Skip to content

Commit

Permalink
work around 12.6.1 cudafe++ bug
Browse files Browse the repository at this point in the history
  • Loading branch information
trxcllnt authored and robertmaynard committed Oct 9, 2024
1 parent 5e18c85 commit 137f182
Showing 1 changed file with 1 addition and 19 deletions.
20 changes: 1 addition & 19 deletions cpp/include/raft/matrix/detail/columnWiseSort.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -76,8 +76,7 @@ template <
typename InType,
typename OutType,
int BLOCK_SIZE,
int ITEMS_PER_THREAD,
typename std::enable_if<TemplateChecker<InType, BLOCK_SIZE>::IsValid, InType>::type* = nullptr>
int ITEMS_PER_THREAD>
RAFT_KERNEL __launch_bounds__(1024, 1) devKeyValSortColumnPerRow(const InType* inputKeys,
InType* outputKeys,
OutType* inputVals,
Expand Down Expand Up @@ -120,23 +119,6 @@ RAFT_KERNEL __launch_bounds__(1024, 1) devKeyValSortColumnPerRow(const InType* i
}
}

template <
typename InType,
typename OutType,
int BLOCK_SIZE,
int ITEMS_PER_THREAD,
typename std::enable_if<!(TemplateChecker<InType, BLOCK_SIZE>::IsValid), InType>::type* = nullptr>
RAFT_KERNEL devKeyValSortColumnPerRow(const InType* inputKeys,
InType* outputKeys,
OutType* inputVals,
int n_rows,
int n_cols,
InType MAX_VALUE)
{
// place holder function
// so that compiler unrolls for all template types successfully
}

// helper function to layout values (index's) for key-value sort
template <typename OutType>
cudaError_t layoutIdx(OutType* in, int n_rows, int n_columns, cudaStream_t stream)
Expand Down

0 comments on commit 137f182

Please sign in to comment.