Merge branch 'dri2'
[mesa.git] / src / mesa / drivers / dri / savage / savage_xmesa.c
index 721047bb8933d5d2efbd1ccc947706592bbc77ab..013e88216fa70c190eeaead402e34eb2860b91f3 100644 (file)
  */
 
 
-#ifdef GLX_DIRECT_RENDERING
-
 #include <X11/Xlibint.h>
 #include <stdio.h>
 
 #include "savagecontext.h"
 #include "context.h"
 #include "matrix.h"
-
+#include "framebuffer.h"
+#include "renderbuffer.h"
 #include "simple_list.h"
 
+#include "utils.h"
+
+#include "extensions.h"
+
 #include "swrast/swrast.h"
 #include "swrast_setup/swrast_setup.h"
 #include "tnl/tnl.h"
-#include "array_cache/acache.h"
+#include "vbo/vbo.h"
 
 #include "tnl/t_pipeline.h"
 
 
 #include "savage_dri.h"
 
-#include "savagedma.h"
+#include "drirenderbuffer.h"
+#include "texmem.h"
 
+#define need_GL_ARB_multisample
+#define need_GL_ARB_texture_compression
+#define need_GL_ARB_vertex_buffer_object
+#define need_GL_EXT_secondary_color
+#include "extension_helper.h"
+
+#include "xmlpool.h"
+
+/* Driver-specific options
+ */
+#define SAVAGE_ENABLE_VDMA(def) \
+DRI_CONF_OPT_BEGIN(enable_vdma,bool,def) \
+       DRI_CONF_DESC(en,"Use DMA for vertex transfers") \
+       DRI_CONF_DESC(de,"Benutze DMA für Vertextransfers") \
+DRI_CONF_OPT_END
+#define SAVAGE_ENABLE_FASTPATH(def) \
+DRI_CONF_OPT_BEGIN(enable_fastpath,bool,def) \
+       DRI_CONF_DESC(en,"Use fast path for unclipped primitives") \
+       DRI_CONF_DESC(de,"Schneller Codepfad für ungeschnittene Polygone") \
+DRI_CONF_OPT_END
+#define SAVAGE_SYNC_FRAMES(def) \
+DRI_CONF_OPT_BEGIN(sync_frames,bool,def) \
+       DRI_CONF_DESC(en,"Synchronize with graphics hardware after each frame") \
+       DRI_CONF_DESC(de,"Synchronisiere nach jedem Frame mit Grafikhardware") \
+DRI_CONF_OPT_END
+
+/* Configuration
+ */
+PUBLIC const char __driConfigOptions[] =
+DRI_CONF_BEGIN
+    DRI_CONF_SECTION_QUALITY
+        DRI_CONF_TEXTURE_DEPTH(DRI_CONF_TEXTURE_DEPTH_FB)
+        DRI_CONF_COLOR_REDUCTION(DRI_CONF_COLOR_REDUCTION_DITHER)
+        DRI_CONF_FLOAT_DEPTH(false)
+    DRI_CONF_SECTION_END
+    DRI_CONF_SECTION_PERFORMANCE
+        SAVAGE_ENABLE_VDMA(true)
+        SAVAGE_ENABLE_FASTPATH(true)
+        SAVAGE_SYNC_FRAMES(false)
+        DRI_CONF_MAX_TEXTURE_UNITS(2,1,2)
+       DRI_CONF_TEXTURE_HEAPS(DRI_CONF_TEXTURE_HEAPS_ALL)
+        DRI_CONF_FORCE_S3TC_ENABLE(false)
+    DRI_CONF_SECTION_END
+    DRI_CONF_SECTION_DEBUG
+        DRI_CONF_NO_RAST(false)
+    DRI_CONF_SECTION_END
+DRI_CONF_END;
+static const GLuint __driNConfigOptions = 10;
+
+
+static const struct dri_debug_control debug_control[] =
+{
+    { "fall",  DEBUG_FALLBACKS },
+    { "api",   DEBUG_VERBOSE_API },
+    { "tex",   DEBUG_VERBOSE_TEX },
+    { "verb",  DEBUG_VERBOSE_MSG },
+    { "dma",   DEBUG_DMA },
+    { "state", DEBUG_STATE },
+    { NULL,    0 }
+};
 #ifndef SAVAGE_DEBUG
-int SAVAGE_DEBUG = (0
-/*                       | DEBUG_ALWAYS_SYNC  */
-/*               | DEBUG_VERBOSE_RING    */
-/*               | DEBUG_VERBOSE_OUTREG  */
-/*               | DEBUG_VERBOSE_MSG */
-/*               | DEBUG_NO_OUTRING */
-/*               | DEBUG_NO_OUTREG */
-/*               | DEBUG_VERBOSE_API */
-/*               | DEBUG_VERBOSE_2D */
-/*               | DEBUG_VERBOSE_DRI */
-/*               | DEBUG_VALIDATE_RING */
-/*               | DEBUG_VERBOSE_IOCTL */
-                 );
+int SAVAGE_DEBUG = 0;
 #endif
 
 
@@ -79,6 +131,47 @@ unsigned long time_sum=0;
 struct timeval tv_s1,tv_f1;
 #endif
 
+static const struct dri_extension card_extensions[] =
+{
+    { "GL_ARB_multisample",                GL_ARB_multisample_functions },
+    { "GL_ARB_multitexture",               NULL },
+    { "GL_ARB_texture_compression",        GL_ARB_texture_compression_functions },
+    { "GL_ARB_vertex_buffer_object",       GL_ARB_vertex_buffer_object_functions },
+    { "GL_EXT_stencil_wrap",               NULL },
+    { "GL_EXT_texture_lod_bias",           NULL },
+    { "GL_EXT_secondary_color",            GL_EXT_secondary_color_functions },
+    { NULL,                                NULL }
+};
+
+static const struct dri_extension s4_extensions[] =
+{
+    { "GL_ARB_texture_env_add",            NULL },
+    { "GL_ARB_texture_mirrored_repeat",    NULL },
+    { NULL,                                NULL }
+};
+
+extern struct tnl_pipeline_stage _savage_texnorm_stage;
+extern struct tnl_pipeline_stage _savage_render_stage;
+
+static const struct tnl_pipeline_stage *savage_pipeline[] = {
+
+   &_tnl_vertex_transform_stage,
+   &_tnl_normal_transform_stage,
+   &_tnl_lighting_stage,
+   &_tnl_fog_coordinate_stage,
+   &_tnl_texgen_stage,
+   &_tnl_texture_transform_stage,
+   &_savage_texnorm_stage,
+   &_savage_render_stage,
+   &_tnl_render_stage,
+   0,
+};
+
+
+static const __DRIextension *savageExtensions[] = {
+    &driReadDrawableExtension,
+};
+
 /* this is first function called in dirver*/
 
 static GLboolean
@@ -87,32 +180,11 @@ savageInitDriver(__DRIscreenPrivate *sPriv)
   savageScreenPrivate *savageScreen;
   SAVAGEDRIPtr         gDRIPriv = (SAVAGEDRIPtr)sPriv->pDevPriv;
 
-
-  /* Check the DRI version */
-   {
-      int major, minor, patch;
-      if (XF86DRIQueryVersion(sPriv->display, &major, &minor, &patch)) {
-         if (major != 4 || minor < 0) {
-           __driUtilMessage("savage DRI driver expected DRI version 4.0.x but got version %d.%d.%d", major, minor, patch);
-            return GL_FALSE;
-         }
-      }
-   }
-
-   /* Check that the DDX driver version is compatible */
-   if (sPriv->ddxMajor != 1 ||
-       sPriv->ddxMinor < 0) {
-      __driUtilMessage("savage DRI driver expected DDX driver version 1.0.x but got version %d.%d.%d", sPriv->ddxMajor, sPriv->ddxMinor, sPriv->ddxPatch);
+   if (sPriv->devPrivSize != sizeof(SAVAGEDRIRec)) {
+      fprintf(stderr,"\nERROR!  sizeof(SAVAGEDRIRec) does not match passed size from device driver\n");
       return GL_FALSE;
    }
 
-   /* Check that the DRM driver version is compatible */
-   if (sPriv->drmMajor != 1 ||
-       sPriv->drmMinor < 0) {
-      __driUtilMessage("savage DRI driver expected DRM driver version 1.1.x but got version %d.%d.%d", sPriv->drmMajor, sPriv->drmMinor, sPriv->drmPatch);
-      return GL_FALSE;
-   }
-       
    /* Allocate the private area */
    savageScreen = (savageScreenPrivate *)Xmalloc(sizeof(savageScreenPrivate));
    if (!savageScreen)
@@ -127,23 +199,19 @@ savageInitDriver(__DRIscreenPrivate *sPriv)
    savageScreen->mem=gDRIPriv->mem;
    savageScreen->cpp=gDRIPriv->cpp;
    savageScreen->zpp=gDRIPriv->zpp;
-   savageScreen->frontPitch=gDRIPriv->frontPitch;
-   savageScreen->frontOffset=gDRIPriv->frontOffset;
-   savageScreen->frontBitmapDesc = gDRIPriv->frontBitmapDesc;
-   
+
+   savageScreen->agpMode=gDRIPriv->agpMode;
+
+   savageScreen->bufferSize=gDRIPriv->bufferSize;
+
    if (gDRIPriv->cpp == 4) 
        savageScreen->frontFormat = DV_PF_8888;
    else
        savageScreen->frontFormat = DV_PF_565;
-
+   savageScreen->frontOffset=gDRIPriv->frontOffset;
    savageScreen->backOffset = gDRIPriv->backOffset; 
-   savageScreen->backBitmapDesc = gDRIPriv->backBitmapDesc; 
    savageScreen->depthOffset=gDRIPriv->depthOffset;
-   savageScreen->depthBitmapDesc = gDRIPriv->depthBitmapDesc; 
-#if 0   
-   savageScreen->backPitch = gDRIPriv->auxPitch;
-   savageScreen->backPitchBits = gDRIPriv->auxPitchBits;
-#endif   
+
    savageScreen->textureOffset[SAVAGE_CARD_HEAP] = 
                                    gDRIPriv->textureOffset;
    savageScreen->textureSize[SAVAGE_CARD_HEAP] = 
@@ -152,27 +220,54 @@ savageInitDriver(__DRIscreenPrivate *sPriv)
                                    gDRIPriv->logTextureGranularity;
 
    savageScreen->textureOffset[SAVAGE_AGP_HEAP] = 
-                                   gDRIPriv->agpTextures.handle;
+                                   gDRIPriv->agpTextureHandle;
    savageScreen->textureSize[SAVAGE_AGP_HEAP] = 
-                                   gDRIPriv->agpTextures.size;
+                                   gDRIPriv->agpTextureSize;
    savageScreen->logTextureGranularity[SAVAGE_AGP_HEAP] =
                                    gDRIPriv->logAgpTextureGranularity;
-   
-   savageScreen->back.handle = gDRIPriv->backbuffer;
-   savageScreen->back.size = gDRIPriv->backbufferSize;
-   savageScreen->back.map = 
-       (drmAddress)(((unsigned int)sPriv->pFB)+gDRIPriv->backOffset);
-   
-   savageScreen->depth.handle = gDRIPriv->depthbuffer;
-   savageScreen->depth.size = gDRIPriv->depthbufferSize;
 
-   savageScreen->depth.map = 
-              (drmAddress)(((unsigned int)sPriv->pFB)+gDRIPriv->depthOffset);
-   
-   savageScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
+   savageScreen->agpTextures.handle = gDRIPriv->agpTextureHandle;
+   savageScreen->agpTextures.size   = gDRIPriv->agpTextureSize;
+   if (gDRIPriv->agpTextureSize) {
+       if (drmMap(sPriv->fd, 
+                 savageScreen->agpTextures.handle,
+                 savageScreen->agpTextures.size,
+                 (drmAddress *)&(savageScreen->agpTextures.map)) != 0) {
+          Xfree(savageScreen);
+          sPriv->private = NULL;
+          return GL_FALSE;
+       }
+   } else
+       savageScreen->agpTextures.map = NULL;
 
    savageScreen->texVirtual[SAVAGE_CARD_HEAP] = 
-             (drmAddress)(((unsigned int)sPriv->pFB)+gDRIPriv->textureOffset);
+             (drmAddress)(((GLubyte *)sPriv->pFB)+gDRIPriv->textureOffset);
+   savageScreen->texVirtual[SAVAGE_AGP_HEAP] = 
+                        (drmAddress)(savageScreen->agpTextures.map);
+
+   savageScreen->aperture.handle = gDRIPriv->apertureHandle;
+   savageScreen->aperture.size   = gDRIPriv->apertureSize;
+   savageScreen->aperturePitch   = gDRIPriv->aperturePitch;
+   if (drmMap(sPriv->fd, 
+             savageScreen->aperture.handle, 
+             savageScreen->aperture.size, 
+             (drmAddress *)&savageScreen->aperture.map) != 0) 
+   {
+      Xfree(savageScreen);
+      sPriv->private = NULL;
+      return GL_FALSE;
+   }
+
+   savageScreen->bufs = drmMapBufs(sPriv->fd);
+
+   savageScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
+
+   /* parse information in __driConfigOptions */
+   driParseOptionInfo (&savageScreen->optionCache,
+                      __driConfigOptions, __driNConfigOptions);
+
+   sPriv->extensions = savageExtensions;
+
 #if 0
    savageDDFastPathInit();
    savageDDTrifuncInit();
@@ -188,7 +283,12 @@ savageDestroyScreen(__DRIscreenPrivate *sPriv)
 {
    savageScreenPrivate *savageScreen = (savageScreenPrivate *)sPriv->private;
 
-   
+   if (savageScreen->bufs)
+       drmUnmapBufs(savageScreen->bufs);
+
+   /* free all option information */
+   driDestroyOptionInfo (&savageScreen->optionCache);
+
    Xfree(savageScreen);
    sPriv->private = NULL;
 }
@@ -230,11 +330,10 @@ savageCreateContext( const __GLcontextModes *mesaVis,
    savageContextPtr imesa;
    __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
    struct dd_function_table functions;
-   SAVAGEDRIPtr         gDRIPriv = (SAVAGEDRIPtr)sPriv->pDevPriv;
    savageScreenPrivate *savageScreen = (savageScreenPrivate *)sPriv->private;
    drm_savage_sarea_t *saPriv=(drm_savage_sarea_t *)(((char*)sPriv->pSAREA)+
                                                 savageScreen->sarea_priv_offset);
-   GLuint maxTextureSize, minTextureSize, maxTextureLevels;
+   int textureSize[SAVAGE_NR_TEX_HEAPS];
    int i;
    imesa = (savageContextPtr)Xcalloc(sizeof(savageContext), 1);
    if (!imesa) {
@@ -258,45 +357,19 @@ savageCreateContext( const __GLcontextModes *mesaVis,
    }
    driContextPriv->driverPrivate = imesa;
 
-   if (savageScreen->chipset >= S3_SAVAGE4)
-       ctx->Const.MaxTextureUnits = 2;
-   else
-       ctx->Const.MaxTextureUnits = 1;
-   ctx->Const.MaxTextureImageUnits = ctx->Const.MaxTextureUnits;
-   ctx->Const.MaxTextureCoordUnits = ctx->Const.MaxTextureUnits;
+   imesa->cmdBuf.size = SAVAGE_CMDBUF_SIZE;
+   imesa->cmdBuf.base = imesa->cmdBuf.write =
+       malloc(SAVAGE_CMDBUF_SIZE * sizeof(drm_savage_cmd_header_t));
+   if (!imesa->cmdBuf.base)
+       return GL_FALSE;
 
-   /* Set the maximum texture size small enough that we can guarentee
-    * that all texture units can bind a maximal texture and have them
-    * in memory at once.
-    */
-   if (savageScreen->textureSize[SAVAGE_CARD_HEAP] >
-       savageScreen->textureSize[SAVAGE_AGP_HEAP]) {
-       maxTextureSize = savageScreen->textureSize[SAVAGE_CARD_HEAP];
-       minTextureSize = savageScreen->textureSize[SAVAGE_AGP_HEAP];
-   } else {
-       maxTextureSize = savageScreen->textureSize[SAVAGE_AGP_HEAP];
-       minTextureSize = savageScreen->textureSize[SAVAGE_CARD_HEAP];
-   }
-   if (ctx->Const.MaxTextureUnits == 2) {
-       /* How to distribute two maximum sized textures to two texture heaps?
-       * If the smaller heap is less then half the size of the bigger one
-       * then the maximum size is half the size of the bigger heap.
-       * Otherwise it's the size of the smaller heap. */
-       if (minTextureSize < maxTextureSize / 2)
-          maxTextureSize = maxTextureSize / 2;
-       else
-          maxTextureSize = minTextureSize;
-   }
-   for (maxTextureLevels = 1; maxTextureLevels <= 11; ++maxTextureLevels) {
-       GLuint size = 1 << maxTextureLevels;
-       size *= size * 4; /* 4 bytes per texel */
-       size *= 2; /* all mipmap levels together take roughly twice the size of
-                    the biggest level */
-       if (size > maxTextureSize)
-          break;
-   }
-   ctx->Const.MaxTextureLevels = maxTextureLevels;
-   assert (ctx->Const.MaxTextureLevels > 6); /*spec requires at least 64x64*/
+   /* Parse configuration files */
+   driParseConfigFiles (&imesa->optionCache, &savageScreen->optionCache,
+                        sPriv->myNum, "savage");
+
+   imesa->float_depth = driQueryOptionb(&imesa->optionCache, "float_depth") &&
+       savageScreen->chipset >= S3_SAVAGE4;
+   imesa->no_rast = driQueryOptionb(&imesa->optionCache, "no_rast");
 
 #if 0
    ctx->Const.MinLineWidth = 1.0;
@@ -319,116 +392,104 @@ savageCreateContext( const __GLcontextModes *mesaVis,
    
    /* DMA buffer */
 
-   /*The shadow pointer*/
-   imesa->shadowPointer = 
-     (volatile GLuint *)((((GLuint)(&saPriv->shadow_status)) + 31) & 0xffffffe0L) ;
-   /* here we use eventTag1 because eventTag0 is used by HWXvMC*/
-   imesa->eventTag1 = (volatile GLuint *)(imesa->shadowPointer + 6);
-   /*   imesa->eventTag1=(volatile GLuint *)(imesa->MMIO_BASE+0x48c04);*/
-   imesa->shadowCounter = MAX_SHADOWCOUNTER;
-   imesa->shadowStatus = GL_TRUE;/*Will judge by 2d message */
-
-   if (drmMap(sPriv->fd, 
-             gDRIPriv->registers.handle, 
-             gDRIPriv->registers.size, 
-             (drmAddress *)&(gDRIPriv->registers.map)) != 0) 
-   {
-      Xfree(savageScreen);
-      sPriv->private = NULL;
-      return GL_FALSE;
-   }
-   
-   if (drmMap(sPriv->fd, 
-             gDRIPriv->agpTextures.handle, 
-             gDRIPriv->agpTextures.size, 
-             (drmAddress *)&(gDRIPriv->agpTextures.map)) != 0) 
-   {
-      Xfree(savageScreen);
-      sPriv->private = NULL;
-      return GL_FALSE;
-   }
-
-/* agp texture*/
-   savageScreen->texVirtual[SAVAGE_AGP_HEAP] = 
-                        (drmAddress)(gDRIPriv->agpTextures.map);
-
-   
-
-   gDRIPriv->BCIcmdBuf.map = (drmAddress *)
-                           ((unsigned int)gDRIPriv->registers.map+0x00010000);
-      
-   imesa->MMIO_BASE = (GLuint)gDRIPriv->registers.map;
-   imesa->BCIBase= (GLuint)gDRIPriv->BCIcmdBuf.map;
-
-   savageScreen->aperture.handle = gDRIPriv->aperture.handle;
-   savageScreen->aperture.size   = gDRIPriv->aperture.size;
-   if (drmMap(sPriv->fd, 
-             savageScreen->aperture.handle, 
-             savageScreen->aperture.size, 
-             (drmAddress *)&savageScreen->aperture.map) != 0) 
-   {
-      Xfree(savageScreen);
-      sPriv->private = NULL;
-      return GL_FALSE;
-   } 
-   
-   
-
-
-   
    for(i=0;i<5;i++)
    {
-       imesa->apertureBase[i] = ((GLuint)savageScreen->aperture.map + 
-                                 0x01000000 * i );
-       
-   
+       imesa->apertureBase[i] = (GLubyte *)savageScreen->aperture.map + 
+          0x01000000 * i;
    }
    
-   {
-      volatile unsigned int * tmp;
-      
-      tmp=(volatile unsigned int *)(imesa->MMIO_BASE + 0x850C);
-      
-      
-      tmp=(volatile unsigned int *)(imesa->MMIO_BASE + 0x48C40);
-      
-   
-      tmp=(volatile unsigned int *)(imesa->MMIO_BASE + 0x48C44);
-
-   
-      tmp=(volatile unsigned int *)(imesa->MMIO_BASE + 0x48C48);
+   imesa->aperturePitch = savageScreen->aperturePitch;
 
-   
-   
-   }
-   
-   imesa->aperturePitch = gDRIPriv->aperturePitch;
-   
-   
    /* change texHeap initialize to support two kind of texture heap*/
    /* here is some parts of initialization, others in InitDriver() */
     
+   (void) memset( imesa->textureHeaps, 0, sizeof( imesa->textureHeaps ) );
+   make_empty_list( & imesa->swapped );
+
+   textureSize[SAVAGE_CARD_HEAP] = savageScreen->textureSize[SAVAGE_CARD_HEAP];
+   textureSize[SAVAGE_AGP_HEAP] = savageScreen->textureSize[SAVAGE_AGP_HEAP];
    imesa->lastTexHeap = savageScreen->texVirtual[SAVAGE_AGP_HEAP] ? 2 : 1;
+   switch(driQueryOptioni (&imesa->optionCache, "texture_heaps")) {
+   case DRI_CONF_TEXTURE_HEAPS_CARD: /* only use card memory, if available */
+       if (textureSize[SAVAGE_CARD_HEAP])
+          imesa->lastTexHeap = 1;
+       break;
+   case DRI_CONF_TEXTURE_HEAPS_GART: /* only use gart memory, if available */
+       if (imesa->lastTexHeap == 2 && textureSize[SAVAGE_AGP_HEAP])
+          textureSize[SAVAGE_CARD_HEAP] = 0;
+       break;
+   /*default: Nothing to do, use all available memory. */
+   }
    
-   /*allocate texHeap for multi-tex*/ 
-   {
-     int i;
-     
-     for(i=0;i<SAVAGE_NR_TEX_HEAPS;i++)
-     {
-       imesa->texHeap[i] = mmInit( 0, savageScreen->textureSize[i] );
-       make_empty_list(&imesa->TexObjList[i]);
-     }
-     
-     make_empty_list(&imesa->SwappedOut);
+   for (i = 0; i < imesa->lastTexHeap; i++) {
+       imesa->textureHeaps[i] = driCreateTextureHeap(
+          i, imesa,
+          textureSize[i],
+          11,                                  /* 2^11 = 2k alignment */
+          SAVAGE_NR_TEX_REGIONS,
+          (drmTextureRegionPtr)imesa->sarea->texList[i],
+           &imesa->sarea->texAge[i],
+           &imesa->swapped,
+           sizeof( savageTexObj ),
+           (destroy_texture_object_t *) savageDestroyTexObj );
+       /* If textureSize[i] == 0 textureHeaps[i] is NULL. This can happen
+       * if there is not enough card memory for a card texture heap. */
+       if (imesa->textureHeaps[i])
+          driSetTextureSwapCounterLocation( imesa->textureHeaps[i],
+                                            & imesa->c_textureSwaps );
+   }
+   imesa->texture_depth = driQueryOptioni (&imesa->optionCache,
+                                          "texture_depth");
+   if (imesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB)
+       imesa->texture_depth = ( savageScreen->cpp == 4 ) ?
+          DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16;
+
+   if (savageScreen->chipset >= S3_SAVAGE4)
+       ctx->Const.MaxTextureUnits = 2;
+   else
+       ctx->Const.MaxTextureUnits = 1;
+   if (driQueryOptioni(&imesa->optionCache, "texture_units") <
+       ctx->Const.MaxTextureUnits)
+       ctx->Const.MaxTextureUnits =
+          driQueryOptioni(&imesa->optionCache, "texture_units");
+   ctx->Const.MaxTextureImageUnits = ctx->Const.MaxTextureUnits;
+   ctx->Const.MaxTextureCoordUnits = ctx->Const.MaxTextureUnits;
+
+   driCalculateMaxTextureLevels( imesa->textureHeaps,
+                                imesa->lastTexHeap,
+                                & ctx->Const,
+                                4,
+                                11, /* max 2D texture size is 2048x2048 */
+                                0,  /* 3D textures unsupported. */
+                                0,  /* cube textures unsupported. */
+                                0,  /* texture rectangles unsupported. */
+                                12,
+                                GL_FALSE,
+                                0 );
+   if (ctx->Const.MaxTextureLevels <= 6) { /*spec requires at least 64x64*/
+       __driUtilMessage("Not enough texture memory. "
+                       "Falling back to indirect rendering.");
+       Xfree(imesa);
+       return GL_FALSE;
    }
 
-   imesa->hw_stencil = GL_FALSE;
-#if HW_STENCIL
    imesa->hw_stencil = mesaVis->stencilBits && mesaVis->depthBits == 24;
-#endif
    imesa->depth_scale = (imesa->savageScreen->zpp == 2) ?
-       (1.0F/0x10000):(1.0F/0x1000000);
+       (1.0F/0xffff):(1.0F/0xffffff);
+
+   imesa->bufferSize = savageScreen->bufferSize;
+   imesa->dmaVtxBuf.total = 0;
+   imesa->dmaVtxBuf.used = 0;
+   imesa->dmaVtxBuf.flushed = 0;
+
+   imesa->clientVtxBuf.total = imesa->bufferSize / 4;
+   imesa->clientVtxBuf.used = 0;
+   imesa->clientVtxBuf.flushed = 0;
+   imesa->clientVtxBuf.buf = (u_int32_t *)malloc(imesa->bufferSize);
+
+   imesa->vtxBuf = &imesa->clientVtxBuf;
+
+   imesa->firstElt = -1;
 
    /* Uninitialized vertex format. Force setting the vertex state in
     * savageRenderStart.
@@ -438,29 +499,44 @@ savageCreateContext( const __GLcontextModes *mesaVis,
    /* Utah stuff
     */
    imesa->new_state = ~0;
+   imesa->new_gl_state = ~0;
    imesa->RenderIndex = ~0;
    imesa->dirty = ~0;
    imesa->lostContext = GL_TRUE;
-   imesa->TextureMode = ctx->Texture.Unit[0].EnvMode;
    imesa->CurrentTexObj[0] = 0;
    imesa->CurrentTexObj[1] = 0;
-   imesa->texAge[SAVAGE_CARD_HEAP]=0;
-   imesa->texAge[SAVAGE_AGP_HEAP]=0;
 
    /* Initialize the software rasterizer and helper modules.
     */
    _swrast_CreateContext( ctx );
-   _ac_CreateContext( ctx );
+   _vbo_CreateContext( ctx );
    _tnl_CreateContext( ctx );
    
    _swsetup_CreateContext( ctx );
 
    /* Install the customized pipeline:
     */
-#if 0
    _tnl_destroy_pipeline( ctx );
    _tnl_install_pipeline( ctx, savage_pipeline );
-#endif
+
+   imesa->enable_fastpath = driQueryOptionb(&imesa->optionCache,
+                                           "enable_fastpath");
+   /* DRM versions before 2.1.3 would only render triangle lists. ELTS
+    * support was added in 2.2.0. */
+   if (imesa->enable_fastpath && sPriv->drm_version.minor < 2) {
+      fprintf (stderr,
+              "*** Disabling fast path because your DRM version is buggy "
+              "or doesn't\n*** support ELTS. You need at least Savage DRM "
+              "version 2.2.\n");
+      imesa->enable_fastpath = GL_FALSE;
+   }
+
+   if (!savageScreen->bufs || savageScreen->chipset == S3_SUPERSAVAGE)
+       imesa->enable_vdma = GL_FALSE;
+   else
+       imesa->enable_vdma = driQueryOptionb(&imesa->optionCache, "enable_vdma");
+
+   imesa->sync_frames = driQueryOptionb(&imesa->optionCache, "sync_frames");
 
    /* Configure swrast to match hardware characteristics:
     */
@@ -471,10 +547,23 @@ savageCreateContext( const __GLcontextModes *mesaVis,
 
    ctx->DriverCtx = (void *) imesa;
    imesa->glCtx = ctx;
-   if (savageDMAInit(imesa) == GL_FALSE)
-       return GL_FALSE;  
-   
-   savageDDExtensionsInit( ctx );
+
+#ifndef SAVAGE_DEBUG
+   SAVAGE_DEBUG = driParseDebugString( getenv( "SAVAGE_DEBUG" ),
+                                      debug_control );
+#endif
+
+   driInitExtensions( ctx, card_extensions, GL_TRUE );
+   if (savageScreen->chipset >= S3_SAVAGE4)
+       driInitExtensions( ctx, s4_extensions, GL_FALSE );
+   if (ctx->Mesa_DXTn ||
+       driQueryOptionb (&imesa->optionCache, "force_s3tc_enable")) {
+       _mesa_enable_extension( ctx, "GL_S3_s3tc" );
+       if (savageScreen->chipset >= S3_SAVAGE4)
+          /* This extension needs DXT3 and DTX5 support in hardware.
+           * Not available on Savage3D/MX/IX. */
+          _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
+   }
 
    savageDDInitStateFuncs( ctx );
    savageDDInitSpanFuncs( ctx );
@@ -493,25 +582,24 @@ static void
 savageDestroyContext(__DRIcontextPrivate *driContextPriv)
 {
    savageContextPtr imesa = (savageContextPtr) driContextPriv->driverPrivate;
+   GLuint i;
 
    assert (imesa); /* should never be NULL */
    if (imesa) {
-      savageTextureObjectPtr next_t, t;
+      savageFlushVertices(imesa);
+      savageReleaseIndexedVerts(imesa);
+      savageFlushCmdBuf(imesa, GL_TRUE); /* release DMA buffer */
+      WAIT_IDLE_EMPTY(imesa);
+
+      for (i = 0; i < imesa->lastTexHeap; i++)
+        driDestroyTextureHeap(imesa->textureHeaps[i]);
+
+      free(imesa->cmdBuf.base);
+      free(imesa->clientVtxBuf.buf);
 
-      /* update for multi-tex*/ 
-      {
-       int i;
-       for(i=0;i<SAVAGE_NR_TEX_HEAPS;i++)
-          foreach_s (t, next_t, &(imesa->TexObjList[i]))
-        savageDestroyTexObj(imesa, t);
-      }
-      foreach_s (t, next_t, &(imesa->SwappedOut))
-        savageDestroyTexObj(imesa, t);
-      /*free the dma buffer*/
-      savageDMAClose(imesa);
       _swsetup_DestroyContext(imesa->glCtx );
       _tnl_DestroyContext( imesa->glCtx );
-      _ac_DestroyContext( imesa->glCtx );
+      _vbo_DestroyContext( imesa->glCtx );
       _swrast_DestroyContext( imesa->glCtx );
 
       /* free the Mesa context */
@@ -523,27 +611,97 @@ savageDestroyContext(__DRIcontextPrivate *driContextPriv)
    }
 }
 
+
 static GLboolean
 savageCreateBuffer( __DRIscreenPrivate *driScrnPriv,
                    __DRIdrawablePrivate *driDrawPriv,
                    const __GLcontextModes *mesaVis,
                    GLboolean isPixmap)
 {
+   savageScreenPrivate *screen = (savageScreenPrivate *) driScrnPriv->private;
+
    if (isPixmap) {
       return GL_FALSE; /* not implemented */
    }
    else {
-#if HW_STENCIL
-       GLboolean swStencil = mesaVis->stencilBits > 0 && mesaVis->depthBits != 24;
-#else
-       GLboolean swStencil = mesaVis->stencilBits > 0;
-#endif
-      driDrawPriv->driverPrivate = (void *) 
-         _mesa_create_framebuffer(mesaVis,
-                                  GL_FALSE,  /* software depth buffer? */
-                                  swStencil,
-                                  mesaVis->accumRedBits > 0,
-                                  mesaVis->alphaBits > 0 );
+      GLboolean swStencil = mesaVis->stencilBits > 0 && mesaVis->depthBits != 24;
+      struct gl_framebuffer *fb = _mesa_create_framebuffer(mesaVis);
+      /*
+       * XXX: this value needs to be set according to the config file
+       * setting.  But we don't get that until we create a rendering
+       * context!!!!
+       */
+      GLboolean float_depth = GL_FALSE;
+
+      {
+         driRenderbuffer *frontRb
+            = driNewRenderbuffer(GL_RGBA,
+                                 (GLubyte *) screen->aperture.map
+                                 + 0x01000000 * TARGET_FRONT,
+                                 screen->cpp,
+                                 screen->frontOffset, screen->aperturePitch,
+                                 driDrawPriv);
+         savageSetSpanFunctions(frontRb, mesaVis, float_depth);
+         assert(frontRb->Base.Data);
+         _mesa_add_renderbuffer(fb, BUFFER_FRONT_LEFT, &frontRb->Base);
+      }
+
+      if (mesaVis->doubleBufferMode) {
+         driRenderbuffer *backRb
+            = driNewRenderbuffer(GL_RGBA,
+                                 (GLubyte *) screen->aperture.map
+                                 + 0x01000000 * TARGET_BACK,
+                                 screen->cpp,
+                                 screen->backOffset, screen->aperturePitch,
+                                 driDrawPriv);
+         savageSetSpanFunctions(backRb, mesaVis, float_depth);
+         assert(backRb->Base.Data);
+         _mesa_add_renderbuffer(fb, BUFFER_BACK_LEFT, &backRb->Base);
+      }
+
+      if (mesaVis->depthBits == 16) {
+         driRenderbuffer *depthRb
+            = driNewRenderbuffer(GL_DEPTH_COMPONENT16,
+                                 (GLubyte *) screen->aperture.map
+                                 + 0x01000000 * TARGET_DEPTH,
+                                 screen->zpp,
+                                 screen->depthOffset, screen->aperturePitch,
+                                 driDrawPriv);
+         savageSetSpanFunctions(depthRb, mesaVis, float_depth);
+         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
+      }
+      else if (mesaVis->depthBits == 24) {
+         driRenderbuffer *depthRb
+            = driNewRenderbuffer(GL_DEPTH_COMPONENT24,
+                                 (GLubyte *) screen->aperture.map
+                                 + 0x01000000 * TARGET_DEPTH,
+                                 screen->zpp,
+                                 screen->depthOffset, screen->aperturePitch,
+                                 driDrawPriv);
+         savageSetSpanFunctions(depthRb, mesaVis, float_depth);
+         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
+      }
+
+      if (mesaVis->stencilBits > 0 && !swStencil) {
+         driRenderbuffer *stencilRb
+            = driNewRenderbuffer(GL_STENCIL_INDEX8_EXT,
+                                 (GLubyte *) screen->aperture.map
+                                 + 0x01000000 * TARGET_DEPTH,
+                                 screen->zpp,
+                                 screen->depthOffset, screen->aperturePitch,
+                                 driDrawPriv);
+         savageSetSpanFunctions(stencilRb, mesaVis, float_depth);
+         _mesa_add_renderbuffer(fb, BUFFER_STENCIL, &stencilRb->Base);
+      }
+
+      _mesa_add_soft_renderbuffers(fb,
+                                   GL_FALSE, /* color */
+                                   GL_FALSE, /* depth */
+                                   swStencil,
+                                   mesaVis->accumRedBits > 0,
+                                   GL_FALSE, /* alpha */
+                                   GL_FALSE /* aux */);
+      driDrawPriv->driverPrivate = (void *) fb;
 
       return (driDrawPriv->driverPrivate != NULL);
    }
@@ -552,7 +710,7 @@ savageCreateBuffer( __DRIscreenPrivate *driScrnPriv,
 static void
 savageDestroyBuffer(__DRIdrawablePrivate *driDrawPriv)
 {
-   _mesa_destroy_framebuffer((GLframebuffer *) (driDrawPriv->driverPrivate));
+   _mesa_unreference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)));
 }
 
 #if 0
@@ -566,62 +724,41 @@ void XMesaSwapBuffers(__DRIdrawablePrivate *driDrawPriv)
 }
 #endif
 
-void savageXMesaSetFrontClipRects( savageContextPtr imesa )
-{
-   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
-
-   imesa->numClipRects = dPriv->numClipRects;
-   imesa->pClipRects = dPriv->pClipRects;
-   imesa->dirty |= SAVAGE_UPLOAD_CLIPRECTS;
-   imesa->drawX = dPriv->x;
-   imesa->drawY = dPriv->y;
-
-   savageEmitDrawingRectangle( imesa );
-}
-
 
-void savageXMesaSetBackClipRects( savageContextPtr imesa )
+void savageXMesaSetClipRects(savageContextPtr imesa)
 {
    __DRIdrawablePrivate *dPriv = imesa->driDrawable;
 
-   if (dPriv->numBackClipRects == 0) 
-   {
-
-
+   if ((dPriv->numBackClipRects == 0)
+       || (imesa->glCtx->DrawBuffer->_ColorDrawBufferMask[0] == BUFFER_BIT_FRONT_LEFT)) {
       imesa->numClipRects = dPriv->numClipRects;
       imesa->pClipRects = dPriv->pClipRects;
       imesa->drawX = dPriv->x;
       imesa->drawY = dPriv->y;
    } else {
-
-
       imesa->numClipRects = dPriv->numBackClipRects;
       imesa->pClipRects = dPriv->pBackClipRects;
       imesa->drawX = dPriv->backX;
       imesa->drawY = dPriv->backY;
    }
 
-   savageEmitDrawingRectangle( imesa );
-   imesa->dirty |= SAVAGE_UPLOAD_CLIPRECTS;
-
-
+   savageCalcViewport( imesa->glCtx );
 }
 
 
 static void savageXMesaWindowMoved( savageContextPtr imesa ) 
 {
+   __DRIdrawablePrivate *const drawable = imesa->driDrawable;
+   __DRIdrawablePrivate *const readable = imesa->driReadable;
+
    if (0)
       fprintf(stderr, "savageXMesaWindowMoved\n\n");
 
-   switch (imesa->glCtx->Color._DrawDestMask) {
-   case DD_FRONT_LEFT_BIT:
-      savageXMesaSetFrontClipRects( imesa );
-      break;
-   case DD_BACK_LEFT_BIT:
-      savageXMesaSetBackClipRects( imesa );
-      break;
-   default:
-       break;
+   savageXMesaSetClipRects(imesa);
+
+   driUpdateFramebufferSize(imesa->glCtx, drawable);
+   if (drawable != readable) {
+      driUpdateFramebufferSize(imesa->glCtx, readable);
    }
 }
 
@@ -636,6 +773,7 @@ savageUnbindContext(__DRIcontextPrivate *driContextPriv)
    return GL_TRUE;
 }
 
+#if 0
 static GLboolean
 savageOpenFullScreen(__DRIcontextPrivate *driContextPriv)
 {
@@ -661,7 +799,7 @@ savageCloseFullScreen(__DRIcontextPrivate *driContextPriv)
     
     if (driContextPriv) {
       savageContextPtr imesa = (savageContextPtr) driContextPriv->driverPrivate;
-      WAIT_IDLE_EMPTY;
+      WAIT_IDLE_EMPTY(imesa);
       imesa->IsFullScreen = GL_FALSE;   
       imesa->savageScreen->frontOffset = imesa->backup_frontOffset;
       imesa->savageScreen->backOffset = imesa->backup_backOffset;
@@ -669,7 +807,7 @@ savageCloseFullScreen(__DRIcontextPrivate *driContextPriv)
    }
     return GL_TRUE;
 }
-
+#endif
 
 static GLboolean
 savageMakeCurrent(__DRIcontextPrivate *driContextPriv,
@@ -677,26 +815,33 @@ savageMakeCurrent(__DRIcontextPrivate *driContextPriv,
                  __DRIdrawablePrivate *driReadPriv)
 {
    if (driContextPriv) {
-      savageContextPtr imesa = (savageContextPtr) driContextPriv->driverPrivate;
-      
+      savageContextPtr imesa
+         = (savageContextPtr) driContextPriv->driverPrivate;
+      struct gl_framebuffer *drawBuffer
+         = (GLframebuffer *) driDrawPriv->driverPrivate;
+      struct gl_framebuffer *readBuffer
+         = (GLframebuffer *) driReadPriv->driverPrivate;
+      driRenderbuffer *frontRb = (driRenderbuffer *)
+         drawBuffer->Attachment[BUFFER_FRONT_LEFT].Renderbuffer;
+      driRenderbuffer *backRb = (driRenderbuffer *)
+         drawBuffer->Attachment[BUFFER_BACK_LEFT].Renderbuffer;
+
+      assert(frontRb->Base.Data);
+      if (imesa->glCtx->Visual.doubleBufferMode) {
+         assert(backRb->Base.Data);
+      }
+
       imesa->driReadable = driReadPriv;
       imesa->driDrawable = driDrawPriv;
-      imesa->mesa_drawable = driDrawPriv;
       imesa->dirty = ~0;
       
-      _mesa_make_current2(imesa->glCtx,
-                          (GLframebuffer *) driDrawPriv->driverPrivate,
-                          (GLframebuffer *) driReadPriv->driverPrivate);
+      _mesa_make_current(imesa->glCtx, drawBuffer, readBuffer);
       
       savageXMesaWindowMoved( imesa );
-      
-      if (!imesa->glCtx->Viewport.Width)
-        _mesa_set_viewport(imesa->glCtx, 0, 0,
-                            driDrawPriv->w, driDrawPriv->h);
    }
    else 
    {
-      _mesa_make_current(NULL, NULL);
+      _mesa_make_current(NULL, NULL, NULL);
    }
    return GL_TRUE;
 }
@@ -704,12 +849,14 @@ savageMakeCurrent(__DRIcontextPrivate *driContextPriv,
 
 void savageGetLock( savageContextPtr imesa, GLuint flags ) 
 {
-   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+   __DRIdrawablePrivate *const drawable = imesa->driDrawable;
+   __DRIdrawablePrivate *const readable = imesa->driReadable;
    __DRIscreenPrivate *sPriv = imesa->driScreen;
    drm_savage_sarea_t *sarea = imesa->sarea;
    int me = imesa->hHWContext;
-   int stamp = dPriv->lastStamp; 
+   int stamp = drawable->lastStamp; 
    int heap;
+   unsigned int timestamp = 0;
 
   
 
@@ -727,75 +874,50 @@ void savageGetLock( savageContextPtr imesa, GLuint flags )
     * NOTE: This releases and regains the hw lock, so all state
     * checking must be done *after* this call:
     */
-   DRI_VALIDATE_DRAWABLE_INFO(sPriv, dPriv);           
+   DRI_VALIDATE_DRAWABLE_INFO(sPriv, drawable);
+   if (drawable != readable) {
+      DRI_VALIDATE_DRAWABLE_INFO(sPriv, readable);
+   }
 
 
-  
-
    /* If we lost context, need to dump all registers to hardware.
     * Note that we don't care about 2d contexts, even if they perform
     * accelerated commands, so the DRI locking in the X server is even
     * more broken than usual.
     */
    if (sarea->ctxOwner != me) {
-      imesa->dirty |= (SAVAGE_UPLOAD_CTX |
-                      SAVAGE_UPLOAD_CLIPRECTS |
+      imesa->dirty |= (SAVAGE_UPLOAD_LOCAL |
+                      SAVAGE_UPLOAD_GLOBAL |
+                      SAVAGE_UPLOAD_FOGTBL |
                       SAVAGE_UPLOAD_TEX0 |
-                      SAVAGE_UPLOAD_TEX1);
+                      SAVAGE_UPLOAD_TEX1 |
+                      SAVAGE_UPLOAD_TEXGLOBAL);
       imesa->lostContext = GL_TRUE;
       sarea->ctxOwner = me;
    }
 
-   /* Shared texture managment - if another client has played with
-    * texture space, figure out which if any of our textures have been
-    * ejected, and update our global LRU.
-    */
-   /*frank just for compiling,texAge,texList,AGP*/ 
-   
-   for(heap= 0 ;heap < imesa->lastTexHeap ; heap++)
-   {
-       if (sarea->texAge[heap] != imesa->texAge[heap]) {
-           int sz = 1 << (imesa->savageScreen->logTextureGranularity[heap]);
-           int idx, nr = 0;
+   for (heap = 0; heap < imesa->lastTexHeap; ++heap) {
+      /* If a heap was changed, update its timestamp. Do this before
+       * DRI_AGE_TEXTURES updates the local_age. */
+      if (imesa->textureHeaps[heap] &&
+         imesa->textureHeaps[heap]->global_age[0] >
+         imesa->textureHeaps[heap]->local_age) {
+        if (timestamp == 0)
+           timestamp = savageEmitEventLocked(imesa, 0);
+        imesa->textureHeaps[heap]->timestamp = timestamp;
+      }
+      DRI_AGE_TEXTURES( imesa->textureHeaps[heap] );
+   }
 
-      /* Have to go right round from the back to ensure stuff ends up
-       * LRU in our local list...
-       */
-           for (idx = sarea->texList[heap][SAVAGE_NR_TEX_REGIONS].prev ; 
-              idx != SAVAGE_NR_TEX_REGIONS && nr < SAVAGE_NR_TEX_REGIONS ; 
-              idx = sarea->texList[heap][idx].prev, nr++)
-           {
-              if (sarea->texList[heap][idx].age > imesa->texAge[heap])
-              {
-                  savageTexturesGone(imesa, heap ,idx * sz, sz, 
-                                     sarea->texList[heap][idx].in_use);
-              }
-           }
-
-           if (nr == SAVAGE_NR_TEX_REGIONS) 
-           {
-             savageTexturesGone(imesa, heap, 0, 
-                                imesa->savageScreen->textureSize[heap], 0);
-             savageResetGlobalLRU( imesa , heap );
-           }
-
-           imesa->dirty |= SAVAGE_UPLOAD_TEX0IMAGE;
-           imesa->dirty |= SAVAGE_UPLOAD_TEX1IMAGE;
-           imesa->texAge[heap] = sarea->texAge[heap];
-       }
-   } /* end of for loop */ 
-   
-   if (dPriv->lastStamp != stamp)
+   if (drawable->lastStamp != stamp) {
+      driUpdateFramebufferSize(imesa->glCtx, drawable);
       savageXMesaWindowMoved( imesa );
-
-  
-   
+   }
 }
 
 
 
 static const struct __DriverAPIRec savageAPI = {
-   savageInitDriver,
    savageDestroyScreen,
    savageCreateContext,
    savageDestroyContext,
@@ -803,25 +925,133 @@ static const struct __DriverAPIRec savageAPI = {
    savageDestroyBuffer,
    savageSwapBuffers,
    savageMakeCurrent,
-   savageUnbindContext,
-   savageOpenFullScreen,
-   savageCloseFullScreen
+   savageUnbindContext
 };
 
 
+static __GLcontextModes *
+savageFillInModes( unsigned pixel_bits, unsigned depth_bits,
+                  unsigned stencil_bits, GLboolean have_back_buffer )
+{
+    __GLcontextModes * modes;
+    __GLcontextModes * m;
+    unsigned num_modes;
+    unsigned depth_buffer_factor;
+    unsigned back_buffer_factor;
+    GLenum fb_format;
+    GLenum fb_type;
+
+    /* Right now GLX_SWAP_COPY_OML isn't supported, but it would be easy
+     * enough to add support.  Basically, if a context is created with an
+     * fbconfig where the swap method is GLX_SWAP_COPY_OML, pageflipping
+     * will never be used.
+     *
+     * FK: What about drivers that don't use page flipping? Could they
+     * just expose GLX_SWAP_COPY_OML?
+     */
+    static const GLenum back_buffer_modes[] = {
+       GLX_NONE, GLX_SWAP_UNDEFINED_OML /*, GLX_SWAP_COPY_OML */
+    };
+
+    u_int8_t depth_bits_array[2];
+    u_int8_t stencil_bits_array[2];
+
+
+    depth_bits_array[0] = depth_bits;
+    depth_bits_array[1] = depth_bits;
+    
+    /* Just like with the accumulation buffer, always provide some modes
+     * with a stencil buffer.  It will be a sw fallback, but some apps won't
+     * care about that.
+     */
+    stencil_bits_array[0] = 0;
+    stencil_bits_array[1] = (stencil_bits == 0) ? 8 : stencil_bits;
+
+    depth_buffer_factor = ((depth_bits != 0) || (stencil_bits != 0)) ? 2 : 1;
+    back_buffer_factor  = (have_back_buffer) ? 2 : 1;
+
+    num_modes = depth_buffer_factor * back_buffer_factor * 4;
+
+    if ( pixel_bits == 16 ) {
+        fb_format = GL_RGB;
+        fb_type = GL_UNSIGNED_SHORT_5_6_5;
+    }
+    else {
+        fb_format = GL_BGR;
+        fb_type = GL_UNSIGNED_INT_8_8_8_8_REV;
+    }
+
+    modes = (*dri_interface->createContextModes)( num_modes, sizeof( __GLcontextModes ) );
+    m = modes;
+    if ( ! driFillInModes( & m, fb_format, fb_type,
+                          depth_bits_array, stencil_bits_array, depth_buffer_factor,
+                          back_buffer_modes, back_buffer_factor,
+                          GLX_TRUE_COLOR ) ) {
+       fprintf( stderr, "[%s:%u] Error creating FBConfig!\n",
+                __func__, __LINE__ );
+       return NULL;
+    }
+
+    if ( ! driFillInModes( & m, fb_format, fb_type,
+                          depth_bits_array, stencil_bits_array, depth_buffer_factor,
+                          back_buffer_modes, back_buffer_factor,
+                          GLX_DIRECT_COLOR ) ) {
+       fprintf( stderr, "[%s:%u] Error creating FBConfig!\n",
+                __func__, __LINE__ );
+       return NULL;
+    }
+
+    /* Mark the visual as slow if there are "fake" stencil bits.
+     */
+    for ( m = modes ; m != NULL ; m = m->next ) {
+       if ( (m->stencilBits != 0) && (m->stencilBits != stencil_bits) ) {
+           m->visualRating = GLX_SLOW_CONFIG;
+       }
+    }
+
+    return modes;
+}
 
-/*
- * This is the bootstrap function for the driver.
- * The __driCreateScreen name is the symbol that libGL.so fetches.
- * Return:  pointer to a __DRIscreenPrivate.
+
+/**
+ * This is the driver specific part of the createNewScreen entry point.
+ * 
+ * \todo maybe fold this into intelInitDriver
+ *
+ * \return the __GLcontextModes supported by this driver
  */
-void *__driCreateScreen(Display *dpy, int scrn, __DRIscreen *psc,
-                        int numConfigs, __GLXvisualConfig *config)
+__GLcontextModes *__driDriverInitScreen(__DRIscreenPrivate *psp)
 {
-   __DRIscreenPrivate *psp;
-   psp = __driUtilCreateScreen(dpy, scrn, psc, numConfigs, config, &savageAPI);
-   return (void *) psp;
-}
+   static const __DRIversion ddx_expected = { 2, 0, 0 };
+   static const __DRIversion dri_expected = { 4, 0, 0 };
+   static const __DRIversion drm_expected = { 2, 1, 0 };
+   SAVAGEDRIPtr dri_priv = (SAVAGEDRIPtr)psp->pDevPriv;
+
+   if ( ! driCheckDriDdxDrmVersions2( "Savage",
+                                     &psp->dri_version, & dri_expected,
+                                     &psp->ddx_version, & ddx_expected,
+                                     &psp->drm_version, & drm_expected ) )
+      return NULL;
+
+   psp->DriverAPI = savageAPI;
+
+   /* Calling driInitExtensions here, with a NULL context pointer,
+    * does not actually enable the extensions.  It just makes sure
+    * that all the dispatch offsets for all the extensions that
+    * *might* be enables are known.  This is needed because the
+    * dispatch offsets need to be known when _mesa_context_create is
+    * called, but we can't enable the extensions until we have a
+    * context pointer.
+    *
+    * Hello chicken.  Hello egg.  How are you two today?
+    */
+   driInitExtensions( NULL, card_extensions, GL_FALSE );
 
+   if (!savageInitDriver(psp))
+       return NULL;
 
-#endif
+   return savageFillInModes( dri_priv->cpp*8,
+                            (dri_priv->cpp == 2) ? 16 : 24,
+                            (dri_priv->cpp == 2) ? 0  : 8,
+                            (dri_priv->backOffset != dri_priv->depthOffset) );
+}