diff --git a/cmake/external/xpu.cmake b/cmake/external/xpu.cmake
index cfbe68eecbaca55c5a288aae2c985bbc33d37be2..dbc629a3e3fd2f3b5d99bba78954cd00986310af 100644
--- a/cmake/external/xpu.cmake
+++ b/cmake/external/xpu.cmake
@@ -36,7 +36,7 @@ ENDIF()
 
 if(NOT DEFINED XPU_BASE_URL)
   SET(XPU_BASE_URL_WITHOUT_DATE "https://baidu-kunlun-product.cdn.bcebos.com/KL-SDK/klsdk-dev")
-  SET(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220307")
+  SET(XPU_BASE_URL "${XPU_BASE_URL_WITHOUT_DATE}/20220324")
 else()
   SET(XPU_BASE_URL "${XPU_BASE_URL}")
 endif()
diff --git a/paddle/fluid/memory/allocation/naive_best_fit_allocator.cc b/paddle/fluid/memory/allocation/naive_best_fit_allocator.cc
index 0bfbe2c6962294fc7e4aa2fff079e9cf411f26f8..5efbfce7fedd600a4e679e5b10ee67454a569f4f 100644
--- a/paddle/fluid/memory/allocation/naive_best_fit_allocator.cc
+++ b/paddle/fluid/memory/allocation/naive_best_fit_allocator.cc
@@ -150,7 +150,7 @@ void *Alloc<platform::XPUPlace>(const platform::XPUPlace &place, size_t size) {
   platform::XPUDeviceGuard gurad(place.device);
   int ret = xpu_malloc(reinterpret_cast<void **>(&p), size);
   if (ret != XPU_SUCCESS) {
-    std::cout << "xpu memory malloc(" << size << ") failed, try again\n";
+    VLOG(10) << "xpu memory malloc(" << size << ") failed, try again";
     xpu_wait();
     ret = xpu_malloc(reinterpret_cast<void **>(&p), size);
   }
diff --git a/python/paddle/fluid/tests/unittests/xpu/test_tril_triu_op_xpu.py b/python/paddle/fluid/tests/unittests/xpu/test_tril_triu_op_xpu.py
index 785549abba8f3ad732dd49263532fcf453f0517b..fb6b28d9c282512caf3405d56a4321df9e2668dd 100644
--- a/python/paddle/fluid/tests/unittests/xpu/test_tril_triu_op_xpu.py
+++ b/python/paddle/fluid/tests/unittests/xpu/test_tril_triu_op_xpu.py
@@ -94,17 +94,27 @@ class XPUTestTrilTriuOp(XPUOpTestWrapper):
     class TestTrilTriuOp3(TestTrilTriuOp):
         def initTestCase(self):
             self.diagonal = 10
-            self.Xshape = (25, 25)
+            self.Xshape = (2, 25, 25)
 
     class TestTrilTriuOp4(TestTrilTriuOp):
         def initTestCase(self):
             self.diagonal = -10
-            self.Xshape = (33, 11)
+            self.Xshape = (1, 2, 33, 11)
 
     class TestTrilTriuOp5(TestTrilTriuOp):
         def initTestCase(self):
             self.diagonal = 11
-            self.Xshape = (1, 99)
+            self.Xshape = (1, 1, 99)
+
+    class TestTrilTriuOp6(TestTrilTriuOp):
+        def initTestCase(self):
+            self.diagonal = 5
+            self.Xshape = (1, 2, 3, 5, 99)
+
+    class TestTrilTriuOp7(TestTrilTriuOp):
+        def initTestCase(self):
+            self.diagonal = -100
+            self.Xshape = (2, 2, 3, 4, 5)
 
 
 class TestTrilTriuOpError(unittest.TestCase):