experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix code format

Browse Source
This commit is contained in:
yunfeima 2020-12-23 16:47:36 +08:00
parent dfb45cd2ea
commit 04e7e1da0a
4 changed files with 3 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tensorflow/core/grappler/optimizers/remapper.cc
View File

@ -472,7 +472,7 @@ bool FindContractionWithBiasAndActivation(
if (!IsMatMul(*contraction_node_def) && IsTanh(*node_def)) return false;
// Currently, only (conv | matmul) + bias + leakyrelu is enabled
if ((!IsConv2D(*contraction_node_def) && !IsMatMul(*contraction_node_def)) &&
if (!(IsConv2D(*contraction_node_def) || IsMatMul(*contraction_node_def)) &&
IsLeakyRelu(*node_def))
return false;

1
tensorflow/core/grappler/optimizers/remapper_test.cc
View File

@ -712,7 +712,6 @@ class RemapperFuseMatMulWithBiasAndActivationTest : public RemapperTest {
if (activation == "LeakyRelu") {
EXPECT_EQ(node.attr().at("leakyrelu_alpha").f(), leakyrelu_alpha);
}
found++;
}
}

3
tensorflow/core/kernels/matmul_op_fused.cc
View File

@ -108,8 +108,7 @@ struct LaunchFusedMatMulOp<CPUDevice, T> {
executeWithOutputKernel(WithBiasAddAndElu<T>(bias_add_args));
break;
case FusedComputationType::kBiasAddWithLeakyRelu:
out.device(d) = lhs.contract(rhs, dim_pair,
WithBiasAddAndLeakyRelu<T>(bias_add_args));
executeWithOutputKernel(WithBiasAddAndLeakyRelu<T>(bias_add_args));
break;
case FusedComputationType::kUndefined:
OP_REQUIRES_OK(context, errors::Internal("Fusion type is undefined"));

2
tensorflow/core/ops/math_ops.cc
View File

@ -954,7 +954,7 @@ REGISTER_OP("_FusedMatMul")
.Attr("fused_ops: list(string) = []")
// Attributes for the FusedBatchNorm ----------- //
.Attr("epsilon: float = 0.0001")
// Attributes for the LeakyRelu ----------------------------------------- //
// Attributes for the LeakyRelu ---------------- //
.Attr("leakyrelu_alpha: float = 0.2")
// --------------------------------------------- //
.SetShapeFn(shape_inference::MatMulShape)