parallel processing - With OpenMP parallelized nested loops run slow -


i've got part of fortran program consisting of nested loops want parallelize openmp. 

integer :: nstates , n, i, dima, dimb, dimc, a_row, b_row, b_col, c_row, row, col double complex, dimension(4,4):: mat double complex, dimension(:), allocatable :: vecin,vecout   nstates = 2 n = 24  allocate(vecin(nstates**n), vecout(nstates**n)) vecin = ...some data vecout = 0  mat = reshape([...some data...],[4,4])  dimb=nstates**2  !$omp parallel private(dima,dimc,row,col,a_row,b_row,c_row,b_col)  i=1,n-1     dima=nstates**(i-1)     dimc=nstates**(n-i-1)      a_row = 1, dima         b_row = 1,dimb             c_row = 1,dimc                 row = ((a_row-1)*dimb + b_row - 1)*dimc + c_row                 b_col = 1,dimb                     col = ((a_row-1)*dimb + b_col - 1)*dimc + c_row                     !$omp atomic                     vecout(row) = vecout(row) + vecin(col)*mat(b_row,b_col)                 end             end         end     end end !$omp end parallel

the program runs , result correct, it's incredible slow. slower without openmp. don't know openmp. have done wrong use of private or omp atomic? grateful every advice how improve performance of code.

if arrays large , stack overflows automatic reduction, can implement reduction allocatable temporary arrays.

as francois jacq pointed out, have race condition caused dima , dimb should private.

double complex, dimension(:), allocatable :: tmp  !$omp parallel private(dima,dimb,row,col,a_row,b_row,c_row,b_col,tmp)  allocate(tmp(size(vecout))) tmp = 0  !$omp do i=1,n-1     dima=nstates**(i-1)     dimc=nstates**(n-i-1)      a_row = 1, dima         b_row = 1,dimb             c_row = 1,dimc                 row = ((a_row-1)*dimb + b_row - 1)*dimc + c_row                 b_col = 1,dimb                     col = ((a_row-1)*dimb + b_col - 1)*dimc + c_row                     tmp(row) = tmp(row) + vecin(col)*mat(b_row,b_col)                 end             end         end     end end !$omp end  !$omp critical vecout = vecout + tmp !$omp end critical !$omp end parallel

Comments

Post a Comment

Popular posts from this blog

Payment information shows nothing in one page checkout page magento -

i using magento ver. 1.9.1.0 want remove shipping information , shipping method step 1 page checkout page. changed code in core file like in app\code\core\mage\checkout\block\onepage.php public function getsteps() { $steps = array(); $stepcodes = $this->_getstepcodes(); /* remove shipping address , shipping method */ $excludestep=array_diff($stepcodes,array('shipping','shipping_method')); //$excludestep=$stepcodes; if ($this->iscustomerloggedin()) { $stepcodes = array_diff($excludestep, array('login')); } foreach ($excludestep $step) { $steps[$step] = $this->getcheckout()->getstepdata($step); } return $steps; } and in app\code\core\mage\checkout\controllers\onepagecontroller.php public function savebillingaction() { if ($this->_expireajax()) { return; } if ($this->getrequest()-...
Read more

tcpdump - How to check if server received packet (acknowledged) -

Image
i running tcpdump on server capture traffic between server (servera) , remote server (serverb). not know how read output tcpdump. point out if serverb acknowledging packets being sent servera being received serverb? here tcpdump log: 2015-03-31 11:32:18.038284 ip (tos 0x0, ttl 64, id 30781, offset 0, flags [df], proto tcp (6), length 60) servera.14033 > serverb.8004: flags [s], cksum 0x02bd (correct), seq 337482636, win 14600, options [mss 1460,nop,nop,ts val 506588081 ecr 0,nop,wscale 7], length 0 0x0000: 4500 003c 783d 4000 4006 bae4 0a04 7a0e e.. servera.14033: flags [s.], cksum 0x3b2a (correct), seq 1742982091, ack 337482637, win 4140, options [mss 1380,nop,wscale 0], length 0 0x0000: 4500 0030 5ff2 0000 f606 5d3b cefd b48a e..0_.....];.... 0x0010: 0a04 7a0e 1f44 36d1 67e3 cbcb 141d 938d ..z..d6.g....... 0x0020: 7012 102c 3b2a 0000 0204 0564 0103 0300 p..,;*.....d.... 2015-03-31 11:32:18.078622 ip (tos 0x0, ttl 64, id 30782, offset 0, flags [...
Read more