Project 2: Yunhao Qian

.clang-format

@@ -0,0 +1,7 @@
+BasedOnStyle: LLVM
+
+BreakTemplateDeclarations: Yes
+ColumnLimit: 100
+FixNamespaceComments: false
+IndentWidth: 4
+NamespaceIndentation: All

CMakeLists.txt

@@ -44,3 +44,6 @@ source_group(Sources FILES ${sources})
 add_executable(${CMAKE_PROJECT_NAME} ${sources} ${headers})
 target_link_libraries(${CMAKE_PROJECT_NAME} stream_compaction)
 set_property(DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTY VS_STARTUP_PROJECT ${CMAKE_PROJECT_NAME})
+
+add_executable(measure_time src/measure_time.cpp ${headers})
+target_link_libraries(measure_time stream_compaction)

scripts/.gitignore

@@ -0,0 +1 @@
+__pycache__

scripts/helper.py

@@ -0,0 +1,32 @@
+import subprocess
+from pathlib import Path
+from typing import Literal
+
+ROOT_DIR = Path(__file__).parent.parent.absolute()
+MEASURE_TIME_EXE = ROOT_DIR / "build" / "bin" / "Release" / "measure_time.exe"
+assert MEASURE_TIME_EXE.is_file()
+
+
+def measure_time(
+    operation: Literal["scan", "compact"],
+    implementation: Literal["cpu", "naive", "efficient", "thrust"],
+    input_size: int,
+    block_size: int = -1,
+    elements_per_thread: int = -1,
+) -> float:
+    result = subprocess.run(
+        [
+            str(MEASURE_TIME_EXE),
+            operation,
+            implementation,
+            str(input_size),
+            str(block_size),
+            str(elements_per_thread),
+        ],
+        capture_output=True,
+        cwd=ROOT_DIR,
+        check=True,
+        encoding="utf-8",
+        text=True,
+    )
+    return float(result.stdout)

scripts/measure_performance.py

@@ -0,0 +1,53 @@
+import json
+from pathlib import Path
+
+from helper import measure_time
+
+
+def main() -> None:
+    performance_json_path = Path(__file__).parent / "performance.json"
+    if performance_json_path.exists():
+        with performance_json_path.open(encoding="utf-8") as f:
+            performance_data = json.load(f)
+    else:
+        performance_data = []
+
+    def find_existing_entry(
+        operation: str, implementation: str, input_size: int
+    ) -> dict | None:
+        for entry in performance_data:
+            if (
+                entry["operation"] == operation
+                and entry["implementation"] == implementation
+                and entry["input_size"] == input_size
+            ):
+                return entry
+        return None
+
+    for implementation in ["cpu", "naive", "efficient", "efficient_plus", "thrust"]:
+        for exponent in range(4, 28):
+            base_input_size = 1 << exponent
+            for input_size in [base_input_size, base_input_size - 3]:
+                config = {
+                    "operation": "scan",
+                    "implementation": implementation,
+                    "input_size": input_size,
+                }
+                if find_existing_entry(**config) is not None:
+                    print(f"Skipping existing entry: {config}")
+                    continue
+
+                time_sum = 0.0
+                time_count = 10
+                for _ in range(time_count):
+                    time_sum += measure_time("scan", implementation, input_size)
+                time = time_sum / time_count
+                print(f"Measured: {config}, time: {time} ms")
+                performance_data.append({**config, "time": time})
+
+                with performance_json_path.open("w", encoding="utf-8") as f:
+                    json.dump(performance_data, f, indent=4)
+
+
+if __name__ == "__main__":
+    main()