From: Masashi Yoshimura <redacted>
Date: Wed, 13 May 2026 17:22:44 +0000 (+0900)
Subject: flush the gpu profile timestamp before the queryset is overflowed (#22995)
X-Git-Tag: upstream/0.0.10438~1299
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=527045bfb03ec339150f67d1bb5a581a8e91ec0a;p=pkg%2Fggml%2Fsources%2Fllama.cpp

flush the gpu profile timestamp before the queryset is overflowed (#22995)
---

diff --git a/ggml/src/ggml-webgpu/ggml-webgpu.cpp b/ggml/src/ggml-webgpu/ggml-webgpu.cpp
index b24101c78..401c75c12 100644
--- a/ggml/src/ggml-webgpu/ggml-webgpu.cpp
+++ b/ggml/src/ggml-webgpu/ggml-webgpu.cpp
@@ -3148,6 +3148,16 @@ static ggml_status ggml_backend_webgpu_graph_compute(ggml_backend_t backend, str
             }
             ctx->param_arena.reset();
             commands.clear();
+#ifdef GGML_WEBGPU_GPU_PROFILE
+            // flush before the next batch can overflow the QuerySet
+            if (ctx->profile_timestamp_query_count + 2 * ctx->global_ctx->command_submit_batch_size >=
+                WEBGPU_MAX_PROFILE_QUERY_COUNT) {
+                ggml_backend_webgpu_collect_profile_results(ctx, profile_pipeline_names, num_inflight_batches);
+                // reset profile timestamp state
+                ctx->profile_timestamp_query_count = 0;
+                profile_pipeline_names.clear();
+            }
+#endif
         }
 
         node_idx += num_encoded_ops;