PR24518 and PR24520 - Improvements to MOD and

[François-Xavier Coudert]

Paul,

Thanks for the updated patch. Below are the results of my testing on
ppc-darwin, for both real(8) and real(16). To be seen in fixed-width
font. Basically, there is very little difference between the library
and the inline version.

FX


real(8) on powerpc-apple-darwin7.9.0

n		-O1	-O2	[1]	[2]
---------------------------------------------
4	lib.	72.80	62.20	66.40	66.60
	inline	73.30	60.80	67.70	65.70
8	lib.	64.40	54.00	59.50	59.40
	inline	68.10	52.70	58.60	58.70
16	lib.	110.9	101.4	106.8	105.7
	inline	116.6	99.30	104.9	109.5
32	lib.	204.2	193.1	199.0	198.9
	inline	212.6	194.4	198.5	198.0
64	lib.	390.0	379.2	385.1	385.3
	inline	409.1	394.8	384.5	385.0
128	lib.	767.2	751.5	757.0	756.9
	inline	793.3	767.0	756.2	756.8
256	lib.	1557.	1495.	1502.	1504.
	inline	1569.	1519.	1502.	1503.
512	lib.	3015.	2984.	2989.	2989.
	inline	3125	3027.	3009.	3032.
1024	lib.	5970.	5961.	5966.	5967.
	inline	6048	5981.	6089.	5994.

[1] -O2 -funroll-loops
[2] -O2 -funroll-loops -ffast-math

################################################################

real(16) on powerpc-apple-darwin7.9.0

n		-O1	-O2	[1]	[2]
---------------------------------------------
4	lib.	1085	1083	1044	1067
	inline	1057	1040	1057	1053
8	lib.	2074	1948	1940	1947
	inline	1976	1940	1951	1949
16	lib.	3980	3872	3876	3832
	inline	3881	3850	3841	3836
32	lib.	7869	7692	7681	7644
	inline	7723	7683	7654	7654
64	lib.	15452	15347	XXX	XXX
	inline	15376	15558	XXX	XXX

[1] -O2 -funroll-loops
[2] -O2 -funroll-loops -ffast-math
-------------- next part --------------
A non-text attachment was scrubbed...
Name: thomas_timing
Type: application/octet-stream
Size: 1247 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20051116/f48f11e9/attachment.obj>