Precompute embedded string literals hash code #10596
Conversation
etiennebarrie
force-pushed
the
interned-string-hashcode
branch
from
5281aa9
to
303929b
Compare
This comment has been minimized.
This comment has been minimized.
string.c
Outdated
| RUBY_ASSERT(free_bytes >= sizeof(st_index_t)); | ||
| #endif | ||
|
|
||
| *(st_index_t *)(RSTRING_END(str) + TERM_LEN(str)) = rb_str_hash(str); |
There was a problem hiding this comment.
I am afraid if this may cause a bus error on some platform. Is it safe?
There was a problem hiding this comment.
this may cause a bus error on some platform
Because of alignment? If it's a concern, I think we could ensure it's always aligned, the overhead would be very slightly larger.
There was a problem hiding this comment.
Yes > because of alignment
It would be safer to use aligned address if possible.
There was a problem hiding this comment.
Thanks, I'll look into it.
There was a problem hiding this comment.
| *(st_index_t *)(RSTRING_END(str) + TERM_LEN(str)) = rb_str_hash(str); | |
| typedef struct {char bytes[sizeof(st_index_t)];} unaligned_index; | |
| union {st_index_t i; unaligned_index b;} u = {.i = rb_str_hash(str)}; | |
| *(unaligned_index *)(RSTRING_END(str) + TERM_LEN(str)) = u.b; |
There was a problem hiding this comment.
Thank @nobu, interesting trick. I suppose we need it on read as well?
casperisfine
force-pushed
the
interned-string-hashcode
branch
from
303929b
to
54c8bda
Compare
|
Looks like we broke something after the refactor, I'll investigate. |
casperisfine
force-pushed
the
interned-string-hashcode
branch
from
54c8bda
to
6255749
Compare
| if (FL_TEST_RAW(str, STR_PRECOMPUTED_HASH)) { | ||
| real_size += sizeof(st_index_t); | ||
| } |
There was a problem hiding this comment.
This was the bug. The compactor thought the string could fit in a smaller slot, corrupting the precomputed hash.
casperisfine
force-pushed
the
interned-string-hashcode
branch
5 times, most recently
from
2bd83d4
to
8f949a4
Compare
|
Alright, I think the bugs have been squashed now. |
etiennebarrie
force-pushed
the
interned-string-hashcode
branch
from
8f949a4
to
39fae7c
Compare
casperisfine
force-pushed
the
interned-string-hashcode
branch
from
39fae7c
to
44bc1f4
Compare
With embedded strings we often have some space left in the slot, which we can use to store the string Hash code. It's probably only worth it for string literals, as they are the ones likely to be used as hash keys. We chose to store the Hash code right after the string terminator as to make it easy/fast to compute, and not require one more union in RString. ``` compare-ruby: ruby 3.4.0dev (2024-04-22T06:32:21Z main f77618c) [arm64-darwin23] built-ruby: ruby 3.4.0dev (2024-04-22T10:13:03Z interned-string-ha.. 8a1a32331b) [arm64-darwin23] last_commit=Precompute embedded string literals hash code | |compare-ruby|built-ruby| |:-----------|-----------:|---------:| |symbol | 39.275M| 39.753M| | | -| 1.01x| |dyn_symbol | 37.348M| 37.704M| | | -| 1.01x| |small_lit | 29.514M| 33.948M| | | -| 1.15x| |frozen_lit | 27.180M| 33.056M| | | -| 1.22x| |iseq_lit | 27.391M| 32.242M| | | -| 1.18x| ``` Co-Authored-By: Étienne Barrié <etienne.barrie@gmail.com>
etiennebarrie
force-pushed
the
interned-string-hashcode
branch
from
44bc1f4
to
85b5a30
Compare
[Feature #20415]
With embedded strings we often have some space left in the slot, which we can use to store the string Hash code.
It's probably only worth it for string literals, as they are the ones likely to be used as hash keys.
We chose to store the Hash code right after the string terminator as to make it easy/fast to compute, and not require one more union in RString.