Add a new concurrent hashmap benchmark to AI benchmark suite in DCPerf. (#147)

Summary:
Pull Request resolved: #147

This update introduces a new benchmark type, CHM, to the AI benchmark suite. The CHM benchmark simulates the usage pattern of a concurrent hash map in the critical machine learning workloads, where certain key-value pairs are cached in an in-memory key-value concurrent hashmap for retrieval.

#### HOW TO RUN THE NEW BENCHMARK
To install and run the rebatch benchmark, use the following commands:

```
./benchpress_cli.py -b ai install chm_a // Install the chm workload representing workload A
./benchpress_cli.py -b ai install chm_b // Install the chm workload representing workload B

./benchpress_cli.py -b ai run chm_a // Run the chm workload representing workload A
./benchpress_cli.py -b ai run chm_b // Run the chm workload representing workload B
```

#### FILES ADDED AND MODIFIED
```
- benchpress
  - configs
    - benchmarks_ai.yml // Added the new chm benchmark.
    - jobs_ai.yml // Added two new chm benchmark jobs.
  - plugins
    - parsers.py
      - chm.py // Added a new parser for the chm benchmark.
      - __init__.py // Registered the new parser.
- packages
  - chm
    - install_chm.sh // Added a new installation script for chm benchmark.
    - cleanup_chm.sh // Added a new cleanup script for chm benchmark.
    - ChmBenchmark.cpp // Added a benchmark simuates the concurrent hash map workload.
    - CMakeLists.txt // CMake file to compile the benchmark.
    - ConcurrentHashMap.h // A sharded concurrent hash map implementation
    - model_a.dist // Key distribution file for workload A.
    - model_b.dist // Key distribution file for workload B.
- BUCK // Updated to include the new benchmark components.
...
```

Reviewed By: excelle08

Differential Revision: D77214563

fbshipit-source-id: 569b9b623bd3450920e99b004094bab00fd8d57d

Author: Kaiwei Tu

benchpress/config/benchmarks_ai.yml

@@ -14,3 +14,11 @@ rebatch:
   metrics:
     - bandwidth
     - time_per_batch_us
+
+chm:
+  parser: chm
+  install_script: ./packages/chm/install_chm.sh
+  cleanup_script: ./packages/chm/cleanup_chm.sh
+  path: ./benchmarks/chm/chm_bench
+  metrics:
+    - Mops/sec

benchpress/config/jobs_ai.yml

@@ -104,7 +104,7 @@
 
 - benchmark: rebatch
   name: rebatch_a
-  description: rebatch benchmark for Model A.
+  description: Rebatch benchmark for Model A.
   args:
     - '--threads={threads}'
     - '--duration={duration}'
@@ -125,7 +125,7 @@
 
 - benchmark: rebatch
   name: rebatch_b
-  description: rebatch benchmark for Model B.
+  description: Rebatch benchmark for Model B.
   args:
     - '--threads={threads}'
     - '--duration={duration}'
@@ -143,3 +143,30 @@
     - 'prefetch=0'
     - 'output_tensor_size=2421002'
     - 'memory_pool_size=4'
+
+- benchmark: chm
+  name: chm_a
+  description: Concurrent hash map benchmark for Model A.
+  args:
+    - '--distribution_file={distribution_file}'
+    - '--num_threads={num_threads}'
+    - '--duration_seconds={duration_seconds}'
+
+  vars:
+    - 'distribution_file=benchmarks/chm/model_a.dist'
+    - 'num_threads=24'
+    - 'duration_seconds=360'
+
+
+- benchmark: chm
+  name: chm_b
+  description: Concurrent hash map benchmark for Model B.
+  args:
+    - '--distribution_file={distribution_file}'
+    - '--num_threads={num_threads}'
+    - '--duration_seconds={duration_seconds}'
+
+  vars:
+    - 'distribution_file=benchmarks/chm/model_b.dist'
+    - 'num_threads=24'
+    - 'duration_seconds=360'

benchpress/plugins/parsers/__init__.py

@@ -11,6 +11,7 @@
 from .benchdnn import BenchdnnParser
 from .cachebench import CacheBenchParser
 from .checkmark import CheckmarkParser
+from .chm import ChmParser
 from .clang import ClangParser
 from .cloudsuite_graph import CloudSuiteGraphParser
 from .compression_parser import CompressionParser
@@ -104,6 +105,7 @@ def register_parsers(factory):
     factory.register("syscall", SyscallParser)
     factory.register("embedding", EmbeddingParser)
     factory.register("rebatch", RebatchParser)
+    factory.register("chm", ChmParser)
 
     if not open_source:
         factory.register("adsim", AdSimParser)

benchpress/plugins/parsers/chm.py

@@ -0,0 +1,27 @@
+#!/usr/bin/env python3
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+# This source code is licensed under the MIT license found in the
+# LICENSE file in the root directory of this source tree.
+
+# pyre-unsafe
+import re
+
+from benchpress.lib.parser import Parser
+
+
+class ChmParser(Parser):
+    def parse(self, stdout, stderr, returncode):
+        metrics = {}
+
+        # Parse stdout for operations per second
+        for line in stdout:
+            # Match "Millions of Operations per Second: X Mops/sec"
+            match = re.search(
+                r"Millions of Operations per Second:\s*(\d+\.\d+)\s*Mops/sec", line
+            )
+            if match:
+                metrics["Mops/sec"] = float(match.group(1))
+                break
+
+        return metrics

packages/chm/CMakeLists.txt

@@ -0,0 +1,17 @@
+cmake_minimum_required(VERSION 3.10)
+project(ChmBench LANGUAGES CXX)
+
+set(CMAKE_PREFIX_PATH
+    "${CMAKE_SOURCE_DIR}/installed"
+    ${CMAKE_PREFIX_PATH})
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+
+
+find_package(folly CONFIG REQUIRED)
+find_package(gflags CONFIG REQUIRED)
+
+add_executable(chm_bench ChmBenchmark.cpp)
+
+
+target_link_libraries(chm_bench PRIVATE Folly::folly)