p60975123
/
mindspore
forked from mindspore-Ecosystem/mindspore
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

sparse gather v2

This commit is contained in:
zhaozhenlong 2020-06-11 16:35:48 +08:00
parent 95c8de970e
commit 562c2b8d3d
6 changed files with 75 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
mindspore/ccsrc/kernel/tbe/tbe_adapter.cc
View File

@ -53,6 +53,7 @@ static std::map<string, string> tbe_func_adapter_map = {
{"scatter_nd", "scatter_nd_d"},
{"tile", "tile_d"},
{"gather_v2", "gather_v2_d"},
{"sparse_gather_v2", "gather_v2_d"},
{"batch_mat_mul", "batch_matmul"},
{"b_n_training_reduce", "bn_training_reduce"},
{"b_n_training_update", "bn_training_update"},

1
mindspore/ccsrc/pre_activate/pass/const_input_to_attr_registry.cc
View File

@ -47,6 +47,7 @@ ConstInputToAttrInfoRegistry::ConstInputToAttrInfoRegistry() {
Register(prim::kPrimCumProd->name(), {1});
Register(prim::kPrimReduceAll->name(), {1});
Register(prim::kPrimUnsortedSegmentMin->name(), {2});
Register(kSparseGatherV2, {2});
Register(kUnsortedSegmentProdOpName, {2});
Register(kSimpleMeanGradOpName, {1});
Register(kMeanGradOpName, {1});

1
mindspore/ccsrc/utils/utils.h
View File

@ -65,6 +65,7 @@ constexpr auto kScatterNdOpName = "ScatterNd";
constexpr auto kStridedSliceAssignOpName = "StridedSliceAssign";
constexpr auto kStridedSliceOpName = "StridedSlice";
constexpr auto kStridedSliceGradOpName = "StridedSliceGrad";
constexpr auto kSparseGatherV2 = "SparseGatherV2";
constexpr auto kUnsortedSegmentProdOpName = "UnsortedSegmentProd";
constexpr auto kUnsortedSegmentMinOpName = "UnsortedSegmentMin";
constexpr auto kFlattenGradOpName = "FlattenGrad";

1
mindspore/ops/_op_impl/tbe/__init__.py
View File

@ -248,3 +248,4 @@ from .range import _range_tbe
from .fused_mul_add_n_l2loss import _fused_mul_add_n_l2loss_tbe
from .fused_mul_apply_momentum_extern import _fused_mul_apply_momentum_extern_tbe
from .lamb_next_right import _lamb_next_right_tbe
from .sparse_gather_v2 import _sparse_gather_v2_tbe

66
mindspore/ops/_op_impl/tbe/sparse_gather_v2.py Normal file
View File

@ -0,0 +1,66 @@
# Copyright 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.
# ============================================================================
"""SparseGatherV2 op"""
from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
sparse_gather_v2_op_info = TBERegOp("SparseGatherV2") \
.fusion_type("OPAQUE") \
.async_flag(False) \
.binfile_name("gather_v2_d.so") \
.compute_cost(10) \
.kernel_name("gather_v2_d") \
.partial_flag(True) \
.attr("axis", "optional", "int", "all") \
.input(0, "x", False, "required", "all") \
.input(1, "indices", False, "required", "all") \
.output(0, "y", False, "required", "all") \
.dtype_format(DataType.I8_Default, DataType.I32_Default, DataType.I8_Default) \
.dtype_format(DataType.I8_Default, DataType.I64_Default, DataType.I8_Default) \
.dtype_format(DataType.I8_5HD, DataType.I32_5HD, DataType.I8_5HD) \
.dtype_format(DataType.I8_5HD, DataType.I64_5HD, DataType.I8_5HD) \
.dtype_format(DataType.I8_FracZ, DataType.I32_FracZ, DataType.I8_FracZ) \
.dtype_format(DataType.I8_FracZ, DataType.I64_FracZ, DataType.I8_FracZ) \
.dtype_format(DataType.U8_Default, DataType.I32_Default, DataType.U8_Default) \
.dtype_format(DataType.U8_Default, DataType.I64_Default, DataType.U8_Default) \
.dtype_format(DataType.U8_5HD, DataType.I32_5HD, DataType.U8_5HD) \
.dtype_format(DataType.U8_5HD, DataType.I64_5HD, DataType.U8_5HD) \
.dtype_format(DataType.U8_FracZ, DataType.I32_FracZ, DataType.U8_FracZ) \
.dtype_format(DataType.U8_FracZ, DataType.I64_FracZ, DataType.U8_FracZ) \
.dtype_format(DataType.I32_Default, DataType.I32_Default, DataType.I32_Default) \
.dtype_format(DataType.I32_Default, DataType.I64_Default, DataType.I32_Default) \
.dtype_format(DataType.I32_5HD, DataType.I32_5HD, DataType.I32_5HD) \
.dtype_format(DataType.I32_5HD, DataType.I64_5HD, DataType.I32_5HD) \
.dtype_format(DataType.I32_FracZ, DataType.I32_FracZ, DataType.I32_FracZ) \
.dtype_format(DataType.I32_FracZ, DataType.I64_FracZ, DataType.I32_FracZ) \
.dtype_format(DataType.F16_Default, DataType.I32_Default, DataType.F16_Default) \
.dtype_format(DataType.F16_Default, DataType.I64_Default, DataType.F16_Default) \
.dtype_format(DataType.F16_5HD, DataType.I32_5HD, DataType.F16_5HD) \
.dtype_format(DataType.F16_5HD, DataType.I64_5HD, DataType.F16_5HD) \
.dtype_format(DataType.F16_FracZ, DataType.I32_FracZ, DataType.F16_FracZ) \
.dtype_format(DataType.F16_FracZ, DataType.I64_FracZ, DataType.F16_FracZ) \
.dtype_format(DataType.F32_Default, DataType.I32_Default, DataType.F32_Default) \
.dtype_format(DataType.F32_Default, DataType.I64_Default, DataType.F32_Default) \
.dtype_format(DataType.F32_5HD, DataType.I32_5HD, DataType.F32_5HD) \
.dtype_format(DataType.F32_5HD, DataType.I64_5HD, DataType.F32_5HD) \
.dtype_format(DataType.F32_FracZ, DataType.I32_FracZ, DataType.F32_FracZ) \
.dtype_format(DataType.F32_FracZ, DataType.I64_FracZ, DataType.F32_FracZ) \
.get_op_info()
@op_info_register(sparse_gather_v2_op_info)
def _sparse_gather_v2_tbe():
"""SparseGatherV2 TBE register"""
return

5
tests/ut/python/ops/test_ops.py
View File

@ -956,6 +956,11 @@ test_case_nn_ops = [
'desc_const': [0],
'desc_inputs': [[1152], Tensor(np.array(10).astype(np.int32))],
'desc_bprop': [Tensor(np.array(10).astype(np.float32))]}),
('SparseGatherV2_0', {
'block': P.SparseGatherV2(),
'desc_const': [0],
'desc_inputs': [[3, 1, 2], Tensor(np.array([0, 1]).astype(np.int32))],
'desc_bprop': [[2, 1, 2]]}),
('Range', {
'block': P.Range(1.0, 5.0),
'desc_inputs': [Tensor(np.ones([10]).astype(np.float32))],