[CI][BENCHMARK] Merge new metadata implementation + tweaks to benchmarks (#17617)

This PR partially merges changes introduced in
#17229. Specifically, it merges
changes related to:
- Introducing metadata in the result
- Tweaks to the benchmarks itself: benchmarks ran, parameters used to
run, what is enabled/disabled, etc
- Enables/Disables benchmarks we are interested/no longer interested in
  - Disables non-working benchmarks
  - Bumps commit hashes used to fetch each benchmark
- New command line arguments, such as `--build-jobs` and `--redownload`
(to re-fetch the benchmarks)

**Note:** I am relying on this PR having its commits squashed during
merge (which should be the default behavior for intel/llvm)

I did not try to clean up the commit history. Most of these changes are
by Piotr anyway, I wanted to make sure he was properly credited for the
changes.

Updates for current splitting effort of #17229:
#17545 (comment)

---------

Co-authored-by: Piotr Balcer <[email protected]>
Co-authored-by: Łukasz Stolarczuk <[email protected]>

Author: 3 people

devops/scripts/benchmarks/benches/base.py

@@ -1,16 +1,37 @@
-# Copyright (C) 2024 Intel Corporation
+# Copyright (C) 2024-2025 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
 
+from dataclasses import dataclass
 import os
 import shutil
 from pathlib import Path
-from .result import Result
+from utils.result import BenchmarkMetadata, BenchmarkTag, Result
 from options import options
 from utils.utils import download, run
-import urllib.request
-import tarfile
+
+benchmark_tags = [
+    BenchmarkTag("SYCL", "Benchmark uses SYCL runtime"),
+    BenchmarkTag("UR", "Benchmark uses Unified Runtime API"),
+    BenchmarkTag("L0", "Benchmark uses Level Zero API directly"),
+    BenchmarkTag("UMF", "Benchmark uses Unified Memory Framework directly"),
+    BenchmarkTag("micro", "Microbenchmark focusing on a specific functionality"),
+    BenchmarkTag("application", "Real application-based performance test"),
+    BenchmarkTag("proxy", "Benchmark that simulates real application use-cases"),
+    BenchmarkTag("submit", "Tests kernel submission performance"),
+    BenchmarkTag("math", "Tests math computation performance"),
+    BenchmarkTag("memory", "Tests memory transfer or bandwidth performance"),
+    BenchmarkTag("allocation", "Tests memory allocation performance"),
+    BenchmarkTag("graph", "Tests graph-based execution performance"),
+    BenchmarkTag("latency", "Measures operation latency"),
+    BenchmarkTag("throughput", "Measures operation throughput"),
+    BenchmarkTag("inference", "Tests ML/AI inference performance"),
+    BenchmarkTag("image", "Image processing benchmark"),
+    BenchmarkTag("simulation", "Physics or scientific simulation benchmark"),
+]
+
+benchmark_tags_dict = {tag.name: tag for tag in benchmark_tags}
 
 
 class Benchmark:
@@ -55,19 +76,25 @@ def create_data_path(self, name, skip_data_dir=False):
             data_path = os.path.join(self.directory, name)
         else:
             data_path = os.path.join(self.directory, "data", name)
-            if options.rebuild and Path(data_path).exists():
+            if options.redownload and Path(data_path).exists():
                 shutil.rmtree(data_path)
 
         Path(data_path).mkdir(parents=True, exist_ok=True)
 
         return data_path
 
-    def download(self, name, url, file, untar=False, unzip=False, skip_data_dir=False):
+    def download(
+        self,
+        name,
+        url,
+        file,
+        untar=False,
+        unzip=False,
+        skip_data_dir=False,
+        checksum="",
+    ):
         self.data_path = self.create_data_path(name, skip_data_dir)
-        return download(self.data_path, url, file, untar, unzip)
-
-    def name(self):
-        raise NotImplementedError()
+        return download(self.data_path, url, file, untar, unzip, checksum)
 
     def lower_is_better(self):
         return True
@@ -87,6 +114,30 @@ def stddev_threshold(self):
     def get_suite_name(self) -> str:
         return self.suite.name()
 
+    def name(self):
+        raise NotImplementedError()
+
+    def description(self):
+        return ""
+
+    def notes(self) -> str:
+        return None
+
+    def unstable(self) -> str:
+        return None
+
+    def get_tags(self) -> list[str]:
+        return []
+
+    def get_metadata(self) -> BenchmarkMetadata:
+        return BenchmarkMetadata(
+            type="benchmark",
+            description=self.description(),
+            notes=self.notes(),
+            unstable=self.unstable(),
+            tags=self.get_tags(),
+        )
+
 
 class Suite:
     def benchmarks(self) -> list[Benchmark]:
@@ -97,3 +148,6 @@ def name(self) -> str:
 
     def setup(self):
         return
+
+    def additionalMetadata(self) -> dict[str, BenchmarkMetadata]:
+        return {}