diff --git a/src/04kernel/src/kernels/batch_normalization/cnnl_kernel.cc b/src/04kernel/src/kernels/batch_normalization/cnnl_kernel.cc
index e3efcd1d..1330cfa7 100644
--- a/src/04kernel/src/kernels/batch_normalization/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/batch_normalization/cnnl_kernel.cc
@@ -128,8 +128,8 @@ namespace refactor::kernel {
             auto y = outputs[0];
 
             void *xTrans = workspace;
-            void *yTrans = xTrans + xTransSize;
-            void *cursor = yTrans + xTransSize;
+            void *yTrans = reinterpret_cast<uint8_t *>(xTrans) + xTransSize;
+            void *cursor = reinterpret_cast<uint8_t *>(yTrans) + xTransSize;
 
             // transpose NCHW input to NHWC
             CNNL_ASSERT(cnnlTranspose_v2(handle, d->NCHW2NHWC, d->inDesc, x,
@@ -147,7 +147,6 @@ namespace refactor::kernel {
             CNNL_ASSERT(cnnlTranspose_v2(handle, d->NHWC2NCHW, d->inDescTrans, yTrans,
                                          d->inDesc, y, cursor, workspaceSize));
 
-            BANG_ASSERT(cnrtQueueSync(res.fetchOrStore<CnnlContext>()->queue));
         };
 
         return {std::move(routine), totalWorkspaceSize};
diff --git a/src/04kernel/src/kernels/cast/cnnl_kernel.cc b/src/04kernel/src/kernels/cast/cnnl_kernel.cc
index 95120974..54798e67 100644
--- a/src/04kernel/src/kernels/cast/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/cast/cnnl_kernel.cc
@@ -65,7 +65,6 @@ namespace refactor::kernel {
         return [d = std::move(d)](Resources &res, void *workspace, void const *const *inputs, void *const *outputs) {
             CNNL_ASSERT(cnnlCastDataType(res.fetchOrStore<CnnlContext>()->handle,
                                          d->inDesc, inputs[0], d->cast, d->outDesc, outputs[0]));
-            // BANG_ASSERT(cnrtQueueSync(res.fetchOrStore<CnnlContext>()->queue));
         };
     }
 
diff --git a/src/04kernel/src/kernels/clip/cnnl_kernel.cc b/src/04kernel/src/kernels/clip/cnnl_kernel.cc
index 423939bc..fe65e99b 100644
--- a/src/04kernel/src/kernels/clip/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/clip/cnnl_kernel.cc
@@ -57,7 +57,6 @@ namespace refactor::kernel {
                                     CNNL_POINTER_MODE_DEVICE, d->t,
                                     inputs[0], inputs[1], hasMax ? inputs[2] : nullptr,
                                     d->t, outputs[0]));
-            BANG_ASSERT(cnrtQueueSync(res.fetchOrStore<CnnlContext>()->queue));
         };
     }
 
diff --git a/src/04kernel/src/kernels/concat/cnnl_kernel.cc b/src/04kernel/src/kernels/concat/cnnl_kernel.cc
index 6c7a0fba..c35b1c33 100644
--- a/src/04kernel/src/kernels/concat/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/concat/cnnl_kernel.cc
@@ -52,7 +52,7 @@ namespace refactor::kernel {
             }
             ~Descriptors() noexcept(false) {
                 CNNL_ASSERT(cnnlDestroyTensorDescriptor(in));
-                for (auto i = 0; i < out.size(); i++) {
+                for (size_t i = 0; i < out.size(); i++) {
                     CNNL_ASSERT(cnnlDestroyTensorDescriptor(out[i]));
                 }
             }
@@ -62,7 +62,7 @@ namespace refactor::kernel {
         };
         auto d = std::make_shared<Descriptors>(info.num, info.dataType != DT::F64);
         setCnnlTensor(d->in, info.dataType, slice(info.inDim.data(), info.inDim.size()));
-        for (auto i = 0; i < info.outDims.size(); i++) {
+        for (size_t i = 0; i < info.outDims.size(); i++) {
             setCnnlTensor(d->out[i], info.dataType, slice(info.outDims[i].data(), info.outDims[i].size()));
         }
 
diff --git a/src/04kernel/src/kernels/conv/cnnl_kernel.cc b/src/04kernel/src/kernels/conv/cnnl_kernel.cc
index 85dcb951..187ac4b5 100644
--- a/src/04kernel/src/kernels/conv/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/conv/cnnl_kernel.cc
@@ -209,9 +209,9 @@ namespace refactor::kernel {
             // }
 
             void *xTrans = workspace;
-            void *wTrans = xTrans + xTransSize;
-            void *yTrans = wTrans + wTransSize;
-            void *opWorkspace = yTrans + yTransSize;
+            void *wTrans = reinterpret_cast<uint8_t *>(xTrans) + xTransSize;
+            void *yTrans = reinterpret_cast<uint8_t *>(wTrans) + wTransSize;
+            void *opWorkspace = reinterpret_cast<uint8_t *>(yTrans) + yTransSize;
 
             // transpose NCHW input to NHWC
             CNNL_ASSERT(cnnlTranspose_v2(handle, d->NCHW2NHWC, d->x, x,
diff --git a/src/04kernel/src/kernels/expand/cnnl_kernel.cc b/src/04kernel/src/kernels/expand/cnnl_kernel.cc
index 18f58642..426eac49 100644
--- a/src/04kernel/src/kernels/expand/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/expand/cnnl_kernel.cc
@@ -60,7 +60,6 @@ namespace refactor::kernel {
         return [d = std::move(d)](Resources &res, void *workspace, void const *const *inputs, void *const *outputs) {
             CNNL_ASSERT(cnnlExpand(res.fetchOrStore<CnnlContext>()->handle,
                                    d->inDesc, inputs[0], d->outDesc, outputs[0]));
-            // BANG_ASSERT(cnrtQueueSync(res.fetchOrStore<CnnlContext>()->queue));
         };
     }
 #endif
diff --git a/src/04kernel/src/kernels/gather/cnnl_kernel.cc b/src/04kernel/src/kernels/gather/cnnl_kernel.cc
index b4d5aa15..abdf0d5c 100644
--- a/src/04kernel/src/kernels/gather/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/gather/cnnl_kernel.cc
@@ -79,7 +79,6 @@ namespace refactor::kernel {
                                      d->inDesc, inputs[0], reinterpret_cast<const int *>(workspace),
                                      d->indexDesc, reinterpret_cast<const int *>(inputs[1]),
                                      d->outDesc, outputs[0]));
-            BANG_ASSERT(cnrtQueueSync(res.fetchOrStore<CnnlContext>()->queue));
        };
 
         return {std::move(routine), workspaceSize};
diff --git a/src/04kernel/src/kernels/mat_mul/cnnl_kernel.cc b/src/04kernel/src/kernels/mat_mul/cnnl_kernel.cc
index 7be05b7b..3eac3572 100644
--- a/src/04kernel/src/kernels/mat_mul/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/mat_mul/cnnl_kernel.cc
@@ -141,7 +141,6 @@ namespace refactor::kernel {
                     workspace, algoWorkspaceSize));
             }
 
-            BANG_ASSERT(cnrtQueueSync(res.fetchOrStore<CnnlContext>()->queue));
         };
 
         return {std::move(routine), algoWorkspaceSize};
diff --git a/src/04kernel/src/kernels/pool/cnnl_kernel.cc b/src/04kernel/src/kernels/pool/cnnl_kernel.cc
index 083125b1..929ea578 100644
--- a/src/04kernel/src/kernels/pool/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/pool/cnnl_kernel.cc
@@ -130,7 +130,7 @@ namespace refactor::kernel {
             auto handle = res.fetchOrStore<CnnlContext>()->handle;
            
             void *extraInputDev = workspace;
-            void *poolWorkSpace = workspace + extraInputSize;
+            void *poolWorkSpace = reinterpret_cast<uint8_t *>(workspace) + extraInputSize;
 
             void *extraInputHost = malloc(extraInputSize);
             CNNL_ASSERT(cnnlInitPoolingExtraInput(handle, d->pooling, d->x, d->y, extraInputHost));
@@ -145,7 +145,7 @@ namespace refactor::kernel {
                 &b, extraInputDev, d->y, outputs[0],
                 poolWorkSpace, workspaceSize));
 
-            BANG_ASSERT(cnrtQueueSync(res.fetchOrStore<CnnlContext>()->queue));
+            res.fetchOrStore<CnnlContext>()->queueSync();
 
             free(extraInputHost);
         };
diff --git a/src/04kernel/src/kernels/simple_binary/binary_cnnl.cc b/src/04kernel/src/kernels/simple_binary/binary_cnnl.cc
index 11712b4f..8b4647ac 100644
--- a/src/04kernel/src/kernels/simple_binary/binary_cnnl.cc
+++ b/src/04kernel/src/kernels/simple_binary/binary_cnnl.cc
@@ -180,7 +180,6 @@ namespace refactor::kernel {
                                          workspace, workspaceSize));
             }
             
-            BANG_ASSERT(cnrtQueueSync(res.fetchOrStore<CnnlContext>()->queue));
         };
 
         return {std::move(routine), workspaceSize};
diff --git a/src/04kernel/src/kernels/slice/cnnl_kernel.cc b/src/04kernel/src/kernels/slice/cnnl_kernel.cc
index b5f8b4c8..85bc9093 100644
--- a/src/04kernel/src/kernels/slice/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/slice/cnnl_kernel.cc
@@ -64,7 +64,7 @@ namespace refactor::kernel {
         CNNL_ASSERT(cnnlSetTensorDescriptor(d->in, CNNL_LAYOUT_NCHW, cnnlDataTypeConvert(info.dataType), info.inDim.size(), info.inDim.data()));
         CNNL_ASSERT(cnnlSetTensorDescriptor(d->out, CNNL_LAYOUT_NCHW, cnnlDataTypeConvert(info.dataType), info.outDim.size(), info.outDim.data()));
         std::vector<int> begin, end, stride;
-        for (auto i = 0; i < info.dims.size(); i++) {
+        for (size_t i = 0; i < info.dims.size(); i++) {
             // [begin, end), end is not inclued
             begin.push_back(info.dims[i].start);
             auto sign = info.dims[i].step > 0 ? 1 : -1;
diff --git a/src/04kernel/src/kernels/softmax/cnnl_kernel.cc b/src/04kernel/src/kernels/softmax/cnnl_kernel.cc
index 0633195d..babaf33c 100644
--- a/src/04kernel/src/kernels/softmax/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/softmax/cnnl_kernel.cc
@@ -80,7 +80,6 @@ namespace refactor::kernel {
                 CNNL_COMPUTATION_ULTRAHIGH_PRECISION,
                 &a, d->t, inputs[0],
                 &b, d->t, outputs[0]));
-            res.fetchOrStore<CnnlContext>()->queueSync();
         };
     }
 
diff --git a/src/04kernel/src/kernels/split/cnnl_kernel.cc b/src/04kernel/src/kernels/split/cnnl_kernel.cc
index c967bff2..8f686d59 100644
--- a/src/04kernel/src/kernels/split/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/split/cnnl_kernel.cc
@@ -69,7 +69,7 @@ namespace refactor::kernel {
             }
             ~Descriptors() noexcept(false) {
                 CNNL_ASSERT(cnnlDestroyTensorDescriptor(in));
-                for (auto i = 0; i < out.size(); i++) {
+                for (size_t i = 0; i < out.size(); i++) {
                     CNNL_ASSERT(cnnlDestroyTensorDescriptor(out[i]));
                 }
             }
@@ -81,7 +81,7 @@ namespace refactor::kernel {
         // setCnnlTensor(d->in, info.dataType, slice(info.inDim.data(), info.inDim.size()));
         CNNL_ASSERT(cnnlSetTensorDescriptor(d->in, CNNL_LAYOUT_NCHW, cnnlDataTypeConvert(info.dataType), info.inDim.size(), info.inDim.data()));
 
-        for (auto i = 0; i < info.outDims.size(); i++) {
+        for (size_t i = 0; i < info.outDims.size(); i++) {
             // setCnnlTensor(d->out[i], info.dataType, slice(info.outDims[i].data(), info.outDims[i].size()));
             CNNL_ASSERT(cnnlSetTensorDescriptor(d->out[i], CNNL_LAYOUT_NCHW, cnnlDataTypeConvert(info.dataType), info.outDims[i].size(), info.outDims[i].data()));
         }
diff --git a/src/04kernel/src/kernels/where/cnnl_kernel.cc b/src/04kernel/src/kernels/where/cnnl_kernel.cc
index 50b7c9d8..14f8d667 100644
--- a/src/04kernel/src/kernels/where/cnnl_kernel.cc
+++ b/src/04kernel/src/kernels/where/cnnl_kernel.cc
@@ -102,7 +102,6 @@ namespace refactor::kernel {
                 d->y, inputs[2], workspace, workspaceSize,
                 d->ans, outputs[0]));
 
-            res.fetchOrStore<CnnlContext>()->queueSync();
         };
 
         return {std::move(routine), workspaceSize};
diff --git a/src/04kernel/src/utilities/bang/cnrt_functions.cc b/src/04kernel/src/utilities/bang/cnrt_functions.cc
index 2ea66194..26c1b975 100644
--- a/src/04kernel/src/utilities/bang/cnrt_functions.cc
+++ b/src/04kernel/src/utilities/bang/cnrt_functions.cc
@@ -4,7 +4,7 @@
 #include <cnrt.h>
 #include <cstdio>
 
-namespace refactor::kernel::cnnl {
+namespace refactor::kernel::bang {
 
     int currentDevice() {
         int device;
@@ -22,6 +22,6 @@ namespace refactor::kernel::cnnl {
                                CNRT_MEM_TRANS_DIR_DEV2HOST));
     }
 
-}// namespace refactor::kernel::cnnl
+}// namespace refactor::kernel::bang
 
 #endif
diff --git a/src/04kernel/src/utilities/bang/cnrt_functions.h b/src/04kernel/src/utilities/bang/cnrt_functions.h
index ef119819..3a05195c 100644
--- a/src/04kernel/src/utilities/bang/cnrt_functions.h
+++ b/src/04kernel/src/utilities/bang/cnrt_functions.h
@@ -3,7 +3,7 @@
 
 #include "common.h"
 
-namespace refactor::kernel::cnnl {
+namespace refactor::kernel::bang {
 
     int currentDevice();
 
@@ -11,6 +11,6 @@ namespace refactor::kernel::cnnl {
 
     void copyOut(void *dst, const void *src, size_t size);
 
-}// namespace refactor::kernel::cnnl
+}// namespace refactor::kernel::bang
 
 #endif// KERNEL_CNRT_FUNCTIONS_H
diff --git a/src/04kernel/test/kernels/batch_normalization/test_cnnl.cpp b/src/04kernel/test/kernels/batch_normalization/test_cnnl.cpp
index 14a1a07f..d1ad9bd7 100644
--- a/src/04kernel/test/kernels/batch_normalization/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/batch_normalization/test_cnnl.cpp
@@ -2,6 +2,7 @@
 
 #include "../../../src/kernels/batch_normalization/cnnl_kernel.hh"
 #include "../../../src/kernels/batch_normalization/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -57,6 +58,7 @@ TEST(kernel, BatchNormalizationCnnl) {
         void const *inputs[]{*mluIn, *mluScale, *mluBias, *mluMean, *mluVar};
         void *outputs[]{*mluOut};
         rMlu(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     // take output data
     std::vector<float> result(outTensor->elementsSize());
diff --git a/src/04kernel/test/kernels/cast/test_cnnl.cpp b/src/04kernel/test/kernels/cast/test_cnnl.cpp
index 72f84247..94297357 100644
--- a/src/04kernel/test/kernels/cast/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/cast/test_cnnl.cpp
@@ -1,7 +1,8 @@
 #ifdef USE_BANG
 
-#include "../../../src/kernels/cast/cpu_kernel.hh"
 #include "../../../src/kernels/cast/cnnl_kernel.hh"
+#include "../../../src/kernels/cast/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -34,6 +35,7 @@ TEST(kernel, CastCnnl) {
         void const *inputs[]{*xMlu};
         void *outputs[]{*yMlu};
         routine(res, nullptr, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{x_.data()};
diff --git a/src/04kernel/test/kernels/clip/test_cnnl.cpp b/src/04kernel/test/kernels/clip/test_cnnl.cpp
index ab29a1cb..ff2e7729 100644
--- a/src/04kernel/test/kernels/clip/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/clip/test_cnnl.cpp
@@ -1,7 +1,8 @@
 #ifdef USE_BANG
 
-#include "../../../src/kernels/clip/cpu_kernel.hh"
 #include "../../../src/kernels/clip/cnnl_kernel.hh"
+#include "../../../src/kernels/clip/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -36,6 +37,7 @@ TEST(kernel, ClipCnnl) {
         void const *inputs[]{*mluMem, *mluMin, *mluMax};
         void *outputs[]{*mluMem};
         routine(res, nullptr, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{value.data(), &min, &max};
diff --git a/src/04kernel/test/kernels/concat/test_cnnl.cpp b/src/04kernel/test/kernels/concat/test_cnnl.cpp
index 2911c9e9..ecc817ac 100644
--- a/src/04kernel/test/kernels/concat/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/concat/test_cnnl.cpp
@@ -1,7 +1,8 @@
 #ifdef USE_BANG
 
-#include "../../../src/kernels/concat/cpu_kernel.hh"
 #include "../../../src/kernels/concat/cnnl_kernel.hh"
+#include "../../../src/kernels/concat/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -65,6 +66,7 @@ TEST(kernel, ConcatCnnl) {
         void const *inputs[]{*mluIns[0], *mluIns[1], *mluIns[2], *mluIns[3]};
         void *outputs[]{*mluOut};
         routine(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{cpuIns[0].data(), cpuIns[1].data(), cpuIns[2].data(), cpuIns[3].data()};
diff --git a/src/04kernel/test/kernels/conv/test_cnnl.cpp b/src/04kernel/test/kernels/conv/test_cnnl.cpp
index bf4bff3c..74e799f5 100644
--- a/src/04kernel/test/kernels/conv/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/conv/test_cnnl.cpp
@@ -1,6 +1,7 @@
 #ifdef USE_BANG
 
 #include "../../../src/kernels/conv/cnnl_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -39,15 +40,7 @@ void testConvCnnl(int rank, const int64_t *pads, const int64_t *strides, const i
     void const *inputs[]{*xMlu, *wMlu};
     void *outputs[]{*yMlu};
     routine(res, *workspace, inputs, outputs);
-
-    xMlu->copyToHost(xData.data(), xTensor->bytesSize());
-    wMlu->copyToHost(wData.data(), wTensor->bytesSize());
-    // fmt::println("{}", vec2str(xData));
-    // fmt::println("{}", vec2str(wData));
-
-    // std::vector<float> ws(workspaceSize);
-    // workspace->copyToHost(ws.data(), workspaceSize);
-    // fmt::println("{}", vec2str(ws));
+    kernel::bang::sync();
 
     // take output data
     std::vector<float> result(yTensor->elementsSize());
diff --git a/src/04kernel/test/kernels/expand/test_cnnl.cpp b/src/04kernel/test/kernels/expand/test_cnnl.cpp
index e06e5977..43fb07e8 100644
--- a/src/04kernel/test/kernels/expand/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/expand/test_cnnl.cpp
@@ -2,6 +2,7 @@
 
 #include "../../../src/kernels/expand/cnnl_kernel.hh"
 #include "../../../src/kernels/expand/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -36,6 +37,7 @@ TEST(kernel, ExpandCnnl) {
         void const *inputs[]{*mluIn};
         void *outputs[]{*mluOut};
         routine(res, nullptr, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{data.data()};
diff --git a/src/04kernel/test/kernels/gather/test_gather_cnnl.cpp b/src/04kernel/test/kernels/gather/test_gather_cnnl.cpp
index 020b5f91..b63682d2 100644
--- a/src/04kernel/test/kernels/gather/test_gather_cnnl.cpp
+++ b/src/04kernel/test/kernels/gather/test_gather_cnnl.cpp
@@ -2,6 +2,7 @@
 
 #include "../src/kernels/gather/cnnl_kernel.hh"
 #include "../src/kernels/gather/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -39,6 +40,7 @@ TEST(kernel, GatherCnnl) {
             void const *inputs[]{*aMLU, *bMLU};
             void *outputs[]{*cMLU};
             cnnlRoutine(res, *workspace, inputs, outputs);
+            kernel::bang::sync();
         }
         {
             void const *inputs[]{a.data(), b.data()};
@@ -81,6 +83,7 @@ TEST(kernel, GatherCnnl) {
             void const *inputs[]{*aMLU, *bMLU};
             void *outputs[]{*cMLU};
             cnnlRoutine(res, *workspace, inputs, outputs);
+            kernel::bang::sync();
         }
         {
             void const *inputs[]{a.data(), b.data()};
@@ -110,7 +113,7 @@ TEST(kernel, GatherCnnl) {
         auto cpuRoutine = cpuKernel->lower(res).routine;
         // Init inputs and outputs
         std::vector<float> a;
-        for (auto i = 0; i < data->elementsSize(); i++) {
+        for (size_t i = 0; i < data->elementsSize(); i++) {
             a.push_back(i + 0.1f);
         }
         std::vector<int64_t> b(indices->elementsSize(), 0);
@@ -126,6 +129,7 @@ TEST(kernel, GatherCnnl) {
             void const *inputs[]{*aMLU, *bMLU};
             void *outputs[]{*cMLU};
             cnnlRoutine(res, *workspace, inputs, outputs);
+            kernel::bang::sync();
         }
         {
             void const *inputs[]{a.data(), b.data()};
diff --git a/src/04kernel/test/kernels/mat_mul/test_cnnl.cpp b/src/04kernel/test/kernels/mat_mul/test_cnnl.cpp
index 251c5c89..f079b444 100644
--- a/src/04kernel/test/kernels/mat_mul/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/mat_mul/test_cnnl.cpp
@@ -2,6 +2,7 @@
 
 #include "../src/kernels/mat_mul/cnnl_kernel.hh"
 #include "../src/kernels/mat_mul/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -48,6 +49,7 @@ TEST(kernel, MatMulCnnl_OnlyBias) {
     void const *inputs[]{*ma, *mb, *mc};
     void *outputs[]{*my};
     routine(res, *workspace, inputs, outputs);
+    kernel::bang::sync();
     // take output data
     std::vector<float> result(Y->elementsSize());
     my->copyToHost(result.data(), Y->bytesSize());
@@ -91,6 +93,7 @@ TEST(kernel, MatMulCnnl_Broadcast) {
         void const *inputs[]{*ma, *mb, *mc};
         void *outputs[]{*my};
         mluRoutine(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{dataA.data(), dataB.data(), dataC.data()};
@@ -135,6 +138,7 @@ TEST(kernel, MatMulCnnl_TransABNoBias) {
         void const *inputs[]{*ma, *mb};
         void *outputs[]{*my};
         mluRoutine(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{dataA.data(), dataB.data()};
@@ -189,6 +193,7 @@ TEST(kernel, MatMulCnnl_Large) {
         void const *inputs[]{*ma, *mb, *mc};
         void *outputs[]{*my};
         mluRoutine(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{dataA.data(), dataB.data(), dataC.data()};
diff --git a/src/04kernel/test/kernels/pool/test_cnnl.cpp b/src/04kernel/test/kernels/pool/test_cnnl.cpp
index 405bf3f8..1adf4513 100644
--- a/src/04kernel/test/kernels/pool/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/pool/test_cnnl.cpp
@@ -1,6 +1,7 @@
 #ifdef USE_BANG
 
 #include "../../../src/kernels/pool/cnnl_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -32,6 +33,7 @@ void testPoolCnnl(PoolType poolType, int rank, const int64_t *pads, const int64_
     void const *inputs[]{*mluMem};
     void *outputs[]{*mluMem};
     routine(res, *workspace, inputs, outputs);
+    kernel::bang::sync();
     // take output data
     std::vector<float> result(yTensor->elementsSize());
     mluMem->copyToHost(result.data(), yTensor->bytesSize());
diff --git a/src/04kernel/test/kernels/reduce/test_cnnl.cpp b/src/04kernel/test/kernels/reduce/test_cnnl.cpp
index 32952fea..113fe766 100644
--- a/src/04kernel/test/kernels/reduce/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/reduce/test_cnnl.cpp
@@ -1,6 +1,7 @@
 #ifdef USE_BANG
 
 #include "../../../src/kernels/reduce/cnnl_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -28,6 +29,7 @@ static void testReducemean(const Shape &shape, const std::vector<float> &data,
         void const *inputs[]{*mluMemIn};
         void *outputs[]{*mluMemOut};
         routine(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     // take output data
     Shape outDimArray;
diff --git a/src/04kernel/test/kernels/scatter_nd/test_cnnl.cpp b/src/04kernel/test/kernels/scatter_nd/test_cnnl.cpp
index 1cdf7216..cf5b9c36 100644
--- a/src/04kernel/test/kernels/scatter_nd/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/scatter_nd/test_cnnl.cpp
@@ -1,7 +1,8 @@
 #ifdef USE_BANG
 
-#include "../../../src/kernels/scatter_nd/cpu_kernel.hh"
 #include "../../../src/kernels/scatter_nd/cnnl_kernel.hh"
+#include "../../../src/kernels/scatter_nd/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -48,6 +49,7 @@ TEST(kernel, ScatterNDCnnl) {
         void const *inputs[]{*mluData, *mluIndices, *mluUpdates};
         void *outputs[]{*mluOut};
         routine(res, nullptr, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{data_.data(), indices_.data(), updates_.data()};
diff --git a/src/04kernel/test/kernels/simple_binary/test_binary_cnnl.cpp b/src/04kernel/test/kernels/simple_binary/test_binary_cnnl.cpp
index 7bd18bf6..dfd3c9b8 100644
--- a/src/04kernel/test/kernels/simple_binary/test_binary_cnnl.cpp
+++ b/src/04kernel/test/kernels/simple_binary/test_binary_cnnl.cpp
@@ -2,6 +2,7 @@
 
 #include "../src/kernels/simple_binary/binary_cnnl.hh"
 #include "../src/kernels/simple_binary/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -38,6 +39,7 @@ void testBinaryCnnl(SimpleBinaryType binaryOPT, Shape dimA, Shape dimB, Shape di
         void const *inputs[]{*aMLU, *bMLU};
         void *outputs[]{*cMLU};
         routine(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{a.data(), b.data()};
@@ -100,27 +102,6 @@ TEST(kernel, BinaryCnnlFMod) {
                                   Shape{2, 5, 10, 20, 3, 4});
 }
 
-// TEST(kernel, BinaryCnnlMod) {
-//     testBinaryCnnl<DataType::I8>(SimpleBinaryType::Mod,
-//                                  Shape{2, 5, 10, 20, 3, 4},
-//                                  Shape{2, 5, 10, 20, 3, 4},
-//                                  Shape{2, 5, 10, 20, 3, 4});
-// }
-
-// TEST(kernel, BinaryCnnlFmodI8) {
-//     testBinaryCnnl<DataType::I8>(SimpleBinaryType::Fmod,
-//                                  Shape{2, 5, 10, 20, 3, 4},
-//                                  Shape{2, 5, 10, 20, 3, 4},
-//                                  Shape{2, 5, 10, 20, 3, 4});
-// }
-
-// TEST(kernel, BinaryCnnlFmodF32) {
-//     testBinaryCnnl<DataType::F32>(SimpleBinaryType::Fmod,
-//                                   Shape{2, 5, 10, 20, 3, 4},
-//                                   Shape{2, 5, 10, 20, 3, 4},
-//                                   Shape{2, 5, 10, 20, 3, 4});
-// }
-
 TEST(kernel, BinaryCnnlBroadcast) {
     testBinaryCnnl<DataType::F32>(SimpleBinaryType::Add, Shape{1, 2, 3, 4, 5, 6}, Shape{}, Shape{1, 2, 3, 4, 5, 6});
 }
diff --git a/src/04kernel/test/kernels/simple_unary/test_cnnl.cpp b/src/04kernel/test/kernels/simple_unary/test_cnnl.cpp
index e4b9428e..8e1e8a9f 100644
--- a/src/04kernel/test/kernels/simple_unary/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/simple_unary/test_cnnl.cpp
@@ -3,6 +3,7 @@
 #include "../../../src/kernels/simple_unary/cnnl_activation_kernel.hh"
 #include "../../../src/kernels/simple_unary/cnnl_simple_unary_kernel.hh"
 #include "../../../src/kernels/simple_unary/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -32,6 +33,7 @@ static void testOp(SimpleUnaryType opType, bool activation = true) {
         void const *inputs[]{*mluMem};
         void *outputs[]{*mluMem};
         routine(res, nullptr, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{data.data()};
diff --git a/src/04kernel/test/kernels/slice/test_cnnl.cpp b/src/04kernel/test/kernels/slice/test_cnnl.cpp
index 914a8fee..1685d7aa 100644
--- a/src/04kernel/test/kernels/slice/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/slice/test_cnnl.cpp
@@ -1,7 +1,8 @@
 #ifdef USE_BANG
 
-#include "../../../src/kernels/slice/cpu_kernel.hh"
 #include "../../../src/kernels/slice/cnnl_kernel.hh"
+#include "../../../src/kernels/slice/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -45,6 +46,7 @@ TEST(kernel, SliceCnnl) {
         void const *inputs[]{*mluIn};
         void *outputs[]{*mluOut};
         routine(res, nullptr, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{data.data()};
diff --git a/src/04kernel/test/kernels/softmax/test_cnnl.cpp b/src/04kernel/test/kernels/softmax/test_cnnl.cpp
index a8c7fb28..09874d20 100644
--- a/src/04kernel/test/kernels/softmax/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/softmax/test_cnnl.cpp
@@ -1,7 +1,8 @@
 #ifdef USE_BANG
 
-#include "../../../src/kernels/softmax/cpu_kernel.hh"
 #include "../../../src/kernels/softmax/cnnl_kernel.hh"
+#include "../../../src/kernels/softmax/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -39,6 +40,7 @@ TEST(kernel, SoftmaxCnnl) {
         void const *inputs[]{*mluIn};
         void *outputs[]{*mluOut};
         rCnnl(res, nullptr, inputs, outputs);
+        kernel::bang::sync();
     }
     // take output data
     std::vector<float> result(outTensor->elementsSize());
diff --git a/src/04kernel/test/kernels/split/test_cnnl.cpp b/src/04kernel/test/kernels/split/test_cnnl.cpp
index 5afa434e..71e69b21 100644
--- a/src/04kernel/test/kernels/split/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/split/test_cnnl.cpp
@@ -1,7 +1,8 @@
 #ifdef USE_BANG
 
-#include "../../../src/kernels/split/cpu_kernel.hh"
 #include "../../../src/kernels/split/cnnl_kernel.hh"
+#include "../../../src/kernels/split/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -64,6 +65,7 @@ TEST(kernel, SplitCnnl) {
         void const *inputs[]{*mluIn};
         void *outputs[]{*mluOuts[0], *mluOuts[1], *mluOuts[2], *mluOuts[3]};
         routine(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     {
         void const *inputs[]{data.data()};
diff --git a/src/04kernel/test/kernels/transpose/test_cnnl.cpp b/src/04kernel/test/kernels/transpose/test_cnnl.cpp
index 4f4301d8..f4689071 100644
--- a/src/04kernel/test/kernels/transpose/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/transpose/test_cnnl.cpp
@@ -2,6 +2,7 @@
 
 #include "../../../src/kernels/transpose/cnnl_kernel.hh"
 #include "../../../src/kernels/transpose/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 #include <numeric>
@@ -42,6 +43,7 @@ TEST(kernel, TransposeCnnl) {
         void const *inputs[]{*mluIn};
         void *outputs[]{*mluOut};
         routine(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     // take output data
     std::vector<float> result(dataTensor->elementsSize());
diff --git a/src/04kernel/test/kernels/where/test_cnnl.cpp b/src/04kernel/test/kernels/where/test_cnnl.cpp
index aaad6585..6e26ddbd 100644
--- a/src/04kernel/test/kernels/where/test_cnnl.cpp
+++ b/src/04kernel/test/kernels/where/test_cnnl.cpp
@@ -2,6 +2,7 @@
 
 #include "../../../src/kernels/where/cnnl_kernel.hh"
 #include "../../../src/kernels/where/cpu_kernel.hh"
+#include "../src/utilities/bang/cnrt_functions.h"
 #include "hardware/device_manager.h"
 #include <gtest/gtest.h>
 
@@ -49,6 +50,7 @@ void testWhereCnnl(Shape cDim, Shape xDim, Shape yDim, Shape outDim) {
         void const *inputs[]{*mluC, *mluX, *mluY};
         void *outputs[]{*mluOut};
         rCnnl(res, *workspace, inputs, outputs);
+        kernel::bang::sync();
     }
     // take output data
     std::vector<float> result(outTensor->elementsSize());
diff --git a/src/09python_ffi/src/executor.cc b/src/09python_ffi/src/executor.cc
index 947410cc..1d0e543f 100644
--- a/src/09python_ffi/src/executor.cc
+++ b/src/09python_ffi/src/executor.cc
@@ -76,7 +76,7 @@ namespace refactor::python_ffi {
         auto ans = _stream.bench(sync ? kernel::cuda::sync : nullptr);
 #else
     #ifdef USE_BANG
-        auto ans = _stream.bench(sync ? kernel::cnnl::sync : nullptr);
+        auto ans = _stream.bench(sync ? kernel::bang::sync : nullptr);
 #else
         auto ans = _stream.bench(nullptr);
     #endif
@@ -222,7 +222,7 @@ namespace refactor::python_ffi {
                 kernel::cuda::copyOut(buffer.data(), addresses[idx], size);
 #endif
 #ifdef USE_BANG
-                kernel::cnnl::copyOut(buffer.data(), addresses[idx], size);
+                kernel::bang::copyOut(buffer.data(), addresses[idx], size);
 #endif
 
                 auto file = path / fmt::format("data{:06}.{}", dataIdx++, format);