improved texfetch to exploit all of texture hardware

Peter,
Rather than the scaled images, would it be possible for you to post the actual min/max times and standard deviations of the times for each image? I find it hard to interpret the scaled images since if I understand you correctly, they are each scaled to their own min/max timing values. I think min/max/stddev would be more informative for me personally…

Cheers,
John

Yeah, I am just starting to do some timing statistics and it struck me quite a bit to see the slow startup for these operators. I expected a more homogeneous distribution as all processing is not data dependent here. (Maybe that was also true for previous shader implementations, we just never got a chance to see it…)

So here you go: (all in clock() ticks)

[min;mean;max] average/variance @ std.deviation

[1270;3062;5018] 2993/478085 @ 691.437 screenshot1
[1816;3020;6044] 3023/174462 @ 417.686 screenshot2
[35271;45156;50533] 45125/213848 @ 462.437 screenshot3
[2665;3501;5596] 3511/202781 @ 450.312 screenshot4
[48252;77220;89598] 77335/385408 @ 620.812 screenshot5

More to come when I have time (varying block sizes etc)

Peter