From 4dc3249f0d800f9e36ee11ec5c00351e67dbeee3 Mon Sep 17 00:00:00 2001
From: Aapo Tahkola <aet@rasterburn.org>
Date: Wed, 26 Oct 2005 16:42:06 +0000
Subject: [PATCH] Sync with my local tree. Changes to current operation: -Elts
 are no longer converted to 16-bit format -Cube maps

---
 src/mesa/drivers/dri/r300/Makefile          |   2 +
 src/mesa/drivers/dri/r300/r300_cmdbuf.c     |  46 +-
 src/mesa/drivers/dri/r300/r300_context.c    |  96 +--
 src/mesa/drivers/dri/r300/r300_context.h    |  57 +-
 src/mesa/drivers/dri/r300/r300_ioctl.c      | 124 +++-
 src/mesa/drivers/dri/r300/r300_maos.c       | 396 ++++++++++++-
 src/mesa/drivers/dri/r300/r300_maos.h       |  11 +-
 src/mesa/drivers/dri/r300/r300_reg.h        |   1 +
 src/mesa/drivers/dri/r300/r300_render.c     | 280 ++++++++-
 src/mesa/drivers/dri/r300/r300_state.c      |  30 +
 src/mesa/drivers/dri/r300/r300_state.h      |   4 +
 src/mesa/drivers/dri/r300/r300_texstate.c   |   5 +-
 src/mesa/drivers/dri/r300/r300_vertexprog.c |  18 +
 src/mesa/drivers/dri/r300/radeon_mm.c       | 339 +++++++++++
 src/mesa/drivers/dri/r300/radeon_mm.h       |  35 ++
 src/mesa/drivers/dri/r300/radeon_vtxfmt_a.c | 610 ++++++++++++++++++++
 16 files changed, 1898 insertions(+), 156 deletions(-)
 create mode 100644 src/mesa/drivers/dri/r300/radeon_mm.c
 create mode 100644 src/mesa/drivers/dri/r300/radeon_mm.h
 create mode 100644 src/mesa/drivers/dri/r300/radeon_vtxfmt_a.c

diff --git a/src/mesa/drivers/dri/r300/Makefile b/src/mesa/drivers/dri/r300/Makefile
index 02ef3214776..29dd20860c8 100644
--- a/src/mesa/drivers/dri/r300/Makefile
+++ b/src/mesa/drivers/dri/r300/Makefile
@@ -25,6 +25,8 @@ DRIVER_SOURCES = \
 		 radeon_lock.c \
 		 radeon_span.c \
 		 radeon_state.c \
+		 radeon_mm.c \
+		 radeon_vtxfmt_a.c \
 		 \
 		 r300_context.c \
 		 r300_ioctl.c \
diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
index f24ebe1a91c..94c7031baae 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
@@ -128,9 +128,10 @@ int r300FlushCmdBuf(r300ContextPtr r300, const char* caller)
 }
 
 
-static void print_state_atom(struct r300_state_atom *state, int dwords)
+void r300_print_state_atom(r300ContextPtr r300, struct r300_state_atom *state)
 {
 	int i;
+	int dwords = (*state->check)(r300, state);
 
 	fprintf(stderr, "  emit %s/%d/%d\n", state->name, dwords, state->cmd_size);
 
@@ -159,7 +160,7 @@ static __inline__ void r300DoEmitState(r300ContextPtr r300, GLboolean dirty)
 				int dwords = (*atom->check)(r300, atom);
 
 				if (dwords)
-					print_state_atom(atom, dwords);
+					r300_print_state_atom(r300, atom);
 				else
 					fprintf(stderr, "  skip state %s\n",
 						atom->name);
@@ -567,7 +568,7 @@ void r300EmitBlit(r300ContextPtr rmesa,
 		  GLint srcx, GLint srcy,
 		  GLint dstx, GLint dsty, GLuint w, GLuint h)
 {
-	drm_radeon_cmd_header_t *cmd;
+	drm_r300_cmd_header_t *cmd;
 
 	if (RADEON_DEBUG & DEBUG_IOCTL)
 		fprintf(stderr,
@@ -582,14 +583,13 @@ void r300EmitBlit(r300ContextPtr rmesa,
 	assert(w < (1 << 16));
 	assert(h < (1 << 16));
 
-	cmd =
-	    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 8,
+	cmd = (drm_r300_cmd_header_t *) r300AllocCmdBuf(rmesa, 8,
 							__FUNCTION__);
 
 	cmd[0].header.cmd_type = R300_CMD_PACKET3;
 	cmd[0].header.pad0 = R300_CMD_PACKET3_RAW;
-	cmd[1].i = R200_CP_CMD_BITBLT_MULTI | (5 << 16);
-	cmd[2].i = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
+	cmd[1].u = R200_CP_CMD_BITBLT_MULTI | (5 << 16);
+	cmd[2].u = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 		    RADEON_GMC_DST_PITCH_OFFSET_CNTL |
 		    RADEON_GMC_BRUSH_NONE |
 		    (color_fmt << 8) |
@@ -598,28 +598,24 @@ void r300EmitBlit(r300ContextPtr rmesa,
 		    RADEON_DP_SRC_SOURCE_MEMORY |
 		    RADEON_GMC_CLR_CMP_CNTL_DIS | RADEON_GMC_WR_MSK_DIS);
 
-	cmd[3].i = ((src_pitch / 64) << 22) | (src_offset >> 10);
-	cmd[4].i = ((dst_pitch / 64) << 22) | (dst_offset >> 10);
-	cmd[5].i = (srcx << 16) | srcy;
-	cmd[6].i = (dstx << 16) | dsty;	/* dst */
-	cmd[7].i = (w << 16) | h;
+	cmd[3].u = ((src_pitch / 64) << 22) | (src_offset >> 10);
+	cmd[4].u = ((dst_pitch / 64) << 22) | (dst_offset >> 10);
+	cmd[5].u = (srcx << 16) | srcy;
+	cmd[6].u = (dstx << 16) | dsty;	/* dst */
+	cmd[7].u = (w << 16) | h;
 }
 
 void r300EmitWait(r300ContextPtr rmesa, GLuint flags)
 {
-	if (rmesa->radeon.dri.drmMinor >= 6) {
-		drm_radeon_cmd_header_t *cmd;
-
-		assert(!(flags & ~(R300_WAIT_2D | R300_WAIT_3D)));
-
-		cmd =
-		    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa,
-								1,
-								__FUNCTION__);
-		cmd[0].i = 0;
-		cmd[0].wait.cmd_type = R300_CMD_WAIT;
-		cmd[0].wait.flags = flags;
-	}
+	drm_r300_cmd_header_t *cmd;
+
+	assert(!(flags & ~(R300_WAIT_2D | R300_WAIT_3D)));
+
+	cmd = (drm_r300_cmd_header_t *) r300AllocCmdBuf(rmesa, 1,
+							__FUNCTION__);
+	cmd[0].u = 0;
+	cmd[0].wait.cmd_type = R300_CMD_WAIT;
+	cmd[0].wait.flags = flags;
 }
 
 void r300EmitAOS(r300ContextPtr rmesa, GLuint nr, GLuint offset)
diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c
index db8b0a6e11c..07eaaedfbe8 100644
--- a/src/mesa/drivers/dri/r300/r300_context.c
+++ b/src/mesa/drivers/dri/r300/r300_context.c
@@ -59,6 +59,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_ioctl.h"
 #include "r300_tex.h"
 
+#ifdef USER_BUFFERS
+#include "radeon_mm.h"
+#endif
+
 #include "vblank.h"
 #include "utils.h"
 #include "xmlpool.h"		/* for symbolic values of enum-type options */
@@ -83,8 +87,7 @@ const struct dri_extension card_extensions[] = {
   {"GL_ARB_multitexture",		NULL},
   {"GL_ARB_texture_border_clamp",	NULL},
   {"GL_ARB_texture_compression",	GL_ARB_texture_compression_functions},
-/* disable until we support it, fixes a few things in ut2004 */
-/*	{"GL_ARB_texture_cube_map",	NULL}, */
+  {"GL_ARB_texture_cube_map",		NULL},
   {"GL_ARB_texture_env_add",		NULL},
   {"GL_ARB_texture_env_combine",	NULL},
   {"GL_ARB_texture_env_crossbar",	NULL},
@@ -155,75 +158,6 @@ static const struct tnl_pipeline_stage *r300_pipeline[] = {
 	0,
 };
 
-static void r300BufferData(GLcontext *ctx, GLenum target, GLsizeiptrARB size,
-		const GLvoid *data, GLenum usage, struct gl_buffer_object *obj)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	drm_radeon_mem_alloc_t alloc;
-	int offset, ret;
-
-	/* Free previous buffer */
-	if (obj->OnCard) {
-		drm_radeon_mem_free_t memfree;
-		
-		memfree.region = RADEON_MEM_REGION_GART;
-		memfree.region_offset = (char *)obj->Data - (char *)rmesa->radeon.radeonScreen->gartTextures.map;
-
-		ret = drmCommandWrite(rmesa->radeon.radeonScreen->driScreen->fd,
-				      DRM_RADEON_FREE, &memfree, sizeof(memfree));
-
-		if (ret) {
-			WARN_ONCE("Failed to free GART memroy!\n");
-		}
-		obj->OnCard = GL_FALSE;
-	}
-	
-	alloc.region = RADEON_MEM_REGION_GART;
-	alloc.alignment = 4;
-	alloc.size = size;
-	alloc.region_offset = &offset;
-
-	ret = drmCommandWriteRead( rmesa->radeon.dri.fd, DRM_RADEON_ALLOC, &alloc, sizeof(alloc));
-   	if (ret) {
-		WARN_ONCE("Ran out of GART memory!\n");
-		obj->Data = NULL;
-		_mesa_buffer_data(ctx, target, size, data, usage, obj);
-		return ;
-	}
-	obj->Data = ((GLubyte *)rmesa->radeon.radeonScreen->gartTextures.map) + offset;
-	
-	if (data)
-		memcpy(obj->Data, data, size);
-	
-	obj->Size = size;
-	obj->Usage = usage;
-	obj->OnCard = GL_TRUE;
-#if 0
-	fprintf(stderr, "allocated %d bytes at %p, offset=%d\n", size, obj->Data, offset);
-#endif
-}
-
-static void r300DeleteBuffer(GLcontext *ctx, struct gl_buffer_object *obj)
-{
-	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	
-	if(r300IsGartMemory(rmesa, obj->Data, obj->Size)){
-		drm_radeon_mem_free_t memfree;
-		int ret;
-		
-		memfree.region = RADEON_MEM_REGION_GART;
-		memfree.region_offset = (char *)obj->Data - (char *)rmesa->radeon.radeonScreen->gartTextures.map;
-
-		ret = drmCommandWrite(rmesa->radeon.radeonScreen->driScreen->fd,
-				      DRM_RADEON_FREE, &memfree, sizeof(memfree));
-
-		if(ret){
-			WARN_ONCE("Failed to free GART memroy!\n");
-		}
-		obj->Data = NULL;
-	}
-	_mesa_delete_buffer_object(ctx, obj);
-}
 
 /* Create the device specific rendering context.
  */
@@ -263,13 +197,14 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	r300InitTextureFuncs(&functions);
 	r300InitShaderFuncs(&functions);
 	
-#if 0 /* Needs various Mesa changes... */
+#ifdef USER_BUFFERS
+	radeon_mm_init(r300);
+#endif
+#ifdef HW_VBOS
 	if (hw_tcl_on) {
-		functions.BufferData = r300BufferData;
-		functions.DeleteBuffer = r300DeleteBuffer;
+		r300_init_vbo_funcs(&functions);
 	}
-#endif
-	
+#endif	
 	if (!radeonInitContext(&r300->radeon, &functions,
 			       glVisual, driContextPriv, sharedContextPrivate)) {
 		FREE(r300);
@@ -331,6 +266,11 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	ctx->Const.MinLineWidthAA = 1.0;
 	ctx->Const.MaxLineWidth = R300_LINESIZE_MAX;
 	ctx->Const.MaxLineWidthAA = R300_LINESIZE_MAX;
+	
+#ifdef USER_BUFFERS
+	/* Needs further modifications */
+	//ctx->Const.MaxArrayLockSize = (/*512*/RADEON_BUFFER_SIZE*16*1024) / (4*4);
+#endif
 
 	/* Initialize the software rasterizer and helper modules.
 	 */
@@ -391,6 +331,10 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	radeonInitSpanFuncs(ctx);
 	r300InitCmdBuf(r300);
 	r300InitState(r300);
+	
+#ifdef RADEON_VTXFMT_A
+	radeon_init_vtxfmt_a(r300);
+#endif
 
 #if 0
 	/* plug in a few more device driver functions */
diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h
index b4ba612af6b..cb2e4bb3218 100644
--- a/src/mesa/drivers/dri/r300/r300_context.h
+++ b/src/mesa/drivers/dri/r300/r300_context.h
@@ -48,6 +48,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_context.h"
 
 #define USE_ARB_F_P 1
+//#define USER_BUFFERS
+//#define RADEON_VTXFMT_A
+//#define HW_VBOS
+//#define OPTIMIZE_ELTS
 
 struct r300_context;
 typedef struct r300_context r300ContextRec;
@@ -109,12 +113,15 @@ static __inline__ uint32_t r300PackFloat32(float fl)
 struct r300_dma_buffer {
 	int refcount;		/* the number of retained regions in buf */
 	drmBufPtr buf;
+	int id;
 };
-
+#ifdef USER_BUFFERS
+#define GET_START(rvb) (r300GartOffsetFromVirtual(rmesa, (rvb)->address+(rvb)->start))
+#else
 #define GET_START(rvb) (rmesa->radeon.radeonScreen->gart_buffer_offset +		\
 			(rvb)->address - rmesa->dma.buf0_address +	\
 			(rvb)->start)
-
+#endif
 /* A retained region, eg vertices for indexed vertices.
  */
 struct r300_dma_region {
@@ -738,6 +745,30 @@ struct r300_pixel_shader_state {
 #define REG_COLOR0	1
 #define REG_TEX0	2
 
+#ifdef USER_BUFFERS
+struct dt {
+	GLint size;
+	GLenum type;
+	GLsizei stride;
+	void *data;
+};
+
+struct radeon_vertex_buffer {
+	int Count;
+	void *Elts;
+	int elt_size;
+	int elt_min, elt_max; /* debug */
+	
+	struct dt AttribPtr[VERT_ATTRIB_MAX];
+	
+	struct tnl_prim  *Primitive;	              
+	GLuint      PrimitiveCount;
+	GLint LockFirst;
+	GLsizei LockCount;
+	int lock_uptodate;
+};
+#endif
+
 struct r300_aos_rec {
 	GLuint offset;
 	int element_size; /* in dwords */
@@ -761,6 +792,9 @@ struct r300_state {
 #endif
 	struct r300_dma_region aos[R300_MAX_AOS_ARRAYS];
 	int aos_count;
+#ifdef USER_BUFFERS
+	struct radeon_vertex_buffer VB;
+#endif
 
 	GLuint *Elts;
 	struct r300_dma_region elt_dma;
@@ -815,8 +849,19 @@ struct r300_context {
 	GLuint TexGenInputs;
 	GLuint TexGenCompSel;
 	GLmatrix tmpmat;
+#ifdef USER_BUFFERS
+	key_t mm_ipc_key;
+	int mm_shm_id;
+	int mm_sem_id;
+	struct radeon_memory_manager *rmm;
+#endif
 };
 
+struct r300_buffer_object {
+	struct gl_buffer_object mesa_obj;
+	int id;
+};
+		
 #define R300_CONTEXT(ctx)		((r300ContextPtr)(ctx->DriverCtx))
 
 static __inline GLuint r300PackColor( GLuint cpp,
@@ -841,4 +886,12 @@ void translate_vertex_shader(struct r300_vertex_program *vp);
 extern void r300InitShaderFuncs(struct dd_function_table *functions);
 extern void r300VertexProgUpdateParams(GLcontext *ctx, struct r300_vertex_program *vp);
 
+#ifdef RADEON_VTXFMT_A
+extern void radeon_init_vtxfmt_a(r300ContextPtr rmesa);
+#endif
+
+#ifdef HW_VBOS
+extern void r300_init_vbo_funcs(struct dd_function_table *functions);
+#endif
+
 #endif				/* __R300_CONTEXT_H__ */
diff --git a/src/mesa/drivers/dri/r300/r300_ioctl.c b/src/mesa/drivers/dri/r300/r300_ioctl.c
index 82379ebb371..56a5ec9473a 100644
--- a/src/mesa/drivers/dri/r300/r300_ioctl.c
+++ b/src/mesa/drivers/dri/r300/r300_ioctl.c
@@ -71,6 +71,8 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
 	r300ContextPtr rmesa=r300;
 	LOCAL_VARS;
 #else
+	r300ContextPtr rmesa=r300;
+	LOCAL_VARS;
 	int i;
 #endif
 	
@@ -261,6 +263,11 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
 
 	r300EmitState(r300);
 #else
+#if 1
+	cp_wait(r300, R300_WAIT_3D | R300_WAIT_3D_CLEAN);
+	end_3d(PASS_PREFIX_VOID);
+#endif
+	
 	R300_STATECHANGE(r300, cb);
 	reg_start(R300_RB3D_COLOROFFSET0, 0);
 	e32(cboffset);
@@ -270,6 +277,9 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
 	else
 		cbpitch |= R300_COLOR_FORMAT_RGB565;
 	
+	if (r300->radeon.sarea->tiling_enabled)
+		cbpitch |= R300_COLOR_TILE_ENABLE;
+	
 	reg_start(R300_RB3D_COLORPITCH0, 0);
 	e32(cbpitch);
 
@@ -352,6 +362,15 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
 	cmd2[7].u = r300PackFloat32(ctx->Color.ClearColor[2]);
 	cmd2[8].u = r300PackFloat32(ctx->Color.ClearColor[3]);
 
+#if 1
+	reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+	e32(0x0000000a);
+	  
+
+	reg_start(0x4f18,0);
+	e32(0x00000003);
+	cp_wait(rmesa, R300_WAIT_3D | R300_WAIT_3D_CLEAN);
+#endif
 }
 
 #ifdef CB_DPATH
@@ -422,7 +441,7 @@ static void r300EmitClearState(GLcontext * ctx)
 	R300_STATECHANGE(r300, rc);
 	/* The second constant is needed to get glxgears display anything .. */
 	reg_start(R300_RS_CNTL_0, 1);
-	e32(R300_RS_CNTL_0_UNKNOWN_7 | R300_RS_CNTL_0_UNKNOWN_18);
+	e32((1 << R300_RS_CNTL_CI_CNT_SHIFT) | R300_RS_CNTL_0_UNKNOWN_18);
 	e32(0);
 	
 	R300_STATECHANGE(r300, rr);
@@ -477,6 +496,8 @@ static void r300EmitClearState(GLcontext * ctx)
 	e32(VP_ZERO());
 	e32(0);
 	
+	/*reg_start(0x4500,0);
+	e32(2560-1);*/
 }
 #endif
 
@@ -561,6 +582,7 @@ static void r300Clear(GLcontext * ctx, GLbitfield mask, GLboolean all,
 #endif
 }
 
+
 void r300Flush(GLcontext * ctx)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
@@ -572,6 +594,104 @@ void r300Flush(GLcontext * ctx)
 		r300FlushCmdBuf(r300, __FUNCTION__);
 }
 
+#ifdef USER_BUFFERS
+#include "radeon_mm.h"
+
+void r300RefillCurrentDmaRegion(r300ContextPtr rmesa)
+{
+	struct r300_dma_buffer *dmabuf;
+	int fd = rmesa->radeon.dri.fd;
+	int index = 0;
+	int size = 0;
+	drmDMAReq dma;
+	int ret;
+	
+	if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
+		fprintf(stderr, "%s\n", __FUNCTION__);
+
+	if (rmesa->dma.flush) {
+		rmesa->dma.flush(rmesa);
+	}
+
+	if (rmesa->dma.current.buf)
+		r300ReleaseDmaRegion(rmesa, &rmesa->dma.current, __FUNCTION__);
+
+	if (rmesa->dma.nr_released_bufs > 4)
+		r300FlushCmdBuf(rmesa, __FUNCTION__);
+	
+	dmabuf = CALLOC_STRUCT(r300_dma_buffer);
+	dmabuf->buf = (void *)1; /* hack */
+	dmabuf->refcount = 1;
+
+	dmabuf->id = radeon_mm_alloc(rmesa, 4, RADEON_BUFFER_SIZE*16);
+			
+	rmesa->dma.current.buf = dmabuf;
+	rmesa->dma.current.address = radeon_mm_ptr(rmesa, dmabuf->id);
+	rmesa->dma.current.end = RADEON_BUFFER_SIZE*16;
+	rmesa->dma.current.start = 0;
+	rmesa->dma.current.ptr = 0;
+}
+
+void r300ReleaseDmaRegion(r300ContextPtr rmesa,
+			  struct r300_dma_region *region, const char *caller)
+{
+	if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "%s from %s\n", __FUNCTION__, caller);
+
+	if (!region->buf)
+		return;
+
+	if (rmesa->dma.flush)
+		rmesa->dma.flush(rmesa);
+
+	if (--region->buf->refcount == 0) {
+		radeon_mm_free(rmesa, region->buf->id);
+		FREE(region->buf);
+		rmesa->dma.nr_released_bufs++;
+	}
+
+	region->buf = 0;
+	region->start = 0;
+}
+
+/* Allocates a region from rmesa->dma.current.  If there isn't enough
+ * space in current, grab a new buffer (and discard what was left of current)
+ */
+void r300AllocDmaRegion(r300ContextPtr rmesa,
+			struct r300_dma_region *region,
+			int bytes, int alignment)
+{
+	if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "%s %d\n", __FUNCTION__, bytes);
+
+	if (rmesa->dma.flush)
+		rmesa->dma.flush(rmesa);
+
+	if (region->buf)
+		r300ReleaseDmaRegion(rmesa, region, __FUNCTION__);
+
+	alignment--;
+	rmesa->dma.current.start = rmesa->dma.current.ptr =
+	    (rmesa->dma.current.ptr + alignment) & ~alignment;
+
+	if (rmesa->dma.current.ptr + bytes > rmesa->dma.current.end)
+		r300RefillCurrentDmaRegion(rmesa);
+
+	region->start = rmesa->dma.current.start;
+	region->ptr = rmesa->dma.current.start;
+	region->end = rmesa->dma.current.start + bytes;
+	region->address = rmesa->dma.current.address;
+	region->buf = rmesa->dma.current.buf;
+	region->buf->refcount++;
+
+	rmesa->dma.current.ptr += bytes;	/* bug - if alignment > 7 */
+	rmesa->dma.current.start =
+	    rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7;
+
+	assert(rmesa->dma.current.ptr <= rmesa->dma.current.end);
+}
+
+#else
 void r300RefillCurrentDmaRegion(r300ContextPtr rmesa)
 {
 	struct r300_dma_buffer *dmabuf;
@@ -714,6 +834,8 @@ void r300AllocDmaRegion(r300ContextPtr rmesa,
 	assert(rmesa->dma.current.ptr <= rmesa->dma.current.end);
 }
 
+#endif
+
 /* Called via glXGetMemoryOffsetMESA() */
 GLuint r300GetMemoryOffsetMESA(__DRInativeDisplay * dpy, int scrn,
 			       const GLvoid * pointer)
diff --git a/src/mesa/drivers/dri/r300/r300_maos.c b/src/mesa/drivers/dri/r300/r300_maos.c
index 1908ac07acd..52827c2d5dc 100644
--- a/src/mesa/drivers/dri/r300/r300_maos.c
+++ b/src/mesa/drivers/dri/r300/r300_maos.c
@@ -50,6 +50,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_maos.h"
 #include "r300_ioctl.h"
 
+#ifdef USER_BUFFERS
+#include "radeon_mm.h"
+#endif
+
 #define DEBUG_ALL DEBUG_VERTS
 
 
@@ -173,15 +177,11 @@ static void emit_vector(GLcontext * ctx,
 		fprintf(stderr, "%s count %d size %d stride %d\n",
 			__FUNCTION__, count, size, stride);
 
-	if(r300IsGartMemory(rmesa, data, size*stride)){
-		rvb->address = rmesa->radeon.radeonScreen->gartTextures.map;
-		rvb->start = (char *)data - rvb->address;
+	if(r300IsGartMemory(rmesa, data, /*(count-1)*stride */ 4)){
+		rvb->address = data;
+		rvb->start = 0;
 		rvb->aos_offset = r300GartOffsetFromVirtual(rmesa, data);
-		
-		if(stride == 0)
-			rvb->aos_stride	= 0;
-		else
-			rvb->aos_stride	= stride / 4;
+		rvb->aos_stride	= stride / 4 ;
 		
 		rvb->aos_size = size;
 		return;
@@ -226,26 +226,30 @@ static void emit_vector(GLcontext * ctx,
 
 }
 
-void r300EmitElts(GLcontext * ctx, GLuint *elts, unsigned long n_elts)
+void r300EmitElts(GLcontext * ctx, void *elts, unsigned long n_elts, int elt_size)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	struct r300_dma_region *rvb=&rmesa->state.elt_dma;
-	unsigned short int *out;
-	int i;
+	void *out;
 	
-	if(r300IsGartMemory(rmesa, elts, n_elts*sizeof(unsigned short int))){
+	assert(elt_size == 2 || elt_size == 4);
+	
+	if(r300IsGartMemory(rmesa, elts, n_elts * elt_size)){
 		rvb->address = rmesa->radeon.radeonScreen->gartTextures.map;
-		rvb->start = (char *)elts - rvb->address;
+		rvb->start = ((char *)elts) - rvb->address;
 		rvb->aos_offset = rmesa->radeon.radeonScreen->gart_texture_offset + rvb->start;
+		
 		return ;
+	}else if(r300IsGartMemory(rmesa, elts, 1)){
+		WARN_ONCE("Pointer not within GART memory!\n");
+		exit(1);
 	}
 	
-	r300AllocDmaRegion(rmesa, rvb, n_elts*sizeof(unsigned short int), 2);
+	r300AllocDmaRegion(rmesa, rvb, n_elts * elt_size, elt_size);
+	rvb->aos_offset	= GET_START(rvb);
 	
-	out = (unsigned short int *)(rvb->address + rvb->start);
-	
-	for(i=0; i < n_elts; i++)
-		out[i]=(unsigned short int)elts[i];
+	out = rvb->address + rvb->start;
+	memcpy(out, elts, n_elts * elt_size);
 }
 
 /* Emit vertex data to GART memory (unless immediate mode)
@@ -549,6 +553,362 @@ void r300EmitArrays(GLcontext * ctx, GLboolean immd)
 	rmesa->state.aos_count = nr;
 }
 
+#ifdef RADEON_VTXFMT_A
+void r300EmitArraysVtx(GLcontext * ctx, GLboolean immd)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	r300ContextPtr r300 = rmesa;
+	struct radeon_vertex_buffer *VB = &rmesa->state.VB;
+	//struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+	GLuint nr = 0;
+	GLuint count = VB->Count;
+	GLuint dw,mask;
+	GLuint vic_1 = 0;	/* R300_VAP_INPUT_CNTL_1 */
+	GLuint aa_vap_reg = 0; /* VAP register assignment */
+	GLuint i;
+	GLuint inputs = 0;
+	
+
+#define CONFIGURE_AOS(r, f, v, sz, cn) { \
+		if (RADEON_DEBUG & DEBUG_STATE) \
+			fprintf(stderr, "Enabling "#v "\n"); \
+		if (++nr >= R300_MAX_AOS_ARRAYS) { \
+			fprintf(stderr, "Aieee! AOS array count exceeded!\n"); \
+			exit(-1); \
+		} \
+		\
+		if (hw_tcl_on == GL_FALSE) \
+			rmesa->state.aos[nr-1].aos_reg = aa_vap_reg++; \
+		rmesa->state.aos[nr-1].aos_format = f; \
+		if (immd) { \
+			rmesa->state.aos[nr-1].aos_size = 4; \
+			rmesa->state.aos[nr-1].aos_stride = 4; \
+			rmesa->state.aos[nr-1].aos_offset = 0; \
+		} else { \
+			emit_vector(ctx, \
+						&rmesa->state.aos[nr-1], \
+						v.data, \
+						sz, \
+						v.stride, \
+						cn); \
+		rmesa->state.vap_reg.r=rmesa->state.aos[nr-1].aos_reg; \
+		} \
+}
+
+	if (hw_tcl_on) {
+		GLuint InputsRead = CURRENT_VERTEX_SHADER(ctx)->InputsRead;
+		struct r300_vertex_program *prog=(struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx);
+		if (InputsRead & (1<<VERT_ATTRIB_POS)) {
+			inputs |= _TNL_BIT_POS;
+			rmesa->state.aos[nr++].aos_reg = prog->inputs[VERT_ATTRIB_POS];
+		}
+		if (InputsRead & (1<<VERT_ATTRIB_NORMAL)) {
+			inputs |= _TNL_BIT_NORMAL;
+			rmesa->state.aos[nr++].aos_reg = prog->inputs[VERT_ATTRIB_NORMAL];
+		}
+		if (InputsRead & (1<<VERT_ATTRIB_COLOR0)) {
+			inputs |= _TNL_BIT_COLOR0;
+			rmesa->state.aos[nr++].aos_reg = prog->inputs[VERT_ATTRIB_COLOR0];
+		}
+		if (InputsRead & (1<<VERT_ATTRIB_COLOR1)) {
+			inputs |= _TNL_BIT_COLOR1;
+			rmesa->state.aos[nr++].aos_reg = prog->inputs[VERT_ATTRIB_COLOR1];
+		}
+		if (InputsRead & (1<<VERT_ATTRIB_FOG)) {
+			inputs |= _TNL_BIT_FOG;
+			rmesa->state.aos[nr++].aos_reg = prog->inputs[VERT_ATTRIB_FOG];
+		}
+		if(ctx->Const.MaxTextureUnits > 8) { /* Not sure if this can even happen... */
+			fprintf(stderr, "%s: Cant handle that many inputs\n", __FUNCTION__);
+			exit(-1);
+		}
+		for (i=0;i<ctx->Const.MaxTextureUnits;i++) {
+			if (InputsRead & (1<<(VERT_ATTRIB_TEX0+i))) {
+				inputs |= _TNL_BIT_TEX0<<i;
+				rmesa->state.aos[nr++].aos_reg = prog->inputs[VERT_ATTRIB_TEX0+i];
+			}
+		}
+		nr = 0;
+	} else {
+		inputs = TNL_CONTEXT(ctx)->render_inputs;
+	}
+	rmesa->state.render_inputs = inputs;
+
+	if (inputs & _TNL_BIT_POS) {
+		CONFIGURE_AOS(i_coords,	AOS_FORMAT_FLOAT,
+						VB->AttribPtr[VERT_ATTRIB_POS],
+						immd ? 4 : VB->AttribPtr[VERT_ATTRIB_POS].size,
+						count);
+
+		vic_1 |= R300_INPUT_CNTL_POS;
+	}
+
+	if (inputs & _TNL_BIT_NORMAL) {
+		CONFIGURE_AOS(i_normal,	AOS_FORMAT_FLOAT,
+						VB->AttribPtr[VERT_ATTRIB_NORMAL],
+						immd ? 4 : VB->AttribPtr[VERT_ATTRIB_NORMAL].size,
+						count);
+
+		vic_1 |= R300_INPUT_CNTL_NORMAL;
+	}
+
+	if (inputs & _TNL_BIT_COLOR0) {
+		int emitsize=4;
+
+		if (!immd) {
+			if (VB->AttribPtr[VERT_ATTRIB_COLOR0].size == 4 &&
+			    (VB->AttribPtr[VERT_ATTRIB_COLOR0].stride != 0 ||
+			     ((float*)VB->AttribPtr[VERT_ATTRIB_COLOR0].data)[3] != 1.0)) {
+				emitsize = 4;
+			} else {
+				emitsize = 3;
+			}//emitsize = VB->AttribPtr[VERT_ATTRIB_COLOR0].size;
+		}
+		if(VB->AttribPtr[VERT_ATTRIB_COLOR0].type == GL_UNSIGNED_BYTE)
+			emitsize = 1;
+
+		CONFIGURE_AOS(i_color[0], VB->AttribPtr[VERT_ATTRIB_COLOR0].type == GL_UNSIGNED_BYTE ? AOS_FORMAT_UBYTE : AOS_FORMAT_FLOAT_COLOR,
+						VB->AttribPtr[VERT_ATTRIB_COLOR0],
+						immd ? 4 : emitsize,
+						count);
+
+		vic_1 |= R300_INPUT_CNTL_COLOR;
+	}
+
+	if (inputs & _TNL_BIT_COLOR1) {
+		CONFIGURE_AOS(i_color[1], AOS_FORMAT_FLOAT_COLOR,
+						VB->AttribPtr[VERT_ATTRIB_COLOR1],
+						immd ? 4 : VB->AttribPtr[VERT_ATTRIB_COLOR1].size,
+						count);
+	}
+
+#if 0
+	if (inputs & _TNL_BIT_FOG) {
+		CONFIGURE_AOS(	AOS_FORMAT_FLOAT,
+						VB->FogCoordPtr,
+						immd ? 4 : VB->FogCoordPtr->size,
+						count);
+	}
+#endif
+
+	r300->state.texture.tc_count = 0;
+	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
+		if (inputs & (_TNL_BIT_TEX0 << i)) {
+			CONFIGURE_AOS(i_tex[i], AOS_FORMAT_FLOAT,
+							VB->AttribPtr[VERT_ATTRIB_TEX0+i],
+							immd ? 4 : VB->AttribPtr[VERT_ATTRIB_TEX0+i].size,
+							count);
+
+			vic_1 |= R300_INPUT_CNTL_TC0 << i;
+			r300->state.texture.tc_count++;
+		}
+	}
+	for(i=0; i < nr; i++)
+		if(r300->state.aos[i].aos_format == 2){
+			assert(r300->state.aos[i].aos_size == 1);
+			r300->state.aos[i].aos_size=5;
+		}
+	
+#define SHOW_INFO(n) do { \
+	if (RADEON_DEBUG & DEBUG_ALL) { \
+	fprintf(stderr, "RR[%d] - sz=%d, reg=%d, fmt=%d -- st=%d, of=0x%08x\n", \
+		n, \
+		r300->state.aos[n].aos_size, \
+		r300->state.aos[n].aos_reg, \
+		r300->state.aos[n].aos_format, \
+		r300->state.aos[n].aos_stride, \
+		r300->state.aos[n].aos_offset); \
+	} \
+} while(0);
+
+	/* setup INPUT_ROUTE */
+	R300_STATECHANGE(r300, vir[0]);
+	for(i=0;i+1<nr;i+=2){
+		SHOW_INFO(i)
+		SHOW_INFO(i+1)
+		dw=(r300->state.aos[i].aos_size-1)
+		| ((r300->state.aos[i].aos_reg)<<8)
+		| (r300->state.aos[i].aos_format<<14)
+		| (((r300->state.aos[i+1].aos_size-1)
+		| ((r300->state.aos[i+1].aos_reg)<<8)
+		| (r300->state.aos[i+1].aos_format<<14))<<16);
+
+		if(i+2==nr){
+			dw|=(1<<(13+16));
+			}
+		r300->hw.vir[0].cmd[R300_VIR_CNTL_0+(i>>1)]=dw;
+		}
+	if(nr & 1){
+		SHOW_INFO(nr-1)
+		dw=(r300->state.aos[nr-1].aos_size-1)
+		| (r300->state.aos[nr-1].aos_format<<14)
+		| ((r300->state.aos[nr-1].aos_reg)<<8)
+		| (1<<13);
+		r300->hw.vir[0].cmd[R300_VIR_CNTL_0+(nr>>1)]=dw;
+		//fprintf(stderr, "vir0 dw=%08x\n", dw);
+		}
+	/* Set the rest of INPUT_ROUTE_0 to 0 */
+	//for(i=((count+1)>>1); i<8; i++)r300->hw.vir[0].cmd[R300_VIR_CNTL_0+i]=(0x0);
+	((drm_r300_cmd_header_t*)r300->hw.vir[0].cmd)->packet0.count = (nr+1)>>1;
+
+
+	/* Mesa assumes that all missing components are from (0, 0, 0, 1) */
+#define ALL_COMPONENTS ((R300_INPUT_ROUTE_SELECT_X<<R300_INPUT_ROUTE_X_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_Y<<R300_INPUT_ROUTE_Y_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_Z<<R300_INPUT_ROUTE_Z_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_W<<R300_INPUT_ROUTE_W_SHIFT))
+
+#define ALL_DEFAULT ((R300_INPUT_ROUTE_SELECT_ZERO<<R300_INPUT_ROUTE_X_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_ZERO<<R300_INPUT_ROUTE_Y_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_ZERO<<R300_INPUT_ROUTE_Z_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_ONE<<R300_INPUT_ROUTE_W_SHIFT))
+
+	R300_STATECHANGE(r300, vir[1]);
+
+	for(i=0; i < nr; i++)
+		if(r300->state.aos[i].aos_format == 2){
+			assert(r300->state.aos[i].aos_size == 5);
+			r300->state.aos[i].aos_size=/*3*/4; /* XXX */
+		}
+		
+		
+	for(i=0;i+1<nr;i+=2){
+		/* do i first.. */
+		mask=(1<<(r300->state.aos[i].aos_size*3))-1;
+		dw=(ALL_COMPONENTS & mask)
+		| (ALL_DEFAULT & ~mask)
+		| R300_INPUT_ROUTE_ENABLE;
+
+		/* i+1 */
+		mask=(1<<(r300->state.aos[i+1].aos_size*3))-1;
+		dw|=(
+		(ALL_COMPONENTS & mask)
+		| (ALL_DEFAULT & ~mask)
+		| R300_INPUT_ROUTE_ENABLE
+		)<<16;
+
+		//fprintf(stderr, "vir1 dw=%08x\n", dw);
+		r300->hw.vir[1].cmd[R300_VIR_CNTL_0+(i>>1)]=dw;
+		}
+	if(nr & 1){
+		mask=(1<<(r300->state.aos[nr-1].aos_size*3))-1;
+		dw=(ALL_COMPONENTS & mask)
+		| (ALL_DEFAULT & ~mask)
+		| R300_INPUT_ROUTE_ENABLE;
+		r300->hw.vir[1].cmd[R300_VIR_CNTL_0+(nr>>1)]=dw;
+		//fprintf(stderr, "vir1 dw=%08x\n", dw);
+		}
+	/* Set the rest of INPUT_ROUTE_1 to 0 */
+	//for(i=((count+1)>>1); i<8; i++)r300->hw.vir[1].cmd[R300_VIR_CNTL_0+i]=0x0;
+	((drm_r300_cmd_header_t*)r300->hw.vir[1].cmd)->packet0.count = (nr+1)>>1;
+
+	/* Set up input_cntl */
+	/* I don't think this is needed for vertex buffers, but it doesn't hurt anything */
+	R300_STATECHANGE(r300, vic);
+	r300->hw.vic.cmd[R300_VIC_CNTL_0]=0x5555;  /* Hard coded value, no idea what it means */
+	r300->hw.vic.cmd[R300_VIC_CNTL_1]=vic_1;
+
+	for(i=0; i < nr; i++)
+		if(r300->state.aos[i].aos_format == 2){
+			assert(r300->state.aos[i].aos_size == /*3*/4); /* XXX */
+			r300->state.aos[i].aos_size=1;
+		}
+#if 0
+	r300->hw.vic.cmd[R300_VIC_CNTL_1]=0;
+
+	if(r300->state.render_inputs & _TNL_BIT_POS)
+		r300->hw.vic.cmd[R300_VIC_CNTL_1]|=R300_INPUT_CNTL_POS;
+
+	if(r300->state.render_inputs & _TNL_BIT_NORMAL)
+		r300->hw.vic.cmd[R300_VIC_CNTL_1]|=R300_INPUT_CNTL_NORMAL;
+
+	if(r300->state.render_inputs & _TNL_BIT_COLOR0)
+		r300->hw.vic.cmd[R300_VIC_CNTL_1]|=R300_INPUT_CNTL_COLOR;
+
+	for(i=0;i < ctx->Const.MaxTextureUnits;i++)
+		if(r300->state.render_inputs & (_TNL_BIT_TEX0<<i))
+			r300->hw.vic.cmd[R300_VIC_CNTL_1]|=(R300_INPUT_CNTL_TC0<<i);
+#endif
+
+	/* Stage 3: VAP output */
+	
+	R300_STATECHANGE(r300, vof);
+	
+	r300->hw.vof.cmd[R300_VOF_CNTL_0]=0;
+	r300->hw.vof.cmd[R300_VOF_CNTL_1]=0;
+	if (hw_tcl_on){
+		GLuint OutputsWritten = CURRENT_VERTEX_SHADER(ctx)->OutputsWritten;
+		
+		if(OutputsWritten & (1<<VERT_RESULT_HPOS))
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
+		if(OutputsWritten & (1<<VERT_RESULT_COL0))
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT;
+		/*if(OutputsWritten & (1<<VERT_RESULT_COL1))
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT;
+		if(OutputsWritten & (1<<VERT_RESULT_BFC0))
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT;
+		if(OutputsWritten & (1<<VERT_RESULT_BFC1))
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT;*/
+		//if(OutputsWritten & (1<<VERT_RESULT_FOGC))
+
+		if(OutputsWritten & (1<<VERT_RESULT_PSIZ))
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
+
+		for(i=0;i < ctx->Const.MaxTextureUnits;i++)
+			if(OutputsWritten & (1<<(VERT_RESULT_TEX0+i)))
+				r300->hw.vof.cmd[R300_VOF_CNTL_1] |= (4<<(3*i));
+	} else {
+		if(inputs & _TNL_BIT_POS)
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
+		if(inputs & _TNL_BIT_COLOR0)
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT;
+		if(inputs & _TNL_BIT_COLOR1)
+			r300->hw.vof.cmd[R300_VOF_CNTL_0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT;
+
+		for(i=0;i < ctx->Const.MaxTextureUnits;i++)
+			if(inputs & (_TNL_BIT_TEX0<<i))
+				r300->hw.vof.cmd[R300_VOF_CNTL_1]|=(4<<(3*i));
+	}
+
+	rmesa->state.aos_count = nr;
+}
+#endif
+
+#ifdef USER_BUFFERS
+void r300UseArrays(GLcontext * ctx)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	int i;
+	
+	if(rmesa->state.elt_dma.buf)
+		radeon_mm_use(rmesa, rmesa->state.elt_dma.buf->id);
+	
+	for (i=0; i < rmesa->state.aos_count;i++) {
+		if (rmesa->state.aos[i].buf)
+			radeon_mm_use(rmesa, rmesa->state.aos[i].buf->id);
+	}
+	
+#ifdef HW_VBOS
+	
+#define USE_VBO(a) if (ctx->Array.a.BufferObj->Name && ctx->Array.a.Enabled) \
+		radeon_mm_use(rmesa, ((struct r300_buffer_object *)ctx->Array.a.BufferObj)->id)
+	
+	if (ctx->Array.ElementArrayBufferObj->Name && ctx->Array.ElementArrayBufferObj->OnCard)
+		radeon_mm_use(rmesa, ((struct r300_buffer_object *)ctx->Array.ElementArrayBufferObj)->id);
+	
+	USE_VBO(Vertex);
+	USE_VBO(Normal);
+	USE_VBO(Color);
+	USE_VBO(SecondaryColor);
+	USE_VBO(FogCoord);
+	
+	for (i=0; i < MAX_TEXTURE_COORD_UNITS; i++)
+		USE_VBO(TexCoord[i]);
+#endif
+	
+}
+#endif
+
 void r300ReleaseArrays(GLcontext * ctx)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
diff --git a/src/mesa/drivers/dri/r300/r300_maos.h b/src/mesa/drivers/dri/r300/r300_maos.h
index f76c94a2191..c75589085bf 100644
--- a/src/mesa/drivers/dri/r300/r300_maos.h
+++ b/src/mesa/drivers/dri/r300/r300_maos.h
@@ -40,8 +40,17 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 #include "r300_context.h"
 
-extern void r300EmitElts(GLcontext * ctx, GLuint *elts, unsigned long n_elts);
+extern void r300EmitElts(GLcontext * ctx, void *elts, unsigned long n_elts, int elt_size);
 extern void r300EmitArrays(GLcontext * ctx, GLboolean immd);
+
+#ifdef RADEON_VTXFMT_A
+extern void r300EmitArraysVtx(GLcontext * ctx, GLboolean immd);
+#endif
+
+#ifdef USER_BUFFERS
+void r300UseArrays(GLcontext * ctx);
+#endif
+
 extern void r300ReleaseArrays(GLcontext * ctx);
 
 #endif
diff --git a/src/mesa/drivers/dri/r300/r300_reg.h b/src/mesa/drivers/dri/r300/r300_reg.h
index 07c453027ba..32cac350fa5 100644
--- a/src/mesa/drivers/dri/r300/r300_reg.h
+++ b/src/mesa/drivers/dri/r300/r300_reg.h
@@ -728,6 +728,7 @@ I am fairly certain that they are correct unless stated otherwise in comments.
 #	define R300_TX_FORMAT_G8R8_G8B8	    	    0x15     /* no swizzle */
 						  /* 0x16 - some 16 bit green format.. ?? */
 #	define R300_TX_FORMAT_UNK25		   (1 << 25) /* no swizzle */
+#	define R300_TX_FORMAT_CUBIC_MAP		   (1 << 26)
 
 	/* gap */
 	/* Floating point formats */
diff --git a/src/mesa/drivers/dri/r300/r300_render.c b/src/mesa/drivers/dri/r300/r300_render.c
index a0769e27f89..8f74007faa9 100644
--- a/src/mesa/drivers/dri/r300/r300_render.c
+++ b/src/mesa/drivers/dri/r300/r300_render.c
@@ -417,27 +417,81 @@ static GLboolean r300_run_immediate_render(GLcontext *ctx,
 
 /* vertex buffer implementation */
 
-static void inline fire_EB(PREFIX unsigned long addr, int vertex_count, int type)
+static void inline fire_EB(PREFIX unsigned long addr, int vertex_count, int type, int elt_size)
 {
 	LOCAL_VARS
 	unsigned long addr_a;
+	unsigned long t_addr;
+	unsigned long magic_1, magic_2;
+	GLcontext *ctx;
+	ctx = rmesa->radeon.glCtx; 
 	
-	if(addr & 1){
+	assert(elt_size == 2 || elt_size == 4);
+	
+	if(addr & (elt_size-1)){
 		WARN_ONCE("Badly aligned buffer\n");
 		return ;
 	}
-	addr_a = 0; /*addr & 0x1c;*/
+#ifdef OPTIMIZE_ELTS
+	addr_a = 0;
+	
+	magic_1 = (addr % 32) / 4;
+	t_addr = addr & (~0x1d);
+	magic_2 = (vertex_count + 1 + (t_addr & 0x2)) / 2 + magic_1;
+	
+	check_space(6);
+	
+	start_packet3(RADEON_CP_PACKET3_3D_DRAW_INDX_2, 0);
+	if(elt_size == 4){
+		e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
+	} else {
+		e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type);
+	}
+
+	start_packet3(RADEON_CP_PACKET3_INDX_BUFFER, 2);
+	if(elt_size == 4){
+		e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
+		e32(addr /*& 0xffffffe3*/);
+	} else {
+		e32(R300_EB_UNK1 | (magic_1 << 16) | R300_EB_UNK2);
+		e32(t_addr);
+	}
+	
+	if(elt_size == 4){
+		e32(vertex_count /*+ addr_a/4*/); /* Total number of dwords needed? */
+	} else {
+		e32(magic_2); /* Total number of dwords needed? */
+	}
+	//cp_delay(PASS_PREFIX 1);
+#if 0
+	fprintf(stderr, "magic_1 %d\n", magic_1);
+	fprintf(stderr, "t_addr %x\n", t_addr);
+	fprintf(stderr, "magic_2 %d\n", magic_2);
+	exit(1);
+#endif
+#else
+	addr_a = 0;
 	
 	check_space(6);
 	
 	start_packet3(RADEON_CP_PACKET3_3D_DRAW_INDX_2, 0);
-	/* TODO: R300_VAP_VF_CNTL__INDEX_SIZE_32bit . */
-	e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type);
+	if(elt_size == 4){
+		e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type | R300_VAP_VF_CNTL__INDEX_SIZE_32bit);
+	} else {
+		e32(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (vertex_count<<16) | type);
+	}
 
 	start_packet3(RADEON_CP_PACKET3_INDX_BUFFER, 2);
-	e32(R300_EB_UNK1 | (addr_a << 16) | R300_EB_UNK2);
+	e32(R300_EB_UNK1 | (0 << 16) | R300_EB_UNK2);
 	e32(addr /*& 0xffffffe3*/);
-	e32((vertex_count+1)/2 /*+ addr_a/4*/); /* Total number of dwords needed? */
+	
+	if(elt_size == 4){
+		e32(vertex_count /*+ addr_a/4*/); /* Total number of dwords needed? */
+	} else {
+		e32((vertex_count+1)/2 /*+ addr_a/4*/); /* Total number of dwords needed? */
+	}
+	//cp_delay(PASS_PREFIX 1);
+#endif	
 }
 
 static void r300_render_vb_primitive(r300ContextPtr rmesa,
@@ -476,8 +530,8 @@ static void r300_render_vb_primitive(r300ContextPtr rmesa,
 		WARN_ONCE("Too many elts\n");
 		return;
 	}
-	r300EmitElts(ctx, rmesa->state.Elts+start, num_verts);
-	fire_EB(PASS_PREFIX GET_START(&(rmesa->state.elt_dma)), num_verts, type);
+	r300EmitElts(ctx, rmesa->state.Elts+start, num_verts, 4);
+	fire_EB(PASS_PREFIX GET_START(&(rmesa->state.elt_dma)), num_verts, type, 4);
 #endif
    }else{
 	   r300EmitAOS(rmesa, rmesa->state.aos_count, start);
@@ -500,7 +554,7 @@ static GLboolean r300_run_vb_render(GLcontext *ctx,
 
    	r300ReleaseArrays(ctx);
 	r300EmitArrays(ctx, GL_FALSE);
-
+	
 //	LOCK_HARDWARE(&(rmesa->radeon));
 
 	reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
@@ -526,6 +580,9 @@ static GLboolean r300_run_vb_render(GLcontext *ctx,
 	reg_start(0x4f18,0);
 	e32(0x00000003);
 
+#ifdef USER_BUFFERS
+	r300UseArrays(ctx);
+#endif
 //	end_3d(PASS_PREFIX_VOID);
 
    /* Flush state - we are done drawing.. */
@@ -536,6 +593,186 @@ static GLboolean r300_run_vb_render(GLcontext *ctx,
 	return GL_FALSE;
 }
 
+#ifdef RADEON_VTXFMT_A
+
+static void r300_render_vb_primitive_vtxfmt_a(r300ContextPtr rmesa,
+	GLcontext *ctx,
+	int start,
+	int end,
+	int prim)
+{
+   int type, num_verts;
+   radeonScreenPtr rsp=rmesa->radeon.radeonScreen;
+   LOCAL_VARS
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &tnl->vb;
+   int i;
+
+   type=r300_get_primitive_type(rmesa, ctx, prim);
+   num_verts=r300_get_num_verts(rmesa, ctx, end-start, prim);
+
+   if(type<0 || num_verts <= 0)return;
+
+   if(rmesa->state.VB.Elts){
+	r300EmitAOS(rmesa, rmesa->state.aos_count, /*0*/start);
+#if 0
+	start_index32_packet(num_verts, type);
+	for(i=0; i < num_verts; i++)
+		e32(((unsigned long *)rmesa->state.VB.Elts)[i]/*rmesa->state.Elts[start+i]*/); /* start ? */
+#else
+	WARN_ONCE("Rendering with elt buffers\n");
+	if(num_verts == 1){
+		//start_index32_packet(num_verts, type);
+		//e32(rmesa->state.Elts[start]);
+		return;
+	}
+	
+	if(num_verts > 65535){ /* not implemented yet */
+		WARN_ONCE("Too many elts\n");
+		return;
+	}
+	
+	r300EmitElts(ctx, rmesa->state.VB.Elts, num_verts, rmesa->state.VB.elt_size);
+	fire_EB(PASS_PREFIX rmesa->state.elt_dma.aos_offset, num_verts, type, rmesa->state.VB.elt_size);
+#endif
+   }else{
+	   r300EmitAOS(rmesa, rmesa->state.aos_count, start);
+	   fire_AOS(PASS_PREFIX num_verts, type);
+   }
+}
+
+void dump_array(struct r300_dma_region *rvb, int count)
+{
+	int *out = (int *)(rvb->address + rvb->start);
+	int i, ci;
+	
+	for (i=0; i < count; i++) {
+		fprintf(stderr, "{");
+		if (rvb->aos_format == AOS_FORMAT_FLOAT)
+			for (ci=0; ci < rvb->aos_size; ci++)
+				fprintf(stderr, "%f ", ((float *)out)[ci]);
+		else
+			for (ci=0; ci < rvb->aos_size; ci++)
+				fprintf(stderr, "%d ", ((unsigned char *)out)[ci]);
+		fprintf(stderr, "}");
+		
+		out += rvb->aos_stride;
+	}
+
+	fprintf(stderr, "\n");
+}
+		
+void dump_dt(struct dt *dt, int count)
+{
+	int *out = dt->data;
+	int i, ci;
+	
+	fprintf(stderr, "base at %p ", out);
+	
+	for (i=0; i < count; i++){
+		fprintf(stderr, "{");
+		if (dt->type == GL_FLOAT)
+			for (ci=0; ci < dt->size; ci++)
+				fprintf(stderr, "%f ", ((float *)out)[ci]);
+		else
+			for (ci=0; ci < dt->size; ci++)
+				fprintf(stderr, "%d ", ((unsigned char *)out)[ci]);
+		fprintf(stderr, "}");
+		
+		out = (char *)out + dt->stride;
+	}
+	
+	fprintf(stderr, "\n");
+}
+
+/*static */GLboolean r300_run_vb_render_vtxfmt_a(GLcontext *ctx,
+				 struct tnl_pipeline_stage *stage)
+{
+   r300ContextPtr rmesa = R300_CONTEXT(ctx);
+   //TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct radeon_vertex_buffer *VB = &rmesa->state.VB; //&tnl->vb;
+   int i, j;
+   LOCAL_VARS
+   
+	if (RADEON_DEBUG & DEBUG_PRIMS)
+		fprintf(stderr, "%s\n", __FUNCTION__);
+	
+	if (rmesa->state.VB.LockCount == 0) {
+ 	  	r300ReleaseArrays(ctx);
+		r300EmitArraysVtx(ctx, GL_FALSE);
+	} else {
+		/* TODO: Figure out why do we need these. */
+		R300_STATECHANGE(rmesa, vir[0]);
+		R300_STATECHANGE(rmesa, vir[1]);
+		R300_STATECHANGE(rmesa, vic);
+		R300_STATECHANGE(rmesa, vof);
+		
+#if 0		
+		fprintf(stderr, "dt:\n");
+		for(i=0; i < VERT_ATTRIB_MAX; i++){
+			fprintf(stderr, "dt %d:", i);
+			dump_dt(&rmesa->state.VB.AttribPtr[i], VB->Count);
+		}
+		
+		fprintf(stderr, "before:\n");
+		for(i=0; i < rmesa->state.aos_count; i++){
+			fprintf(stderr, "aos %d:", i);
+			dump_array(&rmesa->state.aos[i], VB->Count);
+		}
+#endif
+#if 0
+ 	  	r300ReleaseArrays(ctx);
+		r300EmitArraysVtx(ctx, GL_FALSE);
+			
+		fprintf(stderr, "after:\n");
+		for(i=0; i < rmesa->state.aos_count; i++){
+			fprintf(stderr, "aos %d:", i);
+			dump_array(&rmesa->state.aos[i], VB->Count);
+		}
+#endif
+	}
+	
+//	LOCK_HARDWARE(&(rmesa->radeon));
+
+	reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+	e32(0x0000000a);
+
+	reg_start(0x4f18,0);
+	e32(0x00000003);
+#if 0
+	reg_start(R300_VAP_PVS_WAITIDLE,0);
+		e32(0x00000000);
+#endif
+	r300EmitState(rmesa);
+	
+	for(i=0; i < VB->PrimitiveCount; i++){
+		GLuint prim = VB->Primitive[i].mode;
+		GLuint start = VB->Primitive[i].start;
+		GLuint length = VB->Primitive[i].count;
+		
+		r300_render_vb_primitive_vtxfmt_a(rmesa, ctx, start, start + length, prim);
+	}
+
+	reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+	e32(0x0000000a/*0x2*/);
+
+	reg_start(0x4f18,0);
+	e32(0x00000003/*0x1*/);
+
+#ifdef USER_BUFFERS
+	r300UseArrays(ctx);
+#endif
+//	end_3d(PASS_PREFIX_VOID);
+
+   /* Flush state - we are done drawing.. */
+//	r300FlushCmdBufLocked(rmesa, __FUNCTION__);
+//	radeonWaitForIdleLocked(&(rmesa->radeon));
+
+//	UNLOCK_HARDWARE(&(rmesa->radeon));
+	return GL_FALSE;
+}
+#endif
+
 /**
  * Called by the pipeline manager to render a batch of primitives.
  * We can return true to pass on to the next stage (i.e. software
@@ -678,26 +915,9 @@ static GLboolean r300_run_tcl_render(GLcontext *ctx,
 		fprintf(stderr, "%s\n", __FUNCTION__);
 	if(hw_tcl_on == GL_FALSE)
 		return GL_TRUE;
-	if(ctx->VertexProgram._Enabled == GL_FALSE){
-		_tnl_UpdateFixedFunctionProgram(ctx);
-	}
-	vp = (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx);
-	if(vp->translated == GL_FALSE)
-		translate_vertex_shader(vp);
-	if(vp->translated == GL_FALSE){
-		fprintf(stderr, "Failing back to sw-tcl\n");
-		debug_vp(ctx, &vp->mesa_program);
-		hw_tcl_on=future_hw_tcl_on=0;
-		r300ResetHwState(rmesa);
-		return GL_TRUE;
-	}
-		
-	r300_setup_textures(ctx);
-	r300_setup_rs_unit(ctx);
-
-	r300SetupVertexShader(rmesa);
-	r300SetupPixelShader(rmesa);
-
+	
+	r300UpdateShaderStates(rmesa);
+	
 	return r300_run_vb_render(ctx, stage);
 }
 
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index c0b4dcc8cab..e0f357d8c45 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1553,6 +1553,36 @@ void r300SetupVertexProgram(r300ContextPtr rmesa)
 #endif
 }
 
+extern int future_hw_tcl_on;
+void r300UpdateShaderStates(r300ContextPtr rmesa)
+{
+	GLcontext *ctx;
+	struct r300_vertex_program *vp;
+	
+	ctx = rmesa->radeon.glCtx; 
+	
+	if(ctx->VertexProgram._Enabled == GL_FALSE){
+		_tnl_UpdateFixedFunctionProgram(ctx);
+	}
+	vp = (struct r300_vertex_program *)CURRENT_VERTEX_SHADER(ctx);
+	if(vp->translated == GL_FALSE)
+		translate_vertex_shader(vp);
+	if(vp->translated == GL_FALSE){
+		fprintf(stderr, "Failing back to sw-tcl\n");
+		debug_vp(ctx, &vp->mesa_program);
+		hw_tcl_on=future_hw_tcl_on=0;
+		r300ResetHwState(rmesa);
+		
+		return ;
+	}
+		
+	r300_setup_textures(ctx);
+	r300_setup_rs_unit(ctx);
+
+	r300SetupVertexShader(rmesa);
+	r300SetupPixelShader(rmesa);
+}
+
 /* This is probably wrong for some values, I need to test this
  * some more.  Range checking would be a good idea also..
  * 
diff --git a/src/mesa/drivers/dri/r300/r300_state.h b/src/mesa/drivers/dri/r300/r300_state.h
index 131834fa782..3e157943711 100644
--- a/src/mesa/drivers/dri/r300/r300_state.h
+++ b/src/mesa/drivers/dri/r300/r300_state.h
@@ -43,6 +43,9 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 		r300->hw.is_dirty = GL_TRUE;		\
 	} while(0)
 
+#define R300_PRINT_STATE(r300, atom) \
+		r300_print_state_atom(r300, &r300->hw.atom)
+
 /* Fire the buffered vertices no matter what.
    TODO: This has not been implemented yet
  */
@@ -65,5 +68,6 @@ extern void r300SetupPixelShader(r300ContextPtr rmesa);
 
 extern void r300_setup_textures(GLcontext *ctx);
 extern void r300_setup_rs_unit(GLcontext *ctx);
+extern void r300UpdateShaderStates(r300ContextPtr rmesa);
 
 #endif /* __R300_STATE_H__ */
diff --git a/src/mesa/drivers/dri/r300/r300_texstate.c b/src/mesa/drivers/dri/r300/r300_texstate.c
index 9d929516c96..359a0b1b8af 100644
--- a/src/mesa/drivers/dri/r300/r300_texstate.c
+++ b/src/mesa/drivers/dri/r300/r300_texstate.c
@@ -341,9 +341,8 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 		t->format_x |= R200_TEXCOORD_VOLUME;
 	} else if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
 		ASSERT(log2Width == log2Height);
-		t->format |= ((log2Width << R200_TXFORMAT_F5_WIDTH_SHIFT) |
-				   (log2Height << R200_TXFORMAT_F5_HEIGHT_SHIFT)
-				   | (R200_TXFORMAT_CUBIC_MAP_ENABLE));
+		t->format |= R300_TX_FORMAT_CUBIC_MAP;
+		
 		t->format_x |= R200_TEXCOORD_CUBIC_ENV;
 		t->pp_cubic_faces = ((log2Width << R200_FACE_WIDTH_1_SHIFT) |
 				     (log2Height << R200_FACE_HEIGHT_1_SHIFT) |
diff --git a/src/mesa/drivers/dri/r300/r300_vertexprog.c b/src/mesa/drivers/dri/r300/r300_vertexprog.c
index 717832048cf..10fa22d5615 100644
--- a/src/mesa/drivers/dri/r300/r300_vertexprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertexprog.c
@@ -597,6 +597,7 @@ void translate_vertex_shader(struct r300_vertex_program *vp)
 		   Ops that need temp vars should probably be given reg indexes starting at the end of tmp area. */
 		switch(vpi->Opcode){
 		case VP_OPCODE_MOV://ADD RESULT 1.X Y Z W PARAM 0{} {X Y Z W} PARAM 0{} {ZERO ZERO ZERO ZERO} 
+#if 1
 			o_inst->op=MAKE_VSF_OP(R300_VPI_OUT_OP_ADD, t_dst_index(vp, &vpi->DstReg),
 					t_dst_mask(vpi->DstReg.WriteMask), t_dst_class(vpi->DstReg.File));
 			o_inst->src1=t_src(vp, &src[0]);
@@ -606,6 +607,23 @@ void translate_vertex_shader(struct r300_vertex_program *vp)
 					t_src_class(src[0].File), VSF_FLAG_NONE);
 
 			o_inst->src3=0;
+#else
+			hw_op=(src[0].File == PROGRAM_TEMPORARY) ? R300_VPI_OUT_OP_MAD_2 : R300_VPI_OUT_OP_MAD;
+			
+			o_inst->op=MAKE_VSF_OP(hw_op, t_dst_index(vp, &vpi->DstReg),
+				t_dst_mask(vpi->DstReg.WriteMask), t_dst_class(vpi->DstReg.File));
+			o_inst->src1=t_src(vp, &src[0]);
+			o_inst->src2=MAKE_VSF_SOURCE(t_src_index(vp, &src[0]),
+					SWIZZLE_ONE, SWIZZLE_ONE,
+					SWIZZLE_ONE, SWIZZLE_ONE,
+					t_src_class(src[0].File), VSF_FLAG_NONE);
+
+
+			o_inst->src3=MAKE_VSF_SOURCE(t_src_index(vp, &src[0]),
+					SWIZZLE_ZERO, SWIZZLE_ZERO,
+					SWIZZLE_ZERO, SWIZZLE_ZERO,
+					t_src_class(src[0].File), VSF_FLAG_NONE);
+#endif			
 
 			goto next;
 			
diff --git a/src/mesa/drivers/dri/r300/radeon_mm.c b/src/mesa/drivers/dri/r300/radeon_mm.c
new file mode 100644
index 00000000000..b69929e6642
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/radeon_mm.c
@@ -0,0 +1,339 @@
+#include <sys/ipc.h>
+#include <sys/shm.h>
+#include <sys/types.h>
+#include <sys/sem.h>
+
+#include "r300_context.h"
+#include "r300_cmdbuf.h"
+#include "radeon_mm.h"
+
+#ifdef USER_BUFFERS
+void radeon_mm_reset(r300ContextPtr rmesa)
+{
+	drm_r300_cmd_header_t *cmd;
+	int ret;
+	
+	memset(rmesa->rmm, 0, sizeof(struct radeon_memory_manager));
+		
+	rmesa->rmm->u_size = 1024; //2048;
+	//rmesa->radeon.radeonScreen->scratch[2] = rmesa->rmm->vb_age;
+#if 0 /* FIXME */
+	cmd = r300AllocCmdBuf(rmesa, 4, __FUNCTION__);
+	cmd[0].scratch.cmd_type = R300_CMD_SCRATCH;
+	cmd[0].scratch.reg = 2;
+	cmd[0].scratch.n_bufs = 1;
+	cmd[0].scratch.flags = 0;
+	cmd[1].u = (unsigned long)(&rmesa->rmm->vb_age);
+	cmd[2].u = (unsigned long)(&rmesa->rmm->u_list[0].age);
+	cmd[3].u = /*id*/0;
+
+	/* Protect from DRM. */	
+	LOCK_HARDWARE(&rmesa->radeon);
+	rmesa->rmm->u_list[0].h_pending ++;
+	ret = r300FlushCmdBufLocked(rmesa, __FUNCTION__);
+	UNLOCK_HARDWARE(&rmesa->radeon);
+
+	if (ret) {
+		WARN_ONCE("r300FlushCmdBufLocked\n");
+		exit(1);
+	}
+#endif
+}
+
+void radeon_mm_init(r300ContextPtr rmesa)
+{
+	
+	rmesa->mm_ipc_key = 0xdeadbeed; //ftok("/tmp/.r300.mm_lock", "x");
+	if(rmesa->mm_ipc_key == -1){
+		perror("ftok");
+		exit(1);
+	}
+	
+	rmesa->mm_shm_id = shmget(rmesa->mm_ipc_key, sizeof(struct radeon_memory_manager), 0644);
+	if (rmesa->mm_shm_id == -1) {
+		rmesa->mm_shm_id = shmget(rmesa->mm_ipc_key, sizeof(struct radeon_memory_manager), 0644 | IPC_CREAT);
+		
+		rmesa->rmm = shmat(rmesa->mm_shm_id, (void *)0, 0);
+		if (rmesa->rmm == (char *)(-1)) {
+			perror("shmat");
+			exit(1);
+		}
+		
+		radeon_mm_reset(rmesa);
+		
+		rmesa->mm_sem_id = semget(rmesa->mm_ipc_key, 2, 0666 | IPC_CREAT);
+		if (rmesa->mm_sem_id == -1) {
+			perror("semget");
+			exit(1);
+		}
+		
+		return ;
+	}
+	
+	rmesa->rmm = shmat(rmesa->mm_shm_id, (void *)0, 0);
+	if (rmesa->rmm == (char *)(-1)) {
+		perror("shmat");
+		exit(1);
+	}
+	/* FIXME */
+	radeon_mm_reset(rmesa);
+			
+	rmesa->mm_sem_id = semget(rmesa->mm_ipc_key, 2, 0666);
+	if (rmesa->mm_sem_id == -1) {
+		perror("semget");
+		exit(1);
+	}
+}
+
+static void radeon_mm_lock(r300ContextPtr rmesa, int sem)
+{
+	struct sembuf sb = { 0, 1, 0 };
+	
+	sb.sem_num = sem;
+	
+	if (semop(rmesa->mm_sem_id, &sb, 1) == -1) {
+		perror("semop");
+		exit(1);
+	}
+}
+
+static void radeon_mm_unlock(r300ContextPtr rmesa, int sem)
+{
+	struct sembuf sb = { 0, -1, 0 };
+	
+	sb.sem_num = sem;
+			
+	if (semop(rmesa->mm_sem_id, &sb, 1) == -1) {
+		perror("semop");
+		exit(1);
+	}	
+}
+
+void *radeon_mm_ptr(r300ContextPtr rmesa, int id)
+{
+	return rmesa->rmm->u_list[id].ptr;
+}
+
+//#define MM_DEBUG
+int radeon_mm_alloc(r300ContextPtr rmesa, int alignment, int size)
+{
+	drm_radeon_mem_alloc_t alloc;
+	int offset, ret;
+	int i, end, free=-1;
+	int done_age;
+	drm_radeon_mem_free_t memfree;
+	int tries=0, tries2=0;
+	
+	memfree.region = RADEON_MEM_REGION_GART;
+			
+	radeon_mm_lock(rmesa, RADEON_MM_UL);
+	
+	again:
+	
+	done_age = rmesa->radeon.radeonScreen->scratch[2];
+	
+	i = 1; //rmesa->rmm->u_head + 1;
+	//i &= rmesa->rmm->u_size - 1;
+	
+	end = i + rmesa->rmm->u_size;
+	//end &= rmesa->rmm->u_size - 1;
+	
+	for (; i != end; i ++/*, i &= rmesa->rmm->u_size-1*/) {
+		if (rmesa->rmm->u_list[i].ptr == NULL){
+			free = i;
+			continue;
+		}
+		
+		if (rmesa->rmm->u_list[i].h_pending == 0 &&
+			rmesa->rmm->u_list[i].pending && rmesa->rmm->u_list[i].age <= done_age) {
+			memfree.region_offset = (char *)rmesa->rmm->u_list[i].ptr -
+						(char *)rmesa->radeon.radeonScreen->gartTextures.map;
+
+			ret = drmCommandWrite(rmesa->radeon.radeonScreen->driScreen->fd,
+					      DRM_RADEON_FREE, &memfree, sizeof(memfree));
+
+			if (ret) {
+				//fprintf(stderr, "Failed to free at %p\n", rmesa->rmm->u_list[i].ptr);
+				//fprintf(stderr, "ret = %s\n", strerror(-ret));
+				
+				//radeon_mm_unlock(rmesa, RADEON_MM_UL);
+				//exit(1);
+			} else {
+#ifdef MM_DEBUG
+				fprintf(stderr, "really freed %d at age %x\n", i, rmesa->radeon.radeonScreen->scratch[2]);
+#endif
+				rmesa->rmm->u_list[i].pending = 0;
+				rmesa->rmm->u_list[i].ptr = NULL;
+				free = i;
+			}
+		}
+	}
+	done:
+	rmesa->rmm->u_head = i;
+	
+	if (free == -1) {
+		//usleep(100);
+		r300FlushCmdBuf(rmesa, __FUNCTION__);
+		tries++;
+		if(tries>100){
+			WARN_ONCE("Ran out of slots!\n");
+			exit(1);
+		}
+		goto again;
+	}
+		
+	alloc.region = RADEON_MEM_REGION_GART;
+	alloc.alignment = alignment;
+	alloc.size = size;
+	alloc.region_offset = &offset;
+
+	ret = drmCommandWriteRead( rmesa->radeon.dri.fd, DRM_RADEON_ALLOC, &alloc, sizeof(alloc));
+   	if (ret) {
+		r300FlushCmdBuf(rmesa, __FUNCTION__);
+		//usleep(100);
+		tries2++;
+		tries = 0;
+		if(tries2>100){
+			WARN_ONCE("Ran out of GART memory!\n");
+			exit(1);
+		}
+		goto again;
+	}
+	
+	i = free;
+	rmesa->rmm->u_list[i].ptr = ((GLubyte *)rmesa->radeon.radeonScreen->gartTextures.map) + offset;
+	rmesa->rmm->u_list[i].size = size;
+	rmesa->rmm->u_list[i].age = 0;
+	
+#ifdef MM_DEBUG
+	fprintf(stderr, "allocated %d at age %x\n", i, rmesa->radeon.radeonScreen->scratch[2]);
+#endif
+	
+	radeon_mm_unlock(rmesa, RADEON_MM_UL);
+	
+	return i;
+}
+
+void radeon_mm_use(r300ContextPtr rmesa, int id)
+{
+#ifdef MM_DEBUG
+	fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id, rmesa->radeon.radeonScreen->scratch[2]);
+#endif	
+	drm_r300_cmd_header_t *cmd;
+	
+	if(id == 0)
+		return;
+	
+	radeon_mm_lock(rmesa, RADEON_MM_UL);
+		
+	cmd = r300AllocCmdBuf(rmesa, 4, __FUNCTION__);
+	cmd[0].scratch.cmd_type = R300_CMD_SCRATCH;
+	cmd[0].scratch.reg = 2;
+	cmd[0].scratch.n_bufs = 1;
+	cmd[0].scratch.flags = 0;
+	cmd[1].u = (unsigned long)(&rmesa->rmm->vb_age);
+	cmd[2].u = (unsigned long)(&rmesa->rmm->u_list[id].age);
+	cmd[3].u = /*id*/0;
+	
+	LOCK_HARDWARE(&rmesa->radeon); /* Protect from DRM. */
+	rmesa->rmm->u_list[id].h_pending ++;
+	UNLOCK_HARDWARE(&rmesa->radeon);
+			
+	radeon_mm_unlock(rmesa, RADEON_MM_UL);
+}
+
+void *radeon_mm_map(r300ContextPtr rmesa, int id, int access)
+{
+#ifdef MM_DEBUG
+	fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id, rmesa->radeon.radeonScreen->scratch[2]);
+#endif	
+	void *ptr;
+	int tries = 0;
+	
+	if (access == RADEON_MM_R) {
+		radeon_mm_lock(rmesa, RADEON_MM_UL);
+		
+		if(rmesa->rmm->u_list[id].mapped == 1)
+			WARN_ONCE("buffer %d already mapped\n", id);
+	
+		rmesa->rmm->u_list[id].mapped = 1;
+		ptr = radeon_mm_ptr(rmesa, id);
+		
+		radeon_mm_unlock(rmesa, RADEON_MM_UL);
+		
+		return ptr;
+	}
+	
+	radeon_mm_lock(rmesa, RADEON_MM_UL);
+	
+	if (rmesa->rmm->u_list[id].h_pending)
+		r300FlushCmdBuf(rmesa, __FUNCTION__);
+	
+	if (rmesa->rmm->u_list[id].h_pending) {
+		radeon_mm_unlock(rmesa, RADEON_MM_UL);
+		return NULL;
+	}
+	
+	while(rmesa->rmm->u_list[id].age > rmesa->radeon.radeonScreen->scratch[2] && tries++ < 1000)
+		usleep(10);
+	
+	if (tries >= 1000) {
+		fprintf(stderr, "Idling failed (%x vs %x)\n",
+				rmesa->rmm->u_list[id].age, rmesa->radeon.radeonScreen->scratch[2]);
+		radeon_mm_unlock(rmesa, RADEON_MM_UL);
+		return NULL;
+	}
+	
+	if(rmesa->rmm->u_list[id].mapped == 1)
+		WARN_ONCE("buffer %d already mapped\n", id);
+	
+	rmesa->rmm->u_list[id].mapped = 1;
+	ptr = radeon_mm_ptr(rmesa, id);
+	
+	radeon_mm_unlock(rmesa, RADEON_MM_UL);
+	
+	return ptr;
+}
+
+void radeon_mm_unmap(r300ContextPtr rmesa, int id)
+{
+#ifdef MM_DEBUG
+	fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id, rmesa->radeon.radeonScreen->scratch[2]);
+#endif	
+	
+	radeon_mm_lock(rmesa, RADEON_MM_UL);
+	
+	if(rmesa->rmm->u_list[id].mapped == 0)
+		WARN_ONCE("buffer %d not mapped\n", id);
+	
+	rmesa->rmm->u_list[id].mapped = 0;
+	
+	radeon_mm_unlock(rmesa, RADEON_MM_UL);
+}
+
+void radeon_mm_free(r300ContextPtr rmesa, int id)
+{
+#ifdef MM_DEBUG
+	fprintf(stderr, "%s: %d at age %x\n", __FUNCTION__, id, rmesa->radeon.radeonScreen->scratch[2]);
+#endif	
+	
+	if(id == 0)
+		return;
+	
+	radeon_mm_lock(rmesa, RADEON_MM_UL);
+	if(rmesa->rmm->u_list[id].ptr == NULL){
+		radeon_mm_unlock(rmesa, RADEON_MM_UL);
+		WARN_ONCE("Not allocated!\n");
+		return ;
+	}
+	
+	if(rmesa->rmm->u_list[id].pending){
+		radeon_mm_unlock(rmesa, RADEON_MM_UL);
+		WARN_ONCE("%p already pended!\n", rmesa->rmm->u_list[id].ptr);
+		return ;
+	}
+			
+	rmesa->rmm->u_list[id].pending = 1;
+	radeon_mm_unlock(rmesa, RADEON_MM_UL);
+}
+#endif
diff --git a/src/mesa/drivers/dri/r300/radeon_mm.h b/src/mesa/drivers/dri/r300/radeon_mm.h
new file mode 100644
index 00000000000..637ecb1ddaf
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/radeon_mm.h
@@ -0,0 +1,35 @@
+#ifndef __RADEON_MM_H__
+#define __RADEON_MM_H__
+
+//#define RADEON_MM_PDL 0
+#define RADEON_MM_UL 1
+
+#define RADEON_MM_R 1
+#define RADEON_MM_W 2
+#define RADEON_MM_RW (RADEON_MM_R | RADEON_MM_W)
+    
+struct radeon_memory_manager {
+	uint32_t vb_age;
+	/*uint32_t ages[1024];*/
+	
+	struct {
+		void *ptr;
+		uint32_t size;
+		uint32_t age;
+		uint32_t h_pending;
+		int pending;
+		int mapped;
+	} u_list[/*4096*/2048];
+	int u_head, u_tail, u_size;
+	
+};
+
+extern void radeon_mm_init(r300ContextPtr rmesa);
+extern void *radeon_mm_ptr(r300ContextPtr rmesa, int id);
+extern int radeon_mm_alloc(r300ContextPtr rmesa, int alignment, int size);
+extern void radeon_mm_use(r300ContextPtr rmesa, int id);
+extern void *radeon_mm_map(r300ContextPtr rmesa, int id, int access);
+extern void radeon_mm_unmap(r300ContextPtr rmesa, int id);
+extern void radeon_mm_free(r300ContextPtr rmesa, int id);
+
+#endif
diff --git a/src/mesa/drivers/dri/r300/radeon_vtxfmt_a.c b/src/mesa/drivers/dri/r300/radeon_vtxfmt_a.c
new file mode 100644
index 00000000000..d1f90abfb24
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/radeon_vtxfmt_a.c
@@ -0,0 +1,610 @@
+#include "context.h"
+#include "r300_context.h"
+#include "r300_cmdbuf.h"
+#include "radeon_mm.h"
+
+#ifdef RADEON_VTXFMT_A
+
+#define CONV(a, b) rmesa->state.VB.AttribPtr[(a)].size = ctx->Array.b.Size, \
+			rmesa->state.VB.AttribPtr[(a)].data = ctx->Array.b.BufferObj->Name ? \
+			ADD_POINTERS(ctx->Array.b.Ptr, ctx->Array.b.BufferObj->Data) : ctx->Array.b.Ptr, \
+			rmesa->state.VB.AttribPtr[(a)].stride = ctx->Array.b.StrideB, \
+			rmesa->state.VB.AttribPtr[(a)].type = ctx->Array.b.Type
+
+static int setup_arrays(r300ContextPtr rmesa, GLint start)
+{
+	int i;
+	struct dt def = { 4, GL_FLOAT, 0, NULL };
+	GLcontext *ctx;
+	GLuint enabled = 0;
+	
+	ctx = rmesa->radeon.glCtx; 
+
+	memset(rmesa->state.VB.AttribPtr, 0, VERT_ATTRIB_MAX*sizeof(struct dt));
+	
+	CONV(VERT_ATTRIB_POS, Vertex);
+	if (ctx->Array.Vertex.Enabled)
+		enabled |= 1 << VERT_ATTRIB_POS;
+	
+	CONV(VERT_ATTRIB_NORMAL, Normal);
+	if (ctx->Array.Normal.Enabled)
+		enabled |= 1 << VERT_ATTRIB_NORMAL;
+	
+	CONV(VERT_ATTRIB_COLOR0, Color);
+	if (ctx->Array.Color.Enabled)
+		enabled |= 1 << VERT_ATTRIB_COLOR0;
+	
+	CONV(VERT_ATTRIB_COLOR1, SecondaryColor);
+	if (ctx->Array.SecondaryColor.Enabled)
+		enabled |= 1 << VERT_ATTRIB_COLOR1;
+	
+	CONV(VERT_ATTRIB_FOG, FogCoord);
+	if (ctx->Array.FogCoord.Enabled)
+		enabled |= 1 << VERT_ATTRIB_FOG;
+	
+	for (i=0; i < MAX_TEXTURE_COORD_UNITS; i++) {
+		CONV(VERT_ATTRIB_TEX0 + i, TexCoord[i]);
+		
+		if(ctx->Array.TexCoord[i].Enabled) {
+			enabled |= 1 << (VERT_ATTRIB_TEX0+i);
+		}
+		
+	}
+
+	for (i=0; i < VERT_ATTRIB_MAX; i++) {
+		if (enabled & (1 << i)) {
+			rmesa->state.VB.AttribPtr[i].data += rmesa->state.VB.AttribPtr[i].stride * start;
+		} else {
+			def.data = ctx->Current.Attrib[i];
+			memcpy(&rmesa->state.VB.AttribPtr[i], &def, sizeof(struct dt));
+		}
+		
+		/*if(rmesa->state.VB.AttribPtr[i].data == ctx->Current.Attrib[i])
+			fprintf(stderr, "%d is default coord\n", i);*/
+	}
+	
+	for(i=0; i < VERT_ATTRIB_MAX; i++){
+		if(rmesa->state.VB.AttribPtr[i].type != GL_UNSIGNED_BYTE &&
+			rmesa->state.VB.AttribPtr[i].type != GL_FLOAT){
+			WARN_ONCE("Unsupported format %d at index %d\n", rmesa->state.VB.AttribPtr[i].type, i);
+			return -1;
+		}
+		if(rmesa->state.VB.AttribPtr[i].type == GL_UNSIGNED_BYTE &&
+			rmesa->state.VB.AttribPtr[i].size != 4){
+			WARN_ONCE("Unsupported component count for ub colors\n");
+			return -1;
+		}
+		
+		/*fprintf(stderr, "%d: ", i);
+		
+		switch(rmesa->state.VB.AttribPtr[i].type){
+		case GL_BYTE: fprintf(stderr, "byte "); break;
+		case GL_UNSIGNED_BYTE: fprintf(stderr, "u byte "); break;
+		case GL_SHORT: fprintf(stderr, "short "); break;
+		case GL_UNSIGNED_SHORT: fprintf(stderr, "u short "); break;
+		case GL_INT: fprintf(stderr, "int "); break;
+		case GL_UNSIGNED_INT: fprintf(stderr, "u int "); break;
+		case GL_FLOAT: fprintf(stderr, "float "); break;
+		case GL_2_BYTES: fprintf(stderr, "2 bytes "); break;
+		case GL_3_BYTES: fprintf(stderr, "3 bytes "); break;
+		case GL_4_BYTES: fprintf(stderr, "4 bytes "); break;
+		case GL_DOUBLE: fprintf(stderr, "double "); break;
+		default: fprintf(stderr, "unknown "); break;
+		}
+
+		fprintf(stderr, "Size %d ", rmesa->state.VB.AttribPtr[i].size);
+		fprintf(stderr, "Ptr %p ", rmesa->state.VB.AttribPtr[i].data);
+		fprintf(stderr, "Stride %d ", rmesa->state.VB.AttribPtr[i].stride);
+		fprintf(stderr, "\n");*/
+	}
+	return 0;
+}
+
+void radeonDrawElements( GLenum mode, GLsizei count, GLenum type, const GLvoid *indices )
+{
+	GET_CURRENT_CONTEXT(ctx);
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	int elt_size;
+	int i;
+	unsigned int min = ~0, max = 0;
+	struct tnl_prim prim;
+	static void *ptr = NULL;
+	static struct r300_dma_region rvb;
+	
+	if (ctx->Array.ElementArrayBufferObj->Name) {
+		/* use indices in the buffer object */
+		if (!ctx->Array.ElementArrayBufferObj->Data) {
+			_mesa_warning(ctx, "DrawRangeElements with empty vertex elements buffer!");
+			return;
+		}
+		/* actual address is the sum of pointers */
+		indices = (const GLvoid *)
+		ADD_POINTERS(ctx->Array.ElementArrayBufferObj->Data, (const GLubyte *) indices);
+	}
+	
+	if (!_mesa_validate_DrawElements( ctx, mode, count, type, indices ))
+		return;
+	
+	FLUSH_CURRENT( ctx, 0 );
+	/*
+		fprintf(stderr, "dt at %s:\n", __FUNCTION__);
+		for(i=0; i < VERT_ATTRIB_MAX; i++){
+			fprintf(stderr, "dt %d:", i);
+			dump_dt(&rmesa->state.VB.AttribPtr[i], rmesa->state.VB.Count);
+		}*/
+	r300ReleaseDmaRegion(rmesa, &rvb, __FUNCTION__);
+	
+	switch (type) {
+	case GL_UNSIGNED_BYTE:
+		elt_size = 2;
+		
+		r300AllocDmaRegion(rmesa, &rvb, count * elt_size, elt_size);
+		rvb.aos_offset = GET_START(&rvb);
+		ptr = rvb.address + rvb.start;
+			
+		for (i=0; i < count; i++) {
+			if(((unsigned char *)indices)[i] < min)
+				min = ((unsigned char *)indices)[i];
+			if(((unsigned char *)indices)[i] > max)
+				max = ((unsigned char *)indices)[i];
+		}
+		
+		for (i=0; i < count; i++)
+			((unsigned short int *)ptr)[i] = ((unsigned char *)indices)[i] - min;
+	break;
+		
+	case GL_UNSIGNED_SHORT:
+		elt_size = 2;
+		
+		r300AllocDmaRegion(rmesa, &rvb, count * elt_size, elt_size);
+		rvb.aos_offset = GET_START(&rvb);
+		ptr = rvb.address + rvb.start;
+		
+		for (i=0; i < count; i++) {
+			if(((unsigned short int *)indices)[i] < min)
+				min = ((unsigned short int *)indices)[i];
+			if(((unsigned short int *)indices)[i] > max)
+				max = ((unsigned short int *)indices)[i];
+		}
+		
+		for (i=0; i < count; i++)
+			((unsigned short int *)ptr)[i] = ((unsigned short int *)indices)[i] - min;
+	break;
+	
+	case GL_UNSIGNED_INT:
+		elt_size = 4;
+		
+		r300AllocDmaRegion(rmesa, &rvb, count * elt_size, elt_size);
+		rvb.aos_offset = GET_START(&rvb);
+		ptr = rvb.address + rvb.start;
+		
+		for (i=0; i < count; i++) {
+			if(((unsigned int *)indices)[i] < min)
+				min = ((unsigned int *)indices)[i];
+			if(((unsigned int *)indices)[i] > max)
+				max = ((unsigned int *)indices)[i];
+		}
+		
+		for (i=0; i < count; i++)
+			((unsigned int *)ptr)[i] = ((unsigned int *)indices)[i] - min;
+	break;
+	
+	default:
+		fprintf(stderr, "Unknown elt type!\n");
+	return;
+	
+
+	}
+	
+	if (ctx->NewState) 
+		_mesa_update_state( ctx );
+	
+	r300UpdateShaderStates(rmesa);
+	
+	if (rmesa->state.VB.LockCount) {
+		if (rmesa->state.VB.lock_uptodate == GL_FALSE) {
+			if (setup_arrays(rmesa, rmesa->state.VB.LockFirst))
+				return;
+			
+			rmesa->state.VB.Count = rmesa->state.VB.LockCount;
+			
+			r300ReleaseArrays(ctx);
+			r300EmitArraysVtx(ctx, GL_FALSE);
+			
+			rmesa->state.VB.lock_uptodate = GL_TRUE;
+		}
+		
+		if (min < rmesa->state.VB.LockFirst) {
+			WARN_ONCE("Out of range min %d vs %d!\n", min, rmesa->state.VB.LockFirst);
+			return;
+		}
+		
+		if (max >= rmesa->state.VB.LockFirst + rmesa->state.VB.LockCount) {
+			WARN_ONCE("Out of range max %d vs %d!\n", max, rmesa->state.VB.LockFirst +
+					rmesa->state.VB.LockCount);
+			return;
+		}
+	} else {
+		if (setup_arrays(rmesa, min))
+			return;
+		rmesa->state.VB.Count = max - min + 1;
+	}
+	
+	rmesa->state.VB.Primitive = &prim;
+	rmesa->state.VB.PrimitiveCount = 1;
+	
+	prim.mode = mode | PRIM_BEGIN | PRIM_END;
+	if (rmesa->state.VB.LockCount)
+		prim.start = min - rmesa->state.VB.LockFirst;
+	else
+		prim.start = 0;
+	prim.count = count;
+	
+	rmesa->state.VB.Elts = ptr;
+	rmesa->state.VB.elt_size = elt_size;
+	
+	r300_run_vb_render_vtxfmt_a(ctx, NULL);
+	
+	if(rvb.buf)
+		radeon_mm_use(rmesa, rvb.buf->id);
+}
+
+void radeonDrawRangeElements(GLenum mode, GLuint start, GLuint end, GLsizei count, GLenum type, const GLvoid *indices)
+{
+	GET_CURRENT_CONTEXT(ctx);
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct tnl_prim prim;
+	int elt_size;
+	int i;
+	static void *ptr = NULL;
+	static struct r300_dma_region rvb;
+	
+	if (ctx->Array.ElementArrayBufferObj->Name) {
+		/* use indices in the buffer object */
+		if (!ctx->Array.ElementArrayBufferObj->Data) {
+			_mesa_warning(ctx, "DrawRangeElements with empty vertex elements buffer!");
+			return;
+		}
+		/* actual address is the sum of pointers */
+		indices = (const GLvoid *)
+		ADD_POINTERS(ctx->Array.ElementArrayBufferObj->Data, (const GLubyte *) indices);
+	}
+	
+	if (!_mesa_validate_DrawRangeElements( ctx, mode, start, end, count, type, indices ))
+		return;
+	
+	FLUSH_CURRENT( ctx, 0 );
+#ifdef OPTIMIZE_ELTS
+	start = 0;
+#endif
+	r300ReleaseDmaRegion(rmesa, &rvb, __FUNCTION__);
+	
+	switch(type){
+	case GL_UNSIGNED_BYTE:
+		elt_size = 2;
+		
+		r300AllocDmaRegion(rmesa, &rvb, count * elt_size, elt_size);
+		rvb.aos_offset = GET_START(&rvb);
+		ptr = rvb.address + rvb.start;
+		
+		for(i=0; i < count; i++)
+			((unsigned short int *)ptr)[i] = ((unsigned char *)indices)[i] - start;
+	break;
+	
+	case GL_UNSIGNED_SHORT:
+		elt_size = 2;
+		
+#ifdef OPTIMIZE_ELTS
+		if (start == 0 && ctx->Array.ElementArrayBufferObj->Name){
+			ptr = indices;
+			break;
+		}
+#endif
+		r300AllocDmaRegion(rmesa, &rvb, count * elt_size, elt_size);
+		rvb.aos_offset = GET_START(&rvb);
+		ptr = rvb.address + rvb.start;
+
+		for(i=0; i < count; i++)
+			((unsigned short int *)ptr)[i] = ((unsigned short int *)indices)[i] - start;
+	break;
+	
+	case GL_UNSIGNED_INT:
+		elt_size = 4;
+		
+		r300AllocDmaRegion(rmesa, &rvb, count * elt_size, elt_size);
+		rvb.aos_offset = GET_START(&rvb);
+		ptr = rvb.address + rvb.start;
+		
+		for(i=0; i < count; i++)
+			((unsigned int *)ptr)[i] = ((unsigned int *)indices)[i] - start;
+	break;
+	
+	default:
+		fprintf(stderr, "Unknown elt type!\n");
+	return;
+	
+	}
+	
+	if(setup_arrays(rmesa, start))
+		return;
+	
+	if (ctx->NewState) 
+		_mesa_update_state( ctx );
+	
+	r300UpdateShaderStates(rmesa);
+
+	rmesa->state.VB.Count = (end - start) + 1;
+	rmesa->state.VB.Primitive = &prim;
+	rmesa->state.VB.PrimitiveCount = 1;
+	
+	prim.mode = mode | PRIM_BEGIN | PRIM_END;
+	prim.start = 0;
+	prim.count = count;
+	
+	rmesa->state.VB.Elts = ptr;
+	rmesa->state.VB.elt_size = elt_size;
+	rmesa->state.VB.elt_min = start;
+	rmesa->state.VB.elt_max = end;
+	
+	r300_run_vb_render_vtxfmt_a(ctx, NULL);
+	
+	if(rvb.buf)
+		radeon_mm_use(rmesa, rvb.buf->id);
+}
+
+void radeonDrawArrays( GLenum mode, GLint start, GLsizei count )
+{
+	GET_CURRENT_CONTEXT(ctx);
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct tnl_prim prim;
+	
+	if (!_mesa_validate_DrawArrays( ctx, mode, start, count ))
+		return;
+	
+	FLUSH_CURRENT( ctx, 0 );
+	
+	if (ctx->NewState) 
+		_mesa_update_state( ctx );
+	
+	if (rmesa->state.VB.LockCount == 0)
+		if (setup_arrays(rmesa, start))
+			return;
+	
+	r300UpdateShaderStates(rmesa);
+	
+	if (rmesa->state.VB.LockCount) {
+		start -= rmesa->state.VB.LockFirst;
+		if (start < 0) { /* Generate error */
+			WARN_ONCE("Out of range!\n");
+			return;
+		}
+	}
+		
+	if (rmesa->state.VB.LockCount == 0)
+		rmesa->state.VB.Count = count;
+	rmesa->state.VB.Primitive = &prim;
+	rmesa->state.VB.PrimitiveCount = 1;
+	
+	prim.mode = mode | PRIM_BEGIN | PRIM_END;
+	if (ctx->Array.LockCount == 0)
+		prim.start = 0;
+	else
+		prim.start = start;
+	
+	prim.count = count;
+	
+	rmesa->state.VB.Elts = NULL;
+	rmesa->state.VB.elt_size = 0;
+	rmesa->state.VB.elt_min = 0;
+	rmesa->state.VB.elt_max = 0;
+	
+	r300_run_vb_render_vtxfmt_a(ctx, NULL);
+}
+
+void radeon_init_vtxfmt_a(r300ContextPtr rmesa)
+{
+	GLcontext *ctx;
+	GLvertexformat *vfmt;
+	
+	ctx = rmesa->radeon.glCtx; 
+	vfmt = ctx->TnlModule.Current;
+   
+	vfmt->DrawElements = radeonDrawElements;
+	vfmt->DrawArrays = radeonDrawArrays;
+	vfmt->DrawRangeElements = radeonDrawRangeElements;
+	
+}
+#endif
+
+#ifdef HW_VBOS
+
+void radeonLockArraysEXT(GLcontext *ctx, GLint first, GLsizei count)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	int i;
+	
+	/* Disabled as array changes arent properly handled yet. */
+	first = 0; count = 0;
+	
+	if (first < 0 || count <= 0) {
+		rmesa->state.VB.LockFirst = 0;
+		rmesa->state.VB.LockCount = 0;
+		return ;
+	}
+	
+	rmesa->state.VB.LockFirst = first;
+	rmesa->state.VB.LockCount = count;
+	rmesa->state.VB.lock_uptodate = GL_FALSE;
+}
+
+void radeonUnlockArraysEXT(GLcontext *ctx)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	
+	rmesa->state.VB.LockFirst = 0;
+	rmesa->state.VB.LockCount = 0;
+}
+
+struct gl_buffer_object *
+r300NewBufferObject(GLcontext *ctx, GLuint name, GLenum target )
+{
+	struct r300_buffer_object *obj;
+
+	(void) ctx;
+
+	obj = MALLOC_STRUCT(r300_buffer_object);
+	_mesa_initialize_buffer_object(&obj->mesa_obj, name, target);
+	return &obj->mesa_obj;
+}
+
+void r300BufferData(GLcontext *ctx, GLenum target, GLsizeiptrARB size,
+		const GLvoid *data, GLenum usage, struct gl_buffer_object *obj)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct r300_buffer_object *r300_obj = (struct r300_buffer_object *)obj;
+	drm_radeon_mem_alloc_t alloc;
+	int offset, ret;
+
+	/* Free previous buffer */
+	if (obj->OnCard) {
+		radeon_mm_free(rmesa, r300_obj->id);
+		obj->OnCard = GL_FALSE;
+	} else {
+		if (obj->Data)
+			free(obj->Data);
+	}
+#ifdef OPTIMIZE_ELTS
+	if (0) {
+#else
+	if (target == GL_ELEMENT_ARRAY_BUFFER_ARB) {
+#endif
+		obj->Data = malloc(size);
+		
+		if (data)
+			memcpy(obj->Data, data, size);
+		
+		obj->OnCard = GL_FALSE;
+	} else {
+		r300_obj->id = radeon_mm_alloc(rmesa, 4, size);
+		obj->Data = radeon_mm_map(rmesa, r300_obj->id, RADEON_MM_W);
+	
+		if (data)
+			memcpy(obj->Data, data, size);
+	
+		radeon_mm_unmap(rmesa, r300_obj->id);
+		obj->OnCard = GL_TRUE;
+	}
+	
+	obj->Size = size;
+	obj->Usage = usage;
+}
+
+void r300BufferSubData(GLcontext *ctx, GLenum target, GLintptrARB offset,
+		GLsizeiptrARB size, const GLvoid * data, struct gl_buffer_object * bufObj)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct r300_buffer_object *r300_obj = (struct r300_buffer_object *)bufObj;
+	(void) ctx; (void) target;
+	void *ptr;
+
+	if (bufObj->Data && ((GLuint) (size + offset) <= bufObj->Size)) {
+		if (bufObj->OnCard){
+			ptr = radeon_mm_map(rmesa, r300_obj->id, RADEON_MM_W);
+		
+			_mesa_memcpy( (GLubyte *) ptr + offset, data, size );
+		
+			radeon_mm_unmap(rmesa, r300_obj->id);
+		} else {
+			_mesa_memcpy( (GLubyte *) bufObj->Data + offset, data, size );
+		}
+	}
+}
+
+void *r300MapBuffer(GLcontext *ctx, GLenum target, GLenum access,
+		struct gl_buffer_object *bufObj)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct r300_buffer_object *r300_obj = (struct r300_buffer_object *)bufObj;
+	
+	(void) ctx;
+	(void) target;
+	(void) access;
+	//ASSERT(!bufObj->OnCard);
+	/* Just return a direct pointer to the data */
+	if (bufObj->Pointer) {
+		/* already mapped! */
+		return NULL;
+	}
+	
+	if (!bufObj->OnCard) {
+		bufObj->Pointer = bufObj->Data;
+		return bufObj->Pointer;
+	}
+	
+	switch (access) {
+	case GL_READ_ONLY:
+		bufObj->Pointer = radeon_mm_map(rmesa, r300_obj->id, RADEON_MM_R);
+	break;
+	
+	case GL_WRITE_ONLY:
+		bufObj->Pointer = radeon_mm_map(rmesa, r300_obj->id, RADEON_MM_W);
+	break;
+	
+	case GL_READ_WRITE:
+		bufObj->Pointer = radeon_mm_map(rmesa, r300_obj->id, RADEON_MM_RW);
+	break;
+	
+	default:
+		WARN_ONCE("Unknown access type\n");
+		bufObj->Pointer = NULL;
+	break;
+	}
+	
+	return bufObj->Pointer;
+}
+
+GLboolean r300UnmapBuffer(GLcontext *ctx, GLenum target, struct gl_buffer_object *bufObj)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct r300_buffer_object *r300_obj = (struct r300_buffer_object *)bufObj;
+	
+	(void) ctx;
+	(void) target;
+	//ASSERT(!bufObj->OnCard);
+	/* XXX we might assert here that bufObj->Pointer is non-null */
+	if (!bufObj->OnCard) {
+		bufObj->Pointer = NULL;
+		return GL_TRUE;
+	}
+	radeon_mm_unmap(rmesa, r300_obj->id);
+	
+	bufObj->Pointer = NULL;
+	return GL_TRUE;
+}
+
+void r300DeleteBuffer(GLcontext *ctx, struct gl_buffer_object *obj)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	struct r300_buffer_object *r300_obj = (struct r300_buffer_object *)obj;
+	
+	if (obj->OnCard) {
+		radeon_mm_free(rmesa, r300_obj->id);
+		obj->Data = NULL;
+	}
+	_mesa_delete_buffer_object(ctx, obj);
+}
+
+void r300_init_vbo_funcs(struct dd_function_table *functions)
+{
+	functions->NewBufferObject = r300NewBufferObject;
+	functions->BufferData = r300BufferData;
+	functions->BufferSubData = r300BufferSubData;
+	functions->MapBuffer = r300MapBuffer;
+	functions->UnmapBuffer = r300UnmapBuffer;
+	functions->DeleteBuffer = r300DeleteBuffer;
+	
+	functions->LockArraysEXT = radeonLockArraysEXT;
+	functions->UnlockArraysEXT = radeonUnlockArraysEXT;
+}
+
+#endif
-- 
2.30.2