AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

rocprofiler_kernel_dispatch_info_t + header record for buffered counter collection (#758)

Browse Source 
* Update include/rocprofiler-sdk

- defines.h
  - ROCPROFILER_VERSION_10_0 -> ROCPROFILER_SDK_VERSION_0_0
- fwd.h
  - rocprofiler_counter_record_kind_t
  - rocprofiler_kernel_dispatch_info_t
  - rocprofiler_record_counter_t
    - has dispatch id instead of correlation id
  - rocprofiler_counter_info_v0_t
    - added rocprofiler_counter_id_t field
    - added is_constant field
    - reordered better packing
- dispatch_profile.h
  - added rocprofiler_profile_counting_dispatch_record_t for use as a header record for rocprofiler_profile_counting_dispatch_data_t
- callback_tracing.h
  - rocprofiler_callback_tracing_kernel_dispatch_data_t uses rocprofiler_kernel_dispatch_info_t
- buffer_tracing.h
  - rocprofiler_buffer_tracing_kernel_dispatch_record_t uses rocprofiler_kernel_dispatch_info_t

* Update lib/rocprofiler-sdk/*

- transition to rocprofiler_kernel_dispatch_info_t
- set id and is_constant values for rocprofiler_counter_info_v0_t in rocprofiler_query_counter_info

* Update lib/rocprofiler-sdk-tool

- transition to rocprofiler_kernel_dispatch_info_t

* Update lib/rocprofiler-sdk/counters/tests/core.cpp

- transition to rocprofiler_kernel_dispatch_info_t

* Update samples

- transition to rocprofiler_kernel_dispatch_info_t
- transition to rocprofiler_counter_record_kind_t

* Update tests

- transition to rocprofiler_kernel_dispatch_info_t
- transition to rocprofiler_counter_record_kind_t
- improve integration test validation for counter-collection
- update serialization for new/additional types

* Fix tests/counter-collection/validate.py

- loosen restrictions on the length of counter description

* Update include/rocprofiler-sdk/buffer_tracing.h

- remove accidental packed attribute

* Update lib/rocprofiler-sdk/counters/xml/derived_counters.xml

- Add description for TCC_TAG_STALL_sum (reference: https://rocm.docs.amd.com/en/develop/conceptual/gpu-arch/mi300-mi200-performance-counters.html)

* Update tests/page-migration/validate.py

[ROCm/rocprofiler-sdk commit: 07537b6231]
This commit is contained in:
Jonathan R. Madsen
2024-04-12 17:30:34 -05:00
committed by GitHub
parent 4f99edbad5
commit 2aef3c3d15
23 changed files with 489 additions and 273 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocprofiler-sdk/samples/api_buffered_tracing/client.cpp
+16 -9
View File
@@ -60,6 +60,7 @@
#include <iostream>
#include <map>
#include <mutex>
#include <sstream>
#include <string>
#include <string_view>
#include <thread>
@@ -222,19 +223,22 @@ tool_tracing_callback(rocprofiler_context_id_t context,
auto info = std::stringstream{};
info << "agent_id=" << record->agent_id.handle
<< ", queue_id=" << record->queue_id.handle << ", kernel_id=" << record->kernel_id
<< ", kernel=" << client_kernels.at(record->kernel_id).kernel_name
info << "agent_id=" << record->dispatch_info.agent_id.handle
<< ", queue_id=" << record->dispatch_info.queue_id.handle
<< ", kernel_id=" << record->dispatch_info.kernel_id
<< ", kernel=" << client_kernels.at(record->dispatch_info.kernel_id).kernel_name
<< ", context=" << context.handle << ", buffer_id=" << buffer_id.handle
<< ", cid=" << record->correlation_id.internal
<< ", extern_cid=" << record->correlation_id.external.value
<< ", kind=" << record->kind << ", start=" << record->start_timestamp
<< ", stop=" << record->end_timestamp
<< ", private_segment_size=" << record->private_segment_size
<< ", group_segment_size=" << record->group_segment_size << ", workgroup_size=("
<< record->workgroup_size.x << "," << record->workgroup_size.y << ","
<< record->workgroup_size.z << "), grid_size=(" << record->grid_size.x << ","
<< record->grid_size.y << "," << record->grid_size.z << ")";
<< ", private_segment_size=" << record->dispatch_info.private_segment_size
<< ", group_segment_size=" << record->dispatch_info.group_segment_size
<< ", workgroup_size=(" << record->dispatch_info.workgroup_size.x << ","
<< record->dispatch_info.workgroup_size.y << ","
<< record->dispatch_info.workgroup_size.z << "), grid_size=("
<< record->dispatch_info.grid_size.x << "," << record->dispatch_info.grid_size.y
<< "," << record->dispatch_info.grid_size.z << ")";
if(record->start_timestamp > record->end_timestamp)
throw std::runtime_error("kernel dispatch: start > end");
@@ -267,7 +271,10 @@ tool_tracing_callback(rocprofiler_context_id_t context,
}
else
{
throw std::runtime_error{"unexpected rocprofiler_record_header_t category + kind"};
auto _msg = std::stringstream{};
_msg << "unexpected rocprofiler_record_header_t category + kind: (" << header->category
<< " + " << header->kind << ")";
throw std::runtime_error{_msg.str()};
}
}
}
projects/rocprofiler-sdk/samples/counter_collection/callback_client.cpp
+14 -11
View File
@@ -74,12 +74,13 @@ record_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
void* callback_data_args)
{
std::stringstream ss;
ss << "Kernel_id " << dispatch_data.kernel_id << ": ";
ss << "Dispatch_Id=" << dispatch_data.dispatch_info.dispatch_id
<< ", Kernel_id=" << dispatch_data.dispatch_info.kernel_id
<< ", Corr_Id=" << dispatch_data.correlation_id.internal << ": ";
for(size_t i = 0; i < record_count; ++i)
{
ss << "(Id: " << record_data[i].id << " Value [D]: " << record_data[i].counter_value
<< " Corr_Id: " << record_data[i].correlation_id.internal << "),";
}
<< "),";
auto* output_stream = static_cast<std::ostream*>(callback_data_args);
if(!output_stream) throw std::runtime_error{"nullptr to output stream"};
*output_stream << "[" << __FUNCTION__ << "] " << ss.str() << "\n";
@@ -110,7 +111,8 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
static std::unordered_map<uint64_t, rocprofiler_profile_config_id_t> profile_cache = {};
auto search_cache = [&]() {
if(auto pos = profile_cache.find(dispatch_data.agent_id.handle); pos != profile_cache.end())
if(auto pos = profile_cache.find(dispatch_data.dispatch_info.agent_id.handle);
pos != profile_cache.end())
{
*config = pos->second;
return true;
@@ -133,7 +135,7 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
// Iterate through the agents and get the counters available on that agent
ROCPROFILER_CALL(rocprofiler_iterate_agent_supported_counters(
dispatch_data.agent_id,
dispatch_data.dispatch_info.agent_id,
[](rocprofiler_agent_id_t,
rocprofiler_counter_id_t* counters,
size_t num_counters,
@@ -167,12 +169,13 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
// Create a colleciton profile for the counters
rocprofiler_profile_config_id_t profile;
ROCPROFILER_CALL(
rocprofiler_create_profile_config(
dispatch_data.agent_id, collect_counters.data(), collect_counters.size(), &profile),
"Could not construct profile cfg");
ROCPROFILER_CALL(rocprofiler_create_profile_config(dispatch_data.dispatch_info.agent_id,
collect_counters.data(),
collect_counters.size(),
&profile),
"Could not construct profile cfg");
profile_cache.emplace(dispatch_data.agent_id.handle, profile);
profile_cache.emplace(dispatch_data.dispatch_info.agent_id.handle, profile);
// Return the profile to collect those counters for this dispatch
*config = profile;
}
projects/rocprofiler-sdk/samples/counter_collection/client.cpp
+23 -10
View File
@@ -95,12 +95,23 @@ buffered_callback(rocprofiler_context_id_t,
for(size_t i = 0; i < num_headers; ++i)
{
auto* header = headers[i];
if(header->category == ROCPROFILER_BUFFER_CATEGORY_COUNTERS && header->kind == 0)
if(header->category == ROCPROFILER_BUFFER_CATEGORY_COUNTERS &&
header->kind == ROCPROFILER_COUNTER_RECORD_PROFILE_COUNTING_DISPATCH_HEADER)
{
// Print the returned counter data.
auto* record =
static_cast<rocprofiler_profile_counting_dispatch_record_t*>(header->payload);
ss << "[Dispatch_Id: " << record->dispatch_info.dispatch_id
<< " Kernel_ID: " << record->dispatch_info.kernel_id
<< " Corr_Id: " << record->correlation_id.internal << ")]\n";
}
else if(header->category == ROCPROFILER_BUFFER_CATEGORY_COUNTERS &&
header->kind == ROCPROFILER_COUNTER_RECORD_VALUE)
{
// Print the returned counter data.
auto* record = static_cast<rocprofiler_record_counter_t*>(header->payload);
ss << "(Id: " << record->id << " Value [D]: " << record->counter_value
<< " Corr_Id: " << record->correlation_id.internal << "),";
ss << " (Dispatch_Id: " << record->dispatch_id << " Id: " << record->id
<< " Value [D]: " << record->counter_value << "),";
}
}
@@ -133,7 +144,8 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
static std::unordered_map<uint64_t, rocprofiler_profile_config_id_t> profile_cache = {};
auto search_cache = [&]() {
if(auto pos = profile_cache.find(dispatch_data.agent_id.handle); pos != profile_cache.end())
if(auto pos = profile_cache.find(dispatch_data.dispatch_info.agent_id.handle);
pos != profile_cache.end())
{
*config = pos->second;
return true;
@@ -156,7 +168,7 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
// Iterate through the agents and get the counters available on that agent
ROCPROFILER_CALL(rocprofiler_iterate_agent_supported_counters(
dispatch_data.agent_id,
dispatch_data.dispatch_info.agent_id,
[](rocprofiler_agent_id_t,
rocprofiler_counter_id_t* counters,
size_t num_counters,
@@ -190,12 +202,13 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
// Create a colleciton profile for the counters
rocprofiler_profile_config_id_t profile;
ROCPROFILER_CALL(
rocprofiler_create_profile_config(
dispatch_data.agent_id, collect_counters.data(), collect_counters.size(), &profile),
"Could not construct profile cfg");
ROCPROFILER_CALL(rocprofiler_create_profile_config(dispatch_data.dispatch_info.agent_id,
collect_counters.data(),
collect_counters.size(),
&profile),
"Could not construct profile cfg");
profile_cache.emplace(dispatch_data.agent_id.handle, profile);
profile_cache.emplace(dispatch_data.dispatch_info.agent_id.handle, profile);
// Return the profile to collect those counters for this dispatch
*config = profile;
}
projects/rocprofiler-sdk/samples/counter_collection/print_functional_counters.cpp
+11 -8
View File
@@ -170,7 +170,8 @@ buffered_callback(rocprofiler_context_id_t,
for(size_t i = 0; i < num_headers; ++i)
{
auto* header = headers[i];
if(header->category == ROCPROFILER_BUFFER_CATEGORY_COUNTERS && header->kind == 0)
if(header->category == ROCPROFILER_BUFFER_CATEGORY_COUNTERS &&
header->kind == ROCPROFILER_COUNTER_RECORD_VALUE)
{
// Record the counters we have in the buffer and the number of instances of
// the counter we have seen.
@@ -242,7 +243,7 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
{
std::vector<rocprofiler_counter_id_t> counters_needed;
ROCPROFILER_CALL(rocprofiler_iterate_agent_supported_counters(
dispatch_data.agent_id,
dispatch_data.dispatch_info.agent_id,
[](rocprofiler_agent_id_t,
rocprofiler_counter_id_t* counters,
size_t num_counters,
@@ -269,7 +270,7 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
cap.expected_counter_names.emplace(found_counter.handle, std::string(version.name));
size_t expected = 0;
ROCPROFILER_CALL(rocprofiler_query_counter_instance_count(
dispatch_data.agent_id, found_counter, &expected),
dispatch_data.dispatch_info.agent_id, found_counter, &expected),
"COULD NOT QUERY INSTANCES");
cap.remaining.push_back(found_counter);
cap.expected.emplace(found_counter.handle, expected);
@@ -297,8 +298,9 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
}
if(cap.expected.empty())
{
std::clog << "No counters found for agent " << dispatch_data.agent_id.handle << " ("
<< agents.at(dispatch_data.agent_id.handle)->name << ")";
std::clog << "No counters found for agent "
<< dispatch_data.dispatch_info.agent_id.handle << " ("
<< agents.at(dispatch_data.dispatch_info.agent_id.handle)->name << ")";
}
}
if(cap.remaining.empty()) return;
@@ -306,9 +308,10 @@ dispatch_callback(rocprofiler_profile_counting_dispatch_data_t dispatch_data,
rocprofiler_profile_config_id_t profile;
// Select the next counter to collect.
ROCPROFILER_CALL(rocprofiler_create_profile_config(
dispatch_data.agent_id, &(cap.remaining.back()), 1, &profile),
"Could not construct profile cfg");
ROCPROFILER_CALL(
rocprofiler_create_profile_config(
dispatch_data.dispatch_info.agent_id, &(cap.remaining.back()), 1, &profile),
"Could not construct profile cfg");
cap.remaining.pop_back();
*config = profile;