arm_compute v18.05

Author: Jenkins

README.md

@@ -1,5 +1,5 @@
 
-: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:
+: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:
 
 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**
@@ -15,7 +15,7 @@ News:
         - Familiarity in developing compute-intensive applications and ideally industry experience of product development
         - Experience programming in assembly language.
 
-    Interested ? Contact us: developer@arm.com
+    Interested ? Contact us: developer-compute@arm.com
 - 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:
@@ -27,6 +27,7 @@ Related projects:
 
 Documentation available here:
 
+- [v18.05](https://arm-software.github.io/ComputeLibrary/v18.05/)
 - [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/)
@@ -40,6 +41,8 @@ Documentation available here:
 
 Binaries available here:
 
+- [v18.05-linux](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.05/arm_compute-v18.05-bin-linux.tar.gz)
+- [v18.05-android](https://github.com/ARM-software/ComputeLibrary/releases/download/v18.05/arm_compute-v18.05-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)
@@ -53,6 +56,6 @@ Binaries available here:
 - [v17.04](https://github.com/ARM-software/ComputeLibrary/releases/download/v17.04/arm_compute-v17.04-bin.tar.gz)
 - [v17.03.1](https://github.com/ARM-software/ComputeLibrary/releases/download/v17.03.1/arm_compute-v17.03.1-bin.tar.gz)
 
-Support: developer@arm.com
+Support: developer-compute@arm.com
 
 License & Contributions: The software is provided under MIT license. Contributions to this project are accepted under the same license.

SConscript

@@ -24,8 +24,8 @@ import os.path
 import re
 import subprocess
 
-VERSION = "v18.03"
-SONAME_VERSION="10.0.0"
+VERSION = "v18.05"
+SONAME_VERSION="11.0.0"
 
 Import('env')
 Import('vars')
@@ -153,9 +153,6 @@ if env["build"] == "embed_only":
 arm_compute_env.Append(LINKFLAGS=['-Wl,--no-undefined'])
 arm_compute_env.Append(CPPPATH =[Dir("./src/core/").path] )
 
-if env["os"] not in ["android", "bare_metal"]:
-    arm_compute_env.Append(LIBS = ['pthread'])
-
 arm_compute_env.Append(LIBS = ['dl'])
 
 core_files = Glob('src/core/*.cpp')
@@ -170,6 +167,9 @@ runtime_files += Glob('src/runtime/CPP/functions/*.cpp')
 # CLHarrisCorners uses the Scheduler to run CPP kernels
 runtime_files += Glob('src/runtime/CPP/SingleThreadScheduler.cpp')
 
+graph_files = Glob('src/graph/*.cpp')
+graph_files += Glob('src/graph/*/*.cpp')
+
 if env['cppthreads']:
      runtime_files += Glob('src/runtime/CPP/CPPScheduler.cpp')
 
@@ -182,21 +182,30 @@ if env['opencl']:
 
     runtime_files += Glob('src/runtime/CL/*.cpp')
     runtime_files += Glob('src/runtime/CL/functions/*.cpp')
+    runtime_files += Glob('src/runtime/CL/tuners/*.cpp')
+
+    graph_files += Glob('src/graph/backends/CL/*.cpp')
+
 
 if env['neon']:
     core_files += Glob('src/core/NEON/*.cpp')
     core_files += Glob('src/core/NEON/kernels/*.cpp')
 
+    core_files += Glob('src/core/NEON/kernels/arm_gemm/*.cpp')
+
     # build winograd sources for either v7a / v8a
     core_files += Glob('src/core/NEON/kernels/convolution/*/*.cpp')
     core_files += Glob('src/core/NEON/kernels/convolution/winograd/*/*.cpp')
     arm_compute_env.Append(CPPPATH = ["arm_compute/core/NEON/kernels/winograd/", "arm_compute/core/NEON/kernels/assembly/"])
 
+    graph_files += Glob('src/graph/backends/NEON/*.cpp')
+
     if env['arch'] == "armv7a":
-        core_files += Glob('src/core/NEON/kernels/arm32/*.cpp')
+        core_files += Glob('src/core/NEON/kernels/arm_gemm/kernels/a32_*/*.cpp')
+
 
     if "arm64-v8" in env['arch']:
-        core_files += Glob('src/core/NEON/kernels/arm64/*.cpp')
+        core_files += Glob('src/core/NEON/kernels/arm_gemm/kernels/a64_*/*.cpp')
 
     runtime_files += Glob('src/runtime/NEON/*.cpp')
     runtime_files += Glob('src/runtime/NEON/functions/*.cpp')
@@ -211,6 +220,8 @@ if env['gles_compute']:
     runtime_files += Glob('src/runtime/GLES_COMPUTE/*.cpp')
     runtime_files += Glob('src/runtime/GLES_COMPUTE/functions/*.cpp')
 
+    graph_files += Glob('src/graph/backends/GLES/*.cpp')
+
 arm_compute_core_a = build_library('arm_compute_core-static', core_files, static=True)
 Export('arm_compute_core_a')
 
@@ -226,30 +237,14 @@ if env['os'] != 'bare_metal' and not env['standalone']:
     Depends(arm_compute_so, arm_compute_core_so)
     Export('arm_compute_so')
 
-if env['neon'] and env['opencl']:
-    Import('opencl')
-    graph_files = Glob('src/graph/*.cpp')
-    graph_files += Glob('src/graph/nodes/*.cpp')
-    graph_files += Glob('src/graph/operations/*.cpp')
-
-    graph_files += Glob('src/graph/CL/*.cpp')
-    graph_files += Glob('src/graph/NEON/*.cpp')
-
-    shared_graph_objects = [arm_compute_env.SharedObject(f) for f in graph_files]
-    static_graph_objects = [arm_compute_env.StaticObject(f) for f in graph_files]
+arm_compute_graph_a = build_library('arm_compute_graph-static', graph_files, static=True, libs = [ arm_compute_a])
+Export('arm_compute_graph_a')
 
-    arm_compute_graph_a = build_library('arm_compute_graph-static', static_graph_objects, static=True, libs = [ arm_compute_a ])
-    Export('arm_compute_graph_a')
-
-    arm_compute_env.Append(LIBPATH = ["#build/%s/opencl-1.2-stubs" % env['build_dir']])
-    arm_compute_graph_so = build_library('arm_compute_graph', shared_graph_objects, static=False, libs = [ "arm_compute", "arm_compute_core"])
+if env['os'] != 'bare_metal' and not env['standalone']:
+    arm_compute_graph_so = build_library('arm_compute_graph', graph_files, static=False, libs = [ "arm_compute" , "arm_compute_core"])
     Depends(arm_compute_graph_so, arm_compute_so)
-    Depends(arm_compute_graph_so, opencl)
     Export('arm_compute_graph_so')
 
-    graph_alias = arm_compute_env.Alias("arm_compute_graph", [arm_compute_graph_a, arm_compute_graph_so])
-    Default(graph_alias)
-
 if env['standalone']:
     alias = arm_compute_env.Alias("arm_compute", [arm_compute_a])
 else:

SConstruct

@@ -54,7 +54,8 @@ vars.AddVariables(
     BoolVariable("openmp", "Enable OpenMP backend", False),
     BoolVariable("cppthreads", "Enable C++11 threads backend", True),
     PathVariable("build_dir", "Specify sub-folder for the build", ".", PathVariable.PathAccept),
-    ("extra_cxx_flags", "Extra CXX flags to be appended to the build command", "")
+    ("extra_cxx_flags", "Extra CXX flags to be appended to the build command", ""),
+    ("compiler_cache", "Command to prefix to the C and C++ compiler (e.g ccache)", "")
 )
 
 env = Environment(platform="posix", variables=vars, ENV = os.environ)
@@ -97,10 +98,10 @@ default_c_compiler = 'gcc' if env['os'] != 'android' else 'clang'
 cpp_compiler = os.environ.get('CXX', default_cpp_compiler)
 c_compiler = os.environ.get('CC', default_c_compiler)
 
-if env['os'] == 'android' and ( cpp_compiler != 'clang++' or c_compiler != 'clang'):
+if env['os'] == 'android' and ( 'clang++' not in cpp_compiler or 'clang' not in c_compiler ):
     print "WARNING: Only clang is officially supported to build the Compute Library for Android"
 
-if cpp_compiler == 'clang++':
+if 'clang++' in cpp_compiler:
     env.Append(CXXFLAGS = ['-Wno-format-nonliteral','-Wno-deprecated-increment-bool','-Wno-vla-extension','-Wno-mismatched-tags'])
 else:
     env.Append(CXXFLAGS = ['-Wlogical-op','-Wnoexcept','-Wstrict-null-sentinel'])
@@ -109,7 +110,7 @@ if env['cppthreads']:
     env.Append(CPPDEFINES = [('ARM_COMPUTE_CPP_SCHEDULER', 1)])
 
 if env['openmp']:
-    if cpp_compiler == 'clang++':
+    if 'clang++' in cpp_compiler:
         print "Clang does not support OpenMP. Use scheduler=cpp."
         Exit(1)
 
@@ -132,19 +133,20 @@ if env['arch'] == 'armv7a':
         env.Append(CXXFLAGS = ['-mfloat-abi=softfp'])
 elif env['arch'] == 'arm64-v8a':
     env.Append(CXXFLAGS = ['-march=armv8-a'])
-    env.Append(CPPDEFINES = ['ARM_COMPUTE_AARCH64_V8A'])
+    env.Append(CPPDEFINES = ['ARM_COMPUTE_AARCH64_V8A','NO_DOT_IN_TOOLCHAIN'])
     if env['os'] == 'linux':
         prefix = "aarch64-linux-gnu-"
     elif env['os'] == 'bare_metal':
         prefix = "aarch64-elf-"
     elif env['os'] == 'android':
         prefix = "aarch64-linux-android-"
+    if 'clang++' in cpp_compiler:
+        env.Append(CXXFLAGS = ['-no-integrated-as'])
 elif env['arch'] == 'arm64-v8.2-a':
     env.Append(CXXFLAGS = ['-march=armv8.2-a+fp16']) # explicitly enable fp16 extension otherwise __ARM_FEATURE_FP16_VECTOR_ARITHMETIC is undefined
-    env.Append(CPPDEFINES = ['ARM_COMPUTE_AARCH64_V8_2'])
-    if cpp_compiler == 'clang++':
-        env.Append(CXXFLAGS = ['-fno-integrated-as'])
-
+    env.Append(CPPDEFINES = ['ARM_COMPUTE_AARCH64_V8_2','NO_DOT_IN_TOOLCHAIN'])
+    if 'clang++' in cpp_compiler:
+        env.Append(CXXFLAGS = ['-no-integrated-as'])
     if env['os'] == 'linux':
         prefix = "aarch64-linux-gnu-"
     elif env['os'] == 'bare_metal':
@@ -161,8 +163,8 @@ elif env['arch'] == 'x86_64':
 if env['build'] == 'native':
     prefix = ""
 
-env['CC'] = prefix + c_compiler
-env['CXX'] = prefix + cpp_compiler
+env['CC'] = env['compiler_cache']+" "+prefix + c_compiler
+env['CXX'] = env['compiler_cache']+" "+prefix + cpp_compiler
 env['LD'] = prefix + "ld"
 env['AS'] = prefix + "as"
 env['AR'] = prefix + "ar"
@@ -175,7 +177,7 @@ if not GetOption("help"):
         print("ERROR: Compiler '%s' not found" % env['CXX'])
         Exit(1)
 
-    if cpp_compiler == 'g++':
+    if 'clang++' not in cpp_compiler:
         if env['arch'] == 'arm64-v8.2-a' and not version_at_least(compiler_ver, '6.2.1'):
             print "GCC 6.2.1 or newer is required to compile armv8.2-a code"
             Exit(1)
@@ -192,8 +194,6 @@ if not GetOption("help"):
 if env['standalone']:
     env.Append(CXXFLAGS = ['-fPIC'])
     env.Append(LINKFLAGS = ['-static-libgcc','-static-libstdc++'])
-    if env['cppthreads']:
-        env.Append(LINKFLAGS = ['-lpthread'])
 
 if env['Werror']:
     env.Append(CXXFLAGS = ['-Werror'])
@@ -210,9 +210,17 @@ elif env['os'] == 'bare_metal':
 
 if env['opencl']:
     if env['os'] in ['bare_metal'] or env['standalone']:
-        print("Cannot link OpenCL statically, which is required on bare metal")
+        print("Cannot link OpenCL statically, which is required for bare metal / standalone builds")
         Exit(1)
 
+if env['gles_compute']:
+    if env['os'] in ['bare_metal'] or env['standalone']:
+        print("Cannot link OpenGLES statically, which is required for bare metal / standalone builds")
+        Exit(1)
+
+if env["os"] not in ["android", "bare_metal"] and (env['opencl'] or env['cppthreads']):
+    env.Append(LIBS = ['pthread'])
+
 if env['opencl'] or env['gles_compute']:
     if env['embed_kernels']:
         env.Append(CPPDEFINES = ['EMBEDDED_KERNELS'])
@@ -242,7 +250,6 @@ if env['opencl']:
 
 if env['gles_compute'] and env['os'] != 'android':
     env.Append(CPPPATH = ['#/include/linux'])
-    env.Append(LIBPATH = ["#build/%s/opengles-3.1-stubs" % env['build_dir']])
     SConscript("./opengles-3.1-stubs/SConscript", variant_dir="build/%s/opengles-3.1-stubs" % env['build_dir'], duplicate=0)
 
 SConscript('./SConscript', variant_dir='#build/%s' % env['build_dir'], duplicate=0)

arm_compute/core/AccessWindowAutoPadding.h

@@ -52,16 +52,25 @@ class AccessWindowAutoPadding : public IAccessWindow
      * @param[in,out] info Tensor info of the accessed kernel.
      */
     AccessWindowAutoPadding(ITensorInfo *info);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
     AccessWindowAutoPadding(const AccessWindowAutoPadding &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
     AccessWindowAutoPadding &operator=(const AccessWindowAutoPadding &) = delete;
-    AccessWindowAutoPadding(AccessWindowAutoPadding &&)                 = default;
+    /** Allow instances of this class to be move constructed */
+    AccessWindowAutoPadding(AccessWindowAutoPadding &&) = default;
+    /** Allow instances of this class to be moved */
     AccessWindowAutoPadding &operator=(AccessWindowAutoPadding &&) = default;
-    ~AccessWindowAutoPadding()                                     = default;
+    /** Default destructor */
+    ~AccessWindowAutoPadding() = default;
 
     /** Set the valid region to match the entire tensor. */
     void set_valid_region();
 
-    /** Return a valid region that spans across the entire tensor. */
+    /** Return a valid region that spans across the entire tensor.
+     *
+     * @return a valid region.
+     *
+     */
     ValidRegion compute_valid_region() const;
 
     // Inherited methods overridden:

arm_compute/core/AccessWindowStatic.h

@@ -56,11 +56,16 @@ class AccessWindowStatic : public IAccessWindow
      */
     AccessWindowStatic(ITensorInfo *info, int start_x, int start_y, int end_x, int end_y);
 
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
     AccessWindowStatic(const AccessWindowStatic &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
     AccessWindowStatic &operator=(const AccessWindowStatic &) = delete;
-    AccessWindowStatic(AccessWindowStatic &&)                 = default;
+    /** Allow instances of this class to be move constructed */
+    AccessWindowStatic(AccessWindowStatic &&) = default;
+    /** Allow instances of this class to be moved */
     AccessWindowStatic &operator=(AccessWindowStatic &&) = default;
-    ~AccessWindowStatic()                                = default;
+    /** Default destructor */
+    ~AccessWindowStatic() = default;
 
     /** Set the valid region based on the static access pattern and valid
      *  region of the inputs.
@@ -74,6 +79,9 @@ class AccessWindowStatic : public IAccessWindow
      *
      * @param[in] window             Execution window of the kernel.
      * @param[in] input_valid_region Combined valid region of all inputs.
+     *
+     * @return a valid region.
+     *
      */
     ValidRegion compute_valid_region(const Window &window, ValidRegion input_valid_region) const;
 
@@ -82,6 +90,7 @@ class AccessWindowStatic : public IAccessWindow
     bool update_padding_if_needed(const Window &window) override;
     ValidRegion compute_valid_region(const Window &window, ValidRegion input_valid_region, bool border_undefined, BorderSize border_size) const override;
 
+private:
     ITensorInfo *_info;
     int          _start_x;
     int          _start_y;