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
The first call to vector_sum(x) produces an incorrect result of tensor([79561.5703], device='cuda:6').
Subsequent calls to vector_sum(x) produce correct results.
The issue occurs with other reduction functions such as tl.max and atomic functions like tl.atomic_max.
However, using only one of them does not raise the issue.
The text was updated successfully, but these errors were encountered:
Environment
Issue Description
When using both reduction operations and atomic operations
triton.autotune
, the output is incorrect upon encountering a new input shape.Reproduction Code
Conclusion
The first call to
vector_sum(x)
produces an incorrect result oftensor([79561.5703], device='cuda:6')
.Subsequent calls to
vector_sum(x)
produce correct results.The issue occurs with other reduction functions such as
tl.max
and atomic functions liketl.atomic_max
.However, using only one of them does not raise the issue.
The text was updated successfully, but these errors were encountered: