X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fsvga%2Fsvga_state_need_swtnl.c;h=704866861b1ad1d1f74b1b940ae3988c339aafa3;hb=36ea744f58e5b436c45ed857bd92bbb472e641d5;hp=00201b8091d9bba23b79489e1c24e79d6fb96ccf;hpb=a3eb0f718e19653a2ad8e49396c904183be456f3;p=mesa.git

diff --git a/src/gallium/drivers/svga/svga_state_need_swtnl.c b/src/gallium/drivers/svga/svga_state_need_swtnl.c
index 00201b8091d..704866861b1 100644
--- a/src/gallium/drivers/svga/svga_state_need_swtnl.c
+++ b/src/gallium/drivers/svga/svga_state_need_swtnl.c
@@ -23,7 +23,7 @@
  *
  **********************************************************/
 
-#include "pipe/p_inlines.h"
+#include "util/u_inlines.h"
 #include "pipe/p_state.h"
 
 
@@ -35,6 +35,11 @@
 /***********************************************************************
  */
 
+
+/**
+ * Given a gallium vertex element format, return the corresponding SVGA3D
+ * format.  Return SVGA3D_DECLTYPE_MAX for unsupported gallium formats.
+ */
 static INLINE SVGA3dDeclType 
 svga_translate_vertex_format(enum pipe_format format)
 {
@@ -52,14 +57,10 @@ svga_translate_vertex_format(enum pipe_format format)
    case PIPE_FORMAT_R16G16B16A16_SNORM:   return SVGA3D_DECLTYPE_SHORT4N;
    case PIPE_FORMAT_R16G16_UNORM:         return SVGA3D_DECLTYPE_USHORT2N;
    case PIPE_FORMAT_R16G16B16A16_UNORM:   return SVGA3D_DECLTYPE_USHORT4N;
-
-   /* These formats don't exist yet:
-    * 
-   case PIPE_FORMAT_R10G10B10_USCALED:    return SVGA3D_DECLTYPE_UDEC3;
-   case PIPE_FORMAT_R10G10B10_SNORM:      return SVGA3D_DECLTYPE_DEC3N;
+   case PIPE_FORMAT_R10G10B10X2_USCALED:  return SVGA3D_DECLTYPE_UDEC3;
+   case PIPE_FORMAT_R10G10B10X2_SNORM:    return SVGA3D_DECLTYPE_DEC3N;
    case PIPE_FORMAT_R16G16_FLOAT:         return SVGA3D_DECLTYPE_FLOAT16_2;
    case PIPE_FORMAT_R16G16B16A16_FLOAT:   return SVGA3D_DECLTYPE_FLOAT16_4;
-   */
 
    default:
       /* There are many formats without hardware support.  This case
@@ -70,15 +71,22 @@ svga_translate_vertex_format(enum pipe_format format)
 }
 
 
-static int update_need_swvfetch( struct svga_context *svga,
-                                 unsigned dirty )
+static enum pipe_error
+update_need_swvfetch( struct svga_context *svga,
+                      unsigned dirty )
 {
    unsigned i;
    boolean need_swvfetch = FALSE;
 
-   for (i = 0; i < svga->curr.num_vertex_elements; i++) {
-      svga->state.sw.ve_format[i] = svga_translate_vertex_format(svga->curr.ve[i].src_format);
+   if (!svga->curr.velems) {
+      /* No vertex elements bound. */
+      return 0;
+   }
+
+   for (i = 0; i < svga->curr.velems->count; i++) {
+      svga->state.sw.ve_format[i] = svga_translate_vertex_format(svga->curr.velems->velem[i].src_format);
       if (svga->state.sw.ve_format[i] == SVGA3D_DECLTYPE_MAX) {
+         /* Unsupported format - use software fetch */
          need_swvfetch = TRUE;
          break;
       }
@@ -89,7 +97,7 @@ static int update_need_swvfetch( struct svga_context *svga,
       svga->dirty |= SVGA_NEW_NEED_SWVFETCH;
    }
    
-   return 0;
+   return PIPE_OK;
 }
 
 struct svga_tracked_state svga_update_need_swvfetch = 
@@ -103,37 +111,56 @@ struct svga_tracked_state svga_update_need_swvfetch =
 /*********************************************************************** 
  */
 
-static int update_need_pipeline( struct svga_context *svga,
-                                 unsigned dirty )
+static enum pipe_error
+update_need_pipeline( struct svga_context *svga,
+                      unsigned dirty )
 {
    
    boolean need_pipeline = FALSE;
+   struct svga_vertex_shader *vs = svga->curr.vs;
 
    /* SVGA_NEW_RAST, SVGA_NEW_REDUCED_PRIMITIVE
     */
    if (svga->curr.rast->need_pipeline & (1 << svga->curr.reduced_prim)) {
-      SVGA_DBG(DEBUG_SWTNL, "%s: rast need_pipeline (%d) & prim (%x)\n", 
+      SVGA_DBG(DEBUG_SWTNL, "%s: rast need_pipeline (0x%x) & prim (0x%x)\n",
                  __FUNCTION__,
                  svga->curr.rast->need_pipeline,
                  (1 << svga->curr.reduced_prim) );
+      SVGA_DBG(DEBUG_SWTNL, "%s: rast need_pipeline tris (%s), lines (%s), points (%s)\n",
+                 __FUNCTION__,
+                 svga->curr.rast->need_pipeline_tris_str,
+                 svga->curr.rast->need_pipeline_lines_str,
+                 svga->curr.rast->need_pipeline_points_str);
       need_pipeline = TRUE;
    }
 
-   /* SVGA_NEW_EDGEFLAGS
+   /* EDGEFLAGS
     */
-   if (svga->curr.rast->hw_unfilled != PIPE_POLYGON_MODE_FILL &&
-       svga->curr.reduced_prim == PIPE_PRIM_TRIANGLES && 
-       svga->curr.edgeflags != NULL) {
+    if (vs && vs->base.info.writes_edgeflag) {
       SVGA_DBG(DEBUG_SWTNL, "%s: edgeflags\n", __FUNCTION__);
       need_pipeline = TRUE;
    }
 
-   /* SVGA_NEW_CLIP 
+   /* SVGA_NEW_FS, SVGA_NEW_RAST, SVGA_NEW_REDUCED_PRIMITIVE
     */
-   if (!svga->curr.rast->templ.bypass_vs_clip_and_viewport &&
-       svga->curr.clip.nr) {
-      SVGA_DBG(DEBUG_SWTNL, "%s: userclip\n", __FUNCTION__);
-      need_pipeline = TRUE;
+   if (svga->curr.reduced_prim == PIPE_PRIM_POINTS) {
+      unsigned sprite_coord_gen = svga->curr.rast->templ.sprite_coord_enable;
+      unsigned generic_inputs =
+         svga->curr.fs ? svga->curr.fs->generic_inputs : 0;
+
+      if (sprite_coord_gen &&
+          (generic_inputs & ~sprite_coord_gen)) {
+         /* The fragment shader is using some generic inputs that are
+          * not being replaced by auto-generated point/sprite coords (and
+          * auto sprite coord generation is turned on).
+          * The SVGA3D interface does not support that: if we enable
+          * SVGA3D_RS_POINTSPRITEENABLE it gets enabled for _all_
+          * texture coordinate sets.
+          * To solve this, we have to use the draw-module's wide/sprite
+          * point stage.
+          */
+         need_pipeline = TRUE;
+      }
    }
 
    if (need_pipeline != svga->state.sw.need_pipeline) {
@@ -141,7 +168,11 @@ static int update_need_pipeline( struct svga_context *svga,
       svga->dirty |= SVGA_NEW_NEED_PIPELINE;
    }
 
-   return 0;
+   /* DEBUG */
+   if (0 && svga->state.sw.need_pipeline)
+      debug_printf("sw.need_pipeline = %d\n", svga->state.sw.need_pipeline);
+
+   return PIPE_OK;
 }
 
 
@@ -149,7 +180,8 @@ struct svga_tracked_state svga_update_need_pipeline =
 {
    "need pipeline",
    (SVGA_NEW_RAST |
-    SVGA_NEW_CLIP |
+    SVGA_NEW_FS |
+    SVGA_NEW_VS |
     SVGA_NEW_REDUCED_PRIMITIVE),
    update_need_pipeline
 };
@@ -158,26 +190,35 @@ struct svga_tracked_state svga_update_need_pipeline =
 /*********************************************************************** 
  */
 
-static int update_need_swtnl( struct svga_context *svga,
-                              unsigned dirty )
+static enum pipe_error
+update_need_swtnl( struct svga_context *svga,
+                   unsigned dirty )
 {
    boolean need_swtnl;
 
    if (svga->debug.no_swtnl) {
-      svga->state.sw.need_swvfetch = 0;
-      svga->state.sw.need_pipeline = 0;
+      svga->state.sw.need_swvfetch = FALSE;
+      svga->state.sw.need_pipeline = FALSE;
    }
 
    need_swtnl = (svga->state.sw.need_swvfetch ||
                  svga->state.sw.need_pipeline);
 
    if (svga->debug.force_swtnl) {
-      need_swtnl = 1;
+      need_swtnl = TRUE;
    }
 
+   /*
+    * Some state changes the draw module does makes us believe we
+    * we don't need swtnl. This causes the vdecl code to pickup
+    * the wrong buffers and vertex formats. Try trivial/line-wide.
+    */
+   if (svga->state.sw.in_swtnl_draw)
+      need_swtnl = TRUE;
+
    if (need_swtnl != svga->state.sw.need_swtnl) {
       SVGA_DBG(DEBUG_SWTNL|DEBUG_PERF,
-               "%s need_swvfetch: %s, need_pipeline %s\n",
+               "%s: need_swvfetch %s, need_pipeline %s\n",
                __FUNCTION__,
                svga->state.sw.need_swvfetch ? "true" : "false",
                svga->state.sw.need_pipeline ? "true" : "false");
@@ -187,7 +228,7 @@ static int update_need_swtnl( struct svga_context *svga,
       svga->swtnl.new_vdecl = TRUE;
    }
   
-   return 0;
+   return PIPE_OK;
 }