Opencl cheet sheet

OpenCL API 1.2 Reference Card - Page 1
OpenCL (Open Computing Language) is a multi-vendor open standard The OpenCL Runtime cl_int clReleaseCommandQueue (
cl_command_queue command_queue)
for general-purpose parallel programming of heterogeneous systems
that include CPUs, GPUs, and other processors. OpenCL provides Command Queues [5.1]
a uniform programming environment for software developers to cl_command_queue clCreateCommandQueue ( cl_int clGetCommandQueueInfo (
cl_context context, cl_device_id device, cl_command_queue command_queue,
write efficient, portable code for high-performance compute servers, cl_command_queue_info param_name,
cl_command_queue_properties properties,
desktop computer systems, and handheld devices. cl_int *errcode_ret) size_t param_value_size,
void *param_value,
[n.n.n] refers to the section in the OpenCL Specification. properties: CL_QUEUE_PROFILING_ENABLE, size_t *param_value_size_ret)
CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ ENABLE
[n.n.n] refers to the section in the OpenCL Extension Specification param_name: CL_QUEUE_CONTEXT,
Text shown in purple is as per the OpenCL Extension Specification. CL_QUEUE_DEVICE,
cl_int clRetainCommandQueue ( CL_QUEUE_REFERENCE_COUNT,
cl_command_queue command_queue) CL_QUEUE_PROPERTIES
Specifications are available at www.khronos.org/opencl.

The OpenCL Platform Layer Buffer Objects cl_int clEnqueueWriteBuffer (
cl_command_queue command_queue, cl_mem buffer,
The OpenCL platform layer implements platform-specific Elements of a buffer object are stored sequentially and accessed cl_bool blocking_write, size_t offset, size_t size,
features that allow applications to query OpenCL devices, using a pointer by a kernel executing on a device. Data is stored const void *ptr, cl_uint num_events_in_wait_list,
device configuration information, and to create OpenCL in the same format as it is accessed by the kernel. const cl_event *event_wait_list, cl_event *event)
contexts using one or more devices. cl_int clEnqueueWriteBufferRect (
Querying Platform Info and Devices [4.1, 4.2] Create Buffer Objects [5.2.1] cl_command_queue command_queue,
cl_mem clCreateBuffer (cl_context context, cl_mem buffer, cl_bool blocking_write,
cl_int clGetPlatformIDs (cl_uint num_entries, const size_t *buffer_origin, const size_t *host_origin,
cl_platform_id *platforms, cl_uint *num_platforms) cl_mem_flags flags, size_t size, void *host_ptr,
cl_int *errcode_ret) const size_t *region, size_t buffer_row_pitch,
cl_int clGetPlatformInfo (cl_platform_id platform, size_t buffer_slice_pitch, size_t host_row_pitch,
flags: CL_MEM_READ_WRITE, size_t host_slice_pitch, const void *ptr,
cl_platform_info param_name, CL_MEM_{WRITE, READ}_ONLY, cl_uint num_events_in_wait_list,
size_t param_value_size, void *param_value, CL_MEM_HOST_NO_ACCESS, const cl_event *event_wait_list, cl_event *event)
size_t *param_value_size_ret) CL_MEM_HOST_{READ, WRITE}_ONLY,
param_name: CL_PLATFORM_{PROFILE, VERSION}, CL_MEM_{USE, ALLOC, COPY}_HOST_PTR cl_int clEnqueueFillBuffer (
CL_PLATFORM_{NAME, VENDOR, EXTENSIONS} cl_command_queue command_queue,
cl_mem clCreateSubBuffer (cl_mem buffer, cl_mem buffer, const void *pattern,
cl_int clGetDeviceIDs (cl_platform_id platform, cl_mem_flags flags, size_t pattern_size, size_t offset, size_t size,
cl_device_type device_type, cl_uint num_entries, cl_buffer_create_type buffer_create_type, cl_uint num_events_in_wait_list,
cl_device_id *devices, cl_uint *num_devices) const void *buffer_create_info, cl_int *errcode_ret) const cl_event *event_wait_list, cl_event *event)
device_type: CL_DEVICE_TYPE_{ACCELERATOR, ALL, CPU}, cl_int clEnqueueCopyBuffer (
CL_DEVICE_TYPE_{CUSTOM, DEFAULT, GPU} flags: same as for clCreateBuffer
buffer_create_type: CL_BUFFER_CREATE_TYPE_REGION cl_command_queue command_queue,
cl_int clGetDeviceInfo (cl_device_id device, cl_mem src_buffer, cl_mem dst_buffer,
cl_device_info param_name, size_t param_value_size, size_t src_offset, size_t dst_offset, size_t size,
void *param_value, size_t *param_value_size_ret) Read, Write, Copy Buffer Objects [5.2.2] cl_uint num_events_in_wait_list,
cl_int clEnqueueReadBuffer ( const cl_event *event_wait_list, cl_event *event)
param_name: cl_command_queue command_queue, cl_mem buffer,
CL_DEVICE_{NAME, VENDOR, PROFILE, TYPE}, cl_int clEnqueueCopyBufferRect (
CL_DEVICE_NATIVE_VECTOR_WIDTH_{CHAR, INT}, cl_bool blocking_read, size_t offset, size_t size, cl_command_queue command_queue,
CL_DEVICE_NATIVE_VECTOR_WIDTH_{LONG, SHORT}, void *ptr, cl_uint num_events_in_wait_list, cl_mem src_buffer, cl_mem dst_buffer,
CL_DEVICE_NATIVE_VECTOR_WIDTH_{DOUBLE, HALF}, const cl_event *event_wait_list, cl_event *event) const size_t *src_origin, const size_t *dst_origin,
CL_DEVICE_NATIVE_VECTOR_WIDTH_FLOAT, const size_t *region, size_t src_row_pitch,
CL_DEVICE_PREFERRED_VECTOR_WIDTH_{CHAR, INT},
cl_int clEnqueueReadBufferRect ( size_t src_slice_pitch, size_t dst_row_pitch,
CL_DEVICE_PREFERRED_VECTOR_WIDTH_{LONG, SHORT},
cl_command_queue command_queue, cl_mem buffer, size_t dst_slice_pitch, cl_uint num_events_in_wait_list,
CL_DEVICE_PREFERRED_VECTOR_WIDTH_{DOUBLE, HALF},
cl_bool blocking_read, const size_t *buffer_origin, const cl_event *event_wait_list, cl_event *event)
CL_DEVICE_PREFERRED_VECTOR_WIDTH_FLOAT,
const size_t *host_origin, const size_t *region,
CL_DEVICE_PREFERRED_INTEROP_USER_SYNC,
size_t buffer_row_pitch, size_t buffer_slice_pitch, Map Buffer Objects [5.2.3]
CL_DEVICE_ADDRESS_BITS, CL_DEVICE_AVAILABLE, size_t host_row_pitch, size_t host_slice_pitch,
void *ptr, cl_uint num_events_in_wait_list, void * clEnqueueMapBuffer (
CL_DEVICE_BUILT_IN_KERNELS, cl_command_queue command_queue, cl_mem buffer,
CL_DEVICE_COMPILER_AVAILABLE, const cl_event *event_wait_list, cl_event *event)
cl_bool blocking_map, cl_map_flags map_flags,
CL_DEVICE_{DOUBLE, HALF, SINGLE}_FP_CONFIG, size_t offset, size_t size, cl_uint num_events_in_wait_list,
CL_DEVICE_ENDIAN_LITTLE, CL_DEVICE_EXTENSIONS, cl_int clReleaseDevice (cl_device_id device) const cl_event *event_wait_list, cl_event *event,
CL_DEVICE_ERROR_CORRECTION_SUPPORT, cl_int *errcode_ret)
CL_DEVICE_EXECUTION_CAPABILITIES,
CL_DEVICE_GLOBAL_MEM_CACHE_{SIZE, TYPE}, Contexts [4.4] map_flags: CL_MAP_{READ, WRITE},
CL_DEVICE_GLOBAL_MEM_{CACHELINE_SIZE, SIZE}, cl_context clCreateContext ( CL_MAP_WRITE_INVALIDATE_REGION
CL_DEVICE_HOST_UNIFIED_MEMORY, const cl_context_properties *properties,
CL_DEVICE_IMAGE_MAX_{ARRAY, BUFFER}_SIZE, cl_uint num_devices, const cl_device_id *devices, Memory Objects [5.4.1, 5.4.2]
CL_DEVICE_IMAGE_SUPPORT, void (CL_CALLBACK*pfn_notify) cl_int clRetainMemObject (cl_mem memobj)
CL_DEVICE_IMAGE2D_MAX_{WIDTH, HEIGHT}, (const char *errinfo, const void *private_info, cl_int clReleaseMemObject (cl_mem memobj)
CL_DEVICE_IMAGE3D_MAX_{WIDTH, HEIGHT, DEPTH}, size_t cb, void *user_data),
CL_DEVICE_LOCAL_MEM_{TYPE, SIZE}, void *user_data, cl_int *errcode_ret) cl_int clSetMemObjectDestructorCallback (
CL_DEVICE_MAX_{READ, WRITE}_IMAGE_ARGS, cl_mem memobj, void (CL_CALLBACK *pfn_notify)
CL_DEVICE_MAX_CLOCK_FREQUENCY, properties: NULL or CL_CONTEXT_PLATFORM,
CL_CONTEXT_INTEROP_USER_SYNC, (cl_mem memobj, void *user_data),
CL_DEVICE_MAX_COMPUTE_UNITS, void *user_data)
CL_DEVICE_MAX_CONSTANT_{ARGS,BUFFER_SIZE}, CL_CONTEXT_{D3D10, D3D11}_DEVICE_KHR,
CL_DEVICE_MAX_{MEM_ALLOC, PARAMETER}_SIZE, CL_CONTEXT_ADAPTER_{D3D9, D3D9EX, DXVA}_KHR, cl_int clEnqueueUnmapMemObject (
CL_DEVICE_MAX_SAMPLERS, CL_GL_CONTEXT_KHR, CL_CGL_SHAREGROUP_KHR, cl_command_queue command_queue,
CL_DEVICE_MAX_WORK_GROUP_SIZE, CL_{EGL, GLX}_DISPLAY_KHR, CL_WGL_HDC_KHR cl_mem memobj, void *mapped_ptr,
CL_DEVICE_MAX_WORK_ITEM_{DIMENSIONS, SIZES}, cl_uint num_events_in_wait_list,
CL_DEVICE_MEM_BASE_ADDR_ALIGN, cl_context clCreateContextFromType ( const cl_event *event_wait_list, cl_event *event)
CL_DEVICE_OPENCL_C_VERSION, CL_DEVICE_PARENT_DEVICE, const cl_context_properties *properties,
CL_DEVICE_PARTITION_AFFINITY_DOMAIN, cl_device_type device_type, Migrate Memory Objects [5.4.4]
CL_DEVICE_PARTITION_MAX_SUB_DEVICES, void (CL_CALLBACK *pfn_notify) cl_int clEnqueueMigrateMemObjects (
CL_DEVICE_PARTITION_{PROPERTIES, TYPE}, (const char *errinfo, const void *private_info, cl_command_queue command_queue,
CL_DEVICE_PLATFORM, CL_DEVICE_PRINTF_BUFFER_SIZE, size_t cb, void *user_data), cl_uint num_mem_objects,
CL_DEVICE_PROFILING_TIMER_RESOLUTION, void *user_data, cl_int *errcode_ret) const cl_mem *mem_objects,
CL_DEVICE_QUEUE_PROPERTIES, properties: See clCreateContext cl_mem_migration_flags flags,
CL_DEVICE_REFERENCE_COUNT, cl_uint num_events_in_wait_list,
CL_DEVICE_VENDOR_ID, CL_{DEVICE, DRIVER}_VERSION cl_int clRetainContext (cl_context context)
const cl_event *event_wait_list, cl_event *event)
cl_int clReleaseContext (cl_context context)
Partitioning a Device [4.3] flags: CL_MIGRATE_MEM_OBJECT_HOST,
cl_int clGetContextInfo (cl_context context, CL_MIGRATE_MEM_OBJECT_CONTENT_UNDEFINED
cl_int clCreateSubDevices (cl_device_id in_device, cl_context_info param_name, size_t param_value_size,
const cl_device_partition_property *properties, void *param_value, size_t *param_value_size_ret) Query Memory Object [5.4.5]
cl_uint num_devices, cl_device_id *out_devices, param_name: CL_CONTEXT_REFERENCE_COUNT, cl_int clGetMemObjectInfo (cl_mem memobj,
cl_uint *num_devices_ret) CL_CONTEXT_{DEVICES, NUM_DEVICES, PROPERTIES}, cl_mem_info param_name, size_t param_value_size,
properties: CL_DEVICE_PARTITION_EQUALLY, CL_CONTEXT_D3D10_PREFER_SHARED_RESOURCES_KHR, void *param_value, size_t *param_value_size_ret)
CL_DEVICE_PARTITION_BY_{COUNTS, AFFINITY_DOMAIN} CL_CONTEXT_D3D11_PREFER_SHARED_RESOURCES_KHR param_name: CL_MEM_{TYPE, FLAGS, SIZE, HOST_PTR},
(Affinity domains may be: CL_MEM_{MAP, REFERENCE}_COUNT, CL_MEM_OFFSET,
CL_DEVICE_AFFINITY_DOMAIN_NUMA,
CL_DEVICE_AFFINITY_DOMAIN_{L4, L3, L2, L1}_CACHE, Get CL Extension Function Pointers [9.2] CL_MEM_CONTEXT, CL_MEM_ASSOCIATED_MEMOBJECT,
CL_MEM_{D3D10, D3D11}_RESOURCE_KHR,
CL_DEVICE_AFFINITY_DOMAIN_NEXT_PARTITIONABLE) void* clGetExtensionFunctionAddressForPlatform ( CL_MEM_DX9_MEDIA_ADAPTER_TYPE_KHR,
cl_int clRetainDevice (cl_device_id device) cl_platform_id platform, const char *funcname) CL_MEM_DX9_MEDIA_SURFACE_INFO_KHR

©2011 Khronos Group - Rev. 1111 www.khronos.org/opencl

OpenCL Class Diagram [2.1] OpenCL Device Architecture Diagram [3.3]
The figure below describes the OpenCL specification as a class diagram using the Unified Modeling The table below shows memory regions with allocation and memory access capabilities.
Language1 (UML) notation. The diagram shows both nodes and edges which are classes and their
relationships. As a simplification it shows only classes, and no attributes or operations. Global Constant Local Private
Host Dynamic allocation Dynamic allocation Dynamic allocation No allocation
Annotations Read/Write access Read/Write access No access No access
Relationships Kernel No allocation Static allocation Static allocation Static allocation
abstract classes {abstract} Read/Write access Read-only access Read/Write access Read/Write access
aggregations
This conceptual OpenCL device architecture diagram shows processing elements (PE),
inheritance compute units (CU), and devices. The host is not shown.
relationship
navigability

Cardinality
many *
one and only one 1
optionally one 0..1
one or more 1..*

1
Unified Modeling Language (http://www.uml.org/) is a trademark of Object Management Group (OMG).

Program Objects cl_program clLinkProgram (cl_context context,
cl_uint num_devices, const cl_device_id *device_list,
Optimization options:
-cl-opt-disable -cl-mad-enable
Create Program Objects [5.6.1] const char *options, cl_uint num_input_programs, -cl-no-signed-zeros -cl-finite-math-only
cl_program clCreateProgramWithSource ( const cl_program *input_programs, -cl-unsafe-math-optimizations -cl-fast-relaxed-math
cl_context context, cl_uint count, const char **strings, void (CL_CALLBACK*pfn_notify)
(cl_program program, void *user_data), Warning request/suppress:
const size_t *lengths, cl_int *errcode_ret) void *user_data, cl_int *errcode_ret) -w -Werror
cl_program clCreateProgramWithBinary ( Control OpenCL C language version:
cl_context context, cl_uint num_devices, Unload the OpenCL Compiler [5.6.6] -cl-std=CL1.1 // OpenCL 1.1 specification.
const cl_device_id *device_list, const size_t *lengths, cl_int clUnloadPlatformCompiler (
const unsigned char **binaries, cl_int *binary_status, -cl-std=CL1.2 // OpenCL 1.2 specification.
cl_platform_id platform)
cl_int *errcode_ret) Query kernel argument information:
cl_program clCreateProgramWithBuiltInKernels ( Query Program Objects [5.6.7] -cl-kernel-arg-info
cl_context context, cl_uint num_devices, cl_int clGetProgramInfo (cl_program program,
const cl_device_id *device_list, cl_program_info param_name, size_t param_value_size, Linker Options [5.6.5]
const char *kernel_names, cl_int *errcode_ret) void *param_value, size_t *param_value_size_ret)
Library linking options: Program linking options:
cl_int clRetainProgram (cl_program program) param_name: CL_PROGRAM_REFERENCE_COUNT, -create-library -cl-denorms-are-zero
CL_PROGRAM_{CONTEXT, NUM_DEVICES, DEVICES},
cl_int clReleaseProgram (cl_program program) CL_PROGRAM_{SOURCE, BINARY_SIZES, BINARIES}, -enable-link-options -cl-no-signed-zeroes
CL_PROGRAM_{NUM_KERNELS, KERNEL_NAMES} -cl-unsafe-math-optimizations
Building Program Executables [5.6.2] cl_int clGetProgramBuildInfo ( -cl-finite-math-only
cl_int clBuildProgram (cl_program program, cl_program program, cl_device_id device, -cl-fast-relaxed-math
cl_uint num_devices, const cl_device_id *device_list, cl_program_build_info param_name,
const char *options, void (CL_CALLBACK*pfn_notify) size_t param_value_size, void *param_value,
(cl_program program, void *user_data), size_t *param_value_size_ret)
void *user_data) cl_int clSetUserEventStatus (cl_event event,
param_name: CL_PROGRAM_BINARY_TYPE, cl_int execution_status)
Separate Compilation and Linking [5.6.3] CL_PROGRAM_BUILD_{STATUS, OPTIONS, LOG}
cl_int clWaitForEvents (cl_uint num_events,
cl_int clCompileProgram (cl_program program,
cl_uint num_devices, const cl_device_id *device_list, Compiler Options [5.6.4] const cl_event *event_list)
const char *options, cl_uint num_input_headers, Preprocessor: (-D processed in order listed in cl_int clGetEventInfo (cl_event event,
const cl_program *input_headers, clBuildProgram or clCompileProgram) cl_event_info param_name, size_t param_value_size,
const char **header_include_names, -D name -D name=definition -I dir void *param_value, size_t *param_value_size_ret)
void (CL_CALLBACK*pfn_notify) Math intrinsics: param_name: CL_EVENT_COMMAND_{QUEUE, TYPE},
(cl_program program, void *user_data), -cl-single-precision-constant -cl-denorms-are-zero CL_EVENT_{CONTEXT, REFERENCE_COUNT},
void *user_data) -cl-fp32-correctly-rounded-divide-sqrt CL_EVENT_COMMAND_EXECUTION_STATUS
cl_int clSetEventCallback (cl_event event,
cl_int command_exec_callback_type,
Kernel and Event Objects cl_int clGetKernelArgInfo (cl_kernel kernel,
cl_uint arg_indx, cl_kernel_arg_info param_name, void (CL_CALLBACK *pfn_event_notify)
(cl_event event, cl_int event_command_exec_status,
Create Kernel Objects [5.7.1] size_t param_value_size, void *param_value,
void *user_data),
cl_kernel clCreateKernel (cl_program program, size_t *param_value_size_ret)
void *user_data)
const char *kernel_name, cl_int *errcode_ret) param_name:
CL_KERNEL_ARG_{ACCESS, ADDRESS, TYPE}_QUALIFIER, cl_int clRetainEvent (cl_event event)
cl_int clCreateKernelsInProgram (cl_program program, CL_KERNEL_ARG_NAME, CL_KERNEL_ARG_TYPE_NAME cl_int clReleaseEvent (cl_event event)
cl_uint num_kernels, cl_kernel *kernels,
cl_uint *num_kernels_ret) Execute Kernels [5.8] Markers, Barriers, and Waiting for Events [5.10]
cl_int clRetainKernel (cl_kernel kernel) cl_int clEnqueueNDRangeKernel (
cl_command_queue command_queue, cl_int clEnqueueMarkerWithWaitList (
cl_int clReleaseKernel (cl_kernel kernel) cl_kernel kernel, cl_uint work_dim, cl_command_queue command_queue,
const size_t *global_work_offset, cl_uint num_events_in_wait_list,
Kernel Arguments and Queries [5.7.2, 5.7.3] const size_t *global_work_size, const cl_event *event_wait_list, cl_event *event)
cl_int clSetKernelArg (cl_kernel kernel, cl_uint arg_index, const size_t *local_work_size,
size_t arg_size, const void *arg_value) cl_uint num_events_in_wait_list, cl_int clEnqueueBarrierWithWaitList (
const cl_event *event_wait_list, cl_event *event) cl_command_queue command_queue,
cl_int clGetKernelInfo (cl_kernel kernel, cl_uint num_events_in_wait_list,
cl_kernel_info param_name, size_t param_value_size, cl_int clEnqueueTask ( const cl_event *event_wait_list, cl_event *event)
void *param_value, size_t *param_value_size_ret) cl_command_queue command_queue,
param_name: CL_KERNEL_FUNCTION_NAME, cl_kernel kernel, cl_uint num_events_in_wait_list, Profiling Operations [5.12]
CL_KERNEL_NUM_ARGS, CL_KERNEL_REFERENCE_COUNT, const cl_event *event_wait_list, cl_event *event)
CL_KERNEL_{ATTRIBUTES, CONTEXT, PROGRAM} cl_int clGetEventProfilingInfo (cl_event event,
cl_int clEnqueueNativeKernel (cl_command_queue cl_profiling_info param_name,
cl_int clGetKernelWorkGroupInfo ( command_queue, void (*user_func)(void *), size_t param_value_size, void *param_value,
cl_kernel kernel, cl_device_id device, void *args, size_t cb_args, cl_uint num_mem_objects, size_t *param_value_size_ret)
cl_kernel_work_group_info param_name, const cl_mem *mem_list, const void **args_mem_loc, param_name: CL_PROFILING_COMMAND_QUEUED,
size_t param_value_size, void *param_value, cl_uint num_events_in_wait_list, CL_PROFILING_COMMAND_{SUBMIT, START, END}
size_t *param_value_size_ret) const cl_event *event_wait_list, cl_event *event)
param_name: CL_KERNEL_GLOBAL_WORK_SIZE, Flush and Finish [5.13]
CL_KERNEL_[COMPILE_]WORK_GROUP_SIZE, Event Objects [5.9] cl_int clFlush (cl_command_queue command_queue)
CL_KERNEL_{LOCAL, PRIVATE}_MEM_SIZE, cl_event clCreateUserEvent (cl_context context,
CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE cl_int *errcode_ret) cl_int clFinish (cl_command_queue command_queue)


Supported Data Types Vector Component Addressing [6.1.7]
The optional double scalar and vector types are supported if Vector Components
CL_DEVICE_DOUBLE_FP_CONFIG is not zero.
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Built-in Scalar Data Types [6.1.1] float2 v; v.x, v.s0 v.y, v.s1
OpenCL Type API Type Description float3 v; v.x, v.s0 v.y, v.s1 v.z, v.s2
bool -- true (1) or false (0) float4 v; v.x, v.s0 v.y, v.s1 v.z, v.s2 v.w, v.s3
char cl_char 8-bit signed
float8 v; v.s0 v.s1 v.s2 v.s3 v.s4 v.s5 v.s6 v.s7
unsigned char, uchar cl_uchar 8-bit unsigned
float16 v; v.s0 v.s1 v.s2 v.s3 v.s4 v.s5 v.s6 v.s7 v.s8 v.s9 v.sa, v.sb, v.sc, v.sd, v.se, v.sf,
short cl_short 16-bit signed v.sA v.sB v.sC v.sD v.sE v.sF
unsigned short, ushort cl_ushort 16-bit unsigned
int cl_int 32-bit signed Vector Addressing Equivalences
unsigned int, uint cl_uint 32-bit unsigned Numeric indices are preceded by the letter s or S, e.g.: s1. Swizzling, duplication, and nesting are allowed, e.g.: v.yx, v.xx, v.lo.x
long cl_long 64-bit signed v.lo v.hi v.odd v.even v.lo v.hi v.odd v.even
unsigned long, ulong cl_ulong 64-bit unsigned
float2 v.x, v.s0 v.y, v.s1 v.y, v.s1 v.x, v.s0 float8 v.s0123 v.s4567 v.s1357 v.s0246
float cl_float 32-bit float float3 * v.s01, v.xy v.s23, v.zw v.s13, v.yw v.s02, v.xz float16 v.s01234567 v.s89abcdef v.s13579bdf v.s02468ace
double OPTIONAL cl_double 64-bit. IEEE 754
float4 v.s01, v.xy v.s23, v.zw v.s13, v.yw v.s02, v.xz *When using .lo or .hi with a 3-component vector, the .w component is undefined.
half cl_half 16-bit float (storage only)
size_t -- 32- or 64-bit unsigned integer
ptrdiff_t -- 32- or 64-bit signed integer Operators and Qualifiers Address Space Qualifiers [6.5]
Operators [6.3] __global, global __local, local
intptr_t -- 32- or 64-bit signed integer __constant, constant __private, private
uintptr_t -- 32- or 64-bit unsigned integer These operators behave similarly as in C99 except that
void void void
operands may include vector types when possible: Function Qualifiers [6.7]
+ - * % / -- ++ == != & __kernel, kernel
__attribute__((vec_type_hint(type))) //type defaults to int
Built-in Vector Data Types [6.1.2] ~ ^ > < >= <= | ! && || __attribute__((work_group_size_hint(X, Y, Z)))
OpenCL Type API Type Description ?: >> << = , op= sizeof __attribute__((reqd_work_group_size(X, Y, Z)))
charn cl_charn 8-bit signed
ucharn cl_ucharn 8-bit unsigned
Specify Type Attributes [6.11.1]
__attribute__((aligned(n))) __attribute__((endian(host)))
shortn cl_shortn 16-bit signed Use to specify special attributes of enum, __attribute__((aligned)) __attribute__((endian(device)))
ushortn cl_ushortn 16-bit unsigned struct and union types. __attribute__((packed)) __attribute__((endian))
intn cl_intn 32-bit signed
uintn cl_uintn 32-bit unsigned
Math Constants [6.12.2] [9.5.2] INFINITY Constant float expression,
positive or unsigned
M_LN2_F Value of loge2
longn cl_longn 64-bit signed
The values of the following symbolic infinity. M_LN10_F Value of loge10
ulongn cl_ulongn 64-bit unsigned constants are type float, accurate within
the precision of a single precision NAN Constant float expression, M_PI_F Value of π
floatn cl_floatn 32-bit float quiet NaN.
floating-point number. M_PI_2_F Value of π / 2
doublen OPTIONAL cl_doublen 64-bit float When double is supported, macros
MAXFLOAT Value of maximum non- ending in _F are available in type double M_PI_4_F Value of π / 4
infinite single-precision by removing _F from the macro name,
Other Built-in Data Types [6.1.3] floating-point number.
and in type half when the half extension M_1_PI_F Value of 1 / π
The optional types listed here other than event_t are only is enabled by replacing _F with _H. M_2_PI_F Value of 2 / π
defined if CL_DEVICE_IMAGE_SUPPORT is CL_TRUE. HUGE_VALF Positive float expression,
evaluates to +infinity. M_E_F Value of e M_2_SQRTPI_F Value of 2 / √π
OpenCL Type Description HUGE_VAL Positive double M_LOG2E_F Value of log2e M_SQRT2_F Value of √2
image2d_t OPTIONAL 2D image handle expression, evals. to
+infinity. OPTIONAL M_LOG10E_F Value of log10e M_SQRT1_2_F Value of 1 / √2
image3d_t OPTIONAL 3D image handle
image2d_array_t OPTIONAL 2D image array
image1d_t OPTIONAL 1D image handle Integer Built-in Functions [6.12.3] T sub_sat (T x, T y) x - y and saturates the result
image1d_buffer_t OPTIONAL 1D image buffer T is type char, charn, uchar, ucharn, short, shortn, ushort, T popcount (T x) Number of non-zero bits in x
image1d_array_t OPTIONAL 1D image array ushortn, int, intn, uint, uintn, long, longn, ulong, or ulongn,
sampler_t OPTIONAL sampler handle where n is 2, 3, 4, 8, or 16. Tu is the unsigned version of T.
Tsc is the scalar version of T. For upsample, return type is scalar when the parameters are scalar.
event_t event handle
Tu abs (T x) |x| short[n] upsample ( result[i]= ((short)hi[i]<< 8)|lo[i]
Reserved Data Types [6.1.4] char[n] hi, uchar[n] lo)
Tu abs_diff (T x, T y) | x – y | without modulo overflow
OpenCL Type Description ushort[n] upsample ( result[i]=((ushort)hi[i]<< 8)|lo[i]
T add_sat (T x, T y) x + y and saturates the result uchar[n] hi, uchar[n] lo)
booln boolean vector T hadd (T x, T y) (x + y) >> 1 without mod. overflow int[n] upsample ( result[i]=((int)hi[i]<< 16)|lo[i]
halfn 16-bit, vector T rhadd (T x, T y) (x + y + 1) >> 1 short[n] hi, ushort[n] lo)
quad, quadn 128-bit float, vector T clamp (T x, T min, T max) min(max(x, minval), maxval) uint[n] upsample (
T clamp (T x, Tsc min, Tsc max) ushort[n] hi, ushort[n] lo) result[i]=((uint)hi[i]<< 16)|lo[i]
complex half, complex halfn 16-bit complex, vector
imaginary half, imaginary halfn T clz (T x) number of leading 0-bits in x long[n] upsample (
int[n] hi, uint[n] lo) result[i]=((long)hi[i]<< 32)|lo[i]
complex float, complex floatn 32-bit complex, vector T mad_hi (T a, T b, T c) mul_hi(a, b) + c
imaginary float, imaginary floatn ulong[n] upsample (
T mad_sat (T a, T b, T c) a * b + c and saturates the result result[i]=((ulong)hi[i]<< 32)|lo[i]
complex double, complex doublen 64-bit complex, vector uint[n] hi, uint[n] lo)
imaginary double, imaginary doublen T max (T x, T y) y if x < y, otherwise it returns x
complex quad, complex quadn T max (T x, Tsc y) The following fast integer functions optimize the performance
imaginary quad, imaginary quadn 128-bit complex, vector of kernels. In these functions, T is type int, uint, intn or
T min (T x, T y) y if y < x, otherwise it returns x
intn,where n is 2, 3, 4, 8, or 16.
n*m matrix of 32-bit
floatnxm T min (T x, Tsc y) y if y < x, otherwise it returns x Multiply 24-bit integer values x, y,
floats T mad24 (T x, T y, T z)
n*m matrix of 64-bit T mul_hi (T x, T y) high half of the product of x and y add 32-bit int. result to 32-bit int. z
doublenxm T mul24 (T x, T y) Multiply 24-bit integer values x and y
floats T rotate (T v, T i) result[indx] = v[indx] << i[indx]

Preprocessor Directives & Macros [6.10] __LINE__ Integer line number __FAST_RELAXED_MATH__ 1 if –cl-fast-relaxed-math
__OPENCL_VERSION__ Integer version number, e.g: 120 optimization option is specified
#pragma OPENCL FP_CONTRACT on-off-switch
on-off-switch: ON, OFF, DEFAULT __CL_VERSION_1_0 Substitutes integer 100 for 1.0 FP_FAST_FMA Defined if double fma is fast
#pragma OPENCL EXTENSION extensionname : behavior __CL_VERSION_1_1 Substitutes integer 110 for 1.1 FP_FAST_FMAF Defined if float fma is fast
__CL_VERSION_1_2 Substitutes integer 120 for 1.2 FP_FAST_FMA_HALF Defined if half fma is fast
#pragma OPENCL EXTENSION all : behavior
__OPENCL_C_VERSION__ Sub. integer for OpenCL C version. __kernel_exec(X, typen) Same as:
__FILE__ Current source file __ENDIAN_LITTLE__ 1 if device is little endian __kernel __attribute__((work_group_size_hint(X, 1, 1)))
__func__ Current function name __IMAGE_SUPPORT__ 1 if images are supported __attribute__((vec_type_hint(typen)))


Math Built-in Functions [6.12.2] [9.5.2] T exp10 (T) HN Exponential base 10
T modf (T x, Q T *iptr) Decompose floating-point
Ts is type float, optionally double, or half if the half extension is number
T expm1 (T x) e x -1.0
enabled. Tn is the vector form of Ts, where n is 2, 3, 4, 8, or 16. float[n] nan (uint[n] nancode)
T is Ts and Tn. Q is qualifier __global, __local, or __private. T fabs (T) Absolute value Quiet NaN
HN indicates that half and native variants are available using only half[n] nan (ushort[n] nancode) (Return is scalar when nancode
the float or floatn types by prepending “half_” or “native_” to the Positive difference between x double[n] nan (ulong[n] nancode) is scalar)
T fdim (T x, T y) and y
function name. Prototypes shown in brown text are available in
half_ and native_ forms only using the float or floatn types. T nextafter (T x, T y) Next representable floating-point
T floor (T) Round to integer toward - infinity value after x in the direction of y
T acos (T) Arc cosine T fma (T a, T b, T c) Multiply and add, then round T pow (T x, T y) Compute x to the power of y
T acosh (T) Inverse hyperbolic cosine T fmax (T x, T y) Return y if x < y, Ts pown (T x, int y) Compute x y, where y is an
Tn fmax (Tn x, Ts y) otherwise it returns x Tn pown (T x, intn y) integer
T acospi (T x) acos (x) / π
T fmin (T x, T y) Return y if y < x, T powr (T x, T y) HN Compute x y, where x is >= 0
T asin (T) Arc sine
Tn fmin (Tn x, Ts y) otherwise it returns x
T asinh (T) Inverse hyperbolic sine T half_recip (T x) 1/x
Modulus. Returns x – y * trunc T native_recip (T x) (T may only be float or floatn)
T asinpi (T x) asin (x) / π T fmod (T x, T y) (x/y)
T remainder (T x, T y) Floating point remainder
T atan (T y_over_x) Arc tangent T fract (T x, Q T *iptr) Fractional value in x
Ts remquo (T x, T y, Q int *quo)
T atan2 (T y, T x) Arc tangent of y / x Ts frexp (T x, Q int *exp) Remainder and quotient
Extract mantissa and exponent Tn remquo (T x, T y, Q intn *quo)
T atanh (T) Hyperbolic arc tangent Tn frexp (T x, Q intn *exp)
T rint (T) Round to nearest even integer
T atanpi (T x) atan (x) / π T hypot (T x, T y) Square root of x2 + y2
Ts rootn (T x, int y) Compute x to the power of 1/y
T atan2pi (T x, T y) atan2 (y, x) / π Return exponent as an integer Tn rootn (T x, intn y)
int[n] ilogb (T x) value
T cbrt (T) Cube root T round (T x) Integral value nearest to x
Ts ldexp (T x, int n) rounding
x * 2n
Round to integer toward + Tn ldexp (T x, intn n)
T ceil (T) T rsqrt (T) HN Inverse square root
infinity
T lgamma (T x)
T copysign (T x, T y) x with sign changed to sign of y T sin (T) HN Sine
Ts lgamma_r (T x, Q int *signp) Log gamma function
T cos (T) HN Cosine Tn lgamma_r (T x, Q intn *signp) T sincos (T x, Q T *cosval) Sine and cosine of x

T cosh (T) Hyperbolic cosine T log (T) HN Natural logarithm T sinh (T) Hyperbolic sine

T cospi (T x) cos (π x) T log2 (T) HN Base 2 logarithm T sinpi (T x) sin (π x)

T half_divide (T x, T y) T log10 (T) HN Base 10 logarithm T sqrt (T) HN Square root
x/y
T native_divide (T x, T y) (T may only be float or floatn) T log1p (T x) ln (1.0 + x) T tan (T) HN Tangent
T erfc (T) Complementary error function T logb (T x) Exponent of x T tanh (T) Hyperbolic tangent
T erf (T) Calculates error function of T T mad (T a, T b, T c) Approximates a * b + c T tanpi (T x) tan (π x)
T exp (T x) HN Exponential base e T maxmag (T x, T y) Maximum magnitude of x and y T tgamma (T) Gamma function
T exp2 (T) HN Exponential base 2 T minmag (T x, T y) Minimum magnitude of x and y T trunc (T) Round to integer toward zero

Geometric Built-in Functions [6.12.5] [9.5.4] Vector Data Load/Store [6.12.7] [9.5.6] void vstore_half_R (double data, Write a half to address
Ts is scalar type float, optionally double, or half if the half T is type char, uchar, short, ushort, int, uint, long, ulong, or float, size_t offset, Q half *p) (p + offset)
extension is enabled. T is Ts and the 2-, 3-, or 4-component optionally double, or half if the half extension is enabled. Tn
vector forms of Ts. refers to the vector form of type T, where n is 2, 3, 4, 8, or 16. void vstore_halfn (floatn data,
Q is an Address Space Qualifier listed in 6.5 unless otherwise size_t offset, Q half *p)
float{3,4} cross (float{3,4} p0, float{3,4} p1) noted. When red, Q cannot be __constant. R defaults to the void vstore_halfn_R (floatn data,
double{3,4} cross (double{3,4} p0, double{3,4} p1) Cross product current rounding mode, or is one of the Rounding Modes listed size_t offset, Q half *p) Write a half vector to
in 6.2.3.2. void vstore_halfn (doublen data, address (p + (offset * n))
half{3,4} cross (half{3,4} p0, half{3,4} p1) size_t offset, Q half *p)
Tn vloadn (size_t offset, Read vector data from void vstore_halfn_R (doublen
Ts distance (T p0, T p1) Vector distance const Q T *p) address (p + (offset * n)) data, size_t offset, Q half *p)
Ts dot (T p0, T p1) Dot product
void vstoren (Tn data, Write vector data to
floatn vloada_halfn (size_t offset, Read half vector datahalf3,
from
Ts length (T p) Vector length size_t offset, Q T *p) address (p + (offset * n) (p + (offset * n)). For
const Q half *p)
float vload_half (size_t offset, Read a half from address read from (p + (offset * 4)).
T normalize (T p) Normal vector
length 1 const Q half *p) (p + offset)
void vstorea_halfn (floatn data,
float fast_distance (float p0, float p1) floatn vload_halfn (size_t offset, Read a halfn from address size_t offset, Q half *p)
Vector distance const Q half *p) (p + (offset * n)) void vstorea_halfn_R (floatn data,
float fast_distance (floatn p0, floatn p1)
size_t offset, Q half *p) Write half vector data to
float fast_length (float p) void vstore_half (float data, void vstorea_halfn (doublen data, (p + (offset * n)). For half3,
Vector length size_t offset, Q half *p) size_t offset, Q half *p) write to (p + (offset * 4)).
float fast_length (floatn p) void vstore_half_R (float data, Write a half to address void vstorea_halfn_R (
float fast_normalize (float p) size_t offset, Q half *p) (p + offset) doublen data, size_t offset,
Normal vector void vstore_half (double data,
floatn fast_normalize (floatn p) length 1 Q half *p)
size_t offset, Q half *p)

Async Copies and Prefetch Functions [6.12.10] [9.5.7] Work-Item Built-in Functions [6.12.1]
T is type char, charn, uchar, ucharn, short, shortn, ushort, ushortn, int, intn, uint, uintn, long, longn, ulong, ulongn, float, floatn, These functions query the number of dimensions, the global
optionally double or doublen, or half or halfn if the half extension is enabled. and local work size specified to clEnqueueNDRangeKernel,
event_t async_work_group_copy (__local T *dst, const __global T *src, and the global and local identifier of each work-item when
size_t num_gentypes, event_t event) this kernel is executed on a device. D is the dimension index.
Copies num_gentypes T elements from src to dst uint get_work_dim () Number of dimensions in use
event_t async_work_group_copy (__global T *dst, const __local T *src,
size_t num_gentypes, event_t event) size_t get_global_size (uint D) Number of global work-items
event_t async_work_group_strided_copy (__local T *dst, const __global T *src, size_t get_global_id (uint D) Global work-item ID value
size_t num_gentypes, size_t src_stride, event_t event)
Copies num_gentypes T elements from src to dst size_t get_local_size (uint D) Number of local work-items
event_t async_work_group_strided_copy (__global T *dst, const __local T *src,
size_t num_gentypes, size_t dst_stride, event_t event) size_t get_local_id (uint D) Local work-item ID
void wait_group_events (int num_events, event_t *event_list) Wait for events that identify the size_t get_num_groups (uint D) Number of work-groups
async_work_group_copy operations to complete
size_t get_group_id (uint D) Returns the work-group ID
void prefetch (const __global T *p, size_t num_gentypes) Prefetch num_gentypes * sizeof(T) bytes into the
global cache size_t get_global_offset (uint D) Returns global offset


Common Built-in Functions [6.12.4] [9.5.3] T min (T x, T y) Min of x and y Atomic Functions [6.12.11] [9.3]
These functions operate component-wise and use round Tn min (Tn x, Ts y)
These functions functions provide atomic operations on 32-bit signed
to nearest even rounding mode. Ts is type float, optionally T mix (T x, T y, T a) and unsigned integers and single precision floating-point to locations
Linear blend of x and y
double, or half if the half extension is enabled. Tn is the Tn mix (Tn x, Tn y, Ts a) in __global or __local memory. T is type int or unsigned int. T may also
vector form of Ts, where n is 2, 3, 4, 8, or 16. T is Ts and Tn. T radians (T degrees) degrees to radians be type float for atomic_xchg, and type long or ulong for extended 64-bit
T clamp (T x, T min, T max) Clamp x to range given by T step (T edge, T x) atomic functions. Q is volatile __global or volatile __local.
Tn clamp (Tn x, Ts min, Ts max) min, max 0.0 if x < edge, else 1.0
Tn step (Ts edge, Tn x)
T atomic_add (Q T *p, T val) Read, add, and store
T degrees (T radians) radians to degrees T smoothstep (T edge0, T edge1, T x)
Step and interpolate T atomic_sub (Q T *p, T val) Read, subtract, and store
T smoothstep (Ts edge0, Ts edge1, T x)
T max (T x, T y) Max of x and y T sign (T x) Sign of x T atomic_xchg (Q T *p, T val) Read, swap, and store
Tn max (Tn x, Ts y)
T atomic_inc (Q T *p) Read, increment, and store
T atomic_dec (Q T *p)
Relational Built-in Functions [6.12.6] int isinf (float)
Read, decrement, and store
These functions can be used with built-in scalar or vector intn isinf (floatn) T atomic_cmpxchg (Q T *p, Read, store
types as arguments and return a scalar or vector integer int isinf (double) Test for + or T cmp, T val) (*p ==cmp) ? val : *p
result. T is type float, floatn, char, charn, uchar, ucharn, longn isinf (doublen) – infinity
int isinf (half) T atomic_min (Q T *p, T val) Read, store min(*p, val)
short, shortn, ushort, ushortn, int, intn, uint, uintn, long,
longn, ulong, ulongn, or optionally double or doublen. Ti is shortn isinf (halfn) T atomic_max (Q T *p, T val) Read, store max(*p, val)
type char, charn, short, shortn, int, intn, long, or longn. Tu
is type uchar, ucharn, ushort, ushortn, uint, uintn, ulong, or int isnan (float) T atomic_and (Q T *p, T val) Read, store (*p & val)
ulongn. n is 2, 3, 4, 8, or 16. Optional extension enables half intn isnan (floatn)
and halfn types. int isnan (double) T atomic_or (Q T *p, T val) Read, store (*p | val)
Test for a NaN
longn isnan (doublen) T atomic_xor (Q T *p, T val) Read, store (*p ^ val)
int isequal (float x, float y) int isnan (half)
intn isequal (floatn x, floatn y) shortn isnan (halfn) Optional extensions enable forms of these functions using the
int isequal (double x, double y) Compare of atom_ prefix that implement atomic operations on 64-bit signed and
longn isequal (doublen x, doublen y) x == y int isnormal (float)
intn isnormal (floatn) Test for a unsigned integers. To use any of these forms, include the following in
int isequal (half x, half y) normal value the OpenCL program source:
shortn isequal (halfn x, halfn y) int isnormal (double)
#pragma OPENCL EXTENSION extension-name : enable
int isnotequal (float x, float y) longn isnormal (doublen) Test for a
intn isnotequal (floatn x, floatn y) int isnormal (half) normal Use cl_khr_int64_base_atomics for extension-name to enable 64-bit
shortn isnormal (halfn) value versions of the following functions:
int isnotequal (double x, double y) Compare of
longn isnotequal (doublen x, doublen y) x != y atom_add atom_sub atom_inc
int isnotequal (half x, half y) int isordered (float x, float y) atom_dec atom_xchg atom_cmpxchg
shortn isnotequal (halfn x, halfn y) intn isordered (floatn x, floatn y)
int isordered (double x, double y) Test if
int isgreater (float x, float y) arguments Use cl_khr_int64_extended_atomics for extension-name to enable 64-
longn isordered (doublen x, doublen y) are ordered
intn isgreater (floatn x, floatn y) int isordered (half x, half y) bit versions of the following functions:
int isgreater (double x, double y) Compare of shortn isordered (halfn x, halfn y) atom_min atom_max atom_and
longn isgreater (doublen x, doublen y) x>y
int isgreater (half x, half y) int isunordered (float x, float y) atom_or atom_xor
shortn isgreater (halfn x, halfn y) intn isunordered (floatn x, floatn y) Test if
int isgreaterequal (float x, float y) int isunordered (double x, double y) arguments
intn isgreaterequal (floatn x, floatn y) longn isunordered (doublen x, doublen y) are Conversions and Type Casting Examples [6.2]
int isgreaterequal (double x, double y) Compare of int isunordered (half x, half y) unordered
T a = (T)b; // Scalar to scalar, or scalar to vector
longn isgreaterequal (doublen x, doublen y) x >= y shortn isunordered (halfn x, halfn y)
T a = convert_T(b); T a = convert_T_R(b);
int isgreaterequal (half x, half y) T a = as_T(b); T a = convert_T_sat_R(b);
shortn isgreaterequal (halfn x, halfn y) int signbit (float)
intn signbit (floatn)
int isless (float x, float y) int signbit (double) R can be one of the following rounding modes:
intn isless (floatn x, floatn y) Test for sign bit
longn signbit (doublen) _rte to nearest even _rtp toward + infinity
int isless (double x, double y) Compare of int signbit (half)
longn isless (doublen x, doublen y) x<y shortn signbit (halfn) _rtz toward zero _rtn toward - infinity
int isless (half x, half y)
shortn isless (halfn x, halfn y) 1 if MSB in
int islessequal (float x, float y)
int any (Ti x) component of x
is set; else 0
Synchronization and Explicit Memory Fence
intn islessequal (floatn x, floatn y) Functions 6.12.8, 6.12.9]
int islessequal (double x, double y) Compare of 1 if MSB in all flags argument is the memory address space, set to a combination
longn islessequal (doublen x, doublen y) x <= y int all (Ti x) components of of CLK_LOCAL_MEM_FENCE and CLK_GLOBAL_MEM_FENCE. Explicit
int islessequal (half x, half y) x are set; else 0 memory fence functions provide ordering between memory operations
shortn islessequal (halfn x, halfn y) Each bit of result
of a work-item.
int islessgreater (float x, float y) T bitselect (T a, T b, T c) is corresponding void barrier ( Work-items in a work-group must
intn islessgreater (floatn x, floatn y) half bitselect (half a, half b, half c) bit of a if cl_mem_fence_flags flags) execute this before any can continue
int islessgreater (double x, double y) Compare of halfn bitselect (halfn a, halfn b, halfn c) corresponding
longn islessgreater (doublen x, doublen y) (x < y) || (x > y) bit of c is 0 void mem_fence ( Orders loads and stores of a work-
int islessgreater (half x, half y) For each cl_mem_fence_flags flags) item executing a kernel
shortn islessgreater (halfn x, halfn y) T select (T a, T b, Ti c)
component of a
T select (T a, T b, Tu c) vector type,
int isfinite (float) void read_mem_fence ( Orders memory loads
intn isfinite (floatn) halfn select (halfn a, halfn b, shortn c) result[i] = if cl_mem_fence_flags flags)
int isfinite (double) Test for finite half select (half a, half b, short c) MSB of c[i] is
longn isfinite (doublen) value halfn select (halfn a, halfn b, ushortn c) set ? b[i] : a[i]
For scalar type, void write_mem_fence ( Orders memory stores
int isfinite (half) cl_mem_fence_flags flags)
shortn isfinite (halfn) half select (half a, half b, ushort c) result = c ? b : a

Miscellaneous Vector Functions [6.12.12] printf Function [6.12.13] Examples:
Tm and Tn are type charn, ucharn, shortn, ushortn, intn, uintn, longn, Writes output to an implementation-defined stream. The following examples show the use of the vector specifier
ulongn, floatn, optionally doublen, or halfn if the half extension is in the printf format string.
int printf (constant char * restrict format, …)
enabled, where n is 2,4,8, or 16 except in vec_step it may also be 3. TUn float4 f = (float4)(1.0f, 2.0f, 3.0f, 4.0f);
is ucharn, ushortn, uintn, or ulongn. printf(“f4 = %2.2v4fn”, f);
printf output synchronization
Takes a built-in scalar or vector data type When the event associated with a particular kernel invocation Output: f4 = 1.00,2.00,3.00,4.00
int vec_step (Tn a) argument, returns an integer showing number completes, the output of applicable printf() calls is flushed to
of elements in the scalar or vector. Returns 1 for the implementation-defined output stream. uchar4 uc = (uchar4)(0xFA, 0xFB, 0xFC, 0xFD);
int vec_step (typename) scalar, 4 for 3-component vector, else number of printf(“uc = %#v4xn”, uc);
elements in the specified type.
printf format string Output: uc = 0xfa,0xfb,0xfc,0xfd
Tn shuffle (Tm x, Construct permutation of elements from one The format string follows C99 conventions and supports an
TUn mask) or two input vectors, return a vector with same optional vector specifier: uint2 ui = (uint2)(0x12345678, 0x87654321);
Tn shuffle2 (Tm x, element type as input and length that is the printf(“unsigned short value = (%#v2hx)n”, ui);
same as the shuffle mask. %[flags][width][.precision][vector][length]conversion
Tm y, TUn mask)
Output: unsigned short value = (0x5678,0x4321)

OpenCL Image Processing: Following is a subset of the OpenCL specification that pertains to image processing and graphics.

Image Objects cl_int clEnqueueFillImage (
cl_command_queue command_queue,
Query Image Objects [5.3.6]
Create Image Objects [5.3.1] cl_int clGetImageInfo (cl_mem image,
cl_mem image, const void *fill_color, cl_image_info param_name, size_t param_value_size,
cl_mem clCreateImage (cl_context context, const size_t *origin, const size_t *region, void *param_value, size_t *param_value_size_ret)
cl_mem_flags flags, cl_uint num_events_in_wait_list,
const cl_image_format *image_format, const cl_event *event_wait_list, param_name: CL_IMAGE_{ARRAY, ELEMENT}_SIZE,
const cl_image_desc *image_desc, cl_event *event) CL_IMAGE_{ROW, SLICE}_PITCH,
void *host_ptr, cl_int *errcode_ret) CL_IMAGE_{FORMAT, BUFFER, HEIGHT, WIDTH, DEPTH},
cl_int clEnqueueCopyImage ( CL_IMAGE_NUM_{SAMPLES, MIP_LEVELS},
flags: cl_command_queue command_queue, CL_IMAGE_DX9_MEDIA_PLANE_KHR,
CL_MEM_READ_WRITE, cl_mem src_image, cl_mem dst_image, CL_IMAGE_{D3D10, D3D11}_SUBRESOURCE_KHR
CL_MEM_{WRITE, READ}_ONLY, const size_t *src_origin, const size_t *dst_origin,
CL_MEM_HOST_{WRITE, READ}_ONLY, const size_t *region, cl_uint num_events_in_wait_list,
CL_MEM_HOST_NO_ACCESS, const cl_event *event_wait_list, cl_event *event)
CL_MEM_{USE, ALLOC, COPY}_HOST_PTR Image Formats [5.3.1.1, 9.5]
Query List of Supported Image Formats [5.3.2] Copy Between Image, Buffer Objects [5.3.4] Supported image formats: image_channel_order with
cl_int clEnqueueCopyImageToBuffer ( image_channel_data_type.
cl_int clGetSupportedImageFormats ( cl_command_queue command_queue,
cl_context context, cl_mem_flags flags, cl_mem src_image, cl_mem dst_buffer, Built-in support: [Table 5.8]
cl_mem_object_type image_type, const size_t *src_origin, const size_t *region, CL_RGBA: CL_HALF_FLOAT, CL_FLOAT, CL_UNORM_INT{8,16} ,
cl_uint num_entries, cl_image_format *image_formats, size_t dst_offset, cl_uint num_events_in_wait_list, CL_SIGNED_INT{8,16,32}, CL_UNSIGNED_INT{8,16,32}
cl_uint *num_image_formats) const cl_event *event_wait_list, cl_event *event)
flags: See clCreateImage CL_BGRA: CL_UNORM_INT8
image_type: CL_MEM_OBJECT_IMAGE{1D, 2D, 3D}, cl_int clEnqueueCopyBufferToImage ( Optional support: [Table 5.6]
CL_MEM_OBJECT_IMAGE1D_BUFFER, cl_command_queue command_queue,
CL_MEM_OBJECT_IMAGE{1D, 2D}_ARRAY cl_mem src_buffer, cl_mem dst_image, CL_R, CL_A: CL_HALF_FLOAT, CL_FLOAT, CL_UNORM_INT{8,16},
size_t src_offset, CL_SIGNED_INT{8,16,32} , CL_UNSIGNED_INT{8,16,32} ,
Read, Write, Copy Image Objects [5.3.3] const size_t *dst_origin, const size_t *region, CL_SNORM_INT{8,16}
cl_int clEnqueueReadImage ( cl_uint num_events_in_wait_list, CL_INTENSITY: CL_HALF_FLOAT, CL_FLOAT, CL_UNORM_INT{8,16},
cl_command_queue command_queue, const cl_event *event_wait_list, cl_event *event) CL_SNORM_INT{8|16}
cl_mem image, cl_bool blocking_read, Map and Unmap Image Objects [5.3.5]
const size_t *origin, const size_t *region, CL_LUMINANCE: CL_UNORM_INT{8,16}, CL_HALF_FLOAT,
size_t row_pitch, size_t slice_pitch, void *ptr, void * clEnqueueMapImage ( CL_FLOAT, CL_SNORM_INT{8,16}
cl_uint num_events_in_wait_list, cl_command_queue command_queue, cl_mem image, CL_RG, CL_RA: CL_HALF_FLOAT, CL_FLOAT, CL_UNORM_INT{8,16},
const cl_event *event_wait_list, cl_event *event) cl_bool blocking_map, cl_map_flags map_flags, CL_SIGNED_INT{8,16, 32} , CL_UNSIGNED_INT{8,16,32} ,
const size_t *origin, const size_t *region, CL_SNORM_INT{8,16}
cl_int clEnqueueWriteImage ( size_t *image_row_pitch, size_t *image_slice_pitch,
cl_command_queue command_queue, cl_uint num_events_in_wait_list, CL_RGB: CL_UNORM_SHORT_{555,565} , CL_UNORM_INT_101010
cl_mem image, cl_bool blocking_write, const cl_event *event_wait_list, cl_event *event,
const size_t *origin, const size_t *region, cl_int *errcode_ret) CL_ARGB: CL_UNORM_INT8, CL_SIGNED_INT8,
size_t input_row_pitch, size_t input_slice_pitch, CL_UNSIGNED_INT8, CL_SNORM_INT8
const void *ptr, cl_uint num_events_in_wait_list, Also see clGetMemObjectInfo [5.4.5]
const cl_event *event_wait_list, cl_event *event) CL_BGRA: CL_{SIGNED, UNSIGNED}_INT8, CL_SNORM_INT8

Image Read and Write Built-in Functions Read and write functions for 1D images (continued) Read and write functions for 2D images (continued)
[6.12.14] [9.4, 9.5.8] void write_imagef (image2d_t image, int2 coord, float4 color)
The built-in functions defined in this section can only be used void write_imagei (image1d_t image, int coord, int4 color)
void write_imagei (image1d_array_t image, int2 coord, void write_imagef (image2d_array_t image, int4 coord,
with image memory objects created with clCreateImage. float4 color)
sampler specifies the addressing and filtering mode to use. int4 color)
To enable the read_imageh and write_imageh forms, enable void write_imagei (image1d_buffer_t image, int coord, void write_imagei (image2d_t image, int2 coord, int4 color)
the extension cl_khr_fp16. To enable the type image3d_t in int4 color) void write_imagei (image2d_array_t image, int4 coord,
functions write_image{f, i, ui, h}, enable the extension int4 color)
cl_khr_3d_image_writes. void write_imageh (image1d_t image, int coord, half4 color)
void write_imageh (image1d_array_t image, int2 coord, void write_imageui (image2d_t image, int2 coord,
Read and write functions for 1D images half4 color) uint4 color)
Read an element from a 1D image, or write a color value to a void write_imageh (image1d_buffer_t image, int coord, void write_imageui (image2d_array_t image, int4 coord,
location in a 1D image. half4 color) uint4 color)
void write_imageui (image1d_t image, int coord, uint4 color) void write_imageh (image2d_t image, int2 coord, half4 color)
float4 read_imagef (image1d_t image, sampler_t sampler, void write_imageh (image2d_array_t image, int4 coord,
{int, float} coord) void write_imageui (image1d_array_t image, int2 coord,
uint4 color) half4 color)
float4 read_imagef (image1d_t image, int coord)
float4 read_imagef (image1d_array_t image, void write_imageui (image1d_buffer_t image, int coord,
uint4 color) Read and write functions for 3D images
sampler_t sampler, {int2, float4} coord) Read an element from a 3D image, or write a color value to a
float4 read_imagef (image1d_array_t image, int2 coord) location in a 3D image.
float4 read_imagef (image1d_buffer_t image, int coord) Read and write functions for 2D images
Read an element from a 2D image, or write a color value to a float4 read_imagef (image3d_t image, sampler_t sampler,
int4 read_imagei (image1d_t image, sampler_t sampler, location in a 2D image. {int4, float4} coord)
{int, float} coord) float4 read_imagef (image3d_t image, int4 coord)
int4 read_imagei (image1d_t image, int coord) float4 read_imagef (image2d_t image, sampler_t sampler, int4 read_imagei (image3d_t image, sampler_t sampler,
int4 read_imagei (image1d_array_t image, sampler_t sampler, {int2, float2} coord) {int4, float4} coord)
{int2, float2} coord) float4 read_imagef (image2d_t image, int2 coord) int4 read_imagei (image3d_t image, int4 coord)
int4 read_imagei (image1d_array_t image, int2 coord) float4 read_imagef (image2d_array_t image, uint4 read_imageui (image3d_t image, sampler_t sampler,
int4 read_imagei (image1d_buffer_t image, int coord) sampler_t sampler, {int4, float4} coord) {int4, float4} coord)
float4 read_imagef (image2d_array_t image, int4 coord) uint4 read_imageui (image3d_t image, int4 coord)
uint4 read_imageui (image1d_t image, sampler_t sampler,
{int, float} coord) int4 read_imagei (image2d_t image, sampler_t sampler, half4 read_imageh (image3d_t image, sampler_t sampler,
uint4 read_imageui (image1d_t image, int coord) {int2, float2} coord) {int4, float4} coord)
uint4 read_imageui (image1d_array_t image, int4 read_imagei (image2d_t image, int2 coord) half4 read_imageh (image3d_t image, int4 coord)
sampler_t sampler, {int2, float2} coord) int4 read_imagei (image2d_array_t image, sampler_t sampler,
Use this pragma to enable writes to type image3d_t:
uint4 read_imageui (image1d_array_t image, int2 coord) {int4, float4} coord)
int4 read_imagei (image2d_array_t image, int4 coord) #pragma OPENCL EXTENSION cl_khr_3d_image_writes : enable
uint4 read_imageui (image1d_buffer_t image, int coord)
void write_imagef (image3d_t image, int4 coord,
half4 read_imageh (image1d_t image, sampler_t sampler, uint4 read_imageui (image2d_t image, sampler_t sampler, float4 color)
{int, float} coord) {int2, float2} coord)
half4 read_imageh (image1d_t image, int coord) void write_imagei (image3d_t image, int4 coord, int4 color)
uint4 read_imageui (image2d_t image, int2 coord)
half4 read_imageh (image1d_array_t image, uint4 read_imageui (image2d_array_t image, void write_imageui (image3d_t image, int4 coord, uint4 color)
sampler_t sampler, {int2, float4} coord) sampler_t sampler, {int4, float4} coord) void write_imageh (image3d_t image, int4 coord, half4 color)
half4 read_imageh (image1d_array_t image, int2 coord) uint4 read_imageui (image2d_array_t image, int4 coord)
half4 read_imageh (image1d_buffer_t image, int coord)
half4 read_imageh (image2d_t image, sampler_t sampler,
void write_imagef (image1d_t image, int coord, float4 color) {int2, float2} coord) Access Qualifiers [6.6]
void write_imagef (image1d_array_t image, int2 coord, half4 read_imageh (image2d_t image, int2 coord) Apply to 2D and 3D image types to declare if the image
half4 read_imageh (image2d_array_t image, memory object is being read or written by a kernel.
float4 color)
void write_imagef (image1d_buffer_t image, int coord, sampler_t sampler, {int4, float4} coord) __read_only, read_only
float4 color) half4 read_imageh (image2d_array_t image, int4 coord) __write_only, write_only


OpenCL Image Processing (continued): Following is a subset of the OpenCL specification that pertains to image processing and graphics.

Sampler Objects [5.5] Image Query Functions [6.12.14.5] Query image dimensions
cl_sampler clCreateSampler ( Query image width, height, and depth in pixels int2 get_image_dim (image2d_t image)
cl_context context, cl_bool normalized_coords, int2 get_image_dim (image2d_array_t image)
cl_addressing_mode addressing_mode, int get_image_width (image{1,2,3}d_t image) int4 get_image_dim (image3d_t image)
cl_filter_mode filter_mode, cl_int *errcode_ret) int get_image_width (image1d_buffer_t image)
addressing_mode: CL_ADDRESS_[MIRRORED_]REPEAT, int get_image_width (image{1,2}d_array_t image) Query image Channel data type and order
CL_ADDRESS_CLAMP[_TO_EDGE], CL_ADDRESS_NONE int get_image_height (image{2,3}d_t image)
filter_mode: CL_FILTER_{NEAREST, LINEAR} int get_image_channel_data_type (image{1,2,3}d_t image)
int get_image_height (image2d_array_t image) int get_image_channel_data_type (image1d_buffer_t image)
cl_int clRetainSampler (cl_sampler sampler)
int get_image_depth (image3d_t image) int get_image_channel_data_type (image{1,2}d_array_t image)
cl_int clReleaseSampler (cl_sampler sampler)
cl_int clGetSamplerInfo (cl_sampler sampler, Query image array size int get_image_channel_order (image{1,2,3}d_t image)
cl_sampler_info param_name, int get_image_channel_order (image1d_buffer_t image)
size_t param_value_size, void *param_value, size_t get_image_array_size (image1d_array_t image) int get_image_channel_order (image{1,2}d_array_t image)
size_t *param_value_size_ret) size_t get_image_array_size (image2d_array_t image)
param_name: CL_SAMPLER_REFERENCE_COUNT,
CL_SAMPLER_{CONTEXT, FILTER_MODE},
CL_SAMPLER_ADDRESSING_MODE,
CL_SAMPLER_NORMALIZED_COORDS OpenGL Sharing *gl_object_type returns:
CL_GL_OBJECT_TEXTURE_BUFFER,
Functions available if cl_khr_gl_sharing or cl_apple_gl_sharing CL_GL_OBJECT_TEXTURE{1D, 2D, 3D},
Sampler Declaration Fields [6.12.14.1] is supported. Creating OpenCL memory objects from OpenGL CL_GL_OBJECT_TEXTURE{1D, 2D}_ARRAY,
The sampler can be passed as an argument to the kernel objects using clCreateFromGLBuffer, clCreateFromGLTexture, and CL_GL_OBJECT_{BUFFER, RENDERBUFFER}
using clSetKernelArg, or can be declared in the outermost clCreateFromGLRenderbuffer ensure the OpenGL object will not be
scope of kernel functions, or it can be a constant variable of deleted while the corresponding OpenCL memory object exists. cl_int clGetGLTextureInfo (cl_mem memobj,
type sampler_t declared in the program source. cl_gl_texture_info param_name,
size_t param_value_size, void *param_value,
const sampler_t <sampler-name> = CL Buffer Objects > GL Buffer Objects [9.7.2] size_t *param_value_size_ret)
<normalized-mode> | <address-mode> | <filter-mode> cl_mem clCreateFromGLBuffer (cl_context context,
normalized-mode: cl_mem_flags flags, GLuint bufobj, cl_int *errcode_ret) param_name:
CLK_NORMALIZED_COORDS_{TRUE, FALSE} CL_GL_{TEXTURE_TARGET, MIPMAP_LEVEL}
flags: CL_MEM_{READ, WRITE}_ONLY, CL_MEM_READ_WRITE
address-mode: Share Objects [9.7.6]
CLK_ADDRESS_{REPEAT, CLAMP, NONE}, CL Image Objects > GL Textures [9.7.3]
CLK_ADDRESS_{CLAMP_TO_EDGE, MIRRORED_REPEAT} cl_int clEnqueueAcquireGLObjects (
cl_mem clCreateFromGLTexture (cl_context context, cl_command_queue command_queue,
filter-mode: CLK_FILTER_NEAREST, CLK_FILTER_LINEAR cl_mem_flags flags, GLenum texture_target, cl_uint num_objects, const cl_mem *mem_objects,
GLint miplevel, GLuint texture, cl_int *errcode_ret) cl_uint num_events_in_wait_list,
Direct3D 10 Sharing [9.9] flags: See clCreateFromGLBuffer
Provide interoperability between OpenCL and Direct3D 10. texture_target: GL_TEXTURE_{1D, 2D}[_ARRAY], cl_int clEnqueueReleaseGLObjects (
If supported, cl_khr_d3d10_sharing will be present in GL_TEXTURE_{3D, BUFFER, RECTANGLE}, cl_command_queue command_queue,
CL_PLATFORM_EXTENSIONS or CL_DEVICE_EXTENSIONS. GL_TEXTURE_CUBE_MAP_POSITIVE_{X, Y, Z}, cl_uint num_objects, const cl_mem *mem_objects,
GL_TEXTURE_CUBE_MAP_NEGATIVE_{X, Y, Z} cl_uint num_events_in_wait_list,
cl_int clGetDeviceIDsFromD3D10KHR ( const cl_event *event_wait_list, cl_event *event)
cl_platform_id platform,
cl_d3d10_device_source_khr d3d_device_source, CL Image Objects > GL Renderbuffers [9.7.4] CL Event Objects > GL Sync Objects [9.8.2]
void *d3d_object, cl_mem clCreateFromGLRenderbuffer ( cl_event clCreateEventFromGLsyncKHR (
cl_d3d10_device_set_khr d3d_device_set, cl_context context, cl_mem_flags flags, cl_context context, GLsync sync, cl_int *errcode_ret)
cl_uint num_entries, cl_device_id *devices, GLuint renderbuffer, cl_int *errcode_ret)
cl_uint *num_devices) flags: CL_MEM_{READ, WRITE}_ONLY, CL_MEM_READ_WRITE CL Context > GL Context, Sharegroup [9.6.5]
d3d_device_source: cl_int clGetGLContextInfoKHR (
CL_D3D10_{DEVICE, DXGI_ADAPTER}_KHR const cl_context_properties *properties,
d3d_device_set:
Query Information [9.7.5] cl_gl_context_info param_name, size_t param_value_size,
CL_{ALL, PREFERRED}_DEVICES_FOR_D3D10_KHR cl_int clGetGLObjectInfo (cl_mem memobj, void *param_value, size_t *param_value_size_ret)
cl_gl_object_type *gl_object_type,
cl_mem clCreateFromD3D10BufferKHR ( GLuint *gl_object_name) param_name: CL_DEVICES_FOR_GL_CONTEXT_KHR,
cl_context context, cl_mem_flags flags, CL_CURRENT_DEVICE_FOR_GL_CONTEXT_KHR
ID3D10Buffer *resource, cl_int *errcode_ret)
cl_mem clCreateFromD3D10Texture2DKHR ( DX9 Media Surface Sharing [9.10] cl_mem clCreateFromDX9MediaSurfaceKHR (
cl_context context, cl_mem_flags flags,
cl_context context, cl_mem_flags flags, These functions allow applications to use media surfaces cl_dx9_media_adapter_type_khr adapter_type,
ID3D10Texture2D *resource, UINT subresource, as OpenCL memory objects. If this extension in supported, void *surface_info, cl_uint plane, cl_int *errcode_ret)
cl_int *errcode_ret) cl_khr_dx9_media_sharing will be present in
flags: See clCreateFromD3D10BufferKHR CL_PLATFORM_EXTENSIONS or CL_DEVICE_EXTENSIONS.
adapter_type: CL_ADAPTER_{D3D9, D3D9EX, DXVA}_KHR
cl_mem clCreateFromD3D10Texture3DKHR ( cl_int clGetDeviceIDsFromDX9MediaAdapterKHR (
cl_context context, cl_mem_flags flags, cl_platform_id platform, cl_uint num_media_adapters, cl_int clEnqueueAcquireDX9MediaSurfacesKHR (
ID3D10Texture3D *resource, UINT subresource, cl_dx9_media_adapter_type_khr *media_adapters_type, cl_command_queue command_queue,
cl_int *errcode_ret) void *media_adapters, cl_uint num_objects, const cl_mem *mem_objects,
cl_dx9_mem_adapter_set_khr media_adapter_set, cl_uint num_events_in_wait_list,
flags: See clCreateFromD3D10BufferKHR const cl_event *event_wait_list, cl_event *event)
cl_uint num_entries, cl_device_id *devices,
cl_int clEnqueueAcquireD3D10ObjectsKHR ( cl_int *num_devices) cl_int clEnqueueReleaseDX9MediaSurfacesKHR (
cl_ command_queue command_queue, media_adapter_type: CL_ADAPTER_{D3D9, D3D9EX, DXVA}_KHR cl_command_queue command_queue,
cl_uint num_objects, const cl_mem *mem_objects, cl_uint num_objects, const cl_mem *mem_objects,
cl_uint num_events_in_wait_list, media_adapter_set: CL_ALL_DEVICES_FOR_DXP_MEDIA_ADAPTER_KHR,
const cl_event *event_wait_list, cl_event *event) CL_PREFERRED_DEVICES_FOR_DX9_MEDIA_ADAPTER_KHR cl_uint num_events_in_wait_list,
cl_int clEnqueueReleaseD3D10ObjectsKHR (
cl_ command_queue command_queue,
cl_uint num_objects, const cl_mem *mem_objects, cl_int clGetDeviceIDsFromD3D11KHR ( cl_mem clCreateFromD3D11Texture3DKHR (
cl_uint num_events_in_wait_list, cl_platform_id platform, cl_context context, cl_mem_flags flags,
const cl_event *event_wait_list, cl_event *event) cl_d3d11_device_source_khr d3d_device_source, ID3D11Texture3D *resource, UINT subresource,
void *d3d_object, cl_int *errcode_ret)
cl_d3d11_device_set_khr d3d_device_set, flags: CL_MEM_{READ, WRITE}_ONLY, CL_MEM_READ_WRITE
cl_uint num_entries, cl_device_id *devices,
Direct3D 11 Sharing [9.11] cl_uint *num_devices) cl_int clEnqueueAcquireD3D11ObjectsKHR (
Provide interoperability between OpenCL and Direct3D 11. If d3d_device_source: CL_D3D11_DEVICE_KHR, cl_command_queue command_queue,
supported, cl_khr_d3d11_sharing will be present in CL_D3D11_DXGI_ADAPTER_KHR cl_uint num_objects, const cl_mem *mem_objects,
CL_PLATFORM_EXTENSIONS or CL_DEVICE_EXTENSIONS. d3d_device_set: CL_PREFERRED_DEVICES_FOR_D3D11_KHR, cl_uint num_events_in_wait_list,
CL_ALL_DEVICES_FOR_D3D11_KHR const cl_event *event_wait_list, cl_event *event)
cl_mem clCreateFromD3D11Texture2DKHR (
cl_context context, cl_mem_flags flags, cl_mem clCreateFromD3D11BufferKHR ( cl_int clEnqueueReleaseD3D11ObjectsKHR (
ID3D11Texture2D *resource, cl_context context, cl_mem_flags flags, cl_command_queue command_queue,
UINT subresource, cl_int *errcode_ret) ID3D11Buffer *resource, cl_int *errcode_ret) cl_uint num_objects, const cl_mem *mem_objects,
flags: CL_MEM_{READ, WRITE}_ONLY, CL_MEM_READ_WRITE flags: CL_MEM_{READ, WRITE}_ONLY, CL_MEM_READ_WRITE cl_uint num_events_in_wait_list,


OpenCL Reference Card Index
The following index shows each item included on this card along with the page on which it is described. The color of the row in the table below is the color of the box to which you should refer.
A clEnqueueCopyImage 6 clWaitForEvents 2 Image Query Functions 7 Q
abs, abs_diff 3 clEnqueueCopyImageToBuffer 6 clz 3 Image Processing 6,7 Qualifiers 3
Access Qualifiers 6 clEnqueueFillBuffer 1 Command Queues 1 Image Read and Write Functions 6 Query image information 7
acos, acosh, acospi 4 clEnqueueFillImage 6 Common Functions 5 INFINITY 3 Query Image Objects 6
add_sat 3 clEnqueueMapBuffer 1 Compiler Options 2 Integer Functions 3 Query List Supported Image Formats 6
Address Space Qualifiers 3 clEnqueueMapImage 6 Contexts 1 isequal 5 Query Memory Object 1
all 5 clEnqueueMarkerWithWaitList 2 Conversions and Type Casting 5 isfinite 5 Query Program Objects 2
any 5 clEnqueueMigrateMemObjects 1 convert_T 5 Querying Platform Info, Devices 1
isgreater, isgreaterequal 5
Architecture Diagram 7 clEnqueueNativeKernel 2 Copy Between Image, Buffer 6
isinf 5 R
asin, asinh, asinpi 4 clEnqueueNDRangeKernel 2 copysign 4
isless, islessequal, islessgreater 5 radians 5
Async Copies and Prefetch 4 clEnqueueReadBuffer 1 cos, cosh, cospi 4
Create Buffer Objects 1 isnan 5 Read, Write, Copy Buffer Objects 1
async_work_group_copy 4 clEnqueueReadBufferRect 1
Create Image Objects 6 isnormal 5 Read, Write, Copy Image Objects 6
async_work_group_strided_copy 4 clEnqueueReadImage 6
Create Kernel Objects 2 isnotequal 5 read_image{f, i, ui, h} 6
atan, atanh, atanpi 4 clEnqueue{Acquire, Release} 7 Create Program Objects 2 isordered, isunordered 5 read_mem_fence 5
atan2, atan2pi 4 D3D10Objects KHR
cross 4 __read_only 6
Atomic Functions 5 clEnqueueReleaseGLObjects 7 K recip (native, half) 4
atomic_, atom_ 5 clEnqueueTask 2 Kernel and Event Objects 2
D Relational Functions 5
Attributes (Type) 3 clEnqueueUnmapMemObject 1 Kernel Args. & Object Queries 2
D3D10 Sharing 7 remainder 4
Attributes (Function) 3 clEnqueueWriteBuffer 1 D3D11 Sharing 7 remquo 4
clEnqueueWriteBufferRect 1 L
B Data Types 3 Reserved Data Types 3
clEnqueueWriteImage 6 ldexp 4
barrier 5 degrees 5 rhadd 3
clFinish, clFlush 2 Device Architecture Diagram 7 length 4
bitselect 5 rint 4
clGetCommandQueueInfo 1 distance 4 lgamma, lgamma_r 4 rootn 4
Buffer Objects 1
clGetContextInfo 1 divide, {half_, native_} 4 Linker Options 2 rotate 3
Building Program Executables 2
clGetDeviceIDs 1 dot 4 log, log2, log10, log1p, logb 4 round 4
C clGetDeviceIDsFromD3D10KHR 7 DX9 Media Surface Sharing 7 Rounding modes 5
M
cbrt 4 clGetDeviceIDsFromD3D11KHR 7 rsqrt 4
E mad 4
ceil 4 clGetDeviceIDsFromDX9Media... 7 Runtime 1
erf, erfc 4 mad_hi, mad_sat, mad24 3
clamp (Common) 5 clGetDeviceInfo 1
clamp (Integer) 3 Event Objects 2 Map and Unmap Image Objects 6 S
clGetEventInfo 2
Class Diagram 7 Execute Kernels 2 Map Buffer Objects 1 Sampler Declaration Fields 7
clGetEventProfilingInfo 2
cl_apple_gl_sharing 7 exp, exp2, exp10, expm1 4 Markers, Barriers, Wait for Events 2 Sampler Objects 7
clGetExtensionFunctionAddress 1
cl_khr_3d_image_writes 6 ForPlatform EXTENSION 3 Math Constants 3 sampler_t 7
cl_khr_d3d10_sharing 7 clGetGLContextInfoKHR 7 Math Functions 4 Scalar Data Types (Built-in) 3
F select 5
cl_khr_d3d11_sharing 7 clGetGLObjectInfo 7 Math intrinsics options 2
fabs 4 Separate Compilation, Linking 2
cl_khr_dx9_media_sharing 7 clGetGLTextureInfo 7 max (common) 5
fast_{distance, length, normalize} 4 shuffle, shuffle2 5
cl_khr_fp16 6 clGetImageInfo 6 fdim 4 max (integer) 3
clGetKernelArgInfo 2 MAXFLOAT 3 sign 5
cl_khr_gl_sharing 7 floor 4
clGetKernelInfo 2 maxmag 4 signbit 5
cl_khr_int64_base_atomics 5 Flush and Finish 2 sin, sincos, sinh, sinpi 4
cl_khr_int64_extended_atomics 5 clGetKernelWorkGroupInfo 2 fma 4 mem_fence 5
smoothstep 5
clBuildProgram 2 clGetMemObjectInfo 1 fmin, fmax 4 Memory Objects 1
sqrt 4
clCompileProgram 2 clGetPlatformIDs 1 fmod 4 Migrate Memory Objects 1
step 5
clCreateBuffer 1 clGetPlatformInfo 1 FP_CONTRACT 3 min (common) 5 sub_sat 3
clCreateCommandQueue 1 clGetProgramBuildInfo 2 FP_FAST_FMA* 3 min (integer) 3 Synchronization, Explicit
clCreateContext 1 clGetProgramInfo 2 fract 4 minmag 4 Memory Fence Functions 5
clCreateContextFromType 1 clGetSamplerInfo 7 frexp 4 mix 5 T
clCreateEventFromGLsyncKHR 7 clGetSupportedImageFormats 6 Function Qualifiers 3 modf 4
tan, tanh, tanpi 4
clCreateFromD3D10* 7 clLinkProgram 2 mul_hi, mul24 3
G tgamma 4
clCreateFromD3D11* 7 clReleaseCommandQueue 1
Geometric Functions 4 N trunc 4
clCreateFromDX9MediaSurface... 7 clReleaseContext 1 Type Attributes 3
get_array_size 7 NAN 3
clCreateFromGL* 7 clReleaseDevice 1
get_global_{id, offset, size} 4 nan 4 U
clCreateImage 6 clReleaseEvent 2
get_group_id 4 nextafter 4
clCreateKernel 2 clReleaseKernel 2 Unload OpenCL Compiler 2
get_image_{width, height, depth} 7 normalize 4
clCreateKernelsInProgram 2 clReleaseMemObject 1 upsample 3
get_image_channel_data_type 7
clCreateProgramWith clReleaseProgram 2 O V
{Binary, BuiltInKernels, Source} 2 get_image_channel_order 7
clReleaseSampler 7 OpenGL Sharing 7 vec_step 5
clCreateSampler 7 get_image_dim 7
clRetainCommandQueue 1 Operators 3 Vector Components 3
1 get_local_{id, size} 4
clCreateSubBuffer clRetainContext 1 Optimization options 2 Vector Data Types (Built-in) 3
get_num_groups 4
clCreateSubDevices 1 clRetainDevice 1 Vector Data Load/Store Functions 4
get_work_dim 4
clCreateUserEvent 2 clRetainEvent 2 P Vector Functions 5
GL Sharing 7
clEnqueue{Acquire, Release} 7 clRetainKernel 2 Partitioning a Device 1 vload* 4
D3D10Objects KHR Platform layer 1
clRetainMemObject 1 H vstore* 4
clEnqueue{Acquire, Release} 7 popcount 3
D3D11ObjectsKHR clRetainProgram 2 hadd 3 W
clRetainSampler 7 HUGE_VAL, HUGE_VALF 3 pow, pown, powr 4
clEnqueue{Acquire, Release} 7 wait_group_events 4
DX9MediaSurfaceKHR clSetEventCallback 2 hypot 4 prefetch 4
Warning request/suppress 2
clEnqueueAcquireGLObjects 7 clSetKernelArg 2 Preprocessor Directives, Macros 3 options
clEnqueueBarrierWithWaitList 2 I Preprocessor options 2 Work-Item Functions 4
clSetMemObjectDestructor 1
clEnqueueCopyBuffer 1 Callback ilogb 4 printf 5 write_image{f, i, ui, h} 6
clEnqueueCopyBufferRect 1 clSetUserEventStatus 2 Image Formats 6 Profiling Operations 2 write_mem_fence 5
clEnqueueCopyBufferToImage 6 clUnloadPlatformCompiler 2 Image Objects 6 Program Objects 2 __write_only 6

The Khronos Group is an industry consortium creating open standards for the authoring and
acceleration of parallel computing, graphics and dynamic media on a wide variety of platforms and
devices. See www.khronos.org to learn more about the Khronos Group.
OpenCL is a trademark of Apple Inc. and is used under license by Khronos.

©2011 Khronos Group - Rev. 1111 Reference card production by Miller & Mattson www.millermattson.com www.khronos.org/opencl

Opencl cheet sheet

More Related Content

What's hot (20)

Similar to Opencl cheet sheet (20)

More from Piyush Mittal (20)

Recently uploaded (20)

Opencl cheet sheet