Compute Library v22.11

ARM-software · Nov 18, 2022 · 1b3192e · 1b3192e
1 parent aabef6c
commit 1b3192e
Show file tree

Hide file tree

Showing 227 changed files with 12,775 additions and 3,223 deletions.
diff --git a/Android.bp b/Android.bp
@@ -447,6 +447,7 @@ cc_library_static {
         "src/cpu/kernels/CpuWinogradConv2dKernel.cpp",
         "src/cpu/kernels/activation/generic/neon/fp16.cpp",
         "src/cpu/kernels/activation/generic/neon/fp32.cpp",
+        "src/cpu/kernels/activation/generic/neon/lut.cpp",
         "src/cpu/kernels/activation/generic/neon/qasymm8.cpp",
         "src/cpu/kernels/activation/generic/neon/qasymm8_signed.cpp",
         "src/cpu/kernels/activation/generic/neon/qsymm16.cpp",
@@ -519,6 +520,7 @@ cc_library_static {
         "src/cpu/kernels/meanstddevnorm/generic/neon/fp16.cpp",
         "src/cpu/kernels/meanstddevnorm/generic/neon/fp32.cpp",
         "src/cpu/kernels/meanstddevnorm/generic/neon/impl.cpp",
+        "src/cpu/kernels/meanstddevnorm/generic/neon/qasymm8.cpp",
         "src/cpu/kernels/pool2d/neon/fp16.cpp",
         "src/cpu/kernels/pool2d/neon/fp32.cpp",
         "src/cpu/kernels/pool2d/neon/nchw/all.cpp",
@@ -590,6 +592,24 @@ cc_library_static {
         "src/cpu/operators/CpuTranspose.cpp",
         "src/cpu/operators/CpuWinogradConv2d.cpp",
         "src/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp",
+        "src/dynamic_fusion/runtime/gpu/cl/ClKernelRuntime.cpp",
+        "src/dynamic_fusion/runtime/gpu/cl/ClWorkloadRuntime.cpp",
+        "src/dynamic_fusion/sketch/OperatorAttributes.cpp",
+        "src/dynamic_fusion/sketch/gpu/GpuKernelArgument.cpp",
+        "src/dynamic_fusion/sketch/gpu/GpuKernelComponentGraph.cpp",
+        "src/dynamic_fusion/sketch/gpu/GpuKernelComponentGroup.cpp",
+        "src/dynamic_fusion/sketch/gpu/GpuKernelComponentStream.cpp",
+        "src/dynamic_fusion/sketch/gpu/GpuLogicalKernel.cpp",
+        "src/dynamic_fusion/sketch/gpu/GpuOperatorGroup.cpp",
+        "src/dynamic_fusion/sketch/gpu/GpuWorkloadContext.cpp",
+        "src/dynamic_fusion/sketch/gpu/GpuWorkloadSketch.cpp",
+        "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDirectConv2d.cpp",
+        "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentStore.cpp",
+        "src/dynamic_fusion/sketch/gpu/operators/GpuConv2d.cpp",
+        "src/dynamic_fusion/sketch/gpu/template_writer/GpuKernelVariableTable.cpp",
+        "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateDirectConv2d.cpp",
+        "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateStore.cpp",
+        "src/dynamic_fusion/sketch/gpu/template_writer/cl/ClTemplateWriter.cpp",
         "src/gpu/cl/ClContext.cpp",
         "src/gpu/cl/ClKernelLibrary.cpp",
         "src/gpu/cl/ClQueue.cpp",

diff --git a/README.md b/README.md
@@ -10,7 +10,7 @@
  <img src="https://raw.githubusercontent.com/ARM-software/ComputeLibrary/gh-pages/ACL_logo.png"/><br><br>
 </div>
 
-# Compute Library ![](https://img.shields.io/badge/latest_release-22.08-green)
+# Compute Library ![](https://img.shields.io/badge/latest_release-22.11-green)
 
 
 The Compute Library is a collection of low-level machine learning functions optimized for Arm® Cortex®-A, Arm® Neoverse® and Arm® Mali™ GPUs architectures.<br>
@@ -30,15 +30,15 @@ Key Features:
 
 <br>
 
-| Repository  | Link |
-| ----------- | ----------- |
-| Release     | https://github.com/arm-software/ComputeLibrary       |
-| Development | https://review.mlplatform.org/#/admin/projects/ml/ComputeLibrary        |
+| Repository  | Link                                                             |
+| ----------- | ---------------------------------------------------------------- |
+| Release     | https://github.com/arm-software/ComputeLibrary                   |
+| Development | https://review.mlplatform.org/#/admin/projects/ml/ComputeLibrary |
 
 <br>
 
 ## Documentation
-[![Documentation](https://img.shields.io/badge/documentation-22.08-green)](https://arm-software.github.io/ComputeLibrary/latest)
+[![Documentation](https://img.shields.io/badge/documentation-22.11-green)](https://arm-software.github.io/ComputeLibrary/latest)
 
 > Note: The documentation includes the reference API, changelogs, build guide, contribution guide, errata, etc.
 
@@ -49,25 +49,26 @@ All the binaries can be downloaded from [here](https://github.com/ARM-software/C
 
 <br>
 
-| Platform    | Operating System | Release archive (Download) |
-| ----------- | ----------- | ----------- |
-| Raspberry Pi 4 | Linux 32bit | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-armv7a-neon.tar.gz) |
-| Raspberry Pi 4 | Linux 64bit | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-neon.tar.gz) |
-| Odroid N2 | Linux 64bit | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-neon-cl.tar.gz) |
-| HiKey960 | Linux 64bit | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-neon-cl.tar.gz) |
+| Platform       | Operating System | Release archive (Download) |
+| -------------- | ---------------- | -------------------------- |
+| Raspberry Pi 4 | Linux 32bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-armv7a-neon.tar.gz) |
+| Raspberry Pi 4 | Linux 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-neon.tar.gz) |
+| Odroid N2      | Linux 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-neon-cl.tar.gz) |
+| HiKey960       | Linux 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-neon-cl.tar.gz) |
 
 <br>
 
-| Architecture    | Operating System | Release archive (Download) |
-| ----------- | ----------- | ----------- |
-| armv7 | Linux | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-armv7a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-armv7a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-armv7a-neon-cl.tar.gz) |
-| arm64-v8a | Android | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-android-arm64-v8a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-android-arm64-v8a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-android-arm64-v8a-neon-cl.tar.gz) |
-| arm64-v8a | Linux | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8a-neon-cl.tar.gz) |
-| arm64-v8.2-a | Android | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-android-arm64-v8.2-a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-android-arm64-v8.2-a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-android-arm64-v8.2-a-neon-cl.tar.gz) |
-| arm64-v8.2-a | Linux | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8.2-a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8.2-a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.08/arm_compute-v22.08-bin-linux-arm64-v8.2-a-neon-cl.tar.gz) |
+| Architecture | Operating System | Release archive (Download) |
+| ------------ | ---------------- | -------------------------- |
+| armv7        | Linux            | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-armv7a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-armv7a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-armv7a-neon-cl.tar.gz) |
+| arm64-v8a    | Android          | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-android-arm64-v8a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-android-arm64-v8a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-android-arm64-v8a-neon-cl.tar.gz) |
+| arm64-v8a    | Linux            | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8a-neon-cl.tar.gz) |
+| arm64-v8.2-a | Android          | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-android-arm64-v8.2-a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-android-arm64-v8.2-a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-android-arm64-v8.2-a-neon-cl.tar.gz) |
+| arm64-v8.2-a | Linux            | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8.2-a-neon.tar.gz) [![](https://img.shields.io/badge/build-opencl-blue)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8.2-a-cl.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v22.11/arm_compute-v22.11-bin-linux-arm64-v8.2-a-neon-cl.tar.gz) |
 
 <br>
-Please refer to the following link for more pre-built binaries: [![](https://img.shields.io/badge/v22.08-bins-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/tag/v22.08)
+
+Please refer to the following link for more pre-built binaries: [![](https://img.shields.io/badge/v22.11-bins-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/tag/v22.11)
 
 Pre-build binaries are generated with the following security / good coding practices related flags:
 > -Wall, -Wextra, -Wformat=2, -Winit-self, -Wstrict-overflow=2, -Wswitch-default, -Woverloaded-virtual, -Wformat-security, -Wctor-dtor-privacy, -Wsign-promo, -Weffc++, -pedantic, -fstack-protector-strong

diff --git a/SConscript b/SConscript
@@ -31,8 +31,8 @@ import zlib
 import json
 import codecs
 
-VERSION = "v22.08"
-LIBRARY_VERSION_MAJOR = 28
+VERSION = "v22.11"
+LIBRARY_VERSION_MAJOR = 29
 LIBRARY_VERSION_MINOR =  0
 LIBRARY_VERSION_PATCH =  0
 SONAME_VERSION = str(LIBRARY_VERSION_MAJOR) + "." + str(LIBRARY_VERSION_MINOR) + "." + str(LIBRARY_VERSION_PATCH)
@@ -124,6 +124,12 @@ def build_library(name, build_env, sources, static=False, libs=[]):
         else:
             obj = cloned_build_env.SharedLibrary(name, source=sources, LIBS = arm_compute_env["LIBS"] + libs)
 
+    if env['mapfile']:
+        if not 'windows' in env['os'] and not 'macos' in env['os']:
+            cloned_build_env['LINKFLAGS'].append('"-Wl,-Map='+ name + '.map"')
+        else:
+            cloned_build_env['LINKFLAGS'].append('-Wl,-map,' + name + '.map')
+
     obj = install_lib(obj)
     build_env.Default(obj)
     return obj

diff --git a/SConstruct b/SConstruct
@@ -127,6 +127,7 @@ vars.AddVariables(
             └── Neon\n""", "", PathVariable.PathAccept),
     BoolVariable("experimental_dynamic_fusion", "Build the experimental dynamic fusion files", False),
     BoolVariable("experimental_fixed_format_kernels", "Enable fixed format kernels for GEMM", False),
+    BoolVariable("mapfile", "Generate a map file", False),
     ListVariable("custom_options", "Custom options that can be used to turn on/off features", "none", ["disable_mmla_fp"]),
     ListVariable("data_type_support", "Enable a list of data types to support", "all", ["qasymm8", "qasymm8_signed", "qsymm16", "fp16", "fp32", "integer"]),
     ListVariable("data_layout_support", "Enable a list of data layout to support", "all", ["nhwc", "nchw"]),
@@ -312,7 +313,7 @@ else: # NONE "multi_isa" builds
         elif 'armv8.6-a-sve' == env['arch']:
             env.Append(CXXFLAGS = ['-march=armv8.6-a+sve'])
         elif 'armv8.6-a' == env['arch']:
-            env.Append(CXXFLAGS = ['-march=armv8.6-a'])
+            env.Append(CXXFLAGS = ['-march=armv8.6-a+fp16'])
 
         env.Append(CPPDEFINES = ['ARM_COMPUTE_ENABLE_I8MM', 'ARM_COMPUTE_ENABLE_BF16','ARM_COMPUTE_ENABLE_FP16'])
         if "disable_mmla_fp" not in env['custom_options']:

diff --git a/arm_compute/core/CL/CLCompileContext.h b/arm_compute/core/CL/CLCompileContext.h
@@ -304,6 +304,18 @@ class CLCompileContext final
     */
     bool is_wbsm_supported() const;
 
+    /** Return the DDK version. If the DDK version cannot be detected, return -1.
+     *
+     * @return The DDK version.
+     */
+    int32_t get_ddk_version() const;
+
+    /** Return the Gpu target of the associated device
+     *
+     * @return GPUTarget
+     */
+    GPUTarget get_gpu_target() const;
+
 private:
     /** Load program and its dependencies.
      *

diff --git a/arm_compute/core/CL/CLHelpers.h b/arm_compute/core/CL/CLHelpers.h
@@ -242,13 +242,13 @@ bool get_wbsm_support_info(const cl::Device &device);
  */
 void set_wbsm(cl::Kernel &kernel, cl_int wbsm_hint);
 
-/* Helper function to check if we can export the weights to cl_image
+/* Helper function to check if we can export the tensor to cl_image
  *
- * @param[in] tensor Weights tensor
+ * @param[in] input tensor
  *
- * @return true if we can export the weights to cl_image
+ * @return true if we can export the tensor to cl_image
  */
-bool export_weights_to_cl_image(const ITensorInfo *tensor);
+bool export_to_cl_image(const ITensorInfo *tensor);
 
 /* Helper function to force unroll with pragma when any of the input values (iterations) are greater than @ref max_manual_loop_unrolling
  *

diff --git a/arm_compute/core/ITensorInfo.h b/arm_compute/core/ITensorInfo.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2021 Arm Limited.
+ * Copyright (c) 2016-2022 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -41,6 +41,11 @@ class ITensorInfo : public misc::ICloneable<ITensorInfo>
 {
 public:
     using TensorDimsState = std::vector<int>;
+    /** An id that uniquely identifies an ITensorInfo within some domain (e.g. a workload)
+     */
+    using Id = int32_t;
+    /** An invalid tensor id within a domain */
+    static constexpr Id invalid_tensor_id = -1;
     /** Get the value representing dynamic dimension state
      *
      * @return Value representing dynamic dimension state
@@ -280,7 +285,20 @@ class ITensorInfo : public misc::ICloneable<ITensorInfo>
     * @return A DataLayout containing the layout data information.
     */
     virtual DataLayout data_layout() const = 0;
-
+    /** Get the workload tensor id of the tensor.
+    *
+    * @return Workload tensor id of the tensor
+    */
+    virtual Id id() const = 0;
+    /** Set the tensor id
+    */
+    virtual ITensorInfo &set_id(ITensorInfo::Id id) = 0;
+    /** Check if the tensor id is valid
+     */
+    bool has_valid_id() const
+    {
+        return id() != invalid_tensor_id;
+    }
     /** If infos are broadcast compatible tensor info's, return the broadcasted shape and the intersection of
      * the broadcasted valid regions of the tensors.
      *

diff --git a/arm_compute/core/KernelDescriptors.h b/arm_compute/core/KernelDescriptors.h
@@ -106,6 +106,7 @@ struct DWCComputeKernelInfo
 {
     unsigned int n0{ 0 };                             /**< Number of columns processed by each thread */
     unsigned int m0{ 0 };                             /**< Number of rows processed by each thread */
+    bool         export_input_to_cl_image{ false };   /**< Export input to cl_image */
     bool         export_weights_to_cl_image{ false }; /**< Export the weights to cl_image */
 };