[SOLVED] Need help to use slegsy in NumLib correctly

[ikel]

Hi,

I'm running the following example with slegsy, but throw an access violation error.
The original example is located here; https://wiki.freepascal.org/NumLib_Documentation#Solving_systems_of_linear_equations_(unit_sle)
Here is the code.

Code: Pascal  [Select][+][-]

1. program solve_sys_lin_eq;
2.  
3. uses
4.   typ, sle, omv;
5.  
6. const
7.   n = 4;
8.   A: array[1..n, 1..n] of ArbFloat = (
9.     ( 5,  7,  6,  5),
10.     ( 7, 10,  8,  7),
11.     ( 6,  8, 10,  9),
12.     ( 5,  7,  9, 10)
13.   );
14.   b: array[1..n] of ArbFloat = (
15.     57, 79, 88, 86
16.   );
17.  
18. var
19.   x: array[1..n] of ArbFloat;
20.   b_test: array[1..n] of ArbFloat;
21.   ca: ArbFloat;
22.   i, j: Integer;
23.   term: Integer;
24.  
25. begin
26.   WriteLn('Solve matrix system A x = b');
27.   WriteLn;
28.  
29.   // Print A
30.   WriteLn('Matrix A = ');
31.   for i:= 1 to n do begin
32.     for j := 1 to n do
33.       Write(A[i, j]:10:0);
34.     WriteLn;
35.   end;
36.   WriteLn;
37.  
38.   // Print b
39.   WriteLn('Vector b = ');
40.   for i:= 1 to n do
41.     Write(b[i]:10:0);
42.   WriteLn;
43.   WriteLn;
44.  
45.   // Solve - select one of these methods
46.   //slegen(n, n, A[1, 1], b[1], x[1], ca, term);
47.   slegsy(n, n, A[1, 1], b[1], x[1], ca, term);
48.   //slegpd(n, n, A[1, 1], b[1], x[1], ca, term);
49.  
50.   if term = 1 then begin
51.     WriteLn('Solution vector x = ');
52.     for i:= 1 to n do
53.       Write(x[i]:10:0);
54.     WriteLn;
55.     WriteLn;
56.  
57.     omvmmv(A[1,1], n, n, n, x[1], b_test[1]);
58.  
59.     WriteLn('Check result: A x = (must be equal to b)');
60.     for i:= 1 to n do
61.       Write(b_test[i]:10:0);
62.     WriteLn;
63.   end
64.   else
65.     WriteLn('Error');
66.  
67.   ReadLn;
68.  
69. end.
70.  

The error I get is ...

Code: Bash  [Select][+][-]

1. $ ./SolveLinearEq.exe
2. Solve matrix system A x = b
3.  
4. Matrix A =
5.          5         7         6         5
6.          7        10         8         7
7.          6         8        10         9
8.          5         7         9        10
9.  
10. Vector b =
11.         57        79        88        86
12.  
13. An unhandled exception occurred at $0000000100015539:
14. EAccessViolation: Access violation
15.   $0000000100015539
16.   $000000010000F35A
17.   $00000001000018D3
18.   $0000000100001AD6
19.   $000000010000DA20
20.   $00000001000016C0
21.   $00007FFC4F09257D
22.   $00007FFC5052AF28
23.  

Can someone help me to use slegsy correctly?

Thanks.

[wp]

I tested your code with Laz 3.4/FPC 3.2.2 (32 bit and 64 bit), Laz main/FPC 3.2.2 (32 bit and 64 bit) and with Laz main/FPC main, all on Windows 11, and with Laz 3.4/FPC 3.2.2 on Kubuntu 24.04, and did not get an error. Which Laz/FPC versions are you using, and on with operating system?

[ikel]

Hi wp,

I'm sorry, I mised out the important bit out. I am uing the following.

Lazarus 3.5 (rev lazarus_3_4-96-gcfd24b5c32)
FPC 3.2.3 x86_64-win64-win32/win64
Win 11.

Let me try the snippet on FPC 3.2.2. Thank you.

-ikel

[ikel]

I tested your code with Laz 3.4/FPC 3.2.2 (32 bit and 64 bit), Laz main/FPC 3.2.2 (32 bit and 64 bit) and with Laz main/FPC main, all on Windows 11, and with Laz 3.4/FPC 3.2.2 on Kubuntu 24.04, and did not get an error. Which Laz/FPC versions are you using, and on with operating system?

wp, the snippet works when I use FPC 3.2.2 x86_64-win64-win32/win64. The snippet doesn't work on FPC 3.2.3 (fixes).

Thanks for your help!

-ikel

[jamie]

So, I have an enquiry, all those other units you have in the USES list, were they also compiled with the 3.2.3 release?

[ikel]

So, I have an enquiry, all those other units you have in the USES list, were they also compiled with the 3.2.3 release?

Good question. I didn't think of that.
I believe not. This is the log from fpcupdeluxe.

Code: Pascal  [Select][+][-]

1. info: FPC Native Installer (BuildModule: FPC): To compile this FPC, we will use the (already available) compiler with version : 3.2.2.
2.  

[Thaddy]

I can reproduce the exact error on windows 11-64 and a recent 3.3.1 trunk. The error does not appear in a recent Ubuntu64 + a recent trunk.
Both compiled like
fpc -CX -XXs -O4 solve_sys_lin_eq.pas
So what has changed since you marked it as solved?
Same trunk version

Error is around line 47.
If I compile in fpc - which is the default - mode instead of objfpc, I get a compile time error instead of a runtime error:
solve_sys_lin_eq.pas(47,45) Error: (3069) Call by var for arg no. 7 has to match exactly: Got "SmallInt" expected "LongInt", but that is likely due to integer size in fpc mode.
I am more worried about which Arbfloat is chosen for win64, but looks ok: size 8 so double.

[wp]

Strange. I can confirm Thaddy's observation that the crash happens with FPC 3.3.1/64 bit on Windows (it's at line 254 in procedure mdtgtr() of unit mdt) . But when I copy the units det, mdt, and sle from the numlib folder into the project folder (for easier debugging) the crash is gone (attached this project)... Then I rebuilt FPC, but it does not fix the issue in the original project.

[marcov]

win64 uses SSE for floating point, it might be a problem there. Or some FPU masking not done for win64.