IntenseWebs/openvino
3
0 Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GPU] Fix FC 3D input size propagation and bias fusion (#10249)

Browse Source
This commit is contained in:
Sergey Shlyapnikov authored and GitHub committed 2022-02-10 16:16:12 +03:00
1 parent d94cff59a3
commit bc21e52912
2 files changed
+22 -3
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/plugins/intel_gpu/src/graph/graph_optimizer/prepare_primitive_fusing.cpp
+17 -1
View File
@@ -285,8 +285,21 @@ void prepare_primitive_fusing::fuse_bias(program &p) {
if (!is_bias)
continue;
auto is_3d_fully_connected = [](program_node& node) {
if (!node.is_type<fully_connected>())
return false;
return node.as<fully_connected>().get_primitive()->input_size == 3;
};
size_t out_features = static_cast<size_t>(node->get_output_layout().size.feature[0]);
// Change out_features value to proper dimension for 3D FC case
if (is_3d_fully_connected(node->get_dependency(0)))
out_features = static_cast<size_t>(node->get_dependency(0).get_output_layout().size.spatial[1]);
else if (is_3d_fully_connected(node->get_dependency(1)))
out_features = static_cast<size_t>(node->get_dependency(1).get_output_layout().size.spatial[1]);
int bias_idx = -1;
for (size_t i = 0; i < eltw_node.get_dependencies().size(); i++) {
auto& dep = eltw_node.get_dependency(i);
@@ -448,7 +461,10 @@ void prepare_primitive_fusing::fuse_bias(program &p) {
desc->input[0],
desc->weights,
bias_name,
fc.get_output_layout().data_type);
fc.get_output_layout().data_type,
desc->ext_prim_id,
desc->output_padding,
desc->input_size);
auto& new_fc_node = p.get_or_create(fc_with_bias_prim);
fuse_bias_f(fc, new_fc_node, bias_node, eltw_node);
src/plugins/intel_gpu/tests/fusions/fully_connected_fusion_test.cpp
+5 -2
View File
@@ -84,7 +84,7 @@ public:
layout get_bias_layout(fully_connected_test_params& p) {
if (p.out_shape.spatial[1] > 1) {
// 3d case
return layout{ p.default_type, format::bfyx, tensor{ 1, p.out_shape.spatial[1], 1, 1 } };
return layout{ p.default_type, format::bfyx, tensor{ 1, 1, 1, p.out_shape.spatial[1] } };
}
else {
return layout{ p.default_type, format::bfyx, tensor{ 1, p.out_shape.feature[0], 1, 1 } };
@@ -207,7 +207,7 @@ TEST_P(fc_fp32_bias, basic) {
input_layout("input", get_input_layout(p)),
data("weights", get_mem(get_weights_layout(p))),
data("bias", get_mem(get_bias_layout(p))),
fully_connected("fc_prim", "input", "weights", ""),
fully_connected("fc_prim", "input", "weights", "", "", padding(), get_output_dim_size(p)),
eltwise("bias_add", { "fc_prim", "bias" }, eltwise_mode::sum),
reorder("reorder_bfyx", "bias_add", p.default_format, data_types::f32)
);
@@ -220,6 +220,9 @@ INSTANTIATE_TEST_SUITE_P(fusings_gpu, fc_fp32_bias, ::testing::ValuesIn(std::vec
fully_connected_test_params{ CASE_FC_FP32_1, 2, 3 },
fully_connected_test_params{ CASE_FC_FP32_2, 2, 3 },
fully_connected_test_params{ CASE_FC_FP32_3, 2, 3 },
fully_connected_test_params{ CASE_FC_FP32_3D_1, 2, 3 },
fully_connected_test_params{ CASE_FC_FP32_3D_2, 2, 3 },
fully_connected_test_params{ CASE_FC_FP32_3D_3, 2, 3 },
}));
class fc_int8_scale : public FullyConnectedFusingTest {};