feat: add profiling tracepoints to CPU kernel implementations (Part 1)

Instrument key CPU kernel entry points in src/cpu/kernels/* with
tracepoints to enable lightweight runtime profiling. These tracepoints
leverage the ACL_PROFILE macros and form the basis for collecting
execution timing and behavior metrics. This is the first step in
integrating end-to-end profiling support.

Partially Resolves: COMPMID-8330
Signed-off-by: Walid Ben Romdhane <[email protected]>
Change-Id: Ifc7e3d6f1d049f7ad051ee8f58b8934013456dc8
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/14773
Benchmark: Arm Jenkins <[email protected]>
Tested-by: Arm Jenkins <[email protected]>
Comments-Addressed: Arm Jenkins <[email protected]>
Reviewed-by: Dongsung Kim <[email protected]>

Author: walidbr

src/common/utils/profile/acl_profile.h

@@ -37,9 +37,8 @@
 #define ARM_COMPUTE_PROF_CAT_SCHEDULER "SCHEDULER"
 
 // Define ACL profile levels
-#define ARM_COMPUTE_PROF_L0 0
-#define ARM_COMPUTE_PROF_L1 1
-#define ARM_COMPUTE_PROF_L2 2
+#define ARM_COMPUTE_PROF_LVL_CPU 0
+#define ARM_COMPUTE_PROF_LVL_GPU 1
 
 #define ARM_COMPUTE_TRACE_EVENT(category, level, name) \
     (void)category;                                    \

src/cpu/kernels/CpuActivationKernel.cpp

@@ -27,6 +27,7 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Utils.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/common/Registrars.h"
 #include "src/core/CPP/Validate.h"
 #include "src/core/helpers/AutoConfiguration.h"
@@ -199,6 +200,7 @@ void init_lut(ActivationLayerInfo::ActivationFunction act_func,
 
 void CpuActivationKernel::configure(const ITensorInfo *src, ITensorInfo *dst, ActivationLayerInfo activation_info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuActivationKernel::configure");
     ARM_COMPUTE_UNUSED(dst);
     ARM_COMPUTE_ERROR_ON_NULLPTR(src);
     ARM_COMPUTE_ERROR_THROW_ON(CpuActivationKernel::validate(src, dst, activation_info));
@@ -248,9 +250,9 @@ void CpuActivationKernel::configure(const ITensorInfo *src, ITensorInfo *dst, Ac
 Status
 CpuActivationKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuActivationKernel::validate");
     ARM_COMPUTE_UNUSED(act_info);
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, act_info));
-
     return Status{};
 }
 
@@ -264,6 +266,7 @@ size_t CpuActivationKernel::get_mws(const CPUInfo &platform, size_t thread_count
 
 void CpuActivationKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuActivationKernel::run_op");
     // Early exit on disabled activation
     if (!_act_info.enabled())
     {

src/cpu/kernels/CpuAddKernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021-2022, 2024 Arm Limited.
+ * Copyright (c) 2021-2022, 2024-2025 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -27,6 +27,7 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Validate.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/common/Registrars.h"
 #include "src/core/CPP/Validate.h"
 #include "src/core/helpers/AutoConfiguration.h"
@@ -157,6 +158,7 @@ validate_arguments(const ITensorInfo &src0, const ITensorInfo &src1, const ITens
 
 void CpuAddKernel::configure(const ITensorInfo *src0, const ITensorInfo *src1, ITensorInfo *dst, ConvertPolicy policy)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuAddKernel::configure");
     ARM_COMPUTE_ERROR_ON_NULLPTR(src0, src1, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(*src0, *src1, *dst, policy));
 
@@ -191,6 +193,7 @@ void CpuAddKernel::configure(const ITensorInfo *src0, const ITensorInfo *src1, I
 Status
 CpuAddKernel::validate(const ITensorInfo *src0, const ITensorInfo *src1, const ITensorInfo *dst, ConvertPolicy policy)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuAddKernel::validate");
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src0, src1, dst);
 
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(*src0, *src1, *dst, policy));
@@ -200,6 +203,7 @@ CpuAddKernel::validate(const ITensorInfo *src0, const ITensorInfo *src1, const I
 
 void CpuAddKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuAddKernel::run_op");
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);

src/cpu/kernels/CpuAddMulAddKernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2023 Arm Limited.
+ * Copyright (c) 2023, 2025 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -27,6 +27,7 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Validate.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/common/Registrars.h"
 #include "src/core/CPP/Validate.h"
 #include "src/core/helpers/AutoConfiguration.h"
@@ -127,6 +128,7 @@ void CpuAddMulAddKernel::configure(const ITensorInfo         *input1,
                                    ConvertPolicy              policy,
                                    const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuAddMulAddKernel::configure");
     ARM_COMPUTE_UNUSED(bn_mul, bn_add, input2);
     ARM_COMPUTE_ERROR_ON_NULLPTR(input1, input2, bn_add, bn_mul, final_output);
     ARM_COMPUTE_ERROR_THROW_ON(
@@ -167,6 +169,7 @@ Status CpuAddMulAddKernel::validate(const ITensorInfo         *input1,
                                     ConvertPolicy              policy,
                                     const ActivationLayerInfo &act_info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuAddMulAddKernel::validate");
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input1, input2, bn_mul, bn_add, final_output);
 
     ARM_COMPUTE_RETURN_ON_ERROR(
@@ -177,6 +180,7 @@ Status CpuAddMulAddKernel::validate(const ITensorInfo         *input1,
 
 void CpuAddMulAddKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuAddMulAddKernel::run_op");
     ARM_COMPUTE_UNUSED(info);
 
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);

src/cpu/kernels/CpuCastKernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2024 Arm Limited.
+ * Copyright (c) 2016-2025 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,6 +29,7 @@
 #include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Validate.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/common/Registrars.h"
 #include "src/core/CPP/Validate.h"
 #include "src/core/helpers/AutoConfiguration.h"
@@ -162,6 +163,7 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, Conver
 
 void CpuCastKernel::configure(const ITensorInfo *src, ITensorInfo *dst, ConvertPolicy policy)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuCastKernel::configure");
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
 
     // Auto initialize dst shape if not initialized (We can only auto-configure the shape, datatype must be given)
@@ -179,6 +181,7 @@ void CpuCastKernel::configure(const ITensorInfo *src, ITensorInfo *dst, ConvertP
 
 Status CpuCastKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, ConvertPolicy policy)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuCastKernel::validate");
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, policy));
     return Status{};
 }
@@ -270,6 +273,7 @@ convert64(Iterator &src, Iterator &dst, const Window &win, int window_start_x, i
 
 void CpuCastKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuCastKernel::run_op");
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);

src/cpu/kernels/CpuCol2ImKernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2021 Arm Limited.
+ * Copyright (c) 2017-2021, 2025 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -32,6 +32,7 @@
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/core/Validate.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/helpers/AutoConfiguration.h"
 #include "src/core/helpers/WindowHelpers.h"
 
@@ -64,6 +65,7 @@ Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const
 
 void CpuCol2ImKernel::configure(const ITensorInfo *src, ITensorInfo *dst, const Size2D &convolved_dims)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuCol2ImKernel::configure");
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, convolved_dims));
 
@@ -81,12 +83,14 @@ void CpuCol2ImKernel::configure(const ITensorInfo *src, ITensorInfo *dst, const
 
 Status CpuCol2ImKernel::validate(const ITensorInfo *src, const ITensorInfo *output, const Size2D &convolved_dims)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuCol2ImKernel::validate");
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, output, convolved_dims));
     return Status{};
 }
 
 void CpuCol2ImKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuCol2ImKernel::run_op");
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);

src/cpu/kernels/CpuConcatenateBatchKernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2025 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -31,6 +31,7 @@
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/core/Window.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/helpers/AutoConfiguration.h"
 #include "src/core/helpers/WindowHelpers.h"
 #include "src/core/NEON/NEAsymm.h"
@@ -158,6 +159,7 @@ Status validate_arguments(const ITensorInfo *src, unsigned int batch_offset, con
 
 void CpuConcatenateBatchKernel::configure(const ITensorInfo *src, unsigned int batch_offset, ITensorInfo *dst)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateBatchKernel::configure");
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, batch_offset, dst));
 
@@ -195,12 +197,14 @@ Status CpuConcatenateBatchKernel::validate(const arm_compute::ITensorInfo *src,
                                            unsigned int                    batch_offset,
                                            const arm_compute::ITensorInfo *dst)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateBatchKernel::validate");
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, batch_offset, dst));
     return Status{};
 }
 
 void CpuConcatenateBatchKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateBatchKernel::run_op");
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);

src/cpu/kernels/CpuConcatenateDepthKernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2021 Arm Limited.
+ * Copyright (c) 2017-2021, 2025 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -31,6 +31,7 @@
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/core/Window.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/helpers/AutoConfiguration.h"
 #include "src/core/helpers/WindowHelpers.h"
 #include "src/core/NEON/NEAsymm.h"
@@ -160,6 +161,7 @@ Status validate_arguments(const ITensorInfo *input, unsigned int depth_offset, c
 
 void CpuConcatenateDepthKernel::configure(const ITensorInfo *src, unsigned int depth_offset, ITensorInfo *dst)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateDepthKernel::configure");
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, depth_offset, dst));
 
@@ -193,12 +195,14 @@ Status CpuConcatenateDepthKernel::validate(const arm_compute::ITensorInfo *src,
                                            unsigned int                    depth_offset,
                                            const arm_compute::ITensorInfo *dst)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateDepthKernel::validate");
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, depth_offset, dst));
     return Status{};
 }
 
 void CpuConcatenateDepthKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateDepthKernel::run_op");
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);

src/cpu/kernels/CpuConcatenateHeightKernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021 Arm Limited.
+ * Copyright (c) 2019-2021, 2025 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -31,6 +31,7 @@
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/core/Window.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/helpers/AutoConfiguration.h"
 #include "src/core/helpers/WindowHelpers.h"
 #include "src/core/NEON/NEAsymm.h"
@@ -65,6 +66,8 @@ Status validate_arguments(const ITensorInfo *src, unsigned int height_offset, co
 
 void CpuConcatenateHeightKernel::configure(const ITensorInfo *src, unsigned int height_offset, ITensorInfo *dst)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU,
+                            "CpuConcatenateHeightKernel::configure");
     ARM_COMPUTE_UNUSED(src);
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, height_offset, dst));
@@ -78,12 +81,14 @@ void CpuConcatenateHeightKernel::configure(const ITensorInfo *src, unsigned int
 
 Status CpuConcatenateHeightKernel::validate(const ITensorInfo *src, unsigned int height_offset, const ITensorInfo *dst)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateHeightKernel::validate");
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, height_offset, dst));
     return Status{};
 }
 
 void CpuConcatenateHeightKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateHeightKernel::run_op");
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);

src/cpu/kernels/CpuConcatenateWidthKernel.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2021, 2023 Arm Limited.
+ * Copyright (c) 2018-2021, 2023, 2025 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,6 +28,7 @@
 #include "arm_compute/core/Steps.h"
 #include "arm_compute/core/Validate.h"
 
+#include "src/common/utils/profile/acl_profile.h"
 #include "src/core/helpers/WindowHelpers.h"
 #include "src/core/NEON/NEAsymm.h"
 
@@ -58,6 +59,7 @@ Status validate_arguments(const ITensorInfo *src, unsigned int width_offset, con
 
 void CpuConcatenateWidthKernel::configure(const ITensorInfo *src, unsigned int width_offset, ITensorInfo *dst)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateWidthKernel::configure");
     ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
     ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, width_offset, dst));
     ARM_COMPUTE_UNUSED(dst);
@@ -72,12 +74,14 @@ void CpuConcatenateWidthKernel::configure(const ITensorInfo *src, unsigned int w
 
 Status CpuConcatenateWidthKernel::validate(const ITensorInfo *src, unsigned int width_offset, const ITensorInfo *dst)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateWidthKernel::validate");
     ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, width_offset, dst));
     return Status{};
 }
 
 void CpuConcatenateWidthKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
+    ARM_COMPUTE_TRACE_EVENT(ARM_COMPUTE_PROF_CAT_CPU, ARM_COMPUTE_PROF_LVL_CPU, "CpuConcatenateWidthKernel::run_op");
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICpuKernel::window(), window);