Enable ATT continuous mode and code object tracing registration (#850)

* Adding ATT continuous mode and ATT code object tracking

* Fixing aql_packet.cpp

* Updating to aqlprofile codeobj changes

* Removing kernel packet from ATT dispatch callback

* Changing getSymbolMap() to return relative vaddr

* Tidy fixes

* Formatting

* Fix shadowing

* Fixing packet test

* Updating tests

* Simplifying multi-agent traces

* Adding dynamic codeobj tracking

* leftover book-keeping for codeobj markers

* Formatting

* Formatting

* Temporary removing codeobj marker

* Formatting

* Re-enabling codeobj tracking

* Making copy of coreapi table

* Fixing issues with toolData lifetile

* Formatting

* Fixing issues with ASAN

* Improving memory profile

* Removing misplaced annotation

* Fixing queue type and allowing shared_locks in globalThreadTracer

* Update logging

* Changing ATT formats to be more in line with the SDk (#883)

* Fixing some merge conflicts

* Fixing cmakelists

* Fixing merge conflicts

* Formatting
This commit is contained in:
Giovanni Lenzi Baraldi
2024-05-29 13:09:28 -03:00
committed by GitHub
parent 2225153c23
commit 1b95089c28
24 changed files with 1696 additions and 824 deletions
@@ -69,12 +69,15 @@ CoreApiTable&
get_api_table()
{
static auto _v = []() {
auto val = CoreApiTable{};
val.hsa_iterate_agents_fn = hsa_iterate_agents;
val.hsa_agent_get_info_fn = hsa_agent_get_info;
val.hsa_queue_create_fn = hsa_queue_create;
val.hsa_queue_destroy_fn = hsa_queue_destroy;
val.hsa_signal_wait_relaxed_fn = hsa_signal_wait_relaxed;
auto val = CoreApiTable{};
val.hsa_iterate_agents_fn = hsa_iterate_agents;
val.hsa_agent_get_info_fn = hsa_agent_get_info;
val.hsa_queue_create_fn = hsa_queue_create;
val.hsa_queue_destroy_fn = hsa_queue_destroy;
val.hsa_signal_wait_relaxed_fn = hsa_signal_wait_relaxed;
val.hsa_queue_load_read_index_relaxed_fn = hsa_queue_load_read_index_relaxed;
val.hsa_queue_add_write_index_relaxed_fn = hsa_queue_add_write_index_relaxed;
val.hsa_signal_store_screlease_fn = hsa_signal_store_screlease;
return val;
}();
return _v;
@@ -83,32 +86,40 @@ get_api_table()
void
test_init()
{
HsaApiTable table;
table.amd_ext_ = &get_ext_table();
table.core_ = &get_api_table();
agent::construct_agent_cache(&table);
ASSERT_TRUE(hsa::get_queue_controller() != nullptr);
hsa::get_queue_controller()->init(get_api_table(), get_ext_table());
auto init = []() -> bool {
HsaApiTable table;
table.amd_ext_ = &get_ext_table();
table.core_ = &get_api_table();
agent::construct_agent_cache(&table);
hsa::get_queue_controller()->init(get_api_table(), get_ext_table());
return true;
};
[[maybe_unused]] static bool run_ince = init();
}
} // namespace rocprofiler
using namespace rocprofiler::aql;
using namespace rocprofiler;
TEST(thread_trace, construct_default_packets)
TEST(thread_trace, resource_creation)
{
ASSERT_EQ(hsa_init(), HSA_STATUS_SUCCESS);
rocprofiler::test_init();
auto agents = rocprofiler::hsa::get_queue_controller()->get_supported_agents();
test_init();
registration::init_logging();
registration::set_init_status(-1);
auto agents = hsa::get_queue_controller()->get_supported_agents();
ASSERT_GT(agents.size(), 0);
for(const auto& [_, agent] : agents)
{
auto params = std::make_shared<rocprofiler::thread_trace_parameter_pack>();
auto params = thread_trace_parameter_pack{};
ThreadTraceAQLPacketFactory factory(
agent, params, rocprofiler::get_api_table(), rocprofiler::get_ext_table());
aql::ThreadTraceAQLPacketFactory factory(agent, params, get_api_table(), get_ext_table());
auto packet = factory.construct_packet();
packet->populate_before();
packet->populate_after();
size_t vendor_packet = HSA_PACKET_TYPE_VENDOR_SPECIFIC << HSA_PACKET_HEADER_TYPE;
ASSERT_TRUE(packet->start.header == vendor_packet);
@@ -116,16 +127,42 @@ TEST(thread_trace, construct_default_packets)
ASSERT_TRUE(packet->before_krn_pkt.size() > 0);
ASSERT_TRUE(packet->after_krn_pkt.size() > 0);
}
{
thread_trace_parameter_pack params{};
GlobalThreadTracer tracer(std::move(params));
for(const auto& [_, agent] : agents)
{
// Init twice to simulate two queues
tracer.resource_init(agent, get_api_table(), get_ext_table());
tracer.resource_init(agent, get_api_table(), get_ext_table());
}
for(auto& [_, agenttracer] : tracer.agents)
{
agenttracer->load_codeobj(1, 0x1000, 0x1000);
agenttracer->load_codeobj(2, 0x3000, 0x1000);
agenttracer->unload_codeobj(1);
}
for(const auto& [_, agent] : agents)
{
// Deinit twice to remove both queues
tracer.resource_deinit(agent);
tracer.resource_deinit(agent);
}
}
hsa_shut_down();
}
TEST(thread_trace, configure_test)
{
rocprofiler::test_init();
test_init();
rocprofiler::registration::init_logging();
rocprofiler::registration::set_init_status(-1);
rocprofiler::context::push_client(1);
registration::init_logging();
registration::set_init_status(-1);
context::push_client(1);
rocprofiler_context_id_t ctx;
ROCPROFILER_CALL(rocprofiler_create_context(&ctx), "context creation failed");
@@ -142,8 +179,7 @@ TEST(thread_trace, configure_test)
[](rocprofiler_queue_id_t,
const rocprofiler_agent_t*,
rocprofiler_correlation_id_t,
const hsa_kernel_dispatch_packet_t*,
uint64_t,
rocprofiler_kernel_id_t,
void*) { return ROCPROFILER_ATT_CONTROL_NONE; },
[](int64_t, void*, size_t, void*) {},
nullptr);
@@ -151,4 +187,5 @@ TEST(thread_trace, configure_test)
ASSERT_EQ(hsa_init(), HSA_STATUS_SUCCESS);
ROCPROFILER_CALL(rocprofiler_start_context(ctx), "context start failed");
ROCPROFILER_CALL(rocprofiler_stop_context(ctx), "context stop failed");
hsa_shut_down();
}