ifort gives wrong answer for simple addition!

[Boris_B_]

Here is a program to add 1 twenty times. Answer should be 20, but ifort 10.1 with default optimization gives 74 - try it please:

real a,c

a=1.

c=0.0

do i=1,20

c=a+c

end do

write(*,*)'sum=',c

end

Switching off optimization gives the right answer. Help appreciated.

[TimP]

What's worse, it still does that in more recent versions, but the problem goes away when diagnostics are set.
Of course, it's associated with vectorization of sum reduction.
You're certainly entitled to file a problem report.

[Steven_L_Intel1]

I can reproduce this in the latest 11.0, but the beta of 11.1 chooses not to vectorize this and gets the right answer. I will report this to the developers.

Edit: Even if vectorization is forced in 11.1 the answer is correct. I have reported this as issue DPD200111340.

[TimP]

Quoting - Steve Lionel (Intel)

I can reproduce this in the latest 11.0, but the beta of 11.1 chooses not to vectorize this and gets the right answer. I will report this to the developers.

Edit: Even if vectorization is forced in 11.1 the answer is correct. I have reported this as issue DPD200118979.

Intel Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Beta Build 20081213 Package ID: l_cprof_b_11.1.016
Copyright (C) 1985-2008 Intel Corporation. All rights reserved.

ifort: NOTE: The Beta evaluation period for this product ends on 28-jul-2009 UTC.
Intel Fortran 11.1-2332
bb.f90(4): (col. 1) remark: LOOP WAS VECTORIZED.
GNU ld (Linux/GNU Binutils) 2.19.50.0.1.20081007
tim@tim-t61:~/src/net> ./a.out
sum= 74.00000

Doesn't attempt vectorization if -fp-model source is set.

[Steven_L_Intel1]

Interesting. I had tried IA-32. I'll add that to my notes.

[Steven_L_Intel1]

This bug did get fixed for 11.1. Sorry for the late notice.