opencl: add functionality to query CL_KERNEL_WORK_GROUP_SIZE

dumbbell · Apr 6, 2012 · fbfe645 · fbfe645
1 parent b0b4cd7
commit fbfe645
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 0 deletions.
diff --git a/src/common/dlopencl.c b/src/common/dlopencl.c
@@ -126,6 +126,7 @@ int dt_dlopencl_init(const char *name, dt_dlopencl_t **ocl)
     success = success && dt_gmodule_symbol(module, "clGetEventProfilingInfo", (void (**)(void))&d->symbols->dt_clGetEventProfilingInfo); 
     success = success && dt_gmodule_symbol(module, "clGetKernelInfo", (void (**)(void))&d->symbols->dt_clGetKernelInfo); 
     success = success && dt_gmodule_symbol(module, "clEnqueueBarrier", (void (**)(void))&d->symbols->dt_clEnqueueBarrier); 
+    success = success && dt_gmodule_symbol(module, "clGetKernelWorkGroupInfo", (void (**)(void))&d->symbols->dt_clGetKernelWorkGroupInfo); 
 
     if (!success) dt_print(DT_DEBUG_OPENCL, "[opencl_init] could not load all required symbols from library\n");
     d->have_opencl = success;

diff --git a/src/common/opencl.c b/src/common/opencl.c
@@ -498,6 +498,15 @@ int dt_opencl_get_work_group_limits(const int dev, size_t *sizes, size_t *workgr
 }
 
 
+int dt_opencl_get_kernel_work_group_size(const int dev, const int kernel, size_t *kernelworkgroupsize)
+{
+  dt_opencl_t *cl = darktable.opencl;
+  if(!cl->inited || dev < 0) return -1;
+  if(kernel < 0 || kernel >= DT_OPENCL_MAX_KERNELS) return -1;
+
+  return (cl->dlocl->symbols->dt_clGetKernelWorkGroupInfo)(cl->dev[dev].kernel[kernel], cl->dev[dev].devid, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), kernelworkgroupsize, NULL);
+}
+
 
 int dt_opencl_set_kernel_arg(const int dev, const int kernel, const int num, const size_t size, const void *arg)
 {

diff --git a/src/common/opencl.h b/src/common/opencl.h
@@ -134,6 +134,9 @@ int dt_opencl_get_max_work_item_sizes(const int dev, size_t *sizes);
 /** return max size per dimension in sizes[3] and max total size in workgroupsize */
 int dt_opencl_get_work_group_limits(const int dev, size_t *sizes, size_t *workgroupsize, unsigned long *localmemsize);
 
+/** return max workgroup size for a specifc kernel */
+int dt_opencl_get_kernel_work_group_size(const int dev, const int kernel, size_t *kernelworkgroupsize);
+
 /** attach arg. */
 int dt_opencl_set_kernel_arg(const int dev, const int kernel, const int num, const size_t size, const void *arg);
 
@@ -262,6 +265,10 @@ static inline int dt_opencl_get_work_group_limits(const int dev, size_t *sizes,
 {
   return -1;
 }
+static inline int dt_opencl_get_kernel_work_group_size(const int dev, const int kernel, size_t *kernelworkgroupsize)
+{
+  return -1;
+}
 static inline int dt_opencl_set_kernel_arg(const int dev, const int kernel, const size_t size, const void *arg)
 {
   return -1;