aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/gemm
diff options
context:
space:
mode:
authorGeorgios Pinitas <georgios.pinitas@arm.com>2021-04-22 21:13:21 +0100
committerGeorgios Pinitas <georgios.pinitas@arm.com>2021-05-18 14:48:39 +0000
commit856f66e6c61b77d03f754cd0fa8439891f0e4aca (patch)
treef9379cd0853ac407109e54c3d53b385ceee066c2 /src/core/CL/gemm
parent37f4b2ef1ea225a90ccb563fcb2c08f8fb0fb5d5 (diff)
downloadComputeLibrary-856f66e6c61b77d03f754cd0fa8439891f0e4aca.tar.gz
Port CLGEMM to memory injecting interface
Moves the following kernels: - CLGEMMMatrixMultiplyKernel - CLGEMMMatrixMultiplyNativeKernel - CLGEMMMatrixMultipluReshapedKernel - CLGEMMMatrixMultiplyReshapedOnlyRHSKernel Moves the following functions - CLGEMM Introduces facilities to easy handling of auxiliary temporary buffers under then new run interface. Such are: - CLAuxTensorHandler: That allows wrapping of workspace buffers memory to CLBuffer objects - Ability to inject TensorInfo to allocator without transferring ownership. This reduce the copy overhead if needed. Resolves: COMPMID-4188 Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com> Change-Id: I7055435d831b05b749b26302082e4ac45f26dfb0 Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5498 Tested-by: Arm Jenkins <bsgcomp@arm.com> Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com> Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/gemm')
-rw-r--r--src/core/CL/gemm/CLGEMMHelpers.cpp113
-rw-r--r--src/core/CL/gemm/CLGEMMHelpers.h92
-rw-r--r--src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.cpp240
-rw-r--r--src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.h56
-rw-r--r--src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.cpp67
-rw-r--r--src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.h51
-rw-r--r--src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.cpp162
-rw-r--r--src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.h53
-rw-r--r--src/core/CL/gemm/native/CLGEMMNativeKernelConfiguration.h65
-rw-r--r--src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.cpp350
-rw-r--r--src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.h58
-rw-r--r--src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.cpp532
-rw-r--r--src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.h55
-rw-r--r--src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfiguration.h63
-rw-r--r--src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.cpp512
-rw-r--r--src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.h61
-rw-r--r--src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.cpp564
-rw-r--r--src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.h55
-rw-r--r--src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfiguration.h63
19 files changed, 0 insertions, 3212 deletions
diff --git a/src/core/CL/gemm/CLGEMMHelpers.cpp b/src/core/CL/gemm/CLGEMMHelpers.cpp
deleted file mode 100644
index 61aa962198..0000000000
--- a/src/core/CL/gemm/CLGEMMHelpers.cpp
+++ /dev/null
@@ -1,113 +0,0 @@
-/*
- * Copyright (c) 2019-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/gemm/CLGEMMHelpers.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/CL/OpenCL.h"
-#include "arm_compute/core/ITensorInfo.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
-
-#include <utility>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-using namespace arm_compute::misc::shape_calculator;
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_lhs_rhs_info(unsigned int m, unsigned int n, unsigned int m0, unsigned int n0, unsigned int k0, unsigned int v0, unsigned int h0,
- bool lhs_interleave, bool rhs_interleave, bool lhs_transpose, bool rhs_transpose, bool export_to_cl_image)
-{
- ARM_COMPUTE_ERROR_ON(m0 == 0 || n0 == 0);
- v0 = std::max(std::min(static_cast<int>(m / m0), static_cast<int>(v0)), static_cast<int>(1));
- h0 = std::max(std::min(static_cast<int>(n / n0), static_cast<int>(h0)), static_cast<int>(1));
-
- const GEMMLHSMatrixInfo lhs_info(m0, k0, v0, lhs_transpose, lhs_interleave);
- const GEMMRHSMatrixInfo rhs_info(n0, k0, h0, rhs_transpose, rhs_interleave, export_to_cl_image);
-
- return std::make_pair(lhs_info, rhs_info);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> select_lhs_rhs_info(std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_img,
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_buf,
- unsigned int n, unsigned int k, unsigned int b, DataType data_type)
-{
- const TensorInfo tensor_rhs_info(TensorShape(n, k, b), 1, data_type);
- const TensorShape shape = compute_rhs_reshaped_shape(tensor_rhs_info, info_img.second);
- const TensorInfo tensor_reshaped_info(shape, 1, data_type);
-
- if(bool(validate_image2d_support_on_rhs(tensor_reshaped_info, info_img.second)))
- {
- return info_img;
- }
- else
- {
- return info_buf;
- }
-}
-
-void update_padding_for_cl_image(ITensorInfo *tensor)
-{
- constexpr unsigned int num_floats_per_pixel = 4;
-
- const unsigned int stride_y_in_elements = tensor->strides_in_bytes()[1] / tensor->element_size();
- const unsigned int pixel_alignment = get_cl_image_pitch_alignment(CLKernelLibrary::get().get_device());
-
- ARM_COMPUTE_ERROR_ON_MSG(pixel_alignment == 0, "Cannot retrieve cl_image pitch alignment");
- if(pixel_alignment == 0)
- {
- return;
- }
-
- const unsigned int row_pitch_alignment = pixel_alignment * num_floats_per_pixel;
- const unsigned int round_up_width = ((stride_y_in_elements + row_pitch_alignment - 1) / row_pitch_alignment) * row_pitch_alignment;
- const unsigned int padding = round_up_width - stride_y_in_elements;
-
- tensor->extend_padding(PaddingSize(0, padding, 0, 0));
-}
-
-Status validate_image2d_support_on_rhs(const ITensorInfo &tensor_reshaped_info, const GEMMRHSMatrixInfo &rhs_info)
-{
- 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_DATA_TYPE_NOT_IN(&tensor_reshaped_info, DataType::F32, DataType::F16);
- 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 on the second dimension
- const size_t max_image_w = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_WIDTH>();
- const size_t max_image_h = CLKernelLibrary::get().get_device().getInfo<CL_DEVICE_IMAGE2D_MAX_HEIGHT>();
-
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(tensor_reshaped_info.tensor_shape()[0] > max_image_w * 4, "Not supported width for cl_image");
- ARM_COMPUTE_RETURN_ERROR_ON_MSG(tensor_reshaped_info.tensor_shape()[1] * tensor_reshaped_info.tensor_shape()[2] > max_image_h, "Not supported height for cl_image");
- }
-
- return Status{};
-}
-} // namespace cl_gemm
-} // namespace arm_compute
diff --git a/src/core/CL/gemm/CLGEMMHelpers.h b/src/core/CL/gemm/CLGEMMHelpers.h
deleted file mode 100644
index 57624673c0..0000000000
--- a/src/core/CL/gemm/CLGEMMHelpers.h
+++ /dev/null
@@ -1,92 +0,0 @@
-/*
- * Copyright (c) 2019-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMHELPERS_H
-#define ARM_COMPUTE_CLGEMMHELPERS_H
-
-#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/Types.h"
-
-namespace arm_compute
-{
-class ITensorInfo;
-struct GEMMRHSMatrixInfo;
-
-namespace cl_gemm
-{
-/** Configure @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo
- *
- * @param[in] m Number of rows (M) in the LHS matrix not reshaped
- * @param[in] n Number of columns (N) in the RHS matrix not reshaped
- * @param[in] m0 Number of rows processed by each thread/work-item
- * @param[in] n0 Number of columns processed by each thread/work-item
- * @param[in] k0 Number of inner accumulation performed by each thread/work-item
- * @param[in] v0 Number of vertical blocks of size (m0xk0) stored on the same output row
- * @param[in] h0 Number of horizontal blocks of size (k0xn0) stored on the same output row
- * @param[in] lhs_interleave True if the v0 (m0xk0) blocks have to be interleaved in the output row
- * @param[in] rhs_interleave True if the h0 (k0xn0) blocks have to be interleaved in the output row
- * @param[in] lhs_transpose True if the (m0xk0) block has to be transposed before been stored
- * @param[in] rhs_transpose True if the (k0xn0) block has to be transposed before been stored
- * @param[in] export_to_cl_image (Optional) True if the RHS reshaped matrix has to be exported to cl_image
- *
- * @return @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo
- */
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_lhs_rhs_info(unsigned int m, unsigned int n, unsigned int m0, unsigned int n0, unsigned int k0, unsigned int v0, unsigned int h0,
- bool lhs_interleave, bool rhs_interleave, bool lhs_transpose, bool rhs_transpose, bool export_to_cl_image = false);
-
-/** Select @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo
- *
- * This function accepts two pairs of GEMMLHSMatrixInfo/GEMMRHSMatrixInfo where only the first is with cl_image2d support,
- * and selects the valid one validating the GEMMRHSMatrixInfo. If the validation passes, the functions will return
- * the first GEMMLHSMatrixInfo/GEMMRHSMatrixInfo pair with cl_image2d support.
- *
- * @param[in] info_img GEMMLHSMatrixInfo/GEMMRHSMatrixInfo with cl_image2d support
- * @param[in] info_buf GEMMLHSMatrixInfo/GEMMRHSMatrixInfo to fall-back if cl_image2d cannot be used
- * @param[in] n Number of columns (N) in the RHS matrix not reshaped
- * @param[in] k Number of rows (K) in the RHS matrix not reshaped
- * @param[in] b Batch size
- * @param[in] data_type Data type
- *
- * @return @ref GEMMLHSMatrixInfo and @ref GEMMRHSMatrixInfo
- */
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> select_lhs_rhs_info(std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_img,
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_buf,
- unsigned int n, unsigned int k, unsigned int b, DataType data_type);
-
-/** Update padding required to export the OpenCL buffer to OpenCL image2d
- *
- * @param[in,out] tensor ITensorInfo of the tensor required to be exported to OpenCL image2d
- */
-void update_padding_for_cl_image(ITensorInfo *tensor);
-
-/** Utility function to validate the image2d OpenCL object support on the RHS reshaped matrix
- *
- * @param[in] tensor_reshaped_info TensorInfo for the RHS reshaped matrix
- * @param[in] rhs_info @ref GEMMRHSMatrixInfo
- *
- * @return Status reporting if we can use the image2d OpenCL object on the RHS reshaped matrix
- */
-Status validate_image2d_support_on_rhs(const ITensorInfo &tensor_reshaped_info, const GEMMRHSMatrixInfo &rhs_info);
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMHELPERS_H */
diff --git a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.cpp b/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.cpp
deleted file mode 100644
index 52023dd835..0000000000
--- a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.cpp
+++ /dev/null
@@ -1,240 +0,0 @@
-/*
- * Copyright (c) 2019-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/GPUTarget.h"
-#include "src/core/CL/gemm/CLGEMMHelpers.h"
-
-#include <utility>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-CLGEMMDefaultConfigNativeBifrost::CLGEMMDefaultConfigNativeBifrost(GPUTarget gpu)
- : ICLGEMMKernelConfiguration(gpu)
-{
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeBifrost::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
-{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (CLGEMMDefaultConfigNativeBifrost::*)(unsigned int m, unsigned int n, unsigned int k,
- unsigned int b);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G71(&CLGEMMDefaultConfigNativeBifrost::configure_G71_f32,
- &CLGEMMDefaultConfigNativeBifrost::configure_G71_f32, // We use the F32 heuristic
- &CLGEMMDefaultConfigNativeBifrost::configure_G71_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G76(&CLGEMMDefaultConfigNativeBifrost::configure_G76_f32,
- &CLGEMMDefaultConfigNativeBifrost::configure_G76_f32, // We use the F32 heuristic
- &CLGEMMDefaultConfigNativeBifrost::configure_G76_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G7x(&CLGEMMDefaultConfigNativeBifrost::configure_default_f32,
- &CLGEMMDefaultConfigNativeBifrost::configure_default_f32, // We use the F32 heuristic
- &CLGEMMDefaultConfigNativeBifrost::configure_default_u8);
-
- ConfigurationFunctionExecutorPtr func = nullptr;
-
- switch(_target)
- {
- case GPUTarget::G76:
- func = configs_G76.get_function(data_type);
- break;
- case GPUTarget::G71:
- func = configs_G71.get_function(data_type);
- break;
- default:
- func = configs_G7x.get_function(data_type);
- break;
- }
-
- ARM_COMPUTE_ERROR_ON_MSG(func == nullptr, "Data type not support for GEMM");
- return (this->*func)(m, n, k, b);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeBifrost::configure_G71_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- if(n < 2048)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 1, false, false, false, false);
- }
- else if(n >= 2048 && n < 8192)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 4, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 4, 1, 1, false, false, false, false);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 4, 2, 1, 1, false, false, false, false);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeBifrost::configure_G71_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(dot8_supported(CLKernelLibrary::get().get_device()))
- {
- if(m == 1)
- {
- if(n < 2048)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 1, false, false, false, false);
- }
- else if(n >= 2048 && n < 16384)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 16, 1, 1, false, false, false, false);
- }
- }
- else
- {
- if(m < 64)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 16, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 2, 16, 1, 1, false, false, false, false);
- }
- }
- }
- else
- {
- if(m == 1)
- {
- if(n < 8192)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 16, 1, 1, false, false, false, false);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 16, 1, 1, false, false, false, false);
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeBifrost::configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- if(n > 4196)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 2, 1, 1, false, false, false, false);
- }
- else
- {
- if(k < 2048)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 2, 1, 1, false, false, false, false);
- }
- else if(k >= 2048 && k < 16384)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, 1, false, false, false, false);
- }
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 2, 1, 1, false, false, false, false);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeBifrost::configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- if(n < 2048)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 1, false, false, false, false);
- }
- else if(n >= 2048 && n < 16384)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 16, 1, 1, false, false, false, false);
- }
- }
- else
- {
- if(m < 64)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 16, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 2, 16, 1, 1, false, false, false, false);
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeBifrost::configure_default_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- return configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 1, false, false, false, false);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeBifrost::configure_default_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- return configure_lhs_rhs_info(m, n, 5, 2, 16, 1, 1, false, false, false, false);
-}
-} // namespace cl_gemm
-} // namespace arm_compute \ No newline at end of file
diff --git a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.h b/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.h
deleted file mode 100644
index 78d47a8195..0000000000
--- a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.h
+++ /dev/null
@@ -1,56 +0,0 @@
-/*
- * Copyright (c) 2019-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEBIFROST_H
-#define ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEBIFROST_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** Bifrost based OpenCL GEMMNative configuration */
-class CLGEMMDefaultConfigNativeBifrost final : public ICLGEMMKernelConfiguration
-{
-public:
- /** Constructor
- *
- * @param[in] gpu GPU target
- */
- CLGEMMDefaultConfigNativeBifrost(GPUTarget gpu);
-
- // Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
-
-private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G71_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G71_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_default_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_default_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEBIFROST_H */
diff --git a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.cpp b/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.cpp
deleted file mode 100644
index cf9bb1828f..0000000000
--- a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.cpp
+++ /dev/null
@@ -1,67 +0,0 @@
-/*
- * Copyright (c) 2020-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/GPUTarget.h"
-#include "src/core/CL/gemm/CLGEMMHelpers.h"
-
-#include <utility>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-CLGEMMDefaultConfigNativeMidgard::CLGEMMDefaultConfigNativeMidgard(GPUTarget gpu)
- : ICLGEMMKernelConfiguration(gpu)
-{
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeMidgard::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
-{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (CLGEMMDefaultConfigNativeMidgard::*)(unsigned int m, unsigned int n, unsigned int k,
- unsigned int b);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_default(nullptr,
- nullptr,
- &CLGEMMDefaultConfigNativeMidgard::default_q8);
-
- auto func = configs_default.get_function(data_type);
- ARM_COMPUTE_ERROR_ON_MSG(func == nullptr, "Data type not support for GEMM");
- return (this->*func)(m, n, k, b);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeMidgard::default_q8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- const unsigned int m0 = std::min(m, static_cast<unsigned int>(4));
- const unsigned int n0 = std::min(n, static_cast<unsigned int>(4));
-
- return configure_lhs_rhs_info(m, n, m0, n0, 2, 1, 1, false, false, false, false);
-}
-} // namespace cl_gemm
-} // namespace arm_compute \ No newline at end of file
diff --git a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.h b/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.h
deleted file mode 100644
index 40c91d42b1..0000000000
--- a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.h
+++ /dev/null
@@ -1,51 +0,0 @@
-/*
- * Copyright (c) 2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEMIDGARD_H
-#define ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEMIDGARD_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** Midgard based OpenCL GEMMNative configuration */
-class CLGEMMDefaultConfigNativeMidgard final : public ICLGEMMKernelConfiguration
-{
-public:
- /** Constructor
- *
- * @param[in] gpu GPU target
- */
- CLGEMMDefaultConfigNativeMidgard(GPUTarget gpu);
-
- // Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
-
-private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> default_q8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEMIDGARD_H */
diff --git a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.cpp b/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.cpp
deleted file mode 100644
index 3b55be747f..0000000000
--- a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.cpp
+++ /dev/null
@@ -1,162 +0,0 @@
-/*
- * Copyright (c) 2020-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/GPUTarget.h"
-#include "src/core/CL/gemm/CLGEMMHelpers.h"
-
-#include <utility>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-CLGEMMDefaultConfigNativeValhall::CLGEMMDefaultConfigNativeValhall(GPUTarget gpu)
- : ICLGEMMKernelConfiguration(gpu)
-{
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeValhall::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
-{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (CLGEMMDefaultConfigNativeValhall::*)(unsigned int m, unsigned int n, unsigned int k,
- unsigned int b);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_default(&CLGEMMDefaultConfigNativeValhall::configure_G77_f32,
- &CLGEMMDefaultConfigNativeValhall::configure_G77_f16,
- &CLGEMMDefaultConfigNativeValhall::configure_G77_u8);
-
- auto func = configs_default.get_function(data_type);
- ARM_COMPUTE_ERROR_ON_MSG(func == nullptr, "Data type not support for GEMM");
- return (this->*func)(m, n, k, b);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeValhall::configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- if(n < 2048)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 1, false, false, false, false);
- }
- else if(n >= 2048 && n < 8192)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 4, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 4, 1, 1, false, false, false, false);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 4, 2, 1, 1, false, false, false, false);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeValhall::configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- if(n < 2048)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 1, false, false, false, false);
- }
- else if(n >= 2048 && n < 8192)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 4, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 4, 1, 1, false, false, false, false);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 8, 2, 1, 1, false, false, false, false);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigNativeValhall::configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(dot8_supported(CLKernelLibrary::get().get_device()))
- {
- if(m == 1)
- {
- if(n < 2048)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 1, false, false, false, false);
- }
- else if(n >= 2048 && n < 16384)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 16, 1, 1, false, false, false, false);
- }
- }
- else
- {
- if(m < 64)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 16, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 2, 16, 1, 1, false, false, false, false);
- }
- }
- }
- else
- {
- if(m == 1)
- {
- if(n < 8192)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 1, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 16, 1, 1, false, false, false, false);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 16, 1, 1, false, false, false, false);
- }
- }
-}
-} // namespace cl_gemm
-} // namespace arm_compute \ No newline at end of file
diff --git a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.h b/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.h
deleted file mode 100644
index 08d2d57a3e..0000000000
--- a/src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.h
+++ /dev/null
@@ -1,53 +0,0 @@
-/*
- * Copyright (c) 2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEVALHALL_H
-#define ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEVALHALL_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** Valhall based OpenCL GEMMNative configuration */
-class CLGEMMDefaultConfigNativeValhall final : public ICLGEMMKernelConfiguration
-{
-public:
- /** Constructor
- *
- * @param[in] gpu GPU target
- */
- CLGEMMDefaultConfigNativeValhall(GPUTarget gpu);
-
- // Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
-
-private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMDEFAULTCONFIGNATIVEVALHALL_H */
diff --git a/src/core/CL/gemm/native/CLGEMMNativeKernelConfiguration.h b/src/core/CL/gemm/native/CLGEMMNativeKernelConfiguration.h
deleted file mode 100644
index 39a534e817..0000000000
--- a/src/core/CL/gemm/native/CLGEMMNativeKernelConfiguration.h
+++ /dev/null
@@ -1,65 +0,0 @@
-/*
- * Copyright (c) 2019-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMNATIVEKERNELCONFIGURATION_H
-#define ARM_COMPUTE_CLGEMMNATIVEKERNELCONFIGURATION_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-#include "src/core/CL/gemm/native/CLGEMMDefaultConfigNativeBifrost.h"
-#include "src/core/CL/gemm/native/CLGEMMDefaultConfigNativeMidgard.h"
-#include "src/core/CL/gemm/native/CLGEMMDefaultConfigNativeValhall.h"
-
-#include <memory>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** CLGEMMNative factory class */
-class CLGEMMNativeKernelConfigurationFactory final
-{
-public:
- /** Static method to construct CLGEMMNative kernel object accordingly with the GPU target
- *
- * @param[in] gpu GPU target
- *
- * @return CLGEMMNative kernel configuration class
- */
- static std::unique_ptr<ICLGEMMKernelConfiguration> create(GPUTarget gpu)
- {
- switch(get_arch_from_target(gpu))
- {
- case GPUTarget::MIDGARD:
- return std::make_unique<CLGEMMDefaultConfigNativeMidgard>(gpu);
- case GPUTarget::BIFROST:
- return std::make_unique<CLGEMMDefaultConfigNativeBifrost>(gpu);
- case GPUTarget::VALHALL:
- return std::make_unique<CLGEMMDefaultConfigNativeValhall>(gpu);
- default:
- ARM_COMPUTE_ERROR("Not supported GPU target");
- }
- }
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMNATIVEKERNELCONFIGURATION_H */
diff --git a/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.cpp b/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.cpp
deleted file mode 100644
index 5877ab96e7..0000000000
--- a/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.cpp
+++ /dev/null
@@ -1,350 +0,0 @@
-/*
- * Copyright (c) 2019-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/GPUTarget.h"
-#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/TensorShape.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
-#include "src/core/CL/gemm/CLGEMMHelpers.h"
-
-#include <utility>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-using namespace arm_compute::misc::shape_calculator;
-
-CLGEMMDefaultConfigReshapedBifrost::CLGEMMDefaultConfigReshapedBifrost(GPUTarget gpu)
- : ICLGEMMKernelConfiguration(gpu)
-{
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
-{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (CLGEMMDefaultConfigReshapedBifrost::*)(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G7x(&CLGEMMDefaultConfigReshapedBifrost::configure_G7x_f32,
- &CLGEMMDefaultConfigReshapedBifrost::configure_G7x_f16,
- &CLGEMMDefaultConfigReshapedBifrost::configure_G7x_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G52(&CLGEMMDefaultConfigReshapedBifrost::configure_G52_f32,
- &CLGEMMDefaultConfigReshapedBifrost::configure_G52_f16,
- &CLGEMMDefaultConfigReshapedBifrost::configure_G7x_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G76(&CLGEMMDefaultConfigReshapedBifrost::configure_G76_f32,
- &CLGEMMDefaultConfigReshapedBifrost::configure_G76_f16,
- &CLGEMMDefaultConfigReshapedBifrost::configure_G76_u8);
-
- ConfigurationFunctionExecutorPtr func = nullptr;
-
- switch(_target)
- {
- case GPUTarget::G76:
- func = configs_G76.get_function(data_type);
- break;
- case GPUTarget::G52:
- func = configs_G52.get_function(data_type);
- break;
- default:
- func = configs_G7x.get_function(data_type);
- break;
- }
-
- ARM_COMPUTE_ERROR_ON_MSG(func == nullptr, "Data type not support for GEMM");
- return (this->*func)(m, n, k, b);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(n <= 4)
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 8, 16, 16, true, false, false, true);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 4, 4, 2, 16, false, true, false, true);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure_G7x_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(n <= 4)
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 8, 8, 2, true, true, true, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 8, 4, 4, 2, true, true, true, false);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(dot8_supported(CLKernelLibrary::get().get_device()))
- {
- if(n <= 4)
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 16, 2, 2, true, false, false, true);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 16, 2, 2, true, false, false, true);
- }
- }
- else
- {
- if(n <= 4)
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 8, 2, 2, true, false, false, true);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 6, 4, 4, 2, 2, true, true, false, true);
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure_G52_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
- const float r_nk = static_cast<float>(n) / static_cast<float>(k);
-
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- if(workload <= 274.4000f)
- {
- if(r_nk <= 0.7461f)
- {
- if(r_mn <= 21.1667f)
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 4, 4, 4, false, true, true, false, false);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- else
- {
- if(r_mk <= 17.3926f)
- {
- if(workload <= 542.4000f)
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- else
- {
- if(r_nk <= 0.5463f)
- {
- if(workload <= 11767.6001f)
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, true, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, true, true, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure_G52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
-
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
-
- if(workload <= 323.4000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 8, 4, 8, false, false, false, true, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 8, 4, 2, 2, true, true, true, false, false);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- // Get lhs_info/rhs_info in case of OpenCL buffer
- if(n <= 4)
- {
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 2, 8, 16, 16, true, false, false, true);
- }
- else
- {
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 2, 8, 16, false, false, false, true);
- }
-
- // Get lhs_info/rhs_info in case of OpenCL image
- // Condition on the GPU workload
- if((m / 4) * (n / 4) >= 2560)
- {
- // Big workload
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 8, true, true, true, false, true);
- }
- else
- {
- // Small workload
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 1, true, true, true, false, true);
- }
-
- const TensorInfo tensor_rhs_info(TensorShape(n, k, b), 1, DataType::F32);
- const TensorShape shape = compute_rhs_reshaped_shape(tensor_rhs_info, rhs_info_img);
- const TensorInfo tensor_reshaped_info(shape, 1, DataType::F32);
-
- // In case of vector by matrix with few work-items, we use the OpenCL buffer rather than the OpenCL image2d
- const bool use_cl_image2d = (n <= 4) ? false : true;
-
- if(bool(validate_image2d_support_on_rhs(tensor_reshaped_info, rhs_info_img)) && use_cl_image2d)
- {
- return std::make_pair(lhs_info_img, rhs_info_img);
- }
- else
- {
- return std::make_pair(lhs_info_buf, rhs_info_buf);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
-
- if(workload <= 1595.2000f)
- {
- if(r_mk <= 2.1044f)
- {
- if(workload <= 870.4000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 2, true, false, true, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 4, 2, 2, false, false, true, false, false);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 4, 2, 2, false, false, true, false, false);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 8, 4, 4, 2, true, true, true, false, false);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedBifrost::configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(n <= 4)
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 16, 4, 1, false, false, false, true);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 16, 2, 2, false, true, false, true);
- }
-}
-} // namespace cl_gemm
-} // namespace arm_compute
diff --git a/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.h b/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.h
deleted file mode 100644
index 814b831b69..0000000000
--- a/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.h
+++ /dev/null
@@ -1,58 +0,0 @@
-/*
- * Copyright (c) 2019-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDBIFROST_H
-#define ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDBIFROST_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** Bifrost based OpenCL GEMMReshaped configuration */
-class CLGEMMDefaultConfigReshapedBifrost final : public ICLGEMMKernelConfiguration
-{
-public:
- /** Constructor
- *
- * @param[in] gpu GPU target
- */
- CLGEMMDefaultConfigReshapedBifrost(GPUTarget gpu);
-
- // Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
-
-private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G52_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDBIFROST_H */
diff --git a/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.cpp b/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.cpp
deleted file mode 100644
index b07092ab83..0000000000
--- a/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.cpp
+++ /dev/null
@@ -1,532 +0,0 @@
-/*
- * Copyright (c) 2020-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/GPUTarget.h"
-#include "src/core/CL/gemm/CLGEMMHelpers.h"
-
-#include <utility>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-CLGEMMDefaultConfigReshapedValhall::CLGEMMDefaultConfigReshapedValhall(GPUTarget gpu)
- : ICLGEMMKernelConfiguration(gpu)
-{
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedValhall::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
-{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (CLGEMMDefaultConfigReshapedValhall::*)(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G77(&CLGEMMDefaultConfigReshapedValhall::configure_G77_f32,
- &CLGEMMDefaultConfigReshapedValhall::configure_G77_f16,
- &CLGEMMDefaultConfigReshapedValhall::configure_G77_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G78(&CLGEMMDefaultConfigReshapedValhall::configure_G78_f32,
- &CLGEMMDefaultConfigReshapedValhall::configure_G78_f16,
- &CLGEMMDefaultConfigReshapedValhall::configure_G77_u8);
-
- ConfigurationFunctionExecutorPtr func = nullptr;
-
- switch(_target)
- {
- case GPUTarget::G78:
- func = configs_G78.get_function(data_type);
- break;
- case GPUTarget::G77:
- default:
- func = configs_G77.get_function(data_type);
- break;
- }
-
- ARM_COMPUTE_ERROR_ON_MSG(func == nullptr, "Data type not support for GEMM");
- return (this->*func)(m, n, k, b);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedValhall::configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(n <= 4)
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 8, 16, 16, 1, 0, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 4, 4, 2, 16, 0, 1, 0, 1);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedValhall::configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
- const float r_nk = static_cast<float>(n) / static_cast<float>(k);
-
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 0);
-
- if(r_mk <= 0.11824845522642136)
- {
- if(workload <= 880.0)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 4, 0, 0, 1, 0, 0);
- }
- else
- {
- if(r_nk <= 0.42521367967128754)
- {
- if(workload <= 1726.4000244140625)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 0);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 0, 1, 1, 0, 1);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- }
- else
- {
- if(workload <= 1241.6000366210938)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 4, 0, 0, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 0);
- }
- }
- }
- }
- else
- {
- if(workload <= 11404.7998046875)
- {
- if(r_mk <= 1.0126488208770752)
- {
- if(r_mn <= 2.545312523841858)
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 0, 1, 1, 0, 1);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 4, 0, 0, 1, 0, 0);
- }
- }
- else
- {
- if(workload <= 2881.199951171875)
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, 0, 0, 1, 0, 1);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 0, 1, 1, 0, 1);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- }
- }
- else
- {
- if(r_nk <= 0.5765306055545807)
- {
- if(r_mn <= 6.010416746139526)
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 0, 1, 1, 0, 1);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 1, 0, 1, 0, 1);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, 1, 0, 1, 0, 1);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedValhall::configure_G78_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
- const float r_nk = static_cast<float>(n) / static_cast<float>(k);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
-
- if(workload <= 1288.0000f)
- {
- if(workload <= 505.6000f)
- {
- if(r_mn <= 0.4466f)
- {
- if(r_nk <= 0.2384f)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 4, 4, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 4, 2, 2, 0, 0, 1, 0, 0);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 4, 2, 2, 0, 0, 1, 0, 0);
- }
- }
- else
- {
- if(r_mn <= 0.2250f)
- {
- if(r_mn <= 0.1599f)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 4, 4, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- }
- else
- {
- if(r_mk <= 0.7609f)
- {
- if(r_mn <= 2.5453f)
- {
- if(workload <= 1089.6000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 4, 4, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 2, 4, 0, 0, 1, 0, 1);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 16, 4, 4, 0, 0, 1, 0, 1);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 4, 4, 0, 0, 1, 0, 1);
- }
- }
- }
- }
- else
- {
- if(workload <= 5434.4001f)
- {
- if(workload <= 1603.2000f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- if(r_nk <= 0.6192f)
- {
- if(r_mn <= 16.1016f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- if(workload <= 2750.0000f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- if(r_mk <= 6.3151f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 0, 1, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- }
- }
- }
- else
- {
- if(r_mk <= 0.0387f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 1);
- }
- else
- {
- if(r_mk <= 2.5859f)
- {
- if(r_mk <= 0.2734f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- }
- }
- }
- }
- else
- {
- if(r_mk <= 25.7500f)
- {
- if(r_mk <= 0.3615f)
- {
- if(r_mn <= 0.0913f)
- {
- if(r_mk <= 0.0683f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 4, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 4, 4, 0, 0, 1, 0, 1);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- }
- else
- {
- if(workload <= 11174.3999f)
- {
- if(r_mk <= 0.8047f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- if(workload <= 7185.5999f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 4, 2, 0, 0, 1, 0, 1);
- }
- }
- }
- else
- {
- if(workload <= 17917.5000f)
- {
- if(r_mk <= 1.5078f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 1, 0, 1);
- }
- }
- else
- {
- if(workload <= 34449.6016f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 4, 0, 0, 1, 0, 1);
- }
- }
- }
- }
- }
- else
- {
- if(r_mk <= 331.1111f)
- {
- if(workload <= 53397.5996f)
- {
- if(r_mn <= 57.8063f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 0, 1, 1);
- }
- }
- else
- {
- if(r_nk <= 0.9211f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 4, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 0, 1, 1);
- }
- }
- }
- else
- {
- if(workload <= 38070.4004f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, 0, 0, 0, 1, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- }
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedValhall::configure_G78_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float r_nk = static_cast<float>(n) / static_cast<float>(k);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
-
- if(workload <= 801.6000f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 1, 1, 0, 0, 1, 0, 1);
- }
- else
- {
- if(r_mn <= 0.1211f)
- {
- if(workload <= 3296.0000f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- if(r_nk <= 1.0625f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 4, 0, 0, 1, 0, 1);
- }
- }
- }
- else
- {
- if(workload <= 5068.8000f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 1, 1, 0, 0, 1, 0, 1);
- }
- else
- {
- if(r_nk <= 0.2361f)
- {
- if(workload <= 12630.0000f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 1, 1, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 1, 0, 0, 1, 0, 1);
- }
- }
- else
- {
- if(workload <= 178790.3984f)
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 2, 2, 0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 8, 4, 4, 1, 1, 0, 0, 1, 0, 1);
- }
- }
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedValhall::configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(n <= 4)
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 16, 4, 1, 0, 0, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 16, 2, 2, 0, 1, 0, 1);
- }
-}
-} // namespace cl_gemm
-} // namespace arm_compute
diff --git a/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.h b/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.h
deleted file mode 100644
index 52b83b09b6..0000000000
--- a/src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.h
+++ /dev/null
@@ -1,55 +0,0 @@
-/*
- * Copyright (c) 2020-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDVALHALL_H
-#define ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDVALHALL_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** Valhall based OpenCL GEMMReshaped configuration */
-class CLGEMMDefaultConfigReshapedValhall final : public ICLGEMMKernelConfiguration
-{
-public:
- /** Constructor
- *
- * @param[in] gpu GPU target
- */
- CLGEMMDefaultConfigReshapedValhall(GPUTarget gpu);
-
- // Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
-
-private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G78_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G78_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDVALHALL_H */
diff --git a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfiguration.h b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfiguration.h
deleted file mode 100644
index de60698a91..0000000000
--- a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfiguration.h
+++ /dev/null
@@ -1,63 +0,0 @@
-/*
- * Copyright (c) 2019-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMRESHAPEDKERNELCONFIGURATION_H
-#define ARM_COMPUTE_CLGEMMRESHAPEDKERNELCONFIGURATION_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-#include "src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedBifrost.h"
-#include "src/core/CL/gemm/reshaped/CLGEMMDefaultConfigReshapedValhall.h"
-
-#include <memory>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** CLGEMMReshaped factory class */
-class CLGEMMReshapedKernelConfigurationFactory final
-{
-public:
- /** Static method to call the CLGEMMReshaped kernel configuration class accordingly with the GPU target
- *
- * @param[in] gpu GPU target
- *
- * @return CLGEMMReshaped kernel configuration class
- */
- static std::unique_ptr<ICLGEMMKernelConfiguration> create(GPUTarget gpu)
- {
- switch(get_arch_from_target(gpu))
- {
- case GPUTarget::MIDGARD:
- case GPUTarget::BIFROST:
- return std::make_unique<CLGEMMDefaultConfigReshapedBifrost>(gpu);
- case GPUTarget::VALHALL:
- return std::make_unique<CLGEMMDefaultConfigReshapedValhall>(gpu);
- default:
- ARM_COMPUTE_ERROR("Not supported GPU target");
- }
- }
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMRESHAPEDKERNELCONFIGURATION_H */
diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.cpp b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.cpp
deleted file mode 100644
index 3645a0e141..0000000000
--- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.cpp
+++ /dev/null
@@ -1,512 +0,0 @@
-/*
- * Copyright (c) 2019-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/GPUTarget.h"
-#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/TensorShape.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
-#include "src/core/CL/gemm/CLGEMMHelpers.h"
-
-#include <utility>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-using namespace arm_compute::misc::shape_calculator;
-
-CLGEMMDefaultConfigReshapedRHSOnlyBifrost::CLGEMMDefaultConfigReshapedRHSOnlyBifrost(GPUTarget gpu)
- : ICLGEMMKernelConfiguration(gpu)
-{
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
-{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (CLGEMMDefaultConfigReshapedRHSOnlyBifrost::*)(unsigned int m, unsigned int n, unsigned int k,
- unsigned int b);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G51(&CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G51_f32,
- &CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G51_f16,
- &CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G51_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G52(&CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G52_f32,
- &CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G52_f16,
- &CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G7x_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G76(&CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G76_f32,
- &CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G76_f16,
- &CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G76_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G7x(&CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G7x_f32,
- &CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G7x_f16,
- &CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G7x_u8);
-
- ConfigurationFunctionExecutorPtr func = nullptr;
-
- switch(_target)
- {
- case GPUTarget::G76:
- func = configs_G76.get_function(data_type);
- break;
- case GPUTarget::G51:
- func = configs_G51.get_function(data_type);
- break;
- case GPUTarget::G52:
- func = configs_G52.get_function(data_type);
- break;
- default:
- func = configs_G7x.get_function(data_type);
- break;
- }
-
- ARM_COMPUTE_ERROR_ON_MSG(func == nullptr, "Data type not support for GEMM");
- return (this->*func)(m, n, k, b);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- if(n <= 2548)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 4, false, true, false, true, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 8, false, true, false, true, false);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 4, false, true, false, true);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- const bool is_workload_big = ((m * n * b) / 16) >= 2048;
-
- if(m == 1)
- {
- if(n >= 8192)
- {
- const unsigned int h0 = std::max(n / 4, 1U);
- return configure_lhs_rhs_info(m, n, 1, 4, 8, 1, h0, false, true, false, true, false);
- }
- else
- {
- const unsigned int h0 = std::max(n / 2, 1U);
- if(n <= 204)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, h0, false, true, false, true, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, h0, false, true, false, true, false);
- }
- }
- }
- else
- {
- const int h0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(16)), static_cast<int>(1));
- if(is_workload_big)
- {
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 1, h0, false, true, false, true);
- }
- else
- {
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 2, 4, 8, 1, h0, false, true, false, true);
- }
- }
-
- // Get lhs_info/rhs_info in case of OpenCL image
- const int h0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(16)), static_cast<int>(1));
- if(is_workload_big)
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 1, h0, false, true, false, false, true);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 2, 4, 8, 1, h0, false, true, false, true, true);
- }
-
- const TensorInfo tensor_rhs_info(TensorShape(n, k, b), 1, DataType::F32);
- const TensorShape shape = compute_rhs_reshaped_shape(tensor_rhs_info, rhs_info_img);
- const TensorInfo tensor_reshaped_info(shape, 1, DataType::F32);
-
- // In case of vector by matrix or small workloads, we use the OpenCL buffer rather than the OpenCL image2d
- const bool use_cl_image2d = ((m == 1) || ((((m * n * b) / 16) < 2048) && n < 128)) ? false : true;
-
- if(bool(validate_image2d_support_on_rhs(tensor_reshaped_info, rhs_info_img)) && use_cl_image2d)
- {
- return std::make_pair(lhs_info_img, rhs_info_img);
- }
- else
- {
- return std::make_pair(lhs_info_buf, rhs_info_buf);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G52_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- const float r_nk = static_cast<float>(n) / static_cast<float>(k);
-
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- if(m == 1)
- {
- if(r_nk <= 0.4664f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 16, false, true, false, true, false);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 1, 4, 8, 1, 16, false, true, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 1, 4, 8, 1, 16, false, true, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- else
- {
- if(workload <= 274.4000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 4, 1, 16, false, false, false, true, false);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 2, false, false, false, true, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 2, false, false, false, true, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G51_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- const unsigned int n0 = n < 1280 ? 2 : 4;
- const unsigned int h0 = std::max(n / n0, 1U);
- return configure_lhs_rhs_info(m, n, 1, n0, 4, 1, h0, false, true, false, true);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 2, false, true, false, true);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G7x_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- if(n > 2048)
- {
- const unsigned int h0 = std::max(n / 4, 1U);
- return configure_lhs_rhs_info(m, n, 1, 4, 4, 1, h0, false, true, false, true);
- }
- else
- {
- const unsigned int h0 = std::max(n / 2, 1U);
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, h0, false, true, false, true);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 4, false, true, false, true);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
- const float r_nk = static_cast<float>(n) / static_cast<float>(k);
-
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- if(m == 1)
- {
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, false, true, false, false, false);
-
- if(r_mk <= 0.0026f)
- {
- if(r_nk <= 0.4664f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 32, false, true, false, true, false);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, false, true, false, false, true);
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- }
- else
- {
- if(r_mk <= 0.0148f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 32, false, true, false, true, false);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, false, true, false, false, true);
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- }
- }
- else
- {
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 5, 8, 4, 1, 2, false, false, false, false, false);
-
- if(workload <= 362.6000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 8, 1, 16, false, false, false, true, false);
- }
- else
- {
- if(r_mn <= 22.6067f)
- {
- if(workload <= 708.8000f)
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 2, false, false, false, false, true);
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 8, 2, 1, 16, false, false, false, false, false);
- }
- }
- else
- {
- if(r_nk <= 0.0917f)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 8, 1, 16, false, false, false, true, false);
- }
- else
- {
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 2, false, false, false, false, true);
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
-
- if(m == 1)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 32, false, true, false, true, false);
- }
- else
- {
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
-
- if(workload <= 7449.60f)
- {
- if(workload <= 691.60f)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 8, 1, 8, false, false, false, false, false);
- }
- else
- {
- if(workload <= 4155.20f)
- {
- return configure_lhs_rhs_info(m, n, 5, 2, 8, 1, 16, false, false, false, false, false);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 8, 2, 1, 32, false, false, false, false, false);
- }
- }
- }
- else
- {
- if(workload <= 16300.80f)
- {
- if(r_mn <= 44.56f)
- {
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 8, 4, 4, 1, 1, false, true, false, false, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 5, 2, 8, 1, 16, false, false, false, false, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 5, 2, 8, 1, 16, false, false, false, false, false);
- }
- }
- else
- {
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 2, false, true, false, false, true);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 5, 2, 8, 1, 16, false, false, false, false, false);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F16);
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G51_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- const unsigned int n0 = n < 1280 ? 2 : 4;
- const unsigned int h0 = std::max(n / n0, 1U);
- return configure_lhs_rhs_info(m, n, 1, n0, 8, 1, h0, false, true, false, true);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 2, false, true, false, true);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(dot8_supported(CLKernelLibrary::get().get_device()))
- {
- if(m == 1)
- {
- const unsigned int h0 = std::max(n / 2, 1U);
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, h0, false, true, false, true);
- }
- else
- {
- const unsigned int h0 = std::max(n / 4, 1U);
- return configure_lhs_rhs_info(m, n, 4, 4, 16, 1, h0, false, true, false, true);
- }
- }
- else
- {
- const int h0 = std::max(std::min(static_cast<int>(n / 2), static_cast<int>(128)), static_cast<int>(1));
- if(m == 1)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, h0, false, true, false, true);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 2, 16, 1, h0, false, true, false, true);
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- const unsigned int h0 = std::max(n / 2, 1U);
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, h0, false, true, false, true);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 16, 1, 2, false, true, false, true);
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyBifrost::configure_G51_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- const unsigned int h0 = std::max(n / 2, 1U);
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, h0, false, true, false, true);
- }
- else
- {
- const unsigned int h0 = std::max(n / 2, 1U);
- return configure_lhs_rhs_info(m, n, 4, 2, 16, 1, h0, false, true, false, true);
- }
-}
-
-} // namespace cl_gemm
-} // namespace arm_compute
diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.h b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.h
deleted file mode 100644
index db89d8317c..0000000000
--- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.h
+++ /dev/null
@@ -1,61 +0,0 @@
-/*
- * Copyright (c) 2019-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDRHSONLYBIFROST_H
-#define ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDRHSONLYBIFROST_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** Bifrost based OpenCL GEMMReshapedOnlyRHS configuration */
-class CLGEMMDefaultConfigReshapedRHSOnlyBifrost final : public ICLGEMMKernelConfiguration
-{
-public:
- /** Constructor
- *
- * @param[in] gpu GPU target
- */
- CLGEMMDefaultConfigReshapedRHSOnlyBifrost(GPUTarget gpu);
-
- // Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
-
-private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G52_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G51_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G51_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G51_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDRHSONLYBIFROST_H */
diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.cpp b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.cpp
deleted file mode 100644
index a3f0509eda..0000000000
--- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.cpp
+++ /dev/null
@@ -1,564 +0,0 @@
-/*
- * Copyright (c) 2020-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#include "src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.h"
-
-#include "arm_compute/core/CL/CLHelpers.h"
-#include "arm_compute/core/CL/CLKernelLibrary.h"
-#include "arm_compute/core/GPUTarget.h"
-#include "arm_compute/core/TensorInfo.h"
-#include "arm_compute/core/TensorShape.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
-#include "src/core/CL/gemm/CLGEMMHelpers.h"
-
-#include <utility>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-using namespace arm_compute::misc::shape_calculator;
-
-CLGEMMDefaultConfigReshapedRHSOnlyValhall::CLGEMMDefaultConfigReshapedRHSOnlyValhall(GPUTarget gpu)
- : ICLGEMMKernelConfiguration(gpu)
-{
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type)
-{
- using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (CLGEMMDefaultConfigReshapedRHSOnlyValhall::*)(unsigned int m, unsigned int n, unsigned int k,
- unsigned int b);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G77(&CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G77_f32,
- &CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G77_f16,
- &CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G77_u8);
-
- CLGEMMConfigArray<ConfigurationFunctionExecutorPtr> configs_G78(&CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G78_f32,
- &CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G78_f16,
- &CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G77_u8);
-
- ConfigurationFunctionExecutorPtr func = nullptr;
-
- switch(_target)
- {
- case GPUTarget::G78:
- func = configs_G78.get_function(data_type);
- break;
- case GPUTarget::G77:
- default:
- func = configs_G77.get_function(data_type);
- break;
- }
-
- ARM_COMPUTE_ERROR_ON_MSG(func == nullptr, "Data type not support for GEMM");
- return (this->*func)(m, n, k, b);
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- if(m == 1)
- {
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
-
- if(r_mk <= 0.0064484127797186375)
- {
- if(r_mn <= 0.0028273810748942196)
- {
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
-
- const unsigned int h0 = std::max(n / 4, 1U);
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 1, 4, 8, 1, 16, 0, 1, 0, 0, 1);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 1, 4, 4, 1, h0, 0, 1, 0, 1, 0);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 8, 0, 1, 0, 0, 0);
- }
- }
- else
- {
- if(r_mk <= 0.020312500186264515)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 4, 0, 1, 0, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, 0, 1, 0, 1, 0);
- }
- }
- }
- else
- {
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
-
- if(workload <= 1999.2000122070312)
- {
- if(workload <= 747.1999816894531)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 4, 1, 8, 0, 1, 0, 1, 0);
- }
- else
- {
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 2, 4, 8, 1, 2, 0, 0, 0, 1, 1);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 2, 2, 4, 1, 8, 0, 1, 0, 1, 0);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- else
- {
- if(r_mn <= 0.03348214365541935)
- {
- if(r_mk <= 0.028125000186264515)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 4, 1, 8, 0, 1, 0, 1, 0);
- }
- else
- {
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 2, 4, 8, 1, 2, 0, 0, 0, 1, 1);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 2, 2, 4, 1, 8, 0, 1, 0, 1, 0);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- else
- {
- GEMMLHSMatrixInfo lhs_info_buf;
- GEMMRHSMatrixInfo rhs_info_buf;
- GEMMLHSMatrixInfo lhs_info_img;
- GEMMRHSMatrixInfo rhs_info_img;
- std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 2, 0, 1, 0, 0, 1);
- std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 16, 0, 1, 0, 1, 0);
-
- return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
- std::make_pair(lhs_info_buf, rhs_info_buf),
- n, k, b, DataType::F32);
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- const unsigned int h0 = std::max(n / 2, 1U);
- if(n <= 836.0)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, h0, 0, 1, 0, 1, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, h0, 0, 1, 0, 1, 0);
- }
- }
- else if(m < 128)
- {
- const int h0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(256)), static_cast<int>(1));
- if(k >= 512)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 16, 1, h0, 0, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 1, h0, 0, 1, 0, 0);
- }
- }
- else
- {
- const int h0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(256)), static_cast<int>(1));
- if(n >= 64)
- {
- return configure_lhs_rhs_info(m, n, 4, 8, 4, 1, h0, 0, 1, 0, 0);
- }
- else
- {
- if(k >= 512)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 16, 1, h0, 0, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 1, h0, 0, 1, 0, 0);
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- ARM_COMPUTE_UNUSED(k);
- ARM_COMPUTE_UNUSED(b);
-
- if(m == 1)
- {
- const unsigned int h0 = std::max(n / 2, 1U);
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, h0, 0, 1, 0, 1);
- }
- else
- {
- const int h0 = std::max(std::min(static_cast<int>(n / 4), static_cast<int>(256)), static_cast<int>(1));
- if(m >= 28)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 16, 1, h0, 0, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 16, 1, h0, 0, 1, 0, 1);
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G78_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
- const float r_nk = static_cast<float>(n) / static_cast<float>(k);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
-
- if(m == 1)
- {
- if(workload <= 278.7000f)
- {
- if(workload <= 7.5000f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, 2, 0, 1, 1, 0, 0);
- }
- else
- {
- if(r_mn <= 0.0031f)
- {
- if(workload <= 256.6000f)
- {
- if(workload <= 16.7500f)
- {
- if(r_nk <= 1.6671f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 2, 1, 32, 0, 0, 0, 1, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, 2, 0, 1, 1, 0, 0);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 2, 1, 32, 0, 0, 0, 1, 0);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 2, 1, 32, 0, 0, 0, 1, 0);
- }
- }
- else
- {
- if(r_mk <= 0.0027f)
- {
- if(r_mk <= 0.0014f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 2, 1, 32, 0, 0, 0, 1, 0);
- }
- else
- {
- if(workload <= 8.9500f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, 2, 0, 1, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 2, 1, 32, 0, 0, 0, 1, 0);
- }
- }
- }
- else
- {
- if(workload <= 14.1500f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, 2, 0, 1, 1, 0, 0);
- }
- else
- {
- if(r_mk <= 0.0041f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 2, 1, 32, 0, 0, 0, 1, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 8, 1, 2, 0, 1, 1, 0, 0);
- }
- }
- }
- }
- }
- }
- else
- {
- if(workload <= 363.7000f)
- {
- if(r_mk <= 0.0031f)
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 2, 1, 32, 0, 1, 0, 1, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 4, 1, 32, 0, 1, 0, 1, 0);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 4, 2, 1, 32, 0, 1, 0, 1, 0);
- }
- }
- }
- else
- {
- if(workload <= 1384.8000f)
- {
- if(workload <= 704.0000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 4, 1, 32, 0, 1, 0, 1, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 1, 4, 0, 0, 0, 1, 1);
- }
- }
- else
- {
- if(workload <= 16761.6006f)
- {
- if(r_mn <= 187.1250f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 16, 0, 0, 0, 1, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 1, 4, 0, 0, 0, 1, 1);
- }
- }
- else
- {
- if(r_mk <= 432.4630f)
- {
- return configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 16, 0, 0, 0, 1, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 16, 0, 1, 0, 1, 1);
- }
- }
- }
- }
-}
-
-std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMDefaultConfigReshapedRHSOnlyValhall::configure_G78_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
-{
- const float r_mn = static_cast<float>(m) / static_cast<float>(n);
- const float r_mk = static_cast<float>(m) / static_cast<float>(k);
- const float r_nk = static_cast<float>(n) / static_cast<float>(k);
- const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
-
- if(m == 1)
- {
- if(r_mn <= 0.0038f)
- {
- if(workload <= 353.9000f)
- {
- if(workload <= 278.7000f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 32, 0, 0, 1, 0, 0);
- }
- else
- {
- if(r_mk <= 0.0004f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 32, 0, 0, 1, 0, 0);
- }
- else
- {
- if(r_mk <= 0.0030f)
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 4, 1, 8, 0, 1, 1, 0, 1);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 32, 0, 0, 1, 0, 0);
- }
- }
- }
- }
- else
- {
- if(r_nk <= 1.9384f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 32, 0, 0, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 8, 4, 1, 8, 0, 1, 1, 0, 1);
- }
- }
- }
- else
- {
- if(r_nk <= 1.0368f)
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 16, 1, 32, 0, 0, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, 32, 0, 0, 1, 0, 0);
- }
- }
- }
- else
- {
- if(workload <= 1422.4000f)
- {
- if(workload <= 704.0000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 2, 8, 1, 32, 0, 0, 1, 0, 0);
- }
- else
- {
- if(workload <= 1197.6000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 1, 8, 0, 1, 1, 0, 1);
- }
- else
- {
- if(workload <= 1241.6000f)
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 8, 1, 16, 0, 1, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 1, 8, 0, 1, 1, 0, 1);
- }
- }
- }
- }
- else
- {
- if(workload <= 2769.6000f)
- {
- if(workload <= 1846.4000f)
- {
- if(r_mn <= 2.4927f)
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 8, 1, 16, 0, 1, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 8, 1, 32, 0, 1, 1, 0, 0);
- }
- }
- else
- {
- if(r_mn <= 0.6261f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 8, 1, 32, 0, 1, 1, 0, 0);
- }
- else
- {
- if(r_mk <= 3.4453f)
- {
- if(r_mn <= 1.4135f)
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 8, 1, 16, 0, 1, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 8, 1, 32, 0, 1, 1, 0, 0);
- }
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 8, 1, 16, 0, 1, 1, 0, 0);
- }
- }
- }
- }
- else
- {
- if(r_nk <= 0.0302f)
- {
- return configure_lhs_rhs_info(m, n, 2, 4, 8, 1, 8, 0, 1, 1, 0, 1);
- }
- else
- {
- if(r_mk <= 181.3750f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 8, 1, 32, 0, 1, 1, 0, 0);
- }
- else
- {
- if(workload <= 28035.2002f)
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 8, 1, 16, 0, 1, 1, 0, 0);
- }
- else
- {
- if(r_mk <= 808.6667f)
- {
- return configure_lhs_rhs_info(m, n, 4, 4, 8, 1, 32, 0, 1, 1, 0, 0);
- }
- else
- {
- return configure_lhs_rhs_info(m, n, 2, 8, 8, 1, 16, 0, 1, 1, 0, 0);
- }
- }
- }
- }
- }
- }
- }
-}
-} // namespace cl_gemm
-} // namespace arm_compute
diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.h b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.h
deleted file mode 100644
index a3b556c441..0000000000
--- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.h
+++ /dev/null
@@ -1,55 +0,0 @@
-/*
- * Copyright (c) 2020-2021 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDRHSONLYVALHALL_H
-#define ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDRHSONLYVALHALL_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** Valhall based OpenCL GEMMReshapedOnlyRHS configuration */
-class CLGEMMDefaultConfigReshapedRHSOnlyValhall final : public ICLGEMMKernelConfiguration
-{
-public:
- /** Constructor
- *
- * @param[in] gpu GPU target
- */
- CLGEMMDefaultConfigReshapedRHSOnlyValhall(GPUTarget gpu);
-
- // Inherited overridden method
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure(unsigned int m, unsigned int n, unsigned int k, unsigned int b, DataType data_type) override;
-
-private:
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G78_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G78_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
- std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G77_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMDEFAULTCONFIGRESHAPEDRHSONLYVALHALL_H */
diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfiguration.h b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfiguration.h
deleted file mode 100644
index 001b98dca8..0000000000
--- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfiguration.h
+++ /dev/null
@@ -1,63 +0,0 @@
-/*
- * Copyright (c) 2019-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLGEMMRESHAPEDONLYRHSKERNELCONFIGURATION_H
-#define ARM_COMPUTE_CLGEMMRESHAPEDONLYRHSKERNELCONFIGURATION_H
-
-#include "src/core/CL/ICLGEMMKernelConfiguration.h"
-#include "src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyBifrost.h"
-#include "src/core/CL/gemm/reshaped_only_rhs/CLGEMMDefaultConfigReshapedRHSOnlyValhall.h"
-
-#include <memory>
-
-namespace arm_compute
-{
-namespace cl_gemm
-{
-/** CLGEMMReshapedOnlyRHS factory class */
-class CLGEMMReshapedOnlyRHSKernelConfigurationFactory final
-{
-public:
- /** Static method to call the CLGEMMReshapedOnlyRHS kernel configuration class accordingly with the GPU target
- *
- * @param[in] gpu GPU target
- *
- * @return CLGEMMReshapedOnlyRHS kernel configuration class
- */
- static std::unique_ptr<ICLGEMMKernelConfiguration> create(GPUTarget gpu)
- {
- switch(get_arch_from_target(gpu))
- {
- case GPUTarget::MIDGARD:
- case GPUTarget::BIFROST:
- return std::make_unique<CLGEMMDefaultConfigReshapedRHSOnlyBifrost>(gpu);
- case GPUTarget::VALHALL:
- return std::make_unique<CLGEMMDefaultConfigReshapedRHSOnlyValhall>(gpu);
- default:
- ARM_COMPUTE_ERROR("Not supported GPU target");
- }
- }
-};
-} // namespace cl_gemm
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLGEMMRESHAPEDONLYRHSKERNELCONFIGURATION_H */