diff options
Diffstat (limited to 'src')
| -rw-r--r-- | src/core/CL/cl_kernels/arg_min_max.cl | 9 | ||||
| -rw-r--r-- | src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp | 2 |
2 files changed, 7 insertions, 4 deletions
diff --git a/src/core/CL/cl_kernels/arg_min_max.cl b/src/core/CL/cl_kernels/arg_min_max.cl index 06dcc8ddde..104d30d8f3 100644 --- a/src/core/CL/cl_kernels/arg_min_max.cl +++ b/src/core/CL/cl_kernels/arg_min_max.cl @@ -254,10 +254,15 @@ __kernel void arg_min_max_x( barrier(CLK_LOCAL_MEM_FENCE); + // Looking for the next highest power of 2 (maximum value of lsize is 8) + unsigned int middle = lsize - 1; + middle |= middle >> 1; + middle |= middle >> 2; + middle += 1; // Perform parallel reduction - for(unsigned int i = lsize >> 1; i > 0; i >>= 1) + for(unsigned int i = middle; i > 0; i >>= 1) { - if(lid < i) + if( lid < i && lid + i < lsize) { DATA_TYPE tmp0 = *(src_in_row + local_results[lid]); DATA_TYPE tmp1 = *(src_in_row + local_results[lid + i]); diff --git a/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp b/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp index bdb7ab76d5..4845d60487 100644 --- a/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp +++ b/src/core/CL/kernels/CLArgMinMaxLayerKernel.cpp @@ -198,8 +198,6 @@ void CLArgMinMaxLayerKernel::run(const Window &window, cl::CommandQueue &queue) Window out_slice = out_window.first_slice_window_2D(); // Reshape window - const unsigned int border_width = ((in_slice.x().end() % vector_size) != 0) ? vector_size - in_slice.x().end() % vector_size : 0; - in_slice.set(Window::DimX, Window::Dimension(in_slice.x().start(), in_slice.x().end() + border_width, in_slice.x().step())); const unsigned int num_tensors = _prev_output != nullptr ? 3 : 2; // Set local sums buffer |