diffct 2.0.0__py3-none-any.whl

diffct/constants.py

@@ -0,0 +1,72 @@
+"""Global constants and configuration for diffct package.
+
+This module defines core constants used throughout the diffct package,
+including data types, CUDA thread block configurations, and numerical
+precision parameters.
+"""
+
+import numpy as np
+from numba import cuda
+
+# ---------------------------------------------------------------------------
+# Data Types and Numerical Constants
+# ---------------------------------------------------------------------------
+
+_DTYPE = np.float32
+"""Default data type for numerical computations (numpy.float32)."""
+
+_INF = _DTYPE(np.inf)
+"""Floating-point infinity in default data type."""
+
+_NEG_INF = _DTYPE(-np.inf)
+"""Negative floating-point infinity in default data type."""
+
+_BIG = _DTYPE(1e30)
+"""Finite stand-in for infinity in fastmath kernels, where LLVM may assume no inf occurs."""
+
+_TINY = _DTYPE(1e-20)
+"""Smallest ray direction component that the Siddon kernels divide by; ray parameters then stay below _BIG."""
+
+_ZERO = _DTYPE(0.0)
+"""Floating-point zero in default data type."""
+
+_ONE = _DTYPE(1.0)
+"""Floating-point one in default data type."""
+
+_HALF = _DTYPE(0.5)
+"""Floating-point one half in default data type."""
+
+_EPSILON = _DTYPE(1e-6)
+"""Small epsilon value for numerical comparisons to avoid division by zero."""
+
+# ---------------------------------------------------------------------------
+# CUDA Thread Block Configurations
+# ---------------------------------------------------------------------------
+
+# CUDA thread block configurations optimized for different dimensionalities
+# 2D blocks: 16x16 = 256 threads per block, optimal for 2D ray-tracing kernels
+# Balances occupancy with shared memory usage for parallel/fan beam projections
+_TPB_2D = (16, 16)
+"""CUDA threads-per-block for 2D kernels (parallel/fan beam): (16, 16) = 256 threads."""
+
+# 3D blocks: 8x8x8 = 512 threads per block, optimal for 3D cone beam kernels
+# Smaller per-dimension size accommodates higher register usage in 3D algorithms
+_TPB_3D = (8, 8, 8)
+"""CUDA threads-per-block for 3D kernels (cone beam): (8, 8, 8) = 512 threads."""
+
+# ---------------------------------------------------------------------------
+# CUDA JIT Decorators
+# ---------------------------------------------------------------------------
+
+# CUDA fastmath optimization: enables aggressive floating-point optimizations
+# Trades numerical precision for performance in ray-tracing calculations
+# Safe for CT reconstruction where slight precision loss is acceptable for speed gains
+_FASTMATH_DECORATOR = cuda.jit(cache=True, fastmath=True)
+"""Numba CUDA JIT decorator with fastmath enabled for forward kernels."""
+
+# Analytical reconstruction kernels (FBP / FDK voxel-driven gather) run at
+# the tightest available precision - the Fourier-convention constant and
+# the (sid/U)^2 weight amplify rounding errors, and the kernels are only
+# called once per reconstruction so the speed penalty is negligible.
+_FDK_ACCURACY_DECORATOR = cuda.jit(cache=True, fastmath=False)
+"""Numba CUDA JIT decorator with fastmath disabled for FBP/FDK gather kernels."""

diffct/differentiable.py

@@ -0,0 +1,74 @@
+"""Backward compatibility module for diffct.differentiable.
+
+This module provides backward compatibility by re-exporting all functions
+and classes from the refactored modular structure. New code should import
+directly from the specific modules (projectors, geometry, etc.) or from
+the top-level diffct package.
+
+DEPRECATED: This module exists only for backward compatibility.
+Prefer importing from 'diffct' directly:
+    from diffct import ConeProjectorFunction, circular_trajectory_3d
+"""
+
+import warnings
+
+# Show deprecation warning when this module is imported
+warnings.warn(
+    "Importing from 'diffct.differentiable' is deprecated. "
+    "Please import from 'diffct' directly, e.g., 'from diffct import ConeProjectorFunction'",
+    DeprecationWarning,
+    stacklevel=2
+)
+
+# Re-export all public API from refactored modules
+from .projectors import (
+    ParallelProjectorFunction,
+    ParallelBackprojectorFunction,
+    FanProjectorFunction,
+    FanBackprojectorFunction,
+    ConeProjectorFunction,
+    ConeBackprojectorFunction,
+)
+
+from .geometry import (
+    circular_trajectory_3d,
+    random_trajectory_3d,
+)
+
+# Also export utility functions that were previously accessible
+from .utils import (
+    DeviceManager,
+    TorchCUDABridge,
+)
+
+from .constants import (
+    _DTYPE,
+    _TPB_2D,
+    _TPB_3D,
+    _FASTMATH_DECORATOR,
+    _INF,
+    _EPSILON,
+)
+
+__all__ = [
+    # Projector Functions
+    'ParallelProjectorFunction',
+    'ParallelBackprojectorFunction',
+    'FanProjectorFunction',
+    'FanBackprojectorFunction',
+    'ConeProjectorFunction',
+    'ConeBackprojectorFunction',
+    # Geometry Functions
+    'circular_trajectory_3d',
+    'random_trajectory_3d',
+    # Utility Classes
+    'DeviceManager',
+    'TorchCUDABridge',
+    # Constants (for advanced users)
+    '_DTYPE',
+    '_TPB_2D',
+    '_TPB_3D',
+    '_FASTMATH_DECORATOR',
+    '_INF',
+    '_EPSILON',
+]