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

[GPU] Add reorder if the input format of eltwise is different from the output format (#19068)

Browse Source 
* [GPU] Add reorder if the input format of eltwise is different from the output format

* Fix failed TCs for eltwise's mixed layout
This commit is contained in:
Andrew Kwangwoong Park 2023-08-09 15:21:54 +09:00 committed by GitHub
parent e60da26894
commit df659daf44
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 53 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/plugins/intel_gpu/src/graph/graph_optimizer/add_required_reorders.cpp
View File

@ -76,14 +76,15 @@ void add_required_reorders::run(program& p) {
}
}
if (usr->is_type<eltwise>() && usr->is_in_shape_of_subgraph()) {
if (usr->is_type<eltwise>()) {
for (size_t i = 0; i < usr->get_dependencies().size(); i++) {
auto& dep = usr->get_dependency(i);
if (!dep.is_in_data_flow() || dep.is_constant())
continue;
auto dep_layout = dep.get_output_layout();
auto out_layout = usr->get_output_layout();
bool required_reorder = out_layout.data_type != dep_layout.data_type;
bool required_reorder = (format::dimension(out_layout.format) != format::dimension(dep_layout.format)) ||
(usr->is_in_shape_of_subgraph() && (out_layout.data_type != dep_layout.data_type));
if (required_reorder) {
auto new_reorder = std::make_shared<reorder>(dep.id() + "_reorder_" + usr->id(), dep.id(), out_layout.format, out_layout.data_type);
auto& new_reorder_node = p.get_or_create(new_reorder);

50
src/plugins/intel_gpu/tests/unit/passes/add_required_reorders_test.cpp
View File

@ -17,6 +17,8 @@
#include "permute_inst.h"
#include "reshape_inst.h"
#include "shape_of_inst.h"
#include "convolution_inst.h"
#include "dft_inst.h"
#include "to_string_utils.h"
#include "program_wrapper.h"
@ -66,6 +68,54 @@ TEST(add_required_reorders, input_reorder_inside_shape_of_subgraph) {
ASSERT_EQ(eltwise_in_layout.data_type, data_types::f32);
}
TEST(add_required_reorders, eltwise_input_reorder) {
// Topology:
// (bfyx)input weights
// \ /
// conv1 irdft_input(bfzyx)
// / | \ /
// (bfzyx)rdft | irdft(bfyx)
// | /
// eltwise
// |
// conv2
//
// Expectation:
// The selected format of eltwise should be selected as bfzyx (reorder_inputs)
// A new reorder that converts from b_fs_yx_fsv16 to bfzyx should be inserted after convolution (reorder_inputs)
// If the input format of eltwise is different from the output format, reorder(bfyx->bfzyx) should be added (add_required_reorders)
auto& engine = get_test_engine();
auto conv1_weight_mem = engine.allocate_memory(layout{ { 192, 384, 1, 1 }, data_types::f16, format::bfyx });
auto conv2_weight_mem = engine.allocate_memory(layout{ { 384, 192, 1, 1 }, data_types::f16, format::bfyx });
topology topology;
topology.add(data("conv1_weights", conv1_weight_mem));
topology.add(data("conv2_weights", conv2_weight_mem));
topology.add(input_layout("input", layout{ { 1, 384, 36, 64 }, data_types::f16, format::bfyx }));
topology.add(input_layout("irdft_input", layout{ { 1, 192, 36, 33, 2 }, data_types::f16, format::bfzyx }));
topology.add(convolution("conv1", input_info("input"), "conv1_weights", "", 1, {1, 1}, {1, 1}, {0, 0}, {0, 0}, false));
topology.add(dft("rdft", input_info("conv1"), {2, 3}, {36, 64}, {1, 192, 36, 33, 2}, dft_direction::forward, dft_mode::real));
topology.add(dft("irdft", input_info("irdft_input"), {2, 3}, {36, 64}, {1, 192, 36, 64}, dft_direction::inverse, dft_mode::real));
topology.add(eltwise("eltwise", input_info("conv1"), input_info("irdft"), eltwise_mode::sum));
topology.add(convolution("conv2", input_info("eltwise"), "conv2_weights", "", 1, {1, 1}, {1, 1}, {0, 0}, {0, 0}, false));
ExecutionConfig config = get_test_default_config(engine);
ov::intel_gpu::ImplementationDesc conv1_impl = { format::b_fs_yx_fsv16, "" };
ov::intel_gpu::ImplementationDesc conv2_impl = { format::b_fs_yx_fsv16, "" };
config.set_property(ov::intel_gpu::force_implementations(ov::intel_gpu::ImplForcingMap{ { "conv1", conv1_impl }, { "conv2", conv2_impl } }));
config.set_property(ov::intel_gpu::optimize_data(true));
program::ptr prog = nullptr;
ASSERT_NO_THROW(prog = program::build_program(engine, topology, config));
ASSERT_NE(prog, nullptr);
auto prog_impl = prog.get();
auto& eltwise_node = prog_impl->get_node("eltwise");
ASSERT_EQ(eltwise_node.get_input_layouts()[1].format, format::bfzyx);
ASSERT_EQ(eltwise_node.get_output_layout().format, format::bfzyx);
}
TEST(add_required_reorders, skip_adding_reorder_batch_axis_padding) {
auto& engine = get_test_engine();