PatchSet 7733 Date: 2008/02/02 16:17:47 Author: robilad Branch: HEAD Tag: (none) Log: cleaned up docs from GNU MP references
2008-02-02 Dalibor Topic <[EMAIL PROTECTED]> * FAQ/FAQ.bignum: Removed. * Makefile.am (EXTRA_DIST): Removed FAQ/FAQ.bignum. * THIRDPARTY: Remove section on gmp.m4. * FAQ/FAQ.requiredlibraries: Removed section on ESD, ALSA, GNU MP, and libmd. * FAQ/FAQ.ps2linux: Removed paragraph on GNU MP. Members: ChangeLog:1.5233->1.5234 Makefile.am:1.150->1.151 Makefile.in:1.313->1.314 THIRDPARTY:1.60->1.61 FAQ/FAQ.bignum:1.1->1.2(DEAD) FAQ/FAQ.ps2linux:INITIAL->1.3 FAQ/FAQ.requiredlibraries:1.16->1.17 Index: kaffe/ChangeLog diff -u kaffe/ChangeLog:1.5233 kaffe/ChangeLog:1.5234 --- kaffe/ChangeLog:1.5233 Sat Feb 2 16:05:46 2008 +++ kaffe/ChangeLog Sat Feb 2 16:17:47 2008 @@ -1,5 +1,22 @@ 2008-02-02 Dalibor Topic <[EMAIL PROTECTED]> + * FAQ/FAQ.bignum: Removed. + + * Makefile.am (EXTRA_DIST): Removed FAQ/FAQ.bignum. + + * THIRDPARTY: Remove section on gmp.m4. + + * FAQ/FAQ.requiredlibraries: Removed section on ESD, ALSA, + GNU MP, and libmd. + + * FAQ/FAQ.ps2linux: Removed paragraph on GNU MP. + +2008-02-02 Dalibor Topic <[EMAIL PROTECTED]> + + * WHATSNEW, TODO: Updated for native big math purge. + +2008-02-02 Dalibor Topic <[EMAIL PROTECTED]> + Fix for build on arm-oabi-linux-ffi. * kaffe/kaffevm/systems/unix-pthreads/thread-impl.c (__USE_GNU) Index: kaffe/Makefile.am diff -u kaffe/Makefile.am:1.150 kaffe/Makefile.am:1.151 --- kaffe/Makefile.am:1.150 Sat Jan 19 22:53:06 2008 +++ kaffe/Makefile.am Sat Feb 2 16:17:48 2008 @@ -67,7 +67,6 @@ FAQ/FAQ.automake \ FAQ/FAQ.awt \ FAQ/FAQ.benchmarking \ - FAQ/FAQ.bignum \ FAQ/FAQ.blackfin \ FAQ/FAQ.charsets \ FAQ/FAQ.class-states \ Index: kaffe/Makefile.in diff -u kaffe/Makefile.in:1.313 kaffe/Makefile.in:1.314 --- kaffe/Makefile.in:1.313 Sat Feb 2 15:58:10 2008 +++ kaffe/Makefile.in Sat Feb 2 16:17:48 2008 @@ -356,7 +356,6 @@ FAQ/FAQ.automake \ FAQ/FAQ.awt \ FAQ/FAQ.benchmarking \ - FAQ/FAQ.bignum \ FAQ/FAQ.blackfin \ FAQ/FAQ.charsets \ FAQ/FAQ.class-states \ Index: kaffe/THIRDPARTY diff -u kaffe/THIRDPARTY:1.60 kaffe/THIRDPARTY:1.61 --- kaffe/THIRDPARTY:1.60 Sun Jan 6 19:05:14 2008 +++ kaffe/THIRDPARTY Sat Feb 2 16:17:48 2008 @@ -147,15 +147,6 @@ agrees to be bound by the terms and conditions of this License Agreement. -* m4/gmp.m4 - -This code has been taken from GNU m4. Kaffe is tracking this project's CVS. -The code is licensed under GNU General Public License v2 or later. - -See license.terms for the text of GNU General Public License v2. - -More information on GNU m4 is available at http://www.gnu.org/software/m4/. - * libraries/clib/native/VMRuntime.c, m4/classpath.m4. This file comes from CACAO. More information on CACAO is available from =================================================================== Checking out kaffe/FAQ/FAQ.bignum RCS: /home/cvs/kaffe/kaffe/FAQ/Attic/FAQ.bignum,v VERS: 1.1 *************** --- kaffe/FAQ/FAQ.bignum Sat Feb 2 16:19:38 2008 +++ /dev/null Sun Aug 4 19:57:58 2002 @@ -1,43 +0,0 @@ -Using BigIntegers and BigDecimal on Kaffe -========================================= - -Author: Dalibor Topic <[EMAIL PROTECTED]> - -Implementations ---------------- - - Kaffe comes with two implementations of java.math package. The default one, - from GNU Classpath, is written in Java. It is available on all platforms. - - The other implementation uses GNU MP for the actual calculations on - BigInteger. Using the very fast GNU MP library makes it much faster than the - default implementation. - -Configuration -------------- - - When Kaffe's configure script detects the presence of GNU MP library, it - automatically takes care of building kaffe with GNU MP support enabled. - - Support for GNU MP can be explicitely turned off using the --without-gmp - configure script option. - -Usage ------ - - Kaffe uses the pure java implementation by default. If you want to use the - fast java.math implementation, you need to run kaffe with the - -Xnative-big-math option. - -Performance ------------ - - Running the NativeBigInteger benchmark from i2p [1] shows that kaffe on i686 - on GNU/Linux with jit3 and jthreads is about 4 times slower than JDK 1.4.2 - using the default, pure java implementation of java.math. - - Using the native implementation, the tables turn, and kaffe is 5 times faster - than JDK 1.4.2. - -[1] I2P is available at http://www.i2p.net. The benchmark is in the net.i2p.util - package. =================================================================== Checking out kaffe/FAQ/FAQ.ps2linux RCS: /home/cvs/kaffe/kaffe/FAQ/FAQ.ps2linux,v VERS: 1.3 *************** --- /dev/null Sun Aug 4 19:57:58 2002 +++ kaffe/FAQ/FAQ.ps2linux Sat Feb 2 16:19:40 2008 @@ -0,0 +1,89 @@ +PS2 Linux Kaffe Building FAQ +============================ + +This document applies to kaffe-1.1.4 (and also contains hints +for building from CVS). + +FAQ.ps2linux Status +=================== + +Last updated 28-Jan-2005 by Patryk Laurent ([EMAIL PROTECTED]) +Original document 18-dec-2002 Dylan Schell ([EMAIL PROTECTED]) + + +Test Notes [not updated or confirmed on 28-Jan-2005] +==================================================== + +- kaffe in interpreter mode currently fails 5 regression tests related to + IEEE compliancy. ( See Notes ) +- kaffe in jit3 mode currently fails a 27 regression tests since the + jit is emitting 64 bit floating point instructions that are not supported + on the PS2. ( see Notes ) + +Building kaffe on the PS2 +========================= + +1. Install jikes ( version 1.21 was tested ) on your PS2 + +http://ibm.com/developerworks/opensource/jikes/ + +Compiling jikes will take some time on the PS2 but building the +kaffe runtime library with the supplied compiler is a bit of a problem. +Even if you get it to work, it is a lot slower than compiling jikes) + +2. Untar kaffe. Edit kaffe's libraries/javalib/rebuildLib.in + +Add -target 1.2 to the VERBOSE variable to force jikes to generate +class files compatible with kaffe. + +VERBOSE=-target 1.2 + +3. Run configure to enable jikes and select the interpreter engine: +./configure --enable-pure-java-math --with-jikes --with-engine=intrp + +IF YOU ARE BUILDING FROM CVS, you probably should add --without-kaffe-qt-awt and --disable-native-awt (otherwise you will get gtk+-2.0 missing errors in the configure step.) +January 22, 2005 From CVS: ./configure --enable-pure-java-math --with-jikes --witth-engine=intrp --without-kaffe-qt-awt --disable-native-awt + +Or if you don't need double precision floating point support, you can +build the jit3 version: +./configure --with-jikes + +(Remember to run make clean if you change between interpreter and jit!) + +4. Compile/install in the regular fashion. +make +sudo make install + +Notes about the PS2 Linux port +============================== + +The ps2 port is a work in progress, it works with the following limitations: + +1. Some IEEE constructs are not supported by the FPU: NaN/+Inf/-Inf/+0.0/-0.0 +If the PS2 kernel is ever updated to include a FPU emulator to add these missing constructs, then this port should pass all regression tests. +(This goes for the interpreter as well as for the jit version) + +1. jit3 is functional but will cause problems when using double precision floating point operations. This is because the PS2 lacks a double precsion FPU. If, or when the linux kernel is updated with a FPU emulator (and a matching gcc is installed) to make up for this problem, the port will have to be updated ( the PS2 specific patches in config/mips/o32-sysdepCallMethod.h can go away ). + +To make sure interpreter bugs don't interfere with building kaffe on the PS2 use jikes to compile the class library/tests. Some versions of jikes generate incorrect class files or class files that are too new for kaffe to handle. Using jikes 1.13 seems to work OK. + +The PS2 also lacks the MIPS II "LL" and "SC" instructions. This means locking is somewhat slower. the MIPS II instructions are disabled in config/mips/common.h + +The PS2 register set is also different from basic mips processors. This means the FP_OFFSET is much larger ( general purpose registers are 128 bit wide for special multimedia/vector processing instructions ) + +Note that the changes in o32-sysdepCallMethod.h pass 64 bit floats in a pair +of general purpose registers. This is the convention gcc uses on the PS2 ( +This might be the default for all mips systems that lack these registers, I +don't know...) + +Files with PS2 specific changes +=============================== +config/mips/o32-sysdepCallMethod.h ( work around lack of 64 bit FPU regs ) +config/mips/common.h ( set FP_OFFSET / disable MIPS II instructions ) +config/mips/linux/md.h ( FP_OFFSET=36 ) + +Dylan Schell [EMAIL PROTECTED] + +Patryk Laurent [EMAIL PROTECTED] Index: kaffe/FAQ/FAQ.requiredlibraries diff -u kaffe/FAQ/FAQ.requiredlibraries:1.16 kaffe/FAQ/FAQ.requiredlibraries:1.17 --- kaffe/FAQ/FAQ.requiredlibraries:1.16 Sat Feb 2 15:38:23 2008 +++ kaffe/FAQ/FAQ.requiredlibraries Sat Feb 2 16:17:49 2008 @@ -99,53 +99,6 @@ Ditto for png images. Available at http://www.libpng.org/pub/png/libpng.html -Sound libraries ---------------- - -You need at least one of these libraries if you want to use kaffe's -Java Sound implementation, which has been merged in from the tritonus -project. More information on sound support in kaffe is available in -FAQ.sound. - -ALSA: - The Advanced Linux Sound Architecture backend provides both - sampled sound and MIDI functionality. You need to have ALSA - installed first, which is available at: - http://www.alsa-project.org/ - -Esd: - The Enlightened Sound Daemon provides sound playback - functionality on many Unix implementations. It is available - at: - http://www.tux.org/~ricdude/EsounD.html - -Mathematical libraries ----------------------- - -If you want to use arbitrary precision numbers provided by the -java.math package, you can install these libraries. If kaffe can not -find them, it will automatically use a pure java implementation from -GNU Classpath. - -GNU MP: - libgmp is used in the native implementation of - java.math.BigInteger. It is available at: - http://www.swox.com/gmp/ - -Cryptographical libraries -------------------------- - -If you want to use cryptographical hashing algorithms, you need to -install these libraries. - -libmd: - This library is included with FreeBSD and may be included with - other operating systems. It supplies implementations of the MD2, - MD4, and MD5 hashing algorithms. If this library is not found by - configure, then SHA is the only hashing algorithm that will be - available. It can be found at: - http://www.penguin.cz/~mhi/libmd/ - Internationalization libraries ------------------------------ _______________________________________________ kaffe mailing list kaffe@kaffe.org http://kaffe.org/cgi-bin/mailman/listinfo/kaffe