Specify the CONV behavior in the presence of groups more precisely (#3763)

* specify behaviour with groups more precisely
Signed-off-by: NAlexandre Eichenberger <alexe@us.ibm.com>

* update before regen the files
Signed-off-by: NAlexandre Eichenberger <alexe@us.ibm.com>

* added the auto-generated files
Signed-off-by: NAlexandre Eichenberger <alexe@us.ibm.com>

* updates of generated docs with latest onnx installed
Signed-off-by: NAlexandre Eichenberger <alexe@us.ibm.com>
Co-authored-by: NG. Ramalingam <grama@microsoft.com>

docs/Changelog.md

@@ -10716,7 +10716,7 @@ This version of the operator has been available since version 11 of the default
 <dt><tt>X</tt> (differentiable) : T</dt>
 <dd>Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].</dd>
 <dt><tt>W</tt> (differentiable) : T</dt>
-<dd>The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL. </dd>
+<dd>The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.</dd>
 <dt><tt>B</tt> (optional, differentiable) : T</dt>
 <dd>Optional 1D bias to be added to the convolution, has size of M.</dd>
 </dl>

docs/Operators.md

@@ -3454,7 +3454,7 @@ Other versions of this operator: <a href="Changelog.md#Conv-1">1</a>
 <dt><tt>X</tt> (differentiable) : T</dt>
 <dd>Input data tensor from previous layer; has size (N x C x H x W), where N is the batch size, C is the number of channels, and H and W are the height and width. Note that this is for the 2D image. Otherwise the size is (N x C x D1 x D2 ... x Dn). Optionally, if dimension denotation is in effect, the operation expects input data tensor to arrive with the dimension denotation of [DATA_BATCH, DATA_CHANNEL, DATA_FEATURE, DATA_FEATURE ...].</dd>
 <dt><tt>W</tt> (differentiable) : T</dt>
-<dd>The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. X.shape[1] == (W.shape[1] * group) == C (assuming zero based indices for the shape array). Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL. </dd>
+<dd>The weight tensor that will be used in the convolutions; has size (M x C/group x kH x kW), where C is the number of channels, and kH and kW are the height and width of the kernel, and M is the number of feature maps. For more than 2 dimensions, the kernel shape will be (M x C/group x k1 x k2 x ... x kn), where (k1 x k2 x ... kn) is the dimension of the kernel. Optionally, if dimension denotation is in effect, the operation expects the weight tensor to arrive with the dimension denotation of [FILTER_OUT_CHANNEL, FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. Assuming zero based indices for the shape array, X.shape[1] == (W.shape[1] * group) == C and W.shape[0] mod G == 0. Or in other words FILTER_IN_CHANNEL multiplied by the number of groups should be equal to DATA_CHANNEL and the number of feature maps M should be a multiple of the number of groups G.</dd>
 <dt><tt>B</tt> (optional, differentiable) : T</dt>
 <dd>Optional 1D bias to be added to the convolution, has size of M.</dd>
 </dl>

onnx/defs/nn/defs.cc

@@ -796,9 +796,13 @@ computes the output.)DOC";
         "the operation expects the weight tensor to arrive "
         "with the dimension denotation of [FILTER_OUT_CHANNEL, "
         "FILTER_IN_CHANNEL, FILTER_SPATIAL, FILTER_SPATIAL ...]. "
-        "X.shape[1] == (W.shape[1] * group) == C "
-        "(assuming zero based indices for the shape array). "
-        "Or in other words FILTER_IN_CHANNEL should be equal to DATA_CHANNEL. ",
+        "Assuming zero based indices for the shape array, "
+        "X.shape[1] == (W.shape[1] * group) == C and "
+        "W.shape[0] mod G == 0. Or in other words "
+        "FILTER_IN_CHANNEL multiplied by the number of groups "
+        "should be equal to DATA_CHANNEL and the number of "
+        "feature maps M should be a multiple of the number of "
+        "groups G.",
         "T",
         OpSchema::Single,
         true,