[Benchmarks] add MT MemcpyExecute benchmark

Author: igchor

scripts/benchmarks/benches/compute.py

@@ -20,7 +20,7 @@ def setup(self):
         if self.built:
             return
 
-        repo_path = git_clone(self.directory, "compute-benchmarks-repo", "https://github.com/intel/compute-benchmarks.git", "08c41bb8bc1762ad53c6194df6d36bfcceff4aa2")
+        repo_path = git_clone(self.directory, "compute-benchmarks-repo", "https://github.com/intel/compute-benchmarks.git", "aa6a3b2108bb86202b654ad28129156fa746d41d")
         build_path = create_build_path(self.directory, 'compute-benchmarks-build')
 
         configure_command = [
@@ -231,3 +231,26 @@ def bin_args(self) -> list[str]:
             "--numberOfElementsY=256",
             "--numberOfElementsZ=256",
         ]
+
+class MemcpyExecute(ComputeBenchmark):
+    def __init__(self, bench, numOpsPerThread, numThreads, allocSize, iterations):
+        self.numOpsPerThread = numOpsPerThread
+        self.numThreads = numThreads
+        self.allocSize = allocSize
+        self.iterations = iterations
+        super().__init__(bench, "multithread_benchmark_ur", "MemcpyExecute")
+
+    def name(self):
+        return f"multithread_benchmark_ur MemcpyExecute opsPerThread:{self.numOpsPerThread}, numThreads:{self.numThreads}, allocSize:{self.allocSize}"
+
+    def bin_args(self) -> list[str]:
+        return [
+            "--Ioq=1",
+            "--UseEvents=1",
+            "--MeasureCompletion=1",
+            "--UseQueuePerThread=1",
+            f"--AllocSize={self.allocSize}",
+            f"--NumThreads={self.numThreads}",
+            f"--NumOpsPerThread={self.numOpsPerThread}",
+            f"--iterations={self.iterations}"
+        ]

scripts/benchmarks/main.py

@@ -44,6 +44,12 @@ def main(directory, additional_env_vars, save_name, compare_names, filter):
         ExecImmediateCopyQueue(cb, 0, 1, 'Device', 'Device', 1024),
         ExecImmediateCopyQueue(cb, 1, 1, 'Device', 'Host', 1024),
         VectorSum(cb),
+        MemcpyExecute(cb, 400, 8, 1024, 100),
+        MemcpyExecute(cb, 400, 8, 102400, 10),
+        MemcpyExecute(cb, 500, 8, 102400, 10),
+        MemcpyExecute(cb, 400, 1, 1024, 1000),
+        MemcpyExecute(cb, 10, 16, 1024, 1000),
+        MemcpyExecute(cb, 10, 16, 102400, 100),
 
         # *** Velocity benchmarks
         Hashtable(vb),