need help dgetc2 and bi.Daxpy

[jorgeneo560]

hi, im trying to implement dtgsyl and dtgsyl2, i am runing this example

https://www.nag.com/numeric/nl/nagdoc_26.1/nagdoc_fl26.1/pdf/f08/f08yhf.pdf

the problem that i had with dgetc2 is that return k while i should get 0 and ipiv and jpiv are 0;1 while in fortran i get 1;2.

and with daxpy i dont get the same results in the matrix, i am runing almost the same values than in fortran and the resoult are very diferent, i am doing

bi.Daxpy(is, alpha, a[is:], lda, c[js:], ldc)

and i also try

bi.Daxpy(is, alpha, a[is:], 1, c[js:], 1)

and i get the same resoult

with dgetc2 i am doing

k = impl.Dgetc2(zdim, z, ldz, ipiv[:(zdim)], jpiv[:(zdim)])

[Patricio Whittingslow]

Follow up on this, I've dug into the DTGSY2 code doing side by side debugging and I'm finding discrepancies in results of Dgemm, which uses Daxpy under the hood.
This is the code:
Go:

bi.Dgemm(blas.NoTrans, blas.NoTrans, is, nb, mb, -1,

a[is:], lda, rhs, mb, 1, c[js:], ldc)

Fortran:

CALL DGEMM( 'N', 'N', IS-1, NB, MB, -ONE,

$ A( 1, IS ), LDA, RHS( 1 ), MB, ONE,

$ C( 1, JS ), LDC )

Here are the inputs which differ: (All leading dimensions are = 4)

FORTRAN:

A(1,1)@16

[16]

[1]: 4

[2]: 0

[3]: 0

[4]: 0

[5]: 1

[6]: 3

[7]: 1

[8]: 0

[9]: 1

[10]: 4

[11]: 3

[12]: 0

[13]: 2

[14]: 1

[15]: 1

[16]: 6

RHS(1)@8

[8]

[1]: 2

[2]: 1

[3]: -1

[4]: 3

[5]: 3

[6]: 1

[7]: -1

[8]: 2

C(1,1)@16

[16]

[1]: 1.0000000000000004

[2]: -1.0000000000000004

[3]: -0.99999999999999956

[4]: -0.99999999999999978

[5]: 9.0000000000000018

[6]: 2

[7]: 1

[8]: 1

[9]: 7.0000000000000018

[10]: -1

[11]: 3

[12]: -0.99999999999999978

[13]: 16

[14]: -0.99999999999999978

[15]: 7

[16]: 20

IS

2

JS

2

LDA

4

LDC

4

GO:

a

[]float64 len: 16, cap: 16, [4,1,1,2,0,3,4,1,0,1,3,1,0,0,0,6]

rhs

[]float64 len: 8, cap: 8, [2,1,-1,3,3,1,-1,2]

c

[]float64 len: 16, cap: 16, [1.0000000000000004,9.000000000000002,7.000000000000002,16,-1.0000000000000004,2,-1,-0.9999999999999998,-0.9999999999999996,1,3,7,-0.9999999999999998,1,-0.9999999999999998,20]

is

1

nb

2

mb

2

lda

4

js

1

The result is as follows: Fortran:

C(1,1)@16 [16] [1]: 1.0000000000000004 [2]: -1.0000000000000004 [3]: -0.99999999999999956 [4]: -0.99999999999999978 [5]: 6.0000000000000018 [6]: 2 [7]: 1 [8]: 1 [9]: 5.0000000000000018 [10]: -1 [11]: 3 [12]: -0.99999999999999978 [13]: 16 [14]: -0.99999999999999978 [15]: 7 [16]: 20

GO: 

c []float64 len: 16, cap: 16, [1.0000000000000004,8.000000000000002,3.0000000000000018,16,-1.0000000000000004,2,-1,-0.9999999999999998,-0.9999999999999996,1,3,7,-0.9999999999999998,1,-0.9999999999999998,20]

As you can see, most values match, except for the second value in Go (fifth in Fortran) and the third value in Go.

Here is the Go output array (C)  in column major representation for the discrepancy to be more visual:

call reorder(c,4,4)

[0]: 1.0000000000000004

[1]: -1.0000000000000004

[2]: -0.9999999999999996

[3]: -0.9999999999999998

[4]: 8.000000000000002

[5]: 2

[6]: 1

[7]: 1

[8]: 3.0000000000000018

[9]: -1

[10]: 3

[11]: -0.9999999999999998

[12]: 16

[13]: -0.9999999999999998

[14]: 7

[15]: 20

[Patricio Whittingslow]

I said "inputs which differ", I meant to clarify the inputs are exactly the same to my knowledge and should have the same output.