A small-string optimization is a way of reusing inline storage space for
sufficiently small strings, rather than allocating them on the heap. The
current approach takes after an old Facebook string class: it reuses the
highest-order byte for flags and small-string size, in such a way that a
maximally-sized small string will have its last byte zeroed, making it a
null terminator for the C string.
The only flag we have is in the highest-order bit, that says whether the
string is big (set) or small (cleared.) Most of the logic switches based
on the value of this bit; e.g. data() returns big()->p if it's set, else
small()->buf if it's cleared. For a small string, the capacity is always
fixed at sizeof(string) - 1 bytes; we store the length in the last byte,
but we store it as the number of remaining bytes of capacity, so that at
max size, the last byte will read zero and serve as our null terminator.
Morally speaking, our class's storage is a union over two POD C structs.
For now I gravitated towards a slightly more obtuse approach: the string
class itself contains a blob of the right size, and we alias that blob's
pointer for the two structs, taking some care not to run afoul of object
lifetime rules in C++. If anyone wants to improve on this, contributions
are welcome.
This commit also introduces the `ctl::__` namespace. It can't be legally
spelled by library users, and serves as our version of boost's "detail".
We introduced a string::swap function, and we now use that in operator=.
operator= now takes its argument by value, so we never need to check for
the case where the pointers are equal and can just swap the entire store
of the argument with our own, leaving the C++ destructor to free our old
storage afterwards.
There are probably still a few places where our capacity is slightly off
and we grow too fast, although there don't appear to be any where we are
too slow. I will leave these to be fixed in future changes.
If pthread_create() is linked into the binary, then the cosmo runtime
will create an independent dlmalloc arena for each core. Whenever the
malloc() function is used it will index `g_heaps[sched_getcpu() / 2]`
to find the arena with the greatest hyperthread / numa locality. This
may be configured via an environment variable. For example if you say
`export COSMOPOLITAN_HEAP_COUNT=1` then you can restore the old ways.
Your process may be configured to have anywhere between 1 - 128 heaps
We need this revision because it makes multithreaded C++ applications
faster. For example, an HTTP server I'm working on that makes extreme
use of the STL went from 16k to 2000k requests per second, after this
change was made. To understand why, try out the malloc_test benchmark
which calls malloc() + realloc() in a loop across many threads, which
sees a a 250x improvement in process clock time and 200x on wall time
The tradeoff is this adds ~25ns of latency to individual malloc calls
compared to MODE=tiny, once the cosmo runtime has transitioned into a
fully multi-threaded state. If you don't need malloc() to be scalable
then cosmo provides many options for you. For starters the heap count
variable above can be set to put the process back in single heap mode
plus you can go even faster still, if you include tinymalloc.inc like
many of the programs in tool/build/.. are already doing since that'll
shave tens of kb off your binary footprint too. Theres also MODE=tiny
which is configured to use just 1 plain old dlmalloc arena by default
Another tradeoff is we need more memory now (except in MODE=tiny), to
track the provenance of memory allocation. This is so allocations can
be freely shared across threads, and because OSes can reschedule code
to different CPUs at any time.
