gpu - Linking with 3rd party CUDA libraries slows down cudaMalloc -


it not secret on cuda 4.x first call cudamalloc can ridiculously slow (which reported several times), seemingly bug in cuda drivers.

recently, noticed weird behaviour: running time of cudamalloc directly depends on how many 3rd-party cuda libraries linked program (note not use these libraries, link program them)

i ran tests using following program:

int main() {   cudasetdevice(0);   unsigned int *ptr = 0;   cudamalloc((void **)&ptr, 2000000 * sizeof(unsigned int));      cudafree(ptr); return 1; }

the results follows:

  • linked with: -lcudart -lnpp -lcufft -lcublas -lcusparse -lcurand running time: 5.852449

  • linked with: -lcudart -lnpp -lcufft -lcublas running time: 1.425120

  • linked with: -lcudart -lnpp -lcufft running time: 0.905424

  • linked with: -lcudart running time: 0.394558

according 'gdb', time indeed goes cudamalloc, it's not caused library initialization routine..

i wonder if has plausible explanation ?

in example, cudamalloc call initiates lazy context establishment on gpu. when runtime api libraries included, binary payloads have inspected , gpu elf symbols , objects contain merged context. more libraries there are, longer can expect process take. further, if there architecture mismatch in of cubins , have backwards compatible gpu, can trigger driver recompilation of device code target gpu. in extreme case, have seen old application linked old version of cublas take 10s of seconds load , initialise when run on fermi gpu.

you can explicitly force lazy context establishment issuing cudafree call this:

int main() {     cudasetdevice(0);     cudafree(0); // context establishment happens here     unsigned int *ptr = 0;     cudamalloc((void **)&ptr, 2000000 * sizeof(unsigned int));        cudafree(ptr);   return 1; }

if profile or instrument version timers should find first cudafree call consumes of runtime , cudamalloc call becomes free.


Comments

Post a Comment

Popular posts from this blog

jquery - How can I dynamically add a browser tab? -

since using iframes house external pages are, accounts "not preferred method" , disallowed sites (flickr, one), want dynamically generate browser tabs. e.g., when user submits form, want dynamically generate tab on browser in response submitted. e.g., might want add new tab url like: http://www.bigsurgarrapata.com/contact how in jquery? you want form target attribute action attribute. http://www.w3schools.com/tags/att_form_target.asp browsers have ultimate control on result of target attribute open new tab value of _blank. <form action="contact_submit.php" target="_blank" method="post">
Read more

node.js - Getting the socket id,user id pair of a logged in user(s) -

i have web application done using node js , in it,i have authentication system requires users log in before can access pages. have 2 pages,the homepage , updates page. from read docs,a new socket id issued when user opens tab in browser. me being admin,i know socket id of logged in user , more so,the socket id,logged in user id pair logged in users. what method can use socket id,logged in user id pair make sure broadcast realtime info users online?.
Read more

keyboard - C++ GetAsyncKeyState alternative -

i using windows.h library's getasynckeystate function define key pushed on keyboard . here program returns int value of pushed key example. #include <iostream> #include <windows.h> using namespace std; int main (){ char i; for(i=8;i<190;i++){ if(getasynckeystate(i)== -32767){ cout<<int (i)<<endl; } } return 0; } but trying make opengl simple game , function appeared slow . there function not need cycle 180 times , if statement . thanks you're trying make opengl game, you're going use glfw. if that's case can away using glfwgetkey() such: if (glfwgetkey(glfw_key_up) == glfw_press) {...} if (glfwgetkey(glfw_key_down) == glfw_press) {...} that method nicely explained in tutorial 6 opengl-tutorial.org . should start first tutorial of beginners tutorials , later intermediate tutorials . if you're going make win32 window can listen wm_keydown or wm_keyup messages , virtua
Read more