From b0c5037d94ba7073ccabb0ebaff54db320f184c4 Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Fri, 15 Mar 2019 10:13:05 +0000 Subject: COMPMID-2043: Add support for "dummy threads" in CLGEMMReshaped Change-Id: I89403b97503fbb99f6a32f5d62b8c535ab26a7be Signed-off-by: Gian Marco Iodice Reviewed-on: https://review.mlplatform.org/c/877 Reviewed-by: Michele Di Giorgio Tested-by: Arm Jenkins Reviewed-by: Georgios Pinitas Comments-Addressed: Arm Jenkins --- .../CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp') diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp index 6dc2a9e610..af06fecd00 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp @@ -161,7 +161,7 @@ std::pair validate_and_configure_window(ITensorInfo *input0, ITe } // namespace CLGEMMMatrixMultiplyReshapedOnlyRHSKernel::CLGEMMMatrixMultiplyReshapedOnlyRHSKernel() - : _input0(nullptr), _input1(nullptr), _output(nullptr), _slide_matrix_b(true), _reinterpret_input_as_3d(false), _reinterpret_output_as_3d(false) + : _input0(nullptr), _input1(nullptr), _output(nullptr), _slide_matrix_b(true), _reinterpret_input_as_3d(false), _reinterpret_output_as_3d(false), _use_dummy_work_items(false) { } @@ -177,6 +177,7 @@ void CLGEMMMatrixMultiplyReshapedOnlyRHSKernel::configure(const ICLTensor *input _output = output; _reinterpret_input_as_3d = gemm_info.reinterpret_input_as_3d(); _reinterpret_output_as_3d = (gemm_info.depth_output_gemm3d() != 0); + _use_dummy_work_items = preferred_dummy_work_items_support(CLKernelLibrary::get().get_device()); // In case both input and output have to be reinterpreted as 3D tensors, // force reinterpret_input_as_3d and reinterpret_output_as_3d to be false. @@ -202,13 +203,15 @@ void CLGEMMMatrixMultiplyReshapedOnlyRHSKernel::configure(const ICLTensor *input build_opts.add_option_if(_reinterpret_input_as_3d || _reinterpret_output_as_3d, "-DDEPTH_GEMM3D=" + support::cpp11::to_string(output->info()->dimension(2))); build_opts.add_option_if(!_slide_matrix_b, "-DMATRIX_B_DEPTH=" + support::cpp11::to_string(input1->info()->dimension(2))); build_opts.add_option_if(rhs_info.interleave, "-DRHS_INTERLEAVE"); + build_opts.add_option_if(_use_dummy_work_items, "-DDUMMY_WORK_ITEMS"); + build_opts.add_option("-DM=" + support::cpp11::to_string(gemm_info.m())); + build_opts.add_option("-DN=" + support::cpp11::to_string(gemm_info.n())); build_opts.add_option("-DK=" + support::cpp11::to_string(gemm_info.k())); build_opts.add_option("-DM0=" + support::cpp11::to_string(lhs_info.m0)); build_opts.add_option("-DN0=" + support::cpp11::to_string(rhs_info.n0)); build_opts.add_option("-DK0=" + support::cpp11::to_string(rhs_info.k0)); build_opts.add_option("-DH0=" + support::cpp11::to_string(rhs_info.h0)); - std::string kernel_name("gemm_mm_reshaped_only_rhs_"); kernel_name += rhs_info.transpose ? "t" : "nt"; @@ -308,7 +311,7 @@ void CLGEMMMatrixMultiplyReshapedOnlyRHSKernel::run(const Window &window, cl::Co _kernel.setArg(idx++, static_cast(_input0->info()->strides_in_bytes()[2])); _kernel.setArg(idx++, static_cast(_input1->info()->strides_in_bytes()[2])); _kernel.setArg(idx++, static_cast(_output->info()->strides_in_bytes()[2])); - enqueue(queue, *this, slice, lws_hint()); + enqueue(queue, *this, slice, lws_hint(), _use_dummy_work_items); } while(window.slide_window_slice_3D(slice)); } -- cgit v1.2.1