gallium: add PIPE_SHADER_CAP_GLSL_16BIT_TEMPS for LowerPrecisionTemporaries
[mesa.git] / src / gallium / drivers / svga / svga_state_need_swtnl.c
index 3c35a8579f714f0826f33b45472a56ad482b00c0..5a52c25a4c1ec30536f9cf134e8d0c0dce898011 100644 (file)
  *
  **********************************************************/
 
-#include "pipe/p_inlines.h"
+#include "util/u_inlines.h"
 #include "pipe/p_state.h"
-
-
 #include "svga_context.h"
+#include "svga_shader.h"
 #include "svga_state.h"
 #include "svga_debug.h"
 #include "svga_hw_reg.h"
 
-/***********************************************************************
- */
 
-static INLINE SVGA3dDeclType 
-svga_translate_vertex_format(enum pipe_format format)
+static enum pipe_error
+update_need_swvfetch(struct svga_context *svga, uint64_t dirty)
 {
-   switch (format) {
-   case PIPE_FORMAT_R32_FLOAT:            return SVGA3D_DECLTYPE_FLOAT1;
-   case PIPE_FORMAT_R32G32_FLOAT:         return SVGA3D_DECLTYPE_FLOAT2;
-   case PIPE_FORMAT_R32G32B32_FLOAT:      return SVGA3D_DECLTYPE_FLOAT3;
-   case PIPE_FORMAT_R32G32B32A32_FLOAT:   return SVGA3D_DECLTYPE_FLOAT4;
-   case PIPE_FORMAT_B8G8R8A8_UNORM:       return SVGA3D_DECLTYPE_D3DCOLOR;
-   case PIPE_FORMAT_R8G8B8A8_USCALED:     return SVGA3D_DECLTYPE_UBYTE4;
-   case PIPE_FORMAT_R16G16_SSCALED:       return SVGA3D_DECLTYPE_SHORT2;
-   case PIPE_FORMAT_R16G16B16A16_SSCALED: return SVGA3D_DECLTYPE_SHORT4;
-   case PIPE_FORMAT_R8G8B8A8_UNORM:       return SVGA3D_DECLTYPE_UBYTE4N;
-   case PIPE_FORMAT_R16G16_SNORM:         return SVGA3D_DECLTYPE_SHORT2N;
-   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_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
-       * will be hit regularly, meaning we'll need swvfetch.
-       */
-      return SVGA3D_DECLTYPE_MAX;
+   if (!svga->curr.velems) {
+      /* No vertex elements bound. */
+      return PIPE_OK;
    }
-}
-
 
-static int 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->state.sw.ve_format[i] == SVGA3D_DECLTYPE_MAX) {
-         need_swvfetch = TRUE;
-         break;
-      }
-   }
-
-   if (need_swvfetch != svga->state.sw.need_swvfetch) {
-      svga->state.sw.need_swvfetch = need_swvfetch;
+   if (svga->state.sw.need_swvfetch != svga->curr.velems->need_swvfetch) {
+      svga->state.sw.need_swvfetch = svga->curr.velems->need_swvfetch;
       svga->dirty |= SVGA_NEW_NEED_SWVFETCH;
    }
-   
-   return 0;
+
+   return PIPE_OK;
 }
 
-struct svga_tracked_state svga_update_need_swvfetch = 
+struct svga_tracked_state svga_update_need_swvfetch =
 {
    "update need_swvfetch",
    ( SVGA_NEW_VELEMENT ),
@@ -100,39 +56,73 @@ 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, uint64_t dirty)
 {
-   
    boolean need_pipeline = FALSE;
    struct svga_vertex_shader *vs = svga->curr.vs;
+   const char *reason = "";
 
    /* 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", 
+   if (svga->curr.rast &&
+       (svga->curr.rast->need_pipeline & (1 << svga->curr.reduced_prim))) {
+      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;
+
+      switch (svga->curr.reduced_prim) {
+      case PIPE_PRIM_POINTS:
+         reason = svga->curr.rast->need_pipeline_points_str;
+         break;
+      case PIPE_PRIM_LINES:
+         reason = svga->curr.rast->need_pipeline_lines_str;
+         break;
+      case PIPE_PRIM_TRIANGLES:
+         reason = svga->curr.rast->need_pipeline_tris_str;
+         break;
+      default:
+         assert(!"Unexpected reduced prim type");
+      }
    }
 
    /* EDGEFLAGS
     */
-    if (vs->base.info.writes_edgeflag) {
+    if (vs && vs->base.info.writes_edgeflag) {
       SVGA_DBG(DEBUG_SWTNL, "%s: edgeflags\n", __FUNCTION__);
       need_pipeline = TRUE;
+      reason = "edge flags";
    }
 
-   /* 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.rast && 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 (!svga_have_vgpu10(svga) && 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;
+         reason = "point sprite coordinate generation";
+      }
    }
 
    if (need_pipeline != svga->state.sw.need_pipeline) {
@@ -140,44 +130,59 @@ 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);
+
+   if (svga->state.sw.need_pipeline) {
+      assert(reason);
+      pipe_debug_message(&svga->debug.callback, FALLBACK,
+                         "Using semi-fallback for %s", reason);
+   }
+
+   return PIPE_OK;
 }
 
 
-struct svga_tracked_state svga_update_need_pipeline = 
+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
 };
 
 
-/*********************************************************************** 
- */
-
-static int update_need_swtnl( struct svga_context *svga,
-                              unsigned dirty )
+static enum pipe_error
+update_need_swtnl(struct svga_context *svga, uint64_t 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");
@@ -186,8 +191,8 @@ static int update_need_swtnl( struct svga_context *svga,
       svga->dirty |= SVGA_NEW_NEED_SWTNL;
       svga->swtnl.new_vdecl = TRUE;
    }
-  
-   return 0;
+
+   return PIPE_OK;
 }