Databases
StackOverflow2013
Postsdbo
POCode runs faster using placement new
 

Note that there are some explanatory texts on larger screens.

plurals
  1. POCode runs faster using placement new
    primarykey
    data
    text
    <p>I got this class,</p> <p><strong>Approach 1:</strong></p> <pre><code>typedef float v4sf __attribute__ (vector_size(16)) class Unit { public: Unit(int num) { u = new float[num]; v = new float[num]; } void update() { for(int i =0 ; i &lt; num; i+=4) { *(v4sf*)&amp;u[i] = *(v4sf*)&amp;v[i] + *(v4sf*)&amp;t[i]; //many other equations } } float*u,*v,*t; //and many other variables } </code></pre> <p><strong>Approach 2:</strong></p> <p>Same as approach 1. Except that in approach 2, <code>v</code>,<code>u</code>, and all other variables are allocated on a big chunk pre-allocated on heap, using placement <code>new</code>.</p> <pre><code>typedef float v4sf __attribute__ (vector_size(16)) class Unit { public: Unit(int num) { buffer = new char[num*sizeof(*u) + sizeof(*v) /*..and so on for other variables..*/] u = new(buffer) float[num]; v = new(buffer+sizeof(float)*num) float[num]; //And so on for other variables } void update() { for(int i =0 ; i &lt; num; i+=4) { *(v4sf*)&amp;u[i] = *(v4sf*)&amp;v[i] + *(v4sf*)&amp;t[i]; //many other equations } } char* buffer; float*u,*v,*t; //and many other variables } </code></pre> <p>However, approach 2 is 2x faster. Why is that?</p> <p>There are around 12 float variables and num is 500K. update() is called <code>1k</code> times. The speed doesnt factor in the memory allocation. I measure the speed like this:</p> <pre><code>double start = getTime(); for( int i = 0; i &lt; 1000; i++) { unit-&gt;update(); } double end = getTime(); cout&lt;&lt;end - start; </code></pre> <p>And this is around 2x faster in approach 2. </p> <p>Compiler options: <code>gcc -msse4 -o3 -ftree-vectorize.</code></p> <p>L1 cache is 256K, Ram is 8GB, pagesize is 4K. </p> <p>Edit: Corrected the mistake in allocating the variables in approach 2. All variables are allocated in different sections, correctly. Processor is Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz</p> <p>Edit: added the source here - <a href="https://gist.github.com/11e94d3fba7e7f9ea2a4" rel="nofollow">Source</a>. Approach 1) gives 69.58s , Approach 2) gives 46.74s. Though not 2x faster, it is still fast. </p>
    singulars
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. PTQuestion
    1. USexcray
    1. USexcray
    plurals
    1. PL
      singulars
      1. LTLinked
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. PO
      singulars
      1. PTAnswer
    2. PO
      singulars
      1. PTAnswer
    3. PO
      singulars
      1. PTAnswer
    1. VO
      singulars
      1. POCode runs faster using placement new
      1. This table or related slice is empty.
      1. VTUpMod
    2. VO
      singulars
      1. POCode runs faster using placement new
      1. This table or related slice is empty.
      1. VTDownMod
    3. VO
      singulars
      1. POCode runs faster using placement new
      1. USrgngl
      1. VTFavorite
    1. COCan you show actual code for approach 2?
      singulars
      1. POCode runs faster using placement new
      1. USR. Martinho Fernandes
    2. COWhat is the processor? I have a strong feeling that you may be running into a similar situation as: [Why is one loop so much slower than two loops?](http://stackoverflow.com/questions/8547778/why-is-one-loop-so-much-slower-than-two-loops)
      singulars
      1. POCode runs faster using placement new
      1. USMysticial
    3. COu and v are the same now. I think you should use v = new(buffer + sizeof(float)*num)
      singulars
      1. POCode runs faster using placement new
      1. USholgac
 

Querying!

 
Guidance
A row detail

Detail views are divided into sections. All the information in the data section comes from columns in the selected row. The other sections display data from other, related rows.

Related data can be related in a to-one or a to-many fashion. Captions of data related in a to-many fashion link to a list view showing a filtered view of the table.

Try moving around until you find a non-empty to-many entry and click on the label to get to one. You can move back to the root by clicking on the database name in the header.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload