matmul (#2084)

* matmul

* np.allclose

oneflow/python/keras/maths.py

@@ -8,38 +8,6 @@ import oneflow.core.register.logical_blob_id_pb2 as logical_blob_id_util
 
 from oneflow.python.oneflow_export import oneflow_export
 
-@oneflow_export('keras.maths.matmul')
-def matmul( a,
-            b,
-            transpose_a=False,
-            transpose_b=False,
-            adjoint_a=False,
-            adjoint_b=False,
-            a_is_sparse=False,
-            b_is_sparse=False,
-            name=None):
-
-    assert adjoint_a == False
-    assert adjoint_b == False
-    assert a_is_sparse == False
-    assert b_is_sparse == False
-
-    op_conf = op_conf_util.OperatorConf()
-    if name is None:
-        op_conf.name = id_util.UniqueStr('Matmul_')
-    else:
-        op_conf.name = name
-    op_conf.matmul_conf.a = a.logical_blob_name
-    op_conf.matmul_conf.b = b.logical_blob_name
-    op_conf.matmul_conf.transpose_a = transpose_a
-    op_conf.matmul_conf.transpose_b = transpose_b
-    op_conf.matmul_conf.out = "out"
-    compile_context.CurJobAddOp(op_conf)
-    lbi = logical_blob_id_util.LogicalBlobId()
-    lbi.op_name = op_conf.name
-    lbi.blob_name = "out"
-    return remote_blob_util.RemoteBlob(lbi)
-
 
 @oneflow_export('keras.maths.add')
 def add(x,

oneflow/python/ops/linalg.py

+from __future__ import absolute_import
+
+import oneflow.python.framework.compile_context as compile_context
+import oneflow.python.framework.remote_blob as remote_blob_util
+import oneflow.python.framework.id_util as id_util
+import oneflow.core.operator.op_conf_pb2 as op_conf_util
+import oneflow.core.register.logical_blob_id_pb2 as logical_blob_id_util
+from oneflow.python.oneflow_export import oneflow_export
+
+# TODO: Support export multiple interfaces, eg. @oneflow_export("matmul", "linalg.matmul")
+@oneflow_export("matmul")
+def matmul(a, b, transpose_a=False, transpose_b=False, name=None):
+    op_conf = op_conf_util.OperatorConf()
+    setattr(op_conf, "name", name if name is not None else id_util.UniqueStr("Matmul_"))
+    setattr(op_conf.matmul_conf, "a", a.logical_blob_name)
+    setattr(op_conf.matmul_conf, "b", b.logical_blob_name)
+    setattr(op_conf.matmul_conf, "transpose_a", transpose_a)
+    setattr(op_conf.matmul_conf, "transpose_b", transpose_b)
+    setattr(op_conf.matmul_conf, "out", "out")
+    compile_context.CurJobAddOp(op_conf)
+    out_lbi = logical_blob_id_util.LogicalBlobId()
+    setattr(out_lbi, "op_name", op_conf.name)
+    setattr(out_lbi, "blob_name", "out")
+    return remote_blob_util.RemoteBlob(out_lbi)

oneflow/python/test/matmul.py

-import oneflow as flow
-import numpy as np
-import torch
-
-config = flow.ConfigProtoBuilder()
-config.gpu_device_num(1)
-flow.init(config)
-
-def MatmulJob(a = flow.input_blob_def((4, 5)), b = flow.input_blob_def((5,4))):
-    job_conf = flow.get_cur_job_conf_builder()
-    job_conf.batch_size(4).data_part_num(1).default_data_type(flow.float)
-    return flow.keras.maths.matmul(a, b)
-
-flow.add_job(MatmulJob)
-
-a = np.arange(1,21).reshape((4,5)).astype(np.float32)
-b = np.arange(1,21).reshape((5,4)).astype(np.float32)
-
-with flow.Session() as sess:
-    x = sess.run(MatmulJob, a, b).get()
-
-y = torch.matmul(torch.Tensor(a), torch.Tensor(b))
-
-result = np.isclose(np.array(x), y.numpy(), rtol=1e-03, atol=1e-05)
-for i in result.ravel():
-    assert i, "the matmul test is wrong!"

oneflow/python/test/matmul_demo.py

+import tensorflow as tf
+import oneflow as flow
+import numpy as np
+
+tf.enable_eager_execution()
+assert tf.executing_eagerly()
+
+
+def test_matmul(a_shape, b_shape, transpose_a=False, transpose_b=False):
+    a = np.random.random_sample(a_shape).astype(np.float32)
+    b = np.random.random_sample(b_shape).astype(np.float32)
+
+    # OneFlow
+    config = flow.ConfigProtoBuilder()
+    config.gpu_device_num(1)
+    flow.init(config)
+
+    def MatmulTestJob(a=flow.input_blob_def(a_shape), b=flow.input_blob_def(b_shape)):
+        job_conf = flow.get_cur_job_conf_builder()
+        job_conf.batch_size(1).data_part_num(1).default_data_type(flow.float)
+        return flow.matmul(a, b, transpose_a, transpose_b)
+
+    flow.add_job(MatmulTestJob)
+    with flow.Session() as sess:
+        of_out = sess.run(MatmulTestJob, a, b).get()
+
+    # TensorFlow
+    tf_out = tf.matmul(tf.Variable(a), tf.Variable(b), transpose_a, transpose_b).numpy()
+
+    assert np.allclose(of_out, tf_out, atol=1e-7)
+
+
+# run one example each time
+if __name__ == "__main__":
+    test_matmul(a_shape=(512, 256), b_shape=(256, 1024))
+    # test_matmul(a_shape=(256, 512), b_shape=(256, 1024), transpose_a=True)
+    # test_matmul(a_shape=(512, 256), b_shape=(1024, 256), transpose_b=True)
+    # test_matmul(
+    #     a_shape=(256, 512), b_shape=(1024, 256), transpose_a=True, transpose_b=True
+    # )
+    # test_matmul(a_shape=(10, 10, 64, 32), b_shape=(10, 10, 32, 128))
+    # test_matmul(a_shape=(10, 10, 32, 64), b_shape=(10, 10, 32, 128), transpose_a=True)
+    # test_matmul(a_shape=(10, 10, 64, 32), b_shape=(10, 10, 128, 32), transpose_b=True)
+    # test_matmul(
+    #     a_shape=(10, 10, 32, 64),
+    #     b_shape=(10, 10, 128, 32),
+    #     transpose_a=True,
+    #     transpose_b=True,
+    # )