Add support for CLVK

This patch enables CLVK through the graph API and inside the
CLScheduler. By default the Native platform is selected.
Selecting CLVK can be done via --target=clvk.

Resolves COMPMID-4205 and COMPMID-4206

Change-Id: Ic60744980c6b8a60e776627ea677ed46be88f656
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5475
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

8 files changed, 59 insertions, 23 deletions

diff --git a/arm_compute/core/CL/OpenCL.h b/arm_compute/core/CL/OpenCL.h
index 155c3e4eef..1e6b04c042 100644
--- a/arm_compute/core/CL/OpenCL.h
+++ b/arm_compute/core/CL/OpenCL.h
@@ -123,6 +123,7 @@ public:
     DECLARE_FUNCTION_PTR(clGetDeviceIDs);
     DECLARE_FUNCTION_PTR(clGetMemObjectInfo);
     DECLARE_FUNCTION_PTR(clRetainEvent);
+    DECLARE_FUNCTION_PTR(clGetPlatformInfo);
     DECLARE_FUNCTION_PTR(clGetPlatformIDs);
     DECLARE_FUNCTION_PTR(clGetKernelWorkGroupInfo);
     DECLARE_FUNCTION_PTR(clGetCommandQueueInfo);
diff --git a/arm_compute/graph/TypePrinter.h b/arm_compute/graph/TypePrinter.h
index 317c5a7596..0bbb4695de 100644
--- a/arm_compute/graph/TypePrinter.h
+++ b/arm_compute/graph/TypePrinter.h
@@ -48,6 +48,9 @@ inline ::std::ostream &operator<<(::std::ostream &os, const Target &target)
         case Target::CL:
             os << "CL";
             break;
+        case Target::CLVK:
+            os << "CLVK";
+            break;
         default:
             ARM_COMPUTE_ERROR("NOT_SUPPORTED!");
     }
diff --git a/arm_compute/graph/Types.h b/arm_compute/graph/Types.h
index 4f3ce76370..7306b82a1e 100644
--- a/arm_compute/graph/Types.h
+++ b/arm_compute/graph/Types.h
@@ -28,6 +28,7 @@
 #include "arm_compute/core/PixelValue.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTunerTypes.h"
+#include "arm_compute/runtime/CL/CLTypes.h"
 
 #include <limits>
 #include <string>
@@ -37,6 +38,7 @@ namespace arm_compute
 namespace graph
 {
 using arm_compute::CLTunerMode;
+using arm_compute::CLBackendType;
 using arm_compute::Status;
 
 using arm_compute::Coordinates;
@@ -80,16 +82,17 @@ struct TensorDescriptor;
 /** Graph configuration structure */
 struct GraphConfig
 {
-    bool        use_function_memory_manager{ true };   /**< Use a memory manager to manage per-function auxilary memory */
-    bool        use_function_weights_manager{ true };  /**< Use a weights manager to manage transformed weights */
-    bool        use_transition_memory_manager{ true }; /**< Use a memory manager to manager transition buffer memory */
-    bool        use_tuner{ false };                    /**< Use a tuner in tunable backends */
-    bool        use_synthetic_type{ false };           /**< Convert graph to a synthetic graph for a data type */
-    DataType    synthetic_type{ DataType::QASYMM8 };   /**< The data type of the synthetic graph  */
-    CLTunerMode tuner_mode{ CLTunerMode::EXHAUSTIVE }; /**< Tuner mode to be used by the CL tuner */
-    int         num_threads{ -1 };                     /**< Number of threads to use (thread capable backends), if 0 the backend will auto-initialize, if -1 the backend will stay as it is. */
-    std::string tuner_file{ "acl_tuner.csv" };         /**< File to load/store tuning values from */
-    std::string mlgo_file{ "heuristics.mlgo" };        /**< Filename to load MLGO heuristics from */
+    bool          use_function_memory_manager{ true };   /**< Use a memory manager to manage per-function auxilary memory */
+    bool          use_function_weights_manager{ true };  /**< Use a weights manager to manage transformed weights */
+    bool          use_transition_memory_manager{ true }; /**< Use a memory manager to manager transition buffer memory */
+    bool          use_tuner{ false };                    /**< Use a tuner in tunable backends */
+    bool          use_synthetic_type{ false };           /**< Convert graph to a synthetic graph for a data type */
+    DataType      synthetic_type{ DataType::QASYMM8 };   /**< The data type of the synthetic graph  */
+    CLTunerMode   tuner_mode{ CLTunerMode::EXHAUSTIVE }; /**< Tuner mode to be used by the CL tuner */
+    int           num_threads{ -1 };                     /**< Number of threads to use (thread capable backends), if 0 the backend will auto-initialize, if -1 the backend will stay as it is. */
+    std::string   tuner_file{ "acl_tuner.csv" };         /**< File to load/store tuning values from */
+    std::string   mlgo_file{ "heuristics.mlgo" };        /**< Filename to load MLGO heuristics from */
+    CLBackendType backend_type{ CLBackendType::Native }; /**< CL backend type to use */
 };
 
 /**< Device target types */
@@ -98,6 +101,7 @@ enum class Target
     UNSPECIFIED, /**< Unspecified Target */
     NEON,        /**< Arm® Neon™ capable target device */
     CL,          /**< OpenCL capable target device */
+    CLVK,        /**< CLVK capable target device */
 };
 
 /** Supported Element-wise operations */
diff --git a/arm_compute/graph/backends/CL/CLDeviceBackend.h b/arm_compute/graph/backends/CL/CLDeviceBackend.h
index 82c0eacd11..278a8e5031 100644
--- a/arm_compute/graph/backends/CL/CLDeviceBackend.h
+++ b/arm_compute/graph/backends/CL/CLDeviceBackend.h
@@ -28,6 +28,7 @@
 
 #include "arm_compute/runtime/CL/CLBufferAllocator.h"
 #include "arm_compute/runtime/CL/CLGEMMHeuristicsHandle.h"
+#include "arm_compute/runtime/CL/CLTypes.h"
 #include "arm_compute/runtime/CL/CLTuner.h"
 
 namespace arm_compute
@@ -76,6 +77,7 @@ private:
     CLGEMMHeuristicsHandle             _gemm_heuristics; /**< GEMM heuristics */
     std::unique_ptr<CLBufferAllocator> _allocator;       /**< CL buffer affinity allocator */
     std::string                        _tuner_file;      /**< Filename to load/store the tuner's values from */
+    CLBackendType                      _backend_type;    /**< OpenCL backend type to use */
 };
 } // namespace backends
 } // namespace graph
diff --git a/arm_compute/runtime/CL/CLHelpers.h b/arm_compute/runtime/CL/CLHelpers.h
index 9b71561c39..fef2619071 100644
--- a/arm_compute/runtime/CL/CLHelpers.h
+++ b/arm_compute/runtime/CL/CLHelpers.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,6 +25,7 @@
 #define ARM_COMPUTE_CL_HELPERS_H
 
 #include "arm_compute/core/CL/OpenCL.h"
+#include "arm_compute/runtime/CL/CLTypes.h"
 #include "arm_compute/runtime/IScheduler.h"
 
 namespace arm_compute
@@ -37,11 +38,13 @@ class ICLKernel;
  *
  * @note In debug builds, the function will automatically enable cl_arm_printf if the driver/device supports it.
  *
+ * @param[in] cl_backend_type The OpenCL backend type to use.
+ *
  * @return A std::tuple where the first element is the opencl context, the second element is the opencl device
  *         and the third one an error code. The error code will be CL_SUCCESS upon successful creation, otherwise
  *         a value telling why the function failed.
  */
-std::tuple<cl::Context, cl::Device, cl_int> create_opencl_context_and_device();
+std::tuple<cl::Context, cl::Device, cl_int> create_opencl_context_and_device(CLBackendType cl_backend_type);
 /** Schedules a kernel using the context if not nullptr else uses the legacy scheduling flow.
  *
  * @param[in] ctx    Context to use.
@@ -49,5 +52,13 @@ std::tuple<cl::Context, cl::Device, cl_int> create_opencl_context_and_device();
  * @param[in] flush  (Optional) Specifies if the command queue will be flushed after running the kernel.
  */
 void schedule_kernel_on_ctx(CLRuntimeContext *ctx, ICLKernel *kernel, bool flush = true);
+
+/** This function selects the OpenCL platform based on the backend type.
+ *
+ * @param[in] cl_backend_type The OpenCL backend type to use.
+ *
+ * @return A cl::Platform object.
+ */
+cl::Platform select_preferable_platform(CLBackendType cl_backend_type);
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CL_HELPERS_H */
diff --git a/arm_compute/runtime/CL/CLRuntimeContext.h b/arm_compute/runtime/CL/CLRuntimeContext.h
index 083ac0ab88..4ab8f70887 100644
--- a/arm_compute/runtime/CL/CLRuntimeContext.h
+++ b/arm_compute/runtime/CL/CLRuntimeContext.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,6 +29,7 @@
 #include "arm_compute/core/CL/OpenCL.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CL/CLTuner.h"
+#include "arm_compute/runtime/CL/CLTypes.h"
 #include "arm_compute/runtime/IScheduler.h"
 #include "arm_compute/runtime/RuntimeContext.h"
 
@@ -60,6 +61,7 @@ private:
     CLTuner                      _tuner{ false };
     CLSymbols                    _symbols{};
     CLCoreRuntimeContext         _core_context{};
+    CLBackendType                _backend_type{ CLBackendType::Native };
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLRUNTIME_CONTEXT_H */
diff --git a/arm_compute/runtime/CL/CLScheduler.h b/arm_compute/runtime/CL/CLScheduler.h
index 41a074089e..56852aec6e 100644
--- a/arm_compute/runtime/CL/CLScheduler.h
+++ b/arm_compute/runtime/CL/CLScheduler.h
@@ -31,6 +31,8 @@
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/experimental/Types.h"
 #include "arm_compute/runtime/CL/CLGEMMHeuristicsHandle.h"
+#include "arm_compute/runtime/CL/CLHelpers.h"
+#include "arm_compute/runtime/CL/CLTypes.h"
 #include "arm_compute/runtime/CL/ICLTuner.h"
 
 namespace arm_compute
@@ -57,10 +59,11 @@ public:
     /** Initialises the context and command queue used by the scheduler to default values
      *  and sets a default device and kernel path for the @ref CLKernelLibrary.
      *
-     * @param[in] cl_tuner (Optional) Pointer to ICLTuner (default=nullptr)
-     * @param[in] gemm_h   (Optional) Pointer to CLGEMMHeuristicsHandle (default = nullptr)
+     * @param[in] cl_tuner        (Optional) Pointer to ICLTuner (default=nullptr)
+     * @param[in] gemm_h          (Optional) Pointer to CLGEMMHeuristicsHandle (default = nullptr)
+     * @param[in] cl_backend_type (Optional) Type of backend to use (default = CLBackendType::Native)
      */
-    void default_init(ICLTuner *cl_tuner = nullptr, CLGEMMHeuristicsHandle *gemm_h = nullptr);
+    void default_init(ICLTuner *cl_tuner = nullptr, CLGEMMHeuristicsHandle *gemm_h = nullptr, CLBackendType cl_backend_type = CLBackendType::Native);
     /** Initialises the scheduler with context and device provided by the user
      *
      * @param[in] device   OpenCL device to be used
@@ -86,14 +89,16 @@ public:
 
     /** Initialises the context and command queue to be used by the scheduler.
      *
-     * @param[in] context  A CL context.
-     * @param[in] queue    A CL command queue.
-     * @param[in] device   A CL device.
-     * @param[in] cl_tuner (Optional) Pointer to OpenCL tuner (default=nullptr)
-     *                     Note: It is caller's responsibility to release the allocated memory for CLTuner
-     * @param[in] gemm_h   (Optional) Pointer to CLGEMMHeuristicsHandle (default = nullptr)
+     * @param[in] context         A CL context.
+     * @param[in] queue           A CL command queue.
+     * @param[in] device          A CL device.
+     * @param[in] cl_tuner        (Optional) Pointer to OpenCL tuner (default=nullptr)
+     *                            Note: It is caller's responsibility to release the allocated memory for CLTuner
+     * @param[in] gemm_h          (Optional) Pointer to CLGEMMHeuristicsHandle (default = nullptr)
+     * @param[in] cl_backend_type (Optional) Type of backend to use (default = CLBackendType::Native)
      */
-    void init(cl::Context context, cl::CommandQueue queue, const cl::Device &device, ICLTuner *cl_tuner = nullptr, CLGEMMHeuristicsHandle *gemm_h = nullptr);
+    void init(cl::Context context, cl::CommandQueue queue, const cl::Device &device, ICLTuner *cl_tuner = nullptr, CLGEMMHeuristicsHandle *gemm_h = nullptr,
+              CLBackendType cl_backend_type = CLBackendType::Native);
 
     /** Accessor for the associated CL context.
      *
@@ -171,6 +176,7 @@ private:
     bool                    _is_initialised;
     ICLTuner               *_cl_tuner;
     CLGEMMHeuristicsHandle *_gemm_heuristics;
+    CLBackendType           _backend_type;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLSCHEDULER_H */
diff --git a/arm_compute/runtime/CL/CLTypes.h b/arm_compute/runtime/CL/CLTypes.h
index ab973f973c..cf0486c8c3 100644
--- a/arm_compute/runtime/CL/CLTypes.h
+++ b/arm_compute/runtime/CL/CLTypes.h
@@ -58,5 +58,12 @@ struct CLGEMMKernelSelectionParams
     bool         is_rhs_constant{ false };       /**< True if the content of the rhs matrix is constant */
     DataType     data_type{ DataType::UNKNOWN }; /**< Data type */
 };
+
+/** List the possible OpenCL backends */
+enum class CLBackendType
+{
+    Native, /**< OpenCL native backend */
+    Clvk,   /**< CLVK backend */
+};
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_RUNTIME_CLTYPES_H */