Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[TEST][TMP]Luwei/test buffer fusing #28208

Open
wants to merge 2 commits into
base: master
Choose a base branch
from
+54 −27
Open

[TEST][TMP]Luwei/test buffer fusing #28208

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
852448e
Add some comments.
luweizhou2016 Dec 24, 2024
a9ae5d9
test.
luweizhou2016 Dec 26, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
81 changes: 54 additions & 27 deletions src/plugins/intel_gpu/src/graph/graph_optimizer/prepare_buffer_fusing.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,7 +68,7 @@ auto available_pred = [](const program_node& input) {
!input.is_type<activation>() && !input.is_type<deconvolution>() && !input.is_type<concatenation>() &&
!input.is_type<crop>() && !input.is_type<eltwise>() && !input.is_type<resample>() &&
!input.is_type<reorder>() && !(input.is_type<permute>() && !input.as<permute>().is_rotating_except_batch()) &&
!input.is_type<strided_slice>())
!input.is_type<strided_slice>() && !input.is_type<reshape>())
return false;
return true;
};
Expand All @@ -77,8 +77,10 @@ bool concat_in_place_optimization::match(const program_node& concat_node,
kernel_impl_params& concat_params,
std::vector<kernel_impl_params>& pred_params,
bool is_runtime) {
if (concat_node.is_output() || concat_params.fused_desc.size() > 0 || concat_node.is_in_shape_of_subgraph())
if (concat_node.is_output() || concat_params.fused_desc.size() > 0 || concat_node.is_in_shape_of_subgraph()) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << concat_node.id() << std::endl;
return false;
}
bool do_runtime_buffer_fusing = true;
GPU_DEBUG_GET_INSTANCE(debug_config);
GPU_DEBUG_IF(debug_config->disable_runtime_buffer_fusing) {
Expand All @@ -93,8 +95,10 @@ bool concat_in_place_optimization::match(const program_node& concat_node,
for (size_t j = 0; j < concat_params.get_output_layout().get_rank(); j++) {
if (j != concat_axis_index) {
if ((concat_params.get_output_layout().data_padding._lower_size[j] != 0)
|| (concat_params.get_output_layout().data_padding._upper_size[j] != 0))
return false;
|| (concat_params.get_output_layout().data_padding._upper_size[j] != 0)) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << concat_node.id() << std::endl;
return false;
}
}
}
}
Expand All @@ -109,14 +113,20 @@ bool concat_in_place_optimization::match(const program_node& concat_node,
// for simple patterns where the concat is the only user of all the preds.
// Also cascaded concat is not handled for dynamic shape. for now.
// If we have more flexible exec order handling in the future we'll be able to remove this condition below
if (p.first->is_dynamic() && (!do_runtime_buffer_fusing || p.first->get_users().size() > 1))
return false;
if (concat_node.is_dynamic() && (!do_runtime_buffer_fusing || !p.first->is_dynamic()))
if (p.first->is_dynamic() && (!do_runtime_buffer_fusing || p.first->get_users().size() > 1)) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << p.first->id() << " , " << concat_node.id() << std::endl;
return false;
}
if (concat_node.is_dynamic() && (!do_runtime_buffer_fusing || !p.first->is_dynamic())) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << p.first->id() << " , " << concat_node.id() << std::endl;
return false;
}
}
// if this is called in primitive_inst::execute() and concat is static, that concat should already be optimized in build time, not in runtime.
if (is_runtime && !concat_node.is_dynamic())
return false;
if (is_runtime && !concat_node.is_dynamic()) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << concat_node.id() << std::endl;
return false;
}
bool is_onednn_impl = false;

// For in place concatenation input layouts and data types must match.
Expand All @@ -131,50 +141,67 @@ bool concat_in_place_optimization::match(const program_node& concat_node,

size_t idx = 0;
for (const auto& pred : pred_nodes) {
if (!available_pred(*pred.first))
if (!available_pred(*pred.first)) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << pred.first->id() << " , " << concat_node.id() << std::endl;
return false;
if (pred.first->is_output())
}
if (pred.first->is_output()) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << pred.first->id() << " , " << concat_node.id() << std::endl;
return false;
// if an input is marked as network output, prevent optimizations
} // if an input is marked as network output, prevent optimizations
// which would affect a form of its output (unless debug flag is set),
// we also need to restrict input types to those which support padding on all axis
if (!pred.first->is_dynamic() || is_runtime) {
if (!pred.first->is_padding_supported(static_cast<int>(concat_axis), lower_padd_in_axis))
if (!pred.first->is_padding_supported(static_cast<int>(concat_axis), lower_padd_in_axis)) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << pred.first->id() << " , " << concat_node.id() << std::endl;
return false;
}
}
// TODO: handle optimized reshape
if (pred.first->is_type<reshape>() && pred.first->can_be_optimized())
return false;
// if (pred.first->is_type<reshape>() && pred.first->can_be_optimized()) {
// GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << pred.first->id() << " , " << concat_node.id() << std::endl;
// return false;
// }
// TODO: Investigate if this condition is needed
if (pred.first->get_users().size() > 2)
return false;

if (pred.first->get_users().size() > 2) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << pred.first->id() << " , " << concat_node.id() << std::endl;
return false;
}
// Check that input isn't optimized out concatenation along different axis.
if (pred.first->is_type<concatenation>() && pred.first->can_be_optimized()) {
// cascaded concat opt is not supported for dynamic shape yet
if (concat_node.is_dynamic() || is_runtime)
if (concat_node.is_dynamic() || is_runtime) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << pred.first->id() << " , " << concat_node.id() << std::endl;
return false;
else if (pred.first->as<concatenation>().get_primitive()->axis != concat_axis)
} else if (pred.first->as<concatenation>().get_primitive()->axis != concat_axis) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << pred.first->id() << " , " << concat_node.id() << std::endl;
return false;
}
}
// Check that input isn't optimized out non-concatenation.
if (!pred.first->is_type<concatenation>() && pred.first->can_be_optimized())
return false;

if (!pred.first->is_type<concatenation>() && !pred.first->is_type<reshape>() && pred.first->can_be_optimized()) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << pred.first->id() << " , " << concat_node.id() << std::endl;
return false;
}
size_t concat_users = 0;
for (const auto& user : pred.first->get_users())
if (user->is_type<concatenation>())
concat_users += 1;

// If input is used by more than one concatenation then they may require different paddings.
if (concat_users != 1)
if (concat_users != 1) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << concat_node.id() << std::endl;
return false;

}
const layout& pred_l = pred_params[idx].get_output_layout();
if (output_format != pred_l.format || output_datatype != pred_l.data_type)
if (output_format != pred_l.format || output_datatype != pred_l.data_type) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << concat_node.id() << std::endl;
return false;
if (pred_l.format.block_sizes().size() > 1)
}
if (pred_l.format.block_sizes().size() > 1) {
GPU_DEBUG_TRACE_DETAIL << "[prepare_buffer_fusing_match_fail] : " << concat_node.id() << std::endl;
return false;
}
// TODO: Below condition should be moved to program_node::supports_padding.
// This however will need updating the algorithm as it may make cascade adjustment impossible in some cases.
// It however would make normal optimizations possible in others, so this is a trade-off to be investigated.
Expand Down
Loading