Skip to content

Vishakh1/round mode br #353

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 8 commits into from
Dec 11, 2019
Merged

Vishakh1/round mode br #353

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
67472ee
Added TF rounding modes
vishakha-nervana Nov 26, 2019
266b090
Apply suggestions from code review
vishakha-nervana Nov 26, 2019
d59dc9b
Update test/opexecuter.cpp
vishakha-nervana Nov 26, 2019
5ce7c5f
Update test/opexecuter.cpp
vishakha-nervana Nov 26, 2019
d676780
Merge branch 'r0.19' into vishakh1/round_mode_br
sayantan-nervana Dec 3, 2019
1fb9b51
Merge branch 'r0.19' into vishakh1/round_mode_br
sayantan-nervana Dec 4, 2019
943e99c
Merge branch 'r0.19' into vishakh1/round_mode_br
vishakha-nervana Dec 6, 2019
8571218
Merge branch 'r0.19' into vishakh1/round_mode_br
sayantan-nervana Dec 10, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
28 changes: 22 additions & 6 deletions ngraph_bridge/ngraph_builder.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3327,8 +3327,17 @@ static Status TranslateQuantizeAndDequantizeV2Op(
op->name(), ng_r_et, ng::Shape(), std::vector<float>({scale}));
auto ng_offset = ConstructNgNode<ng::op::Constant>(
op->name(), ng_q_et, ng::Shape(), std::vector<int>({0}));
ng::op::Quantize::RoundMode ng_round_mode =
ng::op::Quantize::RoundMode::ROUND_NEAREST_TOWARD_INFINITY;
ng::op::Quantize::RoundMode ng_round_mode;
string round_mode_string;
TF_RETURN_IF_ERROR(
GetNodeAttr(op->attrs(), "round_mode", &round_mode_string));
if (round_mode_string == "HALF_UP") {
ng_round_mode = ng::op::Quantize::RoundMode::ROUND_NEAREST_UPWARD;
} else if (round_mode_string == "HALF_TO_EVEN") {
ng_round_mode = ng::op::Quantize::RoundMode::ROUND_NEAREST_TOWARD_EVEN;
} else {
return errors::Internal("Tensorflow Rounding Mode not supported by Ngraph");
}
auto ng_quant = ConstructNgNode<ng::op::Quantize>(
op->name(), ng_input, ng_scale, ng_offset, ng_q_et, ng::AxisSet(),
ng_round_mode);
Expand Down Expand Up @@ -3608,10 +3617,17 @@ static Status TranslateQuantizeV2Op(const Node* op,
ng::element::Type ng_et;
TF_RETURN_IF_ERROR(TFDataTypeToNGraphElementType(dtype, &ng_et));

// TODO: Only RoundMode = ROUND_NEAREST_TOWARD_EVEN is supported, for now.
// Support other modes later
ng::op::Quantize::RoundMode ng_round_mode =
ng::op::Quantize::RoundMode::ROUND_NEAREST_TOWARD_EVEN;
ng::op::Quantize::RoundMode ng_round_mode;
string round_mode_string;
TF_RETURN_IF_ERROR(
GetNodeAttr(op->attrs(), "round_mode", &round_mode_string));
if (round_mode_string == "HALF_UP") {
ng_round_mode = ng::op::Quantize::RoundMode::ROUND_NEAREST_UPWARD;
} else if (round_mode_string == "HALF_TO_EVEN") {
ng_round_mode = ng::op::Quantize::RoundMode::ROUND_NEAREST_TOWARD_EVEN;
} else {
return errors::Internal("Tensorflow Rounding Mode not supported by Ngraph");
}

auto ng_node = ng::builder::ScaledQuantize(ng_input, ng_min, ng_max, ng_et,
ng::AxisSet(), ng_round_mode);
Expand Down
56 changes: 55 additions & 1 deletion test/test_array_ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -797,7 +797,61 @@ TEST(ArrayOps, QuantizeAndDequantizeV2x8xtruexfalse) {
output_datatypes, sess_run_fetchoutputs);

opexecuter.RunTest();
} // end of test op QuantizeAndDequantizeV2x8xtruexfalse
}

TEST(ArrayOps, QuantizeAndDequantizeV2RoundingMode1) {
Scope root = Scope::NewRootScope();
int dim1 = 2;
int dim2 = 3;

Tensor A(DT_FLOAT, TensorShape({dim1, dim2}));
AssignInputValues<float>(A, {0.9, 3.4, 2.6, 5.4, 4.2, 4.5});

auto attrs = ops::QuantizeAndDequantizeV2::Attrs();
attrs.num_bits_ = 8;
attrs.range_given_ = true;
attrs.signed_input_ = true;
attrs.round_mode_ = "HALF_UP";

vector<int> static_input_indexes = {1, 2};
ops::QuantizeAndDequantizeV2 R =
ops::QuantizeAndDequantizeV2(root, A, 0.0f, 127.0f, attrs);

vector<DataType> output_datatypes = {DT_FLOAT};

std::vector<Output> sess_run_fetchoutputs = {R.output};
OpExecuter opexecuter(root, "QuantizeAndDequantizeV2", static_input_indexes,
output_datatypes, sess_run_fetchoutputs);

opexecuter.RunTest();
}

TEST(ArrayOps, QuantizeAndDequantizeV2RoundingMode2) {
Scope root = Scope::NewRootScope();
int dim1 = 2;
int dim2 = 3;

Tensor A(DT_FLOAT, TensorShape({dim1, dim2}));
AssignInputValues<float>(A, {0.9, 3.4, 2.6, 5.4, 4.2, 4.5});

auto attrs = ops::QuantizeAndDequantizeV2::Attrs();
attrs.num_bits_ = 8;
attrs.range_given_ = true;
attrs.signed_input_ = true;
attrs.round_mode_ = "HALF_TO_EVEN";

vector<int> static_input_indexes = {1, 2};
ops::QuantizeAndDequantizeV2 R =
ops::QuantizeAndDequantizeV2(root, A, 0.0f, 127.0f, attrs);

vector<DataType> output_datatypes = {DT_FLOAT};

std::vector<Output> sess_run_fetchoutputs = {R.output};
OpExecuter opexecuter(root, "QuantizeAndDequantizeV2", static_input_indexes,
output_datatypes, sess_run_fetchoutputs);

opexecuter.RunTest();
} // end of test op QuantizeAndDequantizeV2x8xtruextrue

// CPU only supports QuantizedConcat with DT_QINT32 and DT_QUINT8
TEST(ArrayOps, QuantizedConcat) {
Expand Down