Maybe this is a stupid question, but say I've got like two functions, void F1(int x) and void F2(int x), and I want to execute them in each iteration of a for loop. How much would it difer (performance-wise) if I did one big for loop, like this:
`
for(int i = 0; i < 100 ; i++)
{
F1(i);
F2(i);
}
`
compared to doing two separate loops, one in which I call F1, one in which I call F2:
`
for(int i = 0 ; i < 100 ; i++)
F1(i);
for(int i = 0 ; i < 100 ; i++)
F2(i);
`
While writing this, it occured to me that the first way is probably faster because there are only aproximately 100 increments and 100 comparisons, while in the second case, we'll get 200 of each.
Say my loop only has to run for 200 iterations. Would the two-for loops approach be pretty much the same in terms of performance, considering, say, CPUs from 2007 and after:)?
It depends on what F1 and F2 do.
It could not matter at all or you can experience dramatic slow down by having both functions called one after the other.
As an example of the latter case consider F1 and F2 accessing two different arrays. At each run of F1 and F2 they read enough data to cause the whole cache to be overwritten. That would probably cause a good slow down.
But is always better not to speculate and measure and benchmark your code instead. If performance is equivalent for both versions go for the most readable.
Well, as you pointed out the amount of operations differ. However, if you go with the second solution you can use multiple threads and achieve the same or better performance. Also, consider readability, testability, expandability, encapsulation, usually those factors are more important than any small performance gain you might get. And also, the compiler is usually very great at making your code run more effectively, so my advice is focus on your readability more than your performance in most cases.
Related
I've been working on my implementation of BigInteger, and when I was contemplating the solution for addition, I decided to go with cleaner one, which had in mind adding corresponding digits in function and "normalizing" them later. Like in the following example
999 999 + 111 111
= 10 10 10 10 10 10 (value after addition)
= 1 111 110 (value after normalization)
But since then I was wondering about how it affects the efficiency of the program. Are several loops doing small things each generally going to work faster than one big nested loop?
For example, using
int a[7]={0,9,9,9,9,9,9};
int b[7]={0,1,1,1,1,1,1};
int c[7];
Is this,
for(int q=0; q<7; ++q){
c[q]=a[q]+b[q];
if(c[q]>9){
c[q-1]=c[q]/10;
c[q]%=10;
}
}
better than this
for(int q=0; q<7; ++q){
c[q]=a[q]+b[q];
}
for(int q=0;q<7;++q){
if(c[q]>9){
c[q-1]=c[q]/10;
c[q]%=10;
}
}
And what about bigger loops, that have much more things to go through on each iteration?
UPD.
As someone suggested I did measure performance time for both examples. For two loops the average time (for 100mil. elements) ~4.85sec. For one loop ~3.72sec
It is very difficult to tell which one of the two approaches will be more efficient. It probably varies among C++ compiler vendors and within a single vendor, from version to version of their compiler.
The bottom line is:
You will never know unless you benchmark.
As usual, it is almost certain that it does not matter anyway, and you are most probably unduly concerned about performance, like the vast majority of programmers do, in the vast majority of cases.
At the end of the day, all that matters is what is more readable and more maintainable. Code maintainability is far more important than saving clock cycles.
If you decide to follow the wise path of "what is more readable" keep in mind that different folks find different things more readable. For example, I personally hate surprises when I am reading code, so I would be rather annoyed to read your first loop which allows decimal digits to receive erroneous values outside of the 0-9 range, only to find out later that you are finally remedying that with another loop.
This question already has an answer here:
How to prevent optimization of busy-wait
(1 answer)
Closed 7 years ago.
I am doing some experiments on CPU's performance. I wonder if anyone know a formal way or a tool to generate simple code that can run for a period of time (several seconds) and consumes significant computation resource of a CPU.
I know there are a lot of CPU benchmarks but the code of them is pretty complicated. What I want is a program more straight forward.
As the compiler is very smart, writing some redundant code as following will not work.
for (int i = 0; i < 100; i++) {
int a = i * 200 + 100;
}
Put the benchmark code in a function in a separate translation unit from the code that calls it. This prevents the code from being inlined, which can lead to aggressive optimizations.
Use parameters for the fixed values (e.g., the number of iterations to run) and return the resulting value. This prevents the optimizer from doing too much constant folding and it keeps it from eliminating calculations for a variable that it determines you never use.
Building on the example from the question:
int TheTest(int iterations) {
int a;
for (int i = 0; i < iterations; i++) {
a = i * 200 + 100;
}
return a;
}
Even in this example, there's still a chance that the compiler might realize that only the last iteration matters and completely omit the loop and just return 200*(iterations - 1) + 100, but I wouldn't expect that to happen in many real-life cases. Examine the generated code to be certain.
Other ideas, like using volatile on certain variables can inhibit some reasonable optimizations, which might make your benchmark perform worse that actual code.
There are also frameworks, like this one, for writing benchmarks like these.
It's not necessarily your optimiser that removes the code. CPU's these days are very powerful, and you need to increase the challenge level. However, note that your original code is not a good general benchmark: you only use a very subset of a CPU's instruction set. A good benchmark will try to challenge the CPU on different kinds of operations, to predict the performance in real world scenarios. Very good benchmarks will even put load on various components of your computer, to test their interplay.
Therefore, just stick to a well known published benchmark for your problem. There is a very good reason why they are more involved. However, if you really just want to benchmark your setup and code, then this time, just go for higher counter values:
double j=10000;
for (double i = 0; i < j*j*j*j*j; i++)
{
}
This should work better for now. Note that there a just more iterations. Change j according to your needs.
I have two huge arrays (int source[1000], dest[1000] in the code below, but having millions of elements in reality). The source array contains a series of ints of which I want to copy 3 out of every 4.
For example, if the source array is:
int source[1000] = {1,2,3,4,5,6,7,8....};
int dest[1000];
Here is my code:
for (int count_small = 0, count_large = 0; count_large < 1000; count_small += 3, count_large +=4)
{
dest[count_small] = source[count_large];
dest[count_small+1] = source[count_large+1];
dest[count_small+2] = source[count_large+2];
}
In the end, dest console output would be:
1 2 3 5 6 7 9 10 11...
But this algorithm is so slow! Is there an algorithm or an open source function that I can use / include?
Thank you :)
Edit: The actual length of my array would be about 1 million (640*480*3)
Edit 2: Processing this for loop takes about 0.98 seconds to 2.28 seconds, while the other code only take 0.08 seconds to 0.14 seconds, so the device uses at least 90 % cpu time only for the loop
Well, the asymptotic complexity there is as good as it's going to get. You might be able to achieve slightly better performance by loading in the values as four 4-way SIMD integers, shuffling them into three 4-way SIMD integers, and writing them back out, but even that's not likely to be hugely faster.
With that said, though, the time to process 1000 elements (Edit: or one million elements) is going to be utterly trivial. If you think this is the bottleneck in your program, you are incorrect.
Before you do much more, try profiling your application and determine if this is the best place to spend your time. Then, if this is a hot spot, determine how fast is it, and how fast you need it to be/might achieve? Then test the alternatives; the overhead of threading or OpenMP might even slow it down (especially, as you now have noted, if you are on a single core processor - in which case it won't help at all). For single threading, I would look to memcpy as per Sean's answer.
#Sneftel has also reference other options below involving SIMD integers.
One option would be to try parallel processing the loop, and see if that helps. You could try using the OpenMP standard (see Wikipedia link here), but you would have to try it for your specific situation and see if it helped. I used this recently on an AI implementation and it helped us a lot.
#pragma omp parallel for
for (...)
{
... do work
}
Other than that, you are limited to the compiler's own optimisations.
You could also look at the recent threading support in C11, though you might be better off using pre-implemented framework tools like parallel_for (available in the new Windows Concurrency Runtime through the PPL in Visual Studio, if that's what you're using) than rolling your own.
parallel_for(0, max_iterations,
[...] (int i)
{
... do stuff
}
);
Inside the for loop, you still have other options. You could try a for loop that iterates and skips every for, instead of doing 3 copies per iteration (just skip when (i+1) % 4 == 0), or doing block memcopy operations for groups of 3 integers as per Seans answer. You might achieve slightly different compiler optimisations for some of these, but it is unlikely (memcpy is probably as fast as you'll get).
for (int i = 0, int j = 0; i < 1000; i++)
{
if ((i+1) % 4 != 0)
{
dest[j] = source[i];
j++;
}
}
You should then develop a test rig so you can quickly performance test and decide on the best one for you. Above all, decide how much time is worth spending on this before optimising elsewhere.
You could try memcpy instead of the individual assignments:
memcpy(&dest[count_small], &source[count_large], sizeof(int) * 3);
Is your array size only a 1000? If so, how is it slow? It should be done in no time!
As long as you are creating a new array and for a single threaded application, this is the only away AFAIK.
However, if the datasets are huge, you could try a multi threaded application.
Also you could explore having a bigger data type holding the value, such that the array size decreases... That is if this is viable to your real life application.
If you have Nvidia card you can consider using CUDA. If thats not the case you can try other parallel programming methods/environments as well.
Let's say I want to benchmark two competing implementations of some function double a(double b, double c). I already have a large array <double, 1000000> vals from which I can take input values, so my benchmarking would look roughly like this:
//start timer here
double r;
for (int i = 0; i < 1000000; i+=2) {
r = a(vals[i], vals[i+1]);
}
//stop timer here
Now, a clever compiler could realize that I can only ever use the result of the last iteration and simply kill the rest, leaving me with double r = a(vals[999998], vals[999999]). This of course defeats the purpose of benchmarking.
Is there a good way (bonus points if it works on multiple compilers) to prevent this kind of optimization while keeping all other optimizations in place?
(I have seen other threads about inserting empty asm blocks but I'm worried that might prevent inlining or reordering. I'm also not particularly fond of the idea of adding the results sum += r; during each iteration because that's extra work that should not be included in the resulting timings. For the purposes of this question, it would be great if we could focus on other alternative solutions, although for anyone interested in this there is a lively discussion in the comments where the consensus is that += is the most appropriate method in many cases. )
Put a in a separate compilation unit and do not use LTO (link-time optimizations). That way:
The loop is always identical (no difference due to optimizations based on a)
The overhead of the function call is always the same
To measure the pure overhead and to have a baseline to compare implementations, just benchmark an empty version of a
Note that the compiler can not assume that the call to a has no side-effect, so it can not optimize the loop away and replace it with just the last call.
A totally different approach could use RDTSC, which is a hardware register in the CPU core that measures the clock cycles. It's sometimes useful for micro-benchmarks, but it's not exactly trivial to understand the results correctly. For example, check out this and goggle/search SO for more information on RDTSCs.
Or is it all about semantics?
Short answer: no, they are exactly the same.
Guess it could in theory depend on the compiler; a really broken one might do something slightly different but I'd be surprised.
Just for fun here are two variants that compile down to exactly the same assembly code for me using x86 gcc version 4.3.3 as shipped with Ubuntu. You can check the assembly produced on the final binary with objdump on linux.
int main()
{
#if 1
int i = 10;
do { printf("%d\n", i); } while(--i);
#else
int i = 10;
for (; i; --i) printf("%d\n", i);
#endif
}
EDIT: Here is an "oranges with oranges" while loop example that also compiles down to the same thing:
while(i) { printf("%d\n", i); --i; }
If your for and while loops do the same things, the machine code generated by the compiler should be (nearly) the same.
For instance in some testing I did a few years ago,
for (int i = 0; i < 10; i++)
{
...
}
and
int i = 0;
do
{
...
i++;
}
while (i < 10);
would generate exactly the same code, or (and Neil pointed out in the comments) with one extra jmp, which won't make a big enough difference in performance to worry about.
There is no semantic difference, there need not be any compiled difference. But it depends on the compiler. So I tried with with g++ 4.3.2, CC 5.5, and xlc6.
g++, CC were identical, xlc WAS NOT
The difference in xlc was in the initial loop entry.
extern int doit( int );
void loop1( ) {
for ( int ii = 0; ii < 10; ii++ ) {
doit( ii );
}
}
void loop2() {
int ii = 0;
while ( ii < 10 ) {
doit( ii );
ii++;
}
}
XLC OUTPUT
.loop2: # 0x00000000 (H.10.NO_SYMBOL)
mfspr r0,LR
stu SP,-80(SP)
st r0,88(SP)
cal r3,0(r0)
st r3,64(SP)
l r3,64(SP) ### DIFFERENCE ###
cmpi 0,r3,10
bc BO_IF_NOT,CR0_LT,__L40
...
enter code here
.loop1: # 0x0000006c (H.10.NO_SYMBOL+0x6c)
mfspr r0,LR
stu SP,-80(SP)
st r0,88(SP)
cal r3,0(r0)
cmpi 0,r3,10 ### DIFFERENCE ###
st r3,64(SP)
bc BO_IF_NOT,CR0_LT,__La8
...
The scope of the variable in the test of the while loop is wider than the scope of variables declared in the header of the for loop.
Therefore, if there are performance implications as a side-effect of keeping a variable alive longer, then there will be performance implications in choosing between a while and a for loop ( and not wrapping the while up in {} to reduce the scope of its variables ).
An example might be a concurrent collection which counts the number of iterators referring to it, and if more than one iterator exists, it applies locking to prevent concurrent modification, but as an optimisation elides the locking if only one iterator refers to it. If you then had two for loops in a function using differently named iterators on the same container, the fast path would be taken, but with two while loops the slow path would be taken. Similarly there may be performance implications if the objects are large (more cache traffic), or use system resources. But I can't think of a real example that I've ever seen where it would make a difference.
Compilers that optimize using loop unrolling will probably only do so in the for-loop case.
Both are equivalent. It's a matter of semantics.
The only difference may lie in the do... while construct, where you postpone the evaluation of the condition until after the body, and thus may save 1 evaluation.
i = 1; do { ... i--; } while( i > 0 );
as opposed to
for( i = 1; i > 0; --i )
{ ....
}
I write compilers. We compile all "structured" control flow (if, while, for, switch, do...while) into conditional and unconditional branches. Then we analyze the control-flow graph. Since a C compiler has to deal with general goto anyway, it is easiest to reduce everything to branch and conditional-branch instructions, then be sure to handle that case well. (A C compiler has to do a good job not just on handwritten code but also on automatically generated code, which may have many, many goto statements.)
No. If they're doing equivalent things, they'll compile to the same code - as you say, it's about semantics. Choose the one that best represents what you're trying to express.
Ideally it should be the same, but eventually it depends on your compiler/interpreter. To be sure, you must measure or examine the generated assembly code.
Proof that there may be a difference: These lines produce different assembly code using cc65.
for (; i < 1000; ++i);
while (i < 1000) ++i;
On Atmel ATMega while() is faster than for(). Why is this is explained in AVR035: Efficient C Coding for AVR.
P.S. Original platform was not mentioned in question.
continue behaves differently in for and while: in for, it alters the counter, in while, it usually doesn't
To add another answer: In my experience, optimizing software is like a big, bushy beard being shaved off a man.
First you lop it off in big chunks with scissors (prune whole limbs off the call tree).
Then you make it short with an electric clipper (tweak algorithms).
Finally you shave it with a razor to get rid of the last little bit (low-level optimization).
The last is where the difference between for() and while() might, but probably won't, make a difference.
P.S. The programmers I know (who are all very good, and I suspect are a representative sample) basically go at it from the other direction.
They are the same as far as performance goes. I tend to use while when waiting for a state change (such as waiting for a buffer to be filled) and for when processing a number of discrete objects (such as going through each item in a collection).
There is a difference in some cases.
If you are at the point where that difference matters, you either need to pick a better algorithm or begin coding in assembly language. Trust me, coding in assembly is preferable to fixing your compiler version.
Is while() faster/slower than for()? Let's review a few things about optimization:
Compiler-writers work very hard to shave cycles by having fewer calls to jump, compare, increment, and the other kinds of instructions that they generate.
Call instructions, on the other hand, consume many magnitudes more cycles, but the compiler is nearly powerless to do anything to remove those.
As programmers, we write lots of function calls, some because we mean to, some because we're lazy, and some because the compiler slips them in without being obvious.
Most of the time, it doesn't matter, because the hardware is so fast, and our jobs are so small, that the computer is like a beagle dog who wolfes her food and begs for more.
Sometimes, however, the job is big enough that performance is an issue.
What do we do then? Where's the bigger payoff?
Getting the compiler to shave a few cycles off loops & such?
Finding function calls that don't -really- need to be done so much?
The compiler can't do the latter. Only we the programmers can.
We need to learn or be taught how to do this. It doesn't come naturally.
We are congenitally inclined to make wrong guesses and then bet on them.
Getting better algorithms is a start, but only a start. Our teachers need to teach this, if indeed they know how.
Profilers are a start. I do this.
The apocryphal quote of Willie Sutton when asked Why do you rob banks?:
Because that's where the money is.
If you want to save cycles, find out where they are.
Probably only coding style.
for if you know the number of iterations.
while if you do not know the number of iterations.