Arm Performance Studio vs MicroProfile

Arm Performance Studio is a free profiling suite for applications running on devices with Arm CPUs and GPUs. Streamline records CPU, GPU, and application activity from a device, and Performance Advisor turns those captures into reports highlighting potential problems and optimization advice. Frame Advisor examines API calls and rendering data for a problem frame and provides geometry metrics. Mali Offline Compiler estimates shader and compute-kernel performance across Arm GPUs, including register usage, cycle cost, and stage-specific recommendations. The suite also includes RenderDoc for Arm GPUs, an Arm fork of the graphics API debugger with support for Arm GPU extensions and Android features. Tools can run without a graphical interface in continuous-integration workflows and export machine-readable data for custom reporting; Unity and Godot integrations can add application markers or counters. The product describes profiling Android and Linux applications and lists Android 11 or later for Streamline, Performance Advisor, Frame Advisor, and RenderDoc for Arm GPUs. Graphics profiling supports Mali GPUs from the Mali-T700 series onward, but not Mali-400 or Mali-T600 products. Community support is provided through Arm’s Gaming, Graphics, and VR forum.

MicroProfile is a free, embeddable C++ profiler supplied in a few files for adding profiling to a project. Developers mark code with timer macros and include microprofile.cpp. Its live view is opened in Chrome by visiting the profiler host on port 1338, and it can capture the last 30 frames by default. Captures are self-contained webpages that can be saved and shared; profile data can also be written to HTML or CSV. GPU timers support OpenGL, D3D11, D3D12, and Vulkan. Counters appear in live and capture views, and timeline markers cover longer-running unique timers. A detailed view can compare two captures, but not captures made with earlier versions. Experimental dynamic instrumentation supports Intel x86-64 on Windows, macOS, and Linux, with MICROPROFILE_DYNAMIC_INSTRUMENT_ENABLE and distorm required; Linux builds also need -rdynamic. Default per-thread profiling buffers allocate 2 MB, and GPU buffers default to 128 KB each. The project is MIT-licensed and lists Windows, macOS, Linux, and unspecified console support.