Databases
StackOverflow2013
Postsdbo
PO
 

Note that there are some explanatory texts on larger screens.

plurals
  1. PO
    primarykey
    data
    text
    <p>In optimized code you run the kernel with blocks half as large as in the non-optimized implementation.</p> <p>Let's call the size of the block in non-optimized code <code>work</code>, let half of this size be called <code>unit</code>, and let these sizes have same numerical value for the optimized code as well. </p> <p>In non-optimized code you run the kernel with as many threads as the <code>work</code> is, that is <code>blockDim = 2 * unit</code>. The code in each block just copies part of <code>g_idata</code> to an array in shared memory, of size <code>2 * unit</code>. </p> <p>In the optimized code <code>blockDim = unit</code>, so there are now 1/2 of the threads, and the array in shared memory is 2x smaller. In line 3 first summand comes from even units, while second from odd units. In this way all the data required for reduction is taken into account. </p> <p>Example: If you run non-optimized kernel with <code>blockDim=256=work</code> (single block, <code>unit=128</code>), then optimized code has a single block of <code>blockDim=128=unit</code>. Since this block gets <code>blockIdx=0</code>, the <code>*2</code> does not matter; the first thread does <code>g_idata[0] + g_idata[0 + 128]</code>. </p> <p>If you had 512 elements, and run non-optimized with 2 blocks of size 256 (<code>work=256</code>, <code>unit=128</code>), then optimized code has 2 blocks, but now of size 128. The first thread in second block (<code>blockIdx=1</code>) does <code>g_idata[2*128] + g_idata[2*128+128]</code>. </p>
    singulars
    1. This table or related slice is empty.
    1. POParallel Reduction
      singulars
      1. PTQuestion
    1. PTAnswer
    1. USP Marecki
    1. USP Marecki
    plurals
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. This table or related slice is empty.
    1. CO@P Marecki : Thank you very much. Your answer really helps me understanding the solution, but if you can let me know that in first paragraph, what is the total elements. If it is 256, then how 256 elements would be occupied by single block? Same question for 2nd paragraph 512 elements and only 2 blocks with 128 threads.
      singulars
      1. PO
      1. USrobot
    2. CO@robot : Total number of elements in `g_idata` is 256 in 1st paragraph, and 512 in second. True: in optimized code `sdata` is `2x` smaller (you only have `128` elements there, or `2 * 128` in 2nd paragraph), but this is enough for the purpose of reduction.
      singulars
      1. PO
      1. USP Marecki
    3. CO@P Marecki : Thanks for your reply. But if there are 256 elements then we have to process 256 elements, how elements are reduced to 128 elements? You mean there are 256 elements, and we are having 1 block with block size of 128 for processing 256 elements?
      singulars
      1. PO
      1. USrobot
 

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