Skip to content

Latest commit

 

History

History
281 lines (210 loc) · 7.05 KB

TROUBLESHOOTING.md

File metadata and controls

281 lines (210 loc) · 7.05 KB

TorchSparse Windows Troubleshooting Guide

This guide addresses common issues encountered when building or installing TorchSparse on Windows.

🚨 Common Build Errors

1. Missing 'google/dense_hash_map' Error

Error Message:

fatal error C1083: Cannot open include file: 'google/dense_hash_map': No such file or directory

Root Cause: Missing sparsehash dependency

Solutions:

Option A: Quick Fix (Recommended)

# Download pre-built headers
Invoke-WebRequest -Uri "https://github.com/sparsehash/sparsehash/archive/refs/tags/sparsehash-2.0.4.zip" -OutFile "sparsehash.zip"
Expand-Archive -Path "sparsehash.zip" -DestinationPath "C:\"
Rename-Item "C:\sparsehash-sparsehash-2.0.4" "C:\sparsehash"

# Set environment variable
$env:INCLUDE = "$env:INCLUDE;C:\sparsehash\src"

Option B: Using vcpkg

git clone https://github.com/Microsoft/vcpkg.git C:\vcpkg
cd C:\vcpkg
.\bootstrap-vcpkg.bat
.\vcpkg install sparsehash:x64-windows
$env:CMAKE_TOOLCHAIN_FILE = "C:\vcpkg\scripts\buildsystems\vcpkg.cmake"

Option C: Manual Build

git clone https://github.com/sparsehash/sparsehash.git C:\sparsehash-src
cd C:\sparsehash-src
mkdir build && cd build
cmake .. -DCMAKE_INSTALL_PREFIX=C:\sparsehash -A x64
cmake --build . --config Release
cmake --install .
$env:INCLUDE = "$env:INCLUDE;C:\sparsehash\include"

2. Compiler Out of Heap Space

Error Message:

fatal error C1060: compiler is out of heap space
catastrophic error: out of memory

Root Cause: Insufficient memory during compilation

Solutions:

Immediate Fixes:

# 1. Reduce optimization level
$env:CL = "/O1 /MP2"  # Use O1 instead of O2, limit parallel jobs

# 2. Increase virtual memory
# Go to System Properties > Advanced > Performance Settings > Advanced > Virtual Memory
# Set to 16GB+ or let Windows manage automatically

# 3. Close unnecessary applications
Get-Process | Where-Object {$_.WorkingSet -gt 500MB} | Stop-Process -Force

Long-term Solutions:

  1. Upgrade RAM: 16GB+ recommended for building
  2. Use SSD: Faster virtual memory access
  3. Use pre-built wheels: Avoid compilation entirely

3. CUDA Architecture Mismatch

Error Message:

nvcc fatal : Unsupported gpu architecture 'compute_XX'

Solution:

# Check your GPU architecture
nvidia-smi

# Set appropriate architectures (adjust based on your GPU)
$env:TORCH_CUDA_ARCH_LIST = "7.5;8.0;8.6;8.9"

# Common GPU architectures:
# RTX 20xx series: 7.5
# RTX 30xx series: 8.6
# RTX 40xx series: 8.9

4. Visual Studio Not Found

Error Message:

Microsoft Visual C++ 14.0 is required

Solutions:

Option A: Install Visual Studio Build Tools

# Download and install Visual Studio Build Tools 2019 or 2022
# Include: C++ build tools, Windows 10/11 SDK, CMake tools

Option B: Set Environment Variables

$env:DISTUTILS_USE_SDK = "1"
$env:MSSdk = "1"
$env:VS160COMNTOOLS = "C:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\Common7\Tools\"

5. PyTorch Version Conflicts

Error Message:

RuntimeError: The detected CUDA version (X.X) mismatches the version that was used to compile PyTorch (Y.Y)

Solution:

# Uninstall existing PyTorch
pip uninstall torch torchvision torchaudio

# Install matching CUDA version
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
# or for CUDA 12.1:
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

🔧 Environment Setup Issues

Python Environment Problems

Issue: Multiple Python installations causing conflicts

Solution:

# Use virtual environments
python -m venv torchsparse_env
.\torchsparse_env\Scripts\activate
pip install --upgrade pip setuptools wheel

Path Issues

Issue: Tools not found in PATH

Solution:

# Add to PATH (adjust paths as needed)
$env:PATH += ";C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin"
$env:PATH += ";C:\Program Files (x86)\Microsoft Visual Studio\2019\BuildTools\MSBuild\Current\Bin"

🚀 Performance Optimization

Build Speed Improvements

# Use ninja build system
pip install ninja

# Parallel compilation (adjust based on CPU cores)
$env:CL = "/MP8"  # Use 8 parallel jobs

# Use ccache if available
$env:CC = "ccache cl"
$env:CXX = "ccache cl"

Memory Usage Optimization

# Reduce memory usage during build
$env:CL = "/O1 /MP2 /Zm200"  # Limit memory per compilation unit

# Build in release mode only
$env:CMAKE_BUILD_TYPE = "Release"

🔍 Debugging Build Issues

Enable Verbose Output

# Verbose pip installation
pip install . --verbose --no-build-isolation

# Verbose setup.py
python setup.py build_ext --verbose

Check System Information

# System info script
import torch
import platform
import subprocess

print(f"Platform: {platform.platform()}")
print(f"Python: {platform.python_version()}")
print(f"PyTorch: {torch.__version__}")
print(f"CUDA Available: {torch.cuda.is_available()}")
if torch.cuda.is_available():
    print(f"CUDA Version: {torch.version.cuda}")
    print(f"GPU: {torch.cuda.get_device_name()}")

# Check CUDA toolkit
try:
    result = subprocess.run(["nvcc", "--version"], capture_output=True, text=True)
    print(f"NVCC: {result.stdout}")
except:
    print("NVCC: Not found")

🆘 When All Else Fails

Use Pre-built Wheels

Instead of building from source, use our pre-built wheels:

# Download appropriate wheel from releases
pip install https://github.com/Deathdadev/torchsparse/releases/download/v2.1.0-windows/torchsparse-2.1.0-cp310-cp310-win_amd64.whl

Alternative Installation Methods

# Try different installation flags
pip install git+https://github.com/Deathdadev/torchsparse.git --no-build-isolation --no-cache-dir

# Or use conda-forge (if available)
conda install -c conda-forge torchsparse

Docker Alternative

If native Windows installation continues to fail:

# Use Windows containers with pre-built environment
FROM mcr.microsoft.com/windows/servercore:ltsc2019
# ... setup Python, CUDA, and dependencies

📞 Getting Help

If you're still experiencing issues:

  1. Check existing issues: GitHub Issues

  2. Create new issue with:

    • Complete error log
    • System information (use script above)
    • Steps to reproduce
    • Environment details

  3. Include this information:

    python -c "import torch; print(f'PyTorch: {torch.__version__}, CUDA: {torch.version.cuda}')"
nvcc --version
cl

📚 Additional Resources