Skip to content

[ET-VK][qconv] Dynamically select between im2col path and general path #17387

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

Open
SS-JIA wants to merge 3 commits into
base: gh/SS-JIA/418/base
Choose a base branch
from
Open

[ET-VK][qconv] Dynamically select between im2col path and general path #17387

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
8394aa4
[ET-VK][qconv] Dynamically select between im2col path and general path
Feb 11, 2026
4622463
Update on "[ET-VK][qconv] Dynamically select between im2col path and …
Feb 11, 2026
1e5cfdd
Update on "[ET-VK][qconv] Dynamically select between im2col path and …
Feb 11, 2026
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
26 changes: 25 additions & 1 deletion backends/vulkan/runtime/graph/ops/impl/Q8taConv2d.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -323,7 +323,9 @@ void add_q8ta_conv2d_node(
// High level operator impl
//

void q8ta_conv2d(ComputeGraph& graph, const std::vector<ValueRef>& args) {
void q8ta_conv2d_general(
ComputeGraph& graph,
const std::vector<ValueRef>& args) {
int32_t idx = 0;
const ValueRef packed_int8_input = args.at(idx++);
const ValueRef input_scale = args.at(idx++);
Expand Down Expand Up @@ -398,8 +400,30 @@ void q8ta_conv2d(ComputeGraph& graph, const std::vector<ValueRef>& args) {
packed_int8_output);
}

void q8ta_conv2d(ComputeGraph& graph, const std::vector<ValueRef>& args) {
// Index into args to extract values needed for dispatch decision
const ValueRef packed_int8_input = args.at(0);
const ValueRef kernel_size = args.at(9);
const ValueRef groups = args.at(13);

const int32_t groups_val = graph.get_int(groups);
const int64_t IC = graph.size_at<int64_t>(-3, packed_int8_input);

const int64_t K_h = graph.get_int_list(kernel_size)->at(0);
const int64_t K_w = graph.get_int_list(kernel_size)->at(1);

// Use im2col path when: non-grouped, input channels multiple of 4, small
// kernel
if (groups_val == 1 && IC % 4 == 0 && K_h <= 3 && K_w <= 3) {
q8ta_conv2d_im2col(graph, args);
} else {
q8ta_conv2d_general(graph, args);
}
}

REGISTER_OPERATORS {
VK_REGISTER_OP(etvk.q8ta_conv2d.default, q8ta_conv2d);
VK_REGISTER_OP(etvk.q8ta_conv2d_general.default, q8ta_conv2d_general);
}

} // namespace vkcompute
2 changes: 2 additions & 0 deletions backends/vulkan/runtime/graph/ops/impl/Q8taConv2d.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -113,4 +113,6 @@ void add_q8ta_conv2d_pw_node(
const ValueRef packed_bias,
const ValueRef packed_int8_output);

void q8ta_conv2d_im2col(ComputeGraph& graph, const std::vector<ValueRef>& args);

} // namespace vkcompute
3 changes: 3 additions & 0 deletions backends/vulkan/test/custom_ops/impl/TestQ8taConv2d.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -157,6 +157,9 @@ void test_q8ta_conv2d(ComputeGraph& graph, const std::vector<ValueRef>& args) {
} else if (impl_selector == "im2col") {
// Use the im2col-based conv2d operator
VK_GET_OP_FN("etvk.q8ta_conv2d_im2col.default")(graph, conv_args);
} else if (impl_selector == "general") {
// Use the general q8ta_conv2d operator (no im2col dispatch)
VK_GET_OP_FN("etvk.q8ta_conv2d_general.default")(graph, conv_args);
} else {
// Use the new general q8ta_conv2d operator
VK_GET_OP_FN("etvk.q8ta_conv2d.default")(graph, conv_args);
Expand Down
33 changes: 19 additions & 14 deletions backends/vulkan/test/custom_ops/test_q8ta_conv2d.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@ static TestCase create_test_case_from_config(
vkapi::ScalarType input_dtype,
utils::StorageType fp_storage_type,
utils::GPUMemoryLayout int8_memory_layout,
const std::string& impl_selector = "general") {
const std::string& impl_selector = "") {
TestCase test_case;

// Calculate output dimensions
Expand All @@ -53,7 +53,6 @@ static TestCase create_test_case_from_config(
std::to_string(config.input_size.w) + " " +
"g=" + std::to_string(config.groups) + " " +
"k=" + std::to_string(config.kernel.h) + " " +
repr_str(fp_storage_type, fp_memory_layout) + "->" +
repr_str(utils::kBuffer, int8_memory_layout);
if (!impl_selector.empty()) {
test_name += " [" + impl_selector + "]";
Expand Down Expand Up @@ -218,8 +217,7 @@ std::vector<TestCase> generate_quantized_conv2d_easy_cases() {
};
config.op_name = "conv2d_q8ta_q8csw_q8to";

std::vector<utils::StorageType> fp_storage_types = {
utils::kTexture3D, utils::kBuffer};
std::vector<utils::StorageType> fp_storage_types = {utils::kTexture3D};

// Memory layouts for int8 tensors - test both optimized (4W4C) and general
// paths
Expand Down Expand Up @@ -379,8 +377,7 @@ static std::vector<TestCase> generate_quantized_conv2d_test_cases() {
4}};

// Test with different storage types and memory layouts
std::vector<utils::StorageType> fp_storage_types = {
utils::kTexture3D, utils::kBuffer};
std::vector<utils::StorageType> fp_storage_types = {utils::kTexture3D};

// Memory layouts for int8 tensors - test both optimized (4W4C) and general
// paths
Expand All @@ -401,29 +398,37 @@ static std::vector<TestCase> generate_quantized_conv2d_test_cases() {
int8_memory_layouts) {
config.test_case_name = make_test_case_name(
config, is_performance, fp_storage_type, utils::kBuffer);

test_cases.push_back(create_test_case_from_config(
config, vkapi::kFloat, fp_storage_type, int8_memory_layout));
config,
vkapi::kFloat,
fp_storage_type,
int8_memory_layout,
/*impl_selector=*/"general"));

// For 4W4C layout, also test the legacy implementation
if (int8_memory_layout == utils::kPackedInt8_4W4C) {
// Test im2col implementation for non-grouped convolutions with input
// channels that are a multiple of 4 and stride_w == 1
if (config.groups == 1 && config.channels.in % 4 == 0) {
test_cases.push_back(create_test_case_from_config(
config,
vkapi::kFloat,
fp_storage_type,
int8_memory_layout,
/*impl_selector=*/"legacy_4w4c"));
/*impl_selector=*/"im2col"));
}

// Test im2col implementation for non-grouped convolutions with input
// channels that are a multiple of 4 and stride_w == 1
if (config.groups == 1 && config.channels.in % 4 == 0) {
// For 4W4C layout, also test the legacy implementation
if (int8_memory_layout == utils::kPackedInt8_4W4C) {
test_cases.push_back(create_test_case_from_config(
config,
vkapi::kFloat,
fp_storage_type,
int8_memory_layout,
/*impl_selector=*/"im2col"));
/*impl_selector=*/"legacy_4w4c"));
}

test_cases.push_back(create_test_case_from_config(
config, vkapi::kFloat, fp_storage_type, int8_memory_layout));
}
}
}
Expand Down
7 changes: 2 additions & 5 deletions backends/vulkan/test/custom_ops/test_q8ta_conv2d_dw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -54,7 +54,6 @@ TestCase create_test_case_from_config(
std::to_string(config.input_size.w) + " " +
"g=" + std::to_string(config.groups) + " " +
"k=" + std::to_string(config.kernel.h) + " " +
repr_str(fp_storage_type, fp_memory_layout) + "->" +
repr_str(utils::kBuffer, int8_memory_layout);
if (!impl_selector.empty()) {
test_name += " [" + impl_selector + "]";
Expand Down Expand Up @@ -228,8 +227,7 @@ std::vector<TestCase> generate_quantized_conv2d_dw_easy_cases() {
};
config.op_name = "conv2d_q8ta_q8csw_q8to";

std::vector<utils::StorageType> fp_storage_types = {
utils::kTexture3D, utils::kBuffer};
std::vector<utils::StorageType> fp_storage_types = {utils::kTexture3D};

// Memory layouts for int8 tensors - test both optimized (4W4C) and general
// paths
Expand Down Expand Up @@ -351,8 +349,7 @@ std::vector<TestCase> generate_quantized_conv2d_dw_test_cases() {
32}};

// Test with different storage types and data types
std::vector<utils::StorageType> fp_storage_types = {
utils::kTexture3D, utils::kBuffer};
std::vector<utils::StorageType> fp_storage_types = {utils::kTexture3D};

// Memory layouts for int8 tensors - test both optimized (4W4C) and general
// paths
Expand Down
4 changes: 1 addition & 3 deletions backends/vulkan/test/custom_ops/test_q8ta_conv2d_pw.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,7 +53,6 @@ static TestCase create_test_case_from_config(
std::to_string(config.input_size.w) + " " +
"g=" + std::to_string(config.groups) + " " +
"k=" + std::to_string(config.kernel.h) + " " +
repr_str(fp_storage_type, fp_memory_layout) + "->" +
repr_str(utils::kBuffer, int8_memory_layout);
if (!impl_selector.empty()) {
test_name += " [" + impl_selector + "]";
Expand Down Expand Up @@ -286,8 +285,7 @@ static std::vector<TestCase> generate_quantized_conv2d_pw_test_cases() {
};

// Test with different storage types and memory layouts
std::vector<utils::StorageType> fp_storage_types = {
utils::kTexture3D, utils::kBuffer};
std::vector<utils::StorageType> fp_storage_types = {utils::kTexture3D};

// Memory layouts for int8 tensors - test both optimized (4W4C) and general
// paths
Expand Down
Loading