Update ops-related pbtxt files.

Change: 146318441

1 files changed, 5 insertions, 5 deletions

diff --git a/tensorflow/core/ops/ops.pbtxt b/tensorflow/core/ops/ops.pbtxt
index b5c670328a..805ba2c662 100644
--- a/tensorflow/core/ops/ops.pbtxt
+++ b/tensorflow/core/ops/ops.pbtxt
@@ -3128,7 +3128,7 @@ op {
   }
   output_arg {
     name: "output"
-    description: "4-D with shape `[batch, height, width, depth]`, where:\n\n      height = height_pad - crop_top - crop_bottom\n      width = width_pad - crop_left - crop_right\n\nThe attr `block_size` must be greater than one. It indicates the block size.\n\nSome examples:\n\n(1) For the following input of shape `[4, 1, 1, 1]` and block_size of 2:\n\n```prettyprint\n[[[[1]]], [[[2]]], [[[3]]], [[[4]]]]\n```\n\nThe output tensor has shape `[1, 2, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [2]], [[3], [4]]]]\n```\n\n(2) For the following input of shape `[4, 1, 1, 3]` and block_size of 2:\n\n```prettyprint\n[[[1, 2, 3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]]\n```\n\nThe output tensor has shape `[1, 2, 2, 3]` and value:\n\n```prettyprint\nx = [[[[1, 2, 3], [4, 5, 6]],\n      [[7, 8, 9], [10, 11, 12]]]]\n```\n\n(3) For the following input of shape `[4, 2, 2, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1], [3]], [[5], [7]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```\n\nThe output tensor has shape `[1, 4, 4, 1]` and value:\n\n```prettyprint\nx = [[[1],   [2],  [3],  [4]],\n     [[5],   [6],  [7],  [8]],\n     [[9],  [10], [11],  [12]],\n     [[13], [14], [15],  [16]]]\n```\n\n(4) For the following input of shape `[8, 1, 2, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1], [3]]], [[[9], [11]]], [[[2], [4]]], [[[10], [12]]],\n     [[[5], [7]]], [[[13], [15]]], [[[6], [8]]], [[[14], [16]]]]\n```\n\nThe output tensor has shape `[2, 2, 4, 1]` and value:\n\n```prettyprint\nx = [[[[1], [3]], [[5], [7]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```"
+    description: "4-D with shape `[batch, height, width, depth]`, where:\n\n      height = height_pad - crop_top - crop_bottom\n      width = width_pad - crop_left - crop_right\n\nThe attr `block_size` must be greater than one. It indicates the block size.\n\nSome examples:\n\n(1) For the following input of shape `[4, 1, 1, 1]` and block_size of 2:\n\n```prettyprint\n[[[[1]]], [[[2]]], [[[3]]], [[[4]]]]\n```\n\nThe output tensor has shape `[1, 2, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [2]], [[3], [4]]]]\n```\n\n(2) For the following input of shape `[4, 1, 1, 3]` and block_size of 2:\n\n```prettyprint\n[[[1, 2, 3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]]\n```\n\nThe output tensor has shape `[1, 2, 2, 3]` and value:\n\n```prettyprint\nx = [[[[1, 2, 3], [4, 5, 6]],\n      [[7, 8, 9], [10, 11, 12]]]]\n```\n\n(3) For the following input of shape `[4, 2, 2, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1], [3]], [[9], [11]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```\n\nThe output tensor has shape `[1, 4, 4, 1]` and value:\n\n```prettyprint\nx = [[[1],   [2],  [3],  [4]],\n     [[5],   [6],  [7],  [8]],\n     [[9],  [10], [11],  [12]],\n     [[13], [14], [15],  [16]]]\n```\n\n(4) For the following input of shape `[8, 1, 2, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1], [3]]], [[[9], [11]]], [[[2], [4]]], [[[10], [12]]],\n     [[[5], [7]]], [[[13], [15]]], [[[6], [8]]], [[[14], [16]]]]\n```\n\nThe output tensor has shape `[2, 2, 4, 1]` and value:\n\n```prettyprint\nx = [[[[1], [3]], [[5], [7]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```"
     type_attr: "T"
   }
   attr {
@@ -3171,7 +3171,7 @@ op {
   }
   input_arg {
     name: "crops"
-    description: "2-D with shape `[M, 2]`, all values must be >= 0.\n  `crops[i] = [crop_start, crop_end]` specifies the amount to crop from input\n  dimension `i + 1`, which corresponds to spatial dimension `i`.  It is\n  required that\n  `crop_start[i] + crop_end[i] <= block_shape[i] * input_shape[i + 1]`.\n\nThis operation is equivalent to the following steps:\n\n1. Reshape `input` to `reshaped` of shape:\n     [block_shape[0], ..., block_shape[M-1],\n      batch / prod(block_shape),\n      input_shape[1], ..., input_shape[N-1]]\n\n2. Permute dimensions of `reshaped` to produce `permuted` of shape\n     [batch / prod(block_shape),\n\n      input_shape[1], block_shape[0],\n      ...,\n      input_shape[M], block_shape[M-1],\n\n      input_shape[M+1], ..., input_shape[N-1]]\n\n3. Reshape `permuted` to produce `reshaped_permuted` of shape\n     [batch / prod(block_shape),\n\n      input_shape[1] * block_shape[0],\n      ...,\n      input_shape[M] * block_shape[M-1],\n\n      input_shape[M+1],\n      ...,\n      input_shape[N-1]]\n\n4. Crop the start and end of dimensions `[1, ..., M]` of\n   `reshaped_permuted` according to `crops` to produce the output of shape:\n     [batch / prod(block_shape),\n\n      input_shape[1] * block_shape[0] - crops[0,0] - crops[0,1],\n      ...,\n      input_shape[M] * block_shape[M-1] - crops[M-1,0] - crops[M-1,1],\n\n      input_shape[M+1], ..., input_shape[N-1]]\n\nSome examples:\n\n(1) For the following input of shape `[4, 1, 1, 1]`, `block_shape = [2, 2]`, and\n    `crops = [[0, 0], [0, 0]]`:\n\n```prettyprint\n[[[[1]]], [[[2]]], [[[3]]], [[[4]]]]\n```\n\nThe output tensor has shape `[1, 2, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [2]], [[3], [4]]]]\n```\n\n(2) For the following input of shape `[4, 1, 1, 3]`, `block_shape = [2, 2]`, and\n    `crops = [[0, 0], [0, 0]]`:\n\n```prettyprint\n[[[1, 2, 3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]]\n```\n\nThe output tensor has shape `[1, 2, 2, 3]` and value:\n\n```prettyprint\nx = [[[[1, 2, 3], [4, 5, 6]],\n      [[7, 8, 9], [10, 11, 12]]]]\n```\n\n(3) For the following input of shape `[4, 2, 2, 1]`, `block_shape = [2, 2]`, and\n    `crops = [[0, 0], [0, 0]]`:\n\n```prettyprint\nx = [[[[1], [3]], [[5], [7]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```\n\nThe output tensor has shape `[1, 4, 4, 1]` and value:\n\n```prettyprint\nx = [[[1],   [2],  [3],  [4]],\n     [[5],   [6],  [7],  [8]],\n     [[9],  [10], [11],  [12]],\n     [[13], [14], [15],  [16]]]\n```\n\n(4) For the following input of shape `[8, 1, 3, 1]`, `block_shape = [2, 2]`, and\n    `crops = [[0, 0], [2, 0]]`:\n\n```prettyprint\nx = [[[[0], [1], [3]]], [[[0], [9], [11]]],\n     [[[0], [2], [4]]], [[[0], [10], [12]]],\n     [[[0], [5], [7]]], [[[0], [13], [15]]],\n     [[[0], [6], [8]]], [[[0], [14], [16]]]]\n```\n\nThe output tensor has shape `[2, 2, 4, 1]` and value:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]]],\n     [[[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```"
+    description: "2-D with shape `[M, 2]`, all values must be >= 0.\n  `crops[i] = [crop_start, crop_end]` specifies the amount to crop from input\n  dimension `i + 1`, which corresponds to spatial dimension `i`.  It is\n  required that\n  `crop_start[i] + crop_end[i] <= block_shape[i] * input_shape[i + 1]`.\n\nThis operation is equivalent to the following steps:\n\n1. Reshape `input` to `reshaped` of shape:\n     [block_shape[0], ..., block_shape[M-1],\n      batch / prod(block_shape),\n      input_shape[1], ..., input_shape[N-1]]\n\n2. Permute dimensions of `reshaped` to produce `permuted` of shape\n     [batch / prod(block_shape),\n\n      input_shape[1], block_shape[0],\n      ...,\n      input_shape[M], block_shape[M-1],\n\n      input_shape[M+1], ..., input_shape[N-1]]\n\n3. Reshape `permuted` to produce `reshaped_permuted` of shape\n     [batch / prod(block_shape),\n\n      input_shape[1] * block_shape[0],\n      ...,\n      input_shape[M] * block_shape[M-1],\n\n      input_shape[M+1],\n      ...,\n      input_shape[N-1]]\n\n4. Crop the start and end of dimensions `[1, ..., M]` of\n   `reshaped_permuted` according to `crops` to produce the output of shape:\n     [batch / prod(block_shape),\n\n      input_shape[1] * block_shape[0] - crops[0,0] - crops[0,1],\n      ...,\n      input_shape[M] * block_shape[M-1] - crops[M-1,0] - crops[M-1,1],\n\n      input_shape[M+1], ..., input_shape[N-1]]\n\nSome examples:\n\n(1) For the following input of shape `[4, 1, 1, 1]`, `block_shape = [2, 2]`, and\n    `crops = [[0, 0], [0, 0]]`:\n\n```prettyprint\n[[[[1]]], [[[2]]], [[[3]]], [[[4]]]]\n```\n\nThe output tensor has shape `[1, 2, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [2]], [[3], [4]]]]\n```\n\n(2) For the following input of shape `[4, 1, 1, 3]`, `block_shape = [2, 2]`, and\n    `crops = [[0, 0], [0, 0]]`:\n\n```prettyprint\n[[[1, 2, 3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]]\n```\n\nThe output tensor has shape `[1, 2, 2, 3]` and value:\n\n```prettyprint\nx = [[[[1, 2, 3], [4, 5, 6]],\n      [[7, 8, 9], [10, 11, 12]]]]\n```\n\n(3) For the following input of shape `[4, 2, 2, 1]`, `block_shape = [2, 2]`, and\n    `crops = [[0, 0], [0, 0]]`:\n\n```prettyprint\nx = [[[[1], [3]], [[9], [11]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```\n\nThe output tensor has shape `[1, 4, 4, 1]` and value:\n\n```prettyprint\nx = [[[1],   [2],  [3],  [4]],\n     [[5],   [6],  [7],  [8]],\n     [[9],  [10], [11],  [12]],\n     [[13], [14], [15],  [16]]]\n```\n\n(4) For the following input of shape `[8, 1, 3, 1]`, `block_shape = [2, 2]`, and\n    `crops = [[0, 0], [2, 0]]`:\n\n```prettyprint\nx = [[[[0], [1], [3]]], [[[0], [9], [11]]],\n     [[[0], [2], [4]]], [[[0], [10], [12]]],\n     [[[0], [5], [7]]], [[[0], [13], [15]]],\n     [[[0], [6], [8]]], [[[0], [14], [16]]]]\n```\n\nThe output tensor has shape `[2, 2, 4, 1]` and value:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]]],\n     [[[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```"
     type_attr: "Tcrops"
   }
   output_arg {
@@ -3613,7 +3613,7 @@ op {
     description: "If True, merge repeated classes in output."
   }
   summary: "Performs greedy decoding on the logits given in inputs."
-  description: "A note about the attribute merge_repeated: if enabled, when\nconsecutive logits\' maximum indices are the same, only the first of\nthese is emitted.  Labeling the blank \'*\', the sequence \"A B B * B B\"\nbecomes \"A B\" if merge_repeated = True and \"A B B B B\" if\nmerge_repeated = False.\n\nRegardless of the value of merge_repeated, if the maximum index of a given\ntime and batch corresponds to the blank, index `(num_classes - 1)`, no new\nelement is emitted."
+  description: "A note about the attribute merge_repeated: if enabled, when\nconsecutive logits\' maximum indices are the same, only the first of\nthese is emitted.  Labeling the blank \'*\', the sequence \"A B B * B B\"\nbecomes \"A B B\" if merge_repeated = True and \"A B B B B\" if\nmerge_repeated = False.\n\nRegardless of the value of merge_repeated, if the maximum index of a given\ntime and batch corresponds to the blank, index `(num_classes - 1)`, no new\nelement is emitted."
 }
 op {
   name: "CTCLoss"
@@ -19780,7 +19780,7 @@ op {
   }
   input_arg {
     name: "paddings"
-    description: "2-D tensor of non-negative integers with shape `[2, 2]`. It specifies\n  the padding of the input with zeros across the spatial dimensions as follows:\n\n      paddings = [[pad_top, pad_bottom], [pad_left, pad_right]]\n\n  The effective spatial dimensions of the zero-padded input tensor will be:\n\n      height_pad = pad_top + height + pad_bottom\n      width_pad = pad_left + width + pad_right\n\nThe attr `block_size` must be greater than one. It indicates the block size.\n\n  * Non-overlapping blocks of size `block_size x block size` in the height and\n    width dimensions are rearranged into the batch dimension at each location.\n  * The batch of the output tensor is `batch * block_size * block_size`.\n  * Both height_pad and width_pad must be divisible by block_size.\n\nThe shape of the output will be:\n\n    [batch*block_size*block_size, height_pad/block_size, width_pad/block_size,\n     depth]\n\nSome examples:\n\n(1) For the following input of shape `[1, 2, 2, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1], [2]], [[3], [4]]]]\n```\n\nThe output tensor has shape `[4, 1, 1, 1]` and value:\n\n```prettyprint\n[[[[1]]], [[[2]]], [[[3]]], [[[4]]]]\n```\n\n(2) For the following input of shape `[1, 2, 2, 3]` and block_size of 2:\n\n```prettyprint\nx = [[[[1, 2, 3], [4, 5, 6]],\n      [[7, 8, 9], [10, 11, 12]]]]\n```\n\nThe output tensor has shape `[4, 1, 1, 3]` and value:\n\n```prettyprint\n[[[1, 2, 3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]]\n```\n\n(3) For the following input of shape `[1, 4, 4, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]],\n      [[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```\n\nThe output tensor has shape `[4, 2, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [3]], [[5], [7]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```\n\n(4) For the following input of shape `[2, 2, 4, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]]],\n     [[[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```\n\nThe output tensor has shape `[8, 1, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [3]]], [[[9], [11]]], [[[2], [4]]], [[[10], [12]]],\n     [[[5], [7]]], [[[13], [15]]], [[[6], [8]]], [[[14], [16]]]]\n```\n\nAmong others, this operation is useful for reducing atrous convolution into\nregular convolution."
+    description: "2-D tensor of non-negative integers with shape `[2, 2]`. It specifies\n  the padding of the input with zeros across the spatial dimensions as follows:\n\n      paddings = [[pad_top, pad_bottom], [pad_left, pad_right]]\n\n  The effective spatial dimensions of the zero-padded input tensor will be:\n\n      height_pad = pad_top + height + pad_bottom\n      width_pad = pad_left + width + pad_right\n\nThe attr `block_size` must be greater than one. It indicates the block size.\n\n  * Non-overlapping blocks of size `block_size x block size` in the height and\n    width dimensions are rearranged into the batch dimension at each location.\n  * The batch of the output tensor is `batch * block_size * block_size`.\n  * Both height_pad and width_pad must be divisible by block_size.\n\nThe shape of the output will be:\n\n    [batch*block_size*block_size, height_pad/block_size, width_pad/block_size,\n     depth]\n\nSome examples:\n\n(1) For the following input of shape `[1, 2, 2, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1], [2]], [[3], [4]]]]\n```\n\nThe output tensor has shape `[4, 1, 1, 1]` and value:\n\n```prettyprint\n[[[[1]]], [[[2]]], [[[3]]], [[[4]]]]\n```\n\n(2) For the following input of shape `[1, 2, 2, 3]` and block_size of 2:\n\n```prettyprint\nx = [[[[1, 2, 3], [4, 5, 6]],\n      [[7, 8, 9], [10, 11, 12]]]]\n```\n\nThe output tensor has shape `[4, 1, 1, 3]` and value:\n\n```prettyprint\n[[[1, 2, 3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]]\n```\n\n(3) For the following input of shape `[1, 4, 4, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]],\n      [[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```\n\nThe output tensor has shape `[4, 2, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [3]], [[9], [11]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```\n\n(4) For the following input of shape `[2, 2, 4, 1]` and block_size of 2:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]]],\n     [[[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```\n\nThe output tensor has shape `[8, 1, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [3]]], [[[9], [11]]], [[[2], [4]]], [[[10], [12]]],\n     [[[5], [7]]], [[[13], [15]]], [[[6], [8]]], [[[14], [16]]]]\n```\n\nAmong others, this operation is useful for reducing atrous convolution into\nregular convolution."
     type_attr: "Tpaddings"
   }
   output_arg {
@@ -19827,7 +19827,7 @@ op {
   }
   input_arg {
     name: "paddings"
-    description: "2-D with shape `[M, 2]`, all values must be >= 0.\n  `paddings[i] = [pad_start, pad_end]` specifies the padding for input dimension\n  `i + 1`, which corresponds to spatial dimension `i`.  It is required that\n  `block_shape[i]` divides `input_shape[i + 1] + pad_start + pad_end`.\n\nThis operation is equivalent to the following steps:\n\n1. Zero-pad the start and end of dimensions `[1, ..., M]` of the\n   input according to `paddings` to produce `padded` of shape `padded_shape`.\n\n2. Reshape `padded` to `reshaped_padded` of shape:\n\n     [batch] +\n     [padded_shape[1] / block_shape[0],\n       block_shape[0],\n      ...,\n      padded_shape[M] / block_shape[M-1],\n      block_shape[M-1]] +\n     remaining_shape\n\n3. Permute dimensions of `reshaped_padded` to produce\n   `permuted_reshaped_padded` of shape:\n\n     block_shape +\n     [batch] +\n     [padded_shape[1] / block_shape[0],\n      ...,\n      padded_shape[M] / block_shape[M-1]] +\n     remaining_shape\n\n4. Reshape `permuted_reshaped_padded` to flatten `block_shape` into the batch\n   dimension, producing an output tensor of shape:\n\n     [batch * prod(block_shape)] +\n     [padded_shape[1] / block_shape[0],\n      ...,\n      padded_shape[M] / block_shape[M-1]] +\n     remaining_shape\n\nSome examples:\n\n(1) For the following input of shape `[1, 2, 2, 1]`, `block_shape = [2, 2]`, and\n    `paddings = [[0, 0], [0, 0]]`:\n\n```prettyprint\nx = [[[[1], [2]], [[3], [4]]]]\n```\n\nThe output tensor has shape `[4, 1, 1, 1]` and value:\n\n```prettyprint\n[[[[1]]], [[[2]]], [[[3]]], [[[4]]]]\n```\n\n(2) For the following input of shape `[1, 2, 2, 3]`, `block_shape = [2, 2]`, and\n    `paddings = [[0, 0], [0, 0]]`:\n\n```prettyprint\nx = [[[[1, 2, 3], [4, 5, 6]],\n      [[7, 8, 9], [10, 11, 12]]]]\n```\n\nThe output tensor has shape `[4, 1, 1, 3]` and value:\n\n```prettyprint\n[[[1, 2, 3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]]\n```\n\n(3) For the following input of shape `[1, 4, 4, 1]`, `block_shape = [2, 2]`, and\n    `paddings = [[0, 0], [0, 0]]`:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]],\n      [[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```\n\nThe output tensor has shape `[4, 2, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [3]], [[5], [7]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```\n\n(4) For the following input of shape `[2, 2, 4, 1]`, block_shape = `[2, 2]`, and\n    paddings = `[[0, 0], [2, 0]]`:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]]],\n     [[[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```\n\nThe output tensor has shape `[8, 1, 3, 1]` and value:\n\n```prettyprint\nx = [[[[0], [1], [3]]], [[[0], [9], [11]]],\n     [[[0], [2], [4]]], [[[0], [10], [12]]],\n     [[[0], [5], [7]]], [[[0], [13], [15]]],\n     [[[0], [6], [8]]], [[[0], [14], [16]]]]\n```\n\nAmong others, this operation is useful for reducing atrous convolution into\nregular convolution."
+    description: "2-D with shape `[M, 2]`, all values must be >= 0.\n  `paddings[i] = [pad_start, pad_end]` specifies the padding for input dimension\n  `i + 1`, which corresponds to spatial dimension `i`.  It is required that\n  `block_shape[i]` divides `input_shape[i + 1] + pad_start + pad_end`.\n\nThis operation is equivalent to the following steps:\n\n1. Zero-pad the start and end of dimensions `[1, ..., M]` of the\n   input according to `paddings` to produce `padded` of shape `padded_shape`.\n\n2. Reshape `padded` to `reshaped_padded` of shape:\n\n     [batch] +\n     [padded_shape[1] / block_shape[0],\n       block_shape[0],\n      ...,\n      padded_shape[M] / block_shape[M-1],\n      block_shape[M-1]] +\n     remaining_shape\n\n3. Permute dimensions of `reshaped_padded` to produce\n   `permuted_reshaped_padded` of shape:\n\n     block_shape +\n     [batch] +\n     [padded_shape[1] / block_shape[0],\n      ...,\n      padded_shape[M] / block_shape[M-1]] +\n     remaining_shape\n\n4. Reshape `permuted_reshaped_padded` to flatten `block_shape` into the batch\n   dimension, producing an output tensor of shape:\n\n     [batch * prod(block_shape)] +\n     [padded_shape[1] / block_shape[0],\n      ...,\n      padded_shape[M] / block_shape[M-1]] +\n     remaining_shape\n\nSome examples:\n\n(1) For the following input of shape `[1, 2, 2, 1]`, `block_shape = [2, 2]`, and\n    `paddings = [[0, 0], [0, 0]]`:\n\n```prettyprint\nx = [[[[1], [2]], [[3], [4]]]]\n```\n\nThe output tensor has shape `[4, 1, 1, 1]` and value:\n\n```prettyprint\n[[[[1]]], [[[2]]], [[[3]]], [[[4]]]]\n```\n\n(2) For the following input of shape `[1, 2, 2, 3]`, `block_shape = [2, 2]`, and\n    `paddings = [[0, 0], [0, 0]]`:\n\n```prettyprint\nx = [[[[1, 2, 3], [4, 5, 6]],\n      [[7, 8, 9], [10, 11, 12]]]]\n```\n\nThe output tensor has shape `[4, 1, 1, 3]` and value:\n\n```prettyprint\n[[[1, 2, 3]], [[4, 5, 6]], [[7, 8, 9]], [[10, 11, 12]]]\n```\n\n(3) For the following input of shape `[1, 4, 4, 1]`, `block_shape = [2, 2]`, and\n    `paddings = [[0, 0], [0, 0]]`:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]],\n      [[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```\n\nThe output tensor has shape `[4, 2, 2, 1]` and value:\n\n```prettyprint\nx = [[[[1], [3]], [[9], [11]]],\n     [[[2], [4]], [[10], [12]]],\n     [[[5], [7]], [[13], [15]]],\n     [[[6], [8]], [[14], [16]]]]\n```\n\n(4) For the following input of shape `[2, 2, 4, 1]`, block_shape = `[2, 2]`, and\n    paddings = `[[0, 0], [2, 0]]`:\n\n```prettyprint\nx = [[[[1],   [2],  [3],  [4]],\n      [[5],   [6],  [7],  [8]]],\n     [[[9],  [10], [11],  [12]],\n      [[13], [14], [15],  [16]]]]\n```\n\nThe output tensor has shape `[8, 1, 3, 1]` and value:\n\n```prettyprint\nx = [[[[0], [1], [3]]], [[[0], [9], [11]]],\n     [[[0], [2], [4]]], [[[0], [10], [12]]],\n     [[[0], [5], [7]]], [[[0], [13], [15]]],\n     [[[0], [6], [8]]], [[[0], [14], [16]]]]\n```\n\nAmong others, this operation is useful for reducing atrous convolution into\nregular convolution."
     type_attr: "Tpaddings"
   }
   output_arg {