You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Applying dot to views of two three-dimensional CuArrays gives a substantially different result than dot over the equivalent slice. Behaviour is correct with CPU arrays.
To reproduce
using LinearAlgebra
using CUDA
A =rand(100, 100, 100)
B =rand(100, 100, 100)
# Consistent down to machine epsilon on CPU@showdot(A[2:99, 2:99, 2:99], B[2:99, 2:99, 2:99])
@showdot(view(A, 2:99, 2:99, 2:99), view(B, 2:99, 2:99, 2:99))
cA =CuArray(A)
cB =CuArray(B)
# Different results on GPU@showdot(cA[2:99, 2:99, 2:99], cB[2:99, 2:99, 2:99])
@showdot(view(cA, 2:99, 2:99, 2:99), view(cB, 2:99, 2:99, 2:99))
Describe the bug
Applying dot to views of two three-dimensional CuArrays gives a substantially different result than dot over the equivalent slice. Behaviour is correct with CPU arrays.
To reproduce
Manifest.toml
Expected behavior
The same answer from views and slicing.
Version info
Details on Julia:
Details on CUDA:
The text was updated successfully, but these errors were encountered: