From ed5fe69b6612a5cf0dd52340f6781885d77afbc9 Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Thu, 9 Jul 2020 08:41:10 +0100 Subject: COMPMID-3326: Update heuristic for GEMMReshaped and GEMMReshapedOnlyRHS - Update the heuristic for Arm Mali-G76 (F32) in order to use the OpenCL image2d object on GEMM - Create utility function to validate the support for image2d Change-Id: I0913ac5f27fd07992b0ac188af753a2abeb034ca Signed-off-by: Gian Marco Iodice Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3559 Tested-by: Arm Jenkins Reviewed-by: Georgios Pinitas Comments-Addressed: Arm Jenkins --- .../CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp | 19 ++----------------- 1 file changed, 2 insertions(+), 17 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 8b4f930ea4..7d76ffd86c 100644 --- a/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp +++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.cpp @@ -26,6 +26,7 @@ #include "arm_compute/core/AccessWindowStatic.h" #include "arm_compute/core/CL/CLValidate.h" #include "arm_compute/core/CL/ICLTensor.h" +#include "arm_compute/core/CL/gemm/CLGEMMHelpers.h" #include "arm_compute/core/Error.h" #include "arm_compute/core/Helpers.h" #include "arm_compute/core/TensorInfo.h" @@ -65,23 +66,7 @@ Status validate_arguments(const ITensorInfo *input0, const ITensorInfo *input1, && (!gemm_info.broadcast_bias), "Bias addition only supported with broadcast mode in case the input or output has to be reinterpreted as 3D"); ARM_COMPUTE_RETURN_ERROR_ON_MSG(gemm_info.fp_mixed_precision, "Mixed precision not supported"); - - if(rhs_info.export_to_cl_image) - { - ARM_COMPUTE_RETURN_ERROR_ON_MSG((rhs_info.n0 == 2) || (rhs_info.n0 == 3), "Export to cl_image only supported with n0 = 4, 8 or 16"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG((rhs_info.k0 == 2) || (rhs_info.k0 == 3), "Export to cl_image only supported with k0 = 4, 8 or 16"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(input1->data_type() != DataType::F32, "Export to cl_image only supported with F32 data type"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(!image2d_from_buffer_supported(CLKernelLibrary::get().get_device()), "The extension cl_khr_image2d_from_buffer is not supported on the target platform"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device()) == 0, "Impossible to retrieve the cl_image pitch alignment"); - - // Check the width and height of the output tensor. - // Since we cannot create a 3d image from a buffer, the third dimension is collapsed with the second dimension - size_t max_image_w = CLKernelLibrary::get().get_device().getInfo(); - size_t max_image_h = CLKernelLibrary::get().get_device().getInfo(); - - ARM_COMPUTE_RETURN_ERROR_ON_MSG(input1->tensor_shape()[0] > max_image_w * 4, "Not supported width for cl_image"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(input1->tensor_shape()[1] * input1->tensor_shape()[2] > max_image_h, "Not supported height for cl_image"); - } + ARM_COMPUTE_RETURN_ON_ERROR(cl_gemm::validate_image2d_support_on_rhs(*input1, rhs_info)); const unsigned int m = gemm_info.m; const unsigned int n = gemm_info.n; -- cgit v1.2.1