i965/gs: Add a case to brwNewProgram() for geometry shaders.
[mesa.git] / src / mesa / drivers / dri / i965 / brw_tex_layout.c
index e7b5046c06548365abb8820e83cd9c9aa7d053c6..e4e66b4219c2128fa343ad6d4f38987f33b32135 100644 (file)
  */
 
 #include "intel_mipmap_tree.h"
-#include "intel_tex_layout.h"
-#include "intel_context.h"
+#include "brw_context.h"
 #include "main/macros.h"
 
 #define FILE_DEBUG_FLAG DEBUG_MIPTREE
 
+static unsigned int
+intel_horizontal_texture_alignment_unit(struct brw_context *brw,
+                                       gl_format format)
+{
+   /**
+    * From the "Alignment Unit Size" section of various specs, namely:
+    * - Gen3 Spec: "Memory Data Formats" Volume,         Section 1.20.1.4
+    * - i965 and G45 PRMs:             Volume 1,         Section 6.17.3.4.
+    * - Ironlake and Sandybridge PRMs: Volume 1, Part 1, Section 7.18.3.4
+    * - BSpec (for Ivybridge and slight variations in separate stencil)
+    *
+    * +----------------------------------------------------------------------+
+    * |                                        | alignment unit width  ("i") |
+    * | Surface Property                       |-----------------------------|
+    * |                                        | 915 | 965 | ILK | SNB | IVB |
+    * +----------------------------------------------------------------------+
+    * | YUV 4:2:2 format                       |  8  |  4  |  4  |  4  |  4  |
+    * | BC1-5 compressed format (DXTn/S3TC)    |  4  |  4  |  4  |  4  |  4  |
+    * | FXT1  compressed format                |  8  |  8  |  8  |  8  |  8  |
+    * | Depth Buffer (16-bit)                  |  4  |  4  |  4  |  4  |  8  |
+    * | Depth Buffer (other)                   |  4  |  4  |  4  |  4  |  4  |
+    * | Separate Stencil Buffer                | N/A | N/A |  8  |  8  |  8  |
+    * | All Others                             |  4  |  4  |  4  |  4  |  4  |
+    * +----------------------------------------------------------------------+
+    *
+    * On IVB+, non-special cases can be overridden by setting the SURFACE_STATE
+    * "Surface Horizontal Alignment" field to HALIGN_4 or HALIGN_8.
+    */
+    if (_mesa_is_format_compressed(format)) {
+       /* The hardware alignment requirements for compressed textures
+        * happen to match the block boundaries.
+        */
+      unsigned int i, j;
+      _mesa_get_format_block_size(format, &i, &j);
+      return i;
+    }
+
+   if (format == MESA_FORMAT_S8)
+      return 8;
+
+   if (brw->gen >= 7 && format == MESA_FORMAT_Z16)
+      return 8;
+
+   return 4;
+}
+
+static unsigned int
+intel_vertical_texture_alignment_unit(struct brw_context *brw,
+                                     gl_format format)
+{
+   /**
+    * From the "Alignment Unit Size" section of various specs, namely:
+    * - Gen3 Spec: "Memory Data Formats" Volume,         Section 1.20.1.4
+    * - i965 and G45 PRMs:             Volume 1,         Section 6.17.3.4.
+    * - Ironlake and Sandybridge PRMs: Volume 1, Part 1, Section 7.18.3.4
+    * - BSpec (for Ivybridge and slight variations in separate stencil)
+    *
+    * +----------------------------------------------------------------------+
+    * |                                        | alignment unit height ("j") |
+    * | Surface Property                       |-----------------------------|
+    * |                                        | 915 | 965 | ILK | SNB | IVB |
+    * +----------------------------------------------------------------------+
+    * | BC1-5 compressed format (DXTn/S3TC)    |  4  |  4  |  4  |  4  |  4  |
+    * | FXT1  compressed format                |  4  |  4  |  4  |  4  |  4  |
+    * | Depth Buffer                           |  2  |  2  |  2  |  4  |  4  |
+    * | Separate Stencil Buffer                | N/A | N/A | N/A |  4  |  8  |
+    * | Multisampled (4x or 8x) render target  | N/A | N/A | N/A |  4  |  4  |
+    * | All Others                             |  2  |  2  |  2  |  2  |  2  |
+    * +----------------------------------------------------------------------+
+    *
+    * On SNB+, non-special cases can be overridden by setting the SURFACE_STATE
+    * "Surface Vertical Alignment" field to VALIGN_2 or VALIGN_4.
+    *
+    * We currently don't support multisampling.
+    */
+   if (_mesa_is_format_compressed(format))
+      return 4;
+
+   if (format == MESA_FORMAT_S8)
+      return brw->gen >= 7 ? 8 : 4;
+
+   GLenum base_format = _mesa_get_format_base_format(format);
+
+   if (brw->gen >= 6 &&
+       (base_format == GL_DEPTH_COMPONENT ||
+       base_format == GL_DEPTH_STENCIL)) {
+      return 4;
+   }
+
+   return 2;
+}
+
 static void
 brw_miptree_layout_2d(struct intel_mipmap_tree *mt)
 {
-   GLuint level;
-   GLuint x = 0;
-   GLuint y = 0;
-   GLuint width = mt->physical_width0;
-   GLuint height = mt->physical_height0;
-   GLuint depth = mt->physical_depth0; /* number of array layers. */
+   unsigned x = 0;
+   unsigned y = 0;
+   unsigned width = mt->physical_width0;
+   unsigned height = mt->physical_height0;
+   unsigned depth = mt->physical_depth0; /* number of array layers. */
 
    mt->total_width = mt->physical_width0;
 
@@ -61,7 +151,7 @@ brw_miptree_layout_2d(struct intel_mipmap_tree *mt)
     * 2nd mipmap out past the width of its parent.
     */
    if (mt->first_level != mt->last_level) {
-       GLuint mip1_width;
+       unsigned mip1_width;
 
        if (mt->compressed) {
           mip1_width = ALIGN(minify(mt->physical_width0, 1), mt->align_w) +
@@ -78,8 +168,8 @@ brw_miptree_layout_2d(struct intel_mipmap_tree *mt)
 
    mt->total_height = 0;
 
-   for ( level = mt->first_level ; level <= mt->last_level ; level++ ) {
-      GLuint img_height;
+   for (unsigned level = mt->first_level; level <= mt->last_level; level++) {
+      unsigned img_height;
 
       intel_miptree_set_level_info(mt, level, x, y, width,
                                   height, depth);
@@ -107,26 +197,25 @@ brw_miptree_layout_2d(struct intel_mipmap_tree *mt)
 }
 
 static void
-brw_miptree_layout_texture_array(struct intel_context *intel,
+brw_miptree_layout_texture_array(struct brw_context *brw,
                                 struct intel_mipmap_tree *mt)
 {
-   GLuint level;
-   GLuint qpitch = 0;
-   int h0, h1, q;
+   unsigned qpitch = 0;
+   int h0, h1;
 
    h0 = ALIGN(mt->physical_height0, mt->align_h);
    h1 = ALIGN(minify(mt->physical_height0, 1), mt->align_h);
    if (mt->array_spacing_lod0)
       qpitch = h0;
    else
-      qpitch = (h0 + h1 + (intel->gen >= 7 ? 12 : 11) * mt->align_h);
+      qpitch = (h0 + h1 + (brw->gen >= 7 ? 12 : 11) * mt->align_h);
    if (mt->compressed)
       qpitch /= 4;
 
    brw_miptree_layout_2d(mt);
 
-   for (level = mt->first_level; level <= mt->last_level; level++) {
-      for (q = 0; q < mt->physical_depth0; q++) {
+   for (unsigned level = mt->first_level; level <= mt->last_level; level++) {
+      for (int q = 0; q < mt->physical_depth0; q++) {
         intel_miptree_set_image_offset(mt, level, q, 0, q * qpitch);
       }
    }
@@ -134,15 +223,14 @@ brw_miptree_layout_texture_array(struct intel_context *intel,
 }
 
 static void
-brw_miptree_layout_texture_3d(struct intel_context *intel,
+brw_miptree_layout_texture_3d(struct brw_context *brw,
                               struct intel_mipmap_tree *mt)
 {
-   GLuint width  = mt->physical_width0;
-   GLuint height = mt->physical_height0;
-   GLuint depth = mt->physical_depth0;
-   GLuint pack_x_pitch, pack_x_nr;
-   GLuint pack_y_pitch;
-   GLuint level;
+   unsigned width  = mt->physical_width0;
+   unsigned height = mt->physical_height0;
+   unsigned depth = mt->physical_depth0;
+   unsigned pack_x_pitch, pack_x_nr;
+   unsigned pack_y_pitch;
 
    mt->total_height = 0;
 
@@ -157,17 +245,16 @@ brw_miptree_layout_texture_3d(struct intel_context *intel,
    pack_x_pitch = width;
    pack_x_nr = 1;
 
-   for (level = mt->first_level ; level <= mt->last_level ; level++) {
-      GLint x = 0;
-      GLint y = 0;
-      GLint q, j;
+   for (unsigned level = mt->first_level; level <= mt->last_level; level++) {
+      int x = 0;
+      int y = 0;
 
       intel_miptree_set_level_info(mt, level,
                                    0, mt->total_height,
                                    width, height, depth);
 
-      for (q = 0; q < depth; /* empty */) {
-         for (j = 0; j < pack_x_nr && q < depth; j++, q++) {
+      for (int q = 0; q < depth; /* empty */) {
+         for (int j = 0; j < pack_x_nr && q < depth; j++, q++) {
             intel_miptree_set_image_offset(mt, level, q, x, y);
             x += pack_x_pitch;
          }
@@ -214,36 +301,39 @@ brw_miptree_layout_texture_3d(struct intel_context *intel,
 }
 
 void
-brw_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree *mt)
+brw_miptree_layout(struct brw_context *brw, struct intel_mipmap_tree *mt)
 {
+   mt->align_w = intel_horizontal_texture_alignment_unit(brw, mt->format);
+   mt->align_h = intel_vertical_texture_alignment_unit(brw, mt->format);
+
    switch (mt->target) {
    case GL_TEXTURE_CUBE_MAP:
-      if (intel->gen == 4) {
+      if (brw->gen == 4) {
          /* Gen4 stores cube maps as 3D textures. */
          assert(mt->physical_depth0 == 6);
-         brw_miptree_layout_texture_3d(intel, mt);
+         brw_miptree_layout_texture_3d(brw, mt);
       } else {
          /* All other hardware stores cube maps as 2D arrays. */
-        brw_miptree_layout_texture_array(intel, mt);
+        brw_miptree_layout_texture_array(brw, mt);
       }
       break;
 
    case GL_TEXTURE_3D:
-      brw_miptree_layout_texture_3d(intel, mt);
+      brw_miptree_layout_texture_3d(brw, mt);
       break;
 
    case GL_TEXTURE_1D_ARRAY:
    case GL_TEXTURE_2D_ARRAY:
    case GL_TEXTURE_2D_MULTISAMPLE_ARRAY:
    case GL_TEXTURE_CUBE_MAP_ARRAY:
-      brw_miptree_layout_texture_array(intel, mt);
+      brw_miptree_layout_texture_array(brw, mt);
       break;
 
    default:
       switch (mt->msaa_layout) {
       case INTEL_MSAA_LAYOUT_UMS:
       case INTEL_MSAA_LAYOUT_CMS:
-         brw_miptree_layout_texture_array(intel, mt);
+         brw_miptree_layout_texture_array(brw, mt);
          break;
       case INTEL_MSAA_LAYOUT_NONE:
       case INTEL_MSAA_LAYOUT_IMS: