From 22eab509078dd7a806a5eeb4ee92add84ce99814 Mon Sep 17 00:00:00 2001
From: Seungha Yang <seungha@centricular.com>
Date: Sun, 15 Mar 2020 19:20:47 +0900
Subject: [PATCH] nvdec: Add fallback for CUDA/OpenGL interop failure

It happens when local OpenGL context belongs to non-nvidia GPU.
---
 sys/nvcodec/gstcudaloader.c | 14 +++++++++++++
 sys/nvcodec/gstcudaloader.h |  6 ++++++
 sys/nvcodec/gstnvdec.c      | 49 ++++++++++++++++++++++++++++++++++++++++++---
 sys/nvcodec/stub/cuda.h     |  6 ++++++
 4 files changed, 72 insertions(+), 3 deletions(-)

diff --git a/sys/nvcodec/gstcudaloader.c b/sys/nvcodec/gstcudaloader.c
index 7c6df7c..965717a 100644
--- a/sys/nvcodec/gstcudaloader.c
+++ b/sys/nvcodec/gstcudaloader.c
@@ -92,6 +92,9 @@ typedef struct _GstNvCodecCudaVTable
     CUresult (CUDAAPI *
       CuGraphicsResourceSetMapFlags) (CUgraphicsResource resource,
       unsigned int flags);
+    CUresult (CUDAAPI * CuGLGetDevices) (unsigned int *pCudaDeviceCount,
+      CUdevice * pCudaDevices, unsigned int cudaDeviceCount,
+      CUGLDeviceList deviceList);
 } GstNvCodecCudaVTable;
 
 static GstNvCodecCudaVTable gst_cuda_vtable = { 0, };
@@ -150,6 +153,7 @@ gst_cuda_load_library (void)
   LOAD_SYMBOL (cuGraphicsGLRegisterImage, CuGraphicsGLRegisterImage);
   LOAD_SYMBOL (cuGraphicsGLRegisterBuffer, CuGraphicsGLRegisterBuffer);
   LOAD_SYMBOL (cuGraphicsResourceSetMapFlags, CuGraphicsResourceSetMapFlags);
+  LOAD_SYMBOL (cuGLGetDevices, CuGLGetDevices);
 
   vtable->loaded = TRUE;
 
@@ -389,3 +393,13 @@ CuGraphicsResourceSetMapFlags (CUgraphicsResource resource, unsigned int flags)
 
   return gst_cuda_vtable.CuGraphicsResourceSetMapFlags (resource, flags);
 }
+
+CUresult CUDAAPI
+CuGLGetDevices (unsigned int *pCudaDeviceCount, CUdevice * pCudaDevices,
+    unsigned int cudaDeviceCount, CUGLDeviceList deviceList)
+{
+  g_assert (gst_cuda_vtable.CuGLGetDevices != NULL);
+
+  return gst_cuda_vtable.CuGLGetDevices (pCudaDeviceCount, pCudaDevices,
+      cudaDeviceCount, deviceList);
+}
diff --git a/sys/nvcodec/gstcudaloader.h b/sys/nvcodec/gstcudaloader.h
index ef89986..39cacbb 100644
--- a/sys/nvcodec/gstcudaloader.h
+++ b/sys/nvcodec/gstcudaloader.h
@@ -142,5 +142,11 @@ G_GNUC_INTERNAL
 CUresult CUDAAPI CuGraphicsResourceSetMapFlags (CUgraphicsResource resource,
                                                 unsigned int flags);
 
+G_GNUC_INTERNAL
+CUresult CUDAAPI CuGLGetDevices (unsigned int * pCudaDeviceCount,
+                                 CUdevice * pCudaDevices,
+                                 unsigned int cudaDeviceCount,
+                                 CUGLDeviceList deviceList);
+
 G_END_DECLS
 #endif /* __GST_CUDA_LOADER_H__ */
diff --git a/sys/nvcodec/gstnvdec.c b/sys/nvcodec/gstnvdec.c
index 6a4d102..70a7c0c 100644
--- a/sys/nvcodec/gstnvdec.c
+++ b/sys/nvcodec/gstnvdec.c
@@ -479,7 +479,7 @@ gst_nvdec_negotiate (GstVideoDecoder * decoder)
   if (nvdec->mem_type == GST_NVDEC_MEM_TYPE_GL &&
       !gst_nvdec_ensure_gl_context (nvdec)) {
     GST_WARNING_OBJECT (nvdec,
-        "OpenGL context cannot support PBO memory, fallback to system memory");
+        "OpenGL context is not CUDA-compatible, fallback to system memory");
     nvdec->mem_type = GST_NVDEC_MEM_TYPE_SYSTEM;
   }
 
@@ -579,7 +579,7 @@ parser_display_callback (GstNvDec * nvdec, CUVIDPARSERDISPINFO * dispinfo)
   GstVideoCodecFrame *frame = NULL;
   GstBuffer *output_buffer = NULL;
   GstFlowReturn ret = GST_FLOW_OK;
-  gboolean copy_ret;
+  gboolean copy_ret = FALSE;
 
   GST_LOG_OBJECT (nvdec, "picture index: %u", dispinfo->picture_index);
 
@@ -639,7 +639,19 @@ parser_display_callback (GstNvDec * nvdec, CUVIDPARSERDISPINFO * dispinfo)
 #ifdef HAVE_NVCODEC_GST_GL
   if (nvdec->mem_type == GST_NVDEC_MEM_TYPE_GL) {
     copy_ret = gst_nvdec_copy_device_to_gl (nvdec, dispinfo, output_buffer);
-  } else
+
+    /* FIXME: This is the case where OpenGL context of downstream glbufferpool
+     * belongs to non-nvidia (or different device).
+     * There should be enhancement to ensure nvdec has compatible OpenGL context
+     */
+    if (!copy_ret) {
+      GST_WARNING_OBJECT (nvdec,
+          "Couldn't copy frame to GL memory, fallback to system memory");
+      nvdec->mem_type = GST_NVDEC_MEM_TYPE_SYSTEM;
+    }
+  }
+
+  if (!copy_ret)
 #endif
   {
     copy_ret = gst_nvdec_copy_device_to_system (nvdec, dispinfo, output_buffer);
@@ -1203,9 +1215,32 @@ gst_nvdec_finish (GstVideoDecoder * decoder)
 }
 
 #ifdef HAVE_NVCODEC_GST_GL
+static void
+gst_nvdec_check_cuda_device_from_context (GstGLContext * context,
+    gboolean * ret)
+{
+  guint device_count = 0;
+  CUdevice device_list[1] = { 0, };
+  CUresult cuda_ret;
+
+  *ret = FALSE;
+
+  cuda_ret = CuGLGetDevices (&device_count,
+      device_list, 1, CU_GL_DEVICE_LIST_ALL);
+
+  if (!gst_cuda_result (cuda_ret) || device_count == 0)
+    return;
+
+  *ret = TRUE;
+
+  return;
+}
+
 static gboolean
 gst_nvdec_ensure_gl_context (GstNvDec * nvdec)
 {
+  gboolean ret;
+
   if (!nvdec->gl_display) {
     GST_DEBUG_OBJECT (nvdec, "No available OpenGL display");
     return FALSE;
@@ -1241,6 +1276,14 @@ gst_nvdec_ensure_gl_context (GstNvDec * nvdec)
     return FALSE;
   }
 
+  gst_gl_context_thread_add (nvdec->gl_context,
+      (GstGLContextThreadFunc) gst_nvdec_check_cuda_device_from_context, &ret);
+
+  if (!ret) {
+    GST_WARNING_OBJECT (nvdec, "Current OpenGL context is not CUDA-compatible");
+    return FALSE;
+  }
+
   return TRUE;
 }
 
diff --git a/sys/nvcodec/stub/cuda.h b/sys/nvcodec/stub/cuda.h
index 91dbc35..985d882 100644
--- a/sys/nvcodec/stub/cuda.h
+++ b/sys/nvcodec/stub/cuda.h
@@ -92,6 +92,11 @@ typedef struct
   gsize Height;
 } CUDA_MEMCPY2D;
 
+typedef enum
+{
+  CU_GL_DEVICE_LIST_ALL = 0x01,
+} CUGLDeviceList;
+
 #define CUDA_VERSION 10000
 
 #ifdef _WIN32
@@ -112,6 +117,7 @@ typedef struct
 #define cuMemcpy2D cuMemcpy2D_v2
 #define cuMemcpy2DAsync cuMemcpy2DAsync_v2
 #define cuMemFree cuMemFree_v2
+#define cuGLGetDevices cuGLGetDevices_v2
 
 G_END_DECLS
 
-- 
2.7.4