Here is a way to test GetDuo. It is a simple Monte Carlo for estimating Pi. gcc cannot interfere with the loop too much because lCount is used outside the loop when the estimate of Pi is executed.
Code: Select all
#Include "MsWsII.bas"
Dim As Uint32 n = 10^9, i, lCount
Dim As Double t, x, y, z
lCount = 0
t = Timer
For i = 1 To n
msws.getduo(x,y)
' x = Rnd
' y = Rnd
If Sqr( x*x + y*y) <= 1 Then lCount += 1
Next
Print Timer - t;" seconds"
z = 4*lCount/n
Print "Estimate of Pi:";z
Sleep
One run gave me this:
Code: Select all
32 bit 64 bit
2xRnd 9.217s 3.606s
GetDuo 11.402s 3.602s
2 x Rnd is faster than GetDuo in 32-bit. They are neck and neck in 64-bit.
GetDuo is then not living to expectations either mine or Widynski's.
It could be his Intel Core i7-9700 (Q2'19) is cleverer than my Intel Core i7-3770 (Q2'12). Perhaps your CPU is cleverer than mine, so I will leave GetDuo in MsWsII.
From Widynski's Table 1 we have:
Code: Select all
Table 1: Time to generate and
sum one billion random numbers
RNG Precision Time
msws64 32 0.87
xoroshiro 53 1.09
msws32 32 1.18
xorwow 32 1.26
msws64 53 1.29
I should like to know how he derived the 'msws64 32' time.
The xoroshiro is xoroshiro128+ which is one of Sebastiano Vigna's older generators. His xoroshiro128** is very much better, randomness quality, and is faster. In fact it is a better 64-bit output generator than MsWsII. However, MsWsII's Rnd (randd) is no slouch, and it is handy to have it in the same package as the 32-bit output generator. Engine2's method of 64-bit output is definitely faster than 2 x Engine1 32-bit output.
If any of you have MsWsII working, then give the Monte Carlo code a test to see how your machine handles GetDuo. The reason I ask is that I have another procedure which splits the 64-bit output, but that is not living to expectations either on my machine, so I have not added it.
