Fix doxygen-related CI failure

There was a problem with defgroup syntax.
rapidsai · Sep 6, 2022 · 36e5865 · 36e5865
1 parent 87319de
commit 36e5865
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 6 deletions.
diff --git a/cpp/bench/spatial/fused_l2_nn.cu b/cpp/bench/spatial/fused_l2_nn.cu
@@ -22,8 +22,12 @@
 #include <raft/linalg/norm.cuh>
 #include <raft/random/rng.cuh>
 
-// Note: do not include raft/spatial/knn/specializations.hpp based on
-// RAFT_NN_COMPILED, as fusedL2NN is not specialized and not defined there.
+// TODO: Once fusedL2NN is specialized in the raft_distance shared library, add
+// back
+//
+// #if defined RAFT_NN_COMPILED
+// #include <raft/spatial/knn/specializations.hpp>
+// #endif
 
 namespace raft::bench::spatial {
 

diff --git a/cpp/include/raft/distance/detail/fused_l2_nn.cuh b/cpp/include/raft/distance/detail/fused_l2_nn.cuh
@@ -202,6 +202,8 @@ __global__ __launch_bounds__(P::Nthreads, 2) void fusedL2NNkernel(OutT* min,
       for (int i = 0; i < P::AccRowsPerTh; ++i) {
 #pragma unroll
         for (int j = P::AccThCols / 2; j > 0; j >>= 1) {
+          // Actually, the srcLane (lid +j) should be (lid +j) % P:AccThCols,
+          // but the shfl op applies the modulo internally.
           auto tmpkey   = raft::shfl(val[i].key, lid + j, P::AccThCols);
           auto tmpvalue = raft::shfl(val[i].value, lid + j, P::AccThCols);
           KVPair tmp    = {tmpkey, tmpvalue};

diff --git a/cpp/include/raft/linalg/contractions.cuh b/cpp/include/raft/linalg/contractions.cuh
@@ -168,13 +168,10 @@ struct Policy4x4<double, _veclen> {
 /** @} */
 
 /**
- * @defgroup Policy4x4Skinny
- *
  * A smaller k-block (8 instead of 32) with fewer threads per block (8x8 instead
  * of 16x16), which is faster for raft::distance::fusedL2NN on skinny matrices,
  * i.e., matrices with a small k dimension.
  *
- * @{
  */
 template <typename DataT, int _veclen>
 struct Policy4x4Skinny {
@@ -191,7 +188,6 @@ struct Policy4x4Skinny<double, _veclen> {
   typedef KernelPolicy<double, _veclen, 8, 4, 4, 8, 8> Policy;
   typedef ColKernelPolicy<double, _veclen, 8, 4, 4, 8, 8> ColPolicy;
 };
-/** @} */
 
 /**
  * @defgroup Policy2x8 16 elements per thread Policy with k-block = 16

diff --git a/cpp/test/distance/fused_l2_nn.cu b/cpp/test/distance/fused_l2_nn.cu
@@ -23,6 +23,13 @@
 #include <raft/linalg/norm.cuh>
 #include <raft/random/rng.cuh>
 
+// TODO: Once fusedL2NN is specialized in the raft_distance shared library, add
+// the following:
+//
+// #if defined RAFT_NN_COMPILED
+// #include <raft/spatial/knn/specializations.hpp>
+// #endif
+
 namespace raft {
 namespace distance {