[feat] [assistant] [I40GGM] add new ascend operator Ger

mindspore/core/base/core_ops.h

@@ -67,6 +67,7 @@ constexpr auto kAbs = "Abs";
 constexpr auto kTrunc = "Trunc";
 constexpr auto kSquare = "Square";
 constexpr auto kReal = "Real";
+constexpr auto kGer = "Ger";
 
 // Arrays
 constexpr auto kDynamicShape = "DynamicShape";
@@ -467,6 +468,7 @@ inline const PrimitivePtr kPrimTensorListStack = std::make_shared<Primitive>("Te
 inline const PrimitivePtr kPrimTensorListSetItem = std::make_shared<Primitive>("TensorListSetItem");
 
 // Maths
+inline const PrimitivePtr kPrimGer = std::make_shared<Primitive>("Ger");
 inline const PrimitivePtr kPrimCeil = std::make_shared<Primitive>("Ceil");
 inline const PrimitivePtr kPrimTensorAdd = std::make_shared<Primitive>("TensorAdd");
 inline const PrimitivePtr kPrimAdd = std::make_shared<Primitive>(kAdd);

mindspore/core/ops/ger.cc

@@ -0,0 +1,67 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <algorithm>
+#include <set>
+
+#include "ops/ger.h"
+#include "ops/op_utils.h"
+#include "abstract/primitive_infer_map.h"
+
+namespace mindspore {
+namespace ops {
+namespace {
+abstract::ShapePtr InferShape(const PrimitivePtr &primitive, const std::vector<AbstractBasePtr> &input_args) {
+  MS_EXCEPTION_IF_NULL(primitive);
+  auto prim_name = primitive->name();
+  for (const auto &item : input_args) {
+    MS_EXCEPTION_IF_NULL(item);
+  }
+
+  auto first_input_shape = CheckAndConvertUtils::ConvertShapePtrToShapeMap(input_args[0]->BuildShape())[kShape];
+  auto second_input_shape = CheckAndConvertUtils::ConvertShapePtrToShapeMap(input_args[1]->BuildShape())[kShape];
+  (void)CheckAndConvertUtils::CheckInteger("x1 rank", SizeToLong(first_input_shape.size()), kEqual, 1, prim_name);
+  (void)CheckAndConvertUtils::CheckInteger("x2 rank", SizeToLong(second_input_shape.size()), kEqual, 1, prim_name);
+
+  std::vector<int64_t> out_shape = {first_input_shape[0], second_input_shape[0]};
+  return std::make_shared<abstract::Shape>(out_shape);
+}
+
+TypePtr InferType(const PrimitivePtr &prim, const std::vector<AbstractBasePtr> &input_args) {
+  MS_EXCEPTION_IF_NULL(prim);
+  auto prim_name = prim->name();
+  const std::set<TypePtr> valid_types = {kFloat16, kFloat32};
+  for (const auto &item : input_args) {
+    MS_EXCEPTION_IF_NULL(item);
+  }
+  std::map<std::string, TypePtr> types;
+  types.emplace("x1", input_args[0]->BuildType());
+  types.emplace("x2", input_args[1]->BuildType());
+  return CheckAndConvertUtils::CheckTensorTypeSame(types, valid_types, prim_name);
+}
+}  // namespace
+AbstractBasePtr GerInfer(const abstract::AnalysisEnginePtr &, const PrimitivePtr &primitive,
+                         const std::vector<AbstractBasePtr> &input_args) {
+  MS_EXCEPTION_IF_NULL(primitive);
+  const int64_t input_num = 2;
+  CheckAndConvertUtils::CheckInputArgs(input_args, kEqual, input_num, primitive->name());
+  auto type = InferType(primitive, input_args);
+  auto shape = InferShape(primitive, input_args);
+  return abstract::MakeAbstract(shape, type);
+}
+REGISTER_PRIMITIVE_EVAL_IMPL(Ger, prim::kPrimGer, GerInfer, nullptr, true);
+}  // namespace ops
+}  // namespace mindspore

mindspore/core/ops/ger.h

@@ -0,0 +1,44 @@
+/**
+ * Copyright 2021 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_CORE_OPS_GER_H_
+#define MINDSPORE_CORE_OPS_GER_H_
+
+#include <map>
+#include <memory>
+#include <string>
+#include <vector>
+#include "ops/primitive_c.h"
+#include "abstract/abstract_value.h"
+#include "utils/check_convert_utils.h"
+
+namespace mindspore {
+namespace ops {
+constexpr auto kNameGer = "Ger";
+class MS_CORE_API Ger : public PrimitiveC {
+ public:
+  Ger() : PrimitiveC(kNameGer) { InitIOName({"x", "y"}, {"output"}); }
+  ~Ger() = default;
+  MS_DECLARE_PARENT(Ger, PrimitiveC);
+};
+
+AbstractBasePtr GerInfer(const abstract::AnalysisEnginePtr &, const PrimitivePtr &primitive,
+                         const std::vector<AbstractBasePtr> &input_args);
+using PrimGerPtr = std::shared_ptr<Ger>;
+}  // namespace ops
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CORE_OPS_GER_H_

mindspore/ops/_grad_experimental/grad_math_ops.py

@@ -108,3 +108,19 @@ def get_bprop_trunc(self):
         return (bc_x,)
 
     return bprop
+
+
+@bprop_getters.register(P.Ger)
+def get_bprop_ger(self):
+    """Grad definition for 'Ger' operation"""
+    transpose_op = P.Transpose()
+    matmul = P.MatMul()
+    expand_dims = P.ExpandDims()
+    squeeze = P.Squeeze(1)
+
+    def bprop(input_x, input_y, out, dout):
+        dx = squeeze(matmul(dout, expand_dims(input_y, 1)))
+        dy = squeeze(matmul(transpose_op(dout, (1, 0)), expand_dims(input_x, 1)))
+        return dx, dy
+
+    return bprop

mindspore/ops/_op_impl/tbe/__init__.py

@@ -65,6 +65,7 @@ from .dropout_do_mask import _dropout_do_mask_tbe
 from .dropout_do_mask_ds import _dropout_do_mask_ds_tbe
 from .gelu import _gelu_tbe
 from .gelu_grad import _gelu_grad_tbe
+from .ger import _ger_tbe
 from .fast_gelu import _fast_gelu_tbe
 from .fast_gelu_grad import _fast_gelu_grad_tbe
 from .max_pool import _max_pool_tbe

mindspore/ops/_op_impl/tbe/ger.py

@@ -0,0 +1,43 @@
+# Copyright 2021 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+"""Ger op"""
+from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
+
+ger_op_info = TBERegOp("Ger") \
+    .fusion_type("OPAQUE") \
+    .async_flag(False) \
+    .binfile_name("ger.so") \
+    .compute_cost(10) \
+    .kernel_name("ger") \
+    .partial_flag(True) \
+    .input(0, "x1", False, "required", "all") \
+    .input(0, "x2", False, "required", "all") \
+    .output(0, "y", False, "required", "all") \
+    .dtype_format(DataType.F16_Default, DataType.F16_Default, DataType.F16_Default) \
+    .dtype_format(DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD) \
+    .dtype_format(DataType.F16_NHWC, DataType.F16_NHWC, DataType.F16_NHWC) \
+    .dtype_format(DataType.F16_Default, DataType.F16_Default, DataType.F16_Default) \
+    .dtype_format(DataType.F32_Default, DataType.F32_Default, DataType.F32_Default) \
+    .dtype_format(DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD) \
+    .dtype_format(DataType.F32_NHWC, DataType.F32_NHWC, DataType.F32_NHWC) \
+    .dtype_format(DataType.F32_Default, DataType.F32_Default, DataType.F32_Default) \
+    .get_op_info()
+
+
+@op_info_register(ger_op_info)
+def _ger_tbe():
+    """Ger TBE register"""
+    return

mindspore/ops/operations/__init__.py

@@ -47,7 +47,7 @@ from .inner_ops import (ScalarCast, Randperm, NoRepeatNGram, LambApplyOptimizerA
                         FusedWeightScaleApplyMomentum, FusedCastAdamWeightDecay)
 
 from .math_ops import (Abs, ACos, Asin, Asinh, AddN, AccumulateNV2, AssignAdd, AssignSub, Atan2, BatchMatMul,
-                       BitwiseAnd, BitwiseOr,
+                       BitwiseAnd, BitwiseOr, Ger,
                        BitwiseXor, Inv, Invert, ApproximateEqual, InplaceAdd, InplaceSub,
                        ReduceMax, ReduceMin, ReduceMean, ReduceSum, ReduceAll, ReduceProd, CumProd, Cdist, ReduceAny,
                        Cos, Div, DivNoNan, Equal, EqualCount, Exp, Expm1, Erf, Erfc, Floor, FloorDiv, FloorMod, Ceil,
@@ -123,6 +123,7 @@ from .rl_ops import (BufferAppend, BufferGetItem, BufferSample)
 from ._inner_ops import (MatmulDDS, DSDMatmul, NonZero)
 
 __all__ = [
+    'Ger',
     'Unique',
     'ReverseSequence',
     'Sort',

mindspore/ops/operations/math_ops.py

@@ -136,6 +136,46 @@ class _BitwiseBinaryOp(_MathBinaryOp):
         return _BitwiseBinaryOp._check_bitwise_op_input_type(x1_type, x2_type, self.name)
 
 
+class Ger(Primitive):
+    r"""
+    Ger product of `x1` and `x2`. Calculate the outer product of two one-dimensional arrays.If `x1` is a 1D Tensor of
+    shape :math:`(m,)` and `x2` is a 1D Tensor of shape :math:`(n,)`,then `output` must be a Tensor of shape
+    :math:`(m * n)`.
+
+
+    Inputs:
+        - **x1** - (Tensor) - 1-D input Tensor, with dtype of float16 or float32.
+        - **x2** - (Tensor) - 1-D input Tensor, with dtype of float16 or float32.
+
+    Outputs:
+        Tensor, output matrix with the same dtype as inputs.With `x1` shape :math:`(m,)` and
+        `x2` shape of :math:`(n,)`,`output` has shape :math:`(m * n)`.
+
+    Raises:
+        TypeError: If `x1` or `x2` is not a Tensor.
+        TypeError: If the dtype of `x1` and `x2` is neither float16 nor float32.
+        ValueError: If `x1` or `x2` is not a 1D Tensor.
+
+    Supported Platforms:
+        ``Ascend``
+
+    Examples:
+        >>> x1 = Tensor([1., 2., 3., 4.], mindspore.float32)
+        >>> x2 = Tensor([1., 2., 3.], mindspore.float32)
+        >>> ger = ops.Ger()
+        >>> output = ger(x1, x2)
+        >>> print(output)
+        [[ 1.  2.  3.]
+         [ 2.  4.  6.]
+         [ 3.  6.  9.]
+         [ 4.  8. 12.]]
+    """
+
+    @prim_attr_register
+    def __init__(self):
+        """Initialize Ger"""
+        self.init_prim_io_names(inputs=['x1', 'x2'], outputs=['y'])
+
 
 class Add(_MathBinaryOp):
     r"""

tests/ut/python/ops/test_ops.py

@@ -1043,6 +1043,10 @@ class SparseApplyRMSPropNet(nn.Cell):
         return out
 
 test_case_math_ops = [
+    ('Ger', {
+        'block': P.Ger(),
+        'desc_inputs': [[3,], [4,]],
+        'desc_bprop': [[3, 4]]}),
     ('BitwiseAnd', {
         'block': P.BitwiseAnd(),
         'desc_inputs': [Tensor(np.array([0, 0, 1, -1, 1, 1, 1]), mstype.int16),