Subversion Repositories Open64

[/] [branches/] [open64-3.0/] - Rev 672

Rev

Filtering Options

Clear current filter

Rev Log message Author Age Path
672 copy this fix into branches/open64-3.0, when there are several bug fixes accumulated, we can tags this branches as open64-3.0.1 tianwei 2364d 08h /branches/open64-3.0/
668 One .0 to much ributzka 2373d 05h /branches/open64-3.0/
666 Create branch open64-3.0.0 from trunk ributzka 2373d 05h /branches/open64-3.0.0/
665 this workaround is used to prevent a block from growing too big. It is up to region formation to control the region size syang 2373d 19h /trunk/
664 so long as the dedicated TN bound to the granted register is not live at the exit point of the block, it can be granted to LRA. syang 2373d 19h /trunk/
662 This change installs crt{begin|end}[S].o coming from gcc suite instead of prebuilt archives. Currently, fortran compiler use ld to link and it looks for crt{begin|end}[S].o from phase dir. Those crts should also be found by gcc --print-file-name when the compiler is on native environment. We are preparing the upcoming 3.0 release and hence reluctant to change code syang 2376d 05h /trunk/
661 1. fix for ipa_link error, rpath is useless, we just delete it as sun suggested
2. also we need to add " -Wl,-dynamic-linker -Wl,/lib/ld-linux-ia64.so.2" for SVR4
tianwei 2376d 13h /trunk/
660 change the driver info of "opencc -v " tianwei 2376d 13h /trunk/
659 ebo fix,just prevents the EBO from generating "cmp p0,p0=r0,r0" tianwei 2376d 13h /trunk/
658 turn off alias type for wn_fio.cxx tianwei 2376d 13h /trunk/
656 we no longer need to install crt{begin,end}[S].o syang 2377d 04h /trunk/
654 roll back this changes, only turn off turn off type_alias for wn_fio.cxx tianwei 2377d 16h /trunk/
653 Sun decide the release number is 3.0 rather than 2.1 syang 2377d 17h /trunk/
652 turn off type_alias for bootstraping compiler tianwei 2379d 15h /trunk/
651 This is a temporary fix for GRA. It shouldn't grant some registers to LRA. However, even if the wrong registers are granted to LRA, LRA should recognize it and not allocate them. This should also be fixed. tianwei 2379d 15h /trunk/
650 1. yhd's fix for bug 209.
2. yhd's fix for bug 211
tianwei 2379d 20h /trunk/
649 our compiler would generate a temp .o file for dwarf info during compiler,so if
a previous same .o object file exsited, it will be deleted, such as
opencc expr.o -c expr.cxx
opencc c-expr.o -c expr.cxx
ar a.so expr.o c-expr.o
we will found that the first expr.o is missing
tianwei 2379d 20h /trunk/
648 some small modification for Kernel boot tianwei 2379d 20h /trunk/
647 fix to BUG-200 syang 2381d 06h /trunk/
646 Install libstdc++.so into $TOOLROOT/usr/.../lib to avoid linker's complain of
"undefined versioned symbol name _ZSt10time_put_w@@GLIBCPP_3.2"
hucheng 2381d 20h /trunk/
645 Changing the copyright of the newly added WOVP files to BSD style.
-- Jianian
tianwei 2381d 20h /trunk/
640 use prebuilt mfef95 by default syang 2383d 18h /trunk/
639 prebuilt library check-in so that users can build pre-built library from source.

This is NOT turned on by default. If users want to have a try, they can build it with BUILD_COMPILER=OSP var.
fengzhou 2384d 11h /trunk/
638 change minor release number syang 2385d 02h /trunk/
637 merge kernel branch into the main truck tianwei 2385d 19h /trunk/
636 1. gcommondefs fix for building FORTRAN FE with ICC in static link mode.
2. dag.cxx is fix for bug 198
3. s_utils.c is for fixing tonto and galgel bug.
These bugs are caused by creating dope vector in FORTRAN library.
The stride mulitplier in dope vector(or dynamic allocated array) for an array is sizeof(array type)/sizeof(int) (sizeof(int)==32 on native).
The stride multiplier for normal array is 1.
When we create a new dope vector from an existing dope vector for FORTRAN function calls that use dope vector (like MATMUL), we do should just copy the stride multiplier from source.
fengzhou 2385d 21h /trunk/
635 1.dehao's fix for inline enhancement's side effect.
2.450.soplex's RE's fix, shuxin's 573 checkin didn't solve the problem
perfectly, he classify the problem, and found that WN_object_ty for "Mload"
is not very right.
tianwei 2385d 21h /trunk/
633 Remove the wrongly checked in fixes in revision 632, sorry for my stupid mistakes. tianwei 2387d 13h /trunk/
632 Adds the EH related documents, including EH's implementation framework,
and runtime specification on IA-64.
Besides, also adds the document "HOW-TO-DEBUG-OPEN64".
tianwei 2387d 14h /trunk/
631 This is a temporary tuning of inliner. It will:
1. Set correct value to loopnest, which is always 0 previously.
2. Let more PUs with simple structure be inlined.
This will be replaced by more sophisticated interprocedural analysis.
tianwei 2387d 20h /trunk/