svga: implement user index buffers
[mesa.git] / src / gallium / drivers / svga / svga_pipe_draw.c
index d53edcb23c534de320e9f880a5cf1c823e7e3a8d..bbd443049507da7dcbaa0bb0b655c780fc7189e4 100644 (file)
  *
  **********************************************************/
 
-#include "svga_cmd.h"
 
 #include "util/u_format.h"
+#include "util/u_helpers.h"
 #include "util/u_inlines.h"
 #include "util/u_prim.h"
+#include "util/u_prim_restart.h"
 #include "util/u_time.h"
+#include "util/u_upload_mgr.h"
 #include "indices/u_indices.h"
 
 #include "svga_hw_reg.h"
+#include "svga_cmd.h"
 #include "svga_context.h"
 #include "svga_screen.h"
 #include "svga_draw.h"
+#include "svga_shader.h"
 #include "svga_state.h"
+#include "svga_surface.h"
 #include "svga_swtnl.h"
 #include "svga_debug.h"
 #include "svga_resource_buffer.h"
-#include "util/u_upload_mgr.h"
-
-/**
- * Determine the ranges to upload for the user-buffers referenced
- * by the next draw command.
- *
- * TODO: It might be beneficial to support multiple ranges. In that case,
- * the struct svga_buffer::uploaded member should be made an array or a
- * list, since we need to account for the possibility that different ranges
- * may be uploaded to different hardware buffers chosen by the utility
- * upload manager.
- */
-
-static void
-svga_user_buffer_range(struct svga_context *svga,
-                       unsigned start,
-                       unsigned count,
-                       unsigned instance_count)
-{
-   const struct pipe_vertex_element *ve = svga->curr.velems->velem;
-   int i;
-
-   /*
-    * Release old uploaded range (if not done already) and
-    * initialize new ranges.
-    */
-
-   for (i=0; i < svga->curr.velems->count; i++) {
-      struct pipe_vertex_buffer *vb =
-         &svga->curr.vb[ve[i].vertex_buffer_index];
-
-      if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
-         struct svga_buffer *buffer = svga_buffer(vb->buffer);
-
-         pipe_resource_reference(&buffer->uploaded.buffer, NULL);
-         buffer->uploaded.start = ~0;
-         buffer->uploaded.end = 0;
-      }
-   }
-
-   for (i=0; i < svga->curr.velems->count; i++) {
-      struct pipe_vertex_buffer *vb =
-         &svga->curr.vb[ve[i].vertex_buffer_index];
-
-      if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
-         struct svga_buffer *buffer = svga_buffer(vb->buffer);
-         unsigned first, size;
-         unsigned instance_div = ve[i].instance_divisor;
-         unsigned elemSize = util_format_get_blocksize(ve[i].src_format);
-
-         svga->dirty |= SVGA_NEW_VBUFFER;
-
-         if (instance_div) {
-            first = ve[i].src_offset;
-            count = (instance_count + instance_div - 1) / instance_div;
-            size = vb->stride * (count - 1) + elemSize;
-         } else if (vb->stride) {
-            first = vb->stride * start + ve[i].src_offset;
-            size = vb->stride * (count - 1) + elemSize;
-         } else {
-            /* Only a single vertex!
-             * Upload with the largest vertex size the hw supports,
-             * if possible.
-             */
-            first = ve[i].src_offset;
-            size = MIN2(16, vb->buffer->width0);
-         }
-
-         buffer->uploaded.start = MIN2(buffer->uploaded.start, first);
-         buffer->uploaded.end = MAX2(buffer->uploaded.end, first + size);
-      }
-   }
-}
-
-/**
- * svga_upload_user_buffers - upload parts of user buffers
- *
- * This function streams a part of a user buffer to hw and fills
- * svga_buffer::uploaded with information on the upload.
- */
-
-static int
-svga_upload_user_buffers(struct svga_context *svga,
-                         unsigned start,
-                         unsigned count,
-                         unsigned instance_count)
-{
-   const struct pipe_vertex_element *ve = svga->curr.velems->velem;
-   unsigned i;
-   int ret;
-
-   svga_user_buffer_range(svga, start, count, instance_count);
-
-   for (i=0; i < svga->curr.velems->count; i++) {
-      struct pipe_vertex_buffer *vb =
-         &svga->curr.vb[ve[i].vertex_buffer_index];
-
-      if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
-         struct svga_buffer *buffer = svga_buffer(vb->buffer);
-         boolean flushed;
-
-         /*
-          * Check if already uploaded. Otherwise go ahead and upload.
-          */
-
-         if (buffer->uploaded.buffer)
-            continue;
-
-         ret = u_upload_buffer( svga->upload_vb,
-                                0,
-                                buffer->uploaded.start,
-                                buffer->uploaded.end - buffer->uploaded.start,
-                                &buffer->b.b,
-                                &buffer->uploaded.offset,
-                                &buffer->uploaded.buffer,
-                                &flushed);
-
-         if (ret)
-            return ret;
-
-         if (0)
-            debug_printf("%s: %d: orig buf %p upl buf %p ofs %d sofs %d"
-                         " sz %d\n",
-                         __FUNCTION__,
-                         i,
-                         buffer,
-                         buffer->uploaded.buffer,
-                         buffer->uploaded.offset,
-                         buffer->uploaded.start,
-                         buffer->uploaded.end - buffer->uploaded.start);
-
-         vb->buffer_offset = buffer->uploaded.offset;
-      }
-   }
-
-   return PIPE_OK;
-}
-
-/**
- * svga_release_user_upl_buffers - release uploaded parts of user buffers
- *
- * This function releases the hw copy of the uploaded fraction of the
- * user-buffer. It's important to do this as soon as all draw calls
- * affecting the uploaded fraction are issued, as this allows for
- * efficient reuse of the hardware surface backing the uploaded fraction.
- *
- * svga_buffer::source_offset is set to 0, and svga_buffer::uploaded::buffer
- * is set to 0.
- */
-
-static void
-svga_release_user_upl_buffers(struct svga_context *svga)
-{
-   unsigned i;
-   unsigned nr;
-
-   nr = svga->curr.num_vertex_buffers;
-
-   for (i = 0; i < nr; ++i) {
-      struct pipe_vertex_buffer *vb = &svga->curr.vb[i];
-
-      if (vb->buffer && svga_buffer_is_user_buffer(vb->buffer)) {
-         struct svga_buffer *buffer = svga_buffer(vb->buffer);
-
-         buffer->uploaded.start = ~0;
-         buffer->uploaded.end = 0;
-         if (buffer->uploaded.buffer)
-            pipe_resource_reference(&buffer->uploaded.buffer, NULL);
-      }
-   }
-}
-
-
 
 static enum pipe_error
 retry_draw_range_elements( struct svga_context *svga,
@@ -220,105 +52,132 @@ retry_draw_range_elements( struct svga_context *svga,
                            int index_bias,
                            unsigned min_index,
                            unsigned max_index,
-                           unsigned prim, 
-                           unsigned start, 
+                           enum pipe_prim_type prim,
+                           unsigned start,
                            unsigned count,
+                           unsigned start_instance,
                            unsigned instance_count,
                            boolean do_retry )
 {
-   enum pipe_error ret = 0;
+   enum pipe_error ret = PIPE_OK;
 
-   svga_hwtnl_set_unfilled( svga->hwtnl,
-                            svga->curr.rast->hw_unfilled );
+   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_DRAWELEMENTS);
 
-   svga_hwtnl_set_flatshade( svga->hwtnl,
-                             svga->curr.rast->templ.flatshade,
-                             svga->curr.rast->templ.flatshade_first );
+   svga_hwtnl_set_fillmode(svga->hwtnl, svga->curr.rast->hw_fillmode);
 
-   ret = svga_upload_user_buffers( svga, min_index + index_bias,
-                                   max_index - min_index + 1, instance_count );
+   ret = svga_update_state( svga, SVGA_STATE_HW_DRAW );
    if (ret != PIPE_OK)
       goto retry;
 
-   ret = svga_update_state( svga, SVGA_STATE_HW_DRAW );
-   if (ret)
-      goto retry;
+   /** determine if flatshade is to be used after svga_update_state()
+    *  in case the fragment shader is changed.
+    */
+   svga_hwtnl_set_flatshade(svga->hwtnl,
+                            svga->curr.rast->templ.flatshade ||
+                            svga->state.hw_draw.fs->uses_flat_interp,
+                            svga->curr.rast->templ.flatshade_first);
 
    ret = svga_hwtnl_draw_range_elements( svga->hwtnl,
                                          index_buffer, index_size, index_bias,
                                          min_index, max_index,
-                                         prim, start, count );
-   if (ret)
+                                         prim, start, count,
+                                         start_instance, instance_count);
+   if (ret != PIPE_OK)
       goto retry;
 
-   return PIPE_OK;
+   goto done;
 
 retry:
    svga_context_flush( svga, NULL );
 
    if (do_retry)
    {
-      return retry_draw_range_elements( svga,
-                                        index_buffer, index_size, index_bias,
-                                        min_index, max_index,
-                                        prim, start, count,
-                                        instance_count, FALSE );
+      ret = retry_draw_range_elements(svga,
+                                      index_buffer, index_size, index_bias,
+                                      min_index, max_index,
+                                      prim, start, count,
+                                      start_instance, instance_count, FALSE);
    }
 
+done:
+   SVGA_STATS_TIME_POP(svga_sws(svga));
    return ret;
 }
 
 
 static enum pipe_error
 retry_draw_arrays( struct svga_context *svga,
-                   unsigned prim, 
-                   unsigned start, 
-                   unsigned count,
-                   unsigned instance_count,
+                   enum pipe_prim_type prim, unsigned start, unsigned count,
+                   unsigned start_instance, unsigned instance_count,
                    boolean do_retry )
 {
    enum pipe_error ret;
 
-   svga_hwtnl_set_unfilled( svga->hwtnl,
-                            svga->curr.rast->hw_unfilled );
-
-   svga_hwtnl_set_flatshade( svga->hwtnl,
-                             svga->curr.rast->templ.flatshade,
-                             svga->curr.rast->templ.flatshade_first );
+   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_DRAWARRAYS);
 
-   ret = svga_upload_user_buffers( svga, start, count, instance_count );
+   svga_hwtnl_set_fillmode(svga->hwtnl, svga->curr.rast->hw_fillmode);
 
+   ret = svga_update_state( svga, SVGA_STATE_HW_DRAW );
    if (ret != PIPE_OK)
       goto retry;
 
-   ret = svga_update_state( svga, SVGA_STATE_HW_DRAW );
-   if (ret)
-      goto retry;
+   /** determine if flatshade is to be used after svga_update_state()
+    *  in case the fragment shader is changed.
+    */
+   svga_hwtnl_set_flatshade(svga->hwtnl,
+                            svga->curr.rast->templ.flatshade ||
+                            svga->state.hw_draw.fs->uses_flat_interp,
+                            svga->curr.rast->templ.flatshade_first);
 
-   ret = svga_hwtnl_draw_arrays( svga->hwtnl, prim,
-                                 start, count );
-   if (ret)
+   ret = svga_hwtnl_draw_arrays(svga->hwtnl, prim, start, count,
+                                start_instance, instance_count);
+   if (ret != PIPE_OK)
       goto retry;
 
-   return 0;
+   goto done;
 
 retry:
-   if (ret == PIPE_ERROR_OUT_OF_MEMORY && do_retry) 
+   if (ret == PIPE_ERROR_OUT_OF_MEMORY && do_retry)
    {
       svga_context_flush( svga, NULL );
 
-      return retry_draw_arrays( svga,
-                                prim,
-                                start,
-                                count,
-                                instance_count,
-                                FALSE );
+      ret = retry_draw_arrays(svga, prim, start, count,
+                              start_instance, instance_count,
+                              FALSE);
    }
 
+done:
+   SVGA_STATS_TIME_POP(svga_sws(svga));
    return ret;
 }
 
 
+/**
+ * Determine if we need to implement primitive restart with a fallback
+ * path which breaks the original primitive into sub-primitive at the
+ * restart indexes.
+ */
+static boolean
+need_fallback_prim_restart(const struct svga_context *svga,
+                           const struct pipe_draw_info *info)
+{
+   if (info->primitive_restart && info->indexed) {
+      if (!svga_have_vgpu10(svga))
+         return TRUE;
+      else if (!svga->state.sw.need_swtnl) {
+         if (svga->curr.ib.index_size == 1)
+            return TRUE; /* no device support for 1-byte indexes */
+         else if (svga->curr.ib.index_size == 2)
+            return info->restart_index != 0xffff;
+         else
+            return info->restart_index != 0xffffffff;
+      }
+   }
+
+   return FALSE;
+}
+
+
 static void
 svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
 {
@@ -328,8 +187,21 @@ svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
    enum pipe_error ret = 0;
    boolean needed_swtnl;
 
-   if (!u_trim_pipe_prim( info->mode, &count ))
+   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_DRAWVBO);
+
+   svga->hud.num_draw_calls++;  /* for SVGA_QUERY_NUM_DRAW_CALLS */
+
+   if (u_reduced_prim(info->mode) == PIPE_PRIM_TRIANGLES &&
+       svga->curr.rast->templ.cull_face == PIPE_FACE_FRONT_AND_BACK)
+      goto done;
+
+   /* Upload a user index buffer. */
+   struct pipe_index_buffer ibuffer_saved = {};
+   if (info->indexed && svga->curr.ib.user_buffer &&
+       !util_save_and_upload_index_buffer(pipe, info, &svga->curr.ib,
+                                          &ibuffer_saved)) {
       return;
+   }
 
    /*
     * Mark currently bound target surfaces as dirty
@@ -344,18 +216,24 @@ svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
       svga->curr.reduced_prim = reduced_prim;
       svga->dirty |= SVGA_NEW_REDUCED_PRIMITIVE;
    }
-   
+
+   if (need_fallback_prim_restart(svga, info)) {
+      enum pipe_error r;
+      r = util_draw_vbo_without_prim_restart(pipe, &svga->curr.ib, info);
+      assert(r == PIPE_OK);
+      (void) r;
+      goto done;
+   }
+
+   if (!u_trim_pipe_prim( info->mode, &count ))
+      goto done;
+
    needed_swtnl = svga->state.sw.need_swtnl;
 
    svga_update_state_retry( svga, SVGA_STATE_NEED_SWTNL );
 
-#ifdef DEBUG
-   if (svga->curr.vs->base.id == svga->debug.disable_shader ||
-       svga->curr.fs->base.id == svga->debug.disable_shader)
-      return;
-#endif
-
    if (svga->state.sw.need_swtnl) {
+      svga->hud.num_fallbacks++;  /* for SVGA_QUERY_NUM_FALLBACKS */
       if (!needed_swtnl) {
          /*
           * We're switching from HW to SW TNL.  SW TNL will require mapping all
@@ -387,26 +265,31 @@ svga_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
                                           info->max_index,
                                           info->mode,
                                           info->start + offset,
-                                          info->count,
+                                          count,
+                                          info->start_instance,
                                           info->instance_count,
                                           TRUE );
       }
       else {
-         ret = retry_draw_arrays( svga,
-                                  info->mode,
-                                  info->start,
-                                  info->count,
-                                  info->instance_count,
-                                  TRUE );
+         ret = retry_draw_arrays(svga, info->mode, info->start, count,
+                                 info->start_instance, info->instance_count,
+                                 TRUE);
       }
    }
 
-   svga_release_user_upl_buffers( svga );
+   /* XXX: Silence warnings, do something sensible here? */
+   (void)ret;
 
    if (SVGA_DEBUG & DEBUG_FLUSH) {
       svga_hwtnl_flush_retry( svga );
       svga_context_flush(svga, NULL);
    }
+
+done:
+   if (info->indexed && ibuffer_saved.user_buffer)
+      pipe->set_index_buffer(pipe, &ibuffer_saved);
+
+   SVGA_STATS_TIME_POP(svga_sws(svga));
 }