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

Public C++ header files and samples updates (#819)

Browse Source 
* Public C++ header files (source/include/rocprofiler-sdk/cxx)

* Update samples/api_buffered_tracing

- scratch memory and page migration
- README

* Update samples/api_buffered_tracing

- page migration component in sample

* Update tests/page-migration/validate.py

- fix checks for page migration operation names

* Update tests/page-migration/validate.py

- fix get_allocated_pages

* Update scratch memory and page migration validations

* Fix include/rocprofiler-sdk/cxx installation

* Rework include/rocprofiler-sdk/cxx

- Improve name_info to support const char*, string_view, string

* Update samples/api_{buffered,callback}_tracing

* External correlation ID request sample

- includes correlation ID retirement demo

* Update samples/api_buffered_tracing/README.md

* Update lib/rocprofiler-sdk/hsa/queue.cpp

- generate correlation ID for kernel launch if one doesn't exist

* Remove priority check from tool libraries (samples/tests)

- if(priority > 0) return nullptr check in rocprofiler_configure has proliferated beyond its intended use

* Apply suggestions from code review
This commit is contained in:
Jonathan R. Madsen
2024-04-25 20:09:11 -05:00
committed by GitHub
parent e2bce49655
commit de13d2ac5d
37 changed files with 2232 additions and 512 deletions
Download Patch File Download Diff File Expand all files Collapse all files
source/lib/rocprofiler-sdk/hsa/queue.cpp
+37 -20
View File
@@ -21,6 +21,7 @@
THE SOFTWARE. */
#include "lib/rocprofiler-sdk/hsa/queue.hpp"
#include "lib/common/scope_destructor.hpp"
#include "lib/common/utility.hpp"
#include "lib/rocprofiler-sdk/agent.hpp"
#include "lib/rocprofiler-sdk/buffer.hpp"
@@ -156,8 +157,8 @@ AsyncSignalHandler(hsa_signal_value_t /*signal_v*/, void* data)
LOG_IF(FATAL, _corr_id->get_ref_count() == 0)
<< "reference counter for correlation id " << _corr_id->internal << " from thread "
<< _corr_id->thread_idx << " has no reference count";
_corr_id->sub_ref_count();
_corr_id->sub_kern_count();
_corr_id->sub_ref_count();
}
queue_info_session.queue.async_complete();
@@ -239,18 +240,6 @@ WriteInterceptor(const void* packets,
ROCPROFILER_BUFFER_TRACING_KERNEL_DISPATCH,
tracing_data_v);
auto* corr_id = context::get_latest_correlation_id();
auto thr_id = (corr_id) ? corr_id->thread_idx : common::get_tid();
auto user_data = rocprofiler_user_data_t{.value = 0};
auto internal_corr_id = (corr_id) ? corr_id->internal : 0;
tracing::populate_external_correlation_ids(
tracing_data_v.external_correlation_ids,
thr_id,
ROCPROFILER_EXTERNAL_CORRELATION_REQUEST_KERNEL_DISPATCH,
ROCPROFILER_KERNEL_DISPATCH_ENQUEUE,
internal_corr_id);
const auto* packets_arr = static_cast<const rocprofiler_packet*>(packets);
auto transformed_packets = std::vector<rocprofiler_packet>{};
@@ -267,6 +256,41 @@ WriteInterceptor(const void* packets,
continue;
}
auto* corr_id = context::get_latest_correlation_id();
context::correlation_id* _corr_id_pop = nullptr;
if(!corr_id)
{
constexpr auto ref_count = 1;
corr_id = context::correlation_tracing_service::construct(ref_count);
_corr_id_pop = corr_id;
}
// increase the reference count to denote that this correlation id is being used in a kernel
corr_id->add_ref_count();
corr_id->add_kern_count();
auto thr_id = (corr_id) ? corr_id->thread_idx : common::get_tid();
auto user_data = rocprofiler_user_data_t{.value = 0};
auto internal_corr_id = (corr_id) ? corr_id->internal : 0;
// if we constructed a correlation id, this decrements the reference count after the
// underlying function returns
auto _corr_id_dtor = common::scope_destructor{[_corr_id_pop]() {
if(_corr_id_pop)
{
context::pop_latest_correlation_id(_corr_id_pop);
_corr_id_pop->sub_ref_count();
}
}};
tracing::populate_external_correlation_ids(
tracing_data_v.external_correlation_ids,
thr_id,
ROCPROFILER_EXTERNAL_CORRELATION_REQUEST_KERNEL_DISPATCH,
ROCPROFILER_KERNEL_DISPATCH_ENQUEUE,
internal_corr_id);
queue.async_started();
// Copy kernel pkt, copy is to allow for signal to be modified
rocprofiler_packet kernel_pkt = packets_arr[i];
@@ -274,13 +298,6 @@ WriteInterceptor(const void* packets,
queue.create_signal(HSA_AMD_SIGNAL_AMD_GPU_ONLY,
&kernel_pkt.ext_amd_aql_pm4.completion_signal);
// increase the reference count to denote that this correlation id is being used in a kernel
if(corr_id)
{
corr_id->add_ref_count();
corr_id->add_kern_count();
}
// computes the "size" based on the offset of reserved_padding field
constexpr auto kernel_dispatch_info_rt_size =
common::compute_runtime_sizeof<rocprofiler_kernel_dispatch_info_t>();