[INFRA] Import NVIDIA/CCCL upstream as optimization reference library

CCCL (CUDA C++ Core Libraries) provides:
- CUB: device/block/warp-level GPU primitives (reduce, scan, sort, topk)
- Thrust: high-level parallel algorithms (transform_reduce, sort, scan)
- libcudacxx: CUDA C++ standard library (atomics, barriers, memory)
- cudax: experimental features (memory resources, allocators)
- Tuning policies: per-SM hardware-specific algorithm parameters

Competition optimization vectors mapped to CCCL:
- Output TPS (83% weight): warp_reduce, block_reduce, device_topk
- Input TPS (14% weight): device_scan, block_load, prefetch
- Cache TPS (3% weight): prefix caching strategy patterns
- Memory (0.9 util): pooled/cached/buddy allocators

Source: https://github.com/NVIDIA/cccl (shallow clone, HEAD only)
License: Apache-2.0

cccl_upstream/benchmarks/scripts/cccl/bench/cmake.py

@@ -0,0 +1,138 @@
+import os
+import signal
+import subprocess
+import time
+
+from .build import Build
+from .config import Config
+from .logger import Logger
+from .storage import Storage
+
+
+def create_builds_table(conn):
+    with conn:
+        conn.execute("""
+        CREATE TABLE IF NOT EXISTS builds (
+            ctk TEXT NOT NULL,
+            cccl TEXT NOT NULL,
+            bench TEXT NOT NULL,
+            code TEXT NOT NULL,
+            elapsed REAL
+        );
+        """)
+
+
+class CMakeCache:
+    _instance = None
+
+    def __new__(cls, *args, **kwargs):
+        if cls._instance is None:
+            cls._instance = super().__new__(cls, *args, **kwargs)
+            create_builds_table(Storage().connection())
+        return cls._instance
+
+    def pull_build(self, bench):
+        config = Config()
+        ctk = config.ctk
+        cccl = config.cccl
+        conn = Storage().connection()
+
+        with conn:
+            query = "SELECT code, elapsed FROM builds WHERE ctk = ? AND cccl = ? AND bench = ?;"
+            result = conn.execute(query, (ctk, cccl, bench.label())).fetchone()
+
+            if result:
+                code, elapsed = result
+                return Build(int(code), float(elapsed))
+
+            return result
+
+    def push_build(self, bench, build):
+        config = Config()
+        ctk = config.ctk
+        cccl = config.cccl
+        conn = Storage().connection()
+
+        with conn:
+            conn.execute(
+                "INSERT INTO builds (ctk, cccl, bench, code, elapsed) VALUES (?, ?, ?, ?, ?);",
+                (ctk, cccl, bench.label(), build.code, build.elapsed),
+            )
+
+
+class CMake:
+    def __init__(self):
+        pass
+
+    def do_build(self, bench, timeout):
+        logger = Logger()
+
+        try:
+            if not bench.is_base():
+                with open(bench.exe_name() + ".h", "w") as f:
+                    f.writelines(bench.definitions())
+
+            cmd = ["cmake", "--build", ".", "--target", bench.exe_name()]
+            logger.info(
+                "starting build for {}: {}".format(bench.label(), " ".join(cmd))
+            )
+
+            begin = time.time()
+            p = subprocess.Popen(
+                cmd,
+                start_new_session=True,
+                stdout=subprocess.DEVNULL,
+                stderr=subprocess.DEVNULL,
+            )
+            p.wait(timeout=timeout)
+            elapsed = time.time() - begin
+            logger.info(
+                "finished build for {} (exit code: {}) in {:.3f}s".format(
+                    bench.label(), p.returncode, elapsed
+                )
+            )
+
+            return Build(p.returncode, elapsed)
+        except subprocess.TimeoutExpired:
+            logger.info(
+                "build for {} reached timeout of {}s".format(bench.label(), timeout)
+            )
+            os.killpg(os.getpgid(p.pid), signal.SIGTERM)
+            return Build(424242, float("inf"))
+
+    def build(self, bench):
+        logger = Logger()
+        timeout = None
+
+        cache = CMakeCache()
+
+        if bench.is_base():
+            # Only base build can be pulled from cache
+            build = cache.pull_build(bench)
+
+            if build:
+                logger.info("found cached base build for {}".format(bench.label()))
+                if bench.is_base():
+                    if not os.path.exists("bin/{}".format(bench.exe_name())):
+                        self.do_build(bench, None)
+
+                return build
+        else:
+            base_build = self.build(bench.get_base())
+
+            if base_build.code != 0:
+                raise Exception("Base build failed")
+
+            timeout = base_build.elapsed * 10
+
+        build = self.do_build(bench, timeout)
+        cache.push_build(bench, build)
+        return build
+
+    def clean():
+        cmd = ["cmake", "--build", ".", "--target", "clean"]
+        p = subprocess.Popen(cmd, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+        p.wait()
+
+        if p.returncode != 0:
+            raise Exception("Unable to clean build directory")