[VA] SRC #015b - HP-15C & clones: COMPLEX Matrix Inverse up to 8x8

[Werner]

Thank you, Gerson!
In the meantime I updated my post with a new, shorter and faster (for solving, at least) version.
It seems that every time I look at my code I find ways to improve it..

I also did some accuracy tests, with the example from pg 128 in the Advanced Functions Handbook, which you can now do on your original 15C:
(I multiplied all values by 3 - then they are all integers):

A 4x4: 
  300    0    0    0 
    0  3E6 -3E6    0
    0 -3E6  3E6    0
    0    0    0  3E5

B 4x4:
 -350  800    0    0
  800 -350    0    0
    0    0  442 -450
    0    0 -450  442

C 4x1:
 30
  0
  0
  0

D 4x1:
  0
  0 
  0
  0

exact result, rounded to 10 digits:
 ( 1.995795141E-4 ,  4.096399085E-3)   
 (-1.448833619E-3 , -3.563298308E-2)   
 (-1.454083174E-3 , -3.563276083E-2)   
 ( 5.344581171E-5 , -2.259868256E-6)

built-in result (on the 15CE, 8x8 matrix):
 ( 1.995820000E-4 ,  4.096401051E-3 )
 (-1.448812372E-3 , -3.563300015E-2 )
 (-1.454061929E-3 , -3.563277790E-2 )
 ( 5.344583732E-5 , -2.259863892E-6 )

split solve result:
 ( 1.995574520E-4 ,  4.096386294E-3 )
 (-1.448834528E-3 , -3.563298920E-2 )
 (-1.454084082E-3 , -3.563276695E-2 )
 ( 5.344583458E-5 , -2.259876193E-6 )


These results are comparable. The matrix is badly conditioned, so the results are only accurate to about 5 digits.
The split solve routine is slightly slower than the built-in one (without counting the necessary transformations to produce the 8x8 matrix, though!): 0.45 seconds vs. 0.36 seconds on the 15CE, but this latest version should be about 10% faster than the previous one.

Cheers, Werner