diff --git a/.github/workflows/integration-tests.yml b/.github/workflows/integration-tests.yml
index 544bce95b..fb69bae2f 100644
--- a/.github/workflows/integration-tests.yml
+++ b/.github/workflows/integration-tests.yml
@@ -23,6 +23,7 @@ jobs:
           alias python='python3'
           cd python
           pip3 install -e .
+          rm -r /tmp/triton/
 
       - name: Unit tests
         run: |
@@ -32,6 +33,7 @@ jobs:
       - name: Regression tests
         run: |
           cd python/test/regression
+          sudo nvidia-smi -i 0 -pm 1
           sudo nvidia-smi -i 0 --lock-gpu-clocks=1350,1350
           sudo nvidia-smi -i 0 --lock-memory-clocks=877,877
           pytest -vs .
diff --git a/lib/codegen/transform/coalesce.cc b/lib/codegen/transform/coalesce.cc
index d736d3570..c3ecbe165 100644
--- a/lib/codegen/transform/coalesce.cc
+++ b/lib/codegen/transform/coalesce.cc
@@ -58,14 +58,14 @@ void coalesce::run(ir::module &mod) {
   for(ir::basic_block *block: fn->blocks())
   for(ir::instruction* i: block->get_inst_list()){
     // coalesce before store
-    if(auto x = dynamic_cast<ir::store_inst*>(i))
-    if(ir::value* op = x->get_value_operand())
+    if(dynamic_cast<ir::store_inst*>(i) || dynamic_cast<ir::atomic_rmw_inst*>(i))
+    if(ir::value* op = i->get_operand(1))
     if(op->get_type()->is_block_ty())
     if(layout_->get(op)->to_mma()){
-      builder.set_insert_point(x);
+      builder.set_insert_point(i);
       ir::instruction* new_op = ir::cvt_layout_inst::create(op);
       builder.insert(new_op);
-      x->replace_uses_of_with(op, new_op);
+      i->replace_uses_of_with(op, new_op);
     }
     // uncoalesce after load
     if(auto x = dynamic_cast<ir::load_inst*>(i))