From afb9951ed8e2550667bd2dfd0ebacb94e62857be Mon Sep 17 00:00:00 2001
From: Oleksandr Pavlyk <21087696+oleksandr-pavlyk@users.noreply.github.com>
Date: Mon, 28 Jul 2025 11:41:55 -0500
Subject: [PATCH] Enable building of NVBench as part of buildign extension

1. Download and include CPM.cmake, version 0.42.0
2. Use CPM.make to get Pybind11
3. Update to use pybind11=3.0.0
4. Also use CPM to configure/build nvbench
---
 python/CMakeLists.txt | 41 +++++++++++++++++++++++++++--------------
 python/README.md      | 39 ++++++++++++++++++++-------------------
 2 files changed, 47 insertions(+), 33 deletions(-)

diff --git a/python/CMakeLists.txt b/python/CMakeLists.txt
index b61da52..abfc59a 100644
--- a/python/CMakeLists.txt
+++ b/python/CMakeLists.txt
@@ -1,29 +1,42 @@
 cmake_minimum_required(VERSION 3.30...4.0)
 
-project(${SKBUILD_PROJECT_NAME} LANGUAGES CXX)
-
-set(CMAKE_CXX_STANDARD 20)
-set(CMAKE_POSITION_INDEPENDENT_CODE ON)
+# CUDA is transitive dependency of nvbench
+project(${SKBUILD_PROJECT_NAME} LANGUAGES CXX CUDA)
 
 find_package(Python REQUIRED COMPONENTS Development.Module)
 find_package(CUDAToolkit REQUIRED)
 
-include(FetchContent)
-
-FetchContent_Declare(
-    pybind11
-    URL https://github.com/pybind/pybind11/archive/refs/tags/v2.13.6.tar.gz
-    URL_HASH SHA256=e08cb87f4773da97fa7b5f035de8763abc656d87d5773e62f6da0587d1f0ec20
-    FIND_PACKAGE_ARGS NAMES pybind11
+# Get CMake package manager
+set(_cpm_download_location ${CMAKE_CURRENT_BINARY_DIR}/cmake/CPM.cmake)
+file(
+  DOWNLOAD
+  https://github.com/cpm-cmake/CPM.cmake/releases/download/v0.42.0/CPM.cmake
+  ${_cpm_download_location}
+  EXPECTED_HASH SHA256=2020b4fc42dba44817983e06342e682ecfc3d2f484a581f11cc5731fbe4dce8a
 )
-FetchContent_MakeAvailable(pybind11)
+include(${_cpm_download_location})
 
-find_package(nvbench CONFIG REQUIRED)
+CPMAddPackage(
+   NAME nvbench
+   SOURCE_DIR ${CMAKE_CURRENT_SOURCE_DIR}/..
+   OPTIONS "NVBench_INSTALL_RULES ON"
+   FIND_PACKAGE_ARGS CONFIG REQUIRED
+)
+
+CPMAddPackage("gh:pybind/pybind11@3.0.0")
 
 pybind11_add_module(_nvbench MODULE src/py_nvbench.cpp)
 target_link_libraries(_nvbench PUBLIC nvbench::nvbench)
 target_link_libraries(_nvbench PRIVATE CUDA::cudart_static)
+
 set_target_properties(_nvbench PROPERTIES INSTALL_RPATH "$ORIGIN")
+set_target_properties(_nvbench PROPERTIES INTERPROCEDURAL_OPTIMIZATION ON)
+set_target_properties(_nvbench PROPERTIES POSITION_INDEPENDENT_CODE ON)
+set_target_properties(_nvbench PROPERTIES CXX_STANDARD 20)
 
 install(TARGETS _nvbench DESTINATION cuda/nvbench)
-install(IMPORTED_RUNTIME_ARTIFACTS nvbench::nvbench DESTINATION cuda/nvbench)
+
+# Determine target that nvbench::nvbench is an alias of,
+# necessary because ALIAS targets cannot be installed
+get_target_property(_aliased_target_name nvbench::nvbench ALIASED_TARGET)
+install(IMPORTED_RUNTIME_ARTIFACTS ${_aliased_target_name} DESTINATION cuda/nvbench)
diff --git a/python/README.md b/python/README.md
index dffd70e..bd812f5 100644
--- a/python/README.md
+++ b/python/README.md
@@ -4,7 +4,7 @@ This package provides Python API to CUDA Kernel Benchmarking Library `NVBench`.
 
 ## Building
 
-### Build `NVBench` project
+### Ensure recent version of CMake
 
 Since `nvbench` requires a rather new version of CMake (>=3.30.4), either build CMake from sources, or create a conda environment with a recent version of CMake, using
 
@@ -13,48 +13,49 @@ conda create -n build_env --yes  cmake ninja
 conda activate build_env
 ```
 
+### Ensure CUDA compiler
+
+Since building `NVBench` library requires CUDA compiler, ensure that appropriate environment variables
+are set. For example, assuming CUDA toolkit is installedsystem-wide, and assuming Ampere GPU architecture:
+
+```bash
+export CUDACXX=/usr/local/cuda/bin/nvcc
+export CUDAARCHS=86
+``
+
+### Build Python project
+
 Now switch to python folder, configure and install NVBench library, and install the package in editable mode:
 
-```
+```bash
 cd nvbench/python
-cmake -B nvbench_build --preset nvbench-ci -S $(pwd)/.. -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc -DNVBench_ENABLE_EXAMPLES=OFF -DCMAKE_INSTALL_PREFIX=$(pwd)/nvbench_install
-cmake --build nvbench_build/ --config Release --target install
+pip install -e .
 ```
 
-### Build Python extension
-
-Specify location local installation of `NVBench` library and perform editable `pip install`:
-
-```
-nvbench_DIR=$(pwd)/nvbench_install/lib/cmake CUDACXX=/usr/local/cuda/bin/nvcc pip install -e .
-```
-
-Note that `CUDACXX` must be set for NVBench cmake script to work, but Python extension itself only uses host compiler.
-
 ### Verify that package works
 
-```
+```bash
 python test/run_1.py
 ```
 
 ### Run examples
 
-```
+```bash
 # Example benchmarking numba.cuda kernel
 python examples/throughput.py
 ```
 
-```
+```bash
 # Example benchmarking kernels authored using cuda.core
 python examples/axes.py
 ```
 
-```
+```bash
 # Example benchmarking algorithms from cuda.cccl.parallel
 python examples/cccl_parallel_segmented_reduce.py
 ```
 
-```
+```bash
 # Example benchmarking CuPy function
 python examples/cupy_extract.py
 ```