From 856f66e6c61b77d03f754cd0fa8439891f0e4aca Mon Sep 17 00:00:00 2001
From: Georgios Pinitas <georgios.pinitas@arm.com>
Date: Thu, 22 Apr 2021 21:13:21 +0100
Subject: 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>
---
 tests/CL/Helper.h | 83 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 83 insertions(+)

(limited to 'tests/CL/Helper.h')

diff --git a/tests/CL/Helper.h b/tests/CL/Helper.h
index 5153e98add..b99911e1e6 100644
--- a/tests/CL/Helper.h
+++ b/tests/CL/Helper.h
@@ -29,8 +29,11 @@
 #include "arm_compute/runtime/CL/functions/CLFill.h"
 #include "arm_compute/runtime/IFunction.h"
 #include "src/core/CL/kernels/CLFillBorderKernel.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
+#include "src/runtime/gpu/cl/operators/ClFill.h"
 
 #include "src/core/CL/ICLKernel.h"
+#include "support/Cast.h"
 
 #include <memory>
 
@@ -38,6 +41,86 @@ namespace arm_compute
 {
 namespace test
 {
+/** This template synthetizes a simple IOperator which runs the given kernel K */
+template <typename K>
+class CLSynthetizeOperator : public opencl::IClOperator
+{
+public:
+    /** Configure the kernel.
+     *
+     * @param[in] args Configuration arguments.
+     */
+    template <typename... Args>
+    void configure(Args &&... args)
+    {
+        auto k = std::make_unique<K>();
+        k->configure(CLKernelLibrary::get().get_compile_context(), std::forward<Args>(args)...);
+        _kernel = std::move(k);
+    }
+    /** Configure the kernel setting the GPU target as well
+     *
+     * @param[in] gpu_target GPUTarget to set
+     * @param[in] args       Configuration arguments.
+     */
+    template <typename... Args>
+    void configure(GPUTarget gpu_target, Args &&... args)
+    {
+        auto k = std::make_unique<K>();
+        k->set_target(gpu_target);
+        k->configure(CLKernelLibrary::get().get_compile_context(), std::forward<Args>(args)...);
+        _kernel = std::move(k);
+    }
+    /** Validate input arguments
+     *
+     * @param[in] args Configuration arguments.
+     */
+    template <typename... Args>
+    static Status validate(Args &&... args)
+    {
+        return K::validate(std::forward<Args>(args)...);
+    }
+};
+
+/** As above but this also initializes to zero the input tensor */
+template <typename K, int bordersize>
+class CLSynthetizeOperatorInitOutputWithZeroAndWithZeroConstantBorder : public opencl::IClOperator
+{
+public:
+    /** Configure the kernel.
+     *
+     * @param[in] first  First input argument.
+     * @param[in] second Second input argument.
+     * @param[in] args   Rest of the configuration arguments.
+     */
+    template <typename T, typename... Args>
+    void configure(T first, T second, Args &&... args)
+    {
+        auto cctx = CLKernelLibrary::get().get_compile_context();
+        auto k    = std::make_unique<K>();
+        k->set_target(CLScheduler::get().target());
+        k->configure(cctx, first, second, std::forward<Args>(args)...);
+        _kernel = std::move(k);
+        _border_handler.configure(cctx, first, BorderSize(bordersize), BorderMode::CONSTANT, PixelValue());
+        _fill.configure(cctx, second, PixelValue());
+    }
+
+    // Inherited method overridden:
+    void run(ITensorPack &tensors) override final
+    {
+        ARM_COMPUTE_ERROR_ON_MSG(!_kernel, "The CL kernel or function isn't configured");
+
+        ITensorPack fill_pack = { { ACL_SRC, tensors.get_tensor(TensorType::ACL_DST) } };
+        _fill.run(fill_pack);
+        CLScheduler::get().enqueue_op(_border_handler, tensors);
+        CLScheduler::get().enqueue_op(*_kernel, tensors);
+    }
+
+private:
+    opencl::ClFill             _fill{};           /**< Kernel to initialize the tensor */
+    CLFillBorderKernel         _border_handler{}; /**< Kernel to handle  borders */
+    std::unique_ptr<ICLKernel> _kernel{};         /**< Kernel to run */
+};
+
 /** This template synthetizes an ICLSimpleFunction which runs the given kernel K */
 template <typename K>
 class CLSynthetizeFunction : public ICLSimpleFunction
-- 
cgit v1.2.1