Is there an efficient way to do a signed/unsigned 16/32 bit add or sub in C with the C2000 compiler?
In assembly, this is relatively easy by setting the OVM bit and doing the 16 bit add in AH.
Unfortunately, there are no intrinsics for saturated add/sub available.
So when doing this from C, I would write something like
int a,b,c;
asm(" setc SXM, OVM");
c = (int)((((long)a << 16) + ((long)b << 16)) >> 16);
to do the add in AH and save the saturated 16 bit result of c = a+b.
This does not work. The C-Compiler generates this code with -o3:
setc SXM, OVM
MOV ACC,@_a
ADD ACC,@_b
LSL ACC,16
MOVH @_c << 0
The code does no saturation - the add is performed in the low 16 bit of ACC.
Without optimization, the code is okay:
setc SXM, OVM
MOV ACC,@_a << 16
MOVL XAR6,ACC
MOV ACC,@_b << 16
ADDL ACC,XAR6
SFR ACC,16
MOV @_c,AL
So is there a solution that works with all optimization levels and will work with all future compiler versions?
I would highly appreciate intrinsics for this... - but any other working and efficient solution is okay.
