PureBasic Forums - English

Try this. It replaces the main loop in your code but will still need you to add the part before (while getting into alignment) and after (for the leftover bytes). Because it's done 2048 bytes at a time the leftover bytes may be as many as 2047.

'on entry to this part xmm0 contains 16x the byte to ...

!pxor xmm6,xmm6 'zero xmm6
!psadbw xmm6,xmm7 'sum the 8 upper and 8 lower bytes
!movdqa xmm7,xmm6 'take a copy
!psrldq xmm6,8 'shift to line up
!paddw xmm7,xmm6 'add the 2 8byte sums to give a 16 byte sum

'the low word of xmm7 should now contain the sum required.
'but it needs correcting ...

I know little about the i7 and I don't have one to test but if the code is to be run on a range of

processors then i7 quirks can't be relied on. PREFETCH will benefit most CPUs and it shouldn't harm the i7 performance.

I did see the 100,000 but that gave unreasonably high figures. Maximum ...

What am I missing? There was a contest a few years ago here:
http://www.powerbasic.com/support/forums/Forum12/HTML/001733.html

Some of the code submitted there scans a 4MB file and counts ALL characters, not just 1, and it runs in 200ms on a really old processor, faster than any of the code posted ...

Paul,
you're going to have to look at the underlying ASM code to see what's going on.
Create the simplest demo you can that demonstrates the problem then compile it and look at the ASM to see how the compiled code for the integers differs in the 2 cases.

Both code and data alignment are important to timings but I can't imagine a modern compiler not aligning 32 bit data such as longs and floats, on a 32 bit (4 byte) boundary. Some compilers may misalign larger data such as doubles as they should be aligned on an 8 byte boundary for best performance ...

Code alignment/misalignment can easily change timings for small loops by as much 20%.
You can demonstrate alignment differences by inserting ASM NOP statements immediately ahead of your loop. Time the loop, insert one more NOP, time the loop, insert one more NOP .. until you have 16 or 32 NOPs and ...

Please, do add an unused padding byte element at the end so your structure ends up being 32bits long instead of 24. You can benchmark if you want, but you'll always see it's faster this way.
This is not always the case.
Using 4 bytes/structure instead of 3 requires 33% more memory reads which can ...

Well, for one collision and intersections are two different, separate things.

How does the calculation differ?

Untested pseudo code follows
'define the line by the co-ordinates of any 2 points on that line, (x1,y1) and (x2,y2)


'define the circle by the co-ordinates of its centre xc,yc and its radius, r

'now calculate the length of the sides of the triangle formed by the 3 points, call them a,b,c
a=SQR ...

The only other information you should need to decode your bit stream is the symbol table you just generated:S = 0, I = 11, P = 101, M = 100

You scan your coded data from the start and compare with each symbol in your symbol table. When you get a match, you have found the next symbol, you remove ...

I'm surprised any standard, modern compiler "wins". Surely they all use the same FPU and it's that which sets the accuracy.
FPU arithmetic has been standardised for 20+ years.

Sospel.
Can you post the instructions in POWERBASIC?

The following is the complete PB Console Compiler code (Currently $169)
DEFDBL a-z 'you want to test DOUBLES so set default variable type to that
FUNCTION PBMAIN () AS LONG

value = 1.0
FOR i=1 TO 10
FOR j=1 TO 1000
FOR k=1 TO 1000
value = TAN ...

Psychophanta,

i'd like to demonstrate to blueznl and to all that these 2 users are wrong

Just because you don't understand what these 2 users are trying to explain doesn't mean they are wrong.

Psychophanta,
The original post in this thread stated "error in % = 0.000010005" so we are dealing here with percentage errors.

In your example the exact answer is zero so if your "value" is anything other than zero then the % error is infinite (not undefined).

It may not be a useful result ...