X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fsvga%2Fsvga_state_fs.c;h=74c19926bc99259cc64080219db551f225bca4b8;hb=e9cf8c1d3010e9eada5a7921239ad7895f249b5c;hp=ec2886348b9c24aaa9e8f8b7131bf8679ef94c1a;hpb=7b5ad23c7f7f9016f725cb1caa3cf8971aeedbc8;p=mesa.git

diff --git a/src/gallium/drivers/svga/svga_state_fs.c b/src/gallium/drivers/svga/svga_state_fs.c
index ec2886348b9..74c19926bc9 100644
--- a/src/gallium/drivers/svga/svga_state_fs.c
+++ b/src/gallium/drivers/svga/svga_state_fs.c
@@ -23,153 +23,154 @@
  *
  **********************************************************/
 
-#include "pipe/p_inlines.h"
+#include "util/u_inlines.h"
 #include "pipe/p_defines.h"
+#include "util/u_format.h"
 #include "util/u_math.h"
+#include "util/u_memory.h"
 #include "util/u_bitmask.h"
+#include "tgsi/tgsi_ureg.h"
 
 #include "svga_context.h"
 #include "svga_state.h"
 #include "svga_cmd.h"
+#include "svga_shader.h"
+#include "svga_resource_texture.h"
 #include "svga_tgsi.h"
+#include "svga_format.h"
 
 #include "svga_hw_reg.h"
 
 
 
-static INLINE int compare_fs_keys( const struct svga_fs_compile_key *a,
-                                   const struct svga_fs_compile_key *b )
+/**
+ * If we fail to compile a fragment shader (because it uses too many
+ * registers, for example) we'll use a dummy/fallback shader that
+ * simply emits a constant color (red for debug, black for release).
+ * We hit this with the Unigine/Heaven demo when Shaders = High.
+ * With black, the demo still looks good.
+ */
+static const struct tgsi_token *
+get_dummy_fragment_shader(void)
 {
-   unsigned keysize_a = svga_fs_key_size( a );
-   unsigned keysize_b = svga_fs_key_size( b );
+#ifdef DEBUG
+   static const float color[4] = { 1.0, 0.0, 0.0, 0.0 }; /* red */
+#else
+   static const float color[4] = { 0.0, 0.0, 0.0, 0.0 }; /* black */
+#endif
+   struct ureg_program *ureg;
+   const struct tgsi_token *tokens;
+   struct ureg_src src;
+   struct ureg_dst dst;
+
+   ureg = ureg_create(PIPE_SHADER_FRAGMENT);
+   if (!ureg)
+      return NULL;
+
+   dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
+   src = ureg_DECL_immediate(ureg, color, 4);
+   ureg_MOV(ureg, dst, src);
+   ureg_END(ureg);
+
+   tokens = ureg_get_tokens(ureg, NULL);
+
+   ureg_destroy(ureg);
+
+   return tokens;
+}
+
 
-   if (keysize_a != keysize_b) {
-      return (int)(keysize_a - keysize_b);
+static struct svga_shader_variant *
+translate_fragment_program(struct svga_context *svga,
+                           const struct svga_fragment_shader *fs,
+                           const struct svga_compile_key *key)
+{
+   if (svga_have_vgpu10(svga)) {
+      return svga_tgsi_vgpu10_translate(svga, &fs->base, key,
+                                        PIPE_SHADER_FRAGMENT);
+   }
+   else {
+      return svga_tgsi_vgpu9_translate(svga, &fs->base, key,
+                                       PIPE_SHADER_FRAGMENT);
    }
-   return memcmp( a, b, keysize_a );
 }
 
 
-static struct svga_shader_result *search_fs_key( struct svga_fragment_shader *fs,
-                                                 const struct svga_fs_compile_key *key )
+/**
+ * Replace the given shader's instruction with a simple constant-color
+ * shader.  We use this when normal shader translation fails.
+ */
+static struct svga_shader_variant *
+get_compiled_dummy_shader(struct svga_context *svga,
+                          struct svga_fragment_shader *fs,
+                          const struct svga_compile_key *key)
 {
-   struct svga_shader_result *result = fs->base.results;
+   const struct tgsi_token *dummy = get_dummy_fragment_shader();
+   struct svga_shader_variant *variant;
 
-   assert(key);
-
-   for ( ; result; result = result->next) {
-      if (compare_fs_keys( key, &result->key.fkey ) == 0)
-         return result;
+   if (!dummy) {
+      return NULL;
    }
-   
-   return NULL;
+
+   FREE((void *) fs->base.tokens);
+   fs->base.tokens = dummy;
+
+   tgsi_scan_shader(fs->base.tokens, &fs->base.info);
+   fs->generic_inputs = svga_get_generic_inputs_mask(&fs->base.info);
+   svga_remap_generics(fs->generic_inputs, fs->generic_remap_table);
+
+   variant = translate_fragment_program(svga, fs, key);
+   return variant;
 }
 
 
-static enum pipe_error compile_fs( struct svga_context *svga,
-                                   struct svga_fragment_shader *fs,
-                                   const struct svga_fs_compile_key *key,
-                                   struct svga_shader_result **out_result )
+/**
+ * Translate TGSI shader into an svga shader variant.
+ */
+static enum pipe_error
+compile_fs(struct svga_context *svga,
+           struct svga_fragment_shader *fs,
+           const struct svga_compile_key *key,
+           struct svga_shader_variant **out_variant)
 {
-   struct svga_shader_result *result;
+   struct svga_shader_variant *variant;
    enum pipe_error ret = PIPE_ERROR;
 
-   result = svga_translate_fragment_program( fs, key );
-   if (result == NULL) {
-      ret = PIPE_ERROR_OUT_OF_MEMORY;
-      goto fail;
+   variant = translate_fragment_program(svga, fs, key);
+   if (variant == NULL) {
+      debug_printf("Failed to compile fragment shader,"
+                   " using dummy shader instead.\n");
+      variant = get_compiled_dummy_shader(svga, fs, key);
    }
-
-   result->id = util_bitmask_add(svga->fs_bm);
-   if(result->id == UTIL_BITMASK_INVALID_INDEX) {
-      ret = PIPE_ERROR_OUT_OF_MEMORY;
-      goto fail;
+   else if (svga_shader_too_large(svga, variant)) {
+      /* too big, use dummy shader */
+      debug_printf("Shader too large (%u bytes),"
+                   " using dummy shader instead.\n",
+                   (unsigned) (variant->nr_tokens
+                               * sizeof(variant->tokens[0])));
+      /* Free the too-large variant */
+      svga_destroy_shader_variant(svga, variant);
+      /* Use simple pass-through shader instead */
+      variant = get_compiled_dummy_shader(svga, fs, key);
    }
 
-   ret = SVGA3D_DefineShader(svga->swc, 
-                             result->id,
-                             SVGA3D_SHADERTYPE_PS,
-                             result->tokens, 
-                             result->nr_tokens * sizeof result->tokens[0]);
-   if (ret)
-      goto fail;
-
-   *out_result = result;
-   result->next = fs->base.results;
-   fs->base.results = result;
-   return PIPE_OK;
+   if (!variant) {
+      return PIPE_ERROR;
+   }
 
-fail:
-   if (result) {
-      if (result->id != UTIL_BITMASK_INVALID_INDEX)
-         util_bitmask_clear( svga->fs_bm, result->id );
-      svga_destroy_shader_result( result );
+   ret = svga_define_shader(svga, variant);
+   if (ret != PIPE_OK) {
+      svga_destroy_shader_variant(svga, variant);
+      return ret;
    }
-   return ret;
-}
 
-/* The blend workaround for simulating logicop xor behaviour requires
- * that the incoming fragment color be white.  This change achieves
- * that by hooking up a hard-wired fragment shader that just emits
- * color 1,1,1,1
- *   
- * This is a slightly incomplete solution as it assumes that the
- * actual bound shader has no other effects beyond generating a
- * fragment color.  In particular shaders containing TEXKIL and/or
- * depth-write will not have the correct behaviour, nor will those
- * expecting to use alphatest.
- *   
- * These are avoidable issues, but they are not much worse than the
- * unavoidable ones associated with this technique, so it's not clear
- * how much effort should be expended trying to resolve them - the
- * ultimate result will still not be correct in most cases.
- *
- * Shader below was generated with:
- *   SVGA_DEBUG=tgsi ./mesa/progs/fp/fp-tri white.txt
- */
-static int emit_white_fs( struct svga_context *svga )
-{
-   int ret = PIPE_ERROR;
+   *out_variant = variant;
 
-   /* ps_3_0
-    * def c0, 1.000000, 0.000000, 0.000000, 1.000000
-    * mov oC0, c0.x
-    * end
-    */
-   static const unsigned white_tokens[] = {
-      0xffff0300,
-      0x05000051,
-      0xa00f0000,
-      0x3f800000,
-      0x00000000,
-      0x00000000,
-      0x3f800000,
-      0x02000001,
-      0x800f0800,
-      0xa0000000,
-      0x0000ffff,
-   };
-
-   assert(SVGA3D_INVALID_ID == UTIL_BITMASK_INVALID_INDEX);
-   svga->state.white_fs_id = util_bitmask_add(svga->fs_bm);
-   if(svga->state.white_fs_id == SVGA3D_INVALID_ID)
-      goto no_fs_id;
-
-   ret = SVGA3D_DefineShader(svga->swc, 
-                             svga->state.white_fs_id,
-                             SVGA3D_SHADERTYPE_PS,
-                             white_tokens, 
-                             sizeof(white_tokens));
-   if (ret)
-      goto no_definition;
-
-   return 0;
-
-no_definition:
-   util_bitmask_clear(svga->fs_bm, svga->state.white_fs_id);
-   svga->state.white_fs_id = SVGA3D_INVALID_ID;
-no_fs_id:
-   return ret;
+   /* insert variant at head of linked list */
+   variant->next = fs->base.variants;
+   fs->base.variants = variant;
+
+   return PIPE_OK;
 }
 
 
@@ -178,124 +179,310 @@ no_fs_id:
  * SVGA_NEW_NEED_SWTNL
  * SVGA_NEW_SAMPLER
  */
-static int make_fs_key( const struct svga_context *svga,
-                        struct svga_fs_compile_key *key )
+static enum pipe_error
+make_fs_key(const struct svga_context *svga,
+            struct svga_fragment_shader *fs,
+            struct svga_compile_key *key)
 {
-   int i;
-   int idx = 0;
+   const enum pipe_shader_type shader = PIPE_SHADER_FRAGMENT;
+   unsigned i;
 
    memset(key, 0, sizeof *key);
 
+   memcpy(key->generic_remap_table, fs->generic_remap_table,
+          sizeof(fs->generic_remap_table));
+
+   /* SVGA_NEW_GS, SVGA_NEW_VS
+    */
+   if (svga->curr.gs) {
+      key->fs.gs_generic_outputs = svga->curr.gs->generic_outputs;
+   } else {
+      key->fs.vs_generic_outputs = svga->curr.vs->generic_outputs;
+   }
+
    /* Only need fragment shader fixup for twoside lighting if doing
     * hwtnl.  Otherwise the draw module does the whole job for us.
     *
     * SVGA_NEW_SWTNL
     */
    if (!svga->state.sw.need_swtnl) {
-      /* SVGA_NEW_RAST
+      /* SVGA_NEW_RAST, SVGA_NEW_REDUCED_PRIMITIVE
        */
-      key->light_twoside = svga->curr.rast->templ.light_twoside;
-      key->front_cw = (svga->curr.rast->templ.front_winding == 
-                       PIPE_WINDING_CW);
+      key->fs.light_twoside = svga->curr.rast->templ.light_twoside;
+      key->fs.front_ccw = svga->curr.rast->templ.front_ccw;
+      key->fs.pstipple = (svga->curr.rast->templ.poly_stipple_enable &&
+                          svga->curr.reduced_prim == PIPE_PRIM_TRIANGLES);
+      key->fs.aa_point = (svga->curr.rast->templ.point_smooth &&
+                          svga->curr.reduced_prim == PIPE_PRIM_POINTS &&
+                          (svga->curr.rast->pointsize > 1.0 ||
+                           svga->curr.vs->base.info.writes_psize));
+      if (key->fs.aa_point) {
+         assert(svga->curr.gs != NULL);
+         assert(svga->curr.gs->aa_point_coord_index != -1);
+         key->fs.aa_point_coord_index = svga->curr.gs->aa_point_coord_index;
+      }
+   }
+
+   /* The blend workaround for simulating logicop xor behaviour
+    * requires that the incoming fragment color be white.  This change
+    * achieves that by creating a variant of the current fragment
+    * shader that overrides all output colors with 1,1,1,1
+    *   
+    * This will work for most shaders, including those containing
+    * TEXKIL and/or depth-write.  However, it will break on the
+    * combination of xor-logicop plus alphatest.
+    *
+    * Ultimately, we could implement alphatest in the shader using
+    * texkil prior to overriding the outgoing fragment color.
+    *   
+    * SVGA_NEW_BLEND
+    */
+   key->fs.white_fragments = svga->curr.blend->need_white_fragments;
+
+   key->fs.alpha_to_one = svga->curr.blend->alpha_to_one;
+
+#ifdef DEBUG
+   /*
+    * We expect a consistent set of samplers and sampler views.
+    * Do some debug checks/warnings here.
+    */
+   {
+      static boolean warned = FALSE;
+      unsigned i, n = MAX2(svga->curr.num_sampler_views[shader],
+                           svga->curr.num_samplers[shader]);
+      /* Only warn once to prevent too much debug output */
+      if (!warned) {
+         if (svga->curr.num_sampler_views[shader] !=
+             svga->curr.num_samplers[shader]) {
+            debug_printf("svga: mismatched number of sampler views (%u) "
+                         "vs. samplers (%u)\n",
+                         svga->curr.num_sampler_views[shader],
+                         svga->curr.num_samplers[shader]);
+         }
+         for (i = 0; i < n; i++) {
+            if ((svga->curr.sampler_views[shader][i] == NULL) !=
+                (svga->curr.sampler[shader][i] == NULL))
+               debug_printf("sampler_view[%u] = %p but sampler[%u] = %p\n",
+                            i, svga->curr.sampler_views[shader][i],
+                            i, svga->curr.sampler[shader][i]);
+         }
+         warned = TRUE;
+      }
    }
+#endif
 
-   
    /* XXX: want to limit this to the textures that the shader actually
     * refers to.
     *
     * SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
     */
-   for (i = 0; i < svga->curr.num_textures; i++) {
-      if (svga->curr.texture[i]) {
-         assert(svga->curr.sampler[i]);
-         key->tex[i].texture_target = svga->curr.texture[i]->target;
-         if (!svga->curr.sampler[i]->normalized_coords) {
-            key->tex[i].width_height_idx = idx++;
-            key->tex[i].unnormalized = TRUE;
-            ++key->num_unnormalized_coords;
+   svga_init_shader_key_common(svga, shader, key);
+
+   for (i = 0; i < svga->curr.num_samplers[shader]; ++i) {
+      struct pipe_sampler_view *view = svga->curr.sampler_views[shader][i];
+      const struct svga_sampler_state *sampler = svga->curr.sampler[shader][i];
+      if (view) {
+         struct pipe_resource *tex = view->texture;
+         if (tex->target != PIPE_BUFFER) {
+            struct svga_texture *stex = svga_texture(tex);
+            SVGA3dSurfaceFormat format = stex->key.format;
+
+            if (!svga_have_vgpu10(svga) &&
+                (format == SVGA3D_Z_D16 ||
+                 format == SVGA3D_Z_D24X8 ||
+                 format == SVGA3D_Z_D24S8)) {
+               /* If we're sampling from a SVGA3D_Z_D16, SVGA3D_Z_D24X8,
+                * or SVGA3D_Z_D24S8 surface, we'll automatically get
+                * shadow comparison.  But we only get LEQUAL mode.
+                * Set TEX_COMPARE_NONE here so we don't emit the extra FS
+                * code for shadow comparison.
+                */
+               key->tex[i].compare_mode = PIPE_TEX_COMPARE_NONE;
+               key->tex[i].compare_func = PIPE_FUNC_NEVER;
+               /* These depth formats _only_ support comparison mode and
+                * not ordinary sampling so warn if the later is expected.
+                */
+               if (sampler->compare_mode != PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+                  debug_warn_once("Unsupported shadow compare mode");
+               }
+               /* The shader translation code can emit code to
+                * handle ALWAYS and NEVER compare functions
+                */
+               else if (sampler->compare_func == PIPE_FUNC_ALWAYS ||
+                        sampler->compare_func == PIPE_FUNC_NEVER) {
+                  key->tex[i].compare_mode = sampler->compare_mode;
+                  key->tex[i].compare_func = sampler->compare_func;
+               }
+               else if (sampler->compare_func != PIPE_FUNC_LEQUAL) {
+                  debug_warn_once("Unsupported shadow compare function");
+               }
+            }
+            else {
+               /* For other texture formats, just use the compare func/mode
+                * as-is.  Should be no-ops for color textures.  For depth
+                * textures, we do not get automatic depth compare.  We have
+                * to do it ourselves in the shader.  And we don't get PCF.
+                */
+               key->tex[i].compare_mode = sampler->compare_mode;
+               key->tex[i].compare_func = sampler->compare_func;
+            }
          }
       }
    }
-   key->num_textures = svga->curr.num_textures;
 
-   idx = 0;
-   for (i = 0; i < svga->curr.num_samplers; ++i) {
-      if (svga->curr.sampler[i]) {
-         key->tex[i].compare_mode = svga->curr.sampler[i]->compare_mode;
-         key->tex[i].compare_func = svga->curr.sampler[i]->compare_func;
+   /* sprite coord gen state */
+   key->sprite_coord_enable = svga->curr.rast->templ.sprite_coord_enable;
+
+   key->sprite_origin_lower_left = (svga->curr.rast->templ.sprite_coord_mode
+                                    == PIPE_SPRITE_COORD_LOWER_LEFT);
+
+   key->fs.flatshade = svga->curr.rast->templ.flatshade;
+
+   /* SVGA_NEW_DEPTH_STENCIL_ALPHA */
+   if (svga_have_vgpu10(svga)) {
+      /* Alpha testing is not supported in integer-valued render targets. */
+      if (svga_has_any_integer_cbufs(svga)) {
+         key->fs.alpha_func = SVGA3D_CMP_ALWAYS;
+         key->fs.alpha_ref = 0;
+      }
+      else {
+         key->fs.alpha_func = svga->curr.depth->alphafunc;
+         key->fs.alpha_ref = svga->curr.depth->alpharef;
       }
    }
 
-   return 0;
-}
+   /* SVGA_NEW_FRAME_BUFFER | SVGA_NEW_BLEND */
+   if (fs->base.info.properties[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS] ||
+       svga->curr.blend->need_white_fragments) {
+      /* Replicate color0 output (or white) to N colorbuffers */
+      key->fs.write_color0_to_n_cbufs = svga->curr.framebuffer.nr_cbufs;
+   }
 
+   return PIPE_OK;
+}
 
 
-static int emit_hw_fs( struct svga_context *svga,
-                       unsigned dirty )
+/**
+ * svga_reemit_fs_bindings - Reemit the fragment shader bindings
+ */
+enum pipe_error
+svga_reemit_fs_bindings(struct svga_context *svga)
 {
-   struct svga_shader_result *result = NULL;
-   unsigned id = SVGA3D_INVALID_ID;
-   int ret = 0;
+   enum pipe_error ret;
 
-   /* SVGA_NEW_BLEND
-    */
-   if (svga->curr.blend->need_white_fragments) {
-      if (svga->state.white_fs_id == SVGA3D_INVALID_ID) {
-         ret = emit_white_fs( svga );
-         if (ret)
-            return ret;
-      }
-      id = svga->state.white_fs_id;
+   assert(svga->rebind.flags.fs);
+   assert(svga_have_gb_objects(svga));
+
+   if (!svga->state.hw_draw.fs)
+      return PIPE_OK;
+
+   if (!svga_need_to_rebind_resources(svga)) {
+      ret =  svga->swc->resource_rebind(svga->swc, NULL,
+                                        svga->state.hw_draw.fs->gb_shader,
+                                        SVGA_RELOC_READ);
    }
    else {
-      struct svga_fragment_shader *fs = svga->curr.fs;
-      struct svga_fs_compile_key key;
+      if (svga_have_vgpu10(svga))
+         ret = SVGA3D_vgpu10_SetShader(svga->swc, SVGA3D_SHADERTYPE_PS,
+                                       svga->state.hw_draw.fs->gb_shader,
+                                       svga->state.hw_draw.fs->id);
+      else
+         ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_PS,
+                                  svga->state.hw_draw.fs->gb_shader);
+   }
 
-      /* SVGA_NEW_TEXTURE_BINDING
-       * SVGA_NEW_RAST
-       * SVGA_NEW_NEED_SWTNL
-       * SVGA_NEW_SAMPLER
-       */
-      ret = make_fs_key( svga, &key );
-      if (ret)
-         return ret;
-
-      result = search_fs_key( fs, &key );
-      if (!result) {
-         ret = compile_fs( svga, fs, &key, &result );
-         if (ret)
-            return ret;
+   if (ret != PIPE_OK)
+      return ret;
+
+   svga->rebind.flags.fs = FALSE;
+   return PIPE_OK;
+}
+
+
+
+static enum pipe_error
+emit_hw_fs(struct svga_context *svga, unsigned dirty)
+{
+   struct svga_shader_variant *variant = NULL;
+   enum pipe_error ret = PIPE_OK;
+   struct svga_fragment_shader *fs = svga->curr.fs;
+   struct svga_compile_key key;
+
+   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_EMITFS);
+
+   /* Disable rasterization if rasterizer_discard flag is set or
+    * vs/gs does not output position.
+    */
+   svga->disable_rasterizer =
+      svga->curr.rast->templ.rasterizer_discard ||
+      (svga->curr.gs && !svga->curr.gs->base.info.writes_position) ||
+      (!svga->curr.gs && !svga->curr.vs->base.info.writes_position);
+
+   /* Set FS to NULL when rasterization is to be disabled */
+   if (svga->disable_rasterizer) {
+      /* Set FS to NULL if it has not been done */
+      if (svga->state.hw_draw.fs) {
+         ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_PS, NULL);
+         if (ret != PIPE_OK)
+            goto done;
       }
+      svga->rebind.flags.fs = FALSE;
+      svga->state.hw_draw.fs = NULL;
+      goto done;
+   }
 
-      assert (result);
-      id = result->id;
+   /* SVGA_NEW_BLEND
+    * SVGA_NEW_TEXTURE_BINDING
+    * SVGA_NEW_RAST
+    * SVGA_NEW_NEED_SWTNL
+    * SVGA_NEW_SAMPLER
+    * SVGA_NEW_FRAME_BUFFER
+    * SVGA_NEW_DEPTH_STENCIL_ALPHA
+    * SVGA_NEW_VS
+    */
+   ret = make_fs_key(svga, fs, &key);
+   if (ret != PIPE_OK)
+      goto done;
+
+   variant = svga_search_shader_key(&fs->base, &key);
+   if (!variant) {
+      ret = compile_fs(svga, fs, &key, &variant);
+      if (ret != PIPE_OK)
+         goto done;
    }
 
-   assert(id != SVGA3D_INVALID_ID);
+   assert(variant);
 
-   if (result != svga->state.hw_draw.fs) {
-      ret = SVGA3D_SetShader(svga->swc,
-                             SVGA3D_SHADERTYPE_PS,
-                             id );
-      if (ret)
-         return ret;
+   if (variant != svga->state.hw_draw.fs) {
+      ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_PS, variant);
+      if (ret != PIPE_OK)
+         goto done;
 
-      svga->dirty |= SVGA_NEW_FS_RESULT;
-      svga->state.hw_draw.fs = result;      
+      svga->rebind.flags.fs = FALSE;
+
+      svga->dirty |= SVGA_NEW_FS_VARIANT;
+      svga->state.hw_draw.fs = variant;
    }
 
-   return 0;
+done:
+   SVGA_STATS_TIME_POP(svga_sws(svga));
+   return ret;
 }
 
 struct svga_tracked_state svga_hw_fs = 
 {
    "fragment shader (hwtnl)",
    (SVGA_NEW_FS |
+    SVGA_NEW_GS |
+    SVGA_NEW_VS |
     SVGA_NEW_TEXTURE_BINDING |
     SVGA_NEW_NEED_SWTNL |
     SVGA_NEW_RAST |
+    SVGA_NEW_STIPPLE |
+    SVGA_NEW_REDUCED_PRIMITIVE |
     SVGA_NEW_SAMPLER |
+    SVGA_NEW_FRAME_BUFFER |
+    SVGA_NEW_DEPTH_STENCIL_ALPHA |
     SVGA_NEW_BLEND),
    emit_hw_fs
 };