svn merge ^/branches/released .

20 files changed, 1094 insertions, 494 deletions

diff --git a/mesalib/src/mesa/drivers/common/meta.c b/mesalib/src/mesa/drivers/common/meta.c
index 2b8b45bce..06d4337f8 100644
--- a/mesalib/src/mesa/drivers/common/meta.c
+++ b/mesalib/src/mesa/drivers/common/meta.c
@@ -1871,7 +1871,8 @@ _mesa_meta_DrawPixels(struct gl_context *ctx,
        * just going for the matching set of channels, in floating

        * point.

        */

-      if (ctx->Color.ClampFragmentColor != GL_TRUE)

+      if (ctx->Color.ClampFragmentColor != GL_TRUE &&

+	  ctx->Extensions.ARB_texture_float)

 	 texIntFormat = GL_RGBA32F;

    }

    else if (_mesa_is_stencil_format(format)) {

diff --git a/mesalib/src/mesa/main/extensions.c b/mesalib/src/mesa/main/extensions.c
index 109cadd95..d31477731 100644
--- a/mesalib/src/mesa/main/extensions.c
+++ b/mesalib/src/mesa/main/extensions.c
@@ -502,6 +502,7 @@ _mesa_enable_sw_extensions(struct gl_context *ctx)
    ctx->Extensions.EXT_texture_env_dot3 = GL_TRUE;

    ctx->Extensions.EXT_texture_mirror_clamp = GL_TRUE;

    ctx->Extensions.EXT_texture_lod_bias = GL_TRUE;

+   ctx->Extensions.EXT_texture_shared_exponent = GL_TRUE;

 #if FEATURE_EXT_texture_sRGB

    ctx->Extensions.EXT_texture_sRGB = GL_TRUE;

    ctx->Extensions.EXT_texture_sRGB_decode = GL_TRUE;

diff --git a/mesalib/src/mesa/main/fbobject.c b/mesalib/src/mesa/main/fbobject.c
index a961e68cb..68a141d29 100644
--- a/mesalib/src/mesa/main/fbobject.c
+++ b/mesalib/src/mesa/main/fbobject.c
@@ -418,6 +418,15 @@ _mesa_validate_framebuffer(struct gl_context *ctx, struct gl_framebuffer *fb)
          case GL_RG:

             fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED;

             return;

+         case GL_RGB:

+            switch (rb->Format) {

+            case MESA_FORMAT_RGB9_E5_FLOAT:

+               fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED;

+               return;

+            default:;

+            }

+            break;

+

          default:

             /* render buffer format is supported by software rendering */

             ;

@@ -1175,6 +1184,10 @@ _mesa_base_fbo_format(struct gl_context *ctx, GLenum internalFormat)
    case GL_INTENSITY32F_ARB:

       return ctx->Extensions.ARB_texture_float &&

              ctx->Extensions.ARB_framebuffer_object ? GL_INTENSITY : 0;

+   case GL_RGB9_E5:

+      return ctx->Extensions.EXT_texture_shared_exponent ? GL_RGB : 0;

+   case GL_R11F_G11F_B10F:

+      return ctx->Extensions.EXT_packed_float ? GL_RGB : 0;

    /* XXX add integer formats eventually */

    default:

       return 0;

diff --git a/mesalib/src/mesa/main/formats.c b/mesalib/src/mesa/main/formats.c
index 5ae8e2e29..f0f30fe0a 100644
--- a/mesalib/src/mesa/main/formats.c
+++ b/mesalib/src/mesa/main/formats.c
@@ -1072,7 +1072,25 @@ static struct gl_format_info format_info[MESA_FORMAT_COUNT] =
       0, 0, 0, 0,

       0, 16, 0, 0, 0,

       1, 1, 2

-   }

+   },

+   {

+      MESA_FORMAT_RGB9_E5_FLOAT,

+      "MESA_FORMAT_RGB9_E5",

+      GL_RGB,

+      GL_FLOAT,

+      9, 9, 9, 0,

+      0, 0, 0, 0, 0,

+      1, 1, 4

+   },

+   {

+      MESA_FORMAT_R11_G11_B10_FLOAT,

+      "MESA_FORMAT_R11_G11_B10_FLOAT",

+      GL_RGB,

+      GL_FLOAT,

+      11, 11, 10, 0,

+      0, 0, 0, 0, 0,

+      1, 1, 4

+   },

 };

 

 

@@ -1803,6 +1821,16 @@ _mesa_format_to_type_and_comps(gl_format format,
       *comps = 4;

       return;

 

+   case MESA_FORMAT_RGB9_E5_FLOAT:

+      *datatype = GL_UNSIGNED_INT_5_9_9_9_REV;

+      *comps = 3;

+      return;

+

+   case MESA_FORMAT_R11_G11_B10_FLOAT:

+      *datatype = GL_UNSIGNED_INT_10F_11F_11F_REV;

+      *comps = 3;

+      return;

+

    case MESA_FORMAT_COUNT:

       assert(0);

       return;

diff --git a/mesalib/src/mesa/main/formats.h b/mesalib/src/mesa/main/formats.h
index 492eef917..c9fd0baab 100644
--- a/mesalib/src/mesa/main/formats.h
+++ b/mesalib/src/mesa/main/formats.h
@@ -206,6 +206,9 @@ typedef enum
    MESA_FORMAT_SIGNED_AL1616,     /* AAAA AAAA AAAA AAAA LLLL LLLL LLLL LLLL */

    MESA_FORMAT_SIGNED_I16,        /*                     IIII IIII IIII IIII */

 

+   MESA_FORMAT_RGB9_E5_FLOAT,

+   MESA_FORMAT_R11_G11_B10_FLOAT,

+

    MESA_FORMAT_COUNT

 } gl_format;

 

diff --git a/mesalib/src/mesa/main/image.c b/mesalib/src/mesa/main/image.c
index 4bbbbdb37..a5cd5ac03 100644
--- a/mesalib/src/mesa/main/image.c
+++ b/mesalib/src/mesa/main/image.c
@@ -82,6 +82,8 @@ _mesa_type_is_packed(GLenum type)
    case GL_UNSIGNED_SHORT_8_8_MESA:

    case GL_UNSIGNED_SHORT_8_8_REV_MESA:

    case GL_UNSIGNED_INT_24_8_EXT:

+   case GL_UNSIGNED_INT_5_9_9_9_REV:

+   case GL_UNSIGNED_INT_10F_11F_11F_REV:

       return GL_TRUE;

    }

 

@@ -222,6 +224,10 @@ _mesa_sizeof_packed_type( GLenum type )
          return sizeof(GLushort);      

       case GL_UNSIGNED_INT_24_8_EXT:

          return sizeof(GLuint);

+      case GL_UNSIGNED_INT_5_9_9_9_REV:

+         return sizeof(GLuint);

+      case GL_UNSIGNED_INT_10F_11F_11F_REV:

+         return sizeof(GLuint);

       default:

          return -1;

    }

@@ -363,6 +369,16 @@ _mesa_bytes_per_pixel( GLenum format, GLenum type )
             return sizeof(GLuint);

          else

             return -1;

+      case GL_UNSIGNED_INT_5_9_9_9_REV:

+         if (format == GL_RGB)

+            return sizeof(GLuint);

+         else

+            return -1;

+      case GL_UNSIGNED_INT_10F_11F_11F_REV:

+         if (format == GL_RGB)

+            return sizeof(GLuint);

+         else

+            return -1;

       default:

          return -1;

    }

@@ -458,6 +474,10 @@ _mesa_is_legal_format_and_type(const struct gl_context *ctx,
                return GL_TRUE;

             case GL_HALF_FLOAT_ARB:

                return ctx->Extensions.ARB_half_float_pixel;

+            case GL_UNSIGNED_INT_5_9_9_9_REV:

+               return ctx->Extensions.EXT_texture_shared_exponent;

+            case GL_UNSIGNED_INT_10F_11F_11F_REV:

+               return ctx->Extensions.EXT_packed_float;

             default:

                return GL_FALSE;

          }

@@ -821,6 +841,8 @@ _mesa_is_color_format(GLenum format)
       case GL_INTENSITY_SNORM:

       case GL_INTENSITY8_SNORM:

       case GL_INTENSITY16_SNORM:

+      case GL_RGB9_E5:

+      case GL_R11F_G11F_B10F:

          return GL_TRUE;

       case GL_YCBCR_MESA:  /* not considered to be RGB */

          /* fall-through */

diff --git a/mesalib/src/mesa/main/mipmap.c b/mesalib/src/mesa/main/mipmap.c
index b0b0c0c13..ed93cbc6c 100644
--- a/mesalib/src/mesa/main/mipmap.c
+++ b/mesalib/src/mesa/main/mipmap.c
@@ -34,6 +34,9 @@
 #include "teximage.h"

 #include "texstore.h"

 #include "image.h"

+#include "macros.h"

+#include "../../gallium/auxiliary/util/u_format_rgb9e5.h"

+#include "../../gallium/auxiliary/util/u_format_r11g11b10f.h"

 

 

 

@@ -665,6 +668,44 @@ do_row(GLenum datatype, GLuint comps, GLint srcWidth,
       }

    }

 

+   else if (datatype == GL_UNSIGNED_INT_5_9_9_9_REV && comps == 3) {

+      GLuint i, j, k;

+      const GLuint *rowA = (const GLuint*) srcRowA;

+      const GLuint *rowB = (const GLuint*) srcRowB;

+      GLuint *dst = (GLuint*)dstRow;

+      GLfloat res[3], rowAj[3], rowBj[3], rowAk[3], rowBk[3];

+      for (i = j = 0, k = k0; i < (GLuint) dstWidth;

+           i++, j += colStride, k += colStride) {

+         rgb9e5_to_float3(rowA[j], rowAj);

+         rgb9e5_to_float3(rowB[j], rowBj);

+         rgb9e5_to_float3(rowA[k], rowAk);

+         rgb9e5_to_float3(rowB[k], rowBk);

+         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0]) * 0.25F;

+         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1]) * 0.25F;

+         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2]) * 0.25F;

+         dst[i] = float3_to_rgb9e5(res);

+      }

+   }

+

+   else if (datatype == GL_UNSIGNED_INT_10F_11F_11F_REV && comps == 3) {

+      GLuint i, j, k;

+      const GLuint *rowA = (const GLuint*) srcRowA;

+      const GLuint *rowB = (const GLuint*) srcRowB;

+      GLuint *dst = (GLuint*)dstRow;

+      GLfloat res[3], rowAj[3], rowBj[3], rowAk[3], rowBk[3];

+      for (i = j = 0, k = k0; i < (GLuint) dstWidth;

+           i++, j += colStride, k += colStride) {

+         r11g11b10f_to_float3(rowA[j], rowAj);

+         r11g11b10f_to_float3(rowB[j], rowBj);

+         r11g11b10f_to_float3(rowA[k], rowAk);

+         r11g11b10f_to_float3(rowB[k], rowBk);

+         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0]) * 0.25F;

+         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1]) * 0.25F;

+         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2]) * 0.25F;

+         dst[i] = float3_to_r11g11b10f(res);

+      }

+   }

+

    else {

       _mesa_problem(NULL, "bad format in do_row()");

    }

@@ -1245,6 +1286,61 @@ do_row_3D(GLenum datatype, GLuint comps, GLint srcWidth,
          dst[i] = (a << 30) | (b << 20) | (g << 10) | r;

       }

    }

+

+   else if (datatype == GL_UNSIGNED_INT_5_9_9_9_REV && comps == 3) {

+      DECLARE_ROW_POINTERS0(GLuint);

+

+      GLfloat res[3];

+      GLfloat rowAj[3], rowBj[3], rowCj[3], rowDj[3];

+      GLfloat rowAk[3], rowBk[3], rowCk[3], rowDk[3];

+

+      for (i = j = 0, k = k0; i < (GLuint) dstWidth;

+           i++, j += colStride, k += colStride) {

+         rgb9e5_to_float3(rowA[j], rowAj);

+         rgb9e5_to_float3(rowB[j], rowBj);

+         rgb9e5_to_float3(rowC[j], rowCj);

+         rgb9e5_to_float3(rowD[j], rowDj);

+         rgb9e5_to_float3(rowA[k], rowAk);

+         rgb9e5_to_float3(rowB[k], rowBk);

+         rgb9e5_to_float3(rowC[k], rowCk);

+         rgb9e5_to_float3(rowD[k], rowDk);

+         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0] +

+                   rowCj[0] + rowCk[0] + rowDj[0] + rowDk[0]) * 0.125F;

+         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1] +

+                   rowCj[1] + rowCk[1] + rowDj[1] + rowDk[1]) * 0.125F;

+         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2] +

+                   rowCj[2] + rowCk[2] + rowDj[2] + rowDk[2]) * 0.125F;

+         dst[i] = float3_to_rgb9e5(res);

+      }

+   }

+

+   else if (datatype == GL_UNSIGNED_INT_10F_11F_11F_REV && comps == 3) {

+      DECLARE_ROW_POINTERS0(GLuint);

+

+      GLfloat res[3];

+      GLfloat rowAj[3], rowBj[3], rowCj[3], rowDj[3];

+      GLfloat rowAk[3], rowBk[3], rowCk[3], rowDk[3];

+

+      for (i = j = 0, k = k0; i < (GLuint) dstWidth;

+           i++, j += colStride, k += colStride) {

+         r11g11b10f_to_float3(rowA[j], rowAj);

+         r11g11b10f_to_float3(rowB[j], rowBj);

+         r11g11b10f_to_float3(rowC[j], rowCj);

+         r11g11b10f_to_float3(rowD[j], rowDj);

+         r11g11b10f_to_float3(rowA[k], rowAk);

+         r11g11b10f_to_float3(rowB[k], rowBk);

+         r11g11b10f_to_float3(rowC[k], rowCk);

+         r11g11b10f_to_float3(rowD[k], rowDk);

+         res[0] = (rowAj[0] + rowAk[0] + rowBj[0] + rowBk[0] +

+                   rowCj[0] + rowCk[0] + rowDj[0] + rowDk[0]) * 0.125F;

+         res[1] = (rowAj[1] + rowAk[1] + rowBj[1] + rowBk[1] +

+                   rowCj[1] + rowCk[1] + rowDj[1] + rowDk[1]) * 0.125F;

+         res[2] = (rowAj[2] + rowAk[2] + rowBj[2] + rowBk[2] +

+                   rowCj[2] + rowCk[2] + rowDj[2] + rowDk[2]) * 0.125F;

+         dst[i] = float3_to_r11g11b10f(res);

+      }

+   }

+

    else {

       _mesa_problem(NULL, "bad format in do_row()");

    }

diff --git a/mesalib/src/mesa/main/pack.c b/mesalib/src/mesa/main/pack.c
index 620c8a5cd..e6734bbbc 100644
--- a/mesalib/src/mesa/main/pack.c
+++ b/mesalib/src/mesa/main/pack.c
@@ -37,6 +37,9 @@
 #include "mtypes.h"

 #include "pack.h"

 #include "pixeltransfer.h"

+#include "imports.h"

+#include "../../gallium/auxiliary/util/u_format_rgb9e5.h"

+#include "../../gallium/auxiliary/util/u_format_r11g11b10f.h"

 

 

 /**

@@ -1892,6 +1895,22 @@ _mesa_pack_rgba_span_float(struct gl_context *ctx, GLuint n, GLfloat rgba[][4],
             }

          }

          break;

+      case GL_UNSIGNED_INT_5_9_9_9_REV:

+         {

+            GLuint *dst = (GLuint *) dstAddr;

+            for (i = 0; i < n; i++) {

+               dst[i] = float3_to_rgb9e5(rgba[i]);

+            }

+         }

+         break;

+      case GL_UNSIGNED_INT_10F_11F_11F_REV:

+         {

+            GLuint *dst = (GLuint *) dstAddr;

+            for (i = 0; i < n; i++) {

+               dst[i] = float3_to_r11g11b10f(rgba[i]);

+            }

+         }

+         break;

       default:

          _mesa_problem(ctx, "bad type in _mesa_pack_rgba_span_float");

          return;

@@ -2330,7 +2349,9 @@ extract_float_rgba(GLuint n, GLfloat rgba[][4],
           srcType == GL_UNSIGNED_INT_8_8_8_8 ||

           srcType == GL_UNSIGNED_INT_8_8_8_8_REV ||

           srcType == GL_UNSIGNED_INT_10_10_10_2 ||

-          srcType == GL_UNSIGNED_INT_2_10_10_10_REV);

+          srcType == GL_UNSIGNED_INT_2_10_10_10_REV ||

+          srcType == GL_UNSIGNED_INT_5_9_9_9_REV ||

+          srcType == GL_UNSIGNED_INT_10F_11F_11F_REV);

 

    get_component_mapping(srcFormat,

                          &rSrc, &gSrc, &bSrc, &aSrc,

@@ -2800,6 +2821,62 @@ extract_float_rgba(GLuint n, GLfloat rgba[][4],
             }

          }

          break;

+      case GL_UNSIGNED_INT_5_9_9_9_REV:

+         if (swapBytes) {

+            const GLuint *uisrc = (const GLuint *) src;

+            GLuint i;

+            GLfloat f[3];

+            for (i = 0; i < n; i ++) {

+               GLuint p = uisrc[i];

+               SWAP4BYTE(p);

+               rgb9e5_to_float3(p, f);

+               rgba[i][rDst] = f[0];

+               rgba[i][gDst] = f[1];

+               rgba[i][bDst] = f[2];

+               rgba[i][aDst] = 1.0F;

+            }

+         }

+         else {

+            const GLuint *uisrc = (const GLuint *) src;

+            GLuint i;

+            GLfloat f[3];

+            for (i = 0; i < n; i ++) {

+               rgb9e5_to_float3(uisrc[i], f);

+               rgba[i][rDst] = f[0];

+               rgba[i][gDst] = f[1];

+               rgba[i][bDst] = f[2];

+               rgba[i][aDst] = 1.0F;

+            }

+         }

+         break;

+      case GL_UNSIGNED_INT_10F_11F_11F_REV:

+         if (swapBytes) {

+            const GLuint *uisrc = (const GLuint *) src;

+            GLuint i;

+            GLfloat f[3];

+            for (i = 0; i < n; i ++) {

+               GLuint p = uisrc[i];

+               SWAP4BYTE(p);

+               r11g11b10f_to_float3(p, f);

+               rgba[i][rDst] = f[0];

+               rgba[i][gDst] = f[1];

+               rgba[i][bDst] = f[2];

+               rgba[i][aDst] = 1.0F;

+            }

+         }

+         else {

+            const GLuint *uisrc = (const GLuint *) src;

+            GLuint i;

+            GLfloat f[3];

+            for (i = 0; i < n; i ++) {

+               r11g11b10f_to_float3(uisrc[i], f);

+               rgba[i][rDst] = f[0];

+               rgba[i][gDst] = f[1];

+               rgba[i][bDst] = f[2];

+               rgba[i][aDst] = 1.0F;

+            }

+         }

+         break;

       default:

          _mesa_problem(NULL, "bad srcType in extract float data");

          break;

@@ -2902,7 +2979,9 @@ extract_uint_rgba(GLuint n, GLuint rgba[][4],
           srcType == GL_UNSIGNED_INT_8_8_8_8 ||

           srcType == GL_UNSIGNED_INT_8_8_8_8_REV ||

           srcType == GL_UNSIGNED_INT_10_10_10_2 ||

-          srcType == GL_UNSIGNED_INT_2_10_10_10_REV);

+          srcType == GL_UNSIGNED_INT_2_10_10_10_REV ||

+          srcType == GL_UNSIGNED_INT_5_9_9_9_REV ||

+          srcType == GL_UNSIGNED_INT_10F_11F_11F_REV);

 

    get_component_mapping(srcFormat,

                          &rSrc, &gSrc, &bSrc, &aSrc,

@@ -3266,6 +3345,64 @@ extract_uint_rgba(GLuint n, GLuint rgba[][4],
             }

          }

          break;

+      case GL_UNSIGNED_INT_5_9_9_9_REV:

+         if (swapBytes) {

+            const GLuint *uisrc = (const GLuint *) src;

+            GLuint i;

+            float f[3];

+            for (i = 0; i < n; i ++) {

+               GLuint p = uisrc[i];

+               SWAP4BYTE(p);

+               rgb9e5_to_float3(p, f);

+               rgba[i][rDst] = clamp_float_to_uint(f[0]);

+               rgba[i][gDst] = clamp_float_to_uint(f[1]);

+               rgba[i][bDst] = clamp_float_to_uint(f[2]);

+               rgba[i][aDst] = 1;

+            }

+         }

+         else {

+            const GLuint *uisrc = (const GLuint *) src;

+            GLuint i;

+            float f[3];

+            for (i = 0; i < n; i ++) {

+               GLuint p = uisrc[i];

+               rgb9e5_to_float3(p, f);

+               rgba[i][rDst] = clamp_float_to_uint(f[0]);

+               rgba[i][gDst] = clamp_float_to_uint(f[1]);

+               rgba[i][bDst] = clamp_float_to_uint(f[2]);

+               rgba[i][aDst] = 1;

+            }

+         }

+         break;

+      case GL_UNSIGNED_INT_10F_11F_11F_REV:

+         if (swapBytes) {

+            const GLuint *uisrc = (const GLuint *) src;

+            GLuint i;

+            float f[3];

+            for (i = 0; i < n; i ++) {

+               GLuint p = uisrc[i];

+               SWAP4BYTE(p);

+               r11g11b10f_to_float3(p, f);

+               rgba[i][rDst] = clamp_float_to_uint(f[0]);

+               rgba[i][gDst] = clamp_float_to_uint(f[1]);

+               rgba[i][bDst] = clamp_float_to_uint(f[2]);

+               rgba[i][aDst] = 1;

+            }

+         }

+         else {

+            const GLuint *uisrc = (const GLuint *) src;

+            GLuint i;

+            float f[3];

+            for (i = 0; i < n; i ++) {

+               GLuint p = uisrc[i];

+               r11g11b10f_to_float3(p, f);

+               rgba[i][rDst] = clamp_float_to_uint(f[0]);

+               rgba[i][gDst] = clamp_float_to_uint(f[1]);

+               rgba[i][bDst] = clamp_float_to_uint(f[2]);

+               rgba[i][aDst] = 1;

+            }

+         }

+         break;

       default:

          _mesa_problem(NULL, "bad srcType in extract uint data");

          break;

@@ -3345,7 +3482,9 @@ _mesa_unpack_color_span_chan( struct gl_context *ctx,
           srcType == GL_UNSIGNED_INT_8_8_8_8 ||

           srcType == GL_UNSIGNED_INT_8_8_8_8_REV ||

           srcType == GL_UNSIGNED_INT_10_10_10_2 ||

-          srcType == GL_UNSIGNED_INT_2_10_10_10_REV);

+          srcType == GL_UNSIGNED_INT_2_10_10_10_REV ||

+          srcType == GL_UNSIGNED_INT_5_9_9_9_REV ||

+          srcType == GL_UNSIGNED_INT_10F_11F_11F_REV);

 

    /* Try simple cases first */

    if (transferOps == 0) {

@@ -3667,7 +3806,9 @@ _mesa_unpack_color_span_float( struct gl_context *ctx,
           srcType == GL_UNSIGNED_INT_8_8_8_8 ||

           srcType == GL_UNSIGNED_INT_8_8_8_8_REV ||

           srcType == GL_UNSIGNED_INT_10_10_10_2 ||

-          srcType == GL_UNSIGNED_INT_2_10_10_10_REV);

+          srcType == GL_UNSIGNED_INT_2_10_10_10_REV ||

+          srcType == GL_UNSIGNED_INT_5_9_9_9_REV ||

+          srcType == GL_UNSIGNED_INT_10F_11F_11F_REV);

 

    /* general solution, no special cases, yet */

    {

@@ -3873,7 +4014,9 @@ _mesa_unpack_color_span_uint(struct gl_context *ctx,
           srcType == GL_UNSIGNED_INT_8_8_8_8 ||

           srcType == GL_UNSIGNED_INT_8_8_8_8_REV ||

           srcType == GL_UNSIGNED_INT_10_10_10_2 ||

-          srcType == GL_UNSIGNED_INT_2_10_10_10_REV);

+          srcType == GL_UNSIGNED_INT_2_10_10_10_REV ||

+          srcType == GL_UNSIGNED_INT_5_9_9_9_REV ||

+          srcType == GL_UNSIGNED_INT_10F_11F_11F_REV);

 

 

    /* Extract image data as uint[4] pixels */

diff --git a/mesalib/src/mesa/main/texfetch.c b/mesalib/src/mesa/main/texfetch.c
index cb2716619..ab3abf657 100644
--- a/mesalib/src/mesa/main/texfetch.c
+++ b/mesalib/src/mesa/main/texfetch.c
@@ -41,6 +41,8 @@
 #include "texcompress_rgtc.h"

 #include "texfetch.h"

 #include "teximage.h"

+#include "../../gallium/auxiliary/util/u_format_rgb9e5.h"

+#include "../../gallium/auxiliary/util/u_format_r11g11b10f.h"

 

 

 /**

@@ -898,6 +900,20 @@ texfetch_funcs[MESA_FORMAT_COUNT] =
       fetch_texel_3d_signed_i16,

       store_texel_signed_i16

    },

+   {

+      MESA_FORMAT_RGB9_E5_FLOAT,

+      fetch_texel_1d_rgb9_e5,

+      fetch_texel_2d_rgb9_e5,

+      fetch_texel_3d_rgb9_e5,

+      store_texel_rgb9_e5

+   },

+   {

+      MESA_FORMAT_R11_G11_B10_FLOAT,

+      fetch_texel_1d_r11_g11_b10f,

+      fetch_texel_2d_r11_g11_b10f,

+      fetch_texel_3d_r11_g11_b10f,

+      store_texel_r11_g11_b10f

+   }

 };

 

 

diff --git a/mesalib/src/mesa/main/texfetch_tmp.h b/mesalib/src/mesa/main/texfetch_tmp.h
index 57bb94c68..278becc11 100644
--- a/mesalib/src/mesa/main/texfetch_tmp.h
+++ b/mesalib/src/mesa/main/texfetch_tmp.h
@@ -2332,6 +2332,48 @@ static void store_texel_s8_z24(struct gl_texture_image *texImage,
 #endif

 

 

+/* MESA_FORMAT_RGB9_E5 ******************************************************/

+

+static void FETCH(rgb9_e5)( const struct gl_texture_image *texImage,

+                            GLint i, GLint j, GLint k, GLfloat *texel )

+{

+   const GLuint *src = TEXEL_ADDR(GLuint, texImage, i, j, k, 1);

+   rgb9e5_to_float3(*src, texel);

+   texel[ACOMP] = 1.0F;

+}

+

+#if DIM == 3

+static void store_texel_rgb9_e5(struct gl_texture_image *texImage,

+                                GLint i, GLint j, GLint k, const void *texel)

+{

+   const GLfloat *src = (const GLfloat *) texel;

+   GLuint *dst = TEXEL_ADDR(GLuint, texImage, i, j, k, 1);

+   *dst = float3_to_rgb9e5(src);

+}

+#endif

+

+

+/* MESA_FORMAT_R11_G11_B10_FLOAT *********************************************/

+

+static void FETCH(r11_g11_b10f)( const struct gl_texture_image *texImage,

+                                 GLint i, GLint j, GLint k, GLfloat *texel )

+{

+   const GLuint *src = TEXEL_ADDR(GLuint, texImage, i, j, k, 1);

+   r11g11b10f_to_float3(*src, texel);

+   texel[ACOMP] = 1.0F;

+}

+

+#if DIM == 3

+static void store_texel_r11_g11_b10f(struct gl_texture_image *texImage,

+                                     GLint i, GLint j, GLint k, const void *texel)

+{

+   const GLfloat *src = (const GLfloat *) texel;

+   GLuint *dst = TEXEL_ADDR(GLuint, texImage, i, j, k, 1);

+   *dst = float3_to_r11g11b10f(src);

+}

+#endif

+

+

 #undef TEXEL_ADDR

 #undef DIM

 #undef FETCH

diff --git a/mesalib/src/mesa/main/texformat.c b/mesalib/src/mesa/main/texformat.c
index 24c4f1a92..41d9e9599 100644
--- a/mesalib/src/mesa/main/texformat.c
+++ b/mesalib/src/mesa/main/texformat.c
@@ -382,6 +382,26 @@ _mesa_choose_tex_format( struct gl_context *ctx, GLint internalFormat,
       }

    }

 

+   if (ctx->Extensions.EXT_texture_shared_exponent) {

+      switch (internalFormat) {

+         case GL_RGB9_E5:

+            ASSERT(ctx->TextureFormatSupported[MESA_FORMAT_RGB9_E5_FLOAT]);

+            return MESA_FORMAT_RGB9_E5_FLOAT;

+         default:

+            ; /* fallthrough */

+      }

+   }

+

+   if (ctx->Extensions.EXT_packed_float) {

+      switch (internalFormat) {

+         case GL_R11F_G11F_B10F:

+            ASSERT(ctx->TextureFormatSupported[MESA_FORMAT_R11_G11_B10_FLOAT]);

+            return MESA_FORMAT_R11_G11_B10_FLOAT;

+         default:

+            ; /* fallthrough */

+      }

+   }

+

    if (ctx->Extensions.EXT_packed_depth_stencil) {

       switch (internalFormat) {

          case GL_DEPTH_STENCIL_EXT:

diff --git a/mesalib/src/mesa/main/texparam.c b/mesalib/src/mesa/main/texparam.c
index c6683980a..74e1f4a1b 100644
--- a/mesalib/src/mesa/main/texparam.c
+++ b/mesalib/src/mesa/main/texparam.c
@@ -979,11 +979,9 @@ _mesa_GetTexLevelParameteriv( GLenum target, GLint level,
          }

          break;

       case GL_TEXTURE_SHARED_SIZE:

-         if (ctx->VersionMajor >= 3) {

-            /* XXX return number of exponent bits for shared exponent texture

-             * formats, like GL_RGB9_E5.

-             */

-            *params = 0;

+         if (ctx->VersionMajor >= 3 ||

+             ctx->Extensions.EXT_texture_shared_exponent) {

+            *params = texFormat == MESA_FORMAT_RGB9_E5_FLOAT ? 5 : 0;

          }

          else {

             goto invalid_pname;

diff --git a/mesalib/src/mesa/main/texrender.c b/mesalib/src/mesa/main/texrender.c
index 2766bd6c5..4bcef9899 100644
--- a/mesalib/src/mesa/main/texrender.c
+++ b/mesalib/src/mesa/main/texrender.c
@@ -530,6 +530,7 @@ update_wrapper(struct gl_context *ctx, struct gl_renderbuffer_attachment *att)
 {

    struct texture_renderbuffer *trb

       = (struct texture_renderbuffer *) att->Renderbuffer;

+   GLuint unused;

 

    (void) ctx;

    ASSERT(trb);

@@ -602,8 +603,10 @@ update_wrapper(struct gl_context *ctx, struct gl_renderbuffer_attachment *att)
       trb->Base._BaseFormat = GL_RGBA;

       break;

    default:

-      trb->Base.DataType = CHAN_TYPE;

-      trb->Base._BaseFormat = GL_RGBA;

+      _mesa_format_to_type_and_comps(trb->TexImage->TexFormat,

+                                     &trb->Base.DataType, &unused);

+      trb->Base._BaseFormat =

+         _mesa_base_fbo_format(ctx, trb->TexImage->InternalFormat);

    }

    trb->Base.Data = trb->TexImage->Data;

 }

diff --git a/mesalib/src/mesa/main/texstore.c b/mesalib/src/mesa/main/texstore.c
index bf2b6b6e4..ea116a011 100644
--- a/mesalib/src/mesa/main/texstore.c
+++ b/mesalib/src/mesa/main/texstore.c
@@ -70,6 +70,8 @@
 #include "teximage.h"

 #include "texstore.h"

 #include "enums.h"

+#include "../../gallium/auxiliary/util/u_format_rgb9e5.h"

+#include "../../gallium/auxiliary/util/u_format_r11g11b10f.h"

 

 

 enum {

@@ -4176,6 +4178,111 @@ _mesa_texstore_sla8(TEXSTORE_PARAMS)
 

 #endif /* FEATURE_EXT_texture_sRGB */

 

+static GLboolean

+_mesa_texstore_rgb9_e5(TEXSTORE_PARAMS)

+{

+   const GLenum baseFormat = _mesa_get_format_base_format(dstFormat);

+

+   ASSERT(dstFormat == MESA_FORMAT_RGB9_E5_FLOAT);

+   ASSERT(baseInternalFormat == GL_RGB);

+

+   if (!ctx->_ImageTransferState &&

+       !srcPacking->SwapBytes &&

+       srcFormat == GL_RGB &&

+       srcType == GL_UNSIGNED_INT_5_9_9_9_REV) {

+      /* simple memcpy path */

+      memcpy_texture(ctx, dims,

+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,

+                     dstRowStride,

+                     dstImageOffsets,

+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,

+                     srcAddr, srcPacking);

+   }

+   else {

+      /* general path */

+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,

+                                                 baseInternalFormat,

+                                                 baseFormat,

+                                                 srcWidth, srcHeight, srcDepth,

+                                                 srcFormat, srcType, srcAddr,

+                                                 srcPacking,

+                                                 ctx->_ImageTransferState);

+      const GLfloat *srcRow = tempImage;

+      GLint img, row, col;

+      if (!tempImage)

+         return GL_FALSE;

+      for (img = 0; img < srcDepth; img++) {

+         GLubyte *dstRow = (GLubyte *) dstAddr

+            + dstImageOffsets[dstZoffset + img] * 4

+            + dstYoffset * dstRowStride

+            + dstXoffset * 4;

+         for (row = 0; row < srcHeight; row++) {

+            GLuint *dstUI = (GLuint*)dstRow;

+            for (col = 0; col < srcWidth; col++) {

+               dstUI[col] = float3_to_rgb9e5(&srcRow[col * 3]);

+            }

+            dstRow += dstRowStride;

+            srcRow += srcWidth * 3;

+         }

+      }

+

+      free((void *) tempImage);

+   }

+   return GL_TRUE;

+}

+

+static GLboolean

+_mesa_texstore_r11_g11_b10f(TEXSTORE_PARAMS)

+{

+   const GLenum baseFormat = _mesa_get_format_base_format(dstFormat);

+

+   ASSERT(dstFormat == MESA_FORMAT_R11_G11_B10_FLOAT);

+   ASSERT(baseInternalFormat == GL_RGB);

+

+   if (!ctx->_ImageTransferState &&

+       !srcPacking->SwapBytes &&

+       srcFormat == GL_RGB &&

+       srcType == GL_UNSIGNED_INT_10F_11F_11F_REV) {

+      /* simple memcpy path */

+      memcpy_texture(ctx, dims,

+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,

+                     dstRowStride,

+                     dstImageOffsets,

+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,

+                     srcAddr, srcPacking);

+   }

+   else {

+      /* general path */

+      const GLfloat *tempImage = _mesa_make_temp_float_image(ctx, dims,

+                                                 baseInternalFormat,

+                                                 baseFormat,

+                                                 srcWidth, srcHeight, srcDepth,

+                                                 srcFormat, srcType, srcAddr,

+                                                 srcPacking,

+                                                 ctx->_ImageTransferState);

+      const GLfloat *srcRow = tempImage;

+      GLint img, row, col;

+      if (!tempImage)

+         return GL_FALSE;

+      for (img = 0; img < srcDepth; img++) {

+         GLubyte *dstRow = (GLubyte *) dstAddr

+            + dstImageOffsets[dstZoffset + img] * 4

+            + dstYoffset * dstRowStride

+            + dstXoffset * 4;

+         for (row = 0; row < srcHeight; row++) {

+            GLuint *dstUI = (GLuint*)dstRow;

+            for (col = 0; col < srcWidth; col++) {

+               dstUI[col] = float3_to_r11g11b10f(&srcRow[col * 3]);

+            }

+            dstRow += dstRowStride;

+            srcRow += srcWidth * 3;

+         }

+      }

+

+      free((void *) tempImage);

+   }

+   return GL_TRUE;

+}

 

 

 

@@ -4309,6 +4416,9 @@ texstore_funcs[MESA_FORMAT_COUNT] =
    { MESA_FORMAT_SIGNED_L16, _mesa_texstore_snorm16 },

    { MESA_FORMAT_SIGNED_AL1616, _mesa_texstore_snorm1616 },

    { MESA_FORMAT_SIGNED_I16, _mesa_texstore_snorm16 },

+

+   { MESA_FORMAT_RGB9_E5_FLOAT, _mesa_texstore_rgb9_e5 },

+   { MESA_FORMAT_R11_G11_B10_FLOAT, _mesa_texstore_r11_g11_b10f },

 };

 

 

diff --git a/mesalib/src/mesa/main/transformfeedback.c b/mesalib/src/mesa/main/transformfeedback.c
index 2ff262fc6..85c213552 100644
--- a/mesalib/src/mesa/main/transformfeedback.c
+++ b/mesalib/src/mesa/main/transformfeedback.c
@@ -42,7 +42,6 @@
 #include "main/dispatch.h"

 

 #include "program/prog_parameter.h"

-//#include "program/shader_api.h"

 

 

 #if FEATURE_EXT_transform_feedback

@@ -507,7 +506,7 @@ _mesa_BindBufferBase(GLenum target, GLuint index, GLuint buffer)
 

    if (obj->Active) {

       _mesa_error(ctx, GL_INVALID_OPERATION,

-                  "glBindBufferRange(transform feedback active)");

+                  "glBindBufferBase(transform feedback active)");

       return;

    }

 

@@ -555,7 +554,7 @@ _mesa_BindBufferOffsetEXT(GLenum target, GLuint index, GLuint buffer,
 

    if (obj->Active) {

       _mesa_error(ctx, GL_INVALID_OPERATION,

-                  "glBindBufferRange(transform feedback active)");

+                  "glBindBufferOffsetEXT(transform feedback active)");

       return;

    }

 

@@ -885,7 +884,7 @@ _mesa_ResumeTransformFeedback(void)
 

    if (!obj->Active || !obj->Paused) {

       _mesa_error(ctx, GL_INVALID_OPERATION,

-               "glPauseTransformFeedback(feedback not active or not paused)");

+               "glResumeTransformFeedback(feedback not active or not paused)");

       return;

    }

 

@@ -932,15 +931,4 @@ _mesa_DrawTransformFeedback(GLenum mode, GLuint name)
 }

 

 

-/*

-XXX misc to do:

-

-glGet*() for

-

-GL_TRANSFORM_FEEDBACK_BUFFER_PAUSED

-GL_TRANSFORM_FEEDBACK_BUFFER_ACTIVE

-GL_TRANSFORM_FEEDBACK_BINDING

-*/

-

-

 #endif /* FEATURE_EXT_transform_feedback */

diff --git a/mesalib/src/mesa/main/varray.c b/mesalib/src/mesa/main/varray.c
index 1f771a585..6b888b485 100644
--- a/mesalib/src/mesa/main/varray.c
+++ b/mesalib/src/mesa/main/varray.c
@@ -57,8 +57,8 @@
 #define HALF_BIT             0x80

 #define FLOAT_BIT            0x100

 #define DOUBLE_BIT           0x200

-#define FIXED_BIT            0x400

-

+#define FIXED_ES_BIT         0x400

+#define FIXED_GL_BIT         0x800

 

 

 /** Convert GL datatype enum into a <type>_BIT value seen above */

@@ -90,7 +90,7 @@ type_to_bit(const struct gl_context *ctx, GLenum type)
    case GL_DOUBLE:

       return DOUBLE_BIT;

    case GL_FIXED:

-      return FIXED_BIT;

+      return ctx->API == API_OPENGL ? FIXED_GL_BIT : FIXED_ES_BIT;

    default:

       return 0;

    }

@@ -130,7 +130,10 @@ update_array(struct gl_context *ctx,
 

    if (ctx->API != API_OPENGLES && ctx->API != API_OPENGLES2) {

       /* fixed point arrays / data is only allowed with OpenGL ES 1.x/2.0 */

-      legalTypesMask &= ~FIXED_BIT;

+      legalTypesMask &= ~FIXED_ES_BIT;

+   }

+   if (!ctx->Extensions.ARB_ES2_compatibility) {

+      legalTypesMask &= ~FIXED_GL_BIT;

    }

 

    typeBit = type_to_bit(ctx, type);

@@ -198,7 +201,7 @@ void GLAPIENTRY
 _mesa_VertexPointer(GLint size, GLenum type, GLsizei stride, const GLvoid *ptr)

 {

    GLbitfield legalTypes = (SHORT_BIT | INT_BIT | FLOAT_BIT |

-                            DOUBLE_BIT | HALF_BIT | FIXED_BIT);

+                            DOUBLE_BIT | HALF_BIT | FIXED_ES_BIT);

    GET_CURRENT_CONTEXT(ctx);

    ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx);

 

@@ -217,7 +220,7 @@ _mesa_NormalPointer(GLenum type, GLsizei stride, const GLvoid *ptr )
 {

    const GLbitfield legalTypes = (BYTE_BIT | SHORT_BIT | INT_BIT |

                                   HALF_BIT | FLOAT_BIT | DOUBLE_BIT |

-                                  FIXED_BIT);

+                                  FIXED_ES_BIT);

    GET_CURRENT_CONTEXT(ctx);

    ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx);

 

@@ -235,7 +238,7 @@ _mesa_ColorPointer(GLint size, GLenum type, GLsizei stride, const GLvoid *ptr)
                                   SHORT_BIT | UNSIGNED_SHORT_BIT |

                                   INT_BIT | UNSIGNED_INT_BIT |

                                   HALF_BIT | FLOAT_BIT | DOUBLE_BIT |

-                                  FIXED_BIT);

+                                  FIXED_ES_BIT);

    GET_CURRENT_CONTEXT(ctx);

    ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx);

 

@@ -299,7 +302,7 @@ _mesa_TexCoordPointer(GLint size, GLenum type, GLsizei stride,
 {

    GLbitfield legalTypes = (SHORT_BIT | INT_BIT |

                             HALF_BIT | FLOAT_BIT | DOUBLE_BIT |

-                            FIXED_BIT);

+                            FIXED_ES_BIT);

    GET_CURRENT_CONTEXT(ctx);

    const GLuint unit = ctx->Array.ActiveTexture;

    ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx);

@@ -337,7 +340,7 @@ _mesa_EdgeFlagPointer(GLsizei stride, const GLvoid *ptr)
 void GLAPIENTRY

 _mesa_PointSizePointer(GLenum type, GLsizei stride, const GLvoid *ptr)

 {

-   const GLbitfield legalTypes = (FLOAT_BIT | FIXED_BIT);

+   const GLbitfield legalTypes = (FLOAT_BIT | FIXED_ES_BIT);

    GET_CURRENT_CONTEXT(ctx);

    ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx);

 

@@ -405,7 +408,7 @@ _mesa_VertexAttribPointerARB(GLuint index, GLint size, GLenum type,
                                   SHORT_BIT | UNSIGNED_SHORT_BIT |

                                   INT_BIT | UNSIGNED_INT_BIT |

                                   HALF_BIT | FLOAT_BIT | DOUBLE_BIT |

-                                  FIXED_BIT);

+                                  FIXED_ES_BIT | FIXED_GL_BIT);

    GET_CURRENT_CONTEXT(ctx);

    ASSERT_OUTSIDE_BEGIN_END(ctx);

 

diff --git a/mesalib/src/mesa/program/register_allocate.c b/mesalib/src/mesa/program/register_allocate.c
index 95a9bde40..e78db24a4 100644
--- a/mesalib/src/mesa/program/register_allocate.c
+++ b/mesalib/src/mesa/program/register_allocate.c
@@ -1,455 +1,517 @@
-/*
- * Copyright © 2010 Intel Corporation
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS 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:
- *    Eric Anholt <eric@anholt.net>
- *
- */
-
-/** @file register_allocate.c
- *
- * Graph-coloring register allocator.
- */
-
-#include <ralloc.h>
-
-#include "main/imports.h"
-#include "main/macros.h"
-#include "main/mtypes.h"
-#include "register_allocate.h"
-
-struct ra_reg {
-   GLboolean *conflicts;
-   unsigned int *conflict_list;
-   unsigned int conflict_list_size;
-   unsigned int num_conflicts;
-};
-
-struct ra_regs {
-   struct ra_reg *regs;
-   unsigned int count;
-
-   struct ra_class **classes;
-   unsigned int class_count;
-};
-
-struct ra_class {
-   GLboolean *regs;
-
-   /**
-    * p_B in Runeson/Nyström paper.
-    *
-    * This is "how many regs are in the set."
-    */
-   unsigned int p;
-
-   /**
-    * q_B,C in Runeson/Nyström paper.
-    */
-   unsigned int *q;
-};
-
-struct ra_node {
-   GLboolean *adjacency;
-   unsigned int *adjacency_list;
-   unsigned int class;
-   unsigned int adjacency_count;
-   unsigned int reg;
-   GLboolean in_stack;
-   float spill_cost;
-};
-
-struct ra_graph {
-   struct ra_regs *regs;
-   /**
-    * the variables that need register allocation.
-    */
-   struct ra_node *nodes;
-   unsigned int count; /**< count of nodes. */
-
-   unsigned int *stack;
-   unsigned int stack_count;
-};
-
-struct ra_regs *
-ra_alloc_reg_set(unsigned int count)
-{
-   unsigned int i;
-   struct ra_regs *regs;
-
-   regs = rzalloc(NULL, struct ra_regs);
-   regs->count = count;
-   regs->regs = rzalloc_array(regs, struct ra_reg, count);
-
-   for (i = 0; i < count; i++) {
-      regs->regs[i].conflicts = rzalloc_array(regs->regs, GLboolean, count);
-      regs->regs[i].conflicts[i] = GL_TRUE;
-
-      regs->regs[i].conflict_list = ralloc_array(regs->regs, unsigned int, 4);
-      regs->regs[i].conflict_list_size = 4;
-      regs->regs[i].conflict_list[0] = i;
-      regs->regs[i].num_conflicts = 1;
-   }
-
-   return regs;
-}
-
-static void
-ra_add_conflict_list(struct ra_regs *regs, unsigned int r1, unsigned int r2)
-{
-   struct ra_reg *reg1 = &regs->regs[r1];
-
-   if (reg1->conflict_list_size == reg1->num_conflicts) {
-      reg1->conflict_list_size *= 2;
-      reg1->conflict_list = reralloc(regs->regs, reg1->conflict_list,
-				     unsigned int, reg1->conflict_list_size);
-   }
-   reg1->conflict_list[reg1->num_conflicts++] = r2;
-   reg1->conflicts[r2] = GL_TRUE;
-}
-
-void
-ra_add_reg_conflict(struct ra_regs *regs, unsigned int r1, unsigned int r2)
-{
-   if (!regs->regs[r1].conflicts[r2]) {
-      ra_add_conflict_list(regs, r1, r2);
-      ra_add_conflict_list(regs, r2, r1);
-   }
-}
-
-unsigned int
-ra_alloc_reg_class(struct ra_regs *regs)
-{
-   struct ra_class *class;
-
-   regs->classes = reralloc(regs->regs, regs->classes, struct ra_class *,
-			    regs->class_count + 1);
-
-   class = rzalloc(regs, struct ra_class);
-   regs->classes[regs->class_count] = class;
-
-   class->regs = rzalloc_array(class, GLboolean, regs->count);
-
-   return regs->class_count++;
-}
-
-void
-ra_class_add_reg(struct ra_regs *regs, unsigned int c, unsigned int r)
-{
-   struct ra_class *class = regs->classes[c];
-
-   class->regs[r] = GL_TRUE;
-   class->p++;
-}
-
-/**
- * Must be called after all conflicts and register classes have been
- * set up and before the register set is used for allocation.
- */
-void
-ra_set_finalize(struct ra_regs *regs)
-{
-   unsigned int b, c;
-
-   for (b = 0; b < regs->class_count; b++) {
-      regs->classes[b]->q = ralloc_array(regs, unsigned int, regs->class_count);
-   }
-
-   /* Compute, for each class B and C, how many regs of B an
-    * allocation to C could conflict with.
-    */
-   for (b = 0; b < regs->class_count; b++) {
-      for (c = 0; c < regs->class_count; c++) {
-	 unsigned int rc;
-	 int max_conflicts = 0;
-
-	 for (rc = 0; rc < regs->count; rc++) {
-	    int conflicts = 0;
-	    int i;
-
-	    if (!regs->classes[c]->regs[rc])
-	       continue;
-
-	    for (i = 0; i < regs->regs[rc].num_conflicts; i++) {
-	       unsigned int rb = regs->regs[rc].conflict_list[i];
-	       if (regs->classes[b]->regs[rb])
-		  conflicts++;
-	    }
-	    max_conflicts = MAX2(max_conflicts, conflicts);
-	 }
-	 regs->classes[b]->q[c] = max_conflicts;
-      }
-   }
-}
-
-static void
-ra_add_node_adjacency(struct ra_graph *g, unsigned int n1, unsigned int n2)
-{
-   g->nodes[n1].adjacency[n2] = GL_TRUE;
-   g->nodes[n1].adjacency_list[g->nodes[n1].adjacency_count] = n2;
-   g->nodes[n1].adjacency_count++;
-}
-
-struct ra_graph *
-ra_alloc_interference_graph(struct ra_regs *regs, unsigned int count)
-{
-   struct ra_graph *g;
-   unsigned int i;
-
-   g = rzalloc(regs, struct ra_graph);
-   g->regs = regs;
-   g->nodes = rzalloc_array(g, struct ra_node, count);
-   g->count = count;
-
-   g->stack = rzalloc_array(g, unsigned int, count);
-
-   for (i = 0; i < count; i++) {
-      g->nodes[i].adjacency = rzalloc_array(g, GLboolean, count);
-      g->nodes[i].adjacency_list = ralloc_array(g, unsigned int, count);
-      g->nodes[i].adjacency_count = 0;
-      ra_add_node_adjacency(g, i, i);
-      g->nodes[i].reg = ~0;
-   }
-
-   return g;
-}
-
-void
-ra_set_node_class(struct ra_graph *g,
-		  unsigned int n, unsigned int class)
-{
-   g->nodes[n].class = class;
-}
-
-void
-ra_add_node_interference(struct ra_graph *g,
-			 unsigned int n1, unsigned int n2)
-{
-   if (!g->nodes[n1].adjacency[n2]) {
-      ra_add_node_adjacency(g, n1, n2);
-      ra_add_node_adjacency(g, n2, n1);
-   }
-}
-
-static GLboolean pq_test(struct ra_graph *g, unsigned int n)
-{
-   unsigned int j;
-   unsigned int q = 0;
-   int n_class = g->nodes[n].class;
-
-   for (j = 0; j < g->nodes[n].adjacency_count; j++) {
-      unsigned int n2 = g->nodes[n].adjacency_list[j];
-      unsigned int n2_class = g->nodes[n2].class;
-
-      if (n != n2 && !g->nodes[n2].in_stack) {
-	 q += g->regs->classes[n_class]->q[n2_class];
-      }
-   }
-
-   return q < g->regs->classes[n_class]->p;
-}
-
-/**
- * Simplifies the interference graph by pushing all
- * trivially-colorable nodes into a stack of nodes to be colored,
- * removing them from the graph, and rinsing and repeating.
- *
- * Returns GL_TRUE if all nodes were removed from the graph.  GL_FALSE
- * means that either spilling will be required, or optimistic coloring
- * should be applied.
- */
-GLboolean
-ra_simplify(struct ra_graph *g)
-{
-   GLboolean progress = GL_TRUE;
-   int i;
-
-   while (progress) {
-      progress = GL_FALSE;
-
-      for (i = g->count - 1; i >= 0; i--) {
-	 if (g->nodes[i].in_stack)
-	    continue;
-
-	 if (pq_test(g, i)) {
-	    g->stack[g->stack_count] = i;
-	    g->stack_count++;
-	    g->nodes[i].in_stack = GL_TRUE;
-	    progress = GL_TRUE;
-	 }
-      }
-   }
-
-   for (i = 0; i < g->count; i++) {
-      if (!g->nodes[i].in_stack)
-	 return GL_FALSE;
-   }
-
-   return GL_TRUE;
-}
-
-/**
- * Pops nodes from the stack back into the graph, coloring them with
- * registers as they go.
- *
- * If all nodes were trivially colorable, then this must succeed.  If
- * not (optimistic coloring), then it may return GL_FALSE;
- */
-GLboolean
-ra_select(struct ra_graph *g)
-{
-   int i;
-
-   while (g->stack_count != 0) {
-      unsigned int r;
-      int n = g->stack[g->stack_count - 1];
-      struct ra_class *c = g->regs->classes[g->nodes[n].class];
-
-      /* Find the lowest-numbered reg which is not used by a member
-       * of the graph adjacent to us.
-       */
-      for (r = 0; r < g->regs->count; r++) {
-	 if (!c->regs[r])
-	    continue;
-
-	 /* Check if any of our neighbors conflict with this register choice. */
-	 for (i = 0; i < g->nodes[n].adjacency_count; i++) {
-	    unsigned int n2 = g->nodes[n].adjacency_list[i];
-
-	    if (!g->nodes[n2].in_stack &&
-		g->regs->regs[r].conflicts[g->nodes[n2].reg]) {
-	       break;
-	    }
-	 }
-	 if (i == g->nodes[n].adjacency_count)
-	    break;
-      }
-      if (r == g->regs->count)
-	 return GL_FALSE;
-
-      g->nodes[n].reg = r;
-      g->nodes[n].in_stack = GL_FALSE;
-      g->stack_count--;
-   }
-
-   return GL_TRUE;
-}
-
-/**
- * Optimistic register coloring: Just push the remaining nodes
- * on the stack.  They'll be colored first in ra_select(), and
- * if they succeed then the locally-colorable nodes are still
- * locally-colorable and the rest of the register allocation
- * will succeed.
- */
-void
-ra_optimistic_color(struct ra_graph *g)
-{
-   unsigned int i;
-
-   for (i = 0; i < g->count; i++) {
-      if (g->nodes[i].in_stack)
-	 continue;
-
-      g->stack[g->stack_count] = i;
-      g->stack_count++;
-      g->nodes[i].in_stack = GL_TRUE;
-   }
-}
-
-GLboolean
-ra_allocate_no_spills(struct ra_graph *g)
-{
-   if (!ra_simplify(g)) {
-      ra_optimistic_color(g);
-   }
-   return ra_select(g);
-}
-
-unsigned int
-ra_get_node_reg(struct ra_graph *g, unsigned int n)
-{
-   return g->nodes[n].reg;
-}
-
-static float
-ra_get_spill_benefit(struct ra_graph *g, unsigned int n)
-{
-   int j;
-   float benefit = 0;
-   int n_class = g->nodes[n].class;
-
-   /* Define the benefit of eliminating an interference between n, n2
-    * through spilling as q(C, B) / p(C).  This is similar to the
-    * "count number of edges" approach of traditional graph coloring,
-    * but takes classes into account.
-    */
-   for (j = 0; j < g->nodes[n].adjacency_count; j++) {
-      unsigned int n2 = g->nodes[n].adjacency_list[j];
-      if (n != n2) {
-	 unsigned int n2_class = g->nodes[n2].class;
-	 benefit += ((float)g->regs->classes[n_class]->q[n2_class] /
-		     g->regs->classes[n_class]->p);
-      }
-   }
-
-   return benefit;
-}
-
-/**
- * Returns a node number to be spilled according to the cost/benefit using
- * the pq test, or -1 if there are no spillable nodes.
- */
-int
-ra_get_best_spill_node(struct ra_graph *g)
-{
-   unsigned int best_node = -1;
-   unsigned int best_benefit = 0.0;
-   unsigned int n;
-
-   for (n = 0; n < g->count; n++) {
-      float cost = g->nodes[n].spill_cost;
-      float benefit;
-
-      if (cost <= 0.0)
-	 continue;
-
-      benefit = ra_get_spill_benefit(g, n);
-
-      if (benefit / cost > best_benefit) {
-	 best_benefit = benefit / cost;
-	 best_node = n;
-      }
-   }
-
-   return best_node;
-}
-
-/**
- * Only nodes with a spill cost set (cost != 0.0) will be considered
- * for register spilling.
- */
-void
-ra_set_node_spill_cost(struct ra_graph *g, unsigned int n, float cost)
-{
-   g->nodes[n].spill_cost = cost;
-}
+/*

+ * Copyright © 2010 Intel Corporation

+ *

+ * Permission is hereby granted, free of charge, to any person obtaining a

+ * copy of this software and associated documentation files (the "Software"),

+ * to deal in the Software without restriction, including without limitation

+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,

+ * and/or sell copies of the Software, and to permit persons to whom the

+ * Software is furnished to do so, subject to the following conditions:

+ *

+ * The above copyright notice and this permission notice (including the next

+ * paragraph) shall be included in all copies or substantial portions of the

+ * Software.

+ *

+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL

+ * THE AUTHORS OR COPYRIGHT HOLDERS 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:

+ *    Eric Anholt <eric@anholt.net>

+ *

+ */

+

+/** @file register_allocate.c

+ *

+ * Graph-coloring register allocator.

+ *

+ * The basic idea of graph coloring is to make a node in a graph for

+ * every thing that needs a register (color) number assigned, and make

+ * edges in the graph between nodes that interfere (can't be allocated

+ * to the same register at the same time).

+ *

+ * During the "simplify" process, any any node with fewer edges than

+ * there are registers means that that edge can get assigned a

+ * register regardless of what its neighbors choose, so that node is

+ * pushed on a stack and removed (with its edges) from the graph.

+ * That likely causes other nodes to become trivially colorable as well.

+ *

+ * Then during the "select" process, nodes are popped off of that

+ * stack, their edges restored, and assigned a color different from

+ * their neighbors.  Because they were pushed on the stack only when

+ * they were trivially colorable, any color chosen won't interfere

+ * with the registers to be popped later.

+ *

+ * The downside to most graph coloring is that real hardware often has

+ * limitations, like registers that need to be allocated to a node in

+ * pairs, or aligned on some boundary.  This implementation follows

+ * the paper "Retargetable Graph-Coloring Register Allocation for

+ * Irregular Architectures" by Johan Runeson and Sven-Olof Nyström.

+ *

+ * In this system, there are register classes each containing various

+ * registers, and registers may interfere with other registers.  For

+ * example, one might have a class of base registers, and a class of

+ * aligned register pairs that would each interfere with their pair of

+ * the base registers.  Each node has a register class it needs to be

+ * assigned to.  Define p(B) to be the size of register class B, and

+ * q(B,C) to be the number of registers in B that the worst choice

+ * register in C could conflict with.  Then, this system replaces the

+ * basic graph coloring test of "fewer edges from this node than there

+ * are registers" with "For this node of class B, the sum of q(B,C)

+ * for each neighbor node of class C is less than pB".

+ *

+ * A nice feature of the pq test is that q(B,C) can be computed once

+ * up front and stored in a 2-dimensional array, so that the cost of

+ * coloring a node is constant with the number of registers.  We do

+ * this during ra_set_finalize().

+ */

+

+#include <ralloc.h>

+

+#include "main/imports.h"

+#include "main/macros.h"

+#include "main/mtypes.h"

+#include "register_allocate.h"

+

+struct ra_reg {

+   GLboolean *conflicts;

+   unsigned int *conflict_list;

+   unsigned int conflict_list_size;

+   unsigned int num_conflicts;

+};

+

+struct ra_regs {

+   struct ra_reg *regs;

+   unsigned int count;

+

+   struct ra_class **classes;

+   unsigned int class_count;

+};

+

+struct ra_class {

+   GLboolean *regs;

+

+   /**

+    * p(B) in Runeson/Nyström paper.

+    *

+    * This is "how many regs are in the set."

+    */

+   unsigned int p;

+

+   /**

+    * q(B,C) (indexed by C, B is this register class) in

+    * Runeson/Nyström paper.  This is "how many registers of B could

+    * the worst choice register from C conflict with".

+    */

+   unsigned int *q;

+};

+

+struct ra_node {

+   /** @{

+    *

+    * List of which nodes this node interferes with.  This should be

+    * symmetric with the other node.

+    */

+   GLboolean *adjacency;

+   unsigned int *adjacency_list;

+   unsigned int adjacency_count;

+   /** @} */

+

+   unsigned int class;

+

+   /* Register, if assigned, or ~0. */

+   unsigned int reg;

+

+   /**

+    * Set when the node is in the trivially colorable stack.  When

+    * set, the adjacency to this node is ignored, to implement the

+    * "remove the edge from the graph" in simplification without

+    * having to actually modify the adjacency_list.

+    */

+   GLboolean in_stack;

+

+   /* For an implementation that needs register spilling, this is the

+    * approximate cost of spilling this node.

+    */

+   float spill_cost;

+};

+

+struct ra_graph {

+   struct ra_regs *regs;

+   /**

+    * the variables that need register allocation.

+    */

+   struct ra_node *nodes;

+   unsigned int count; /**< count of nodes. */

+

+   unsigned int *stack;

+   unsigned int stack_count;

+};

+

+struct ra_regs *

+ra_alloc_reg_set(unsigned int count)

+{

+   unsigned int i;

+   struct ra_regs *regs;

+

+   regs = rzalloc(NULL, struct ra_regs);

+   regs->count = count;

+   regs->regs = rzalloc_array(regs, struct ra_reg, count);

+

+   for (i = 0; i < count; i++) {

+      regs->regs[i].conflicts = rzalloc_array(regs->regs, GLboolean, count);

+      regs->regs[i].conflicts[i] = GL_TRUE;

+

+      regs->regs[i].conflict_list = ralloc_array(regs->regs, unsigned int, 4);

+      regs->regs[i].conflict_list_size = 4;

+      regs->regs[i].conflict_list[0] = i;

+      regs->regs[i].num_conflicts = 1;

+   }

+

+   return regs;

+}

+

+static void

+ra_add_conflict_list(struct ra_regs *regs, unsigned int r1, unsigned int r2)

+{

+   struct ra_reg *reg1 = &regs->regs[r1];

+

+   if (reg1->conflict_list_size == reg1->num_conflicts) {

+      reg1->conflict_list_size *= 2;

+      reg1->conflict_list = reralloc(regs->regs, reg1->conflict_list,

+				     unsigned int, reg1->conflict_list_size);

+   }

+   reg1->conflict_list[reg1->num_conflicts++] = r2;

+   reg1->conflicts[r2] = GL_TRUE;

+}

+

+void

+ra_add_reg_conflict(struct ra_regs *regs, unsigned int r1, unsigned int r2)

+{

+   if (!regs->regs[r1].conflicts[r2]) {

+      ra_add_conflict_list(regs, r1, r2);

+      ra_add_conflict_list(regs, r2, r1);

+   }

+}

+

+unsigned int

+ra_alloc_reg_class(struct ra_regs *regs)

+{

+   struct ra_class *class;

+

+   regs->classes = reralloc(regs->regs, regs->classes, struct ra_class *,

+			    regs->class_count + 1);

+

+   class = rzalloc(regs, struct ra_class);

+   regs->classes[regs->class_count] = class;

+

+   class->regs = rzalloc_array(class, GLboolean, regs->count);

+

+   return regs->class_count++;

+}

+

+void

+ra_class_add_reg(struct ra_regs *regs, unsigned int c, unsigned int r)

+{

+   struct ra_class *class = regs->classes[c];

+

+   class->regs[r] = GL_TRUE;

+   class->p++;

+}

+

+/**

+ * Must be called after all conflicts and register classes have been

+ * set up and before the register set is used for allocation.

+ */

+void

+ra_set_finalize(struct ra_regs *regs)

+{

+   unsigned int b, c;

+

+   for (b = 0; b < regs->class_count; b++) {

+      regs->classes[b]->q = ralloc_array(regs, unsigned int, regs->class_count);

+   }

+

+   /* Compute, for each class B and C, how many regs of B an

+    * allocation to C could conflict with.

+    */

+   for (b = 0; b < regs->class_count; b++) {

+      for (c = 0; c < regs->class_count; c++) {

+	 unsigned int rc;

+	 int max_conflicts = 0;

+

+	 for (rc = 0; rc < regs->count; rc++) {

+	    int conflicts = 0;

+	    int i;

+

+	    if (!regs->classes[c]->regs[rc])

+	       continue;

+

+	    for (i = 0; i < regs->regs[rc].num_conflicts; i++) {

+	       unsigned int rb = regs->regs[rc].conflict_list[i];

+	       if (regs->classes[b]->regs[rb])

+		  conflicts++;

+	    }

+	    max_conflicts = MAX2(max_conflicts, conflicts);

+	 }

+	 regs->classes[b]->q[c] = max_conflicts;

+      }

+   }

+}

+

+static void

+ra_add_node_adjacency(struct ra_graph *g, unsigned int n1, unsigned int n2)

+{

+   g->nodes[n1].adjacency[n2] = GL_TRUE;

+   g->nodes[n1].adjacency_list[g->nodes[n1].adjacency_count] = n2;

+   g->nodes[n1].adjacency_count++;

+}

+

+struct ra_graph *

+ra_alloc_interference_graph(struct ra_regs *regs, unsigned int count)

+{

+   struct ra_graph *g;

+   unsigned int i;

+

+   g = rzalloc(regs, struct ra_graph);

+   g->regs = regs;

+   g->nodes = rzalloc_array(g, struct ra_node, count);

+   g->count = count;

+

+   g->stack = rzalloc_array(g, unsigned int, count);

+

+   for (i = 0; i < count; i++) {

+      g->nodes[i].adjacency = rzalloc_array(g, GLboolean, count);

+      g->nodes[i].adjacency_list = ralloc_array(g, unsigned int, count);

+      g->nodes[i].adjacency_count = 0;

+      ra_add_node_adjacency(g, i, i);

+      g->nodes[i].reg = ~0;

+   }

+

+   return g;

+}

+

+void

+ra_set_node_class(struct ra_graph *g,

+		  unsigned int n, unsigned int class)

+{

+   g->nodes[n].class = class;

+}

+

+void

+ra_add_node_interference(struct ra_graph *g,

+			 unsigned int n1, unsigned int n2)

+{

+   if (!g->nodes[n1].adjacency[n2]) {

+      ra_add_node_adjacency(g, n1, n2);

+      ra_add_node_adjacency(g, n2, n1);

+   }

+}

+

+static GLboolean pq_test(struct ra_graph *g, unsigned int n)

+{

+   unsigned int j;

+   unsigned int q = 0;

+   int n_class = g->nodes[n].class;

+

+   for (j = 0; j < g->nodes[n].adjacency_count; j++) {

+      unsigned int n2 = g->nodes[n].adjacency_list[j];

+      unsigned int n2_class = g->nodes[n2].class;

+

+      if (n != n2 && !g->nodes[n2].in_stack) {

+	 q += g->regs->classes[n_class]->q[n2_class];

+      }

+   }

+

+   return q < g->regs->classes[n_class]->p;

+}

+

+/**

+ * Simplifies the interference graph by pushing all

+ * trivially-colorable nodes into a stack of nodes to be colored,

+ * removing them from the graph, and rinsing and repeating.

+ *

+ * Returns GL_TRUE if all nodes were removed from the graph.  GL_FALSE

+ * means that either spilling will be required, or optimistic coloring

+ * should be applied.

+ */

+GLboolean

+ra_simplify(struct ra_graph *g)

+{

+   GLboolean progress = GL_TRUE;

+   int i;

+

+   while (progress) {

+      progress = GL_FALSE;

+

+      for (i = g->count - 1; i >= 0; i--) {

+	 if (g->nodes[i].in_stack)

+	    continue;

+

+	 if (pq_test(g, i)) {

+	    g->stack[g->stack_count] = i;

+	    g->stack_count++;

+	    g->nodes[i].in_stack = GL_TRUE;

+	    progress = GL_TRUE;

+	 }

+      }

+   }

+

+   for (i = 0; i < g->count; i++) {

+      if (!g->nodes[i].in_stack)

+	 return GL_FALSE;

+   }

+

+   return GL_TRUE;

+}

+

+/**

+ * Pops nodes from the stack back into the graph, coloring them with

+ * registers as they go.

+ *

+ * If all nodes were trivially colorable, then this must succeed.  If

+ * not (optimistic coloring), then it may return GL_FALSE;

+ */

+GLboolean

+ra_select(struct ra_graph *g)

+{

+   int i;

+

+   while (g->stack_count != 0) {

+      unsigned int r;

+      int n = g->stack[g->stack_count - 1];

+      struct ra_class *c = g->regs->classes[g->nodes[n].class];

+

+      /* Find the lowest-numbered reg which is not used by a member

+       * of the graph adjacent to us.

+       */

+      for (r = 0; r < g->regs->count; r++) {

+	 if (!c->regs[r])

+	    continue;

+

+	 /* Check if any of our neighbors conflict with this register choice. */

+	 for (i = 0; i < g->nodes[n].adjacency_count; i++) {

+	    unsigned int n2 = g->nodes[n].adjacency_list[i];

+

+	    if (!g->nodes[n2].in_stack &&

+		g->regs->regs[r].conflicts[g->nodes[n2].reg]) {

+	       break;

+	    }

+	 }

+	 if (i == g->nodes[n].adjacency_count)

+	    break;

+      }

+      if (r == g->regs->count)

+	 return GL_FALSE;

+

+      g->nodes[n].reg = r;

+      g->nodes[n].in_stack = GL_FALSE;

+      g->stack_count--;

+   }

+

+   return GL_TRUE;

+}

+

+/**

+ * Optimistic register coloring: Just push the remaining nodes

+ * on the stack.  They'll be colored first in ra_select(), and

+ * if they succeed then the locally-colorable nodes are still

+ * locally-colorable and the rest of the register allocation

+ * will succeed.

+ */

+void

+ra_optimistic_color(struct ra_graph *g)

+{

+   unsigned int i;

+

+   for (i = 0; i < g->count; i++) {

+      if (g->nodes[i].in_stack)

+	 continue;

+

+      g->stack[g->stack_count] = i;

+      g->stack_count++;

+      g->nodes[i].in_stack = GL_TRUE;

+   }

+}

+

+GLboolean

+ra_allocate_no_spills(struct ra_graph *g)

+{

+   if (!ra_simplify(g)) {

+      ra_optimistic_color(g);

+   }

+   return ra_select(g);

+}

+

+unsigned int

+ra_get_node_reg(struct ra_graph *g, unsigned int n)

+{

+   return g->nodes[n].reg;

+}

+

+static float

+ra_get_spill_benefit(struct ra_graph *g, unsigned int n)

+{

+   int j;

+   float benefit = 0;

+   int n_class = g->nodes[n].class;

+

+   /* Define the benefit of eliminating an interference between n, n2

+    * through spilling as q(C, B) / p(C).  This is similar to the

+    * "count number of edges" approach of traditional graph coloring,

+    * but takes classes into account.

+    */

+   for (j = 0; j < g->nodes[n].adjacency_count; j++) {

+      unsigned int n2 = g->nodes[n].adjacency_list[j];

+      if (n != n2) {

+	 unsigned int n2_class = g->nodes[n2].class;

+	 benefit += ((float)g->regs->classes[n_class]->q[n2_class] /

+		     g->regs->classes[n_class]->p);

+      }

+   }

+

+   return benefit;

+}

+

+/**

+ * Returns a node number to be spilled according to the cost/benefit using

+ * the pq test, or -1 if there are no spillable nodes.

+ */

+int

+ra_get_best_spill_node(struct ra_graph *g)

+{

+   unsigned int best_node = -1;

+   unsigned int best_benefit = 0.0;

+   unsigned int n;

+

+   for (n = 0; n < g->count; n++) {

+      float cost = g->nodes[n].spill_cost;

+      float benefit;

+

+      if (cost <= 0.0)

+	 continue;

+

+      benefit = ra_get_spill_benefit(g, n);

+

+      if (benefit / cost > best_benefit) {

+	 best_benefit = benefit / cost;

+	 best_node = n;

+      }

+   }

+

+   return best_node;

+}

+

+/**

+ * Only nodes with a spill cost set (cost != 0.0) will be considered

+ * for register spilling.

+ */

+void

+ra_set_node_spill_cost(struct ra_graph *g, unsigned int n, float cost)

+{

+   g->nodes[n].spill_cost = cost;

+}

diff --git a/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c b/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
index f159ed2db..9cc40c3d9 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -372,6 +372,14 @@ internal_format(struct gl_context *ctx, GLenum format, GLenum type)
             return

                ctx->Extensions.ARB_texture_float ? GL_RGBA32F :

                ctx->Extensions.EXT_texture_snorm ? GL_RGBA16_SNORM : GL_RGBA16;

+

+         case GL_UNSIGNED_INT_5_9_9_9_REV:

+            assert(ctx->Extensions.EXT_texture_shared_exponent);

+            return GL_RGB9_E5;

+

+         case GL_UNSIGNED_INT_10F_11F_11F_REV:

+            assert(ctx->Extensions.EXT_packed_float);

+            return GL_R11F_G11F_B10F;

          }

       }

    }

diff --git a/mesalib/src/mesa/state_tracker/st_extensions.c b/mesalib/src/mesa/state_tracker/st_extensions.c
index 8f697fa51..0c5e1991c 100644
--- a/mesalib/src/mesa/state_tracker/st_extensions.c
+++ b/mesalib/src/mesa/state_tracker/st_extensions.c
@@ -515,6 +515,12 @@ void st_init_extensions(struct st_context *st)
       ctx->Extensions.ARB_half_float_vertex = GL_TRUE;

    }

 

+   if (screen->is_format_supported(screen, PIPE_FORMAT_R32G32B32A32_FIXED,

+                                   PIPE_BUFFER, 0,

+                                   PIPE_BIND_VERTEX_BUFFER)) {

+      ctx->Extensions.ARB_ES2_compatibility = GL_TRUE;

+   }

+

    if (screen->get_shader_param(screen, PIPE_SHADER_GEOMETRY, PIPE_SHADER_CAP_MAX_INSTRUCTIONS) > 0) {

 #if 0 /* XXX re-enable when GLSL compiler again supports geometry shaders */

       ctx->Extensions.ARB_geometry_shader4 = GL_TRUE;

@@ -576,4 +582,17 @@ void st_init_extensions(struct st_context *st)
    if (st->pipe->texture_barrier) {

       ctx->Extensions.NV_texture_barrier = GL_TRUE;

    }

+

+   if (screen->is_format_supported(screen, PIPE_FORMAT_R9G9B9E5_FLOAT,

+                                   PIPE_TEXTURE_2D, 0,

+                                   PIPE_BIND_SAMPLER_VIEW)) {

+      ctx->Extensions.EXT_texture_shared_exponent = GL_TRUE;

+   }

+

+   if (screen->is_format_supported(screen, PIPE_FORMAT_R11G11B10_FLOAT,

+                                   PIPE_TEXTURE_2D, 0,

+                                   PIPE_BIND_RENDER_TARGET |

+                                   PIPE_BIND_SAMPLER_VIEW)) {

+      ctx->Extensions.EXT_packed_float = GL_TRUE;

+   }

 }

diff --git a/mesalib/src/mesa/state_tracker/st_format.c b/mesalib/src/mesa/state_tracker/st_format.c
index 32b44b806..8568c68e1 100644
--- a/mesalib/src/mesa/state_tracker/st_format.c
+++ b/mesalib/src/mesa/state_tracker/st_format.c
@@ -325,6 +325,11 @@ st_mesa_format_to_pipe_format(gl_format mesaFormat)
    case MESA_FORMAT_SIGNED_I16:

       return PIPE_FORMAT_I16_SNORM;

 

+   case MESA_FORMAT_RGB9_E5_FLOAT:

+      return PIPE_FORMAT_R9G9B9E5_FLOAT;

+   case MESA_FORMAT_R11_G11_B10_FLOAT:

+      return PIPE_FORMAT_R11G11B10_FLOAT;

+

    default:

       assert(0);

       return PIPE_FORMAT_NONE;

@@ -545,6 +550,11 @@ st_pipe_format_to_mesa_format(enum pipe_format format)
    case PIPE_FORMAT_I16_SNORM:

       return MESA_FORMAT_SIGNED_I16;

 

+   case PIPE_FORMAT_R9G9B9E5_FLOAT:

+      return MESA_FORMAT_RGB9_E5_FLOAT;

+   case PIPE_FORMAT_R11G11B10_FLOAT:

+      return MESA_FORMAT_R11_G11_B10_FLOAT;

+

    default:

       assert(0);

       return MESA_FORMAT_NONE;

@@ -1483,6 +1493,20 @@ st_choose_format(struct pipe_screen *screen, GLenum internalFormat,
                target, sample_count, bindings);

       }

 

+   case GL_RGB9_E5:

+      if (screen->is_format_supported(screen, PIPE_FORMAT_R9G9B9E5_FLOAT, target,

+                                      sample_count, bindings)) {

+         return PIPE_FORMAT_R9G9B9E5_FLOAT;

+      }

+      return PIPE_FORMAT_NONE;

+

+   case GL_R11F_G11F_B10F:

+      if (screen->is_format_supported(screen, PIPE_FORMAT_R11G11B10_FLOAT, target,

+                                      sample_count, bindings)) {

+         return PIPE_FORMAT_R11G11B10_FLOAT;

+      }

+      return PIPE_FORMAT_NONE;

+

    default:

       return PIPE_FORMAT_NONE;

    }