feat: Add NETopKV function.

Android.bp

@@ -462,6 +462,7 @@ cc_library_static {
         "src/cpu/kernels/CpuScatterKernel.cpp",
         "src/cpu/kernels/CpuSoftmaxKernel.cpp",
         "src/cpu/kernels/CpuSubKernel.cpp",
+        "src/cpu/kernels/CpuTopKVKernel.cpp",
         "src/cpu/kernels/CpuTransposeKernel.cpp",
         "src/cpu/kernels/CpuWeightsReshapeKernel.cpp",
         "src/cpu/kernels/CpuWinogradConv2dKernel.cpp",
@@ -611,6 +612,10 @@ cc_library_static {
         "src/cpu/kernels/sub/neon/qasymm8.cpp",
         "src/cpu/kernels/sub/neon/qasymm8_signed.cpp",
         "src/cpu/kernels/sub/neon/qsymm16.cpp",
+        "src/cpu/kernels/topkv/generic/neon/fp16.cpp",
+        "src/cpu/kernels/topkv/generic/neon/fp32.cpp",
+        "src/cpu/kernels/topkv/generic/neon/qasymm8.cpp",
+        "src/cpu/kernels/topkv/generic/neon/qasymm8_signed.cpp",
         "src/cpu/operators/CpuActivation.cpp",
         "src/cpu/operators/CpuAdd.cpp",
         "src/cpu/operators/CpuAddMulAdd.cpp",
@@ -649,6 +654,7 @@ cc_library_static {
         "src/cpu/operators/CpuScatter.cpp",
         "src/cpu/operators/CpuSoftmax.cpp",
         "src/cpu/operators/CpuSub.cpp",
+        "src/cpu/operators/CpuTopKV.cpp",
         "src/cpu/operators/CpuTranspose.cpp",
         "src/cpu/operators/CpuWinogradConv2d.cpp",
         "src/cpu/operators/internal/CpuGemmAssemblyDispatch.cpp",
@@ -978,6 +984,7 @@ cc_library_static {
         "src/runtime/NEON/functions/NEStackLayer.cpp",
         "src/runtime/NEON/functions/NEStridedSlice.cpp",
         "src/runtime/NEON/functions/NETile.cpp",
+        "src/runtime/NEON/functions/NETopKV.cpp",
         "src/runtime/NEON/functions/NETranspose.cpp",
         "src/runtime/NEON/functions/NEUnstack.cpp",
         "src/runtime/NEON/functions/NEWinogradConvolutionLayer.cpp",

arm_compute/runtime/NEON/NEFunctions.h

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2025 Arm Limited.
+ * Copyright (c) 2016-2026 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -112,6 +112,7 @@
 #include "arm_compute/runtime/NEON/functions/NEStackLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEStridedSlice.h"
 #include "arm_compute/runtime/NEON/functions/NETile.h"
+#include "arm_compute/runtime/NEON/functions/NETopKV.h"
 #include "arm_compute/runtime/NEON/functions/NETranspose.h"
 #include "arm_compute/runtime/NEON/functions/NEUnstack.h"
 #include "arm_compute/runtime/NEON/functions/NEWinogradConvolutionLayer.h"

arm_compute/runtime/NEON/functions/NETopKV.h

@@ -0,0 +1,87 @@
+/*
+ * Copyright (c) 2026 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_ARM_COMPUTE_RUNTIME_NEON_FUNCTIONS_NETOPKV_H
+#define ACL_ARM_COMPUTE_RUNTIME_NEON_FUNCTIONS_NETOPKV_H
+
+/** @file
+ * @publicapi
+ */
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/runtime/IFunction.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+// Forward declarations
+class ITensor;
+class ITensorInfo;
+
+/** Basic function to run cpu::kernels::CpuActivationKernel
+ *
+ * @note The function simulates an activation layer with the specified activation function.
+ */
+class NETopKV : public IFunction
+{
+public:
+    /** Constructor */
+    NETopKV();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NETopKV(const NETopKV &) = delete;
+    /** Default move constructor */
+    NETopKV(NETopKV &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NETopKV &operator=(const NETopKV &) = delete;
+    /** Default move assignment operator */
+    NETopKV &operator=(NETopKV &&);
+    /** Destructor */
+    ~NETopKV();
+    /** Set the input and output of the kernel.
+     *
+     * @param[in]  predictions A batch_size x classes tensor. Data types supported: F16/F32/QASYMM8/QASYMM8_SIGNED
+     * @param[in]  targets     A batch_size 1D tensor of class ids. Data types supported: U32
+     * @param[out] output      Computed precision at @p k as a bool 1D tensor. Data types supported: U8
+     * @param[in]  k           Number of top elements to look at for computing precision.
+     */
+    void configure(const ITensor *predictions, const ITensor *targets, ITensor *output, const unsigned int k);
+
+    /** Static function to check if given info will lead to a valid configuration.
+     *
+     * Similar to @ref NETopKV::configure()
+     *
+     * @return a status
+     */
+    static Status
+    validate(const ITensorInfo *predictions, const ITensorInfo *targets, ITensorInfo *output, const unsigned int k);
+
+    // Inherited methods overridden
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
+};
+} // namespace arm_compute
+#endif // ACL_ARM_COMPUTE_RUNTIME_NEON_FUNCTIONS_NETOPKV_H

filelist.json

@@ -2453,6 +2453,26 @@
           ]
         }
       },
+      "TopKV": {
+        "files": {
+          "common": [
+            "src/cpu/kernels/CpuTopKVKernel.cpp",
+            "src/cpu/operators/CpuTopKV.cpp",
+            "src/runtime/NEON/functions/NETopKV.cpp"
+          ],
+          "neon": {
+            "fp16": [ "src/cpu/kernels/topkv/generic/neon/fp16.cpp" ],
+            "fp32": [ "src/cpu/kernels/topkv/generic/neon/fp32.cpp" ],
+            "qasymm8": [
+              "src/cpu/kernels/topkv/generic/neon/qasymm8.cpp"
+            ],
+            "qasymm8_signed": [
+              "src/cpu/kernels/topkv/generic/neon/qasymm8_signed.cpp"
+            ]
+          }
+
+        }
+      },
       "Transpose": {
         "files": {
           "common": [

src/BUILD.bazel

@@ -738,6 +738,7 @@ filegroup(
 	"cpu/kernels/CpuScatterKernel.cpp",
 	"cpu/kernels/CpuSoftmaxKernel.cpp",
 	"cpu/kernels/CpuSubKernel.cpp",
+	"cpu/kernels/CpuTopKVKernel.cpp",
 	"cpu/kernels/CpuTransposeKernel.cpp",
 	"cpu/kernels/CpuWeightsReshapeKernel.cpp",
 	"cpu/kernels/CpuWinogradConv2dKernel.cpp",
@@ -848,6 +849,9 @@ filegroup(
 	"cpu/kernels/sub/neon/qasymm8.cpp",
 	"cpu/kernels/sub/neon/qasymm8_signed.cpp",
 	"cpu/kernels/sub/neon/qsymm16.cpp",
+	"cpu/kernels/topkv/generic/neon/fp32.cpp",
+	"cpu/kernels/topkv/generic/neon/qasymm8.cpp",
+	"cpu/kernels/topkv/generic/neon/qasymm8_signed.cpp",
 	"cpu/operators/CpuActivation.cpp",
 	"cpu/operators/CpuAdd.cpp",
 	"cpu/operators/CpuAddMulAdd.cpp",
@@ -886,6 +890,7 @@ filegroup(
 	"cpu/operators/CpuScatter.cpp",
 	"cpu/operators/CpuSoftmax.cpp",
 	"cpu/operators/CpuSub.cpp",
+	"cpu/operators/CpuTopKV.cpp",
 	"cpu/operators/CpuTranspose.cpp",
 	"cpu/operators/CpuWinogradConv2d.cpp",
 	"cpu/operators/internal/CpuGemmAssemblyDispatch.cpp",
@@ -994,6 +999,7 @@ filegroup(
 	"runtime/NEON/functions/NEStackLayer.cpp",
 	"runtime/NEON/functions/NEStridedSlice.cpp",
 	"runtime/NEON/functions/NETile.cpp",
+	"runtime/NEON/functions/NETopKV.cpp",
 	"runtime/NEON/functions/NETranspose.cpp",
 	"runtime/NEON/functions/NEUnstack.cpp",
 	"runtime/NEON/functions/NEWinogradConvolutionLayer.cpp",
@@ -1109,7 +1115,8 @@ filegroup(
 	"cpu/kernels/scatter/generic/neon/fp16.cpp",
 	"cpu/kernels/select/generic/neon/fp16.cpp",
 	"cpu/kernels/softmax/generic/neon/fp16.cpp",
-	"cpu/kernels/sub/neon/fp16.cpp"]  +
+	"cpu/kernels/sub/neon/fp16.cpp",
+	"cpu/kernels/topkv/generic/neon/fp16.cpp"]  +
     glob(["**/*.h",
     "**/*.hpp",
     "**/*.inl"]),

src/CMakeLists.txt

@@ -732,6 +732,7 @@ target_sources(
 	cpu/kernels/CpuScatterKernel.cpp
 	cpu/kernels/CpuSoftmaxKernel.cpp
 	cpu/kernels/CpuSubKernel.cpp
+	cpu/kernels/CpuTopKVKernel.cpp
 	cpu/kernels/CpuTransposeKernel.cpp
 	cpu/kernels/CpuWeightsReshapeKernel.cpp
 	cpu/kernels/CpuWinogradConv2dKernel.cpp
@@ -842,6 +843,9 @@ target_sources(
 	cpu/kernels/sub/neon/qasymm8.cpp
 	cpu/kernels/sub/neon/qasymm8_signed.cpp
 	cpu/kernels/sub/neon/qsymm16.cpp
+	cpu/kernels/topkv/generic/neon/fp32.cpp
+	cpu/kernels/topkv/generic/neon/qasymm8.cpp
+	cpu/kernels/topkv/generic/neon/qasymm8_signed.cpp
 	cpu/operators/CpuActivation.cpp
 	cpu/operators/CpuAdd.cpp
 	cpu/operators/CpuAddMulAdd.cpp
@@ -880,6 +884,7 @@ target_sources(
 	cpu/operators/CpuScatter.cpp
 	cpu/operators/CpuSoftmax.cpp
 	cpu/operators/CpuSub.cpp
+	cpu/operators/CpuTopKV.cpp
 	cpu/operators/CpuTranspose.cpp
 	cpu/operators/CpuWinogradConv2d.cpp
 	cpu/operators/internal/CpuGemmAssemblyDispatch.cpp
@@ -988,6 +993,7 @@ target_sources(
 	runtime/NEON/functions/NEStackLayer.cpp
 	runtime/NEON/functions/NEStridedSlice.cpp
 	runtime/NEON/functions/NETile.cpp
+	runtime/NEON/functions/NETopKV.cpp
 	runtime/NEON/functions/NETranspose.cpp
 	runtime/NEON/functions/NEUnstack.cpp
 	runtime/NEON/functions/NEWinogradConvolutionLayer.cpp
@@ -1109,4 +1115,5 @@ target_sources(
 	cpu/kernels/select/generic/neon/fp16.cpp
 	cpu/kernels/softmax/generic/neon/fp16.cpp
 	cpu/kernels/sub/neon/fp16.cpp
+	cpu/kernels/topkv/generic/neon/fp16.cpp
 )