Skip to content

[ET-VK][ez] Add AOT support for PackedInt8_4C1W dtype #17389

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/420/base
Choose a base branch
from
+19 −2
Open

[ET-VK][ez] Add AOT support for PackedInt8_4C1W dtype #17389

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
d936ff3
[ET-VK][ez] Add AOT support for PackedInt8_4C1W dtype
Feb 11, 2026
63926da
Update on "[ET-VK][ez] Add AOT support for PackedInt8_4C1W dtype"
Feb 11, 2026
32015e1
Update on "[ET-VK][ez] Add AOT support for PackedInt8_4C1W dtype"
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
9 changes: 7 additions & 2 deletions backends/vulkan/op_registry.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -431,6 +431,11 @@ def register_torchao_quantize_dequantize():
)


# =============================================================================
# Q8taQuantizeDequantize.cpp
# =============================================================================


@update_features(
[
exir_ops.edge.quantized_decomposed.quantize_per_tensor.default,
Expand All @@ -443,7 +448,7 @@ def register_quantize_per_tensor():
utils.CHANNELS_PACKED_TEXTURE_OR_CONTIGUOUS_BUFFER,
],
outputs_storage=[
utils.PACKED_INT8_4W4C_BUFFER,
utils.PACKED_INT8_BUFFER,
],
)

Expand All @@ -457,7 +462,7 @@ def register_quantize_per_tensor():
def register_dequantize_per_tensor():
return OpFeatures(
inputs_storage=[
utils.PACKED_INT8_4W4C_BUFFER,
utils.PACKED_INT8_BUFFER,
],
outputs_storage=[
utils.CHANNELS_PACKED_TEXTURE_OR_CONTIGUOUS_BUFFER,
Expand Down
2 changes: 2 additions & 0 deletions backends/vulkan/runtime/VulkanBackend.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,6 +144,8 @@ utils::GPUMemoryLayout get_memory_layout(
return utils::kPackedInt8_4W4C;
case vkgraph::VkMemoryLayout::PACKED_INT8_4H4W:
return utils::kPackedInt8_4H4W;
case vkgraph::VkMemoryLayout::PACKED_INT8_4C1W:
return utils::kPackedInt8_4C1W;
default:
break;
}
Expand Down
1 change: 1 addition & 0 deletions backends/vulkan/serialization/schema.fbs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,6 +42,7 @@ enum VkMemoryLayout : ubyte {
TENSOR_CHANNELS_PACKED = 2,
PACKED_INT8_4W4C = 3,
PACKED_INT8_4H4W = 4,
PACKED_INT8_4C1W = 8,
DEFAULT_LAYOUT = 255,
}

Expand Down
1 change: 1 addition & 0 deletions backends/vulkan/serialization/vulkan_graph_schema.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,6 +50,7 @@ class VkMemoryLayout(IntEnum):
TENSOR_CHANNELS_PACKED = 2
PACKED_INT8_4W4C = 3
PACKED_INT8_4H4W = 4
PACKED_INT8_4C1W = 8
DEFAULT_LAYOUT = 255

def __str__(self) -> str:
Expand Down
8 changes: 8 additions & 0 deletions backends/vulkan/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -594,6 +594,7 @@ def node_has_target(node: Any, target: str):
all_quantized_memory_layouts: Set[VkMemoryLayout] = {
VkMemoryLayout.PACKED_INT8_4W4C,
VkMemoryLayout.PACKED_INT8_4H4W,
VkMemoryLayout.PACKED_INT8_4C1W,
}

universal_memory_layout_set: Set[VkMemoryLayout] = {
Expand Down Expand Up @@ -983,7 +984,14 @@ def make_filtered_tensor_repset(

# Only includes memory layouts that can be used by quantized tensors

PACKED_INT8_BUFFER = TensorRepSet(all_quantized_memory_layouts, set())
PACKED_INT8_4W4C_BUFFER = TensorRepSet({VkMemoryLayout.PACKED_INT8_4W4C}, set())
PACKED_INT8_4C1W_BUFFER = TensorRepSet({VkMemoryLayout.PACKED_INT8_4C1W}, set())

PACKED_INT8_CHANNELS_PACKED_BUFFER = TensorRepSet(
{VkMemoryLayout.PACKED_INT8_4W4C, VkMemoryLayout.PACKED_INT8_4C1W}, set()
)


# Special use RepSets

Expand Down
Loading