Fix rocprofiler plugin

- Replace non-working fields with working ones
    - remove CU_OCCUPANCY completely as it isn't well supported
- Fix rocprofiler initialization with shared_ptr and rdc_module_init
- Replace env var ROCPROFILER_METRICS_PATH with ROCP_METRICS
    - ROCPROFILER_METRICS_PATH is only relevant for rocprofv2
    - ROCP_METRICS is only relevant for rocprofv1 (which we are using)

Change-Id: I21e6fa3f0e1694c38f44ca0e5659d672559f7380
Signed-off-by: Galantsev, Dmitrii <dmitrii.galantsev@amd.com>
This commit is contained in:
Galantsev, Dmitrii
2024-06-04 19:37:42 -05:00
parent 07c414af5e
commit 20ca2ce574
8 changed files with 62 additions and 37 deletions
@@ -243,7 +243,6 @@ RdcRocpBase::RdcRocpBase() {
// all fields
static const std::map<rdc_field_t, const char*> temp_field_map_k = {
{RDC_FI_PROF_CU_OCCUPANCY, "CU_OCCUPANCY"},
{RDC_FI_PROF_MEAN_OCCUPANCY_PER_CU, "MEAN_OCCUPANCY_PER_CU"},
{RDC_FI_PROF_MEAN_OCCUPANCY_PER_ACTIVE_CU, "MEAN_OCCUPANCY_PER_ACTIVE_CU"},
{RDC_FI_PROF_ACTIVE_CYCLES, "ACTIVE_CYCLES"},
@@ -36,14 +36,24 @@ THE SOFTWARE.
#include "rdc_lib/rdc_common.h"
#include "rdc_modules/rdc_rocp/RdcRocpBase.h"
amd::rdc::RdcRocpBase rocp;
std::unique_ptr<amd::rdc::RdcRocpBase> rocp_p;
rdc_status_t rdc_module_init(uint64_t flags) { return RDC_ST_OK; }
rdc_status_t rdc_module_init(uint64_t flags) {
rocp_p = std::unique_ptr<amd::rdc::RdcRocpBase>(new amd::rdc::RdcRocpBase);
return RDC_ST_OK;
}
rdc_status_t rdc_module_destroy() {
rocp_p.reset();
return RDC_ST_OK;
}
// get supported field ids
rdc_status_t rdc_telemetry_fields_query(uint32_t field_ids[MAX_NUM_FIELDS], uint32_t* field_count) {
// extract all keys from counter_map
std::vector<rdc_field_t> fields = rocp.get_field_ids();
if (rocp_p == nullptr) {
return RDC_ST_FAIL_LOAD_MODULE;
}
std::vector<rdc_field_t> fields = rocp_p->get_field_ids();
std::vector<uint32_t> counter_keys(fields.begin(), fields.end());
*field_count = counter_keys.size();
@@ -56,7 +66,10 @@ rdc_status_t rdc_telemetry_fields_query(uint32_t field_ids[MAX_NUM_FIELDS], uint
// Fetch
rdc_status_t rdc_telemetry_fields_value_get(rdc_gpu_field_t* fields, const uint32_t fields_count,
rdc_field_value_f callback, void* user_data) {
//
if (rocp_p == nullptr) {
return RDC_ST_FAIL_LOAD_MODULE;
}
// Bulk fetch fields
std::vector<rdc_gpu_field_value_t> bulk_results;
@@ -81,7 +94,7 @@ rdc_status_t rdc_telemetry_fields_value_get(rdc_gpu_field_t* fields, const uint3
bulk_count = 0;
}
status = rocp.rocp_lookup(fields[i].gpu_index, fields[i].field_id, &data);
status = rocp_p->rocp_lookup(fields[i].gpu_index, fields[i].field_id, &data);
// get value
values[bulk_count].gpu_index = fields[i].gpu_index;
values[bulk_count].field_value.type = DOUBLE;