ggml: add graph_reused (#21764)

* ggml: add graph_reused

* use versioning instead of reuse flag

* increment version with atomic

* use top bits for split numbering

* add assert

* move counter to ggml.c

* set uid in split_graph only

* fix windows

* address further review comments

* get next_uid rather than doing bit manipulation

* rename + add comment about uid
This commit is contained in:
Aman Gupta
2026-04-16 17:21:28 +08:00
committed by GitHub
parent ae2d34899e
commit 3f7c29d318
5 changed files with 35 additions and 0 deletions
+1
View File
@@ -1186,6 +1186,7 @@ struct ggml_cuda_graph {
std::vector<cudaGraphNode_t> nodes;
bool disable_due_to_gpu_arch = false;
bool warmup_complete = false;
uint64_t uid = 0;
struct node_properties {
ggml_tensor node;
void * node_src_data_ptrs[GGML_MAX_SRC];
+9
View File
@@ -3108,6 +3108,15 @@ static bool ggml_cuda_graph_update_required(ggml_backend_cuda_context * cuda_ctx
const void * graph_key = ggml_cuda_graph_get_key(cgraph);
ggml_cuda_graph * graph = cuda_ctx->cuda_graph(graph_key);
if (cgraph->uid != 0 &&
cgraph->uid == graph->uid) {
GGML_LOG_DEBUG("CUDA Graph id %zu reused\n", cgraph->uid);
GGML_ASSERT((int)graph->node_props.size() == cgraph->n_nodes);
return false;
}
graph->uid = cgraph->uid;
// Check if the graph size has changed
if ((int)graph->node_props.size() != cgraph->n_nodes) {
res = true;