Software Pipeline Depth

Problem

Choose the software pipeline depth that fits both register and shared-memory budgets: min(max_stages, registers_per_thread // regs_per_stage, smem_bytes // stage_smem_bytes), clamped to at least 1.

The signature is software_pipeline_depth(registers_per_thread, regs_per_stage, max_stages, smem_bytes, stage_smem_bytes). Return an integer.

Hint

Deeper pipelines hide more global-memory latency but consume registers and shared memory.

Test cases

case 1

in: [255,64,4,100000,8192]

expected: 3

case 2

in: [128,64,8,100000,8192]

expected: 2

case 3

in: [64,64,4,1000,10000]

expected: 1

case 4

in: [512,32,6,65536,8192]

expected: 6

Starter codePython

def software_pipeline_depth(registers_per_thread, regs_per_stage, max_stages, smem_bytes, stage_smem_bytes):
    # Your code here
    pass

4 test cases · run your code against them in the editor