I hate to necro-post on a two year old thread,but I am noticing problems with the ACOS and F28335 also. I am finding that on occasion the processor is returning 0 for the arc cosine (-1). The result is repeatable, but the same code does not always produce the error.
In this particular case, I compute the phase angle between two vectors and take the acos() of the value to get the angle in radians. In the code below, result will sometimes return 0 when the input vectors are 180 degrees apart. I have broken the vAngle function into separate statements and verified that the input into the acos is indeed -1. If I test for -1 +/- a small epsilon and make it -0.9999999 I get the correct result. The problem appears to be caused by an input of exactly -1.0 (which is representable as a even power of 2 in floating point, while 0.9999 is not).
As I indicated, it does not always happen. One test case that does cause it to happen is to use an angle set of 6 and 186 degrees for the two vector inputs. The vectors are simply a double[] with the cos and sin values of the angle in radians (i,e. the x,y components).
I have upgraded to code generation tools version 5.2.6 (April 2010) which apparently corrects some problems in math.h. I did include the run time library in the project and I did include the proper header files (math.h). I tried this in C and C++ and received the same result. I have run this on a 6711 simulator and the code works perfectly.
The error can be duplicated easily with the following code: using a step size of 6. Break when tcount = 92, then step and the next call to vAngle will produce an erroneous result (the inputs will be 6 and 186 degrees and the cos of the angle between them is -1).
float vDot(const float a[ ], const float b[ ] , const size_t n){
float sum = 0;
size_t i;
for(i=0; i<n; ++i)
sum += a[i] * b[i];
return sum;
}
float vNorm(const float a[ ], const size_t n){
return sqrt(vDot(a, a, n));
}
float vAngle(const float a[ ], const float b[ ], const size_t n) {
// cos theta = a dot b / |a| x |b|
return acos (vDot(a,b,n) / (vNorm(a, n) * vNorm(b ,n) ));
}
//main code
void vector_test(void)
{
const float pi=3.1415926535897932384626433832795;
//const float halfpi = pi/2;
const float twopi = 2*pi;
int num_steps = 60; //Do each 6 degrees - fails right after tcount=92
const float step = twopi / num_steps;
int tcount = 0; //count the test number to reproduce failures
float aVec[2] = {0,0};
float bVec[2] = {0,0};
float loop1, loop2;
float result;
float eresult;
float error;
float error_max = 0;
int flag = 0;
for (loop1 = 0; loop1 < twopi; loop1 += step)
{
aVec[0] = cos(loop1); //r=1 implied
aVec[1] = sin(loop1);
for (loop2 = 0; loop2 < twopi; loop2 += step)
{
bVec[0] = cos(loop2);
bVec[1] = sin(loop2);
//calculate the angle between them
eresult = fabs((loop1 - loop2));
//Make sure we have the smaller of the two angles
if(eresult > pi)
eresult = twopi - eresult; //take the smaller path
result = vAngle(aVec, bVec, 2);
error = fabs(result-eresult);
if( error > error_max)
error_max = error;
if (error > .0005)
{
while(1)
{
flag = 1;
}
}
tcount++;
}
}
}
