[Mesa-dev] [PATCH 5/7] util: Use SSE intrinsics in _mesa_lroundeven{f, }.
Roland Scheidegger
sroland at vmware.com
Mon Aug 3 16:17:57 PDT 2015
Am 03.08.2015 um 18:43 schrieb Matt Turner:
> gcc actually generates this for us now that we use -fno-math-errno
> (which is weird, since lrintf()/lrint() don't set errno) but clang still
> does not. Presumably helps MSVC as well.
>
> Reduced .text size by 8.5k with gcc before -fno-math-errno.
>
> text data bss dec hex filename
> 4935850 195136 26192 5157178 4eb13a i965_dri.so before
> 4927225 195128 26192 5148545 4e8f81 i965_dri.so after
> ---
> src/util/rounding.h | 22 ++++++++++++++++++++++
> 1 file changed, 22 insertions(+)
>
> diff --git a/src/util/rounding.h b/src/util/rounding.h
> index 088cf86..b0c9918 100644
> --- a/src/util/rounding.h
> +++ b/src/util/rounding.h
> @@ -25,6 +25,12 @@
> #define _ROUNDING_H
>
> #include <math.h>
> +#include <limits.h>
> +
> +#ifdef __x86_64__
> +#include <xmmintrin.h>
> +#include <emmintrin.h>
> +#endif
>
> #ifdef __SSE4_1__
> #include <smmintrin.h>
> @@ -87,7 +93,15 @@ _mesa_roundeven(double x)
> static inline long
> _mesa_lroundevenf(float x)
> {
> +#ifdef __x86_64__
> +#if LONG_BIT == 64
> + return _mm_cvtss_si64(_mm_load_ss(&x));
> +#elif LONG_BIT == 32
> + return _mm_cvtss_si32(_mm_load_ss(&x));
> +#endif
> +#else
> return lrintf(x);
> +#endif
> }
>
> /**
> @@ -97,7 +111,15 @@ _mesa_lroundevenf(float x)
> static inline long
> _mesa_lroundeven(double x)
> {
> +#ifdef __x86_64__
> +#if LONG_BIT == 64
> + return _mm_cvtsd_si64(_mm_load_sd(&x));
> +#elif LONG_BIT == 32
> + return _mm_cvtsd_si32(_mm_load_sd(&x));
> +#endif
> +#else
> return lrint(x);
> +#endif
> }
>
> #endif
>
Reviewed-by: Roland Scheidegger <sroland at vmware.com>
I'm wondering though if this is really exactly implementing lrintf why
we're using a different name and not just something like _mesa_lrintf().
Though arguably _mesa_roundeven is the same... A different name would be
more justifiable if this is not quite the same as lrintf (e.g. returning
int instead of long).
But anyway, that's not a deal breaker.
Roland
More information about the mesa-dev
mailing list