Macro fusion merges two instructions to a single micro-op. Intel core
platform performs this hardware optimization under limited
circumstances.
For example, CMP + JCC can be "fused" and executed /retired together.
While with sampling this can result in the sample sometimes being on the
JCC and sometimes on the CMP. So for the fused instruction pair, they
could be considered together.
On Nehalem, fused instruction pairs:
cmp/test + jcc.
On other new CPU:
cmp/test/add/sub/and/inc/dec + jcc.
This patch adds an x86-specific function which checks if 2 instructions
are in a "fused" pair. For non-x86 arch, the function is just NULL.
Changelog:
v4: Move the CPU model checking to symbol__disassemble and save the CPU
family/model in arch structure.
It avoids checking every time when jump arrow printed.
v3: Add checking for Nehalem (CMP, TEST). For other newer Intel CPUs
just check it by default (CMP, TEST, ADD, SUB, AND, INC, DEC).
v2: Remove the original weak function. Arnaldo points out that doing it
as a weak function that will be overridden by the host arch doesn't
work. So now it's implemented as an arch-specific function.
Committer fix:
Do not access evsel->evlist->env->cpuid, ->env can be null, introduce
perf_evsel__env_cpuid(), just like perf_evsel__env_arch(), also used in
this function call.
The original patch was segfaulting 'perf top' + annotation.
But this essentially disables this fused instructions augmentation in
'perf top', the right thing is to get the cpuid from the running kernel,
left for a later patch tho.
Signed-off-by: Yao Jin <yao.jin@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1499403995-19857-2-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The annotate browser is divided into 2 frames. Left frame contains 3
columns (some platforms only have one column).
For example:
│26 int compute_flag()
│27 {
22.80 1.20 │ sub $0x8,%rsp
│25 int i;
│
│27 i = rand() % 2;
22.78 1.20 1 │ → callq rand@plt
While it's hard for user to understand what the data is.
This patch adds the titles "Percent", "IPC" and "Cycle" on columns.
Percent IPC Cycle │
│25 __attribute__((noinline))
│26 int compute_flag()
│27 {
22.80 1.20 │ sub $0x8,%rsp
│25 int i;
│
│27 i = rand() % 2;
22.78 1.20 1 │ → callq rand@plt
The titles are displayed at row 0 of annotate browser if row 0 doesn't
have values of percent, ipc and cycle.
Signed-off-by: Yao Jin <yao.jin@linux.intel.com>
Acked-by: Milian Wolff <milian.wolff@kdab.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Yao Jin <yao.jin@linux.intel.com>
Link: http://lkml.kernel.org/r/1493909895-9668-3-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The __hpp__sort_acc() sorts entries using callchain depth in order to
put callers above in children mode. But it assumed the callchain order
was callee-first. Now default (for children) is caller-first so the
order of entries is reverted.
For example, consider following case:
$ perf report --no-children
..l
# Overhead Command Shared Object Symbol
# ........ ....... ................... ..........................
#
99.44% a.out a.out [.] main
|
---main
__libc_start_main
_start
Then children mode should show 'start' above '__libc_start_main' since
it's the caller (parent) of the __libc_start_main. But it's reversed:
# Children Self Command Shared Object Symbol
# ........ ........ ....... ............... .....................
#
99.61% 0.00% a.out libc-2.25.so [.] __libc_start_main
99.61% 0.00% a.out a.out [.] _start
99.54% 99.44% a.out a.out [.] main
This patch fixes it.
# Children Self Command Shared Object Symbol
# ........ ........ ....... ............... .....................
#
99.61% 0.00% a.out a.out [.] _start
99.61% 0.00% a.out libc-2.25.so [.] __libc_start_main
99.54% 99.44% a.out a.out [.] main
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Milian Wolff <milian.wolff@kdab.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yao Jin <yao.jin@linux.intel.com>
Cc: kernel-team@lge.com
Link: http://lkml.kernel.org/r/20170524062129.32529-8-namhyung@kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Often it is interesting to know how costly a given source line is in
total. Previously, one had to build these sums manually based on all
addresses that pointed to the same source line. This patch introduces
srcline as a sort key, which will do the aggregation for us.
Paired with the recent addition of showing inline frames, this makes
perf report much more useful for many C++ work loads.
The following shows the new feature in action. First, let's show the
status quo output when we sort by address. The result contains many hist
entries that generate the same output:
~~~~~~~~~~~~~~~~
$ perf report --stdio --inline -g address
# Children Self Command Shared Object Symbol
# ........ ........ ............ ................... .........................................
#
99.89% 35.34% cpp-inlining cpp-inlining [.] main
|
|--64.55%--main complex:655
| /home/milian/projects/kdab/rnd/hotspot/tests/test-clients/cpp-inlining/main.cpp:39 (inline)
| /usr/include/c++/6.3.1/complex:664 (inline)
| |
| |--60.31%--hypot +20
| | |
| | |--8.52%--__hypot_finite +273
| | |
| | |--7.32%--__hypot_finite +411
...
--35.34%--_start +4194346
__libc_start_main +241
|
|--6.65%--main random.tcc:3326
| /home/milian/projects/kdab/rnd/hotspot/tests/test-clients/cpp-inlining/main.cpp:39 (inline)
| /usr/include/c++/6.3.1/bits/random.h:1809 (inline)
| /usr/include/c++/6.3.1/bits/random.h:1818 (inline)
| /usr/include/c++/6.3.1/bits/random.h:185 (inline)
|
|--2.70%--main random.tcc:3326
| /home/milian/projects/kdab/rnd/hotspot/tests/test-clients/cpp-inlining/main.cpp:39 (inline)
| /usr/include/c++/6.3.1/bits/random.h:1809 (inline)
| /usr/include/c++/6.3.1/bits/random.h:1818 (inline)
| /usr/include/c++/6.3.1/bits/random.h:185 (inline)
|
|--1.69%--main random.tcc:3326
| /home/milian/projects/kdab/rnd/hotspot/tests/test-clients/cpp-inlining/main.cpp:39 (inline)
| /usr/include/c++/6.3.1/bits/random.h:1809 (inline)
| /usr/include/c++/6.3.1/bits/random.h:1818 (inline)
| /usr/include/c++/6.3.1/bits/random.h:185 (inline)
...
~~~~~~~~~~~~~~~~
With this patch and `-g srcline` we instead get the following output:
~~~~~~~~~~~~~~~~
$ perf report --stdio --inline -g srcline
# Children Self Command Shared Object Symbol
# ........ ........ ............ ................... .........................................
#
99.89% 35.34% cpp-inlining cpp-inlining [.] main
|
|--64.55%--main complex:655
| /home/milian/projects/kdab/rnd/hotspot/tests/test-clients/cpp-inlining/main.cpp:39 (inline)
| /usr/include/c++/6.3.1/complex:664 (inline)
| |
| |--64.02%--hypot
| | |
| | --59.81%--__hypot_finite
| |
| --0.53%--cabs
|
--35.34%--_start
__libc_start_main
|
|--12.48%--main random.tcc:3326
| /home/milian/projects/kdab/rnd/hotspot/tests/test-clients/cpp-inlining/main.cpp:39 (inline)
| /usr/include/c++/6.3.1/bits/random.h:1809 (inline)
| /usr/include/c++/6.3.1/bits/random.h:1818 (inline)
| /usr/include/c++/6.3.1/bits/random.h:185 (inline)
...
~~~~~~~~~~~~~~~~
Signed-off-by: Milian Wolff <milian.wolff@kdab.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Yao Jin <yao.jin@linux.intel.com>
Link: http://lkml.kernel.org/r/20170318214928.9047-1-milian.wolff@kdab.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
If the address belongs to an inlined function, the source information
back to the first non-inlined function will be printed.
For example:
1. Show inlined function name
perf report -g function --inline
- 0.69% 0.00% inline ld-2.23.so [.] dl_main
- dl_main
0.56% _dl_relocate_object
_dl_relocate_object (inline)
elf_dynamic_do_Rela (inline)
2. Show the file/line information
perf report -g address --inline
- 0.69% 0.00% inline ld-2.23.so [.] _dl_start
_dl_start rtld.c:307
/build/glibc-GKVZIf/glibc-2.23/elf/rtld.c:413 (inline)
+ _dl_sysdep_start dl-sysdep.c:250
Signed-off-by: Yao Jin <yao.jin@linux.intel.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tested-by: Milian Wolff <milian.wolff@kdab.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@intel.com>
Link: http://lkml.kernel.org/r/1490474069-15823-6-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Commit 21e6d84286 ("perf diff: Use perf_hpp__register_sort_field
interface") changed list_add() to perf_hpp__register_sort_field().
This resulted in a behavior change since the field was added to the tail
instead of the head. So the -o option is mostly ignored due to its
order in the list.
This patch fixes it by adding perf_hpp__prepend_sort_field().
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Fixes: 21e6d84286 ("perf diff: Use perf_hpp__register_sort_field interface")
Link: http://lkml.kernel.org/r/20170118051457.30946-2-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The -o/--order option is to select column number to sort a diff result.
It does the job by adding a hpp field at the beginning of the sort list.
But it should not be added to the output field list as it has no
callbacks required by a output field.
During the setup_sorting(), the perf_hpp__setup_output_field() appends
the given sort keys to the output field if it's not there already.
Originally it was checked by fmt->list being non-empty. But commit
3f931f2c42 ("perf hists: Make hpp setup function generic") changed it
to check the ->equal callback.
Anyways, we don't need to add the pseudo hpp field to the output field
list since it won't be used for output. So just skip fields if they
have no ->color or ->entry callbacks.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Fixes: 3f931f2c42 ("perf hists: Make hpp setup function generic")
Link: http://lkml.kernel.org/r/20170118051457.30946-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The use_browser and perf_version_string variables are both declared in
perf.c but they are also referenced by other functions of libperf.a.
Therefore a user linking an own main() with libperf.a must declare those
two variables in their files even if the files never use the browser or
the version information.
This patch fixes this issue by moving use_browser and
perf_version_string out of perf.c to some other files.
Signed-off-by: Soramichi Akiyama <akiyama@m.soramichi.jp>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20170117002237.c1aec0ce3b4d675dca018deb@m.soramichi.jp
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
If jump target is outside of function range, perf is not handling it
correctly. Especially when target address is lesser than function start
address, target offset will be negative. But, target address declared to
be unsigned, converts negative number into 2's complement. See below
example. Here target of 'jumpq' instruction at 34cf8 is 34ac0 which is
lesser than function start address(34cf0).
34ac0 - 34cf0 = -0x230 = 0xfffffffffffffdd0
Objdump output:
0000000000034cf0 <__sigaction>:
__GI___sigaction():
34cf0: lea -0x20(%rdi),%eax
34cf3: cmp -bashx1,%eax
34cf6: jbe 34d00 <__sigaction+0x10>
34cf8: jmpq 34ac0 <__GI___libc_sigaction>
34cfd: nopl (%rax)
34d00: mov 0x386161(%rip),%rax # 3bae68 <_DYNAMIC+0x2e8>
34d07: movl -bashx16,%fs:(%rax)
34d0e: mov -bashxffffffff,%eax
34d13: retq
perf annotate before applying patch:
__GI___sigaction /usr/lib64/libc-2.22.so
lea -0x20(%rdi),%eax
cmp -bashx1,%eax
v jbe 10
v jmpq fffffffffffffdd0
nop
10: mov _DYNAMIC+0x2e8,%rax
movl -bashx16,%fs:(%rax)
mov -bashxffffffff,%eax
retq
perf annotate after applying patch:
__GI___sigaction /usr/lib64/libc-2.22.so
lea -0x20(%rdi),%eax
cmp -bashx1,%eax
v jbe 10
^ jmpq 34ac0 <__GI___libc_sigaction>
nop
10: mov _DYNAMIC+0x2e8,%rax
movl -bashx16,%fs:(%rax)
mov -bashxffffffff,%eax
retq
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Chris Riyder <chris.ryder@arm.com>
Cc: Kim Phillips <kim.phillips@arm.com>
Cc: Markus Trippelsdorf <markus@trippelsdorf.de>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Taeung Song <treeze.taeung@gmail.com>
Cc: linuxppc-dev@lists.ozlabs.org
Link: http://lkml.kernel.org/r/1480953407-7605-3-git-send-email-ravi.bangoria@linux.vnet.ibm.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
New tool:
- 'perf sched timehist' provides an analysis of scheduling events.
Example usage:
perf sched record -- sleep 1
perf sched timehist
By default it shows the individual schedule events, including the wait
time (time between sched-out and next sched-in events for the task), the
task scheduling delay (time between wakeup and actually running) and run
time for the task:
time cpu task name wait time sch delay run time
[tid/pid] (msec) (msec) (msec)
-------- ------ ---------------- --------- --------- --------
1.874569 [0011] gcc[31949] 0.014 0.000 1.148
1.874591 [0010] gcc[31951] 0.000 0.000 0.024
1.874603 [0010] migration/10[59] 3.350 0.004 0.011
1.874604 [0011] <idle> 1.148 0.000 0.035
1.874723 [0005] <idle> 0.016 0.000 1.383
1.874746 [0005] gcc[31949] 0.153 0.078 0.022
...
Times are in msec.usec. (David Ahern, Namhyung Kim)
Improvements:
- Make 'perf c2c report' support -f/--force, to allow skipping the
ownership check for root users, for instance, just like the other
tools (Jiri Olsa)
- Allow sorting cachelines by total number of HITMs, in addition to
local and remote numbers (Jiri Olsa)
Fixes:
- Make sure errors aren't suppressed by the TUI reset at the end of
a 'perf c2c report' session (Jiri Olsa)
Infrastructure changes:
- Initial work on having the annotate code better support multiple
architectures, including the ability to cross-annotate, i.e. to
annotate perf.data files collected on an ARM system on a x86_64
workstation (Arnaldo Carvalho de Melo, Ravi Bangoria, Kim Phillips)
- Use USECS_PER_SEC instead of hard coded number in libtraceevent (Steven Rostedt)
- Add retrieval of preempt count and latency flags in libtraceevent (Steven Rostedt)
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
If the branch is 100% predicted then the "predicted" is hidden.
Similarly, if there is no branch tsx abort, the "abort" is hidden.
There is only cycles shown (cycle is supported on skylake platform,
older platform would be 0).
If no iterations, the "iterations" is hidden.
For example:
|--29.93%--main div.c:39 (predicted:50.6%, cycles:1, iterations:18)
| main div.c:44 (predicted:50.6%, cycles:1)
| |
| --22.69%--main div.c:42 (cycles:2, iterations:17)
| compute_flag div.c:28 (cycles:2)
| |
| --10.52%--compute_flag div.c:27 (cycles:1)
| rand rand.c:28 (cycles:1)
| rand rand.c:28 (cycles:1)
| __random random.c:298 (cycles:1)
| __random random.c:297 (cycles:1)
| __random random.c:295 (cycles:1)
| __random random.c:295 (cycles:1)
| __random random.c:295 (cycles:1)
| __random random.c:295 (cycles:6)
Signed-off-by: Yao Jin <yao.jin@linux.intel.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Linux-kernel@vger.kernel.org
Cc: Yao Jin <yao.jin@linux.intel.com>
Link: http://lkml.kernel.org/r/1477876794-30749-5-git-send-email-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
When horizontal scrolling is used in hierarchy mode, the folded signed
disappears at the right most column.
Committer note:
To test it, run 'perf top --hierarchy, see the '+' symbol at the first
column, then press the right arrow key, the '+' symbol will disappear,
this patch fixes that.
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Tested-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20161108130833.9263-3-namhyung@kernel.org
[ Move 'width -= 2' invariant to right after the if/else ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The perf report/top on TUI supports horizontal scrolling using LEFT and
RIGHT keys.
But it calculate the number of columns incorrectly when hierarchy mode
is enabled so that keep pressing RIGHT key can make the output
disappeared.
In the hierarchy mode, all sort keys are collapsed into a single column,
so it needs to be applied when calculating column numbers.
Reported-and-Tested-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20161024162110.17918-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The perf report/top on TUI supports horizontal scrolling using LEFT and
RIGHT keys.
But it calculate the number of columns incorrectly when hierarchy mode
is enabled so that keep pressing RIGHT key can make the output
disappeared.
In the hierarchy mode, all sort keys are collapsed into a single column,
so it needs to be applied when calculating column numbers.
Reported-and-Tested-by: Markus Trippelsdorf <markus@trippelsdorf.de>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20161024162110.17918-1-namhyung@kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Here is a small patch which tries to fulfill a point in the perf todo
list:
* Make pressing 'V' multiple times to go on cycling thru various
verbosity levels in 'perf top', so that info that is present in
'perf top -v' can be obtained without having to restart the tool
(acme).
After a small grep in the code, the max verbosity level seems 3; so,
we cycle at 4; I did not dare define a MAX_VERBOSE_LEVEL constant.
Signed-off-by: Alexis Berlemont <alexis.berlemont@gmail.com>
Suggested-and-Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20161012214823.14324-2-alexis.berlemont@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
