Sean Doran wrote:

> I'd be interested in how your results change for various
> -mmemory-latency=  values, and for no -mmemory-latency=
> option at all.

No huge difference, except a quickish time for the second L2 value:

	-mmemory-latency=main:
		26.096u 0.189s 0:26.33 99.7%    0+0k 0+110io 2pf+0w
		25.764u 0.217s 0:26.02 99.8%    0+0k 0+110io 0pf+0w
		25.979u 0.182s 0:26.21 99.7%    0+0k 0+110io 0pf+0w
	-mmemory-latency=L0:
		26.149u 0.185s 0:26.38 99.7%    0+0k 0+110io 2pf+0w
		25.848u 0.190s 0:26.08 99.8%    0+0k 0+110io 0pf+0w
		25.789u 0.202s 0:26.04 99.7%    0+0k 0+110io 0pf+0w
	-mmemory-latency=L1:
		26.301u 0.175s 0:26.53 99.7%    0+0k 0+110io 2pf+0w
		26.159u 0.190s 0:26.39 99.8%    0+0k 0+110io 0pf+0w
		26.256u 0.202s 0:26.51 99.7%    0+0k 0+110io 0pf+0w
	-mmemory-latency=L2:
		26.234u 0.212s 0:26.49 99.8%    0+0k 0+110io 2pf+0w
		24.822u 0.164s 0:25.02 99.8%    0+0k 0+110io 0pf+0w
		26.041u 0.205s 0:26.28 99.8%    0+0k 0+110io 0pf+0w
	-mmemory-latency=L3:
		26.524u 0.211s 0:26.77 99.8%    0+0k 0+110io 2pf+0w
		26.257u 0.196s 0:26.50 99.7%    0+0k 0+110io 0pf+0w
		25.696u 0.161s 0:25.90 99.8%    0+0k 0+110io 0pf+0w
	nothing:
		25.988u 0.190s 0:26.22 99.8%    0+0k 0+110io 2pf+0w
		26.131u 0.194s 0:26.37 99.8%    0+0k 0+110io 0pf+0w
		26.108u 0.158s 0:26.31 99.7%    0+0k 0+110io 0pf+0w

This still uses all the other options (-mcpu=21164a -Wa,-m21164a
-mmemory-latency=XXX -fsched-interblock -fsched-spec -fsched-spec-load
-fbranch-count-reg)

Simon.