avutil/hwcontext_cuda: explicitly synchronize cuMemcpy calls
authorTimo Rothenpieler <timo@rothenpieler.org>
Tue, 8 May 2018 10:12:01 +0000 (12:12 +0200)
committerTimo Rothenpieler <timo@rothenpieler.org>
Wed, 9 May 2018 22:34:22 +0000 (00:34 +0200)
libavutil/hwcontext_cuda.c

index f3e8168..668293b 100644 (file)
@@ -258,13 +258,19 @@ static int cuda_transfer_data_from(AVHWFramesContext *ctx, AVFrame *dst,
             .Height        = src->height >> (i ? priv->shift_height : 0),
         };
 
-        err = cu->cuMemcpy2D(&cpy);
+        err = cu->cuMemcpy2DAsync(&cpy, device_hwctx->stream);
         if (err != CUDA_SUCCESS) {
             av_log(ctx, AV_LOG_ERROR, "Error transferring the data from the CUDA frame\n");
             return AVERROR_UNKNOWN;
         }
     }
 
+    err = cu->cuStreamSynchronize(device_hwctx->stream);
+    if (err != CUDA_SUCCESS) {
+        av_log(ctx, AV_LOG_ERROR, "Error synchronizing CUDA stream\n");
+        return AVERROR_UNKNOWN;
+    }
+
     cu->cuCtxPopCurrent(&dummy);
 
     return 0;
@@ -297,13 +303,19 @@ static int cuda_transfer_data_to(AVHWFramesContext *ctx, AVFrame *dst,
             .Height        = src->height >> (i ? priv->shift_height : 0),
         };
 
-        err = cu->cuMemcpy2D(&cpy);
+        err = cu->cuMemcpy2DAsync(&cpy, device_hwctx->stream);
         if (err != CUDA_SUCCESS) {
-            av_log(ctx, AV_LOG_ERROR, "Error transferring the data from the CUDA frame\n");
+            av_log(ctx, AV_LOG_ERROR, "Error transferring the data to the CUDA frame\n");
             return AVERROR_UNKNOWN;
         }
     }
 
+    err = cu->cuStreamSynchronize(device_hwctx->stream);
+    if (err != CUDA_SUCCESS) {
+        av_log(ctx, AV_LOG_ERROR, "Error synchronizing CUDA stream\n");
+        return AVERROR_UNKNOWN;
+    }
+
     cu->cuCtxPopCurrent(&dummy);
 
     return 0;