Description of problem:
Backward substitution for triangular systems of linear equations as implemented in the function backsolve() in R returns erroneous results with openblas-0.3.19-1.fc35.x86_64.

Version-Release number of selected component (if applicable):
openblas-0.3.19-1.fc35.x86_64

How reproducible:
Always on my system

Steps to Reproduce:
Run the following R code:

n <- 8

R <- diag(n)
b <- rep(1, n)
x <- backsolve(R, b)

drop(R %*% x)
## returns:
## [1] 1 0 0 0 0 0 0 1
## but should return:
## [1] 1 1 1 1 1 1 1 1

Actual results:
A vector of ones and zeros in the example above, but backsolve() also breaks with other data.

Expected results:
A vector of only ones in the example above.

Additional info:
I think this is a pretty critical bug for scientific computing on Fedora.