gemma4: check graph_compute return in prefill_bsa per-layer embed

howard0su · Copilot · howard0su · commit a318f9df18af · 2026-05-22T00:00:30.000+08:00
Co-authored-by: Copilot &lt;223556219+Copilot@users.noreply.github.com&gt;
diff --git a/dflash/src/gemma4/gemma4_graph.cpp b/dflash/src/gemma4/gemma4_graph.cpp
@@ -960,7 +960,12 @@ bool gemma4_prefill_bsa(
             return false;
         }
         ggml_backend_tensor_set(tok, token_ids, 0, (size_t)S * sizeof(int32_t));
-        ggml_backend_graph_compute(backend, gf);
+        if (ggml_backend_graph_compute(backend, gf) != GGML_STATUS_SUCCESS) {
+            std::fprintf(stderr, "gemma4_prefill_bsa: per-layer embed graph_compute failed\n");
+            ggml_gallocr_free(ga); ggml_free(ctx);
+            g4_free_pers(per_layer_buf); cleanup_all();
+            return false;
+        }
         ggml_gallocr_free(ga);
         ggml_free(ctx);
     }

Original file line number	Diff line number	Diff line change
`@@ -960,7 +960,12 @@ bool gemma4_prefill_bsa(`
`960`	`960`	`return false;`
`961`	`961`	`}`
`962`	`962`	`ggml_backend_tensor_set(tok, token_ids, 0, (size_t)S * sizeof(int32_t));`
`963`		`- ggml_backend_graph_compute(backend, gf);`
	`963`	`+ if (ggml_backend_graph_compute(backend, gf) != GGML_STATUS_SUCCESS) {`
	`964`	`+ std::fprintf(stderr, "gemma4_prefill_bsa: per-layer embed graph_compute failed\n");`
	`965`	`+ ggml_gallocr_free(ga); ggml_free(ctx);`
	`966`	`+ g4_free_pers(per_layer_buf); cleanup_all();`
	`967`	`+ return false;`
	`968`	`+ }`
`964`	`969`	`ggml_gallocr_free(ga);`
`965`	`970`	`ggml_free(ctx);`
`966`	`971`	`}`