va:add synchornization fence for HW execution

add fence in and fence out for vaEndPicture which is using to submit command buffer
fence out will be signaled after HW execution, fence in is the dependencies

Signed-off-by: Carl Zhang <carl.zhang@intel.com>

Author: XinfengZhang

va/va.c

@@ -1643,7 +1643,7 @@ VAStatus vaEndPicture(
     ctx = CTX(dpy);
 
     VA_TRACE_V(dpy, END_PICTURE, TRACE_BEGIN, context);
-    VA_TRACE_ALL(va_TraceEndPicture, dpy, context, 0);
+    VA_TRACE_ALL(va_TraceEndPicture, dpy, context, NULL, 0, 0);
     va_status = ctx->vtable->vaEndPicture(ctx, context);
     VA_TRACE_RET(dpy, va_status);
     /* dump surface content */
@@ -1653,6 +1653,32 @@ VAStatus vaEndPicture(
     return va_status;
 }
 
+VAStatus vaEndPicture2(
+    VADisplay dpy,
+    VAContextID context,
+    int32_t *sync_fds,
+    int32_t sync_num
+
+)
+{
+    VAStatus va_status = VA_STATUS_SUCCESS;
+    VADriverContextP ctx;
+
+    CHECK_DISPLAY(dpy);
+    ctx = CTX(dpy);
+
+    VA_TRACE_V(dpy, END_PICTURE, TRACE_BEGIN, context);
+    VA_TRACE_ALL(va_TraceEndPicture, dpy, context, sync_fds, sync_num, 0);
+    va_status = ctx->vtable->vaEndPicture2(ctx, context, sync_fds, sync_num);
+    VA_TRACE_RET(dpy, va_status);
+    /* dump surface content */
+    VA_TRACE_ALL(va_TraceEndPictureExt, dpy, context, 1);
+    VA_TRACE_V(dpy, END_PICTURE, TRACE_END, va_status);
+
+    return va_status;
+}
+
+
 VAStatus vaSyncSurface(
     VADisplay dpy,
     VASurfaceID render_target

va/va.h

@@ -4174,6 +4174,24 @@ VAStatus vaEndPicture(
     VAContextID context
 );
 
+/**
+ * Same behavior with vaEndPicture except a sync fd list for synchronizations.
+ * if sync_num = 0 or sync_fds == NULL, the behavior should be same with vaEndPicture.
+ * if sync_num >= 1, the sync_fds[0] should be the fence out, this fd(or the fence
+ * behind this fd)  will be overridden by this call, and will be signaled to indicate
+ * current frame finishing.
+ * sync_fds[1] to sync_fds[1 ~ sync_num-1] is fence in, current HW execution
+ * will be blocked until all these fences are signaled.
+ * these fence fds is file descriptor of dma_fence.
+ */
+
+VAStatus vaEndPicture2(
+    VADisplay dpy,
+    VAContextID context,
+    int32_t *sync_fds,
+    int32_t sync_num
+);
+
 /**
  * Make the end of rendering for a pictures in contexts passed with submission.
  * The server should start processing all pending operations for contexts.

va/va_backend.h

@@ -511,8 +511,16 @@ struct VADriverVTable {
         void **pbuf,                        /* out */
         uint32_t flags                      /* in */
     );
+
+    VAStatus(*vaEndPicture2)(
+        VADriverContextP ctx,              /* in */
+        VAContextID context,               /* in */
+        int32_t *sync_fds,
+        int32_t sync_num
+    );
+
     /** \brief Reserved bytes for future use, must be zero */
-    unsigned long reserved[53];
+    unsigned long reserved[52];
 
 };
 

va/va_trace.c

@@ -5943,6 +5943,8 @@ void va_TraceRenderPicture(
 void va_TraceEndPicture(
     VADisplay dpy,
     VAContextID context,
+    int32_t *sync_fds,
+    int32_t sync_num,
     int endpic_done
 )
 {
@@ -5952,6 +5954,12 @@ void va_TraceEndPicture(
 
     va_TraceMsg(trace_ctx, "\tcontext = 0x%08x\n", context);
     va_TraceMsg(trace_ctx, "\trender_targets = 0x%08x\n", trace_ctx->trace_rendertarget);
+    if (sync_num && sync_fds) {
+        va_TraceMsg(trace_ctx, "\t fence out = %d \n", sync_fds[0]);
+        for (int i = 1; i < sync_num && sync_num > 1; i ++) {
+            va_TraceMsg(trace_ctx, "\t fence_in index %d  = %d \n", i - 1, sync_fds[i]);
+        }
+    }
     va_TraceMsg(trace_ctx, NULL);
 }
 

va/va_trace.h

@@ -377,6 +377,8 @@ DLL_HIDDEN
 void va_TraceEndPicture(
     VADisplay dpy,
     VAContextID context,
+    int32_t *sync_fds,
+    int32_t sync_num,
     int endpic_done
 );