Compute Library v23.08

Author: Jenkins

Android.bp

@@ -51,6 +51,7 @@ opencl_srcs = [
         "src/core/CL/cl_kernels/common/instance_normalization.cl",
         "src/core/CL/cl_kernels/common/l2_normalize.cl",
         "src/core/CL/cl_kernels/common/mat_mul.cl",
+        "src/core/CL/cl_kernels/common/mat_mul_mmul.cl",
         "src/core/CL/cl_kernels/common/mat_mul_quantized.cl",
         "src/core/CL/cl_kernels/common/mean_stddev_normalization.cl",
         "src/core/CL/cl_kernels/common/memset.cl",
@@ -165,7 +166,9 @@ arm_compute_library_defaults {
             "-DARM_COMPUTE_ENABLE_NEON",
             "-Wno-unused-parameter",
             "-DNO_DOT_IN_TOOLCHAIN",
-            "-Wno-implicit-fallthrough"
+            "-Wno-implicit-fallthrough",
+            "-fPIC",
+            "-DACL_INTERNAL_TEST_CKW_IN_DF"
     ],
     rtti: true,
 }
@@ -176,6 +179,8 @@ cc_library_static {
     proprietary: true,
     local_include_dirs: ["build/android-arm64v8a/src/core",
                          "build/android-arm64v8a/src/core/CL",
+                         "compute_kernel_writer/prototype/include",
+                         "compute_kernel_writer/prototype",
                          "src/core/common",
                          "src/core/helpers",
                          "src/core/NEON/kernels/arm_gemm",
@@ -185,6 +190,15 @@ cc_library_static {
                          "src/cpu/kernels/assembly"],
     export_include_dirs: [".", "./include"],
     srcs: [
+        "compute_kernel_writer/prototype/src/Kernel.cpp",
+        "compute_kernel_writer/prototype/src/KernelArgument.cpp",
+        "compute_kernel_writer/prototype/src/KernelWriter.cpp",
+        "compute_kernel_writer/prototype/src/OperandBase.cpp",
+        "compute_kernel_writer/prototype/src/TensorInfo.cpp",
+        "compute_kernel_writer/prototype/src/TensorOperand.cpp",
+        "compute_kernel_writer/prototype/src/TensorTileSampler.cpp",
+        "compute_kernel_writer/prototype/src/TileInfo.cpp",
+        "compute_kernel_writer/prototype/src/TileOperand.cpp",
         "src/c/AclContext.cpp",
         "src/c/AclOperator.cpp",
         "src/c/AclQueue.cpp",
@@ -313,9 +327,9 @@ cc_library_static {
         "src/core/NEON/kernels/arm_conv/depthwise/depthwise_strategies_common.cpp",
         "src/core/NEON/kernels/arm_conv/depthwise/depthwise_u8q.cpp",
         "src/core/NEON/kernels/arm_conv/depthwise/depthwise_u8s8u8q.cpp",
-        "src/core/NEON/kernels/arm_conv/depthwise/interleaves/8b_mla.cpp",
         "src/core/NEON/kernels/arm_conv/depthwise/interleaves/generic.cpp",
         "src/core/NEON/kernels/arm_conv/depthwise/interleaves/generic_quantized_dot_product.cpp",
+        "src/core/NEON/kernels/arm_conv/depthwise/premultiply.cpp",
         "src/core/NEON/kernels/arm_conv/pooling/kernels/cpp_nhwc_1x1_stride_any_depthfirst/generic.cpp",
         "src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp",
         "src/core/NEON/kernels/arm_conv/pooling/pooling_fp32.cpp",
@@ -363,7 +377,6 @@ cc_library_static {
         "src/core/NEON/kernels/convolution/winograd/output_transforms/arm_fp32_4x4_3x3.cpp",
         "src/core/NEON/kernels/convolution/winograd/output_transforms_fp16.cpp",
         "src/core/NEON/kernels/convolution/winograd/output_transforms_fp32.cpp",
-        "src/core/NEON/kernels/convolution/winograd/padding.cpp",
         "src/core/NEON/kernels/convolution/winograd/weight_transforms/arm_fp32_2x2_3x3.cpp",
         "src/core/NEON/kernels/convolution/winograd/weight_transforms/arm_fp32_2x2_5x5.cpp",
         "src/core/NEON/kernels/convolution/winograd/weight_transforms/arm_fp32_4x4_3x3.cpp",
@@ -384,8 +397,14 @@ cc_library_static {
         "src/core/Version.cpp",
         "src/core/helpers/SoftmaxHelpers.cpp",
         "src/core/helpers/WindowHelpers.cpp",
+        "src/core/utils/ActivationFunctionUtils.cpp",
         "src/core/utils/AssemblyUtils.cpp",
+        "src/core/utils/DataLayoutUtils.cpp",
+        "src/core/utils/DataTypeUtils.cpp",
+        "src/core/utils/FormatUtils.cpp",
+        "src/core/utils/InterpolationPolicyUtils.cpp",
         "src/core/utils/ScaleUtils.cpp",
+        "src/core/utils/StringUtils.cpp",
         "src/core/utils/helpers/fft.cpp",
         "src/core/utils/helpers/tensor_transform.cpp",
         "src/core/utils/io/FileHandler.cpp",
@@ -604,6 +623,7 @@ cc_library_static {
         "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/runtime/gpu/cl/ckw_driver/GpuCkwKernelArgumentsHelpers.cpp",
         "src/dynamic_fusion/sketch/attributes/CastAttributes.cpp",
         "src/dynamic_fusion/sketch/attributes/ClampAttributes.cpp",
         "src/dynamic_fusion/sketch/attributes/Conv2dAttributes.cpp",
@@ -620,6 +640,16 @@ cc_library_static {
         "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/ckw_driver/GpuCkwComponentArgument.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/GpuCkwDriver.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/GpuCkwKernelWriter.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/GpuCkwScopedKernelWriter.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/GpuCkwVariableTable.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwActivation.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwCast.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwDirectConv2d.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwElementwiseBinary.cpp",
+        "src/dynamic_fusion/sketch/gpu/ckw_driver/components/GpuCkwStore.cpp",
         "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentActivation.cpp",
         "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentCast.cpp",
         "src/dynamic_fusion/sketch/gpu/components/cl/ClComponentDepthwiseConv2d.cpp",
@@ -700,6 +730,7 @@ cc_library_static {
         "src/gpu/cl/kernels/ClIndirectConv2dKernel.cpp",
         "src/gpu/cl/kernels/ClMatMulLowpNativeKernel.cpp",
         "src/gpu/cl/kernels/ClMatMulNativeKernel.cpp",
+        "src/gpu/cl/kernels/ClMatMulNativeMMULKernel.cpp",
         "src/gpu/cl/kernels/ClMulKernel.cpp",
         "src/gpu/cl/kernels/ClPermuteKernel.cpp",
         "src/gpu/cl/kernels/ClPool2dKernel.cpp",
@@ -1020,7 +1051,6 @@ cc_library_static {
             srcs: [
                 "src/core/NEON/kernels/arm_conv/depthwise/interleaves/a64_s8q_3x3_dot.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/interleaves/a64_u8q_3x3_dot.cpp",
-                "src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_8b_mla.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_s8q_3x3_dot.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/interleaves/sve_u8q_3x3_dot.cpp",
                 "src/core/NEON/kernels/arm_conv/depthwise/kernels/a64_fp16_nhwc_3x3_s1_output2x2_mla_depthfirst/generic_direct.cpp",

BUILD.bazel

@@ -382,6 +382,7 @@ cc_library(
             }),
     includes = [
         "arm_compute/runtime",
+        "src/core/NEON/kernels/arm_gemm",
         "src/core/NEON/kernels/assembly",
         "src/core/NEON/kernels/convolution/common",
         "src/core/NEON/kernels/convolution/winograd",

CMakeLists.txt

@@ -28,7 +28,7 @@ cmake_minimum_required(VERSION 3.13 FATAL_ERROR)
 list(APPEND CMAKE_MESSAGE_CONTEXT ArmCompute)
 project(
   ArmCompute
-  VERSION 31.0.1
+  VERSION 32.0.0
   DESCRIPTION
     "The Arm Compute Library is a collection of low-level machine learning functions optimized for Arm® Cortex®-A CPU and Arm® Mali™ GPU architectures"
   LANGUAGES C CXX ASM)
@@ -197,6 +197,7 @@ target_include_directories(
          ${CMAKE_CURRENT_SOURCE_DIR}
   PRIVATE src
           src/cpu/kernels/assembly
+          src/core/NEON/kernels/arm_gemm
           src/core/NEON/kernels/assembly
           src/core/NEON/kernels/convolution/common
           src/core/NEON/kernels/arm_conv/depthwise
@@ -223,6 +224,7 @@ target_include_directories(
          ${CMAKE_CURRENT_SOURCE_DIR}
   PRIVATE src
           src/cpu/kernels/assembly
+          src/core/NEON/kernels/arm_gemm
           src/core/NEON/kernels/assembly
           src/core/NEON/kernels/convolution/common
           src/core/NEON/kernels/arm_conv/depthwise

CONTRIBUTING.md

@@ -1,4 +1,4 @@
-Please read https://arm-software.github.io/ComputeLibrary/v23.02.1/contribution_guidelines.xhtml
+Please read https://arm-software.github.io/ComputeLibrary/latest/contribution_guidelines.xhtml
 
 Here on github we only publish a snapshot of the main development branch for each release, that's the reason why we don't accept pull requests.