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>

23 files changed, 258 insertions, 22 deletions

diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 2fb9bacf4c..0f98021247 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 1d25b7f255..94fac880c6 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 d95a4c5705..f8cb04a8f3 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 757ed2645e..d6a3e7c2ce 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 853bd6716e..10f32fd929 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 942ddc4d7b..b3d3546402 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 72d391257a..e17f16147b 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 ec0f758d45..d312dc4828 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 f10d185868..bf1c73cad5 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 d62c636f2e..99bb66c9b8 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;
    }
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index 6ee7b99f47..d605dba226 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -94,6 +94,9 @@ struct llvmpipe_context {
    struct pipe_query_data_so_statistics so_stats;
    unsigned num_primitives_generated;
 
+   struct pipe_query_data_pipeline_statistics pipeline_statistics;
+   unsigned active_statistics_queries;
+
    unsigned dirty; /**< Mask of LP_NEW_x flags */
 
    unsigned active_occlusion_query;
diff --git a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
index efeca25575..e8a5880dee 100644
--- a/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
+++ b/src/gallium/drivers/llvmpipe/lp_draw_arrays.c
@@ -108,6 +108,8 @@ llvmpipe_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
          draw_vs_attach_so(lp->vs->draw_data, &lp->gs->shader.stream_output);
       }
    }
+   draw_collect_pipeline_statistics(draw,
+                                    lp->active_statistics_queries > 0);
 
    /* draw! */
    draw_vbo(draw, info);
diff --git a/src/gallium/drivers/llvmpipe/lp_query.c b/src/gallium/drivers/llvmpipe/lp_query.c
index 013d1928c1..96e1e3ff1f 100644
--- a/src/gallium/drivers/llvmpipe/lp_query.c
+++ b/src/gallium/drivers/llvmpipe/lp_query.c
@@ -144,6 +144,12 @@ llvmpipe_get_query_result(struct pipe_context *pipe,
       stats->primitives_storage_needed = pq->num_primitives_generated;
    }
       break;
+   case PIPE_QUERY_PIPELINE_STATISTICS: {
+      struct pipe_query_data_pipeline_statistics *stats =
+         (struct pipe_query_data_pipeline_statistics *)vresult;
+      *stats = pq->stats;
+   }
+      break;
    default:
       assert(0);
       break;
@@ -188,6 +194,16 @@ llvmpipe_begin_query(struct pipe_context *pipe, struct pipe_query *q)
       llvmpipe->num_primitives_generated = 0;
    }
 
+   if (pq->type == PIPE_QUERY_PIPELINE_STATISTICS) {
+      /* reset our cache */
+      if (llvmpipe->active_statistics_queries == 0) {
+         memset(&llvmpipe->pipeline_statistics, 0,
+                sizeof(llvmpipe->pipeline_statistics));
+      }
+      memcpy(&pq->stats, &llvmpipe->pipeline_statistics, sizeof(pq->stats));
+      llvmpipe->active_statistics_queries++;
+   }
+
    if (pq->type == PIPE_QUERY_OCCLUSION_COUNTER) {
       llvmpipe->active_occlusion_query = TRUE;
       llvmpipe->dirty |= LP_NEW_OCCLUSION_QUERY;
@@ -216,6 +232,27 @@ llvmpipe_end_query(struct pipe_context *pipe, struct pipe_query *q)
       pq->num_primitives_generated = llvmpipe->num_primitives_generated;
    }
 
+   if (pq->type == PIPE_QUERY_PIPELINE_STATISTICS) {
+      pq->stats.ia_vertices =
+         llvmpipe->pipeline_statistics.ia_vertices - pq->stats.ia_vertices;
+      pq->stats.ia_primitives =
+         llvmpipe->pipeline_statistics.ia_primitives - pq->stats.ia_primitives;
+      pq->stats.vs_invocations =
+         llvmpipe->pipeline_statistics.vs_invocations - pq->stats.vs_invocations;
+      pq->stats.gs_invocations =
+         llvmpipe->pipeline_statistics.gs_invocations - pq->stats.gs_invocations;
+      pq->stats.gs_primitives =
+         llvmpipe->pipeline_statistics.gs_primitives - pq->stats.gs_primitives;
+      pq->stats.c_invocations =
+         llvmpipe->pipeline_statistics.c_invocations - pq->stats.c_invocations;
+      pq->stats.c_primitives =
+         llvmpipe->pipeline_statistics.c_primitives - pq->stats.c_primitives;
+      pq->stats.ps_invocations =
+         llvmpipe->pipeline_statistics.ps_invocations - pq->stats.ps_invocations;
+
+      llvmpipe->active_statistics_queries--;
+   }
+
    if (pq->type == PIPE_QUERY_OCCLUSION_COUNTER) {
       assert(llvmpipe->active_occlusion_query);
       llvmpipe->active_occlusion_query = FALSE;
diff --git a/src/gallium/drivers/llvmpipe/lp_query.h b/src/gallium/drivers/llvmpipe/lp_query.h
index 19d977fd2e..68dfe99f42 100644
--- a/src/gallium/drivers/llvmpipe/lp_query.h
+++ b/src/gallium/drivers/llvmpipe/lp_query.h
@@ -47,6 +47,8 @@ struct llvmpipe_query {
    unsigned type;                   /* PIPE_QUERY_* */
    unsigned num_primitives_generated;
    unsigned num_primitives_written;
+
+   struct pipe_query_data_pipeline_statistics stats;
 };
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c
index 6cb2702056..29dd933c64 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast.c
@@ -477,6 +477,7 @@ lp_rast_begin_query(struct lp_rasterizer_task *task,
    case PIPE_QUERY_PRIMITIVES_GENERATED:
    case PIPE_QUERY_PRIMITIVES_EMITTED:
    case PIPE_QUERY_SO_STATISTICS:
+   case PIPE_QUERY_PIPELINE_STATISTICS:
       break;
    default:
       assert(0);
@@ -509,6 +510,7 @@ lp_rast_end_query(struct lp_rasterizer_task *task,
    case PIPE_QUERY_PRIMITIVES_GENERATED:
    case PIPE_QUERY_PRIMITIVES_EMITTED:
    case PIPE_QUERY_SO_STATISTICS:
+   case PIPE_QUERY_PIPELINE_STATISTICS:
       break;
    default:
       assert(0);
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
index 9e0a5b3995..8173994e5c 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
@@ -547,6 +547,28 @@ lp_setup_so_info(struct vbuf_render *vbr, uint primitives, uint vertices,
    lp->num_primitives_generated += prim_generated;
 }
 
+static void
+lp_setup_pipeline_statistics(
+   struct vbuf_render *vbr,
+   const struct pipe_query_data_pipeline_statistics *stats)
+{
+   struct lp_setup_context *setup = lp_setup_context(vbr);
+   struct llvmpipe_context *llvmpipe = llvmpipe_context(setup->pipe);
+
+   llvmpipe->pipeline_statistics.ia_vertices +=
+      stats->ia_vertices;
+   llvmpipe->pipeline_statistics.ia_primitives +=
+      stats->ia_primitives;
+   llvmpipe->pipeline_statistics.vs_invocations +=
+      stats->vs_invocations;
+   llvmpipe->pipeline_statistics.gs_invocations +=
+      stats->gs_invocations;
+   llvmpipe->pipeline_statistics.gs_primitives +=
+      stats->gs_primitives;
+   llvmpipe->pipeline_statistics.c_invocations +=
+      stats->c_invocations;
+}
+
 /**
  * Create the post-transform vertex handler for the given context.
  */
@@ -566,4 +588,5 @@ lp_setup_init_vbuf(struct lp_setup_context *setup)
    setup->base.release_vertices = lp_setup_release_vertices;
    setup->base.destroy = lp_setup_vbuf_destroy;
    setup->base.set_stream_output_info = lp_setup_so_info;
+   setup->base.pipeline_statistics = lp_setup_pipeline_statistics;
 }
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index d78352ab63..431864ae08 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -92,6 +92,9 @@ struct softpipe_context {
    struct pipe_query_data_so_statistics so_stats;
    unsigned num_primitives_generated;
 
+   struct pipe_query_data_pipeline_statistics pipeline_statistics;
+   unsigned active_statistics_queries;
+
    unsigned num_samplers[PIPE_SHADER_TYPES];
    unsigned num_sampler_views[PIPE_SHADER_TYPES];
 
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
index ba9e7817ce..4378312bc3 100644
--- a/src/gallium/drivers/softpipe/sp_draw_arrays.c
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -105,6 +105,9 @@ softpipe_draw_vbo(struct pipe_context *pipe,
    draw_set_mapped_so_targets(draw, sp->num_so_targets,
                               sp->so_targets);
 
+   draw_collect_pipeline_statistics(draw,
+                                    sp->active_statistics_queries > 0);
+
    /* draw! */
    draw_vbo(draw, info);
 
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index bb5ebcc7ae..5d0b5e1c58 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -600,6 +600,28 @@ sp_vbuf_so_info(struct vbuf_render *vbr, uint primitives, uint vertices,
    softpipe->num_primitives_generated += prim_generated;
 }
 
+static void
+sp_vbuf_pipeline_statistics(
+   struct vbuf_render *vbr, 
+   const struct pipe_query_data_pipeline_statistics *stats)
+{
+   struct softpipe_vbuf_render *cvbr = softpipe_vbuf_render(vbr);
+   struct softpipe_context *softpipe = cvbr->softpipe;
+
+   softpipe->pipeline_statistics.ia_vertices +=
+      stats->ia_vertices;
+   softpipe->pipeline_statistics.ia_primitives +=
+      stats->ia_primitives;
+   softpipe->pipeline_statistics.vs_invocations +=
+      stats->vs_invocations;
+   softpipe->pipeline_statistics.gs_invocations +=
+      stats->gs_invocations;
+   softpipe->pipeline_statistics.gs_primitives +=
+      stats->gs_primitives;
+   softpipe->pipeline_statistics.c_invocations +=
+      stats->c_invocations;
+}
+
 
 static void
 sp_vbuf_destroy(struct vbuf_render *vbr)
@@ -634,6 +656,7 @@ sp_create_vbuf_backend(struct softpipe_context *sp)
    cvbr->base.draw_arrays = sp_vbuf_draw_arrays;
    cvbr->base.release_vertices = sp_vbuf_release_vertices;
    cvbr->base.set_stream_output_info = sp_vbuf_so_info;
+   cvbr->base.pipeline_statistics = sp_vbuf_pipeline_statistics;
    cvbr->base.destroy = sp_vbuf_destroy;
 
    cvbr->softpipe = sp;
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index f32fe5f406..d379f6457e 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -73,6 +73,11 @@ shade_quad(struct quad_stage *qs, struct quad_header *quad)
    struct softpipe_context *softpipe = qs->softpipe;
    struct tgsi_exec_machine *machine = softpipe->fs_machine;
 
+   if (softpipe->active_statistics_queries) {
+      softpipe->pipeline_statistics.ps_invocations +=
+         util_bitcount(quad->inout.mask);         
+   }
+
    /* run shader */
    machine->flatshade_color = softpipe->rasterizer->flatshade ? TRUE : FALSE;
    return softpipe->fs_variant->run( softpipe->fs_variant, machine, quad );
diff --git a/src/gallium/drivers/softpipe/sp_query.c b/src/gallium/drivers/softpipe/sp_query.c
index 2dd82c81d6..62411de503 100644
--- a/src/gallium/drivers/softpipe/sp_query.c
+++ b/src/gallium/drivers/softpipe/sp_query.c
@@ -43,6 +43,8 @@ struct softpipe_query {
    uint64_t end;
    struct pipe_query_data_so_statistics so;
    unsigned num_primitives_generated;
+
+   struct pipe_query_data_pipeline_statistics stats;
 };
 
 
@@ -61,7 +63,8 @@ softpipe_create_query(struct pipe_context *pipe,
           type == PIPE_QUERY_TIME_ELAPSED ||
           type == PIPE_QUERY_SO_STATISTICS ||
           type == PIPE_QUERY_PRIMITIVES_EMITTED ||
-          type == PIPE_QUERY_PRIMITIVES_GENERATED ||
+          type == PIPE_QUERY_PRIMITIVES_GENERATED || 
+          type == PIPE_QUERY_PIPELINE_STATISTICS ||
           type == PIPE_QUERY_GPU_FINISHED ||
           type == PIPE_QUERY_TIMESTAMP ||
           type == PIPE_QUERY_TIMESTAMP_DISJOINT);
@@ -106,6 +109,16 @@ softpipe_begin_query(struct pipe_context *pipe, struct pipe_query *q)
    case PIPE_QUERY_TIMESTAMP:
    case PIPE_QUERY_GPU_FINISHED:
       break;
+   case PIPE_QUERY_PIPELINE_STATISTICS:
+      /* reset our cache */
+      if (softpipe->active_statistics_queries == 0) {
+         memset(&softpipe->pipeline_statistics, 0,
+                sizeof(softpipe->pipeline_statistics));
+      }
+      memcpy(&sq->stats, &softpipe->pipeline_statistics,
+             sizeof(sq->stats));
+      softpipe->active_statistics_queries++;
+      break;
    default:
       assert(0);
       break;
@@ -145,6 +158,26 @@ softpipe_end_query(struct pipe_context *pipe, struct pipe_query *q)
       break;
    case PIPE_QUERY_GPU_FINISHED:
       break;
+   case PIPE_QUERY_PIPELINE_STATISTICS:
+      sq->stats.ia_vertices =
+         softpipe->pipeline_statistics.ia_vertices - sq->stats.ia_vertices;
+      sq->stats.ia_primitives =
+         softpipe->pipeline_statistics.ia_primitives - sq->stats.ia_primitives;
+      sq->stats.vs_invocations =
+         softpipe->pipeline_statistics.vs_invocations - sq->stats.vs_invocations;
+      sq->stats.gs_invocations =
+         softpipe->pipeline_statistics.gs_invocations - sq->stats.gs_invocations;
+      sq->stats.gs_primitives =
+         softpipe->pipeline_statistics.gs_primitives - sq->stats.gs_primitives;
+      sq->stats.c_invocations =
+         softpipe->pipeline_statistics.c_invocations - sq->stats.c_invocations;
+      sq->stats.c_primitives =
+         softpipe->pipeline_statistics.c_primitives - sq->stats.c_primitives;
+      sq->stats.ps_invocations =
+         softpipe->pipeline_statistics.ps_invocations - sq->stats.ps_invocations;
+
+      softpipe->active_statistics_queries--;
+      break;
    default:
       assert(0);
       break;
@@ -167,6 +200,10 @@ softpipe_get_query_result(struct pipe_context *pipe,
       memcpy(vresult, &sq->so,
              sizeof(struct pipe_query_data_so_statistics));
       break;
+   case PIPE_QUERY_PIPELINE_STATISTICS:
+      memcpy(vresult, &sq->stats,
+             sizeof(struct pipe_query_data_pipeline_statistics));;
+      break;
    case PIPE_QUERY_GPU_FINISHED:
       *result = TRUE;
       break;
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index 6915f91800..b2b7f2e320 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -87,7 +87,7 @@ softpipe_get_param(struct pipe_screen *screen, enum pipe_cap param)
    case PIPE_CAP_QUERY_TIME_ELAPSED:
       return 1;
    case PIPE_CAP_QUERY_PIPELINE_STATISTICS:
-      return 0;
+      return 1;
    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
       return 1;
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
index 3dad131d88..08cd4da5bd 100644
--- a/src/gallium/drivers/softpipe/sp_setup.c
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -857,6 +857,10 @@ sp_setup_tri(struct setup_context *setup,
 
    flush_spans( setup );
 
+   if (setup->softpipe->active_statistics_queries) {
+      setup->softpipe->pipeline_statistics.c_primitives++;
+   }
+
 #if DEBUG_FRAGS
    printf("Tri: %u frags emitted, %u written\n",
           setup->numFragsEmitted,