Benchmark Configurations

The scripts/benchmarks/ directory contains microarchitecture-specific configurations modeled after real hardware. These are used by the analysis scripts (particularly top_down.py) and can serve as starting points for your own experiments.

Available Configurations

SiFive Performance P550

Based on published microarchitecture analysis (Chips and Cheese, SiFive specs).

from scripts.benchmarks.p550.config import p550_config
config = p550_config()

Parameter	Value	Notes
Width	3	Triple-issue
Backend	OutOfOrder	72-entry ROB, 32-entry IQ
Branch Predictor	Tournament	9.1 KiB budget
L1D	32KB, 8-way, 3cy	8 MSHRs, stride prefetch
L2	256KB, 8-way, 10cy	16 MSHRs
L3	4MB, 16-way, 30cy	32 MSHRs
Memory	DRAM	t_cas=14, row_miss=120

ARM Cortex-A72

Based on publicly documented microarchitecture.

from scripts.benchmarks.cortex_a72.config import cortex_a72_config
config = cortex_a72_config()

Parameter	Value	Notes
Width	3	Triple-issue
Backend	OutOfOrder	128-entry ROB, 60-entry IQ
Branch Predictor	TAGE	4 banks, 2048-entry tables
L1I	48KB, 3-way	NextLine prefetch
L1D	32KB, 2-way	8 MSHRs, stride prefetch
L2	1MB, 16-way, 12cy	16 MSHRs

Apple M1

Modeled after Apple's Firestorm (performance) core.

from scripts.benchmarks.m1.config import m1_config
config = m1_config()

Parameter	Value	Notes
Width	4	Quad-issue
Branch Predictor	TAGE	4 banks, 4096-entry tables
L1I	128KB, 8-way	NextLine prefetch (degree=2)
L1D	128KB, 8-way	12 MSHRs, stride prefetch (degree=2)
L2	4MB, 16-way, 12cy	32 MSHRs

Using Benchmark Configs

With Environment

from scripts.benchmarks.p550.config import p550_config
from rvsim import Environment

result = Environment(
    binary="software/bin/programs/qsort.elf",
    config=p550_config(),
).run()

print(result.stats.query("ipc|stall|miss"))

With Sweep

from scripts.benchmarks.p550.config import p550_config
from scripts.benchmarks.cortex_a72.config import cortex_a72_config
from scripts.benchmarks.m1.config import m1_config
from rvsim import Sweep

results = Sweep(
    binaries=["software/bin/programs/qsort.elf"],
    configs={
        "P550": p550_config(),
        "A72": cortex_a72_config(),
        "M1": m1_config(),
    },
).run(parallel=True)

results.compare(metrics=["ipc", "cycles", "dcache_misses", "branch_accuracy_pct"])

Deriving Variants

Use replace() to create variants of a benchmark config:

base = p550_config()
wide = base.replace(width=4)
big_cache = base.replace(l1d=Cache("64KB", ways=8, latency=3, mshr_count=8))

Creating Your Own Config

Follow the pattern in the existing configs. A config file should:

1. Define a function that returns a Config object
2. Assign it to a module-level config variable (for Simulator.config("path/to/config.py") support)

"""My custom machine config."""
from rvsim import Config, Cache, Backend, BranchPredictor, Fu

def my_config():
    return Config(
        width=2,
        backend=Backend.OutOfOrder(
            rob_size=64,
            issue_queue_size=24,
            fu_config=Fu([
                Fu.IntAlu(count=2, latency=1),
                Fu.IntMul(count=1, latency=3),
                Fu.Branch(count=1, latency=1),
                Fu.Mem(count=1, latency=1),
            ]),
        ),
        branch_predictor=BranchPredictor.GShare(),
        l1d=Cache("16KB", ways=4, latency=1, mshr_count=4),
        l2=Cache("128KB", ways=8, latency=8),
    )

config = my_config