Skip to content

Commit

Permalink
uuid based timeline
Browse files Browse the repository at this point in the history
  • Loading branch information
@solo2abera
solo2abera committed Jul 10, 2024
1 parent cf74a67 commit b76ca86
Show file tree
Hide file tree
Showing 7 changed files with 188 additions and 63 deletions.
6 changes: 2 additions & 4 deletions utils/xprof_utils.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -57,11 +57,9 @@ typedef std::tuple<hostname_t, process_id_t, thread_id_t, thapi_device_id, thapi
hpt_device_function_name_t;
typedef std::tuple<hostname_t, process_id_t, thapi_device_id> hp_device_t;
typedef std::tuple<hostname_t, thapi_device_id> h_device_t;
typedef std::tuple<hostname_t, process_id_t, thapi_device_id, thapi_fabricPort_id> hpd_fabricPort_t;
typedef std::tuple<hostname_t, process_id_t, thapi_device_id, thapi_device_id> hp_dsd_t;
typedef std::tuple<hostname_t, process_id_t, thapi_device_id, thapi_telemetry_handle> hp_ddomain_t;
typedef std::tuple<hostname_t, process_id_t, thapi_device_id, thapi_telemetry_handle, thapi_sdevice_idx> hp_dsdev_t;
typedef std::tuple<hostname_t, process_id_t, thapi_device_id, thapi_telemetry_handle, bool> hp_dfsdev_t;
typedef std::tuple<hostname_t, process_id_t, thapi_device_id, thapi_telemetry_handle, thapi_domain_idx> hp_ddomain_t;
typedef std::tuple<hostname_t, process_id_t, thapi_device_id, thapi_telemetry_handle, thapi_sdevice_idx, bool> hp_dfsdev_t;
typedef std::tuple<long, long> sd_t;
typedef std::tuple<thread_id_t, thapi_function_name, long> tfn_ts_t;
typedef std::tuple<thapi_function_name, long> fn_ts_t;
Expand Down
25 changes: 25 additions & 0 deletions xprof/btx_interval_model.yaml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -102,6 +102,11 @@
:type: integer_unsigned
:field_value_range: 64
:cast_type: uint64_t
- :name: deviceIdx
:field_class:
:type: integer_unsigned
:field_value_range: 32
:cast_type: uint32_t
- :name: hFrequency
:field_class:
:type: integer_unsigned
Expand All @@ -126,6 +131,11 @@
:type: integer_unsigned
:field_value_range: 64
:cast_type: uint64_t
- :name: deviceIdx
:field_class:
:type: integer_unsigned
:field_value_range: 32
:cast_type: uint32_t
- :name: hPower
:field_class:
:type: integer_unsigned
Expand All @@ -150,6 +160,11 @@
:type: integer_unsigned
:field_value_range: 64
:cast_type: uint64_t
- :name: deviceIdx
:field_class:
:type: integer_unsigned
:field_value_range: 32
:cast_type: uint32_t
- :name: hEngine
:field_class:
:type: integer_unsigned
Expand All @@ -173,6 +188,11 @@
:type: integer_unsigned
:field_value_range: 64
:cast_type: uint64_t
- :name: deviceIdx
:field_class:
:type: integer_unsigned
:field_value_range: 32
:cast_type: uint32_t
- :name: hEngine
:field_class:
:type: integer_unsigned
Expand All @@ -196,6 +216,11 @@
:type: integer_unsigned
:field_value_range: 64
:cast_type: uint64_t
- :name: deviceIdx
:field_class:
:type: integer_unsigned
:field_value_range: 32
:cast_type: uint32_t
- :name: hFabricPort
:field_class:
:type: integer_unsigned
Expand Down
82 changes: 41 additions & 41 deletions xprof/btx_timeline.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -43,7 +43,7 @@ struct FabricDetails {
};

using timeline_dispatch_t = struct timeline_dispatch_s;
using uuid_getter_t = perfetto_uuid_t (*)(timeline_dispatch_t *, const std::string &, uint64_t, uint64_t, uint64_t,
using uuid_getter_t = perfetto_uuid_t (*)(timeline_dispatch_t *, const std::string &, uint64_t, uint64_t, uint32_t, uint64_t,
uint32_t, std::optional<FabricDetails>);

static perfetto_uuid_t gen_perfetto_uuid() {
Expand All @@ -54,7 +54,7 @@ static perfetto_uuid_t gen_perfetto_uuid() {

static perfetto_uuid_t get_parent_counter_track_uuid(timeline_dispatch_t *dispatch,
std::string hostname, uint64_t process_id,
thapi_device_id did) {
thapi_device_id did, uint32_t deviceIdx) {
perfetto_uuid_t hp_uuid = 0;
auto [it, inserted] =
dispatch->hp_device2countertracks.insert({{hostname, did}, hp_uuid});
Expand All @@ -77,7 +77,7 @@ static perfetto_uuid_t get_parent_counter_track_uuid(timeline_dispatch_t *dispat
auto *process = track_descriptor->mutable_process();
process->set_pid(hp_uuid);
std::ostringstream oss;
oss << "Hostname " << hostname << " | Device " << did;
oss << "Hostname " << hostname << " | Device " << deviceIdx;
// oss << " | " << track_name << " | uuid ";
process->set_process_name(oss.str());
return hp_uuid;
Expand All @@ -86,29 +86,29 @@ static perfetto_uuid_t get_parent_counter_track_uuid(timeline_dispatch_t *dispat
static perfetto_uuid_t get_counter_track_uuuid(timeline_dispatch_t *dispatch,
std::unordered_map<hp_ddomain_t, perfetto_uuid_t> &counter_tracks,
const std::string &track_name, const std::string &hostname, uint64_t process_id,
thapi_device_id did, uint64_t tHandle, thapi_domain_idx domain,
thapi_device_id did, uint32_t deviceIdx, uint64_t tHandle, thapi_domain_idx domain,
std::optional<FabricDetails> details = std::nullopt,
std::unordered_map<hp_dfsdev_t, perfetto_uuid_t> *counter_tracks_fp = nullptr) {
perfetto_uuid_t hp_dev_uuid = 0;
perfetto_uuid_t hp_uuid;

if (details && counter_tracks_fp != nullptr) {
auto [it, inserted] = counter_tracks_fp->insert({{hostname, process_id, did, tHandle, details->RxTx},
auto [it, inserted] = counter_tracks_fp->insert({{hostname, process_id, did, tHandle, domain, details->RxTx},
hp_dev_uuid});
auto &potential_uuid = it->second;
if (!inserted)
return potential_uuid;

hp_uuid = get_parent_counter_track_uuid(dispatch, hostname, process_id, did);
hp_uuid = get_parent_counter_track_uuid(dispatch, hostname, process_id, did, deviceIdx);
hp_dev_uuid = gen_perfetto_uuid();
potential_uuid = hp_dev_uuid;
} else {
auto [it, inserted] = counter_tracks.insert({{hostname, process_id, did, tHandle}, hp_dev_uuid});
auto [it, inserted] = counter_tracks.insert({{hostname, process_id, did, domain, tHandle}, hp_dev_uuid});
auto &potential_uuid = it->second;
if (!inserted)
return potential_uuid;

hp_uuid = get_parent_counter_track_uuid(dispatch, hostname, process_id, did);
hp_uuid = get_parent_counter_track_uuid(dispatch, hostname, process_id, did, deviceIdx);
hp_dev_uuid = gen_perfetto_uuid();
potential_uuid = hp_dev_uuid;
}
Expand Down Expand Up @@ -136,38 +136,38 @@ static perfetto_uuid_t get_counter_track_uuuid(timeline_dispatch_t *dispatch,
}

static perfetto_uuid_t get_copyEU_track_uuuid(timeline_dispatch_t *dispatch, const std::string &hostname, uint64_t process_id,
uint64_t did, uint64_t hEngine, uint32_t subDevice, std::optional<FabricDetails> options) {
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, "CopyEngine (%)", hostname, process_id, did, hEngine, subDevice);
uint64_t did, uint32_t deviceIdx, uint64_t hEngine, uint32_t subDevice, std::optional<FabricDetails> options) {
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, "CopyEngine (%)", hostname, process_id, did,deviceIdx, hEngine, subDevice);
}

static perfetto_uuid_t get_computeEU_track_uuuid(timeline_dispatch_t *dispatch, const std::string &hostname, uint64_t process_id,
uint64_t did, uint64_t hEngine, uint32_t subDevice, std::optional<FabricDetails> options ) {
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, "ComputeEngine (%)", hostname, process_id, did, hEngine, subDevice);
uint64_t did, uint32_t deviceIdx, uint64_t hEngine, uint32_t subDevice, std::optional<FabricDetails> options ) {
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, "ComputeEngine (%)", hostname, process_id, did, deviceIdx, hEngine, subDevice);
}

static perfetto_uuid_t get_fpThroughput_track_uuuid(timeline_dispatch_t *dispatch, const std::string &hostname, uint64_t process_id,
uint64_t did, uint64_t hFabricPort, uint32_t subDevice, std::optional<FabricDetails> options) {
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, "FabricT", hostname, process_id, did, hFabricPort, subDevice, options, &dispatch->hp_dfsdev2fptracks);
uint64_t did, uint32_t deviceIdx, uint64_t hFabricPort, uint32_t subDevice, std::optional<FabricDetails> options) {
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, "FabricT", hostname, process_id, did, deviceIdx, hFabricPort, subDevice, options, &dispatch->hp_dfsdev2fptracks);
}

static perfetto_uuid_t get_power_track_uuuid(timeline_dispatch_t *dispatch, const std::string &hostname, uint64_t process_id,
uint64_t did, uint64_t hPower, uint32_t subDevice, std::optional<FabricDetails> options) {
uint64_t did, uint32_t deviceIdx, uint64_t hPower, uint32_t subDevice, std::optional<FabricDetails> options) {
// Extra space to maintain track sequence in the timeline
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, " Power", hostname, process_id, did, hPower, subDevice);
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, " Power", hostname, process_id, did, deviceIdx, hPower, subDevice);
}


static perfetto_uuid_t get_frequency_track_uuuid(timeline_dispatch_t *dispatch, const std::string &hostname, uint64_t process_id,
uint64_t did, uint64_t hFrequency, uint32_t subDevice, std::optional<FabricDetails> options) {
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, " Ferquency", hostname, process_id, did, hFrequency, subDevice);
uint64_t did, uint32_t deviceIdx, uint64_t hFrequency, uint32_t subDevice, std::optional<FabricDetails> options) {
return get_counter_track_uuuid(dispatch, dispatch->hp_ddomain2telmtracks, " Ferquency", hostname, process_id, did, deviceIdx, hFrequency, subDevice);
}

static void add_event_DTelemetry(timeline_dispatch_t *dispatch, const std::string &hostname, uint64_t process_id,
uint64_t thread_id, uint64_t did, uint64_t tHandle, uint32_t subDevice,
uint64_t thread_id, uint64_t did, uint32_t deviceIdx, uint64_t tHandle, uint32_t subDevice,
uint64_t timestamp, float value, uuid_getter_t uuid_getter, const std::string &eventName,
std::optional<FabricDetails> options = std::nullopt) {
perfetto_uuid_t track_uuid;
track_uuid = uuid_getter(dispatch, hostname, process_id, did, tHandle, subDevice, options);
track_uuid = uuid_getter(dispatch, hostname, process_id, did, deviceIdx, tHandle, subDevice, options);

auto *packet = dispatch->trace.add_packet();
packet->set_trusted_packet_sequence_id(TRUSTED_PACKED_SEQUENCE_ID);
Expand All @@ -179,44 +179,44 @@ static void add_event_DTelemetry(timeline_dispatch_t *dispatch, const std::strin
}

static void add_event_fabricPort( timeline_dispatch_t *dispatch, std::string hostname,
uint64_t process_id, uint64_t thread_id, uint64_t did, uintptr_t hFabricPort,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint32_t deviceIdx, uintptr_t hFabricPort,
uint32_t subDevice, uint64_t timestamp, uint32_t fabricId, uint32_t remotePortId,
float rxThroughput, float txThroughput, float rxSpeed, float txSpeed) {
// Define details for RX throughput.
FabricDetails details = {false, fabricId, remotePortId};
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, hFabricPort, subDevice, timestamp,
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, deviceIdx, hFabricPort, subDevice, timestamp,
rxThroughput, get_fpThroughput_track_uuuid, "Fabric ThroughputRX", details);

details.RxTx = true;
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, hFabricPort, subDevice, timestamp,
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, deviceIdx, hFabricPort, subDevice, timestamp,
txThroughput, get_fpThroughput_track_uuuid, "Fabric ThroughputTX", details);
}

static void add_event_frequency(timeline_dispatch_t *dispatch, std::string hostname,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint64_t hFrequency,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint32_t deviceIdx, uint64_t hFrequency,
uint32_t subDevice, uint64_t timestamp, float frequency) {
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, hFrequency, subDevice, timestamp,
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, deviceIdx, hFrequency, subDevice, timestamp,
frequency, get_frequency_track_uuuid, "Frequency");
}

static void add_event_power(timeline_dispatch_t *dispatch, std::string hostname,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint64_t hPower,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint32_t deviceIdx, uint64_t hPower,
uint32_t subDevice, uint64_t timestamp, float power) {
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, hPower, subDevice, timestamp,
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, deviceIdx, hPower, subDevice, timestamp,
power, get_power_track_uuuid, "Power");
}

static void add_event_computeEU(timeline_dispatch_t *dispatch, std::string hostname,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint64_t hEngine,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint32_t deviceIdx, uint64_t hEngine,
uint32_t subDevice, uint64_t timestamp, float activeTime) {
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, hEngine, subDevice, timestamp,
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, deviceIdx, hEngine, subDevice, timestamp,
activeTime, get_computeEU_track_uuuid, "ComputeEngine");
}

static void add_event_copyEU(timeline_dispatch_t *dispatch, std::string hostname,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint64_t hEngine,
uint64_t process_id, uint64_t thread_id, uint64_t did, uint32_t deviceIdx, uint64_t hEngine,
uint32_t subDevice, uint64_t timestamp, float activeTime) {
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, hEngine, subDevice, timestamp,
add_event_DTelemetry(dispatch, hostname, process_id, thread_id, did, deviceIdx, hEngine, subDevice, timestamp,
activeTime, get_copyEU_track_uuuid, "CopyEngine");
}

Expand Down Expand Up @@ -431,39 +431,39 @@ static void device_usr_callback(void *btx_handle, void *usr_data, const char *ho

static void frequency_usr_callback(void *btx_handle, void *usr_data, const char *hostname,
int64_t vpid, uint64_t vtid, int64_t ts, int64_t backend,
uint64_t did, uint64_t hFrequency, uint32_t domain, uint64_t frequency) {
uint64_t did, uint32_t deviceIdx, uint64_t hFrequency, uint32_t domain, uint64_t frequency) {
auto *dispatch = static_cast<timeline_dispatch_t *>(usr_data);
add_event_frequency(dispatch, hostname, vpid, vtid, did, hFrequency, domain, ts, frequency);
add_event_frequency(dispatch, hostname, vpid, vtid, did, deviceIdx, hFrequency, domain, ts, frequency);
}

static void power_usr_callback(void *btx_handle, void *usr_data, const char *hostname, int64_t vpid,
uint64_t vtid, int64_t ts, int64_t backend, uint64_t did, uint64_t hPower,
uint64_t vtid, int64_t ts, int64_t backend, uint64_t did, uint32_t deviceIdx, uint64_t hPower,
uint32_t domain, uint64_t power) {
auto *dispatch = static_cast<timeline_dispatch_t *>(usr_data);
add_event_power(dispatch, hostname, vpid, vtid, did, hPower, domain, ts, power);
add_event_power(dispatch, hostname, vpid, vtid, did, hPower, deviceIdx, domain, ts, power);
}

static void computeEU_usr_callback(void *btx_handle, void *usr_data, const char *hostname,
int64_t vpid, uint64_t vtid, int64_t ts, int64_t backend,
uint64_t did, uint64_t hEngine, uint32_t subDevice, float activeTime) {
uint64_t did, uint32_t deviceIdx, uint64_t hEngine, uint32_t subDevice, float activeTime) {
auto *dispatch = static_cast<timeline_dispatch_t *>(usr_data);
add_event_computeEU(dispatch, hostname, vpid, vtid, did, hEngine, subDevice, ts, activeTime);
add_event_computeEU(dispatch, hostname, vpid, vtid, did, deviceIdx, hEngine, subDevice, ts, activeTime);
}

static void copyEU_usr_callback(void *btx_handle, void *usr_data, const char *hostname,
int64_t vpid, uint64_t vtid, int64_t ts, int64_t backend,
uint64_t did, uint64_t hEngine, uint32_t subDevice, float activeTime) {
uint64_t did, uint32_t deviceIdx, uint64_t hEngine, uint32_t subDevice, float activeTime) {
auto *dispatch = static_cast<timeline_dispatch_t *>(usr_data);
add_event_copyEU(dispatch, hostname, vpid, vtid, did, hEngine, subDevice, ts, activeTime);
add_event_copyEU(dispatch, hostname, vpid, vtid, did, deviceIdx, hEngine, subDevice, ts, activeTime);
}

static void fabricPort_usr_callback(void *btx_handle, void *usr_data, const char *hostname,
int64_t vpid, uint64_t vtid, int64_t ts, int64_t backend,
uint64_t did, uint64_t hFabricPort, uint32_t subDevice,
uint64_t did, uint32_t deviceIdx, uint64_t hFabricPort, uint32_t subDevice,
uint32_t fabricId, uint32_t remotePortId, float rxThroughput,
float txThroughput, float rxSpeed, float txSpeed) {
auto *dispatch = static_cast<timeline_dispatch_t *>(usr_data);
add_event_fabricPort(dispatch, hostname, vpid, vtid, did, hFabricPort, subDevice, ts, fabricId,
add_event_fabricPort(dispatch, hostname, vpid, vtid, did, deviceIdx, hFabricPort, subDevice, ts, fabricId,
remotePortId, rxThroughput, txThroughput, rxSpeed, txSpeed);
}

Expand Down
Loading

0 comments on commit b76ca86

Please sign in to comment.