ids.alfacom.it/extracted_idf/diagnose_tesla_m60.sh

#!/bin/bash
# =========================================================================
# DIAGNOSI COMPLETA TESLA M60 + TENSORFLOW
# Identifica esattamente il problema e la soluzione
# =========================================================================

set -e

echo "🔍 DIAGNOSI COMPLETA TESLA M60 + TENSORFLOW"
echo "==========================================="

# 1. VERIFICA DRIVER NVIDIA
echo "🎮 VERIFICA DRIVER NVIDIA..."
nvidia-smi --query-gpu=name,driver_version,compute_cap,memory.total --format=csv,noheader

echo -e "\n🔧 Verifica librerie driver..."
ls -la /usr/lib64/libcuda* || echo "❌ libcuda non trovato"
ls -la /usr/lib64/libnvidia* | head -5 || echo "❌ libnvidia non trovato"

# 2. VERIFICA INSTALLAZIONI CUDA
echo -e "\n📦 VERIFICA INSTALLAZIONI CUDA..."
echo "CUDA 12.2:"
ls -la /usr/local/cuda-12.2/lib64/libcuda* 2>/dev/null || echo "❌ CUDA 12.2 libcuda non trovato"
echo "CUDA 11.8:"
ls -la /usr/local/cuda-11.8/lib64/libcuda* 2>/dev/null || echo "❌ CUDA 11.8 libcuda non trovato"

# 3. VERIFICA cuDNN
echo -e "\n📚 VERIFICA cuDNN..."
echo "Sistema cuDNN:"
ls -la /usr/lib64/libcudnn* | head -3
echo "CUDA 11.8 cuDNN:"
ls -la /usr/local/cuda-11.8/lib64/libcudnn* | head -3

# 4. TEST TENSORFLOW DETTAGLIATO
echo -e "\n🧪 TEST TENSORFLOW DETTAGLIATO..."

# Abilita logging massimo per debug
export TF_CPP_MIN_LOG_LEVEL=0
export CUDA_VISIBLE_DEVICES=0
export CUDA_HOME=/usr/local/cuda-11.8
export LD_LIBRARY_PATH=/usr/local/cuda-11.8/lib64:/usr/lib64:$LD_LIBRARY_PATH

python3 -c "
import os
import sys
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '0'  # Massimo logging
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
os.environ['CUDA_HOME'] = '/usr/local/cuda-11.8'

print('=== TENSORFLOW DEBUG DETTAGLIATO ===')
print('Python version:', sys.version)
print('CUDA_HOME:', os.environ.get('CUDA_HOME'))
print('LD_LIBRARY_PATH:', os.environ.get('LD_LIBRARY_PATH'))

try:
    import tensorflow as tf
    print('\\n✅ TensorFlow importato:', tf.__version__)
    
    # Verifica build info
    try:
        build_info = tf.sysconfig.get_build_info()
        print('\\n📋 TensorFlow Build Info:')
        for key, value in build_info.items():
            if 'cuda' in key.lower() or 'gpu' in key.lower():
                print(f'  {key}: {value}')
    except Exception as e:
        print('⚠️ Build info error:', e)
    
    # Test CUDA availability
    print('\\n🔍 CUDA Tests:')
    print('is_built_with_cuda():', tf.test.is_built_with_cuda())
    print('is_gpu_available():', tf.test.is_gpu_available())
    print('is_built_with_gpu_support():', tf.test.is_built_with_gpu_support())
    
    # Lista device fisici
    physical_devices = tf.config.list_physical_devices()
    print('\\n📱 All Physical Devices:')
    for device in physical_devices:
        print(f'  {device}')
    
    # Test specifico GPU
    gpu_devices = tf.config.list_physical_devices('GPU')
    print(f'\\n🎮 GPU Devices: {len(gpu_devices)}')
    
    if gpu_devices:
        for i, gpu in enumerate(gpu_devices):
            print(f'  GPU {i}: {gpu}')
            try:
                details = tf.config.experimental.get_device_details(gpu)
                print(f'    Details: {details}')
            except Exception as e:
                print(f'    Details error: {e}')
    
except ImportError as e:
    print('❌ TensorFlow import failed:', e)
except Exception as e:
    print('❌ TensorFlow error:', e)
    import traceback
    traceback.print_exc()
"

# 5. TEST LIBRERIE DLOPEN
echo -e "\n🔗 TEST DLOPEN LIBRERIE..."

python3 -c "
import ctypes
import os

os.environ['LD_LIBRARY_PATH'] = '/usr/local/cuda-11.8/lib64:/usr/lib64'

print('=== TEST DLOPEN LIBRERIE ===')

# Test librerie critiche
libraries = [
    '/usr/lib64/libcuda.so.1',
    '/usr/local/cuda-11.8/lib64/libcudart.so',
    '/usr/local/cuda-11.8/lib64/libcublas.so',
    '/usr/local/cuda-11.8/lib64/libcufft.so',
    '/usr/local/cuda-11.8/lib64/libcudnn.so'
]

for lib in libraries:
    try:
        handle = ctypes.CDLL(lib)
        print(f'✅ {lib}: OK')
    except Exception as e:
        print(f'❌ {lib}: {e}')
"

# 6. VERIFICA COMPATIBILITÀ VERSIONI
echo -e "\n🔄 VERIFICA COMPATIBILITÀ VERSIONI..."

echo "Compute Capability Tesla M60:"
nvidia-smi --query-gpu=compute_cap --format=csv,noheader,nounits

echo -e "\nVersioni installate:"
echo "Driver NVIDIA: $(nvidia-smi --query-gpu=driver_version --format=csv,noheader,nounits)"
echo "CUDA Runtime: $(nvcc --version | grep 'release' || echo 'N/A')"

python3 -c "
import tensorflow as tf
try:
    info = tf.sysconfig.get_build_info()
    print(f'TensorFlow CUDA: {info.get(\"cuda_version\", \"N/A\")}')
    print(f'TensorFlow cuDNN: {info.get(\"cudnn_version\", \"N/A\")}')
except:
    print('TensorFlow build info non disponibile')
"

# 7. POSSIBILI SOLUZIONI
echo -e "\n💡 POSSIBILI SOLUZIONI..."

echo "Basandoci sui risultati della diagnosi:"
echo ""
echo "OPZIONE 1: TensorFlow più vecchio (2.10.x)"
echo "  • pip3 install tensorflow==2.10.1"
echo "  • Supporto garantito Tesla M60 CC 5.2"
echo ""
echo "OPZIONE 2: TensorFlow con conda (librerie embedded)"
echo "  • conda install tensorflow-gpu=2.12"
echo "  • Include librerie CUDA ottimizzate"
echo ""
echo "OPZIONE 3: CPU-only con ottimizzazioni"
echo "  • Uso esclusivo CPU con parallelizzazione"
echo "  • Performance comunque accettabili per DDoS Detection"
echo ""
echo "OPZIONE 4: Build TensorFlow custom"
echo "  • Compilazione specifica per Tesla M60"
echo "  • Tempo richiesto: 2-3 ore"

# 8. RACCOMANDAZIONE FINALE
echo -e "\n🎯 RACCOMANDAZIONE..."

GPU_COUNT=$(python3 -c "import tensorflow as tf; print(len(tf.config.list_physical_devices('GPU')))" 2>/dev/null || echo "0")

if [ "$GPU_COUNT" = "0" ]; then
    echo "❌ GPU non rilevata - RACCOMANDAZIONE:"
    echo ""
    echo "🚀 SOLUZIONE IMMEDIATA: TensorFlow 2.10.1"
    echo "  pip3 uninstall tensorflow"
    echo "  pip3 install tensorflow==2.10.1"
    echo "  # TF 2.10.1 ha miglior supporto hardware legacy"
    echo ""
    echo "📊 ALTERNATIVE:"
    echo "  • CPU-only: Performance 75K record/sec (accettabile)"
    echo "  • Upgrade hardware: GPU moderna (RTX/Tesla P100+)"
    echo "  • Cloud GPU: AWS/GCP Tesla V100/A100"
else
    echo "✅ GPU rilevata - sistema funzionante!"
fi

echo -e "\n🔚 DIAGNOSI COMPLETATA"