# Clean PATH setup (no duplicates)
export PATH="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"

# CUDA environment (12.8 default, also available: 12.9, 13.0, 13.1, 13.2)
# Switch with: export CUDA_HOME=/usr/local/cuda-13.2 && export PATH="$CUDA_HOME/bin:$PATH"
export CUDA_HOME=/usr/local/cuda-12.8
export PATH="$CUDA_HOME/bin:$PATH"
export LD_LIBRARY_PATH="$CUDA_HOME/lib64:${LD_LIBRARY_PATH:-}"

# EFA and OpenMPI environment for multi-node GPU communication
export PATH="/opt/amazon/openmpi/bin:/opt/amazon/efa/bin:$PATH"
export LD_LIBRARY_PATH="/opt/amazon/ofi-nccl/lib:/opt/amazon/openmpi/lib:/opt/amazon/efa/lib:${LD_LIBRARY_PATH:-}"

# NCCL settings for EFA with GPUDirect RDMA
export FI_PROVIDER=efa
export FI_EFA_USE_DEVICE_RDMA=1
export NCCL_NET_GDR_LEVEL=SYS
export NCCL_DEBUG=INFO
export NCCL_ALGO=ring,tree
export NCCL_SOCKET_IFNAME=enp71s0
export NCCL_IB_HCA=^mlx
export NCCL_CROSS_NIC=0
export NCCL_IB_PCI_RELAXED_ORDERING=1
export NCCL_ASYNC_ERROR_HANDLING=1

# Node.js user global packages (for Claude CLI)
export PATH="$HOME/.npm-global/bin:$PATH"

# ccache configuration for faster C++ compilation
export PATH="/usr/lib/ccache:$PATH"
export CCACHE_DIR="/ccache_shared"
export CCACHE_MAXSIZE="10G"

# Claude Code configuration for Bedrock
export CLAUDE_CODE_USE_BEDROCK=1