drivers/llvmpipe/lp_setup.c

/**************************************************************************
 *
 * Copyright 2007 VMware, Inc.
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sub license, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial portions
 * of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 **************************************************************************/

/**
 * Tiling engine.
 *
 * Builds per-tile display lists and executes them on calls to
 * lp_setup_flush().
 */

#include <limits.h>

#include "pipe/p_defines.h"
#include "util/u_framebuffer.h"
#include "util/u_inlines.h"
#include "util/u_memory.h"
#include "util/u_pack_color.h"
#include "draw/draw_pipe.h"
#include "os/os_time.h"
#include "lp_context.h"
#include "lp_memory.h"
#include "lp_scene.h"
#include "lp_texture.h"
#include "lp_debug.h"
#include "lp_fence.h"
#include "lp_query.h"
#include "lp_rast.h"
#include "lp_setup_context.h"
#include "lp_screen.h"
#include "lp_state.h"
#include "state_tracker/sw_winsys.h"

#include "draw/draw_context.h"
#include "draw/draw_vbuf.h"


static boolean set_scene_state( struct lp_setup_context *, enum setup_state,
                             const char *reason);
static boolean try_update_scene_state( struct lp_setup_context *setup );


static void
lp_setup_get_empty_scene(struct lp_setup_context *setup)
{
   assert(setup->scene == NULL);

   setup->scene_idx++;
   setup->scene_idx %= Elements(setup->scenes);

   setup->scene = setup->scenes[setup->scene_idx];

   if (setup->scene->fence) {
      if (LP_DEBUG & DEBUG_SETUP)
         debug_printf("%s: wait for scene %d\n",
                      __FUNCTION__, setup->scene->fence->id);

      lp_fence_wait(setup->scene->fence);
   }

   lp_scene_begin_binning(setup->scene, &setup->fb, setup->rasterizer_discard);

}


static void
first_triangle( struct lp_setup_context *setup,
                const float (*v0)[4],
                const float (*v1)[4],
                const float (*v2)[4])
{
   assert(setup->state == SETUP_ACTIVE);
   lp_setup_choose_triangle( setup );
   setup->triangle( setup, v0, v1, v2 );
}

static void
first_line( struct lp_setup_context *setup,
	    const float (*v0)[4],
	    const float (*v1)[4])
{
   assert(setup->state == SETUP_ACTIVE);
   lp_setup_choose_line( setup );
   setup->line( setup, v0, v1 );
}

static void
first_point( struct lp_setup_context *setup,
	     const float (*v0)[4])
{
   assert(setup->state == SETUP_ACTIVE);
   lp_setup_choose_point( setup );
   setup->point( setup, v0 );
}

void lp_setup_reset( struct lp_setup_context *setup )
{
   unsigned i;

   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   /* Reset derived state */
   for (i = 0; i < Elements(setup->constants); ++i) {
      setup->constants[i].stored_size = 0;
      setup->constants[i].stored_data = NULL;
   }
   setup->fs.stored = NULL;
   setup->dirty = ~0;

   /* no current bin */
   setup->scene = NULL;

   /* Reset some state:
    */
   memset(&setup->clear, 0, sizeof setup->clear);

   /* Have an explicit "start-binning" call and get rid of this
    * pointer twiddling?
    */
   setup->line = first_line;
   setup->point = first_point;
   setup->triangle = first_triangle;
}


/** Rasterize all scene's bins */
static void
lp_setup_rasterize_scene( struct lp_setup_context *setup )
{
   struct lp_scene *scene = setup->scene;
   struct llvmpipe_screen *screen = llvmpipe_screen(scene->pipe->screen);

   scene->num_active_queries = setup->active_binned_queries;
   memcpy(scene->active_queries, setup->active_queries,
          scene->num_active_queries * sizeof(scene->active_queries[0]));

   lp_scene_end_binning(scene);

   lp_fence_reference(&setup->last_fence, scene->fence);

   if (setup->last_fence)
      setup->last_fence->issued = TRUE;

   pipe_mutex_lock(screen->rast_mutex);
   lp_rast_queue_scene(screen->rast, scene);
   lp_rast_finish(screen->rast);
   pipe_mutex_unlock(screen->rast_mutex);

   lp_scene_end_rasterization(setup->scene);
   lp_setup_reset( setup );

   LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
}


static boolean
begin_binning( struct lp_setup_context *setup )
{
   struct lp_scene *scene = setup->scene;
   boolean need_zsload = FALSE;
   boolean ok;

   assert(scene);
   assert(scene->fence == NULL);

   /* Always create a fence:
    */
   scene->fence = lp_fence_create(MAX2(1, setup->num_threads));
   if (!scene->fence)
      return FALSE;

   ok = try_update_scene_state(setup);
   if (!ok)
      return FALSE;

   if (setup->fb.zsbuf &&
       ((setup->clear.flags & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL) &&
        util_format_is_depth_and_stencil(setup->fb.zsbuf->format))
      need_zsload = TRUE;

   LP_DBG(DEBUG_SETUP, "%s color clear bufs: %x depth: %s\n", __FUNCTION__,
          setup->clear.flags >> 2,
          need_zsload ? "clear": "load");

   if (setup->clear.flags & PIPE_CLEAR_COLOR) {
      unsigned cbuf;
      for (cbuf = 0; cbuf < setup->fb.nr_cbufs; cbuf++) {
         assert(PIPE_CLEAR_COLOR0 == 1 << 2);
         if (setup->clear.flags & (1 << (2 + cbuf))) {
            union lp_rast_cmd_arg clearrb_arg;
            struct lp_rast_clear_rb *cc_scene =
               (struct lp_rast_clear_rb *)
                  lp_scene_alloc(scene, sizeof(struct lp_rast_clear_rb));

            if (!cc_scene) {
               return FALSE;
            }

            cc_scene->cbuf = cbuf;
            cc_scene->color_val = setup->clear.color_val[cbuf];
            clearrb_arg.clear_rb = cc_scene;

            if (!lp_scene_bin_everywhere(scene,
                                         LP_RAST_OP_CLEAR_COLOR,
                                         clearrb_arg))
               return FALSE;
         }
      }
   }

   if (setup->fb.zsbuf) {
      if (setup->clear.flags & PIPE_CLEAR_DEPTHSTENCIL) {
         ok = lp_scene_bin_everywhere( scene,
                                       LP_RAST_OP_CLEAR_ZSTENCIL,
                                       lp_rast_arg_clearzs(
                                          setup->clear.zsvalue,
                                          setup->clear.zsmask));
         if (!ok)
            return FALSE;
      }
   }

   setup->clear.flags = 0;
   setup->clear.zsmask = 0;
   setup->clear.zsvalue = 0;

   scene->had_queries = !!setup->active_binned_queries;

   LP_DBG(DEBUG_SETUP, "%s done\n", __FUNCTION__);
   return TRUE;
}


/* This basically bins and then flushes any outstanding full-screen
 * clears.
 *
 * TODO: fast path for fullscreen clears and no triangles.
 */
static boolean
execute_clears( struct lp_setup_context *setup )
{
   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   return begin_binning( setup );
}

const char *states[] = {
   "FLUSHED",
   "CLEARED",
   "ACTIVE "
};


static boolean
set_scene_state( struct lp_setup_context *setup,
                 enum setup_state new_state,
                 const char *reason)
{
   unsigned old_state = setup->state;

   if (old_state == new_state)
      return TRUE;

   if (LP_DEBUG & DEBUG_SCENE) {
      debug_printf("%s old %s new %s%s%s\n",
                   __FUNCTION__,
                   states[old_state],
                   states[new_state],
                   (new_state == SETUP_FLUSHED) ? ": " : "",
                   (new_state == SETUP_FLUSHED) ? reason : "");

      if (new_state == SETUP_FLUSHED && setup->scene)
         lp_debug_draw_bins_by_cmd_length(setup->scene);
   }

   /* wait for a free/empty scene
    */
   if (old_state == SETUP_FLUSHED)
      lp_setup_get_empty_scene(setup);

   switch (new_state) {
   case SETUP_CLEARED:
      break;

   case SETUP_ACTIVE:
      if (!begin_binning( setup ))
         goto fail;
      break;

   case SETUP_FLUSHED:
      if (old_state == SETUP_CLEARED)
         if (!execute_clears( setup ))
            goto fail;

      lp_setup_rasterize_scene( setup );
      assert(setup->scene == NULL);
      break;

   default:
      assert(0 && "invalid setup state mode");
      goto fail;
   }

   setup->state = new_state;
   return TRUE;

fail:
   if (setup->scene) {
      lp_scene_end_rasterization(setup->scene);
      setup->scene = NULL;
   }

   setup->state = SETUP_FLUSHED;
   lp_setup_reset( setup );
   return FALSE;
}


void
lp_setup_flush( struct lp_setup_context *setup,
                struct pipe_fence_handle **fence,
                const char *reason)
{
   set_scene_state( setup, SETUP_FLUSHED, reason );

   if (fence) {
      lp_fence_reference((struct lp_fence **)fence, setup->last_fence);
   }
}


void
lp_setup_bind_framebuffer( struct lp_setup_context *setup,
                           const struct pipe_framebuffer_state *fb )
{
   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   /* Flush any old scene.
    */
   set_scene_state( setup, SETUP_FLUSHED, __FUNCTION__ );

   /*
    * Ensure the old scene is not reused.
    */
   assert(!setup->scene);

   /* Set new state.  This will be picked up later when we next need a
    * scene.
    */
   util_copy_framebuffer_state(&setup->fb, fb);
   setup->framebuffer.x0 = 0;
   setup->framebuffer.y0 = 0;
   setup->framebuffer.x1 = fb->width-1;
   setup->framebuffer.y1 = fb->height-1;
   setup->dirty |= LP_SETUP_NEW_SCISSOR;
}


/*
 * Try to clear one color buffer of the attached fb, either by binning a clear
 * command or queuing up the clear for later (when binning is started).
 */
static boolean
lp_setup_try_clear_color_buffer(struct lp_setup_context *setup,
                                const union pipe_color_union *color,
                                unsigned cbuf)
{
   union lp_rast_cmd_arg clearrb_arg;
   union util_color uc;
   enum pipe_format format = setup->fb.cbufs[cbuf]->format;

   LP_DBG(DEBUG_SETUP, "%s state %d\n", __FUNCTION__, setup->state);

   if (util_format_is_pure_integer(format)) {
      /*
       * We expect int/uint clear values here, though some APIs
       * might disagree (but in any case util_pack_color()
       * couldn't handle it)...
       */
      if (util_format_is_pure_sint(format)) {
         util_format_write_4i(format, color->i, 0, &uc, 0, 0, 0, 1, 1);
      }
      else {
         assert(util_format_is_pure_uint(format));
         util_format_write_4ui(format, color->ui, 0, &uc, 0, 0, 0, 1, 1);
      }
   }
   else {
      util_pack_color(color->f, format, &uc);
   }

   if (setup->state == SETUP_ACTIVE) {
      struct lp_scene *scene = setup->scene;

      /* Add the clear to existing scene.  In the unusual case where
       * both color and depth-stencil are being cleared when there's
       * already been some rendering, we could discard the currently
       * binned scene and start again, but I don't see that as being
       * a common usage.
       */
      struct lp_rast_clear_rb *cc_scene =
         (struct lp_rast_clear_rb *)
            lp_scene_alloc_aligned(scene, sizeof(struct lp_rast_clear_rb), 8);

      if (!cc_scene) {
         return FALSE;
      }

      cc_scene->cbuf = cbuf;
      cc_scene->color_val = uc;
      clearrb_arg.clear_rb = cc_scene;

      if (!lp_scene_bin_everywhere(scene,
                                   LP_RAST_OP_CLEAR_COLOR,
                                   clearrb_arg))
         return FALSE;
   }
   else {
      /* Put ourselves into the 'pre-clear' state, specifically to try
       * and accumulate multiple clears to color and depth_stencil
       * buffers which the app or state-tracker might issue
       * separately.
       */
      set_scene_state( setup, SETUP_CLEARED, __FUNCTION__ );

      assert(PIPE_CLEAR_COLOR0 == (1 << 2));
      setup->clear.flags |= 1 << (cbuf + 2);
      setup->clear.color_val[cbuf] = uc;
   }

   return TRUE;
}

static boolean
lp_setup_try_clear_zs(struct lp_setup_context *setup,
                      double depth,
                      unsigned stencil,
                      unsigned flags)
{
   uint64_t zsmask = 0;
   uint64_t zsvalue = 0;
   uint32_t zmask32;
   uint8_t smask8;

   LP_DBG(DEBUG_SETUP, "%s state %d\n", __FUNCTION__, setup->state);

   zmask32 = (flags & PIPE_CLEAR_DEPTH) ? ~0 : 0;
   smask8 = (flags & PIPE_CLEAR_STENCIL) ? ~0 : 0;

   zsvalue = util_pack64_z_stencil(setup->fb.zsbuf->format,
                                   depth,
                                   stencil);

   zsmask = util_pack64_mask_z_stencil(setup->fb.zsbuf->format,
                                       zmask32,
                                       smask8);

   zsvalue &= zsmask;

   if (setup->state == SETUP_ACTIVE) {
      struct lp_scene *scene = setup->scene;

      /* Add the clear to existing scene.  In the unusual case where
       * both color and depth-stencil are being cleared when there's
       * already been some rendering, we could discard the currently
       * binned scene and start again, but I don't see that as being
       * a common usage.
       */
      if (!lp_scene_bin_everywhere(scene,
                                   LP_RAST_OP_CLEAR_ZSTENCIL,
                                   lp_rast_arg_clearzs(zsvalue, zsmask)))
         return FALSE;
   }
   else {
      /* Put ourselves into the 'pre-clear' state, specifically to try
       * and accumulate multiple clears to color and depth_stencil
       * buffers which the app or state-tracker might issue
       * separately.
       */
      set_scene_state( setup, SETUP_CLEARED, __FUNCTION__ );

      setup->clear.flags |= flags;

      setup->clear.zsmask |= zsmask;
      setup->clear.zsvalue =
         (setup->clear.zsvalue & ~zsmask) | (zsvalue & zsmask);
   }

   return TRUE;
}

void
lp_setup_clear( struct lp_setup_context *setup,
                const union pipe_color_union *color,
                double depth,
                unsigned stencil,
                unsigned flags )
{
   unsigned i;

   /*
    * Note any of these (max 9) clears could fail (but at most there should
    * be just one failure!). This avoids doing the previous succeeded
    * clears again (we still clear tiles twice if a clear command succeeded
    * partially for one buffer).
    */
   if (flags & PIPE_CLEAR_DEPTHSTENCIL) {
      unsigned flagszs = flags & PIPE_CLEAR_DEPTHSTENCIL;
      if (!lp_setup_try_clear_zs(setup, depth, stencil, flagszs)) {
         lp_setup_flush(setup, NULL, __FUNCTION__);

         if (!lp_setup_try_clear_zs(setup, depth, stencil, flagszs))
            assert(0);
      }
   }

   if (flags & PIPE_CLEAR_COLOR) {
      assert(PIPE_CLEAR_COLOR0 == (1 << 2));
      for (i = 0; i < setup->fb.nr_cbufs; i++) {
         if ((flags & (1 << (2 + i))) && setup->fb.cbufs[i]) {
            if (!lp_setup_try_clear_color_buffer(setup, color, i)) {
               lp_setup_flush(setup, NULL, __FUNCTION__);

               if (!lp_setup_try_clear_color_buffer(setup, color, i))
                  assert(0);
            }
         }
      }
   }
}


void
lp_setup_set_triangle_state( struct lp_setup_context *setup,
                             unsigned cull_mode,
                             boolean ccw_is_frontface,
                             boolean scissor,
                             boolean half_pixel_center,
                             boolean bottom_edge_rule)
{
   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   setup->ccw_is_frontface = ccw_is_frontface;
   setup->cullmode = cull_mode;
   setup->triangle = first_triangle;
   setup->pixel_offset = half_pixel_center ? 0.5f : 0.0f;
   setup->bottom_edge_rule = bottom_edge_rule;

   if (setup->scissor_test != scissor) {
      setup->dirty |= LP_SETUP_NEW_SCISSOR;
      setup->scissor_test = scissor;
   }
}

void
lp_setup_set_line_state( struct lp_setup_context *setup,
			 float line_width)
{
   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   setup->line_width = line_width;
}

void
lp_setup_set_point_state( struct lp_setup_context *setup,
                          float point_size,
                          boolean point_size_per_vertex,
                          uint sprite_coord_enable,
                          uint sprite_coord_origin)
{
   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   setup->point_size = point_size;
   setup->sprite_coord_enable = sprite_coord_enable;
   setup->sprite_coord_origin = sprite_coord_origin;
   setup->point_size_per_vertex = point_size_per_vertex;
}

void
lp_setup_set_setup_variant( struct lp_setup_context *setup,
			    const struct lp_setup_variant *variant)
{
   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   setup->setup.variant = variant;
}

void
lp_setup_set_fs_variant( struct lp_setup_context *setup,
                         struct lp_fragment_shader_variant *variant)
{
   LP_DBG(DEBUG_SETUP, "%s %p\n", __FUNCTION__,
          variant);
   /* FIXME: reference count */

   setup->fs.current.variant = variant;
   setup->dirty |= LP_SETUP_NEW_FS;
}

void
lp_setup_set_fs_constants(struct lp_setup_context *setup,
                          unsigned num,
                          struct pipe_constant_buffer *buffers)
{
   unsigned i;

   LP_DBG(DEBUG_SETUP, "%s %p\n", __FUNCTION__, (void *) buffers);

   assert(num <= Elements(setup->constants));

   for (i = 0; i < num; ++i) {
      util_copy_constant_buffer(&setup->constants[i].current, &buffers[i]);
   }
   for (; i < Elements(setup->constants); i++) {
      util_copy_constant_buffer(&setup->constants[i].current, NULL);
   }
   setup->dirty |= LP_SETUP_NEW_CONSTANTS;
}


void
lp_setup_set_alpha_ref_value( struct lp_setup_context *setup,
                              float alpha_ref_value )
{
   LP_DBG(DEBUG_SETUP, "%s %f\n", __FUNCTION__, alpha_ref_value);

   if(setup->fs.current.jit_context.alpha_ref_value != alpha_ref_value) {
      setup->fs.current.jit_context.alpha_ref_value = alpha_ref_value;
      setup->dirty |= LP_SETUP_NEW_FS;
   }
}

void
lp_setup_set_stencil_ref_values( struct lp_setup_context *setup,
                                 const ubyte refs[2] )
{
   LP_DBG(DEBUG_SETUP, "%s %d %d\n", __FUNCTION__, refs[0], refs[1]);

   if (setup->fs.current.jit_context.stencil_ref_front != refs[0] ||
       setup->fs.current.jit_context.stencil_ref_back != refs[1]) {
      setup->fs.current.jit_context.stencil_ref_front = refs[0];
      setup->fs.current.jit_context.stencil_ref_back = refs[1];
      setup->dirty |= LP_SETUP_NEW_FS;
   }
}

void
lp_setup_set_blend_color( struct lp_setup_context *setup,
                          const struct pipe_blend_color *blend_color )
{
   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   assert(blend_color);

   if(memcmp(&setup->blend_color.current, blend_color, sizeof *blend_color) != 0) {
      memcpy(&setup->blend_color.current, blend_color, sizeof *blend_color);
      setup->dirty |= LP_SETUP_NEW_BLEND_COLOR;
   }
}


void
lp_setup_set_scissors( struct lp_setup_context *setup,
                       const struct pipe_scissor_state *scissors )
{
   unsigned i;
   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   assert(scissors);

   for (i = 0; i < PIPE_MAX_VIEWPORTS; ++i) {
      setup->scissors[i].x0 = scissors[i].minx;
      setup->scissors[i].x1 = scissors[i].maxx-1;
      setup->scissors[i].y0 = scissors[i].miny;
      setup->scissors[i].y1 = scissors[i].maxy-1;
   }
   setup->dirty |= LP_SETUP_NEW_SCISSOR;
}


void
lp_setup_set_flatshade_first( struct lp_setup_context *setup,
                              boolean flatshade_first )
{
   setup->flatshade_first = flatshade_first;
}

void
lp_setup_set_rasterizer_discard( struct lp_setup_context *setup,
                                 boolean rasterizer_discard )
{
   if (setup->rasterizer_discard != rasterizer_discard) {
      setup->rasterizer_discard = rasterizer_discard;
      set_scene_state( setup, SETUP_FLUSHED, __FUNCTION__ );
   }
}

void
lp_setup_set_vertex_info( struct lp_setup_context *setup,
                          struct vertex_info *vertex_info )
{
   /* XXX: just silently holding onto the pointer:
    */
   setup->vertex_info = vertex_info;
}


/**
 * Called during state validation when LP_NEW_VIEWPORT is set.
 */
void
lp_setup_set_viewports(struct lp_setup_context *setup,
                       unsigned num_viewports,
                       const struct pipe_viewport_state *viewports)
{
   struct llvmpipe_context *lp = llvmpipe_context(setup->pipe);
   unsigned i;

   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   assert(num_viewports <= PIPE_MAX_VIEWPORTS);
   assert(viewports);

   /*
    * For use in lp_state_fs.c, propagate the viewport values for all viewports.
    */
   for (i = 0; i < num_viewports; i++) {
      float min_depth;
      float max_depth;

      if (lp->rasterizer->clip_halfz == 0) {
         float half_depth = viewports[i].scale[2];
         min_depth = viewports[i].translate[2] - half_depth;
         max_depth = min_depth + half_depth * 2.0f;
      } else {
         min_depth = viewports[i].translate[2];
         max_depth = min_depth + viewports[i].scale[2];
      }

      if (setup->viewports[i].min_depth != min_depth ||
          setup->viewports[i].max_depth != max_depth) {
          setup->viewports[i].min_depth = min_depth;
          setup->viewports[i].max_depth = max_depth;
          setup->dirty |= LP_SETUP_NEW_VIEWPORTS;
      }
   }
}


/**
 * Called during state validation when LP_NEW_SAMPLER_VIEW is set.
 */
void
lp_setup_set_fragment_sampler_views(struct lp_setup_context *setup,
                                    unsigned num,
                                    struct pipe_sampler_view **views)
{
   unsigned i;

   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   assert(num <= PIPE_MAX_SHADER_SAMPLER_VIEWS);

   for (i = 0; i < PIPE_MAX_SHADER_SAMPLER_VIEWS; i++) {
      struct pipe_sampler_view *view = i < num ? views[i] : NULL;

      if (view) {
         struct pipe_resource *res = view->texture;
         struct llvmpipe_resource *lp_tex = llvmpipe_resource(res);
         struct lp_jit_texture *jit_tex;
         jit_tex = &setup->fs.current.jit_context.textures[i];

         /* We're referencing the texture's internal data, so save a
          * reference to it.
          */
         pipe_resource_reference(&setup->fs.current_tex[i], res);

         if (!lp_tex->dt) {
            /* regular texture - setup array of mipmap level offsets */
            int j;
            unsigned first_level = 0;
            unsigned last_level = 0;

            if (llvmpipe_resource_is_texture(res)) {
               first_level = view->u.tex.first_level;
               last_level = view->u.tex.last_level;
               assert(first_level <= last_level);
               assert(last_level <= res->last_level);
               jit_tex->base = lp_tex->tex_data;
            }
            else {
              jit_tex->base = lp_tex->data;
            }

            if (LP_PERF & PERF_TEX_MEM) {
               /* use dummy tile memory */
               jit_tex->base = lp_dummy_tile;
               jit_tex->width = TILE_SIZE/8;
               jit_tex->height = TILE_SIZE/8;
               jit_tex->depth = 1;
               jit_tex->first_level = 0;
               jit_tex->last_level = 0;
               jit_tex->mip_offsets[0] = 0;
               jit_tex->row_stride[0] = 0;
               jit_tex->img_stride[0] = 0;
            }
            else {
               jit_tex->width = res->width0;
               jit_tex->height = res->height0;
               jit_tex->depth = res->depth0;
               jit_tex->first_level = first_level;
               jit_tex->last_level = last_level;

               if (llvmpipe_resource_is_texture(res)) {
                  for (j = first_level; j <= last_level; j++) {
                     jit_tex->mip_offsets[j] = lp_tex->mip_offsets[j];
                     jit_tex->row_stride[j] = lp_tex->row_stride[j];
                     jit_tex->img_stride[j] = lp_tex->img_stride[j];
                  }

                  if (res->target == PIPE_TEXTURE_1D_ARRAY ||
                      res->target == PIPE_TEXTURE_2D_ARRAY ||
                      res->target == PIPE_TEXTURE_CUBE_ARRAY) {
                     /*
                      * For array textures, we don't have first_layer, instead
                      * adjust last_layer (stored as depth) plus the mip level offsets
                      * (as we have mip-first layout can't just adjust base ptr).
                      * XXX For mip levels, could do something similar.
                      */
                     jit_tex->depth = view->u.tex.last_layer - view->u.tex.first_layer + 1;
                     for (j = first_level; j <= last_level; j++) {
                        jit_tex->mip_offsets[j] += view->u.tex.first_layer *
                                                   lp_tex->img_stride[j];
                     }
                     if (res->target == PIPE_TEXTURE_CUBE_ARRAY) {
                        assert(jit_tex->depth % 6 == 0);
                     }
                     assert(view->u.tex.first_layer <= view->u.tex.last_layer);
                     assert(view->u.tex.last_layer < res->array_size);
                  }
               }
               else {
                  /*
                   * For buffers, we don't have first_element, instead adjust
                   * last_element (stored as width) plus the base pointer.
                   */
                  unsigned view_blocksize = util_format_get_blocksize(view->format);
                  /* probably don't really need to fill that out */
                  jit_tex->mip_offsets[0] = 0;
                  jit_tex->row_stride[0] = 0;
                  jit_tex->row_stride[0] = 0;

                  /* everything specified in number of elements here. */
                  jit_tex->width = view->u.buf.last_element - view->u.buf.first_element + 1;
                  jit_tex->base = (uint8_t *)jit_tex->base + view->u.buf.first_element *
                                  view_blocksize;
                  /* XXX Unsure if we need to sanitize parameters? */
                  assert(view->u.buf.first_element <= view->u.buf.last_element);
                  assert(view->u.buf.last_element * view_blocksize < res->width0);
               }
            }
         }
         else {
            /* display target texture/surface */
            /*
             * XXX: Where should this be unmapped?
             */
            struct llvmpipe_screen *screen = llvmpipe_screen(res->screen);
            struct sw_winsys *winsys = screen->winsys;
            jit_tex->base = winsys->displaytarget_map(winsys, lp_tex->dt,
                                                         PIPE_TRANSFER_READ);
            jit_tex->row_stride[0] = lp_tex->row_stride[0];
            jit_tex->img_stride[0] = lp_tex->img_stride[0];
            jit_tex->mip_offsets[0] = 0;
            jit_tex->width = res->width0;
            jit_tex->height = res->height0;
            jit_tex->depth = res->depth0;
            jit_tex->first_level = jit_tex->last_level = 0;
            assert(jit_tex->base);
         }
      }
   }

   setup->dirty |= LP_SETUP_NEW_FS;
}


/**
 * Called during state validation when LP_NEW_SAMPLER is set.
 */
void
lp_setup_set_fragment_sampler_state(struct lp_setup_context *setup,
                                    unsigned num,
                                    struct pipe_sampler_state **samplers)
{
   unsigned i;

   LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);

   assert(num <= PIPE_MAX_SAMPLERS);

   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
      const struct pipe_sampler_state *sampler = i < num ? samplers[i] : NULL;

      if (sampler) {
         struct lp_jit_sampler *jit_sam;
         jit_sam = &setup->fs.current.jit_context.samplers[i];

         jit_sam->min_lod = sampler->min_lod;
         jit_sam->max_lod = sampler->max_lod;
         jit_sam->lod_bias = sampler->lod_bias;
         COPY_4V(jit_sam->border_color, sampler->border_color.f);
      }
   }

   setup->dirty |= LP_SETUP_NEW_FS;
}


/**
 * Is the given texture referenced by any scene?
 * Note: we have to check all scenes including any scenes currently
 * being rendered and the current scene being built.
 */
unsigned
lp_setup_is_resource_referenced( const struct lp_setup_context *setup,
                                const struct pipe_resource *texture )
{
   unsigned i;

   /* check the render targets */
   for (i = 0; i < setup->fb.nr_cbufs; i++) {
      if (setup->fb.cbufs[i] && setup->fb.cbufs[i]->texture == texture)
         return LP_REFERENCED_FOR_READ | LP_REFERENCED_FOR_WRITE;
   }
   if (setup->fb.zsbuf && setup->fb.zsbuf->texture == texture) {
      return LP_REFERENCED_FOR_READ | LP_REFERENCED_FOR_WRITE;
   }

   /* check textures referenced by the scene */
   for (i = 0; i < Elements(setup->scenes); i++) {
      if (lp_scene_is_resource_referenced(setup->scenes[i], texture)) {
         return LP_REFERENCED_FOR_READ;
      }
   }

   return LP_UNREFERENCED;
}


/**
 * Called by vbuf code when we're about to draw something.
 *
 * This function stores all dirty state in the current scene's display list
 * memory, via lp_scene_alloc().  We can not pass pointers of mutable state to
 * the JIT functions, as the JIT functions will be called later on, most likely
 * on a different thread.
 *
 * When processing dirty state it is imperative that we don't refer to any
 * pointers previously allocated with lp_scene_alloc() in this function (or any
 * function) as they may belong to a scene freed since then.
 */
static boolean
try_update_scene_state( struct lp_setup_context *setup )
{
   boolean new_scene = (setup->fs.stored == NULL);
   struct lp_scene *scene = setup->scene;
   unsigned i;

   assert(scene);

   if (setup->dirty & LP_SETUP_NEW_VIEWPORTS) {
      /*
       * Record new depth range state for changes due to viewport updates.
       *
       * TODO: Collapse the existing viewport and depth range information
       *       into one structure, for access by JIT.
       */
      struct lp_jit_viewport *stored;

      stored = (struct lp_jit_viewport *)
         lp_scene_alloc(scene, sizeof setup->viewports);

      if (!stored) {
         assert(!new_scene);
         return FALSE;
      }

      memcpy(stored, setup->viewports, sizeof setup->viewports);

      setup->fs.current.jit_context.viewports = stored;
      setup->dirty |= LP_SETUP_NEW_FS;
   }

   if(setup->dirty & LP_SETUP_NEW_BLEND_COLOR) {
      uint8_t *stored;
      float* fstored;
      unsigned i, j;
      unsigned size;

      /* Alloc u8_blend_color (16 x i8) and f_blend_color (4 or 8 x f32) */
      size  = 4 * 16 * sizeof(uint8_t);
      size += (LP_MAX_VECTOR_LENGTH / 4) * sizeof(float);
      stored = lp_scene_alloc_aligned(scene, size, LP_MIN_VECTOR_ALIGN);

      if (!stored) {
         assert(!new_scene);
         return FALSE;
      }

      /* Store floating point colour */
      fstored = (float*)(stored + 4*16);
      for (i = 0; i < (LP_MAX_VECTOR_LENGTH / 4); ++i) {
         fstored[i] = setup->blend_color.current.color[i % 4];
      }

      /* smear each blend color component across 16 ubyte elements */
      for (i = 0; i < 4; ++i) {
         uint8_t c = float_to_ubyte(setup->blend_color.current.color[i]);
         for (j = 0; j < 16; ++j)
            stored[i*16 + j] = c;
      }

      setup->blend_color.stored = stored;
      setup->fs.current.jit_context.u8_blend_color = stored;
      setup->fs.current.jit_context.f_blend_color = fstored;
      setup->dirty |= LP_SETUP_NEW_FS;
   }

   if (setup->dirty & LP_SETUP_NEW_CONSTANTS) {
      for (i = 0; i < Elements(setup->constants); ++i) {
         struct pipe_resource *buffer = setup->constants[i].current.buffer;
         const unsigned current_size = setup->constants[i].current.buffer_size;
         const ubyte *current_data = NULL;
         int num_constants;

         if (buffer) {
            /* resource buffer */
            current_data = (ubyte *) llvmpipe_resource_data(buffer);
         }
         else if (setup->constants[i].current.user_buffer) {
            /* user-space buffer */
            current_data = (ubyte *) setup->constants[i].current.user_buffer;
         }

         if (current_data) {
            current_data += setup->constants[i].current.buffer_offset;

            /* TODO: copy only the actually used constants? */

            if (setup->constants[i].stored_size != current_size ||
               !setup->constants[i].stored_data ||
               memcmp(setup->constants[i].stored_data,
                      current_data,
                      current_size) != 0) {
               void *stored;

               stored = lp_scene_alloc(scene, current_size);
               if (!stored) {
                  assert(!new_scene);
                  return FALSE;
               }

               memcpy(stored,
                      current_data,
                      current_size);
               setup->constants[i].stored_size = current_size;
               setup->constants[i].stored_data = stored;
            }
         }
         else {
            setup->constants[i].stored_size = 0;
            setup->constants[i].stored_data = NULL;
         }

         setup->fs.current.jit_context.constants[i] =
            setup->constants[i].stored_data;
         num_constants =
            setup->constants[i].stored_size / (sizeof(float) * 4);
         setup->fs.current.jit_context.num_constants[i] = num_constants;
         setup->dirty |= LP_SETUP_NEW_FS;
      }
   }


   if (setup->dirty & LP_SETUP_NEW_FS) {
      if (!setup->fs.stored ||
          memcmp(setup->fs.stored,
                 &setup->fs.current,
                 sizeof setup->fs.current) != 0)
      {
         struct lp_rast_state *stored;

         /* The fs state that's been stored in the scene is different from
          * the new, current state.  So allocate a new lp_rast_state object
          * and append it to the bin's setup data buffer.
          */
         stored = (struct lp_rast_state *) lp_scene_alloc(scene, sizeof *stored);
         if (!stored) {
            assert(!new_scene);
            return FALSE;
         }

         memcpy(stored,
                &setup->fs.current,
                sizeof setup->fs.current);
         setup->fs.stored = stored;

         /* The scene now references the textures in the rasterization
          * state record.  Note that now.
          */
         for (i = 0; i < Elements(setup->fs.current_tex); i++) {
            if (setup->fs.current_tex[i]) {
               if (!lp_scene_add_resource_reference(scene,
                                                    setup->fs.current_tex[i],
                                                    new_scene)) {
                  assert(!new_scene);
                  return FALSE;
               }
            }
         }
      }
   }

   if (setup->dirty & LP_SETUP_NEW_SCISSOR) {
      unsigned i;
      for (i = 0; i < PIPE_MAX_VIEWPORTS; ++i) {
         setup->draw_regions[i] = setup->framebuffer;
         if (setup->scissor_test) {
            u_rect_possible_intersection(&setup->scissors[i],
                                         &setup->draw_regions[i]);
         }
      }
   }

   setup->dirty = 0;

   assert(setup->fs.stored);
   return TRUE;
}

boolean
lp_setup_update_state( struct lp_setup_context *setup,
                       boolean update_scene )
{
   /* Some of the 'draw' pipeline stages may have changed some driver state.
    * Make sure we've processed those state changes before anything else.
    *
    * XXX this is the only place where llvmpipe_context is used in the
    * setup code.  This may get refactored/changed...
    */
   {
      struct llvmpipe_context *lp = llvmpipe_context(setup->pipe);
      if (lp->dirty) {
         llvmpipe_update_derived(lp);
      }

      if (lp->setup->dirty) {
         llvmpipe_update_setup(lp);
      }

      assert(setup->setup.variant);

      /* Will probably need to move this somewhere else, just need
       * to know about vertex shader point size attribute.
       */
      setup->psize = lp->psize_slot;
      setup->viewport_index_slot = lp->viewport_index_slot;
      setup->layer_slot = lp->layer_slot;
      setup->face_slot = lp->face_slot;

      assert(lp->dirty == 0);

      assert(lp->setup_variant.key.size ==
	     setup->setup.variant->key.size);

      assert(memcmp(&lp->setup_variant.key,
		    &setup->setup.variant->key,
		    setup->setup.variant->key.size) == 0);
   }

   if (update_scene && setup->state != SETUP_ACTIVE) {
      if (!set_scene_state( setup, SETUP_ACTIVE, __FUNCTION__ ))
         return FALSE;
   }

   /* Only call into update_scene_state() if we already have a
    * scene:
    */
   if (update_scene && setup->scene) {
      assert(setup->state == SETUP_ACTIVE);

      if (try_update_scene_state(setup))
         return TRUE;

      /* Update failed, try to restart the scene.
       *
       * Cannot call lp_setup_flush_and_restart() directly here
       * because of potential recursion.
       */
      if (!set_scene_state(setup, SETUP_FLUSHED, __FUNCTION__))
         return FALSE;

      if (!set_scene_state(setup, SETUP_ACTIVE, __FUNCTION__))
         return FALSE;

      if (!setup->scene)
         return FALSE;

      return try_update_scene_state(setup);
   }

   return TRUE;
}


/* Only caller is lp_setup_vbuf_destroy()
 */
void
lp_setup_destroy( struct lp_setup_context *setup )
{
   uint i;

   lp_setup_reset( setup );

   util_unreference_framebuffer_state(&setup->fb);

   for (i = 0; i < Elements(setup->fs.current_tex); i++) {
      pipe_resource_reference(&setup->fs.current_tex[i], NULL);
   }

   for (i = 0; i < Elements(setup->constants); i++) {
      pipe_resource_reference(&setup->constants[i].current.buffer, NULL);
   }

   /* free the scenes in the 'empty' queue */
   for (i = 0; i < Elements(setup->scenes); i++) {
      struct lp_scene *scene = setup->scenes[i];

      if (scene->fence)
         lp_fence_wait(scene->fence);

      lp_scene_destroy(scene);
   }

   lp_fence_reference(&setup->last_fence, NULL);

   FREE( setup );
}


/**
 * Create a new primitive tiling engine.  Plug it into the backend of
 * the draw module.  Currently also creates a rasterizer to use with
 * it.
 */
struct lp_setup_context *
lp_setup_create( struct pipe_context *pipe,
                 struct draw_context *draw )
{
   struct llvmpipe_screen *screen = llvmpipe_screen(pipe->screen);
   struct lp_setup_context *setup;
   unsigned i;

   setup = CALLOC_STRUCT(lp_setup_context);
   if (!setup) {
      goto no_setup;
   }

   lp_setup_init_vbuf(setup);

   /* Used only in update_state():
    */
   setup->pipe = pipe;


   setup->num_threads = screen->num_threads;
   setup->vbuf = draw_vbuf_stage(draw, &setup->base);
   if (!setup->vbuf) {
      goto no_vbuf;
   }

   draw_set_rasterize_stage(draw, setup->vbuf);
   draw_set_render(draw, &setup->base);

   /* create some empty scenes */
   for (i = 0; i < MAX_SCENES; i++) {
      setup->scenes[i] = lp_scene_create( pipe );
      if (!setup->scenes[i]) {
         goto no_scenes;
      }
   }

   setup->triangle = first_triangle;
   setup->line     = first_line;
   setup->point    = first_point;

   setup->dirty = ~0;

   return setup;

no_scenes:
   for (i = 0; i < MAX_SCENES; i++) {
      if (setup->scenes[i]) {
         lp_scene_destroy(setup->scenes[i]);
      }
   }

   setup->vbuf->destroy(setup->vbuf);
no_vbuf:
   FREE(setup);
no_setup:
   return NULL;
}


/**
 * Put a BeginQuery command into all bins.
 */
void
lp_setup_begin_query(struct lp_setup_context *setup,
                     struct llvmpipe_query *pq)
{

   set_scene_state(setup, SETUP_ACTIVE, "begin_query");

   if (!(pq->type == PIPE_QUERY_OCCLUSION_COUNTER ||
         pq->type == PIPE_QUERY_OCCLUSION_PREDICATE ||
         pq->type == PIPE_QUERY_PIPELINE_STATISTICS))
      return;

   /* init the query to its beginning state */
   assert(setup->active_binned_queries < LP_MAX_ACTIVE_BINNED_QUERIES);
   /* exceeding list size so just ignore the query */
   if (setup->active_binned_queries >= LP_MAX_ACTIVE_BINNED_QUERIES) {
      return;
   }
   assert(setup->active_queries[setup->active_binned_queries] == NULL);
   setup->active_queries[setup->active_binned_queries] = pq;
   setup->active_binned_queries++;

   assert(setup->scene);
   if (setup->scene) {
      if (!lp_scene_bin_everywhere(setup->scene,
                                   LP_RAST_OP_BEGIN_QUERY,
                                   lp_rast_arg_query(pq))) {

         if (!lp_setup_flush_and_restart(setup))
            return;

         if (!lp_scene_bin_everywhere(setup->scene,
                                      LP_RAST_OP_BEGIN_QUERY,
                                      lp_rast_arg_query(pq))) {
            return;
         }
      }
      setup->scene->had_queries |= TRUE;
   }
}


/**
 * Put an EndQuery command into all bins.
 */
void
lp_setup_end_query(struct lp_setup_context *setup, struct llvmpipe_query *pq)
{
   set_scene_state(setup, SETUP_ACTIVE, "end_query");

   assert(setup->scene);
   if (setup->scene) {
      /* pq->fence should be the fence of the *last* scene which
       * contributed to the query result.
       */
      lp_fence_reference(&pq->fence, setup->scene->fence);

      if (pq->type == PIPE_QUERY_OCCLUSION_COUNTER ||
          pq->type == PIPE_QUERY_OCCLUSION_PREDICATE ||
          pq->type == PIPE_QUERY_PIPELINE_STATISTICS ||
          pq->type == PIPE_QUERY_TIMESTAMP) {
         if (pq->type == PIPE_QUERY_TIMESTAMP &&
               !(setup->scene->tiles_x | setup->scene->tiles_y)) {
            /*
             * If there's a zero width/height framebuffer, there's no bins and
             * hence no rast task is ever run. So fill in something here instead.
             */
            pq->end[0] = os_time_get_nano();
         }

         if (!lp_scene_bin_everywhere(setup->scene,
                                      LP_RAST_OP_END_QUERY,
                                      lp_rast_arg_query(pq))) {
            if (!lp_setup_flush_and_restart(setup))
               goto fail;

            if (!lp_scene_bin_everywhere(setup->scene,
                                         LP_RAST_OP_END_QUERY,
                                         lp_rast_arg_query(pq))) {
               goto fail;
            }
         }
         setup->scene->had_queries |= TRUE;
      }
   }
   else {
      lp_fence_reference(&pq->fence, setup->last_fence);
   }

fail:
   /* Need to do this now not earlier since it still needs to be marked as
    * active when binning it would cause a flush.
    */
   if (pq->type == PIPE_QUERY_OCCLUSION_COUNTER ||
      pq->type == PIPE_QUERY_OCCLUSION_PREDICATE ||
      pq->type == PIPE_QUERY_PIPELINE_STATISTICS) {
      unsigned i;

      /* remove from active binned query list */
      for (i = 0; i < setup->active_binned_queries; i++) {
         if (setup->active_queries[i] == pq)
            break;
      }
      assert(i < setup->active_binned_queries);
      if (i == setup->active_binned_queries)
         return;
      setup->active_binned_queries--;
      setup->active_queries[i] = setup->active_queries[setup->active_binned_queries];
      setup->active_queries[setup->active_binned_queries] = NULL;
   }
}


boolean
lp_setup_flush_and_restart(struct lp_setup_context *setup)
{
   if (0) debug_printf("%s\n", __FUNCTION__);

   assert(setup->state == SETUP_ACTIVE);

   if (!set_scene_state(setup, SETUP_FLUSHED, __FUNCTION__))
      return FALSE;

   if (!lp_setup_update_state(setup, TRUE))
      return FALSE;

   return TRUE;
}