diff --git a/src/api/c/assign.cpp b/src/api/c/assign.cpp
index 4dda94086b..f863214759 100644
--- a/src/api/c/assign.cpp
+++ b/src/api/c/assign.cpp
@@ -127,7 +127,10 @@ af_err af_assign_seq(af_array *out,
             AF_CHECK(af_assign_seq(&tmp_out, tmp_in, ndims, index, rhs));
             AF_CHECK(af_moddims(out, tmp_out, lInfo.ndims(), lInfo.dims().get()));
             AF_CHECK(af_release_array(tmp_in));
-            AF_CHECK(af_release_array(tmp_out));
+            // This can run into a double free issue if tmp_in == tmp_out
+            // The condition ensures release only if both are different
+            // Issue found on Tegra X1
+            if(tmp_in != tmp_out) AF_CHECK(af_release_array(tmp_out));
             return AF_SUCCESS;
         }
 
@@ -244,7 +247,10 @@ af_err af_assign_gen(af_array *out,
             AF_CHECK(af_assign_gen(&tmp_out, tmp_in, ndims, indexs, rhs_));
             AF_CHECK(af_moddims(out, tmp_out, lInfo.ndims(), lInfo.dims().get()));
             AF_CHECK(af_release_array(tmp_in));
-            AF_CHECK(af_release_array(tmp_out));
+            // This can run into a double free issue if tmp_in == tmp_out
+            // The condition ensures release only if both are different
+            // Issue found on Tegra X1
+            if(tmp_in != tmp_out) AF_CHECK(af_release_array(tmp_out));
             return AF_SUCCESS;
         }
 
diff --git a/src/backend/cpu/scan_by_key.cpp b/src/backend/cpu/scan_by_key.cpp
index d00db6cceb..3f0aad80a9 100644
--- a/src/backend/cpu/scan_by_key.cpp
+++ b/src/backend/cpu/scan_by_key.cpp
@@ -31,6 +31,8 @@ namespace cpu
         kernel::scan_dim_by_key<op, Ti, Tk, To, 4> func4(inclusive_scan);
 
         in.eval();
+        key.eval();
+
         switch (in.ndims()) {
         case 1:
             getQueue().enqueue(func1, out, 0, key, 0, in, 0, dim);
diff --git a/src/backend/cuda/kernel/nearest_neighbour.hpp b/src/backend/cuda/kernel/nearest_neighbour.hpp
index b76c70ee2d..eb2bebed80 100644
--- a/src/backend/cuda/kernel/nearest_neighbour.hpp
+++ b/src/backend/cuda/kernel/nearest_neighbour.hpp
@@ -411,8 +411,8 @@ __global__ void select_matches(
                 s_dist[sid] = dist;
                 s_idx[sid]  = s_idx[sid + i];
             }
-            __syncthreads();
         }
+        __syncthreads();
     }
 
     // Store best matches and indexes to training dataset
diff --git a/src/backend/opencl/kernel/bilateral.hpp b/src/backend/opencl/kernel/bilateral.hpp
index 0493389645..2dc2a5932d 100644
--- a/src/backend/opencl/kernel/bilateral.hpp
+++ b/src/backend/opencl/kernel/bilateral.hpp
@@ -50,7 +50,6 @@ void bilateral(Param out, const Param in, float s_sigma, float c_sigma)
 
         std::call_once( compileFlags[device], [device] () {
                 bool use_native_exp = getActivePlatform() != AFCL_PLATFORM_POCL;
-                printf("NATIVE_EXP: %d\n", use_native_exp);
                 std::ostringstream options;
                 options << " -D inType=" << dtype_traits<inType>::getName()
                         << " -D outType=" << dtype_traits<outType>::getName();