m53297601
/
mindspore
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

!12786 adapt some 3d ops.

Browse Source 
From: @liu_xiao_93
Reviewed-by: @liangchenghui,@wuxuejian
Signed-off-by: @liangchenghui
pull/12786/MERGE
mindspore-ci-bot 4 years ago committed by Gitee
parent 7f71525b0a 6459229fff
commit ccc343f8da
9 changed files with 26 additions and 11 deletions
Show Stats Download Patch File Download Diff File

4
mindspore/ccsrc/backend/kernel_compiler/tbe/tbe_kernel_build.cc
Unescape View File

@ -99,8 +99,8 @@ constexpr auto kJDynamicIndex = "dynamic_index";
bool IsNeedChangeDefaultFormat(const CNodePtr &cnode) { bool IsNeedChangeDefaultFormat(const CNodePtr &cnode) {
MS_EXCEPTION_IF_NULL(cnode); MS_EXCEPTION_IF_NULL(cnode);
if (AnfAlgo::HasNodeAttr("io_format", cnode->cast<CNodePtr>())) { if (AnfAlgo::HasNodeAttr(kAttrFormat, cnode->cast<CNodePtr>())) {
auto attr = AnfAlgo::GetNodeAttr<std::string>(cnode, "io_format"); auto attr = AnfAlgo::GetNodeAttr<std::string>(cnode, kAttrFormat);
return attr == kOpFormat_NCDHW; return attr == kOpFormat_NCDHW;
} }
return false; return false;

2
mindspore/ccsrc/backend/optimizer/ascend/enhancer/add_attr_for_3d_graph.cc
Unescape View File

@ -39,7 +39,7 @@ const AnfNodePtr AddIoFormatAttrFor3DGraph::Process(const FuncGraphPtr &func_gra
auto formats = AnfAlgo::GetAllOutputFormats(node); auto formats = AnfAlgo::GetAllOutputFormats(node);
if (std::any_of(formats.begin(), formats.end(), if (std::any_of(formats.begin(), formats.end(),
[](const std::string &format) { return k3DFormatSet.find(format) != k3DFormatSet.end(); })) { [](const std::string &format) { return k3DFormatSet.find(format) != k3DFormatSet.end(); })) {
AnfAlgo::SetNodeAttr("io_format", MakeValue(kOpFormat_NCDHW), node); AnfAlgo::SetNodeAttr(kAttrFormat, MakeValue(kOpFormat_NCDHW), node);
} }
return node; return node;
} }

8
mindspore/ops/_op_impl/tbe/apply_momentum.py
Unescape View File

@ -47,6 +47,14 @@ apply_momentum_op_info = TBERegOp("ApplyMomentum") \
DataType.F32_Default, DataType.F32_C1HWNCoC0, DataType.F32_C1HWNCoC0) \ DataType.F32_Default, DataType.F32_C1HWNCoC0, DataType.F32_C1HWNCoC0) \
.dtype_format(DataType.F32_FracZ, DataType.F32_FracZ, DataType.F32_Default, DataType.F32_FracZ, .dtype_format(DataType.F32_FracZ, DataType.F32_FracZ, DataType.F32_Default, DataType.F32_FracZ,
DataType.F32_Default, DataType.F32_FracZ, DataType.F32_FracZ) \ DataType.F32_Default, DataType.F32_FracZ, DataType.F32_FracZ) \
.dtype_format(DataType.F16_NDC1HWC0, DataType.F16_NDC1HWC0, DataType.F16_Default, DataType.F16_NDC1HWC0,
DataType.F16_Default, DataType.F16_NDC1HWC0, DataType.F16_NDC1HWC0) \
.dtype_format(DataType.F16_FRACTAL_Z_3D, DataType.F16_FRACTAL_Z_3D, DataType.F16_Default, DataType.F16_FRACTAL_Z_3D,
DataType.F16_Default, DataType.F16_FRACTAL_Z_3D, DataType.F16_FRACTAL_Z_3D) \
.dtype_format(DataType.F32_NDC1HWC0, DataType.F32_NDC1HWC0, DataType.F32_Default, DataType.F32_NDC1HWC0,
DataType.F32_Default, DataType.F32_NDC1HWC0, DataType.F32_NDC1HWC0) \
.dtype_format(DataType.F32_FRACTAL_Z_3D, DataType.F32_FRACTAL_Z_3D, DataType.F32_Default, DataType.F32_FRACTAL_Z_3D,
DataType.F32_Default, DataType.F32_FRACTAL_Z_3D, DataType.F32_FRACTAL_Z_3D) \
.get_op_info() .get_op_info()

2
mindspore/ops/_op_impl/tbe/sigmoid_cross_entropy_with_logits.py
Unescape View File

@ -30,6 +30,8 @@ sigmoid_cross_entropy_with_logits_op_info = TBERegOp("SigmoidCrossEntropyWithLog
.dtype_format(DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD) \ .dtype_format(DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD) \
.dtype_format(DataType.F32_Default, DataType.F32_Default, DataType.F32_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_5HD, DataType.F32_5HD, DataType.F32_5HD) \
.dtype_format(DataType.F16_NDC1HWC0, DataType.F16_NDC1HWC0, DataType.F16_NDC1HWC0) \
.dtype_format(DataType.F32_NDC1HWC0, DataType.F32_NDC1HWC0, DataType.F32_NDC1HWC0) \
.get_op_info() .get_op_info()

2
mindspore/ops/_op_impl/tbe/sigmoid_cross_entropy_with_logits_grad.py
Unescape View File

@ -29,8 +29,10 @@ sigmoid_cross_entropy_with_logits_grad_op_info = TBERegOp("SigmoidCrossEntropyWi
.output(0, "gradient", False, "required", "all") \ .output(0, "gradient", False, "required", "all") \
.dtype_format(DataType.F16_Default, DataType.F16_Default, DataType.F16_Default, DataType.F16_Default) \ .dtype_format(DataType.F16_Default, DataType.F16_Default, DataType.F16_Default, DataType.F16_Default) \
.dtype_format(DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD) \ .dtype_format(DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD) \
.dtype_format(DataType.F16_NDC1HWC0, DataType.F16_NDC1HWC0, DataType.F16_NDC1HWC0, DataType.F16_NDC1HWC0) \
.dtype_format(DataType.F32_Default, DataType.F32_Default, DataType.F32_Default, DataType.F32_Default) \ .dtype_format(DataType.F32_Default, DataType.F32_Default, DataType.F32_Default, DataType.F32_Default) \
.dtype_format(DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD) \ .dtype_format(DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD) \
.dtype_format(DataType.F32_NDC1HWC0, DataType.F32_NDC1HWC0, DataType.F32_NDC1HWC0, DataType.F32_NDC1HWC0) \
.get_op_info() .get_op_info()

8
mindspore/ops/_op_impl/tbe/strided_slice_d.py
Unescape View File

@ -33,12 +33,8 @@ strided_slice_d_op_info = TBERegOp("StridedSlice") \
.attr("shrink_axis_mask", "required", "int", "all") \ .attr("shrink_axis_mask", "required", "int", "all") \
.input(0, "x", False, "required", "all") \ .input(0, "x", False, "required", "all") \
.output(0, "y", False, "required", "all") \ .output(0, "y", False, "required", "all") \
.dtype_format(DataType.BOOL_Default, DataType.BOOL_Default) \ .op_pattern("dynamicFormat") \
.dtype_format(DataType.I8_Default, DataType.I8_Default) \ .dtype_format(DataType.None_None, DataType.None_None) \
.dtype_format(DataType.U8_Default, DataType.U8_Default) \
.dtype_format(DataType.I32_Default, DataType.I32_Default) \
.dtype_format(DataType.F16_Default, DataType.F16_Default) \
.dtype_format(DataType.F32_Default, DataType.F32_Default) \
.get_op_info() .get_op_info()

2
mindspore/ops/_op_impl/tbe/trans_data.py
Unescape View File

@ -151,6 +151,8 @@ trans_data_op_info = TBERegOp("TransData") \
.dtype_format(DataType.F32_FRACTAL_Z_3D, DataType.F32_DHWCN) \ .dtype_format(DataType.F32_FRACTAL_Z_3D, DataType.F32_DHWCN) \
.dtype_format(DataType.F32_NDC1HWC0, DataType.F32_NDHWC) \ .dtype_format(DataType.F32_NDC1HWC0, DataType.F32_NDHWC) \
.dtype_format(DataType.F32_NDHWC, DataType.F32_NDC1HWC0) \ .dtype_format(DataType.F32_NDHWC, DataType.F32_NDC1HWC0) \
.dtype_format(DataType.F32_NDC1HWC0, DataType.F32_NCDHW) \
.dtype_format(DataType.F32_NCDHW, DataType.F32_NDC1HWC0) \
.get_op_info() .get_op_info()

2
mindspore/ops/operations/nn_ops.py
Unescape View File

@ -3783,7 +3783,7 @@ class BCEWithLogitsLoss(PrimitiveWithInfer):
for i, v in enumerate(reversed_pos_shape): for i, v in enumerate(reversed_pos_shape):
if v not in (reversed_target[i], 1): if v not in (reversed_target[i], 1):
raise ValueError(f"For {self.name}, shapes can not broadcast. " raise ValueError(f"For {self.name}, shapes can not broadcast. "
f"predict: {tuple(predict)}, weight shape {tuple(pos_weight)}.") f"predict: {tuple(predict)}, pos_weight shape {tuple(pos_weight)}.")
if self.reduction in ('mean', 'sum'): if self.reduction in ('mean', 'sum'):
shape = [] shape = []

7
tests/st/ops/ascend/test_dynamic_gru_v2.py
Unescape View File

@ -13,6 +13,7 @@
# limitations under the License. # limitations under the License.
# ============================================================================ # ============================================================================
import numpy as np import numpy as np
import pytest
import mindspore.context as context import mindspore.context as context
import mindspore.nn as nn import mindspore.nn as nn
@ -31,6 +32,10 @@ class DynamicGRUV2(nn.Cell):
return self.dynamic_gru(x, weight_i, weight_h, bias_i, bias_h, None, init_h) return self.dynamic_gru(x, weight_i, weight_h, bias_i, bias_h, None, init_h)
@pytest.mark.level0
@pytest.mark.env_onecard
@pytest.mark.platform_arm_ascend_training
@pytest.mark.platform_x86_ascend_training
def test_dynamic_gru_v2(): def test_dynamic_gru_v2():
x = Tensor(np.random.rand(2, 8, 64).astype(np.float16)) x = Tensor(np.random.rand(2, 8, 64).astype(np.float16))
weight_i = Tensor(np.random.rand(64, 48).astype(np.float16)) weight_i = Tensor(np.random.rand(64, 48).astype(np.float16))
@ -40,4 +45,4 @@ def test_dynamic_gru_v2():
init_h = Tensor(np.random.rand(8, 16).astype(np.float16)) init_h = Tensor(np.random.rand(8, 16).astype(np.float16))
gru_net = DynamicGRUV2() gru_net = DynamicGRUV2()
output = gru_net(x, weight_i, weight_h, bias_i, bias_h, init_h) output = gru_net(x, weight_i, weight_h, bias_i, bias_h, init_h)
print(output) assert output[0].shape == (2, 8, 16)

Write Preview
Loading…
Cancel
Save