diff --git a/bench/benchmark-blocking-sizes.cpp b/bench/benchmark-blocking-sizes.cpp
index 0bf9c07f8..465f5d2c8 100644
--- a/bench/benchmark-blocking-sizes.cpp
+++ b/bench/benchmark-blocking-sizes.cpp
@@ -199,7 +199,7 @@ void benchmark_t::run()
 
     double starttime = timer.getCpuTime();
     for (int i = 0; i < iters_at_a_time; i++) {
-      dst[matrix_index] = lhs[matrix_index] * rhs[matrix_index];
+      dst[matrix_index].noalias() = lhs[matrix_index] * rhs[matrix_index];
       matrix_index++;
       if (matrix_index == matrix_pool_size) {
         matrix_index = 0;
diff --git a/bench/perf_monitoring/gemm/changesets.txt b/bench/perf_monitoring/gemm/changesets.txt
index b379d7bd2..a5b63bc89 100644
--- a/bench/perf_monitoring/gemm/changesets.txt
+++ b/bench/perf_monitoring/gemm/changesets.txt
@@ -40,3 +40,4 @@ before-evaluators
 6937:c8c042f286b2   # avoid redundant pack_rhs
 6981:7e5d6f78da59   # dynamic loop swapping
 6984:45f26866c091   # rm dynamic loop swapping, adjust lhs's micro panel height to fully exploit L1 cache
+6986:a675d05b6f8f   # blocking heuristic: block on the rhs in L1 if the lhs fit in L1.