Wrong answers from dot with a SubArray that reverts the order of elements on Julia v1.6 and before

[junyuan-chen]

I am surprised by some wrong answers from dot that are illustrated below:

julia> using LinearAlgebra

julia> v1 = [1.0, 2.0, 3.0]
3-element Vector{Float64}:
 1.0
 2.0
 3.0

julia> dot([1.0, 2.0], view(v1, 2:-1:1))
7.0

The 7.0 returned above seems like coming from a bug that computes what I would expect for dot([1.0, 2.0], view(v1, 3:-1:2)).

Such behavior is not present if the parent vector contains integers:

julia> v2 = [1, 2, 3]
3-element Vector{Int64}:
 1
 2
 3

julia> dot([1.0, 2.0], view(v2, 2:-1:1))
4.0

I noticed the above unexpected results when I was trying to figure out why the CI for a package fails on Julia v1.6 for all platforms. It seems that the above behavior happens on Julia v1.6.7 and older versions but not v1.7 or later.

I am posting my local version info here but I don't think this is platform-specific:

julia> versioninfo()
Julia Version 1.6.7
Commit 3b76b25b64 (2022-07-19 15:11 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin21.4.0)
  CPU: Apple M1 Pro
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-11.0.1 (ORCJIT, westmere)
Environment:
  JULIA_NUM_THREADS = 8

[N5N3]

The neg-stride support for BLAS.dot was added in 1.7 (#39751), and #42957 extends it to more BLAS funs in 1.8.
Since #41513 was backported to 1.6. So perhaps we should also backport them.