-
Notifications
You must be signed in to change notification settings - Fork 667
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Switching on CNTVCT for ThunderX2 platform and setting some optimization options. #150
base: master
Are you sure you want to change the base?
Conversation
…. Commented by now.
…ilation slow down.
…eduling flags, removed
AX_CHECK_COMPILER_FLAGS(-fsched-critical-path-heuristic, CFLAGS="$CFLAGS -fsched-critical-path-heuristic") | ||
AX_CHECK_COMPILER_FLAGS(-fsel-sched-pipelining, CFLAGS="$CFLAGS -fsel-sched-pipelining") | ||
AX_CHECK_COMPILER_FLAGS(-fselective-scheduling, CFLAGS="$CFLAGS -fselective-scheduling") | ||
AX_CHECK_COMPILER_FLAGS(-ftrapping-math, CFLAGS="$CFLAGS -ftrapping-math") |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Are all of these flags necessary for the performance improvements, or is it only one or two of them that really matters?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Are all of these flags necessary for the performance improvements, or is it only one or two of them that really matters?
Among about a hundred of checked flags only those were left, that have statistically significant influence on performance for the target platform.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Here are charts with several benchFFT tests for optimized vs default versions:
https://github.com/aur-ml/fftw3/wiki
Tested on ThunderX2 servers: the performance on 2^20 i-FFT problem (ib1048576) improves for about 20-25%.
The proposed changes does not influence other hardware architectures.