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

[GPU] Memory reuse false for dynamic and null impl case (#19354)

Browse Source
This commit is contained in:
Kelvin Choi 2023-08-24 17:32:07 +09:00 committed by GitHub
parent c89b9edfe7
commit ce47522165
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 54 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/plugins/intel_gpu/src/graph/primitive_inst.cpp
View File

@ -1229,6 +1229,9 @@ static bool user_requesting_mem_reuse_false(const program_node& node) {
if ((user->get_selected_impl() != nullptr) && (user->get_selected_impl()->can_reuse_memory == false)) {
return true;
} else if (user->get_selected_impl() == nullptr) {
if (user->is_dynamic()) {
return true;
}
if (user_requesting_mem_reuse_false(*user)) {
return true;
}

51
src/plugins/intel_gpu/tests/unit/test_cases/memory_test.cpp
View File

@ -14,6 +14,7 @@
#include <intel_gpu/primitives/reshape.hpp>
#include <intel_gpu/primitives/crop.hpp>
#include <intel_gpu/primitives/eltwise.hpp>
#include <intel_gpu/primitives/grid_sample.hpp>
#include <fully_connected_inst.h>
using namespace cldnn;
@ -595,6 +596,52 @@ public:
std::static_pointer_cast<fully_connected_inst>(network.get_primitive("relu2"))->output_memory_ptr()->buffer_ptr());
}
}
void test_dynamic_mem_reuse_for_null_sel_impl() {
auto& engine = get_test_engine();
const int32_t input_f = 3;
auto input_static_layout = layout{ ov::PartialShape{ 2, input_f, 1, 2 }, data_types::f32, format::bfyx };
auto input_dyn_layout = layout{ ov::PartialShape{ 2, input_f, ov::Dimension(), 2 }, data_types::f32, format::bfyx };
auto input_data1 = engine.allocate_memory(input_static_layout);
auto fc_weights_data = engine.allocate_memory({ ov::PartialShape{ 2, input_f, 1, 1 }, data_types::f32, format::bfyx});
set_values(input_data1, { 0.5f, -2.0f, -0.5f, -0.5f, 2.0f, 0.5f,
-0.5f, 2.0f, 0.5f, -0.5f, 2.0f, 0.5f });
set_values(fc_weights_data, { 1.5f, 1.0f, 0.5f,
-1.0f, 0.0f, 0.5f});
GridSampleOp::Attributes attributes(false, GridSampleOp::InterpolationMode::NEAREST, GridSampleOp::PaddingMode::ZEROS);
cldnn::topology topology{
input_layout("input", input_static_layout),
input_layout("input_dyn", input_dyn_layout),
activation("relu1", input_info("input"), activation_func::relu),
eltwise("elt1", { input_info("input"), input_info("relu1") }, eltwise_mode::prod),
activation("relu2", input_info("elt1"), activation_func::sqrt),
// The 'grid_sample' layer should be dynamic, not 'shape agnostic kernel' and user of 'elt1'. This is a key condition of this test.
grid_sample("grid_sample", { input_info("elt1"), input_info("input_dyn") }, attributes),
data("fc_weights", fc_weights_data),
fully_connected("fc", input_info("grid_sample"), "fc_weights")
};
ExecutionConfig config = get_test_default_config(engine);
config.set_property(ov::intel_gpu::optimize_data(true));
config.set_property(ov::intel_gpu::allow_new_shape_infer(true));
network network(engine, topology, config);
{
network.set_input_data("input", input_data1);
network.set_input_data("input_dyn", input_data1);
auto outputs = network.execute();
// Should be false for memory-reuse
ASSERT_NE(std::static_pointer_cast<fully_connected_inst>(network.get_primitive("relu1"))->output_memory_ptr()->buffer_ptr(),
std::static_pointer_cast<fully_connected_inst>(network.get_primitive("relu2"))->output_memory_ptr()->buffer_ptr());
}
}
};
TEST_F(memory_pool, basic_non_padded_relu_pipe) {
@ -641,6 +688,10 @@ TEST_F(memory_pool, dynamic_mem_reuse) {
this->test_dynamic_mem_reuse();
}
TEST_F(memory_pool, dynamic_mem_reuse_for_null_sel_impl) {
this->test_dynamic_mem_reuse_for_null_sel_impl();
}
#ifdef RUN_ALL_MODEL_CACHING_TESTS
TEST_F(memory_pool, basic_non_padded_relu_pipe_cached) {
this->test_basic_non_padded_relu_pipe(true);