* (speedup) addMemEvent: generate inline code to check whether a
memory access is within 16k of the stack pointer, and if so
don't bother to call the helper
* (speedup) find_Block_containing: cache the most recently seen 2
blocks, and check new references in them first. This gives a
worthwhile speedup.
* (fix) at the end of the run, merge stats from un-freed blocks
back into APs. This fixes misleading stats that cause un-freed
blocks to appear to not have been accessed at all.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11480
on bug #256600 because the original version turned out to somewhat
fragile across different glibc versions.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11479
hyper-optimised versions which defeat valgrind in recent versions
of glibc.
Fixes#256600 based on patch from Dmitry Djachenko.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11478
* looks_like_a_pointer: x86-linux: halve lower threshold value
to take account of evident address space randomisation done
by Fedora 14 on x86
* handle __NR_lgetxattr
* move __NR_mq_* entries to the correct place (tidyup only)
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11435
* change the per-block-byte freq count type from 8- to 16-bit
so as to reduce the misleadingness of eventual numbers for
frequently accessed blocks
* disable debug printing
* add command line parameters to control the number of APs shown
in the final output, and to control the sorting order
* show average block size for each AP
* avoid possible problems when retiring a block and merging its
per-byte access counts into the AP's counts
* add a trailer message giving some important hints to the user
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11434
filtering out thereof, so as to make Cachegrind and Callgrind
pass their regressiont tests on ARM-Linux.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11433
don't assert in their regtests on ARM. (Value is the same in both
ARM and Thumb mode, fortunately.)
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11432
is a heap profiler that is complementary to Massif. DHAT tracks heap
allocations, and connects which memory accesses are to which blocks.
It can find the following information:
* total allocation and max liveness
* average block lifetime (# instructions between allocation and
freeing)
* average number of reads and writes to each byte in the block
("access ratios")
* average of longest interval of non-access to a block, also
measured in instructions
* which fields of blocks are used a lot, and which aren't
(hot-field profiling)
Using these stats it is possible to identify allocation points with
the following characteristics:
* potential process-lifetime leaks (blocks allocated by the point just
accumulate, and are freed only at the end of the run)
* excessive turnover: points which chew through a lot of heap, even if
it is not held onto for very long
* excessively transient: points which allocate very short lived blocks
* useless or underused allocations: blocks which are allocated but not
completely filled in, or are filled in but not subsequently read.
* blocks which see extended periods of inactivity. Could these
perhaps be allocated later or freed sooner?
* blocks with inefficient layout (hot fields spread out over
multiple cache lines), or with alignment holes
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11431
behave more like the original proposal in #245535. This makes it
more flexible and general. Also rename it.
* new name is --fullpath-after=
* allow multiple instances of --fullpath-after=
* don't require the specified strings to be prefixes, only substrings
But retain the elegant backwards-compatibility trick in Bart's r11312
commit: if --fullpath-after= is not specified at all, then behave
exactly as before.
Fixes#245535. A mixture of patches from Bart Van Assche
(bart.vanassche@gmail.com), Alexander Potapenko (glider@google.com),
and me (integration and documentation).
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11430
- "*(int *)0 = " is apparently ignored by LLVM for who-knows-why
reason. Cast the zero to a volatile int * instead.
- remove an unused function that gcc failed to mention was unused
(why? because it was marked __attribute__((noreturn)) ?)
As an aside, clang/llvm-2.8 seemed to be able to successfully
compile Valgrind.
git-svn-id: svn://svn.valgrind.org/valgrind/trunk@11429
