[prev in list] [next in list] [prev in thread] [next in thread]
List: wine-devel
Subject: [PATCH v2 1/4] msvcrt: Fix fmaf not to depend on rounding mode.
From: Piotr Caban <piotr () codeweavers ! com>
Date: 2022-01-31 20:36:16
Message-ID: c63e3c88-12e5-8052-6072-b517bf96f1d2 () codeweavers ! com
[Download RAW message or body]
Signed-off-by: Piotr Caban <piotr@codeweavers.com>
---
v2:
- remove nearbyint msvcr120 specific behavior and tests - it was fixed
in newer version of the dll
dlls/msvcrt/math.c | 16 ++++++++++------
1 file changed, 10 insertions(+), 6 deletions(-)
["0001-msvcrt-Fix-fmaf-not-to-depend-on-rounding-mode.txt" (text/x-patch)]
diff --git a/dlls/msvcrt/math.c b/dlls/msvcrt/math.c
index 9bb3e26570f..d03b4def801 100644
--- a/dlls/msvcrt/math.c
+++ b/dlls/msvcrt/math.c
@@ -5058,8 +5058,8 @@ double CDECL fma( double x, double y, double z )
float CDECL fmaf( float x, float y, float z )
{
union { double f; UINT64 i; } u;
- double xy, adjust;
- int e;
+ double xy, err;
+ int e, neg;
xy = (double)x * y;
u.f = xy + z;
@@ -5083,11 +5083,15 @@ float CDECL fmaf( float x, float y, float z )
* If result is inexact, and exactly halfway between two float values,
* we need to adjust the low-order bit in the direction of the error.
*/
- _controlfp(_RC_CHOP, _MCW_RC);
- adjust = fp_barrier(xy + z);
- _controlfp(_RC_NEAR, _MCW_RC);
- if (u.f == adjust)
+ neg = u.i >> 63;
+ if (neg == (z > xy))
+ err = xy - u.f + z;
+ else
+ err = z - u.f + xy;
+ if (neg == (err < 0))
u.i++;
+ else
+ u.i--;
return u.f;
}
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic