The Smell of Simulator in the Morning

When one first clones a simulator, it is so fascinating to be at the state where research begins again. It is fresh like napalm in the morning. Just as explosive and unforgiving. Mostly I will be wrangling the obscene amounts of work done by others to hack together a piece of work that will hopefully produce a paper. Either way the napalm will be ignited. the question is always, where do I start. 3 years ago, this was an easy question to answer, but these days, it just feels like more work then the benefit.

New Framework for GPU Task Parallelism

Soon I will release a framework built on top of OpenCL that  supports generic task parallelism and synchronization without the use of persistent threads. I wonder who would want to use such a thing. This framework is best for APU systems except for their poor memory capacity. I should buy a ps4 just so I can get GDDR5 shared across an entire chip so that this framework really screams. Let me know if you want to check it out.

Portability

Trying to understand the delicate nature of OpenCL portability between AMD and NVIDIA GPUs is a very complex subject. One would think that as long as there is no resource difference that code should not consistency seg fault on one machine but not on the other machine. Once again this is a problem with persistent threads (btw, not good for GPUs). Who knows?

Persistent Threads

The concept of persistent threads frustrates me on GPUs.

Welcome to my blog about about GPU programming. I will be sharing the frustrations of developing a programmable interface to GPUs. This work is built on top of OpenCL which is about as easy to debug as it is to climb half-dome in Yosemite 3 days in a row. Sometimes it is nice just to have a place to vent to world about struggles.