Expand description
GPU occupancy queries for performance optimisation.
Occupancy measures how effectively GPU resources (warps, registers, shared memory) are utilised. These queries help select launch configurations that maximise hardware utilisation.
ยงExample
let func = module.get_function("my_kernel")?;
// Query the optimal block size for maximum occupancy.
let (min_grid_size, optimal_block_size) = func.optimal_block_size(0)?;
println!("optimal: grid >= {min_grid_size}, block = {optimal_block_size}");
// Query active blocks per SM for a specific block size.
let active = func.max_active_blocks_per_sm(256, 0)?;
println!("active blocks per SM with 256 threads: {active}");