Performance improvements, support userspace simulatorsHEADmaster

17 files changed, 313 insertions, 90 deletions

diff --git a/src/mesa/Makefile b/src/mesa/Makefile
index 1315e35..d612675 100644
--- a/src/mesa/Makefile
+++ b/src/mesa/Makefile
@@ -28,6 +28,10 @@ default:
 		$(MAKE) beos ; \
 	elif [ "${DRIVER_DIRS}" = "verite" ] ; then \
 		$(MAKE) verite-alone ; \
+	elif [ "${DRIVER_DIRS}" = "verite-sim" ] ; then \
+		$(MAKE) verite-sim ; \
+	elif [ "${DRIVER_DIRS}" = "glide-sim" ] ; then \
+		$(MAKE) glide-sim ; \
 	else \
 		$(MAKE) stand-alone ; \
 	fi
@@ -171,6 +175,50 @@ verite-alone: depend $(VERITE_ALONE_OBJECTS) $(OSMESA_DRIVER_OBJECTS)
 
 
 ######################################################################
+# Verite driver over the userspace simulator
+
+VERITE_SIM_SOURCES = \
+	$(CORE_SOURCES) \
+	$(COMMON_DRIVER_SOURCES) \
+	$(VERITE_COMMON_DRIVER_SOURCES) \
+	$(PPC_SOURCES)
+
+VERITE_SIM_OBJECTS = $(VERITE_SIM_SOURCES:.c=.o)
+
+-include $(VERITE_SIM_OBJECTS:.o=.d)
+
+# rm -f first: mklib's Darwin static case updates with "ar -ruv", which
+# would keep stale members (e.g. libv3d.o) from a previous config.
+verite-sim: depend $(VERITE_SIM_OBJECTS)
+	rm -f $(LIB_DIR)/$(GL_LIB_NAME)
+	CC="$(CC)" CXX="$(CXX)" $(TOP)/bin/mklib -o $(GL_LIB) \
+		-major $(GL_MAJOR) -minor $(GL_MINOR) -patch $(GL_TINY) \
+		-install $(LIB_DIR) $(MKLIB_OPTIONS) $(GL_LIB_DEPS) \
+		$(VERITE_SIM_OBJECTS)
+
+
+######################################################################
+# fxMesa (3dfx Glide driver) over the userspace Voodoo simulator
+
+GLIDE_SIM_SOURCES = \
+	$(CORE_SOURCES) \
+	$(COMMON_DRIVER_SOURCES) \
+	$(GLIDE_DRIVER_SOURCES) \
+	$(PPC_SOURCES)
+
+GLIDE_SIM_OBJECTS = $(GLIDE_SIM_SOURCES:.c=.o)
+
+-include $(GLIDE_SIM_OBJECTS:.o=.d)
+
+glide-sim: depend $(GLIDE_SIM_OBJECTS)
+	rm -f $(LIB_DIR)/$(GL_LIB_NAME)
+	CC="$(CC)" CXX="$(CXX)" $(TOP)/bin/mklib -o $(GL_LIB) \
+		-major $(GL_MAJOR) -minor $(GL_MINOR) -patch $(GL_TINY) \
+		-install $(LIB_DIR) $(MKLIB_OPTIONS) $(GL_LIB_DEPS) \
+		$(GLIDE_SIM_OBJECTS)
+
+
+######################################################################
 # Generic stuff
 
 ALL_SOURCES = \
@@ -205,8 +253,10 @@ tags:
 clean:
 	-rm -f */*.o
 	-rm -f depend.bak mesa.a
-	-rm -f */*.d drivers/*/*.d
-	-rm -f drivers/*/*.o
-	(cd drivers/dri ; $(MAKE) clean)
+	-rm -f */*.d drivers/*/*.d drivers/*/*/*.d
+	-rm -f drivers/*/*.o drivers/*/*/*.o
+	@if [ -f drivers/dri/Makefile ] ; then \
+		(cd drivers/dri ; $(MAKE) clean) ; \
+	fi
 
 include depend
diff --git a/src/mesa/drivers/verite/libv3d.c b/src/mesa/drivers/verite/libv3d.c
index b09f960..8d8050b 100644
--- a/src/mesa/drivers/verite/libv3d.c
+++ b/src/mesa/drivers/verite/libv3d.c
@@ -284,6 +284,13 @@ v3d_alloc(struct v3d *v, uint32_t size, uint32_t align, uint32_t *addr)
 }
 
 int
+v3d_free(struct v3d *v, uint32_t addr)
+{
+
+	return ioctl(v->fd, V3D_FREE, &addr);
+}
+
+int
 v3d_mode(struct v3d *v, uint32_t depth, uint32_t frame_base)
 {
 	struct v3d_mode vm;
diff --git a/src/mesa/drivers/verite/libv3d.h b/src/mesa/drivers/verite/libv3d.h
index 4ee44b1..e27175a 100644
--- a/src/mesa/drivers/verite/libv3d.h
+++ b/src/mesa/drivers/verite/libv3d.h
@@ -122,6 +122,7 @@ void	v3d_commit(struct v3d *, uint32_t nwords);
 int	v3d_flush(struct v3d *);
 int	v3d_sync(struct v3d *);
 int	v3d_alloc(struct v3d *, uint32_t, uint32_t, uint32_t *);
+int	v3d_free(struct v3d *, uint32_t addr);
 int	v3d_mode(struct v3d *, uint32_t depth, uint32_t frame_base);
 int	v3d_flip(struct v3d *, uint32_t frame_base);
 
diff --git a/src/mesa/drivers/verite/rlgl.c b/src/mesa/drivers/verite/rlgl.c
index 1da14f2..4e25d93 100644
--- a/src/mesa/drivers/verite/rlgl.c
+++ b/src/mesa/drivers/verite/rlgl.c
@@ -27,7 +27,6 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <sys/ioctl.h>
 #include <stdio.h>
 #include <string.h>
 
@@ -289,7 +288,7 @@ rlgl_tex_free(struct rlgl *rl, struct rlgl_tex *t)
 		return;
 	if (rl->cur_tex == addr)
 		rl->cur_tex = (uint32_t)-1;
-	(void)ioctl(rl->v.fd, V3D_FREE, &addr);
+	(void)v3d_free(&rl->v, addr);
 	t->addr = 0;
 }
 
diff --git a/src/mesa/drivers/verite/vrspan.c b/src/mesa/drivers/verite/vrspan.c
index 82a9f73..b666f6b 100644
--- a/src/mesa/drivers/verite/vrspan.c
+++ b/src/mesa/drivers/verite/vrspan.c
@@ -89,7 +89,11 @@ vr_set_buffer(GLcontext *ctx, GLframebuffer *buffer, GLuint bufferBit)
 	}
 }
 
+#if defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
 #define VR_SWAP16(x)	((GLushort)((((x) & 0xff) << 8) | ((x) >> 8)))
+#else
+#define VR_SWAP16(x)	((GLushort)(x))
+#endif
 
 #define NAME(PREFIX) PREFIX##_565
 #define SPAN_VARS \
diff --git a/src/mesa/main/imports.h b/src/mesa/main/imports.h
index 86480c8..98e5960 100644
--- a/src/mesa/main/imports.h
+++ b/src/mesa/main/imports.h
@@ -228,7 +228,7 @@ extern void _ext_mesa_free_pixelbuffer( void *pb );
     defined(ia64) || defined(__ia64__) || \
     defined(__hppa__) || defined(hpux) || \
     defined(__mips) || defined(_MIPS_ARCH) || \
-    defined(__arm__) || \
+    defined(__arm__) || defined(__aarch64__) || \
     (defined(__alpha__) && (defined(__IEEE_FLOAT) || !defined(VMS)))
 #define USE_IEEE
 #define IEEE_ONE 0x3f800000
diff --git a/src/mesa/ppc/ppc_xform.c b/src/mesa/ppc/ppc_xform.c
index 966798c..cbc122a 100644
--- a/src/mesa/ppc/ppc_xform.c
+++ b/src/mesa/ppc/ppc_xform.c
@@ -26,21 +26,10 @@ _mesa_init_all_ppc_transform_asm(void)
 
 	rlvec_init();
 
-	/*
-	 * points3 = the glVertex3f-class path.
-	 */
 	_mesa_transform_tab[3][MATRIX_GENERAL] =
 	    ppc_transform_points3_general;
 	_mesa_transform_tab[3][MATRIX_PERSPECTIVE] =
 	    ppc_transform_points3_general;
-	_mesa_transform_tab[3][MATRIX_3D] =
-	    ppc_transform_points3_general;
-	_mesa_transform_tab[3][MATRIX_3D_NO_ROT] =
-	    ppc_transform_points3_general;
-	_mesa_transform_tab[3][MATRIX_2D] =
-	    ppc_transform_points3_general;
-	_mesa_transform_tab[3][MATRIX_2D_NO_ROT] =
-	    ppc_transform_points3_general;
 }
 
 #endif /* USE_PPC_ASM */
diff --git a/src/mesa/sources b/src/mesa/sources
index 8490906..4b88f6b 100644
--- a/src/mesa/sources
+++ b/src/mesa/sources
@@ -219,13 +219,16 @@ GLIDE_DRIVER_SOURCES =			\
 	drivers/glide/fxglidew.c	\
 	drivers/glide/fxg.c
 
-VERITE_DRIVER_SOURCES =			\
+VERITE_COMMON_DRIVER_SOURCES =		\
 	drivers/verite/vrapi.c		\
 	drivers/verite/vrdd.c		\
 	drivers/verite/vrspan.c		\
 	drivers/verite/vrtex.c		\
 	drivers/verite/vrtris.c		\
-	drivers/verite/rlgl.c		\
+	drivers/verite/rlgl.c
+
+VERITE_DRIVER_SOURCES =			\
+	$(VERITE_COMMON_DRIVER_SOURCES)	\
 	drivers/verite/libv3d.c
 
 SVGA_DRIVER_SOURCES =			\
diff --git a/src/mesa/swrast/swrast.h b/src/mesa/swrast/swrast.h
index 7fb0d2a..a3cd9e0 100644
--- a/src/mesa/swrast/swrast.h
+++ b/src/mesa/swrast/swrast.h
@@ -62,14 +62,19 @@
  */
 typedef struct {
    /** win[0], win[1] are the screen-coords of SWvertex. win[2] is the
-    * z-coord. what is win[3]? */
+    * z-coord. what is win[3]?
+    *
+    * The texcoord array (most of the struct) trails so a
+    * fixed-function vertex's hot fields share fewer cache lines.  All
+    * access is by field name, no code depends on member order.
+    */
    GLfloat win[4];
-   GLfloat texcoord[MAX_TEXTURE_COORD_UNITS][4];
    GLchan color[4];
    GLchan specular[4];
    GLfloat fog;
    GLfloat index;
    GLfloat pointSize;
+   GLfloat texcoord[MAX_TEXTURE_COORD_UNITS][4];
 } SWvertex;
 
 
diff --git a/src/mesa/tnl/t_context.c b/src/mesa/tnl/t_context.c
index e4d0d04..8db1c78 100644
--- a/src/mesa/tnl/t_context.c
+++ b/src/mesa/tnl/t_context.c
@@ -54,6 +54,13 @@ _tnl_MakeCurrent( GLcontext *ctx,
    (void) ctx; (void) drawBuffer; (void) readBuffer;
 }
 
+static void
+_tnl_validate_material( GLcontext *ctx )
+{
+   _mesa_update_material( ctx, ~0 );
+   _mesa_validate_all_lighting_tables( ctx );
+}
+
 
 static void
 install_driver_callbacks( GLcontext *ctx )
@@ -123,7 +130,7 @@ _tnl_CreateContext( GLcontext *ctx )
 
    tnl->Driver.Render.PrimTabElts = _tnl_render_tab_elts;
    tnl->Driver.Render.PrimTabVerts = _tnl_render_tab_verts;
-   tnl->Driver.NotifyMaterialChange = _mesa_validate_all_lighting_tables;
+   tnl->Driver.NotifyMaterialChange = _tnl_validate_material;
    
    return GL_TRUE;
 }
diff --git a/src/mesa/tnl/t_context.h b/src/mesa/tnl/t_context.h
index 55b405e..ae565fb 100644
--- a/src/mesa/tnl/t_context.h
+++ b/src/mesa/tnl/t_context.h
@@ -278,6 +278,9 @@ struct _tnl_dynfn_generators {
 struct tnl_vtx {
    GLfloat buffer[VERT_BUFFER_SIZE];
    GLubyte attrsz[_TNL_ATTRIB_MAX];
+   GLubyte active_sz[_TNL_ATTRIB_MAX];	/* size the generic entrypoints
+					 * currently store per attribute;
+					 * 0 forces _tnl_choose_attr() */
    GLuint vertex_size;
    struct tnl_prim prim[TNL_MAX_PRIM];
    GLuint prim_count;
diff --git a/src/mesa/tnl/t_vb_light.c b/src/mesa/tnl/t_vb_light.c
index 1a89bf3..7d19c15 100644
--- a/src/mesa/tnl/t_vb_light.c
+++ b/src/mesa/tnl/t_vb_light.c
@@ -118,12 +118,11 @@ static GLuint prepare_materials( GLcontext *ctx,
 	 store->mat_bitmask |= (1<<attr);
       }
    }
-   
 
-   /* FIXME: Is this already done?
-    */
-   _mesa_update_material( ctx, ~0 );
-   _mesa_validate_all_lighting_tables( ctx );
+   if (store->mat_count) {
+      _mesa_update_material( ctx, ~0 );
+      _mesa_validate_all_lighting_tables( ctx );
+   }
 
    return store->mat_count;
 }
diff --git a/src/mesa/tnl/t_vb_program.c b/src/mesa/tnl/t_vb_program.c
index 38b04b4..aedfb4e 100644
--- a/src/mesa/tnl/t_vb_program.c
+++ b/src/mesa/tnl/t_vb_program.c
@@ -355,7 +355,10 @@ static void dtr( struct tnl_pipeline_stage *stage )
 const struct tnl_pipeline_stage _tnl_vertex_program_stage =
 {
    "vertex-program",
-   _NEW_ALL,	/*XXX FIX */	/* recheck */
+   _NEW_PROGRAM,		/* recheck: check_vp() only reads
+				 * VertexProgram._Enabled and
+				 * Current->InputsRead, both always
+				 * flushed with _NEW_PROGRAM */
    _NEW_ALL,	/*XXX FIX */    /* recalc */
    GL_FALSE,			/* active */
    0,				/* inputs - calculated on the fly */
diff --git a/src/mesa/tnl/t_vertex.c b/src/mesa/tnl/t_vertex.c
index 900648a..6245f70 100644
--- a/src/mesa/tnl/t_vertex.c
+++ b/src/mesa/tnl/t_vertex.c
@@ -871,6 +871,107 @@ static void generic_emit( GLcontext *ctx,
 }
 
 
+#define FUSED_EMIT( NAME, COLSZ, HAVE_TEX )				\
+static void NAME( GLcontext *ctx, GLuint start, GLuint end,		\
+		  void *dest )						\
+{									\
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);			\
+   const struct tnl_clipspace_attr *a = vtx->attr;			\
+   const GLfloat * const vp = a[0].vp;					\
+   const GLubyte *pos_ptr = a[0].inputptr;				\
+   const GLubyte *col_ptr = a[1].inputptr;				\
+   const GLubyte *tex_ptr = HAVE_TEX ? a[2].inputptr : NULL;		\
+   const GLuint pos_stride = a[0].inputstride;				\
+   const GLuint col_stride = a[1].inputstride;				\
+   const GLuint tex_stride = HAVE_TEX ? a[2].inputstride : 0;		\
+   const GLuint pos_off = a[0].vertoffset;				\
+   const GLuint col_off = a[1].vertoffset;				\
+   const GLuint tex_off = HAVE_TEX ? a[2].vertoffset : 0;		\
+   const GLuint stride = vtx->vertex_size;				\
+   GLubyte *v = (GLubyte *)dest;					\
+   GLuint i;								\
+									\
+   end -= start;							\
+									\
+   for (i = 0; i < end; i++, v += stride) {				\
+      {									\
+	 const GLfloat *in = (const GLfloat *)pos_ptr;			\
+	 GLfloat *out = (GLfloat *)(v + pos_off);			\
+									\
+	 out[0] = vp[0] * in[0] + vp[12];				\
+	 out[1] = vp[5] * in[1] + vp[13];				\
+	 out[2] = vp[10] * in[2] + vp[14];				\
+	 out[3] = in[3];						\
+	 pos_ptr += pos_stride;						\
+      }									\
+      {									\
+	 const GLfloat *in = (const GLfloat *)col_ptr;			\
+	 GLchan *c = (GLchan *)(v + col_off);				\
+									\
+	 UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]);				\
+	 UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]);				\
+	 UNCLAMPED_FLOAT_TO_CHAN(c[2], in[2]);				\
+	 if (COLSZ == 4) {						\
+	    UNCLAMPED_FLOAT_TO_CHAN(c[3], in[3]);			\
+	 }								\
+	 else								\
+	    c[3] = CHAN_MAX;						\
+	 col_ptr += col_stride;						\
+      }									\
+      if (HAVE_TEX) {							\
+	 const GLfloat *in = (const GLfloat *)tex_ptr;			\
+	 GLfloat *out = (GLfloat *)(v + tex_off);			\
+									\
+	 out[0] = in[0];						\
+	 out[1] = in[1];						\
+	 out[2] = 0;							\
+	 out[3] = 1;							\
+	 tex_ptr += tex_stride;						\
+      }									\
+   }									\
+}
+
+FUSED_EMIT( emit_viewport4_rgba3,      3, 0 )
+FUSED_EMIT( emit_viewport4_rgba4,      4, 0 )
+FUSED_EMIT( emit_viewport4_rgba3_st,   3, 1 )
+FUSED_EMIT( emit_viewport4_rgba4_st,   4, 1 )
+
+#undef FUSED_EMIT
+
+/* Match the installed attribute map plus this flush's input vector
+ * sizes against the fused layouts above; anything else takes
+ * generic_emit.
+ */
+static tnl_emit_func choose_emit_func( struct tnl_clipspace *vtx,
+				       struct vertex_buffer *VB )
+{
+   const struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint count = vtx->attr_count;
+   GLuint colsz;
+
+   if (count < 2 || count > 3 ||
+       a[0].format != EMIT_4F_VIEWPORT ||
+       VB->AttribPtr[a[0].attrib]->size != 4 ||
+       a[1].attrib != VERT_ATTRIB_COLOR0 ||
+       a[1].format != EMIT_4CHAN_4F_RGBA)
+      return generic_emit;
+
+   colsz = VB->AttribPtr[a[1].attrib]->size;
+   if (colsz != 3 && colsz != 4)
+      return generic_emit;
+
+   if (count == 3) {
+      if (a[2].format != EMIT_4F ||
+	  VB->AttribPtr[a[2].attrib]->size != 2)
+	 return generic_emit;
+      return (colsz == 4) ? emit_viewport4_rgba4_st
+			  : emit_viewport4_rgba3_st;
+   }
+
+   return (colsz == 4) ? emit_viewport4_rgba4 : emit_viewport4_rgba3;
+}
+
+
 static void generic_interp( GLcontext *ctx,
 			    GLfloat t,
 			    GLuint edst, GLuint eout, GLuint ein,
@@ -1030,13 +1131,7 @@ static void do_emit( GLcontext *ctx, GLuint start, GLuint end,
       a[j].emit = a[j].insert[vptr->size - 1];
    }
 
-   vtx->emit = 0;
-   
-   if (0) 
-      vtx->emit = _tnl_codegen_emit(ctx);
-   
-   if (!vtx->emit)
-      vtx->emit = generic_emit;
+   vtx->emit = choose_emit_func( vtx, VB );
 
    vtx->emit( ctx, start, end, dest );
 }
diff --git a/src/mesa/tnl/t_vtx_api.c b/src/mesa/tnl/t_vtx_api.c
index 56b1e9f..f2e7857 100644
--- a/src/mesa/tnl/t_vtx_api.c
+++ b/src/mesa/tnl/t_vtx_api.c
@@ -312,14 +312,20 @@ static void _tnl_wrap_upgrade_vertex( GLcontext *ctx,
     * codegen.  Might be a reasonable place to try & detect attributes
     * in the vertex which aren't being submitted any more.
     */
-   for (i = 0 ; i < _TNL_ATTRIB_MAX ; i++) 
+   for (i = 0 ; i < _TNL_ATTRIB_MAX ; i++) {
+      /* Force the generic entrypoints through _tnl_choose_attr() again
+       * so shrunken attributes get their identity components refilled
+       * (mirrors the tabfv reset below).
+       */
+      tnl->vtx.active_sz[i] = 0;
+
       if (tnl->vtx.attrsz[i]) {
 	 GLuint j = tnl->vtx.attrsz[i] - 1;
 
 	 if (i < _TNL_MAX_ATTR_CODEGEN)
 	    tnl->vtx.tabfv[i][j] = choose[i][j];
       }
-
+   }
 }
 
 
@@ -440,6 +446,22 @@ static tnl_attrfv_func do_choose( GLuint attr, GLuint sz )
 
 
 
+/* Helper for the single-dispatch generic entrypoints
+ */
+void _tnl_choose_attr( GLcontext *ctx, GLuint attr, GLuint sz )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+
+   assert(attr < _TNL_ATTRIB_MAX);
+   assert(sz >= 1 && sz <= 4);
+
+   if (tnl->vtx.attrsz[attr] != sz)
+      _tnl_fixup_vertex( ctx, attr, sz );
+
+   tnl->vtx.active_sz[attr] = (GLubyte) sz;
+}
+
+
 #define CHOOSE( ATTR, N )				\
 static void choose_##ATTR##_##N( const GLfloat *v )	\
 {							\
@@ -487,10 +509,12 @@ static void error_attrib( const GLfloat *unused )
 
 
 static void reset_attrfv( TNLcontext *tnl )
-{   
+{
    GLuint i;
 
-   for (i = 0 ; i < _TNL_ATTRIB_MAX ; i++) 
+   for (i = 0 ; i < _TNL_ATTRIB_MAX ; i++) {
+      tnl->vtx.active_sz[i] = 0;
+
       if (tnl->vtx.attrsz[i]) {
 	 GLint j = tnl->vtx.attrsz[i] - 1;
 	 tnl->vtx.attrsz[i] = 0;
@@ -502,6 +526,7 @@ static void reset_attrfv( TNLcontext *tnl )
             }
          }
       }
+   }
 
    tnl->vtx.vertex_size = 0;
    tnl->vtx.have_materials = 0;
diff --git a/src/mesa/tnl/t_vtx_api.h b/src/mesa/tnl/t_vtx_api.h
index 9818c08..0480bad 100644
--- a/src/mesa/tnl/t_vtx_api.h
+++ b/src/mesa/tnl/t_vtx_api.h
@@ -51,6 +51,8 @@ extern void _tnl_flush_vtx( GLcontext *ctx );
 
 extern void GLAPIENTRY _tnl_wrap_filled_vertex( GLcontext *ctx );
 
+extern void _tnl_choose_attr( GLcontext *ctx, GLuint attr, GLuint sz );
+
 /* t_vtx_exec.c:
  */
 
diff --git a/src/mesa/tnl/t_vtx_generic.c b/src/mesa/tnl/t_vtx_generic.c
index daa7dea..faf7e19 100644
--- a/src/mesa/tnl/t_vtx_generic.c
+++ b/src/mesa/tnl/t_vtx_generic.c
@@ -133,52 +133,51 @@ void _tnl_generic_attr_table_init( tnl_attrfv_func (*tab)[4] )
    INIT( tab, 15 );
 }
 
-/* These can be made efficient with codegen.  Further, by adding more
- * logic to do_choose(), the double-dispatch for legacy entrypoints
- * like glVertex3f() can be removed.
- */
-#define DISPATCH_ATTRFV( ATTR, COUNT, P )	\
-do {						\
-   GET_CURRENT_CONTEXT( ctx ); 			\
-   TNLcontext *tnl = TNL_CONTEXT(ctx); 		\
-   tnl->vtx.tabfv[ATTR][COUNT-1]( P );		\
+/* Single-dispatch entrypoints */
+#define ATTRF( A, N, X, Y, Z, W )			\
+do {							\
+   GET_CURRENT_CONTEXT( ctx );				\
+   TNLcontext *tnl = TNL_CONTEXT(ctx);			\
+							\
+   if (tnl->vtx.active_sz[A] != (N))			\
+      _tnl_choose_attr( ctx, (A), (N) );		\
+							\
+   if ((A) == 0) {					\
+      GLfloat *dest = tnl->vtx.vbptr;			\
+      GLuint i;						\
+							\
+      if ((N)>0) dest[0] = (X);				\
+      if ((N)>1) dest[1] = (Y);				\
+      if ((N)>2) dest[2] = (Z);				\
+      if ((N)>3) dest[3] = (W);				\
+							\
+      for (i = (N); i < tnl->vtx.vertex_size; i++)	\
+	 dest[i] = tnl->vtx.vertex[i];			\
+							\
+      tnl->vtx.vbptr = dest + tnl->vtx.vertex_size;	\
+							\
+      if (--tnl->vtx.counter == 0)			\
+	 _tnl_wrap_filled_vertex( ctx );		\
+   }							\
+   else {						\
+      GLfloat *dest = tnl->vtx.attrptr[A];		\
+							\
+      if ((N)>0) dest[0] = (X);				\
+      if ((N)>1) dest[1] = (Y);				\
+      if ((N)>2) dest[2] = (Z);				\
+      if ((N)>3) dest[3] = (W);				\
+   }							\
 } while (0)
 
-#define DISPATCH_ATTR1FV( ATTR, V ) DISPATCH_ATTRFV( ATTR, 1, V )
-#define DISPATCH_ATTR2FV( ATTR, V ) DISPATCH_ATTRFV( ATTR, 2, V )
-#define DISPATCH_ATTR3FV( ATTR, V ) DISPATCH_ATTRFV( ATTR, 3, V )
-#define DISPATCH_ATTR4FV( ATTR, V ) DISPATCH_ATTRFV( ATTR, 4, V )
-
-#define DISPATCH_ATTR1F( ATTR, S ) DISPATCH_ATTRFV( ATTR, 1, &(S) )
+#define DISPATCH_ATTR1FV( ATTR, V ) ATTRF( ATTR, 1, (V)[0], 0, 0, 1 )
+#define DISPATCH_ATTR2FV( ATTR, V ) ATTRF( ATTR, 2, (V)[0], (V)[1], 0, 1 )
+#define DISPATCH_ATTR3FV( ATTR, V ) ATTRF( ATTR, 3, (V)[0], (V)[1], (V)[2], 1 )
+#define DISPATCH_ATTR4FV( ATTR, V ) ATTRF( ATTR, 4, (V)[0], (V)[1], (V)[2], (V)[3] )
 
-#if defined(USE_X86_ASM) && 0 /* will break register calling convention */
-/* Naughty cheat:
- */
-#define DISPATCH_ATTR2F( ATTR, S,T ) DISPATCH_ATTRFV( ATTR, 2, &(S) )
-#define DISPATCH_ATTR3F( ATTR, S,T,R ) DISPATCH_ATTRFV( ATTR, 3, &(S) )
-#define DISPATCH_ATTR4F( ATTR, S,T,R,Q ) DISPATCH_ATTRFV( ATTR, 4, &(S) )
-#else
-/* Safe:
- */
-#define DISPATCH_ATTR2F( ATTR, S,T ) 		\
-do { 						\
-   GLfloat v[2]; 				\
-   v[0] = S; v[1] = T;				\
-   DISPATCH_ATTR2FV( ATTR, v );			\
-} while (0)
-#define DISPATCH_ATTR3F( ATTR, S,T,R ) 		\
-do { 						\
-   GLfloat v[3]; 				\
-   v[0] = S; v[1] = T; v[2] = R;		\
-   DISPATCH_ATTR3FV( ATTR, v );			\
-} while (0)
-#define DISPATCH_ATTR4F( ATTR, S,T,R,Q )	\
-do { 						\
-   GLfloat v[4]; 				\
-   v[0] = S; v[1] = T; v[2] = R; v[3] = Q;	\
-   DISPATCH_ATTR4FV( ATTR, v );			\
-} while (0)
-#endif
+#define DISPATCH_ATTR1F( ATTR, S )       ATTRF( ATTR, 1, S, 0, 0, 1 )
+#define DISPATCH_ATTR2F( ATTR, S,T )     ATTRF( ATTR, 2, S, T, 0, 1 )
+#define DISPATCH_ATTR3F( ATTR, S,T,R )   ATTRF( ATTR, 3, S, T, R, 1 )
+#define DISPATCH_ATTR4F( ATTR, S,T,R,Q ) ATTRF( ATTR, 4, S, T, R, Q )
 
 
 static void GLAPIENTRY _tnl_Vertex2f( GLfloat x, GLfloat y )
@@ -363,42 +362,66 @@ static void GLAPIENTRY _tnl_MultiTexCoord4fv( GLenum target,
 
 static void GLAPIENTRY _tnl_VertexAttrib1fNV( GLuint index, GLfloat x )
 {
-   if (index >= VERT_ATTRIB_MAX) index = ERROR_ATTRIB;
+   if (index >= VERT_ATTRIB_MAX) {
+      GET_CURRENT_CONTEXT( ctx );
+      _mesa_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+      return;
+   }
    DISPATCH_ATTR1F( index, x );
 }
 
 static void GLAPIENTRY _tnl_VertexAttrib1fvNV( GLuint index, 
 					       const GLfloat *v )
 {
-   if (index >= VERT_ATTRIB_MAX) index = ERROR_ATTRIB;
+   if (index >= VERT_ATTRIB_MAX) {
+      GET_CURRENT_CONTEXT( ctx );
+      _mesa_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+      return;
+   }
    DISPATCH_ATTR1FV( index, v );
 }
 
 static void GLAPIENTRY _tnl_VertexAttrib2fNV( GLuint index, GLfloat x, 
 					      GLfloat y )
 {
-   if (index >= VERT_ATTRIB_MAX) index = ERROR_ATTRIB;
+   if (index >= VERT_ATTRIB_MAX) {
+      GET_CURRENT_CONTEXT( ctx );
+      _mesa_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+      return;
+   }
    DISPATCH_ATTR2F( index, x, y );
 }
 
 static void GLAPIENTRY _tnl_VertexAttrib2fvNV( GLuint index,
 					       const GLfloat *v )
 {
-   if (index >= VERT_ATTRIB_MAX) index = ERROR_ATTRIB;
+   if (index >= VERT_ATTRIB_MAX) {
+      GET_CURRENT_CONTEXT( ctx );
+      _mesa_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+      return;
+   }
    DISPATCH_ATTR2FV( index, v );
 }
 
 static void GLAPIENTRY _tnl_VertexAttrib3fNV( GLuint index, GLfloat x,
 					      GLfloat y, GLfloat z )
 {
-   if (index >= VERT_ATTRIB_MAX) index = ERROR_ATTRIB;
+   if (index >= VERT_ATTRIB_MAX) {
+      GET_CURRENT_CONTEXT( ctx );
+      _mesa_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+      return;
+   }
    DISPATCH_ATTR3F( index, x, y, z );
 }
 
 static void GLAPIENTRY _tnl_VertexAttrib3fvNV( GLuint index,
 					       const GLfloat *v )
 {
-   if (index >= VERT_ATTRIB_MAX) index = ERROR_ATTRIB;
+   if (index >= VERT_ATTRIB_MAX) {
+      GET_CURRENT_CONTEXT( ctx );
+      _mesa_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+      return;
+   }
    DISPATCH_ATTR3FV( index, v );
 }
 
@@ -406,14 +429,22 @@ static void GLAPIENTRY _tnl_VertexAttrib4fNV( GLuint index, GLfloat x,
 					      GLfloat y, GLfloat z,
 					      GLfloat w )
 {
-   if (index >= VERT_ATTRIB_MAX) index = ERROR_ATTRIB;
+   if (index >= VERT_ATTRIB_MAX) {
+      GET_CURRENT_CONTEXT( ctx );
+      _mesa_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+      return;
+   }
    DISPATCH_ATTR4F( index, x, y, z, w );
 }
 
 static void GLAPIENTRY _tnl_VertexAttrib4fvNV( GLuint index, 
 					       const GLfloat *v )
 {
-   if (index >= VERT_ATTRIB_MAX) index = ERROR_ATTRIB;
+   if (index >= VERT_ATTRIB_MAX) {
+      GET_CURRENT_CONTEXT( ctx );
+      _mesa_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+      return;
+   }
    DISPATCH_ATTR4FV( index, v );
 }