mesa: move num_instances and base_instance out of _mesa_prim

They are never used by multi draws and internal draws.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3990>

26 files changed, 120 insertions, 78 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c
index 10c599e796c..2b702297a68 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.c
+++ b/src/mesa/drivers/dri/i965/brw_draw.c
@@ -160,6 +160,7 @@ brw_emit_prim(struct brw_context *brw,
               const struct _mesa_prim *prim,
               uint32_t hw_prim,
               bool is_indexed,
+              GLuint num_instances, GLuint base_instance,
               struct brw_transform_feedback_object *xfb_obj,
               unsigned stream,
               bool is_indirect,
@@ -217,7 +218,7 @@ brw_emit_prim(struct brw_context *brw,
       BEGIN_BATCH(9);
       OUT_BATCH(MI_LOAD_REGISTER_IMM | (9 - 2));
       OUT_BATCH(GEN7_3DPRIM_INSTANCE_COUNT);
-      OUT_BATCH(prim->num_instances);
+      OUT_BATCH(num_instances);
       OUT_BATCH(GEN7_3DPRIM_START_VERTEX);
       OUT_BATCH(0);
       OUT_BATCH(GEN7_3DPRIM_BASE_VERTEX);
@@ -270,8 +271,8 @@ brw_emit_prim(struct brw_context *brw,
    }
    OUT_BATCH(verts_per_instance);
    OUT_BATCH(start_vertex_location);
-   OUT_BATCH(prim->num_instances);
-   OUT_BATCH(prim->base_instance);
+   OUT_BATCH(num_instances);
+   OUT_BATCH(base_instance);
    OUT_BATCH(base_vertex_location);
    ADVANCE_BATCH();
 
@@ -899,7 +900,7 @@ brw_finish_drawing(struct gl_context *ctx)
  */
 static void
 gen9_emit_preempt_wa(struct brw_context *brw,
-                     const struct _mesa_prim *prim)
+                     const struct _mesa_prim *prim, GLuint num_instances)
 {
    bool object_preemption = true;
    ASSERTED const struct gen_device_info *devinfo = &brw->screen->devinfo;
@@ -944,7 +945,7 @@ gen9_emit_preempt_wa(struct brw_context *brw,
     *
     *    WA: Disable preemption when using instanceing.
     */
-   if (prim->num_instances > 1)
+   if (num_instances > 1)
       object_preemption = false;
 
    brw_enable_obj_preemption(brw, object_preemption);
@@ -958,6 +959,7 @@ brw_draw_single_prim(struct gl_context *ctx,
                      const struct _mesa_prim *prim,
                      unsigned prim_id,
                      bool is_indexed,
+                     GLuint num_instances, GLuint base_instance,
                      struct brw_transform_feedback_object *xfb_obj,
                      unsigned stream,
                      GLsizeiptr indirect_offset)
@@ -980,12 +982,12 @@ brw_draw_single_prim(struct gl_context *ctx,
    intel_batchbuffer_save_state(brw);
    fail_next = intel_batchbuffer_saved_state_is_empty(brw);
 
-   if (brw->num_instances != prim->num_instances ||
+   if (brw->num_instances != num_instances ||
        brw->basevertex != prim->basevertex ||
-       brw->baseinstance != prim->base_instance) {
-      brw->num_instances = prim->num_instances;
+       brw->baseinstance != base_instance) {
+      brw->num_instances = num_instances;
       brw->basevertex = prim->basevertex;
-      brw->baseinstance = prim->base_instance;
+      brw->baseinstance = base_instance;
       if (prim_id > 0) { /* For i == 0 we just did this before the loop */
          brw->ctx.NewDriverState |= BRW_NEW_VERTICES;
          brw_merge_inputs(brw);
@@ -999,7 +1001,7 @@ brw_draw_single_prim(struct gl_context *ctx,
     */
    const int new_firstvertex =
       is_indexed ? prim->basevertex : prim->start;
-   const int new_baseinstance = prim->base_instance;
+   const int new_baseinstance = base_instance;
    const struct brw_vs_prog_data *vs_prog_data =
       brw_vs_prog_data(brw->vs.base.prog_data);
    if (prim_id > 0) {
@@ -1068,10 +1070,11 @@ retry:
    }
 
    if (devinfo->gen == 9)
-      gen9_emit_preempt_wa(brw, prim);
+      gen9_emit_preempt_wa(brw, prim, num_instances);
 
-   brw_emit_prim(brw, prim, brw->primitive, is_indexed, xfb_obj, stream,
-                 is_indirect, indirect_offset);
+   brw_emit_prim(brw, prim, brw->primitive, is_indexed, num_instances,
+                 base_instance, xfb_obj, stream, is_indirect,
+                 indirect_offset);
 
    brw->batch.no_wrap = false;
 
@@ -1108,6 +1111,8 @@ brw_draw_prims(struct gl_context *ctx,
                GLboolean index_bounds_valid,
                GLuint min_index,
                GLuint max_index,
+               GLuint num_instances,
+               GLuint base_instance,
                struct gl_transform_feedback_object *gl_xfb_obj,
                unsigned stream)
 {
@@ -1121,7 +1126,8 @@ brw_draw_prims(struct gl_context *ctx,
       return;
 
    /* Handle primitive restart if needed */
-   if (brw_handle_primitive_restart(ctx, prims, nr_prims, ib)) {
+   if (brw_handle_primitive_restart(ctx, prims, nr_prims, ib, num_instances,
+                                    base_instance)) {
       /* The draw was handled, so we can exit now */
       return;
    }
@@ -1134,8 +1140,8 @@ brw_draw_prims(struct gl_context *ctx,
                  _mesa_enum_to_string(ctx->RenderMode));
       _swsetup_Wakeup(ctx);
       _tnl_wakeup(ctx);
-      _tnl_draw(ctx, prims, nr_prims, ib,
-                index_bounds_valid, min_index, max_index, NULL, 0);
+      _tnl_draw(ctx, prims, nr_prims, ib, index_bounds_valid, min_index,
+                max_index, num_instances, base_instance, NULL, 0);
       return;
    }
 
@@ -1187,7 +1193,8 @@ brw_draw_prims(struct gl_context *ctx,
          brw->predicate.state = BRW_PREDICATE_STATE_USE_BIT;
       }
 
-      brw_draw_single_prim(ctx, &prims[i], i, ib != NULL, xfb_obj, stream,
+      brw_draw_single_prim(ctx, &prims[i], i, ib != NULL, num_instances,
+                           base_instance, xfb_obj, stream,
                            brw->draw.draw_indirect_offset +
                            brw->draw.draw_indirect_stride * i);
    }
@@ -1239,9 +1246,7 @@ brw_draw_indirect_prims(struct gl_context *ctx,
 
    brw->draw.draw_indirect_data = indirect_data;
 
-   brw_draw_prims(ctx, prim, draw_count,
-                  ib, false, 0, ~0,
-                  NULL, 0);
+   brw_draw_prims(ctx, prim, draw_count, ib, false, 0, ~0, 0, 0, NULL, 0);
 
    brw->draw.draw_indirect_data = NULL;
    free(prim);
diff --git a/src/mesa/drivers/dri/i965/brw_draw.h b/src/mesa/drivers/dri/i965/brw_draw.h
index ddbf6238a17..c2e7d26f3ca 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.h
+++ b/src/mesa/drivers/dri/i965/brw_draw.h
@@ -51,6 +51,8 @@ void brw_draw_prims(struct gl_context *ctx,
 		     GLboolean index_bounds_valid,
 		     GLuint min_index,
 		     GLuint max_index,
+                     GLuint num_instances,
+                     GLuint base_instance,
 		     struct gl_transform_feedback_object *unused_tfb_object,
                      unsigned stream);
 
@@ -65,7 +67,8 @@ GLboolean
 brw_handle_primitive_restart(struct gl_context *ctx,
                              const struct _mesa_prim *prims,
                              GLuint nr_prims,
-                             const struct _mesa_index_buffer *ib);
+                             const struct _mesa_index_buffer *ib,
+                             GLuint num_instances, GLuint base_instance);
 
 void
 brw_draw_indirect_prims(struct gl_context *ctx,
diff --git a/src/mesa/drivers/dri/i965/brw_primitive_restart.c b/src/mesa/drivers/dri/i965/brw_primitive_restart.c
index 158f9477939..c539cd78476 100644
--- a/src/mesa/drivers/dri/i965/brw_primitive_restart.c
+++ b/src/mesa/drivers/dri/i965/brw_primitive_restart.c
@@ -130,7 +130,8 @@ GLboolean
 brw_handle_primitive_restart(struct gl_context *ctx,
                              const struct _mesa_prim *prims,
                              GLuint nr_prims,
-                             const struct _mesa_index_buffer *ib)
+                             const struct _mesa_index_buffer *ib,
+                             GLuint num_instances, GLuint base_instance)
 {
    struct brw_context *brw = brw_context(ctx);
 
@@ -162,7 +163,8 @@ brw_handle_primitive_restart(struct gl_context *ctx,
       /* Cut index should work for primitive restart, so use it
        */
       brw->prim_restart.enable_cut_index = true;
-      brw_draw_prims(ctx, prims, nr_prims, ib, GL_FALSE, -1, -1, NULL, 0);
+      brw_draw_prims(ctx, prims, nr_prims, ib, GL_FALSE, -1, -1,
+                     num_instances, base_instance, NULL, 0);
       brw->prim_restart.enable_cut_index = false;
    } else {
       /* Not all the primitive draw modes are supported by the cut index,
@@ -173,8 +175,8 @@ brw_handle_primitive_restart(struct gl_context *ctx,
       /* Clear this to make the draw direct. */
       brw->draw.draw_indirect_data = NULL;
 
-      vbo_sw_primitive_restart(ctx, prims, nr_prims, ib,
-                               indirect_data,
+      vbo_sw_primitive_restart(ctx, prims, nr_prims, ib, num_instances,
+                               base_instance, indirect_data,
                                brw->draw.draw_indirect_offset);
    }
 
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
index d365f7a531a..83fb9596d58 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
@@ -244,6 +244,7 @@ TAG(vbo_render_prims)(struct gl_context *ctx,
 		      const struct _mesa_index_buffer *ib,
 		      GLboolean index_bounds_valid,
 		      GLuint min_index, GLuint max_index,
+                      GLuint num_instances, GLuint base_instance,
 		      struct gl_transform_feedback_object *tfb_vertcount,
                       unsigned stream);
 
@@ -251,7 +252,8 @@ static GLboolean
 vbo_maybe_split(struct gl_context *ctx, const struct tnl_vertex_array *arrays,
 	    const struct _mesa_prim *prims, GLuint nr_prims,
 	    const struct _mesa_index_buffer *ib,
-	    GLuint min_index, GLuint max_index)
+	    GLuint min_index, GLuint max_index,
+            GLuint num_instances, GLuint base_instance)
 {
 	struct nouveau_context *nctx = to_nouveau_context(ctx);
 	struct nouveau_render_state *render = to_render_state(ctx);
@@ -277,7 +279,8 @@ vbo_maybe_split(struct gl_context *ctx, const struct tnl_vertex_array *arrays,
 		};
 
 		_tnl_split_prims(ctx, arrays, prims, nr_prims, ib, min_index,
-				 max_index, TAG(vbo_render_prims), &limits);
+				 max_index, num_instances, base_instance,
+                                 TAG(vbo_render_prims), &limits);
 		return GL_TRUE;
 	}
 
@@ -481,6 +484,7 @@ TAG(vbo_render_prims)(struct gl_context *ctx,
 		      const struct _mesa_index_buffer *ib,
 		      GLboolean index_bounds_valid,
 		      GLuint min_index, GLuint max_index,
+                      GLuint num_instances, GLuint base_instance,
 		      struct gl_transform_feedback_object *tfb_vertcount,
                       unsigned stream)
 {
@@ -494,7 +498,7 @@ TAG(vbo_render_prims)(struct gl_context *ctx,
 	vbo_choose_attrs(ctx, arrays);
 
 	if (vbo_maybe_split(ctx, arrays, prims, nr_prims, ib, min_index,
-			    max_index))
+			    max_index, num_instances, base_instance))
 		return;
 
 	vbo_init_arrays(ctx, ib, arrays);
@@ -518,6 +522,7 @@ TAG(vbo_check_render_prims)(struct gl_context *ctx,
 			    const struct _mesa_index_buffer *ib,
 			    GLboolean index_bounds_valid,
 			    GLuint min_index, GLuint max_index,
+                            GLuint num_instances, GLuint base_instance,
 			    struct gl_transform_feedback_object *tfb_vertcount,
                             unsigned stream)
 {
@@ -528,11 +533,13 @@ TAG(vbo_check_render_prims)(struct gl_context *ctx,
 	if (nctx->fallback == HWTNL)
 		TAG(vbo_render_prims)(ctx, arrays, prims, nr_prims, ib,
 				      index_bounds_valid, min_index, max_index,
-				      tfb_vertcount, stream);
+				      num_instances, base_instance,
+                                      tfb_vertcount, stream);
 
 	if (nctx->fallback == SWTNL)
 		_tnl_draw_prims(ctx, arrays, prims, nr_prims, ib,
 				index_bounds_valid, min_index, max_index,
+                                num_instances, base_instance,
 				tfb_vertcount, stream);
 }
 
@@ -542,6 +549,7 @@ TAG(vbo_draw)(struct gl_context *ctx,
 	      const struct _mesa_index_buffer *ib,
 	      GLboolean index_bounds_valid,
 	      GLuint min_index, GLuint max_index,
+              GLuint num_instances, GLuint base_instance,
 	      struct gl_transform_feedback_object *tfb_vertcount,
 	      unsigned stream)
 {
@@ -551,6 +559,7 @@ TAG(vbo_draw)(struct gl_context *ctx,
 	TAG(vbo_check_render_prims)(ctx, arrays,
 				    prims, nr_prims, ib,
 				    index_bounds_valid, min_index, max_index,
+                                    num_instances, base_instance,
 				    tfb_vertcount, stream);
 }
 
diff --git a/src/mesa/main/dd.h b/src/mesa/main/dd.h
index 824f320668b..3fd47ba6f9e 100644
--- a/src/mesa/main/dd.h
+++ b/src/mesa/main/dd.h
@@ -528,6 +528,8 @@ struct dd_function_table {
     * \param index_bounds_valid  are min_index and max_index valid?
     * \param min_index  lowest vertex index used
     * \param max_index  highest vertex index used
+    * \param num_instances  instance count from ARB_draw_instanced
+    * \param base_instance  base instance from ARB_base_instance
     * \param tfb_vertcount  if non-null, indicates which transform feedback
     *                       object has the vertex count.
     * \param tfb_stream  If called via DrawTransformFeedbackStream, specifies
@@ -542,6 +544,7 @@ struct dd_function_table {
                 const struct _mesa_index_buffer *ib,
                 GLboolean index_bounds_valid,
                 GLuint min_index, GLuint max_index,
+                GLuint num_instances, GLuint base_instance,
                 struct gl_transform_feedback_object *tfb_vertcount,
                 unsigned tfb_stream);
 
diff --git a/src/mesa/main/draw.c b/src/mesa/main/draw.c
index 7fb57f67290..02efab74db1 100644
--- a/src/mesa/main/draw.c
+++ b/src/mesa/main/draw.c
@@ -362,15 +362,14 @@ _mesa_draw_arrays(struct gl_context *ctx, GLenum mode, GLint start,
       .begin = 1,
       .end = 1,
       .mode = mode,
-      .num_instances = numInstances,
-      .base_instance = baseInstance,
       .draw_id = drawID,
       .start = start,
       .count = count,
    };
 
    ctx->Driver.Draw(ctx, &prim, 1, NULL,
-                    GL_TRUE, start, start + count - 1, NULL, 0);
+                    GL_TRUE, start, start + count - 1,
+                    numInstances, baseInstance, NULL, 0);
 
    if (MESA_DEBUG_FLAGS & DEBUG_ALWAYS_FLUSH) {
       _mesa_flush(ctx);
@@ -773,8 +772,6 @@ _mesa_validated_drawrangeelements(struct gl_context *ctx, GLenum mode,
    prim.start = 0;
    prim.count = count;
    prim.basevertex = basevertex;
-   prim.num_instances = numInstances;
-   prim.base_instance = baseInstance;
    prim.draw_id = 0;
 
    /* Need to give special consideration to rendering a range of
@@ -809,7 +806,8 @@ _mesa_validated_drawrangeelements(struct gl_context *ctx, GLenum mode,
     */
 
    ctx->Driver.Draw(ctx, &prim, 1, &ib,
-                    index_bounds_valid, start, end, NULL, 0);
+                    index_bounds_valid, start, end,
+                    numInstances, baseInstance, NULL, 0);
 
    if (MESA_DEBUG_FLAGS & DEBUG_ALWAYS_FLUSH) {
       _mesa_flush(ctx);
@@ -1229,8 +1227,6 @@ _mesa_validated_multidrawelements(struct gl_context *ctx, GLenum mode,
          prim[i].start =
             ((uintptr_t) indices[i] - min_index_ptr) / index_type_size;
          prim[i].count = count[i];
-         prim[i].num_instances = 1;
-         prim[i].base_instance = 0;
          prim[i].draw_id = i;
          if (basevertex != NULL)
             prim[i].basevertex = basevertex[i];
@@ -1239,7 +1235,7 @@ _mesa_validated_multidrawelements(struct gl_context *ctx, GLenum mode,
       }
 
       ctx->Driver.Draw(ctx, prim, primcount, &ib,
-                       false, 0, ~0, NULL, 0);
+                       false, 0, ~0, 1, 0, NULL, 0);
    }
    else {
       /* render one prim at a time */
@@ -1256,15 +1252,13 @@ _mesa_validated_multidrawelements(struct gl_context *ctx, GLenum mode,
          prim[0].mode = mode;
          prim[0].start = 0;
          prim[0].count = count[i];
-         prim[0].num_instances = 1;
-         prim[0].base_instance = 0;
          prim[0].draw_id = i;
          if (basevertex != NULL)
             prim[0].basevertex = basevertex[i];
          else
             prim[0].basevertex = 0;
 
-         ctx->Driver.Draw(ctx, prim, 1, &ib, false, 0, ~0, NULL, 0);
+         ctx->Driver.Draw(ctx, prim, 1, &ib, false, 0, ~0, 1, 0, NULL, 0);
       }
    }
 
@@ -1374,14 +1368,13 @@ _mesa_draw_transform_feedback(struct gl_context *ctx, GLenum mode,
    prim.begin = 1;
    prim.end = 1;
    prim.mode = mode;
-   prim.num_instances = numInstances;
-   prim.base_instance = 0;
 
    /* Maybe we should do some primitive splitting for primitive restart
     * (like in DrawArrays), but we have no way to know how many vertices
     * will be rendered. */
 
-   ctx->Driver.Draw(ctx, &prim, 1, NULL, GL_FALSE, 0, ~0, obj, stream);
+   ctx->Driver.Draw(ctx, &prim, 1, NULL, GL_FALSE, 0, ~0, numInstances, 0,
+                    obj, stream);
 
    if (MESA_DEBUG_FLAGS & DEBUG_ALWAYS_FLUSH) {
       _mesa_flush(ctx);
diff --git a/src/mesa/main/draw.h b/src/mesa/main/draw.h
index 1363ff58df2..3a7dedea529 100644
--- a/src/mesa/main/draw.h
+++ b/src/mesa/main/draw.h
@@ -61,8 +61,6 @@ struct _mesa_prim
    GLuint start;
    GLuint count;
    GLint basevertex;
-   GLuint num_instances;
-   GLuint base_instance;
    GLuint draw_id;
 };
 
diff --git a/src/mesa/state_tracker/st_cb_rasterpos.c b/src/mesa/state_tracker/st_cb_rasterpos.c
index 7a009d3a0d4..f45e616abfa 100644
--- a/src/mesa/state_tracker/st_cb_rasterpos.c
+++ b/src/mesa/state_tracker/st_cb_rasterpos.c
@@ -209,8 +209,6 @@ new_draw_rastpos_stage(struct gl_context *ctx, struct draw_context *draw)
    rs->prim.end = 1;
    rs->prim.start = 0;
    rs->prim.count = 1;
-   rs->prim.num_instances = 1;
-   rs->prim.base_instance = 0;
 
    return rs;
 }
@@ -262,7 +260,7 @@ st_RasterPos(struct gl_context *ctx, const GLfloat v[4])
    _mesa_set_draw_vao(ctx, rs->VAO, VERT_BIT_POS);
 
    /* Draw the point. */
-   st_feedback_draw_vbo(ctx, &rs->prim, 1, NULL, GL_TRUE, 0, 1,
+   st_feedback_draw_vbo(ctx, &rs->prim, 1, NULL, GL_TRUE, 0, 1, 1, 0,
                         NULL, 0);
 
    /* restore draw's rasterization stage depending on rendermode */
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
index d0a27a01d26..b370e552e79 100644
--- a/src/mesa/state_tracker/st_draw.c
+++ b/src/mesa/state_tracker/st_draw.c
@@ -172,6 +172,8 @@ st_draw_vbo(struct gl_context *ctx,
 	    GLboolean index_bounds_valid,
             GLuint min_index,
             GLuint max_index,
+            GLuint num_instances,
+            GLuint base_instance,
             struct gl_transform_feedback_object *tfb_vertcount,
             unsigned stream)
 {
@@ -188,6 +190,8 @@ st_draw_vbo(struct gl_context *ctx,
    info.indirect = NULL;
    info.count_from_stream_output = NULL;
    info.restart_index = 0;
+   info.start_instance = base_instance;
+   info.instance_count = num_instances;
 
    if (ib) {
       struct gl_buffer_object *bufobj = ib->obj;
@@ -244,8 +248,6 @@ st_draw_vbo(struct gl_context *ctx,
 
       info.mode = translate_prim(ctx, prims[i].mode);
       info.start = start + prims[i].start;
-      info.start_instance = prims[i].base_instance;
-      info.instance_count = prims[i].num_instances;
       info.index_bias = prims[i].basevertex;
       info.drawid = prims[i].draw_id;
       if (!ib) {
diff --git a/src/mesa/state_tracker/st_draw.h b/src/mesa/state_tracker/st_draw.h
index 5c1dd033cd4..d04277a2325 100644
--- a/src/mesa/state_tracker/st_draw.h
+++ b/src/mesa/state_tracker/st_draw.h
@@ -55,6 +55,8 @@ st_feedback_draw_vbo(struct gl_context *ctx,
 		     GLboolean index_bounds_valid,
                      GLuint min_index,
                      GLuint max_index,
+                     GLuint num_instances,
+                     GLuint base_instance,
                      struct gl_transform_feedback_object *tfb_vertcount,
                      unsigned stream);
 
diff --git a/src/mesa/state_tracker/st_draw_feedback.c b/src/mesa/state_tracker/st_draw_feedback.c
index 16c88efa2ae..14c809c8a09 100644
--- a/src/mesa/state_tracker/st_draw_feedback.c
+++ b/src/mesa/state_tracker/st_draw_feedback.c
@@ -100,6 +100,8 @@ st_feedback_draw_vbo(struct gl_context *ctx,
 		     GLboolean index_bounds_valid,
                      GLuint min_index,
                      GLuint max_index,
+                     GLuint num_instances,
+                     GLuint base_instance,
                      struct gl_transform_feedback_object *tfb_vertcount,
                      unsigned stream)
 {
@@ -419,6 +421,9 @@ st_feedback_draw_vbo(struct gl_context *ctx,
    }
    draw_set_images(draw, PIPE_SHADER_VERTEX, images, prog->info.num_images);
 
+   info.start_instance = base_instance;
+   info.instance_count = num_instances;
+
    /* draw here */
    for (i = 0; i < nr_prims; i++) {
       info.count = prims[i].count;
@@ -428,8 +433,6 @@ st_feedback_draw_vbo(struct gl_context *ctx,
 
       info.mode = prims[i].mode;
       info.start = start + prims[i].start;
-      info.start_instance = prims[i].base_instance;
-      info.instance_count = prims[i].num_instances;
       info.index_bias = prims[i].basevertex;
       info.drawid = prims[i].draw_id;
       if (!ib) {
diff --git a/src/mesa/tnl/t_draw.c b/src/mesa/tnl/t_draw.c
index a2388b9b14f..b39c8443814 100644
--- a/src/mesa/tnl/t_draw.c
+++ b/src/mesa/tnl/t_draw.c
@@ -434,6 +434,8 @@ void _tnl_draw_prims(struct gl_context *ctx,
 			 GLboolean index_bounds_valid,
 			 GLuint min_index,
 			 GLuint max_index,
+                         GLuint num_instances,
+                         GLuint base_instance,
 			 struct gl_transform_feedback_object *tfb_vertcount,
                          unsigned stream)
 {
@@ -469,7 +471,7 @@ void _tnl_draw_prims(struct gl_context *ctx,
       /* We always translate away calls with min_index != 0. 
        */
       t_rebase_prims( ctx, arrays, prim, nr_prims, ib,
-                      min_index, max_index,
+                      min_index, max_index, num_instances, base_instance,
                       _tnl_draw_prims );
       return;
    }
@@ -488,6 +490,7 @@ void _tnl_draw_prims(struct gl_context *ctx,
        */
       _tnl_split_prims( ctx, arrays, prim, nr_prims, ib,
                         0, max_index + prim->basevertex,
+                        num_instances, base_instance,
                         _tnl_draw_prims,
                         &limits );
    }
@@ -499,6 +502,8 @@ void _tnl_draw_prims(struct gl_context *ctx,
       GLuint nr_bo = 0;
       GLuint inst;
 
+      assert(num_instances > 0);
+
       for (i = 0; i < nr_prims;) {
 	 GLuint this_nr_prims;
 
@@ -512,12 +517,10 @@ void _tnl_draw_prims(struct gl_context *ctx,
 	       break;
 	 }
 
-         assert(prim[i].num_instances > 0);
-
 	 /* Binding inputs may imply mapping some vertex buffer objects.
 	  * They will need to be unmapped below.
 	  */
-         for (inst = 0; inst < prim[i].num_instances; inst++) {
+         for (inst = 0; inst < num_instances; inst++) {
 
             bind_prims(ctx, &prim[i], this_nr_prims);
             bind_inputs(ctx, arrays, max_index + prim[i].basevertex + 1,
@@ -637,6 +640,7 @@ _tnl_draw(struct gl_context *ctx,
           const struct _mesa_prim *prim, GLuint nr_prims,
           const struct _mesa_index_buffer *ib,
           GLboolean index_bounds_valid, GLuint min_index, GLuint max_index,
+          GLuint num_instances, GLuint base_instance,
           struct gl_transform_feedback_object *tfb_vertcount,
           unsigned stream)
 {
@@ -646,7 +650,7 @@ _tnl_draw(struct gl_context *ctx,
 
    _tnl_draw_prims(ctx, arrays, prim, nr_prims, ib,
                    index_bounds_valid, min_index, max_index,
-                   tfb_vertcount, stream);
+                   num_instances, base_instance, tfb_vertcount, stream);
 }
 
 
diff --git a/src/mesa/tnl/t_rebase.c b/src/mesa/tnl/t_rebase.c
index 24abcef98a2..06d67620f63 100644
--- a/src/mesa/tnl/t_rebase.c
+++ b/src/mesa/tnl/t_rebase.c
@@ -110,6 +110,8 @@ void t_rebase_prims( struct gl_context *ctx,
                      const struct _mesa_index_buffer *ib,
                      GLuint min_index,
                      GLuint max_index,
+                     GLuint num_instances,
+                     GLuint base_instance,
                      tnl_draw_func draw )
 {
    struct gl_array_attributes tmp_attribs[VERT_ATTRIB_MAX];
@@ -242,6 +244,7 @@ void t_rebase_prims( struct gl_context *ctx,
 	 GL_TRUE,
 	 0, 
 	 max_index - min_index,
+         num_instances, base_instance,
 	 NULL, 0);
 
    free(tmp_indices);
diff --git a/src/mesa/tnl/t_rebase.h b/src/mesa/tnl/t_rebase.h
index d0aa9e18905..b596d93995b 100644
--- a/src/mesa/tnl/t_rebase.h
+++ b/src/mesa/tnl/t_rebase.h
@@ -34,6 +34,8 @@ void t_rebase_prims( struct gl_context *ctx,
                      const struct _mesa_index_buffer *ib,
                      GLuint min_index,
                      GLuint max_index,
+                     GLuint num_instances,
+                     GLuint base_instance,
                      tnl_draw_func draw );
 
 #endif
diff --git a/src/mesa/tnl/t_split.c b/src/mesa/tnl/t_split.c
index d7aac10e4c8..9c59380c831 100644
--- a/src/mesa/tnl/t_split.c
+++ b/src/mesa/tnl/t_split.c
@@ -106,6 +106,8 @@ _tnl_split_prims(struct gl_context *ctx,
                  const struct _mesa_index_buffer *ib,
                  GLuint min_index,
                  GLuint max_index,
+                 GLuint num_instances,
+                 GLuint base_instance,
                  tnl_draw_func draw,
                  const struct split_limits *limits)
 {
@@ -135,7 +137,8 @@ _tnl_split_prims(struct gl_context *ctx,
           * individual primitives.
           */
          _tnl_split_inplace(ctx, arrays, prim, nr_prims, ib,
-                            min_index, max_index, draw, limits);
+                            min_index, max_index, num_instances,
+                            base_instance, draw, limits);
       }
       else {
          /* Why were we called? */
@@ -149,7 +152,8 @@ _tnl_split_prims(struct gl_context *ctx,
           * otherwise try to split the individual primitives.
           */
          _tnl_split_inplace(ctx, arrays, prim, nr_prims, ib,
-                            min_index, max_index, draw, limits);
+                            min_index, max_index, num_instances,
+                            base_instance, draw, limits);
       }
       else {
          /* Why were we called? */
diff --git a/src/mesa/tnl/t_split.h b/src/mesa/tnl/t_split.h
index 49017e5dfb8..7634e553a87 100644
--- a/src/mesa/tnl/t_split.h
+++ b/src/mesa/tnl/t_split.h
@@ -57,6 +57,8 @@ _tnl_split_inplace(struct gl_context *ctx,
                    const struct _mesa_index_buffer *ib,
                    GLuint min_index,
                    GLuint max_index,
+                   GLuint num_instances,
+                   GLuint base_instance,
                    tnl_draw_func draw,
                    const struct split_limits *limits);
 
diff --git a/src/mesa/tnl/t_split_copy.c b/src/mesa/tnl/t_split_copy.c
index 511e46c1e2f..45b20fbd49c 100644
--- a/src/mesa/tnl/t_split_copy.c
+++ b/src/mesa/tnl/t_split_copy.c
@@ -211,6 +211,8 @@ flush(struct copy_context *copy)
               GL_TRUE,
               0,
               copy->dstbuf_nr - 1,
+              1,
+              0,
               NULL, 0);
 
    /* Reset all pointers:
@@ -237,7 +239,6 @@ begin(struct copy_context *copy, GLenum mode, GLboolean begin_flag)
 
    prim->mode = mode;
    prim->begin = begin_flag;
-   prim->num_instances = 1;
 }
 
 
diff --git a/src/mesa/tnl/t_split_inplace.c b/src/mesa/tnl/t_split_inplace.c
index 048f87309ee..d9ea52dfaf1 100644
--- a/src/mesa/tnl/t_split_inplace.c
+++ b/src/mesa/tnl/t_split_inplace.c
@@ -49,6 +49,8 @@ struct split_context {
    const struct _mesa_index_buffer *ib;
    GLuint min_index;
    GLuint max_index;
+   GLuint num_instances;
+   GLuint base_instance;
    tnl_draw_func draw;
 
    const struct split_limits *limits;
@@ -93,6 +95,8 @@ flush_vertex( struct split_context *split)
                !split->ib,
                split->min_index,
                split->max_index,
+               split->num_instances,
+               split->base_instance,
                NULL, 0);
 
    split->dstprim_nr = 0;
@@ -190,8 +194,6 @@ split_prims(struct split_context *split)
             outprim->end = (nr == remaining && prim->end);
             outprim->start = prim->start + j;
             outprim->count = nr;
-            outprim->num_instances = prim->num_instances;
-            outprim->base_instance = prim->base_instance;
 
             update_index_bounds(split, outprim);
 
@@ -232,8 +234,6 @@ split_prims(struct split_context *split)
          tmpprim = *prim;
          tmpprim.start = 0;
          tmpprim.count = count;
-         tmpprim.num_instances = 1;
-         tmpprim.base_instance = 0;
 
          flush_vertex(split);
 
@@ -270,6 +270,8 @@ _tnl_split_inplace(struct gl_context *ctx,
                    const struct _mesa_index_buffer *ib,
                    GLuint min_index,
                    GLuint max_index,
+                   GLuint num_instances,
+                   GLuint base_instance,
                    tnl_draw_func draw,
                    const struct split_limits *limits)
 {
@@ -286,6 +288,8 @@ _tnl_split_inplace(struct gl_context *ctx,
    /* Empty interval, makes calculations simpler. */
    split.min_index = ~0;
    split.max_index = 0;
+   split.num_instances = num_instances;
+   split.base_instance = base_instance;
 
    split.draw = draw;
    split.limits = limits;
diff --git a/src/mesa/tnl/tnl.h b/src/mesa/tnl/tnl.h
index c3927536b8b..4bf8ee20f68 100644
--- a/src/mesa/tnl/tnl.h
+++ b/src/mesa/tnl/tnl.h
@@ -107,6 +107,8 @@ _tnl_draw_prims(struct gl_context *ctx,
 		     GLboolean index_bounds_valid,
 		     GLuint min_index,
 		     GLuint max_index,
+                     GLuint num_instances,
+                     GLuint base_instance,
 		     struct gl_transform_feedback_object *tfb_vertcount,
                      unsigned stream);
 
@@ -115,6 +117,7 @@ _tnl_draw(struct gl_context *ctx,
           const struct _mesa_prim *prim, GLuint nr_prims,
           const struct _mesa_index_buffer *ib,
           GLboolean index_bounds_valid, GLuint min_index, GLuint max_index,
+          GLuint num_instances, GLuint base_instance,
           struct gl_transform_feedback_object *tfb_vertcount, unsigned stream);
 
 extern void
@@ -172,6 +175,8 @@ typedef void (*tnl_draw_func)(struct gl_context *ctx,
                               GLboolean index_bounds_valid,
                               GLuint min_index,
                               GLuint max_index,
+                              GLuint num_instances,
+                              GLuint base_instance,
                               struct gl_transform_feedback_object *tfb_vertcount,
                               unsigned tfb_stream);
 
@@ -198,6 +203,8 @@ _tnl_split_prims(struct gl_context *ctx,
                  const struct _mesa_index_buffer *ib,
                  GLuint min_index,
                  GLuint max_index,
+                 GLuint num_instances,
+                 GLuint base_instance,
                  tnl_draw_func draw,
                  const struct split_limits *limits);
 
diff --git a/src/mesa/vbo/vbo.h b/src/mesa/vbo/vbo.h
index a03a4cec70b..688e8fd83eb 100644
--- a/src/mesa/vbo/vbo.h
+++ b/src/mesa/vbo/vbo.h
@@ -97,6 +97,7 @@ vbo_sw_primitive_restart(struct gl_context *ctx,
                          const struct _mesa_prim *prim,
                          GLuint nr_prims,
                          const struct _mesa_index_buffer *ib,
+                         GLuint num_instances, GLuint base_instance,
                          struct gl_buffer_object *indirect,
                          GLsizeiptr indirect_offset);
 
diff --git a/src/mesa/vbo/vbo_exec.c b/src/mesa/vbo/vbo_exec.c
index f1b1f8bff95..66ad8888bcd 100644
--- a/src/mesa/vbo/vbo_exec.c
+++ b/src/mesa/vbo/vbo_exec.c
@@ -191,9 +191,7 @@ vbo_can_merge_prims(const struct _mesa_prim *p0, const struct _mesa_prim *p1)
    if (p0->start + p0->count != p1->start)
       return false;
 
-   assert(p0->basevertex == p1->basevertex &&
-          p0->num_instances == p1->num_instances &&
-          p0->base_instance == p1->base_instance);
+   assert(p0->basevertex == p1->basevertex);
 
    /* can always merge subsequent GL_POINTS primitives */
    if (p0->mode == GL_POINTS)
diff --git a/src/mesa/vbo/vbo_exec_api.c b/src/mesa/vbo/vbo_exec_api.c
index d724f4efe11..2f049ccd450 100644
--- a/src/mesa/vbo/vbo_exec_api.c
+++ b/src/mesa/vbo/vbo_exec_api.c
@@ -812,8 +812,6 @@ vbo_exec_Begin(GLenum mode)
    exec->vtx.prim[i].end = 0;
    exec->vtx.prim[i].start = exec->vtx.vert_count;
    exec->vtx.prim[i].count = 0;
-   exec->vtx.prim[i].num_instances = 1;
-   exec->vtx.prim[i].base_instance = 0;
 
    ctx->Driver.CurrentExecPrimitive = mode;
 
diff --git a/src/mesa/vbo/vbo_exec_draw.c b/src/mesa/vbo/vbo_exec_draw.c
index 10e330eb816..87e7db18d67 100644
--- a/src/mesa/vbo/vbo_exec_draw.c
+++ b/src/mesa/vbo/vbo_exec_draw.c
@@ -413,7 +413,7 @@ vbo_exec_vtx_flush(struct vbo_exec_context *exec)
                    exec->vtx.vert_count);
 
          ctx->Driver.Draw(ctx, exec->vtx.prim, exec->vtx.prim_count,
-                          NULL, GL_TRUE, 0, exec->vtx.vert_count - 1,
+                          NULL, GL_TRUE, 0, exec->vtx.vert_count - 1, 1, 0,
                           NULL, 0);
 
          /* Get new storage -- unless asked not to. */
diff --git a/src/mesa/vbo/vbo_primitive_restart.c b/src/mesa/vbo/vbo_primitive_restart.c
index b4b3cdf8d4c..2b6bcd46e3e 100644
--- a/src/mesa/vbo/vbo_primitive_restart.c
+++ b/src/mesa/vbo/vbo_primitive_restart.c
@@ -164,6 +164,7 @@ vbo_sw_primitive_restart(struct gl_context *ctx,
                          const struct _mesa_prim *prims,
                          GLuint nr_prims,
                          const struct _mesa_index_buffer *ib,
+                         GLuint num_instances, GLuint base_instance,
                          struct gl_buffer_object *indirect,
                          GLsizeiptr indirect_offset)
 {
@@ -200,10 +201,11 @@ vbo_sw_primitive_restart(struct gl_context *ctx,
                                      indirect_offset);
 
       new_prim.count = indirect_params[0];
-      new_prim.num_instances = indirect_params[1];
       new_prim.start = indirect_params[2];
       new_prim.basevertex = indirect_params[3];
-      new_prim.base_instance = indirect_params[4];
+
+      num_instances = indirect_params[1];
+      base_instance = indirect_params[4];
 
       new_ib = *ib;
       new_ib.count = new_prim.count;
@@ -249,10 +251,12 @@ vbo_sw_primitive_restart(struct gl_context *ctx,
                 (temp_prim.count == sub_prim->count)) {
                ctx->Driver.Draw(ctx, &temp_prim, 1, ib, GL_TRUE,
                                 sub_prim->min_index, sub_prim->max_index,
+                                num_instances, base_instance,
                                 NULL, 0);
             } else {
                ctx->Driver.Draw(ctx, &temp_prim, 1, ib,
                                 GL_FALSE, -1, -1,
+                                num_instances, base_instance,
                                 NULL, 0);
             }
          }
diff --git a/src/mesa/vbo/vbo_save_api.c b/src/mesa/vbo/vbo_save_api.c
index 8e7ab86bae1..1ef6043ad0c 100644
--- a/src/mesa/vbo/vbo_save_api.c
+++ b/src/mesa/vbo/vbo_save_api.c
@@ -739,8 +739,6 @@ wrap_buffers(struct gl_context *ctx)
    save->prims[0].end = 0;
    save->prims[0].start = 0;
    save->prims[0].count = 0;
-   save->prims[0].num_instances = 1;
-   save->prims[0].base_instance = 0;
    save->prim_count = 1;
 }
 
@@ -1218,8 +1216,6 @@ vbo_save_NotifyBegin(struct gl_context *ctx, GLenum mode,
    save->prims[i].end = 0;
    save->prims[i].start = save->vert_count;
    save->prims[i].count = 0;
-   save->prims[i].num_instances = 1;
-   save->prims[i].base_instance = 0;
 
    save->no_current_update = no_current_update;
 
diff --git a/src/mesa/vbo/vbo_save_draw.c b/src/mesa/vbo/vbo_save_draw.c
index f6298ee2c9b..e193a5a32af 100644
--- a/src/mesa/vbo/vbo_save_draw.c
+++ b/src/mesa/vbo/vbo_save_draw.c
@@ -213,7 +213,7 @@ vbo_save_playback_vertex_list(struct gl_context *ctx, void *data)
          GLuint min_index = _vbo_save_get_min_index(node);
          GLuint max_index = _vbo_save_get_max_index(node);
          ctx->Driver.Draw(ctx, node->prims, node->prim_count, NULL, GL_TRUE,
-                          min_index, max_index, NULL, 0);
+                          min_index, max_index, 1, 0, NULL, 0);
       }
    }