Skip to content

Vulkan: use the 1x1 depthwise output tile for 5x5 kernels on Mali too #22784

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
msluszniak wants to merge 1 commit into
base: main
Choose a base branch
from
+13 −4
Open

Vulkan: use the 1x1 depthwise output tile for 5x5 kernels on Mali too #22784

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 9 additions & 0 deletions backends/vulkan/runtime/graph/ops/glsl/conv2d_dw_output_tile.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,3 +32,12 @@ conv2d_dw_output_tile:
OPERATOR: clamp(X, A, B)
BATCH_SIZE_X: 1
BATCH_SIZE_Y: 1
- NAME: conv2d_dw_output_tile_5x5_b1x1
TILE_SIZE: 5
BATCH_SIZE_X: 1
BATCH_SIZE_Y: 1
- NAME: conv2d_dw_output_tile_5x5_b1x1_clamp
OPERATOR: clamp(X, A, B)
TILE_SIZE: 5
BATCH_SIZE_X: 1
BATCH_SIZE_Y: 1
8 changes: 4 additions & 4 deletions backends/vulkan/runtime/graph/ops/impl/Conv2dDW.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -81,13 +81,13 @@ std::string pick_conv2d_dw_shader(
kernel_name += "_sned";
}

if (is_3x3) {
kernel_name += "_output_tile_3x3";
if (is_3x3 || is_5x5) {
kernel_name += is_3x3 ? "_output_tile_3x3" : "_output_tile_5x5";
// Mali spills with the default 4x2 output tile: each invocation holds 8
// accumulators and loads an overlapping input window for each of them.
if (stride_equals_dilation && graph.device_is_mali()) {
kernel_name += "_b1x1";
}
} else if (is_5x5) {
kernel_name += "_output_tile_5x5";
}

if (clamp_out) {
Expand Down
Loading