Q: Producer/Consumer ring is a common pattern for high perf comm between 2 CPU cores or CPU core & device. Thus, I expected Intel to have non-temporal store instruction that write to LLC without polluting L1/L2. Useful also with device DDIO. But MOVNT* also bypass LLC. Why? (1/3)
-
-
Correct that today there is no instruction that allows you to directly place data in a particular cache level like you describe. In general, cache management via software is hard (my opinion is similar with SW prefetching although that is starting to become more mature finally).
-
You may want to look at the CLDEMOTE instruction coming in future HW. Then you could do a store+CLDEMOTE to get behavior similar to what you are looking for.
- Još 5 drugih odgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.