Re: Vectorization paper / kernels.
On 30 April 2012 15:10, Singh, Ravi Kumar (Ravi) wrote: > Ramana, > > My tests S115 is hanging in the vectorized version of TSVC benchmark. Thanks for trying it out - I hadn't really tried it out when I had sent the mail. Uli is planning to look into these in the next few days. Ramana > > Regards > RKS > > -Original Message- > From: linaro-toolchain-boun...@lists.linaro.org > [mailto:linaro-toolchain-boun...@lists.linaro.org] On Behalf Of Ramana > Radhakrishnan > Sent: Monday, April 30, 2012 3:44 AM > To: Ulrich Weigand > Cc: linaro-toolchain > Subject: Vectorization paper / kernels. > > Hi Uli, > > While looking into something else I ran into these - I wonder how many > of these GCC manages to vectorize ... > > http://www.netlib.org/benchmark/livermorec . These look interesting > from a vectorizer kernels point of view. > > The other interesting paper of note was this PACT paper on > vectorization benchmarks comparing icc , xlc and GCC which might > provide some interesting hints / reading. > http://polaris.cs.uiuc.edu/~garzaran/doc/pact11.pdf . The appropriate > benchmarks kernels are linked to below. > > http://polaris.cs.uiuc.edu/ ̃maleki1/TSVC.tar.gz. > > > regards, > Ramana > > ___ > linaro-toolchain mailing list > linaro-toolchain@lists.linaro.org > http://lists.linaro.org/mailman/listinfo/linaro-toolchain ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain
RE: Vectorization paper / kernels.
Actually its not hanging, I tried to run it in isolation. Without VectorizationWith Vectorization Time Value Time Value S1115 67.98 0.065529 1186.440.065536 That means degradation of -1645.28% Regards RKS -Original Message- From: Ramana Radhakrishnan [mailto:ramana.radhakrish...@linaro.org] Sent: Thursday, May 03, 2012 12:38 PM To: Singh, Ravi Kumar (Ravi) Cc: Ulrich Weigand; linaro-toolchain Subject: Re: Vectorization paper / kernels. On 30 April 2012 15:10, Singh, Ravi Kumar (Ravi) wrote: > Ramana, > > My tests S115 is hanging in the vectorized version of TSVC benchmark. Thanks for trying it out - I hadn't really tried it out when I had sent the mail. Uli is planning to look into these in the next few days. Ramana > > Regards > RKS > > -Original Message- > From: linaro-toolchain-boun...@lists.linaro.org > [mailto:linaro-toolchain-boun...@lists.linaro.org] On Behalf Of Ramana > Radhakrishnan > Sent: Monday, April 30, 2012 3:44 AM > To: Ulrich Weigand > Cc: linaro-toolchain > Subject: Vectorization paper / kernels. > > Hi Uli, > > While looking into something else I ran into these - I wonder how many > of these GCC manages to vectorize ... > > http://www.netlib.org/benchmark/livermorec . These look interesting > from a vectorizer kernels point of view. > > The other interesting paper of note was this PACT paper on > vectorization benchmarks comparing icc , xlc and GCC which might > provide some interesting hints / reading. > http://polaris.cs.uiuc.edu/~garzaran/doc/pact11.pdf . The appropriate > benchmarks kernels are linked to below. > > http://polaris.cs.uiuc.edu/ ̃maleki1/TSVC.tar.gz. > > > regards, > Ramana > > ___ > linaro-toolchain mailing list > linaro-toolchain@lists.linaro.org > http://lists.linaro.org/mailman/listinfo/linaro-toolchain ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain
Re: Vectorization paper / kernels.
On 3 May 2012 19:30, Singh, Ravi Kumar (Ravi) wrote: > Actually its not hanging, I tried to run it in isolation. > > Without Vectorization With Vectorization > Time Value Time Value > S1115 67.98 0.065529 1186.44 0.065536 > > > That means degradation of -1645.28% Sounds to me like vectorization might have gone awry - What happens if you use something like -fvect-cost-model ? Does that make any difference ? It sounds to me like a pathological runaway vectorization which might have resulted in massive spills causing such a regression. For the record this was just a pointer to a paper that might be of interest . I haven't really investigated the kernels to see how useful they really are :) . regards, Ramana > > Regards > RKS > -Original Message- > From: Ramana Radhakrishnan [mailto:ramana.radhakrish...@linaro.org] > Sent: Thursday, May 03, 2012 12:38 PM > To: Singh, Ravi Kumar (Ravi) > Cc: Ulrich Weigand; linaro-toolchain > Subject: Re: Vectorization paper / kernels. > > On 30 April 2012 15:10, Singh, Ravi Kumar (Ravi) wrote: >> Ramana, >> >> My tests S115 is hanging in the vectorized version of TSVC benchmark. > > Thanks for trying it out - I hadn't really tried it out when I had > sent the mail. Uli is planning to look into these in the next few > days. > > Ramana > >> >> Regards >> RKS >> >> -Original Message- >> From: linaro-toolchain-boun...@lists.linaro.org >> [mailto:linaro-toolchain-boun...@lists.linaro.org] On Behalf Of Ramana >> Radhakrishnan >> Sent: Monday, April 30, 2012 3:44 AM >> To: Ulrich Weigand >> Cc: linaro-toolchain >> Subject: Vectorization paper / kernels. >> >> Hi Uli, >> >> While looking into something else I ran into these - I wonder how many >> of these GCC manages to vectorize ... >> >> http://www.netlib.org/benchmark/livermorec . These look interesting >> from a vectorizer kernels point of view. >> >> The other interesting paper of note was this PACT paper on >> vectorization benchmarks comparing icc , xlc and GCC which might >> provide some interesting hints / reading. >> http://polaris.cs.uiuc.edu/~garzaran/doc/pact11.pdf . The appropriate >> benchmarks kernels are linked to below. >> >> http://polaris.cs.uiuc.edu/ ̃maleki1/TSVC.tar.gz. >> >> >> regards, >> Ramana >> >> ___ >> linaro-toolchain mailing list >> linaro-toolchain@lists.linaro.org >> http://lists.linaro.org/mailman/listinfo/linaro-toolchain ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain
RE: Vectorization paper / kernels.
Ramana, If I use -fvect-cost-model it reverts bacl to novec perf regards RKS From: Ramana Radhakrishnan [ramana.radhakrish...@linaro.org] Sent: Thursday, May 03, 2012 3:03 PM To: Singh, Ravi Kumar (Ravi) Cc: Ulrich Weigand; linaro-toolchain Subject: Re: Vectorization paper / kernels. On 3 May 2012 19:30, Singh, Ravi Kumar (Ravi) wrote: > Actually its not hanging, I tried to run it in isolation. > > Without VectorizationWith Vectorization > Time Value Time Value > S1115 67.98 0.065529 1186.440.065536 > > > That means degradation of -1645.28% Sounds to me like vectorization might have gone awry - What happens if you use something like -fvect-cost-model ? Does that make any difference ? It sounds to me like a pathological runaway vectorization which might have resulted in massive spills causing such a regression. For the record this was just a pointer to a paper that might be of interest . I haven't really investigated the kernels to see how useful they really are :) . regards, Ramana > > Regards > RKS > -Original Message- > From: Ramana Radhakrishnan [mailto:ramana.radhakrish...@linaro.org] > Sent: Thursday, May 03, 2012 12:38 PM > To: Singh, Ravi Kumar (Ravi) > Cc: Ulrich Weigand; linaro-toolchain > Subject: Re: Vectorization paper / kernels. > > On 30 April 2012 15:10, Singh, Ravi Kumar (Ravi) wrote: >> Ramana, >> >> My tests S115 is hanging in the vectorized version of TSVC benchmark. > > Thanks for trying it out - I hadn't really tried it out when I had > sent the mail. Uli is planning to look into these in the next few > days. > > Ramana > >> >> Regards >> RKS >> >> -Original Message- >> From: linaro-toolchain-boun...@lists.linaro.org >> [mailto:linaro-toolchain-boun...@lists.linaro.org] On Behalf Of Ramana >> Radhakrishnan >> Sent: Monday, April 30, 2012 3:44 AM >> To: Ulrich Weigand >> Cc: linaro-toolchain >> Subject: Vectorization paper / kernels. >> >> Hi Uli, >> >> While looking into something else I ran into these - I wonder how many >> of these GCC manages to vectorize ... >> >> http://www.netlib.org/benchmark/livermorec . These look interesting >> from a vectorizer kernels point of view. >> >> The other interesting paper of note was this PACT paper on >> vectorization benchmarks comparing icc , xlc and GCC which might >> provide some interesting hints / reading. >> http://polaris.cs.uiuc.edu/~garzaran/doc/pact11.pdf . The appropriate >> benchmarks kernels are linked to below. >> >> http://polaris.cs.uiuc.edu/ ̃maleki1/TSVC.tar.gz. >> >> >> regards, >> Ramana >> >> ___ >> linaro-toolchain mailing list >> linaro-toolchain@lists.linaro.org >> http://lists.linaro.org/mailman/listinfo/linaro-toolchain ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain
RE: Vectorization paper / kernels.
Ramana, Can you suggest the flags used for optimal vectorization on A15 using GNU GCC 4.7. Perhaps specific optimization flags for A15 would also help regards RKSingh From: Ramana Radhakrishnan [ramana.radhakrish...@linaro.org] Sent: Thursday, May 03, 2012 3:03 PM To: Singh, Ravi Kumar (Ravi) Cc: Ulrich Weigand; linaro-toolchain Subject: Re: Vectorization paper / kernels. On 3 May 2012 19:30, Singh, Ravi Kumar (Ravi) wrote: > Actually its not hanging, I tried to run it in isolation. > > Without VectorizationWith Vectorization > Time Value Time Value > S1115 67.98 0.065529 1186.440.065536 > > > That means degradation of -1645.28% Sounds to me like vectorization might have gone awry - What happens if you use something like -fvect-cost-model ? Does that make any difference ? It sounds to me like a pathological runaway vectorization which might have resulted in massive spills causing such a regression. For the record this was just a pointer to a paper that might be of interest . I haven't really investigated the kernels to see how useful they really are :) . regards, Ramana > > Regards > RKS > -Original Message- > From: Ramana Radhakrishnan [mailto:ramana.radhakrish...@linaro.org] > Sent: Thursday, May 03, 2012 12:38 PM > To: Singh, Ravi Kumar (Ravi) > Cc: Ulrich Weigand; linaro-toolchain > Subject: Re: Vectorization paper / kernels. > > On 30 April 2012 15:10, Singh, Ravi Kumar (Ravi) wrote: >> Ramana, >> >> My tests S115 is hanging in the vectorized version of TSVC benchmark. > > Thanks for trying it out - I hadn't really tried it out when I had > sent the mail. Uli is planning to look into these in the next few > days. > > Ramana > >> >> Regards >> RKS >> >> -Original Message- >> From: linaro-toolchain-boun...@lists.linaro.org >> [mailto:linaro-toolchain-boun...@lists.linaro.org] On Behalf Of Ramana >> Radhakrishnan >> Sent: Monday, April 30, 2012 3:44 AM >> To: Ulrich Weigand >> Cc: linaro-toolchain >> Subject: Vectorization paper / kernels. >> >> Hi Uli, >> >> While looking into something else I ran into these - I wonder how many >> of these GCC manages to vectorize ... >> >> http://www.netlib.org/benchmark/livermorec . These look interesting >> from a vectorizer kernels point of view. >> >> The other interesting paper of note was this PACT paper on >> vectorization benchmarks comparing icc , xlc and GCC which might >> provide some interesting hints / reading. >> http://polaris.cs.uiuc.edu/~garzaran/doc/pact11.pdf . The appropriate >> benchmarks kernels are linked to below. >> >> http://polaris.cs.uiuc.edu/ ̃maleki1/TSVC.tar.gz. >> >> >> regards, >> Ramana >> >> ___ >> linaro-toolchain mailing list >> linaro-toolchain@lists.linaro.org >> http://lists.linaro.org/mailman/listinfo/linaro-toolchain ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain
Updates to the BzrTips wiki page
I've done some minor updates to the instructions for working with gcc-linaro, bzr, and merge requests at: https://wiki.linaro.org/WorkingGroups/ToolChain/BzrTips The interesting changes are using: * bzr commit --fixes=lp: to link a branch to a bug number * bzr branch --hardlink to cut down on branch time and disk usage * bzr-merge-changelog to automatically merge ChangeLog.linaro -- Michael ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain
Simpler benchmark comparison
I've made a couple of changes to the processing of benchmark results. The server now automatically tabulates the raw results like: http://ex.seabright.co.nz/benchmarks/gcc-linaro-4.7+bzr114974/logs/armv7l-natty-cbuild292-tcpanda02-cortexa9r1/benchmarks.txt diffbench[1] can now take a URL instead of a local file such as: python diffbench.py \ http://ex.seabright.co.nz/benchmarks/gcc-linaro-4.7+bzr114974/logs/armv7l-natty-cbuild292-tcpanda02-cortexa9r1/benchmarks.txt \ http://ex.seabright.co.nz/benchmarks/gcc-linaro-4.7+bzr114977~uweigand~mov-vld1.64-4.7/logs/armv7l-natty-cbuild292-tcpanda06-cortexa9r1/benchmarks.txt You'll need to put the username and password from https://wiki.linaro.org/Internal/ToolChain into ~/.netrc to authenticate. diffbench now tries to pull a readable name from the results to make the results easier to understand. Running the above command now shows: Significant differences --- * bzr114977 scored xx.yy % (t=38.9) against bzr114974 on popular embedded benchmark rgbcmykv2data3 ... All other results - == = = testname subnamebzr114974 bzr114977 improvement in bzr114977 == = = coremark 2xxx.096 2xxx.653100.0 ... -- Michael [1] http://code.launchpad.net/linaro-toolchain-benchmarks scripts/diffbench.py ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain
Agenda for 2012-05-07
I've put the agenda for Monday's call at: https://wiki.linaro.org/WorkingGroups/ToolChain/Meetings/2012-05-07 which is: * Review action items from last meeting * Connect sessions: * GDB for Android * GCC performance call - Live! * KVM performance * Renderscript * Android benchmarking * Dalvik improvements * Hard float switchover status * multiarch upstreaming * Next week is release week * vld1, EEMBC results, and noise * twolf result variance * KVM minimum features * UP/UP, Ubuntu * [[MichaelHope/Sandbox/KVMUseCase]] Feel free to edit, -- Michael ___ linaro-toolchain mailing list linaro-toolchain@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-toolchain
