From d19255d81892dea491b990e17aa5237bd1b82087 Mon Sep 17 00:00:00 2001
From: Yaman Umuroglu <maltanar@gmail.com>
Date: Wed, 1 Feb 2023 11:32:17 +0100
Subject: [PATCH] [Build] calculate stable_throughput metric as part of
step_measure_rtlsim_performance
---
src/finn/builder/build_dataflow_steps.py | 20 +++++++++++++++++---
1 file changed, 17 insertions(+), 3 deletions(-)
diff --git a/src/finn/builder/build_dataflow_steps.py b/src/finn/builder/build_dataflow_steps.py
index b0f7b6ec6..fb4d60c1e 100644
--- a/src/finn/builder/build_dataflow_steps.py
+++ b/src/finn/builder/build_dataflow_steps.py
@@ -667,7 +667,6 @@ def step_measure_rtlsim_performance(model: ModelWrapper, cfg: DataflowBuildConfi
)
rtlsim_bs = int(cfg.rtlsim_batch_size)
if force_python_rtlsim:
- # run with single input to get latency
orig_rtlsim_trace_depth = get_rtlsim_trace_depth()
assert rtlsim_bs > 0, "rtlsim batch size must be >0"
if cfg.verify_save_rtlsim_waveforms:
@@ -680,9 +679,11 @@ def step_measure_rtlsim_performance(model: ModelWrapper, cfg: DataflowBuildConfi
rtlsim_model.set_metadata_prop(
"extra_verilator_args", str(["-CFLAGS", "-O3"])
)
+ # run with single input to get latency
+ rtlsim_latency_dict = throughput_test_rtlsim(rtlsim_model, 1)
+ # run with batch to get stable-state throughput
rtlsim_perf_dict = throughput_test_rtlsim(rtlsim_model, rtlsim_bs)
- rtlsim_latency = rtlsim_perf_dict["cycles"]
- rtlsim_perf_dict["latency_cycles"] = rtlsim_latency
+ rtlsim_perf_dict["latency_cycles"] = rtlsim_latency_dict["cycles"]
else:
rtlsim_perf_dict = verilator_fifosim(model, rtlsim_bs)
# keep keys consistent between the Python and C++-styles
@@ -696,6 +697,19 @@ def step_measure_rtlsim_performance(model: ModelWrapper, cfg: DataflowBuildConfi
for (key, val) in rtlsim_perf_dict.items():
if "max_count" in key:
del rtlsim_perf_dict[key]
+ # estimate stable-state throughput based on latency+throughput
+ if rtlsim_bs == 1:
+ rtlsim_perf_dict["stable_throughput[images/s]"] = rtlsim_perf_dict[
+ "throughput[images/s]"
+ ]
+ else:
+ total_cycles = rtlsim_perf_dict["cycles"]
+ latency_cycles = rtlsim_perf_dict["latency_cycles"]
+ stablestate_cycles = total_cycles - latency_cycles
+ clk_ns = float(model.get_metadata_prop("clk_ns"))
+ fclk_mhz = 1 / (clk_ns * 0.001)
+ runtime_s = (stablestate_cycles * clk_ns) * (10**-9)
+ rtlsim_perf_dict["stable_throughput[images/s]"] = rtlsim_bs / runtime_s
with open(report_dir + "/rtlsim_performance.json", "w") as f:
json.dump(rtlsim_perf_dict, f, indent=2)
--
GitLab