fft/tune/tune-fft.c is now included in the distribution tarball. I also discovered that make speed is working after all. I simply hadn't built the library.
I also checked the fft cutoffs by hand and it is producing reasonable cutoffs. It's surprising to me that the toom8h code is competitive (with some crossovers along the way) all the way up to 11000 limbs. So the 7500 limbs that the tuning code selects on my machine is quite reasonable, if not a little large. I will make one final check that I have the code for timing the fft set up correctly. But probably it is just due to not having an iterative fft implementation to speed things up for small sizes, and a very fast toom8h (which we added from the GMP project). I'd like to get this crossover much lower, but for an initial release with the new fft this will be fine. We'll have to work on some fast assembly code for it and an iterative version for small convolution lengths. Anyhow, this brings the list of todo items to the following: mpir-todo: ========== * fft_combine_bits const warning (see below) * ../mpn/generic/gcdext.c:238: warning: integer overflow in expression * _mp_release_macro volker braun reported * implicit declaration warning (see below) * integer overflow warning (see below) * test start/end warning (see below) * update library version number * test on platforms we have access to * make tune on platforms we have access to * release paperwork * make mpir-2.6.0 branch in svn * list pavel's Multiprecision Computing Toolbox for MATLAB, http://www.advanpix.com on website I will now try to fix all these compiler warnings. They should all be easy one line fixes. Bill. On 27 September 2012 22:41, Bill Hart <goodwillh...@googlemail.com> wrote: > OK, thanks. I'll add that program to the relevant makefile.am. > > Bill. > > On 27 September 2012 22:32, Brian Gladman <b...@gladman.plus.com> wrote: >> -----Original Message----- From: Bill Hart >> Sent: Thursday, September 27, 2012 10:06 PM >> To: mpir-devel@googlegroups.com >> Cc: Brian Gladman >> >> Subject: Re: [mpir-devel] Re: MPIR 2.6 release progress >> >> Hi Brian, >> >> After discovering that the libtool in MinGW32 is different in version >> number by 0.001 and therefore completely *incompatible*, according to >> autoreconf on Linux, I did manage to get it to reconf and am happy to >> report it builds and passes tests on Linux. >> >> I also added tests/fft to the relevant Makefile.am and the fft tests >> now run. We've not run these before on Linux, but I'm happy to report >> they pass on x86_64. I'll try them shortly on MinGW32. Then I'll pack >> it in for the night. >> >> Brian, I've attached a tarball produced by make dist. Could you verify >> that this builds on Windows and that make dist hasn't omitted >> anything. If it has, can you let me know what and I'll try to modify >> it so that it is picked up. This is not an svn issue, but an autotools >> one. Yes, autotools even interacts with the MSVC build!! >> >> ============== >> The only problem I had was that tune-fft.c was not present (or not where I >> expected it). >> >> Otherwise it worked fine on x64. >> >> >> Brian >> >> -- >> You received this message because you are subscribed to the Google Groups >> "mpir-devel" group. >> To post to this group, send email to mpir-devel@googlegroups.com. >> To unsubscribe from this group, send email to >> mpir-devel+unsubscr...@googlegroups.com. >> For more options, visit this group at >> http://groups.google.com/group/mpir-devel?hl=en. >> -- You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to mpir-devel@googlegroups.com. To unsubscribe from this group, send email to mpir-devel+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en.
