git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH] Properly align memory allocations and temporary buffers

From
René Scharfe <l.s.r@web.de>
Date
Jan 7, 2022, 20:56 UTC
Message-ID
<40077881-e91a-dfba-5950-cdc03719264d@web.de>
In-Reply-To
<xmqq7dbb9wcz.fsf@gitster.g>
Am 07.01.22 um 20:30 schrieb Junio C Hamano:
Show 16 quoted lines
> "brian m. carlson" <sandals@crustytoothpaste.net> writes:
>
>> I'm not opposed to a small amount of finagling for this case, but I am
>> very much opposed to defining your C ABI in an intentionally difficult
>> way.
>
> I was looking at git_qsort_s() today and thought that its use of
> "char buf[1k]" on stack is something we would write for a system
> without alloca().
>
> We already have xalloca() wrapper defined in the compat-util header.
> Perhaps it is a good idea to use it instead?
>
> Both the true alloca() and xmalloc() we use as a fallback (when
> <alloca.h> is not available) are supposed to return a block of
> memory that is suitably aligned for any use, no?

If falling back to malloc(3) is fine in general then I wonder if we can just it use always. This would save two branches and make buffer management trivial here. How much worse is malloc(3) on platforms with alloca(3)? Do we sort lots of short lists somewhere? In other words: Does this stack allocation optimization actually make a measurable difference?

Heap fragmention should not be a concern here, at least, because the pattern of requesting, using and releasing a single allocation won't leave holes.

Show 41 quoted lines
>
>  compat/qsort_s.c | 20 +++++++++++---------
>  1 file changed, 11 insertions(+), 9 deletions(-)
>
> diff --git c/compat/qsort_s.c w/compat/qsort_s.c
> index 52d1f0a73d..63660a4304 100644
> --- c/compat/qsort_s.c
> +++ w/compat/qsort_s.c
> @@ -49,21 +49,23 @@ int git_qsort_s(void *b, size_t n, size_t s,
>  		int (*cmp)(const void *, const void *, void *), void *ctx)
>  {
>  	const size_t size = st_mult(n, s);
> -	char buf[1024];
> +	char *tmp;
> +	int use_alloca;
>
>  	if (!n)
>  		return 0;
>  	if (!b || !cmp)
>  		return -1;
>
> -	if (size < sizeof(buf)) {
> -		/* The temporary array fits on the small on-stack buffer. */
> -		msort_with_tmp(b, n, s, cmp, buf, ctx);
> -	} else {
> -		/* It's somewhat large, so malloc it.  */
> -		char *tmp = xmalloc(size);
> -		msort_with_tmp(b, n, s, cmp, tmp, ctx);
> +	use_alloca = (size < 1024);
> +
> +	tmp = use_alloca ? xalloca(size) : xmalloc(size);
> +
> +	msort_with_tmp(b, n, s, cmp, tmp, ctx);
> +
> +	if (use_alloca)
> +		xalloca_free(tmp);
> +	else
>  		free(tmp);
> -	}
>  	return 0;
>  }
Previous: Jessica ClarkeNext: Junio C Hamano
Message 15 of 32 in “Properly align memory allocations and temporary buffers”
  1. Properly align memory allocations and temporary buffersJessica Clarke, Jan 5, 2022
  2. Taylor BlauJan 6, 2022
  3. Jessica ClarkeJan 6, 2022
  4. Junio C HamanoJan 6, 2022
  5. Jessica ClarkeJan 6, 2022
  6. Junio C HamanoJan 7, 2022
  7. Jessica ClarkeJan 7, 2022
  8. brian m. carlsonJan 7, 2022
  9. Jessica ClarkeJan 7, 2022
  10. brian m. carlsonJan 7, 2022
  11. Jessica ClarkeJan 7, 2022
  12. Jessica ClarkeJan 7, 2022
  13. Junio C HamanoJan 7, 2022
  14. Jessica ClarkeJan 7, 2022
  15. René ScharfeJan 7, 2022
  16. Junio C HamanoJan 7, 2022
  17. René ScharfeJan 7, 2022
  18. Elijah NewrenJan 8, 2022
  19. brian m. carlsonJan 6, 2022
  20. Jessica ClarkeJan 6, 2022
  21. Philip OakleyJan 7, 2022
  22. René ScharfeJan 7, 2022
  23. Jessica ClarkeJan 7, 2022
  24. Jessica ClarkeJan 12, 2022
  25. René ScharfeJan 12, 2022
  26. Jessica ClarkeJan 12, 2022
  27. mem-pool: Don't assume uintmax_t is aligned enough for all typesJessica Clarke, Jan 23, 2022
  28. Junio C HamanoJan 23, 2022
  29. Jessica ClarkeJan 23, 2022
  30. Junio C HamanoJan 23, 2022
  31. mem-pool: Don't assume uintmax_t is aligned enough for all typesJessica Clarke, Jan 23, 2022
  32. Junio C HamanoJan 24, 2022

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.