diff options
| -rw-r--r-- | src/arena.c | 24 |
1 files changed, 22 insertions, 2 deletions
diff --git a/src/arena.c b/src/arena.c index 20083b5..23d374e 100644 --- a/src/arena.c +++ b/src/arena.c @@ -91,12 +91,32 @@ _arena_grow(arena_t *a, void *ptr, size_t old_nmemb, size_t new_nmemb, more trailing free space in the region to avoid a memcpy(). */ size_t oldsz = old_nmemb * size; if ((char *)ptr == (char *)p->free - oldsz) { - size_t rest = p->cap - ((char *)p->free - (char *)p->data); + size_t rem = p->cap - ((char *)p->free - (char *)p->data); size_t need = (new_nmemb - old_nmemb) * size; - if (need <= rest) { + if (need <= rem) { p->free = (char *)p->free + need; return ptr; } + + /* NOTE: After benchmarking, (albiet rather naïvely) it seems + that the closer to the original mapping size we are, the + faster mremap() runs. With really large mappings it + converges with mmap() + memcpy() on performance though, so + it seems this is always the better solution on systems + with mremap() since it’s also a bit more memory efficient. + + We can’t pass MREMAP_MAYMOVE though, because it would + invalidate existing pointers to data in the mapping. */ +#if __linux__ + size_t ncap = p->cap + need - rem; + void *nptr = mremap(p->data, p->cap, ncap, 0); + if (nptr != MAP_FAILED) { + p->cap = ncap; + return ptr; + } + if (errno != ENOMEM) + err("%s:", __func__); +#endif } void *dst = arena_alloc(a, new_nmemb, size, align); |