Merge branch 'master' of ssh://git.freedesktop.org/git/mesa/mesa into pipe-video

Conflicts:
	src/gallium/include/pipe/p_format.h

configs/autoconf.in

@@ -53,7 +53,7 @@ MKDEP_OPTIONS = @MKDEP_OPTIONS@
 INSTALL = @INSTALL@
 
 # Python and flags (generally only needed by the developers)
-PYTHON2 = python
+PYTHON2 = @PYTHON2@
 PYTHON_FLAGS = -t -O -O
 
 # Library names (base name)

configure.ac

@@ -30,6 +30,7 @@ AC_PROG_CPP
 AC_PROG_CC
 AC_PROG_CXX
 AC_CHECK_PROGS([MAKE], [gmake make])
+AC_CHECK_PROGS([PYTHON2], [python2 python])
 AC_PATH_PROG([MKDEP], [makedepend])
 AC_PATH_PROG([SED], [sed])
 
@@ -1720,6 +1721,8 @@ echo ""
 echo "        CFLAGS:          $cflags"
 echo "        CXXFLAGS:        $cxxflags"
 echo "        Macros:          $defines"
+echo ""
+echo "        PYTHON2:         $PYTHON2"
 
 echo ""
 echo "        Run '${MAKE-make}' to build Mesa"

docs/GL3.txt

@@ -25,7 +25,7 @@ Non-normalized Integer texture/framebuffer formats    not started
 Packed depth/stencil formats                          DONE
 Per-buffer blend and masks (GL_EXT_draw_buffers2)     DONE
 GL_EXT_texture_compression_rgtc                       not started
-Red and red/green texture formats                     DONE (swrast, i965)
+Red and red/green texture formats                     DONE (swrast, i965, gallium)
 Transform feedback (GL_EXT_transform_feedback)        ~50% done
    glBindFragDataLocation, glGetFragDataLocation,
    glBindBufferRange, glBindBufferBase commands
@@ -42,7 +42,7 @@ GL 3.1:
 GLSL 1.30 and 1.40                                    not started
 Instanced drawing (GL_ARB_draw_instanced)             ~50% done
 Buffer copying (GL_ARB_copy_buffer)                   DONE
-Primitive restart (GL_NV_primitive_restart)           not started
+Primitive restart (GL_NV_primitive_restart)           DONE (gallium)
 16 vertex texture image units                         not started
 Texture buffer objs (GL_ARB_texture_buffer_object)    not started
 Rectangular textures (GL_ARB_texture_rectangle)       DONE

docs/relnotes-7.10.html

@@ -35,6 +35,9 @@ tbd
 <h2>New features</h2>
 <ul>
 <li>GL_ARB_explicit_attrib_location extension (Intel and software drivers).
+<li>GL_ARB_texture_rg (Intel, software drivers, gallium drivers).
+<li>GL_EXT_separate_shader_objects extension (Intel and software drivers).
+<li>GL_NV_primitive_restart extension (Gallium softpipe, llvmpipe).
 </ul>
 
 

src/egl/drivers/Makefile.template

@@ -24,8 +24,8 @@ $(EGL_DRIVER_PATH): $(EGL_DRIVER)
 
 $(EGL_DRIVER): $(EGL_OBJECTS) Makefile $(TOP)/src/egl/drivers/Makefile.template
 	@$(MKLIB) -o $(EGL_DRIVER) -noprefix \
-		-linker '$(CC)' -ldflags '$(LDFLAGS)' \
+		-linker '$(CC)' -ldflags '-L$(TOP)/$(LIB_DIR) $(LDFLAGS)' \
-		-L$(TOP)/$(LIB_DIR) $(MKLIB_OPTIONS) \
+		$(MKLIB_OPTIONS) \
 		$(EGL_OBJECTS) $(EGL_LIBS) -l$(EGL_LIB)
 
 .c.o:

src/egl/drivers/dri2/egl_dri2.c

@@ -248,21 +248,20 @@ dri2_add_config(_EGLDisplay *disp, const __DRIconfig *dri_config, int id,
    if (double_buffer)
       return NULL;
 
-   if (depth > 0 && depth != _eglGetConfigKey(&base, EGL_BUFFER_SIZE))
+   if (depth > 0 && depth != base.BufferSize)
       return NULL;
 
-   _eglSetConfigKey(&base, EGL_NATIVE_RENDERABLE, EGL_TRUE);
+   base.NativeRenderable = EGL_TRUE;
 
-   _eglSetConfigKey(&base, EGL_SURFACE_TYPE, surface_type);
+   base.SurfaceType = surface_type;
    if (surface_type & (EGL_PIXMAP_BIT | EGL_PBUFFER_BIT)) {
-      _eglSetConfigKey(&base, EGL_BIND_TO_TEXTURE_RGB, bind_to_texture_rgb);
+      base.BindToTextureRGB = bind_to_texture_rgb;
-      if (_eglGetConfigKey(&base, EGL_ALPHA_SIZE) > 0)
+      if (base.AlphaSize > 0)
-	 _eglSetConfigKey(&base,
+         base.BindToTextureRGBA = bind_to_texture_rgba;
-			  EGL_BIND_TO_TEXTURE_RGBA, bind_to_texture_rgba);
    }
 
-   _eglSetConfigKey(&base, EGL_RENDERABLE_TYPE, disp->ClientAPIsMask);
+   base.RenderableType = disp->ClientAPIsMask;
-   _eglSetConfigKey(&base, EGL_CONFORMANT, disp->ClientAPIsMask);
+   base.Conformant = disp->ClientAPIsMask;
 
    if (!_eglValidateConfig(&base, EGL_FALSE)) {
       _eglLog(_EGL_DEBUG, "DRI2: failed to validate config %d", id);
@@ -273,7 +272,7 @@ dri2_add_config(_EGLDisplay *disp, const __DRIconfig *dri_config, int id,
    if (conf != NULL) {
       memcpy(&conf->base, &base, sizeof base);
       conf->dri_config = dri_config;
-      _eglAddConfig(disp, &conf->base);
+      _eglLinkConfig(&conf->base);
    }
 
    return conf;
@@ -1161,7 +1160,7 @@ dri2_destroy_surface(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *surf)
 
    (void) drv;
 
-   if (_eglIsSurfaceBound(surf))
+   if (!_eglPutSurface(surf))
       return EGL_TRUE;
 
    (*dri2_dpy->core->destroyDrawable)(dri2_surf->dri_drawable);
@@ -1188,15 +1187,17 @@ dri2_make_current(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *dsurf,
    struct dri2_egl_surface *dri2_dsurf = dri2_egl_surface(dsurf);
    struct dri2_egl_surface *dri2_rsurf = dri2_egl_surface(rsurf);
    struct dri2_egl_context *dri2_ctx = dri2_egl_context(ctx);
+   _EGLContext *old_ctx;
+   _EGLSurface *old_dsurf, *old_rsurf;
    __DRIdrawable *ddraw, *rdraw;
    __DRIcontext *cctx;
 
-   /* bind the new context and return the "orphaned" one */
+   /* make new bindings */
-   if (!_eglBindContext(&ctx, &dsurf, &rsurf))
+   if (!_eglBindContext(ctx, dsurf, rsurf, &old_ctx, &old_dsurf, &old_rsurf))
       return EGL_FALSE;
 
    /* flush before context switch */
-   if (ctx && dri2_drv->glFlush)
+   if (old_ctx && dri2_drv->glFlush)
       dri2_drv->glFlush();
 
    ddraw = (dri2_dsurf) ? dri2_dsurf->dri_drawable : NULL;
@@ -1205,16 +1206,29 @@ dri2_make_current(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *dsurf,
 
    if ((cctx == NULL && ddraw == NULL && rdraw == NULL) ||
        dri2_dpy->core->bindContext(cctx, ddraw, rdraw)) {
-      if (dsurf && !_eglIsSurfaceLinked(dsurf))
+      dri2_destroy_surface(drv, disp, old_dsurf);
-	 dri2_destroy_surface(drv, disp, dsurf);
+      dri2_destroy_surface(drv, disp, old_rsurf);
-      if (rsurf && rsurf != dsurf && !_eglIsSurfaceLinked(dsurf))
+      if (old_ctx) {
-	 dri2_destroy_surface(drv, disp, rsurf);
+	 dri2_dpy->core->unbindContext(dri2_egl_context(old_ctx)->dri_context);
-      if (ctx != NULL && !_eglIsContextLinked(ctx))
+         /* no destroy? */
-	 dri2_dpy->core->unbindContext(dri2_egl_context(ctx)->dri_context);
+         _eglPutContext(old_ctx);
+      }
 
       return EGL_TRUE;
    } else {
-      _eglBindContext(&ctx, &dsurf, &rsurf);
+      /* undo the previous _eglBindContext */
+      _eglBindContext(old_ctx, old_dsurf, old_rsurf, &ctx, &dsurf, &rsurf);
+      assert(&dri2_ctx->base == ctx &&
+             &dri2_dsurf->base == dsurf &&
+             &dri2_rsurf->base == rsurf);
+
+      _eglPutSurface(dsurf);
+      _eglPutSurface(rsurf);
+      _eglPutContext(ctx);
+
+      _eglPutSurface(old_dsurf);
+      _eglPutSurface(old_rsurf);
+      _eglPutContext(old_ctx);
 
       return EGL_FALSE;
    }
@@ -1251,8 +1265,7 @@ dri2_create_surface(_EGLDriver *drv, _EGLDisplay *disp, EGLint type,
    if (type == EGL_PBUFFER_BIT) {
       dri2_surf->drawable = xcb_generate_id(dri2_dpy->conn);
       s = xcb_setup_roots_iterator(xcb_get_setup(dri2_dpy->conn));
-      xcb_create_pixmap(dri2_dpy->conn,
+      xcb_create_pixmap(dri2_dpy->conn, conf->BufferSize,
-			_eglGetConfigKey(conf, EGL_BUFFER_SIZE),
 			dri2_surf->drawable, s.data->root,
 			dri2_surf->base.Width, dri2_surf->base.Height);
    } else {
@@ -1601,7 +1614,7 @@ dri2_create_image_khr_pixmap(_EGLDisplay *disp, _EGLContext *ctx,
       return EGL_NO_IMAGE_KHR;
    }
 
-   if (!_eglInitImage(&dri2_img->base, disp, attr_list)) {
+   if (!_eglInitImage(&dri2_img->base, disp)) {
       free(buffers_reply);
       free(geometry_reply);
       return EGL_NO_IMAGE_KHR;
@@ -1644,7 +1657,7 @@ dri2_create_image_khr_renderbuffer(_EGLDisplay *disp, _EGLContext *ctx,
       return EGL_NO_IMAGE_KHR;
    }
 
-   if (!_eglInitImage(&dri2_img->base, disp, attr_list))
+   if (!_eglInitImage(&dri2_img->base, disp))
       return EGL_NO_IMAGE_KHR;
 
    dri2_img->dri_image = 
@@ -1661,56 +1674,28 @@ dri2_create_image_mesa_drm_buffer(_EGLDisplay *disp, _EGLContext *ctx,
 {
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
    struct dri2_egl_image *dri2_img;
-   EGLint width, height, format, name, stride, pitch, i, err;
+   EGLint format, name, pitch, err;
+   _EGLImageAttribs attrs;
 
    (void) ctx;
 
    name = (EGLint) buffer;
 
-   err = EGL_SUCCESS;
+   err = _eglParseImageAttribList(&attrs, disp, attr_list);
-   width = 0;
+   if (err != EGL_SUCCESS)
-   height = 0;
-   format = 0;
-   stride = 0;
-
-   for (i = 0; attr_list[i] != EGL_NONE; i++) {
-      EGLint attr = attr_list[i++];
-      EGLint val = attr_list[i];
-
-      switch (attr) {
-      case EGL_WIDTH:
-	 width = val;
-         break;
-      case EGL_HEIGHT:
-	 height = val;
-         break;
-      case EGL_DRM_BUFFER_FORMAT_MESA:
-	 format = val;
-         break;
-      case EGL_DRM_BUFFER_STRIDE_MESA:
-	 stride = val;
-         break;
-      default:
-         err = EGL_BAD_ATTRIBUTE;
-         break;
-      }
-
-      if (err != EGL_SUCCESS) {
-         _eglLog(_EGL_WARNING, "bad image attribute 0x%04x", attr);
       return NULL;
-      }
-   }
 
-   if (width <= 0 || height <= 0 || stride <= 0) {
+   if (attrs.Width <= 0 || attrs.Height <= 0 ||
+       attrs.DRMBufferStrideMESA <= 0) {
       _eglError(EGL_BAD_PARAMETER,
 		"bad width, height or stride");
       return NULL;
    }
 
-   switch (format) {
+   switch (attrs.DRMBufferFormatMESA) {
    case EGL_DRM_BUFFER_FORMAT_ARGB32_MESA:
       format = __DRI_IMAGE_FORMAT_ARGB8888;
-      pitch = stride;
+      pitch = attrs.DRMBufferStrideMESA;
       break;
    default:
       _eglError(EGL_BAD_PARAMETER,
@@ -1724,15 +1709,15 @@ dri2_create_image_mesa_drm_buffer(_EGLDisplay *disp, _EGLContext *ctx,
       return NULL;
    }
 
-   if (!_eglInitImage(&dri2_img->base, disp, attr_list)) {
+   if (!_eglInitImage(&dri2_img->base, disp)) {
       free(dri2_img);
       return NULL;
    }
 
    dri2_img->dri_image =
       dri2_dpy->image->createImageFromName(dri2_dpy->dri_screen,
-					   width,
+					   attrs.Width,
-					   height,
+					   attrs.Height,
 					   format,
 					   name,
 					   pitch,
@@ -1786,8 +1771,9 @@ dri2_create_drm_image_mesa(_EGLDriver *drv, _EGLDisplay *disp,
 {
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
    struct dri2_egl_image *dri2_img;
-   int width, height, format, i;
+   _EGLImageAttribs attrs;
-   unsigned int use, dri_use, valid_mask;
+   unsigned int dri_use, valid_mask;
+   int format;
    EGLint err = EGL_SUCCESS;
 
    (void) drv;
@@ -1803,74 +1789,50 @@ dri2_create_drm_image_mesa(_EGLDriver *drv, _EGLDisplay *disp,
       goto cleanup_img;
    }
 
-   if (!_eglInitImage(&dri2_img->base, disp, attr_list)) {
+   if (!_eglInitImage(&dri2_img->base, disp)) {
       err = EGL_BAD_PARAMETER;
       goto cleanup_img;
    }
 
-   width = 0;
+   err = _eglParseImageAttribList(&attrs, disp, attr_list);
-   height = 0;
+   if (err != EGL_SUCCESS)
-   format = 0;
-   use = 0;
-   for (i = 0; attr_list[i] != EGL_NONE; i++) {
-      EGLint attr = attr_list[i++];
-      EGLint val = attr_list[i];
-
-      switch (attr) {
-      case EGL_WIDTH:
-	 width = val;
-         break;
-      case EGL_HEIGHT:
-	 height = val;
-         break;
-      case EGL_DRM_BUFFER_FORMAT_MESA:
-	 format = val;
-         break;
-      case EGL_DRM_BUFFER_USE_MESA:
-	 use = val;
-         break;
-      default:
-         err = EGL_BAD_ATTRIBUTE;
-         break;
-      }
-
-      if (err != EGL_SUCCESS) {
-         _eglLog(_EGL_WARNING, "bad image attribute 0x%04x", attr);
       goto cleanup_img;
-      }
-   }
 
-   if (width <= 0 || height <= 0) {
+   if (attrs.Width <= 0 || attrs.Height <= 0) {
-      _eglLog(_EGL_WARNING, "bad width or height (%dx%d)", width, height);
+      _eglLog(_EGL_WARNING, "bad width or height (%dx%d)",
+            attrs.Width, attrs.Height);
       goto cleanup_img;
    }
 
-   switch (format) {
+   switch (attrs.DRMBufferFormatMESA) {
    case EGL_DRM_BUFFER_FORMAT_ARGB32_MESA:
       format = __DRI_IMAGE_FORMAT_ARGB8888;
       break;
    default:
-      _eglLog(_EGL_WARNING, "bad image format value 0x%04x", format);
+      _eglLog(_EGL_WARNING, "bad image format value 0x%04x",
+            attrs.DRMBufferFormatMESA);
       goto cleanup_img;
    }
 
    valid_mask =
       EGL_DRM_BUFFER_USE_SCANOUT_MESA |
       EGL_DRM_BUFFER_USE_SHARE_MESA; 
-   if (use & ~valid_mask) {
+   if (attrs.DRMBufferUseMESA & ~valid_mask) {
-      _eglLog(_EGL_WARNING, "bad image use bit 0x%04x", use & ~valid_mask);
+      _eglLog(_EGL_WARNING, "bad image use bit 0x%04x",
+            attrs.DRMBufferUseMESA & ~valid_mask);
       goto cleanup_img;
    }
 
    dri_use = 0;
-   if (use & EGL_DRM_BUFFER_USE_SHARE_MESA)
+   if (attrs.DRMBufferUseMESA & EGL_DRM_BUFFER_USE_SHARE_MESA)
       dri_use |= __DRI_IMAGE_USE_SHARE;
-   if (use & EGL_DRM_BUFFER_USE_SCANOUT_MESA)
+   if (attrs.DRMBufferUseMESA & EGL_DRM_BUFFER_USE_SCANOUT_MESA)
       dri_use |= __DRI_IMAGE_USE_SCANOUT;
 
    dri2_img->dri_image = 
       dri2_dpy->image->createImage(dri2_dpy->dri_screen,
-				   width, height, format, dri_use, dri2_img);
+				   attrs.Width, attrs.Height,
+                                   format, dri_use, dri2_img);
    if (dri2_img->dri_image == NULL) {
       err = EGL_BAD_ALLOC;
       goto cleanup_img;

src/egl/drivers/glx/egl_glx.c

@@ -132,29 +132,38 @@ static const struct {
    int egl_attr;
 } fbconfig_attributes[] = {
    /* table 3.1 of GLX 1.4 */
+   { GLX_FBCONFIG_ID,                  0 },
    { GLX_BUFFER_SIZE,                  EGL_BUFFER_SIZE },
    { GLX_LEVEL,                        EGL_LEVEL },
+   { GLX_DOUBLEBUFFER,                 0 },
+   { GLX_STEREO,                       0 },
+   { GLX_AUX_BUFFERS,                  0 },
    { GLX_RED_SIZE,                     EGL_RED_SIZE },
    { GLX_GREEN_SIZE,                   EGL_GREEN_SIZE },
    { GLX_BLUE_SIZE,                    EGL_BLUE_SIZE },
    { GLX_ALPHA_SIZE,                   EGL_ALPHA_SIZE },
    { GLX_DEPTH_SIZE,                   EGL_DEPTH_SIZE },
    { GLX_STENCIL_SIZE,                 EGL_STENCIL_SIZE },
+   { GLX_ACCUM_RED_SIZE,               0 },
+   { GLX_ACCUM_GREEN_SIZE,             0 },
+   { GLX_ACCUM_BLUE_SIZE,              0 },
+   { GLX_ACCUM_ALPHA_SIZE,             0 },
    { GLX_SAMPLE_BUFFERS,               EGL_SAMPLE_BUFFERS },
    { GLX_SAMPLES,                      EGL_SAMPLES },
-   { GLX_RENDER_TYPE,			EGL_RENDERABLE_TYPE },
+   { GLX_RENDER_TYPE,                  0 },
+   { GLX_DRAWABLE_TYPE,                EGL_SURFACE_TYPE },
    { GLX_X_RENDERABLE,                 EGL_NATIVE_RENDERABLE },
    { GLX_X_VISUAL_TYPE,                EGL_NATIVE_VISUAL_TYPE },
    { GLX_CONFIG_CAVEAT,                EGL_CONFIG_CAVEAT },
    { GLX_TRANSPARENT_TYPE,             EGL_TRANSPARENT_TYPE },
+   { GLX_TRANSPARENT_INDEX_VALUE,      0 },
    { GLX_TRANSPARENT_RED_VALUE,        EGL_TRANSPARENT_RED_VALUE },
    { GLX_TRANSPARENT_GREEN_VALUE,      EGL_TRANSPARENT_GREEN_VALUE },
    { GLX_TRANSPARENT_BLUE_VALUE,       EGL_TRANSPARENT_BLUE_VALUE },
    { GLX_MAX_PBUFFER_WIDTH,            EGL_MAX_PBUFFER_WIDTH },
    { GLX_MAX_PBUFFER_HEIGHT,           EGL_MAX_PBUFFER_HEIGHT },
    { GLX_MAX_PBUFFER_PIXELS,           EGL_MAX_PBUFFER_PIXELS },
-   { GLX_VISUAL_ID,			EGL_NATIVE_VISUAL_ID },
+   { GLX_VISUAL_ID,                    EGL_NATIVE_VISUAL_ID }
-   { GLX_X_VISUAL_TYPE,			EGL_NATIVE_VISUAL_TYPE },
 };
 
 
@@ -162,13 +171,31 @@ static EGLBoolean
 convert_fbconfig(Display *dpy, GLXFBConfig fbconfig,
                  struct GLX_egl_config *GLX_conf)
 {
-   int err = 0, attr, egl_attr, val;
+   int err, attr, val;
    unsigned i;
-   EGLint conformant, config_caveat, surface_type;
+
+   /* must have rgba bit */
+   err = glXGetFBConfigAttrib(dpy, fbconfig, GLX_RENDER_TYPE, &val);
+   if (err || !(val & GLX_RGBA_BIT))
+      return EGL_FALSE;
+
+   /* must know whether it is double-buffered */
+   err = glXGetFBConfigAttrib(dpy, fbconfig, GLX_DOUBLEBUFFER, &val);
+   if (err)
+      return EGL_FALSE;
+   GLX_conf->double_buffered = val;
+
+   GLX_conf->Base.RenderableType = EGL_OPENGL_BIT;
+   GLX_conf->Base.Conformant = EGL_OPENGL_BIT;
 
    for (i = 0; i < ARRAY_SIZE(fbconfig_attributes); i++) {
+      EGLint egl_attr, egl_val;
+
       attr = fbconfig_attributes[i].attr;
       egl_attr = fbconfig_attributes[i].egl_attr;
+      if (!egl_attr)
+         continue;
+
       err = glXGetFBConfigAttrib(dpy, fbconfig, attr, &val);
       if (err) {
          if (err == GLX_BAD_ATTRIBUTE) {
@@ -178,47 +205,71 @@ convert_fbconfig(Display *dpy, GLXFBConfig fbconfig,
          break;
       }
 
-      _eglSetConfigKey(&GLX_conf->Base, egl_attr, val);
+      switch (egl_attr) {
+      case EGL_SURFACE_TYPE:
+         egl_val = 0;
+         if (val & GLX_WINDOW_BIT)
+            egl_val |= EGL_WINDOW_BIT;
+         /* pixmap and pbuffer surfaces must be single-buffered in EGL */
+         if (!GLX_conf->double_buffered) {
+            if (val & GLX_PIXMAP_BIT)
+               egl_val |= EGL_PIXMAP_BIT;
+            if (val & GLX_PBUFFER_BIT)
+               egl_val |= EGL_PBUFFER_BIT;
+         }
+         break;
+      case EGL_NATIVE_VISUAL_TYPE:
+         switch (val) {
+         case GLX_TRUE_COLOR:
+            egl_val = TrueColor;
+            break;
+         case GLX_DIRECT_COLOR:
+            egl_val = DirectColor;
+            break;
+         case GLX_PSEUDO_COLOR:
+            egl_val = PseudoColor;
+            break;
+         case GLX_STATIC_COLOR:
+            egl_val = StaticColor;
+            break;
+         case GLX_GRAY_SCALE:
+            egl_val = GrayScale;
+            break;
+         case GLX_STATIC_GRAY:
+            egl_val = StaticGray;
+            break;
+         default:
+            egl_val = EGL_NONE;
+            break;
+         }
+         break;
+      case EGL_CONFIG_CAVEAT:
+         egl_val = EGL_NONE;
+         if (val == GLX_SLOW_CONFIG) {
+            egl_val = EGL_SLOW_CONFIG;
+         }
+         else if (val == GLX_NON_CONFORMANT_CONFIG) {
+            GLX_conf->Base.Conformant &= ~EGL_OPENGL_BIT;
+            egl_val = EGL_NONE;
+         }
+         break;
+      case EGL_TRANSPARENT_TYPE:
+         egl_val = (val == GLX_TRANSPARENT_RGB) ?
+            EGL_TRANSPARENT_RGB : EGL_NONE;
+         break;
+      default:
+         egl_val = val;
+         break;
+      }
+
+      _eglSetConfigKey(&GLX_conf->Base, egl_attr, egl_val);
    }
    if (err)
       return EGL_FALSE;
 
-   /* must have rgba bit */
+   if (!GLX_conf->Base.SurfaceType)
-   glXGetFBConfigAttrib(dpy, fbconfig, GLX_RENDER_TYPE, &val);
-   if (!(val & GLX_RGBA_BIT))
       return EGL_FALSE;
 
-   conformant = EGL_OPENGL_BIT;
-   glXGetFBConfigAttrib(dpy, fbconfig, GLX_CONFIG_CAVEAT, &val);
-   if (val == GLX_SLOW_CONFIG)
-      config_caveat = EGL_SLOW_CONFIG;
-   if (val == GLX_NON_CONFORMANT_CONFIG)
-      conformant &= ~EGL_OPENGL_BIT;
-   if (!(conformant & EGL_OPENGL_ES_BIT))
-      config_caveat = EGL_NON_CONFORMANT_CONFIG;
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_CONFIG_CAVEAT, config_caveat);
-
-   surface_type = 0;
-   glXGetFBConfigAttrib(dpy, fbconfig, GLX_DRAWABLE_TYPE, &val);
-   if (val & GLX_WINDOW_BIT)
-      surface_type |= EGL_WINDOW_BIT;
-   if (val & GLX_PIXMAP_BIT)
-      surface_type |= EGL_PIXMAP_BIT;
-   if (val & GLX_PBUFFER_BIT)
-      surface_type |= EGL_PBUFFER_BIT;
-
-   /* pixmap and pbuffer surfaces must be single-buffered in EGL */
-   glXGetFBConfigAttrib(dpy, fbconfig, GLX_DOUBLEBUFFER, &val);
-   GLX_conf->double_buffered = val;
-   if (GLX_conf->double_buffered) {
-      surface_type &= ~(EGL_PIXMAP_BIT | EGL_PBUFFER_BIT);
-      if (!surface_type)
-         return EGL_FALSE;
-   }
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_SURFACE_TYPE, surface_type);
-
    return EGL_TRUE;
 }
 
@@ -227,35 +278,69 @@ static const struct {
    int egl_attr;
 } visual_attributes[] = {
    /* table 3.7 of GLX 1.4 */
-   /* no GLX_USE_GL */
+   { GLX_USE_GL,              0 },
    { GLX_BUFFER_SIZE,         EGL_BUFFER_SIZE },
    { GLX_LEVEL,               EGL_LEVEL },
+   { GLX_RGBA,                0 },
+   { GLX_DOUBLEBUFFER,        0 },
+   { GLX_STEREO,              0 },
+   { GLX_AUX_BUFFERS,         0 },
    { GLX_RED_SIZE,            EGL_RED_SIZE },
    { GLX_GREEN_SIZE,          EGL_GREEN_SIZE },
    { GLX_BLUE_SIZE,           EGL_BLUE_SIZE },
    { GLX_ALPHA_SIZE,          EGL_ALPHA_SIZE },
    { GLX_DEPTH_SIZE,          EGL_DEPTH_SIZE },
    { GLX_STENCIL_SIZE,        EGL_STENCIL_SIZE },
+   { GLX_ACCUM_RED_SIZE,      0 },
+   { GLX_ACCUM_GREEN_SIZE,    0 },
+   { GLX_ACCUM_BLUE_SIZE,     0 },
+   { GLX_ACCUM_ALPHA_SIZE,    0 },
    { GLX_SAMPLE_BUFFERS,      EGL_SAMPLE_BUFFERS },
    { GLX_SAMPLES,             EGL_SAMPLES },
+   { GLX_FBCONFIG_ID,         0 },
+   /* GLX_EXT_visual_rating */
+   { GLX_VISUAL_CAVEAT_EXT,   EGL_CONFIG_CAVEAT }
 };
 
 static EGLBoolean
 convert_visual(Display *dpy, XVisualInfo *vinfo,
                struct GLX_egl_config *GLX_conf)
 {
-   int err, attr, egl_attr, val;
+   int err, attr, val;
    unsigned i;
-   EGLint conformant, config_caveat, surface_type;
 
-   /* the visual must support OpenGL */
+   /* the visual must support OpenGL and RGBA buffer */
    err = glXGetConfig(dpy, vinfo, GLX_USE_GL, &val);
+   if (!err && val)
+      err = glXGetConfig(dpy, vinfo, GLX_RGBA, &val);
    if (err || !val)
       return EGL_FALSE;
 
+   /* must know whether it is double-buffered */
+   err = glXGetConfig(dpy, vinfo, GLX_DOUBLEBUFFER, &val);
+   if (err)
+      return EGL_FALSE;
+   GLX_conf->double_buffered = val;
+
+   GLX_conf->Base.RenderableType = EGL_OPENGL_BIT;
+   GLX_conf->Base.Conformant = EGL_OPENGL_BIT;
+   GLX_conf->Base.SurfaceType = EGL_WINDOW_BIT;
+   /* pixmap surfaces must be single-buffered in EGL */
+   if (!GLX_conf->double_buffered)
+      GLX_conf->Base.SurfaceType |= EGL_PIXMAP_BIT;
+
+   GLX_conf->Base.NativeVisualID = vinfo->visualid;
+   GLX_conf->Base.NativeVisualType = vinfo->class;
+   GLX_conf->Base.NativeRenderable = EGL_TRUE;
+
    for (i = 0; i < ARRAY_SIZE(visual_attributes); i++) {
+      EGLint egl_attr, egl_val;
+
       attr = visual_attributes[i].attr;
-      egl_attr = fbconfig_attributes[i].egl_attr;
+      egl_attr = visual_attributes[i].egl_attr;
+      if (!egl_attr)
+         continue;
+
       err = glXGetConfig(dpy, vinfo, attr, &val);
       if (err) {
          if (err == GLX_BAD_ATTRIBUTE) {
@@ -265,41 +350,26 @@ convert_visual(Display *dpy, XVisualInfo *vinfo,
          break;
       }
 
-      _eglSetConfigKey(&GLX_conf->Base, egl_attr, val);
+      switch (egl_attr) {
+      case EGL_CONFIG_CAVEAT:
+         egl_val = EGL_NONE;
+         if (val == GLX_SLOW_VISUAL_EXT) {
+            egl_val = EGL_SLOW_CONFIG;
+         }
+         else if (val == GLX_NON_CONFORMANT_VISUAL_EXT) {
+            GLX_conf->Base.Conformant &= ~EGL_OPENGL_BIT;
+            egl_val = EGL_NONE;
+         }
+         break;
+         break;
+      default:
+         egl_val = val;
+         break;
+      }
+      _eglSetConfigKey(&GLX_conf->Base, egl_attr, egl_val);
    }
-   if (err)
-      return EGL_FALSE;
 
-   glXGetConfig(dpy, vinfo, GLX_RGBA, &val);
+   return (err) ? EGL_FALSE : EGL_TRUE;
-   if (!val)
-      return EGL_FALSE;
-
-   conformant = EGL_OPENGL_BIT;
-   glXGetConfig(dpy, vinfo, GLX_VISUAL_CAVEAT_EXT, &val);
-   if (val == GLX_SLOW_CONFIG)
-      config_caveat = EGL_SLOW_CONFIG;
-   if (val == GLX_NON_CONFORMANT_CONFIG)
-      conformant &= ~EGL_OPENGL_BIT;
-   if (!(conformant & EGL_OPENGL_ES_BIT))
-      config_caveat = EGL_NON_CONFORMANT_CONFIG;
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_CONFIG_CAVEAT, config_caveat);
-   _eglSetConfigKey(&GLX_conf->Base, EGL_NATIVE_VISUAL_ID, vinfo->visualid);
-   _eglSetConfigKey(&GLX_conf->Base, EGL_NATIVE_VISUAL_TYPE, vinfo->class);
-
-   /* pixmap and pbuffer surfaces must be single-buffered in EGL */
-   glXGetConfig(dpy, vinfo, GLX_DOUBLEBUFFER, &val);
-   GLX_conf->double_buffered = val;
-   surface_type = EGL_WINDOW_BIT;
-   /* pixmap surfaces must be single-buffered in EGL */
-   if (!GLX_conf->double_buffered)
-      surface_type |= EGL_PIXMAP_BIT;
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_SURFACE_TYPE, surface_type);
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_NATIVE_RENDERABLE, EGL_TRUE);
-
-   return EGL_TRUE;
 }
 
 
@@ -307,30 +377,31 @@ static void
 fix_config(struct GLX_egl_display *GLX_dpy, struct GLX_egl_config *GLX_conf)
 {
    _EGLConfig *conf = &GLX_conf->Base;
-   EGLint surface_type, r, g, b, a;
 
-   surface_type = GET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE);
    if (!GLX_conf->double_buffered && GLX_dpy->single_buffered_quirk) {
       /* some GLX impls do not like single-buffered window surface */
-      surface_type &= ~EGL_WINDOW_BIT;
+      conf->SurfaceType &= ~EGL_WINDOW_BIT;
       /* pbuffer bit is usually not set */
       if (GLX_dpy->have_pbuffer)
-         surface_type |= EGL_PBUFFER_BIT;
+         conf->SurfaceType |= EGL_PBUFFER_BIT;
-      SET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE, surface_type);
    }
 
    /* no visual attribs unless window bit is set */
-   if (!(surface_type & EGL_WINDOW_BIT)) {
+   if (!(conf->SurfaceType & EGL_WINDOW_BIT)) {
-      SET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_ID, 0);
+      conf->NativeVisualID = 0;
-      SET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_TYPE, EGL_NONE);
+      conf->NativeVisualType = EGL_NONE;
+   }
+
+   if (conf->TransparentType != EGL_TRANSPARENT_RGB) {
+      /* some impls set them to -1 (GLX_DONT_CARE) */
+      conf->TransparentRedValue = 0;
+      conf->TransparentGreenValue = 0;
+      conf->TransparentBlueValue = 0;
    }
 
    /* make sure buffer size is set correctly */
-   r = GET_CONFIG_ATTRIB(conf, EGL_RED_SIZE);
+   conf->BufferSize =
-   g = GET_CONFIG_ATTRIB(conf, EGL_GREEN_SIZE);
+      conf->RedSize + conf->GreenSize + conf->BlueSize + conf->AlphaSize;
-   b = GET_CONFIG_ATTRIB(conf, EGL_BLUE_SIZE);
-   a = GET_CONFIG_ATTRIB(conf, EGL_ALPHA_SIZE);
-   SET_CONFIG_ATTRIB(conf, EGL_BUFFER_SIZE, r + g + b + a);
 }
 
 
@@ -381,7 +452,7 @@ create_configs(_EGLDisplay *dpy, struct GLX_egl_display *GLX_dpy,
          memcpy(GLX_conf, &template, sizeof(template));
          GLX_conf->index = i;
 
-         _eglAddConfig(dpy, &GLX_conf->Base);
+         _eglLinkConfig(&GLX_conf->Base);
          id++;
       }
    }
@@ -606,14 +677,16 @@ GLX_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *dsurf,
    struct GLX_egl_surface *GLX_dsurf = GLX_egl_surface(dsurf);
    struct GLX_egl_surface *GLX_rsurf = GLX_egl_surface(rsurf);
    struct GLX_egl_context *GLX_ctx = GLX_egl_context(ctx);
+   _EGLContext *old_ctx;
+   _EGLSurface *old_dsurf, *old_rsurf;
    GLXDrawable ddraw, rdraw;
    GLXContext cctx;
    EGLBoolean ret = EGL_FALSE;
 
    (void) drv;
 
-   /* bind the new context and return the "orphaned" one */
+   /* make new bindings */
-   if (!_eglBindContext(&ctx, &dsurf, &rsurf))
+   if (!_eglBindContext(ctx, dsurf, rsurf, &old_ctx, &old_dsurf, &old_rsurf))
       return EGL_FALSE;
 
    ddraw = (GLX_dsurf) ? GLX_dsurf->glx_drawable : None;
@@ -626,13 +699,27 @@ GLX_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *dsurf,
       ret = glXMakeCurrent(GLX_dpy->dpy, ddraw, cctx);
 
    if (ret) {
-      if (dsurf && !_eglIsSurfaceLinked(dsurf))
+      if (_eglPutSurface(old_dsurf))
-         destroy_surface(disp, dsurf);
+         destroy_surface(disp, old_dsurf);
-      if (rsurf && rsurf != dsurf && !_eglIsSurfaceLinked(rsurf))
+      if (_eglPutSurface(old_rsurf))
-         destroy_surface(disp, rsurf);
+         destroy_surface(disp, old_rsurf);
+      /* no destroy? */
+      _eglPutContext(old_ctx);
    }
    else {
-      _eglBindContext(&ctx, &dsurf, &rsurf);
+      /* undo the previous _eglBindContext */
+      _eglBindContext(old_ctx, old_dsurf, old_rsurf, &ctx, &dsurf, &rsurf);
+      assert(&GLX_ctx->Base == ctx &&
+             &GLX_dsurf->Base == dsurf &&
+             &GLX_rsurf->Base == rsurf);
+
+      _eglPutSurface(dsurf);
+      _eglPutSurface(rsurf);
+      _eglPutContext(ctx);
+
+      _eglPutSurface(old_dsurf);
+      _eglPutSurface(old_rsurf);
+      _eglPutContext(old_ctx);
    }
 
    return ret;
@@ -836,7 +923,7 @@ GLX_eglDestroySurface(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *surf)
 {
    (void) drv;
 
-   if (!_eglIsSurfaceBound(surf))
+   if (_eglPutSurface(surf))
       destroy_surface(disp, surf);
 
    return EGL_TRUE;

src/egl/main/Makefile

@@ -36,6 +36,7 @@ SOURCES = \
 	eglcurrent.c \
 	egldisplay.c \
 	egldriver.c \
+	eglfallbacks.c \
 	eglglobals.c \
 	eglimage.c \
 	egllog.c \

src/egl/main/SConscript

@@ -28,6 +28,7 @@ if env['platform'] != 'winddk':
 		'eglcurrent.c',
 		'egldisplay.c',
 		'egldriver.c',
+		'eglfallbacks.c',
 		'eglglobals.c',
 		'eglimage.c',
 		'egllog.c',

src/egl/main/eglapi.c

@@ -416,7 +416,7 @@ eglCreateContext(EGLDisplay dpy, EGLConfig config, EGLContext share_list,
       RETURN_EGL_ERROR(disp, EGL_BAD_CONTEXT, EGL_NO_CONTEXT);
 
    context = drv->API.CreateContext(drv, disp, conf, share, attrib_list);
-   ret = (context) ? _eglLinkContext(context, disp) : EGL_NO_CONTEXT;
+   ret = (context) ? _eglLinkContext(context) : EGL_NO_CONTEXT;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -515,7 +515,7 @@ eglCreateWindowSurface(EGLDisplay dpy, EGLConfig config,
       RETURN_EGL_ERROR(disp, EGL_BAD_NATIVE_WINDOW, EGL_NO_SURFACE);
 
    surf = drv->API.CreateWindowSurface(drv, disp, conf, window, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -536,7 +536,7 @@ eglCreatePixmapSurface(EGLDisplay dpy, EGLConfig config,
       RETURN_EGL_ERROR(disp, EGL_BAD_NATIVE_PIXMAP, EGL_NO_SURFACE);
 
    surf = drv->API.CreatePixmapSurface(drv, disp, conf, pixmap, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -555,7 +555,7 @@ eglCreatePbufferSurface(EGLDisplay dpy, EGLConfig config,
    _EGL_CHECK_CONFIG(disp, conf, EGL_NO_SURFACE, drv);
 
    surf = drv->API.CreatePbufferSurface(drv, disp, conf, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -648,11 +648,12 @@ eglSwapInterval(EGLDisplay dpy, EGLint interval)
 
    _EGL_CHECK_DISPLAY(disp, EGL_FALSE, drv);
 
-   if (!ctx || !_eglIsContextLinked(ctx) || ctx->Resource.Display != disp)
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       ctx->Resource.Display != disp)
       RETURN_EGL_ERROR(disp, EGL_BAD_CONTEXT, EGL_FALSE);
 
    surf = ctx->DrawSurface;
-   if (!_eglIsSurfaceLinked(surf))
+   if (_eglGetSurfaceHandle(surf) == EGL_NO_SURFACE)
       RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
 
    ret = drv->API.SwapInterval(drv, disp, surf, interval);
@@ -673,7 +674,8 @@ eglSwapBuffers(EGLDisplay dpy, EGLSurface surface)
    _EGL_CHECK_SURFACE(disp, surf, EGL_FALSE, drv);
 
    /* surface must be bound to current context in EGL 1.4 */
-   if (!ctx || !_eglIsContextLinked(ctx) || surf != ctx->DrawSurface)
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       surf != ctx->DrawSurface)
       RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
 
    ret = drv->API.SwapBuffers(drv, disp, surf);
@@ -714,7 +716,8 @@ eglWaitClient(void)
    _eglLockMutex(&disp->Mutex);
 
    /* let bad current context imply bad current surface */
-   if (!_eglIsContextLinked(ctx) || !_eglIsSurfaceLinked(ctx->DrawSurface))
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       _eglGetSurfaceHandle(ctx->DrawSurface) == EGL_NO_SURFACE)
       RETURN_EGL_ERROR(disp, EGL_BAD_CURRENT_SURFACE, EGL_FALSE);
 
    /* a valid current context implies an initialized current display */
@@ -763,7 +766,8 @@ eglWaitNative(EGLint engine)
    _eglLockMutex(&disp->Mutex);
 
    /* let bad current context imply bad current surface */
-   if (!_eglIsContextLinked(ctx) || !_eglIsSurfaceLinked(ctx->DrawSurface))
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       _eglGetSurfaceHandle(ctx->DrawSurface) == EGL_NO_SURFACE)
       RETURN_EGL_ERROR(disp, EGL_BAD_CURRENT_SURFACE, EGL_FALSE);
 
    /* a valid current context implies an initialized current display */
@@ -1043,7 +1047,7 @@ eglCreateScreenSurfaceMESA(EGLDisplay dpy, EGLConfig config,
    _EGL_CHECK_CONFIG(disp, conf, EGL_NO_SURFACE, drv);
 
    surf = drv->API.CreateScreenSurfaceMESA(drv, disp, conf, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1235,7 +1239,7 @@ eglCreatePbufferFromClientBuffer(EGLDisplay dpy, EGLenum buftype,
 
    surf = drv->API.CreatePbufferFromClientBuffer(drv, disp, buftype, buffer,
                                                  conf, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1298,7 +1302,7 @@ eglCreateImageKHR(EGLDisplay dpy, EGLContext ctx, EGLenum target,
 
    img = drv->API.CreateImageKHR(drv,
          disp, context, target, buffer, attr_list);
-   ret = (img) ? _eglLinkImage(img, disp) : EGL_NO_IMAGE_KHR;
+   ret = (img) ? _eglLinkImage(img) : EGL_NO_IMAGE_KHR;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1344,7 +1348,7 @@ eglCreateSyncKHR(EGLDisplay dpy, EGLenum type, const EGLint *attrib_list)
       RETURN_EGL_EVAL(disp, EGL_NO_SYNC_KHR);
 
    sync = drv->API.CreateSyncKHR(drv, disp, type, attrib_list);
-   ret = (sync) ? _eglLinkSync(sync, disp) : EGL_NO_SYNC_KHR;
+   ret = (sync) ? _eglLinkSync(sync) : EGL_NO_SYNC_KHR;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1437,7 +1441,8 @@ eglSwapBuffersRegionNOK(EGLDisplay dpy, EGLSurface surface,
       RETURN_EGL_EVAL(disp, EGL_FALSE);
 
    /* surface must be bound to current context in EGL 1.4 */
-   if (!ctx || !_eglIsContextLinked(ctx) || surf != ctx->DrawSurface)
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       surf != ctx->DrawSurface)
       RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
 
    ret = drv->API.SwapBuffersRegionNOK(drv, disp, surf, numRects, rects);
@@ -1463,7 +1468,7 @@ eglCreateDRMImageMESA(EGLDisplay dpy, const EGLint *attr_list)
       RETURN_EGL_EVAL(disp, EGL_NO_IMAGE_KHR);
 
    img = drv->API.CreateDRMImageMESA(drv, disp, attr_list);
-   ret = (img) ? _eglLinkImage(img, disp) : EGL_NO_IMAGE_KHR;
+   ret = (img) ? _eglLinkImage(img) : EGL_NO_IMAGE_KHR;
 
    RETURN_EGL_EVAL(disp, ret);
 }

src/egl/main/eglconfig.c

@@ -24,34 +24,34 @@
  * IDs are from 1 to N respectively.
  */
 void
-_eglInitConfig(_EGLConfig *config, _EGLDisplay *dpy, EGLint id)
+_eglInitConfig(_EGLConfig *conf, _EGLDisplay *dpy, EGLint id)
 {
-   memset(config, 0, sizeof(*config));
+   memset(conf, 0, sizeof(*conf));
 
-   config->Display = dpy;
+   conf->Display = dpy;
 
    /* some attributes take non-zero default values */
-   SET_CONFIG_ATTRIB(config, EGL_CONFIG_ID,               id);
+   conf->ConfigID = id;
-   SET_CONFIG_ATTRIB(config, EGL_CONFIG_CAVEAT,           EGL_NONE);
+   conf->ConfigCaveat = EGL_NONE;
-   SET_CONFIG_ATTRIB(config, EGL_TRANSPARENT_TYPE,        EGL_NONE);
+   conf->TransparentType = EGL_NONE;
-   SET_CONFIG_ATTRIB(config, EGL_NATIVE_VISUAL_TYPE,      EGL_NONE);
+   conf->NativeVisualType = EGL_NONE;
-#ifdef EGL_VERSION_1_2
+   conf->ColorBufferType = EGL_RGB_BUFFER;
-   SET_CONFIG_ATTRIB(config, EGL_COLOR_BUFFER_TYPE,       EGL_RGB_BUFFER);
-#endif /* EGL_VERSION_1_2 */
 }
 
 
 /**
- * Link a config to a display and return the handle of the link.
+ * Link a config to its display and return the handle of the link.
  * The handle can be passed to client directly.
  *
  * Note that we just save the ptr to the config (we don't copy the config).
  */
-EGLConfig
+PUBLIC EGLConfig
-_eglAddConfig(_EGLDisplay *dpy, _EGLConfig *conf)
+_eglLinkConfig(_EGLConfig *conf)
 {
+   _EGLDisplay *dpy = conf->Display;
+
    /* sanity check */
-   assert(GET_CONFIG_ATTRIB(conf, EGL_CONFIG_ID) > 0);
+   assert(dpy && conf->ConfigID > 0);
 
    if (!dpy->Configs) {
       dpy->Configs = _eglCreateArray("Config", 16);
@@ -59,23 +59,29 @@ _eglAddConfig(_EGLDisplay *dpy, _EGLConfig *conf)
          return (EGLConfig) NULL;
    }
 
-   conf->Display = dpy;
    _eglAppendArray(dpy->Configs, (void *) conf);
 
    return (EGLConfig) conf;
 }
 
 
-EGLBoolean
+/**
-_eglCheckConfigHandle(EGLConfig config, _EGLDisplay *dpy)
+ * Lookup a handle to find the linked config.
+ * Return NULL if the handle has no corresponding linked config.
+ */
+_EGLConfig *
+_eglLookupConfig(EGLConfig config, _EGLDisplay *dpy)
 {
    _EGLConfig *conf;
 
+   if (!dpy)
+      return NULL;
+
    conf = (_EGLConfig *) _eglFindArray(dpy->Configs, (void *) config);
    if (conf)
       assert(conf->Display == dpy);
 
-   return (conf != NULL);
+   return conf;
 }
 
 
@@ -104,6 +110,7 @@ static const struct {
    EGLint default_value;
 } _eglValidationTable[] =
 {
+   /* core */
    { EGL_BUFFER_SIZE,               ATTRIB_TYPE_INTEGER,
                                     ATTRIB_CRITERION_ATLEAST,
                                     0 },
@@ -200,22 +207,13 @@ static const struct {
    { EGL_TRANSPARENT_BLUE_VALUE,    ATTRIB_TYPE_INTEGER,
                                     ATTRIB_CRITERION_EXACT,
                                     EGL_DONT_CARE },
-   /* these are not real attributes */
    { EGL_MATCH_NATIVE_PIXMAP,       ATTRIB_TYPE_PSEUDO,
                                     ATTRIB_CRITERION_SPECIAL,
                                     EGL_NONE },
-   /* there is a gap before EGL_SAMPLES */
+   /* extensions */
-   { 0x3030,                        ATTRIB_TYPE_PSEUDO,
-                                    ATTRIB_CRITERION_IGNORE,
-                                    0 },
-   { EGL_NONE,                      ATTRIB_TYPE_PSEUDO,
-                                    ATTRIB_CRITERION_IGNORE,
-                                    0 },
-
    { EGL_Y_INVERTED_NOK,            ATTRIB_TYPE_BOOLEAN,
                                     ATTRIB_CRITERION_EXACT,
-                                    EGL_DONT_CARE },
+                                    EGL_DONT_CARE }
-
 };
 
 
@@ -232,18 +230,13 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
 {
    EGLint i, attr, val;
    EGLBoolean valid = EGL_TRUE;
-   EGLint red_size = 0, green_size = 0, blue_size = 0, luminance_size = 0;
-   EGLint alpha_size = 0, buffer_size = 0;
-
-   /* all attributes should have been listed */
-   assert(ARRAY_SIZE(_eglValidationTable) == _EGL_CONFIG_NUM_ATTRIBS);
 
    /* check attributes by their types */
    for (i = 0; i < ARRAY_SIZE(_eglValidationTable); i++) {
       EGLint mask;
 
       attr = _eglValidationTable[i].attr;
-      val = GET_CONFIG_ATTRIB(conf, attr);
+      val = _eglGetConfigKey(conf, attr);
 
       switch (_eglValidationTable[i].type) {
       case ATTRIB_TYPE_INTEGER:
@@ -255,31 +248,15 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
             break;
          case EGL_SAMPLE_BUFFERS:
             /* there can be at most 1 sample buffer */
-            if (val > 1)
+            if (val > 1 || val < 0)
                valid = EGL_FALSE;
             break;
-         case EGL_RED_SIZE:
+         default:
-            red_size = val;
-            break;
-         case EGL_GREEN_SIZE:
-            green_size = val;
-            break;
-         case EGL_BLUE_SIZE:
-            blue_size = val;
-            break;
-         case EGL_LUMINANCE_SIZE:
-            luminance_size = val;
-            break;
-         case EGL_ALPHA_SIZE:
-            alpha_size = val;
-            break;
-         case EGL_BUFFER_SIZE:
-            buffer_size = val;
-            break;
-         }
             if (val < 0)
                valid = EGL_FALSE;
             break;
+         }
+         break;
       case ATTRIB_TYPE_BOOLEAN:
          if (val != EGL_TRUE && val != EGL_FALSE)
             valid = EGL_FALSE;
@@ -366,17 +343,18 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
 
    /* now check for conflicting attribute values */
 
-   switch (GET_CONFIG_ATTRIB(conf, EGL_COLOR_BUFFER_TYPE)) {
+   switch (conf->ColorBufferType) {
    case EGL_RGB_BUFFER:
-      if (luminance_size)
+      if (conf->LuminanceSize)
          valid = EGL_FALSE;
-      if (red_size + green_size + blue_size + alpha_size != buffer_size)
+      if (conf->RedSize + conf->GreenSize +
+            conf->BlueSize + conf->AlphaSize != conf->BufferSize)
          valid = EGL_FALSE;
       break;
    case EGL_LUMINANCE_BUFFER:
-      if (red_size || green_size || blue_size)
+      if (conf->RedSize || conf->GreenSize || conf->BlueSize)
          valid = EGL_FALSE;
-      if (luminance_size + alpha_size != buffer_size)
+      if (conf->LuminanceSize + conf->AlphaSize != conf->BufferSize)
          valid = EGL_FALSE;
       break;
    }
@@ -385,23 +363,19 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
       return EGL_FALSE;
    }
 
-   val = GET_CONFIG_ATTRIB(conf, EGL_SAMPLE_BUFFERS);
+   if (!conf->SampleBuffers && conf->Samples)
-   if (!val && GET_CONFIG_ATTRIB(conf, EGL_SAMPLES))
       valid = EGL_FALSE;
    if (!valid) {
       _eglLog(_EGL_DEBUG, "conflicting samples and sample buffers");
       return EGL_FALSE;
    }
 
-   val = GET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE);
+   if (!(conf->SurfaceType & EGL_WINDOW_BIT)) {
-   if (!(val & EGL_WINDOW_BIT)) {
+      if (conf->NativeVisualID != 0 || conf->NativeVisualType != EGL_NONE)
-      if (GET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_ID) != 0 ||
-          GET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_TYPE) != EGL_NONE)
          valid = EGL_FALSE;
    }
-   if (!(val & EGL_PBUFFER_BIT)) {
+   if (!(conf->SurfaceType & EGL_PBUFFER_BIT)) {
-      if (GET_CONFIG_ATTRIB(conf, EGL_BIND_TO_TEXTURE_RGB) ||
+      if (conf->BindToTextureRGB || conf->BindToTextureRGBA)
-          GET_CONFIG_ATTRIB(conf, EGL_BIND_TO_TEXTURE_RGBA))
          valid = EGL_FALSE;
    }
    if (!valid) {
@@ -433,11 +407,11 @@ _eglMatchConfig(const _EGLConfig *conf, const _EGLConfig *criteria)
          continue;
 
       attr = _eglValidationTable[i].attr;
-      cmp = GET_CONFIG_ATTRIB(criteria, attr);
+      cmp = _eglGetConfigKey(criteria, attr);
       if (cmp == EGL_DONT_CARE)
          continue;
 
-      val = GET_CONFIG_ATTRIB(conf, attr);
+      val = _eglGetConfigKey(conf, attr);
       switch (_eglValidationTable[i].criterion) {
       case ATTRIB_CRITERION_EXACT:
          if (val != cmp)
@@ -478,16 +452,11 @@ _eglMatchConfig(const _EGLConfig *conf, const _EGLConfig *criteria)
 static INLINE EGLBoolean
 _eglIsConfigAttribValid(_EGLConfig *conf, EGLint attr)
 {
-   if (_eglIndexConfig(conf, attr) < 0)
+   if (_eglOffsetOfConfig(attr) < 0)
       return EGL_FALSE;
 
-   /* there are some holes in the range */
    switch (attr) {
-   case 0x3030 /* a gap before EGL_SAMPLES */:
-   case EGL_NONE:
-#ifdef EGL_VERSION_1_4
    case EGL_MATCH_NATIVE_PIXMAP:
-#endif
       return EGL_FALSE;
    case EGL_Y_INVERTED_NOK:
       return conf->Display->Extensions.NOK_texture_from_pixmap;
@@ -503,18 +472,18 @@ _eglIsConfigAttribValid(_EGLConfig *conf, EGLint attr)
  * Return EGL_FALSE if any of the attribute is invalid.
  */
 EGLBoolean
-_eglParseConfigAttribList(_EGLConfig *conf, const EGLint *attrib_list)
+_eglParseConfigAttribList(_EGLConfig *conf, _EGLDisplay *dpy,
+                          const EGLint *attrib_list)
 {
    EGLint attr, val, i;
-   EGLint config_id = 0, level = 0;
+
-   EGLBoolean has_native_visual_type = EGL_FALSE;
+   _eglInitConfig(conf, dpy, EGL_DONT_CARE);
-   EGLBoolean has_transparent_color = EGL_FALSE;
 
    /* reset to default values */
    for (i = 0; i < ARRAY_SIZE(_eglValidationTable); i++) {
       attr = _eglValidationTable[i].attr;
       val = _eglValidationTable[i].default_value;
-      SET_CONFIG_ATTRIB(conf, attr, val);
+      _eglSetConfigKey(conf, attr, val);
    }
 
    /* parse the list */
@@ -525,58 +494,32 @@ _eglParseConfigAttribList(_EGLConfig *conf, const EGLint *attrib_list)
       if (!_eglIsConfigAttribValid(conf, attr))
 	 return EGL_FALSE;
 
-      SET_CONFIG_ATTRIB(conf, attr, val);
+      _eglSetConfigKey(conf, attr, val);
-
-      /* rememeber some attributes for post-processing */
-      switch (attr) {
-      case EGL_CONFIG_ID:
-         config_id = val;
-         break;
-      case EGL_LEVEL:
-         level = val;
-         break;
-      case EGL_NATIVE_VISUAL_TYPE:
-         has_native_visual_type = EGL_TRUE;
-         break;
-      case EGL_TRANSPARENT_RED_VALUE:
-      case EGL_TRANSPARENT_GREEN_VALUE:
-      case EGL_TRANSPARENT_BLUE_VALUE:
-         has_transparent_color = EGL_TRUE;
-         break;
-      default:
-         break;
-      }
    }
 
    if (!_eglValidateConfig(conf, EGL_TRUE))
       return EGL_FALSE;
 
    /* the spec says that EGL_LEVEL cannot be EGL_DONT_CARE */
-   if (level == EGL_DONT_CARE)
+   if (conf->Level == EGL_DONT_CARE)
       return EGL_FALSE;
 
    /* ignore other attributes when EGL_CONFIG_ID is given */
-   if (config_id > 0) {
+   if (conf->ConfigID != EGL_DONT_CARE) {
-      _eglResetConfigKeys(conf, EGL_DONT_CARE);
+      for (i = 0; i < ARRAY_SIZE(_eglValidationTable); i++) {
-      SET_CONFIG_ATTRIB(conf, EGL_CONFIG_ID, config_id);
+         attr = _eglValidationTable[i].attr;
+         if (attr != EGL_CONFIG_ID)
+            _eglSetConfigKey(conf, attr, EGL_DONT_CARE);
+      }
    }
    else {
-      if (has_native_visual_type) {
+      if (!(conf->SurfaceType & EGL_WINDOW_BIT))
-         val = GET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE);
+         conf->NativeVisualType = EGL_DONT_CARE;
-         if (!(val & EGL_WINDOW_BIT))
-            SET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_TYPE, EGL_DONT_CARE);
-      }
 
-      if (has_transparent_color) {
+      if (conf->TransparentType == EGL_NONE) {
-         val = GET_CONFIG_ATTRIB(conf, EGL_TRANSPARENT_TYPE);
+         conf->TransparentRedValue = EGL_DONT_CARE;
-         if (val == EGL_NONE) {
+         conf->TransparentGreenValue = EGL_DONT_CARE;
-            SET_CONFIG_ATTRIB(conf, EGL_TRANSPARENT_RED_VALUE,
+         conf->TransparentBlueValue = EGL_DONT_CARE;
-                              EGL_DONT_CARE);
-            SET_CONFIG_ATTRIB(conf, EGL_TRANSPARENT_GREEN_VALUE,
-                              EGL_DONT_CARE);
-            SET_CONFIG_ATTRIB(conf, EGL_TRANSPARENT_BLUE_VALUE,
-                              EGL_DONT_CARE);
-         }
       }
    }
 
@@ -610,7 +553,6 @@ _eglCompareConfigs(const _EGLConfig *conf1, const _EGLConfig *conf2,
       EGL_ALPHA_MASK_SIZE,
    };
    EGLint val1, val2;
-   EGLBoolean rgb_buffer;
    EGLint i;
 
    if (conf1 == conf2)
@@ -619,44 +561,41 @@ _eglCompareConfigs(const _EGLConfig *conf1, const _EGLConfig *conf2,
    /* the enum values have the desired ordering */
    assert(EGL_NONE < EGL_SLOW_CONFIG);
    assert(EGL_SLOW_CONFIG < EGL_NON_CONFORMANT_CONFIG);
-   val1 = GET_CONFIG_ATTRIB(conf1, EGL_CONFIG_CAVEAT);
+   val1 = conf1->ConfigCaveat - conf2->ConfigCaveat;
-   val2 = GET_CONFIG_ATTRIB(conf2, EGL_CONFIG_CAVEAT);
+   if (val1)
-   if (val1 != val2)
+      return val1;
-      return (val1 - val2);
 
    /* the enum values have the desired ordering */
    assert(EGL_RGB_BUFFER < EGL_LUMINANCE_BUFFER);
-   val1 = GET_CONFIG_ATTRIB(conf1, EGL_COLOR_BUFFER_TYPE);
+   val1 = conf1->ColorBufferType - conf2->ColorBufferType;
-   val2 = GET_CONFIG_ATTRIB(conf2, EGL_COLOR_BUFFER_TYPE);
+   if (val1)
-   if (val1 != val2)
+      return val1;
-      return (val1 - val2);
-   rgb_buffer = (val1 == EGL_RGB_BUFFER);
 
    if (criteria) {
       val1 = val2 = 0;
-      if (rgb_buffer) {
+      if (conf1->ColorBufferType == EGL_RGB_BUFFER) {
-         if (GET_CONFIG_ATTRIB(criteria, EGL_RED_SIZE) > 0) {
+         if (criteria->RedSize > 0) {
-            val1 += GET_CONFIG_ATTRIB(conf1, EGL_RED_SIZE);
+            val1 += conf1->RedSize;
-            val2 += GET_CONFIG_ATTRIB(conf2, EGL_RED_SIZE);
+            val2 += conf2->RedSize;
          }
-         if (GET_CONFIG_ATTRIB(criteria, EGL_GREEN_SIZE) > 0) {
+         if (criteria->GreenSize > 0) {
-            val1 += GET_CONFIG_ATTRIB(conf1, EGL_GREEN_SIZE);
+            val1 += conf1->GreenSize;
-            val2 += GET_CONFIG_ATTRIB(conf2, EGL_GREEN_SIZE);
+            val2 += conf2->GreenSize;
          }
-         if (GET_CONFIG_ATTRIB(criteria, EGL_BLUE_SIZE) > 0) {
+         if (criteria->BlueSize > 0) {
-            val1 += GET_CONFIG_ATTRIB(conf1, EGL_BLUE_SIZE);
+            val1 += conf1->BlueSize;
-            val2 += GET_CONFIG_ATTRIB(conf2, EGL_BLUE_SIZE);
+            val2 += conf2->BlueSize;
          }
       }
       else {
-         if (GET_CONFIG_ATTRIB(criteria, EGL_LUMINANCE_SIZE) > 0) {
+         if (criteria->LuminanceSize > 0) {
-            val1 += GET_CONFIG_ATTRIB(conf1, EGL_LUMINANCE_SIZE);
+            val1 += conf1->LuminanceSize;
-            val2 += GET_CONFIG_ATTRIB(conf2, EGL_LUMINANCE_SIZE);
+            val2 += conf2->LuminanceSize;
          }
       }
-      if (GET_CONFIG_ATTRIB(criteria, EGL_ALPHA_SIZE) > 0) {
+      if (criteria->AlphaSize > 0) {
-         val1 += GET_CONFIG_ATTRIB(conf1, EGL_ALPHA_SIZE);
+         val1 += conf1->AlphaSize;
-         val2 += GET_CONFIG_ATTRIB(conf2, EGL_ALPHA_SIZE);
+         val2 += conf2->AlphaSize;
       }
    }
    else {
@@ -669,24 +608,15 @@ _eglCompareConfigs(const _EGLConfig *conf1, const _EGLConfig *conf2,
       return (val2 - val1);
 
    for (i = 0; i < ARRAY_SIZE(compare_attribs); i++) {
-      val1 = GET_CONFIG_ATTRIB(conf1, compare_attribs[i]);
+      val1 = _eglGetConfigKey(conf1, compare_attribs[i]);
-      val2 = GET_CONFIG_ATTRIB(conf2, compare_attribs[i]);
+      val2 = _eglGetConfigKey(conf2, compare_attribs[i]);
       if (val1 != val2)
          return (val1 - val2);
    }
 
    /* EGL_NATIVE_VISUAL_TYPE cannot be compared here */
 
-   if (compare_id) {
+   return (compare_id) ? (conf1->ConfigID - conf2->ConfigID) : 0;
-      val1 = GET_CONFIG_ATTRIB(conf1, EGL_CONFIG_ID);
-      val2 = GET_CONFIG_ATTRIB(conf2, EGL_CONFIG_ID);
-      assert(val1 != val2);
-   }
-   else {
-      val1 = val2 = 0;
-   }
-
-   return (val1 - val2);
 }
 
 
@@ -764,8 +694,7 @@ _eglChooseConfig(_EGLDriver *drv, _EGLDisplay *disp, const EGLint *attrib_list,
    if (!num_configs)
       return _eglError(EGL_BAD_PARAMETER, "eglChooseConfigs");
 
-   _eglInitConfig(&criteria, disp, 0);
+   if (!_eglParseConfigAttribList(&criteria, disp, attrib_list))
-   if (!_eglParseConfigAttribList(&criteria, attrib_list))
       return _eglError(EGL_BAD_ATTRIBUTE, "eglChooseConfig");
 
    configList = (_EGLConfig **) _eglFilterArray(disp->Configs, &count,
@@ -802,7 +731,7 @@ _eglGetConfigAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
    if (!value)
       return _eglError(EGL_BAD_PARAMETER, "eglGetConfigAttrib");
 
-   *value = GET_CONFIG_ATTRIB(conf, attribute);
+   *value = _eglGetConfigKey(conf, attribute);
    return EGL_TRUE;
 }
 

src/egl/main/eglconfig.h

@@ -6,69 +6,103 @@
 #include "egltypedefs.h"
 
 
-#define _EGL_CONFIG_FIRST_ATTRIB EGL_BUFFER_SIZE
+/* update _eglValidationTable and _eglOffsetOfConfig before updating this
-#define _EGL_CONFIG_LAST_ATTRIB EGL_CONFORMANT
+ * struct */
-#define _EGL_CONFIG_NUM_CONTIGUOUS_ATTRIBS \
-   (_EGL_CONFIG_LAST_ATTRIB - _EGL_CONFIG_FIRST_ATTRIB + 1)
-
-/* Attributes outside the contiguous block:
- *
- *   EGL_Y_INVERTED_NOK
- */
-#define _EGL_CONFIG_FIRST_EXTRA_ATTRIB _EGL_CONFIG_NUM_CONTIGUOUS_ATTRIBS
-#define _EGL_CONFIG_NUM_EXTRA_ATTRIBS 1
-
-#define _EGL_CONFIG_NUM_ATTRIBS \
-   _EGL_CONFIG_NUM_CONTIGUOUS_ATTRIBS + _EGL_CONFIG_NUM_EXTRA_ATTRIBS
-
-
 struct _egl_config
 {
    _EGLDisplay *Display;
-   EGLint Storage[_EGL_CONFIG_NUM_ATTRIBS];
+
+   /* core */
+   EGLint BufferSize;
+   EGLint AlphaSize;
+   EGLint BlueSize;
+   EGLint GreenSize;
+   EGLint RedSize;
+   EGLint DepthSize;
+   EGLint StencilSize;
+   EGLint ConfigCaveat;
+   EGLint ConfigID;
+   EGLint Level;
+   EGLint MaxPbufferHeight;
+   EGLint MaxPbufferPixels;
+   EGLint MaxPbufferWidth;
+   EGLint NativeRenderable;
+   EGLint NativeVisualID;
+   EGLint NativeVisualType;
+   EGLint Samples;
+   EGLint SampleBuffers;
+   EGLint SurfaceType;
+   EGLint TransparentType;
+   EGLint TransparentBlueValue;
+   EGLint TransparentGreenValue;
+   EGLint TransparentRedValue;
+   EGLint BindToTextureRGB;
+   EGLint BindToTextureRGBA;
+   EGLint MinSwapInterval;
+   EGLint MaxSwapInterval;
+   EGLint LuminanceSize;
+   EGLint AlphaMaskSize;
+   EGLint ColorBufferType;
+   EGLint RenderableType;
+   EGLint MatchNativePixmap;
+   EGLint Conformant;
+
+   /* extensions */
+   EGLint YInvertedNOK;
 };
 
 
 /**
- * Macros for source level compatibility.
+ * Map an EGL attribute enum to the offset of the member in _EGLConfig.
- */
-#define SET_CONFIG_ATTRIB(CONF, ATTR, VAL) _eglSetConfigKey(CONF, ATTR, VAL)
-#define GET_CONFIG_ATTRIB(CONF, ATTR) _eglGetConfigKey(CONF, ATTR)
-
-
-/**
- * Given a key, return an index into the storage of the config.
- * Return -1 if the key is invalid.
  */
 static INLINE EGLint
-_eglIndexConfig(const _EGLConfig *conf, EGLint key)
+_eglOffsetOfConfig(EGLint attr)
 {
-   (void) conf;
+   switch (attr) {
-   if (key >= _EGL_CONFIG_FIRST_ATTRIB &&
+#define ATTRIB_MAP(attr, memb) case attr: return offsetof(_EGLConfig, memb)
-       key < _EGL_CONFIG_FIRST_ATTRIB + _EGL_CONFIG_NUM_CONTIGUOUS_ATTRIBS)
+   /* core */
-      return key - _EGL_CONFIG_FIRST_ATTRIB;
+   ATTRIB_MAP(EGL_BUFFER_SIZE,               BufferSize);
-   
+   ATTRIB_MAP(EGL_ALPHA_SIZE,                AlphaSize);
-   switch (key) {
+   ATTRIB_MAP(EGL_BLUE_SIZE,                 BlueSize);
-   case EGL_Y_INVERTED_NOK:
+   ATTRIB_MAP(EGL_GREEN_SIZE,                GreenSize);
-      return _EGL_CONFIG_FIRST_EXTRA_ATTRIB;
+   ATTRIB_MAP(EGL_RED_SIZE,                  RedSize);
+   ATTRIB_MAP(EGL_DEPTH_SIZE,                DepthSize);
+   ATTRIB_MAP(EGL_STENCIL_SIZE,              StencilSize);
+   ATTRIB_MAP(EGL_CONFIG_CAVEAT,             ConfigCaveat);
+   ATTRIB_MAP(EGL_CONFIG_ID,                 ConfigID);
+   ATTRIB_MAP(EGL_LEVEL,                     Level);
+   ATTRIB_MAP(EGL_MAX_PBUFFER_HEIGHT,        MaxPbufferHeight);
+   ATTRIB_MAP(EGL_MAX_PBUFFER_PIXELS,        MaxPbufferPixels);
+   ATTRIB_MAP(EGL_MAX_PBUFFER_WIDTH,         MaxPbufferWidth);
+   ATTRIB_MAP(EGL_NATIVE_RENDERABLE,         NativeRenderable);
+   ATTRIB_MAP(EGL_NATIVE_VISUAL_ID,          NativeVisualID);
+   ATTRIB_MAP(EGL_NATIVE_VISUAL_TYPE,        NativeVisualType);
+   ATTRIB_MAP(EGL_SAMPLES,                   Samples);
+   ATTRIB_MAP(EGL_SAMPLE_BUFFERS,            SampleBuffers);
+   ATTRIB_MAP(EGL_SURFACE_TYPE,              SurfaceType);
+   ATTRIB_MAP(EGL_TRANSPARENT_TYPE,          TransparentType);
+   ATTRIB_MAP(EGL_TRANSPARENT_BLUE_VALUE,    TransparentBlueValue);
+   ATTRIB_MAP(EGL_TRANSPARENT_GREEN_VALUE,   TransparentGreenValue);
+   ATTRIB_MAP(EGL_TRANSPARENT_RED_VALUE,     TransparentRedValue);
+   ATTRIB_MAP(EGL_BIND_TO_TEXTURE_RGB,       BindToTextureRGB);
+   ATTRIB_MAP(EGL_BIND_TO_TEXTURE_RGBA,      BindToTextureRGBA);
+   ATTRIB_MAP(EGL_MIN_SWAP_INTERVAL,         MinSwapInterval);
+   ATTRIB_MAP(EGL_MAX_SWAP_INTERVAL,         MaxSwapInterval);
+   ATTRIB_MAP(EGL_LUMINANCE_SIZE,            LuminanceSize);
+   ATTRIB_MAP(EGL_ALPHA_MASK_SIZE,           AlphaMaskSize);
+   ATTRIB_MAP(EGL_COLOR_BUFFER_TYPE,         ColorBufferType);
+   ATTRIB_MAP(EGL_RENDERABLE_TYPE,           RenderableType);
+   ATTRIB_MAP(EGL_MATCH_NATIVE_PIXMAP,       MatchNativePixmap);
+   ATTRIB_MAP(EGL_CONFORMANT,                Conformant);
+   /* extensions */
+   ATTRIB_MAP(EGL_Y_INVERTED_NOK,            YInvertedNOK);
+#undef ATTRIB_MAP
    default:
       return -1;
    }
 }
 
 
-/**
- * Reset all keys in the config to a given value.
- */
-static INLINE void
-_eglResetConfigKeys(_EGLConfig *conf, EGLint val)
-{
-   EGLint i;
-   for (i = 0; i < _EGL_CONFIG_NUM_ATTRIBS; i++)
-      conf->Storage[i] = val;
-}
-
-
 /**
  * Update a config for a given key.
  *
@@ -79,9 +113,9 @@ _eglResetConfigKeys(_EGLConfig *conf, EGLint val)
 static INLINE void
 _eglSetConfigKey(_EGLConfig *conf, EGLint key, EGLint val)
 {
-   EGLint idx = _eglIndexConfig(conf, key);
+   EGLint offset = _eglOffsetOfConfig(key);
-   assert(idx >= 0);
+   assert(offset >= 0);
-   conf->Storage[idx] = val;
+   *((EGLint *) ((char *) conf + offset)) = val;
 }
 
 
@@ -91,9 +125,9 @@ _eglSetConfigKey(_EGLConfig *conf, EGLint key, EGLint val)
 static INLINE EGLint
 _eglGetConfigKey(const _EGLConfig *conf, EGLint key)
 {
-   EGLint idx = _eglIndexConfig(conf, key);
+   EGLint offset = _eglOffsetOfConfig(key);
-   assert(idx >= 0);
+   assert(offset >= 0);
-   return conf->Storage[idx];
+   return *((EGLint *) ((char *) conf + offset));
 }
 
 
@@ -102,34 +136,20 @@ _eglInitConfig(_EGLConfig *config, _EGLDisplay *dpy, EGLint id);
 
 
 PUBLIC EGLConfig
-_eglAddConfig(_EGLDisplay *dpy, _EGLConfig *conf);
+_eglLinkConfig(_EGLConfig *conf);
 
 
-extern EGLBoolean
+extern _EGLConfig *
-_eglCheckConfigHandle(EGLConfig config, _EGLDisplay *dpy);
+_eglLookupConfig(EGLConfig config, _EGLDisplay *dpy);
 
 
 /**
- * Lookup a handle to find the linked config.
+ * Return the handle of a linked config.
- * Return NULL if the handle has no corresponding linked config.
- */
-static INLINE _EGLConfig *
-_eglLookupConfig(EGLConfig config, _EGLDisplay *dpy)
-{
-   _EGLConfig *conf = (_EGLConfig *) config;
-   if (!dpy || !_eglCheckConfigHandle(config, dpy))
-      conf = NULL;
-   return conf;
-}
-
-
-/**
- * Return the handle of a linked config, or NULL.
  */
 static INLINE EGLConfig
 _eglGetConfigHandle(_EGLConfig *conf)
 {
-   return (EGLConfig) ((conf && conf->Display) ? conf : NULL);
+   return (EGLConfig) conf;
 }
 
 
@@ -142,7 +162,8 @@ _eglMatchConfig(const _EGLConfig *conf, const _EGLConfig *criteria);
 
 
 PUBLIC EGLBoolean
-_eglParseConfigAttribList(_EGLConfig *conf, const EGLint *attrib_list);
+_eglParseConfigAttribList(_EGLConfig *conf, _EGLDisplay *dpy,
+                          const EGLint *attrib_list);
 
 
 PUBLIC EGLint

src/egl/main/eglcontext.c

@@ -103,8 +103,7 @@ _eglInitContext(_EGLContext *ctx, _EGLDisplay *dpy, _EGLConfig *conf,
       return EGL_FALSE;
    }
 
-   memset(ctx, 0, sizeof(_EGLContext));
+   _eglInitResource(&ctx->Resource, sizeof(*ctx), dpy);
-   ctx->Resource.Display = dpy;
    ctx->ClientAPI = api;
    ctx->Config = conf;
    ctx->WindowRenderBuffer = EGL_NONE;
@@ -113,13 +112,12 @@ _eglInitContext(_EGLContext *ctx, _EGLDisplay *dpy, _EGLConfig *conf,
 
    err = _eglParseContextAttribList(ctx, attrib_list);
    if (err == EGL_SUCCESS && ctx->Config) {
-      EGLint renderable_type, api_bit;
+      EGLint api_bit;
 
-      renderable_type = GET_CONFIG_ATTRIB(ctx->Config, EGL_RENDERABLE_TYPE);
       api_bit = _eglGetContextAPIBit(ctx);
-      if (!(renderable_type & api_bit)) {
+      if (!(ctx->Config->RenderableType & api_bit)) {
          _eglLog(_EGL_DEBUG, "context api is 0x%x while config supports 0x%x",
-               api_bit, renderable_type);
+               api_bit, ctx->Config->RenderableType);
          err = EGL_BAD_CONFIG;
       }
    }
@@ -130,29 +128,6 @@ _eglInitContext(_EGLContext *ctx, _EGLDisplay *dpy, _EGLConfig *conf,
 }
 
 
-/**
- * Just a placeholder/demo function.  Real driver will never use this!
- */
-_EGLContext *
-_eglCreateContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                  _EGLContext *share_list, const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Default fallback routine - drivers should usually override this.
- */
-EGLBoolean
-_eglDestroyContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
-{
-   if (!_eglIsContextBound(ctx))
-      free(ctx);
-   return EGL_TRUE;
-}
-
-
 #ifdef EGL_VERSION_1_2
 static EGLint
 _eglQueryContextRenderBuffer(_EGLContext *ctx)
@@ -183,7 +158,9 @@ _eglQueryContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *c,
 
    switch (attribute) {
    case EGL_CONFIG_ID:
-      *value = GET_CONFIG_ATTRIB(c->Config, EGL_CONFIG_ID);
+      if (!c->Config)
+         return _eglError(EGL_BAD_ATTRIBUTE, "eglQueryContext");
+      *value = c->Config->ConfigID;
       break;
    case EGL_CONTEXT_CLIENT_VERSION:
       *value = c->ClientVersion;
@@ -272,10 +249,6 @@ _eglCheckMakeCurrent(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read)
    if (!surfaceless && (draw == NULL || read == NULL))
       return _eglError(EGL_BAD_MATCH, "eglMakeCurrent");
 
-   /* context stealing from another thread is not allowed */
-   if (ctx->Binding && ctx->Binding != t)
-      return _eglError(EGL_BAD_ACCESS, "eglMakeCurrent");
-
    /*
     * The spec says
     *
@@ -283,16 +256,23 @@ _eglCheckMakeCurrent(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read)
     * bound to contexts in another thread, an EGL_BAD_ACCESS error is
     * generated."
     *
-    * But it also says
+    * and
     *
     * "at most one context may be bound to a particular surface at a given
     * time"
-    *
-    * The latter is more restrictive so we can check only the latter case.
     */
-   if ((draw && draw->CurrentContext && draw->CurrentContext != ctx) ||
+   if (ctx->Binding && ctx->Binding != t)
-       (read && read->CurrentContext && read->CurrentContext != ctx))
       return _eglError(EGL_BAD_ACCESS, "eglMakeCurrent");
+   if (draw && draw->CurrentContext && draw->CurrentContext != ctx) {
+      if (draw->CurrentContext->Binding != t ||
+          draw->CurrentContext->ClientAPI != ctx->ClientAPI)
+         return _eglError(EGL_BAD_ACCESS, "eglMakeCurrent");
+   }
+   if (read && read->CurrentContext && read->CurrentContext != ctx) {
+      if (read->CurrentContext->Binding != t ||
+          read->CurrentContext->ClientAPI != ctx->ClientAPI)
+         return _eglError(EGL_BAD_ACCESS, "eglMakeCurrent");
+   }
 
    /* simply require the configs to be equal */
    if ((draw && draw->Config != ctx->Config) ||
@@ -323,79 +303,65 @@ _eglCheckMakeCurrent(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read)
 
 /**
  * Bind the context to the current thread and given surfaces.  Return the
- * "orphaned" context and surfaces.  Each argument is both input and output.
+ * previous bound context and surfaces.  The caller should unreference the
+ * returned context and surfaces.
+ *
+ * Making a second call with the resources returned by the first call
+ * unsurprisingly undoes the first call, except for the resouce reference
+ * counts.
  */
 EGLBoolean
-_eglBindContext(_EGLContext **ctx, _EGLSurface **draw, _EGLSurface **read)
+_eglBindContext(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read,
+                _EGLContext **old_ctx,
+                _EGLSurface **old_draw, _EGLSurface **old_read)
 {
    _EGLThreadInfo *t = _eglGetCurrentThread();
-   _EGLContext *newCtx = *ctx, *oldCtx;
+   _EGLContext *prev_ctx;
-   _EGLSurface *newDraw = *draw, *newRead = *read;
+   _EGLSurface *prev_draw, *prev_read;
 
-   if (!_eglCheckMakeCurrent(newCtx, newDraw, newRead))
+   if (!_eglCheckMakeCurrent(ctx, draw, read))
       return EGL_FALSE;
 
+   /* increment refcounts before binding */
+   _eglGetContext(ctx);
+   _eglGetSurface(draw);
+   _eglGetSurface(read);
+
    /* bind the new context */
-   oldCtx = _eglBindContextToThread(newCtx, t);
+   prev_ctx = _eglBindContextToThread(ctx, t);
 
-   /* break old bindings */
+   /* break previous bindings */
-   if (oldCtx) {
+   if (prev_ctx) {
-      *ctx = oldCtx;
+      prev_draw = prev_ctx->DrawSurface;
-      *draw = oldCtx->DrawSurface;
+      prev_read = prev_ctx->ReadSurface;
-      *read = oldCtx->ReadSurface;
 
-      if (*draw)
+      if (prev_draw)
-         (*draw)->CurrentContext = NULL;
+         prev_draw->CurrentContext = NULL;
-      if (*read)
+      if (prev_read)
-         (*read)->CurrentContext = NULL;
+         prev_read->CurrentContext = NULL;
 
-      oldCtx->DrawSurface = NULL;
+      prev_ctx->DrawSurface = NULL;
-      oldCtx->ReadSurface = NULL;
+      prev_ctx->ReadSurface = NULL;
+   }
+   else {
+      prev_draw = prev_read = NULL;
    }
 
    /* establish new bindings */
-   if (newCtx) {
+   if (ctx) {
-      if (newDraw)
+      if (draw)
-         newDraw->CurrentContext = newCtx;
+         draw->CurrentContext = ctx;
-      if (newRead)
+      if (read)
-         newRead->CurrentContext = newCtx;
+         read->CurrentContext = ctx;
 
-      newCtx->DrawSurface = newDraw;
+      ctx->DrawSurface = draw;
-      newCtx->ReadSurface = newRead;
+      ctx->ReadSurface = read;
    }
 
-   /* an old context or surface is not orphaned if it is still bound */
+   assert(old_ctx && old_draw && old_read);
-   if (*ctx == newCtx)
+   *old_ctx = prev_ctx;
-      *ctx = NULL;
+   *old_draw = prev_draw;
-   if (*draw == newDraw || *draw == newRead)
+   *old_read = prev_read;
-      *draw = NULL;
-   if (*read == newDraw || *read == newRead)
-      *read = NULL;
 
    return EGL_TRUE;
 }
-
-
-/**
- * Just a placeholder/demo function.  Drivers should override this.
- */
-EGLBoolean
-_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw,
-                _EGLSurface *read, _EGLContext *ctx)
-{
-   return EGL_FALSE;
-}
-
-
-/**
- * This is defined by the EGL_MESA_copy_context extension.
- */
-EGLBoolean
-_eglCopyContextMESA(_EGLDriver *drv, EGLDisplay dpy, EGLContext source,
-                    EGLContext dest, EGLint mask)
-{
-   /* This function will always have to be overridden/implemented in the
-    * device driver.  If the driver is based on Mesa, use _mesa_copy_context().
-    */
-   return EGL_FALSE;
-}

src/egl/main/eglcontext.h

@@ -34,51 +34,46 @@ _eglInitContext(_EGLContext *ctx, _EGLDisplay *dpy,
                 _EGLConfig *config, const EGLint *attrib_list);
 
 
-extern _EGLContext *
-_eglCreateContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, _EGLContext *share_list, const EGLint *attrib_list);
-
-
-extern EGLBoolean
-_eglDestroyContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx);
-
-
 extern EGLBoolean
 _eglQueryContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx, EGLint attribute, EGLint *value);
 
 
 PUBLIC EGLBoolean
-_eglBindContext(_EGLContext **ctx, _EGLSurface **draw, _EGLSurface **read);
+_eglBindContext(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read,
-
+                _EGLContext **old_ctx,
-
+                _EGLSurface **old_draw, _EGLSurface **old_read);
-extern EGLBoolean
-_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw, _EGLSurface *read, _EGLContext *ctx);
-
-
-extern EGLBoolean
-_eglCopyContextMESA(_EGLDriver *drv, EGLDisplay dpy, EGLContext source, EGLContext dest, EGLint mask);
 
 
 /**
- * Return true if the context is bound to a thread.
+ * Increment reference count for the context.
- *
- * The binding is considered a reference to the context.  Drivers should not
- * destroy a context when it is bound.
  */
-static INLINE EGLBoolean
+static INLINE _EGLContext *
-_eglIsContextBound(_EGLContext *ctx)
+_eglGetContext(_EGLContext *ctx)
 {
-   return (ctx->Binding != NULL);
+   if (ctx)
+      _eglGetResource(&ctx->Resource);
+   return ctx;
 }
 
 
 /**
- * Link a context to a display and return the handle of the link.
+ * Decrement reference count for the context.
+ */
+static INLINE EGLBoolean
+_eglPutContext(_EGLContext *ctx)
+{
+   return (ctx) ? _eglPutResource(&ctx->Resource) : EGL_FALSE;
+}
+
+
+/**
+ * Link a context to its display and return the handle of the link.
  * The handle can be passed to client directly.
  */
 static INLINE EGLContext
-_eglLinkContext(_EGLContext *ctx, _EGLDisplay *dpy)
+_eglLinkContext(_EGLContext *ctx)
 {
-   _eglLinkResource(&ctx->Resource, _EGL_RESOURCE_CONTEXT, dpy);
+   _eglLinkResource(&ctx->Resource, _EGL_RESOURCE_CONTEXT);
    return (EGLContext) ctx;
 }
 
@@ -120,18 +115,4 @@ _eglGetContextHandle(_EGLContext *ctx)
 }
 
 
-/**
- * Return true if the context is linked to a display.
- *
- * The link is considered a reference to the context (the display is owning the
- * context).  Drivers should not destroy a context when it is linked.
- */
-static INLINE EGLBoolean
-_eglIsContextLinked(_EGLContext *ctx)
-{
-   _EGLResource *res = (_EGLResource *) ctx;
-   return (res && _eglIsResourceLinked(res));
-}
-
-
 #endif /* EGLCONTEXT_INCLUDED */

src/egl/main/egldisplay.c

@@ -233,17 +233,53 @@ _eglCheckResource(void *res, _EGLResourceType type, _EGLDisplay *dpy)
 
 
 /**
- * Link a resource to a display.
+ * Initialize a display resource.
  */
 void
-_eglLinkResource(_EGLResource *res, _EGLResourceType type, _EGLDisplay *dpy)
+_eglInitResource(_EGLResource *res, EGLint size, _EGLDisplay *dpy)
 {
-   assert(!res->Display || res->Display == dpy);
+   memset(res, 0, size);
-
    res->Display = dpy;
+   res->RefCount = 1;
+}
+
+
+/**
+ * Increment reference count for the resource.
+ */
+void
+_eglGetResource(_EGLResource *res)
+{
+   assert(res && res->RefCount > 0);
+   /* hopefully a resource is always manipulated with its display locked */
+   res->RefCount++;
+}
+
+
+/**
+ * Decrement reference count for the resource.
+ */
+EGLBoolean
+_eglPutResource(_EGLResource *res)
+{
+   assert(res && res->RefCount > 0);
+   res->RefCount--;
+   return (!res->RefCount);
+}
+
+
+/**
+ * Link a resource to its display.
+ */
+void
+_eglLinkResource(_EGLResource *res, _EGLResourceType type)
+{
+   assert(res->Display);
+
    res->IsLinked = EGL_TRUE;
-   res->Next = dpy->ResourceLists[type];
+   res->Next = res->Display->ResourceLists[type];
-   dpy->ResourceLists[type] = res;
+   res->Display->ResourceLists[type] = res;
+   _eglGetResource(res);
 }
 
 
@@ -270,6 +306,9 @@ _eglUnlinkResource(_EGLResource *res, _EGLResourceType type)
    }
 
    res->Next = NULL;
-   /* do not reset res->Display */
    res->IsLinked = EGL_FALSE;
+   _eglPutResource(res);
+
+   /* We always unlink before destroy.  The driver still owns a reference */
+   assert(res->RefCount);
 }

src/egl/main/egldisplay.h

@@ -40,6 +40,7 @@ struct _egl_resource
    /* which display the resource belongs to */
    _EGLDisplay *Display;
    EGLBoolean IsLinked;
+   EGLint RefCount;
 
    /* used to link resources of the same type */
    _EGLResource *Next;
@@ -162,7 +163,19 @@ _eglGetDisplayHandle(_EGLDisplay *dpy)
 
 
 extern void
-_eglLinkResource(_EGLResource *res, _EGLResourceType type, _EGLDisplay *dpy);
+_eglInitResource(_EGLResource *res, EGLint size, _EGLDisplay *dpy);
+
+
+PUBLIC void
+_eglGetResource(_EGLResource *res);
+
+
+PUBLIC EGLBoolean
+_eglPutResource(_EGLResource *res);
+
+
+extern void
+_eglLinkResource(_EGLResource *res, _EGLResourceType type);
 
 
 extern void

src/egl/main/egldriver.c

@@ -9,18 +9,10 @@
 #include <stdlib.h>
 
 #include "eglstring.h"
-#include "eglconfig.h"
-#include "eglcontext.h"
 #include "egldefines.h"
 #include "egldisplay.h"
 #include "egldriver.h"
 #include "egllog.h"
-#include "eglmisc.h"
-#include "eglmode.h"
-#include "eglscreen.h"
-#include "eglsurface.h"
-#include "eglimage.h"
-#include "eglsync.h"
 #include "eglmutex.h"
 
 #if defined(_EGL_OS_UNIX)
@@ -662,77 +654,6 @@ _eglUnloadDrivers(void)
 }
 
 
-/**
- * Plug all the available fallback routines into the given driver's
- * dispatch table.
- */
-void
-_eglInitDriverFallbacks(_EGLDriver *drv)
-{
-   /* If a pointer is set to NULL, then the device driver _really_ has
-    * to implement it.
-    */
-   drv->API.Initialize = NULL;
-   drv->API.Terminate = NULL;
-
-   drv->API.GetConfigs = _eglGetConfigs;
-   drv->API.ChooseConfig = _eglChooseConfig;
-   drv->API.GetConfigAttrib = _eglGetConfigAttrib;
-
-   drv->API.CreateContext = _eglCreateContext;
-   drv->API.DestroyContext = _eglDestroyContext;
-   drv->API.MakeCurrent = _eglMakeCurrent;
-   drv->API.QueryContext = _eglQueryContext;
-
-   drv->API.CreateWindowSurface = _eglCreateWindowSurface;
-   drv->API.CreatePixmapSurface = _eglCreatePixmapSurface;
-   drv->API.CreatePbufferSurface = _eglCreatePbufferSurface;
-   drv->API.DestroySurface = _eglDestroySurface;
-   drv->API.QuerySurface = _eglQuerySurface;
-   drv->API.SurfaceAttrib = _eglSurfaceAttrib;
-   drv->API.BindTexImage = _eglBindTexImage;
-   drv->API.ReleaseTexImage = _eglReleaseTexImage;
-   drv->API.SwapInterval = _eglSwapInterval;
-   drv->API.SwapBuffers = _eglSwapBuffers;
-   drv->API.CopyBuffers = _eglCopyBuffers;
-
-   drv->API.QueryString = _eglQueryString;
-   drv->API.WaitClient = _eglWaitClient;
-   drv->API.WaitNative = _eglWaitNative;
-
-#ifdef EGL_MESA_screen_surface
-   drv->API.ChooseModeMESA = _eglChooseModeMESA; 
-   drv->API.GetModesMESA = _eglGetModesMESA;
-   drv->API.GetModeAttribMESA = _eglGetModeAttribMESA;
-   drv->API.GetScreensMESA = _eglGetScreensMESA;
-   drv->API.CreateScreenSurfaceMESA = _eglCreateScreenSurfaceMESA;
-   drv->API.ShowScreenSurfaceMESA = _eglShowScreenSurfaceMESA;
-   drv->API.ScreenPositionMESA = _eglScreenPositionMESA;
-   drv->API.QueryScreenMESA = _eglQueryScreenMESA;
-   drv->API.QueryScreenSurfaceMESA = _eglQueryScreenSurfaceMESA;
-   drv->API.QueryScreenModeMESA = _eglQueryScreenModeMESA;
-   drv->API.QueryModeStringMESA = _eglQueryModeStringMESA;
-#endif /* EGL_MESA_screen_surface */
-
-#ifdef EGL_VERSION_1_2
-   drv->API.CreatePbufferFromClientBuffer = _eglCreatePbufferFromClientBuffer;
-#endif /* EGL_VERSION_1_2 */
-
-#ifdef EGL_KHR_image_base
-   drv->API.CreateImageKHR = _eglCreateImageKHR;
-   drv->API.DestroyImageKHR = _eglDestroyImageKHR;
-#endif /* EGL_KHR_image_base */
-
-#ifdef EGL_KHR_reusable_sync
-   drv->API.CreateSyncKHR = _eglCreateSyncKHR;
-   drv->API.DestroySyncKHR = _eglDestroySyncKHR;
-   drv->API.ClientWaitSyncKHR = _eglClientWaitSyncKHR;
-   drv->API.SignalSyncKHR = _eglSignalSyncKHR;
-   drv->API.GetSyncAttribKHR = _eglGetSyncAttribKHR;
-#endif /* EGL_KHR_reusable_sync */
-}
-
-
 /**
  * Invoke a callback function on each EGL search path.
  *

src/egl/main/egldriver.h

@@ -80,6 +80,7 @@ extern void
 _eglUnloadDrivers(void);
 
 
+/* defined in eglfallbacks.c */
 PUBLIC void
 _eglInitDriverFallbacks(_EGLDriver *drv);
 

src/egl/main/eglfallbacks.c

@@ -0,0 +1,99 @@
+#include <string.h>
+#include "egltypedefs.h"
+#include "egldriver.h"
+#include "eglconfig.h"
+#include "eglcontext.h"
+#include "eglsurface.h"
+#include "eglmisc.h"
+#include "eglscreen.h"
+#include "eglmode.h"
+#include "eglsync.h"
+
+
+static EGLBoolean
+_eglReturnFalse(void)
+{
+   return EGL_FALSE;
+}
+
+
+/**
+ * Plug all the available fallback routines into the given driver's
+ * dispatch table.
+ */
+void
+_eglInitDriverFallbacks(_EGLDriver *drv)
+{
+   memset(&drv->API, 0, sizeof(drv->API));
+
+   /* the driver has to implement these */
+   drv->API.Initialize = NULL;
+   drv->API.Terminate = NULL;
+
+   drv->API.GetConfigs = _eglGetConfigs;
+   drv->API.ChooseConfig = _eglChooseConfig;
+   drv->API.GetConfigAttrib = _eglGetConfigAttrib;
+
+   drv->API.CreateContext = (CreateContext_t) _eglReturnFalse;
+   drv->API.DestroyContext = (DestroyContext_t) _eglReturnFalse;
+   drv->API.MakeCurrent = (MakeCurrent_t) _eglReturnFalse;
+   drv->API.QueryContext = _eglQueryContext;
+
+   drv->API.CreateWindowSurface = (CreateWindowSurface_t) _eglReturnFalse;
+   drv->API.CreatePixmapSurface = (CreatePixmapSurface_t) _eglReturnFalse;
+   drv->API.CreatePbufferSurface = (CreatePbufferSurface_t) _eglReturnFalse;
+   drv->API.CreatePbufferFromClientBuffer =
+      (CreatePbufferFromClientBuffer_t) _eglReturnFalse;
+   drv->API.DestroySurface = (DestroySurface_t) _eglReturnFalse;
+   drv->API.QuerySurface = _eglQuerySurface;
+   drv->API.SurfaceAttrib = _eglSurfaceAttrib;
+
+   drv->API.BindTexImage = (BindTexImage_t) _eglReturnFalse;
+   drv->API.ReleaseTexImage = (ReleaseTexImage_t) _eglReturnFalse;
+   drv->API.CopyBuffers = (CopyBuffers_t) _eglReturnFalse;
+   drv->API.SwapBuffers = (SwapBuffers_t) _eglReturnFalse;
+   drv->API.SwapInterval = _eglSwapInterval;
+
+   drv->API.WaitClient = (WaitClient_t) _eglReturnFalse;
+   drv->API.WaitNative = (WaitNative_t) _eglReturnFalse;
+   drv->API.GetProcAddress = (GetProcAddress_t) _eglReturnFalse;
+   drv->API.QueryString = _eglQueryString;
+
+#ifdef EGL_MESA_screen_surface
+   drv->API.CopyContextMESA = (CopyContextMESA_t) _eglReturnFalse;
+   drv->API.CreateScreenSurfaceMESA =
+      (CreateScreenSurfaceMESA_t) _eglReturnFalse;
+   drv->API.ShowScreenSurfaceMESA = (ShowScreenSurfaceMESA_t) _eglReturnFalse;
+   drv->API.ChooseModeMESA = _eglChooseModeMESA;
+   drv->API.GetModesMESA = _eglGetModesMESA;
+   drv->API.GetModeAttribMESA = _eglGetModeAttribMESA;
+   drv->API.GetScreensMESA = _eglGetScreensMESA;
+   drv->API.ScreenPositionMESA = _eglScreenPositionMESA;
+   drv->API.QueryScreenMESA = _eglQueryScreenMESA;
+   drv->API.QueryScreenSurfaceMESA = _eglQueryScreenSurfaceMESA;
+   drv->API.QueryScreenModeMESA = _eglQueryScreenModeMESA;
+   drv->API.QueryModeStringMESA = _eglQueryModeStringMESA;
+#endif /* EGL_MESA_screen_surface */
+
+#ifdef EGL_KHR_image_base
+   drv->API.CreateImageKHR = NULL;
+   drv->API.DestroyImageKHR = NULL;
+#endif /* EGL_KHR_image_base */
+
+#ifdef EGL_KHR_reusable_sync
+   drv->API.CreateSyncKHR = NULL;
+   drv->API.DestroySyncKHR = NULL;
+   drv->API.ClientWaitSyncKHR = NULL;
+   drv->API.SignalSyncKHR = NULL;
+   drv->API.GetSyncAttribKHR = _eglGetSyncAttribKHR;
+#endif /* EGL_KHR_reusable_sync */
+
+#ifdef EGL_MESA_drm_image
+   drv->API.CreateDRMImageMESA = NULL;
+   drv->API.ExportDRMImageMESA = NULL;
+#endif
+
+#ifdef EGL_NOK_swap_region
+   drv->API.SwapBuffersRegionNOK = NULL;
+#endif
+}

src/egl/main/eglimage.c

@@ -2,7 +2,6 @@
 #include <string.h>
 
 #include "eglimage.h"
-#include "eglcurrent.h"
 #include "egllog.h"
 
 
@@ -12,28 +11,57 @@
 /**
  * Parse the list of image attributes and return the proper error code.
  */
-static EGLint
+EGLint
-_eglParseImageAttribList(_EGLImage *img, const EGLint *attrib_list)
+_eglParseImageAttribList(_EGLImageAttribs *attrs, _EGLDisplay *dpy,
+                         const EGLint *attrib_list)
 {
    EGLint i, err = EGL_SUCCESS;
 
+   (void) dpy;
+
+   memset(attrs, 0, sizeof(attrs));
+   attrs->ImagePreserved = EGL_FALSE;
+   attrs->GLTextureLevel = 0;
+   attrs->GLTextureZOffset = 0;
+
    if (!attrib_list)
-      return EGL_SUCCESS;
+      return err;
 
    for (i = 0; attrib_list[i] != EGL_NONE; i++) {
       EGLint attr = attrib_list[i++];
       EGLint val = attrib_list[i];
 
       switch (attr) {
+      /* EGL_KHR_image_base */
       case EGL_IMAGE_PRESERVED_KHR:
-         img->Preserved = val;
+         attrs->ImagePreserved = val;
          break;
+
+      /* EGL_KHR_gl_image */
       case EGL_GL_TEXTURE_LEVEL_KHR:
-         img->GLTextureLevel = val;
+         attrs->GLTextureLevel = val;
          break;
       case EGL_GL_TEXTURE_ZOFFSET_KHR:
-         img->GLTextureZOffset = val;
+         attrs->GLTextureZOffset = val;
          break;
+
+      /* EGL_MESA_drm_image */
+      case EGL_WIDTH:
+         attrs->Width = val;
+         break;
+      case EGL_HEIGHT:
+         attrs->Height = val;
+         break;
+      case EGL_DRM_BUFFER_FORMAT_MESA:
+         attrs->DRMBufferFormatMESA = val;
+         break;
+      case EGL_DRM_BUFFER_USE_MESA:
+         attrs->DRMBufferUseMESA = val;
+         break;
+      case EGL_DRM_BUFFER_STRIDE_MESA:
+         attrs->DRMBufferStrideMESA = val;
+         break;
+
       default:
          /* unknown attrs are ignored */
          break;
@@ -50,41 +78,12 @@ _eglParseImageAttribList(_EGLImage *img, const EGLint *attrib_list)
 
 
 EGLBoolean
-_eglInitImage(_EGLImage *img, _EGLDisplay *dpy, const EGLint *attrib_list)
+_eglInitImage(_EGLImage *img, _EGLDisplay *dpy)
 {
-   EGLint err;
+   _eglInitResource(&img->Resource, sizeof(*img), dpy);
-
-   memset(img, 0, sizeof(_EGLImage));
-   img->Resource.Display = dpy;
-
-   img->Preserved = EGL_FALSE;
-   img->GLTextureLevel = 0;
-   img->GLTextureZOffset = 0;
-
-   err = _eglParseImageAttribList(img, attrib_list);
-   if (err != EGL_SUCCESS)
-      return _eglError(err, "eglCreateImageKHR");
 
    return EGL_TRUE;
 }
 
 
-_EGLImage *
-_eglCreateImageKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx,
-                   EGLenum target, EGLClientBuffer buffer,
-                   const EGLint *attr_list)
-{
-   /* driver should override this function */
-   return NULL;
-}
-
-
-EGLBoolean
-_eglDestroyImageKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLImage *image)
-{
-   /* driver should override this function */
-   return EGL_FALSE;
-}
-
-
 #endif /* EGL_KHR_image_base */

src/egl/main/eglimage.h

@@ -6,6 +6,23 @@
 #include "egldisplay.h"
 
 
+struct _egl_image_attribs
+{
+   /* EGL_KHR_image_base */
+   EGLBoolean ImagePreserved;
+
+   /* EGL_KHR_gl_image */
+   EGLint GLTextureLevel;
+   EGLint GLTextureZOffset;
+
+   /* EGL_MESA_drm_image */
+   EGLint Width;
+   EGLint Height;
+   EGLint DRMBufferFormatMESA;
+   EGLint DRMBufferUseMESA;
+   EGLint DRMBufferStrideMESA;
+};
+
 /**
  * "Base" class for device driver images.
  */
@@ -13,34 +30,48 @@ struct _egl_image
 {
    /* An image is a display resource */
    _EGLResource Resource;
-
-   EGLBoolean Preserved;
-   EGLint GLTextureLevel;
-   EGLint GLTextureZOffset;
 };
 
 
+PUBLIC EGLint
+_eglParseImageAttribList(_EGLImageAttribs *attrs, _EGLDisplay *dpy,
+                         const EGLint *attrib_list);
+
+
 PUBLIC EGLBoolean
-_eglInitImage(_EGLImage *img, _EGLDisplay *dpy, const EGLint *attrib_list);
+_eglInitImage(_EGLImage *img, _EGLDisplay *dpy);
-
-
-extern _EGLImage *
-_eglCreateImageKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx,
-                   EGLenum target, EGLClientBuffer buffer, const EGLint *attr_list);
-
-
-extern EGLBoolean
-_eglDestroyImageKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLImage *image);
 
 
 /**
- * Link an image to a display and return the handle of the link.
+ * Increment reference count for the image.
+ */
+static INLINE _EGLImage *
+_eglGetImage(_EGLImage *img)
+{
+   if (img)
+      _eglGetResource(&img->Resource);
+   return img;
+}
+
+
+/**
+ * Decrement reference count for the image.
+ */
+static INLINE EGLBoolean
+_eglPutImage(_EGLImage *img)
+{
+   return (img) ? _eglPutResource(&img->Resource) : EGL_FALSE;
+}
+
+
+/**
+ * Link an image to its display and return the handle of the link.
  * The handle can be passed to client directly.
  */
 static INLINE EGLImageKHR
-_eglLinkImage(_EGLImage *img, _EGLDisplay *dpy)
+_eglLinkImage(_EGLImage *img)
 {
-   _eglLinkResource(&img->Resource, _EGL_RESOURCE_IMAGE, dpy);
+   _eglLinkResource(&img->Resource, _EGL_RESOURCE_IMAGE);
    return (EGLImageKHR) img;
 }
 
@@ -82,15 +113,4 @@ _eglGetImageHandle(_EGLImage *img)
 }
 
 
-/**
- * Return true if the image is linked to a display.
- */
-static INLINE EGLBoolean
-_eglIsImageLinked(_EGLImage *img)
-{
-   _EGLResource *res = (_EGLResource *) img;
-   return (res && _eglIsResourceLinked(res));
-}
-
-
 #endif /* EGLIMAGE_INCLUDED */

src/egl/main/eglmisc.c

@@ -158,32 +158,3 @@ _eglQueryString(_EGLDriver *drv, _EGLDisplay *dpy, EGLint name)
       return NULL;
    }
 }
-
-
-EGLBoolean
-_eglWaitClient(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
-{
-   /* just a placeholder */
-   (void) drv;
-   (void) dpy;
-   (void) ctx;
-   return EGL_TRUE;
-}
-
-
-EGLBoolean
-_eglWaitNative(_EGLDriver *drv, _EGLDisplay *dpy, EGLint engine)
-{
-   /* just a placeholder */
-   (void) drv;
-   (void) dpy;
-   switch (engine) {
-   case EGL_CORE_NATIVE_ENGINE:
-      break;
-   default:
-      _eglError(EGL_BAD_PARAMETER, "eglWaitNative(engine)");
-      return EGL_FALSE;
-   }
-
-   return EGL_TRUE;
-}

src/egl/main/eglmisc.h

@@ -37,12 +37,4 @@ extern const char *
 _eglQueryString(_EGLDriver *drv, _EGLDisplay *dpy, EGLint name);
 
 
-extern EGLBoolean
-_eglWaitClient(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx);
-
-
-extern EGLBoolean
-_eglWaitNative(_EGLDriver *drv, _EGLDisplay *dpy, EGLint engine);
-
-
 #endif /* EGLMISC_INCLUDED */

src/egl/main/eglmode.c

@@ -3,11 +3,9 @@
 #include <string.h>
 
 #include "egldisplay.h"
-#include "egldriver.h"
 #include "eglmode.h"
 #include "eglcurrent.h"
 #include "eglscreen.h"
-#include "eglstring.h"
 
 
 #ifdef EGL_MESA_screen_surface
@@ -31,56 +29,24 @@ _eglLookupMode(EGLModeMESA mode, _EGLDisplay *disp)
    /* loop over all screens on the display */
    for (scrnum = 0; scrnum < disp->Screens->Size; scrnum++) {
       const _EGLScreen *scrn = disp->Screens->Elements[scrnum];
-      EGLint i;
+      EGLint idx;
-      /* search list of modes for handle */
-      for (i = 0; i < scrn->NumModes; i++) {
-         if (scrn->Modes[i].Handle == mode) {
-            return scrn->Modes + i;
-         }
-      }
-   }
 
-   return NULL;
+      /*
-}
+       * the mode ids of a screen ranges from scrn->Handle to scrn->Handle +
-
+       * scrn->NumModes
-
-/**
- * Add a new mode with the given attributes (width, height, depth, refreshRate)
- * to the given screen.
- * Assign a new mode ID/handle to the mode as well.
- * \return pointer to the new _EGLMode
        */
-_EGLMode *
+      if (mode >= scrn->Handle &&
-_eglAddNewMode(_EGLScreen *screen, EGLint width, EGLint height,
+          mode < scrn->Handle + _EGL_SCREEN_MAX_MODES) {
-               EGLint refreshRate, const char *name)
+         idx = mode - scrn->Handle;
-{
-   EGLint n;
-   _EGLMode *newModes;
 
-   assert(screen);
+         assert(idx < scrn->NumModes && scrn->Modes[idx].Handle == mode);
-   assert(width > 0);
-   assert(height > 0);
-   assert(refreshRate > 0);
 
-   n = screen->NumModes;
+         return &scrn->Modes[idx];
-   newModes = (_EGLMode *) realloc(screen->Modes, (n+1) * sizeof(_EGLMode));
-   if (newModes) {
-      screen->Modes = newModes;
-      screen->Modes[n].Handle = n + 1;
-      screen->Modes[n].Width = width;
-      screen->Modes[n].Height = height;
-      screen->Modes[n].RefreshRate = refreshRate;
-      screen->Modes[n].Optimal = EGL_FALSE;
-      screen->Modes[n].Interlaced = EGL_FALSE;
-      screen->Modes[n].Name = _eglstrdup(name);
-      screen->NumModes++;
-      return screen->Modes + n;
       }
-   else {
+   }
+
    return NULL;
 }
-}
-
 
 
 /**

src/egl/main/eglmode.h

@@ -32,11 +32,6 @@ extern _EGLMode *
 _eglLookupMode(EGLModeMESA mode, _EGLDisplay *dpy);
 
 
-PUBLIC _EGLMode *
-_eglAddNewMode(_EGLScreen *screen, EGLint width, EGLint height,
-               EGLint refreshRate, const char *name);
-
-
 extern EGLBoolean
 _eglChooseModeMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn,
                    const EGLint *attrib_list, EGLModeMESA *modes,

src/egl/main/eglscreen.c

@@ -18,7 +18,6 @@
 #include "egldisplay.h"
 #include "eglcurrent.h"
 #include "eglmode.h"
-#include "eglconfig.h"
 #include "eglsurface.h"
 #include "eglscreen.h"
 #include "eglmutex.h"
@@ -42,7 +41,8 @@ _eglAllocScreenHandle(void)
    EGLScreenMESA s;
 
    _eglLockMutex(&_eglNextScreenHandleMutex);
-   s = _eglNextScreenHandle++;
+   s = _eglNextScreenHandle;
+   _eglNextScreenHandle += _EGL_SCREEN_MAX_MODES;
    _eglUnlockMutex(&_eglNextScreenHandleMutex);
 
    return s;
@@ -53,16 +53,54 @@ _eglAllocScreenHandle(void)
  * Initialize an _EGLScreen object to default values.
  */
 void
-_eglInitScreen(_EGLScreen *screen)
+_eglInitScreen(_EGLScreen *screen, _EGLDisplay *dpy, EGLint num_modes)
 {
    memset(screen, 0, sizeof(_EGLScreen));
+
+   screen->Display = dpy;
+   screen->NumModes = num_modes;
    screen->StepX = 1;
    screen->StepY = 1;
+
+   if (num_modes > _EGL_SCREEN_MAX_MODES)
+      num_modes = _EGL_SCREEN_MAX_MODES;
+   screen->Modes = (_EGLMode *) calloc(num_modes, sizeof(*screen->Modes));
+   screen->NumModes = (screen->Modes) ? num_modes : 0;
 }
 
 
 /**
- * Given a public screen handle, return the internal _EGLScreen object.
+ * Link a screen to its display and return the handle of the link.
+ * The handle can be passed to client directly.
+ */
+EGLScreenMESA
+_eglLinkScreen(_EGLScreen *screen)
+{
+   _EGLDisplay *display;
+   EGLint i;
+
+   assert(screen && screen->Display);
+   display = screen->Display;
+
+   if (!display->Screens) {
+      display->Screens = _eglCreateArray("Screen", 4);
+      if (!display->Screens)
+         return (EGLScreenMESA) 0;
+   }
+
+   screen->Handle = _eglAllocScreenHandle();
+   for (i = 0; i < screen->NumModes; i++)
+      screen->Modes[i].Handle = screen->Handle + i;
+
+   _eglAppendArray(display->Screens, (void *) screen);
+
+   return screen->Handle;
+}
+
+
+/**
+ * Lookup a handle to find the linked config.
+ * Return NULL if the handle has no corresponding linked config.
  */
 _EGLScreen *
 _eglLookupScreen(EGLScreenMESA screen, _EGLDisplay *display)
@@ -74,39 +112,21 @@ _eglLookupScreen(EGLScreenMESA screen, _EGLDisplay *display)
 
    for (i = 0; i < display->Screens->Size; i++) {
       _EGLScreen *scr = (_EGLScreen *) display->Screens->Elements[i];
-      if (scr->Handle == screen)
+      if (scr->Handle == screen) {
+         assert(scr->Display == display);
          return scr;
       }
+   }
    return NULL;
 }
 
 
-/**
- * Add the given _EGLScreen to the display's list of screens.
- */
-void
-_eglAddScreen(_EGLDisplay *display, _EGLScreen *screen)
-{
-   assert(display);
-   assert(screen);
-
-   if (!display->Screens) {
-      display->Screens = _eglCreateArray("Screen", 4);
-      if (!display->Screens)
-         return;
-   }
-   screen->Handle = _eglAllocScreenHandle();
-   _eglAppendArray(display->Screens, (void *) screen);
-}
-
-
-
 static EGLBoolean
 _eglFlattenScreen(void *elem, void *buffer)
 {
    _EGLScreen *scr = (_EGLScreen *) elem;
    EGLScreenMESA *handle = (EGLScreenMESA *) buffer;
-   *handle = scr->Handle;
+   *handle = _eglGetScreenHandle(scr);
    return EGL_TRUE;
 }
 
@@ -122,66 +142,6 @@ _eglGetScreensMESA(_EGLDriver *drv, _EGLDisplay *display, EGLScreenMESA *screens
 }
 
 
-/**
- * Drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreateScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                            const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Show the given surface on the named screen.
- * If surface is EGL_NO_SURFACE, disable the screen's output.
- * 
- * This is just a placeholder function; drivers will always override
- * this with code that _really_ shows the surface.
- */
-EGLBoolean
-_eglShowScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy,
-                          _EGLScreen *scrn, _EGLSurface *surf,
-                          _EGLMode *mode)
-{
-   if (!surf) {
-      scrn->CurrentSurface = NULL;
-   }
-   else {
-      if (surf->Type != EGL_SCREEN_BIT_MESA) {
-         _eglError(EGL_BAD_SURFACE, "eglShowSurfaceMESA");
-         return EGL_FALSE;
-      }
-      if (surf->Width < mode->Width || surf->Height < mode->Height) {
-         _eglError(EGL_BAD_SURFACE,
-                   "eglShowSurfaceMESA(surface smaller than screen size)");
-         return EGL_FALSE;
-      }
-
-      scrn->CurrentSurface = surf;
-      scrn->CurrentMode = mode;
-   }
-   return EGL_TRUE;
-}
-
-
-/**
- * Set a screen's current display mode.
- * Note: mode = EGL_NO_MODE is valid (turns off the screen)
- *
- * This is just a placeholder function; drivers will always override
- * this with code that _really_ sets the mode.
- */
-EGLBoolean
-_eglScreenModeMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn,
-                   _EGLMode *m)
-{
-   scrn->CurrentMode = m;
-   return EGL_TRUE;
-}
-
-
 /**
  * Set a screen's surface origin.
  */
@@ -242,33 +202,4 @@ _eglQueryScreenMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn,
 }
 
 
-/**
- * Delete the modes associated with given screen.
- */
-void
-_eglDestroyScreenModes(_EGLScreen *scrn)
-{
-   EGLint i;
-   for (i = 0; i < scrn->NumModes; i++) {
-      if (scrn->Modes[i].Name)
-         free((char *) scrn->Modes[i].Name); /* cast away const */
-   }
-   if (scrn->Modes)
-      free(scrn->Modes);
-   scrn->Modes = NULL;
-   scrn->NumModes = 0;
-}
-
-      
-/**
- * Default fallback routine - drivers should usually override this.
- */
-void
-_eglDestroyScreen(_EGLScreen *scrn)
-{
-   _eglDestroyScreenModes(scrn);
-   free(scrn);
-}
-
-
 #endif /* EGL_MESA_screen_surface */

src/egl/main/eglscreen.h

@@ -8,6 +8,9 @@
 #ifdef EGL_MESA_screen_surface
 
 
+#define _EGL_SCREEN_MAX_MODES 16
+
+
 /**
  * Per-screen information.
  * Note that an EGL screen doesn't have a size.  A screen may be set to
@@ -19,6 +22,8 @@
  */
 struct _egl_screen
 {
+   _EGLDisplay *Display;
+
    EGLScreenMESA Handle; /* The public/opaque handle which names this object */
 
    _EGLMode *CurrentMode;
@@ -33,41 +38,35 @@ struct _egl_screen
 
 
 PUBLIC void
-_eglInitScreen(_EGLScreen *screen);
+_eglInitScreen(_EGLScreen *screen, _EGLDisplay *dpy, EGLint num_modes);
+
+
+PUBLIC EGLScreenMESA
+_eglLinkScreen(_EGLScreen *screen);
 
 
 extern _EGLScreen *
 _eglLookupScreen(EGLScreenMESA screen, _EGLDisplay *dpy);
 
 
-PUBLIC void
+/**
-_eglAddScreen(_EGLDisplay *display, _EGLScreen *screen);
+ * Return the handle of a linked screen.
+ */
+static INLINE EGLScreenMESA
+_eglGetScreenHandle(_EGLScreen *screen)
+{
+   return (screen) ? screen->Handle : (EGLScreenMESA) 0;
+}
 
 
 extern EGLBoolean
 _eglGetScreensMESA(_EGLDriver *drv, _EGLDisplay *dpy, EGLScreenMESA *screens, EGLint max_screens, EGLint *num_screens);
 
 
-extern _EGLSurface *
-_eglCreateScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list);
-
-
-extern EGLBoolean
-_eglShowScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn, _EGLSurface *surf, _EGLMode *m);
-
-
-extern EGLBoolean
-_eglScreenModeMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn, _EGLMode *m);
-
-
 extern EGLBoolean
 _eglScreenPositionMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn, EGLint x, EGLint y);
 
 
-extern EGLBoolean
-_eglQueryDisplayMESA(_EGLDriver *drv, _EGLDisplay *dpy, EGLint attribute, EGLint *value);
-
-
 extern EGLBoolean
 _eglQueryScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy,
                            _EGLScreen *scrn, _EGLSurface **surface);
@@ -81,14 +80,6 @@ extern EGLBoolean
 _eglQueryScreenMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn, EGLint attribute, EGLint *value);
 
 
-extern void
-_eglDestroyScreenModes(_EGLScreen *scrn);
-
-
-PUBLIC void
-_eglDestroyScreen(_EGLScreen *scrn);
-
-
 #endif /* EGL_MESA_screen_surface */
 
 

src/egl/main/eglsurface.c

@@ -17,12 +17,12 @@
 static void
 _eglClampSwapInterval(_EGLSurface *surf, EGLint interval)
 {
-   EGLint bound = GET_CONFIG_ATTRIB(surf->Config, EGL_MAX_SWAP_INTERVAL);
+   EGLint bound = surf->Config->MaxSwapInterval;
    if (interval >= bound) {
       interval = bound;
    }
    else {
-      bound = GET_CONFIG_ATTRIB(surf->Config, EGL_MIN_SWAP_INTERVAL);
+      bound = surf->Config->MinSwapInterval;
       if (interval < bound)
          interval = bound;
    }
@@ -263,14 +263,13 @@ _eglInitSurface(_EGLSurface *surf, _EGLDisplay *dpy, EGLint type,
       return EGL_FALSE;
    }
 
-   if ((GET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE) & type) == 0) {
+   if ((conf->SurfaceType & type) == 0) {
       /* The config can't be used to create a surface of this type */
       _eglError(EGL_BAD_CONFIG, func);
       return EGL_FALSE;
    }
 
-   memset(surf, 0, sizeof(_EGLSurface));
+   _eglInitResource(&surf->Resource, sizeof(*surf), dpy);
-   surf->Resource.Display = dpy;
    surf->Type = type;
    surf->Config = conf;
 
@@ -303,24 +302,6 @@ _eglInitSurface(_EGLSurface *surf, _EGLDisplay *dpy, EGLint type,
 }
 
 
-EGLBoolean
-_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
-{
-   /* Drivers have to do the actual buffer swap.  */
-   return EGL_TRUE;
-}
-
-
-EGLBoolean
-_eglCopyBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
-                EGLNativePixmapType target)
-{
-   /* copy surface to native pixmap */
-   /* All implementation burdon for this is in the device driver */
-   return EGL_FALSE;
-}
-
-
 EGLBoolean
 _eglQuerySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
                  EGLint attribute, EGLint *value)
@@ -333,7 +314,7 @@ _eglQuerySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
       *value = surface->Height;
       break;
    case EGL_CONFIG_ID:
-      *value = GET_CONFIG_ATTRIB(surface->Config, EGL_CONFIG_ID);
+      *value = surface->Config->ConfigID;
       break;
    case EGL_LARGEST_PBUFFER:
       *value = surface->LargestPbuffer;
@@ -388,51 +369,6 @@ _eglQuerySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
 }
 
 
-/**
- * Drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreateWindowSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                        EGLNativeWindowType window, const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreatePixmapSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                        EGLNativePixmapType pixmap, const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreatePbufferSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                         const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Default fallback routine - drivers should usually override this.
- */
-EGLBoolean
-_eglDestroySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
-{
-   if (!_eglIsSurfaceBound(surf))
-      free(surf);
-   return EGL_TRUE;
-}
-
-
 /**
  * Default fallback routine - drivers might override this.
  */
@@ -445,7 +381,7 @@ _eglSurfaceAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
 
    switch (attribute) {
    case EGL_MIPMAP_LEVEL:
-      confval = GET_CONFIG_ATTRIB(surface->Config, EGL_RENDERABLE_TYPE);
+      confval = surface->Config->RenderableType;
       if (!(confval & (EGL_OPENGL_ES_BIT | EGL_OPENGL_ES2_BIT))) {
          err = EGL_BAD_PARAMETER;
          break;
@@ -457,7 +393,7 @@ _eglSurfaceAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
       case EGL_MULTISAMPLE_RESOLVE_DEFAULT:
          break;
       case EGL_MULTISAMPLE_RESOLVE_BOX:
-         confval = GET_CONFIG_ATTRIB(surface->Config, EGL_SURFACE_TYPE);
+         confval = surface->Config->SurfaceType;
          if (!(confval & EGL_MULTISAMPLE_RESOLVE_BOX_BIT))
             err = EGL_BAD_MATCH;
          break;
@@ -474,7 +410,7 @@ _eglSurfaceAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
       case EGL_BUFFER_DESTROYED:
          break;
       case EGL_BUFFER_PRESERVED:
-         confval = GET_CONFIG_ATTRIB(surface->Config, EGL_SURFACE_TYPE);
+         confval = surface->Config->SurfaceType;
          if (!(confval & EGL_SWAP_BEHAVIOR_PRESERVED_BIT))
             err = EGL_BAD_MATCH;
          break;
@@ -536,40 +472,6 @@ _eglBindTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
 }
 
 
-EGLBoolean
-_eglReleaseTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
-                    EGLint buffer)
-{
-   /* Just do basic error checking and return success/fail.
-    * Drivers must implement the real stuff.
-    */
-
-   if (surface->Type != EGL_PBUFFER_BIT) {
-      _eglError(EGL_BAD_SURFACE, "eglBindTexImage");
-      return EGL_FALSE;
-   }
-
-   if (surface->TextureFormat == EGL_NO_TEXTURE) {
-      _eglError(EGL_BAD_MATCH, "eglBindTexImage");
-      return EGL_FALSE;
-   }
-
-   if (buffer != EGL_BACK_BUFFER) {
-      _eglError(EGL_BAD_PARAMETER, "eglReleaseTexImage");
-      return EGL_FALSE;
-   }
-
-   if (!surface->BoundToTexture) {
-      _eglError(EGL_BAD_SURFACE, "eglReleaseTexImage");
-      return EGL_FALSE;
-   }
-
-   surface->BoundToTexture = EGL_FALSE;
-
-   return EGL_TRUE;
-}
-
-
 EGLBoolean
 _eglSwapInterval(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
                  EGLint interval)
@@ -577,24 +479,3 @@ _eglSwapInterval(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
    _eglClampSwapInterval(surf, interval);
    return EGL_TRUE;
 }
-
-
-#ifdef EGL_VERSION_1_2
-
-/**
- * Example function - drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreatePbufferFromClientBuffer(_EGLDriver *drv, _EGLDisplay *dpy,
-                                  EGLenum buftype, EGLClientBuffer buffer,
-                                  _EGLConfig *conf, const EGLint *attrib_list)
-{
-   if (buftype != EGL_OPENVG_IMAGE) {
-      _eglError(EGL_BAD_PARAMETER, "eglCreatePbufferFromClientBuffer");
-      return NULL;
-   }
-
-   return NULL;
-}
-
-#endif /* EGL_VERSION_1_2 */

src/egl/main/eglsurface.h

@@ -51,34 +51,10 @@ _eglInitSurface(_EGLSurface *surf, _EGLDisplay *dpy, EGLint type,
                 _EGLConfig *config, const EGLint *attrib_list);
 
 
-extern EGLBoolean
-_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf);
-
-
-extern EGLBoolean
-_eglCopyBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLNativePixmapType target);
-
-
 extern EGLBoolean
 _eglQuerySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint attribute, EGLint *value);
 
 
-extern _EGLSurface *
-_eglCreateWindowSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, EGLNativeWindowType window, const EGLint *attrib_list);
-
-
-extern _EGLSurface *
-_eglCreatePixmapSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, EGLNativePixmapType pixmap, const EGLint *attrib_list);
-
-
-extern _EGLSurface *
-_eglCreatePbufferSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list);
-
-
-extern EGLBoolean
-_eglDestroySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf);
-
-
 extern EGLBoolean
 _eglSurfaceAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint attribute, EGLint value);
 
@@ -87,45 +63,40 @@ PUBLIC extern EGLBoolean
 _eglBindTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint buffer);
 
 
-extern EGLBoolean
-_eglReleaseTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint buffer);
-
-
 extern EGLBoolean
 _eglSwapInterval(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint interval);
 
 
-#ifdef EGL_VERSION_1_2
-
-extern _EGLSurface *
-_eglCreatePbufferFromClientBuffer(_EGLDriver *drv, _EGLDisplay *dpy,
-                                  EGLenum buftype, EGLClientBuffer buffer,
-                                  _EGLConfig *conf, const EGLint *attrib_list);
-
-#endif /* EGL_VERSION_1_2 */
-
-
 /**
- * Return true if there is a context bound to the surface.
+ * Increment reference count for the surface.
- *
- * The binding is considered a reference to the surface.  Drivers should not
- * destroy a surface when it is bound.
  */
-static INLINE EGLBoolean
+static INLINE _EGLSurface *
-_eglIsSurfaceBound(_EGLSurface *surf)
+_eglGetSurface(_EGLSurface *surf)
 {
-   return (surf->CurrentContext != NULL);
+   if (surf)
+      _eglGetResource(&surf->Resource);
+   return surf;
 }
 
 
 /**
- * Link a surface to a display and return the handle of the link.
+ * Decrement reference count for the surface.
+ */
+static INLINE EGLBoolean
+_eglPutSurface(_EGLSurface *surf)
+{
+   return (surf) ? _eglPutResource(&surf->Resource) : EGL_FALSE;
+}
+
+
+/**
+ * Link a surface to its display and return the handle of the link.
  * The handle can be passed to client directly.
  */
 static INLINE EGLSurface
-_eglLinkSurface(_EGLSurface *surf, _EGLDisplay *dpy)
+_eglLinkSurface(_EGLSurface *surf)
 {
-   _eglLinkResource(&surf->Resource, _EGL_RESOURCE_SURFACE, dpy);
+   _eglLinkResource(&surf->Resource, _EGL_RESOURCE_SURFACE);
    return (EGLSurface) surf;
 }
 
@@ -167,18 +138,4 @@ _eglGetSurfaceHandle(_EGLSurface *surf)
 }
 
 
-/**
- * Return true if the surface is linked to a display.
- *
- * The link is considered a reference to the surface (the display is owning the
- * surface).  Drivers should not destroy a surface when it is linked.
- */
-static INLINE EGLBoolean
-_eglIsSurfaceLinked(_EGLSurface *surf)
-{
-   _EGLResource *res = (_EGLResource *) surf;
-   return (res && _eglIsResourceLinked(res));
-}
-
-
 #endif /* EGLSURFACE_INCLUDED */

src/egl/main/eglsync.c

@@ -50,10 +50,7 @@ _eglInitSync(_EGLSync *sync, _EGLDisplay *dpy, EGLenum type,
        !(type == EGL_SYNC_FENCE_KHR && dpy->Extensions.KHR_fence_sync))
       return _eglError(EGL_BAD_ATTRIBUTE, "eglCreateSyncKHR");
 
-   memset(sync, 0, sizeof(*sync));
+   _eglInitResource(&sync->Resource, sizeof(*sync), dpy);
-
-   sync->Resource.Display = dpy;
-
    sync->Type = type;
    sync->SyncStatus = EGL_UNSIGNALED_KHR;
    sync->SyncCondition = EGL_SYNC_PRIOR_COMMANDS_COMPLETE_KHR;
@@ -66,37 +63,6 @@ _eglInitSync(_EGLSync *sync, _EGLDisplay *dpy, EGLenum type,
 }
 
 
-_EGLSync *
-_eglCreateSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy,
-                  EGLenum type, const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-EGLBoolean
-_eglDestroySyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync)
-{
-   return EGL_TRUE;
-}
-
-
-EGLint
-_eglClientWaitSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
-                      EGLint flags, EGLTimeKHR timeout)
-{
-   return EGL_FALSE;
-}
-
-
-EGLBoolean
-_eglSignalSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
-                  EGLenum mode)
-{
-   return EGL_FALSE;
-}
-
-
 EGLBoolean
 _eglGetSyncAttribKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
                      EGLint attribute, EGLint *value)

src/egl/main/eglsync.h

@@ -28,38 +28,41 @@ _eglInitSync(_EGLSync *sync, _EGLDisplay *dpy, EGLenum type,
              const EGLint *attrib_list);
 
 
-extern _EGLSync *
-_eglCreateSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy,
-                  EGLenum type, const EGLint *attrib_list);
-
-
-extern EGLBoolean
-_eglDestroySyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync);
-
-
-extern EGLint
-_eglClientWaitSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
-                      EGLint flags, EGLTimeKHR timeout);
-
-
-extern EGLBoolean
-_eglSignalSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
-                  EGLenum mode);
-
-
 extern EGLBoolean
 _eglGetSyncAttribKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
                      EGLint attribute, EGLint *value);
 
 
 /**
- * Link a sync to a display and return the handle of the link.
+ * Increment reference count for the sync.
+ */
+static INLINE _EGLSync *
+_eglGetSync(_EGLSync *sync)
+{
+   if (sync)
+      _eglGetResource(&sync->Resource);
+   return sync;
+}
+
+
+/**
+ * Decrement reference count for the sync.
+ */
+static INLINE EGLBoolean
+_eglPutSync(_EGLSync *sync)
+{
+   return (sync) ? _eglPutResource(&sync->Resource) : EGL_FALSE;
+}
+
+
+/**
+ * Link a sync to its display and return the handle of the link.
  * The handle can be passed to client directly.
  */
 static INLINE EGLSyncKHR
-_eglLinkSync(_EGLSync *sync, _EGLDisplay *dpy)
+_eglLinkSync(_EGLSync *sync)
 {
-   _eglLinkResource(&sync->Resource, _EGL_RESOURCE_SYNC, dpy);
+   _eglLinkResource(&sync->Resource, _EGL_RESOURCE_SYNC);
    return (EGLSyncKHR) sync;
 }
 
@@ -100,20 +103,6 @@ _eglGetSyncHandle(_EGLSync *sync)
 }
 
 
-/**
- * Return true if the sync is linked to a display.
- *
- * The link is considered a reference to the sync (the display is owning the
- * sync).  Drivers should not destroy a sync when it is linked.
- */
-static INLINE EGLBoolean
-_eglIsSyncLinked(_EGLSync *sync)
-{
-   _EGLResource *res = (_EGLResource *) sync;
-   return (res && _eglIsResourceLinked(res));
-}
-
-
 #endif /* EGL_KHR_reusable_sync */
 
 

src/egl/main/egltypedefs.h

@@ -24,6 +24,8 @@ typedef struct _egl_extensions _EGLExtensions;
 
 typedef struct _egl_image _EGLImage;
 
+typedef struct _egl_image_attribs _EGLImageAttribs;
+
 typedef struct _egl_mode _EGLMode;
 
 typedef struct _egl_resource _EGLResource;

src/gallium/auxiliary/Makefile

@@ -173,6 +173,7 @@ GALLIVM_SOURCES = \
         gallivm/lp_bld_struct.c \
         gallivm/lp_bld_swizzle.c \
         gallivm/lp_bld_tgsi_aos.c \
+        gallivm/lp_bld_tgsi_info.c \
         gallivm/lp_bld_tgsi_soa.c \
         gallivm/lp_bld_type.c \
         draw/draw_llvm.c \
@@ -207,16 +208,16 @@ include ../Makefile.template
 
 
 indices/u_indices_gen.c: indices/u_indices_gen.py
-	python $< > $@
+	$(PYTHON2) $< > $@
 
 indices/u_unfilled_gen.c: indices/u_unfilled_gen.py
-	python $< > $@
+	$(PYTHON2) $< > $@
 
 util/u_format_srgb.c: util/u_format_srgb.py
-	python $< > $@
+	$(PYTHON2) $< > $@
 
 util/u_format_table.c: util/u_format_table.py util/u_format_pack.py util/u_format_parse.py util/u_format.csv
-	python util/u_format_table.py util/u_format.csv > $@
+	$(PYTHON2) util/u_format_table.py util/u_format.csv > $@
 
 util/u_half.c: util/u_half.py
-	python util/u_half.py > $@
+	$(PYTHON2) util/u_half.py > $@

src/gallium/auxiliary/SConscript

@@ -225,6 +225,7 @@ if env['llvm']:
     'gallivm/lp_bld_struct.c',
     'gallivm/lp_bld_swizzle.c',
     'gallivm/lp_bld_tgsi_aos.c',
+    'gallivm/lp_bld_tgsi_info.c',
     'gallivm/lp_bld_tgsi_soa.c',
     'gallivm/lp_bld_type.c',
     'draw/draw_llvm.c',

src/gallium/auxiliary/draw/draw_context.c

@@ -335,6 +335,7 @@ draw_set_mapped_constant_buffer(struct draw_context *draw,
    case PIPE_SHADER_VERTEX:
       draw->pt.user.vs_constants[slot] = buffer;
       draw->pt.user.vs_constants_size[slot] = size;
+      draw->pt.user.planes = (float (*) [12][4]) &(draw->plane[0]);
       draw_vs_set_constants(draw, slot, buffer, size);
       break;
    case PIPE_SHADER_GEOMETRY:
@@ -721,9 +722,9 @@ draw_set_mapped_texture(struct draw_context *draw,
                         unsigned sampler_idx,
                         uint32_t width, uint32_t height, uint32_t depth,
                         uint32_t last_level,
-                        uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
+                        uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
-                        uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
+                        uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
-                        const void *data[DRAW_MAX_TEXTURE_LEVELS])
+                        const void *data[PIPE_MAX_TEXTURE_LEVELS])
 {
 #ifdef HAVE_LLVM
    if(draw->llvm)

src/gallium/auxiliary/draw/draw_context.h

@@ -49,7 +49,6 @@ struct draw_geometry_shader;
 struct draw_fragment_shader;
 struct tgsi_sampler;
 
-#define DRAW_MAX_TEXTURE_LEVELS 13  /* 4K x 4K for now */
 
 struct draw_context *draw_create( struct pipe_context *pipe );
 
@@ -120,9 +119,9 @@ draw_set_mapped_texture(struct draw_context *draw,
                         unsigned sampler_idx,
                         uint32_t width, uint32_t height, uint32_t depth,
                         uint32_t last_level,
-                        uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
+                        uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
-                        uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
+                        uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
-                        const void *data[DRAW_MAX_TEXTURE_LEVELS]);
+                        const void *data[PIPE_MAX_TEXTURE_LEVELS]);
 
 
 /*

src/gallium/auxiliary/draw/draw_llvm.c

@@ -31,6 +31,9 @@
 #include "draw_vs.h"
 
 #include "gallivm/lp_bld_arit.h"
+#include "gallivm/lp_bld_logic.h"
+#include "gallivm/lp_bld_const.h"
+#include "gallivm/lp_bld_swizzle.h"
 #include "gallivm/lp_bld_struct.h"
 #include "gallivm/lp_bld_type.h"
 #include "gallivm/lp_bld_flow.h"
@@ -43,7 +46,6 @@
 #include "tgsi/tgsi_exec.h"
 #include "tgsi/tgsi_dump.h"
 
-#include "util/u_cpu_detect.h"
 #include "util/u_math.h"
 #include "util/u_pointer.h"
 #include "util/u_string.h"
@@ -72,12 +74,12 @@ init_globals(struct draw_llvm *llvm)
       elem_types[DRAW_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
       elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
       elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), DRAW_MAX_TEXTURE_LEVELS);
+         LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS);
       elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), DRAW_MAX_TEXTURE_LEVELS);
+         LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS);
       elem_types[DRAW_JIT_TEXTURE_DATA] =
          LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
-                       DRAW_MAX_TEXTURE_LEVELS);
+                       PIPE_MAX_TEXTURE_LEVELS);
       elem_types[DRAW_JIT_TEXTURE_MIN_LOD] = LLVMFloatType();
       elem_types[DRAW_JIT_TEXTURE_MAX_LOD] = LLVMFloatType();
       elem_types[DRAW_JIT_TEXTURE_LOD_BIAS] = LLVMFloatType();
@@ -128,12 +130,14 @@ init_globals(struct draw_llvm *llvm)
 
    /* struct draw_jit_context */
    {
-      LLVMTypeRef elem_types[3];
+      LLVMTypeRef elem_types[5];
       LLVMTypeRef context_type;
 
       elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */
-      elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */
+      elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* gs_constants */
-      elem_types[2] = LLVMArrayType(texture_type,
+      elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(LLVMFloatType(), 4), 12), 0); /* planes */
+      elem_types[3] = LLVMPointerType(LLVMFloatType(), 0); /* viewport */
+      elem_types[4] = LLVMArrayType(texture_type,
                                     PIPE_MAX_VERTEX_SAMPLERS); /* textures */
 
       context_type = LLVMStructType(elem_types, Elements(elem_types), 0);
@@ -142,6 +146,8 @@ init_globals(struct draw_llvm *llvm)
                              llvm->target, context_type, 0);
       LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, gs_constants,
                              llvm->target, context_type, 1);
+      LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes,
+                             llvm->target, context_type, 2);
       LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures,
                              llvm->target, context_type,
                              DRAW_JIT_CTX_TEXTURES);
@@ -267,13 +273,7 @@ draw_llvm_create(struct draw_context *draw)
          LLVMAddConstantPropagationPass(llvm->pass);
       }
 
-      if(util_cpu_caps.has_sse4_1) {
-         /* FIXME: There is a bug in this pass, whereby the combination of fptosi
-          * and sitofp (necessary for trunc/floor/ceil/round implementation)
-          * somehow becomes invalid code.
-          */
       LLVMAddInstructionCombiningPass(llvm->pass);
-      }
       LLVMAddGVNPass(llvm->pass);
    } else {
       /* We need at least this pass to prevent the backends to fail in
@@ -421,7 +421,7 @@ generate_fetch(LLVMBuilderRef builder,
                             "instance_divisor");
    }
 
-   /* limit index to min(inex, vb_max_index) */
+   /* limit index to min(index, vb_max_index) */
    cond = LLVMBuildICmp(builder, LLVMIntULE, index, vb_max_index, "");
    index = LLVMBuildSelect(builder, cond, index, vb_max_index, "");
 
@@ -550,19 +550,28 @@ static void
 store_aos(LLVMBuilderRef builder,
           LLVMValueRef io_ptr,
           LLVMValueRef index,
-          LLVMValueRef value)
+          LLVMValueRef value,
+          LLVMValueRef clipmask)
 {
    LLVMValueRef id_ptr = draw_jit_header_id(builder, io_ptr);
    LLVMValueRef data_ptr = draw_jit_header_data(builder, io_ptr);
    LLVMValueRef indices[3];
+   LLVMValueRef val, shift;
 
    indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
    indices[1] = index;
    indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0);
 
-   /* undefined vertex */
+   /* initialize vertex id:16 = 0xffff, pad:3 = 0, edgeflag:1 = 1 */
-   LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(),
+   val = LLVMConstInt(LLVMInt32Type(), 0xffff1, 0); 
-                                        0xffff, 0), id_ptr);
+   shift  = LLVMConstInt(LLVMInt32Type(), 12, 0);          
+   val = LLVMBuildShl(builder, val, shift, "");
+   /* add clipmask:12 */   
+   val = LLVMBuildOr(builder, val, clipmask, "");               
+
+   /* store vertex header */
+   LLVMBuildStore(builder, val, id_ptr);
+
 
 #if DEBUG_STORE
    lp_build_printf(builder, "    ---- %p storing attribute %d (io = %p)\n", data_ptr, index, io_ptr);
@@ -617,7 +626,8 @@ store_aos_array(LLVMBuilderRef builder,
                 LLVMValueRef io_ptr,
                 LLVMValueRef aos[NUM_CHANNELS],
                 int attrib,
-                int num_outputs)
+                int num_outputs,
+                LLVMValueRef clipmask)
 {
    LLVMValueRef attr_index = LLVMConstInt(LLVMInt32Type(), attrib, 0);
    LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
@@ -625,6 +635,7 @@ store_aos_array(LLVMBuilderRef builder,
    LLVMValueRef ind2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
    LLVMValueRef ind3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
    LLVMValueRef io0_ptr, io1_ptr, io2_ptr, io3_ptr;
+   LLVMValueRef clipmask0, clipmask1, clipmask2, clipmask3;
    
    debug_assert(NUM_CHANNELS == 4);
 
@@ -637,21 +648,31 @@ store_aos_array(LLVMBuilderRef builder,
    io3_ptr = LLVMBuildGEP(builder, io_ptr,
                           &ind3, 1, "");
 
-#if DEBUG_STORE
+   clipmask0 = LLVMBuildExtractElement(builder, clipmask,
-   lp_build_printf(builder, "   io = %p, indexes[%d, %d, %d, %d]\n",
+                                       ind0, "");
-                   io_ptr, ind0, ind1, ind2, ind3);
+   clipmask1 = LLVMBuildExtractElement(builder, clipmask,
-#endif
+                                       ind1, "");
+   clipmask2 = LLVMBuildExtractElement(builder, clipmask,
+                                       ind2, "");
+   clipmask3 = LLVMBuildExtractElement(builder, clipmask,
+                                       ind3, "");
 
-   store_aos(builder, io0_ptr, attr_index, aos[0]);
+#if DEBUG_STORE
-   store_aos(builder, io1_ptr, attr_index, aos[1]);
+   lp_build_printf(builder, "io = %p, indexes[%d, %d, %d, %d]\n, clipmask0 = %x, clipmask1 = %x, clipmask2 = %x, clipmask3 = %x\n",
-   store_aos(builder, io2_ptr, attr_index, aos[2]);
+                   io_ptr, ind0, ind1, ind2, ind3, clipmask0, clipmask1, clipmask2, clipmask3);
-   store_aos(builder, io3_ptr, attr_index, aos[3]);
+#endif
+   /* store for each of the 4 vertices */
+   store_aos(builder, io0_ptr, attr_index, aos[0], clipmask0);
+   store_aos(builder, io1_ptr, attr_index, aos[1], clipmask1);
+   store_aos(builder, io2_ptr, attr_index, aos[2], clipmask2);
+   store_aos(builder, io3_ptr, attr_index, aos[3], clipmask3);
 }
 
 static void
 convert_to_aos(LLVMBuilderRef builder,
                LLVMValueRef io,
                LLVMValueRef (*outputs)[NUM_CHANNELS],
+               LLVMValueRef clipmask,
                int num_outputs,
                int max_vertices)
 {
@@ -680,13 +701,305 @@ convert_to_aos(LLVMBuilderRef builder,
                       io,
                       aos,
                       attrib,
-                      num_outputs);
+                      num_outputs,
+                      clipmask);
    }
 #if DEBUG_STORE
    lp_build_printf(builder, "   # storing end\n");
 #endif
 }
 
+/*
+ * Stores original vertex positions in clip coordinates
+ * There is probably a more efficient way to do this, 4 floats at once
+ * rather than extracting each element one by one.
+ */
+static void
+store_clip(LLVMBuilderRef builder,
+           LLVMValueRef io_ptr,           
+           LLVMValueRef (*outputs)[NUM_CHANNELS])
+{
+   LLVMValueRef out[4];
+   LLVMValueRef indices[2]; 
+   LLVMValueRef io0_ptr, io1_ptr, io2_ptr, io3_ptr;
+   LLVMValueRef clip_ptr0, clip_ptr1, clip_ptr2, clip_ptr3;
+   LLVMValueRef clip0_ptr, clip1_ptr, clip2_ptr, clip3_ptr;    
+   LLVMValueRef out0elem, out1elem, out2elem, out3elem;
+   int i;
+
+   LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   LLVMValueRef ind1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
+   LLVMValueRef ind2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
+   LLVMValueRef ind3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+   
+   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[1] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   
+   out[0] = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/
+   out[1] = LLVMBuildLoad(builder, outputs[0][1], ""); /*y0 y1 y2 y3*/
+   out[2] = LLVMBuildLoad(builder, outputs[0][2], ""); /*z0 z1 z2 z3*/
+   out[3] = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/  
+
+   io0_ptr = LLVMBuildGEP(builder, io_ptr, &ind0, 1, "");
+   io1_ptr = LLVMBuildGEP(builder, io_ptr, &ind1, 1, "");
+   io2_ptr = LLVMBuildGEP(builder, io_ptr, &ind2, 1, "");
+   io3_ptr = LLVMBuildGEP(builder, io_ptr, &ind3, 1, "");
+
+   clip_ptr0 = draw_jit_header_clip(builder, io0_ptr);
+   clip_ptr1 = draw_jit_header_clip(builder, io1_ptr);
+   clip_ptr2 = draw_jit_header_clip(builder, io2_ptr);
+   clip_ptr3 = draw_jit_header_clip(builder, io3_ptr);
+
+   for (i = 0; i<4; i++){
+      clip0_ptr = LLVMBuildGEP(builder, clip_ptr0,
+                               indices, 2, ""); //x0
+      clip1_ptr = LLVMBuildGEP(builder, clip_ptr1,
+                               indices, 2, ""); //x1
+      clip2_ptr = LLVMBuildGEP(builder, clip_ptr2,
+                               indices, 2, ""); //x2
+      clip3_ptr = LLVMBuildGEP(builder, clip_ptr3,
+                               indices, 2, ""); //x3
+
+      out0elem = LLVMBuildExtractElement(builder, out[i],
+                                         ind0, ""); //x0
+      out1elem = LLVMBuildExtractElement(builder, out[i],
+                                         ind1, ""); //x1
+      out2elem = LLVMBuildExtractElement(builder, out[i],
+                                         ind2, ""); //x2
+      out3elem = LLVMBuildExtractElement(builder, out[i],
+                                         ind3, ""); //x3
+  
+      LLVMBuildStore(builder, out0elem, clip0_ptr);
+      LLVMBuildStore(builder, out1elem, clip1_ptr);
+      LLVMBuildStore(builder, out2elem, clip2_ptr);
+      LLVMBuildStore(builder, out3elem, clip3_ptr);
+
+      indices[1]= LLVMBuildAdd(builder, indices[1], ind1, "");
+   }
+
+}
+
+/* Equivalent of _mm_set1_ps(a)
+ */
+static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
+				      LLVMValueRef a,
+				      const char *name)
+{
+   LLVMValueRef res = LLVMGetUndef(LLVMVectorType(LLVMFloatType(), 4));
+   int i;
+
+   for(i = 0; i < 4; ++i) {
+      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      res = LLVMBuildInsertElement(bld, res, a, index, i == 3 ? name : "");
+   }
+
+   return res;
+}
+
+/*
+ * Transforms the outputs for viewport mapping
+ */
+static void
+generate_viewport(struct draw_llvm *llvm,
+                  LLVMBuilderRef builder,
+                  LLVMValueRef (*outputs)[NUM_CHANNELS],
+                  LLVMValueRef context_ptr)
+{
+   int i;
+   struct lp_type f32_type = lp_type_float_vec(32);
+   LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/   
+   LLVMValueRef const1 = lp_build_const_vec(f32_type, 1.0);       /*1.0 1.0 1.0 1.0*/ 
+   LLVMValueRef vp_ptr = draw_jit_context_viewport(builder, context_ptr);
+
+   /* for 1/w convention*/
+   out3 = LLVMBuildFDiv(builder, const1, out3, "");
+   LLVMBuildStore(builder, out3, outputs[0][3]);
+  
+   /* Viewport Mapping */
+   for (i=0; i<3; i++){
+      LLVMValueRef out = LLVMBuildLoad(builder, outputs[0][i], ""); /*x0 x1 x2 x3*/
+      LLVMValueRef scale;
+      LLVMValueRef trans;
+      LLVMValueRef scale_i;
+      LLVMValueRef trans_i;
+      LLVMValueRef index;
+      
+      index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      scale_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
+
+      index = LLVMConstInt(LLVMInt32Type(), i+4, 0);
+      trans_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
+
+      scale = vec4f_from_scalar(builder, LLVMBuildLoad(builder, scale_i, ""), "scale");
+      trans = vec4f_from_scalar(builder, LLVMBuildLoad(builder, trans_i, ""), "trans");
+
+      /* divide by w */
+      out = LLVMBuildFMul(builder, out, out3, "");
+      /* mult by scale */
+      out = LLVMBuildFMul(builder, out, scale, "");
+      /* add translation */
+      out = LLVMBuildFAdd(builder, out, trans, "");
+
+      /* store transformed outputs */
+      LLVMBuildStore(builder, out, outputs[0][i]);
+   }
+   
+}
+
+
+/*
+ * Returns clipmask as 4xi32 bitmask for the 4 vertices
+ */
+static LLVMValueRef 
+generate_clipmask(LLVMBuilderRef builder,
+                  LLVMValueRef (*outputs)[NUM_CHANNELS],
+                  boolean clip_xy,
+                  boolean clip_z,
+                  boolean clip_user,
+                  boolean clip_halfz,
+                  unsigned nr,
+                  LLVMValueRef context_ptr)
+{
+   LLVMValueRef mask; /* stores the <4xi32> clipmasks */     
+   LLVMValueRef test, temp; 
+   LLVMValueRef zero, shift;
+   LLVMValueRef pos_x, pos_y, pos_z, pos_w;
+   LLVMValueRef plane1, planes, plane_ptr, sum;
+
+   unsigned i;
+
+   struct lp_type f32_type = lp_type_float_vec(32); 
+
+   mask = lp_build_const_int_vec(lp_type_int_vec(32), 0);
+   temp = lp_build_const_int_vec(lp_type_int_vec(32), 0);
+   zero = lp_build_const_vec(f32_type, 0);                    /* 0.0f 0.0f 0.0f 0.0f */
+   shift = lp_build_const_int_vec(lp_type_int_vec(32), 1);    /* 1 1 1 1 */
+
+   /* Assuming position stored at output[0] */
+   pos_x = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/
+   pos_y = LLVMBuildLoad(builder, outputs[0][1], ""); /*y0 y1 y2 y3*/
+   pos_z = LLVMBuildLoad(builder, outputs[0][2], ""); /*z0 z1 z2 z3*/
+   pos_w = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/   
+
+   /* Cliptest, for hardwired planes */
+   if (clip_xy){
+      /* plane 1 */
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
+      temp = shift;
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = test;
+   
+      /* plane 2 */
+      test = LLVMBuildFAdd(builder, pos_x, pos_w, "");
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+      temp = LLVMBuildShl(builder, temp, shift, "");
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = LLVMBuildOr(builder, mask, test, "");
+   
+      /* plane 3 */
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w);
+      temp = LLVMBuildShl(builder, temp, shift, "");
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = LLVMBuildOr(builder, mask, test, "");
+
+      /* plane 4 */
+      test = LLVMBuildFAdd(builder, pos_y, pos_w, "");
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+      temp = LLVMBuildShl(builder, temp, shift, "");
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = LLVMBuildOr(builder, mask, test, "");
+   }
+
+   if (clip_z){
+      temp = lp_build_const_int_vec(lp_type_int_vec(32), 16);
+      if (clip_halfz){
+         /* plane 5 */
+         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, pos_z);
+         test = LLVMBuildAnd(builder, test, temp, ""); 
+         mask = LLVMBuildOr(builder, mask, test, "");
+      }  
+      else{
+         /* plane 5 */
+         test = LLVMBuildFAdd(builder, pos_z, pos_w, "");
+         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+         test = LLVMBuildAnd(builder, test, temp, ""); 
+         mask = LLVMBuildOr(builder, mask, test, "");
+      }
+      /* plane 6 */
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_z, pos_w);
+      temp = LLVMBuildShl(builder, temp, shift, "");
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = LLVMBuildOr(builder, mask, test, "");
+   }   
+
+   if (clip_user){
+      LLVMValueRef planes_ptr = draw_jit_context_planes(builder, context_ptr);
+      LLVMValueRef indices[3];
+      temp = lp_build_const_int_vec(lp_type_int_vec(32), 32);
+
+      /* userclip planes */
+      for (i = 6; i < nr; i++) {
+         indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+         indices[1] = LLVMConstInt(LLVMInt32Type(), i, 0);
+
+         indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+         plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+         plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_x");
+         planes = vec4f_from_scalar(builder, plane1, "plane4_x");
+         sum = LLVMBuildFMul(builder, planes, pos_x, "");
+
+         indices[2] = LLVMConstInt(LLVMInt32Type(), 1, 0);
+         plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+         plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_y"); 
+         planes = vec4f_from_scalar(builder, plane1, "plane4_y");
+         test = LLVMBuildFMul(builder, planes, pos_y, "");
+         sum = LLVMBuildFAdd(builder, sum, test, "");
+         
+         indices[2] = LLVMConstInt(LLVMInt32Type(), 2, 0);
+         plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+         plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_z"); 
+         planes = vec4f_from_scalar(builder, plane1, "plane4_z");
+         test = LLVMBuildFMul(builder, planes, pos_z, "");
+         sum = LLVMBuildFAdd(builder, sum, test, "");
+
+         indices[2] = LLVMConstInt(LLVMInt32Type(), 3, 0);
+         plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+         plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_w"); 
+         planes = vec4f_from_scalar(builder, plane1, "plane4_w");
+         test = LLVMBuildFMul(builder, planes, pos_w, "");
+         sum = LLVMBuildFAdd(builder, sum, test, "");
+
+         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, sum);
+         temp = LLVMBuildShl(builder, temp, shift, "");
+         test = LLVMBuildAnd(builder, test, temp, ""); 
+         mask = LLVMBuildOr(builder, mask, test, "");
+      }
+   }
+   return mask;
+}
+
+/*
+ * Returns boolean if any clipping has occurred
+ * Used zero/non-zero i32 value to represent boolean 
+ */
+static void
+clipmask_bool(LLVMBuilderRef builder, 
+              LLVMValueRef clipmask,
+              LLVMValueRef ret_ptr)
+{
+   LLVMValueRef ret = LLVMBuildLoad(builder, ret_ptr, "");   
+   LLVMValueRef temp;
+   int i;
+
+   for (i=0; i<4; i++){   
+      temp = LLVMBuildExtractElement(builder, clipmask,
+                                     LLVMConstInt(LLVMInt32Type(), i, 0) , "");
+      ret = LLVMBuildOr(builder, ret, temp, "");
+   }
+   
+   LLVMBuildStore(builder, ret, ret_ptr);
+}
+
 static void
 draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
 {
@@ -706,6 +1019,11 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
    LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS];
    void *code;
    struct lp_build_sampler_soa *sampler = 0;
+   LLVMValueRef ret, ret_ptr;
+   boolean bypass_viewport = variant->key.bypass_viewport;
+   boolean enable_cliptest = variant->key.clip_xy || 
+                             variant->key.clip_z  ||
+                             variant->key.clip_user;
    
    arg_types[0] = llvm->context_ptr_type;           /* context */
    arg_types[1] = llvm->vertex_header_ptr_type;     /* vertex_header */
@@ -716,7 +1034,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
    arg_types[6] = llvm->vb_ptr_type;                /* pipe_vertex_buffer's */
    arg_types[7] = LLVMInt32Type();                  /* instance_id */
 
-   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+   func_type = LLVMFunctionType(LLVMInt32Type(), arg_types, Elements(arg_types), 0);
 
    variant->function = LLVMAddFunction(llvm->module, "draw_llvm_shader", func_type);
    LLVMSetFunctionCallConv(variant->function, LLVMCCallConv);
@@ -756,6 +1074,10 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
 
    step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0);
 
+   /* function will return non-zero i32 value if any clipped vertices */     
+   ret_ptr = lp_build_alloca(builder, LLVMInt32Type(), "");   
+   LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), ret_ptr);
+
    /* code generated texture sampling */
    sampler = draw_llvm_sampler_soa_create(
       draw_llvm_variant_key_samplers(&variant->key),
@@ -770,6 +1092,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
       LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
       LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } };
       LLVMValueRef io;
+      LLVMValueRef clipmask;   /* holds the clipmask value */
       const LLVMValueRef (*ptr_aos)[NUM_CHANNELS];
 
       io_itr = LLVMBuildSub(builder, lp_loop.counter, start, "");
@@ -806,10 +1129,37 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
                   context_ptr,
                   sampler);
 
-      convert_to_aos(builder, io, outputs,
+      /* store original positions in clip before further manipulation */
+      store_clip(builder, io, outputs);
+
+      /* do cliptest */
+      if (enable_cliptest){
+         /* allocate clipmask, assign it integer type */
+         clipmask = generate_clipmask(builder, outputs,
+                                      variant->key.clip_xy,
+                                      variant->key.clip_z, 
+                                      variant->key.clip_user,
+                                      variant->key.clip_halfz,
+                                      variant->key.nr_planes,
+                                      context_ptr);
+         /* return clipping boolean value for function */
+         clipmask_bool(builder, clipmask, ret_ptr);
+      }
+      else{
+         clipmask = lp_build_const_int_vec(lp_type_int_vec(32), 0);    
+      }
+      
+      /* do viewport mapping */
+      if (!bypass_viewport){
+         generate_viewport(llvm, builder, outputs, context_ptr);
+      }
+
+      /* store clipmask in vertex header and positions in data */
+      convert_to_aos(builder, io, outputs, clipmask,
                      draw->vs.vertex_shader->info.num_outputs,
                      max_vertices);
    }
+
    lp_build_loop_end_cond(builder, end, step, LLVMIntUGE, &lp_loop);
 
    sampler->destroy(sampler);
@@ -819,7 +1169,8 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
    lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
 #endif
 
-   LLVMBuildRetVoid(builder);
+   ret = LLVMBuildLoad(builder, ret_ptr,"");
+   LLVMBuildRet(builder, ret);
       
    LLVMDisposeBuilder(builder);
 
@@ -870,6 +1221,11 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    LLVMValueRef fetch_max;
    void *code;
    struct lp_build_sampler_soa *sampler = 0;
+   LLVMValueRef ret, ret_ptr;
+   boolean bypass_viewport = variant->key.bypass_viewport;
+   boolean enable_cliptest = variant->key.clip_xy || 
+                             variant->key.clip_z  ||
+                             variant->key.clip_user;
    
    arg_types[0] = llvm->context_ptr_type;               /* context */
    arg_types[1] = llvm->vertex_header_ptr_type;         /* vertex_header */
@@ -880,10 +1236,9 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    arg_types[6] = llvm->vb_ptr_type;                    /* pipe_vertex_buffer's */
    arg_types[7] = LLVMInt32Type();                      /* instance_id */
 
-   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+   func_type = LLVMFunctionType(LLVMInt32Type(), arg_types, Elements(arg_types), 0);
 
-   variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts",
+   variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts", func_type);
-                                            func_type);
    LLVMSetFunctionCallConv(variant->function_elts, LLVMCCallConv);
    for(i = 0; i < Elements(arg_types); ++i)
       if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
@@ -929,11 +1284,16 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
                             LLVMConstInt(LLVMInt32Type(), 1, 0),
                             "fetch_max");
 
+   /* function returns non-zero i32 value if any clipped vertices */
+   ret_ptr = lp_build_alloca(builder, LLVMInt32Type(), ""); 
+   LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), ret_ptr);
+
    lp_build_loop_begin(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), &lp_loop);
    {
       LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
       LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } };
       LLVMValueRef io;
+      LLVMValueRef clipmask;   /* holds the clipmask value */
       const LLVMValueRef (*ptr_aos)[NUM_CHANNELS];
 
       io_itr = lp_loop.counter;
@@ -980,10 +1340,40 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
                   context_ptr,
                   sampler);
 
-      convert_to_aos(builder, io, outputs,
+      /* store original positions in clip before further manipulation */
+      store_clip(builder, io, outputs);
+
+      /* do cliptest */
+      if (enable_cliptest){
+         /* allocate clipmask, assign it integer type */
+         clipmask = generate_clipmask(builder, outputs,
+                                      variant->key.clip_xy,
+                                      variant->key.clip_z, 
+                                      variant->key.clip_user,
+                                      variant->key.clip_halfz,
+                                      variant->key.nr_planes,
+                                      context_ptr);
+         /* return clipping boolean value for function */
+         clipmask_bool(builder, clipmask, ret_ptr);
+      }
+      else{
+         clipmask = lp_build_const_int_vec(lp_type_int_vec(32), 0);
+      }
+      
+      /* do viewport mapping */
+      if (!bypass_viewport){
+         generate_viewport(llvm, builder, outputs, context_ptr);
+      }
+
+      /* store clipmask in vertex header, 
+       * original positions in clip 
+       * and transformed positions in data 
+       */   
+      convert_to_aos(builder, io, outputs, clipmask,
                      draw->vs.vertex_shader->info.num_outputs,
                      max_vertices);
    }
+
    lp_build_loop_end_cond(builder, fetch_count, step, LLVMIntUGE, &lp_loop);
 
    sampler->destroy(sampler);
@@ -993,7 +1383,8 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
 #endif
 
-   LLVMBuildRetVoid(builder);
+   ret = LLVMBuildLoad(builder, ret_ptr,"");   
+   LLVMBuildRet(builder, ret);
    
    LLVMDisposeBuilder(builder);
 
@@ -1038,6 +1429,16 @@ draw_llvm_make_variant_key(struct draw_llvm *llvm, char *store)
     */
    key->nr_vertex_elements = llvm->draw->pt.nr_vertex_elements;
 
+   /* will have to rig this up properly later */
+   key->clip_xy = llvm->draw->clip_xy;
+   key->clip_z = llvm->draw->clip_z;
+   key->clip_user = llvm->draw->clip_user;
+   key->bypass_viewport = llvm->draw->identity_viewport;
+   key->clip_halfz = !llvm->draw->rasterizer->gl_rasterization_rules;
+   key->need_edgeflags = (llvm->draw->vs.edgeflag_output ? TRUE : FALSE);
+   key->nr_planes = llvm->draw->nr_planes;
+   key->pad = 0;
+
    /* All variants of this shader will have the same value for
     * nr_samplers.  Not yet trying to compact away holes in the
     * sampler array.
@@ -1066,9 +1467,9 @@ draw_llvm_set_mapped_texture(struct draw_context *draw,
                              unsigned sampler_idx,
                              uint32_t width, uint32_t height, uint32_t depth,
                              uint32_t last_level,
-                             uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
+                             uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
-                             uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
+                             uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
-                             const void *data[DRAW_MAX_TEXTURE_LEVELS])
+                             const void *data[PIPE_MAX_TEXTURE_LEVELS])
 {
    unsigned j;
    struct draw_jit_texture *jit_tex;

src/gallium/auxiliary/draw/draw_llvm.h

@@ -41,7 +41,6 @@
 #include <llvm-c/Target.h>
 #include <llvm-c/ExecutionEngine.h>
 
-#define DRAW_MAX_TEXTURE_LEVELS 13  /* 4K x 4K for now */
 
 struct draw_llvm;
 struct llvm_vertex_shader;
@@ -52,9 +51,9 @@ struct draw_jit_texture
    uint32_t height;
    uint32_t depth;
    uint32_t last_level;
-   uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS];
+   uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS];
-   uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS];
+   uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS];
-   const void *data[DRAW_MAX_TEXTURE_LEVELS];
+   const void *data[PIPE_MAX_TEXTURE_LEVELS];
    float min_lod;
    float max_lod;
    float lod_bias;
@@ -97,7 +96,8 @@ struct draw_jit_context
 {
    const float *vs_constants;
    const float *gs_constants;
-
+   float (*planes) [12][4];
+   float *viewport;
 
    struct draw_jit_texture textures[PIPE_MAX_VERTEX_SAMPLERS];
 };
@@ -109,18 +109,22 @@ struct draw_jit_context
 #define draw_jit_context_gs_constants(_builder, _ptr) \
    lp_build_struct_get(_builder, _ptr, 1, "gs_constants")
 
-#define DRAW_JIT_CTX_TEXTURES 2
+#define draw_jit_context_planes(_builder, _ptr) \
+   lp_build_struct_get(_builder, _ptr, 2, "planes")
+
+#define draw_jit_context_viewport(_builder, _ptr) \
+   lp_build_struct_get(_builder, _ptr, 3, "viewport")
+
+#define DRAW_JIT_CTX_TEXTURES 4
 
 #define draw_jit_context_textures(_builder, _ptr) \
    lp_build_struct_get_ptr(_builder, _ptr, DRAW_JIT_CTX_TEXTURES, "textures")
 
-
-
 #define draw_jit_header_id(_builder, _ptr)              \
    lp_build_struct_get_ptr(_builder, _ptr, 0, "id")
 
 #define draw_jit_header_clip(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, 1, "clip")
+   lp_build_struct_get_ptr(_builder, _ptr, 1, "clip")
 
 #define draw_jit_header_data(_builder, _ptr)            \
    lp_build_struct_get_ptr(_builder, _ptr, 2, "data")
@@ -136,7 +140,7 @@ struct draw_jit_context
    lp_build_struct_get(_builder, _ptr, 2, "buffer_offset")
 
 
-typedef void
+typedef int
 (*draw_jit_vert_func)(struct draw_jit_context *context,
                       struct vertex_header *io,
                       const char *vbuffers[PIPE_MAX_ATTRIBS],
@@ -147,7 +151,7 @@ typedef void
                       unsigned instance_id);
 
 
-typedef void
+typedef int
 (*draw_jit_vert_func_elts)(struct draw_jit_context *context,
                            struct vertex_header *io,
                            const char *vbuffers[PIPE_MAX_ATTRIBS],
@@ -159,8 +163,16 @@ typedef void
 
 struct draw_llvm_variant_key
 {
-   unsigned nr_vertex_elements:16;
+   unsigned nr_vertex_elements:8;
-   unsigned nr_samplers:16;
+   unsigned nr_samplers:8;
+   unsigned clip_xy:1;
+   unsigned clip_z:1;
+   unsigned clip_user:1;
+   unsigned clip_halfz:1;
+   unsigned bypass_viewport:1;
+   unsigned need_edgeflags:1;
+   unsigned nr_planes:4;
+   unsigned pad:6;
 
    /* Variable number of vertex elements:
     */
@@ -290,8 +302,8 @@ draw_llvm_set_mapped_texture(struct draw_context *draw,
                              unsigned sampler_idx,
                              uint32_t width, uint32_t height, uint32_t depth,
                              uint32_t last_level,
-                             uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
+                             uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
-                             uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
+                             uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
-                             const void *data[DRAW_MAX_TEXTURE_LEVELS]);
+                             const void *data[PIPE_MAX_TEXTURE_LEVELS]);
 
 #endif

src/gallium/auxiliary/draw/draw_private.h

@@ -169,6 +169,9 @@ struct draw_context
          unsigned vs_constants_size[PIPE_MAX_CONSTANT_BUFFERS];
          const void *gs_constants[PIPE_MAX_CONSTANT_BUFFERS];
          unsigned gs_constants_size[PIPE_MAX_CONSTANT_BUFFERS];
+         
+         /* pointer to planes */
+         float (*planes)[12][4]; 
       } user;
 
       boolean test_fse;         /* enable FSE even though its not correct (eg for softpipe) */

src/gallium/auxiliary/draw/draw_pt.c

@@ -287,6 +287,84 @@ draw_print_arrays(struct draw_context *draw, uint prim, int start, uint count)
 }
 
 
+/** Helper code for below */
+#define PRIM_RESTART_LOOP(elements) \
+   do { \
+      for (i = start; i < end; i++) { \
+         if (elements[i] == info->restart_index) { \
+            if (cur_count > 0) { \
+               /* draw elts up to prev pos */ \
+               draw_pt_arrays(draw, prim, cur_start, cur_count); \
+            } \
+            /* begin new prim at next elt */ \
+            cur_start = i + 1; \
+            cur_count = 0; \
+         } \
+         else { \
+            cur_count++; \
+         } \
+      } \
+      if (cur_count > 0) { \
+         draw_pt_arrays(draw, prim, cur_start, cur_count); \
+      } \
+   } while (0)
+
+
+/**
+ * For drawing prims with primitive restart enabled.
+ * Scan for restart indexes and draw the runs of elements/vertices between
+ * the restarts.
+ */
+static void
+draw_pt_arrays_restart(struct draw_context *draw,
+                       const struct pipe_draw_info *info)
+{
+   const unsigned prim = info->mode;
+   const unsigned start = info->start;
+   const unsigned count = info->count;
+   const unsigned end = start + count;
+   unsigned i, cur_start, cur_count;
+
+   assert(info->primitive_restart);
+
+   if (draw->pt.user.elts) {
+      /* indexed prims (draw_elements) */
+      cur_start = start;
+      cur_count = 0;
+
+      switch (draw->pt.user.eltSize) {
+      case 1:
+         {
+            const ubyte *elt_ub = (const ubyte *) draw->pt.user.elts;
+            PRIM_RESTART_LOOP(elt_ub);
+         }
+         break;
+      case 2:
+         {
+            const ushort *elt_us = (const ushort *) draw->pt.user.elts;
+            PRIM_RESTART_LOOP(elt_us);
+         }
+         break;
+      case 4:
+         {
+            const uint *elt_ui = (const uint *) draw->pt.user.elts;
+            PRIM_RESTART_LOOP(elt_ui);
+         }
+         break;
+      default:
+         assert(0 && "bad eltSize in draw_arrays()");
+      }
+   }
+   else {
+      /* Non-indexed prims (draw_arrays).
+       * Primitive restart should have been handled in the state tracker.
+       */
+      draw_pt_arrays(draw, prim, start, count);
+   }
+}
+
+
+
 /**
  * Non-instanced drawing.
  * \sa draw_arrays_instanced
@@ -395,6 +473,12 @@ draw_vbo(struct draw_context *draw,
 
    for (instance = 0; instance < info->instance_count; instance++) {
       draw->instance_id = instance + info->start_instance;
+
+      if (info->primitive_restart) {
+         draw_pt_arrays_restart(draw, info);
+      }
+      else {
          draw_pt_arrays(draw, info->mode, info->start, info->count);
       }
    }
+}

src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c

@@ -175,6 +175,11 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
       draw->pt.user.vs_constants[0];
    fpme->llvm->jit_context.gs_constants =
       draw->pt.user.gs_constants[0];
+   fpme->llvm->jit_context.planes =
+      (float (*) [12][4]) draw->pt.user.planes[0];
+   fpme->llvm->jit_context.viewport =
+      (float *)draw->viewport.scale;
+    
 }
 
 
@@ -217,6 +222,7 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
    struct draw_vertex_info gs_vert_info;
    struct draw_vertex_info *vert_info;
    unsigned opt = fpme->opt;
+   unsigned clipped = 0;
 
    llvm_vert_info.count = fetch_info->count;
    llvm_vert_info.vertex_size = fpme->vertex_size;
@@ -230,7 +236,7 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
    }
 
    if (fetch_info->linear)
-      fpme->current_variant->jit_func( &fpme->llvm->jit_context,
+      clipped = fpme->current_variant->jit_func( &fpme->llvm->jit_context,
                                        llvm_vert_info.verts,
                                        (const char **)draw->pt.user.vbuffer,
                                        fetch_info->start,
@@ -239,7 +245,7 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
                                        draw->pt.vertex_buffer,
                                        draw->instance_id);
    else
-      fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
+      clipped = fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
                                             llvm_vert_info.verts,
                                             (const char **)draw->pt.user.vbuffer,
                                             fetch_info->elts,
@@ -266,6 +272,9 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
       FREE(vert_info->verts);
       vert_info = &gs_vert_info;
       prim_info = &gs_prim_info;
+
+      clipped = draw_pt_post_vs_run( fpme->post_vs, vert_info );
+
    }
 
    /* stream output needs to be done before clipping */
@@ -273,11 +282,11 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
 		    vert_info,
                     prim_info );
 
-   if (draw_pt_post_vs_run( fpme->post_vs, vert_info )) {
+   if (clipped) {
       opt |= PT_PIPELINE;
    }
 
-   /* Do we need to run the pipeline?
+   /* Do we need to run the pipeline? Now will come here if clipped
     */
    if (opt & PT_PIPELINE) {
       pipeline( fpme,

src/gallium/auxiliary/gallivm/lp_bld_arit.c

@@ -983,6 +983,12 @@ enum lp_build_round_sse41_mode
 };
 
 
+/**
+ * Helper for SSE4.1's ROUNDxx instructions.
+ *
+ * NOTE: In the SSE4.1's nearest mode, if two values are equally close, the
+ * result is the even value.  That is, rounding 2.5 will be 2.0, and not 3.0.
+ */
 static INLINE LLVMValueRef
 lp_build_round_sse41(struct lp_build_context *bld,
                      LLVMValueRef a,
@@ -1053,10 +1059,58 @@ lp_build_round_sse41(struct lp_build_context *bld,
 }
 
 
+static INLINE LLVMValueRef
+lp_build_iround_nearest_sse2(struct lp_build_context *bld,
+                             LLVMValueRef a)
+{
+   const struct lp_type type = bld->type;
+   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMTypeRef ret_type = lp_build_int_vec_type(type);
+   const char *intrinsic;
+   LLVMValueRef res;
+
+   assert(type.floating);
+   /* using the double precision conversions is a bit more complicated */
+   assert(type.width == 32);
+
+   assert(lp_check_value(type, a));
+   assert(util_cpu_caps.has_sse2);
+
+   /* This is relying on MXCSR rounding mode, which should always be nearest. */
+   if (type.length == 1) {
+      LLVMTypeRef vec_type;
+      LLVMValueRef undef;
+      LLVMValueRef arg;
+      LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
+
+      vec_type = LLVMVectorType(bld->elem_type, 4);
+
+      intrinsic = "llvm.x86.sse.cvtss2si";
+
+      undef = LLVMGetUndef(vec_type);
+
+      arg = LLVMBuildInsertElement(bld->builder, undef, a, index0, "");
+
+      res = lp_build_intrinsic_unary(bld->builder, intrinsic,
+                                     ret_type, arg);
+   }
+   else {
+      assert(type.width*type.length == 128);
+
+      intrinsic = "llvm.x86.sse2.cvtps2dq";
+
+      res = lp_build_intrinsic_unary(bld->builder, intrinsic,
+                                     ret_type, a);
+   }
+
+   return res;
+}
+
+
 /**
- * Return the integer part of a float (vector) value.  The returned value is
+ * Return the integer part of a float (vector) value (== round toward zero).
- * a float (vector).
+ * The returned value is a float (vector).
- * Ex: trunc(-1.5) = 1.0
+ * Ex: trunc(-1.5) = -1.0
  */
 LLVMValueRef
 lp_build_trunc(struct lp_build_context *bld,
@@ -1181,9 +1235,9 @@ lp_build_fract(struct lp_build_context *bld,
 
 
 /**
- * Return the integer part of a float (vector) value.  The returned value is
+ * Return the integer part of a float (vector) value (== round toward zero).
- * an integer (vector).
+ * The returned value is an integer (vector).
- * Ex: itrunc(-1.5) = 1
+ * Ex: itrunc(-1.5) = -1
  */
 LLVMValueRef
 lp_build_itrunc(struct lp_build_context *bld,
@@ -1217,7 +1271,11 @@ lp_build_iround(struct lp_build_context *bld,
 
    assert(lp_check_value(type, a));
 
-   if (util_cpu_caps.has_sse4_1 &&
+   if (util_cpu_caps.has_sse2 &&
+       ((type.width == 32) && (type.length == 1 || type.length == 4))) {
+      return lp_build_iround_nearest_sse2(bld, a);
+   }
+   else if (util_cpu_caps.has_sse4_1 &&
        (type.length == 1 || type.width*type.length == 128)) {
       res = lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
    }
@@ -1371,8 +1429,6 @@ lp_build_ifloor_fract(struct lp_build_context *bld,
                       LLVMValueRef *out_ipart,
                       LLVMValueRef *out_fpart)
 {
-
-
    const struct lp_type type = bld->type;
    LLVMValueRef ipart;
 
@@ -2209,6 +2265,71 @@ lp_build_exp2(struct lp_build_context *bld,
 }
 
 
+/**
+ * Extract the exponent of a IEEE-754 floating point value.
+ *
+ * Optionally apply an integer bias.
+ *
+ * Result is an integer value with
+ *
+ *   ifloor(log2(x)) + bias
+ */
+LLVMValueRef
+lp_build_extract_exponent(struct lp_build_context *bld,
+                          LLVMValueRef x,
+                          int bias)
+{
+   const struct lp_type type = bld->type;
+   unsigned mantissa = lp_mantissa(type);
+   LLVMValueRef res;
+
+   assert(type.floating);
+
+   assert(lp_check_value(bld->type, x));
+
+   x = LLVMBuildBitCast(bld->builder, x, bld->int_vec_type, "");
+
+   res = LLVMBuildLShr(bld->builder, x, lp_build_const_int_vec(type, mantissa), "");
+   res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(type, 255), "");
+   res = LLVMBuildSub(bld->builder, res, lp_build_const_int_vec(type, 127 - bias), "");
+
+   return res;
+}
+
+
+/**
+ * Extract the mantissa of the a floating.
+ *
+ * Result is a floating point value with
+ *
+ *   x / floor(log2(x))
+ */
+LLVMValueRef
+lp_build_extract_mantissa(struct lp_build_context *bld,
+                          LLVMValueRef x)
+{
+   const struct lp_type type = bld->type;
+   unsigned mantissa = lp_mantissa(type);
+   LLVMValueRef mantmask = lp_build_const_int_vec(type, (1ULL << mantissa) - 1);
+   LLVMValueRef one = LLVMConstBitCast(bld->one, bld->int_vec_type);
+   LLVMValueRef res;
+
+   assert(lp_check_value(bld->type, x));
+
+   assert(type.floating);
+
+   x = LLVMBuildBitCast(bld->builder, x, bld->int_vec_type, "");
+
+   /* res = x / 2**ipart */
+   res = LLVMBuildAnd(bld->builder, x, mantmask, "");
+   res = LLVMBuildOr(bld->builder, res, one, "");
+   res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+
+   return res;
+}
+
+
+
 /**
  * Minimax polynomial fit of log2(x)/(x - 1), for x in range [1, 2[
  * These coefficients can be generate with
@@ -2333,7 +2454,10 @@ lp_build_log2(struct lp_build_context *bld,
 /**
  * Faster (and less accurate) log2.
  *
- *    log2(x) = floor(log2(x)) + frac(x)
+ *    log2(x) = floor(log2(x)) - 1 + x / 2**floor(log2(x))
+ *
+ * Piece-wise linear approximation, with exact results when x is a
+ * power of two.
  *
  * See http://www.flipcode.com/archives/Fast_log_Function.shtml
  */
@@ -2341,35 +2465,21 @@ LLVMValueRef
 lp_build_fast_log2(struct lp_build_context *bld,
                    LLVMValueRef x)
 {
-   const struct lp_type type = bld->type;
-   LLVMTypeRef vec_type = bld->vec_type;
-   LLVMTypeRef int_vec_type = bld->int_vec_type;
-
-   unsigned mantissa = lp_mantissa(type);
-   LLVMValueRef mantmask = lp_build_const_int_vec(type, (1ULL << mantissa) - 1);
-   LLVMValueRef one = LLVMConstBitCast(bld->one, int_vec_type);
-
    LLVMValueRef ipart;
    LLVMValueRef fpart;
 
    assert(lp_check_value(bld->type, x));
 
-   assert(type.floating);
+   assert(bld->type.floating);
-
-   x = LLVMBuildBitCast(bld->builder, x, int_vec_type, "");
 
    /* ipart = floor(log2(x)) - 1 */
-   ipart = LLVMBuildLShr(bld->builder, x, lp_build_const_int_vec(type, mantissa), "");
+   ipart = lp_build_extract_exponent(bld, x, -1);
-   ipart = LLVMBuildAnd(bld->builder, ipart, lp_build_const_int_vec(type, 255), "");
+   ipart = LLVMBuildSIToFP(bld->builder, ipart, bld->vec_type, "");
-   ipart = LLVMBuildSub(bld->builder, ipart, lp_build_const_int_vec(type, 128), "");
-   ipart = LLVMBuildSIToFP(bld->builder, ipart, vec_type, "");
 
-   /* fpart = 1.0 + frac(x) */
+   /* fpart = x / 2**ipart */
-   fpart = LLVMBuildAnd(bld->builder, x, mantmask, "");
+   fpart = lp_build_extract_mantissa(bld, x);
-   fpart = LLVMBuildOr(bld->builder, fpart, one, "");
-   fpart = LLVMBuildBitCast(bld->builder, fpart, vec_type, "");
 
-   /* floor(log2(x)) + frac(x) */
+   /* ipart + fpart */
    return LLVMBuildFAdd(bld->builder, ipart, fpart, "");
 }
 
@@ -2383,27 +2493,18 @@ LLVMValueRef
 lp_build_ilog2(struct lp_build_context *bld,
                LLVMValueRef x)
 {
-   const struct lp_type type = bld->type;
+   LLVMValueRef sqrt2 = lp_build_const_vec(bld->type, M_SQRT2);
-   LLVMTypeRef int_vec_type = bld->int_vec_type;
-
-   unsigned mantissa = lp_mantissa(type);
-   LLVMValueRef sqrt2 = lp_build_const_vec(type, 1.4142135623730951);
-
    LLVMValueRef ipart;
 
-   assert(lp_check_value(bld->type, x));
+   assert(bld->type.floating);
 
-   assert(type.floating);
+   assert(lp_check_value(bld->type, x));
 
    /* x * 2^(0.5)   i.e., add 0.5 to the log2(x) */
    x = LLVMBuildFMul(bld->builder, x, sqrt2, "");
 
-   x = LLVMBuildBitCast(bld->builder, x, int_vec_type, "");
-
    /* ipart = floor(log2(x) + 0.5)  */
-   ipart = LLVMBuildLShr(bld->builder, x, lp_build_const_int_vec(type, mantissa), "");
+   ipart = lp_build_extract_exponent(bld, x, 0);
-   ipart = LLVMBuildAnd(bld->builder, ipart, lp_build_const_int_vec(type, 255), "");
-   ipart = LLVMBuildSub(bld->builder, ipart, lp_build_const_int_vec(type, 127), "");
 
    return ipart;
 }

src/gallium/auxiliary/gallivm/lp_bld_arit.h

@@ -214,6 +214,15 @@ LLVMValueRef
 lp_build_exp2(struct lp_build_context *bld,
               LLVMValueRef a);
 
+LLVMValueRef
+lp_build_extract_exponent(struct lp_build_context *bld,
+                          LLVMValueRef x,
+                          int bias);
+
+LLVMValueRef
+lp_build_extract_mantissa(struct lp_build_context *bld,
+                          LLVMValueRef x);
+
 LLVMValueRef
 lp_build_log2(struct lp_build_context *bld,
               LLVMValueRef a);
@@ -226,7 +235,6 @@ LLVMValueRef
 lp_build_ilog2(struct lp_build_context *bld,
                LLVMValueRef x);
 
-
 void
 lp_build_exp2_approx(struct lp_build_context *bld,
                      LLVMValueRef x,

src/gallium/auxiliary/gallivm/lp_bld_conv.c

@@ -97,58 +97,104 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
    LLVMTypeRef int_vec_type = lp_build_int_vec_type(src_type);
    LLVMValueRef res;
    unsigned mantissa;
-   unsigned n;
+
+   assert(src_type.floating);
+   assert(dst_width <= src_type.width);
+   src_type.sign = FALSE;
+
+   mantissa = lp_mantissa(src_type);
+
+   if (dst_width <= mantissa) {
+      /*
+       * Apply magic coefficients that will make the desired result to appear
+       * in the lowest significant bits of the mantissa, with correct rounding.
+       *
+       * This only works if the destination width fits in the mantissa.
+       */
+
       unsigned long long ubound;
       unsigned long long mask;
       double scale;
       double bias;
 
-   assert(src_type.floating);
+      ubound = (1ULL << dst_width);
-
-   mantissa = lp_mantissa(src_type);
-
-   /* We cannot carry more bits than the mantissa */
-   n = MIN2(mantissa, dst_width);
-
-   /* This magic coefficients will make the desired result to appear in the
-    * lowest significant bits of the mantissa.
-    */
-   ubound = ((unsigned long long)1 << n);
       mask = ubound - 1;
       scale = (double)mask/ubound;
-   bias = (double)((unsigned long long)1 << (mantissa - n));
+      bias = (double)(1ULL << (mantissa - dst_width));
 
       res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
       res = LLVMBuildFAdd(builder, res, lp_build_const_vec(src_type, bias), "");
       res = LLVMBuildBitCast(builder, res, int_vec_type, "");
-
-   if(dst_width > n) {
-      int shift = dst_width - n;
-      res = LLVMBuildShl(builder, res, lp_build_const_int_vec(src_type, shift), "");
-
-      /* TODO: Fill in the empty lower bits for additional precision? */
-      /* YES: this fixes progs/trivial/tri-z-eq.c.
-       * Otherwise vertex Z=1.0 values get converted to something like
-       * 0xfffffb00 and the test for equality with 0xffffffff fails.
-       */
-#if 0
-      {
-         LLVMValueRef msb;
-         msb = LLVMBuildLShr(builder, res, lp_build_const_int_vec(src_type, dst_width - 1), "");
-         msb = LLVMBuildShl(builder, msb, lp_build_const_int_vec(src_type, shift), "");
-         msb = LLVMBuildSub(builder, msb, lp_build_const_int_vec(src_type, 1), "");
-         res = LLVMBuildOr(builder, res, msb, "");
-      }
-#elif 0
-      while(shift > 0) {
-         res = LLVMBuildOr(builder, res, LLVMBuildLShr(builder, res, lp_build_const_int_vec(src_type, n), ""), "");
-         shift -= n;
-         n *= 2;
-      }
-#endif
-   }
-   else
       res = LLVMBuildAnd(builder, res, lp_build_const_int_vec(src_type, mask), "");
+   }
+   else if (dst_width == (mantissa + 1)) {
+      /*
+       * The destination width matches exactly what can be represented in
+       * floating point (i.e., mantissa + 1 bits). So do a straight
+       * multiplication followed by casting. No further rounding is necessary.
+       */
+
+      double scale;
+
+      scale = (double)((1ULL << dst_width) - 1);
+
+      res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
+      res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
+   }
+   else {
+      /*
+       * The destination exceeds what can be represented in the floating point.
+       * So multiply by the largest power two we get away with, and when
+       * subtract the most significant bit to rescale to normalized values.
+       *
+       * The largest power of two factor we can get away is
+       * (1 << (src_type.width - 1)), because we need to use signed . In theory it
+       * should be (1 << (src_type.width - 2)), but IEEE 754 rules states
+       * INT_MIN should be returned in FPToSI, which is the correct result for
+       * values near 1.0!
+       *
+       * This means we get (src_type.width - 1) correct bits for values near 0.0,
+       * and (mantissa + 1) correct bits for values near 1.0. Equally or more
+       * important, we also get exact results for 0.0 and 1.0.
+       */
+
+      unsigned n = MIN2(src_type.width - 1, dst_width);
+
+      double scale = (double)(1ULL << n);
+      unsigned lshift = dst_width - n;
+      unsigned rshift = n;
+      LLVMValueRef lshifted;
+      LLVMValueRef rshifted;
+
+      res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
+      res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
+
+      /*
+       * Align the most significant bit to its final place.
+       *
+       * This will cause 1.0 to overflow to 0, but the later adjustment will
+       * get it right.
+       */
+      if (lshift) {
+         lshifted = LLVMBuildShl(builder, res,
+                                 lp_build_const_int_vec(src_type, lshift), "");
+      } else {
+         lshifted = res;
+      }
+
+      /*
+       * Align the most significant bit to the right.
+       */
+      rshifted =  LLVMBuildAShr(builder, res,
+                                lp_build_const_int_vec(src_type, rshift), "");
+
+      /*
+       * Subtract the MSB to the LSB, therefore re-scaling from
+       * (1 << dst_width) to ((1 << dst_width) - 1).
+       */
+
+      res = LLVMBuildSub(builder, lshifted, rshifted, "");
+   }
 
    return res;
 }
@@ -178,6 +224,16 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
 
    assert(dst_type.floating);
 
+   /* Special-case int8->float, though most cases could be handled
+    * this way:
+    */
+   if (src_width == 8) {
+      scale = 1.0/255.0;
+      res = LLVMBuildSIToFP(builder, src, vec_type, "");
+      res = LLVMBuildFMul(builder, res, lp_build_const_vec(dst_type, scale), "");
+      return res;
+   }
+
    mantissa = lp_mantissa(dst_type);
 
    n = MIN2(mantissa, src_width);
@@ -257,7 +313,9 @@ lp_build_conv(LLVMBuilderRef builder,
        dst_type.sign     == 0 &&
        dst_type.norm     == 1 &&
        dst_type.width    == 8 &&
-       dst_type.length   == 16)
+       dst_type.length   == 16 &&
+
+       util_cpu_caps.has_sse2)
    {
       int i;
 
@@ -296,23 +354,7 @@ lp_build_conv(LLVMBuilderRef builder,
          c = LLVMBuildFMul(builder, src[2], const_255f, "");
          d = LLVMBuildFMul(builder, src[3], const_255f, "");
 
-         /* lp_build_round generates excessively general code without
+         {
-          * sse4, so do rounding manually.
-          */
-         if (!util_cpu_caps.has_sse4_1) {
-            LLVMValueRef const_half = lp_build_const_vec(src_type, 0.5f);
-
-            a = LLVMBuildFAdd(builder, a, const_half, "");
-            b = LLVMBuildFAdd(builder, b, const_half, "");
-            c = LLVMBuildFAdd(builder, c, const_half, "");
-            d = LLVMBuildFAdd(builder, d, const_half, "");
-            
-            src_int0 = LLVMBuildFPToSI(builder, a, int32_vec_type, "");
-            src_int1 = LLVMBuildFPToSI(builder, b, int32_vec_type, "");
-            src_int2 = LLVMBuildFPToSI(builder, c, int32_vec_type, "");
-            src_int3 = LLVMBuildFPToSI(builder, d, int32_vec_type, "");
-         }
-         else {
             struct lp_build_context bld;
 
             bld.builder = builder;
@@ -329,7 +371,7 @@ lp_build_conv(LLVMBuilderRef builder,
             src_int2 = lp_build_iround(&bld, c);
             src_int3 = lp_build_iround(&bld, d);
          }
-
+         /* relying on clamping behavior of sse2 intrinsics here */
          lo = lp_build_pack2(builder, int32_type, int16_type, src_int0, src_int1);
          hi = lp_build_pack2(builder, int32_type, int16_type, src_int2, src_int3);
          dst[i] = lp_build_pack2(builder, int16_type, dst_type, lo, hi);

src/gallium/auxiliary/gallivm/lp_bld_debug.c

@@ -57,6 +57,8 @@ lp_disassemble(const void* func)
 #ifdef HAVE_UDIS86
    ud_t ud_obj;
    uint64_t max_jmp_pc;
+   uint inst_no;
+   boolean emit_addrs = TRUE, emit_line_nos = FALSE;
 
    ud_init(&ud_obj);
 
@@ -76,13 +78,18 @@ lp_disassemble(const void* func)
 
    while (ud_disassemble(&ud_obj)) {
 
+      if (emit_addrs) {
 #ifdef PIPE_ARCH_X86
          debug_printf("0x%08lx:\t", (unsigned long)ud_insn_off(&ud_obj));
 #endif
 #ifdef PIPE_ARCH_X86_64
          debug_printf("0x%016llx:\t", (unsigned long long)ud_insn_off(&ud_obj));
 #endif
-
+      }
+      else if (emit_line_nos) {
+         debug_printf("%6d:\t", inst_no);
+         inst_no++;
+      }
 #if 0
       debug_printf("%-16s ", ud_insn_hex(&ud_obj));
 #endif
@@ -115,8 +122,10 @@ lp_disassemble(const void* func)
          }
       }
 
-      if ((ud_insn_off(&ud_obj) >= max_jmp_pc && ud_obj.mnemonic == UD_Iret) ||
+      if (ud_obj.mnemonic == UD_Iinvalid ||
-           ud_obj.mnemonic == UD_Iinvalid)
+          (ud_insn_off(&ud_obj) >= max_jmp_pc &&
+           (ud_obj.mnemonic == UD_Iret ||
+            ud_obj.mnemonic == UD_Ijmp)))
          break;
    }
 

src/gallium/auxiliary/gallivm/lp_bld_debug.h

@@ -36,11 +36,12 @@
 #include "util/u_string.h"
 
 
-#define GALLIVM_DEBUG_TGSI      0x1
+#define GALLIVM_DEBUG_TGSI          (1 << 0)
-#define GALLIVM_DEBUG_IR        0x2
+#define GALLIVM_DEBUG_IR            (1 << 1)
-#define GALLIVM_DEBUG_ASM       0x4
+#define GALLIVM_DEBUG_ASM           (1 << 2)
-#define GALLIVM_DEBUG_NO_OPT    0x8
+#define GALLIVM_DEBUG_NO_OPT        (1 << 3)
-#define GALLIVM_DEBUG_PERF      0x10
+#define GALLIVM_DEBUG_PERF          (1 << 4)
+#define GALLIVM_DEBUG_NO_BRILINEAR  (1 << 5)
 
 
 #ifdef DEBUG

src/gallium/auxiliary/gallivm/lp_bld_flow.c

@@ -38,273 +38,15 @@
 #include "lp_bld_flow.h"
 
 
-#define LP_BUILD_FLOW_MAX_VARIABLES 64
-#define LP_BUILD_FLOW_MAX_DEPTH 32
-
 /**
- * Enumeration of all possible flow constructs.
+ * Insert a new block, right where builder is pointing to.
- */
-enum lp_build_flow_construct_kind {
-   LP_BUILD_FLOW_SCOPE,
-   LP_BUILD_FLOW_SKIP,
-   LP_BUILD_FLOW_IF
-};
-
-
-/**
- * Variable declaration scope.
- */
-struct lp_build_flow_scope
-{
-   /** Number of variables declared in this scope */
-   unsigned num_variables;
-};
-
-
-/**
- * Early exit. Useful to skip to the end of a function or block when
- * the execution mask becomes zero or when there is an error condition.
- */
-struct lp_build_flow_skip
-{
-   /** Block to skip to */
-   LLVMBasicBlockRef block;
-
-   /** Number of variables declared at the beginning */
-   unsigned num_variables;
-
-   LLVMValueRef *phi;  /**< array [num_variables] */
-};
-
-
-/**
- * if/else/endif.
- */
-struct lp_build_flow_if
-{
-   unsigned num_variables;
-
-   LLVMValueRef *phi;  /**< array [num_variables] */
-
-   LLVMValueRef condition;
-   LLVMBasicBlockRef entry_block, true_block, false_block, merge_block;
-};
-
-
-/**
- * Union of all possible flow constructs' data
- */
-union lp_build_flow_construct_data
-{
-   struct lp_build_flow_scope scope;
-   struct lp_build_flow_skip skip;
-   struct lp_build_flow_if ifthen;
-};
-
-
-/**
- * Element of the flow construct stack.
- */
-struct lp_build_flow_construct
-{
-   enum lp_build_flow_construct_kind kind;
-   union lp_build_flow_construct_data data;
-};
-
-
-/**
- * All necessary data to generate LLVM control flow constructs.
  *
- * Besides keeping track of the control flow construct themselves we also
+ * This is useful important not only for aesthetic reasons, but also for
- * need to keep track of variables in order to generate SSA Phi values.
+ * performance reasons, as frequently run blocks should be laid out next to
- */
+ * each other and fall-throughs maximized.
-struct lp_build_flow_context
-{
-   LLVMBuilderRef builder;
-
-   /**
-    * Control flow stack.
-    */
-   struct lp_build_flow_construct constructs[LP_BUILD_FLOW_MAX_DEPTH];
-   unsigned num_constructs;
-
-   /**
-    * Variable stack
-    */
-   LLVMValueRef *variables[LP_BUILD_FLOW_MAX_VARIABLES];
-   unsigned num_variables;
-};
-
-
-struct lp_build_flow_context *
-lp_build_flow_create(LLVMBuilderRef builder)
-{
-   struct lp_build_flow_context *flow;
-
-   flow = CALLOC_STRUCT(lp_build_flow_context);
-   if(!flow)
-      return NULL;
-
-   flow->builder = builder;
-
-   return flow;
-}
-
-
-void
-lp_build_flow_destroy(struct lp_build_flow_context *flow)
-{
-   assert(flow->num_constructs == 0);
-   assert(flow->num_variables == 0);
-   FREE(flow);
-}
-
-
-/**
- * Begin/push a new flow control construct, such as a loop, skip block
- * or variable scope.
- */
-static union lp_build_flow_construct_data *
-lp_build_flow_push(struct lp_build_flow_context *flow,
-                   enum lp_build_flow_construct_kind kind)
-{
-   assert(flow->num_constructs < LP_BUILD_FLOW_MAX_DEPTH);
-   if(flow->num_constructs >= LP_BUILD_FLOW_MAX_DEPTH)
-      return NULL;
-
-   flow->constructs[flow->num_constructs].kind = kind;
-   return &flow->constructs[flow->num_constructs++].data;
-}
-
-
-/**
- * Return the current/top flow control construct on the stack.
- * \param kind  the expected type of the top-most construct
- */
-static union lp_build_flow_construct_data *
-lp_build_flow_peek(struct lp_build_flow_context *flow,
-                   enum lp_build_flow_construct_kind kind)
-{
-   assert(flow->num_constructs);
-   if(!flow->num_constructs)
-      return NULL;
-
-   assert(flow->constructs[flow->num_constructs - 1].kind == kind);
-   if(flow->constructs[flow->num_constructs - 1].kind != kind)
-      return NULL;
-
-   return &flow->constructs[flow->num_constructs - 1].data;
-}
-
-
-/**
- * End/pop the current/top flow control construct on the stack.
- * \param kind  the expected type of the top-most construct
- */
-static union lp_build_flow_construct_data *
-lp_build_flow_pop(struct lp_build_flow_context *flow,
-                  enum lp_build_flow_construct_kind kind)
-{
-   assert(flow->num_constructs);
-   if(!flow->num_constructs)
-      return NULL;
-
-   assert(flow->constructs[flow->num_constructs - 1].kind == kind);
-   if(flow->constructs[flow->num_constructs - 1].kind != kind)
-      return NULL;
-
-   return &flow->constructs[--flow->num_constructs].data;
-}
-
-
-/**
- * Begin a variable scope.
  *
+ * See also llvm/lib/Transforms/Scalar/BasicBlockPlacement.cpp.
  *
- */
-void
-lp_build_flow_scope_begin(struct lp_build_flow_context *flow)
-{
-   struct lp_build_flow_scope *scope;
-
-   scope = &lp_build_flow_push(flow, LP_BUILD_FLOW_SCOPE)->scope;
-   if(!scope)
-      return;
-
-   scope->num_variables = 0;
-}
-
-
-/**
- * Declare a variable.
- *
- * A variable is a named entity which can have different LLVMValueRef's at
- * different points of the program. This is relevant for control flow because
- * when there are multiple branches to a same location we need to replace
- * the variable's value with a Phi function as explained in
- * http://en.wikipedia.org/wiki/Static_single_assignment_form .
- *
- * We keep track of variables by keeping around a pointer to where they're
- * current.
- *
- * There are a few cautions to observe:
- *
- * - Variable's value must not be NULL. If there is no initial value then
- *   LLVMGetUndef() should be used.
- *
- * - Variable's value must be kept up-to-date. If the variable is going to be
- *   modified by a function then a pointer should be passed so that its value
- *   is accurate. Failure to do this will cause some of the variables'
- *   transient values to be lost, leading to wrong results.
- *
- * - A program should be written from top to bottom, by always appending
- *   instructions to the bottom with a single LLVMBuilderRef. Inserting and/or
- *   modifying existing statements will most likely lead to wrong results.
- *
- */
-void
-lp_build_flow_scope_declare(struct lp_build_flow_context *flow,
-                            LLVMValueRef *variable)
-{
-   struct lp_build_flow_scope *scope;
-
-   scope = &lp_build_flow_peek(flow, LP_BUILD_FLOW_SCOPE)->scope;
-   if(!scope)
-      return;
-
-   assert(*variable);
-   if(!*variable)
-      return;
-
-   assert(flow->num_variables < LP_BUILD_FLOW_MAX_VARIABLES);
-   if(flow->num_variables >= LP_BUILD_FLOW_MAX_VARIABLES)
-      return;
-
-   flow->variables[flow->num_variables++] = variable;
-   ++scope->num_variables;
-}
-
-
-void
-lp_build_flow_scope_end(struct lp_build_flow_context *flow)
-{
-   struct lp_build_flow_scope *scope;
-
-   scope = &lp_build_flow_pop(flow, LP_BUILD_FLOW_SCOPE)->scope;
-   if(!scope)
-      return;
-
-   assert(flow->num_variables >= scope->num_variables);
-   if(flow->num_variables < scope->num_variables) {
-      flow->num_variables = 0;
-      return;
-   }
-
-   flow->num_variables -= scope->num_variables;
-}
-
-
-/**
  * Note: this function has no dependencies on the flow code and could
  * be used elsewhere.
  */
@@ -334,52 +76,18 @@ lp_build_insert_new_block(LLVMBuilderRef builder, const char *name)
 }
 
 
-static LLVMBasicBlockRef
-lp_build_flow_insert_block(struct lp_build_flow_context *flow)
-{
-   return lp_build_insert_new_block(flow->builder, "");
-}
-
-
 /**
  * Begin a "skip" block.  Inside this block we can test a condition and
  * skip to the end of the block if the condition is false.
  */
 void
-lp_build_flow_skip_begin(struct lp_build_flow_context *flow)
+lp_build_flow_skip_begin(struct lp_build_skip_context *skip,
+                         LLVMBuilderRef builder)
 {
-   struct lp_build_flow_skip *skip;
+   skip->builder = builder;
-   LLVMBuilderRef builder;
-   unsigned i;
-
-   skip = &lp_build_flow_push(flow, LP_BUILD_FLOW_SKIP)->skip;
-   if(!skip)
-      return;
 
    /* create new basic block */
-   skip->block = lp_build_flow_insert_block(flow);
+   skip->block = lp_build_insert_new_block(skip->builder, "skip");
-
-   skip->num_variables = flow->num_variables;
-   if(!skip->num_variables) {
-      skip->phi = NULL;
-      return;
-   }
-
-   /* Allocate a Phi node for each variable in this skip scope */
-   skip->phi = MALLOC(skip->num_variables * sizeof *skip->phi);
-   if(!skip->phi) {
-      skip->num_variables = 0;
-      return;
-   }
-
-   builder = LLVMCreateBuilder();
-   LLVMPositionBuilderAtEnd(builder, skip->block);
-
-   /* create a Phi node for each variable */
-   for(i = 0; i < skip->num_variables; ++i)
-      skip->phi[i] = LLVMBuildPhi(builder, LLVMTypeOf(*flow->variables[i]), "");
-
-   LLVMDisposeBuilder(builder);
 }
 
 
@@ -388,83 +96,50 @@ lp_build_flow_skip_begin(struct lp_build_flow_context *flow)
  * skip block if the condition is true.
  */
 void
-lp_build_flow_skip_cond_break(struct lp_build_flow_context *flow,
+lp_build_flow_skip_cond_break(struct lp_build_skip_context *skip,
                               LLVMValueRef cond)
 {
-   struct lp_build_flow_skip *skip;
-   LLVMBasicBlockRef current_block;
    LLVMBasicBlockRef new_block;
-   unsigned i;
 
-   skip = &lp_build_flow_peek(flow, LP_BUILD_FLOW_SKIP)->skip;
+   new_block = lp_build_insert_new_block(skip->builder, "");
-   if(!skip)
-      return;
-
-   current_block = LLVMGetInsertBlock(flow->builder);
-
-   new_block = lp_build_flow_insert_block(flow);
-
-   /* for each variable, update the Phi node with a (variable, block) pair */
-   for(i = 0; i < skip->num_variables; ++i) {
-      assert(*flow->variables[i]);
-      assert(LLVMTypeOf(skip->phi[i]) == LLVMTypeOf(*flow->variables[i]));
-      LLVMAddIncoming(skip->phi[i], flow->variables[i], &current_block, 1);
-   }
 
    /* if cond is true, goto skip->block, else goto new_block */
-   LLVMBuildCondBr(flow->builder, cond, skip->block, new_block);
+   LLVMBuildCondBr(skip->builder, cond, skip->block, new_block);
 
-   LLVMPositionBuilderAtEnd(flow->builder, new_block);
+   LLVMPositionBuilderAtEnd(skip->builder, new_block);
 }
 
 
 void
-lp_build_flow_skip_end(struct lp_build_flow_context *flow)
+lp_build_flow_skip_end(struct lp_build_skip_context *skip)
 {
-   struct lp_build_flow_skip *skip;
-   LLVMBasicBlockRef current_block;
-   unsigned i;
-
-   skip = &lp_build_flow_pop(flow, LP_BUILD_FLOW_SKIP)->skip;
-   if(!skip)
-      return;
-
-   current_block = LLVMGetInsertBlock(flow->builder);
-
-   /* add (variable, block) tuples to the phi nodes */
-   for(i = 0; i < skip->num_variables; ++i) {
-      assert(*flow->variables[i]);
-      assert(LLVMTypeOf(skip->phi[i]) == LLVMTypeOf(*flow->variables[i]));
-      LLVMAddIncoming(skip->phi[i], flow->variables[i], &current_block, 1);
-      *flow->variables[i] = skip->phi[i];
-   }
-
    /* goto block */
-   LLVMBuildBr(flow->builder, skip->block);
+   LLVMBuildBr(skip->builder, skip->block);
-   LLVMPositionBuilderAtEnd(flow->builder, skip->block);
+   LLVMPositionBuilderAtEnd(skip->builder, skip->block);
-
-   FREE(skip->phi);
 }
 
 
 /**
  * Check if the mask predicate is zero.  If so, jump to the end of the block.
  */
-static void
+void
 lp_build_mask_check(struct lp_build_mask_context *mask)
 {
-   LLVMBuilderRef builder = mask->flow->builder;
+   LLVMBuilderRef builder = mask->skip.builder;
+   LLVMValueRef value;
    LLVMValueRef cond;
 
+   value = lp_build_mask_value(mask);
+
    /* cond = (mask == 0) */
    cond = LLVMBuildICmp(builder,
                         LLVMIntEQ,
-                        LLVMBuildBitCast(builder, mask->value, mask->reg_type, ""),
+                        LLVMBuildBitCast(builder, value, mask->reg_type, ""),
                         LLVMConstNull(mask->reg_type),
                         "");
 
    /* if cond, goto end of block */
-   lp_build_flow_skip_cond_break(mask->flow, cond);
+   lp_build_flow_skip_cond_break(&mask->skip, cond);
 }
 
 
@@ -477,21 +152,27 @@ lp_build_mask_check(struct lp_build_mask_context *mask)
  */
 void
 lp_build_mask_begin(struct lp_build_mask_context *mask,
-                    struct lp_build_flow_context *flow,
+                    LLVMBuilderRef builder,
                     struct lp_type type,
                     LLVMValueRef value)
 {
    memset(mask, 0, sizeof *mask);
 
-   mask->flow = flow;
    mask->reg_type = LLVMIntType(type.width * type.length);
-   mask->value = value;
+   mask->var = lp_build_alloca(builder,
+                               lp_build_int_vec_type(type),
+                               "execution_mask");
 
-   lp_build_flow_scope_begin(flow);
+   LLVMBuildStore(builder, value, mask->var);
-   lp_build_flow_scope_declare(flow, &mask->value);
-   lp_build_flow_skip_begin(flow);
 
-   lp_build_mask_check(mask);
+   lp_build_flow_skip_begin(&mask->skip, builder);
+}
+
+
+LLVMValueRef
+lp_build_mask_value(struct lp_build_mask_context *mask)
+{
+   return LLVMBuildLoad(mask->skip.builder, mask->var, "");
 }
 
 
@@ -504,9 +185,10 @@ void
 lp_build_mask_update(struct lp_build_mask_context *mask,
                      LLVMValueRef value)
 {
-   mask->value = LLVMBuildAnd( mask->flow->builder, mask->value, value, "");
+   value = LLVMBuildAnd(mask->skip.builder,
-
+                        lp_build_mask_value(mask),
-   lp_build_mask_check(mask);
+                        value, "");
+   LLVMBuildStore(mask->skip.builder, value, mask->var);
 }
 
 
@@ -516,9 +198,8 @@ lp_build_mask_update(struct lp_build_mask_context *mask,
 LLVMValueRef
 lp_build_mask_end(struct lp_build_mask_context *mask)
 {
-   lp_build_flow_skip_end(mask->flow);
+   lp_build_flow_skip_end(&mask->skip);
-   lp_build_flow_scope_end(mask->flow);
+   return lp_build_mask_value(mask);
-   return mask->value;
 }
 
 
@@ -528,19 +209,47 @@ lp_build_loop_begin(LLVMBuilderRef builder,
                     LLVMValueRef start,
                     struct lp_build_loop_state *state)
 {
-   LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
+   state->block = lp_build_insert_new_block(builder, "loop_begin");
-   LLVMValueRef function = LLVMGetBasicBlockParent(block);
 
-   state->block = LLVMAppendBasicBlock(function, "loop");
+   state->counter_var = lp_build_alloca(builder, LLVMTypeOf(start), "loop_counter");
+
+   LLVMBuildStore(builder, start, state->counter_var);
 
    LLVMBuildBr(builder, state->block);
 
    LLVMPositionBuilderAtEnd(builder, state->block);
 
-   state->counter = LLVMBuildPhi(builder, LLVMTypeOf(start), "");
+   state->counter = LLVMBuildLoad(builder, state->counter_var, "");
+}
 
-   LLVMAddIncoming(state->counter, &start, &block, 1);
 
+void
+lp_build_loop_end_cond(LLVMBuilderRef builder,
+                       LLVMValueRef end,
+                       LLVMValueRef step,
+                       LLVMIntPredicate llvm_cond,
+                       struct lp_build_loop_state *state)
+{
+   LLVMValueRef next;
+   LLVMValueRef cond;
+   LLVMBasicBlockRef after_block;
+
+   if (!step)
+      step = LLVMConstInt(LLVMTypeOf(end), 1, 0);
+
+   next = LLVMBuildAdd(builder, state->counter, step, "");
+
+   LLVMBuildStore(builder, next, state->counter_var);
+
+   cond = LLVMBuildICmp(builder, llvm_cond, next, end, "");
+
+   after_block = lp_build_insert_new_block(builder, "loop_end");
+
+   LLVMBuildCondBr(builder, cond, after_block, state->block);
+
+   LLVMPositionBuilderAtEnd(builder, after_block);
+
+   state->counter = LLVMBuildLoad(builder, state->counter_var, "");
 }
 
 
@@ -550,55 +259,7 @@ lp_build_loop_end(LLVMBuilderRef builder,
                   LLVMValueRef step,
                   struct lp_build_loop_state *state)
 {
-   LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
+   lp_build_loop_end_cond(builder, end, step, LLVMIntNE, state);
-   LLVMValueRef function = LLVMGetBasicBlockParent(block);
-   LLVMValueRef next;
-   LLVMValueRef cond;
-   LLVMBasicBlockRef after_block;
-
-   if (!step)
-      step = LLVMConstInt(LLVMTypeOf(end), 1, 0);
-
-   next = LLVMBuildAdd(builder, state->counter, step, "");
-
-   cond = LLVMBuildICmp(builder, LLVMIntNE, next, end, "");
-
-   after_block = LLVMAppendBasicBlock(function, "");
-
-   LLVMBuildCondBr(builder, cond, after_block, state->block);
-
-   LLVMAddIncoming(state->counter, &next, &block, 1);
-
-   LLVMPositionBuilderAtEnd(builder, after_block);
-}
-
-void
-lp_build_loop_end_cond(LLVMBuilderRef builder,
-                       LLVMValueRef end,
-                       LLVMValueRef step,
-                       int llvm_cond,
-                       struct lp_build_loop_state *state)
-{
-   LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
-   LLVMValueRef function = LLVMGetBasicBlockParent(block);
-   LLVMValueRef next;
-   LLVMValueRef cond;
-   LLVMBasicBlockRef after_block;
-
-   if (!step)
-      step = LLVMConstInt(LLVMTypeOf(end), 1, 0);
-
-   next = LLVMBuildAdd(builder, state->counter, step, "");
-
-   cond = LLVMBuildICmp(builder, llvm_cond, next, end, "");
-
-   after_block = LLVMAppendBasicBlock(function, "");
-
-   LLVMBuildCondBr(builder, cond, after_block, state->block);
-
-   LLVMAddIncoming(state->counter, &next, &block, 1);
-
-   LLVMPositionBuilderAtEnd(builder, after_block);
 }
 
 
@@ -616,24 +277,16 @@ lp_build_loop_end_cond(LLVMBuilderRef builder,
 
   Is built with:
 
-     LLVMValueRef x = LLVMGetUndef();  // or something else
+     // x needs an alloca variable
+     x = lp_build_alloca(builder, type, "x");
 
-     flow = lp_build_flow_create(builder);
 
-        lp_build_flow_scope_begin(flow);
+     lp_build_if(ctx, builder, cond);
-
+        LLVMBuildStore(LLVMBuildAdd(1, 2), x);
-           // x needs a phi node
-           lp_build_flow_scope_declare(flow, &x);
-
-           lp_build_if(ctx, flow, builder, cond);
-              x = LLVMAdd(1, 2);
      lp_build_else(ctx);
-              x = LLVMAdd(2, 3);
+        LLVMBuildStore(LLVMBuildAdd(2, 3). x);
      lp_build_endif(ctx);
 
-        lp_build_flow_scope_end(flow);
-
-     lp_build_flow_destroy(flow);
  */
 
 
@@ -642,47 +295,19 @@ lp_build_loop_end_cond(LLVMBuilderRef builder,
  * Begin an if/else/endif construct.
  */
 void
-lp_build_if(struct lp_build_if_state *ctx,
+lp_build_if(struct lp_build_if_state *ifthen,
-            struct lp_build_flow_context *flow,
             LLVMBuilderRef builder,
             LLVMValueRef condition)
 {
    LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
-   struct lp_build_flow_if *ifthen;
-   unsigned i;
 
-   memset(ctx, 0, sizeof(*ctx));
+   memset(ifthen, 0, sizeof *ifthen);
-   ctx->builder = builder;
+   ifthen->builder = builder;
-   ctx->flow = flow;
-
-   /* push/create new scope */
-   ifthen = &lp_build_flow_push(flow, LP_BUILD_FLOW_IF)->ifthen;
-   assert(ifthen);
-
-   ifthen->num_variables = flow->num_variables;
    ifthen->condition = condition;
    ifthen->entry_block = block;
 
-   /* create a Phi node for each variable in this flow scope */
-   ifthen->phi = MALLOC(ifthen->num_variables * sizeof(*ifthen->phi));
-   if (!ifthen->phi) {
-      ifthen->num_variables = 0;
-      return;
-   }
-
    /* create endif/merge basic block for the phi functions */
    ifthen->merge_block = lp_build_insert_new_block(builder, "endif-block");
-   LLVMPositionBuilderAtEnd(builder, ifthen->merge_block);
-
-   /* create a phi node for each variable */
-   for (i = 0; i < flow->num_variables; i++) {
-      ifthen->phi[i] = LLVMBuildPhi(builder, LLVMTypeOf(*flow->variables[i]), "");
-
-      /* add add the initial value of the var from the entry block */
-      if (!LLVMIsUndef(*flow->variables[i]))
-         LLVMAddIncoming(ifthen->phi[i], flow->variables[i],
-                         &ifthen->entry_block, 1);
-   }
 
    /* create/insert true_block before merge_block */
    ifthen->true_block = LLVMInsertBasicBlock(ifthen->merge_block, "if-true-block");
@@ -696,27 +321,16 @@ lp_build_if(struct lp_build_if_state *ctx,
  * Begin else-part of a conditional
  */
 void
-lp_build_else(struct lp_build_if_state *ctx)
+lp_build_else(struct lp_build_if_state *ifthen)
 {
-   struct lp_build_flow_context *flow = ctx->flow;
+   /* Append an unconditional Br(anch) instruction on the true_block */
-   struct lp_build_flow_if *ifthen;
+   LLVMBuildBr(ifthen->builder, ifthen->merge_block);
-   unsigned i;
-
-   ifthen = &lp_build_flow_peek(flow, LP_BUILD_FLOW_IF)->ifthen;
-   assert(ifthen);
-
-   /* for each variable, update the Phi node with a (variable, block) pair */
-   LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
-   for (i = 0; i < flow->num_variables; i++) {
-      assert(*flow->variables[i]);
-      LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &ifthen->true_block, 1);
-   }
 
    /* create/insert false_block before the merge block */
    ifthen->false_block = LLVMInsertBasicBlock(ifthen->merge_block, "if-false-block");
 
    /* successive code goes into the else block */
-   LLVMPositionBuilderAtEnd(ctx->builder, ifthen->false_block);
+   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->false_block);
 }
 
 
@@ -724,75 +338,30 @@ lp_build_else(struct lp_build_if_state *ctx)
  * End a conditional.
  */
 void
-lp_build_endif(struct lp_build_if_state *ctx)
+lp_build_endif(struct lp_build_if_state *ifthen)
 {
-   struct lp_build_flow_context *flow = ctx->flow;
-   struct lp_build_flow_if *ifthen;
-   LLVMBasicBlockRef curBlock = LLVMGetInsertBlock(ctx->builder);
-   unsigned i;
-
-   ifthen = &lp_build_flow_pop(flow, LP_BUILD_FLOW_IF)->ifthen;
-   assert(ifthen);
-
    /* Insert branch to the merge block from current block */
-   LLVMBuildBr(ctx->builder, ifthen->merge_block);
+   LLVMBuildBr(ifthen->builder, ifthen->merge_block);
 
-   if (ifthen->false_block) {
+   /*
-      LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
+    * Now patch in the various branch instructions.
-      /* for each variable, update the Phi node with a (variable, block) pair */
+    */
-      for (i = 0; i < flow->num_variables; i++) {
-         assert(*flow->variables[i]);
-         LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &curBlock, 1);
-         /* replace the variable ref with the phi function */
-         *flow->variables[i] = ifthen->phi[i];
-      }
-   }
-   else {
-      /* no else clause */
-      LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
-      for (i = 0; i < flow->num_variables; i++) {
-         assert(*flow->variables[i]);
-         LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &ifthen->true_block, 1);
-
-         /* replace the variable ref with the phi function */
-         *flow->variables[i] = ifthen->phi[i];
-      }
-   }
-
-   FREE(ifthen->phi);
-
-   /***
-    *** Now patch in the various branch instructions.
-    ***/
 
    /* Insert the conditional branch instruction at the end of entry_block */
-   LLVMPositionBuilderAtEnd(ctx->builder, ifthen->entry_block);
+   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->entry_block);
    if (ifthen->false_block) {
       /* we have an else clause */
-      LLVMBuildCondBr(ctx->builder, ifthen->condition,
+      LLVMBuildCondBr(ifthen->builder, ifthen->condition,
                       ifthen->true_block, ifthen->false_block);
    }
    else {
       /* no else clause */
-      LLVMBuildCondBr(ctx->builder, ifthen->condition,
+      LLVMBuildCondBr(ifthen->builder, ifthen->condition,
                       ifthen->true_block, ifthen->merge_block);
    }
 
-   /* Insert branch from end of true_block to merge_block */
-   if (ifthen->false_block) {
-      /* Append an unconditional Br(anch) instruction on the true_block */
-      LLVMPositionBuilderAtEnd(ctx->builder, ifthen->true_block);
-      LLVMBuildBr(ctx->builder, ifthen->merge_block);
-   }
-   else {
-      /* No else clause.
-       * Note that we've already inserted the branch at the end of
-       * true_block.  See the very first LLVMBuildBr() call in this function.
-       */
-   }
-
    /* Resume building code at end of the ifthen->merge_block */
-   LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
+   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->merge_block);
 }
 
 
@@ -830,6 +399,7 @@ lp_build_alloca(LLVMBuilderRef builder,
    }
 
    res = LLVMBuildAlloca(first_builder, type, name);
+   LLVMBuildStore(builder, LLVMConstNull(type), res);
 
    LLVMDisposeBuilder(first_builder);
 

src/gallium/auxiliary/gallivm/lp_bld_flow.h

@@ -41,52 +41,49 @@
 struct lp_type;
 
 
-struct lp_build_flow_context;
+/**
+ * Early exit. Useful to skip to the end of a function or block when
+ * the execution mask becomes zero or when there is an error condition.
+ */
+struct lp_build_skip_context
+{
+   LLVMBuilderRef builder;
 
-
+   /** Block to skip to */
-struct lp_build_flow_context *
+   LLVMBasicBlockRef block;
-lp_build_flow_create(LLVMBuilderRef builder);
+};
 
 void
-lp_build_flow_destroy(struct lp_build_flow_context *flow);
+lp_build_flow_skip_begin(struct lp_build_skip_context *ctx,
+                         LLVMBuilderRef builder);
 
 void
-lp_build_flow_scope_begin(struct lp_build_flow_context *flow);
+lp_build_flow_skip_cond_break(struct lp_build_skip_context *ctx,
-
-void
-lp_build_flow_scope_declare(struct lp_build_flow_context *flow,
-                            LLVMValueRef *variable);
-
-void
-lp_build_flow_scope_end(struct lp_build_flow_context *flow);
-
-void
-lp_build_flow_skip_begin(struct lp_build_flow_context *flow);
-
-void
-lp_build_flow_skip_cond_break(struct lp_build_flow_context *flow,
                               LLVMValueRef cond);
 
 void
-lp_build_flow_skip_end(struct lp_build_flow_context *flow);
+lp_build_flow_skip_end(struct lp_build_skip_context *ctx);
 
 
 struct lp_build_mask_context
 {
-   struct lp_build_flow_context *flow;
+   struct lp_build_skip_context skip;
 
    LLVMTypeRef reg_type;
 
-   LLVMValueRef value;
+   LLVMValueRef var;
 };
 
 
 void
 lp_build_mask_begin(struct lp_build_mask_context *mask,
-                    struct lp_build_flow_context *flow,
+                    LLVMBuilderRef builder,
                     struct lp_type type,
                     LLVMValueRef value);
 
+LLVMValueRef
+lp_build_mask_value(struct lp_build_mask_context *mask);
+
 /**
  * Bitwise AND the mask with the given value, if a previous mask was set.
  */
@@ -94,6 +91,9 @@ void
 lp_build_mask_update(struct lp_build_mask_context *mask,
                      LLVMValueRef value);
 
+void
+lp_build_mask_check(struct lp_build_mask_context *mask);
+
 LLVMValueRef
 lp_build_mask_end(struct lp_build_mask_context *mask);
 
@@ -108,6 +108,7 @@ lp_build_mask_end(struct lp_build_mask_context *mask);
 struct lp_build_loop_state
 {
   LLVMBasicBlockRef block;
+  LLVMValueRef counter_var;
   LLVMValueRef counter;
 };
 
@@ -128,22 +129,28 @@ void
 lp_build_loop_end_cond(LLVMBuilderRef builder,
                        LLVMValueRef end,
                        LLVMValueRef step,
-                       int cond, /* LLVM condition */
+                       LLVMIntPredicate cond,
                        struct lp_build_loop_state *state);
 
 
 
 
+/**
+ * if/else/endif.
+ */
 struct lp_build_if_state
 {
    LLVMBuilderRef builder;
-   struct lp_build_flow_context *flow;
+   LLVMValueRef condition;
+   LLVMBasicBlockRef entry_block;
+   LLVMBasicBlockRef true_block;
+   LLVMBasicBlockRef false_block;
+   LLVMBasicBlockRef merge_block;
 };
 
 
 void
 lp_build_if(struct lp_build_if_state *ctx,
-            struct lp_build_flow_context *flow,
             LLVMBuilderRef builder,
             LLVMValueRef condition);
 

src/gallium/auxiliary/gallivm/lp_bld_init.c

@@ -44,6 +44,7 @@ static const struct debug_named_value lp_bld_debug_flags[] = {
    { "asm",    GALLIVM_DEBUG_ASM, NULL },
    { "nopt",   GALLIVM_DEBUG_NO_OPT, NULL },
    { "perf",   GALLIVM_DEBUG_PERF, NULL },
+   { "no_brilinear", GALLIVM_DEBUG_NO_BRILINEAR, NULL },
    DEBUG_NAMED_VALUE_END
 };
 

src/gallium/auxiliary/gallivm/lp_bld_init.h

@@ -47,4 +47,10 @@ lp_build_init(void);
 extern void
 lp_func_delete_body(LLVMValueRef func);
 
+
+extern LLVMValueRef
+lp_build_load_volatile(LLVMBuilderRef B, LLVMValueRef PointerVal,
+                       const char *Name);
+
+
 #endif /* !LP_BLD_INIT_H */

src/gallium/auxiliary/gallivm/lp_bld_logic.c

@@ -462,10 +462,12 @@ lp_build_select(struct lp_build_context *bld,
       LLVMTypeRef arg_type;
       LLVMValueRef args[3];
 
-      if (type.width == 64) {
+      if (type.floating &&
+          type.width == 64) {
          intrinsic = "llvm.x86.sse41.blendvpd";
          arg_type = LLVMVectorType(LLVMDoubleType(), 2);
-      } else if (type.width == 32) {
+      } else if (type.floating &&
+                 type.width == 32) {
          intrinsic = "llvm.x86.sse41.blendvps";
          arg_type = LLVMVectorType(LLVMFloatType(), 4);
       } else {

src/gallium/auxiliary/gallivm/lp_bld_misc.cpp

@@ -178,3 +178,13 @@ lp_func_delete_body(LLVMValueRef FF)
    llvm::Function *func = llvm::unwrap<llvm::Function>(FF);
    func->deleteBody();
 }
+
+
+extern "C"
+LLVMValueRef
+lp_build_load_volatile(LLVMBuilderRef B, LLVMValueRef PointerVal,
+                       const char *Name)
+{
+   return llvm::wrap(llvm::unwrap(B)->CreateLoad(llvm::unwrap(PointerVal), true, Name));
+}
+

src/gallium/auxiliary/gallivm/lp_bld_printf.c

@@ -29,6 +29,8 @@
 
 #include "util/u_debug.h"
 #include "util/u_memory.h"
+#include "util/u_string.h"
+#include "lp_bld_const.h"
 #include "lp_bld_printf.h"
 
 
@@ -119,3 +121,22 @@ lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...)
    return LLVMBuildCall(builder, func_printf, params, argcount + 1, "");
 }
 
+
+
+/**
+ * Print a float[4] vector.
+ */
+LLVMValueRef
+lp_build_print_vec4(LLVMBuilderRef builder, const char *msg, LLVMValueRef vec)
+{
+   char format[1000];
+   LLVMValueRef x, y, z, w;
+
+   x = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(0), "");
+   y = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(1), "");
+   z = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(2), "");
+   w = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(3), "");
+
+   util_snprintf(format, sizeof(format), "%s %%f %%f %%f %%f\n", msg);
+   return lp_build_printf(builder, format, x, y, z, w);
+}

src/gallium/auxiliary/gallivm/lp_bld_printf.h

@@ -35,5 +35,9 @@
 LLVMValueRef lp_build_const_string_variable(LLVMModuleRef module, const char *str, int len);
 LLVMValueRef lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...);
 
+LLVMValueRef
+lp_build_print_vec4(LLVMBuilderRef builder, const char *msg, LLVMValueRef vec);
+
+
 #endif
 

src/gallium/auxiliary/gallivm/lp_bld_quad.c

@@ -81,11 +81,15 @@ LLVMValueRef
 lp_build_scalar_ddx(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
-   LLVMValueRef idx_left  = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_LEFT, 0);
+   LLVMTypeRef i32t = LLVMInt32Type();
-   LLVMValueRef idx_right = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_RIGHT, 0);
+   LLVMValueRef idx_left  = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0);
-   LLVMValueRef a_left  = LLVMBuildExtractElement(bld->builder, a, idx_left, "");
+   LLVMValueRef idx_right = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_RIGHT, 0);
-   LLVMValueRef a_right = LLVMBuildExtractElement(bld->builder, a, idx_right, "");
+   LLVMValueRef a_left  = LLVMBuildExtractElement(bld->builder, a, idx_left, "left");
-   return lp_build_sub(bld, a_right, a_left);
+   LLVMValueRef a_right = LLVMBuildExtractElement(bld->builder, a, idx_right, "right");
+   if (bld->type.floating)
+      return LLVMBuildFSub(bld->builder, a_right, a_left, "ddx");
+   else
+      return LLVMBuildSub(bld->builder, a_right, a_left, "ddx");
 }
 
 
@@ -93,9 +97,13 @@ LLVMValueRef
 lp_build_scalar_ddy(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
-   LLVMValueRef idx_top    = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_LEFT, 0);
+   LLVMTypeRef i32t = LLVMInt32Type();
-   LLVMValueRef idx_bottom = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_BOTTOM_LEFT, 0);
+   LLVMValueRef idx_top    = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0);
-   LLVMValueRef a_top    = LLVMBuildExtractElement(bld->builder, a, idx_top, "");
+   LLVMValueRef idx_bottom = LLVMConstInt(i32t, LP_BLD_QUAD_BOTTOM_LEFT, 0);
-   LLVMValueRef a_bottom = LLVMBuildExtractElement(bld->builder, a, idx_bottom, "");
+   LLVMValueRef a_top    = LLVMBuildExtractElement(bld->builder, a, idx_top, "top");
-   return lp_build_sub(bld, a_bottom, a_top);
+   LLVMValueRef a_bottom = LLVMBuildExtractElement(bld->builder, a, idx_bottom, "bottom");
+   if (bld->type.floating)
+      return LLVMBuildFSub(bld->builder, a_bottom, a_top, "ddy");
+   else
+      return LLVMBuildSub(bld->builder, a_bottom, a_top, "ddy");
 }

src/gallium/auxiliary/gallivm/lp_bld_sample.c

@@ -47,8 +47,7 @@
 
 
 /*
- * Bri-linear factor. Use zero or any other number less than one to force
+ * Bri-linear factor. Should be greater than one.
- * tri-linear filtering.
  */
 #define BRILINEAR_FACTOR 2
 
@@ -201,8 +200,8 @@ lp_build_rho(struct lp_build_sample_context *bld,
    LLVMValueRef float_size;
    LLVMValueRef rho;
 
-   dsdx = LLVMBuildExtractElement(bld->builder, ddx[0], index0, "dsdx");
+   dsdx = ddx[0];
-   dsdy = LLVMBuildExtractElement(bld->builder, ddy[0], index0, "dsdy");
+   dsdy = ddy[0];
 
    if (dims <= 1) {
       rho_x = dsdx;
@@ -215,15 +214,15 @@ lp_build_rho(struct lp_build_sample_context *bld,
       rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dsdx, index0, "");
       rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dsdy, index0, "");
 
-      dtdx = LLVMBuildExtractElement(bld->builder, ddx[1], index0, "dtdx");
+      dtdx = ddx[1];
-      dtdy = LLVMBuildExtractElement(bld->builder, ddy[1], index0, "dtdy");
+      dtdy = ddy[1];
 
       rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dtdx, index1, "");
       rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dtdy, index1, "");
 
       if (dims >= 3) {
-         drdx = LLVMBuildExtractElement(bld->builder, ddx[2], index0, "drdx");
+         drdx = ddx[2];
-         drdy = LLVMBuildExtractElement(bld->builder, ddy[2], index0, "drdy");
+         drdy = ddy[2];
 
          rho_x = LLVMBuildInsertElement(bld->builder, rho_x, drdx, index2, "");
          rho_y = LLVMBuildInsertElement(bld->builder, rho_y, drdy, index2, "");
@@ -294,31 +293,30 @@ lp_build_rho(struct lp_build_sample_context *bld,
  * TODO: This could be done in fixed point, where applicable.
  */
 static void
-lp_build_brilinear_lod(struct lp_build_sample_context *bld,
+lp_build_brilinear_lod(struct lp_build_context *bld,
                        LLVMValueRef lod,
                        double factor,
                        LLVMValueRef *out_lod_ipart,
                        LLVMValueRef *out_lod_fpart)
 {
-   struct lp_build_context *float_bld = &bld->float_bld;
    LLVMValueRef lod_fpart;
-   float pre_offset = (factor - 0.5)/factor - 0.5;
+   double pre_offset = (factor - 0.5)/factor - 0.5;
-   float post_offset = 1 - factor;
+   double post_offset = 1 - factor;
 
    if (0) {
       lp_build_printf(bld->builder, "lod = %f\n", lod);
    }
 
-   lod = lp_build_add(float_bld, lod,
+   lod = lp_build_add(bld, lod,
-                      lp_build_const_vec(float_bld->type, pre_offset));
+                      lp_build_const_vec(bld->type, pre_offset));
 
-   lp_build_ifloor_fract(float_bld, lod, out_lod_ipart, &lod_fpart);
+   lp_build_ifloor_fract(bld, lod, out_lod_ipart, &lod_fpart);
 
-   lod_fpart = lp_build_mul(float_bld, lod_fpart,
+   lod_fpart = lp_build_mul(bld, lod_fpart,
-                            lp_build_const_vec(float_bld->type, factor));
+                            lp_build_const_vec(bld->type, factor));
 
-   lod_fpart = lp_build_add(float_bld, lod_fpart,
+   lod_fpart = lp_build_add(bld, lod_fpart,
-                            lp_build_const_vec(float_bld->type, post_offset));
+                            lp_build_const_vec(bld->type, post_offset));
 
    /*
     * It's not necessary to clamp lod_fpart since:
@@ -335,6 +333,61 @@ lp_build_brilinear_lod(struct lp_build_sample_context *bld,
 }
 
 
+/*
+ * Combined log2 and brilinear lod computation.
+ *
+ * It's in all identical to calling lp_build_fast_log2() and
+ * lp_build_brilinear_lod() above, but by combining we can compute the interger
+ * and fractional part independently.
+ */
+static void
+lp_build_brilinear_rho(struct lp_build_context *bld,
+                       LLVMValueRef rho,
+                       double factor,
+                       LLVMValueRef *out_lod_ipart,
+                       LLVMValueRef *out_lod_fpart)
+{
+   LLVMValueRef lod_ipart;
+   LLVMValueRef lod_fpart;
+
+   const double pre_factor = (2*factor - 0.5)/(M_SQRT2*factor);
+   const double post_offset = 1 - 2*factor;
+
+   assert(bld->type.floating);
+
+   assert(lp_check_value(bld->type, rho));
+
+   /*
+    * The pre factor will make the intersections with the exact powers of two
+    * happen precisely where we want then to be, which means that the integer
+    * part will not need any post adjustments.
+    */
+   rho = lp_build_mul(bld, rho,
+                      lp_build_const_vec(bld->type, pre_factor));
+
+   /* ipart = ifloor(log2(rho)) */
+   lod_ipart = lp_build_extract_exponent(bld, rho, 0);
+
+   /* fpart = rho / 2**ipart */
+   lod_fpart = lp_build_extract_mantissa(bld, rho);
+
+   lod_fpart = lp_build_mul(bld, lod_fpart,
+                            lp_build_const_vec(bld->type, factor));
+
+   lod_fpart = lp_build_add(bld, lod_fpart,
+                            lp_build_const_vec(bld->type, post_offset));
+
+   /*
+    * Like lp_build_brilinear_lod, it's not necessary to clamp lod_fpart since:
+    * - the above expression will never produce numbers greater than one.
+    * - the mip filtering branch is only taken if lod_fpart is positive
+    */
+
+   *out_lod_ipart = lod_ipart;
+   *out_lod_fpart = lod_fpart;
+}
+
+
 /**
  * Generate code to compute texture level of detail (lambda).
  * \param ddx  partial derivatives of (s, t, r, q) with respect to X
@@ -389,17 +442,33 @@ lp_build_lod_selector(struct lp_build_sample_context *bld,
 
          rho = lp_build_rho(bld, ddx, ddy);
 
-         /* compute lod = log2(rho) */
+         /*
-         if ((mip_filter == PIPE_TEX_MIPFILTER_NONE ||
+          * Compute lod = log2(rho)
-              mip_filter == PIPE_TEX_MIPFILTER_NEAREST) &&
+          */
-             !lod_bias &&
+
+         if (!lod_bias &&
              !bld->static_state->lod_bias_non_zero &&
              !bld->static_state->apply_max_lod &&
              !bld->static_state->apply_min_lod) {
+            /*
+             * Special case when there are no post-log2 adjustments, which
+             * saves instructions but keeping the integer and fractional lod
+             * computations separate from the start.
+             */
+
+            if (mip_filter == PIPE_TEX_MIPFILTER_NONE ||
+                mip_filter == PIPE_TEX_MIPFILTER_NEAREST) {
                *out_lod_ipart = lp_build_ilog2(float_bld, rho);
                *out_lod_fpart = bld->float_bld.zero;
                return;
             }
+            if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR &&
+                !(gallivm_debug & GALLIVM_DEBUG_NO_BRILINEAR)) {
+               lp_build_brilinear_rho(float_bld, rho, BRILINEAR_FACTOR,
+                                      out_lod_ipart, out_lod_fpart);
+               return;
+            }
+         }
 
          if (0) {
             lod = lp_build_log2(float_bld, rho);
@@ -437,21 +506,22 @@ lp_build_lod_selector(struct lp_build_sample_context *bld,
    }
 
    if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
-      if (BRILINEAR_FACTOR > 1.0) {
+      if (!(gallivm_debug & GALLIVM_DEBUG_NO_BRILINEAR)) {
-         lp_build_brilinear_lod(bld, lod, BRILINEAR_FACTOR,
+         lp_build_brilinear_lod(float_bld, lod, BRILINEAR_FACTOR,
                                 out_lod_ipart, out_lod_fpart);
       }
       else {
          lp_build_ifloor_fract(float_bld, lod, out_lod_ipart, out_lod_fpart);
       }
 
-      lp_build_name(*out_lod_ipart, "lod_ipart");
       lp_build_name(*out_lod_fpart, "lod_fpart");
    }
    else {
       *out_lod_ipart = lp_build_iround(float_bld, lod);
    }
 
+   lp_build_name(*out_lod_ipart, "lod_ipart");
+
    return;
 }
 
@@ -630,37 +700,21 @@ lp_build_get_level_stride_vec(struct lp_build_sample_context *bld,
 void
 lp_build_mipmap_level_sizes(struct lp_build_sample_context *bld,
                             LLVMValueRef ilevel,
-                            LLVMValueRef *out_width_vec,
+                            LLVMValueRef *out_size,
-                            LLVMValueRef *out_height_vec,
-                            LLVMValueRef *out_depth_vec,
                             LLVMValueRef *row_stride_vec,
                             LLVMValueRef *img_stride_vec)
 {
    const unsigned dims = bld->dims;
    LLVMValueRef ilevel_vec;
-   LLVMValueRef size_vec;
-   LLVMTypeRef i32t = LLVMInt32Type();
 
    ilevel_vec = lp_build_broadcast_scalar(&bld->int_size_bld, ilevel);
 
    /*
     * Compute width, height, depth at mipmap level 'ilevel'
     */
-   size_vec = lp_build_minify(&bld->int_size_bld, bld->int_size, ilevel_vec);
+   *out_size = lp_build_minify(&bld->int_size_bld, bld->int_size, ilevel_vec);
 
-   *out_width_vec = lp_build_extract_broadcast(bld->builder,
-                                               bld->int_size_type,
-                                               bld->int_coord_type,
-                                               size_vec,
-                                               LLVMConstInt(i32t, 0, 0));
    if (dims >= 2) {
-
-      *out_height_vec = lp_build_extract_broadcast(bld->builder,
-                                                   bld->int_size_type,
-                                                   bld->int_coord_type,
-                                                   size_vec,
-                                                   LLVMConstInt(i32t, 1, 0));
-
       *row_stride_vec = lp_build_get_level_stride_vec(bld,
                                                       bld->row_stride_array,
                                                       ilevel);
@@ -668,18 +722,90 @@ lp_build_mipmap_level_sizes(struct lp_build_sample_context *bld,
          *img_stride_vec = lp_build_get_level_stride_vec(bld,
                                                          bld->img_stride_array,
                                                          ilevel);
-         if (dims == 3) {
-            *out_depth_vec = lp_build_extract_broadcast(bld->builder,
-                                                        bld->int_size_type,
-                                                        bld->int_coord_type,
-                                                        size_vec,
-                                                        LLVMConstInt(i32t, 2, 0));
-         }
       }
    }
 }
 
 
+/**
+ * Extract and broadcast texture size.
+ *
+ * @param size_type   type of the texture size vector (either
+ *                    bld->int_size_type or bld->float_size_type)
+ * @param coord_type  type of the texture size vector (either
+ *                    bld->int_coord_type or bld->coord_type)
+ * @param int_size    vector with the integer texture size (width, height,
+ *                    depth)
+ */
+void
+lp_build_extract_image_sizes(struct lp_build_sample_context *bld,
+                             struct lp_type size_type,
+                             struct lp_type coord_type,
+                             LLVMValueRef size,
+                             LLVMValueRef *out_width,
+                             LLVMValueRef *out_height,
+                             LLVMValueRef *out_depth)
+{
+   const unsigned dims = bld->dims;
+   LLVMTypeRef i32t = LLVMInt32Type();
+
+   *out_width = lp_build_extract_broadcast(bld->builder,
+                                           size_type,
+                                           coord_type,
+                                           size,
+                                           LLVMConstInt(i32t, 0, 0));
+   if (dims >= 2) {
+      *out_height = lp_build_extract_broadcast(bld->builder,
+                                               size_type,
+                                               coord_type,
+                                               size,
+                                               LLVMConstInt(i32t, 1, 0));
+      if (dims == 3) {
+         *out_depth = lp_build_extract_broadcast(bld->builder,
+                                                 size_type,
+                                                 coord_type,
+                                                 size,
+                                                 LLVMConstInt(i32t, 2, 0));
+      }
+   }
+}
+
+
+/**
+ * Unnormalize coords.
+ *
+ * @param int_size  vector with the integer texture size (width, height, depth)
+ */
+void
+lp_build_unnormalized_coords(struct lp_build_sample_context *bld,
+                             LLVMValueRef flt_size,
+                             LLVMValueRef *s,
+                             LLVMValueRef *t,
+                             LLVMValueRef *r)
+{
+   const unsigned dims = bld->dims;
+   LLVMValueRef width;
+   LLVMValueRef height;
+   LLVMValueRef depth;
+
+   lp_build_extract_image_sizes(bld,
+                                bld->float_size_type,
+                                bld->coord_type,
+                                flt_size,
+                                &width,
+                                &height,
+                                &depth);
+
+   /* s = s * width, t = t * height */
+   *s = lp_build_mul(&bld->coord_bld, *s, width);
+   if (dims >= 2) {
+      *t = lp_build_mul(&bld->coord_bld, *t, height);
+      if (dims >= 3) {
+         *r = lp_build_mul(&bld->coord_bld, *r, depth);
+      }
+   }
+}
+
 
 /** Helper used by lp_build_cube_lookup() */
 static LLVMValueRef
@@ -798,25 +924,16 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
    rz_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rz, float_bld->zero, "");
 
    {
-      struct lp_build_flow_context *flow_ctx;
       struct lp_build_if_state if_ctx;
+      LLVMValueRef face_s_var;
+      LLVMValueRef face_t_var;
+      LLVMValueRef face_var;
 
-      flow_ctx = lp_build_flow_create(bld->builder);
+      face_s_var = lp_build_alloca(bld->builder, bld->coord_bld.vec_type, "face_s_var");
-      lp_build_flow_scope_begin(flow_ctx);
+      face_t_var = lp_build_alloca(bld->builder, bld->coord_bld.vec_type, "face_t_var");
+      face_var = lp_build_alloca(bld->builder, bld->int_bld.vec_type, "face_var");
 
-      *face_s = bld->coord_bld.undef;
+      lp_build_if(&if_ctx, bld->builder, arx_ge_ary_arz);
-      *face_t = bld->coord_bld.undef;
-      *face = bld->int_bld.undef;
-
-      lp_build_name(*face_s, "face_s");
-      lp_build_name(*face_t, "face_t");
-      lp_build_name(*face, "face");
-
-      lp_build_flow_scope_declare(flow_ctx, face_s);
-      lp_build_flow_scope_declare(flow_ctx, face_t);
-      lp_build_flow_scope_declare(flow_ctx, face);
-
-      lp_build_if(&if_ctx, flow_ctx, bld->builder, arx_ge_ary_arz);
       {
          /* +/- X face */
          LLVMValueRef sign = lp_build_sgn(float_bld, rx);
@@ -826,57 +943,52 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld,
          *face = lp_build_cube_face(bld, rx,
                                     PIPE_TEX_FACE_POS_X,
                                     PIPE_TEX_FACE_NEG_X);
+         LLVMBuildStore(bld->builder, *face_s, face_s_var);
+         LLVMBuildStore(bld->builder, *face_t, face_t_var);
+         LLVMBuildStore(bld->builder, *face, face_var);
       }
       lp_build_else(&if_ctx);
       {
-         struct lp_build_flow_context *flow_ctx2;
          struct lp_build_if_state if_ctx2;
 
-         LLVMValueRef face_s2 = bld->coord_bld.undef;
-         LLVMValueRef face_t2 = bld->coord_bld.undef;
-         LLVMValueRef face2 = bld->int_bld.undef;
-
-         flow_ctx2 = lp_build_flow_create(bld->builder);
-         lp_build_flow_scope_begin(flow_ctx2);
-         lp_build_flow_scope_declare(flow_ctx2, &face_s2);
-         lp_build_flow_scope_declare(flow_ctx2, &face_t2);
-         lp_build_flow_scope_declare(flow_ctx2, &face2);
-
          ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, "");
 
-         lp_build_if(&if_ctx2, flow_ctx2, bld->builder, ary_ge_arx_arz);
+         lp_build_if(&if_ctx2, bld->builder, ary_ge_arx_arz);
          {
             /* +/- Y face */
             LLVMValueRef sign = lp_build_sgn(float_bld, ry);
             LLVMValueRef ima = lp_build_cube_ima(coord_bld, t);
-            face_s2 = lp_build_cube_coord(coord_bld, NULL, -1, s, ima);
+            *face_s = lp_build_cube_coord(coord_bld, NULL, -1, s, ima);
-            face_t2 = lp_build_cube_coord(coord_bld, sign, -1, r, ima);
+            *face_t = lp_build_cube_coord(coord_bld, sign, -1, r, ima);
-            face2 = lp_build_cube_face(bld, ry,
+            *face = lp_build_cube_face(bld, ry,
                                        PIPE_TEX_FACE_POS_Y,
                                        PIPE_TEX_FACE_NEG_Y);
+            LLVMBuildStore(bld->builder, *face_s, face_s_var);
+            LLVMBuildStore(bld->builder, *face_t, face_t_var);
+            LLVMBuildStore(bld->builder, *face, face_var);
          }
          lp_build_else(&if_ctx2);
          {
             /* +/- Z face */
             LLVMValueRef sign = lp_build_sgn(float_bld, rz);
             LLVMValueRef ima = lp_build_cube_ima(coord_bld, r);
-            face_s2 = lp_build_cube_coord(coord_bld, sign, -1, s, ima);
+            *face_s = lp_build_cube_coord(coord_bld, sign, -1, s, ima);
-            face_t2 = lp_build_cube_coord(coord_bld, NULL, +1, t, ima);
+            *face_t = lp_build_cube_coord(coord_bld, NULL, +1, t, ima);
-            face2 = lp_build_cube_face(bld, rz,
+            *face = lp_build_cube_face(bld, rz,
                                        PIPE_TEX_FACE_POS_Z,
                                        PIPE_TEX_FACE_NEG_Z);
+            LLVMBuildStore(bld->builder, *face_s, face_s_var);
+            LLVMBuildStore(bld->builder, *face_t, face_t_var);
+            LLVMBuildStore(bld->builder, *face, face_var);
          }
          lp_build_endif(&if_ctx2);
-         lp_build_flow_scope_end(flow_ctx2);
-         lp_build_flow_destroy(flow_ctx2);
-         *face_s = face_s2;
-         *face_t = face_t2;
-         *face = face2;
       }
 
       lp_build_endif(&if_ctx);
-      lp_build_flow_scope_end(flow_ctx);
+
-      lp_build_flow_destroy(flow_ctx);
+      *face_s = LLVMBuildLoad(bld->builder, face_s_var, "face_s");
+      *face_t = LLVMBuildLoad(bld->builder, face_t_var, "face_t");
+      *face   = LLVMBuildLoad(bld->builder, face_var, "face");
    }
 }
 

src/gallium/auxiliary/gallivm/lp_bld_sample.h

@@ -197,10 +197,6 @@ struct lp_build_sample_context
    struct lp_type coord_type;
    struct lp_build_context coord_bld;
 
-   /** Unsigned integer coordinates */
-   struct lp_type uint_coord_type;
-   struct lp_build_context uint_coord_bld;
-
    /** Signed integer coordinates */
    struct lp_type int_coord_type;
    struct lp_build_context int_coord_bld;
@@ -333,13 +329,29 @@ lp_build_get_const_mipmap_level(struct lp_build_sample_context *bld,
 void
 lp_build_mipmap_level_sizes(struct lp_build_sample_context *bld,
                             LLVMValueRef ilevel,
-                            LLVMValueRef *out_width_vec,
+                            LLVMValueRef *out_size_vec,
-                            LLVMValueRef *out_height_vec,
-                            LLVMValueRef *out_depth_vec,
                             LLVMValueRef *row_stride_vec,
                             LLVMValueRef *img_stride_vec);
 
 
+void
+lp_build_extract_image_sizes(struct lp_build_sample_context *bld,
+                             struct lp_type size_type,
+                             struct lp_type coord_type,
+                             LLVMValueRef size,
+                             LLVMValueRef *out_width,
+                             LLVMValueRef *out_height,
+                             LLVMValueRef *out_depth);
+
+
+void
+lp_build_unnormalized_coords(struct lp_build_sample_context *bld,
+                             LLVMValueRef flt_size,
+                             LLVMValueRef *s,
+                             LLVMValueRef *t,
+                             LLVMValueRef *r);
+
+
 void
 lp_build_cube_lookup(struct lp_build_sample_context *bld,
                      LLVMValueRef s,

src/gallium/auxiliary/gallivm/lp_bld_sample_aos.c

@@ -45,6 +45,7 @@
 #include "lp_bld_const.h"
 #include "lp_bld_conv.h"
 #include "lp_bld_arit.h"
+#include "lp_bld_bitarit.h"
 #include "lp_bld_logic.h"
 #include "lp_bld_swizzle.h"
 #include "lp_bld_pack.h"
@@ -80,11 +81,10 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld,
                                  LLVMValueRef *out_offset,
                                  LLVMValueRef *out_i)
 {
-   struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
    LLVMValueRef length_minus_one;
 
-   length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one);
+   length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
 
    switch(wrap_mode) {
    case PIPE_TEX_WRAP_REPEAT:
@@ -92,7 +92,7 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld,
          coord = LLVMBuildAnd(bld->builder, coord, length_minus_one, "");
       else {
          /* Add a bias to the texcoord to handle negative coords */
-         LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
+         LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
          coord = LLVMBuildAdd(bld->builder, coord, bias, "");
          coord = LLVMBuildURem(bld->builder, coord, length, "");
       }
@@ -113,7 +113,7 @@ lp_build_sample_wrap_nearest_int(struct lp_build_sample_context *bld,
       assert(0);
    }
 
-   lp_build_sample_partial_offset(uint_coord_bld, block_length, coord, stride,
+   lp_build_sample_partial_offset(int_coord_bld, block_length, coord, stride,
                                   out_offset, out_i);
 }
 
@@ -146,7 +146,6 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
                                 LLVMValueRef *i0,
                                 LLVMValueRef *i1)
 {
-   struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
    LLVMValueRef length_minus_one;
    LLVMValueRef lmask, umask, mask;
@@ -188,8 +187,8 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
     * multiplication.
     */
 
-   *i0 = uint_coord_bld->zero;
+   *i0 = int_coord_bld->zero;
-   *i1 = uint_coord_bld->zero;
+   *i1 = int_coord_bld->zero;
 
    length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
 
@@ -200,7 +199,7 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
       }
       else {
          /* Add a bias to the texcoord to handle negative coords */
-         LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
+         LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
          coord0 = LLVMBuildAdd(bld->builder, coord0, bias, "");
          coord0 = LLVMBuildURem(bld->builder, coord0, length, "");
       }
@@ -208,9 +207,9 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
       mask = lp_build_compare(bld->builder, int_coord_bld->type,
                               PIPE_FUNC_NOTEQUAL, coord0, length_minus_one);
 
-      *offset0 = lp_build_mul(uint_coord_bld, coord0, stride);
+      *offset0 = lp_build_mul(int_coord_bld, coord0, stride);
       *offset1 = LLVMBuildAnd(bld->builder,
-                              lp_build_add(uint_coord_bld, *offset0, stride),
+                              lp_build_add(int_coord_bld, *offset0, stride),
                               mask, "");
       break;
 
@@ -225,8 +224,8 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
 
       mask = LLVMBuildAnd(bld->builder, lmask, umask, "");
 
-      *offset0 = lp_build_mul(uint_coord_bld, coord0, stride);
+      *offset0 = lp_build_mul(int_coord_bld, coord0, stride);
-      *offset1 = lp_build_add(uint_coord_bld,
+      *offset1 = lp_build_add(int_coord_bld,
                               *offset0,
                               LLVMBuildAnd(bld->builder, stride, mask, ""));
       break;
@@ -239,8 +238,8 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
    default:
       assert(0);
-      *offset0 = uint_coord_bld->zero;
+      *offset0 = int_coord_bld->zero;
-      *offset1 = uint_coord_bld->zero;
+      *offset1 = int_coord_bld->zero;
       break;
    }
 }
@@ -253,9 +252,7 @@ lp_build_sample_wrap_linear_int(struct lp_build_sample_context *bld,
  */
 static void
 lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
-                              LLVMValueRef width_vec,
+                              LLVMValueRef int_size,
-                              LLVMValueRef height_vec,
-                              LLVMValueRef depth_vec,
                               LLVMValueRef row_stride_vec,
                               LLVMValueRef img_stride_vec,
                               LLVMValueRef data_ptr,
@@ -270,7 +267,8 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
    struct lp_build_context i32, h16, u8n;
    LLVMTypeRef i32_vec_type, h16_vec_type, u8n_vec_type;
    LLVMValueRef i32_c8;
-   LLVMValueRef s_ipart, t_ipart, r_ipart;
+   LLVMValueRef width_vec, height_vec, depth_vec;
+   LLVMValueRef s_ipart, t_ipart = NULL, r_ipart = NULL;
    LLVMValueRef x_stride;
    LLVMValueRef x_offset, offset;
    LLVMValueRef x_subcoord, y_subcoord, z_subcoord;
@@ -283,30 +281,33 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
    h16_vec_type = lp_build_vec_type(h16.type);
    u8n_vec_type = lp_build_vec_type(u8n.type);
 
-   if (bld->static_state->normalized_coords) {
+   lp_build_extract_image_sizes(bld,
-      /* s = s * width, t = t * height */
+                                bld->int_size_type,
-      LLVMTypeRef coord_vec_type = lp_build_vec_type(bld->coord_type);
+                                bld->int_coord_type,
-      LLVMValueRef fp_width = LLVMBuildSIToFP(bld->builder, width_vec,
+                                int_size,
-                                              coord_vec_type, "");
+                                &width_vec,
-      s = lp_build_mul(&bld->coord_bld, s, fp_width);
+                                &height_vec,
-      if (dims >= 2) {
+                                &depth_vec);
-         LLVMValueRef fp_height = LLVMBuildSIToFP(bld->builder, height_vec,
-                                                  coord_vec_type, "");
-         t = lp_build_mul(&bld->coord_bld, t, fp_height);
-         if (dims >= 3) {
-            LLVMValueRef fp_depth = LLVMBuildSIToFP(bld->builder, depth_vec,
-                                                    coord_vec_type, "");
-            r = lp_build_mul(&bld->coord_bld, r, fp_depth);
-         }
-      }
-   }
 
+   if (bld->static_state->normalized_coords) {
+      LLVMValueRef scaled_size;
+      LLVMValueRef flt_size;
+
+      /* scale size by 256 (8 fractional bits) */
+      scaled_size = lp_build_shl_imm(&bld->int_size_bld, int_size, 8);
+
+      flt_size = lp_build_int_to_float(&bld->float_size_bld, scaled_size);
+
+      lp_build_unnormalized_coords(bld, flt_size, &s, &t, &r);
+   }
+   else {
       /* scale coords by 256 (8 fractional bits) */
       s = lp_build_mul_imm(&bld->coord_bld, s, 256);
       if (dims >= 2)
          t = lp_build_mul_imm(&bld->coord_bld, t, 256);
       if (dims >= 3)
          r = lp_build_mul_imm(&bld->coord_bld, r, 256);
+   }
 
    /* convert float to int */
    s = LLVMBuildFPToSI(builder, s, i32_vec_type, "");
@@ -324,7 +325,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
       r_ipart = LLVMBuildAShr(builder, r, i32_c8, "");
 
    /* get pixel, row, image strides */
-   x_stride = lp_build_const_vec(bld->uint_coord_bld.type,
+   x_stride = lp_build_const_vec(bld->int_coord_bld.type,
                                  bld->format_desc->block.bits/8);
 
    /* Do texcoord wrapping, compute texel offset */
@@ -343,7 +344,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
                                        bld->static_state->pot_height,
                                        bld->static_state->wrap_t,
                                        &y_offset, &y_subcoord);
-      offset = lp_build_add(&bld->uint_coord_bld, offset, y_offset);
+      offset = lp_build_add(&bld->int_coord_bld, offset, y_offset);
       if (dims >= 3) {
          LLVMValueRef z_offset;
          lp_build_sample_wrap_nearest_int(bld,
@@ -352,13 +353,13 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
                                           bld->static_state->pot_height,
                                           bld->static_state->wrap_r,
                                           &z_offset, &z_subcoord);
-         offset = lp_build_add(&bld->uint_coord_bld, offset, z_offset);
+         offset = lp_build_add(&bld->int_coord_bld, offset, z_offset);
       }
       else if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
          LLVMValueRef z_offset;
          /* The r coord is the cube face in [0,5] */
-         z_offset = lp_build_mul(&bld->uint_coord_bld, r, img_stride_vec);
+         z_offset = lp_build_mul(&bld->int_coord_bld, r, img_stride_vec);
-         offset = lp_build_add(&bld->uint_coord_bld, offset, z_offset);
+         offset = lp_build_add(&bld->int_coord_bld, offset, z_offset);
       }
    }
 
@@ -417,9 +418,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
  */
 static void
 lp_build_sample_image_linear(struct lp_build_sample_context *bld,
-                             LLVMValueRef width_vec,
+                             LLVMValueRef int_size,
-                             LLVMValueRef height_vec,
-                             LLVMValueRef depth_vec,
                              LLVMValueRef row_stride_vec,
                              LLVMValueRef img_stride_vec,
                              LLVMValueRef data_ptr,
@@ -434,9 +433,10 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
    struct lp_build_context i32, h16, u8n;
    LLVMTypeRef i32_vec_type, h16_vec_type, u8n_vec_type;
    LLVMValueRef i32_c8, i32_c128, i32_c255;
+   LLVMValueRef width_vec, height_vec, depth_vec;
    LLVMValueRef s_ipart, s_fpart, s_fpart_lo, s_fpart_hi;
-   LLVMValueRef t_ipart, t_fpart, t_fpart_lo, t_fpart_hi;
+   LLVMValueRef t_ipart = NULL, t_fpart = NULL, t_fpart_lo = NULL, t_fpart_hi = NULL;
-   LLVMValueRef r_ipart, r_fpart, r_fpart_lo, r_fpart_hi;
+   LLVMValueRef r_ipart = NULL, r_fpart = NULL, r_fpart_lo = NULL, r_fpart_hi = NULL;
    LLVMValueRef x_stride, y_stride, z_stride;
    LLVMValueRef x_offset0, x_offset1;
    LLVMValueRef y_offset0, y_offset1;
@@ -458,30 +458,33 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
    h16_vec_type = lp_build_vec_type(h16.type);
    u8n_vec_type = lp_build_vec_type(u8n.type);
 
-   if (bld->static_state->normalized_coords) {
+   lp_build_extract_image_sizes(bld,
-      /* s = s * width, t = t * height */
+                                bld->int_size_type,
-      LLVMTypeRef coord_vec_type = lp_build_vec_type(bld->coord_type);
+                                bld->int_coord_type,
-      LLVMValueRef fp_width = LLVMBuildSIToFP(bld->builder, width_vec,
+                                int_size,
-                                              coord_vec_type, "");
+                                &width_vec,
-      s = lp_build_mul(&bld->coord_bld, s, fp_width);
+                                &height_vec,
-      if (dims >= 2) {
+                                &depth_vec);
-         LLVMValueRef fp_height = LLVMBuildSIToFP(bld->builder, height_vec,
-                                                  coord_vec_type, "");
-         t = lp_build_mul(&bld->coord_bld, t, fp_height);
-      }
-      if (dims >= 3) {
-         LLVMValueRef fp_depth = LLVMBuildSIToFP(bld->builder, depth_vec,
-                                                 coord_vec_type, "");
-         r = lp_build_mul(&bld->coord_bld, r, fp_depth);
-      }
-   }
 
+   if (bld->static_state->normalized_coords) {
+      LLVMValueRef scaled_size;
+      LLVMValueRef flt_size;
+
+      /* scale size by 256 (8 fractional bits) */
+      scaled_size = lp_build_shl_imm(&bld->int_size_bld, int_size, 8);
+
+      flt_size = lp_build_int_to_float(&bld->float_size_bld, scaled_size);
+
+      lp_build_unnormalized_coords(bld, flt_size, &s, &t, &r);
+   }
+   else {
       /* scale coords by 256 (8 fractional bits) */
       s = lp_build_mul_imm(&bld->coord_bld, s, 256);
       if (dims >= 2)
          t = lp_build_mul_imm(&bld->coord_bld, t, 256);
       if (dims >= 3)
          r = lp_build_mul_imm(&bld->coord_bld, r, 256);
+   }
 
    /* convert float to int */
    s = LLVMBuildFPToSI(builder, s, i32_vec_type, "");
@@ -517,7 +520,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
       r_fpart = LLVMBuildAnd(builder, r, i32_c255, "");
 
    /* get pixel, row and image strides */
-   x_stride = lp_build_const_vec(bld->uint_coord_bld.type,
+   x_stride = lp_build_const_vec(bld->int_coord_bld.type,
                                  bld->format_desc->block.bits/8);
    y_stride = row_stride_vec;
    z_stride = img_stride_vec;
@@ -548,9 +551,9 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
 
       for (z = 0; z < 2; z++) {
          for (x = 0; x < 2; x++) {
-            offset[z][0][x] = lp_build_add(&bld->uint_coord_bld,
+            offset[z][0][x] = lp_build_add(&bld->int_coord_bld,
                                            offset[z][0][x], y_offset0);
-            offset[z][1][x] = lp_build_add(&bld->uint_coord_bld,
+            offset[z][1][x] = lp_build_add(&bld->int_coord_bld,
                                            offset[z][1][x], y_offset1);
          }
       }
@@ -566,20 +569,20 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
                                       &z_subcoord[0], &z_subcoord[1]);
       for (y = 0; y < 2; y++) {
          for (x = 0; x < 2; x++) {
-            offset[0][y][x] = lp_build_add(&bld->uint_coord_bld,
+            offset[0][y][x] = lp_build_add(&bld->int_coord_bld,
                                            offset[0][y][x], z_offset0);
-            offset[1][y][x] = lp_build_add(&bld->uint_coord_bld,
+            offset[1][y][x] = lp_build_add(&bld->int_coord_bld,
                                            offset[1][y][x], z_offset1);
          }
       }
    }
    else if (bld->static_state->target == PIPE_TEXTURE_CUBE) {
       LLVMValueRef z_offset;
-      z_offset = lp_build_mul(&bld->uint_coord_bld, r, img_stride_vec);
+      z_offset = lp_build_mul(&bld->int_coord_bld, r, img_stride_vec);
       for (y = 0; y < 2; y++) {
          for (x = 0; x < 2; x++) {
             /* The r coord is the cube face in [0,5] */
-            offset[0][y][x] = lp_build_add(&bld->uint_coord_bld,
+            offset[0][y][x] = lp_build_add(&bld->int_coord_bld,
                                            offset[0][y][x], z_offset);
          }
       }
@@ -788,12 +791,8 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
                        LLVMValueRef colors_hi_var)
 {
    LLVMBuilderRef builder = bld->builder;
-   LLVMValueRef width0_vec;
+   LLVMValueRef size0;
-   LLVMValueRef width1_vec;
+   LLVMValueRef size1;
-   LLVMValueRef height0_vec;
-   LLVMValueRef height1_vec;
-   LLVMValueRef depth0_vec;
-   LLVMValueRef depth1_vec;
    LLVMValueRef row_stride0_vec;
    LLVMValueRef row_stride1_vec;
    LLVMValueRef img_stride0_vec;
@@ -806,12 +805,12 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
 
    /* sample the first mipmap level */
    lp_build_mipmap_level_sizes(bld, ilevel0,
-                               &width0_vec, &height0_vec, &depth0_vec,
+                               &size0,
                                &row_stride0_vec, &img_stride0_vec);
    data_ptr0 = lp_build_get_mipmap_level(bld, ilevel0);
    if (img_filter == PIPE_TEX_FILTER_NEAREST) {
       lp_build_sample_image_nearest(bld,
-                                    width0_vec, height0_vec, depth0_vec,
+                                    size0,
                                     row_stride0_vec, img_stride0_vec,
                                     data_ptr0, s, t, r,
                                     &colors0_lo, &colors0_hi);
@@ -819,7 +818,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
    else {
       assert(img_filter == PIPE_TEX_FILTER_LINEAR);
       lp_build_sample_image_linear(bld,
-                                   width0_vec, height0_vec, depth0_vec,
+                                   size0,
                                    row_stride0_vec, img_stride0_vec,
                                    data_ptr0, s, t, r,
                                    &colors0_lo, &colors0_hi);
@@ -832,12 +831,9 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
    if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
       LLVMValueRef h16_scale = LLVMConstReal(LLVMFloatType(), 256.0);
       LLVMTypeRef i32_type = LLVMIntType(32);
-      struct lp_build_flow_context *flow_ctx;
       struct lp_build_if_state if_ctx;
       LLVMValueRef need_lerp;
 
-      flow_ctx = lp_build_flow_create(builder);
-
       lod_fpart = LLVMBuildFMul(builder, lod_fpart, h16_scale, "");
       lod_fpart = LLVMBuildFPToSI(builder, lod_fpart, i32_type, "lod_fpart.fixed16");
 
@@ -846,7 +842,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
                                 lod_fpart, LLVMConstNull(i32_type),
                                 "need_lerp");
 
-      lp_build_if(&if_ctx, flow_ctx, builder, need_lerp);
+      lp_build_if(&if_ctx, builder, need_lerp);
       {
          struct lp_build_context h16_bld;
 
@@ -854,19 +850,19 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
 
          /* sample the second mipmap level */
          lp_build_mipmap_level_sizes(bld, ilevel1,
-                                     &width1_vec, &height1_vec, &depth1_vec,
+                                     &size1,
                                      &row_stride1_vec, &img_stride1_vec);
          data_ptr1 = lp_build_get_mipmap_level(bld, ilevel1);
          if (img_filter == PIPE_TEX_FILTER_NEAREST) {
             lp_build_sample_image_nearest(bld,
-                                          width1_vec, height1_vec, depth1_vec,
+                                          size1,
                                           row_stride1_vec, img_stride1_vec,
                                           data_ptr1, s, t, r,
                                           &colors1_lo, &colors1_hi);
          }
          else {
             lp_build_sample_image_linear(bld,
-                                         width1_vec, height1_vec, depth1_vec,
+                                         size1,
                                          row_stride1_vec, img_stride1_vec,
                                          data_ptr1, s, t, r,
                                          &colors1_lo, &colors1_hi);
@@ -877,6 +873,26 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
          lod_fpart = LLVMBuildTrunc(builder, lod_fpart, h16_bld.elem_type, "");
          lod_fpart = lp_build_broadcast_scalar(&h16_bld, lod_fpart);
 
+#if HAVE_LLVM == 0x208
+         /* This is a work-around for a bug in LLVM 2.8.
+          * Evidently, something goes wrong in the construction of the
+          * lod_fpart short[8] vector.  Adding this no-effect shuffle seems
+          * to force the vector to be properly constructed.
+          * Tested with mesa-demos/src/tests/mipmap_limits.c (press t, f).
+          */
+         {
+            LLVMValueRef shuffles[8], shuffle;
+            int i;
+            assert(h16_bld.type.length <= Elements(shuffles));
+            for (i = 0; i < h16_bld.type.length; i++)
+               shuffles[i] = lp_build_const_int32(2 * (i & 1));
+            shuffle = LLVMConstVector(shuffles, h16_bld.type.length);
+            lod_fpart = LLVMBuildShuffleVector(builder,
+                                               lod_fpart, lod_fpart,
+                                               shuffle, "");
+         }
+#endif
+
          colors0_lo = lp_build_lerp(&h16_bld, lod_fpart,
                                     colors0_lo, colors1_lo);
          colors0_hi = lp_build_lerp(&h16_bld, lod_fpart,
@@ -886,8 +902,6 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
          LLVMBuildStore(builder, colors0_hi, colors_hi_var);
       }
       lp_build_endif(&if_ctx);
-
-      lp_build_flow_destroy(flow_ctx);
    }
 }
 
@@ -946,12 +960,12 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
       r = lp_build_broadcast_scalar(&bld->int_coord_bld, face); /* vec */
 
       /* recompute ddx, ddy using the new (s,t) face texcoords */
-      face_ddx[0] = lp_build_ddx(&bld->coord_bld, s);
+      face_ddx[0] = lp_build_scalar_ddx(&bld->coord_bld, s);
-      face_ddx[1] = lp_build_ddx(&bld->coord_bld, t);
+      face_ddx[1] = lp_build_scalar_ddx(&bld->coord_bld, t);
       face_ddx[2] = NULL;
       face_ddx[3] = NULL;
-      face_ddy[0] = lp_build_ddy(&bld->coord_bld, s);
+      face_ddy[0] = lp_build_scalar_ddy(&bld->coord_bld, s);
-      face_ddy[1] = lp_build_ddy(&bld->coord_bld, t);
+      face_ddy[1] = lp_build_scalar_ddy(&bld->coord_bld, t);
       face_ddy[2] = NULL;
       face_ddy[3] = NULL;
       ddx = face_ddx;
@@ -1027,17 +1041,14 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
       /* Emit conditional to choose min image filter or mag image filter
        * depending on the lod being > 0 or <= 0, respectively.
        */
-      struct lp_build_flow_context *flow_ctx;
       struct lp_build_if_state if_ctx;
       LLVMValueRef minify;
 
-      flow_ctx = lp_build_flow_create(builder);
-
       /* minify = lod >= 0.0 */
       minify = LLVMBuildICmp(builder, LLVMIntSGE,
                              lod_ipart, int_bld->zero, "");
 
-      lp_build_if(&if_ctx, flow_ctx, builder, minify);
+      lp_build_if(&if_ctx, builder, minify);
       {
          /* Use the minification filter */
          lp_build_sample_mipmap(bld,
@@ -1056,8 +1067,6 @@ lp_build_sample_aos(struct lp_build_sample_context *bld,
                                 packed_lo, packed_hi);
       }
       lp_build_endif(&if_ctx);
-
-      lp_build_flow_destroy(flow_ctx);
    }
 
    /*

src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c

@@ -131,7 +131,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
    }
 
    /* convert x,y,z coords to linear offset from start of texture, in bytes */
-   lp_build_sample_offset(&bld->uint_coord_bld,
+   lp_build_sample_offset(&bld->int_coord_bld,
                           bld->format_desc,
                           x, y, z, y_stride, z_stride,
                           &offset, &i, &j);
@@ -145,7 +145,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld,
        * coords which are out of bounds to become zero.  Zero's guaranteed
        * to be inside the texture image.
        */
-      offset = lp_build_andnot(&bld->uint_coord_bld, offset, use_border);
+      offset = lp_build_andnot(&bld->int_coord_bld, offset, use_border);
    }
 
    lp_build_fetch_rgba_soa(bld->builder,
@@ -202,11 +202,7 @@ lp_build_coord_mirror(struct lp_build_sample_context *bld,
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
    LLVMValueRef fract, flr, isOdd;
 
-   /* fract = coord - floor(coord) */
+   lp_build_ifloor_fract(coord_bld, coord, &flr, &fract);
-   fract = lp_build_sub(coord_bld, coord, lp_build_floor(coord_bld, coord));
-
-   /* flr = ifloor(coord); */
-   flr = lp_build_ifloor(coord_bld, coord);
 
    /* isOdd = flr & 1 */
    isOdd = LLVMBuildAnd(bld->builder, flr, int_coord_bld->one, "");
@@ -234,6 +230,7 @@ static void
 lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
                             LLVMValueRef coord,
                             LLVMValueRef length,
+                            LLVMValueRef length_f,
                             boolean is_pot,
                             unsigned wrap_mode,
                             LLVMValueRef *x0_out,
@@ -242,10 +239,8 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
 {
    struct lp_build_context *coord_bld = &bld->coord_bld;
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
-   struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
    LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5);
-   LLVMValueRef length_f = lp_build_int_to_float(coord_bld, length);
+   LLVMValueRef length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
-   LLVMValueRef length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one);
    LLVMValueRef coord0, coord1, weight;
 
    switch(wrap_mode) {
@@ -255,19 +250,23 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
       coord = lp_build_sub(coord_bld, coord, half);
       /* convert to int, compute lerp weight */
       lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
-      coord1 = lp_build_add(uint_coord_bld, coord0, uint_coord_bld->one);
       /* repeat wrap */
       if (is_pot) {
+         coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
          coord0 = LLVMBuildAnd(bld->builder, coord0, length_minus_one, "");
          coord1 = LLVMBuildAnd(bld->builder, coord1, length_minus_one, "");
       }
       else {
          /* Add a bias to the texcoord to handle negative coords */
-         LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
+         LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
+         LLVMValueRef mask;
          coord0 = LLVMBuildAdd(bld->builder, coord0, bias, "");
-         coord1 = LLVMBuildAdd(bld->builder, coord1, bias, "");
          coord0 = LLVMBuildURem(bld->builder, coord0, length, "");
-         coord1 = LLVMBuildURem(bld->builder, coord1, length, "");
+         mask = lp_build_compare(bld->builder, int_coord_bld->type,
+                                 PIPE_FUNC_NOTEQUAL, coord0, length_minus_one);
+         coord1 = LLVMBuildAnd(bld->builder,
+                              lp_build_add(int_coord_bld, coord0, int_coord_bld->one),
+                              mask, "");
       }
       break;
 
@@ -288,41 +287,39 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
       break;
 
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      {
+         struct lp_build_context abs_coord_bld = bld->coord_bld;
+         abs_coord_bld.type.sign = FALSE;
+
          if (bld->static_state->normalized_coords) {
-         /* clamp to [0,1] */
+            /* mul by tex size */
-         coord = lp_build_clamp(coord_bld, coord, coord_bld->zero, coord_bld->one);
-         /* mul by tex size and subtract 0.5 */
             coord = lp_build_mul(coord_bld, coord, length_f);
+         }
+         /* clamp to length max */
+         coord = lp_build_min(coord_bld, coord, length_f);
+         /* subtract 0.5 */
          coord = lp_build_sub(coord_bld, coord, half);
-      }
+         /* clamp to [0, length - 0.5] */
-      else {
+         coord = lp_build_max(coord_bld, coord, coord_bld->zero);
-         LLVMValueRef min, max;
-         /* clamp to [0.5, length - 0.5] */
-         min = half;
-         max = lp_build_sub(coord_bld, length_f, min);
-         coord = lp_build_clamp(coord_bld, coord, min, max);
-      }
          /* convert to int, compute lerp weight */
-      lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
+         lp_build_ifloor_fract(&abs_coord_bld, coord, &coord0, &weight);
          coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
-      /* coord0 = max(coord0, 0) */
-      coord0 = lp_build_max(int_coord_bld, coord0, int_coord_bld->zero);
          /* coord1 = min(coord1, length-1) */
          coord1 = lp_build_min(int_coord_bld, coord1, length_minus_one);
          break;
+      }
 
    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
       {
-         LLVMValueRef min, max;
+         LLVMValueRef min;
          if (bld->static_state->normalized_coords) {
             /* scale coord to length */
             coord = lp_build_mul(coord_bld, coord, length_f);
          }
-         /* clamp to [-0.5, length + 0.5] */
+         /* was: clamp to [-0.5, length + 0.5], then sub 0.5 */
-         min = lp_build_const_vec(coord_bld->type, -0.5F);
-         max = lp_build_sub(coord_bld, length_f, min);
-         coord = lp_build_clamp(coord_bld, coord, min, max);
          coord = lp_build_sub(coord_bld, coord, half);
+         min = lp_build_const_vec(coord_bld->type, -1.0F);
+         coord = lp_build_clamp(coord_bld, coord, min, length_f);
          /* convert to int, compute lerp weight */
          lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
          coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
@@ -368,7 +365,8 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
       {
          LLVMValueRef min, max;
-
+         struct lp_build_context abs_coord_bld = bld->coord_bld;
+         abs_coord_bld.type.sign = FALSE;
          coord = lp_build_abs(coord_bld, coord);
 
          if (bld->static_state->normalized_coords) {
@@ -384,15 +382,13 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
          coord = lp_build_sub(coord_bld, coord, half);
 
          /* convert to int, compute lerp weight */
-         lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
+         lp_build_ifloor_fract(&abs_coord_bld, coord, &coord0, &weight);
          coord1 = lp_build_add(int_coord_bld, coord0, int_coord_bld->one);
       }
       break;
 
    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
       {
-         LLVMValueRef min, max;
-
          coord = lp_build_abs(coord_bld, coord);
 
          if (bld->static_state->normalized_coords) {
@@ -400,12 +396,10 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld,
             coord = lp_build_mul(coord_bld, coord, length_f);
          }
 
-         /* clamp to [-0.5, length + 0.5] */
+         /* was: clamp to [-0.5, length + 0.5] then sub 0.5 */
-         min = lp_build_negate(coord_bld, half);
+         /* skip -0.5 clamp (always positive), do sub first */
-         max = lp_build_sub(coord_bld, length_f, min);
-         coord = lp_build_clamp(coord_bld, coord, min, max);
-
          coord = lp_build_sub(coord_bld, coord, half);
+         coord = lp_build_min(coord_bld, coord, length_f);
 
          /* convert to int, compute lerp weight */
          lp_build_ifloor_fract(coord_bld, coord, &coord0, &weight);
@@ -437,14 +431,13 @@ static LLVMValueRef
 lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
                              LLVMValueRef coord,
                              LLVMValueRef length,
+                             LLVMValueRef length_f,
                              boolean is_pot,
                              unsigned wrap_mode)
 {
    struct lp_build_context *coord_bld = &bld->coord_bld;
    struct lp_build_context *int_coord_bld = &bld->int_coord_bld;
-   struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld;
+   LLVMValueRef length_minus_one = lp_build_sub(int_coord_bld, length, int_coord_bld->one);
-   LLVMValueRef length_f = lp_build_int_to_float(coord_bld, length);
-   LLVMValueRef length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one);
    LLVMValueRef icoord;
    
    switch(wrap_mode) {
@@ -455,7 +448,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
          icoord = LLVMBuildAnd(bld->builder, icoord, length_minus_one, "");
       else {
          /* Add a bias to the texcoord to handle negative coords */
-         LLVMValueRef bias = lp_build_mul_imm(uint_coord_bld, length, 1024);
+         LLVMValueRef bias = lp_build_mul_imm(int_coord_bld, length, 1024);
          icoord = LLVMBuildAdd(bld->builder, icoord, bias, "");
          icoord = LLVMBuildURem(bld->builder, icoord, length, "");
       }
@@ -469,7 +462,8 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
       }
 
       /* floor */
-      icoord = lp_build_ifloor(coord_bld, coord);
+      /* use itrunc instead since we clamp to 0 anyway */
+      icoord = lp_build_itrunc(coord_bld, coord);
 
       /* clamp to [0, length - 1]. */
       icoord = lp_build_clamp(int_coord_bld, icoord, int_coord_bld->zero,
@@ -503,7 +497,8 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
       assert(bld->static_state->normalized_coords);
       coord = lp_build_mul(coord_bld, coord, length_f);
 
-      icoord = lp_build_ifloor(coord_bld, coord);
+      /* itrunc == ifloor here */
+      icoord = lp_build_itrunc(coord_bld, coord);
 
       /* clamp to [0, length - 1] */
       icoord = lp_build_min(int_coord_bld, icoord, length_minus_one);
@@ -518,7 +513,8 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
          coord = lp_build_mul(coord_bld, coord, length_f);
       }
 
-      icoord = lp_build_ifloor(coord_bld, coord);
+      /* itrunc == ifloor here */
+      icoord = lp_build_itrunc(coord_bld, coord);
 
       /* clamp to [0, length - 1] */
       icoord = lp_build_min(int_coord_bld, icoord, length_minus_one);
@@ -532,7 +528,8 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
          coord = lp_build_mul(coord_bld, coord, length_f);
       }
 
-      icoord = lp_build_ifloor(coord_bld, coord);
+      /* itrunc == ifloor here */
+      icoord = lp_build_itrunc(coord_bld, coord);
 
       /* clamp to [0, length] */
       icoord = lp_build_min(int_coord_bld, icoord, length);
@@ -554,9 +551,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld,
 static void
 lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
                               unsigned unit,
-                              LLVMValueRef width_vec,
+                              LLVMValueRef size,
-                              LLVMValueRef height_vec,
-                              LLVMValueRef depth_vec,
                               LLVMValueRef row_stride_vec,
                               LLVMValueRef img_stride_vec,
                               LLVMValueRef data_ptr,
@@ -566,24 +561,45 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
                               LLVMValueRef colors_out[4])
 {
    const unsigned dims = bld->dims;
+   LLVMValueRef width_vec;
+   LLVMValueRef height_vec;
+   LLVMValueRef depth_vec;
+   LLVMValueRef flt_size;
+   LLVMValueRef flt_width_vec;
+   LLVMValueRef flt_height_vec;
+   LLVMValueRef flt_depth_vec;
    LLVMValueRef x, y, z;
 
+   lp_build_extract_image_sizes(bld,
+                                bld->int_size_type,
+                                bld->int_coord_type,
+                                size,
+                                &width_vec, &height_vec, &depth_vec);
+
+   flt_size = lp_build_int_to_float(&bld->float_size_bld, size);
+
+   lp_build_extract_image_sizes(bld,
+                                bld->float_size_type,
+                                bld->coord_type,
+                                flt_size,
+                                &flt_width_vec, &flt_height_vec, &flt_depth_vec);
+
    /*
     * Compute integer texcoords.
     */
-   x = lp_build_sample_wrap_nearest(bld, s, width_vec,
+   x = lp_build_sample_wrap_nearest(bld, s, width_vec, flt_width_vec,
                                     bld->static_state->pot_width,
                                     bld->static_state->wrap_s);
    lp_build_name(x, "tex.x.wrapped");
 
    if (dims >= 2) {
-      y = lp_build_sample_wrap_nearest(bld, t, height_vec,
+      y = lp_build_sample_wrap_nearest(bld, t, height_vec, flt_height_vec,
                                        bld->static_state->pot_height,
                                        bld->static_state->wrap_t);
       lp_build_name(y, "tex.y.wrapped");
 
       if (dims == 3) {
-         z = lp_build_sample_wrap_nearest(bld, r, depth_vec,
+         z = lp_build_sample_wrap_nearest(bld, r, depth_vec, flt_depth_vec,
                                           bld->static_state->pot_depth,
                                           bld->static_state->wrap_r);
          lp_build_name(z, "tex.z.wrapped");
@@ -617,9 +633,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld,
 static void
 lp_build_sample_image_linear(struct lp_build_sample_context *bld,
                              unsigned unit,
-                             LLVMValueRef width_vec,
+                             LLVMValueRef size,
-                             LLVMValueRef height_vec,
-                             LLVMValueRef depth_vec,
                              LLVMValueRef row_stride_vec,
                              LLVMValueRef img_stride_vec,
                              LLVMValueRef data_ptr,
@@ -629,15 +643,36 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
                              LLVMValueRef colors_out[4])
 {
    const unsigned dims = bld->dims;
+   LLVMValueRef width_vec;
+   LLVMValueRef height_vec;
+   LLVMValueRef depth_vec;
+   LLVMValueRef flt_size;
+   LLVMValueRef flt_width_vec;
+   LLVMValueRef flt_height_vec;
+   LLVMValueRef flt_depth_vec;
    LLVMValueRef x0, y0, z0, x1, y1, z1;
    LLVMValueRef s_fpart, t_fpart, r_fpart;
    LLVMValueRef neighbors[2][2][4];
    int chan;
 
+   lp_build_extract_image_sizes(bld,
+                                bld->int_size_type,
+                                bld->int_coord_type,
+                                size,
+                                &width_vec, &height_vec, &depth_vec);
+
+   flt_size = lp_build_int_to_float(&bld->float_size_bld, size);
+
+   lp_build_extract_image_sizes(bld,
+                                bld->float_size_type,
+                                bld->coord_type,
+                                flt_size,
+                                &flt_width_vec, &flt_height_vec, &flt_depth_vec);
+
    /*
     * Compute integer texcoords.
     */
-   lp_build_sample_wrap_linear(bld, s, width_vec,
+   lp_build_sample_wrap_linear(bld, s, width_vec, flt_width_vec,
                                bld->static_state->pot_width,
                                bld->static_state->wrap_s,
                                &x0, &x1, &s_fpart);
@@ -645,7 +680,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
    lp_build_name(x1, "tex.x1.wrapped");
 
    if (dims >= 2) {
-      lp_build_sample_wrap_linear(bld, t, height_vec,
+      lp_build_sample_wrap_linear(bld, t, height_vec, flt_height_vec,
                                   bld->static_state->pot_height,
                                   bld->static_state->wrap_t,
                                   &y0, &y1, &t_fpart);
@@ -653,7 +688,7 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld,
       lp_build_name(y1, "tex.y1.wrapped");
 
       if (dims == 3) {
-         lp_build_sample_wrap_linear(bld, r, depth_vec,
+         lp_build_sample_wrap_linear(bld, r, depth_vec, flt_depth_vec,
                                      bld->static_state->pot_depth,
                                      bld->static_state->wrap_r,
                                      &z0, &z1, &r_fpart);
@@ -796,12 +831,8 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
                        LLVMValueRef *colors_out)
 {
    LLVMBuilderRef builder = bld->builder;
-   LLVMValueRef width0_vec;
+   LLVMValueRef size0;
-   LLVMValueRef width1_vec;
+   LLVMValueRef size1;
-   LLVMValueRef height0_vec;
-   LLVMValueRef height1_vec;
-   LLVMValueRef depth0_vec;
-   LLVMValueRef depth1_vec;
    LLVMValueRef row_stride0_vec;
    LLVMValueRef row_stride1_vec;
    LLVMValueRef img_stride0_vec;
@@ -813,12 +844,12 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
 
    /* sample the first mipmap level */
    lp_build_mipmap_level_sizes(bld, ilevel0,
-                               &width0_vec, &height0_vec, &depth0_vec,
+                               &size0,
                                &row_stride0_vec, &img_stride0_vec);
    data_ptr0 = lp_build_get_mipmap_level(bld, ilevel0);
    if (img_filter == PIPE_TEX_FILTER_NEAREST) {
       lp_build_sample_image_nearest(bld, unit,
-                                    width0_vec, height0_vec, depth0_vec,
+                                    size0,
                                     row_stride0_vec, img_stride0_vec,
                                     data_ptr0, s, t, r,
                                     colors0);
@@ -826,7 +857,7 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
    else {
       assert(img_filter == PIPE_TEX_FILTER_LINEAR);
       lp_build_sample_image_linear(bld, unit,
-                                   width0_vec, height0_vec, depth0_vec,
+                                   size0,
                                    row_stride0_vec, img_stride0_vec,
                                    data_ptr0, s, t, r,
                                    colors0);
@@ -838,35 +869,32 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
    }
 
    if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
-      struct lp_build_flow_context *flow_ctx;
       struct lp_build_if_state if_ctx;
       LLVMValueRef need_lerp;
 
-      flow_ctx = lp_build_flow_create(builder);
-
       /* need_lerp = lod_fpart > 0 */
       need_lerp = LLVMBuildFCmp(builder, LLVMRealUGT,
                                 lod_fpart,
                                 bld->float_bld.zero,
                                 "need_lerp");
 
-      lp_build_if(&if_ctx, flow_ctx, builder, need_lerp);
+      lp_build_if(&if_ctx, builder, need_lerp);
       {
          /* sample the second mipmap level */
          lp_build_mipmap_level_sizes(bld, ilevel1,
-                                     &width1_vec, &height1_vec, &depth1_vec,
+                                     &size1,
                                      &row_stride1_vec, &img_stride1_vec);
          data_ptr1 = lp_build_get_mipmap_level(bld, ilevel1);
          if (img_filter == PIPE_TEX_FILTER_NEAREST) {
             lp_build_sample_image_nearest(bld, unit,
-                                          width1_vec, height1_vec, depth1_vec,
+                                          size1,
                                           row_stride1_vec, img_stride1_vec,
                                           data_ptr1, s, t, r,
                                           colors1);
          }
          else {
             lp_build_sample_image_linear(bld, unit,
-                                         width1_vec, height1_vec, depth1_vec,
+                                         size1,
                                          row_stride1_vec, img_stride1_vec,
                                          data_ptr1, s, t, r,
                                          colors1);
@@ -883,8 +911,6 @@ lp_build_sample_mipmap(struct lp_build_sample_context *bld,
          }
       }
       lp_build_endif(&if_ctx);
-
-      lp_build_flow_destroy(flow_ctx);
    }
 }
 
@@ -937,12 +963,12 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
       r = lp_build_broadcast_scalar(&bld->int_coord_bld, face); /* vec */
 
       /* recompute ddx, ddy using the new (s,t) face texcoords */
-      face_ddx[0] = lp_build_ddx(&bld->coord_bld, s);
+      face_ddx[0] = lp_build_scalar_ddx(&bld->coord_bld, s);
-      face_ddx[1] = lp_build_ddx(&bld->coord_bld, t);
+      face_ddx[1] = lp_build_scalar_ddx(&bld->coord_bld, t);
       face_ddx[2] = NULL;
       face_ddx[3] = NULL;
-      face_ddy[0] = lp_build_ddy(&bld->coord_bld, s);
+      face_ddy[0] = lp_build_scalar_ddy(&bld->coord_bld, s);
-      face_ddy[1] = lp_build_ddy(&bld->coord_bld, t);
+      face_ddy[1] = lp_build_scalar_ddy(&bld->coord_bld, t);
       face_ddy[2] = NULL;
       face_ddy[3] = NULL;
       ddx = face_ddx;
@@ -1020,17 +1046,14 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
       /* Emit conditional to choose min image filter or mag image filter
        * depending on the lod being > 0 or <= 0, respectively.
        */
-      struct lp_build_flow_context *flow_ctx;
       struct lp_build_if_state if_ctx;
       LLVMValueRef minify;
 
-      flow_ctx = lp_build_flow_create(builder);
-
       /* minify = lod >= 0.0 */
       minify = LLVMBuildICmp(builder, LLVMIntSGE,
                              lod_ipart, int_bld->zero, "");
 
-      lp_build_if(&if_ctx, flow_ctx, builder, minify);
+      lp_build_if(&if_ctx, builder, minify);
       {
          /* Use the minification filter */
          lp_build_sample_mipmap(bld, unit,
@@ -1049,8 +1072,6 @@ lp_build_sample_general(struct lp_build_sample_context *bld,
                                 texels);
       }
       lp_build_endif(&if_ctx);
-
-      lp_build_flow_destroy(flow_ctx);
    }
 
    for (chan = 0; chan < 4; ++chan) {
@@ -1166,7 +1187,6 @@ lp_build_sample_soa(LLVMBuilderRef builder,
    bld.float_type = lp_type_float(32);
    bld.int_type = lp_type_int(32);
    bld.coord_type = type;
-   bld.uint_coord_type = lp_uint_type(type);
    bld.int_coord_type = lp_int_type(type);
    bld.float_size_type = lp_type_float(32);
    bld.float_size_type.length = dims > 1 ? 4 : 1;
@@ -1179,7 +1199,6 @@ lp_build_sample_soa(LLVMBuilderRef builder,
    lp_build_context_init(&bld.float_vec_bld, builder, float_vec_type);
    lp_build_context_init(&bld.int_bld, builder, bld.int_type);
    lp_build_context_init(&bld.coord_bld, builder, bld.coord_type);
-   lp_build_context_init(&bld.uint_coord_bld, builder, bld.uint_coord_type);
    lp_build_context_init(&bld.int_coord_bld, builder, bld.int_coord_type);
    lp_build_context_init(&bld.int_size_bld, builder, bld.int_size_type);
    lp_build_context_init(&bld.float_size_bld, builder, bld.float_size_type);

src/gallium/auxiliary/gallivm/lp_bld_tgsi.h

@@ -36,6 +36,9 @@
 #define LP_BLD_TGSI_H
 
 #include "gallivm/lp_bld.h"
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
+#include "tgsi/tgsi_scan.h"
 
 
 struct tgsi_token;
@@ -54,6 +57,75 @@ enum lp_build_tex_modifier {
 };
 
 
+/**
+ * Describe a channel of a register.
+ *
+ * The value can be a:
+ * - immediate value (i.e. derived from a IMM register)
+ * - CONST[n].x/y/z/w
+ * - IN[n].x/y/z/w
+ * - undetermined (when .file == TGSI_FILE_NULL)
+ *
+ * This is one of the analysis results, and is used to described
+ * the output color in terms of inputs.
+ */
+struct lp_tgsi_channel_info
+{
+   unsigned file:4; /* TGSI_FILE_* */
+   unsigned swizzle:3; /* PIPE_SWIZZLE_x */
+   union {
+      uint32_t index;
+      float value; /* for TGSI_FILE_IMMEDIATE */
+   } u;
+};
+
+
+/**
+ * Describe a texture sampler interpolator.
+ *
+ * The interpolation is described in terms of regular inputs.
+ */
+struct lp_tgsi_texture_info
+{
+   struct lp_tgsi_channel_info coord[4];
+   unsigned target:8; /* TGSI_TEXTURE_* */
+   unsigned unit:8;  /* Sampler unit */
+   unsigned modifier:8; /* LP_BLD_TEX_MODIFIER_* */
+};
+
+
+struct lp_tgsi_info
+{
+   struct tgsi_shader_info base;
+
+   /*
+    * Whether any of the texture opcodes access a register file other than
+    * TGSI_FILE_INPUT.
+    *
+    * We could also handle TGSI_FILE_CONST/IMMEDIATE here, but there is little
+    * benefit.
+    */
+   unsigned indirect_textures:1;
+
+   /*
+    * Texture opcode description. Aimed at detecting and described direct
+    * texture opcodes.
+    */
+   unsigned num_texs;
+   struct lp_tgsi_texture_info tex[PIPE_MAX_SAMPLERS];
+
+   /*
+    * Output description. Aimed at detecting and describing simple blit
+    * shaders.
+    */
+   struct lp_tgsi_channel_info output[PIPE_MAX_SHADER_OUTPUTS][4];
+
+   /*
+    * Shortcut pointers into the above (for fragment shaders).
+    */
+   const struct lp_tgsi_channel_info *cbuf[PIPE_MAX_COLOR_BUFS];
+};
+
 /**
  * Sampler code generation interface.
  *
@@ -96,6 +168,11 @@ struct lp_build_sampler_aos
 };
 
 
+void
+lp_build_tgsi_info(const struct tgsi_token *tokens,
+                   struct lp_tgsi_info *info);
+
+
 void
 lp_build_tgsi_soa(LLVMBuilderRef builder,
                   const struct tgsi_token *tokens,

src/gallium/auxiliary/gallivm/lp_bld_tgsi_aos.c

@@ -513,7 +513,7 @@ emit_instruction(
 {
    LLVMValueRef src0, src1, src2;
    LLVMValueRef tmp0, tmp1;
-   LLVMValueRef dst0;
+   LLVMValueRef dst0 = NULL;
 
    /*
     * Stores and write masks are handled in a general fashion after the long

src/gallium/auxiliary/gallivm/lp_bld_tgsi_info.c

@@ -0,0 +1,479 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * 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, sub license, 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 SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ **************************************************************************/
+
+
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_util.h"
+#include "tgsi/tgsi_dump.h"
+#include "lp_bld_debug.h"
+#include "lp_bld_tgsi.h"
+
+
+/**
+ * Analysis context.
+ *
+ * This is where we keep store the value of each channel of the IMM/TEMP/OUT
+ * register values, as we walk the shader.
+ */
+struct analysis_context
+{
+   struct lp_tgsi_info *info;
+
+   unsigned num_imms;
+   float imm[32][4];
+
+   struct lp_tgsi_channel_info temp[32][4];
+};
+
+
+/**
+ * Describe the specified channel of the src register.
+ */
+static void
+analyse_src(struct analysis_context *ctx,
+            struct lp_tgsi_channel_info *chan_info,
+            const struct tgsi_src_register *src,
+            unsigned chan)
+{
+   chan_info->file = TGSI_FILE_NULL;
+   if (!src->Indirect && !src->Absolute && !src->Negate) {
+      unsigned swizzle = tgsi_util_get_src_register_swizzle(src, chan);
+      if (src->File == TGSI_FILE_TEMPORARY) {
+         if (src->Index < Elements(ctx->temp)) {
+            *chan_info = ctx->temp[src->Index][swizzle];
+         }
+      } else {
+         chan_info->file = src->File;
+         if (src->File == TGSI_FILE_IMMEDIATE) {
+            assert(src->Index < Elements(ctx->imm));
+            if (src->Index < Elements(ctx->imm)) {
+               chan_info->u.value = ctx->imm[src->Index][swizzle];
+            }
+         } else {
+            chan_info->u.index = src->Index;
+            chan_info->swizzle = swizzle;
+         }
+      }
+   }
+}
+
+
+/**
+ * Whether this register channel refers to a specific immediate value.
+ */
+static boolean
+is_immediate(const struct lp_tgsi_channel_info *chan_info, float value)
+{
+   return chan_info->file == TGSI_FILE_IMMEDIATE &&
+          chan_info->u.value == value;
+}
+
+
+static void
+analyse_tex(struct analysis_context *ctx,
+            const struct tgsi_full_instruction *inst,
+            enum lp_build_tex_modifier modifier)
+{
+   struct lp_tgsi_info *info = ctx->info;
+   unsigned chan;
+
+   if (info->num_texs < Elements(info->tex)) {
+      struct lp_tgsi_texture_info *tex_info = &info->tex[info->num_texs];
+      bool indirect = FALSE;
+      unsigned readmask = 0;
+
+      tex_info->target = inst->Texture.Texture;
+      switch (inst->Texture.Texture) {
+      case TGSI_TEXTURE_1D:
+         readmask = TGSI_WRITEMASK_X;
+         break;
+      case TGSI_TEXTURE_2D:
+      case TGSI_TEXTURE_RECT:
+         readmask = TGSI_WRITEMASK_XY;
+         break;
+      case TGSI_TEXTURE_SHADOW1D:
+      case TGSI_TEXTURE_SHADOW2D:
+      case TGSI_TEXTURE_SHADOWRECT:
+      case TGSI_TEXTURE_3D:
+      case TGSI_TEXTURE_CUBE:
+         readmask = TGSI_WRITEMASK_XYZ;
+         break;
+      default:
+         assert(0);
+         return;
+      }
+
+      if (modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV) {
+         /* We don't track explicit derivatives, although we could */
+         indirect = TRUE;
+         tex_info->unit = inst->Src[3].Register.Index;
+      }  else {
+         if (modifier == LP_BLD_TEX_MODIFIER_PROJECTED ||
+             modifier == LP_BLD_TEX_MODIFIER_LOD_BIAS ||
+             modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_LOD) {
+            readmask |= TGSI_WRITEMASK_W;
+         }
+         tex_info->unit = inst->Src[1].Register.Index;
+      }
+
+      for (chan = 0; chan < 4; ++chan) {
+         struct lp_tgsi_channel_info *chan_info = &tex_info->coord[chan];
+         if (readmask & (1 << chan)) {
+            analyse_src(ctx, chan_info, &inst->Src[0].Register, chan);
+            if (chan_info->file != TGSI_FILE_INPUT) {
+               indirect = TRUE;
+            }
+         } else {
+            memset(chan_info, 0, sizeof *chan_info);
+         }
+      }
+
+      if (indirect) {
+         info->indirect_textures = TRUE;
+      }
+
+      ++info->num_texs;
+   } else {
+      info->indirect_textures = TRUE;
+   }
+}
+
+
+/**
+ * Process an instruction, and update the register values accordingly.
+ */
+static void
+analyse_instruction(struct analysis_context *ctx,
+                    struct tgsi_full_instruction *inst)
+{
+   struct lp_tgsi_info *info = ctx->info;
+   struct lp_tgsi_channel_info (*regs)[4];
+   unsigned max_regs;
+   unsigned i;
+   unsigned index;
+   unsigned chan;
+
+   for (i = 0; i < inst->Instruction.NumDstRegs; ++i) {
+      const struct tgsi_dst_register *dst = &inst->Dst[i].Register;
+
+      /*
+       * Get the lp_tgsi_channel_info array corresponding to the destination
+       * register file.
+       */
+
+      if (dst->File == TGSI_FILE_TEMPORARY) {
+         regs = ctx->temp;
+         max_regs = Elements(ctx->temp);
+      } else if (dst->File == TGSI_FILE_OUTPUT) {
+         regs = info->output;
+         max_regs = Elements(info->output);
+      } else if (dst->File == TGSI_FILE_ADDRESS ||
+                 dst->File == TGSI_FILE_PREDICATE) {
+         continue;
+      } else {
+         assert(0);
+         continue;
+      }
+
+      /*
+       * Detect direct TEX instructions
+       */
+
+      switch (inst->Instruction.Opcode) {
+      case TGSI_OPCODE_TEX:
+         analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_NONE);
+         break;
+      case TGSI_OPCODE_TXD:
+         analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV);
+         break;
+      case TGSI_OPCODE_TXB:
+         analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_LOD_BIAS);
+         break;
+      case TGSI_OPCODE_TXL:
+         analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_EXPLICIT_LOD);
+         break;
+      case TGSI_OPCODE_TXP:
+         analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_PROJECTED);
+         break;
+      default:
+         break;
+      }
+
+      /*
+       * Keep track of assignments and writes
+       */
+
+      if (dst->Indirect) {
+         /*
+          * It could be any register index so clear all register indices.
+          */
+
+         for (chan = 0; chan < 4; ++chan) {
+            if (dst->WriteMask & (1 << chan)) {
+               for (index = 0; index < max_regs; ++index) {
+                  regs[index][chan].file = TGSI_FILE_NULL;
+               }
+            }
+         }
+      } else if (dst->Index < max_regs) {
+         /*
+          * Update this destination register value.
+          */
+
+         struct lp_tgsi_channel_info res[4];
+
+         memset(res, 0, sizeof res);
+
+         if (!inst->Instruction.Predicate &&
+             !inst->Instruction.Saturate) {
+            for (chan = 0; chan < 4; ++chan) {
+               if (dst->WriteMask & (1 << chan)) {
+                  if (inst->Instruction.Opcode == TGSI_OPCODE_MOV) {
+                     analyse_src(ctx, &res[chan],
+                                 &inst->Src[0].Register, chan);
+                  } else if (inst->Instruction.Opcode == TGSI_OPCODE_MUL) {
+                     /*
+                      * Propagate values across 1.0 and 0.0 multiplications.
+                      */
+
+                     struct lp_tgsi_channel_info src0;
+                     struct lp_tgsi_channel_info src1;
+
+                     analyse_src(ctx, &src0, &inst->Src[0].Register, chan);
+                     analyse_src(ctx, &src1, &inst->Src[1].Register, chan);
+
+                     if (is_immediate(&src0, 0.0f)) {
+                        res[chan] = src0;
+                     } else if (is_immediate(&src1, 0.0f)) {
+                        res[chan] = src1;
+                     } else if (is_immediate(&src0, 1.0f)) {
+                        res[chan] = src1;
+                     } else if (is_immediate(&src1, 1.0f)) {
+                        res[chan] = src0;
+                     }
+                  }
+               }
+            }
+         }
+
+         for (chan = 0; chan < 4; ++chan) {
+            if (dst->WriteMask & (1 << chan)) {
+               regs[dst->Index][chan] = res[chan];
+            }
+         }
+      }
+   }
+
+   /*
+    * Clear all temporaries information in presence of a control flow opcode.
+    */
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_IF:
+   case TGSI_OPCODE_IFC:
+   case TGSI_OPCODE_ELSE:
+   case TGSI_OPCODE_ENDIF:
+   case TGSI_OPCODE_BGNLOOP:
+   case TGSI_OPCODE_BRK:
+   case TGSI_OPCODE_BREAKC:
+   case TGSI_OPCODE_CONT:
+   case TGSI_OPCODE_ENDLOOP:
+   case TGSI_OPCODE_CALLNZ:
+   case TGSI_OPCODE_CAL:
+   case TGSI_OPCODE_BGNSUB:
+   case TGSI_OPCODE_ENDSUB:
+   case TGSI_OPCODE_SWITCH:
+   case TGSI_OPCODE_CASE:
+   case TGSI_OPCODE_DEFAULT:
+   case TGSI_OPCODE_ENDSWITCH:
+   case TGSI_OPCODE_RET:
+   case TGSI_OPCODE_END:
+      /* XXX: Are there more cases? */
+      memset(&ctx->temp, 0, sizeof ctx->temp);
+      memset(&info->output, 0, sizeof info->output);
+   default:
+      break;
+   }
+}
+
+
+static INLINE void
+dump_info(const struct tgsi_token *tokens,
+          struct lp_tgsi_info *info)
+{
+   unsigned index;
+   unsigned chan;
+
+   tgsi_dump(tokens, 0);
+
+   for (index = 0; index < info->num_texs; ++index) {
+      const struct lp_tgsi_texture_info *tex_info = &info->tex[index];
+      debug_printf("TEX[%u] =", index);
+      for (chan = 0; chan < 4; ++chan) {
+         const struct lp_tgsi_channel_info *chan_info =
+               &tex_info->coord[chan];
+         if (chan_info->file != TGSI_FILE_NULL) {
+            debug_printf(" %s[%u].%c",
+                         tgsi_file_names[chan_info->file],
+                         chan_info->u.index,
+                         "xyzw01"[chan_info->swizzle]);
+         } else {
+            debug_printf(" _");
+         }
+      }
+      debug_printf(", SAMP[%u], %s\n",
+                   tex_info->unit,
+                   tgsi_texture_names[tex_info->target]);
+   }
+
+   for (index = 0; index < PIPE_MAX_SHADER_OUTPUTS; ++index) {
+      for (chan = 0; chan < 4; ++chan) {
+         const struct lp_tgsi_channel_info *chan_info =
+               &info->output[index][chan];
+         if (chan_info->file != TGSI_FILE_NULL) {
+            debug_printf("OUT[%u].%c = ", index, "xyzw"[chan]);
+            if (chan_info->file == TGSI_FILE_IMMEDIATE) {
+               debug_printf("%f", chan_info->u.value);
+            } else {
+               const char *file_name;
+               switch (chan_info->file) {
+               case TGSI_FILE_CONSTANT:
+                  file_name = "CONST";
+                  break;
+               case TGSI_FILE_INPUT:
+                  file_name = "IN";
+                  break;
+               default:
+                  file_name = "???";
+                  break;
+               }
+               debug_printf("%s[%u].%c",
+                            file_name,
+                            chan_info->u.index,
+                            "xyzw01"[chan_info->swizzle]);
+            }
+            debug_printf("\n");
+         }
+      }
+   }
+}
+
+
+/**
+ * Detect any direct relationship between the output color
+ */
+void
+lp_build_tgsi_info(const struct tgsi_token *tokens,
+                   struct lp_tgsi_info *info)
+{
+   struct tgsi_parse_context parse;
+   struct analysis_context ctx;
+   unsigned index;
+   unsigned chan;
+
+   memset(info, 0, sizeof *info);
+
+   tgsi_scan_shader(tokens, &info->base);
+
+   memset(&ctx, 0, sizeof ctx);
+   ctx.info = info;
+
+   tgsi_parse_init(&parse, tokens);
+
+   while (!tgsi_parse_end_of_tokens(&parse)) {
+      tgsi_parse_token(&parse);
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         {
+            struct tgsi_full_instruction *inst =
+                  &parse.FullToken.FullInstruction;
+
+            if (inst->Instruction.Opcode == TGSI_OPCODE_END ||
+                inst->Instruction.Opcode == TGSI_OPCODE_BGNSUB) {
+               /* We reached the end of main function body. */
+               goto finished;
+            }
+
+            analyse_instruction(&ctx, inst);
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         {
+            const unsigned size =
+                  parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
+            assert(size <= 4);
+            if (ctx.num_imms < Elements(ctx.imm)) {
+               for (chan = 0; chan < size; ++chan) {
+                  ctx.imm[ctx.num_imms][chan] =
+                        parse.FullToken.FullImmediate.u[chan].Float;
+               }
+               ++ctx.num_imms;
+            }
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_PROPERTY:
+         break;
+
+      default:
+         assert(0);
+      }
+   }
+finished:
+
+   tgsi_parse_free(&parse);
+
+
+   /*
+    * Link the output color values.
+    */
+
+   for (index = 0; index < PIPE_MAX_COLOR_BUFS; ++index) {
+      const struct lp_tgsi_channel_info null_output[4];
+      info->cbuf[index] = null_output;
+   }
+
+   for (index = 0; index < info->base.num_outputs; ++index) {
+      unsigned semantic_name = info->base.output_semantic_name[index];
+      unsigned semantic_index = info->base.output_semantic_index[index];
+      if (semantic_name == TGSI_SEMANTIC_COLOR &&
+          semantic_index < PIPE_MAX_COLOR_BUFS) {
+         info->cbuf[semantic_index] = info->output[index];
+      }
+   }
+
+   if (gallivm_debug & GALLIVM_DEBUG_TGSI) {
+      dump_info(tokens, info);
+   }
+}

src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c

@@ -887,21 +887,25 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
    }
 
    if (modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV) {
+      LLVMTypeRef i32t = LLVMInt32Type();
+      LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
       for (i = 0; i < num_coords; i++) {
-         ddx[i] = emit_fetch( bld, inst, 1, i );
+         LLVMValueRef src1 = emit_fetch( bld, inst, 1, i );
-         ddy[i] = emit_fetch( bld, inst, 2, i );
+         LLVMValueRef src2 = emit_fetch( bld, inst, 2, i );
+         ddx[i] = LLVMBuildExtractElement(bld->base.builder, src1, index0, "");
+         ddy[i] = LLVMBuildExtractElement(bld->base.builder, src2, index0, "");
       }
       unit = inst->Src[3].Register.Index;
    }  else {
       for (i = 0; i < num_coords; i++) {
-         ddx[i] = lp_build_ddx( &bld->base, coords[i] );
+         ddx[i] = lp_build_scalar_ddx( &bld->base, coords[i] );
-         ddy[i] = lp_build_ddy( &bld->base, coords[i] );
+         ddy[i] = lp_build_scalar_ddy( &bld->base, coords[i] );
       }
       unit = inst->Src[1].Register.Index;
    }
    for (i = num_coords; i < 3; i++) {
-      ddx[i] = bld->base.undef;
+      ddx[i] = LLVMGetUndef(bld->base.elem_type);
-      ddy[i] = bld->base.undef;
+      ddy[i] = LLVMGetUndef(bld->base.elem_type);
    }
 
    bld->sampler->emit_fetch_texel(bld->sampler,
@@ -913,6 +917,43 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
                                   texel);
 }
 
+static boolean
+near_end_of_shader(struct lp_build_tgsi_soa_context *bld,
+		   int pc)
+{
+   int i;
+
+   for (i = 0; i < 5; i++) {
+      unsigned opcode;
+
+      if (pc + i >= bld->info->num_instructions)
+	 return TRUE;
+
+      opcode = bld->instructions[pc + i].Instruction.Opcode;
+
+      if (opcode == TGSI_OPCODE_END)
+	 return TRUE;
+
+      if (opcode == TGSI_OPCODE_TEX ||
+	  opcode == TGSI_OPCODE_TXP ||
+	  opcode == TGSI_OPCODE_TXD ||
+	  opcode == TGSI_OPCODE_TXB ||
+	  opcode == TGSI_OPCODE_TXL ||
+	  opcode == TGSI_OPCODE_TXF ||
+	  opcode == TGSI_OPCODE_TXQ ||
+	  opcode == TGSI_OPCODE_CAL ||
+	  opcode == TGSI_OPCODE_CALLNZ ||
+	  opcode == TGSI_OPCODE_IF ||
+	  opcode == TGSI_OPCODE_IFC ||
+	  opcode == TGSI_OPCODE_BGNLOOP ||
+	  opcode == TGSI_OPCODE_SWITCH)
+	 return FALSE;
+   }
+
+   return TRUE;
+}
+
+
 
 /**
  * Kill fragment if any of the src register values are negative.
@@ -920,7 +961,8 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
 static void
 emit_kil(
    struct lp_build_tgsi_soa_context *bld,
-   const struct tgsi_full_instruction *inst )
+   const struct tgsi_full_instruction *inst,
+   int pc)
 {
    const struct tgsi_full_src_register *reg = &inst->Src[0];
    LLVMValueRef terms[NUM_CHANNELS];
@@ -959,8 +1001,12 @@ emit_kil(
       }
    }
 
-   if(mask)
+   if(mask) {
       lp_build_mask_update(bld->mask, mask);
+
+      if (!near_end_of_shader(bld, pc))
+	 lp_build_mask_check(bld->mask);
+   }
 }
 
 
@@ -972,7 +1018,8 @@ emit_kil(
  */
 static void
 emit_kilp(struct lp_build_tgsi_soa_context *bld,
-          const struct tgsi_full_instruction *inst)
+          const struct tgsi_full_instruction *inst,
+	  int pc)
 {
    LLVMValueRef mask;
 
@@ -983,10 +1030,14 @@ emit_kilp(struct lp_build_tgsi_soa_context *bld,
       mask = LLVMBuildNot(bld->base.builder, bld->exec_mask.exec_mask, "kilp");
    }
    else {
-      mask = bld->base.zero;
+      LLVMValueRef zero = LLVMConstNull(bld->base.int_vec_type);
+      mask = zero;
    }
 
    lp_build_mask_update(bld->mask, mask);
+
+   if (!near_end_of_shader(bld, pc))
+      lp_build_mask_check(bld->mask);
 }
 
 static void
@@ -1535,12 +1586,12 @@ emit_instruction(
 
    case TGSI_OPCODE_KILP:
       /* predicated kill */
-      emit_kilp( bld, inst );
+      emit_kilp( bld, inst, (*pc)-1 );
       break;
 
    case TGSI_OPCODE_KIL:
       /* conditional kill */
-      emit_kil( bld, inst );
+      emit_kil( bld, inst, (*pc)-1 );
       break;
 
    case TGSI_OPCODE_PK2H:

src/gallium/auxiliary/rtasm/rtasm_ppc.c

@@ -97,7 +97,7 @@ void (*ppc_get_func(struct ppc_function *p))(void)
       return (void (*)(void)) NULL;
    else
 #endif
-      return (void (*)(void)) p->store;
+      return (void (*)(void)) pointer_to_func(p->store);
 }
 
 

src/gallium/auxiliary/target-helpers/inline_sw_helper.h

@@ -23,25 +23,12 @@
 #include "cell/ppu/cell_public.h"
 #endif
 
+
 static INLINE struct pipe_screen *
-sw_screen_create(struct sw_winsys *winsys)
+sw_screen_create_named(struct sw_winsys *winsys, const char *driver)
 {
-   const char *default_driver;
-   const char *driver;
    struct pipe_screen *screen = NULL;
 
-#if defined(GALLIUM_CELL)
-   default_driver = "cell";
-#elif defined(GALLIUM_LLVMPIPE)
-   default_driver = "llvmpipe";
-#elif defined(GALLIUM_SOFTPIPE)
-   default_driver = "softpipe";
-#else
-   default_driver = "";
-#endif
-
-   driver = debug_get_option("GALLIUM_DRIVER", default_driver);
-
 #if defined(GALLIUM_CELL)
    if (screen == NULL && strcmp(driver, "cell") == 0)
       screen = cell_create_screen(winsys);
@@ -60,4 +47,26 @@ sw_screen_create(struct sw_winsys *winsys)
    return screen;
 }
 
+
+static INLINE struct pipe_screen *
+sw_screen_create(struct sw_winsys *winsys)
+{
+   const char *default_driver;
+   const char *driver;
+
+#if defined(GALLIUM_CELL)
+   default_driver = "cell";
+#elif defined(GALLIUM_LLVMPIPE)
+   default_driver = "llvmpipe";
+#elif defined(GALLIUM_SOFTPIPE)
+   default_driver = "softpipe";
+#else
+   default_driver = "";
+#endif
+
+   driver = debug_get_option("GALLIUM_DRIVER", default_driver);
+   return sw_screen_create_named(winsys, driver);
+}
+
+
 #endif

src/gallium/auxiliary/target-helpers/inline_wrapper_sw_helper.h

@@ -13,20 +13,26 @@ static INLINE struct pipe_screen *
 sw_screen_wrap(struct pipe_screen *screen)
 {
    struct sw_winsys *sws;
-   struct pipe_screen *sw_screen;
+   struct pipe_screen *sw_screen = NULL;
+   const char *driver;
 
-   sws = wrapper_sw_winsys_warp_pipe_screen(screen);
+   driver = debug_get_option("GALLIUM_DRIVER", "native");
+   if (strcmp(driver, "native") == 0)
+      return screen;
+
+   sws = wrapper_sw_winsys_wrap_pipe_screen(screen);
    if (!sws)
       goto err;
 
-   sw_screen = sw_screen_create(sws);
+   sw_screen = sw_screen_create_named(sws, driver);
-   if (sw_screen == screen)
+
+   if (!sw_screen)
       goto err_winsys;
 
    return sw_screen;
 
 err_winsys:
-   sws->destroy(sws);
+   return wrapper_sw_winsys_dewrap_pipe_screen(sws);
 err:
    return screen;
 }

src/gallium/auxiliary/tgsi/tgsi_dump.c

@@ -90,7 +90,8 @@ static const char *processor_type_names[] =
    "GEOM"
 };
 
-static const char *file_names[TGSI_FILE_COUNT] =
+const char *
+tgsi_file_names[TGSI_FILE_COUNT] =
 {
    "NULL",
    "CONST",
@@ -125,7 +126,8 @@ static const char *semantic_names[] =
    "FACE",
    "EDGEFLAG",
    "PRIM_ID",
-   "INSTANCEID"
+   "INSTANCEID",
+   "STENCIL"
 };
 
 static const char *immediate_type_names[] =
@@ -135,7 +137,8 @@ static const char *immediate_type_names[] =
    "INT32"
 };
 
-static const char *swizzle_names[] =
+const char *
+tgsi_swizzle_names[] =
 {
    "x",
    "y",
@@ -143,7 +146,8 @@ static const char *swizzle_names[] =
    "w"
 };
 
-static const char *texture_names[] =
+const char *
+tgsi_texture_names[] =
 {
    "UNKNOWN",
    "1D",
@@ -201,15 +205,15 @@ _dump_register_src(
    struct dump_ctx *ctx,
    const struct tgsi_full_src_register *src )
 {
-   ENM(src->Register.File, file_names);
+   ENM(src->Register.File, tgsi_file_names);
    if (src->Register.Dimension) {
       if (src->Dimension.Indirect) {
          CHR( '[' );
-         ENM( src->DimIndirect.File, file_names );
+         ENM( src->DimIndirect.File, tgsi_file_names );
          CHR( '[' );
          SID( src->DimIndirect.Index );
          TXT( "]." );
-         ENM( src->DimIndirect.SwizzleX, swizzle_names );
+         ENM( src->DimIndirect.SwizzleX, tgsi_swizzle_names );
          if (src->Dimension.Index != 0) {
             if (src->Dimension.Index > 0)
                CHR( '+' );
@@ -224,11 +228,11 @@ _dump_register_src(
    }
    if (src->Register.Indirect) {
       CHR( '[' );
-      ENM( src->Indirect.File, file_names );
+      ENM( src->Indirect.File, tgsi_file_names );
       CHR( '[' );
       SID( src->Indirect.Index );
       TXT( "]." );
-      ENM( src->Indirect.SwizzleX, swizzle_names );
+      ENM( src->Indirect.SwizzleX, tgsi_swizzle_names );
       if (src->Register.Index != 0) {
          if (src->Register.Index > 0)
             CHR( '+' );
@@ -248,15 +252,15 @@ _dump_register_dst(
    struct dump_ctx *ctx,
    const struct tgsi_full_dst_register *dst )
 {
-   ENM(dst->Register.File, file_names);
+   ENM(dst->Register.File, tgsi_file_names);
    if (dst->Register.Dimension) {
       if (dst->Dimension.Indirect) {
          CHR( '[' );
-         ENM( dst->DimIndirect.File, file_names );
+         ENM( dst->DimIndirect.File, tgsi_file_names );
          CHR( '[' );
          SID( dst->DimIndirect.Index );
          TXT( "]." );
-         ENM( dst->DimIndirect.SwizzleX, swizzle_names );
+         ENM( dst->DimIndirect.SwizzleX, tgsi_swizzle_names );
          if (dst->Dimension.Index != 0) {
             if (dst->Dimension.Index > 0)
                CHR( '+' );
@@ -271,11 +275,11 @@ _dump_register_dst(
    }
    if (dst->Register.Indirect) {
       CHR( '[' );
-      ENM( dst->Indirect.File, file_names );
+      ENM( dst->Indirect.File, tgsi_file_names );
       CHR( '[' );
       SID( dst->Indirect.Index );
       TXT( "]." );
-      ENM( dst->Indirect.SwizzleX, swizzle_names );
+      ENM( dst->Indirect.SwizzleX, tgsi_swizzle_names );
       if (dst->Register.Index != 0) {
          if (dst->Register.Index > 0)
             CHR( '+' );
@@ -351,7 +355,7 @@ iter_declaration(
 
    TXT( "DCL " );
 
-   ENM(decl->Declaration.File, file_names);
+   ENM(decl->Declaration.File, tgsi_file_names);
 
    /* all geometry shader inputs are two dimensional */
    if (decl->Declaration.File == TGSI_FILE_INPUT &&
@@ -585,10 +589,10 @@ iter_instruction(
           inst->Predicate.SwizzleZ != TGSI_SWIZZLE_Z ||
           inst->Predicate.SwizzleW != TGSI_SWIZZLE_W) {
          CHR( '.' );
-         ENM( inst->Predicate.SwizzleX, swizzle_names );
+         ENM( inst->Predicate.SwizzleX, tgsi_swizzle_names );
-         ENM( inst->Predicate.SwizzleY, swizzle_names );
+         ENM( inst->Predicate.SwizzleY, tgsi_swizzle_names );
-         ENM( inst->Predicate.SwizzleZ, swizzle_names );
+         ENM( inst->Predicate.SwizzleZ, tgsi_swizzle_names );
-         ENM( inst->Predicate.SwizzleW, swizzle_names );
+         ENM( inst->Predicate.SwizzleW, tgsi_swizzle_names );
       }
 
       TXT( ") " );
@@ -641,10 +645,10 @@ iter_instruction(
           src->Register.SwizzleZ != TGSI_SWIZZLE_Z ||
           src->Register.SwizzleW != TGSI_SWIZZLE_W) {
          CHR( '.' );
-         ENM( src->Register.SwizzleX, swizzle_names );
+         ENM( src->Register.SwizzleX, tgsi_swizzle_names );
-         ENM( src->Register.SwizzleY, swizzle_names );
+         ENM( src->Register.SwizzleY, tgsi_swizzle_names );
-         ENM( src->Register.SwizzleZ, swizzle_names );
+         ENM( src->Register.SwizzleZ, tgsi_swizzle_names );
-         ENM( src->Register.SwizzleW, swizzle_names );
+         ENM( src->Register.SwizzleW, tgsi_swizzle_names );
       }
 
       if (src->Register.Absolute)
@@ -655,7 +659,7 @@ iter_instruction(
 
    if (inst->Instruction.Texture) {
       TXT( ", " );
-      ENM( inst->Texture.Texture, texture_names );
+      ENM( inst->Texture.Texture, tgsi_texture_names );
    }
 
    switch (inst->Instruction.Opcode) {

src/gallium/auxiliary/tgsi/tgsi_dump.h

@@ -35,6 +35,15 @@
 extern "C" {
 #endif
 
+extern const char *
+tgsi_file_names[TGSI_FILE_COUNT];
+
+extern const char *
+tgsi_swizzle_names[];
+
+extern const char *
+tgsi_texture_names[];
+
 void
 tgsi_dump_str(
    const struct tgsi_token *tokens,

src/gallium/auxiliary/tgsi/tgsi_scan.c

@@ -147,6 +147,7 @@ tgsi_scan_shader(const struct tgsi_token *tokens,
                   info->input_semantic_name[reg] = (ubyte)fulldecl->Semantic.Name;
                   info->input_semantic_index[reg] = (ubyte)fulldecl->Semantic.Index;
                   info->input_interpolate[reg] = (ubyte)fulldecl->Declaration.Interpolate;
+                  info->input_centroid[reg] = (ubyte)fulldecl->Declaration.Centroid;
                   info->input_cylindrical_wrap[reg] = (ubyte)fulldecl->Declaration.CylindricalWrap;
                   info->num_inputs++;
                }
@@ -157,9 +158,11 @@ tgsi_scan_shader(const struct tgsi_token *tokens,
 
                   /* extra info for special outputs */
                   if (procType == TGSI_PROCESSOR_FRAGMENT &&
-                      fulldecl->Semantic.Name == TGSI_SEMANTIC_POSITION) {
+                      fulldecl->Semantic.Name == TGSI_SEMANTIC_POSITION)
                         info->writes_z = TRUE;
-                  }
+                  if (procType == TGSI_PROCESSOR_FRAGMENT &&
+                      fulldecl->Semantic.Name == TGSI_SEMANTIC_STENCIL)
+                        info->writes_stencil = TRUE;
                   if (procType == TGSI_PROCESSOR_VERTEX &&
                       fulldecl->Semantic.Name == TGSI_SEMANTIC_EDGEFLAG) {
                      info->writes_edgeflag = TRUE;

src/gallium/auxiliary/tgsi/tgsi_scan.h

@@ -45,6 +45,7 @@ struct tgsi_shader_info
    ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; /**< TGSI_SEMANTIC_x */
    ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
    ubyte input_interpolate[PIPE_MAX_SHADER_INPUTS];
+   ubyte input_centroid[PIPE_MAX_SHADER_INPUTS];
    ubyte input_usage_mask[PIPE_MAX_SHADER_INPUTS];
    ubyte input_cylindrical_wrap[PIPE_MAX_SHADER_INPUTS];
    ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; /**< TGSI_SEMANTIC_x */
@@ -60,6 +61,7 @@ struct tgsi_shader_info
    uint opcode_count[TGSI_OPCODE_LAST];  /**< opcode histogram */
 
    boolean writes_z;  /**< does fragment shader write Z value? */
+   boolean writes_stencil; /**< does fragment shader write stencil value? */
    boolean writes_edgeflag; /**< vertex shader outputs edgeflag */
    boolean uses_kill;  /**< KIL or KILP instruction used? */
 

src/gallium/auxiliary/translate/translate.h

@@ -68,6 +68,33 @@ struct translate_key {
 };
 
 
+struct translate;
+
+
+typedef void (PIPE_CDECL *run_elts_func)(struct translate *,
+                                         const unsigned *elts,
+                                         unsigned count,
+                                         unsigned instance_id,
+                                         void *output_buffer);
+
+typedef void (PIPE_CDECL *run_elts16_func)(struct translate *,
+                                           const uint16_t *elts,
+                                           unsigned count,
+                                           unsigned instance_id,
+                                           void *output_buffer);
+
+typedef void (PIPE_CDECL *run_elts8_func)(struct translate *,
+                                          const uint8_t *elts,
+                                          unsigned count,
+                                          unsigned instance_id,
+                                          void *output_buffer);
+
+typedef void (PIPE_CDECL *run_func)(struct translate *,
+                                    unsigned start,
+                                    unsigned count,
+                                    unsigned instance_id,
+                                    void *output_buffer);
+
 struct translate {
    struct translate_key key;
 
@@ -79,42 +106,14 @@ struct translate {
 		       unsigned stride,
 		       unsigned max_index );
 
-   void (PIPE_CDECL *run_elts)( struct translate *,
+   run_elts_func run_elts;
-                                const unsigned *elts,
+   run_elts16_func run_elts16;
-                                unsigned count,
+   run_elts8_func run_elts8;
-                                unsigned instance_id,
+   run_func run;
-                                void *output_buffer);
-
-   void (PIPE_CDECL *run_elts16)( struct translate *,
-                                const uint16_t *elts,
-                                unsigned count,
-                                unsigned instance_id,
-                                void *output_buffer);
-
-   void (PIPE_CDECL *run_elts8)( struct translate *,
-                                const uint8_t *elts,
-                                unsigned count,
-                                unsigned instance_id,
-                                void *output_buffer);
-
-   void (PIPE_CDECL *run)( struct translate *,
-                           unsigned start,
-                           unsigned count,
-                           unsigned instance_id,
-                           void *output_buffer);
 };
 
 
 
-#if 0
-struct translate_context *translate_context_create( void );
-void translate_context_destroy( struct translate_context * );
-
-struct translate *translate_lookup_or_create( struct translate_context *tctx,
-					      const struct translate_key *key );
-#endif
-
-
 struct translate *translate_create( const struct translate_key *key );
 
 boolean translate_is_output_format_supported(enum pipe_format format);

src/gallium/auxiliary/translate/translate_sse.c

@@ -1495,19 +1495,19 @@ struct translate *translate_sse2_create( const struct translate_key *key )
    if (!build_vertex_emit(p, &p->elt8_func, 1))
       goto fail;
 
-   p->translate.run = (void*)x86_get_func(&p->linear_func);
+   p->translate.run = (run_func) x86_get_func(&p->linear_func);
    if (p->translate.run == NULL)
       goto fail;
 
-   p->translate.run_elts = (void*)x86_get_func(&p->elt_func);
+   p->translate.run_elts = (run_elts_func) x86_get_func(&p->elt_func);
    if (p->translate.run_elts == NULL)
       goto fail;
 
-   p->translate.run_elts16 = (void*)x86_get_func(&p->elt16_func);
+   p->translate.run_elts16 = (run_elts16_func) x86_get_func(&p->elt16_func);
    if (p->translate.run_elts16 == NULL)
       goto fail;
 
-   p->translate.run_elts8 = (void*)x86_get_func(&p->elt8_func);
+   p->translate.run_elts8 = (run_elts8_func) x86_get_func(&p->elt8_func);
    if (p->translate.run_elts8 == NULL)
       goto fail;
 

src/gallium/auxiliary/util/u_dl.c

@@ -38,6 +38,7 @@
 #endif
 
 #include "u_dl.h"
+#include "u_pointer.h"
 
 
 struct util_dl_library *
@@ -58,7 +59,7 @@ util_dl_get_proc_address(struct util_dl_library *library,
                          const char *procname)
 {
 #if defined(PIPE_OS_UNIX)
-   return (util_dl_proc)dlsym((void *)library, procname);
+   return (util_dl_proc) pointer_to_func(dlsym((void *)library, procname));
 #elif defined(PIPE_OS_WINDOWS)
    return (util_dl_proc)GetProcAddress((HMODULE)library, procname);
 #else

src/gallium/auxiliary/util/u_format.csv

@@ -109,9 +109,12 @@ PIPE_FORMAT_Z32_UNORM               , plain, 1, 1, un32,     ,     ,     , x___,
 PIPE_FORMAT_Z32_FLOAT               , plain, 1, 1, f32 ,     ,     ,     , x___, zs
 PIPE_FORMAT_Z24_UNORM_S8_USCALED    , plain, 1, 1, un24, u8  ,     ,     , xy__, zs
 PIPE_FORMAT_S8_USCALED_Z24_UNORM    , plain, 1, 1, u8 ,  un24,     ,     , yx__, zs
+PIPE_FORMAT_X24S8_USCALED           , plain, 1, 1, x24,  u8  ,     ,     , _y__, zs
+PIPE_FORMAT_S8X24_USCALED           , plain, 1, 1, u8  , x24 ,     ,     , _x__, zs
 PIPE_FORMAT_Z24X8_UNORM             , plain, 1, 1, un24, x8  ,     ,     , x___, zs
 PIPE_FORMAT_X8Z24_UNORM             , plain, 1, 1, x8  , un24,     ,     , y___, zs
 PIPE_FORMAT_Z32_FLOAT_S8X24_USCALED , plain, 1, 1, f32,  u8  , x24 ,     , xy__, zs
+PIPE_FORMAT_X32_S8X24_USCALED       , plain, 1, 1, x32,  u8  , x24 ,     , _y__, zs
 
 # YUV formats
 # http://www.fourcc.org/yuv.php#UYVY

src/gallium/auxiliary/util/u_format_srgb.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 
-'''
+CopyRight = '''
 /**************************************************************************
  *
  * Copyright 2010 VMware, Inc.
@@ -89,7 +89,7 @@ def main():
     print '/* This file is autogenerated by u_format_srgb.py. Do not edit directly. */'
     print
     # This will print the copyright message on the top of this file
-    print __doc__.strip()
+    print CopyRight.strip()
     print
     print '#include "u_format_srgb.h"'
     print

src/gallium/auxiliary/util/u_format_table.py

@@ -1,6 +1,6 @@
 #!/usr/bin/env python
 
-'''
+CopyRight = '''
 /**************************************************************************
  *
  * Copyright 2010 VMware, Inc.
@@ -83,7 +83,7 @@ def write_format_table(formats):
     print '/* This file is autogenerated by u_format_table.py from u_format.csv. Do not edit directly. */'
     print
     # This will print the copyright message on the top of this file
-    print __doc__.strip()
+    print CopyRight.strip()
     print
     print '#include "u_format.h"'
     print '#include "u_format_s3tc.h"'

src/gallium/auxiliary/util/u_format_zs.c

@@ -918,3 +918,56 @@ util_format_z32_float_s8x24_uscaled_pack_s_8uscaled(uint8_t *dst_row, unsigned d
    }
 }
 
+
+void
+util_format_x24s8_uscaled_unpack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   util_format_z24_unorm_s8_uscaled_unpack_s_8uscaled(dst_row, dst_stride,
+						      src_row, src_stride,
+						      width, height);
+}
+
+void
+util_format_x24s8_uscaled_pack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   util_format_z24_unorm_s8_uscaled_pack_s_8uscaled(dst_row, dst_stride,
+						    src_row, src_stride,
+						    width, height);
+}
+
+void
+util_format_s8x24_uscaled_unpack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   util_format_s8_uscaled_z24_unorm_unpack_s_8uscaled(dst_row, dst_stride,
+						      src_row, src_stride,
+						      width, height);
+}
+
+void
+util_format_s8x24_uscaled_pack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   util_format_s8_uscaled_z24_unorm_pack_s_8uscaled(dst_row, dst_stride,
+						      src_row, src_stride,
+						      width, height);
+}
+
+void
+util_format_x32_s8x24_uscaled_unpack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride,
+						const uint8_t *src_row, unsigned src_stride,
+						unsigned width, unsigned height)
+{
+   util_format_z32_float_s8x24_uscaled_unpack_s_8uscaled(dst_row, dst_stride,
+							 src_row, src_stride,
+							 width, height);
+
+}
+
+void
+util_format_x32_s8x24_uscaled_pack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride,
+					      const uint8_t *src_row, unsigned src_stride,
+					      unsigned width, unsigned height)
+{
+   util_format_z32_float_s8x24_uscaled_pack_s_8uscaled(dst_row, dst_stride,
+                                                       src_row, src_stride,
+						       width, height);
+}

src/gallium/auxiliary/util/u_format_zs.h

@@ -192,5 +192,21 @@ util_format_z32_float_s8x24_uscaled_unpack_s_8uscaled(uint8_t *dst_row, unsigned
 void
 util_format_z32_float_s8x24_uscaled_pack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
 
+void
+util_format_x24s8_uscaled_unpack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
 
+void
+util_format_x24s8_uscaled_pack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_s8x24_uscaled_unpack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_s8x24_uscaled_pack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_x32_s8x24_uscaled_unpack_s_8uscaled(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_x32_s8x24_uscaled_pack_s_8uscaled(uint8_t *dst_row, unsigned dst_sride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
 #endif /* U_FORMAT_ZS_H_ */

src/gallium/auxiliary/util/u_math.h

@@ -118,6 +118,11 @@ __inline double __cdecl atan2(double val)
 #endif
 
 
+#ifndef M_SQRT2
+#define M_SQRT2 1.41421356237309504880
+#endif
+
+
 #if defined(_MSC_VER) 
 
 #if _MSC_VER < 1400 && !defined(__cplusplus) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)

src/gallium/auxiliary/util/u_sse.h

@@ -71,6 +71,96 @@ _mm_castps_si128(__m128 a)
 
 #endif /* defined(_MSC_VER) && _MSC_VER < 1500 */
 
+union m128i {
+   __m128i m;
+   ubyte ub[16];
+   ushort us[8];
+   uint ui[4];
+};
+
+static INLINE void u_print_epi8(const char *name, __m128i r)
+{
+   union { __m128i m; ubyte ub[16]; } u;
+   u.m = r;
+
+   debug_printf("%s: "
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x/"
+                "%02x\n",
+                name,
+                u.ub[0],  u.ub[1],  u.ub[2],  u.ub[3],
+                u.ub[4],  u.ub[5],  u.ub[6],  u.ub[7],
+                u.ub[8],  u.ub[9],  u.ub[10], u.ub[11],
+                u.ub[12], u.ub[13], u.ub[14], u.ub[15]);
+}
+
+static INLINE void u_print_epi16(const char *name, __m128i r)
+{
+   union { __m128i m; ushort us[8]; } u;
+   u.m = r;
+
+   debug_printf("%s: "
+                "%04x/"
+                "%04x/"
+                "%04x/"
+                "%04x/"
+                "%04x/"
+                "%04x/"
+                "%04x/"
+                "%04x\n",
+                name,
+                u.us[0],  u.us[1],  u.us[2],  u.us[3],
+                u.us[4],  u.us[5],  u.us[6],  u.us[7]);
+}
+
+static INLINE void u_print_epi32(const char *name, __m128i r)
+{
+   union { __m128i m; uint ui[4]; } u;
+   u.m = r;
+
+   debug_printf("%s: "
+                "%08x/"
+                "%08x/"
+                "%08x/"
+                "%08x\n",
+                name,
+                u.ui[0],  u.ui[1],  u.ui[2],  u.ui[3]);
+}
+
+static INLINE void u_print_ps(const char *name, __m128 r)
+{
+   union { __m128 m; float f[4]; } u;
+   u.m = r;
+
+   debug_printf("%s: "
+                "%f/"
+                "%f/"
+                "%f/"
+                "%f\n",
+                name,
+                u.f[0],  u.f[1],  u.f[2],  u.f[3]);
+}
+
+
+#define U_DUMP_EPI32(a) u_print_epi32(#a, a)
+#define U_DUMP_EPI16(a) u_print_epi16(#a, a)
+#define U_DUMP_EPI8(a)  u_print_epi8(#a, a)
+#define U_DUMP_PS(a)    u_print_ps(#a, a)
+
+
 
 #if defined(PIPE_ARCH_SSSE3)
 
@@ -98,6 +188,68 @@ _mm_shuffle_epi8(__m128i a, __m128i mask)
 #endif /* !PIPE_ARCH_SSSE3 */
 
 
-#endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */
+
+
+/* Provide an SSE2 implementation of _mm_mullo_epi32() in terms of
+ * _mm_mul_epu32().
+ *
+ * I suspect this works fine for us because one of our operands is
+ * always positive, but not sure that this can be used for general
+ * signed integer multiplication.
+ *
+ * This seems close enough to the speed of SSE4 and the real
+ * _mm_mullo_epi32() intrinsic as to not justify adding an sse4
+ * dependency at this point.
+ */
+static INLINE __m128i mm_mullo_epi32(const __m128i a, const __m128i b)
+{
+   __m128i a4   = _mm_srli_epi64(a, 32);  /* shift by one dword */
+   __m128i b4   = _mm_srli_epi64(b, 32);  /* shift by one dword */
+   __m128i ba   = _mm_mul_epu32(b, a);   /* multply dwords 0, 2 */
+   __m128i b4a4 = _mm_mul_epu32(b4, a4); /* multiply dwords 1, 3 */
+
+   /* Interleave the results, either with shuffles or (slightly
+    * faster) direct bit operations:
+    */
+#if 0
+   __m128i ba8             = _mm_shuffle_epi32(ba, 8);
+   __m128i b4a48           = _mm_shuffle_epi32(b4a4, 8);
+   __m128i result          = _mm_unpacklo_epi32(ba8, b4a48);
+#else
+   __m128i mask            = _mm_setr_epi32(~0,0,~0,0);
+   __m128i ba_mask         = _mm_and_si128(ba, mask);
+   __m128i b4a4_mask_shift = _mm_slli_epi64(b4a4, 32);
+   __m128i result          = _mm_or_si128(ba_mask, b4a4_mask_shift);
+#endif
+
+   return result;
+}
+
+
+static INLINE void
+transpose4_epi32(const __m128i * restrict a,
+                 const __m128i * restrict b,
+                 const __m128i * restrict c,
+                 const __m128i * restrict d,
+                 __m128i * restrict o,
+                 __m128i * restrict p,
+                 __m128i * restrict q,
+                 __m128i * restrict r)
+{
+  __m128i t0 = _mm_unpacklo_epi32(*a, *b);
+  __m128i t1 = _mm_unpacklo_epi32(*c, *d);
+  __m128i t2 = _mm_unpackhi_epi32(*a, *b);
+  __m128i t3 = _mm_unpackhi_epi32(*c, *d);
+
+  *o = _mm_unpacklo_epi64(t0, t1);
+  *p = _mm_unpackhi_epi64(t0, t1);
+  *q = _mm_unpacklo_epi64(t2, t3);
+  *r = _mm_unpackhi_epi64(t2, t3);
+}
+
+#define SCALAR_EPI32(m, i) _mm_shuffle_epi32((m), _MM_SHUFFLE(i,i,i,i))
+
+
+#endif /* PIPE_ARCH_SSE */
 
 #endif /* U_SSE_H_ */

src/gallium/auxiliary/util/u_tile.c

@@ -217,6 +217,81 @@ z24s8_get_tile_rgba(const unsigned *src,
    }
 }
 
+/*** PIPE_FORMAT_S8X24_USCALED ***/
+
+/**
+ * Return S component as four uint32_t in [0..255].  Z part ignored.
+ */
+static void
+s8x24_get_tile_rgba(const unsigned *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float)((*src++ >> 24) & 0xff);
+      }
+
+      p += dst_stride;
+   }
+}
+
+/*** PIPE_FORMAT_X24S8_USCALED ***/
+
+/**
+ * Return S component as four uint32_t in [0..255].  Z part ignored.
+ */
+static void
+x24s8_get_tile_rgba(const unsigned *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float)(*src++ & 0xff);
+      }
+      p += dst_stride;
+   }
+}
+
+
+/**
+ * Return S component as four uint32_t in [0..255].  Z part ignored.
+ */
+static void
+s8_get_tile_rgba(const unsigned char *src,
+		 unsigned w, unsigned h,
+		 float *p,
+		 unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float)(*src++ & 0xff);
+      }
+      p += dst_stride;
+   }
+}
 
 /*** PIPE_FORMAT_Z32_FLOAT ***/
 
@@ -261,10 +336,19 @@ pipe_tile_raw_to_rgba(enum pipe_format format,
    case PIPE_FORMAT_Z24X8_UNORM:
       s8z24_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
       break;
+   case PIPE_FORMAT_S8_USCALED:
+      s8_get_tile_rgba((unsigned char *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_X24S8_USCALED:
+      s8x24_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
+      break;
    case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
    case PIPE_FORMAT_X8Z24_UNORM:
       z24s8_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
       break;
+   case PIPE_FORMAT_S8X24_USCALED:
+      x24s8_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
+      break;
    case PIPE_FORMAT_Z32_FLOAT:
       z32f_get_tile_rgba((float *) src, w, h, dst, dst_stride);
       break;

src/gallium/docs/source/context.rst

@@ -156,6 +156,15 @@ If there is an index buffer bound, and ``indexed`` field is true, all vertex
 indices will be looked up in the index buffer.  ``min_index``, ``max_index``,
 and ``index_bias`` apply after index lookup.
 
+When drawing indexed primitives, the primitive restart index can be
+used to draw disjoint primitive strips.  For example, several separate
+line strips can be drawn by designating a special index value as the
+restart index.  The ``primitive_restart`` flag enables/disables this
+feature.  The ``restart_index`` field specifies the restart index value.
+
+When primitive restart is in use, array indexes are compared to the
+restart index before adding the index_bias offset.
+
 If a given vertex element has ``instance_divisor`` set to 0, it is said
 it contains per-vertex data and effective vertex attribute address needs
 to be recalculated for every index.

src/gallium/docs/source/tgsi.rst

@@ -1415,6 +1415,12 @@ Edge flags are used to control which lines or points are actually
 drawn when the polygon mode converts triangles/quads/polygons into
 points or lines.
 
+TGSI_SEMANTIC_STENCIL
+""""""""""""""""""""""
+
+For fragment shaders, this semantic label indicates than an output
+is a writable stencil reference value. Only the Y component is writable.
+This allows the fragment shader to change the fragments stencilref value.
 
 
 Properties
@@ -1493,6 +1499,8 @@ well.
 | Z                  | XXX TBD      | (z, z, z, 1)       | (0, z, 0, 1) |
 |                    |              | [#depth-tex-mode]_ |              |
 +--------------------+--------------+--------------------+--------------+
+| S                  | (s, s, s, s) | unknown            | unknown      |
++--------------------+--------------+--------------------+--------------+
 
 .. [#envmap-bumpmap] http://www.opengl.org/registry/specs/ATI/envmap_bumpmap.txt
 .. [#depth-tex-mode] the default is (z, z, z, 1) but may also be (0, 0, 0, z)

src/gallium/drivers/galahad/glhd_context.c

@@ -641,7 +641,7 @@ galahad_set_index_buffer(struct pipe_context *_pipe,
          break;
       default:
          glhd_warn("index buffer %p has unrecognized index size %d",
-               _ib->buffer, _ib->index_size);
+                   (void *) _ib->buffer, _ib->index_size);
          break;
       }
    }
@@ -1013,7 +1013,7 @@ galahad_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
 
    glhd_pipe->pipe = pipe;
 
-   glhd_warn("Created context %p", glhd_pipe);
+   glhd_warn("Created context %p", (void *) glhd_pipe);
 
    return &glhd_pipe->base;
 }

src/gallium/drivers/galahad/glhd_screen.c

@@ -370,7 +370,7 @@ galahad_screen_create(struct pipe_screen *screen)
 
    glhd_screen->screen = screen;
 
-   glhd_warn("Created screen %p", glhd_screen);
+   glhd_warn("Created screen %p", (void *) glhd_screen);
 
    return &glhd_screen->base;
 }

src/gallium/drivers/llvmpipe/Makefile

@@ -28,8 +28,6 @@ C_SOURCES = \
 	lp_scene_queue.c \
 	lp_screen.c \
 	lp_setup.c \
-	lp_setup_coef.c \
-	lp_setup_coef_intrin.c \
 	lp_setup_line.c \
 	lp_setup_point.c \
 	lp_setup_tri.c \
@@ -38,6 +36,7 @@ C_SOURCES = \
 	lp_state_clip.c \
 	lp_state_derived.c \
 	lp_state_fs.c \
+	lp_state_setup.c \
 	lp_state_gs.c \
 	lp_state_rasterizer.c \
 	lp_state_sampler.c \
@@ -63,14 +62,14 @@ PROGS := lp_test_format	\
 # Need this for the lp_test_*.o files
 CLEAN_EXTRA = *.o
 
+include ../../Makefile.template
+
 lp_test_sincos.o : sse_mathfun.h
 
 PROGS_DEPS := ../../auxiliary/libgallium.a
 
-include ../../Makefile.template
-
 lp_tile_soa.c: lp_tile_soa.py ../../auxiliary/util/u_format_parse.py ../../auxiliary/util/u_format_pack.py ../../auxiliary/util/u_format.csv
-	python lp_tile_soa.py ../../auxiliary/util/u_format.csv > $@
+	$(PYTHON2) $(PYTHON_FLAGS) lp_tile_soa.py ../../auxiliary/util/u_format.csv > $@
 
 LDFLAGS += $(LLVM_LDFLAGS)
 LIBS += -L../../auxiliary/ -lgallium libllvmpipe.a $(LLVM_LIBS) $(GL_LIB_DEPS)

src/gallium/drivers/llvmpipe/SConscript

@@ -27,13 +27,7 @@ env.Depends('lp_tile_soa.c', [
 ])
 
 
-# Only enable SSSE3 for lp_tile_soa_sse3.c
+lp_tile_soa_os = env.SharedObject('lp_tile_soa.c')
-ssse3_env = env.Clone()
-if env['gcc'] \
-   and distutils.version.LooseVersion(env['CCVERSION']) >= distutils.version.LooseVersion('4.3') \
-   and env['machine'] in ('x86', 'x86_64') :
-    ssse3_env.Append(CCFLAGS = ['-mssse3'])
-lp_tile_soa_os = ssse3_env.SharedObject('lp_tile_soa.c')
 
 
 llvmpipe = env.ConvenienceLibrary(
@@ -64,13 +58,12 @@ llvmpipe = env.ConvenienceLibrary(
 		'lp_setup_line.c',
 		'lp_setup_point.c',
 		'lp_setup_tri.c',
-		'lp_setup_coef.c',
-		'lp_setup_coef_intrin.c',
 		'lp_setup_vbuf.c',
 		'lp_state_blend.c',
 		'lp_state_clip.c',
 		'lp_state_derived.c',
 		'lp_state_fs.c',
+		'lp_state_setup.c',
 		'lp_state_gs.c',
 		'lp_state_rasterizer.c',
 		'lp_state_sampler.c',

src/gallium/drivers/llvmpipe/lp_bld_alpha.c

@@ -48,7 +48,8 @@ lp_build_alpha_test(LLVMBuilderRef builder,
                     struct lp_type type,
                     struct lp_build_mask_context *mask,
                     LLVMValueRef alpha,
-                    LLVMValueRef ref)
+                    LLVMValueRef ref,
+                    boolean do_branch)
 {
    struct lp_build_context bld;
    LLVMValueRef test;
@@ -60,4 +61,7 @@ lp_build_alpha_test(LLVMBuilderRef builder,
    lp_build_name(test, "alpha_mask");
 
    lp_build_mask_update(mask, test);
+
+   if (do_branch)
+      lp_build_mask_check(mask);
 }

src/gallium/drivers/llvmpipe/lp_bld_alpha.h

@@ -48,7 +48,8 @@ lp_build_alpha_test(LLVMBuilderRef builder,
                     struct lp_type type,
                     struct lp_build_mask_context *mask,
                     LLVMValueRef alpha,
-                    LLVMValueRef ref);
+                    LLVMValueRef ref,
+                    boolean do_branch);
 
 
 #endif /* !LP_BLD_ALPHA_H */

src/gallium/drivers/llvmpipe/lp_bld_depth.c

@@ -1,6 +1,6 @@
 /**************************************************************************
  *
- * Copyright 2009 VMware, Inc.
+ * Copyright 2009-2010 VMware, Inc.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -53,15 +53,8 @@
  *  ... ... ... ... ... ... ... ... ...
  *
  *
- * Stencil test:
- * Two-sided stencil test is supported but probably not as efficient as
- * it could be.  Currently, we use if/then/else constructs to do the
- * operations for front vs. back-facing polygons.  We could probably do
- * both the front and back arithmetic then use a Select() instruction to
- * choose the result depending on polyon orientation.  We'd have to
- * measure performance both ways and see which is better.
- *
  * @author Jose Fonseca <jfonseca@vmware.com>
+ * @author Brian Paul <jfonseca@vmware.com>
  */
 
 #include "pipe/p_state.h"
@@ -71,6 +64,7 @@
 #include "gallivm/lp_bld_arit.h"
 #include "gallivm/lp_bld_bitarit.h"
 #include "gallivm/lp_bld_const.h"
+#include "gallivm/lp_bld_conv.h"
 #include "gallivm/lp_bld_logic.h"
 #include "gallivm/lp_bld_flow.h"
 #include "gallivm/lp_bld_intr.h"
@@ -310,8 +304,13 @@ lp_depth_type(const struct util_format_description *format_desc,
    }
    else if(format_desc->channel[swizzle].type == UTIL_FORMAT_TYPE_UNSIGNED) {
       assert(format_desc->block.bits <= 32);
-      if(format_desc->channel[swizzle].normalized)
+      assert(format_desc->channel[swizzle].normalized);
-         type.norm = TRUE;
+      if (format_desc->channel[swizzle].size < format_desc->block.bits) {
+         /* Prefer signed integers when possible, as SSE has less support
+          * for unsigned comparison;
+          */
+         type.sign = TRUE;
+      }
    }
    else
       assert(0);
@@ -333,7 +332,7 @@ lp_depth_type(const struct util_format_description *format_desc,
  */
 static boolean
 get_z_shift_and_mask(const struct util_format_description *format_desc,
-                     unsigned *shift, unsigned *mask)
+                     unsigned *shift, unsigned *width, unsigned *mask)
 {
    const unsigned total_bits = format_desc->block.bits;
    unsigned z_swizzle;
@@ -349,12 +348,14 @@ get_z_shift_and_mask(const struct util_format_description *format_desc,
    if (z_swizzle == UTIL_FORMAT_SWIZZLE_NONE)
       return FALSE;
 
+   *width = format_desc->channel[z_swizzle].size;
+
    padding_right = 0;
    for (chan = 0; chan < z_swizzle; ++chan)
       padding_right += format_desc->channel[chan].size;
 
    padding_left =
-      total_bits - (padding_right + format_desc->channel[z_swizzle].size);
+      total_bits - (padding_right + *width);
 
    if (padding_left || padding_right) {
       unsigned long long mask_left = (1ULL << (total_bits - padding_left)) - 1;
@@ -365,7 +366,7 @@ get_z_shift_and_mask(const struct util_format_description *format_desc,
       *mask = 0xffffffff;
    }
 
-   *shift = padding_left;
+   *shift = padding_right;
 
    return TRUE;
 }
@@ -409,7 +410,7 @@ get_s_shift_and_mask(const struct util_format_description *format_desc,
  * \param maskvalue is the depth test mask.
  * \param counter is a pointer of the uint32 counter.
  */
-static void
+void
 lp_build_occlusion_count(LLVMBuilderRef builder,
                          struct lp_type type,
                          LLVMValueRef maskvalue,
@@ -446,33 +447,58 @@ lp_build_occlusion_count(LLVMBuilderRef builder,
  * \param format_desc  description of the depth/stencil surface
  * \param mask  the alive/dead pixel mask for the quad (vector)
  * \param stencil_refs  the front/back stencil ref values (scalar)
- * \param z_src  the incoming depth/stencil values (a 2x2 quad)
+ * \param z_src  the incoming depth/stencil values (a 2x2 quad, float32)
  * \param zs_dst_ptr  pointer to depth/stencil values in framebuffer
- * \param facing  contains float value indicating front/back facing polygon
+ * \param facing  contains boolean value indicating front/back facing polygon
  */
 void
 lp_build_depth_stencil_test(LLVMBuilderRef builder,
                             const struct pipe_depth_state *depth,
                             const struct pipe_stencil_state stencil[2],
-                            struct lp_type type,
+                            struct lp_type z_src_type,
                             const struct util_format_description *format_desc,
                             struct lp_build_mask_context *mask,
                             LLVMValueRef stencil_refs[2],
                             LLVMValueRef z_src,
                             LLVMValueRef zs_dst_ptr,
                             LLVMValueRef face,
-                            LLVMValueRef counter)
+                            LLVMValueRef *zs_value,
+                            boolean do_branch)
 {
-   struct lp_build_context bld;
+   struct lp_type z_type;
-   struct lp_build_context sbld;
+   struct lp_build_context z_bld;
+   struct lp_build_context s_bld;
    struct lp_type s_type;
+   unsigned z_shift = 0, z_width = 0, z_mask = 0;
    LLVMValueRef zs_dst, z_dst = NULL;
    LLVMValueRef stencil_vals = NULL;
    LLVMValueRef z_bitmask = NULL, stencil_shift = NULL;
    LLVMValueRef z_pass = NULL, s_pass_mask = NULL;
-   LLVMValueRef orig_mask = mask->value;
+   LLVMValueRef orig_mask = lp_build_mask_value(mask);
    LLVMValueRef front_facing = NULL;
 
+
+   /*
+    * Depths are expected to be between 0 and 1, even if they are stored in
+    * floats. Setting these bits here will ensure that the lp_build_conv() call
+    * below won't try to unnecessarily clamp the incoming values.
+    */
+   if(z_src_type.floating) {
+      z_src_type.sign = FALSE;
+      z_src_type.norm = TRUE;
+   }
+   else {
+      assert(!z_src_type.sign);
+      assert(z_src_type.norm);
+   }
+
+   /* Pick the depth type. */
+   z_type = lp_depth_type(format_desc, z_src_type.width*z_src_type.length);
+
+   /* FIXME: Cope with a depth test type with a different bit width. */
+   assert(z_type.width == z_src_type.width);
+   assert(z_type.length == z_src_type.length);
+
    /* Sanity checking */
    {
       const unsigned z_swizzle = format_desc->swizzle[0];
@@ -493,8 +519,8 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
       }
 
       assert(z_swizzle < 4);
-      assert(format_desc->block.bits == type.width);
+      assert(format_desc->block.bits == z_type.width);
-      if (type.floating) {
+      if (z_type.floating) {
          assert(z_swizzle == 0);
          assert(format_desc->channel[z_swizzle].type ==
                 UTIL_FORMAT_TYPE_FLOAT);
@@ -505,54 +531,56 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
          assert(format_desc->channel[z_swizzle].type ==
                 UTIL_FORMAT_TYPE_UNSIGNED);
          assert(format_desc->channel[z_swizzle].normalized);
-         assert(!type.fixed);
+         assert(!z_type.fixed);
-         assert(!type.sign);
-         assert(type.norm);
       }
    }
 
 
    /* Setup build context for Z vals */
-   lp_build_context_init(&bld, builder, type);
+   lp_build_context_init(&z_bld, builder, z_type);
 
    /* Setup build context for stencil vals */
-   s_type = lp_type_int_vec(type.width);
+   s_type = lp_type_int_vec(z_type.width);
-   lp_build_context_init(&sbld, builder, s_type);
+   lp_build_context_init(&s_bld, builder, s_type);
 
    /* Load current z/stencil value from z/stencil buffer */
+   zs_dst_ptr = LLVMBuildBitCast(builder,
+                                 zs_dst_ptr,
+                                 LLVMPointerType(z_bld.vec_type, 0), "");
    zs_dst = LLVMBuildLoad(builder, zs_dst_ptr, "");
 
-   lp_build_name(zs_dst, "zsbufval");
+   lp_build_name(zs_dst, "zs_dst");
 
 
    /* Compute and apply the Z/stencil bitmasks and shifts.
     */
    {
-      unsigned z_shift, z_mask;
       unsigned s_shift, s_mask;
 
-      if (get_z_shift_and_mask(format_desc, &z_shift, &z_mask)) {
+      if (get_z_shift_and_mask(format_desc, &z_shift, &z_width, &z_mask)) {
-         if (z_shift) {
-            LLVMValueRef shift = lp_build_const_int_vec(type, z_shift);
-            z_src = LLVMBuildLShr(builder, z_src, shift, "");
-         }
-
          if (z_mask != 0xffffffff) {
-            LLVMValueRef mask = lp_build_const_int_vec(type, z_mask);
+            z_bitmask = lp_build_const_int_vec(z_type, z_mask);
-            z_src = LLVMBuildAnd(builder, z_src, mask, "");
-            z_dst = LLVMBuildAnd(builder, zs_dst, mask, "");
-            z_bitmask = mask;  /* used below */
-         }
-         else {
-            z_dst = zs_dst;
          }
 
-         lp_build_name(z_dst, "zsbuf.z");
+         /*
+          * Align the framebuffer Z 's LSB to the right.
+          */
+         if (z_shift) {
+            LLVMValueRef shift = lp_build_const_int_vec(z_type, z_shift);
+            z_dst = LLVMBuildLShr(builder, zs_dst, shift, "z_dst");
+         } else if (z_bitmask) {
+	    /* TODO: Instead of loading a mask from memory and ANDing, it's
+	     * probably faster to just shake the bits with two shifts. */
+            z_dst = LLVMBuildAnd(builder, zs_dst, z_bitmask, "z_dst");
+         } else {
+            z_dst = zs_dst;
+            lp_build_name(z_dst, "z_dst");
+         }
       }
 
       if (get_s_shift_and_mask(format_desc, &s_shift, &s_mask)) {
          if (s_shift) {
-            LLVMValueRef shift = lp_build_const_int_vec(type, s_shift);
+            LLVMValueRef shift = lp_build_const_int_vec(s_type, s_shift);
             stencil_vals = LLVMBuildLShr(builder, zs_dst, shift, "");
             stencil_shift = shift;  /* used below */
          }
@@ -561,48 +589,85 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
          }
 
          if (s_mask != 0xffffffff) {
-            LLVMValueRef mask = lp_build_const_int_vec(type, s_mask);
+            LLVMValueRef mask = lp_build_const_int_vec(s_type, s_mask);
             stencil_vals = LLVMBuildAnd(builder, stencil_vals, mask, "");
          }
 
-         lp_build_name(stencil_vals, "stencil");
+         lp_build_name(stencil_vals, "s_dst");
       }
    }
 
    if (stencil[0].enabled) {
 
       if (face) {
-         LLVMValueRef zero = LLVMConstReal(LLVMFloatType(), 0.0);
+         LLVMValueRef zero = LLVMConstInt(LLVMInt32Type(), 0, 0);
 
-         /* front_facing = face > 0.0 ? ~0 : 0 */
+         /* front_facing = face != 0 ? ~0 : 0 */
-         front_facing = LLVMBuildFCmp(builder, LLVMRealUGT, face, zero, "");
+         front_facing = LLVMBuildICmp(builder, LLVMIntNE, face, zero, "");
          front_facing = LLVMBuildSExt(builder, front_facing,
-                                      LLVMIntType(bld.type.length*bld.type.width),
+                                      LLVMIntType(s_bld.type.length*s_bld.type.width),
                                       "");
          front_facing = LLVMBuildBitCast(builder, front_facing,
-                                         bld.int_vec_type, "");
+                                         s_bld.int_vec_type, "");
       }
 
       /* convert scalar stencil refs into vectors */
-      stencil_refs[0] = lp_build_broadcast_scalar(&bld, stencil_refs[0]);
+      stencil_refs[0] = lp_build_broadcast_scalar(&s_bld, stencil_refs[0]);
-      stencil_refs[1] = lp_build_broadcast_scalar(&bld, stencil_refs[1]);
+      stencil_refs[1] = lp_build_broadcast_scalar(&s_bld, stencil_refs[1]);
 
-      s_pass_mask = lp_build_stencil_test(&sbld, stencil,
+      s_pass_mask = lp_build_stencil_test(&s_bld, stencil,
                                           stencil_refs, stencil_vals,
                                           front_facing);
 
       /* apply stencil-fail operator */
       {
-         LLVMValueRef s_fail_mask = lp_build_andnot(&bld, orig_mask, s_pass_mask);
+         LLVMValueRef s_fail_mask = lp_build_andnot(&s_bld, orig_mask, s_pass_mask);
-         stencil_vals = lp_build_stencil_op(&sbld, stencil, S_FAIL_OP,
+         stencil_vals = lp_build_stencil_op(&s_bld, stencil, S_FAIL_OP,
                                             stencil_refs, stencil_vals,
                                             s_fail_mask, front_facing);
       }
    }
 
    if (depth->enabled) {
+      /*
+       * Convert fragment Z to the desired type, aligning the LSB to the right.
+       */
+
+      assert(z_type.width == z_src_type.width);
+      assert(z_type.length == z_src_type.length);
+      assert(lp_check_value(z_src_type, z_src));
+      if (z_src_type.floating) {
+         /*
+          * Convert from floating point values
+          */
+
+         if (!z_type.floating) {
+            z_src = lp_build_clamped_float_to_unsigned_norm(builder,
+                                                            z_src_type,
+                                                            z_width,
+                                                            z_src);
+         }
+      } else {
+         /*
+          * Convert from unsigned normalized values.
+          */
+
+         assert(!z_src_type.sign);
+         assert(!z_src_type.fixed);
+         assert(z_src_type.norm);
+         assert(!z_type.floating);
+         if (z_src_type.width > z_width) {
+            LLVMValueRef shift = lp_build_const_int_vec(z_src_type,
+                                                        z_src_type.width - z_width);
+            z_src = LLVMBuildLShr(builder, z_src, shift, "");
+         }
+      }
+      assert(lp_check_value(z_type, z_src));
+
+      lp_build_name(z_src, "z_src");
+
       /* compare src Z to dst Z, returning 'pass' mask */
-      z_pass = lp_build_cmp(&bld, depth->func, z_src, z_dst);
+      z_pass = lp_build_cmp(&z_bld, depth->func, z_src, z_dst);
 
       if (!stencil[0].enabled) {
          /* We can potentially skip all remaining operations here, but only
@@ -610,28 +675,28 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
           * buffer values.  Don't need to update Z buffer values.
           */
          lp_build_mask_update(mask, z_pass);
+
+         if (do_branch) {
+            lp_build_mask_check(mask);
+            do_branch = FALSE;
+         }
       }
 
       if (depth->writemask) {
-         LLVMValueRef zselectmask = mask->value;
+         LLVMValueRef zselectmask;
 
          /* mask off bits that failed Z test */
-         zselectmask = LLVMBuildAnd(builder, zselectmask, z_pass, "");
+         zselectmask = LLVMBuildAnd(builder, orig_mask, z_pass, "");
 
          /* mask off bits that failed stencil test */
          if (s_pass_mask) {
             zselectmask = LLVMBuildAnd(builder, zselectmask, s_pass_mask, "");
          }
 
-         /* if combined Z/stencil format, mask off the stencil bits */
-         if (z_bitmask) {
-            zselectmask = LLVMBuildAnd(builder, zselectmask, z_bitmask, "");
-         }
-
          /* Mix the old and new Z buffer values.
-          * z_dst[i] = (zselectmask[i] & z_src[i]) | (~zselectmask[i] & z_dst[i])
+          * z_dst[i] = zselectmask[i] ? z_src[i] : z_dst[i]
           */
-         z_dst = lp_build_select_bitwise(&bld, zselectmask, z_src, z_dst);
+         z_dst = lp_build_select(&z_bld, zselectmask, z_src, z_dst);
       }
 
       if (stencil[0].enabled) {
@@ -639,14 +704,14 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
          LLVMValueRef z_fail_mask, z_pass_mask;
 
          /* apply Z-fail operator */
-         z_fail_mask = lp_build_andnot(&bld, orig_mask, z_pass);
+         z_fail_mask = lp_build_andnot(&z_bld, orig_mask, z_pass);
-         stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_FAIL_OP,
+         stencil_vals = lp_build_stencil_op(&s_bld, stencil, Z_FAIL_OP,
                                             stencil_refs, stencil_vals,
                                             z_fail_mask, front_facing);
 
          /* apply Z-pass operator */
-         z_pass_mask = LLVMBuildAnd(bld.builder, orig_mask, z_pass, "");
+         z_pass_mask = LLVMBuildAnd(z_bld.builder, orig_mask, z_pass, "");
-         stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_PASS_OP,
+         stencil_vals = lp_build_stencil_op(&s_bld, stencil, Z_PASS_OP,
                                             stencil_refs, stencil_vals,
                                             z_pass_mask, front_facing);
       }
@@ -655,17 +720,19 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
       /* No depth test: apply Z-pass operator to stencil buffer values which
        * passed the stencil test.
        */
-      s_pass_mask = LLVMBuildAnd(bld.builder, orig_mask, s_pass_mask, "");
+      s_pass_mask = LLVMBuildAnd(s_bld.builder, orig_mask, s_pass_mask, "");
-      stencil_vals = lp_build_stencil_op(&sbld, stencil, Z_PASS_OP,
+      stencil_vals = lp_build_stencil_op(&s_bld, stencil, Z_PASS_OP,
                                          stencil_refs, stencil_vals,
                                          s_pass_mask, front_facing);
    }
 
-   /* The Z bits are already in the right place but we may need to shift the
+   /* Put Z and ztencil bits in the right place */
-    * stencil bits before ORing Z with Stencil to make the final pixel value.
+   if (z_dst && z_shift) {
-    */
+      LLVMValueRef shift = lp_build_const_int_vec(z_type, z_shift);
+      z_dst = LLVMBuildShl(builder, z_dst, shift, "");
+   }
    if (stencil_vals && stencil_shift)
-      stencil_vals = LLVMBuildShl(bld.builder, stencil_vals,
+      stencil_vals = LLVMBuildShl(s_bld.builder, stencil_vals,
                                   stencil_shift, "");
 
    /* Finally, merge/store the z/stencil values */
@@ -673,13 +740,13 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
        (stencil[0].enabled && stencil[0].writemask)) {
 
       if (z_dst && stencil_vals)
-         zs_dst = LLVMBuildOr(bld.builder, z_dst, stencil_vals, "");
+         zs_dst = LLVMBuildOr(z_bld.builder, z_dst, stencil_vals, "");
       else if (z_dst)
          zs_dst = z_dst;
       else
          zs_dst = stencil_vals;
 
-      LLVMBuildStore(builder, zs_dst, zs_dst_ptr);
+      *zs_value = zs_dst;
    }
 
    if (s_pass_mask)
@@ -688,6 +755,47 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
    if (depth->enabled && stencil[0].enabled)
       lp_build_mask_update(mask, z_pass);
 
-   if (counter)
+   if (do_branch)
-      lp_build_occlusion_count(builder, type, mask->value, counter);
+      lp_build_mask_check(mask);
+
+}
+
+
+void
+lp_build_depth_write(LLVMBuilderRef builder,
+                     const struct util_format_description *format_desc,
+                     LLVMValueRef zs_dst_ptr,
+                     LLVMValueRef zs_value)
+{
+   zs_dst_ptr = LLVMBuildBitCast(builder, zs_dst_ptr,
+                                 LLVMPointerType(LLVMTypeOf(zs_value), 0), "");
+
+   LLVMBuildStore(builder, zs_value, zs_dst_ptr);
+}
+
+
+void
+lp_build_deferred_depth_write(LLVMBuilderRef builder,
+                              struct lp_type z_src_type,
+                              const struct util_format_description *format_desc,
+                              struct lp_build_mask_context *mask,
+                              LLVMValueRef zs_dst_ptr,
+                              LLVMValueRef zs_value)
+{
+   struct lp_type z_type;
+   struct lp_build_context z_bld;
+   LLVMValueRef z_dst;
+
+   /* XXX: pointlessly redo type logic:
+    */
+   z_type = lp_depth_type(format_desc, z_src_type.width*z_src_type.length);
+   lp_build_context_init(&z_bld, builder, z_type);
+
+   zs_dst_ptr = LLVMBuildBitCast(builder, zs_dst_ptr,
+                                 LLVMPointerType(z_bld.vec_type, 0), "");
+
+   z_dst = LLVMBuildLoad(builder, zs_dst_ptr, "zsbufval");
+   z_dst = lp_build_select(&z_bld, lp_build_mask_value(mask), zs_value, z_dst);
+
+   LLVMBuildStore(builder, z_dst, zs_dst_ptr);
 }

src/gallium/drivers/llvmpipe/lp_bld_depth.h

@@ -61,7 +61,27 @@ lp_build_depth_stencil_test(LLVMBuilderRef builder,
                             LLVMValueRef zs_src,
                             LLVMValueRef zs_dst_ptr,
                             LLVMValueRef facing,
+                            LLVMValueRef *zs_value,
+                            boolean do_branch);
+
+void
+lp_build_depth_write(LLVMBuilderRef builder,
+                     const struct util_format_description *format_desc,
+                     LLVMValueRef zs_dst_ptr,
+                     LLVMValueRef zs_value);
+
+void
+lp_build_deferred_depth_write(LLVMBuilderRef builder,
+                              struct lp_type z_src_type,
+                              const struct util_format_description *format_desc,
+                              struct lp_build_mask_context *mask,
+                              LLVMValueRef zs_dst_ptr,
+                              LLVMValueRef zs_value);
+
+void
+lp_build_occlusion_count(LLVMBuilderRef builder,
+                         struct lp_type type,
+                         LLVMValueRef maskvalue,
                          LLVMValueRef counter);
 
-
 #endif /* !LP_BLD_DEPTH_H */

src/gallium/drivers/llvmpipe/lp_bld_interp.c

@@ -206,7 +206,7 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
             dadq2 = LLVMBuildFAdd(builder, dadq, dadq, "");
 
             /*
-             * a = a0 + x * dadx + y * dady
+             * a = a0 + (x * dadx + y * dady)
              */
 
             if (attrib == 0 && chan == 0) {
@@ -219,11 +219,11 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
                a = a0;
                if (interp != LP_INTERP_CONSTANT &&
                    interp != LP_INTERP_FACING) {
-                  LLVMValueRef tmp;
+                  LLVMValueRef ax, ay, axy;
-                  tmp = LLVMBuildFMul(builder, bld->x, dadx, "");
+                  ax = LLVMBuildFMul(builder, bld->x, dadx, "");
-                  a = LLVMBuildFAdd(builder, a, tmp, "");
+                  ay = LLVMBuildFMul(builder, bld->y, dady, "");
-                  tmp = LLVMBuildFMul(builder, bld->y, dady, "");
+                  axy = LLVMBuildFAdd(builder, ax, ay, "");
-                  a = LLVMBuildFAdd(builder, a, tmp, "");
+                  a = LLVMBuildFAdd(builder, a, axy, "");
                }
             }
 
@@ -272,7 +272,10 @@ coeffs_init(struct lp_build_interp_soa_context *bld,
  * This is called when we move from one quad to the next.
  */
 static void
-attribs_update(struct lp_build_interp_soa_context *bld, int quad_index)
+attribs_update(struct lp_build_interp_soa_context *bld,
+               int quad_index,
+               int start,
+               int end)
 {
    struct lp_build_context *coeff_bld = &bld->coeff_bld;
    LLVMValueRef shuffle = lp_build_const_int_vec(coeff_bld->type, quad_index);
@@ -282,7 +285,7 @@ attribs_update(struct lp_build_interp_soa_context *bld, int quad_index)
 
    assert(quad_index < 4);
 
-   for(attrib = 0; attrib < bld->num_attribs; ++attrib) {
+   for(attrib = start; attrib < end; ++attrib) {
       const unsigned mask = bld->mask[attrib];
       const unsigned interp = bld->interp[attrib];
       for(chan = 0; chan < NUM_CHANNELS; ++chan) {
@@ -350,6 +353,14 @@ attribs_update(struct lp_build_interp_soa_context *bld, int quad_index)
                }
 #endif
 
+               if (attrib == 0 && chan == 2) {
+                  /* FIXME: Depth values can exceed 1.0, due to the fact that
+                   * setup interpolation coefficients refer to (0,0) which causes
+                   * precision loss. So we must clamp to 1.0 here to avoid artifacts
+                   */
+                  a = lp_build_min(coeff_bld, a, coeff_bld->one);
+               }
+
                attrib_name(a, attrib, chan, "");
             }
             bld->attribs[attrib][chan] = a;
@@ -434,8 +445,6 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
    pos_init(bld, x0, y0);
 
    coeffs_init(bld, a0_ptr, dadx_ptr, dady_ptr);
-
-   attribs_update(bld, 0);
 }
 
 
@@ -443,10 +452,20 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
  * Advance the position and inputs to the given quad within the block.
  */
 void
-lp_build_interp_soa_update(struct lp_build_interp_soa_context *bld,
+lp_build_interp_soa_update_inputs(struct lp_build_interp_soa_context *bld,
                                   int quad_index)
 {
    assert(quad_index < 4);
 
-   attribs_update(bld, quad_index);
+   attribs_update(bld, quad_index, 1, bld->num_attribs);
 }
+
+void
+lp_build_interp_soa_update_pos(struct lp_build_interp_soa_context *bld,
+                                  int quad_index)
+{
+   assert(quad_index < 4);
+
+   attribs_update(bld, quad_index, 0, 1);
+}
+

src/gallium/drivers/llvmpipe/lp_bld_interp.h

@@ -46,7 +46,31 @@
 
 #include "tgsi/tgsi_exec.h"
 
-#include "lp_setup.h"
+/**
+ * Describes how to compute the interpolation coefficients (a0, dadx, dady)
+ * from the vertices passed into our triangle/line/point functions by the
+ * draw module.
+ *
+ * Vertices are treated as an array of float[4] values, indexed by
+ * src_index.
+ *
+ * LP_INTERP_COLOR is translated to either LP_INTERP_CONSTANT or
+ * LINEAR depending on flatshade state.
+ */
+enum lp_interp {
+   LP_INTERP_CONSTANT,
+   LP_INTERP_COLOR,
+   LP_INTERP_LINEAR,
+   LP_INTERP_PERSPECTIVE,
+   LP_INTERP_POSITION,
+   LP_INTERP_FACING
+};
+
+struct lp_shader_input {
+   ushort interp:4;       /* enum lp_interp */
+   ushort usage_mask:4;   /* bitmask of TGSI_WRITEMASK_x flags */
+   ushort src_index:8;    /* where to find values in incoming vertices */
+};
 
 
 struct lp_build_interp_soa_context
@@ -89,7 +113,11 @@ lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld,
                          LLVMValueRef y);
 
 void
-lp_build_interp_soa_update(struct lp_build_interp_soa_context *bld,
+lp_build_interp_soa_update_inputs(struct lp_build_interp_soa_context *bld,
+                           int quad_index);
+
+void
+lp_build_interp_soa_update_pos(struct lp_build_interp_soa_context *bld,
                            int quad_index);
 
 

src/gallium/drivers/llvmpipe/lp_context.c

@@ -82,6 +82,8 @@ static void llvmpipe_destroy( struct pipe_context *pipe )
       }
    }
 
+   lp_delete_setup_variants(llvmpipe);
+
    align_free( llvmpipe );
 }
 
@@ -108,6 +110,7 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    memset(llvmpipe, 0, sizeof *llvmpipe);
 
    make_empty_list(&llvmpipe->fs_variants_list);
+   make_empty_list(&llvmpipe->setup_variants_list);
 
    llvmpipe->pipe.winsys = screen->winsys;
    llvmpipe->pipe.screen = screen;

src/gallium/drivers/llvmpipe/lp_context.h

@@ -39,6 +39,7 @@
 #include "lp_jit.h"
 #include "lp_setup.h"
 #include "lp_state_fs.h"
+#include "lp_state_setup.h"
 
 
 struct llvmpipe_vbuf_render;
@@ -48,6 +49,7 @@ struct lp_fragment_shader;
 struct lp_vertex_shader;
 struct lp_blend_state;
 struct lp_setup_context;
+struct lp_setup_variant;
 struct lp_velems_state;
 
 struct llvmpipe_context {
@@ -105,12 +107,9 @@ struct llvmpipe_context {
    /** Which vertex shader output slot contains point size */
    int psize_slot;
 
-   /** Fragment shader input interpolation info */
-   unsigned num_inputs;
-   struct lp_shader_input inputs[PIPE_MAX_SHADER_INPUTS];
-
    /** The tiling engine */
    struct lp_setup_context *setup;
+   struct lp_setup_variant setup_variant;
 
    /** The primitive drawing context */
    struct draw_context *draw;
@@ -120,6 +119,9 @@ struct llvmpipe_context {
 
    struct lp_fs_variant_list_item fs_variants_list;
    unsigned nr_fs_variants;
+
+   struct lp_setup_variant_list_item setup_variants_list;
+   unsigned nr_setup_variants;
 };
 
 

src/gallium/drivers/llvmpipe/lp_flush.h

@@ -32,6 +32,7 @@
 
 struct pipe_context;
 struct pipe_fence_handle;
+struct pipe_resource;
 
 void
 llvmpipe_flush(struct pipe_context *pipe,

src/gallium/drivers/llvmpipe/lp_jit.c

@@ -36,7 +36,6 @@
 #include <llvm-c/Transforms/Scalar.h>
 
 #include "util/u_memory.h"
-#include "util/u_cpu_detect.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_debug.h"
 #include "lp_screen.h"
@@ -162,9 +161,6 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
 void
 lp_jit_screen_cleanup(struct llvmpipe_screen *screen)
 {
-   if(screen->engine)
-      LLVMDisposeExecutionEngine(screen->engine);
-
    if(screen->pass)
       LLVMDisposePassManager(screen->pass);
 }
@@ -190,13 +186,7 @@ lp_jit_screen_init(struct llvmpipe_screen *screen)
       LLVMAddCFGSimplificationPass(screen->pass);
       LLVMAddPromoteMemoryToRegisterPass(screen->pass);
       LLVMAddConstantPropagationPass(screen->pass);
-      if(util_cpu_caps.has_sse4_1) {
-         /* FIXME: There is a bug in this pass, whereby the combination of fptosi
-          * and sitofp (necessary for trunc/floor/ceil/round implementation)
-          * somehow becomes invalid code.
-          */
       LLVMAddInstructionCombiningPass(screen->pass);
-      }
       LLVMAddGVNPass(screen->pass);
    } else {
       /* We need at least this pass to prevent the backends to fail in

src/gallium/drivers/llvmpipe/lp_jit.h

@@ -144,7 +144,7 @@ typedef void
 (*lp_jit_frag_func)(const struct lp_jit_context *context,
                     uint32_t x,
                     uint32_t y,
-                    float facing,
+                    uint32_t facing,
                     const void *a0,
                     const void *dadx,
                     const void *dady,

src/gallium/drivers/llvmpipe/lp_limits.h

@@ -72,4 +72,14 @@
  */
 #define LP_MAX_SHADER_VARIANTS 1024
 
+/**
+ * Max number of setup variants that will be kept around.
+ *
+ * These are determined by the combination of the fragment shader
+ * input signature and a small amount of rasterization state (eg
+ * flatshading).  It is likely that many active fragment shaders will
+ * share the same setup variant.
+ */
+#define LP_MAX_SETUP_VARIANTS 64
+
 #endif /* LP_LIMITS_H */

src/gallium/drivers/llvmpipe/lp_rast.c

@@ -334,7 +334,7 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task,
 {
    const struct lp_scene *scene = task->scene;
    const struct lp_rast_shader_inputs *inputs = arg.shade_tile;
-   const struct lp_rast_state *state = inputs->state;
+   const struct lp_rast_state *state = task->state;
    struct lp_fragment_shader_variant *variant = state->variant;
    const unsigned tile_x = task->x, tile_y = task->y;
    unsigned x, y;
@@ -365,10 +365,10 @@ lp_rast_shade_tile(struct lp_rasterizer_task *task,
          BEGIN_JIT_CALL(state);
          variant->jit_function[RAST_WHOLE]( &state->jit_context,
                                             tile_x + x, tile_y + y,
-                                            inputs->facing,
+                                            inputs->frontfacing,
-                                            inputs->a0,
+                                            GET_A0(inputs),
-                                            inputs->dadx,
+                                            GET_DADX(inputs),
-                                            inputs->dady,
+                                            GET_DADY(inputs),
                                             color,
                                             depth,
                                             0xffff,
@@ -414,7 +414,7 @@ lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
                          unsigned x, unsigned y,
                          unsigned mask)
 {
-   const struct lp_rast_state *state = inputs->state;
+   const struct lp_rast_state *state = task->state;
    struct lp_fragment_shader_variant *variant = state->variant;
    const struct lp_scene *scene = task->scene;
    uint8_t *color[PIPE_MAX_COLOR_BUFS];
@@ -446,10 +446,10 @@ lp_rast_shade_quads_mask(struct lp_rasterizer_task *task,
    BEGIN_JIT_CALL(state);
    variant->jit_function[RAST_EDGE_TEST](&state->jit_context,
                                          x, y,
-                                         inputs->facing,
+                                         inputs->frontfacing,
-                                         inputs->a0,
+                                         GET_A0(inputs),
-                                         inputs->dadx,
+                                         GET_DADX(inputs),
-                                         inputs->dady,
+                                         GET_DADY(inputs),
                                          color,
                                          depth,
                                          mask,
@@ -490,6 +490,14 @@ lp_rast_end_query(struct lp_rasterizer_task *task,
 }
 
 
+void
+lp_rast_set_state(struct lp_rasterizer_task *task,
+                  const union lp_rast_cmd_arg arg)
+{
+   task->state = arg.state;
+}
+
+
 
 /**
  * Set top row and left column of the tile's pixels to white.  For debugging.
@@ -602,6 +610,7 @@ static lp_rast_cmd_func dispatch[LP_RAST_OP_MAX] =
    lp_rast_shade_tile_opaque,
    lp_rast_begin_query,
    lp_rast_end_query,
+   lp_rast_set_state,
 };