This patch prints the stddev and hist for the cycles diff of program
block. It can help us to understand if the cycles is noisy or not.
This patch is inspired by Andi Kleen's patch:
https://lwn.net/Articles/600471/
We create new option '--cycles-hist'.
Example:
perf record -b ./div
perf record -b ./div
perf diff -c cycles
# Baseline [Program Block Range] Cycles Diff Shared Object Symbol
# ........ .......................................................... .... ................. ............................
#
46.72% [div.c:40 -> div.c:40] 0 div [.] main
46.72% [div.c:42 -> div.c:44] 0 div [.] main
46.72% [div.c:42 -> div.c:39] 0 div [.] main
20.54% [random_r.c:357 -> random_r.c:394] 1 libc-2.27.so [.] __random_r
20.54% [random_r.c:357 -> random_r.c:380] 0 libc-2.27.so [.] __random_r
20.54% [random_r.c:388 -> random_r.c:388] 0 libc-2.27.so [.] __random_r
20.54% [random_r.c:388 -> random_r.c:391] 0 libc-2.27.so [.] __random_r
17.04% [random.c:288 -> random.c:291] 0 libc-2.27.so [.] __random
17.04% [random.c:291 -> random.c:291] 0 libc-2.27.so [.] __random
17.04% [random.c:293 -> random.c:293] 0 libc-2.27.so [.] __random
17.04% [random.c:295 -> random.c:295] 0 libc-2.27.so [.] __random
17.04% [random.c:295 -> random.c:295] 0 libc-2.27.so [.] __random
17.04% [random.c:298 -> random.c:298] 0 libc-2.27.so [.] __random
8.40% [div.c:22 -> div.c:25] 0 div [.] compute_flag
8.40% [div.c:27 -> div.c:28] 0 div [.] compute_flag
5.14% [rand.c:26 -> rand.c:27] 0 libc-2.27.so [.] rand
5.14% [rand.c:28 -> rand.c:28] 0 libc-2.27.so [.] rand
2.15% [rand@plt+0 -> rand@plt+0] 0 div [.] rand@plt
0.00% [kernel.kallsyms] [k] __x86_indirect_thunk_rax
0.00% [do_mmap+714 -> do_mmap+732] -10 [kernel.kallsyms] [k] do_mmap
0.00% [do_mmap+737 -> do_mmap+765] 1 [kernel.kallsyms] [k] do_mmap
0.00% [do_mmap+262 -> do_mmap+299] 0 [kernel.kallsyms] [k] do_mmap
0.00% [__x86_indirect_thunk_r15+0 -> __x86_indirect_thunk_r15+0] 7 [kernel.kallsyms] [k] __x86_indirect_thunk_r15
0.00% [native_sched_clock+0 -> native_sched_clock+119] -1 [kernel.kallsyms] [k] native_sched_clock
0.00% [native_write_msr+0 -> native_write_msr+16] -13 [kernel.kallsyms] [k] native_write_msr
When we enable the option '--cycles-hist', the output is
perf diff -c cycles --cycles-hist
# Baseline [Program Block Range] Cycles Diff stddev/Hist Shared Object Symbol
# ........ .......................................................... .... ................. ................. ............................
#
46.72% [div.c:40 -> div.c:40] 0 ± 37.8% ▁█▁▁██▁█ div [.] main
46.72% [div.c:42 -> div.c:44] 0 ± 49.4% ▁▁▂█▂▂▂▂ div [.] main
46.72% [div.c:42 -> div.c:39] 0 ± 24.1% ▃█▂▄▁▃▂▁ div [.] main
20.54% [random_r.c:357 -> random_r.c:394] 1 ± 33.5% ▅▂▁█▃▁▂▁ libc-2.27.so [.] __random_r
20.54% [random_r.c:357 -> random_r.c:380] 0 ± 39.4% ▁▁█▁██▅▁ libc-2.27.so [.] __random_r
20.54% [random_r.c:388 -> random_r.c:388] 0 libc-2.27.so [.] __random_r
20.54% [random_r.c:388 -> random_r.c:391] 0 ± 41.2% ▁▃▁▂█▄▃▁ libc-2.27.so [.] __random_r
17.04% [random.c:288 -> random.c:291] 0 ± 48.8% ▁▁▁▁███▁ libc-2.27.so [.] __random
17.04% [random.c:291 -> random.c:291] 0 ±100.0% ▁█▁▁▁▁▁▁ libc-2.27.so [.] __random
17.04% [random.c:293 -> random.c:293] 0 ±100.0% ▁█▁▁▁▁▁▁ libc-2.27.so [.] __random
17.04% [random.c:295 -> random.c:295] 0 ±100.0% ▁█▁▁▁▁▁▁ libc-2.27.so [.] __random
17.04% [random.c:295 -> random.c:295] 0 libc-2.27.so [.] __random
17.04% [random.c:298 -> random.c:298] 0 ± 75.6% ▃█▁▁▁▁▁▁ libc-2.27.so [.] __random
8.40% [div.c:22 -> div.c:25] 0 ± 42.1% ▁▃▁▁███▁ div [.] compute_flag
8.40% [div.c:27 -> div.c:28] 0 ± 41.8% ██▁▁▄▁▁▄ div [.] compute_flag
5.14% [rand.c:26 -> rand.c:27] 0 ± 37.8% ▁▁▁████▁ libc-2.27.so [.] rand
5.14% [rand.c:28 -> rand.c:28] 0 libc-2.27.so [.] rand
2.15% [rand@plt+0 -> rand@plt+0] 0 div [.] rand@plt
0.00% [kernel.kallsyms] [k] __x86_indirect_thunk_rax
0.00% [do_mmap+714 -> do_mmap+732] -10 [kernel.kallsyms] [k] do_mmap
0.00% [do_mmap+737 -> do_mmap+765] 1 [kernel.kallsyms] [k] do_mmap
0.00% [do_mmap+262 -> do_mmap+299] 0 [kernel.kallsyms] [k] do_mmap
0.00% [__x86_indirect_thunk_r15+0 -> __x86_indirect_thunk_r15+0] 7 [kernel.kallsyms] [k] __x86_indirect_thunk_r15
0.00% [native_sched_clock+0 -> native_sched_clock+119] -1 ± 38.5% ▄█▁ [kernel.kallsyms] [k] native_sched_clock
0.00% [native_write_msr+0 -> native_write_msr+16] -13 ± 47.1% ▁█▇▃▁▁ [kernel.kallsyms] [k] native_write_msr
v8:
---
Rebase to perf/core branch
v7:
---
1. v6 got Jiri's ACK.
2. Rebase to latest perf/core branch.
v6:
---
1. Jiri provides better code for using data__hpp_register() in ui_init().
Use this code in v6.
v5:
---
1. Refine the use of data__hpp_register() in ui_init() according to
Jiri's suggestion.
v4:
---
1. Rename the new option from '--noisy' to '--cycles-hist'
2. Remove the option '-n'.
3. Only update the spark value and stats when '--cycles-hist' is enabled.
4. Remove the code of printing '..'.
v3:
---
1. Move the histogram to a separate column
2. Move the svals[] out of struct stats
v2:
---
Jiri got a compile error,
CC builtin-diff.o
builtin-diff.c: In function ‘compute_cycles_diff’:
builtin-diff.c:712:10: error: taking the absolute value of unsigned type ‘u64’ {aka ‘long unsigned int’} has no effect [-Werror=absolute-value]
712 | labs(pair->block_info->cycles_spark[i] -
| ^~~~
Because the result of u64 - u64 is still u64. Now we change the type of
cycles_spark[] to s64.
Signed-off-by: Jin Yao <yao.jin@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lore.kernel.org/lkml/20190925011446.30678-1-yao.jin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This will allow us to untangle the header dependency a bit more, as some places will not need event.h anymore. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: https://lkml.kernel.org/n/tip-enqncj29ovzaat3cd9203rwl@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Reducing the includes hell a bit more, speeding up the build and avoiding needless rebuilds when just one of those files gets updated. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: https://lkml.kernel.org/n/tip-u63el2vqsovsmnhebx1rcixo@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
To disentangle util/sort.h a bit more. Cc: Adrian Hunter <adrian.hunter@intel.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Namhyung Kim <namhyung@kernel.org> Link: https://lkml.kernel.org/n/tip-6kbf2cauas06rbqp15pyter5@git.kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Rename struct perf_evlist to struct evlist, so we don't have a name clash when we add struct perf_evlist in libperf. Committer notes: Added fixes to build on arm64, from Jiri and from me (tools/perf/util/cs-etm.c) Signed-off-by: Jiri Olsa <jolsa@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Alexey Budankov <alexey.budankov@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Michael Petlan <mpetlan@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/20190721112506.12306-6-jolsa@kernel.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The hist__account_cycles() can account cycles per basic block. The basic block information is saved in cycles_hist structure. This patch processes each symbol, get basic blocks from cycles_hist and add the basic block entries to a new hists (in 'struct block_hist'). Using a hists is because we need to compare, sort and print the basic blocks later. v6: --- Since 'ops' argument is removed from hists__add_entry_block, update the code accordingly. No functional change. v5: --- Since now we still carry block_info in 'struct hist_entry' we don't need to use our own new/free ops for hist entries. And the block_info is released in hist_entry__delete. v3: --- 1. In v2, we put block stuffs in 'struct hist_entry', but it's not a good design. In v3, we create a new 'struct block_hist' and cast the 'struct hist_entry' to 'struct block_hist' in some places, which can avoid adding new stuffs in 'struct hist_entry'. 2. abs() -> labs(), in block_cycles_diff_cmp(). v2: --- v1 adds the basic block entries to per data-file hists but v2 adds the basic block entries to per symbol hists. That is to keep current perf-diff format. Will show the result in next patches. Signed-off-by: Jin Yao <yao.jin@linux.intel.com> Reviewed-by: Jiri Olsa <jolsa@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Jin Yao <yao.jin@intel.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1561713784-30533-5-git-send-email-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
The block_info contains the program basic block information, i.e, contains the start address and the end address of this basic block and how much cycles it takes. We need to compare, sort and even print out the basic block by some orders, i.e. sort by cycles. For this purpose, we add block_info field to hist_entry. In order not to impact current interface, we creates a new function hists__add_entry_block. v6: --- Remove the 'ops' argument in hists__add_entry_block Signed-off-by: Jin Yao <yao.jin@linux.intel.com> Reviewed-by: Jiri Olsa <jolsa@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Andi Kleen <ak@linux.intel.com> Cc: Jin Yao <yao.jin@intel.com> Cc: Kan Liang <kan.liang@linux.intel.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: http://lkml.kernel.org/r/1561713784-30533-3-git-send-email-yao.jin@linux.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Show all the supported sort keys in the command line help output, so
that it's not needed to refer to the manpage.
Before:
% perf report -h
...
-s, --sort <key[,key2...]>
sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ... Please refer the man page for the complete list.
After:
% perf report -h
...
-s, --sort <key[,key2...]>
sort by key(s): overhead overhead_sys overhead_us overhead_guest_sys overhead_guest_us overhead_children sample period pid comm dso symbol parent cpu ...
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
LPU-Reference: 20190314225002.30108-5-andi@firstfloor.org
Link: https://lkml.kernel.org/n/tip-9r3uz2ch4izoi1uln3f889co@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Now 'perf report' can show whole time periods with 'perf script', but the user still has to find individual samples of interest manually. It would be expensive and complicated to search for the right samples in the whole perf file. Typically users only need to look at a small number of samples for useful analysis. Also the full scripts tend to show samples of all CPUs and all threads mixed up, which can be very confusing on larger systems. Add a new --samples option to save a small random number of samples per hist entry. Use a reservoir sample technique to select a representatve number of samples. Then allow browsing the samples using 'perf script' as part of the hist entry context menu. This automatically adds the right filters, so only the thread or cpu of the sample is displayed. Then we use less' search functionality to directly jump the to the time stamp of the selected sample. It uses different menus for assembler and source display. Assembler needs xed installed and source needs debuginfo. Currently it only supports as many samples as fit on the screen due to some limitations in the slang ui code. Signed-off-by: Andi Kleen <ak@linux.intel.com> Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com> Acked-by: Jiri Olsa <jolsa@kernel.org> Link: http://lkml.kernel.org/r/20190311174605.GA29294@tassilo.jf.intel.com Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Add a time sort key to perf report to display samples for different time
quantums separately. This allows easier analysis of workloads that
change over time, and also will allow looking at the context of samples.
% perf record ...
% perf report --sort time,overhead,symbol --time-quantum 1ms --stdio
...
0.67% 277061.87300 [.] _dl_start
0.50% 277061.87300 [.] f1
0.50% 277061.87300 [.] f2
0.33% 277061.87300 [.] main
0.29% 277061.87300 [.] _dl_lookup_symbol_x
0.29% 277061.87300 [.] dl_main
0.29% 277061.87300 [.] do_lookup_x
0.17% 277061.87300 [.] _dl_debug_initialize
0.17% 277061.87300 [.] _dl_init_paths
0.08% 277061.87300 [.] check_match
0.04% 277061.87300 [.] _dl_count_modids
1.33% 277061.87400 [.] f1
1.33% 277061.87400 [.] f2
1.33% 277061.87400 [.] main
1.17% 277061.87500 [.] main
1.08% 277061.87500 [.] f1
1.08% 277061.87500 [.] f2
1.00% 277061.87600 [.] main
0.83% 277061.87600 [.] f1
0.83% 277061.87600 [.] f2
1.00% 277061.87700 [.] main
Committer notes:
Rename 'time' argument to hist_time() to htime to overcome this in older
distros:
cc1: warnings being treated as errors
util/hist.c: In function 'hist_time':
util/hist.c:251: error: declaration of 'time' shadows a global declaration
/usr/include/time.h:186: error: shadowed declaration is here
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Acked-by: Jiri Olsa <jolsa@kernel.org>
Link: http://lkml.kernel.org/r/20190311144502.15423-4-andi@firstfloor.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>