Flip statistics on ST231 processor

Flip statistics on ST231 processor

Floating-Point Library for Integer Processors:

with different rounding types (IEEE: nearest[=RN], towards +Inf[=RU] -Inf[=RD] 0[=RZ] ; or faithful[=RF])
with or without subnormals (=+/-0.xx...xxx*2^-Expmin)
previous benchmarks, FLIP v0.3 and STlib

Performance (unit: cycles):

Function	Flip Last Development Version								FLIP v0.3		1st lib
	Without Subnormals				With Subnormals					Without
	RN	RU	RZ	RD	RN	RU	RZ	RD	RN	RN	RN
add	26	26	23	26	26	26	23	26	44		61
div	27	30	28	30	32	35	34	35	132	51	177
mul	21	21	18	21	21	21	18	21	45	36	45
sqrt	21	21	21	21	23	23	23	23	123	51	127
sub	26	26	23	26	26	26	23	26	45		62

Complexity (unit: number of integer operations):

Function	Flip Last Development Version
	Without Subnormals				With Subnormals
	RN	RU	RZ	RD	RN	RU	RZ	RD
add	82	86	71	86	82	86	71	86
div	86	99	83	97	100	104	98	103
mul	71	73	59	72	71	73	59	72
sqrt	56	57	59	59	62	63	65	65
sub	84	88	73	88	84	88	73	88

Efficency (unit: percentage of VLIW bundle usage):

Function	Flip Last Development Version
	Without Subnormals				With Subnormals
	RN	RU	RZ	RD	RN	RU	RZ	RD
add	78	82	77	82	78	82	77	82
div	79	82	74	80	78	74	72	73
mul	84	86	81	85	84	86	81	85
sqrt	66	67	70	70	67	68	70	70
sub	80	84	79	84	80	84	79	84

Return to Flip's homepage

Claude-Pierre Jeannerod, Guillaume Revy
Last updated: LASTUPDATE