Skip to content

[ET-VK] Reduced int precision for texture coordinates in conv2d_pw op, to reduce shader register pressure. #6766

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 1 commit into from
Nov 12, 2024
Merged

[ET-VK] Reduced int precision for texture coordinates in conv2d_pw op, to reduce shader register pressure. #6766

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
11 changes: 6 additions & 5 deletions backends/vulkan/runtime/graph/ops/glsl/conv2d_pw.glsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,15 +77,16 @@ void main() {
sum[i] = sum[0];
}

int z4 = 0;
// Since the kernel is 1x1, we only have to loop over the depth dimension.
for (int z = 0, z4 = 0; z < in_group_size; z += 4, ++z4) {
for (uint16_t z = uint16_t(0); z < uint16_t(in_group_size); z += uint16_t(4), ++z4) {
// During prepacking, the weight tensor has been permuted so that the
// channel (IC) dim is along the x-axis, and the batch (OC) dim is along
// the z-axis.
const vec4 ktex_0 = texelFetch(t_kernel, u16vec2(z + 0, gpos.z), 0);
const vec4 ktex_1 = texelFetch(t_kernel, u16vec2(z + 1, gpos.z), 0);
const vec4 ktex_2 = texelFetch(t_kernel, u16vec2(z + 2, gpos.z), 0);
const vec4 ktex_3 = texelFetch(t_kernel, u16vec2(z + 3, gpos.z), 0);
const vec4 ktex_0 = texelFetchOffset(t_kernel, u16vec2(z, gpos.z), 0, u16vec2(0, 0));
const vec4 ktex_1 = texelFetchOffset(t_kernel, u16vec2(z, gpos.z), 0, u16vec2(1, 0));
const vec4 ktex_2 = texelFetchOffset(t_kernel, u16vec2(z, gpos.z), 0, u16vec2(2, 0));
const vec4 ktex_3 = texelFetchOffset(t_kernel, u16vec2(z, gpos.z), 0, u16vec2(3, 0));


#pragma unroll
Expand Down
Loading