va_trace_surface: dumpe surface content after vaEndPicture call

vaEndPicture will kick HW decode/encode, and at this time use CPU
to dump surface content can keep GPU/CPU in parallel.

Also for encode trace, don't need to wait current frame completions
but for decode, we must wait decode till it is completed

Signed-off-by: Austin Yuan <shengquan.yuan@intel.com>
(cherry picked from commit 0af365de4c03c12fc20183093a944f7513a1c7be)

2 files changed, 16 insertions, 21 deletions

diff --git a/va/va.c b/va/va.c
index e331f74..6210112 100644
--- a/va/va.c
+++ b/va/va.c
@@ -1069,19 +1069,16 @@ VAStatus vaEndPicture (
     VAContextID context
 )
 {
-  VAStatus va_status;
+  VAStatus va_status = VA_STATUS_SUCCESS;
   VADriverContextP ctx;
 
   CHECK_DISPLAY(dpy);
   ctx = CTX(dpy);
 
-  /* dump encode source surface */
-  VA_TRACE_SURFACE(va_TraceEndPicture, dpy, context, 0);
-  /* return directly if do dummy operation */
-  VA_FOOL_RETURN();
-  
-  va_status = ctx->vtable->vaEndPicture( ctx, context );
-  /* dump decode dest surface */
+  if (fool_codec == 0)
+      va_status = ctx->vtable->vaEndPicture( ctx, context );
+
+  /* dump surface content */
   VA_TRACE_SURFACE(va_TraceEndPicture, dpy, context, 1);
 
   return va_status;
diff --git a/va/va_trace.c b/va/va_trace.c
index 45856d7..6f2478f 100644
--- a/va/va_trace.c
+++ b/va/va_trace.c
@@ -2175,10 +2175,8 @@ void va_TraceEndPicture(
 
     TRACE_FUNCNAME(idx);
 
-    if (endpic_done == 0) {
-        va_TraceMsg(idx, "\tcontext = 0x%08x\n", context);
-        va_TraceMsg(idx, "\trender_targets = 0x%08x\n", trace_context[idx].trace_rendertarget);
-    }
+    va_TraceMsg(idx, "\tcontext = 0x%08x\n", context);
+    va_TraceMsg(idx, "\trender_targets = 0x%08x\n", trace_context[idx].trace_rendertarget);
 
     encode = (trace_context[idx].trace_entrypoint == VAEntrypointEncSlice) &&
         (trace_flag & VA_TRACE_FLAG_SURFACE_ENCODE);
@@ -2186,27 +2184,27 @@ void va_TraceEndPicture(
         (trace_flag & VA_TRACE_FLAG_SURFACE_DECODE);
     jpeg = (trace_context[idx].trace_entrypoint == VAEntrypointEncPicture) &&
         (trace_flag & VA_TRACE_FLAG_SURFACE_JPEG);
-    
-    /* want to trace encode source surface, do it before vaEndPicture */
-    if ((encode || jpeg) && (endpic_done == 0))
+
+    /* trace encode source surface, can do it before HW completes rendering */
+    if (encode || jpeg)
         va_TraceSurface(dpy);
     
-    /* want to trace encoode codedbuf, do it after vaEndPicture */
-    if ((encode || jpeg) && (endpic_done == 1)) {
-        /* force the pipleline finish rendering */
+    /* trace coded buffer, do it after HW completes rendering */
+    if ((encode || jpeg) && (trace_flag & VA_TRACE_FLAG_CODEDBUF)) {
         vaSyncSurface(dpy, trace_context[idx].trace_rendertarget);
         va_TraceCodedBuf(dpy);
     }
 
-    /* want to trace decode dest surface, do it after vaEndPicture */
-    if (decode && (endpic_done == 1)) {
-        /* force the pipleline finish rendering */
+    /* trace decoded surface, do it after HW completes rendering */
+    if (decode) {
         vaSyncSurface(dpy, trace_context[idx].trace_rendertarget);
         va_TraceSurface(dpy);
     }
+
     va_TraceMsg(idx, NULL);
 }
 
+
 void va_TraceSyncSurface(
     VADisplay dpy,
     VASurfaceID render_target