aboutsummaryrefslogtreecommitdiff
path: root/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/runtime/CL/functions/CLPixelWiseMultiplication.cpp')
-rw-r--r--src/runtime/CL/functions/CLPixelWiseMultiplication.cpp147
1 files changed, 106 insertions, 41 deletions
diff --git a/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp b/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
index 3c1a7de76d..6aa9d9cbb3 100644
--- a/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
+++ b/src/runtime/CL/functions/CLPixelWiseMultiplication.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2016-2020 ARM Limited.
+ * Copyright (c) 2016-2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -24,67 +24,132 @@
#include "arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h"
#include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/CL/kernels/CLPixelWiseMultiplicationKernel.h"
-#include "support/MemorySupport.h"
+#include "arm_compute/runtime/CL/CLScheduler.h"
+
+#include "src/core/CL/ICLKernel.h"
+#include "src/gpu/cl/operators/ClMul.h"
#include <utility>
namespace arm_compute
{
-void CLPixelWiseMultiplication::configure(ICLTensor *input1, ICLTensor *input2, ICLTensor *output, float scale,
- ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
+struct CLPixelWiseMultiplication::Impl
+{
+ const ICLTensor *src_0{nullptr};
+ const ICLTensor *src_1{nullptr};
+ ICLTensor *dst{nullptr};
+ std::unique_ptr<opencl::ClMul> op{nullptr};
+};
+
+CLPixelWiseMultiplication::CLPixelWiseMultiplication() : _impl(std::make_unique<Impl>())
+{
+}
+CLPixelWiseMultiplication::CLPixelWiseMultiplication(CLPixelWiseMultiplication &&) = default;
+CLPixelWiseMultiplication &CLPixelWiseMultiplication::operator=(CLPixelWiseMultiplication &&) = default;
+CLPixelWiseMultiplication::~CLPixelWiseMultiplication() = default;
+
+void CLPixelWiseMultiplication::configure(ICLTensor *input1,
+ ICLTensor *input2,
+ ICLTensor *output,
+ float scale,
+ ConvertPolicy overflow_policy,
+ RoundingPolicy rounding_policy,
+ const ActivationLayerInfo &act_info)
+{
+ configure(CLKernelLibrary::get().get_compile_context(), input1, input2, output, scale, overflow_policy,
+ rounding_policy, act_info);
+}
+
+void CLPixelWiseMultiplication::configure(const CLCompileContext &compile_context,
+ ICLTensor *input1,
+ ICLTensor *input2,
+ ICLTensor *output,
+ float scale,
+ ConvertPolicy overflow_policy,
+ RoundingPolicy rounding_policy,
+ const ActivationLayerInfo &act_info)
{
- configure(CLKernelLibrary::get().get_compile_context(), input1, input2, output, scale, overflow_policy, rounding_policy, act_info);
+ _impl->src_0 = input1;
+ _impl->src_1 = input2;
+ _impl->dst = output;
+ _impl->op = std::make_unique<opencl::ClMul>();
+ _impl->op->configure(compile_context, input1->info(), input2->info(), output->info(), scale, overflow_policy,
+ rounding_policy, act_info);
}
-void CLPixelWiseMultiplication::configure(const CLCompileContext &compile_context, ICLTensor *input1, ICLTensor *input2, ICLTensor *output, float scale,
- ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
+Status CLPixelWiseMultiplication::validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ float scale,
+ ConvertPolicy overflow_policy,
+ RoundingPolicy rounding_policy,
+ const ActivationLayerInfo &act_info)
{
- auto k = arm_compute::support::cpp14::make_unique<CLPixelWiseMultiplicationKernel>();
- k->configure(compile_context, input1, input2, output, scale, overflow_policy, rounding_policy, act_info);
- _kernel = std::move(k);
-
- if(output->info()->dimension(0) > 1)
- {
- ICLTensor *broadcasted_info = (input1->info()->dimension(0) == 1) ? input1 : input2;
-
- if(broadcasted_info->info()->dimension(0) == 1)
- {
- _border_handler.configure(compile_context, broadcasted_info, _kernel->border_size(), BorderMode::REPLICATE);
- }
- }
+ return opencl::ClMul::validate(input1, input2, output, scale, overflow_policy, rounding_policy, act_info);
}
-Status CLPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, float scale,
- ConvertPolicy overflow_policy, RoundingPolicy rounding_policy, const ActivationLayerInfo &act_info)
+void CLPixelWiseMultiplication::run()
{
- return CLPixelWiseMultiplicationKernel::validate(input1, input2, output, scale, overflow_policy, rounding_policy, act_info);
+ ITensorPack pack;
+ pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0);
+ pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1);
+ pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+
+ _impl->op->run(pack);
}
-void CLComplexPixelWiseMultiplication::configure(ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info)
+struct CLComplexPixelWiseMultiplication::Impl
+{
+ const ICLTensor *src_0{nullptr};
+ const ICLTensor *src_1{nullptr};
+ ICLTensor *dst{nullptr};
+ std::unique_ptr<opencl::ClComplexMul> op{nullptr};
+};
+
+CLComplexPixelWiseMultiplication::CLComplexPixelWiseMultiplication() : _impl(std::make_unique<Impl>())
+{
+}
+CLComplexPixelWiseMultiplication::CLComplexPixelWiseMultiplication(CLComplexPixelWiseMultiplication &&) = default;
+CLComplexPixelWiseMultiplication &
+CLComplexPixelWiseMultiplication::operator=(CLComplexPixelWiseMultiplication &&) = default;
+CLComplexPixelWiseMultiplication::~CLComplexPixelWiseMultiplication() = default;
+
+void CLComplexPixelWiseMultiplication::configure(ICLTensor *input1,
+ ICLTensor *input2,
+ ICLTensor *output,
+ const ActivationLayerInfo &act_info)
{
configure(CLKernelLibrary::get().get_compile_context(), input1, input2, output, act_info);
}
-void CLComplexPixelWiseMultiplication::configure(const CLCompileContext &compile_context, ICLTensor *input1, ICLTensor *input2, ICLTensor *output, const ActivationLayerInfo &act_info)
+void CLComplexPixelWiseMultiplication::configure(const CLCompileContext &compile_context,
+ ICLTensor *input1,
+ ICLTensor *input2,
+ ICLTensor *output,
+ const ActivationLayerInfo &act_info)
{
- auto k = arm_compute::support::cpp14::make_unique<CLComplexPixelWiseMultiplicationKernel>();
- k->configure(compile_context, input1, input2, output, act_info);
- _kernel = std::move(k);
-
- if(output->info()->dimension(0) > 1)
- {
- ICLTensor *broadcasted_info = (input1->info()->dimension(0) == 1) ? input1 : input2;
-
- if(broadcasted_info->info()->dimension(0) == 1)
- {
- _border_handler.configure(compile_context, broadcasted_info, _kernel->border_size(), BorderMode::REPLICATE);
- }
- }
+ _impl->src_0 = input1;
+ _impl->src_1 = input2;
+ _impl->dst = output;
+ _impl->op = std::make_unique<opencl::ClComplexMul>();
+ _impl->op->configure(compile_context, input1->info(), input2->info(), output->info(), act_info);
}
-Status CLComplexPixelWiseMultiplication::validate(const ITensorInfo *input1, const ITensorInfo *input2, const ITensorInfo *output, const ActivationLayerInfo &act_info)
+Status CLComplexPixelWiseMultiplication::validate(const ITensorInfo *input1,
+ const ITensorInfo *input2,
+ const ITensorInfo *output,
+ const ActivationLayerInfo &act_info)
{
- return CLComplexPixelWiseMultiplicationKernel::validate(input1, input2, output, act_info);
+ return opencl::ClComplexMul::validate(input1, input2, output, act_info);
+}
+
+void CLComplexPixelWiseMultiplication::run()
+{
+ ITensorPack pack;
+ pack.add_tensor(TensorType::ACL_SRC_0, _impl->src_0);
+ pack.add_tensor(TensorType::ACL_SRC_1, _impl->src_1);
+ pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+
+ _impl->op->run(pack);
}
} // namespace arm_compute
generated by cgit v1.2.1 (git 2.23.0) at 2024-05-05 10:39:29 +0000