[fs_connector][feat]: Add storage_offload CUDA kernel and setup config

Signed-off-by: Kfir Toledo <kfir.toledo@ibm.com>

Author: kfirtoledo

kv_connectors/llmd_fs_backend/setup.py

@@ -0,0 +1,26 @@
+from setuptools import setup, find_packages
+from torch.utils.cpp_extension import CUDAExtension, BuildExtension
+
+setup(
+    name="storage_offload",
+    packages=find_packages("src"),
+    package_dir={"": "src"},
+    ext_modules=[
+        CUDAExtension(
+            "storage_offload",
+             sources=[
+                "src/csrc/storage/storage_offload.cu",
+                "src/csrc/storage/buffer.cpp",
+                "src/csrc/storage/file_io.cpp",
+                "src/csrc/storage/thread_pool.cpp",
+                "src/csrc/storage/tensor_copy.cu",
+            ],
+            libraries=['nvidia-ml', 'numa', 'cuda'],
+            extra_compile_args={
+                "cxx": ["-O3", "-std=c++17", "-fopenmp"],
+                "nvcc": ["-O3", "-std=c++17", "-Xcompiler", "-std=c++17","-Xcompiler", "-fopenmp"]
+            }
+        ),
+    ],
+    cmdclass={"build_ext": BuildExtension},
+)

kv_connectors/llmd_fs_backend/src/csrc/storage/debug_utils.hpp

@@ -0,0 +1,33 @@
+#pragma once
+
+#include <cstdlib>
+#include <iostream>
+#include <string>
+#include <chrono>
+
+// -------------------------------------
+// Debugging and timing macros
+// -------------------------------------
+
+// Debug print - enabled when STORAGE_CONNECTOR_DEBUG is set and not "0"
+#define DEBUG_PRINT(msg)                                                     \
+    do {                                                                     \
+        const char* env = std::getenv("STORAGE_CONNECTOR_DEBUG");            \
+        if (env && std::string(env) != "0")                                 \
+            std::cout << "[DEBUG] " << msg << std::endl;                     \
+    } while (0)
+
+// Timing macro - measures execution time when STORAGE_CONNECTOR_DEBUG  is set and not "0"
+#define TIME_EXPR(label, expr, info_str) ([&]() {                                  \
+    const char* env = std::getenv("STORAGE_CONNECTOR_DEBUG");                      \
+    if (!(env && std::string(env) != "0")) {                                       \
+        return (expr);                                                             \
+    }                                                                              \
+    auto __t0 = std::chrono::high_resolution_clock::now();                         \
+    auto __ret = (expr);                                                           \
+    auto __t1 = std::chrono::high_resolution_clock::now();                         \
+    double __ms = std::chrono::duration<double, std::milli>(__t1 - __t0).count();  \
+    std::cout << "[DEBUG][TIME] " << label << " took " << __ms << " ms | "         \
+              << info_str << std::endl;                                            \
+    return __ret;                                                                  \
+})()