radeon/r200/r300/r600: Warn if we emit more than prediction was.

Prediction code making too small prediction may cause space check aserttion
failure later in rendering. So warning about any failure to predict correctly
should be fixed.

4 files changed, 44 insertions, 12 deletions

diff --git a/src/mesa/drivers/dri/r200/r200_tcl.c b/src/mesa/drivers/dri/r200/r200_tcl.c
index 0f5e501c1a6..7b3a9fc859d 100644
--- a/src/mesa/drivers/dri/r200/r200_tcl.c
+++ b/src/mesa/drivers/dri/r200/r200_tcl.c
@@ -51,6 +51,8 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r200_swtcl.h"
 #include "r200_maos.h"
 
+#include "radeon_common_context.h"
+
 
 
 #define HAVE_POINTS      1
@@ -368,12 +370,13 @@ r200ComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord )
  * Predict total emit size for next rendering operation so there is no flush in middle of rendering
  * Prediction has to aim towards the best possible value that is worse than worst case scenario
  */
-static void r200EnsureEmitSize( GLcontext * ctx , GLubyte* vimap_rev )
+static GLuint r200EnsureEmitSize( GLcontext * ctx , GLubyte* vimap_rev )
 {
   r200ContextPtr rmesa = R200_CONTEXT(ctx);
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct vertex_buffer *VB = &tnl->vb;
   GLuint space_required;
+  GLuint state_size;
   GLuint nr_aos = 0;
   int i;
   /* predict number of aos to emit */
@@ -387,10 +390,11 @@ static void r200EnsureEmitSize( GLcontext * ctx , GLubyte* vimap_rev )
 
   {
     /* count the prediction for state size */
-    space_required = radeonCountStateEmitSize( &rmesa->radeon );
+    space_required = 0;
+    state_size = radeonCountStateEmitSize( &rmesa->radeon );
     /* vtx may be changed in r200EmitArrays so account for it if not dirty */
     if (!rmesa->hw.vtx.dirty)
-      space_required += rmesa->hw.vtx.check(rmesa->radeon.glCtx, &rmesa->hw.vtx);
+      state_size += rmesa->hw.vtx.check(rmesa->radeon.glCtx, &rmesa->hw.vtx);
     /* predict size for elements */
     for (i = 0; i < VB->PrimitiveCount; ++i)
     {
@@ -413,7 +417,10 @@ static void r200EnsureEmitSize( GLcontext * ctx , GLubyte* vimap_rev )
     space_required += SCISSOR_BUFSZ;
   }
   /* flush the buffer in case we need more than is left. */
-  rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required, __FUNCTION__);
+  if (rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required + state_size, __FUNCTION__))
+    return space_required + radeonCountStateEmitSize( &rmesa->radeon );
+  else
+    return space_required + state_size;
 }
 
 
@@ -530,7 +537,8 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx,
    /* Do the actual work:
     */
    radeonReleaseArrays( ctx, ~0 /* stage->changed_inputs */ );
-   r200EnsureEmitSize( ctx, vimap_rev );
+   GLuint emit_end = r200EnsureEmitSize( ctx, vimap_rev )
+     + rmesa->radeon.cmdbuf.cs->cdw;
    r200EmitArrays( ctx, vimap_rev );
 
    for (i = 0 ; i < VB->PrimitiveCount ; i++)
@@ -547,6 +555,9 @@ static GLboolean r200_run_tcl_render( GLcontext *ctx,
       else
 	 r200EmitPrimitive( ctx, start, start+length, prim );
    }
+   if ( emit_end < rmesa->radeon.cmdbuf.cs->cdw )
+     WARN_ONCE("Rendering was %d commands larger than predicted size."
+	 " We might overflow  command buffer.\n", rmesa->radeon.cmdbuf.cs->cdw - emit_end);
 
    return GL_FALSE;		/* finished the pipe */
 }
diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c
index 2ac898bd59d..56680516c83 100644
--- a/src/mesa/drivers/dri/r300/r300_draw.c
+++ b/src/mesa/drivers/dri/r300/r300_draw.c
@@ -42,6 +42,7 @@
 #include "r300_cmdbuf.h"
 
 #include "radeon_buffer_objects.h"
+#include "radeon_common_context.h"
 
 #include "tnl/tnl.h"
 #include "tnl/t_vp_build.h"
@@ -632,7 +633,8 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx,
 
 	/* ensure we have the cmd buf space in advance to cover
 	 * the state + DMA AOS pointers */
-	r300PredictTryDrawPrimsSize(ctx, nr_prims);
+	GLuint emit_end = r300PredictTryDrawPrimsSize(ctx, nr_prims)
+		+ r300->radeon.cmdbuf.cs->cdw;
 
 	r300SetupIndexBuffer(ctx, ib);
 
@@ -656,6 +658,10 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx,
 		fprintf(stderr, "%s: %u (%d-%d) cs ending at %d\n",
 			__FUNCTION__, nr_prims, min_index, max_index, r300->radeon.cmdbuf.cs->cdw );
 
+	if (emit_end < r300->radeon.cmdbuf.cs->cdw)
+		WARN_ONCE("Rendering was %d commands larger than predicted size."
+				" We might overflow  command buffer.\n", r300->radeon.cmdbuf.cs->cdw - emit_end);
+
 	return GL_TRUE;
 }
 
diff --git a/src/mesa/drivers/dri/r600/r700_render.c b/src/mesa/drivers/dri/r600/r700_render.c
index 899c2cf05c3..9b0b5443d6a 100644
--- a/src/mesa/drivers/dri/r600/r700_render.c
+++ b/src/mesa/drivers/dri/r600/r700_render.c
@@ -53,6 +53,8 @@
 #include "r700_fragprog.h"
 #include "r700_state.h"
 
+#include "radeon_common_context.h"
+
 void r700WaitForIdle(context_t *context);
 void r700WaitForIdleClean(context_t *context);
 GLboolean r700SendTextureState(context_t *context);
@@ -384,7 +386,9 @@ static GLboolean r700RunRender(GLcontext * ctx,
         fprintf(stderr, "%s: cs end at %d\n",
                 __func__, context->radeon.cmdbuf.cs->cdw);
 
-    assert(context->radeon.cmdbuf.cs->cdw <= emit_end);
+    if ( emit_end < context->radeon.cmdbuf.cs->cdw )
+       WARN_ONCE("Rendering was %d commands larger than predicted size."
+	       " We might overflow  command buffer.\n", context->radeon.cmdbuf.cs->cdw - emit_end);
 
     return GL_FALSE;
 }
diff --git a/src/mesa/drivers/dri/radeon/radeon_tcl.c b/src/mesa/drivers/dri/radeon/radeon_tcl.c
index 908b3c9f061..2404f284505 100644
--- a/src/mesa/drivers/dri/radeon/radeon_tcl.c
+++ b/src/mesa/drivers/dri/radeon/radeon_tcl.c
@@ -50,6 +50,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_tcl.h"
 #include "radeon_swtcl.h"
 #include "radeon_maos.h"
+#include "radeon_common_context.h"
 
 
 
@@ -359,12 +360,13 @@ radeonComputeFogBlendFactor( GLcontext *ctx, GLfloat fogcoord )
  * Predict total emit size for next rendering operation so there is no flush in middle of rendering
  * Prediction has to aim towards the best possible value that is worse than worst case scenario
  */
-static void radeonEnsureEmitSize( GLcontext * ctx , GLuint inputs )
+static GLuint radeonEnsureEmitSize( GLcontext * ctx , GLuint inputs )
 {
   r100ContextPtr rmesa = R100_CONTEXT(ctx);
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct vertex_buffer *VB = &tnl->vb;
   GLuint space_required;
+  GLuint state_size;
   GLuint nr_aos = 1; /* radeonEmitArrays does always emit one */
   int i;
   /* list of flags that are allocating aos object */
@@ -388,10 +390,11 @@ static void radeonEnsureEmitSize( GLcontext * ctx , GLuint inputs )
 
   {
     /* count the prediction for state size */
-    space_required = radeonCountStateEmitSize( &rmesa->radeon );
+    space_required = 0;
+    state_size = radeonCountStateEmitSize( &rmesa->radeon );
     /* tcl may be changed in radeonEmitArrays so account for it if not dirty */
     if (!rmesa->hw.tcl.dirty)
-      space_required += rmesa->hw.tcl.check( rmesa->radeon.glCtx, &rmesa->hw.tcl );
+      state_size += rmesa->hw.tcl.check( rmesa->radeon.glCtx, &rmesa->hw.tcl );
     /* predict size for elements */
     for (i = 0; i < VB->PrimitiveCount; ++i)
     {
@@ -414,7 +417,10 @@ static void radeonEnsureEmitSize( GLcontext * ctx , GLuint inputs )
     space_required += SCISSOR_BUFSZ;
   }
   /* flush the buffer in case we need more than is left. */
-  rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required, __FUNCTION__);
+  if (rcommonEnsureCmdBufSpace(&rmesa->radeon, space_required, __FUNCTION__))
+    return space_required + radeonCountStateEmitSize( &rmesa->radeon );
+  else
+    return space_required + state_size;
 }
 
 /**********************************************************************/
@@ -467,7 +473,8 @@ static GLboolean radeon_run_tcl_render( GLcontext *ctx,
    }
 
    radeonReleaseArrays( ctx, ~0 );
-   radeonEnsureEmitSize( ctx, inputs );
+   GLuint emit_end = radeonEnsureEmitSize( ctx, inputs )
+     + rmesa->radeon.cmdbuf.cs->cdw;
    radeonEmitArrays( ctx, inputs );
 
    rmesa->tcl.Elts = VB->Elts;
@@ -487,6 +494,10 @@ static GLboolean radeon_run_tcl_render( GLcontext *ctx,
 	 radeonEmitPrimitive( ctx, start, start+length, prim );
    }
 
+   if (emit_end < rmesa->radeon.cmdbuf.cs->cdw)
+      WARN_ONCE("Rendering was %d commands larger than predicted size."
+	  " We might overflow  command buffer.\n", rmesa->radeon.cmdbuf.cs->cdw - emit_end);
+
    return GL_FALSE;		/* finished the pipe */
 }