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>

5 files changed, 54 insertions, 30 deletions

diff --git a/examples/gemm_tuner/CommonGemmExampleOptions.cpp b/examples/gemm_tuner/CommonGemmExampleOptions.cpp
index f1306ccf5c..bee202b99e 100644
--- a/examples/gemm_tuner/CommonGemmExampleOptions.cpp
+++ b/examples/gemm_tuner/CommonGemmExampleOptions.cpp
@@ -39,7 +39,7 @@ using namespace utils;
     return os;
 }
 
-CommonGemmExampleOptions::CommonGemmExampleOptions(CommandLineParser &parser, DataType default_data_type)
+CommonGemmExampleOptions::CommonGemmExampleOptions(arm_compute::utils::CommandLineParser &parser, arm_compute::DataType default_data_type)
     : help(parser.add_option<ToggleOption>("help")),
       M(parser.add_positional_option<SimpleOption<size_t>>("M", 100)),
       N(parser.add_positional_option<SimpleOption<size_t>>("N", 100)),
diff --git a/examples/gemm_tuner/cl_gemm_native.cpp b/examples/gemm_tuner/cl_gemm_native.cpp
index 5a144dabf7..093935f716 100644
--- a/examples/gemm_tuner/cl_gemm_native.cpp
+++ b/examples/gemm_tuner/cl_gemm_native.cpp
@@ -32,7 +32,7 @@
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CL/CLTuner.h"
-#include "src/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.h"
+#include "src/core/gpu/cl/kernels/ClGemmMatrixMultiplyNativeKernel.h"
 #include "tests/CL/Helper.h"
 #include "utils/Utils.h"
 #include "utils/command_line/CommandLineOptions.h"
@@ -41,6 +41,7 @@
 #include <cstdlib>
 
 using namespace arm_compute;
+using namespace arm_compute::opencl::kernels;
 using namespace utils;
 using namespace arm_compute::misc::shape_calculator;
 using namespace gemm_tuner;
@@ -122,8 +123,8 @@ GemmConfigs consume_gemm_configs(const GemmConfigOptions &options)
 }
 
 } // namespace
-// Create function for CLGEMMMatrixMultiplyNativeKernel
-using CLGEMMMatrixMultiplyNative = test::CLSynthetizeFunction<CLGEMMMatrixMultiplyNativeKernel>;
+// Create function for ClGemmMatrixMultiplyNativeKernel
+using CLGEMMMatrixMultiplyNative = test::CLSynthetizeOperator<ClGemmMatrixMultiplyNativeKernel>;
 
 class CLGEMMMatrixMultiplyNativeExample : public Example
 {
@@ -197,7 +198,7 @@ public:
 
         // Validate argments
         Status status{};
-        status = gemm.validate((&lhs)->info(), (&rhs)->info(), (&bias)->info(), (&dst)->info(), alpha, beta, lhs_info, rhs_info, kernel_info);
+        status = gemm.validate(lhs.info(), rhs.info(), bias.info(), dst.info(), alpha, beta, lhs_info, rhs_info, kernel_info);
         if(!status)
         {
             // Unsupported arguments
@@ -207,7 +208,7 @@ public:
         }
 
         // Configure function
-        gemm.configure(&lhs, &rhs, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
+        gemm.configure(lhs.info(), rhs.info(), bias.info(), dst.info(), alpha, beta, lhs_info, rhs_info, kernel_info);
 
         // Allocate tensors
         lhs.allocator()->allocate();
@@ -220,7 +221,12 @@ public:
     void do_run() override
     {
         // Execute the function
-        gemm.run();
+        ITensorPack gemm_pack({ { ACL_SRC_0, &lhs },
+            { ACL_SRC_1, &rhs },
+            { ACL_SRC_2, &bias },
+            { ACL_DST, &dst }
+        });
+        gemm.run(gemm_pack);
 
         // Make sure all the OpenCL jobs are done executing:
         CLScheduler::get().sync();
diff --git a/examples/gemm_tuner/cl_gemm_reshaped.cpp b/examples/gemm_tuner/cl_gemm_reshaped.cpp
index 444a342d74..e6caeec873 100644
--- a/examples/gemm_tuner/cl_gemm_reshaped.cpp
+++ b/examples/gemm_tuner/cl_gemm_reshaped.cpp
@@ -33,8 +33,8 @@
 #include "arm_compute/runtime/CL/CLTuner.h"
 #include "examples/gemm_tuner/CommonGemmExampleOptions.h"
 #include "examples/gemm_tuner/GemmTunerHelpers.h"
-#include "src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.h"
-#include "src/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h"
+#include "src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedKernel.h"
+#include "src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.h"
 #include "tests/CL/Helper.h"
 #include "utils/Utils.h"
 #include "utils/command_line/CommandLineOptions.h"
@@ -43,6 +43,7 @@
 #include <cstdlib>
 
 using namespace arm_compute;
+using namespace arm_compute::opencl::kernels;
 using namespace utils;
 using namespace arm_compute::misc::shape_calculator;
 using namespace gemm_tuner;
@@ -172,10 +173,11 @@ GemmConfigs consume_gemm_configs(const GemmConfigOptions &options)
 }
 
 } // namespace
-// Create function for CLGEMMReshapeLHSMatrixKernel
-using CLGEMMReshapeLHSMatrix = test::CLSynthetizeFunction<CLGEMMReshapeLHSMatrixKernel>;
-// Create function for CLGEMMMatrixMultiplyReshapedKernel
-using CLGEMMMatrixMultiplyReshaped = test::CLSynthetizeFunction<CLGEMMMatrixMultiplyReshapedKernel>;
+
+// Create function for ClGemmReshapeLhsMatrixKernel
+using CLGEMMReshapeLHSMatrix = test::CLSynthetizeOperator<ClGemmReshapeLhsMatrixKernel>;
+// Create function for ClGemmMatrixMultiplyReshapedKernel
+using CLGEMMMatrixMultiplyReshaped = test::CLSynthetizeOperator<ClGemmMatrixMultiplyReshapedKernel>;
 
 class CLGEMMMatrixMultiplyReshapedExample : public Example
 {
@@ -271,7 +273,7 @@ public:
 
         // Validate argments
         Status status{};
-        status = reshape_lhs.validate((&lhs)->info(), (&lhs_reshaped)->info(), lhs_info, kernel_info.reinterpret_input_as_3d);
+        status = reshape_lhs.validate(lhs.info(), lhs_reshaped.info(), lhs_info, kernel_info.reinterpret_input_as_3d);
         if(!status)
         {
             // Unsupported arguments
@@ -280,7 +282,7 @@ public:
             return false;
         }
 
-        status = gemm.validate((&lhs_reshaped)->info(), (&rhs_reshaped)->info(), (&bias)->info(), (&dst)->info(), alpha, beta, lhs_info, rhs_info, kernel_info);
+        status = gemm.validate(lhs_reshaped.info(), rhs_reshaped.info(), bias.info(), dst.info(), alpha, beta, lhs_info, rhs_info, kernel_info);
         if(!status)
         {
             // Unsupported arguments
@@ -290,10 +292,10 @@ public:
         }
 
         // Configure reshape lhs function
-        reshape_lhs.configure(&lhs, &lhs_reshaped, lhs_info);
+        reshape_lhs.configure(lhs.info(), lhs_reshaped.info(), lhs_info);
 
         // Configure function
-        gemm.configure(&lhs_reshaped, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
+        gemm.configure(lhs_reshaped.info(), rhs_reshaped.info(), bias.info(), dst.info(), alpha, beta, lhs_info, rhs_info, kernel_info);
 
         // Allocate tensors
         lhs.allocator()->allocate();
@@ -307,9 +309,16 @@ public:
     }
     void do_run() override
     {
-        // Execute the function
-        reshape_lhs.run();
-        gemm.run();
+        // Execute the functions
+        ITensorPack reshape_lsh_pack({ { ACL_SRC, &lhs }, { ACL_DST, &lhs_reshaped } });
+        reshape_lhs.run(reshape_lsh_pack);
+
+        ITensorPack gemm_pack({ { ACL_SRC_0, &lhs_reshaped },
+            { ACL_SRC_1, &rhs_reshaped },
+            { ACL_SRC_2, &bias },
+            { ACL_DST, &dst }
+        });
+        reshape_lhs.run(gemm_pack);
 
         // Make sure all the OpenCL jobs are done executing:
         CLScheduler::get().sync();
diff --git a/examples/gemm_tuner/cl_gemm_reshaped_rhs_only.cpp b/examples/gemm_tuner/cl_gemm_reshaped_rhs_only.cpp
index 68bec9da6e..dbaaca6048 100644
--- a/examples/gemm_tuner/cl_gemm_reshaped_rhs_only.cpp
+++ b/examples/gemm_tuner/cl_gemm_reshaped_rhs_only.cpp
@@ -33,7 +33,7 @@
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CL/CLTuner.h"
-#include "src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.h"
+#include "src/core/gpu/cl/kernels/ClGemmMatrixMultiplyReshapedOnlyRhsKernel.h"
 #include "tests/CL/Helper.h"
 #include "utils/Utils.h"
 #include "utils/command_line/CommandLineOptions.h"
@@ -42,6 +42,7 @@
 #include <cstdlib>
 
 using namespace arm_compute;
+using namespace arm_compute::opencl::kernels;
 using namespace utils;
 using namespace arm_compute::misc::shape_calculator;
 using namespace gemm_tuner;
@@ -147,8 +148,8 @@ GemmConfigs consume_gemm_configs(const GemmConfigOptions &options)
 }
 
 } // namespace
-// Create function for CLGEMMMatrixMultiplyReshapedOnlyRHSKernel
-using CLGEMMMatrixMultiplyReshapedOnlyRHS = test::CLSynthetizeFunction<CLGEMMMatrixMultiplyReshapedOnlyRHSKernel>;
+// Create function for ClGemmMatrixMultiplyReshapedOnlyRhsKernel
+using CLGEMMMatrixMultiplyReshapedOnlyRHS = test::CLSynthetizeOperator<ClGemmMatrixMultiplyReshapedOnlyRhsKernel>;
 
 class CLGEMMMatrixMultiplyReshapedOnlyRHSExample : public Example
 {
@@ -238,7 +239,7 @@ public:
 
         // Validate argments
         Status status{};
-        status = gemm.validate((&lhs)->info(), (&rhs_reshaped)->info(), (&bias)->info(), (&dst)->info(), alpha, beta, lhs_info, rhs_info, kernel_info);
+        status = gemm.validate(lhs.info(), rhs_reshaped.info(), bias.info(), dst.info(), alpha, beta, lhs_info, rhs_info, kernel_info);
         if(!status)
         {
             // Unsupported arguments
@@ -248,7 +249,7 @@ public:
         }
 
         // Configure function
-        gemm.configure(&lhs, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
+        gemm.configure(lhs.info(), rhs_reshaped.info(), bias.info(), dst.info(), alpha, beta, lhs_info, rhs_info, kernel_info);
 
         // Allocate tensors
         lhs.allocator()->allocate();
@@ -262,7 +263,12 @@ public:
     void do_run() override
     {
         // Execute the function
-        gemm.run();
+        ITensorPack gemm_pack({ { ACL_SRC_0, &lhs },
+            { ACL_SRC_1, &rhs_reshaped },
+            { ACL_SRC_2, &bias },
+            { ACL_DST, &dst }
+        });
+        gemm.run(gemm_pack);
 
         // Make sure all the OpenCL jobs are done executing:
         CLScheduler::get().sync();
diff --git a/examples/gemm_tuner/cl_gemmlowp_reshaped.cpp b/examples/gemm_tuner/cl_gemmlowp_reshaped.cpp
index 5b81963752..3d3f7fef1e 100644
--- a/examples/gemm_tuner/cl_gemmlowp_reshaped.cpp
+++ b/examples/gemm_tuner/cl_gemmlowp_reshaped.cpp
@@ -34,7 +34,7 @@
 #include "examples/gemm_tuner/CommonGemmExampleOptions.h"
 #include "examples/gemm_tuner/GemmTunerHelpers.h"
 #include "src/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.h"
-#include "src/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h"
+#include "src/core/gpu/cl/kernels/ClGemmReshapeLhsMatrixKernel.h"
 #include "tests/CL/Helper.h"
 #include "utils/Utils.h"
 #include "utils/command_line/CommandLineOptions.h"
@@ -43,6 +43,7 @@
 #include <cstdlib>
 
 using namespace arm_compute;
+using namespace arm_compute::opencl::kernels;
 using namespace utils;
 using namespace arm_compute::misc::shape_calculator;
 using namespace gemm_tuner;
@@ -167,7 +168,7 @@ GemmConfigs consume_gemm_configs(const GemmConfigOptions &options)
 
 } // namespace
 
-using CLGEMMReshapeLHSMatrix           = test::CLSynthetizeFunction<CLGEMMReshapeLHSMatrixKernel>;
+using CLGEMMReshapeLHSMatrix           = test::CLSynthetizeOperator<ClGemmReshapeLhsMatrixKernel>;
 using CLGEMMLowpMatrixMultiplyReshaped = test::CLSynthetizeFunction<CLGEMMLowpMatrixMultiplyReshapedKernel>;
 
 class CLGEMMLowpMatrixMultiplyReshapedExample : public Example
@@ -279,7 +280,7 @@ public:
         }
 
         // Configure functions
-        reshape_lhs.configure(&lhs, &lhs_reshaped, lhs_info);
+        reshape_lhs.configure(lhs.info(), lhs_reshaped.info(), lhs_info);
 
         gemm.configure(&lhs_reshaped, &rhs_reshaped, &dst, lhs_info, rhs_info, gemm_info);
 
@@ -294,7 +295,9 @@ public:
     }
     void do_run() override
     {
-        reshape_lhs.run();
+        ITensorPack reshape_lsh_pack({ { ACL_SRC, &lhs }, { ACL_DST, &lhs_reshaped } });
+        reshape_lhs.run(reshape_lsh_pack);
+
         gemm.run();
 
         // Make sure all the OpenCL jobs are done executing: