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

Enable tensor offset to GemmKernelRef for input padding support (#12133)

Browse Source 
Signed-off-by: Andrew Park <andrew.park@intel.com>
This commit is contained in:
Andrew Kwangwoong Park
2022-07-14 12:22:36 +09:00
committed by GitHub
parent 99dbb35835
commit dc374ca1bf
2 changed files with 57 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/plugins/intel_gpu/src/kernel_selector/core/actual_kernels/gemm/gemm_kernel_ref.cpp
View File

@@ -26,6 +26,7 @@ ParamsKey GemmKernelRef::GetSupportedKey() const {
k.EnableBatching();
k.EnableDifferentTypes();
k.EnableTensorPitches();
k.EnableTensorOffset();
k.EnableQuantization(QuantizationType::SYMMETRIC);
return k;

56
src/plugins/intel_gpu/tests/test_cases/gemm_gpu_test.cpp
View File

@@ -8,6 +8,7 @@
#include <intel_gpu/primitives/input_layout.hpp>
#include <intel_gpu/primitives/gemm.hpp>
#include <intel_gpu/primitives/crop.hpp>
#include <cstddef>
@@ -111,6 +112,61 @@ TEST(gemm_gpu, basic_bfyx_t2) {
}
}
TEST(gemm_gpu, basic_bfyx_t2_inplace_crop_with_pad) {
auto& engine = get_test_engine();
auto input = engine.allocate_memory({ data_types::f32, format::bfyx, { 1, 2, 4, 3 } });
auto input2 = engine.allocate_memory({ data_types::f32, format::bfyx, { 1, 1, 4, 1 } });
std::vector<float> input_data = {
1.f, -2.f, 3.f, -4.f,
5.f, 6.f, 1.f, 2.f,
3.f, 3.f, 2.f, -1.f,
1.f, -2.f, 3.f, -4.f,
5.f, 6.f, 1.f, 2.f,
3.f, 3.f, 2.f, -1.f,
};
std::vector<float> input_data2 = {
2.f, 5.f, -4.f, -7.f,
};
set_values(input, input_data);
set_values(input2, input_data2);
std::vector<float> out_data = {
8.f, 22.f, 20.f
};
topology topology;
topology.add(
input_layout("input", input->get_layout())
);
topology.add(
input_layout("input2", input2->get_layout())
);
topology.add(
crop("crop.1", "input", { 1, 1, 4, 3 }, { 0, 1, 0, 0 })
);
topology.add(
gemm("output", { "crop.1", "input2" }, data_types::f32, false, true)
);
build_options options;
options.set_option(build_option::optimize_data(true));
network network(engine, topology, options);
network.set_input_data("input", input);
network.set_input_data("input2", input2);
auto outputs = network.execute();
auto output = outputs.at("output").get_memory();
cldnn::mem_lock<float> output_ptr(output, get_test_stream());
EXPECT_EQ(output_ptr.size(), (uint32_t)3);
for (uint32_t i = 0; i < out_data.size(); ++i) {
EXPECT_FLOAT_EQ(output_ptr[i], out_data[i]);
}
}
TEST(gemm_gpu, basic_bfyx_t1t2) {
auto& engine = get_test_engine();
auto input = engine.allocate_memory({ data_types::f32, format::bfyx, { 2, 1, 3, 4 } });