ad762e682d
- Suppression de generate_solar (éclairage solaire) des visualisations - Accélération GPU de hillshade, slope, aspect, curvature, depressions, anomalies, roughness, texture GLCM, flow (sink filling) - Nettoyage mémoire GPU entre visualisations (gpu_cleanup) - Correction OOM texture GLCM: calcul entropie bin par bin au lieu d'un tableau 3D massif sur GPU - Correction bug: xp_minimum_filter manquant dans imports visualizations - Option --file accepte plusieurs noms complets sans extension - run.sh affiche l'aide si appelé sans arguments - Option --test pour exécuter les tests unitaires dans Docker - Filtre ReturnNumber>=1 intégré dans le pipeline PDAL (plus d'erreur SMRF) - 60 tests unitaires: GPU, visualisations, rendering, DTM, pipeline, CLI - Ajout pytest au Dockerfile Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
79 lines
2.4 KiB
Python
79 lines
2.4 KiB
Python
"""GPU acceleration helpers for LiDAR pipeline.
|
|
|
|
Provides CuPy/numpy abstraction layer. If CuPy is available and a CUDA GPU
|
|
is detected, array operations are accelerated on the GPU. Otherwise, all
|
|
operations fall back to numpy/scipy on CPU.
|
|
"""
|
|
|
|
import logging
|
|
import numpy as np
|
|
from scipy import ndimage
|
|
|
|
logger = logging.getLogger("lidar")
|
|
|
|
# GPU detection - must happen at import time
|
|
HAS_GPU = False
|
|
_gpu_name = None
|
|
_gpu_mem_gb = 0
|
|
_xp = np # Default: CPU
|
|
|
|
try:
|
|
import cupy as cp
|
|
import cupyx.scipy.ndimage as cp_ndimage
|
|
|
|
_gpu_info = cp.cuda.runtime.getDeviceProperties(0)
|
|
_gpu_name = _gpu_info['name'].decode() if isinstance(_gpu_info['name'], bytes) else str(_gpu_info['name'])
|
|
_gpu_mem_gb = _gpu_info['totalGlobalMem'] // (1024 ** 3)
|
|
HAS_GPU = True
|
|
_xp = cp
|
|
except (ImportError, Exception):
|
|
pass
|
|
|
|
|
|
def log_gpu_status():
|
|
"""Log GPU detection result. Called after logging is configured."""
|
|
if HAS_GPU:
|
|
logger.info(f"GPU détectée: {_gpu_name} ({_gpu_mem_gb} Go VRAM)")
|
|
else:
|
|
logger.info("Pas de GPU — mode CPU uniquement")
|
|
|
|
|
|
def to_gpu(arr):
|
|
"""Send array to GPU if available, otherwise return as float64 numpy."""
|
|
if HAS_GPU:
|
|
return cp.asarray(arr.astype(np.float64))
|
|
return arr.astype(np.float64)
|
|
|
|
|
|
def to_cpu(arr):
|
|
"""Bring array back to CPU (numpy). No-op if already on CPU."""
|
|
if HAS_GPU and isinstance(arr, cp.ndarray):
|
|
return cp.asnumpy(arr)
|
|
return arr
|
|
|
|
|
|
def xp_gaussian_filter(arr, sigma):
|
|
"""Gaussian filter — uses GPU if array is on GPU, CPU otherwise."""
|
|
if HAS_GPU and isinstance(arr, cp.ndarray):
|
|
return cp_ndimage.gaussian_filter(arr, sigma)
|
|
return ndimage.gaussian_filter(arr, sigma)
|
|
|
|
|
|
def xp_uniform_filter(arr, size):
|
|
"""Uniform filter — uses GPU if array is on GPU, CPU otherwise."""
|
|
if HAS_GPU and isinstance(arr, cp.ndarray):
|
|
return cp_ndimage.uniform_filter(arr, size)
|
|
return ndimage.uniform_filter(arr, size)
|
|
|
|
|
|
def xp_minimum_filter(arr, footprint=None, size=None):
|
|
"""Minimum filter — uses GPU if array is on GPU, CPU otherwise."""
|
|
if HAS_GPU and isinstance(arr, cp.ndarray):
|
|
return cp_ndimage.minimum_filter(arr, footprint=footprint, size=size)
|
|
return ndimage.minimum_filter(arr, footprint=footprint, size=size)
|
|
|
|
|
|
def gpu_cleanup():
|
|
"""Free GPU memory. Call between visualizations to prevent OOM."""
|
|
if HAS_GPU:
|
|
cp.get_default_memory_pool().free_all_blocks() |