From b5b133723aaf863e080557df6e64601d30e1a8e4 Mon Sep 17 00:00:00 2001
From: 0cc4m <picard12@live.de>
Date: Thu, 20 Jul 2023 19:32:17 +0200
Subject: [PATCH] Don't free before queue done

---
 ggml-vulkan.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/ggml-vulkan.cpp b/ggml-vulkan.cpp
index 9781e286c..b7365e5d1 100644
--- a/ggml-vulkan.cpp
+++ b/ggml-vulkan.cpp
@@ -1602,8 +1602,6 @@ static void ggml_vk_mul_mat_f16(const ggml_tensor * src0, const ggml_tensor * sr
         }
     }
 
-    ggml_vk_host_free(fp16_staging);
-
     ggml_vk_submit(vk_transfer_queues[0], transfer_0_seqs, VK_NULL_HANDLE);
 
     // cleanup waits for the queue to be done
@@ -1611,6 +1609,8 @@ static void ggml_vk_mul_mat_f16(const ggml_tensor * src0, const ggml_tensor * sr
     ggml_vk_queue_cleanup(vk_transfer_queues[1]);
     ggml_vk_queue_cleanup(vk_compute_queue);
 
+    ggml_vk_host_free(fp16_staging);
+
     if (src0->backend != GGML_BACKEND_GPU) {
         ggml_vk_pool_free(d_X);
     }