Sitos, Have you tried making your own cross compiler tool-chain? Try these two projects: http://freshmeat.net/projects/crosstool-ng/ http://kegel.com/crosstool/
It might be the one on the M5 site does not have exactly everything you need. Also try compiling the OpenMP apps with the -static flag if they haven't been already. To allow M5 to simulate more than 2GB of main memory, the machine you run on M5 needs to have more than 2GB of main memory otherwise it'll fail. As a side note the x86 support in M5 is far, far from complete. Stick with the Alpha or SPARC support which are known to be stable. Geoff -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Sitos Lin Sent: Thursday, January 22, 2009 11:33 AM To: M5 users mailing list Subject: [m5-users] Run SpecOMP on M5 Dear Isuru: Thank you for the useful link. I have build the patch and the correspondent cross-compiler. However, I encounter several problems. The goal of my work is to run SpecOMP on M5. For now, I tried several approaches, and all of them do not work. 1) I tried to use the newest M5 and the pre-compiled cross-compiler from the M5's website. The cross-compiler can compile the SpecOMP without error. But the compiled binaries can not run on neither ALPHA_SE nor ALPHA_FS mode. In the ALPHA_SE mode, some threaded related system calls are not emulated and the threaded application can not be executed correctly. In the ALPHA_FS mode, full system simulation allow the binaries run with the system call. However, I always get Killed or Segmentation fault when I try to execute the programs. I guess that is because the memory is not enough for the SpecOMP benchmarks. The memory requirement of the SpecOMP is 2GB memory for the medium benchmarking. However, I can not configure the memory size to 2GB in ALPHA_FS mode. Therefore, I didn't run threaded SpecOMP correctly for now. On the other hand, the non-threaded version (compiled without -fopenmp flag) seems can be executed in ALPHA_SE mode normally. Therefore, I think the compilation has no problem. However, I have no idea about how to execute them. 2) I tried to use the newest M5 and run the X86_SE or X86_FS mode with the binaries compiled from conventional x86 gcc compiler. The status is even worse.... 3) I tried to use the link you provide. However, there are several problems: (a) The modified cross-compiler in the M5 patch for multi-threaded application seems only has C compiler without Fortran compiler. However, most of the benchmarks in SpecOMP is written in Fortran. (b) When I want to use it to run the benchmark written in C in SpecOMP, the cross-compiler does not support OpenMP. The gcc compilers support OpenMP since gcc-4.2, and the cross-compiler is based on gcc-4.1. The cross-compiler can not link the APIs of OpenMP correctly. I have tried to search the source code of OpenMP library. Because I think that I can compile the library of OpenMP by the cross-compiler and link it to the SpecOMP applications. However, I didn't find any one. (c) I tried to compile the application with the pre-compiled cross-compiler from M5 website instead of that provided from the M5 multi-thread patch. Then, I execute the binaries in the multi-thread M5 patch. However, the error message shows up to say that the M5 simulator can not support TLS mode (maybe thread level speculation), and ask me to compile the binary with non-TLS toolchain. However, I have no idea about how to build a non-TLS toolchain of gcc-4.2 which supports OpenMP. (a non-TLS cross-compiler for gcc-4.1 is available, e.g. the one from the multi-thread M5 patch) I don't know how can I achieve my goal. Can you give me some advise? best regards, sitos 2009/1/21 Isuru Herath <[email protected]>: > Dear Sitos, > > Check the following link. > > http://www.cs.virginia.edu/~jm6dg/fractal/smp.htm > and this forum post. > http://comments.gmane.org/gmane.comp.emulators.m5.users/3953 > > It might be helpful. to be honest, still I havent tried this. Hopefully within > coming two weeks will do. > > regards, > Isuru > > > > > _______________________________________________ > m5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/m5-users > _______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users No virus found in this incoming message. Checked by AVG - http://www.avg.com Version: 8.0.176 / Virus Database: 270.10.12/1909 - Release Date: 1/22/2009 7:08 AM _______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
