voxcity.simulator_gpu.solar.reflection¶

Optimized Radiation Computation for GPU

This module provides optimized GPU kernels for radiation computation that minimize kernel launches and synchronization overhead.

Key optimizations: 1. Fused kernels - combine multiple operations into single kernel launches 2. Reduced synchronization - batch operations to minimize ti.sync() calls 3. Better memory access patterns - coalesced memory access 4. Reduced atomic operations - use local accumulation where possible

Attributes¶

`Vector3`
`gpu_times`

Classes¶

OptimizedReflectionSolver

Optimized GPU solver for multi-bounce radiation reflections.

Functions¶

`fused_reflection_step_kernel`(surfins_in, surfins_out, ...)	Single fused kernel for one reflection step.
`compute_initial_and_reflections_fused`(surf_direction, ...)	Fully fused kernel: initial radiation + all reflection iterations.
`benchmark_reflections`()	Benchmark the reflection solver.

Module Contents¶

voxcity.simulator_gpu.solar.reflection.Vector3¶

voxcity.simulator_gpu.solar.reflection.fused_reflection_step_kernel(surfins_in: ti.template(), surfins_out: ti.template(), surfout: ti.template(), albedo: ti.template(), svf: ti.template(), total_incoming: ti.template(), total_outgoing: ti.template(), svf_source: ti.template(), svf_target: ti.template(), svf_vf: ti.template(), svf_trans: ti.template(), svf_nnz: taichi.i32, n_surfaces: taichi.i32)¶

Single fused kernel for one reflection step.

Combines: outgoing computation + distribution + accumulation into fewer synchronization points.

voxcity.simulator_gpu.solar.reflection.compute_initial_and_reflections_fused(surf_direction: ti.template(), surf_svf: ti.template(), surf_shadow: ti.template(), surf_canopy_trans: ti.template(), surf_albedo: ti.template(), surf_normal: ti.template(), sun_dir_x: taichi.f32, sun_dir_y: taichi.f32, sun_dir_z: taichi.f32, cos_zenith: taichi.f32, sw_direct: taichi.f32, sw_diffuse: taichi.f32, svf_source: ti.template(), svf_target: ti.template(), svf_vf: ti.template(), svf_trans: ti.template(), svf_nnz: taichi.i32, n_surfaces: taichi.i32, n_ref_steps: taichi.i32, sw_in_direct: ti.template(), sw_in_diffuse: ti.template(), sw_in_reflected: ti.template(), sw_out_total: ti.template(), surfins_a: ti.template(), surfins_b: ti.template(), surfout: ti.template())¶

Fully fused kernel: initial radiation + all reflection iterations.

This is the most optimized version that runs everything in one kernel.

voxcity.simulator_gpu.solar.reflection.benchmark_reflections()¶: Benchmark the reflection solver.

voxcity.simulator_gpu.solar.reflection.gpu_times¶