From c6ba86cdc3da8ab9599fff5483cedbe5b1422687 Mon Sep 17 00:00:00 2001 From: Cedric Nugteren Date: Tue, 7 May 2019 20:47:31 +0200 Subject: Enabled avc_motion_estimation extension for Intel subgroup shuffling --- src/kernels/level3/xgemm_part1.opencl | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'src/kernels/level3/xgemm_part1.opencl') diff --git a/src/kernels/level3/xgemm_part1.opencl b/src/kernels/level3/xgemm_part1.opencl index cc03696e..306280bc 100644 --- a/src/kernels/level3/xgemm_part1.opencl +++ b/src/kernels/level3/xgemm_part1.opencl @@ -125,8 +125,9 @@ R"( #define USE_SUBGROUP_SHUFFLING 0 // Optionally enables subgroup shuffling for Intel GPUs #endif -// Intel subgroups (https://www.khronos.org/registry/OpenCL/extensions/intel/cl_intel_subgroups.txt) +// Intel subgroups (https://www.khronos.org/registry/OpenCL/extensions/intel/cl_intel_subgroups.html) #if USE_SUBGROUP_SHUFFLING == 1 && SUBGROUP_SHUFFLING_INTEL == 1 + #pragma OPENCL EXTENSION cl_intel_device_side_avc_motion_estimation: enable #define SUBGROUP_SIZE 8 // Assumes subgroup size is always 8 on Intel GPUs #endif -- cgit v1.2.3