Add a BLAS wrapper for axpby #23291
Add a BLAS wrapper for axpby #23291
Conversation
base/linalg/blas.jl
Outdated
| @@ -463,6 +463,55 @@ function axpy!(alpha::Number, x::Array{T}, rx::Union{UnitRange{Ti},Range{Ti}}, | |||
| y | |||
| end | |||
|
|
|||
| """ | |||
| axpby!(a, X, b, Y) | |||
There was a problem hiding this comment.
4 space indent for the signature.
base/linalg/blas.jl
Outdated
| """ | ||
| axpby!(a, X, b, Y) | ||
|
|
||
| Overwrite `Y` with `a*X + b*Y`, where `a` and `b` are scalars. Returns `Y`. |
base/linalg/blas.jl
Outdated
|
|
||
| Overwrite `Y` with `a*X + b*Y`, where `a` and `b` are scalars. Returns `Y`. | ||
|
|
||
| # Example: |
base/linalg/blas.jl
Outdated
|
|
||
| # Example: | ||
| ```jldoctest | ||
| julia> x = [1.; 2; 3]; |
There was a problem hiding this comment.
Perhaps change to commas in the vector, i.e. [1., 2, 3] and likewise for y?
base/linalg/blas.jl
Outdated
| # INTEGER INCX,INCY,N | ||
| #* .. Array Arguments .. | ||
| # DOUBLE PRECISION DX(*),DY(*) | ||
| function axpby!(n::Integer, alpha::($elty), dx::Union{Ptr{$elty}, DenseArray{$elty}}, incx::Integer, beta::($elty), dy::Union{Ptr{$elty}, DenseArray{$elty}}, incy::Integer) |
There was a problem hiding this comment.
getting a bit beyond the line length recommendation
would also be more conventional to indent the whole @eval block since it's inside a for loop (or do @eval function since the begin is somewhat superfluous)
There was a problem hiding this comment.
Thanks :) don't know why the indenting got messed up
|
I discussed #13640 with @andreasnoack and I think it makes sense indeed to add a generic Also the ccall should probably not use |
|
Bump |
| # INTEGER INCX,INCY,N | ||
| #* .. Array Arguments .. | ||
| # DOUBLE PRECISION DX(*),DY(*) | ||
| function axpby!(n::Integer, alpha::($elty), dx::Union{Ptr{$elty}, |
There was a problem hiding this comment.
does this exist in every blas implementation? if it isn't in netlib I don't think we can unconditionally have a wrapper for it here
The
axpby!routine computesy = a * x + b * yand is an extension of BLAS. The current BLAS-style alternative is to doscale!(b, y); axpy!(a, x, y), which does the same amount of computational work, but requires going through memory twice.I have not added generic fallbacks for this method because I'm unsure about #13640 which made
axpy!basicallyxapy!. It also does not come withaxpby!(a, x, indices_x, b, y, indices_y)because I think that's superfluous when when there are views.