AI/rocm-systems
2
0 Vork 0
Code Kwesties Pull-aanvragen Actions Paketten Projecten Publicaties Wiki Activiteit

ATT API changes - add user_data field and separation of dispatch vs agent profiling (#893)

Bladeren bron 
* DRM Issue Fix for SLES 15 (#897)

* DRM Issue Fix

* Formatting Fix

* PC sampling: CID manager unit test (#898)

* Adding per-dispatch userdata field to ATT

* Clang tidy

* Formatting

* Update source/lib/rocprofiler-sdk/hsa/aql_packet.hpp

Co-authored-by: Vladimir Indic <139573562+vlaindic@users.noreply.github.com>

* Adding dispatch_id, fixing user_data and update aql_profile_v2

* Formatting

* Tidy fixes

* Second fix for userdata

* removing assert for union

* Adding serialization. Created agent profiling-like thread trace

* Implemented agent thread trace

* Update source/lib/rocprofiler-sdk/hsa/aql_packet.hpp

Co-authored-by: Vladimir Indic <139573562+vlaindic@users.noreply.github.com>

* Restructured thread trace packets

* Added agent API tests

* Fixing multigpu for agent test

* Formatting

* Formatting

* Improving header locations

* Fixing merge conflicts

* Tidy

* Tidy

* Tidy

---------

Co-authored-by: Ammar ELWazir <ammar.elwazir@amd.com>
Co-authored-by: Vladimir Indic <139573562+vlaindic@users.noreply.github.com>
This commit is contained in:
Giovanni Lenzi BaraldiVladimir IndicAmmar ELWazir
 authored and GitHub committed 2024-06-13 15:29:29 -03:00
1 bovenliggende 2cfd88e571
commit 9676295d3d
24 files changed
+1154 -635
Download Patch-bestand Download Diff-bestand Expand all files Collapse all files

No files matched your search

source/lib/rocprofiler-sdk/aql/packet_construct.cpp
+15 -18
Bestand weergeven
@@ -219,31 +219,28 @@ ThreadTraceAQLPacketFactory::ThreadTraceAQLPacketFactory(const hsa::AgentCache&
uint32_t perf_ctrl = static_cast<uint32_t>(params.perfcounter_ctrl);
aql_params.clear();
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_COMPUTE_UNIT_TARGET, cu});
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_SE_MASK, shader_engine_mask});
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_SIMD_SELECTION, simd});
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_ATT_BUFFER_SIZE, buffer_size});
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_PERFCOUNTER_CTRL, perf_ctrl});
for(uint32_t perf_counter : params.perfcounters)
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_COMPUTE_UNIT_TARGET, {cu}});
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_SE_MASK, {shader_engine_mask}});
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_SIMD_SELECTION, {simd}});
aql_params.push_back({HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_ATT_BUFFER_SIZE, {buffer_size}});
if(perf_ctrl != 0 && !params.perfcounters.empty())
{
aql_params.push_back(
{HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_PERFCOUNTER_NAME, perf_counter});
{HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_PERFCOUNTER_CTRL, {perf_ctrl - 1}});
auto perf_param = HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_PERFCOUNTER_NAME;
for(uint32_t perf_counter : params.perfcounters)
aql_params.push_back({perf_param, {perf_counter}});
}
}
const std::vector<hsa_ven_amd_aqlprofile_parameter_t>&
ThreadTraceAQLPacketFactory::get_aql_params()
{
return aql_params;
}
std::unique_ptr<hsa::TraceControlAQLPacket>
ThreadTraceAQLPacketFactory::construct_packet()
ThreadTraceAQLPacketFactory::construct_control_packet()
{
uint32_t num_params = static_cast<uint32_t>(get_aql_params().size());
auto profile =
aqlprofile_att_profile_t{tracepool.gpu_agent, get_aql_params().data(), num_params};
auto packet = std::make_unique<hsa::TraceControlAQLPacket>(this->tracepool, profile);
auto num_params = static_cast<uint32_t>(aql_params.size());
auto profile = aqlprofile_att_profile_t{tracepool.gpu_agent, aql_params.data(), num_params};
auto packet = std::make_unique<hsa::TraceControlAQLPacket>(this->tracepool, profile);
packet->clear();
return packet;
}