Compute Library v52.3.0

Author: Jenkins

Android.bp

@@ -1000,6 +1000,7 @@ cc_library_static {
         "src/runtime/experimental/operators/CpuDepthwiseConv2d.cpp",
         "src/runtime/experimental/operators/CpuDequantize.cpp",
         "src/runtime/experimental/operators/CpuElementwise.cpp",
+        "src/runtime/experimental/operators/CpuFullyConnected.cpp",
         "src/runtime/experimental/operators/CpuGEMMLowp.cpp",
         "src/runtime/experimental/operators/CpuGemm.cpp",
         "src/runtime/experimental/operators/CpuGemmConv2d.cpp",

CMakeLists.txt

@@ -24,7 +24,7 @@ cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
 
 project(
   ArmCompute
-  VERSION 52.2.0
+  VERSION 52.3.0
   DESCRIPTION
   "The Compute Library is a collection of low-level machine learning functions \
    optimized for Arm® Cortex®-A, Arm® Neoverse™ CPU and Arm® Mali™ GPU \

README.md

@@ -14,7 +14,7 @@
  <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-52.2.0-green)
+# Compute Library ![](https://img.shields.io/badge/latest_release-52.3.0-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>
@@ -42,7 +42,7 @@ Key Features:
 <br>
 
 ## Documentation
-[![Documentation](https://img.shields.io/badge/documentation-52.2.0-green)](https://artificial-intelligence.sites.arm.com/computelibrary/v52.2.0/index.xhtml)
+[![Documentation](https://img.shields.io/badge/documentation-52.3.0-green)](https://artificial-intelligence.sites.arm.com/computelibrary/v52.3.0/index.xhtml)
 
 > Note: The documentation includes the reference API, changelogs, build guide, contribution guide, errata, etc.
 
@@ -55,22 +55,22 @@ All the binaries can be downloaded from [here](https://github.com/ARM-software/C
 
 | 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/v52.2.0/arm_compute-v52.2.0-linux-armv7a-cpu-bin.tar.gz) |
-| Raspberry Pi 4 | Linux® 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-linux-aarch64-cpu-bin.tar.gz) |
-| Odroid N2      | Linux® 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-linux-aarch64-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-linux-aarch64-cpu-gpu-bin.tar.gz) |
-| HiKey960       | Linux® 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-linux-aarch64-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-linux-aarch64-cpu-gpu-bin.tar.gz) |
+| Raspberry Pi 4 | Linux® 32bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-armv7a-cpu-bin.tar.gz) |
+| Raspberry Pi 4 | Linux® 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-aarch64-cpu-bin.tar.gz) |
+| Odroid N2      | Linux® 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-aarch64-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-aarch64-cpu-gpu-bin.tar.gz) |
+| HiKey960       | Linux® 64bit      | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-aarch64-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-aarch64-cpu-gpu-bin.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/v52.2.0/arm_compute-v52.2.0-linux-armv7a-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-linux-armv7a-cpu-gpu-bin.tar.gz) |
-| arm64-v8a    | Android™          | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-android-aarch64-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-android-aarch64-cpu-gpu-bin.tar.gz) |
-| arm64-v8a    | Linux®            | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-linux-aarch64-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.2.0/arm_compute-v52.2.0-linux-aarch64-cpu-gpu-bin.tar.gz) |
+| armv7        | Linux®            | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-armv7a-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-armv7a-cpu-gpu-bin.tar.gz) |
+| arm64-v8a    | Android™          | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-android-aarch64-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-android-aarch64-cpu-gpu-bin.tar.gz) |
+| arm64-v8a    | Linux®            | [![](https://img.shields.io/badge/build-neon-orange)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-aarch64-cpu-bin.tar.gz) [![](https://img.shields.io/badge/build-neon+cl-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/download/v52.3.0/arm_compute-v52.3.0-linux-aarch64-cpu-gpu-bin.tar.gz) |
 
 <br>
 
-Please refer to the following link for more pre-built binaries: [![](https://img.shields.io/badge/v52.2.0-bins-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/tag/v52.2.0)
+Please refer to the following link for more pre-built binaries: [![](https://img.shields.io/badge/v52.3.0-bins-yellowgreen)](https://github.com/ARM-software/ComputeLibrary/releases/tag/v52.3.0)
 
 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
@@ -113,13 +113,13 @@ Pre-build binaries are generated with the following security / good coding pract
 
 ## Experimental builds
 
-**⚠ Important** Bazel and CMake builds are experimental CPU only builds, please see the [documentation](https://artificial-intelligence.sites.arm.com/computelibrary/v52.2.0/how_to_build.xhtml) for more details.
+**⚠ Important** Bazel and CMake builds are experimental CPU only builds, please see the [documentation](https://artificial-intelligence.sites.arm.com/computelibrary/v52.3.0/how_to_build.xhtml) for more details.
 
 <br>
 
 ## How to contribute
 
-Contributions to the Compute Library are more than welcome. If you are interested on contributing, please have a look at our [how to contribute guidelines](https://artificial-intelligence.sites.arm.com/computelibrary/v52.2.0/contribution_guidelines.xhtml).
+Contributions to the Compute Library are more than welcome. If you are interested on contributing, please have a look at our [how to contribute guidelines](https://artificial-intelligence.sites.arm.com/computelibrary/v52.3.0/contribution_guidelines.xhtml).
 
 ### Developer Certificate of Origin (DCO)
 Before the Compute Library accepts your contribution, you need to certify its origin and give us your permission. To manage this process we use the Developer Certificate of Origin (DCO) V1.1 (https://developercertificate.org/)

SConscript

@@ -33,9 +33,9 @@ import codecs
 import platform
 import SCons
 
-VERSION = "v52.2.0"
+VERSION = "v52.3.0"
 LIBRARY_VERSION_MAJOR = 52
-LIBRARY_VERSION_MINOR = 2
+LIBRARY_VERSION_MINOR = 3
 LIBRARY_VERSION_PATCH = 0
 SONAME_VERSION = str(LIBRARY_VERSION_MAJOR) + "." + str(LIBRARY_VERSION_MINOR) + "." + str(LIBRARY_VERSION_PATCH)
 

arm_compute/core/Dimensions.h

@@ -167,15 +167,17 @@ class Dimensions
 
         if (last > (first + 1))
         {
+            auto it_first = _id.begin() + first;
+            auto it_last  = _id.begin() + last;
             // Collapse dimensions into the first
-            _id[first] = std::accumulate(&_id[first], &_id[last], 1, std::multiplies<T>());
+            _id[first] = std::accumulate(it_first, it_last, 1, std::multiplies<T>());
             // Shift the remaining dimensions down
-            std::copy(&_id[last], &_id[_num_dimensions], &_id[first + 1]);
+            std::copy(it_last, _id.begin() + _num_dimensions, it_first + 1);
             // Reduce the number of dimensions
             const size_t old_num_dimensions = _num_dimensions;
             _num_dimensions -= last - first - 1;
             // Fill the now empty dimensions with zero
-            std::fill(&_id[_num_dimensions], &_id[old_num_dimensions], 0);
+            std::fill(_id.begin() + _num_dimensions, _id.begin() + old_num_dimensions, 0);
         }
     }
 

arm_compute/core/Error.h

@@ -24,10 +24,6 @@
 #ifndef ACL_ARM_COMPUTE_CORE_ERROR_H
 #define ACL_ARM_COMPUTE_CORE_ERROR_H
 
-/** @file
- * @publicapi
- */
-
 #include <array>
 #include <string>
 

arm_compute/core/QuantizationInfo.h

@@ -394,15 +394,19 @@ inline int8_t quantize_qsymm8(float value, const QuantizationInfo &qinfo)
 
 /** Quantize a value given a 8-bit symmetric per channel quantization scheme
  *
- * @param[in] value      Value to quantize
- * @param[in] qinfo      Quantization information to use for quantizing
- * @param[in] channel_id channel index into the scale vector of quantization info
+ * @param[in] value           Value to quantize
+ * @param[in] qinfo           Quantization information to use for quantizing
+ * @param[in] channel_id      channel index into the scale vector of quantization info
+ * @param[in] rounding_policy (Optional) Rounding policy to use. Default: nearest up
  *
  * @return Quantized value
  */
-inline int8_t quantize_qsymm8_per_channel(float value, const QuantizationInfo &qinfo, size_t channel_id = 0)
+inline int8_t quantize_qsymm8_per_channel(float                   value,
+                                          const QuantizationInfo &qinfo,
+                                          size_t                  channel_id      = 0,
+                                          RoundingPolicy          rounding_policy = RoundingPolicy::TO_NEAREST_UP)
 {
-    int quantized = arm_compute::round(value / qinfo.scale()[channel_id], RoundingPolicy::TO_NEAREST_UP);
+    int quantized = arm_compute::round(value / qinfo.scale()[channel_id], rounding_policy);
     quantized     = std::max(-128, std::min(quantized, 127));
     return quantized;
 }

arm_compute/runtime/NEON/functions/NEQuantizationLayer.h

@@ -60,21 +60,21 @@ class NEQuantizationLayer : public IFunction
      * - All
      *
      * Valid data type configurations:
-     * |src                |dst                                    |
-     * |:------------------|:--------------------------------------|
-     * |QASYMM8            |QASYMM8, QASYMM8_SIGNED, QASYMM16      |
-     * |QASYMM8_SIGNED     |QASYMM8, QASYMM8_SIGNED, QASYMM16      |
-     * |F16                |QASYMM8, QASYMM8_SIGNED, QASYMM16      |
-     * |F32                |QASYMM8, QASYMM8_SIGNED, QASYMM16      |
+     * |src                |dst                                                        |
+     * |:------------------|:----------------------------------------------------------|
+     * |QASYMM8            |QASYMM8, QASYMM8_SIGNED, QASYMM16                          |
+     * |QASYMM8_SIGNED     |QASYMM8, QASYMM8_SIGNED, QASYMM16                          |
+     * |F16                |QASYMM8, QASYMM8_SIGNED, QASYMM16                          |
+     * |F32                |QASYMM8, QASYMM8_SIGNED, QSYMM8_PER_CHANNEL, QASYMM16      |
      *
      * @param[in]  input  Source tensor. The dimensions over the third will be interpreted as batches. Data types supported: QASYMM8/QASYMM8_SIGNED/F32/F16.
-     * @param[out] output Destination tensor with the same dimensions of input. Data types supported: QASYMM8/QASYMM8_SIGNED/QASYMM16
+     * @param[out] output Destination tensor with the same dimensions of input. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/QASYMM16
      */
     void configure(const ITensor *input, ITensor *output);
     /** Static function to check if given info will lead to a valid configuration of @ref NEQuantizationLayer
      *
      * @param[in] input  Input tensor info. The dimensions over the third will be interpreted as batches. Data types supported: QASYMM8/QASYMM8_SIGNED/F32/F16.
-     * @param[in] output Output tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/QASYMM16
+     * @param[in] output Output tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/QASYMM16
      *
      * @return a status
      */