st/mesa: don't unbind sampler states if none are used
[mesa.git] / src / mesa / state_tracker / st_atom_sampler.c
index 9d63f1c6ab6583b251e8d4d6514b0311477f3421..539b28933b9f4ce6b596c323b9e25cc75be1654a 100644 (file)
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2007 VMware, Inc.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -18,7 +18,7 @@
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
  /*
   * Authors:
-  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Keith Whitwell <keithw@vmware.com>
   *   Brian Paul
   */
  
 
 #include "main/macros.h"
+#include "main/mtypes.h"
+#include "main/glformats.h"
+#include "main/samplerobj.h"
+#include "main/teximage.h"
+#include "main/texobj.h"
 
 #include "st_context.h"
 #include "st_cb_texture.h"
+#include "st_format.h"
 #include "st_atom.h"
+#include "st_texture.h"
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 
 #include "cso_cache/cso_context.h"
 
+#include "util/u_format.h"
+
 
 /**
  * Convert GLenum texcoord wrap tokens to pipe tokens.
@@ -117,138 +126,306 @@ gl_filter_to_img_filter(GLenum filter)
 }
 
 
-static void
-xlate_border_color(const GLfloat *colorIn, GLenum baseFormat, GLfloat *colorOut)
+/**
+ * Convert a gl_sampler_object to a pipe_sampler_state object.
+ */
+void
+st_convert_sampler(const struct st_context *st,
+                   const struct gl_texture_object *texobj,
+                   const struct gl_sampler_object *msamp,
+                   struct pipe_sampler_state *sampler)
 {
-   switch (baseFormat) {
-   case GL_RGB:
-      colorOut[0] = colorIn[0];
-      colorOut[1] = colorIn[1];
-      colorOut[2] = colorIn[2];
-      colorOut[3] = 1.0F;
-      break;
-   case GL_ALPHA:
-      colorOut[0] = colorOut[1] = colorOut[2] = 0.0;
-      colorOut[3] = colorIn[3];
-      break;
-   case GL_LUMINANCE:
-      colorOut[0] = colorOut[1] = colorOut[2] = colorIn[0];
-      colorOut[3] = 1.0;
-      break;
-   case GL_LUMINANCE_ALPHA:
-      colorOut[0] = colorOut[1] = colorOut[2] = colorIn[0];
-      colorOut[3] = colorIn[3];
-      break;
-   case GL_INTENSITY:
-      colorOut[0] = colorOut[1] = colorOut[2] = colorOut[3] = colorIn[0];
-      break;
-   default:
-      COPY_4V(colorOut, colorIn);
+   GLenum texBaseFormat;
+
+   texBaseFormat = _mesa_texture_base_format(texobj);
+
+   memset(sampler, 0, sizeof(*sampler));
+   sampler->wrap_s = gl_wrap_xlate(msamp->WrapS);
+   sampler->wrap_t = gl_wrap_xlate(msamp->WrapT);
+   sampler->wrap_r = gl_wrap_xlate(msamp->WrapR);
+
+   sampler->min_img_filter = gl_filter_to_img_filter(msamp->MinFilter);
+   sampler->min_mip_filter = gl_filter_to_mip_filter(msamp->MinFilter);
+   sampler->mag_img_filter = gl_filter_to_img_filter(msamp->MagFilter);
+
+   if (texobj->Target != GL_TEXTURE_RECTANGLE_ARB)
+      sampler->normalized_coords = 1;
+
+   sampler->lod_bias = msamp->LodBias;
+   /* Reduce the number of states by allowing only the values that AMD GCN
+    * can represent. Apps use lod_bias for smooth transitions to bigger mipmap
+    * levels.
+    */
+   sampler->lod_bias = CLAMP(sampler->lod_bias, -16, 16);
+   sampler->lod_bias = floorf(sampler->lod_bias * 256) / 256;
+
+   sampler->min_lod = MAX2(msamp->MinLod, 0.0f);
+   sampler->max_lod = msamp->MaxLod;
+   if (sampler->max_lod < sampler->min_lod) {
+      /* The GL spec doesn't seem to specify what to do in this case.
+       * Swap the values.
+       */
+      float tmp = sampler->max_lod;
+      sampler->max_lod = sampler->min_lod;
+      sampler->min_lod = tmp;
+      assert(sampler->min_lod <= sampler->max_lod);
    }
-}
 
+   /* For non-black borders... */
+   if (msamp->BorderColor.ui[0] ||
+       msamp->BorderColor.ui[1] ||
+       msamp->BorderColor.ui[2] ||
+       msamp->BorderColor.ui[3]) {
+      const struct st_texture_object *stobj = st_texture_object_const(texobj);
+      const GLboolean is_integer = texobj->_IsIntegerFormat;
+      const struct pipe_sampler_view *sv = NULL;
+      union pipe_color_union border_color;
+      GLuint i;
+
+      /* Just search for the first used view. We can do this because the
+         swizzle is per-texture, not per context. */
+      /* XXX: clean that up to not use the sampler view at all */
+      for (i = 0; i < stobj->num_sampler_views; ++i) {
+         if (stobj->sampler_views[i]) {
+            sv = stobj->sampler_views[i];
+            break;
+         }
+      }
 
-static void 
-update_samplers(struct st_context *st)
-{
-   struct gl_vertex_program *vprog = st->ctx->VertexProgram._Current;
-   struct gl_fragment_program *fprog = st->ctx->FragmentProgram._Current;
-   const GLbitfield samplersUsed = (vprog->Base.SamplersUsed |
-                                    fprog->Base.SamplersUsed);
-   GLuint su;
+      if (st->apply_texture_swizzle_to_border_color && sv) {
+         const unsigned char swz[4] =
+         {
+            sv->swizzle_r,
+            sv->swizzle_g,
+            sv->swizzle_b,
+            sv->swizzle_a,
+         };
+
+         st_translate_color(&msamp->BorderColor,
+                            &border_color,
+                            texBaseFormat, is_integer);
+
+         util_format_apply_color_swizzle(&sampler->border_color,
+                                         &border_color, swz, is_integer);
+      } else {
+         st_translate_color(&msamp->BorderColor,
+                            &sampler->border_color,
+                            texBaseFormat, is_integer);
+      }
+   }
 
-   st->state.num_samplers = 0;
+   sampler->max_anisotropy = (msamp->MaxAnisotropy == 1.0 ?
+                              0 : (GLuint) msamp->MaxAnisotropy);
 
-   /* loop over sampler units (aka tex image units) */
-   for (su = 0; su < st->ctx->Const.MaxTextureImageUnits; su++) {
-      struct pipe_sampler_state *sampler = st->state.samplers + su;
+   /* If sampling a depth texture and using shadow comparison */
+   if ((texBaseFormat == GL_DEPTH_COMPONENT ||
+        (texBaseFormat == GL_DEPTH_STENCIL && !texobj->StencilSampling)) &&
+       msamp->CompareMode == GL_COMPARE_R_TO_TEXTURE) {
+      sampler->compare_mode = PIPE_TEX_COMPARE_R_TO_TEXTURE;
+      sampler->compare_func = st_compare_func_to_pipe(msamp->CompareFunc);
+   }
 
-      memset(sampler, 0, sizeof(*sampler));
+   /* Only set the seamless cube map texture parameter because the per-context
+    * enable should be ignored and treated as disabled when using texture
+    * handles, as specified by ARB_bindless_texture.
+    */
+   sampler->seamless_cube_map = msamp->CubeMapSeamless;
+}
 
-      if (samplersUsed & (1 << su)) {
-         struct gl_texture_object *texobj;
-         struct gl_texture_image *teximg;
-         GLuint texUnit;
+/**
+ * Get a pipe_sampler_state object from a texture unit.
+ */
+void
+st_convert_sampler_from_unit(const struct st_context *st,
+                             struct pipe_sampler_state *sampler,
+                             GLuint texUnit)
+{
+   const struct gl_texture_object *texobj;
+   struct gl_context *ctx = st->ctx;
+   const struct gl_sampler_object *msamp;
 
-         if (fprog->Base.SamplersUsed & (1 << su))
-            texUnit = fprog->Base.SamplerUnits[su];
-         else
-            texUnit = vprog->Base.SamplerUnits[su];
+   texobj = ctx->Texture.Unit[texUnit]._Current;
+   assert(texobj);
 
-         texobj = st->ctx->Texture.Unit[texUnit]._Current;
-         if (!texobj) {
-            texobj = st_get_default_texture(st);
-         }
+   msamp = _mesa_get_samplerobj(ctx, texUnit);
 
-         teximg = texobj->Image[0][texobj->BaseLevel];
-
-         sampler->wrap_s = gl_wrap_xlate(texobj->WrapS);
-         sampler->wrap_t = gl_wrap_xlate(texobj->WrapT);
-         sampler->wrap_r = gl_wrap_xlate(texobj->WrapR);
-
-         sampler->min_img_filter = gl_filter_to_img_filter(texobj->MinFilter);
-         sampler->min_mip_filter = gl_filter_to_mip_filter(texobj->MinFilter);
-         sampler->mag_img_filter = gl_filter_to_img_filter(texobj->MagFilter);
-
-         if (texobj->Target != GL_TEXTURE_RECTANGLE_ARB)
-            sampler->normalized_coords = 1;
-
-         sampler->lod_bias = st->ctx->Texture.Unit[su].LodBias;
-         sampler->min_lod = MAX2(0.0f, texobj->MinLod);
-         sampler->max_lod = MIN2(texobj->MaxLevel - texobj->BaseLevel,
-                                 texobj->MaxLod);
-         if (sampler->max_lod < sampler->min_lod) {
-            /* The GL spec doesn't seem to specify what to do in this case.
-             * Swap the values.
-             */
-            float tmp = sampler->max_lod;
-            sampler->max_lod = sampler->min_lod;
-            sampler->min_lod = tmp;
-            assert(sampler->min_lod <= sampler->max_lod);
-         }
+   st_convert_sampler(st, texobj, msamp, sampler);
 
-         xlate_border_color(texobj->BorderColor.f,
-                            teximg ? teximg->_BaseFormat : GL_RGBA,
-                            sampler->border_color);
+   sampler->lod_bias += ctx->Texture.Unit[texUnit].LodBias;
+   sampler->seamless_cube_map |= ctx->Texture.CubeMapSeamless;
+}
 
-        sampler->max_anisotropy = texobj->MaxAnisotropy;
 
-         /* only care about ARB_shadow, not SGI shadow */
-         if (texobj->CompareMode == GL_COMPARE_R_TO_TEXTURE) {
-            sampler->compare_mode = PIPE_TEX_COMPARE_R_TO_TEXTURE;
-            sampler->compare_func
-               = st_compare_func_to_pipe(texobj->CompareFunc);
-         }
+/**
+ * Update the gallium driver's sampler state for fragment, vertex or
+ * geometry shader stage.
+ */
+static void
+update_shader_samplers(struct st_context *st,
+                       enum pipe_shader_type shader_stage,
+                       const struct gl_program *prog,
+                       unsigned max_units,
+                       struct pipe_sampler_state *samplers,
+                       unsigned *num_samplers)
+{
+   GLbitfield samplers_used = prog->SamplersUsed;
+   GLbitfield free_slots = ~prog->SamplersUsed;
+   GLbitfield external_samplers_used = prog->ExternalSamplersUsed;
+   GLuint unit;
+   const GLuint old_max = *num_samplers;
+   const struct pipe_sampler_state *states[PIPE_MAX_SAMPLERS];
 
-         st->state.num_samplers = su + 1;
+   if (samplers_used == 0x0)
+      return;
 
-         /*printf("%s su=%u non-null\n", __FUNCTION__, su);*/
-         cso_single_sampler(st->cso_context, su, sampler);
-         if (su < st->ctx->Const.MaxVertexTextureImageUnits) {
-            cso_single_vertex_sampler(st->cso_context, su, sampler);
-         }
+   *num_samplers = 0;
+
+   /* loop over sampler units (aka tex image units) */
+   for (unit = 0; unit < max_units; unit++, samplers_used >>= 1) {
+      struct pipe_sampler_state *sampler = samplers + unit;
+
+      if (samplers_used & 1) {
+         const GLuint texUnit = prog->SamplerUnits[unit];
+
+         st_convert_sampler_from_unit(st, sampler, texUnit);
+         states[unit] = sampler;
+         *num_samplers = unit + 1;
+      }
+      else if (samplers_used != 0 || unit < old_max) {
+         states[unit] = NULL;
       }
       else {
-         /*printf("%s su=%u null\n", __FUNCTION__, su);*/
-         cso_single_sampler(st->cso_context, su, NULL);
-         if (su < st->ctx->Const.MaxVertexTextureImageUnits) {
-            cso_single_vertex_sampler(st->cso_context, su, NULL);
-         }
+         /* if we've reset all the old samplers and we have no more new ones */
+         break;
+      }
+   }
+
+   /* For any external samplers with multiplaner YUV, stuff the additional
+    * sampler states we need at the end.
+    *
+    * Just re-use the existing sampler-state from the primary slot.
+    */
+   while (unlikely(external_samplers_used)) {
+      GLuint unit = u_bit_scan(&external_samplers_used);
+      GLuint extra = 0;
+      struct st_texture_object *stObj =
+            st_get_texture_object(st->ctx, prog, unit);
+      struct pipe_sampler_state *sampler = samplers + unit;
+
+      if (!stObj)
+         continue;
+
+      switch (st_get_view_format(stObj)) {
+      case PIPE_FORMAT_NV12:
+         /* we need one additional sampler: */
+         extra = u_bit_scan(&free_slots);
+         states[extra] = sampler;
+         break;
+      case PIPE_FORMAT_IYUV:
+         /* we need two additional samplers: */
+         extra = u_bit_scan(&free_slots);
+         states[extra] = sampler;
+         extra = u_bit_scan(&free_slots);
+         states[extra] = sampler;
+         break;
+      default:
+         break;
       }
+
+      *num_samplers = MAX2(*num_samplers, extra + 1);
+   }
+
+   cso_set_samplers(st->cso_context, shader_stage, *num_samplers, states);
+}
+
+
+void
+st_update_vertex_samplers(struct st_context *st)
+{
+   const struct gl_context *ctx = st->ctx;
+
+   update_shader_samplers(st,
+                          PIPE_SHADER_VERTEX,
+                          ctx->VertexProgram._Current,
+                          ctx->Const.Program[MESA_SHADER_VERTEX].MaxTextureImageUnits,
+                          st->state.samplers[PIPE_SHADER_VERTEX],
+                          &st->state.num_samplers[PIPE_SHADER_VERTEX]);
+}
+
+
+void
+st_update_tessctrl_samplers(struct st_context *st)
+{
+   const struct gl_context *ctx = st->ctx;
+
+   if (ctx->TessCtrlProgram._Current) {
+      update_shader_samplers(st,
+                             PIPE_SHADER_TESS_CTRL,
+                             ctx->TessCtrlProgram._Current,
+                             ctx->Const.Program[MESA_SHADER_TESS_CTRL].MaxTextureImageUnits,
+                             st->state.samplers[PIPE_SHADER_TESS_CTRL],
+                             &st->state.num_samplers[PIPE_SHADER_TESS_CTRL]);
    }
+}
+
+
+void
+st_update_tesseval_samplers(struct st_context *st)
+{
+   const struct gl_context *ctx = st->ctx;
+
+   if (ctx->TessEvalProgram._Current) {
+      update_shader_samplers(st,
+                             PIPE_SHADER_TESS_EVAL,
+                             ctx->TessEvalProgram._Current,
+                             ctx->Const.Program[MESA_SHADER_TESS_EVAL].MaxTextureImageUnits,
+                             st->state.samplers[PIPE_SHADER_TESS_EVAL],
+                             &st->state.num_samplers[PIPE_SHADER_TESS_EVAL]);
+   }
+}
 
-   cso_single_sampler_done(st->cso_context);
-   if (st->ctx->Const.MaxVertexTextureImageUnits > 0) {
-      cso_single_vertex_sampler_done(st->cso_context);
+
+void
+st_update_geometry_samplers(struct st_context *st)
+{
+   const struct gl_context *ctx = st->ctx;
+
+   if (ctx->GeometryProgram._Current) {
+      update_shader_samplers(st,
+                             PIPE_SHADER_GEOMETRY,
+                             ctx->GeometryProgram._Current,
+                             ctx->Const.Program[MESA_SHADER_GEOMETRY].MaxTextureImageUnits,
+                             st->state.samplers[PIPE_SHADER_GEOMETRY],
+                             &st->state.num_samplers[PIPE_SHADER_GEOMETRY]);
    }
 }
 
 
-const struct st_tracked_state st_update_sampler = {
-   "st_update_sampler",                                        /* name */
-   {                                                   /* dirty */
-      _NEW_TEXTURE,                                    /* mesa */
-      0,                                               /* st */
-   },
-   update_samplers                                     /* update */
-};
+void
+st_update_fragment_samplers(struct st_context *st)
+{
+   const struct gl_context *ctx = st->ctx;
+
+   update_shader_samplers(st,
+                          PIPE_SHADER_FRAGMENT,
+                          ctx->FragmentProgram._Current,
+                          ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits,
+                          st->state.samplers[PIPE_SHADER_FRAGMENT],
+                          &st->state.num_samplers[PIPE_SHADER_FRAGMENT]);
+}
+
+
+void
+st_update_compute_samplers(struct st_context *st)
+{
+   const struct gl_context *ctx = st->ctx;
+
+   if (ctx->ComputeProgram._Current) {
+      update_shader_samplers(st,
+                             PIPE_SHADER_COMPUTE,
+                             ctx->ComputeProgram._Current,
+                             ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits,
+                             st->state.samplers[PIPE_SHADER_COMPUTE],
+                             &st->state.num_samplers[PIPE_SHADER_COMPUTE]);
+   }
+}