LightningX

Profiling Results

Benchmark	Cosimulation Cycles	LightningSim Cycles	OmniSim Cycles	Cosimulation Runtime (in s)	LightningSim Runtime (in s)	OmniSim Runtime (in s)	LightningSim Speedup (vs. Cosim)	OmniSim Speedup (vs. Cosim)	OmniSim Speedup (vs. LightningSim)
Fixed-point square root	30	30	30	27.25	4.97	3.65	5.48×	7.47×	1.36×
FIR filter	172	172	172	20.12	2.43	1.94	8.23×	10.37×	1.25×
Fixed-point window conv	35	35	35	28.30	3.69	3.15	7.67×	8.98×	1.71×
Floating-point conv	35	35	35	49.78	2.42	2.46	20.57×	20.24×	0.98×
Arbitrary Precision ALU	36	36	36	24.17	2.12	2.03	11.40×	11.91×	1.04×
Parallel loops	32	32	32	26.81	2.34	2.16	11.48×	12.41×	1.08×
Imperfect loops	34	34	34	25.80	2.24	2.13	11.52×	12.11×	1.05×
Loop with max bound	31	31	31	24.76	2.25	2.14	11.01×	11.57×	1.05×
Perfect nested loops	406	406	406	24.76	2.27	2.12	10.91×	11.68×	1.07×
Pipelined nested loops	405	405	405	24.92	2.23	2.19	11.18×	11.38×	1.02×
Sequential accumulators	32	32	32	26.59	2.29	2.20	11.61×	12.09×	1.04×
Accumulators + asserts	33	33	33	27.13	2.30	2.30	11.80×	11.80×	1.00×
Accumulators + dataflow	31	31	32	27.26	2.29	2.19	11.90×	12.45×	1.05×
Static memory example	66	66	66	33.23	2.18	2.12	15.24×	15.67×	1.03×
Pointer casting example	408	408	408	32.55	2.15	2.13	15.14×	15.28×	1.01×
Double pointer example	25	25	25	31.70	2.14	1.91	14.81×	16.60×	1.12×
AXI-4 master	178	177	177	21.06	2.19	2.07	9.62×	16.60×	10.17×
AXIS w/o side channel	52	51	51	19.12	2.06	1.94	9.28×	9.86×	1.06×
Multiple array access	252	252	252	24.32	2.18	2.08	11.16×	11.69×	1.05×
Resolved array access	131	131	131	24.36	2.20	2.05	11.07×	11.88×	1.07×
URAM with ECC	115	115	115	22.07	2.21	2.05	9.99×	10.77×	1.08×
Fixed-point Hamming	259	259	259	33.28	2.37	2.46	14.04×	13.53×	0.96×
Unoptimized FFT	261781	261150	261150	153.53	2.78	2.91	55.23×	52.76×	0.96×
Multi-stage FFT	3770	3772	3722	61.43	2.67	2.93	23.01×	20.97×	0.92×
Huffman encoding	10283	10272	10272	46.89	2.63	2.32	17.83×	20.21×	1.13×
Matrix Multiplication	1036	1036	1036	26.33	2.61	2.59	10.09×	10.17×	1.01×
Parallelized merge sort	131	131	131	48.79	2.27	2.15	21.49×	22.69×	1.06×
Vector add with stream	4261	4261	4261	27.21	4.48	3.56	6.07×	7.64×	1.26×
FlowGNN GIN	260359	260337	260337	4219.85	28.90	11.97	146.02×	352.53×	2.41×
FlowGNN GCN	112836	112561	112561	534.33	30.90	17.18	17.29×	31.10×	1.80×
FlowGNN GAT	17282	17282	17282	838.24	41.60	24.60	20.15×	34.07×	1.69×
FlowGNN PNA	344206	344206	344206	3285.45	30.50	29.00	107.72×	113.29×	1.05×
FlowGNN DGN	110710	110710	110710	996.13	26.90	11.71	37.03×	85.07×	2.30×

Important Takeaways!

The above results have been published for version 2021.1 of the Vitis Development Suite.

The timing estimates provided by both LightningSim and OmniSim are 99.9% accurate with respect to the results from C/RTL Co-simulation.

LightningSim achieves a speedup of up to 146.02× over C/RTL Co-simulation.

OmniSim achieves a speedup of up to 352.53× over C/RTL Co-simulation

Both LightningSim and OmniSim provide the incremental design space exploration features, thereby achieving a speedup of up to 577× over C/RTL Co-simulation.