arm_compute v21.05

Author: Jenkins

Android.bp

@@ -42,7 +42,7 @@ Binaries available at https://github.com/ARM-software/ComputeLibrary/releases.
 
 - x86
 
-### Supported OS
+### Supported Systems
 
 - Android™
 - Bare Metal
@@ -72,7 +72,7 @@ You must use your real name, no pseudonyms or anonymous contributions are accept
 
 Android is a trademark of Google LLC.
 
-Arm, Cortex and Mali are registered trademarks or trademarks of Arm Limited (or its subsidiaries) in the US and/or elsewhere.
+Arm, Cortex, Mali and Neon are registered trademarks or trademarks of Arm Limited (or its subsidiaries) in the US and/or elsewhere.
 
 Linux® is the registered trademark of Linus Torvalds in the U.S. and other countries.
 

README.md

@@ -27,8 +27,8 @@ import zlib
 import base64
 import string
 
-VERSION = "v21.02"
-LIBRARY_VERSION_MAJOR = 22
+VERSION = "v21.05"
+LIBRARY_VERSION_MAJOR = 23
 LIBRARY_VERSION_MINOR =  0
 LIBRARY_VERSION_PATCH =  0
 SONAME_VERSION = str(LIBRARY_VERSION_MAJOR) + "." + str(LIBRARY_VERSION_MINOR) + "." + str(LIBRARY_VERSION_PATCH)
@@ -156,15 +156,6 @@ if env['opencl'] and env['embed_kernels']:
 
     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()
@@ -174,7 +165,6 @@ arm_compute_env.Append(CPPDEFINES = [('ARM_COMPUTE_VERSION_MAJOR', LIBRARY_VERSI
                                      ('ARM_COMPUTE_VERSION_MINOR', LIBRARY_VERSION_MINOR),
                                      ('ARM_COMPUTE_VERSION_PATCH', LIBRARY_VERSION_PATCH)])
 
-
 # Don't allow undefined references in the libraries:
 undefined_flag = '-Wl,-undefined,error' if 'macos' in arm_compute_env["os"] else '-Wl,--no-undefined'
 arm_compute_env.Append(LINKFLAGS=[undefined_flag])
@@ -198,6 +188,25 @@ runtime_files = Glob('src/runtime/*.cpp')
 runtime_files += Glob('src/runtime/CPP/ICPPSimpleFunction.cpp')
 runtime_files += Glob('src/runtime/CPP/functions/*.cpp')
 
+# C API files
+c_api_files = ['src/c/AclContext.cpp',
+               'src/c/AclQueue.cpp',
+               'src/c/AclTensor.cpp',
+               'src/c/AclTensorPack.cpp',
+               'src/c/AclVersion.cpp',
+               ]
+if env['opencl']:
+    c_api_files += ['src/c/cl/AclOpenClExt.cpp']
+
+# Common backend files
+common_backend_files = ['src/common/utils/LegacySupport.cpp',
+                        'src/common/AllocatorWrapper.cpp',
+                        'src/common/ITensorV2.cpp',
+                        'src/common/TensorPack.cpp',
+                        ]
+
+core_files += common_backend_files
+runtime_files += c_api_files
 # CLHarrisCorners uses the Scheduler to run CPP kernels
 runtime_files += Glob('src/runtime/CPP/SingleThreadScheduler.cpp')
 
@@ -229,6 +238,8 @@ if env['opencl']:
     runtime_files += Glob('src/runtime/CL/mlgo/*.cpp')
     runtime_files += Glob('src/runtime/CL/gemm_auto_heuristics/*.cpp')
 
+    runtime_files += Glob('src/gpu/cl/*.cpp')
+
     graph_files += Glob('src/graph/backends/CL/*.cpp')
 
 
@@ -280,8 +291,36 @@ if env['neon']:
     runtime_files += Glob('src/runtime/NEON/functions/*.cpp')
     runtime_files += Glob('src/runtime/NEON/functions/assembly/*.cpp')
 
-    core_files += Glob('src/core/cpu/*.cpp')
-    core_files += Glob('src/core/cpu/kernels/*.cpp')
+    cpu_kernel_hp_files = ['src/core/cpu/kernels/CpuActivationKernel.cpp',
+                           'src/core/cpu/kernels/CpuDepthwiseConvolutionNativeKernel.cpp',
+                           'src/core/cpu/kernels/CpuDirectConvolutionKernel.cpp',
+                           'src/core/cpu/kernels/CpuDirectConvolutionOutputStageKernel.cpp',
+                           'src/core/cpu/kernels/CpuPermuteKernel.cpp',
+                           'src/core/cpu/kernels/CpuPoolingAssemblyWrapperKernel.cpp',
+                           'src/core/cpu/kernels/CpuPoolingKernel.cpp',
+                           'src/core/cpu/kernels/CpuReshapeKernel.cpp',
+                          ]
+    cpu_kernel_files = ['src/core/cpu/kernels/CpuAddKernel.cpp',
+                        'src/core/cpu/kernels/CpuConcatenateBatchKernel.cpp',
+                        'src/core/cpu/kernels/CpuConcatenateDepthKernel.cpp',
+                        'src/core/cpu/kernels/CpuConcatenateHeightKernel.cpp',
+                        'src/core/cpu/kernels/CpuConcatenateWidthKernel.cpp',
+                        'src/core/cpu/kernels/CpuConvertFullyConnectedWeightsKernel.cpp',
+                        'src/core/cpu/kernels/CpuCopyKernel.cpp',
+                        'src/core/cpu/kernels/CpuDequantizationKernel.cpp',
+                        'src/core/cpu/kernels/CpuElementwiseKernel.cpp',
+                        'src/core/cpu/kernels/CpuElementwiseUnaryKernel.cpp',
+                        'src/core/cpu/kernels/CpuFillKernel.cpp',
+                        'src/core/cpu/kernels/CpuFloorKernel.cpp',
+                        'src/core/cpu/kernels/CpuMulKernel.cpp',
+                        'src/core/cpu/kernels/CpuQuantizationKernel.cpp',
+                        'src/core/cpu/kernels/CpuScaleKernel.cpp',
+                        'src/core/cpu/kernels/CpuSoftmaxKernel.cpp',
+                        'src/core/cpu/kernels/CpuSubKernel.cpp',
+                        'src/core/cpu/kernels/CpuTransposeKernel.cpp',
+                       ]
+    core_files += [cpu_kernel_hp_files, cpu_kernel_files]
+
     core_files += Glob('src/core/cpu/kernels/*/*.cpp')
     if any(i in env['data_type_support'] for i in ['all', 'fp16']):
         core_files += Glob('src/core/cpu/kernels/*/*/fp16.cpp')
@@ -295,30 +334,40 @@ if env['neon']:
         core_files += Glob('src/core/cpu/kernels/*/*/qsymm16.cpp')
     if any(i in env['data_type_support'] for i in ['all', 'integer']):
         core_files += Glob('src/core/cpu/kernels/*/*/integer.cpp')
-   
+
     if any(i in env['data_layout_support'] for i in ['all', 'nchw']):
         core_files += Glob('src/core/cpu/kernels/*/*/nchw/all.cpp')
 
-    runtime_files += Glob('src/runtime/cpu/*.cpp')
-    runtime_files += Glob('src/runtime/cpu/operators/*.cpp')
-
-if env['gles_compute']:
-    if env['os'] != 'android':
-        arm_compute_env.Append(CPPPATH = ["#opengles-3.1/include", "#opengles-3.1/mali_include"])
-
-    core_files += Glob('src/core/GLES_COMPUTE/*.cpp')
-    core_files += Glob('src/core/GLES_COMPUTE/kernels/*.cpp')
-
-    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')
-if env['tracing']:
-    arm_compute_env.Append(CPPDEFINES = ['ARM_COMPUTE_TRACING_ENABLED'])
-else:
-    # Remove TracePoint files if tracing is disabled:
-    core_files = [ f for f in core_files if not "TracePoint" in str(f)]
-    runtime_files = [ f for f in runtime_files if not "TracePoint" in str(f)]
+    cpu_rt_files = ['src/cpu/CpuContext.cpp',
+                    'src/cpu/CpuQueue.cpp',
+                    'src/cpu/CpuTensor.cpp'
+                   ]
+    cpu_operator_hp_files = ['src/runtime/cpu/operators/CpuActivation.cpp',
+                             'src/runtime/cpu/operators/CpuDepthwiseConvolution.cpp',
+                             'src/runtime/cpu/operators/CpuDepthwiseConvolutionAssemblyDispatch.cpp',
+                             'src/runtime/cpu/operators/CpuDirectConvolution.cpp',
+                             'src/runtime/cpu/operators/CpuPermute.cpp',
+                             'src/runtime/cpu/operators/CpuPooling.cpp',
+                             'src/runtime/cpu/operators/CpuPoolingAssemblyDispatch.cpp',
+                            ]
+    cpu_operator_files = ['src/runtime/cpu/operators/CpuAdd.cpp',
+                          'src/runtime/cpu/operators/CpuConcatenate.cpp',
+                          'src/runtime/cpu/operators/CpuConvertFullyConnectedWeights.cpp',
+                          'src/runtime/cpu/operators/CpuCopy.cpp',
+                          'src/runtime/cpu/operators/CpuDequantization.cpp',
+                          'src/runtime/cpu/operators/CpuElementwise.cpp',
+                          'src/runtime/cpu/operators/CpuElementwiseUnary.cpp',
+                          'src/runtime/cpu/operators/CpuFill.cpp',
+                          'src/runtime/cpu/operators/CpuFloor.cpp',
+                          'src/runtime/cpu/operators/CpuMul.cpp',
+                          'src/runtime/cpu/operators/CpuQuantization.cpp',
+                          'src/runtime/cpu/operators/CpuReshape.cpp',
+                          'src/runtime/cpu/operators/CpuScale.cpp',
+                          'src/runtime/cpu/operators/CpuSoftmax.cpp',
+                          'src/runtime/cpu/operators/CpuSub.cpp',
+                          'src/runtime/cpu/operators/CpuTranspose.cpp',
+                         ]
+    runtime_files += [ cpu_rt_files, cpu_operator_hp_files, cpu_operator_files ]
 
 bootcode_o = []
 if env['os'] == 'bare_metal':

SConscript

@@ -1,4 +1,6 @@
-# Copyright (c) 2016, 2017 Arm Limited.
+# -*- coding: utf-8 -*-
+
+# Copyright (c) 2016-2021 Arm Limited.
 #
 # SPDX-License-Identifier: MIT
 #
@@ -43,7 +45,7 @@ vars.AddVariables(
     BoolVariable("asserts", "Enable asserts (this flag is forced to 1 for debug=1)", False),
     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", "arm64-v8.2-a-sve", "arm64-v8.2-a-sve2", "x86_32", "x86_64",
+                  allowed_values=("armv7a", "armv7a-hf", "arm64-v8a", "arm64-v8.2-a", "arm64-v8.2-a-sve", "arm64-v8.2-a-sve2", "x86_32", "x86_64",
                                   "armv8a", "armv8.2-a", "armv8.2-a-sve", "armv8.6-a", "armv8.6-a-sve", "armv8.6-a-sve2", "armv8r64", "x86")),
     EnumVariable("estate", "Execution State", "auto", allowed_values=("auto", "32", "64")),
     EnumVariable("os", "Target OS", "linux", allowed_values=("linux", "android", "tizen", "macos", "bare_metal")),
@@ -53,12 +55,10 @@ vars.AddVariables(
     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),
     BoolVariable("opencl", "Enable OpenCL support", True),
-    BoolVariable("neon", "Enable Neon support", False),
-    BoolVariable("gles_compute", "Enable OpenGL ES Compute Shader support", False),
+    BoolVariable("neon", "Enable Arm® Neon™ support", False),
     BoolVariable("embed_kernels", "Embed OpenCL kernels and OpenGL ES compute shaders in library binary", True),
     BoolVariable("compress_kernels", "Compress embedded OpenCL kernels in library binary. Note embed_kernels should be enabled", False),
     BoolVariable("set_soname", "Set the library's soname and shlibversion (requires SCons 2.4 or above)", False),
-    BoolVariable("tracing", "Enable runtime tracing", False),
     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),
@@ -124,7 +124,7 @@ if env['build'] == "embed_only":
     Return()
 
 if env['neon'] and 'x86' in env['arch']:
-    print("Cannot compile Neon for x86")
+    print("Cannot compile Arm® Neon™ for x86")
     Exit(1)
 
 if env['set_soname'] and not version_at_least(SCons.__version__, "2.4"):
@@ -142,8 +142,8 @@ if env['opencl'] and env['embed_kernels'] and env['compress_kernels'] and env['o
     Exit(1)
 
 if not env['exceptions']:
-    if env['opencl'] or env['gles_compute']:
-         print("ERROR: OpenCL and GLES are not supported when building without exceptions. Use opencl=0 gles_compute=0")
+    if env['opencl']:
+         print("ERROR: OpenCL is not supported when building without exceptions. Use opencl=0")
          Exit(1)
 
     env.Append(CPPDEFINES = ['ARM_COMPUTE_EXCEPTIONS_DISABLED'])
@@ -180,10 +180,6 @@ if env['cppthreads']:
     env.Append(CPPDEFINES = [('ARM_COMPUTE_CPP_SCHEDULER', 1)])
 
 if env['openmp']:
-    if 'clang++' in cpp_compiler:
-        print( "Clang does not support OpenMP. Use scheduler=cpp.")
-        Exit(1)
-
     env.Append(CPPDEFINES = [('ARM_COMPUTE_OPENMP_SCHEDULER', 1)])
     env.Append(CXXFLAGS = ['-fopenmp'])
     env.Append(LINKFLAGS = ['-fopenmp'])
@@ -207,7 +203,7 @@ if 'v7a' in env['estate'] and env['estate'] == '64':
 prefix = ""
 if 'v7a' in env['arch']:
     env.Append(CXXFLAGS = ['-march=armv7-a', '-mthumb', '-mfpu=neon'])
-    if env['os'] == 'android' or env['os'] == 'tizen':
+    if (env['os'] == 'android' or env['os'] == 'tizen') and not 'hf' in env['arch']:
         env.Append(CXXFLAGS = ['-mfloat-abi=softfp'])
     else:
         env.Append(CXXFLAGS = ['-mfloat-abi=hard'])
@@ -290,7 +286,7 @@ if not GetOption("help"):
             print("GCC 6.2.1 or newer is required to compile armv8.2-a code")
             Exit(1)
         elif env['arch'] == 'arm64-v8a' and not version_at_least(compiler_ver, '4.9'):
-            print("GCC 4.9 or newer is required to compile Neon code for AArch64")
+            print("GCC 4.9 or newer is required to compile Arm® Neon™ code for AArch64")
             Exit(1)
 
         if version_at_least(compiler_ver, '6.1'):
@@ -345,20 +341,19 @@ if env['os'] == 'linux' and env['arch'] == 'armv7a':
 if env['specs_file'] != "":
     env.Append(LINKFLAGS = ['-specs='+env['specs_file']])
 
+if env['neon']:
+    env.Append(CPPDEFINES = ['ARM_COMPUTE_CPU_ENABLED'])
+
 if env['opencl']:
+    env.Append(CPPDEFINES = ['ARM_COMPUTE_OPENCL_ENABLED'])
     if env['os'] in ['bare_metal'] or env['standalone']:
         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['opencl']:
     if env['embed_kernels']:
         env.Append(CPPDEFINES = ['EMBEDDED_KERNELS'])
     if env['compress_kernels']:
@@ -392,9 +387,6 @@ for dirname in os.listdir("./include"):
 
 Export('version_at_least')
 
-if env['gles_compute'] and env['os'] != 'android':
-    env.Append(CPPPATH = ['#/include/linux'])
-
 SConscript('./SConscript', variant_dir=build_path, duplicate=0)
 
 if env['examples'] and env['exceptions']:

SConstruct

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2017 Arm Limited.
+ * Copyright (c) 2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,19 +21,20 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "arm_compute/core/CL/ICLMultiImage.h"
+#ifndef ARM_COMPUTE_ACL_H_
+#define ARM_COMPUTE_ACL_H_
 
-#include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/ITensor.h"
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
 
-using namespace arm_compute;
+/* Core headers */
+#include "arm_compute/AclEntrypoints.h"
+#include "arm_compute/AclTypes.h"
+#include "arm_compute/AclUtils.h"
+#include "arm_compute/AclVersion.h"
 
-IImage *ICLMultiImage::plane(unsigned int index)
-{
-    return cl_plane(index);
-}
-
-const IImage *ICLMultiImage::plane(unsigned int index) const
-{
-    return cl_plane(index);
+#ifdef __cplusplus
 }
+#endif /* __cplusplus */
+#endif /* ARM_COMPUTE_ACL_H_ */