Major check-in of changes for GL_EXT_framebuffer_object extension.

Main driver impacts:
- new code for creating the Mesa GLframebuffer
- new span/pixel read/write code
Some drivers not yet updated/tested.

6 files changed, 329 insertions, 89 deletions

diff --git a/src/mesa/drivers/dri/common/depthtmp.h b/src/mesa/drivers/dri/common/depthtmp.h
index ad26e6b891..74273165e4 100644
--- a/src/mesa/drivers/dri/common/depthtmp.h
+++ b/src/mesa/drivers/dri/common/depthtmp.h
@@ -1,5 +1,13 @@
 /* $XFree86: xc/lib/GL/mesa/src/drv/common/depthtmp.h,v 1.5 2001/03/21 16:14:20 dawes Exp $ */
 
+/*
+ * Notes:
+ * 1. These functions plug into the gl_renderbuffer structure.
+ * 2. The 'values' parameter always points to GLuint values, regardless of
+ *    the actual Z buffer depth.
+ */
+
+
 #ifndef DBG
 #define DBG 0
 #endif
@@ -20,12 +28,14 @@
 #endif
 
 static void TAG(WriteDepthSpan)( GLcontext *ctx,
-                             GLuint n, GLint x, GLint y,
-				 const GLdepth *depth,
+                                 struct gl_renderbuffer *rb,
+                                 GLuint n, GLint x, GLint y,
+				 const void *values,
 				 const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLuint *depth = (const GLuint *) values;
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_DEPTH_VARS;
@@ -64,14 +74,31 @@ static void TAG(WriteDepthSpan)( GLcontext *ctx,
    HW_WRITE_UNLOCK();
 }
 
-#if !HAVE_HW_DEPTH_SPANS
+
+#if HAVE_HW_DEPTH_SPANS
+/* implement MonoWriteDepthSpan() in terms of WriteDepthSpan() */
+static void
+TAG(WriteMonoDepthSpan)( GLcontext *ctx, struct gl_renderbuffer *rb,
+                         GLuint n, GLint x, GLint y,
+                         const void *value, const GLubyte mask[] )
+{
+   const GLuint depthVal = *((GLuint *) value);
+   GLuint depths[MAX_WIDTH];
+   GLuint i;
+   for (i = 0; i < n; i++)
+      depths[i] = depthVal;
+   TAG(WriteDepthSpan)(ctx, rb, n, x, y, depths, mask);
+}
+#else
 static void TAG(WriteMonoDepthSpan)( GLcontext *ctx,
-                                 GLuint n, GLint x, GLint y,
-				 const GLdepth depth,
-				 const GLubyte mask[] )
+                                     struct gl_renderbuffer *rb,
+                                     GLuint n, GLint x, GLint y,
+                                     const void *value,
+                                     const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLuint depth = *((GLuint *) value);
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_DEPTH_VARS;
@@ -102,15 +129,18 @@ static void TAG(WriteMonoDepthSpan)( GLcontext *ctx,
 }
 #endif
 
+
 static void TAG(WriteDepthPixels)( GLcontext *ctx,
+                                   struct gl_renderbuffer *rb,
 				   GLuint n,
 				   const GLint x[],
 				   const GLint y[],
-				   const GLdepth depth[],
+				   const void *values,
 				   const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLuint *depth = (const GLuint *) values;
 	 GLuint i;
 	 LOCAL_DEPTH_VARS;
 
@@ -141,11 +171,13 @@ static void TAG(WriteDepthPixels)( GLcontext *ctx,
 /* Read depth spans and pixels
  */
 static void TAG(ReadDepthSpan)( GLcontext *ctx,
+                                struct gl_renderbuffer *rb,
 				GLuint n, GLint x, GLint y,
-				GLdepth depth[] )
+				void *values )
 {
    HW_READ_LOCK()
       {
+         GLuint *depth = (GLuint *) values;
 	 GLint x1, n1;
 	 LOCAL_DEPTH_VARS;
 
@@ -172,12 +204,15 @@ static void TAG(ReadDepthSpan)( GLcontext *ctx,
    HW_READ_UNLOCK();
 }
 
-static void TAG(ReadDepthPixels)( GLcontext *ctx, GLuint n,
+static void TAG(ReadDepthPixels)( GLcontext *ctx,
+                                  struct gl_renderbuffer *rb,
+                                  GLuint n,
 				  const GLint x[], const GLint y[],
-				  GLdepth depth[] )
+				  void *values )
 {
    HW_READ_LOCK()
       {
+         GLuint *depth = (GLuint *) values;
 	 GLuint i;
 	 LOCAL_DEPTH_VARS;
 
diff --git a/src/mesa/drivers/dri/common/drirenderbuffer.c b/src/mesa/drivers/dri/common/drirenderbuffer.c
new file mode 100644
index 0000000000..3d9574f0cf
--- /dev/null
+++ b/src/mesa/drivers/dri/common/drirenderbuffer.c
@@ -0,0 +1,96 @@
+
+#include "mtypes.h"
+#include "drirenderbuffer.h"
+#include "renderbuffer.h"
+#include "imports.h"
+
+
+/**
+ * This will get called when a window is resized.
+ * Just update width, height and internal format fields for now.
+ */
+static GLboolean
+driRenderbufferStorage(GLcontext *ctx, struct gl_renderbuffer *rb,
+                       GLenum internalFormat, GLuint width, GLuint height)
+{
+   rb->Width = width;
+   rb->Height = height;
+   rb->InternalFormat = internalFormat;
+   return GL_TRUE;
+}
+
+
+/**
+ * Allocate a new driRenderbuffer object.
+ * Individual drivers are free to implement different versions of
+ * this function.
+ * \param format  Either GL_RGBA, GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT24,
+ *                GL_DEPTH_COMPONENT32, or GL_STENCIL_INDEX8_EXT (for now).
+ * \param cpp  chars or bytes per pixel
+ * \param offset  start of buffer with respect to framebuffer address
+ * \param pitch   pixels per row
+ */
+driRenderbuffer *
+driNewRenderbuffer(GLenum format, GLint cpp, GLint offset, GLint pitch)
+{
+   driRenderbuffer *drb;
+
+   assert(format == GL_RGBA ||
+          format == GL_DEPTH_COMPONENT16 ||
+          format == GL_DEPTH_COMPONENT24 ||
+          format == GL_DEPTH_COMPONENT32 ||
+          format == GL_STENCIL_INDEX8_EXT);
+
+   assert(cpp > 0);
+   assert(pitch > 0);
+
+   drb = _mesa_calloc(sizeof(driRenderbuffer));
+   if (drb) {
+      const GLuint name = 0;
+
+      _mesa_init_renderbuffer(&drb->Base, name);
+
+      /* Make sure we're using a null-valued GetPointer routine */
+      assert(drb->Base.GetPointer(NULL, &drb->Base, 0, 0) == NULL);
+
+      drb->Base.InternalFormat = format;
+
+      if (format == GL_RGBA) {
+         /* Color */
+         drb->Base._BaseFormat = GL_RGBA;
+         drb->Base.DataType = GL_UNSIGNED_BYTE;
+      }
+      else if (format == GL_DEPTH_COMPONENT16) {
+         /* Depth */
+         drb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+         /* we always Get/Put 32-bit Z values */
+         drb->Base.DataType = GL_UNSIGNED_INT;
+      }
+      else if (format == GL_DEPTH_COMPONENT24) {
+         /* Depth */
+         drb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+         /* we always Get/Put 32-bit Z values */
+         drb->Base.DataType = GL_UNSIGNED_INT;
+      }
+      else {
+         /* Stencil */
+         ASSERT(format == GL_STENCIL_INDEX8);
+         drb->Base._BaseFormat = GL_STENCIL_INDEX;
+         drb->Base.DataType = GL_UNSIGNED_BYTE;
+      }
+
+      /* XXX if we were allocating a user-created renderbuffer, we'd have
+       * to fill in the ComponentSizes[] array too.
+       */
+
+      drb->Base.AllocStorage = driRenderbufferStorage;
+      /* using default Delete function */
+
+
+      /* DRI renderbuffer-specific fields: */
+      drb->offset = offset;
+      drb->pitch = pitch;
+      drb->cpp = cpp;
+   }
+   return drb;
+}
diff --git a/src/mesa/drivers/dri/common/drirenderbuffer.h b/src/mesa/drivers/dri/common/drirenderbuffer.h
new file mode 100644
index 0000000000..627f1d41d9
--- /dev/null
+++ b/src/mesa/drivers/dri/common/drirenderbuffer.h
@@ -0,0 +1,39 @@
+
+/**
+ * A driRenderbuffer is dervied from gl_renderbuffer.
+ * It describes a color buffer (front or back), a depth buffer, or stencil
+ * buffer etc.
+ * Specific to DRI drivers are the offset and pitch fields.
+ */
+
+
+#ifndef DRIRENDERBUFFER_H
+#define DRIRENDERBUFFER_H
+
+#include "mtypes.h"
+
+typedef struct {
+   struct gl_renderbuffer Base;
+
+   /* Chars or bytes per pixel.  If Z and Stencil are stored together this
+    * will typically be 32 whether this a depth or stencil renderbuffer.
+    */
+   GLint cpp;
+
+   /* Buffer position and pitch (row stride).  Recall that for today's DRI
+    * drivers, we have statically allocated color/depth/stencil buffers.
+    * So this information describes the whole screen, not just a window.
+    * To address pixels in a window, we need to know the window's position
+    * and size with respect to the screen.
+    */
+   GLint offset;  /* in bytes */
+   GLint pitch;   /* in pixels */
+
+} driRenderbuffer;
+
+
+driRenderbuffer *
+driNewRenderbuffer(GLenum format, GLint cpp, GLint offset, GLint pitch);
+
+
+#endif /* DRIRENDERBUFFER_H */
diff --git a/src/mesa/drivers/dri/common/spantmp.h b/src/mesa/drivers/dri/common/spantmp.h
index 96f26333fc..b15bbb2577 100644
--- a/src/mesa/drivers/dri/common/spantmp.h
+++ b/src/mesa/drivers/dri/common/spantmp.h
@@ -56,13 +56,14 @@
 #endif
 
 
-static void TAG(WriteRGBASpan)( const GLcontext *ctx,
+static void TAG(WriteRGBASpan)( GLcontext *ctx,
+                                struct gl_renderbuffer *rb,
 				GLuint n, GLint x, GLint y,
-				const GLubyte rgba[][4],
-				const GLubyte mask[] )
+				const void *values, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte (*rgba)[4] = (const GLubyte (*)[4]) values;
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_VARS;
@@ -98,13 +99,14 @@ static void TAG(WriteRGBASpan)( const GLcontext *ctx,
    HW_WRITE_UNLOCK();
 }
 
-static void TAG(WriteRGBSpan)( const GLcontext *ctx,
+static void TAG(WriteRGBSpan)( GLcontext *ctx,
+                               struct gl_renderbuffer *rb,
 			       GLuint n, GLint x, GLint y,
-			       const GLubyte rgb[][3],
-			       const GLubyte mask[] )
+			       const void *values, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte (*rgb)[3] = (const GLubyte (*)[3]) values;
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_VARS;
@@ -136,15 +138,14 @@ static void TAG(WriteRGBSpan)( const GLcontext *ctx,
    HW_WRITE_UNLOCK();
 }
 
-static void TAG(WriteRGBAPixels)( const GLcontext *ctx,
-			       GLuint n,
-			       const GLint x[],
-			       const GLint y[],
-			       const GLubyte rgba[][4],
-			       const GLubyte mask[] )
+static void TAG(WriteRGBAPixels)( GLcontext *ctx,
+                                  struct gl_renderbuffer *rb,
+                                  GLuint n, const GLint x[], const GLint y[],
+                                  const void *values, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte (*rgba)[4] = (const GLubyte (*)[4]) values;
 	 GLuint i;
 	 LOCAL_VARS;
 
@@ -183,13 +184,15 @@ static void TAG(WriteRGBAPixels)( const GLcontext *ctx,
 }
 
 
-static void TAG(WriteMonoRGBASpan)( const GLcontext *ctx,	
+static void TAG(WriteMonoRGBASpan)( GLcontext *ctx,	
+                                    struct gl_renderbuffer *rb,
 				    GLuint n, GLint x, GLint y, 
-				    const GLchan color[4],
+				    const void *value,
 				    const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte *color = (const GLubyte *) value;
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_VARS;
@@ -221,14 +224,16 @@ static void TAG(WriteMonoRGBASpan)( const GLcontext *ctx,
 }
 
 
-static void TAG(WriteMonoRGBAPixels)( const GLcontext *ctx,
+static void TAG(WriteMonoRGBAPixels)( GLcontext *ctx,
+                                      struct gl_renderbuffer *rb,
 				      GLuint n,
-				      const GLint x[], const GLint y[],
-				      const GLchan color[],
-				      const GLubyte mask[] ) 
+                                      const GLint x[], const GLint y[],
+				      const void *value,
+                                      const GLubyte mask[] ) 
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte *color = (const GLubyte *) value;
 	 GLuint i;
 	 LOCAL_VARS;
 	 INIT_MONO_PIXEL(p, color);
@@ -261,12 +266,14 @@ static void TAG(WriteMonoRGBAPixels)( const GLcontext *ctx,
 }
 
 
-static void TAG(ReadRGBASpan)( const GLcontext *ctx,
+static void TAG(ReadRGBASpan)( GLcontext *ctx,
+                               struct gl_renderbuffer *rb,
 			       GLuint n, GLint x, GLint y,
-			       GLubyte rgba[][4])
+			       void *values)
 {
    HW_READ_LOCK()
       {
+         GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
 	 GLint x1,n1;
 	 LOCAL_VARS;
 
@@ -287,12 +294,15 @@ static void TAG(ReadRGBASpan)( const GLcontext *ctx,
 }
 
 
-static void TAG(ReadRGBAPixels)( const GLcontext *ctx,
+static void TAG(ReadRGBAPixels)( GLcontext *ctx,
+                                 struct gl_renderbuffer *rb,
 				 GLuint n, const GLint x[], const GLint y[],
-				 GLubyte rgba[][4], const GLubyte mask[] )
+				 void *values )
 {
    HW_READ_LOCK()
       {
+         GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
+         const GLubyte *mask = NULL; /* remove someday */
 	 GLuint i;
 	 LOCAL_VARS;
 
diff --git a/src/mesa/drivers/dri/common/spantmp2.h b/src/mesa/drivers/dri/common/spantmp2.h
index fa737a9623..e15491d1ad 100644
--- a/src/mesa/drivers/dri/common/spantmp2.h
+++ b/src/mesa/drivers/dri/common/spantmp2.h
@@ -145,13 +145,14 @@
 #include "x86/common_x86_asm.h"
 #endif
 
-static void TAG(WriteRGBASpan)( const GLcontext *ctx,
+static void TAG(WriteRGBASpan)( GLcontext *ctx,
+                                struct gl_renderbuffer *rb,
 				GLuint n, GLint x, GLint y,
-				const GLubyte rgba[][4],
-				const GLubyte mask[] )
+				const void *values, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte (*rgba)[4] = (const GLubyte (*)[4]) values;
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_VARS;
@@ -187,13 +188,14 @@ static void TAG(WriteRGBASpan)( const GLcontext *ctx,
    HW_WRITE_UNLOCK();
 }
 
-static void TAG(WriteRGBSpan)( const GLcontext *ctx,
+static void TAG(WriteRGBSpan)( GLcontext *ctx,
+                               struct gl_renderbuffer *rb,
 			       GLuint n, GLint x, GLint y,
-			       const GLubyte rgb[][3],
-			       const GLubyte mask[] )
+			       const void *values, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte (*rgb)[3] = (const GLubyte (*)[3]) values;
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_VARS;
@@ -225,15 +227,14 @@ static void TAG(WriteRGBSpan)( const GLcontext *ctx,
    HW_WRITE_UNLOCK();
 }
 
-static void TAG(WriteRGBAPixels)( const GLcontext *ctx,
-			       GLuint n,
-			       const GLint x[],
-			       const GLint y[],
-			       const GLubyte rgba[][4],
-			       const GLubyte mask[] )
+static void TAG(WriteRGBAPixels)( GLcontext *ctx,
+                                  struct gl_renderbuffer *rb,
+                                  GLuint n, const GLint x[], const GLint y[],
+                                  const void *values, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte (*rgba)[4] = (const GLubyte (*)[4]) values;
 	 GLint i;
 	 LOCAL_VARS;
 
@@ -272,13 +273,14 @@ static void TAG(WriteRGBAPixels)( const GLcontext *ctx,
 }
 
 
-static void TAG(WriteMonoRGBASpan)( const GLcontext *ctx,	
+static void TAG(WriteMonoRGBASpan)( GLcontext *ctx,	
+                                    struct gl_renderbuffer *rb,
 				    GLuint n, GLint x, GLint y, 
-				    const GLchan color[4],
-				    const GLubyte mask[] )
+				    const void *value, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte *color = (const GLubyte *) value;
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_VARS;
@@ -310,14 +312,16 @@ static void TAG(WriteMonoRGBASpan)( const GLcontext *ctx,
 }
 
 
-static void TAG(WriteMonoRGBAPixels)( const GLcontext *ctx,
+static void TAG(WriteMonoRGBAPixels)( GLcontext *ctx,
+                                      struct gl_renderbuffer *rb,
 				      GLuint n,
 				      const GLint x[], const GLint y[],
-				      const GLchan color[],
+				      const void *value,
 				      const GLubyte mask[] ) 
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte *color = (const GLubyte *) value;
 	 GLint i;
 	 LOCAL_VARS;
 	 INIT_MONO_PIXEL(p, color);
@@ -350,12 +354,13 @@ static void TAG(WriteMonoRGBAPixels)( const GLcontext *ctx,
 }
 
 
-static void TAG(ReadRGBASpan)( const GLcontext *ctx,
-			       GLuint n, GLint x, GLint y,
-			       GLubyte rgba[][4])
+static void TAG(ReadRGBASpan)( GLcontext *ctx,
+                               struct gl_renderbuffer *rb,
+			       GLuint n, GLint x, GLint y, void *values)
 {
    HW_READ_LOCK()
       {
+         GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
 	 GLint x1,n1;
 	 LOCAL_VARS;
 
@@ -381,9 +386,9 @@ static void TAG(ReadRGBASpan)( const GLcontext *ctx,
 	(SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)) || \
     ((SPANTMP_PIXEL_FMT == GL_RGB) && \
 	(SPANTMP_PIXEL_TYPE == GL_UNSIGNED_SHORT_5_6_5)))
-static void TAG2(ReadRGBASpan,_MMX)( const GLcontext *ctx,
-			       GLuint n, GLint x, GLint y,
-			       GLubyte rgba[][4])
+static void TAG2(ReadRGBASpan,_MMX)( GLcontext *ctx,
+                                     GLuint n, GLint x, GLint y,
+                                     GLubyte rgba[][4])
 {
 #ifndef USE_INNER_EMMS
    /* The EMMS instruction is directly in-lined here because using GCC's
@@ -394,6 +399,7 @@ static void TAG2(ReadRGBASpan,_MMX)( const GLcontext *ctx,
 
    HW_READ_LOCK()
      {
+        GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
 	GLint x1,n1;
 	LOCAL_VARS;
 
@@ -429,12 +435,14 @@ static void TAG2(ReadRGBASpan,_MMX)( const GLcontext *ctx,
 #if defined(USE_SSE_ASM) && \
    (SPANTMP_PIXEL_FMT == GL_BGRA) && \
      (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
-static void TAG2(ReadRGBASpan,_SSE2)( const GLcontext *ctx,
-			       GLuint n, GLint x, GLint y,
-			       GLubyte rgba[][4])
+static void TAG2(ReadRGBASpan,_SSE2)( GLcontext *ctx,
+                                      struct gl_renderbuffer *rb,
+                                      GLuint n, GLint x, GLint y,
+                                      void *values)
 {
    HW_READ_LOCK()
      {
+        GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
 	GLint x1,n1;
 	LOCAL_VARS;
 
@@ -461,9 +469,10 @@ static void TAG2(ReadRGBASpan,_SSE2)( const GLcontext *ctx,
 #if defined(USE_SSE_ASM) && \
    (SPANTMP_PIXEL_FMT == GL_BGRA) && \
      (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
-static void TAG2(ReadRGBASpan,_SSE)( const GLcontext *ctx,
-			       GLuint n, GLint x, GLint y,
-			       GLubyte rgba[][4])
+static void TAG2(ReadRGBASpan,_SSE)( GLcontext *ctx,
+                                     struct gl_renderbuffer *rb,
+                                     GLuint n, GLint x, GLint y,
+                                     void *values)
 {
 #ifndef USE_INNER_EMMS
    /* The EMMS instruction is directly in-lined here because using GCC's
@@ -474,6 +483,7 @@ static void TAG2(ReadRGBASpan,_SSE)( const GLcontext *ctx,
 
    HW_READ_LOCK()
      {
+        GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
 	GLint x1,n1;
 	LOCAL_VARS;
 
@@ -501,12 +511,15 @@ static void TAG2(ReadRGBASpan,_SSE)( const GLcontext *ctx,
 #endif
 
 
-static void TAG(ReadRGBAPixels)( const GLcontext *ctx,
+static void TAG(ReadRGBAPixels)( GLcontext *ctx,
+                                 struct gl_renderbuffer *rb,
 				 GLuint n, const GLint x[], const GLint y[],
-				 GLubyte rgba[][4], const GLubyte mask[] )
+				 void *values )
 {
    HW_READ_LOCK()
       {
+         GLubyte (*rgba)[4] = (GLubyte (*)[4]) values;
+         GLubyte *mask = NULL; /* remove someday */
 	 GLint i;
 	 LOCAL_VARS;
 
@@ -537,21 +550,21 @@ static void TAG(ReadRGBAPixels)( const GLcontext *ctx,
    HW_READ_UNLOCK();
 }
 
-static void TAG(InitPointers)(struct swrast_device_driver *swdd)
+static void TAG(InitPointers)(struct gl_renderbuffer *rb)
 {
-   swdd->WriteRGBASpan = TAG(WriteRGBASpan);
-   swdd->WriteRGBSpan = TAG(WriteRGBSpan);
-   swdd->WriteMonoRGBASpan = TAG(WriteMonoRGBASpan);
-   swdd->WriteRGBAPixels = TAG(WriteRGBAPixels);
-   swdd->WriteMonoRGBAPixels = TAG(WriteMonoRGBAPixels);
-   swdd->ReadRGBAPixels = TAG(ReadRGBAPixels);
+   rb->PutRow = TAG(WriteRGBASpan);
+   rb->PutRowRGB = TAG(WriteRGBSpan);
+   rb->PutMonoRow = TAG(WriteMonoRGBASpan);
+   rb->PutValues = TAG(WriteRGBAPixels);
+   rb->PutMonoValues = TAG(WriteMonoRGBAPixels);
+   rb->GetValues = TAG(ReadRGBAPixels);
 
 #if defined(USE_SSE_ASM) && \
    (SPANTMP_PIXEL_FMT == GL_BGRA) && \
      (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
    if ( cpu_has_xmm2 ) {
-      if (DBG) fprintf( stderr, "Using %s version of ReadRGBASpan\n", "SSE2" );
-      swdd->ReadRGBASpan = TAG2(ReadRGBASpan, _SSE2);
+      if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "SSE2" );
+      rb->GetRow = TAG2(ReadRGBASpan, _SSE2);
    }
    else
 #endif
@@ -559,8 +572,8 @@ static void TAG(InitPointers)(struct swrast_device_driver *swdd)
    (SPANTMP_PIXEL_FMT == GL_BGRA) && \
      (SPANTMP_PIXEL_TYPE == GL_UNSIGNED_INT_8_8_8_8_REV)
    if ( cpu_has_xmm ) {
-      if (DBG) fprintf( stderr, "Using %s version of ReadRGBASpan\n", "SSE" );
-      swdd->ReadRGBASpan = TAG2(ReadRGBASpan, _SSE);
+      if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "SSE" );
+      rb->GetRow = TAG2(ReadRGBASpan, _SSE);
    }
    else
 #endif
@@ -570,14 +583,14 @@ static void TAG(InitPointers)(struct swrast_device_driver *swdd)
     ((SPANTMP_PIXEL_FMT == GL_RGB) && \
 	(SPANTMP_PIXEL_TYPE == GL_UNSIGNED_SHORT_5_6_5)))
    if ( cpu_has_mmx ) {
-      if (DBG) fprintf( stderr, "Using %s version of ReadRGBASpan\n", "MMX" );
-      swdd->ReadRGBASpan = TAG2(ReadRGBASpan, _MMX);
+      if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "MMX" );
+      rb->GetRow = TAG2(ReadRGBASpan, _MMX);
    }
    else
 #endif
    {
-      if (DBG) fprintf( stderr, "Using %s version of ReadRGBASpan\n", "C" );
-      swdd->ReadRGBASpan = TAG(ReadRGBASpan);
+      if (DBG) fprintf( stderr, "Using %s version of GetRow\n", "C" );
+      rb->GetRow = TAG(ReadRGBASpan);
    }
 
 }
diff --git a/src/mesa/drivers/dri/common/stenciltmp.h b/src/mesa/drivers/dri/common/stenciltmp.h
index dfbe665407..ffcb1a4c1e 100644
--- a/src/mesa/drivers/dri/common/stenciltmp.h
+++ b/src/mesa/drivers/dri/common/stenciltmp.h
@@ -19,12 +19,13 @@
 #endif
 
 static void TAG(WriteStencilSpan)( GLcontext *ctx,
+                                   struct gl_renderbuffer *rb,
 				   GLuint n, GLint x, GLint y,
-				   const GLstencil *stencil, 
-				   const GLubyte mask[] )
+				   const void *values, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte *stencil = (const GLubyte *) values;
 	 GLint x1;
 	 GLint n1;
 	 LOCAL_STENCIL_VARS;
@@ -57,15 +58,57 @@ static void TAG(WriteStencilSpan)( GLcontext *ctx,
 }
 
 
+static void TAG(WriteMonoStencilSpan)( GLcontext *ctx,
+                                       struct gl_renderbuffer *rb,
+                                       GLuint n, GLint x, GLint y,
+                                       const void *value,
+                                       const GLubyte mask[] )
+{
+   HW_WRITE_LOCK()
+      {
+         const GLubyte stencil = *((const GLubyte *) value);
+	 GLint x1;
+	 GLint n1;
+	 LOCAL_STENCIL_VARS;
+
+	 y = Y_FLIP(y);
+
+	 HW_CLIPLOOP() 
+	    {
+	       GLint i = 0;
+	       CLIPSPAN(x,y,n,x1,n1,i);
+
+	       if (DBG) fprintf(stderr, "WriteStencilSpan %d..%d (x1 %d)\n",
+				(int)i, (int)n1, (int)x1);
+
+	       if (mask)
+	       {
+		  for (;n1>0;i++,x1++,n1--)
+		     if (mask[i])
+			WRITE_STENCIL( x1, y, stencil );
+	       }
+	       else
+	       {
+		  for (;n1>0;i++,x1++,n1--)
+		     WRITE_STENCIL( x1, y, stencil );
+	       }
+	    }
+	 HW_ENDCLIPLOOP();
+      }
+   HW_WRITE_UNLOCK();
+}
+
+
+
 static void TAG(WriteStencilPixels)( GLcontext *ctx,
-				     GLuint n, 
-				     const GLint x[], 
-				     const GLint y[],
-				     const GLstencil stencil[], 
-				     const GLubyte mask[] )
+                                     struct gl_renderbuffer *rb,
+				     GLuint n,
+				     const GLint x[], const GLint y[],
+				     const void *values, const GLubyte mask[] )
 {
    HW_WRITE_LOCK()
       {
+         const GLubyte *stencil = (const GLubyte *) values;
 	 GLuint i;
 	 LOCAL_STENCIL_VARS;
 
@@ -91,11 +134,13 @@ static void TAG(WriteStencilPixels)( GLcontext *ctx,
 /* Read stencil spans and pixels
  */
 static void TAG(ReadStencilSpan)( GLcontext *ctx,
+                                  struct gl_renderbuffer *rb,
 				  GLuint n, GLint x, GLint y,
-				  GLstencil stencil[])
+				  void *values)
 {
    HW_READ_LOCK()
       {
+         GLubyte *stencil = (GLubyte *) values;
 	 GLint x1,n1;
 	 LOCAL_STENCIL_VARS;
 
@@ -115,12 +160,14 @@ static void TAG(ReadStencilSpan)( GLcontext *ctx,
    HW_READ_UNLOCK();
 }
 
-static void TAG(ReadStencilPixels)( GLcontext *ctx, GLuint n, 
-				    const GLint x[], const GLint y[],
-				    GLstencil stencil[] )
+static void TAG(ReadStencilPixels)( GLcontext *ctx,
+                                    struct gl_renderbuffer *rb,
+                                    GLuint n, const GLint x[], const GLint y[],
+				    void *values )
 {
    HW_READ_LOCK()
       {
+         GLubyte *stencil = (GLubyte *) values;
 	 GLuint i;
 	 LOCAL_STENCIL_VARS;