Of course, it is. Thanks.
Allocating large blocks of memory: bare-metal C++ speeds https://lemire.me/blog/2020/01/17/allocating-large-blocks-of-memory-bare-metal-c-speeds/ …
with credit to @trav_downs
-
-
Tweet je nedostupan.
-
-
-
-
for large allocations should you not be using huge pages as well? Doesn't make sense to do large allocations 4KB at a time. What I'd be even more interested in is performance impact of memory deallocation.
-
I enable transparent huge pages in some of the test.
Kraj razgovora
Novi razgovor -
-
-
When I wrote my own malloc lib (2007), the speed barrier I had was zerofilling by the kernel. If new/malloc implements huge memory allocation/deallocation with mmap, you get zerofilled memory already. A "user" zerofill could be wasteful.
- Još 4 druga odgovora
Novi razgovor -
-
-
Thank you for this benchmark. Out of curiousity, I added a case allocating`std::vector<char>(size)`, and to my surprise it runs as fast as `calloc` or `new_and_memset`. (g++-7 on Ubuntu 18.04, Xeon E5-2620). Are there any drawbacks to using it, as the speed hit is minor enough?
-
Interesting. Under the covers, vector is probably using something like std::unitialized_fill, for gcc seems to be able figure to use a memset, whereas for a simple loop, not.https://godbolt.org/z/iYz2kL
- 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.