vulkan: fuse rms_norm + mul + rope (+ view + set_rows) (#16977)

This change combines the rms_norm+mul and rope+view+set_rows fusions to allow fusing the whole sequence together. This comes up in Qwen3, Bailing, and some other models.
2025-11-08 08:52:15 +01:00
parent d6fe40fa00
commit b4e335d8dc
12 changed files with 695 additions and 281 deletions
@@ -0,0 +1,27 @@
+#if !defined(GGML_ROPE_PARAMS)
+#define GGML_ROPE_PARAMS
+
+#include "rte.glsl"
+
+struct rope_params {
+    uint rope_mode;
+    uint ncols;
+    uint n_dims;
+    float freq_scale;
+    uint p_delta_rows;
+    float freq_base;
+    float ext_factor;
+    float attn_factor;
+    float corr_dims[2];
+    float theta_scale;
+    uint has_ff;
+    uint ne02;
+    uint nb01;
+    uint nb02;
+    int sections[4];
+    uint is_imrope;
+    uint is_back;
+    uint set_rows_stride;
+};
+
+#endif // !defined(GGML_ROPE_PARAMS)