- Merge from trunk into branchsarea-1-0-0-20010307

47 files changed, 5083 insertions, 5097 deletions

diff --git a/xc/lib/GL/mesa/src/Imakefile b/xc/lib/GL/mesa/src/Imakefile
index cc88803f4..7ccc16fd7 100644
--- a/xc/lib/GL/mesa/src/Imakefile
+++ b/xc/lib/GL/mesa/src/Imakefile
@@ -152,6 +152,8 @@ LinkSourceFile(state.c, $(MESASRCDIR)/src)
 LinkSourceFile(state.h, $(MESASRCDIR)/src)
 LinkSourceFile(stencil.c, $(MESASRCDIR)/src)
 LinkSourceFile(stencil.h, $(MESASRCDIR)/src)
+LinkSourceFile(texformat.c, $(MESASRCDIR)/src)
+LinkSourceFile(texformat.h, $(MESASRCDIR)/src)
 LinkSourceFile(texgen_tmp.h, $(MESASRCDIR)/src)
 LinkSourceFile(teximage.c, $(MESASRCDIR)/src)
 LinkSourceFile(teximage.h, $(MESASRCDIR)/src)
@@ -163,6 +165,7 @@ LinkSourceFile(texture.c, $(MESASRCDIR)/src)
 LinkSourceFile(texture.h, $(MESASRCDIR)/src)
 LinkSourceFile(texutil.c, $(MESASRCDIR)/src)
 LinkSourceFile(texutil.h, $(MESASRCDIR)/src)
+LinkSourceFile(texutil_tmp.h, $(MESASRCDIR)/src)
 LinkSourceFile(trans_tmp.h, $(MESASRCDIR)/src)
 LinkSourceFile(translate.c, $(MESASRCDIR)/src)
 LinkSourceFile(translate.h, $(MESASRCDIR)/src)
@@ -252,6 +255,7 @@ LinkSourceFile(zoom.h, $(MESASRCDIR)/src)
 		stages.c \
 		state.c \
 		stencil.c \
+		texformat.c \
 		teximage.c \
 		texobj.c \
 		texstate.c \
@@ -325,6 +329,7 @@ LinkSourceFile(zoom.h, $(MESASRCDIR)/src)
 		stages.o \
 		state.o \
 		stencil.o \
+		texformat.o \
 		teximage.o \
 		texobj.o \
 		texstate.o \
@@ -346,8 +351,8 @@ LinkSourceFile(zoom.h, $(MESASRCDIR)/src)
 		zoom.o
 
 #ifdef i386Architecture
-     ASM_SRCS = 
-     ASM_OBJS = 
+     ASM_SRCS =
+     ASM_OBJS =
 #if MesaUseMMX
      MMX_DEFS = -DUSE_MMX_ASM
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h b/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
index a7200f1b8..cad45626d 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810_3d_reg.h
@@ -113,8 +113,8 @@
 #define LCS_LINEWIDTH_2_0       (0x4<<12)
 #define LCS_LINEWIDTH_3_0       (0x6<<12)
 #define LCS_UPDATE_ALPHA_INTERP (0x1<<11)
-#define LCS_ALPHA_FLAT          (0x0<<10)
-#define LCS_ALPHA_INTERP        (0x1<<10)
+#define LCS_ALPHA_FLAT          (0x1<<10)
+#define LCS_ALPHA_INTERP        (0x0<<10)
 #define LCS_UPDATE_FOG_INTERP   (0x1<<9)
 #define LCS_FOG_INTERP          (0x0<<8)
 #define LCS_FOG_FLAT            (0x1<<8)
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
index 2f16e33d7..146b6c1b9 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810ioctl.c
@@ -200,7 +200,6 @@ void i810DmaFinish( i810ContextPtr imesa  )
    FLUSH_BATCH( imesa );
 
    if (imesa->sarea->last_quiescent != imesa->sarea->last_enqueue) {
-     
       if (I810_DEBUG&DEBUG_VERBOSE_IOCTL) 
 	 fprintf(stderr, "i810DmaFinish\n");
 
@@ -214,14 +213,22 @@ void i810DmaFinish( i810ContextPtr imesa  )
 
 void i810RegetLockQuiescent( i810ContextPtr imesa  ) 
 {
+   /* XXX I disabled this conditional.  Doing so fixes all the readpixels
+    * problems.  The problem was that we'd sometimes read from the frame
+    * buffer (via the span functions) before rendering was completed.
+    * Taking out this conditional solves that problem.  (BrianP)
+    *
    if (imesa->sarea->last_quiescent != imesa->sarea->last_enqueue) {
+   */
       if (I810_DEBUG&DEBUG_VERBOSE_IOCTL)
 	 fprintf(stderr, "i810RegetLockQuiescent\n");
 
       drmUnlock(imesa->driFd, imesa->hHWContext);
       i810GetLock( imesa, DRM_LOCK_QUIESCENT ); 
       imesa->sarea->last_quiescent = imesa->sarea->last_enqueue;
+   /*
    }
+   */
 }
 
 void i810WaitAgeLocked( i810ContextPtr imesa, int age  ) 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810span.c b/xc/lib/GL/mesa/src/drv/i810/i810span.c
index 9e86e9b68..4bf02ad13 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810span.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810span.c
@@ -30,7 +30,7 @@
 			dPriv->x * 2 +			\
 			dPriv->y * pitch)
 
-#define INIT_MONO_PIXEL(p) 
+#define INIT_MONO_PIXEL(p)
 
 #define CLIPPIXEL(_x,_y) (_x >= minx && _x < maxx && \
 			  _y >= miny && _y < maxy)
@@ -50,6 +50,8 @@
 
 #define HW_LOCK()				\
    i810ContextPtr imesa = I810_CONTEXT(ctx);	\
+   FLUSH_BATCH(imesa);				\
+   i810DmaFinish(imesa);			\
    LOCK_HARDWARE_QUIESCENT(imesa);
 
 #define HW_CLIPLOOP()						\
@@ -85,9 +87,9 @@
 #define READ_RGBA( rgba, _x, _y )				\
 do {								\
    GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);	\
-   rgba[0] = (p >> 8) & 0xf8;					\
-   rgba[1] = (p >> 3) & 0xfc;					\
-   rgba[2] = (p << 3) & 0xf8;					\
+   rgba[0] = (((p >> 11) & 0x1f) * 255) / 31;			\
+   rgba[1] = (((p >>  5) & 0x3f) * 255) / 63;			\
+   rgba[2] = (((p >>  0) & 0x1f) * 255) / 31;			\
    rgba[3] = 255;						\
 } while(0)
 
@@ -128,7 +130,7 @@ do {								\
    *(GLushort *)(buf + _x*2 + _y*pitch)  = d;
 
 #define READ_DEPTH( d, _x, _y )	\
-   d = *(GLushort *)(buf + _x*2 + _y*pitch);	 
+   d = *(GLushort *)(buf + _x*2 + _y*pitch);
 
 /*     d = 0xffff; */
 
@@ -144,7 +146,7 @@ void i810DDInitSpanFuncs( GLcontext *ctx )
       ctx->Driver.WriteRGBSpan = i810WriteRGBSpan_565;
       ctx->Driver.WriteMonoRGBASpan = i810WriteMonoRGBASpan_565;
       ctx->Driver.WriteRGBAPixels = i810WriteRGBAPixels_565;
-      ctx->Driver.WriteMonoRGBAPixels = i810WriteMonoRGBAPixels_565; 
+      ctx->Driver.WriteMonoRGBAPixels = i810WriteMonoRGBAPixels_565;
       ctx->Driver.ReadRGBASpan = i810ReadRGBASpan_565;
       ctx->Driver.ReadRGBAPixels = i810ReadRGBAPixels_565;
    } else {
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810state.c b/xc/lib/GL/mesa/src/drv/i810/i810state.c
index cdccbd5ec..50a916fb7 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810state.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810state.c
@@ -182,6 +182,7 @@ static void i810DDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
    GLubyte p[4];
    int i,j,k;
    int active = (ctx->Polygon.StippleFlag && ctx->PB->primitive == GL_POLYGON);
+   GLuint newMask;
 
    FLUSH_BATCH(imesa);
    ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
@@ -191,10 +192,10 @@ static void i810DDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
       imesa->Setup[I810_CTXREG_ST1] &= ~ST1_ENABLE;
    }
 
-   p[0] = mask[0] & 0xf; p[0] |= p[0] << 4;
-   p[1] = mask[4] & 0xf; p[1] |= p[1] << 4;
-   p[2] = mask[8] & 0xf; p[2] |= p[2] << 4;
-   p[3] = mask[12] & 0xf; p[3] |= p[3] << 4;
+   p[0] = mask[12] & 0xf; p[0] |= p[0] << 4;
+   p[1] = mask[8] & 0xf; p[1] |= p[1] << 4;
+   p[2] = mask[4] & 0xf; p[2] |= p[2] << 4;
+   p[3] = mask[0] & 0xf; p[3] |= p[3] << 4;
 
    for (k = 0 ; k < 8 ; k++)
       for (j = 0 ; j < 4; j++) 
@@ -203,12 +204,19 @@ static void i810DDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
 	       ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
 	       return;
 	    }
-   
+
+   newMask = ((p[0] & 0xf) << 0) |
+             ((p[1] & 0xf) << 4) |
+             ((p[2] & 0xf) << 8) |
+             ((p[3] & 0xf) << 12);
+   if (newMask == 0xffff) {
+      /* do opaque stipple in software for conformance */
+      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+      return;
+   }
+
    imesa->Setup[I810_CTXREG_ST1] &= ~0xffff;
-   imesa->Setup[I810_CTXREG_ST1] |= ( ((p[0] & 0xf) << 0) |
-				      ((p[1] & 0xf) << 4) |
-				      ((p[2] & 0xf) << 8) |
-				      ((p[3] & 0xf) << 12) );   
+   imesa->Setup[I810_CTXREG_ST1] |= newMask;
 
    if (active)
       imesa->Setup[I810_CTXREG_ST1] |= ST1_ENABLE;
@@ -439,26 +447,23 @@ static GLboolean i810DDColorMask(GLcontext *ctx,
 				 GLboolean b, GLboolean a )
 {
    i810ContextPtr imesa = I810_CONTEXT( ctx );
-   GLuint tmp = 0;
-   GLuint rv = 1;
-
-   imesa->Fallback &= ~I810_FALLBACK_COLORMASK;
+   GLuint tmp;
 
-   if (r && g && b) {
-      tmp = imesa->Setup[I810_CTXREG_B2] | B2_FB_WRITE_ENABLE;
-   } else if (!r && !g && !b) {
-      tmp = imesa->Setup[I810_CTXREG_B2] & ~B2_FB_WRITE_ENABLE;
-   } else {
-      rv = 0;
+   if (r && g && b)
+      imesa->Fallback &= ~I810_FALLBACK_COLORMASK;
+   else
       imesa->Fallback |= I810_FALLBACK_COLORMASK;
-   }
-      
+
+   tmp = imesa->Setup[I810_CTXREG_B2] |
+      (B2_FB_WRITE_ENABLE | B2_UPDATE_FB_WRITE_ENABLE);
+
    if (tmp != imesa->Setup[I810_CTXREG_B2]) {
       FLUSH_BATCH(imesa);
       imesa->Setup[I810_CTXREG_B2] = tmp;
+      imesa->dirty |= I810_UPLOAD_CTX;
    }
 
-   return rv;
+   return GL_FALSE;  /* makes s/w path always do s/w masking */
 }
 
 /* Seperate specular not fully implemented in hardware...  Needs
@@ -625,16 +630,6 @@ static void i810DDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
    case GL_TEXTURE_2D:      
       FLUSH_BATCH(imesa);
       imesa->new_state |= I810_NEW_TEXTURE;
-      imesa->dirty |= I810_UPLOAD_CTX;
-      if (ctx->Texture.CurrentUnit == 0) {
-	 imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL0_ENABLE;
-	 if (state)
-	    imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL0_ENABLE;
-      } else {
-	 imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL1_ENABLE;
-	 if (state)
-	    imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL1_ENABLE;
-      }
       break;
    case GL_COLOR_LOGIC_OP:
       FLUSH_BATCH( imesa );
@@ -995,8 +990,8 @@ void i810DDInitState( i810ContextPtr imesa )
 
 #define INTERESTED (~(NEW_MODELVIEW|NEW_PROJECTION|\
                       NEW_TEXTURE_MATRIX|\
-                      NEW_USER_CLIP|NEW_CLIENT_STATE|\
-                      NEW_TEXTURE_ENABLE))
+                      NEW_USER_CLIP|NEW_CLIENT_STATE))
+
 
 void i810DDUpdateState( GLcontext *ctx )
 {
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tex.c b/xc/lib/GL/mesa/src/drv/i810/i810tex.c
index 5c1615128..ff537aece 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tex.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tex.c
@@ -705,44 +705,50 @@ static void i810UpdateTex0State( GLcontext *ctx )
    struct gl_texture_object	*tObj;
    i810TextureObjectPtr t;
    int ma_modulate_op;
+   int format;
 
-
-   tObj = ctx->Texture.Unit[0].Current;
-
-   if ( tObj != ctx->Texture.Unit[0].CurrentD[2] ) 
-      tObj = 0;
-
-
+   /* disable */
+   imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL0_ENABLE;
    imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
-				     MC_STAGE_0 |
-				     MC_UPDATE_DEST |
-				     MC_DEST_CURRENT |
-				     MC_UPDATE_ARG1 |
-				     MC_ARG1_ITERATED_COLOR | 
-				     MC_UPDATE_ARG2 |
-				     MC_ARG2_ONE |
-				     MC_UPDATE_OP |
-				     MC_OP_ARG1 );
-
+                                     MC_STAGE_0 |
+                                     MC_UPDATE_DEST |
+                                     MC_DEST_CURRENT |
+                                     MC_UPDATE_ARG1 |
+                                     MC_ARG1_ITERATED_COLOR | 
+                                     MC_UPDATE_ARG2 |
+                                     MC_ARG2_ONE |
+                                     MC_UPDATE_OP |
+                                     MC_OP_ARG1 );
    imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
-				     MA_STAGE_0 |
-				     MA_UPDATE_ARG1 |
-				     MA_ARG1_ITERATED_ALPHA |
-				     MA_UPDATE_ARG2 |
-				     MA_ARG2_TEX0_ALPHA |
-				     MA_UPDATE_OP |
-				     MA_OP_ARG1 );
-    
+                                     MA_STAGE_0 |
+                                     MA_UPDATE_ARG1 |
+                                     MA_ARG1_ITERATED_ALPHA |
+                                     MA_UPDATE_ARG2 |
+                                     MA_ARG2_TEX0_ALPHA |
+                                     MA_UPDATE_OP |
+                                     MA_OP_ARG1 );
+
+   if (ctx->Texture.Unit[0].ReallyEnabled == 0) {
+      return;
+   }
 
-   if (!(ctx->Texture.ReallyEnabled & 0xf) || !tObj || !tObj->Complete) {
+   tObj = ctx->Texture.Unit[0].Current;
+   if (ctx->Texture.Unit[0].ReallyEnabled != TEXTURE0_2D ||
+       tObj->Image[tObj->BaseLevel]->Border > 0) {
+      /* 1D or 3D texturing enabled, or texture border - fallback */
+      imesa->Fallback |= I810_FALLBACK_TEXTURE;
       return;
    }
 
+   /* Do 2D texture setup */
+
+   imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL0_ENABLE;
+
    t = tObj->DriverData;
-  
    if (!t) {
       t = i810CreateTexObj( imesa, tObj );
-      if (!t) return;
+      if (!t)
+         return;
    }
 
    if (t->current_unit != 0)
@@ -757,9 +763,11 @@ static void i810UpdateTex0State( GLcontext *ctx )
    if (t->MemBlock)
       i810UpdateTexLRU( imesa, t );
   
+   format = t->image[0].internalFormat;
+
    switch (ctx->Texture.Unit[0].EnvMode) {
    case GL_REPLACE:
-      if (t->image[0].internalFormat == GL_ALPHA) 
+      if (format == GL_ALPHA) 
 	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_0 |
 					   MC_UPDATE_DEST |
@@ -782,7 +790,7 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					   MC_UPDATE_OP |
 					   MC_OP_ARG1 );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_ARG2;
@@ -809,7 +817,7 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_MODULATE );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_MODULATE;
@@ -826,30 +834,74 @@ static void i810UpdateTex0State( GLcontext *ctx )
       break;
 
    case GL_ADD:
-      imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
-					MC_STAGE_0 |
-					MC_UPDATE_DEST |
-					MC_DEST_CURRENT |
-					MC_UPDATE_ARG1 |
-					MC_ARG1_TEX0_COLOR | 
-					MC_UPDATE_ARG2 |
-					MC_ARG2_ITERATED_COLOR |
-					MC_UPDATE_OP |
-					MC_OP_ADD );
+      if (format == GL_ALPHA) {
+         /* Cv = Cf */
+	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+					   MC_STAGE_0 |
+					   MC_UPDATE_DEST |
+					   MC_DEST_CURRENT |
+					   MC_UPDATE_ARG1 |
+					   MC_ARG1_TEX0_COLOR | 
+					   MC_UPDATE_ARG2 |
+					   MC_ARG2_ITERATED_COLOR |
+					   MC_UPDATE_OP |
+					   MC_OP_ARG2 );
+      }
+      else {
+         /* Cv = Cf + Ct */
+         imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
+                                           MC_STAGE_0 |
+                                           MC_UPDATE_DEST |
+                                           MC_DEST_CURRENT |
+                                           MC_UPDATE_ARG1 |
+                                           MC_ARG1_TEX0_COLOR | 
+                                           MC_UPDATE_ARG2 |
+                                           MC_ARG2_ITERATED_COLOR |
+                                           MC_UPDATE_OP |
+                                           MC_OP_ADD );
+      }
 
-      imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
-					MA_STAGE_0 |
-					MA_UPDATE_ARG1 |
-					MA_ARG1_ITERATED_ALPHA |
-					MA_UPDATE_ARG2 |
-					MA_ARG2_TEX0_ALPHA |
-					MA_UPDATE_OP |
-					MA_OP_ADD );
+      /* alpha */
+      if (format == GL_ALPHA ||
+          format == GL_LUMINANCE_ALPHA ||
+          format == GL_RGBA) {
+         /* Av = Af * At */
+         imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+                                           MA_STAGE_0 |
+                                           MA_UPDATE_ARG1 |
+                                           MA_ARG1_ITERATED_ALPHA |
+                                           MA_UPDATE_ARG2 |
+                                           MA_ARG2_TEX0_ALPHA |
+                                           MA_UPDATE_OP |
+                                           MA_OP_MODULATE );
+      }
+      else if (format == GL_LUMINANCE || format == GL_RGB) {
+         /* Av = Af */
+         imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+                                           MA_STAGE_0 |
+                                           MA_UPDATE_ARG1 |
+                                           MA_ARG1_ITERATED_ALPHA |
+                                           MA_UPDATE_ARG2 |
+                                           MA_ARG2_ITERATED_ALPHA |
+                                           MA_UPDATE_OP |
+                                           MA_OP_ARG1 );
+      }
+      else {
+         /* Av = Af + At */
+         imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+                                           MA_STAGE_0 |
+                                           MA_UPDATE_ARG1 |
+                                           MA_ARG1_ITERATED_ALPHA |
+                                           MA_UPDATE_ARG2 |
+                                           MA_ARG2_TEX0_ALPHA |
+                                           MA_UPDATE_OP |
+                                           MA_OP_ADD );
+      }
       break;
 
    case GL_DECAL:
-
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
+         /* C = Ct */
 	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_0 |
 					   MC_UPDATE_DEST |
@@ -862,29 +914,32 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					   MC_OP_ARG2 );
 
       } else {
+         /* RGBA or undefined result */
+         /* C = Cf*(1-At)+Ct*At */
 	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_0 |
 					   MC_UPDATE_DEST |
 					   MC_DEST_CURRENT |
 					   MC_UPDATE_ARG1 |
-					   MC_ARG1_COLOR_FACTOR | 
+					   MC_ARG1_TEX0_COLOR |
 					   MC_UPDATE_ARG2 |
-					   MC_ARG2_TEX0_COLOR |
+					   MC_ARG2_ITERATED_COLOR | 
 					   MC_UPDATE_OP |
 					   MC_OP_LIN_BLEND_TEX0_ALPHA );
       }
 
+      /* Av = Af */
       imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
 					MA_STAGE_0 |
 					MA_UPDATE_ARG1 |
-					MA_ARG1_ALPHA_FACTOR |
+					MA_ARG1_ITERATED_ALPHA |
 					MA_UPDATE_ARG2 |
-					MA_ARG2_ALPHA_FACTOR |
+					MA_ARG2_ITERATED_ALPHA |
 					MA_UPDATE_OP |
 					MA_OP_ARG1 );
       break;
    case GL_BLEND:
-      if (t->image[0].internalFormat == GL_ALPHA) 
+      if (format == GL_ALPHA) 
 	 imesa->Setup[I810_CTXREG_MC0] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_0 |
 					   MC_UPDATE_DEST |
@@ -907,7 +962,20 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					   MC_UPDATE_OP |
 					   MC_OP_LIN_BLEND_TEX0_COLOR );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      /* alpha */
+      if (format == GL_LUMINANCE || format == GL_RGB) {
+         /* Av = Af */
+         imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
+                                           MA_STAGE_0 |
+                                           MA_UPDATE_ARG1 |
+                                           MA_ARG1_ITERATED_ALPHA |
+                                           MA_UPDATE_ARG2 |
+                                           MA_ARG2_ITERATED_ALPHA |
+                                           MA_UPDATE_OP |
+                                           MA_OP_ARG1 );
+      }
+      else if (format == GL_INTENSITY) {
+         /* Av = Af(1-It)+AcIt */
 	 imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
 					   MA_STAGE_0 |
 					   MA_UPDATE_ARG1 |
@@ -915,16 +983,17 @@ static void i810UpdateTex0State( GLcontext *ctx )
 					   MA_UPDATE_ARG2 |
 					   MA_ARG2_ITERATED_ALPHA |
 					   MA_UPDATE_OP |
-					   MA_OP_ARG1 );
+					   MA_OP_LIN_BLEND_TEX0_ALPHA );
       } else {
+         /* Av = AfAt */
 	 imesa->Setup[I810_CTXREG_MA0] = ( GFX_OP_MAP_ALPHA_STAGES |
 					   MA_STAGE_0 |
 					   MA_UPDATE_ARG1 |
-					   MA_ARG1_ALPHA_FACTOR |
+					   MA_ARG1_TEX0_ALPHA |
 					   MA_UPDATE_ARG2 |
 					   MA_ARG2_ITERATED_ALPHA |
 					   MA_UPDATE_OP |
-					   MA_OP_LIN_BLEND_TEX0_ALPHA );
+					   MA_OP_MODULATE );
       }
       break;
 
@@ -942,50 +1011,56 @@ static void i810UpdateTex1State( GLcontext *ctx )
    i810ContextPtr imesa = I810_CONTEXT(ctx);
    struct gl_texture_object	*tObj;
    i810TextureObjectPtr t;
-   int ma_modulate_op;
-
-
-   tObj = ctx->Texture.Unit[1].Current;
-
-   if ( tObj != ctx->Texture.Unit[1].CurrentD[2] ) 
-      tObj = 0;
-
+   int ma_modulate_op, format;
 
+   /* disable */
+   imesa->Setup[I810_CTXREG_MT] &= ~MT_TEXEL1_ENABLE;
    imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
-				     MC_STAGE_1 |
-				     MC_UPDATE_DEST |
-				     MC_DEST_CURRENT |
-				     MC_UPDATE_ARG1 |
-				     MC_ARG1_ONE | 
-				     MC_ARG1_DONT_REPLICATE_ALPHA |
-				     MC_ARG1_DONT_INVERT |
-				     MC_UPDATE_ARG2 |
-				     MC_ARG2_ONE |
-				     MC_ARG2_DONT_REPLICATE_ALPHA |
-				     MC_ARG2_DONT_INVERT |
-				     MC_UPDATE_OP |
-				     MC_OP_DISABLE );
-
+                                     MC_STAGE_1 |
+                                     MC_UPDATE_DEST |
+                                     MC_DEST_CURRENT |
+                                     MC_UPDATE_ARG1 |
+                                     MC_ARG1_ONE | 
+                                     MC_ARG1_DONT_REPLICATE_ALPHA |
+                                     MC_ARG1_DONT_INVERT |
+                                     MC_UPDATE_ARG2 |
+                                     MC_ARG2_ONE |
+                                     MC_ARG2_DONT_REPLICATE_ALPHA |
+                                     MC_ARG2_DONT_INVERT |
+                                     MC_UPDATE_OP |
+                                     MC_OP_DISABLE );
    imesa->Setup[I810_CTXREG_MA1] = ( GFX_OP_MAP_ALPHA_STAGES |
-				     MA_STAGE_1 |
-				     MA_UPDATE_ARG1 |
-				     MA_ARG1_CURRENT_ALPHA |
-				     MA_ARG1_DONT_INVERT |
-				     MA_UPDATE_ARG2 |
-				     MA_ARG2_CURRENT_ALPHA |
-				     MA_ARG2_DONT_INVERT |
-				     MA_UPDATE_OP |
-				     MA_OP_ARG1 );
-
-   if (!(ctx->Texture.ReallyEnabled & 0xf0) || !tObj || !tObj->Complete) {
+                                     MA_STAGE_1 |
+                                     MA_UPDATE_ARG1 |
+                                     MA_ARG1_CURRENT_ALPHA |
+                                     MA_ARG1_DONT_INVERT |
+                                     MA_UPDATE_ARG2 |
+                                     MA_ARG2_CURRENT_ALPHA |
+                                     MA_ARG2_DONT_INVERT |
+                                     MA_UPDATE_OP |
+                                     MA_OP_ARG1 );
+
+   if (ctx->Texture.Unit[1].ReallyEnabled == 0) {
+      return;
+   }
+
+   tObj = ctx->Texture.Unit[1].Current;
+   if (ctx->Texture.Unit[1].ReallyEnabled != TEXTURE0_2D ||
+       tObj->Image[tObj->BaseLevel]->Border > 0) {
+      /* 1D or 3D texturing enabled, or texture border - fallback */
+      imesa->Fallback |= I810_FALLBACK_TEXTURE;
       return;
    }
 
+   /* Do 2D texture setup */
+
+   imesa->Setup[I810_CTXREG_MT] |= MT_TEXEL1_ENABLE;
+
    t = tObj->DriverData;
-  
    if (!t) {
       t = i810CreateTexObj( imesa, tObj );
-      if (!t) return;
+      if (!t)
+         return;
    }
     
    if (t->current_unit != 1)
@@ -1000,6 +1075,8 @@ static void i810UpdateTex1State( GLcontext *ctx )
    if (t->MemBlock)
       i810UpdateTexLRU( imesa, t );
 
+   format = t->image[0].internalFormat;
+
    switch (ctx->Texture.Unit[1].EnvMode) {
    case GL_REPLACE:
       imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
@@ -1013,7 +1090,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_ARG1 );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_ARG2;
@@ -1040,7 +1117,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_MODULATE );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_MODULATE;
@@ -1068,7 +1145,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_ADD );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 ma_modulate_op = MA_OP_ARG1;
       } else {
 	 ma_modulate_op = MA_OP_ADD;
@@ -1086,7 +1163,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 
 
    case GL_DECAL:
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 imesa->Setup[I810_CTXREG_MC1] = ( GFX_OP_MAP_COLOR_STAGES |
 					   MC_STAGE_1 |
 					   MC_UPDATE_DEST |
@@ -1133,7 +1210,7 @@ static void i810UpdateTex1State( GLcontext *ctx )
 					MC_UPDATE_OP |
 					MC_OP_LIN_BLEND_TEX1_COLOR );
 
-      if (t->image[0].internalFormat == GL_RGB) {
+      if (format == GL_RGB) {
 	 imesa->Setup[I810_CTXREG_MA1] = ( GFX_OP_MAP_ALPHA_STAGES |
 					   MA_STAGE_1 |
 					   MA_UPDATE_ARG1 |
@@ -1169,11 +1246,12 @@ void i810UpdateTextureState( GLcontext *ctx )
    if (imesa->CurrentTexObj[1]) imesa->CurrentTexObj[1]->bound = 0;
    imesa->CurrentTexObj[0] = 0;
    imesa->CurrentTexObj[1] = 0;   
+   imesa->Fallback &= ~I810_FALLBACK_TEXTURE;
    i810UpdateTex0State( ctx );
    i810UpdateTex1State( ctx );
    I810_CONTEXT( ctx )->dirty |= (I810_UPLOAD_CTX |
-				  I810_UPLOAD_TEX0 | 
-				  I810_UPLOAD_TEX1);
+                                  I810_UPLOAD_TEX0 | 
+                                  I810_UPLOAD_TEX1);
 }
 
 
@@ -1196,13 +1274,17 @@ static void i810TexEnv( GLcontext *ctx, GLenum target,
 
       struct gl_texture_unit *texUnit = 
 	 &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
-      GLfloat *fc = texUnit->EnvColor;
-      GLuint col;
-
-      col = ((((GLubyte)fc[3])<<24) | 
-	     (((GLubyte)fc[0])<<16) | 
-	     (((GLubyte)fc[1])<<8) | 
-	     (((GLubyte)fc[2])<<0));
+      const GLfloat *fc = texUnit->EnvColor;
+      GLuint r, g, b, a, col;
+      FLOAT_COLOR_TO_UBYTE_COLOR(r, fc[0]);
+      FLOAT_COLOR_TO_UBYTE_COLOR(g, fc[1]);
+      FLOAT_COLOR_TO_UBYTE_COLOR(b, fc[2]);
+      FLOAT_COLOR_TO_UBYTE_COLOR(a, fc[3]);
+
+      col = ((a << 24) | 
+	     (r << 16) | 
+	     (g <<  8) | 
+	     (b <<  0));
     
       if (imesa->Setup[I810_CTXREG_CF1] != col) {
 	 FLUSH_BATCH(imesa);	
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810tris.h b/xc/lib/GL/mesa/src/drv/i810/i810tris.h
index a83adad95..a416f82dc 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810tris.h
+++ b/xc/lib/GL/mesa/src/drv/i810/i810tris.h
@@ -87,40 +87,42 @@ static __inline__ void i810_draw_point( i810ContextPtr imesa,
 {
    int vertsize = imesa->vertsize;
    GLuint *vb = i810AllocDwordsInline( imesa, 6 * vertsize );
+   const GLfloat x = tmp->v.x + 0.125;
+   const GLfloat y = tmp->v.y - 0.5F;
    int j;
 
-   *(float *)&vb[0] = tmp->v.x - sz;
-   *(float *)&vb[1] = tmp->v.y - sz;
+   *(float *)&vb[0] = x - sz;
+   *(float *)&vb[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x + sz;
-   *(float *)&vb[1] = tmp->v.y - sz;
+   *(float *)&vb[0] = x + sz;
+   *(float *)&vb[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x + sz;
-   *(float *)&vb[1] = tmp->v.y + sz;
+   *(float *)&vb[0] = x + sz;
+   *(float *)&vb[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x + sz;
-   *(float *)&vb[1] = tmp->v.y + sz;
+   *(float *)&vb[0] = x + sz;
+   *(float *)&vb[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x - sz;
-   *(float *)&vb[1] = tmp->v.y + sz;
+   *(float *)&vb[0] = x - sz;
+   *(float *)&vb[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
    vb += vertsize;
 
-   *(float *)&vb[0] = tmp->v.x - sz;
-   *(float *)&vb[1] = tmp->v.y - sz;
+   *(float *)&vb[0] = x - sz;
+   *(float *)&vb[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++)
       vb[j] = tmp->ui[j];
 }
@@ -143,13 +145,33 @@ static __inline__ void i810_draw_line( i810ContextPtr imesa,
 			 : "=%c" (j)
 			 : "0" (vertsize), "S" ((long)v1)
 			 : "memory" );
+#elif 0
+   for (j = 0 ; j < vertsize ; j++)
+      vb[j] = v0->ui[j];
+
+   vb += vertsize;
+   for (j = 0 ; j < vertsize ; j++)
+      vb[j] = v1->ui[j];
 #else
+   const GLfloat dx = -0.5;
+   const GLfloat dy = -0.5;
+
+   v0->v.x += dx;
+   v0->v.y += dy;
+   v1->v.x += dx;
+   v1->v.y += dy;
+
    for (j = 0 ; j < vertsize ; j++)
       vb[j] = v0->ui[j];
 
    vb += vertsize;
    for (j = 0 ; j < vertsize ; j++)
       vb[j] = v1->ui[j];
+
+   v0->v.x -= dx;
+   v0->v.y -= dy;
+   v1->v.x -= dx;
+   v1->v.y -= dy;
 #endif
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/i810/i810vb.c b/xc/lib/GL/mesa/src/drv/i810/i810vb.c
index a6ad760a6..5ecc3e2af 100644
--- a/xc/lib/GL/mesa/src/drv/i810/i810vb.c
+++ b/xc/lib/GL/mesa/src/drv/i810/i810vb.c
@@ -87,15 +87,18 @@
 
 #define COORD					\
       GLfloat *win = VB->Win.data[i];		\
-      v->v.x =                 win[0];	\
-      v->v.y =    i810height - win[1];	\
-      v->v.z = (1.0/0x10000) * win[2];	\
+      v->v.x =       xoffset + win[0];		\
+      v->v.y =       yoffset - win[1];		\
+      v->v.z = (1.0/0x10000) * win[2];		\
       v->v.oow =               win[3]; 
 
 
 
 #define NOP
 
+#define SUBPIXEL_X -0.5
+#define SUBPIXEL_Y -0.375
+
 
 #define SETUPFUNC(name,win,col,tex0,tex1,tex0_4,spec,fog)		\
 static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
@@ -105,9 +108,12 @@ static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
    i810VertexPtr v;							\
    GLfloat (*tc0)[4];							\
    GLfloat (*tc1)[4];							\
-   GLfloat i810height = dPriv->h;					\
+   const GLfloat xoffset = SUBPIXEL_X;					\
+   const GLfloat yoffset = dPriv->h + SUBPIXEL_Y;			\
    int i;								\
-   (void) i810height; (void) imesa;					\
+   (void) xoffset;							\
+   (void) yoffset;							\
+   (void) imesa;							\
 									\
 									\
    gl_import_client_data( VB, VB->ctx->RenderFlags,			\
@@ -269,7 +275,7 @@ void i810ChooseRasterSetupFunc(GLcontext *ctx)
      funcindex |= I810_TEX0_BIT;
   
   if (ctx->Texture.ReallyEnabled & 0xf0) {
-     funcindex |= I810_TEX1_BIT;
+     funcindex |= (I810_TEX0_BIT | I810_TEX1_BIT);
      imesa->vertsize = 10;
      imesa->Setup[I810_CTXREG_VF] = I810_VFMT_T0T1;
   }
diff --git a/xc/lib/GL/mesa/src/drv/mga/Imakefile b/xc/lib/GL/mesa/src/drv/mga/Imakefile
index 2630d8159..e4c73ddc0 100644
--- a/xc/lib/GL/mesa/src/drv/mga/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/mga/Imakefile
@@ -151,6 +151,7 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		../../stages.c \
 		../../state.c \
 		../../stencil.c \
+		../../texformat.c \
 		../../teximage.c \
 		../../texobj.c \
 		../../texstate.c \
@@ -224,6 +225,7 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		../../stages.o \
 		../../state.o \
 		../../stencil.o \
+		../../texformat.o \
 		../../teximage.o \
 		../../texobj.o \
 		../../texstate.o \
diff --git a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
index f428614f0..8e7bb0f24 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c
@@ -314,6 +314,7 @@ GLboolean XMesaCreateContext( Display *dpy, GLvisual *mesaVis,
    mmesa->hHWContext = driContextPriv->hHWContext;
    mmesa->driFd = sPriv->fd;
    mmesa->driHwLock = &sPriv->pSAREA->lock;
+
    mmesa->mgaScreen = mgaScreen;
    mmesa->driScreen = sPriv;
    mmesa->sarea = saPriv;
@@ -374,7 +375,7 @@ GLboolean XMesaCreateContext( Display *dpy, GLvisual *mesaVis,
       break;
    };
 
-
+   mmesa->canDoStipple = GL_FALSE;
    mmesa->renderindex = -1;		/* impossible value */
    mmesa->new_state = ~0;
    mmesa->dirty = ~0;
@@ -444,14 +445,16 @@ GLframebuffer *XMesaCreateWindowBuffer( Display *dpy,
                                         __DRIdrawablePrivate *driDrawPriv,
                                         GLvisual *mesaVis)
 {
+   GLboolean swStencil = mesaVis->StencilBits > 0 && mesaVis->DepthBits != 24;
+
    if (MGA_DEBUG&DEBUG_VERBOSE_DRI)
       fprintf(stderr, "XMesaCreateWindowBuffer\n");
 
    return gl_create_framebuffer(mesaVis,
                                 GL_FALSE,  /* software depth buffer? */
-                                mesaVis->StencilBits > 0,
+                                swStencil,
                                 mesaVis->AccumRedBits > 0,
-                                mesaVis->AlphaBits > 0
+                                GL_FALSE   /* software alpha buffer/ */
                                 );
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgacontext.h b/xc/lib/GL/mesa/src/drv/mga/mgacontext.h
index 1ae30deec..e1aacb193 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgacontext.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgacontext.h
@@ -57,11 +57,13 @@
  *    - texture env GL_BLEND -- can be fixed
  *    - 1D and 3D textures
  *    - incomplete textures
+ *    - GL_DEPTH_FUNC == GL_NEVER not in h/w
  */
 #define MGA_FALLBACK_TEXTURE   0x1
 #define MGA_FALLBACK_BUFFER    0x2
 #define MGA_FALLBACK_LOGICOP   0x4
 #define MGA_FALLBACK_STENCIL   0x8
+#define MGA_FALLBACK_DEPTH     0x10
 
 
 /* For mgaCtx->new_state.
@@ -191,6 +193,7 @@ struct mga_context_t {
    GLuint        depth_clear_mask;
    GLuint        stencil_clear_mask;
    GLuint        hw_stencil;
+   GLboolean     canDoStipple;
 
    /* Dma buffers
     */
@@ -220,7 +223,6 @@ struct mga_context_t {
     */
    unsigned int texAge[MGA_NR_TEX_HEAPS];/* texture LRU age  */
    unsigned int dirtyAge;		/* buffer age for synchronization */
-   unsigned int lastSwap;		/* throttling runaway apps */
    GLuint *status;
    GLuint status_offset;
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgadd.c b/xc/lib/GL/mesa/src/drv/mga/mgadd.c
index dced4235b..6f055c5be 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgadd.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgadd.c
@@ -51,7 +51,7 @@
 #include "X86/common_x86_asm.h"
 #endif
 
-#define MGA_DATE	"20010215"
+#define MGA_DATE	"20010306"
 
 
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
index c803ac5b5..c31d23eca 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
@@ -28,7 +28,7 @@ static void mga_iload_dma_ioctl(mgaContextPtr mmesa,
 
    if (MGA_DEBUG&DEBUG_VERBOSE_IOCTL)
       fprintf(stderr, "DRM_IOCTL_MGA_ILOAD idx %d dst %x length %d\n",
-	      buf->idx, dest, length);
+	      buf->idx, (int) dest, length);
 
    ret = drmMGATextureLoad( mmesa->driFd, buf->idx, dest, length );
    if ( ret < 0 ) {
@@ -156,8 +156,9 @@ drmBufPtr mga_get_buffer_ioctl( mgaContextPtr mmesa )
 
 
 
-GLbitfield mgaDDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
-		       GLint cx, GLint cy, GLint cw, GLint ch )
+static GLbitfield
+mgaDDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
+            GLint cx, GLint cy, GLint cw, GLint ch )
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
    __DRIdrawablePrivate *dPriv = mmesa->driDrawable;
@@ -290,7 +291,6 @@ void mgaSwapBuffers( mgaContextPtr mmesa )
    int nbox;
    int ret;
    int i;
-   int tmp;
    GLuint last_frame, last_wrap;
 
    FLUSH_BATCH( mmesa );
@@ -300,6 +300,16 @@ void mgaSwapBuffers( mgaContextPtr mmesa )
    last_frame = mmesa->sarea->last_frame.head;
    last_wrap = mmesa->sarea->last_frame.wrap;
 
+   /* FIXME: Add a timeout to this loop...
+    */
+   while ( 1 ) {
+      if ( last_wrap < mmesa->sarea->last_wrap ||
+	   ( last_wrap == mmesa->sarea->last_wrap &&
+	     last_frame <= *mmesa->status - mmesa->status_offset ) ) {
+	 break;
+      }
+   }
+
    /* Use the frontbuffer cliprects
     */
    if (mmesa->dirty_cliprects & MGA_FRONT)
@@ -329,32 +339,8 @@ void mgaSwapBuffers( mgaContextPtr mmesa )
       }
    }
 
-   if ( 0 ) {
-      fprintf( stderr, "waiting...\n" );
-   }
-
-   while ( 1 ) {
-      if ( last_wrap < mmesa->sarea->last_wrap ||
-	   ( last_wrap == mmesa->sarea->last_wrap &&
-	      last_frame <= *mmesa->status - mmesa->status_offset ) ) {
-	 break;
-      }
-      if ( 0 ) {
-	 fprintf( stderr, "   head=0x%06x wrap=%d\n",
-		  *mmesa->status - mmesa->status_offset,
-		  mmesa->sarea->last_wrap );
-	 fprintf( stderr, "  frame=0x%06x wrap=%d\n\n",
-		  last_frame, last_wrap );
-      }
-   }
-
-   if ( 0 )
-      fprintf( stderr, "waiting... done.\n" );
-
    UNLOCK_HARDWARE( mmesa );
 
-
-   mmesa->lastSwap = tmp;
    mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
 }
 
@@ -408,8 +394,8 @@ static int intersect_rect( XF86DRIClipRectPtr out,
    if (b->y1 > out->y1) out->y1 = b->y1;
    if (b->x2 < out->x2) out->x2 = b->x2;
    if (b->y2 < out->y2) out->y2 = b->y2;
-   if (out->x1 >= out->x2) return 0;
-   if (out->y1 >= out->y2) return 0;
+   if (out->x1 > out->x2) return 0;
+   if (out->y1 > out->y2) return 0;
    return 1;
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgapixel.c b/xc/lib/GL/mesa/src/drv/mga/mgapixel.c
index d970e49b4..72903c209 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgapixel.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgapixel.c
@@ -545,9 +545,9 @@ mgaDDDrawPixels( GLcontext *ctx,
 	  * negligble speedup when the buffers/images don't exactly
 	  * match:
 	  */
-#if 1
+#if 0
 	 if (cpp == 2) {
-	    if (!_mesa_convert_teximage( MESA_R5_G6_B5,
+	    if (!_mesa_convert_teximage( MESA_FORMAT_RGB565,
 					 width, rows,
 					 address, bufferpitch,
 					 width, rows,
@@ -558,7 +558,7 @@ mgaDDDrawPixels( GLcontext *ctx,
 	       return GL_FALSE;
 	    }
 	 } else {
-	    if (!_mesa_convert_teximage( MESA_A8_R8_G8_B8,
+	    if (!_mesa_convert_teximage( MESA_FORMAT_ARGB8888,
 					 width, rows,
 					 address, bufferpitch,
 					 width, rows,
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaspan.c b/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
index b707bc406..83a78f2dc 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
@@ -25,7 +25,6 @@
 			dPriv->y * pitch);		\
    GLuint p = MGA_CONTEXT( ctx )->MonoColor;		\
    (void) read_buf; (void) buf; (void) p
-   
 
 
 #define LOCAL_DEPTH_VARS				\
@@ -59,8 +58,10 @@
 
 #define HW_LOCK()				\
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);	\
+   FLUSH_BATCH(mmesa);				\
    LOCK_HARDWARE_QUIESCENT(mmesa);
 
+
 #define HW_CLIPLOOP()						\
   do {								\
     int _nc = mmesa->numClipRects;				\
@@ -97,9 +98,9 @@
 #define READ_RGBA( rgba, _x, _y )				\
 do {								\
    GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);	\
-   rgba[0] = (p >> 8) & 0xf8;					\
-   rgba[1] = (p >> 3) & 0xfc;					\
-   rgba[2] = (p << 3) & 0xf8;					\
+   rgba[0] = (((p >> 11) & 0x1f) * 255) / 31;			\
+   rgba[1] = (((p >>  5) & 0x3f) * 255) / 63;			\
+   rgba[2] = (((p >>  0) & 0x1f) * 255) / 31;			\
    rgba[3] = 255;						\
 } while(0)
 
@@ -127,7 +128,7 @@ do {								\
 	rgba[0] = (p >> 16) & 0xff;				\
 	rgba[1] = (p >> 8)  & 0xff;				\
 	rgba[2] = (p >> 0)  & 0xff;				\
-	rgba[3] = (p >> 24) & 0xff;				\
+	rgba[3] = 0xff;						\
     } while (0)
 
 #define TAG(x) mga##x##_8888
@@ -168,13 +169,13 @@ do {								\
 #define WRITE_DEPTH( _x, _y, d ) {			\
    GLuint tmp = *(GLuint *)(buf + _x*4 + _y*pitch);	\
    tmp &= 0xff;						\
-   tmp |= (d) & 0xffffff00;				\
+   tmp |= (d) << 8;					\
    *(GLuint *)(buf + _x*4 + _y*pitch) = tmp;		\
 }
 
-#define READ_DEPTH( d, _x, _y )		\
-   d = *(GLuint *)(buf + _x*4 + _y*pitch) & ~0xff;	
-
+#define READ_DEPTH( d, _x, _y )	{				\
+   d = (*(GLuint *)(buf + _x*4 + _y*pitch) & ~0xff) >> 8;	\
+}
 
 #define TAG(x) mga##x##_24_8
 #include "depthtmp.h"
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgastate.c b/xc/lib/GL/mesa/src/drv/mga/mgastate.c
index eb45233e0..54cc4de29 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgastate.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgastate.c
@@ -190,7 +190,9 @@ static void mgaUpdateZMode(const GLcontext *ctx)
    if (ctx->Depth.Test) {
       switch(ctx->Depth.Func)  {
       case GL_NEVER:
-	 zmode = DC_zmode_nozcmp; break;
+         /* can't do this in h/w, we'll use a s/w fallback */
+	 zmode = DC_zmode_nozcmp;
+         break;
       case GL_ALWAYS:
 	 zmode = DC_zmode_nozcmp; break;
       case GL_LESS:
@@ -208,15 +210,17 @@ static void mgaUpdateZMode(const GLcontext *ctx)
       default:
 	 break;
       }
-   } else {
+
+      if (ctx->Depth.Mask)
+         zmode |= DC_atype_zi;
+      else
+         zmode |= DC_atype_i;
+   }
+   else {
       zmode |= DC_zmode_nozcmp;
+      zmode |= DC_atype_i;  /* don't write to zbuffer */
    }
 
-   if (ctx->Depth.Mask)
-      zmode |= DC_atype_zi;
-   else
-      zmode |= DC_atype_i;
-
 #if defined(ACCEL_ROP)
    mmesa->setup.dwgctl &= DC_bop_MASK;
    if (ctx->Color.ColorLogicOpEnabled)
@@ -287,6 +291,10 @@ static void mgaDDDepthFunc(GLcontext *ctx, GLenum func)
 {
    FLUSH_BATCH( MGA_CONTEXT(ctx) );
    MGA_CONTEXT(ctx)->new_state |= MGA_NEW_DEPTH;
+   if (func == GL_NEVER && ctx->Depth.Test)
+      MGA_CONTEXT(ctx)->Fallback |= MGA_FALLBACK_DEPTH;
+   else
+      MGA_CONTEXT(ctx)->Fallback &= ~MGA_FALLBACK_DEPTH;
 }
 
 static void mgaDDDepthMask(GLcontext *ctx, GLboolean flag)
@@ -439,7 +447,10 @@ static void mgaUpdateAlphaMode(GLcontext *ctx)
 	    a |= AC_src_zero;
 	 break;
       case GL_SRC_ALPHA_SATURATE:
-	 a |= AC_src_src_alpha_sat;
+         if (ctx->Visual->AlphaBits > 0)
+            a |= AC_src_src_alpha_sat;
+         else
+            a |= AC_src_zero;
 	 break;
       default:		/* never happens */
 	 break;
@@ -499,10 +510,10 @@ void mgaUpdateClipping(const GLcontext *ctx)
    if (mmesa->driDrawable)
    {
       int x1 = mmesa->driDrawable->x + ctx->Scissor.X;
-      int y1 = mmesa->driDrawable->y + mmesa->driDrawable->h - (ctx->Scissor.Y+
-								ctx->Scissor.Height);
-      int x2 = mmesa->driDrawable->x + ctx->Scissor.X+ctx->Scissor.Width;
-      int y2 = mmesa->driDrawable->y + mmesa->driDrawable->h - ctx->Scissor.Y;
+      int y1 = mmesa->driDrawable->y + mmesa->driDrawable->h
+               - (ctx->Scissor.Y + ctx->Scissor.Height);
+      int x2 = x1 + ctx->Scissor.Width - 1;
+      int y2 = y1 + ctx->Scissor.Height - 1;
 
       if (x1 < 0) x1 = 0;
       if (y1 < 0) y1 = 0;
@@ -636,7 +647,7 @@ static GLboolean mgaDDColorMask(GLcontext *ctx,
       mmesa->dirty |= MGA_UPLOAD_CONTEXT;
    }
 
-   return 1;
+   return 0;  /* Kind of a hack */
 }
 
 /* =============================================================
@@ -644,9 +655,12 @@ static GLboolean mgaDDColorMask(GLcontext *ctx,
  *
  * The mga supports a subset of possible 4x4 stipples natively, GL
  * wants 32x32.  Fortunately stipple is usually a repeating pattern.
+ *
+ * Note: the fully opaque pattern (0xffff) has been disabled in order
+ * to work around a conformance issue.
  */
 static int mgaStipples[16] = {
-   0xffff,
+   0xffff1,  /* See above note */
    0xa5a5,
    0x5a5a,
    0xa0a0,
@@ -674,7 +688,12 @@ static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
    GLuint stipple;
 
    FLUSH_BATCH(mmesa);
-   ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
+
+   /* Turn off flags.  We'll turn them on below if this stipple pattern
+    * works in h/w.
+    */
+   ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+   mmesa->canDoStipple = GL_FALSE;
 
    if (active) {
       mmesa->dirty |= MGA_UPLOAD_CONTEXT;
@@ -690,7 +709,6 @@ static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
       for (j = 0 ; j < 4; j++)
 	 for (i = 0 ; i < 4 ; i++)
 	    if (*m++ != p[j]) {
-	       ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
 	       return;
 	    }
 
@@ -706,13 +724,15 @@ static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
       }
 
    if (i == 16) {
-      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
       return;
    }
 
+   mmesa->canDoStipple = GL_TRUE;
+
    if (active) {
       mmesa->setup.dwgctl &= ~(0xf<<20);
       mmesa->setup.dwgctl |= mmesa->poly_stipple;
+      ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
    }
 }
 
@@ -778,6 +798,16 @@ void mgaEmitHwStateLocked( mgaContextPtr mmesa )
    mmesa->sarea->dirty |= mmesa->dirty;
 
    mmesa->dirty &= (MGA_UPLOAD_CLIPRECTS|MGA_WAIT_AGE);
+
+   /* This is a bit of a hack but seems to be the best place to ensure
+    * that separate specular is disabled when not needed.
+    */
+   if (mmesa->glCtx->Texture.ReallyEnabled == 0 ||
+       !mmesa->glCtx->Light.Enabled ||
+       mmesa->glCtx->Light.Model.ColorControl == GL_SINGLE_COLOR) {
+      sarea->TexState[0].texctl2 &= ~TMC_specen_enable;
+      sarea->TexState[1].texctl2 &= ~TMC_specen_enable;
+   }
 }
 
 
@@ -808,6 +838,10 @@ static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
    case GL_DEPTH_TEST:
       FLUSH_BATCH( mmesa );
       mmesa->new_state |= MGA_NEW_DEPTH;
+      if (ctx->Depth.Func == GL_NEVER && ctx->Depth.Test)
+         mmesa->Fallback |= MGA_FALLBACK_DEPTH;
+      else
+         mmesa->Fallback &= ~MGA_FALLBACK_DEPTH;
       break;
    case GL_SCISSOR_TEST:
       FLUSH_BATCH( mmesa );
@@ -829,14 +863,20 @@ static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
       mmesa->new_state |= (MGA_NEW_TEXTURE|MGA_NEW_ALPHA);
       break;
    case GL_POLYGON_STIPPLE:
-      if ((ctx->Driver.TriangleCaps & DD_TRI_STIPPLE) &&
-	  ctx->PB->primitive == GL_POLYGON)
-      {
-	 FLUSH_BATCH(mmesa);
-	 mmesa->dirty |= MGA_UPLOAD_CONTEXT;
-	 mmesa->setup.dwgctl &= ~(0xf<<20);
-	 if (state)
+      FLUSH_BATCH(mmesa);
+      mmesa->dirty |= MGA_UPLOAD_CONTEXT;
+      mmesa->setup.dwgctl &= ~(0xf<<20);
+      if (state) {
+         if (mmesa->canDoStipple && ctx->PB->primitive == GL_POLYGON) {
 	    mmesa->setup.dwgctl |= mmesa->poly_stipple;
+            ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
+         }
+         else {
+            ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+         }
+      }
+      else {
+         ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
       }
       break;
    case GL_COLOR_LOGIC_OP:
@@ -948,11 +988,6 @@ void mgaDDUpdateHwState( GLcontext *ctx )
 }
 
 
-
-
-
-
-
 void mgaDDReducedPrimitiveChange( GLcontext *ctx, GLenum prim )
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
@@ -960,20 +995,24 @@ void mgaDDReducedPrimitiveChange( GLcontext *ctx, GLenum prim )
    FLUSH_BATCH( mmesa );
    mgaUpdateCull(ctx);
 
-   if (ctx->Polygon.StippleFlag && (ctx->Driver.TriangleCaps & DD_TRI_STIPPLE))
-   {
+   if (ctx->Polygon.StippleFlag) {
       mmesa->dirty |= MGA_UPLOAD_CONTEXT;
       mmesa->setup.dwgctl &= ~(0xf<<20);
-      if (ctx->PB->primitive == GL_POLYGON)
+      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
+      if (ctx->PB->primitive == GL_POLYGON && mmesa->canDoStipple) {
 	 mmesa->setup.dwgctl |= mmesa->poly_stipple;
+         ctx->Driver.TriangleCaps |= DD_TRI_STIPPLE;
+      }
+   }
+   else {
+      ctx->Driver.TriangleCaps &= ~DD_TRI_STIPPLE;
    }
 }
 
 
 #define INTERESTED (~(NEW_MODELVIEW|NEW_PROJECTION|\
                       NEW_TEXTURE_MATRIX|\
-                      NEW_USER_CLIP|NEW_CLIENT_STATE|\
-                      NEW_TEXTURE_ENABLE))
+                      NEW_USER_CLIP|NEW_CLIENT_STATE))
 
 void mgaDDUpdateState( GLcontext *ctx )
 {
@@ -999,6 +1038,9 @@ void mgaDDUpdateState( GLcontext *ctx )
       ctx->Driver.TriangleFunc=mmesa->TriangleFunc;
       ctx->Driver.QuadFunc=mmesa->QuadFunc;
    }
+   else {
+      ctx->IndirectTriangles |= mmesa->IndirectTriangles;
+   }
 }
 
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatex.c b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
index 03f1e39f4..35945224a 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatex.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
@@ -37,9 +37,11 @@
 #include "mgaregs.h"
 #include "mgaioctl.h"
 
+#include "context.h"
 #include "enums.h"
 #include "simple_list.h"
 #include "mem.h"
+#include "texutil.h"
 
 #define TEX_0 1
 #define TEX_1 2
@@ -144,11 +146,166 @@ static void mgaSetTexBorderColor(mgaTextureObjectPtr t, GLubyte color[4])
 }
 
 
+static GLint mgaChooseTexFormat( mgaContextPtr mmesa,
+				 struct gl_texture_image *texImage,
+				 GLenum format, GLenum type )
+{
+   const GLboolean do32bpt = mmesa->default32BitTextures;
+   const struct gl_texture_format *texFormat;
+   GLint ret;
+
+   if ( 0 )
+      fprintf( stderr, "internal=%s format=%s type=%s\n",
+	       texImage->IntFormat == 3 ? "GL_RGB (3)" :
+	       texImage->IntFormat == 4 ? "GL_RGBA (4)" :
+	       gl_lookup_enum_by_nr( texImage->IntFormat ),
+	       gl_lookup_enum_by_nr( format ),
+	       gl_lookup_enum_by_nr( type ) );
+
+#define SET_FORMAT( r, gl )						\
+   do {									\
+      ret = (r);							\
+      texFormat = &(gl);						\
+   } while (0)
+
+#define SET_FORMAT_32BPT( r32, gl32, r16, gl16 )			\
+   do {									\
+      if ( do32bpt ) {							\
+	 ret = (r32);							\
+	 texFormat = &(gl32);						\
+      } else {								\
+	 ret = (r16);							\
+	 texFormat = &(gl16);						\
+      }									\
+   } while (0)
+
+   switch ( texImage->IntFormat ) {
+      /* GH: Bias towards GL_RGB, GL_RGBA texture formats.  This has
+       * got to be better than sticking them way down the end of this
+       * huge list.
+       */
+   case GL_RGBA:
+   case 4:
+      if ( format == GL_BGRA ) {
+	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
+	    SET_FORMAT( TMC_tformat_tw32, _mesa_texformat_argb8888 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
+	    SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
+	    SET_FORMAT( TMC_tformat_tw15, _mesa_texformat_argb1555 );
+	    break;
+	 }
+      }
+      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
+			TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGB:
+   case 3:
+      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
+	 SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
+	 break;
+      }
+      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
+			TMC_tformat_tw16, _mesa_texformat_rgb565 );
+      break;
+
+      /* GH: Okay, keep checking as normal.  Still test for GL_RGB,
+       * GL_RGBA formats first.
+       */
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
+			TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGBA4:
+   case GL_RGBA2:
+      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGB5_A1:
+      SET_FORMAT( TMC_tformat_tw15, _mesa_texformat_argb1555 );
+      break;
+
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
+			TMC_tformat_tw16, _mesa_texformat_rgb565 );
+      break;
+
+   case GL_RGB5:
+   case GL_RGB4:
+   case GL_R3_G3_B2:
+      SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
+      break;
+
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+      /* FIXME: This will report incorrect component sizes... */
+      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
+
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+      /* FIXME: This will report incorrect component sizes... */
+      SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
+      break;
 
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+      /* FIXME: This will report incorrect component sizes... */
+      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
 
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+      /* FIXME: This will report incorrect component sizes... */
+      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
+      break;
 
+   case GL_COLOR_INDEX:
+   case GL_COLOR_INDEX1_EXT:
+   case GL_COLOR_INDEX2_EXT:
+   case GL_COLOR_INDEX4_EXT:
+   case GL_COLOR_INDEX8_EXT:
+   case GL_COLOR_INDEX12_EXT:
+   case GL_COLOR_INDEX16_EXT:
+      SET_FORMAT( TMC_tformat_tw8, _mesa_texformat_ci8 );
+      break;
+
+   default:
+      fprintf( stderr, "bad texture format in mgaChooseTexFormat() %d",
+	       texImage->IntFormat );
+      return -1;
+   }
 
+   texImage->TexFormat = texFormat;
 
+   return ret;
+}
 
 
 /*
@@ -160,60 +317,27 @@ static void mgaSetTexBorderColor(mgaTextureObjectPtr t, GLubyte color[4])
 static void mgaCreateTexObj(mgaContextPtr mmesa,
 			    struct gl_texture_object *tObj)
 {
-   const struct gl_texture_image *image = tObj->Image[ 0 ];
+   const GLint baseLevel = tObj->BaseLevel;
+   struct gl_texture_image *image = tObj->Image[baseLevel];
    mgaTextureObjectPtr t;
    int i, ofs;
    int LastLevel;
    int s, s2;
-   int textureFormat;
-
+   int tformat;
 
    if (!image) return;
 
-
    tObj->DriverData = t = CALLOC( sizeof( *t ) );
    if (!t) {
       fprintf(stderr, "mgaCreateTexObj: Failed to malloc mgaTextureObject\n" );
       return;
    }
 
-   switch( image->Format ) {
-   case GL_RGB:
-   case GL_LUMINANCE:
-      if ( image->IntFormat != GL_RGB5 && ( image->IntFormat == GL_RGB8 ||
-					    mmesa->default32BitTextures ) ) {
-	 t->texelBytes = 4;
-	 textureFormat = TMC_tformat_tw32;
-      } else {
-	 t->texelBytes = 2;
-	 textureFormat = TMC_tformat_tw16;
-      }
-      break;
-   case GL_ALPHA:
-   case GL_LUMINANCE_ALPHA:
-   case GL_INTENSITY:
-   case GL_RGBA:
-      if ( image->IntFormat != GL_RGBA4 && ( image->IntFormat == GL_RGBA8 ||
-					     mmesa->default32BitTextures ) ) {
-	 t->texelBytes = 4;
-	 textureFormat = TMC_tformat_tw32;
-      } else {
-	 t->texelBytes = 2;
-	 textureFormat = TMC_tformat_tw12;
-      }
-      break;
-   case GL_COLOR_INDEX:
-      textureFormat = TMC_tformat_tw8;
-      t->texelBytes = 1;
-      break;
-   default:
-      fprintf(stderr, "mgaCreateTexObj: bad image->Format %x/%s\n",
-	      image->Format,
-	      gl_lookup_enum_by_nr(image->Format));
-      FREE( t );
-      tObj->DriverData = 0;
-      return;
-   }
+   /* FIXME: Use the real DD interface...
+    */
+   tformat = mgaChooseTexFormat( mmesa, image, image->Format,
+				 GL_UNSIGNED_BYTE );
+   t->texelBytes = image->TexFormat->TexelBytes;
 
    /* We are going to upload all levels that are present, even if
     * later levels wouldn't be used by the current filtering mode.  This
@@ -236,7 +360,6 @@ static void mgaCreateTexObj(mgaContextPtr mmesa,
 	       MAX2( tObj->Image[i]->Height, 8 ) *
 	       t->texelBytes) + 31) & ~31;
    }
-
    t->totalSize = ofs;
    t->lastLevel = LastLevel;
    t->tObj = tObj;
@@ -249,9 +372,7 @@ static void mgaCreateTexObj(mgaContextPtr mmesa,
 
 
    /* setup hardware register values */
-   t->setup.texctl = (TMC_takey_1 |
-		      TMC_tamask_0 |
-		      textureFormat );
+   t->setup.texctl = TMC_takey_1 | TMC_tamask_0 | tformat;
 
    if (image->WidthLog2 >= 3)
       t->setup.texctl |= ((image->WidthLog2 - 3) << TMC_tpitch_SHIFT);
@@ -331,6 +452,7 @@ static void mgaUpdateTextureEnvG400( GLcontext *ctx, int unit )
    GLuint *reg = ((GLuint *)&mmesa->setup.tdualstage0 + unit);
    GLuint source = mmesa->tmu_source[unit];
    struct gl_texture_object *tObj = ctx->Texture.Unit[source].Current;
+   GLenum format;
 
    if ( tObj != ctx->Texture.Unit[source].CurrentD[2] ||
 	!tObj ||
@@ -338,95 +460,183 @@ static void mgaUpdateTextureEnvG400( GLcontext *ctx, int unit )
 	((ctx->Enabled>>(source*4))&TEXTURE0_ANY) != TEXTURE0_2D )
       return;
 
+   format = tObj->Image[tObj->BaseLevel]->Format;
 
    switch (ctx->Texture.Unit[source].EnvMode) {
    case GL_REPLACE:
-      *reg = (TD0_color_sel_arg1 |
-	      TD0_alpha_sel_arg1 );
+      if (format == GL_RGB || format == GL_LUMINANCE) {
+	 *reg = (TD0_color_sel_arg1 |
+                 TD0_alpha_arg2_diffuse |
+		 TD0_alpha_sel_arg2 );
+      }
+      else if (format == GL_ALPHA) {
+         *reg = (TD0_color_sel_arg2 |
+                 TD0_color_arg2_diffuse |
+                 TD0_alpha_sel_arg1 );
+      }
+      else {
+         *reg = (TD0_color_sel_arg1 |
+                 TD0_alpha_sel_arg1 );
+      }
       break;
 
    case GL_MODULATE:
-      if (unit == 0)
+      if (unit == 0) {
 	 *reg = ( TD0_color_arg2_diffuse |
 		  TD0_color_sel_mul |
 		  TD0_alpha_arg2_diffuse |
 		  TD0_alpha_sel_mul);
-      else
+      }
+      else {
 	 *reg = ( TD0_color_arg2_prevstage |
 		  TD0_color_alpha_prevstage |
 		  TD0_color_sel_mul |
 		  TD0_alpha_arg2_prevstage |
 		  TD0_alpha_sel_mul);
+      }
       break;
    case GL_DECAL:
-      if (tObj->Image[0]->Format == GL_RGB)
-	 *reg = (TD0_color_sel_arg1 |
-		 TD0_alpha_sel_arg1 );
-      else if (unit == 0)
-	 *reg = (TD0_color_arg2_diffuse |
-		 TD0_color_alpha_currtex |
-		 TD0_color_alpha2inv_enable |
-		 TD0_color_arg2mul_alpha2 |
-		 TD0_color_arg1mul_alpha1 |
-		 TD0_color_add_add |
-		 TD0_color_sel_add |
-		 TD0_alpha_arg2_diffuse |
-		 TD0_alpha_sel_arg2 );
-      else
-	 *reg = (TD0_color_arg2_prevstage |
-		 TD0_color_alpha_currtex |
-		 TD0_color_alpha2inv_enable |
-		 TD0_color_arg2mul_alpha2 |
-		 TD0_color_arg1mul_alpha1 |
-		 TD0_color_add_add |
-		 TD0_color_sel_add |
-		 TD0_alpha_arg2_prevstage |
-		 TD0_alpha_sel_arg2 );
-
+      if (format == GL_RGB) {
+         if (unit == 0) {
+            *reg = (TD0_color_sel_arg1 |
+                    TD0_alpha_arg2_diffuse |
+                    TD0_alpha_sel_arg2 );
+         }
+         else {
+            *reg = (TD0_color_sel_arg1 |
+                    TD0_alpha_arg2_prevstage |
+                    TD0_alpha_sel_arg2 );
+         }
+      }
+      else if ( format == GL_RGBA ) {
+#if 0
+         if (unit == 0) {
+            /* this doesn't work */
+            *reg = (TD0_color_arg2_diffuse |
+                    TD0_color_alpha_currtex |
+                    TD0_color_alpha2inv_enable |
+                    TD0_color_arg2mul_alpha2 |
+                    TD0_color_arg1mul_alpha1 |
+                    TD0_color_blend_enable |
+                    TD0_color_arg1add_mulout |
+                    TD0_color_arg2add_mulout |
+                    TD0_color_add_add |
+                    TD0_color_sel_mul |
+                    TD0_alpha_arg2_diffuse |
+                    TD0_alpha_sel_arg2 );
+         }
+         else {
+            *reg = (TD0_color_arg2_prevstage |
+                    TD0_color_alpha_currtex |
+                    TD0_color_alpha2inv_enable |
+                    TD0_color_arg2mul_alpha2 |
+                    TD0_color_arg1mul_alpha1 |
+                    TD0_color_add_add |
+                    TD0_color_sel_add |
+                    TD0_alpha_arg2_prevstage |
+                    TD0_alpha_sel_arg2 );
+         }
+#else
+         /* s/w fallback, pretty sure we can't do in h/w */
+	 mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
+#endif
+      }
+      else {
+	 if (unit == 0) {
+	    *reg = ( TD0_color_arg2_diffuse |
+		     TD0_color_sel_arg2 |
+		     TD0_alpha_arg2_diffuse |
+		     TD0_alpha_sel_arg2);
+	 }
+	 else {
+	    *reg = ( TD0_color_arg2_prevstage |
+		     TD0_color_sel_arg2 |
+		     TD0_alpha_arg2_prevstage |
+		     TD0_alpha_sel_arg2);
+	 }
+      }
       break;
 
    case GL_ADD:
-      if (unit == 0)
-	 *reg = ( TD0_color_arg2_diffuse |
-		  TD0_color_add_add |
-		  TD0_color_sel_add |
-		  TD0_alpha_arg2_diffuse |
-		  TD0_alpha_sel_add);
-      else
-	 *reg = ( TD0_color_arg2_prevstage |
-		  TD0_color_alpha_prevstage |
-		  TD0_color_add_add |
-		  TD0_color_sel_add |
-		  TD0_alpha_arg2_prevstage |
-		  TD0_alpha_sel_add);
+      if (unit == 0) {
+         if (format == GL_INTENSITY)
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_add_enable |
+                     TD0_alpha_sel_add);
+         else if (format == GL_ALPHA)
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_sel_mul |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_sel_mul);
+         else
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_sel_mul);
+      }
+      else {
+         if (format == GL_INTENSITY) {
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_add_enable |
+                     TD0_alpha_sel_add);
+         }
+         else if (format == GL_ALPHA) {
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_sel_mul |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_sel_mul);
+         }
+         else {
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_alpha_prevstage |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_sel_mul);
+         }
+      }
       break;
 
    case GL_BLEND:
-      if (mmesa->blend_flags)
-	 mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
-
-      /* Do singletexture GL_BLEND with 'all ones' env-color
-       * by using both texture units.  Multitexture gl_blend
-       * is a fallback.
-       */
-      if (unit == 0) {
-	 /* Part 1: R1 = Rf ( 1 - Rt )
-	  *         A1 = Af At
-	  */
+      if (format == GL_ALPHA) {
 	 *reg = ( TD0_color_arg2_diffuse |
-		  TD0_color_arg1_inv_enable |
 		  TD0_color_sel_mul |
 		  TD0_alpha_arg2_diffuse |
-		  TD0_alpha_sel_arg1);
-      } else {
-	 /* Part 2: R2 = R1 + Rt
-	  *         A2 = A1
-	  */
-	 *reg = ( TD0_color_arg2_prevstage |
-		  TD0_color_add_add |
-		  TD0_color_sel_add |
-		  TD0_alpha_arg2_prevstage |
-		  TD0_alpha_sel_arg2);
+		  TD0_alpha_sel_mul);
+      }
+      else {
+	 mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
+
+         /* Do singletexture GL_BLEND with 'all ones' env-color
+          * by using both texture units.  Multitexture gl_blend
+          * is a fallback.
+          */
+         if (unit == 0) {
+            /* Part 1: R1 = Rf ( 1 - Rt )
+             *         A1 = Af At
+             */
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_arg1_inv_enable |
+                     TD0_color_sel_mul |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_sel_arg1);
+         } else {
+            /* Part 2: R2 = R1 + Rt
+             *         A2 = A1
+             */
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_sel_arg2);
+         }
       }
       break;
    default:
@@ -459,6 +669,11 @@ static void mgaUpdateTextureObject( GLcontext *ctx, int unit )
       return;
    }
 
+   if (tObj->Image[tObj->BaseLevel]->Border > 0) {
+      mmesa->Fallback |= MGA_FALLBACK_TEXTURE;
+      return;
+   }
+
 /*     if (!tObj) tObj = ctx->Texture.Unit[0].Current; */
 /*     if (!tObj) return; */
 
@@ -597,6 +812,9 @@ static void mgaDDTexImage( GLcontext *ctx, GLenum target,
 {
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
    mgaTextureObjectPtr t;
+   GLint tformat;
+   /* hack: cast-away const */
+   struct gl_texture_image *img = (struct gl_texture_image *) image;
 
    /* just free the mga texture if it exists, it will be recreated at
       mgaUpdateTextureState time. */
@@ -608,6 +826,9 @@ static void mgaDDTexImage( GLcontext *ctx, GLenum target,
       mmesa->new_state |= MGA_NEW_TEXTURE;
    }
 
+   tformat = mgaChooseTexFormat( mmesa, img, img->Format,
+				 GL_UNSIGNED_BYTE );
+
    if (0)
       fprintf(stderr, "mgaDDTexImage tObj %p, level %d, image %p\n",
 	      tObj, level, image);
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatris.c b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
index cd051bf26..4caa1bfa4 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatris.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
@@ -38,25 +38,6 @@
 #include "mgavb.h"
 
 
-static void mga_null_quad( GLcontext *ctx, GLuint v0,
-			   GLuint v1, GLuint v2, GLuint v3, GLuint pv ) 
-{
-}     
-
-static void mga_null_triangle( GLcontext *ctx, GLuint v0,
-			       GLuint v1, GLuint v2, GLuint pv ) 
-{
-}     
-
-static void mga_null_line( GLcontext *ctx, GLuint v1, GLuint v2, GLuint pv ) 
-{
-}
-
-static void mga_null_points( GLcontext *ctx, GLuint first, GLuint last ) 
-{
-}
-
-
 #define MGA_COLOR(to, from) {			\
   (to)[0] = (from)[2];				\
   (to)[1] = (from)[1];				\
@@ -64,6 +45,12 @@ static void mga_null_points( GLcontext *ctx, GLuint first, GLuint last )
   (to)[3] = (from)[3];				\
 }
 
+#define MGA_COLOR3(to, from) {			\
+  (to)[0] = (from)[2];				\
+  (to)[1] = (from)[1];				\
+  (to)[2] = (from)[0];				\
+}
+
 
 
 static triangle_func tri_tab[0x10];   
@@ -106,9 +93,6 @@ static points_func   points_tab[0x10];
 
 void mgaDDTrifuncInit()
 {
-   int i;
-
-
    init();
    init_flat();
    init_offset();
@@ -117,14 +101,6 @@ void mgaDDTrifuncInit()
    init_twoside_flat();
    init_twoside_offset();
    init_twoside_offset_flat();
-
-   for (i = 0 ; i < 0x20 ; i++) 
-      if (i & MGA_NODRAW_BIT) {
-	 quad_tab[i] = mga_null_quad; 
-	 tri_tab[i] = mga_null_triangle; 
-	 line_tab[i] = mga_null_line;
-	 points_tab[i] = mga_null_points;
-      }
 }
 
 
@@ -134,7 +110,7 @@ void mgaDDTrifuncInit()
 #define LINE_FALLBACK (ALL_FALLBACK | DD_LINE_SMOOTH | DD_LINE_STIPPLE)
 #define TRI_FALLBACK (ALL_FALLBACK | DD_TRI_SMOOTH | DD_TRI_UNFILLED)
 #define ANY_FALLBACK (POINT_FALLBACK|LINE_FALLBACK|TRI_FALLBACK|DD_TRI_STIPPLE)
-#define ANY_RASTER_FLAGS (DD_FLATSHADE|DD_TRI_LIGHT_TWOSIDE|DD_TRI_OFFSET|DD_Z_NEVER)
+#define ANY_RASTER_FLAGS (DD_FLATSHADE|DD_TRI_LIGHT_TWOSIDE|DD_TRI_OFFSET)
 
 /* Setup the Point, Line, Triangle and Quad functions based on the
    current rendering state.  Wherever possible, use the hardware to
@@ -147,6 +123,9 @@ void mgaDDChooseRenderState(GLcontext *ctx)
 
     if (mmesa->Fallback) {
 	mmesa->renderindex = MGA_FALLBACK_BIT;
+        if (flags & DD_TRI_LIGHT_TWOSIDE) {
+           mmesa->IndirectTriangles = DD_TRI_LIGHT_TWOSIDE;
+        }
 	return;
     }
 
@@ -154,7 +133,6 @@ void mgaDDChooseRenderState(GLcontext *ctx)
 	if (flags & DD_FLATSHADE)               index |= MGA_FLAT_BIT;
 	if (flags & DD_TRI_LIGHT_TWOSIDE)       index |= MGA_TWOSIDE_BIT;
 	if (flags & DD_TRI_OFFSET)              index |= MGA_OFFSET_BIT; 
-	if (flags & DD_Z_NEVER)                 index |= MGA_NODRAW_BIT; 
     }
 	
     mmesa->PointsFunc = points_tab[index];
@@ -194,6 +172,9 @@ void mgaDDChooseRenderState(GLcontext *ctx)
 	    mmesa->QuadFunc = 0;
 	    mmesa->IndirectTriangles |= (DD_TRI_SW_RASTERIZE |
 					 DD_QUAD_SW_RASTERIZE);
+            if (flags & DD_TRI_LIGHT_TWOSIDE) {
+               mmesa->IndirectTriangles |= DD_TRI_LIGHT_TWOSIDE;
+            }
 	}
     }
 }
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatris.h b/xc/lib/GL/mesa/src/drv/mga/mgatris.h
index f50c83974..9880bbdf1 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatris.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatris.h
@@ -38,8 +38,7 @@ extern void mgaDDTrifuncInit( void );
 #define MGA_FLAT_BIT	    0x1
 #define MGA_OFFSET_BIT	    0x2	
 #define MGA_TWOSIDE_BIT	    0x4	
-#define MGA_NODRAW_BIT	    0x8
-#define MGA_FALLBACK_BIT    0x10
+#define MGA_FALLBACK_BIT    0x8
 
 static __inline void mga_draw_triangle( mgaContextPtr mmesa,
 				      mgaVertex *v0, 
@@ -90,40 +89,42 @@ static __inline void mga_draw_point( mgaContextPtr mmesa,
 {
    GLuint vertsize = mmesa->vertsize;
    GLuint *wv = mgaAllocVertexDwords( mmesa, 6*vertsize);
-   int j;
+   GLuint j;
+   const GLfloat x = tmp->v.x + 0.125;
+   const GLfloat y = tmp->v.y - 0.125;
 
-   *(float *)&wv[0] = tmp->v.x - sz;
-   *(float *)&wv[1] = tmp->v.y - sz;
+   *(float *)&wv[0] = x - sz;
+   *(float *)&wv[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x + sz;
-   *(float *)&wv[1] = tmp->v.y - sz;
+   *(float *)&wv[0] = x + sz;
+   *(float *)&wv[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x + sz;
-   *(float *)&wv[1] = tmp->v.y + sz;
+   *(float *)&wv[0] = x + sz;
+   *(float *)&wv[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x + sz;
-   *(float *)&wv[1] = tmp->v.y + sz;
+   *(float *)&wv[0] = x + sz;
+   *(float *)&wv[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x - sz;
-   *(float *)&wv[1] = tmp->v.y + sz;
+   *(float *)&wv[0] = x - sz;
+   *(float *)&wv[1] = y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->v.x - sz;
-   *(float *)&wv[1] = tmp->v.y - sz;
+   *(float *)&wv[0] = x - sz;
+   *(float *)&wv[1] = y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
 }
@@ -134,55 +135,79 @@ static __inline void mga_draw_line( mgaContextPtr mmesa,
 				  const mgaVertex *tmp1,
 				  float width )
 {
-   GLuint vertsize = mmesa->vertsize;
+   const GLuint vertsize = mmesa->vertsize;
    GLuint *wv = mgaAllocVertexDwords( mmesa, 6 * vertsize );
-   float dx, dy, ix, iy;
-   int j;
+   GLuint j;
+   GLfloat x0 = tmp0->v.x;
+   GLfloat y0 = tmp0->v.y;
+   GLfloat x1 = tmp1->v.x;
+   GLfloat y1 = tmp1->v.y;
+   GLfloat dx, dy, ix, iy;
+   GLfloat hw;
+
+   hw = 0.5F * width;
+   if (hw > 0.1F && hw < 0.5F) {
+      hw = 0.5F;
+   }
 
+   /* adjust vertices depending on line direction */
    dx = tmp0->v.x - tmp1->v.x;
    dy = tmp0->v.y - tmp1->v.y;
-
-   ix = width * .5; iy = 0;
-  
-   if ((ix<.5) && (ix>0.1)) ix = .5; /* I want to see lines with width
-                                        0.5 also */
-  
    if (dx * dx > dy * dy) {
-      iy = ix; ix = 0;
+      /* X-major line */
+      ix = 0.0F;
+      iy = hw;
+      if (x1 < x0) {
+         x0 += 0.5F;
+         x1 += 0.5F;
+      }
+      y0 -= 0.5F;
+      y1 -= 0.5F;
+   }
+   else {
+      /* Y-major line */
+      ix = hw;
+      iy = 0.0F;
+      if (y1 > y0) {
+         y0 -= 0.5F;
+         y1 -= 0.5F;
+      }
+      x0 += 0.5F;
+      x1 += 0.5F;
    }
-  
-   *(float *)&wv[0] = tmp0->v.x - ix;
-   *(float *)&wv[1] = tmp0->v.y - iy;
+
+   *(float *)&wv[0] = x0 - ix;
+   *(float *)&wv[1] = y0 - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->v.x + ix;
-   *(float *)&wv[1] = tmp1->v.y + iy;
+   *(float *)&wv[0] = x1 + ix;
+   *(float *)&wv[1] = y1 + iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp1->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp0->v.x + ix;
-   *(float *)&wv[1] = tmp0->v.y + iy;
+   *(float *)&wv[0] = x0 + ix;
+   *(float *)&wv[1] = y0 + iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 	 
-   *(float *)&wv[0] = tmp0->v.x - ix;
-   *(float *)&wv[1] = tmp0->v.y - iy;
+   *(float *)&wv[0] = x0 - ix;
+   *(float *)&wv[1] = y0 - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->v.x - ix;
-   *(float *)&wv[1] = tmp1->v.y - iy;
+   *(float *)&wv[0] = x1 - ix;
+   *(float *)&wv[1] = y1 - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp1->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->v.x + ix;
-   *(float *)&wv[1] = tmp1->v.y + iy;
+   *(float *)&wv[0] = x1 + ix;
+   *(float *)&wv[1] = y1 + iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp1->ui[j];
    wv += vertsize;
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
index f578871ce..a041d887d 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
@@ -16,6 +16,7 @@ static __inline void TAG(triangle)(GLcontext *ctx,
 
 #if (IND & (MGA_TWOSIDE_BIT | MGA_FLAT_BIT))
    GLuint c[3];
+   GLuint s[3];
 #endif
 
    v[0] = &mgaverts[e0];
@@ -26,6 +27,9 @@ static __inline void TAG(triangle)(GLcontext *ctx,
    c[0] = v[0]->ui[4];
    c[1] = v[1]->ui[4];
    c[2] = v[2]->ui[4];
+   s[0] = v[0]->ui[5];
+   s[1] = v[1]->ui[5];
+   s[2] = v[2]->ui[5];
 #endif
 
 
@@ -41,13 +45,19 @@ static __inline void TAG(triangle)(GLcontext *ctx,
       {
 	 GLuint  facing        = (cc > 0.0) ^ ctx->Polygon.FrontBit;
 	 GLubyte (*vbcolor)[4] = VB->Color[facing]->data;
+	 GLubyte (*vbspec)[4] = VB->Spec[facing];
 	 if (IND & MGA_FLAT_BIT) {
 	    MGA_COLOR((char *)&v[0]->ui[4], vbcolor[pv]);
 	    v[2]->ui[4] = v[1]->ui[4] = v[0]->ui[4];
+            MGA_COLOR3((char *)&v[0]->ui[5], vbspec[pv]);
+	    v[2]->ui[5] = v[1]->ui[5] = v[0]->ui[5];
 	 } else {
 	    MGA_COLOR((char *)&v[0]->ui[4], vbcolor[e0]);
 	    MGA_COLOR((char *)&v[1]->ui[4], vbcolor[e1]);
 	    MGA_COLOR((char *)&v[2]->ui[4], vbcolor[e2]);
+	    MGA_COLOR3((char *)&v[0]->ui[5], vbspec[e0]);
+	    MGA_COLOR3((char *)&v[1]->ui[5], vbspec[e1]);
+	    MGA_COLOR3((char *)&v[2]->ui[5], vbspec[e2]);
 	 }
       }
 #endif
@@ -79,9 +89,13 @@ static __inline void TAG(triangle)(GLcontext *ctx,
 #elif (IND & MGA_FLAT_BIT)
    {
       GLuint color = mgaverts[pv].ui[4];
+      GLuint spec = mgaverts[pv].ui[5];
       v[0]->ui[4] = color;
       v[1]->ui[4] = color;
       v[2]->ui[4] = color;
+      v[0]->ui[5] = spec;
+      v[1]->ui[5] = spec;
+      v[2]->ui[5] = spec;
    }
 #endif
 
@@ -97,6 +111,9 @@ static __inline void TAG(triangle)(GLcontext *ctx,
    v[0]->ui[4] = c[0];
    v[1]->ui[4] = c[1];
    v[2]->ui[4] = c[2];    
+   v[0]->ui[5] = s[0];
+   v[1]->ui[5] = s[1];
+   v[2]->ui[5] = s[2];    
 #endif
 
 }
@@ -120,25 +137,34 @@ static void TAG(line)( GLcontext *ctx, GLuint v0, GLuint v1, GLuint pv )
    float width = ctx->Line.Width;
    GLfloat z0, z1;
    GLuint c0, c1;
+   GLuint s0, s1;
    mgaVertex *vert0 = &mgaVB[v0];
    mgaVertex *vert1 = &mgaVB[v1];
 
-
    if (IND & MGA_TWOSIDE_BIT) {
       GLubyte (*vbcolor)[4] = ctx->VB->ColorPtr->data;
+      GLubyte (*vbspec)[4] = ctx->VB->Specular;
 
       if (IND & MGA_FLAT_BIT) {
 	 MGA_COLOR((char *)&vert0->v.color,vbcolor[pv]);
 	 *(int *)&vert1->v.color = *(int *)&vert0->v.color;
+         MGA_COLOR3((char *)&vert0->v.specular, vbspec[pv]);
+         *(int *)&vert1->v.specular = *(int *)&vert0->v.specular;
       } else {
 	 MGA_COLOR((char *)&vert0->v.color,vbcolor[v0]);
 	 MGA_COLOR((char *)&vert1->v.color,vbcolor[v1]);
+         MGA_COLOR3((char *)&vert0->v.specular, vbspec[v0]);
+         MGA_COLOR3((char *)&vert1->v.specular, vbspec[v1]);
       }
    } else if (IND & MGA_FLAT_BIT) {
       c0 = *(GLuint *) &(vert0->v.color);
       c1 = *(GLuint *) &(vert1->v.color);
       *(int *)&vert0->v.color = 
 	 *(int *)&vert1->v.color = *(int *)&mgaVB[pv].v.color;
+      s0 = *(GLuint *) &(vert0->v.specular);
+      s1 = *(GLuint *) &(vert1->v.specular);
+      *(int *)&vert0->v.specular = 
+         *(int *)&vert1->v.specular = *(int *)&mgaVB[pv].v.specular;
    }
 
    if (IND & MGA_OFFSET_BIT) {
@@ -159,6 +185,8 @@ static void TAG(line)( GLcontext *ctx, GLuint v0, GLuint v1, GLuint pv )
    if ((IND & MGA_FLAT_BIT) && !(IND & MGA_TWOSIDE_BIT)) {
       *(GLuint *) &(vert0->v.color) = c0;
       *(GLuint *) &(vert1->v.color) = c1;
+      *(GLuint *) &(vert0->v.specular) = s0;
+      *(GLuint *) &(vert1->v.specular) = s1;
    }
 }
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgavb.c b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
index ae6316fc8..1dfc89d2b 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgavb.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
@@ -35,34 +35,34 @@
 #include <stdio.h>
 #include <stdlib.h>
 
-#define TEX0 {					\
+#define TEX0 {				\
   v->v.tu0 = tc0[i][0];			\
   v->v.tv0 = tc0[i][1];			\
 }
 
-#define TEX1 {					\
+#define TEX1 {				\
   v->v.tu1 = tc1[i][0];			\
   v->v.tv1 = tc1[i][1];			\
 }
 
-#define SPC {					\
-  GLubyte *spec = &(VB->Spec[0][i][0]);		\
+#define SPC {				\
+  GLubyte *spec = &(VB->Spec[0][i][0]);	\
   v->v.specular.red = spec[0];		\
-  v->v.specular.green = spec[1];		\
+  v->v.specular.green = spec[1];	\
   v->v.specular.blue = spec[2];		\
 }
 
-#define FOG {					\
-  GLubyte *spec = &(VB->Spec[0][i][0]);		\
-  v->v.specular.alpha = spec[3];		\
+#define FOG {				\
+  GLubyte *spec = &(VB->Spec[0][i][0]);	\
+  v->v.specular.alpha = spec[3];	\
 }
 
 #define COL {					\
   GLubyte *col = &(VB->Color[0]->data[i][0]);	\
-  v->v.color.blue  = col[2];		\
-  v->v.color.green = col[1];		\
-  v->v.color.red   = col[0];		\
-  v->v.color.alpha = col[3];		\
+  v->v.color.blue  = col[2];			\
+  v->v.color.green = col[1];			\
+  v->v.color.red   = col[0];			\
+  v->v.color.alpha = col[3];			\
 }
 
 /* The v code we have doesn't seem to support projective texturing
@@ -70,14 +70,13 @@
  * second set of texcoords).  This may be a problem for the g400.  
  */
 #define TEX0_4						\
-  if (VB->TexCoordPtr[0]->size == 4)			\
-  {							\
+  if (VB->TexCoordPtr[0]->size == 4) {			\
      GLfloat (*tc)[4] = VB->TexCoordPtr[0]->data;	\
      v = &(MGA_DRIVER_DATA(VB)->verts[start]);		\
      mmesa->setupdone &= ~MGA_WIN_BIT;			\
-     for (i=start; i < end; i++, v++)	{		\
-        float oow = 1.0 / tc[i][3];			\
-	v->v.rhw *= tc[i][3];			\
+     for (i = start; i < end; i++, v++) {		\
+        GLfloat oow = 1.0 / tc[i][3];			\
+	v->v.rhw *= tc[i][3];				\
 	v->v.tu0 *= oow;				\
 	v->v.tv0 *= oow;				\
      }							\
@@ -138,7 +137,7 @@ static void name(struct vertex_buffer *VB, GLuint start, GLuint end)	\
 	    spec;							\
 	    fog;							\
 	 }								\
-	    col;							\
+         col;								\
       }									\
    tex0_4;								\
 }
diff --git a/xc/lib/GL/mesa/src/drv/radeon/Imakefile b/xc/lib/GL/mesa/src/drv/radeon/Imakefile
index d862565d0..2dc0b3083 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/radeon/Imakefile
@@ -59,6 +59,8 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		radeon_span.c \
 		radeon_state.c \
 		radeon_tex.c \
+		radeon_texmem.c \
+		radeon_texstate.c \
 		radeon_tris.c \
 		radeon_vb.c \
 		radeon_xmesa.c
@@ -74,6 +76,8 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		radeon_span.o \
 		radeon_state.o \
 		radeon_tex.o \
+		radeon_texmem.o \
+		radeon_texstate.o \
 		radeon_tris.o \
 		radeon_vb.o \
 		radeon_xmesa.o
@@ -152,6 +156,7 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		../../stages.c \
 		../../state.c \
 		../../stencil.c \
+		../../texformat.c \
 		../../teximage.c \
 		../../texobj.c \
 		../../texstate.c \
@@ -225,6 +230,7 @@ MESA_INCLUDES = -I. -I.. -I../../include \
 		../../stages.o \
 		../../state.o \
 		../../stencil.o \
+		../../texformat.o \
 		../../teximage.o \
 		../../texobj.o \
 		../../texstate.o \
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_context.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_context.c
index 9b2c5b2f1..172e0f62a 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_context.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_context.c
@@ -163,16 +163,8 @@ GLboolean radeonCreateContext( Display *dpy, GLvisual *glVisual,
       ctx->Const.MaxTextureSize = (1 << 10);
    }
 
-   /* FIXME: Support all available texture units... */
    ctx->Const.MaxTextureUnits = 2;
 
-#if ENABLE_PERF_BOXES
-   if (getenv("LIBGL_PERFORMANCE_BOXES"))
-      rmesa->boxes = 1;
-   else
-      rmesa->boxes = 0;
-#endif
-
    ctx->DriverCtx = (void *)rmesa;
 
    radeonDDInitExtensions( ctx );
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_dd.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_dd.c
index b86506a8d..4ef0ebd30 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_dd.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_dd.c
@@ -46,7 +46,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "X86/common_x86_asm.h"
 #endif
 
-#define RADEON_DATE	"20010105"
+#define RADEON_DATE	"20010305"
 
 
 /* Return the width and height of the current color buffer.
@@ -100,6 +100,11 @@ static const GLubyte *radeonDDGetString( GLcontext *ctx, GLenum name )
 	 strncat( buffer, " x86", 4 );
       }
 #endif
+#ifdef USE_MMX_ASM
+      if ( cpu_has_mmx ) {
+	 strncat( buffer, "/MMX", 4 );
+      }
+#endif
 #ifdef USE_3DNOW_ASM
       if ( cpu_has_3dnow ) {
 	 strncat( buffer, "/3DNow!", 7 );
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c
index a02794e49..c2cd6a203 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c
@@ -321,6 +321,7 @@ void radeonFireBlitLocked( radeonContextPtr rmesa, drmBufPtr buffer,
 			   GLint offset, GLint pitch, GLint format,
 			   GLint x, GLint y, GLint width, GLint height )
 {
+#if 0
    GLint ret;
 
    ret = drmRadeonTextureBlit( rmesa->driFd, buffer->idx,
@@ -332,6 +333,7 @@ void radeonFireBlitLocked( radeonContextPtr rmesa, drmBufPtr buffer,
       fprintf( stderr, "drmRadeonTextureBlit: return = %d\n", ret );
       exit( 1 );
    }
+#endif
 }
 
 
@@ -616,12 +618,6 @@ static GLbitfield radeonDDClear( GLcontext *ctx, GLbitfield mask,
    }
 #endif
 
-#if 0
-   if ( rmesa->dirty & ~RADEON_UPLOAD_CLIPRECTS ) {
-      radeonEmitHwStateLocked( rmesa );
-   }
-#endif
-
    for ( i = 0 ; i < rmesa->numClipRects ; ) {
       GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, rmesa->numClipRects );
       XF86DRIClipRectPtr box = rmesa->pClipRects;
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_state.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_state.c
index d82013338..d80004dd6 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_state.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_state.c
@@ -511,7 +511,8 @@ static void radeonDDLightModelfv( GLcontext *ctx, GLenum pname,
 
       FLUSH_BATCH( rmesa );
 
-      if ( ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR ) {
+      if ( ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR &&
+           ctx->Light.Enabled && ctx->Texture.ReallyEnabled) {
 	 p |=  RADEON_SPECULAR_ENABLE;
       } else {
 	 p &= ~RADEON_SPECULAR_ENABLE;
@@ -805,6 +806,22 @@ static void radeonDDEnable( GLcontext *ctx, GLenum cap, GLboolean state )
       }
       break;
 
+   case GL_LIGHTING:
+      {
+         GLuint p = rmesa->setup.pp_cntl;
+         if ( ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR &&
+              ctx->Light.Enabled && ctx->Texture.ReallyEnabled) {
+            p |=  RADEON_SPECULAR_ENABLE;
+         } else {
+            p &= ~RADEON_SPECULAR_ENABLE;
+         }
+         if ( rmesa->setup.pp_cntl != p ) {
+            rmesa->setup.pp_cntl = p;
+            rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+         }
+         break;
+      }
+
    case GL_SCISSOR_TEST:
       FLUSH_BATCH( rmesa );
       rmesa->scissor = state;
@@ -919,29 +936,29 @@ void radeonEmitHwStateLocked( radeonContextPtr rmesa )
    if ( (rmesa->dirty & RADEON_UPLOAD_TEX0) && t0 ) {
       radeon_texture_regs_t *tex = &sarea->TexState[0];
 
-      tex->pp_txfilter = t0->setup.pp_txfilter | rmesa->lod_bias[0] << 8;
-      tex->pp_txformat = t0->setup.pp_txformat | RADEON_TXF_ST_ROUTE_STQ0;
-      tex->pp_txoffset = t0->setup.pp_txoffset;
+      tex->pp_txfilter = t0->pp_txfilter | rmesa->lod_bias[0] << 8;
+      tex->pp_txformat = t0->pp_txformat | RADEON_TXFORMAT_ST_ROUTE_STQ0;
+      tex->pp_txoffset = t0->pp_txoffset;
 
       tex->pp_txcblend = rmesa->color_combine[0];
       tex->pp_txablend = rmesa->alpha_combine[0];
       tex->pp_tfactor = rmesa->env_color[0];
 
-      tex->pp_border_color = t0->setup.pp_border_color;
+      tex->pp_border_color = t0->pp_border_color;
    }
 
    if ( (rmesa->dirty & RADEON_UPLOAD_TEX1) && t1 ) {
       radeon_texture_regs_t *tex = &sarea->TexState[1];
 
-      tex->pp_txfilter = t1->setup.pp_txfilter | rmesa->lod_bias[1] << 8;
-      tex->pp_txformat = t1->setup.pp_txformat | RADEON_TXF_ST_ROUTE_STQ1;
-      tex->pp_txoffset = t1->setup.pp_txoffset;
+      tex->pp_txfilter = t1->pp_txfilter | rmesa->lod_bias[1] << 8;
+      tex->pp_txformat = t1->pp_txformat | RADEON_TXFORMAT_ST_ROUTE_STQ1;
+      tex->pp_txoffset = t1->pp_txoffset;
 
       tex->pp_txcblend = rmesa->color_combine[1];
       tex->pp_txablend = rmesa->alpha_combine[1];
       tex->pp_tfactor = rmesa->env_color[1];
 
-      tex->pp_border_color = t1->setup.pp_border_color;
+      tex->pp_border_color = t1->pp_border_color;
    }
 
    if ( rmesa->dirty & RADEON_UPLOAD_TEX2 ) {
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c
index dd9c4557b..a77a7f154 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c
@@ -44,1963 +44,701 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "simple_list.h"
 #include "enums.h"
 #include "mem.h"
+#include "texutil.h"
 
 
 static void radeonSetTexWrap( radeonTexObjPtr t, GLenum swrap, GLenum twrap )
 {
-   t->setup.pp_txfilter &= ~(RADEON_CLAMP_S_MASK | RADEON_CLAMP_T_MASK);
+   t->pp_txfilter &= ~(RADEON_CLAMP_S_MASK | RADEON_CLAMP_T_MASK);
 
    switch ( swrap ) {
    case GL_REPEAT:
-      t->setup.pp_txfilter |= RADEON_CLAMP_S_WRAP;
+      t->pp_txfilter |= RADEON_CLAMP_S_WRAP;
       break;
    case GL_CLAMP:
-      t->setup.pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
+      t->pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
       break;
    case GL_CLAMP_TO_EDGE:
-      t->setup.pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
+      t->pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
       break;
    }
 
    switch ( twrap ) {
    case GL_REPEAT:
-      t->setup.pp_txfilter |= RADEON_CLAMP_T_WRAP;
+      t->pp_txfilter |= RADEON_CLAMP_T_WRAP;
       break;
    case GL_CLAMP:
-      t->setup.pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
+      t->pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
       break;
    case GL_CLAMP_TO_EDGE:
-      t->setup.pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
+      t->pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
       break;
    }
 }
 
 static void radeonSetTexFilter( radeonTexObjPtr t, GLenum minf, GLenum magf )
 {
-   t->setup.pp_txfilter &= ~(RADEON_MIN_FILTER_MASK | RADEON_MAG_FILTER_MASK);
+   t->pp_txfilter &= ~(RADEON_MIN_FILTER_MASK | RADEON_MAG_FILTER_MASK);
 
    switch ( minf ) {
    case GL_NEAREST:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_NEAREST;
+      t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST;
       break;
    case GL_LINEAR:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_LINEAR;
+      t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR;
       break;
    case GL_NEAREST_MIPMAP_NEAREST:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_NEAREST;
+      t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_NEAREST;
       break;
    case GL_LINEAR_MIPMAP_NEAREST:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_NEAREST;
+      t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_NEAREST;
       break;
    case GL_NEAREST_MIPMAP_LINEAR:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_LINEAR;
+      t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_LINEAR;
       break;
    case GL_LINEAR_MIPMAP_LINEAR:
-      t->setup.pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_LINEAR;
+      t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_LINEAR;
       break;
    }
 
    switch ( magf ) {
    case GL_NEAREST:
-      t->setup.pp_txfilter |= RADEON_MAG_FILTER_NEAREST;
+      t->pp_txfilter |= RADEON_MAG_FILTER_NEAREST;
       break;
    case GL_LINEAR:
-      t->setup.pp_txfilter |= RADEON_MAG_FILTER_LINEAR;
+      t->pp_txfilter |= RADEON_MAG_FILTER_LINEAR;
       break;
    }
 }
 
 static void radeonSetTexBorderColor( radeonTexObjPtr t, GLubyte c[4] )
 {
-   t->setup.pp_border_color = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
+   t->pp_border_color = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
 }
 
-
-/* Allocate and initialize hardware state associated with texture `t'.
- */
-static radeonTexObjPtr radeonCreateTexObj( radeonContextPtr rmesa,
-					   struct gl_texture_object *tObj )
+static radeonTexObjPtr radeonAllocTexObj( struct gl_texture_object *tObj )
 {
    radeonTexObjPtr t;
-   struct gl_texture_image *image;
-   GLint log2Width, log2Height, log2Size, log2MinSize;
-   GLint totalSize;
-   GLint texelsPerDword = 0, blitWidth = 0, blitPitch = 0;
-   GLint x, y, width, height;
-   GLint i;
-   GLuint txformat, txalpha;
-
-   image = tObj->Image[0];
-   if ( !image )
-      return NULL;
-
-   t = (radeonTexObjPtr) CALLOC( sizeof(*t) );
-   if ( !t )
-      return NULL;
-
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API )
-      fprintf( stderr, "%s( %p )\n", __FUNCTION__, tObj );
-
-   switch ( image->Format ) {
-   case GL_RGBA:
-      if ( image->IntFormat != GL_RGBA4 &&
-	   ( image->IntFormat == GL_RGBA8 ||
-	     rmesa->radeonScreen->cpp == 4 ) ) {
-	 t->texelBytes = 4;
-	 txformat = RADEON_TXF_32BPP_ARGB8888;
-      } else {
-	 t->texelBytes = 2;
-	 txformat = RADEON_TXF_16BPP_ARGB4444;
-      }
-      txalpha = RADEON_TXF_ALPHA_IN_MAP;
-      break;
-
-   case GL_RGB:
-      if ( image->IntFormat != GL_RGB5 &&
-	   ( image->IntFormat == GL_RGB8 ||
-	     rmesa->radeonScreen->cpp == 4 ) ) {
-	 t->texelBytes = 4;
-	 txformat = RADEON_TXF_32BPP_ARGB8888;
-      } else {
-	 t->texelBytes = 2;
-	 txformat = RADEON_TXF_16BPP_RGB565;
-      }
-      txalpha = 0;
-      break;
-
-   case GL_ALPHA:
-   case GL_LUMINANCE_ALPHA:
-      t->texelBytes = 2;
-      txformat = RADEON_TXF_16BPP_AI88;
-      txalpha = RADEON_TXF_ALPHA_IN_MAP;
-      break;
-
-   case GL_LUMINANCE:
-      t->texelBytes = 2;
-      txformat = RADEON_TXF_16BPP_AI88;
-      txalpha = 0;
-      break;
-
-   case GL_INTENSITY:
-      t->texelBytes = 1;
-      txformat = RADEON_TXF_8BPP_I;
-      txalpha = 0;
-      break;
-
-   case GL_COLOR_INDEX:
-   default:
-      fprintf( stderr, "%s: bad image->Format\n", __FUNCTION__ );
-      FREE( t );
-      return NULL;
-   }
-
-   /* Calculate dimensions in log domain.
-    */
-   for ( i = 1, log2Height = 0 ; i < image->Height ; i *= 2 ) {
-      log2Height++;
-   }
-   for ( i = 1, log2Width = 0 ;  i < image->Width ;  i *= 2 ) {
-      log2Width++;
-   }
-   if ( image->Width > image->Height ) {
-      log2Size = log2Width;
-   } else {
-      log2Size = log2Height;
-   }
-
-   t->dirty_images = 0;
-
-   /* The Radeon has a 64-byte minimum pitch for all blits.  We
-    * calculate the equivalent number of texels to simplify the
-    * calculation of the texture image area.
-    */
-   switch ( t->texelBytes ) {
-   case 1:
-      texelsPerDword = 4;
-      blitPitch = 64;
-      break;
-   case 2:
-      texelsPerDword = 2;
-      blitPitch = 32;
-      break;
-   case 4:
-      texelsPerDword = 1;
-      blitPitch = 16;
-      break;
-   }
-
-   /* Select the larger of the two widths for our global texture image
-    * coordinate space.  As the Radeon has very strict offset rules, we
-    * can't upload mipmaps directly and have to reference their location
-    * from the aligned start of the whole image.
-    */
-   blitWidth = MAX2( image->Width, blitPitch );
-
-   /* Calculate mipmap offsets and dimensions.
-    */
-   totalSize = 0;
-   x = 0;
-   y = 0;
-
-   for ( i = 0 ; i <= log2Size ; i++ ) {
-      GLuint size;
 
-      image = tObj->Image[i];
-      if ( !image )
-	 break;
-
-      width = image->Width;
-      height = image->Height;
-
-      /* Texture images have a minimum pitch of 32 bytes (half of the
-       * 64-byte minimum pitch for blits).  For images that have a
-       * width smaller than this, we must pad each texture image
-       * scanline out to this amount.
-       */
-      if ( width < blitPitch / 2 ) {
-	 width = blitPitch / 2;
-      }
-
-      size = width * height * t->texelBytes;
-      totalSize += size;
-
-      t->dirty_images |= (1 << i);
-
-      while ( width < blitWidth && height > 1 ) {
-	 width *= 2;
-	 height /= 2;
-      }
-
-      t->image[i].x = x;
-      t->image[i].y = y;
-
-      t->image[i].width  = width;
-      t->image[i].height = height;
-
-      t->image[i].dwords = size / sizeof(CARD32);
-
-      /* While blits must have a pitch of at least 64 bytes, mipmaps
-       * must be aligned on a 32-byte boundary (just like each texture
-       * image scanline).
-       */
-      if ( width >= blitWidth ) {
-	 y += height;
-      } else {
-	 x += width;
-	 if ( x >= blitWidth ) {
-	    x = 0;
-	    y++;
-	 }
-      }
+   t = CALLOC_STRUCT( radeon_tex_obj );
 
-      if ( 0 )
-	 fprintf( stderr, "level=%d p=%d   %dx%d -> %dx%d at (%d,%d)  %d dwords\n",
-		  i, blitWidth, image->Width, image->Height,
-		  t->image[i].width, t->image[i].height,
-		  t->image[i].x, t->image[i].y,
-		  t->image[i].dwords );
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p )\n", __FUNCTION__, t );
    }
 
-   log2MinSize = log2Size - i + 1;
-
-   /* Align the total size of texture memory block.
+   /* Initialize non-image-dependent parts of the state:
     */
-   totalSize = (totalSize + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
-
-   t->totalSize = totalSize;
-
-   t->bound = 0;
-   t->heap = 0;
    t->tObj = tObj;
+   t->dirty_images = ~0;
+   t->pp_txfilter = RADEON_BORDER_MODE_OGL;
+   t->pp_txformat = (RADEON_TXFORMAT_ENDIAN_NO_SWAP |
+		     RADEON_TXFORMAT_PERSPECTIVE_ENABLE);
 
-   t->memBlock = NULL;
-   t->bufAddr = 0;
-
-   /* Hardware state:
-    */
-   t->setup.pp_txfilter = ((log2Size << RADEON_MAX_MIP_LEVEL_SHIFT) |
-			   RADEON_BORDER_MODE_OGL);
-
-   t->setup.pp_txformat = (txformat | txalpha |
-			   (log2Width << RADEON_TXF_WIDTH_SHIFT) |
-			   (log2Height << RADEON_TXF_HEIGHT_SHIFT) |
-			   RADEON_TXF_ENDIAN_NO_SWAP |
-			   RADEON_TXF_PERSPECTIVE_ENABLE);
-
-   t->setup.pp_txoffset		= 0x00000000;
-   t->setup.pp_txcblend		= 0x00000000;
-   t->setup.pp_txablend		= 0x00000000;
-   t->setup.pp_tfactor		= 0x00000000;
-   t->setup.pp_border_color	= 0x00000000;
+   make_empty_list( t );
 
    radeonSetTexWrap( t, tObj->WrapS, tObj->WrapT );
    radeonSetTexFilter( t, tObj->MinFilter, tObj->MagFilter );
    radeonSetTexBorderColor( t, tObj->BorderColor );
 
-   tObj->DriverData = t;
-
-   make_empty_list( t );
-
    return t;
 }
 
-/* Destroy hardware state associated with texture `t'.
- */
-void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-#if ENABLE_PERF_BOXES
-   /* Bump the performace counter */
-   rmesa->c_textureSwaps++;
-#endif
-   if ( !t ) return;
-
-   if ( t->memBlock ) {
-      mmFreeMem( t->memBlock );
-      t->memBlock = NULL;
-   }
-
-   if ( t->tObj )
-      t->tObj->DriverData = NULL;
-   if ( t->bound )
-      rmesa->CurrentTexObj[t->bound-1] = NULL;
-
-   remove_from_list( t );
-   FREE( t );
-}
-
-/* Keep track of swapped out texture objects.
- */
-static void radeonSwapOutTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-#if ENABLE_PERF_BOXES
-   /* Bump the performace counter */
-   rmesa->c_textureSwaps++;
-#endif
-   if ( t->memBlock ) {
-      mmFreeMem( t->memBlock );
-      t->memBlock = NULL;
-   }
-
-   t->dirty_images = ~0;
-   move_to_tail( &rmesa->SwappedOut, t );
-}
-
-/* Print out debugging information about texture LRU.
- */
-void radeonPrintLocalLRU( radeonContextPtr rmesa, int heap )
-{
-   radeonTexObjPtr t;
-   int sz = 1 << (rmesa->radeonScreen->logTexGranularity[heap]);
-
-   fprintf( stderr, "\nLocal LRU, heap %d:\n", heap );
-
-   foreach ( t, &rmesa->TexObjList[heap] ) {
-      if (!t->tObj) {
-	 fprintf( stderr, "Placeholder %d at 0x%x sz 0x%x\n",
-		  t->memBlock->ofs / sz,
-		  t->memBlock->ofs,
-		  t->memBlock->size );
-      } else {
-	 fprintf( stderr, "Texture (bound %d) at 0x%x sz 0x%x\n",
-		  t->bound,
-		  t->memBlock->ofs,
-		  t->memBlock->size );
-      }
-   }
-
-   fprintf( stderr, "\n" );
-}
-
-void radeonPrintGlobalLRU( radeonContextPtr rmesa, int heap )
-{
-   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
-   int i, j;
-
-   fprintf( stderr, "\nGlobal LRU, heap %d list %p:\n", heap, list );
-
-   for ( i = 0, j = RADEON_NR_TEX_REGIONS ; i < RADEON_NR_TEX_REGIONS ; i++ ) {
-      fprintf( stderr, "list[%d] age %d next %d prev %d\n",
-	       j, list[j].age, list[j].next, list[j].prev );
-      j = list[j].next;
-      if ( j == RADEON_NR_TEX_REGIONS ) break;
-   }
-
-   if ( j != RADEON_NR_TEX_REGIONS ) {
-      fprintf( stderr, "Loop detected in global LRU\n" );
-      for ( i = 0 ; i < RADEON_NR_TEX_REGIONS ; i++ ) {
-	 fprintf( stderr, "list[%d] age %d next %d prev %d\n",
-		  i, list[i].age, list[i].next, list[i].prev );
-      }
-   }
-
-   fprintf( stderr, "\n" );
-}
-
-/* Reset the global texture LRU.
- */
-static void radeonResetGlobalLRU( radeonContextPtr rmesa, int heap )
-{
-   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
-   int sz = 1 << rmesa->radeonScreen->logTexGranularity[heap];
-   int i;
-
-   /*
-    * (Re)initialize the global circular LRU list.  The last element in
-    * the array (RADEON_NR_TEX_REGIONS) is the sentinal.  Keeping it at
-    * the end of the array allows it to be addressed rationally when
-    * looking up objects at a particular location in texture memory.
-    */
-   for ( i = 0 ; (i+1) * sz <= rmesa->radeonScreen->texSize[heap] ; i++ ) {
-      list[i].prev = i-1;
-      list[i].next = i+1;
-      list[i].age = 0;
-   }
-
-   i--;
-   list[0].prev = RADEON_NR_TEX_REGIONS;
-   list[i].prev = i-1;
-   list[i].next = RADEON_NR_TEX_REGIONS;
-   list[RADEON_NR_TEX_REGIONS].prev = i;
-   list[RADEON_NR_TEX_REGIONS].next = 0;
-   rmesa->sarea->texAge[heap] = 0;
-}
-
-/* Update the local and glock texture LRUs.
- */
-static void radeonUpdateTexLRU(radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-   int heap = t->heap;
-   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
-   int sz = rmesa->radeonScreen->logTexGranularity[heap];
-   int start = t->memBlock->ofs >> sz;
-   int end = (t->memBlock->ofs + t->memBlock->size-1) >> sz;
-   int i;
-
-   rmesa->lastTexAge[heap] = ++rmesa->sarea->texAge[heap];
-
-   if ( !t->memBlock ) {
-      fprintf( stderr, "no memblock\n\n" );
-      return;
-   }
-
-   /* Update our local LRU */
-   move_to_head( &rmesa->TexObjList[heap], t );
-
-   /* Update the global LRU */
-   for ( i = start ; i <= end ; i++ ) {
-      list[i].in_use = 1;
-      list[i].age = rmesa->lastTexAge[heap];
-
-      /* remove_from_list(i) */
-      list[(CARD32)list[i].next].prev = list[i].prev;
-      list[(CARD32)list[i].prev].next = list[i].next;
-
-      /* insert_at_head(list, i) */
-      list[i].prev = RADEON_NR_TEX_REGIONS;
-      list[i].next = list[RADEON_NR_TEX_REGIONS].next;
-      list[(CARD32)list[RADEON_NR_TEX_REGIONS].next].prev = i;
-      list[RADEON_NR_TEX_REGIONS].next = i;
-   }
-
-   if ( 0 ) {
-      radeonPrintGlobalLRU( rmesa, t->heap );
-      radeonPrintLocalLRU( rmesa, t->heap );
-   }
-}
-
-/* Update our notion of what textures have been changed since we last
- * held the lock.  This pertains to both our local textures and the
- * textures belonging to other clients.  Keep track of other client's
- * textures by pushing a placeholder texture onto the LRU list -- these
- * are denoted by (tObj == NULL).
- */
-static void radeonTexturesGone( radeonContextPtr rmesa, int heap,
-				int offset, int size, int in_use )
-{
-   radeonTexObjPtr t, tmp;
-
-   foreach_s ( t, tmp, &rmesa->TexObjList[heap] ) {
-      if ( t->memBlock->ofs >= offset + size ||
-	   t->memBlock->ofs + t->memBlock->size <= offset )
-	 continue;
-
-      /* It overlaps - kick it out.  Need to hold onto the currently
-       * bound objects, however.
-       */
-      if ( t->bound ) {
-	 radeonSwapOutTexObj( rmesa, t );
-      } else {
-	 radeonDestroyTexObj( rmesa, t );
-      }
-   }
-
-   if ( in_use ) {
-      t = (radeonTexObjPtr) CALLOC( sizeof(*t) );
-      if ( !t ) return;
 
-      t->memBlock = mmAllocMem( rmesa->texHeap[heap], size, 0, offset );
-      if ( !t->memBlock ) {
-	 fprintf( stderr, "Couldn't alloc placeholder sz %x ofs %x\n",
-		  (int)size, (int)offset );
-	 mmDumpMemInfo( rmesa->texHeap[heap] );
-	 return;
-      }
-      insert_at_head( &rmesa->TexObjList[heap], t );
-   }
-}
-
-/* Update our client's shared texture state.  If another client has
- * modified a region in which we have textures, then we need to figure
- * out which of our textures has been removed, and update our global
- * LRU.
- */
-void radeonAgeTextures( radeonContextPtr rmesa, int heap )
+static GLint radeonChooseTexFormat( radeonContextPtr rmesa,
+				    struct gl_texture_image *texImage,
+				    GLenum format, GLenum type )
 {
-   RADEONSAREAPrivPtr sarea = rmesa->sarea;
-
-   if ( sarea->texAge[heap] != rmesa->lastTexAge[heap] ) {
-      int sz = 1 << rmesa->radeonScreen->logTexGranularity[heap];
-      int nr = 0;
-      int idx;
-
-      for ( idx = sarea->texList[heap][RADEON_NR_TEX_REGIONS].prev ;
-	    idx != RADEON_NR_TEX_REGIONS && nr < RADEON_NR_TEX_REGIONS ;
-	    idx = sarea->texList[heap][idx].prev, nr++ )
-      {
-	 /* If switching texturing schemes, then the SAREA might not
-	  * have been properly cleared, so we need to reset the
-	  * global texture LRU.
-	  */
-	 if ( idx * sz > rmesa->radeonScreen->texSize[heap] ) {
-	    nr = RADEON_NR_TEX_REGIONS;
-	    break;
-	 }
-
-	 if ( sarea->texList[heap][idx].age > rmesa->lastTexAge[heap] ) {
-	    radeonTexturesGone( rmesa, heap, idx * sz, sz,
-				sarea->texList[heap][idx].in_use );
-	 }
-      }
-
-      if ( nr == RADEON_NR_TEX_REGIONS ) {
-	 radeonTexturesGone( rmesa, heap, 0,
-			     rmesa->radeonScreen->texSize[heap], 0 );
-	 radeonResetGlobalLRU( rmesa, heap );
-      }
-
-      rmesa->dirty |= (RADEON_UPLOAD_CONTEXT |
-		       RADEON_UPLOAD_TEX0IMAGES |
-		       RADEON_UPLOAD_TEX1IMAGES);
-      rmesa->lastTexAge[heap] = sarea->texAge[heap];
-   }
-}
+   const GLboolean do32bpt = ( rmesa->radeonScreen->cpp == 4 );
+   const struct gl_texture_format *texFormat;
+   GLint ret;
 
+   if ( 0 )
+      fprintf( stderr, "internal=%s format=%s type=%s\n",
+	       texImage->IntFormat == 3 ? "GL_RGB (3)" :
+	       texImage->IntFormat == 4 ? "GL_RGBA (4)" :
+	       gl_lookup_enum_by_nr( texImage->IntFormat ),
+	       gl_lookup_enum_by_nr( format ),
+	       gl_lookup_enum_by_nr( type ) );
 
-/* ================================================================
- * Texture image conversions
- */
-
-/* Convert a block of Mesa-formatted texture to an 8bpp hardware format.
- */
-static void radeonConvertTexture8bpp( CARD32 *dst,
-				      struct gl_texture_image *image,
-				      int x, int y, int width, int height,
-				      int pitch )
-{
-   CARD8 *src;
-   int i, j;
-
-   if ( width < 4 ) {
-      width = 4;
-   }
-
-#define ALIGN_DST							\
+#define SET_FORMAT( r, gl )						\
    do {									\
-      if ( width < 32 ) {						\
-	 dst += ((32 - width) / 4);					\
-      }									\
+      ret = (r);							\
+      texFormat = &(gl);						\
    } while (0)
 
-   switch ( image->Format ) {
-   case GL_INTENSITY:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x);
-	 for ( j = width >> 2 ; j ; j-- ) {
-	    *dst++ = src[0] | (src[1] << 8) | (src[2] << 16) | (src[3] << 24);
-	    src += 4;
-	 }
-	 ALIGN_DST;
-      }
-      break;
-
-   case GL_COLOR_INDEX:
-   default:
-      fprintf( stderr, "%s: unsupported format 0x%x\n",
-	       __FUNCTION__, image->Format );
-      break;
-   }
-}
-#undef ALIGN_DST
-
-/* Convert a block of Mesa-formatted texture to a 16bpp hardware format.
- */
-static void radeonConvertTexture16bpp( CARD32 *dst,
-				       struct gl_texture_image *image,
-				       int x, int y, int width, int height,
-				       int pitch )
-{
-   CARD8 *src;
-   int i, j;
-
-   if ( width < 2 ) {
-      width = 2;
-   }
-
-#define ALIGN_DST							\
+#define SET_FORMAT_32BPT( r32, gl32, r16, gl16 )			\
    do {									\
-      if ( width < 16 ) {						\
-	 dst += ((16 - width) / 2);					\
+      if ( do32bpt ) {							\
+	 ret = (r32);							\
+	 texFormat = &(gl32);						\
+      } else {								\
+	 ret = (r16);							\
+	 texFormat = &(gl16);						\
       }									\
    } while (0)
 
-   switch ( image->Format ) {
+   switch ( texImage->IntFormat ) {
+      /* GH: Bias towards GL_RGB, GL_RGBA texture formats.  This has
+       * got to be better than sticking them way down the end of this
+       * huge list.
+       */
    case GL_RGBA:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 4;
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR4444( src[0], src[1], src[2], src[3] )) |
-		      (RADEONPACKCOLOR4444( src[4], src[5], src[6], src[7] ) << 16));
-	    src += 8;
+   case 4:
+      if ( format == GL_BGRA ) {
+	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
+	    SET_FORMAT( RADEON_TXFORMAT_ARGB8888, _mesa_texformat_argb8888 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
+	    SET_FORMAT( RADEON_TXFORMAT_ARGB4444, _mesa_texformat_argb4444 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
+	    SET_FORMAT( RADEON_TXFORMAT_ARGB1555, _mesa_texformat_argb1555 );
+	    break;
 	 }
-	 ALIGN_DST;
       }
+      SET_FORMAT_32BPT( RADEON_TXFORMAT_RGBA8888, _mesa_texformat_rgba8888,
+			RADEON_TXFORMAT_ARGB4444, _mesa_texformat_argb4444 );
       break;
 
    case GL_RGB:
-   {
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 3;
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR565( src[0], src[1], src[2] )) |
-		      (RADEONPACKCOLOR565( src[3], src[4], src[5] ) << 16));
-	    src += 6;
-	 }
-	 ALIGN_DST;
+   case 3:
+      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
+	 SET_FORMAT( RADEON_TXFORMAT_RGB565, _mesa_texformat_rgb565 );
+	 break;
       }
-   }
+      SET_FORMAT_32BPT( RADEON_TXFORMAT_RGBA8888, _mesa_texformat_rgba8888,
+			RADEON_TXFORMAT_RGB565,   _mesa_texformat_rgb565 );
       break;
 
-   case GL_ALPHA:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x);
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR88( 0x00, src[0] )) |
-		      (RADEONPACKCOLOR88( 0x00, src[1] ) << 16));
-	    src += 2;
-	 }
-	 ALIGN_DST;
-      }
+      /* GH: Okay, keep checking as normal.  Still test for GL_RGB,
+       * GL_RGBA formats first.
+       */
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      SET_FORMAT_32BPT( RADEON_TXFORMAT_RGBA8888, _mesa_texformat_rgba8888,
+			RADEON_TXFORMAT_ARGB4444, _mesa_texformat_argb4444 );
       break;
 
-   case GL_LUMINANCE:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x);
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR88( src[0], 0xff )) |
-		      (RADEONPACKCOLOR88( src[1], 0xff ) << 16));
-	    src += 2;
-	 }
-	 ALIGN_DST;
-      }
+   case GL_RGBA4:
+   case GL_RGBA2:
+      SET_FORMAT( RADEON_TXFORMAT_ARGB4444, _mesa_texformat_argb4444 );
       break;
 
-   case GL_LUMINANCE_ALPHA:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 2;
-	 for ( j = width >> 1 ; j ; j-- ) {
-	    *dst++ = ((RADEONPACKCOLOR88( src[0], src[1] )) |
-		      (RADEONPACKCOLOR88( src[2], src[3] ) << 16));
-	    src += 4;
-	 }
-	 ALIGN_DST;
-      }
+   case GL_RGB5_A1:
+      SET_FORMAT( RADEON_TXFORMAT_ARGB1555, _mesa_texformat_argb1555 );
       break;
 
-   default:
-      fprintf( stderr, "%s: unsupported format 0x%x\n",
-	       __FUNCTION__, image->Format );
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      SET_FORMAT_32BPT( RADEON_TXFORMAT_RGBA8888, _mesa_texformat_rgba8888,
+			RADEON_TXFORMAT_RGB565,   _mesa_texformat_rgb565 );
       break;
-   }
-}
-#undef ALIGN_DST
 
-/* Convert a block of Mesa-formatted texture to a 32bpp hardware format.
- */
-static void radeonConvertTexture32bpp( CARD32 *dst,
-				       struct gl_texture_image *image,
-				       int x, int y, int width, int height,
-				       int pitch )
-{
-   CARD8 *src;
-   int i, j;
+   case GL_RGB5:
+   case GL_RGB4:
+   case GL_R3_G3_B2:
+      SET_FORMAT( RADEON_TXFORMAT_RGB565, _mesa_texformat_rgb565 );
+      break;
 
-#define ALIGN_DST							\
-   do {									\
-      if ( width < 8 ) {						\
-	 dst += (8 - width);						\
-      }									\
-   } while (0)
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+      SET_FORMAT( RADEON_TXFORMAT_AI88, _mesa_texformat_al88 );
+      break;
 
-   switch ( image->Format ) {
-   case GL_RGBA:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 4;
-	 for ( j = width ; j ; j-- ) {
-	    *dst++ = RADEONPACKCOLOR8888( src[0], src[1], src[2], src[3] );
-	    src += 4;
-	 }
-	 ALIGN_DST;
-      }
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+      SET_FORMAT( RADEON_TXFORMAT_AI88, _mesa_texformat_al88 );
       break;
 
-   case GL_RGB:
-      for ( i = 0 ; i < height ; i++ ) {
-	 src = (CARD8 *)image->Data + ((y + i) * pitch + x) * 3;
-	 for ( j = width ; j ; j-- ) {
-	    *dst++ = RADEONPACKCOLOR8888( src[0], src[1], src[2], 0xff );
-	    src += 3;
-	 }
-	 ALIGN_DST;
-      }
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+      SET_FORMAT( RADEON_TXFORMAT_AI88, _mesa_texformat_al88 );
       break;
 
-   default:
-      fprintf( stderr, "%s: unsupported format 0x%x\n",
-	       __FUNCTION__, image->Format );
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+      SET_FORMAT( RADEON_TXFORMAT_I8, _mesa_texformat_i8 );
       break;
+
+   case GL_COLOR_INDEX:
+   case GL_COLOR_INDEX1_EXT:
+   case GL_COLOR_INDEX2_EXT:
+   case GL_COLOR_INDEX4_EXT:
+   case GL_COLOR_INDEX8_EXT:
+   case GL_COLOR_INDEX12_EXT:
+   case GL_COLOR_INDEX16_EXT:
+   default:
+      fprintf( stderr, "bad texture format in radeonChooseTexFormat() %d",
+	       texImage->IntFormat );
+      return -1;
    }
+
+   texImage->TexFormat = texFormat;
+
+   return ret;
 }
 
-/* Upload the texture image associated with texture `t' at level `level'
- * at the address relative to `start'.
+
+/* ================================================================
+ * Texture image callbacks
  */
-static void radeonUploadSubImage( radeonContextPtr rmesa,
-				  radeonTexObjPtr t, GLint level,
-				  GLint x, GLint y, GLint width, GLint height )
-{
-   struct gl_texture_image *image;
-   GLint texelsPerDword = 0;
-   GLint imageX, imageY, imageWidth, imageHeight;
-   GLint blitX, blitY, blitWidth, blitHeight;
-   GLint imageRows, blitRows;
-   GLint remaining ;
-   GLint format, dwords;
-   CARD32 pitch, offset;
-   drmBufPtr buffer;
-   CARD32 *dst;
-
-   /* Ensure we have a valid texture to upload */
-   if ( ( level < 0 ) || ( level >= RADEON_MAX_TEXTURE_LEVELS ) )
-      return;
 
-   image = t->tObj->Image[level];
-   if ( !image )
-      return;
+static GLboolean
+radeonDDTexImage1D( GLcontext *ctx, GLenum target, GLint level,
+		    GLenum format, GLenum type, const GLvoid *pixels,
+		    const struct gl_pixelstore_attrib *packing,
+		    struct gl_texture_object *texObj,
+		    struct gl_texture_image *texImage,
+		    GLboolean *retainInternalCopy )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
+   GLuint unit = ctx->Texture.CurrentUnit;
+   GLuint texSize;
+   GLint txformat;
+   GLubyte *data;
 
-   switch ( t->texelBytes ) {
-   case 1:
-      texelsPerDword = 4;
-      break;
-   case 2:
-      texelsPerDword = 2;
-      break;
-   case 4:
-      texelsPerDword = 1;
-      break;
-   }
+   if ( target != GL_TEXTURE_1D )
+      return GL_FALSE;
 
-   format = t->setup.pp_txformat & RADEON_TXF_FORMAT_MASK;
-
-   imageX = 0;
-   imageY = 0;
-   imageWidth = image->Width;
-   imageHeight = image->Height;
-
-   blitX = t->image[level].x;
-   blitY = t->image[level].y;
-   blitWidth = t->image[level].width;
-   blitHeight = t->image[level].height;
-
-   dwords = t->image[level].dwords;
-   offset = t->bufAddr;
-   pitch = (t->image[0].width * t->texelBytes) / 64;
-
-#if ENABLE_PERF_BOXES
-   /* Bump the performace counter */
-   rmesa->c_textureBytes += (dwords << 2);
-#endif
-
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
-      fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
-	       imageWidth, imageHeight, imageX, imageY );
-      fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
-	       blitWidth, blitHeight, blitX, blitY );
-      fprintf( stderr, "       blit ofs: 0x%07x pitch: 0x%x dwords: %d "
-	       "level: %d format: %x\n",
-	       (GLuint)offset, (GLuint)pitch, dwords, level, format );
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p, %d )\n", __FUNCTION__, texObj, level );
    }
 
-   /* Subdivide the texture if required */
-   if ( dwords <= RADEON_BUFFER_MAX_DWORDS / 2 ) {
-      imageRows = imageHeight;
-      blitRows = blitHeight;
+   if ( t ) {
+      if ( t->bound ) FLUSH_BATCH( rmesa );
+      radeonSwapOutTexObj( rmesa, t );
    } else {
-      imageRows = (RADEON_BUFFER_MAX_DWORDS * texelsPerDword)/(2 * imageWidth);
-      blitRows = (RADEON_BUFFER_MAX_DWORDS * texelsPerDword) / (2 * blitWidth);
+      t = radeonAllocTexObj( texObj );
+      texObj->DriverData = t;
    }
 
-   for ( remaining = imageHeight ;
-	 remaining > 0 ;
-	 remaining -= imageRows, imageY += imageRows, blitY += blitRows )
-   {
-      if ( remaining >= imageRows ) {
-	 imageHeight = imageRows;
-	 blitHeight = blitRows;
-      } else {
-	 imageHeight = remaining;
-	 blitHeight = blitRows;
-      }
-      dwords = blitWidth * blitHeight / texelsPerDword;
+   txformat = radeonChooseTexFormat( rmesa, texImage, format, type );
+   if ( txformat < 0 )
+      return GL_FALSE;
 
-      if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
-	 fprintf( stderr, "       blitting: %d,%d at %d,%d\n",
-		  imageWidth, imageHeight, imageX, imageY );
-	 fprintf( stderr, "                 %d,%d at %d,%d - %d dwords\n",
-		  blitWidth, blitHeight, blitX, blitY, dwords );
-      }
+   texFormat = texImage->TexFormat;
+   texSize = texImage->Width * texFormat->TexelBytes;
 
-      /* Grab the indirect buffer for the texture blit */
-      buffer = radeonGetBufferLocked( rmesa );
+   /* We really shouldn't have to keep the texture image, it should be
+    * hung from the main texImage structure.
+    */
+   if ( t->image[level].data ) {
+      FREE( t->image[level].data );
+      t->image[level].data = NULL;
+   }
 
-      dst = (CARD32 *)((char *)buffer->address + RADEON_HOSTDATA_BLIT_OFFSET);
+   data = (GLubyte *) MALLOC( texSize );
+   if ( !data )
+      return GL_FALSE;
 
-      /* Actually do the texture conversion */
-      switch ( t->texelBytes ) {
-      case 1:
-	 radeonConvertTexture8bpp( dst, image,
-				   imageX, imageY, imageWidth, imageHeight,
-				   imageWidth );
-	 break;
-      case 2:
-	 radeonConvertTexture16bpp( dst, image,
-				    imageX, imageY, imageWidth, imageHeight,
-				    imageWidth );
-	 break;
-      case 4:
-	 radeonConvertTexture32bpp( dst, image,
-				    imageX, imageY, imageWidth, imageHeight,
-				    imageWidth );
-	 break;
-      }
-
-      radeonFireBlitLocked( rmesa, buffer,
-			    offset, pitch, format,
-			    blitX, blitY, blitWidth, blitHeight );
+   if ( !_mesa_convert_texsubimage1d( texFormat->IntFormat,
+				      0, texImage->Width,
+				      format, type, packing,
+				      pixels, data ) ) {
+      /*fprintf( stderr, "   *** convert failed!\n" );*/
+      FREE( data );
+      return GL_FALSE;
    }
 
-   rmesa->new_state |= RADEON_NEW_CONTEXT;
-   rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_MASKS;
-}
+   t->image[level].data = data;
+   t->dirty_images |= (1 << level);
 
-/* Upload the texture images associated with texture `t'.  This might
- * require removing our own and/or other client's texture objects to
- * make room for these images.
- */
-/* NOTE: This function is only called while holding the hardware lock */
-int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t )
-{
-   int i;
-   int heap;
+   /* Format-specific hardware state:
+    */
+   t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
+		       RADEON_TXFORMAT_ALPHA_IN_MAP);
+   t->pp_txformat |= txformat;
 
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p, %p )\n",
-	       __FUNCTION__, rmesa->glCtx, t );
+   if ( txformat == RADEON_TXFORMAT_RGBA8888 ||
+	txformat == RADEON_TXFORMAT_ARGB4444 ||
+	txformat == RADEON_TXFORMAT_ARGB1555 ||
+	txformat == RADEON_TXFORMAT_AI88 ) {
+      t->pp_txformat |= RADEON_TXFORMAT_ALPHA_IN_MAP;
    }
 
-   if ( !t )
-      return 0;
-
-   /* Choose the heap appropriately */
-   heap = t->heap = RADEON_CARD_HEAP;
-#if 0
-   if ( !rmesa->radeonScreen->IsPCI &&
-	t->totalSize > rmesa->radeonScreen->texSize[heap] ) {
-      heap = t->heap = RADEON_AGP_HEAP;
+   /* If we don't have a texture bound, bind this one.
+    */
+   if ( !rmesa->CurrentTexObj[unit] ) {
+      rmesa->CurrentTexObj[unit] = t;
+      t->bound |= unit + 1;
    }
-#endif
-
-   /* Do we need to eject LRU texture objects? */
-   if ( !t->memBlock ) {
-      /* Allocate a memory block on a 4k boundary (1<<12 == 4096) */
-      t->memBlock = mmAllocMem( rmesa->texHeap[heap],
-				t->totalSize, 12, 0 );
-
-#if 0
-      /* Try AGP before kicking anything out of local mem */
-      if ( !t->memBlock && heap == RADEON_CARD_HEAP ) {
-	 t->memBlock = mmAllocMem( rmesa->texHeap[RADEON_AGP_HEAP],
-				   t->totalSize, 12, 0 );
-
-	 if ( t->memBlock )
-	    heap = t->heap = RADEON_AGP_HEAP;
-      }
-#endif
-
-      /* Kick out textures until the requested texture fits */
-      while ( !t->memBlock ) {
-	 if ( rmesa->TexObjList[heap].prev->bound ) {
-	    fprintf( stderr,
-		     "radeonUploadTexImages: ran into bound texture\n" );
-	    return -1;
-	 }
-	 if ( rmesa->TexObjList[heap].prev ==
-	      &rmesa->TexObjList[heap] ) {
-	    if ( rmesa->radeonScreen->IsPCI ) {
-	       fprintf( stderr, "radeonUploadTexImages: upload texture "
-			"failure on local texture heaps, sz=%d\n",
-			t->totalSize );
-	       return -1;
-#if 0
-	    } else if ( heap == RADEON_CARD_HEAP ) {
-	       heap = t->heap = RADEON_AGP_HEAP;
-	       continue;
-#endif
-	    } else {
-	       fprintf( stderr, "radeonUploadTexImages: upload texture "
-			"failure on both local and AGP texture heaps, "
-			"sz=%d\n",
-			t->totalSize );
-	       return -1;
-	    }
-	 }
-
-	 radeonDestroyTexObj( rmesa, rmesa->TexObjList[heap].prev );
 
-	 t->memBlock = mmAllocMem( rmesa->texHeap[heap],
-				   t->totalSize, 12, 0 );
-      }
-
-      /* Set the base offset of the texture image */
-      t->bufAddr = rmesa->radeonScreen->texOffset[heap] + t->memBlock->ofs;
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-      t->setup.pp_txoffset = t->bufAddr;
-#if 0
-      /* Fix AGP texture offsets */
-      if ( heap == RADEON_AGP_HEAP ) {
-	  t->setup.pp_tx_offset += RADEON_AGP_TEX_OFFSET +
-	      rmesa->radeonScreen->agpTexOffset;
-      }
-#endif
+   *retainInternalCopy = GL_FALSE;
+   return GL_TRUE;
+}
 
-      /* Force loading the new state into the hardware */
-      switch ( t->bound ) {
-      case 1:
-	 rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_TEX0;
-	 break;
+static GLboolean
+radeonDDTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+		    GLenum format, GLenum type, const GLvoid *pixels,
+		    const struct gl_pixelstore_attrib *packing,
+		    struct gl_texture_object *texObj,
+		    struct gl_texture_image *texImage,
+		    GLboolean *retainInternalCopy )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
+   GLuint unit = ctx->Texture.CurrentUnit;
+   GLuint texSize;
+   GLint txformat;
+   GLubyte *data;
 
-      case 2:
-	 rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_TEX1;
-	 break;
+   if ( target != GL_TEXTURE_2D )
+      return GL_FALSE;
 
-      default:
-	 return -1;
-      }
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p, %d )\n", __FUNCTION__, texObj, level );
    }
 
-   /* Let the world know we've used this memory recently */
-   radeonUpdateTexLRU( rmesa, t );
+   if ( t ) {
+      if ( t->bound ) FLUSH_BATCH( rmesa );
+      radeonSwapOutTexObj( rmesa, t );
+   } else {
+      t = radeonAllocTexObj( texObj );
+      texObj->DriverData = t;
+   }
 
-   /* Upload any images that are new */
-   if ( t->dirty_images ) {
-      int num_levels = ((t->setup.pp_txfilter & RADEON_MAX_MIP_LEVEL_MASK) >>
-			RADEON_MAX_MIP_LEVEL_SHIFT);
+   txformat = radeonChooseTexFormat( rmesa, texImage, format, type );
+   if ( txformat < 0 )
+      return GL_FALSE;
 
-      for ( i = 0 ; i <= num_levels ; i++ ) {
-	 if ( t->dirty_images & (1 << i) ) {
-	    radeonUploadSubImage( rmesa, t, i, 0, 0,
-				  t->image[i].width, t->image[i].height );
-	 }
-      }
+   texFormat = texImage->TexFormat;
+   texSize = texImage->Width * texImage->Height * texFormat->TexelBytes;
 
-      rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+   /* We really shouldn't have to keep the texture image, it should be
+    * hung from the main texImage structure.
+    */
+   if ( t->image[level].data ) {
+      FREE( t->image[level].data );
+      t->image[level].data = NULL;
    }
 
-   t->dirty_images = 0;
-   return 0;
-}
-
+   data = (GLubyte *) MALLOC( texSize );
+   if ( !data )
+      return GL_FALSE;
 
-/* ================================================================
- * Texture combine functions
- */
+   if ( !_mesa_convert_texsubimage2d( texFormat->IntFormat,
+				      0, 0, texImage->Width, texImage->Height,
+				      texImage->Width, format, type, packing,
+				      pixels, data ) ) {
+      if ( 0 )
+	 fprintf( stderr, "   *** convert failed!  %s/%s-> %s\n",
+		  gl_lookup_enum_by_nr( format ),
+		  gl_lookup_enum_by_nr( type ),
+		  gl_lookup_enum_by_nr( texImage->IntFormat ) );
+      FREE( data );
+      return GL_FALSE;
+   }
 
-#define RADEON_DISABLE		0
-#define RADEON_REPLACE		1
-#define RADEON_MODULATE		2
-#define RADEON_DECAL		3
-#define RADEON_BLEND		4
-#define RADEON_ADD		5
-#define RADEON_MAX_COMBFUNC	6
+   t->image[level].data = data;
+   t->dirty_images |= (1 << level);
 
-static GLuint radeon_color_combine[][RADEON_MAX_COMBFUNC] =
-{
-   /* Unit 0:
-    */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_CURRENT_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00802800
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T0_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800142
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T0_COLOR |
-       RADEON_COLOR_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x008c2d42
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T0_COLOR |
-       RADEON_COLOR_ARG_C_T0_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x008c2902
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
-       RADEON_COLOR_ARG_C_T0_COLOR |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00812802
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T0_COLOR |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-   },
-
-   /* Unit 1:
-    */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_CURRENT_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00803000
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T1_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800182
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T1_COLOR |
-       RADEON_COLOR_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x008c3582
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T1_COLOR |
-       RADEON_COLOR_ARG_C_T1_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x008c3102
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
-       RADEON_COLOR_ARG_C_T1_COLOR |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00813002
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T1_COLOR |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-   },
-
-   /* Unit 2:
+   /* Format-specific hardware state:
     */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_CURRENT_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00803800
-       */
-      (RADEON_COLOR_ARG_A_ZERO |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T2_COLOR |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x008001c2
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T2_COLOR |
-       RADEON_COLOR_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x008c3dc2
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_T2_COLOR |
-       RADEON_COLOR_ARG_C_T2_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x008c3902
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
-       RADEON_COLOR_ARG_C_T2_COLOR |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00813802
-       */
-      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
-       RADEON_COLOR_ARG_B_ZERO |
-       RADEON_COLOR_ARG_C_T2_COLOR |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
+   t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
+		       RADEON_TXFORMAT_ALPHA_IN_MAP);
+   t->pp_txformat |= txformat;
+
+   if ( txformat == RADEON_TXFORMAT_RGBA8888 ||
+	txformat == RADEON_TXFORMAT_ARGB4444 ||
+	txformat == RADEON_TXFORMAT_ARGB1555 ||
+	txformat == RADEON_TXFORMAT_AI88 ) {
+      t->pp_txformat |= RADEON_TXFORMAT_ALPHA_IN_MAP;
    }
-};
 
-static GLuint radeon_alpha_combine[][RADEON_MAX_COMBFUNC] =
-{
-   /* Unit 0:
-    */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00800500
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T0_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800051
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_T0_ALPHA |
-       RADEON_ALPHA_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x00800100
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x00800051
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
-       RADEON_ALPHA_ARG_C_T0_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00800051
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T0_ALPHA |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-   },
-
-   /* Unit 1:
+   /* If we don't have a texture bound, bind this one.
     */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00800600
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T1_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800061
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_T1_ALPHA |
-       RADEON_ALPHA_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x00800100
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x00800061
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
-       RADEON_ALPHA_ARG_C_T1_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00800061
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T1_ALPHA |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-   },
-
-   /* Unit 2:
-    */
-   {
-      /* Disable combiner stage
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_REPLACE = 0x00800700
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T2_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_MODULATE = 0x00800071
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_T2_ALPHA |
-       RADEON_ALPHA_ARG_C_ZERO |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_DECAL = 0x00800100
-       */
-      (RADEON_ALPHA_ARG_A_ZERO |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_BLEND = 0x00800071
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
-       RADEON_ALPHA_ARG_C_T2_ALPHA |
-       RADEON_BLEND_CTL_BLEND |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
-
-      /* GL_ADD = 0x00800021
-       */
-      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
-       RADEON_ALPHA_ARG_B_ZERO |
-       RADEON_ALPHA_ARG_C_T2_ALPHA |
-       RADEON_COMP_ARG_B |
-       RADEON_BLEND_CTL_ADD |
-       RADEON_SCALE_1X |
-       RADEON_CLAMP_TX),
+   if ( !rmesa->CurrentTexObj[unit] ) {
+      rmesa->CurrentTexObj[unit] = t;
+      t->bound |= unit + 1;
    }
-};
 
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-/* GL_EXT_texture_env_combine support
- */
+   *retainInternalCopy = GL_FALSE;
+   return GL_TRUE;
+}
 
-/* The color tables have combine functions for GL_SRC_COLOR,
- * GL_ONE_MINUS_SRC_COLOR, GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
+/* GH: This is undoubtedly broken...
  */
-static GLuint radeon_texture_color[][RADEON_MAX_TEXTURE_UNITS] =
-{
-   {
-      RADEON_COLOR_ARG_A_T0_COLOR,
-      RADEON_COLOR_ARG_A_T1_COLOR,
-      RADEON_COLOR_ARG_A_T2_COLOR
-   },
-   {
-      RADEON_COLOR_ARG_A_T0_COLOR | RADEON_COMP_ARG_A,
-      RADEON_COLOR_ARG_A_T1_COLOR | RADEON_COMP_ARG_A,
-      RADEON_COLOR_ARG_A_T2_COLOR | RADEON_COMP_ARG_A
-   },
-   {
-      RADEON_COLOR_ARG_A_T0_ALPHA,
-      RADEON_COLOR_ARG_A_T1_ALPHA,
-      RADEON_COLOR_ARG_A_T2_ALPHA
-   },
-   {
-      RADEON_COLOR_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
-      RADEON_COLOR_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
-      RADEON_COLOR_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
-   },
-};
-
-static GLuint radeon_tfactor_color[] =
-{
-   RADEON_COLOR_ARG_A_TFACTOR_COLOR,
-   RADEON_COLOR_ARG_A_TFACTOR_COLOR | RADEON_COMP_ARG_A,
-   RADEON_COLOR_ARG_A_TFACTOR_ALPHA,
-   RADEON_COLOR_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
-};
-
-static GLuint radeon_primary_color[] =
-{
-   RADEON_COLOR_ARG_A_DIFFUSE_COLOR,
-   RADEON_COLOR_ARG_A_DIFFUSE_COLOR | RADEON_COMP_ARG_A,
-   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA,
-   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
-};
-
-static GLuint radeon_previous_color[] =
+static GLboolean
+radeonDDTexImage3D( GLcontext *ctx, GLenum target, GLint level,
+		    GLenum format, GLenum type, const GLvoid *pixels,
+		    const struct gl_pixelstore_attrib *packing,
+		    struct gl_texture_object *texObj,
+		    struct gl_texture_image *texImage,
+		    GLboolean *retainInternalCopy )
 {
-   RADEON_COLOR_ARG_A_CURRENT_COLOR,
-   RADEON_COLOR_ARG_A_CURRENT_COLOR | RADEON_COMP_ARG_A,
-   RADEON_COLOR_ARG_A_CURRENT_ALPHA,
-   RADEON_COLOR_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
-};
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
+   GLuint unit = ctx->Texture.CurrentUnit;
+   GLuint texSize;
+   GLint txformat;
+   GLubyte *data;
 
-/* The alpha tables only have GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
- */
-static GLuint radeon_texture_alpha[][RADEON_MAX_TEXTURE_UNITS] =
-{
-   {
-      RADEON_ALPHA_ARG_A_T0_ALPHA,
-      RADEON_ALPHA_ARG_A_T1_ALPHA,
-      RADEON_ALPHA_ARG_A_T2_ALPHA
-   },
-   {
-      RADEON_ALPHA_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
-      RADEON_ALPHA_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
-      RADEON_ALPHA_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
-   },
-};
-
-static GLuint radeon_tfactor_alpha[] =
-{
-   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA,
-   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
-};
+   if ( target != GL_TEXTURE_3D )
+      return GL_FALSE;
 
-static GLuint radeon_primary_alpha[] =
-{
-   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA,
-   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
-};
+   if ( t ) {
+      if ( t->bound ) FLUSH_BATCH( rmesa );
+      radeonSwapOutTexObj( rmesa, t );
+   } else {
+      t = radeonAllocTexObj( texObj );
+      texObj->DriverData = t;
+   }
 
-static GLuint radeon_previous_alpha[] =
-{
-   RADEON_ALPHA_ARG_A_CURRENT_ALPHA,
-   RADEON_ALPHA_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
-};
+   txformat = radeonChooseTexFormat( rmesa, texImage, format, type );
+   if ( txformat < 0 )
+      return GL_FALSE;
 
+   texFormat = texImage->TexFormat;
+   texSize = (texImage->Width * texImage->Height *
+	      texImage->Depth * texFormat->TexelBytes);
 
-/* Extract the arg from slot A, shift it into the correct argument slot
- * and set the corresponding complement bit.
- */
-#define RADEON_COLOR_ARG( n, arg )					\
-do {									\
-   color_combine |=							\
-      ((color_arg[n] & RADEON_COLOR_ARG_MASK)				\
-       << RADEON_COLOR_ARG_##arg##_SHIFT);				\
-   color_combine |=							\
-      ((color_arg[n] >> RADEON_COMP_ARG_SHIFT)				\
-       << RADEON_COMP_ARG_##arg##_SHIFT);				\
-} while (0)
-
-#define RADEON_ALPHA_ARG( n, arg )					\
-do {									\
-   alpha_combine |=							\
-      ((alpha_arg[n] & RADEON_ALPHA_ARG_MASK)				\
-       << RADEON_ALPHA_ARG_##arg##_SHIFT);				\
-   alpha_combine |=							\
-      ((alpha_arg[n] >> RADEON_COMP_ARG_SHIFT)				\
-       << RADEON_COMP_ARG_##arg##_SHIFT);				\
-} while (0)
+   /* We really shouldn't have to keep the texture image, it should be
+    * hung from the main texImage structure.
+    */
+   if ( t->image[level].data ) {
+      FREE( t->image[level].data );
+      t->image[level].data = NULL;
+   }
 
+   data = (GLubyte *) MALLOC( texSize );
+   if ( !data )
+      return GL_FALSE;
 
-/* ================================================================
- * Texture unit state management
- */
+   if ( !_mesa_convert_texsubimage3d( texFormat->IntFormat,
+				      0, 0, 0, texImage->Width,
+				      texImage->Height, texImage->Depth,
+				      texImage->Width, texImage->Height,
+				      format, type, packing,
+				      pixels, data ) ) {
+      FREE( data );
+      return GL_FALSE;
+   }
 
-static void radeonUpdateTextureEnv( GLcontext *ctx, int unit )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   int source = rmesa->tmu_source[unit];
-   struct gl_texture_object *tObj;
-   struct gl_texture_unit *texUnit;
-   GLuint enabled;
-   GLuint color_combine, alpha_combine;
-   GLuint color_arg[3], alpha_arg[3];
-   GLuint i, numColorArgs = 0, numAlphaArgs = 0;
-   GLuint op;
+   t->image[level].data = data;
+   t->dirty_images |= (1 << level);
 
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p, %d )\n",
-	       __FUNCTION__, ctx, unit );
+   /* If we don't have a texture bound, bind this one.
+    */
+   if ( !rmesa->CurrentTexObj[unit] ) {
+      rmesa->CurrentTexObj[unit] = t;
+      t->bound |= unit + 1;
    }
 
-   enabled = (ctx->Texture.ReallyEnabled >> (source * 4)) & TEXTURE0_ANY;
-   if ( enabled != TEXTURE0_2D && enabled != TEXTURE0_1D )
-      return;
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-   /* Only update the hardware texture state if the texture is current,
-    * complete and enabled.
-    */
-   texUnit = &ctx->Texture.Unit[source];
-   tObj = texUnit->Current;
-   if ( !tObj || !tObj->Complete )
-      return;
+   *retainInternalCopy = GL_FALSE;
+   return GL_TRUE;
+}
 
-   if ( ( tObj != texUnit->CurrentD[2] ) &&
-	( tObj != texUnit->CurrentD[1] ) )
-      return;
 
-   /* Set the texture environment state.  Isn't this nice and clean?
-    * The Radeon will automagically set the texture alpha to 0xff when
-    * the texture format does not include an alpha component.  This
-    * reduces the amount of special-casing we have to do, alpha-only
-    * textures being a notable exception.
-    */
-   switch ( texUnit->EnvMode ) {
-   case GL_REPLACE:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_REPLACE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
-	 break;
-      case GL_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+/* ================================================================
+ * Texture subimage callbacks
+ */
 
-   case GL_MODULATE:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_MODULATE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+static GLboolean
+radeonDDTexSubImage1D( GLcontext *ctx, GLenum target, GLint level,
+		       GLint xoffset, GLsizei width,
+		       GLenum format, GLenum type,
+		       const GLvoid *pixels,
+		       const struct gl_pixelstore_attrib *packing,
+		       struct gl_texture_object *texObj,
+		       struct gl_texture_image *texImage )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
 
-   case GL_DECAL:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-	 color_combine = radeon_color_combine[unit][RADEON_DECAL];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
-	 break;
-      case GL_ALPHA:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+   if ( target != GL_TEXTURE_1D )
+      return GL_FALSE;
 
-   case GL_BLEND:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_BLEND];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+   /* FIXME: Can this ever be NULL???
+    */
+   ASSERT( t );
 
-   case GL_ADD:
-      switch ( tObj->Image[0]->Format ) {
-      case GL_RGBA:
-      case GL_RGB:
-      case GL_LUMINANCE:
-      case GL_LUMINANCE_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_ADD];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_ALPHA:
-	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
-	 break;
-      case GL_INTENSITY:
-	 color_combine = radeon_color_combine[unit][RADEON_ADD];
-	 alpha_combine = radeon_alpha_combine[unit][RADEON_ADD];
-	 break;
-      case GL_COLOR_INDEX:
-      default:
-	 return;
-      }
-      break;
+   if ( t->bound )
+      FLUSH_BATCH( rmesa );
 
-   case GL_COMBINE_EXT:
-      /* Step 0:
-       * Calculate how many arguments we need to process.
-       */
-      switch ( texUnit->CombineModeRGB ) {
-      case GL_REPLACE:
-	 numColorArgs = 1;
-	 break;
-      case GL_MODULATE:
-      case GL_ADD:
-      case GL_ADD_SIGNED_EXT:
-      case GL_DOT3_RGB_EXT:
-      case GL_DOT3_RGBA_EXT:
-	 numColorArgs = 2;
-	 break;
-      case GL_INTERPOLATE_EXT:
-	 numColorArgs = 3;
-	 break;
-      default:
-	 return;
-      }
+   texFormat = texImage->TexFormat;
 
-      switch ( texUnit->CombineModeA ) {
-      case GL_REPLACE:
-	 numAlphaArgs = 1;
-	 break;
-      case GL_MODULATE:
-      case GL_ADD:
-      case GL_ADD_SIGNED_EXT:
-	 numAlphaArgs = 2;
-	 break;
-      case GL_INTERPOLATE_EXT:
-	 numAlphaArgs = 3;
-	 break;
-      default:
-	 return;
-      }
+   if ( !_mesa_convert_texsubimage1d( texFormat->IntFormat,
+				      xoffset, width, format, type, packing,
+				      pixels, t->image[level].data ) ) {
+      /*fprintf( stderr, "   *** convert failed!\n" );*/
+      return GL_FALSE;
+   }
 
-      /* Step 1:
-       * Extract the color and alpha combine function arguments.
-       */
-      for ( i = 0 ; i < numColorArgs ; i++ ) {
-	 op = texUnit->CombineOperandRGB[i] - GL_SRC_COLOR;
-	 switch ( texUnit->CombineSourceRGB[i] ) {
-	 case GL_TEXTURE:
-	    color_arg[i] = radeon_texture_color[op][unit];
-	    break;
-	 case GL_CONSTANT_EXT:
-	    color_arg[i] = radeon_tfactor_color[op];
-	    break;
-	 case GL_PRIMARY_COLOR_EXT:
-	    color_arg[i] = radeon_primary_color[op];
-	    break;
-	 case GL_PREVIOUS_EXT:
-	    color_arg[i] = radeon_previous_color[op];
-	    break;
-	 default:
-	    return;
-	 }
-      }
+   t->dirty_images |= (1 << level);
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-      for ( i = 0 ; i < numAlphaArgs ; i++ ) {
-	 op = texUnit->CombineOperandA[i] - GL_SRC_ALPHA;
-	 switch ( texUnit->CombineSourceA[i] ) {
-	 case GL_TEXTURE:
-	    alpha_arg[i] = radeon_texture_alpha[op][unit];
-	    break;
-	 case GL_CONSTANT_EXT:
-	    alpha_arg[i] = radeon_tfactor_alpha[op];
-	    break;
-	 case GL_PRIMARY_COLOR_EXT:
-	    alpha_arg[i] = radeon_primary_alpha[op];
-	    break;
-	 case GL_PREVIOUS_EXT:
-	    alpha_arg[i] = radeon_previous_alpha[op];
-	    break;
-	 default:
-	    return;
-	 }
-      }
+   return GL_TRUE;
+}
 
-      /* Step 2:
-       * Build up the color and alpha combine functions.
-       */
-      switch ( texUnit->CombineModeRGB ) {
-      case GL_REPLACE:
-	 color_combine = (RADEON_COLOR_ARG_A_ZERO |
-			  RADEON_COLOR_ARG_B_ZERO |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, C );
-	 break;
-      case GL_MODULATE:
-	 color_combine = (RADEON_COLOR_ARG_C_ZERO |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, A );
-	 RADEON_COLOR_ARG( 1, B );
-	 break;
-      case GL_ADD:
-	 color_combine = (RADEON_COLOR_ARG_B_ZERO |
-			  RADEON_COMP_ARG_B |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, A );
-	 RADEON_COLOR_ARG( 1, C );
-	 break;
-      case GL_ADD_SIGNED_EXT:
-	 color_combine = (RADEON_COLOR_ARG_B_ZERO |
-			  RADEON_COMP_ARG_B |
-			  RADEON_BLEND_CTL_ADDSIGNED |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, A );
-	 RADEON_COLOR_ARG( 1, C );
-	 break;
-      case GL_INTERPOLATE_EXT:
-	 color_combine = (RADEON_BLEND_CTL_BLEND |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, B );
-	 RADEON_COLOR_ARG( 1, A );
-	 RADEON_COLOR_ARG( 2, C );
-	 break;
-      case GL_DOT3_RGB_EXT:
-      case GL_DOT3_RGBA_EXT:
-	 color_combine = (RADEON_COLOR_ARG_C_ZERO |
-			  RADEON_BLEND_CTL_DOT3 |
-			  RADEON_CLAMP_TX);
-	 RADEON_COLOR_ARG( 0, A );
-	 RADEON_COLOR_ARG( 1, B );
-	 break;
-      default:
-	 return;
-      }
+static GLboolean
+radeonDDTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
+		       GLint xoffset, GLint yoffset,
+		       GLsizei width, GLsizei height,
+		       GLenum format, GLenum type,
+		       const GLvoid *pixels,
+		       const struct gl_pixelstore_attrib *packing,
+		       struct gl_texture_object *texObj,
+		       struct gl_texture_image *texImage )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
 
-      switch ( texUnit->CombineModeA ) {
-      case GL_REPLACE:
-	 alpha_combine = (RADEON_ALPHA_ARG_A_ZERO |
-			  RADEON_ALPHA_ARG_B_ZERO |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, C );
-	 break;
-      case GL_MODULATE:
-	 alpha_combine = (RADEON_ALPHA_ARG_C_ZERO |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, A );
-	 RADEON_ALPHA_ARG( 1, B );
-	 break;
-      case GL_ADD:
-	 alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
-			  RADEON_COMP_ARG_B |
-			  RADEON_BLEND_CTL_ADD |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, A );
-	 RADEON_ALPHA_ARG( 1, C );
-	 break;
-      case GL_ADD_SIGNED_EXT:
-	 alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
-			  RADEON_COMP_ARG_B |
-			  RADEON_BLEND_CTL_ADDSIGNED |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, A );
-	 RADEON_ALPHA_ARG( 1, C );
-	 break;
-      case GL_INTERPOLATE_EXT:
-	 alpha_combine = (RADEON_BLEND_CTL_BLEND |
-			  RADEON_CLAMP_TX);
-	 RADEON_ALPHA_ARG( 0, B );
-	 RADEON_ALPHA_ARG( 1, A );
-	 RADEON_ALPHA_ARG( 2, C );
-	 break;
-      default:
-	 return;
-      }
+   if ( target != GL_TEXTURE_2D )
+      return GL_FALSE;
 
-      if ( texUnit->CombineModeRGB == GL_DOT3_RGB_EXT ) {
-	 alpha_combine |= RADEON_DOT_ALPHA_DONT_REPLICATE;
-      }
+   /* FIXME: Can this ever be NULL???
+    */
+   ASSERT( t );
 
-      /* Step 3:
-       * Apply the scale factor.  The EXT extension has a somewhat
-       * unnecessary restriction that the scale must be 4x.  The ARB
-       * extension will likely drop this and we can just apply the
-       * scale factors regardless.
-       */
-      if ( texUnit->CombineModeRGB != GL_DOT3_RGB_EXT &&
-	   texUnit->CombineModeRGB != GL_DOT3_RGBA_EXT ) {
-	 color_combine |= (texUnit->CombineScaleShiftRGB << 21);
-	 alpha_combine |= (texUnit->CombineScaleShiftA << 21);
-      } else {
-	 color_combine |= RADEON_SCALE_4X;
-	 alpha_combine |= RADEON_SCALE_4X;
-      }
+   if ( t->bound )
+      FLUSH_BATCH( rmesa );
 
-      /* All done!
-       */
-      break;
+   texFormat = texImage->TexFormat;
 
-   default:
-      return;
+   if ( !_mesa_convert_texsubimage2d( texFormat->IntFormat,
+				      xoffset, yoffset, width, height,
+				      texImage->Width, format, type, packing,
+				      pixels, t->image[level].data ) ) {
+      /*fprintf( stderr, "   *** convert failed!\n" );*/
+      return GL_FALSE;
    }
 
-   rmesa->color_combine[source] = color_combine;
-   rmesa->alpha_combine[source] = alpha_combine;
+   t->dirty_images |= (1 << level);
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
+
+   return GL_TRUE;
 }
 
-static void radeonUpdateTextureObject( GLcontext *ctx, int unit )
+/* GH: This is undoubtedly broken...
+ */
+static GLboolean
+radeonDDTexSubImage3D( GLcontext *ctx, GLenum target, GLint level,
+		       GLint xoffset, GLint yoffset, GLint zoffset,
+		       GLsizei width, GLsizei height, GLint depth,
+		       GLenum format, GLenum type,
+		       const GLvoid *pixels,
+		       const struct gl_pixelstore_attrib *packing,
+		       struct gl_texture_object *texObj,
+		       struct gl_texture_image *texImage )
 {
    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   int source = rmesa->tmu_source[unit];
-   struct gl_texture_object *tObj;
-   radeonTexObjPtr t;
-   GLuint enabled;
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   const struct gl_texture_format *texFormat;
 
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p, %d )\n",
-	       __FUNCTION__, ctx, unit );
-   }
-
-   enabled = (ctx->Texture.ReallyEnabled >> (source * 4)) & TEXTURE0_ANY;
-   if ( enabled != TEXTURE0_2D && enabled != TEXTURE0_1D ) {
-      if ( enabled )
-	 rmesa->Fallback |= RADEON_FALLBACK_TEXTURE;
-      return;
-   }
+   if ( target != GL_TEXTURE_3D )
+      return GL_FALSE;
 
-   /* Only update the hardware texture state if the texture is current,
-    * complete and enabled.
+   /* FIXME: Can this ever be NULL???
     */
-   tObj = ctx->Texture.Unit[source].Current;
-   if ( !tObj || !tObj->Complete )
-      return;
-
-   if ( ( tObj != ctx->Texture.Unit[source].CurrentD[2] ) &&
-	( tObj != ctx->Texture.Unit[source].CurrentD[1] ) )
-      return;
+   ASSERT( t );
 
-   if ( !tObj->DriverData ) {
-      /* If this is the first time the texture has been used, then create
-       * a new texture object for it.
-       */
-      radeonCreateTexObj( rmesa, tObj );
-
-      if ( !tObj->DriverData ) {
-	 /* Can't create a texture object... */
-	 fprintf( stderr, "%s: texture object creation failed!\n",
-		  __FUNCTION__ );
-	 rmesa->Fallback |= RADEON_FALLBACK_TEXTURE;
-	 return;
-      }
-   }
-
-   /* We definately have a valid texture now */
-   t = tObj->DriverData;
-
-   /* Force the texture unit state to be loaded into the hardware */
-   rmesa->dirty |= RADEON_UPLOAD_CONTEXT | (RADEON_UPLOAD_TEX0 << unit);
+   if ( t->bound )
+      FLUSH_BATCH( rmesa );
 
-   /* Force any texture images to be loaded into the hardware */
-   if ( t->dirty_images )
-      rmesa->dirty |= (RADEON_UPLOAD_TEX0IMAGES << unit);
+   texFormat = texImage->TexFormat;
 
-   /* Bind to the given texture unit */
-   rmesa->CurrentTexObj[unit] = t;
-   t->bound = unit + 1;
+   if ( !_mesa_convert_texsubimage3d( texFormat->IntFormat,
+				      xoffset, yoffset, zoffset,
+				      width, height, depth,
+				      texImage->Width, texImage->Height,
+				      format, type, packing,
+				      pixels, t->image[level].data ) ) {
+      /*fprintf( stderr, "   *** convert failed!\n" );*/
+      return GL_FALSE;
+   }
 
-   if ( t->memBlock )
-      radeonUpdateTexLRU( rmesa, t );
+   t->dirty_images |= (1 << level);
+   rmesa->new_state |= RADEON_NEW_TEXTURE;
 
-   switch ( unit ) {
-   case 0:
-      rmesa->setup.pp_cntl |= (RADEON_TEX_0_ENABLE |
-			       RADEON_TEX_BLEND_0_ENABLE);
-      break;
-   case 1:
-      rmesa->setup.pp_cntl |= (RADEON_TEX_1_ENABLE |
-			       RADEON_TEX_BLEND_1_ENABLE);
-      break;
-   }
+   return GL_TRUE;
 }
 
-void radeonUpdateTextureState( GLcontext *ctx )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
 
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p ) en=0x%x\n",
-	       __FUNCTION__, ctx, ctx->Texture.ReallyEnabled );
-   }
-
-   /* Clear any texturing fallbacks */
-   rmesa->Fallback &= ~RADEON_FALLBACK_TEXTURE;
+/* ================================================================
+ * DEPRECATED...
+ */
 
-   /* Unbind any currently bound textures */
-   if ( rmesa->CurrentTexObj[0] ) rmesa->CurrentTexObj[0]->bound = 0;
-   if ( rmesa->CurrentTexObj[1] ) rmesa->CurrentTexObj[1]->bound = 0;
-   rmesa->CurrentTexObj[0] = NULL;
-   rmesa->CurrentTexObj[1] = NULL;
+static GLvoid *radeonDDGetTexImage( GLcontext *ctx, GLenum target, GLint level,
+				    const struct gl_texture_object *texObj,
+				    GLenum *formatOut, GLenum *typeOut,
+				    GLboolean *freeImageOut )
+{
+   const struct gl_texture_image *texImage = texObj->Image[level];
+   const struct gl_texture_format *texFormat = texImage->TexFormat;
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+   GLubyte *data;
 
-   if ( ctx->Enabled & (TEXTURE0_3D|TEXTURE1_3D) )
-      rmesa->Fallback |= RADEON_FALLBACK_TEXTURE;
+   if ( !t || !t->image[level].data )
+      return NULL;
 
-   /* Disable all texturing until it is known to be good */
-   rmesa->setup.pp_cntl &= ~(RADEON_TEX_ENABLE_MASK |
-			     RADEON_TEX_BLEND_ENABLE_MASK);
+   data = (GLubyte *) MALLOC( texImage->Width * texImage->Height * 4 );
+   if ( !data )
+      return NULL;
 
-   radeonUpdateTextureObject( ctx, 0 );
-   radeonUpdateTextureEnv( ctx, 0 );
+   if ( 0 )
+      fprintf( stderr, "   in=%d out=%s\n",
+	       texFormat->IntFormat,
+	       gl_lookup_enum_by_nr( texImage->Format ) );
 
-   if ( rmesa->multitex ) {
-      radeonUpdateTextureObject( ctx, 1 );
-      radeonUpdateTextureEnv( ctx, 1 );
+   switch ( target ) {
+   case GL_TEXTURE_1D:
+      _mesa_unconvert_teximage1d( texFormat->IntFormat, texImage->Format,
+				  texImage->Width,
+				  t->image[level].data, data );
+      break;
+   case GL_TEXTURE_2D:
+      _mesa_unconvert_teximage2d( texFormat->IntFormat, texImage->Format,
+				  texImage->Width, texImage->Height,
+				  t->image[level].data, data );
+      break;
+   default:
+      return NULL;
    }
 
-   rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+   *formatOut = texImage->Format;
+   *typeOut = GL_UNSIGNED_BYTE;
+   *freeImageOut = GL_TRUE;
+
+   return data;
 }
 
 
 /* ================================================================
- * DD interface texturing functions
- *
- * FIXME: Many of these are deprecated -- we should move to the new
- * single-copy texture interface.
+ * Texture state callbacks
  */
+
 #define SCALED_FLOAT_TO_BYTE( x, scale ) \
 		((((GLint)((256.0F / scale) * (x))) - 1) / 2)
 
@@ -2067,79 +805,6 @@ static void radeonDDTexEnv( GLcontext *ctx, GLenum target,
    }
 }
 
-static void radeonDDTexImage( GLcontext *ctx, GLenum target,
-			      struct gl_texture_object *tObj, GLint level,
-			      GLint internalFormat,
-			      const struct gl_texture_image *image )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   radeonTexObjPtr t;
-
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API )
-      fprintf( stderr, "%s( %p, level %d )\n", __FUNCTION__, tObj, level );
-
-   if ( ( target != GL_TEXTURE_2D ) &&
-	( target != GL_TEXTURE_1D ) )
-      return;
-
-   if ( level >= RADEON_MAX_TEXTURE_LEVELS )
-      return;
-
-   t = (radeonTexObjPtr)tObj->DriverData;
-   if ( t ) {
-      if ( t->bound ) FLUSH_BATCH( rmesa );
-
-      /* Destroy the old texture, and upload a new one.  The actual
-       * uploading of the texture image occurs in the UploadSubImage
-       * function.
-       */
-      radeonDestroyTexObj( rmesa, t );
-      rmesa->new_state |= RADEON_NEW_TEXTURE;
-   }
-}
-
-static void radeonDDTexSubImage( GLcontext *ctx, GLenum target,
-				 struct gl_texture_object *tObj, GLint level,
-				 GLint xoffset, GLint yoffset,
-				 GLsizei width, GLsizei height,
-				 GLint internalFormat,
-				 const struct gl_texture_image *image )
-{
-   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   radeonTexObjPtr t;
-
-   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
-      fprintf( stderr, "%s( %p, level %d ) size: %d,%d of %d,%d\n",
-	       __FUNCTION__, tObj, level, width, height,
-	       image->Width, image->Height );
-   }
-
-   if ( ( target != GL_TEXTURE_2D ) &&
-	( target != GL_TEXTURE_1D ) )
-      return;
-
-   if ( level >= RADEON_MAX_TEXTURE_LEVELS )
-      return;
-
-   t = (radeonTexObjPtr)tObj->DriverData;
-   if ( t ) {
-      if ( t->bound ) FLUSH_BATCH( rmesa );
-
-#if 0
-      /* FIXME: Only upload textures if we already have space in the heap.
-       */
-      LOCK_HARDWARE( rmesa );
-      radeonUploadSubImage( rmesa, t, level,
-			    xoffset, yoffset, width, height );
-      UNLOCK_HARDWARE( rmesa );
-#else
-      radeonDestroyTexObj( rmesa, t );
-#endif
-      /* Update the context state */
-      rmesa->new_state |= RADEON_NEW_TEXTURE;
-   }
-}
-
 static void radeonDDTexParameter( GLcontext *ctx, GLenum target,
 				  struct gl_texture_object *tObj,
 				  GLenum pname, const GLfloat *params )
@@ -2156,7 +821,7 @@ static void radeonDDTexParameter( GLcontext *ctx, GLenum target,
     * created with current state before it is used, so we don't have
     * to do anything now.
     */
-   if ( !t || !t->bound )
+   if ( !t )
       return;
 
    if ( ( target != GL_TEXTURE_2D ) &&
@@ -2192,34 +857,51 @@ static void radeonDDBindTexture( GLcontext *ctx, GLenum target,
 				 struct gl_texture_object *tObj )
 {
    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
-   GLint unit = ctx->Texture.CurrentUnit;
+   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
+   GLuint unit = ctx->Texture.CurrentUnit;
 
    if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
       fprintf( stderr, "%s( %p ) unit=%d\n",
-	       __FUNCTION__, tObj, ctx->Texture.CurrentUnit );
+	       __FUNCTION__, tObj, unit );
    }
 
    FLUSH_BATCH( rmesa );
 
+   if ( !t ) {
+      t = radeonAllocTexObj( tObj );
+      tObj->DriverData = t;
+   }
+
+   /* Unbind a currently bound texture.
+    */
    if ( rmesa->CurrentTexObj[unit] ) {
-      rmesa->CurrentTexObj[unit]->bound = 0;
+      rmesa->CurrentTexObj[unit]->bound &= ~(unit + 1);
       rmesa->CurrentTexObj[unit] = NULL;
    }
 
+   /* Bind to the given texture unit.
+    */
+   rmesa->CurrentTexObj[unit] = t;
+   t->bound |= unit + 1;
+
    rmesa->new_state |= RADEON_NEW_TEXTURE;
 }
 
 static void radeonDDDeleteTexture( GLcontext *ctx,
-				 struct gl_texture_object *tObj )
+				   struct gl_texture_object *tObj )
 {
    radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
    radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData;
 
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, __FUNCTION__ "( %p )\n", tObj );
+   }
+
    if ( t ) {
       if ( t->bound ) {
 	 FLUSH_BATCH( rmesa );
-
-	 rmesa->CurrentTexObj[t->bound-1] = 0;
+	 if ( t->bound & TEX_0 ) rmesa->CurrentTexObj[0] = NULL;
+	 if ( t->bound & TEX_1 ) rmesa->CurrentTexObj[1] = NULL;
 	 rmesa->new_state |= RADEON_NEW_TEXTURE;
       }
 
@@ -2240,14 +922,19 @@ static GLboolean radeonDDIsTextureResident( GLcontext *ctx,
 
 void radeonDDInitTextureFuncs( GLcontext *ctx )
 {
+   ctx->Driver.TexImage1D		= radeonDDTexImage1D;
+   ctx->Driver.TexImage2D		= radeonDDTexImage2D;
+   ctx->Driver.TexImage3D		= NULL /*radeonDDTexImage3D*/;
+   ctx->Driver.TexSubImage1D		= radeonDDTexSubImage1D;
+   ctx->Driver.TexSubImage2D		= radeonDDTexSubImage2D;
+   ctx->Driver.TexSubImage3D		= NULL /*radeonDDTexSubImage3D*/;
+   ctx->Driver.GetTexImage		= radeonDDGetTexImage;
    ctx->Driver.TexEnv			= radeonDDTexEnv;
-   ctx->Driver.TexImage			= radeonDDTexImage;
-   ctx->Driver.TexSubImage		= radeonDDTexSubImage;
    ctx->Driver.TexParameter		= radeonDDTexParameter;
    ctx->Driver.BindTexture		= radeonDDBindTexture;
    ctx->Driver.DeleteTexture		= radeonDDDeleteTexture;
-   ctx->Driver.UpdateTexturePalette	= NULL;
-   ctx->Driver.ActiveTexture		= NULL;
    ctx->Driver.IsTextureResident	= radeonDDIsTextureResident;
    ctx->Driver.PrioritizeTexture	= NULL;
+   ctx->Driver.ActiveTexture		= NULL;
+   ctx->Driver.UpdateTexturePalette	= NULL;
 }
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h b/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h
index 6bf967761..967048dd1 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h
@@ -45,9 +45,11 @@ extern int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t );
 
 extern void radeonAgeTextures( radeonContextPtr rmesa, int heap );
 extern void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t );
+extern void radeonSwapOutTexObj( radeonContextPtr rmesa, radeonTexObjPtr t );
 
 extern void radeonPrintLocalLRU( radeonContextPtr rmesa, int heap );
 extern void radeonPrintGlobalLRU( radeonContextPtr rmesa, int heap );
+extern void radeonUpdateTexLRU(radeonContextPtr rmesa, radeonTexObjPtr t );
 
 extern void radeonDDInitTextureFuncs( GLcontext *ctx );
 
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_texmem.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_texmem.c
new file mode 100644
index 000000000..13f6d4b59
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_texmem.c
@@ -0,0 +1,508 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+on 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 above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, VA LINUX SYSTEMS AND/OR THEIR 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.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <martin@valinux.com>
+ *   Gareth Hughes <gareth@valinux.com>
+ *
+ */
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_vb.h"
+#include "radeon_tex.h"
+
+#include "mmath.h"
+#include "simple_list.h"
+#include "enums.h"
+#include "mem.h"
+
+
+/* Destroy hardware state associated with texture `t'.
+ */
+void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+#if ENABLE_PERF_BOXES
+   /* Bump the performace counter */
+   rmesa->c_textureSwaps++;
+#endif
+   if ( !t ) return;
+
+   if ( t->memBlock ) {
+      mmFreeMem( t->memBlock );
+      t->memBlock = NULL;
+   }
+
+   if ( t->tObj )
+      t->tObj->DriverData = NULL;
+
+   if ( t->bound & TEX_0 ) rmesa->CurrentTexObj[0] = NULL;
+   if ( t->bound & TEX_1 ) rmesa->CurrentTexObj[1] = NULL;
+
+   remove_from_list( t );
+   FREE( t );
+}
+
+/* Keep track of swapped out texture objects.
+ */
+void radeonSwapOutTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+#if ENABLE_PERF_BOXES
+   /* Bump the performace counter */
+   rmesa->c_textureSwaps++;
+#endif
+   if ( t->memBlock ) {
+      mmFreeMem( t->memBlock );
+      t->memBlock = NULL;
+   }
+
+   t->dirty_images = ~0;
+   move_to_tail( &rmesa->SwappedOut, t );
+}
+
+/* Print out debugging information about texture LRU.
+ */
+void radeonPrintLocalLRU( radeonContextPtr rmesa, int heap )
+{
+   radeonTexObjPtr t;
+   int sz = 1 << (rmesa->radeonScreen->logTexGranularity[heap]);
+
+   fprintf( stderr, "\nLocal LRU, heap %d:\n", heap );
+
+   foreach ( t, &rmesa->TexObjList[heap] ) {
+      if (!t->tObj) {
+	 fprintf( stderr, "Placeholder %d at 0x%x sz 0x%x\n",
+		  t->memBlock->ofs / sz,
+		  t->memBlock->ofs,
+		  t->memBlock->size );
+      } else {
+	 fprintf( stderr, "Texture (bound %d) at 0x%x sz 0x%x\n",
+		  t->bound,
+		  t->memBlock->ofs,
+		  t->memBlock->size );
+      }
+   }
+
+   fprintf( stderr, "\n" );
+}
+
+void radeonPrintGlobalLRU( radeonContextPtr rmesa, int heap )
+{
+   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
+   int i, j;
+
+   fprintf( stderr, "\nGlobal LRU, heap %d list %p:\n", heap, list );
+
+   for ( i = 0, j = RADEON_NR_TEX_REGIONS ; i < RADEON_NR_TEX_REGIONS ; i++ ) {
+      fprintf( stderr, "list[%d] age %d next %d prev %d\n",
+	       j, list[j].age, list[j].next, list[j].prev );
+      j = list[j].next;
+      if ( j == RADEON_NR_TEX_REGIONS ) break;
+   }
+
+   if ( j != RADEON_NR_TEX_REGIONS ) {
+      fprintf( stderr, "Loop detected in global LRU\n" );
+      for ( i = 0 ; i < RADEON_NR_TEX_REGIONS ; i++ ) {
+	 fprintf( stderr, "list[%d] age %d next %d prev %d\n",
+		  i, list[i].age, list[i].next, list[i].prev );
+      }
+   }
+
+   fprintf( stderr, "\n" );
+}
+
+/* Reset the global texture LRU.
+ */
+static void radeonResetGlobalLRU( radeonContextPtr rmesa, int heap )
+{
+   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
+   int sz = 1 << rmesa->radeonScreen->logTexGranularity[heap];
+   int i;
+
+   /*
+    * (Re)initialize the global circular LRU list.  The last element in
+    * the array (RADEON_NR_TEX_REGIONS) is the sentinal.  Keeping it at
+    * the end of the array allows it to be addressed rationally when
+    * looking up objects at a particular location in texture memory.
+    */
+   for ( i = 0 ; (i+1) * sz <= rmesa->radeonScreen->texSize[heap] ; i++ ) {
+      list[i].prev = i-1;
+      list[i].next = i+1;
+      list[i].age = 0;
+   }
+
+   i--;
+   list[0].prev = RADEON_NR_TEX_REGIONS;
+   list[i].prev = i-1;
+   list[i].next = RADEON_NR_TEX_REGIONS;
+   list[RADEON_NR_TEX_REGIONS].prev = i;
+   list[RADEON_NR_TEX_REGIONS].next = 0;
+   rmesa->sarea->texAge[heap] = 0;
+}
+
+/* Update the local and glock texture LRUs.
+ */
+void radeonUpdateTexLRU(radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+   int heap = t->heap;
+   radeon_tex_region_t *list = rmesa->sarea->texList[heap];
+   int sz = rmesa->radeonScreen->logTexGranularity[heap];
+   int start = t->memBlock->ofs >> sz;
+   int end = (t->memBlock->ofs + t->memBlock->size-1) >> sz;
+   int i;
+
+   rmesa->lastTexAge[heap] = ++rmesa->sarea->texAge[heap];
+
+   if ( !t->memBlock ) {
+      fprintf( stderr, "no memblock\n\n" );
+      return;
+   }
+
+   /* Update our local LRU */
+   move_to_head( &rmesa->TexObjList[heap], t );
+
+   /* Update the global LRU */
+   for ( i = start ; i <= end ; i++ ) {
+      list[i].in_use = 1;
+      list[i].age = rmesa->lastTexAge[heap];
+
+      /* remove_from_list(i) */
+      list[(CARD32)list[i].next].prev = list[i].prev;
+      list[(CARD32)list[i].prev].next = list[i].next;
+
+      /* insert_at_head(list, i) */
+      list[i].prev = RADEON_NR_TEX_REGIONS;
+      list[i].next = list[RADEON_NR_TEX_REGIONS].next;
+      list[(CARD32)list[RADEON_NR_TEX_REGIONS].next].prev = i;
+      list[RADEON_NR_TEX_REGIONS].next = i;
+   }
+
+   if ( 0 ) {
+      radeonPrintGlobalLRU( rmesa, t->heap );
+      radeonPrintLocalLRU( rmesa, t->heap );
+   }
+}
+
+/* Update our notion of what textures have been changed since we last
+ * held the lock.  This pertains to both our local textures and the
+ * textures belonging to other clients.  Keep track of other client's
+ * textures by pushing a placeholder texture onto the LRU list -- these
+ * are denoted by (tObj == NULL).
+ */
+static void radeonTexturesGone( radeonContextPtr rmesa, int heap,
+				int offset, int size, int in_use )
+{
+   radeonTexObjPtr t, tmp;
+
+   foreach_s ( t, tmp, &rmesa->TexObjList[heap] ) {
+      if ( t->memBlock->ofs >= offset + size ||
+	   t->memBlock->ofs + t->memBlock->size <= offset )
+	 continue;
+
+      /* It overlaps - kick it out.  Need to hold onto the currently
+       * bound objects, however.
+       */
+      if ( t->bound ) {
+	 radeonSwapOutTexObj( rmesa, t );
+      } else {
+	 radeonDestroyTexObj( rmesa, t );
+      }
+   }
+
+   if ( in_use ) {
+      t = (radeonTexObjPtr) CALLOC( sizeof(*t) );
+      if ( !t ) return;
+
+      t->memBlock = mmAllocMem( rmesa->texHeap[heap], size, 0, offset );
+      if ( !t->memBlock ) {
+	 fprintf( stderr, "Couldn't alloc placeholder sz %x ofs %x\n",
+		  (int)size, (int)offset );
+	 mmDumpMemInfo( rmesa->texHeap[heap] );
+	 return;
+      }
+      insert_at_head( &rmesa->TexObjList[heap], t );
+   }
+}
+
+/* Update our client's shared texture state.  If another client has
+ * modified a region in which we have textures, then we need to figure
+ * out which of our textures has been removed, and update our global
+ * LRU.
+ */
+void radeonAgeTextures( radeonContextPtr rmesa, int heap )
+{
+   RADEONSAREAPrivPtr sarea = rmesa->sarea;
+
+   if ( sarea->texAge[heap] != rmesa->lastTexAge[heap] ) {
+      int sz = 1 << rmesa->radeonScreen->logTexGranularity[heap];
+      int nr = 0;
+      int idx;
+
+      for ( idx = sarea->texList[heap][RADEON_NR_TEX_REGIONS].prev ;
+	    idx != RADEON_NR_TEX_REGIONS && nr < RADEON_NR_TEX_REGIONS ;
+	    idx = sarea->texList[heap][idx].prev, nr++ )
+      {
+	 /* If switching texturing schemes, then the SAREA might not
+	  * have been properly cleared, so we need to reset the
+	  * global texture LRU.
+	  */
+	 if ( idx * sz > rmesa->radeonScreen->texSize[heap] ) {
+	    nr = RADEON_NR_TEX_REGIONS;
+	    break;
+	 }
+
+	 if ( sarea->texList[heap][idx].age > rmesa->lastTexAge[heap] ) {
+	    radeonTexturesGone( rmesa, heap, idx * sz, sz,
+				sarea->texList[heap][idx].in_use );
+	 }
+      }
+
+      if ( nr == RADEON_NR_TEX_REGIONS ) {
+	 radeonTexturesGone( rmesa, heap, 0,
+			     rmesa->radeonScreen->texSize[heap], 0 );
+	 radeonResetGlobalLRU( rmesa, heap );
+      }
+
+      rmesa->dirty |= (RADEON_UPLOAD_CONTEXT |
+		       RADEON_UPLOAD_TEX0IMAGES |
+		       RADEON_UPLOAD_TEX1IMAGES);
+      rmesa->lastTexAge[heap] = sarea->texAge[heap];
+   }
+}
+
+
+/* ================================================================
+ * Texture image uploads
+ */
+
+/* Upload the texture image associated with texture `t' at level `level'
+ * at the address relative to `start'.
+ */
+static void radeonUploadSubImage( radeonContextPtr rmesa,
+				  radeonTexObjPtr t, GLint level,
+				  GLint x, GLint y, GLint width, GLint height )
+{
+   struct gl_texture_image *texImage;
+   const struct gl_texture_format *texFormat;
+   GLint texelsPerDword = 0;
+   GLint imageX, imageY, imageWidth, imageHeight;
+   GLint blitX, blitY, blitWidth, blitHeight;
+   GLint imageRows, blitRows;
+   GLint remaining;
+   GLint format, dwords;
+   CARD32 pitch, offset;
+   drmBufPtr buffer;
+   CARD32 *dst;
+   GLint ret;
+
+   /* Ensure we have a valid texture to upload */
+   texImage = t->tObj->Image[level];
+   if ( !texImage )
+      return;
+
+   texFormat = texImage->TexFormat;
+
+   switch ( texFormat->TexelBytes ) {
+   case 1:
+      texelsPerDword = 4;
+      break;
+   case 2:
+      texelsPerDword = 2;
+      break;
+   case 4:
+      texelsPerDword = 1;
+      break;
+   }
+
+   format = t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK;
+
+   imageX = 0;
+   imageY = 0;
+   imageWidth = texImage->Width;
+   imageHeight = texImage->Height;
+
+   blitX = t->image[level].x;
+   blitY = t->image[level].y;
+   blitWidth = t->image[level].width;
+   blitHeight = t->image[level].height;
+
+   /* dwords = t->image[level].dwords; */
+   offset = t->bufAddr;
+   pitch = (t->image[0].width * texFormat->TexelBytes) / 64;
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
+      fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
+	       imageWidth, imageHeight, imageX, imageY );
+      fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
+	       blitWidth, blitHeight, blitX, blitY );
+      fprintf( stderr, "       blit ofs: 0x%07x pitch: 0x%x "
+	       "level: %d format: %x\n",
+	       (GLuint)offset, (GLuint)pitch, level, format );
+   }
+
+   ret = drmRadeonLoadTexture( rmesa->driFd, offset, pitch, format,
+			       imageWidth, imageHeight, &t->image[level] );
+
+   if ( ret ) {
+      UNLOCK_HARDWARE( rmesa );
+      fprintf( stderr, "drmRadeonTextureBlit: return = %d\n", ret );
+      exit( 1 );
+   }
+
+   rmesa->new_state |= RADEON_NEW_CONTEXT;
+   rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_MASKS;
+}
+
+/* Upload the texture images associated with texture `t'.  This might
+ * require removing our own and/or other client's texture objects to
+ * make room for these images.
+ */
+int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+   int i;
+   int heap;
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p, %p )\n",
+	       __FUNCTION__, rmesa->glCtx, t->tObj );
+   }
+
+   if ( !t )
+      return 0;
+
+   /* Choose the heap appropriately */
+   heap = t->heap = RADEON_CARD_HEAP;
+#if 0
+   if ( !rmesa->radeonScreen->IsPCI &&
+	t->totalSize > rmesa->radeonScreen->texSize[heap] ) {
+      heap = t->heap = RADEON_AGP_HEAP;
+   }
+#endif
+
+   /* Do we need to eject LRU texture objects? */
+   if ( !t->memBlock ) {
+      /* Allocate a memory block on a 4k boundary (1<<12 == 4096) */
+      t->memBlock = mmAllocMem( rmesa->texHeap[heap],
+				t->totalSize, 12, 0 );
+
+#if 0
+      /* Try AGP before kicking anything out of local mem */
+      if ( !t->memBlock && heap == RADEON_CARD_HEAP ) {
+	 t->memBlock = mmAllocMem( rmesa->texHeap[RADEON_AGP_HEAP],
+				   t->totalSize, 12, 0 );
+
+	 if ( t->memBlock )
+	    heap = t->heap = RADEON_AGP_HEAP;
+      }
+#endif
+
+      /* Kick out textures until the requested texture fits */
+      while ( !t->memBlock ) {
+	 if ( rmesa->TexObjList[heap].prev->bound ) {
+	    fprintf( stderr,
+		     "radeonUploadTexImages: ran into bound texture\n" );
+	    return -1;
+	 }
+	 if ( rmesa->TexObjList[heap].prev ==
+	      &rmesa->TexObjList[heap] ) {
+	    if ( rmesa->radeonScreen->IsPCI ) {
+	       fprintf( stderr, "radeonUploadTexImages: upload texture "
+			"failure on local texture heaps, sz=%d\n",
+			t->totalSize );
+	       return -1;
+#if 0
+	    } else if ( heap == RADEON_CARD_HEAP ) {
+	       heap = t->heap = RADEON_AGP_HEAP;
+	       continue;
+#endif
+	    } else {
+	       fprintf( stderr, "radeonUploadTexImages: upload texture "
+			"failure on both local and AGP texture heaps, "
+			"sz=%d\n",
+			t->totalSize );
+	       return -1;
+	    }
+	 }
+
+	 radeonDestroyTexObj( rmesa, rmesa->TexObjList[heap].prev );
+
+	 t->memBlock = mmAllocMem( rmesa->texHeap[heap],
+				   t->totalSize, 12, 0 );
+      }
+
+      /* Set the base offset of the texture image */
+      t->bufAddr = rmesa->radeonScreen->texOffset[heap] + t->memBlock->ofs;
+
+      t->pp_txoffset = t->bufAddr;
+#if 0
+      /* Fix AGP texture offsets */
+      if ( heap == RADEON_AGP_HEAP ) {
+	  t->setup.pp_tx_offset += RADEON_AGP_TEX_OFFSET +
+	      rmesa->radeonScreen->agpTexOffset;
+      }
+#endif
+
+      /* Force loading the new state into the hardware */
+      switch ( t->bound ) {
+      case 1:
+	 rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_TEX0;
+	 break;
+
+      case 2:
+	 rmesa->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_TEX1;
+	 break;
+
+      default:
+	 return -1;
+      }
+   }
+
+   /* Let the world know we've used this memory recently */
+   radeonUpdateTexLRU( rmesa, t );
+
+   /* Upload any images that are new */
+   if ( t->dirty_images ) {
+      int levels = ((t->pp_txfilter & RADEON_MAX_MIP_LEVEL_MASK) >>
+		    RADEON_MAX_MIP_LEVEL_SHIFT);
+
+      for ( i = 0 ; i <= levels ; i++ ) {
+	 if ( t->dirty_images & (1 << i) ) {
+	    radeonUploadSubImage( rmesa, t, i, 0, 0,
+				  t->image[i].width, t->image[i].height );
+	 }
+      }
+
+      rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+   }
+
+   t->dirty_images = 0;
+   return 0;
+}
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_texobj.h b/xc/lib/GL/mesa/src/drv/radeon/radeon_texobj.h
index b63d5c6a7..7cec84e9a 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_texobj.h
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_texobj.h
@@ -40,17 +40,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_sarea.h"
 #include "mm.h"
 
-/* Handle the Radeon's tightly packed mipmaps and strict offset,
- * pitch rules for blits by assigning each mipmap a set of
- * coordinates that can be used for a hostdata blit.
- */
-typedef struct {
-   GLuint x;				/* Blit coordinates */
-   GLuint y;
-   GLuint width;			/* Blit dimensions */
-   GLuint height;
-   GLuint dwords;			/* Size of image level */
-} radeonTexImage;
+#define TEX_0 1
+#define TEX_1 2
 
 typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr;
 
@@ -72,16 +63,15 @@ struct radeon_tex_obj {
    GLint bound;				/* Texture unit currently bound to */
    GLint heap;				/* Texture heap currently stored in */
 
-   radeonTexImage image[RADEON_MAX_TEXTURE_LEVELS]; /* Image data for all
-						      mipmap levels */
+   drmRadeonTexImage image[RADEON_MAX_TEXTURE_LEVELS];
 
    GLint totalSize;			/* Total size of the texture
 					   including all mipmap levels */
-   GLint texelBytes;			/* Number of bytes per texel */
-
-   GLboolean hasAlpha;
 
-   radeon_texture_regs_t setup;		/* Setup regs for texture */
+   GLuint pp_txfilter;		        /* Hardware register values */
+   GLuint pp_txformat;
+   GLuint pp_txoffset;
+   GLuint pp_border_color;
 };
 
 #endif /* __RADEON_TEXOBJ_H__ */
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_texstate.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_texstate.c
new file mode 100644
index 000000000..2a076717e
--- /dev/null
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_texstate.c
@@ -0,0 +1,1103 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+on 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 above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ATI, VA LINUX SYSTEMS AND/OR THEIR 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.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Kevin E. Martin <martin@valinux.com>
+ *   Gareth Hughes <gareth@valinux.com>
+ *
+ */
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_vb.h"
+#include "radeon_tex.h"
+
+#include "mmath.h"
+#include "simple_list.h"
+#include "enums.h"
+#include "mem.h"
+
+static void radeonSetTexImages( radeonContextPtr rmesa,
+				struct gl_texture_object *tObj )
+{
+   radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData;
+   struct gl_texture_image *texImage = tObj->Image[0];
+   const struct gl_texture_format *texFormat = texImage->TexFormat;
+   GLint log2Width, log2Height, log2Size;
+   GLint totalSize;
+   GLint texelsPerDword = 0, blitWidth = 0, blitPitch = 0;
+   GLint x, y, width, height;
+   GLint i;
+
+   /* Calculate dimensions in log domain.
+    */
+   for ( i = 1, log2Height = 0 ; i < texImage->Height ; i *= 2 ) {
+      log2Height++;
+   }
+   for ( i = 1, log2Width = 0 ;  i < texImage->Width ;  i *= 2 ) {
+      log2Width++;
+   }
+   log2Size = MAX2( log2Width, log2Height );
+
+   /* The Radeon has a 64-byte minimum pitch for all blits.  We
+    * calculate the equivalent number of texels to simplify the
+    * calculation of the texture image area.
+    */
+   switch ( texFormat->TexelBytes ) {
+   case 4:
+      texelsPerDword = 1;
+      blitPitch = 16;
+      break;
+   case 2:
+      texelsPerDword = 2;
+      blitPitch = 32;
+      break;
+   case 1:
+      texelsPerDword = 4;
+      blitPitch = 64;
+      break;
+   }
+
+   /* Select the larger of the two widths for our global texture image
+    * coordinate space.  As the Radeon has very strict offset rules, we
+    * can't upload mipmaps directly and have to reference their location
+    * from the aligned start of the whole image.
+    */
+   blitWidth = MAX2( texImage->Width, blitPitch );
+
+   /* Calculate mipmap offsets and dimensions.
+    */
+   totalSize = 0;
+   x = 0;
+   y = 0;
+
+   for ( i = 0 ; i <= log2Size ; i++ ) {
+      GLuint size;
+
+      texImage = tObj->Image[i];
+      if ( !texImage )
+	 break;
+
+      width = texImage->Width;
+      height = texImage->Height;
+
+      /* Texture images have a minimum pitch of 32 bytes (half of the
+       * 64-byte minimum pitch for blits).  For images that have a
+       * width smaller than this, we must pad each texture image
+       * scanline out to this amount.
+       */
+      if ( width < blitPitch / 2 ) {
+	 width = blitPitch / 2;
+      }
+
+      size = width * height * texFormat->TexelBytes;
+      totalSize += size;
+      ASSERT( (totalSize & 31) == 0 );
+
+      while ( width < blitWidth && height > 1 ) {
+	 width *= 2;
+	 height /= 2;
+      }
+
+      t->image[i].x = x;
+      t->image[i].y = y;
+
+      t->image[i].width  = width;
+      t->image[i].height = height;
+
+      /* While blits must have a pitch of at least 64 bytes, mipmaps
+       * must be aligned on a 32-byte boundary (just like each texture
+       * image scanline).
+       */
+      if ( width >= blitWidth ) {
+	 y += height;
+      } else {
+	 x += width;
+	 if ( x >= blitWidth ) {
+	    x = 0;
+	    y++;
+	 }
+      }
+
+      if ( 0 ) {
+	 fprintf( stderr, "level=%d p=%d   %dx%d -> %dx%d at (%d,%d)\n",
+		  i, blitWidth, texImage->Width, texImage->Height,
+		  t->image[i].width, t->image[i].height,
+		  t->image[i].x, t->image[i].y );
+      }
+   }
+
+   /* Align the total size of texture memory block.
+    */
+   t->totalSize = (totalSize + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
+
+   /* Hardware state:
+    */
+   t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK;
+   t->pp_txfilter |= i << RADEON_MAX_MIP_LEVEL_SHIFT;
+
+   t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK |
+		       RADEON_TXFORMAT_HEIGHT_MASK);
+   t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) |
+		      (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT));
+}
+
+
+/* ================================================================
+ * Texture combine functions
+ */
+
+#define RADEON_DISABLE		0
+#define RADEON_REPLACE		1
+#define RADEON_MODULATE		2
+#define RADEON_DECAL		3
+#define RADEON_BLEND		4
+#define RADEON_ADD		5
+#define RADEON_MAX_COMBFUNC	6
+
+static GLuint radeon_color_combine[][RADEON_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00802800
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800142
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T0_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c2d42
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T0_COLOR |
+       RADEON_COLOR_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c2902
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00812802
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 1:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00803000
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800182
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T1_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c3582
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T1_COLOR |
+       RADEON_COLOR_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c3102
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00813002
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 2:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00803800
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x008001c2
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T2_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c3dc2
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T2_COLOR |
+       RADEON_COLOR_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c3902
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00813802
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   }
+};
+
+static GLuint radeon_alpha_combine[][RADEON_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800500
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T0_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 1:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800600
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T1_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 2:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800700
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800071
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T2_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800071
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800021
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   }
+};
+
+
+/* GL_EXT_texture_env_combine support
+ */
+
+/* The color tables have combine functions for GL_SRC_COLOR,
+ * GL_ONE_MINUS_SRC_COLOR, GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
+ */
+static GLuint radeon_texture_color[][RADEON_MAX_TEXTURE_UNITS] =
+{
+   {
+      RADEON_COLOR_ARG_A_T0_COLOR,
+      RADEON_COLOR_ARG_A_T1_COLOR,
+      RADEON_COLOR_ARG_A_T2_COLOR
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_COLOR | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T1_COLOR | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T2_COLOR | RADEON_COMP_ARG_A
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_ALPHA,
+      RADEON_COLOR_ARG_A_T1_ALPHA,
+      RADEON_COLOR_ARG_A_T2_ALPHA
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
+   },
+};
+
+static GLuint radeon_tfactor_color[] =
+{
+   RADEON_COLOR_ARG_A_TFACTOR_COLOR,
+   RADEON_COLOR_ARG_A_TFACTOR_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_TFACTOR_ALPHA,
+   RADEON_COLOR_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_primary_color[] =
+{
+   RADEON_COLOR_ARG_A_DIFFUSE_COLOR,
+   RADEON_COLOR_ARG_A_DIFFUSE_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA,
+   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_previous_color[] =
+{
+   RADEON_COLOR_ARG_A_CURRENT_COLOR,
+   RADEON_COLOR_ARG_A_CURRENT_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_CURRENT_ALPHA,
+   RADEON_COLOR_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
+};
+
+/* The alpha tables only have GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
+ */
+static GLuint radeon_texture_alpha[][RADEON_MAX_TEXTURE_UNITS] =
+{
+   {
+      RADEON_ALPHA_ARG_A_T0_ALPHA,
+      RADEON_ALPHA_ARG_A_T1_ALPHA,
+      RADEON_ALPHA_ARG_A_T2_ALPHA
+   },
+   {
+      RADEON_ALPHA_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_ALPHA_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_ALPHA_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
+   },
+};
+
+static GLuint radeon_tfactor_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA,
+   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_primary_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA,
+   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_previous_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_CURRENT_ALPHA,
+   RADEON_ALPHA_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
+};
+
+
+/* Extract the arg from slot A, shift it into the correct argument slot
+ * and set the corresponding complement bit.
+ */
+#define RADEON_COLOR_ARG( n, arg )					\
+do {									\
+   color_combine |=							\
+      ((color_arg[n] & RADEON_COLOR_ARG_MASK)				\
+       << RADEON_COLOR_ARG_##arg##_SHIFT);				\
+   color_combine |=							\
+      ((color_arg[n] >> RADEON_COMP_ARG_SHIFT)				\
+       << RADEON_COMP_ARG_##arg##_SHIFT);				\
+} while (0)
+
+#define RADEON_ALPHA_ARG( n, arg )					\
+do {									\
+   alpha_combine |=							\
+      ((alpha_arg[n] & RADEON_ALPHA_ARG_MASK)				\
+       << RADEON_ALPHA_ARG_##arg##_SHIFT);				\
+   alpha_combine |=							\
+      ((alpha_arg[n] >> RADEON_COMP_ARG_SHIFT)				\
+       << RADEON_COMP_ARG_##arg##_SHIFT);				\
+} while (0)
+
+
+/* ================================================================
+ * Texture unit state management
+ */
+
+static void radeonUpdateTextureEnv( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int source = rmesa->tmu_source[unit];
+   struct gl_texture_object *tObj;
+   struct gl_texture_unit *texUnit;
+   GLuint enabled;
+   GLuint color_combine, alpha_combine;
+   GLuint color_arg[3], alpha_arg[3];
+   GLuint i, numColorArgs = 0, numAlphaArgs = 0;
+   GLuint op;
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
+      fprintf( stderr, "%s( %p, %d )\n",
+	       __FUNCTION__, ctx, unit );
+   }
+
+   enabled = (ctx->Texture.ReallyEnabled >> (source * 4)) & TEXTURE0_ANY;
+   if ( enabled != TEXTURE0_2D && enabled != TEXTURE0_1D )
+      return;
+
+   /* Only update the hardware texture state if the texture is current,
+    * complete and enabled.
+    */
+   texUnit = &ctx->Texture.Unit[source];
+   tObj = texUnit->Current;
+   if ( !tObj || !tObj->Complete )
+      return;
+
+   if ( ( tObj != texUnit->CurrentD[2] ) &&
+	( tObj != texUnit->CurrentD[1] ) )
+      return;
+
+   /* Set the texture environment state.  Isn't this nice and clean?
+    * The Radeon will automagically set the texture alpha to 0xff when
+    * the texture format does not include an alpha component.  This
+    * reduces the amount of special-casing we have to do, alpha-only
+    * textures being a notable exception.
+    */
+   switch ( texUnit->EnvMode ) {
+   case GL_REPLACE:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_REPLACE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
+	 break;
+      case GL_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_MODULATE:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_MODULATE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_DECAL:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+	 color_combine = radeon_color_combine[unit][RADEON_DECAL];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	 break;
+      case GL_ALPHA:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_BLEND:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_BLEND];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_ADD:
+      switch ( tObj->Image[0]->Format ) {
+      case GL_RGBA:
+      case GL_RGB:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_ADD];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_ALPHA:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_ADD];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_ADD];
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return;
+      }
+      break;
+
+   case GL_COMBINE_EXT:
+      /* Step 0:
+       * Calculate how many arguments we need to process.
+       */
+      switch ( texUnit->CombineModeRGB ) {
+      case GL_REPLACE:
+	 numColorArgs = 1;
+	 break;
+      case GL_MODULATE:
+      case GL_ADD:
+      case GL_ADD_SIGNED_EXT:
+      case GL_DOT3_RGB_EXT:
+      case GL_DOT3_RGBA_EXT:
+	 numColorArgs = 2;
+	 break;
+      case GL_INTERPOLATE_EXT:
+	 numColorArgs = 3;
+	 break;
+      default:
+	 return;
+      }
+
+      switch ( texUnit->CombineModeA ) {
+      case GL_REPLACE:
+	 numAlphaArgs = 1;
+	 break;
+      case GL_MODULATE:
+      case GL_ADD:
+      case GL_ADD_SIGNED_EXT:
+	 numAlphaArgs = 2;
+	 break;
+      case GL_INTERPOLATE_EXT:
+	 numAlphaArgs = 3;
+	 break;
+      default:
+	 return;
+      }
+
+      /* Step 1:
+       * Extract the color and alpha combine function arguments.
+       */
+      for ( i = 0 ; i < numColorArgs ; i++ ) {
+	 op = texUnit->CombineOperandRGB[i] - GL_SRC_COLOR;
+	 switch ( texUnit->CombineSourceRGB[i] ) {
+	 case GL_TEXTURE:
+	    color_arg[i] = radeon_texture_color[op][unit];
+	    break;
+	 case GL_CONSTANT_EXT:
+	    color_arg[i] = radeon_tfactor_color[op];
+	    break;
+	 case GL_PRIMARY_COLOR_EXT:
+	    color_arg[i] = radeon_primary_color[op];
+	    break;
+	 case GL_PREVIOUS_EXT:
+	    color_arg[i] = radeon_previous_color[op];
+	    break;
+	 default:
+	    return;
+	 }
+      }
+
+      for ( i = 0 ; i < numAlphaArgs ; i++ ) {
+	 op = texUnit->CombineOperandA[i] - GL_SRC_ALPHA;
+	 switch ( texUnit->CombineSourceA[i] ) {
+	 case GL_TEXTURE:
+	    alpha_arg[i] = radeon_texture_alpha[op][unit];
+	    break;
+	 case GL_CONSTANT_EXT:
+	    alpha_arg[i] = radeon_tfactor_alpha[op];
+	    break;
+	 case GL_PRIMARY_COLOR_EXT:
+	    alpha_arg[i] = radeon_primary_alpha[op];
+	    break;
+	 case GL_PREVIOUS_EXT:
+	    alpha_arg[i] = radeon_previous_alpha[op];
+	    break;
+	 default:
+	    return;
+	 }
+      }
+
+      /* Step 2:
+       * Build up the color and alpha combine functions.
+       */
+      switch ( texUnit->CombineModeRGB ) {
+      case GL_REPLACE:
+	 color_combine = (RADEON_COLOR_ARG_A_ZERO |
+			  RADEON_COLOR_ARG_B_ZERO |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, C );
+	 break;
+      case GL_MODULATE:
+	 color_combine = (RADEON_COLOR_ARG_C_ZERO |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, A );
+	 RADEON_COLOR_ARG( 1, B );
+	 break;
+      case GL_ADD:
+	 color_combine = (RADEON_COLOR_ARG_B_ZERO |
+			  RADEON_COMP_ARG_B |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, A );
+	 RADEON_COLOR_ARG( 1, C );
+	 break;
+      case GL_ADD_SIGNED_EXT:
+	 color_combine = (RADEON_COLOR_ARG_B_ZERO |
+			  RADEON_COMP_ARG_B |
+			  RADEON_BLEND_CTL_ADDSIGNED |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, A );
+	 RADEON_COLOR_ARG( 1, C );
+	 break;
+      case GL_INTERPOLATE_EXT:
+	 color_combine = (RADEON_BLEND_CTL_BLEND |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, B );
+	 RADEON_COLOR_ARG( 1, A );
+	 RADEON_COLOR_ARG( 2, C );
+	 break;
+      case GL_DOT3_RGB_EXT:
+      case GL_DOT3_RGBA_EXT:
+	 color_combine = (RADEON_COLOR_ARG_C_ZERO |
+			  RADEON_BLEND_CTL_DOT3 |
+			  RADEON_CLAMP_TX);
+	 RADEON_COLOR_ARG( 0, A );
+	 RADEON_COLOR_ARG( 1, B );
+	 break;
+      default:
+	 return;
+      }
+
+      switch ( texUnit->CombineModeA ) {
+      case GL_REPLACE:
+	 alpha_combine = (RADEON_ALPHA_ARG_A_ZERO |
+			  RADEON_ALPHA_ARG_B_ZERO |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, C );
+	 break;
+      case GL_MODULATE:
+	 alpha_combine = (RADEON_ALPHA_ARG_C_ZERO |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, A );
+	 RADEON_ALPHA_ARG( 1, B );
+	 break;
+      case GL_ADD:
+	 alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
+			  RADEON_COMP_ARG_B |
+			  RADEON_BLEND_CTL_ADD |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, A );
+	 RADEON_ALPHA_ARG( 1, C );
+	 break;
+      case GL_ADD_SIGNED_EXT:
+	 alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
+			  RADEON_COMP_ARG_B |
+			  RADEON_BLEND_CTL_ADDSIGNED |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, A );
+	 RADEON_ALPHA_ARG( 1, C );
+	 break;
+      case GL_INTERPOLATE_EXT:
+	 alpha_combine = (RADEON_BLEND_CTL_BLEND |
+			  RADEON_CLAMP_TX);
+	 RADEON_ALPHA_ARG( 0, B );
+	 RADEON_ALPHA_ARG( 1, A );
+	 RADEON_ALPHA_ARG( 2, C );
+	 break;
+      default:
+	 return;
+      }
+
+      if ( texUnit->CombineModeRGB == GL_DOT3_RGB_EXT ) {
+	 alpha_combine |= RADEON_DOT_ALPHA_DONT_REPLICATE;
+      }
+
+      /* Step 3:
+       * Apply the scale factor.  The EXT extension has a somewhat
+       * unnecessary restriction that the scale must be 4x.  The ARB
+       * extension will likely drop this and we can just apply the
+       * scale factors regardless.
+       */
+      if ( texUnit->CombineModeRGB != GL_DOT3_RGB_EXT &&
+	   texUnit->CombineModeRGB != GL_DOT3_RGBA_EXT ) {
+	 color_combine |= (texUnit->CombineScaleShiftRGB << 21);
+	 alpha_combine |= (texUnit->CombineScaleShiftA << 21);
+      } else {
+	 color_combine |= RADEON_SCALE_4X;
+	 alpha_combine |= RADEON_SCALE_4X;
+      }
+
+      /* All done!
+       */
+      break;
+
+   default:
+      return;
+   }
+
+   rmesa->color_combine[source] = color_combine;
+   rmesa->alpha_combine[source] = alpha_combine;
+}
+
+static void radeonUpdateTextureObject( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int source = rmesa->tmu_source[unit];
+   struct gl_texture_object *tObj;
+   radeonTexObjPtr t;
+   GLuint enabled;
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_MSG ) {
+      fprintf( stderr, "%s( %p, %d )\n",
+	       __FUNCTION__, ctx, unit );
+   }
+
+   enabled = (ctx->Texture.ReallyEnabled >> (source * 4)) & TEXTURE0_ANY;
+   if ( enabled != TEXTURE0_2D && enabled != TEXTURE0_1D ) {
+      if ( enabled )
+	 rmesa->Fallback |= RADEON_FALLBACK_TEXTURE;
+      return;
+   }
+
+   /* Only update the hardware texture state if the texture is current,
+    * complete and enabled.
+    */
+   tObj = ctx->Texture.Unit[source].Current;
+   if ( !tObj || !tObj->Complete )
+      return;
+
+   if ( ( tObj != ctx->Texture.Unit[source].CurrentD[2] ) &&
+	( tObj != ctx->Texture.Unit[source].CurrentD[1] ) )
+      return;
+
+   /* We definately have a valid texture now */
+   t = tObj->DriverData;
+
+   /* Force the texture unit state to be loaded into the hardware */
+   rmesa->dirty |= RADEON_UPLOAD_CONTEXT | (RADEON_UPLOAD_TEX0 << unit);
+
+   /* Force any texture images to be loaded into the hardware */
+   if ( t->dirty_images ) {
+      if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+	 fprintf( stderr, "   t->dirty_images = 0x%x\n", t->dirty_images );
+      }
+      radeonSetTexImages( rmesa, tObj );
+      rmesa->dirty |= (RADEON_UPLOAD_TEX0IMAGES << unit);
+   }
+
+   if ( t->memBlock )
+      radeonUpdateTexLRU( rmesa, t );
+
+   switch ( unit ) {
+   case 0:
+      rmesa->setup.pp_cntl |= (RADEON_TEX_0_ENABLE |
+			       RADEON_TEX_BLEND_0_ENABLE);
+      break;
+   case 1:
+      rmesa->setup.pp_cntl |= (RADEON_TEX_1_ENABLE |
+			       RADEON_TEX_BLEND_1_ENABLE);
+      break;
+   }
+}
+
+void radeonUpdateTextureState( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p ) en=0x%x\n",
+	       __FUNCTION__, ctx, ctx->Texture.ReallyEnabled );
+   }
+
+   /* Clear any texturing fallbacks */
+   rmesa->Fallback &= ~RADEON_FALLBACK_TEXTURE;
+
+   /* Disable all texturing until it is known to be good */
+   rmesa->setup.pp_cntl &= ~(RADEON_TEX_ENABLE_MASK |
+			     RADEON_TEX_BLEND_ENABLE_MASK);
+
+   radeonUpdateTextureObject( ctx, 0 );
+   radeonUpdateTextureEnv( ctx, 0 );
+
+   if ( rmesa->multitex ) {
+      radeonUpdateTextureObject( ctx, 1 );
+      radeonUpdateTextureEnv( ctx, 1 );
+   }
+
+   rmesa->dirty |= RADEON_UPLOAD_CONTEXT;
+}
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_tris.c b/xc/lib/GL/mesa/src/drv/radeon/radeon_tris.c
index 71ed6910b..e4b6394ee 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_tris.c
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_tris.c
@@ -55,6 +55,13 @@ do {									\
    *(GLuint *)(to) = *(GLuint *)(from);					\
 } while (0)
 
+#define RADEON_COLOR3( to, from )		\
+do {						\
+  (to)[0] = (from)[2];				\
+  (to)[1] = (from)[1];				\
+  (to)[2] = (from)[0];				\
+} while (0)
+
 
 static void radeon_null_quad( GLcontext *ctx, GLuint v0,
 			      GLuint v1, GLuint v2, GLuint v3, GLuint pv )
@@ -162,6 +169,10 @@ void radeonDDChooseRenderState( GLcontext *ctx )
 
    if ( rmesa->Fallback ) {
       rmesa->RenderIndex = RADEON_FALLBACK_BIT;
+      /* fixes vorder.c failure: */
+      if (flags & DD_TRI_LIGHT_TWOSIDE) {
+         rmesa->IndirectTriangles = DD_TRI_LIGHT_TWOSIDE;
+      }
       return;
    }
 
@@ -200,6 +211,11 @@ void radeonDDChooseRenderState( GLcontext *ctx )
 	 rmesa->IndirectTriangles |= (DD_TRI_SW_RASTERIZE |
 				      DD_QUAD_SW_RASTERIZE);
       }
+
+      /* fixes vorder.c failure: */
+      if (flags & DD_TRI_LIGHT_TWOSIDE) {
+         rmesa->IndirectTriangles |= DD_TRI_LIGHT_TWOSIDE;
+      }
    }
 
    if ( 0 ) {
diff --git a/xc/lib/GL/mesa/src/drv/radeon/radeon_tritmp.h b/xc/lib/GL/mesa/src/drv/radeon/radeon_tritmp.h
index ed5006ea2..a53512a30 100644
--- a/xc/lib/GL/mesa/src/drv/radeon/radeon_tritmp.h
+++ b/xc/lib/GL/mesa/src/drv/radeon/radeon_tritmp.h
@@ -247,7 +247,8 @@ static void TAG(line)( GLcontext *ctx,
    GLfloat z[2];
 #endif
 #if (IND & RADEON_TWOSIDE_BIT)
-   int c[2];
+   GLuint c[2];
+   GLuint s[2];
 #endif
 
    v[0] = &verts[e0];
@@ -256,17 +257,24 @@ static void TAG(line)( GLcontext *ctx,
 #if (IND & RADEON_TWOSIDE_BIT)
    c[0] = v[0]->ui[4];
    c[1] = v[1]->ui[4];
+   s[0] = v[0]->ui[5];
+   s[1] = v[1]->ui[5];
 #endif
 
 #if (IND & RADEON_TWOSIDE_BIT)
    {
       GLubyte (*vbcolor)[4] = ctx->VB->ColorPtr->data;
+      GLubyte (*vbspec)[4] = ctx->VB->Specular;
       if ( IND & RADEON_FLAT_BIT ) {
 	 RADEON_COLOR( (char *)&v[0]->ui[4], vbcolor[pv] );
 	 v[1]->ui[4] = v[0]->ui[4];
+	 RADEON_COLOR3( (char *)&v[0]->ui[5], vbspec[pv] );
+	 v[1]->ui[5] = v[0]->ui[5];
       } else {
 	 RADEON_COLOR( (char *)&v[0]->ui[4], vbcolor[e0] );
 	 RADEON_COLOR( (char *)&v[1]->ui[4], vbcolor[e1] );
+	 RADEON_COLOR3( (char *)&v[0]->ui[5], vbspec[e0] );
+	 RADEON_COLOR3( (char *)&v[1]->ui[5], vbspec[e1] );
       }
    }
 #endif
@@ -289,6 +297,8 @@ static void TAG(line)( GLcontext *ctx,
 #if (IND & RADEON_TWOSIDE_BIT)
    v[0]->ui[4] = c[0];
    v[1]->ui[4] = c[1];
+   v[0]->ui[5] = s[0];
+   v[1]->ui[5] = s[1];
 #endif
 }
 
@@ -309,7 +319,10 @@ static void TAG(points)( GLcontext *ctx,
 
 	    if ( IND & RADEON_TWOSIDE_BIT ) {
 	       GLubyte (*vbcolor)[4] = VB->ColorPtr->data;
+               GLubyte (*vbspec)[4] = VB->Specular;
 	       RADEON_COLOR( (char *)&tmp0.v.color, vbcolor[i] );
+               if (vbspec)
+                  RADEON_COLOR3( (char *)&tmp0.v.specular, vbspec[i] );
 	    }
 	    if ( IND & RADEON_OFFSET_BIT ) {
 	       GLfloat offset = ctx->PointZoffset * rmesa->depth_scale;
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/Imakefile b/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
index ebb353844..9f45df60d 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/tdfx/Imakefile
@@ -150,6 +150,7 @@ MESA_INCLUDES = -I. -I.. -I../../include
 		../../stages.c \
 		../../state.c \
 		../../stencil.c \
+		../../texformat.c \
 		../../teximage.c \
 		../../texobj.c \
 		../../texstate.c \
@@ -223,6 +224,7 @@ MESA_INCLUDES = -I. -I.. -I../../include
 		../../stages.o \
 		../../state.o \
 		../../stencil.o \
+		../../texformat.o \
 		../../teximage.o \
 		../../texobj.o \
 		../../texstate.o \
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.c
index 4cb7efaba..293c42fd7 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.c
@@ -43,9 +43,22 @@
 #include "tdfx_render.h"
 #include "tdfx_pipeline.h"
 #include "tdfx_span.h"
+#include "tdfx_tex.h"
 #include "tdfx_texman.h"
 #include "extensions.h"
 
+#ifndef TDFX_DEBUG
+int TDFX_DEBUG = (0
+/*  		  | DEBUG_ALWAYS_SYNC */
+/*		  | DEBUG_VERBOSE_API */
+/*		  | DEBUG_VERBOSE_MSG */
+/*		  | DEBUG_VERBOSE_LRU */
+/*  		  | DEBUG_VERBOSE_DRI */
+/*  		  | DEBUG_VERBOSE_IOCTL */
+/*   		  | DEBUG_VERBOSE_2D */
+/*   		  | DEBUG_VERBOSE_TEXTURE */
+   );
+#endif
 
 
 #if 0
@@ -75,15 +88,13 @@ static void tdfxDDInitExtensions( GLcontext *ctx )
    gl_extensions_disable( ctx, "GL_INGR_blend_func_separate" );
    gl_extensions_enable( ctx, "GL_HP_occlusion_test" );
 
-   if ( !fxMesa->haveTwoTMUs ) {
+   if ( fxMesa->numTMUs == 1 ) {
       gl_extensions_disable( ctx, "GL_EXT_texture_env_add" );
       gl_extensions_disable( ctx, "GL_ARB_multitexture" );
    }
 
    if ( TDFX_IS_NAPALM( fxMesa ) ) {
-      gl_extensions_enable( ctx, "GL_ARB_texture_compression" );
       gl_extensions_enable( ctx, "GL_EXT_texture_env_combine" );
-      gl_extensions_enable( ctx, "GL_3DFX_texture_compression_FXT1" );
    }
 
    if (fxMesa->haveHwStencil) {
@@ -178,16 +189,10 @@ GLboolean tdfxCreateContext( Display *dpy, GLvisual *mesaVis,
    fxMesa->Glide.Initialized = GL_FALSE;
    fxMesa->Glide.Board = 0;
 
-
-   if (getenv("FX_EMULATE_SINGLE_TMU")) {
-      fxMesa->haveTwoTMUs = GL_FALSE;
-   }
-   else {
-      if ( TDFX_IS_BANSHEE( fxMesa ) ) {
-         fxMesa->haveTwoTMUs = GL_FALSE;
-      } else {
-         fxMesa->haveTwoTMUs = GL_TRUE;
-      }
+   if ( getenv( "FX_EMULATE_SINGLE_TMU" ) || TDFX_IS_BANSHEE( fxMesa ) ) {
+      fxMesa->numTMUs = 1;
+   } else {
+      fxMesa->numTMUs = 2;
    }
 
    fxMesa->stats.swapBuffer = 0;
@@ -218,6 +223,7 @@ GLboolean tdfxCreateContext( Display *dpy, GLvisual *mesaVis,
    tdfxDDInitStateFuncs( ctx );
    tdfxDDInitRenderFuncs( ctx );
    tdfxDDInitSpanFuncs( ctx );
+   tdfxDDInitTextureFuncs( ctx );
 
    ctx->Driver.TriangleCaps = (DD_TRI_CULL |
 			       DD_TRI_LIGHT_TWOSIDE |
@@ -448,19 +454,21 @@ void tdfxDestroyContext( tdfxContextPtr fxMesa )
    }
 
    if ( fxMesa ) {
-      if (fxMesa->glCtx->Shared->RefCount == 1) {
+      GLcontext *ctx = fxMesa->glCtx;
+      struct gl_texture_object *tObj;
+
+      if ( ctx->Shared->RefCount == 1 ) {
          /* This share group is about to go away, free our private
           * texture object data.
           */
-         struct gl_texture_object *tObj;
-         tObj = fxMesa->glCtx->Shared->TexObjectList;
-         while (tObj) {
-            tdfxTMFreeTexture(fxMesa, tObj);
-            tObj = tObj->Next;
+	 LOCK_HARDWARE( fxMesa );
+         for ( tObj = ctx->Shared->TexObjectList ; tObj ; tObj = tObj->Next ) {
+            tdfxTMFreeTextureLocked( fxMesa, tObj );
          }
+	 UNLOCK_HARDWARE( fxMesa );
       }
 
-      tdfxTMClose(fxMesa);  /* free texture memory */
+      tdfxTMClose( fxMesa );  /* free texture memory */
       XFree( fxMesa );
    }
 
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.h
index d630184bd..a73347435 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.h
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_context.h
@@ -280,9 +280,7 @@ typedef struct {
    volatile int fifoOwner;
    volatile int ctxOwner;
    volatile int texOwner;
-}
-TDFXSAREAPriv;
-
+} TDFXSAREAPriv;
 
 typedef struct {
    GLuint swapBuffer;
@@ -292,73 +290,72 @@ typedef struct {
    GLuint texSwaps;
 } tdfxStats;
 
-
-
 /*
  *  Memory range from startAddr to endAddr-1
  */
 typedef struct mem_range {
    struct mem_range *next;
    FxU32 startAddr, endAddr;
-}
-tdfxMemRange;
+} tdfxMemRange;
 
+typedef struct {
+   GLvoid *data;
+   GLsizei width, height;
+   FxU32 size;
+} tdfxTexRawData;
 
 typedef struct {
-   GLsizei		width, height;	/* image size */
-   GLint		texelSize;	/* How many bytes to a texel */
-   GrTextureFormat_t	glideFormat;	/* Glide image format */
-   void			*data;		/* Glide-formated texture image */
-   FxU32		dataSize;	/* image size in bytes */
-}
-tdfxMipMapLevel;
+   tdfxTexRawData original;		/* Mesa-formatted texture image */
+   tdfxTexRawData rescaled;		/* Only needed if aspect ratio > 8:1 */
+
+   GLvoid *data;			/* Final version of texture image */
+   FxU32 size;				/* image size in bytes */
+
+   GrTextureFormat_t glideFormat;	/* Glide image format */
+   GLint wScale, hScale;		/* Broken hardware... */
+} tdfxTexImage, *tdfxTexImagePtr;
 
 
 #define TDFX_NUM_TMU		2
 
 
-typedef struct tdfxTexInfo_t
-{
+typedef struct {
    GLboolean isInTM;
    GLboolean reloadImages;  /* if true, resend images to Glide */
    GLuint lastTimeUsed;
    FxU32 whichTMU;
 
    GrTexInfo info;
-   GrAspectRatio_t aspectRatio;
-   tdfxMipMapLevel mipmapLevel[MAX_TEXTURE_LEVELS];
-   tdfxMemRange *tm[TDFX_NUM_TMU];
+   tdfxTexImage image[MAX_TEXTURE_LEVELS];
+   tdfxMemRange *range[TDFX_NUM_TMU];
 
    GLint minLevel, maxLevel;
+   GrMipMapMode_t mmMode;
+   GrAspectRatio_t aspectRatio;
+   FxBool LODblend;
    GrTextureFilterMode_t minFilt;
    GrTextureFilterMode_t magFilt;
    GrTextureClampMode_t sClamp;
    GrTextureClampMode_t tClamp;
-   FxBool LODblend;
-   GrMipMapMode_t mmMode;
 
-   GLfloat sScale, tScale;  /* texcoord scale factor */
+   GLfloat sScale, tScale;		/* texcoord scale factor */
 
    GuTexPalette palette;
-}
-tdfxTexInfo;
+} tdfxTexObj, *tdfxTexObjPtr;
 
+#define TDFX_TEXTURE_DATA(tObj)		((tdfxTexObjPtr)((tObj)->DriverData))
 
-#define TDFX_TEXTURE_DATA(mesaObj) ((tdfxTexInfo *)((mesaObj)->DriverData))
 
-
-
-/*
- * This is state which may be shared by several tdfx contexts.
+/* This is state which may be shared by several tdfx contexts.
  * It hangs off of Mesa's gl_shared_state object (ctx->Shared->DriverData).
  */
-struct tdfxSharedState {
+typedef struct tdfx_shared_state {
    GLboolean umaTexMemory;
-   GLuint totalTexMem[TDFX_NUM_TMU]; /* constant */
-   GLuint freeTexMem[TDFX_NUM_TMU]; /* changes as we go */
-   tdfxMemRange *tmPool;
-   tdfxMemRange *tmFree[TDFX_NUM_TMU];
-};
+   GLuint totalTexMem[TDFX_NUM_TMU];	/* constant */
+   GLuint freeTexMem[TDFX_NUM_TMU];	/* changes as we go */
+   tdfxMemRange *rangePool;
+   tdfxMemRange *freeRanges[TDFX_NUM_TMU];
+} tdfxSharedState, *tdfxSharedStatePtr;
 
 
 
@@ -516,6 +513,11 @@ struct tdfx_depth {
    FxBool Mask;				/* Write enable flag */
 };
 
+#ifndef GR_STIPPLE_PATTERN
+#error "You MUST upgrade your Glide3 libraries and headers."
+#error "Get the latest from http://dri.sourceforge.net/resources.html"
+#endif
+
 struct tdfx_stipple {
    GrStippleMode_t Mode;		/* Stipple enable/disable */
    FxU32 Pattern;			/* 8x4 Stipple Pattern */
@@ -630,6 +632,7 @@ struct tdfx_context {
 
    GLuint tmu_source[TDFX_NUM_TMU];
    GLuint tex_dest[MAX_TEXTURE_UNITS];
+   GLuint numTMUs;
 
    GLuint SetupIndex;
    GLuint SetupDone;
@@ -641,10 +644,6 @@ struct tdfx_context {
    GLfloat sScale0, tScale0;
    GLfloat sScale1, tScale1;
 
-#if 0
-   GLuint last_tri_caps;
-   GLuint stw_hint_state;      /* for grHints */
-#endif
    GLuint using_fast_path, passes, multipass;
    GLuint texBindNumber;
    GLint tmuSrc;
@@ -652,13 +651,6 @@ struct tdfx_context {
    int screen_width;
    int screen_height;
 
-#if 0
-   void *state;                 /* Glide state buffer */
-
-   GLint textureAlign;
-   GLboolean verbose;
-#endif
-   GLboolean haveTwoTMUs;      /* True if we have 2 tmu's  */
    GLboolean haveHwStencil;
 
    GLint maxPendingSwapBuffers;
@@ -768,6 +760,7 @@ extern int TDFX_DEBUG;
 #define DEBUG_VERBOSE_DRI	0x10
 #define DEBUG_VERBOSE_IOCTL	0x20
 #define DEBUG_VERBOSE_2D	0x40
+#define DEBUG_VERBOSE_TEXTURE	0x80
 
 #endif /* GLX_DIRECT_RENDERING */
 
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_dd.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_dd.c
index ec5b25871..3f76fb88d 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_dd.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_dd.c
@@ -43,9 +43,11 @@
 
 #include "enums.h"
 #include "pb.h"
+#if defined(USE_X86_ASM) || defined(USE_3DNOW_ASM) || defined(USE_KATMAI_ASM)
+#include "X86/common_x86_asm.h"
+#endif
 
-
-#define TDFX_DATE	"20010104"
+#define TDFX_DATE	"20010305"
 
 
 /* These are used in calls to FX_grColorMaskv() */
@@ -63,10 +65,12 @@ static const GLubyte *tdfxDDGetString( GLcontext *ctx, GLenum name )
    tdfxContextPtr fxMesa = (tdfxContextPtr) ctx->DriverCtx;
 
    switch ( name ) {
-   case GL_RENDERER:
-   {
-      static char buffer[100];
-      char hardware[100];
+   case GL_VENDOR:
+      return "VA Linux Systems, Inc.";
+
+   case GL_RENDERER: {
+      static char buffer[128];
+      char hardware[128];
 
       strcpy( hardware, FX_grGetString( fxMesa, GR_HARDWARE ) );
 
@@ -83,19 +87,41 @@ static const GLubyte *tdfxDDGetString( GLcontext *ctx, GLenum name )
 	 strcpy( hardware, "Voodoo5" );
       }
       else {
-	 /* unexpected result: replace spaces with hyphens */
+	 /* Unexpected result: replace spaces with hyphens */
 	 int i;
 	 for ( i = 0 ; hardware[i] ; i++ ) {
 	    if ( hardware[i] == ' ' || hardware[i] == '\t' )
 	       hardware[i] = '-';
 	 }
       }
-      /* now make the GL_RENDERER string */
+      /* Now make the GL_RENDERER string */
       sprintf( buffer, "Mesa DRI %s " TDFX_DATE, hardware );
+
+      /* Append any CPU-specific information.
+       */
+#ifdef USE_X86_ASM
+      if ( gl_x86_cpu_features ) {
+	 strncat( buffer, " x86", 4 );
+      }
+#endif
+#ifdef USE_MMX_ASM
+      if ( cpu_has_mmx ) {
+	 strncat( buffer, "/MMX", 4 );
+      }
+#endif
+#ifdef USE_3DNOW_ASM
+      if ( cpu_has_3dnow ) {
+	 strncat( buffer, "/3DNow!", 7 );
+      }
+#endif
+#ifdef USE_KATMAI_ASM
+      if ( cpu_has_xmm ) {
+	 strncat( buffer, "/SSE", 4 );
+      }
+#endif
       return buffer;
    }
-   case GL_VENDOR:
-      return "VA Linux Systems, Inc.";
+
    default:
       return NULL;
    }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_render.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_render.c
index 97931a351..da60bee16 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_render.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_render.c
@@ -594,16 +594,13 @@ static void uploadTextureImages( tdfxContextPtr fxMesa )
 {
    GLcontext *ctx = fxMesa->glCtx;
    int unit;
-   for (unit = 0; unit < TDFX_NUM_TMU; unit++) {
-      if (ctx->Texture.Unit[unit].ReallyEnabled == TEXTURE0_2D) {
+   for ( unit = 0 ; unit < TDFX_NUM_TMU ; unit++ ) {
+      if ( ctx->Texture.Unit[unit].ReallyEnabled == TEXTURE0_2D ) {
          struct gl_texture_object *tObj = ctx->Texture.Unit[unit].CurrentD[2];
-         tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-         if (ti && ti->reloadImages) {
-            /*
-            printf("download texture image on unit %d\n", unit);
-            */
-            tdfxTMDownloadTexture(fxMesa, tObj);
-            ti->reloadImages = GL_FALSE;
+         tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+         if ( t && t->reloadImages ) {
+            tdfxTMDownloadTextureLocked( fxMesa, tObj );
+            t->reloadImages = GL_FALSE;
          }
       }
    }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_state.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_state.c
index eb9a6a98b..d630fbd8c 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_state.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_state.c
@@ -45,7 +45,6 @@
 #include "tdfx_vb.h"
 #include "tdfx_tex.h"
 #include "tdfx_texman.h"
-#include "tdfx_texstate.h"
 #include "tdfx_tris.h"
 #include "tdfx_render.h"
 
@@ -1535,15 +1534,6 @@ void tdfxDDInitStateFuncs( GLcontext *ctx )
    ctx->Driver.Scissor			= tdfxDDScissor;
    ctx->Driver.ShadeModel		= tdfxDDShadeModel;
 
-   ctx->Driver.BindTexture		= tdfxDDBindTexture;
-   ctx->Driver.DeleteTexture		= tdfxDDDeleteTexture;
-   ctx->Driver.TexEnv			= tdfxDDTexEnv;
-   ctx->Driver.TexParameter		= tdfxDDTexParameter;
-   ctx->Driver.TexImage2D		= tdfxDDTexImage2D;
-   ctx->Driver.TexSubImage2D		= tdfxDDTexSubImage2D;
-   ctx->Driver.GetTexImage		= tdfxDDGetTexImage;
-   ctx->Driver.UpdateTexturePalette	= tdfxDDTexturePalette;
-
    if ( fxMesa->haveHwStencil ) {
       ctx->Driver.StencilFunc		= tdfxDDStencilFunc;
       ctx->Driver.StencilMask		= tdfxDDStencilMask;
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.c
index c572e22a8..e0f2d2eb0 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.c
@@ -35,39 +35,37 @@
  *
  */
 
-#include "image.h"
-#include "texutil.h"
 #include "tdfx_context.h"
 #include "tdfx_tex.h"
 #include "tdfx_texman.h"
 
+#include "enums.h"
+#include "image.h"
+#include "texutil.h"
+
+#define TX_DITHER_NONE		0x00000000
 
 
-static int
-logbase2(int n)
+static int logbase2( int n )
 {
-    GLint i = 1;
-    GLint log2 = 0;
-
-    if (n < 0) {
-        return -1;
-    }
-
-    while (n > i) {
-        i *= 2;
-        log2++;
-    }
-    if (i != n) {
-        return -1;
-    }
-    else {
-        return log2;
-    }
+   GLint i = 1;
+   GLint log2 = 0;
+
+   if ( n < 0 )
+      return -1;
+
+   while ( n > i ) {
+      i *= 2;
+      log2++;
+   }
+   if ( i != n ) {
+      return -1;
+   } else {
+      return log2;
+   }
 }
 
-
-/*
- * Compute various texture image parameters.
+/* Compute various texture image parameters.
  * Input:  w, h - source texture width and height
  * Output:  lodlevel - Glide lod level token for the larger texture dimension
  *          aspectratio - Glide aspect ratio token
@@ -84,1520 +82,977 @@ logbase2(int n)
  *      32   64  GR_LOD_LOG2_64 (=6)   GR_ASPECT_LOG2_1x2 (=-1)
  *      32   32  GR_LOD_LOG2_32 (=5)   GR_ASPECT_LOG2_1x1 (=0)
  */
-static void
-tdfxTexGetInfo(const GLcontext *ctx, int w, int h,
-               GrLOD_t *lodlevel, GrAspectRatio_t *aspectratio,
-               float *sscale, float *tscale,
-               int *wscale, int *hscale)
+static void tdfxTexGetInfo( const GLcontext *ctx, int w, int h,
+			    GrLOD_t *lodlevel, GrAspectRatio_t *aspectratio,
+			    float *sscale, float *tscale,
+			    int *wscale, int *hscale )
 {
-    int logw, logh, ar, lod, ws, hs;
-    float s, t;
-
-    ASSERT(w >= 1);
-    ASSERT(h >= 1);
-
-    logw = logbase2(w);
-    logh = logbase2(h);
-    ar = logw - logh;  /* aspect ratio = difference in log dimensions */
-
-    /* Hardware only allows a maximum aspect ratio of 8x1, so handle
-       |ar| > 3 by scaling the image and using an 8x1 aspect ratio */
-    if (ar >= 0) {
-        ASSERT(width >= height);
-        lod = logw;
-        s = 256.0;
-        ws = 1;
-        if (ar <= GR_ASPECT_LOG2_8x1) {
-            t = 256 >> ar;
-            hs = 1;
-        }
-        else {
-            /* have to stretch image height */
-            t = 32.0;
-            hs = 1 << (ar - 3);
-        }
-    }
-    else {
-        ASSERT(width < height);
-        lod = logh;
-        t = 256.0;
-        hs = 1;
-        if (ar >= GR_ASPECT_LOG2_1x8) {
-            s = 256 >> -ar;
-            ws = 1;
-        }
-        else {
-            /* have to stretch image width */
-            s = 32.0;
-            ws = 1 << (-ar - 3);
-        }
-    }
-
-    if (ar < GR_ASPECT_LOG2_1x8)
-        ar = GR_ASPECT_LOG2_1x8;
-    else if (ar > GR_ASPECT_LOG2_8x1)
-        ar = GR_ASPECT_LOG2_8x1;
-
-    if (lodlevel)
-        *lodlevel = (GrLOD_t) lod;
-    if (aspectratio)
-        *aspectratio = (GrAspectRatio_t) ar;
-    if (sscale)
-        *sscale = s;
-    if (tscale)
-        *tscale = t;
-    if (wscale)
-        *wscale = ws;
-    if (hscale)
-        *hscale = hs;
+   int logw, logh, ar, lod, ws, hs;
+   float s, t;
+
+   ASSERT( w >= 1 );
+   ASSERT( h >= 1 );
+
+   logw = logbase2( w );
+   logh = logbase2( h );
+   ar = logw - logh;  /* aspect ratio = difference in log dimensions */
+
+   /* Hardware only allows a maximum aspect ratio of 8x1, so handle
+    * |ar| > 3 by scaling the image and using an 8x1 aspect ratio.
+    */
+   if ( ar >= 0 ) {
+      ASSERT( width >= height );
+      lod = logw;
+      s = 256.0;
+      ws = 1;
+      if ( ar <= GR_ASPECT_LOG2_8x1 ) {
+	 t = 256 >> ar;
+	 hs = 1;
+      } else {
+	 /* have to stretch image height */
+	 t = 32.0;
+	 hs = 1 << (ar - 3);
+      }
+   } else {
+      ASSERT( width < height );
+      lod = logh;
+      t = 256.0;
+      hs = 1;
+      if ( ar >= GR_ASPECT_LOG2_1x8 ) {
+	 s = 256 >> -ar;
+	 ws = 1;
+      } else {
+	 /* have to stretch image width */
+	 s = 32.0;
+	 ws = 1 << (-ar - 3);
+      }
+   }
+
+   if ( ar < GR_ASPECT_LOG2_1x8 ) {
+      ar = GR_ASPECT_LOG2_1x8;
+   } else if ( ar > GR_ASPECT_LOG2_8x1 ) {
+      ar = GR_ASPECT_LOG2_8x1;
+   }
+
+   if ( lodlevel )
+      *lodlevel = (GrLOD_t)lod;
+   if ( aspectratio )
+      *aspectratio = (GrAspectRatio_t)ar;
+   if ( sscale )
+      *sscale = s;
+   if ( tscale )
+      *tscale = t;
+   if ( wscale )
+      *wscale = ws;
+   if ( hscale )
+      *hscale = hs;
 }
 
 
-/*
- * We need to call this when a texture object's minification filter
+/* We need to call this when a texture object's minification filter
  * or texture image sizes change.
  */
-static void RevalidateTexture(GLcontext *ctx, struct gl_texture_object *tObj)
+static void tdfxRevalidateTexture( GLcontext *ctx,
+				   struct gl_texture_object *tObj )
 {
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    GLint minl, maxl;
-
-    if (!ti)
-       return;
-
-    minl = maxl = tObj->BaseLevel;
-
-    if (tObj->Image[minl]) {
-       maxl = MIN2(tObj->MaxLevel, tObj->Image[minl]->MaxLog2);
-
-       /* compute largeLodLog2, aspect ratio and texcoord scale factors */
-       tdfxTexGetInfo(ctx, tObj->Image[minl]->Width, tObj->Image[minl]->Height,
-                      &ti->info.largeLodLog2,
-                      &ti->info.aspectRatioLog2,
-                      &(ti->sScale), &(ti->tScale), NULL, NULL);
-    }
-
-    if (tObj->Image[maxl] && (tObj->MinFilter != GL_NEAREST) && (tObj->MinFilter != GL_LINEAR)) {
-        /* mipmapping: need to compute smallLodLog2 */
-        tdfxTexGetInfo(ctx, tObj->Image[maxl]->Width,
-                       tObj->Image[maxl]->Height,
-                       &ti->info.smallLodLog2, NULL,
-                       NULL, NULL, NULL, NULL);
-    }
-    else {
-        /* not mipmapping: smallLodLog2 = largeLodLog2 */
-        ti->info.smallLodLog2 = ti->info.largeLodLog2;
-    }
-
-    ti->minLevel = minl;
-    ti->maxLevel = maxl;
-    ti->info.data = NULL;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+   GLint minl, maxl;
+
+   if ( !t )
+      return;
+
+   minl = maxl = tObj->BaseLevel;
+
+   if ( tObj->Image[minl] ) {
+      maxl = MIN2( tObj->MaxLevel, tObj->Image[minl]->MaxLog2 );
+
+      /* Compute largeLodLog2, aspect ratio and texcoord scale factors.
+       */
+      tdfxTexGetInfo( ctx,
+		      tObj->Image[minl]->Width, tObj->Image[minl]->Height,
+		      &t->info.largeLodLog2, &t->info.aspectRatioLog2,
+		      &t->sScale, &t->tScale, NULL, NULL );
+   }
+
+   if ( tObj->Image[maxl] &&
+	tObj->MinFilter != GL_NEAREST  &&
+	tObj->MinFilter != GL_LINEAR ) {
+      /* Mipmapping: need to compute smallLodLog2 */
+      tdfxTexGetInfo( ctx,
+		      tObj->Image[maxl]->Width, tObj->Image[maxl]->Height,
+		      &t->info.smallLodLog2,
+		      NULL, NULL, NULL, NULL, NULL );
+   } else {
+      /* Not mipmapping: smallLodLog2 = largeLodLog2 */
+      t->info.smallLodLog2 = t->info.largeLodLog2;
+   }
+
+   t->minLevel = minl;
+   t->maxLevel = maxl;
+   t->info.data = NULL;
 }
 
 
-static tdfxTexInfo *
-fxAllocTexObjData(tdfxContextPtr fxMesa)
+static tdfxTexObjPtr tdfxAllocTexObj( tdfxContextPtr fxMesa )
 {
-    tdfxTexInfo *ti;
-    int i;
+   tdfxTexObjPtr t;
+   int i;
 
-    if (!(ti = CALLOC(sizeof(tdfxTexInfo)))) {
-        gl_problem(NULL, "tdfx driver: out of memory");
-        return NULL;
-    }
+   t = CALLOC( sizeof(tdfxTexObj) );
+   if ( !t ) {
+      gl_problem( NULL, "tdfx driver: out of memory" );
+      return NULL;
+   }
 
-    ti->isInTM = GL_FALSE;
+   t->isInTM = GL_FALSE;
 
-    ti->whichTMU = TDFX_TMU_NONE;
+   t->whichTMU = TDFX_TMU_NONE;
 
-    ti->tm[TDFX_TMU0] = NULL;
-    ti->tm[TDFX_TMU1] = NULL;
+   t->range[TDFX_TMU0] = NULL;
+   t->range[TDFX_TMU1] = NULL;
 
-    ti->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-    ti->magFilt = GR_TEXTUREFILTER_BILINEAR;
+   t->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+   t->magFilt = GR_TEXTUREFILTER_BILINEAR;
 
-    ti->sClamp = GR_TEXTURECLAMP_WRAP;
-    ti->tClamp = GR_TEXTURECLAMP_WRAP;
+   t->sClamp = GR_TEXTURECLAMP_WRAP;
+   t->tClamp = GR_TEXTURECLAMP_WRAP;
 
-    ti->mmMode = GR_MIPMAP_NEAREST;
-    ti->LODblend = FXFALSE;
+   t->mmMode = GR_MIPMAP_NEAREST;
+   t->LODblend = FXFALSE;
 
-    for (i = 0; i < MAX_TEXTURE_LEVELS; i++) {
-        ti->mipmapLevel[i].data = NULL;
-    }
+   for ( i = 0 ; i < MAX_TEXTURE_LEVELS ; i++ ) {
+      t->image[i].original.data = NULL;
+      t->image[i].rescaled.data = NULL;
+   }
 
-    return ti;
+   return t;
 }
 
 
-/*
- * Called via glBindTexture.
+/* Given an OpenGL internal texture format, return the corresponding
+ * Glide internal texture format and MesaIntTexFormat.
+ * If allow32bpp is true, we'll return 32-bit texel formats when
+ * appropriate.
  */
-
-void
-tdfxDDBindTexture(GLcontext * ctx, GLenum target,
-                  struct gl_texture_object *tObj)
+static GrTextureFormat_t
+tdfxTexGetFormat( tdfxContextPtr fxMesa, struct gl_texture_image *texImage,
+		  GLenum format, GLenum type )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTexInfo *ti;
+   const GLboolean allow32bpp = TDFX_IS_NAPALM(fxMesa);
+   const GLboolean is32bpp = ( fxMesa->fxScreen->cpp == 4 );
+   const struct gl_texture_format *texFormat;
+   GrTextureFormat_t ret;
+
+   if ( 0 )
+      fprintf( stderr, "internal=%s format=%s type=%s\n",
+	       texImage->IntFormat == 3 ? "GL_RGB (3)" :
+	       texImage->IntFormat == 4 ? "GL_RGBA (4)" :
+	       gl_lookup_enum_by_nr( texImage->IntFormat ),
+	       gl_lookup_enum_by_nr( format ),
+	       gl_lookup_enum_by_nr( type ) );
+
+#define SET_FORMAT( gr, gl )						\
+   do {									\
+      ret = (gr);							\
+      texFormat = &(gl);						\
+   } while (0)
+
+#define SET_FORMAT_32BPP( gr32, gl32, gr16, gl16 )			\
+   do {									\
+      if ( allow32bpp ) {						\
+	 ret = (gr32);							\
+	 texFormat = &(gl32);						\
+      } else {								\
+	 ret = (gr16);							\
+	 texFormat = &(gl16);						\
+      }									\
+   } while (0)
+
+   switch ( texImage->IntFormat ) {
+      /* GH: Bias towards GL_RGB, GL_RGBA texture formats.  This has
+       * got to be better than sticking them way down the end of this
+       * huge list.
+       */
+   case GL_RGBA:
+   case 4:
+      if ( format == GL_BGRA ) {
+	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV && allow32bpp ) {
+	    SET_FORMAT( GR_TEXFMT_ARGB_8888, _mesa_texformat_argb8888 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
+	    SET_FORMAT( GR_TEXFMT_ARGB_4444, _mesa_texformat_argb4444 );
+	    break;
+	 } else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
+	    SET_FORMAT( GR_TEXFMT_ARGB_1555, _mesa_texformat_argb1555 );
+	    break;
+	 }
+      }
+      if ( allow32bpp && is32bpp ) {
+	 SET_FORMAT( GR_TEXFMT_ARGB_8888, _mesa_texformat_rgba8888 );
+      } else {
+	 SET_FORMAT( GR_TEXFMT_ARGB_4444, _mesa_texformat_argb4444 );
+      }
+      break;
+
+   case GL_RGB:
+   case 3:
+      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
+	 SET_FORMAT( GR_TEXFMT_RGB_565, _mesa_texformat_rgb565 );
+	 break;
+      }
+      if ( allow32bpp && is32bpp ) {
+	 SET_FORMAT( GR_TEXFMT_ARGB_8888, _mesa_texformat_rgba8888 );
+      } else {
+	 SET_FORMAT( GR_TEXFMT_RGB_565, _mesa_texformat_rgb565 );
+      }
+      break;
+
+      /* GH: Okay, keep checking as normal.  Still test for GL_RGB,
+       * GL_RGBA formats first.
+       */
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      SET_FORMAT_32BPP( GR_TEXFMT_ARGB_8888, _mesa_texformat_argb8888,
+			GR_TEXFMT_ARGB_4444, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGBA4:
+   case GL_RGBA2:
+      SET_FORMAT( GR_TEXFMT_ARGB_4444, _mesa_texformat_argb4444 );
+      break;
+
+   case GL_RGB5_A1:
+      SET_FORMAT( GR_TEXFMT_ARGB_1555, _mesa_texformat_argb1555 );
+      break;
+
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      SET_FORMAT_32BPP( GR_TEXFMT_ARGB_8888, _mesa_texformat_argb8888,
+			GR_TEXFMT_RGB_565,   _mesa_texformat_rgb565 );
+      break;
+
+   case GL_RGB5:
+   case GL_RGB4:
+   case GL_R3_G3_B2:
+      SET_FORMAT( GR_TEXFMT_RGB_565, _mesa_texformat_rgb565 );
+      break;
+
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+      SET_FORMAT( GR_TEXFMT_ALPHA_8, _mesa_texformat_a8 );
+      break;
+
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+      SET_FORMAT( GR_TEXFMT_INTENSITY_8, _mesa_texformat_l8 );
+      break;
+
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+      SET_FORMAT( GR_TEXFMT_ALPHA_INTENSITY_88, _mesa_texformat_al88 );
+      break;
+
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+      SET_FORMAT( GR_TEXFMT_ALPHA_8, _mesa_texformat_i8 );
+      break;
+
+   case GL_COLOR_INDEX:
+   case GL_COLOR_INDEX1_EXT:
+   case GL_COLOR_INDEX2_EXT:
+   case GL_COLOR_INDEX4_EXT:
+   case GL_COLOR_INDEX8_EXT:
+   case GL_COLOR_INDEX12_EXT:
+   case GL_COLOR_INDEX16_EXT:
+      SET_FORMAT( GR_TEXFMT_P_8, _mesa_texformat_ci8 );
+      break;
+
+   default:
+      fprintf( stderr, "bad texture format in fxTexGetFormat() %d",
+	       texImage->IntFormat );
+      return -1;
+   }
+
+   texImage->TexFormat = texFormat;
+
+   return ret;
+}
 
-    if (MESA_VERBOSE & VERBOSE_DRIVER) {
-        fprintf(stderr, "fxmesa: fxDDTexBind(%d,%p)\n", tObj->Name,
-                tObj->DriverData);
-    }
 
-    if (target != GL_TEXTURE_2D)
-        return;
+static GLboolean
+tdfxDDTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+		  GLenum format, GLenum type, const GLvoid *pixels,
+		  const struct gl_pixelstore_attrib *packing,
+		  struct gl_texture_object *texObj,
+		  struct gl_texture_image *texImage,
+		  GLboolean *retainInternalCopy )
+{
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   const struct gl_texture_format *texFormat;
+   GrTextureFormat_t glideFormat;
+   tdfxTexObjPtr t;
+   tdfxTexImagePtr image;
+   GLint dstWidth, dstHeight, wScale, hScale;
+   GLint size;
+   void *data;
+
+   if ( 0 ) {
+      printf("TexImage id=%d int 0x%x  format 0x%x  type 0x%x  %dx%d\n",
+	     texObj->Name, texImage->IntFormat, format, type,
+	     texImage->Width, texImage->Height);
+   }
+
+   if ( target != GL_TEXTURE_2D || texImage->Border > 0 )
+      return GL_FALSE;
+
+   if ( !texObj->DriverData )
+      texObj->DriverData = tdfxAllocTexObj( fxMesa );
+
+   t = TDFX_TEXTURE_DATA(texObj);
+   image = &t->image[level];
+
+   /* Determine the appropriate GL internal texel format, Mesa internal
+    * texel format, and texelSize (bytes) given the user's internal
+    * texture format hint.
+    */
+   glideFormat = tdfxTexGetFormat( fxMesa, texImage, format, type );
 
-    if (!tObj->DriverData) {
-        tObj->DriverData = fxAllocTexObjData(fxMesa);
-    }
+   /* Get the destination internal format.
+    */
+   texFormat = texImage->TexFormat;
 
-    ti = TDFX_TEXTURE_DATA(tObj);
-    ti->lastTimeUsed = fxMesa->texBindNumber++;
+   /* Determine width and height scale factors for texture.  Remember,
+    * Glide is limited to 8:1 aspect ratios.
+    */
+   tdfxTexGetInfo( ctx,
+		   texImage->Width, texImage->Height,
+		   NULL, NULL, NULL, NULL,
+		   &wScale, &hScale );
+   dstWidth = texImage->Width * wScale;
+   dstHeight = texImage->Height * hScale;
+
+   /* Allocate new storage for texture image, if needed.  This
+    * conditional wants to set uncompressedImage to point to the
+    * uncompressed image, and mml->data to the texture data.  If the
+    * image is uncompressed, these are identical.  If the image is not
+    * compressed, these are different.
+    */
+   if ( !image->original.data || image->glideFormat != glideFormat ||
+	image->original.width != texImage->Width ||
+	image->original.height != texImage->Height )
+   {
+      if ( image->original.data ) {
+	 FREE( image->original.data );
+	 image->original.data = NULL;
+      }
+      if ( image->rescaled.data ) {
+	 FREE( image->rescaled.data );
+	 image->rescaled.data = NULL;
+      }
+
+      size = texImage->Width * texImage->Height * texFormat->TexelBytes;
+      image->original.data = (void *) MALLOC( size );
+      if ( !image->original.data )
+	 return GL_FALSE;
+
+      image->original.width = texImage->Width;
+      image->original.height = texImage->Height;
+      image->original.size = size;
+
+      image->glideFormat = glideFormat;
+      image->wScale = wScale;
+      image->hScale = hScale;
+
+      t->info.format = glideFormat;
+      tdfxTMMoveOutTM( fxMesa, texObj );
+   }
+
+   /* Store the texture image into the 'original' space.
+    */
+   if ( !_mesa_convert_texsubimage2d( texFormat->IntFormat,
+				      0, 0, texImage->Width,
+				      texImage->Height, texImage->Width,
+				      format, type, packing, pixels,
+				      image->original.data ) ) {
+      return GL_FALSE;
+   }
+
+   data = image->original.data;
+   size = image->original.size;
+
+   /* GH: Sigh...
+    */
+   if ( wScale > 1 || hScale > 1 ) {
+      if ( image->rescaled.data ) {
+	 FREE( image->rescaled.data );
+	 image->rescaled.data = NULL;
+      }
 
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-}
+      size = dstWidth * dstHeight * texFormat->TexelBytes;
+      image->rescaled.data = (void *) MALLOC( size );
+      if ( !image->rescaled.data )
+	 return GL_FALSE;
 
+      image->rescaled.width = dstWidth;
+      image->rescaled.height = dstHeight;
+      image->rescaled.size = size;
 
-/*
- * Called via glTexEnv.
- */
-void
-tdfxDDTexEnv(GLcontext * ctx, GLenum target, GLenum pname,
-             const GLfloat * param)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-
-    if ( TDFX_DEBUG & DEBUG_VERBOSE_API ) {
-        if (param)
-            fprintf(stderr, "fxmesa: texenv(%x,%x)\n", pname,
-                    (GLint) (*param));
-        else
-            fprintf(stderr, "fxmesa: texenv(%x)\n", pname);
-    }
+      _mesa_rescale_teximage2d( texFormat,
+				texImage->Width, texImage->Height,
+				dstWidth, dstHeight,
+				image->original.data, image->rescaled.data );
 
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-}
+      data = image->rescaled.data;
+   }
 
+   image->data = data;
+   image->size = size;
 
-/*
- * Called via glTexParameter.
- */
-void
-tdfxDDTexParameter(GLcontext * ctx, GLenum target,
-                   struct gl_texture_object *tObj,
-                   GLenum pname, const GLfloat * params)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    GLenum param = (GLenum) (GLint) params[0];
-    tdfxTexInfo *ti;
-
-    if (MESA_VERBOSE & VERBOSE_DRIVER) {
-        fprintf(stderr, "fxmesa: fxDDTexParam(%d,%p,%x,%x)\n", tObj->Name,
-                tObj->DriverData, pname, param);
-    }
-
-    if (target != GL_TEXTURE_2D)
-        return;
-
-    if (!tObj->DriverData)
-        tObj->DriverData = fxAllocTexObjData(fxMesa);
-
-    ti = TDFX_TEXTURE_DATA(tObj);
-
-    switch (pname) {
-    case GL_TEXTURE_MIN_FILTER:
-        switch (param) {
-        case GL_NEAREST:
-            ti->mmMode = GR_MIPMAP_DISABLE;
-            ti->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-            ti->LODblend = FXFALSE;
-            break;
-        case GL_LINEAR:
-            ti->mmMode = GR_MIPMAP_DISABLE;
-            ti->minFilt = GR_TEXTUREFILTER_BILINEAR;
-            ti->LODblend = FXFALSE;
-            break;
-        case GL_NEAREST_MIPMAP_LINEAR:
-            if (TDFX_IS_NAPALM(fxMesa)) {
-                 if (fxMesa->haveTwoTMUs) {
-                     ti->mmMode = GR_MIPMAP_NEAREST;
-                     ti->LODblend = FXTRUE;
-                 }
-                 else {
-                     ti->mmMode = GR_MIPMAP_NEAREST_DITHER;
-                     ti->LODblend = FXFALSE;
-                 }
-                 ti->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-                 break;
-            }
-            /* XXX Voodoo3/Banshee mipmap blending seems to produce
-             * incorrectly filtered colors for the smallest mipmap levels.
-             * To work-around we fall-through here and use a different filter.
-             */
-        case GL_NEAREST_MIPMAP_NEAREST:
-            ti->mmMode = GR_MIPMAP_NEAREST;
-            ti->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-            ti->LODblend = FXFALSE;
-            break;
-        case GL_LINEAR_MIPMAP_LINEAR:
-            if (TDFX_IS_NAPALM(fxMesa)) {
-                if (fxMesa->haveTwoTMUs) {
-                    ti->mmMode = GR_MIPMAP_NEAREST;
-                    ti->LODblend = FXTRUE;
-                }
-                else {
-                    ti->mmMode = GR_MIPMAP_NEAREST_DITHER;
-                    ti->LODblend = FXFALSE;
-                }
-                ti->minFilt = GR_TEXTUREFILTER_BILINEAR;
-                break;
-            }
-            /* XXX Voodoo3/Banshee mipmap blending seems to produce
-             * incorrectly filtered colors for the smallest mipmap levels.
-             * To work-around we fall-through here and use a different filter.
-             */
-        case GL_LINEAR_MIPMAP_NEAREST:
-            ti->mmMode = GR_MIPMAP_NEAREST;
-            ti->minFilt = GR_TEXTUREFILTER_BILINEAR;
-            ti->LODblend = FXFALSE;
-            break;
-        default:
-            break;
-        }
-        RevalidateTexture(ctx, tObj);
-        fxMesa->new_state |= TDFX_NEW_TEXTURE;
-        break;
-
-    case GL_TEXTURE_MAG_FILTER:
-        switch (param) {
-        case GL_NEAREST:
-            ti->magFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
-            break;
-        case GL_LINEAR:
-            ti->magFilt = GR_TEXTUREFILTER_BILINEAR;
-            break;
-        default:
-            break;
-        }
-        fxMesa->new_state |= TDFX_NEW_TEXTURE;
-        break;
-
-    case GL_TEXTURE_WRAP_S:
-        switch (param) {
-        case GL_CLAMP:
-            ti->sClamp = GR_TEXTURECLAMP_CLAMP;
-            break;
-        case GL_REPEAT:
-            ti->sClamp = GR_TEXTURECLAMP_WRAP;
-            break;
-        default:
-            break;
-        }
-        fxMesa->new_state |= TDFX_NEW_TEXTURE;
-        break;
-
-    case GL_TEXTURE_WRAP_T:
-        switch (param) {
-        case GL_CLAMP:
-            ti->tClamp = GR_TEXTURECLAMP_CLAMP;
-            break;
-        case GL_REPEAT:
-            ti->tClamp = GR_TEXTURECLAMP_WRAP;
-            break;
-        default:
-            break;
-        }
-        fxMesa->new_state |= TDFX_NEW_TEXTURE;
-        break;
-
-    case GL_TEXTURE_BORDER_COLOR:
-        /* TO DO */
-        break;
-    case GL_TEXTURE_MIN_LOD:
-        /* TO DO */
-        break;
-    case GL_TEXTURE_MAX_LOD:
-        /* TO DO */
-        break;
-    case GL_TEXTURE_BASE_LEVEL:
-        RevalidateTexture(ctx, tObj);
-        break;
-    case GL_TEXTURE_MAX_LEVEL:
-        RevalidateTexture(ctx, tObj);
-        break;
-
-    default:
-        break;
-    }
-}
+   tdfxRevalidateTexture( ctx, texObj );
 
+   t->reloadImages = GL_TRUE;
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;
 
-/*
- * Called via glDeleteTextures to delete a texture object.
- * Here, we delete the Glide data associated with the texture.
- */
-void
-tdfxDDDeleteTexture(GLcontext * ctx, struct gl_texture_object *tObj)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTMFreeTexture(fxMesa, tObj);
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
+   *retainInternalCopy = GL_FALSE;
+   return GL_TRUE;
 }
 
 
-/*
- * Return true if texture is resident, false otherwise.
- */
-GLboolean
-tdfxDDIsTextureResident(GLcontext *ctx, struct gl_texture_object *tObj)
+static GLboolean
+tdfxDDTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
+		     GLint xoffset, GLint yoffset,
+		     GLsizei width, GLsizei height,
+		     GLenum format, GLenum type, const GLvoid *pixels,
+		     const struct gl_pixelstore_attrib *packing,
+		     struct gl_texture_object *texObj,
+		     struct gl_texture_image *texImage )
 {
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    return (GLboolean) (ti && ti->isInTM);
-}
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+   tdfxTexImagePtr image;
 
+   if ( target != GL_TEXTURE_2D )
+      return GL_FALSE;
 
+   if ( !t )
+      return GL_FALSE;
 
-/*
- * Convert a gl_color_table texture palette to Glide's format.
- */
-static void
-convertPalette(FxU32 data[256], const struct gl_color_table *table)
-{
-    const GLubyte *tableUB = (const GLubyte *) table->Table;
-    GLint width = table->Size;
-    FxU32 r, g, b, a;
-    GLint i;
-
-    ASSERT(table->TableType == GL_UNSIGNED_BYTE);
-
-    switch (table->Format) {
-    case GL_INTENSITY:
-        for (i = 0; i < width; i++) {
-            r = tableUB[i];
-            g = tableUB[i];
-            b = tableUB[i];
-            a = tableUB[i];
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_LUMINANCE:
-        for (i = 0; i < width; i++) {
-            r = tableUB[i];
-            g = tableUB[i];
-            b = tableUB[i];
-            a = 255;
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_ALPHA:
-        for (i = 0; i < width; i++) {
-            r = g = b = 255;
-            a = tableUB[i];
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_LUMINANCE_ALPHA:
-        for (i = 0; i < width; i++) {
-            r = g = b = tableUB[i * 2 + 0];
-            a = tableUB[i * 2 + 1];
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_RGB:
-        for (i = 0; i < width; i++) {
-            r = tableUB[i * 3 + 0];
-            g = tableUB[i * 3 + 1];
-            b = tableUB[i * 3 + 2];
-            a = 255;
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    case GL_RGBA:
-        for (i = 0; i < width; i++) {
-            r = tableUB[i * 4 + 0];
-            g = tableUB[i * 4 + 1];
-            b = tableUB[i * 4 + 2];
-            a = tableUB[i * 4 + 3];
-            data[i] = (a << 24) | (r << 16) | (g << 8) | b;
-        }
-        break;
-    }
-}
+   if ( 0 ) {
+      fprintf( stderr, "TexSubImage id=%d lvl=%d int=0x%x format=0x%x type=0x%x  x=%d y=%d w=%d h=%d fullW=%d fullH=%d\n",
+	       texObj->Name, level, texImage->IntFormat, format, type,
+	       xoffset, yoffset, width, height,
+	       texImage->Width, texImage->Height );
+   }
 
+   image = &t->image[level];
 
+   /* Must have an existing texture image!
+    */
+   assert( image->original.data );
 
-void
-tdfxDDTexturePalette(GLcontext * ctx, struct gl_texture_object *tObj)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-
-    if (tObj) {
-        /* per-texture palette */
-        tdfxTexInfo *ti;
-        if (!tObj->DriverData)
-            tObj->DriverData = fxAllocTexObjData(fxMesa);
-        ti = TDFX_TEXTURE_DATA(tObj);
-        convertPalette(ti->palette.data, &tObj->Palette);
-        /*tdfxTexInvalidate(ctx, tObj);*/
-    }
-    else {
-        /* global texture palette */
-        convertPalette(fxMesa->glbPalette.data, &ctx->Texture.Palette);
-    }
-    fxMesa->new_state |= TDFX_NEW_TEXTURE; /* XXX too heavy-handed */
+   if ( !_mesa_convert_texsubimage2d( texImage->TexFormat->IntFormat,
+				      xoffset, yoffset, width, height,
+				      texImage->Width,
+				      format, type, packing,
+				      pixels, image->original.data ) ) {
+      return GL_FALSE;
+   }
+
+   /* Rescale the original image again if we have to.
+    */
+   if ( image->wScale > 1 || image->hScale > 1 ) {
+      assert( image->rescaled.data );
+      _mesa_rescale_teximage2d( texImage->TexFormat,
+				image->original.width, image->original.height,
+				image->rescaled.width, image->rescaled.height,
+				image->original.data, image->rescaled.data );
+   }
+
+   t->reloadImages = GL_TRUE; /* signal the image needs to be reloaded */
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;  /* XXX this might be a bit much */
+
+   return GL_TRUE;
 }
 
 
-/*
- * Given an OpenGL internal texture format, return the corresponding
- * Glide internal texture format and MesaIntTexFormat.
- * If allow32bpp is true, we'll return 32-bit texel formats when
- * appropriate.
- */
-static void
-tdfxTexGetFormat(GLenum intFormatHint, GLboolean allow32bpp,
-                 GrTextureFormat_t *glideFormat,
-                 MesaIntTexFormat *mesaFormat,
-                 GLint *texelSize)
-{
-    switch (intFormatHint) {
-    case 1:
-    case GL_LUMINANCE:
-    case GL_LUMINANCE4:
-    case GL_LUMINANCE8:
-    case GL_LUMINANCE12:
-    case GL_LUMINANCE16:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_INTENSITY_8;
-        if (mesaFormat)
-            *mesaFormat = MESA_L8;
-        if (texelSize)
-            *texelSize = 1;
-        break;
-    case 2:
-    case GL_LUMINANCE_ALPHA:
-    case GL_LUMINANCE4_ALPHA4:
-    case GL_LUMINANCE6_ALPHA2:
-    case GL_LUMINANCE8_ALPHA8:
-    case GL_LUMINANCE12_ALPHA4:
-    case GL_LUMINANCE12_ALPHA12:
-    case GL_LUMINANCE16_ALPHA16:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ALPHA_INTENSITY_88;
-        if (mesaFormat)
-            *mesaFormat = MESA_A8_L8;
-        if (texelSize)
-            *texelSize = 2;
-        break;
-    case GL_INTENSITY:
-    case GL_INTENSITY4:
-    case GL_INTENSITY8:
-    case GL_INTENSITY12:
-    case GL_INTENSITY16:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ALPHA_8;
-        if (mesaFormat)
-            *mesaFormat = MESA_I8;
-        if (texelSize)
-            *texelSize = 1;
-       break;
-    case GL_ALPHA:
-    case GL_ALPHA4:
-    case GL_ALPHA8:
-    case GL_ALPHA12:
-    case GL_ALPHA16:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ALPHA_8;
-        if (mesaFormat)
-            *mesaFormat = MESA_A8;
-        if (texelSize)
-            *texelSize = 1;
-        break;
-    case GL_R3_G3_B2:
-    case GL_RGB4:
-    case GL_RGB5:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_RGB_565;
-        if (mesaFormat)
-            *mesaFormat = MESA_R5_G6_B5;
-        if (texelSize)
-            *texelSize = 2;
-        break;
-    case 3:
-    case GL_RGB:
-    case GL_RGB8:
-    case GL_RGB10:
-    case GL_RGB12:
-    case GL_RGB16:
-        if (allow32bpp) {
-            if (glideFormat)
-                *glideFormat = GR_TEXFMT_ARGB_8888;
-            if (mesaFormat)
-                *mesaFormat = MESA_FF_R8_G8_B8;
-            if (texelSize)
-                *texelSize = 4;
-        }
-        else {
-            if (glideFormat)
-                *glideFormat = GR_TEXFMT_RGB_565;
-            if (mesaFormat)
-                *mesaFormat = MESA_R5_G6_B5;
-            if (texelSize)
-                *texelSize = 2;
-        }
-        break;
-    case GL_RGBA2:
-    case GL_RGBA4:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ARGB_4444;
-        if (mesaFormat)
-            *mesaFormat = MESA_A4_R4_G4_B4;
-        if (texelSize)
-            *texelSize = 2;
-        break;
-    case 4:
-    case GL_RGBA:
-    case GL_RGBA8:
-    case GL_RGB10_A2:
-    case GL_RGBA12:
-    case GL_RGBA16:
-        if (allow32bpp) {
-            if (glideFormat)
-                *glideFormat = GR_TEXFMT_ARGB_8888;
-            if (mesaFormat)
-                *mesaFormat = MESA_A8_R8_G8_B8;
-            if (texelSize)
-                *texelSize = 4;
-        }
-        else {
-            if (glideFormat)
-                *glideFormat = GR_TEXFMT_ARGB_4444;
-            if (mesaFormat)
-                *mesaFormat = MESA_A4_R4_G4_B4;
-            if (texelSize)
-                *texelSize = 2;
-        }
-        break;
-    case GL_RGB5_A1:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ARGB_1555;
-        if (mesaFormat)
-            *mesaFormat = MESA_A1_R5_G5_B5;
-        if (texelSize)
-           *texelSize = 2;
-        break;
-    case GL_COLOR_INDEX:
-    case GL_COLOR_INDEX1_EXT:
-    case GL_COLOR_INDEX2_EXT:
-    case GL_COLOR_INDEX4_EXT:
-    case GL_COLOR_INDEX8_EXT:
-    case GL_COLOR_INDEX12_EXT:
-    case GL_COLOR_INDEX16_EXT:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_P_8;
-        if (mesaFormat)
-            *mesaFormat = MESA_C8;
-        if (texelSize)
-            *texelSize = 1;
-        break;
-    case GL_COMPRESSED_RGB_FXT1_3DFX:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ARGB_CMP_FXT1;
-        if (mesaFormat)
-            *mesaFormat = MESA_A8_R8_G8_B8;
-        if (texelSize)
-            *texelSize = 4;
-        break;
-    case GL_COMPRESSED_RGBA_FXT1_3DFX:
-        if (glideFormat)
-            *glideFormat = GR_TEXFMT_ARGB_CMP_FXT1;
-        if (mesaFormat)
-            *mesaFormat = MESA_A8_R8_G8_B8;
-        if (texelSize)
-            *texelSize = 4;
-        break;
-    default:
-        fprintf(stderr, "bad texture format in fxTexGetFormat() %d", intFormatHint);
-        break;
-    }
-}
 
+/* ================================================================
+ *
+ */
 
-/**********************************************************************/
-/**** NEW TEXTURE IMAGE FUNCTIONS                                  ****/
-/**********************************************************************/
+static void tdfxDDTexEnv( GLcontext *ctx, GLenum target,
+			  GLenum pname, const GLfloat *param )
+{
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
 
-static FxBool TexusFatalError = FXFALSE;
-static FxBool TexusError = FXFALSE;
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_API ) {
+      if ( param ) {
+	 fprintf( stderr, __FUNCTION__"( %x, %x )\n", pname, (GLint)(*param) );
+      } else {
+	 fprintf( stderr, __FUNCTION__"( %x )\n", pname );
+      }
+   }
 
-#define TX_DITHER_NONE                                  0x00000000
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;
+}
 
-static void
-fxTexusError(const char *string, FxBool fatal)
+static void tdfxDDTexParameter( GLcontext *ctx, GLenum target,
+				struct gl_texture_object *tObj,
+				GLenum pname, const GLfloat *params )
 {
-    gl_problem(NULL, string);
-   /*
-    * Just propagate the fatal value up.
-    */
-    TexusError = FXTRUE;
-    TexusFatalError = fatal;
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   GLenum param = (GLenum) (GLint) params[0];
+   tdfxTexObjPtr t;
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, __FUNCTION__ "( %d, %p, %x, %x )\n",
+	       tObj->Name, tObj->DriverData, pname, param );
+   }
+
+   if ( target != GL_TEXTURE_2D )
+      return;
+
+   if ( !tObj->DriverData )
+      tObj->DriverData = tdfxAllocTexObj( fxMesa );
+
+   t = TDFX_TEXTURE_DATA(tObj);
+
+   switch ( pname ) {
+   case GL_TEXTURE_MIN_FILTER:
+      switch ( param ) {
+      case GL_NEAREST:
+	 t->mmMode = GR_MIPMAP_DISABLE;
+	 t->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+	 t->LODblend = FXFALSE;
+	 break;
+
+      case GL_LINEAR:
+	 t->mmMode = GR_MIPMAP_DISABLE;
+	 t->minFilt = GR_TEXTUREFILTER_BILINEAR;
+	 t->LODblend = FXFALSE;
+	 break;
+
+      case GL_NEAREST_MIPMAP_LINEAR:
+	 if ( TDFX_IS_NAPALM(fxMesa) ) {
+	    if ( fxMesa->numTMUs > 1 ) {
+	       t->mmMode = GR_MIPMAP_NEAREST;
+	       t->LODblend = FXTRUE;
+	    } else {
+	       t->mmMode = GR_MIPMAP_NEAREST_DITHER;
+	       t->LODblend = FXFALSE;
+	    }
+	    t->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+	    break;
+	 }
+	 /* XXX Voodoo3/Banshee mipmap blending seems to produce
+	  * incorrectly filtered colors for the smallest mipmap levels.
+	  * To work-around we fall-through here and use a different filter.
+	  */
+      case GL_NEAREST_MIPMAP_NEAREST:
+	 t->mmMode = GR_MIPMAP_NEAREST;
+	 t->minFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+	 t->LODblend = FXFALSE;
+	 break;
+
+      case GL_LINEAR_MIPMAP_LINEAR:
+	 if ( TDFX_IS_NAPALM(fxMesa) ) {
+	    if ( fxMesa->numTMUs > 1 ) {
+	       t->mmMode = GR_MIPMAP_NEAREST;
+	       t->LODblend = FXTRUE;
+	    } else {
+	       t->mmMode = GR_MIPMAP_NEAREST_DITHER;
+	       t->LODblend = FXFALSE;
+	    }
+	    t->minFilt = GR_TEXTUREFILTER_BILINEAR;
+	    break;
+	 }
+	 /* XXX Voodoo3/Banshee mipmap blending seems to produce
+	  * incorrectly filtered colors for the smallest mipmap levels.
+	  * To work-around we fall-through here and use a different filter.
+	  */
+      case GL_LINEAR_MIPMAP_NEAREST:
+	 t->mmMode = GR_MIPMAP_NEAREST;
+	 t->minFilt = GR_TEXTUREFILTER_BILINEAR;
+	 t->LODblend = FXFALSE;
+	 break;
+      default:
+	 break;
+      }
+      tdfxRevalidateTexture( ctx, tObj );
+      fxMesa->new_state |= TDFX_NEW_TEXTURE;
+      break;
+
+   case GL_TEXTURE_MAG_FILTER:
+      switch ( param ) {
+      case GL_NEAREST:
+	 t->magFilt = GR_TEXTUREFILTER_POINT_SAMPLED;
+	 break;
+      case GL_LINEAR:
+	 t->magFilt = GR_TEXTUREFILTER_BILINEAR;
+	 break;
+      default:
+	 break;
+      }
+      fxMesa->new_state |= TDFX_NEW_TEXTURE;
+      break;
+
+   case GL_TEXTURE_WRAP_S:
+      switch ( param ) {
+      case GL_CLAMP:
+	 t->sClamp = GR_TEXTURECLAMP_CLAMP;
+	 break;
+      case GL_REPEAT:
+	 t->sClamp = GR_TEXTURECLAMP_WRAP;
+	 break;
+      default:
+	 break;
+      }
+      fxMesa->new_state |= TDFX_NEW_TEXTURE;
+      break;
+
+   case GL_TEXTURE_WRAP_T:
+      switch ( param ) {
+      case GL_CLAMP:
+	 t->tClamp = GR_TEXTURECLAMP_CLAMP;
+	 break;
+      case GL_REPEAT:
+	 t->tClamp = GR_TEXTURECLAMP_WRAP;
+	 break;
+      default:
+	 break;
+      }
+      fxMesa->new_state |= TDFX_NEW_TEXTURE;
+      break;
+
+   case GL_TEXTURE_BASE_LEVEL:
+      tdfxRevalidateTexture( ctx, tObj );
+      break;
+
+   case GL_TEXTURE_MAX_LEVEL:
+      tdfxRevalidateTexture( ctx, tObj );
+      break;
+
+   case GL_TEXTURE_BORDER_COLOR:
+      /* TO DO */
+      break;
+   case GL_TEXTURE_MIN_LOD:
+      /* TO DO */
+      break;
+   case GL_TEXTURE_MAX_LOD:
+      /* TO DO */
+      break;
+
+   default:
+      break;
+   }
 }
 
-GLboolean
-tdfxDDTexImage2D(GLcontext * ctx, GLenum target, GLint level,
-                 GLenum format, GLenum type, const GLvoid * pixels,
-                 const struct gl_pixelstore_attrib * packing,
-                 struct gl_texture_object * texObj,
-                 struct gl_texture_image * texImage,
-                 GLboolean * retainInternalCopy)
+static void tdfxDDBindTexture( GLcontext *ctx, GLenum target,
+			       struct gl_texture_object *tObj )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    const GLboolean allow32bpt = TDFX_IS_NAPALM(fxMesa);
-    GrTextureFormat_t gldformat;
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
-    GLint dstWidth, dstHeight, wScale, hScale, texelSize, dstStride;
-    MesaIntTexFormat intFormat;
-    GLboolean isCompressedFormat;
-    GLint texsize;
-    void *uncompressedImage;
-
-    /*
-    printf("TexImage id=%d int 0x%x  format 0x%x  type 0x%x  %dx%d\n",
-           texObj->Name, texImage->IntFormat, format, type,
-           texImage->Width, texImage->Height);
-    */
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   tdfxTexObjPtr t;
 
-    isCompressedFormat = texImage->IsCompressed;
-    if (target != GL_TEXTURE_2D || texImage->Border > 0)
-        return GL_FALSE;
-
-    if (!texObj->DriverData)
-        texObj->DriverData = fxAllocTexObjData(fxMesa);
-
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-
-    /* Determine the appropriate GL internal texel format, Mesa internal
-     * texel format, and texelSize (bytes) given the user's internal
-     * texture format hint.
-     */
-    tdfxTexGetFormat(texImage->IntFormat, allow32bpt,
-                     &gldformat, &intFormat, &texelSize);
-
-    /* Determine width and height scale factors for texture.
-     * Remember, Glide is limited to 8:1 aspect ratios.
-     */
-    tdfxTexGetInfo(ctx,
-                   texImage->Width, texImage->Height,
-                   NULL,       /* lod level          */
-                   NULL,       /* aspect ratio       */
-                   NULL, NULL, /* sscale, tscale     */
-                   &wScale, &hScale);
-    dstWidth = texImage->Width * wScale;
-    dstHeight = texImage->Height * hScale;
-    if (isCompressedFormat) {
-        texsize = tdfxDDCompressedImageSize(ctx,
-                                            texImage->IntFormat,
-                                            2,
-                                            texImage->Width,
-                                            texImage->Height,
-                                            1);
-    }
-    else {
-        texsize = dstWidth * dstHeight * texelSize;
-    }
-
-   /*
-    * If the image is not compressed, this doesn't
-    * matter, but it might as well have a sensible
-    * value, and it might save a failure later on.
-    */
-    texImage->CompressedSize = texsize;
-    /* housekeeping */
-    _mesa_set_teximage_component_sizes(intFormat, texImage);
-
-    /*
-     * allocate new storage for texture image, if needed.
-     * This conditional wants to set uncompressedImage to
-     * point to the uncompressed image, and mml->data to
-     * the texture data.  If the image is uncompressed,
-     * these are identical.  If the image is not compressed,
-     * these are different.
-     */
-    if (!mml->data || mml->glideFormat != gldformat ||
-        mml->width != dstWidth || mml->height != dstHeight ||
-        texsize != mml->dataSize ) {
-        if (mml->data) {
-            FREE(mml->data);
-        }
-        uncompressedImage
-            = (void *)MALLOC(dstWidth * dstHeight * texelSize);
-        if (!uncompressedImage) {
-            return GL_FALSE;
-        }
-        if (isCompressedFormat) {
-            mml->data = MALLOC(texsize);
-            if (!mml->data) {
-                FREE(uncompressedImage);
-                return GL_FALSE;
-            }
-        } else {
-            mml->data = uncompressedImage;
-        }
-        mml->texelSize = texelSize;
-        mml->glideFormat = gldformat;
-        mml->width = dstWidth;
-        mml->height = dstHeight;
-        mml->dataSize = texsize;
-        ti->info.format = gldformat;
-        tdfxTMMoveOutTM(fxMesa, texObj);
-        /*tdfxTexInvalidate(ctx, texObj);*/
-    }
-    else {
-       /*
-        * Here we don't have to allocate anything, but we
-        * do have to point uncompressedImage to the uncompressed
-        * data.
-        */
-        if (isCompressedFormat) {
-            uncompressedImage
-                = (void *)MALLOC(dstWidth * dstHeight * texelSize);
-            if (!uncompressedImage) {
-                return GL_FALSE;
-            }
-        } else {
-            uncompressedImage = mml->data;
-        }
-    }
-
-    dstStride = dstWidth * texelSize;
-
-    /* store the texture image into uncompressedImage */
-    if (!_mesa_convert_teximage(intFormat,
-                                dstWidth, dstHeight,
-                                uncompressedImage,
-                                dstStride,
-                                texImage->Width, texImage->Height,
-                                format, type, pixels, packing)) {
-        /*printf("convert failed\n");*/
-        return GL_FALSE;  /* not necessarily an error */
-    }
-
-    /*
-     * Now compress it if necessary.
-     */
-    if (isCompressedFormat) {
-        TxErrorCallbackFnc_t oldErrorCallback;
-        (*txErrorSetCallbackProc)(fxTexusError, &oldErrorCallback);
-        (*txImgQuantizeProc)((char *)mml->data,
-                             (char *)uncompressedImage,
-                             texImage->Width,
-                             texImage->Height,
-                             gldformat,
-                             TX_DITHER_NONE);
-        (*txErrorSetCallbackProc)(oldErrorCallback, NULL);
-        if (uncompressedImage != mml->data) {
-            /*
-             * We do not need this any more, errors or no.
-             */
-            FREE(uncompressedImage);
-        }
-        TexusError = FXFALSE;
-        if (TexusFatalError) {
-            FREE(mml->data);
-            mml->data = NULL;
-            TexusFatalError = FXFALSE;
-            return GL_FALSE;
-        }
-    }
-
-
-    RevalidateTexture(ctx, texObj);
-
-    ti->reloadImages = GL_TRUE;
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-
-    *retainInternalCopy = GL_FALSE;
-    return GL_TRUE;
-}
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, __FUNCTION__ "( %d, %p )\n",
+	       tObj->Name, tObj->DriverData );
+   }
 
+   if ( target != GL_TEXTURE_2D )
+      return;
 
-GLboolean
-tdfxDDTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
-                    GLint xoffset, GLint yoffset,
-                    GLsizei width, GLsizei height,
-                    GLenum format, GLenum type, const GLvoid * pixels,
-                    const struct gl_pixelstore_attrib * packing,
-                    struct gl_texture_object * texObj,
-                    struct gl_texture_image * texImage)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTexInfo *ti;
-    GLint wscale, hscale, dstStride = 0;
-    tdfxMipMapLevel *mml;
-    GLboolean result;
-    void *uncompressedImage = (void *)0;
-    FxU32 uncompressedSize;
-    TxErrorCallbackFnc_t oldErrorCallback;
-
-    if (target != GL_TEXTURE_2D)
-        return GL_FALSE;
-
-    if (!texObj->DriverData)
-        return GL_FALSE;
-
-    /*
-    printf("TexSubImage id=%d lvl=%d int=0x%x format=0x%x type=0x%x  x=%d y=%d w=%d h=%d fullW=%d fullH=%d\n",
-           texObj->Name, level,
-           texImage->IntFormat, format, type, xoffset, yoffset, width, height,
-           texImage->Width, texImage->Height);
-    */
+   if ( !tObj->DriverData )
+      tObj->DriverData = tdfxAllocTexObj( fxMesa );
 
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-
-    tdfxTexGetInfo(ctx, texImage->Width, texImage->Height, NULL, NULL,
-                   NULL, NULL, &wscale, &hscale);
-
-    /*
-     * Must have an existing texture image!
-     */
-    assert(mml->data);
-
-    switch (mml->glideFormat) {
-    case GR_TEXFMT_INTENSITY_8:
-        dstStride = mml->width;
-        result = _mesa_convert_texsubimage(MESA_I8, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_ALPHA_8:
-        dstStride = mml->width;
-        result = _mesa_convert_texsubimage(MESA_A8, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_P_8:
-        dstStride = mml->width;
-        result = _mesa_convert_texsubimage(MESA_C8, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_ALPHA_INTENSITY_88:
-        dstStride = mml->width * 2;
-        result = _mesa_convert_texsubimage(MESA_A8_L8, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_RGB_565:
-        dstStride = mml->width * 2;
-        result = _mesa_convert_texsubimage(MESA_R5_G6_B5, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_ARGB_4444:
-        dstStride = mml->width * 2;
-        result = _mesa_convert_texsubimage(MESA_A4_R4_G4_B4, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    case GR_TEXFMT_ARGB_CMP_FXT1:
-        /*
-         * There are some special legality constraints for compressed
-         * textures.
-         */
-        if ((xoffset != texImage->Border)
-            || (yoffset != texImage->Border)) {
-            gl_error( ctx,
-                      GL_INVALID_OPERATION,
-                      "glTexSubImage2D(offset)" );
-            return GL_FALSE;
-        }
-        if ((width != texImage->Width)
-            || (height != texImage->Height)) {
-            gl_error( ctx,
-                      GL_INVALID_VALUE,
-                      "glTexSubImage2D(image size)" );
-            return GL_FALSE;
-        }
-        /*
-         * The width and height have to be multiples of
-         * 8 and 4 respectively.
-         */
-        width  = (mml->width  + 0x7) &~ 0x7;
-        height = (mml->height + 0x3) &~ 0x3;
-        /*
-         * A texel is 8888 for this format.
-         */
-        uncompressedSize = mml->width * mml->height * 4;
-        uncompressedImage = (void *)MALLOC(uncompressedSize);
-        /*
-         * Convert the data.
-         */
-        dstStride = mml->width * 4;
-        result = _mesa_convert_texsubimage(MESA_A8_R8_G8_B8, xoffset, yoffset,
-                                           mml->width, mml->height, uncompressedImage,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        if (!result) {
-            FREE(uncompressedImage);
-            printf("TexSubImage convert failed\n");
-            return GL_FALSE;
-        }
-        /*
-         * Now that we have converted the data, then compress it.
-         */
-        (*txErrorSetCallbackProc)(fxTexusError, &oldErrorCallback);
-        (*txImgQuantizeProc)((char *)mml->data,
-                            (char *)uncompressedImage,
-                            mml->width,
-                            mml->height,
-                            mml->glideFormat,
-                            TX_DITHER_NONE);
-        (*txErrorSetCallbackProc)(oldErrorCallback, NULL);
-        result = TexusFatalError;
-        TexusFatalError = TexusError = FXFALSE;
-        /*
-         * We don't need this any more.
-         */
-        FREE(uncompressedImage);
-        break;
-    case GR_TEXFMT_ARGB_8888:
-        {
-            MesaIntTexFormat intFormat;
-            if (texImage->Format == GL_RGB) {
-                /* An RGB image padded out to 4 bytes/texel */
-                intFormat = MESA_FF_R8_G8_B8;
-            }
-            else {
-                intFormat = MESA_A8_R8_G8_B8;
-            }
-            dstStride = mml->width * 4;
-            result = _mesa_convert_texsubimage(intFormat, xoffset, yoffset,
-                                            mml->width, mml->height, mml->data,
-                                            dstStride, width, height,
-                                            texImage->Width, texImage->Height,
-                                            format, type, pixels, packing);
-        }
-        break;
-    case GR_TEXFMT_ARGB_1555:
-        dstStride = mml->width * 2;
-        result = _mesa_convert_texsubimage(MESA_A1_R5_G5_B5, xoffset, yoffset,
-                                           mml->width, mml->height, mml->data,
-                                           dstStride, width, height,
-                                           texImage->Width, texImage->Height,
-                                           format, type, pixels, packing);
-        break;
-    default:
-        gl_problem(NULL, "tdfx driver: fxTexBuildSubImageMap() bad format");
-        result = GL_FALSE;
-    }
-
-    if (!result) {
-        return GL_FALSE;
-    }
-
-    ti->reloadImages = GL_TRUE; /* signal the image needs to be reloaded */
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;  /* XXX this might be a bit much */
-
-    return GL_TRUE;
+   t = TDFX_TEXTURE_DATA(tObj);
+   t->lastTimeUsed = fxMesa->texBindNumber++;
+
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;
 }
 
+static void tdfxDDDeleteTexture( GLcontext *ctx,
+				 struct gl_texture_object *tObj )
+{
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
 
-/**********************************************************************/
-/**** COMPRESSED TEXTURE IMAGE FUNCTIONS                           ****/
-/**********************************************************************/
+   LOCK_HARDWARE( fxMesa );
+   tdfxTMFreeTextureLocked( fxMesa, tObj );
+   UNLOCK_HARDWARE( fxMesa );
 
-GLboolean
-tdfxDDCompressedTexImage2D( GLcontext *ctx, GLenum target,
-                            GLint level, GLsizei imageSize,
-                            const GLvoid *data,
-                            struct gl_texture_object *texObj,
-                            struct gl_texture_image *texImage,
-                            GLboolean *retainInternalCopy)
-{
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    const GLboolean allow32bpt = TDFX_IS_NAPALM(fxMesa);
-    GrTextureFormat_t gldformat;
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
-    GLint dstWidth, dstHeight, wScale, hScale, texelSize;
-    MesaIntTexFormat intFormat;
-    GLboolean isCompressedFormat;
-    GLsizei texsize;
-
-    if (target != GL_TEXTURE_2D || texImage->Border > 0)
-        return GL_FALSE;
-
-    if (!texObj->DriverData)
-        texObj->DriverData = fxAllocTexObjData(fxMesa);
-
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-
-    isCompressedFormat = tdfxDDIsCompressedGlideFormatMacro(texImage->IntFormat);
-    if (!isCompressedFormat) {
-        gl_error( ctx, GL_INVALID_ENUM, "glCompressedTexImage2D(format)" );
-        return GL_FALSE;
-    }
-    /* Determine the apporpriate GL internal texel format, Mesa internal
-     * texel format, and texelSize (bytes) given the user's internal
-     * texture format hint.
-     */
-    tdfxTexGetFormat(texImage->IntFormat, allow32bpt,
-                     &gldformat, &intFormat, &texelSize);
-
-    /* Determine width and height scale factors for texture.
-     * Remember, Glide is limited to 8:1 aspect ratios.
-     */
-    tdfxTexGetInfo(ctx,
-                   texImage->Width, texImage->Height,
-                   NULL,       /* lod level          */
-                   NULL,       /* aspect ratio       */
-                   NULL, NULL, /* sscale, tscale     */
-                   &wScale, &hScale);
-    dstWidth = texImage->Width * wScale;
-    dstHeight = texImage->Height * hScale;
-    /* housekeeping */
-    _mesa_set_teximage_component_sizes(intFormat, texImage);
-
-    texsize = tdfxDDCompressedImageSize(ctx,
-                                        texImage->IntFormat,
-                                        2,
-                                        texImage->Width,
-                                        texImage->Height,
-                                        1);
-    if (texsize != imageSize) {
-        gl_error(ctx,
-                 GL_INVALID_VALUE,
-                 "glCompressedTexImage2D(texsize)");
-        return GL_FALSE;
-    }
-
-    /* allocate new storage for texture image, if needed */
-    if (!mml->data || mml->glideFormat != gldformat ||
-        mml->width != dstWidth || mml->height != dstHeight ||
-        texsize != mml->dataSize) {
-        if (mml->data) {
-            FREE(mml->data);
-        }
-        mml->data = MALLOC(texsize);
-        if (!mml->data) {
-            return GL_FALSE;
-        }
-        mml->texelSize = texelSize;
-        mml->glideFormat = gldformat;
-        mml->width = dstWidth;
-        mml->height = dstHeight;
-        tdfxTMMoveOutTM(fxMesa, texObj);
-        /*tdfxTexInvalidate(ctx, texObj);*/
-    }
-
-    /* save the texture data */
-    MEMCPY(mml->data, data, imageSize);
-
-    RevalidateTexture(ctx, texObj);
-
-    ti->reloadImages = GL_TRUE;
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-
-    *retainInternalCopy = GL_FALSE;
-    return GL_TRUE;
+   fxMesa->new_state |= TDFX_NEW_TEXTURE;
 }
 
-GLboolean
-tdfxDDCompressedTexSubImage2D( GLcontext *ctx, GLenum target,
-                               GLint level, GLint xoffset,
-                               GLint yoffset, GLsizei width,
-                               GLint height, GLenum format,
-                               GLsizei imageSize, const GLvoid *data,
-                               struct gl_texture_object *texObj,
-                               struct gl_texture_image *texImage )
+static GLboolean tdfxDDIsTextureResident( GLcontext *ctx,
+					  struct gl_texture_object *tObj )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
-
-    /*
-     * We punt if we are not replacing the entire image.  This
-     * is allowed by the spec.
-     */
-    if ((xoffset != 0) && (yoffset != 0)
-        && (width != texImage->Width)
-        && (height != texImage->Height)) {
-        return GL_FALSE;
-    }
-
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-    if (imageSize != mml->dataSize) {
-        return GL_FALSE;
-    }
-    MEMCPY(data, mml->data, imageSize);
-
-    ti->reloadImages = GL_TRUE;
-    fxMesa->new_state |= TDFX_NEW_TEXTURE;
-
-    return GL_TRUE;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+
+   return ( t && t->isInTM );
 }
 
 
-#if	0
+/* Convert a gl_color_table texture palette to Glide's format.
+ */
 static void
-PrintTexture(int w, int h, int c, const GLubyte * data)
+tdfxConvertPalette( FxU32 data[256], const struct gl_color_table *table )
 {
-    int i, j;
-    for (i = 0; i < h; i++) {
-        for (j = 0; j < w; j++) {
-            if (c == 2)
-                printf("%02x %02x  ", data[0], data[1]);
-            else if (c == 3)
-                printf("%02x %02x %02x  ", data[0], data[1], data[2]);
-            data += c;
-        }
-        printf("\n");
-    }
+   const GLubyte *tableUB = (const GLubyte *) table->Table;
+   GLint width = table->Size;
+   FxU32 r, g, b, a;
+   GLint i;
+
+   ASSERT( table->TableType == GL_UNSIGNED_BYTE );
+
+   switch ( table->Format ) {
+   case GL_RGBA:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = tableUB[i * 4 + 0];
+	 g = tableUB[i * 4 + 1];
+	 b = tableUB[i * 4 + 2];
+	 a = tableUB[i * 4 + 3];
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_RGB:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = tableUB[i * 3 + 0];
+	 g = tableUB[i * 3 + 1];
+	 b = tableUB[i * 3 + 2];
+	 a = 255;
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_LUMINANCE:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = tableUB[i];
+	 g = tableUB[i];
+	 b = tableUB[i];
+	 a = 255;
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_ALPHA:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = g = b = 255;
+	 a = tableUB[i];
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_LUMINANCE_ALPHA:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = g = b = tableUB[i * 2 + 0];
+	 a = tableUB[i * 2 + 1];
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   case GL_INTENSITY:
+      for ( i = 0 ; i < width ; i++ ) {
+	 r = tableUB[i];
+	 g = tableUB[i];
+	 b = tableUB[i];
+	 a = tableUB[i];
+	 data[i] = PACK_COLOR_8888( a, r, g, b );
+      }
+      break;
+   }
 }
-#endif
 
-
-GLboolean
-tdfxDDTestProxyTexImage(GLcontext *ctx, GLenum target,
-                        GLint level, GLint internalFormat,
-                        GLenum format, GLenum type,
-                        GLint width, GLint height,
-                        GLint depth, GLint border)
+static void
+tdfxDDTexturePalette( GLcontext *ctx, struct gl_texture_object *tObj )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-
-    switch (target) {
-    case GL_PROXY_TEXTURE_1D:
-        return GL_TRUE;  /* software rendering */
-    case GL_PROXY_TEXTURE_2D:
-        {
-            struct gl_texture_object *tObj;
-            tdfxTexInfo *ti;
-            int memNeeded;
-
-            tObj = ctx->Texture.Proxy2D;
-            if (!tObj->DriverData)
-                tObj->DriverData = fxAllocTexObjData(fxMesa);
-            ti = TDFX_TEXTURE_DATA(tObj);
-
-            /* assign the parameters to test against */
-            tObj->Image[level]->Width = width;
-            tObj->Image[level]->Height = height;
-            tObj->Image[level]->Border = border;
-            tObj->Image[level]->IntFormat = internalFormat;
-            if (level == 0) {
-               /* don't use mipmap levels > 0 */
-               tObj->MinFilter = tObj->MagFilter = GL_NEAREST;
-            }
-            else {
-               /* test with all mipmap levels */
-               tObj->MinFilter = GL_LINEAR_MIPMAP_LINEAR;
-               tObj->MagFilter = GL_NEAREST;
-            }
-            RevalidateTexture(ctx, tObj);
-
-            /*
-            printf("small lodlog2 0x%x\n", ti->info.smallLodLog2);
-            printf("large lodlog2 0x%x\n", ti->info.largeLodLog2);
-            printf("aspect ratio 0x%x\n", ti->info.aspectRatioLog2);
-            printf("glide format 0x%x\n", ti->info.format);
-            printf("data %p\n", ti->info.data);
-            printf("lodblend %d\n", (int) ti->LODblend);
-            */
-
-            /* determine where texture will reside */
-            if (ti->LODblend && !shared->umaTexMemory) {
-                /* XXX GR_MIPMAPLEVELMASK_BOTH might not be right, but works */
-                memNeeded = FX_grTexTextureMemRequired_NoLock(
-                                        GR_MIPMAPLEVELMASK_BOTH, &(ti->info));
-            }
-            else {
-                /* XXX GR_MIPMAPLEVELMASK_BOTH might not be right, but works */
-                memNeeded = FX_grTexTextureMemRequired_NoLock(
-                                        GR_MIPMAPLEVELMASK_BOTH, &(ti->info));
-            }
-            /*
-            printf("Proxy test %d > %d\n", memNeeded, shared->totalTexMem[0]);
-            */
-            if (memNeeded > shared->totalTexMem[0])
-                return GL_FALSE;
-            else
-                return GL_TRUE;
-        }
-    case GL_PROXY_TEXTURE_3D:
-        return GL_TRUE;  /* software rendering */
-    default:
-        return GL_TRUE;  /* never happens, silence compiler */
-    }
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   tdfxTexObjPtr t;
+
+   if ( tObj ) {
+      /* Per-texture palette */
+      if ( !tObj->DriverData )
+	 tObj->DriverData = tdfxAllocTexObj(fxMesa);
+
+      t = TDFX_TEXTURE_DATA(tObj);
+      tdfxConvertPalette( t->palette.data, &tObj->Palette );
+      /*tdfxTexInvalidate( ctx, tObj );*/
+   } else {
+      /* Global texture palette */
+      tdfxConvertPalette( fxMesa->glbPalette.data, &ctx->Texture.Palette );
+   }
+
+   fxMesa->new_state |= TDFX_NEW_TEXTURE; /* XXX too heavy-handed */
 }
 
 
-/*
- * Return a texture image to Mesa.  This is either to satisfy
- * a glGetTexImage() call or to prepare for software texturing.
- */
-GLvoid *
-tdfxDDGetTexImage(GLcontext * ctx, GLenum target, GLint level,
-                  const struct gl_texture_object *texObj,
-                  GLenum * formatOut, GLenum * typeOut,
-                  GLboolean * freeImageOut)
-{
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
-
-    if (target != GL_TEXTURE_2D)
-        return NULL;
-
-    if (!texObj->DriverData)
-        return NULL;
-
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[level];
-    if (mml->data) {
-        MesaIntTexFormat mesaFormat;
-        GLenum glFormat;
-        struct gl_texture_image *texImage = texObj->Image[level];
-        GLint srcStride;
-        void *uncompressedImage = NULL;
-
-        GLubyte *data =
-            (GLubyte *) MALLOC(texImage->Width * texImage->Height * 4);
-        if (!data)
-            return NULL;
-
-        uncompressedImage = (void *)mml->data;
-        switch (mml->glideFormat) {
-        case GR_TEXFMT_INTENSITY_8:
-            mesaFormat = MESA_I8;
-            glFormat = GL_INTENSITY;
-            srcStride = mml->width;
-            break;
-        case GR_TEXFMT_ALPHA_INTENSITY_88:
-            mesaFormat = MESA_A8_L8;
-            glFormat = GL_LUMINANCE_ALPHA;
-            srcStride = mml->width;
-            break;
-        case GR_TEXFMT_ALPHA_8:
-            if (texImage->Format == GL_INTENSITY) {
-                mesaFormat = MESA_I8;
-                glFormat = GL_INTENSITY;
-            }
-            else {
-                mesaFormat = MESA_A8;
-                glFormat = GL_ALPHA;
-            }
-            srcStride = mml->width;
-            break;
-        case GR_TEXFMT_RGB_565:
-            mesaFormat = MESA_R5_G6_B5;
-            glFormat = GL_RGB;
-            srcStride = mml->width * 2;
-            break;
-        case GR_TEXFMT_ARGB_8888:
-            mesaFormat = MESA_A8_R8_G8_B8;
-            glFormat = GL_RGBA;
-            srcStride = mml->width * 4;
-            break;
-        case GR_TEXFMT_ARGB_4444:
-            mesaFormat = MESA_A4_R4_G4_B4;
-            glFormat = GL_RGBA;
-            srcStride = mml->width * 2;
-            break;
-        case GR_TEXFMT_ARGB_1555:
-            mesaFormat = MESA_A1_R5_G5_B5;
-            glFormat = GL_RGBA;
-            srcStride = mml->width * 2;
-            break;
-        case GR_TEXFMT_P_8:
-            mesaFormat = MESA_C8;
-            glFormat = GL_COLOR_INDEX;
-            srcStride = mml->width;
-            break;
-        case GR_TEXFMT_ARGB_CMP_FXT1:
-            mesaFormat = MESA_A8_R8_G8_B8;
-            glFormat = GL_RGBA;
-            srcStride = mml->width * 4;
-           /*
-            * Allocate data for the uncompressed image,
-            * decompress the image.  The data will be deallocated
-            * after it is converted to the mesa format.
-            */
-            uncompressedImage = MALLOC(mml->width * mml->height * 4);
-            if (!uncompressedImage) {
-                gl_problem(NULL, "can't get memory in tdfxDDGetTexImage");
-                return NULL;
-            }
-            (*txImgDequantizeFXT1Proc)((FxU32 *)uncompressedImage,
-                                       (FxU32 *)mml->data,
-                                       mml->width,
-                                       mml->height);
-            break;
-        default:
-            gl_problem(NULL, "Bad glideFormat in tdfxDDGetTexImage");
-            return NULL;
-        }
-        _mesa_unconvert_teximage(mesaFormat, mml->width, mml->height,
-                                 uncompressedImage, srcStride, texImage->Width,
-                                 texImage->Height, glFormat, data);
-        if (uncompressedImage != mml->data) {
-            FREE(uncompressedImage);
-        }
-        *formatOut = glFormat;
-        *typeOut = GL_UNSIGNED_BYTE;
-        *freeImageOut = GL_TRUE;
-        return data;
-    }
-    else {
-        return NULL;
-    }
-}
 
-/*
- * This is called from _mesa_GetCompressedTexImage.  We just
- * copy out the compressed data.
- */
-void
-tdfxDDGetCompressedTexImage( GLcontext *ctx, GLenum target,
-                             GLint lod, void *image,
-                             const struct gl_texture_object *texObj,
-                             struct gl_texture_image *texImage )
-{
-    tdfxTexInfo *ti;
-    tdfxMipMapLevel *mml;
 
-    if (target != GL_TEXTURE_2D)
-        return;
+/**********************************************************************/
+/**** NEW TEXTURE IMAGE FUNCTIONS                                  ****/
+/**********************************************************************/
 
-    if (!texObj->DriverData)
-        return;
 
-    ti = TDFX_TEXTURE_DATA(texObj);
-    mml = &ti->mipmapLevel[lod];
-    if (mml->data) {
-        MEMCPY(image, mml->data, mml->dataSize);
-    }
-}
 
-/*
- * Calculate a specific texture format given a generic
- * texture format.
- */
-GLint
-tdfxDDSpecificCompressedTexFormat(GLcontext *ctx,
-                                  GLint internalFormat,
-                                  GLint numDimensions)
+#if 0
+static void
+PrintTexture(int w, int h, int c, const GLubyte * data)
 {
-    if (numDimensions != 2) {
-        return internalFormat;
-    }
-   /*
-    * If we don't have pointers to the functions, then
-    * we drop back to uncompressed format.  The logic
-    * in Mesa proper handles this for us.
-    *
-    * This is just to ease the transition to a Glide with
-    * the texus2 library.
-    */
-    if (!txImgQuantizeProc || !txImgDequantizeFXT1Proc) {
-        return internalFormat;
-    }
-    switch (internalFormat) {
-    case GL_COMPRESSED_RGB_ARB:
-        return GL_COMPRESSED_RGB_FXT1_3DFX;
-    case GL_COMPRESSED_RGBA_ARB:
-        return GL_COMPRESSED_RGBA_FXT1_3DFX;
-    }
-    return internalFormat;
+   int i, j;
+   for (i = 0; i < h; i++) {
+      for (j = 0; j < w; j++) {
+	 if (c == 2)
+	    printf("%02x %02x  ", data[0], data[1]);
+	 else if (c == 3)
+	    printf("%02x %02x %02x  ", data[0], data[1], data[2]);
+	 data += c;
+      }
+      printf("\n");
+   }
 }
+#endif
 
-/*
- * Calculate a specific texture format given a generic
- * texture format.
- */
-GLint
-tdfxDDBaseCompressedTexFormat(GLcontext *ctx,
-                              GLint internalFormat)
+
+static GLboolean
+tdfxDDTestProxyTexImage( GLcontext *ctx, GLenum target,
+			 GLint level, GLint internalFormat,
+			 GLenum format, GLenum type,
+			 GLint width, GLint height,
+			 GLint depth, GLint border )
 {
-    switch (internalFormat) {
-    case GL_COMPRESSED_RGB_FXT1_3DFX:
-        return GL_RGB;
-    case GL_COMPRESSED_RGBA_FXT1_3DFX:
-        return GL_RGBA;
-    }
-    return -1;
+   tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+
+   switch (target) {
+   case GL_PROXY_TEXTURE_1D:
+      return GL_TRUE;  /* software rendering */
+   case GL_PROXY_TEXTURE_2D:
+   {
+      struct gl_texture_object *tObj;
+      tdfxTexObjPtr t;
+      int memNeeded;
+
+      tObj = ctx->Texture.Proxy2D;
+      if (!tObj->DriverData)
+	 tObj->DriverData = tdfxAllocTexObj(fxMesa);
+      t = TDFX_TEXTURE_DATA(tObj);
+
+      /* assign the parameters to test against */
+      tObj->Image[level]->Width = width;
+      tObj->Image[level]->Height = height;
+      tObj->Image[level]->Border = border;
+      tObj->Image[level]->IntFormat = internalFormat;
+      if (level == 0) {
+	 /* don't use mipmap levels > 0 */
+	 tObj->MinFilter = tObj->MagFilter = GL_NEAREST;
+      }
+      else {
+	 /* test with all mipmap levels */
+	 tObj->MinFilter = GL_LINEAR_MIPMAP_LINEAR;
+	 tObj->MagFilter = GL_NEAREST;
+      }
+      tdfxRevalidateTexture(ctx, tObj);
+
+      /*
+	printf("small lodlog2 0x%x\n", t->info.smallLodLog2);
+	printf("large lodlog2 0x%x\n", t->info.largeLodLog2);
+	printf("aspect ratio 0x%x\n", t->info.aspectRatioLog2);
+	printf("glide format 0x%x\n", t->info.format);
+	printf("data %p\n", t->info.data);
+	printf("lodblend %d\n", (int) t->LODblend);
+      */
+
+      /* determine where texture will reside */
+      if (t->LODblend && !tss->umaTexMemory) {
+	 /* XXX GR_MIPMAPLEVELMASK_BOTH might not be right, but works */
+	 memNeeded = FX_grTexTextureMemRequired_NoLock(
+	    GR_MIPMAPLEVELMASK_BOTH, &(t->info));
+      }
+      else {
+	 /* XXX GR_MIPMAPLEVELMASK_BOTH might not be right, but works */
+	 memNeeded = FX_grTexTextureMemRequired_NoLock(
+	    GR_MIPMAPLEVELMASK_BOTH, &(t->info));
+      }
+      /*
+	printf("Proxy test %d > %d\n", memNeeded, tss->totalTexMem[0]);
+      */
+      if (memNeeded > tss->totalTexMem[0])
+	 return GL_FALSE;
+      else
+	 return GL_TRUE;
+   }
+   case GL_PROXY_TEXTURE_3D:
+      return GL_TRUE;  /* software rendering */
+   default:
+      return GL_TRUE;  /* never happens, silence compiler */
+   }
 }
 
-/*
- * Tell us if an image is compressed.  The real work is done
- * in a macro, but we need to have a function to create a
- * function pointer.
+
+/* Return a texture image to Mesa.  This is either to satisfy
+ * a glGetTexImage() call or to prepare for software texturing.
  */
-GLboolean
-tdfxDDIsCompressedFormat(GLcontext *ctx, GLint internalFormat)
+static GLvoid *
+tdfxDDGetTexImage( GLcontext *ctx, GLenum target, GLint level,
+		   const struct gl_texture_object *texObj,
+		   GLenum *formatOut, GLenum *typeOut,
+		   GLboolean *freeImageOut )
 {
-    return tdfxDDIsCompressedFormatMacro(internalFormat);
+   const struct gl_texture_image *texImage = texObj->Image[level];
+   const struct gl_texture_format *texFormat = texImage->TexFormat;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+   tdfxTexImagePtr image;
+   GLubyte *data;
+
+   if ( target != GL_TEXTURE_2D )
+      return NULL;
+   if ( !t )
+      return NULL;
+
+   image = &t->image[level];
+   if ( !image->original.data )
+      return NULL;
+
+   data = (GLubyte *) MALLOC( texImage->Width * texImage->Height * 4 );
+   if ( !data )
+      return NULL;
+
+   _mesa_unconvert_teximage2d( texFormat->IntFormat, texImage->Format,
+			       texImage->Width, texImage->Height,
+			       image->original.data, data );
+
+   *formatOut = texImage->Format;
+   *typeOut = GL_UNSIGNED_BYTE;
+   *freeImageOut = GL_TRUE;
+
+   return data;
 }
 
 
-/*
- * Calculate the image size of a compressed texture.
- *
- * The current compressed format, the FXT1 family, all
- * map 8x32 texel blocks into 128 bits.
- *
- * We return 0 if we can't calculate the size.
- *
- * Glide would report this out to us, but we don't have
- * exactly the right parameters.
- */
-GLsizei
-tdfxDDCompressedImageSize(GLcontext *ctx,
-                          GLenum intFormat,
-                          GLuint numDimensions,
-                          GLuint width,
-                          GLuint height,
-                          GLuint depth)
+void tdfxDDInitTextureFuncs( GLcontext *ctx )
 {
-    if (numDimensions != 2) {
-        return 0;
-    }
-    switch (intFormat) {
-    case GL_COMPRESSED_RGB_FXT1_3DFX:
-    case GL_COMPRESSED_RGBA_FXT1_3DFX:
-       /*
-        * Round height and width to multiples of 4 and 8,
-        * divide the resulting product by 32 to get the number
-        * of blocks, and multiply by 32 = 128/8 to get the.
-        * number of bytes required.  That is to say, just
-        * return the product.  Remember that we are returning
-        * bytes, not texels, so we have shrunk the texture
-        * by a factor of the texel size.
-        */
-        width = (width   + 0x7) &~ 0x7;
-        height = (height + 0x3) &~ 0x3;
-        return width * height;
-    }
-    return 0;
+   ctx->Driver.TexImage2D		= tdfxDDTexImage2D;
+   ctx->Driver.TexSubImage2D		= tdfxDDTexSubImage2D;
+   ctx->Driver.GetTexImage		= tdfxDDGetTexImage;
+   ctx->Driver.TexEnv			= tdfxDDTexEnv;
+   ctx->Driver.TexParameter		= tdfxDDTexParameter;
+   ctx->Driver.BindTexture		= tdfxDDBindTexture;
+   ctx->Driver.DeleteTexture		= tdfxDDDeleteTexture;
+   ctx->Driver.IsTextureResident	= tdfxDDIsTextureResident;
+   ctx->Driver.UpdateTexturePalette	= tdfxDDTexturePalette;
 }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.h
index 95ef3b018..1132f3f35 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.h
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_tex.h
@@ -35,123 +35,12 @@
  *
  */
 
-#ifndef _TDFX_TEX_H_
-#define _TDFX_TEX_H_
+#ifndef __TDFX_TEX_H__
+#define __TDFX_TEX_H__
 
+extern void tdfxUpdateTextureState( GLcontext *ctx );
+extern void tdfxUpdateTextureBinding( GLcontext *ctx );
 
-#include "texutil.h"
-
-
-#define tdfxDDIsCompressedFormatMacro(internalFormat) \
-    (((internalFormat) == GL_COMPRESSED_RGB_FXT1_3DFX) || \
-     ((internalFormat) == GL_COMPRESSED_RGBA_FXT1_3DFX))
-#define tdfxDDIsCompressedGlideFormatMacro(internalFormat) \
-    ((internalFormat) == GR_TEXFMT_ARGB_CMP_FXT1)
-
-
-
-extern void
-tdfxTexValidate(GLcontext * ctx, struct gl_texture_object *tObj);
-
-extern void
-tdfxDDBindTexture(GLcontext * ctx, GLenum target,
-                  struct gl_texture_object *tObj);
-
-extern void
-tdfxDDDeleteTexture(GLcontext * ctx, struct gl_texture_object *tObj);
-
-extern GLboolean
-tdfxDDIsTextureResident(GLcontext *ctx, struct gl_texture_object *tObj);
-
-extern void
-tdfxDDTexturePalette(GLcontext * ctx, struct gl_texture_object *tObj);
-
-#if 000 /* DEAD? */
-extern void
-fxDDTexUseGlobalPalette(GLcontext * ctx, GLboolean state);
-#endif
-
-extern void
-tdfxDDTexEnv(GLcontext * ctx, GLenum target, GLenum pname,
-             const GLfloat * param);
-
-extern void
-tdfxDDTexParameter(GLcontext * ctx, GLenum target,
-                   struct gl_texture_object *tObj,
-                   GLenum pname, const GLfloat * params);
-
-extern GLboolean
-tdfxDDTexImage2D(GLcontext * ctx, GLenum target, GLint level,
-                 GLenum format, GLenum type, const GLvoid * pixels,
-                 const struct gl_pixelstore_attrib * packing,
-                 struct gl_texture_object * texObj,
-                 struct gl_texture_image * texImage,
-                 GLboolean * retainInternalCopy);
-
-extern GLboolean
-tdfxDDTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
-                    GLint xoffset, GLint yoffset,
-                    GLsizei width, GLsizei height,
-                    GLenum format, GLenum type, const GLvoid * pixels,
-                    const struct gl_pixelstore_attrib * packing,
-                    struct gl_texture_object * texObj,
-                    struct gl_texture_image * texImage);
-
-extern GLboolean
-tdfxDDCompressedTexImage2D( GLcontext *ctx, GLenum target,
-                            GLint level, GLsizei imageSize,
-                            const GLvoid *data,
-                            struct gl_texture_object *texObj,
-                            struct gl_texture_image *texImage,
-                            GLboolean *retainInternalCopy);
-
-extern GLboolean
-tdfxDDCompressedTexSubImage2D( GLcontext *ctx, GLenum target,
-                               GLint level, GLint xoffset,
-                               GLint yoffset, GLsizei width,
-                               GLint height, GLenum format,
-                               GLsizei imageSize, const GLvoid *data,
-                               struct gl_texture_object *texObj,
-                               struct gl_texture_image *texImage );
-
-extern GLboolean
-tdfxDDTestProxyTexImage(GLcontext *ctx, GLenum target,
-                        GLint level, GLint internalFormat,
-                        GLenum format, GLenum type,
-                        GLint width, GLint height,
-                        GLint depth, GLint border);
-
-extern GLvoid *
-tdfxDDGetTexImage(GLcontext * ctx, GLenum target, GLint level,
-                  const struct gl_texture_object *texObj,
-                  GLenum * formatOut, GLenum * typeOut,
-                  GLboolean * freeImageOut);
-
-extern void
-tdfxDDGetCompressedTexImage( GLcontext *ctx, GLenum target,
-                             GLint lod, void *image,
-                             const struct gl_texture_object *texObj,
-                             struct gl_texture_image *texImage );
-
-extern GLint
-tdfxDDSpecificCompressedTexFormat(GLcontext *ctx,
-                                  GLint internalFormat,
-                                  GLint numDimensions);
-
-extern GLint
-tdfxDDBaseCompressedTexFormat(GLcontext *ctx,
-                              GLint internalFormat);
-
-extern GLboolean
-tdfxDDIsCompressedFormat(GLcontext *ctx, GLint internalFormat);
-
-extern GLsizei
-tdfxDDCompressedImageSize(GLcontext *ctx,
-                          GLenum intFormat,
-                          GLuint numDimensions,
-                          GLuint width,
-                          GLuint height,
-                          GLuint depth);
-
+extern void tdfxDDInitTextureFuncs( GLcontext *ctx );
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.c
index 97fe9d4c6..8ca93f91a 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.c
@@ -39,930 +39,882 @@
 #include "tdfx_tex.h"
 #include "tdfx_texman.h"
 
+#define BAD_ADDRESS	((FxU32) -1)
 
-#define BAD_ADDRESS ((FxU32) -1)
-
-
-#if 0 /* DEBUG use */
-/*
- * Verify the consistancy of the texture memory manager.
+/* Verify the consistancy of the texture memory manager.
  * This involves:
  *    Traversing all texture objects and computing total memory used.
  *    Traverse the free block list and computing total memory free.
  *    Compare the total free and total used amounts to the total memory size.
  *    Make various assertions about the results.
  */
-static void
-VerifyFreeList(tdfxContextPtr fxMesa, FxU32 tmu)
+static void tdfxTMVerifyFreeList( tdfxContextPtr fxMesa, FxU32 unit )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxMemRange *block;
-    int prevStart = -1, prevEnd = -1;
-    int totalFree = 0;
-    int numObj = 0, numRes = 0;
-    int totalUsed = 0;
-
-    for (block = shared->tmFree[tmu]; block; block = block->next) {
-       assert( block->endAddr > 0 );
-       assert( block->startAddr <= shared->totalTexMem[tmu] );
-       assert( block->endAddr <= shared->totalTexMem[tmu] );
-       assert( (int) block->startAddr > prevStart );
-       assert( (int) block->startAddr >= prevEnd );
-       prevStart = (int) block->startAddr;
-       prevEnd = (int) block->endAddr;
-       totalFree += (block->endAddr - block->startAddr);
-    }
-    assert(totalFree == shared->freeTexMem[tmu]);
-
-    {
-       struct gl_texture_object *obj;
-       for (obj = mesaShared->TexObjectList; obj; obj = obj->Next) {
-          tdfxTexInfo *ti = TDFX_TEXTURE_DATA(obj);
-          numObj++;
-          if (ti) {
-             if (ti->isInTM) {
-                numRes++;
-                assert(ti->tm[0]);
-                if (ti->tm[tmu])
-                   totalUsed += (ti->tm[tmu]->endAddr - ti->tm[tmu]->startAddr);
-             }
-             else {
-                assert(!ti->tm[0]);
-             }
-          }
-       }
-    }
-
-    printf("totalFree: %d  totalUsed: %d  totalMem: %d #objs=%d  #res=%d\n",
-           shared->freeTexMem[tmu], totalUsed, shared->totalTexMem[tmu],
-           numObj, numRes);
-
-    assert(totalUsed + totalFree == shared->totalTexMem[tmu]);
-}
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   struct gl_texture_object *texObj;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   tdfxMemRange *block;
+   int prevStart = -1, prevEnd = -1;
+   int totalFree = 0;
+   int numObj = 0, numRes = 0;
+   int totalUsed = 0;
+
+   for ( block = tss->freeRanges[unit] ; block ; block = block->next ) {
+      assert( block->endAddr > 0 );
+      assert( block->startAddr <= tss->totalTexMem[unit] );
+      assert( block->endAddr <= tss->totalTexMem[unit] );
+      assert( (int) block->startAddr > prevStart );
+      assert( (int) block->startAddr >= prevEnd );
+      prevStart = (int) block->startAddr;
+      prevEnd = (int) block->endAddr;
+      totalFree += (block->endAddr - block->startAddr);
+   }
+   assert( totalFree == tss->freeTexMem[unit] );
+
+   for ( texObj = ss->TexObjectList ; texObj ; texObj = texObj->Next ) {
+      tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+      numObj++;
+      if ( t ) {
+	 if ( t->isInTM ) {
+	    numRes++;
+	    assert( t->range[0] );
+	    if ( t->range[unit] )
+	       totalUsed += (t->range[unit]->endAddr - t->range[unit]->startAddr);
+	 } else {
+	    assert(!t->range[0]);
+	 }
+      }
+   }
 
+   fprintf( stderr,
+	    "totalFree: %d  totalUsed: %d  totalMem: %d #objs=%d  #res=%d\n",
+	    tss->freeTexMem[unit], totalUsed, tss->totalTexMem[unit],
+	    numObj, numRes );
+
+   assert( totalUsed + totalFree == tss->totalTexMem[unit] );
+}
 
-static void
-dump_texmem(tdfxContextPtr fxMesa)
+static void tdfxTMDumpTexMem( tdfxContextPtr fxMesa )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    struct gl_texture_object *oldestObj, *obj, *lowestPriorityObj;
-    tdfxMemRange *r;
-    FxU32 prev;
-
-    printf("DUMP Objects:\n");
-    for (obj = mesaShared->TexObjectList; obj; obj = obj->Next) {
-        tdfxTexInfo *info = TDFX_TEXTURE_DATA(obj);
-
-        if (info && info->isInTM) {
-        printf("Obj %8p: %4d  info = %p\n", obj, obj->Name, info);
-
-           printf("  isInTM=%d  whichTMU=%d  lastTimeUsed=%d\n",
-                  info->isInTM, info->whichTMU, info->lastTimeUsed);
-           printf("    tm[0] = %p", info->tm[0]);
-           assert(info->tm[0]);
-           if (info->tm[0]) {
-              printf("  tm startAddr = %d  endAddr = %d",
-                     info->tm[0]->startAddr,
-                     info->tm[0]->endAddr);
-           }
-           printf("\n");
-           printf("    tm[1] = %p", info->tm[1]);
-           if (info->tm[1]) {
-              printf("  tm startAddr = %d  endAddr = %d",
-                     info->tm[1]->startAddr,
-                     info->tm[1]->endAddr);
-           }
-           printf("\n");
-        }
-    }
-
-    VerifyFreeList(fxMesa, 0);
-    VerifyFreeList(fxMesa, 1);
-
-    printf("Free memory unit 0:  %d bytes\n", shared->freeTexMem[0]);
-    prev = 0;
-    for (r = shared->tmFree[0]; r; r = r->next) {
-       printf("%8p:  start %8d  end %8d  size %8d  gap %8d\n", r, r->startAddr, r->endAddr, r->endAddr - r->startAddr, r->startAddr - prev);
-       prev = r->endAddr;
-    }
-
-    printf("Free memory unit 1:  %d bytes\n", shared->freeTexMem[1]);
-    prev = 0;
-    for (r = shared->tmFree[1]; r; r = r->next) {
-       printf("%8p:  start %8d  end %8d  size %8d  gap %8d\n", r, r->startAddr, r->endAddr, r->endAddr - r->startAddr, r->startAddr - prev);
-       prev = r->endAddr;
-    }
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   struct gl_texture_object *texObj;
+   tdfxMemRange *r;
+   FxU32 prev;
+
+   printf( "DUMP Objects:\n" );
+   for ( texObj = ss->TexObjectList ; texObj ; texObj = texObj->Next ) {
+      tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+
+      if ( t && t->isInTM ) {
+	 printf( "Obj %8p: %4d  info = %p\n", texObj, texObj->Name, t );
+
+	 printf( "  isInTM=%d  whichTMU=%ld  lastTimeUsed=%d\n",
+		 t->isInTM, t->whichTMU, t->lastTimeUsed );
+	 printf( "    tm[0] = %p", t->range[0] );
+	 assert( t->range[0] );
+	 if ( t->range[0] ) {
+	    printf( "  tm startAddr = %ld  endAddr = %ld",
+		    t->range[0]->startAddr,
+		    t->range[0]->endAddr );
+	 }
+	 printf( "\n" );
+	 printf( "    tm[1] = %p", t->range[1] );
+	 if ( t->range[1] ) {
+	    printf( "  tm startAddr = %ld  endAddr = %ld",
+		    t->range[1]->startAddr,
+		    t->range[1]->endAddr );
+	 }
+	 printf( "\n" );
+      }
+   }
 
-}
-#endif
+   tdfxTMVerifyFreeList( fxMesa, 0 );
+   tdfxTMVerifyFreeList( fxMesa, 1 );
+
+   printf( "Free memory unit 0:  %d bytes\n", tss->freeTexMem[0] );
+   prev = 0;
+   for ( r = tss->freeRanges[0] ; r ; r = r->next ) {
+      printf( "%8p:  start %8ld  end %8ld  size %8ld  gap %8ld\n",
+	      r, r->startAddr, r->endAddr, r->endAddr - r->startAddr,
+	      r->startAddr - prev );
+      prev = r->endAddr;
+   }
 
+   printf( "Free memory unit 1:  %d bytes\n", tss->freeTexMem[1] );
+   prev = 0;
+   for ( r = tss->freeRanges[1] ; r ; r = r->next ) {
+      printf( "%8p:  start %8ld  end %8ld  size %8ld  gap %8ld\n",
+	      r, r->startAddr, r->endAddr, r->endAddr - r->startAddr,
+	      r->startAddr - prev );
+      prev = r->endAddr;
+   }
+}
 
 
 #ifdef TEXSANITY
-static void
-fubar(void)
+static void fubar( void )
 {
+   /* GH: What am I meant to do??? */
 }
 
-/*
- * Sanity Check
+/* Sanity Check
  */
-static void
-sanity(tdfxContextPtr fxMesa)
+static void sanity( tdfxContextPtr fxMesa )
 {
-    tdfxMemRange *tmp, *prev, *pos;
-
-    prev = 0;
-    tmp = fxMesa->tmFree[0];
-    while (tmp) {
-        if (!tmp->startAddr && !tmp->endAddr) {
-            fprintf(stderr, "Textures fubar\n");
-            fubar();
-        }
-        if (tmp->startAddr >= tmp->endAddr) {
-            fprintf(stderr, "Node fubar\n");
-            fubar();
-        }
-        if (prev && (prev->startAddr >= tmp->startAddr ||
-                     prev->endAddr > tmp->startAddr)) {
-            fprintf(stderr, "Sorting fubar\n");
-            fubar();
-        }
-        prev = tmp;
-        tmp = tmp->next;
-    }
-    prev = 0;
-    tmp = fxMesa->tmFree[1];
-    while (tmp) {
-        if (!tmp->startAddr && !tmp->endAddr) {
-            fprintf(stderr, "Textures fubar\n");
-            fubar();
-        }
-        if (tmp->startAddr >= tmp->endAddr) {
-            fprintf(stderr, "Node fubar\n");
-            fubar();
-        }
-        if (prev && (prev->startAddr >= tmp->startAddr ||
-                     prev->endAddr > tmp->startAddr)) {
-            fprintf(stderr, "Sorting fubar\n");
-            fubar();
-        }
-        prev = tmp;
-        tmp = tmp->next;
-    }
+   tdfxMemRange *tmp, *prev, *pos;
+
+   prev = 0;
+   tmp = fxMesa->freeRanges[0];
+   while ( tmp ) {
+      if ( !tmp->startAddr && !tmp->endAddr ) {
+	 fprintf( stderr, "Textures fubar\n" );
+	 fubar();
+      }
+      if ( tmp->startAddr >= tmp->endAddr ) {
+	 fprintf( stderr, "Node fubar\n" );
+	 fubar();
+      }
+      if ( prev && ( prev->startAddr >= tmp->startAddr ||
+		     prev->endAddr > tmp->startAddr ) ) {
+	 fprintf( stderr, "Sorting fubar\n" );
+	 fubar();
+      }
+      prev = tmp;
+      tmp = tmp->next;
+   }
+
+   prev = 0;
+   tmp = fxMesa->freeRanges[1];
+   while ( tmp ) {
+      if ( !tmp->startAddr && !tmp->endAddr ) {
+	 fprintf( stderr, "Textures fubar\n" );
+	 fubar();
+      }
+      if ( tmp->startAddr >= tmp->endAddr ) {
+	 fprintf( stderr, "Node fubar\n" );
+	 fubar();
+      }
+      if ( prev && ( prev->startAddr >= tmp->startAddr ||
+		     prev->endAddr > tmp->startAddr ) ) {
+	 fprintf( stderr, "Sorting fubar\n" );
+	 fubar();
+      }
+      prev = tmp;
+      tmp = tmp->next;
+   }
 }
 #endif
 
 
-
-
-
-/*
- * Allocate and initialize a new MemRange struct.
- * Try to allocate it from the pool of free MemRange nodes rather than malloc.
+/* Allocate and initialize a new MemRange struct.  Try to allocate it
+ * from the pool of free MemRange nodes rather than malloc.
  */
 static tdfxMemRange *
-NewRangeNode(tdfxContextPtr fxMesa, FxU32 start, FxU32 end)
+tdfxTMNewRangeNode( tdfxContextPtr fxMesa, FxU32 start, FxU32 end )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxMemRange *result;
-
-    _glthread_LOCK_MUTEX(mesaShared->Mutex);
-    if (shared && shared->tmPool) {
-        result = shared->tmPool;
-        shared->tmPool = shared->tmPool->next;
-    }
-    else {
-        result = MALLOC(sizeof(tdfxMemRange));
-
-    }
-    _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-
-    if (!result) {
-        /*fprintf(stderr, "fxDriver: out of memory!\n");*/
-        return NULL;
-    }
-
-    result->startAddr = start;
-    result->endAddr = end;
-    result->next = NULL;
-
-    return result;
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   tdfxMemRange *range;
+
+   _glthread_LOCK_MUTEX( ss->Mutex );
+   if ( tss && tss->rangePool ) {
+      range = tss->rangePool;
+      tss->rangePool = tss->rangePool->next;
+   } else {
+      range = MALLOC( sizeof(tdfxMemRange) );
+   }
+   _glthread_UNLOCK_MUTEX( ss->Mutex );
+
+   if ( !range ) {
+      if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf( stderr, __FUNCTION__ ": out of memory!\n" );
+      return NULL;
+   }
+
+   range->startAddr = start;
+   range->endAddr = end;
+   range->next = NULL;
+
+   return range;
 }
 
 
-/*
- * Initialize texture memory.
- * We take care of one or both TMU's here.
+/* Initialize texture memory.  We take care of one or both TMU's here.
  */
-void
-tdfxTMInit(tdfxContextPtr fxMesa)
+void tdfxTMInit( tdfxContextPtr fxMesa )
 {
-    if (!fxMesa->glCtx->Shared->DriverData) {
-        const char *extensions;
-        struct tdfxSharedState *shared = CALLOC_STRUCT(tdfxSharedState);
-        if (!shared)
-           return;
-
-        extensions = FX_grGetString(fxMesa, GR_EXTENSION);
-        if (strstr(extensions, "TEXUMA")) {
-            FxU32 start, end;
-            shared->umaTexMemory = GL_TRUE;
-            FX_grEnable(fxMesa, GR_TEXTURE_UMA_EXT);
-            start = FX_grTexMinAddress(fxMesa, 0);
-            end = FX_grTexMaxAddress(fxMesa, 0);
-            shared->totalTexMem[0] = end - start;
-            shared->totalTexMem[1] = 0;
-            shared->freeTexMem[0] = end - start;
-            shared->freeTexMem[1] = 0;
-            shared->tmFree[0] = NewRangeNode(fxMesa, start, end);
-            shared->tmFree[1] = NULL;
-            /*printf("UMA tex memory: %d\n", (int) (end - start));*/
-        }
-        else {
-            const int numTMUs = fxMesa->haveTwoTMUs ? 2 : 1;
-            int tmu;
-            shared->umaTexMemory = GL_FALSE;
-            for (tmu = 0; tmu < numTMUs; tmu++) {
-                FxU32 start = FX_grTexMinAddress(fxMesa, tmu);
-                FxU32 end = FX_grTexMaxAddress(fxMesa, tmu);
-                shared->totalTexMem[tmu] = end - start;
-                shared->freeTexMem[tmu] = end - start;
-                shared->tmFree[tmu] = NewRangeNode(fxMesa, start, end);
-                /*printf("Split tex memory: %d\n", (int) (end - start));*/
-            }
-        }
-
-        shared->tmPool = NULL;
-        fxMesa->glCtx->Shared->DriverData = shared;
-        /*printf("Texture memory init UMA: %d\n", shared->umaTexMemory);*/
-    }
+   GLcontext *ctx = fxMesa->glCtx;
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+      fprintf( stderr, __FUNCTION__ "\n" );
+
+   if ( !ctx->Shared->DriverData ) {
+      const char *extensions;
+      tdfxSharedStatePtr tss = CALLOC_STRUCT( tdfx_shared_state );
+
+      if ( !tss )
+	 return;
+
+      LOCK_HARDWARE( fxMesa );
+
+      extensions = grGetString( GR_EXTENSION );
+
+      if ( strstr( extensions, " TEXUMA " ) ) {
+	 FxU32 start, end;
+
+	 tss->umaTexMemory = GL_TRUE;
+
+	 grEnable( GR_TEXTURE_UMA_EXT );
+
+	 start = grTexMinAddress( 0 );
+	 end = grTexMaxAddress( 0 );
+
+	 if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	    fprintf( stderr, "   UMA tex memory: %d\n", (int)(end - start) );
+
+	 tss->totalTexMem[0] = end - start;
+	 tss->totalTexMem[1] = 0;
+	 tss->freeTexMem[0] = end - start;
+	 tss->freeTexMem[1] = 0;
+	 tss->freeRanges[0] = tdfxTMNewRangeNode( fxMesa, start, end );
+	 tss->freeRanges[1] = NULL;
+      } else {
+	 int unit;
+
+	 tss->umaTexMemory = GL_FALSE;
+
+	 for ( unit = 0 ; unit < fxMesa->numTMUs ; unit++ ) {
+	    FxU32 start, end;
+
+	    start = grTexMinAddress( unit );
+	    end = grTexMaxAddress( unit );
+
+	    tss->totalTexMem[unit] = end - start;
+	    tss->freeTexMem[unit] = end - start;
+	    tss->freeRanges[unit] = tdfxTMNewRangeNode( fxMesa, start, end );
+
+	    if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	       fprintf( stderr, "   Split tex memory: %d\n",
+			(int)(end - start) );
+	 }
+      }
+
+      UNLOCK_HARDWARE( fxMesa );
+
+      tss->rangePool = NULL;
+      ctx->Shared->DriverData = tss;
+
+      if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf( stderr, "  init UMA: %d\n", tss->umaTexMemory );
+   }
 }
 
 
-/*
- * Clean-up texture memory before destroying context.
+/* Clean-up texture memory before destroying context.
  */
-void
-tdfxTMClose(tdfxContextPtr fxMesa)
+void tdfxTMClose( tdfxContextPtr fxMesa )
 {
-    if (fxMesa->glCtx->Shared->RefCount == 1) {
-        /* refcount will soon go to zero, free our 3dfx stuff */
-        struct tdfxSharedState *shared = (struct tdfxSharedState *) fxMesa->glCtx->Shared->DriverData;
-
-        const int numTMUs = fxMesa->haveTwoTMUs ? 2 : 1;
-        int tmu;
-        tdfxMemRange *tmp, *next;
-
-        /* Deallocate the pool of free tdfxMemRange nodes */
-        tmp = shared->tmPool;
-        while (tmp) {
-            next = tmp->next;
-            FREE(tmp);
-            tmp = next;
-        }
-
-        /* Delete the texture memory block tdfxMemRange nodes */
-        for (tmu = 0; tmu < numTMUs; tmu++) {
-            tmp = shared->tmFree[tmu];
-            while (tmp) {
-                next = tmp->next;
-                FREE(tmp);
-                tmp = next;
-            }
-        }
-
-        FREE(shared);
-        fxMesa->glCtx->Shared->DriverData = NULL;
-    }
-}
+   GLcontext *ctx = fxMesa->glCtx;
 
+   if ( ctx->Shared->RefCount == 1 ) {
+      /* RefCount will soon go to zero, free our 3dfx stuff */
+      tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ctx->Shared->DriverData;
+      int unit;
+      tdfxMemRange *tmp, *next;
+
+      /* Deallocate the pool of free tdfxMemRange nodes */
+      tmp = tss->rangePool;
+      while ( tmp ) {
+	 next = tmp->next;
+	 FREE( tmp );
+	 tmp = next;
+      }
 
+      /* Delete the texture memory block tdfxMemRange nodes */
+      for ( unit = 0 ; unit < fxMesa->numTMUs ; unit++ ) {
+	 tmp = tss->freeRanges[unit];
+	 while ( tmp ) {
+	    next = tmp->next;
+	    FREE( tmp );
+	    tmp = next;
+	 }
+      }
 
-/*
- * Delete a tdfxMemRange struct.
- * We keep a linked list of free/available tdfxMemRange structs to
- * avoid extra malloc/free calls.
- */
-#if 0
-static void
-DeleteRangeNode_NoLock(struct TdfxSharedState *shared, tdfxMemRange *range)
-{
-    /* insert at head of list */
-    range->next = shared->tmPool;
-    shared->tmPool = range;
+      FREE( tss );
+      ctx->Shared->DriverData = NULL;
+   }
 }
-#endif
 
-#define DELETE_RANGE_NODE(shared, range) \
-    (range)->next = (shared)->tmPool;    \
-    (shared)->tmPool = (range)
 
 
+/* Delete a tdfxMemRange struct.
+ * We keep a linked list of free/available tdfxMemRange structs to
+ * avoid extra malloc/free calls.
+ */
+#define DELETE_RANGE_NODE( tss, range )					\
+do {									\
+   (range)->next = (tss)->rangePool;					\
+   (tss)->rangePool = (range);						\
+} while (0)
 
-/*
- * When we've run out of texture memory we have to throw out an
+/* When we've run out of texture memory we have to throw out an
  * existing texture to make room for the new one.  This function
  * determins the texture to throw out.
  */
 static struct gl_texture_object *
-FindOldestObject(tdfxContextPtr fxMesa, FxU32 tmu)
+tdfxTMFindOldestObject( tdfxContextPtr fxMesa, FxU32 unit )
 {
-    const GLuint bindnumber = fxMesa->texBindNumber;
-    struct gl_texture_object *oldestObj, *obj, *lowestPriorityObj;
-    GLfloat lowestPriority;
-    GLuint oldestAge;
-
-    oldestObj = NULL;
-    oldestAge = 0;
-
-    lowestPriority = 1.0F;
-    lowestPriorityObj = NULL;
-
-    for (obj = fxMesa->glCtx->Shared->TexObjectList; obj; obj = obj->Next) {
-        tdfxTexInfo *info = TDFX_TEXTURE_DATA(obj);
-
-        if (info && info->isInTM &&
-            ((info->whichTMU == tmu) || (info->whichTMU == TDFX_TMU_BOTH) ||
-             (info->whichTMU == TDFX_TMU_SPLIT))) {
-            GLuint age, lasttime;
-
-            assert(info->tm[0]);
-            lasttime = info->lastTimeUsed;
-
-            if (lasttime > bindnumber)
-                age = bindnumber + (UINT_MAX - lasttime + 1); /* TO DO: check wrap around */
-            else
-                age = bindnumber - lasttime;
-
-            if (age >= oldestAge) {
-                oldestAge = age;
-                oldestObj = obj;
-            }
-
-            /* examine priority */
-            if (obj->Priority < lowestPriority) {
-                lowestPriority = obj->Priority;
-                lowestPriorityObj = obj;
-            }
-        }
-    }
-
-    if (lowestPriority < 1.0) {
-        ASSERT(lowestPriorityObj);
-        /*
-        printf("discard %d pri=%f\n", lowestPriorityObj->Name, lowestPriority);
-        */
-        return lowestPriorityObj;
-    }
-    else {
-        /*
-        printf("discard %d age=%d\n", oldestObj->Name, oldestAge);
-        */
-        return oldestObj;
-    }
-}
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   const GLuint bindNumber = fxMesa->texBindNumber;
+   struct gl_texture_object *oldestObj, *texObj, *lowestPriorityObj;
+   GLfloat lowestPriority;
+   GLuint oldestAge;
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+      fprintf( stderr, __FUNCTION__ "\n" );
+
+   oldestObj = NULL;
+   oldestAge = 0;
+
+   lowestPriority = 1.0F;
+   lowestPriorityObj = NULL;
+
+   for ( texObj = ss->TexObjectList ; texObj ; texObj = texObj->Next ) {
+      tdfxTexObjPtr t = TDFX_TEXTURE_DATA(texObj);
+
+      if ( t && t->isInTM &&
+	   ( ( t->whichTMU == unit ) ||
+	     ( t->whichTMU == TDFX_TMU_BOTH ) ||
+	     ( t->whichTMU == TDFX_TMU_SPLIT ) ) ) {
+	 GLuint age, lastTime;
+
+	 assert( t->range[0] );
+	 lastTime = t->lastTimeUsed;
+
+	 if ( lastTime > bindNumber ) {
+	    /* TODO: check wrap around */
+	    age = bindNumber + (UINT_MAX - lastTime + 1);
+	 } else {
+	    age = bindNumber - lastTime;
+	 }
+	 if ( age >= oldestAge ) {
+	    oldestAge = age;
+	    oldestObj = texObj;
+	 }
 
+	 /* examine priority */
+	 if ( texObj->Priority < lowestPriority ) {
+	    lowestPriority = texObj->Priority;
+	    lowestPriorityObj = texObj;
+	 }
+      }
+   }
 
-#if 0
-static void
-FlushTexMemory(tdfxContextPtr fxMesa)
-{
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    struct gl_texture_object *obj;
-
-    for (obj = mesaShared->TexObjectList; obj; obj = obj->Next) {
-       if (obj->RefCount < 2) {
-          /* don't flush currently bound textures */
-          tdfxTMMoveOutTM_NoLock(fxMesa, obj);
-       }
-    }
+   if ( lowestPriority < 1.0 ) {
+      ASSERT( lowestPriorityObj );
+      if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf( stderr, "discard %d pri=%f\n",
+		  lowestPriorityObj->Name, lowestPriority );
+      return lowestPriorityObj;
+   } else {
+      if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf( stderr, "discard %d age=%d\n",
+		  oldestObj->Name, oldestAge );
+      return oldestObj;
+   }
 }
-#endif
 
 
-/*
- * Find the address (offset?) at which we can store a new texture.
- * <tmu> is the texture unit.
+/* Find the address (offset?) at which we can store a new texture.
+ * <unit> is the texture unit.
  * <size> is the texture size in bytes.
  */
-static FxU32
-FindStartAddr(tdfxContextPtr fxMesa, FxU32 tmu, FxU32 size)
+static FxU32 tdfxTMFindStartAddr( tdfxContextPtr fxMesa,
+				  FxU32 unit, FxU32 size )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxMemRange *prev, *block;
-    FxU32 result;
-#if 0
-    int discardedCount = 0;
-#define MAX_DISCARDS 10
-#endif
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   struct gl_texture_object *texObj;
+   tdfxMemRange *prev, *block;
+   FxU32 result;
+
+   if ( tss->umaTexMemory ) {
+      assert( unit == TDFX_TMU0 );
+   }
 
-    if (shared->umaTexMemory) {
-        assert(tmu == TDFX_TMU0);
-    }
-
-    _glthread_LOCK_MUTEX(mesaShared->Mutex);
-    while (1) {
-        prev = NULL;
-        block = shared->tmFree[tmu];
-        while (block) {
-            if (block->endAddr - block->startAddr >= size) {
-                /* The texture will fit here */
-                result = block->startAddr;
-                block->startAddr += size;
-                if (block->startAddr == block->endAddr) {
-                    /* Remove this node since it's empty */
-                    if (prev) {
-                        prev->next = block->next;
-                    }
-                    else {
-                        shared->tmFree[tmu] = block->next;
-                    }
-                    DELETE_RANGE_NODE(shared, block);
-                }
-                shared->freeTexMem[tmu] -= size;
-                _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-                return result;
-            }
-            prev = block;
-            block = block->next;
-        }
-        /* We failed to find a block large enough to accomodate <size> bytes.
-         * Find the oldest texObject and free it.
-         */
-#if 0
-        discardedCount++;
-        if (discardedCount > MAX_DISCARDS + 1) {
-            gl_problem(NULL, "tdfx driver: extreme texmem fragmentation");
-            _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-            return BAD_ADDRESS;
-        }
-        else if (discardedCount > MAX_DISCARDS) {
-            /* texture memory is probably really fragmented, flush it */
-            FlushTexMemory(fxMesa);
-        }
-        else
-#endif
-        {
-            struct gl_texture_object *obj = FindOldestObject(fxMesa, tmu);
-            if (obj) {
-                tdfxTMMoveOutTM_NoLock(fxMesa, obj);
-                fxMesa->stats.texSwaps++;
-            }
-            else {
-                gl_problem(NULL, "tdfx driver: extreme texmem fragmentation");
-                _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-                return BAD_ADDRESS;
-            }
-        }
-    }
-
-    /* never get here, but play it safe */
-    _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
-    return BAD_ADDRESS;
+   _glthread_LOCK_MUTEX( ss->Mutex );
+   while ( 1 ) {
+      prev = NULL;
+      block = tss->freeRanges[unit];
+
+      while ( block ) {
+	 if ( block->endAddr - block->startAddr >= size ) {
+	    /* The texture will fit here */
+	    result = block->startAddr;
+	    block->startAddr += size;
+	    if ( block->startAddr == block->endAddr ) {
+	       /* Remove this node since it's empty */
+	       if ( prev ) {
+		  prev->next = block->next;
+	       } else {
+		  tss->freeRanges[unit] = block->next;
+	       }
+	       DELETE_RANGE_NODE( tss, block );
+	    }
+	    tss->freeTexMem[unit] -= size;
+	    _glthread_UNLOCK_MUTEX( ss->Mutex );
+	    return result;
+	 }
+	 prev = block;
+	 block = block->next;
+      }
+
+      /* We failed to find a block large enough to accomodate <size> bytes.
+       * Find the oldest texObject and free it.
+       */
+      texObj = tdfxTMFindOldestObject( fxMesa, unit );
+      if ( texObj ) {
+	 tdfxTMMoveOutTMLocked( fxMesa, texObj );
+	 fxMesa->stats.texSwaps++;
+      } else {
+	 gl_problem( NULL, "tdfx driver: extreme texmem fragmentation" );
+	 _glthread_UNLOCK_MUTEX( ss->Mutex );
+	 return BAD_ADDRESS;
+      }
+   }
+
+   /* never get here, but play it safe */
+   _glthread_UNLOCK_MUTEX( ss->Mutex );
+   return BAD_ADDRESS;
 }
 
 
-/*
- * Remove the given tdfxMemRange node from hardware texture memory.
+/* Remove the given tdfxMemRange node from hardware texture memory.
  */
-static void
-RemoveRange_NoLock(tdfxContextPtr fxMesa, FxU32 tmu, tdfxMemRange *range)
+static void tdfxTMRemoveRangeLocked( tdfxContextPtr fxMesa,
+				     FxU32 unit, tdfxMemRange *range )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxMemRange *block, *prev;
-
-    if (shared->umaTexMemory) {
-       assert(tmu == TDFX_TMU0);
-    }
-
-    if (!range)
-        return;
-
-    if (range->startAddr == range->endAddr) {
-        DELETE_RANGE_NODE(shared, range);
-        return;
-    }
-    shared->freeTexMem[tmu] += range->endAddr - range->startAddr;
-
-    /* find position in linked list to insert this tdfxMemRange node */
-    prev = NULL;
-    block = shared->tmFree[tmu];
-    while (block) {
-        assert(range->startAddr != block->startAddr);
-        if (range->startAddr > block->startAddr) {
-            prev = block;
-            block = block->next;
-        }
-        else {
-            break;
-        }
-    }
-
-    /* Insert the free block, combine with adjacent blocks when possible */
-    range->next = block;
-    if (block) {
-        if (range->endAddr == block->startAddr) {
-            /* Combine */
-            block->startAddr = range->startAddr;
-            DELETE_RANGE_NODE(shared, range);
-            range = block;
-        }
-    }
-    if (prev) {
-        if (prev->endAddr == range->startAddr) {
-            /* Combine */
-            prev->endAddr = range->endAddr;
-            prev->next = range->next;
-            DELETE_RANGE_NODE(shared, range);
-        }
-        else {
-            prev->next = range;
-        }
-    }
-    else {
-        shared->tmFree[tmu] = range;
-    }
-}
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   tdfxMemRange *block, *prev;
 
+   if ( tss->umaTexMemory ) {
+      assert( unit == TDFX_TMU0 );
+   }
 
-#if 0 /* NOT USED */
-static void
-RemoveRange(tdfxContextPtr fxMesa, FxU32 tmu, tdfxMemRange *range)
-{
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    _glthread_LOCK_MUTEX(mesaShared->Mutex);
-    RemoveRange_NoLock(fxMesa, tmu, range);
-    _glthread_UNLOCK_MUTEX(mesaShared->Mutex);
+   if ( !range )
+      return;
+
+   if ( range->startAddr == range->endAddr ) {
+      DELETE_RANGE_NODE( tss, range );
+      return;
+   }
+   tss->freeTexMem[unit] += range->endAddr - range->startAddr;
+
+   /* find position in linked list to insert this tdfxMemRange node */
+   prev = NULL;
+   block = tss->freeRanges[unit];
+   while ( block ) {
+      assert( range->startAddr != block->startAddr );
+      if ( range->startAddr > block->startAddr ) {
+	 prev = block;
+	 block = block->next;
+      } else {
+	 break;
+      }
+   }
+
+   /* Insert the free block, combine with adjacent blocks when possible */
+   range->next = block;
+   if ( block ) {
+      if ( range->endAddr == block->startAddr ) {
+	 /* Combine */
+	 block->startAddr = range->startAddr;
+	 DELETE_RANGE_NODE( tss, range );
+	 range = block;
+      }
+   }
+   if ( prev ) {
+      if ( prev->endAddr == range->startAddr ) {
+	 /* Combine */
+	 prev->endAddr = range->endAddr;
+	 prev->next = range->next;
+	 DELETE_RANGE_NODE( tss, range );
+      } else {
+	 prev->next = range;
+      }
+   } else {
+      tss->freeRanges[unit] = range;
+   }
 }
-#endif
 
 
-/*
- * Allocate space for a texture image.
+/* Allocate space for a texture image.
  * <tmu> is the texture unit
  * <texmemsize> is the number of bytes to allocate
  */
 static tdfxMemRange *
-AllocTexMem(tdfxContextPtr fxMesa, FxU32 tmu, FxU32 texmemsize)
+tdfxTMAllocTexMem( tdfxContextPtr fxMesa, FxU32 unit, FxU32 size )
 {
-    FxU32 startAddr;
-    startAddr = FindStartAddr(fxMesa, tmu, texmemsize);
-    if (startAddr == BAD_ADDRESS) {
-       printf("AllocTexMem returned NULL!  tmu=%d texmemsize=%d\n",
-              tmu, texmemsize);
-        return NULL;
-    }
-    else {
-        tdfxMemRange *range;
-        range = NewRangeNode(fxMesa, startAddr, startAddr + texmemsize);
-        return range;
-    }
+   tdfxMemRange *range = NULL;
+   FxU32 start;
+
+   start = tdfxTMFindStartAddr( fxMesa, unit, size );
+
+   if ( start != BAD_ADDRESS ) {
+      range = tdfxTMNewRangeNode( fxMesa, start, start + size );
+   } else {
+      fprintf( stderr,
+	       "tdfxTMAllocTexMem returned NULL!  unit=%ld size=%ld\n",
+	       unit, size );
+   }
+   return range;
 }
 
 
-/*
- * Download (copy) the given texture data (all mipmap levels) into the
- * Voodoo's texture memory.
- * The texture memory must have already been allocated.
+/* Download (copy) the given texture data (all mipmap levels) into the
+ * Voodoo's texture memory.  The texture memory must have already been
+ * allocated.
  */
-void
-tdfxTMDownloadTexture(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
+void tdfxTMDownloadTextureLocked( tdfxContextPtr fxMesa,
+				  struct gl_texture_object *tObj )
 {
-    tdfxTexInfo *ti;
-    GLint l;
-    FxU32 targetTMU;
-
-    assert(tObj);
-    ti = TDFX_TEXTURE_DATA(tObj);
-    assert(ti);
-    targetTMU = ti->whichTMU;
-
-    switch (targetTMU) {
-    case TDFX_TMU0:
-    case TDFX_TMU1:
-        if (ti->tm[targetTMU]) {
-            for (l = ti->minLevel; l <= ti->maxLevel
-                    && ti->mipmapLevel[l].data; l++) {
-                GrLOD_t glideLod = ti->info.largeLodLog2 - l + tObj->BaseLevel;
-                FX_grTexDownloadMipMapLevel_NoLock(targetTMU,
-                                                  ti->tm[targetTMU]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_BOTH,
-                                                  ti->mipmapLevel[l].data);
-            }
-        }
-        break;
-    case TDFX_TMU_SPLIT:
-        if (ti->tm[TDFX_TMU0] && ti->tm[TDFX_TMU1]) {
-            for (l = ti->minLevel; l <= ti->maxLevel
-                    && ti->mipmapLevel[l].data; l++) {
-                GrLOD_t glideLod = ti->info.largeLodLog2 - l + tObj->BaseLevel;
-                FX_grTexDownloadMipMapLevel_NoLock(GR_TMU0,
-                                                  ti->tm[TDFX_TMU0]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_ODD,
-                                                  ti->mipmapLevel[l].data);
-
-                FX_grTexDownloadMipMapLevel_NoLock(GR_TMU1,
-                                                  ti->tm[TDFX_TMU1]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_EVEN,
-                                                  ti->mipmapLevel[l].data);
-            }
-        }
-        break;
-    case TDFX_TMU_BOTH:
-        if (ti->tm[TDFX_TMU0] && ti->tm[TDFX_TMU1]) {
-            for (l = ti->minLevel; l <= ti->maxLevel
-                    && ti->mipmapLevel[l].data; l++) {
-                GrLOD_t glideLod = ti->info.largeLodLog2 - l + tObj->BaseLevel;
-                FX_grTexDownloadMipMapLevel_NoLock(GR_TMU0,
-                                                  ti->tm[TDFX_TMU0]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_BOTH,
-                                                  ti->mipmapLevel[l].data);
-
-                FX_grTexDownloadMipMapLevel_NoLock(GR_TMU1,
-                                                  ti->tm[TDFX_TMU1]->startAddr,
-                                                  glideLod,
-                                                  ti->info.largeLodLog2,
-                                                  ti->info.aspectRatioLog2,
-                                                  ti->info.format,
-                                                  GR_MIPMAPLEVELMASK_BOTH,
-                                                  ti->mipmapLevel[l].data);
-            }
-        }
-        break;
-    default:
-        gl_problem(NULL, "error in tdfxTMDownloadTexture: bad tmu");
-        return;
-    }
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+   FxU32 targetTMU;
+   GLint l;
+
+   assert( tObj );
+   assert( t );
+
+   targetTMU = t->whichTMU;
+
+   switch ( targetTMU ) {
+   case TDFX_TMU0:
+   case TDFX_TMU1:
+      if ( t->range[targetTMU] ) {
+	 for ( l = t->minLevel ; l <= t->maxLevel && t->image[l].data ; l++ ) {
+	    GrLOD_t glideLod = t->info.largeLodLog2 - l + tObj->BaseLevel;
+
+	    grTexDownloadMipMapLevel( targetTMU,
+				      t->range[targetTMU]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_BOTH,
+				      t->image[l].data );
+	 }
+      }
+      break;
+
+   case TDFX_TMU_SPLIT:
+      if ( t->range[TDFX_TMU0] && t->range[TDFX_TMU1] ) {
+	 for ( l = t->minLevel ; l <= t->maxLevel && t->image[l].data ; l++ ) {
+	    GrLOD_t glideLod = t->info.largeLodLog2 - l + tObj->BaseLevel;
+
+	    grTexDownloadMipMapLevel( GR_TMU0,
+				      t->range[TDFX_TMU0]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_ODD,
+				      t->image[l].data );
+
+	    grTexDownloadMipMapLevel( GR_TMU1,
+				      t->range[TDFX_TMU1]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_EVEN,
+				      t->image[l].data );
+	 }
+      }
+      break;
+
+   case TDFX_TMU_BOTH:
+      if ( t->range[TDFX_TMU0] && t->range[TDFX_TMU1] ) {
+	 for ( l = t->minLevel ; l <= t->maxLevel && t->image[l].data ; l++ ) {
+	    GrLOD_t glideLod = t->info.largeLodLog2 - l + tObj->BaseLevel;
+
+	    grTexDownloadMipMapLevel( GR_TMU0,
+				      t->range[TDFX_TMU0]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_BOTH,
+				      t->image[l].data );
+
+	    grTexDownloadMipMapLevel( GR_TMU1,
+				      t->range[TDFX_TMU1]->startAddr,
+				      glideLod,
+				      t->info.largeLodLog2,
+				      t->info.aspectRatioLog2,
+				      t->info.format,
+				      GR_MIPMAPLEVELMASK_BOTH,
+				      t->image[l].data );
+	 }
+      }
+      break;
+
+   default:
+      gl_problem( NULL, "error in tdfxTMDownloadTexture: bad unit" );
+      return;
+   }
 }
 
 
-void
-tdfxTMReloadMipMapLevel(GLcontext *ctx, struct gl_texture_object *tObj,
-                        GLint level)
+void tdfxTMReloadMipMapLevelLocked( GLcontext *ctx,
+				    struct gl_texture_object *tObj,
+				    GLint level )
 {
-    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    GrLOD_t glideLod;
-    FxU32 tmu;
-
-    tmu = ti->whichTMU;
-    glideLod =  ti->info.largeLodLog2 - level + tObj->BaseLevel;
-    ASSERT(ti->isInTM);
-
-    switch (tmu) {
-    case TDFX_TMU0:
-    case TDFX_TMU1:
-        FX_grTexDownloadMipMapLevel(fxMesa, tmu,
-                                    ti->tm[tmu]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_BOTH,
-                                    ti->mipmapLevel[level].data);
-        break;
-    case TDFX_TMU_SPLIT:
-        FX_grTexDownloadMipMapLevel(fxMesa, GR_TMU0,
-                                    ti->tm[GR_TMU0]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_ODD,
-                                    ti->mipmapLevel[level].data);
-
-        FX_grTexDownloadMipMapLevel(fxMesa, GR_TMU1,
-                                    ti->tm[GR_TMU1]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_EVEN,
-                                    ti->mipmapLevel[level].data);
-        break;
-    case TDFX_TMU_BOTH:
-        FX_grTexDownloadMipMapLevel(fxMesa, GR_TMU0,
-                                    ti->tm[GR_TMU0]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_BOTH,
-                                    ti->mipmapLevel[level].data);
-
-        FX_grTexDownloadMipMapLevel(fxMesa, GR_TMU1,
-                                    ti->tm[GR_TMU1]->startAddr,
-                                    glideLod,
-                                    ti->info.largeLodLog2,
-                                    ti->info.aspectRatioLog2,
-                                    ti->info.format,
-                                    GR_MIPMAPLEVELMASK_BOTH,
-                                    ti->mipmapLevel[level].data);
-        break;
-
-    default:
-        gl_problem(ctx, "error in tdfxTMReloadMipMapLevel(): wrong tmu");
-        break;
-    }
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+   GrLOD_t glideLod;
+   FxU32 unit;
+
+   ASSERT( t->isInTM );
+
+   unit = t->whichTMU;
+   glideLod =  t->info.largeLodLog2 - level + tObj->BaseLevel;
+
+   switch ( unit ) {
+   case TDFX_TMU0:
+   case TDFX_TMU1:
+      grTexDownloadMipMapLevel( unit,
+				t->range[unit]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_BOTH,
+				t->image[level].data );
+      break;
+
+   case TDFX_TMU_SPLIT:
+      grTexDownloadMipMapLevel( GR_TMU0,
+				t->range[GR_TMU0]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_ODD,
+				t->image[level].data );
+
+      grTexDownloadMipMapLevel( GR_TMU1,
+				t->range[GR_TMU1]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_EVEN,
+				t->image[level].data );
+      break;
+
+   case TDFX_TMU_BOTH:
+      grTexDownloadMipMapLevel( GR_TMU0,
+				t->range[GR_TMU0]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_BOTH,
+				t->image[level].data );
+
+      grTexDownloadMipMapLevel( GR_TMU1,
+				t->range[GR_TMU1]->startAddr,
+				glideLod,
+				t->info.largeLodLog2,
+				t->info.aspectRatioLog2,
+				t->info.format,
+				GR_MIPMAPLEVELMASK_BOTH,
+				t->image[level].data );
+      break;
+
+   default:
+      gl_problem( ctx, "error in tdfxTMReloadMipMapLevel(): wrong unit" );
+      break;
+   }
 }
 
 
-/*
- * Allocate space for the given texture in texture memory then
+/* Allocate space for the given texture in texture memory then
  * download (copy) it into that space.
  */
-void
-tdfxTMMoveInTM_NoLock( tdfxContextPtr fxMesa, struct gl_texture_object *tObj,
-                       FxU32 targetTMU )
+void tdfxTMMoveInTMLocked( tdfxContextPtr fxMesa,
+			    struct gl_texture_object *tObj, FxU32 targetTMU )
 {
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    FxU32 texmemsize;
-
-    fxMesa->stats.reqTexUpload++;
-
-    if (ti->isInTM) {
-        if (ti->whichTMU == targetTMU)
-            return;
-        if (targetTMU == TDFX_TMU_SPLIT || ti->whichTMU == TDFX_TMU_SPLIT) {
-            tdfxTMMoveOutTM_NoLock(fxMesa, tObj);
-        }
-        else {
-            if (ti->whichTMU == TDFX_TMU_BOTH)
-                return;
-            targetTMU = TDFX_TMU_BOTH;
-        }
-    }
-
-    ti->whichTMU = targetTMU;
-
-    switch (targetTMU) {
-    case TDFX_TMU0:
-    case TDFX_TMU1:
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_BOTH,
-                                                       &(ti->info));
-        ti->tm[targetTMU] = AllocTexMem(fxMesa, targetTMU, texmemsize);
-        break;
-    case TDFX_TMU_SPLIT:
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_ODD,
-                                                       &(ti->info));
-        ti->tm[TDFX_TMU0] = AllocTexMem(fxMesa, TDFX_TMU0, texmemsize);
-        if (ti->tm[TDFX_TMU0])
-           fxMesa->stats.memTexUpload += texmemsize;
-
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_EVEN,
-                                                       &(ti->info));
-        ti->tm[TDFX_TMU1] = AllocTexMem(fxMesa, TDFX_TMU1, texmemsize);
-        break;
-    case TDFX_TMU_BOTH:
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_BOTH,
-                                                       &(ti->info));
-        ti->tm[TDFX_TMU0] = AllocTexMem(fxMesa, TDFX_TMU0, texmemsize);
-        if (ti->tm[TDFX_TMU0])
-           fxMesa->stats.memTexUpload += texmemsize;
-
-        texmemsize = FX_grTexTextureMemRequired_NoLock(GR_MIPMAPLEVELMASK_BOTH,
-                                                       &(ti->info));
-        ti->tm[TDFX_TMU1] = AllocTexMem(fxMesa, TDFX_TMU1, texmemsize);
-        break;
-    default:
-        gl_problem(NULL, "error in tdfxTMMoveInTM() -> bad tmu (%d)");
-        return;
-    }
-
-    ti->reloadImages = GL_TRUE;
-    ti->isInTM = GL_TRUE;
-
-    fxMesa->stats.texUpload++;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+   FxU32 size;
+
+   fxMesa->stats.reqTexUpload++;
+
+   if ( t->isInTM ) {
+      if ( t->whichTMU == targetTMU )
+	 return;
+
+      if ( targetTMU == TDFX_TMU_SPLIT || t->whichTMU == TDFX_TMU_SPLIT ) {
+	 tdfxTMMoveOutTMLocked( fxMesa, tObj );
+      } else {
+	 if ( t->whichTMU == TDFX_TMU_BOTH )
+	    return;
+	 targetTMU = TDFX_TMU_BOTH;
+      }
+   }
+
+   t->whichTMU = targetTMU;
+
+   switch ( targetTMU ) {
+   case TDFX_TMU0:
+   case TDFX_TMU1:
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_BOTH, &t->info );
+      t->range[targetTMU] = tdfxTMAllocTexMem(fxMesa, targetTMU, size);
+      break;
+
+   case TDFX_TMU_SPLIT:
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_ODD, &t->info );
+      t->range[TDFX_TMU0] = tdfxTMAllocTexMem( fxMesa, TDFX_TMU0, size );
+      if ( t->range[TDFX_TMU0] )
+	 fxMesa->stats.memTexUpload += size;
+
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_EVEN, &t->info );
+      t->range[TDFX_TMU1] = tdfxTMAllocTexMem( fxMesa, TDFX_TMU1, size );
+      break;
+
+   case TDFX_TMU_BOTH:
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_BOTH, &t->info );
+      t->range[TDFX_TMU0] = tdfxTMAllocTexMem( fxMesa, TDFX_TMU0, size );
+      if ( t->range[TDFX_TMU0] )
+	 fxMesa->stats.memTexUpload += size;
+
+      size = grTexTextureMemRequired( GR_MIPMAPLEVELMASK_BOTH, &t->info );
+      t->range[TDFX_TMU1] = tdfxTMAllocTexMem( fxMesa, TDFX_TMU1, size );
+      break;
+
+   default:
+      gl_problem( NULL, "error in tdfxTMMoveInTM() -> bad unit (%d)" );
+      return;
+   }
+
+   t->reloadImages = GL_TRUE;
+   t->isInTM = GL_TRUE;
+
+   fxMesa->stats.texUpload++;
 }
 
 
-/*
- * Move the given texture out of hardware texture memory.
+/* Move the given texture out of hardware texture memory.
  * This deallocates the texture's memory space.
  */
-void
-tdfxTMMoveOutTM_NoLock( tdfxContextPtr fxMesa, struct gl_texture_object *tObj )
+void tdfxTMMoveOutTMLocked( tdfxContextPtr fxMesa,
+			    struct gl_texture_object *tObj )
 {
-    struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-
-    if (MESA_VERBOSE & VERBOSE_DRIVER) {
-        fprintf(stderr, "fxmesa: fxTMMoveOutTM(%p (%d))\n", tObj, tObj->Name);
-    }
-
-    /*
-    VerifyFreeList(fxMesa, 0);
-    VerifyFreeList(fxMesa, 1);
-    */
-
-    if (!ti || !ti->isInTM)
-        return;
-
-    switch (ti->whichTMU) {
-    case TDFX_TMU0:
-    case TDFX_TMU1:
-        RemoveRange_NoLock(fxMesa, ti->whichTMU, ti->tm[ti->whichTMU]);
-        break;
-    case TDFX_TMU_SPLIT:
-    case TDFX_TMU_BOTH:
-        assert(!shared->umaTexMemory);
-        RemoveRange_NoLock(fxMesa, TDFX_TMU0, ti->tm[TDFX_TMU0]);
-        RemoveRange_NoLock(fxMesa, TDFX_TMU1, ti->tm[TDFX_TMU1]);
-        break;
-    default:
-        gl_problem(NULL, "tdfx driver: bad tmu in tdfxTMMOveOutTM()");
-        return;
-    }
-
-    ti->isInTM = GL_FALSE;
-    ti->tm[0] = NULL;
-    ti->tm[1] = NULL;
-    ti->whichTMU = TDFX_TMU_NONE;
-
-    /*
-    VerifyFreeList(fxMesa, 0);
-    VerifyFreeList(fxMesa, 1);
-    */
+   struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE ) {
+      fprintf( stderr, __FUNCTION__ "( %p (%d) )\n", tObj, tObj->Name );
+      tdfxTMVerifyFreeList( fxMesa, 0 );
+      tdfxTMVerifyFreeList( fxMesa, 1 );
+   }
+
+   if ( !t || !t->isInTM )
+      return;
+
+   switch ( t->whichTMU ) {
+   case TDFX_TMU0:
+   case TDFX_TMU1:
+      tdfxTMRemoveRangeLocked( fxMesa, t->whichTMU, t->range[t->whichTMU] );
+      break;
+
+   case TDFX_TMU_SPLIT:
+   case TDFX_TMU_BOTH:
+      assert( !tss->umaTexMemory );
+      tdfxTMRemoveRangeLocked( fxMesa, TDFX_TMU0, t->range[TDFX_TMU0] );
+      tdfxTMRemoveRangeLocked( fxMesa, TDFX_TMU1, t->range[TDFX_TMU1] );
+      break;
+
+   default:
+      gl_problem( NULL, "tdfx driver: bad unit in tdfxTMMOveOutTM()" );
+      return;
+   }
+
+   t->isInTM = GL_FALSE;
+   t->range[0] = NULL;
+   t->range[1] = NULL;
+   t->whichTMU = TDFX_TMU_NONE;
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE ) {
+      tdfxTMVerifyFreeList( fxMesa, 0 );
+      tdfxTMVerifyFreeList( fxMesa, 1 );
+   }
 }
 
 
-/*
- * Called via glDeleteTexture to delete a texture object.
- */
-void
-tdfxTMFreeTexture(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
+void tdfxTMFreeTextureLocked( tdfxContextPtr fxMesa,
+			      struct gl_texture_object *tObj )
 {
-    tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
-    if (ti) {
-        int i;
-        tdfxTMMoveOutTM(fxMesa, tObj);
-        for (i = 0; i < MAX_TEXTURE_LEVELS; i++) {
-            if (ti->mipmapLevel[i].data) {
-                FREE(ti->mipmapLevel[i].data);
-                ti->mipmapLevel[i].data = NULL;
-            }
-        }
-        FREE(ti);
-        tObj->DriverData = NULL;
-    }
-    /*
-    VerifyFreeList(fxMesa, 0);
-    VerifyFreeList(fxMesa, 1);
-    */
-}
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
+
+   if ( t ) {
+      int i;
+      tdfxTMMoveOutTMLocked( fxMesa, tObj );
+      for ( i = 0 ; i < MAX_TEXTURE_LEVELS ; i++ ) {
+	 if ( t->image[i].original.data ) {
+	    FREE( t->image[i].original.data );
+	    t->image[i].original.data = NULL;
+	    t->image[i].original.width = 0;
+	    t->image[i].original.height = 0;
+	    t->image[i].original.size = 0;
+	 }
+	 if ( t->image[i].rescaled.data ) {
+	    FREE( t->image[i].rescaled.data );
+	    t->image[i].rescaled.data = NULL;
+	    t->image[i].rescaled.width = 0;
+	    t->image[i].rescaled.height = 0;
+	    t->image[i].rescaled.size = 0;
+	 }
+      }
+      FREE( t );
+      tObj->DriverData = NULL;
+   }
 
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE ) {
+      tdfxTMVerifyFreeList( fxMesa, 0 );
+      tdfxTMVerifyFreeList( fxMesa, 1 );
+   }
+}
 
 
-/*
- * After a context switch this function will be called to restore
+/* After a context switch this function will be called to restore
  * texture memory for the new context.
  */
-void tdfxTMRestoreTextures_NoLock( tdfxContextPtr fxMesa )
+void tdfxTMRestoreTexturesLocked( tdfxContextPtr fxMesa )
 {
    GLcontext *ctx = fxMesa->glCtx;
    struct gl_texture_object *tObj;
    int i;
 
    for ( tObj = ctx->Shared->TexObjectList ; tObj ; tObj = tObj->Next ) {
-      tdfxTexInfo *ti = TDFX_TEXTURE_DATA( tObj );
-      if ( ti && ti->isInTM ) {
+      tdfxTexObjPtr t = TDFX_TEXTURE_DATA( tObj );
+      if ( t && t->isInTM ) {
 	 for ( i = 0 ; i < MAX_TEXTURE_UNITS ; i++ ) {
 	    if ( ctx->Texture.Unit[i].Current == tObj ) {
-	       tdfxTMDownloadTexture( fxMesa, tObj );
+	       tdfxTMDownloadTextureLocked( fxMesa, tObj );
 	       break;
 	    }
 	 }
 	 if ( i == MAX_TEXTURE_UNITS ) {
-	    tdfxTMMoveOutTM_NoLock( fxMesa, tObj );
+	    tdfxTMMoveOutTMLocked( fxMesa, tObj );
 	 }
       }
    }
-   /*
-   VerifyFreeList(fxMesa, 0);
-   VerifyFreeList(fxMesa, 1);
-   */
+
+   if ( TDFX_DEBUG & DEBUG_VERBOSE_TEXTURE ) {
+      tdfxTMVerifyFreeList( fxMesa, 0 );
+      tdfxTMVerifyFreeList( fxMesa, 1 );
+   }
 }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.h
index 1965aaa84..6979367a0 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.h
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texman.h
@@ -38,47 +38,39 @@
 #ifndef __TDFX_TEXMAN_H__
 #define __TDFX_TEXMAN_H__
 
-
 #include "tdfx_lock.h"
 
-
 extern void tdfxTMInit( tdfxContextPtr fxMesa );
-
 extern void tdfxTMClose( tdfxContextPtr fxMesa );
 
-extern void tdfxTMDownloadTexture(tdfxContextPtr fxMesa,
-                                  struct gl_texture_object *tObj);
-
-extern void tdfxTMReloadMipMapLevel( GLcontext *ctx,
-				     struct gl_texture_object *tObj,
-				     GLint level );
-
-extern void tdfxTMMoveInTM_NoLock( tdfxContextPtr fxMesa,
-                                   struct gl_texture_object *tObj,
-                                   FxU32 targetTMU );
-
-extern void tdfxTMMoveOutTM_NoLock( tdfxContextPtr fxMesa,
-                                    struct gl_texture_object *tObj );
-
-extern void tdfxTMFreeTexture( tdfxContextPtr fxMesa,
-			       struct gl_texture_object *tObj );
-
-extern void tdfxTMRestoreTextures_NoLock( tdfxContextPtr fxMesa );
-
-
-#define tdfxTMMoveInTM( fxMesa, tObj, targetTMU )		\
-   do {								\
-      LOCK_HARDWARE( fxMesa );					\
-      tdfxTMMoveInTM_NoLock( fxMesa, tObj, targetTMU );		\
-      UNLOCK_HARDWARE( fxMesa );				\
-   } while (0)
-
-#define tdfxTMMoveOutTM( fxMesa, tObj )				\
-   do {								\
-      LOCK_HARDWARE( fxMesa );					\
-      tdfxTMMoveOutTM_NoLock( fxMesa, tObj );			\
-      UNLOCK_HARDWARE( fxMesa );				\
-   } while (0)
-
+extern void tdfxTMDownloadTextureLocked( tdfxContextPtr fxMesa,
+					 struct gl_texture_object *tObj );
+extern void tdfxTMReloadMipMapLevelLocked( GLcontext *ctx,
+					   struct gl_texture_object *tObj,
+					   GLint level );
+extern void tdfxTMMoveInTMLocked( tdfxContextPtr fxMesa,
+				  struct gl_texture_object *tObj,
+				  FxU32 targetTMU );
+extern void tdfxTMMoveOutTMLocked( tdfxContextPtr fxMesa,
+				   struct gl_texture_object *tObj );
+extern void tdfxTMRestoreTexturesLocked( tdfxContextPtr fxMesa );
+
+extern void tdfxTMFreeTextureLocked( tdfxContextPtr fxMesa,
+				     struct gl_texture_object *tObj );
+
+
+#define tdfxTMMoveInTM( fxMesa, tObj, targetTMU )			\
+do {									\
+   LOCK_HARDWARE( fxMesa );						\
+   tdfxTMMoveInTMLocked( fxMesa, tObj, targetTMU );			\
+   UNLOCK_HARDWARE( fxMesa );						\
+} while (0)
+
+#define tdfxTMMoveOutTM( fxMesa, tObj )					\
+do {									\
+   LOCK_HARDWARE( fxMesa );						\
+   tdfxTMMoveOutTMLocked( fxMesa, tObj );				\
+   UNLOCK_HARDWARE( fxMesa );						\
+} while (0)
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.c
index 5493eea8a..368f051a3 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.c
@@ -38,7 +38,6 @@
 #include "tdfx_state.h"
 #include "tdfx_tex.h"
 #include "tdfx_texman.h"
-#include "tdfx_texstate.h"
 
 
 /* =============================================================
@@ -1270,105 +1269,105 @@ SetupDoubleTexEnvVoodoo3(GLcontext *ctx, int tmu0,
 static void
 setupSingleTMU(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
 {
-   struct tdfxSharedState *shared = (struct tdfxSharedState *) fxMesa->glCtx->Shared->DriverData;
-   tdfxTexInfo *ti = TDFX_TEXTURE_DATA(tObj);
    const GLcontext *ctx = fxMesa->glCtx;
+   tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ctx->Shared->DriverData;
+   tdfxTexObjPtr t = TDFX_TEXTURE_DATA(tObj);
 
    /* Make sure we're not loaded incorrectly */
-   if (ti->isInTM && !shared->umaTexMemory) {
+   if (t->isInTM && !tss->umaTexMemory) {
       /* if doing filtering between mipmap levels, alternate mipmap levels
        * must be in alternate TMUs.
        */
-      if (ti->LODblend) {
-         if (ti->whichTMU != TDFX_TMU_SPLIT)
-            tdfxTMMoveOutTM_NoLock(fxMesa, tObj);
+      if (t->LODblend) {
+         if (t->whichTMU != TDFX_TMU_SPLIT)
+            tdfxTMMoveOutTMLocked(fxMesa, tObj);
       }
       else {
-         if (ti->whichTMU == TDFX_TMU_SPLIT)
-            tdfxTMMoveOutTM_NoLock(fxMesa, tObj);
+         if (t->whichTMU == TDFX_TMU_SPLIT)
+            tdfxTMMoveOutTMLocked(fxMesa, tObj);
       }
    }
 
    /* Make sure we're loaded correctly */
-   if (!ti->isInTM) {
+   if (!t->isInTM) {
       /* Have to download the texture */
-      if (shared->umaTexMemory) {
-         tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU0);
+      if (tss->umaTexMemory) {
+         tdfxTMMoveInTMLocked(fxMesa, tObj, TDFX_TMU0);
       }
       else {
          /* Voodoo3 (split texture memory) */
-         if (ti->LODblend) {
-            tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU_SPLIT);
+         if (t->LODblend) {
+            tdfxTMMoveInTMLocked(fxMesa, tObj, TDFX_TMU_SPLIT);
          }
          else {
 #if 0
             /* XXX putting textures into the second memory bank when the
              * first bank is full is not working at this time.
              */
-            if (fxMesa->haveTwoTMUs) {
-               GLint memReq = FX_grTexTextureMemRequired_NoLock(
-                                       GR_MIPMAPLEVELMASK_BOTH, &(ti->info));
-               if (shared->freeTexMem[TDFX_TMU0] > memReq) {
-                  tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU0);
+            if (fxMesa->numTMUs > 1) {
+               GLint memReq =
+		  grTexTextureMemRequired( GR_MIPMAPLEVELMASK_BOTH, &t->info );
+               if (tss->freeTexMem[TDFX_TMU0] > memReq) {
+                  tdfxTMMoveInTMLocked(fxMesa, tObj, TDFX_TMU0);
                }
                else {
-                  tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU1);
+                  tdfxTMMoveInTMLocked(fxMesa, tObj, TDFX_TMU1);
                }
             }
             else
 #endif
             {
-               tdfxTMMoveInTM_NoLock(fxMesa, tObj, TDFX_TMU0);
+               tdfxTMMoveInTMLocked( fxMesa, tObj, TDFX_TMU0 );
             }
          }
       }
    }
 
-   if (ti->LODblend && ti->whichTMU == TDFX_TMU_SPLIT) {
+   if (t->LODblend && t->whichTMU == TDFX_TMU_SPLIT) {
       /* mipmap levels split between texture banks */
       GLint u;
 
-      if (ti->info.format == GR_TEXFMT_P_8 && !ctx->Texture.SharedPalette) {
+      if (t->info.format == GR_TEXFMT_P_8 && !ctx->Texture.SharedPalette) {
          fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-         fxMesa->TexPalette.Data = &(ti->palette);
+         fxMesa->TexPalette.Data = &(t->palette);
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
       }
 
       for (u = 0; u < 2; u++) {
-         fxMesa->TexParams[u].sClamp = ti->sClamp;
-         fxMesa->TexParams[u].tClamp = ti->tClamp;
-         fxMesa->TexParams[u].minFilt = ti->minFilt;
-         fxMesa->TexParams[u].magFilt = ti->magFilt;
-         fxMesa->TexParams[u].mmMode = ti->mmMode;
-         fxMesa->TexParams[u].LODblend = ti->LODblend;
+         fxMesa->TexParams[u].sClamp = t->sClamp;
+         fxMesa->TexParams[u].tClamp = t->tClamp;
+         fxMesa->TexParams[u].minFilt = t->minFilt;
+         fxMesa->TexParams[u].magFilt = t->magFilt;
+         fxMesa->TexParams[u].mmMode = t->mmMode;
+         fxMesa->TexParams[u].LODblend = t->LODblend;
          fxMesa->TexParams[u].LodBias = ctx->Texture.Unit[u].LodBias;
       }
       fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PARAMS;
 
-      fxMesa->TexSource[0].StartAddress = ti->tm[TDFX_TMU0]->startAddr;
+      fxMesa->TexSource[0].StartAddress = t->range[TDFX_TMU0]->startAddr;
       fxMesa->TexSource[0].EvenOdd = GR_MIPMAPLEVELMASK_ODD;
-      fxMesa->TexSource[0].Info = &(ti->info);
-      fxMesa->TexSource[1].StartAddress = ti->tm[TDFX_TMU1]->startAddr;
+      fxMesa->TexSource[0].Info = &(t->info);
+      fxMesa->TexSource[1].StartAddress = t->range[TDFX_TMU1]->startAddr;
       fxMesa->TexSource[1].EvenOdd = GR_MIPMAPLEVELMASK_EVEN;
-      fxMesa->TexSource[1].Info = &(ti->info);
+      fxMesa->TexSource[1].Info = &(t->info);
       fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_SOURCE;
    }
    else {
       FxU32 tmu;
 
-      if (ti->whichTMU == TDFX_TMU_BOTH)
+      if (t->whichTMU == TDFX_TMU_BOTH)
          tmu = TDFX_TMU0;
       else
-         tmu = ti->whichTMU;
+         tmu = t->whichTMU;
 
-      if (shared->umaTexMemory) {
-         assert(ti->whichTMU == TDFX_TMU0);
+      if (tss->umaTexMemory) {
+         assert(t->whichTMU == TDFX_TMU0);
          assert(tmu == TDFX_TMU0);
       }
 
-      if (ti->info.format == GR_TEXFMT_P_8 && !ctx->Texture.SharedPalette) {
+      if (t->info.format == GR_TEXFMT_P_8 && !ctx->Texture.SharedPalette) {
          fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-         fxMesa->TexPalette.Data = &(ti->palette);
+         fxMesa->TexPalette.Data = &(t->palette);
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
       }
 
@@ -1377,18 +1376,18 @@ setupSingleTMU(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
        * texture memory, so perhaps it's not a good idea.
        */
 
-      if (fxMesa->TexParams[tmu].sClamp != ti->sClamp ||
-          fxMesa->TexParams[tmu].tClamp != ti->tClamp ||
-          fxMesa->TexParams[tmu].minFilt != ti->minFilt ||
-          fxMesa->TexParams[tmu].magFilt != ti->magFilt ||
-          fxMesa->TexParams[tmu].mmMode != ti->mmMode ||
+      if (fxMesa->TexParams[tmu].sClamp != t->sClamp ||
+          fxMesa->TexParams[tmu].tClamp != t->tClamp ||
+          fxMesa->TexParams[tmu].minFilt != t->minFilt ||
+          fxMesa->TexParams[tmu].magFilt != t->magFilt ||
+          fxMesa->TexParams[tmu].mmMode != t->mmMode ||
           fxMesa->TexParams[tmu].LODblend != FXFALSE ||
           fxMesa->TexParams[tmu].LodBias != ctx->Texture.Unit[tmu].LodBias) {
-         fxMesa->TexParams[tmu].sClamp = ti->sClamp;
-         fxMesa->TexParams[tmu].tClamp = ti->tClamp;
-         fxMesa->TexParams[tmu].minFilt = ti->minFilt;
-         fxMesa->TexParams[tmu].magFilt = ti->magFilt;
-         fxMesa->TexParams[tmu].mmMode = ti->mmMode;
+         fxMesa->TexParams[tmu].sClamp = t->sClamp;
+         fxMesa->TexParams[tmu].tClamp = t->tClamp;
+         fxMesa->TexParams[tmu].minFilt = t->minFilt;
+         fxMesa->TexParams[tmu].magFilt = t->magFilt;
+         fxMesa->TexParams[tmu].mmMode = t->mmMode;
          fxMesa->TexParams[tmu].LODblend = FXFALSE;
          fxMesa->TexParams[tmu].LodBias = ctx->Texture.Unit[tmu].LodBias;
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PARAMS;
@@ -1397,16 +1396,16 @@ setupSingleTMU(tdfxContextPtr fxMesa, struct gl_texture_object *tObj)
       /* Glide texture source info */
       fxMesa->TexSource[0].Info = NULL;
       fxMesa->TexSource[1].Info = NULL;
-      if (ti->tm[tmu]) {
-         fxMesa->TexSource[tmu].StartAddress = ti->tm[tmu]->startAddr;
+      if (t->range[tmu]) {
+         fxMesa->TexSource[tmu].StartAddress = t->range[tmu]->startAddr;
          fxMesa->TexSource[tmu].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu].Info = &(ti->info);
+         fxMesa->TexSource[tmu].Info = &(t->info);
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_SOURCE;
       }
    }
 
-   fxMesa->sScale0 = ti->sScale;
-   fxMesa->tScale0 = ti->tScale;
+   fxMesa->sScale0 = t->sScale;
+   fxMesa->tScale0 = t->tScale;
 }
 
 static void
@@ -1420,12 +1419,12 @@ selectSingleTMUSrc(tdfxContextPtr fxMesa, GLint tmu, FxBool LODblend)
       fxMesa->TexCombine[0].InvertRGB = FXFALSE;
       fxMesa->TexCombine[0].InvertAlpha = FXFALSE;
 
-      if (fxMesa->haveTwoTMUs) {
-         const struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-         const struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
+      if ( fxMesa->numTMUs > 1 ) {
+         const struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+         const tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
          int tmu;
 
-         if (shared->umaTexMemory)
+         if (tss->umaTexMemory)
             tmu = GR_TMU0;
          else
             tmu = GR_TMU1;
@@ -1447,7 +1446,7 @@ selectSingleTMUSrc(tdfxContextPtr fxMesa, GLint tmu, FxBool LODblend)
          fxMesa->TexCombine[0].FactorAlpha = GR_COMBINE_FACTOR_NONE;
          fxMesa->TexCombine[0].InvertRGB = FXFALSE;
          fxMesa->TexCombine[0].InvertAlpha = FXFALSE;
-         if (fxMesa->haveTwoTMUs) {
+         if ( fxMesa->numTMUs > 1 ) {
             fxMesa->TexCombine[1].FunctionRGB = GR_COMBINE_FUNCTION_ZERO;
             fxMesa->TexCombine[1].FactorRGB = GR_COMBINE_FACTOR_NONE;
             fxMesa->TexCombine[1].FunctionAlpha = GR_COMBINE_FUNCTION_ZERO;
@@ -1511,7 +1510,7 @@ static void print_state(tdfxContextPtr fxMesa)
 static void setupTextureSingleTMU(GLcontext * ctx, GLuint unit)
 {
    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-   tdfxTexInfo *ti;
+   tdfxTexObjPtr t;
    struct gl_texture_object *tObj;
    int tmu;
    GLenum envMode, baseFormat;
@@ -1530,17 +1529,17 @@ static void setupTextureSingleTMU(GLcontext * ctx, GLuint unit)
 
    setupSingleTMU(fxMesa, tObj);
 
-   ti = TDFX_TEXTURE_DATA(tObj);
-   if (ti->whichTMU == TDFX_TMU_BOTH)
+   t = TDFX_TEXTURE_DATA(tObj);
+   if (t->whichTMU == TDFX_TMU_BOTH)
       tmu = TDFX_TMU0;
    else
-      tmu = ti->whichTMU;
+      tmu = t->whichTMU;
 
    if (fxMesa->tmuSrc != tmu) {
-      selectSingleTMUSrc(fxMesa, tmu, ti->LODblend);
+      selectSingleTMUSrc(fxMesa, tmu, t->LODblend);
    }
 
-   if (ti->reloadImages)
+   if (t->reloadImages)
       fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_IMAGES;
 
    /* Some texture environments not supported */
@@ -1627,35 +1626,35 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
 #define T0_IN_TMU1     0x10
 #define T1_IN_TMU1     0x20
 
-    const struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    const struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
+    const struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+    const tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
     const GLcontext *ctx = fxMesa->glCtx;
-    tdfxTexInfo *ti0 = TDFX_TEXTURE_DATA(tObj0);
-    tdfxTexInfo *ti1 = TDFX_TEXTURE_DATA(tObj1);
+    tdfxTexObjPtr t0 = TDFX_TEXTURE_DATA(tObj0);
+    tdfxTexObjPtr t1 = TDFX_TEXTURE_DATA(tObj1);
     GLuint tstate = 0;
     int tmu0 = 0, tmu1 = 1;
 
-    if (shared->umaTexMemory) {
-       if (!ti0->isInTM) {
-          tdfxTMMoveInTM_NoLock(fxMesa, tObj0, TDFX_TMU0);
-          assert(ti0->isInTM);
+    if (tss->umaTexMemory) {
+       if (!t0->isInTM) {
+          tdfxTMMoveInTMLocked(fxMesa, tObj0, TDFX_TMU0);
+          assert(t0->isInTM);
        }
-       if (!ti1->isInTM) {
-          tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU0);
-          assert(ti1->isInTM);
+       if (!t1->isInTM) {
+          tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU0);
+          assert(t1->isInTM);
        }
     }
     else {
        /* We shouldn't need to do this. There is something wrong with
           multitexturing when the TMUs are swapped. So, we're forcing
           them to always be loaded correctly. !!! */
-       if (ti0->whichTMU == TDFX_TMU1)
-           tdfxTMMoveOutTM_NoLock(fxMesa, tObj0);
-       if (ti1->whichTMU == TDFX_TMU0)
-           tdfxTMMoveOutTM_NoLock(fxMesa, tObj1);
+       if (t0->whichTMU == TDFX_TMU1)
+           tdfxTMMoveOutTMLocked(fxMesa, tObj0);
+       if (t1->whichTMU == TDFX_TMU0)
+           tdfxTMMoveOutTMLocked(fxMesa, tObj1);
 
-       if (ti0->isInTM) {
-           switch (ti0->whichTMU) {
+       if (t0->isInTM) {
+           switch (t0->whichTMU) {
            case TDFX_TMU0:
                tstate |= T0_IN_TMU0;
                break;
@@ -1673,8 +1672,8 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
        else
            tstate |= T0_NOT_IN_TMU;
 
-       if (ti1->isInTM) {
-           switch (ti1->whichTMU) {
+       if (t1->isInTM) {
+           switch (t1->whichTMU) {
            case TDFX_TMU0:
                tstate |= T1_IN_TMU0;
                break;
@@ -1697,7 +1696,7 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
        if (!(((tstate & T0_IN_TMU0) && (tstate & T1_IN_TMU1)) ||
              ((tstate & T0_IN_TMU1) && (tstate & T1_IN_TMU0)))) {
            if (tObj0 == tObj1) {
-              tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU_BOTH);
+              tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU_BOTH);
            }
            else {
                /* Find the minimal way to correct the situation */
@@ -1705,10 +1704,10 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
                    /* We have one in the standard order, setup the other */
                    if (tstate & T0_IN_TMU0) {
                       /* T0 is in TMU0, put T1 in TMU1 */
-                      tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU1);
+                      tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU1);
                    }
                    else {
-                       tdfxTMMoveInTM_NoLock(fxMesa, tObj0, TDFX_TMU0);
+                       tdfxTMMoveInTMLocked(fxMesa, tObj0, TDFX_TMU0);
                    }
                    /* tmu0 and tmu1 are setup */
                }
@@ -1716,36 +1715,36 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
                    /* we have one in the reverse order, setup the other */
                    if (tstate & T1_IN_TMU0) {
                       /* T1 is in TMU0, put T0 in TMU1 */
-                      tdfxTMMoveInTM_NoLock(fxMesa, tObj0, TDFX_TMU1);
+                      tdfxTMMoveInTMLocked(fxMesa, tObj0, TDFX_TMU1);
                    }
                    else {
-                       tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU0);
+                       tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU0);
                    }
                    tmu0 = 1;
                    tmu1 = 0;
                }
                else {              /* Nothing is loaded */
-                   tdfxTMMoveInTM_NoLock(fxMesa, tObj0, TDFX_TMU0);
-                   tdfxTMMoveInTM_NoLock(fxMesa, tObj1, TDFX_TMU1);
+                   tdfxTMMoveInTMLocked(fxMesa, tObj0, TDFX_TMU0);
+                   tdfxTMMoveInTMLocked(fxMesa, tObj1, TDFX_TMU1);
                    /* tmu0 and tmu1 are setup */
                }
            }
        }
     }
 
-    ti0->lastTimeUsed = fxMesa->texBindNumber;
-    ti1->lastTimeUsed = fxMesa->texBindNumber;
+    t0->lastTimeUsed = fxMesa->texBindNumber;
+    t1->lastTimeUsed = fxMesa->texBindNumber;
 
 
     if (!ctx->Texture.SharedPalette) {
-        if (ti0->info.format == GR_TEXFMT_P_8) {
+        if (t0->info.format == GR_TEXFMT_P_8) {
             fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-            fxMesa->TexPalette.Data = &(ti0->palette);
+            fxMesa->TexPalette.Data = &(t0->palette);
             fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
         }
-        else if (ti1->info.format == GR_TEXFMT_P_8) {
+        else if (t1->info.format == GR_TEXFMT_P_8) {
             fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-            fxMesa->TexPalette.Data = &(ti1->palette);
+            fxMesa->TexPalette.Data = &(t1->palette);
             fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
         }
         else {
@@ -1756,25 +1755,25 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
     /*
      * Setup Unit 0
      */
-    assert(ti0->isInTM);
-    assert(ti0->tm[tmu0]);
-    fxMesa->TexSource[tmu0].StartAddress = ti0->tm[tmu0]->startAddr;
+    assert(t0->isInTM);
+    assert(t0->range[tmu0]);
+    fxMesa->TexSource[tmu0].StartAddress = t0->range[tmu0]->startAddr;
     fxMesa->TexSource[tmu0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-    fxMesa->TexSource[tmu0].Info = &(ti0->info);
+    fxMesa->TexSource[tmu0].Info = &(t0->info);
     fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_SOURCE;
 
-    if (fxMesa->TexParams[tmu0].sClamp != ti0->sClamp ||
-        fxMesa->TexParams[tmu0].tClamp != ti0->tClamp ||
-        fxMesa->TexParams[tmu0].minFilt != ti0->minFilt ||
-        fxMesa->TexParams[tmu0].magFilt != ti0->magFilt ||
-        fxMesa->TexParams[tmu0].mmMode != ti0->mmMode ||
+    if (fxMesa->TexParams[tmu0].sClamp != t0->sClamp ||
+        fxMesa->TexParams[tmu0].tClamp != t0->tClamp ||
+        fxMesa->TexParams[tmu0].minFilt != t0->minFilt ||
+        fxMesa->TexParams[tmu0].magFilt != t0->magFilt ||
+        fxMesa->TexParams[tmu0].mmMode != t0->mmMode ||
         fxMesa->TexParams[tmu0].LODblend != FXFALSE ||
         fxMesa->TexParams[tmu0].LodBias != ctx->Texture.Unit[tmu0].LodBias) {
-       fxMesa->TexParams[tmu0].sClamp = ti0->sClamp;
-       fxMesa->TexParams[tmu0].tClamp = ti0->tClamp;
-       fxMesa->TexParams[tmu0].minFilt = ti0->minFilt;
-       fxMesa->TexParams[tmu0].magFilt = ti0->magFilt;
-       fxMesa->TexParams[tmu0].mmMode = ti0->mmMode;
+       fxMesa->TexParams[tmu0].sClamp = t0->sClamp;
+       fxMesa->TexParams[tmu0].tClamp = t0->tClamp;
+       fxMesa->TexParams[tmu0].minFilt = t0->minFilt;
+       fxMesa->TexParams[tmu0].magFilt = t0->magFilt;
+       fxMesa->TexParams[tmu0].mmMode = t0->mmMode;
        fxMesa->TexParams[tmu0].LODblend = FXFALSE;
        fxMesa->TexParams[tmu0].LodBias = ctx->Texture.Unit[tmu0].LodBias;
        fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PARAMS;
@@ -1783,42 +1782,42 @@ setupDoubleTMU(tdfxContextPtr fxMesa,
     /*
      * Setup Unit 1
      */
-    if (shared->umaTexMemory) {
-        ASSERT(ti1->isInTM);
-        ASSERT(ti1->tm[0]);
-        fxMesa->TexSource[tmu1].StartAddress = ti1->tm[0]->startAddr;
+    if (tss->umaTexMemory) {
+        ASSERT(t1->isInTM);
+        ASSERT(t1->range[0]);
+        fxMesa->TexSource[tmu1].StartAddress = t1->range[0]->startAddr;
         fxMesa->TexSource[tmu1].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-        fxMesa->TexSource[tmu1].Info = &(ti1->info);
+        fxMesa->TexSource[tmu1].Info = &(t1->info);
     }
     else {
-        ASSERT(ti1->isInTM);
-        ASSERT(ti1->tm[tmu1]);
-        fxMesa->TexSource[tmu1].StartAddress = ti1->tm[tmu1]->startAddr;
+        ASSERT(t1->isInTM);
+        ASSERT(t1->range[tmu1]);
+        fxMesa->TexSource[tmu1].StartAddress = t1->range[tmu1]->startAddr;
         fxMesa->TexSource[tmu1].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-        fxMesa->TexSource[tmu1].Info = &(ti1->info);
+        fxMesa->TexSource[tmu1].Info = &(t1->info);
     }
 
-    if (fxMesa->TexParams[tmu1].sClamp != ti1->sClamp ||
-        fxMesa->TexParams[tmu1].tClamp != ti1->tClamp ||
-        fxMesa->TexParams[tmu1].minFilt != ti1->minFilt ||
-        fxMesa->TexParams[tmu1].magFilt != ti1->magFilt ||
-        fxMesa->TexParams[tmu1].mmMode != ti1->mmMode ||
+    if (fxMesa->TexParams[tmu1].sClamp != t1->sClamp ||
+        fxMesa->TexParams[tmu1].tClamp != t1->tClamp ||
+        fxMesa->TexParams[tmu1].minFilt != t1->minFilt ||
+        fxMesa->TexParams[tmu1].magFilt != t1->magFilt ||
+        fxMesa->TexParams[tmu1].mmMode != t1->mmMode ||
         fxMesa->TexParams[tmu1].LODblend != FXFALSE ||
         fxMesa->TexParams[tmu1].LodBias != ctx->Texture.Unit[tmu1].LodBias) {
-       fxMesa->TexParams[tmu1].sClamp = ti1->sClamp;
-       fxMesa->TexParams[tmu1].tClamp = ti1->tClamp;
-       fxMesa->TexParams[tmu1].minFilt = ti1->minFilt;
-       fxMesa->TexParams[tmu1].magFilt = ti1->magFilt;
-       fxMesa->TexParams[tmu1].mmMode = ti1->mmMode;
+       fxMesa->TexParams[tmu1].sClamp = t1->sClamp;
+       fxMesa->TexParams[tmu1].tClamp = t1->tClamp;
+       fxMesa->TexParams[tmu1].minFilt = t1->minFilt;
+       fxMesa->TexParams[tmu1].magFilt = t1->magFilt;
+       fxMesa->TexParams[tmu1].mmMode = t1->mmMode;
        fxMesa->TexParams[tmu1].LODblend = FXFALSE;
        fxMesa->TexParams[tmu1].LodBias = ctx->Texture.Unit[tmu1].LodBias;
        fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PARAMS;
     }
 
-    fxMesa->sScale0 = ti0->sScale;
-    fxMesa->tScale0 = ti0->tScale;
-    fxMesa->sScale1 = ti1->sScale;
-    fxMesa->tScale1 = ti1->tScale;
+    fxMesa->sScale0 = t0->sScale;
+    fxMesa->tScale0 = t0->tScale;
+    fxMesa->sScale1 = t1->sScale;
+    fxMesa->tScale1 = t1->tScale;
 
 #undef T0_NOT_IN_TMU
 #undef T1_NOT_IN_TMU
@@ -1833,8 +1832,8 @@ static void setupTextureDoubleTMU(GLcontext * ctx)
    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
    struct gl_texture_object *tObj0 = ctx->Texture.Unit[0].CurrentD[2];
    struct gl_texture_object *tObj1 = ctx->Texture.Unit[1].CurrentD[2];
-   tdfxTexInfo *ti0 = TDFX_TEXTURE_DATA(tObj0);
-   tdfxTexInfo *ti1 = TDFX_TEXTURE_DATA(tObj1);
+   tdfxTexObjPtr t0 = TDFX_TEXTURE_DATA(tObj0);
+   tdfxTexObjPtr t1 = TDFX_TEXTURE_DATA(tObj1);
    struct gl_texture_image *baseImage0 = tObj0->Image[tObj0->BaseLevel];
    struct gl_texture_image *baseImage1 = tObj1->Image[tObj1->BaseLevel];
    const GLenum envMode0 = ctx->Texture.Unit[0].EnvMode;
@@ -1847,7 +1846,7 @@ static void setupTextureDoubleTMU(GLcontext * ctx)
 
    setupDoubleTMU(fxMesa, tObj0, tObj1);
 
-   if (ti0->reloadImages || ti1->reloadImages)
+   if (t0->reloadImages || t1->reloadImages)
       fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_IMAGES;
 
    fxMesa->tmuSrc = TDFX_TMU_BOTH;
@@ -1893,7 +1892,7 @@ static void setupTextureDoubleTMU(GLcontext * ctx)
    }
    else {
       int unit0, unit1;
-      if ((ti0->whichTMU == TDFX_TMU1) || (ti1->whichTMU == TDFX_TMU0))
+      if ((t0->whichTMU == TDFX_TMU1) || (t1->whichTMU == TDFX_TMU0))
          unit0 = 1;
       else
          unit0 = 0;
@@ -1924,34 +1923,34 @@ static void setupTextureDoubleTMU(GLcontext * ctx)
 }
 
 
-void
-tdfxUpdateTextureState( GLcontext *ctx )
+void tdfxUpdateTextureState( GLcontext *ctx )
 {
    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
-   GLuint tex2Denabled = ctx->Texture.ReallyEnabled;
+   GLuint enabled = ctx->Texture.ReallyEnabled;
 
-   if (!fxMesa->haveTwoTMUs)
-      tex2Denabled &= TEXTURE0_2D;
+   if ( fxMesa->numTMUs == 1 )
+      enabled &= TEXTURE0_2D;
 
-   switch (tex2Denabled) {
+   switch ( enabled ) {
    case TEXTURE0_2D:
       LOCK_HARDWARE( fxMesa );  /* XXX remove locking eventually */
-      setupTextureSingleTMU(ctx, 0);
+      setupTextureSingleTMU( ctx, 0 );
       UNLOCK_HARDWARE( fxMesa );
       break;
    case TEXTURE1_2D:
       LOCK_HARDWARE( fxMesa );
-      setupTextureSingleTMU(ctx, 1);
+      setupTextureSingleTMU( ctx, 1 );
       UNLOCK_HARDWARE( fxMesa );
       break;
-   case (TEXTURE0_2D | TEXTURE1_2D):
+   case TEXTURE0_2D | TEXTURE1_2D:
       LOCK_HARDWARE( fxMesa );
-      setupTextureDoubleTMU(ctx);
+      setupTextureDoubleTMU( ctx );
       UNLOCK_HARDWARE( fxMesa );
       break;
+
    default:
-      /* disable hardware texturing */
-      if (TDFX_IS_NAPALM(fxMesa)) {
+      /* Disable hardware texturing */
+      if ( TDFX_IS_NAPALM( fxMesa ) ) {
          fxMesa->ColorCombineExt.SourceA = GR_CMBX_ITRGB;
          fxMesa->ColorCombineExt.ModeA = GR_FUNC_MODE_X;
          fxMesa->ColorCombineExt.SourceB = GR_CMBX_ZERO;
@@ -1972,9 +1971,8 @@ tdfxUpdateTextureState( GLcontext *ctx )
          fxMesa->AlphaCombineExt.InvertD = FXFALSE;
          fxMesa->AlphaCombineExt.Shift = 0;
          fxMesa->AlphaCombineExt.Invert = FXFALSE;
-      }
-      else {
-         /* Voodoo 3*/
+      } else {
+         /* Voodoo 3 */
          fxMesa->ColorCombine.Function = GR_COMBINE_FUNCTION_LOCAL;
          fxMesa->ColorCombine.Factor = GR_COMBINE_FACTOR_NONE;
          fxMesa->ColorCombine.Local = GR_COMBINE_LOCAL_ITERATED;
@@ -1997,106 +1995,103 @@ tdfxUpdateTextureState( GLcontext *ctx )
 }
 
 
-
-/*
- * This is a special case of texture state update.
+/* This is a special case of texture state update.
  * It's used when we've simply bound a new texture to a texture
  * unit and the new texture has the exact same attributes as the
  * previously bound texture.
  * This is very common in Quake3.
  */
-void
-tdfxUpdateTextureBinding( GLcontext *ctx )
+void tdfxUpdateTextureBinding( GLcontext *ctx )
 {
    tdfxContextPtr fxMesa = TDFX_CONTEXT(ctx);
+   const struct gl_shared_state *ss = fxMesa->glCtx->Shared;
+   const tdfxSharedStatePtr tss = (tdfxSharedStatePtr)ss->DriverData;
    struct gl_texture_object *tObj0 = ctx->Texture.Unit[0].CurrentD[2];
    struct gl_texture_object *tObj1 = ctx->Texture.Unit[1].CurrentD[2];
-   tdfxTexInfo *ti0 = TDFX_TEXTURE_DATA(tObj0);
-   tdfxTexInfo *ti1 = TDFX_TEXTURE_DATA(tObj1);
-
-    const struct gl_shared_state *mesaShared = fxMesa->glCtx->Shared;
-    const struct tdfxSharedState *shared = (struct tdfxSharedState *) mesaShared->DriverData;
+   tdfxTexObjPtr t0 = TDFX_TEXTURE_DATA(tObj0);
+   tdfxTexObjPtr t1 = TDFX_TEXTURE_DATA(tObj1);
 
-   if (ti0) {
-      fxMesa->sScale0 = ti0->sScale;
-      fxMesa->tScale0 = ti0->tScale;
-      if (ti0->info.format == GR_TEXFMT_P_8) {
+   if ( t0 ) {
+      fxMesa->sScale0 = t0->sScale;
+      fxMesa->tScale0 = t0->tScale;
+      if ( t0->info.format == GR_TEXFMT_P_8 ) {
          fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-         fxMesa->TexPalette.Data = &(ti0->palette);
+         fxMesa->TexPalette.Data = &t0->palette;
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
-      }
-      else if (ti1 && ti1->info.format == GR_TEXFMT_P_8) {
+      } else if ( t1 && t1->info.format == GR_TEXFMT_P_8 ) {
          fxMesa->TexPalette.Type = GR_TEXTABLE_PALETTE_6666_EXT;
-         fxMesa->TexPalette.Data = &(ti1->palette);
+         fxMesa->TexPalette.Data = &t1->palette;
          fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_PALETTE;
       }
    }
-   if (ti1) {
-      fxMesa->sScale1 = ti1->sScale;
-      fxMesa->tScale1 = ti1->tScale;
+   if ( t1 ) {
+      fxMesa->sScale1 = t1->sScale;
+      fxMesa->tScale1 = t1->tScale;
    }
 
-   if (ctx->Texture.ReallyEnabled == TEXTURE0_2D) {
-      if (shared->umaTexMemory) {
-         fxMesa->TexSource[0].StartAddress = ti0->tm[0]->startAddr;
+   switch  ( ctx->Texture.ReallyEnabled ) {
+   case TEXTURE0_2D:
+      if ( tss->umaTexMemory ) {
+         fxMesa->TexSource[0].StartAddress = t0->range[0]->startAddr;
          fxMesa->TexSource[0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[0].Info = &(ti0->info);
-      }
-      else {
-         if (ti0->LODblend && ti0->whichTMU == TDFX_TMU_SPLIT) {
-            fxMesa->TexSource[0].StartAddress = ti0->tm[TDFX_TMU0]->startAddr;
+         fxMesa->TexSource[0].Info = &(t0->info);
+      } else {
+         if ( t0->LODblend && t0->whichTMU == TDFX_TMU_SPLIT ) {
+            fxMesa->TexSource[0].StartAddress = t0->range[TDFX_TMU0]->startAddr;
             fxMesa->TexSource[0].EvenOdd = GR_MIPMAPLEVELMASK_ODD;
-            fxMesa->TexSource[0].Info = &(ti0->info);
-            fxMesa->TexSource[1].StartAddress = ti0->tm[TDFX_TMU1]->startAddr;
+            fxMesa->TexSource[0].Info = &t0->info;
+            fxMesa->TexSource[1].StartAddress = t0->range[TDFX_TMU1]->startAddr;
             fxMesa->TexSource[1].EvenOdd = GR_MIPMAPLEVELMASK_EVEN;
-            fxMesa->TexSource[1].Info = &(ti0->info);
-         }
-         else {
-            FxU32 tmu;
-            if (ti0->whichTMU == TDFX_TMU_BOTH)
-               tmu = TDFX_TMU0;
-            else
-               tmu = ti0->whichTMU;
+            fxMesa->TexSource[1].Info = &t0->info;
+         } else {
+            FxU32 unit;
+            if ( t0->whichTMU == TDFX_TMU_BOTH ) {
+               unit = TDFX_TMU0;
+            } else {
+               unit = t0->whichTMU;
+	    }
             fxMesa->TexSource[0].Info = NULL;
             fxMesa->TexSource[1].Info = NULL;
-            if (ti0->tm[tmu]) {
-               fxMesa->TexSource[tmu].StartAddress = ti0->tm[tmu]->startAddr;
-               fxMesa->TexSource[tmu].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-               fxMesa->TexSource[tmu].Info = &(ti0->info);
+            if ( t0->range[unit] ) {
+               fxMesa->TexSource[unit].StartAddress = t0->range[unit]->startAddr;
+               fxMesa->TexSource[unit].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
+               fxMesa->TexSource[unit].Info = &t0->info;
             }
          }
       }
-   }
-   else if (ctx->Texture.ReallyEnabled == TEXTURE1_2D) {
-      if (shared->umaTexMemory) {
-         fxMesa->TexSource[0].StartAddress = ti1->tm[0]->startAddr;
+      break;
+
+   case TEXTURE1_2D:
+      if ( tss->umaTexMemory ) {
+         fxMesa->TexSource[0].StartAddress = t1->range[0]->startAddr;
          fxMesa->TexSource[0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[0].Info = &(ti1->info);
+         fxMesa->TexSource[0].Info = &t1->info;
       }
-   }
-   else if (ctx->Texture.ReallyEnabled == (TEXTURE0_2D | TEXTURE1_2D)) {
-      if (shared->umaTexMemory) {
+      break;
+
+   case TEXTURE0_2D | TEXTURE1_2D:
+      if ( tss->umaTexMemory ) {
          const FxU32 tmu0 = 0, tmu1 = 1;
-         fxMesa->TexSource[tmu0].StartAddress = ti0->tm[0]->startAddr;
+         fxMesa->TexSource[tmu0].StartAddress = t0->range[0]->startAddr;
          fxMesa->TexSource[tmu0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu0].Info = &(ti0->info);
+         fxMesa->TexSource[tmu0].Info = &t0->info;
 
-         fxMesa->TexSource[tmu1].StartAddress = ti1->tm[0]->startAddr;
+         fxMesa->TexSource[tmu1].StartAddress = t1->range[0]->startAddr;
          fxMesa->TexSource[tmu1].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu1].Info = &(ti1->info);
+         fxMesa->TexSource[tmu1].Info = &t1->info;
       }
       else {
          const FxU32 tmu0 = 0, tmu1 = 1;
-         fxMesa->TexSource[tmu0].StartAddress = ti0->tm[tmu0]->startAddr;
+         fxMesa->TexSource[tmu0].StartAddress = t0->range[tmu0]->startAddr;
          fxMesa->TexSource[tmu0].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu0].Info = &(ti0->info);
+         fxMesa->TexSource[tmu0].Info = &t0->info;
 
-         fxMesa->TexSource[tmu1].StartAddress = ti1->tm[tmu1]->startAddr;
+         fxMesa->TexSource[tmu1].StartAddress = t1->range[tmu1]->startAddr;
          fxMesa->TexSource[tmu1].EvenOdd = GR_MIPMAPLEVELMASK_BOTH;
-         fxMesa->TexSource[tmu1].Info = &(ti1->info);
+         fxMesa->TexSource[tmu1].Info = &t1->info;
       }
+      break;
    }
 
-
    fxMesa->dirty |= TDFX_UPLOAD_TEXTURE_SOURCE;
 }
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.h
deleted file mode 100644
index 0c0d4adcb..000000000
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_texstate.h
+++ /dev/null
@@ -1,44 +0,0 @@
-/* -*- mode: c; c-basic-offset: 3 -*-
- *
- * Copyright 2000 VA Linux Systems Inc., Fremont, California.
- *
- * 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, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * VA LINUX SYSTEMS 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.
- */
-/* $XFree86$ */
-
-/*
- * Original rewrite:
- *	Gareth Hughes <gareth@valinux.com>, 29 Sep - 1 Oct 2000
- *
- * Authors:
- *	Gareth Hughes <gareth@valinux.com>
- *	Brian Paul <brianp@valinux.com>
- *
- */
-
-#ifndef __TDFX_TEXSTATE_H__
-#define __TDFX_TEXSTATE_H__
-
-extern void tdfxUpdateTextureState( GLcontext *ctx );
-extern void tdfxUpdateTextureBinding( GLcontext *ctx );
-
-#endif
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_wrapper.h b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_wrapper.h
index 20e5530c5..0adb85b41 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_wrapper.h
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_wrapper.h
@@ -552,8 +552,6 @@ extern FxBool FX_grLfbLock(tdfxContextPtr fxMesa,
     UNLOCK_HARDWARE(fxMesa);						\
   } while (0)
 
-#define FX_grTexDownloadMipMapLevel_NoLock grTexDownloadMipMapLevel
-
 #define FX_grTexDownloadMipMapLevelPartial(fxMesa, t, sa, tlod, llod, ar, f, eo, d, s, e);	\
   do {									    \
     LOCK_HARDWARE(fxMesa);						    \
diff --git a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_xmesa.c b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_xmesa.c
index a0b4304e4..00935e2de 100644
--- a/xc/lib/GL/mesa/src/drv/tdfx/tdfx_xmesa.c
+++ b/xc/lib/GL/mesa/src/drv/tdfx/tdfx_xmesa.c
@@ -52,19 +52,6 @@
 #include "tdfx_texman.h"
 
 
-#ifndef TDFX_DEBUG
-int TDFX_DEBUG = (0
-/*  		  | DEBUG_ALWAYS_SYNC */
-/*		  | DEBUG_VERBOSE_API */
-/*		  | DEBUG_VERBOSE_MSG */
-/*		  | DEBUG_VERBOSE_LRU */
-/*  		  | DEBUG_VERBOSE_DRI */
-/*  		  | DEBUG_VERBOSE_IOCTL */
-/*   		  | DEBUG_VERBOSE_2D */
-   );
-#endif
-
-
 GLboolean
 XMesaInitDriver( __DRIscreenPrivate *sPriv )
 {
@@ -393,11 +380,12 @@ XMesaMakeCurrent( __DRIcontextPrivate *driContextPriv,
 GLboolean
 XMesaOpenFullScreen(__DRIcontextPrivate *driContextPriv)
 {
-    fprintf(stderr,"***** XMesaOpenFullScreen *****\n");
+   if ( 0 )
+      fprintf( stderr, "***** XMesaOpenFullScreen *****\n" );
 #if 0 /* When new glide3 calls exist */
-    return((GLboolean)grDRISetupFullScreen(GL_TRUE));
+   return (GLboolean)grDRISetupFullScreen( GL_TRUE );
 #else
-    return GL_TRUE;
+   return GL_TRUE;
 #endif
 }
 
@@ -405,11 +393,12 @@ XMesaOpenFullScreen(__DRIcontextPrivate *driContextPriv)
 GLboolean
 XMesaCloseFullScreen(__DRIcontextPrivate *driContextPriv)
 {
-    fprintf(stderr,"***** XMesaCloseFullScreen *****\n");
+   if ( 0 )
+      fprintf( stderr, "***** XMesaCloseFullScreen *****\n" );
 #if 0 /* When new glide3 calls exist */
-    return((GLboolean)grDRISetupFullScreen(GL_FALSE));
+   return (GLboolean)grDRISetupFullScreen( GL_FALSE );
 #else
-    return GL_TRUE;
+   return GL_TRUE;
 #endif
 }