C#

Leslie Sanford wrote:

Any thoughts or insights?

Luc Pattyn wrote:

I like open questions... they indicate wisdom: either awareness of ignorance,
or a truely open mind!

Luc Pattyn wrote:

Quite often, when comparing two alternatives, the second wins, not because
it is faster, but just because it gets the better starting conditions.

So the minimum one should do is swap both to see how much of a difference that
gives.

Luc Pattyn wrote:

My approach to cover both issues typically is to perform all N different tests
one after another, but then loop over it for say 100 times.
Then either ignore the first iteration entirely, and/or do some statistical
analysis on those N result arrays.

But even then, for N>2 the order of the tests may also be significant, so it
might be better to sequence them at random.

Luc Pattyn wrote:

Hope this helps.

for(int i = 0; i < Size; i++) {
   p++ = 1.0f;
}

Guffa wrote:

p++ = 1.0f;

fixed(float* pSource = data)
{                            
    float* p = pSource;
 
    for(int i = 0; i < Size; i++)
    {
        *p = 1.0f;
        p++;
    }
}

Guffa wrote:

That should be a bit faster, as it doesn't produce any code to check the boundary of the array.

Robert Surtees wrote:

If I just reverse the order of the tests I get opposite results.

Robert Surtees wrote:

The optimizer seems pretty stupid however. Changing
*p = 1.0f;
p++;

to

*p++ = 1.0f

nearly halves the result. Perhaps there is some switch to improve that sort of thing.

*p++ = 1.0f;

Robert Surtees wrote:

The optimizer seems pretty stupid however. Changing
*p = 1.0f;
p++;

to

*p++ = 1.0f

nearly halves the result. Perhaps there is some switch to improve that sort of thing.

*p = 42;
p++;

*p++ = 42;

ldloc.1
ldc.i4.s 0x2a
stind.i4
ldloc.1
ldc.i4.4
conv.i
add
stloc.1

ldloc.1
dup
ldc.i4.4
conv.i
add
stloc.1
ldc.i4.s 0x2a
stind.i4

mov dword ptr [ebx],2Ah
add ebx,4

mov edi,ebx
add ebx,4
mov dword ptr [edi],2Ah

Luc Pattyn wrote:

One of the factors is the tags that are present inside the < PRE> </PRE> block.
In your message the lines are separated by <br> tags; most other messages have
just regular newlines.

Could you please tell me how you get that, i.e. what tools you use, and how you go
about it. Thanks in advance.

fixed(float* pSource = data)
{
    float* p = pSource;

    // A blank line should come before the loop.
    for(int i = 0; i < Size; i++)
    {
        *p = 1.0f;
        p++;
    }
}

fixed(float* pSource = data)
{
    float* p = pSource;
 
    // A blank line should come before the loop.
    for(int i = 0; i < Size; i++)
    {
        *p = 1.0f;
        p++;
    }
}

Luc Pattyn wrote:

Yep, but you walked around what I really wanted to know:
which browser do you use for posting CP messages that include code snippets ?
and do you add any tags inside the PRE tags, such as CODE, BR, other ?

Luc Pattyn wrote:

So I still need to figure out why your snippets are harder to copy
wiht formatting than most others... (that's with IE6/IE7 towards Visual Studio).

fixed(float* pSource = data)
{
    float* p = pSource;

    pointerWatch.Start();

    for(int i = 0; i < Size; i++)
    {
        p[i] = 1.0f;
    }

    pointerWatch.Stop();
}