I will document my progress here, perhaps experts here can add their advice and thoughts.
I went ahead and recompiled using this website as a guide: http://www.alphalinux.org/docs/perf.html (I am using the precompiled alpha cross compiler on the m5 website) I tried the flags: -mcpu=EV5 (didnt work) -mcpu=21164 (m5 gave illegal instruction error) -mno-max (disables max instr set, gave a segmentation fault of the benchmark, not in m5) The flag that worked worked: -mcpu=21064 (a variant of EV4 without extra instruction sets ) Progress so far: The problem is fixed, but ferret has yet to finish compiling, hopefully no more problems. Thanks, EF On Tue, Jul 21, 2009 at 11:06 PM, Ali Saidi <[email protected]> wrote: > > If you do implement the instruction a quick google search turns up the > following: > > http://mattst88.com/svn/listing.php?repname=alpha_mmintrin&path=%2Ftrunk%2F&rev=5&sc=0 > which seems to be a testsuite that would validate your implementation. > > Ali > > > On Tue, 21 Jul 2009 15:39:34 -0700, Lisa Hsu <[email protected]> wrote: > > I believe EV5 did not have such instructions, so if you use a > crosscompiler > > targeted to EV5 (which you can specify when you build with > > http://freshmeat.net/projects/crosstool-ng/). > > > > Alternatively, I don't know if there is a compiler switch for turning off > > the generation of MVI instructions (of which maxuw4 is a member), so you > > could look into that if you don't want to build a new crosscompiler. > > > > Lisa > > > > On Tue, Jul 21, 2009 at 3:21 PM, ef > > > >> Hello, > >> > >> I am compiling Ferret from PARSEC 2.0 and I got this panic in m5: > >> > >> panic: attempt to execute unimplemented instruction 'maxuw4' (inst > >> 0x723007a4, opcode 0x1c) > >> > @ cycle 2367067420500 > >> > [execute:build/ALPHA_FS/arch/alpha/o3_cpu_exec.cc, line 70] > >> > >> Any clues on how fix this? One way is to implement the instructions, > >> other > >> way is probably to find some sort of compiling flag to not produce this > >> instructions. Anyways any advice, experience or thoughts on this problem > >> is > >> appreciated. > >> > >> Thanks > >> EF > >> > >> _______________________________________________ > >> 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 >
_______________________________________________ m5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
