On 2/27/2015 5:49 PM, Bengt Larsson wrote:
Below are two benchmarks that explore maximum floating point
performance. loopm6 is double precision floating point and loopm6fp is
parallell single-precision. They are manually unrolled multiply-add
loops.
I used to reach 2.8 and 11 GFlops on these. Now I only get
2 and 6.
If you explore the inner loop with gcc -O2 -S you can see that it seems
to use few registers.
If you run them, there is a parameter expected. I use 30000 - 50000.
gcc 4.9.2-3 on 64-bit. I use gcc -O2.
May be
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53967
--
Problem reports: http://cygwin.com/problems.html
FAQ: http://cygwin.com/faq/
Documentation: http://cygwin.com/docs.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple