Skip to content

[ET-VK] 6/n Split dispatches between multiple command buffers. Replaced encode_execute function with invalidate_execute_encoding and moved encoding logic to execute function(). #13016

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

Merged
merged 4 commits into from
Aug 1, 2025
Merged

[ET-VK] 6/n Split dispatches between multiple command buffers. Replaced encode_execute function with invalidate_execute_encoding and moved encoding logic to execute function(). #13016

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
9f03195
[ET-VK] 6/n Split dispatches between multiple command buffers. Replac…
trivedivivek Jul 30, 2025
c867ffd
Update on "[ET-VK] 6/n Split dispatches between multiple command buff…
trivedivivek Jul 31, 2025
216ff46
Update on "[ET-VK] 6/n Split dispatches between multiple command buff…
trivedivivek Jul 31, 2025
2592f49
Update on "[ET-VK] 6/n Split dispatches between multiple command buff…
trivedivivek Jul 31, 2025
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
7 changes: 0 additions & 7 deletions backends/vulkan/runtime/VulkanBackend.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -509,13 +509,6 @@ class VulkanBackend final : public ::executorch::runtime::BackendInterface {

compute_graph->prepack();

// If dynamic shapes are not expected, then the command buffer only needs to
// be encoded once. Otherwise, wait until the first inference to encode the
// the command buffer, when actual input shapes are known.
if (!compute_graph->graphconfig().expect_dynamic_shapes) {
compute_graph->encode_execute();
}

return Error::Ok;
}

Expand Down
23 changes: 11 additions & 12 deletions backends/vulkan/runtime/graph/ComputeGraph.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -860,21 +860,20 @@ void ComputeGraph::prepack() {
staging_nbytes_in_cmd_ = 0;
}

void ComputeGraph::encode_execute() {
clear_deferred_cmds();
context_->flush();
context_->set_cmd(/*reusable = */ true);
void ComputeGraph::execute() {
if (deferred_cmd_list_.empty()) {
context_->flush();
context_->set_cmd(/*reusable = */ true);

context_->cmd_reset_querypool();
context_->cmd_reset_querypool();

for (std::unique_ptr<ExecuteNode>& node : execute_nodes_) {
node->encode(this);
}
for (std::unique_ptr<ExecuteNode>& node : execute_nodes_) {
node->encode(this);
}

deferred_cmd_list_.emplace_back(std::move(context_->extract_cmd()));
}
deferred_cmd_list_.emplace_back(std::move(context_->extract_cmd()));
}

void ComputeGraph::execute() {
submit_deferred_cmds_and_wait();
execute_count_++;
}
Expand All @@ -898,7 +897,7 @@ void ComputeGraph::propagate_resize() {
}
// Only re-encode on resize if dynamic shapes are expected
if (config_.expect_dynamic_shapes) {
encode_execute();
clear_deferred_cmds();
}
}

Expand Down
1 change: 0 additions & 1 deletion backends/vulkan/runtime/graph/ComputeGraph.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -892,7 +892,6 @@ class ComputeGraph final {
// Graph Execution
//

void encode_execute();
void execute();

//
Expand Down
2 changes: 0 additions & 2 deletions backends/vulkan/test/op_tests/choose_qparams_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -458,7 +458,6 @@ void test_vulkan_choose_qparams_tensor_impl(
graph.prepare();

graph.prepack();
graph.encode_execute();

// Run Vulkan choose_qparams_tensor
graph.copy_into_staging(
Expand Down Expand Up @@ -678,7 +677,6 @@ void test_vulkan_choose_qparams_per_token_asymmetric_impl(
graph.prepare();

graph.prepack();
graph.encode_execute();

// Run Vulkan choose_qparams_per_token_asymmetric
graph.copy_into_staging(
Expand Down
3 changes: 0 additions & 3 deletions backends/vulkan/test/op_tests/dequantize_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1140,7 +1140,6 @@ void test_vulkan_dequantize_per_token_impl(
graph.prepare();

graph.prepack();
graph.encode_execute();

// Copy input data to GPU
graph.copy_into_staging(
Expand Down Expand Up @@ -1671,7 +1670,6 @@ void test_vulkan_dequantize_per_channel_impl(

graph.prepare();
graph.prepack();
graph.encode_execute();

// Copy input data to GPU
graph.copy_into_staging(
Expand Down Expand Up @@ -2345,7 +2343,6 @@ void test_vulkan_dequantize_per_tensor_tensor_impl(

graph.prepare();
graph.prepack();
graph.encode_execute();

// Run Vulkan dequantize_per_tensor.tensor
graph.copy_into_staging(
Expand Down
3 changes: 0 additions & 3 deletions backends/vulkan/test/op_tests/quantize_affine_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -491,7 +491,6 @@ void test_vulkan_quantize_affine_impl(

graph.prepare();
graph.prepack();
graph.encode_execute();

// Copy input data to GPU
graph.copy_into_staging(
Expand Down Expand Up @@ -789,7 +788,6 @@ void test_vulkan_dequantize_affine_impl(

graph.prepare();
graph.prepack();
graph.encode_execute();

// Copy input data to GPU
graph.copy_into_staging(
Expand Down Expand Up @@ -1079,7 +1077,6 @@ void test_vulkan_choose_qparams_affine_impl(

graph.prepare();
graph.prepack();
graph.encode_execute();

// Copy input data to GPU
graph.copy_into_staging(
Expand Down
3 changes: 0 additions & 3 deletions backends/vulkan/test/op_tests/quantize_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -931,7 +931,6 @@ void test_vulkan_quantize_per_token_impl(
graph.prepare();

graph.prepack();
graph.encode_execute();

// Copy input data to GPU
graph.copy_into_staging(
Expand Down Expand Up @@ -1413,7 +1412,6 @@ void test_vulkan_quantize_per_channel_impl(

graph.prepare();
graph.prepack();
graph.encode_execute();

// Copy input data to GPU
graph.copy_into_staging(
Expand Down Expand Up @@ -2042,7 +2040,6 @@ void test_vulkan_quantize_per_tensor_tensor_impl(

graph.prepare();
graph.prepack();
graph.encode_execute();

// Run Vulkan quantize_per_tensor.tensor
graph.copy_into_staging(
Expand Down
5 changes: 1 addition & 4 deletions backends/vulkan/test/op_tests/quantized_linear_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -456,7 +456,6 @@ void test_vulkan_linear_qga4w_impl(
graph.prepare();

graph.prepack();
graph.encode_execute();

//
// Run model
Expand Down Expand Up @@ -551,7 +550,6 @@ void test_vulkan_linear_qcs4w_impl(
graph.prepare();

graph.prepack();
graph.encode_execute();

//
// Run model
Expand Down Expand Up @@ -685,7 +683,6 @@ void test_vulkan_linear_qta8a_qga4w_impl(
graph.prepare();

graph.prepack();
graph.encode_execute();

//
// Run model
Expand Down Expand Up @@ -900,4 +897,4 @@ TEST_F(VulkanLinearQTA8AQGA4WTest, test_vulkan_linear_quant_gemv) {
/*M = */ 1,
/*K = */ 256,
/*N = */ 256);
}
}
1 change: 0 additions & 1 deletion backends/vulkan/test/op_tests/rotary_embedding_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,7 +114,6 @@ void test_reference(
graph.prepare();

graph.prepack();
graph.encode_execute();

//
// Run model
Expand Down
3 changes: 0 additions & 3 deletions backends/vulkan/test/op_tests/sdpa_test.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -352,7 +352,6 @@ void test_vulkan_sdpa(
graph.prepare();

graph.prepack();
graph.encode_execute();

//
// Run model
Expand Down Expand Up @@ -586,7 +585,6 @@ void test_vulkan_flash_attention(
graph.prepare();
graph.encode_prepack();
graph.prepack();
graph.encode_execute();

// Copy inputs and run
graph.copy_into_staging(r_q.staging, q.const_data_ptr(), q.numel());
Expand Down Expand Up @@ -845,7 +843,6 @@ void test_reference_flash_attention(
graph.prepare();
graph.encode_prepack();
graph.prepack();
graph.encode_execute();

graph.copy_into_staging(r_q.staging, q.const_data_ptr(), q.numel());
graph.copy_into_staging(r_k.staging, k.const_data_ptr(), k.numel());
Expand Down
1 change: 0 additions & 1 deletion backends/vulkan/test/op_tests/utils/gen_computegraph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -682,7 +682,6 @@ def gen_graph_build_code(self, include_declarations: bool = True) -> str:

graph_build += f"{self.graph}{self.dot}prepare();\n"
graph_build += f"{self.graph}{self.dot}prepack();\n"
graph_build += f"{self.graph}{self.dot}encode_execute();\n"

graph_build += "\n"
return graph_build
Expand Down
Loading
Loading