What’s New in AMD APP SDK v2.8
Key features supported in SDK 2.8 andthe Catalyst 12.8 drivers include:
• Support for Direct3D 11 sharing.
• Performance improvement in the OpenCL GPU run-time.
• Performance improvement in the OpenCL CPU run-time.
• Support added for 64-bit atomics.
• Asynchronous dispatch support added for Radeon HD™ 7000-series devices.
• Support for LLVM 3.0.
• Support added for bus-addressable memory.
• 2D max image size increased to 16k*16k.
• UVD performance improvements.
• Samples are now included for BOLT, Aparapi, and C++ AMP.
Bolt data parallel primitives C++ template library, preview release for OpenCL™
• Preview includes the following templates:
– Transform reduce
The CodeXL beta integrated hybrid computing (HC) tool suite is recommended for use with AMD
APP SDK and can be downloaded here. Features include:
• Unified profiling and debugging on AMD CPUs/APUs/GPUs.
• New user interface with unified look-and-feel, available as a stand-alone Windows or Linux
application, and as an extension for Microsoft®Visual Studio®2010 on Windows.
• GPU Debugger.
– GPU debugging includes OpenCL™ kernel and API calls.
– Single-step debugging and breakpoints on a single GPU (running display).
– Trace API call history (combined C/C++ and OpenCL).
– Multi-watch view of memory, variable values/types.
• CPU Profiler.
– Profiling suite to identify, investigate, and tune application performance on AMD CPUs.
– Precise application hot-spot analysis with time-based, event-based, and instruction-based sampling.
– Call graph profiling, which is a butterfly view offunction calls.
– OpenCL™ source profiling for CPU.
• GPU Profiler.
– OpenCL™ and DirectCompute™ profiler for AMD Radeon™ GPUs.
– API trace, kernel occupancy, and hot spots analysis.
– Comprehensive visualization: timeline and summary views of host, kernel, and data
– Supports GPU performance counters.
• Static Analyzer.
– Static analysis of OpenCL or IL kernels for AMD Radeon™ GPUs, without having to run
the application on actual hardware.
– Estimate accurate performance of OpenCL kernels.
– View statistics and disassembly of the generated hardware kernel.
APP Math Libraries (APPML) 1.8 new features include:
– Public interfaces for transpose function.
– Completed support for all BLAS L2 and 3 functions by adding:
– Modified error return codes.