* [PATCH v2 1/4] perf event: Add simd_flags field to perf_sample
2023-03-20 15:15 [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data James Clark
@ 2023-03-20 15:15 ` James Clark
2023-03-20 15:15 ` [PATCH v2 2/4] perf arm-spe: Refactor arm-spe to support operation packet type James Clark
` (3 subsequent siblings)
4 siblings, 0 replies; 8+ messages in thread
From: James Clark @ 2023-03-20 15:15 UTC (permalink / raw)
To: linux-perf-users, Anshuman.Khandual
Cc: German Gomez, James Clark, Peter Zijlstra, Ingo Molnar,
Arnaldo Carvalho de Melo, Mark Rutland, Alexander Shishkin,
Jiri Olsa, Namhyung Kim, Ian Rogers, Adrian Hunter, John Garry,
Will Deacon, Mike Leach, Leo Yan, linux-kernel, linux-arm-kernel
From: German Gomez <german.gomez@arm.com>
Add new field to the struct perf_sample to store flags related to SIMD
ops.
It will be used to store SIMD information from SVE and NEON when
profiling using ARM SPE.
Signed-off-by: German Gomez <german.gomez@arm.com>
Signed-off-by: James Clark <james.clark@arm.com>
---
tools/perf/util/sample.h | 13 +++++++++++++
1 file changed, 13 insertions(+)
diff --git a/tools/perf/util/sample.h b/tools/perf/util/sample.h
index 33b08e0ac746..c92ad0f51ecd 100644
--- a/tools/perf/util/sample.h
+++ b/tools/perf/util/sample.h
@@ -66,6 +66,18 @@ struct aux_sample {
void *data;
};
+struct simd_flags {
+ u64 arch:1, /* architecture (isa) */
+ pred:2; /* predication */
+};
+
+/* simd architecture flags */
+#define SIMD_OP_FLAGS_ARCH_SVE 0x01 /* ARM SVE */
+
+/* simd predicate flags */
+#define SIMD_OP_FLAGS_PRED_PARTIAL 0x01 /* partial predicate */
+#define SIMD_OP_FLAGS_PRED_EMPTY 0x02 /* empty predicate */
+
struct perf_sample {
u64 ip;
u32 pid, tid;
@@ -106,6 +118,7 @@ struct perf_sample {
struct stack_dump user_stack;
struct sample_read read;
struct aux_sample aux_sample;
+ struct simd_flags simd_flags;
};
/*
--
2.34.1
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [PATCH v2 2/4] perf arm-spe: Refactor arm-spe to support operation packet type
2023-03-20 15:15 [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data James Clark
2023-03-20 15:15 ` [PATCH v2 1/4] perf event: Add simd_flags field to perf_sample James Clark
@ 2023-03-20 15:15 ` James Clark
2023-03-20 15:15 ` [PATCH v2 3/4] perf arm-spe: Add SVE flags to the SPE samples James Clark
` (2 subsequent siblings)
4 siblings, 0 replies; 8+ messages in thread
From: James Clark @ 2023-03-20 15:15 UTC (permalink / raw)
To: linux-perf-users, Anshuman.Khandual
Cc: German Gomez, Leo Yan, James Clark, Peter Zijlstra, Ingo Molnar,
Arnaldo Carvalho de Melo, Mark Rutland, Alexander Shishkin,
Jiri Olsa, Namhyung Kim, Ian Rogers, Adrian Hunter, John Garry,
Will Deacon, Mike Leach, linux-kernel, linux-arm-kernel
From: German Gomez <german.gomez@arm.com>
Extend the decoder of Arm SPE records to support more fields from the
operation packet type.
Not all fields are being decoded by this commit. Only those needed to
support the use-case SVE load/store/other operations.
Suggested-by: Leo Yan <leo.yan@linaro.org>
Signed-off-by: German Gomez <german.gomez@arm.com>
Signed-off-by: James Clark <james.clark@arm.com>
---
.../util/arm-spe-decoder/arm-spe-decoder.c | 30 ++++++++++--
.../util/arm-spe-decoder/arm-spe-decoder.h | 47 +++++++++++++++----
tools/perf/util/arm-spe.c | 8 ++--
3 files changed, 67 insertions(+), 18 deletions(-)
diff --git a/tools/perf/util/arm-spe-decoder/arm-spe-decoder.c b/tools/perf/util/arm-spe-decoder/arm-spe-decoder.c
index 40dcedfd75cd..f3918f290df5 100644
--- a/tools/perf/util/arm-spe-decoder/arm-spe-decoder.c
+++ b/tools/perf/util/arm-spe-decoder/arm-spe-decoder.c
@@ -190,11 +190,27 @@ static int arm_spe_read_record(struct arm_spe_decoder *decoder)
decoder->record.context_id = payload;
break;
case ARM_SPE_OP_TYPE:
- if (idx == SPE_OP_PKT_HDR_CLASS_LD_ST_ATOMIC) {
- if (payload & 0x1)
- decoder->record.op = ARM_SPE_ST;
+ switch (idx) {
+ case SPE_OP_PKT_HDR_CLASS_LD_ST_ATOMIC:
+ decoder->record.op |= ARM_SPE_OP_LDST;
+ if (payload & SPE_OP_PKT_ST)
+ decoder->record.op |= ARM_SPE_OP_ST;
else
- decoder->record.op = ARM_SPE_LD;
+ decoder->record.op |= ARM_SPE_OP_LD;
+ if (SPE_OP_PKT_IS_LDST_SVE(payload))
+ decoder->record.op |= ARM_SPE_OP_SVE_LDST;
+ break;
+ case SPE_OP_PKT_HDR_CLASS_OTHER:
+ decoder->record.op |= ARM_SPE_OP_OTHER;
+ if (SPE_OP_PKT_IS_OTHER_SVE_OP(payload))
+ decoder->record.op |= ARM_SPE_OP_SVE_OTHER;
+ break;
+ case SPE_OP_PKT_HDR_CLASS_BR_ERET:
+ decoder->record.op |= ARM_SPE_OP_BRANCH_ERET;
+ break;
+ default:
+ pr_err("Get packet error!\n");
+ return -1;
}
break;
case ARM_SPE_EVENTS:
@@ -222,6 +238,12 @@ static int arm_spe_read_record(struct arm_spe_decoder *decoder)
if (payload & BIT(EV_MISPRED))
decoder->record.type |= ARM_SPE_BRANCH_MISS;
+ if (payload & BIT(EV_PARTIAL_PREDICATE))
+ decoder->record.type |= ARM_SPE_SVE_PARTIAL_PRED;
+
+ if (payload & BIT(EV_EMPTY_PREDICATE))
+ decoder->record.type |= ARM_SPE_SVE_EMPTY_PRED;
+
break;
case ARM_SPE_DATA_SOURCE:
decoder->record.source = payload;
diff --git a/tools/perf/util/arm-spe-decoder/arm-spe-decoder.h b/tools/perf/util/arm-spe-decoder/arm-spe-decoder.h
index 46a61df1145b..1443c28545a9 100644
--- a/tools/perf/util/arm-spe-decoder/arm-spe-decoder.h
+++ b/tools/perf/util/arm-spe-decoder/arm-spe-decoder.h
@@ -14,19 +14,46 @@
#include "arm-spe-pkt-decoder.h"
enum arm_spe_sample_type {
- ARM_SPE_L1D_ACCESS = 1 << 0,
- ARM_SPE_L1D_MISS = 1 << 1,
- ARM_SPE_LLC_ACCESS = 1 << 2,
- ARM_SPE_LLC_MISS = 1 << 3,
- ARM_SPE_TLB_ACCESS = 1 << 4,
- ARM_SPE_TLB_MISS = 1 << 5,
- ARM_SPE_BRANCH_MISS = 1 << 6,
- ARM_SPE_REMOTE_ACCESS = 1 << 7,
+ ARM_SPE_L1D_ACCESS = 1 << 0,
+ ARM_SPE_L1D_MISS = 1 << 1,
+ ARM_SPE_LLC_ACCESS = 1 << 2,
+ ARM_SPE_LLC_MISS = 1 << 3,
+ ARM_SPE_TLB_ACCESS = 1 << 4,
+ ARM_SPE_TLB_MISS = 1 << 5,
+ ARM_SPE_BRANCH_MISS = 1 << 6,
+ ARM_SPE_REMOTE_ACCESS = 1 << 7,
+ ARM_SPE_SVE_PARTIAL_PRED = 1 << 8,
+ ARM_SPE_SVE_EMPTY_PRED = 1 << 9,
};
enum arm_spe_op_type {
- ARM_SPE_LD = 1 << 0,
- ARM_SPE_ST = 1 << 1,
+ /* First level operation type */
+ ARM_SPE_OP_OTHER = 1 << 0,
+ ARM_SPE_OP_LDST = 1 << 1,
+ ARM_SPE_OP_BRANCH_ERET = 1 << 2,
+
+ /* Second level operation type for OTHER */
+ ARM_SPE_OP_SVE_OTHER = 1 << 16,
+ ARM_SPE_OP_SVE_FP = 1 << 17,
+ ARM_SPE_OP_SVE_PRED_OTHER = 1 << 18,
+
+ /* Second level operation type for LDST */
+ ARM_SPE_OP_LD = 1 << 16,
+ ARM_SPE_OP_ST = 1 << 17,
+ ARM_SPE_OP_ATOMIC = 1 << 18,
+ ARM_SPE_OP_EXCL = 1 << 19,
+ ARM_SPE_OP_AR = 1 << 20,
+ ARM_SPE_OP_SIMD_FP = 1 << 21,
+ ARM_SPE_OP_GP_REG = 1 << 22,
+ ARM_SPE_OP_UNSPEC_REG = 1 << 23,
+ ARM_SPE_OP_NV_SYSREG = 1 << 24,
+ ARM_SPE_OP_SVE_LDST = 1 << 25,
+ ARM_SPE_OP_SVE_PRED_LDST = 1 << 26,
+ ARM_SPE_OP_SVE_SG = 1 << 27,
+
+ /* Second level operation type for BRANCH_ERET */
+ ARM_SPE_OP_BR_COND = 1 << 16,
+ ARM_SPE_OP_BR_INDIRECT = 1 << 17,
};
enum arm_spe_neoverse_data_source {
diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index 906476a839e1..bfae4731a47a 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -411,7 +411,7 @@ static void arm_spe__synth_data_source_neoverse(const struct arm_spe_record *rec
* We have no data on the hit level or data source for stores in the
* Neoverse SPE records.
*/
- if (record->op & ARM_SPE_ST) {
+ if (record->op & ARM_SPE_OP_ST) {
data_src->mem_lvl = PERF_MEM_LVL_NA;
data_src->mem_lvl_num = PERF_MEM_LVLNUM_NA;
data_src->mem_snoop = PERF_MEM_SNOOP_NA;
@@ -497,12 +497,12 @@ static void arm_spe__synth_data_source_generic(const struct arm_spe_record *reco
static u64 arm_spe__synth_data_source(const struct arm_spe_record *record, u64 midr)
{
- union perf_mem_data_src data_src = { 0 };
+ union perf_mem_data_src data_src = { .mem_op = PERF_MEM_OP_NA };
bool is_neoverse = is_midr_in_range_list(midr, neoverse_spe);
- if (record->op == ARM_SPE_LD)
+ if (record->op & ARM_SPE_OP_LD)
data_src.mem_op = PERF_MEM_OP_LOAD;
- else if (record->op == ARM_SPE_ST)
+ else if (record->op & ARM_SPE_OP_ST)
data_src.mem_op = PERF_MEM_OP_STORE;
else
return 0;
--
2.34.1
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [PATCH v2 3/4] perf arm-spe: Add SVE flags to the SPE samples
2023-03-20 15:15 [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data James Clark
2023-03-20 15:15 ` [PATCH v2 1/4] perf event: Add simd_flags field to perf_sample James Clark
2023-03-20 15:15 ` [PATCH v2 2/4] perf arm-spe: Refactor arm-spe to support operation packet type James Clark
@ 2023-03-20 15:15 ` James Clark
2023-03-20 15:15 ` [PATCH v2 4/4] perf report: Add 'simd' sort field James Clark
2023-03-20 17:50 ` [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data Ian Rogers
4 siblings, 0 replies; 8+ messages in thread
From: James Clark @ 2023-03-20 15:15 UTC (permalink / raw)
To: linux-perf-users, Anshuman.Khandual
Cc: German Gomez, James Clark, Peter Zijlstra, Ingo Molnar,
Arnaldo Carvalho de Melo, Mark Rutland, Alexander Shishkin,
Jiri Olsa, Namhyung Kim, Ian Rogers, Adrian Hunter, John Garry,
Will Deacon, Mike Leach, Leo Yan, linux-kernel, linux-arm-kernel
From: German Gomez <german.gomez@arm.com>
Add flags from the Scalable Vector Extension (SVE) to the SPE samples
which are available from Armv8.3 (FEAT_SPEv1p1).
These will be displayed in a new SIMD sort field in a later commit.
Signed-off-by: German Gomez <german.gomez@arm.com>
Signed-off-by: James Clark <james.clark@arm.com>
---
tools/perf/util/arm-spe.c | 20 ++++++++++++++++++++
1 file changed, 20 insertions(+)
diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index bfae4731a47a..7b36ba6b4079 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -273,6 +273,25 @@ static int arm_spe_set_tid(struct arm_spe_queue *speq, pid_t tid)
return 0;
}
+static struct simd_flags arm_spe__synth_simd_flags(const struct arm_spe_record *record)
+{
+ struct simd_flags simd_flags = {};
+
+ if ((record->op & ARM_SPE_OP_LDST) && (record->op & ARM_SPE_OP_SVE_LDST))
+ simd_flags.arch |= SIMD_OP_FLAGS_ARCH_SVE;
+
+ if ((record->op & ARM_SPE_OP_OTHER) && (record->op & ARM_SPE_OP_SVE_OTHER))
+ simd_flags.arch |= SIMD_OP_FLAGS_ARCH_SVE;
+
+ if (record->type & ARM_SPE_SVE_PARTIAL_PRED)
+ simd_flags.pred |= SIMD_OP_FLAGS_PRED_PARTIAL;
+
+ if (record->type & ARM_SPE_SVE_EMPTY_PRED)
+ simd_flags.pred |= SIMD_OP_FLAGS_PRED_EMPTY;
+
+ return simd_flags;
+}
+
static void arm_spe_prep_sample(struct arm_spe *spe,
struct arm_spe_queue *speq,
union perf_event *event,
@@ -289,6 +308,7 @@ static void arm_spe_prep_sample(struct arm_spe *spe,
sample->tid = speq->tid;
sample->period = 1;
sample->cpu = speq->cpu;
+ sample->simd_flags = arm_spe__synth_simd_flags(record);
event->sample.header.type = PERF_RECORD_SAMPLE;
event->sample.header.misc = sample->cpumode;
--
2.34.1
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply related [flat|nested] 8+ messages in thread
* [PATCH v2 4/4] perf report: Add 'simd' sort field
2023-03-20 15:15 [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data James Clark
` (2 preceding siblings ...)
2023-03-20 15:15 ` [PATCH v2 3/4] perf arm-spe: Add SVE flags to the SPE samples James Clark
@ 2023-03-20 15:15 ` James Clark
2023-03-20 17:50 ` Ian Rogers
2023-03-20 17:50 ` [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data Ian Rogers
4 siblings, 1 reply; 8+ messages in thread
From: James Clark @ 2023-03-20 15:15 UTC (permalink / raw)
To: linux-perf-users, Anshuman.Khandual
Cc: German Gomez, James Clark, Peter Zijlstra, Ingo Molnar,
Arnaldo Carvalho de Melo, Mark Rutland, Alexander Shishkin,
Jiri Olsa, Namhyung Kim, Ian Rogers, Adrian Hunter, John Garry,
Will Deacon, Mike Leach, Leo Yan, linux-kernel, linux-arm-kernel
From: German Gomez <german.gomez@arm.com>
Add 'simd' sort field to visualize SIMD ops in perf-report.
Rows are labeled with the SIMD isa, and the type of predicate (if any):
- [p] partial predicate
- [e] empty predicate (no elements in the vector being used)
Example with Arm SPE and SVE (Scalable Vector Extension):
#include <arm_sve.h>
double src[1025], dst[1025];
int main(void) {
svfloat64_t vc = svdup_f64(1);
for(;;)
for(int i = 0; i < 1025; i += svcntd())
{
svbool_t pg = svwhilelt_b64(i, 1025);
svfloat64_t vsrc = svld1(pg, &src[i]);
svfloat64_t vdst = svadd_x(pg, vsrc, vc);
svst1(pg, &dst[i], vdst);
}
return 0;
}
... compiled using "gcc-11 -march=armv8-a+sve -O3"
Profiling on a platform that implements FEAT_SVE and FEAT_SPEv1p1:
$ perf record -e arm_spe_0// -- ./a.out
$ perf report --itrace=i1i -s overhead,pid,simd,sym
Overhead Pid:Command Simd Symbol
........ ................ ....... ......................
53.76% 10758:program [.] main
46.14% 10758:program [.] SVE [.] main
0.09% 10758:program [p] SVE [.] main
The report shows 0.09% of the sampled SVE operations use partial
predicates due to src and dst arrays not being multiples of the vector
register lengths.
Signed-off-by: German Gomez <german.gomez@arm.com>
Signed-off-by: James Clark <james.clark@arm.com>
---
tools/perf/Documentation/perf-report.txt | 1 +
tools/perf/util/hist.c | 1 +
tools/perf/util/hist.h | 1 +
tools/perf/util/sort.c | 47 ++++++++++++++++++++++++
tools/perf/util/sort.h | 2 +
5 files changed, 52 insertions(+)
diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
index c242e8da6b1a..cfd502f7e6da 100644
--- a/tools/perf/Documentation/perf-report.txt
+++ b/tools/perf/Documentation/perf-report.txt
@@ -117,6 +117,7 @@ OPTIONS
- addr: (Full) virtual address of the sampled instruction
- retire_lat: On X86, this reports pipeline stall of this instruction compared
to the previous instruction in cycles. And currently supported only on X86
+ - simd: Flags describing a SIMD operation. "e" for empty Arm SVE predicate. "p" for partial Arm SVE predicate
By default, comm, dso and symbol keys are used.
(i.e. --sort comm,dso,symbol)
diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c
index 3670136a0074..0c11f50abfec 100644
--- a/tools/perf/util/hist.c
+++ b/tools/perf/util/hist.c
@@ -745,6 +745,7 @@ __hists__add_entry(struct hists *hists,
.weight = sample->weight,
.ins_lat = sample->ins_lat,
.p_stage_cyc = sample->p_stage_cyc,
+ .simd_flags = sample->simd_flags,
}, *he = hists__findnew_entry(hists, &entry, al, sample_self);
if (!hists->has_callchains && he && he->callchain_size != 0)
diff --git a/tools/perf/util/hist.h b/tools/perf/util/hist.h
index 86a677954279..afc9f1c7f4dc 100644
--- a/tools/perf/util/hist.h
+++ b/tools/perf/util/hist.h
@@ -81,6 +81,7 @@ enum hist_column {
HISTC_ADDR_FROM,
HISTC_ADDR_TO,
HISTC_ADDR,
+ HISTC_SIMD,
HISTC_NR_COLS, /* Last entry */
};
diff --git a/tools/perf/util/sort.c b/tools/perf/util/sort.c
index 093a0c8b2e3d..e11e68ecf0a2 100644
--- a/tools/perf/util/sort.c
+++ b/tools/perf/util/sort.c
@@ -139,6 +139,52 @@ struct sort_entry sort_thread = {
.se_width_idx = HISTC_THREAD,
};
+/* --sort simd */
+
+static int64_t
+sort__simd_cmp(struct hist_entry *left, struct hist_entry *right)
+{
+ if (left->simd_flags.arch != right->simd_flags.arch)
+ return (int64_t) left->simd_flags.arch - right->simd_flags.arch;
+
+ return (int64_t) left->simd_flags.pred - right->simd_flags.pred;
+}
+
+static const char *hist_entry__get_simd_name(struct simd_flags *simd_flags)
+{
+ u64 arch = simd_flags->arch;
+
+ if (arch & SIMD_OP_FLAGS_ARCH_SVE)
+ return "SVE";
+ else
+ return "n/a";
+}
+
+static int hist_entry__simd_snprintf(struct hist_entry *he, char *bf,
+ size_t size, unsigned int width __maybe_unused)
+{
+ const char *name;
+
+ if (!he->simd_flags.arch)
+ return repsep_snprintf(bf, size, "");
+
+ name = hist_entry__get_simd_name(&he->simd_flags);
+
+ if (he->simd_flags.pred & SIMD_OP_FLAGS_PRED_EMPTY)
+ return repsep_snprintf(bf, size, "[e] %s", name);
+ else if (he->simd_flags.pred & SIMD_OP_FLAGS_PRED_PARTIAL)
+ return repsep_snprintf(bf, size, "[p] %s", name);
+
+ return repsep_snprintf(bf, size, "[.] %s", name);
+}
+
+struct sort_entry sort_simd = {
+ .se_header = "Simd ",
+ .se_cmp = sort__simd_cmp,
+ .se_snprintf = hist_entry__simd_snprintf,
+ .se_width_idx = HISTC_SIMD,
+};
+
/* --sort comm */
/*
@@ -2142,6 +2188,7 @@ static struct sort_dimension common_sort_dimensions[] = {
DIM(SORT_ADDR, "addr", sort_addr),
DIM(SORT_LOCAL_RETIRE_LAT, "local_retire_lat", sort_local_p_stage_cyc),
DIM(SORT_GLOBAL_RETIRE_LAT, "retire_lat", sort_global_p_stage_cyc),
+ DIM(SORT_SIMD, "simd", sort_simd)
};
#undef DIM
diff --git a/tools/perf/util/sort.h b/tools/perf/util/sort.h
index 22f437c3476f..ecfb7f1359d5 100644
--- a/tools/perf/util/sort.h
+++ b/tools/perf/util/sort.h
@@ -111,6 +111,7 @@ struct hist_entry {
u64 p_stage_cyc;
u8 cpumode;
u8 depth;
+ struct simd_flags simd_flags;
/* We are added by hists__add_dummy_entry. */
bool dummy;
@@ -241,6 +242,7 @@ enum sort_type {
SORT_ADDR,
SORT_LOCAL_RETIRE_LAT,
SORT_GLOBAL_RETIRE_LAT,
+ SORT_SIMD,
/* branch stack specific sort keys */
__SORT_BRANCH_STACK,
--
2.34.1
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply related [flat|nested] 8+ messages in thread
* Re: [PATCH v2 4/4] perf report: Add 'simd' sort field
2023-03-20 15:15 ` [PATCH v2 4/4] perf report: Add 'simd' sort field James Clark
@ 2023-03-20 17:50 ` Ian Rogers
0 siblings, 0 replies; 8+ messages in thread
From: Ian Rogers @ 2023-03-20 17:50 UTC (permalink / raw)
To: James Clark
Cc: linux-perf-users, Anshuman.Khandual, German Gomez,
Peter Zijlstra, Ingo Molnar, Arnaldo Carvalho de Melo,
Mark Rutland, Alexander Shishkin, Jiri Olsa, Namhyung Kim,
Adrian Hunter, John Garry, Will Deacon, Mike Leach, Leo Yan,
linux-kernel, linux-arm-kernel
On Mon, Mar 20, 2023 at 8:15 AM James Clark <james.clark@arm.com> wrote:
>
> From: German Gomez <german.gomez@arm.com>
>
> Add 'simd' sort field to visualize SIMD ops in perf-report.
>
> Rows are labeled with the SIMD isa, and the type of predicate (if any):
>
> - [p] partial predicate
> - [e] empty predicate (no elements in the vector being used)
>
> Example with Arm SPE and SVE (Scalable Vector Extension):
>
> #include <arm_sve.h>
>
> double src[1025], dst[1025];
>
> int main(void) {
> svfloat64_t vc = svdup_f64(1);
> for(;;)
> for(int i = 0; i < 1025; i += svcntd())
> {
> svbool_t pg = svwhilelt_b64(i, 1025);
> svfloat64_t vsrc = svld1(pg, &src[i]);
> svfloat64_t vdst = svadd_x(pg, vsrc, vc);
> svst1(pg, &dst[i], vdst);
> }
> return 0;
> }
>
> ... compiled using "gcc-11 -march=armv8-a+sve -O3"
>
> Profiling on a platform that implements FEAT_SVE and FEAT_SPEv1p1:
>
> $ perf record -e arm_spe_0// -- ./a.out
> $ perf report --itrace=i1i -s overhead,pid,simd,sym
>
> Overhead Pid:Command Simd Symbol
> ........ ................ ....... ......................
>
> 53.76% 10758:program [.] main
> 46.14% 10758:program [.] SVE [.] main
> 0.09% 10758:program [p] SVE [.] main
>
> The report shows 0.09% of the sampled SVE operations use partial
> predicates due to src and dst arrays not being multiples of the vector
> register lengths.
>
> Signed-off-by: German Gomez <german.gomez@arm.com>
> Signed-off-by: James Clark <james.clark@arm.com>
> ---
> tools/perf/Documentation/perf-report.txt | 1 +
> tools/perf/util/hist.c | 1 +
> tools/perf/util/hist.h | 1 +
> tools/perf/util/sort.c | 47 ++++++++++++++++++++++++
> tools/perf/util/sort.h | 2 +
> 5 files changed, 52 insertions(+)
>
> diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
> index c242e8da6b1a..cfd502f7e6da 100644
> --- a/tools/perf/Documentation/perf-report.txt
> +++ b/tools/perf/Documentation/perf-report.txt
> @@ -117,6 +117,7 @@ OPTIONS
> - addr: (Full) virtual address of the sampled instruction
> - retire_lat: On X86, this reports pipeline stall of this instruction compared
> to the previous instruction in cycles. And currently supported only on X86
> + - simd: Flags describing a SIMD operation. "e" for empty Arm SVE predicate. "p" for partial Arm SVE predicate
nit: the line wrap looks off here.
Thanks,
Ian
>
> By default, comm, dso and symbol keys are used.
> (i.e. --sort comm,dso,symbol)
> diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c
> index 3670136a0074..0c11f50abfec 100644
> --- a/tools/perf/util/hist.c
> +++ b/tools/perf/util/hist.c
> @@ -745,6 +745,7 @@ __hists__add_entry(struct hists *hists,
> .weight = sample->weight,
> .ins_lat = sample->ins_lat,
> .p_stage_cyc = sample->p_stage_cyc,
> + .simd_flags = sample->simd_flags,
> }, *he = hists__findnew_entry(hists, &entry, al, sample_self);
>
> if (!hists->has_callchains && he && he->callchain_size != 0)
> diff --git a/tools/perf/util/hist.h b/tools/perf/util/hist.h
> index 86a677954279..afc9f1c7f4dc 100644
> --- a/tools/perf/util/hist.h
> +++ b/tools/perf/util/hist.h
> @@ -81,6 +81,7 @@ enum hist_column {
> HISTC_ADDR_FROM,
> HISTC_ADDR_TO,
> HISTC_ADDR,
> + HISTC_SIMD,
> HISTC_NR_COLS, /* Last entry */
> };
>
> diff --git a/tools/perf/util/sort.c b/tools/perf/util/sort.c
> index 093a0c8b2e3d..e11e68ecf0a2 100644
> --- a/tools/perf/util/sort.c
> +++ b/tools/perf/util/sort.c
> @@ -139,6 +139,52 @@ struct sort_entry sort_thread = {
> .se_width_idx = HISTC_THREAD,
> };
>
> +/* --sort simd */
> +
> +static int64_t
> +sort__simd_cmp(struct hist_entry *left, struct hist_entry *right)
> +{
> + if (left->simd_flags.arch != right->simd_flags.arch)
> + return (int64_t) left->simd_flags.arch - right->simd_flags.arch;
> +
> + return (int64_t) left->simd_flags.pred - right->simd_flags.pred;
> +}
> +
> +static const char *hist_entry__get_simd_name(struct simd_flags *simd_flags)
> +{
> + u64 arch = simd_flags->arch;
> +
> + if (arch & SIMD_OP_FLAGS_ARCH_SVE)
> + return "SVE";
> + else
> + return "n/a";
> +}
> +
> +static int hist_entry__simd_snprintf(struct hist_entry *he, char *bf,
> + size_t size, unsigned int width __maybe_unused)
> +{
> + const char *name;
> +
> + if (!he->simd_flags.arch)
> + return repsep_snprintf(bf, size, "");
> +
> + name = hist_entry__get_simd_name(&he->simd_flags);
> +
> + if (he->simd_flags.pred & SIMD_OP_FLAGS_PRED_EMPTY)
> + return repsep_snprintf(bf, size, "[e] %s", name);
> + else if (he->simd_flags.pred & SIMD_OP_FLAGS_PRED_PARTIAL)
> + return repsep_snprintf(bf, size, "[p] %s", name);
> +
> + return repsep_snprintf(bf, size, "[.] %s", name);
> +}
> +
> +struct sort_entry sort_simd = {
> + .se_header = "Simd ",
> + .se_cmp = sort__simd_cmp,
> + .se_snprintf = hist_entry__simd_snprintf,
> + .se_width_idx = HISTC_SIMD,
> +};
> +
> /* --sort comm */
>
> /*
> @@ -2142,6 +2188,7 @@ static struct sort_dimension common_sort_dimensions[] = {
> DIM(SORT_ADDR, "addr", sort_addr),
> DIM(SORT_LOCAL_RETIRE_LAT, "local_retire_lat", sort_local_p_stage_cyc),
> DIM(SORT_GLOBAL_RETIRE_LAT, "retire_lat", sort_global_p_stage_cyc),
> + DIM(SORT_SIMD, "simd", sort_simd)
> };
>
> #undef DIM
> diff --git a/tools/perf/util/sort.h b/tools/perf/util/sort.h
> index 22f437c3476f..ecfb7f1359d5 100644
> --- a/tools/perf/util/sort.h
> +++ b/tools/perf/util/sort.h
> @@ -111,6 +111,7 @@ struct hist_entry {
> u64 p_stage_cyc;
> u8 cpumode;
> u8 depth;
> + struct simd_flags simd_flags;
>
> /* We are added by hists__add_dummy_entry. */
> bool dummy;
> @@ -241,6 +242,7 @@ enum sort_type {
> SORT_ADDR,
> SORT_LOCAL_RETIRE_LAT,
> SORT_GLOBAL_RETIRE_LAT,
> + SORT_SIMD,
>
> /* branch stack specific sort keys */
> __SORT_BRANCH_STACK,
> --
> 2.34.1
>
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data
2023-03-20 15:15 [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data James Clark
` (3 preceding siblings ...)
2023-03-20 15:15 ` [PATCH v2 4/4] perf report: Add 'simd' sort field James Clark
@ 2023-03-20 17:50 ` Ian Rogers
2023-03-20 22:29 ` Arnaldo Carvalho de Melo
4 siblings, 1 reply; 8+ messages in thread
From: Ian Rogers @ 2023-03-20 17:50 UTC (permalink / raw)
To: James Clark
Cc: linux-perf-users, Anshuman.Khandual, Peter Zijlstra, Ingo Molnar,
Arnaldo Carvalho de Melo, Mark Rutland, Alexander Shishkin,
Jiri Olsa, Namhyung Kim, Adrian Hunter, John Garry, Will Deacon,
Mike Leach, Leo Yan, linux-kernel, linux-arm-kernel
On Mon, Mar 20, 2023 at 8:15 AM James Clark <james.clark@arm.com> wrote:
>
> Changes since v1:
> * Rebase onto perf/core because it no longer applied cleanly
>
> ------------
>
> Hi,
>
> I'm submitting this on behalf of German who moved on to work on other
> things in Arm before he could finish it off.
>
> The predicate information is available on SPE samples from
> Armv8.3 (FEAT_SPEv1p1), this could be useful info for profiling SVE
> code as partial and empty predicates indicate that the full vector
> width isn't being used. There is a good example in the last commit
> message.
>
> Though currently, there isn't a suitable field to store the info
> on Perf samples, so this change also adds a new SIMD field.
> This field could be used by other architectures, but currently there
> is only one bit reserved to identify SVE. It's only added to
> struct perf_sample on the userspace side, and isn't part of the kernel
> ABI, so it doesn't survive a perf inject. Although this is the
> same behavior for some other fields like branch flags, so I don't
> think it should be an issue to do something similar here. Perhaps in
> the future we could make sure everything that is synthesised from
> auxtrace data also makes it back into the new Perf inject file without
> being lost.
Series:
Acked-by: Ian Rogers <irogers@google.com>
Thanks,
Ian
> German Gomez (4):
> perf event: Add simd_flags field to perf_sample
> perf arm-spe: Refactor arm-spe to support operation packet type
> perf arm-spe: Add SVE flags to the SPE samples
> perf report: Add 'simd' sort field
>
> tools/perf/Documentation/perf-report.txt | 1 +
> .../util/arm-spe-decoder/arm-spe-decoder.c | 30 ++++++++++--
> .../util/arm-spe-decoder/arm-spe-decoder.h | 47 +++++++++++++++----
> tools/perf/util/arm-spe.c | 28 +++++++++--
> tools/perf/util/hist.c | 1 +
> tools/perf/util/hist.h | 1 +
> tools/perf/util/sample.h | 13 +++++
> tools/perf/util/sort.c | 47 +++++++++++++++++++
> tools/perf/util/sort.h | 2 +
> 9 files changed, 152 insertions(+), 18 deletions(-)
>
>
> base-commit: 96d541699e5c50b1bc2d50c83cd7145994d5f071
> --
> 2.34.1
>
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data
2023-03-20 17:50 ` [PATCH v2 0/4] Enable display of partial and empty SVE predicates from Arm SPE data Ian Rogers
@ 2023-03-20 22:29 ` Arnaldo Carvalho de Melo
0 siblings, 0 replies; 8+ messages in thread
From: Arnaldo Carvalho de Melo @ 2023-03-20 22:29 UTC (permalink / raw)
To: Ian Rogers
Cc: James Clark, linux-perf-users, Anshuman.Khandual, Peter Zijlstra,
Ingo Molnar, Mark Rutland, Alexander Shishkin, Jiri Olsa,
Namhyung Kim, Adrian Hunter, John Garry, Will Deacon, Mike Leach,
Leo Yan, linux-kernel, linux-arm-kernel
Em Mon, Mar 20, 2023 at 10:50:59AM -0700, Ian Rogers escreveu:
> On Mon, Mar 20, 2023 at 8:15 AM James Clark <james.clark@arm.com> wrote:
> >
> > Changes since v1:
> > * Rebase onto perf/core because it no longer applied cleanly
> >
> > ------------
> >
> > Hi,
> >
> > I'm submitting this on behalf of German who moved on to work on other
> > things in Arm before he could finish it off.
> >
> > The predicate information is available on SPE samples from
> > Armv8.3 (FEAT_SPEv1p1), this could be useful info for profiling SVE
> > code as partial and empty predicates indicate that the full vector
> > width isn't being used. There is a good example in the last commit
> > message.
> >
> > Though currently, there isn't a suitable field to store the info
> > on Perf samples, so this change also adds a new SIMD field.
> > This field could be used by other architectures, but currently there
> > is only one bit reserved to identify SVE. It's only added to
> > struct perf_sample on the userspace side, and isn't part of the kernel
> > ABI, so it doesn't survive a perf inject. Although this is the
> > same behavior for some other fields like branch flags, so I don't
> > think it should be an issue to do something similar here. Perhaps in
> > the future we could make sure everything that is synthesised from
> > auxtrace data also makes it back into the new Perf inject file without
> > being lost.
>
> Series:
> Acked-by: Ian Rogers <irogers@google.com>
Thanks, applied.
- Arnaldo
> Thanks,
> Ian
>
> > German Gomez (4):
> > perf event: Add simd_flags field to perf_sample
> > perf arm-spe: Refactor arm-spe to support operation packet type
> > perf arm-spe: Add SVE flags to the SPE samples
> > perf report: Add 'simd' sort field
> >
> > tools/perf/Documentation/perf-report.txt | 1 +
> > .../util/arm-spe-decoder/arm-spe-decoder.c | 30 ++++++++++--
> > .../util/arm-spe-decoder/arm-spe-decoder.h | 47 +++++++++++++++----
> > tools/perf/util/arm-spe.c | 28 +++++++++--
> > tools/perf/util/hist.c | 1 +
> > tools/perf/util/hist.h | 1 +
> > tools/perf/util/sample.h | 13 +++++
> > tools/perf/util/sort.c | 47 +++++++++++++++++++
> > tools/perf/util/sort.h | 2 +
> > 9 files changed, 152 insertions(+), 18 deletions(-)
> >
> >
> > base-commit: 96d541699e5c50b1bc2d50c83cd7145994d5f071
> > --
> > 2.34.1
> >
--
- Arnaldo
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 8+ messages in thread