i965: Enable L3 caching of buffer surfaces.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_sampler_state.c
index 7ccf14640f8a8b3928559a7272aed84eff54b159..0fe0853420363beca9597a660f8be2c3bfe11bf4 100644 (file)
@@ -70,7 +70,82 @@ gen7_emit_sampler_state_pointers_xs(struct brw_context *brw,
    ADVANCE_BATCH();
 }
 
-uint32_t
+/**
+ * Emit a SAMPLER_STATE structure, given all the fields.
+ */
+void
+brw_emit_sampler_state(struct brw_context *brw,
+                       uint32_t *ss,
+                       uint32_t batch_offset_for_sampler_state,
+                       unsigned min_filter,
+                       unsigned mag_filter,
+                       unsigned mip_filter,
+                       unsigned max_anisotropy,
+                       unsigned address_rounding,
+                       unsigned wrap_s,
+                       unsigned wrap_t,
+                       unsigned wrap_r,
+                       unsigned min_lod,
+                       unsigned max_lod,
+                       int lod_bias,
+                       unsigned base_level,
+                       unsigned shadow_function,
+                       bool non_normalized_coordinates,
+                       uint32_t border_color_offset)
+{
+   ss[0] = BRW_SAMPLER_LOD_PRECLAMP_ENABLE |
+           SET_FIELD(base_level, BRW_SAMPLER_BASE_MIPLEVEL) |
+           SET_FIELD(mip_filter, BRW_SAMPLER_MIP_FILTER) |
+           SET_FIELD(mag_filter, BRW_SAMPLER_MAG_FILTER) |
+           SET_FIELD(min_filter, BRW_SAMPLER_MIN_FILTER);
+
+   ss[2] = border_color_offset;
+   if (brw->gen < 6) {
+      ss[2] += brw->batch.bo->offset64; /* reloc */
+      drm_intel_bo_emit_reloc(brw->batch.bo,
+                              batch_offset_for_sampler_state + 8,
+                              brw->batch.bo, border_color_offset,
+                              I915_GEM_DOMAIN_SAMPLER, 0);
+   }
+
+   ss[3] = SET_FIELD(max_anisotropy, BRW_SAMPLER_MAX_ANISOTROPY) |
+           SET_FIELD(address_rounding, BRW_SAMPLER_ADDRESS_ROUNDING);
+
+   if (brw->gen >= 7) {
+      ss[0] |= SET_FIELD(lod_bias & 0x1fff, GEN7_SAMPLER_LOD_BIAS);
+
+      if (min_filter == BRW_MAPFILTER_ANISOTROPIC)
+         ss[0] |= GEN7_SAMPLER_EWA_ANISOTROPIC_ALGORITHM;
+
+      ss[1] = SET_FIELD(min_lod, GEN7_SAMPLER_MIN_LOD) |
+              SET_FIELD(max_lod, GEN7_SAMPLER_MAX_LOD) |
+              SET_FIELD(shadow_function, GEN7_SAMPLER_SHADOW_FUNCTION);
+
+      ss[3] |= SET_FIELD(wrap_s, BRW_SAMPLER_TCX_WRAP_MODE) |
+               SET_FIELD(wrap_t, BRW_SAMPLER_TCY_WRAP_MODE) |
+               SET_FIELD(wrap_r, BRW_SAMPLER_TCZ_WRAP_MODE);
+
+      if (non_normalized_coordinates)
+         ss[3] |= GEN7_SAMPLER_NON_NORMALIZED_COORDINATES;
+   } else {
+      ss[0] |= SET_FIELD(lod_bias & 0x7ff, GEN4_SAMPLER_LOD_BIAS) |
+               SET_FIELD(shadow_function, GEN4_SAMPLER_SHADOW_FUNCTION);
+
+      if (brw->gen == 6 && min_filter != mag_filter)
+         ss[0] |= GEN6_SAMPLER_MIN_MAG_NOT_EQUAL;
+
+      ss[1] = SET_FIELD(min_lod, GEN4_SAMPLER_MIN_LOD) |
+              SET_FIELD(max_lod, GEN4_SAMPLER_MAX_LOD) |
+              SET_FIELD(wrap_s, BRW_SAMPLER_TCX_WRAP_MODE) |
+              SET_FIELD(wrap_t, BRW_SAMPLER_TCY_WRAP_MODE) |
+              SET_FIELD(wrap_r, BRW_SAMPLER_TCZ_WRAP_MODE);
+
+      if (brw->gen >= 6 && non_normalized_coordinates)
+         ss[3] |= GEN6_SAMPLER_NON_NORMALIZED_COORDINATES;
+   }
+}
+
+static uint32_t
 translate_wrap_mode(struct brw_context *brw, GLenum wrap, bool using_nearest)
 {
    switch( wrap ) {
@@ -110,12 +185,22 @@ translate_wrap_mode(struct brw_context *brw, GLenum wrap, bool using_nearest)
    }
 }
 
+/**
+ * Return true if the given wrap mode requires the border color to exist.
+ */
+static bool
+wrap_mode_needs_border_color(unsigned wrap_mode)
+{
+   return wrap_mode == BRW_TEXCOORDMODE_CLAMP_BORDER ||
+          wrap_mode == GEN8_TEXCOORDMODE_HALF_BORDER;
+}
+
 /**
  * Upload SAMPLER_BORDER_COLOR_STATE.
  */
-void
+static void
 upload_default_color(struct brw_context *brw,
-                     struct gl_sampler_object *sampler,
+                     const struct gl_sampler_object *sampler,
                      int unit,
                      uint32_t *sdc_offset)
 {
@@ -236,100 +321,102 @@ upload_default_color(struct brw_context *brw,
 static void
 brw_update_sampler_state(struct brw_context *brw,
                          int unit,
-                         struct brw_sampler_state *sampler,
+                         uint32_t *sampler_state,
                          uint32_t batch_offset_for_sampler_state)
 {
    struct gl_context *ctx = &brw->ctx;
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *texObj = texUnit->_Current;
-   struct gl_sampler_object *gl_sampler = _mesa_get_samplerobj(ctx, unit);
-   bool using_nearest = false;
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   const struct gl_texture_object *texObj = texUnit->_Current;
+   const struct gl_sampler_object *sampler = _mesa_get_samplerobj(ctx, unit);
 
    /* These don't use samplers at all. */
    if (texObj->Target == GL_TEXTURE_BUFFER)
       return;
 
-   switch (gl_sampler->MinFilter) {
+   unsigned min_filter, mag_filter, mip_filter;
+
+   /* Select min and mip filters. */
+   switch (sampler->MinFilter) {
    case GL_NEAREST:
-      sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
-      sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
-      using_nearest = true;
+      min_filter = BRW_MAPFILTER_NEAREST;
+      mip_filter = BRW_MIPFILTER_NONE;
       break;
    case GL_LINEAR:
-      sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
-      sampler->ss0.mip_filter = BRW_MIPFILTER_NONE;
+      min_filter = BRW_MAPFILTER_LINEAR;
+      mip_filter = BRW_MIPFILTER_NONE;
       break;
    case GL_NEAREST_MIPMAP_NEAREST:
-      sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
-      sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
+      min_filter = BRW_MAPFILTER_NEAREST;
+      mip_filter = BRW_MIPFILTER_NEAREST;
       break;
    case GL_LINEAR_MIPMAP_NEAREST:
-      sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
-      sampler->ss0.mip_filter = BRW_MIPFILTER_NEAREST;
+      min_filter = BRW_MAPFILTER_LINEAR;
+      mip_filter = BRW_MIPFILTER_NEAREST;
       break;
    case GL_NEAREST_MIPMAP_LINEAR:
-      sampler->ss0.min_filter = BRW_MAPFILTER_NEAREST;
-      sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
+      min_filter = BRW_MAPFILTER_NEAREST;
+      mip_filter = BRW_MIPFILTER_LINEAR;
       break;
    case GL_LINEAR_MIPMAP_LINEAR:
-      sampler->ss0.min_filter = BRW_MAPFILTER_LINEAR;
-      sampler->ss0.mip_filter = BRW_MIPFILTER_LINEAR;
+      min_filter = BRW_MAPFILTER_LINEAR;
+      mip_filter = BRW_MIPFILTER_LINEAR;
       break;
    default:
-      break;
+      unreachable("not reached");
    }
 
-   /* Set Anisotropy:
-    */
-   if (gl_sampler->MaxAnisotropy > 1.0) {
-      sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC;
-      sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
-
-      if (gl_sampler->MaxAnisotropy > 2.0) {
-        sampler->ss3.max_aniso = MIN2((gl_sampler->MaxAnisotropy - 2) / 2,
-                                      BRW_ANISORATIO_16);
-      }
-   }
-   else {
-      switch (gl_sampler->MagFilter) {
-      case GL_NEAREST:
-        sampler->ss0.mag_filter = BRW_MAPFILTER_NEAREST;
-        using_nearest = true;
-        break;
-      case GL_LINEAR:
-        sampler->ss0.mag_filter = BRW_MAPFILTER_LINEAR;
-        break;
-      default:
-        break;
+   /* Select mag filter. */
+   if (sampler->MagFilter == GL_LINEAR)
+      mag_filter = BRW_MAPFILTER_LINEAR;
+   else
+      mag_filter = BRW_MAPFILTER_NEAREST;
+
+   /* Enable anisotropic filtering if desired. */
+   unsigned max_anisotropy = BRW_ANISORATIO_2;
+   if (sampler->MaxAnisotropy > 1.0) {
+      min_filter = BRW_MAPFILTER_ANISOTROPIC;
+      mag_filter = BRW_MAPFILTER_ANISOTROPIC;
+
+      if (sampler->MaxAnisotropy > 2.0) {
+        max_anisotropy =
+            MIN2((sampler->MaxAnisotropy - 2) / 2, BRW_ANISORATIO_16);
       }
    }
 
-   sampler->ss1.r_wrap_mode = translate_wrap_mode(brw, gl_sampler->WrapR,
-                                                 using_nearest);
-   sampler->ss1.s_wrap_mode = translate_wrap_mode(brw, gl_sampler->WrapS,
-                                                 using_nearest);
-   sampler->ss1.t_wrap_mode = translate_wrap_mode(brw, gl_sampler->WrapT,
-                                                 using_nearest);
+   /* Set address rounding bits if not using nearest filtering. */
+   unsigned address_rounding = 0;
+   if (min_filter != BRW_MAPFILTER_NEAREST) {
+      address_rounding |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
+                          BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
+                          BRW_ADDRESS_ROUNDING_ENABLE_R_MIN;
+   }
+   if (mag_filter != BRW_MAPFILTER_NEAREST) {
+      address_rounding |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
+                          BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
+                          BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
+   }
 
-   if (brw->gen >= 6 &&
-       sampler->ss0.min_filter != sampler->ss0.mag_filter)
-       sampler->ss0.min_mag_neq = 1;
+   bool either_nearest =
+      sampler->MinFilter == GL_NEAREST || sampler->MagFilter == GL_NEAREST;
+   unsigned wrap_s = translate_wrap_mode(brw, sampler->WrapS, either_nearest);
+   unsigned wrap_t = translate_wrap_mode(brw, sampler->WrapT, either_nearest);
+   unsigned wrap_r = translate_wrap_mode(brw, sampler->WrapR, either_nearest);
 
-   /* 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 (texObj->Target == GL_TEXTURE_CUBE_MAP ||
        texObj->Target == GL_TEXTURE_CUBE_MAP_ARRAY) {
-      if ((ctx->Texture.CubeMapSeamless || gl_sampler->CubeMapSeamless) &&
-         (gl_sampler->MinFilter != GL_NEAREST ||
-          gl_sampler->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;
+      /* Cube maps must use the same wrap mode for all three coordinate
+       * dimensions.  Prior to Haswell, only CUBE and CLAMP are valid.
+       */
+      if ((ctx->Texture.CubeMapSeamless || sampler->CubeMapSeamless) &&
+         (sampler->MinFilter != GL_NEAREST ||
+          sampler->MagFilter != GL_NEAREST)) {
+        wrap_s = BRW_TEXCOORDMODE_CUBE;
+        wrap_t = BRW_TEXCOORDMODE_CUBE;
+        wrap_r = 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;
+        wrap_s = BRW_TEXCOORDMODE_CLAMP;
+        wrap_t = BRW_TEXCOORDMODE_CLAMP;
+        wrap_r = BRW_TEXCOORDMODE_CLAMP;
       }
    } else if (texObj->Target == GL_TEXTURE_1D) {
       /* There's a bug in 1D texture sampling - it actually pays
@@ -337,66 +424,47 @@ brw_update_sampler_state(struct brw_context *brw,
        * 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;
+      wrap_t = BRW_TEXCOORDMODE_WRAP;
    }
 
-
-   /* Set shadow function:
-    */
-   if (gl_sampler->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB) {
-      /* Shadowing is "enabled" by emitting a particular sampler
-       * message (sample_c).  So need to recompile WM program when
-       * shadow comparison is enabled on each/any texture unit.
-       */
-      sampler->ss0.shadow_function =
-        intel_translate_shadow_compare_func(gl_sampler->CompareFunc);
+   /* Set shadow function. */
+   unsigned shadow_function = 0;
+   if (sampler->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB) {
+      shadow_function =
+        intel_translate_shadow_compare_func(sampler->CompareFunc);
    }
 
-   /* Set LOD bias:
-    */
-   sampler->ss0.lod_bias = S_FIXED(CLAMP(texUnit->LodBias +
-                                        gl_sampler->LodBias, -16, 15), 6);
-
-   sampler->ss0.lod_preclamp = 1; /* OpenGL mode */
-   sampler->ss0.default_color_mode = 0; /* OpenGL/DX10 mode */
-
-   sampler->ss0.base_level = U_FIXED(0, 1);
+   const int lod_bits = brw->gen >= 7 ? 8 : 6;
+   const unsigned min_lod = U_FIXED(CLAMP(sampler->MinLod, 0, 13), lod_bits);
+   const unsigned max_lod = U_FIXED(CLAMP(sampler->MaxLod, 0, 13), lod_bits);
+   const int lod_bias =
+      S_FIXED(CLAMP(texUnit->LodBias + sampler->LodBias, -16, 15), lod_bits);
+   const unsigned base_level = U_FIXED(0, 1);
 
-   sampler->ss1.max_lod = U_FIXED(CLAMP(gl_sampler->MaxLod, 0, 13), 6);
-   sampler->ss1.min_lod = U_FIXED(CLAMP(gl_sampler->MinLod, 0, 13), 6);
-
-   /* On Gen6+, the sampler can handle non-normalized texture
-    * rectangle coordinates natively
+   /* Upload the border color if necessary.  If not, just point it at
+    * offset 0 (the start of the batch) - the color should be ignored,
+    * but that address won't fault in case something reads it anyway.
     */
-   if (brw->gen >= 6 && texObj->Target == GL_TEXTURE_RECTANGLE) {
-      sampler->ss3.non_normalized_coord = 1;
-   }
-
-   uint32_t sdc_offset;
-   upload_default_color(brw, gl_sampler, unit, &sdc_offset);
-
-   if (brw->gen >= 6) {
-      sampler->ss2.default_color_pointer = sdc_offset >> 5;
-   } else {
-      /* reloc */
-      sampler->ss2.default_color_pointer =
-         (brw->batch.bo->offset64 + sdc_offset) >> 5;
-
-      drm_intel_bo_emit_reloc(brw->batch.bo,
-                             batch_offset_for_sampler_state +
-                             offsetof(struct brw_sampler_state, ss2),
-                             brw->batch.bo, sdc_offset,
-                             I915_GEM_DOMAIN_SAMPLER, 0);
+   uint32_t border_color_offset = 0;
+   if (wrap_mode_needs_border_color(wrap_s) ||
+       wrap_mode_needs_border_color(wrap_t) ||
+       wrap_mode_needs_border_color(wrap_r)) {
+      upload_default_color(brw, sampler, unit, &border_color_offset);
    }
 
-   if (sampler->ss0.min_filter != BRW_MAPFILTER_NEAREST)
-      sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MIN |
-                                    BRW_ADDRESS_ROUNDING_ENABLE_V_MIN |
-                                    BRW_ADDRESS_ROUNDING_ENABLE_R_MIN;
-   if (sampler->ss0.mag_filter != BRW_MAPFILTER_NEAREST)
-      sampler->ss3.address_round |= BRW_ADDRESS_ROUNDING_ENABLE_U_MAG |
-                                    BRW_ADDRESS_ROUNDING_ENABLE_V_MAG |
-                                    BRW_ADDRESS_ROUNDING_ENABLE_R_MAG;
+   const bool non_normalized_coords = texObj->Target == GL_TEXTURE_RECTANGLE;
+
+   brw_emit_sampler_state(brw,
+                          sampler_state,
+                          batch_offset_for_sampler_state,
+                          min_filter, mag_filter, mip_filter,
+                          max_anisotropy,
+                          address_rounding,
+                          wrap_s, wrap_t, wrap_r,
+                          min_lod, max_lod, lod_bias, base_level,
+                          shadow_function,
+                          non_normalized_coords,
+                          border_color_offset);
 }
 
 
@@ -428,16 +496,8 @@ brw_upload_sampler_state_table(struct brw_context *brw,
       if (SamplersUsed & (1 << s)) {
          const unsigned unit = prog->SamplerUnits[s];
          if (ctx->Texture.Unit[unit]._Current) {
-            if (brw->gen >= 7) {
-               gen7_update_sampler_state(brw, unit,
-                                         (struct gen7_sampler_state *)
-                                         sampler_state);
-            } else {
-               brw_update_sampler_state(brw, unit,
-                                        (struct brw_sampler_state *)
-                                        sampler_state,
-                                        batch_offset_for_sampler_state);
-            }
+            brw_update_sampler_state(brw, unit, sampler_state,
+                                     batch_offset_for_sampler_state);
          }
       }
 
@@ -452,7 +512,7 @@ brw_upload_sampler_state_table(struct brw_context *brw,
       /* Flag that the sampler state table pointer has changed; later atoms
        * will handle it.
        */
-      brw->state.dirty.cache |= CACHE_NEW_SAMPLER;
+      brw->state.dirty.brw |= BRW_NEW_SAMPLER_STATE_TABLE;
    }
 }
 
@@ -461,7 +521,7 @@ brw_upload_fs_samplers(struct brw_context *brw)
 {
    /* BRW_NEW_FRAGMENT_PROGRAM */
    struct gl_program *fs = (struct gl_program *) brw->fragment_program;
-   brw->vtbl.upload_sampler_state_table(brw, fs, &brw->wm.base);
+   brw_upload_sampler_state_table(brw, fs, &brw->wm.base);
 }
 
 const struct brw_tracked_state brw_fs_samplers = {
@@ -469,7 +529,6 @@ const struct brw_tracked_state brw_fs_samplers = {
       .mesa = _NEW_TEXTURE,
       .brw = BRW_NEW_BATCH |
              BRW_NEW_FRAGMENT_PROGRAM,
-      .cache = 0
    },
    .emit = brw_upload_fs_samplers,
 };
@@ -479,7 +538,7 @@ brw_upload_vs_samplers(struct brw_context *brw)
 {
    /* BRW_NEW_VERTEX_PROGRAM */
    struct gl_program *vs = (struct gl_program *) brw->vertex_program;
-   brw->vtbl.upload_sampler_state_table(brw, vs, &brw->vs.base);
+   brw_upload_sampler_state_table(brw, vs, &brw->vs.base);
 }
 
 
@@ -488,7 +547,6 @@ const struct brw_tracked_state brw_vs_samplers = {
       .mesa = _NEW_TEXTURE,
       .brw = BRW_NEW_BATCH |
              BRW_NEW_VERTEX_PROGRAM,
-      .cache = 0
    },
    .emit = brw_upload_vs_samplers,
 };
@@ -502,7 +560,7 @@ brw_upload_gs_samplers(struct brw_context *brw)
    if (!gs)
       return;
 
-   brw->vtbl.upload_sampler_state_table(brw, gs, &brw->gs.base);
+   brw_upload_sampler_state_table(brw, gs, &brw->gs.base);
 }
 
 
@@ -511,14 +569,6 @@ const struct brw_tracked_state brw_gs_samplers = {
       .mesa = _NEW_TEXTURE,
       .brw = BRW_NEW_BATCH |
              BRW_NEW_GEOMETRY_PROGRAM,
-      .cache = 0
    },
    .emit = brw_upload_gs_samplers,
 };
-
-
-void
-gen4_init_vtable_sampler_functions(struct brw_context *brw)
-{
-   brw->vtbl.upload_sampler_state_table = brw_upload_sampler_state_table;
-}