mesa: add Driver.DrawTransformFeedback

to remove some overhead from Driver.Draw.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7441>

16 files changed, 111 insertions, 74 deletions

diff --git a/src/mesa/drivers/common/driverfuncs.c b/src/mesa/drivers/common/driverfuncs.c
index 13dbd113a29..21771d2cfc6 100644
--- a/src/mesa/drivers/common/driverfuncs.c
+++ b/src/mesa/drivers/common/driverfuncs.c
@@ -122,6 +122,7 @@ _mesa_init_driver_functions(struct dd_function_table *driver)
    /* Draw functions */
    driver->Draw = NULL;
    driver->DrawIndirect = NULL;
+   driver->DrawTransformFeedback = NULL;
 
    /* simple state commands */
    driver->AlphaFunc = NULL;
diff --git a/src/mesa/drivers/dri/i965/brw_draw.c b/src/mesa/drivers/dri/i965/brw_draw.c
index cf4778e8abc..25162ca7825 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.c
+++ b/src/mesa/drivers/dri/i965/brw_draw.c
@@ -1140,15 +1140,11 @@ brw_draw_prims(struct gl_context *ctx,
                GLuint min_index,
                GLuint max_index,
                GLuint num_instances,
-               GLuint base_instance,
-               struct gl_transform_feedback_object *gl_xfb_obj,
-               unsigned stream)
+               GLuint base_instance)
 {
    unsigned i;
    struct brw_context *brw = brw_context(ctx);
    int predicate_state = brw->predicate.state;
-   struct brw_transform_feedback_object *xfb_obj =
-      (struct brw_transform_feedback_object *) gl_xfb_obj;
 
    if (!brw_check_conditional_render(brw))
       return;
@@ -1169,7 +1165,7 @@ brw_draw_prims(struct gl_context *ctx,
       _swsetup_Wakeup(ctx);
       _tnl_wakeup(ctx);
       _tnl_draw(ctx, prims, nr_prims, ib, index_bounds_valid, min_index,
-                max_index, num_instances, base_instance, NULL, 0);
+                max_index, num_instances, base_instance);
       return;
    }
 
@@ -1222,7 +1218,7 @@ brw_draw_prims(struct gl_context *ctx,
       }
 
       brw_draw_single_prim(ctx, &prims[i], i, ib != NULL, num_instances,
-                           base_instance, xfb_obj, stream,
+                           base_instance, NULL, 0,
                            brw->draw.draw_indirect_offset +
                            brw->draw.draw_indirect_stride * i);
    }
@@ -1231,6 +1227,45 @@ brw_draw_prims(struct gl_context *ctx,
    brw->predicate.state = predicate_state;
 }
 
+static void
+brw_draw_transform_feedback(struct gl_context *ctx, GLenum mode,
+                            unsigned num_instances, unsigned stream,
+                            struct gl_transform_feedback_object *gl_xfb_obj)
+{
+   struct brw_context *brw = brw_context(ctx);
+   struct brw_transform_feedback_object *xfb_obj =
+      (struct brw_transform_feedback_object *) gl_xfb_obj;
+
+   if (!brw_check_conditional_render(brw))
+      return;
+
+   /* Do GL_SELECT and GL_FEEDBACK rendering using swrast, even though it
+    * won't support all the extensions we support.
+    */
+   if (ctx->RenderMode != GL_RENDER) {
+      perf_debug("%s render mode not supported in hardware\n",
+                 _mesa_enum_to_string(ctx->RenderMode));
+      /* swrast doesn't support DrawTransformFeedback. Nothing to do. */
+      return;
+   }
+
+   brw_prepare_drawing(ctx, NULL, false, 0, ~0);
+
+   struct _mesa_prim prim;
+   memset(&prim, 0, sizeof(prim));
+   prim.begin = 1;
+   prim.end = 1;
+   prim.mode = mode;
+
+   /* Try drawing with the hardware, but don't do anything else if we can't
+    * manage it.  swrast doesn't support our featureset, so we can't fall back
+    * to it.
+    */
+   brw_draw_single_prim(ctx, &prim, 0, false, num_instances, 0, xfb_obj,
+                        stream, 0);
+   brw_finish_drawing(ctx);
+}
+
 void
 brw_draw_indirect_prims(struct gl_context *ctx,
                         GLuint mode,
@@ -1274,7 +1309,7 @@ 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, 0, 0, NULL, 0);
+   brw_draw_prims(ctx, prim, draw_count, ib, false, 0, ~0, 0, 0);
 
    brw->draw.draw_indirect_data = NULL;
    free(prim);
@@ -1286,6 +1321,7 @@ brw_init_draw_functions(struct dd_function_table *functions)
    /* Register our drawing function:
     */
    functions->Draw = brw_draw_prims;
+   functions->DrawTransformFeedback = brw_draw_transform_feedback;
    functions->DrawIndirect = brw_draw_indirect_prims;
 }
 
diff --git a/src/mesa/drivers/dri/i965/brw_draw.h b/src/mesa/drivers/dri/i965/brw_draw.h
index c2e7d26f3ca..9331d5468d2 100644
--- a/src/mesa/drivers/dri/i965/brw_draw.h
+++ b/src/mesa/drivers/dri/i965/brw_draw.h
@@ -52,9 +52,7 @@ void brw_draw_prims(struct gl_context *ctx,
 		     GLuint min_index,
 		     GLuint max_index,
                      GLuint num_instances,
-                     GLuint base_instance,
-		     struct gl_transform_feedback_object *unused_tfb_object,
-                     unsigned stream);
+                     GLuint base_instance);
 
 void brw_init_draw_functions(struct dd_function_table *functions);
 void brw_draw_init( struct brw_context *brw );
diff --git a/src/mesa/drivers/dri/i965/brw_primitive_restart.c b/src/mesa/drivers/dri/i965/brw_primitive_restart.c
index e73f32720e1..a571123bcef 100644
--- a/src/mesa/drivers/dri/i965/brw_primitive_restart.c
+++ b/src/mesa/drivers/dri/i965/brw_primitive_restart.c
@@ -163,7 +163,7 @@ brw_handle_primitive_restart(struct gl_context *ctx,
        */
       brw->prim_restart.enable_cut_index = true;
       brw_draw_prims(ctx, prims, nr_prims, ib, GL_FALSE, -1, -1,
-                     num_instances, base_instance, NULL, 0);
+                     num_instances, base_instance);
       brw->prim_restart.enable_cut_index = false;
    } else {
       /* Not all the primitive draw modes are supported by the cut index,
diff --git a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
index e6274d45cf4..9248d097119 100644
--- a/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
+++ b/src/mesa/drivers/dri/nouveau/nouveau_vbo_t.c
@@ -546,9 +546,7 @@ 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,
-	      UNUSED struct gl_transform_feedback_object *tfb_vertcount,
-	      UNUSED unsigned stream)
+              GLuint num_instances, GLuint base_instance)
 {
 	/* Borrow and update the inputs list from the tnl context */
 	const struct tnl_vertex_array* arrays = _tnl_bind_inputs(ctx);
diff --git a/src/mesa/main/dd.h b/src/mesa/main/dd.h
index 506a18240cc..ae30f2b1423 100644
--- a/src/mesa/main/dd.h
+++ b/src/mesa/main/dd.h
@@ -544,23 +544,13 @@ struct dd_function_table {
     * \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
-    *                    the vertex stream buffer from which to get the vertex
-    *                    count.
-    * \param indirect  If any prims are indirect, this specifies the buffer
-    *                  to find the "DrawArrays/ElementsIndirectCommand" data.
-    *                  This may be deprecated in the future
     */
    void (*Draw)(struct gl_context *ctx,
                 const struct _mesa_prim *prims, 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 tfb_stream);
+                GLuint num_instances, GLuint base_instance);
 
 
    /**
@@ -587,6 +577,21 @@ struct dd_function_table {
                         struct gl_buffer_object *indirect_draw_count_buffer,
                         GLsizeiptr indirect_draw_count_offset,
                         const struct _mesa_index_buffer *ib);
+
+   /**
+    * Driver implementation of glDrawTransformFeedback.
+    *
+    * \param mode    Primitive type
+    * \param num_instances  instance count from ARB_draw_instanced
+    * \param stream  If called via DrawTransformFeedbackStream, specifies
+    *                the vertex stream buffer from which to get the vertex
+    *                count.
+    * \param tfb_vertcount  if non-null, indicates which transform feedback
+    *                       object has the vertex count.
+    */
+   void (*DrawTransformFeedback)(struct gl_context *ctx, GLenum mode,
+                                 unsigned num_instances, unsigned stream,
+                                 struct gl_transform_feedback_object *tfb_vertcount);
    /*@}*/
 
 
diff --git a/src/mesa/main/draw.c b/src/mesa/main/draw.c
index 74a619e3022..0d7258eaa3b 100644
--- a/src/mesa/main/draw.c
+++ b/src/mesa/main/draw.c
@@ -366,7 +366,7 @@ _mesa_draw_arrays(struct gl_context *ctx, GLenum mode, GLint start,
 
    ctx->Driver.Draw(ctx, &prim, 1, NULL,
                     GL_TRUE, start, start + count - 1,
-                    numInstances, baseInstance, NULL, 0);
+                    numInstances, baseInstance);
 
    if (MESA_DEBUG_FLAGS & DEBUG_ALWAYS_FLUSH) {
       _mesa_flush(ctx);
@@ -729,8 +729,7 @@ _mesa_exec_MultiDrawArrays(GLenum mode, const GLint *first,
       prim[i].basevertex = 0;
    }
 
-   ctx->Driver.Draw(ctx, prim, primcount, NULL, GL_FALSE, 0, 0, 1, 0,
-                    NULL, 0);
+   ctx->Driver.Draw(ctx, prim, primcount, NULL, GL_FALSE, 0, 0, 1, 0);
 
    if (MESA_DEBUG_FLAGS & DEBUG_ALWAYS_FLUSH)
       _mesa_flush(ctx);
@@ -890,7 +889,7 @@ _mesa_validated_drawrangeelements(struct gl_context *ctx, GLenum mode,
 
    ctx->Driver.Draw(ctx, &prim, 1, &ib,
                     index_bounds_valid, start, end,
-                    numInstances, baseInstance, NULL, 0);
+                    numInstances, baseInstance);
 
    if (MESA_DEBUG_FLAGS & DEBUG_ALWAYS_FLUSH) {
       _mesa_flush(ctx);
@@ -1312,7 +1311,7 @@ _mesa_validated_multidrawelements(struct gl_context *ctx, GLenum mode,
       }
 
       ctx->Driver.Draw(ctx, prim, primcount, &ib,
-                       false, 0, ~0, 1, 0, NULL, 0);
+                       false, 0, ~0, 1, 0);
       FREE_PRIMS(prim, primcount);
    }
    else {
@@ -1337,7 +1336,7 @@ _mesa_validated_multidrawelements(struct gl_context *ctx, GLenum mode,
          else
             prim.basevertex = 0;
 
-         ctx->Driver.Draw(ctx, &prim, 1, &ib, false, 0, ~0, 1, 0, NULL, 0);
+         ctx->Driver.Draw(ctx, &prim, 1, &ib, false, 0, ~0, 1, 0);
       }
    }
 
@@ -1412,8 +1411,6 @@ _mesa_draw_transform_feedback(struct gl_context *ctx, GLenum mode,
                               struct gl_transform_feedback_object *obj,
                               GLuint stream, GLuint numInstances)
 {
-   struct _mesa_prim prim;
-
    FLUSH_FOR_DRAW(ctx);
 
    _mesa_set_draw_vao(ctx, ctx->Array.VAO, enabled_filter(ctx));
@@ -1440,18 +1437,11 @@ _mesa_draw_transform_feedback(struct gl_context *ctx, GLenum mode,
    if (skip_validated_draw(ctx))
       return;
 
-   /* init most fields to zero */
-   memset(&prim, 0, sizeof(prim));
-   prim.begin = 1;
-   prim.end = 1;
-   prim.mode = mode;
-
    /* 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, numInstances, 0,
-                    obj, stream);
+   ctx->Driver.DrawTransformFeedback(ctx, mode, numInstances, stream, obj);
 
    if (MESA_DEBUG_FLAGS & DEBUG_ALWAYS_FLUSH) {
       _mesa_flush(ctx);
diff --git a/src/mesa/state_tracker/st_cb_rasterpos.c b/src/mesa/state_tracker/st_cb_rasterpos.c
index c3684176071..691d5995109 100644
--- a/src/mesa/state_tracker/st_cb_rasterpos.c
+++ b/src/mesa/state_tracker/st_cb_rasterpos.c
@@ -260,8 +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, 1, 0,
-                        NULL, 0);
+   st_feedback_draw_vbo(ctx, &rs->prim, 1, NULL, GL_TRUE, 0, 1, 1, 0);
 
    /* restore draw's rasterization stage depending on rendermode */
    if (ctx->RenderMode == GL_FEEDBACK) {
diff --git a/src/mesa/state_tracker/st_draw.c b/src/mesa/state_tracker/st_draw.c
index 996d985510c..1745a8d1948 100644
--- a/src/mesa/state_tracker/st_draw.c
+++ b/src/mesa/state_tracker/st_draw.c
@@ -164,9 +164,7 @@ st_draw_vbo(struct gl_context *ctx,
             GLuint min_index,
             GLuint max_index,
             GLuint num_instances,
-            GLuint base_instance,
-            struct gl_transform_feedback_object *tfb_vertcount,
-            unsigned stream)
+            GLuint base_instance)
 {
    struct st_context *st = st_context(ctx);
    struct pipe_draw_info info;
@@ -221,13 +219,6 @@ st_draw_vbo(struct gl_context *ctx,
    else {
       info.index_size = 0;
       info.has_user_indices = false;
-
-      /* Transform feedback drawing is always non-indexed. */
-      /* Set info.count_from_stream_output. */
-      if (tfb_vertcount) {
-         if (!st_transform_feedback_draw_init(tfb_vertcount, stream, &info))
-            return;
-      }
    }
 
    /* do actual drawing */
@@ -235,7 +226,7 @@ st_draw_vbo(struct gl_context *ctx,
       info.count = prims[i].count;
 
       /* Skip no-op draw calls. */
-      if (!info.count && !tfb_vertcount)
+      if (!info.count)
          continue;
 
       info.mode = translate_prim(ctx, prims[i].mode);
@@ -332,12 +323,42 @@ st_indirect_draw_vbo(struct gl_context *ctx,
    }
 }
 
+static void
+st_draw_transform_feedback(struct gl_context *ctx, GLenum mode,
+                           unsigned num_instances, unsigned stream,
+                           struct gl_transform_feedback_object *tfb_vertcount)
+{
+   struct st_context *st = st_context(ctx);
+   struct pipe_draw_info info;
+
+   prepare_draw(st, ctx);
+
+   util_draw_init_info(&info);
+   info.start = 0; /* index offset / index size */
+   info.max_index = ~0u; /* so that u_vbuf can tell that it's unknown */
+   info.mode = translate_prim(ctx, mode);
+   info.vertices_per_patch = ctx->TessCtrlProgram.patch_vertices;
+   info.instance_count = num_instances;
+
+   if (ST_DEBUG & DEBUG_DRAW) {
+      debug_printf("st/draw transform feedback: mode %s\n",
+                   u_prim_name(info.mode));
+   }
+
+   /* Transform feedback drawing is always non-indexed. */
+   /* Set info.count_from_stream_output. */
+   if (!st_transform_feedback_draw_init(tfb_vertcount, stream, &info))
+      return;
+
+   cso_draw_vbo(st->cso_context, &info);
+}
 
 void
 st_init_draw_functions(struct dd_function_table *functions)
 {
    functions->Draw = st_draw_vbo;
    functions->DrawIndirect = st_indirect_draw_vbo;
+   functions->DrawTransformFeedback = st_draw_transform_feedback;
 }
 
 
diff --git a/src/mesa/state_tracker/st_draw.h b/src/mesa/state_tracker/st_draw.h
index d04277a2325..c87727d28f8 100644
--- a/src/mesa/state_tracker/st_draw.h
+++ b/src/mesa/state_tracker/st_draw.h
@@ -56,9 +56,7 @@ st_feedback_draw_vbo(struct gl_context *ctx,
                      GLuint min_index,
                      GLuint max_index,
                      GLuint num_instances,
-                     GLuint base_instance,
-                     struct gl_transform_feedback_object *tfb_vertcount,
-                     unsigned stream);
+                     GLuint base_instance);
 
 /**
  * When drawing with VBOs, the addresses specified with
diff --git a/src/mesa/state_tracker/st_draw_feedback.c b/src/mesa/state_tracker/st_draw_feedback.c
index a2abc9b4b33..9129162554c 100644
--- a/src/mesa/state_tracker/st_draw_feedback.c
+++ b/src/mesa/state_tracker/st_draw_feedback.c
@@ -101,9 +101,7 @@ st_feedback_draw_vbo(struct gl_context *ctx,
                      GLuint min_index,
                      GLuint max_index,
                      GLuint num_instances,
-                     GLuint base_instance,
-                     struct gl_transform_feedback_object *tfb_vertcount,
-                     unsigned stream)
+                     GLuint base_instance)
 {
    struct st_context *st = st_context(ctx);
    struct pipe_context *pipe = st->pipe;
@@ -126,7 +124,6 @@ st_feedback_draw_vbo(struct gl_context *ctx,
    info.primitive_restart = false;
    info.vertices_per_patch = ctx->TessCtrlProgram.patch_vertices;
    info.indirect = NULL;
-   info.count_from_stream_output = NULL;
    info.restart_index = 0;
 
    st_flush_bitmap_cache(st);
diff --git a/src/mesa/tnl/t_draw.c b/src/mesa/tnl/t_draw.c
index e8b45af081c..cf23c01ba6d 100644
--- a/src/mesa/tnl/t_draw.c
+++ b/src/mesa/tnl/t_draw.c
@@ -631,9 +631,7 @@ _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,
-          UNUSED struct gl_transform_feedback_object *tfb_vertcount,
-          UNUSED unsigned stream)
+          GLuint num_instances, GLuint base_instance)
 {
    /* Update TNLcontext::draw_arrays and return that pointer.
     */
diff --git a/src/mesa/tnl/tnl.h b/src/mesa/tnl/tnl.h
index 58503016a9b..5c87ab6a4b7 100644
--- a/src/mesa/tnl/tnl.h
+++ b/src/mesa/tnl/tnl.h
@@ -115,8 +115,7 @@ _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);
+          GLuint num_instances, GLuint base_instance);
 
 extern void
 _tnl_RasterPos(struct gl_context *ctx, const GLfloat vObj[4]);
diff --git a/src/mesa/vbo/vbo_exec_draw.c b/src/mesa/vbo/vbo_exec_draw.c
index b31c1f90156..ae14478c260 100644
--- a/src/mesa/vbo/vbo_exec_draw.c
+++ b/src/mesa/vbo/vbo_exec_draw.c
@@ -325,8 +325,7 @@ 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, 1, 0,
-                          NULL, 0);
+                          NULL, GL_TRUE, 0, exec->vtx.vert_count - 1, 1, 0);
 
          /* Get new storage -- unless asked not to. */
          if (!persistent_mapping)
diff --git a/src/mesa/vbo/vbo_primitive_restart.c b/src/mesa/vbo/vbo_primitive_restart.c
index ff5df0c2e79..8dce7d67366 100644
--- a/src/mesa/vbo/vbo_primitive_restart.c
+++ b/src/mesa/vbo/vbo_primitive_restart.c
@@ -254,13 +254,11 @@ 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);
+                                num_instances, base_instance);
             } else {
                ctx->Driver.Draw(ctx, &temp_prim, 1, ib,
                                 GL_FALSE, -1, -1,
-                                num_instances, base_instance,
-                                NULL, 0);
+                                num_instances, base_instance);
             }
          }
          if (sub_end_index >= end_index) {
diff --git a/src/mesa/vbo/vbo_save_draw.c b/src/mesa/vbo/vbo_save_draw.c
index 5d0c6c54ed3..8df082fb0ab 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, 1, 0, NULL, 0);
+                          min_index, max_index, 1, 0);
       }
    }