Merge branch 'feature/ntt_benchmark_roofline_3' of https://github.com…

…/kendryte/nncase into feature/ntt_benchmark_roofline_3
kendryte · Sep 30, 2024 · bfcbeb1 · bfcbeb1
2 parents a553d9a + 0faa9c3
commit bfcbeb1
Showing 1 changed file with 3 additions and 2 deletions.
diff --git a/src/Native/include/nncase/ntt/ukernels/u_matmul.h b/src/Native/include/nncase/ntt/ukernels/u_matmul.h
@@ -210,8 +210,9 @@ struct u_matmul<ukernels::mamtul_pack_kind::pack_mkn, AccumulateC, M0Tile,
                 });
 
                 for (size_t k1 = 0; k1 < K; k1++) {
-                    volatile size_t sk1_max = TLhsElem::shape()[1];
-                    [[assume(sk1_max > 0)]];
+                    // Force compiler do not unroll the loop
+                    size_t sk1_max = TLhsElem::shape()[1];
+                    #pragma GCC unroll 1
                     for (size_t sk1 = 0; sk1 < sk1_max; sk1++) {
                         using TSubLhsElem = typename TLhsElem::element_type;
                         using TSubRhsElem =