Hi,
Using Code composer 3.3 and 4.0 (compilers TMS470 4.6.4 and 4.6.5)
I have been trying to track down an issue with that appears to be a compiler issue. I was wondering if any one can comment.
I have the following code (the 75: and 76 are line numbers for line reference in the assembly):
unsigned long a, b;
75: a = 0x956b3db8;
76: b = 0x7fffe695;
if (((int) (a - b)) > 0)
{
status = 1;
}
else
{
status = 2;
}
My understanding is that (a - b) should happen first, the result should be typecast to an int, and then that result compared to 0.
With compiler optimizations (no optimization), o1, o2, and o3 I get the following output (similar for each optimization):
LDR V9, $C$CON2 ; |75|
STR V9, [SP, #48] ; |75|
LDR V9, $C$CON3 ; |76|
STR V9, [SP, #52] ; |76|
LDR V9, [SP, #52] ; |78|
LDR A1, [SP, #48] ; |78|
CMP V9, A1 ; |78|
BGE ||$C$L1|| ; |78|
If I use optimization level 0 (o0) I get this:
LDR V9, $C$CON2 ; |75|
STR V9, [SP, #32] ; |75|
LDR V9, $C$CON3 ; |76|
STR V9, [SP, #36] ; |76|
LDR V9, [SP, #36] ; |78|
LDR A1, [SP, #32] ; |78|
CMP V9, A1 ; |78|
BLT ||$C$L1|| ; |78|
The problem is this: the first case the first code works correctly most of the time. However, when a is a very large number (high bit set), the type cast is apparently happening first, so it becomes a very small number, and the equality doesn't work the way the 'C' code says it should.
Even with the difference for the o0 optimization, I will still get the wrong answer if a and b are equal. I would expect the branch instruction to be BLE.
Can someone tell me what I am missing, or is this an issue with the compiler?
Thanks,
Mike
