Workaround for SWDEV-559598. Enabling more thread trace tests. (#1336)

* Workaround for SWDEV-559598

* gfx11 fix
Этот коммит содержится в:
Giovanni Lenzi Baraldi
2025-11-27 20:03:38 +01:00
коммит произвёл GitHub
родитель 259010f2d5
Коммит 0e04fdd571
4 изменённых файлов: 49 добавлений и 15 удалений
+15 -3
Просмотреть файл
@@ -94,16 +94,28 @@ rocprofiler_add_integration_execute_test(
LABELS "integration-tests"
PRELOAD "${ROCPROFILER_MEMCHECK_PRELOAD_ENV_VALUE}")
# Feature introduced in ROCm 7.1, however some pre-release CI builds for 7.1 dont have
# this feature in aqlprofile. Disable the next tests for ROCm < 7.2.
set(IS_DISABLED True)
find_package(hip CONFIG REQUIRED)
if(hip_FOUND)
if(${hip_VERSION_MAJOR} EQUAL 7 AND ${hip_VERSION_MINOR} GREATER 1)
set(IS_DISABLED False)
elseif(${hip_VERSION_MAJOR} GREATER 7)
set(IS_DISABLED False)
endif()
endif()
# Test large buffer sizes. 5120 == 5GB
rocprofiler_add_integration_execute_test(
thread-trace-api-large-buffer-test
COMMAND $<TARGET_FILE:thread-trace-api-agent-test>
DEPENDS thread-trace-api-agent-test
TIMEOUT 10
TIMEOUT 30
LABELS "integration-tests"
PRELOAD "${PRELOAD_ENV}"
ENVIRONMENT "ATT_BUFFER_SIZE_MB=5120"
DISABLED ${ROCPROFILER_DISABLE_UNSTABLE_CTESTS})
ENVIRONMENT "ATT_LARGE_BUFFER_TEST=1"
DISABLED ${IS_DISABLED})
# Test occupancy mode
rocprofiler_add_integration_execute_test(
+24 -6
Просмотреть файл
@@ -102,15 +102,33 @@ query_available_agents(rocprofiler_agent_version_t /* version */,
const auto* agent = static_cast<const rocprofiler_agent_v0_t*>(agents[idx]);
if(agent->type != ROCPROFILER_AGENT_TYPE_GPU) continue;
// Check if we are testing for large buffers
static const char* var = std::getenv("ATT_BUFFER_SIZE_MB");
static uint64_t buffer_size_mb = (var ? atoi(var) : 96) * 1024ul * 1024ul;
uint64_t buffer_size_gb = 1;
std::vector<rocprofiler_thread_trace_parameter_t> parameters;
// Are we testing for larger buffers?
if(const char* var = std::getenv("ATT_LARGE_BUFFER_TEST"); var && atoi(var))
{
// To fully test this feature, we need >4GB per shader engine (>8GB total).
// Some RDNA GPUs only have 8GB of VRAM, so we have to use 5GB total = 2.5GB per SE.
uint64_t total_memory = 0;
for(uint32_t i = 0; i < agent->mem_banks_count; i++)
total_memory += agent->mem_banks[i].size_in_bytes;
// Check we have >11GB VRAM. If so, allocate 10GB.
if(total_memory > (11ul << 30))
buffer_size_gb = 10;
else
buffer_size_gb = 5;
}
uint64_t buffer_size_bytes = buffer_size_gb << 30;
if(agent->gfx_target_version / 10000 == 11u)
buffer_size_bytes = 255ul << 20; // gfx11 limititation
auto parameters = std::vector<rocprofiler_thread_trace_parameter_t>{};
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_TARGET_CU, {1}});
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_SIMD_SELECT, {0xF}});
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_BUFFER_SIZE, {buffer_size_mb}});
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_SHADER_ENGINE_MASK, {0x1}});
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_BUFFER_SIZE, {buffer_size_bytes}});
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_SHADER_ENGINE_MASK, {0x3}});
static const bool extra_args =
std::getenv("ATT_NODETAIL") ? std::stoi(std::getenv("ATT_NODETAIL")) != 0 : false;
+5 -4
Просмотреть файл
@@ -71,8 +71,9 @@ tool_init(rocprofiler_client_finalize_t /* fini_func */, void* /* tool_data */)
nullptr),
"code object tracing service configure");
std::vector<rocprofiler_thread_trace_parameter_t> params{};
params.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_SERIALIZE_ALL, {1}});
auto parameters = std::vector<rocprofiler_thread_trace_parameter_t>{};
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_SERIALIZE_ALL, {1}});
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_SHADER_ENGINE_MASK, {0x3}});
std::vector<rocprofiler_agent_id_t> agents{};
@@ -97,8 +98,8 @@ tool_init(rocprofiler_client_finalize_t /* fini_func */, void* /* tool_data */)
ROCPROFILER_CALL(
rocprofiler_configure_dispatch_thread_trace_service(client_ctx,
id,
params.data(),
params.size(),
parameters.data(),
parameters.size(),
dispatch_callback,
Callbacks::shader_data_callback,
nullptr),
+5 -2
Просмотреть файл
@@ -84,13 +84,16 @@ tool_init(rocprofiler_client_finalize_t /* fini_func */, void* /* tool_data */)
&agents),
"Failed to iterate agents");
auto parameters = std::vector<rocprofiler_thread_trace_parameter_t>{};
parameters.push_back({ROCPROFILER_THREAD_TRACE_PARAMETER_SHADER_ENGINE_MASK, 0x3});
for(auto id : agents)
{
ROCPROFILER_CALL(
rocprofiler_configure_dispatch_thread_trace_service(client_ctx,
id,
nullptr,
0,
parameters.data(),
parameters.size(),
dispatch_callback,
Callbacks::shader_data_callback,
nullptr),