0_(CL_HPP_DECLARE_PARAM_TRAITS_)
1448
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 120
1449
#if CL_HPP_TARGET_OPENCL_VERSION > 120 && CL_HPP_MINIMUM_OPENCL_VERSION < 200
1450
CL_HPP_PARAM_NAME_INFO_1_2_DEPRECATED_IN_2_0_(CL_HPP_DECLARE_PARAM_TRAITS_)
1451
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 200
1453
#if defined(CL_HPP_USE_CL_DEVICE_FISSION)
1454
CL_HPP_PARAM_NAME_DEVICE_FISSION_(CL_HPP_DECLARE_PARAM_TRAITS_);
1455
#endif // CL_HPP_USE_CL_DEVICE_FISSION
1457
#ifdef CL_PLATFORM_ICD_SUFFIX_KHR
1458
CL_HPP_DECLARE_PARAM_TRAITS_(cl_platform_info, CL_PLATFORM_ICD_SUFFIX_KHR, string)
1461
#ifdef CL_DEVICE_PROFILING_TIMER_OFFSET_AMD
1462
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_PROFILING_TIMER_OFFSET_AMD, cl_ulong)
1465
#ifdef CL_DEVICE_GLOBAL_FREE_MEMORY_AMD
1466
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_GLOBAL_FREE_MEMORY_AMD, vector<size_type>)
1468
#ifdef CL_DEVICE_SIMD_PER_COMPUTE_UNIT_AMD
1469
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_SIMD_PER_COMPUTE_UNIT_AMD, cl_uint)
1471
#ifdef CL_DEVICE_SIMD_WIDTH_AMD
1472
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_SIMD_WIDTH_AMD, cl_uint)
1474
#ifdef CL_DEVICE_SIMD_INSTRUCTION_WIDTH_AMD
1475
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_SIMD_INSTRUCTION_WIDTH_AMD, cl_uint)
1477
#ifdef CL_DEVICE_WAVEFRONT_WIDTH_AMD
1478
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_WAVEFRONT_WIDTH_AMD, cl_uint)
1480
#ifdef CL_DEVICE_GLOBAL_MEM_CHANNELS_AMD
1481
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_GLOBAL_MEM_CHANNELS_AMD, cl_uint)
1483
#ifdef CL_DEVICE_GLOBAL_MEM_CHANNEL_BANKS_AMD
1484
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_GLOBAL_MEM_CHANNEL_BANKS_AMD, cl_uint)
1486
#ifdef CL_DEVICE_GLOBAL_MEM_CHANNEL_BANK_WIDTH_AMD
1487
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_GLOBAL_MEM_CHANNEL_BANK_WIDTH_AMD, cl_uint)
1489
#ifdef CL_DEVICE_LOCAL_MEM_SIZE_PER_COMPUTE_UNIT_AMD
1490
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_LOCAL_MEM_SIZE_PER_COMPUTE_UNIT_AMD, cl_uint)
1492
#ifdef CL_DEVICE_LOCAL_MEM_BANKS_AMD
1493
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_LOCAL_MEM_BANKS_AMD, cl_uint)
1496
#ifdef CL_DEVICE_COMPUTE_UNITS_BITFIELD_ARM
1497
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_COMPUTE_UNITS_BITFIELD_ARM, cl_ulong)
1499
#ifdef CL_DEVICE_JOB_SLOTS_ARM
1500
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_JOB_SLOTS_ARM, cl_uint)
1503
#ifdef CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV
1504
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_COMPUTE_CAPABILITY_MAJOR_NV, cl_uint)
1506
#ifdef CL_DEVICE_COMPUTE_CAPABILITY_MINOR_NV
1507
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_COMPUTE_CAPABILITY_MINOR_NV, cl_uint)
1509
#ifdef CL_DEVICE_REGISTERS_PER_BLOCK_NV
1510
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_REGISTERS_PER_BLOCK_NV, cl_uint)
1512
#ifdef CL_DEVICE_WARP_SIZE_NV
1513
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_WARP_SIZE_NV, cl_uint)
1515
#ifdef CL_DEVICE_GPU_OVERLAP_NV
1516
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_GPU_OVERLAP_NV, cl_bool)
1518
#ifdef CL_DEVICE_KERNEL_EXEC_TIMEOUT_NV
1519
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_KERNEL_EXEC_TIMEOUT_NV, cl_bool)
1521
#ifdef CL_DEVICE_INTEGRATED_MEMORY_NV
1522
CL_HPP_DECLARE_PARAM_TRAITS_(cl_device_info, CL_DEVICE_INTEGRATED_MEMORY_NV, cl_bool)
1525
// Convenience functions
1527
template <typename Func, typename T>
1529
getInfo(Func f, cl_uint name, T* param)
1531
return getInfoHelper(f, name, param, 0);
1534
template <typename Func, typename Arg0>
1535
struct GetInfoFunctor0
1537
Func f_; const Arg0& arg0_;
1539
cl_uint param, size_type size, void* value, size_type* size_ret)
1540
{ return f_(arg0_, param, size, value, size_ret); }
1543
template <typename Func, typename Arg0, typename Arg1>
1544
struct GetInfoFunctor1
1546
Func f_; const Arg0& arg0_; const Arg1& arg1_;
1548
cl_uint param, size_type size, void* value, size_type* size_ret)
1549
{ return f_(arg0_, arg1_, param, size, value, size_ret); }
1552
template <typename Func, typename Arg0, typename T>
1554
getInfo(Func f, const Arg0& arg0, cl_uint name, T* param)
1556
GetInfoFunctor0<Func, Arg0> f0 = { f, arg0 };
1557
return getInfoHelper(f0, name, param, 0);
1560
template <typename Func, typename Arg0, typename Arg1, typename T>
1562
getInfo(Func f, const Arg0& arg0, const Arg1& arg1, cl_uint name, T* param)
1564
GetInfoFunctor1<Func, Arg0, Arg1> f0 = { f, arg0, arg1 };
1565
return getInfoHelper(f0, name, param, 0);
1569
template<typename T>
1570
struct ReferenceHandler
1573
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
1575
* OpenCL 1.2 devices do have retain/release.
1578
struct ReferenceHandler<cl_device_id>
1581
* Retain the device.
1582
* \param device A valid device created using createSubDevices
1584
* CL_SUCCESS if the function executed successfully.
1585
* CL_INVALID_DEVICE if device was not a valid subdevice
1586
* CL_OUT_OF_RESOURCES
1587
* CL_OUT_OF_HOST_MEMORY
1589
static cl_int retain(cl_device_id device)
1590
{ return ::clRetainDevice(device); }
1592
* Retain the device.
1593
* \param device A valid device created using createSubDevices
1595
* CL_SUCCESS if the function executed successfully.
1596
* CL_INVALID_DEVICE if device was not a valid subdevice
1597
* CL_OUT_OF_RESOURCES
1598
* CL_OUT_OF_HOST_MEMORY
1600
static cl_int release(cl_device_id device)
1601
{ return ::clReleaseDevice(device); }
1603
#else // CL_HPP_TARGET_OPENCL_VERSION >= 120
1605
* OpenCL 1.1 devices do not have retain/release.
1608
struct ReferenceHandler<cl_device_id>
1610
// cl_device_id does not have retain().
1611
static cl_int retain(cl_device_id)
1612
{ return CL_SUCCESS; }
1613
// cl_device_id does not have release().
1614
static cl_int release(cl_device_id)
1615
{ return CL_SUCCESS; }
1617
#endif // ! (CL_HPP_TARGET_OPENCL_VERSION >= 120)
1620
struct ReferenceHandler<cl_platform_id>
1622
// cl_platform_id does not have retain().
1623
static cl_int retain(cl_platform_id)
1624
{ return CL_SUCCESS; }
1625
// cl_platform_id does not have release().
1626
static cl_int release(cl_platform_id)
1627
{ return CL_SUCCESS; }
1631
struct ReferenceHandler<cl_context>
1633
static cl_int retain(cl_context context)
1634
{ return ::clRetainContext(context); }
1635
static cl_int release(cl_context context)
1636
{ return ::clReleaseContext(context); }
1640
struct ReferenceHandler<cl_command_queue>
1642
static cl_int retain(cl_command_queue queue)
1643
{ return ::clRetainCommandQueue(queue); }
1644
static cl_int release(cl_command_queue queue)
1645
{ return ::clReleaseCommandQueue(queue); }
1649
struct ReferenceHandler<cl_mem>
1651
static cl_int retain(cl_mem memory)
1652
{ return ::clRetainMemObject(memory); }
1653
static cl_int release(cl_mem memory)
1654
{ return ::clReleaseMemObject(memory); }
1658
struct ReferenceHandler<cl_sampler>
1660
static cl_int retain(cl_sampler sampler)
1661
{ return ::clRetainSampler(sampler); }
1662
static cl_int release(cl_sampler sampler)
1663
{ return ::clReleaseSampler(sampler); }
1667
struct ReferenceHandler<cl_program>
1669
static cl_int retain(cl_program program)
1670
{ return ::clRetainProgram(program); }
1671
static cl_int release(cl_program program)
1672
{ return ::clReleaseProgram(program); }
1676
struct ReferenceHandler<cl_kernel>
1678
static cl_int retain(cl_kernel kernel)
1679
{ return ::clRetainKernel(kernel); }
1680
static cl_int release(cl_kernel kernel)
1681
{ return ::clReleaseKernel(kernel); }
1685
struct ReferenceHandler<cl_event>
1687
static cl_int retain(cl_event event)
1688
{ return ::clRetainEvent(event); }
1689
static cl_int release(cl_event event)
1690
{ return ::clReleaseEvent(event); }
1694
#if CL_HPP_TARGET_OPENCL_VERSION >= 120 && CL_HPP_MINIMUM_OPENCL_VERSION < 120
1695
// Extracts version number with major in the upper 16 bits, minor in the lower 16
1696
static cl_uint getVersion(const vector<char> &versionInfo)
1698
int highVersion = 0;
1701
while(versionInfo[index] != '.' ) {
1703
highVersion += versionInfo[index]-'0';
1707
while(versionInfo[index] != ' ' && versionInfo[index] != '\0') {
1709
lowVersion += versionInfo[index]-'0';
1712
return (highVersion << 16) | lowVersion;
1715
static cl_uint getPlatformVersion(cl_platform_id platform)
1718
clGetPlatformInfo(platform, CL_PLATFORM_VERSION, 0, NULL, &size);
1720
vector<char> versionInfo(size);
1721
clGetPlatformInfo(platform, CL_PLATFORM_VERSION, size, versionInfo.data(), &size);
1722
return getVersion(versionInfo);
1725
static cl_uint getDevicePlatformVersion(cl_device_id device)
1727
cl_platform_id platform;
1728
clGetDeviceInfo(device, CL_DEVICE_PLATFORM, sizeof(platform), &platform, NULL);
1729
return getPlatformVersion(platform);
1732
static cl_uint getContextPlatformVersion(cl_context context)
1734
// The platform cannot be queried directly, so we first have to grab a
1735
// device and obtain its context
1737
clGetContextInfo(context, CL_CONTEXT_DEVICES, 0, NULL, &size);
1740
vector<cl_device_id> devices(size/sizeof(cl_device_id));
1741
clGetContextInfo(context, CL_CONTEXT_DEVICES, size, devices.data(), NULL);
1742
return getDevicePlatformVersion(devices[0]);
1744
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120 && CL_HPP_MINIMUM_OPENCL_VERSION < 120
1746
template <typename T>
1756
Wrapper() : object_(NULL) { }
1758
Wrapper(const cl_type &obj, bool retainObject) : object_(obj)
1761
detail::errHandler(retain(), __RETAIN_ERR);
1767
if (object_ != NULL) { release(); }
1770
Wrapper(const Wrapper<cl_type>& rhs)
1772
object_ = rhs.object_;
1773
detail::errHandler(retain(), __RETAIN_ERR);
1776
Wrapper(Wrapper<cl_type>&& rhs) CL_HPP_NOEXCEPT_
1778
object_ = rhs.object_;
1782
Wrapper<cl_type>& operator = (const Wrapper<cl_type>& rhs)
1785
detail::errHandler(release(), __RELEASE_ERR);
1786
object_ = rhs.object_;
1787
detail::errHandler(retain(), __RETAIN_ERR);
1792
Wrapper<cl_type>& operator = (Wrapper<cl_type>&& rhs)
1795
detail::errHandler(release(), __RELEASE_ERR);
1796
object_ = rhs.object_;
1802
Wrapper<cl_type>& operator = (const cl_type &rhs)
1804
detail::errHandler(release(), __RELEASE_ERR);
1809
const cl_type& operator ()() const { return object_; }
1811
cl_type& operator ()() { return object_; }
1813
cl_type get() const { return object_; }
1816
template<typename Func, typename U>
1817
friend inline cl_int getInfoHelper(Func, cl_uint, U*, int, typename U::cl_type);
1819
cl_int retain() const
1821
if (object_ != nullptr) {
1822
return ReferenceHandler<cl_type>::retain(object_);
1829
cl_int release() const
1831
if (object_ != nullptr) {
1832
return ReferenceHandler<cl_type>::release(object_);
1841
class Wrapper<cl_device_id>
1844
typedef cl_device_id cl_type;
1848
bool referenceCountable_;
1850
static bool isReferenceCountable(cl_device_id device)
1852
bool retVal = false;
1853
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
1854
#if CL_HPP_MINIMUM_OPENCL_VERSION < 120
1855
if (device != NULL) {
1856
int version = getDevicePlatformVersion(device);
1857
if(version > ((1 << 16) + 1)) {
1861
#else // CL_HPP_MINIMUM_OPENCL_VERSION < 120
1863
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 120
1864
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
1869
Wrapper() : object_(NULL), referenceCountable_(false)
1873
Wrapper(const cl_type &obj, bool retainObject) :
1875
referenceCountable_(false)
1877
referenceCountable_ = isReferenceCountable(obj);
1880
detail::errHandler(retain(), __RETAIN_ERR);
1889
Wrapper(const Wrapper<cl_type>& rhs)
1891
object_ = rhs.object_;
1892
referenceCountable_ = isReferenceCountable(object_);
1893
detail::errHandler(retain(), __RETAIN_ERR);
1896
Wrapper(Wrapper<cl_type>&& rhs) CL_HPP_NOEXCEPT_
1898
object_ = rhs.object_;
1899
referenceCountable_ = rhs.referenceCountable_;
1901
rhs.referenceCountable_ = false;
1904
Wrapper<cl_type>& operator = (const Wrapper<cl_type>& rhs)
1907
detail::errHandler(release(), __RELEASE_ERR);
1908
object_ = rhs.object_;
1909
referenceCountable_ = rhs.referenceCountable_;
1910
detail::errHandler(retain(), __RETAIN_ERR);
1915
Wrapper<cl_type>& operator = (Wrapper<cl_type>&& rhs)
1918
detail::errHandler(release(), __RELEASE_ERR);
1919
object_ = rhs.object_;
1920
referenceCountable_ = rhs.referenceCountable_;
1922
rhs.referenceCountable_ = false;
1927
Wrapper<cl_type>& operator = (const cl_type &rhs)
1929
detail::errHandler(release(), __RELEASE_ERR);
1931
referenceCountable_ = isReferenceCountable(object_);
1935
const cl_type& operator ()() const { return object_; }
1937
cl_type& operator ()() { return object_; }
1939
cl_type get() const { return object_; }
1942
template<typename Func, typename U>
1943
friend inline cl_int getInfoHelper(Func, cl_uint, U*, int, typename U::cl_type);
1945
template<typename Func, typename U>
1946
friend inline cl_int getInfoHelper(Func, cl_uint, vector<U>*, int, typename U::cl_type);
1948
cl_int retain() const
1950
if( object_ != nullptr && referenceCountable_ ) {
1951
return ReferenceHandler<cl_type>::retain(object_);
1958
cl_int release() const
1960
if (object_ != nullptr && referenceCountable_) {
1961
return ReferenceHandler<cl_type>::release(object_);
1969
template <typename T>
1970
inline bool operator==(const Wrapper<T> &lhs, const Wrapper<T> &rhs)
1972
return lhs() == rhs();
1975
template <typename T>
1976
inline bool operator!=(const Wrapper<T> &lhs, const Wrapper<T> &rhs)
1978
return !operator==(lhs, rhs);
1981
} // namespace detail
1985
using BuildLogType = vector<std::pair<cl::Device, typename detail::param_traits<detail::cl_program_build_info, CL_PROGRAM_BUILD_LOG>::param_type>>;
1986
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
1988
* Exception class for build errors to carry build info
1990
class BuildError : public Error
1993
BuildLogType buildLogs;
1995
BuildError(cl_int err, const char * errStr, const BuildLogType &vec) : Error(err, errStr), buildLogs(vec)
1999
BuildLogType getBuildLog() const
2005
static inline cl_int buildErrHandler(
2007
const char * errStr,
2008
const BuildLogType &buildLogs)
2010
if (err != CL_SUCCESS) {
2011
throw BuildError(err, errStr, buildLogs);
2015
} // namespace detail
2019
static inline cl_int buildErrHandler(
2021
const char * errStr,
2022
const BuildLogType &buildLogs)
2024
(void)buildLogs; // suppress unused variable warning
2028
} // namespace detail
2029
#endif // #if defined(CL_HPP_ENABLE_EXCEPTIONS)
2032
/*! \stuct ImageFormat
2033
* \brief Adds constructors and member functions for cl_image_format.
2035
* \see cl_image_format
2037
struct ImageFormat : public cl_image_format
2039
//! \brief Default constructor - performs no initialization.
2042
//! \brief Initializing constructor.
2043
ImageFormat(cl_channel_order order, cl_channel_type type)
2045
image_channel_order = order;
2046
image_channel_data_type = type;
2049
//! \brief Assignment operator.
2050
ImageFormat& operator = (const ImageFormat& rhs)
2053
this->image_channel_data_type = rhs.image_channel_data_type;
2054
this->image_channel_order = rhs.image_channel_order;
2060
/*! \brief Class interface for cl_device_id.
2062
* \note Copies of these objects are inexpensive, since they don't 'own'
2063
* any underlying resources or data structures.
2067
class Device : public detail::Wrapper<cl_device_id>
2070
static std::once_flag default_initialized_;
2071
static Device default_;
2072
static cl_int default_error_;
2074
/*! \brief Create the default context.
2076
* This sets @c default_ and @c default_error_. It does not throw
2079
static void makeDefault();
2081
/*! \brief Create the default platform from a provided platform.
2083
* This sets @c default_. It does not throw
2086
static void makeDefaultProvided(const Device &p) {
2091
#ifdef CL_HPP_UNIT_TEST_ENABLE
2092
/*! \brief Reset the default.
2094
* This sets @c default_ to an empty value to support cleanup in
2095
* the unit test framework.
2096
* This function is not thread safe.
2098
static void unitTestClearDefault() {
2099
default_ = Device();
2101
#endif // #ifdef CL_HPP_UNIT_TEST_ENABLE
2103
//! \brief Default constructor - initializes to NULL.
2104
Device() : detail::Wrapper<cl_type>() { }
2106
/*! \brief Constructor from cl_device_id.
2108
* This simply copies the device ID value, which is an inexpensive operation.
2110
explicit Device(const cl_device_id &device, bool retainObject = false) :
2111
detail::Wrapper<cl_type>(device, retainObject) { }
2113
/*! \brief Returns the first device on the default context.
2115
* \see Context::getDefault()
2117
static Device getDefault(
2118
cl_int *errResult = NULL)
2120
std::call_once(default_initialized_, makeDefault);
2121
detail::errHandler(default_error_);
2122
if (errResult != NULL) {
2123
*errResult = default_error_;
2129
* Modify the default device to be used by
2130
* subsequent operations.
2131
* Will only set the default if no default was previously created.
2132
* @return updated default device.
2133
* Should be compared to the passed value to ensure that it was updated.
2135
static Device setDefault(const Device &default_device)
2137
std::call_once(default_initialized_, makeDefaultProvided, std::cref(default_device));
2138
detail::errHandler(default_error_);
2142
/*! \brief Assignment operator from cl_device_id.
2144
* This simply copies the device ID value, which is an inexpensive operation.
2146
Device& operator = (const cl_device_id& rhs)
2148
detail::Wrapper<cl_type>::operator=(rhs);
2152
/*! \brief Copy constructor to forward copy to the superclass correctly.
2153
* Required for MSVC.
2155
Device(const Device& dev) : detail::Wrapper<cl_type>(dev) {}
2157
/*! \brief Copy assignment to forward copy to the superclass correctly.
2158
* Required for MSVC.
2160
Device& operator = (const Device &dev)
2162
detail::Wrapper<cl_type>::operator=(dev);
2166
/*! \brief Move constructor to forward move to the superclass correctly.
2167
* Required for MSVC.
2169
Device(Device&& dev) CL_HPP_NOEXCEPT_ : detail::Wrapper<cl_type>(std::move(dev)) {}
2171
/*! \brief Move assignment to forward move to the superclass correctly.
2172
* Required for MSVC.
2174
Device& operator = (Device &&dev)
2176
detail::Wrapper<cl_type>::operator=(std::move(dev));
2180
//! \brief Wrapper for clGetDeviceInfo().
2181
template <typename T>
2182
cl_int getInfo(cl_device_info name, T* param) const
2184
return detail::errHandler(
2185
detail::getInfo(&::clGetDeviceInfo, object_, name, param),
2186
__GET_DEVICE_INFO_ERR);
2189
//! \brief Wrapper for clGetDeviceInfo() that returns by value.
2190
template <cl_int name> typename
2191
detail::param_traits<detail::cl_device_info, name>::param_type
2192
getInfo(cl_int* err = NULL) const
2194
typename detail::param_traits<
2195
detail::cl_device_info, name>::param_type param;
2196
cl_int result = getInfo(name, ¶m);
2204
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
2206
* Return the current value of the host clock as seen by the device.
2207
* The resolution of the device timer may be queried with the
2208
* CL_DEVICE_PROFILING_TIMER_RESOLUTION query.
2209
* @return The host timer value.
2211
cl_ulong getHostTimer(cl_int *error = nullptr)
2213
cl_ulong retVal = 0;
2215
clGetHostTimer(this->get(), &retVal);
2218
__GET_HOST_TIMER_ERR);
2226
* Return a synchronized pair of host and device timestamps as seen by device.
2227
* Use to correlate the clocks and get the host timer only using getHostTimer
2228
* as a lower cost mechanism in between calls.
2229
* The resolution of the host timer may be queried with the
2230
* CL_PLATFORM_HOST_TIMER_RESOLUTION query.
2231
* The resolution of the device timer may be queried with the
2232
* CL_DEVICE_PROFILING_TIMER_RESOLUTION query.
2233
* @return A pair of (device timer, host timer) timer values.
2235
std::pair<cl_ulong, cl_ulong> getDeviceAndHostTimer(cl_int *error = nullptr)
2237
std::pair<cl_ulong, cl_ulong> retVal;
2239
clGetDeviceAndHostTimer(this->get(), &(retVal.first), &(retVal.second));
2242
__GET_DEVICE_AND_HOST_TIMER_ERR);
2248
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
2253
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
2254
//! \brief Wrapper for clCreateSubDevices().
2255
cl_int createSubDevices(
2256
const cl_device_partition_property * properties,
2257
vector<Device>* devices)
2260
cl_int err = clCreateSubDevices(object_, properties, 0, NULL, &n);
2261
if (err != CL_SUCCESS) {
2262
return detail::errHandler(err, __CREATE_SUB_DEVICES_ERR);
2265
vector<cl_device_id> ids(n);
2266
err = clCreateSubDevices(object_, properties, n, ids.data(), NULL);
2267
if (err != CL_SUCCESS) {
2268
return detail::errHandler(err, __CREATE_SUB_DEVICES_ERR);
2271
// Cannot trivially assign because we need to capture intermediates
2272
// with safe construction
2274
devices->resize(ids.size());
2276
// Assign to param, constructing with retain behaviour
2277
// to correctly capture each underlying CL object
2278
for (size_type i = 0; i < ids.size(); i++) {
2279
// We do not need to retain because this device is being created
2281
(*devices)[i] = Device(ids[i], false);
2287
#elif defined(CL_HPP_USE_CL_DEVICE_FISSION)
2290
* CL 1.1 version that uses device fission extension.
2292
cl_int createSubDevices(
2293
const cl_device_partition_property_ext * properties,
2294
vector<Device>* devices)
2296
typedef CL_API_ENTRY cl_int
2297
( CL_API_CALL * PFN_clCreateSubDevicesEXT)(
2298
cl_device_id /*in_device*/,
2299
const cl_device_partition_property_ext * /* properties */,
2300
cl_uint /*num_entries*/,
2301
cl_device_id * /*out_devices*/,
2302
cl_uint * /*num_devices*/ ) CL_EXT_SUFFIX__VERSION_1_1;
2304
static PFN_clCreateSubDevicesEXT pfn_clCreateSubDevicesEXT = NULL;
2305
CL_HPP_INIT_CL_EXT_FCN_PTR_(clCreateSubDevicesEXT);
2308
cl_int err = pfn_clCreateSubDevicesEXT(object_, properties, 0, NULL, &n);
2309
if (err != CL_SUCCESS) {
2310
return detail::errHandler(err, __CREATE_SUB_DEVICES_ERR);
2313
vector<cl_device_id> ids(n);
2314
err = pfn_clCreateSubDevicesEXT(object_, properties, n, ids.data(), NULL);
2315
if (err != CL_SUCCESS) {
2316
return detail::errHandler(err, __CREATE_SUB_DEVICES_ERR);
2318
// Cannot trivially assign because we need to capture intermediates
2319
// with safe construction
2321
devices->resize(ids.size());
2323
// Assign to param, constructing with retain behaviour
2324
// to correctly capture each underlying CL object
2325
for (size_type i = 0; i < ids.size(); i++) {
2326
// We do not need to retain because this device is being created
2328
(*devices)[i] = Device(ids[i], false);
2333
#endif // defined(CL_HPP_USE_CL_DEVICE_FISSION)
2336
CL_HPP_DEFINE_STATIC_MEMBER_ std::once_flag Device::default_initialized_;
2337
CL_HPP_DEFINE_STATIC_MEMBER_ Device Device::default_;
2338
CL_HPP_DEFINE_STATIC_MEMBER_ cl_int Device::default_error_ = CL_SUCCESS;
2340
/*! \brief Class interface for cl_platform_id.
2342
* \note Copies of these objects are inexpensive, since they don't 'own'
2343
* any underlying resources or data structures.
2345
* \see cl_platform_id
2347
class Platform : public detail::Wrapper<cl_platform_id>
2350
static std::once_flag default_initialized_;
2351
static Platform default_;
2352
static cl_int default_error_;
2354
/*! \brief Create the default context.
2356
* This sets @c default_ and @c default_error_. It does not throw
2359
static void makeDefault() {
2360
/* Throwing an exception from a call_once invocation does not do
2361
* what we wish, so we catch it and save the error.
2363
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
2367
// If default wasn't passed ,generate one
2371
cl_int err = ::clGetPlatformIDs(0, NULL, &n);
2372
if (err != CL_SUCCESS) {
2373
default_error_ = err;
2377
default_error_ = CL_INVALID_PLATFORM;
2381
vector<cl_platform_id> ids(n);
2382
err = ::clGetPlatformIDs(n, ids.data(), NULL);
2383
if (err != CL_SUCCESS) {
2384
default_error_ = err;
2388
default_ = Platform(ids[0]);
2390
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
2391
catch (cl::Error &e) {
2392
default_error_ = e.err();
2397
/*! \brief Create the default platform from a provided platform.
2399
* This sets @c default_. It does not throw
2402
static void makeDefaultProvided(const Platform &p) {
2407
#ifdef CL_HPP_UNIT_TEST_ENABLE
2408
/*! \brief Reset the default.
2410
* This sets @c default_ to an empty value to support cleanup in
2411
* the unit test framework.
2412
* This function is not thread safe.
2414
static void unitTestClearDefault() {
2415
default_ = Platform();
2417
#endif // #ifdef CL_HPP_UNIT_TEST_ENABLE
2419
//! \brief Default constructor - initializes to NULL.
2420
Platform() : detail::Wrapper<cl_type>() { }
2422
/*! \brief Constructor from cl_platform_id.
2424
* \param retainObject will cause the constructor to retain its cl object.
2425
* Defaults to false to maintain compatibility with
2427
* This simply copies the platform ID value, which is an inexpensive operation.
2429
explicit Platform(const cl_platform_id &platform, bool retainObject = false) :
2430
detail::Wrapper<cl_type>(platform, retainObject) { }
2432
/*! \brief Assignment operator from cl_platform_id.
2434
* This simply copies the platform ID value, which is an inexpensive operation.
2436
Platform& operator = (const cl_platform_id& rhs)
2438
detail::Wrapper<cl_type>::operator=(rhs);
2442
static Platform getDefault(
2443
cl_int *errResult = NULL)
2445
std::call_once(default_initialized_, makeDefault);
2446
detail::errHandler(default_error_);
2447
if (errResult != NULL) {
2448
*errResult = default_error_;
2454
* Modify the default platform to be used by
2455
* subsequent operations.
2456
* Will only set the default if no default was previously created.
2457
* @return updated default platform.
2458
* Should be compared to the passed value to ensure that it was updated.
2460
static Platform setDefault(const Platform &default_platform)
2462
std::call_once(default_initialized_, makeDefaultProvided, std::cref(default_platform));
2463
detail::errHandler(default_error_);
2467
//! \brief Wrapper for clGetPlatformInfo().
2468
cl_int getInfo(cl_platform_info name, string* param) const
2470
return detail::errHandler(
2471
detail::getInfo(&::clGetPlatformInfo, object_, name, param),
2472
__GET_PLATFORM_INFO_ERR);
2475
//! \brief Wrapper for clGetPlatformInfo() that returns by value.
2476
template <cl_int name> typename
2477
detail::param_traits<detail::cl_platform_info, name>::param_type
2478
getInfo(cl_int* err = NULL) const
2480
typename detail::param_traits<
2481
detail::cl_platform_info, name>::param_type param;
2482
cl_int result = getInfo(name, ¶m);
2489
/*! \brief Gets a list of devices for this platform.
2491
* Wraps clGetDeviceIDs().
2494
cl_device_type type,
2495
vector<Device>* devices) const
2498
if( devices == NULL ) {
2499
return detail::errHandler(CL_INVALID_ARG_VALUE, __GET_DEVICE_IDS_ERR);
2501
cl_int err = ::clGetDeviceIDs(object_, type, 0, NULL, &n);
2502
if (err != CL_SUCCESS) {
2503
return detail::errHandler(err, __GET_DEVICE_IDS_ERR);
2506
vector<cl_device_id> ids(n);
2507
err = ::clGetDeviceIDs(object_, type, n, ids.data(), NULL);
2508
if (err != CL_SUCCESS) {
2509
return detail::errHandler(err, __GET_DEVICE_IDS_ERR);
2512
// Cannot trivially assign because we need to capture intermediates
2513
// with safe construction
2514
// We must retain things we obtain from the API to avoid releasing
2515
// API-owned objects.
2517
devices->resize(ids.size());
2519
// Assign to param, constructing with retain behaviour
2520
// to correctly capture each underlying CL object
2521
for (size_type i = 0; i < ids.size(); i++) {
2522
(*devices)[i] = Device(ids[i], true);
2528
#if defined(CL_HPP_USE_DX_INTEROP)
2529
/*! \brief Get the list of available D3D10 devices.
2531
* \param d3d_device_source.
2533
* \param d3d_object.
2535
* \param d3d_device_set.
2537
* \param devices returns a vector of OpenCL D3D10 devices found. The cl::Device
2538
* values returned in devices can be used to identify a specific OpenCL
2539
* device. If \a devices argument is NULL, this argument is ignored.
2541
* \return One of the following values:
2542
* - CL_SUCCESS if the function is executed successfully.
2544
* The application can query specific capabilities of the OpenCL device(s)
2545
* returned by cl::getDevices. This can be used by the application to
2546
* determine which device(s) to use.
2548
* \note In the case that exceptions are enabled and a return value
2549
* other than CL_SUCCESS is generated, then cl::Error exception is
2553
cl_d3d10_device_source_khr d3d_device_source,
2555
cl_d3d10_device_set_khr d3d_device_set,
2556
vector<Device>* devices) const
2558
typedef CL_API_ENTRY cl_int (CL_API_CALL *PFN_clGetDeviceIDsFromD3D10KHR)(
2559
cl_platform_id platform,
2560
cl_d3d10_device_source_khr d3d_device_source,
2562
cl_d3d10_device_set_khr d3d_device_set,
2563
cl_uint num_entries,
2564
cl_device_id * devices,
2565
cl_uint* num_devices);
2567
if( devices == NULL ) {
2568
return detail::errHandler(CL_INVALID_ARG_VALUE, __GET_DEVICE_IDS_ERR);
2571
static PFN_clGetDeviceIDsFromD3D10KHR pfn_clGetDeviceIDsFromD3D10KHR = NULL;
2572
CL_HPP_INIT_CL_EXT_FCN_PTR_PLATFORM_(object_, clGetDeviceIDsFromD3D10KHR);
2575
cl_int err = pfn_clGetDeviceIDsFromD3D10KHR(
2583
if (err != CL_SUCCESS) {
2584
return detail::errHandler(err, __GET_DEVICE_IDS_ERR);
2587
vector<cl_device_id> ids(n);
2588
err = pfn_clGetDeviceIDsFromD3D10KHR(
2596
if (err != CL_SUCCESS) {
2597
return detail::errHandler(err, __GET_DEVICE_IDS_ERR);
2600
// Cannot trivially assign because we need to capture intermediates
2601
// with safe construction
2602
// We must retain things we obtain from the API to avoid releasing
2603
// API-owned objects.
2605
devices->resize(ids.size());
2607
// Assign to param, constructing with retain behaviour
2608
// to correctly capture each underlying CL object
2609
for (size_type i = 0; i < ids.size(); i++) {
2610
(*devices)[i] = Device(ids[i], true);
2617
/*! \brief Gets a list of available platforms.
2619
* Wraps clGetPlatformIDs().
2622
vector<Platform>* platforms)
2626
if( platforms == NULL ) {
2627
return detail::errHandler(CL_INVALID_ARG_VALUE, __GET_PLATFORM_IDS_ERR);
2630
cl_int err = ::clGetPlatformIDs(0, NULL, &n);
2631
if (err != CL_SUCCESS) {
2632
return detail::errHandler(err, __GET_PLATFORM_IDS_ERR);
2635
vector<cl_platform_id> ids(n);
2636
err = ::clGetPlatformIDs(n, ids.data(), NULL);
2637
if (err != CL_SUCCESS) {
2638
return detail::errHandler(err, __GET_PLATFORM_IDS_ERR);
2642
platforms->resize(ids.size());
2644
// Platforms don't reference count
2645
for (size_type i = 0; i < ids.size(); i++) {
2646
(*platforms)[i] = Platform(ids[i]);
2652
/*! \brief Gets the first available platform.
2654
* Wraps clGetPlatformIDs(), returning the first result.
2657
Platform * platform)
2660
Platform default_platform = Platform::getDefault(&err);
2662
*platform = default_platform;
2667
/*! \brief Gets the first available platform, returning it by value.
2669
* \return Returns a valid platform if one is available.
2670
* If no platform is available will return a null platform.
2671
* Throws an exception if no platforms are available
2672
* or an error condition occurs.
2673
* Wraps clGetPlatformIDs(), returning the first result.
2675
static Platform get(
2676
cl_int * errResult = NULL)
2679
Platform default_platform = Platform::getDefault(&err);
2683
return default_platform;
2686
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
2687
//! \brief Wrapper for clUnloadCompiler().
2691
return ::clUnloadPlatformCompiler(object_);
2693
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
2694
}; // class Platform
2696
CL_HPP_DEFINE_STATIC_MEMBER_ std::once_flag Platform::default_initialized_;
2697
CL_HPP_DEFINE_STATIC_MEMBER_ Platform Platform::default_;
2698
CL_HPP_DEFINE_STATIC_MEMBER_ cl_int Platform::default_error_ = CL_SUCCESS;
2702
* Deprecated APIs for 1.2
2704
#if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
2706
* Unload the OpenCL compiler.
2707
* \note Deprecated for OpenCL 1.2. Use Platform::unloadCompiler instead.
2709
inline CL_EXT_PREFIX__VERSION_1_1_DEPRECATED cl_int
2710
UnloadCompiler() CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED;
2714
return ::clUnloadCompiler();
2716
#endif // #if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
2718
/*! \brief Class interface for cl_context.
2720
* \note Copies of these objects are shallow, meaning that the copy will refer
2721
* to the same underlying cl_context as the original. For details, see
2722
* clRetainContext() and clReleaseContext().
2727
: public detail::Wrapper<cl_context>
2730
static std::once_flag default_initialized_;
2731
static Context default_;
2732
static cl_int default_error_;
2734
/*! \brief Create the default context from the default device type in the default platform.
2736
* This sets @c default_ and @c default_error_. It does not throw
2739
static void makeDefault() {
2740
/* Throwing an exception from a call_once invocation does not do
2741
* what we wish, so we catch it and save the error.
2743
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
2747
#if !defined(__APPLE__) && !defined(__MACOS)
2748
const Platform &p = Platform::getDefault();
2749
cl_platform_id defaultPlatform = p();
2750
cl_context_properties properties[3] = {
2751
CL_CONTEXT_PLATFORM, (cl_context_properties)defaultPlatform, 0
2753
#else // #if !defined(__APPLE__) && !defined(__MACOS)
2754
cl_context_properties *properties = nullptr;
2755
#endif // #if !defined(__APPLE__) && !defined(__MACOS)
2758
CL_DEVICE_TYPE_DEFAULT,
2764
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
2765
catch (cl::Error &e) {
2766
default_error_ = e.err();
2772
/*! \brief Create the default context from a provided Context.
2774
* This sets @c default_. It does not throw
2777
static void makeDefaultProvided(const Context &c) {
2782
#ifdef CL_HPP_UNIT_TEST_ENABLE
2783
/*! \brief Reset the default.
2785
* This sets @c default_ to an empty value to support cleanup in
2786
* the unit test framework.
2787
* This function is not thread safe.
2789
static void unitTestClearDefault() {
2790
default_ = Context();
2792
#endif // #ifdef CL_HPP_UNIT_TEST_ENABLE
2794
/*! \brief Constructs a context including a list of specified devices.
2796
* Wraps clCreateContext().
2799
const vector<Device>& devices,
2800
cl_context_properties* properties = NULL,
2801
void (CL_CALLBACK * notifyFptr)(
2811
size_type numDevices = devices.size();
2812
vector<cl_device_id> deviceIDs(numDevices);
2814
for( size_type deviceIndex = 0; deviceIndex < numDevices; ++deviceIndex ) {
2815
deviceIDs[deviceIndex] = (devices[deviceIndex])();
2818
object_ = ::clCreateContext(
2819
properties, (cl_uint) numDevices,
2821
notifyFptr, data, &error);
2823
detail::errHandler(error, __CREATE_CONTEXT_ERR);
2830
const Device& device,
2831
cl_context_properties* properties = NULL,
2832
void (CL_CALLBACK * notifyFptr)(
2842
cl_device_id deviceID = device();
2844
object_ = ::clCreateContext(
2847
notifyFptr, data, &error);
2849
detail::errHandler(error, __CREATE_CONTEXT_ERR);
2855
/*! \brief Constructs a context including all or a subset of devices of a specified type.
2857
* Wraps clCreateContextFromType().
2860
cl_device_type type,
2861
cl_context_properties* properties = NULL,
2862
void (CL_CALLBACK * notifyFptr)(
2872
#if !defined(__APPLE__) && !defined(__MACOS)
2873
cl_context_properties prop[4] = {CL_CONTEXT_PLATFORM, 0, 0, 0 };
2875
if (properties == NULL) {
2876
// Get a valid platform ID as we cannot send in a blank one
2877
vector<Platform> platforms;
2878
error = Platform::get(&platforms);
2879
if (error != CL_SUCCESS) {
2880
detail::errHandler(error, __CREATE_CONTEXT_FROM_TYPE_ERR);
2887
// Check the platforms we found for a device of our specified type
2888
cl_context_properties platform_id = 0;
2889
for (unsigned int i = 0; i < platforms.size(); i++) {
2891
vector<Device> devices;
2893
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
2897
error = platforms[i].getDevices(type, &devices);
2899
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
2900
} catch (cl::Error& e) {
2903
// Catch if exceptions are enabled as we don't want to exit if first platform has no devices of type
2904
// We do error checking next anyway, and can throw there if needed
2907
// Only squash CL_SUCCESS and CL_DEVICE_NOT_FOUND
2908
if (error != CL_SUCCESS && error != CL_DEVICE_NOT_FOUND) {
2909
detail::errHandler(error, __CREATE_CONTEXT_FROM_TYPE_ERR);
2915
if (devices.size() > 0) {
2916
platform_id = (cl_context_properties)platforms[i]();
2921
if (platform_id == 0) {
2922
detail::errHandler(CL_DEVICE_NOT_FOUND, __CREATE_CONTEXT_FROM_TYPE_ERR);
2924
*err = CL_DEVICE_NOT_FOUND;
2929
prop[1] = platform_id;
2930
properties = &prop[0];
2933
object_ = ::clCreateContextFromType(
2934
properties, type, notifyFptr, data, &error);
2936
detail::errHandler(error, __CREATE_CONTEXT_FROM_TYPE_ERR);
2942
/*! \brief Copy constructor to forward copy to the superclass correctly.
2943
* Required for MSVC.
2945
Context(const Context& ctx) : detail::Wrapper<cl_type>(ctx) {}
2947
/*! \brief Copy assignment to forward copy to the superclass correctly.
2948
* Required for MSVC.
2950
Context& operator = (const Context &ctx)
2952
detail::Wrapper<cl_type>::operator=(ctx);
2956
/*! \brief Move constructor to forward move to the superclass correctly.
2957
* Required for MSVC.
2959
Context(Context&& ctx) CL_HPP_NOEXCEPT_ : detail::Wrapper<cl_type>(std::move(ctx)) {}
2961
/*! \brief Move assignment to forward move to the superclass correctly.
2962
* Required for MSVC.
2964
Context& operator = (Context &&ctx)
2966
detail::Wrapper<cl_type>::operator=(std::move(ctx));
2971
/*! \brief Returns a singleton context including all devices of CL_DEVICE_TYPE_DEFAULT.
2973
* \note All calls to this function return the same cl_context as the first.
2975
static Context getDefault(cl_int * err = NULL)
2977
std::call_once(default_initialized_, makeDefault);
2978
detail::errHandler(default_error_);
2980
*err = default_error_;
2986
* Modify the default context to be used by
2987
* subsequent operations.
2988
* Will only set the default if no default was previously created.
2989
* @return updated default context.
2990
* Should be compared to the passed value to ensure that it was updated.
2992
static Context setDefault(const Context &default_context)
2994
std::call_once(default_initialized_, makeDefaultProvided, std::cref(default_context));
2995
detail::errHandler(default_error_);
2999
//! \brief Default constructor - initializes to NULL.
3000
Context() : detail::Wrapper<cl_type>() { }
3002
/*! \brief Constructor from cl_context - takes ownership.
3004
* This effectively transfers ownership of a refcount on the cl_context
3005
* into the new Context object.
3007
explicit Context(const cl_context& context, bool retainObject = false) :
3008
detail::Wrapper<cl_type>(context, retainObject) { }
3010
/*! \brief Assignment operator from cl_context - takes ownership.
3012
* This effectively transfers ownership of a refcount on the rhs and calls
3013
* clReleaseContext() on the value previously held by this instance.
3015
Context& operator = (const cl_context& rhs)
3017
detail::Wrapper<cl_type>::operator=(rhs);
3021
//! \brief Wrapper for clGetContextInfo().
3022
template <typename T>
3023
cl_int getInfo(cl_context_info name, T* param) const
3025
return detail::errHandler(
3026
detail::getInfo(&::clGetContextInfo, object_, name, param),
3027
__GET_CONTEXT_INFO_ERR);
3030
//! \brief Wrapper for clGetContextInfo() that returns by value.
3031
template <cl_int name> typename
3032
detail::param_traits<detail::cl_context_info, name>::param_type
3033
getInfo(cl_int* err = NULL) const
3035
typename detail::param_traits<
3036
detail::cl_context_info, name>::param_type param;
3037
cl_int result = getInfo(name, ¶m);
3044
/*! \brief Gets a list of supported image formats.
3046
* Wraps clGetSupportedImageFormats().
3048
cl_int getSupportedImageFormats(
3050
cl_mem_object_type type,
3051
vector<ImageFormat>* formats) const
3059
cl_int err = ::clGetSupportedImageFormats(
3066
if (err != CL_SUCCESS) {
3067
return detail::errHandler(err, __GET_SUPPORTED_IMAGE_FORMATS_ERR);
3070
if (numEntries > 0) {
3071
vector<ImageFormat> value(numEntries);
3072
err = ::clGetSupportedImageFormats(
3077
(cl_image_format*)value.data(),
3079
if (err != CL_SUCCESS) {
3080
return detail::errHandler(err, __GET_SUPPORTED_IMAGE_FORMATS_ERR);
3083
formats->assign(begin(value), end(value));
3086
// If no values are being returned, ensure an empty vector comes back
3094
inline void Device::makeDefault()
3096
/* Throwing an exception from a call_once invocation does not do
3097
* what we wish, so we catch it and save the error.
3099
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
3105
Context context = Context::getDefault(&error);
3106
detail::errHandler(error, __CREATE_CONTEXT_ERR);
3108
if (error != CL_SUCCESS) {
3109
default_error_ = error;
3112
default_ = context.getInfo<CL_CONTEXT_DEVICES>()[0];
3113
default_error_ = CL_SUCCESS;
3116
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
3117
catch (cl::Error &e) {
3118
default_error_ = e.err();
3123
CL_HPP_DEFINE_STATIC_MEMBER_ std::once_flag Context::default_initialized_;
3124
CL_HPP_DEFINE_STATIC_MEMBER_ Context Context::default_;
3125
CL_HPP_DEFINE_STATIC_MEMBER_ cl_int Context::default_error_ = CL_SUCCESS;
3127
/*! \brief Class interface for cl_event.
3129
* \note Copies of these objects are shallow, meaning that the copy will refer
3130
* to the same underlying cl_event as the original. For details, see
3131
* clRetainEvent() and clReleaseEvent().
3135
class Event : public detail::Wrapper<cl_event>
3138
//! \brief Default constructor - initializes to NULL.
3139
Event() : detail::Wrapper<cl_type>() { }
3141
/*! \brief Constructor from cl_event - takes ownership.
3143
* \param retainObject will cause the constructor to retain its cl object.
3144
* Defaults to false to maintain compatibility with
3146
* This effectively transfers ownership of a refcount on the cl_event
3147
* into the new Event object.
3149
explicit Event(const cl_event& event, bool retainObject = false) :
3150
detail::Wrapper<cl_type>(event, retainObject) { }
3152
/*! \brief Assignment operator from cl_event - takes ownership.
3154
* This effectively transfers ownership of a refcount on the rhs and calls
3155
* clReleaseEvent() on the value previously held by this instance.
3157
Event& operator = (const cl_event& rhs)
3159
detail::Wrapper<cl_type>::operator=(rhs);
3163
//! \brief Wrapper for clGetEventInfo().
3164
template <typename T>
3165
cl_int getInfo(cl_event_info name, T* param) const
3167
return detail::errHandler(
3168
detail::getInfo(&::clGetEventInfo, object_, name, param),
3169
__GET_EVENT_INFO_ERR);
3172
//! \brief Wrapper for clGetEventInfo() that returns by value.
3173
template <cl_int name> typename
3174
detail::param_traits<detail::cl_event_info, name>::param_type
3175
getInfo(cl_int* err = NULL) const
3177
typename detail::param_traits<
3178
detail::cl_event_info, name>::param_type param;
3179
cl_int result = getInfo(name, ¶m);
3186
//! \brief Wrapper for clGetEventProfilingInfo().
3187
template <typename T>
3188
cl_int getProfilingInfo(cl_profiling_info name, T* param) const
3190
return detail::errHandler(detail::getInfo(
3191
&::clGetEventProfilingInfo, object_, name, param),
3192
__GET_EVENT_PROFILE_INFO_ERR);
3195
//! \brief Wrapper for clGetEventProfilingInfo() that returns by value.
3196
template <cl_int name> typename
3197
detail::param_traits<detail::cl_profiling_info, name>::param_type
3198
getProfilingInfo(cl_int* err = NULL) const
3200
typename detail::param_traits<
3201
detail::cl_profiling_info, name>::param_type param;
3202
cl_int result = getProfilingInfo(name, ¶m);
3209
/*! \brief Blocks the calling thread until this event completes.
3211
* Wraps clWaitForEvents().
3215
return detail::errHandler(
3216
::clWaitForEvents(1, &object_),
3217
__WAIT_FOR_EVENTS_ERR);
3220
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
3221
/*! \brief Registers a user callback function for a specific command execution status.
3223
* Wraps clSetEventCallback().
3227
void (CL_CALLBACK * pfn_notify)(cl_event, cl_int, void *),
3228
void * user_data = NULL)
3230
return detail::errHandler(
3231
::clSetEventCallback(
3236
__SET_EVENT_CALLBACK_ERR);
3238
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 110
3240
/*! \brief Blocks the calling thread until every event specified is complete.
3242
* Wraps clWaitForEvents().
3245
waitForEvents(const vector<Event>& events)
3247
return detail::errHandler(
3249
(cl_uint) events.size(), (events.size() > 0) ? (cl_event*)&events.front() : NULL),
3250
__WAIT_FOR_EVENTS_ERR);
3254
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
3255
/*! \brief Class interface for user events (a subset of cl_event's).
3257
* See Event for details about copy semantics, etc.
3259
class UserEvent : public Event
3262
/*! \brief Constructs a user event on a given context.
3264
* Wraps clCreateUserEvent().
3267
const Context& context,
3268
cl_int * err = NULL)
3271
object_ = ::clCreateUserEvent(
3275
detail::errHandler(error, __CREATE_USER_EVENT_ERR);
3281
//! \brief Default constructor - initializes to NULL.
3282
UserEvent() : Event() { }
3284
/*! \brief Sets the execution status of a user event object.
3286
* Wraps clSetUserEventStatus().
3288
cl_int setStatus(cl_int status)
3290
return detail::errHandler(
3291
::clSetUserEventStatus(object_,status),
3292
__SET_USER_EVENT_STATUS_ERR);
3295
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 110
3297
/*! \brief Blocks the calling thread until every event specified is complete.
3299
* Wraps clWaitForEvents().
3301
inline static cl_int
3302
WaitForEvents(const vector<Event>& events)
3304
return detail::errHandler(
3306
(cl_uint) events.size(), (events.size() > 0) ? (cl_event*)&events.front() : NULL),
3307
__WAIT_FOR_EVENTS_ERR);
3310
/*! \brief Class interface for cl_mem.
3312
* \note Copies of these objects are shallow, meaning that the copy will refer
3313
* to the same underlying cl_mem as the original. For details, see
3314
* clRetainMemObject() and clReleaseMemObject().
3318
class Memory : public detail::Wrapper<cl_mem>
3321
//! \brief Default constructor - initializes to NULL.
3322
Memory() : detail::Wrapper<cl_type>() { }
3324
/*! \brief Constructor from cl_mem - takes ownership.
3326
* Optionally transfer ownership of a refcount on the cl_mem
3327
* into the new Memory object.
3329
* \param retainObject will cause the constructor to retain its cl object.
3330
* Defaults to false to maintain compatibility with
3333
* See Memory for further details.
3335
explicit Memory(const cl_mem& memory, bool retainObject) :
3336
detail::Wrapper<cl_type>(memory, retainObject) { }
3338
/*! \brief Assignment operator from cl_mem - takes ownership.
3340
* This effectively transfers ownership of a refcount on the rhs and calls
3341
* clReleaseMemObject() on the value previously held by this instance.
3343
Memory& operator = (const cl_mem& rhs)
3345
detail::Wrapper<cl_type>::operator=(rhs);
3349
/*! \brief Copy constructor to forward copy to the superclass correctly.
3350
* Required for MSVC.
3352
Memory(const Memory& mem) : detail::Wrapper<cl_type>(mem) {}
3354
/*! \brief Copy assignment to forward copy to the superclass correctly.
3355
* Required for MSVC.
3357
Memory& operator = (const Memory &mem)
3359
detail::Wrapper<cl_type>::operator=(mem);
3363
/*! \brief Move constructor to forward move to the superclass correctly.
3364
* Required for MSVC.
3366
Memory(Memory&& mem) CL_HPP_NOEXCEPT_ : detail::Wrapper<cl_type>(std::move(mem)) {}
3368
/*! \brief Move assignment to forward move to the superclass correctly.
3369
* Required for MSVC.
3371
Memory& operator = (Memory &&mem)
3373
detail::Wrapper<cl_type>::operator=(std::move(mem));
3378
//! \brief Wrapper for clGetMemObjectInfo().
3379
template <typename T>
3380
cl_int getInfo(cl_mem_info name, T* param) const
3382
return detail::errHandler(
3383
detail::getInfo(&::clGetMemObjectInfo, object_, name, param),
3384
__GET_MEM_OBJECT_INFO_ERR);
3387
//! \brief Wrapper for clGetMemObjectInfo() that returns by value.
3388
template <cl_int name> typename
3389
detail::param_traits<detail::cl_mem_info, name>::param_type
3390
getInfo(cl_int* err = NULL) const
3392
typename detail::param_traits<
3393
detail::cl_mem_info, name>::param_type param;
3394
cl_int result = getInfo(name, ¶m);
3401
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
3402
/*! \brief Registers a callback function to be called when the memory object
3403
* is no longer needed.
3405
* Wraps clSetMemObjectDestructorCallback().
3407
* Repeated calls to this function, for a given cl_mem value, will append
3408
* to the list of functions called (in reverse order) when memory object's
3409
* resources are freed and the memory object is deleted.
3412
* The registered callbacks are associated with the underlying cl_mem
3413
* value - not the Memory class instance.
3415
cl_int setDestructorCallback(
3416
void (CL_CALLBACK * pfn_notify)(cl_mem, void *),
3417
void * user_data = NULL)
3419
return detail::errHandler(
3420
::clSetMemObjectDestructorCallback(
3424
__SET_MEM_OBJECT_DESTRUCTOR_CALLBACK_ERR);
3426
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 110
3430
// Pre-declare copy functions
3432
template< typename IteratorType >
3433
cl_int copy( IteratorType startIterator, IteratorType endIterator, cl::Buffer &buffer );
3434
template< typename IteratorType >
3435
cl_int copy( const cl::Buffer &buffer, IteratorType startIterator, IteratorType endIterator );
3436
template< typename IteratorType >
3437
cl_int copy( const CommandQueue &queue, IteratorType startIterator, IteratorType endIterator, cl::Buffer &buffer );
3438
template< typename IteratorType >
3439
cl_int copy( const CommandQueue &queue, const cl::Buffer &buffer, IteratorType startIterator, IteratorType endIterator );
3442
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
3448
static cl_svm_mem_flags getSVMMemFlags()
3453
} // namespace detail
3455
template<class Trait = detail::SVMTraitNull>
3456
class SVMTraitReadWrite
3459
static cl_svm_mem_flags getSVMMemFlags()
3461
return CL_MEM_READ_WRITE |
3462
Trait::getSVMMemFlags();
3466
template<class Trait = detail::SVMTraitNull>
3467
class SVMTraitReadOnly
3470
static cl_svm_mem_flags getSVMMemFlags()
3472
return CL_MEM_READ_ONLY |
3473
Trait::getSVMMemFlags();
3477
template<class Trait = detail::SVMTraitNull>
3478
class SVMTraitWriteOnly
3481
static cl_svm_mem_flags getSVMMemFlags()
3483
return CL_MEM_WRITE_ONLY |
3484
Trait::getSVMMemFlags();
3488
template<class Trait = SVMTraitReadWrite<>>
3489
class SVMTraitCoarse
3492
static cl_svm_mem_flags getSVMMemFlags()
3494
return Trait::getSVMMemFlags();
3498
template<class Trait = SVMTraitReadWrite<>>
3502
static cl_svm_mem_flags getSVMMemFlags()
3504
return CL_MEM_SVM_FINE_GRAIN_BUFFER |
3505
Trait::getSVMMemFlags();
3509
template<class Trait = SVMTraitReadWrite<>>
3510
class SVMTraitAtomic
3513
static cl_svm_mem_flags getSVMMemFlags()
3516
CL_MEM_SVM_FINE_GRAIN_BUFFER |
3517
CL_MEM_SVM_ATOMICS |
3518
Trait::getSVMMemFlags();
3522
// Pre-declare SVM map function
3523
template<typename T>
3524
inline cl_int enqueueMapSVM(
3529
const vector<Event>* events = NULL,
3530
Event* event = NULL);
3533
* STL-like allocator class for managing SVM objects provided for convenience.
3535
* Note that while this behaves like an allocator for the purposes of constructing vectors and similar objects,
3536
* care must be taken when using with smart pointers.
3537
* The allocator should not be used to construct a unique_ptr if we are using coarse-grained SVM mode because
3538
* the coarse-grained management behaviour would behave incorrectly with respect to reference counting.
3540
* Instead the allocator embeds a Deleter which may be used with unique_ptr and is used
3541
* with the allocate_shared and allocate_ptr supplied operations.
3543
template<typename T, class SVMTrait>
3544
class SVMAllocator {
3549
typedef T value_type;
3550
typedef value_type* pointer;
3551
typedef const value_type* const_pointer;
3552
typedef value_type& reference;
3553
typedef const value_type& const_reference;
3554
typedef std::size_t size_type;
3555
typedef std::ptrdiff_t difference_type;
3557
template<typename U>
3560
typedef SVMAllocator<U, SVMTrait> other;
3563
template<typename U, typename V>
3564
friend class SVMAllocator;
3567
context_(Context::getDefault())
3571
explicit SVMAllocator(cl::Context context) :
3577
SVMAllocator(const SVMAllocator &other) :
3578
context_(other.context_)
3582
template<typename U>
3583
SVMAllocator(const SVMAllocator<U, SVMTrait> &other) :
3584
context_(other.context_)
3592
pointer address(reference r) CL_HPP_NOEXCEPT_
3594
return std::addressof(r);
3597
const_pointer address(const_reference r) CL_HPP_NOEXCEPT_
3599
return std::addressof(r);
3603
* Allocate an SVM pointer.
3605
* If the allocator is coarse-grained, this will take ownership to allow
3606
* containers to correctly construct data in place.
3610
typename cl::SVMAllocator<void, SVMTrait>::const_pointer = 0)
3612
// Allocate memory with default alignment matching the size of the type
3616
SVMTrait::getSVMMemFlags(),
3619
pointer retValue = reinterpret_cast<pointer>(
3621
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
3623
std::bad_alloc excep;
3626
#endif // #if defined(CL_HPP_ENABLE_EXCEPTIONS)
3628
// If allocation was coarse-grained then map it
3629
if (!(SVMTrait::getSVMMemFlags() & CL_MEM_SVM_FINE_GRAIN_BUFFER)) {
3630
cl_int err = enqueueMapSVM(retValue, CL_TRUE, CL_MAP_READ | CL_MAP_WRITE, size*sizeof(T));
3631
if (err != CL_SUCCESS) {
3632
std::bad_alloc excep;
3637
// If exceptions disabled, return null pointer from allocator
3641
void deallocate(pointer p, size_type)
3643
clSVMFree(context_(), p);
3647
* Return the maximum possible allocation size.
3648
* This is the minimum of the maximum sizes of all devices in the context.
3650
size_type max_size() const CL_HPP_NOEXCEPT_
3652
size_type maxSize = std::numeric_limits<size_type>::max() / sizeof(T);
3654
for (const Device &d : context_.getInfo<CL_CONTEXT_DEVICES>()) {
3657
static_cast<size_type>(d.getInfo<CL_DEVICE_MAX_MEM_ALLOC_SIZE>()));
3663
template< class U, class... Args >
3664
void construct(U* p, Args&&... args)
3676
* Returns true if the contexts match.
3678
inline bool operator==(SVMAllocator const& rhs)
3680
return (context_==rhs.context_);
3683
inline bool operator!=(SVMAllocator const& a)
3685
return !operator==(a);
3687
}; // class SVMAllocator return cl::pointer<T>(tmp, detail::Deleter<T, Alloc>{alloc, copies});
3690
template<class SVMTrait>
3691
class SVMAllocator<void, SVMTrait> {
3693
typedef void value_type;
3694
typedef value_type* pointer;
3695
typedef const value_type* const_pointer;
3697
template<typename U>
3700
typedef SVMAllocator<U, SVMTrait> other;
3703
template<typename U, typename V>
3704
friend class SVMAllocator;
3707
#if !defined(CL_HPP_NO_STD_UNIQUE_PTR)
3710
template<class Alloc>
3717
typedef typename std::allocator_traits<Alloc>::pointer pointer;
3719
Deleter(const Alloc &alloc, size_type copies) : alloc_{ alloc }, copies_{ copies }
3723
void operator()(pointer ptr) const {
3724
Alloc tmpAlloc{ alloc_ };
3725
std::allocator_traits<Alloc>::destroy(tmpAlloc, std::addressof(*ptr));
3726
std::allocator_traits<Alloc>::deallocate(tmpAlloc, ptr, copies_);
3729
} // namespace detail
3732
* Allocation operation compatible with std::allocate_ptr.
3733
* Creates a unique_ptr<T> by default.
3734
* This requirement is to ensure that the control block is not
3735
* allocated in memory inaccessible to the host.
3737
template <class T, class Alloc, class... Args>
3738
cl::pointer<T, detail::Deleter<Alloc>> allocate_pointer(const Alloc &alloc_, Args&&... args)
3740
Alloc alloc(alloc_);
3741
static const size_type copies = 1;
3743
// Ensure that creation of the management block and the
3744
// object are dealt with separately such that we only provide a deleter
3746
T* tmp = std::allocator_traits<Alloc>::allocate(alloc, copies);
3748
std::bad_alloc excep;
3752
std::allocator_traits<Alloc>::construct(
3754
std::addressof(*tmp),
3755
std::forward<Args>(args)...);
3757
return cl::pointer<T, detail::Deleter<Alloc>>(tmp, detail::Deleter<Alloc>{alloc, copies});
3759
catch (std::bad_alloc& b)
3761
std::allocator_traits<Alloc>::deallocate(alloc, tmp, copies);
3766
template< class T, class SVMTrait, class... Args >
3767
cl::pointer<T, detail::Deleter<SVMAllocator<T, SVMTrait>>> allocate_svm(Args... args)
3769
SVMAllocator<T, SVMTrait> alloc;
3770
return cl::allocate_pointer<T>(alloc, args...);
3773
template< class T, class SVMTrait, class... Args >
3774
cl::pointer<T, detail::Deleter<SVMAllocator<T, SVMTrait>>> allocate_svm(const cl::Context &c, Args... args)
3776
SVMAllocator<T, SVMTrait> alloc(c);
3777
return cl::allocate_pointer<T>(alloc, args...);
3779
#endif // #if !defined(CL_HPP_NO_STD_UNIQUE_PTR)
3781
/*! \brief Vector alias to simplify contruction of coarse-grained SVM containers.
3784
template < class T >
3785
using coarse_svm_vector = vector<T, cl::SVMAllocator<int, cl::SVMTraitCoarse<>>>;
3787
/*! \brief Vector alias to simplify contruction of fine-grained SVM containers.
3790
template < class T >
3791
using fine_svm_vector = vector<T, cl::SVMAllocator<int, cl::SVMTraitFine<>>>;
3793
/*! \brief Vector alias to simplify contruction of fine-grained SVM containers that support platform atomics.
3796
template < class T >
3797
using atomic_svm_vector = vector<T, cl::SVMAllocator<int, cl::SVMTraitAtomic<>>>;
3799
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
3802
/*! \brief Class interface for Buffer Memory Objects.
3804
* See Memory for details about copy semantics, etc.
3808
class Buffer : public Memory
3812
/*! \brief Constructs a Buffer in a specified context.
3814
* Wraps clCreateBuffer().
3816
* \param host_ptr Storage to be used if the CL_MEM_USE_HOST_PTR flag was
3817
* specified. Note alignment & exclusivity requirements.
3820
const Context& context,
3823
void* host_ptr = NULL,
3827
object_ = ::clCreateBuffer(context(), flags, size, host_ptr, &error);
3829
detail::errHandler(error, __CREATE_BUFFER_ERR);
3835
/*! \brief Constructs a Buffer in the default context.
3837
* Wraps clCreateBuffer().
3839
* \param host_ptr Storage to be used if the CL_MEM_USE_HOST_PTR flag was
3840
* specified. Note alignment & exclusivity requirements.
3842
* \see Context::getDefault()
3847
void* host_ptr = NULL,
3852
Context context = Context::getDefault(err);
3854
object_ = ::clCreateBuffer(context(), flags, size, host_ptr, &error);
3856
detail::errHandler(error, __CREATE_BUFFER_ERR);
3863
* \brief Construct a Buffer from a host container via iterators.
3864
* IteratorType must be random access.
3865
* If useHostPtr is specified iterators must represent contiguous data.
3867
template< typename IteratorType >
3869
IteratorType startIterator,
3870
IteratorType endIterator,
3872
bool useHostPtr = false,
3875
typedef typename std::iterator_traits<IteratorType>::value_type DataType;
3878
cl_mem_flags flags = 0;
3880
flags |= CL_MEM_READ_ONLY;
3883
flags |= CL_MEM_READ_WRITE;
3886
flags |= CL_MEM_USE_HOST_PTR;
3889
size_type size = sizeof(DataType)*(endIterator - startIterator);
3891
Context context = Context::getDefault(err);
3894
object_ = ::clCreateBuffer(context(), flags, size, static_cast<DataType*>(&*startIterator), &error);
3896
object_ = ::clCreateBuffer(context(), flags, size, 0, &error);
3899
detail::errHandler(error, __CREATE_BUFFER_ERR);
3905
error = cl::copy(startIterator, endIterator, *this);
3906
detail::errHandler(error, __CREATE_BUFFER_ERR);
3914
* \brief Construct a Buffer from a host container via iterators using a specified context.
3915
* IteratorType must be random access.
3916
* If useHostPtr is specified iterators must represent contiguous data.
3918
template< typename IteratorType >
3919
Buffer(const Context &context, IteratorType startIterator, IteratorType endIterator,
3920
bool readOnly, bool useHostPtr = false, cl_int* err = NULL);
3923
* \brief Construct a Buffer from a host container via iterators using a specified queue.
3924
* If useHostPtr is specified iterators must be random access.
3926
template< typename IteratorType >
3927
Buffer(const CommandQueue &queue, IteratorType startIterator, IteratorType endIterator,
3928
bool readOnly, bool useHostPtr = false, cl_int* err = NULL);
3930
//! \brief Default constructor - initializes to NULL.
3931
Buffer() : Memory() { }
3933
/*! \brief Constructor from cl_mem - takes ownership.
3935
* \param retainObject will cause the constructor to retain its cl object.
3936
* Defaults to false to maintain compatibility with earlier versions.
3938
* See Memory for further details.
3940
explicit Buffer(const cl_mem& buffer, bool retainObject = false) :
3941
Memory(buffer, retainObject) { }
3943
/*! \brief Assignment from cl_mem - performs shallow copy.
3945
* See Memory for further details.
3947
Buffer& operator = (const cl_mem& rhs)
3949
Memory::operator=(rhs);
3953
/*! \brief Copy constructor to forward copy to the superclass correctly.
3954
* Required for MSVC.
3956
Buffer(const Buffer& buf) : Memory(buf) {}
3958
/*! \brief Copy assignment to forward copy to the superclass correctly.
3959
* Required for MSVC.
3961
Buffer& operator = (const Buffer &buf)
3963
Memory::operator=(buf);
3967
/*! \brief Move constructor to forward move to the superclass correctly.
3968
* Required for MSVC.
3970
Buffer(Buffer&& buf) CL_HPP_NOEXCEPT_ : Memory(std::move(buf)) {}
3972
/*! \brief Move assignment to forward move to the superclass correctly.
3973
* Required for MSVC.
3975
Buffer& operator = (Buffer &&buf)
3977
Memory::operator=(std::move(buf));
3981
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
3982
/*! \brief Creates a new buffer object from this.
3984
* Wraps clCreateSubBuffer().
3986
Buffer createSubBuffer(
3988
cl_buffer_create_type buffer_create_type,
3989
const void * buffer_create_info,
3990
cl_int * err = NULL)
3994
result.object_ = ::clCreateSubBuffer(
4001
detail::errHandler(error, __CREATE_SUBBUFFER_ERR);
4008
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 110
4011
#if defined (CL_HPP_USE_DX_INTEROP)
4012
/*! \brief Class interface for creating OpenCL buffers from ID3D10Buffer's.
4014
* This is provided to facilitate interoperability with Direct3D.
4016
* See Memory for details about copy semantics, etc.
4020
class BufferD3D10 : public Buffer
4025
/*! \brief Constructs a BufferD3D10, in a specified context, from a
4026
* given ID3D10Buffer.
4028
* Wraps clCreateFromD3D10BufferKHR().
4031
const Context& context,
4033
ID3D10Buffer* bufobj,
4034
cl_int * err = NULL) : pfn_clCreateFromD3D10BufferKHR(nullptr)
4036
typedef CL_API_ENTRY cl_mem (CL_API_CALL *PFN_clCreateFromD3D10BufferKHR)(
4037
cl_context context, cl_mem_flags flags, ID3D10Buffer* buffer,
4038
cl_int* errcode_ret);
4039
PFN_clCreateFromD3D10BufferKHR pfn_clCreateFromD3D10BufferKHR;
4040
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
4041
vector<cl_context_properties> props = context.getInfo<CL_CONTEXT_PROPERTIES>();
4042
cl_platform platform = -1;
4043
for( int i = 0; i < props.size(); ++i ) {
4044
if( props[i] == CL_CONTEXT_PLATFORM ) {
4045
platform = props[i+1];
4048
CL_HPP_INIT_CL_EXT_FCN_PTR_PLATFORM_(platform, clCreateFromD3D10BufferKHR);
4049
#elif CL_HPP_TARGET_OPENCL_VERSION >= 110
4050
CL_HPP_INIT_CL_EXT_FCN_PTR_(clCreateFromD3D10BufferKHR);
4054
object_ = pfn_clCreateFromD3D10BufferKHR(
4060
detail::errHandler(error, __CREATE_GL_BUFFER_ERR);
4066
//! \brief Default constructor - initializes to NULL.
4067
BufferD3D10() : Buffer() { }
4069
/*! \brief Constructor from cl_mem - takes ownership.
4071
* \param retainObject will cause the constructor to retain its cl object.
4072
* Defaults to false to maintain compatibility with
4074
* See Memory for further details.
4076
explicit BufferD3D10(const cl_mem& buffer, bool retainObject = false) :
4077
Buffer(buffer, retainObject) { }
4079
/*! \brief Assignment from cl_mem - performs shallow copy.
4081
* See Memory for further details.
4083
BufferD3D10& operator = (const cl_mem& rhs)
4085
Buffer::operator=(rhs);
4089
/*! \brief Copy constructor to forward copy to the superclass correctly.
4090
* Required for MSVC.
4092
BufferD3D10(const BufferD3D10& buf) :
4095
/*! \brief Copy assignment to forward copy to the superclass correctly.
4096
* Required for MSVC.
4098
BufferD3D10& operator = (const BufferD3D10 &buf)
4100
Buffer::operator=(buf);
4104
/*! \brief Move constructor to forward move to the superclass correctly.
4105
* Required for MSVC.
4107
BufferD3D10(BufferD3D10&& buf) CL_HPP_NOEXCEPT_ : Buffer(std::move(buf)) {}
4109
/*! \brief Move assignment to forward move to the superclass correctly.
4110
* Required for MSVC.
4112
BufferD3D10& operator = (BufferD3D10 &&buf)
4114
Buffer::operator=(std::move(buf));
4120
/*! \brief Class interface for GL Buffer Memory Objects.
4122
* This is provided to facilitate interoperability with OpenGL.
4124
* See Memory for details about copy semantics, etc.
4128
class BufferGL : public Buffer
4131
/*! \brief Constructs a BufferGL in a specified context, from a given
4134
* Wraps clCreateFromGLBuffer().
4137
const Context& context,
4140
cl_int * err = NULL)
4143
object_ = ::clCreateFromGLBuffer(
4149
detail::errHandler(error, __CREATE_GL_BUFFER_ERR);
4155
//! \brief Default constructor - initializes to NULL.
4156
BufferGL() : Buffer() { }
4158
/*! \brief Constructor from cl_mem - takes ownership.
4160
* \param retainObject will cause the constructor to retain its cl object.
4161
* Defaults to false to maintain compatibility with
4163
* See Memory for further details.
4165
explicit BufferGL(const cl_mem& buffer, bool retainObject = false) :
4166
Buffer(buffer, retainObject) { }
4168
/*! \brief Assignment from cl_mem - performs shallow copy.
4170
* See Memory for further details.
4172
BufferGL& operator = (const cl_mem& rhs)
4174
Buffer::operator=(rhs);
4178
/*! \brief Copy constructor to forward copy to the superclass correctly.
4179
* Required for MSVC.
4181
BufferGL(const BufferGL& buf) : Buffer(buf) {}
4183
/*! \brief Copy assignment to forward copy to the superclass correctly.
4184
* Required for MSVC.
4186
BufferGL& operator = (const BufferGL &buf)
4188
Buffer::operator=(buf);
4192
/*! \brief Move constructor to forward move to the superclass correctly.
4193
* Required for MSVC.
4195
BufferGL(BufferGL&& buf) CL_HPP_NOEXCEPT_ : Buffer(std::move(buf)) {}
4197
/*! \brief Move assignment to forward move to the superclass correctly.
4198
* Required for MSVC.
4200
BufferGL& operator = (BufferGL &&buf)
4202
Buffer::operator=(std::move(buf));
4206
//! \brief Wrapper for clGetGLObjectInfo().
4207
cl_int getObjectInfo(
4208
cl_gl_object_type *type,
4209
cl_GLuint * gl_object_name)
4211
return detail::errHandler(
4212
::clGetGLObjectInfo(object_,type,gl_object_name),
4213
__GET_GL_OBJECT_INFO_ERR);
4217
/*! \brief Class interface for GL Render Buffer Memory Objects.
4219
* This is provided to facilitate interoperability with OpenGL.
4221
* See Memory for details about copy semantics, etc.
4225
class BufferRenderGL : public Buffer
4228
/*! \brief Constructs a BufferRenderGL in a specified context, from a given
4231
* Wraps clCreateFromGLRenderbuffer().
4234
const Context& context,
4237
cl_int * err = NULL)
4240
object_ = ::clCreateFromGLRenderbuffer(
4246
detail::errHandler(error, __CREATE_GL_RENDER_BUFFER_ERR);
4252
//! \brief Default constructor - initializes to NULL.
4253
BufferRenderGL() : Buffer() { }
4255
/*! \brief Constructor from cl_mem - takes ownership.
4257
* \param retainObject will cause the constructor to retain its cl object.
4258
* Defaults to false to maintain compatibility with
4260
* See Memory for further details.
4262
explicit BufferRenderGL(const cl_mem& buffer, bool retainObject = false) :
4263
Buffer(buffer, retainObject) { }
4265
/*! \brief Assignment from cl_mem - performs shallow copy.
4267
* See Memory for further details.
4269
BufferRenderGL& operator = (const cl_mem& rhs)
4271
Buffer::operator=(rhs);
4275
/*! \brief Copy constructor to forward copy to the superclass correctly.
4276
* Required for MSVC.
4278
BufferRenderGL(const BufferRenderGL& buf) : Buffer(buf) {}
4280
/*! \brief Copy assignment to forward copy to the superclass correctly.
4281
* Required for MSVC.
4283
BufferRenderGL& operator = (const BufferRenderGL &buf)
4285
Buffer::operator=(buf);
4289
/*! \brief Move constructor to forward move to the superclass correctly.
4290
* Required for MSVC.
4292
BufferRenderGL(BufferRenderGL&& buf) CL_HPP_NOEXCEPT_ : Buffer(std::move(buf)) {}
4294
/*! \brief Move assignment to forward move to the superclass correctly.
4295
* Required for MSVC.
4297
BufferRenderGL& operator = (BufferRenderGL &&buf)
4299
Buffer::operator=(std::move(buf));
4303
//! \brief Wrapper for clGetGLObjectInfo().
4304
cl_int getObjectInfo(
4305
cl_gl_object_type *type,
4306
cl_GLuint * gl_object_name)
4308
return detail::errHandler(
4309
::clGetGLObjectInfo(object_,type,gl_object_name),
4310
__GET_GL_OBJECT_INFO_ERR);
4314
/*! \brief C++ base class for Image Memory objects.
4316
* See Memory for details about copy semantics, etc.
4320
class Image : public Memory
4323
//! \brief Default constructor - initializes to NULL.
4324
Image() : Memory() { }
4326
/*! \brief Constructor from cl_mem - takes ownership.
4328
* \param retainObject will cause the constructor to retain its cl object.
4329
* Defaults to false to maintain compatibility with
4331
* See Memory for further details.
4333
explicit Image(const cl_mem& image, bool retainObject = false) :
4334
Memory(image, retainObject) { }
4336
/*! \brief Assignment from cl_mem - performs shallow copy.
4338
* See Memory for further details.
4340
Image& operator = (const cl_mem& rhs)
4342
Memory::operator=(rhs);
4346
/*! \brief Copy constructor to forward copy to the superclass correctly.
4347
* Required for MSVC.
4349
Image(const Image& img) : Memory(img) {}
4351
/*! \brief Copy assignment to forward copy to the superclass correctly.
4352
* Required for MSVC.
4354
Image& operator = (const Image &img)
4356
Memory::operator=(img);
4360
/*! \brief Move constructor to forward move to the superclass correctly.
4361
* Required for MSVC.
4363
Image(Image&& img) CL_HPP_NOEXCEPT_ : Memory(std::move(img)) {}
4365
/*! \brief Move assignment to forward move to the superclass correctly.
4366
* Required for MSVC.
4368
Image& operator = (Image &&img)
4370
Memory::operator=(std::move(img));
4376
//! \brief Wrapper for clGetImageInfo().
4377
template <typename T>
4378
cl_int getImageInfo(cl_image_info name, T* param) const
4380
return detail::errHandler(
4381
detail::getInfo(&::clGetImageInfo, object_, name, param),
4382
__GET_IMAGE_INFO_ERR);
4385
//! \brief Wrapper for clGetImageInfo() that returns by value.
4386
template <cl_int name> typename
4387
detail::param_traits<detail::cl_image_info, name>::param_type
4388
getImageInfo(cl_int* err = NULL) const
4390
typename detail::param_traits<
4391
detail::cl_image_info, name>::param_type param;
4392
cl_int result = getImageInfo(name, ¶m);
4400
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
4401
/*! \brief Class interface for 1D Image Memory objects.
4403
* See Memory for details about copy semantics, etc.
4407
class Image1D : public Image
4410
/*! \brief Constructs a 1D Image in a specified context.
4412
* Wraps clCreateImage().
4415
const Context& context,
4419
void* host_ptr = NULL,
4423
cl_image_desc desc =
4425
CL_MEM_OBJECT_IMAGE1D,
4427
0, 0, 0, 0, 0, 0, 0, 0
4429
object_ = ::clCreateImage(
4437
detail::errHandler(error, __CREATE_IMAGE_ERR);
4443
//! \brief Default constructor - initializes to NULL.
4446
/*! \brief Constructor from cl_mem - takes ownership.
4448
* \param retainObject will cause the constructor to retain its cl object.
4449
* Defaults to false to maintain compatibility with
4451
* See Memory for further details.
4453
explicit Image1D(const cl_mem& image1D, bool retainObject = false) :
4454
Image(image1D, retainObject) { }
4456
/*! \brief Assignment from cl_mem - performs shallow copy.
4458
* See Memory for further details.
4460
Image1D& operator = (const cl_mem& rhs)
4462
Image::operator=(rhs);
4466
/*! \brief Copy constructor to forward copy to the superclass correctly.
4467
* Required for MSVC.
4469
Image1D(const Image1D& img) : Image(img) {}
4471
/*! \brief Copy assignment to forward copy to the superclass correctly.
4472
* Required for MSVC.
4474
Image1D& operator = (const Image1D &img)
4476
Image::operator=(img);
4480
/*! \brief Move constructor to forward move to the superclass correctly.
4481
* Required for MSVC.
4483
Image1D(Image1D&& img) CL_HPP_NOEXCEPT_ : Image(std::move(img)) {}
4485
/*! \brief Move assignment to forward move to the superclass correctly.
4486
* Required for MSVC.
4488
Image1D& operator = (Image1D &&img)
4490
Image::operator=(std::move(img));
4496
/*! \class Image1DBuffer
4497
* \brief Image interface for 1D buffer images.
4499
class Image1DBuffer : public Image
4503
const Context& context,
4507
const Buffer &buffer,
4511
cl_image_desc desc =
4513
CL_MEM_OBJECT_IMAGE1D_BUFFER,
4515
0, 0, 0, 0, 0, 0, 0,
4518
object_ = ::clCreateImage(
4526
detail::errHandler(error, __CREATE_IMAGE_ERR);
4534
/*! \brief Constructor from cl_mem - takes ownership.
4536
* \param retainObject will cause the constructor to retain its cl object.
4537
* Defaults to false to maintain compatibility with
4539
* See Memory for further details.
4541
explicit Image1DBuffer(const cl_mem& image1D, bool retainObject = false) :
4542
Image(image1D, retainObject) { }
4544
Image1DBuffer& operator = (const cl_mem& rhs)
4546
Image::operator=(rhs);
4550
/*! \brief Copy constructor to forward copy to the superclass correctly.
4551
* Required for MSVC.
4553
Image1DBuffer(const Image1DBuffer& img) : Image(img) {}
4555
/*! \brief Copy assignment to forward copy to the superclass correctly.
4556
* Required for MSVC.
4558
Image1DBuffer& operator = (const Image1DBuffer &img)
4560
Image::operator=(img);
4564
/*! \brief Move constructor to forward move to the superclass correctly.
4565
* Required for MSVC.
4567
Image1DBuffer(Image1DBuffer&& img) CL_HPP_NOEXCEPT_ : Image(std::move(img)) {}
4569
/*! \brief Move assignment to forward move to the superclass correctly.
4570
* Required for MSVC.
4572
Image1DBuffer& operator = (Image1DBuffer &&img)
4574
Image::operator=(std::move(img));
4580
/*! \class Image1DArray
4581
* \brief Image interface for arrays of 1D images.
4583
class Image1DArray : public Image
4587
const Context& context,
4590
size_type arraySize,
4593
void* host_ptr = NULL,
4597
cl_image_desc desc =
4599
CL_MEM_OBJECT_IMAGE1D_ARRAY,
4601
0, 0, // height, depth (unused)
4606
object_ = ::clCreateImage(
4614
detail::errHandler(error, __CREATE_IMAGE_ERR);
4622
/*! \brief Constructor from cl_mem - takes ownership.
4624
* \param retainObject will cause the constructor to retain its cl object.
4625
* Defaults to false to maintain compatibility with
4627
* See Memory for further details.
4629
explicit Image1DArray(const cl_mem& imageArray, bool retainObject = false) :
4630
Image(imageArray, retainObject) { }
4633
Image1DArray& operator = (const cl_mem& rhs)
4635
Image::operator=(rhs);
4639
/*! \brief Copy constructor to forward copy to the superclass correctly.
4640
* Required for MSVC.
4642
Image1DArray(const Image1DArray& img) : Image(img) {}
4644
/*! \brief Copy assignment to forward copy to the superclass correctly.
4645
* Required for MSVC.
4647
Image1DArray& operator = (const Image1DArray &img)
4649
Image::operator=(img);
4653
/*! \brief Move constructor to forward move to the superclass correctly.
4654
* Required for MSVC.
4656
Image1DArray(Image1DArray&& img) CL_HPP_NOEXCEPT_ : Image(std::move(img)) {}
4658
/*! \brief Move assignment to forward move to the superclass correctly.
4659
* Required for MSVC.
4661
Image1DArray& operator = (Image1DArray &&img)
4663
Image::operator=(std::move(img));
4668
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 120
4671
/*! \brief Class interface for 2D Image Memory objects.
4673
* See Memory for details about copy semantics, etc.
4677
class Image2D : public Image
4680
/*! \brief Constructs a 2D Image in a specified context.
4682
* Wraps clCreateImage().
4685
const Context& context,
4690
size_type row_pitch = 0,
4691
void* host_ptr = NULL,
4695
bool useCreateImage;
4697
#if CL_HPP_TARGET_OPENCL_VERSION >= 120 && CL_HPP_MINIMUM_OPENCL_VERSION < 120
4698
// Run-time decision based on the actual platform
4700
cl_uint version = detail::getContextPlatformVersion(context());
4701
useCreateImage = (version >= 0x10002); // OpenCL 1.2 or above
4703
#elif CL_HPP_TARGET_OPENCL_VERSION >= 120
4704
useCreateImage = true;
4706
useCreateImage = false;
4709
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
4712
cl_image_desc desc =
4714
CL_MEM_OBJECT_IMAGE2D,
4717
0, 0, // depth, array size (unused)
4721
object_ = ::clCreateImage(
4729
detail::errHandler(error, __CREATE_IMAGE_ERR);
4734
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
4735
#if CL_HPP_MINIMUM_OPENCL_VERSION < 120
4736
if (!useCreateImage)
4738
object_ = ::clCreateImage2D(
4739
context(), flags,&format, width, height, row_pitch, host_ptr, &error);
4741
detail::errHandler(error, __CREATE_IMAGE2D_ERR);
4746
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 120
4749
#if CL_HPP_TARGET_OPENCL_VERSION >= 200 || defined(CL_HPP_USE_CL_IMAGE2D_FROM_BUFFER_KHR)
4750
/*! \brief Constructs a 2D Image from a buffer.
4751
* \note This will share storage with the underlying buffer.
4753
* Wraps clCreateImage().
4756
const Context& context,
4758
const Buffer &sourceBuffer,
4761
size_type row_pitch = 0,
4762
cl_int* err = nullptr)
4766
cl_image_desc desc =
4768
CL_MEM_OBJECT_IMAGE2D,
4771
0, 0, // depth, array size (unused)
4774
// Use buffer as input to image
4777
object_ = ::clCreateImage(
4779
0, // flags inherited from buffer
4785
detail::errHandler(error, __CREATE_IMAGE_ERR);
4786
if (err != nullptr) {
4790
#endif //#if CL_HPP_TARGET_OPENCL_VERSION >= 200 || defined(CL_HPP_USE_CL_IMAGE2D_FROM_BUFFER_KHR)
4792
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
4793
/*! \brief Constructs a 2D Image from an image.
4794
* \note This will share storage with the underlying image but may
4795
* reinterpret the channel order and type.
4797
* The image will be created matching with a descriptor matching the source.
4799
* \param order is the channel order to reinterpret the image data as.
4800
* The channel order may differ as described in the OpenCL
4801
* 2.0 API specification.
4803
* Wraps clCreateImage().
4806
const Context& context,
4807
cl_channel_order order,
4808
const Image &sourceImage,
4809
cl_int* err = nullptr)
4813
// Descriptor fields have to match source image
4814
size_type sourceWidth =
4815
sourceImage.getImageInfo<CL_IMAGE_WIDTH>();
4816
size_type sourceHeight =
4817
sourceImage.getImageInfo<CL_IMAGE_HEIGHT>();
4818
size_type sourceRowPitch =
4819
sourceImage.getImageInfo<CL_IMAGE_ROW_PITCH>();
4820
cl_uint sourceNumMIPLevels =
4821
sourceImage.getImageInfo<CL_IMAGE_NUM_MIP_LEVELS>();
4822
cl_uint sourceNumSamples =
4823
sourceImage.getImageInfo<CL_IMAGE_NUM_SAMPLES>();
4824
cl_image_format sourceFormat =
4825
sourceImage.getImageInfo<CL_IMAGE_FORMAT>();
4827
// Update only the channel order.
4828
// Channel format inherited from source.
4829
sourceFormat.image_channel_order = order;
4830
cl_image_desc desc =
4832
CL_MEM_OBJECT_IMAGE2D,
4835
0, 0, // depth (unused), array size (unused)
4837
0, // slice pitch (unused)
4840
// Use buffer as input to image
4843
object_ = ::clCreateImage(
4845
0, // flags should be inherited from mem_object
4851
detail::errHandler(error, __CREATE_IMAGE_ERR);
4852
if (err != nullptr) {
4856
#endif //#if CL_HPP_TARGET_OPENCL_VERSION >= 200
4858
//! \brief Default constructor - initializes to NULL.
4861
/*! \brief Constructor from cl_mem - takes ownership.
4863
* \param retainObject will cause the constructor to retain its cl object.
4864
* Defaults to false to maintain compatibility with
4866
* See Memory for further details.
4868
explicit Image2D(const cl_mem& image2D, bool retainObject = false) :
4869
Image(image2D, retainObject) { }
4871
/*! \brief Assignment from cl_mem - performs shallow copy.
4873
* See Memory for further details.
4875
Image2D& operator = (const cl_mem& rhs)
4877
Image::operator=(rhs);
4881
/*! \brief Copy constructor to forward copy to the superclass correctly.
4882
* Required for MSVC.
4884
Image2D(const Image2D& img) : Image(img) {}
4886
/*! \brief Copy assignment to forward copy to the superclass correctly.
4887
* Required for MSVC.
4889
Image2D& operator = (const Image2D &img)
4891
Image::operator=(img);
4895
/*! \brief Move constructor to forward move to the superclass correctly.
4896
* Required for MSVC.
4898
Image2D(Image2D&& img) CL_HPP_NOEXCEPT_ : Image(std::move(img)) {}
4900
/*! \brief Move assignment to forward move to the superclass correctly.
4901
* Required for MSVC.
4903
Image2D& operator = (Image2D &&img)
4905
Image::operator=(std::move(img));
4912
#if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
4913
/*! \brief Class interface for GL 2D Image Memory objects.
4915
* This is provided to facilitate interoperability with OpenGL.
4917
* See Memory for details about copy semantics, etc.
4920
* \note Deprecated for OpenCL 1.2. Please use ImageGL instead.
4922
class CL_EXT_PREFIX__VERSION_1_1_DEPRECATED Image2DGL : public Image2D
4925
/*! \brief Constructs an Image2DGL in a specified context, from a given
4928
* Wraps clCreateFromGLTexture2D().
4931
const Context& context,
4936
cl_int * err = NULL)
4939
object_ = ::clCreateFromGLTexture2D(
4947
detail::errHandler(error, __CREATE_GL_TEXTURE_2D_ERR);
4954
//! \brief Default constructor - initializes to NULL.
4955
Image2DGL() : Image2D() { }
4957
/*! \brief Constructor from cl_mem - takes ownership.
4959
* \param retainObject will cause the constructor to retain its cl object.
4960
* Defaults to false to maintain compatibility with
4962
* See Memory for further details.
4964
explicit Image2DGL(const cl_mem& image, bool retainObject = false) :
4965
Image2D(image, retainObject) { }
4967
/*! \brief Assignment from cl_mem - performs shallow copy.
4969
* See Memory for further details.
4971
Image2DGL& operator = (const cl_mem& rhs)
4973
Image2D::operator=(rhs);
4977
/*! \brief Copy constructor to forward copy to the superclass correctly.
4978
* Required for MSVC.
4980
Image2DGL(const Image2DGL& img) : Image2D(img) {}
4982
/*! \brief Copy assignment to forward copy to the superclass correctly.
4983
* Required for MSVC.
4985
Image2DGL& operator = (const Image2DGL &img)
4987
Image2D::operator=(img);
4991
/*! \brief Move constructor to forward move to the superclass correctly.
4992
* Required for MSVC.
4994
Image2DGL(Image2DGL&& img) CL_HPP_NOEXCEPT_ : Image2D(std::move(img)) {}
4996
/*! \brief Move assignment to forward move to the superclass correctly.
4997
* Required for MSVC.
4999
Image2DGL& operator = (Image2DGL &&img)
5001
Image2D::operator=(std::move(img));
5005
} CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED;
5006
#endif // CL_USE_DEPRECATED_OPENCL_1_1_APIS
5008
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
5009
/*! \class Image2DArray
5010
* \brief Image interface for arrays of 2D images.
5012
class Image2DArray : public Image
5016
const Context& context,
5019
size_type arraySize,
5023
size_type slicePitch,
5024
void* host_ptr = NULL,
5028
cl_image_desc desc =
5030
CL_MEM_OBJECT_IMAGE2D_ARRAY,
5033
0, // depth (unused)
5039
object_ = ::clCreateImage(
5047
detail::errHandler(error, __CREATE_IMAGE_ERR);
5055
/*! \brief Constructor from cl_mem - takes ownership.
5057
* \param retainObject will cause the constructor to retain its cl object.
5058
* Defaults to false to maintain compatibility with
5060
* See Memory for further details.
5062
explicit Image2DArray(const cl_mem& imageArray, bool retainObject = false) : Image(imageArray, retainObject) { }
5064
Image2DArray& operator = (const cl_mem& rhs)
5066
Image::operator=(rhs);
5070
/*! \brief Copy constructor to forward copy to the superclass correctly.
5071
* Required for MSVC.
5073
Image2DArray(const Image2DArray& img) : Image(img) {}
5075
/*! \brief Copy assignment to forward copy to the superclass correctly.
5076
* Required for MSVC.
5078
Image2DArray& operator = (const Image2DArray &img)
5080
Image::operator=(img);
5084
/*! \brief Move constructor to forward move to the superclass correctly.
5085
* Required for MSVC.
5087
Image2DArray(Image2DArray&& img) CL_HPP_NOEXCEPT_ : Image(std::move(img)) {}
5089
/*! \brief Move assignment to forward move to the superclass correctly.
5090
* Required for MSVC.
5092
Image2DArray& operator = (Image2DArray &&img)
5094
Image::operator=(std::move(img));
5098
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 120
5100
/*! \brief Class interface for 3D Image Memory objects.
5102
* See Memory for details about copy semantics, etc.
5106
class Image3D : public Image
5109
/*! \brief Constructs a 3D Image in a specified context.
5111
* Wraps clCreateImage().
5114
const Context& context,
5120
size_type row_pitch = 0,
5121
size_type slice_pitch = 0,
5122
void* host_ptr = NULL,
5126
bool useCreateImage;
5128
#if CL_HPP_TARGET_OPENCL_VERSION >= 120 && CL_HPP_MINIMUM_OPENCL_VERSION < 120
5129
// Run-time decision based on the actual platform
5131
cl_uint version = detail::getContextPlatformVersion(context());
5132
useCreateImage = (version >= 0x10002); // OpenCL 1.2 or above
5134
#elif CL_HPP_TARGET_OPENCL_VERSION >= 120
5135
useCreateImage = true;
5137
useCreateImage = false;
5140
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
5143
cl_image_desc desc =
5145
CL_MEM_OBJECT_IMAGE3D,
5149
0, // array size (unused)
5154
object_ = ::clCreateImage(
5162
detail::errHandler(error, __CREATE_IMAGE_ERR);
5167
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
5168
#if CL_HPP_MINIMUM_OPENCL_VERSION < 120
5169
if (!useCreateImage)
5171
object_ = ::clCreateImage3D(
5172
context(), flags, &format, width, height, depth, row_pitch,
5173
slice_pitch, host_ptr, &error);
5175
detail::errHandler(error, __CREATE_IMAGE3D_ERR);
5180
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 120
5183
//! \brief Default constructor - initializes to NULL.
5184
Image3D() : Image() { }
5186
/*! \brief Constructor from cl_mem - takes ownership.
5188
* \param retainObject will cause the constructor to retain its cl object.
5189
* Defaults to false to maintain compatibility with
5191
* See Memory for further details.
5193
explicit Image3D(const cl_mem& image3D, bool retainObject = false) :
5194
Image(image3D, retainObject) { }
5196
/*! \brief Assignment from cl_mem - performs shallow copy.
5198
* See Memory for further details.
5200
Image3D& operator = (const cl_mem& rhs)
5202
Image::operator=(rhs);
5206
/*! \brief Copy constructor to forward copy to the superclass correctly.
5207
* Required for MSVC.
5209
Image3D(const Image3D& img) : Image(img) {}
5211
/*! \brief Copy assignment to forward copy to the superclass correctly.
5212
* Required for MSVC.
5214
Image3D& operator = (const Image3D &img)
5216
Image::operator=(img);
5220
/*! \brief Move constructor to forward move to the superclass correctly.
5221
* Required for MSVC.
5223
Image3D(Image3D&& img) CL_HPP_NOEXCEPT_ : Image(std::move(img)) {}
5225
/*! \brief Move assignment to forward move to the superclass correctly.
5226
* Required for MSVC.
5228
Image3D& operator = (Image3D &&img)
5230
Image::operator=(std::move(img));
5235
#if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
5236
/*! \brief Class interface for GL 3D Image Memory objects.
5238
* This is provided to facilitate interoperability with OpenGL.
5240
* See Memory for details about copy semantics, etc.
5244
class Image3DGL : public Image3D
5247
/*! \brief Constructs an Image3DGL in a specified context, from a given
5250
* Wraps clCreateFromGLTexture3D().
5253
const Context& context,
5258
cl_int * err = NULL)
5261
object_ = ::clCreateFromGLTexture3D(
5269
detail::errHandler(error, __CREATE_GL_TEXTURE_3D_ERR);
5275
//! \brief Default constructor - initializes to NULL.
5276
Image3DGL() : Image3D() { }
5278
/*! \brief Constructor from cl_mem - takes ownership.
5280
* \param retainObject will cause the constructor to retain its cl object.
5281
* Defaults to false to maintain compatibility with
5283
* See Memory for further details.
5285
explicit Image3DGL(const cl_mem& image, bool retainObject = false) :
5286
Image3D(image, retainObject) { }
5288
/*! \brief Assignment from cl_mem - performs shallow copy.
5290
* See Memory for further details.
5292
Image3DGL& operator = (const cl_mem& rhs)
5294
Image3D::operator=(rhs);
5298
/*! \brief Copy constructor to forward copy to the superclass correctly.
5299
* Required for MSVC.
5301
Image3DGL(const Image3DGL& img) : Image3D(img) {}
5303
/*! \brief Copy assignment to forward copy to the superclass correctly.
5304
* Required for MSVC.
5306
Image3DGL& operator = (const Image3DGL &img)
5308
Image3D::operator=(img);
5312
/*! \brief Move constructor to forward move to the superclass correctly.
5313
* Required for MSVC.
5315
Image3DGL(Image3DGL&& img) CL_HPP_NOEXCEPT_ : Image3D(std::move(img)) {}
5317
/*! \brief Move assignment to forward move to the superclass correctly.
5318
* Required for MSVC.
5320
Image3DGL& operator = (Image3DGL &&img)
5322
Image3D::operator=(std::move(img));
5326
#endif // CL_USE_DEPRECATED_OPENCL_1_1_APIS
5328
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
5330
* \brief general image interface for GL interop.
5331
* We abstract the 2D and 3D GL images into a single instance here
5332
* that wraps all GL sourced images on the grounds that setup information
5333
* was performed by OpenCL anyway.
5335
class ImageGL : public Image
5339
const Context& context,
5344
cl_int * err = NULL)
5347
object_ = ::clCreateFromGLTexture(
5355
detail::errHandler(error, __CREATE_GL_TEXTURE_ERR);
5361
ImageGL() : Image() { }
5363
/*! \brief Constructor from cl_mem - takes ownership.
5365
* \param retainObject will cause the constructor to retain its cl object.
5366
* Defaults to false to maintain compatibility with
5368
* See Memory for further details.
5370
explicit ImageGL(const cl_mem& image, bool retainObject = false) :
5371
Image(image, retainObject) { }
5373
ImageGL& operator = (const cl_mem& rhs)
5375
Image::operator=(rhs);
5379
/*! \brief Copy constructor to forward copy to the superclass correctly.
5380
* Required for MSVC.
5382
ImageGL(const ImageGL& img) : Image(img) {}
5384
/*! \brief Copy assignment to forward copy to the superclass correctly.
5385
* Required for MSVC.
5387
ImageGL& operator = (const ImageGL &img)
5389
Image::operator=(img);
5393
/*! \brief Move constructor to forward move to the superclass correctly.
5394
* Required for MSVC.
5396
ImageGL(ImageGL&& img) CL_HPP_NOEXCEPT_ : Image(std::move(img)) {}
5398
/*! \brief Move assignment to forward move to the superclass correctly.
5399
* Required for MSVC.
5401
ImageGL& operator = (ImageGL &&img)
5403
Image::operator=(std::move(img));
5407
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
5411
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
5412
/*! \brief Class interface for Pipe Memory Objects.
5414
* See Memory for details about copy semantics, etc.
5418
class Pipe : public Memory
5422
/*! \brief Constructs a Pipe in a specified context.
5424
* Wraps clCreatePipe().
5425
* @param context Context in which to create the pipe.
5426
* @param flags Bitfield. Only CL_MEM_READ_WRITE and CL_MEM_HOST_NO_ACCESS are valid.
5427
* @param packet_size Size in bytes of a single packet of the pipe.
5428
* @param max_packets Number of packets that may be stored in the pipe.
5432
const Context& context,
5433
cl_uint packet_size,
5434
cl_uint max_packets,
5439
cl_mem_flags flags = CL_MEM_READ_WRITE | CL_MEM_HOST_NO_ACCESS;
5440
object_ = ::clCreatePipe(context(), flags, packet_size, max_packets, nullptr, &error);
5442
detail::errHandler(error, __CREATE_PIPE_ERR);
5448
/*! \brief Constructs a Pipe in a the default context.
5450
* Wraps clCreatePipe().
5451
* @param flags Bitfield. Only CL_MEM_READ_WRITE and CL_MEM_HOST_NO_ACCESS are valid.
5452
* @param packet_size Size in bytes of a single packet of the pipe.
5453
* @param max_packets Number of packets that may be stored in the pipe.
5457
cl_uint packet_size,
5458
cl_uint max_packets,
5463
Context context = Context::getDefault(err);
5465
cl_mem_flags flags = CL_MEM_READ_WRITE | CL_MEM_HOST_NO_ACCESS;
5466
object_ = ::clCreatePipe(context(), flags, packet_size, max_packets, nullptr, &error);
5468
detail::errHandler(error, __CREATE_PIPE_ERR);
5474
//! \brief Default constructor - initializes to NULL.
5475
Pipe() : Memory() { }
5477
/*! \brief Constructor from cl_mem - takes ownership.
5479
* \param retainObject will cause the constructor to retain its cl object.
5480
* Defaults to false to maintain compatibility with earlier versions.
5482
* See Memory for further details.
5484
explicit Pipe(const cl_mem& pipe, bool retainObject = false) :
5485
Memory(pipe, retainObject) { }
5487
/*! \brief Assignment from cl_mem - performs shallow copy.
5489
* See Memory for further details.
5491
Pipe& operator = (const cl_mem& rhs)
5493
Memory::operator=(rhs);
5497
/*! \brief Copy constructor to forward copy to the superclass correctly.
5498
* Required for MSVC.
5500
Pipe(const Pipe& pipe) : Memory(pipe) {}
5502
/*! \brief Copy assignment to forward copy to the superclass correctly.
5503
* Required for MSVC.
5505
Pipe& operator = (const Pipe &pipe)
5507
Memory::operator=(pipe);
5511
/*! \brief Move constructor to forward move to the superclass correctly.
5512
* Required for MSVC.
5514
Pipe(Pipe&& pipe) CL_HPP_NOEXCEPT_ : Memory(std::move(pipe)) {}
5516
/*! \brief Move assignment to forward move to the superclass correctly.
5517
* Required for MSVC.
5519
Pipe& operator = (Pipe &&pipe)
5521
Memory::operator=(std::move(pipe));
5525
//! \brief Wrapper for clGetMemObjectInfo().
5526
template <typename T>
5527
cl_int getInfo(cl_pipe_info name, T* param) const
5529
return detail::errHandler(
5530
detail::getInfo(&::clGetPipeInfo, object_, name, param),
5531
__GET_PIPE_INFO_ERR);
5534
//! \brief Wrapper for clGetMemObjectInfo() that returns by value.
5535
template <cl_int name> typename
5536
detail::param_traits<detail::cl_pipe_info, name>::param_type
5537
getInfo(cl_int* err = NULL) const
5539
typename detail::param_traits<
5540
detail::cl_pipe_info, name>::param_type param;
5541
cl_int result = getInfo(name, ¶m);
5548
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
5551
/*! \brief Class interface for cl_sampler.
5553
* \note Copies of these objects are shallow, meaning that the copy will refer
5554
* to the same underlying cl_sampler as the original. For details, see
5555
* clRetainSampler() and clReleaseSampler().
5559
class Sampler : public detail::Wrapper<cl_sampler>
5562
//! \brief Default constructor - initializes to NULL.
5565
/*! \brief Constructs a Sampler in a specified context.
5567
* Wraps clCreateSampler().
5570
const Context& context,
5571
cl_bool normalized_coords,
5572
cl_addressing_mode addressing_mode,
5573
cl_filter_mode filter_mode,
5578
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
5579
cl_sampler_properties sampler_properties[] = {
5580
CL_SAMPLER_NORMALIZED_COORDS, normalized_coords,
5581
CL_SAMPLER_ADDRESSING_MODE, addressing_mode,
5582
CL_SAMPLER_FILTER_MODE, filter_mode,
5584
object_ = ::clCreateSamplerWithProperties(
5589
detail::errHandler(error, __CREATE_SAMPLER_WITH_PROPERTIES_ERR);
5594
object_ = ::clCreateSampler(
5601
detail::errHandler(error, __CREATE_SAMPLER_ERR);
5608
/*! \brief Constructor from cl_sampler - takes ownership.
5610
* \param retainObject will cause the constructor to retain its cl object.
5611
* Defaults to false to maintain compatibility with
5613
* This effectively transfers ownership of a refcount on the cl_sampler
5614
* into the new Sampler object.
5616
explicit Sampler(const cl_sampler& sampler, bool retainObject = false) :
5617
detail::Wrapper<cl_type>(sampler, retainObject) { }
5619
/*! \brief Assignment operator from cl_sampler - takes ownership.
5621
* This effectively transfers ownership of a refcount on the rhs and calls
5622
* clReleaseSampler() on the value previously held by this instance.
5624
Sampler& operator = (const cl_sampler& rhs)
5626
detail::Wrapper<cl_type>::operator=(rhs);
5630
/*! \brief Copy constructor to forward copy to the superclass correctly.
5631
* Required for MSVC.
5633
Sampler(const Sampler& sam) : detail::Wrapper<cl_type>(sam) {}
5635
/*! \brief Copy assignment to forward copy to the superclass correctly.
5636
* Required for MSVC.
5638
Sampler& operator = (const Sampler &sam)
5640
detail::Wrapper<cl_type>::operator=(sam);
5644
/*! \brief Move constructor to forward move to the superclass correctly.
5645
* Required for MSVC.
5647
Sampler(Sampler&& sam) CL_HPP_NOEXCEPT_ : detail::Wrapper<cl_type>(std::move(sam)) {}
5649
/*! \brief Move assignment to forward move to the superclass correctly.
5650
* Required for MSVC.
5652
Sampler& operator = (Sampler &&sam)
5654
detail::Wrapper<cl_type>::operator=(std::move(sam));
5658
//! \brief Wrapper for clGetSamplerInfo().
5659
template <typename T>
5660
cl_int getInfo(cl_sampler_info name, T* param) const
5662
return detail::errHandler(
5663
detail::getInfo(&::clGetSamplerInfo, object_, name, param),
5664
__GET_SAMPLER_INFO_ERR);
5667
//! \brief Wrapper for clGetSamplerInfo() that returns by value.
5668
template <cl_int name> typename
5669
detail::param_traits<detail::cl_sampler_info, name>::param_type
5670
getInfo(cl_int* err = NULL) const
5672
typename detail::param_traits<
5673
detail::cl_sampler_info, name>::param_type param;
5674
cl_int result = getInfo(name, ¶m);
5684
class DeviceCommandQueue;
5687
//! \brief Class interface for specifying NDRange values.
5691
size_type sizes_[3];
5692
cl_uint dimensions_;
5695
//! \brief Default constructor - resulting range has zero dimensions.
5704
//! \brief Constructs one-dimensional range.
5705
NDRange(size_type size0)
5713
//! \brief Constructs two-dimensional range.
5714
NDRange(size_type size0, size_type size1)
5722
//! \brief Constructs three-dimensional range.
5723
NDRange(size_type size0, size_type size1, size_type size2)
5731
/*! \brief Conversion operator to const size_type *.
5733
* \returns a pointer to the size of the first dimension.
5735
operator const size_type*() const {
5739
//! \brief Queries the number of dimensions in the range.
5740
size_type dimensions() const
5745
//! \brief Returns the size of the object in bytes based on the
5746
// runtime number of dimensions
5747
size_type size() const
5749
return dimensions_*sizeof(size_type);
5757
const size_type* get() const
5763
//! \brief A zero-dimensional range.
5764
static const NDRange NullRange;
5766
//! \brief Local address wrapper for use with Kernel::setArg
5767
struct LocalSpaceArg
5774
template <typename T, class Enable = void>
5775
struct KernelArgumentHandler;
5777
// Enable for objects that are not subclasses of memory
5778
// Pointers, constants etc
5779
template <typename T>
5780
struct KernelArgumentHandler<T, typename std::enable_if<!std::is_base_of<cl::Memory, T>::value>::type>
5782
static size_type size(const T&) { return sizeof(T); }
5783
static const T* ptr(const T& value) { return &value; }
5786
// Enable for subclasses of memory where we want to get a reference to the cl_mem out
5787
// and pass that in for safety
5788
template <typename T>
5789
struct KernelArgumentHandler<T, typename std::enable_if<std::is_base_of<cl::Memory, T>::value>::type>
5791
static size_type size(const T&) { return sizeof(cl_mem); }
5792
static const cl_mem* ptr(const T& value) { return &(value()); }
5795
// Specialization for DeviceCommandQueue defined later
5798
struct KernelArgumentHandler<LocalSpaceArg, void>
5800
static size_type size(const LocalSpaceArg& value) { return value.size_; }
5801
static const void* ptr(const LocalSpaceArg&) { return NULL; }
5808
* \brief Helper function for generating LocalSpaceArg objects.
5810
inline LocalSpaceArg
5811
Local(size_type size)
5813
LocalSpaceArg ret = { size };
5817
/*! \brief Class interface for cl_kernel.
5819
* \note Copies of these objects are shallow, meaning that the copy will refer
5820
* to the same underlying cl_kernel as the original. For details, see
5821
* clRetainKernel() and clReleaseKernel().
5825
class Kernel : public detail::Wrapper<cl_kernel>
5828
inline Kernel(const Program& program, const char* name, cl_int* err = NULL);
5830
//! \brief Default constructor - initializes to NULL.
5833
/*! \brief Constructor from cl_kernel - takes ownership.
5835
* \param retainObject will cause the constructor to retain its cl object.
5836
* Defaults to false to maintain compatibility with
5838
* This effectively transfers ownership of a refcount on the cl_kernel
5839
* into the new Kernel object.
5841
explicit Kernel(const cl_kernel& kernel, bool retainObject = false) :
5842
detail::Wrapper<cl_type>(kernel, retainObject) { }
5844
/*! \brief Assignment operator from cl_kernel - takes ownership.
5846
* This effectively transfers ownership of a refcount on the rhs and calls
5847
* clReleaseKernel() on the value previously held by this instance.
5849
Kernel& operator = (const cl_kernel& rhs)
5851
detail::Wrapper<cl_type>::operator=(rhs);
5855
/*! \brief Copy constructor to forward copy to the superclass correctly.
5856
* Required for MSVC.
5858
Kernel(const Kernel& kernel) : detail::Wrapper<cl_type>(kernel) {}
5860
/*! \brief Copy assignment to forward copy to the superclass correctly.
5861
* Required for MSVC.
5863
Kernel& operator = (const Kernel &kernel)
5865
detail::Wrapper<cl_type>::operator=(kernel);
5869
/*! \brief Move constructor to forward move to the superclass correctly.
5870
* Required for MSVC.
5872
Kernel(Kernel&& kernel) CL_HPP_NOEXCEPT_ : detail::Wrapper<cl_type>(std::move(kernel)) {}
5874
/*! \brief Move assignment to forward move to the superclass correctly.
5875
* Required for MSVC.
5877
Kernel& operator = (Kernel &&kernel)
5879
detail::Wrapper<cl_type>::operator=(std::move(kernel));
5883
template <typename T>
5884
cl_int getInfo(cl_kernel_info name, T* param) const
5886
return detail::errHandler(
5887
detail::getInfo(&::clGetKernelInfo, object_, name, param),
5888
__GET_KERNEL_INFO_ERR);
5891
template <cl_int name> typename
5892
detail::param_traits<detail::cl_kernel_info, name>::param_type
5893
getInfo(cl_int* err = NULL) const
5895
typename detail::param_traits<
5896
detail::cl_kernel_info, name>::param_type param;
5897
cl_int result = getInfo(name, ¶m);
5904
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
5905
template <typename T>
5906
cl_int getArgInfo(cl_uint argIndex, cl_kernel_arg_info name, T* param) const
5908
return detail::errHandler(
5909
detail::getInfo(&::clGetKernelArgInfo, object_, argIndex, name, param),
5910
__GET_KERNEL_ARG_INFO_ERR);
5913
template <cl_int name> typename
5914
detail::param_traits<detail::cl_kernel_arg_info, name>::param_type
5915
getArgInfo(cl_uint argIndex, cl_int* err = NULL) const
5917
typename detail::param_traits<
5918
detail::cl_kernel_arg_info, name>::param_type param;
5919
cl_int result = getArgInfo(argIndex, name, ¶m);
5925
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
5927
template <typename T>
5928
cl_int getWorkGroupInfo(
5929
const Device& device, cl_kernel_work_group_info name, T* param) const
5931
return detail::errHandler(
5933
&::clGetKernelWorkGroupInfo, object_, device(), name, param),
5934
__GET_KERNEL_WORK_GROUP_INFO_ERR);
5937
template <cl_int name> typename
5938
detail::param_traits<detail::cl_kernel_work_group_info, name>::param_type
5939
getWorkGroupInfo(const Device& device, cl_int* err = NULL) const
5941
typename detail::param_traits<
5942
detail::cl_kernel_work_group_info, name>::param_type param;
5943
cl_int result = getWorkGroupInfo(device, name, ¶m);
5950
#if (CL_HPP_TARGET_OPENCL_VERSION >= 200 && defined(CL_HPP_USE_CL_SUB_GROUPS_KHR)) || CL_HPP_TARGET_OPENCL_VERSION >= 210
5951
cl_int getSubGroupInfo(const cl::Device &dev, cl_kernel_sub_group_info name, const cl::NDRange &range, size_type* param) const
5953
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
5955
return detail::errHandler(
5956
clGetKernelSubGroupInfo(object_, dev(), name, range.size(), range.get(), sizeof(size_type), param, nullptr),
5957
__GET_KERNEL_SUB_GROUP_INFO_ERR);
5959
#else // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
5961
typedef clGetKernelSubGroupInfoKHR_fn PFN_clGetKernelSubGroupInfoKHR;
5962
static PFN_clGetKernelSubGroupInfoKHR pfn_clGetKernelSubGroupInfoKHR = NULL;
5963
CL_HPP_INIT_CL_EXT_FCN_PTR_(clGetKernelSubGroupInfoKHR);
5965
return detail::errHandler(
5966
pfn_clGetKernelSubGroupInfoKHR(object_, dev(), name, range.size(), range.get(), sizeof(size_type), param, nullptr),
5967
__GET_KERNEL_SUB_GROUP_INFO_ERR);
5969
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
5972
template <cl_int name>
5973
size_type getSubGroupInfo(const cl::Device &dev, const cl::NDRange &range, cl_int* err = NULL) const
5976
cl_int result = getSubGroupInfo(dev, name, range, ¶m);
5982
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
5984
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
5985
/*! \brief setArg overload taking a shared_ptr type
5987
template<typename T, class D>
5988
cl_int setArg(cl_uint index, const cl::pointer<T, D> &argPtr)
5990
return detail::errHandler(
5991
::clSetKernelArgSVMPointer(object_, index, argPtr.get()),
5992
__SET_KERNEL_ARGS_ERR);
5995
/*! \brief setArg overload taking a vector type.
5997
template<typename T, class Alloc>
5998
cl_int setArg(cl_uint index, const cl::vector<T, Alloc> &argPtr)
6000
return detail::errHandler(
6001
::clSetKernelArgSVMPointer(object_, index, argPtr.data()),
6002
__SET_KERNEL_ARGS_ERR);
6005
/*! \brief setArg overload taking a pointer type
6007
template<typename T>
6008
typename std::enable_if<std::is_pointer<T>::value, cl_int>::type
6009
setArg(cl_uint index, const T argPtr)
6011
return detail::errHandler(
6012
::clSetKernelArgSVMPointer(object_, index, argPtr),
6013
__SET_KERNEL_ARGS_ERR);
6015
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
6017
/*! \brief setArg overload taking a POD type
6019
template <typename T>
6020
typename std::enable_if<!std::is_pointer<T>::value, cl_int>::type
6021
setArg(cl_uint index, const T &value)
6023
return detail::errHandler(
6027
detail::KernelArgumentHandler<T>::size(value),
6028
detail::KernelArgumentHandler<T>::ptr(value)),
6029
__SET_KERNEL_ARGS_ERR);
6032
cl_int setArg(cl_uint index, size_type size, const void* argPtr)
6034
return detail::errHandler(
6035
::clSetKernelArg(object_, index, size, argPtr),
6036
__SET_KERNEL_ARGS_ERR);
6039
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
6041
* Specify a vector of SVM pointers that the kernel may access in
6042
* addition to its arguments.
6044
cl_int setSVMPointers(const vector<void*> &pointerList)
6046
return detail::errHandler(
6047
::clSetKernelExecInfo(
6049
CL_KERNEL_EXEC_INFO_SVM_PTRS,
6050
sizeof(void*)*pointerList.size(),
6051
pointerList.data()));
6055
* Specify a std::array of SVM pointers that the kernel may access in
6056
* addition to its arguments.
6058
template<int ArrayLength>
6059
cl_int setSVMPointers(const std::array<void*, ArrayLength> &pointerList)
6061
return detail::errHandler(
6062
::clSetKernelExecInfo(
6064
CL_KERNEL_EXEC_INFO_SVM_PTRS,
6065
sizeof(void*)*pointerList.size(),
6066
pointerList.data()));
6069
/*! \brief Enable fine-grained system SVM.
6071
* \note It is only possible to enable fine-grained system SVM if all devices
6072
* in the context associated with kernel support it.
6074
* \param svmEnabled True if fine-grained system SVM is requested. False otherwise.
6075
* \return CL_SUCCESS if the function was executed succesfully. CL_INVALID_OPERATION
6076
* if no devices in the context support fine-grained system SVM.
6078
* \see clSetKernelExecInfo
6080
cl_int enableFineGrainedSystemSVM(bool svmEnabled)
6082
cl_bool svmEnabled_ = svmEnabled ? CL_TRUE : CL_FALSE;
6083
return detail::errHandler(
6084
::clSetKernelExecInfo(
6086
CL_KERNEL_EXEC_INFO_SVM_FINE_GRAIN_SYSTEM,
6093
template<int index, int ArrayLength, class D, typename T0, typename T1, typename... Ts>
6094
void setSVMPointersHelper(std::array<void*, ArrayLength> &pointerList, const pointer<T0, D> &t0, const pointer<T1, D> &t1, Ts & ... ts)
6096
pointerList[index] = static_cast<void*>(t0.get());
6097
setSVMPointersHelper<index + 1, ArrayLength>(pointerList, t1, ts...);
6100
template<int index, int ArrayLength, typename T0, typename T1, typename... Ts>
6101
typename std::enable_if<std::is_pointer<T0>::value, void>::type
6102
setSVMPointersHelper(std::array<void*, ArrayLength> &pointerList, T0 t0, T1 t1, Ts... ts)
6104
pointerList[index] = static_cast<void*>(t0);
6105
setSVMPointersHelper<index + 1, ArrayLength>(pointerList, t1, ts...);
6108
template<int index, int ArrayLength, typename T0, class D>
6109
void setSVMPointersHelper(std::array<void*, ArrayLength> &pointerList, const pointer<T0, D> &t0)
6111
pointerList[index] = static_cast<void*>(t0.get());
6115
template<int index, int ArrayLength, typename T0>
6116
typename std::enable_if<std::is_pointer<T0>::value, void>::type
6117
setSVMPointersHelper(std::array<void*, ArrayLength> &pointerList, T0 t0)
6119
pointerList[index] = static_cast<void*>(t0);
6122
template<typename T0, typename... Ts>
6123
cl_int setSVMPointers(const T0 &t0, Ts & ... ts)
6125
std::array<void*, 1 + sizeof...(Ts)> pointerList;
6127
setSVMPointersHelper<0, 1 + sizeof...(Ts)>(pointerList, t0, ts...);
6128
return detail::errHandler(
6129
::clSetKernelExecInfo(
6131
CL_KERNEL_EXEC_INFO_SVM_PTRS,
6132
sizeof(void*)*(1 + sizeof...(Ts)),
6133
pointerList.data()));
6135
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
6137
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
6139
* Make a deep copy of the kernel object including its arguments.
6140
* @return A new kernel object with internal state entirely separate from that
6141
* of the original but with any arguments set on the original intact.
6146
Kernel retValue(clCloneKernel(this->get(), &error));
6148
detail::errHandler(error, __CLONE_KERNEL_ERR);
6151
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
6155
* \brief Program interface that implements cl_program.
6157
class Program : public detail::Wrapper<cl_program>
6160
#if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6161
typedef vector<vector<unsigned char>> Binaries;
6162
typedef vector<string> Sources;
6163
#else // #if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6164
typedef vector<std::pair<const void*, size_type> > Binaries;
6165
typedef vector<std::pair<const char*, size_type> > Sources;
6166
#endif // #if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6169
const string& source,
6175
const char * strings = source.c_str();
6176
const size_type length = source.size();
6178
Context context = Context::getDefault(err);
6180
object_ = ::clCreateProgramWithSource(
6181
context(), (cl_uint)1, &strings, &length, &error);
6183
detail::errHandler(error, __CREATE_PROGRAM_WITH_SOURCE_ERR);
6185
if (error == CL_SUCCESS && build) {
6187
error = ::clBuildProgram(
6191
#if !defined(CL_HPP_CL_1_2_DEFAULT_BUILD)
6195
#endif // #if !defined(CL_HPP_CL_1_2_DEFAULT_BUILD)
6199
detail::buildErrHandler(error, __BUILD_PROGRAM_ERR, getBuildInfo<CL_PROGRAM_BUILD_LOG>());
6208
const Context& context,
6209
const string& source,
6215
const char * strings = source.c_str();
6216
const size_type length = source.size();
6218
object_ = ::clCreateProgramWithSource(
6219
context(), (cl_uint)1, &strings, &length, &error);
6221
detail::errHandler(error, __CREATE_PROGRAM_WITH_SOURCE_ERR);
6223
if (error == CL_SUCCESS && build) {
6224
error = ::clBuildProgram(
6228
#if !defined(CL_HPP_CL_1_2_DEFAULT_BUILD)
6232
#endif // #if !defined(CL_HPP_CL_1_2_DEFAULT_BUILD)
6236
detail::buildErrHandler(error, __BUILD_PROGRAM_ERR, getBuildInfo<CL_PROGRAM_BUILD_LOG>());
6245
* Create a program from a vector of source strings and the default context.
6246
* Does not compile or link the program.
6249
const Sources& sources,
6253
Context context = Context::getDefault(err);
6255
const size_type n = (size_type)sources.size();
6257
vector<size_type> lengths(n);
6258
vector<const char*> strings(n);
6260
for (size_type i = 0; i < n; ++i) {
6261
#if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6262
strings[i] = sources[(int)i].data();
6263
lengths[i] = sources[(int)i].length();
6264
#else // #if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6265
strings[i] = sources[(int)i].first;
6266
lengths[i] = sources[(int)i].second;
6267
#endif // #if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6270
object_ = ::clCreateProgramWithSource(
6271
context(), (cl_uint)n, strings.data(), lengths.data(), &error);
6273
detail::errHandler(error, __CREATE_PROGRAM_WITH_SOURCE_ERR);
6280
* Create a program from a vector of source strings and a provided context.
6281
* Does not compile or link the program.
6284
const Context& context,
6285
const Sources& sources,
6290
const size_type n = (size_type)sources.size();
6292
vector<size_type> lengths(n);
6293
vector<const char*> strings(n);
6295
for (size_type i = 0; i < n; ++i) {
6296
#if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6297
strings[i] = sources[(int)i].data();
6298
lengths[i] = sources[(int)i].length();
6299
#else // #if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6300
strings[i] = sources[(int)i].first;
6301
lengths[i] = sources[(int)i].second;
6302
#endif // #if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6305
object_ = ::clCreateProgramWithSource(
6306
context(), (cl_uint)n, strings.data(), lengths.data(), &error);
6308
detail::errHandler(error, __CREATE_PROGRAM_WITH_SOURCE_ERR);
6315
#if CL_HPP_TARGET_OPENCL_VERSION >= 210 || (CL_HPP_TARGET_OPENCL_VERSION==200 && defined(CL_HPP_USE_IL_KHR))
6317
* Program constructor to allow construction of program from SPIR-V or another IL.
6318
* Valid for either OpenCL >= 2.1 or when CL_HPP_USE_IL_KHR is defined.
6321
const vector<char>& IL,
6327
Context context = Context::getDefault(err);
6329
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
6331
object_ = ::clCreateProgramWithIL(
6332
context(), static_cast<const void*>(IL.data()), IL.size(), &error);
6334
#else // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
6336
typedef clCreateProgramWithILKHR_fn PFN_clCreateProgramWithILKHR;
6337
static PFN_clCreateProgramWithILKHR pfn_clCreateProgramWithILKHR = NULL;
6338
CL_HPP_INIT_CL_EXT_FCN_PTR_(clCreateProgramWithILKHR);
6340
return detail::errHandler(
6341
pfn_clCreateProgramWithILKHR(
6342
context(), static_cast<const void*>(IL.data()), IL.size(), &error);
6344
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
6346
detail::errHandler(error, __CREATE_PROGRAM_WITH_IL_ERR);
6348
if (error == CL_SUCCESS && build) {
6350
error = ::clBuildProgram(
6354
#if !defined(CL_HPP_CL_1_2_DEFAULT_BUILD)
6358
#endif // #if !defined(CL_HPP_CL_1_2_DEFAULT_BUILD)
6362
detail::buildErrHandler(error, __BUILD_PROGRAM_ERR, getBuildInfo<CL_PROGRAM_BUILD_LOG>());
6371
* Program constructor to allow construction of program from SPIR-V or another IL
6372
* for a specific context.
6373
* Valid for either OpenCL >= 2.1 or when CL_HPP_USE_IL_KHR is defined.
6376
const Context& context,
6377
const vector<char>& IL,
6383
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
6385
object_ = ::clCreateProgramWithIL(
6386
context(), static_cast<const void*>(IL.data()), IL.size(), &error);
6388
#else // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
6390
typedef clCreateProgramWithILKHR_fn PFN_clCreateProgramWithILKHR;
6391
static PFN_clCreateProgramWithILKHR pfn_clCreateProgramWithILKHR = NULL;
6392
CL_HPP_INIT_CL_EXT_FCN_PTR_(clCreateProgramWithILKHR);
6394
return detail::errHandler(
6395
pfn_clCreateProgramWithILKHR(
6396
context(), static_cast<const void*>(IL.data()), IL.size(), &error);
6398
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
6400
detail::errHandler(error, __CREATE_PROGRAM_WITH_IL_ERR);
6402
if (error == CL_SUCCESS && build) {
6403
error = ::clBuildProgram(
6407
#if !defined(CL_HPP_CL_1_2_DEFAULT_BUILD)
6411
#endif // #if !defined(CL_HPP_CL_1_2_DEFAULT_BUILD)
6415
detail::buildErrHandler(error, __BUILD_PROGRAM_ERR, getBuildInfo<CL_PROGRAM_BUILD_LOG>());
6422
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
6425
* Construct a program object from a list of devices and a per-device list of binaries.
6426
* \param context A valid OpenCL context in which to construct the program.
6427
* \param devices A vector of OpenCL device objects for which the program will be created.
6428
* \param binaries A vector of pairs of a pointer to a binary object and its length.
6429
* \param binaryStatus An optional vector that on completion will be resized to
6430
* match the size of binaries and filled with values to specify if each binary
6431
* was successfully loaded.
6432
* Set to CL_SUCCESS if the binary was successfully loaded.
6433
* Set to CL_INVALID_VALUE if the length is 0 or the binary pointer is NULL.
6434
* Set to CL_INVALID_BINARY if the binary provided is not valid for the matching device.
6435
* \param err if non-NULL will be set to CL_SUCCESS on successful operation or one of the following errors:
6436
* CL_INVALID_CONTEXT if context is not a valid context.
6437
* CL_INVALID_VALUE if the length of devices is zero; or if the length of binaries does not match the length of devices;
6438
* or if any entry in binaries is NULL or has length 0.
6439
* CL_INVALID_DEVICE if OpenCL devices listed in devices are not in the list of devices associated with context.
6440
* CL_INVALID_BINARY if an invalid program binary was encountered for any device. binaryStatus will return specific status for each device.
6441
* CL_OUT_OF_HOST_MEMORY if there is a failure to allocate resources required by the OpenCL implementation on the host.
6444
const Context& context,
6445
const vector<Device>& devices,
6446
const Binaries& binaries,
6447
vector<cl_int>* binaryStatus = NULL,
6452
const size_type numDevices = devices.size();
6454
// Catch size mismatch early and return
6455
if(binaries.size() != numDevices) {
6456
error = CL_INVALID_VALUE;
6457
detail::errHandler(error, __CREATE_PROGRAM_WITH_BINARY_ERR);
6465
vector<size_type> lengths(numDevices);
6466
vector<const unsigned char*> images(numDevices);
6467
#if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6468
for (size_type i = 0; i < numDevices; ++i) {
6469
images[i] = binaries[i].data();
6470
lengths[i] = binaries[(int)i].size();
6472
#else // #if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6473
for (size_type i = 0; i < numDevices; ++i) {
6474
images[i] = (const unsigned char*)binaries[i].first;
6475
lengths[i] = binaries[(int)i].second;
6477
#endif // #if !defined(CL_HPP_ENABLE_PROGRAM_CONSTRUCTION_FROM_ARRAY_COMPATIBILITY)
6479
vector<cl_device_id> deviceIDs(numDevices);
6480
for( size_type deviceIndex = 0; deviceIndex < numDevices; ++deviceIndex ) {
6481
deviceIDs[deviceIndex] = (devices[deviceIndex])();
6485
binaryStatus->resize(numDevices);
6488
object_ = ::clCreateProgramWithBinary(
6489
context(), (cl_uint) devices.size(),
6491
lengths.data(), images.data(), (binaryStatus != NULL && numDevices > 0)
6492
? &binaryStatus->front()
6495
detail::errHandler(error, __CREATE_PROGRAM_WITH_BINARY_ERR);
6502
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
6504
* Create program using builtin kernels.
6505
* \param kernelNames Semi-colon separated list of builtin kernel names
6508
const Context& context,
6509
const vector<Device>& devices,
6510
const string& kernelNames,
6516
size_type numDevices = devices.size();
6517
vector<cl_device_id> deviceIDs(numDevices);
6518
for( size_type deviceIndex = 0; deviceIndex < numDevices; ++deviceIndex ) {
6519
deviceIDs[deviceIndex] = (devices[deviceIndex])();
6522
object_ = ::clCreateProgramWithBuiltInKernels(
6524
(cl_uint) devices.size(),
6526
kernelNames.c_str(),
6529
detail::errHandler(error, __CREATE_PROGRAM_WITH_BUILT_IN_KERNELS_ERR);
6534
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
6539
/*! \brief Constructor from cl_mem - takes ownership.
6541
* \param retainObject will cause the constructor to retain its cl object.
6542
* Defaults to false to maintain compatibility with
6545
explicit Program(const cl_program& program, bool retainObject = false) :
6546
detail::Wrapper<cl_type>(program, retainObject) { }
6548
Program& operator = (const cl_program& rhs)
6550
detail::Wrapper<cl_type>::operator=(rhs);
6554
/*! \brief Copy constructor to forward copy to the superclass correctly.
6555
* Required for MSVC.
6557
Program(const Program& program) : detail::Wrapper<cl_type>(program) {}
6559
/*! \brief Copy assignment to forward copy to the superclass correctly.
6560
* Required for MSVC.
6562
Program& operator = (const Program &program)
6564
detail::Wrapper<cl_type>::operator=(program);
6568
/*! \brief Move constructor to forward move to the superclass correctly.
6569
* Required for MSVC.
6571
Program(Program&& program) CL_HPP_NOEXCEPT_ : detail::Wrapper<cl_type>(std::move(program)) {}
6573
/*! \brief Move assignment to forward move to the superclass correctly.
6574
* Required for MSVC.
6576
Program& operator = (Program &&program)
6578
detail::Wrapper<cl_type>::operator=(std::move(program));
6583
const vector<Device>& devices,
6584
const char* options = NULL,
6585
void (CL_CALLBACK * notifyFptr)(cl_program, void *) = NULL,
6586
void* data = NULL) const
6588
size_type numDevices = devices.size();
6589
vector<cl_device_id> deviceIDs(numDevices);
6591
for( size_type deviceIndex = 0; deviceIndex < numDevices; ++deviceIndex ) {
6592
deviceIDs[deviceIndex] = (devices[deviceIndex])();
6595
cl_int buildError = ::clBuildProgram(
6604
return detail::buildErrHandler(buildError, __BUILD_PROGRAM_ERR, getBuildInfo<CL_PROGRAM_BUILD_LOG>());
6608
const char* options = NULL,
6609
void (CL_CALLBACK * notifyFptr)(cl_program, void *) = NULL,
6610
void* data = NULL) const
6612
cl_int buildError = ::clBuildProgram(
6621
return detail::buildErrHandler(buildError, __BUILD_PROGRAM_ERR, getBuildInfo<CL_PROGRAM_BUILD_LOG>());
6624
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
6626
const char* options = NULL,
6627
void (CL_CALLBACK * notifyFptr)(cl_program, void *) = NULL,
6628
void* data = NULL) const
6630
cl_int error = ::clCompileProgram(
6640
return detail::buildErrHandler(error, __COMPILE_PROGRAM_ERR, getBuildInfo<CL_PROGRAM_BUILD_LOG>());
6642
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
6644
template <typename T>
6645
cl_int getInfo(cl_program_info name, T* param) const
6647
return detail::errHandler(
6648
detail::getInfo(&::clGetProgramInfo, object_, name, param),
6649
__GET_PROGRAM_INFO_ERR);
6652
template <cl_int name> typename
6653
detail::param_traits<detail::cl_program_info, name>::param_type
6654
getInfo(cl_int* err = NULL) const
6656
typename detail::param_traits<
6657
detail::cl_program_info, name>::param_type param;
6658
cl_int result = getInfo(name, ¶m);
6665
template <typename T>
6666
cl_int getBuildInfo(
6667
const Device& device, cl_program_build_info name, T* param) const
6669
return detail::errHandler(
6671
&::clGetProgramBuildInfo, object_, device(), name, param),
6672
__GET_PROGRAM_BUILD_INFO_ERR);
6675
template <cl_int name> typename
6676
detail::param_traits<detail::cl_program_build_info, name>::param_type
6677
getBuildInfo(const Device& device, cl_int* err = NULL) const
6679
typename detail::param_traits<
6680
detail::cl_program_build_info, name>::param_type param;
6681
cl_int result = getBuildInfo(device, name, ¶m);
6689
* Build info function that returns a vector of device/info pairs for the specified
6690
* info type and for all devices in the program.
6691
* On an error reading the info for any device, an empty vector of info will be returned.
6693
template <cl_int name>
6694
vector<std::pair<cl::Device, typename detail::param_traits<detail::cl_program_build_info, name>::param_type>>
6695
getBuildInfo(cl_int *err = NULL) const
6697
cl_int result = CL_SUCCESS;
6699
auto devs = getInfo<CL_PROGRAM_DEVICES>(&result);
6700
vector<std::pair<cl::Device, typename detail::param_traits<detail::cl_program_build_info, name>::param_type>>
6703
// If there was an initial error from getInfo return the error
6704
if (result != CL_SUCCESS) {
6711
for (const cl::Device &d : devs) {
6712
typename detail::param_traits<
6713
detail::cl_program_build_info, name>::param_type param;
6714
result = getBuildInfo(d, name, ¶m);
6716
std::pair<cl::Device, typename detail::param_traits<detail::cl_program_build_info, name>::param_type>
6718
if (result != CL_SUCCESS) {
6719
// On error, leave the loop and return the error code
6726
if (result != CL_SUCCESS) {
6732
cl_int createKernels(vector<Kernel>* kernels)
6735
cl_int err = ::clCreateKernelsInProgram(object_, 0, NULL, &numKernels);
6736
if (err != CL_SUCCESS) {
6737
return detail::errHandler(err, __CREATE_KERNELS_IN_PROGRAM_ERR);
6740
vector<cl_kernel> value(numKernels);
6742
err = ::clCreateKernelsInProgram(
6743
object_, numKernels, value.data(), NULL);
6744
if (err != CL_SUCCESS) {
6745
return detail::errHandler(err, __CREATE_KERNELS_IN_PROGRAM_ERR);
6749
kernels->resize(value.size());
6751
// Assign to param, constructing with retain behaviour
6752
// to correctly capture each underlying CL object
6753
for (size_type i = 0; i < value.size(); i++) {
6754
// We do not need to retain because this kernel is being created
6756
(*kernels)[i] = Kernel(value[i], false);
6762
#if CL_HPP_TARGET_OPENCL_VERSION >= 220
6763
/*! \brief Registers a callback function to be called when destructors for
6764
* program scope global variables are complete and before the
6765
* program is released.
6767
* Wraps clSetProgramReleaseCallback().
6769
* Each call to this function registers the specified user callback function
6770
* on a callback stack associated with program. The registered user callback
6771
* functions are called in the reverse order in which they were registered.
6773
cl_int setReleaseCallback(
6774
void (CL_CALLBACK * pfn_notify)(cl_program program, void * user_data),
6775
void * user_data = NULL)
6777
return detail::errHandler(
6778
::clSetProgramReleaseCallback(
6782
__SET_PROGRAM_RELEASE_CALLBACK_ERR);
6785
/*! \brief Sets a SPIR-V specialization constant.
6787
* Wraps clSetProgramSpecializationConstant().
6789
template <typename T>
6790
typename std::enable_if<!std::is_pointer<T>::value, cl_int>::type
6791
setSpecializationConstant(cl_uint index, const T &value)
6793
return detail::errHandler(
6794
::clSetProgramSpecializationConstant(
6799
__SET_PROGRAM_SPECIALIZATION_CONSTANT_ERR);
6802
/*! \brief Sets a SPIR-V specialization constant.
6804
* Wraps clSetProgramSpecializationConstant().
6806
cl_int setSpecializationConstant(cl_uint index, size_type size, const void* value)
6808
return detail::errHandler(
6809
::clSetProgramSpecializationConstant(
6814
__SET_PROGRAM_SPECIALIZATION_CONSTANT_ERR);
6816
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 220
6819
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
6820
inline Program linkProgram(
6823
const char* options = NULL,
6824
void (CL_CALLBACK * notifyFptr)(cl_program, void *) = NULL,
6828
cl_int error_local = CL_SUCCESS;
6830
cl_program programs[2] = { input1(), input2() };
6832
Context ctx = input1.getInfo<CL_PROGRAM_CONTEXT>(&error_local);
6833
if(error_local!=CL_SUCCESS) {
6834
detail::errHandler(error_local, __LINK_PROGRAM_ERR);
6837
cl_program prog = ::clLinkProgram(
6848
detail::errHandler(error_local,__COMPILE_PROGRAM_ERR);
6853
return Program(prog);
6856
inline Program linkProgram(
6857
vector<Program> inputPrograms,
6858
const char* options = NULL,
6859
void (CL_CALLBACK * notifyFptr)(cl_program, void *) = NULL,
6863
cl_int error_local = CL_SUCCESS;
6865
vector<cl_program> programs(inputPrograms.size());
6867
for (unsigned int i = 0; i < inputPrograms.size(); i++) {
6868
programs[i] = inputPrograms[i]();
6872
if(inputPrograms.size() > 0) {
6873
ctx = inputPrograms[0].getInfo<CL_PROGRAM_CONTEXT>(&error_local);
6874
if(error_local!=CL_SUCCESS) {
6875
detail::errHandler(error_local, __LINK_PROGRAM_ERR);
6878
cl_program prog = ::clLinkProgram(
6883
(cl_uint)inputPrograms.size(),
6889
detail::errHandler(error_local,__COMPILE_PROGRAM_ERR);
6894
return Program(prog, false);
6896
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
6898
// Template specialization for CL_PROGRAM_BINARIES
6900
inline cl_int cl::Program::getInfo(cl_program_info name, vector<vector<unsigned char>>* param) const
6902
if (name != CL_PROGRAM_BINARIES) {
6903
return CL_INVALID_VALUE;
6906
// Resize the parameter array appropriately for each allocation
6907
// and pass down to the helper
6909
vector<size_type> sizes = getInfo<CL_PROGRAM_BINARY_SIZES>();
6910
size_type numBinaries = sizes.size();
6912
// Resize the parameter array and constituent arrays
6913
param->resize(numBinaries);
6914
for (size_type i = 0; i < numBinaries; ++i) {
6915
(*param)[i].resize(sizes[i]);
6918
return detail::errHandler(
6919
detail::getInfo(&::clGetProgramInfo, object_, name, param),
6920
__GET_PROGRAM_INFO_ERR);
6927
inline vector<vector<unsigned char>> cl::Program::getInfo<CL_PROGRAM_BINARIES>(cl_int* err) const
6929
vector<vector<unsigned char>> binariesVectors;
6931
cl_int result = getInfo(CL_PROGRAM_BINARIES, &binariesVectors);
6935
return binariesVectors;
6938
#if CL_HPP_TARGET_OPENCL_VERSION >= 220
6939
// Template specialization for clSetProgramSpecializationConstant
6941
inline cl_int cl::Program::setSpecializationConstant(cl_uint index, const bool &value)
6943
cl_uchar ucValue = value ? CL_UCHAR_MAX : 0;
6944
return detail::errHandler(
6945
::clSetProgramSpecializationConstant(
6950
__SET_PROGRAM_SPECIALIZATION_CONSTANT_ERR);
6952
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 220
6954
inline Kernel::Kernel(const Program& program, const char* name, cl_int* err)
6958
object_ = ::clCreateKernel(program(), name, &error);
6959
detail::errHandler(error, __CREATE_KERNEL_ERR);
6967
enum class QueueProperties : cl_command_queue_properties
6970
Profiling = CL_QUEUE_PROFILING_ENABLE,
6971
OutOfOrder = CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE,
6974
inline QueueProperties operator|(QueueProperties lhs, QueueProperties rhs)
6976
return static_cast<QueueProperties>(static_cast<cl_command_queue_properties>(lhs) | static_cast<cl_command_queue_properties>(rhs));
6979
/*! \class CommandQueue
6980
* \brief CommandQueue interface for cl_command_queue.
6982
class CommandQueue : public detail::Wrapper<cl_command_queue>
6985
static std::once_flag default_initialized_;
6986
static CommandQueue default_;
6987
static cl_int default_error_;
6989
/*! \brief Create the default command queue returned by @ref getDefault.
6991
* It sets default_error_ to indicate success or failure. It does not throw
6994
static void makeDefault()
6996
/* We don't want to throw an error from this function, so we have to
6997
* catch and set the error flag.
6999
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
7004
Context context = Context::getDefault(&error);
7006
if (error != CL_SUCCESS) {
7007
default_error_ = error;
7010
Device device = Device::getDefault();
7011
default_ = CommandQueue(context, device, 0, &default_error_);
7014
#if defined(CL_HPP_ENABLE_EXCEPTIONS)
7015
catch (cl::Error &e) {
7016
default_error_ = e.err();
7021
/*! \brief Create the default command queue.
7023
* This sets @c default_. It does not throw
7026
static void makeDefaultProvided(const CommandQueue &c) {
7031
#ifdef CL_HPP_UNIT_TEST_ENABLE
7032
/*! \brief Reset the default.
7034
* This sets @c default_ to an empty value to support cleanup in
7035
* the unit test framework.
7036
* This function is not thread safe.
7038
static void unitTestClearDefault() {
7039
default_ = CommandQueue();
7041
#endif // #ifdef CL_HPP_UNIT_TEST_ENABLE
7045
* \brief Constructs a CommandQueue based on passed properties.
7046
* Will return an CL_INVALID_QUEUE_PROPERTIES error if CL_QUEUE_ON_DEVICE is specified.
7049
cl_command_queue_properties properties,
7054
Context context = Context::getDefault(&error);
7055
detail::errHandler(error, __CREATE_CONTEXT_ERR);
7057
if (error != CL_SUCCESS) {
7063
Device device = context.getInfo<CL_CONTEXT_DEVICES>()[0];
7064
bool useWithProperties;
7066
#if CL_HPP_TARGET_OPENCL_VERSION >= 200 && CL_HPP_MINIMUM_OPENCL_VERSION < 200
7067
// Run-time decision based on the actual platform
7069
cl_uint version = detail::getContextPlatformVersion(context());
7070
useWithProperties = (version >= 0x20000); // OpenCL 2.0 or above
7072
#elif CL_HPP_TARGET_OPENCL_VERSION >= 200
7073
useWithProperties = true;
7075
useWithProperties = false;
7078
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
7079
if (useWithProperties) {
7080
cl_queue_properties queue_properties[] = {
7081
CL_QUEUE_PROPERTIES, properties, 0 };
7082
if ((properties & CL_QUEUE_ON_DEVICE) == 0) {
7083
object_ = ::clCreateCommandQueueWithProperties(
7084
context(), device(), queue_properties, &error);
7087
error = CL_INVALID_QUEUE_PROPERTIES;
7090
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
7095
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
7096
#if CL_HPP_MINIMUM_OPENCL_VERSION < 200
7097
if (!useWithProperties) {
7098
object_ = ::clCreateCommandQueue(
7099
context(), device(), properties, &error);
7101
detail::errHandler(error, __CREATE_COMMAND_QUEUE_ERR);
7106
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 200
7111
* \brief Constructs a CommandQueue based on passed properties.
7112
* Will return an CL_INVALID_QUEUE_PROPERTIES error if CL_QUEUE_ON_DEVICE is specified.
7115
QueueProperties properties,
7120
Context context = Context::getDefault(&error);
7121
detail::errHandler(error, __CREATE_CONTEXT_ERR);
7123
if (error != CL_SUCCESS) {
7129
Device device = context.getInfo<CL_CONTEXT_DEVICES>()[0];
7130
bool useWithProperties;
7132
#if CL_HPP_TARGET_OPENCL_VERSION >= 200 && CL_HPP_MINIMUM_OPENCL_VERSION < 200
7133
// Run-time decision based on the actual platform
7135
cl_uint version = detail::getContextPlatformVersion(context());
7136
useWithProperties = (version >= 0x20000); // OpenCL 2.0 or above
7138
#elif CL_HPP_TARGET_OPENCL_VERSION >= 200
7139
useWithProperties = true;
7141
useWithProperties = false;
7144
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
7145
if (useWithProperties) {
7146
cl_queue_properties queue_properties[] = {
7147
CL_QUEUE_PROPERTIES, static_cast<cl_queue_properties>(properties), 0 };
7149
object_ = ::clCreateCommandQueueWithProperties(
7150
context(), device(), queue_properties, &error);
7152
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
7157
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
7158
#if CL_HPP_MINIMUM_OPENCL_VERSION < 200
7159
if (!useWithProperties) {
7160
object_ = ::clCreateCommandQueue(
7161
context(), device(), static_cast<cl_command_queue_properties>(properties), &error);
7163
detail::errHandler(error, __CREATE_COMMAND_QUEUE_ERR);
7168
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 200
7174
* \brief Constructs a CommandQueue for an implementation defined device in the given context
7175
* Will return an CL_INVALID_QUEUE_PROPERTIES error if CL_QUEUE_ON_DEVICE is specified.
7177
explicit CommandQueue(
7178
const Context& context,
7179
cl_command_queue_properties properties = 0,
7183
bool useWithProperties;
7184
vector<cl::Device> devices;
7185
error = context.getInfo(CL_CONTEXT_DEVICES, &devices);
7187
detail::errHandler(error, __CREATE_CONTEXT_ERR);
7189
if (error != CL_SUCCESS)
7197
#if CL_HPP_TARGET_OPENCL_VERSION >= 200 && CL_HPP_MINIMUM_OPENCL_VERSION < 200
7198
// Run-time decision based on the actual platform
7200
cl_uint version = detail::getContextPlatformVersion(context());
7201
useWithProperties = (version >= 0x20000); // OpenCL 2.0 or above
7203
#elif CL_HPP_TARGET_OPENCL_VERSION >= 200
7204
useWithProperties = true;
7206
useWithProperties = false;
7209
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
7210
if (useWithProperties) {
7211
cl_queue_properties queue_properties[] = {
7212
CL_QUEUE_PROPERTIES, properties, 0 };
7213
if ((properties & CL_QUEUE_ON_DEVICE) == 0) {
7214
object_ = ::clCreateCommandQueueWithProperties(
7215
context(), devices[0](), queue_properties, &error);
7218
error = CL_INVALID_QUEUE_PROPERTIES;
7221
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
7226
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
7227
#if CL_HPP_MINIMUM_OPENCL_VERSION < 200
7228
if (!useWithProperties) {
7229
object_ = ::clCreateCommandQueue(
7230
context(), devices[0](), properties, &error);
7232
detail::errHandler(error, __CREATE_COMMAND_QUEUE_ERR);
7237
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 200
7241
* \brief Constructs a CommandQueue for an implementation defined device in the given context
7242
* Will return an CL_INVALID_QUEUE_PROPERTIES error if CL_QUEUE_ON_DEVICE is specified.
7244
explicit CommandQueue(
7245
const Context& context,
7246
QueueProperties properties,
7250
bool useWithProperties;
7251
vector<cl::Device> devices;
7252
error = context.getInfo(CL_CONTEXT_DEVICES, &devices);
7254
detail::errHandler(error, __CREATE_CONTEXT_ERR);
7256
if (error != CL_SUCCESS)
7264
#if CL_HPP_TARGET_OPENCL_VERSION >= 200 && CL_HPP_MINIMUM_OPENCL_VERSION < 200
7265
// Run-time decision based on the actual platform
7267
cl_uint version = detail::getContextPlatformVersion(context());
7268
useWithProperties = (version >= 0x20000); // OpenCL 2.0 or above
7270
#elif CL_HPP_TARGET_OPENCL_VERSION >= 200
7271
useWithProperties = true;
7273
useWithProperties = false;
7276
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
7277
if (useWithProperties) {
7278
cl_queue_properties queue_properties[] = {
7279
CL_QUEUE_PROPERTIES, static_cast<cl_queue_properties>(properties), 0 };
7280
object_ = ::clCreateCommandQueueWithProperties(
7281
context(), devices[0](), queue_properties, &error);
7283
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
7288
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
7289
#if CL_HPP_MINIMUM_OPENCL_VERSION < 200
7290
if (!useWithProperties) {
7291
object_ = ::clCreateCommandQueue(
7292
context(), devices[0](), static_cast<cl_command_queue_properties>(properties), &error);
7294
detail::errHandler(error, __CREATE_COMMAND_QUEUE_ERR);
7299
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 200
7303
* \brief Constructs a CommandQueue for a passed device and context
7304
* Will return an CL_INVALID_QUEUE_PROPERTIES error if CL_QUEUE_ON_DEVICE is specified.
7307
const Context& context,
7308
const Device& device,
7309
cl_command_queue_properties properties = 0,
7313
bool useWithProperties;
7315
#if CL_HPP_TARGET_OPENCL_VERSION >= 200 && CL_HPP_MINIMUM_OPENCL_VERSION < 200
7316
// Run-time decision based on the actual platform
7318
cl_uint version = detail::getContextPlatformVersion(context());
7319
useWithProperties = (version >= 0x20000); // OpenCL 2.0 or above
7321
#elif CL_HPP_TARGET_OPENCL_VERSION >= 200
7322
useWithProperties = true;
7324
useWithProperties = false;
7327
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
7328
if (useWithProperties) {
7329
cl_queue_properties queue_properties[] = {
7330
CL_QUEUE_PROPERTIES, properties, 0 };
7331
object_ = ::clCreateCommandQueueWithProperties(
7332
context(), device(), queue_properties, &error);
7334
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
7339
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
7340
#if CL_HPP_MINIMUM_OPENCL_VERSION < 200
7341
if (!useWithProperties) {
7342
object_ = ::clCreateCommandQueue(
7343
context(), device(), properties, &error);
7345
detail::errHandler(error, __CREATE_COMMAND_QUEUE_ERR);
7350
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 200
7354
* \brief Constructs a CommandQueue for a passed device and context
7355
* Will return an CL_INVALID_QUEUE_PROPERTIES error if CL_QUEUE_ON_DEVICE is specified.
7358
const Context& context,
7359
const Device& device,
7360
QueueProperties properties,
7364
bool useWithProperties;
7366
#if CL_HPP_TARGET_OPENCL_VERSION >= 200 && CL_HPP_MINIMUM_OPENCL_VERSION < 200
7367
// Run-time decision based on the actual platform
7369
cl_uint version = detail::getContextPlatformVersion(context());
7370
useWithProperties = (version >= 0x20000); // OpenCL 2.0 or above
7372
#elif CL_HPP_TARGET_OPENCL_VERSION >= 200
7373
useWithProperties = true;
7375
useWithProperties = false;
7378
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
7379
if (useWithProperties) {
7380
cl_queue_properties queue_properties[] = {
7381
CL_QUEUE_PROPERTIES, static_cast<cl_queue_properties>(properties), 0 };
7382
object_ = ::clCreateCommandQueueWithProperties(
7383
context(), device(), queue_properties, &error);
7385
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
7390
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
7391
#if CL_HPP_MINIMUM_OPENCL_VERSION < 200
7392
if (!useWithProperties) {
7393
object_ = ::clCreateCommandQueue(
7394
context(), device(), static_cast<cl_command_queue_properties>(properties), &error);
7396
detail::errHandler(error, __CREATE_COMMAND_QUEUE_ERR);
7401
#endif // CL_HPP_MINIMUM_OPENCL_VERSION < 200
7404
static CommandQueue getDefault(cl_int * err = NULL)
7406
std::call_once(default_initialized_, makeDefault);
7407
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
7408
detail::errHandler(default_error_, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
7409
#else // CL_HPP_TARGET_OPENCL_VERSION >= 200
7410
detail::errHandler(default_error_, __CREATE_COMMAND_QUEUE_ERR);
7411
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 200
7413
*err = default_error_;
7419
* Modify the default command queue to be used by
7420
* subsequent operations.
7421
* Will only set the default if no default was previously created.
7422
* @return updated default command queue.
7423
* Should be compared to the passed value to ensure that it was updated.
7425
static CommandQueue setDefault(const CommandQueue &default_queue)
7427
std::call_once(default_initialized_, makeDefaultProvided, std::cref(default_queue));
7428
detail::errHandler(default_error_);
7435
/*! \brief Constructor from cl_mem - takes ownership.
7437
* \param retainObject will cause the constructor to retain its cl object.
7438
* Defaults to false to maintain compatibility with
7441
explicit CommandQueue(const cl_command_queue& commandQueue, bool retainObject = false) :
7442
detail::Wrapper<cl_type>(commandQueue, retainObject) { }
7444
CommandQueue& operator = (const cl_command_queue& rhs)
7446
detail::Wrapper<cl_type>::operator=(rhs);
7450
/*! \brief Copy constructor to forward copy to the superclass correctly.
7451
* Required for MSVC.
7453
CommandQueue(const CommandQueue& queue) : detail::Wrapper<cl_type>(queue) {}
7455
/*! \brief Copy assignment to forward copy to the superclass correctly.
7456
* Required for MSVC.
7458
CommandQueue& operator = (const CommandQueue &queue)
7460
detail::Wrapper<cl_type>::operator=(queue);
7464
/*! \brief Move constructor to forward move to the superclass correctly.
7465
* Required for MSVC.
7467
CommandQueue(CommandQueue&& queue) CL_HPP_NOEXCEPT_ : detail::Wrapper<cl_type>(std::move(queue)) {}
7469
/*! \brief Move assignment to forward move to the superclass correctly.
7470
* Required for MSVC.
7472
CommandQueue& operator = (CommandQueue &&queue)
7474
detail::Wrapper<cl_type>::operator=(std::move(queue));
7478
template <typename T>
7479
cl_int getInfo(cl_command_queue_info name, T* param) const
7481
return detail::errHandler(
7483
&::clGetCommandQueueInfo, object_, name, param),
7484
__GET_COMMAND_QUEUE_INFO_ERR);
7487
template <cl_int name> typename
7488
detail::param_traits<detail::cl_command_queue_info, name>::param_type
7489
getInfo(cl_int* err = NULL) const
7491
typename detail::param_traits<
7492
detail::cl_command_queue_info, name>::param_type param;
7493
cl_int result = getInfo(name, ¶m);
7500
cl_int enqueueReadBuffer(
7501
const Buffer& buffer,
7506
const vector<Event>* events = NULL,
7507
Event* event = NULL) const
7510
cl_int err = detail::errHandler(
7511
::clEnqueueReadBuffer(
7512
object_, buffer(), blocking, offset, size,
7514
(events != NULL) ? (cl_uint) events->size() : 0,
7515
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7516
(event != NULL) ? &tmp : NULL),
7517
__ENQUEUE_READ_BUFFER_ERR);
7519
if (event != NULL && err == CL_SUCCESS)
7525
cl_int enqueueWriteBuffer(
7526
const Buffer& buffer,
7531
const vector<Event>* events = NULL,
7532
Event* event = NULL) const
7535
cl_int err = detail::errHandler(
7536
::clEnqueueWriteBuffer(
7537
object_, buffer(), blocking, offset, size,
7539
(events != NULL) ? (cl_uint) events->size() : 0,
7540
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7541
(event != NULL) ? &tmp : NULL),
7542
__ENQUEUE_WRITE_BUFFER_ERR);
7544
if (event != NULL && err == CL_SUCCESS)
7550
cl_int enqueueCopyBuffer(
7553
size_type src_offset,
7554
size_type dst_offset,
7556
const vector<Event>* events = NULL,
7557
Event* event = NULL) const
7560
cl_int err = detail::errHandler(
7561
::clEnqueueCopyBuffer(
7562
object_, src(), dst(), src_offset, dst_offset, size,
7563
(events != NULL) ? (cl_uint) events->size() : 0,
7564
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7565
(event != NULL) ? &tmp : NULL),
7566
__ENQEUE_COPY_BUFFER_ERR);
7568
if (event != NULL && err == CL_SUCCESS)
7573
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
7574
cl_int enqueueReadBufferRect(
7575
const Buffer& buffer,
7577
const array<size_type, 3>& buffer_offset,
7578
const array<size_type, 3>& host_offset,
7579
const array<size_type, 3>& region,
7580
size_type buffer_row_pitch,
7581
size_type buffer_slice_pitch,
7582
size_type host_row_pitch,
7583
size_type host_slice_pitch,
7585
const vector<Event>* events = NULL,
7586
Event* event = NULL) const
7589
cl_int err = detail::errHandler(
7590
::clEnqueueReadBufferRect(
7594
buffer_offset.data(),
7602
(events != NULL) ? (cl_uint) events->size() : 0,
7603
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7604
(event != NULL) ? &tmp : NULL),
7605
__ENQUEUE_READ_BUFFER_RECT_ERR);
7607
if (event != NULL && err == CL_SUCCESS)
7613
cl_int enqueueWriteBufferRect(
7614
const Buffer& buffer,
7616
const array<size_type, 3>& buffer_offset,
7617
const array<size_type, 3>& host_offset,
7618
const array<size_type, 3>& region,
7619
size_type buffer_row_pitch,
7620
size_type buffer_slice_pitch,
7621
size_type host_row_pitch,
7622
size_type host_slice_pitch,
7624
const vector<Event>* events = NULL,
7625
Event* event = NULL) const
7628
cl_int err = detail::errHandler(
7629
::clEnqueueWriteBufferRect(
7633
buffer_offset.data(),
7641
(events != NULL) ? (cl_uint) events->size() : 0,
7642
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7643
(event != NULL) ? &tmp : NULL),
7644
__ENQUEUE_WRITE_BUFFER_RECT_ERR);
7646
if (event != NULL && err == CL_SUCCESS)
7652
cl_int enqueueCopyBufferRect(
7655
const array<size_type, 3>& src_origin,
7656
const array<size_type, 3>& dst_origin,
7657
const array<size_type, 3>& region,
7658
size_type src_row_pitch,
7659
size_type src_slice_pitch,
7660
size_type dst_row_pitch,
7661
size_type dst_slice_pitch,
7662
const vector<Event>* events = NULL,
7663
Event* event = NULL) const
7666
cl_int err = detail::errHandler(
7667
::clEnqueueCopyBufferRect(
7678
(events != NULL) ? (cl_uint) events->size() : 0,
7679
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7680
(event != NULL) ? &tmp : NULL),
7681
__ENQEUE_COPY_BUFFER_RECT_ERR);
7683
if (event != NULL && err == CL_SUCCESS)
7688
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 110
7689
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
7691
* Enqueue a command to fill a buffer object with a pattern
7692
* of a given size. The pattern is specified as a vector type.
7693
* \tparam PatternType The datatype of the pattern field.
7694
* The pattern type must be an accepted OpenCL data type.
7695
* \tparam offset Is the offset in bytes into the buffer at
7696
* which to start filling. This must be a multiple of
7698
* \tparam size Is the size in bytes of the region to fill.
7699
* This must be a multiple of the pattern size.
7701
template<typename PatternType>
7702
cl_int enqueueFillBuffer(
7703
const Buffer& buffer,
7704
PatternType pattern,
7707
const vector<Event>* events = NULL,
7708
Event* event = NULL) const
7711
cl_int err = detail::errHandler(
7712
::clEnqueueFillBuffer(
7715
static_cast<void*>(&pattern),
7716
sizeof(PatternType),
7719
(events != NULL) ? (cl_uint) events->size() : 0,
7720
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7721
(event != NULL) ? &tmp : NULL),
7722
__ENQUEUE_FILL_BUFFER_ERR);
7724
if (event != NULL && err == CL_SUCCESS)
7729
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
7731
cl_int enqueueReadImage(
7734
const array<size_type, 3>& origin,
7735
const array<size_type, 3>& region,
7736
size_type row_pitch,
7737
size_type slice_pitch,
7739
const vector<Event>* events = NULL,
7740
Event* event = NULL) const
7743
cl_int err = detail::errHandler(
7744
::clEnqueueReadImage(
7753
(events != NULL) ? (cl_uint) events->size() : 0,
7754
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7755
(event != NULL) ? &tmp : NULL),
7756
__ENQUEUE_READ_IMAGE_ERR);
7758
if (event != NULL && err == CL_SUCCESS)
7764
cl_int enqueueWriteImage(
7767
const array<size_type, 3>& origin,
7768
const array<size_type, 3>& region,
7769
size_type row_pitch,
7770
size_type slice_pitch,
7772
const vector<Event>* events = NULL,
7773
Event* event = NULL) const
7776
cl_int err = detail::errHandler(
7777
::clEnqueueWriteImage(
7786
(events != NULL) ? (cl_uint) events->size() : 0,
7787
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7788
(event != NULL) ? &tmp : NULL),
7789
__ENQUEUE_WRITE_IMAGE_ERR);
7791
if (event != NULL && err == CL_SUCCESS)
7797
cl_int enqueueCopyImage(
7800
const array<size_type, 3>& src_origin,
7801
const array<size_type, 3>& dst_origin,
7802
const array<size_type, 3>& region,
7803
const vector<Event>* events = NULL,
7804
Event* event = NULL) const
7807
cl_int err = detail::errHandler(
7808
::clEnqueueCopyImage(
7815
(events != NULL) ? (cl_uint) events->size() : 0,
7816
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7817
(event != NULL) ? &tmp : NULL),
7818
__ENQUEUE_COPY_IMAGE_ERR);
7820
if (event != NULL && err == CL_SUCCESS)
7826
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
7828
* Enqueue a command to fill an image object with a specified color.
7829
* \param fillColor is the color to use to fill the image.
7830
* This is a four component RGBA floating-point color value if
7831
* the image channel data type is not an unnormalized signed or
7832
* unsigned data type.
7834
cl_int enqueueFillImage(
7836
cl_float4 fillColor,
7837
const array<size_type, 3>& origin,
7838
const array<size_type, 3>& region,
7839
const vector<Event>* events = NULL,
7840
Event* event = NULL) const
7843
cl_int err = detail::errHandler(
7844
::clEnqueueFillImage(
7847
static_cast<void*>(&fillColor),
7850
(events != NULL) ? (cl_uint) events->size() : 0,
7851
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7852
(event != NULL) ? &tmp : NULL),
7853
__ENQUEUE_FILL_IMAGE_ERR);
7855
if (event != NULL && err == CL_SUCCESS)
7862
* Enqueue a command to fill an image object with a specified color.
7863
* \param fillColor is the color to use to fill the image.
7864
* This is a four component RGBA signed integer color value if
7865
* the image channel data type is an unnormalized signed integer
7868
cl_int enqueueFillImage(
7871
const array<size_type, 3>& origin,
7872
const array<size_type, 3>& region,
7873
const vector<Event>* events = NULL,
7874
Event* event = NULL) const
7877
cl_int err = detail::errHandler(
7878
::clEnqueueFillImage(
7881
static_cast<void*>(&fillColor),
7884
(events != NULL) ? (cl_uint) events->size() : 0,
7885
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7886
(event != NULL) ? &tmp : NULL),
7887
__ENQUEUE_FILL_IMAGE_ERR);
7889
if (event != NULL && err == CL_SUCCESS)
7896
* Enqueue a command to fill an image object with a specified color.
7897
* \param fillColor is the color to use to fill the image.
7898
* This is a four component RGBA unsigned integer color value if
7899
* the image channel data type is an unnormalized unsigned integer
7902
cl_int enqueueFillImage(
7905
const array<size_type, 3>& origin,
7906
const array<size_type, 3>& region,
7907
const vector<Event>* events = NULL,
7908
Event* event = NULL) const
7911
cl_int err = detail::errHandler(
7912
::clEnqueueFillImage(
7915
static_cast<void*>(&fillColor),
7918
(events != NULL) ? (cl_uint) events->size() : 0,
7919
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7920
(event != NULL) ? &tmp : NULL),
7921
__ENQUEUE_FILL_IMAGE_ERR);
7923
if (event != NULL && err == CL_SUCCESS)
7928
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
7930
cl_int enqueueCopyImageToBuffer(
7933
const array<size_type, 3>& src_origin,
7934
const array<size_type, 3>& region,
7935
size_type dst_offset,
7936
const vector<Event>* events = NULL,
7937
Event* event = NULL) const
7940
cl_int err = detail::errHandler(
7941
::clEnqueueCopyImageToBuffer(
7948
(events != NULL) ? (cl_uint) events->size() : 0,
7949
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7950
(event != NULL) ? &tmp : NULL),
7951
__ENQUEUE_COPY_IMAGE_TO_BUFFER_ERR);
7953
if (event != NULL && err == CL_SUCCESS)
7959
cl_int enqueueCopyBufferToImage(
7962
size_type src_offset,
7963
const array<size_type, 3>& dst_origin,
7964
const array<size_type, 3>& region,
7965
const vector<Event>* events = NULL,
7966
Event* event = NULL) const
7969
cl_int err = detail::errHandler(
7970
::clEnqueueCopyBufferToImage(
7977
(events != NULL) ? (cl_uint) events->size() : 0,
7978
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
7979
(event != NULL) ? &tmp : NULL),
7980
__ENQUEUE_COPY_BUFFER_TO_IMAGE_ERR);
7982
if (event != NULL && err == CL_SUCCESS)
7988
void* enqueueMapBuffer(
7989
const Buffer& buffer,
7994
const vector<Event>* events = NULL,
7995
Event* event = NULL,
7996
cl_int* err = NULL) const
8000
void * result = ::clEnqueueMapBuffer(
8001
object_, buffer(), blocking, flags, offset, size,
8002
(events != NULL) ? (cl_uint) events->size() : 0,
8003
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8004
(event != NULL) ? &tmp : NULL,
8007
detail::errHandler(error, __ENQUEUE_MAP_BUFFER_ERR);
8011
if (event != NULL && error == CL_SUCCESS)
8017
void* enqueueMapImage(
8018
const Image& buffer,
8021
const array<size_type, 3>& origin,
8022
const array<size_type, 3>& region,
8023
size_type * row_pitch,
8024
size_type * slice_pitch,
8025
const vector<Event>* events = NULL,
8026
Event* event = NULL,
8027
cl_int* err = NULL) const
8031
void * result = ::clEnqueueMapImage(
8032
object_, buffer(), blocking, flags,
8035
row_pitch, slice_pitch,
8036
(events != NULL) ? (cl_uint) events->size() : 0,
8037
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8038
(event != NULL) ? &tmp : NULL,
8041
detail::errHandler(error, __ENQUEUE_MAP_IMAGE_ERR);
8045
if (event != NULL && error == CL_SUCCESS)
8050
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
8052
* Enqueues a command that will allow the host to update a region of a coarse-grained SVM buffer.
8053
* This variant takes a raw SVM pointer.
8055
template<typename T>
8056
cl_int enqueueMapSVM(
8061
const vector<Event>* events = NULL,
8062
Event* event = NULL) const
8065
cl_int err = detail::errHandler(::clEnqueueSVMMap(
8066
object_, blocking, flags, static_cast<void*>(ptr), size,
8067
(events != NULL) ? (cl_uint)events->size() : 0,
8068
(events != NULL && events->size() > 0) ? (cl_event*)&events->front() : NULL,
8069
(event != NULL) ? &tmp : NULL),
8070
__ENQUEUE_MAP_BUFFER_ERR);
8072
if (event != NULL && err == CL_SUCCESS)
8080
* Enqueues a command that will allow the host to update a region of a coarse-grained SVM buffer.
8081
* This variant takes a cl::pointer instance.
8083
template<typename T, class D>
8084
cl_int enqueueMapSVM(
8085
cl::pointer<T, D> &ptr,
8089
const vector<Event>* events = NULL,
8090
Event* event = NULL) const
8093
cl_int err = detail::errHandler(::clEnqueueSVMMap(
8094
object_, blocking, flags, static_cast<void*>(ptr.get()), size,
8095
(events != NULL) ? (cl_uint)events->size() : 0,
8096
(events != NULL && events->size() > 0) ? (cl_event*)&events->front() : NULL,
8097
(event != NULL) ? &tmp : NULL),
8098
__ENQUEUE_MAP_BUFFER_ERR);
8100
if (event != NULL && err == CL_SUCCESS)
8107
* Enqueues a command that will allow the host to update a region of a coarse-grained SVM buffer.
8108
* This variant takes a cl::vector instance.
8110
template<typename T, class Alloc>
8111
cl_int enqueueMapSVM(
8112
cl::vector<T, Alloc> &container,
8115
const vector<Event>* events = NULL,
8116
Event* event = NULL) const
8119
cl_int err = detail::errHandler(::clEnqueueSVMMap(
8120
object_, blocking, flags, static_cast<void*>(container.data()), container.size(),
8121
(events != NULL) ? (cl_uint)events->size() : 0,
8122
(events != NULL && events->size() > 0) ? (cl_event*)&events->front() : NULL,
8123
(event != NULL) ? &tmp : NULL),
8124
__ENQUEUE_MAP_BUFFER_ERR);
8126
if (event != NULL && err == CL_SUCCESS)
8131
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
8133
cl_int enqueueUnmapMemObject(
8134
const Memory& memory,
8136
const vector<Event>* events = NULL,
8137
Event* event = NULL) const
8140
cl_int err = detail::errHandler(
8141
::clEnqueueUnmapMemObject(
8142
object_, memory(), mapped_ptr,
8143
(events != NULL) ? (cl_uint) events->size() : 0,
8144
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8145
(event != NULL) ? &tmp : NULL),
8146
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
8148
if (event != NULL && err == CL_SUCCESS)
8155
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
8157
* Enqueues a command that will release a coarse-grained SVM buffer back to the OpenCL runtime.
8158
* This variant takes a raw SVM pointer.
8160
template<typename T>
8161
cl_int enqueueUnmapSVM(
8163
const vector<Event>* events = NULL,
8164
Event* event = NULL) const
8167
cl_int err = detail::errHandler(
8168
::clEnqueueSVMUnmap(
8169
object_, static_cast<void*>(ptr),
8170
(events != NULL) ? (cl_uint)events->size() : 0,
8171
(events != NULL && events->size() > 0) ? (cl_event*)&events->front() : NULL,
8172
(event != NULL) ? &tmp : NULL),
8173
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
8175
if (event != NULL && err == CL_SUCCESS)
8182
* Enqueues a command that will release a coarse-grained SVM buffer back to the OpenCL runtime.
8183
* This variant takes a cl::pointer instance.
8185
template<typename T, class D>
8186
cl_int enqueueUnmapSVM(
8187
cl::pointer<T, D> &ptr,
8188
const vector<Event>* events = NULL,
8189
Event* event = NULL) const
8192
cl_int err = detail::errHandler(
8193
::clEnqueueSVMUnmap(
8194
object_, static_cast<void*>(ptr.get()),
8195
(events != NULL) ? (cl_uint)events->size() : 0,
8196
(events != NULL && events->size() > 0) ? (cl_event*)&events->front() : NULL,
8197
(event != NULL) ? &tmp : NULL),
8198
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
8200
if (event != NULL && err == CL_SUCCESS)
8207
* Enqueues a command that will release a coarse-grained SVM buffer back to the OpenCL runtime.
8208
* This variant takes a cl::vector instance.
8210
template<typename T, class Alloc>
8211
cl_int enqueueUnmapSVM(
8212
cl::vector<T, Alloc> &container,
8213
const vector<Event>* events = NULL,
8214
Event* event = NULL) const
8217
cl_int err = detail::errHandler(
8218
::clEnqueueSVMUnmap(
8219
object_, static_cast<void*>(container.data()),
8220
(events != NULL) ? (cl_uint)events->size() : 0,
8221
(events != NULL && events->size() > 0) ? (cl_event*)&events->front() : NULL,
8222
(event != NULL) ? &tmp : NULL),
8223
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
8225
if (event != NULL && err == CL_SUCCESS)
8230
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
8232
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
8234
* Enqueues a marker command which waits for either a list of events to complete,
8235
* or all previously enqueued commands to complete.
8237
* Enqueues a marker command which waits for either a list of events to complete,
8238
* or if the list is empty it waits for all commands previously enqueued in command_queue
8239
* to complete before it completes. This command returns an event which can be waited on,
8240
* i.e. this event can be waited on to insure that all events either in the event_wait_list
8241
* or all previously enqueued commands, queued before this command to command_queue,
8244
cl_int enqueueMarkerWithWaitList(
8245
const vector<Event> *events = 0,
8246
Event *event = 0) const
8249
cl_int err = detail::errHandler(
8250
::clEnqueueMarkerWithWaitList(
8252
(events != NULL) ? (cl_uint) events->size() : 0,
8253
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8254
(event != NULL) ? &tmp : NULL),
8255
__ENQUEUE_MARKER_WAIT_LIST_ERR);
8257
if (event != NULL && err == CL_SUCCESS)
8264
* A synchronization point that enqueues a barrier operation.
8266
* Enqueues a barrier command which waits for either a list of events to complete,
8267
* or if the list is empty it waits for all commands previously enqueued in command_queue
8268
* to complete before it completes. This command blocks command execution, that is, any
8269
* following commands enqueued after it do not execute until it completes. This command
8270
* returns an event which can be waited on, i.e. this event can be waited on to insure that
8271
* all events either in the event_wait_list or all previously enqueued commands, queued
8272
* before this command to command_queue, have completed.
8274
cl_int enqueueBarrierWithWaitList(
8275
const vector<Event> *events = 0,
8276
Event *event = 0) const
8279
cl_int err = detail::errHandler(
8280
::clEnqueueBarrierWithWaitList(
8282
(events != NULL) ? (cl_uint) events->size() : 0,
8283
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8284
(event != NULL) ? &tmp : NULL),
8285
__ENQUEUE_BARRIER_WAIT_LIST_ERR);
8287
if (event != NULL && err == CL_SUCCESS)
8294
* Enqueues a command to indicate with which device a set of memory objects
8295
* should be associated.
8297
cl_int enqueueMigrateMemObjects(
8298
const vector<Memory> &memObjects,
8299
cl_mem_migration_flags flags,
8300
const vector<Event>* events = NULL,
8306
vector<cl_mem> localMemObjects(memObjects.size());
8308
for( int i = 0; i < (int)memObjects.size(); ++i ) {
8309
localMemObjects[i] = memObjects[i]();
8312
cl_int err = detail::errHandler(
8313
::clEnqueueMigrateMemObjects(
8315
(cl_uint)memObjects.size(),
8316
localMemObjects.data(),
8318
(events != NULL) ? (cl_uint) events->size() : 0,
8319
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8320
(event != NULL) ? &tmp : NULL),
8321
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
8323
if (event != NULL && err == CL_SUCCESS)
8328
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
8331
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
8333
* Enqueues a command that will allow the host associate ranges within a set of
8334
* SVM allocations with a device.
8335
* @param sizes - The length from each pointer to migrate.
8337
template<typename T>
8338
cl_int enqueueMigrateSVM(
8339
const cl::vector<T*> &svmRawPointers,
8340
const cl::vector<size_type> &sizes,
8341
cl_mem_migration_flags flags = 0,
8342
const vector<Event>* events = NULL,
8343
Event* event = NULL) const
8346
cl_int err = detail::errHandler(::clEnqueueSVMMigrateMem(
8348
svmRawPointers.size(), static_cast<void**>(svmRawPointers.data()),
8349
sizes.data(), // array of sizes not passed
8351
(events != NULL) ? (cl_uint)events->size() : 0,
8352
(events != NULL && events->size() > 0) ? (cl_event*)&events->front() : NULL,
8353
(event != NULL) ? &tmp : NULL),
8354
__ENQUEUE_MIGRATE_SVM_ERR);
8356
if (event != NULL && err == CL_SUCCESS)
8363
* Enqueues a command that will allow the host associate a set of SVM allocations with
8366
template<typename T>
8367
cl_int enqueueMigrateSVM(
8368
const cl::vector<T*> &svmRawPointers,
8369
cl_mem_migration_flags flags = 0,
8370
const vector<Event>* events = NULL,
8371
Event* event = NULL) const
8373
return enqueueMigrateSVM(svmRawPointers, cl::vector<size_type>(svmRawPointers.size()), flags, events, event);
8378
* Enqueues a command that will allow the host associate ranges within a set of
8379
* SVM allocations with a device.
8380
* @param sizes - The length from each pointer to migrate.
8382
template<typename T, class D>
8383
cl_int enqueueMigrateSVM(
8384
const cl::vector<cl::pointer<T, D>> &svmPointers,
8385
const cl::vector<size_type> &sizes,
8386
cl_mem_migration_flags flags = 0,
8387
const vector<Event>* events = NULL,
8388
Event* event = NULL) const
8390
cl::vector<void*> svmRawPointers;
8391
svmRawPointers.reserve(svmPointers.size());
8392
for (auto p : svmPointers) {
8393
svmRawPointers.push_back(static_cast<void*>(p.get()));
8396
return enqueueMigrateSVM(svmRawPointers, sizes, flags, events, event);
8401
* Enqueues a command that will allow the host associate a set of SVM allocations with
8404
template<typename T, class D>
8405
cl_int enqueueMigrateSVM(
8406
const cl::vector<cl::pointer<T, D>> &svmPointers,
8407
cl_mem_migration_flags flags = 0,
8408
const vector<Event>* events = NULL,
8409
Event* event = NULL) const
8411
return enqueueMigrateSVM(svmPointers, cl::vector<size_type>(svmPointers.size()), flags, events, event);
8415
* Enqueues a command that will allow the host associate ranges within a set of
8416
* SVM allocations with a device.
8417
* @param sizes - The length from the beginning of each container to migrate.
8419
template<typename T, class Alloc>
8420
cl_int enqueueMigrateSVM(
8421
const cl::vector<cl::vector<T, Alloc>> &svmContainers,
8422
const cl::vector<size_type> &sizes,
8423
cl_mem_migration_flags flags = 0,
8424
const vector<Event>* events = NULL,
8425
Event* event = NULL) const
8427
cl::vector<void*> svmRawPointers;
8428
svmRawPointers.reserve(svmContainers.size());
8429
for (auto p : svmContainers) {
8430
svmRawPointers.push_back(static_cast<void*>(p.data()));
8433
return enqueueMigrateSVM(svmRawPointers, sizes, flags, events, event);
8437
* Enqueues a command that will allow the host associate a set of SVM allocations with
8440
template<typename T, class Alloc>
8441
cl_int enqueueMigrateSVM(
8442
const cl::vector<cl::vector<T, Alloc>> &svmContainers,
8443
cl_mem_migration_flags flags = 0,
8444
const vector<Event>* events = NULL,
8445
Event* event = NULL) const
8447
return enqueueMigrateSVM(svmContainers, cl::vector<size_type>(svmContainers.size()), flags, events, event);
8450
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
8452
cl_int enqueueNDRangeKernel(
8453
const Kernel& kernel,
8454
const NDRange& offset,
8455
const NDRange& global,
8456
const NDRange& local = NullRange,
8457
const vector<Event>* events = NULL,
8458
Event* event = NULL) const
8461
cl_int err = detail::errHandler(
8462
::clEnqueueNDRangeKernel(
8463
object_, kernel(), (cl_uint) global.dimensions(),
8464
offset.dimensions() != 0 ? (const size_type*) offset : NULL,
8465
(const size_type*) global,
8466
local.dimensions() != 0 ? (const size_type*) local : NULL,
8467
(events != NULL) ? (cl_uint) events->size() : 0,
8468
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8469
(event != NULL) ? &tmp : NULL),
8470
__ENQUEUE_NDRANGE_KERNEL_ERR);
8472
if (event != NULL && err == CL_SUCCESS)
8478
#if defined(CL_USE_DEPRECATED_OPENCL_1_2_APIS)
8479
CL_EXT_PREFIX__VERSION_1_2_DEPRECATED cl_int enqueueTask(
8480
const Kernel& kernel,
8481
const vector<Event>* events = NULL,
8482
Event* event = NULL) const CL_EXT_SUFFIX__VERSION_1_2_DEPRECATED
8485
cl_int err = detail::errHandler(
8488
(events != NULL) ? (cl_uint) events->size() : 0,
8489
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8490
(event != NULL) ? &tmp : NULL),
8491
__ENQUEUE_TASK_ERR);
8493
if (event != NULL && err == CL_SUCCESS)
8498
#endif // #if defined(CL_USE_DEPRECATED_OPENCL_1_2_APIS)
8500
cl_int enqueueNativeKernel(
8501
void (CL_CALLBACK *userFptr)(void *),
8502
std::pair<void*, size_type> args,
8503
const vector<Memory>* mem_objects = NULL,
8504
const vector<const void*>* mem_locs = NULL,
8505
const vector<Event>* events = NULL,
8506
Event* event = NULL) const
8508
size_type elements = 0;
8509
if (mem_objects != NULL) {
8510
elements = mem_objects->size();
8512
vector<cl_mem> mems(elements);
8513
for (unsigned int i = 0; i < elements; i++) {
8514
mems[i] = ((*mem_objects)[i])();
8518
cl_int err = detail::errHandler(
8519
::clEnqueueNativeKernel(
8520
object_, userFptr, args.first, args.second,
8521
(mem_objects != NULL) ? (cl_uint) mem_objects->size() : 0,
8523
(mem_locs != NULL && mem_locs->size() > 0) ? (const void **) &mem_locs->front() : NULL,
8524
(events != NULL) ? (cl_uint) events->size() : 0,
8525
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8526
(event != NULL) ? &tmp : NULL),
8527
__ENQUEUE_NATIVE_KERNEL);
8529
if (event != NULL && err == CL_SUCCESS)
8536
* Deprecated APIs for 1.2
8538
#if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
8539
CL_EXT_PREFIX__VERSION_1_1_DEPRECATED
8540
cl_int enqueueMarker(Event* event = NULL) const CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED
8543
cl_int err = detail::errHandler(
8546
(event != NULL) ? &tmp : NULL),
8547
__ENQUEUE_MARKER_ERR);
8549
if (event != NULL && err == CL_SUCCESS)
8555
CL_EXT_PREFIX__VERSION_1_1_DEPRECATED
8556
cl_int enqueueWaitForEvents(const vector<Event>& events) const CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED
8558
return detail::errHandler(
8559
::clEnqueueWaitForEvents(
8561
(cl_uint) events.size(),
8562
events.size() > 0 ? (const cl_event*) &events.front() : NULL),
8563
__ENQUEUE_WAIT_FOR_EVENTS_ERR);
8565
#endif // defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
8567
cl_int enqueueAcquireGLObjects(
8568
const vector<Memory>* mem_objects = NULL,
8569
const vector<Event>* events = NULL,
8570
Event* event = NULL) const
8573
cl_int err = detail::errHandler(
8574
::clEnqueueAcquireGLObjects(
8576
(mem_objects != NULL) ? (cl_uint) mem_objects->size() : 0,
8577
(mem_objects != NULL && mem_objects->size() > 0) ? (const cl_mem *) &mem_objects->front(): NULL,
8578
(events != NULL) ? (cl_uint) events->size() : 0,
8579
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8580
(event != NULL) ? &tmp : NULL),
8581
__ENQUEUE_ACQUIRE_GL_ERR);
8583
if (event != NULL && err == CL_SUCCESS)
8589
cl_int enqueueReleaseGLObjects(
8590
const vector<Memory>* mem_objects = NULL,
8591
const vector<Event>* events = NULL,
8592
Event* event = NULL) const
8595
cl_int err = detail::errHandler(
8596
::clEnqueueReleaseGLObjects(
8598
(mem_objects != NULL) ? (cl_uint) mem_objects->size() : 0,
8599
(mem_objects != NULL && mem_objects->size() > 0) ? (const cl_mem *) &mem_objects->front(): NULL,
8600
(events != NULL) ? (cl_uint) events->size() : 0,
8601
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8602
(event != NULL) ? &tmp : NULL),
8603
__ENQUEUE_RELEASE_GL_ERR);
8605
if (event != NULL && err == CL_SUCCESS)
8611
#if defined (CL_HPP_USE_DX_INTEROP)
8612
typedef CL_API_ENTRY cl_int (CL_API_CALL *PFN_clEnqueueAcquireD3D10ObjectsKHR)(
8613
cl_command_queue command_queue, cl_uint num_objects,
8614
const cl_mem* mem_objects, cl_uint num_events_in_wait_list,
8615
const cl_event* event_wait_list, cl_event* event);
8616
typedef CL_API_ENTRY cl_int (CL_API_CALL *PFN_clEnqueueReleaseD3D10ObjectsKHR)(
8617
cl_command_queue command_queue, cl_uint num_objects,
8618
const cl_mem* mem_objects, cl_uint num_events_in_wait_list,
8619
const cl_event* event_wait_list, cl_event* event);
8621
cl_int enqueueAcquireD3D10Objects(
8622
const vector<Memory>* mem_objects = NULL,
8623
const vector<Event>* events = NULL,
8624
Event* event = NULL) const
8626
static PFN_clEnqueueAcquireD3D10ObjectsKHR pfn_clEnqueueAcquireD3D10ObjectsKHR = NULL;
8627
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
8628
cl_context context = getInfo<CL_QUEUE_CONTEXT>();
8629
cl::Device device(getInfo<CL_QUEUE_DEVICE>());
8630
cl_platform_id platform = device.getInfo<CL_DEVICE_PLATFORM>();
8631
CL_HPP_INIT_CL_EXT_FCN_PTR_PLATFORM_(platform, clEnqueueAcquireD3D10ObjectsKHR);
8633
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
8634
CL_HPP_INIT_CL_EXT_FCN_PTR_(clEnqueueAcquireD3D10ObjectsKHR);
8638
cl_int err = detail::errHandler(
8639
pfn_clEnqueueAcquireD3D10ObjectsKHR(
8641
(mem_objects != NULL) ? (cl_uint) mem_objects->size() : 0,
8642
(mem_objects != NULL && mem_objects->size() > 0) ? (const cl_mem *) &mem_objects->front(): NULL,
8643
(events != NULL) ? (cl_uint) events->size() : 0,
8644
(events != NULL) ? (cl_event*) &events->front() : NULL,
8645
(event != NULL) ? &tmp : NULL),
8646
__ENQUEUE_ACQUIRE_GL_ERR);
8648
if (event != NULL && err == CL_SUCCESS)
8654
cl_int enqueueReleaseD3D10Objects(
8655
const vector<Memory>* mem_objects = NULL,
8656
const vector<Event>* events = NULL,
8657
Event* event = NULL) const
8659
static PFN_clEnqueueReleaseD3D10ObjectsKHR pfn_clEnqueueReleaseD3D10ObjectsKHR = NULL;
8660
#if CL_HPP_TARGET_OPENCL_VERSION >= 120
8661
cl_context context = getInfo<CL_QUEUE_CONTEXT>();
8662
cl::Device device(getInfo<CL_QUEUE_DEVICE>());
8663
cl_platform_id platform = device.getInfo<CL_DEVICE_PLATFORM>();
8664
CL_HPP_INIT_CL_EXT_FCN_PTR_PLATFORM_(platform, clEnqueueReleaseD3D10ObjectsKHR);
8665
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 120
8666
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
8667
CL_HPP_INIT_CL_EXT_FCN_PTR_(clEnqueueReleaseD3D10ObjectsKHR);
8668
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 110
8671
cl_int err = detail::errHandler(
8672
pfn_clEnqueueReleaseD3D10ObjectsKHR(
8674
(mem_objects != NULL) ? (cl_uint) mem_objects->size() : 0,
8675
(mem_objects != NULL && mem_objects->size() > 0) ? (const cl_mem *) &mem_objects->front(): NULL,
8676
(events != NULL) ? (cl_uint) events->size() : 0,
8677
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
8678
(event != NULL) ? &tmp : NULL),
8679
__ENQUEUE_RELEASE_GL_ERR);
8681
if (event != NULL && err == CL_SUCCESS)
8689
* Deprecated APIs for 1.2
8691
#if defined(CL_USE_DEPRECATED_OPENCL_1_1_APIS)
8692
CL_EXT_PREFIX__VERSION_1_1_DEPRECATED
8693
cl_int enqueueBarrier() const CL_EXT_SUFFIX__VERSION_1_1_DEPRECATED
8695
return detail::errHandler(
8696
::clEnqueueBarrier(object_),
8697
__ENQUEUE_BARRIER_ERR);
8699
#endif // CL_USE_DEPRECATED_OPENCL_1_1_APIS
8701
cl_int flush() const
8703
return detail::errHandler(::clFlush(object_), __FLUSH_ERR);
8706
cl_int finish() const
8708
return detail::errHandler(::clFinish(object_), __FINISH_ERR);
8712
CL_HPP_DEFINE_STATIC_MEMBER_ std::once_flag CommandQueue::default_initialized_;
8713
CL_HPP_DEFINE_STATIC_MEMBER_ CommandQueue CommandQueue::default_;
8714
CL_HPP_DEFINE_STATIC_MEMBER_ cl_int CommandQueue::default_error_ = CL_SUCCESS;
8717
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
8718
enum class DeviceQueueProperties : cl_command_queue_properties
8721
Profiling = CL_QUEUE_PROFILING_ENABLE,
8724
inline DeviceQueueProperties operator|(DeviceQueueProperties lhs, DeviceQueueProperties rhs)
8726
return static_cast<DeviceQueueProperties>(static_cast<cl_command_queue_properties>(lhs) | static_cast<cl_command_queue_properties>(rhs));
8729
/*! \class DeviceCommandQueue
8730
* \brief DeviceCommandQueue interface for device cl_command_queues.
8732
class DeviceCommandQueue : public detail::Wrapper<cl_command_queue>
8737
* Trivial empty constructor to create a null queue.
8739
DeviceCommandQueue() { }
8742
* Default construct device command queue on default context and device
8744
DeviceCommandQueue(DeviceQueueProperties properties, cl_int* err = NULL)
8747
cl::Context context = cl::Context::getDefault();
8748
cl::Device device = cl::Device::getDefault();
8750
cl_command_queue_properties mergedProperties =
8751
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_ON_DEVICE | static_cast<cl_command_queue_properties>(properties);
8753
cl_queue_properties queue_properties[] = {
8754
CL_QUEUE_PROPERTIES, mergedProperties, 0 };
8755
object_ = ::clCreateCommandQueueWithProperties(
8756
context(), device(), queue_properties, &error);
8758
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
8765
* Create a device command queue for a specified device in the passed context.
8768
const Context& context,
8769
const Device& device,
8770
DeviceQueueProperties properties = DeviceQueueProperties::None,
8775
cl_command_queue_properties mergedProperties =
8776
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_ON_DEVICE | static_cast<cl_command_queue_properties>(properties);
8777
cl_queue_properties queue_properties[] = {
8778
CL_QUEUE_PROPERTIES, mergedProperties, 0 };
8779
object_ = ::clCreateCommandQueueWithProperties(
8780
context(), device(), queue_properties, &error);
8782
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
8789
* Create a device command queue for a specified device in the passed context.
8792
const Context& context,
8793
const Device& device,
8795
DeviceQueueProperties properties = DeviceQueueProperties::None,
8800
cl_command_queue_properties mergedProperties =
8801
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_ON_DEVICE | static_cast<cl_command_queue_properties>(properties);
8802
cl_queue_properties queue_properties[] = {
8803
CL_QUEUE_PROPERTIES, mergedProperties,
8804
CL_QUEUE_SIZE, queueSize,
8806
object_ = ::clCreateCommandQueueWithProperties(
8807
context(), device(), queue_properties, &error);
8809
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
8815
/*! \brief Constructor from cl_command_queue - takes ownership.
8817
* \param retainObject will cause the constructor to retain its cl object.
8818
* Defaults to false to maintain compatibility with
8821
explicit DeviceCommandQueue(const cl_command_queue& commandQueue, bool retainObject = false) :
8822
detail::Wrapper<cl_type>(commandQueue, retainObject) { }
8824
DeviceCommandQueue& operator = (const cl_command_queue& rhs)
8826
detail::Wrapper<cl_type>::operator=(rhs);
8830
/*! \brief Copy constructor to forward copy to the superclass correctly.
8831
* Required for MSVC.
8833
DeviceCommandQueue(const DeviceCommandQueue& queue) : detail::Wrapper<cl_type>(queue) {}
8835
/*! \brief Copy assignment to forward copy to the superclass correctly.
8836
* Required for MSVC.
8838
DeviceCommandQueue& operator = (const DeviceCommandQueue &queue)
8840
detail::Wrapper<cl_type>::operator=(queue);
8844
/*! \brief Move constructor to forward move to the superclass correctly.
8845
* Required for MSVC.
8847
DeviceCommandQueue(DeviceCommandQueue&& queue) CL_HPP_NOEXCEPT_ : detail::Wrapper<cl_type>(std::move(queue)) {}
8849
/*! \brief Move assignment to forward move to the superclass correctly.
8850
* Required for MSVC.
8852
DeviceCommandQueue& operator = (DeviceCommandQueue &&queue)
8854
detail::Wrapper<cl_type>::operator=(std::move(queue));
8858
template <typename T>
8859
cl_int getInfo(cl_command_queue_info name, T* param) const
8861
return detail::errHandler(
8863
&::clGetCommandQueueInfo, object_, name, param),
8864
__GET_COMMAND_QUEUE_INFO_ERR);
8867
template <cl_int name> typename
8868
detail::param_traits<detail::cl_command_queue_info, name>::param_type
8869
getInfo(cl_int* err = NULL) const
8871
typename detail::param_traits<
8872
detail::cl_command_queue_info, name>::param_type param;
8873
cl_int result = getInfo(name, ¶m);
8881
* Create a new default device command queue for the default device,
8882
* in the default context and of the default size.
8883
* If there is already a default queue for the specified device this
8884
* function will return the pre-existing queue.
8886
static DeviceCommandQueue makeDefault(
8887
cl_int *err = nullptr)
8890
cl::Context context = cl::Context::getDefault();
8891
cl::Device device = cl::Device::getDefault();
8893
cl_command_queue_properties properties =
8894
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_ON_DEVICE | CL_QUEUE_ON_DEVICE_DEFAULT;
8895
cl_queue_properties queue_properties[] = {
8896
CL_QUEUE_PROPERTIES, properties,
8898
DeviceCommandQueue deviceQueue(
8899
::clCreateCommandQueueWithProperties(
8900
context(), device(), queue_properties, &error));
8902
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
8911
* Create a new default device command queue for the specified device
8912
* and of the default size.
8913
* If there is already a default queue for the specified device this
8914
* function will return the pre-existing queue.
8916
static DeviceCommandQueue makeDefault(
8917
const Context &context, const Device &device, cl_int *err = nullptr)
8921
cl_command_queue_properties properties =
8922
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_ON_DEVICE | CL_QUEUE_ON_DEVICE_DEFAULT;
8923
cl_queue_properties queue_properties[] = {
8924
CL_QUEUE_PROPERTIES, properties,
8926
DeviceCommandQueue deviceQueue(
8927
::clCreateCommandQueueWithProperties(
8928
context(), device(), queue_properties, &error));
8930
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
8939
* Create a new default device command queue for the specified device
8940
* and of the requested size in bytes.
8941
* If there is already a default queue for the specified device this
8942
* function will return the pre-existing queue.
8944
static DeviceCommandQueue makeDefault(
8945
const Context &context, const Device &device, cl_uint queueSize, cl_int *err = nullptr)
8949
cl_command_queue_properties properties =
8950
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE | CL_QUEUE_ON_DEVICE | CL_QUEUE_ON_DEVICE_DEFAULT;
8951
cl_queue_properties queue_properties[] = {
8952
CL_QUEUE_PROPERTIES, properties,
8953
CL_QUEUE_SIZE, queueSize,
8955
DeviceCommandQueue deviceQueue(
8956
::clCreateCommandQueueWithProperties(
8957
context(), device(), queue_properties, &error));
8959
detail::errHandler(error, __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR);
8969
#if CL_HPP_TARGET_OPENCL_VERSION >= 210
8971
* Modify the default device command queue to be used for subsequent kernels.
8972
* This can update the default command queue for a device repeatedly to account
8973
* for kernels that rely on the default.
8974
* @return updated default device command queue.
8976
static DeviceCommandQueue updateDefault(const Context &context, const Device &device, const DeviceCommandQueue &default_queue, cl_int *err = nullptr)
8979
error = clSetDefaultDeviceCommandQueue(context.get(), device.get(), default_queue.get());
8981
detail::errHandler(error, __SET_DEFAULT_DEVICE_COMMAND_QUEUE_ERR);
8985
return default_queue;
8989
* Return the current default command queue for the specified command queue
8991
static DeviceCommandQueue getDefault(const CommandQueue &queue, cl_int * err = NULL)
8993
return queue.getInfo<CL_QUEUE_DEVICE_DEFAULT>(err);
8996
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 210
8997
}; // DeviceCommandQueue
9001
// Specialization for device command queue
9003
struct KernelArgumentHandler<cl::DeviceCommandQueue, void>
9005
static size_type size(const cl::DeviceCommandQueue&) { return sizeof(cl_command_queue); }
9006
static const cl_command_queue* ptr(const cl::DeviceCommandQueue& value) { return &(value()); }
9008
} // namespace detail
9010
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
9013
template< typename IteratorType >
9015
const Context &context,
9016
IteratorType startIterator,
9017
IteratorType endIterator,
9022
typedef typename std::iterator_traits<IteratorType>::value_type DataType;
9025
cl_mem_flags flags = 0;
9027
flags |= CL_MEM_READ_ONLY;
9030
flags |= CL_MEM_READ_WRITE;
9033
flags |= CL_MEM_USE_HOST_PTR;
9036
size_type size = sizeof(DataType)*(endIterator - startIterator);
9039
object_ = ::clCreateBuffer(context(), flags, size, static_cast<DataType*>(&*startIterator), &error);
9041
object_ = ::clCreateBuffer(context(), flags, size, 0, &error);
9044
detail::errHandler(error, __CREATE_BUFFER_ERR);
9050
CommandQueue queue(context, 0, &error);
9051
detail::errHandler(error, __CREATE_BUFFER_ERR);
9056
error = cl::copy(queue, startIterator, endIterator, *this);
9057
detail::errHandler(error, __CREATE_BUFFER_ERR);
9064
template< typename IteratorType >
9066
const CommandQueue &queue,
9067
IteratorType startIterator,
9068
IteratorType endIterator,
9073
typedef typename std::iterator_traits<IteratorType>::value_type DataType;
9076
cl_mem_flags flags = 0;
9078
flags |= CL_MEM_READ_ONLY;
9081
flags |= CL_MEM_READ_WRITE;
9084
flags |= CL_MEM_USE_HOST_PTR;
9087
size_type size = sizeof(DataType)*(endIterator - startIterator);
9089
Context context = queue.getInfo<CL_QUEUE_CONTEXT>();
9092
object_ = ::clCreateBuffer(context(), flags, size, static_cast<DataType*>(&*startIterator), &error);
9095
object_ = ::clCreateBuffer(context(), flags, size, 0, &error);
9098
detail::errHandler(error, __CREATE_BUFFER_ERR);
9104
error = cl::copy(queue, startIterator, endIterator, *this);
9105
detail::errHandler(error, __CREATE_BUFFER_ERR);
9112
inline cl_int enqueueReadBuffer(
9113
const Buffer& buffer,
9118
const vector<Event>* events = NULL,
9119
Event* event = NULL)
9122
CommandQueue queue = CommandQueue::getDefault(&error);
9124
if (error != CL_SUCCESS) {
9128
return queue.enqueueReadBuffer(buffer, blocking, offset, size, ptr, events, event);
9131
inline cl_int enqueueWriteBuffer(
9132
const Buffer& buffer,
9137
const vector<Event>* events = NULL,
9138
Event* event = NULL)
9141
CommandQueue queue = CommandQueue::getDefault(&error);
9143
if (error != CL_SUCCESS) {
9147
return queue.enqueueWriteBuffer(buffer, blocking, offset, size, ptr, events, event);
9150
inline void* enqueueMapBuffer(
9151
const Buffer& buffer,
9156
const vector<Event>* events = NULL,
9157
Event* event = NULL,
9161
CommandQueue queue = CommandQueue::getDefault(&error);
9162
detail::errHandler(error, __ENQUEUE_MAP_BUFFER_ERR);
9167
void * result = ::clEnqueueMapBuffer(
9168
queue(), buffer(), blocking, flags, offset, size,
9169
(events != NULL) ? (cl_uint) events->size() : 0,
9170
(events != NULL && events->size() > 0) ? (cl_event*) &events->front() : NULL,
9174
detail::errHandler(error, __ENQUEUE_MAP_BUFFER_ERR);
9182
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
9184
* Enqueues to the default queue a command that will allow the host to
9185
* update a region of a coarse-grained SVM buffer.
9186
* This variant takes a raw SVM pointer.
9188
template<typename T>
9189
inline cl_int enqueueMapSVM(
9194
const vector<Event>* events,
9198
CommandQueue queue = CommandQueue::getDefault(&error);
9199
if (error != CL_SUCCESS) {
9200
return detail::errHandler(error, __ENQUEUE_MAP_BUFFER_ERR);
9203
return queue.enqueueMapSVM(
9204
ptr, blocking, flags, size, events, event);
9208
* Enqueues to the default queue a command that will allow the host to
9209
* update a region of a coarse-grained SVM buffer.
9210
* This variant takes a cl::pointer instance.
9212
template<typename T, class D>
9213
inline cl_int enqueueMapSVM(
9214
cl::pointer<T, D> ptr,
9218
const vector<Event>* events = NULL,
9219
Event* event = NULL)
9222
CommandQueue queue = CommandQueue::getDefault(&error);
9223
if (error != CL_SUCCESS) {
9224
return detail::errHandler(error, __ENQUEUE_MAP_BUFFER_ERR);
9227
return queue.enqueueMapSVM(
9228
ptr, blocking, flags, size, events, event);
9232
* Enqueues to the default queue a command that will allow the host to
9233
* update a region of a coarse-grained SVM buffer.
9234
* This variant takes a cl::vector instance.
9236
template<typename T, class Alloc>
9237
inline cl_int enqueueMapSVM(
9238
cl::vector<T, Alloc> container,
9241
const vector<Event>* events = NULL,
9242
Event* event = NULL)
9245
CommandQueue queue = CommandQueue::getDefault(&error);
9246
if (error != CL_SUCCESS) {
9247
return detail::errHandler(error, __ENQUEUE_MAP_BUFFER_ERR);
9250
return queue.enqueueMapSVM(
9251
container, blocking, flags, events, event);
9254
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
9256
inline cl_int enqueueUnmapMemObject(
9257
const Memory& memory,
9259
const vector<Event>* events = NULL,
9260
Event* event = NULL)
9263
CommandQueue queue = CommandQueue::getDefault(&error);
9264
detail::errHandler(error, __ENQUEUE_MAP_BUFFER_ERR);
9265
if (error != CL_SUCCESS) {
9270
cl_int err = detail::errHandler(
9271
::clEnqueueUnmapMemObject(
9272
queue(), memory(), mapped_ptr,
9273
(events != NULL) ? (cl_uint)events->size() : 0,
9274
(events != NULL && events->size() > 0) ? (cl_event*)&events->front() : NULL,
9275
(event != NULL) ? &tmp : NULL),
9276
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
9278
if (event != NULL && err == CL_SUCCESS)
9284
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
9286
* Enqueues to the default queue a command that will release a coarse-grained
9287
* SVM buffer back to the OpenCL runtime.
9288
* This variant takes a raw SVM pointer.
9290
template<typename T>
9291
inline cl_int enqueueUnmapSVM(
9293
const vector<Event>* events = NULL,
9294
Event* event = NULL)
9297
CommandQueue queue = CommandQueue::getDefault(&error);
9298
if (error != CL_SUCCESS) {
9299
return detail::errHandler(error, __ENQUEUE_UNMAP_MEM_OBJECT_ERR);
9302
return detail::errHandler(queue.enqueueUnmapSVM(ptr, events, event),
9303
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
9308
* Enqueues to the default queue a command that will release a coarse-grained
9309
* SVM buffer back to the OpenCL runtime.
9310
* This variant takes a cl::pointer instance.
9312
template<typename T, class D>
9313
inline cl_int enqueueUnmapSVM(
9314
cl::pointer<T, D> &ptr,
9315
const vector<Event>* events = NULL,
9316
Event* event = NULL)
9319
CommandQueue queue = CommandQueue::getDefault(&error);
9320
if (error != CL_SUCCESS) {
9321
return detail::errHandler(error, __ENQUEUE_UNMAP_MEM_OBJECT_ERR);
9324
return detail::errHandler(queue.enqueueUnmapSVM(ptr, events, event),
9325
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
9329
* Enqueues to the default queue a command that will release a coarse-grained
9330
* SVM buffer back to the OpenCL runtime.
9331
* This variant takes a cl::vector instance.
9333
template<typename T, class Alloc>
9334
inline cl_int enqueueUnmapSVM(
9335
cl::vector<T, Alloc> &container,
9336
const vector<Event>* events = NULL,
9337
Event* event = NULL)
9340
CommandQueue queue = CommandQueue::getDefault(&error);
9341
if (error != CL_SUCCESS) {
9342
return detail::errHandler(error, __ENQUEUE_UNMAP_MEM_OBJECT_ERR);
9345
return detail::errHandler(queue.enqueueUnmapSVM(container, events, event),
9346
__ENQUEUE_UNMAP_MEM_OBJECT_ERR);
9349
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
9351
inline cl_int enqueueCopyBuffer(
9354
size_type src_offset,
9355
size_type dst_offset,
9357
const vector<Event>* events = NULL,
9358
Event* event = NULL)
9361
CommandQueue queue = CommandQueue::getDefault(&error);
9363
if (error != CL_SUCCESS) {
9367
return queue.enqueueCopyBuffer(src, dst, src_offset, dst_offset, size, events, event);
9371
* Blocking copy operation between iterators and a buffer.
9373
* Uses default command queue.
9375
template< typename IteratorType >
9376
inline cl_int copy( IteratorType startIterator, IteratorType endIterator, cl::Buffer &buffer )
9379
CommandQueue queue = CommandQueue::getDefault(&error);
9380
if (error != CL_SUCCESS)
9383
return cl::copy(queue, startIterator, endIterator, buffer);
9387
* Blocking copy operation between iterators and a buffer.
9389
* Uses default command queue.
9391
template< typename IteratorType >
9392
inline cl_int copy( const cl::Buffer &buffer, IteratorType startIterator, IteratorType endIterator )
9395
CommandQueue queue = CommandQueue::getDefault(&error);
9396
if (error != CL_SUCCESS)
9399
return cl::copy(queue, buffer, startIterator, endIterator);
9403
* Blocking copy operation between iterators and a buffer.
9405
* Uses specified queue.
9407
template< typename IteratorType >
9408
inline cl_int copy( const CommandQueue &queue, IteratorType startIterator, IteratorType endIterator, cl::Buffer &buffer )
9410
typedef typename std::iterator_traits<IteratorType>::value_type DataType;
9413
size_type length = endIterator-startIterator;
9414
size_type byteLength = length*sizeof(DataType);
9417
static_cast<DataType*>(queue.enqueueMapBuffer(buffer, CL_TRUE, CL_MAP_WRITE, 0, byteLength, 0, 0, &error));
9418
// if exceptions enabled, enqueueMapBuffer will throw
9419
if( error != CL_SUCCESS ) {
9422
#if defined(_MSC_VER)
9426
stdext::checked_array_iterator<DataType*>(
9429
std::copy(startIterator, endIterator, pointer);
9432
error = queue.enqueueUnmapMemObject(buffer, pointer, 0, &endEvent);
9433
// if exceptions enabled, enqueueUnmapMemObject will throw
9434
if( error != CL_SUCCESS ) {
9442
* Blocking copy operation between iterators and a buffer.
9444
* Uses specified queue.
9446
template< typename IteratorType >
9447
inline cl_int copy( const CommandQueue &queue, const cl::Buffer &buffer, IteratorType startIterator, IteratorType endIterator )
9449
typedef typename std::iterator_traits<IteratorType>::value_type DataType;
9452
size_type length = endIterator-startIterator;
9453
size_type byteLength = length*sizeof(DataType);
9456
static_cast<DataType*>(queue.enqueueMapBuffer(buffer, CL_TRUE, CL_MAP_READ, 0, byteLength, 0, 0, &error));
9457
// if exceptions enabled, enqueueMapBuffer will throw
9458
if( error != CL_SUCCESS ) {
9461
std::copy(pointer, pointer + length, startIterator);
9463
error = queue.enqueueUnmapMemObject(buffer, pointer, 0, &endEvent);
9464
// if exceptions enabled, enqueueUnmapMemObject will throw
9465
if( error != CL_SUCCESS ) {
9473
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
9475
* Blocking SVM map operation - performs a blocking map underneath.
9477
template<typename T, class Alloc>
9478
inline cl_int mapSVM(cl::vector<T, Alloc> &container)
9480
return enqueueMapSVM(container, CL_TRUE, CL_MAP_READ | CL_MAP_WRITE);
9484
* Blocking SVM map operation - performs a blocking map underneath.
9486
template<typename T, class Alloc>
9487
inline cl_int unmapSVM(cl::vector<T, Alloc> &container)
9489
return enqueueUnmapSVM(container);
9492
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
9494
#if CL_HPP_TARGET_OPENCL_VERSION >= 110
9495
inline cl_int enqueueReadBufferRect(
9496
const Buffer& buffer,
9498
const array<size_type, 3>& buffer_offset,
9499
const array<size_type, 3>& host_offset,
9500
const array<size_type, 3>& region,
9501
size_type buffer_row_pitch,
9502
size_type buffer_slice_pitch,
9503
size_type host_row_pitch,
9504
size_type host_slice_pitch,
9506
const vector<Event>* events = NULL,
9507
Event* event = NULL)
9510
CommandQueue queue = CommandQueue::getDefault(&error);
9512
if (error != CL_SUCCESS) {
9516
return queue.enqueueReadBufferRect(
9531
inline cl_int enqueueWriteBufferRect(
9532
const Buffer& buffer,
9534
const array<size_type, 3>& buffer_offset,
9535
const array<size_type, 3>& host_offset,
9536
const array<size_type, 3>& region,
9537
size_type buffer_row_pitch,
9538
size_type buffer_slice_pitch,
9539
size_type host_row_pitch,
9540
size_type host_slice_pitch,
9542
const vector<Event>* events = NULL,
9543
Event* event = NULL)
9546
CommandQueue queue = CommandQueue::getDefault(&error);
9548
if (error != CL_SUCCESS) {
9552
return queue.enqueueWriteBufferRect(
9567
inline cl_int enqueueCopyBufferRect(
9570
const array<size_type, 3>& src_origin,
9571
const array<size_type, 3>& dst_origin,
9572
const array<size_type, 3>& region,
9573
size_type src_row_pitch,
9574
size_type src_slice_pitch,
9575
size_type dst_row_pitch,
9576
size_type dst_slice_pitch,
9577
const vector<Event>* events = NULL,
9578
Event* event = NULL)
9581
CommandQueue queue = CommandQueue::getDefault(&error);
9583
if (error != CL_SUCCESS) {
9587
return queue.enqueueCopyBufferRect(
9600
#endif // CL_HPP_TARGET_OPENCL_VERSION >= 110
9602
inline cl_int enqueueReadImage(
9605
const array<size_type, 3>& origin,
9606
const array<size_type, 3>& region,
9607
size_type row_pitch,
9608
size_type slice_pitch,
9610
const vector<Event>* events = NULL,
9611
Event* event = NULL)
9614
CommandQueue queue = CommandQueue::getDefault(&error);
9616
if (error != CL_SUCCESS) {
9620
return queue.enqueueReadImage(
9632
inline cl_int enqueueWriteImage(
9635
const array<size_type, 3>& origin,
9636
const array<size_type, 3>& region,
9637
size_type row_pitch,
9638
size_type slice_pitch,
9640
const vector<Event>* events = NULL,
9641
Event* event = NULL)
9644
CommandQueue queue = CommandQueue::getDefault(&error);
9646
if (error != CL_SUCCESS) {
9650
return queue.enqueueWriteImage(
9662
inline cl_int enqueueCopyImage(
9665
const array<size_type, 3>& src_origin,
9666
const array<size_type, 3>& dst_origin,
9667
const array<size_type, 3>& region,
9668
const vector<Event>* events = NULL,
9669
Event* event = NULL)
9672
CommandQueue queue = CommandQueue::getDefault(&error);
9674
if (error != CL_SUCCESS) {
9678
return queue.enqueueCopyImage(
9688
inline cl_int enqueueCopyImageToBuffer(
9691
const array<size_type, 3>& src_origin,
9692
const array<size_type, 3>& region,
9693
size_type dst_offset,
9694
const vector<Event>* events = NULL,
9695
Event* event = NULL)
9698
CommandQueue queue = CommandQueue::getDefault(&error);
9700
if (error != CL_SUCCESS) {
9704
return queue.enqueueCopyImageToBuffer(
9714
inline cl_int enqueueCopyBufferToImage(
9717
size_type src_offset,
9718
const array<size_type, 3>& dst_origin,
9719
const array<size_type, 3>& region,
9720
const vector<Event>* events = NULL,
9721
Event* event = NULL)
9724
CommandQueue queue = CommandQueue::getDefault(&error);
9726
if (error != CL_SUCCESS) {
9730
return queue.enqueueCopyBufferToImage(
9741
inline cl_int flush(void)
9744
CommandQueue queue = CommandQueue::getDefault(&error);
9746
if (error != CL_SUCCESS) {
9750
return queue.flush();
9753
inline cl_int finish(void)
9756
CommandQueue queue = CommandQueue::getDefault(&error);
9758
if (error != CL_SUCCESS) {
9763
return queue.finish();
9769
CommandQueue queue_;
9770
const NDRange offset_;
9771
const NDRange global_;
9772
const NDRange local_;
9773
vector<Event> events_;
9775
template<typename... Ts>
9776
friend class KernelFunctor;
9779
EnqueueArgs(NDRange global) :
9780
queue_(CommandQueue::getDefault()),
9788
EnqueueArgs(NDRange global, NDRange local) :
9789
queue_(CommandQueue::getDefault()),
9797
EnqueueArgs(NDRange offset, NDRange global, NDRange local) :
9798
queue_(CommandQueue::getDefault()),
9806
EnqueueArgs(Event e, NDRange global) :
9807
queue_(CommandQueue::getDefault()),
9812
events_.push_back(e);
9815
EnqueueArgs(Event e, NDRange global, NDRange local) :
9816
queue_(CommandQueue::getDefault()),
9821
events_.push_back(e);
9824
EnqueueArgs(Event e, NDRange offset, NDRange global, NDRange local) :
9825
queue_(CommandQueue::getDefault()),
9830
events_.push_back(e);
9833
EnqueueArgs(const vector<Event> &events, NDRange global) :
9834
queue_(CommandQueue::getDefault()),
9843
EnqueueArgs(const vector<Event> &events, NDRange global, NDRange local) :
9844
queue_(CommandQueue::getDefault()),
9853
EnqueueArgs(const vector<Event> &events, NDRange offset, NDRange global, NDRange local) :
9854
queue_(CommandQueue::getDefault()),
9863
EnqueueArgs(CommandQueue &queue, NDRange global) :
9872
EnqueueArgs(CommandQueue &queue, NDRange global, NDRange local) :
9881
EnqueueArgs(CommandQueue &queue, NDRange offset, NDRange global, NDRange local) :
9890
EnqueueArgs(CommandQueue &queue, Event e, NDRange global) :
9896
events_.push_back(e);
9899
EnqueueArgs(CommandQueue &queue, Event e, NDRange global, NDRange local) :
9905
events_.push_back(e);
9908
EnqueueArgs(CommandQueue &queue, Event e, NDRange offset, NDRange global, NDRange local) :
9914
events_.push_back(e);
9917
EnqueueArgs(CommandQueue &queue, const vector<Event> &events, NDRange global) :
9927
EnqueueArgs(CommandQueue &queue, const vector<Event> &events, NDRange global, NDRange local) :
9937
EnqueueArgs(CommandQueue &queue, const vector<Event> &events, NDRange offset, NDRange global, NDRange local) :
9949
//----------------------------------------------------------------------------------------------
9953
* Type safe kernel functor.
9956
template<typename... Ts>
9962
template<int index, typename T0, typename... T1s>
9963
void setArgs(T0&& t0, T1s&&... t1s)
9965
kernel_.setArg(index, t0);
9966
setArgs<index + 1, T1s...>(std::forward<T1s>(t1s)...);
9969
template<int index, typename T0>
9970
void setArgs(T0&& t0)
9972
kernel_.setArg(index, t0);
9982
KernelFunctor(Kernel kernel) : kernel_(kernel)
9986
const Program& program,
9988
cl_int * err = NULL) :
9989
kernel_(program, name.c_str(), err)
9992
//! \brief Return type of the functor
9993
typedef Event result_type;
9997
* @param args Launch parameters of the kernel.
9998
* @param t0... List of kernel arguments based on the template type of the functor.
10001
const EnqueueArgs& args,
10005
setArgs<0>(std::forward<Ts>(ts)...);
10007
args.queue_.enqueueNDRangeKernel(
10019
* Enqueue kernel with support for error code.
10020
* @param args Launch parameters of the kernel.
10021
* @param t0... List of kernel arguments based on the template type of the functor.
10022
* @param error Out parameter returning the error code from the execution.
10025
const EnqueueArgs& args,
10030
setArgs<0>(std::forward<Ts>(ts)...);
10032
error = args.queue_.enqueueNDRangeKernel(
10043
#if CL_HPP_TARGET_OPENCL_VERSION >= 200
10044
cl_int setSVMPointers(const vector<void*> &pointerList)
10046
return kernel_.setSVMPointers(pointerList);
10049
template<typename T0, typename... T1s>
10050
cl_int setSVMPointers(const T0 &t0, T1s &... ts)
10052
return kernel_.setSVMPointers(t0, ts...);
10054
#endif // #if CL_HPP_TARGET_OPENCL_VERSION >= 200
10062
namespace compatibility {
10064
* Backward compatibility class to ensure that cl.hpp code works with cl2.hpp.
10065
* Please use KernelFunctor directly.
10067
template<typename... Ts>
10070
typedef KernelFunctor<Ts...> FunctorType;
10072
FunctorType functor_;
10075
const Program& program,
10077
cl_int * err = NULL) :
10078
functor_(FunctorType(program, name, err))
10082
const Kernel kernel) :
10083
functor_(FunctorType(kernel))
10086
//! \brief Return type of the functor
10087
typedef Event result_type;
10089
//! \brief Function signature of kernel functor with no event dependency.
10090
typedef Event type_(
10091
const EnqueueArgs&,
10095
const EnqueueArgs& enqueueArgs,
10099
enqueueArgs, args...);
10102
} // namespace compatibility
10105
//----------------------------------------------------------------------------------------------------------------------
10107
#undef CL_HPP_ERR_STR_
10108
#if !defined(CL_HPP_USER_OVERRIDE_ERROR_STRINGS)
10109
#undef __GET_DEVICE_INFO_ERR
10110
#undef __GET_PLATFORM_INFO_ERR
10111
#undef __GET_DEVICE_IDS_ERR
10112
#undef __GET_PLATFORM_IDS_ERR
10113
#undef __GET_CONTEXT_INFO_ERR
10114
#undef __GET_EVENT_INFO_ERR
10115
#undef __GET_EVENT_PROFILE_INFO_ERR
10116
#undef __GET_MEM_OBJECT_INFO_ERR
10117
#undef __GET_IMAGE_INFO_ERR
10118
#undef __GET_SAMPLER_INFO_ERR
10119
#undef __GET_KERNEL_INFO_ERR
10120
#undef __GET_KERNEL_ARG_INFO_ERR
10121
#undef __GET_KERNEL_SUB_GROUP_INFO_ERR
10122
#undef __GET_KERNEL_WORK_GROUP_INFO_ERR
10123
#undef __GET_PROGRAM_INFO_ERR
10124
#undef __GET_PROGRAM_BUILD_INFO_ERR
10125
#undef __GET_COMMAND_QUEUE_INFO_ERR
10126
#undef __CREATE_CONTEXT_ERR
10127
#undef __CREATE_CONTEXT_FROM_TYPE_ERR
10128
#undef __GET_SUPPORTED_IMAGE_FORMATS_ERR
10129
#undef __CREATE_BUFFER_ERR
10131
#undef __CREATE_SUBBUFFER_ERR
10132
#undef __CREATE_GL_BUFFER_ERR
10133
#undef __CREATE_GL_RENDER_BUFFER_ERR
10134
#undef __GET_GL_OBJECT_INFO_ERR
10135
#undef __CREATE_IMAGE_ERR
10136
#undef __CREATE_GL_TEXTURE_ERR
10137
#undef __IMAGE_DIMENSION_ERR
10138
#undef __SET_MEM_OBJECT_DESTRUCTOR_CALLBACK_ERR
10139
#undef __CREATE_USER_EVENT_ERR
10140
#undef __SET_USER_EVENT_STATUS_ERR
10141
#undef __SET_EVENT_CALLBACK_ERR
10142
#undef __WAIT_FOR_EVENTS_ERR
10143
#undef __CREATE_KERNEL_ERR
10144
#undef __SET_KERNEL_ARGS_ERR
10145
#undef __CREATE_PROGRAM_WITH_SOURCE_ERR
10146
#undef __CREATE_PROGRAM_WITH_IL_ERR
10147
#undef __CREATE_PROGRAM_WITH_BINARY_ERR
10148
#undef __CREATE_PROGRAM_WITH_IL_ERR
10149
#undef __CREATE_PROGRAM_WITH_BUILT_IN_KERNELS_ERR
10150
#undef __BUILD_PROGRAM_ERR
10151
#undef __COMPILE_PROGRAM_ERR
10152
#undef __LINK_PROGRAM_ERR
10153
#undef __CREATE_KERNELS_IN_PROGRAM_ERR
10154
#undef __CREATE_COMMAND_QUEUE_WITH_PROPERTIES_ERR
10155
#undef __CREATE_SAMPLER_WITH_PROPERTIES_ERR
10156
#undef __SET_COMMAND_QUEUE_PROPERTY_ERR
10157
#undef __ENQUEUE_READ_BUFFER_ERR
10158
#undef __ENQUEUE_READ_BUFFER_RECT_ERR
10159
#undef __ENQUEUE_WRITE_BUFFER_ERR
10160
#undef __ENQUEUE_WRITE_BUFFER_RECT_ERR
10161
#undef __ENQEUE_COPY_BUFFER_ERR
10162
#undef __ENQEUE_COPY_BUFFER_RECT_ERR
10163
#undef __ENQUEUE_FILL_BUFFER_ERR
10164
#undef __ENQUEUE_READ_IMAGE_ERR
10165
#undef __ENQUEUE_WRITE_IMAGE_ERR
10166
#undef __ENQUEUE_COPY_IMAGE_ERR
10167
#undef __ENQUEUE_FILL_IMAGE_ERR
10168
#undef __ENQUEUE_COPY_IMAGE_TO_BUFFER_ERR
10169
#undef __ENQUEUE_COPY_BUFFER_TO_IMAGE_ERR
10170
#undef __ENQUEUE_MAP_BUFFER_ERR
10171
#undef __ENQUEUE_MAP_IMAGE_ERR
10172
#undef __ENQUEUE_UNMAP_MEM_OBJECT_ERR
10173
#undef __ENQUEUE_NDRANGE_KERNEL_ERR
10174
#undef __ENQUEUE_NATIVE_KERNEL
10175
#undef __ENQUEUE_MIGRATE_MEM_OBJECTS_ERR
10176
#undef __ENQUEUE_MIGRATE_SVM_ERR
10177
#undef __ENQUEUE_ACQUIRE_GL_ERR
10178
#undef __ENQUEUE_RELEASE_GL_ERR
10179
#undef __CREATE_PIPE_ERR
10180
#undef __GET_PIPE_INFO_ERR
10181
#undef __RETAIN_ERR
10182
#undef __RELEASE_ERR
10184
#undef __FINISH_ERR
10185
#undef __VECTOR_CAPACITY_ERR
10186
#undef __CREATE_SUB_DEVICES_ERR
10187
#undef __CREATE_SUB_DEVICES_ERR
10188
#undef __ENQUEUE_MARKER_ERR
10189
#undef __ENQUEUE_WAIT_FOR_EVENTS_ERR
10190
#undef __ENQUEUE_BARRIER_ERR
10191
#undef __UNLOAD_COMPILER_ERR
10192
#undef __CREATE_GL_TEXTURE_2D_ERR
10193
#undef __CREATE_GL_TEXTURE_3D_ERR
10194
#undef __CREATE_IMAGE2D_ERR
10195
#undef __CREATE_IMAGE3D_ERR
10196
#undef __CREATE_COMMAND_QUEUE_ERR
10197
#undef __ENQUEUE_TASK_ERR
10198
#undef __CREATE_SAMPLER_ERR
10199
#undef __ENQUEUE_MARKER_WAIT_LIST_ERR
10200
#undef __ENQUEUE_BARRIER_WAIT_LIST_ERR
10201
#undef __CLONE_KERNEL_ERR
10202
#undef __GET_HOST_TIMER_ERR
10203
#undef __GET_DEVICE_AND_HOST_TIMER_ERR
10205
#endif //CL_HPP_USER_OVERRIDE_ERROR_STRINGS
10208
#undef CL_HPP_INIT_CL_EXT_FCN_PTR_
10209
#undef CL_HPP_INIT_CL_EXT_FCN_PTR_PLATFORM_
10211
#if defined(CL_HPP_USE_CL_DEVICE_FISSION)
10212
#undef CL_HPP_PARAM_NAME_DEVICE_FISSION_
10213
#endif // CL_HPP_USE_CL_DEVICE_FISSION
10215
#undef CL_HPP_NOEXCEPT_
10216
#undef CL_HPP_DEFINE_STATIC_MEMBER_