Add tuner mode support in GeMM benchmark

Change-Id: Ie1ad9880d22daa2a8ee4f239b5b23d3a7ba4cb1b
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5210
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>

8 files changed, 61 insertions, 24 deletions

diff --git a/examples/gemm_tuner/CommonGemmExampleOptions.cpp b/examples/gemm_tuner/CommonGemmExampleOptions.cpp
index 440973737c..f1306ccf5c 100644
--- a/examples/gemm_tuner/CommonGemmExampleOptions.cpp
+++ b/examples/gemm_tuner/CommonGemmExampleOptions.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -35,6 +35,7 @@ using namespace utils;
     os << "K : " << common_params.K << std::endl;
     os << "B : " << common_params.B << std::endl;
     os << "Data type : " << common_params.data_type << std::endl;
+    os << "OpenCL tuner mode : " << common_params.tuner_mode << std::endl;
     return os;
 }
 
@@ -44,7 +45,8 @@ CommonGemmExampleOptions::CommonGemmExampleOptions(CommandLineParser &parser, Da
       N(parser.add_positional_option<SimpleOption<size_t>>("N", 100)),
       K(parser.add_positional_option<SimpleOption<size_t>>("K", 50)),
       B(parser.add_positional_option<SimpleOption<size_t>>("B", 1)),
-      data_type()
+      data_type(),
+      tuner_mode()
 {
     const std::set<DataType> supported_data_types
     {
@@ -52,9 +54,18 @@ CommonGemmExampleOptions::CommonGemmExampleOptions(CommandLineParser &parser, Da
         DataType::F32,
         DataType::QASYMM8,
     };
+
+    const std::set<CLTunerMode> supported_tuner_modes
+    {
+        CLTunerMode::EXHAUSTIVE,
+        CLTunerMode::NORMAL,
+        CLTunerMode::RAPID
+    };
+
     ARM_COMPUTE_ERROR_ON_MSG(supported_data_types.find(default_data_type) == supported_data_types.end(), "Default data type unsupported");
 
-    data_type = parser.add_option<EnumOption<DataType>>("type", supported_data_types, default_data_type);
+    data_type  = parser.add_option<EnumOption<DataType>>("type", supported_data_types, default_data_type);
+    tuner_mode = parser.add_option<EnumOption<CLTunerMode>>("tuner-mode", supported_tuner_modes, CLTunerMode::RAPID);
 
     help->set_help("Show this help message.");
     M->set_help("Number of lhs matrix rows.");
@@ -62,16 +73,18 @@ CommonGemmExampleOptions::CommonGemmExampleOptions(CommandLineParser &parser, Da
     K->set_help("Number of lhs matrix columns/rhs matrix rows.");
     B->set_help("Batch size.");
     data_type->set_help("Data type to use");
+    tuner_mode->set_help("OpenCL tuner mode");
 }
 
 CommonGemmExampleParams consume_common_gemm_example_parameters(const CommonGemmExampleOptions &options)
 {
     CommonGemmExampleParams common_params;
-    common_params.M         = options.M->value();
-    common_params.N         = options.N->value();
-    common_params.K         = options.K->value();
-    common_params.B         = options.B->value();
-    common_params.data_type = options.data_type->value();
+    common_params.M          = options.M->value();
+    common_params.N          = options.N->value();
+    common_params.K          = options.K->value();
+    common_params.B          = options.B->value();
+    common_params.data_type  = options.data_type->value();
+    common_params.tuner_mode = options.tuner_mode->value();
     return common_params;
 }
 } // namespace gemm_tuner
diff --git a/examples/gemm_tuner/CommonGemmExampleOptions.h b/examples/gemm_tuner/CommonGemmExampleOptions.h
index 633e9252bd..f7447e3db3 100644
--- a/examples/gemm_tuner/CommonGemmExampleOptions.h
+++ b/examples/gemm_tuner/CommonGemmExampleOptions.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,6 +26,7 @@
 
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/Utils.h"
+#include "arm_compute/runtime/CL/CLTuner.h"
 #include "utils/TypePrinter.h"
 #include "utils/command_line/CommandLineOptions.h"
 #include "utils/command_line/CommandLineParser.h"
@@ -35,11 +36,12 @@ namespace gemm_tuner
 /** Structure holding all the common gemm example parameters */
 struct CommonGemmExampleParams
 {
-    size_t                M{ 100 };                                /**< Number of lhs matrix rows */
-    size_t                N{ 100 };                                /**< Number of rhs matrix columns */
-    size_t                K{ 50 };                                 /**< Number of lhs matrix columns/rhs matrix rows */
-    size_t                B{ 1 };                                  /**< Batch size */
-    arm_compute::DataType data_type{ arm_compute::DataType::F32 }; /**< Data type */
+    size_t                   M{ 100 };                                      /**< Number of lhs matrix rows */
+    size_t                   N{ 100 };                                      /**< Number of rhs matrix columns */
+    size_t                   K{ 50 };                                       /**< Number of lhs matrix columns/rhs matrix rows */
+    size_t                   B{ 1 };                                        /**< Batch size */
+    arm_compute::DataType    data_type{ arm_compute::DataType::F32 };       /**< Data type */
+    arm_compute::CLTunerMode tuner_mode{ arm_compute::CLTunerMode::RAPID }; /**< OpenCL tuner mode */
 };
 
 /** Formatted output of the CommonGemmExampleParams type
@@ -80,12 +82,13 @@ public:
     /** Default destructor */
     ~CommonGemmExampleOptions() = default;
 
-    arm_compute::utils::ToggleOption                      *help;      /**< Show help option */
-    arm_compute::utils::SimpleOption<size_t>              *M;         /**< Number of lhs matrix rows option */
-    arm_compute::utils::SimpleOption<size_t>              *N;         /**< Number of rhs matrix columns option */
-    arm_compute::utils::SimpleOption<size_t>              *K;         /**< Number of lhs matrix columns/rhs matrix rows option */
-    arm_compute::utils::SimpleOption<size_t>              *B;         /**< Batch size option */
-    arm_compute::utils::EnumOption<arm_compute::DataType> *data_type; /**< Data type */
+    arm_compute::utils::ToggleOption                         *help;       /**< Show help option */
+    arm_compute::utils::SimpleOption<size_t>                 *M;          /**< Number of lhs matrix rows option */
+    arm_compute::utils::SimpleOption<size_t>                 *N;          /**< Number of rhs matrix columns option */
+    arm_compute::utils::SimpleOption<size_t>                 *K;          /**< Number of lhs matrix columns/rhs matrix rows option */
+    arm_compute::utils::SimpleOption<size_t>                 *B;          /**< Batch size option */
+    arm_compute::utils::EnumOption<arm_compute::DataType>    *data_type;  /**< Data type */
+    arm_compute::utils::EnumOption<arm_compute::CLTunerMode> *tuner_mode; /**< OpenCL tuner mode */
 };
 
 /** Consumes the common gemm example options and creates a structure containing all information
diff --git a/examples/gemm_tuner/cl_gemm_benchmark.sh b/examples/gemm_tuner/cl_gemm_benchmark.sh
index a49c2e154b..92fe6b194e 100755
--- a/examples/gemm_tuner/cl_gemm_benchmark.sh
+++ b/examples/gemm_tuner/cl_gemm_benchmark.sh
@@ -54,7 +54,10 @@ DEFAULT_ID_EXPERIMENT_START=0
 DEFAULT_NUM_EXPERIMENTS="all"
 
 # Default output file extension
-DEFAULT_OUT_EXTENSION="gemm_benchmark"
+DEFAULT_OUT_EXTENSION="mlgo_benchmark"
+
+# Default OpenCL tuner mode
+DEFAULT_TUNER_MODE="rapid"
 
 # Number of iterations for each benchmark run
 NUM_ITERATION=5
@@ -269,6 +272,10 @@ Options:
         Output file extension.
         Default: ${DEFAULT_OUT_EXTENSION}
 
+        -m <tuner_mode>
+        OpenCL tuner mode.
+        Default: ${DEFAULT_TUNER_MODE}
+
 EOF
 # Print help messages about gemm shapes and various gemm configs
 $HELP && help_gemm_shape_file
@@ -347,6 +354,7 @@ function arr_contains() {
 # Globals:
 #   OUT_DIR
 #   OUT_EXTENSION
+#   TUNER_MODE
 #   EXAMPLE_BIN_DIR
 #   NUM_ITERATION
 #   GEMM_CONFIGS_FILE
@@ -437,7 +445,7 @@ function run() {
 
     echo "Running shape[$array_shapes_idx]=$gemm_shape with config[$array_configs_idx]=$gemm_config" 1>&2
 
-    example_args="${gemm_shape},${gemm_config},--type=${DATA_TYPE}"
+    example_args="${gemm_shape},${gemm_config},--type=${DATA_TYPE},--tuner-mode=${TUNER_MODE}"
     json_filename="${STRATEGY_OPTION}_${gemm_shape}_${gemm_config}_${DATA_TYPE}"
     # Replace "," with "_"
     json_filename=${json_filename//,/_}
@@ -510,11 +518,13 @@ ID_EXPERIMENT_START=${DEFAULT_ID_EXPERIMENT_START}
 NUM_EXPERIMENTS=${DEFAULT_NUM_EXPERIMENTS}
 # Output benchmark result file extension
 OUT_EXTENSION=${DEFAULT_OUT_EXTENSION}
+# OpenCL tuner mode
+TUNER_MODE=${DEFAULT_TUNER_MODE}
 # Toggle help
 HELP=false
 
 # Obtain options
-while getopts "hs:e:g:c:d:o:i:n:t:" opt; do
+while getopts "hs:e:g:c:d:o:i:n:t:m:" opt; do
   case "$opt" in
     h) HELP=true ;;
     s) STRATEGY_OPTION=$(to_lower "${OPTARG}");;
@@ -526,6 +536,7 @@ while getopts "hs:e:g:c:d:o:i:n:t:" opt; do
     i) ID_EXPERIMENT_START="${OPTARG}";;
     n) NUM_EXPERIMENTS="${OPTARG}";;
     t) OUT_EXTENSION="${OPTARG}";;
+    m) TUNER_MODE="${OPTARG}";;
   esac
 done
 shift $((OPTIND - 1))
diff --git a/examples/gemm_tuner/cl_gemm_native.cpp b/examples/gemm_tuner/cl_gemm_native.cpp
index 02f144ea12..5a144dabf7 100644
--- a/examples/gemm_tuner/cl_gemm_native.cpp
+++ b/examples/gemm_tuner/cl_gemm_native.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -170,6 +170,8 @@ public:
         std::cout << "Gemm configurations:" << std::endl;
         std::cout << configs << std::endl;
 
+        tuner.set_tuner_mode(params.tuner_mode);
+
         CLScheduler::get().default_init(&tuner);
 
         lhs.allocator()->init(TensorInfo(TensorShape(params.K, params.M, params.B), 1, params.data_type));
diff --git a/examples/gemm_tuner/cl_gemm_reshaped.cpp b/examples/gemm_tuner/cl_gemm_reshaped.cpp
index 2ea4769dd3..444a342d74 100644
--- a/examples/gemm_tuner/cl_gemm_reshaped.cpp
+++ b/examples/gemm_tuner/cl_gemm_reshaped.cpp
@@ -222,6 +222,8 @@ public:
         std::cout << "Gemm configurations:" << std::endl;
         std::cout << configs << std::endl;
 
+        tuner.set_tuner_mode(params.tuner_mode);
+
         CLScheduler::get().default_init(&tuner);
 
         lhs.allocator()->init(TensorInfo(TensorShape(params.K, params.M, params.B), 1, params.data_type));
diff --git a/examples/gemm_tuner/cl_gemm_reshaped_rhs_only.cpp b/examples/gemm_tuner/cl_gemm_reshaped_rhs_only.cpp
index 10fd2984cf..68bec9da6e 100644
--- a/examples/gemm_tuner/cl_gemm_reshaped_rhs_only.cpp
+++ b/examples/gemm_tuner/cl_gemm_reshaped_rhs_only.cpp
@@ -195,6 +195,8 @@ public:
         std::cout << "Gemm configurations:" << std::endl;
         std::cout << configs << std::endl;
 
+        tuner.set_tuner_mode(params.tuner_mode);
+
         CLScheduler::get().default_init(&tuner);
 
         lhs.allocator()->init(TensorInfo(TensorShape(params.K, params.M, params.B), 1, params.data_type));
diff --git a/examples/gemm_tuner/cl_gemmlowp_reshaped.cpp b/examples/gemm_tuner/cl_gemmlowp_reshaped.cpp
index 7c26e0f389..5b81963752 100644
--- a/examples/gemm_tuner/cl_gemmlowp_reshaped.cpp
+++ b/examples/gemm_tuner/cl_gemmlowp_reshaped.cpp
@@ -208,6 +208,8 @@ public:
         std::cout << "Gemm configurations:" << std::endl;
         std::cout << configs << std::endl;
 
+        tuner.set_tuner_mode(params.tuner_mode);
+
         CLScheduler::get().default_init(&tuner);
 
         lhs.allocator()->init(TensorInfo(TensorShape(params.K, params.M, params.B), 1, params.data_type));
diff --git a/examples/gemm_tuner/cl_gemmlowp_reshaped_rhs_only_fused_output_stage_fixedpoint.cpp b/examples/gemm_tuner/cl_gemmlowp_reshaped_rhs_only_fused_output_stage_fixedpoint.cpp
index f25341ce25..95431ed50c 100644
--- a/examples/gemm_tuner/cl_gemmlowp_reshaped_rhs_only_fused_output_stage_fixedpoint.cpp
+++ b/examples/gemm_tuner/cl_gemmlowp_reshaped_rhs_only_fused_output_stage_fixedpoint.cpp
@@ -187,6 +187,8 @@ public:
         std::cout << "Gemm configurations:" << std::endl;
         std::cout << configs << std::endl;
 
+        tuner.set_tuner_mode(params.tuner_mode);
+
         CLScheduler::get().default_init(&tuner);
 
         lhs.allocator()->init(TensorInfo(TensorShape(params.K, params.M, params.B), 1, params.data_type));