!6378 nn.MatMul support broadcast.

Merge pull request !6378 from liuxiao93/Broadcast-nn.MatMul-Ascend

mindspore/nn/layer/math.py

@@ -25,7 +25,7 @@ from ..._checkparam import Validator as validator
 from ..._checkparam import Rel
 
 
-__all__ = ['ReduceLogSumExp', 'Range', 'LinSpace', 'LGamma']
+__all__ = ['ReduceLogSumExp', 'Range', 'LinSpace', 'LGamma', 'MatMul']
 
 
 class ReduceLogSumExp(Cell):
@@ -302,3 +302,106 @@ class LGamma(Cell):
         result = self.select(need_to_reflect, reflection, log_y)
 
         return self.select(self.isfinite(input_x), result, infinity)
+
+
+@constexpr
+def get_broadcast_matmul_shape(x_shape, y_shape):
+    """get broadcast_matmul shape"""
+    if (len(x_shape) < 2) or (len(y_shape) < 2):
+        raise ValueError('For matmul, rank of x1 and x2 should be equal to or greater than 2, '
+                         + f'but got {x_shape} and {y_shape}.')
+    x_shape_batch = x_shape[:-2]
+    y_shape_batch = y_shape[:-2]
+    if x_shape_batch == y_shape_batch:
+        return x_shape, y_shape
+    x_len = len(x_shape)
+    y_len = len(y_shape)
+    length = x_len if x_len < y_len else y_len
+    broadcast_shape_back = []
+    for i in range(-length, -2):
+        if x_shape[i] == 1:
+            broadcast_shape_back.append(y_shape[i])
+        elif y_shape[i] == 1:
+            broadcast_shape_back.append(x_shape[i])
+        elif x_shape[i] == y_shape[i]:
+            broadcast_shape_back.append(x_shape[i])
+        else:
+            raise ValueError(f"For MatMul, the x1_shape {x_shape} and x2_shape {y_shape} can not broadcast.")
+
+    broadcast_shape_front = y_shape[0: y_len - length] if length == x_len else x_shape[0: x_len - length]
+    x_broadcast_shape = broadcast_shape_front + tuple(broadcast_shape_back) + x_shape[-2:]
+    y_broadcast_shape = broadcast_shape_front + tuple(broadcast_shape_back) + y_shape[-2:]
+    return x_broadcast_shape, y_broadcast_shape
+
+
+@constexpr
+def check_col_row_equal(x1_shape, x2_shape, transpose_x1, transpose_x2):
+    """check col and row equal"""
+    x1_last = x1_shape[-2:]
+    x2_last = x2_shape[-2:]
+    x1_col = x1_last[not transpose_x1]  # x1_col = x1_last[1] if (not transpose_a) else x1_last[0]
+    x2_row = x2_last[transpose_x2]  # x2_row = x2_last[0] if (not transpose_b) else x2_last[1]
+    if x1_col != x2_row:
+        raise ValueError('The column of matrix dimensions of x1 should be equal to '
+                         + f'the row of matrix dimensions of x2, but got {x1_col} and {x2_row}.')
+
+
+class MatMul(Cell):
+    """
+    Multiplies matrix `x1` by matrix `x2`.
+
+    The rank of input tensors must be not less than `2`. The none-matrix dimensions(batch) of inputs
+    will be broadcasted and must be broadcastable.
+
+    Args:
+        transpose_x1 (bool): If True, `a` is transposed before multiplication. Default: False.
+        transpose_x2 (bool): If True, `b` is transposed before multiplication. Default: False.
+
+    Inputs:
+        - **input_x1** (Tensor) - The first tensor to be multiplied. The shape of the tensor is :math:`(*A, N, C)`,
+          where :math:`*A` represents the batch size of `x1` which can be multidimensional.
+          If `transpose_a` is True, its shape should be :math:`(*A, N, C)` after transposing.
+        - **input_x2** (Tensor) - The second tensor to be multiplied. The shape of the tensor is :math:`(*B, C, M)`,
+          where :math:`*B` represents the batch size of `x2` which can be multidimensional.
+          If `transpose_b` is True, its shape should be :math:`(*B, C, M)` after transposing.
+
+    Outputs:
+        Tensor, the shape of the output tensor is :math:`(*L, N, M)`. :math:`*L` is the batch size after broadcasting.
+
+    Examples:
+        >>> net = nn.MatMul()
+        >>> input_x1 = Tensor(np.ones(shape=[3, 2, 3]), mindspore.float32)
+        >>> input_x2 = Tensor(np.ones(shape=[3, 4]), mindspore.float32)
+        >>> output = net(input_x1, input_x2)
+        >>> print(output.shape)
+        (3, 2, 4)
+    """
+
+    def __init__(self, transpose_x1=False, transpose_x2=False):
+        super(MatMul, self).__init__()
+
+        validator.check_value_type('transpose_x1', transpose_x1, [bool], self.cls_name)
+        validator.check_value_type('transpose_x2', transpose_x2, [bool], self.cls_name)
+        self.transpose_x1 = transpose_x1
+        self.transpose_x2 = transpose_x2
+        self.shape_op = P.Shape()
+        self.matmul_op = P.MatMul(self.transpose_x1, self.transpose_x2)
+        self.batch_matmul_op = P.BatchMatMul(self.transpose_x1, self.transpose_x2)
+
+    def construct(self, x1, x2):
+        x1_shape = self.shape_op(x1)
+        x2_shape = self.shape_op(x2)
+        check_col_row_equal(x1_shape, x2_shape, self.transpose_x1, self.transpose_x2)
+
+        x1_broadcast_shape, x2_broadcast_shape = get_broadcast_matmul_shape(x1_shape, x2_shape)
+        x1_broadcast_to = P.BroadcastTo(x1_broadcast_shape)
+        x2_broadcast_to = P.BroadcastTo(x2_broadcast_shape)
+        if x1_broadcast_shape != x1_shape:
+            x1 = x1_broadcast_to(x1)
+        if x2_broadcast_shape != x2_shape:
+            x2 = x2_broadcast_to(x2)
+        if len(x1_broadcast_shape) == 2:
+            matmul_broadcast = self.matmul_op(x1, x2)
+        else:
+            matmul_broadcast = self.batch_matmul_op(x1, x2)
+        return matmul_broadcast

tests/st/ops/ascend/test_nn_matmul.py

@@ -0,0 +1,56 @@
+import numpy as np
+
+import mindspore.context as context
+import mindspore.nn as nn
+from mindspore import Tensor
+
+context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
+
+
+class Net(nn.Cell):
+    def __init__(self, transpose_x1, transpose_x2):
+        super(Net, self).__init__()
+        self.matmul = nn.MatMul(transpose_x1, transpose_x2)
+
+    def construct(self, x1, x2):
+        return self.matmul(x1, x2)
+
+
+def test_x1_2D_x2_3D():
+    x1 = np.random.randn(16, 64).astype(np.float32)
+    x2 = np.random.randn(32, 64, 20).astype(np.float32)
+    transpose_x1 = False
+    transpose_x2 = False
+    net = Net(transpose_x1, transpose_x2)
+    output = net(Tensor(x1), Tensor(x2))
+    assert output.shape == (32, 16, 20)
+
+
+def test_x1_4D_x2_3D_transpose_x2_True():
+    x1 = np.random.randn(3, 2, 3, 4).astype(np.float32)
+    x2 = np.random.randn(1, 5, 4).astype(np.float32)
+    transpose_x1 = False
+    transpose_x2 = True
+    net = Net(transpose_x1, transpose_x2)
+    output = net(Tensor(x1), Tensor(x2))
+    assert output.shape == (3, 2, 3, 5)
+
+
+def test_x1_3D_transpose_x1_True_x2_2D():
+    x1 = np.random.randn(2, 3, 4).astype(np.float32)
+    x2 = np.random.randn(3, 4).astype(np.float32)
+    transpose_x1 = True
+    transpose_x2 = False
+    net = Net(transpose_x1, transpose_x2)
+    output = net(Tensor(x1), Tensor(x2))
+    assert output.shape == (2, 4, 4)
+
+
+def test_x1_3D_transpose_x1_True_x2_3D_transpose_x2_True():
+    x1 = np.random.randn(2, 5, 6).astype(np.float32)
+    x2 = np.random.randn(2, 4, 5).astype(np.float32)
+    transpose_x1 = True
+    transpose_x2 = True
+    net = Net(transpose_x1, transpose_x2)
+    output = net(Tensor(x1), Tensor(x2))
+    assert output.shape == (2, 6, 4)