arm_compute v18.02

Change-Id: I7207aa488e5470f235f39b6c188b4678dc38d1a6

README.md

@@ -1,14 +1,18 @@
 
+: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
+**Make sure you are using the latest version of the library before opening an issue. Thanks**
 
 Related projects:
 
-- [Caffe on Compute Library](https://github.com/OAID/caffeOnACL)
+- [Caffe on Compute Library](https://github.com/OAID/Caffe-HRT)
 - [Tutorial: Cartoonifying Images on Raspberry Pi with the Compute Library](https://community.arm.com/graphics/b/blog/posts/cartoonifying-images-on-raspberry-pi-with-the-compute-library)
+- [Tutorial: Running AlexNet on Raspberry Pi with Compute Library](https://community.arm.com/processors/b/blog/posts/running-alexnet-on-raspberry-pi-with-compute-library)
 
 Documentation available here:
 
+- [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/)
 - [v17.10](https://arm-software.github.io/ComputeLibrary/v17.10/)
@@ -20,6 +24,8 @@ 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)
 - [v18.01](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)

SConscript

@@ -24,8 +24,8 @@ import os.path
 import re
 import subprocess
 
-VERSION = "v18.01"
-SONAME_VERSION="7.0.0"
+VERSION = "v18.02"
+SONAME_VERSION="9.0.0"
 
 Import('env')
 Import('vars')
@@ -123,12 +123,34 @@ def create_version_file(target, source, env):
     with open(target[0].get_path(), "w") as fd:
         fd.write(build_info)
 
-
 arm_compute_env = env.Clone()
-# Don't allow undefined references in the libraries:
-arm_compute_env.Append(LINKFLAGS=['-Wl,--no-undefined','-Wl,--no-allow-shlib-undefined'])
 
+# Generate embed files
 generate_embed = [ arm_compute_env.Command("src/core/arm_compute_version.embed", "", action=create_version_file) ]
+if env['opencl'] and env['embed_kernels']:
+    cl_files = Glob('src/core/CL/cl_kernels/*.cl')
+    cl_files += Glob('src/core/CL/cl_kernels/*.h')
+
+    embed_files = [ f.get_path()+"embed" for f in cl_files ]
+    arm_compute_env.Append(CPPPATH =[Dir("./src/core/CL/").path] )
+
+    generate_embed.append(arm_compute_env.Command(embed_files, cl_files, action=resolve_includes))
+
+if env['gles_compute'] and env['embed_kernels']:
+    cs_files = Glob('src/core/GLES_COMPUTE/cs_shaders/*.cs')
+    cs_files += Glob('src/core/GLES_COMPUTE/cs_shaders/*.h')
+
+    embed_files = [ f.get_path()+"embed" for f in cs_files ]
+    arm_compute_env.Append(CPPPATH =[Dir("./src/core/GLES_COMPUTE/").path] )
+
+    generate_embed.append(arm_compute_env.Command(embed_files, cs_files, action=resolve_includes))
+
+Default(generate_embed)
+if env["build"] == "embed_only":
+    Return()
+
+# Don't allow undefined references in the libraries:
+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"]:
@@ -161,24 +183,14 @@ if env['opencl']:
     runtime_files += Glob('src/runtime/CL/*.cpp')
     runtime_files += Glob('src/runtime/CL/functions/*.cpp')
 
-    # Generate embed files
-    if env['embed_kernels']:
-        cl_files = Glob('src/core/CL/cl_kernels/*.cl')
-        cl_files += Glob('src/core/CL/cl_kernels/*.h')
-
-        embed_files = [ f.get_path()+"embed" for f in cl_files ]
-        arm_compute_env.Append(CPPPATH =[Dir("./src/core/CL/").path] )
-
-        generate_embed.append(arm_compute_env.Command(embed_files, cl_files, action=resolve_includes))
-
 if env['neon']:
     core_files += Glob('src/core/NEON/*.cpp')
     core_files += Glob('src/core/NEON/kernels/*.cpp')
 
     # build winograd sources for either v7a / v8a
-    core_files += Glob('src/core/NEON/kernels/winograd/*.cpp')
-    core_files += Glob('src/core/NEON/kernels/winograd/transforms/*.cpp')
-    arm_compute_env.Append(CPPPATH = ["arm_compute/core/NEON/kernels/winograd/"])
+    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/"])
 
     if env['arch'] == "armv7a":
         core_files += Glob('src/core/NEON/kernels/arm32/*.cpp')
@@ -199,16 +211,6 @@ if env['gles_compute']:
     runtime_files += Glob('src/runtime/GLES_COMPUTE/*.cpp')
     runtime_files += Glob('src/runtime/GLES_COMPUTE/functions/*.cpp')
 
-    # Generate embed files
-    if env['embed_kernels']:
-        cs_files = Glob('src/core/GLES_COMPUTE/cs_shaders/*.cs')
-        cs_files += Glob('src/core/GLES_COMPUTE/cs_shaders/*.h')
-
-        embed_files = [ f.get_path()+"embed" for f in cs_files ]
-        arm_compute_env.Append(CPPPATH =[Dir("./src/core/GLES_COMPUTE/").path] )
-
-        generate_embed.append(arm_compute_env.Command(embed_files, cs_files, action=resolve_includes))
-
 arm_compute_core_a = build_library('arm_compute_core-static', core_files, static=True)
 Export('arm_compute_core_a')
 
@@ -255,8 +257,6 @@ else:
 
 Default(alias)
 
-Default(generate_embed)
-
 if env['standalone']:
     Depends([alias,arm_compute_core_a], generate_embed)
 else:

SConstruct

@@ -42,7 +42,7 @@ vars.AddVariables(
     BoolVariable("logging", "Logging (this flag is forced to 1 for debug=1)", False),
     EnumVariable("arch", "Target Architecture", "armv7a", allowed_values=("armv7a", "arm64-v8a", "arm64-v8.2-a", "x86_32", "x86_64")),
     EnumVariable("os", "Target OS", "linux", allowed_values=("linux", "android", "bare_metal")),
-    EnumVariable("build", "Build type", "cross_compile", allowed_values=("native", "cross_compile")),
+    EnumVariable("build", "Build type", "cross_compile", allowed_values=("native", "cross_compile", "embed_only")),
     BoolVariable("examples", "Build example programs", True),
     BoolVariable("Werror", "Enable/disable the -Werror compilation flag", True),
     BoolVariable("standalone", "Builds the tests as standalone executables, links statically with libgcc, libstdc++ and libarm_compute", False),
@@ -59,11 +59,17 @@ vars.AddVariables(
 
 env = Environment(platform="posix", variables=vars, ENV = os.environ)
 env.Append(LIBPATH = ["#build/%s" % env['build_dir']])
+Export('env')
+Export('vars')
 
 SConsignFile('build/.%s' % env['build_dir'])
 
 Help(vars.GenerateHelpText(env))
 
+if env['build'] == "embed_only":
+    SConscript('./SConscript', variant_dir='#build/%s' % env['build_dir'], duplicate=0)
+    Return()
+
 if env['neon'] and 'x86' in env['arch']:
     print "Cannot compile NEON for x86"
     Exit(1)
@@ -82,7 +88,7 @@ env.Append(CXXFLAGS = ['-Wno-deprecated-declarations','-Wall','-DARCH_ARM',
          '-Wextra','-Wno-unused-parameter','-pedantic','-Wdisabled-optimization','-Wformat=2',
          '-Winit-self','-Wstrict-overflow=2','-Wswitch-default',
          '-fpermissive','-std=gnu++11','-Wno-vla','-Woverloaded-virtual',
-         '-Wctor-dtor-privacy','-Wsign-promo','-Weffc++','-Wno-format-nonliteral','-Wno-overlength-strings','-Wno-strict-overflow'])
+         '-Wctor-dtor-privacy','-Wsign-promo','-Weffc++','-Wno-format-nonliteral','-Wno-overlength-strings','-Wno-strict-overflow','-Wno-implicit-fallthrough'])
 
 env.Append(CPPDEFINES = ['_GLIBCXX_USE_NANOSLEEP'])
 
@@ -134,8 +140,8 @@ elif env['arch'] == 'arm64-v8a':
     elif env['os'] == 'android':
         prefix = "aarch64-linux-android-"
 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'])
 
@@ -229,8 +235,6 @@ if env['logging']:
 env.Append(CPPPATH = ['#/include', "#"])
 env.Append(CXXFLAGS = env['extra_cxx_flags'])
 
-Export('vars')
-Export('env')
 Export('version_at_least')
 
 if env['opencl']:

arm_compute/core/AccessWindowAutoPadding.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -66,7 +66,7 @@ class AccessWindowAutoPadding : public IAccessWindow
 
     // Inherited methods overridden:
     bool update_window_if_needed(Window &window) const override;
-    bool update_padding_if_needed(const Window &window) const override;
+    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:

arm_compute/core/AccessWindowStatic.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -79,7 +79,7 @@ class AccessWindowStatic : public IAccessWindow
 
     // Inherited methods overriden:
     bool update_window_if_needed(Window &window) const override;
-    bool update_padding_if_needed(const Window &window) const override;
+    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;
 
     ITensorInfo *_info;

arm_compute/core/AccessWindowTranspose.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -40,7 +40,7 @@ class AccessWindowTranspose : public AccessWindowRectangle
 public:
     using AccessWindowRectangle::AccessWindowRectangle;
     bool update_window_if_needed(Window &window) const override;
-    bool update_padding_if_needed(const Window &window) const override;
+    bool update_padding_if_needed(const Window &window) override;
     using AccessWindowRectangle::compute_valid_region;
     ValidRegion compute_valid_region(const Window &window, ValidRegion input_valid_region, bool border_undefined, BorderSize border_size) const override;
 };

arm_compute/core/CL/CLKernels.h

@@ -53,6 +53,7 @@
 #include "arm_compute/core/CL/kernels/CLDerivativeKernel.h"
 #include "arm_compute/core/CL/kernels/CLDilateKernel.h"
 #include "arm_compute/core/CL/kernels/CLDirectConvolutionLayerKernel.h"
+#include "arm_compute/core/CL/kernels/CLDirectConvolutionLayerOutputStageKernel.h"
 #include "arm_compute/core/CL/kernels/CLErodeKernel.h"
 #include "arm_compute/core/CL/kernels/CLFastCornersKernel.h"
 #include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"