X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fsvga%2Fsvga_state_fs.c;h=7931528c661421f97fdd48ec32f3edd36a6e0468;hb=52381a7ffba908410f7a53855f082401fca7293a;hp=6ec38ed3e455aa8d951dea636b9f69688cb6b5e9;hpb=0d4a05445c6b47b93269a3829afbe509ffec4817;p=mesa.git

diff --git a/src/gallium/drivers/svga/svga_state_fs.c b/src/gallium/drivers/svga/svga_state_fs.c
index 6ec38ed3e45..7931528c661 100644
--- a/src/gallium/drivers/svga/svga_state_fs.c
+++ b/src/gallium/drivers/svga/svga_state_fs.c
@@ -23,130 +23,171 @@
  *
  **********************************************************/
 
-#include "pipe/p_inlines.h"
+#include "util/u_inlines.h"
 #include "pipe/p_defines.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_hw_reg.h"
 
 
 
-static INLINE int compare_fs_keys( const struct svga_fs_compile_key *a,
-                                   const struct svga_fs_compile_key *b )
+static INLINE int
+compare_fs_keys(const struct svga_fs_compile_key *a,
+                const struct svga_fs_compile_key *b)
 {
-   unsigned keysize = svga_fs_key_size( a );
-   return memcmp( a, b, keysize );
+   unsigned keysize_a = svga_fs_key_size( a );
+   unsigned keysize_b = svga_fs_key_size( b );
+
+   if (keysize_a != keysize_b) {
+      return (int)(keysize_a - keysize_b);
+   }
+   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 )
+/** Search for a fragment shader variant */
+static struct svga_shader_variant *
+search_fs_key(const struct svga_fragment_shader *fs,
+              const struct svga_fs_compile_key *key)
 {
-   struct svga_shader_result *result = fs->base.results;
+   struct svga_shader_variant *variant = fs->base.variants;
 
    assert(key);
 
-   for ( ; result; result = result->next) {
-      if (compare_fs_keys( key, &result->key.fkey ) == 0)
-         return result;
+   for ( ; variant; variant = variant->next) {
+      if (compare_fs_keys( key, &variant->key.fkey ) == 0)
+         return variant;
    }
    
    return NULL;
 }
 
 
-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 )
+/**
+ * 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)
 {
-   struct svga_shader_result *result;
-   enum pipe_error ret;
+#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;
+   unsigned num_tokens;
+
+   ureg = ureg_create(TGSI_PROCESSOR_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, &num_tokens);
+
+   ureg_destroy(ureg);
+
+   return tokens;
+}
 
-   result = svga_translate_fragment_program( fs, key );
-   if (result == NULL) {
-      ret = PIPE_ERROR_OUT_OF_MEMORY;
-      goto fail;
-   }
 
+/**
+ * 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_fragment_shader *fs,
+                          const struct svga_fs_compile_key *key)
+{
+   const struct tgsi_token *dummy = get_dummy_fragment_shader();
+   struct svga_shader_variant *variant;
 
-   ret = SVGA3D_DefineShader(svga->swc, 
-                             svga->state.next_fs_id,
-                             SVGA3D_SHADERTYPE_PS,
-                             result->tokens, 
-                             result->nr_tokens * sizeof result->tokens[0]);
-   if (ret)
-      goto fail;
+   if (!dummy) {
+      return NULL;
+   }
 
-   *out_result = result;
-   result->id = svga->state.next_fs_id++;
-   result->next = fs->base.results;
-   fs->base.results = result;
-   return PIPE_OK;
+   FREE((void *) fs->base.tokens);
+   fs->base.tokens = dummy;
 
-fail:
-   if (result)
-      svga_destroy_shader_result( result );
-   return ret;
+   variant = svga_translate_fragment_program(fs, key);
+   return variant;
 }
 
-/* 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
+
+/**
+ * Translate TGSI shader into an svga shader variant.
  */
-static int emit_white_fs( struct svga_context *svga )
+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_variant **out_variant)
 {
-   int ret;
+   struct svga_shader_variant *variant;
+   enum pipe_error ret = PIPE_ERROR;
+
+   variant = svga_translate_fragment_program( fs, key );
+   if (variant == NULL) {
+      debug_printf("Failed to compile fragment shader,"
+                   " using dummy shader instead.\n");
+      variant = get_compiled_dummy_shader(fs, key);
+      if (!variant) {
+         ret = PIPE_ERROR;
+         goto fail;
+      }
+   }
 
-   /* 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,
-   };
-
-   ret = SVGA3D_DefineShader(svga->swc, 
-                             svga->state.next_fs_id,
-                             SVGA3D_SHADERTYPE_PS,
-                             white_tokens, 
-                             sizeof(white_tokens));
-   if (ret)
-      return ret;
+   if (variant->nr_tokens * sizeof(variant->tokens[0])
+       + sizeof(SVGA3dCmdDefineShader) + sizeof(SVGA3dCmdHeader)
+       >= SVGA_CB_MAX_COMMAND_SIZE) {
+      /* too big, use dummy shader */
+      debug_printf("Shader too large (%lu bytes),"
+                   " using dummy shader instead.\n",
+                   (unsigned long ) variant->nr_tokens * sizeof(variant->tokens[0]));
+      variant = get_compiled_dummy_shader(fs, key);
+      if (!variant) {
+         ret = PIPE_ERROR;
+         goto fail;
+      }
+   }
+
+   ret = svga_define_shader(svga, SVGA3D_SHADERTYPE_PS, variant);
+   if (ret != PIPE_OK)
+      goto fail;
 
-   svga->state.white_fs_id = svga->state.next_fs_id++;
-   return 0;
+   *out_variant = variant;
+
+   /* insert variants at head of linked list */
+   variant->next = fs->base.variants;
+   fs->base.variants = variant;
+
+   return PIPE_OK;
+
+fail:
+   if (variant) {
+      svga_destroy_shader_variant(svga, SVGA3D_SHADERTYPE_PS, variant);
+   }
+   return ret;
 }
 
 
@@ -155,10 +196,12 @@ static int emit_white_fs( struct svga_context *svga )
  * 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_fs_compile_key *key)
 {
-   int i;
+   unsigned i;
    int idx = 0;
 
    memset(key, 0, sizeof *key);
@@ -172,98 +215,221 @@ static int make_fs_key( const struct svga_context *svga,
       /* SVGA_NEW_RAST
        */
       key->light_twoside = svga->curr.rast->templ.light_twoside;
-      key->front_cw = (svga->curr.rast->templ.front_winding == 
-                       PIPE_WINDING_CW);
+      key->front_ccw = svga->curr.rast->templ.front_ccw;
    }
 
-   
+   /* 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
+    */
+   if (svga->curr.blend->need_white_fragments) {
+      key->white_fragments = 1;
+   }
+
+#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,
+                           svga->curr.num_samplers);
+      /* Only warn once to prevent too much debug output */
+      if (!warned) {
+         if (svga->curr.num_sampler_views != svga->curr.num_samplers) {
+            debug_printf("svga: mismatched number of sampler views (%u) "
+                         "vs. samplers (%u)\n",
+                         svga->curr.num_sampler_views,
+                         svga->curr.num_samplers);
+         }
+         for (i = 0; i < n; i++) {
+            if ((svga->curr.sampler_views[i] == NULL) !=
+                (svga->curr.sampler[i] == NULL))
+               debug_printf("sampler_view[%u] = %p but sampler[%u] = %p\n",
+                            i, svga->curr.sampler_views[i],
+                            i, svga->curr.sampler[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;
+   for (i = 0; i < svga->curr.num_sampler_views; i++) {
+      if (svga->curr.sampler_views[i] && svga->curr.sampler[i]) {
+         assert(svga->curr.sampler_views[i]->texture);
+         key->tex[i].texture_target = svga->curr.sampler_views[i]->texture->target;
          if (!svga->curr.sampler[i]->normalized_coords) {
             key->tex[i].width_height_idx = idx++;
             key->tex[i].unnormalized = TRUE;
             ++key->num_unnormalized_coords;
          }
+
+         key->tex[i].swizzle_r = svga->curr.sampler_views[i]->swizzle_r;
+         key->tex[i].swizzle_g = svga->curr.sampler_views[i]->swizzle_g;
+         key->tex[i].swizzle_b = svga->curr.sampler_views[i]->swizzle_b;
+         key->tex[i].swizzle_a = svga->curr.sampler_views[i]->swizzle_a;
       }
    }
-   key->num_textures = svga->curr.num_textures;
+   key->num_textures = svga->curr.num_sampler_views;
 
    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;
+      if (svga->curr.sampler_views[i] && svga->curr.sampler[i]) {
+         struct pipe_resource *tex = svga->curr.sampler_views[i]->texture;
+         struct svga_texture *stex = svga_texture(tex);
+         SVGA3dSurfaceFormat format = stex->key.format;
+
+         if (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 (svga->curr.sampler[i]->compare_mode !=
+                PIPE_TEX_COMPARE_R_TO_TEXTURE) {
+               debug_warn_once("Unsupported shadow compare mode");
+            }                   
+            /* The only supported comparison mode is LEQUAL */
+            if (svga->curr.sampler[i]->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 = svga->curr.sampler[i]->compare_mode;
+            key->tex[i].compare_func = svga->curr.sampler[i]->compare_func;
+         }
       }
    }
 
-   return 0;
+   /* sprite coord gen state */
+   for (i = 0; i < svga->curr.num_samplers; ++i) {
+      key->tex[i].sprite_texgen =
+         svga->curr.rast->templ.sprite_coord_enable & (1 << i);
+   }
+
+   key->sprite_origin_lower_left = (svga->curr.rast->templ.sprite_coord_mode
+                                    == PIPE_SPRITE_COORD_LOWER_LEFT);
+
+   /* SVGA_NEW_FRAME_BUFFER */
+   if (fs->base.info.color0_writes_all_cbufs) {
+      /* Replicate color0 output to N colorbuffers */
+      key->write_color0_to_n_cbufs = svga->curr.framebuffer.nr_cbufs;
+   }
+
+   return PIPE_OK;
 }
 
 
+/**
+ * svga_reemit_fs_bindings - Reemit the fragment shader bindings
+ */
+enum pipe_error
+svga_reemit_fs_bindings(struct svga_context *svga)
+{
+   enum pipe_error ret;
+
+   assert(svga->rebind.fs);
+   assert(svga_have_gb_objects(svga));
+
+   if (!svga->state.hw_draw.fs)
+      return PIPE_OK;
+
+   ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_PS,
+                            svga->state.hw_draw.fs->gb_shader);
+   if (ret != PIPE_OK)
+      return ret;
+
+   svga->rebind.fs = FALSE;
+   return PIPE_OK;
+}
 
-static int emit_hw_fs( struct svga_context *svga,
-                       unsigned dirty )
+
+
+static enum pipe_error
+emit_hw_fs(struct svga_context *svga, unsigned dirty)
 {
-   struct svga_shader_result *result = NULL;
-   unsigned id = SVGA3D_INVALID_ID;
-   int ret = 0;
+   struct svga_shader_variant *variant = NULL;
+   enum pipe_error ret = PIPE_OK;
+   struct svga_fragment_shader *fs = svga->curr.fs;
+   struct svga_fs_compile_key key;
 
    /* SVGA_NEW_BLEND
+    * SVGA_NEW_TEXTURE_BINDING
+    * SVGA_NEW_RAST
+    * SVGA_NEW_NEED_SWTNL
+    * SVGA_NEW_SAMPLER
+    * SVGA_NEW_FRAME_BUFFER
     */
-   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;
-   }
-   else {
-      struct svga_fragment_shader *fs = svga->curr.fs;
-      struct svga_fs_compile_key key;
-
-      /* SVGA_NEW_TEXTURE_BINDING
-       * SVGA_NEW_RAST
-       * SVGA_NEW_NEED_SWTNL
-       * SVGA_NEW_SAMPLER
-       */
-      ret = make_fs_key( svga, &key );
-      if (ret)
+   ret = make_fs_key( svga, fs, &key );
+   if (ret != PIPE_OK)
+      return ret;
+
+   variant = search_fs_key( fs, &key );
+   if (!variant) {
+      ret = compile_fs( svga, fs, &key, &variant );
+      if (ret != PIPE_OK)
          return ret;
+   }
 
-      result = search_fs_key( fs, &key );
-      if (!result) {
-         ret = compile_fs( svga, fs, &key, &result );
-         if (ret)
-            return ret;
-      }
+   assert(variant);
 
-      assert (result);
-      id = result->id;
-   }
+   if (variant != svga->state.hw_draw.fs) {
+      if (svga_have_gb_objects(svga)) {
+         /*
+          * Bind is necessary here only because pipebuffer_fenced may move
+          * the shader contents around....
+          */
+         ret = SVGA3D_BindGBShader(svga->swc, variant->gb_shader);
+         if (ret != PIPE_OK)
+            return ret;
 
-   assert(id != SVGA3D_INVALID_ID);
+         ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_PS,
+                                  variant->gb_shader);
+         if (ret != PIPE_OK)
+            return ret;
 
-   if (id != svga->state.hw_draw.shader_id[PIPE_SHADER_FRAGMENT]) {
-      ret = SVGA3D_SetShader(svga->swc, 
-                             SVGA3D_SHADERTYPE_PS, 
-                             id );
-      if (ret)
-         return ret;
+         svga->rebind.fs = FALSE;
+      }
+      else {
+         ret = SVGA3D_SetShader(svga->swc, SVGA3D_SHADERTYPE_PS, variant->id);
+         if (ret != PIPE_OK)
+            return ret;
+      }
 
-      svga->dirty |= SVGA_NEW_FS_RESULT;
-      svga->state.hw_draw.shader_id[PIPE_SHADER_FRAGMENT] = id;
-      svga->state.hw_draw.fs = result;      
+      svga->dirty |= SVGA_NEW_FS_VARIANT;
+      svga->state.hw_draw.fs = variant;      
    }
 
-   return 0;
+   return PIPE_OK;
 }
 
 struct svga_tracked_state svga_hw_fs = 
@@ -274,6 +440,7 @@ struct svga_tracked_state svga_hw_fs =
     SVGA_NEW_NEED_SWTNL |
     SVGA_NEW_RAST |
     SVGA_NEW_SAMPLER |
+    SVGA_NEW_FRAME_BUFFER |
     SVGA_NEW_BLEND),
    emit_hw_fs
 };