Performance work for the standard (non-eltpath) rendering pipelines. Good

speedups for geometry-bound apps like isosurf.

13 files changed, 218 insertions, 169 deletions

diff --git a/xc/lib/GL/mesa/src/drv/mga/Imakefile b/xc/lib/GL/mesa/src/drv/mga/Imakefile
index 3eec12d29..f01632491 100644
--- a/xc/lib/GL/mesa/src/drv/mga/Imakefile
+++ b/xc/lib/GL/mesa/src/drv/mga/Imakefile
@@ -45,13 +45,6 @@ MESA_INCLUDES = -I. -I.. -I../../include
       DEFINES = $(ALLOC_DEFINES) $(DRI_DEFINES) $(ASM_DEFS)
      INCLUDES = -I$(XLIBSRC) -I$(EXTINCSRC) $(MESA_INCLUDES) $(DRI_INCLUDES) 
 
-#if 0
-	LOSRC = ../../../../lowpc.c
-	LOOBJ = ../../../../lowpc.o
-
-	HISRC = ../../../../highpc.c
-	HIOBJ = ../../../../highpc.o
-#endif
 
       DRISRCS = ../../../dri/dri_mesa.c \
 		../../../../dri/dri_tmm.c
@@ -275,14 +268,23 @@ MESA_INCLUDES = -I. -I.. -I../../include
 #endif
 #endif
 
+#ifdef GlxSoProf
+       HIOBJS = ../../../../highpc.o
+       LOOBJS = ../../../../lowpc.o
+       HISRCS = ../../../../highpc.c
+       LOSRCS = ../../../../lowpc.c
+#endif
+
       ASMSRCS = $(X86_SRCS) $(MMX_SRCS) $(3DNOW_SRCS)
       ASMOBJS = $(X86_OBJS) $(MMX_OBJS) $(3DNOW_OBJS)
 
       COMMONSRCS = ../common/mm.c ../common/hwlog.c
       COMMONOBJS = ../common/mm.o ../common/hwlog.o
 
-         SRCS = $(LOWSRC) $(DRISRCS) $(DRMSRCS) $(MESASRCS) $(ASMSRCS) $(COMMONSRCS) $(MGASRCS) $(HISRC)
-         OBJS = $(LOWOBJ) $(DRIOBJS) $(DRMOBJS) $(MESAOBJS) $(ASMOBJS) $(COMMONOBJS) $(MGAOBJS) $(HIOBJ)
+         SRCS = $(LOSRCS) $(DRISRCS) $(DRMSRCS) $(MESASRCS) $(ASMSRCS)	\
+		$(COMMONSRCS) $(MGASRCS) $(HISRCS)
+         OBJS = $(LOOBJS) $(DRIOBJS) $(DRMOBJS) $(MESAOBJS) $(ASMOBJS)	\
+		$(COMMONOBJS) $(MGAOBJS) $(HIOBJS)
 
 REQUIREDLIBS += -lm
 #if !GlxBuiltInMga
@@ -310,6 +312,12 @@ ALL_OBJS = $(OBJS)
 ALL_DEPS = DONE
 SharedDepModuleTarget($(LIBNAME),$(ALL_DEPS),$(ALL_OBJS))
 InstallDynamicModule($(LIBNAME),$(MODULEDIR),dri)
+
+#ifdef GlxSoProf
+SOPROF_LIBNAME = _mga_dri_p
+NormalDepLibraryTarget($(SOPROF_LIBNAME),$(ALL_DEPS),$(ALL_OBJS))
+InstallLibrary($(SOPROF_LIBNAME),$(MODULEDIR)/dri)
+#endif
 #endif
 
 DependTarget()
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgabuffers.c b/xc/lib/GL/mesa/src/drv/mga/mgabuffers.c
index 851d66032..0431361d2 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgabuffers.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgabuffers.c
@@ -37,6 +37,7 @@ SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "mgalib.h"
 #include "mgabuffers.h"
 #include "mgastate.h"
+#include "mgaioctl.h"
 
 static void mgaXMesaSetFrontClipRects( mgaContextPtr mmesa )
 {
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaelttmp.h b/xc/lib/GL/mesa/src/drv/mga/mgaelttmp.h
index ce758a532..19c913f5a 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaelttmp.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaelttmp.h
@@ -83,8 +83,6 @@ static void TAG(emit_unclipped_verts)( struct vertex_buffer *VB )
 	 }
 
 	 if (TYPE & MGA_TEX0_BIT) {
-/*  	    fprintf(stderr, "i %d tex0 %f, %f\n", i,  */
-/*  		    tex0_data[0], tex0_data[1]); */
 	    *(int*)&f[6] = *(int*)&tex0_data[0];
 	    *(int*)&f[7] = *(int*)&tex0_data[1];
 	 }
@@ -127,19 +125,9 @@ static void TAG(build_tri_verts)( mgaContextPtr mmesa,
 
       if (TYPE & MGA_RGBA_BIT) {
 	 GLubyte *col = VEC_ELT(VB->ColorPtr, GLubyte, elt[i]);
-	 GLubyte *b = (GLubyte *)&O[4];
-	 b[CLIP_UBYTE_R] = col[0];
-	 b[CLIP_UBYTE_G] = col[1];
-	 b[CLIP_UBYTE_B] = col[2];
-	 b[CLIP_UBYTE_A] = col[3];
+	 *(GLuint *)&O[4] = *(GLuint *)col;
       }
 
-      if (0)
-	 fprintf(stderr, 
-	      "build_tri_vert elt[%d]: %d phys: %x (first_phys %x elt_buf %x\n",
-	      i, elt[i], UNCLIPPED_VERT(elt[i]), 
-	      mmesa->first_vert_phys, (GLuint)mmesa->elt_buf);
-
       *(GLuint *)&O[5] = UNCLIPPED_VERT(elt[i]);
 
       if (TYPE & MGA_TEX0_BIT) {
@@ -176,8 +164,6 @@ static void TAG(interp)( GLfloat t,
 		  ((GLubyte *)&(J[4])));
    }
 
-   if (0) fprintf(stderr, "setting 0x%x to ~0\n", (GLuint)&O[5]);
-
    *(GLuint *)&O[5] = ~0;	/* note that this is a new vertex */
  
    if (TYPE & MGA_TEX0_BIT) {
@@ -215,10 +201,6 @@ static void TAG(project_and_emit_verts)( mgaContextPtr mmesa,
       const GLfloat *I = &verts[elt[i] * CLIP_STRIDE];
       GLuint tmp = *(GLuint *)&I[5];
 
-      if (0) fprintf(stderr, "elt[%d] (tmp 0x%x %d) %d --> ", i, (GLuint)&I[5],
-	      tmp, elt[i]);
-
-
       if ((elt[i] = tmp) == ~0) 
       {      
 	 GLfloat oow = 1.0/I[3];
@@ -247,7 +229,6 @@ static void TAG(project_and_emit_verts)( mgaContextPtr mmesa,
 
 	 O -= BUFFER_STRIDE;
       }
-      if (0) fprintf(stderr, "0x%x\n", elt[i]);
    }
 
    mmesa->next_vert = O;
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
index edacb9115..a1875d85d 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.c
@@ -80,7 +80,7 @@ int mgaUpdateLock( mgaContextPtr mmesa, drmLockFlags flags )
    return 0;
 }
 
-static drmBufPtr mga_get_buffer_ioctl( mgaContextPtr mmesa )
+drmBufPtr mga_get_buffer_ioctl( mgaContextPtr mmesa )
 {
    int idx = 0;
    int size = 0;
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h
index 900e4a4f8..96be18ee1 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaioctl.h
@@ -2,6 +2,7 @@
 #define MGA_IOCTL_H
 
 #include "mgalib.h"
+#include "mga_xmesa.h"
 
 GLbitfield mgaClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 		     GLint cx, GLint cy, GLint cw, GLint ch ); 
@@ -54,5 +55,36 @@ void mgaDDInitIoctlFuncs( GLcontext *ctx );
 	else if (mmesa->next_elt != mmesa->first_elt) mgaFlushElts(mmesa);	        	\
 } while (0)
 
+extern drmBufPtr mga_get_buffer_ioctl( mgaContextPtr mmesa );
+
+static __inline
+mgaUI32 *mgaAllocVertexDwordsInline( mgaContextPtr mmesa, int dwords )
+{
+   int bytes = dwords * 4;
+   mgaUI32 *head;
+
+   if (!mmesa->vertex_dma_buffer) {
+      LOCK_HARDWARE( mmesa );
+
+      if (mmesa->first_elt != mmesa->next_elt) 
+	 mgaFlushEltsLocked(mmesa);
+
+      mmesa->vertex_dma_buffer = mga_get_buffer_ioctl( mmesa );
+      UNLOCK_HARDWARE( mmesa );
+   } else if (mmesa->vertex_dma_buffer->used + bytes > 
+	      mmesa->vertex_dma_buffer->total) {
+      LOCK_HARDWARE( mmesa );
+      mgaFlushVerticesLocked( mmesa );
+      mmesa->vertex_dma_buffer = mga_get_buffer_ioctl( mmesa );
+      UNLOCK_HARDWARE( mmesa );
+   }
+
+   head = (mgaUI32 *)((char *)mmesa->vertex_dma_buffer->address + 
+		      mmesa->vertex_dma_buffer->used);
+
+   mmesa->vertex_dma_buffer->used += bytes;
+   return head;
+}
+
 
 #endif
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgalib.h b/xc/lib/GL/mesa/src/drv/mga/mgalib.h
index be843ba8f..1f8c64909 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgalib.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgalib.h
@@ -39,7 +39,6 @@
 #include "mgacommon.h"
 #include "mm.h"
 #include "mgalog.h"
-#include "mgaioctl.h"
 #include "mgatex.h"
 #include "mgavb.h"
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgaspan.c b/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
index 4b0bd9ee0..25553bcff 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgaspan.c
@@ -3,6 +3,7 @@
 #include "mgalib.h"
 #include "mgalog.h"
 #include "mgaspan.h"
+#include "mgaioctl.h"
 
 #define DBG 0
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatex.c b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
index ebd9ce486..b095807fd 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatex.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatex.c
@@ -37,6 +37,7 @@
 #include "mgatex.h"
 #include "mgalog.h"
 #include "mgaregs.h"
+#include "mgaioctl.h"
 
 #include "simple_list.h"
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatris.c b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
index 4f5ff299b..2942cec7e 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatris.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatris.c
@@ -108,6 +108,7 @@ static void mgaPrintRenderState( const char *msg, GLuint state )
 #define TAG(x) x##_twoside_offset_flat
 #include "mgatritmp.h"
 
+
 void mgaDDTrifuncInit()
 {
    int i;
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgatris.h b/xc/lib/GL/mesa/src/drv/mga/mgatris.h
index 8d340e3ad..bbebbc400 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatris.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatris.h
@@ -28,6 +28,7 @@
 #define MGATIS_INC
 
 #include "types.h"
+#include "mgaioctl.h"
 
 extern void mgaDDChooseRenderState(GLcontext *ctx);
 extern void mgaDDTrifuncInit( void );
@@ -48,9 +49,28 @@ static __inline void mga_draw_triangle( mgaContextPtr mmesa,
 				      mgaVertex *v2 )
 {
    mgaUI32 vertsize = mmesa->vertsize;
-   mgaUI32 *wv = mgaAllocVertexDwords( mmesa, 3 * vertsize );
+   mgaUI32 *wv = mgaAllocVertexDwordsInline( mmesa, 3 * vertsize );
    int j;
 
+#if defined (USE_X86_ASM)
+    /* GTH: We can safely assume the vertex stride is some number of
+     * dwords, and thus a "rep movsd" is okay.  The vb pointer is
+     * automagically updated with this instruction, so we don't have
+     * to manually take care of incrementing it.
+     */
+    __asm__ __volatile__( "rep ; movsl"
+			  : "=%c" (j)
+			  : "0" (vertsize), "D" ((long)wv), "S" ((long)v0)
+			  : "memory" );
+    __asm__ __volatile__( "rep ; movsl"
+			  : "=%c" (j)
+			  : "0" (vertsize), "S" ((long)v1)
+			  : "memory" );
+    __asm__ __volatile__( "rep ; movsl"
+			  : "=%c" (j)
+			  : "0" (vertsize), "S" ((long)v2)
+			  : "memory" );
+#else
    for (j = 0 ; j < vertsize ; j++) 
       wv[j] = v0->ui[j];
 
@@ -61,6 +81,7 @@ static __inline void mga_draw_triangle( mgaContextPtr mmesa,
    wv += vertsize;
    for (j = 0 ; j < vertsize ; j++) 
       wv[j] = v2->ui[j];
+#endif
 }
 
 
@@ -71,38 +92,38 @@ static __inline void mga_draw_point( mgaContextPtr mmesa,
    mgaUI32 *wv = mgaAllocVertexDwords( mmesa, 6*vertsize);
    int j;
 
-   *(float *)&wv[0] = tmp->warp1.x - sz;
-   *(float *)&wv[1] = tmp->warp1.y - sz;
+   *(float *)&wv[0] = tmp->v.x - sz;
+   *(float *)&wv[1] = tmp->v.y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->warp1.x + sz;
-   *(float *)&wv[1] = tmp->warp1.y - sz;
+   *(float *)&wv[0] = tmp->v.x + sz;
+   *(float *)&wv[1] = tmp->v.y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->warp1.x + sz;
-   *(float *)&wv[1] = tmp->warp1.y + sz;
+   *(float *)&wv[0] = tmp->v.x + sz;
+   *(float *)&wv[1] = tmp->v.y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->warp1.x + sz;
-   *(float *)&wv[1] = tmp->warp1.y + sz;
+   *(float *)&wv[0] = tmp->v.x + sz;
+   *(float *)&wv[1] = tmp->v.y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->warp1.x - sz;
-   *(float *)&wv[1] = tmp->warp1.y + sz;
+   *(float *)&wv[0] = tmp->v.x - sz;
+   *(float *)&wv[1] = tmp->v.y + sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp->warp1.x - sz;
-   *(float *)&wv[1] = tmp->warp1.y - sz;
+   *(float *)&wv[0] = tmp->v.x - sz;
+   *(float *)&wv[1] = tmp->v.y - sz;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp->ui[j];
 }
@@ -118,8 +139,8 @@ static __inline void mga_draw_line( mgaContextPtr mmesa,
    float dx, dy, ix, iy;
    int j;
 
-   dx = tmp0->warp1.x - tmp1->warp1.x;
-   dy = tmp0->warp1.y - tmp1->warp1.y;
+   dx = tmp0->v.x - tmp1->v.x;
+   dy = tmp0->v.y - tmp1->v.y;
 
    ix = width * .5; iy = 0;
   
@@ -130,38 +151,38 @@ static __inline void mga_draw_line( mgaContextPtr mmesa,
       iy = ix; ix = 0;
    }
   
-   *(float *)&wv[0] = tmp0->warp1.x - ix;
-   *(float *)&wv[1] = tmp0->warp1.y - iy;
+   *(float *)&wv[0] = tmp0->v.x - ix;
+   *(float *)&wv[1] = tmp0->v.y - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->warp1.x + ix;
-   *(float *)&wv[1] = tmp1->warp1.y + iy;
+   *(float *)&wv[0] = tmp1->v.x + ix;
+   *(float *)&wv[1] = tmp1->v.y + iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp1->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp0->warp1.x + ix;
-   *(float *)&wv[1] = tmp0->warp1.y + iy;
+   *(float *)&wv[0] = tmp0->v.x + ix;
+   *(float *)&wv[1] = tmp0->v.y + iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 	 
-   *(float *)&wv[0] = tmp0->warp1.x - ix;
-   *(float *)&wv[1] = tmp0->warp1.y - iy;
+   *(float *)&wv[0] = tmp0->v.x - ix;
+   *(float *)&wv[1] = tmp0->v.y - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp0->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->warp1.x - ix;
-   *(float *)&wv[1] = tmp1->warp1.y - iy;
+   *(float *)&wv[0] = tmp1->v.x - ix;
+   *(float *)&wv[1] = tmp1->v.y - iy;
    for (j = 2 ; j < vertsize ; j++) 
       wv[j] = tmp1->ui[j];
    wv += vertsize;
 
-   *(float *)&wv[0] = tmp1->warp1.x + ix;
-   *(float *)&wv[1] = tmp1->warp1.y + iy;
+   *(float *)&wv[0] = tmp1->v.x + ix;
+   *(float *)&wv[1] = tmp1->v.y + 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 cc2b8a359..c1b81aa4d 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgatritmp.h
@@ -1,94 +1,108 @@
 
-static __inline void TAG(triangle)( GLcontext *ctx, GLuint e0,
-				  GLuint e1, GLuint e2, GLuint pv )
+static __inline void TAG(triangle)(GLcontext *ctx,
+				   GLuint e0, GLuint e1, GLuint e2,
+				   GLuint pv)
 {
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   GLuint vertsize = mmesa->vertsize;
-   mgaUI32 *wv = mgaAllocVertexDwords( mmesa, 3 * vertsize);
-
-   struct vertex_buffer *VB = ctx->VB;
-   mgaVertexPtr mgaVB = MGA_DRIVER_DATA(VB)->verts;
-   const mgaVertex *v[3];
-   int i, j;
+   mgaContextPtr        mgactx   = MGA_CONTEXT(ctx);
+   struct vertex_buffer *VB        = ctx->VB;
+   mgaVertexPtr         mgaverts = MGA_DRIVER_DATA(VB)->verts;
+   mgaVertex     *v[3];
 
 #if (IND & MGA_OFFSET_BIT)
-   GLfloat offset = 0.0;
+   GLfloat offset;
+   GLfloat z[3];
 #endif
 
-#if (IND & (MGA_FLAT_BIT|MGA_TWOSIDE_BIT))
-   mgaUI32 c[3];
-
-   c[2] = c[1] = c[0] = *(mgaUI32 *)&mgaVB[pv].warp2.color; 
+#if (IND & (MGA_TWOSIDE_BIT | MGA_FLAT_BIT))
+   GLuint c[3];
 #endif
 
-   (void) VB;
+   v[0] = &mgaverts[e0];
+   v[1] = &mgaverts[e1];
+   v[2] = &mgaverts[e2];
 
-   v[0] = &mgaVB[e0];  
-   v[1] = &mgaVB[e1];  
-   v[2] = &mgaVB[e2];  
+#if (IND & (MGA_TWOSIDE_BIT | MGA_FLAT_BIT))
+   c[0] = v[0]->ui[4];
+   c[1] = v[1]->ui[4];
+   c[2] = v[2]->ui[4];
+#endif
 
 
-#if (IND & (MGA_TWOSIDE_BIT|MGA_OFFSET_BIT)) 
+#if (IND & (MGA_TWOSIDE_BIT | MGA_OFFSET_BIT))
    {
-      GLfloat ex = v[0]->warp1.x - v[2]->warp1.x;
-      GLfloat ey = v[0]->warp1.y - v[2]->warp1.y;
-      GLfloat fx = v[1]->warp1.x - v[2]->warp1.x;
-      GLfloat fy = v[1]->warp1.y - v[2]->warp1.y;
-      GLfloat cc = ex*fy-ey*fx;
-	 
-#if (IND & MGA_TWOSIDE_BIT) 
+      GLfloat ex = v[0]->v.x - v[2]->v.x;
+      GLfloat ey = v[0]->v.y - v[2]->v.y;
+      GLfloat fx = v[1]->v.x - v[2]->v.x;
+      GLfloat fy = v[1]->v.y - v[2]->v.y;
+      GLfloat cc  = ex*fy - ey*fx;
+
+#if (IND & MGA_TWOSIDE_BIT)
       {
-	 GLuint facing = (cc>0.0) ^ ctx->Polygon.FrontBit;
+	 GLuint  facing        = (cc > 0.0) ^ ctx->Polygon.FrontBit;
 	 GLubyte (*vbcolor)[4] = VB->Color[facing]->data;
 	 if (IND & MGA_FLAT_BIT) {
-	    MGA_COLOR((char *)&c[0],vbcolor[pv]);
-	    c[2] = c[1] = c[0];
+	    MGA_COLOR((char *)&v[0]->ui[4], vbcolor[pv]);
+	    v[2]->ui[4] = v[1]->ui[4] = v[0]->ui[4];
 	 } else {
-	    MGA_COLOR((char *)&c[0],vbcolor[e0]);
-	    MGA_COLOR((char *)&c[1],vbcolor[e1]);
-	    MGA_COLOR((char *)&c[2],vbcolor[e2]);
+	    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]);
 	 }
       }
 #endif
-      
-#if (IND & MGA_OFFSET_BIT) 
-      if (ctx->Polygon.OffsetFill)
+
+#if (IND & MGA_OFFSET_BIT)
       {
-	 offset = ctx->Polygon.OffsetUnits * (1.0/0x10000);
+	 offset = ctx->Polygon.OffsetUnits * 1.0/0x10000;
+	 z[0] = v[0]->v.z;
+	 z[1] = v[1]->v.z;
+	 z[2] = v[2]->v.z;
 	 if (cc * cc > 1e-16) {
-	    GLfloat ez = v[0]->warp1.z - v[2]->warp1.z;
-	    GLfloat fz = v[1]->warp1.z - v[2]->warp1.z;
-	    GLfloat a = ey*fz-ez*fy;
-	    GLfloat b = ez*fx-ex*fz;
-	    GLfloat ic = 1.0 / cc;
-	    GLfloat ac = a * ic;
-	    GLfloat bc = b * ic;
-	    if (ac<0.0F)  ac = -ac;
-	    if (bc<0.0F)  bc = -bc;
-	    offset += MAX2( ac, bc ) * ctx->Polygon.OffsetFactor;
+	    GLfloat ez     = z[0] - z[2];
+	    GLfloat fz     = z[1] - z[2];
+	    GLfloat a      = ey*fz - ez*fy;
+	    GLfloat b      = ez*fx - ex*fz;
+	    GLfloat ic     = 1.0 / cc;
+	    GLfloat ac     = a * ic;
+	    GLfloat bc     = b * ic;
+	    if (ac < 0.0f) ac = -ac;
+	    if (bc < 0.0f) bc = -bc;
+	    offset += MAX2(ac, bc) * ctx->Polygon.OffsetFactor;
 	 }
+	 v[0]->v.z += offset;
+	 v[1]->v.z += offset;
+	 v[2]->v.z += offset;
       }
 #endif
    }
-#endif  
-
-
-   for (j = 0 ; j < 3 ; j++, wv += vertsize) {
+#elif (IND & MGA_FLAT_BIT)
+   {
+      GLuint color = mgaverts[pv].ui[4];
+      v[0]->ui[4] = color;
+      v[1]->ui[4] = color;
+      v[2]->ui[4] = color;
+   }
+#endif
 
-      for (i = 0 ; i < vertsize ; i++)
-	 wv[i] = v[j]->ui[i];
+   mga_draw_triangle( mgactx, v[0], v[1], v[2] );
 
-#if (IND & (MGA_FLAT_BIT|MGA_TWOSIDE_BIT))
-      wv[4] = c[j];		/* color is the fifth element... */
-#endif
 #if (IND & MGA_OFFSET_BIT)
-      *(float *)&wv[2] = v[j]->warp1.z + offset;
+   v[0]->v.z = z[0];
+   v[1]->v.z = z[1];
+   v[2]->v.z = z[2];
 #endif
-   }
+
+#if (IND & (MGA_FLAT_BIT | MGA_TWOSIDE_BIT))
+   v[0]->ui[4] = c[0];
+   v[1]->ui[4] = c[1];
+   v[2]->ui[4] = c[2];    
+#endif
+
 }
 
 
 
+
 static void TAG(quad)( GLcontext *ctx, GLuint v0,
 		       GLuint v1, GLuint v2, GLuint v3, 
 		       GLuint pv )
@@ -113,37 +127,37 @@ static void TAG(line)( GLcontext *ctx, GLuint v0, GLuint v1, GLuint pv )
       GLubyte (*vbcolor)[4] = ctx->VB->ColorPtr->data;
 
       if (IND & MGA_FLAT_BIT) {
-	 MGA_COLOR((char *)&vert0->warp1.color,vbcolor[pv]);
-	 *(int *)&vert1->warp1.color = *(int *)&vert0->warp1.color;
+	 MGA_COLOR((char *)&vert0->v.color,vbcolor[pv]);
+	 *(int *)&vert1->v.color = *(int *)&vert0->v.color;
       } else {
-	 MGA_COLOR((char *)&vert0->warp1.color,vbcolor[v0]);
-	 MGA_COLOR((char *)&vert1->warp1.color,vbcolor[v1]);
+	 MGA_COLOR((char *)&vert0->v.color,vbcolor[v0]);
+	 MGA_COLOR((char *)&vert1->v.color,vbcolor[v1]);
       }
    } else if (IND & MGA_FLAT_BIT) {
-      c0 = *(GLuint *) &(vert0->warp1.color);
-      c1 = *(GLuint *) &(vert1->warp1.color);
-      *(int *)&vert0->warp1.color = 
-	 *(int *)&vert1->warp1.color = *(int *)&mgaVB[pv].warp1.color;
+      c0 = *(GLuint *) &(vert0->v.color);
+      c1 = *(GLuint *) &(vert1->v.color);
+      *(int *)&vert0->v.color = 
+	 *(int *)&vert1->v.color = *(int *)&mgaVB[pv].v.color;
    }
 
    if (IND & MGA_OFFSET_BIT) {
       GLfloat offset = ctx->LineZoffset * (1.0 / 0x10000); 
-      z0 = vert0->warp1.z;
-      z1 = vert1->warp1.z;
-      vert0->warp1.z += offset;
-      vert1->warp1.z += offset;
+      z0 = vert0->v.z;
+      z1 = vert1->v.z;
+      vert0->v.z += offset;
+      vert1->v.z += offset;
    }
 
    mga_draw_line( mmesa, &mgaVB[v0], &mgaVB[v1], width );
 
    if (IND & MGA_OFFSET_BIT) {
-      vert0->warp1.z = z0;
-      vert1->warp1.z = z1;
+      vert0->v.z = z0;
+      vert1->v.z = z1;
    }
    
    if ((IND & MGA_FLAT_BIT) && !(IND & MGA_TWOSIDE_BIT)) {
-      *(GLuint *) &(vert0->warp1.color) = c0;
-      *(GLuint *) &(vert1->warp1.color) = c1;
+      *(GLuint *) &(vert0->v.color) = c0;
+      *(GLuint *) &(vert1->v.color) = c1;
    }
 }
 
@@ -162,11 +176,11 @@ static void TAG(points)( GLcontext *ctx, GLuint first, GLuint last )
 	    mgaVertex tmp0 = mgaVB[i];
 	    if (IND & MGA_TWOSIDE_BIT) {
 	       GLubyte (*vbcolor)[4] = VB->ColorPtr->data;
-	       MGA_COLOR((char *)&tmp0.warp1.color, vbcolor[i]);
+	       MGA_COLOR((char *)&tmp0.v.color, vbcolor[i]);
 	    }
 	    if (IND & MGA_OFFSET_BIT) {
 	       GLfloat offset = ctx->PointZoffset * (1.0 / 0x10000);
-	       tmp0.warp1.z += offset;
+	       tmp0.v.z += offset;
 	    }
 	    mga_draw_point( mmesa, &tmp0, sz );
 	 } else
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgavb.c b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
index 742b557e6..3fea4ffe0 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgavb.c
+++ b/xc/lib/GL/mesa/src/drv/mga/mgavb.c
@@ -34,36 +34,36 @@
 #include <stdlib.h>
 
 #define TEX0 {					\
-  v->warp2.tu0 = tc0[i][0];			\
-  v->warp2.tv0 = tc0[i][1];			\
+  v->v.tu0 = tc0[i][0];			\
+  v->v.tv0 = tc0[i][1];			\
 }
 
 #define TEX1 {					\
-  v->warp2.tu1 = tc1[i][0];			\
-  v->warp2.tv1 = tc1[i][1];			\
+  v->v.tu1 = tc1[i][0];			\
+  v->v.tv1 = tc1[i][1];			\
 }
 
 #define SPC {					\
   GLubyte *spec = &(VB->Spec[0][i][0]);		\
-  v->warp2.specular.red = spec[0];		\
-  v->warp2.specular.green = spec[1];		\
-  v->warp2.specular.blue = spec[2];		\
+  v->v.specular.red = spec[0];		\
+  v->v.specular.green = spec[1];		\
+  v->v.specular.blue = spec[2];		\
 }
 
 #define FOG {					\
   GLubyte *spec = &(VB->Spec[0][i][0]);		\
-  v->warp2.specular.alpha = spec[3];		\
+  v->v.specular.alpha = spec[3];		\
 }
 
 #define COL {					\
   GLubyte *col = &(VB->Color[0]->data[i][0]);	\
-  v->warp2.color.blue  = col[2];		\
-  v->warp2.color.green = col[1];		\
-  v->warp2.color.red   = col[0];		\
-  v->warp2.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 warp2 code we have doesn't seem to support projective texturing
+/* The v code we have doesn't seem to support projective texturing
  * in the multitexture case.  (Would require another 1/w value for the
  * second set of texcoords).  This may be a problem for the g400.  
  */
@@ -75,19 +75,19 @@
      mmesa->setupdone &= ~MGA_WIN_BIT;			\
      for (i=start; i < end; i++, v++)	{		\
         float oow = 1.0 / tc[i][3];			\
-	v->warp2.rhw *= tc[i][3];			\
-	v->warp2.tu0 *= oow;				\
-	v->warp2.tv0 *= oow;				\
+	v->v.rhw *= tc[i][3];			\
+	v->v.tu0 *= oow;				\
+	v->v.tv0 *= oow;				\
      }							\
   }
 
 
 #define COORD							\
       GLfloat *win = VB->Win.data[i];				\
-      v->warp2.rhw =               win[3];			\
-      v->warp2.z = (1.0/0x10000) * win[2];			\
-      v->warp2.x =                 win[0] + xoffset;		\
-      v->warp2.y =          -      win[1] + yoffset; 
+      v->v.rhw =               win[3];			\
+      v->v.z = (1.0/0x10000) * win[2];			\
+      v->v.x =                 win[0] + xoffset;		\
+      v->v.y =          -      win[1] + yoffset; 
 
 #define NOP
 
diff --git a/xc/lib/GL/mesa/src/drv/mga/mgavb.h b/xc/lib/GL/mesa/src/drv/mga/mgavb.h
index 675c1779c..e8926ba1f 100644
--- a/xc/lib/GL/mesa/src/drv/mga/mgavb.h
+++ b/xc/lib/GL/mesa/src/drv/mga/mgavb.h
@@ -46,35 +46,25 @@ typedef struct mga_warp_color_t {
 } mga_warp_color;
 
 
-/* 
- *  vertex type used for the single-warp g200
- */
-typedef struct mga_warp_vertex_t {
-  GLfloat		x,y,z;	        /* coordinates in screen space*/
-  GLfloat 		rhw;	        /* reciprocal homogeneous w */
-  mga_warp_color	color;		/* vertex color */
-  mga_warp_color	specular;	/* specular color, alpha is fog */
-  GLfloat		tu0,tv0;	/* texture coordinates */
-} mga_warp_vertex1;
 
 /*
- *  vertex type used for the dual-warp g400
+ * The vertex structure.  The final tu1/tv1 values only used in multitexture
+ * modes.
  */
-typedef struct mga_warp_vertex2_t {
+typedef struct mga_warp_vertex_t {
   GLfloat		x,y,z;	        /* coordinates in screen space*/
   GLfloat 		rhw;	        /* reciprocal homogeneous w */
   mga_warp_color	color;		/* vertex color */
   mga_warp_color	specular;	/* specular color, alpha is fog */
   GLfloat		tu0,tv0;	/* texture coordinates */
   GLfloat		tu1,tv1;	/* same for second stage */
-} mga_warp_vertex2;
+} mga_warp_vertex;
 
 
 /* The fastpath code still expects a 16-float stride vertex.
  */
 union mga_vertex_t {
-   mga_warp_vertex1 warp1;	
-   mga_warp_vertex2 warp2;
+   mga_warp_vertex v;	
    float f[16];
    mgaUI32 ui[16];
 };