CUDA Driver API :: CUDA Toolkit Documentation

4.15. Execution Control

This section describes the execution control functions of the low-level CUDA driver application programming interface.

Functions

CUresult cuFuncGetAttribute ( int* pi, CUfunction_attribute attrib, CUfunction hfunc ): Returns information about a function.
CUresult cuFuncSetAttribute ( CUfunction hfunc, CUfunction_attribute attrib, int value ): Sets information about a function.
CUresult cuFuncSetCacheConfig ( CUfunction hfunc, CUfunc_cache config ): Sets the preferred cache configuration for a device function.
CUresult cuFuncSetSharedMemConfig ( CUfunction hfunc, CUsharedconfig config ): Sets the shared memory configuration for a device function.
CUresult cuLaunchCooperativeKernel ( CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void** kernelParams ): Launches a CUDA function where thread blocks can cooperate and synchronize as they execute.
CUresult cuLaunchCooperativeKernelMultiDevice ( CUDA_LAUNCH_PARAMS* launchParamsList, unsigned int numDevices, unsigned int flags ): Launches CUDA functions on multiple devices where thread blocks can cooperate and synchronize as they execute.
CUresult cuLaunchKernel ( CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void** kernelParams, void** extra ): Launches a CUDA function.

Functions

CUresult cuFuncGetAttribute ( int* pi, CUfunction_attribute attrib, CUfunction hfunc )

Returns information about a function.

Parameters

pi: - Returned attribute value
attrib: - Attribute requested
hfunc: - Function to query attribute of

Returns

CUDA_SUCCESS, CUDA_ERROR_DEINITIALIZED, CUDA_ERROR_NOT_INITIALIZED, CUDA_ERROR_INVALID_CONTEXT, CUDA_ERROR_INVALID_HANDLE, CUDA_ERROR_INVALID_VALUE

Description

Returns in *pi the integer value of the attribute attrib on the kernel given by hfunc. The supported attributes are:

CU_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK: The maximum number of threads per block, beyond which a launch of the function would fail. This number depends on both the function and the device on which the function is currently loaded.
CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES: The size in bytes of statically-allocated shared memory per block required by this function. This does not include dynamically-allocated shared memory requested by the user at runtime.
CU_FUNC_ATTRIBUTE_CONST_SIZE_BYTES: The size in bytes of user-allocated constant memory required by this function.
CU_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES: The size in bytes of local memory used by each thread of this function.
CU_FUNC_ATTRIBUTE_NUM_REGS: The number of registers used by each thread of this function.
CU_FUNC_ATTRIBUTE_PTX_VERSION: The PTX virtual architecture version for which the function was compiled. This value is the major PTX version * 10 + the minor PTX version, so a PTX version 1.3 function would return the value 13. Note that this may return the undefined value of 0 for cubins compiled prior to CUDA 3.0.
CU_FUNC_ATTRIBUTE_BINARY_VERSION: The binary architecture version for which the function was compiled. This value is the major binary version * 10 + the minor binary version, so a binary version 1.3 function would return the value 13. Note that this will return a value of 10 for legacy cubins that do not have a properly-encoded binary architecture version.
CU_FUNC_CACHE_MODE_CA: The attribute to indicate whether the function has been compiled with user specified option "-Xptxas --dlcm=ca" set .
CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES: The maximum size in bytes of dynamically-allocated shared memory.
CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT: Preferred shared memory-L1 cache split ratio in percent of shared memory.

Note:

Note that this function may also return error codes from previous, asynchronous launches.

CUresult cuFuncSetAttribute ( CUfunction hfunc, CUfunction_attribute attrib, int value )

Sets information about a function.

Parameters

hfunc: - Function to query attribute of
attrib: - Attribute requested
value: - The value to set

Returns

CUDA_SUCCESS, CUDA_ERROR_DEINITIALIZED, CUDA_ERROR_NOT_INITIALIZED, CUDA_ERROR_INVALID_CONTEXT, CUDA_ERROR_INVALID_HANDLE, CUDA_ERROR_INVALID_VALUE

Description

This call sets the value of a specified attribute attrib on the kernel given by hfunc to an integer value specified by val This function returns CUDA_SUCCESS if the new value of the attribute could be successfully set. If the set fails, this call will return an error. Not all attributes can have values set. Attempting to set a value on a read-only attribute will result in an error (CUDA_ERROR_INVALID_VALUE)

Supported attributes for the cuFuncSetAttribute call are:

CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES: This maximum size in bytes of dynamically-allocated shared memory. The value should contain the requested maximum size of dynamically-allocated shared memory. The sum of this value and the function attribute CU_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES cannot exceed the device attribute CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN. The maximal size of requestable dynamic shared memory may differ by GPU architecture.
CU_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT: On devices where the L1 cache and shared memory use the same hardware resources, this sets the shared memory carveout preference, in percent of the total resources. This is only a hint, and the driver can choose a different ratio if required to execute the function.

Note:

Note that this function may also return error codes from previous, asynchronous launches.

CUresult cuFuncSetCacheConfig ( CUfunction hfunc, CUfunc_cache config )

Sets the preferred cache configuration for a device function.

Parameters

hfunc: - Kernel to configure cache for
config: - Requested cache configuration

Returns

CUDA_SUCCESS, CUDA_ERROR_INVALID_VALUE, CUDA_ERROR_DEINITIALIZED, CUDA_ERROR_NOT_INITIALIZED, CUDA_ERROR_INVALID_CONTEXT

Description

On devices where the L1 cache and shared memory use the same hardware resources, this sets through config the preferred cache configuration for the device function hfunc. This is only a preference. The driver will use the requested configuration if possible, but it is free to choose a different configuration if required to execute hfunc. Any context-wide preference set via cuCtxSetCacheConfig() will be overridden by this per-function setting unless the per-function setting is CU_FUNC_CACHE_PREFER_NONE. In that case, the current context-wide setting will be used.

This setting does nothing on devices where the size of the L1 cache and shared memory are fixed.

Launching a kernel with a different preference than the most recent preference setting may insert a device-side synchronization point.

The supported cache configurations are:

CU_FUNC_CACHE_PREFER_NONE: no preference for shared memory or L1 (default)
CU_FUNC_CACHE_PREFER_SHARED: prefer larger shared memory and smaller L1 cache
CU_FUNC_CACHE_PREFER_L1: prefer larger L1 cache and smaller shared memory
CU_FUNC_CACHE_PREFER_EQUAL: prefer equal sized L1 cache and shared memory

Note:

Note that this function may also return error codes from previous, asynchronous launches.

CUresult cuFuncSetSharedMemConfig ( CUfunction hfunc, CUsharedconfig config )

Sets the shared memory configuration for a device function.

Parameters

hfunc: - kernel to be given a shared memory config
config: - requested shared memory configuration

Returns

CUDA_SUCCESS, CUDA_ERROR_INVALID_VALUE, CUDA_ERROR_DEINITIALIZED, CUDA_ERROR_NOT_INITIALIZED, CUDA_ERROR_INVALID_CONTEXT

Description

On devices with configurable shared memory banks, this function will force all subsequent launches of the specified device function to have the given shared memory bank size configuration. On any given launch of the function, the shared memory configuration of the device will be temporarily changed if needed to suit the function's preferred configuration. Changes in shared memory configuration between subsequent launches of functions, may introduce a device side synchronization point.

Any per-function setting of shared memory bank size set via cuFuncSetSharedMemConfig will override the context wide setting set with cuCtxSetSharedMemConfig.

Changing the shared memory bank size will not increase shared memory usage or affect occupancy of kernels, but may have major effects on performance. Larger bank sizes will allow for greater potential bandwidth to shared memory, but will change what kinds of accesses to shared memory will result in bank conflicts.

This function will do nothing on devices with fixed shared memory bank size.

The supported bank configurations are:

CU_SHARED_MEM_CONFIG_DEFAULT_BANK_SIZE: use the context's shared memory configuration when launching this function.
CU_SHARED_MEM_CONFIG_FOUR_BYTE_BANK_SIZE: set shared memory bank width to be natively four bytes when launching this function.
CU_SHARED_MEM_CONFIG_EIGHT_BYTE_BANK_SIZE: set shared memory bank width to be natively eight bytes when launching this function.

Note:

Note that this function may also return error codes from previous, asynchronous launches.

CUresult cuLaunchCooperativeKernel ( CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void** kernelParams )

Launches a CUDA function where thread blocks can cooperate and synchronize as they execute.

Parameters

f: - Kernel to launch
gridDimX: - Width of grid in blocks
gridDimY: - Height of grid in blocks
gridDimZ: - Depth of grid in blocks
blockDimX: - X dimension of each thread block
blockDimY: - Y dimension of each thread block
blockDimZ: - Z dimension of each thread block
sharedMemBytes: - Dynamic shared-memory size per thread block in bytes
hStream: - Stream identifier
kernelParams: - Array of pointers to kernel parameters

Returns

CUDA_SUCCESS, CUDA_ERROR_DEINITIALIZED, CUDA_ERROR_NOT_INITIALIZED, CUDA_ERROR_INVALID_CONTEXT, CUDA_ERROR_INVALID_HANDLE, CUDA_ERROR_INVALID_IMAGE, CUDA_ERROR_INVALID_VALUE, CUDA_ERROR_LAUNCH_FAILED, CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES, CUDA_ERROR_LAUNCH_TIMEOUT, CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING, CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE, CUDA_ERROR_SHARED_OBJECT_INIT_FAILED

Description

Invokes the kernel f on a gridDimX x gridDimY x gridDimZ grid of blocks. Each block contains blockDimX x blockDimY x blockDimZ threads.

sharedMemBytes sets the amount of dynamic shared memory that will be available to each thread block.

The device on which this kernel is invoked must have a non-zero value for the device attribute CU_DEVICE_ATTRIBUTE_COOPERATIVE_LAUNCH.

The total number of blocks launched cannot exceed the maximum number of blocks per multiprocessor as returned by cuOccupancyMaxActiveBlocksPerMultiprocessor (or cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags) times the number of multiprocessors as specified by the device attribute CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT.

The kernel cannot make use of CUDA dynamic parallelism.

Kernel parameters must be specified via kernelParams. If f has N parameters, then kernelParams needs to be an array of N pointers. Each of kernelParams[0] through kernelParams[N-1] must point to a region of memory from which the actual kernel parameter will be copied. The number of kernel parameters and their offsets and sizes do not need to be specified as that information is retrieved directly from the kernel's image.

Calling cuLaunchCooperativeKernel() sets persistent function state that is the same as function state set through cuLaunchKernel API

When the kernel f is launched via cuLaunchCooperativeKernel(), the previous block shape, shared size and parameter info associated with f is overwritten.

Note that to use cuLaunchCooperativeKernel(), the kernel f must either have been compiled with toolchain version 3.2 or later so that it will contain kernel parameter information, or have no kernel parameters. If either of these conditions is not met, then cuLaunchCooperativeKernel() will return CUDA_ERROR_INVALID_IMAGE.

Note:

This function uses standard default stream semantics.
Note that this function may also return error codes from previous, asynchronous launches.

CUresult cuLaunchCooperativeKernelMultiDevice ( CUDA_LAUNCH_PARAMS* launchParamsList, unsigned int numDevices, unsigned int flags )

Launches CUDA functions on multiple devices where thread blocks can cooperate and synchronize as they execute.

Parameters

launchParamsList: - List of launch parameters, one per device
numDevices: - Size of the launchParamsList array
flags: - Flags to control launch behavior

Returns

Description

Invokes kernels as specified in the launchParamsList array where each element of the array specifies all the parameters required to perform a single kernel launch. These kernels can cooperate and synchronize as they execute. The size of the array is specified by numDevices.

No two kernels can be launched on the same device. All the devices targeted by this multi-device launch must be identical. All devices must have a non-zero value for the device attribute CU_DEVICE_ATTRIBUTE_COOPERATIVE_MULTI_DEVICE_LAUNCH.

All kernels launched must be identical with respect to the compiled code. Note that any __device__, __constant__ or __managed__ variables present in the module that owns the kernel launched on each device, are independently instantiated on every device. It is the application's responsiblity to ensure these variables are initialized and used appropriately.

The size of the grids as specified in blocks, the size of the blocks themselves and the amount of shared memory used by each thread block must also match across all launched kernels.

The streams used to launch these kernels must have been created via either cuStreamCreate or cuStreamCreateWithPriority. The NULL stream or CU_STREAM_LEGACY or CU_STREAM_PER_THREAD cannot be used.

The total number of blocks launched per kernel cannot exceed the maximum number of blocks per multiprocessor as returned by cuOccupancyMaxActiveBlocksPerMultiprocessor (or cuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags) times the number of multiprocessors as specified by the device attribute CU_DEVICE_ATTRIBUTE_MULTIPROCESSOR_COUNT. Since the total number of blocks launched per device has to match across all devices, the maximum number of blocks that can be launched per device will be limited by the device with the least number of multiprocessors.

The kernels cannot make use of CUDA dynamic parallelism.

The CUDA_LAUNCH_PARAMS structure is defined as:

‎        typedef struct CUDA_LAUNCH_PARAMS_st
              {
                  CUfunction function;
                  unsigned int gridDimX;
                  unsigned int gridDimY;
                  unsigned int gridDimZ;
                  unsigned int blockDimX;
                  unsigned int blockDimY;
                  unsigned int blockDimZ;
                  unsigned int sharedMemBytes;
                  CUstream hStream;
                  void **kernelParams;
              } CUDA_LAUNCH_PARAMS;

where:

CUDA_LAUNCH_PARAMS::function specifies the kernel to be launched. All functions must be identical with respect to the compiled code.
CUDA_LAUNCH_PARAMS::gridDimX is the width of the grid in blocks. This must match across all kernels launched.
CUDA_LAUNCH_PARAMS::gridDimY is the height of the grid in blocks. This must match across all kernels launched.
CUDA_LAUNCH_PARAMS::gridDimZ is the depth of the grid in blocks. This must match across all kernels launched.
CUDA_LAUNCH_PARAMS::blockDimX is the X dimension of each thread block. This must match across all kernels launched.
CUDA_LAUNCH_PARAMS::blockDimX is the Y dimension of each thread block. This must match across all kernels launched.
CUDA_LAUNCH_PARAMS::blockDimZ is the Z dimension of each thread block. This must match across all kernels launched.
CUDA_LAUNCH_PARAMS::sharedMemBytes is the dynamic shared-memory size per thread block in bytes. This must match across all kernels launched.
CUDA_LAUNCH_PARAMS::hStream is the handle to the stream to perform the launch in. This cannot be the NULL stream or CU_STREAM_LEGACY or CU_STREAM_PER_THREAD. The CUDA context associated with this stream must match that associated with CUDA_LAUNCH_PARAMS::function.
CUDA_LAUNCH_PARAMS::kernelParams is an array of pointers to kernel parameters. If CUDA_LAUNCH_PARAMS::function has N parameters, then CUDA_LAUNCH_PARAMS::kernelParams needs to be an array of N pointers. Each of CUDA_LAUNCH_PARAMS::kernelParams[0] through CUDA_LAUNCH_PARAMS::kernelParams[N-1] must point to a region of memory from which the actual kernel parameter will be copied. The number of kernel parameters and their offsets and sizes do not need to be specified as that information is retrieved directly from the kernel's image.

By default, the kernel won't begin execution on any GPU until all prior work in all the specified streams has completed. This behavior can be overridden by specifying the flag CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_PRE_LAUNCH_SYNC. When this flag is specified, each kernel will only wait for prior work in the stream corresponding to that GPU to complete before it begins execution.

Similarly, by default, any subsequent work pushed in any of the specified streams will not begin execution until the kernels on all GPUs have completed. This behavior can be overridden by specifying the flag CUDA_COOPERATIVE_LAUNCH_MULTI_DEVICE_NO_POST_LAUNCH_SYNC. When this flag is specified, any subsequent work pushed in any of the specified streams will only wait for the kernel launched on the GPU corresponding to that stream to complete before it begins execution.

Calling cuLaunchCooperativeKernelMultiDevice() sets persistent function state that is the same as function state set through cuLaunchKernel API when called individually for each element in launchParamsList.

When kernels are launched via cuLaunchCooperativeKernelMultiDevice(), the previous block shape, shared size and parameter info associated with each CUDA_LAUNCH_PARAMS::function in launchParamsList is overwritten.

Note that to use cuLaunchCooperativeKernelMultiDevice(), the kernels must either have been compiled with toolchain version 3.2 or later so that it will contain kernel parameter information, or have no kernel parameters. If either of these conditions is not met, then cuLaunchCooperativeKernelMultiDevice() will return CUDA_ERROR_INVALID_IMAGE.

Note:

This function uses standard default stream semantics.
Note that this function may also return error codes from previous, asynchronous launches.

CUresult cuLaunchKernel ( CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void** kernelParams, void** extra )

Launches a CUDA function.

Parameters

f: - Kernel to launch
gridDimX: - Width of grid in blocks
gridDimY: - Height of grid in blocks
gridDimZ: - Depth of grid in blocks
blockDimX: - X dimension of each thread block
blockDimY: - Y dimension of each thread block
blockDimZ: - Z dimension of each thread block
sharedMemBytes: - Dynamic shared-memory size per thread block in bytes
hStream: - Stream identifier
kernelParams: - Array of pointers to kernel parameters
extra: - Extra options

Returns

Description

Invokes the kernel f on a gridDimX x gridDimY x gridDimZ grid of blocks. Each block contains blockDimX x blockDimY x blockDimZ threads.

sharedMemBytes sets the amount of dynamic shared memory that will be available to each thread block.

Kernel parameters to f can be specified in one of two ways:

1) Kernel parameters can be specified via kernelParams. If f has N parameters, then kernelParams needs to be an array of N pointers. Each of kernelParams[0] through kernelParams[N-1] must point to a region of memory from which the actual kernel parameter will be copied. The number of kernel parameters and their offsets and sizes do not need to be specified as that information is retrieved directly from the kernel's image.

2) Kernel parameters can also be packaged by the application into a single buffer that is passed in via the extra parameter. This places the burden on the application of knowing each kernel parameter's size and alignment/padding within the buffer. Here is an example of using the extra parameter in this manner:

‎    size_t argBufferSize;
          char argBuffer[256];
      
          // populate argBuffer and argBufferSize
      
          void *config[] = {
              CU_LAUNCH_PARAM_BUFFER_POINTER, argBuffer,
              CU_LAUNCH_PARAM_BUFFER_SIZE,    &argBufferSize,
              CU_LAUNCH_PARAM_END
          };
          status = cuLaunchKernel(f, gx, gy, gz, bx, by, bz, sh, s, NULL, config);

The extra parameter exists to allow cuLaunchKernel to take additional less commonly used arguments. extra specifies a list of names of extra settings and their corresponding values. Each extra setting name is immediately followed by the corresponding value. The list must be terminated with either NULL or CU_LAUNCH_PARAM_END.

CU_LAUNCH_PARAM_END, which indicates the end of the extra array;
CU_LAUNCH_PARAM_BUFFER_POINTER, which specifies that the next value in extra will be a pointer to a buffer containing all the kernel parameters for launching kernel f;
CU_LAUNCH_PARAM_BUFFER_SIZE, which specifies that the next value in extra will be a pointer to a size_t containing the size of the buffer specified with CU_LAUNCH_PARAM_BUFFER_POINTER;

The error CUDA_ERROR_INVALID_VALUE will be returned if kernel parameters are specified with both kernelParams and extra (i.e. both kernelParams and extra are non-NULL).

Calling cuLaunchKernel() sets persistent function state that is the same as function state set through the following deprecated APIs: cuFuncSetBlockShape(), cuFuncSetSharedSize(), cuParamSetSize(), cuParamSeti(), cuParamSetf(), cuParamSetv().

When the kernel f is launched via cuLaunchKernel(), the previous block shape, shared size and parameter info associated with f is overwritten.

Note that to use cuLaunchKernel(), the kernel f must either have been compiled with toolchain version 3.2 or later so that it will contain kernel parameter information, or have no kernel parameters. If either of these conditions is not met, then cuLaunchKernel() will return CUDA_ERROR_INVALID_IMAGE.

Note:

This function uses standard default stream semantics.
Note that this function may also return error codes from previous, asynchronous launches.