You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Problem
By default, the prefetch instruction in x86 is implemented as normal load instruction (with prefetch flag in the request), and blocks the O3 cpu to wait for the response when we are using Ruby. This causes performance problem as the latency is exposed to the pipeline.
Possible Solution
Currently, the classic cache returns early for the prefetch request. We could also implement this behavior in the RubySequencer. This would fix this issue for all CPU models and Ruby protocols.
The text was updated successfully, but these errors were encountered:
Problem
By default, the prefetch instruction in x86 is implemented as normal load instruction (with prefetch flag in the request), and blocks the O3 cpu to wait for the response when we are using Ruby. This causes performance problem as the latency is exposed to the pipeline.
Possible Solution
Currently, the classic cache returns early for the prefetch request. We could also implement this behavior in the RubySequencer. This would fix this issue for all CPU models and Ruby protocols.
The text was updated successfully, but these errors were encountered: