cuda - Is any optimization done if one run the same kernel with the same input again and again? -

-

if run same kernel with same input several times, this

#define n 2000 for(int = 0; < 2000; i++) {     mykernel<<<1,120>>>(...); }

what happens? timed , played around n: halving n (to 1000), halved time took.

yet i'm bit cautious belive runs kernel 2000 times because speed non-cuda code dramatic (~900 sec ~0.9 sec). kind of optimization cuda in case? caching results?

setting cuda_launch_blocking=1 didn't change nothing.

mykernel replaces inner loop in non-cuda code.

hardware geforce gtx 260

cuda doesn't optimization of kind, or caching of results. if launch 2000 kernels, runs 2000 kernels.

however, kernel launches asynchronous, measuring time taken launch 2000 kernel instances in loop isn't same total execution time of 2000 kernel instances. seeing artifact of incorrect time measurement , not true speed-up.


Comments

Post a Comment

Popular posts from this blog

c# - how to write client side events functions for the combobox items -

-
<dx:aspxcombobox id="aspxcombobox2" runat="server" clientidmode="autoid" selectedindex="-1" onselectedindexchanged="aspxcombobox1_selectedindexchanged1" > <clientsideevents selectedindexchanged="function(s, e) { grid.selectrows();}" /> <items> <dx:listedititem text="selecte rows" /> <dx:listedititem text="unselectall" /> </items> </dx:aspxcombobox> i want write event "unselect all" item in combobox. how can this? you'll need write javascript function iterates thru combo-box , marks selected items false. see select/unselect checkbox fields using pure javascript , html sample code msdn:
Read more

exception - Python, pyPdf OCR error: pyPdf.utils.PdfReadError: EOF marker not found -

-
pypdf throws exception: pypdf.utils.pdfreaderror: eof marker not found i don't need fix pypdf, need eof error cause "except" block execute , skip on file, doesn't work. still causes program stop running. background: batch ocr program pdfs python, pypdf, adobe pdf ocr error: unsupported filter /lzwdecode ... saga continues. i got 10,000 pdfs in folder. ocrd, not. can't tell 'em apart. step 1 figure out ones not ocrd , ocr (see other threads details). so i'm using pypdf. exceptions related unrecognized characters , unsupported filters when try read text. guestimated if throws exception, it's got text in , doens't go in list. problem solved, right? so: pypdf import pdffilewriter, pdffilereader import sys, os, pypdf, re path = 'c:\users\homer\documents\my pdfs' filelist = os.listdir(path) has_text_list = [] does_not_have_text_list = [] pdf_name in filelist: pdf_file
Read more