2021-10-26 10:49:29

by Adrian Hunter

[permalink] [raw]
Subject: [PATCH V2 0/6] perf dlfilter: Add dlfilter-show-cycles

Hi

The issue with V1 was that the IPC information used to accumulate cycle
counts has a lower granularity because it is output only when the cycle
count correlates to the IP of the event i.e. only when the IPC is exactly
correct.

To enable more frequent updates to the cycle count, itrace option 'A' is
added, which specifies that IPC information can be approximate.

In addition there are some new miscellaneous patches.


Changes in V2:

perf dlfilter: Add dlfilter-show-cycles
Separate counts for branches, instructions or other events.

New patches:
perf auxtrace: Add missing Z option to ITRACE_HELP
perf auxtrace: Add itrace A option to approximate IPC
perf intel-pt: Support itrace A option to approximate IPC
perf auxtrace: Add itrace d+o option to direct debug log to stdout
perf intel-pt: Support itrace d+o option to direct debug log to stdout


Adrian Hunter (6):
perf auxtrace: Add missing Z option to ITRACE_HELP
perf auxtrace: Add itrace A option to approximate IPC
perf intel-pt: Support itrace A option to approximate IPC
perf dlfilter: Add dlfilter-show-cycles
perf auxtrace: Add itrace d+o option to direct debug log to stdout
perf intel-pt: Support itrace d+o option to direct debug log to stdout

tools/perf/Documentation/itrace.txt | 2 +
tools/perf/Documentation/perf-intel-pt.txt | 23 ++++
tools/perf/Makefile.perf | 2 +-
tools/perf/dlfilters/dlfilter-show-cycles.c | 144 +++++++++++++++++++++
tools/perf/util/auxtrace.c | 3 +
tools/perf/util/auxtrace.h | 6 +
.../perf/util/intel-pt-decoder/intel-pt-decoder.c | 1 +
.../perf/util/intel-pt-decoder/intel-pt-decoder.h | 1 +
tools/perf/util/intel-pt-decoder/intel-pt-log.c | 8 +-
tools/perf/util/intel-pt.c | 21 ++-
10 files changed, 200 insertions(+), 11 deletions(-)
create mode 100644 tools/perf/dlfilters/dlfilter-show-cycles.c


Regards
Adrian


2021-10-26 10:49:33

by Adrian Hunter

[permalink] [raw]
Subject: [PATCH V2 1/6] perf auxtrace: Add missing Z option to ITRACE_HELP

ITRACE_HELP is used by perf commands to display help text for the --itrace
option. Add missing Z option.

Signed-off-by: Adrian Hunter <[email protected]>
---
tools/perf/util/auxtrace.h | 1 +
1 file changed, 1 insertion(+)

diff --git a/tools/perf/util/auxtrace.h b/tools/perf/util/auxtrace.h
index 5f383908ca6e..20dc78d86d54 100644
--- a/tools/perf/util/auxtrace.h
+++ b/tools/perf/util/auxtrace.h
@@ -649,6 +649,7 @@ bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
" L[len]: synthesize last branch entries on existing event records\n" \
" sNUMBER: skip initial number of events\n" \
" q: quicker (less detailed) decoding\n" \
+" Z: prefer to ignore timestamps (so-called \"timeless\" decoding)\n" \
" PERIOD[ns|us|ms|i|t]: specify period to sample stream\n" \
" concatenate multiple options. Default is ibxwpe or cewp\n"

--
2.25.1

2021-10-26 10:51:32

by Adrian Hunter

[permalink] [raw]
Subject: [PATCH V2 2/6] perf auxtrace: Add itrace A option to approximate IPC

Add an option to specify that synthesized IPC can be approximate, rather
than completely accurate.

Signed-off-by: Adrian Hunter <[email protected]>
---
tools/perf/Documentation/itrace.txt | 1 +
tools/perf/util/auxtrace.c | 3 +++
tools/perf/util/auxtrace.h | 3 +++
3 files changed, 7 insertions(+)

diff --git a/tools/perf/Documentation/itrace.txt b/tools/perf/Documentation/itrace.txt
index 2d586fe5e4c5..141449e97bed 100644
--- a/tools/perf/Documentation/itrace.txt
+++ b/tools/perf/Documentation/itrace.txt
@@ -20,6 +20,7 @@
L synthesize last branch entries on existing event records
s skip initial number of events
q quicker (less detailed) decoding
+ A approximate IPC
Z prefer to ignore timestamps (so-called "timeless" decoding)

The default is all events i.e. the same as --itrace=ibxwpe,
diff --git a/tools/perf/util/auxtrace.c b/tools/perf/util/auxtrace.c
index 8d2865b9ade2..c679394b898d 100644
--- a/tools/perf/util/auxtrace.c
+++ b/tools/perf/util/auxtrace.c
@@ -1564,6 +1564,9 @@ int itrace_do_parse_synth_opts(struct itrace_synth_opts *synth_opts,
case 'q':
synth_opts->quick += 1;
break;
+ case 'A':
+ synth_opts->approx_ipc = true;
+ break;
case 'Z':
synth_opts->timeless_decoding = true;
break;
diff --git a/tools/perf/util/auxtrace.h b/tools/perf/util/auxtrace.h
index 20dc78d86d54..889f976ea1a0 100644
--- a/tools/perf/util/auxtrace.h
+++ b/tools/perf/util/auxtrace.h
@@ -84,6 +84,7 @@ enum itrace_period_type {
* @thread_stack: feed branches to the thread_stack
* @last_branch: add branch context to 'instruction' events
* @add_last_branch: add branch context to existing event records
+ * @approx_ipc: approximate IPC
* @flc: whether to synthesize first level cache events
* @llc: whether to synthesize last level cache events
* @tlb: whether to synthesize TLB events
@@ -127,6 +128,7 @@ struct itrace_synth_opts {
bool thread_stack;
bool last_branch;
bool add_last_branch;
+ bool approx_ipc;
bool flc;
bool llc;
bool tlb;
@@ -649,6 +651,7 @@ bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
" L[len]: synthesize last branch entries on existing event records\n" \
" sNUMBER: skip initial number of events\n" \
" q: quicker (less detailed) decoding\n" \
+" A: approximate IPC\n" \
" Z: prefer to ignore timestamps (so-called \"timeless\" decoding)\n" \
" PERIOD[ns|us|ms|i|t]: specify period to sample stream\n" \
" concatenate multiple options. Default is ibxwpe or cewp\n"
--
2.25.1

2021-10-26 10:53:45

by Adrian Hunter

[permalink] [raw]
Subject: [PATCH V2 3/6] perf intel-pt: Support itrace A option to approximate IPC

Normally, for cycle-acccurate mode, IPC values are an exact number of
instructions and cycles. Due to the granularity of timestamps, that happens
only when a CYC packet correlates to the event.

Support the itrace 'A' option, to use instead, the number of cycles
associated with the current timestamp. This provides IPC information for
every change of timestamp, but at the expense of accuracy.

Furthermore, it can be used in conjunction with dlfilter-show-cycles.so
to provide higher granularity cycle information.

Signed-off-by: Adrian Hunter <[email protected]>
---
tools/perf/Documentation/perf-intel-pt.txt | 5 +++++
.../util/intel-pt-decoder/intel-pt-decoder.c | 1 +
.../util/intel-pt-decoder/intel-pt-decoder.h | 1 +
tools/perf/util/intel-pt.c | 16 ++++++++++++----
4 files changed, 19 insertions(+), 4 deletions(-)

diff --git a/tools/perf/Documentation/perf-intel-pt.txt b/tools/perf/Documentation/perf-intel-pt.txt
index 184ba62420f0..31f1f373c463 100644
--- a/tools/perf/Documentation/perf-intel-pt.txt
+++ b/tools/perf/Documentation/perf-intel-pt.txt
@@ -157,6 +157,10 @@ of instructions and number of cycles since the last update, and thus represent
the average IPC since the last IPC for that event type. Note IPC for "branches"
events is calculated separately from IPC for "instructions" events.

+Even with the 'cyc' config term, it is possible to produce IPC information for
+every change of timestamp, but at the expense of accuracy. That is selected by
+specifying the itrace 'A' option.
+
Also note that the IPC instruction count may or may not include the current
instruction. If the cycle count is associated with an asynchronous branch
(e.g. page fault or interrupt), then the instruction count does not include the
@@ -873,6 +877,7 @@ The letters are:
L synthesize last branch entries on existing event records
s skip initial number of events
q quicker (less detailed) decoding
+ A approximate IPC
Z prefer to ignore timestamps (so-called "timeless" decoding)

"Instructions" events look like they were recorded by "perf record -e
diff --git a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c
index 5ab631702769..5f83937bf8f3 100644
--- a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c
+++ b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c
@@ -608,6 +608,7 @@ static inline void intel_pt_update_sample_time(struct intel_pt_decoder *decoder)
{
decoder->sample_timestamp = decoder->timestamp;
decoder->sample_insn_cnt = decoder->timestamp_insn_cnt;
+ decoder->state.cycles = decoder->tot_cyc_cnt;
}

static void intel_pt_reposition(struct intel_pt_decoder *decoder)
diff --git a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h
index 4b5e79fcf557..8fd68f7a0963 100644
--- a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h
+++ b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h
@@ -218,6 +218,7 @@ struct intel_pt_state {
uint64_t to_ip;
uint64_t tot_insn_cnt;
uint64_t tot_cyc_cnt;
+ uint64_t cycles;
uint64_t timestamp;
uint64_t est_timestamp;
uint64_t trace_nr;
diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c
index 6f852b305e92..57e49b23ad25 100644
--- a/tools/perf/util/intel-pt.c
+++ b/tools/perf/util/intel-pt.c
@@ -163,6 +163,7 @@ struct intel_pt_queue {
bool step_through_buffers;
bool use_buffer_pid_tid;
bool sync_switch;
+ bool sample_ipc;
pid_t pid, tid;
int cpu;
int switch_state;
@@ -1571,7 +1572,7 @@ static int intel_pt_synth_branch_sample(struct intel_pt_queue *ptq)
sample.branch_stack = (struct branch_stack *)&dummy_bs;
}

- if (ptq->state->flags & INTEL_PT_SAMPLE_IPC)
+ if (ptq->sample_ipc)
sample.cyc_cnt = ptq->ipc_cyc_cnt - ptq->last_br_cyc_cnt;
if (sample.cyc_cnt) {
sample.insn_cnt = ptq->ipc_insn_cnt - ptq->last_br_insn_cnt;
@@ -1622,7 +1623,7 @@ static int intel_pt_synth_instruction_sample(struct intel_pt_queue *ptq)
else
sample.period = ptq->state->tot_insn_cnt - ptq->last_insn_cnt;

- if (ptq->state->flags & INTEL_PT_SAMPLE_IPC)
+ if (ptq->sample_ipc)
sample.cyc_cnt = ptq->ipc_cyc_cnt - ptq->last_in_cyc_cnt;
if (sample.cyc_cnt) {
sample.insn_cnt = ptq->ipc_insn_cnt - ptq->last_in_insn_cnt;
@@ -2198,8 +2199,15 @@ static int intel_pt_sample(struct intel_pt_queue *ptq)

ptq->have_sample = false;

- ptq->ipc_insn_cnt = ptq->state->tot_insn_cnt;
- ptq->ipc_cyc_cnt = ptq->state->tot_cyc_cnt;
+ if (pt->synth_opts.approx_ipc) {
+ ptq->ipc_insn_cnt = ptq->state->tot_insn_cnt;
+ ptq->ipc_cyc_cnt = ptq->state->cycles;
+ ptq->sample_ipc = true;
+ } else {
+ ptq->ipc_insn_cnt = ptq->state->tot_insn_cnt;
+ ptq->ipc_cyc_cnt = ptq->state->tot_cyc_cnt;
+ ptq->sample_ipc = ptq->state->flags & INTEL_PT_SAMPLE_IPC;
+ }

/*
* Do PEBS first to allow for the possibility that the PEBS timestamp
--
2.25.1

2021-10-26 10:55:41

by Adrian Hunter

[permalink] [raw]
Subject: [PATCH V2 5/6] perf auxtrace: Add itrace d+o option to direct debug log to stdout

It can be useful to see debug output in between normal output.

Add 'o' to the flags of debug option 'd', so that '--itrace=d+o' can
specify output of the debug log to stdout.

Signed-off-by: Adrian Hunter <[email protected]>
---
tools/perf/Documentation/itrace.txt | 1 +
tools/perf/util/auxtrace.h | 2 ++
2 files changed, 3 insertions(+)

diff --git a/tools/perf/Documentation/itrace.txt b/tools/perf/Documentation/itrace.txt
index 141449e97bed..c52755481e2f 100644
--- a/tools/perf/Documentation/itrace.txt
+++ b/tools/perf/Documentation/itrace.txt
@@ -62,5 +62,6 @@
debug messages will or will not be logged. Each flag must be preceded
by either '+' or '-'. The flags are:
a all perf events
+ o output to stdout

If supported, the 'q' option may be repeated to increase the effect.
diff --git a/tools/perf/util/auxtrace.h b/tools/perf/util/auxtrace.h
index 889f976ea1a0..bbf0d78c6401 100644
--- a/tools/perf/util/auxtrace.h
+++ b/tools/perf/util/auxtrace.h
@@ -59,6 +59,7 @@ enum itrace_period_type {
#define AUXTRACE_ERR_FLG_DATA_LOST (1 << ('l' - 'a'))

#define AUXTRACE_LOG_FLG_ALL_PERF_EVTS (1 << ('a' - 'a'))
+#define AUXTRACE_LOG_FLG_USE_STDOUT (1 << ('o' - 'a'))

/**
* struct itrace_synth_opts - AUX area tracing synthesis options.
@@ -641,6 +642,7 @@ bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
" d[flags]: create a debug log\n" \
" each flag must be preceded by + or -\n" \
" log flags are: a (all perf events)\n" \
+" o (output to stdout)\n" \
" f: synthesize first level cache events\n" \
" m: synthesize last level cache events\n" \
" t: synthesize TLB events\n" \
--
2.25.1

2021-10-26 10:55:40

by Adrian Hunter

[permalink] [raw]
Subject: [PATCH V2 6/6] perf intel-pt: Support itrace d+o option to direct debug log to stdout

It can be useful to see debug output in between normal output.

Add support for AUXTRACE_LOG_FLG_USE_STDOUT to Intel PT.

Signed-off-by: Adrian Hunter <[email protected]>
---
tools/perf/Documentation/perf-intel-pt.txt | 1 +
tools/perf/util/intel-pt-decoder/intel-pt-log.c | 8 ++++----
tools/perf/util/intel-pt.c | 5 +++--
3 files changed, 8 insertions(+), 6 deletions(-)

diff --git a/tools/perf/Documentation/perf-intel-pt.txt b/tools/perf/Documentation/perf-intel-pt.txt
index 81dd27be3d09..b94dca105ebd 100644
--- a/tools/perf/Documentation/perf-intel-pt.txt
+++ b/tools/perf/Documentation/perf-intel-pt.txt
@@ -948,6 +948,7 @@ by flags which affect what debug messages will or will not be logged. Each flag
must be preceded by either '+' or '-'. The flags support by Intel PT are:
-a Suppress logging of perf events
+a Log all perf events
+ +o Output to stdout instead of "intel_pt.log"
By default, logged perf events are filtered by any specified time ranges, but
flag +a overrides that.

diff --git a/tools/perf/util/intel-pt-decoder/intel-pt-log.c b/tools/perf/util/intel-pt-decoder/intel-pt-log.c
index 09feb5b07d32..5f5dfc8753f3 100644
--- a/tools/perf/util/intel-pt-decoder/intel-pt-log.c
+++ b/tools/perf/util/intel-pt-decoder/intel-pt-log.c
@@ -82,10 +82,10 @@ static int intel_pt_log_open(void)
if (f)
return 0;

- if (!log_name[0])
- return -1;
-
- f = fopen(log_name, "w+");
+ if (log_name[0])
+ f = fopen(log_name, "w+");
+ else
+ f = stdout;
if (!f) {
intel_pt_enable_logging = false;
return -1;
diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c
index 57e49b23ad25..793bac850268 100644
--- a/tools/perf/util/intel-pt.c
+++ b/tools/perf/util/intel-pt.c
@@ -3659,8 +3659,6 @@ int intel_pt_process_auxtrace_info(union perf_event *event,
if (err)
goto err_free;

- intel_pt_log_set_name(INTEL_PT_PMU_NAME);
-
if (session->itrace_synth_opts->set) {
pt->synth_opts = *session->itrace_synth_opts;
} else {
@@ -3675,6 +3673,9 @@ int intel_pt_process_auxtrace_info(union perf_event *event,
pt->synth_opts.thread_stack = opts->thread_stack;
}

+ if (!(pt->synth_opts.log_plus_flags & AUXTRACE_LOG_FLG_USE_STDOUT))
+ intel_pt_log_set_name(INTEL_PT_PMU_NAME);
+
pt->session = session;
pt->machine = &session->machines.host; /* No kvm support */
pt->auxtrace_type = auxtrace_info->type;
--
2.25.1

2021-10-26 10:55:41

by Adrian Hunter

[permalink] [raw]
Subject: [PATCH V2 4/6] perf dlfilter: Add dlfilter-show-cycles

Add a new dlfilter to show cycles.

Cycle counts are accumulated per CPU (or per thread if CPU is not recorded)
from IPC information, and printed together with the change since the last
print, at the start of each line. Separate counts are kept for branches,
instructions or other events.

Note also, the itrace A option can be useful to provide higher granularity
cycle information.

Example:

$ perf record -e intel_pt/cyc/u uname
Linux
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.044 MB perf.data ]
$ perf script --itrace=A --call-trace --dlfilter dlfilter-show-cycles.so --deltatime | head
0 perf-exec 8509 [001] 0.000000000: psb offs: 0
0 perf-exec 8509 [001] 0.000000000: cbr: 42 freq: 4219 MHz (156%)
833 833 uname 8509 [001] 0.000047689: (/usr/lib/x86_64-linux-gnu/ld-2.31.so ) _start
833 uname 8509 [001] 0.000003261: (/usr/lib/x86_64-linux-gnu/ld-2.31.so ) _dl_start
2015 1182 uname 8509 [001] 0.000000282: (/usr/lib/x86_64-linux-gnu/ld-2.31.so ) _dl_start
2676 661 uname 8509 [001] 0.000002629: (/usr/lib/x86_64-linux-gnu/ld-2.31.so ) _dl_start
3612 936 uname 8509 [001] 0.000001232: (/usr/lib/x86_64-linux-gnu/ld-2.31.so ) _dl_start
4579 967 uname 8509 [001] 0.000002519: (/usr/lib/x86_64-linux-gnu/ld-2.31.so ) _dl_start
6145 1566 uname 8509 [001] 0.000001050: (/usr/lib/x86_64-linux-gnu/ld-2.31.so ) _dl_setup_hash
6239 94 uname 8509 [001] 0.000000023: (/usr/lib/x86_64-linux-gnu/ld-2.31.so ) _dl_sysdep_start

Signed-off-by: Adrian Hunter <[email protected]>
---
tools/perf/Documentation/perf-intel-pt.txt | 19 ++-
tools/perf/Makefile.perf | 2 +-
tools/perf/dlfilters/dlfilter-show-cycles.c | 144 ++++++++++++++++++++
3 files changed, 163 insertions(+), 2 deletions(-)
create mode 100644 tools/perf/dlfilters/dlfilter-show-cycles.c

diff --git a/tools/perf/Documentation/perf-intel-pt.txt b/tools/perf/Documentation/perf-intel-pt.txt
index 31f1f373c463..81dd27be3d09 100644
--- a/tools/perf/Documentation/perf-intel-pt.txt
+++ b/tools/perf/Documentation/perf-intel-pt.txt
@@ -159,7 +159,9 @@ events is calculated separately from IPC for "instructions" events.

Even with the 'cyc' config term, it is possible to produce IPC information for
every change of timestamp, but at the expense of accuracy. That is selected by
-specifying the itrace 'A' option.
+specifying the itrace 'A' option. It may also be useful to use the 'A' option
+in conjunction with dlfilter-show-cycles.so to provide higher granularity cycle
+information.

Also note that the IPC instruction count may or may not include the current
instruction. If the cycle count is associated with an asynchronous branch
@@ -1077,6 +1079,21 @@ The Z option is equivalent to having recorded a trace without TSC
decoding a trace of a virtual machine.


+dlfilter-show-cycles.so
+~~~~~~~~~~~~~~~~~~~~~~~
+
+Cycles can be displayed using dlfilter-show-cycles.so in which case the itrace A
+option can be useful to provide higher granularity cycle information:
+
+ perf script --itrace=A --call-trace --dlfilter dlfilter-show-cycles.so
+
+To see a list of dlfilters:
+
+ perf script -v --list-dlfilters
+
+See also linkperf:perf-dlfilters[1]
+
+
dump option
~~~~~~~~~~~

diff --git a/tools/perf/Makefile.perf b/tools/perf/Makefile.perf
index 7df13e74450c..e155570cb662 100644
--- a/tools/perf/Makefile.perf
+++ b/tools/perf/Makefile.perf
@@ -362,7 +362,7 @@ ifndef NO_JVMTI
PROGRAMS += $(OUTPUT)$(LIBJVMTI)
endif

-DLFILTERS := dlfilter-test-api-v0.so
+DLFILTERS := dlfilter-test-api-v0.so dlfilter-show-cycles.so
DLFILTERS := $(patsubst %,$(OUTPUT)dlfilters/%,$(DLFILTERS))

# what 'all' will build and 'install' will install, in perfexecdir
diff --git a/tools/perf/dlfilters/dlfilter-show-cycles.c b/tools/perf/dlfilters/dlfilter-show-cycles.c
new file mode 100644
index 000000000000..9eccc97bff82
--- /dev/null
+++ b/tools/perf/dlfilters/dlfilter-show-cycles.c
@@ -0,0 +1,144 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * dlfilter-show-cycles.c: Print the number of cycles at the start of each line
+ * Copyright (c) 2021, Intel Corporation.
+ */
+#include <perf/perf_dlfilter.h>
+#include <string.h>
+#include <stdio.h>
+
+#define MAX_CPU 4096
+
+enum {
+ INSTR_CYC,
+ BRNCH_CYC,
+ OTHER_CYC,
+ MAX_ENTRY
+};
+
+static __u64 cycles[MAX_CPU][MAX_ENTRY];
+static __u64 cycles_rpt[MAX_CPU][MAX_ENTRY];
+
+#define BITS 16
+#define TABLESZ (1 << BITS)
+#define TABLEMAX (TABLESZ / 2)
+#define MASK (TABLESZ - 1)
+
+static struct entry {
+ __u32 used;
+ __s32 tid;
+ __u64 cycles[MAX_ENTRY];
+ __u64 cycles_rpt[MAX_ENTRY];
+} table[TABLESZ];
+
+static int tid_cnt;
+
+static int event_entry(const char *event)
+{
+ if (!event)
+ return OTHER_CYC;
+ if (!strncmp(event, "instructions", 12))
+ return INSTR_CYC;
+ if (!strncmp(event, "branches", 8))
+ return BRNCH_CYC;
+ return OTHER_CYC;
+}
+
+static struct entry *find_entry(__s32 tid)
+{
+ __u32 pos = tid & MASK;
+ struct entry *e;
+
+ e = &table[pos];
+ while (e->used) {
+ if (e->tid == tid)
+ return e;
+ if (++pos == TABLESZ)
+ pos = 0;
+ e = &table[pos];
+ }
+
+ if (tid_cnt >= TABLEMAX) {
+ fprintf(stderr, "Too many threads\n");
+ return NULL;
+ }
+
+ tid_cnt += 1;
+ e->used = 1;
+ e->tid = tid;
+ return e;
+}
+
+static void add_entry(__s32 tid, int pos, __u64 cnt)
+{
+ struct entry *e = find_entry(tid);
+
+ if (e)
+ e->cycles[pos] += cnt;
+}
+
+int filter_event_early(void *data, const struct perf_dlfilter_sample *sample, void *ctx)
+{
+ __s32 cpu = sample->cpu;
+ __s32 tid = sample->tid;
+ int pos;
+
+ if (!sample->cyc_cnt)
+ return 0;
+
+ pos = event_entry(sample->event);
+
+ if (cpu >= 0 && cpu < MAX_CPU)
+ cycles[cpu][pos] += sample->cyc_cnt;
+ else if (tid != -1)
+ add_entry(tid, pos, sample->cyc_cnt);
+ return 0;
+}
+
+static void print_vals(__u64 cycles, __u64 delta)
+{
+ if (delta)
+ printf("%10llu %10llu ", cycles, delta);
+ else
+ printf("%10llu %10s ", cycles, "");
+}
+
+int filter_event(void *data, const struct perf_dlfilter_sample *sample, void *ctx)
+{
+ __s32 cpu = sample->cpu;
+ __s32 tid = sample->tid;
+ int pos;
+
+ pos = event_entry(sample->event);
+
+ if (cpu >= 0 && cpu < MAX_CPU) {
+ print_vals(cycles[cpu][pos], cycles[cpu][pos] - cycles_rpt[cpu][pos]);
+ cycles_rpt[cpu][pos] = cycles[cpu][pos];
+ return 0;
+ }
+
+ if (tid != -1) {
+ struct entry *e = find_entry(tid);
+
+ if (e) {
+ print_vals(e->cycles[pos], e->cycles[pos] - e->cycles_rpt[pos]);
+ e->cycles_rpt[pos] = e->cycles[pos];
+ return 0;
+ }
+ }
+
+ printf("%22s", "");
+ return 0;
+}
+
+const char *filter_description(const char **long_description)
+{
+ static char *long_desc = "Cycle counts are accumulated per CPU (or "
+ "per thread if CPU is not recorded) from IPC information, and "
+ "printed together with the change since the last print, at the "
+ "start of each line. Separate counts are kept for branches, "
+ "instructions or other events.";
+
+ *long_description = long_desc;
+ return "Print the number of cycles at the start of each line";
+}
--
2.25.1

2021-10-27 01:36:57

by Andi Kleen

[permalink] [raw]
Subject: Re: [PATCH V2 3/6] perf intel-pt: Support itrace A option to approximate IPC


On 10/26/2021 2:01 AM, Adrian Hunter wrote:
> Normally, for cycle-acccurate mode, IPC values are an exact number of
> instructions and cycles. Due to the granularity of timestamps, that happens
> only when a CYC packet correlates to the event.
>
> Support the itrace 'A' option, to use instead, the number of cycles
> associated with the current timestamp. This provides IPC information for
> every change of timestamp, but at the expense of accuracy.

Can you expand a bit what exactly the accuracy loss it?

Would be good to describe that in the manpage too.

2021-10-27 21:35:52

by Arnaldo Carvalho de Melo

[permalink] [raw]
Subject: Re: [PATCH V2 6/6] perf intel-pt: Support itrace d+o option to direct debug log to stdout

Em Tue, Oct 26, 2021 at 10:05:15AM -0700, Andi Kleen escreveu:
> Except for the documentation comments everything looks good to me. Thanks
> for implementing that.
>
>
> Reviewed-by: Andi Kleen <[email protected]>

Thanks, he can send the expansion on a follow up patch, reviewing and
applying.

- Arnaldo