revise typo

kendryte · Sep 30, 2024 · a392f69 · a392f69
1 parent 6556d01
commit a392f69
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/src/Native/include/nncase/ntt/ukernels/u_matmul.h b/src/Native/include/nncase/ntt/ukernels/u_matmul.h
@@ -154,7 +154,7 @@ struct u_matmul<ukernels::mamtul_pack_kind::pack_kn, AccumulateC, M0Tile,
                 a.view(make_ranked_shape(0, k1), fixed_shape<M0Tile, 1>{});
             auto b0 =
                 b.view(make_ranked_shape(k1, 0), fixed_shape<1, N0Tile>{});
-            for (size_t sk1 = 0; sk1 < TLhsElem::shape()[0]; sk1++) {
+            for (size_t sk1 = 0; sk1 < TLhsElem::shape()[1]; sk1++) {
                 using TSubLhsElem = typename TLhsElem::element_type;
                 using TSubRhsElem = ntt::vector<typename TRhsElem::element_type,
                                                 TRhsElem::shape().last()>;
@@ -210,7 +210,7 @@ struct u_matmul<ukernels::mamtul_pack_kind::pack_mkn, AccumulateC, M0Tile,
                 });
 
                 for (size_t k1 = 0; k1 < K; k1++) {
-                    for (size_t sk1 = 0; sk1 < TLhsElem::shape()[0]; sk1++) {
+                    for (size_t sk1 = 0; sk1 < TLhsElem::shape()[1]; sk1++) {
                         using TSubLhsElem = typename TLhsElem::element_type;
                         using TSubRhsElem =
                             ntt::vector<typename TRhsElem::element_type,

diff --git a/src/Native/test/benchmark_test/benchmark_ntt_matmul_primitive_size.cpp b/src/Native/test/benchmark_test/benchmark_ntt_matmul_primitive_size.cpp
@@ -304,7 +304,7 @@ template <size_t M, size_t K, size_t N> void benchmark_ntt_matmul_pack_M_K_N() {
         ntt::tensor<ntt::vector<float, P, P>, ntt::fixed_shape<K / P, N / P>>
             pb;
     alignas(32)
-        ntt::tensor<ntt::vector<float, P, P>, ntt::fixed_shape<K / P, N / P>>
+        ntt::tensor<ntt::vector<float, P, P>, ntt::fixed_shape<M / P, N / P>>
             pc;
     ntt::pack<0, 1>(ta, pa);
     ntt::pack<0, 1>(tb, pb);