Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

vulkan: Fuse rope+set_rows (#16769)

Browse Source 
This pattern appears in a lot of models, the rope operation is applied right
before storing into the KV cache (usually on the K tensor).

Add a path to some of the rope shaders that computes the destination address
based on the set_rows tensor. Compile variants of the shader with D_TYPE of
f16 (the usual KV cache type).

Add a src3 operand to ggml_vk_op_f32 - sometimes rope uses three srcs and needs
the fourth for the row indices.

Add fused_ops_write_mask to indicate which intermediate tensors need to write
their results to memory. Skipping writing the roped K value helps to allow more
nodes to run concurrently.

Add logic to ggml_vk_graph_optimize to make ROPE+VIEW+SET_ROWS consecutive. It
rarely starts out that way in the graph.

Add new backend tests.
This commit is contained in:
Jeff Bolz
2025-10-29 15:13:10 -05:00
committed by GitHub
parent 3464bdac37
commit b9ce940177
6 changed files with 371 additions and 117 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp
+10 -3
View File
@@ -16,12 +16,19 @@ void main() {
const uint row_x = row_dst % ne1;
const uint channel_x = row_dst / ne1;
const uint idst = row_dst*ne0 + i0/2;
uint idst = row_dst*ne0 + i0/2;
const uint ix = channel_x*p.s2 + row_x*p.s1 + i0/2;
// Fusion optimization: ROPE + VIEW + SET_ROWS..
// The rope output is viewed as a 1D tensor and offset based on a row index in data_i.
if (p.set_rows_stride != 0) {
idst = row_x*ne0 + i0/2;
idst += data_i[channel_x].x * p.set_rows_stride;
}
if (i0 >= p.n_dims) {
data_d[idst + i0/2 + 0] = data_a[ix + i0/2 + 0];
data_d[idst + i0/2 + 1] = data_a[ix + i0/2 + 1];
data_d[idst + i0/2 + 0] = D_TYPE(data_a[ix + i0/2 + 0]);
data_d[idst + i0/2 + 1] = D_TYPE(data_a[ix + i0/2 + 1]);
return;
}