Bill Allombert on Wed, 6 Nov 2002 15:00:59 +0100 |
[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]
CVS branch gmp-kernel-2-2-5 |
Hello PARI developers, I have commited the gmp kernel patch in a new branch named gpm-kernel-2-2-5 This include the change to the t_INT API. To switch your repository to this branch, use cvs update -d -r gpm-kernel-2-2-5. To use the GMP kernel, you need to have gmp in a standard location. you can do ./Configure --kernel=gmp but it is advisable to have both kernel along, so issue ./Configure --kernel=gmp --builddir=O<osname-arch>-gmp to build the gmp kernel version in O<osname-arch>-gmp. make bench should report only report problems in the nfields bench and all of them must be related to 'GMP: ratlift is not yet implemented' This kernel is not finished. Some issue as I remember them: --- Only gmp INT are implemented. REAL are standard PARI reals. --- As a consequence, INT<-->REAL conversion are slower. --- ratlift() is not implemented. This looks difficult to do without an insight of the code. Help appreciate (Gerhart ?). It need to either be ported to the t_INT API, or rewrtten for the gmp kernel. --- The /*HACK*/ in the code need to be investigated and guard values added, else stack overrun will happen. --- No Montgomery multiplication support. The GMP code for that is private to GMP. --- Bit manipulation routine (bitand,bittest,etc...) need to be tested. --- diviiexact is just a call to divii. --- There is no possibility to use the optimised level0 asm code with the gmp kernel. It is mainly a makefile issue. --- Configure --kernel=gmp must look harder to find libgmp. --- Lots of things need tuning/testing. Cheers, Bill