diff --git a/tests/regression/sgemm_gemmini_dma/kernel.cpp b/tests/regression/sgemm_gemmini_dma/kernel.cpp
index 4c37a232..2cbd8de5 100644
--- a/tests/regression/sgemm_gemmini_dma/kernel.cpp
+++ b/tests/regression/sgemm_gemmini_dma/kernel.cpp
@@ -49,6 +49,8 @@
 #define rd_cycles_force(x) asm volatile ("csrr %0, mcycle" : "=r" (x))
 #define rd_cycles(x) rd_cycles_force(x)
 #define HW_TID() ({uint32_t gtid; asm volatile ("csrr %0, mhartid" : "=r" (gtid)); gtid;})
+#define MARK_BEG() asm volatile ("slti x0, x1, -1047")
+#define MARK_END() asm volatile ("slti x0, x1, -499")
 #define PRINTF(...) sprintf(PRINT_BUF, __VA_ARGS__)
 // #define PRINTF(...) vx_printf(__VA_ARGS__)
 #define SWISH(beta, x) ((x) / (1 + exp(-(beta) * (x))))
@@ -98,6 +100,7 @@ void thread_block_matmul_gemmini(kernel_arg_t *__UNIFORM__ arg,
 
   uint32_t marker0, marker1;
   rd_cycles_force(marker0);
+  MARK_BEG();
 
   const uint32_t dim_m = arg->dim_m;
   const uint32_t dim_n = arg->dim_n;
@@ -109,15 +112,6 @@ void thread_block_matmul_gemmini(kernel_arg_t *__UNIFORM__ arg,
 
   const uint32_t num_tile_rows_per_tb = num_tiles_m / NUM_CLUSTERS;
 
-  if (HW_TID() == 0) gemmini_fence();
-  threadblock_barrier(3, NUM_WARPS);
-  if (HW_TID() == 0) gemmini_fence();
-  threadblock_barrier(3, NUM_WARPS);
-  if (HW_TID() == 0) gemmini_fence();
-  threadblock_barrier(3, NUM_WARPS);
-  if (HW_TID() == 0) gemmini_fence();
-  threadblock_barrier(3, NUM_WARPS);
-
   if (HW_TID() == 0) {
     gemmini_extended3_config_ld(dim_k * sizeof(elem_t), MVIN_SCALE_IDENTITY, false, 0);
     gemmini_extended3_config_ld(dim_n * sizeof(elem_t), MVIN_SCALE_IDENTITY, false, 1);
@@ -179,6 +173,7 @@ void thread_block_matmul_gemmini(kernel_arg_t *__UNIFORM__ arg,
   // last thread block complete
   if (threadblock_id == NUM_CLUSTERS - 1) {
     threadblock_barrier(/*barrier_id=*/0, /*count=*/NUM_WARPS);
+    MARK_END();
     rd_cycles_force(marker1);
     if (HW_TID() == 0) {
       #ifdef POWER