ROCm 7.0 RC3 cherry pick (#846)

* Fix memory clock detection with amd-smi (#824)

* bugfix to make amd-smi usage backward compatible (#836)

* Update soc_base.py

Fixes #835

Signed-off-by: fxmarty-amd <felmarty@amd.com>

* address comments

---------

Signed-off-by: fxmarty-amd <felmarty@amd.com>

* Standalone GUI bugfix (#825)

* Fix barchart elements table ids

* Add HBM bandwidth section to L2 cache report for gfx950

* bugfix for standlone GUI

Co-authored-by: Felix Marty <Felix.Marty@amd.com>

* Fix tests and formatting (#826)

* Fix L2 read/write/atomic bandwidths on MI350 (#831)

* Fix rocprofv3 supported counters not being detected (#832)

* Fix rocprofv3 supported counters not being detected

* Fix rocprof interface deprecation warning appearing twice

* Update VERSION due to cherry pick

* quick fix how to call v3 with pc sampling

* Fix pc sampling unit test (#847)

* 7.0.0 Changelog review feedback added

---------

Signed-off-by: fxmarty-amd <felmarty@amd.com>
Co-authored-by: fxmarty-amd <felmarty@amd.com>
Co-authored-by: Felix Marty <Felix.Marty@amd.com>
Co-authored-by: Fei Zheng <44449748+feizheng10@users.noreply.github.com>
Co-authored-by: prbasyal <prbasyal@amd.com>

Author: 4 people

CHANGELOG.md

@@ -2,7 +2,7 @@
 
 Full documentation for ROCm Compute Profiler is available at [https://rocm.docs.amd.com/projects/rocprofiler-compute/en/latest/](https://rocm.docs.amd.com/projects/rocprofiler-compute/en/latest/).
 
-## ROCm Compute Profiler 3.2.1 for ROCm 7.0.0
+## ROCm Compute Profiler 3.2.2 for ROCm 7.0.0
 
 ### Added
 
@@ -69,17 +69,8 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs.
   * ``-b`` option in profile mode also accepts hardware IP block for filtering; however, this filter support will be deprecated soon.
   * ``--list-metrics`` option added in profile mode to list possible metric id(s), similar to analyze mode.
 
-* Interface to ROCprofiler-SDK.
-  * Setting the environment variable ``ROCPROF=rocprofiler-sdk`` will use ROCprofiler-SDK C++ library instead of ``rocprofv3`` python script.
-  * Add --rocprofiler-sdk-library-path runtime option to choose the path to rocprofiler-sdk library to be used
-  * Using rocprof v1 / v2 / v3 interfaces will trigger a deprecation warning to use rocprofiler-sdk interface
-
 * Support MEM chart on CLI (single run)
 
-* Deprecation warning for MongoDB database update mode.
-
-* Deprecation warning for ``rocm-smi``
-
 * ``--specs-correction`` option to provide missing system specifications for analysis.
 
 ### Changed
@@ -102,6 +93,9 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs.
 * Fixed kernel name and kernel dispatch filtering when using ``rocprofv3``.
 * Fixed an issue of TCC channel counters collection in ``rocprofv3``.
 * Fixed peak FLOPS of F8, I8, F16, and BF16 on AMD Instinct MI 300.
+* Fixed not detecting memory clock issue when using amd-smi
+* Fixed standalone GUI crashing
+* Fixed L2 read/write/atomic bandwidths on MI350
 
 ### Known issues
 
@@ -127,12 +121,11 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs.
 
 ### Upcoming changes
 
-* ``rocprof v1/v2/v3`` interfaces will be removed in favor of the ROCprofiler-SDK interface, which directly accesses ``rocprofv3`` C++ tool.
-  * To use ROCprofiler-SDK interface, set environment variable `ROCPROF=rocprofiler-sdk` and optionally provide profile mode option ``--rocprofiler-sdk-library-path /path/to/librocprofiler-sdk.so``
+* ``rocprof v1/v2/v3`` interfaces will be removed in favor of the ROCprofiler-SDK interface, which directly accesses ``rocprofv3`` C++ tool. Using ``rocprof v1/v2/v3`` interfaces will trigger a deprecation warning.
+  * To use ROCprofiler-SDK interface, set environment variable `ROCPROF=rocprofiler-sdk` and optionally provide profile mode option ``--rocprofiler-sdk-library-path /path/to/librocprofiler-sdk.so``. Add ``--rocprofiler-sdk-library-path`` runtime option to choose the path to ROCprofiler-SDK library to be used.
 * Hardware IP block based filtering using ``-b`` option in profile mode will be removed in favor of analysis report block based filtering using ``-b`` option in profile mode.
-* Using rocprof v1 / v2 / v3 interfaces will trigger a deprecation warning to use rocprofiler-sdk interface
-* MongoDB database support will be removed.
-* Usage of ``rocm-smi`` will be removed in favor of ``amd-smi``.
+* MongoDB database support will be removed, and a deprecation warning has been added to the application interface.
+* Usage of ``rocm-smi`` is deprecated in favor of ``amd-smi``, and a deprecation warning has been added to the application interface.
 
 ## ROCm Compute Profiler 3.1.1 for ROCm 6.4.2
 

VERSION

@@ -1 +1 @@
-3.2.1
+3.2.2

src/rocprof_compute_analyze/analysis_webui.py

@@ -36,7 +36,7 @@
 from rocprof_compute_analyze.analysis_base import OmniAnalyze_Base
 from utils import file_io, parser
 from utils.gui import build_bar_chart, build_table_chart
-from utils.logger import console_debug, console_error, demarcate
+from utils.logger import console_debug, console_error, console_warning, demarcate
 
 
 class webui_analysis(OmniAnalyze_Base):
@@ -53,7 +53,9 @@ def __init__(self, args, supported_archs):
         # define different types of bar charts
         self.__barchart_elements = {
             "instr_mix": [1001, 1002],
-            "multi_bar": [1604, 1704],
+            # 1604: L1D - L2 Transactions
+            # 1705: L2 - Fabric Interface Stalls
+            "multi_bar": [1604, 1705],
             "sol": [1101, 1201, 1301, 1401, 1601, 1701],
             # "l2_cache_per_chan": [1802, 1803]
         }
@@ -371,7 +373,11 @@ def determine_chart_type(
 
     # Determine chart type:
     # a) Barchart
-    if table_config["id"] in [x for i in barchart_elements.values() for x in i]:
+    if original_df.empty:
+        console_warning(
+            f"The dataframe with id={table_config['id']} is empty! Not displaying it."
+        )
+    elif table_config["id"] in [x for i in barchart_elements.values() for x in i]:
         d_figs = build_bar_chart(display_df, table_config, barchart_elements, norm_filt)
         # Smaller formatting if barchart yeilds several graphs
         if (

src/rocprof_compute_profile/profiler_base.py

@@ -26,6 +26,7 @@
 import logging
 import os
 import re
+import shlex
 import shutil
 import time
 from abc import ABC, abstractmethod
@@ -453,7 +454,9 @@ def run_profiling(self, version: str, prog: str):
                 method=self.get_args().pc_sampling_method,
                 interval=self.get_args().pc_sampling_interval,
                 workload_dir=self.get_args().path,
-                appcmd=self.get_args().remaining,
+                appcmd=shlex.split(
+                    self.get_args().remaining
+                ),  # FIXME: the right solution is applying it when argparsing once!
                 rocprofiler_sdk_library_path=self.get_args().rocprofiler_sdk_library_path,
             )
             end_run_prof = time.time()

src/rocprof_compute_soc/analysis_configs/gfx950/1700_L2_cache.yaml

@@ -40,6 +40,10 @@ Panel Config:
               * 32)) / (End_Timestamp - Start_Timestamp)))
             unit: GB/s
             tips:
+          HBM Bandwidth:
+            value: $hbmBandwidth
+            unit: GB/s
+            tips:
 
     - metric_table:
         id: 1702
@@ -179,21 +183,21 @@ Panel Config:
             unit: (Bytes + $normUnit)
             tips:
           Read Bandwidth:
-            avg: AVG(TCC_READ_SECTORS_sum / $denom)
-            min: MIN(TCC_READ_SECTORS_sum / $denom)
-            max: MAX(TCC_READ_SECTORS_sum / $denom)
+            avg: AVG(TCC_READ_SECTORS_sum * 32 / $denom)
+            min: MIN(TCC_READ_SECTORS_sum * 32 / $denom)
+            max: MAX(TCC_READ_SECTORS_sum * 32 / $denom)
             unit: (Bytes + $normUnit)
             tips:
           Write Bandwidth:
-            avg: AVG(TCC_WRITE_SECTORS_sum / $denom)
-            min: MIN(TCC_WRITE_SECTORS_sum / $denom)
-            max: MAX(TCC_WRITE_SECTORS_sum / $denom)
+            avg: AVG(TCC_WRITE_SECTORS_sum * 32 / $denom)
+            min: MIN(TCC_WRITE_SECTORS_sum * 32 / $denom)
+            max: MAX(TCC_WRITE_SECTORS_sum * 32 / $denom)
             unit: (Bytes + $normUnit)
             tips:
           Atomic Bandwidth:
-            avg: AVG(TCC_ATOMIC_SECTORS_sum / $denom)
-            min: MIN(TCC_ATOMIC_SECTORS_sum / $denom)
-            max: MAX(TCC_ATOMIC_SECTORS_sum / $denom)
+            avg: AVG(TCC_ATOMIC_SECTORS_sum * 32 / $denom)
+            min: MIN(TCC_ATOMIC_SECTORS_sum * 32 / $denom)
+            max: MAX(TCC_ATOMIC_SECTORS_sum * 32 / $denom)
             unit: (Bytes + $normUnit)
             tips:
           Req:

src/rocprof_compute_soc/soc_base.py

@@ -24,6 +24,7 @@
 
 import ctypes
 import glob
+import json
 import math
 import os
 import re
@@ -166,13 +167,21 @@ def populate_mspec(self):
             )
         )
 
-        # we get the max mclk from rocm-smi --showmclkrange
-        # Regular expression to extract the max memory clock (third frequency level in MEM)
-        memory_clock_pattern = (
-            r"MEM:\s*[^:]*FREQUENCY_LEVELS:\s*(?:\d+: \d+ MHz\s*){2}(\d+)\s*MHz"
-        )
-        amd_smi_mclk = run(["amd-smi", "static"], exit_on_error=True)
-        self._mspec.max_mclk = search(memory_clock_pattern, amd_smi_mclk)
+        # Parse json from amd-smi static --clock
+        amd_smi_mclk = run(["amd-smi", "static", "--clock", "--json"], exit_on_error=True)
+        amd_smi_mclk = json.loads(amd_smi_mclk)
+
+        if isinstance(amd_smi_mclk, dict):
+            # The output of `amd-smi static --clock --json` is a dict with amd-smi>=26.0.0.
+            amd_smi_mclk = amd_smi_mclk["gpu_data"][0]["clock"]["mem"]["frequency_levels"]
+        else:
+            # For backward compatibility: the output of `amd-smi static --clock --json` used to be a list for amd-smi<26.0.0.
+            amd_smi_mclk = amd_smi_mclk[0]["clock"]["mem"]["frequency_levels"]
+
+        # Choose the highest level of memory clock frequency
+        amd_smi_mclk = amd_smi_mclk[sorted(amd_smi_mclk.keys())[-1]]
+        # 100 Mhz -> 100
+        self._mspec.max_mclk = amd_smi_mclk.split(" ")[0]
 
         console_debug("max mem clock is {}".format(self._mspec.max_mclk))
 
@@ -440,6 +449,16 @@ def parse_counters_text(self, text):
 
     def get_rocprof_supported_counters(self):
         rocprof_cmd = detect_rocprof(self.get_args())
+
+        if rocprof_cmd != "rocprofiler-sdk":
+            console_warning(
+                "rocprof v1 / v2 / v3 interfaces will be removed in favor of "
+                "rocprofiler-sdk interface in a future release. To use rocprofiler-sdk "
+                "interface, please set the environment variable ROCPROF to 'rocprofiler-sdk' "
+                "and optionally provide the path to librocprofiler-sdk.so library via the "
+                "--rocprofiler-sdk-library-path option."
+            )
+
         rocprof_counters = set()
 
         if str(rocprof_cmd).endswith("rocprof"):
@@ -489,7 +508,7 @@ def get_rocprof_supported_counters(self):
                     f"Failed to list rocprof supported counters using command: {command}"
                 )
             for line in output.splitlines():
-                if "Name:" in line:
+                if "counter_name" in line:
                     counters, _ = self.parse_counters_text(line.split(":")[1].strip())
                     rocprof_counters.update(counters)
             # Custom counter support for mi100 for rocprofv3

src/utils/gui.py

@@ -52,7 +52,7 @@ def multi_bar_chart(table_id, display_df):
         nested_bar = {"NC": {}, "UC": {}, "RW": {}, "CC": {}}
         for index, row in display_df.iterrows():
             nested_bar[row["Coherency"]][row["Xfer"]] = row["Avg"]
-    if table_id == 1704:
+    if table_id == 1705:  # L2 - Fabric Interface Stalls
         nested_bar = {"Read": {}, "Write": {}}
         for index, row in display_df.iterrows():
             nested_bar[row["Transaction"]][row["Type"]] = row["Avg"]
@@ -197,9 +197,7 @@ def build_bar_chart(display_df, table_config, barchart_elements, norm_filt):
 
     # Speed-of-light bar chart
     elif table_config["id"] in barchart_elements["sol"]:
-        display_df["Avg"] = [
-            x.astype(float) if x != "" else float(0) for x in display_df["Avg"]
-        ]
+        display_df["Avg"] = [float(x) if x != "" else float(0) for x in display_df["Avg"]]
         if table_config["id"] == 1701:
             # special layout for L2 Cache SOL
             d_figs.append(

src/utils/utils.py

@@ -245,14 +245,6 @@ def detect_rocprof(args):
         )
         return rocprof_cmd
 
-    console_warning(
-        "rocprof v1 / v2 / v3 interfaces will be deprecated in favor of "
-        "rocprofiler-sdk interface in a future release. To use rocprofiler-sdk "
-        "interface, please set the environment variable ROCPROF to 'rocprofiler-sdk' "
-        "and optionally provide the path to librocprofiler-sdk.so library via the "
-        "--rocprofiler-sdk-library-path option."
-    )
-
     # detect rocprof
     if not "ROCPROF" in os.environ.keys():
         # default rocprof
@@ -998,8 +990,9 @@ def pc_sampling_prof(
             "-o",
             "ps_file",  # todo: sync up with the name from source in 2100_.yaml
             "--",
-            appcmd,
         ]
+        options.extend(appcmd)
+
         success, output = capture_subprocess_output(
             [rocprof_cmd] + options, new_env=os.environ.copy(), profileMode=True
         )

tests/test_TCP_counters.py

@@ -144,7 +144,7 @@ def test_L1_cache_counters(
 
     # set up two apps: sequential and random access
     app_names = ["vseq", "vrand"]
-    options = ["-b", "TCP"]
+    options = ["-b", "16"]
 
     result = {}
     metrics = ["Read Req", "Write Req", "Cache Hit Rate"]

tests/test_analyze_commands.py

@@ -79,21 +79,6 @@ def test_list_metrics_gfx90a(binary_handler_analyze_rocprof_compute):
     test_utils.clean_output_dir(config["cleanup"], workload_dir)
 
 
-@pytest.mark.list_metrics
-def test_list_metrics_gfx906(binary_handler_analyze_rocprof_compute):
-    code = binary_handler_analyze_rocprof_compute(["analyze", "--list-metrics", "gfx906"])
-    assert code == 1
-
-    for dir in indirs:
-        workload_dir = test_utils.setup_workload_dir(dir)
-        code = binary_handler_analyze_rocprof_compute(
-            ["analyze", "--path", workload_dir, "--list-metrics", "gfx906"]
-        )
-        assert code == 0
-
-    test_utils.clean_output_dir(config["cleanup"], workload_dir)
-
-
 @pytest.mark.list_metrics
 def test_list_metrics_gfx908(binary_handler_analyze_rocprof_compute):
     code = binary_handler_analyze_rocprof_compute(["analyze", "--list-metrics", "gfx908"])