-
Logged In: YES
user_id=254174
Because it's difficult to make the scheduler not to
schedule the code ahead the instrumented code that calls
mcount function, I defer the instrumentation before calling
EMT_Emit_PU() in CG_Generate_Code() in cg.cxx. This is done
by splitting the instrumentation code in old
function 'EETARG_Fixup_Entry_Code()" into a new
function...
2002-03-25 06:34:39 UTC in Open Research Compiler - Aurora
-
This request is to make orc support gcc-like profiling
by enabling option '-pg' in orc.
Current status:
1. orc does have limited support to this option. It
can do instrumentation during cgemit phase by calling
a function "_mcount" in libc.a. But according to the
description of "_mcount", this function needs to be
placed at the head of each pu. Currently ORC puts...
2002-03-20 07:57:46 UTC in Open Research Compiler - Aurora
-
It's a stack overflow.
2002-03-14 00:56:23 UTC in Open Research Compiler - Aurora
-
It's a stack overflow.
2002-03-13 07:05:37 UTC in Open Research Compiler - Aurora
-
Summary:
I managed to get some results after triaging
twolf. With 12 other IPA options turned off, twolf
can pass with option -IPA:picopt=off while fail
without it. The normal triaging process doesn't work
because some of the IPA options can't be turned off
individually. My triaging process is quite a random
selecting process.
Testing process:
All of the below test follows th.
2002-03-13 07:03:35 UTC in Open Research Compiler - Aurora
-
Version:
orc-1.0.0
Impact:
affect all versions after Pro64 .13.
During IPA after ipa_link, the makefile generated in
the xxx.ipakeep contains a line like below:
$(IPA_TMPDIR)/dummy :
cd $(IPA_TMPDIR); /usr/bin/orcc -c -i64
symtab.I -o symtab.o -TENV:emit_global_data=symtab.G -
Gspace 0 -pic1 -TARG:abi=i64 -Wb,-
CG:enable_feedback=off
The compiler path...
2002-03-07 08:09:36 UTC in Open Research Compiler - Aurora
-
Below is from Peng Zhao at ualberta:
Pro64 compiler awlays makes the num_entries
of a switch WN one
less than the actual entries. Thus sometimes it fails
with the follwoing
message:
### XGOTO/COMPGOTO: goto_count=4 n_targ=3
One suspicious place is
in ./osprey1.0/g++fe/wfe_stmt.cxx or the
one in gccfe:
in WFE_Expand_End_Case() there is :
n = case_info_i -...
2002-02-06 08:06:41 UTC in Open Research Compiler - Aurora
-
Following comments from Peng Zhao at ualbeta:
1. in the INSTALL script, it assumes that the orc will
be installed in
some system directories. This is usually very bad: you
can install the RPM
version in /usr/bin etc. But when you hack it, you
often hope to play with
it in common user directory. Thus the INSTALL script
sets the proper ROOT
dir, and in the INSTALL script should...
2002-02-06 07:58:02 UTC in Open Research Compiler - Aurora
-
Logged In: YES
user_id=254174
I disagree with this request. I understand the desire to
improve debuggability of the FmtAssert. The suggested
solution is to use "asm" statements which is a poorly
designed solution by people who don't understand
optimization and refuse to cooperate with compiler writers.
In general, asm statements renders a
program "unoptimizable".
2002-02-06 07:43:56 UTC in Open Research Compiler - Aurora
-
Currently the edge profiling feedback file can only be
dumped to the trace file during compilation. This is
unacceptable because it is too inconvenient and too
cumbersome.
We need a separate tool to dump the profiling feedback
file no matter what the feedback file contains: edge
profiling, value profiling or both.
This tool should be included in the project makefile.
A new...
2002-02-04 07:30:35 UTC in Open Research Compiler - Aurora
