From 85260d8c21e7209d4777150f436b336f85812dce Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Fri, 16 Dec 2022 15:34:27 +0000 Subject: Update direct conv2d kernel in dynamic fusion Resolves COMPMID-5780 Change-Id: I34c764cd1df652f8a938772924dc49baf6ac16db Signed-off-by: Gian Marco Iodice Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8825 Tested-by: Arm Jenkins Reviewed-by: Gunes Bayir Benchmark: Arm Jenkins Comments-Addressed: Arm Jenkins --- src/core/CL/cl_kernels/nhwc/direct_convolution.cl | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'src/core/CL/cl_kernels') diff --git a/src/core/CL/cl_kernels/nhwc/direct_convolution.cl b/src/core/CL/cl_kernels/nhwc/direct_convolution.cl index 8be8e00f0a..4693a1fbcd 100644 --- a/src/core/CL/cl_kernels/nhwc/direct_convolution.cl +++ b/src/core/CL/cl_kernels/nhwc/direct_convolution.cl @@ -188,6 +188,11 @@ __kernel void direct_convolution_nhwc( a[i].v = ZERO_VALUE; }) + LOOP_UNROLLING(int, i, 0, 1, N0, + { + b[i].v = ZERO_VALUE; + }) + // Load tile from the src tensor T_LOAD2D_INDIRECT(SRC_DATA_TYPE, M0, K0, SRC_TENSOR_TYPE, src, ck, src_stride_y, my, a); -- cgit v1.2.1