diff --git a/docs/quick_start/source_en/07_fuse_preprocess.rst b/docs/quick_start/source_en/07_fuse_preprocess.rst
index 4fdc4fe27..b33c67f21 100644
--- a/docs/quick_start/source_en/07_fuse_preprocess.rst
+++ b/docs/quick_start/source_en/07_fuse_preprocess.rst
@@ -19,8 +19,8 @@ At present, the two main series of chips supported by TPU-MLIR are BM168x and CV
    * - NHWC to NCHW
      - True
      - True
-   * - BGR to RGB
-     - False
+   * - BGR/ RGB Conversion
+     - True
      - True
 
 The image cropping will first adjust the image to the size specified by the "--resize_dims" argument of the model_transform tool, and then crop it to the size of the model input. The normalization supports directly converting unpreprocessed image data.
diff --git a/docs/quick_start/source_zh/07_fuse_preprocess.rst b/docs/quick_start/source_zh/07_fuse_preprocess.rst
index fc551a8e9..7db1df7b0 100644
--- a/docs/quick_start/source_zh/07_fuse_preprocess.rst
+++ b/docs/quick_start/source_zh/07_fuse_preprocess.rst
@@ -19,8 +19,8 @@
    * - NHWC to NCHW
      - True
      - True
-   * - BGR to RGB
-     - False
+   * - BGR/RGB 转换
+     - True
      - True
 
 其中图像裁剪会先将图片按使用model_transform工具时输入的“--resize_dims”参数将图片调整为对应的大小, 再裁剪成模型输入的尺寸。而归一化计算支持直接将未进行预处理的图像数据(即unsigned int8格式的数据)做归一化处理。
diff --git a/include/tpu_mlir/Backend/BM168x/Param.h b/include/tpu_mlir/Backend/BM168x/Param.h
index 078167e65..ef634102f 100644
--- a/include/tpu_mlir/Backend/BM168x/Param.h
+++ b/include/tpu_mlir/Backend/BM168x/Param.h
@@ -1084,6 +1084,11 @@ typedef struct swap_dim_spec {
   // int offset[MAX_SHAPE_DIMS];
 } swap_dim_spec_t;
 
+typedef struct swap_channel_param {
+  int order[3];
+  int shape_dim;
+} swap_channel_param_t;
+
 typedef struct {
   int tile_coeff[MAX_SHAPE_DIMS];
   int type;
diff --git a/lib/Dialect/Tpu/Interfaces/BM1684X/ScaleLut.cpp b/lib/Dialect/Tpu/Interfaces/BM1684X/ScaleLut.cpp
index 6682d7fa6..23cf7d012 100644
--- a/lib/Dialect/Tpu/Interfaces/BM1684X/ScaleLut.cpp
+++ b/lib/Dialect/Tpu/Interfaces/BM1684X/ScaleLut.cpp
@@ -11,7 +11,6 @@
 #include "tpu_mlir/Dialect/Tpu/IR/TpuOps.h"
 #include "tpu_mlir/Support/Module.h"
 
-
 // =========================================
 // GlobalGenInterface
 // =========================================
diff --git a/lib/Dialect/Tpu/Interfaces/BM1684X/SwapChannel.cpp b/lib/Dialect/Tpu/Interfaces/BM1684X/SwapChannel.cpp
index ea61984ae..24f09403b 100644
--- a/lib/Dialect/Tpu/Interfaces/BM1684X/SwapChannel.cpp
+++ b/lib/Dialect/Tpu/Interfaces/BM1684X/SwapChannel.cpp
@@ -9,18 +9,32 @@
 
 #include "tpu_mlir/Backend/BM168x/BM1684X.h"
 #include "tpu_mlir/Dialect/Tpu/IR/TpuOps.h"
-
 #include "tpu_mlir/Support/Module.h"
 
 using namespace tpu_mlir::backend;
 
+// =========================================
+// GlobalGenInterface
+// =========================================
 void tpu::SwapChannelOp::codegen_global_bm1684x() {
-  llvm_unreachable("Not Implemented");
+  auto op = getOperation();
+  auto channel_order = module::getI64Array(this->getChannelOrder());
+  auto input_spec = BM168x::get_input_spec(op);
+  auto output_spec = BM168x::get_output_spec(op);
+  swap_channel_param_t param = {0};
+  param.shape_dim = 4;
+  for (int i = 0; i < channel_order->size(); i++) {
+    param.order[i] = channel_order->at(i);
+  }
+  BM168x::call_global_func("backend_api_swap_channel_global", &param,
+                          sizeof(param), input_spec->data(),
+                          output_spec->data());
 }
 
 int64_t tpu::SwapChannelOp::dyn_codegen_global_bm1684x(void *buffer) {
   return 0;
 }
+
 // =========================================
 // LocalGenInterface
 // =========================================
diff --git a/third_party/nntoolchain/README.md b/third_party/nntoolchain/README.md
index 5dac6ee7e..4781c44cf 100644
--- a/third_party/nntoolchain/README.md
+++ b/third_party/nntoolchain/README.md
@@ -3,7 +3,7 @@
 #### 1684x/1686
 
 2023-03-01
-build from TPU1686     d04d5bee07f09783d0911f686b11ad6bd3228e7e
+build from TPU1686     532a31342dd80dbfa23e9a717bd43644a3e44790
 
 #### 1684
 2023-01-12:
diff --git a/third_party/nntoolchain/lib/libbackend_1684x.so b/third_party/nntoolchain/lib/libbackend_1684x.so
index 71dc8e099..4e2bc2163 100755
Binary files a/third_party/nntoolchain/lib/libbackend_1684x.so and b/third_party/nntoolchain/lib/libbackend_1684x.so differ
diff --git a/third_party/nntoolchain/lib/libcmodel_1684x.so b/third_party/nntoolchain/lib/libcmodel_1684x.so
index d4eaa9f70..4f4b4d819 100755
Binary files a/third_party/nntoolchain/lib/libcmodel_1684x.so and b/third_party/nntoolchain/lib/libcmodel_1684x.so differ