Scheduled load-use pairs¶

Generated workload catalogue · Execution contract

Purpose¶

Test whether one independent instruction between load and use hides a pipeline load-use bubble.

Setup¶

Property	Value
Workload ID	`load-scheduled`
Selection	`--workload load-scheduled`
Generator	`buildRV32IBenchmark() in RV32IBenchmarks.cpp`
End condition	EBREAK after the loop and result store

The same n seeded words and initial registers as load-use.

Use the shared address map and parameter bounds. N includes any training iterations; n means size.

Implementation¶

For every word, issue LW, increment independent x7, then add the loaded x5 to x10. No extra operation is inserted.

for (auto v : input) { ++independent_counter; sum += v; }

This is a behavioral summary of the generated assembly, not the host program being timed.

Expected result¶

Result = Nsum(input) modulo 2^32; x7=Nn, exactly as in load-use.

The runner checks the generator's expected result in addition to retirement-by-retirement oracle validation. Output-array assertions, where described, check more than a checksum.

Measurement and running¶

The advanced pipeline should need fewer load-use bubbles when this spacing suffices. Compare observed LOAD_USE counts as well as cycles; memory wait states can change the benefit.

build/rv32i_experiment --core advanced --workload load-scheduled \
  --iterations 2 --size 16 --seed 1  \
  --fidelity behavioral --period 4000 \
  --allocation-limit-mib 512 --max-cycles 4096

The explicit period is for a bounded functional check, not a hardware-speed result. For comparisons, use the serial campaign driver, identical case inputs, and a certified per-target clock.

Source and tests¶

src/analysis/RV32IBenchmarks.cpp: emitted instructions and independent expected results.
tests/cpp/RV32IBenchmarkTest.cpp: generator results, parameter bounds, and edge cases.
tests/checks/RV32IBenchmarkCLITest.py and tests/checks/BenchmarkCampaignCLITest.py: shared execution/campaign checks.