Integrate MLGO into CLGEMM and CLGEMMLowpMatrixMultiplyCore: Part2

* Associate CLScheduler with CLGEMMHeuristicsHandle

* Add option in arm_compute_validation for mlgo file path

* Extend logging for the selection of gemm configurations

Resolves: COMPMID-3843, COMPMID-3844

Signed-off-by: SiCong Li <sicong.li@arm.com>
Change-Id: I869c3a4122414ae6a7bbd721966c1da37621ca11
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5002
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

6 files changed, 86 insertions, 22 deletions

diff --git a/arm_compute/runtime/CL/CLScheduler.h b/arm_compute/runtime/CL/CLScheduler.h
index 8a22832792..d3a91da751 100644
--- a/arm_compute/runtime/CL/CLScheduler.h
+++ b/arm_compute/runtime/CL/CLScheduler.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2020 Arm Limited.
+ * Copyright (c) 2016-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -30,6 +30,7 @@
 #include "arm_compute/core/Error.h"
 #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/ICLTuner.h"
 
 namespace arm_compute
@@ -57,15 +58,17 @@ public:
      *  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)
      */
-    void default_init(ICLTuner *cl_tuner = nullptr);
+    void default_init(ICLTuner *cl_tuner = nullptr, CLGEMMHeuristicsHandle *gemm_h = nullptr);
     /** Initialises the scheduler with context and device provided by the user
      *
      * @param[in] device   OpenCL device to be used
      * @param[in] ctx      OpenCL ctx to be used
      * @param[in] cl_tuner (Optional) Pointer to ICLTuner (default=nullptr)
+     * @param[in] gemm_h   (Optional) Pointer to CLGEMMHeuristicsHandle (default = nullptr)
      */
-    void default_init_with_context(cl::Device &device, cl::Context &ctx, ICLTuner *cl_tuner = nullptr);
+    void default_init_with_context(cl::Device &device, cl::Context &ctx, ICLTuner *cl_tuner = nullptr, CLGEMMHeuristicsHandle *gemm_h = nullptr);
 
     /** Schedule the execution of the passed kernel if possible.
      *
@@ -88,8 +91,9 @@ public:
      * @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)
      */
-    void init(cl::Context context, cl::CommandQueue queue, const cl::Device &device, ICLTuner *cl_tuner = nullptr);
+    void init(cl::Context context, cl::CommandQueue queue, const cl::Device &device, ICLTuner *cl_tuner = nullptr, CLGEMMHeuristicsHandle *gemm_h = nullptr);
 
     /** Accessor for the associated CL context.
      *
@@ -109,6 +113,12 @@ public:
      */
     GPUTarget target() const;
 
+    /** Accessor for the associated CLGEMMHeuristicsHandle
+     *
+     * @return Pointer to CLGEMMHeuristicsHandle
+     */
+    CLGEMMHeuristicsHandle *gemm_heuristics() const;
+
     /** Accessor to set the CL context to be used by the scheduler.
      *
      * @param[in] context A CL context.
@@ -161,6 +171,7 @@ private:
     bool                      _is_initialised;
     ICLTuner                 *_cl_tuner;
     std::unique_ptr<ICLTuner> _cl_default_static_tuner;
+    CLGEMMHeuristicsHandle   *_gemm_heuristics;
 };
 } // 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 19095a5589..ab973f973c 100644
--- a/arm_compute/runtime/CL/CLTypes.h
+++ b/arm_compute/runtime/CL/CLTypes.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,6 +23,7 @@
  */
 #ifndef ARM_COMPUTE_RUNTIME_CLTYPES_H
 #define ARM_COMPUTE_RUNTIME_CLTYPES_H
+#include "arm_compute/core/Types.h"
 
 namespace arm_compute
 {
diff --git a/src/runtime/CL/CLScheduler.cpp b/src/runtime/CL/CLScheduler.cpp
index 6fc7baed63..ef5cb03b32 100644
--- a/src/runtime/CL/CLScheduler.cpp
+++ b/src/runtime/CL/CLScheduler.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2020 Arm Limited.
+ * Copyright (c) 2016-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -49,6 +49,11 @@ GPUTarget CLScheduler::target() const
     return _target;
 }
 
+CLGEMMHeuristicsHandle *CLScheduler::gemm_heuristics() const
+{
+    return _gemm_heuristics;
+}
+
 void CLScheduler::set_queue(cl::CommandQueue queue)
 {
     _queue = std::move(queue);
@@ -92,7 +97,7 @@ bool CLScheduler::is_initialised() const
 std::once_flag CLScheduler::_initialize_symbols;
 
 CLScheduler::CLScheduler()
-    : _context(), _queue(), _target(GPUTarget::MIDGARD), _is_initialised(false), _cl_tuner(nullptr), _cl_default_static_tuner(nullptr)
+    : _context(), _queue(), _target(GPUTarget::MIDGARD), _is_initialised(false), _cl_tuner(nullptr), _cl_default_static_tuner(nullptr), _gemm_heuristics(nullptr)
 {
 }
 
@@ -103,20 +108,20 @@ CLScheduler &CLScheduler::get()
     return scheduler;
 }
 
-void CLScheduler::default_init_with_context(cl::Device &device, cl::Context &ctx, ICLTuner *cl_tuner)
+void CLScheduler::default_init_with_context(cl::Device &device, cl::Context &ctx, ICLTuner *cl_tuner, CLGEMMHeuristicsHandle *gemm_h)
 {
     if(!_is_initialised)
     {
         const std::string cl_kernels_folder("./cl_kernels/");
         cl::CommandQueue  queue = cl::CommandQueue(ctx, device);
         CLKernelLibrary::get().init(cl_kernels_folder, ctx, device);
-        init(ctx, queue, device, cl_tuner);
+        init(ctx, queue, device, cl_tuner, gemm_h);
         _cl_default_static_tuner = tuners::TunerFactory::create_tuner(_target);
         _cl_tuner                = (cl_tuner == nullptr) ? _cl_default_static_tuner.get() : cl_tuner;
     }
 }
 
-void CLScheduler::default_init(ICLTuner *cl_tuner)
+void CLScheduler::default_init(ICLTuner *cl_tuner, CLGEMMHeuristicsHandle *gemm_h)
 {
     if(!_is_initialised)
     {
@@ -127,7 +132,7 @@ void CLScheduler::default_init(ICLTuner *cl_tuner)
         ARM_COMPUTE_ERROR_ON_MSG(err != CL_SUCCESS, "Failed to create OpenCL context");
         cl::CommandQueue queue = cl::CommandQueue(ctx, dev);
         CLKernelLibrary::get().init("./cl_kernels/", ctx, dev);
-        init(ctx, queue, dev, cl_tuner);
+        init(ctx, queue, dev, cl_tuner, gemm_h);
         // Create a default static tuner and set if none was provided
         _cl_default_static_tuner = tuners::TunerFactory::create_tuner(_target);
     }
@@ -142,13 +147,14 @@ void CLScheduler::set_context(cl::Context context)
     CLKernelLibrary::get().set_context(_context);
 }
 
-void CLScheduler::init(cl::Context context, cl::CommandQueue queue, const cl::Device &device, ICLTuner *cl_tuner)
+void CLScheduler::init(cl::Context context, cl::CommandQueue queue, const cl::Device &device, ICLTuner *cl_tuner, CLGEMMHeuristicsHandle *gemm_h)
 {
     set_context(std::move(context));
-    _queue          = std::move(queue);
-    _target         = get_target_from_device(device);
-    _is_initialised = true;
-    _cl_tuner       = cl_tuner;
+    _queue           = std::move(queue);
+    _target          = get_target_from_device(device);
+    _is_initialised  = true;
+    _cl_tuner        = cl_tuner;
+    _gemm_heuristics = gemm_h;
 }
 
 void CLScheduler::enqueue_common(ICLKernel &kernel, ITensorPack &tensors, bool flush)
diff --git a/src/runtime/CL/Utils.cpp b/src/runtime/CL/Utils.cpp
index e04ce84e49..da3d4850bf 100644
--- a/src/runtime/CL/Utils.cpp
+++ b/src/runtime/CL/Utils.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,6 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include "arm_compute/runtime/CL/Utils.h"
+
 #include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 
diff --git a/tests/main.cpp b/tests/main.cpp
index 46a081b6c8..ff641c0762 100644
--- a/tests/main.cpp
+++ b/tests/main.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -38,6 +38,7 @@
 
 #ifdef ARM_COMPUTE_CL
 #include "arm_compute/core/CL/OpenCL.h"
+#include "arm_compute/runtime/CL/CLGEMMHeuristicsHandle.h"
 #include "arm_compute/runtime/CL/CLHelpers.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CL/CLTuner.h"
@@ -143,6 +144,9 @@ int main(int argc, char **argv)
 
     auto tuner_file = parser.add_option<utils::SimpleOption<std::string>>("tuner-file", "");
     tuner_file->set_help("File to load/save CLTuner values");
+
+    auto mlgo_file = parser.add_option<utils::SimpleOption<std::string>>("mlgo-file", "");
+    mlgo_file->set_help("File to load MLGO heuristics");
 #endif /* ARM_COMPUTE_CL */
     auto threads = parser.add_option<utils::SimpleOption<int>>("threads", 1);
     threads->set_help("Number of threads to use");
@@ -190,12 +194,14 @@ int main(int argc, char **argv)
 #endif /* ARM_COMPUTE_GC */
 
 #ifdef ARM_COMPUTE_CL
-        CLTuner cl_tuner(false);
+        CLTuner                cl_tuner(false);
+        CLGEMMHeuristicsHandle gemm_heuristics;
         if(opencl_is_available())
         {
             auto ctx_dev_err = create_opencl_context_and_device();
             ARM_COMPUTE_ERROR_ON_MSG(std::get<2>(ctx_dev_err) != CL_SUCCESS, "Failed to create OpenCL context");
-            CLScheduler::get().default_init_with_context(std::get<1>(ctx_dev_err), std::get<0>(ctx_dev_err), &cl_tuner);
+            gemm_heuristics.reload_from_file(mlgo_file->value());
+            CLScheduler::get().default_init_with_context(std::get<1>(ctx_dev_err), std::get<0>(ctx_dev_err), &cl_tuner, &gemm_heuristics);
         }
 
         if(enable_tuner->is_set())
diff --git a/utils/TypePrinter.h b/utils/TypePrinter.h
index 3996d0ddce..bf5b64a1eb 100644
--- a/utils/TypePrinter.h
+++ b/utils/TypePrinter.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -35,6 +35,7 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTunerTypes.h"
+#include "arm_compute/runtime/CL/CLTypes.h"
 #include "support/StringSupport.h"
 
 #include <ostream>
@@ -282,7 +283,8 @@ inline ::std::ostream &operator<<(::std::ostream &os, const GEMMLHSMatrixInfo &g
  */
 inline ::std::ostream &operator<<(::std::ostream &os, const GEMMRHSMatrixInfo &gemm_info)
 {
-    os << "( n0= " << (unsigned int)gemm_info.n0 << " k0= " << gemm_info.k0 << "  h0= " << gemm_info.h0 << "  trans= " << gemm_info.transpose << "  inter= " << gemm_info.interleave << "})";
+    os << "( n0= " << (unsigned int)gemm_info.n0 << " k0= " << gemm_info.k0 << "  h0= " << gemm_info.h0 << "  trans= " << gemm_info.transpose << "  inter= " << gemm_info.interleave << " exp_img=" <<
+       gemm_info.export_to_cl_image << "})";
     return os;
 }
 
@@ -2441,6 +2443,42 @@ inline std::string to_string(const CLTunerMode val)
         }
     }
 }
+/** Converts a @ref CLGEMMKernelType to string
+ *
+ * @param[in] val CLGEMMKernelType value to be converted
+ *
+ * @return String representing the corresponding CLGEMMKernelType
+ */
+inline std::string to_string(CLGEMMKernelType val)
+{
+    switch(val)
+    {
+        case CLGEMMKernelType::NATIVE_V1:
+        {
+            return "Native_V1";
+        }
+        case CLGEMMKernelType::RESHAPED_V1:
+        {
+            return "Reshaped_V1";
+        }
+        case CLGEMMKernelType::NATIVE:
+        {
+            return "Native";
+        }
+        case CLGEMMKernelType::RESHAPED_ONLY_RHS:
+        {
+            return "Reshaped_Only_RHS";
+        }
+        case CLGEMMKernelType::RESHAPED:
+        {
+            return "Reshaped";
+        }
+        default:
+        {
+            return "Unknown";
+        }
+    }
+}
 /** [Print CLTunerMode type] **/
 /** Formatted output of the CLTunerMode type.
  *