Symbolic evaluation of GPU software

Add to your list(s) Download to your calendar using vCal

• David Huw Jones (Imperial College)
• Monday 21 March 2011, 11:30-12:00
• Level 9 Mahanakorn Lab., EE Dept..

If you have a question about this talk, please contact Grigorios Mingas.

In order to better predict the performance and power consumption of GPU software, we need to understand the dynamics of its execution cycle. Of interest is the sequence and size of global memory access requests and the resulting L1 and L2 cache utilization. Other factors of importance include thread divergence and shared memory bank conflicts.

In this talk I will present preliminary work towards determining all of these factors by means of symbolically evaluating the GPU software. With these we can determine how these factors change with time, thread id and block id. Also we can determine these factors as a function of statistical distributions.

This talk is part of the CAS Talks series.

This talk is included in these lists:

• Andrea Picciau's list
• CAS Talks
• Circuits and Systems Group: Internal Seminars
• Level 9 Mahanakorn Lab., EE Dept.

Note that ex-directory lists are not shown.