diff --git a/src/CodeGen_Xtensa_vectors.template.cpp b/src/CodeGen_Xtensa_vectors.template.cpp
index b3e9fb40f8dd..f3d2f8819c67 100644
--- a/src/CodeGen_Xtensa_vectors.template.cpp
+++ b/src/CodeGen_Xtensa_vectors.template.cpp
@@ -2355,14 +2355,6 @@ HALIDE_ALWAYS_INLINE native_vector_i16 convert<native_vector_i16, native_vector_
                         IVP_SELI_16B_EXTRACT_1_OF_2_OFF_0);
 }
 
-template<>
-HALIDE_ALWAYS_INLINE native_vector_i16_x2 convert<native_vector_i16_x2, native_vector_i32_x4>(const native_vector_i32_x4 &src) {
-    xb_vecNx48 wide0 = IVP_CVT48SNX32(src.native_vector[1], src.native_vector[0]);
-    xb_vecNx48 wide1 = IVP_CVT48SNX32(src.native_vector[3], src.native_vector[2]);
-
-    return native_vector_i16_x2(native_vector_i16_x2::from_native_vector, IVP_PACKLNX48(wide0), IVP_PACKLNX48(wide1));
-}
-
 template<>
 HALIDE_ALWAYS_INLINE native_vector_u16 convert<native_vector_u16, native_vector_i32_x2>(const native_vector_i32_x2 &src) {
     return IVP_SELNX16UI(IVP_MOVNX16_FROMN_2X32(src.native_vector[1]),
diff --git a/test/correctness/simd_op_check_xtensa.cpp b/test/correctness/simd_op_check_xtensa.cpp
index a4cbace99278..f573a3febdee 100644
--- a/test/correctness/simd_op_check_xtensa.cpp
+++ b/test/correctness/simd_op_check_xtensa.cpp
@@ -111,6 +111,8 @@ class SimdOpCheckXtensa : public SimdOpCheckTest {
         // some of these could overflow that limit. (Omitting the spaces is
         // a bit of a band-aid here; a better solution would probably be
         // to allow arbitrary names that don't match, but for now, this will do.)
+        // TODO(vksnk): float16 doesnt't seem to be supported well by cstubs library.
+        // https://github.com/halide/Halide/issues/7858
         // check("convert<float16x32_t,float32x32_t>", vector_width / 2, f16(f32_1));
         // check("convert<float32x32_t,float16x32_t>", vector_width / 2, f32(f16_1));
         check("convert<float32x32_t,int16x32_t>", vector_width / 2, f32(i16_1));