Hello, I have two questions:
1. I've been encountering live-locks frequently(4/10) in gem5[O3CPU, X86, FS, ruby]. I restore from a checkpoint created in atomic mode into timing mode and immediately switch to detailed mode. I've gone through the mail archives and the list of bug fixes,I incorporated - changeset 9513, but it didn't work for me. Now I know just enough gem5 to do my work and hence have been hesitant about debugging the issue as it is not consistently reproducible. I am running a micro-benchmark that uses OpenMP and pthreads and it runs fine on my machine(x86_64), it runs on gem5 in atomic mode, also sort of 6/10 times in detailed. Sometimes just doing random(apparently irrational) stuff like replacing the executable on disk, or re-creating a checkpoint and restoring from it, changing the tick at which I switch cpus solves the problem. Has anybody ever faced a similar issue? NOTE: I am using disk image and kernel from ( http://www.cs.utexas.edu/~cart/parsec_m5/). I'm using changeset 9073. I've made some changes in ruby, but I am sure those changes are not causing these issues. I have verified this. -------------------------------------------------------------------------------------------------------------------------------------------------------- 2. What kind of behaviour can I expect from gem5[O3CPU, X86, FS, ruby]. in the following scenario. a. run my program on gem5, with cache sizes L1d, L1i = 32K and L2 = 128K b. run a slightly modified version of the program on gem5, with cache sizes L1d, L1i = 16K and L2 = 64K c. run my original program(used in step a) on gem5, with cache sizes L1d, L1i = 32K and L2 = 128K If every other parameter is kept constant, can I expect similar results from steps a and c ? If not, is it to do with dynamic scheduling in the linux kernel, or is it to do with things like branch prediction or something that is within the bounds of gem5? Any help would be deeply appreciated. Thanks, Ajit
_______________________________________________ gem5-users mailing list gem5-users@gem5.org http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
