Skip to content

Commit

Permalink
[GPU] Implement bias on internal FC op (openvinotoolkit#23317)
Browse files Browse the repository at this point in the history 
Added bias semantics support for internal FC op

### Details:
 - Added bias semantics support for internal FC op
  • Loading branch information
@dnkurek
dnkurek authored Mar 12, 2024
1 parent 6c55b70 commit ebde6c8
Show file tree
Hide file tree
Showing 13 changed files with 124 additions and 61 deletions.
1 change: 1 addition & 0 deletions src/plugins/intel_gpu/include/intel_gpu/op/fully_connected.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@ class FullyConnected : public ov::op::Op {

FullyConnected(const ov::Output<Node>& A,
const ov::Output<Node>& B,
const ov::Output<Node>& bias,
const ov::element::Type output_type = ov::element::undefined);

bool visit_attributes(ov::AttributeVisitor &visitor) override;
Expand Down
2 changes: 2 additions & 0 deletions src/plugins/intel_gpu/include/intel_gpu/op/fully_connected_compressed.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -18,12 +18,14 @@ class FullyConnectedCompressed : public FullyConnected {

FullyConnectedCompressed(const ov::Output<Node> &A,
const ov::Output<Node> &B,
const ov::Output<Node> &bias,
const ov::Output<Node> &decompression_scale,
const ov::Output<Node> &decompression_zero_point,
const ov::element::Type output_type = ov::element::undefined);

FullyConnectedCompressed(const ov::Output<Node> &A,
const ov::Output<Node> &B,
const ov::Output<Node> &bias,
const ov::Output<Node> &decompression_scale,
const ov::element::Type output_type = ov::element::undefined);

Expand Down
14 changes: 8 additions & 6 deletions src/plugins/intel_gpu/src/plugin/ops/fully_connected.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -26,14 +26,15 @@ namespace ov {
namespace intel_gpu {

static void CreateFullyConnectedCompressedOp(ProgramBuilder& p, const std::shared_ptr<op::FullyConnectedCompressed>& op) {
validate_inputs_count(op, {3, 4});
validate_inputs_count(op, {4, 5});
auto inputs = p.GetInputInfo(op);
std::string primitive_name = layer_type_name_ID(op);

auto input_name = inputs[0].pid;
auto weights_name = inputs[1].pid;
auto scale_name = inputs[2].pid;
auto zp_name = inputs.size() == 4 ? inputs[3].pid : "";
auto bias_name = inputs[2].pid;
auto scale_name = inputs[3].pid;
auto zp_name = inputs.size() == 5 ? inputs[4].pid : "";

float zp_value = 0.0f;
bool has_scalar_zp = false;
Expand All @@ -47,7 +48,7 @@ static void CreateFullyConnectedCompressedOp(ProgramBuilder& p, const std::share
auto fc = cldnn::fully_connected(primitive_name,
cldnn::input_info(input_name),
weights_name,
"",
bias_name,
scale_name,
has_scalar_zp ? "" : zp_name,
cldnn::element_type_to_data_type(op->get_output_element_type(0)),
Expand All @@ -63,12 +64,13 @@ static void CreateFullyConnectedCompressedOp(ProgramBuilder& p, const std::share
}

static void CreateFullyConnectedOp(ProgramBuilder& p, const std::shared_ptr<op::FullyConnected>& op) {
validate_inputs_count(op, {2});
validate_inputs_count(op, {3});
auto inputs = p.GetInputInfo(op);
std::string layerName = layer_type_name_ID(op);

auto input_name = inputs[0].pid;
auto weights_name = inputs[1].pid;
auto bias_name = inputs[2].pid;

auto shape_a = op->get_input_partial_shape(0);
auto shape_b = op->get_input_partial_shape(1);
Expand All @@ -79,7 +81,7 @@ static void CreateFullyConnectedOp(ProgramBuilder& p, const std::shared_ptr<op::
auto fcPrim = cldnn::fully_connected(layerName,
cldnn::input_info(input_name),
weights_name,
"",
bias_name,
cldnn::element_type_to_data_type(op->get_output_element_type(0)),
cldnn::padding(),
rank_a,
Expand Down
7 changes: 6 additions & 1 deletion src/plugins/intel_gpu/src/plugin/transformations/convert_fc_to_compressed.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -59,14 +59,16 @@ ConvertFullyConnectedToFullyConnectedCompressed::ConvertFullyConnectedToFullyCon
auto transpose_m = wrap_type<ov::op::v1::Transpose>({transpose_input, transpose_const_m});

auto data_m = any_input();
auto bias_m = any_input();
auto weights_input_m = std::make_shared<ov::pass::pattern::op::Or>(ov::OutputVector{reshape_m, transpose_m, mul_m});
auto fully_connected_m = wrap_type<op::FullyConnected>({data_m, weights_input_m});
auto fully_connected_m = wrap_type<op::FullyConnected>({data_m, weights_input_m, bias_m});

ov::matcher_pass_callback callback = [=](ov::pass::pattern::Matcher& m) {
const auto& pattern_map = m.get_pattern_value_map();
OPENVINO_ASSERT(pattern_map.count(fully_connected_m));
OPENVINO_ASSERT(pattern_map.count(mul_const_m));
OPENVINO_ASSERT(pattern_map.count(weights_m));
OPENVINO_ASSERT(pattern_map.count(bias_m));
OPENVINO_ASSERT(pattern_map.count(convert_m));
auto fc = std::dynamic_pointer_cast<op::FullyConnected>(pattern_map.at(fully_connected_m).get_node_shared_ptr());
if (!fc || transformation_callback(fc)) {
Expand Down Expand Up @@ -103,6 +105,7 @@ ConvertFullyConnectedToFullyConnectedCompressed::ConvertFullyConnectedToFullyCon
std::shared_ptr<ov::Node> fc_input_b = reshape_const_to_2d(pattern_map.at(weights_m).get_node_shared_ptr());
std::shared_ptr<ov::Node> fc_input_scale = scale;
std::shared_ptr<ov::Node> fc_input_zp = optional_zero_point;
std::shared_ptr<ov::Node> fc_input_bias = pattern_map.at(bias_m).get_node_shared_ptr();
std::vector<std::shared_ptr<ov::Node>> result_nodes = {};
if (has_transpose) {
const auto& transpose = pattern_map.at(transpose_m).get_node_shared_ptr();
Expand All @@ -128,12 +131,14 @@ ConvertFullyConnectedToFullyConnectedCompressed::ConvertFullyConnectedToFullyCon
if (with_zero_point) {
new_fc = std::make_shared<op::FullyConnectedCompressed>(fc_input_a,
fc_input_b,
fc_input_bias,
fc_input_scale,
fc_input_zp,
fc->get_output_type());
} else {
new_fc = std::make_shared<op::FullyConnectedCompressed>(fc_input_a,
fc_input_b,
fc_input_bias,
fc_input_scale,
fc->get_output_type());
}
Expand Down
5 changes: 4 additions & 1 deletion src/plugins/intel_gpu/src/plugin/transformations/convert_matmul_to_fc.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,7 @@
//

#include "intel_gpu/op/fully_connected.hpp"
#include "intel_gpu/op/placeholder.hpp"
#include "convert_matmul_to_fc.hpp"
#include "openvino/op/matmul.hpp"
#include "openvino/op/convert.hpp"
Expand Down Expand Up @@ -177,8 +178,10 @@ ConvertMatMulToFullyConnected::ConvertMatMulToFullyConnected() {
fc_input_b = convert;
}

auto no_bias = std::make_shared<op::Placeholder>();

// Create FullyConnected
auto fc = std::make_shared<op::FullyConnected>(fc_input_a, fc_input_b, matmul->get_output_element_type(0));
auto fc = std::make_shared<op::FullyConnected>(fc_input_a, fc_input_b, no_bias, matmul->get_output_element_type(0));
fc->set_friendly_name(matmul->get_friendly_name());
new_ops.push_back(fc);
ov::copy_runtime_info(matmul, new_ops);
Expand Down
11 changes: 7 additions & 4 deletions src/plugins/intel_gpu/src/plugin/transformations/fc_convert_fusion.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,8 +20,9 @@ FullyConnectedConvertFusion::FullyConnectedConvertFusion() {

auto data = any_input();
auto weights = any_input();
auto fully_connected = wrap_type<op::FullyConnected>({data, weights}, consumers_count(1));
auto fully_connected_compressed = wrap_type<op::FullyConnectedCompressed>({data, weights, any_input(), any_input()}, consumers_count(1));
auto bias = any_input();
auto fully_connected = wrap_type<op::FullyConnected>({data, weights, bias}, consumers_count(1));
auto fully_connected_compressed = wrap_type<op::FullyConnectedCompressed>({data, weights, bias, any_input(), any_input()}, consumers_count(1));
auto fc = std::make_shared<ov::pass::pattern::op::Or>(OutputVector{fully_connected, fully_connected_compressed});
auto convert = wrap_type<ov::op::v0::Convert>({fc}, type_matches(element::f32));

Expand All @@ -30,6 +31,7 @@ FullyConnectedConvertFusion::FullyConnectedConvertFusion() {

const auto& m_data = pattern_map.at(data).get_node_shared_ptr();
const auto& m_weights = pattern_map.at(weights).get_node_shared_ptr();
const auto& m_bias = pattern_map.at(bias).get_node_shared_ptr();
const auto& m_convert = pattern_map.at(convert).get_node_shared_ptr();
auto output_type = m_convert->get_output_element_type(0);

Expand All @@ -38,13 +40,14 @@ FullyConnectedConvertFusion::FullyConnectedConvertFusion() {
auto it = pattern_map.find(fully_connected);
if (it != pattern_map.end()) {
m_fc = it->second.get_node_shared_ptr();
new_fc = std::make_shared<op::FullyConnected>(m_data, m_weights, output_type);
new_fc = std::make_shared<op::FullyConnected>(m_data, m_weights, m_bias, output_type);
} else {
m_fc = pattern_map.at(fully_connected_compressed).get_node_shared_ptr();
new_fc = std::make_shared<op::FullyConnectedCompressed>(m_data,
m_weights,
m_fc->input_value(2),
m_bias,
m_fc->input_value(3),
m_fc->input_value(4),
output_type);
}
new_fc->set_friendly_name(m_convert->get_friendly_name());
Expand Down
2 changes: 1 addition & 1 deletion src/plugins/intel_gpu/src/plugin/transformations/move_fc_reshape_to_weights.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ MoveFCReshapeToWeights::MoveFCReshapeToWeights() {
auto weights_input_m = std::make_shared<ov::pass::pattern::op::Or>(ov::OutputVector{reshape_m, transpose_m});

auto data_m = any_input();
auto fully_connected_m = wrap_type<op::FullyConnected>({data_m, weights_input_m});
auto fully_connected_m = wrap_type<op::FullyConnected>({data_m, weights_input_m, any_input()});

ov::matcher_pass_callback callback = [&](ov::pass::pattern::Matcher& m) {
const auto fully_connected = m.get_match_root();
Expand Down
9 changes: 5 additions & 4 deletions src/plugins/intel_gpu/src/plugin/transformations/op/fully_connected.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -11,24 +11,25 @@ namespace op {

FullyConnected::FullyConnected(const ov::Output<Node>& A,
const ov::Output<Node>& B,
const ov::Output<Node>& bias,
const ov::element::Type output_type)
: Op({A, B}), m_output_type(output_type) {
: Op({A, B, bias}), m_output_type(output_type) {
validate_and_infer_types();
}

std::shared_ptr<ov::Node> FullyConnected::clone_with_new_inputs(const ov::OutputVector& new_args) const {
check_new_args_count(this, new_args);

return std::make_shared<FullyConnected>(new_args.at(0), new_args.at(1), m_output_type);
return std::make_shared<FullyConnected>(new_args.at(0), new_args.at(1), new_args.at(2), m_output_type);
}

void FullyConnected::validate_and_infer_types() {
const auto input_size = get_input_size();
NODE_VALIDATION_CHECK(this,
input_size >= 2,
input_size >= 3,
"Number of inputs is incorrect. Current value is: ",
input_size,
", expected at least 2.");
", expected at least 3.");

ov::op::v0::MatMul op;
op.set_transpose_a(false);
Expand Down
18 changes: 11 additions & 7 deletions src/plugins/intel_gpu/src/plugin/transformations/op/fully_connected_compressed.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,37 +10,41 @@ namespace op {

FullyConnectedCompressed::FullyConnectedCompressed(const ov::Output<Node>& A,
const ov::Output<Node>& B,
const ov::Output<Node>& bias,
const ov::Output<Node>& decompression_scale,
const ov::Output<Node>& decompression_zero_point,
const ov::element::Type output_type)
: FullyConnected(A, B, output_type) {
set_argument(2, decompression_scale);
set_argument(3, decompression_zero_point);
: FullyConnected(A, B, bias, output_type) {
set_argument(3, decompression_scale);
set_argument(4, decompression_zero_point);
validate_and_infer_types();
}

FullyConnectedCompressed::FullyConnectedCompressed(const ov::Output<Node>& A,
const ov::Output<Node>& B,
const ov::Output<Node>& bias,
const ov::Output<Node>& decompression_scale,
const ov::element::Type output_type)
: FullyConnected(A, B, output_type) {
set_argument(2, decompression_scale);
: FullyConnected(A, B, bias, output_type) {
set_argument(3, decompression_scale);
validate_and_infer_types();
}

std::shared_ptr<ov::Node> FullyConnectedCompressed::clone_with_new_inputs(const ov::OutputVector& new_args) const {
check_new_args_count(this, new_args);

if (new_args.size() == 3)
if (new_args.size() == 4)
return std::make_shared<FullyConnectedCompressed>(new_args.at(0),
new_args.at(1),
new_args.at(2),
new_args.at(3),
m_output_type);
else if (new_args.size() == 4)
else if (new_args.size() == 5)
return std::make_shared<FullyConnectedCompressed>(new_args.at(0),
new_args.at(1),
new_args.at(2),
new_args.at(3),
new_args.at(4),
m_output_type);
else
OPENVINO_THROW("Unexpected inputs count for FullyConnectedCompressed op: ", new_args.size());
Expand Down
Loading

0 comments on commit ebde6c8

Please sign in to comment.