At last week's meeting, we decided that we wanted to figure out the static number of instructions for each benchmark program. We did this so that we can have a better comparison and understanding of the benchmark programs. This is especially handy when we look at the stalls due to the instruction cache or instruction duplication, because we want to have the original and static numnber of instructions the program executes. To do this, Akhil and I worked on analyzing and tweaking the benchmark programs, and making a parser to get the number of instructions. First, we looked at the...

While working on setting a unified cache configuration, I noticed something annoying about our GPGPU-Sim output. If we look at the part where we get the L1 cache data, we can notice that we don't get a detailed breakdown of how L1 cache in each core performed like we do with the data cache. For example: L1I_cache: L1I_total_cache_accesses = 2074321 L1I_total_cache_misses = 1745 L1I_total_cache_miss_rate = 0.0008 L1I_total_cache_pending_hits = 0 L1I_total_cache_reservation_fails = 0 L1D_cache: L1D_cache_core[0]: Access = 544, Miss = 272, Miss_rate = 0.500, Pending_hits = 0, Reservation_fails = 0 L1D_cache_core[1]: Access = 544, Miss = 272, Miss_rate = 0.500, Pending_hits...

Last Update: 2/18/2014 9:00PM This blog post will keep a table of all the benchmarks & machine configurations that have been simulated and saved on Charlab machine. Update this table once the benchmark is run & the appropriate output files are moved into the results directory. The output file of the benchmark will be saved in /data/charlab/gpgpu-sim/RESULTS/ directory. Machine BFS CP LIB LPS NN NQU RAY STO GTX 480 default [4:64:4] X X X X X X X X Quadro FX5600 default X X X X Error X X X Quadro FX5800 default Tesla C2050 default _____________________________ _______ _______ _______...

Start time: around 12:10pm In attendance: Prof Spjut, Fabiha, Erik, Paul, Akhil, Sami, Donghyeon __ Paul: -Paul thinks he is oversimplifying his project -Testing with more instructions is the ideal -Instructions should forwards branch as much as possible; if there's a backwards branch, there need to be some code to make it forwards branch again -Getting raspberry pi to work is the hard part, so he is fine; getting to that ASAP would be good -Prof Harris' implementation does have a bug; testing should catch it (Paul should test Prof Harris' thing with his version) __ Sami: -Sami was not...

This week Dong-hyeon and I set out to fiddle with the config files. The config file is long and seemed a little intimidating to me. This was partly because the file was filled with terms I did not understand, magic numbers and was insufficiently commented. However, looking at the comments in the output file from running the benchmarks actually told us much more about the structure of the configuration! Last semester we spoke about employing the'naive method.' This method being simulating and testing a instruction cache configuration in configuration in which all the cores share the same instruction cache as...

Eric spoke about his research in proposing a new data caching scheme. About 8 of the benchmarks work. We can make do with these. We have Cuda 5.5 and Cuda 4.0 as GPGPU Sim cannot use Cuda > 4.9. Fahiba introduced Sami to the project and Sami has a python parser that they plan to use to parse the output files that were generated by the gp-gpu simulations. Fahiba raised an interesting point about only generating part of the output file so that we don't have to parse the ENTIRE output file. Look at Prof Spjut's blog post on 'screen'...

Since the projects for this group are using linux machines through remote connections, I thought it might be useful to give a brief tutorial on a tool called screen that is available on Linux. Screen can be used as a way to manage a shell connection and allow the connection to persist through network problems allowing you to resume work where you left off. For instance, if you name a screen session gpgpusim when working from your desktop at home, you can then connect to the same screen session using that name on your laptop during a group meeting in...

Here are the steps on how to setup and run GPGPU-Sim on charlab machine. Before we begin, here is the summary of current gpgpu-sim setup on Charlab machine. (charlab.eng.hmc.edu) GPGPU-sim is located on /data/charlab/gpgpu-sim. This is where gpgpu-sim will be located at all times. The version of CUDA that is compatible with the current build of GPGPU-sim is CUDA-4.4. The CUDA-4.4 Toolkit is located in /data/charlab/cuda. For running GPGPU-sim, we need always refer to this particular version of cuda. This is different from the CUDA-5.5 located in /usr/local/cuda/. The NVIDIA Computing SDK v4.0 is located in /data/charlab/NVIDIA_GPU_Computing_SDK. This SDK is...