[PATCH v2 0/1] MR10805: Draft: msvcrt: only convert _tolower_l if not exact match already

[Stephan Seitz (＠theHamsta)]

This optimizations brought some speed-ups in
https://gitlab.winehq.org/wine/wine/-/merge_requests/10804 for the
equivalent `memicmp_strW` function as it avoids converting and locale
lookup for exact matches.

--
  v2: msvcrt: only convert _tolower_l if not exact match already
 
https://gitlab.winehq.org/wine/wine/-/merge_requests/10805

[PATCH v2 1/1] msvcrt: only convert _tolower_l if not exact match already

[Stephan Seitz]

From: Stephan Seitz <[e-mail]>

This optimizations brought some speed-ups in
https://gitlab.winehq.org/wine/wine/-/merge_requests/10804 for the
equivalent `memicmp_strW` function.
---
 dlls/msvcrt/string.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/dlls/msvcrt/string.c b/dlls/msvcrt/string.c
index 6b4ffbf45b5..2719ab37e9c 100644
--- a/dlls/msvcrt/string.c
+++ b/dlls/msvcrt/string.c
@@ -3435,8 +3435,11 @@ int __cdecl _memicmp_l(const void *v1, const void *v2, size_t len, _locale_t loc
 
     while (len--)
     {
-        if ((ret = _tolower_l(*s1, locale) - _tolower_l(*s2, locale)))
-            break;
+        if (*s1 != *s2) {
+            /* only convert _tolower_l if not exact match already */
+            if ((ret = _tolower_l(*s1, locale) - _tolower_l(*s2, locale)))
+                break;
+        }
         s1++;
         s2++;
     }
-- 
GitLab

https://gitlab.winehq.org/wine/wine/-/merge_requests/10805

[Alfred Agrell (＠Alcaro)]

If I wanted this function faster, my first idea would be changing _tolower_l (dlls/msvcrt/ctype.h) to ignore locinfo->pctype[c] & _UPPER in the c < 256 path, and always return locinfo->pclmap[c]. One memory access less if it is uppercase, and one branch less for all inputs.

Could be combined with your proposed optimization, of course.

But I would want numbers on how much impact those optimizations have.

-- 
https://gitlab.winehq.org/wine/wine/-/merge_requests/10805#note_138663