This is an automated email from the git hooks/post-receive script.

Git pushed a commit to branch master
in repository ffmpeg.

commit 2a072b4961ac36fae1d4a62432d154f6616add68
Author:     Lynne <[email protected]>
AuthorDate: Sun Aug 9 20:23:58 2026 +0900
Commit:     Lynne <[email protected]>
CommitDate: Sat Aug 15 18:02:17 2026 +0900

    vulkan_encode: do not carry source picture access flags into the barrier
    
    The source picture is written on other queues, whose access flags are
    meaningless on the encode queue: ALL_COMMANDS expands to only the
    stages the queue supports, none of which support e.g. TRANSFER_WRITE,
    making the barrier invalid (VUID 03915). The semaphore wait already
    made those writes available; use no source access, as decoding does.
---
 libavcodec/vulkan_encode.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/libavcodec/vulkan_encode.c b/libavcodec/vulkan_encode.c
index 3a2cf101f4..7dd690ca68 100644
--- a/libavcodec/vulkan_encode.c
+++ b/libavcodec/vulkan_encode.c
@@ -367,12 +367,15 @@ static int vulkan_encode_issue(AVCodecContext *avctx,
     if (err < 0)
         goto fail;
 
-    /* Source image layout conversion */
+    /* Source image layout conversion. Writes to it happened on other
+     * queues, and were made available by the semaphore wait: their access
+     * flags are not to be carried over, and may not even be valid on this
+     * queue. */
     img_bar[nb_img_bar] = (VkImageMemoryBarrier2) {
         .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2,
         .pNext = NULL,
         .srcStageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
-        .srcAccessMask = vkf->access[0],
+        .srcAccessMask = VK_ACCESS_2_NONE,
         .dstStageMask = VK_PIPELINE_STAGE_2_VIDEO_ENCODE_BIT_KHR,
         .dstAccessMask = VK_ACCESS_2_VIDEO_ENCODE_READ_BIT_KHR,
         .oldLayout = vkf->layout[0],

_______________________________________________
ffmpeg-cvslog mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to