On the Feasibility and Limitations of Just-in-Time Instruction Set Extension for FPGA-Based Reconfigurable Processors

Table 5

Specialization process executed for whole applications when targeting the Woolcano architecture without capacity constraints. The performance of the custom instructions has been determined with the PivPav tool. ISE algorithms: MM: MaxMiso, SC: SingleCut, UN: Union. SC and UN search is constrained to 4 inputs and 1 input.

App		Executionruntimes				ISE algorithm runtime				Candidates found				Max ASIP-SP Speedup
		VM	Nat	Ratio		MM	SC	UN		MM	SC	UN		MM	SC	UN
		[s]	[s]	x		[ms]	[ms]	[ms]		#	#	#		x	x	x
164.gzip		23.71	18.47	1.28		40.6	549.0	11170.0		1621	44177	43682		1.172	1.213	1.213
179.art		69.92	74.70	0.94		12.3	55.1	3350.0		371	3534	3513		1.526	21.414	21.414
183.equake		7.97	6.79	1.17		13.5	457.9	4351.0		672	9690	9690		2.147	25.972	25.972
188.ammp		23.18	17.24	1.34		145.7	15840	—		7547	122441	—		3.449	20.826	—
429.mcf		23.94	24.06	1.00		11.1	68.7	200.5		571	3571	3562		1.112	1.112	1.112
433.milc		20.95	16.43	1.28		78.1	5065	—		3573	59450	—		1.301	21.546	—
444.namd		39.94	34.31	1.16		227.5	35854	—		11490	125970	—		1.609	24.846	—
458.sjeng		180.41	155.66	1.16		123.7	6244.1	235195.7		5540	83173	83035		1.118	1.137	1.137
470.lbm		5.68	5.36	1.06		8.6	2777.1	—		490	18216	—		2.554	44.622	—
473.astar		66.00	67.68	0.98		33.4	914.8	303796653		1408	37025	32368		1.159	1.19	1.19
AVG_S		46.17	42.07	1.14		69.45	6783	30405092		3328	50724	17585		1.71	16.39	5.20
adpcm		29.22	28.35	1.03		1.7	15.0	3869.4		83	819	819		1.243	1.309	1.293
fft		18.47	18.49	1.00		1.6	9.7	33.1		87	553	552		3.1	14.413	14.413
sor		15.83	15.85	1.00		0.7	4.3	14.6		35	384	375		14.418	14.422	14.418
whetstone		28.66	28.50	1.01		1.6	9.5	64.0		69	435	435		18.012	18.012	18.012
AVG_E		23.04	22.80	1.01		1.40	9.62	995.27		68.50	547.75	545.25		9.19	12.04	12.03
RATIO		2.00	1.85	1.13		49.61	705.05	30549.59		48.59	92.61	32.25		0.19	1.36	0.43