Skip to content

Zero the output of out-of-place trmm/trsm when beta is zero #761

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
maleadt merged 1 commit into from
Nov 6, 2026
Merged

Zero the output of out-of-place trmm/trsm when beta is zero #761

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
6 changes: 6 additions & 0 deletions lib/mkl/wrappers_blas.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1214,6 +1214,9 @@ for (mmname_variant, smname_variant, elty) in
lda = max(1,stride(A,2))
ldb = max(1,stride(B,2))
ldc = max(1,stride(C,2))
# the variant reads C regardless of beta (unlike gemm), and trsm leaks NaNs
# from an uninitialized C into the result
iszero(beta) && fill!(C, zero($elty))
queue = global_queue(context(A), device(A))
$mmname_variant(sycl_queue(queue), side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb, beta, C, ldc)
C
Expand All @@ -1236,6 +1239,9 @@ for (mmname_variant, smname_variant, elty) in
lda = max(1,stride(A,2))
ldb = max(1,stride(B,2))
ldc = max(1,stride(C,2))
# the variant reads C regardless of beta (unlike gemm), and trsm leaks NaNs
# from an uninitialized C into the result
iszero(beta) && fill!(C, zero($elty))
queue = global_queue(context(A), device(A))
$smname_variant(sycl_queue(queue), side, uplo, transa, diag, m, n, alpha, A, lda, B, ldb, beta, C, ldc)
C
Expand Down
20 changes: 20 additions & 0 deletions test/onemkl.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -782,6 +782,11 @@ end
D = alpha*A*B + beta*C
@test D ≈ h_C
@test B == Array(dB)

# with beta = 0, the contents of C must not matter
dC = oneArray(fill(T(NaN), m, n))
oneMKL.trmm!('L','U','N','N',alpha,dA,dB,dC)
@test alpha*A*B ≈ Array(dC)
end

@testset "right trmm!" begin
Expand Down Expand Up @@ -833,6 +838,11 @@ end
D = alpha*(A\B) + beta*C
@test D ≈ h_C
@test B == Array(dB)

# with beta = 0, the contents of C must not matter
dC = oneArray(fill(T(NaN), m, n))
oneMKL.trsm!('L','U','N','N',alpha,dA,dB,dC)
@test alpha*(A\B) ≈ Array(dC)
end

@testset "left trsm" begin
Expand Down Expand Up @@ -1118,6 +1128,16 @@ end
@test all(iszero, Array(C))
end

@testset "triangular products and solves into uninitialized outputs" begin
S = rand(Float32, 4, 4) + 4I; B = rand(Float32, 4, 5)
dS, dB = oneArray(S), oneArray(B)
for (f, g) in ((mul!, *), (ldiv!, \))
C = oneArray(fill(NaN32, 4, 5))
f(C, UpperTriangular(dS), dB)
@test Array(C) ≈ g(UpperTriangular(S), B)
end
end

@testset "triangular products and solves with reversed views" begin
S = rand(Float32, 4, 4) + 4I; B = rand(Float32, 4, 5); x = rand(Float32, 4)
dS = oneArray(S)
Expand Down
Loading