Allinea Software has announced the availability of Allinea DDT 2.6.1, which supports versions 3.1 and 3.2 of the Nvidia Cuda Toolkit and has been beta tested for some time.
This latest version adds support for multiple GPU devices within the same node, in addition to the existing capabilities for debugging clusters of GPU systems. It also supports the latest hardware, with some of the newer features such as device function call and stepping into and out of nested functions on the Fermi hardware now possible.
DDT's scalable debugging interface includes components that simplify the task of debugging multiple kernels and multiple processes – such as the parallel stack view which intuitively consolidates the display of host and GPU thread stacks, allowing the user to identify rogue threads or processes quickly.
Sanford Russell, general manager, GPU computing software at Nvidia, explains: ‘As the market for GPU computing continues to accelerate, we are seeing increasing demand for cluster-class debuggers. Allinea DDT 2.6.1 responds to this need and delivers an innovative solution for developers looking to harness the massively parallel architecture of GPUs in their applications.’
Allinea DDT also includes support for debugging memory errors in Cuda kernels, such as reading or writing beyond the end of allocated blocks of memory. A ‘stop on kernel launch’ feature allows users to begin debug kernels automatically as soon as they are launched, in addition to the existing support for source code breakpoints.
