Callgrind, Cachegrind, and Lackey call
helpers for memory accesses in bunches, to reduce
register save/restore overhead (and merge load/store
within same instruction into a "modify" event).
The calls should not be done within a RMW section
enclosed by LL/SC instructions, as this reduces the
chance of SC to succeed, and can result in hangs.
For Callgrind, this definitly helped MIPS, and was
committed in r13136. Do the same for Cachegrind/Lackey.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@13143
Because most Ir accesses touch only one line, and this
can be detected at instrumentation time, use a special
handler for that. This handler does not need to check
cache line crossing at runtime.
This does not change the results of the simulator at all,
but improves runtime by around 15% on perf benchmarks.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@13095
Makes modifications simpler in the future, and
should result in the same code.
This patch was already discussed some time ago on the
dev mailing list, and did not make a difference on
various architectures.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@13024
Option sp-at-mem-access can be used by tools which do not
need an up to date BP and IP at each mem access.
It is needed however to have SP up to date at each memory
access, as an up to date SP is needed to grow the stack in
m_signals.c
Tools massif, cachegrind and callgrind are using sp-at-mem-access
as default.
None tool could also use sp-at-mem-access but default is kept
to unwindregs-at-mem-access (similar to memcheck, drd, helgrind, exp-sgcheck).
exp-dhat, exp-bbv, lackey have not been looked at to see if they
could make use of sp-at-mem-access.
Validated on x86, amd64, ppc64 and s390x.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@12872
cachegrind: use the new function to abort startup if the minumum line
size is smaller than the size of the largest guest register.
Partially derived from a patch by Josef Weidendorfer.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@12605
Patch by Philippe Waroquiers, slightly changed.
This actually was a regression from 3.6.1, but the patch
also improves on printed messages, and refactors common
code between cachegrind and callgrind.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@12013
of the L2 cache. This is to accommodate machines with three levels of
cache. We still only simulate two levels, the first and the last.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11404
* do indirect branch prediction simulation on calls
via function pointers
* only call into conditional branch prediction simulation
on real guest code branches (eg. not for VEX emulation of some
instructions using branches of jumpkind Ijk_EmWarn)
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11331
messages now begin with "valgrind: ", and they're more often printed before
the preamble. This required introducing a new message kind, Vg_FailMsg, and
functions VG_(fmsg) and VG_(fmsg_bad_option), and removing
VG_(err_bad_option).
Where we used to have horrible output like this:
[ocean:~/grind/ws2] vg5 --tool=massif --threshold=101 date
==31877== Massif, a heap profiler
==31877== Copyright (C) 2003-2010, and GNU GPL'd, by Nicholas Nethercote
==31877== Using Valgrind-3.6.0.SVN and LibVEX; rerun with -h for copyright info
==31877== Command: date
==31877==
==31877== --threshold must be between 0.0 and 100.0
valgrind: Bad option '--threshold'; aborting.
valgrind: Use --help for more information.
We now have nice output like this:
[ocean:~/grind/ws2] vg2 --tool=massif --threshold=101 date
valgrind: Bad option: --threshold=101
valgrind: --threshold must be between 0.0 and 100.0
valgrind: Use --help for more information or consult the user manual.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11209
* For all tools and the core, don't show statistics when -v is in
effect. Instead, try to restrict -v to mostly user-useful
stuff.
* A new flag --stats=no|yes [no] produces statistics output instead.
* Fix longstanding problem in that Memcheck's leak checker ran after
the core's error manager module shut down, thereby not showing use
counts of leak suppressions. This fixes#186790.
* As a consequence, the leak checker text output of Memcheck has
changed a bit -- leak check is done before the final error
summary is done (much more logical), and the output has been
tidied up a bit.
* Helgrind, Drd and Ptrcheck now also print "For counts of
detected and suppressed errors, rerun with: -v", which makes
them consistent with Memcheck in this regard. These are
filtered out by the regtest filter scripts.
For all tools except Memcheck, the regtests are unchanged. On
Memcheck regtests still fail due to rearrangements of the leak
checker output. This will be fixed by a followup commit.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@10746
This branch adds proper support for atomic instructions, proper in the
sense that the atomicity is preserved through the compilation
pipeline, and thus in the instrumented code.
These changes track the IR changes added by vex r1901. They primarily
update the instrumentation functions in all tools to handle the
changes, with the exception of exp-ptrcheck, which needs some further
work in order to be able to run threaded code.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@10392
I tried using 'svn merge' to do the merge but it did a terrible job and
there were bazillions of conflicts. So instead I just took the diff between
the branch and trunk at r10155, applied the diff to the trunk, 'svn add'ed
the added files (no files needed to be 'svn remove'd) and committed.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@10156
DARWIN branch. A big ugly DARWIN/trunk sync commit, mostly to do with
changing the representation of SysRes and vki_sigset_t. Functionality of
the trunk shouldn't be changed by it.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@9876
Some of our option processing code uses it. This means that eg.
'--log-fd=9xxx' logs to fd 9, and '--log-fd=blahblahblah' logs to 0 (because
atoll() returns 0 if the string doesn't contain a number!)
It turns out that most of our option processing uses VG_(strtoll*) instead
of VG_(atoll). The reason that not all of it does is that the
option-processing macros are underpowered -- they currently work well if you
just want to assign the value to a variable, eg:
VG_BOOL_CLO(arg, "--heap", clo_heap)
else VG_BOOL_CLO(arg, "--stacks", clo_stacks)
else VG_NUM_CLO(arg, "--heap-admin", clo_heap_admin)
else VG_NUM_CLO(arg, "--depth", clo_depth)
(This works because they are actually an if-statement, but it looks odd.)
VG_NUM_CLO uses VG_(stroll10). But if you want to do any checking or
processing, you can't use those macros, leading to code like this:
else if (VG_CLO_STREQN(9, arg, "--log-fd=")) {
log_to = VgLogTo_Fd;
VG_(clo_log_name) = NULL;
tmp_log_fd = (Int)VG_(atoll)(&arg[9]);
}
So this commit:
- Improves the *_CLO_* macros so that they can be used in all circumstances.
They're now just expressions (albeit ones with side-effects, setting the
named variable appropriately). Thus they can be used as if-conditions,
and any post-checking or processing can occur in the then-statement. And
malformed numeric arguments (eg. --log-fd=foo) aren't accepted. This also
means you don't have to specify the lengths of any option strings anywhere
(eg. the 9 in the --log-fd example above). The use of a wrong number
caused at least one bug, in Massif.
- Updates all places where the macros were used.
- Updates Helgrind to use the *_CLO_* macros (it didn't use them).
- Updates Callgrind to use the *_CLO_* macros (it didn't use them), except
for the more esoteric option names (those with numbers in the option
name). This allowed getUInt() and getUWord() to be removed.
- Improves the cache option parsing in Cachegrind and Callgrind -- now uses
VG_(strtoll10)(), detects overflow, and is shorter.
- Uses INT instead of NUM in the macro names, to distinguish better vs. the
DBL macro.
- Removes VG_(atoll*) and the few remaining uses -- they're wretched
functions and VG_(strtoll*) should be used instead.
- Adds the VG_STREQN macro.
- Changes VG_BINT_CLO and VG_BHEX_CLO to abort if the given value is outside
the range -- the current silent truncation is likely to cause confusion as
much as anything.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@9255
The number of sets, ie. number of cache lines divided by associativity,
and the cache line size still have to be powers of two.
This change is needed for default cache parameters used on some Intel
Core 2 and Atom processors.
Includes cachegrind manual update and explicit tests with 24KB D1/3MB L2
Reverts addition of 6MB warning to {cachegrind,callgrind}/tests/filter_stderr
Backporting to VALGRIND_3_4_BRANCH needs r8912
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@9080
relatively minor extensions to m_debuginfo, a major overhaul of
m_debuginfo/readdwarf3.c to get its space usage under control, and
changes throughout the system to enable heap-use profiling.
The majority of the merged changes were committed into
branches/PTRCHECK as the following revs: 8591 8595 8598 8599 8601 and
8161.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@8621
wrote into the parent's output file even if %p was specified.
Josef, I think Callgrind does not have this bug, but you might want to say
something about forking in the manual, as I have done for Massif and
Cachegrind.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@8154
interface and provides full power; and "OSetWord_", which is an
easier-to-use interface for if you just want to store words.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@6841
VG_(record_startup_wd) which records the working directory at startup,
and VG_(get_startup_wd) which later tells you what value was recorded.
This works because all uses of VG_(getcwd) serve only to record the
directory at process start anyway. The motivation is that AIX does
not support sys_getcwd directly, so it's easier for the launcher to
ship in the required value using an environment variable. On Linux
sys_getcwd is used as before.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@6764
--branch-sim=yes is specified, Cachegrind simulates a simple indirect
branch predictor and a conditional branch predictor. The latter
considers both the branch instruction's address and the behaviour of
the last few conditional branches. Return stack prediction is not
modelled.
The new counted events are: conditional branches (Bc), mispredicted
conditional branches (Bcm), indirect branches (Bi) and mispredicted
indirect branches (Bim). Postprocessing tools (cg_annotate, cg_merge)
handle the new events as you would expect. Note that branch
simulation is not enabled by default as it gives a 20%-25% slowdown,
so you need to ask for it explicitly using --branch-sim=yes.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@6733
basename to be something other than "cachegrind.out".
* Observe the core-supplied --log-file-qualifier, if specified,
in creation of output file names.
* To make the above work, move most of the stuff in cg_pre_clo_init
into cg_post_clo_init, so that the core's determination of the
log file qualifier, if any, is done by the time cachegrind comes
to process its arguments.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@6572
