hwcontext_vulkan: wait and signal semaphores when transferring to CUDA

Same as when downloading. Not sure why this isn't done, probably
because the CUDA code predates the sync mechanism we settled on.

diff --git a/libavutil/hwcontext_vulkan.c b/libavutil/hwcontext_vulkan.c
index 6df1db5006232e7c3e4da8768acc333cf467b6a4..d4ff4ae307cd312f8938881d0507cc6acd9a93f2 100644 (file)
--- a/libavutil/hwcontext_vulkan.c
+++ b/libavutil/hwcontext_vulkan.c
@@ -3253,6 +3253,8 @@ static int vulkan_transfer_data_to_cuda(AVHWFramesContext *hwfc, AVFrame *dst,
     AVCUDADeviceContext *cuda_dev = cuda_cu->hwctx;
     AVCUDADeviceContextInternal *cu_internal = cuda_dev->internal;
     CudaFunctions *cu = cu_internal->cuda_dl;
+    CUDA_EXTERNAL_SEMAPHORE_WAIT_PARAMS s_w_par[AV_NUM_DATA_POINTERS] = { 0 };
+    CUDA_EXTERNAL_SEMAPHORE_SIGNAL_PARAMS s_s_par[AV_NUM_DATA_POINTERS] = { 0 };
 
     ret = CHECK_CU(cu->cuCtxPushCurrent(cuda_dev->cuda_ctx));
     if (ret < 0)
@@ -3268,6 +3270,13 @@ static int vulkan_transfer_data_to_cuda(AVHWFramesContext *hwfc, AVFrame *dst,
 
     dst_int = dst_f->internal;
 
+    ret = CHECK_CU(cu->cuWaitExternalSemaphoresAsync(dst_int->cu_sem, s_w_par,
+                                                     planes, cuda_dev->stream));
+    if (ret < 0) {
+        err = AVERROR_EXTERNAL;
+        goto fail;
+    }
+
     for (int i = 0; i < planes; i++) {
         CUDA_MEMCPY2D cpy = {
             .dstMemoryType = CU_MEMORYTYPE_DEVICE,
@@ -3292,6 +3301,13 @@ static int vulkan_transfer_data_to_cuda(AVHWFramesContext *hwfc, AVFrame *dst,
         }
     }
 
+    ret = CHECK_CU(cu->cuSignalExternalSemaphoresAsync(dst_int->cu_sem, s_s_par,
+                                                       planes, cuda_dev->stream));
+    if (ret < 0) {
+        err = AVERROR_EXTERNAL;
+        goto fail;
+    }
+
     CHECK_CU(cu->cuCtxPopCurrent(&dummy));
 
     av_log(hwfc, AV_LOG_VERBOSE, "Transfered Vulkan image to CUDA!\n");