TheRock/rocm-systems/projects/rocprofiler-sdk/source/lib/python/rocpd/summary.py

#!/usr/bin/env python3
###############################################################################
# MIT License
#
# Copyright (c) 2025 Advanced Micro Devices, Inc.
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
# THE SOFTWARE.
###############################################################################

import argparse
import os
import math

from typing import Any, List, Tuple
from .importer import RocpdImportData, execute_statement
from .query import export_sqlite_query

__all__ = [
    "generate_all_summaries",
    "generate_summary_query",
    "generate_domain_query",
    "create_domain_query",
    "create_summary_queries",
    "create_summary_region_queries",
    "export_query",
    "add_args",
    "execute",
    "main",
]


def check_function_availability(connection, function_name):
    """
    Checks if a given function exists in the SQLite database.

    Args:
        connection (sqlite3 db connection): The SQLite database connection handler.
        function_name (str): The name of the function to check.

    Returns:
        bool: True if the function exists, False otherwise.
    """
    cursor = connection.cursor()

    try:
        # Try the modern approach first (SQLite 3.30.0+)
        cursor.execute(
            "SELECT EXISTS(SELECT 1 FROM pragma_function_list WHERE name=?)",
            (function_name,),
        )
        result = cursor.fetchone()[0]
        return bool(result)
    except Exception:
        # Fallback for older SQLite versions (Workaround for RHEL 8)
        # Try to execute a simple query using the function to see if it exists
        try:
            cursor.execute(f"SELECT {function_name}(1)")
            return True
        except Exception:
            return False


def get_temp_view_names(connection: RocpdImportData) -> List[str]:
    """Return the names of all temporary views in the SQLite connection."""
    return [
        v[0]
        for v in execute_statement(
            connection, "SELECT name FROM sqlite_temp_master WHERE type='view'"
        ).fetchall()
    ]


def get_temp_view_columns(connection: RocpdImportData, view_name: str) -> List[str]:
    """Return the column names of a given temporary view."""
    cursor = connection.cursor()
    cursor.execute(f"PRAGMA table_xinfo('{view_name}')")
    return [row[1] for row in cursor.fetchall()]


def export_query(
    connection: RocpdImportData,
    output_path,
    output_file,
    output_format,
    query_name,
    query,
) -> None:
    """Write the contents of a SQL query to an output format."""

    query_not_empty = f"""
        SELECT EXISTS (
            {query}
        )
    """

    # just return if the result is empty
    if not connection.execute(query_not_empty).fetchone()[0]:
        return

    # prepare the output filename
    if not output_file:
        output_filename = query_name
    else:
        output_filename = f"{output_file}_{query_name}"

    if output_format == "console":
        print(f"\n{query_name.upper()}:")

    # call query module to export.  query will append the extension
    export_path = os.path.join(output_path, output_filename)
    export_sqlite_query(
        connection, query, export_format=output_format, export_path=export_path
    )


def generate_summary_query(
    view_name: str,
    view_query="",
    name_column="name",
    by_rank=False,
) -> Tuple[str, str]:
    """Generate the SQL statement to create a summary query."""

    if by_rank:
        view_suffix = "_summary_by_rank"
        group_by_columns = "guid, {name_column}".format(name_column=name_column)
        aggregation_group_by = "T.guid, T.nid, T.{name_column}".format(
            name_column=name_column
        )
        total_duration_group_by = "guid"
        additional_select_columns = "AD.pid AS ProcessID, P.hostname AS Hostname,"
        additional_aggregated_columns = """
                    T.guid,
                    T.nid,
                    T.pid,"""
        join_condition = "T.guid = A.guid AND T.{name_column} = A.name".format(
            name_column=name_column
        )
        total_duration_join = "JOIN total_duration TD ON AD.guid = TD.guid JOIN processes P ON AD.pid = P.pid"
    else:
        view_suffix = "_summary"
        group_by_columns = name_column
        aggregation_group_by = "T.{name_column}".format(name_column=name_column)
        total_duration_group_by = ""
        additional_select_columns = ""
        additional_aggregated_columns = ""
        join_condition = "T.{name_column} = A.name".format(name_column=name_column)
        total_duration_join = "CROSS JOIN total_duration TD"

    full_view_name = f"{view_name}{view_suffix}"

    # Use a different name for the CTE to avoid circular reference when view_query references the view
    source_table_name = f"{view_name}_source" if view_query else view_name

    view_select = (
        f"""
            {source_table_name} AS (
                {view_query}
            ),
    """
        if view_query
        else ""
    )

    summary_query = f"""
        WITH
            {view_select}
            avg_data AS (
                SELECT
                    {group_by_columns.replace(name_column, f"{name_column} AS name")},
                    AVG(duration) AS avg_duration
                FROM {source_table_name}
                GROUP BY {group_by_columns}
            ),
            aggregated_data AS (
                SELECT{additional_aggregated_columns}
                    T.{name_column} as name,
                    COUNT(*) AS calls,
                    SUM(T.duration) AS total_duration,
                    A.avg_duration AS average_duration,
                    MIN(T.duration) AS min_duration,
                    MAX(T.duration) AS max_duration,
                    SQRT(SUM(CAST((T.duration - A.avg_duration) AS REAL) * CAST((T.duration - A.avg_duration) AS REAL)) / (COUNT(*) - 1)) AS std_dev_duration
                FROM {source_table_name} T
                JOIN avg_data A ON {join_condition}
                GROUP BY {aggregation_group_by}
            ),
            total_duration AS (
                SELECT
                    {f"{total_duration_group_by}," if total_duration_group_by else ""}
                    SUM(total_duration) AS grand_total_duration
                FROM
                    aggregated_data
                {f"GROUP BY {total_duration_group_by}" if total_duration_group_by else ""}
            )
        SELECT
            {additional_select_columns}
            AD.name AS Name,
            AD.calls AS Calls,
            AD.total_duration AS "DURATION (nsec)",
            AD.average_duration AS "AVERAGE (nsec)",
            (CAST(AD.total_duration AS REAL) / TD.grand_total_duration) * 100 AS "PERCENT (INC)",
            AD.min_duration AS "MIN (nsec)",
            AD.max_duration AS "MAX (nsec)",
            AD.std_dev_duration AS "STD_DEV"
        FROM
            aggregated_data AD
            {total_duration_join}
        ORDER BY
            {"AD.pid," if by_rank else ""} AD.total_duration DESC
    """

    return (full_view_name, summary_query)


def generate_domain_query(
    connection: RocpdImportData, summary_queries, by_rank=False
) -> Tuple[str, str]:
    """Generate the SQL statement for domain summary by doing union over all summary queries."""

    if by_rank:
        view_suffix = "_summary_by_rank"
        view_name = "domain_summary_by_rank"
        additional_group_columns = "ProcessID, Hostname,"
        additional_select_columns = "GD.ProcessID, GD.Hostname,"
        total_duration_group_by = "GROUP BY ProcessID"
        join_condition = "JOIN total_duration TD ON GD.ProcessID = TD.ProcessID"
        order_by = "ORDER BY GD.ProcessID"
    else:
        view_suffix = "_summary"
        view_name = "domain_summary"
        additional_group_columns = ""
        additional_select_columns = ""
        total_duration_group_by = ""
        join_condition = "CROSS JOIN total_duration TD"
        order_by = 'ORDER BY GD."DURATION (nsec)" DESC'

    summary_dictionary = {
        query_name: query
        for query_name, query in summary_queries.items()
        if query_name.endswith(view_suffix)
    }

    if len(summary_dictionary) < 1:
        return ()

    summary_selects = [
        f"{query_name} AS ({query}) ," for query_name, query in summary_dictionary.items()
    ]

    union_selects = [
        f" SELECT '{query_name.replace(view_suffix, '').upper()}' as domain, * FROM {query_name} "
        for query_name, query in summary_dictionary.items()
    ]

    domain_select = f"""
        WITH
            {f"".join(summary_selects)}
            all_domains AS (
               {f" UNION ALL ".join(union_selects)}
            ),
            grouped_domains AS (
                SELECT
                    domain,
                    {additional_group_columns}
                    SUM(calls) AS calls,
                    SUM("DURATION (nsec)") AS "DURATION (nsec)",
                    SUM("AVERAGE (nsec)") AS "AVERAGE (nsec)",
                    MIN("MIN (nsec)") AS "MIN (nsec)",
                    MAX("MAX (nsec)") AS "MAX (nsec)",
                    SUM("STD_DEV") AS "STD_DEV"
                FROM all_domains
                GROUP BY domain{", ProcessID" if by_rank else ""}
            ),
            total_duration AS (
                SELECT
                    {additional_group_columns}
                    SUM("DURATION (nsec)") AS grand_total_duration
                FROM grouped_domains
                {total_duration_group_by}
            )
        SELECT
            {additional_select_columns}
            GD.domain AS Name,
            GD.calls AS Calls,
            GD."DURATION (nsec)",
            GD."AVERAGE (nsec)",
            (CAST(GD."DURATION (nsec)" AS REAL) / TD.grand_total_duration) * 100 AS "PERCENT (INC)",
            GD."MIN (nsec)",
            GD."MAX (nsec)",
            GD."STD_DEV"
        FROM
            grouped_domains GD
            {join_condition}
        {order_by}
    """

    return (view_name, domain_select)


def _view_matches_category(view_name: str, category: str) -> bool:
    """Return True if view name matches the category."""
    v = view_name.lower()
    c = category.lower()
    plural_map = (c, f"{c}s")
    return v in plural_map or any(v.startswith(f"{itr}_") for itr in plural_map)


def create_summary_queries(
    connection: RocpdImportData,
    by_rank=False,
    only_view_categories=None,
    kernel_name_type=None,
):
    """Create summary queries for eligible temporary views in the database.

    When only_view_categories is set (e.g. ["kernel"]), only include views whose
    name matches one of the categories (e.g. "kernels" matches "kernel").
    """

    NAME_COLUMN_MAP = {
        "memory_allocations": "type",
        "scratch_memory": "operation",
    }

    KERNEL_NAME_TYPE_MAP = {
        "truncated": "truncated_kernel_name",
        "mangled": "kernel_name",
    }

    avoid_view_pattern = ("rocpd", "region", "counter", "pmc")
    required_columns = {"duration"}

    views = get_temp_view_names(connection)

    queries = {}

    for view_name in views:
        if any(pattern in view_name for pattern in avoid_view_pattern):
            continue

        if only_view_categories is not None:
            if not any(
                _view_matches_category(view_name, cat) for cat in only_view_categories
            ):
                continue

        columns = get_temp_view_columns(connection, view_name)
        if not required_columns.issubset(columns):
            continue

        # Determine the name column and view query to use
        name_column = NAME_COLUMN_MAP.get(view_name, "name")
        view_query = ""

        if view_name == "kernels" and kernel_name_type is not None:
            # Use display_name as fallback if kernel_name_type is not in the map
            display_name_column = KERNEL_NAME_TYPE_MAP.get(
                kernel_name_type, "display_name"
            )
            view_query = f"""
                SELECT K.*, COALESCE(KS.{display_name_column}, K.name) AS display_name
                FROM kernels K
                LEFT JOIN kernel_symbols KS
                    ON K.kernel_id = KS.id
                    AND K.guid = KS.guid
            """
            name_column = "display_name"

        # Create regular summary query
        summary_query_name, summary_query = generate_summary_query(
            view_name, view_query, name_column=name_column
        )
        queries[summary_query_name] = summary_query

        # Create per-rank summary query
        if by_rank:
            per_rank_query_name, summary_by_rank_query = generate_summary_query(
                view_name, view_query, name_column=name_column, by_rank=True
            )
            queries[per_rank_query_name] = summary_by_rank_query

    return queries


def create_summary_region_queries(
    connection: RocpdImportData,
    by_rank=False,
    region_categories=None,
):
    """Create summary and region queries"""

    query = "SELECT DISTINCT(category) FROM regions_and_samples"
    categories = execute_statement(connection, query).fetchall()

    category_prefixes = ["rocm_"]

    # Convert all categories and prefixes to lowercase for correct comparison
    categories = [(cat[0].lower(),) for cat in categories]
    category_prefixes = [prefix.lower() for prefix in category_prefixes]

    if region_categories is not None:
        region_categories = [cat.lower() for cat in region_categories]
    else:
        # Automatically retrieve region categories from the database
        region_categories = set()
        for cat in categories:
            category_name = cat[0]
            matching_prefix = next(
                (
                    prefix
                    for prefix in category_prefixes
                    if category_name.startswith(prefix)
                ),
                "",
            )
            first_part = category_name[len(matching_prefix) :].split("_")[0]
            region_categories.add(f"{matching_prefix}{first_part}")

    category_map = {
        cat: [c[0] for c in categories if c[0] == cat or c[0].startswith(cat + "_")]
        for cat in region_categories
        if "MARKER" not in cat.upper()
    }

    queries = {}

    for k, v in category_map.items():
        if len(v) > 0:
            conditions = [f"category LIKE '{c}'" for c in v]
            region_query = f"""
                SELECT *
                FROM regions_and_samples
                WHERE {" OR ".join(conditions)}
            """

            # Create regular summary query
            summary_query_name, summary_query = generate_summary_query(k, region_query)
            queries[summary_query_name] = summary_query

            # Create per-rank summary query
            if by_rank:
                per_rank_query_name, summary_by_rank_query = generate_summary_query(
                    k, region_query, by_rank=True
                )
                queries[per_rank_query_name] = summary_by_rank_query

    # Markers
    if "MARKER" not in region_categories:
        return queries

    markers_query_name = "markers"
    markers_query = """
        SELECT JSON_EXTRACT(extdata, '$.message') AS marker_name, *
        FROM regions_and_samples
        WHERE category LIKE 'MARKER_%'
    """

    # Create regular summary query
    summary_query_name, summary_query = generate_summary_query(
        markers_query_name, markers_query, name_column="marker_name"
    )
    queries[summary_query_name] = summary_query

    # Create per-rank summary query
    if by_rank:
        per_rank_query_name, summary_by_rank_query = generate_summary_query(
            markers_query_name, markers_query, name_column="marker_name", by_rank=True
        )
        queries[per_rank_query_name] = summary_by_rank_query

    return queries


def create_domain_query(connection: RocpdImportData, summary_queries, by_rank=False):
    """Create a domain summary query by aggregating all summary queries."""

    result = generate_domain_query(connection, summary_queries, by_rank=by_rank)
    if not result:
        return {}

    query_name, query = result

    return {query_name: query}


def generate_all_summaries(connection: RocpdImportData, **kwargs: Any) -> None:
    """Generate all summaries and export them to selected format."""

    domain_summary = kwargs.get("domain_summary", False)
    by_rank = kwargs.get("summary_by_rank", False)
    output_file = kwargs.get("output_file", "")
    output_path = kwargs.get("output_path", "./rocpd-output-data")
    region_categories = kwargs.get("region_categories", None)
    output_format = kwargs.get("format", "console")
    mangled_kernels = kwargs.get("mangled_kernels", False)
    truncate_kernels = kwargs.get("truncate_kernels", False)

    if mangled_kernels:
        kernel_name_type = "mangled"
    elif truncate_kernels:
        kernel_name_type = "truncated"
    else:
        kernel_name_type = None

    if not check_function_availability(connection, "sqrt"):
        connection.create_function(
            "sqrt",
            1,
            lambda x: (
                math.sqrt(x)
                if x is not None and isinstance(x, (int, float)) and x >= 0
                else None
            ),
        )

    summary_queries = {}

    # Create the summary queries.
    # When region_categories is specified (and not NONE), only include view-based
    # summaries that match those categories and only region summaries for those categories.
    # With --region-categories NONE we only skip region summaries and still include
    # all view-based summaries.
    is_none_categories = (
        region_categories is not None
        and len(region_categories) == 1
        and str(region_categories[0]).strip().upper() == "NONE"
    )
    if region_categories is None or is_none_categories:
        summary_queries.update(
            create_summary_queries(connection, by_rank, kernel_name_type=kernel_name_type)
        )
    else:
        summary_queries.update(
            create_summary_queries(
                connection,
                by_rank,
                only_view_categories=region_categories,
                kernel_name_type=kernel_name_type,
            )
        )
    summary_queries.update(
        create_summary_region_queries(
            connection, by_rank, region_categories=region_categories
        )
    )

    if domain_summary:
        summary_queries.update(create_domain_query(connection, summary_queries))
        # Create domain summary per rank only if both domain_summary and summary_by_rank are enabled
        if by_rank:
            summary_queries.update(
                create_domain_query(connection, summary_queries, by_rank=True)
            )

    # Export all summary queries
    for query_name, query in summary_queries.items():
        export_query(
            connection, output_path, output_file, output_format, query_name, query
        )


#
# Command-line interface functions
#
def add_args(parser):
    """Add arguments for summary."""

    io_options = parser.add_argument_group("I/O options")
    io_options.add_argument(
        "-f",
        "--format",
        help="Sets the format the summaries are output to (default: console)",
        choices=("console", "csv", "html", "json", "md", "pdf"),
        default="console",
        type=str,
        required=False,
    )

    summary_options = parser.add_argument_group("Summary options")
    summary_options.add_argument(
        "--domain-summary",
        action="store_true",
        default=False,
        help="Generate domain summary view",
    )
    summary_options.add_argument(
        "--summary-by-rank",
        action="store_true",
        default=False,
        help="Generate summary views by-rank (or Process ID)",
    )
    summary_options.add_argument(
        "--region-categories",
        nargs="+",
        default=None,
        help="Specify region categories to include in the summary (example: HIP, HSA, RCCL, ROCDECODE, ROCJPEG, MARKER). If not specified, categories will be automatically retrieved from the database.",
    )
    summary_options.add_argument(
        "--mangled-kernels",
        action="store_true",
        default=False,
        help="Display mangled kernel names (do not demangle)",
    )
    summary_options.add_argument(
        "--truncate-kernels",
        action="store_true",
        default=False,
        help="Display truncated kernel names (function name only, without template parameters)",
    )

    def process_args(input, args):
        valid_args = [
            "format",
            "domain_summary",
            "summary_by_rank",
            "region_categories",
            "mangled_kernels",
            "truncate_kernels",
        ]

        ret = {}
        for itr in valid_args:
            if hasattr(args, itr):
                val = getattr(args, itr)
                if val is not None:
                    ret[itr] = val
        return ret

    return process_args


def execute(input, **kwargs: Any) -> RocpdImportData:

    generate_all_summaries(input, **kwargs)

    return input


def main(argv=None) -> int:
    """Main entry point for command line execution."""
    from . import time_window
    from . import output_config

    parser = argparse.ArgumentParser(description="Generate summary views from rocPD data")
    required_params = parser.add_argument_group("Required options")

    required_params.add_argument(
        "-i",
        "--input",
        required=True,
        type=output_config.check_file_exists,
        nargs="+",
        help="Input path and filename to one or more database(s), separated by spaces",
    )

    process_outcfg_args = output_config.add_args(parser)
    process_summary_args = add_args(parser)
    process_time_window_args = time_window.add_args(parser)

    args = parser.parse_args(argv)

    input = RocpdImportData(
        args.input, automerge_limit=getattr(args, "automerge_limit", None)
    )

    summary_args = process_summary_args(input, args)
    io_args = process_outcfg_args(input, args)
    process_time_window_args(input, args)

    all_args = {**summary_args, **io_args}

    execute(
        input,
        **all_args,
    )


if __name__ == "__main__":
    main()