From 3c949f339957ac9faabac99ee6179279e958007c Mon Sep 17 00:00:00 2001 From: Kawrakow Date: Thu, 13 Aug 2026 15:23:38 +0200 Subject: [PATCH] Add work buffer size calculation for the ds4_comp op (#2307) --- ggml/src/ggml.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/ggml/src/ggml.c b/ggml/src/ggml.c index 4901f37a..189add9b 100644 --- a/ggml/src/ggml.c +++ b/ggml/src/ggml.c @@ -28972,8 +28972,15 @@ struct ggml_cplan ggml_graph_plan(const struct ggml_cgraph * cgraph, int n_threa } break; case GGML_OP_INDEXER_TOPK: { - size_t size = iqk_idx_topk_work_buffer_size(node, n_tasks); - cur = MAX(cur, size); + cur = iqk_idx_topk_work_buffer_size(node, n_tasks); + } break; + case GGML_OP_DS4_COMP: + { + if (node->op_params[0] == 0) { + struct ggml_tensor * idx = node->src[2]; + int ratio = idx->ne[0] / (2*node->ne[1]); + cur = 32*(4*ratio + 3)*sizeof(float)*n_tasks; + } } break; case GGML_OP_CROSS_ENTROPY_LOSS: {