|
And how about the bitwise version of Algorithm X? Is that interesting enough?
|
|
|
|
|
A float has only 24 bits that need to be divided, and one of them is "usually 1", so it requires less iterations than a full 32bit number (or even 64, which is a disaster)
Double's shouldn't really be faster than ints though (edit: ok yes it is possible in some cases, but it depends, see below), something else is going on here. Note that the time a division takes also depends on the magnitude of the result for ints, and on the roundness of the divisor for floats. This makes it harder to accurately compare the speed of integer division and floating point division.
As if that isn't bad enough yet there is also SSE vs good ol' FPU, and with Java you really have no clue which one you're getting, so I suggest you use C++ with assembly subroutines to take full control of what you're testing.
Hope this helps 
edit2: forgot source, Agner Fog's optimization resources[^]
modified on Monday, December 21, 2009 11:36 AM
|
|
|
|
|
Everything - ok maybe not, but pretty much.
Dividing two doubles may take as much as 21 cycles, and as little as 6. Dividing two 32bit ints may take 18 to 42 cycles, but it's usually on the smaller end of that. Whether it is on average better to divide doubles or ints depends on what kinds of values you have.
And doing any kind of math on denormals/NaN's/infinity's can incur a penalty.
Also, for 64bit numbers idiv is lower than div (up to 30%!)
|
|
|
|
|
On most processors (the high-end of them at least) I have seen the speed for a series of numeric operations would be float >= double > int > long; there are several reasons for this:
1.
floating-point units are more highly pipelined, which results in a greater latency but a better throughput, so for a series of FP operations, the FPU would be faster than the same operations handled by integers.
2.
integer divisions compute only a few (say 3 or 4) bits per cycle, so long could take twice as long as int.
similar for real divisions, but more bits per cycle.
3.
float operations leave integer functional units available for other things, such as array indexing, iteration counting, etc. which are most always present in code too; as those units can operate in parallel, having more of them working for you should improve overall speed.
BTW: division (and square root) is inherently sequential, much more so than addition and multiplication.
|
|
|
|
|
Floating point division is not particularly pipelined either though, the difference between the latency and the reciprocal throughput of FDIV is only 1 cycle (on a Core2)
|
|
|
|
|
Like we have FlowLayoutPanel, where controls can automatically get laid out according to their sizes and sequence, I have a requirement of a fluid layout.
In a rectangle or area (say 500w X 100h), there are some fixed small controls. lets say a rectangle control of (100w X 30h) is fixed at point (0 top , 150 left) Now, I need to fill this bigger rectangle with small shaped controls with constant areas but variable widths. Height can be maximum wherever possible.
The controls not adjacent to fixed control will be rectangles So, no problems. The controls adjacent to fixed controls will need to take polygon shape of variable heights to fit in the balance space.
Any logic already existing? Any approaches?
And yes, this is for winforms.
|
|
|
|
|
Assume that there is a vertical cylinder with a diameters as follows
1cm at base
3cm at the height of 2cm
2cm at the height of 4cm
1cm at the height of 7cm
etc etc
Now, what would be the height of different colored fluid columns one over the other (Assuming there are no other forces). The columns are on top of each other and movable by drag and drop. Hence the calculation of each column depends on the bottom column.
Its not necessary to consider volumetric analysis because the solution could be in 2D for area also.
The function should take parameters of volume/area of each column at which it starts.
This is required for a graphical application where the movement of a liquid is shown in 2D.
I am looking for a faster alternative to "for" or "while" loops. It needs to be very fast so that a fluid motion could be achieved. Current algorithm is damn slow.
my current solution looks like:
while (volumeToAdjust > 0)
{
volumeCovered = Min (volumeToAdjust, volumePossibleInCurrentWidth)
if(volumeCovered>volumePossibleInCurrentWidth)
height += heightForCurrentWidth
else
height += CalculateHeightForThisVolumeInCurrentWidth
volumeToAdjust -= volumeCovered
}
This solution works fine for a single column of fluid. Multiple columns, multiple widths, the program slows down. Any better ideas?
|
|
|
|
|
Do the diameters change or can they be hard coded?
What happens in between the values you gave? Is the diameter constant or tapered?
CQ de W5ALT
Walt Fair, Jr., P. E.
Comport Computing
Specializing in Technical Engineering Software
|
|
|
|
|
Yes diameters can be changed. Also, the volume of each fluid column and their sequence can be changed by drag and drop. That is why the need of an algorithm to do it the fastest way.
Diameters are not tapered.
|
|
|
|
|
http://www.prime-numbers.org
Tadeusz Westawic
An ounce of Clever is worth a pound of Experience.
|
|
|
|
|
that list is not complete. 
|
|
|
|
|
Of course it is, they all are.
Tadeusz Westawic
An ounce of Clever is worth a pound of Experience.
|
|
|
|
|
No it isn't, just the ones less than 10 billion. As far as anyone knows, the complete list is infinite. 
You measure democracy by the freedom it gives its dissidents, not the freedom it gives its assimilated conformists.
|
|
|
|
|
Hi every body ,
I am in critical position, can u plz tell me how the plain text to encrypt and decrypt using ecc(Elliptic Curve Cryptography) in C# i need step by step process if not the complete source code i need.
can any one help me plz......
advance thanks to for reply
|
|
|
|
|
I need someone to mow the lawn, paint the house and wash the car. plzzzzzzzzzzzzzzz
|
|
|
|
|
People here will help you with problems, not do your work or more likely homework for you.
You measure democracy by the freedom it gives its dissidents, not the freedom it gives its assimilated conformists.
|
|
|
|
|
i very need to an algorithm for Imaging of cars No in roads.
means how police cameras imaging of car No for forfeit.
because size of cars No is different.
thanx.
|
|
|
|
|
And what has your literature search turned up so far?
You measure democracy by the freedom it gives its dissidents, not the freedom it gives its assimilated conformists.
|
|
|
|
|
question about K-Means algorithm for image
how to code a K-Means algorithm for Image Segmentation In Matlab
and also how to plot the color histogram in matlab??
Thank you
|
|
|
|
|
|
i'm puzzled why it uses sequential search in wndproc to search the handler of a msg.(use swtich..case.. statement). at least, it should use binary search for efficiency, doesn't it? sequential search's time-complexity is O(n), and binary search's is O(logn). they're quite different in efficiency along with the increase of 'n'.
the same puzzle in search the addr. of a function in virtual-table at runtime polymorphic.
|
|
|
|
|
big-O notation gives the dependency law, not the absolute value; there is at least one constant involved, and that tends to be smaller for simpler algorithms. I suggest you create some implementations and test and compare them for a series of reasonable values of n.
|
|
|
|
|
Why is this important? It is a very small part of any program and will have negligible effect on the overall performance.
|
|
|
|
|
IMO both VTABLE lookup and Windows message identification, although small in size, are needed at a high frequency (every virtual method call, every message in an overridden WndProc), so they may very well be relevant to overall performance.
|
|
|
|
|
Possibly, but I suspect the most pressing need for most apps is to make the actual business logic as efficient as possible.
|
|
|
|
General 
News 
Suggestion 
Question 
Bug 
Answer 
Joke 
Praise 
Rant 
Admin 
Submit an article or tip
