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

[Intel MKL] Fix dequantize accuracy issue and re-enable this OP

Browse Source
This commit is contained in:
Li, Guizi 2020-02-14 13:47:29 +08:00
parent 16d4b320f0
commit 60da3fbda7
7 changed files with 146 additions and 84 deletions

32
tensorflow/core/graph/mkl_layout_pass.cc
View File

@ -359,9 +359,9 @@ class MklLayoutRewritePass : public GraphOptimizationPass {
csinfo_.mul = "Mul";
csinfo_.squared_difference = "SquaredDifference";
csinfo_.sub = "Sub";
// End - element-wise ops. See note above.
// End - element-wise ops. See note above.
// NOTE: names are alphabetically sorted.
// NOTE: names are alphabetically sorted.
rinfo_.push_back({csinfo_.addn, mkl_op_registry::GetMklOpName(csinfo_.addn),
CopyAttrsAll, AlwaysRewrite,
kRewriteForLayoutPropagation});
@ -671,18 +671,18 @@ class MklLayoutRewritePass : public GraphOptimizationPass {
rinfo_.push_back(
{csinfo_.requantize, mkl_op_registry::GetMklOpName(csinfo_.requantize),
CopyAttrsAll, AlwaysRewrite, kRewriteForLayoutPropagation});
// Disable these two MKL operators for now due to some test failures caused
// by these two ops
/*
rinfo_.push_back({csinfo_.tanh,
mkl_op_registry::GetMklOpName(csinfo_.tanh),
CopyAttrsAll, AlwaysRewrite,
kRewriteForLayoutPropagation});
rinfo_.push_back({csinfo_.tanh_grad,
mkl_op_registry::GetMklOpName(csinfo_.tanh_grad),
CopyAttrsAll, AlwaysRewrite,
kRewriteForLayoutPropagation});
*/
// Disable these two MKL operators for now due to some test failures caused
// by these two ops
/*
rinfo_.push_back({csinfo_.tanh,
mkl_op_registry::GetMklOpName(csinfo_.tanh),
CopyAttrsAll, AlwaysRewrite,
kRewriteForLayoutPropagation});
rinfo_.push_back({csinfo_.tanh_grad,
mkl_op_registry::GetMklOpName(csinfo_.tanh_grad),
CopyAttrsAll, AlwaysRewrite,
kRewriteForLayoutPropagation});
*/
rinfo_.push_back(
{csinfo_.reshape, mkl_op_registry::GetMklOpName(csinfo_.reshape),
CopyAttrsAll, AlwaysRewrite, kRewriteForLayoutPropagation});
@ -1478,9 +1478,7 @@ rinfo_.push_back({csinfo_.tanh_grad,
"Eigen op for Dequantize op.";
return false;
}
// TODO(sriniva2/mabuzain) Enable the op after verifying support for
// object detection models
return false;
return true;
}
// Rewrite rule for _FusedMatMul.

2
tensorflow/core/kernels/BUILD
View File

@ -7976,6 +7976,7 @@ tf_cc_test_mkl(
srcs = ["mkl_dequantize_op_test.cc"],
deps = [
":mkl_dequantize_op",
":mkl_tfconv_op",
":ops_testutil",
":ops_util",
"//tensorflow/core:array_ops_op_lib",
@ -7984,6 +7985,7 @@ tf_cc_test_mkl(
"//tensorflow/core:mkl_array_ops_op_lib",
"//tensorflow/core:nn_ops_op_lib",
"//tensorflow/core:protos_all_cc",
"//tensorflow/core:tensorflow",
"//tensorflow/core:test",
"//tensorflow/core:test_main",
"//tensorflow/core:testlib",

16
tensorflow/core/kernels/mkl_dequantize_op.cc
View File

@ -92,10 +92,12 @@ class MklDequantizeOp : public OpKernel {
memory::primitive_desc src_pd =
memory::primitive_desc(src_md, cpu_engine);
memory::desc dst_md = src_mkl_shape.IsMklTensor()
? src_md
: memory::desc(src_dims, MklDnnType<float>(),
memory::format::nhwc);
memory::desc dst_md =
src_mkl_shape.IsMklTensor()
? memory::desc(src_dims, MklDnnType<float>(),
static_cast<memory::format>(src_md.data.format))
: memory::desc(src_dims, MklDnnType<float>(),
memory::format::nhwc);
memory::primitive_desc dst_pd =
memory::primitive_desc(dst_md, cpu_engine);
@ -150,9 +152,9 @@ class MklDequantizeOp : public OpKernel {
mkldnn::reorder(reorder_pd, *src.GetUsrMem(), *dst.GetUsrMem()));
stream(stream::kind::eager).submit(net).wait();
} catch (mkldnn::error& e) {
string error_msg = "Status: " + std::to_string(e.status) +
", message: " + string(e.message) + ", in file " +
string(__FILE__) + ":" + std::to_string(__LINE__);
string error_msg = "Status: " + std::to_string(e.status) + ", message: " +
string(e.message) + ", in file " + string(__FILE__) +
":" + std::to_string(__LINE__);
OP_REQUIRES_OK(
ctx, errors::Aborted("Operation received an exception:", error_msg));
}

81
tensorflow/core/kernels/mkl_dequantize_op_test.cc
View File

@ -22,6 +22,8 @@ limitations under the License.
#include "tensorflow/core/platform/test.h"
#include "tensorflow/core/platform/test_benchmark.h"
#include "tensorflow/core/util/mkl_util.h"
namespace tensorflow {
class MklDequantizeOpTest : public OpsTestBase {};
@ -59,4 +61,83 @@ TEST_F(MklDequantizeOpTest, small) {
test::ExpectTensorNear<float>(expected, output, 0.1);
}
Tensor CreateMklInput() {
MklDnnShape mkl_shape;
memory::desc md =
memory::desc({1, 2, 2, 2}, MklDnnType<uint8>(), memory::format::nhwc);
mkl_shape.SetMklTensor(true);
mkl_shape.SetMklLayout(&md);
mkl_shape.SetElemType(MklDnnType<uint8>());
mkl_shape.SetTfLayout(4, {1, 2, 2, 2}, memory::format::nhwc);
DataType dtype = DataTypeToEnum<uint8>::v();
Tensor mkl_tensor(dtype, {mkl_shape.GetSerializeBufferSize()});
mkl_shape.SerializeMklDnnShape(
mkl_tensor.flat<uint8>().data(),
mkl_tensor.flat<uint8>().size() * sizeof(uint8));
return mkl_tensor;
}
template <typename T>
class CommonTestUtilities : public OpsTestBase {
public:
void MklToTF(const Tensor& tensor, const Tensor& mkl_meta_tensor,
Tensor* output) {
// Create an MKL to TF conversion node and execute it
TF_ASSERT_OK(NodeDefBuilder("mkl_to_tf_op", "_MklToTf")
.Input(FakeInput(DataTypeToEnum<T>::v()))
.Input(FakeInput(DT_UINT8)) // MKL second tensor
.Attr("T", DataTypeToEnum<T>::v())
.Attr("_kernel", "MklLayoutDependentOp")
.Finalize(node_def()));
TF_ASSERT_OK(InitOp());
AddInputFromArray<T>(tensor.shape(), tensor.flat<T>());
AddInputFromArray<uint8>(mkl_meta_tensor.shape(),
mkl_meta_tensor.flat<uint8>());
TF_ASSERT_OK(RunOpKernel());
*output = *GetOutput(0);
}
void ConvertAndCompare(const Tensor& tensor, const Tensor& mkl_meta_tensor,
const Tensor& expected) {
Tensor output;
MklToTF(tensor, mkl_meta_tensor, &output);
test::ExpectTensorNear<T>(expected, output, 0.1);
}
void TestBody() {}
};
TEST_F(MklDequantizeOpTest, MKLInput) {
TF_ASSERT_OK(NodeDefBuilder("dequantize_op", "_MklDequantize")
.Input(FakeInput(DT_QUINT8))
.Input(FakeInput(DT_FLOAT))
.Input(FakeInput(DT_FLOAT))
.Input(FakeInput(DT_UINT8)) // MKL second tensor
.Input(FakeInput(DT_UINT8)) // MKL second tensor
.Input(FakeInput(DT_UINT8)) // MKL second tensor
.Attr("T", DataTypeToEnum<quint8>::v())
.Attr("mode", "SCALED")
.Attr("_kernel", "QuantizedMklOp")
.Finalize(node_def()));
TF_ASSERT_OK(InitOp());
AddInputFromArray<quint8>(TensorShape({1, 2, 2, 2}),
{0, 10, 50, 40, 25, 115, 190, 255});
// min_range = 0
AddInputFromArray<float>(TensorShape({1}), {0});
// max_range = 200
AddInputFromArray<float>(TensorShape({1}), {200.0f});
auto mkl_tensor = CreateMklInput();
AddInputFromArray<uint8>(mkl_tensor.shape(), mkl_tensor.flat<uint8>());
AddInputFromArray<uint8>(dummy_shape, dummy_tensor);
AddInputFromArray<uint8>(dummy_shape, dummy_tensor);
TF_ASSERT_OK(RunOpKernel());
Tensor expected(allocator(), DT_FLOAT, TensorShape({1, 2, 2, 2}));
test::FillValues<float>(&expected,
{0.0, 7.84, 39.21, 31.37, 19.6, 90.2, 149.0, 200});
CommonTestUtilities<float> test_util;
test_util.ConvertAndCompare(*GetOutput(0), *GetOutput(1), expected);
}
} // namespace tensorflow

81
tensorflow/core/kernels/mkl_reshape_op.cc
View File

@ -132,7 +132,7 @@ class MklReshapeOp : public OpKernel {
" values, but the requested shape has ",
shape.num_elements()));
if (input_in_mkl_format) {
if (input_in_mkl_format && !SkipReorder(mkl_shape_input, shape)) {
TensorShape& shape_to = shape;
TensorShape shape_from = mkl_shape_input.GetTfShape();
if (shape_from == shape_to) {
@ -152,65 +152,36 @@ class MklReshapeOp : public OpKernel {
// Tensorflow, we don't need to reorder tensor contents, we just
// need to update MklDnnShape object associated with the input
// tensor to reflect the shape change expected by reshape.
if (!SkipReorder(mkl_shape_input, shape_to)) {
// If dimensions that are being expanded or collapsed are not
// maintained contiguously by MKLDNN, then we use reorder.
// If dimensions that are being expanded or collapsed are not
// maintained contiguously by MKLDNN, then we use reorder.
// Get Mkl layout of input tensor.
auto input_mkl_md = mkl_shape_input.GetMklLayout();
// Set input Mkl layout as the user layout.
dnn_data_input.SetUsrMem(input_mkl_md, &input_tensor);
// Get expected Tensorflow layout of input tensor.
auto output_tf_md = mkl_shape_input.GetTfLayout();
auto output_tf_pd =
memory::primitive_desc(output_tf_md, cpu_engine);
// Get Mkl layout of input tensor.
auto input_mkl_md = mkl_shape_input.GetMklLayout();
// Set input Mkl layout as the user layout.
dnn_data_input.SetUsrMem(input_mkl_md, &input_tensor);
// Get expected Tensorflow layout of input tensor.
auto output_tf_md = mkl_shape_input.GetTfLayout();
auto output_tf_pd = memory::primitive_desc(output_tf_md, cpu_engine);
Tensor* output_tensor = nullptr;
MklDnnShape mkl_shape_output;
mkl_shape_output.SetMklTensor(false);
// We allocate output tensor in the shape expected by Reshape.
AllocateOutputSetMklShape(context, kOutputSlotIdx, &output_tensor,
shape_to, mkl_shape_output);
Tensor* output_tensor = nullptr;
MklDnnShape mkl_shape_output;
mkl_shape_output.SetMklTensor(false);
// We allocate output tensor in the shape expected by Reshape.
AllocateOutputSetMklShape(context, kOutputSlotIdx, &output_tensor,
shape_to, mkl_shape_output);
// Insert reorder between Mkl layout and TensorFlow layout if
// needed. If reorder is not needed but reshape is needed (since
// shape_from != shape_to), then we just copy input tensor to
// output tensor with target shape (we cannot forward Mkl layout
// in such case because shape has changed.)
if (dnn_data_input.CheckReorderToOpMem(output_tf_pd,
output_tensor)) {
} else {
OP_REQUIRES(
context, output_tensor->CopyFrom(input_tensor, shape_to),
errors::InvalidArgument("invalid input tensor shape"));
}
return;
// Insert reorder between Mkl layout and TensorFlow layout if
// needed. If reorder is not needed but reshape is needed (since
// shape_from != shape_to), then we just copy input tensor to
// output tensor with target shape (we cannot forward Mkl layout
// in such case because shape has changed.)
if (dnn_data_input.CheckReorderToOpMem(output_tf_pd, output_tensor)) {
} else {
// If dimensions that are being expanded or collapsed are
// maintained contiguously by MKLDNN, then we skip reorder, just
// update MklDnnShape object for the tensorflow tensor, and forward
// Tensorflow tensor as it is to the output.
auto output_dims = TFShapeToMklDnnDims(shape_to);
auto output_strides = CalculateTFStrides(output_dims);
auto output_tf_md = MklDnnData<T>::CreateBlockedMemDesc(
output_dims, output_strides);
auto output_tf_pd =
memory::primitive_desc(output_tf_md, cpu_engine);
// Set MklDnnShape
MklDnnShape mkl_shape_output;
mkl_shape_output.SetMklTensor(true);
mkl_shape_output.SetMklLayout(&output_tf_pd);
mkl_shape_output.SetElemType(MklDnnType<T>());
mkl_shape_output.SetTfLayout(output_dims.size(), output_dims,
memory::format::blocked);
// We now simply forward input Mkl tensor to output and change its
// output MklDnnShape object.
ForwardMklTensorInToOutWithMklShape(
context, kInputSlotIdx, kOutputSlotIdx, mkl_shape_output);
return;
OP_REQUIRES(context,
output_tensor->CopyFrom(input_tensor, shape_to),
errors::InvalidArgument("invalid input tensor shape"));
}
return;
} catch (mkldnn::error& e) {
string error_msg = "Status: " + std::to_string(e.status) +
", message: " + string(e.message) + ", in file " +

3
tensorflow/core/ops/mkl_array_ops.cc
View File

@ -142,7 +142,10 @@ REGISTER_OP("_MklDequantize")
.Output("output: float")
.Output("mkl_output: uint8")
.Attr("T: quantizedtype")
.Attr("narrow_range: bool = false")
.Attr("axis: int = -1")
.Attr("mode: {'MIN_COMBINED', 'MIN_FIRST', 'SCALED'} = 'SCALED'")
.Attr("dtype: {bfloat16, float} = DT_FLOAT")
.SetShapeFn([](InferenceContext* c) {
TF_RETURN_IF_ERROR(shape_inference::UnchangedShape(c));
ShapeHandle unused;

15
tensorflow/core/util/mkl_util.h
View File

@ -728,9 +728,9 @@ inline Status ConvertMklToTF(OpKernelContext* context,
}
return Status::OK();
} catch (mkldnn::error& e) {
string error_msg = "Status: " + std::to_string(e.status) +
", message: " + string(e.message) + ", in file " +
string(__FILE__) + ":" + std::to_string(__LINE__);
string error_msg = "Status: " + std::to_string(e.status) + ", message: " +
string(e.message) + ", in file " + string(__FILE__) +
":" + std::to_string(__LINE__);
LOG(FATAL) << "Operation received an exception: " << error_msg;
}
}
@ -1011,6 +1011,11 @@ memory::data_type MklDnnType<quint8>() {
return memory::data_type::u8;
}
template <>
memory::data_type MklDnnType<uint8>() {
return memory::data_type::u8;
}
template <>
memory::data_type MklDnnType<qint8>() {
return memory::data_type::s8;
@ -1250,8 +1255,8 @@ inline Status CreateBlockedMemDescHelper(const memory::dims& dim,
} catch (mkldnn::error& e) {
return Status(error::Code::INTERNAL,
tensorflow::strings::StrCat(
"Failed to create blocked memory descriptor.",
"Status: ", e.status, ", message: ", e.message));
"Failed to create blocked memory descriptor.", "Status: ",
e.status, ", message: ", e.message));
}
#else
// We have to construct memory descriptor in a C style. This is not at all