Merge pull request #6978 from lindahua/dh/openblas1

Only apply the gemv workaround when OpenBLAS is being used.

base/linalg/matmul.jl

@@ -203,25 +203,24 @@ function copytri!(A::StridedMatrix, uplo::Char, conjugate::Bool=false)
     A
 end
 
-gemv!{T<:BlasFloat}(y::StridedVector{T}, tA::Char, A::StridedMatrix{T}, x::StridedVector{T}) = generic_matvecmul!(y, tA, A, x)
-
-## Avoid calling BLAS.gemv! until #6941 is fixed.
-# function gemv!{T<:BlasFloat}(y::StridedVector{T}, tA::Char, A::StridedMatrix{T}, x::StridedVector{T})
-
-#     stride(A, 1)==1 || return generic_matvecmul!(y, tA, A, x)
-
-#     if tA != 'N'
-#         (nA, mA) = size(A)
-#     else
-#         (mA, nA) = size(A)
-#     end
-
-#     nA==length(x) || throw(DimensionMismatch(""))
-#     mA==length(y) || throw(DimensionMismatch(""))
-#     mA == 0 && return zeros(T, 0)
-#     nA == 0 && return zeros(T, mA)
-#     return BLAS.gemv!(tA, one(T), A, x, zero(T), y)
-# end
+if Base.blas_vendor() == :openblas
+    ## Avoid calling BLAS.gemv! when OpenBLAS is being used until #6941 is fixed.
+    gemv!{T<:BlasFloat}(y::StridedVector{T}, tA::Char, A::StridedMatrix{T}, x::StridedVector{T}) = generic_matvecmul!(y, tA, A, x)
+else
+    function gemv!{T<:BlasFloat}(y::StridedVector{T}, tA::Char, A::StridedMatrix{T}, x::StridedVector{T})
+        stride(A, 1)==1 || return generic_matvecmul!(y, tA, A, x)
+        if tA != 'N'
+            (nA, mA) = size(A)
+        else
+            (mA, nA) = size(A)
+        end
+        nA==length(x) || throw(DimensionMismatch(""))
+        mA==length(y) || throw(DimensionMismatch(""))
+        mA == 0 && return zeros(T, 0)
+        nA == 0 && return zeros(T, mA)
+        return BLAS.gemv!(tA, one(T), A, x, zero(T), y)
+    end
+end
 
 function syrk_wrapper!{T<:BlasFloat}(C::StridedMatrix{T}, tA::Char, A::StridedMatrix{T})
     nC = chksquare(C)