I was trying to be less exhaustive this time, so I one think I didn't mention is that the roughly 30x result holds over a fairly large range of buffer sizes, but not for small buffers (where constant overhead slows things down), and also not for buffers outside the L2 ...
Filling large arrays with zeroes quickly in C++
https://lemire.me/blog/2020/01/20/filling-large-arrays-with-zeroes-quickly-in-c/ …
cc @trav_downs @richardstartin
-
-
-
... where the bottleneck is L3 or DRAM thoughput. The "other" way (byte-by-byte) is so slow is basically the same speed no matter how large the buffer, because the CPU is always the bottleneck.
- Još 1 odgovor
Novi razgovor -
-
-
Is some of this down to a char* being used when wanting to zero the array vs an int*?
-
In a way, int* is a middle ground. At -O2 with gicc, it will never be really fast: it will always use an int-at-a-time loop, but that loop is at least 4x as fast (per byte) as a byte-by-byte loop, since 4 bytes are written at a time. int isn't even eligible for the libstc++ ...
- Još 1 odgovor
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.