replace rocm-smi with amd-smi cmd (#612)

[ROCm/rocprofiler-compute commit: 0c6cec5671]

projects/rocprofiler-compute/CHANGELOG.md

@@ -16,6 +16,7 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs.
 ### Changed
 
 * Change normal_unit default to per_kernel
+* change dependency from rocm-smi to amd-smi
 
 ### Resolved issues
 

projects/rocprofiler-compute/src/rocprof_compute_soc/soc_base.py

@@ -164,8 +164,14 @@ class OmniSoC_Base:
         )
 
         # we get the max mclk from rocm-smi --showmclkrange
-        rocm_smi_mclk = run(["rocm-smi", "--showmclkrange"], exit_on_error=True)
-        self._mspec.max_mclk = search(r"(\d+)Mhz\s*$", rocm_smi_mclk)
+        # 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)
+
+        console_debug("max mem clock is {}".format(self._mspec.max_mclk))
 
         # these are just max's now, because the parsing was broken and this was inconsistent
         # with how we use the clocks elsewhere (all max, all the time)

projects/rocprofiler-compute/src/utils/specs.py

@@ -39,6 +39,7 @@ import pandas as pd
 import config
 from utils.tty import get_table_string
 from utils.utils import (
+    console_debug,
     console_error,
     console_log,
     console_warning,
@@ -135,18 +136,25 @@ def generate_machine_specs(args, sysinfo: dict = None):
         linux_distro = ""
     rocm_version = get_rocm_ver().strip()
     # FIXME: use device
-    vbios = search(r"VBIOS version: (.*?)$", run(["rocm-smi", "-v"], exit_on_error=True))
-    compute_partition = search(
-        r"Compute Partition:\s*(\w+)", run(["rocm-smi", "--showcomputepartition"])
-    )
+
+    vbios_pattern = r"PART_NUMBER:\s*(\S+)"
+    compute_partition_pattern = r"COMPUTE_PARTITION:\s*(\S+)"
+    memory_partition_pattern = r"MEMORY_PARTITION:\s*(\S+)"
+
+    vbios = search(vbios_pattern, run(["amd-smi", "static"], exit_on_error=True))
+    compute_partition = search(compute_partition_pattern, run(["amd-smi", "static"]))
     if compute_partition is None:
         compute_partition = "NA"
-    memory_partition = search(
-        r"Memory Partition:\s*(\w+)", run(["rocm-smi", "--showmemorypartition"])
-    )
+    memory_partition = search(memory_partition_pattern, run(["amd-smi", "static"]))
     if memory_partition is None:
         memory_partition = "NA"
 
+    console_debug(
+        "vbios is {}, compute partition is {}, memory partition is {}".format(
+            vbios, compute_partition, memory_partition
+        )
+    )
+
     ##########################################
     ## B. SoC Specs
     ##########################################
@@ -628,9 +636,9 @@ def run(cmd, exit_on_error=False):
         )
 
     if exit_on_error:
-        if cmd[0] == "rocm-smi":
+        if cmd[0] == "amd-smi":
             if p.returncode != 2 and p.returncode != 0:
-                console_error("No GPU detected. Unable to load rocm-smi")
+                console_error("No GPU detected. Unable to load amd-smi")
         elif p.returncode != 0:
             console_error("Command [%s] failed with non-zero exit code" % cmd)
     return p.stdout.decode("utf-8")

projects/rocprofiler-compute/tests/test_profile_general.py

@@ -263,8 +263,8 @@ def counter_compare(test_name, errors_pd, baseline_df, run_df, threshold=5):
 
 def run(cmd):
     p = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
-    if cmd[0] == "rocm-smi" and p.returncode == 8:
-        print("ERROR: No GPU detected. Unable to load rocm-smi")
+    if cmd[0] == "amd-smi" and p.returncode == 8:
+        print("ERROR: No GPU detected. Unable to load amd-smi")
         assert 0
     return p.stdout.decode("ascii")