draw: implement pipeline statistics in the draw module

This is a basic implementation of the pipeline statistics in the
draw module. The interface is similar to the stream output statistics
and also requires that the callers explicitly enable it.
Included is the implementation of the interface in llvmpipe and
softpipe. Only softpipe enables the pipeline statistics capability
though because llvmpipe is lacking gathering of the fragment shading
and rasterization statistics.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>

10 files changed, 112 insertions, 20 deletions

diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 2fb9bac..0f98021 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -842,3 +842,19 @@ draw_get_shader_param(unsigned shader, enum pipe_shader_cap param)
    return draw_get_shader_param_no_llvm(shader, param);
 }
 
+/**
+ * Enables or disables collection of statistics.
+ *
+ * Draw module is capable of generating statistics for the vertex
+ * processing pipeline. Collection of that data isn't free and so
+ * it's disabled by default. The users of the module can enable
+ * (or disable) this functionality through this function.
+ * The actual data will be emitted through the VBUF interface,
+ * the 'pipeline_statistics' callback to be exact.
+ */
+void
+draw_collect_pipeline_statistics(struct draw_context *draw,
+                                 boolean enable)
+{
+   draw->collect_statistics = enable;
+}
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index 1d25b7f..94fac88 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -264,6 +264,13 @@ void draw_set_driver_clipping( struct draw_context *draw,
 void draw_set_force_passthrough( struct draw_context *draw, 
                                  boolean enable );
 
+
+/*******************************************************************************
+ * Draw statistics
+ */
+void draw_collect_pipeline_statistics(struct draw_context *draw,
+                                      boolean enable);
+
 /*******************************************************************************
  * Draw pipeline 
  */
diff --git a/src/gallium/auxiliary/draw/draw_gs.c b/src/gallium/auxiliary/draw/draw_gs.c
index d95a4c5..f8cb04a 100644
--- a/src/gallium/auxiliary/draw/draw_gs.c
+++ b/src/gallium/auxiliary/draw/draw_gs.c
@@ -353,6 +353,10 @@ static void gs_flush(struct draw_geometry_shader *shader)
 
    unsigned input_primitives = shader->fetched_prim_count;
 
+   if (shader->draw->collect_statistics) {
+      shader->draw->statistics.gs_invocations += input_primitives;
+   }
+
    debug_assert(input_primitives > 0 &&
                 input_primitives <= 4);
 
@@ -493,11 +497,14 @@ int draw_geometry_shader_run(struct draw_geometry_shader *shader,
       input_prim->count;
    unsigned num_in_primitives =
       align(
-         MAX2(u_gs_prims_for_vertices(input_prim->prim, num_input_verts),
-              u_gs_prims_for_vertices(shader->input_primitive, num_input_verts)),
+         MAX2(u_decomposed_prims_for_vertices(input_prim->prim,
+                                              num_input_verts),
+              u_decomposed_prims_for_vertices(shader->input_primitive,
+                                              num_input_verts)),
          shader->vector_length);
-   unsigned max_out_prims = u_gs_prims_for_vertices(shader->output_primitive,
-                                                    shader->max_output_vertices)
+   unsigned max_out_prims =
+      u_decomposed_prims_for_vertices(shader->output_primitive,
+                                      shader->max_output_vertices)
       * num_in_primitives;
 
    //Assume at least one primitive
@@ -593,6 +600,15 @@ int draw_geometry_shader_run(struct draw_geometry_shader *shader,
    output_prims->primitive_count = shader->emitted_primitives;
    output_verts->count = shader->emitted_vertices;
 
+   if (shader->draw->collect_statistics) {
+      unsigned i;
+      for (i = 0; i < shader->emitted_primitives; ++i) {
+         shader->draw->statistics.gs_primitives +=
+            u_decomposed_prims_for_vertices(shader->output_primitive,
+                                            shader->primitive_lengths[i]);
+      }
+   }
+
 #if 0
    debug_printf("GS finished, prims = %d, verts = %d\n",
                 output_prims->primitive_count,
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 757ed26..d6a3e7c 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -312,6 +312,9 @@ struct draw_context
    const struct pipe_sampler_state *samplers[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
    unsigned num_samplers[PIPE_SHADER_TYPES];
 
+   struct pipe_query_data_pipeline_statistics statistics;
+   boolean collect_statistics;
+
    void *driver_private;
 };
 
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 853bd67..10f32fd 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -34,6 +34,7 @@
 #include "draw/draw_gs.h"
 #include "draw/draw_private.h"
 #include "draw/draw_pt.h"
+#include "draw/draw_vbuf.h"
 #include "draw/draw_vs.h"
 #include "tgsi/tgsi_dump.h"
 #include "util/u_math.h"
@@ -545,6 +546,11 @@ draw_vbo(struct draw_context *draw,
       return;
    }
 
+   /* If we're collecting stats then make sure we start from scratch */
+   if (draw->collect_statistics) {
+      memset(&draw->statistics, 0, sizeof(draw->statistics));
+   }
+
    draw->pt.max_index = index_limit - 1;
 
    /*
@@ -562,4 +568,9 @@ draw_vbo(struct draw_context *draw,
          draw_pt_arrays(draw, info->mode, info->start, count);
       }
    }
+
+   /* If requested emit the pipeline statistics for this run */
+   if (draw->collect_statistics) {
+      draw->render->pipeline_statistics(draw->render, &draw->statistics);
+   }
 }
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
index 942ddc4..b3d3546 100644
--- a/src/gallium/auxiliary/draw/draw_pt_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -33,6 +33,7 @@
 #include "draw/draw_pt.h"
 #include "translate/translate.h"
 #include "translate/translate_cache.h"
+#include "util/u_prim.h"
 
 struct pt_emit {
    struct draw_context *draw;
@@ -179,6 +180,12 @@ draw_pt_emit(struct pt_emit *emit,
         i < prim_info->primitive_count;
         start += prim_info->primitive_lengths[i], i++)
    {
+      if (draw->collect_statistics) {
+         draw->statistics.c_invocations +=
+            u_decomposed_prims_for_vertices(prim_info->prim,
+                                            prim_info->primitive_lengths[i]);
+      }
+         
       render->draw_elements(render,
                             elts + start,
                             prim_info->primitive_lengths[i]);
@@ -252,11 +259,17 @@ draw_pt_emit_linear(struct pt_emit *emit,
         i < prim_info->primitive_count;
         start += prim_info->primitive_lengths[i], i++)
    {
+      if (draw->collect_statistics) {
+         draw->statistics.c_invocations +=
+            u_decomposed_prims_for_vertices(prim_info->prim,
+                                            prim_info->primitive_lengths[i]);
+      }
+
       render->draw_arrays(render,
                           start,
                           prim_info->primitive_lengths[i]);
    }
-
+   
    render->release_vertices(render);
 
    return;
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
index 72d3912..e17f161 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
@@ -240,6 +240,12 @@ static void fetch_pipeline_generic( struct draw_pt_middle_end *middle,
       assert(0);
       return;
    }
+   if (draw->collect_statistics) {
+      draw->statistics.ia_vertices += fetch_info->count;
+      draw->statistics.ia_primitives +=
+         u_decomposed_prims_for_vertices(prim_info->prim, fetch_info->count);
+      draw->statistics.vs_invocations += fetch_info->count;
+   }
 
    /* Fetch into our vertex buffer.
     */
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
index ec0f758..d312dc4 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c
@@ -27,6 +27,7 @@
 
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_prim.h"
 #include "draw/draw_context.h"
 #include "draw/draw_gs.h"
 #include "draw/draw_vbuf.h"
@@ -333,6 +334,12 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
       assert(0);
       return;
    }
+   if (draw->collect_statistics) {
+      draw->statistics.ia_vertices += fetch_info->count;
+      draw->statistics.ia_primitives +=
+         u_decomposed_prims_for_vertices(prim_info->prim, fetch_info->count);
+      draw->statistics.vs_invocations += fetch_info->count;
+   }
 
    if (fetch_info->linear)
       clipped = fpme->current_variant->jit_func( &fpme->llvm->jit_context,
diff --git a/src/gallium/auxiliary/draw/draw_vbuf.h b/src/gallium/auxiliary/draw/draw_vbuf.h
index f10d185..bf1c73c 100644
--- a/src/gallium/auxiliary/draw/draw_vbuf.h
+++ b/src/gallium/auxiliary/draw/draw_vbuf.h
@@ -43,6 +43,7 @@
 struct pipe_rasterizer_state;
 struct draw_context;
 struct vertex_info;
+struct pipe_query_data_pipeline_statistics;
 
 
 /**
@@ -126,6 +127,13 @@ struct vbuf_render {
                                    unsigned primitive_count,
                                    unsigned vertices_count,
                                    unsigned primitive_generated );
+
+   /**
+    * Called after all relevant statistics have been accumulated.
+    */
+   void (*pipeline_statistics)(
+      struct vbuf_render *vbufr,
+      const struct pipe_query_data_pipeline_statistics *stats );
 };
 
 
diff --git a/src/gallium/auxiliary/util/u_prim.h b/src/gallium/auxiliary/util/u_prim.h
index d62c636..99bb66c 100644
--- a/src/gallium/auxiliary/util/u_prim.h
+++ b/src/gallium/auxiliary/util/u_prim.h
@@ -165,12 +165,14 @@ u_vertices_per_prim(int primitive)
 /**
  * Returns the number of decomposed primitives for the given
  * vertex count.
- * Geometry shader is invoked once for each triangle in
+ * Parts of the pipline are invoked once for each triangle in
  * triangle strip, triangle fans and triangles and once
- * for each line in line strip, line loop, lines.
+ * for each line in line strip, line loop, lines. Also 
+ * statistics depend on knowing the exact number of decomposed
+ * primitives for a set of vertices.
  */
 static INLINE unsigned
-u_gs_prims_for_vertices(int primitive, int vertices)
+u_decomposed_prims_for_vertices(int primitive, int vertices)
 {
    switch(primitive) {
    case PIPE_PRIM_POINTS:
@@ -180,31 +182,34 @@ u_gs_prims_for_vertices(int primitive, int vertices)
    case PIPE_PRIM_LINE_LOOP:
       return vertices;
    case PIPE_PRIM_LINE_STRIP:
-      return vertices - 1;
+      return (vertices > 1) ? vertices - 1 : 0;
    case PIPE_PRIM_TRIANGLES:
       return vertices /  3;
    case PIPE_PRIM_TRIANGLE_STRIP:
-      return vertices - 2;
+      return (vertices > 2) ? vertices - 2 : 0;
    case PIPE_PRIM_TRIANGLE_FAN:
-      return vertices - 2;
+      return (vertices > 2) ? vertices - 2 : 0;
    case PIPE_PRIM_LINES_ADJACENCY:
       return vertices / 2;
    case PIPE_PRIM_LINE_STRIP_ADJACENCY:
-      return vertices - 1;
+      return (vertices > 1) ? vertices - 1 : 0;
    case PIPE_PRIM_TRIANGLES_ADJACENCY:
       return vertices / 3;
    case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
-      return vertices - 2;
-
-   /* following primitives should never be used
-    * with geometry shaders abd their size is
-    * undefined */
-   case PIPE_PRIM_POLYGON:
+      return (vertices > 2) ? vertices - 2 : 0;
    case PIPE_PRIM_QUADS:
+      return vertices / 4;
    case PIPE_PRIM_QUAD_STRIP:
+      return (vertices > 4) ? (vertices - 2) / 2 : 0;
+   /* Polygons can't be decomposed
+    * because the number of their vertices isn't known so
+    * for them and whatever else we don't recognize just
+    * return 1 if the number of vertices is greater than
+    * 3 and zero otherwise */
+   case PIPE_PRIM_POLYGON:
    default:
-      debug_printf("Unrecognized geometry shader primitive");
-      return 3;
+      debug_printf("Invalid decomposition primitive!\n");
+      return (vertices > 3) ? 1 : 0;
    }
 }