Merge pull request #2456 from EuphoricThinking/benchmark_job6_final_pr

[benchmarks] add umf suite

.github/workflows/benchmarks-reusable.yml

@@ -156,6 +156,26 @@ jobs:
     - name: Install UR
       run: cmake --install ${{github.workspace}}/ur_build
 
+    - name: Checkout UMF
+      uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
+      with:
+        repository: oneapi-src/unified-memory-framework
+        ref: main
+        path: umf-repo
+        fetch-depth: 1
+        fetch-tags: false
+
+    - name: Configure UMF
+      run: >
+        cmake -DCMAKE_BUILD_TYPE=Release
+        -S${{github.workspace}}/umf-repo
+        -B${{github.workspace}}/umf_build
+        -DUMF_BUILD_BENCHMARKS=ON
+        -DUMF_TESTS_FAIL_ON_SKIP=ON
+
+    - name: Build UMF
+      run: cmake --build ${{github.workspace}}/umf_build -j $(nproc)
+
     - name: Run benchmarks
       working-directory: ${{ github.workspace }}/ur-repo/
       id: benchmarks
@@ -164,6 +184,7 @@ jobs:
         ~/bench_workdir
         --sycl ${{ github.workspace }}/sycl_build
         --ur ${{ github.workspace }}/ur_install
+        --umf ${{ github.workspace }}/umf_build
         --adapter ${{ matrix.adapter.str_name }}
         ${{ inputs.upload_report && '--output-html' || '' }}
         ${{ inputs.bench_script_params }}

scripts/benchmarks/benches/base.py

@@ -26,7 +26,7 @@ def get_adapter_full_path():
         assert False, \
             f"could not find adapter file {adapter_path} (and in similar lib paths)"
 
-    def run_bench(self, command, env_vars, ld_library=[]):
+    def run_bench(self, command, env_vars, ld_library=[], add_sycl=True):
         env_vars_with_forced_adapter = env_vars.copy()
         if options.ur is not None:
             env_vars_with_forced_adapter.update(
@@ -35,7 +35,7 @@ def run_bench(self, command, env_vars, ld_library=[]):
         return run(
             command=command,
             env_vars=env_vars_with_forced_adapter,
-            add_sycl=True,
+            add_sycl=add_sycl,
             cwd=options.benchmark_cwd,
             ld_library=ld_library
         ).stdout.decode()
@@ -71,6 +71,9 @@ def run(self, env_vars) -> list[Result]:
     def teardown(self):
         raise NotImplementedError()
 
+    def stddev_threshold(self):
+        return None
+
 class Suite:
     def benchmarks(self) -> list[Benchmark]:
         raise NotImplementedError()

scripts/benchmarks/benches/options.py

@@ -12,6 +12,7 @@ class Options:
     sycl: str = None
     ur: str = None
     ur_adapter: str = None
+    umf: str = None
     rebuild: bool = True
     benchmark_cwd: str = "INVALID"
     timeout: float = 600

scripts/benchmarks/benches/umf.py

@@ -0,0 +1,158 @@
+# Copyright (C) 2024 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
+
+import random
+from utils.utils import git_clone
+from .base import Benchmark, Suite
+from .result import Result
+from utils.utils import run, create_build_path
+from .options import options
+from .oneapi import get_oneapi
+import os
+import csv
+import io
+
+def isUMFAvailable():
+    return options.umf is not None
+
+class UMFSuite(Suite):
+    def __init__(self, directory):
+        self.directory = directory
+        if not isUMFAvailable():
+            print("UMF not provided. Related benchmarks will not run")
+
+    def setup(self):
+        if not isUMFAvailable():
+            return []
+        self.built = True
+
+    def benchmarks(self) -> list[Benchmark]:
+        if not isUMFAvailable():
+            return
+
+        benches = [
+            GBench(self),
+        ]
+
+        return benches
+
+class ComputeUMFBenchmark(Benchmark):
+    def __init__(self, bench, name):
+        self.bench = bench
+        self.bench_name = name
+        self.oneapi = get_oneapi()
+
+        self.col_name = None
+        self.col_iterations = None
+        self.col_real_time = None
+        self.col_cpu_time = None
+        self.col_time_unit = None
+
+        self.col_statistics_time = None
+
+        super().__init__(bench.directory)
+
+    def bin_args(self) -> list[str]:
+        return []
+
+    def extra_env_vars(self) -> dict:
+        return {}
+
+    def setup(self):
+        if not isUMFAvailable():
+            print("UMF prefix path not provided")
+            return
+
+        self.benchmark_bin = os.path.join(options.umf, 'benchmark', self.bench_name)
+
+    def run(self, env_vars) -> list[Result]:
+        command = [
+            f"{self.benchmark_bin}",
+        ]
+
+        command += self.bin_args()
+        env_vars.update(self.extra_env_vars())
+
+        result = self.run_bench(command, env_vars, add_sycl=False, ld_library=[self.oneapi.tbb_lib()])
+        parsed = self.parse_output(result)
+        results = []
+        for r in parsed:
+            (config, pool, mean) = r
+            label = f"{config} {pool}"
+            results.append(Result(label=label, value=mean, command=command, env=env_vars, stdout=result, unit="ns", explicit_group=config))
+        return results
+
+    # Implementation with self.col_* indices could lead to the division by None
+    def get_mean(self, datarow):
+        raise NotImplementedError()
+
+    def teardown(self):
+        return
+
+class GBench(ComputeUMFBenchmark):
+    def __init__(self, bench):
+        super().__init__(bench, "umf-benchmark")
+
+        self.col_name = 0
+        self.col_iterations = 1
+        self.col_real_time = 2
+        self.col_cpu_time = 3
+        self.col_time_unit = 4
+
+        self.idx_pool = 0
+        self.idx_config = 1
+        self.name_separator = '/'
+
+        self.col_statistics_time = self.col_real_time
+
+    def name(self):
+        return self.bench_name
+
+    # --benchmark_format describes stdout output
+    # --benchmark_out=<file> and --benchmark_out_format=<format>
+    # describe output to a file 
+    def bin_args(self):
+        return ["--benchmark_format=csv"]
+
+    # the default unit
+    # might be changed globally with --benchmark_time_unit={ns|us|ms|s}
+    # the change affects only benchmark where time unit has not been set
+    # explicitly
+    def unit(self):
+        return "ns"
+
+    # these benchmarks are not stable, so set this at a large value
+    def stddev_threshold(self) -> float:
+        return 0.2 # 20%
+
+    def get_pool_and_config(self, full_name):
+        list_split = full_name.split(self.name_separator, 1)
+        if len(list_split) != 2:
+            raise ValueError("Incorrect benchmark name format: ", full_name)
+
+        return list_split[self.idx_pool], list_split[self.idx_config]
+
+    def get_mean(self, datarow):
+        return float(datarow[self.col_statistics_time])
+
+    def parse_output(self, output):
+        csv_file = io.StringIO(output)
+        reader = csv.reader(csv_file)
+
+        data_row = next(reader, None)
+        if data_row is None:
+            raise ValueError("Benchmark output does not contain data.")
+
+        results = []
+        for row in reader:
+            try:
+                full_name = row[self.col_name]
+                pool, config = self.get_pool_and_config(full_name)
+                mean = self.get_mean(row)
+                results.append((config, pool, mean))
+            except KeyError as e:
+                raise ValueError(f"Error parsing output: {e}")
+
+        return results

scripts/benchmarks/main.py

@@ -9,6 +9,7 @@
 from benches.velocity import VelocityBench
 from benches.syclbench import *
 from benches.llamacpp import *
+from benches.umf import *
 from benches.test import TestSuite
 from benches.options import Compare, options
 from output_markdown import generate_markdown
@@ -74,7 +75,7 @@ def remove_outliers(results: dict[str, list[Result]], threshold: float = 3.5) ->
 
     return new_results
 
-def process_results(results: dict[str, list[Result]]) -> tuple[bool, list[Result]]:
+def process_results(results: dict[str, list[Result]], stddev_threshold_override) -> tuple[bool, list[Result]]:
     processed: list[Result] = []
     # technically, we can detect whether result is below or above threshold per
     # individual result. However, we can't repeat benchmark runs with that
@@ -94,7 +95,7 @@ def process_results(results: dict[str, list[Result]]) -> tuple[bool, list[Result
         mean_value = statistics.mean(values)
         stddev = statistics.stdev(values)
 
-        threshold = options.stddev_threshold * mean_value
+        threshold = (stddev_threshold_override if stddev_threshold_override is not None else options.stddev_threshold) * mean_value
 
         if stddev > threshold:
             print(f"stddev {stddev} above the threshold {threshold} for {label}")
@@ -120,6 +121,7 @@ def main(directory, additional_env_vars, save_name, compare_names, filter):
         VelocityBench(directory),
         SyclBench(directory),
         LlamaCppBench(directory),
+        UMFSuite(directory),
         #TestSuite()
     ] if not options.dry_run else []
 
@@ -159,7 +161,7 @@ def main(directory, additional_env_vars, save_name, compare_names, filter):
             processed: list[Result] = []
             for _ in range(5):
                 run_iterations(benchmark, merged_env_vars, options.iterations, intermediate_results)
-                valid, processed = process_results(intermediate_results)
+                valid, processed = process_results(intermediate_results, benchmark.stddev_threshold())
                 if valid:
                     break
             results += processed
@@ -231,6 +233,7 @@ def validate_and_parse_env_args(env_args):
     parser.add_argument('benchmark_directory', type=str, help='Working directory to setup benchmarks.')
     parser.add_argument('--sycl', type=str, help='Root directory of the SYCL compiler.', default=None)
     parser.add_argument('--ur', type=str, help='UR install prefix path', default=None)
+    parser.add_argument('--umf', type=str, help='UMF install prefix path', default=None)
     parser.add_argument('--adapter', type=str, help='Options to build the Unified Runtime as part of the benchmark', default="level_zero")
     parser.add_argument("--no-rebuild", help='Rebuild the benchmarks from scratch.', action="store_true")
     parser.add_argument("--env", type=str, help='Use env variable for a benchmark run.', action="append", default=[])
@@ -267,6 +270,7 @@ def validate_and_parse_env_args(env_args):
     options.output_html = args.output_html
     options.output_markdown = args.output_markdown
     options.dry_run = args.dry_run
+    options.umf = args.umf
 
     benchmark_filter = re.compile(args.filter) if args.filter else None
 

scripts/benchmarks/output_html.py

@@ -157,6 +157,9 @@ def create_grouped_bar_charts(groups: list[ExplicitGroup]) -> list[BenchmarkChar
             ax.bar_label(rects, fmt='')
 
             for rect, run, res in zip(rects, run_results.keys(), run_results.values()):
+                if res is None:
+                    continue
+
                 height = rect.get_height()
                 if height > max_height:
                     max_height = height