Skip to content
This repository has been archived by the owner on Mar 12, 2021. It is now read-only.

Add mul! for trmm! and tests. #708

Merged
merged 2 commits into from
May 7, 2020
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
62 changes: 62 additions & 0 deletions src/blas/linalg.jl
Original file line number Diff line number Diff line change
Expand Up @@ -302,6 +302,68 @@ LinearAlgebra.rdiv!(A::CuMatrix{T}, B::Transpose{T,<:LowerTriangular{T, <:CuMatr
LinearAlgebra.rdiv!(A::CuMatrix{T}, B::Transpose{T,<:UnitLowerTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trsm!('R', 'L', 'T', 'U', one(T), parent(parent(B)), A)


# TRMM

# Left mul!
## No transpose/adjoint
LinearAlgebra.mul!(X::CuMatrix{T}, A::UpperTriangular{T, <:CuMatrix{T}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'U', 'N', 'N', one(T), parent(A), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::UnitUpperTriangular{T, <:CuMatrix{T}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'U', 'N', 'U', one(T), parent(A), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::LowerTriangular{T, <:CuMatrix{T}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'L', 'N', 'N', one(T), parent(A), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::UnitLowerTriangular{T, <:CuMatrix{T}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'L', 'N', 'U', one(T), parent(A), B, X)
## Adjoint
LinearAlgebra.mul!(X::CuMatrix{T}, A::Adjoint{T,<:UpperTriangular{T, <:CuMatrix{T}}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'U', 'C', 'N', one(T), parent(parent(A)), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::Adjoint{T,<:UnitUpperTriangular{T, <:CuMatrix{T}}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'U', 'C', 'U', one(T), parent(parent(A)), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::Adjoint{T,<:LowerTriangular{T, <:CuMatrix{T}}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'L', 'C', 'N', one(T), parent(parent(A)), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::Adjoint{T,<:UnitLowerTriangular{T, <:CuMatrix{T}}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'L', 'C', 'U', one(T), parent(parent(A)), B, X)
## Transpose
LinearAlgebra.mul!(X::CuMatrix{T}, A::Transpose{T,<:UpperTriangular{T, <:CuMatrix{T}}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'U', 'T', 'N', one(T), parent(parent(A)), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::Transpose{T,<:UnitUpperTriangular{T, <:CuMatrix{T}}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'U', 'T', 'U', one(T), parent(parent(A)), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::Transpose{T,<:LowerTriangular{T, <:CuMatrix{T}}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'L', 'T', 'N', one(T), parent(parent(A)), B, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::Transpose{T,<:UnitLowerTriangular{T, <:CuMatrix{T}}}, B::CuMatrix{T}) where T<:CublasFloat =
CUBLAS.trmm!('L', 'L', 'T', 'U', one(T), parent(parent(A)), B, X)

# Right mul!
## No transpose/adjoint
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::UpperTriangular{T, <:CuMatrix{T}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'U', 'N', 'N', one(T), parent(B), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::UnitUpperTriangular{T, <:CuMatrix{T}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'U', 'N', 'U', one(T), parent(B), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::LowerTriangular{T, <:CuMatrix{T}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'L', 'N', 'N', one(T), parent(B), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::UnitLowerTriangular{T, <:CuMatrix{T}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'L', 'N', 'U', one(T), parent(B), A, X)
## Adjoint
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::Adjoint{T,<:UpperTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'U', 'C', 'N', one(T), parent(parent(B)), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::Adjoint{T,<:UnitUpperTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'U', 'C', 'U', one(T), parent(parent(B)), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::Adjoint{T,<:LowerTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'L', 'C', 'N', one(T), parent(parent(B)), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::Adjoint{T,<:UnitLowerTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'L', 'C', 'U', one(T), parent(parent(B)), A, X)
## Transpose
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::Transpose{T,<:UpperTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'U', 'T', 'N', one(T), parent(parent(B)), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::Transpose{T,<:UnitUpperTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'U', 'T', 'U', one(T), parent(parent(B)), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::Transpose{T,<:LowerTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'L', 'T', 'N', one(T), parent(parent(B)), A, X)
LinearAlgebra.mul!(X::CuMatrix{T}, A::CuMatrix{T}, B::Transpose{T,<:UnitLowerTriangular{T, <:CuMatrix{T}}}) where T<:CublasFloat =
CUBLAS.trmm!('R', 'L', 'T', 'U', one(T), parent(parent(B)), A, X)


# Direct BLAS calls
for T in Base.uniontypes(CublasFloat) # needed to avoid ambiguous method error
@eval LinearAlgebra.BLAS.trmm!(side::AbstractChar, uplo::AbstractChar, transa::AbstractChar, diag::AbstractChar, alpha::$T, A::CuMatrix{$T}, B::CuMatrix{$T}) =
Expand Down
42 changes: 40 additions & 2 deletions test/blas.jl
Original file line number Diff line number Diff line change
Expand Up @@ -787,13 +787,51 @@ end # level 1 testset
B = copy(B)
dA = CuArray(A)
dB = CuArray(B)
dC = LinearAlgebra.BLAS.trmm!('L','U','N','N',one(elty),dA,dB)
C = LinearAlgebra.BLAS.trmm!('L','U','N','N',one(elty),A,B)
dC = LinearAlgebra.BLAS.trmm!('L','U','N','N',alpha,dA,dB)
C = LinearAlgebra.BLAS.trmm!('L','U','N','N',alpha,A,B)
@test A ≈ Array(dA)
@test B ≈ Array(dB)
@test C ≈ Array(dC)
end

@testset "BLAS.trsm!" begin
A = copy(A)
B = copy(B)
dA = CuArray(A)
dB = CuArray(B)
dC = LinearAlgebra.BLAS.trsm!('L','U','N','N',alpha,dA,dB)
C = LinearAlgebra.BLAS.trsm!('L','U','N','N',alpha,A,B)
@test A ≈ Array(dA)
@test B ≈ Array(dB)
@test C ≈ Array(dC)
end

@testset "mul! trmm!" begin
for t in (identity, transpose, adjoint), TR in (UpperTriangular, LowerTriangular, UnitUpperTriangular, UnitLowerTriangular)
A = copy(sA) |> TR
B_L = copy(B)
B_R = copy(B')
C_L = copy(C)
C_R = copy(C')
dA = CuArray(parent(A)) |> TR
dB_L = CuArray(parent(B_L))
dB_R = CuArray(parent(B_R))
dC_L = CuArray(C_L)
dC_R = CuArray(C_R)

D_L = mul!(C_L, t(A), B_L)
dD_L = mul!(dC_L, t(dA), dB_L)

D_R = mul!(C_R, B_R, t(A))
dD_R = mul!(dC_R, dB_R, t(dA))

@test C_L ≈ Array(dC_L)
@test D_L ≈ Array(dD_L)
@test C_R ≈ Array(dC_R)
@test D_R ≈ Array(dD_R)
end
end

B = rand(elty,m,n)
C = rand(elty,m,n)
d_B = CuArray(B)
Expand Down