segfault in ?gelss with nrhs=0

[eaberendeev]

For example single precision, but there is in all precision.

For nrhs = 0 there is copy (line 530) N elements to B array, but size of B can be less then N

I suggest add condition

IF NRHS > 0

for this condition branch
and for branch on line 638 and 713

[langou]

I agree. I will repeat to make sure that we are on the same page.

You are speaking about these lines of code

lapack/SRC/dgelss.f

Lines 698 to 717 in 4b54a33

	*
	* Multiply B by right singular vectors of A
	* (Workspace: need N, prefer N*NRHS)
	*
	IF( LWORK.GE.LDB*NRHS .AND. NRHS.GT.1 ) THEN
	CALL DGEMM( 'T', 'N', N, NRHS, M, ONE, A, LDA, B, LDB, ZERO,
	$ WORK, LDB )
	CALL DLACPY( 'F', N, NRHS, WORK, LDB, B, LDB )
	ELSE IF( NRHS.GT.1 ) THEN
	CHUNK = LWORK / N
	DO 60 I = 1, NRHS, CHUNK
	BL = MIN( NRHS-I+1, CHUNK )
	CALL DGEMM( 'T', 'N', N, BL, M, ONE, A, LDA, B( 1, I ),
	$ LDB, ZERO, WORK, N )
	CALL DLACPY( 'F', N, BL, WORK, N, B( 1, I ), LDB )
	60 CONTINUE
	ELSE
	CALL DGEMV( 'T', M, N, ONE, A, LDA, B, 1, ZERO, WORK, 1 )
	CALL DCOPY( N, WORK, 1, B, 1 )
	END IF

But there is the same pattern three times throughout the code. As you noted.

Your solution of " add a condition IF NRHS > 0 around the branch" would work.

Other solution: What about changing the ELSE statement in a ELSE IF( NRHS.EQ.1 ) THEN in the line below?

lapack/SRC/dgelss.f

Lines 714 to 717 in 4b54a33

	ELSE
	CALL DGEMV( 'T', M, N, ONE, A, LDA, B, 1, ZERO, WORK, 1 )
	CALL DCOPY( N, WORK, 1, B, 1 )
	END IF

I think that'll do the job too. What do you think? Do you agree?

Would you be willing to submit a PR (pull request)?

Julien

[eaberendeev]

Yes, i will do PR for this.
Your solution
ELSE IF( NRHS.EQ.1 ) THEN is ok

[langou]

I am not sure you are used to LAPACK, so, adding just in case, if you do a PR, (1) thanks a lot, (2) please consider doing all four arithmetics, so that means (1) cgelss.f, (2) dgelss.f, (3) sgelss.f, and (4) zgelss.f. (Not just dgelss.f.) I should have mentioned this in my previous post.