APPCRASH (not when debugging) and Segmentation fault using QtCreator (C/C++) - c++

Question

I'm using QtCreator to code an algorithm that I have already coded on Matlab.
When coding this program, I have two errors. The firts one (APPCRASH) appears just when I build and execute the program normally, but not when I try to debug it (Heisenbug) and it appears on the function 'matriceA'. I tried to make the variables volatile and to write the matrix A term formulas on other function, hoping that that will stop the compiler optimization (I think that the compiler optimization might cause the problem), but I have not been able to solve the problem. I have not tried to to compile the project using the option -o0 because my professor (it's an university project) has to be able to compile it normally (without specific options).
The second one is a SISSEGV segmentation fault. It happens when the code arrives to "DestroyFloatArray(&b, width);" on InpaintingColor.
And here the codes:
clanu_process.cpp (it's little messy because I've tried a lot of things...)
#include "clanu_process.h"
#include "iomanip"
void InpaintingColor(float **Rout, float **Gout, float **Bout, float **Rin, float **Gin, float **Bin, float **Mask, int width, int height, double param)
{
cout << "1" << endl;
float alphak = 0, bethak = 0, res = 0;
float **b = 0, **xk = 0, **dk = 0, **rk = 0, **Ark = 0, **tmp1 = 0,**tmp2 = 0,**tmp3 = 0;
Ark = AllocateFloatArray( width, height);
tmp1 = AllocateFloatArray( width, height);
tmp2 = AllocateFloatArray( width, height);
tmp3 = AllocateFloatArray( width, height);
xk = AllocateFloatArray( width, height);
dk = AllocateFloatArray( width, height);
rk = AllocateFloatArray( width, height);
b = AllocateFloatArray( width, height);
cout << "2" << endl;
res = 1e8;
matrixProductByScalar(b,1.0/(3.0*256),Rin,width,height);
matrixDuplicate(xk, b, width, height);
// APPCRASH error
matriceA(Ark,xk,Mask,width,height);
//More code
// SIGSEGV error
DestroyFloatArray(&b, width);
DestroyFloatArray(&xk, width);
DestroyFloatArray(&dk, width);
DestroyFloatArray(&rk, width);
DestroyFloatArray(&Ark, width);
DestroyFloatArray(&tmp1, width);
DestroyFloatArray(&tmp2, width);
DestroyFloatArray(&tmp3, width);
}
float** matriceA(float **A, float **I, float **Masque, int N2, int N1){
volatile bool bool_iplus = false, bool_imoins = false, bool_jmoins = false, bool_jplus = false;
volatile int iplus = 0, imoins = 0, jplus = 0, jmoins = 0;
for(int i = 1; i <= N1; i++){
bool_iplus = i<N1;
iplus = i+1 < N1 ? i+1 : N1;
bool_imoins = i>1;
imoins = i-1 > 1 ? i-1 : 1;
for(int j = 1; j <= N2; j++){
bool_jplus = j<N2;
jplus = j+1 < N2 ? j+1 : N2;
bool_jmoins = j>1;
jmoins = j -1 > 1 ? j-1 : 1;
if(Masque[i-1][j-1]!=0){
//cout << "if - " << i << ", " << j<< endl;
A[i-1][j-1] = (1.0/36)*(16*I[i-1][j-1]
+ 4*(
(bool_iplus?I[iplus-1][j-1]:0)
+ (bool_imoins?I[imoins-1][j-1]:0)
+ (bool_jplus?I[i-1][jplus-1]:0)
+ (bool_jmoins?I[i-1][jmoins-1]:0)
)+(
(bool_iplus&&bool_jplus?I[iplus-1][jplus-1]:0)
+ (bool_imoins&&bool_jplus?I[imoins-1][jplus-1]:0)
+ (bool_imoins&&bool_jmoins?I[imoins-1][jmoins-1]:0))
+ (bool_iplus&&bool_jmoins?I[iplus-1][jmoins-1]:0));
}else{
//cout << "else - " << i << ", " << j << endl;
A[i-1][j-1]=
-(1.0*N1*N2)*(
-8.0*I[i-1][j-1]
+ I[iplus-1][j-1]
+ I[imoins-1][j-1]
+ I[i-1][jplus-1]
+ I[i-1][jmoins-1]
+ I[iplus-1][jplus-1]
+ I[imoins-1][jplus-1]
+ I[imoins-1][jmoins-1]
+ I[iplus-1][jmoins-1]);
}
}
}
return A;
}
The functions AllocateFloatArray and DestroyFloatArray
float ** AllocateFloatArray(int width, int height)
{
float ** r = new float*[width];
for(int i=0; i<width; i++)
r[i] = new float[height];
return r;
}
void DestroyFloatArray(float ***a, int width)
{
if( *a == 0 ) return;
for(int i=0; i<width; i++)
delete[] a[0][i];
delete[] *a;
*a = 0;
}
Thank you for your time.

Answer 1

I'm no sure that it's the cause of your problem but...
Your function "Matrix operations" (sum(), matrixSubstraction(), matrixAddition(), matrixProductByElement(), matrixProductByScalar(), and matrixDuplicate()) are ranging the first index from zero to width and the second one from zero to height.
If I'm not wrong, this is correct and is consistent with allocation/deallocation (AllocateFloatArray() and DestroyFloatArray()).
But look at the two matriceA() functions; they are defined as
float** matriceA(float **A, float **I, int N2, int N1)
float** matriceA(float **A, float **I, float **Masque, int N2, int N1)
In both functions the first index range from zero to N1 and the second one from zero to N2; by example
for(int i = 1; i <= N1; i++){
// ...
for(int j = 1; j <= N2; j++){
// ...
A[i-1][j-1] = (1.0/36)*(16*I[i-1][j-1] // ...
Good. But you call matriceA() in this way
matriceA(Ark,rk,Mask,width,height);
Briefly: you allocate your matrices as width * height matrices; your "matrix operations" are using they as width * height matrices but your matriceA() function are using they as height * width.
Wonderful way to devastate the memory.
I suppose the solution could be
1) switch N1 and N2 in matriceA() definition
2) or switch width and height in matriceA() calling
p.s.: sorry for my bad English.