arm_compute v18.03

Change-Id: I8f9a2a9d32a6cab019b8504d313216f28671f9f5

Author: Jenkins

Diff for: .github/issue_template.md

@@ -0,0 +1,18 @@
+<!--
+Please fill the fields below in order to help us diagnose the issue. If you have a
+general question or a problem with the scripts, you can ignore these fields.
+-->
+
+**Output of 'strings libarm_compute.so | grep arm_compute_version':**
+
+**Platform:**
+
+**Operating System:**
+
+
+<!--
+Please describe the issue (error, expected behaviour etc) and steps to reproduce it. If possible,
+share the shortest code necessary that reproduces the issue.
+-->
+
+**Problem description:**

Diff for: README.md

@@ -1,9 +1,14 @@
 
-:warning: **Deprecation notice: QS8 and QS16 data types will be removed in the next release** (As far as we know nobody uses these data types, if you do or think they are useful please open an Issue or send us an email):warning:
+:warning: **Deprecation notice: QS8 and QS16 data types will be removed in the 18.05 release** (As far as we know nobody uses these data types, if you do or think they are useful please open an Issue or send us an email):warning:
 
 Please report issues here: https://github.com/ARM-software/ComputeLibrary/issues
 **Make sure you are using the latest version of the library before opening an issue. Thanks**
 
+News:
+
+- We're hiring! [Senior Machine Learning C++ Software Engineer](https://careers.peopleclick.com/careerscp/client_arm/external/jobDetails.do?functionName=getJobDetail&jobPostId=36246&localeCode=en-us)
+- Come talk to us: [Gian Marco will be presenting his work at the EVS](https://www.embedded-vision.com/summit/even-faster-cnns-exploring-new-class-winograd-algorithms)
+
 Related projects:
 
 - [Caffe on Compute Library](https://github.com/OAID/Caffe-HRT)
@@ -12,6 +17,7 @@ Related projects:
 
 Documentation available here:
 
+- [v18.03](https://arm-software.github.io/ComputeLibrary/v18.03/)
 - [v18.02](https://arm-software.github.io/ComputeLibrary/v18.02/)
 - [v18.01](https://arm-software.github.io/ComputeLibrary/v18.01/)
 - [v17.12](https://arm-software.github.io/ComputeLibrary/v17.12/)
@@ -24,8 +30,10 @@ Documentation available here:
 
 Binaries available here:
 
-- [v18.02-linux](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.01/arm_compute-v18.02-bin-linux.tar.gz)
-- [v18.02-android](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.01/arm_compute-v18.02-bin-android.tar.gz)
+- [v18.03-linux](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.03/arm_compute-v18.03-bin-linux.tar.gz)
+- [v18.03-android](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.03/arm_compute-v18.03-bin-android.tar.gz)
+- [v18.02-linux](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.02/arm_compute-v18.02-bin-linux.tar.gz)
+- [v18.02-android](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.02/arm_compute-v18.02-bin-android.tar.gz)
 - [v18.01](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.01/arm_compute-v18.01-bin.tar.gz)
 - [v17.12](https://github.com/ARM-software/ComputeLibrary/releases/download/v17.12/arm_compute-v17.12-bin.tar.gz)
 - [v17.10](https://github.com/ARM-software/ComputeLibrary/releases/download/v17.10/arm_compute-v17.10-bin.tar.gz)

Diff for: SConscript

@@ -24,8 +24,8 @@ import os.path
 import re
 import subprocess
 
-VERSION = "v18.02"
-SONAME_VERSION="9.0.0"
+VERSION = "v18.03"
+SONAME_VERSION="10.0.0"
 
 Import('env')
 Import('vars')

Diff for: arm_compute/core/Dimensions.h

@@ -50,7 +50,7 @@ class Dimensions
      */
     template <typename... Ts>
     explicit Dimensions(Ts... dims)
-        : _id{ { dims... } }, _num_dimensions{ sizeof...(dims) }
+        : _id{ { static_cast<T>(dims)... } }, _num_dimensions{ sizeof...(dims) }
     {
     }
 

Diff for: arm_compute/core/NEON/kernels/convolution/winograd/gemm.hpp

@@ -65,11 +65,11 @@ inline void BlockedGemm(
   const int c_row_stride
 ) {
   // Array access methods
-  const auto A = [a, M, K, a_row_stride] (const int i, const int j) -> TIn {
+  const auto A = [a, a_row_stride] (const int i, const int j) -> TIn {
     return a[i*a_row_stride + j];
   };
 
-  const auto B = [b, K, N, b_row_stride] (const int i, const int j) -> TIn {
+  const auto B = [b, b_row_stride] (const int i, const int j) -> TIn {
     return b[i*b_row_stride + j];
   };
 

Diff for: arm_compute/core/utils/logging/Macros.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -50,14 +50,16 @@
         }                                                                                \
     } while(false)
 
-#define ARM_COMPUTE_LOG_STREAM(logger_name, log_level, stream)                                                 \
-    do                                                                                                         \
-    {                                                                                                          \
-        auto __logger = arm_compute::logging::LoggerRegistry::get().logger(logger_name);                       \
-        if(__logger != nullptr)                                                                                \
-        {                                                                                                      \
-            __logger->log(log_level, static_cast<std::ostringstream &>(std::ostringstream() << stream).str()); \
-        }                                                                                                      \
+#define ARM_COMPUTE_LOG_STREAM(logger_name, log_level, stream)                           \
+    do                                                                                   \
+    {                                                                                    \
+        auto __logger = arm_compute::logging::LoggerRegistry::get().logger(logger_name); \
+        if(__logger != nullptr)                                                          \
+        {                                                                                \
+            std::ostringstream s;                                                        \
+            s << stream;                                                                 \
+            __logger->log(log_level, s.str());                                           \
+        }                                                                                \
     } while(false)
 
 #else /* ARM_COMPUTE_LOGGING_ENABLED */

Diff for: arm_compute/runtime/CL/CLTuner.h

@@ -37,26 +37,43 @@ class ICLKernel;
 class CLTuner : public ICLTuner
 {
 public:
-    /** Constructor */
-    CLTuner();
+    /** Constructor
+     *
+     * @param[in] tune_new_kernels Find the optimal local workgroup size for kernels which are not present in the table ?
+     *
+     */
+    CLTuner(bool tune_new_kernels = true);
 
     /** Destructor */
     ~CLTuner() = default;
 
+    /** Setter for tune_new_kernels option
+     *
+     * @param[in] tune_new_kernels Find the optimal local workgroup size for kernels which are not present in the table ?
+     */
+    void set_tune_new_kernels(bool tune_new_kernels);
+    /** Tune kernels that are not in the LWS table
+     *
+     * @return True if tuning of new kernels is enabled.
+     */
+    bool tune_new_kernels() const;
+    /** Manually add a LWS for a kernel
+     *
+     * @param[in] kernel_id   Unique identifiant of the kernel
+     * @param[in] optimal_lws Optimal local workgroup size to use for the given kernel
+     */
+    void add_lws_to_table(const std::string &kernel_id, cl::NDRange optimal_lws);
     /** Import LWS table
      *
      * @param[in] lws_table The unordered_map container to import
      */
     void import_lws_table(const std::unordered_map<std::string, cl::NDRange> &lws_table);
 
-    /** Export LWS table
+    /** Give read access to the LWS table
      *
      * return The lws table as unordered_map container
      */
-    const std::unordered_map<std::string, cl::NDRange> &export_lws_table();
-
-    // Inherited methods overridden:
-    void tune_kernel(ICLKernel &kernel) override;
+    const std::unordered_map<std::string, cl::NDRange> &lws_table() const;
 
     /** Set the OpenCL kernel event
      *
@@ -66,7 +83,28 @@ class CLTuner : public ICLTuner
      */
     void set_cl_kernel_event(cl_event kernel_event);
 
-    std::function<decltype(clEnqueueNDRangeKernel)> real_function;
+    std::function<decltype(clEnqueueNDRangeKernel)> real_clEnqueueNDRangeKernel;
+
+    /** Load the LWS table from file
+     *
+     * @param[in] filename Load the LWS table from this file.(Must exist)
+     */
+    void load_from_file(const std::string &filename);
+
+    /** Save the content of the LWS table to file
+     *
+     * @param[in] filename Save the LWS table to this file. (Content will be overwritten)
+     */
+    void save_to_file(const std::string &filename) const;
+
+    // Inherited methods overridden:
+    void tune_kernel(ICLKernel &kernel) override;
+
+    /** Is the kernel_event set ?
+     *
+     * @return true if the kernel_event is set.
+     */
+    bool kernel_event_is_set() const;
 
 private:
     /** Find optimal LWS using brute-force approach
@@ -81,33 +119,7 @@ class CLTuner : public ICLTuner
     cl::CommandQueue _queue;
     cl::CommandQueue _queue_profiler;
     cl::Event        _kernel_event;
-};
-
-/* Function to be used to intercept kernel enqueues and store their OpenCL Event */
-class Interceptor
-{
-public:
-    explicit Interceptor(CLTuner &tuner);
-
-    /** clEnqueueNDRangeKernel interface
-     *
-     * @param[in] command_queue           A valid command-queue. The kernel will be queued for execution on the device associated with command_queue.
-     * @param[in] kernel                  A valid kernel object. The OpenCL context associated with kernel and command_queue must be the same.
-     * @param[in] work_dim                The number of dimensions used to specify the global work-items and work-items in the work-group. work_dim must be greater than zero and less than or equal to CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS.
-     * @param[in] gwo                     Global-Workgroup-Offset. It can be used to specify an array of work_dim unsigned values that describe the offset used to calculate the global ID of a work-item. If global_work_offset is NULL, the global IDs start at offset (0, 0, ... 0).
-     * @param[in] gws                     Global-Workgroup-Size. Points to an array of work_dim unsigned values that describe the number of global work-items in work_dim dimensions that will execute the kernel function.
-     * @param[in] lws                     Local-Workgroup-Size. Points to an array of work_dim unsigned values that describe the number of work-items that make up a work-group
-     * @param[in] num_events_in_wait_list Number of events in the waiting list
-     * @param[in] event_wait_list         Event waiting list
-     * @param[in] event                   OpenCL kernel event
-     *
-     * @return the OpenCL status
-     */
-    cl_int operator()(cl_command_queue command_queue, cl_kernel kernel, cl_uint work_dim, const size_t *gwo, const size_t *gws, const size_t *lws, cl_uint num_events_in_wait_list,
-                      const cl_event *event_wait_list, cl_event *event);
-
-private:
-    CLTuner &_tuner;
+    bool             _tune_new_kernels;
 };
 }
 #endif /*__ARM_COMPUTE_CLTUNER_H__ */