Alternating branch loop¶

Generated workload catalogue · Execution contract

Purpose¶

Exercise a repeating taken/not-taken conditional outcome with a small instruction footprint.

Setup¶

Property	Value
Workload ID	`alternating`
Selection	`--workload alternating`
Generator	`legacyProgram() in RV32IWorkloads.cpp`
End condition	EBREAK after the loop

x1 = iterations; x2 and x3 start at zero. No data memory input.

N means the requested iteration count. Registers start from reset; the loop's x2 is not an ABI stack pointer.

Implementation¶

Toggle the low bit of x2. BNE skips an increment of x3 when x2 is nonzero; a separate loop branch decrements x1.

x2 ^= 1;
if (x2 == 0) ++x3;
--x1;

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

Expected result¶

x2=N%2, x3=floor(N/2), x1=0. For N=12: x2=0, x3=6. The first inner branch is taken.

The runner validates the executed architectural state with the instruction oracle. This legacy generator does not publish a separate expected_result checksum; the equations above describe its loop.

Measurement and running¶

There are two static branch sites, with different outcome sequences. Compare policies on identical iterations; do not treat every branch as alternating.

build/rv32i_experiment --core advanced --workload alternating \
  --iterations 2  \
  --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/RV32IWorkloads.cpp: emitted instructions.
tests/checks/RV32IExperimentCLITest.py: legacy loops across all three cores.
tests/checks/RV32IBenchmarkCLITest.py and tests/checks/BenchmarkCampaignCLITest.py: shared execution/campaign checks.