[Bench] Fix torch benchmarks' passing profiler argument

- Fix missing param passing to torch benchmark bin (ref. intel#20978)
- Extend test case to check if the param is passed, as expected
- Move log with full command to execute from Debug to Info

Author: lukaszstolarczuk

devops/scripts/benchmarks/benches/compute.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2024-2025 Intel Corporation
+# Copyright (C) 2024-2026 Intel Corporation
 # Part of the Unified-Runtime Project, under the Apache License v2.0 with LLVM Exceptions.
 # See LICENSE.TXT
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
@@ -1027,9 +1027,11 @@ def _supported_runtimes(self) -> list[RUNTIMES]:
 
     def _bin_args(self, run_trace: TracingType = TracingType.NONE) -> list[str]:
         iters = self._get_iters(run_trace)
-        return [f"--iterations={iters}"] + [
-            f"--{k}={v}" for k, v in self._torch_params.items()
-        ]
+        return (
+            [f"--iterations={iters}"]
+            + [f"--profilerType={self._profiler_type.value}"]
+            + [f"--{k}={v}" for k, v in self._torch_params.items()]
+        )
 
 
 class TorchSingleQueue(TorchBenchmark):

devops/scripts/benchmarks/main.py

@@ -107,8 +107,8 @@ def run_iterations(
     Unless options.exit_on_failure is set, then exception is raised.
     """
 
+    log.info(f"Running '{benchmark.name()}' {iters}x iterations...")
     for iter in range(iters):
-        log.info(f"running {benchmark.name()}, iteration {iter}... ")
         try:
             bench_results = benchmark.run(
                 env_vars, run_trace=run_trace, force_trace=force_trace
@@ -145,7 +145,7 @@ def run_iterations(
                 log.error(f"{failure_label}: verification failed: {str(e)}.")
                 continue
 
-    # Iterations completed successfully
+    log.info(f"Completed '{benchmark.name()}' {iters}x iterations")
     return True
 
 

devops/scripts/benchmarks/tests/test_integration.py

@@ -1,4 +1,4 @@
-# Copyright (C) 2025 Intel Corporation
+# Copyright (C) 2025-2026 Intel Corporation
 # Part of the Unified-Runtime Project, under the Apache License v2.0 with LLVM Exceptions.
 # See LICENSE.TXT
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
@@ -63,7 +63,7 @@ def remove_dirs(self):
             if d is not None:
                 shutil.rmtree(d, ignore_errors=True)
 
-    def run_main(self, *args):
+    def run_main(self, *args) -> subprocess.CompletedProcess:
 
         # TODO: not yet tested: "--detect-version", "sycl,compute_runtime"
 
@@ -101,9 +101,9 @@ def run_main(self, *args):
             "MAIN_PY_STDERR:",
             "\n" + proc.stderr.decode() if proc.stderr else " <empty>",
         )
-        return proc.returncode
+        return proc
 
-    def get_output(self):
+    def get_benchmark_output_data(self):
         with open(os.path.join(self.OUTPUT_DIR, "data.json")) as f:
             out = json.load(f)
             return DataJson(
@@ -169,11 +169,30 @@ def _checkGroup(
     def _checkResultsExist(self, caseName: str, out: DataJson):
         self.assertIn(caseName, [r.name for r in out.runs[0].results])
 
-    def _checkCase(self, caseName: str, groupName: str, tags: set[str]):
-        run_result = self.app.run_main("--filter", caseName + "$")
-        self.assertEqual(run_result, 0, "Subprocess did not exit cleanly")
+    def _checkExistsInProcessOutput(
+        self, proc: subprocess.CompletedProcess, expected: str
+    ):
+        """
+        Check that expected regex string exists in process output.
+        It's useful for checking e.g. if expected params are passed to the benchmark's bin execution.
+        """
+        stdout = proc.stdout.decode()
+        self.assertRegex(stdout, expected, "Expected string not found in output")
+
+    def _checkCase(
+        self,
+        caseName: str,
+        groupName: str,
+        tags: set[str],
+        expected_in_output: str = None,
+    ):
+        return_proc = self.app.run_main("--filter", caseName + "$")
+        self.assertEqual(return_proc.returncode, 0, "Subprocess did not exit cleanly")
+
+        if expected_in_output:
+            self._checkExistsInProcessOutput(return_proc, expected_in_output)
 
-        out = self.app.get_output()
+        out = self.app.get_benchmark_output_data()
         self._checkResultsExist(caseName, out)
 
         metadata = out.metadata[caseName]
@@ -199,11 +218,13 @@ def test_torch_l0(self):
             "torch_benchmark_l0 KernelSubmitSingleQueue KernelBatchSize 512, KernelDataType Int32, KernelName Add, KernelParamsNum 5, KernelSubmitPattern Single, KernelWGCount 4096, KernelWGSize 512",
             "KernelSubmitSingleQueue Int32Large",
             {"pytorch", "L0"},
+            "--test=KernelSubmitSingleQueue.*--profilerType=timer",
         )
         self._checkCase(
             "torch_benchmark_l0 KernelSubmitSingleQueue KernelBatchSize 512, KernelDataType Int32, KernelName Add, KernelParamsNum 5, KernelSubmitPattern Single, KernelWGCount 4096, KernelWGSize 512 CPU count",
             "KernelSubmitSingleQueue Int32Large, CPU count",
             {"pytorch", "L0"},
+            "--test=KernelSubmitSingleQueue.*--profilerType=cpuCounter",
         )
         self._checkCase(
             "torch_benchmark_l0 KernelSubmitMultiQueue kernelsPerQueue 20, workgroupCount 4096, workgroupSize 512",

devops/scripts/benchmarks/utils/utils.py

@@ -71,7 +71,7 @@ def run(
         command_str = " ".join(command)
         env_str = " ".join(f"{key}={value}" for key, value in env_vars.items())
         full_command_str = f"{env_str} {command_str}".strip()
-        log.debug(f"Running: {full_command_str}")
+        log.info(f"Running: {full_command_str}")
 
         for key, value in env_vars.items():
             # Only PATH and LD_LIBRARY_PATH should be prepended to existing values