Below are results of running "the big keybench" benchmark of Xonotic under "perf". They're negligible, still, it might make more difference for slower machines or distinct workloads.
branch-misses: vanilla-wine | patched-wine | improvement ------------------------------|------------ 1,757,439,276 | 1,749,075,739 | -8,363,537 1,750,670,918 | 1,746,169,971 | -4,500,947 1,746,488,762 | 1,748,569,739 | 2,080,977 1,753,328,811 | 1,739,515,555 | -13,813,256
Command line example for testing: export WINEPREFIX=~/.wineTESTING; perf stat -B -e cache-misses,branch-misses,faults ~/Projects/wine/test-build/loader/wine ./xonotic-x86.exe -benchmark demos/the-big-keybench
Changes in v2: * Don't put custom macro into WinAPI header (Alexandre Julliard) * Capitalize macros (Me)
Signed-off-by: Konstantin Kharlamov Hi-Angel@yandex.ru --- include/wine/debug.h | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
diff --git a/include/wine/debug.h b/include/wine/debug.h index 07ac88d264..53d61f16b8 100644 --- a/include/wine/debug.h +++ b/include/wine/debug.h @@ -57,8 +57,17 @@ struct __wine_debug_channel char name[15]; };
+/* Branch hints for performance critical code */ +#ifdef __GNUC__ +# define LIKELY(expr) (__builtin_expect(expr, 1)) +# define UNLIKELY(expr) (__builtin_expect(expr, 0)) +#else +# define LIKELY(expr) (expr) +# define UNLIKELY(expr) (expr) +#endif + #ifndef WINE_NO_TRACE_MSGS -# define __WINE_GET_DEBUGGING_TRACE(dbch) ((dbch)->flags & (1 << __WINE_DBCL_TRACE)) +# define __WINE_GET_DEBUGGING_TRACE(dbch) UNLIKELY((dbch)->flags & (1 << __WINE_DBCL_TRACE)) #else # define __WINE_GET_DEBUGGING_TRACE(dbch) 0 #endif
-
Konstantin Kharlamov