Add Python bindings for CUDA cluster finder

- Add bind_ClusterFinderCUDA.hpp with pybind11 bindings for ClusterFinderCUDA - Build CUDA bindings as separate _aare_cuda.so to avoid segfaults from mixing nvcc and gcc compiled code in the same shared object - Re-export CUDA classes onto _aare in __init__.py so user code uses `from aare import ClusterFinderCUDA` regardless of which .so hosts the class - Factory in ClusterFinder.py selects backend; RuntimeError if GPU requested on CPU-only build - Update python/CMakeLists.txt: _aare_cuda module gated behind AARE_CUDA and AARE_PYTHON_BINDINGS - Add validation notebook: ~20x speedup vs sequential ClusterFinder
2026-10-31 20:51:27 +01:00 · 2026-04-23 11:43:40 +02:00
1 parent 3ed773e520
commit e894bdac9b
@@ -0,0 +1,422 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "789d5aab-0c75-4ed3-94bf-e66172f6737c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import sys; sys.path.append('/home/ferjao_k/aare/build')\n",
+    "\n",
+    "from pathlib import Path\n",
+    "import matplotlib.pyplot as plt\n",
+    "from mpl_toolkits.axes_grid1 import make_axes_locatable\n",
+    "import numpy as np\n",
+    "import boost_histogram as bh\n",
+    "import time\n",
+    "\n",
+    "from aare import File, ClusterFinder, ClusterFinderMT, ClusterCollector, ClusterFinderCUDA"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "3f6089d6-2245-4aca-aad3-2417c16afae2",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def make_hist(clusters):\n",
+    "    h = bh.Histogram(bh.axis.Regular(100, -2, 4000))\n",
+    "    h.fill(clusters.sum())\n",
+    "    return h"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "415b09d4-d0d0-4166-8601-9d84302617bd",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Image size:       (400, 400)\n",
+      "Pedestal frames:  1000\n",
+      "Data frames:      40000\n"
+     ]
+    }
+   ],
+   "source": [
+    "base = Path('/mnt/sls_det_storage/matterhorn_data/aare_test_data/')\n",
+    "f = File(base / 'Moench03new/cu_half_speed_master_4.json')\n",
+    "\n",
+    "n_frames_pd = 1000\n",
+    "N           = 40000\n",
+    "cluster_size = (3, 3)\n",
+    "image_size   = (f.rows, f.cols)\n",
+    "capacity     = 100_000 #3_000_000\n",
+    "\n",
+    "print(f'Image size:       {image_size}')\n",
+    "print(f'Pedestal frames:  {n_frames_pd}')\n",
+    "print(f'Data frames:      {N}')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "5531b79d-12ad-4de6-84b7-0224b10d13c3",
+   "metadata": {},
+   "source": [
+    "## Pedestal (both finders trained on identical frames)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "e33b421f-969a-41d6-8aa6-5bd859a88fb3",
+   "metadata": {},
+   "source": [
+    "- Modify the boolean `SERIAL` to choose between the sequential CPU version (ClusterFinder) and its multi-threaded homologue (ClusterFinderMT)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "6ee7c469-5a23-421a-9bb4-a76e8f57a0c1",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "SERIAL = True"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "9d8f69bf-27f6-48c8-b2da-4157d67e6b04",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "if(SERIAL):\n",
+    "    cf_cpu  = ClusterFinder(image_size, cluster_size, capacity=capacity)\n",
+    "else:\n",
+    "    cf_cpu  = ClusterFinderMT(image_size, cluster_size, capacity=capacity, n_threads=24)\n",
+    "    sink = ClusterCollector(cf_cpu)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "cbdcb805-708b-4205-bda9-2aa163d0e81f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "N_STREAMS  = 1 \n",
+    "cf_cuda = ClusterFinderCUDA(image_size, cluster_size, capacity=capacity, n_streams=N_STREAMS)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "id": "1546f405-1bf6-4073-ab35-8134be695a6c",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Pedestal (1000 frames): 0.498s\n"
+     ]
+    }
+   ],
+   "source": [
+    "t0 = time.perf_counter()\n",
+    "for _ in range(n_frames_pd):\n",
+    "    img = f.read_frame()\n",
+    "    cf_cpu.push_pedestal_frame(img.copy())\n",
+    "    cf_cuda.push_pedestal_frame(img.copy())\n",
+    "print(f'Pedestal ({n_frames_pd} frames): {time.perf_counter() - t0:.3f}s')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "5df035f0-7a27-4d5a-8f8d-c94e745bcd1a",
+   "metadata": {},
+   "source": [
+    "## Read all data frames into memory (I/O out of the timing loop)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "id": "4573be3d-5ba8-4e18-bab0-c874f2b7dcb2",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Reading 40000 frames:        2.002s  (19985 FPS, 6098.804 GB/s)\n"
+     ]
+    }
+   ],
+   "source": [
+    "f.seek(n_frames_pd)\n",
+    "t0 = time.perf_counter()\n",
+    "data = f.read_n(N)\n",
+    "t_io = time.perf_counter() - t0\n",
+    "print(f'Reading {N} frames:        {t_io:.3f}s  ({N/t_io:.0f} FPS, '\n",
+    "      f'{f.bytes_per_frame * N / 1024**2 / t_io:.3f} GB/s)')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "c1e9362e-dcc2-41af-bfa1-c62d0968477e",
+   "metadata": {},
+   "source": [
+    "## CPU clustering"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "id": "fbb14fda-2852-4b73-ba2c-9952abac1d99",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "CPU clustering:          73.978s (541 FPS, 55449596 clusters, 1386.24/frame)\n"
+     ]
+    }
+   ],
+   "source": [
+    "t0 = time.perf_counter()\n",
+    "for frame in data:\n",
+    "    cf_cpu.find_clusters(frame)\n",
+    "t_cpu = time.perf_counter() - t0\n",
+    "\n",
+    "if(SERIAL):\n",
+    "    clusters_cpu = cf_cpu.steal_clusters(realloc_same_capacity=False)\n",
+    "    n_clusters_cpu = clusters_cpu.size\n",
+    "    \n",
+    "    hist_cpu  = make_hist(clusters_cpu)\n",
+    "else:\n",
+    "    cf_cpu.stop()\n",
+    "    sink.stop()\n",
+    "    \n",
+    "    clusters_cpu = sink.steal_clusters() #cf_cpu.steal_clusters(realloc_same_capacity=False)\n",
+    "    \n",
+    "    hist_cpu = bh.Histogram(bh.axis.Regular(100, -2, 4000))\n",
+    "    n_clusters_cpu = 0\n",
+    "    for cv in clusters_cpu:\n",
+    "        hist_cpu.fill(cv.sum())\n",
+    "        n_clusters_cpu += cv.size\n",
+    "        \n",
+    "print(f'CPU clustering:          {t_cpu:.3f}s ({N/t_cpu:.0f} FPS, '\n",
+    "      f'{n_clusters_cpu} clusters, {n_clusters_cpu/N:.2f}/frame)')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "8cfd7091-8020-49ff-b033-28bf773983d8",
+   "metadata": {},
+   "source": [
+    "## CUDA clustering"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "id": "4b8df93b-9a1b-41a5-9fed-9cda295fb523",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "CUDA clustering:          3.880s  (10310 FPS, 55810704 clusters, 1395.27/frame)\n",
+      "Speedup (CPU / CUDA):     19.07×\n"
+     ]
+    }
+   ],
+   "source": [
+    "# Warmup: first kernel launch pays CUDA context + pedestal H2D upload cost\n",
+    "cf_cuda.find_clusters(data[0])\n",
+    "_ = cf_cuda.steal_clusters(realloc_same_capacity=False)\n",
+    "\n",
+    "t0 = time.perf_counter()\n",
+    "for frame in data:\n",
+    "    cf_cuda.find_clusters(frame)\n",
+    "t_cuda = time.perf_counter() - t0\n",
+    "clusters_cuda = cf_cuda.steal_clusters(realloc_same_capacity=False)\n",
+    "n_clusters_cuda = clusters_cuda.size\n",
+    "print(f'CUDA clustering:          {t_cuda:.3f}s  ({N/t_cuda:.0f} FPS, '\n",
+    "      f'{n_clusters_cuda} clusters, {n_clusters_cuda/N:.2f}/frame)')\n",
+    "print(f'Speedup (CPU / CUDA):     {t_cpu / t_cuda:.2f}×')\n",
+    "\n",
+    "hist_cuda = make_hist(clusters_cuda)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "id": "1fda99ee-db54-40c1-8e5b-51168a33fb96",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# BATCH_SIZE = 500\n",
+    "\n",
+    "# # warmup\n",
+    "# _ = cf_cuda.find_clusters_batched(data[:1], first_frame=0)\n",
+    "\n",
+    "# t0 = time.perf_counter()\n",
+    "# clusters_cuda_per_frame = []\n",
+    "# for start in range(0, N, BATCH_SIZE):\n",
+    "#     stop = min(start + BATCH_SIZE, N)\n",
+    "#     clusters_cuda_per_frame.extend(\n",
+    "#         cf_cuda.find_clusters_batched(data[start:stop], first_frame=start)\n",
+    "#     )\n",
+    "# t_cuda = time.perf_counter() - t0\n",
+    "\n",
+    "# n_clusters_cuda = sum(cv.size for cv in clusters_cuda_per_frame)\n",
+    "\n",
+    "# print(f'CUDA clustering:          {t_cuda:.3f}s  ({N/t_cuda:.0f} FPS, '\n",
+    "#       f'{n_clusters_cuda} clusters, {n_clusters_cuda/N:.2f}/frame)')\n",
+    "# print(f'Speedup (CPU / CUDA):     {t_cpu / t_cuda:.2f}×')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "id": "b7df7b9d-6873-46c4-b6bc-7e4238a5bdf7",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# def make_hist_from_batch(result_list):\n",
+    "#     h = bh.Histogram(bh.axis.Regular(100, -2, 4000))\n",
+    "#     energies = [np.asarray(cv.sum()).ravel() for cv in result_list if cv.size > 0]\n",
+    "#     if energies:\n",
+    "#         h.fill(np.concatenate(energies))\n",
+    "#     return h\n",
+    "\n",
+    "# hist_cuda = make_hist_from_batch(clusters_cuda_per_frame)"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "b966cce1-0f73-4565-a707-1aec2a69f75e",
+   "metadata": {},
+   "source": [
+    "## Agreement check: \n",
+    "- Cluster counts should match closely.\n",
+    "- However, as the CUDA CF updates the pedestal once per frame rather than per-pixel, a small divergence after the first few frames is expected."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "id": "d3a850df-7df0-485b-971d-381cfdc6be81",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Cluster count diff:       361108 (0.65%)\n"
+     ]
+    }
+   ],
+   "source": [
+    "diff = abs(n_clusters_cpu - n_clusters_cuda)\n",
+    "rel  = diff / max(n_clusters_cpu, 1)\n",
+    "print(f'Cluster count diff:       {diff} ({rel:.2%})')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "81638ad3-6112-4bd7-a93a-f4d97f1f94cf",
+   "metadata": {},
+   "source": [
+    "## Plots"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "id": "9adeea2f-9309-4c0a-905b-7d1203ba1f4e",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "image/png": "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 truncated
+      "text/plain": [
+       "<Figure size 900x600 with 2 Axes>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "fig, (ax_spec, ax_ratio) = plt.subplots(\n",
+    "    2, 1, figsize=(9, 6), sharex=True,\n",
+    "    gridspec_kw={'height_ratios': [3, 1]}\n",
+    ")\n",
+    "\n",
+    "edges = hist_cpu.axes[0].edges\n",
+    "cpu_vals = hist_cpu.values()\n",
+    "cuda_vals = hist_cuda.values()\n",
+    "\n",
+    "ax_spec.stairs(cpu_vals, edges, label=f'CPU  ({n_clusters_cpu} clusters)')\n",
+    "ax_spec.stairs(cuda_vals, edges, label=f'CUDA ({n_clusters_cuda} clusters)', linestyle='--')\n",
+    "ax_spec.set_ylabel('Counts')\n",
+    "ax_spec.set_title('Cluster energy spectrum: CPU vs CUDA')\n",
+    "ax_spec.legend()\n",
+    "ax_spec.grid(alpha=0.3)\n",
+    "\n",
+    "with np.errstate(divide='ignore', invalid='ignore'):\n",
+    "    ratio = np.where(cpu_vals > 0, cuda_vals / cpu_vals, np.nan)\n",
+    "\n",
+    "ax_ratio.stairs(ratio, edges, color='k')\n",
+    "ax_ratio.axhline(1.0, color='gray', linewidth=0.5)\n",
+    "ax_ratio.set_ylabel('CUDA / CPU')\n",
+    "ax_ratio.set_xlabel('Energy [ADU]')\n",
+    "ax_ratio.set_ylim(0.5, 3.5)\n",
+    "ax_ratio.grid(alpha=0.3)\n",
+    "\n",
+    "plt.tight_layout()\n",
+    "plt.show()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a6ca4581-bb29-4c74-a34a-bd427551a39b",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.11.15"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}