TheRock/rocm-systems/projects/rocprofiler-systems/tests/rocpd-validation-rules/transpose/hw-counter-rules.json

{
    "required_tables": [
        {
            "commit": "Validation rules for GPU hardware counters (rocm_counter_collection)",
            "min_rows": 1,
            "name_prefix": "rocpd_info_pmc_",
            "required_columns": [
                "id",
                "agent_id",
                "target_arch",
                "name",
                "description",
                "units",
                "value_type"
            ],
            "validation_queries": [
                {
                    "comparison": "greater_than",
                    "description": "Check that GPU HW counter metadata was registered",
                    "error_message": "No GPU HW counter info entries found",
                    "expected_result": 0,
                    "query": "SELECT COUNT(*) FROM {table_name} WHERE target_arch = 'GPU' AND name LIKE 'GPU %'"
                }
            ]
        },
        {
            "commit": "Validation rules for GPU hardware counter events",
            "min_rows": 1,
            "name_prefix": "rocpd_pmc_event_",
            "required_columns": [
                "event_id",
                "pmc_id",
                "value"
            ],
            "validation_queries": [
                {
                    "comparison": "greater_than",
                    "description": "Check that GPU HW counter events exist in rocpd_pmc_event",
                    "error_message": "No GPU HW counter event rows found",
                    "expected_result": 0,
                    "query": "SELECT COUNT(*) FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name LIKE 'GPU %'"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check that GPU HW counter end-of-event zero markers exist (each counter interval must close with value=0)",
                    "error_message": "No zero-value end markers found for GPU HW counters \u2014 counter intervals are not properly closed",
                    "expected_result": 0,
                    "query": "SELECT COUNT(*) FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name LIKE 'GPU %' AND event.value = 0.0"
                },
                {
                    "comparison": "greater_than",
                    "description": "Check that GPU HW counter non-zero values exist (actual counter measurements)",
                    "error_message": "No non-zero counter values found \u2014 counters may not be recording",
                    "expected_result": 0,
                    "query": "SELECT COUNT(*) FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name LIKE 'GPU %' AND event.value > 0.0"
                },
                {
                    "comparison": "equals",
                    "description": "Check that every non-zero counter event has a matching zero end marker (paired start/end)",
                    "error_message": "Mismatch between non-zero counter events and zero end markers \u2014 some counter intervals are not closed",
                    "expected_result": 0,
                    "query": "SELECT (SELECT COUNT(*) FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name LIKE 'GPU %' AND event.value > 0.0) - (SELECT COUNT(*) FROM {table_name} event JOIN rocpd_info_pmc info ON event.pmc_id = info.id WHERE info.name LIKE 'GPU %' AND event.value = 0.0) as count"
                }
            ]
        }
    ]
}