We are pleased to announce that a new GPU profiling library called GPU Performance API (or GPUPerfAPI) is available for download at: http://developer.amd.com/gpu/GPUPerfAPI/
Collect performance data about your OpenGL, DirectX 11, or DirectX 10 application running on ATI Radeon HD 2000 series or newer graphics cards and OpenCL applications on ATI Radeon HD 4000 or 5000 series cards. GPUPerfAPI’s flexible interface allows you to profile entire frames, rendering effects, or individual draw calls to identify expensive operations and where you might be able to squeeze additional performance out of our powerful GPUs. GPUPerfAPI is currently used by GPU PerfStudio 2, ATI Stream Profiler, and Graphic Remedy’s gDEBugger, and is now available for you to integrate into your own applications.
GPUPerfAPICL exposes between 13 and 20 counters which can be used in OpenCL applications to:
* Determine the number of ALU, Fetch, and Write instructions executed by a kernel.
* Detect stalls due to the fetch or write units.
* Identify L1 cache usage and LDS memory accesses.
* Identify if memory accesses are causing a decrease in performance.
GPUPerfAPIGL / GPUPerfAPIDX11 / GPUPerfAPIDX10 expose between 70 and 113 counters which can be used in graphics applications to:
* Identify expensive draw operations.
* Determine busiest pipeline stages.
* Obtain the number of ALU and texture instructions executed by a shader.
* Track the number of vertices, primitives, and pixels along the pipeline.
* Identify how many primitives are being clipped or culled from the scene.
* Detect if slow or uncompressed texture formats are being used.
* Gain insight into the actual filtering being performed and texture types being used.
* Understand depth and stencil test operations and performance.
* Identify stalls along the pipeline.
* Much more…
Please post your feedback here.