|
Packit |
577717 |
2010-09-30
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: configure, configure.in: When --with-OS=CLE is enabled,
|
|
Packit |
577717 |
check the kernel version and use perfmon2 for old kernels and
|
|
Packit |
577717 |
perf_events for new kernels.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: configure, configure.in: If no sources of perf counters are
|
|
Packit |
577717 |
available, then use the generic_platform substrate instead.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Currently the code would always fall back on perfctr even if no
|
|
Packit |
577717 |
perfctr support was available.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: configure, configure.in: If you specify --with-perf-events
|
|
Packit |
577717 |
or --with-pe-include but the required perf_event.h header is not
|
|
Packit |
577717 |
available, then have configure fail with an error.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* papi.spec: Bump version number to 4.1.1 in affected files. Also
|
|
Packit |
577717 |
bump requirement for kernel from 2.6.31 to 2.6.32. This in prep
|
|
Packit |
577717 |
for the pending release.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: configure, Makefile.in, configure.in, papi.h: Bump version
|
|
Packit |
577717 |
number to 4.1.1 in affected files. This in prep for the pending
|
|
Packit |
577717 |
release.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* INSTALL.txt: Hope this late commit doesn't interfere with
|
|
Packit |
577717 |
anything. This updates the INSTALL.txt to reflect all of the
|
|
Packit |
577717 |
improvements we've made to perf_event support since the last
|
|
Packit |
577717 |
release.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-09-29
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/Rules.pfm: The -Werror problem was still occurring on
|
|
Packit |
577717 |
ia64/perfmon compiles, as I hadn't updated Rules.pfm
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: configure, configure.in, perf_events.c, perf_events.h,
|
|
Packit |
577717 |
sys_perf_counter_open.c, sys_perf_event_open.c, syscalls.h:
|
|
Packit |
577717 |
Remove support for the perf_counter interface in kernel 2.6.31.
|
|
Packit |
577717 |
Now supports only the perf_event interface in kernel 2.6.32 and
|
|
Packit |
577717 |
above.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-09-22
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: Attempt to add mmtimer support to perf_events
|
|
Packit |
577717 |
substrate.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: multiplex.c, papi.c, papi_protos.h: The multiplex code
|
|
Packit |
577717 |
currently does not make a final adjustment at the time of
|
|
Packit |
577717 |
MPX_read(). This is to avoid the case where counts could be
|
|
Packit |
577717 |
decreasing if you have multiple reads returning estimated values
|
|
Packit |
577717 |
before the next actual counter read.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
While this code works to keep the results non-decreasing, it can
|
|
Packit |
577717 |
cause significant differences from expected results for final
|
|
Packit |
577717 |
reads, especially if many counters are being multiplexed.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This is seen in the sdsc-mpx test. It was failing occasionally
|
|
Packit |
577717 |
on some machines by having error of over 20% (the cutoff for a
|
|
Packit |
577717 |
test error) when multiplexing 11 events.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
What this fix does is to special case the PAPI_stop() case when
|
|
Packit |
577717 |
multiplexing is enabled, having the PAPI_stop() do a final
|
|
Packit |
577717 |
adjustment. The intermediate PAPI_read() case is not changed.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This fixes the sdsc-mpx case, while still passing the mendes-alt
|
|
Packit |
577717 |
case (which checks for non-decreasing values).
|
|
Packit |
577717 |
|
|
Packit |
577717 |
There is a #define that can be set in multiplex.c to restore the
|
|
Packit |
577717 |
previous behavior.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/mendes-alt.c: This is our only test that checks to see
|
|
Packit |
577717 |
if multiplexed values are non-decreasing or not. Unfortunately
|
|
Packit |
577717 |
the test currently doesn't fail if values do go backward.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This change causes the test to fail if it finds multiplexed
|
|
Packit |
577717 |
counts that decrease.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-09-17
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/libpfm-3.y/: config.mk, lib/intel_wsm_events.h: Fix conflicts
|
|
Packit |
577717 |
from merge.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-09-15
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: Makefile.inc, Rules.perfctr-pfm, Rules.pfm_pe: Finally fix
|
|
Packit |
577717 |
the -WExtra problem.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
The issue was -WExtra was being passed to libpfm, but only in the
|
|
Packit |
577717 |
case where the user had a CFLAGS env variable.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
It turns out this is due to the following from section 5.7.2 of
|
|
Packit |
577717 |
the gmake manual: Except by explicit request, make exports a
|
|
Packit |
577717 |
variable only if it is either defined in the environment
|
|
Packit |
577717 |
initially or set on the command line,
|
|
Packit |
577717 |
|
|
Packit |
577717 |
And the fix is also described: If you want to prevent a
|
|
Packit |
577717 |
variable from being exported, use the unexport directive,
|
|
Packit |
577717 |
|
|
Packit |
577717 |
So I've added an "unexport CFLAGS" directive, which seems to be
|
|
Packit |
577717 |
the right thing as our Makefile explicitly passes CFLAGS to the
|
|
Packit |
577717 |
sub-Makefiles that need it. This seems to fix the build.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-09-13
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/libpfm-3.y/: docs/man3/libpfm_westmere.3,
|
|
Packit |
577717 |
lib/intel_wsm_events.h, lib/intel_wsm_unc_events.h,
|
|
Packit |
577717 |
lib/pfmlib_intel_nhm.c, lib/pfmlib_priv.h: Fix the missing files
|
|
Packit |
577717 |
from the import (CVS claims this as a "conflict")
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-09-08
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/Makefile.inc: Fixed the recipies for [c|f]tests and utils.
|
|
Packit |
577717 |
$(LIBRARY) => $(papiLIBS) (this way we don't build libpapi.a if
|
|
Packit |
577717 |
we won't want it)
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-09-03
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/sdsc.c: Had a "%d" instead of "%lld" in that last
|
|
Packit |
577717 |
commit.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/sdsc.c: Give a more detailed error message on the
|
|
Packit |
577717 |
sdsc-mpx test.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
We're seeing sporadic failures (probably due to results being
|
|
Packit |
577717 |
close to the threshold value) but it's hard to tell on buildbot
|
|
Packit |
577717 |
which counter is failing because the error message didn't print
|
|
Packit |
577717 |
the value.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-09-02
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/papi.c: Remove code that reported ENOSUPP if HW multiplexing
|
|
Packit |
577717 |
is not available.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
PAPI can automatically perform SW multiplexing if HW is not
|
|
Packit |
577717 |
available.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
With this part of my previous multiplexing patch reverted,
|
|
Packit |
577717 |
multiplexing seems to work even on 2.6.32 perf_events (by
|
|
Packit |
577717 |
reverting to SW mode on those machines)
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-31
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: Explicitly set the disabled flag to zero in
|
|
Packit |
577717 |
perf_events for new events.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
It was possible with an event set that if you removed an event
|
|
Packit |
577717 |
then added a new one that the disabled flag was obtaining the
|
|
Packit |
577717 |
value from the previously removed event.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This fix doesn't seem to break anything, but the code involved is
|
|
Packit |
577717 |
a bit tricky to follow.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This fixes the sdsc4-mpx test on sol.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/components/coretemp/: Rules.coretemp, linux-coretemp.c,
|
|
Packit |
577717 |
linux-coretemp.h: Initial stab at a coretemp component.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This component exposes every thing that looks like a useful file
|
|
Packit |
577717 |
under /sys/class/hwmon.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-30
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: F_SETOWN_EX is not available until 2.6.32, so
|
|
Packit |
577717 |
don't use it unless we are running on a recent enough kernel.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: Pentium 4 was not supported by perf_events
|
|
Packit |
577717 |
until version 2.6.35. Print an error if we attempt to use it on
|
|
Packit |
577717 |
an older kernel.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-27
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/overflow_allcounters.c: The "overflow_allcounters"
|
|
Packit |
577717 |
test failed on perfmon2 kernels because the behavior of a
|
|
Packit |
577717 |
counter on overflow differs between the various substrates.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Therefore detect if we're running on perfmon2 and print a
|
|
Packit |
577717 |
warning, but still pass the test.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/libpfm-3.y/lib/: intel_wsm_events.h, intel_wsm_unc_events.h,
|
|
Packit |
577717 |
pfmlib_intel_nhm.c, pfmlib_priv.h: updating
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/libpfm-3.y/docs/man3/libpfm_westmere.3: removing westmere
|
|
Packit |
577717 |
documentation
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: Fix warning in compile due to missing
|
|
Packit |
577717 |
parameter in a debug statement.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/test_utils.c: In the ctests, test_skip() was
|
|
Packit |
577717 |
attempting a PAPI_shutdown() before exiting. On multithreaded
|
|
Packit |
577717 |
tests (that had already spawned threads before the decision to
|
|
Packit |
577717 |
skip) this really causes the programs to end up confused and
|
|
Packit |
577717 |
reports spurious memory errors.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
So remove the PAPI_shutdown() from test_skip(). There's a
|
|
Packit |
577717 |
comment in test_fail() that indicates this was already done
|
|
Packit |
577717 |
there for similar reasons.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-26
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/byte_profile.c: byte_profile was failing on systems
|
|
Packit |
577717 |
where fp_ops is a derived event.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
modify the test so it gives a warning instead of failing and
|
|
Packit |
577717 |
avoids using the derived event.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: At PAPI_stop() time a counter with overflow
|
|
Packit |
577717 |
enabled is being adjusted by a value equal to the sampling
|
|
Packit |
577717 |
period.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
It looks like this isn't needed (and is generating an overcount
|
|
Packit |
577717 |
that breaks overflow_allcounters).
|
|
Packit |
577717 |
|
|
Packit |
577717 |
I'm still checking up on this code; if it turns out to be
|
|
Packit |
577717 |
necessary I may have ro revert this later.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/overflow_allcounters.c: Add validation check to
|
|
Packit |
577717 |
overflow_allcounters
|
|
Packit |
577717 |
|
|
Packit |
577717 |
It turns out perf_event kernels overcount overflows for some
|
|
Packit |
577717 |
reason, while perfctr doesn't. I'm investigating.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/: overflow_allcounters.c, papi_test.h, test_utils.c:
|
|
Packit |
577717 |
On Power5 and Power6, hardware counters 5 and 6 cannot generate
|
|
Packit |
577717 |
interrupts.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This means the overflow_allcounters test was failing because
|
|
Packit |
577717 |
overflow could not be generated for events 5 and 6.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Add code that special cases Power5 and Power6 for this test (and
|
|
Packit |
577717 |
generate a warning)
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: Change some debug messages to be warnings
|
|
Packit |
577717 |
instead of errors.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: papi.c, ctests/second.c: Fix ctests/second on bluegrass
|
|
Packit |
577717 |
(POWER6)
|
|
Packit |
577717 |
|
|
Packit |
577717 |
The test was testing domains by trying
|
|
Packit |
577717 |
PAPI_DOM_ALL^PAPI_DOM_SUPERVISOR in an attempt to turn off the
|
|
Packit |
577717 |
SUPERVISOR bit. This fails on Power6 as it leaves the
|
|
Packit |
577717 |
PAPI_DOM_OTHER bit set, which isn't allowed.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
How did the test earlier measure PAPI_DOM_ALL then, which has all
|
|
Packit |
577717 |
bits set? Well it turns out papi.c silently corrects
|
|
Packit |
577717 |
PAPI_DOM_ALL to be available_domains. But if you fiddle any of
|
|
Packit |
577717 |
the bits this correction is lost. This is probably not the
|
|
Packit |
577717 |
right thing to do, but the best way to fix it is not clear.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
For now this modifies the "second" test to clear the DOM_OTHER
|
|
Packit |
577717 |
bit too if the domain setting fails with it set.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-25
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: papi.c, papi.h, perf_events.c, ctests/kufrin.c,
|
|
Packit |
577717 |
ctests/mendes-alt.c, ctests/multiplex1.c,
|
|
Packit |
577717 |
ctests/multiplex1_pthreads.c, ctests/multiplex2.c,
|
|
Packit |
577717 |
ctests/multiplex3_pthreads.c, ctests/sdsc.c, ctests/sdsc2.c,
|
|
Packit |
577717 |
ctests/sdsc4.c, ftests/fmultiplex1.F, ftests/fmultiplex2.F: Add
|
|
Packit |
577717 |
support for including the OS version in the component_info_t
|
|
Packit |
577717 |
struct.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Use this support under perf_events to disable multiplexing
|
|
Packit |
577717 |
support if the kernel is < 2.6.33
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Modify the various multiplexing tests to "skip" if they get a
|
|
Packit |
577717 |
PAPI_ENOSUPP when attempting to set up multiplexing.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/all_native_events.c: Update all_native_events ctest to
|
|
Packit |
577717 |
print warning in the case where we skip events because they
|
|
Packit |
577717 |
aren't implemented yet (offcore and uncore mostly).
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-24
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/: papi_test.h, profile.c, test_utils.c: Adds a new
|
|
Packit |
577717 |
"test_warn()" function for the ctests.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This allows you to let tests pass with a warning.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This is useful in cases where you don't want to forget that an
|
|
Packit |
577717 |
option needs implementing, but that the feature being missed
|
|
Packit |
577717 |
isn't important enough to fail the test.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
The first user of this is the "profile" test. We warn that
|
|
Packit |
577717 |
PAPI_PROFIL_RANDOM is not supported on perf_events.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: From what I can tell, on perf_events the
|
|
Packit |
577717 |
overflow PAPI_OVERFLOW_FORCE_SW case was improperly falling
|
|
Packit |
577717 |
through in _papi_pe_dispatch_timer() to also run the HARDWARE
|
|
Packit |
577717 |
code.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This meant that we were attempting to read non-existant hardware
|
|
Packit |
577717 |
overflow data, causing a lot of errors to be printed to the
|
|
Packit |
577717 |
screen.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This shows up in the overflow_force_software test
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/ctests/: ipc.c, multiplex2.c, multiplex3_pthreads.c,
|
|
Packit |
577717 |
test_utils.c: Some minor changes to the ctests.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
+ ipc -- fail if the reported IPC value is zero + multiplex2 --
|
|
Packit |
577717 |
fail if all 32 counter values report as zero + multiplex3_pthread
|
|
Packit |
577717 |
-- give up sooner if each counter returns zero. otherwise
|
|
Packit |
577717 |
the test can take upwards of an hour to finish and makes the
|
|
Packit |
577717 |
fan on my laptop sound like it's going to explode in the
|
|
Packit |
577717 |
process
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-20
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/Makefile.inc: Disable CFLAGS += $(EXTRA_CFLAGS) (-Wextra) for
|
|
Packit |
577717 |
now. This will get buildbot running again, and if I can manage
|
|
Packit |
577717 |
to figure out exactly what the Makefiles are doing I'll re-enable
|
|
Packit |
577717 |
it again.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: Add support for Pentium 4 under perf events.
|
|
Packit |
577717 |
This requires a 2.6.35 kernel. On p4 perf events requires a
|
|
Packit |
577717 |
special format for the raw event, so we modify the results from
|
|
Packit |
577717 |
libpfm3 to conform to what the kernel expects.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* release_procedure.txt: release_procedure updated to reflect files
|
|
Packit |
577717 |
to keep under /doc
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-18
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: Patch from Gary Mohr that allows PAPI on perf
|
|
Packit |
577717 |
events to catch permissions problems at the time of
|
|
Packit |
577717 |
configuration, rather than only appearing once papi_start() is
|
|
Packit |
577717 |
called.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Quick summary of changes: + Adds a check_permissions() routine
|
|
Packit |
577717 |
PERF_COUNT_HW_INSTRUCTIONS is used as the test event. +
|
|
Packit |
577717 |
check_permissions() is called during PAPI_ATTACH,
|
|
Packit |
577717 |
PAPI_CPU_ATTACH and PAPI_DOMAIN + Various "ctl" structures
|
|
Packit |
577717 |
renamed "pe_ctl" + Some minor debug changes
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-05
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/perf_events.c: Use F_SETOWN_EX instead of F_SETOWN in
|
|
Packit |
577717 |
tune_up_fd()
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This fixes a multi-thread overflow bug found with the Rice
|
|
Packit |
577717 |
test-suite.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
F_SETOWN_EX doesn't exist until Linux 2.6.32. We really need
|
|
Packit |
577717 |
some infrastructure that detects the running kernel at init time
|
|
Packit |
577717 |
and warns that things like F_SETOWN_EX, multiplexing, etc., are
|
|
Packit |
577717 |
unavailable if the kernel is too old.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-04
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: Makefile.inc, cpus.c, cpus.h, genpapifdef.c, papi.c,
|
|
Packit |
577717 |
papi.h, papi_defines.h, papi_internal.c, papi_internal.h,
|
|
Packit |
577717 |
perf_events.c, perf_events.h, threads.h: This is the
|
|
Packit |
577717 |
PAPI_CPU_ATTACH patch from Gary Mohr that also fixes a problem
|
|
Packit |
577717 |
with multiple event sets on perf events.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Changes by file:
|
|
Packit |
577717 |
|
|
Packit |
577717 |
papi.h
|
|
Packit |
577717 |
+ Add PAPI_CPU_ATTACHED
|
|
Packit |
577717 |
+ Add strutctures needed for CPU_ATTACH
|
|
Packit |
577717 |
Makefile.in
|
|
Packit |
577717 |
+ include the new cpus.c file
|
|
Packit |
577717 |
papi_internal.c
|
|
Packit |
577717 |
+ add call to _papi_hwi_shutdown_cpu() in
|
|
Packit |
577717 |
_papi_hwi_free_EventSet()
|
|
Packit |
577717 |
+ make remap_event_position() non-static
|
|
Packit |
577717 |
+ add_native_events() and remove_native_events() use
|
|
Packit |
577717 |
_papi_hwi_get_context()
|
|
Packit |
577717 |
+ _papi_hw_read() has some whitespace and debug message
|
|
Packit |
577717 |
changes,
|
|
Packit |
577717 |
and removes an extraneous loop index
|
|
Packit |
577717 |
papi_internal.h
|
|
Packit |
577717 |
+ a new CPUS_LOCK is added
|
|
Packit |
577717 |
+ cpuinfo struct added to various structures
|
|
Packit |
577717 |
+ an inline call called _papi_hwi_get_context() added
|
|
Packit |
577717 |
perf_events.h
|
|
Packit |
577717 |
+ a cpu_num field added to control_state_t
|
|
Packit |
577717 |
perf_events.c
|
|
Packit |
577717 |
+ open_pe_events() allows per-cpu counting,
|
|
Packit |
577717 |
additional debug was added
|
|
Packit |
577717 |
+ set_cpu() function added
|
|
Packit |
577717 |
+ new debug messages in set_granularity() and
|
|
Packit |
577717 |
_papi_pe_read()
|
|
Packit |
577717 |
+ _papi_pe_ctl() has PAPI_CPU_ATTACH code added
|
|
Packit |
577717 |
+ _papi_pe_update_control_state() has the default domain
|
|
Packit |
577717 |
set to be PAPI_DOM_USER instead of pe_ctl->domain
|
|
Packit |
577717 |
genpapifdef.c
|
|
Packit |
577717 |
+ PAPI_CPU_ATTACHED added
|
|
Packit |
577717 |
threads.h
|
|
Packit |
577717 |
+ an ESI field added to ThreadInfo_t
|
|
Packit |
577717 |
papi.c
|
|
Packit |
577717 |
+ many new ABIDBG() debug messages added
|
|
Packit |
577717 |
+ PAPI_start() updated to check for CPU_ATTACH conflicts,
|
|
Packit |
577717 |
has whitespace fixes, gets context now,
|
|
Packit |
577717 |
if dirty calls update_control_state()
|
|
Packit |
577717 |
+ PAPI_stop(), PAPI_reset(), PAPI_read(), PAPI_read_ts(),
|
|
Packit |
577717 |
PAPI_accum(), PAPI_write(), PAPI_cleanup_eventset(),
|
|
Packit |
577717 |
all use _papi_hwi_get_context() to get context
|
|
Packit |
577717 |
+ PAPI_read() has some braces added
|
|
Packit |
577717 |
+ PAPI_get_opt() and PAPI_set_opt() have CPU_ATTACHED code
|
|
Packit |
577717 |
added.
|
|
Packit |
577717 |
+ PAPI_overflow() and PAPI_sprofil() now report errors if
|
|
Packit |
577717 |
CPU_ATTACH enabled
|
|
Packit |
577717 |
cpus.c, cpus.h
|
|
Packit |
577717 |
+ New files based on threads.c and threads.h
|
|
Packit |
577717 |
|
|
Packit |
577717 |
I made some additional changes, based on warnings given by gcc
|
|
Packit |
577717 |
+ Added a few missing function prototypes in cpus.h + Update
|
|
Packit |
577717 |
PAPI_MAX_LOCK as it wasn't increased to handle the new
|
|
Packit |
577717 |
addition of CPUS_LOCK + Removed various variables and functions
|
|
Packit |
577717 |
reported as being unused.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-08-03
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: papi_internal.h, papi_lock.h: The option
|
|
Packit |
577717 |
--with-no-cpu-counters was not supported on AIX. This has been
|
|
Packit |
577717 |
fixed and works now. Also the get_{real|virt}_{cycles|usec}
|
|
Packit |
577717 |
implementations for AIX (checked in Jul 29) have now been tested
|
|
Packit |
577717 |
and work correctly.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-07-29
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: configure, configure.in, papi_lock.h, papi_vector.c: Added
|
|
Packit |
577717 |
AIX support for the get_{real|virt}_{cycles|usec} functions +++
|
|
Packit |
577717 |
Fortran tests are now compiling on AIX. Wrong compiler flags were
|
|
Packit |
577717 |
used for the AIX compilers.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-07-26
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/papi_events.csv: add PAPI_L1_DCM for atom
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/x86_cache_info.c: Update the x86 cache_info table.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
The data from this table now comes from figure 3-17 in the
|
|
Packit |
577717 |
Intel Architectures Software Reference Manual 2A (cpuid
|
|
Packit |
577717 |
instruction section)
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This fixes an issue on my Atom N270 machine where the L2 cache
|
|
Packit |
577717 |
was not reported.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-07-16
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* INSTALL.txt, src/perf_events.c, src/perf_events.h: Perf Events
|
|
Packit |
577717 |
now support attach and detach.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
The patch for supporting this was written by Gary Mohr
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/papi_events.csv: Add a few missing events to Nehalem, based
|
|
Packit |
577717 |
on reading Intel Volume 3b.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/papi_events.csv: Fix Westmere to not use L1D_ALL_REF:ANY
|
|
Packit |
577717 |
|
|
Packit |
577717 |
I tested this on a Nehalem which has the proper behavior,
|
|
Packit |
577717 |
unfortunately no Westmere here to test on.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: papi_events.csv, papi_pfm_events.c, perfctr-x86.c: Enable
|
|
Packit |
577717 |
support for having more than one CPU block with the same name in
|
|
Packit |
577717 |
the .csv file. This allows easier support for sharing events
|
|
Packit |
577717 |
between similar architectures.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
I *think* this is needed and *think* it shouldn't break anything,
|
|
Packit |
577717 |
but I might have to back it out.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Also fixes event support for Pentium Pro / Pentium III/ P6 on
|
|
Packit |
577717 |
perfmon2 and perf events kernels.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
Also fixed some confusion where perfctr called chips "Intel Core"
|
|
Packit |
577717 |
meaning Core Duo wheras pfmon called "Intel Core" meaning
|
|
Packit |
577717 |
Core2.
|
|
Packit |
577717 |
|
|
Packit |
577717 |
This was tested on actual Pentium Pro and PIII hardware (as well
|
|
Packit |
577717 |
as on a few Pentium 4 machines plus a Core2 machine)
|
|
Packit |
577717 |
|
|
Packit |
577717 |
2010-07-02
|
|
Packit |
577717 |
|
|
Packit |
577717 |
* src/: papi_hl.c, ctests/api.c: Added remaining low-level api
|
|
Packit |
577717 |
tests
|