Rounding broken for negative numbers about libfixmath HOT 5 CLOSED

Fix for the general 64bit cases:

Index: fix16.c
===================================================================
--- fix16.c (revision 51)
+++ fix16.c (working copy)
@@ -35,7 +35,7 @@
    #ifndef FIXMATH_NO_64BIT
    int64_t tempResult = ((int64_t)inArg0 * (int64_t)inArg1);
    #ifndef FIXMATH_NO_ROUNDING
-   tempResult += (fix16_one >> 1);
+   tempResult += (tempResult > 0) ? (fix16_one >> 1) : -(fix16_one >> 1);
    #endif
    tempResult >>= 16;
    return tempResult;
@@ -97,7 +97,7 @@
    int64_t tempResult = inArg0;
    tempResult <<= 16;
    #ifndef FIXMATH_NO_ROUNDING
-   tempResult += (inArg1 >> 1);
+   tempResult += (tempResult > 0) ? (inArg1 >> 1) : -(inArg1 >> 1);
    #endif
    tempResult /= inArg1;
    return tempResult;

Actually, for it to work also for result = 0:

Index: fix16.c
===================================================================
--- fix16.c (revision 51)
+++ fix16.c (working copy)
@@ -35,7 +35,7 @@
    #ifndef FIXMATH_NO_64BIT
    int64_t tempResult = ((int64_t)inArg0 * (int64_t)inArg1);
    #ifndef FIXMATH_NO_ROUNDING
-   tempResult += (fix16_one >> 1);
+   tempResult += (tempResult >= 0) ? (fix16_one >> 1) : -(fix16_one >> 1);
    #endif
    tempResult >>= 16;
    return tempResult;
@@ -97,7 +97,7 @@
    int64_t tempResult = inArg0;
    tempResult <<= 16;
    #ifndef FIXMATH_NO_ROUNDING
-   tempResult += (inArg1 >> 1);
+   tempResult += (tempResult >= 0) ? (inArg1 >> 1) : -(inArg1 >> 1);
    #endif
    tempResult /= inArg1;
    return tempResult;

Looks like we could use some validation framework :)

Agreed, we have some tests but they clearly aren't even close to exhaustive. 
Would you be interested in joining the project? There are no expectations of 
work but it means you can make any reasonable changes you'd like to improve the 
codebase.

I'm currently working on a fix16_t based matrix algebra library, so yes, I 
would be interested.

This issue was closed by revision r52.