/** * Copyright 2019-2020 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 GE_OP_MATRIX_CALCULATION_OPS_H #define GE_OP_MATRIX_CALCULATION_OPS_H #include "../graph/operator_reg.h" namespace ge { /** *@brief Multiplies matrix "a" by matrix "b", producing "a * b". *@par Inputs: *Two inputs, including: * @li x1: A matrix Tensor. 2D. Must be one of the following types: float16, * float32, int32. Has format [ND, NHWC, FRACTAL_NZ]. * @li x2: A matrix Tensor. 2D. Must be one of the following types: float16, * float32, int32. Has format [ND, NHWC, FRACTAL_NZ]. * @li bias: A 1D Tensor. Must be one of the following types: float16, * float32, int32. Has format [ND, NHWC]. *@par Attributes: *@li transpose_a: A bool. If True, changes the shape of "x1" from [M, K] to [K, M]. *@li transpose_b: A bool. If True, changes the shape of "x2" from [M, K] to [K, M]. *@par Outputs: *y: The result matrix Tensor. 2D. Must be one of the following types: float16, * float32, int32. Has format [ND, NHWC, FRACTAL_NZ]. */ REG_OP(MatMul) .INPUT(x1, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .INPUT(x2, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .OPTIONAL_INPUT(bias, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .ATTR(transpose_a, Bool, false) .ATTR(transpose_b, Bool, false) .OP_END_FACTORY_REG(MatMul) REG_OP(MatMulV2) .INPUT(x1, TensorType({DT_FLOAT16, DT_FLOAT16, DT_INT8, DT_INT8})) .INPUT(x2, TensorType({DT_FLOAT16, DT_FLOAT16, DT_INT8, DT_INT8})) .INPUT(alpha, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32, DT_FLOAT})) .INPUT(beta, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32, DT_FLOAT})) .INPUT(bias, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32, DT_FLOAT})) .OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32, DT_FLOAT})) .OP_END_FACTORY_REG(MatMulV2) /** *@brief Multiplies matrix "a" by matrix "b", producing "a * b". *@par Inputs: *Three inputs, including: * @li x1: A matrix Tensor. Must be one of the following types: float16, * float32, int32. 2D or higher. Has format [ND, NHWC, FRACTAL_NZ]. * @li x2: A matrix Tensor. Must be one of the following types: float16, * float32, int32. 2D or higher. Has format [ND, NHWC, FRACTAL_NZ]. *@par Attributes: *@li adj_x: A bool. If True, changes the shape of "x1" from [B, M, K] to [B, K, M]. *@li adj_y: A bool. If True, changes the shape of "x2" from [B, M, K] to [B, K, M]. *@par Outputs: *y: The result matrix Tensor. 2D or higher. Must be one of the following types: float16, * float32, int32. 2D or higher. Has format [ND, NHWC, FRACTAL_NZ]. Has the same shape length as "x1" and "x2". */ REG_OP(BatchMatMul) .INPUT(x1, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .INPUT(x2, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .ATTR(adj_x, Bool, false) .ATTR(adj_y, Bool, false) .OP_END_FACTORY_REG(BatchMatMul) REG_OP(MeanCCE) .INPUT(x, TensorType::ALL()) .INPUT(indices, TensorType::ALL()) .OUTPUT(y, TensorType::ALL()) .ATTR(keep_dims, Bool, false) .ATTR(value1, ListInt, {}) .ATTR(mode, Int, 3) // 0:max pooling or 1:avg pooling .ATTR(pad_mode, Int, 0) .ATTR(global_pooling, Bool, true) .ATTR(window, ListInt, {1,1}) // kernel size .ATTR(pad, ListInt, {0,0,0,0}) // pad size .ATTR(stride, ListInt, {1,1}) // stride size .ATTR(ceil_mode, Int, 0) .ATTR(data_mode, Int, 1) .ATTR(nan_opt, Int, 0) .ATTR(fomart, Int, 0) .OP_END_FACTORY_REG(MeanCCE) REG_OP(MeanGrad) .INPUT(x, TensorType::ALL()) .OUTPUT(y, TensorType::ALL()) .ATTR(mode, Int, 1) // 0:max pooling or 1:avg pooling .ATTR(pad_mode, Int, 0) .ATTR(global_pooling, Bool, false) .ATTR(window, ListInt, {1,1}) // kernel size .ATTR(pad, ListInt, {0,0,0,0}) // pad size .ATTR(stride, ListInt, {1,1}) // stride size .ATTR(ceil_mode, Int, 0) .ATTR(data_mode, Int, 1) .ATTR(nan_opt, Int, 0) .ATTR(mean_grad_output_shape_value, ListInt, {1,1,1,1}) .ATTR(mean_grad_output_shape_format, Int, 1) //must be NHWC .OP_END_FACTORY_REG(MeanGrad) REG_OP(MatMulCCE) .INPUT(x1, TensorType({DT_FLOAT})) .INPUT(x2, TensorType({DT_FLOAT})) .OPTIONAL_INPUT(x3, TensorType({DT_FLOAT})) .OUTPUT(y, TensorType({DT_FLOAT})) .ATTR(transpose_a, Bool, false) .ATTR(transpose_b, Bool, false) .ATTR(has_bias, Bool, false) .OP_END_FACTORY_REG(MatMulCCE) /** *@brief Computes half the L2 norm of a tensor without the sqrt. *@par Inputs: * x: A Tensor. * TensorType::FloatingDataType(). *@par Outputs: *y: A Tensor. Has the same type as "x". */ REG_OP(L2Loss) .INPUT(x, TensorType::FloatingDataType()) .OUTPUT(y, TensorType::FloatingDataType()) .OP_END_FACTORY_REG(L2Loss) REG_OP(MatrixDiag) .INPUT(x, TensorType::BasicType()) .OUTPUT(y, TensorType::BasicType()) .OP_END_FACTORY_REG(MatrixDiag) REG_OP(MatrixDiagD) .INPUT(x, TensorType::BasicType()) .INPUT(assist, TensorType::BasicType()) .OUTPUT(y, TensorType::BasicType()) .OP_END_FACTORY_REG(MatrixDiagD) REG_OP(MatrixDiagPart) .INPUT(x, TensorType::BasicType()) .OUTPUT(y, TensorType::BasicType()) .OP_END_FACTORY_REG(MatrixDiagPart) REG_OP(MatrixDiagPartD) .INPUT(x, TensorType::BasicType()) .INPUT(assist, TensorType::BasicType()) .OUTPUT(y, TensorType::BasicType()) .OP_END_FACTORY_REG(MatrixDiagPartD) REG_OP(MatrixSetDiag) .INPUT(x, TensorType::BasicType()) .INPUT(diagonal, TensorType::BasicType()) .OUTPUT(y, TensorType::BasicType()) .OP_END_FACTORY_REG(MatrixSetDiag) REG_OP(MatrixSetDiagD) .INPUT(x, TensorType::BasicType()) .INPUT(diagonal, TensorType::BasicType()) .INPUT(assist, TensorType::BasicType()) .OUTPUT(y, TensorType::BasicType()) .OP_END_FACTORY_REG(MatrixSetDiagD) REG_OP(ScatterNdUpdate) .INPUT(var, TensorType::BasicType()) .INPUT(indices, TensorType::IndexNumberType()) .INPUT(updates, TensorType::BasicType()) .OUTPUT(var, TensorType::BasicType()) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterNdUpdate) REG_OP(ScatterAdd) .INPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .INPUT(indices, TensorType::IndexNumberType()) .INPUT(updates, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .OUTPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterAdd) REG_OP(ScatterDiv) .INPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .INPUT(indices, TensorType({DT_INT32})) .INPUT(updates, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .OUTPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterDiv) REG_OP(ScatterNdAdd) .INPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .INPUT(indices, TensorType::IndexNumberType()) .INPUT(updates, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .OUTPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterNdAdd) REG_OP(ScatterNdSub) .INPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .INPUT(indices, TensorType::IndexNumberType()) .INPUT(updates, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .OUTPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterNdSub) REG_OP(ScatterSub) .INPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .INPUT(indices, TensorType::IndexNumberType()) .INPUT(updates, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .OUTPUT(var, TensorType({DT_FLOAT16, DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterSub) REG_OP(DiagPartD) .INPUT(x, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .INPUT(assist, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32})) .OP_END_FACTORY_REG(DiagPartD) REG_OP(DiagPart) .INPUT(x, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32, DT_INT64, DT_DOUBLE, DT_COMPLEX64, DT_COMPLEX128})) .OUTPUT(y, TensorType({DT_FLOAT, DT_FLOAT16, DT_INT32, DT_INT64, DT_DOUBLE, DT_COMPLEX64, DT_COMPLEX128})) .OP_END_FACTORY_REG(DiagPart) REG_OP(InnerProduct) .INPUT(x, TensorType({DT_FLOAT16, DT_INT8})) .INPUT(w, TensorType({DT_FLOAT16, DT_INT8})) .OPTIONAL_INPUT(b, TensorType({DT_FLOAT16, DT_INT32})) .OPTIONAL_INPUT(offset_w, TensorType({DT_INT8})) .OUTPUT(y, TensorType({DT_FLOAT16, DT_INT32})) .REQUIRED_ATTR(num_output, Int) .ATTR(transpose, Bool, false) .ATTR(bias_term, Bool, true) .ATTR(axis, Int, 1) .ATTR(offset_a, Int, 0) .OP_END_FACTORY_REG(InnerProduct) REG_OP(ConfusionMatrix) .INPUT(labels, TensorType({DT_FLOAT, DT_INT32, DT_FLOAT16, DT_INT8, DT_UINT8})) .INPUT(predictions, TensorType({DT_FLOAT, DT_INT32, DT_FLOAT16, DT_INT8, DT_UINT8})) .OPTIONAL_INPUT(weights, TensorType({DT_FLOAT, DT_INT32, DT_FLOAT16, DT_INT8, DT_UINT8})) .OUTPUT(y, TensorType({DT_FLOAT, DT_INT32, DT_FLOAT16, DT_INT8, DT_UINT8})) .REQUIRED_ATTR(num_classes, Int) .REQUIRED_ATTR(dtype, String) .OP_END_FACTORY_REG(ConfusionMatrix) REG_OP(ScatterMul) .INPUT(var, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .INPUT(indices, TensorType({DT_INT32})) .INPUT(updates, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .OUTPUT(var, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32,DT_INT8,DT_UINT8})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterMul) REG_OP(ScatterMin) .INPUT(var, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32})) .INPUT(indices, TensorType({DT_INT32})) .INPUT(updates, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32})) .OUTPUT(var, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterMin) REG_OP(ScatterMax) .INPUT(var, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32})) .INPUT(indices, TensorType({DT_INT32})) .INPUT(updates, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32})) .OUTPUT(var, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT32})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterMax) REG_OP(SparseApplyAdagrad) .INPUT(var, TensorType({DT_FLOAT})) .INPUT(accum, TensorType({DT_FLOAT})) .INPUT(lr, TensorType({DT_FLOAT})) .INPUT(grad, TensorType({DT_FLOAT})) .INPUT(indices, TensorType({DT_INT32})) .OUTPUT(var, TensorType({DT_FLOAT})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(SparseApplyAdagrad) REG_OP(SparseApplyAdagradD) .INPUT(var, TensorType({DT_FLOAT})) .INPUT(accum, TensorType({DT_FLOAT})) .INPUT(grad, TensorType({DT_FLOAT})) .INPUT(indices, TensorType({DT_INT32})) .OUTPUT(var, TensorType({DT_FLOAT})) .REQUIRED_ATTR(lr, Float) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(SparseApplyAdagradD) REG_OP(ScatterUpdate) .INPUT(var, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT8,DT_UINT8})) .INPUT(indices, TensorType({DT_INT32})) .INPUT(updates, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT8,DT_UINT8})) .OUTPUT(var, TensorType({DT_FLOAT16,DT_FLOAT,DT_INT8,DT_UINT8})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(ScatterUpdate) /** * @brief Update relevant entries in '*var' according to the Ftrl-proximal scheme. * That is for rows we have grad for, we update var, accum and linear * @par Inputs: * Ten inputs, including: * @li var: A mutable Tensor. Must be of type TensorType::NumberType(). * Should be a Variable Tensor. * @li accum: A mutable Tensor of the same type as "var". * Should be a Variable Tensor. * @li linear: A mutable Tensor of the same type as "var". * Should be a Variable Tensor. * @li grad: A Tensor of the same type as "var", for the gradient. * @li indices: A vector of indices into the first dimension of var and accum. * @li lr: A Tensor of the same type as "var", for the scaling factor. Must be a scalar. * @li l1: A Tensor of the same type as "var", for L1 regulariation. Must be a scalar. * @li l2: A Tensor of the same type as "var", for L2 regulariation. Must be a scalar. * @li l2_shrinkage: A Tensor of the same type as "var", L2 shrinkage regulariation. Must be a scalar. * @li lr_power: A Tensor of the same type as "var", for the scaling factor. Must be a scalar. * @par Attributes: * use_locking: An optional bool. Defaults to "False". * If "True", updating of the "var" and "accum" tensors will be * rotected by a lock; otherwise the behavior is undefined, * but may exhibit less contention. * @par Outputs: * var: A Tensor. Has the same type and format as input "var". */ REG_OP(SparseApplyFtrlV2) .INPUT(var, TensorType({DT_FLOAT})) .INPUT(accum, TensorType({DT_FLOAT})) .INPUT(linear, TensorType({DT_FLOAT})) .INPUT(grad, TensorType({DT_FLOAT})) .INPUT(indices, TensorType({DT_INT32})) .INPUT(lr, TensorType({DT_FLOAT})) .INPUT(l1, TensorType({DT_FLOAT})) .INPUT(l2, TensorType({DT_FLOAT})) .INPUT(l2_shrinkage, TensorType({DT_FLOAT})) .INPUT(lr_power, TensorType({DT_FLOAT})) .OUTPUT(var, TensorType({DT_FLOAT})) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(SparseApplyFtrlV2) /** * @brief Update relevant entries in '*var' according to the Ftrl-proximal scheme. * That is for rows we have grad for, we update var, accum and linear * @par Inputs: * Ten inputs, including: * @li var: A mutable Tensor. Must be of type TensorType::NumberType(). * Should be a Variable Tensor. * @li accum: A mutable Tensor of the same type as "var". * Should be a Variable Tensor. * @li linear: A mutable Tensor of the same type as "var". * Should be a Variable Tensor. * @li grad: A Tensor of the same type as "var", for the gradient. * @li indices: A vector of indices into the first dimension of var and accum. * @par Attributes: * @li lr: A Tensor of the same type as "var", for the scaling factor. Must be a scalar. * @li l1: A Tensor of the same type as "var", for L1 regulariation. Must be a scalar. * @li l2: A Tensor of the same type as "var", for L2 regulariation. Must be a scalar. * @li l2_shrinkage: A Tensor of the same type as "var", L2 shrinkage regulariation. Must be a scalar. * @li lr_power: A Tensor of the same type as "var", for the scaling factor. Must be a scalar. * @li use_locking: An optional bool. Defaults to "False". * If "True", updating of the "var" and "accum" tensors will be * rotected by a lock; otherwise the behavior is undefined, * but may exhibit less contention. * @par Outputs: * var: A Tensor. Has the same type and format as input "var". */ REG_OP(SparseApplyFtrlV2D) .INPUT(var, TensorType({DT_FLOAT})) .INPUT(accum, TensorType({DT_FLOAT})) .INPUT(linear, TensorType({DT_FLOAT})) .INPUT(grad, TensorType({DT_FLOAT})) .INPUT(indices, TensorType({DT_INT32})) .OUTPUT(var, TensorType({DT_FLOAT})) .REQUIRED_ATTR(lr, Float) .REQUIRED_ATTR(l1, Float) .REQUIRED_ATTR(l2, Float) .REQUIRED_ATTR(l2_shrinkage, Float) .REQUIRED_ATTR(lr_power, Float) .ATTR(use_locking, Bool, false) .OP_END_FACTORY_REG(SparseApplyFtrlV2D) } // namespace ge #endif // GE_OP_MATRIX_CALCULATION_OPS_H