CUDA: fuse rope + set_rows (#16884)

* CUDA: add fused rope * move k forward_expand up * create helper function instead of re-using params * make assert statement more in line with comment * rope_norm: coalesced writes to global mem
2025-11-13 08:50:01 +08:00
parent 07751f8d44
commit a90eb94ca9
4 changed files with 215 additions and 61 deletions
@@ -5,3 +5,5 @@
 void ggml_cuda_op_rope(ggml_backend_cuda_context & ctx, ggml_tensor * dst);

 void ggml_cuda_op_rope_back(ggml_backend_cuda_context & ctx, ggml_tensor * dst);
+
+void ggml_cuda_op_rope_fused(ggml_backend_cuda_context & ctx, ggml_tensor * dst, ggml_tensor * set_rows);