i965: Also do constant propagation for the second operand of CMP.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_wm_sampler_state.c
index 2e0aff7ab2820ba954a0da5b03faafa74dd419dd..f9c48140fb6253b7b8a2bd3d3933b34bd5ed8a0f 100644 (file)
@@ -34,7 +34,7 @@
 #include "brw_state.h"
 #include "brw_defines.h"
 
-#include "macros.h"
+#include "main/macros.h"
 
 
 
@@ -66,28 +66,15 @@ static GLuint translate_wrap_mode( GLenum wrap )
    }
 }
 
-
-static GLuint U_FIXED(GLfloat value, GLuint frac_bits)
-{
-   value *= (1<<frac_bits);
-   return value < 0 ? 0 : value;
-}
-
-static GLint S_FIXED(GLfloat value, GLuint frac_bits)
-{
-   return value * (1<<frac_bits);
-}
-
-
-static dri_bo *upload_default_color( struct brw_context *brw,
+static drm_intel_bo *upload_default_color( struct brw_context *brw,
                                     const GLfloat *color )
 {
    struct brw_sampler_default_color sdc;
 
    COPY_4V(sdc.color, color); 
    
-   return brw_cache_data( &brw->cache, BRW_SAMPLER_DEFAULT_COLOR, &sdc,
-                         NULL, 0 );
+   return brw_cache_data(&brw->cache, BRW_SAMPLER_DEFAULT_COLOR,
+                        &sdc, sizeof(sdc));
 }
 
 
@@ -95,13 +82,17 @@ struct wm_sampler_key {
    int sampler_count;
 
    struct wm_sampler_entry {
+      GLenum tex_target;
       GLenum wrap_r, wrap_s, wrap_t;
       float maxlod, minlod;
       float lod_bias;
       float max_aniso;
       GLenum minfilter, magfilter;
       GLenum comparemode, comparefunc;
-      dri_bo *sdc_bo;
+
+      /** If target is cubemap, take context setting.
+       */
+      GLboolean seamless_cube_map;
    } sampler[BRW_MAX_TEX_UNIT];
 };
 
@@ -109,11 +100,14 @@ struct wm_sampler_key {
  * Sets the sampler state for a single unit based off of the sampler key
  * entry.
  */
-static void brw_update_sampler_state(struct wm_sampler_entry *key,
-                                    dri_bo *sdc_bo,
+static void brw_update_sampler_state(struct brw_context *brw,
+                                    struct wm_sampler_entry *key,
+                                    drm_intel_bo *sdc_bo,
                                     struct brw_sampler_state *sampler)
 {
-   _mesa_memset(sampler, 0, sizeof(*sampler));
+   struct intel_context *intel = &brw->intel;
+
+   memset(sampler, 0, sizeof(*sampler));
 
    switch (key->minfilter) {
    case GL_NEAREST:
@@ -151,7 +145,7 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key,
       sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
 
       if (key->max_aniso > 2.0) {
-        sampler->ss3.max_aniso = MAX2((key->max_aniso - 2) / 2,
+        sampler->ss3.max_aniso = MIN2((key->max_aniso - 2) / 2,
                                       BRW_ANISORATIO_16);
       }
    }
@@ -172,15 +166,33 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key,
    sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s);
    sampler->ss1.t_wrap_mode = translate_wrap_mode(key->wrap_t);
 
-   /* Fulsim complains if I don't do this.  Hardware doesn't mind:
+   if (intel->gen >= 6 &&
+       sampler->ss0.min_filter != sampler->ss0.mag_filter)
+       sampler->ss0.min_mag_neq = 1;
+
+   /* Cube-maps on 965 and later must use the same wrap mode for all 3
+    * coordinate dimensions.  Futher, only CUBE and CLAMP are valid.
     */
-#if 0
-   if (texObj->Target == GL_TEXTURE_CUBE_MAP_ARB) {
-      sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
-      sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
-      sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
+   if (key->tex_target == GL_TEXTURE_CUBE_MAP) {
+      if (key->seamless_cube_map &&
+         (key->minfilter != GL_NEAREST || key->magfilter != GL_NEAREST)) {
+        sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CUBE;
+        sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE;
+        sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE;
+      } else {
+        sampler->ss1.r_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
+        sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
+        sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CLAMP;
+      }
+   } else if (key->tex_target == GL_TEXTURE_1D) {
+      /* There's a bug in 1D texture sampling - it actually pays
+       * attention to the wrap_t value, though it should not.
+       * Override the wrap_t value here to GL_REPEAT to keep
+       * any nonexistent border pixels from floating in.
+       */
+      sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP;
    }
-#endif
+
 
    /* Set shadow function: 
     */
@@ -209,26 +221,42 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key,
     */
    sampler->ss0.base_level = U_FIXED(0, 1);
 
-   sampler->ss1.max_lod = U_FIXED(MIN2(MAX2(key->maxlod, 0), 13), 6);
-   sampler->ss1.min_lod = U_FIXED(MIN2(MAX2(key->minlod, 0), 13), 6);
+   sampler->ss1.max_lod = U_FIXED(CLAMP(key->maxlod, 0, 13), 6);
+   sampler->ss1.min_lod = U_FIXED(CLAMP(key->minlod, 0, 13), 6);
    
    sampler->ss2.default_color_pointer = sdc_bo->offset >> 5; /* reloc */
 }
 
+
 /** Sets up the cache key for sampler state for all texture units */
 static void
 brw_wm_sampler_populate_key(struct brw_context *brw,
                            struct wm_sampler_key *key)
 {
+   GLcontext *ctx = &brw->intel.ctx;
    int unit;
+   char *last_entry_end = ((char*)&key->sampler_count) + 
+      sizeof(key->sampler_count);
 
-   memset(key, 0, sizeof(*key));
+   key->sampler_count = 0;
 
    for (unit = 0; unit < BRW_MAX_TEX_UNIT; unit++) {
-      if (brw->attribs.Texture->Unit[unit]._ReallyEnabled) {
+      if (ctx->Texture.Unit[unit]._ReallyEnabled) {
         struct wm_sampler_entry *entry = &key->sampler[unit];
-        struct gl_texture_unit *texUnit = &brw->attribs.Texture->Unit[unit];
+        struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
         struct gl_texture_object *texObj = texUnit->_Current;
+        struct intel_texture_object *intelObj = intel_texture_object(texObj);
+        struct gl_texture_image *firstImage =
+           texObj->Image[0][intelObj->firstLevel];
+
+        memset(last_entry_end, 0, 
+               (char*)entry - last_entry_end + sizeof(*entry));
+        last_entry_end = ((char*)entry) + sizeof(*entry);
+
+         entry->tex_target = texObj->Target;
+
+        entry->seamless_cube_map = (texObj->Target == GL_TEXTURE_CUBE_MAP)
+           ? ctx->Texture.CubeMapSeamless : GL_FALSE;
 
         entry->wrap_r = texObj->WrapR;
         entry->wrap_s = texObj->WrapS;
@@ -241,25 +269,41 @@ brw_wm_sampler_populate_key(struct brw_context *brw,
         entry->minfilter = texObj->MinFilter;
         entry->magfilter = texObj->MagFilter;
         entry->comparemode = texObj->CompareMode;
-    entry->comparefunc = texObj->CompareFunc;
-
-        dri_bo_unreference(brw->wm.sdc_bo[unit]);
-        brw->wm.sdc_bo[unit] = upload_default_color(brw, texObj->BorderColor);
-
+         entry->comparefunc = texObj->CompareFunc;
+
+        drm_intel_bo_unreference(brw->wm.sdc_bo[unit]);
+        if (firstImage->_BaseFormat == GL_DEPTH_COMPONENT) {
+           float bordercolor[4] = {
+              texObj->BorderColor.f[0],
+              texObj->BorderColor.f[0],
+              texObj->BorderColor.f[0],
+              texObj->BorderColor.f[0]
+           };
+           /* GL specs that border color for depth textures is taken from the
+            * R channel, while the hardware uses A.  Spam R into all the
+            * channels for safety.
+            */
+           brw->wm.sdc_bo[unit] = upload_default_color(brw, bordercolor);
+        } else {
+           brw->wm.sdc_bo[unit] = upload_default_color(brw,
+                                                       texObj->BorderColor.f);
+        }
         key->sampler_count = unit + 1;
       }
    }
+   struct wm_sampler_entry *entry = &key->sampler[key->sampler_count];
+   memset(last_entry_end, 0, (char*)entry - last_entry_end);
 }
 
 /* All samplers must be uploaded in a single contiguous array, which
  * complicates various things.  However, this is still too confusing -
  * FIXME: simplify all the different new texture state flags.
  */
-static int upload_wm_samplers( struct brw_context *brw )
+static void upload_wm_samplers( struct brw_context *brw )
 {
+   GLcontext *ctx = &brw->intel.ctx;
    struct wm_sampler_key key;
-   int i;
-   int ret = 0;
+   int i, sampler_key_size;
 
    brw_wm_sampler_populate_key(brw, &key);
 
@@ -268,13 +312,16 @@ static int upload_wm_samplers( struct brw_context *brw )
       brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
    }
 
-   dri_bo_unreference(brw->wm.sampler_bo);
+   drm_intel_bo_unreference(brw->wm.sampler_bo);
    brw->wm.sampler_bo = NULL;
    if (brw->wm.sampler_count == 0)
-      return 0;
+      return;
 
+   /* Only include the populated portion of the key in the search. */
+   sampler_key_size = offsetof(struct wm_sampler_key,
+                              sampler[key.sampler_count]);
    brw->wm.sampler_bo = brw_search_cache(&brw->cache, BRW_SAMPLER,
-                                        &key, sizeof(key),
+                                        &key, sampler_key_size,
                                         brw->wm.sdc_bo, key.sampler_count,
                                         NULL);
 
@@ -289,34 +336,27 @@ static int upload_wm_samplers( struct brw_context *brw )
         if (brw->wm.sdc_bo[i] == NULL)
            continue;
 
-        brw_update_sampler_state(&key.sampler[i], brw->wm.sdc_bo[i],
+        brw_update_sampler_state(brw, &key.sampler[i], brw->wm.sdc_bo[i],
                                  &sampler[i]);
       }
 
       brw->wm.sampler_bo = brw_upload_cache(&brw->cache, BRW_SAMPLER,
-                                           &key, sizeof(key),
+                                           &key, sampler_key_size,
                                            brw->wm.sdc_bo, key.sampler_count,
-                                           &sampler, sizeof(sampler),
-                                           NULL, NULL);
+                                           &sampler, sizeof(sampler));
 
       /* Emit SDC relocations */
       for (i = 0; i < BRW_MAX_TEX_UNIT; i++) {
-        if (!brw->attribs.Texture->Unit[i]._ReallyEnabled)
+        if (!ctx->Texture.Unit[i]._ReallyEnabled)
            continue;
 
-        ret |= dri_bufmgr_check_aperture_space(brw->wm.sdc_bo[i]);
-        intel_bo_emit_reloc(brw->wm.sampler_bo,
-                            DRM_GEM_DOMAIN_I915_INSTRUCTION, 0,
-                            0,
-                            i * sizeof(struct brw_sampler_state) +
-                            offsetof(struct brw_sampler_state, ss2),
-                            brw->wm.sdc_bo[i]);
+        drm_intel_bo_emit_reloc(brw->wm.sampler_bo,
+                                i * sizeof(struct brw_sampler_state) +
+                                offsetof(struct brw_sampler_state, ss2),
+                                brw->wm.sdc_bo[i], 0,
+                                I915_GEM_DOMAIN_SAMPLER, 0);
       }
    }
-
-   ret |= dri_bufmgr_check_aperture_space(brw->wm.sampler_bo);
-   return ret;
-
 }
 
 const struct brw_tracked_state brw_wm_samplers = {