Enable single pass counter collection (#833)

[ROCm/rocprofiler-compute commit: 6a77d241ed]
2025-08-06 10:35:05 -04:00
1 parent 5084aad00b
commit 34dd26fb07
@@ -27,12 +27,11 @@
 import argparse
 import os
 import re
-import shutil
 from pathlib import Path


 def print_avail_arch(avail_arch: list):
-    ret_str = "\t\tList all available metrics for analysis on specified arch:"
+    ret_str = "\t\t\tList all available metrics for analysis on specified arch:"
    for arch in avail_arch:
        ret_str += "\n\t\t\t   {}".format(arch)
    return ret_str
@@ -207,6 +206,19 @@ Examples:
        choices=[""] + list(supported_archs.keys()),  # ["gfx908", "gfx90a"],
        help=print_avail_arch(supported_archs.keys()),
    )
+
+    profile_group.add_argument(
+        "--list-sets",
+        action="store_true",
+        help="\t\t\tDisplay available metric sets and their descriptions",
+    )
+    profile_group.add_argument(
+        "--set",
+        default=None,
+        dest="set_selected",
+        help="\t\t\tProfile a set of metrics of topic of interest by collecting counters in a single pass.\n\t\t\tFor available sets, see --list-sets",
+    )
+
    profile_group.add_argument(
        "--config-dir",
        dest="config_dir",
@@ -27,7 +27,6 @@
 import argparse
 import importlib
 import os
-import shutil
 import socket
 import sys
 import time
@@ -55,6 +54,7 @@ from utils.utils import (
    get_submodules,
    get_version,
    get_version_display,
+    parse_sets_yaml,
    set_locale_encoding,
 )

@@ -243,6 +243,53 @@ class RocProfCompute:
        else:
            console_error("Unsupported arch")

+    @demarcate
+    def list_sets(self):
+        sets_info = parse_sets_yaml(self.__mspec.gpu_arch)
+
+        if not sets_info:
+            console_error("No sets configuration found.")
+
+        print("\nAvailable Sets:")
+        print("=" * 115)
+
+        # Print header
+        print(
+            f"{'Set Option':<35} {'Set Title':<35} {'Metric Name':<30} {'Metric ID':<10}"
+        )
+        print("-" * 115)
+
+        # Print data grouped by set
+        for set_option, set_data in sets_info.items():
+            title = set_data.get("title", set_option)
+            metrics = set_data.get("metric", [])
+
+            first_row = True
+            for metric in metrics:
+                if isinstance(metric, dict) and metric:
+                    metric_id = next(iter(metric.keys()))
+                    metric_name = next(iter(metric.values()))
+
+                    # Only show set info on first row of each set
+                    set_display = set_option if first_row else ""
+                    title_display = title if first_row else ""
+
+                    print(
+                        f"{set_display:<35} {title_display:<35} {metric_name:<30} {metric_id:<10}"
+                    )
+                    first_row = False
+            # Empty line between sets
+            print()
+
+        print("Usage Examples:")
+        if sets_info:
+            first_set = next(iter(sets_info.keys()))
+            print(f"  rocprof-compute profile --set {first_set}  # Profile this set")
+        print(f"  rocprof-compute profile --list-sets        # Show this help")
+        print()
+
+        sys.exit(0)
+
    @demarcate
    def run_profiler(self):
        self.print_graphic()
@@ -250,6 +297,8 @@ class RocProfCompute:

        if self.__args.list_metrics is not None:
            self.list_metrics()
+        elif self.__args.list_sets:
+            self.list_sets()
        elif self.__args.name is None:
            sys.exit("Either --list-name or --name is required")

@@ -26,13 +26,12 @@

 import csv
 import glob
-import logging
 import os
 import re
 import shlex
 import shutil
 import time
-from abc import ABC, abstractmethod
+from abc import abstractmethod
 from pathlib import Path

 import pandas as pd
@@ -350,6 +349,10 @@ class RocProfCompute_Base:
        console_log("Command: " + str(self.__args.remaining))
        console_log("Kernel Selection: " + str(self.__args.kernel))
        console_log("Dispatch Selection: " + str(self.__args.dispatch))
+
+        if self.get_args().set_selected:
+            console_log("Set Selection: " + str(self.__args.set_selected))
+
        if self.get_args().filter_blocks is None:
            console_log("Report Sections: All")
        else:
@@ -0,0 +1,37 @@
+# AUTOGENERATED FILE. Only edit for testing purposes, not for development. Generated from utils/unified_sets.yaml. Generated by utils/split_config.py
+sets:
+- title: Compute Throughput Utilization
+  set_option: compute_thruput_util
+  description: Placeholder
+  metric:
+  - 11.2.2: SALU Utilization
+  - 11.2.3: VALU Utilization
+- title: Compute Throughput FLOPS
+  set_option: compute_thruput_flops
+  description: Placeholder
+  metric:
+  - 2.1.2: MFMA FLOPs (BF16)
+  - 2.1.3: MFMA FLOPs (F16)
+  - 2.1.4: MFMA FLOPs (F32)
+  - 2.1.5: MFMA FLOPs (F64)
+  - 2.1.6: MFMA IOPs (Int8)
+- title: Memory Throughput
+  set_option: mem_thruput
+  description: Placeholder
+  metric:
+  - 2.1.16: Theoretical LDS Bandwidth
+  - 2.1.17: LDS Bank Conflicts/Access
+  - 16.1.2: Utilization
+  - 17.1.0: Utilization
+- title: Launch Stats
+  set_option: launch_stats
+  description: Placeholder
+  metric:
+  - 7.1.0: Grid Size
+  - 7.1.1: Workgroup Size
+  - 7.1.2: Total Wavefronts
+  - 7.1.5: VGPRs
+  - 7.1.6: AGPRs
+  - 7.1.7: SGPRs
+  - 7.1.8: LDS Allocation
+  - 7.1.9: Scratch Allocation
@@ -0,0 +1,39 @@
+# AUTOGENERATED FILE. Only edit for testing purposes, not for development. Generated from utils/unified_sets.yaml. Generated by utils/split_config.py
+sets:
+- title: Compute Throughput Utilization
+  set_option: compute_thruput_util
+  description: Placeholder
+  metric:
+  - 11.2.3: VALU Utilization
+  - 11.2.4: VMEM Utilization
+  - 11.2.5: Branch Utilization
+  - 11.2.6: VALU Active Threads
+- title: Compute Throughput FLOPS
+  set_option: compute_thruput_flops
+  description: Placeholder
+  metric:
+  - 2.1.2: MFMA FLOPs (BF16)
+  - 2.1.3: MFMA FLOPs (F16)
+  - 2.1.4: MFMA FLOPs (F32)
+  - 2.1.5: MFMA FLOPs (F64)
+  - 2.1.6: MFMA IOPs (Int8)
+- title: Memory Throughput
+  set_option: mem_thruput
+  description: Placeholder
+  metric:
+  - 2.1.16: Theoretical LDS Bandwidth
+  - 2.1.17: LDS Bank Conflicts/Access
+  - 16.1.2: Utilization
+  - 17.1.0: Utilization
+- title: Launch Stats
+  set_option: launch_stats
+  description: Placeholder
+  metric:
+  - 7.1.0: Grid Size
+  - 7.1.1: Workgroup Size
+  - 7.1.2: Total Wavefronts
+  - 7.1.5: VGPRs
+  - 7.1.6: AGPRs
+  - 7.1.7: SGPRs
+  - 7.1.8: LDS Allocation
+  - 7.1.9: Scratch Allocation
@@ -0,0 +1,40 @@
+# AUTOGENERATED FILE. Only edit for testing purposes, not for development. Generated from utils/unified_sets.yaml. Generated by utils/split_config.py
+sets:
+- title: Compute Throughput Utilization
+  set_option: compute_thruput_util
+  description: Placeholder
+  metric:
+  - 11.2.2: SALU Utilization
+  - 11.2.3: VALU Utilization
+  - 11.2.4: VMEM Utilization
+  - 11.2.5: Branch Utilization
+- title: Compute Throughput FLOPS
+  set_option: compute_thruput_flops
+  description: Placeholder
+  metric:
+  - 2.1.2: MFMA FLOPs (F8)
+  - 2.1.3: MFMA FLOPs (BF16)
+  - 2.1.4: MFMA FLOPs (F16)
+  - 2.1.5: MFMA FLOPs (F32)
+  - 2.1.6: MFMA FLOPs (F64)
+  - 2.1.7: MFMA IOPs (Int8)
+- title: Memory Throughput
+  set_option: mem_thruput
+  description: Placeholder
+  metric:
+  - 2.1.17: Theoretical LDS Bandwidth
+  - 2.1.18: LDS Bank Conflicts/Access
+  - 16.1.2: Utilization
+  - 17.1.0: Utilization
+- title: Launch Stats
+  set_option: launch_stats
+  description: Placeholder
+  metric:
+  - 7.1.0: Grid Size
+  - 7.1.1: Workgroup Size
+  - 7.1.2: Total Wavefronts
+  - 7.1.5: VGPRs
+  - 7.1.6: AGPRs
+  - 7.1.7: SGPRs
+  - 7.1.8: LDS Allocation
+  - 7.1.9: Scratch Allocation
@@ -0,0 +1,40 @@
+# AUTOGENERATED FILE. Only edit for testing purposes, not for development. Generated from utils/unified_sets.yaml. Generated by utils/split_config.py
+sets:
+- title: Compute Throughput Utilization
+  set_option: compute_thruput_util
+  description: Placeholder
+  metric:
+  - 11.2.2: SALU Utilization
+  - 11.2.3: VALU Utilization
+  - 11.2.4: VMEM Utilization
+  - 11.2.5: Branch Utilization
+- title: Compute Throughput FLOPS
+  set_option: compute_thruput_flops
+  description: Placeholder
+  metric:
+  - 2.1.2: MFMA FLOPs (F8)
+  - 2.1.3: MFMA FLOPs (BF16)
+  - 2.1.4: MFMA FLOPs (F16)
+  - 2.1.5: MFMA FLOPs (F32)
+  - 2.1.6: MFMA FLOPs (F64)
+  - 2.1.7: MFMA IOPs (Int8)
+- title: Memory Throughput
+  set_option: mem_thruput
+  description: Placeholder
+  metric:
+  - 2.1.17: Theoretical LDS Bandwidth
+  - 2.1.18: LDS Bank Conflicts/Access
+  - 16.1.2: Utilization
+  - 17.1.0: Utilization
+- title: Launch Stats
+  set_option: launch_stats
+  description: Placeholder
+  metric:
+  - 7.1.0: Grid Size
+  - 7.1.1: Workgroup Size
+  - 7.1.2: Total Wavefronts
+  - 7.1.5: VGPRs
+  - 7.1.6: AGPRs
+  - 7.1.7: SGPRs
+  - 7.1.8: LDS Allocation
+  - 7.1.9: Scratch Allocation
@@ -0,0 +1,40 @@
+# AUTOGENERATED FILE. Only edit for testing purposes, not for development. Generated from utils/unified_sets.yaml. Generated by utils/split_config.py
+sets:
+- title: Compute Throughput Utilization
+  set_option: compute_thruput_util
+  description: Placeholder
+  metric:
+  - 11.2.2: SALU Utilization
+  - 11.2.3: VALU Utilization
+  - 11.2.4: VMEM Utilization
+  - 11.2.5: Branch Utilization
+- title: Compute Throughput FLOPS
+  set_option: compute_thruput_flops
+  description: Placeholder
+  metric:
+  - 2.1.2: MFMA FLOPs (F8)
+  - 2.1.3: MFMA FLOPs (BF16)
+  - 2.1.4: MFMA FLOPs (F16)
+  - 2.1.5: MFMA FLOPs (F32)
+  - 2.1.6: MFMA FLOPs (F64)
+  - 2.1.7: MFMA IOPs (Int8)
+- title: Memory Throughput
+  set_option: mem_thruput
+  description: Placeholder
+  metric:
+  - 2.1.17: Theoretical LDS Bandwidth
+  - 2.1.18: LDS Bank Conflicts/Access
+  - 16.1.2: Utilization
+  - 17.1.0: Utilization
+- title: Launch Stats
+  set_option: launch_stats
+  description: Placeholder
+  metric:
+  - 7.1.0: Grid Size
+  - 7.1.1: Workgroup Size
+  - 7.1.2: Total Wavefronts
+  - 7.1.5: VGPRs
+  - 7.1.6: AGPRs
+  - 7.1.7: SGPRs
+  - 7.1.8: LDS Allocation
+  - 7.1.9: Scratch Allocation
@@ -0,0 +1,40 @@
+# AUTOGENERATED FILE. Only edit for testing purposes, not for development. Generated from utils/unified_sets.yaml. Generated by utils/split_config.py
+sets:
+- title: Compute Throughput Utilization
+  set_option: compute_thruput_util
+  description: Placeholder
+  metric:
+  - 11.2.2: SALU Utilization
+  - 11.2.3: VALU Utilization
+  - 11.2.5: VMEM Utilization
+  - 11.2.6: Branch Utilization
+- title: Compute Throughput FLOPS
+  set_option: compute_thruput_flops
+  description: Placeholder
+  metric:
+  - 2.1.2: MFMA FLOPs (F8)
+  - 2.1.3: MFMA FLOPs (BF16)
+  - 2.1.4: MFMA FLOPs (F16)
+  - 2.1.5: MFMA FLOPs (F32)
+  - 2.1.6: MFMA FLOPs (F64)
+  - 2.1.8: MFMA IOPs (Int8)
+- title: Memory Throughput
+  set_option: mem_thruput
+  description: Placeholder
+  metric:
+  - 2.1.18: Theoretical LDS Bandwidth
+  - 2.1.19: LDS Bank Conflicts/Access
+  - 16.1.2: Utilization
+  - 17.1.0: Utilization
+- title: Launch Stats
+  set_option: launch_stats
+  description: Placeholder
+  metric:
+  - 7.1.0: Grid Size
+  - 7.1.1: Workgroup Size
+  - 7.1.2: Total Wavefronts
+  - 7.1.5: VGPRs
+  - 7.1.6: AGPRs
+  - 7.1.7: SGPRs
+  - 7.1.8: LDS Allocation
+  - 7.1.9: Scratch Allocation
@@ -24,7 +24,6 @@
 ##############################################################################


-import ctypes
 import glob
 import json
 import math
@@ -32,7 +31,6 @@ import os
 import re
 import shutil
 import sys
-import threading
 from abc import abstractmethod
 from pathlib import Path

@@ -56,6 +54,7 @@ from utils.utils import (
    detect_rocprof,
    get_submodules,
    is_tcc_channel_counter,
+    parse_sets_yaml,
    using_v3,
 )

@@ -277,10 +276,26 @@ class OmniSoC_Base:
            Path(filename).name.split("_")[0]: filename
            for filename in glob.glob(f"{config_root_dir}/*.yaml")
        }
+
        texts = list()

+        set_selected = self.get_args().set_selected
+
+        if set_selected:
+            # NOTE: --blocks and --set are mutually exclusive
+            if self.get_args().filter_blocks:
+                console_error("--block and --set are exclusive options.")
+
+            sets_info = parse_sets_yaml(self.__arch)
+            if set_selected not in set(sets_info.keys()):
+                console_error(
+                    f"argument --set: invalid choice: '{set_selected}' (choose from {sets_info.keys()})"
+                )
+            self.__args.filter_blocks = [
+                next(iter(metric.keys())) for metric in sets_info[set_selected]["metric"]
+            ]
+
        if not self.get_args().filter_blocks:
-            # Read all config files if no filter_blocks are specified
            for filename in config_filename_dict.values():
                with open(filename, "r") as stream:
                    texts.append(stream.read())
@@ -35,11 +35,8 @@ import re
 import selectors
 import shutil
 import subprocess
-import sys
 import tempfile
 import time
-from collections import OrderedDict
-from itertools import product
 from pathlib import Path as path
 from typing import Optional

@@ -1616,3 +1613,25 @@ def format_time(seconds):
    if secs > 0 or not parts:
        parts.append(f"{secs} second{'s' if secs != 1 else ''}")
    return ", ".join(parts[:-1]) + (" and " if len(parts) > 1 else "") + parts[-1]
+
+
+def parse_sets_yaml(arch):
+    filename = (
+        config.rocprof_compute_home
+        / "rocprof_compute_soc"
+        / "profile_configs"
+        / "sets"
+        / f"{arch}_sets.yaml"
+    )
+    with open(filename, "r") as file:
+        content = file.read()
+    data = yaml.safe_load(content)
+
+    sets_data = data.get("sets", [])
+
+    sets_info = {}
+    for set_item in sets_data:
+        set_option = set_item.get("set_option", "")
+        if set_option:
+            sets_info[set_option] = set_item
+    return sets_info