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

[GPU] Fix 1d onednn convolutions (#17038)

Browse Source
This commit is contained in:
Vladimir Paramuzov 2023-04-24 09:24:56 +04:00 committed by GitHub
parent bb0be3c177
commit 219a0eebdc
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 84 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.cpp
View File

@ -163,8 +163,6 @@ protected:
return weights_reorder_params;
}
public:
void save(BinaryOutputBuffer& ob) const override {
#ifdef ONEDNN_PRIMITIVE_SERIALIZATION

9
src/plugins/intel_gpu/src/graph/impls/onednn/convolution_onednn.hpp
View File

@ -37,6 +37,15 @@ static std::shared_ptr<dnnl::convolution_forward::primitive_desc> get_convolutio
pad_r[i] = (os - 1) * stride[i] - is + kernel_range - pad_l[i];
}
// Extend conv parameters in case if spatials rank of output memory doesn't match size of parameters
int64_t insert_count = static_cast<int64_t>(output_md.get_dims().size()) - 2 - stride.size();
if (insert_count > 0) {
stride.insert(stride.end(), insert_count, 1);
dilation.insert(dilation.end(), insert_count, 0);
pad_l.insert(pad_l.end(), insert_count, 0);
pad_r.insert(pad_r.end(), insert_count, 0);
}
if (!prim->bias.empty()) {
auto bias_md = onednn::layout_to_memory_desc(impl_params.get_input_layout(2), dnnl::memory::format_tag::any, true);
return std::make_shared<dnnl::convolution_forward::primitive_desc>(

75
src/plugins/intel_gpu/tests/test_cases/convolution_gpu_test.cpp
View File

@ -9227,6 +9227,81 @@ TEST(convolution_gpu_onednn, padding_for_cldnn_kernel_after_onednn) {
}
}
TEST(convolution_gpu_onednn, spatial_1d) {
auto& engine = get_test_engine();
if (!engine.get_device_info().supports_immad)
return;
ov::PartialShape input_pshape = {1, 16, 6};
ov::PartialShape weights_pshape = {16, 16, 3};
layout in_layout{ input_pshape, data_types::f16, format::bfyx };
layout weights_layout{ weights_pshape, data_types::f16, format::bfyx };
auto input_data = generate_random_1d<FLOAT16>(in_layout.count(), -1, 1);
auto input_mem = engine.allocate_memory(in_layout);
set_values(input_mem, input_data);
auto weights_data = generate_random_1d<FLOAT16>(weights_layout.count(), -1, 1);
auto weights_mem = engine.allocate_memory(weights_layout);
set_values(weights_mem, weights_data);
auto input = input_layout("input", input_mem->get_layout());
auto weights = data("weights", weights_mem);
auto conv = convolution("conv",
input_info("input"),
{ "weights" },
1,
ov::Strides{1},
ov::CoordinateDiff{0},
ov::Strides{1},
ov::CoordinateDiff{0},
ov::CoordinateDiff{0});
auto output_reorder = reorder("reorder", input_info("conv"), format::bfyx, data_types::f32 );
topology t(input, weights, conv, output_reorder);
ExecutionConfig config_test = get_test_default_config(engine);
ov::intel_gpu::ImplementationDesc conv_impl_test = { format::b_fs_yx_fsv16, "", impl_types::onednn };
config_test.set_property(ov::intel_gpu::force_implementations(ov::intel_gpu::ImplForcingMap{ { "conv", conv_impl_test } }));
config_test.set_property(ov::intel_gpu::optimize_data(true));
config_test.set_property(ov::intel_gpu::allow_new_shape_infer(true));
ExecutionConfig config_ref = get_test_default_config(engine);
ov::intel_gpu::ImplementationDesc conv_impl_ref = { format::bfyx, "", impl_types::ocl };
config_ref.set_property(ov::intel_gpu::force_implementations(ov::intel_gpu::ImplForcingMap{{ "conv", conv_impl_ref } }));
config_ref.set_property(ov::intel_gpu::optimize_data(true));
config_ref.set_property(ov::intel_gpu::allow_new_shape_infer(true));
network network_test(engine, t, config_test);
network network_ref(engine, t, config_ref);
network_test.set_input_data("input", input_mem);
network_ref.set_input_data("input", input_mem);
auto outputs_test = network_test.execute();
auto outputs_ref = network_ref.execute();
ASSERT_EQ(outputs_test.size(), size_t(1));
ASSERT_EQ(outputs_test.begin()->first, "reorder");
ASSERT_EQ(outputs_ref.size(), size_t(1));
ASSERT_EQ(outputs_ref.begin()->first, "reorder");
auto output_memory_test = outputs_test.at("reorder").get_memory();
auto output_layout_test = output_memory_test->get_layout();
cldnn::mem_lock<float> output_ptr_test(output_memory_test, get_test_stream());
auto output_memory_ref = outputs_ref.at("reorder").get_memory();
auto output_layout_ref = output_memory_ref->get_layout();
cldnn::mem_lock<float> output_ptr_ref(output_memory_ref, get_test_stream());
ov::PartialShape expected_shape = {1, 16, 4};
ASSERT_EQ(output_layout_test.get_partial_shape(), expected_shape);
ASSERT_EQ(output_layout_ref.get_partial_shape(), expected_shape);
for (size_t i = 0; i < output_memory_ref->count(); i++) {
ASSERT_EQ(output_ptr_ref.data()[i], output_ptr_test.data()[i]);
}
}
TEST(convolution_gpu_onednn, quantized_onednn_convolution_u8s8f32_asymmetric_activations_per_tensor) {
auto& engine = get_test_engine();
if (!engine.get_device_info().supports_immad)