Skip to content

SETUP_WINDOWS.md

Latest commit

 

History

History
363 lines (254 loc) · 12 KB

SETUP_WINDOWS.md

File metadata and controls

363 lines (254 loc) · 12 KB
Raw

Windows Build Environment Setup (Visual Build Tools, CUDA Toolkit, Make)

This guide installs the required tools to build and run the CUDA Matrix Multiplication project on Windows. It targets modern Windows and recommends Visual Studio 2022 for IDE or Build Tools.

Prerequisites

  • Windows 10/11 with admin privileges
  • Internet connection for downloads

Overview

  • Visual C++ Build Tools (MSVC): provides cl.exe used by CUDA's host compiler
  • NVIDIA CUDA Toolkit: provides nvcc, cuBLAS, headers, and libraries
  • GNU Make (optional, used by the provided Makefile) or use CMake + MSVC
  1. Install Visual Studio (or Build Tools) (MSVC) — recommended: 2022

-- Download: https://visualstudio.microsoft.com/downloads/ -- Choose: Visual Studio 2022 (Community/Professional/Enterprise) or Build Tools for Visual Studio 2022.

  • During installation check: Desktop development with C++ workload.
  • Optional but recommended components:
    • MSVC v143 (or latest) toolset
    • Windows 10/11 SDK
    • CMake tools for Windows

Notes:

  • Using Build Tools 2026 is fine — CUDA toolkits support current MSVC toolsets. If you encounter an error, try the 2022 toolset.
  1. Install NVIDIA CUDA Toolkit (includes nvcc, cuBLAS)
  • Download: https://developer.nvidia.com/cuda-downloads -- Choose: OS = Windows, Architecture = x86_64, Installer Type = exe (local) -- Recommended CUDA Toolkit version for GTX 1050 Ti (Pascal / sm_61): CUDA 12.9. Install CUDA 12.9 to ensure offline compile support for sm_61.
  • Run the installer as Administrator. Recommended options:
    • Install CUDA Toolkit (required)
    • Install cuBLAS (required)
    • You can skip the NVIDIA driver step if your driver is current
  • After install, restart your machine if prompted.

Verify CUDA installation in PowerShell:

nvcc --version
nvidia-smi

If nvcc is not found, add to PATH:

# $PS> setx PATH "$Env:PATH;C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0\bin"

Replace v12.0 with the installed CUDA version directory (for this project prefer v12.9).

  1. Install GNU Make (if you prefer Makefile)

Option A — Chocolatey (recommended if you have Chocolatey):

choco install make

Option B — winget:

winget install GnuWin32.Make

Option C — Manual (less recommended):

  • Download GNU Make for Windows from GnuWin32 and add bin to PATH.

Verify:

make --version
  1. (Optional) Install CMake (recommended for cross-platform builds)

Download: https://cmake.org/download/

Verify:

cmake --version
  1. Build the project (Make or CMake)

Using Make (requires make and nvcc on PATH):

cd c:\personal\CUDA\CUDA_Matrix_Multiplication
make all
make test
make bench

Using CMake (MSVC generator):

cd c:\personal\CUDA\CUDA_Matrix_Multiplication
mkdir build
cd build
cmake .. -G "Visual Studio 17 2022"   # use 2026 generator name if available
cmake --build . --config Release
  1. Troubleshooting
  • nvcc: command not found — CUDA Toolkit not added to PATH. Add C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v<version>\bin to PATH.
  • cl.exe: command not found — Visual Build Tools not installed or Developer Command Prompt not used. Ensure MSVC installed and PATH includes MSVC bin (re-open PowerShell after install).
  • cublas_v2.h: No such file or directory — Reinstall CUDA Toolkit, ensure cuBLAS selected.
  • Driver mismatches/crashes — update NVIDIA GPU driver from NVIDIA website.
  • Make problems — use CMake + MSVC if make is unavailable.
  1. Verifying a successful build and run

After building, run the tests and benchmark executables (paths depend on build system):

# From the project root after a Make build
./bin/test_correctness.exe
./bin/matmul_bench.exe
  1. Notes about WSL2
  1. Where to get help

Commit these files to the repository so contributors have exact steps to prepare their machine.

Quick validation: make sure cl.exe, nvcc, and make are available

If you installed Visual Studio Build Tools but cl.exe is not found in a normal PowerShell session, run the MSVC environment setup script (vcvars64.bat) and then re-check. Use the following PowerShell commands to locate your Visual Studio installation and run the environment script:

# locate Visual Studio installation that has the MSVC toolset
& 'C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe' -latest -products * -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath

# if the above prints a path, store it and run vcvars64.bat (adjust path if different)
$vsPath = (& 'C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe' -latest -products * -requires Microsoft.VisualStudio.Component.VC.Tools.x86.x64 -property installationPath).Trim()
& "$vsPath\VC\Auxiliary\Build\vcvars64.bat"

# After the above, verify the compiler is on PATH
cl.exe
where cl.exe

# verify CUDA toolchain and make
nvcc --version
where nvcc
make --version
where make

Notes:

  • If vswhere.exe is not present, check C:\Program Files (x86)\Microsoft Visual Studio\Installer\vswhere.exe manually — it is normally installed with Visual Studio or the Build Tools installer.
  • If cl.exe works only after running vcvars64.bat, always run builds from a Developer Command Prompt or include the vcvars call in CI/build scripts so the MSVC environment variables are set.

Important note about PowerShell vs cmd.exe

  • Calling vcvars64.bat directly from PowerShell using the & operator does not always propagate MSVC environment variables into the PowerShell session in the same way as the Visual Studio Developer Command Prompt. To ensure the environment is set for subsequent build commands, either:

  • Open Developer Command Prompt for VS 2022 or Developer PowerShell for VS 2022 from the Start Menu and run your build there, or

  • Use cmd.exe to call the batch script and run your build in the same cmd process.

Example (run in a normal PowerShell window):

REM Replace the VS path with your installed VS 2022 path if different
cmd.exe /c '"C:\Program Files (x86)\Microsoft Visual Studio\17\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat" && where cl.exe && cl.exe && "C:\Program Files (x86)\GnuWin32\bin\make.exe" --version'

Or open the Developer Command Prompt and run:

"C:\>" cd c:\personal\CUDA\CUDA_Matrix_Multiplication
"C:\>" make all

If you prefer PowerShell, launch Developer PowerShell for VS 2026 from Visual Studio (it runs the setup script and configures the environment correctly for PowerShell sessions).

Convenience scripts

To simplify validation and building on Windows cmd.exe, the repository includes helper scripts at the project root:

  • validate_env.cmd — Runs vcvars64.bat (if found), checks for cl.exe, nvcc, and make. Usage: validate_env.cmd or validate_env.cmd make to also run make all.
  • build_with_make.cmd — Sets the MSVC environment and runs make all.
  • build_with_cmake.cmd — Sets the MSVC environment and runs CMake configure + build (default Release).
  • use_cuda12.cmd — Auto-detects CUDA 12.x, sets CUDA_BIN, validates toolchain, and runs build actions for Pascal GPUs like GTX 1050 Ti. Usage: use_cuda12.cmd validate|all|test|bench.

Example usage from Command Prompt (recommended):

C:\> cd C:\personal\CUDA\CUDA_Matrix_Multiplication
C:\personal\CUDA\CUDA_Matrix_Multiplication> validate_env.cmd
C:\personal\CUDA\CUDA_Matrix_Multiplication> use_cuda12.cmd validate
C:\personal\CUDA\CUDA_Matrix_Multiplication> use_cuda12.cmd all
C:\personal\CUDA\CUDA_Matrix_Multiplication> build_with_make.cmd
C:\personal\CUDA\CUDA_Matrix_Multiplication> build_with_cmake.cmd Release

These scripts are provided so contributors can quickly validate and build without manually starting a Developer Command Prompt.

Nsight / Visual Studio integration note

The CUDA installer displays a message about Nsight "Not Installed" when it cannot find a Visual Studio IDE to integrate with. This is informational only. If you only need to compile and run CUDA code, the MSVC Build Tools are sufficient and you can ignore the Nsight message. If you want full Nsight integration inside Visual Studio, install the full Visual Studio IDE (Community/Professional/Enterprise) and re-run the CUDA installer or install Nsight from the Visual Studio Marketplace.

Windows Setup: CUDA Toolkit, Build Tools, and Make

This document provides complete, reproducible setup steps for building this project on Windows.

Required Tools

  1. NVIDIA GPU driver (latest stable)
  2. Microsoft Visual Studio 2022 Build Tools with C++ workload
  3. NVIDIA CUDA Toolkit (12.x recommended)
  4. GNU Make (or use CMake build flow)
  5. Python 3.9+ (for plotting scripts)

1) Install Visual Studio C++ Build Tools

CUDA on Windows requires MSVC.

  1. Download installer:
  1. In installer, select workload:
  • Desktop development with C++
  1. Ensure these components are included:
  • MSVC v143 - VS 2022 C++ x64/x86 build tools
  • Windows 10/11 SDK
  • C++ CMake tools for Windows (optional but recommended)
  1. Install and restart Windows.

2) Install CUDA Toolkit

  1. Download from:
  1. Choose:
  • OS: Windows
  • Architecture: x86_64
  • Version: Windows 10/11
  • Installer Type: exe (local)
  1. During install:
  • Keep CUDA Toolkit and cuBLAS selected
  • If your display driver is already up to date, you can skip driver installation
  1. Restart Windows after installation.

3) Install GNU Make

Choose one option.

Option A: winget

winget install GnuWin32.Make

Option B: Chocolatey

choco install make

Option C: Manual

  1. Install via GnuWin32 package.
  2. Add make binary path to PATH.

4) Verify Installation

Open a new PowerShell terminal and run:

nvcc --version
make --version

Check MSVC compiler from a Developer Command Prompt, or ensure VS tools are on PATH:

cl.exe

If cl.exe is not recognized in regular PowerShell, launch builds from one of these:

  • x64 Native Tools Command Prompt for VS 2022
  • Developer PowerShell for VS 2022

5) Build This Project

From repository root:

cd c:\personal\CUDA\CUDA_Matrix_Multiplication
make all
make test
make bench

If you prefer CMake:

cd c:\personal\CUDA\CUDA_Matrix_Multiplication
cmake -S . -B build -G "Visual Studio 17 2022"
cmake --build build --config Release

6) Generate Plots

cd c:\personal\CUDA\CUDA_Matrix_Multiplication\python
python benchmark_plot.py
python roofline.py

Output files are written under results/.

Troubleshooting

nvcc not found

  • Reopen terminal after install.
  • Verify CUDA bin path exists, e.g.:
    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.x\bin
  • Add CUDA bin to PATH and reopen terminal.

make not found

  • Reopen terminal.
  • Confirm make installation path is in PATH.
  • Use CMake workflow if make is unavailable.

cl.exe not found

  • Install VS Build Tools C++ workload.
  • Use Developer PowerShell/Prompt for VS 2022.

Link or include errors for cuBLAS

  • Ensure CUDA Toolkit installation included cuBLAS.
  • Verify CUDA install is intact and environment variables are refreshed.

Notes for Contributors

  • These steps are required for local Windows builds.
  • Please include your tool versions when reporting issues:
nvcc --version
make --version
python --version