diff options
Diffstat (limited to 'compute/ARMComputeEx/src/core/CL/cl_kernels/reduce_operation.cl')
| -rw-r--r-- | compute/ARMComputeEx/src/core/CL/cl_kernels/reduce_operation.cl | 18 |
1 files changed, 11 insertions, 7 deletions
diff --git a/compute/ARMComputeEx/src/core/CL/cl_kernels/reduce_operation.cl b/compute/ARMComputeEx/src/core/CL/cl_kernels/reduce_operation.cl index 832ac1270..d0ef31b20 100644 --- a/compute/ARMComputeEx/src/core/CL/cl_kernels/reduce_operation.cl +++ b/compute/ARMComputeEx/src/core/CL/cl_kernels/reduce_operation.cl @@ -100,12 +100,14 @@ __kernel void reduce_min_max(TENSOR4D_DECLARATION(input), TENSOR4D_DECLARATION(o Tensor4D out = CONVERT_TO_TENSOR4D_STRUCT(output, DEPTH_OUT); int indices[4] = { - get_global_id(0), get_global_id(1), get_global_id(2) % DEPTH_OUT, - get_global_id(2) / DEPTH_OUT, + get_global_id(0), + get_global_id(1), + get_global_id(2) % DEPTH_OUT, + get_global_id(2) / DEPTH_OUT, }; DATA_TYPE value = - *((__global DATA_TYPE *)tensor4D_offset(&in, indices[0], indices[1], indices[2], indices[3])); + *((__global DATA_TYPE *)tensor4D_offset(&in, indices[0], indices[1], indices[2], indices[3])); for (int i = 1; i < dim; ++i) { indices[axis] = i; @@ -186,16 +188,18 @@ __kernel void reduce_sum_mean(TENSOR4D_DECLARATION(input), TENSOR4D_DECLARATION( Tensor4D out = CONVERT_TO_TENSOR4D_STRUCT(output, DEPTH_OUT); int indices[4] = { - get_global_id(0), get_global_id(1), get_global_id(2) % DEPTH_OUT, - get_global_id(2) / DEPTH_OUT, + get_global_id(0), + get_global_id(1), + get_global_id(2) % DEPTH_OUT, + get_global_id(2) / DEPTH_OUT, }; DATA_TYPE sum_value = (DATA_TYPE)0; for (int i = 0; i < dim; ++i) { indices[axis] = i; - sum_value += *( - (__global DATA_TYPE *)tensor4D_offset(&in, indices[0], indices[1], indices[2], indices[3])); + sum_value += + *((__global DATA_TYPE *)tensor4D_offset(&in, indices[0], indices[1], indices[2], indices[3])); } #if OP_CODE == 3 // REDUCE_SUM |