Adds dispatch rules for column vector times transpose(row vector) #555
Conversation
src/matrix_multiply.jl
Outdated
| @@ -10,16 +10,18 @@ import LinearAlgebra: BlasFloat, matprod, mul! | |||
| @inline *(A::StaticVector, B::StaticMatrix) = *(reshape(A, Size(Size(A)[1], 1)), B) | |||
| @inline *(A::StaticVector, B::Transpose{<:Any, <:StaticVector}) = _mul(Size(A), Size(B), A, B) | |||
| @inline *(A::StaticVector, B::Adjoint{<:Any, <:StaticVector}) = _mul(Size(A), Size(B), A, B) | |||
| @inline *(A::StaticMatrix, B::Adjoint{<:Any, <:StaticVector}) = *(reshape(A, Size(Size(A)[1],)), B) | |||
There was a problem hiding this comment.
- Wouldn't
_mul(Size(A), Size(B), A, B)work? - Need the same thing for
Transpose.
There was a problem hiding this comment.
If I change that line to _mul(Size(A), Size(B), A, B) I get
ERROR: MethodError: no method matching _mul(::Size{(4, 1)}, ::Size{(1, 4)}, ::SArray{Tuple{4,1},Float64,2,4}, ::LinearAlgebra.Adjoint{Float64,SArray{Tuple{4},Float64,1,4}})
Closest candidates are:
_mul(::Size{sa}, ::Size{sb}, ::StaticArray{Tuple{#s175,#s174},Ta,2} where #s174 where #s175, ::StaticArray{Tuple{#s173},Tb,1} where #s173) where {sa, sb, Ta, Tb} at /home/zygmunt/.julia/dev/StaticArrays/src/matrix_multiply.jl:45
_mul(::Size{sa}, ::Size{sb}, ::StaticArray{Tuple{#s173},Ta,1} where #s173, ::Union{Adjoint{Tb,#s171} where #s171<:(StaticArray{Tuple{N},T,1} where T where N), Transpose{Tb,#s172} where #s172<:(StaticArray{Tuple{N},T,1} where T where N)}) where {sa, sb, Ta, Tb} at /home/zygmunt/.julia/dev/StaticArrays/src/matrix_multiply.jl:65
_mul(::Size{sa}, ::Size{sb}, ::StaticArray{Tuple{#s175,#s174},Ta,2} where #s174 where #s175, ::StaticArray{Tuple{#s173,#s172},Tb,2} where #s172 where #s173) where {sa, sb, Ta, Tb} at /home/zygmunt/.julia/dev/StaticArrays/src/matrix_multiply.jl:77
...
Stacktrace:
[1] *(::SArray{Tuple{4,1},Float64,2,4}, ::LinearAlgebra.Adjoint{Float64,SArray{Tuple{4},Float64,1,4}}) at /home/zygmunt/.julia/dev/StaticArrays/src/matrix_multiply.jl:13
[2] top-level scope at none:0
There was a problem hiding this comment.
what happens if size(A) is (4,2), does reshape fail then? is a DimensionMismatch thrown? I wonder if it wouldn't be cleaner to dispatch on the fact the dimensions 2 of A is 1 (similar to the example here #537 (comment) ). In that case a matrix A thats not a column vector would still go to the fallback method which would ultimately throw a DimensionMismatch.
There was a problem hiding this comment.
It does throw a DimensionMismatch but your solution is definitely neater. I have revised the pull-request as per your suggestion and have included a dispatch rule for transpose as well.
zygmuntszpak
force-pushed
the
dispatch_rules
branch
from
5d0309e to
9470cb9
Compare
|
I don't know if it is covered by other tests already, but it would be nice to test explicitly if a |
zygmuntszpak
force-pushed
the
dispatch_rules
branch
from
9470cb9 to
1ebd02b
Compare
Done. |
Codecov Report
@@ Coverage Diff @@
## master #555 +/- ##
==========================================
+ Coverage 66.17% 66.23% +0.05%
==========================================
Files 38 38
Lines 2903 2908 +5
==========================================
+ Hits 1921 1926 +5
Misses 982 982
Continue to review full report at Codecov.
|
Fixes issue #537