Skip to content

vulkan: fix rms_norm_mul to handle broadcasting dim0 #14817

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
Jul 22, 2025
Merged

vulkan: fix rms_norm_mul to handle broadcasting dim0 #14817

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
2 changes: 1 addition & 1 deletion ggml/src/ggml-vulkan/ggml-vulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10248,7 +10248,7 @@ static bool ggml_vk_can_fuse(const struct ggml_cgraph * cgraph, int node_idx, st
}
// if rms_norm is the B operand, then we don't handle broadcast
if (rms_norm == mul->src[1] &&
mul->src[0]->ne[1] != rms_norm->ne[1]) {
!ggml_are_same_shape(mul->src[0], rms_norm)) {
return false;
}
// rms_norm shader assumes contiguous rows
Expand Down
10 changes: 8 additions & 2 deletions ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,8 +50,14 @@ void main() {
const FLOAT_TYPE scale = inversesqrt(mean + FLOAT_TYPE(p.param1));

if (do_multiply) {
[[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) {
data_d[d_offset + col] = D_TYPE(scale * FLOAT_TYPE(data_a[a_offset + col]) * FLOAT_TYPE(data_b[b_offset + col]));
if (ncols > p.ne10) {
[[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) {
data_d[d_offset + col] = D_TYPE(scale * FLOAT_TYPE(data_a[a_offset + col]) * FLOAT_TYPE(data_b[b_offset + fastmod(col, p.ne10)]));
}
} else {
[[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) {
data_d[d_offset + col] = D_TYPE(scale * FLOAT_TYPE(data_a[a_offset + col]) * FLOAT_TYPE(data_b[b_offset + col]));
}
}
} else {
[[unroll]] for (uint col = tid; col < ncols; col += BLOCK_SIZE) {
Expand Down
Loading