!25339 [assistant][ops] Add new Tril

Merge pull request !25339 from 张凯磊/Tril

mindspore/ccsrc/plugin/device/cpu/kernel/tril_cpu_kernel.cc

@@ -0,0 +1,154 @@
+/**
+ * Copyright 2021-2022 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 "plugin/device/cpu/kernel/tril_cpu_kernel.h"
+
+#include <algorithm>
+#include "Eigen/Core"
+
+#include "plugin/device/cpu/hal/device/cpu_device_address.h"
+
+namespace mindspore {
+namespace kernel {
+namespace {
+constexpr size_t kTrilInputsNum = 1;
+constexpr size_t kTrilOutputsNum = 1;
+constexpr size_t kDim = 2;
+}  // namespace
+
+void TrilCpuKernelMod::InitKernel(const CNodePtr &kernel_node) {
+  MS_EXCEPTION_IF_NULL(kernel_node);
+  kernel_name_ = common::AnfAlgo::GetCNodeName(kernel_node);
+  input_shape_ = common::AnfAlgo::GetPrevNodeOutputInferShape(kernel_node, 0);
+  dtype_ = AnfAlgo::GetInputDeviceDataType(kernel_node, 0);
+
+  input_dims_ = input_shape_.size();
+  if (input_dims_ < kDim) {
+    MS_LOG(EXCEPTION) << "For '" << kernel_name_ << "', the dimension of 'x' should be at least 1-D, but got "
+                      << input_dims_ << "-D.";
+  }
+  if (common::AnfAlgo::HasNodeAttr("diagonal", kernel_node)) {
+    diagonal_ = common::AnfAlgo::GetNodeAttr<int64_t>(kernel_node, "diagonal");
+  }
+}
+
+bool TrilCpuKernelMod::Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
+                              const std::vector<AddressPtr> &outputs) {
+  CHECK_KERNEL_INPUTS_NUM(inputs.size(), kTrilInputsNum, kernel_name_);
+  CHECK_KERNEL_OUTPUTS_NUM(outputs.size(), kTrilOutputsNum, kernel_name_);
+
+  switch (dtype_) {
+    case (kNumberTypeUInt8):
+      LaunchKernel<uint8_t>(inputs, outputs);
+      break;
+    case (kNumberTypeUInt16):
+      LaunchKernel<uint16_t>(inputs, outputs);
+      break;
+    case (kNumberTypeUInt32):
+      LaunchKernel<uint32_t>(inputs, outputs);
+      break;
+    case (kNumberTypeUInt64):
+      LaunchKernel<uint64_t>(inputs, outputs);
+      break;
+    case (kNumberTypeInt8):
+      LaunchKernel<int8_t>(inputs, outputs);
+      break;
+    case (kNumberTypeInt16):
+      LaunchKernel<int16_t>(inputs, outputs);
+      break;
+    case (kNumberTypeInt32):
+      LaunchKernel<int32_t>(inputs, outputs);
+      break;
+    case (kNumberTypeInt64):
+      LaunchKernel<int64_t>(inputs, outputs);
+      break;
+    case (kNumberTypeFloat16):
+      LaunchKernel<float16>(inputs, outputs);
+      break;
+    case (kNumberTypeFloat32):
+      LaunchKernel<float>(inputs, outputs);
+      break;
+    case (kNumberTypeFloat64):
+      LaunchKernel<double>(inputs, outputs);
+      break;
+    case (kNumberTypeBool):
+      LaunchKernel<bool>(inputs, outputs);
+      break;
+    default:
+      MS_LOG(EXCEPTION) << "the datatype of the input not support, support datatype: "
+                           "uint8, uint16, uint32, uint64, int8, int16, int32, int64, "
+                           "float16, float32, float64, bool.";
+  }
+  return true;
+}
+
+template <typename T>
+void TrilCpuKernelMod::LaunchKernel(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &outputs) {
+  auto input_addr = reinterpret_cast<T *>(inputs[0]->addr);
+  auto output_addr = reinterpret_cast<T *>(outputs[0]->addr);
+
+  size_t input_size = 1;
+  for (size_t i = 0; i < input_dims_; ++i) {
+    input_size *= input_shape_[i];
+  }
+
+  using MatrixMap = Eigen::Map<Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor>>;
+
+  auto matrix_width = input_shape_[input_dims_ - 2];
+  auto matrix_height = input_shape_[input_dims_ - 1];
+  auto matrix_size = matrix_width * matrix_height;
+  auto matrixs_num = input_size / matrix_size;
+
+  for (size_t k = 0; k < matrixs_num; ++k) {
+    MatrixMap input(input_addr + k * matrix_size, matrix_width, matrix_height);
+    MatrixMap output(output_addr + k * matrix_size, matrix_width, matrix_height);
+    output = input.template triangularView<Eigen::Lower>();
+    if (diagonal_ > 0) {
+      for (size_t i = 0; i < matrix_width; i++) {
+        for (size_t j = i + 1; j <= i + diagonal_ && j < matrix_height; j++) {
+          output(i, j) = input(i, j);
+        }
+      }
+    } else {
+      for (size_t j = 0; j < matrix_height; j++) {
+        for (size_t i = j; i < j - diagonal_ && i < matrix_width; i++) {
+          output(i, j) = static_cast<T>(0.0);
+        }
+      }
+    }
+  }
+}
+
+std::vector<KernelAttr> TrilCpuKernelMod::GetOpSupport() {
+  static std::vector<KernelAttr> kernel_attr_list = {
+    KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8),
+    KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16),
+    KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32),
+    KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64),
+    KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8),
+    KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16),
+    KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32),
+    KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64),
+    KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
+    KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
+    KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat64),
+    KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool)};
+  return kernel_attr_list;
+}
+
+MS_KERNEL_FACTORY_REG(NativeCpuKernelMod, Tril, TrilCpuKernelMod);
+}  // namespace kernel
+}  // namespace mindspore

mindspore/ccsrc/plugin/device/cpu/kernel/tril_cpu_kernel.h

@@ -0,0 +1,53 @@
+/**
+ * Copyright 2021-2022 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_CCSRC_PLUGIN_DEVICE_CPU_KERNEL_TRIL_CPU_KERNEL_H_
+#define MINDSPORE_CCSRC_PLUGIN_DEVICE_CPU_KERNEL_TRIL_CPU_KERNEL_H_
+
+#include <vector>
+#include <memory>
+
+#include "plugin/device/cpu/kernel/cpu_kernel.h"
+#include "plugin/factory/ms_factory.h"
+
+namespace mindspore {
+namespace kernel {
+class TrilCpuKernelMod : public NativeCpuKernelMod {
+ public:
+  TrilCpuKernelMod() = default;
+  ~TrilCpuKernelMod() override = default;
+
+  void InitKernel(const CNodePtr &kernel_node) override;
+
+  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &workspace,
+              const std::vector<AddressPtr> &outputs) override;
+
+  template <typename T>
+  void LaunchKernel(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &outputs);
+
+ protected:
+  std::vector<KernelAttr> GetOpSupport() override;
+
+ private:
+  int64_t diagonal_{0};
+  std::vector<size_t> input_shape_;
+  size_t input_dims_;
+  TypeId dtype_{kTypeUnknown};
+};
+}  // namespace kernel
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CCSRC_PLUGIN_DEVICE_CPU_KERNEL_TRIL_CPU_KERNEL_H_

mindspore/core/base/core_ops.h

@@ -132,6 +132,7 @@ constexpr auto kLstsq = "Lstsq";
 constexpr auto kLowerBound = "LowerBound";
 constexpr auto kUpperBound = "UpperBound";
 constexpr auto kCummax = "Cummax";
+constexpr auto kTril = "Tril";
 
 // NN
 constexpr auto kCTCLoss = "CTCLoss";
@@ -372,6 +373,7 @@ GVAR_DEF(PrimitivePtr, kPrimLstsq, std::make_shared<Primitive>(kLstsq));
 GVAR_DEF(PrimitivePtr, kPrimLowerBound, std::make_shared<Primitive>(kLowerBound));
 GVAR_DEF(PrimitivePtr, kPrimUpperBound, std::make_shared<Primitive>(kUpperBound));
 GVAR_DEF(PrimitivePtr, kPrimCummax, std::make_shared<Primitive>(kCummax));
+GVAR_DEF(PrimitivePtr, kPrimTril, std::make_shared<Primitive>(kTril));
 
 // image
 GVAR_DEF(PrimitivePtr, kPrimCropAndResizeGradBoxes, std::make_shared<Primitive>(kCropAndResizeGradBoxes));

mindspore/core/ops/tril.cc

@@ -0,0 +1,67 @@
+/**
+ * Copyright 2021-2022 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 "ops/tril.h"
+
+#include <algorithm>
+#include <set>
+
+#include "abstract/primitive_infer_map.h"
+#include "ops/op_utils.h"
+#include "mindapi/src/helper.h"
+
+namespace mindspore {
+namespace ops {
+namespace {
+abstract::ShapePtr TrilInferShape(const PrimitivePtr &primitive, const std::vector<AbstractBasePtr> &input_args) {
+  MS_EXCEPTION_IF_NULL(primitive);
+  auto prim_name = primitive->name();
+
+  const int64_t kShapeSize = 2;
+  auto x_shape = CheckAndConvertUtils::ConvertShapePtrToShapeMap(input_args[0]->BuildShape())[kShape];
+  (void)CheckAndConvertUtils::CheckInteger("x's rank", x_shape.size(), kGreaterEqual, kShapeSize, prim_name);
+
+  return std::make_shared<abstract::Shape>(x_shape);
+}
+
+TypePtr TrilInferType(const PrimitivePtr &primitive, const std::vector<AbstractBasePtr> &input_args) {
+  MS_EXCEPTION_IF_NULL(primitive);
+  auto prim_name = primitive->name();
+
+  MS_EXCEPTION_IF_NULL(input_args[0]);
+  auto x_type = input_args[0]->BuildType();
+  std::set<TypePtr> valid_x_types(common_valid_types);
+  (void)valid_x_types.emplace(kBool);
+  (void)CheckAndConvertUtils::CheckTensorTypeValid("x", x_type, valid_x_types, prim_name);
+  return x_type;
+}
+}  // namespace
+
+AbstractBasePtr TrilInfer(const abstract::AnalysisEnginePtr &, const PrimitivePtr &primitive,
+                          const std::vector<AbstractBasePtr> &input_args) {
+  MS_EXCEPTION_IF_NULL(primitive);
+  const int64_t kInputNum = 1;
+  CheckAndConvertUtils::CheckInputArgs(input_args, kEqual, kInputNum, primitive->name());
+
+  auto infer_type = TrilInferType(primitive, input_args);
+  auto infer_shape = TrilInferShape(primitive, input_args);
+  return abstract::MakeAbstract(infer_shape, infer_type);
+}
+
+MIND_API_BASE_IMPL(Tril, PrimitiveC, BaseOperator);
+REGISTER_PRIMITIVE_EVAL_IMPL(Tril, prim::kPrimTril, TrilInfer, nullptr, true);
+}  // namespace ops
+}  // namespace mindspore

mindspore/core/ops/tril.h

@@ -0,0 +1,42 @@
+/**
+ * Copyright 2021-2022 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_TRIL_H_
+#define MINDSPORE_CORE_OPS_TRIL_H_
+
+#include <map>
+#include <memory>
+#include <string>
+#include <vector>
+
+#include "ops/base_operator.h"
+#include "mindapi/base/types.h"
+
+namespace mindspore {
+namespace ops {
+constexpr auto kNameTril = "Tril";
+class MIND_API Tril : public BaseOperator {
+ public:
+  MIND_API_BASE_MEMBER(Tril);
+  Tril() : BaseOperator(kNameTril) { InitIOName({"x"}, {"y"}); }
+};
+
+abstract::AbstractBasePtr TrilInfer(const abstract::AnalysisEnginePtr &, const PrimitivePtr &primitive,
+                                    const std::vector<AbstractBasePtr> &input_args);
+}  // namespace ops
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CORE_OPS_TRIL_H_

mindspore/python/mindspore/ops/_grad_experimental/grad_array_ops.py

@@ -19,6 +19,7 @@ from ...common import dtype as mstype
 from .._grad.grad_math_ops import binop_grad_common
 from .._grad.grad_base import bprop_getters
 from ..composite.multitype_ops.zeros_like_impl import zeros_like
+from ..operations.array_ops import Tril
 from .. import functional as F
 from .. import operations as P
 
@@ -167,3 +168,16 @@ def get_bprop_extract_volume_patches(self):
         return (dx,)
 
     return bprop
+
+
+@bprop_getters.register(Tril)
+def get_bprop_tril(self):
+    """Grad definition for 'Tril' operation"""
+    diagonal = self.diagonal
+    tril = Tril(diagonal)
+
+    def bprop(x, out, dout):
+        dx = tril(dout)
+        return (dx,)
+
+    return bprop

mindspore/python/mindspore/ops/_op_impl/aicpu/__init__.py

@@ -133,3 +133,4 @@ from .priority_replay_buffer import _prb_create_op_cpu
 from .priority_replay_buffer import _prb_push_op_cpu
 from .priority_replay_buffer import _prb_sample_op_cpu
 from .priority_replay_buffer import _prb_update_op_cpu
+from .tril import _tril_aicpu

mindspore/python/mindspore/ops/_op_impl/aicpu/tril.py

@@ -0,0 +1,42 @@
+# Copyright 2021-2022 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.
+# ============================================================================
+
+"""Tril op"""
+from mindspore.ops.op_info_register import op_info_register, AiCPURegOp, DataType
+
+tril_op_info = AiCPURegOp("Tril") \
+    .fusion_type("OPAQUE") \
+    .attr("diagonal", "int") \
+    .input(0, "x", "required") \
+    .output(0, "y", "required") \
+    .dtype_format(DataType.U8_Default, DataType.U8_Default)   \
+    .dtype_format(DataType.U16_Default, DataType.U16_Default) \
+    .dtype_format(DataType.U32_Default, DataType.U32_Default) \
+    .dtype_format(DataType.U64_Default, DataType.U64_Default) \
+    .dtype_format(DataType.I8_Default, DataType.I8_Default)   \
+    .dtype_format(DataType.I16_Default, DataType.I16_Default) \
+    .dtype_format(DataType.I32_Default, DataType.I32_Default) \
+    .dtype_format(DataType.I64_Default, DataType.I64_Default) \
+    .dtype_format(DataType.F16_Default, DataType.F16_Default) \
+    .dtype_format(DataType.F32_Default, DataType.F32_Default) \
+    .dtype_format(DataType.F64_Default, DataType.F64_Default) \
+    .dtype_format(DataType.BOOL_Default, DataType.BOOL_Default) \
+    .get_op_info()
+
+
+@op_info_register(tril_op_info)
+def _tril_aicpu():
+    """Tril AiCPU register"""
+    return

mindspore/python/mindspore/ops/operations/array_ops.py

@@ -7272,3 +7272,43 @@ class Cummax(Primitive):
         """Initialize Cummax"""
         validator.check_value_type("dim", dim, [int], self.name)
         self.init_prim_io_names(inputs=['x'], outputs=['y', 'indices'])
+
+
+class Tril(Primitive):
+    """
+    Returns the lower triangular part of the matrix (2-D tensor) or batch of matrices input,
+    the other elements of the result tensor out are set to 0.
+    The lower triangular part of the matrix is defined as the elements on and below the diagonal.
+
+    Args:
+        diagonal (int): An optional attribute indicates the diagonal to consider, default to 0.
+
+    Inputs:
+        - **x** (Tensor) - A Tensor with shape :math:`(x_1, x_2, ..., x_R)`. The rank must be at least 2.
+          Supporting all number types including bool.
+
+    Outputs:
+        Tensor, the same shape and data type as the input.
+
+    Raises:
+        TypeError: If `x` is not a Tensor.
+        TypeError: If `diagonal` is not an int.
+        TypeError: If the type of `x` is neither number nor bool.
+        ValueError: If the rank of `x` is less than 2.
+
+    Supported Platforms:
+        ``CPU``
+
+    Examples:
+        >>> tril = ops.Tril()
+        >>> output = tril(Tensor(np.array([[-13.5383, 2.5474, ], [-5.7496, -3.4548]]), mindspore.float32))
+        >>> print(output)
+        [[ -13.5383  0.    ]
+         [  -5.7496 -3.4548]]
+    """
+
+    @prim_attr_register
+    def __init__(self, diagonal=0):
+        """Initialize Tril."""
+        self.init_prim_io_names(inputs=["x"], outputs=["y"])
+        validator.check_value_type("diagonal", diagonal, [int], self.name)

tests/ut/python/ops/test_ops.py

@@ -31,6 +31,7 @@ from mindspore.ops.operations import _grad_ops as G
 from mindspore.ops.operations import _inner_ops as inner
 from mindspore.ops.operations import _quant_ops as Q
 from mindspore.ops.operations import nn_ops as nps
+from mindspore.ops.operations.array_ops import Tril
 from mindspore.ops.operations.random_ops import NonDeterministicInts
 from mindspore.nn.layer import normalization
 from mindspore._c_expression import security
@@ -2823,6 +2824,11 @@ test_case_array_ops = [
         'desc_inputs': [Tensor([[1, 2, 3], [4, 5, 6], [7, 8, 9]])],
         'skip': ['backward'],
     }),
+    ('Tril', {
+        'block': Tril(),
+        'desc_inputs': [Tensor(np.random.rand(3, 8, 9), mstype.float32)],
+        'desc_brop': [Tensor(np.random.rand(5, 6, 6), mstype.float32)]
+    }),
 ]
 
 test_case_image_ops = [