Merge pull request #48082 from njeffrie:f1_depthwise

PiperOrigin-RevId: 365150386
Change-Id: I0cca1c34d2e600cea51ef45b37e5cf808891541a

tensorflow/lite/micro/tools/make/ext_libs/xtensa_download.sh

@@ -65,15 +65,15 @@ else
   fi
 
   unzip -qo /tmp/${TMP_ZIP_ARCHIVE_NAME} -d ${DOWNLOADS_DIR} >&2
-  # TODO(b/183232978): remove these checks when xa_nnlib is updated.
-  sed -i '/XA_NNLIB_ARG_CHK_COND((y_stride > kernel_height), -1);/d' \
-    ${DOWNLOADS_DIR}/xa_nnlib_hifi4/algo/kernels/cnn/hifi4/*.c
-  sed -i '/XA_NNLIB_ARG_CHK_COND((x_stride > kernel_width), -1);/d' \
-    ${DOWNLOADS_DIR}/xa_nnlib_hifi4/algo/kernels/cnn/hifi4/*.c
-  sed -i '/XA_NNLIB_CHK_COND((x_stride <= 0 || x_stride > kernel_width), -1);/d' \
-    ${DOWNLOADS_DIR}/xa_nnlib_hifi4/algo/kernels/cnn/hifi4/*.c
-  sed -i '/XA_NNLIB_CHK_COND((y_stride <= 0 || y_stride > kernel_width), -1);/d' \
-    ${DOWNLOADS_DIR}/xa_nnlib_hifi4/algo/kernels/cnn/hifi4/*.c
+
+  pushd ${DOWNLOADS_DIR}/xa_nnlib_hifi4/
+  git init .
+  git config user.email "tflm@google.com"
+  git config user.name "TensorflowLite Micro"
+  git add *
+  git commit -a -m "Commit for a temporary repository." > /dev/null
+  git apply ../../ext_libs/xtensa_patch.patch
+  popd
 
 fi
 

tensorflow/lite/micro/tools/make/ext_libs/xtensa_patch.patch

@@ -0,0 +1,45 @@
+diff --git a/algo/kernels/cnn/hifi4/xa_nn_conv2d_depthwise.c b/algo/kernels/cnn/hifi4/xa_nn_conv2d_depthwise.c
+index 3e29856..e550ebf 100644
+--- a/algo/kernels/cnn/hifi4/xa_nn_conv2d_depthwise.c
++++ b/algo/kernels/cnn/hifi4/xa_nn_conv2d_depthwise.c
+@@ -249,7 +249,6 @@ WORD32 xa_nn_conv2d_depthwise_getsize
+     XA_NNLIB_CHK_COND((kernel_height <= 0), -1);
+     XA_NNLIB_CHK_COND((kernel_width <= 0), -1);
+     XA_NNLIB_CHK_COND((channels_multiplier <= 0), -1);
+-    XA_NNLIB_CHK_COND((x_stride <= 0 || x_stride > kernel_width), -1);
+     XA_NNLIB_CHK_COND((y_stride <= 0 || y_stride > kernel_height), -1);
+     XA_NNLIB_CHK_COND((x_padding < 0), -1);
+     XA_NNLIB_CHK_COND((y_padding < 0), -1);
+diff --git a/algo/kernels/cnn/hifi4/xa_nn_conv2d_depthwise_sym8sxasym8s.c b/algo/kernels/cnn/hifi4/xa_nn_conv2d_depthwise_sym8sxasym8s.c
+index e719da1..5b7390f 100644
+--- a/algo/kernels/cnn/hifi4/xa_nn_conv2d_depthwise_sym8sxasym8s.c
++++ b/algo/kernels/cnn/hifi4/xa_nn_conv2d_depthwise_sym8sxasym8s.c
+@@ -659,7 +659,6 @@ WORD32 xa_nn_conv2d_depthwise_per_chan_sym8sxasym8s
+     XA_NNLIB_ARG_CHK_COND((input_channels <= 0), -1);
+     XA_NNLIB_ARG_CHK_COND((kernel_height <= 0 || kernel_width <= 0), -1);
+     XA_NNLIB_ARG_CHK_COND((kernel_height > input_height), -1);
+-    XA_NNLIB_ARG_CHK_COND((kernel_width > input_width), -1);
+     XA_NNLIB_ARG_CHK_COND((channels_multiplier <= 0), -1);
+     XA_NNLIB_ARG_CHK_COND((y_stride <= 0 || x_stride <= 0), -1);
+     XA_NNLIB_ARG_CHK_COND((y_padding < 0 || x_padding < 0), -1);
+@@ -671,8 +670,6 @@ WORD32 xa_nn_conv2d_depthwise_per_chan_sym8sxasym8s
+     XA_NNLIB_ARG_CHK_COND((inp_data_format != 0 && inp_data_format != 1), -1);
+     XA_NNLIB_ARG_CHK_COND((out_data_format != 0), -1);
+     /* Implementation dependent checks */
+-    XA_NNLIB_ARG_CHK_COND((y_stride > kernel_height), -1);
+-    XA_NNLIB_ARG_CHK_COND((x_stride > kernel_width), -1);
+
+     if(inp_data_format == 0)
+     {
+diff --git a/algo/kernels/cnn/hifi4/xa_nn_conv2d_std_sym8sxasym8s.c b/algo/kernels/cnn/hifi4/xa_nn_conv2d_std_sym8sxasym8s.c
+index b16b9fc..38e69d3 100644
+--- a/algo/kernels/cnn/hifi4/xa_nn_conv2d_std_sym8sxasym8s.c
++++ b/algo/kernels/cnn/hifi4/xa_nn_conv2d_std_sym8sxasym8s.c
+@@ -198,7 +198,6 @@ WORD32 xa_nn_conv2d_std_per_chan_sym8sxasym8s(
+   XA_NNLIB_ARG_CHK_COND((input_channels <= 0), -1);
+   XA_NNLIB_ARG_CHK_COND((kernel_height <= 0 || kernel_width <= 0), -1);
+   XA_NNLIB_ARG_CHK_COND((kernel_height > input_height), -1);
+-  XA_NNLIB_ARG_CHK_COND((kernel_width > input_width), -1);
+   XA_NNLIB_ARG_CHK_COND((out_channels <= 0), -1);
+   XA_NNLIB_ARG_CHK_COND((y_stride <= 0 || x_stride <= 0), -1);
+   XA_NNLIB_ARG_CHK_COND((y_padding < 0 || x_padding < 0), -1);