Merge remote-tracking branch 'origin/released'

* origin/released:
  fontconfig mesa pixman xserver git update 25 Mar 2013

31 files changed, 733 insertions, 259 deletions

diff --git a/mesalib/src/mesa/drivers/common/meta.c b/mesalib/src/mesa/drivers/common/meta.c
index 29a209ede..8114550ba 100644
--- a/mesalib/src/mesa/drivers/common/meta.c
+++ b/mesalib/src/mesa/drivers/common/meta.c
@@ -3118,6 +3118,7 @@ setup_texture_coords(GLenum faceTarget,
                      GLint slice,
                      GLint width,
                      GLint height,
+                     GLint depth,
                      GLfloat coords0[3],
                      GLfloat coords1[3],
                      GLfloat coords2[3],
@@ -3134,8 +3135,11 @@ setup_texture_coords(GLenum faceTarget,
    case GL_TEXTURE_2D:
    case GL_TEXTURE_3D:
    case GL_TEXTURE_2D_ARRAY:
-      if (faceTarget == GL_TEXTURE_3D)
-         r = 1.0F / slice;
+      if (faceTarget == GL_TEXTURE_3D) {
+         assert(slice < depth);
+         assert(depth >= 1);
+         r = (slice + 0.5f) / depth;
+      }
       else if (faceTarget == GL_TEXTURE_2D_ARRAY)
          r = slice;
       else
@@ -3571,10 +3575,10 @@ _mesa_meta_GenerateMipmap(struct gl_context *ctx, GLenum target,
    else
       assert(!genMipmapSave);
 
-  /* Setup texture coordinates */
+   /* Setup texture coordinates */
    setup_texture_coords(faceTarget,
                         slice,
-                        0, 0, /* width, height never used here */
+                        0, 0, 1, /* width, height never used here */
                         verts[0].tex,
                         verts[1].tex,
                         verts[2].tex,
@@ -3840,6 +3844,7 @@ decompress_texture_image(struct gl_context *ctx,
    struct gl_texture_object *texObj = texImage->TexObject;
    const GLint width = texImage->Width;
    const GLint height = texImage->Height;
+   const GLint depth = texImage->Height;
    const GLenum target = texObj->Target;
    GLenum faceTarget;
    struct vertex {
@@ -3935,7 +3940,7 @@ decompress_texture_image(struct gl_context *ctx,
       _mesa_BindSampler(ctx->Texture.CurrentUnit, decompress->Sampler);
    }
 
-   setup_texture_coords(faceTarget, slice, width, height,
+   setup_texture_coords(faceTarget, slice, width, height, depth,
                         verts[0].tex,
                         verts[1].tex,
                         verts[2].tex,
diff --git a/mesalib/src/mesa/drivers/dri/common/xmlpool/t_options.h b/mesalib/src/mesa/drivers/dri/common/xmlpool/t_options.h
index 1e7eced06..7b441c68f 100644
--- a/mesalib/src/mesa/drivers/dri/common/xmlpool/t_options.h
+++ b/mesalib/src/mesa/drivers/dri/common/xmlpool/t_options.h
@@ -75,6 +75,11 @@ DRI_CONF_OPT_BEGIN(always_flush_cache,bool,def) \
         DRI_CONF_DESC(en,gettext("Enable flushing GPU caches with each draw call")) \
 DRI_CONF_OPT_END
 
+#define DRI_CONF_DISABLE_THROTTLING(def) \
+DRI_CONF_OPT_BEGIN(disable_throttling,bool,def) \
+	DRI_CONF_DESC(en,gettext("Disable throttling on first batch after flush")) \
+DRI_CONF_OPT_END
+
 #define DRI_CONF_FORCE_GLSL_EXTENSIONS_WARN(def) \
 DRI_CONF_OPT_BEGIN(force_glsl_extensions_warn,bool,def) \
         DRI_CONF_DESC(en,gettext("Force GLSL extension default behavior to 'warn'")) \
diff --git a/mesalib/src/mesa/main/compiler.h b/mesalib/src/mesa/main/compiler.h
index 48712485a..8b23665e6 100644
--- a/mesalib/src/mesa/main/compiler.h
+++ b/mesalib/src/mesa/main/compiler.h
@@ -307,8 +307,9 @@ static INLINE GLuint CPU_TO_LE32(GLuint x)
  * USE_IEEE: Determine if we're using IEEE floating point
  */
 #if defined(__i386__) || defined(__386__) || defined(__sparc__) || \
-    defined(__s390x__) || defined(__powerpc__) || \
+    defined(__s390__) || defined(__s390x__) || defined(__powerpc__) || \
     defined(__x86_64__) || \
+    defined(__m68k__) || \
     defined(ia64) || defined(__ia64__) || \
     defined(__hppa__) || defined(hpux) || \
     defined(__mips) || defined(_MIPS_ARCH) || \
diff --git a/mesalib/src/mesa/main/dd.h b/mesalib/src/mesa/main/dd.h
index 4860d4d12..8f3cd3d6b 100644
--- a/mesalib/src/mesa/main/dd.h
+++ b/mesalib/src/mesa/main/dd.h
@@ -195,9 +195,10 @@ struct dd_function_table {
                                      GLenum srcFormat, GLenum srcType );
 
    /**
-    * Determine sample counts support for a particular format
+    * Determine sample counts support for a particular target and format
     *
     * \param ctx            GL context
+    * \param target         GL target enum
     * \param internalFormat GL format enum
     * \param samples        Buffer to hold the returned sample counts.
     *                       Drivers \b must \b not return more than 16 counts.
@@ -207,6 +208,7 @@ struct dd_function_table {
     * \c internaFormat is not renderable, zero is returned.
     */
    size_t (*QuerySamplesForFormat)(struct gl_context *ctx,
+                                   GLenum target,
                                    GLenum internalFormat,
                                    int samples[16]);
 
diff --git a/mesalib/src/mesa/main/fbobject.c b/mesalib/src/mesa/main/fbobject.c
index 0126e2930..3fdf62667 100644
--- a/mesalib/src/mesa/main/fbobject.c
+++ b/mesalib/src/mesa/main/fbobject.c
@@ -43,6 +43,7 @@
 #include "hash.h"
 #include "macros.h"
 #include "mfeatures.h"
+#include "multisample.h"
 #include "mtypes.h"
 #include "renderbuffer.h"
 #include "state.h"
@@ -474,6 +475,32 @@ _mesa_validate_framebuffer(struct gl_context *ctx, struct gl_framebuffer *fb)
 
 
 /**
+ * Return true if the framebuffer has a combined depth/stencil
+ * renderbuffer attached.
+ */
+GLboolean
+_mesa_has_depthstencil_combined(const struct gl_framebuffer *fb)
+{
+   const struct gl_renderbuffer_attachment *depth =
+         &fb->Attachment[BUFFER_DEPTH];
+   const struct gl_renderbuffer_attachment *stencil =
+         &fb->Attachment[BUFFER_STENCIL];
+
+   if (depth->Type == stencil->Type) {
+      if (depth->Type == GL_RENDERBUFFER_EXT &&
+          depth->Renderbuffer == stencil->Renderbuffer)
+         return GL_TRUE;
+
+      if (depth->Type == GL_TEXTURE &&
+          depth->Texture == stencil->Texture)
+         return GL_TRUE;
+   }
+
+   return GL_FALSE;
+}
+
+
+/**
  * For debug only.
  */
 static void
@@ -1466,6 +1493,7 @@ renderbuffer_storage(GLenum target, GLenum internalFormat,
       "glRenderbufferStorage" : "glRenderbufferStorageMultisample";
    struct gl_renderbuffer *rb;
    GLenum baseFormat;
+   GLenum sample_count_error;
    GET_CURRENT_CONTEXT(ctx);
 
    if (MESA_VERBOSE & VERBOSE_API) {
@@ -1509,9 +1537,14 @@ renderbuffer_storage(GLenum target, GLenum internalFormat,
       /* NumSamples == 0 indicates non-multisampling */
       samples = 0;
    }
-   else if (samples > (GLsizei) ctx->Const.MaxSamples) {
-      /* note: driver may choose to use more samples than what's requested */
-      _mesa_error(ctx, GL_INVALID_VALUE, "%s(samples)", func);
+
+   /* check the sample count;
+    * note: driver may choose to use more samples than what's requested
+    */
+   sample_count_error = _mesa_check_sample_count(ctx, target,
+         internalFormat, samples);
+   if (sample_count_error != GL_NO_ERROR) {
+      _mesa_error(ctx, sample_count_error, "%s(samples)", func);
       return;
    }
 
diff --git a/mesalib/src/mesa/main/fbobject.h b/mesalib/src/mesa/main/fbobject.h
index ec8b0afe4..0358864d7 100644
--- a/mesalib/src/mesa/main/fbobject.h
+++ b/mesalib/src/mesa/main/fbobject.h
@@ -113,6 +113,9 @@ _mesa_framebuffer_renderbuffer(struct gl_context *ctx,
 extern void
 _mesa_validate_framebuffer(struct gl_context *ctx, struct gl_framebuffer *fb);
 
+extern GLboolean
+_mesa_has_depthstencil_combined(const struct gl_framebuffer *fb);
+
 extern void
 _mesa_test_framebuffer_completeness(struct gl_context *ctx,
                                     struct gl_framebuffer *fb);
diff --git a/mesalib/src/mesa/main/ff_fragment_shader.cpp b/mesalib/src/mesa/main/ff_fragment_shader.cpp
index 186988bbd..01a4542d7 100644
--- a/mesalib/src/mesa/main/ff_fragment_shader.cpp
+++ b/mesalib/src/mesa/main/ff_fragment_shader.cpp
@@ -1350,22 +1350,6 @@ create_new_program(struct gl_context *ctx, struct state_key *key)
 
    _mesa_glsl_link_shader(ctx, p.shader_program);
 
-   /* Set the sampler uniforms, and relink to get them into the linked
-    * program.
-    */
-   struct gl_shader *const fs =
-      p.shader_program->_LinkedShaders[MESA_SHADER_FRAGMENT];
-   struct gl_program *const fp = fs->Program;
-
-   _mesa_generate_parameters_list_for_uniforms(p.shader_program, fs,
-					       fp->Parameters);
-
-   _mesa_associate_uniform_storage(ctx, p.shader_program, fp->Parameters);
-
-   _mesa_update_shader_textures_used(p.shader_program, fp);
-   if (ctx->Driver.SamplerUniformChange)
-      ctx->Driver.SamplerUniformChange(ctx, fp->Target, fp);
-
    if (!p.shader_program->LinkStatus)
       _mesa_problem(ctx, "Failed to link fixed function fragment shader: %s\n",
 		    p.shader_program->InfoLog);
diff --git a/mesalib/src/mesa/main/formatquery.c b/mesalib/src/mesa/main/formatquery.c
index bd895e874..78c5fbe5e 100644
--- a/mesalib/src/mesa/main/formatquery.c
+++ b/mesalib/src/mesa/main/formatquery.c
@@ -59,9 +59,10 @@ _mesa_GetInternalformativ(GLenum target, GLenum internalformat, GLenum pname,
 
    case GL_TEXTURE_2D_MULTISAMPLE:
    case GL_TEXTURE_2D_MULTISAMPLE_ARRAY:
-      /* Mesa does not currently support GL_ARB_texture_multisample, so these
-       * enums are not valid on this implementation either.
-       */
+      /* These enums are only valid if ARB_texture_multisample is supported */
+      if (_mesa_is_desktop_gl(ctx) && ctx->Extensions.ARB_texture_multisample)
+         break;
+
    default:
       _mesa_error(ctx, GL_INVALID_ENUM,
                   "glGetInternalformativ(target=%s)",
@@ -96,7 +97,8 @@ _mesa_GetInternalformativ(GLenum target, GLenum internalformat, GLenum pname,
 
    switch (pname) {
    case GL_SAMPLES:
-      count = ctx->Driver.QuerySamplesForFormat(ctx, internalformat, buffer);
+      count = ctx->Driver.QuerySamplesForFormat(ctx, target,
+            internalformat, buffer);
       break;
    case GL_NUM_SAMPLE_COUNTS: {
       /* The driver can return 0, and we should pass that along to the
@@ -115,7 +117,7 @@ _mesa_GetInternalformativ(GLenum target, GLenum internalformat, GLenum pname,
        *          returned."
        */
       const size_t num_samples =
-         ctx->Driver.QuerySamplesForFormat(ctx, internalformat, buffer);
+         ctx->Driver.QuerySamplesForFormat(ctx, target, internalformat, buffer);
 
       /* QuerySamplesForFormat writes some stuff to buffer, so we have to
        * separately over-write it with the requested value.
diff --git a/mesalib/src/mesa/main/framebuffer.c b/mesalib/src/mesa/main/framebuffer.c
index d3abc2b30..619aaa337 100644
--- a/mesalib/src/mesa/main/framebuffer.c
+++ b/mesalib/src/mesa/main/framebuffer.c
@@ -923,10 +923,10 @@ _mesa_get_color_read_type(struct gl_context *ctx)
  * Returns the read renderbuffer for the specified format.
  */
 struct gl_renderbuffer *
-_mesa_get_read_renderbuffer_for_format(struct gl_context *ctx,
+_mesa_get_read_renderbuffer_for_format(const struct gl_context *ctx,
                                        GLenum format)
 {
-   struct gl_framebuffer *rfb = ctx->ReadBuffer;
+   const struct gl_framebuffer *rfb = ctx->ReadBuffer;
 
    if (_mesa_is_color_format(format)) {
       return rfb->Attachment[rfb->_ColorReadBufferIndex].Renderbuffer;
diff --git a/mesalib/src/mesa/main/framebuffer.h b/mesalib/src/mesa/main/framebuffer.h
index 06db04925..9b94452d6 100644
--- a/mesalib/src/mesa/main/framebuffer.h
+++ b/mesalib/src/mesa/main/framebuffer.h
@@ -98,7 +98,7 @@ extern GLenum
 _mesa_get_color_read_format(struct gl_context *ctx);
 
 extern struct gl_renderbuffer *
-_mesa_get_read_renderbuffer_for_format(struct gl_context *ctx,
+_mesa_get_read_renderbuffer_for_format(const struct gl_context *ctx,
                                        GLenum format);
 
 extern void
diff --git a/mesalib/src/mesa/main/mtypes.h b/mesalib/src/mesa/main/mtypes.h
index 83b6c8984..a0e7e281d 100644
--- a/mesalib/src/mesa/main/mtypes.h
+++ b/mesalib/src/mesa/main/mtypes.h
@@ -1179,6 +1179,7 @@ struct gl_texture_object
    GLfloat Priority;		/**< in [0,1] */
    GLint BaseLevel;		/**< min mipmap level, OpenGL 1.2 */
    GLint MaxLevel;		/**< max mipmap level, OpenGL 1.2 */
+   GLint ImmutableLevels;       /**< ES 3.0 / ARB_texture_view */
    GLint _MaxLevel;		/**< actual max mipmap level (q in the spec) */
    GLfloat _MaxLambda;		/**< = _MaxLevel - BaseLevel (q - b in spec) */
    GLint CropRect[4];           /**< GL_OES_draw_texture */
@@ -1796,6 +1797,7 @@ struct gl_transform_feedback_state
 typedef enum
 {
    PROGRAM_TEMPORARY,   /**< machine->Temporary[] */
+   PROGRAM_ARRAY,       /**< Arrays & Matrixes */
    PROGRAM_INPUT,       /**< machine->Inputs[] */
    PROGRAM_OUTPUT,      /**< machine->Outputs[] */
    PROGRAM_LOCAL_PARAM, /**< gl_program->LocalParams[] */
diff --git a/mesalib/src/mesa/main/multisample.c b/mesalib/src/mesa/main/multisample.c
index 248494615..b0f45d933 100644
--- a/mesalib/src/mesa/main/multisample.c
+++ b/mesalib/src/mesa/main/multisample.c
@@ -29,6 +29,7 @@
 #include "main/multisample.h"
 #include "main/mtypes.h"
 #include "main/fbobject.h"
+#include "main/glformats.h"
 
 
 /**
@@ -112,3 +113,79 @@ _mesa_SampleMaski(GLuint index, GLbitfield mask)
    FLUSH_VERTICES(ctx, _NEW_MULTISAMPLE);
    ctx->Multisample.SampleMaskValue = mask;
 }
+
+
+/* Helper for checking a requested sample count against the limit
+ * for a particular (target, internalFormat) pair. The limit imposed,
+ * and the error generated, both depend on which extensions are supported.
+ *
+ * Returns a GL error enum, or GL_NO_ERROR if the requested sample count is
+ * acceptable.
+ */
+GLenum
+_mesa_check_sample_count(struct gl_context *ctx, GLenum target,
+                         GLenum internalFormat, GLsizei samples)
+{
+   /* If ARB_internalformat_query is supported, then treat its highest returned sample
+    * count as the absolute maximum for this format; it is allowed to exceed MAX_SAMPLES.
+    *
+    * From the ARB_internalformat_query spec:
+    *
+    * "If <samples is greater than the maximum number of samples supported
+    * for <internalformat> then the error INVALID_OPERATION is generated."
+    */
+   if (ctx->Extensions.ARB_internalformat_query) {
+      GLint buffer[16];
+      int count = ctx->Driver.QuerySamplesForFormat(ctx, target, internalFormat, buffer);
+      int limit = count ? buffer[0] : -1;
+
+      return samples > limit ? GL_INVALID_OPERATION : GL_NO_ERROR;
+   }
+
+   /* If ARB_texture_multisample is supported, we have separate limits,
+    * which may be lower than MAX_SAMPLES:
+    *
+    * From the ARB_texture_multisample spec, when describing the operation
+    * of RenderbufferStorageMultisample:
+    *
+    * "If <internalformat> is a signed or unsigned integer format and
+    * <samples> is greater than the value of MAX_INTEGER_SAMPLES, then the
+    * error INVALID_OPERATION is generated"
+    *
+    * And when describing the operation of TexImage*Multisample:
+    *
+    * "The error INVALID_OPERATION may be generated if any of the following are true:
+    *
+    * * <internalformat> is a depth/stencil-renderable format and <samples>
+    *   is greater than the value of MAX_DEPTH_TEXTURE_SAMPLES
+    * * <internalformat> is a color-renderable format and <samples> is
+    *   grater than the value of MAX_COLOR_TEXTURE_SAMPLES
+    * * <internalformat> is a signed or unsigned integer format and
+    *   <samples> is greater than the value of MAX_INTEGER_SAMPLES
+    */
+
+   if (ctx->Extensions.ARB_texture_multisample) {
+      if (_mesa_is_enum_format_integer(internalFormat))
+         return samples > ctx->Const.MaxIntegerSamples ? GL_INVALID_OPERATION : GL_NO_ERROR;
+
+      if (target == GL_TEXTURE_2D_MULTISAMPLE ||
+          target == GL_TEXTURE_2D_MULTISAMPLE_ARRAY) {
+
+         if (_mesa_is_depth_or_stencil_format(internalFormat))
+            return samples > ctx->Const.MaxDepthTextureSamples
+               ? GL_INVALID_OPERATION : GL_NO_ERROR;
+         else
+            return samples > ctx->Const.MaxColorTextureSamples
+               ? GL_INVALID_OPERATION : GL_NO_ERROR;
+      }
+   }
+
+   /* No more specific limit is available, so just use MAX_SAMPLES:
+    *
+    * On p205 of the GL3.1 spec:
+    *
+    * "... or if samples is greater than MAX_SAMPLES, then the error
+    * INVALID_VALUE is generated"
+    */
+   return samples > ctx->Const.MaxSamples ? GL_INVALID_VALUE : GL_NO_ERROR;
+}
diff --git a/mesalib/src/mesa/main/multisample.h b/mesalib/src/mesa/main/multisample.h
index 9e6b8e0d3..f2f01de5c 100644
--- a/mesalib/src/mesa/main/multisample.h
+++ b/mesalib/src/mesa/main/multisample.h
@@ -44,4 +44,9 @@ _mesa_GetMultisamplefv(GLenum pname, GLuint index, GLfloat* val);
 extern void GLAPIENTRY
 _mesa_SampleMaski(GLuint index, GLbitfield mask);
 
+
+extern GLenum
+_mesa_check_sample_count(struct gl_context *ctx, GLenum target,
+                   GLenum internalFormat, GLsizei samples);
+
 #endif
diff --git a/mesalib/src/mesa/main/readpix.c b/mesalib/src/mesa/main/readpix.c
index 2f130ae9a..d3d09dea3 100644
--- a/mesalib/src/mesa/main/readpix.c
+++ b/mesalib/src/mesa/main/readpix.c
@@ -41,11 +41,212 @@
 
 
 /**
- * Tries to implement glReadPixels() of GL_DEPTH_COMPONENT using memcpy of the
- * mapping.
+ * Return true if the conversion L=R+G+B is needed.
  */
 static GLboolean
-fast_read_depth_pixels( struct gl_context *ctx,
+need_rgb_to_luminance_conversion(gl_format texFormat, GLenum format)
+{
+   GLenum baseTexFormat = _mesa_get_format_base_format(texFormat);
+
+   return (baseTexFormat == GL_RG ||
+           baseTexFormat == GL_RGB ||
+           baseTexFormat == GL_RGBA) &&
+          (format == GL_LUMINANCE || format == GL_LUMINANCE_ALPHA);
+}
+
+
+/**
+ * Return transfer op flags for this ReadPixels operation.
+ */
+static GLbitfield
+get_readpixels_transfer_ops(const struct gl_context *ctx, gl_format texFormat,
+                            GLenum format, GLenum type, GLboolean uses_blit)
+{
+   GLbitfield transferOps = ctx->_ImageTransferState;
+
+   if (format == GL_DEPTH_COMPONENT ||
+       format == GL_DEPTH_STENCIL ||
+       format == GL_STENCIL_INDEX) {
+      return 0;
+   }
+
+   /* Pixel transfer ops (scale, bias, table lookup) do not apply
+    * to integer formats.
+    */
+   if (_mesa_is_enum_format_integer(format)) {
+      return 0;
+   }
+
+   if (uses_blit) {
+      /* For blit-based ReadPixels packing, the clamping is done automatically
+       * unless the type is float. */
+      if (ctx->Color._ClampReadColor == GL_TRUE &&
+          (type == GL_FLOAT || type == GL_HALF_FLOAT)) {
+         transferOps |= IMAGE_CLAMP_BIT;
+      }
+   }
+   else {
+      /* For CPU-based ReadPixels packing, the clamping must always be done
+       * for non-float types, */
+      if (ctx->Color._ClampReadColor == GL_TRUE ||
+          (type != GL_FLOAT && type != GL_HALF_FLOAT)) {
+         transferOps |= IMAGE_CLAMP_BIT;
+      }
+   }
+
+   /* If the format is unsigned normalized, we can ignore clamping
+    * because the values are already in the range [0,1] so it won't
+    * have any effect anyway.
+    */
+   if (_mesa_get_format_datatype(texFormat) == GL_UNSIGNED_NORMALIZED &&
+       !need_rgb_to_luminance_conversion(texFormat, format)) {
+      transferOps &= ~IMAGE_CLAMP_BIT;
+   }
+
+   return transferOps;
+}
+
+
+/**
+ * Return true if memcpy cannot be used for ReadPixels.
+ *
+ * If uses_blit is true, the function returns true if a simple 3D engine blit
+ * cannot be used for ReadPixels packing.
+ *
+ * NOTE: This doesn't take swizzling and format conversions between
+ *       the readbuffer and the pixel pack buffer into account.
+ */
+GLboolean
+_mesa_readpixels_needs_slow_path(const struct gl_context *ctx, GLenum format,
+                                 GLenum type, GLboolean uses_blit)
+{
+   struct gl_renderbuffer *rb =
+         _mesa_get_read_renderbuffer_for_format(ctx, format);
+   GLenum srcType;
+
+   ASSERT(rb);
+
+   /* There are different rules depending on the base format. */
+   switch (format) {
+   case GL_DEPTH_STENCIL:
+      return !_mesa_has_depthstencil_combined(ctx->ReadBuffer) ||
+             ctx->Pixel.DepthScale != 1.0f || ctx->Pixel.DepthBias != 0.0f ||
+             ctx->Pixel.IndexShift || ctx->Pixel.IndexOffset ||
+             ctx->Pixel.MapStencilFlag;
+
+   case GL_DEPTH_COMPONENT:
+      return ctx->Pixel.DepthScale != 1.0f || ctx->Pixel.DepthBias != 0.0f;
+
+   case GL_STENCIL_INDEX:
+      return ctx->Pixel.IndexShift || ctx->Pixel.IndexOffset ||
+             ctx->Pixel.MapStencilFlag;
+
+   default:
+      /* Color formats. */
+      if (need_rgb_to_luminance_conversion(rb->Format, format)) {
+         return GL_TRUE;
+      }
+
+      /* Conversion between signed and unsigned integers needs masking
+       * (it isn't just memcpy). */
+      srcType = _mesa_get_format_datatype(rb->Format);
+
+      if ((srcType == GL_INT &&
+           (type == GL_UNSIGNED_INT ||
+            type == GL_UNSIGNED_SHORT ||
+            type == GL_UNSIGNED_BYTE)) ||
+          (srcType == GL_UNSIGNED_INT &&
+           (type == GL_INT ||
+            type == GL_SHORT ||
+            type == GL_BYTE))) {
+         return GL_TRUE;
+      }
+
+      /* And finally, see if there are any transfer ops. */
+      return get_readpixels_transfer_ops(ctx, rb->Format, format, type,
+                                         uses_blit) != 0;
+   }
+   return GL_FALSE;
+}
+
+
+static GLboolean
+readpixels_can_use_memcpy(const struct gl_context *ctx, GLenum format, GLenum type,
+                          const struct gl_pixelstore_attrib *packing)
+{
+   struct gl_renderbuffer *rb =
+         _mesa_get_read_renderbuffer_for_format(ctx, format);
+
+   ASSERT(rb);
+
+   if (_mesa_readpixels_needs_slow_path(ctx, format, type, GL_FALSE)) {
+      return GL_FALSE;
+   }
+
+   /* The base internal format and the base Mesa format must match. */
+   if (rb->_BaseFormat != _mesa_get_format_base_format(rb->Format)) {
+      return GL_FALSE;
+   }
+
+   /* The Mesa format must match the input format and type. */
+   if (!_mesa_format_matches_format_and_type(rb->Format, format, type,
+                                             packing->SwapBytes)) {
+      return GL_FALSE;
+   }
+
+   return GL_TRUE;
+}
+
+
+static GLboolean
+readpixels_memcpy(struct gl_context *ctx,
+                  GLint x, GLint y,
+                  GLsizei width, GLsizei height,
+                  GLenum format, GLenum type,
+                  GLvoid *pixels,
+                  const struct gl_pixelstore_attrib *packing)
+{
+   struct gl_renderbuffer *rb =
+         _mesa_get_read_renderbuffer_for_format(ctx, format);
+   GLubyte *dst, *map;
+   int dstStride, stride, j, texelBytes;
+
+   /* Fail if memcpy cannot be used. */
+   if (!readpixels_can_use_memcpy(ctx, format, type, packing)) {
+      return GL_FALSE;
+   }
+
+   dstStride = _mesa_image_row_stride(packing, width, format, type);
+   dst = (GLubyte *) _mesa_image_address2d(packing, pixels, width, height,
+					   format, type, 0, 0);
+
+   ctx->Driver.MapRenderbuffer(ctx, rb, x, y, width, height, GL_MAP_READ_BIT,
+			       &map, &stride);
+   if (!map) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glReadPixels");
+      return GL_TRUE;  /* don't bother trying the slow path */
+   }
+
+   texelBytes = _mesa_get_format_bytes(rb->Format);
+
+   /* memcpy*/
+   for (j = 0; j < height; j++) {
+      memcpy(dst, map, width * texelBytes);
+      dst += dstStride;
+      map += stride;
+   }
+
+   ctx->Driver.UnmapRenderbuffer(ctx, rb);
+   return GL_TRUE;
+}
+
+
+/**
+ * Optimized path for conversion of depth values to GL_DEPTH_COMPONENT,
+ * GL_UNSIGNED_INT.
+ */
+static GLboolean
+read_uint_depth_pixels( struct gl_context *ctx,
 			GLint x, GLint y,
 			GLsizei width, GLsizei height,
 			GLenum type, GLvoid *pixels,
@@ -65,10 +266,6 @@ fast_read_depth_pixels( struct gl_context *ctx,
    if (_mesa_get_format_datatype(rb->Format) != GL_UNSIGNED_NORMALIZED)
       return GL_FALSE;
 
-   if (!((type == GL_UNSIGNED_SHORT && rb->Format == MESA_FORMAT_Z16) ||
-	 type == GL_UNSIGNED_INT))
-      return GL_FALSE;
-
    ctx->Driver.MapRenderbuffer(ctx, rb, x, y, width, height, GL_MAP_READ_BIT,
 			       &map, &stride);
 
@@ -82,12 +279,7 @@ fast_read_depth_pixels( struct gl_context *ctx,
 					   GL_DEPTH_COMPONENT, type, 0, 0);
 
    for (j = 0; j < height; j++) {
-      if (type == GL_UNSIGNED_INT) {
-	 _mesa_unpack_uint_z_row(rb->Format, width, map, (GLuint *)dst);
-      } else {
-	 ASSERT(type == GL_UNSIGNED_SHORT && rb->Format == MESA_FORMAT_Z16);
-	 memcpy(dst, map, width * 2);
-      }
+      _mesa_unpack_uint_z_row(rb->Format, width, map, (GLuint *)dst);
 
       map += stride;
       dst += dstStride;
@@ -123,8 +315,10 @@ read_depth_pixels( struct gl_context *ctx,
    ASSERT(x + width <= (GLint) rb->Width);
    ASSERT(y + height <= (GLint) rb->Height);
 
-   if (fast_read_depth_pixels(ctx, x, y, width, height, type, pixels, packing))
+   if (type == GL_UNSIGNED_INT &&
+       read_uint_depth_pixels(ctx, x, y, width, height, type, pixels, packing)) {
       return;
+   }
 
    dstStride = _mesa_image_row_stride(packing, width, GL_DEPTH_COMPONENT, type);
    dst = (GLubyte *) _mesa_image_address2d(packing, pixels, width, height,
@@ -212,21 +406,20 @@ read_stencil_pixels( struct gl_context *ctx,
 
 
 /**
- * Try to do glReadPixels of RGBA data using a simple memcpy or swizzle.
+ * Try to do glReadPixels of RGBA data using swizzle.
  * \return GL_TRUE if successful, GL_FALSE otherwise (use the slow path)
  */
 static GLboolean
-fast_read_rgba_pixels_memcpy( struct gl_context *ctx,
-			      GLint x, GLint y,
-			      GLsizei width, GLsizei height,
-			      GLenum format, GLenum type,
-			      GLvoid *pixels,
-			      const struct gl_pixelstore_attrib *packing,
-			      GLbitfield transferOps )
+read_rgba_pixels_swizzle(struct gl_context *ctx,
+                         GLint x, GLint y,
+                         GLsizei width, GLsizei height,
+                         GLenum format, GLenum type,
+                         GLvoid *pixels,
+                         const struct gl_pixelstore_attrib *packing)
 {
    struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorReadBuffer;
    GLubyte *dst, *map;
-   int dstStride, stride, j, texelBytes;
+   int dstStride, stride, j;
    GLboolean swizzle_rb = GL_FALSE, copy_xrgb = GL_FALSE;
 
    /* XXX we could check for other swizzle/special cases here as needed */
@@ -242,19 +435,9 @@ fast_read_rgba_pixels_memcpy( struct gl_context *ctx,
        !ctx->Pack.SwapBytes) {
       copy_xrgb = GL_TRUE;
    }
-   else if (!_mesa_format_matches_format_and_type(rb->Format, format, type,
-                                                  ctx->Pack.SwapBytes))
-      return GL_FALSE;
-
-   /* If the format is unsigned normalized then we can ignore clamping
-    * because the values are already in the range [0,1] so it won't
-    * have any effect anyway.
-    */
-   if (_mesa_get_format_datatype(rb->Format) == GL_UNSIGNED_NORMALIZED)
-      transferOps &= ~IMAGE_CLAMP_BIT;
-
-   if (transferOps)
+   else {
       return GL_FALSE;
+   }
 
    dstStride = _mesa_image_row_stride(packing, width, format, type);
    dst = (GLubyte *) _mesa_image_address2d(packing, pixels, width, height,
@@ -267,8 +450,6 @@ fast_read_rgba_pixels_memcpy( struct gl_context *ctx,
       return GL_TRUE;  /* don't bother trying the slow path */
    }
 
-   texelBytes = _mesa_get_format_bytes(rb->Format);
-
    if (swizzle_rb) {
       /* swap R/B */
       for (j = 0; j < height; j++) {
@@ -294,13 +475,6 @@ fast_read_rgba_pixels_memcpy( struct gl_context *ctx,
          dst += dstStride;
          map += stride;
       }
-   } else {
-      /* just memcpy */
-      for (j = 0; j < height; j++) {
-         memcpy(dst, map, width * texelBytes);
-         dst += dstStride;
-         map += stride;
-      }
    }
 
    ctx->Driver.UnmapRenderbuffer(ctx, rb);
@@ -379,22 +553,20 @@ read_rgba_pixels( struct gl_context *ctx,
                   GLenum format, GLenum type, GLvoid *pixels,
                   const struct gl_pixelstore_attrib *packing )
 {
-   GLbitfield transferOps = ctx->_ImageTransferState;
+   GLbitfield transferOps;
    struct gl_framebuffer *fb = ctx->ReadBuffer;
    struct gl_renderbuffer *rb = fb->_ColorReadBuffer;
 
    if (!rb)
       return;
 
-   if ((ctx->Color._ClampReadColor == GL_TRUE || type != GL_FLOAT) &&
-       !_mesa_is_enum_format_integer(format)) {
-      transferOps |= IMAGE_CLAMP_BIT;
-   }
+   transferOps = get_readpixels_transfer_ops(ctx, rb->Format, format, type,
+                                             GL_FALSE);
 
    /* Try the optimized paths first. */
-   if (fast_read_rgba_pixels_memcpy(ctx, x, y, width, height,
-                                    format, type, pixels, packing,
-                                    transferOps)) {
+   if (!transferOps &&
+       read_rgba_pixels_swizzle(ctx, x, y, width, height,
+                                    format, type, pixels, packing)) {
       return;
    }
 
@@ -649,6 +821,14 @@ _mesa_readpixels(struct gl_context *ctx,
       pixels = _mesa_map_pbo_dest(ctx, &clippedPacking, pixels);
 
       if (pixels) {
+         /* Try memcpy first. */
+         if (readpixels_memcpy(ctx, x, y, width, height, format, type,
+                               pixels, packing)) {
+            _mesa_unmap_pbo_dest(ctx, &clippedPacking);
+            return;
+         }
+
+         /* Otherwise take the slow path. */
          switch (format) {
          case GL_STENCIL_INDEX:
             read_stencil_pixels(ctx, x, y, width, height, type, pixels,
diff --git a/mesalib/src/mesa/main/readpix.h b/mesalib/src/mesa/main/readpix.h
index 5a5f73f52..7491c22ff 100644
--- a/mesalib/src/mesa/main/readpix.h
+++ b/mesalib/src/mesa/main/readpix.h
@@ -33,6 +33,10 @@ struct gl_context;
 struct gl_pixelstore_attrib;
 
 
+extern GLboolean
+_mesa_readpixels_needs_slow_path(const struct gl_context *ctx, GLenum format,
+                                 GLenum type, GLboolean uses_blit);
+
 extern void
 _mesa_readpixels(struct gl_context *ctx,
                  GLint x, GLint y, GLsizei width, GLsizei height,
diff --git a/mesalib/src/mesa/main/texgetimage.c b/mesalib/src/mesa/main/texgetimage.c
index 7299a4b23..74b09ef2c 100644
--- a/mesalib/src/mesa/main/texgetimage.c
+++ b/mesalib/src/mesa/main/texgetimage.c
@@ -518,6 +518,7 @@ get_tex_rgba(struct gl_context *ctx, GLuint dimensions,
    if (type_needs_clamping(type)) {
       /* the returned image type can't have negative values */
       if (dataType == GL_FLOAT ||
+          dataType == GL_HALF_FLOAT ||
           dataType == GL_SIGNED_NORMALIZED ||
           format == GL_LUMINANCE ||
           format == GL_LUMINANCE_ALPHA) {
diff --git a/mesalib/src/mesa/main/teximage.c b/mesalib/src/mesa/main/teximage.c
index 4042e7969..bc755ae79 100644
--- a/mesalib/src/mesa/main/teximage.c
+++ b/mesalib/src/mesa/main/teximage.c
@@ -40,6 +40,7 @@
 #include "imports.h"
 #include "macros.h"
 #include "mfeatures.h"
+#include "multisample.h"
 #include "state.h"
 #include "texcompress.h"
 #include "texcompress_cpal.h"
@@ -4199,6 +4200,7 @@ teximagemultisample(GLuint dims, GLenum target, GLsizei samples,
    struct gl_texture_image *texImage;
    GLboolean sizeOK, dimensionsOK;
    gl_format texFormat;
+   GLenum sample_count_error;
 
    GET_CURRENT_CONTEXT(ctx);
 
@@ -4225,35 +4227,13 @@ teximagemultisample(GLuint dims, GLenum target, GLsizei samples,
       return;
    }
 
-   if (_mesa_is_enum_format_integer(internalformat)) {
-      if (samples > ctx->Const.MaxIntegerSamples) {
-         _mesa_error(ctx, GL_INVALID_OPERATION,
-               "glTexImage%uDMultisample(samples>GL_MAX_INTEGER_SAMPLES)",
-               dims);
-         return;
-      }
-   }
-   else if (_mesa_is_depth_or_stencil_format(internalformat)) {
-      if (samples > ctx->Const.MaxDepthTextureSamples) {
-         _mesa_error(ctx, GL_INVALID_OPERATION,
-               "glTexImage%uDMultisample(samples>GL_MAX_DEPTH_TEXTURE_SAMPLES)",
-               dims);
-         return;
-      }
-   }
-   else {
-      if (samples > ctx->Const.MaxColorTextureSamples) {
-         _mesa_error(ctx, GL_INVALID_OPERATION,
-               "glTexImage%uDMultisample(samples>GL_MAX_COLOR_TEXTURE_SAMPLES)",
-               dims);
-         return;
-      }
+   sample_count_error = _mesa_check_sample_count(ctx, target,
+         internalformat, samples);
+   if (sample_count_error != GL_NO_ERROR) {
+      _mesa_error(ctx, sample_count_error, "glTexImage%uDMultisample(samples)", dims);
+      return;
    }
 
-   /* TODO: should ask the driver for the exact limit for this internalformat
-    * once IDR's internalformat_query bits land
-    */
-
    texObj = _mesa_get_current_tex_object(ctx, target);
    texImage = _mesa_get_tex_image(ctx, texObj, 0, 0);
 
diff --git a/mesalib/src/mesa/main/texparam.c b/mesalib/src/mesa/main/texparam.c
index 120845b4a..bd2f75170 100644
--- a/mesalib/src/mesa/main/texparam.c
+++ b/mesalib/src/mesa/main/texparam.c
@@ -1460,6 +1460,12 @@ _mesa_GetTexParameterfv( GLenum target, GLenum pname, GLfloat *params )
          *params = (GLfloat) obj->Immutable;
          break;
 
+      case GL_TEXTURE_IMMUTABLE_LEVELS:
+         if (!_mesa_is_gles3(ctx))
+            goto invalid_pname;
+         *params = (GLfloat) obj->ImmutableLevels;
+         break;
+
       case GL_REQUIRED_TEXTURE_IMAGE_UNITS_OES:
          if (!_mesa_is_gles(ctx) || !ctx->Extensions.OES_EGL_image_external)
             goto invalid_pname;
@@ -1637,6 +1643,12 @@ _mesa_GetTexParameteriv( GLenum target, GLenum pname, GLint *params )
          *params = (GLint) obj->Immutable;
          break;
 
+      case GL_TEXTURE_IMMUTABLE_LEVELS:
+         if (!_mesa_is_gles3(ctx))
+            goto invalid_pname;
+         *params = obj->ImmutableLevels;
+         break;
+
       case GL_REQUIRED_TEXTURE_IMAGE_UNITS_OES:
          if (!_mesa_is_gles(ctx) || !ctx->Extensions.OES_EGL_image_external)
             goto invalid_pname;
diff --git a/mesalib/src/mesa/main/texstorage.c b/mesalib/src/mesa/main/texstorage.c
index 00f19bae5..675fd745b 100644
--- a/mesalib/src/mesa/main/texstorage.c
+++ b/mesalib/src/mesa/main/texstorage.c
@@ -397,6 +397,7 @@ texstorage(GLuint dims, GLenum target, GLsizei levels, GLenum internalformat,
       }
 
       texObj->Immutable = GL_TRUE;
+      texObj->ImmutableLevels = levels;
    }
 }
 
diff --git a/mesalib/src/mesa/state_tracker/st_cb_blit.c b/mesalib/src/mesa/state_tracker/st_cb_blit.c
index c463e3b04..50cab4294 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_blit.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_blit.c
@@ -239,31 +239,22 @@ st_BlitFramebuffer(struct gl_context *ctx,
       /* depth and/or stencil blit */
 
       /* get src/dst depth surfaces */
-      struct gl_renderbuffer_attachment *srcDepth =
-         &readFB->Attachment[BUFFER_DEPTH];
-      struct gl_renderbuffer_attachment *dstDepth =
-         &drawFB->Attachment[BUFFER_DEPTH];
-      struct gl_renderbuffer_attachment *srcStencil =
-         &readFB->Attachment[BUFFER_STENCIL];
-      struct gl_renderbuffer_attachment *dstStencil =
-         &drawFB->Attachment[BUFFER_STENCIL];
-
       struct st_renderbuffer *srcDepthRb =
-         st_renderbuffer(srcDepth->Renderbuffer);
+         st_renderbuffer(readFB->Attachment[BUFFER_DEPTH].Renderbuffer);
       struct st_renderbuffer *dstDepthRb = 
-         st_renderbuffer(dstDepth->Renderbuffer);
+         st_renderbuffer(drawFB->Attachment[BUFFER_DEPTH].Renderbuffer);
       struct pipe_surface *dstDepthSurf =
          dstDepthRb ? dstDepthRb->surface : NULL;
 
       struct st_renderbuffer *srcStencilRb =
-         st_renderbuffer(srcStencil->Renderbuffer);
+         st_renderbuffer(readFB->Attachment[BUFFER_STENCIL].Renderbuffer);
       struct st_renderbuffer *dstStencilRb =
-         st_renderbuffer(dstStencil->Renderbuffer);
+         st_renderbuffer(drawFB->Attachment[BUFFER_STENCIL].Renderbuffer);
       struct pipe_surface *dstStencilSurf =
          dstStencilRb ? dstStencilRb->surface : NULL;
 
-      if (st_is_depth_stencil_combined(srcDepth, srcStencil) &&
-          st_is_depth_stencil_combined(dstDepth, dstStencil)) {
+      if (_mesa_has_depthstencil_combined(readFB) &&
+          _mesa_has_depthstencil_combined(drawFB)) {
          blit.mask = 0;
          if (mask & GL_DEPTH_BUFFER_BIT)
             blit.mask |= PIPE_MASK_Z;
diff --git a/mesalib/src/mesa/state_tracker/st_cb_fbo.c b/mesalib/src/mesa/state_tracker/st_cb_fbo.c
index 87c5b048c..4452e523b 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_fbo.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_fbo.c
@@ -547,30 +547,6 @@ st_validate_attachment(struct gl_context *ctx,
 
    return valid;
 }
-
-
-/**
- * Check if two renderbuffer attachments name a combined depth/stencil
- * renderbuffer.
- */
-GLboolean
-st_is_depth_stencil_combined(const struct gl_renderbuffer_attachment *depth,
-                             const struct gl_renderbuffer_attachment *stencil)
-{
-   assert(depth && stencil);
-
-   if (depth->Type == stencil->Type) {
-      if (depth->Type == GL_RENDERBUFFER_EXT &&
-          depth->Renderbuffer == stencil->Renderbuffer)
-         return GL_TRUE;
-
-      if (depth->Type == GL_TEXTURE &&
-          depth->Texture == stencil->Texture)
-         return GL_TRUE;
-   }
-
-   return GL_FALSE;
-}
  
 
 /**
diff --git a/mesalib/src/mesa/state_tracker/st_cb_fbo.h b/mesalib/src/mesa/state_tracker/st_cb_fbo.h
index 506fd06d6..461dbe985 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_fbo.h
+++ b/mesalib/src/mesa/state_tracker/st_cb_fbo.h
@@ -76,9 +76,4 @@ st_new_renderbuffer_fb(enum pipe_format format, int samples, boolean sw);
 extern void
 st_init_fbo_functions(struct dd_function_table *functions);
 
-extern GLboolean
-st_is_depth_stencil_combined(const struct gl_renderbuffer_attachment *depth,
-                             const struct gl_renderbuffer_attachment *stencil);
-
-
 #endif /* ST_CB_FBO_H */
diff --git a/mesalib/src/mesa/state_tracker/st_cb_readpixels.c b/mesalib/src/mesa/state_tracker/st_cb_readpixels.c
index 6b824b161..bfed98870 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_readpixels.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_readpixels.c
@@ -25,35 +25,209 @@
  * 
  **************************************************************************/
 
-
+#include "main/image.h"
+#include "main/pbo.h"
 #include "main/imports.h"
 #include "main/readpix.h"
+#include "main/enums.h"
+#include "main/framebuffer.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
 
+#include "st_cb_fbo.h"
 #include "st_atom.h"
 #include "st_context.h"
 #include "st_cb_bitmap.h"
 #include "st_cb_readpixels.h"
+#include "state_tracker/st_cb_texture.h"
+#include "state_tracker/st_format.h"
+#include "state_tracker/st_texture.h"
 
 
 /**
- * The only special thing we need to do for the state tracker's
- * glReadPixels is to validate state (to be sure we have up-to-date
- * framebuffer surfaces) and flush the bitmap cache prior to reading.
+ * This uses a blit to copy the read buffer to a texture format which matches
+ * the format and type combo and then a fast read-back is done using memcpy.
+ * We can do arbitrary X/Y/Z/W/0/1 swizzling here as long as there is
+ * a format which matches the swizzling.
+ *
+ * If such a format isn't available, we fall back to _mesa_readpixels.
+ *
+ * NOTE: Some drivers use a blit to convert between tiled and linear
+ *       texture layouts during texture uploads/downloads, so the blit
+ *       we do here should be free in such cases.
  */
 static void
 st_readpixels(struct gl_context *ctx, GLint x, GLint y,
               GLsizei width, GLsizei height,
               GLenum format, GLenum type,
               const struct gl_pixelstore_attrib *pack,
-              GLvoid *dest)
+              GLvoid *pixels)
 {
    struct st_context *st = st_context(ctx);
+   struct gl_renderbuffer *rb =
+         _mesa_get_read_renderbuffer_for_format(ctx, format);
+   struct st_renderbuffer *strb = st_renderbuffer(rb);
+   struct pipe_context *pipe = st->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_resource *src;
+   struct pipe_resource *dst = NULL;
+   struct pipe_resource dst_templ;
+   enum pipe_format dst_format, src_format;
+   struct pipe_blit_info blit;
+   unsigned bind = PIPE_BIND_TRANSFER_READ;
+   struct pipe_transfer *tex_xfer;
+   ubyte *map = NULL;
 
+   /* Validate state (to be sure we have up-to-date framebuffer surfaces)
+    * and flush the bitmap cache prior to reading. */
    st_validate_state(st);
    st_flush_bitmap_cache(st);
-   _mesa_readpixels(ctx, x, y, width, height, format, type, pack, dest);
-}
 
+   if (!st->prefer_blit_based_texture_transfer) {
+      goto fallback;
+   }
+
+   /* This must be done after state validation. */
+   src = strb->texture;
+
+   /* XXX Fallback for depth-stencil formats due to an incomplete
+    * stencil blit implementation in some drivers. */
+   if (format == GL_DEPTH_STENCIL) {
+      goto fallback;
+   }
+
+   /* We are creating a texture of the size of the region being read back.
+    * Need to check for NPOT texture support. */
+   if (!screen->get_param(screen, PIPE_CAP_NPOT_TEXTURES) &&
+       (!util_is_power_of_two(width) ||
+        !util_is_power_of_two(height))) {
+      goto fallback;
+   }
+
+   /* If the base internal format and the texture format don't match, we have
+    * to use the slow path. */
+   if (rb->_BaseFormat !=
+       _mesa_get_format_base_format(rb->Format)) {
+      goto fallback;
+   }
+
+   /* See if the texture format already matches the format and type,
+    * in which case the memcpy-based fast path will likely be used and
+    * we don't have to blit. */
+   if (_mesa_format_matches_format_and_type(rb->Format, format,
+                                            type, pack->SwapBytes)) {
+      goto fallback;
+   }
+
+   if (_mesa_readpixels_needs_slow_path(ctx, format, type, GL_TRUE)) {
+      goto fallback;
+   }
+
+   /* Convert the source format to what is expected by ReadPixels
+    * and see if it's supported. */
+   src_format = util_format_linear(src->format);
+   src_format = util_format_luminance_to_red(src_format);
+   src_format = util_format_intensity_to_red(src_format);
+
+   if (!src_format ||
+       !screen->is_format_supported(screen, src_format, src->target,
+                                    src->nr_samples,
+                                    PIPE_BIND_SAMPLER_VIEW)) {
+      printf("fallback: src format unsupported %s\n", util_format_short_name(src_format));
+      goto fallback;
+   }
+
+   if (format == GL_DEPTH_COMPONENT || format == GL_DEPTH_STENCIL)
+      bind |= PIPE_BIND_DEPTH_STENCIL;
+   else
+      bind |= PIPE_BIND_RENDER_TARGET;
+
+   /* Choose the destination format by finding the best match
+    * for the format+type combo. */
+   dst_format = st_choose_matching_format(screen, bind, format, type,
+                                          pack->SwapBytes);
+   if (dst_format == PIPE_FORMAT_NONE) {
+      printf("fallback: no matching format for %s, %s\n",
+             _mesa_lookup_enum_by_nr(format), _mesa_lookup_enum_by_nr(type));
+      goto fallback;
+   }
+
+   /* create the destination texture */
+   memset(&dst_templ, 0, sizeof(dst_templ));
+   dst_templ.target = PIPE_TEXTURE_2D;
+   dst_templ.format = dst_format;
+   dst_templ.bind = bind;
+   dst_templ.usage = PIPE_USAGE_STAGING;
+
+   st_gl_texture_dims_to_pipe_dims(GL_TEXTURE_2D, width, height, 1,
+                                   &dst_templ.width0, &dst_templ.height0,
+                                   &dst_templ.depth0, &dst_templ.array_size);
+
+   dst = screen->resource_create(screen, &dst_templ);
+   if (!dst) {
+      goto fallback;
+   }
+
+   blit.src.resource = src;
+   blit.src.level = strb->rtt_level;
+   blit.src.format = src_format;
+   blit.dst.resource = dst;
+   blit.dst.level = 0;
+   blit.dst.format = dst->format;
+   blit.src.box.x = x;
+   blit.dst.box.x = 0;
+   blit.src.box.y = y;
+   blit.dst.box.y = 0;
+   blit.src.box.z = strb->rtt_face + strb->rtt_slice;
+   blit.dst.box.z = 0;
+   blit.src.box.width = blit.dst.box.width = width;
+   blit.src.box.height = blit.dst.box.height = height;
+   blit.src.box.depth = blit.dst.box.depth = 1;
+   blit.mask = st_get_blit_mask(rb->_BaseFormat, format);
+   blit.filter = PIPE_TEX_FILTER_NEAREST;
+   blit.scissor_enable = FALSE;
+
+   if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) {
+      blit.src.box.y = rb->Height - blit.src.box.y;
+      blit.src.box.height = -blit.src.box.height;
+   }
+
+   /* blit */
+   st->pipe->blit(st->pipe, &blit);
+
+   /* map resources */
+   pixels = _mesa_map_pbo_dest(ctx, pack, pixels);
+
+   map = pipe_transfer_map_3d(pipe, dst, 0, PIPE_TRANSFER_READ,
+                              0, 0, 0, width, height, 1, &tex_xfer);
+   if (!map) {
+      _mesa_unmap_pbo_dest(ctx, pack);
+      pipe_resource_reference(&dst, NULL);
+      goto fallback;
+   }
+
+   /* memcpy data into a user buffer */
+   {
+      const uint bytesPerRow = width * util_format_get_blocksize(dst_format);
+      GLuint row;
+
+      for (row = 0; row < height; row++) {
+         GLvoid *dest = _mesa_image_address3d(pack, pixels,
+                                              width, height, format,
+                                              type, 0, row, 0);
+         memcpy(dest, map, bytesPerRow);
+         map += tex_xfer->stride;
+      }
+   }
+
+   pipe_transfer_unmap(pipe, tex_xfer);
+   _mesa_unmap_pbo_dest(ctx, pack);
+   pipe_resource_reference(&dst, NULL);
+   return;
+
+fallback:
+   _mesa_readpixels(ctx, x, y, width, height, format, type, pack, pixels);
+}
 
 void st_init_readpixels_functions(struct dd_function_table *functions)
 {
diff --git a/mesalib/src/mesa/state_tracker/st_cb_texture.c b/mesalib/src/mesa/state_tracker/st_cb_texture.c
index c922a3164..94fbbf7be 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_texture.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_texture.c
@@ -68,7 +68,7 @@
 #define DBG if (0) printf
 
 
-static enum pipe_texture_target
+enum pipe_texture_target
 gl_target_to_pipe(GLenum target)
 {
    switch (target) {
@@ -542,8 +542,8 @@ prep_teximage(struct gl_context *ctx, struct gl_texture_image *texImage,
  * Return a writemask for the gallium blit. The parameters can be base
  * formats or "format" from glDrawPixels/glTexImage/glGetTexImage.
  */
-static unsigned
-get_blit_mask(GLenum srcFormat, GLenum dstFormat)
+unsigned
+st_get_blit_mask(GLenum srcFormat, GLenum dstFormat)
 {
    switch (dstFormat) {
    case GL_DEPTH_STENCIL:
@@ -608,6 +608,10 @@ st_TexSubImage(struct gl_context *ctx, GLuint dims,
    unsigned bind;
    GLubyte *map;
 
+   if (!st->prefer_blit_based_texture_transfer) {
+      goto fallback;
+   }
+
    if (!dst) {
       goto fallback;
    }
@@ -769,7 +773,7 @@ st_TexSubImage(struct gl_context *ctx, GLuint dims,
    blit.src.box.width = blit.dst.box.width = width;
    blit.src.box.height = blit.dst.box.height = height;
    blit.src.box.depth = blit.dst.box.depth = depth;
-   blit.mask = get_blit_mask(format, texImage->_BaseFormat);
+   blit.mask = st_get_blit_mask(format, texImage->_BaseFormat);
    blit.filter = PIPE_TEX_FILTER_NEAREST;
    blit.scissor_enable = FALSE;
 
@@ -860,6 +864,10 @@ st_GetTexImage(struct gl_context * ctx,
    ubyte *map = NULL;
    boolean done = FALSE;
 
+   if (!st->prefer_blit_based_texture_transfer) {
+      goto fallback;
+   }
+
    if (!stImage->pt) {
       goto fallback;
    }
@@ -996,7 +1004,7 @@ st_GetTexImage(struct gl_context * ctx,
    blit.src.box.width = blit.dst.box.width = width;
    blit.src.box.height = blit.dst.box.height = height;
    blit.src.box.depth = blit.dst.box.depth = depth;
-   blit.mask = get_blit_mask(texImage->_BaseFormat, format);
+   blit.mask = st_get_blit_mask(texImage->_BaseFormat, format);
    blit.filter = PIPE_TEX_FILTER_NEAREST;
    blit.scissor_enable = FALSE;
 
@@ -1370,7 +1378,7 @@ st_CopyTexSubImage(struct gl_context *ctx, GLuint dims,
    blit.dst.box.width = width;
    blit.dst.box.height = height;
    blit.dst.box.depth = 1;
-   blit.mask = get_blit_mask(rb->_BaseFormat, texImage->_BaseFormat);
+   blit.mask = st_get_blit_mask(rb->_BaseFormat, texImage->_BaseFormat);
    blit.filter = PIPE_TEX_FILTER_NEAREST;
 
    /* 1D array textures need special treatment.
diff --git a/mesalib/src/mesa/state_tracker/st_cb_texture.h b/mesalib/src/mesa/state_tracker/st_cb_texture.h
index 27956bcc2..7f70d0b25 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_texture.h
+++ b/mesalib/src/mesa/state_tracker/st_cb_texture.h
@@ -38,6 +38,12 @@ struct gl_texture_object;
 struct pipe_context;
 struct st_context;
 
+extern enum pipe_texture_target
+gl_target_to_pipe(GLenum target);
+
+unsigned
+st_get_blit_mask(GLenum srcFormat, GLenum dstFormat);
+
 extern GLboolean
 st_finalize_texture(struct gl_context *ctx,
 		    struct pipe_context *pipe, 
diff --git a/mesalib/src/mesa/state_tracker/st_context.c b/mesalib/src/mesa/state_tracker/st_context.c
index f9a584ba0..cc87f2bb3 100644
--- a/mesalib/src/mesa/state_tracker/st_context.c
+++ b/mesalib/src/mesa/state_tracker/st_context.c
@@ -182,6 +182,11 @@ st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe,
    st->has_stencil_export =
       screen->get_param(screen, PIPE_CAP_SHADER_STENCIL_EXPORT);
    st->has_shader_model3 = screen->get_param(screen, PIPE_CAP_SM3);
+   st->prefer_blit_based_texture_transfer = screen->get_param(screen,
+                              PIPE_CAP_PREFER_BLIT_BASED_TEXTURE_TRANSFER);
+
+   st->needs_texcoord_semantic =
+      screen->get_param(screen, PIPE_CAP_TGSI_TEXCOORD);
 
    /* GL limits and extensions */
    st_init_limits(st);
diff --git a/mesalib/src/mesa/state_tracker/st_context.h b/mesalib/src/mesa/state_tracker/st_context.h
index b9a98cd05..8786a036f 100644
--- a/mesalib/src/mesa/state_tracker/st_context.h
+++ b/mesalib/src/mesa/state_tracker/st_context.h
@@ -84,6 +84,9 @@ struct st_context
    boolean has_stencil_export; /**< can do shader stencil export? */
    boolean has_time_elapsed;
    boolean has_shader_model3;
+   boolean prefer_blit_based_texture_transfer;
+
+   boolean needs_texcoord_semantic;
 
    /* On old libGL's for linux we need to invalidate the drawables
     * on glViewpport calls, this is set via a option.
diff --git a/mesalib/src/mesa/state_tracker/st_format.c b/mesalib/src/mesa/state_tracker/st_format.c
index 5fd44e76d..a15706a03 100644
--- a/mesalib/src/mesa/state_tracker/st_format.c
+++ b/mesalib/src/mesa/state_tracker/st_format.c
@@ -1769,13 +1769,15 @@ st_ChooseTextureFormat(struct gl_context *ctx, GLenum target,
  * Called via ctx->Driver.ChooseTextureFormat().
  */
 size_t
-st_QuerySamplesForFormat(struct gl_context *ctx, GLenum internalFormat,
-                         int samples[16])
+st_QuerySamplesForFormat(struct gl_context *ctx, GLenum target,
+                         GLenum internalFormat, int samples[16])
 {
    struct st_context *st = st_context(ctx);
    enum pipe_format format;
    unsigned i, bind, num_sample_counts = 0;
 
+   (void) target;
+
    if (_mesa_is_depth_or_stencil_format(internalFormat))
       bind = PIPE_BIND_DEPTH_STENCIL;
    else
diff --git a/mesalib/src/mesa/state_tracker/st_format.h b/mesalib/src/mesa/state_tracker/st_format.h
index 3db409b74..0a1c18d92 100644
--- a/mesalib/src/mesa/state_tracker/st_format.h
+++ b/mesalib/src/mesa/state_tracker/st_format.h
@@ -67,8 +67,8 @@ st_ChooseTextureFormat(struct gl_context * ctx, GLenum target,
                        GLenum format, GLenum type);
 
 size_t
-st_QuerySamplesForFormat(struct gl_context *ctx, GLenum internalFormat,
-                         int samples[16]);
+st_QuerySamplesForFormat(struct gl_context *ctx, GLenum target,
+                         GLenum internalFormat, int samples[16]);
 
 /* can we use a sampler view to translate these formats
    only used to make TFP so far */
diff --git a/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index 0cef092d3..e3718eeda 100644
--- a/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -85,6 +85,11 @@ extern "C" {
  */
 #define MAX_TEMPS         4096
 
+/**
+ * Maximum number of arrays
+ */
+#define MAX_ARRAYS        256
+
 /* will be 4 for GLSL 4.00 */
 #define MAX_GLSL_TEXTURE_OFFSET 1
 
@@ -315,9 +320,11 @@ public:
 
    int next_temp;
 
+   unsigned array_sizes[MAX_ARRAYS];
+   unsigned next_array;
+
    int num_address_regs;
    int samplers_used;
-   bool indirect_addr_temps;
    bool indirect_addr_consts;
    
    int glsl_version;
@@ -549,9 +556,6 @@ glsl_to_tgsi_visitor::emit(ir_instruction *ir, unsigned op,
    /* Update indirect addressing status used by TGSI */
    if (dst.reladdr) {
       switch(dst.file) {
-      case PROGRAM_TEMPORARY:
-         this->indirect_addr_temps = true;
-         break;
       case PROGRAM_LOCAL_PARAM:
       case PROGRAM_ENV_PARAM:
       case PROGRAM_STATE_VAR:
@@ -570,9 +574,6 @@ glsl_to_tgsi_visitor::emit(ir_instruction *ir, unsigned op,
       for (i=0; i<3; i++) {
          if(inst->src[i].reladdr) {
             switch(inst->src[i].file) {
-            case PROGRAM_TEMPORARY:
-               this->indirect_addr_temps = true;
-               break;
             case PROGRAM_LOCAL_PARAM:
             case PROGRAM_ENV_PARAM:
             case PROGRAM_STATE_VAR:
@@ -1005,17 +1006,26 @@ glsl_to_tgsi_visitor::get_temp(const glsl_type *type)
    st_src_reg src;
 
    src.type = native_integers ? type->base_type : GLSL_TYPE_FLOAT;
-   src.file = PROGRAM_TEMPORARY;
-   src.index = next_temp;
    src.reladdr = NULL;
-   next_temp += type_size(type);
+   src.negate = 0;
+
+   if (type->is_array() || type->is_matrix()) {
+      src.file = PROGRAM_ARRAY;
+      src.index = next_array << 16 | 0x8000;
+      array_sizes[next_array] = type_size(type);
+      ++next_array;
+
+   } else {
+      src.file = PROGRAM_TEMPORARY;
+      src.index = next_temp;
+      next_temp += type_size(type);
+   }
 
    if (type->is_array() || type->is_record()) {
       src.swizzle = SWIZZLE_NOOP;
    } else {
       src.swizzle = swizzle_for_size(type->vector_elements);
    }
-   src.negate = 0;
 
    return src;
 }
@@ -1078,13 +1088,11 @@ glsl_to_tgsi_visitor::visit(ir_variable *ir)
           */
          assert((int) ir->num_state_slots == type_size(ir->type));
 
-         storage = new(mem_ctx) variable_storage(ir, PROGRAM_TEMPORARY,
-        					 this->next_temp);
-         this->variables.push_tail(storage);
-         this->next_temp += type_size(ir->type);
+         dst = st_dst_reg(get_temp(ir->type));
+
+         storage = new(mem_ctx) variable_storage(ir, dst.file, dst.index);
 
-         dst = st_dst_reg(st_src_reg(PROGRAM_TEMPORARY, storage->index,
-               native_integers ? ir->type->base_type : GLSL_TYPE_FLOAT));
+         this->variables.push_tail(storage);
       }
 
 
@@ -2052,11 +2060,11 @@ glsl_to_tgsi_visitor::visit(ir_dereference_variable *ir)
          break;
       case ir_var_auto:
       case ir_var_temporary:
-         entry = new(mem_ctx) variable_storage(var, PROGRAM_TEMPORARY,
-        				       this->next_temp);
+         st_src_reg src = get_temp(var->type);
+
+         entry = new(mem_ctx) variable_storage(var, src.file, src.index);
          this->variables.push_tail(entry);
 
-         next_temp += type_size(var->type);
          break;
       }
 
@@ -2574,11 +2582,10 @@ glsl_to_tgsi_visitor::get_function_signature(ir_function_signature *sig)
       storage = find_variable_storage(param);
       assert(!storage);
 
-      storage = new(mem_ctx) variable_storage(param, PROGRAM_TEMPORARY,
-        				      this->next_temp);
-      this->variables.push_tail(storage);
+      st_src_reg src = get_temp(param->type);
 
-      this->next_temp += type_size(param->type);
+      storage = new(mem_ctx) variable_storage(param, src.file, src.index);
+      this->variables.push_tail(storage);
    }
 
    if (!sig->return_type->is_void()) {
@@ -2978,12 +2985,12 @@ glsl_to_tgsi_visitor::glsl_to_tgsi_visitor()
 {
    result.file = PROGRAM_UNDEFINED;
    next_temp = 1;
+   next_array = 0;
    next_signature_id = 1;
    num_immediates = 0;
    current_function = NULL;
    num_address_regs = 0;
    samplers_used = 0;
-   indirect_addr_temps = false;
    indirect_addr_consts = false;
    glsl_version = 0;
    native_integers = false;
@@ -3183,7 +3190,8 @@ glsl_to_tgsi_visitor::simplify_cmp(void)
          assert(inst->dst.index < MAX_TEMPS);
          prevWriteMask = tempWrites[inst->dst.index];
          tempWrites[inst->dst.index] |= inst->dst.writemask;
-      }
+      } else
+         break;
 
       /* For a CMP to be considered a conditional write, the destination
        * register and source register two must be the same. */
@@ -3821,7 +3829,6 @@ get_pixel_transfer_visitor(struct st_fragment_program *fp,
    v->next_temp = original->next_temp;
    v->num_address_regs = original->num_address_regs;
    v->samplers_used = prog->SamplersUsed = original->samplers_used;
-   v->indirect_addr_temps = original->indirect_addr_temps;
    v->indirect_addr_consts = original->indirect_addr_consts;
    memcpy(&v->immediates, &original->immediates, sizeof(v->immediates));
    v->num_immediates = original->num_immediates;
@@ -3952,7 +3959,6 @@ get_bitmap_visitor(struct st_fragment_program *fp,
    v->next_temp = original->next_temp;
    v->num_address_regs = original->num_address_regs;
    v->samplers_used = prog->SamplersUsed = original->samplers_used;
-   v->indirect_addr_temps = original->indirect_addr_temps;
    v->indirect_addr_consts = original->indirect_addr_consts;
    memcpy(&v->immediates, &original->immediates, sizeof(v->immediates));
    v->num_immediates = original->num_immediates;
@@ -4014,6 +4020,7 @@ struct st_translate {
    struct ureg_program *ureg;
 
    struct ureg_dst temps[MAX_TEMPS];
+   struct ureg_dst arrays[MAX_ARRAYS];
    struct ureg_src *constants;
    struct ureg_src *immediates;
    struct ureg_dst outputs[PIPE_MAX_SHADER_OUTPUTS];
@@ -4022,6 +4029,8 @@ struct st_translate {
    struct ureg_src samplers[PIPE_MAX_SAMPLERS];
    struct ureg_src systemValues[SYSTEM_VALUE_MAX];
 
+   unsigned array_sizes[MAX_ARRAYS];
+
    const GLuint *inputMapping;
    const GLuint *outputMapping;
 
@@ -4132,16 +4141,34 @@ dst_register(struct st_translate *t,
              gl_register_file file,
              GLuint index)
 {
+   unsigned array;
+
    switch(file) {
    case PROGRAM_UNDEFINED:
       return ureg_dst_undef();
 
    case PROGRAM_TEMPORARY:
+      assert(index >= 0);
+      assert(index < (int) Elements(t->temps));
+ 
       if (ureg_dst_is_undef(t->temps[index]))
          t->temps[index] = ureg_DECL_local_temporary(t->ureg);
 
       return t->temps[index];
 
+   case PROGRAM_ARRAY:
+      array = index >> 16;
+
+      assert(array >= 0);
+      assert(array < (int) Elements(t->arrays));
+
+      if (ureg_dst_is_undef(t->arrays[array]))
+         t->arrays[array] = ureg_DECL_array_temporary(
+            t->ureg, t->array_sizes[array], TRUE);
+
+      return ureg_dst_array_offset(t->arrays[array],
+                                   (int)(index & 0xFFFF) - 0x8000);
+
    case PROGRAM_OUTPUT:
       if (t->procType == TGSI_PROCESSOR_VERTEX)
          assert(index < VARYING_SLOT_MAX);
@@ -4176,11 +4203,8 @@ src_register(struct st_translate *t,
       return ureg_src_undef();
 
    case PROGRAM_TEMPORARY:
-      assert(index >= 0);
-      assert(index < (int) Elements(t->temps));
-      if (ureg_dst_is_undef(t->temps[index]))
-         t->temps[index] = ureg_DECL_local_temporary(t->ureg);
-      return ureg_src(t->temps[index]);
+   case PROGRAM_ARRAY:
+      return ureg_src(dst_register(t, file, index));
 
    case PROGRAM_ENV_PARAM:
    case PROGRAM_LOCAL_PARAM:
@@ -4262,8 +4286,10 @@ translate_dst(struct st_translate *t,
       }
    }
 
-   if (dst_reg->reladdr != NULL)
+   if (dst_reg->reladdr != NULL) {
+      assert(dst_reg->file != PROGRAM_TEMPORARY);
       dst = ureg_dst_indirect(dst, ureg_src(t->address[0]));
+   }
 
    return dst;
 }
@@ -4286,26 +4312,8 @@ translate_src(struct st_translate *t, const st_src_reg *src_reg)
       src = ureg_negate(src);
 
    if (src_reg->reladdr != NULL) {
-      /* Normally ureg_src_indirect() would be used here, but a stupid compiler 
-       * bug in g++ makes ureg_src_indirect (an inline C function) erroneously 
-       * set the bit for src.Negate.  So we have to do the operation manually
-       * here to work around the compiler's problems. */
-      /*src = ureg_src_indirect(src, ureg_src(t->address[0]));*/
-      struct ureg_src addr = ureg_src(t->address[0]);
-      src.Indirect = 1;
-      src.IndirectFile = addr.File;
-      src.IndirectIndex = addr.Index;
-      src.IndirectSwizzle = addr.SwizzleX;
-      
-      if (src_reg->file != PROGRAM_INPUT &&
-          src_reg->file != PROGRAM_OUTPUT) {
-         /* If src_reg->index was negative, it was set to zero in
-          * src_register().  Reassign it now.  But don't do this
-          * for input/output regs since they get remapped while
-          * const buffers don't.
-          */
-         src.Index = src_reg->index;
-      }
+      assert(src_reg->file != PROGRAM_TEMPORARY);
+      src = ureg_src_indirect(src, ureg_src(t->address[0]));
    }
 
    return src;
@@ -4820,16 +4828,9 @@ st_translate_program(
       }
    }
 
-   if (program->indirect_addr_temps) {
-      /* If temps are accessed with indirect addressing, declare temporaries
-       * in sequential order.  Else, we declare them on demand elsewhere.
-       * (Note: the number of temporaries is equal to program->next_temp)
-       */
-      for (i = 0; i < (unsigned)program->next_temp; i++) {
-         /* XXX use TGSI_FILE_TEMPORARY_ARRAY when it's supported by ureg */
-         t->temps[i] = ureg_DECL_local_temporary(t->ureg);
-      }
-   }
+   /* Copy over array sizes
+    */
+   memcpy(t->array_sizes, program->array_sizes, sizeof(unsigned) * program->next_array);
 
    /* Emit constants and uniforms.  TGSI uses a single index space for these, 
     * so we put all the translated regs in t->constants.
@@ -5064,16 +5065,9 @@ get_mesa_program(struct gl_context *ctx,
    v->copy_propagate();
    while (v->eliminate_dead_code_advanced());
 
-   /* FIXME: These passes to optimize temporary registers don't work when there
-    * is indirect addressing of the temporary register space.  We need proper 
-    * array support so that we don't have to give up these passes in every 
-    * shader that uses arrays.
-    */
-   if (!v->indirect_addr_temps) {
-      v->eliminate_dead_code();
-      v->merge_registers();
-      v->renumber_registers();
-   }
+   v->eliminate_dead_code();
+   v->merge_registers();
+   v->renumber_registers();
    
    /* Write the END instruction. */
    v->emit(NULL, TGSI_OPCODE_END);
diff --git a/mesalib/src/mesa/state_tracker/st_program.c b/mesalib/src/mesa/state_tracker/st_program.c
index 6af8df316..7a38da84f 100644
--- a/mesalib/src/mesa/state_tracker/st_program.c
+++ b/mesalib/src/mesa/state_tracker/st_program.c
@@ -177,6 +177,7 @@ void
 st_prepare_vertex_program(struct gl_context *ctx,
                             struct st_vertex_program *stvp)
 {
+   struct st_context *st = st_context(ctx);
    GLuint attr;
 
    stvp->num_inputs = 0;
@@ -267,7 +268,8 @@ st_prepare_vertex_program(struct gl_context *ctx,
          case VARYING_SLOT_TEX5:
          case VARYING_SLOT_TEX6:
          case VARYING_SLOT_TEX7:
-            stvp->output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
+            stvp->output_semantic_name[slot] = st->needs_texcoord_semantic ?
+               TGSI_SEMANTIC_TEXCOORD : TGSI_SEMANTIC_GENERIC;
             stvp->output_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
             break;
 
@@ -275,10 +277,8 @@ st_prepare_vertex_program(struct gl_context *ctx,
          default:
             assert(attr < VARYING_SLOT_MAX);
             stvp->output_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
-            stvp->output_semantic_index[slot] = (VARYING_SLOT_VAR0 - 
-                                                VARYING_SLOT_TEX0 +
-                                                attr - 
-                                                VARYING_SLOT_VAR0);
+            stvp->output_semantic_index[slot] = st->needs_texcoord_semantic ?
+               (attr - VARYING_SLOT_VAR0) : (attr - VARYING_SLOT_TEX0);
             break;
          }
       }
@@ -585,11 +585,18 @@ st_translate_fragment_program(struct st_context *st,
              * fragment shader plus fixed-function hardware (such as
              * BFC).
              *
-             * There is no requirement that semantic indexes start at
-             * zero or be restricted to a particular range -- nobody
-             * should be building tables based on semantic index.
+             * However, some drivers may need us to identify the PNTC and TEXi
+             * varyings if, for example, their capability to replace them with
+             * sprite coordinates is limited.
              */
          case VARYING_SLOT_PNTC:
+            if (st->needs_texcoord_semantic) {
+               input_semantic_name[slot] = TGSI_SEMANTIC_PCOORD;
+               input_semantic_index[slot] = 0;
+               interpMode[slot] = TGSI_INTERPOLATE_LINEAR;
+               break;
+            }
+            /* fall through */
          case VARYING_SLOT_TEX0:
          case VARYING_SLOT_TEX1:
          case VARYING_SLOT_TEX2:
@@ -598,13 +605,29 @@ st_translate_fragment_program(struct st_context *st,
          case VARYING_SLOT_TEX5:
          case VARYING_SLOT_TEX6:
          case VARYING_SLOT_TEX7:
+            if (st->needs_texcoord_semantic) {
+               input_semantic_name[slot] = TGSI_SEMANTIC_TEXCOORD;
+               input_semantic_index[slot] = attr - VARYING_SLOT_TEX0;
+               interpMode[slot] =
+                  st_translate_interp(stfp->Base.InterpQualifier[attr], FALSE);
+               break;
+            }
+            /* fall through */
          case VARYING_SLOT_VAR0:
          default:
-            /* Actually, let's try and zero-base this just for
-             * readability of the generated TGSI.
+            /* Semantic indices should be zero-based because drivers may choose
+             * to assign a fixed slot determined by that index.
+             * This is useful because ARB_separate_shader_objects uses location
+             * qualifiers for linkage, and if the semantic index corresponds to
+             * these locations, linkage passes in the driver become unecessary.
+             *
+             * If needs_texcoord_semantic is true, no semantic indices will be
+             * consumed for the TEXi varyings, and we can base the locations of
+             * the user varyings on VAR0.  Otherwise, we use TEX0 as base index.
              */
             assert(attr >= VARYING_SLOT_TEX0);
-            input_semantic_index[slot] = (attr - VARYING_SLOT_TEX0);
+            input_semantic_index[slot] = st->needs_texcoord_semantic ?
+               (attr - VARYING_SLOT_VAR0) : (attr - VARYING_SLOT_TEX0);
             input_semantic_name[slot] = TGSI_SEMANTIC_GENERIC;
             if (attr == VARYING_SLOT_PNTC)
                interpMode[slot] = TGSI_INTERPOLATE_LINEAR;