From 016a6f46a03e1e8948d1b355e09869d03e8bef6b Mon Sep 17 00:00:00 2001
From: Evan Harvey <eharvey@sandia.gov>
Date: Thu, 13 Jan 2022 11:32:40 -0700
Subject: [PATCH] src/batched: Use SerialOpt2 for 33 to 39 square matrices

---
 src/batched/dense/KokkosBatched_Gemm_Decl.hpp | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/src/batched/dense/KokkosBatched_Gemm_Decl.hpp b/src/batched/dense/KokkosBatched_Gemm_Decl.hpp
index 9b1eb18e1f..9e147c1253 100644
--- a/src/batched/dense/KokkosBatched_Gemm_Decl.hpp
+++ b/src/batched/dense/KokkosBatched_Gemm_Decl.hpp
@@ -468,8 +468,7 @@ int BatchedGemm(BatchedGemmHandleType *const handle, const ScalarType alpha,
       // } else
       if (on_gpu && ((std::is_same<layout_type, Kokkos::LayoutLeft>::value)
                          ? (c_m >= 16)
-                         : (c_m >= 24))) {  // Vinh's note: use this condition
-                                            // for now, might need to revisit
+                         : (c_m >= 24 && c_m <= 32) || c_m >= 40)) {
         handle->teamSz = handle->vecLen = 8;
         constexpr int tile_m = 32, tile_n = 32, tile_k = 8;
 #ifdef __CUDACC_RDC__