Enable VCN tracing in Perfetto output (#65)
Enable VCN activity tracing on different instances from the GPU metrics fetched using rsmi_dev_gpu_metrics_info_get in the ROCm-SMI. library. The tracing can be controlled with ROCPROFSYS_ROCM_SMI_METRICS by setting the value as vcn_activity, Currently this configuration takes the following values: busy, temp, power, mem_usage, vcn_activity. By default, all the 5 values will be enabled. Signed-off-by: Sajina P Kandy <Sajina.PuthalathKandy@amd.com> Co-authored-by: Sajina Kandy <sputhala-amd@amd.com>
This commit is contained in:
Sajina PK
GitHub
@@ -105,6 +105,7 @@ ROCPROFSYS_DEFINE_CATEGORY(category, rocm_smi_busy, ROCPROFSYS_CATEGORY_ROCM_SMI
|
||||
ROCPROFSYS_DEFINE_CATEGORY(category, rocm_smi_temp, ROCPROFSYS_CATEGORY_ROCM_SMI_TEMP, "device_temp", "Temperature of a GPU device")
|
||||
ROCPROFSYS_DEFINE_CATEGORY(category, rocm_smi_power, ROCPROFSYS_CATEGORY_ROCM_SMI_POWER, "device_power", "Power consumption of a GPU device")
|
||||
ROCPROFSYS_DEFINE_CATEGORY(category, rocm_smi_memory_usage, ROCPROFSYS_CATEGORY_ROCM_SMI_MEMORY_USAGE, "device_memory_usage", "Memory usage of a GPU device")
|
||||
ROCPROFSYS_DEFINE_CATEGORY(category, rocm_smi_vcn_activity, ROCPROFSYS_CATEGORY_ROCM_SMI_VCN_ACTIVITY, "device_vcn_activity", "VCN Activity of a GPU device")
|
||||
ROCPROFSYS_DEFINE_CATEGORY(category, rocm_rccl, ROCPROFSYS_CATEGORY_ROCM_RCCL, "rccl", "ROCm Communication Collectives Library (RCCL) regions")
|
||||
ROCPROFSYS_DEFINE_CATEGORY(category, pthread, ROCPROFSYS_CATEGORY_PTHREAD, "pthread", "POSIX threading functions")
|
||||
ROCPROFSYS_DEFINE_CATEGORY(category, kokkos, ROCPROFSYS_CATEGORY_KOKKOS, "kokkos", "KokkosTools regions")
|
||||
@@ -167,6 +168,7 @@ using name = perfetto_category<Tp...>;
|
||||
ROCPROFSYS_PERFETTO_CATEGORY(category::rocm_smi_temp), \
|
||||
ROCPROFSYS_PERFETTO_CATEGORY(category::rocm_smi_power), \
|
||||
ROCPROFSYS_PERFETTO_CATEGORY(category::rocm_smi_memory_usage), \
|
||||
ROCPROFSYS_PERFETTO_CATEGORY(category::rocm_smi_vcn_activity), \
|
||||
ROCPROFSYS_PERFETTO_CATEGORY(category::rocm_rccl), \
|
||||
ROCPROFSYS_PERFETTO_CATEGORY(category::pthread), \
|
||||
ROCPROFSYS_PERFETTO_CATEGORY(category::kokkos), \
|
||||
|
||||
@@ -82,6 +82,8 @@ struct backtrace_gpu_power
|
||||
{};
|
||||
struct backtrace_gpu_memory
|
||||
{};
|
||||
struct backtrace_gpu_vcn
|
||||
{};
|
||||
using sampling_wall_clock = data_tracker<double, backtrace_wall_clock>;
|
||||
using sampling_cpu_clock = data_tracker<double, backtrace_cpu_clock>;
|
||||
using sampling_percent = data_tracker<double, backtrace_fraction>;
|
||||
@@ -89,6 +91,7 @@ using sampling_gpu_busy = data_tracker<double, backtrace_gpu_busy>;
|
||||
using sampling_gpu_temp = data_tracker<double, backtrace_gpu_temp>;
|
||||
using sampling_gpu_power = data_tracker<double, backtrace_gpu_power>;
|
||||
using sampling_gpu_memory = data_tracker<double, backtrace_gpu_memory>;
|
||||
using sampling_gpu_vcn = data_tracker<double, backtrace_gpu_vcn>;
|
||||
|
||||
template <typename ApiT, typename StartFuncT = default_functor_t,
|
||||
typename StopFuncT = default_functor_t>
|
||||
@@ -121,6 +124,7 @@ ROCPROFSYS_DEFINE_CONCRETE_TRAIT(is_available, component::sampling_gpu_busy, fal
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(is_available, component::sampling_gpu_temp, false_type)
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(is_available, component::sampling_gpu_power, false_type)
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(is_available, component::sampling_gpu_memory, false_type)
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(is_available, component::sampling_gpu_vcn, false_type)
|
||||
#endif
|
||||
|
||||
TIMEMORY_SET_COMPONENT_API(rocprofsys::component::roctracer, project::rocprofsys,
|
||||
@@ -152,6 +156,9 @@ TIMEMORY_SET_COMPONENT_API(rocprofsys::component::sampling_gpu_temp, project::ro
|
||||
tpls::rocm, device::gpu, os::supports_linux,
|
||||
category::temperature, category::sampling,
|
||||
category::process_sampling)
|
||||
TIMEMORY_SET_COMPONENT_API(rocprofsys::component::sampling_gpu_vcn, project::rocprofsys,
|
||||
tpls::rocm, device::gpu, os::supports_linux,
|
||||
category::sampling, category::process_sampling)
|
||||
|
||||
TIMEMORY_METADATA_SPECIALIZATION(rocprofsys::component::roctracer, "roctracer",
|
||||
"High-precision ROCm API and kernel tracing", "")
|
||||
@@ -180,6 +187,10 @@ TIMEMORY_METADATA_SPECIALIZATION(rocprofsys::component::sampling_gpu_power,
|
||||
TIMEMORY_METADATA_SPECIALIZATION(rocprofsys::component::sampling_gpu_temp,
|
||||
"sampling_gpu_temp", "GPU Temperature via ROCm-SMI",
|
||||
"Derived from sampling")
|
||||
TIMEMORY_METADATA_SPECIALIZATION(rocprofsys::component::sampling_gpu_vcn,
|
||||
"sampling_gpu_vcn",
|
||||
"GPU VCN Utilization (% activity) via ROCm-SMI",
|
||||
"Derived from sampling")
|
||||
|
||||
// statistics type
|
||||
TIMEMORY_STATISTICS_TYPE(rocprofsys::component::sampling_wall_clock, double)
|
||||
@@ -188,6 +199,7 @@ TIMEMORY_STATISTICS_TYPE(rocprofsys::component::sampling_gpu_busy, double)
|
||||
TIMEMORY_STATISTICS_TYPE(rocprofsys::component::sampling_gpu_temp, double)
|
||||
TIMEMORY_STATISTICS_TYPE(rocprofsys::component::sampling_gpu_power, double)
|
||||
TIMEMORY_STATISTICS_TYPE(rocprofsys::component::sampling_gpu_memory, double)
|
||||
TIMEMORY_STATISTICS_TYPE(rocprofsys::component::sampling_gpu_vcn, double)
|
||||
TIMEMORY_STATISTICS_TYPE(rocprofsys::component::comm_data_tracker_t, float)
|
||||
|
||||
// enable timing units
|
||||
@@ -219,6 +231,7 @@ ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_sum, component::sampling_gpu_busy, false
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_sum, component::sampling_gpu_temp, false_type)
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_sum, component::sampling_gpu_power, false_type)
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_sum, component::sampling_gpu_memory, false_type)
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_sum, component::sampling_gpu_vcn, false_type)
|
||||
|
||||
// reporting categories (mean)
|
||||
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_mean, component::sampling_percent, false_type)
|
||||
|
||||
@@ -317,8 +317,8 @@ configure_settings(bool _init)
|
||||
|
||||
ROCPROFSYS_CONFIG_SETTING(
|
||||
bool, "ROCPROFSYS_USE_ROCM_SMI",
|
||||
"Enable sampling GPU power, temp, utilization, and memory usage", true, "backend",
|
||||
"rocm_smi", "rocm", "process_sampling");
|
||||
"Enable sampling GPU power, temp, utilization, vcn_activity and memory usage",
|
||||
true, "backend", "rocm_smi", "rocm", "process_sampling");
|
||||
|
||||
ROCPROFSYS_CONFIG_SETTING(bool, "ROCPROFSYS_USE_SAMPLING",
|
||||
"Enable statistical sampling of call-stack", false,
|
||||
@@ -626,10 +626,11 @@ configure_settings(bool _init)
|
||||
|
||||
rocprofiler_sdk::config_settings(_config);
|
||||
|
||||
ROCPROFSYS_CONFIG_SETTING(std::string, "ROCPROFSYS_ROCM_SMI_METRICS",
|
||||
"rocm-smi metrics to collect: busy, temp, power, mem_usage",
|
||||
"busy,temp,power,mem_usage", "backend", "rocm_smi", "rocm",
|
||||
"process_sampling", "advanced");
|
||||
ROCPROFSYS_CONFIG_SETTING(
|
||||
std::string, "ROCPROFSYS_ROCM_SMI_METRICS",
|
||||
"rocm-smi metrics to collect: busy, temp, power, vcn_activity, mem_usage",
|
||||
"busy,temp,power,vcn_activity,mem_usage", "backend", "rocm_smi", "rocm",
|
||||
"process_sampling", "advanced");
|
||||
|
||||
ROCPROFSYS_CONFIG_SETTING(size_t, "ROCPROFSYS_PERFETTO_SHMEM_SIZE_HINT_KB",
|
||||
"Hint for shared-memory buffer size in perfetto (in KB)",
|
||||
|
||||
Reference in New Issue
Block a user