Hello.

use visual studio 2008, 2010.
Why, when using only optimization without SSE, OpenMP and optimization with SSE, OpenMP time of factrorization with double speed up float?

Could you clarify the two configurations you compared? Also make sure you compiled with compiler optimizations ON.

Maximize Speed (/O2)
Streaming SIMD Extensions 2 (/arch:SSE2)
Yes (/openmp)

Configuration Optimization SSE2 OpenMP
A + – –
B – + –
C – – +
D + + +

conf A
float double
CPU 1000 2000 3000 4000 1000 2000 3000 4000
i5 0,61 4,65 16,19 37,16 0,22 1,61 5,14 12,22
i7 0,68 5,16 17,16 40,25 0,22 1,65 5,41 12,59

conf В
i5 2,84 25,76 115,57 171,32 4,77 36,59 121,42 453,13
i7 2,59 20,58 68,55 159,67 4,13 32,03 105,75 248,65

conf C
i5 4,82 47,14 147,86 337,85 4,09 34,57 119,22 300,72
i7 3,58 16,77 47,11 95,35 3,47 15,72 42,62 84,35

conf D

i5 0,19 0,52 1,69 2,86 0,21 0,69 2,24 5,09
i7 0,31 0,57 1,17 2,4 0,34 0,96 2,15 4,17

This looks fine:
- benchmarking without all compiler optimizations is meaningless, so you can ignore the B and C cases
- the result of case D is expected
- the A case is using the x87 FPU which handles double only. So performance should be similar for float and double in this case, but since it is extremely difficult for compilers to generate optimized code for the FPU, you're observing slower perf. with float. Looking at the generated ASM would help to understand how the compiler failed, but since the FPU is deprecated nowadays, this would be waste of time.

thanks.

conf B and C was using only for testing (how eigen works) and do not using for real computation.
P/s/ sorry for my bad english

Why does not always i7 work faster i5?