diff --git a/src/lj_opt_mem.c b/src/lj_opt_mem.c
index c8265b4f..b00c347d 100644
--- a/src/lj_opt_mem.c
+++ b/src/lj_opt_mem.c
@@ -827,7 +827,7 @@ TRef LJ_FASTCALL lj_opt_dse_xstore(jit_State *J)
 	IRIns *ir;
 	/* Check for any intervening guards or any XLOADs (no AA performed). */
 	for (ir = IR(J->cur.nins-1); ir > store; ir--)
-	  if (irt_isguard(ir->t) || ir->o == IR_XLOAD)
+	  if (irt_isguard(ir->t) || ir->o == IR_XLOAD || ir->o == IR_INTRN)
 	    goto doemit;  /* No elimination possible. */
 	/* Remove redundant store from chain and replace with NOP. */
 	*refp = store->prev;
diff --git a/tests/intrinsic_spec.lua b/tests/intrinsic_spec.lua
index e95257ff..58db64a1 100644
--- a/tests/intrinsic_spec.lua
+++ b/tests/intrinsic_spec.lua
@@ -712,6 +712,27 @@ end
     
     assert_jit(brand, testcpuid_brand)
   end)
+  
+  it("no dse between intrinsic", function() 
+    assert_cdef([[int32_t volatileload(int32_t* iptr) __mcode("8brMI");]], "volatileload")
+    
+    local numptr = ffi.new("int32_t[1]", 0)
+    
+    local volatileload = ffi.C.volatileload
+    
+    local function testdse()
+      assert(volatileload)
+      local sum = 0
+      numptr[0] = 0
+      numptr[0] = 1
+      sum = volatileload(numptr)
+      numptr[0] = 2
+      sum = sum + volatileload(numptr)
+      return sum
+    end
+    
+    assert_jit(3, testdse)
+  end)
 end)
 
 context("__reglist", function()