potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 15:53:03 +04:00
Code Packages Releases Wiki Activity

dnn(ocl): fix automatic globalsize adjusting

Browse Source 
- if kernel code doesn't support that
This commit is contained in:
Alexander Alekhin
2021-09-04 16:27:51 +00:00
parent d11f0a709d
commit 5578ad5e14
9 changed files with 50 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/ocl.hpp
+18 -2
View File
@@ -562,7 +562,9 @@ public:
i = set(i, a6); i = set(i, a7); i = set(i, a8); i = set(i, a9); i = set(i, a10); i = set(i, a11);
i = set(i, a12); i = set(i, a13); i = set(i, a14); set(i, a15); return *this;
}
/** @brief Run the OpenCL kernel.
/** @brief Run the OpenCL kernel (globalsize value may be adjusted)
@param dims the work problem dimensions. It is the length of globalsize and localsize. It can be either 1, 2 or 3.
@param globalsize work items for each dimension. It is not the final globalsize passed to
OpenCL. Each dimension will be adjusted to the nearest integer divisible by the corresponding
@@ -571,12 +573,26 @@ public:
@param localsize work-group size for each dimension.
@param sync specify whether to wait for OpenCL computation to finish before return.
@param q command queue
@note Use run_() if your kernel code doesn't support adjusted globalsize.
*/
bool run(int dims, size_t globalsize[],
size_t localsize[], bool sync, const Queue& q=Queue());
/** @brief Run the OpenCL kernel
*
* @param dims the work problem dimensions. It is the length of globalsize and localsize. It can be either 1, 2 or 3.
* @param globalsize work items for each dimension. This value is passed to OpenCL without changes.
* @param localsize work-group size for each dimension.
* @param sync specify whether to wait for OpenCL computation to finish before return.
* @param q command queue
*/
bool run_(int dims, size_t globalsize[], size_t localsize[], bool sync, const Queue& q=Queue());
bool runTask(bool sync, const Queue& q=Queue());
/** @brief Similar to synchronized run() call with returning of kernel execution time
/** @brief Similar to synchronized run_() call with returning of kernel execution time
*
* Separate OpenCL command queue may be used (with CL_QUEUE_PROFILING_ENABLE)
* @return Execution time in nanoseconds or negative number on error
*/