- Best Paper Award at FSE (September 2017)
Our paper “Cooperative Kernels: GPU Multitasking for Blocking Algorithms” was selected for a Best Paper Award at the ESEC/FSE conference this week! Here are my co-authors Tyler Sorensen and Hugues Evrard collecting the award:
Tyler presented the paper, and I also presented on a recent IEEE Transactions on Software Engineering paper about candidate-based invariant generation, which was joint work with Adam Betts, Nathan Chong, Pantazis Deligiannis and Jeroen Ketema, all former members of the Multicore group.
8 days ago
- CONCUR tutorial (September 2017)
This week I gave a tutorial at the CONCUR conference, entitled “Forward Progress on GPU Concurrency”. I gave an overview of what it’s like to program using OpenCL, and demonstrated the GPUVerify tool that we have developed at Imperial. I then did some live coding of a global synchronization barrier, illustrating how one needs to use OpenCL’s memory model to get this to work correctly, and showing how even then there is the possibility of deadlock due to lack of forward progress guarantees by GPU schedulers. I then coded up the discovery protocol we proposed in our OOPSLA 2016 paper to allow a portable barrier to be implemented.
It was fun getting my hands dirty with some low-level OpenCL programming! Thanks to everyone who attended the tutorial.
Jeroen Ketema, Tyler Sorensen, John Wickerson and I wrote an overview paper to accompany the tutorial, giving a summary of the work that we and various collaborators have done over the last 6 years in relation to GPU concurrency. Check out the paper.
9 days ago
- Reader (September 2017)
I have been promoted to Reader, the most wonderfully-named British academic rank!
14 days ago
- Dagstuhl seminar on floating-point (August 2017)
We had a really fun seminar at Schloss
Dagstuhl on “Analysis and Synthesis of
which I co-organised with Eva
Rakamaric and Cindy Rubio Gonzalez.
My colleague George Constantinides has written up a really good
synopsis of many of the topics discussed at the
and we’ll in due course put together a Dagstuhl report with more details.
Thanks to all who attended and made it a great event.
16 days ago
- GLFuzz paper at OOPSLA (August 2017)
I’m really excited that our paper, “Automated Testing of Graphics
will appear at the OOPSLA 2017
This is the culmination of a lot of work, joint with Hugues Evrard, Andrei Lascu,
and Paul Thomson. As well as the paper,
check out our series of blog posts on the
and our repository of shader compiler
a month ago
- Vulnerabilities found using GLFuzz fixed in NVIDIA drivers (August 2017)
We’re really pleased that in a recent security
NVIDIA have confirmed that they have fixed an issue that we found with
our GLFuzz tool, as reported in this blog
crediting us for the bug report. NVIDIA summarise the issue as
follows: “NVIDIA GPU Display Driver contains a vulnerability in the
kernel mode layer handler where an incorrect detection and recovery
from an invalid state produced by specific user actions may lead to a
denial of service.” They have also filed the issue as
a month ago
- Paper at ASE (July 2017)
We have had a paper, “Floating-Point Symbolic Execution: A Case Study
in N-version Programming”, accepted at the 32nd IEEE/ACM
International Conference on Automated Software Engineering (ASE
2017). This is joint work with Daniel
Liew and Cristian
Cadar at Imperial, as well as with
collaborators at RWTH Aachen University.
In this work, we conducted a study in conjunction with Aachen, who we
discovered were independently undertaking some similar work to us on
adding support for floating-point symbolic execution to the KLEE tool.
Rather than racing to try to scoop one another, we teamed up to do a
rigorous case study on N-version programming (with N=2), to
investigate the similar and different design decisions made in each
Two by-products of this work are:
We hope these will be valuable resources for other researchers!
2 months ago