aboutsummaryrefslogtreecommitdiff
path: root/src/core/CL/gemm/reshaped
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/reshaped
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/reshaped')
-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
5 files changed, 0 insertions, 1058 deletions
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 */