JPEG Activity tracing in Perfetto (#108)

- Add JPEG activity track in perfetto trace
- Add JPEG decode tests to the examples
- Change existing videodecode test to include JPEG testing
- Rename videodecode test file to decode to include jpeg tests too
- Fix a bug in the test which checks for total activity of 0
- Disable rocDecode and rocJPEG samples from the github image files
This commit is contained in:
Sajina PK
2025-02-21 10:25:01 -05:00
committed by GitHub
parent c1d8799f92
commit 59d3399901
21 changed files with 1637 additions and 72 deletions
+2
View File
@@ -106,6 +106,7 @@ ROCPROFSYS_DEFINE_CATEGORY(category, rocm_smi_temp, ROCPROFSYS_CATEGORY_ROCM_SMI
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_smi_jpeg_activity, ROCPROFSYS_CATEGORY_ROCM_SMI_JPEG_ACTIVITY, "device_jpeg_activity", "JPEG Activity of a GPU device")
ROCPROFSYS_DEFINE_CATEGORY(category, rocm_rocdecode_api, ROCPROFSYS_CATEGORY_ROCM_ROCDECODE_API, "rocm_rocdecode_api", "ROCm ROCDecode API")
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")
@@ -171,6 +172,7 @@ using name = perfetto_category<Tp...>;
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_smi_jpeg_activity), \
ROCPROFSYS_PERFETTO_CATEGORY(category::rocm_rocdecode_api), \
ROCPROFSYS_PERFETTO_CATEGORY(category::rocm_rccl), \
ROCPROFSYS_PERFETTO_CATEGORY(category::pthread), \
+13
View File
@@ -82,6 +82,8 @@ struct backtrace_gpu_memory
{};
struct backtrace_gpu_vcn
{};
struct backtrace_gpu_jpeg
{};
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>;
@@ -90,6 +92,7 @@ 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>;
using sampling_gpu_jpeg = data_tracker<double, backtrace_gpu_jpeg>;
template <typename ApiT, typename StartFuncT = default_functor_t,
typename StopFuncT = default_functor_t>
@@ -123,6 +126,7 @@ ROCPROFSYS_DEFINE_CONCRETE_TRAIT(is_available, component::sampling_gpu_temp, fal
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)
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(is_available, component::sampling_gpu_jpeg, false_type)
#endif
TIMEMORY_SET_COMPONENT_API(rocprofsys::component::sampling_wall_clock,
@@ -151,6 +155,9 @@ TIMEMORY_SET_COMPONENT_API(rocprofsys::component::sampling_gpu_temp, project::ro
TIMEMORY_SET_COMPONENT_API(rocprofsys::component::sampling_gpu_vcn, project::rocprofsys,
tpls::rocm, device::gpu, os::supports_linux,
category::sampling, category::process_sampling)
TIMEMORY_SET_COMPONENT_API(rocprofsys::component::sampling_gpu_jpeg, project::rocprofsys,
tpls::rocm, device::gpu, os::supports_linux,
category::sampling, category::process_sampling)
TIMEMORY_METADATA_SPECIALIZATION(rocprofsys::component::sampling_wall_clock,
"sampling_wall_clock", "Wall-clock timing",
@@ -179,6 +186,10 @@ TIMEMORY_METADATA_SPECIALIZATION(rocprofsys::component::sampling_gpu_vcn,
"sampling_gpu_vcn",
"GPU VCN Utilization (% activity) via ROCm-SMI",
"Derived from sampling")
TIMEMORY_METADATA_SPECIALIZATION(rocprofsys::component::sampling_gpu_jpeg,
"sampling_gpu_jpeg",
"GPU JPEG 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_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::sampling_gpu_jpeg, double)
TIMEMORY_STATISTICS_TYPE(rocprofsys::component::comm_data_tracker_t, float)
// enable timing units
@@ -220,6 +232,7 @@ ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_sum, component::sampling_gpu_temp, false
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)
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_sum, component::sampling_gpu_jpeg, false_type)
// reporting categories (mean)
ROCPROFSYS_DEFINE_CONCRETE_TRAIT(report_mean, component::sampling_percent, false_type)
+9 -9
View File
@@ -315,10 +315,10 @@ configure_settings(bool _init)
"Enable ROCm API and kernel tracing", true, "backend",
"rocm");
ROCPROFSYS_CONFIG_SETTING(
bool, "ROCPROFSYS_USE_ROCM_SMI",
"Enable sampling GPU power, temp, utilization, vcn_activity and memory usage",
true, "backend", "rocm_smi", "rocm", "process_sampling");
ROCPROFSYS_CONFIG_SETTING(bool, "ROCPROFSYS_USE_ROCM_SMI",
"Enable sampling GPU power, temp, utilization, "
"vcn_activity, jpeg_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,
@@ -626,11 +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, vcn_activity, mem_usage",
"busy,temp,power,vcn_activity,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, jpeg_activity, mem_usage",
"busy,temp,power,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)",