Hello Ali, The implementation of the MVI instructions (13 of them) and ctpop and a bugfix for cttz for ALPHA along with sanity testcases is available here: http://www.csee.usf.edu/~sroy/techres/m5res/m5_alpha_cix_mvi.tar.bz2
The "gsm" benchmark from Mediabench-I generates the "maxsw4" instructions unless compiled with something like a -mcpu=ev56 option. -Soumyaroop On Thu, Oct 29, 2009 at 4:27 PM, Ali Saidi <[email protected]> wrote: > > I'm in no rush... > > Here are some references: > http://www.alphalinux.org/docs/MVI-full.html > http://www.alphalinux.org/wiki/index.php/Motion_Video_Instructions > > libmpeg2 has inverse discrete cosine transform code that makes use of the > MVI instructions. You should be able to extract an inner loop from one > there and compare the output of an x86 and the alpha code executed by M5. > > As far as CTPOP writing a testcase is pretty trivial. The instruction just > counts the number of bits set in a 64 bit register.See > http://en.wikipedia.org/wiki/Hamming_weight for an idea of how to do this > quickly without native popcount support. > > > Thanks, > Ali > > > > On Thu, 29 Oct 2009 16:42:29 -0400, soumyaroop roy <[email protected]> > wrote: >> [I am replying to m5-dev instead of m5-users] >> >> Hello Ali, >> >> Which benchmarks should one use to test the implementation of these >> instructions? >> >> Either I or Vidya Sangkar, a labmate of mine, can work on it. But it >> has to be after Nov 17th. Is that timeline is ok? >> >> regards, >> Soumyaroop >> >> On 10/29/09, Ali Saidi <[email protected]> wrote: >>> >>> ctlz and cttz are available and unless you're doing something crazy I >>> don't >>> think ctpop would be emitted. MVI isn't an instruction but a class of >>> instructions. See >>> > http://en.wikipedia.org/wiki/DEC_Alpha#Motion_Video_Instructions_.28MVI.29 >>> >>> At this point we've probably spent as much time discussing them as it >>> would >>> have taken to implement them. >>> >>> Ali >>> >>> On Thu, 29 Oct 2009 15:55:34 -0400, soumyaroop roy <[email protected]> >>> >>> wrote: >>> > One has to supply the flag, "-mno-cix", during compilation to make >>> > sure that the CIX instructions ("ctlz", "cttz", and "ctpop") are not >>> > emitted in the code. >>> > >>> > Here is the list of default flags passed (excerpt from >>> > "alphaev67-unknown-linux-gnu -v -Q dummy.c"): >>> > >>> > options passed: -v dummy.c -mcpu=ev67 -mtune=ev67 >>> > options enabled: -falign-loops -fargument-alias -fauto-inc-dec >>> > -fbranch-count-reg -fcommon -fearly-inlining >>> > -feliminate-unused-debug-types -ffunction-cse -fgcse-lm -fident >>> > -fivopts >>> > -fkeep-static-consts -fleading-underscore -fmath-errno >>> > -fmerge-debug-strings -fmove-loop-invariants -fpcc-struct-return >>> > -fpeephole -fsched-interblock -fsched-spec -fsched-stalled-insns-dep >>> > -fsigned-zeros -fsplit-ivs-in-unroller -ftoplevel-reorder >>> -ftrapping-math >>> > -ftree-cselim -ftree-loop-im -ftree-loop-ivcanon > -ftree-loop-optimize >>> > -ftree-parallelize-loops= -ftree-reassoc -ftree-scev-cprop >>> > -ftree-vect-loop-version -fvar-tracking -fzero-initialized-in-bss >>> > -mbwx >>> > -mcix -mexplicit-relocs -mfix -mfloat-ieee -mfp-regs -mglibc >>> -mlarge-data >>> > -mlarge-text -mlong-double-128 -mmax >>> > >>> > Note the "-mcix" on the second last line above. >>> > >>> > I could not find anything about the MVI instruction. >>> > >>> > regards, >>> > Soumyaroop >>> > >>> > On 10/29/09, Ali Saidi <[email protected]> wrote: >>> >> >>> >> I've put a link on the d/l page. The file is downloading to >>> >> m5sim.org >>> >> right >>> >> now and should be available within 5 minutes. >>> >> >>> >> Ali >>> >> >>> >> >>> >> >>> >> On Wed, 28 Oct 2009 22:27:11 -0400, soumyaroop roy >>> >> <[email protected]> >>> >> >>> >> wrote: >>> >> > Nate, >>> >> > >>> >> > It is available for download here (33 MB): >>> >> > >>> >> >>> > http://72.187.222.34/~sroy/m5_tests/alphaev67-unknown-linux-gnu-x86-32.tar.bz2 >>> >> > >>> >> > -Soumyaroop >>> >> > >>> >> > On Wed, Oct 28, 2009 at 9:12 PM, soumyaroop roy > <[email protected]> >>> >> > wrote: >>> >> >> On Wed, Oct 28, 2009 at 9:04 PM, nathan binkert >>> >> >> <[email protected]> >>> >> wrote: >>> >> >>>>>> My estimate is that the tar-zipped file should between >>> >> >>>>>> 250-300 >>> >> >>>>>> MB. >>> >> >>>>> No problem. >>> >> >>>> >>> >> >>>> Let me host it somewhere and send you the link so that you may >>> >> download >>> >> >>>> it. >>> >> >>> Sure. Once I get it on the m5 site, you can add the link. >>> >> >> >>> >> >> Is there a way for me to upload it on the M5 wiki directly? Is >>> >> >> that >>> >> >> what you were implying all along that I should do? >>> >> >> >>> >> >>> >>> >> >>>> Do you mean providing the link though the wiki? I do not have >>> >> >>>> permissions to edit this page, >>> >> >>>> http://www.m5sim.org/wiki/index.php/Download, which has links >>> >> >>>> to >>> >> >>>> the >>> >> >>>> cross-compilers. >>> >> >>> >>> >> >>> You should be able to, but you have to create an account. >>> >> >> >>> >> >> I already have an account. But it says: >>> >> >> >>> >> >> You do not have permission to edit this page, for the following >>> >> >> reason: >>> >> >> This page has been locked to prevent editing. >>> >> >> >>> >> >> You can view and copy the source of this page: >>> >> >> >>> >> >>> >>> >> >>> Nate >>> >> >>> _______________________________________________ >>> >> >>> m5-users mailing list >>> >> >>> [email protected] >>> >> >>> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >>> >> >>> >>> >> >> >>> >> >> -Soumyaroop >>> >> >> >>> >> >> -- >>> >> >> Soumyaroop Roy >>> >> >> Ph.D. Candidate >>> >> >> Department of Computer Science and Engineering >>> >> >> University of South Florida, Tampa >>> >> >> http://www.csee.usf.edu/~sroy >>> >> >> >>> >> _______________________________________________ >>> >> m5-users mailing list >>> >> [email protected] >>> >> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users >>> > -- Soumyaroop Roy Ph.D. Candidate Department of Computer Science and Engineering University of South Florida, Tampa http://www.csee.usf.edu/~sroy _______________________________________________ m5-dev mailing list [email protected] http://m5sim.org/mailman/listinfo/m5-dev
