0Pricing
CUDA Academy · Lesson

Register Pressure and Spills

Balancing reuse against occupancy.

Registers Are Precious

Registers are the fastest storage a thread has, but each SM holds only so many. How heavily a kernel uses them is called register pressure.

A Shared, Fixed Pool

Every resident thread draws from one register file per SM. The more registers each thread needs, the fewer threads can stay resident together.

All lessons in this course

  1. Instruction-Level Parallelism
  2. Loop Unrolling with #pragma unroll
  3. Vectorized Loads with float4
  4. Register Pressure and Spills
← Back to CUDA Academy